content
stringlengths 1
15.9M
|
|---|
\section{Introduction}
Recent results from the Super-Kamiokande have sparked tremendous interest
in neutrino physics \cite{skan,sksn}. The deficits seen in the solar and
atmospheric neutrino fluxes can be very naturally explained in terms of
neutrino oscillations. Since the energy and the distance scales of the
solar neutrino problem are widely different from those of the atmospheric
neutrino problem, the neutrino oscillation solution to these problems
requires widely different values of mass-squared differences
\cite{nmru,flm}.
A minimum of three mass eigenstates, and hence three neutrino flavors,
are
needed for a simultaneous solution of solar and atmospheric neutrino
problems. Since LEP has shown that three light, active neutrino species
exist \cite{lep}, it is natural to consider all neutrino data in the
framework of oscillations between all the three active neutrino flavors.
The flavor eigenstates $\nu_\alpha (\alpha = e, \mu, \tau)$ are related
to the mass eigenstates $\nu_i (i = 1,2,3)$ by a unitary matrix $U$
\begin{equation}
|\nu_\alpha \rangle = U | \nu_i \rangle. \label{eq:flmarel}
\end{equation}
As in the quark sector, $U$ can be parametrized in terms of three mixing
angles and one phase. A widely used parametrization, convenient for
analyzing neutrino oscillations with matter effects, is \cite{kupa}
\begin{equation}
U = U^{23} (\psi) \times U^{phase} \times U^{13} (\phi)
\times U^{12} (\omega),
\label{eq:defU}
\end{equation}
where $U^{ij} (\theta_{ij})$ is the two flavor mixing matrix between
the $i$th and $j$th mass eigenstates with the mixing angle $\theta_{ij}$.
We assume that the vacuum mass eigenvalues have the pattern $\mu_3^2 \gg
\mu_2^2 >
\mu_1^2$. Hence $\delta_{31} \simeq \delta_{32} \gg \delta_{21}$, where
$\delta_{ij} = \mu_i^2 - \mu_j^2$. The larger $\delta$ sets the scale for
the
atmospheric neutrino oscillations and the smaller one for solar
neutrino oscillations. The angles $\omega, \phi$ and $\psi$ vary in the
range $\left[0, \pi/2\right]$. For this range of the mixing angles, there
is no loss of generality due to the assumption that $\delta_{31},
\delta_{21} \geq 0$ \cite{flms}.
It has been shown \cite{kupa,jim94} that in the above approximation of one
dominant mass, the oscillation probability for solar neutrinos is a
function of only three parameters ($\delta_{21}, \omega$ and $\phi$) and
that of the atmospheric neutrinos and long baseline neutrinos is also
function of only three
parameters ($\delta_{31}, \phi$ and $\psi$). In each case, the three
flavor nature of the problem is illustrated by the fact that the
oscillation probability is a function of two mixing angles.
The phase is unobservable in the one dominant mass approximation because,
in both solar and atmospheric neutrino oscillations, one of the three
mixing angles can be set to zero \cite{flms}.
The value of $\delta_{31}$ preferred by the Super-K
atmospheric neutrino data is about $2 \times 10^{-3}$ eV$^2$
\cite{skan}. For this
value of $\delta_{31}$, CHOOZ data sets a very strong constraint
\cite{choozex,choozppr}
\begin{equation}
\sin^2 2 \phi \leq 0.2. \label{eq:chzcnst}
\end{equation}
If $\phi = 0$, then the solar and atmospheric neutrino oscillations get
decoupled and become two flavor oscillations with relavant parameters
being $(\delta_{21}, \omega)$ and $(\delta_{31}, \psi)$ respectively.
It is interesting to look for the consequences of non-zero $\phi$. A
non-zero $\phi$ leads to $\nu_\mu \rightarrow \nu_e$ oscillations in
atmospheric neutrinos and long baseline experiments.
Due to theoretical
uncertainty in the calculation of atmospheric neutrino fluxes, it will be
very difficult to discern the effect of a small $\phi$ from the
atmospheric neutrino data. Recently a proposal was made to look
for matter enhanced $\nu_\mu \rightarrow \nu_e$ oscillations in
atmospheric neutrino data, which can be significant even if $\phi$
is small \cite{jim98}. Here we consider the effect of matter on
$\nu_\mu \rightarrow \nu_e$ oscillations in a three flavor framework
at long baseline experiments.
The relation between the flavor states and mass eigenstates can be
written in the simple form,
\begin{equation}
\left[ \begin{array}{c} \nu_e \\ c_\psi \nu_\mu - s_\psi \nu_\tau \\
s_\psi \nu_\mu + c_\psi \nu_\tau \end{array} \right] =
\left( \begin{array}{ccc} c_\phi & 0 & s_\phi \\ 0 & 1 & 0 \\
-s_\phi & 0 & c_\phi \end{array} \right) \left[ \begin{array}{c}
\nu_1 \\ \nu_2 \\ \nu_3 \end{array} \right], \label{eq:matrix}
\end{equation}
where $c$ stands for cosine and $s$ stands for sine.
This equation is obtained by first setting $U^{phase} = I$ and
$\omega = 0$ in equation~(\ref{eq:defU}). Then substitute the
resulting form $U$ in equation~(\ref{eq:flmarel}) and multiply
it from left by $U^{23} (-\psi)$.
From~(\ref{eq:matrix}) we see that $\nu_2$ has no
$\nu_e$ component and hence is decoupled from any oscillation
involving $\nu_e$. Thus the calculation of oscillation probability
is essentially a two flavor problem. The three flavor nature is
present in the fact that the oscillations occur between $(\nu_e
\rightarrow (s_\psi \nu_\mu + c_\psi \nu_\tau))$.
The vacuum oscillation probability is calculated to be
\begin{equation}
P_{\mu e} = \sin^2 \psi \sin^2 2 \phi \sin^2 \left( \frac{1.27 \
\delta_{31}
L}{E} \right), \label{eq:vacP}
\end{equation}
where $\delta_{31}$ is in eV$^2$, the baseline length $L$ is in km and
the neutrino energy $E$ is in GeV. For low energies the phase $1.27
\delta_{31} L/E$ oscillates rapidly and $P_{\mu e}$ becomes insensitive
to it. However, for the range of energies for which the phase is close
to $\pi/2$, $P_{\mu e}$ varies slowly as shown in figures (1)-(3).
If the spectrum of the neutrinos in long baseline experiments peaks in
this range, then $\delta_{31}$ and the mixing angles can be
determined quite accurately.
Given the CHOOZ constraint on
$\phi$, Super-Kamiokande atmospheric neutrino data
fix $\sin^2 2 \psi \simeq 1$ or $\sin^2 \psi \simeq 0.5$.
Substituting this value in equation~(\ref{eq:vacP}),
we find that $P_{\mu e} \leq 0.1$. The neutrino beams at K2K
and MINOS have a $1 \%$ $\nu_e$ contamination. This background
and the systematic uncertainties set a limit to the smallest
value of $P_{\mu e}$ (and $\phi$) that can be measured.
With the current estimates of the systematic uncertainties,
the sensitivity of K2K is similar to that of CHOOZ ($\sin^2 2 \phi
\sim 0.2$) but MINOS is capable of measuring values of $\phi$ as
small as $3^o (\sin^2 2 \phi \geq 0.01)$ \cite{oyama}. Once K2K
starts running, its systematics will be better understood and
its sensitivity is likely to improve.
In both K2K and MINOS experiments, the neutrino beam travels through
earth's crust, where it encounters matter of roughly constant density
3 gm/cc. This traversal leads to the addition of the Wolfenstein term to
the $e-e$ element of the (mass)$^2$ matrix, when it is written in the
flavor basis \cite{wolf}. The Wolfenstein term is given by
\begin{equation}
A = 0.76 \times \rho ({\rm gm/cc}) \times E ({\rm GeV}) \times 10^{-4} \
{\rm eV}^2.
\end{equation}
For $\rho$ of a few gm/cc and $E$ of a few GeV, $A$ is comparable to
the value of $\delta_{31}$ set by atmospheric neutrino data.
This can lead to
interesting and observable matter effects in long baseline experiments.
The interactions of $\nu_\mu$ and $\nu_\tau$ with ordinary matter
are identical. Hence equation~(\ref{eq:matrix}) can be used to
include matter effects in a simple manner because the
problem, once again, is essentially a two flavor one.
It is easy to see that $\psi$ is unaffected by
matter and the matter dependent mixing angle $\phi_m$ and the mass
eigenvalues $m_1^2$ and $m_3^2$ are given by \cite{jim94}
\begin{eqnarray}
\tan 2 \phi_m & = & \frac{\delta_{31} \sin 2 \phi}{
\delta_{31} \cos 2 \phi - A}, \label{eq:tan2phim} \\
m_1^2 & = & \frac{1}{2} \left[ \left(\delta_{31} + A \right)
- \sqrt{\left(\delta_{31} \cos 2 \phi - A \right)^2 +
\left(\delta_{31} \sin 2 \phi \right)^2} \right] ,
\label{eq:m1sq} \\
m_3^2 & = & \frac{1}{2} \left[ \left(\delta_{31} + A \right)
+ \sqrt{\left(\delta_{31} \cos 2 \phi - A \right)^2 +
\left(\delta_{31} \sin 2 \phi \right)^2} \right] .
\label{eq:m3sq}
\end{eqnarray}
The above equations hold for the propagation of neutrinos. For vacuum
propagation, the mass eigenvalues and mixing angles for neutrinos and
anti-neutrinos are the same. However, to include matter effects for
anti-neutrinos, one should replace $A$ by $-A$ in
equations~(\ref{eq:tan2phim}), (\ref{eq:m1sq}) and~(\ref{eq:m3sq}).
Note that, since $A$ is always positive, matter effects enhance the
neutrino mixing angle and suppress the anti-neutrino mixing angle if
$(\delta_{31} \cos 2 \phi)$ is positive and vice-verse if $(\delta_{31}
\cos 2 \phi)$ is negative. Since we have taken $\delta_{31}$
to be positive, matter effects enhance neutrino mixing if $\cos 2 \phi$ is
positive or $\phi < \pi/4$. The CHOOZ constraint from
equation~(\ref{eq:chzcnst}) on $\sin^2 2 \phi$ sets
the limit $\phi \leq 13^o$ or $\phi \geq 77^o$. For the latter
possibility, $\cos 2 \phi$ is negative. However, this large a value of
$\phi$ is forbidden by the three flavor analysis of solar neutrino data,
which yields the independent
constraint $\phi \leq 50^o$ \cite{nmru,flm}. Hence the enhancement of the
mixing angle will be for neutrinos rather than for anti-neutrinos. This is
good news because the beams in long baseline experiments
consist of neutrinos overwhelmingly.
The matter dependent oscillation probability can be calculated in
a straight forward manner from equation~(\ref{eq:matrix}) in terms of
$\psi, \phi_m$ and $\delta^m_{31} = m^2_3 - m^2_1$.
\begin{equation}
P^m_{\mu e} = \frac{1}{2} \sin^2 2 \phi_m \sin^2 \left( \frac{1.27
\delta^m_{31} L}{E} \right), \label{eq:matP}
\end{equation}
where we have substituted the Super-Kamiokande best fit value $\sin^2 \psi
= 1/2$. As discussed above, the matter effects enhance the mixing angle
for neutrinos and the matter modified oscillation probability $P^m_{\mu
e}$ will be greater than the vacuum oscillation probability $P_{\mu e}$,
for a range of energies. For some neutrino energy, the
Mikheyev-Smirnov resonance condition \cite{miksmi}
\begin{equation}
\delta_{31} \cos 2 \phi = A \label{eq:rescon}
\end{equation}
can be satisfied and near this energy $\phi_m \sim \pi/4$. Unfortunately,
this does not lead to any dramatic increase in $P^m_{\mu e}$ because, near
the resonance, when $\sin^2 2 \phi_m$ is maximized, the matter dependent
mass-squared difference $\delta^m_{31}$ is minimized \cite{ajmm}.
In fact, near the
resonance, $P^m_{\mu e} \geq P_{\mu e}$. By differentiating
equation~(\ref{eq:matP}) with respect to $E$, we can calculate the energy
at which $P^m_{\mu e}$ is maximized. The highest energy at which this
occurs is just below the highest energy at which $P_{\mu e}$ is maximised,
that is the energy for which the phase $1.27 \delta_{31} L/E = \pi/2$.
At this energy, the $\phi_m$ is significantly higher than $\phi$ and
hence $P^m_{\mu e}$ will be measurably higher than $P_{\mu e}$.
Since the baseline of K2K is $3$ times smaller than that of MINOS, the
energy where the its phase becomes $\pi/2$ is smaller by a factor of
$3$ compared to similar energy for MINOS. Because the highest energy
maximum (at which the phase is $\pi/2$) occurs at different energies,
the increase in $\nu_\mu
\rightarrow \nu_e$ oscillation probability, due to the energy dependent
enhancement of the mixing angle, is different for the two experiments.
We find that, at their respective highest energy maxima,
$P^m_{\mu e} \simeq 1.1 P_{\mu e}$ for K2K
and $P^m_{\mu e} = 1.25 P_{\mu e}$ for MINOS. This is illustrated
in figure~(1) for $\phi \simeq 12.5^o$ (which is just below the CHOOZ
limit) and $\delta_{31} = 2 \times 10^{-3}$ eV$^2$ (which is the
best fit value for Super-Kamiokande atmospheric neutrino data).
This conclusion is
indenpendent of the value of $\delta_{31}$ and is illustrated in figures
(1)-(3), for different values of $\delta_{31}$ (and $\phi = 12.5^o$).
A similar conclusion was
obtained a recent paper, where it was demonstrated that the relation
between $P^m_{\mu e}$ and $P_{\mu e}$ is almost independent of $\phi$
\cite{lipari}.
We wish to emphasize that data from at least two different experiments
with different baselines is needed to state unambiguously, whether matter
effects are playing a role in neutrino oscillations.
Suppose we have data from only one experiment. We can obtain allowed
values of $\delta_{31}$ and $\phi$ by fitting either $P_{\mu e}$ or
$P^m_{\mu e}$ to this data. The two analyses will give
different values of mixing angle. Since
we don't apriori know the value of vacuum mixing angle, we can't say
which result is correct. However, as mentioned above, matter effects
lead to different enhancement of oscillation probability for K2K and
MINOS as shown in figure~(1). This difference can be exploited to
distinguish matter effects. The data from each experiment, K2K and
MINOS, should be analyzed twice, once using $P_{\mu e}$ as the input
and the second time with $P^m_{\mu e}$ as the input. In the first
analysis, the allowed value of the mixing angle from MINOS will be
significantly higher that from K2K, if the matter effects are
important. Matter effects can be taken to be established, if the
second analysis gives the same values of $\phi$ for both K2K and MINOS.
In conclusion, we find that matter effects enhance $\nu_\mu
\rightarrow \nu_e$ oscillations at long baseline experiments
K2K and MINOS. This enhancement can be large enough to be
observable. However, one must combine the data from both
experiments before making a definite statement about the
effect of the matter term.
Acknowledgement: We thank Sameer Murthy for his help in the preparation
of the figures.
|
\section{Introduction}
The Standard Model is exceedingly successful in describing leptons,
quarks and their interactions. Nevertheless, the Standard Model (SM)
is not considered as the ultimate theory since neither the fundamental
parameters, masses and couplings, nor the symmetry pattern are
predicted. These elements are merely built into the model. Likewise,
the spontaneous electroweak symmetry breaking is simply parametrized
by a single Higgs doublet field.
Even though many aspects of the Standard Model are experimentally
supported to a very high accuracy \cite{Hollik}, the embedding of the
model into a more general framework is to be expected. The argument
is closely related to the mechanism of the electroweak symmetry
breaking. If the Higgs boson is light, the Standard Model can
naturally be embedded in a grand unified theory. The large energy gap
between the low electroweak scale and the high grand unification scale
can be stabilized by supersymmetry. Supersymmetry \cite{susy} actually
provides the link between the experimentally explored interactions at
electroweak energy scales and physics at scales close to the Planck
scale where the gravity is important. If the Higgs boson is very
heavy, or if no fundamental Higgs boson exists, new strong
interactions between the massive electroweak gauge bosons are
predicted by unitarity at the TeV scale. With possibly many more new
layers of matter before the Planck scale is reached, no direct link
between electroweak and Planck scales in such a scenario is expected
at present. In either case, the next generation of accelerators which
will operate in the TeV energy range, can uncover the structure of
physics beyond the Standard Model.
Despite the lack of direct experimental evidence\footnote{The status
of low-energy supersummetry is discussed by S. Pokorski \cite{SP}.}
for supersymmetry (SUSY), the concept of symmetry between bosons and
fermions has so many attractive features that the supersymmetric
extension of the Standard Model is widely considered as a most natural
scenario. SUSY ensures the cancellation of quadratically divergent
quantum corrections from scalar and fermion loops and thus the Higgs
boson mass can be kept in the desired range of order $10^2$ GeV, which
is preferred by precision tests of the SM. The prediction of the
renormalized electroweak mixing angle $\sin^2 \theta_W$, based on the
spectrum of the Minimal Supersymmetric Standard Model (MSSM), is in
striking agreement with the measured value. Last but not least,
supersymmetry provides the opportunity to generate the electroweak
symmetry breaking radiatively.
In the next section the silent features of supersymmetric models are
briefly summarized. We will stress the importance of determining
experimentally all SUSY parameters in a model independent way. For
this purpose the $e^+e^-$ linear colliders \cite{LC} are indispensable
tools. It is illustrated in the next chapter where some of the
recently developed strategies to ``measure'' SUSY parameters in the
gaugino and sfermion sectors are discussed. For a discussion of the
SUSY Higgs sector we refer to \cite{PMZ}.
\section{Low-energy MSSM}
Supersymmetry predicts the quarks and leptons to have scalar partners,
called squarks and sleptons, the gauge bosons to have fermionic
partners, called gauginos. In the MSSM \cite{mssm} two Higgs doublets
with opposite hypercharges, and with their superpartners -- higgsinos
-- are required to give masses to the up and down type fermions and to
ensure anomaly cancellation. Thus the particle content of the MSSM is
given by
\vspace{-0.5cm}
\begin{table}[h]
\begin{center}
\begin{tabular}{|ccccc|cccc|cc|}
\hline \rule{0cm}{4mm}
$(l_a,\nu_a)_L$ & $l^c_{aL}$ &$(u_a,d_a)_L$ &$u^c_{aL}$&$d^c_{aL}$
&$\gamma$& $W^\pm$ &$Z^0$ &$g_{i}$&$H_1$& $H_2$
\\ \hline \rule{0cm}{4mm}
$({\tilde l}_a,{\tilde\nu}_{a})_L$ &${\tilde l}^c_{aL}$ &$
({\tilde u_a},{\tilde d_a})_L$ & ${\tilde u}^c_{aL}$&${\tilde d}^c_{aL}$
&${\tilde\gamma}$& ${\tilde W}^\pm$ &${\tilde Z}^0$ &
${\tilde g}_{i}$ &${\tilde H}_1$& ${\tilde H}_2$ \\
\hline
\end{tabular}
\end{center}
\end{table}
\vspace{-0.5cm}
\noindent where the first row lists the (left-handed) fermion fields
of one generation ($a=$1-3), the gauge fields (for gluons $i=$1-8) and
two Higgs doublets, and the second row -- their superpartners. The
higgsinos and electroweak gauginos mix; the mass eigenstates are
called charginos and neutralinos for electrically charged and neutral
states, respectively. The MSSM is defined by the superpotential
\begin{equation}
W=Y_{ab}^e \hat{L}_a\hat{H}_1 \hat{E}^c_b + Y_{ab}^d \hat{Q}_a
\hat{H}_1\hat{D}^c_b
+Y_{ab}^u\hat{Q}_a\hat{H}_2\hat{U}^c_b
- \mu \hat{H}_1 \hat{H}_2 \label{Rcons}
\end{equation}
where standard notation is used for the superfields of left-handed
doublets of (s)leptons ($\hat{L}_a$) and (s)quarks ($\hat{Q}_a$), the
right-handed singlets of charged (s)leptons ($\hat{E}_a$), up-
($\hat{U}_a$) and down-type (s)quarks ($\hat{D}_a$), and for the Higgs
doublet superfields which couple to the down ($\hat{H}_1$) and up
quarks ($\hat{H}_2$); the indices $a,b$ denote the generations and a
summation is understood, $Y^f_{ab}$ are Yukawa couplings and $\mu$ is
the Higgs mixing mass parameter. The $W$ respects a discrete
multiplicative symmetry under $R$-parity, defined as $
R_p=(-1)^{3B+L+2S} $, where $B$, $L$ and $S$ denote the baryon and
lepton number, and the spin of the particle. The $R_p$ conservation
implies that the lightest supersymmetric particle (LSP -- preferably
the lightest neutralino) is stable and superpartners can be produced
only in pairs in collisions and decays of particles.
If realized in Nature, supersymmetry must be broken at low energy
since no superpartners of ordinary particles have been observed so
far. It is technically achieved \cite{gri} by introducing the
soft--supersymmetry breaking \\
(i) gaugino mass terms for bino
$\tilde{B}$, wino $\tilde{W}^j$ $[j=$1--3] and gluino $\tilde{g}^a$
$[i=$1--8] \begin{eqnarray} {\textstyle \frac{1}{2}} M_1
\,\overline{\tilde{B}}\,\tilde{B} \ + \ {\textstyle \frac{1}{2}} M_2
\,\overline{\tilde{W}}^i \,\tilde{W}^i \ + \ {\textstyle \frac{1}{2}}
M_3 \,\overline{\tilde{g}}^i \,\tilde{g}^i \ , \end{eqnarray} (ii) trilinear
couplings (generation indices are understood) \begin{eqnarray} A^u H_2 \tilde{Q}
\tilde{u}^c + A^d H_1 \tilde{Q} \tilde{d}^c + A^l H_1 \tilde{L}
\tilde{l}^c - \mu B H_1 H_2 \end{eqnarray} (iii) and squark and slepton mass
terms \begin{eqnarray} m_{\tilde{Q}}^2 [\tilde{u}^*_L\tilde{u}_L
+\tilde{d}^*_L\tilde{d}_L] + m_{\tilde{u}}^2 \tilde{u}^*_R \tilde{u}_R
+ m_{\tilde{d}}^2 \tilde{d}^*_R \tilde{d}_R+
\ \cdots
\end{eqnarray}
where the ellipses stand for the soft mass terms for sleptons
and Higgs bosons. \\ \vspace*{-3mm}
The more than doubling the spectrum of states in the MSSM together
with the necessity of including the SUSY breaking terms gives rise to
a large number of parameters. Even with the $R$-parity conserving and
CP-invariant SUSY sector, which we will assume in what follows, in
total more than 100 new parameters are introduced! This number of
parameters can be reduced by additional physical assumptions. The
most radical reduction is achieved in the so called mSUGRA, by
embedding the low--energy supersymmetric theory into a grand unified
(SUSY-GUT) framework by requiring at the GUT scale $M_G$: \\ \vspace*{-3mm}
\noindent $(i)$ the unification of the U(1), SU(2) and SU(3) coupling
constants
\begin{equation} \alpha_3 (M_{\rm G})
= \alpha_2 (M_{\rm G}) = \alpha_1 (M_{\rm G}) =\alpha_G,
\end{equation}
$(ii)$ a common gaugino mass $m_{1/2}$. The gaugino masses $M_i$ at the
electroweak scale are then related through renormalization group
equations (RGEs) to the gauge couplings
\begin{eqnarray} M_i =
\frac{\alpha_i(M_Z)}{\alpha_G} m_{1/2} ,
\end{eqnarray}
$(iii)$ a universal trilinear coupling $A_G$
\begin{eqnarray} A_G = A^u (M_{\rm G}) = A^d (M_{\rm G}) =
A^l (M_{\rm G}) ,
\end{eqnarray}
$(iv)$ a universal scalar mass $m_0$
\begin{eqnarray}
m_0&=&m_{\tilde{Q}}=m_{\tilde{u}}=m_{\tilde{d}}= \cdots,
\end{eqnarray}
$(v)$ radiative breaking of the electroweak symmetry.\\ \vspace*{-3mm}
\noindent
The last requirement allows to solve for $B$ and $\mu$ (to within a
sign) once the values of the GUT parameters $m_{1/2}$, $m_0$, $A_G$ as
well as the ratio of the vacuum expectation values of the fields
$H_2^0$ and $H_1^0$, $\tan\beta=v_2/v_1$, are fixed. As a result, the
mSUGRA is fully specified by $m_{1/2}$, $m_0$, $A_G$, $\tan\beta$ and
sign($\mu$) -- the couplings, masses and mixings at the electroweak
scale are determined by the RGEs \cite{msugra}.
From the experimental point of view, however,
all low-energy parameters should be
measured independently of any theoretical assumptions. Therefore
the experimental program to search for and explore SUSY at present and
future colliders should include the following points:
\begin{itemize}
\item[(a)] discover supersymmetric particles and measure their quantum
numbers to prove that they are {\it the} superpartners of standard
particles,
\item[(b)]
determine the low-energy Lagrangian parameters,
\item[(c)] verify the relations among them in order to distinguish
between various SUSY models.
\end{itemize}
If SUSY is at work it will be a matter of days for the LC to discover
the kinematically accessible supersymmetric particles. Once they are
discovered, the priority will be to measure the low-energy SUSY
parameters independently of theoretical prejudices and then check
whether the correlations among parameters, if any, support a given
theoretical framework, like SUSY-GUT relations. A clear strategy is
needed to deal with so many a prori arbitrary parameters. One should
realize, that the low-energy parameters are of two distinct
categories. The first one includes all the gauge and Yukawa couplings
and the higgsino mass parameter $\mu$. They are related by exact
supersymmetry which is crucial for the cancellation of quadratic
divergencies. For example, at tree-level the $qqZ$,
$\tilde{q}\tilde{q}Z$ gauge and $q\tilde{q}\tilde{Z}$ Yukawa couplings
have to be equal. The relations among these parameters
(with calculable radiative corrections) have to be
confirmed experimentally; if not -- the supersymmetry is excluded.
The second category encompasses all soft supersymmetry breaking
parameters: Higgs, gaugino and sfermion masses and mixings, and
trilinear couplings. They are soft in the sense that they do not
reintroduce dangerous quadratic divergencies. Each one should be
measured independently by experiment to shed light
on the mechanism of supersymmetry breaking.
Particularly in this respect (points (b) and (c) above) the $e^+e^-$
linear colliders are invaluable. An intense activity during last
decade in Europe, the USA and Japan on physics at a linear $e^+e^-$
collider \cite{lcwork}
has convincingly demonstrated the advantages and benefits of
such a machine and its complementarity to the Large Hadron Collider
(LHC). Many studies have shown that the LHC can cover a mass range
for SUSY particles up to $\sim$ 2 TeV, in particular for squarks and
gluinos \cite{SusyLHC}. The problem however is that many different
sparticles will be accessed at once with the heavier ones cascading
into the lighter which will in turn cascade further leading to a
complicated picture. Simulations for the extraction of parameters
have been attempted for the LHC \cite{SusyLHC} and demonstrated that
some of them can be extracted with a good precision. However it must
be stressed that these checks were done with the assumption of an
underlying model, like mSUGRA and it has not been demonstrated so far
that the same can be achieved in a model independent way.
From the practical point of view it is very important that the energy
of the $e^+e^-$ machine can be optimized so that only very few
thresholds are crossed at a time. Another important feature is the
availability of beam polarization as well as a possibility of running
in $e^-e^-$ \cite{emem} or in $e\gamma$ and $\gamma\gamma$
\cite{gamma} modes. Making judicious choices of these features, the
confusing mixing of many final states, unavoidable at the LHC, with
the cascade decays might be avoided and analyses restricted to a
specific subset of processes performed. The measurements that can be
performed in the Higgs sector are discussed in the talk by P. Zerwas
\cite{PMZ}. Here I will discuss some methods of extracting SUSY
parameters from the gaugino (chargino/neutralino) and sfermion
sectors. In contrast to many earlier analyses \cite{other}, we will
not elaborate on global fits
but rather we will discuss attempts at
``measuring'' the fundamental parameters. Such attempts generically
involve two steps:
\begin{enumerate}
\item[$A$:] from the observed quantities: cross sections, asymmetries
etc. \\ $\Longrightarrow$ determine the physical parameters:
the masses, mixings and couplings of sparticles
\item[$B$:] from the physical parameters \\ $\Longrightarrow$ extract
the Lagrangian parameters: $M_i$, $\mu$, $\tan\beta$, $A^u$,
$m_{\tilde{Q}}$ etc.
\end{enumerate}
Each step can suffer from both experimental problems and theoretical
ambiguities. Concentrating first on the theoretical ones, recently
these two steps have been fully realized for the chargino sector
\cite{choi1} and the work on exploiting the neutralinos is in progress
\cite{choi3}. Similar strategies have been developed for sleptons and
squarks \cite{uli,Bartl}.
An alternative approach for the step $B$, based only on
the masses of some of the charginos and neutralinos, can be found in
\cite{KM}.
\section{Determining the Lagrangian parameters}
\subsection{Charginos: extracting $\tan\beta$, $M_2$ and $\mu$}
The spin--1/2 superpartners of the $W$ boson and charged Higgs boson,
$\tilde{W}^\pm$ and $\tilde{H}^\pm$, mix to form chargino mass
eigenstates $\tilde{\chi}^\pm_{1,2}$. Their masses
$m_{\tilde{\chi}_{1,2}^\pm}$ and the mixing angles $\phi_L,\phi_R$ are
determined by the elements of the chargino mass matrix in the
$(\tilde{W}^-,\tilde{H}^-)$ basis \cite{mssm}
\begin{eqnarray}
{\cal M}_C=\left(\begin{array}{cc}
M_2 & \sqrt{2}m_W c_\beta \\
\sqrt{2}m_W s_\beta & \mu
\end{array}\right)
\label{eq:mass matrix}
\end{eqnarray}
which is given in terms of fundamental parameters: $M_2$, $\mu$, and
$\tan\beta=v_2/v_1$; $s_\beta=\sin\beta$, $c_\beta=\cos\beta$.
As outlined above, we will discuss first how to
determine the chargino masses and mixing angles [step $A$] and then
the procedure of extracting $M_2$, $\mu$, and $\tan\beta=v_2/v_1$
[step $B$].
Charginos are produced in $e^+e^-$ collisions, either in diagonal or
in mixed pairs
\begin{eqnarray*}
e^+e^- \ \rightarrow \ \tilde{\chi}^+_i \ \tilde{\chi}^-_j
\end{eqnarray*}
With the second chargino $\tilde{\chi}_2^\pm$ expected to be
significantly heavier than the first one, at LEP2 or even in the first
phase of $e^+e^-$ linear colliders, the
chargino $\tilde{\chi}_1^\pm$ may be, for some time, the only chargino
state that can be studied experimentally in detail. Therefore, we
concentrate on the diagonal pair production of the lightest chargino
$\tilde{\chi}_1^+\tilde{\chi}_1^-$ in $e^+e^-$ collisions. Next,
assuming an upgrade in energy, we consider additional informations
available from $\tilde{\chi}^\pm_1 \ \tilde{\chi}^\mp_2$ and
$\tilde{\chi}^+_2 \ \tilde{\chi}^-_2$ production processes.
Two different matrices acting on the left-- and right--chiral
$(\tilde{W},\tilde{H})$ states are needed to diagonalize the
asymmetric mass matrix (\ref{eq:mass matrix}). The two (positive)
eigenvalues are given by
\begin{eqnarray}
m^2_{\tilde{\chi}^\pm_{1,2}}
=\frac{1}{2}\left[M^2_2+\mu^2+2m^2_W
\mp \Delta \, \right]
\end{eqnarray}
where
\begin{equation}
\Delta=\left[(M^2_2+\mu^2+2m^2_W)^2-4(M_2\mu-m^2_W\sin 2\beta)^2\right]^{1/2}
\end{equation}
The left-- and right--chiral components of the mass eigenstate
$\tilde{\chi}^-_1$ are related to the wino and higgsino components
in the following way,
\begin{eqnarray}
&&\tilde{\chi}^-_{1L}=\tilde{W}^-_L\cos\phi_L
+\tilde{H}^-_{1L}\sin\phi_L \nonumber\\
&&\tilde{\chi}^-_{1R}=\tilde{W}^-_R\cos\phi_R
+\tilde{H}^-_{2R}\sin\phi_R
\end{eqnarray}
with the rotation angles given by
\begin{eqnarray}
&&\cos 2\phi_L=-(M_2^2-\mu^2-2m^2_W\cos 2\beta)/\Delta
\nonumber\\
&&\sin 2\phi_L=-2\sqrt{2}m_W(M_2\cos\beta+\mu\sin\beta)/\Delta
\nonumber\\
&&\cos 2\phi_R=-(M_2^2-\mu^2+2m^2_W\cos 2\beta)/\Delta
\nonumber\\
&&\sin 2\phi_R=-2\sqrt{2}m_W(M_2\sin\beta+\mu\cos\beta)/\Delta
\label{mixing}
\end{eqnarray}
As usual, we take $\tan\beta$ positive, $M_2$ positive and $\mu$ of either
sign.
\vspace{-5mm}
\begin{figure}[htb]
\hspace{1cm}
\epsfig{figure=xrs.eps, height= 2in}
\caption{Total cross section for the chargino pair production for a
representative set of $M_2, \mu$: solid line for the gaugino case,
dashed line for the higgsino case, dot-dashed line for the mixed
case. In the left panel $m_{\tilde{\nu}}=200$ GeV (taken
from~[14]).}
\label{fig:xrs1}
\end{figure}
Light charginos are produced in pairs in $e^+ e^-$ collisions through
s-channel $\gamma$ and $Z$, and t-channel sneutrino exchange. The
production cross section will thus depend on the chargino mass
$m_{\tilde{\chi}^\pm_1}$, the sneutrino mass $m_{\tilde{\nu}}$ and the
mixing angles, eq.(\ref{mixing}), which determine the couplings of the
chargino states to the $Z$ and the sneutrino. The unpolarized total
cross section for $m_{\tilde{\chi}^\pm_1}=95$ GeV is illustrated in
fig.~\ref{fig:xrs1} for representative cases of dominant higgsino,
gaugino or mixed content of the lightest chargino state. The sharp
rise near threshold should allow a precise determination of the
chargino mass. The sensitivity to the sneutrino mass with the typical
destructive interference in the gaugino and mixed cases necessitates
the knowledge of this parameter \cite{gudi}.
Charginos are not stable and each will decay directly to a pair of
matter fermions (leptons or quarks) and the (stable) lightest
neutralino $\tilde{\chi}_1^0$. The decay proceeds through the
exchange of a $W$ boson (charged Higgs exchange is suppressed for
light fermions) or scalar partners of leptons or quarks. The decay
matrix elements will depend on further parameters like the scalar
masses and couplings to the neutralino. In addition, the presence of
two invisible neutralinos in the final state of the process, $ e^+ e^-
\to \tilde{\chi}_1^+ \tilde{\chi}_1^- \to \tilde{\chi}_1^0
\tilde{\chi}_1^0 (f_1 \bar{f}_2) (\bar{f}_3 f_4)$, makes it impossible
to measure directly the chargino production angle $\Theta$ in the
laboratory frame. Integrating over this angle and also over the
invariant masses of the fermionic systems $(f_1 \bar{f}_2)$ and
$(\bar{f}_3 f_4)$, one can write the differential cross section in the
following form:
\begin{equation}
\frac{d^4 \sigma( e^+ e^- \to \tilde{\chi}_1^0
\tilde{\chi}_1^0 (f_1 \bar{f}_2)
(\bar{f}_3 f_4)) }{d\cos \theta^* d\phi^* d\cos \bar{\theta}^*
d\bar{\phi}^*} = \frac{\alpha^2 \beta}{124 \pi s} {\cal B}\, \bar{\cal B}
\, \Sigma(\theta^*, \phi^*, \bar{\theta}^*, \bar{\phi}^*)
\end{equation}
where $\alpha$ is the fine structure constant, $\beta$ the velocity of
the chargino in the c.m. frame. For the $\tilde{\chi}^-_1$ decay we
have ${\cal B}=Br(\tilde{\chi}_1^- \to \tilde{\chi}_1^0 f_1
\bar{f}_2)$, $\theta^*$ is the polar angle of the $f_1 \bar{f}_2$
system in the $\tilde{\chi}^-_1$ rest frame with respect to the
chargino's flight direction in the lab frame, and $\phi^*$ is the
azimuthal angle with respect to the production plane; quantities with
a bar refer to the $\tilde{\chi}^+_1$ decay. The differential cross
section $\Sigma(\theta^*, \phi^*, \bar{\theta}^*, \bar{\phi}^*)$ is
expressed in terms of sixteen independent angular combinations of
helicity production amplitudes
\begin{equation}
\begin{array}{rcl} \Sigma & = & \Sigma_{unpol} + \kappa \cos
\theta^* {\cal P} + \bar{\kappa}\cos \bar{\theta}^* \bar{{\cal P}} +
\cos \theta^* \cos \bar{\theta^*} \kappa \bar{\kappa} {\cal Q} \\[2mm]
&& + \sin \theta^* \sin \bar{\theta^*} \cos (\phi^* + \bar{\phi^*})
\kappa \bar{\kappa} {\cal Y} + \dots
\end{array} \label{Sigma}
\end{equation}
Out of the sixteen terms, corresponding to the unpolarized,
$2 \times 3$ polarization components and $3 \times 3$ spin--spin
correlations in the production process, only 7 are independent
(neglecting small effects from the $Z$-boson width
and loop corrections) and $\kappa=-\bar{\kappa}$ in the
CP-invariant theory.
The polarization component ${\cal P}$ coming
from the $\tilde{\chi}_1^-$ system, for example, reads
\begin{equation}
{\cal P}={\textstyle \frac{1}{4}}\int{\rm d}\cos\Theta\sum_{\sigma=\pm}
[|A_{\sigma;++}|^2+|A_{\sigma;+-}|^2
-|A_{\sigma;-+}|^2-|A_{\sigma;--}|^2 ]
\end{equation}
where $2\pi\alpha A_{\sigma;\lambda\lambda'}$ is the helicity
amplitude with $\sigma;\lambda\lambda'$ denoting the helicities of the
electron and $\tilde{\chi}^-_1\tilde{\chi}^+_1$ pair, respectively.
All the complicted dependence on the chargino decay dynamics
(neutralino and sfermion masses and their couplings) is contained in
the spin analysis-powers $\kappa$ and $\bar{\kappa}$.
\vspace{2cm}
\begin{figure}[htb]
\centerline{ \epsfig{figure=cslrsum.ps, height=4cm}}
\caption{Contours for the ``measured values'' of the total cross
section (solid line), ${\cal P}^2/{\cal Q}$, and ${\cal P}^2/{\cal
Y}$ (dot-dashed line) for $m_{\tilde{\chi}_1^{\pm}}=95 GeV$
[$m_{\tilde{\nu}} = 250$ GeV]. Superimposed are contour lines
(solid, almost vertical lines) for the ``measured'' LR asymmetry. }
\label{fig:cont}
\end{figure}
The crucial observation of \cite{choi1} is that all explicitly written
terms in eq.~(\ref{Sigma}) can be extracted and three
$\kappa$-independent physical observables, $\Sigma_{unpol}, {\cal
P}^2/{\cal Q}$ and ${\cal P}^2/{\cal Y}$, constructed. Indeed, it is
possible by means of kinematical projections, since $\cos \theta^*, \cos
\bar{\theta^*}$ and $\sin \theta^* \sin \bar{\theta^*} \cos( \phi^* +
\bar{\phi^*})$ are fully determined by the measurable parameters $E,
|\vec{p}|$ (the energy and momentum of each of the decay systems $f_i
\bar{f_j}$ in the laboratory frame) and the chargino mass. As a
result, the chargino properties can be determined {\it independently}
of the other sectors of the model.\footnote{Actually to determine the
kinematical variables, $\cos \theta^*$ etc., the knowledge of
$m_{\tilde{\chi}^0_1}$ is also needed, which can be extracted from the
energy distributions of final state particles, see later. However, it
must be stressed that the above procedure does not depend on the
details of decay dynamics nor on the structure of (potentially more
complex) neutralino and sfermion sectors.} The measurements of the
cross section and either of the ratios ${\cal P}^2/{\cal Q}$ or ${\cal
P}^2/{\cal Y}$ can be interpreted as contour lines in the plane
$\{\cos 2\phi_L,\cos 2\phi_R\}$ which intersect with large angles so
that a high precision in the resolution can be achieved. A
representative example for the determination of $\cos 2\phi_L$ and
$\cos 2\phi_R$ is shown in fig.~\ref{fig:cont}. The mass of the light
chargino is set to $m_{\tilde{\chi}^\pm_1}=95$ GeV, and the
``measured'' cross section, ${\cal P}^2/{\cal Q}$ and ${\cal
P}^2/{\cal Y}$ at $\sqrt{s}=500$ GeV are taken to be
\begin{equation}
\sigma(e^+e^-\rightarrow\tilde{\chi}^+_1\tilde{\chi}^-_1)=0.37\ \
{\rm pb},\quad
{\cal P}^2/{\cal Q}=-0.24,\quad
{\cal P}^2/{\cal Y}= -3.66
\label{eq:measured}
\end{equation}
in the left panel of fig.~\ref{fig:cont}. The three contour lines
meet at a single point $\{\cos 2\phi_L,\cos 2\phi_R\}
=\{-0.8,-0.5\}$. The sneutrino mass is set to $m_{\tilde{\nu}}=250$
GeV. Note that the $m_{\tilde{\nu}}$ can be determined together with
the mixing angles by requiring a consistent solution from the
``measured quantities'' $\sigma$, ${\cal P}^2/{\cal Q}$ and ${\cal
P}^2/{\cal Y}$ at several values of incoming energy, as exemplified in
both panels of fig.~\ref{fig:cont} for $\sqrt{s}=500$ and 300 GeV.
If polarized beams are available, the left-right asymmetry $A_{LR}$
can provide an alternative way to extract the mixing angles (or serve
as a consistency check). This is also demonstrated in
fig.~\ref{fig:cont}, where contour lines for the ``measured'' values of
$A_{LR}$ are also shown. Moreover, with right-handed electron beams
one can turn off the sneutrino exchange in the production process and
since at high energy the $\gamma$ and $Z$ ``demix'' back to the
$W^0_3$ and $B^0$ gauge bosons, only the higgsino component of the
chargino is selected. Thus the polarization alone will give us the
composition of charginos. In short, the step $A$ can be fully
realized for the lightest charginos.
Let us now discuss the step $B$ and describe briefly how to determine the
Lagrangian parameters $M_2, \mu$ and $\tan\beta$ from
$m_{\tilde{\chi}^\pm_1}$,
$\cos 2\phi_L$ and $\cos 2\phi_R$. It is most
transparently achieved by introducing the two triangular quantities
\begin{eqnarray}
p=\cot(\phi_R-\phi_L)\ \ {\rm and}\ \ q=\cot(\phi_R+\phi_L)
\end{eqnarray}
They are expressed in terms of the measured values $\cos 2\phi_L$ and
$\cos 2\phi_R$ up to a discrete ambiguity due to undetermined signs
of $\sin 2\phi_L$ and $\sin 2\phi_R$
\begin{eqnarray}
p^2+q^2 &=&\frac{2(\sin^2 2\phi_L+\sin^2 2\phi_R)}{(\cos 2\phi_L-\cos 2
\phi_R)^2}
\nonumber\\
pq &=&\frac{\cos 2\phi_L+\cos 2\phi_R}{\cos 2\phi_L-\cos 2\phi_R}
\nonumber\\
p^2-q^2 &=&\frac{4\sin 2\phi_L\sin 2\phi_R}{(\cos 2\phi_L-\cos 2\phi_R)^2}
\end{eqnarray}
Solving then eqs.~(\ref{mixing}) for $\tan\beta$ one finds at most
two possible solutions, and using
\begin{eqnarray}
M_2&=&{m_W}[(p+q)s_\beta-(p-q)c_\beta]/\sqrt{2}
\nonumber\\
\mu&=&{m_W}[(p-q)s_\beta-(p+q)c_\beta]/\sqrt{2}
\label{eq:M2_mu}
\end{eqnarray}
we arrive at $\tan\beta$, $M_2$ and $\mu$ up to a two-fold
ambiguity. For example, taking the ``measured values'' from
eq.~(\ref{eq:measured}), the following results are found in
\cite{choi1}
\begin{eqnarray}
[\tan\beta; M_2,\mu] =
\left\{\begin{array}{l}
[1.06; \; 83{\rm GeV}, \; -59{\rm GeV}] \\
{ }\\ {}
[3.33; \; 248{\rm GeV}, \; 123{\rm GeV}]
\end{array}\right.
\end{eqnarray}
To summarize, from the lightest chargino pair production, the
measurements of the total production cross section and either the
angular correlations among the chargino decay products (${\cal
P}^2/{\cal Q}$, ${\cal P}^2/{\cal Y}$) or the LR asymmetry, the step
$A$ can be realized and the physical parameters
$m_{\tilde{\chi}^\pm_1}$, $\cos 2\phi_L$ and $\cos 2\phi_R$ determined
unambiguously. Then the fundamental parameters $\tan\beta$,
$M_2$ and $\mu$ are extracted (step $B$) up to a two-fold ambiguity.
If the collider energy is sufficient to produce the two chargino
states in pairs, the above ambiguity can be removed \cite{choi2}.
The new ingredient in this case is the knowledge of the
heavier chargino mass. Like for the ligter one,
$m_{\tilde{\chi}^\pm_{2}}$ can be determined very precisely from the
sharp rise of the production cross sections
$\sigma(e^+e^-\rightarrow\tilde{\chi}^-_i\tilde{\chi}^+_j)$.
Then the value of $\tan\beta$ is uniquely determined in terms of the
mass difference of two chargino states, $\Delta =
m^2_{\tilde{\chi}^\pm_2}-m^2_{\tilde{\chi}^\pm_1}$, and two mixing
angles as follows
\begin{eqnarray}
\tan\beta=\sqrt{\frac{4m^2_W
+\Delta \,
(\cos 2\phi_R-\cos 2\phi_L)}{4m^2_W
-\Delta \,
(\cos 2\phi_R-\cos 2\phi_L)}}
\label{eq:tanb}
\end{eqnarray}
Using the convention $M_2>0$,
the gaugino mass parameter $M_2$ and the modulus of the higgsino
mass parameter are given by
\begin{eqnarray}
M_2&=&\frac{1}{2}
\sqrt{2(m^2_{\tilde{\chi}^\pm_2}+m^2_{\tilde{\chi}^\pm_1}-2m^2_W)
-\Delta \,
(\cos 2\phi_R+\cos 2\phi_L)}\nonumber\\
|\mu|&=&\frac{1}{2}
\sqrt{2(m^2_{\tilde{\chi}^\pm_2}+m^2_{\tilde{\chi}^\pm_1}-2m^2_W)
+\Delta \,
(\cos 2\phi_R+\cos 2\phi_L)}
\label{eq:M2mu}
\end{eqnarray}
The sign of $\mu$ is then determined by the sign of the following
expression
\begin{eqnarray}
\mbox{sign}(\mu)= \mbox{sign}[ \Delta^2
-(M^2_2-\mu^2)^2-4m^2_W(M^2_2+\mu^2)
-4m^4_W\cos^2 2\beta]
\end{eqnarray}
Before leaving the chargino sector, let us note that from the energy
distribution of the final particles in the decay of the charginos
$\tilde{\chi}^\pm_1$, the mass of the lightest neutralino
$\tilde{\chi}^0_1$ can be measured \cite{uli}.
This, as we will see in the next subsection,
allows us to derive the parameter $M_1$
in the CP--invariant theories so that
the neutralino mass matrix, too, can be reconstructed
in a model-independent way.
\subsection{And Neutralinos: extracting also $M_1$}
The spin--1/2 superpartners of the neutral electroweak gauge bosons
and neutral Higgs bosons mix to form four neutralino mass eigenstates
$\tilde{\chi}^0_{1,2,3,4}$. Their masses $m_{\tilde{\chi}_{i}^0}$ and
the mixing angles are determined by the elements of the neutralino
mass matrix given by ($s_W=\sin\theta_W$, $c_W=\cos\theta_W$) \cite{mssm}
\begin{eqnarray}
{\cal M}_N = \left[ \begin{array}{cccc}
M_1 & 0 & -m_Z s_W c_\beta & m_Z s_W s_\beta \\
0 & M_2 & m_Z c_W c_\beta & -m_Z c_W s_\beta \\
-m_Z s_W c_\beta & m_Z c_W c_\beta & 0 & -\mu \\
m_Z s_W s_\beta & -m_Z c_W s_\beta & -\mu & 0
\end{array} \right]
\end{eqnarray}
Since ${\cal M}_N$ is symmetric,
an orthogonal matrix ${\cal N}$ can be constructed that transforms
${\cal M}_N$ to a (positive) diagonal matrix. This
mathematical problem can be solved analytically \cite{esa}.
Due to the large ensemble of four neutralinos, however, the analysis is much
more complex than in the chargino case. In particular, the step $B$,
$i.e.$ the analytical reconstruction of the fundamental SUSY
parameters, is more complicated although,
after measuring the parameters $M_2, \mu$ and $\tan \beta$ from the
chargino production, the only additional parameter in the neutralino
mass matrix is $M_1$.
Neutralinos are produced in $e^+e^-$ collisions either in diagonal or
non-diagonal pairs. The lightest neutralino $\tilde{\chi}^0_1$ is
generally expected to be the lightest SUSY particle (LSP) and
therefore stable in the $R$-parity preserving model. As a result, the
production of the lightest neutralino pairs is difficult to identify
and exploit experimentally. Therefore we consider production
processes where at least one of the neutralinos is not an LSP, for
example $\tilde{\chi}^0_1\tilde{\chi}^0_2$ or
$\tilde{\chi}^0_2\tilde{\chi}^0_2$. These processes are generated by
the $s$-channel $Z$ exchange and the $t$- and $u$-channel selectron
$\tilde{e}_{L,R}$ exchanges. The transition matrix elements will then
depend not only on the neutralino properties but on the selectron
masses as well. The heavier neutralino $\tilde{\chi}^0_2$ will decay
into the LSP and a fermion pair, leptons or quarks,
$\tilde{\chi}^0_2\rightarrow \tilde{\chi}^0_1 f\bar{f}$, throught the
exchange of a $Z$ boson or scalar partners of the fermion (the neutral
Higgs boson exchange is negligible for light fermions). The decay
products will serve as a signature of the production process and from
the fast rise of the cross sections the masses $m_{\tilde{\chi}^0_i}$
can be measured precisely.
Additional informations can be obtained by analysing the angular
correlations among the decay products, like in the chargino sector. In
the case of $\tilde{\chi}^0_2\tilde{\chi}^0_2$, the method developed
for the chargino case can be applied directly. One can attempt to
separate the production from the decay processes and determine the
$Z\tilde{\chi}_i^0\tilde{\chi}_j^0$ and $e\tilde e \tilde{\chi}_i^0$ couplings
(expressed as known combinations of the mixing matrix elements ${\cal
N}_{ij}$ \cite{mssm}). Such a separation is interesting for the hadronic
or $\mu^+\mu^-$ decay modes of $\tilde{\chi}^0_2$ ($\tilde{\chi}^0_2
\rightarrow \tilde{\chi}^0_1 q\bar{q}$ or $\tilde{\chi}^0_1 \mu^+\mu^- $)
because independent
information on the neutralino couplings to electron-selectron and
quark-squark or $\mu^\pm\tilde{\mu}^\mp$ from the production and decay
processes, respectively,
can be inferred. For the $\tilde{\chi}^0_2
\rightarrow \tilde{\chi}^0_1 e^+e^-$ the production/decay separation
might be useful only from the point of view of consistency
checks. In the $\tilde{\chi}^0_1\tilde{\chi}^0_2$ production case, only
one neutralino decays and such a separation is not possible due to a
limited number of measurable kinematical variables.
Nevertheless some information on couplings can be extracted. The
prescription for the step $A$ in the neutralino sector still awaits
a detailed analysis \cite{choi3}.
However, the knowledge of $m_{\tilde{\chi}^0_1}$ in addition to the
measurements performed in the chargino sector is sufficient to
pinpoint the value of $M_1$, the only new parameter. This particular
problem has recently been considered in \cite{KM}, where the emphasis
has been put on the step $B$, namely to what extent
the reconstruction of the
Lagrangian parameters through
a controllable analytical procedure, including all possible ambiguities,
is possible if three of the chargino and neutralino
masses and $\tan \beta$ were known. Two cases have been analysed:
\begin{itemize}
\item[$S_1$:] the two charginos and one neutralino masses are input,
\item[$S_2$:] one chargino and two neutralino masses are input.
\end{itemize}
In case $S_1$, a closed analytical procedure to determine $M_1$
has been found. The crucial observation is to use the four independent
linear combinations of the entries of ${\cal M}_N$ which are invariant
under similarity transformations, and thus relate them simply to the
four eigenvalues of ${\cal M}_N$. As a result, any of the neutralino
masses taken as input, for example $\tilde{\chi}^0_1$, in addition to
$\mu$, $M_2$ and $\tan\beta$ allows the set of these
consistency relations to be
solved for the other three neutralino masses. Then the $M_1$ parameter
is determined as
\begin{equation}
M_1=
-\frac{P_{i j}^2 +
P_{i j} (\mu^2 + m_Z^2 + M_2 S_{i j} - S_{i j}^2)
+ \mu m_Z^2 M_2 s_W^2 \sin 2 \beta}
{P_{i j} (S_{i j} -M_2) + \mu ( c_W^2 m_Z^2 \sin 2 \beta -\mu M_2 )}
\end{equation}
where
\begin{eqnarray}
S_{i j} \equiv \tilde{m}_{i} + \tilde{m}_{j}, \qquad
P_{i j} \equiv \tilde{m}_{i} \tilde{m}_{j} \nonumber
\end{eqnarray}
$i\neq j$, and $\tilde{m}_i=\epsilon_i m_{\tilde{\chi}^0_i}$ (the mass
parameters can be negative, for the details we refer to \cite{KM}).
As an example of the numerical result of such a procedure, the sensitivity to a chargino mass with the other chargino and the neutralino masses fixed
is shown in fig.~\ref{fig:km}.
\begin{figure}
\center
\epsfig{figure=km1.eps, height= 2.5in}
\caption{$\mu, M_1$ and $M_2$ (with the ``higgsino-like'' convention
$|\mu| \leq M_2$) as functions of $m_{\tilde{\chi}^+_2}$ for fixed
$m_{\tilde{\chi}^+_1} = 400$ GeV, $m_{\tilde{\chi}^0_2} = 50$ GeV,
and $\tan\beta = 2$ (taken from~[18]). }
\label{fig:km}
\end{figure}
In case $S_2$, the above consistency relations can be
reformulated in terms of two quadratic equatiuons for $M_2$ and $M_1$
at a given value of $\mu$ (and $\tan\beta$). Without any additional
theoretical input, a numerical (iterative) procedure is used to obtain
at most four distinct solutions for $\mu$, $M_1$ and $M_2$ for a given
set of ${\tilde{\chi}^\pm_1}$ and two neutralino masses.
\subsection{Sfermions: extracting $m_{\tilde{f}_L}$, $m_{\tilde{f}_R}$
and $A^f$}
For each fermion chirality $f_{L,R}$ supersymmetry predicts a
corresponding sfermion $\tilde{f}_{L,R}$. Since SUSY is broken, the
chiral left and right sfermions $\tilde{f}_L$ and $\tilde{f}_R$ may
aquire different mass terms and they can mix. The mass eigenstates
and mixing are determined by the mass matrices (for a given sfermion
flavor $\tilde{f}$)
\begin{eqnarray}
M^2_{\tilde{f}} =
\left[ \begin{array}{cc} m_{\tilde{f}_L}^2 + m_f^2 & m_f (A^f - \mu r_f)
\\ m_f (A^f - \mu r_f) & m_{\tilde{f}_R}^2 + m_f^2 \end{array} \right]
\label{sfmass}
\end{eqnarray}
with
\begin{eqnarray}
&&\hspace{-8mm} m^2_{\tilde f_L} =
m^2_{\tilde Q} + m_Z^2 \cos 2\beta \,(T^3_f - e_f \sin^2\theta_W) ,
\label{eq:msfl}\\
&&\hspace{-8mm} m^2_{\tilde f_R} =
m^2_{\tilde F'} + e_f m_Z^2 \cos 2\beta \sin^2\theta_W ,
\label{eq:msfr}
\end{eqnarray}
where $e_f$ and $T^3_f$ are the charge and the third component
of the weak isospin of the sfermion $\tilde f$,
$m_f$ is the mass of the corresponding fermion,
$m_{\tilde F'} = m_{\tilde u}$, $m_{\tilde u}$ for $\tilde f_R = \tilde u_R$,
$ \tilde d_R$, respectively, and
$r_{f} = 1/\tan\beta$ for up- and $r_{f} = \tan\beta$ for
down-type sfermions.
The matrices (\ref{sfmass}) are diagonalized by orthogonal
transformations with mixing angles
$\theta_f$ defined by
\begin{eqnarray}
\sin 2\theta_f = \frac{2 m_f (A^f -\mu r_f)} { m_{\tilde{f}_1}^2
-m_{\tilde{f}_2}^2 } \ \ , \ \
\cos 2\theta_f = \frac{m_{\tilde{f}_L}^2 -m_{\tilde{f}_R}^2}
{ m_{\tilde{f}_1}^2 -m_{\tilde{f}_2}^2 }
\end{eqnarray}
and the masses of the sfermion eigenstates are given by
\begin{eqnarray}
m_{\tilde{f}_{1,2}}^2 = m_f^2 + \frac{1}{2} \left[
m_{ \tilde{f}_L}^2 + m_{\tilde{f}_R}^2 \mp \sqrt{
(m_{\tilde{f}_L}^2 - m_{\tilde{f}_R}^2)^2 + 4m_f^2 (A^f -\mu r_f)^2 }
\right].
\end{eqnarray}
Since the mixing term is of order $m_f$, it can be substantial
only for the third generation sfermions (for sbottom and stau if
$\tan\beta$ is large), with an important consequence of lowering the
mass of the lighter eigenstate $m_{\tilde{f}_{1}}$. As a result, the
lighter stop ${\tilde{t}_{1}}$ is expected to be the lightest scalar
fermion.
Sfermions are produced in pairs in $e^+e^-$ collisions
\begin{equation}
e^+e^- \rightarrow \tilde f_{i} \bar{\tilde f}_j
\end{equation}
through $s$-channel $\gamma$ and $Z$ exchange; only the selectron
production receives an additional $t$-channel neutralino exchange
contribution. Since the gauge boson couplings respect chirality, the
nondiagonal $\tilde f_{1} \bar{\tilde f}_2$ production can occur only
for nontrivial mixing.
It is important first to verify experimentally the chiral nature of
produced sfermions. This can be easily done at $e^+e^-$ colliders by
using polarized beams (not available at LHC) or by reconstructing the
polarization of the final state fermions from sfermion decays (too
difficult in hadron collisions). As an example, consider the pair
production of right-handed staus which most probably will decay into
the LSP neutralinos and $\tau$'s. The signature is the same as that of
$W$ pair production with the $W$'s decaying into $\tau\nu_\tau$.
However, at high-energy the $Z$ and $\gamma$ ``demix'' back to the
$W_3^0$ and $B^0$ (hypercharge). Since the former does not couple to
right-handed states, only the hypercharge boson is exchanged in right
sfermion production. Therefore the background $W$ pair production can
be suppressed by choosing right-handed electrons. Moreover, as a
result of hypercharge assignements $Y(e_L)=-1$ and $Y(e_R)=-2$, the
signal cross section with right-handed $e^-$ beams will be by factor 4
larger than with the left-handed $e^-$ beams. The beam polarization
therefore is a very powerful tool: allows us not only to tag the
nature of the stau (right-handed) independently of its decays and
increase the signal cross section, but also suppress the
background. All these has been checked by the full simulation of the
Japanese group \cite{JapanSusy}. In addition, reconstruction of the
$\tau$ polarization in the decay process $\ti \tau\rightarrow \tilde{\chi}^0_1
\tau$ will play an important role in exploring the Yukawa couplings.
The $\ti \tau\tau\tilde{\chi}^0_1$ coupling depends on the neutralino
composition. The interaction involving gaugino component ($\tilde{B}$
or $\tilde{W}$) is proportional to gauge couplings and is chiral
conserving, whereas the interaction involving higgsino component
($\tilde{H}_{1,2}$) is proportional to $\tau$ Yukawa coupling
$Y_{\tau}\sim m_{\tau}/\cos\beta$ and chiral flipping. Thus the
polarization of $\tau$ lepton from $\tilde \tau$ decays depends on the
ratio of the chirality flipping and chirality concerving interactions,
and consequently on $\tan\beta$. For a detailed discussion of stau
production we refer to \cite{Nojiri}.
Once the sfermion production has been optimized, one can either infer
the sfermion mass from a threshold scan (which is independent of the
decay) or (as in chargino case) the measurement of the fermion energy
spectrum will give both the $m_{\tilde f}$ and the LSP mass. A combined
fit for a low luminosity option of 10 fb$^{-1}$ and 85\% polarization
of the electron beam shows that a precision of order a few percent for
sfermion masses can easily be obtained \cite{uli}.
A case study of $e^+ e^- \to {\ti t}_1 \bar{\ti t}_1$
with the aim of determining the SUSY parameters has been performed by
the Vienna group \cite{Bartl} at $\sqrt s =$~500~GeV and
${\cal L}=50$ fb$^{-1}$.
The input $m_{{\ti t}_1} =$~180~GeV and
left--right stop mixing angle $|\cos\tht|$ = 0.57
corresponds to
the minimum of the cross section. The cross sections at tree level for
these parameters are $\sigma_L =$ 48.6~f\/b and $\sigma_R =$ 46.1~f\/b
for 90\% left-- and right--polarized $e^-$~beam,
respectively. Based on detailed studies
the experimental errors on these cross sections are
estimated to be $\Delta \sigma_L =
\pm$~6~f\/b and $\Delta \sigma_R = \pm$~4.9~f\/b. Figure~\ref{sab} shows the
resulting error bands and the corresponding error ellipse in the
$m_{{\ti t}_1}$--$\cos\tht$ plane. The experimental accuracy for the stop
mass and mixing angle are $m_{{\ti t}_1} = 180 \pm 7$~GeV, $|\cos\tht| =
0.57 \pm 0.06$.
\begin{figure}
\center
\epsfig{figure=figuerr.eps, height= 2.5in}
\caption{Error bands (dashed) and the corresponding error ellipse as a
function of $m_{{\ti t}_1}$ and $|\cos\theta_{{\ti t}}|$ for the tree--level
cross sections of $e^+ e^- \to {\ti t}_1 {\ti t}_1$ at $E_{cm} =
500$~GeV with 90\% left-- and right--polarized electron beam (taken
from~[17]). }
\label{sab}
\end{figure}
Additional experimental input is needed, however, to determine the
fundamental parameters. The Vienna group decided to exploit the
sbottom system. Assuming that $\tan\beta$ is low and the
${\ti b}_L$--${\ti b}_R$ mixing can be neglected, $i.e.$ $\cos\theta_{\ti b} = 1$,
and taking ${\ti b}_1 = {\ti b}_L=200$ GeV, ${\ti b}_2 =
{\ti b}_R=220$ GeV, the cross sections and the expected
experimental errors are $\sigma_L(e^+ e^- \to {\ti b}_1 \bar{\ti b}_1) = 61.1
\pm 6.4$~f\/b, $\sigma_R(e^+ e^- \to {\ti b}_2 \bar{\ti b}_2) = 6 \pm
2.6$~f\/b for the 90\% left-- and right--polarized $e^-$ beams.
The resulting experimental errors are $m_{{\ti b}_1} = 200 \pm 4$~GeV,
$m_{{\ti b}_2} = 220 \pm 10$~GeV.
With these results
the mass of the
heavier stop can be calculated and is found to be
$m_{{\ti t}_2} = 289 \pm 15$~GeV.
Verifying this prediction experimentally will test the MSSM.
To complete the step $B$, $\mu = - 200$~GeV, $\tan\beta = 2$ and
$m_t = 175$~GeV
have been taken assuming that $\mu$ and $\tan\beta$ are known
from other experiments (from chargino sector, for example).
The soft-supersymmetry breaking parameters of the stop and sbottom systems
can then be determined up to a two-fold ambiguity:
$m_{\tilde Q} = 195 \pm 4$~GeV, $m_{\tilde u} = 138 \pm 26$~GeV,
$m_{\tilde d} = 219 \pm 10$~GeV, $A^t = -236 \pm 38$~GeV
if $\cos\tht > 0$,
and $A^t = 36 \pm 38$~GeV if $\cos\tht < 0$.
\section{Conclusions}
In this talk I have tried to illustrate the discovery power and
precision tools developed to explore supersymmetry at future $e^+e^-$
linear colliders. The LC is an excellent machine for supersymmetry
because a systematic, model-independent determination of the
supersymmetry parameters is possible within a discovery reach that is
limited by the available center-of-mass energy. Although we only
considered real-valued parameters, some of the material presented here
goes through unaltered if phases are allowed \cite{choi1,choi2} even
though extra information will still be needed to determine those
phases.
It should be stressed that the strategies presented here are just at
the theoretical level. A more realistic simulation of the experimental
measurements of physical observables and related errors is still
needed to assess fully the physics potential of LC.
Nevertheless, if
the LC and detectors are built and work as expected, I have no doubt
that the actual measurements will be better than anything I have
presented here -- provided supersymmetry is discovered! After all,
nobody beats experimentalists with real data.
\section*{Acknowledgements}
I would like to thank Gudrid Moortgat-Pick, Stefan Pokorski and Peter
Zerwas for many valuable discussions. This work has been partially
supported by the KBN grant 2 P03B 052 16.
|
\section{Introduction}
A fundamental notion of quantum mechanics is
complementarity which expresses the fact that any
quantum system has at least two properties that cannot
simultaneously be known. One of these
complementary property pairs, and perhaps the
historically most important, is the wave-particle
duality. A quantum system has both particle-like and
wave-like properties. However, observation of one
property precludes the observation of the other.
Recently several quantitative expressions of this
specific duality were derived
\cite{Wootters,Greenberger,Mandel,Walther,Raymer,Kwiat1,Tan,Kwiat2,Jaeger,Englert,Bjork},
Some of these expressions have been experimentally
confirmed \cite{Herzog,Rempe}. In this paper we will
follow Englert's \cite{Englert} definitions most
closely. In general, inequalities quantifying the
wave-particle duality are posed in the context of
interferometry. In any single particle interferometer
it is meaningful both to ask which of the
interferometer paths the particle took, and to record
the visibility of a large number of identically
prepared systems. In this paper we shall discuss the
notions of which path and visibility in a general
framework as to encompass {\em any} system defined in
a two-dimensional Hilbert-space.
Let us begin by establishing some notation. Assume that
we would like to estimate which of two paths, call them
$+$ and $-$, a particle took. The only information we
have are known probabilities $w_+$ and $w_- = 1-w_+$
for the two events. The Maximum Likelihood
(ML) estimation strategy (which is one of many possible
strategies) dictates that we should, for each and every
event, guess that the particle took the most likely
path. The strategy will maximize the likelihood $L$ of
guessing correctly. The likelihood will be $L = {\rm
Max} \{ w_+, w_-\}$, and from this relation it is
evident that $1/2 \leq L \leq 1$. The likelihood can be
renormalized to yield the predictability $P$
\cite{Englert}, given by
\begin{equation}
P = 2 L -1 .
\label{eq:disting}
\end{equation}
It is clear that $0 \leq P \leq 1$,
where $P=0$ corresponds to a random guess of which path
the particle took, and $P=1$ corresponds to absolute
certainty about the path. Take note that
$P$ corresponds to the likelihood of the correct
{\em estimated outcome}. If one were to estimate the
path of an ensemble of identically prepared particles
according to the ML strategy one should guess
identically the same path for each system. That is,
every estimate would be identical so that the
estimated path would have no variance. If, on the
other hand, one made a {\em factual measurement} of
the path, one would get a random outcome characterized
only by the probabilities
$w_+$ and $w_-$.
One could also measure the visibility when the two
path probability amplitudes interfere. The visibility
$V$, too, is a statistical measure which requires an
ensemble of identically prepared systems to
estimate. The classical definition of $V$ is
\begin{equation} V = {I_{max}-I_{min} \over I_{max}+I_{min}} ,
\label{eq:classical visibility definition}
\end{equation} where
$I_{max}$ and $I_{min}$ are the intensities of the
interference fringe maxima and minima. For a single
particle we can only talk about the probability $p$ of
the particle falling on a specific location on a
screen, or exiting one of two interferometer ports.
(Do not confuse this probability $p$ with the
predictability $P$.) The probability
$p$ will vary essentially sinusoidally with position
on the screen, or with the interferometer arm-length
difference. In this case the natural definition of
$V$ is
\begin{equation} V = {p_{max}-p_{min} \over p_{max}+p_{min}} ,
\label{eq:quantum visibility definition}
\end{equation}
It has been shown \cite{Englert} that $P$ and $V$ for a
single particle, satisfies the following inequality:
\begin{equation}
P^2 + V^2 \leq 1 ,
\label{eq:Complementarity relation}
\end{equation}
where the upper bound is saturated for any pure state.
A relevant question to ask is how this inequality is
related to the uncertainty principle, which is also a
quantitative inequality manifesting complementarity.
Furthermore one can ask what observables, if any,
$P$ and $V$ correspond to? We shall attempt to
clarify these issues in this paper. We shall also
show that a relation of the same form as
(\ref{eq:Complementarity relation}) can be derived for
any Hermitian operator by constructing a complementary
(and therefore non-commuting) Hermitian operator.
Finally we shall see that there is a substantial
difference between the non-simultaneous and the
simultaneous Heisenberg-Robertson uncertainty
relations.
\section{Generalized complementarity}
Let us consider the states $\ket{A_+}$ and $\ket{A_-}$,
which are eigenstates of the Hermitian operator
$\hat{A}$ with eigenvalues $A_+ \neq A_-$,
respectively. Therefore, $\langle A_+ | A_- \rangle =
0$. In the following we shall assume that the
two-dimensional Hilbert-space spanned by these two
orthonormal states is sufficient to describe the
system. It is permissible that the operator $\hat{A}$ has
additional eigenstates, but to analyze the
complementarity relation we shall only consider
transitions between, and hence superpositions of, two
of the eigenstates. Below, for the sake of clarity, we
shall refer to the {\em system mode}, which is the
physical entity one can make measurements on, and the
{\em system state}, which is the quantum-mechanical
state of the mode, i.e. a result of a measurement. It
is rather straightforward to extend the relation
(\ref{eq:Complementarity relation}) to the case where
a larger Hilbert-space is needed
\cite{Bjork}, but we will refrain from attempting such
a generalization since two states are sufficient to
elucidate the answers to the questions posed above. We
shall furthermore assume that our system is prepared in
a general state with the associated density operator
\begin{equation}
\hat{\rho} = \left [
\matrix{
w_+ & \rho_{12} e^{-i \theta} \cr
\rho_{12} e^{i \theta} & w_-}
\right ],
\label{eq:state definition}
\end{equation}
where $w_+$, $w_-=1-w_+$, $\rho_{12} \leq \sqrt{w_+
w_-}$ and $\theta$ can be assumed to be real positive
numbers without any loss of generality, and where the
density operator is expressed in matrix form in the
$\ket{A_+}$ and
$\ket{A_-}$ basis. We can identify the parameter $w_+$
as the {\em a priori} probability of finding the
system in the state $\ket{A_+}$, and similarly for
$w_-$.
If we use the maximum likelihood strategy to predict
the outcome of a measurement of $\hat{A}$, then we will
succeed with the likelihood $L={\rm Max}\{w_+,w_-\} =
(w_+ + w_- + |w_+ - w_-|)/2 = (1+|w_+ - w_-|)/2$, where
${\rm Max}\{w_+,w_-\}$ denotes the maximum of the two
probabilities $w_+$ and $w_-$. From this equation, and
(\ref{eq:disting}) follows that
\begin{equation}
P = |w_+ - w_-| = \sqrt{(1-2w_-)^2}= \sqrt{1-4w_+ w_-}
.
\label{eq:predictability}
\end{equation}
Note that this quantity is based only on the
probabilities
$w_+$ and $w_-$ which characterize the {\em
preparation} of the state $\hat{\rho}$.
\begin{figure}
\leavevmode
\epsfxsize=8cm
\epsfbox{fig1lanl.eps}
\vspace{0.3cm}
\caption{The normalized dimensionless minimum
uncertainty product of the operators $\hat{A}$ and
$\hat{B}$ as a function of $w_+=\sin^2(\alpha)$. The
dashed line represents the uncertainty product of the
second class of intelligent states. The uncertainty
product of the initial state for operators
$\hat{A}$ and $\hat{B}$ is bounded from below by the solid
line and from above by the dashed line. The dot-dashed
line represents the predictability of the state and
the long dashed line the visibility of a pure state.}
\label{fig:Uncertainty product}
\end{figure}
Next consider the two unitary operators
\begin{equation}
\hat{U}_{PS} = \left [
\matrix{
1 & 0 \cr
0 & \exp(i \phi)}
\right ] ,
\label{eq:ups}
\end{equation}
and
\begin{equation}
\hat{U}_{BS} = \left [
\matrix{
\cos(\xi) & i \sin(\xi) \cr
i \sin(\xi) & \cos(\xi)}
\right ] ,
\label{eq:ubs}
\end{equation}
which both are expressed in matrix form in the
$\ket{A_+}$ and $\ket{A_-}$ basis. The two unitary
transformations correspond to a generalized phase-shift
(of state $\ket{A_-}$) and a generalized
``beamsplitter'', respectively. The density matrix of
the unitarily transformed state is $\hat{\rho}' =
\hat{U}_{BS} \hat{U}_{PS}
\hat{\rho} \hat{U}_{PS}^\dagger \hat{U}_{BS}^\dagger$.
The probability of obtaining the outcome $A_+$ from
$\hat{\rho}'$ if
$\hat{A}$ is measured is $\bra{A_+} \hat{\rho}' \ket{A_+}$.
We denote the maximum and the minimum of this
probability, as a function of $\phi$ and $\xi$,
$p_{max}$ and $p_{min}$. One finds that the global
maxima and minima are found for $\xi = \pi/4 + n
\pi/2$, where $n$ is an arbitrary integer. This choice
of $\xi$ corresponds to a generalized beamsplitter
with a 50 \% transmittivity. We use the global maxima
and minima to define the visibility according to
(\ref{eq:quantum visibility definition}). It is
straightforward to show that from the obtained
expression follows that
\begin{equation}
V = 2 |\bra{A_+}
\hat{\rho}
\ket{A_-}| = 2 \rho_{12} .
\label{eq:visibility expressed in rho}
\end{equation}
From this expression, and (\ref{eq:predictability})
follows that
\begin{equation}
P^2 + V^2 = w_+^2 - 2 w_+ w_- + w_-^2 + 4 \rho_{12}^2
\leq 1 ,
\label{eq:first complementarity}
\end{equation}
since $w_+ + w_- = 1$ and $0 \leq \rho_{12} \leq
\sqrt{w_+ w_-}$. This re-derivation of relation
(\ref{eq:Complementarity relation}) demonstrates
that every two-state quantum system obeys a
complementary relation even before any attempt has been
made to simultaneously measure both $\hat{A}$ and the
quantity complementary to $\hat{A}$. In Fig.
\ref{fig:Uncertainty product} the
generalized distinguishability and visibility are
plotted dash-dotted and long dashed, respectively, as
a function of $w_+$, parameterized by
$\sin^2\alpha = w_+$.
It is obvious that $P$ in some way corresponds to a
measurement of $\hat{A}$. What is the operator
corresponding to the quantity represented by the
visibility? To answer this question let us construct
the two orthonormal states
\begin{equation}
\ket{B_+} \equiv (\ket{A_+} + e^{i \varrho}
\ket{A_-})/\sqrt{2}
\label{eq:b+ state definition}
\end{equation}
and
\begin{equation}
\ket{B_-}
\equiv (\ket{A_+} - e^{i \varrho} \ket{A_-})/\sqrt{2} .
\label{eq:b- state definition}
\end{equation}
These states, in turn, can be used to construct a
complementary Hermitian operator to
$\hat{A}$ which is
\begin{equation}
\hat{B} = B_+ \ket{B_+}\bra{B_+} + B_-
\ket{B_-}\bra{B_-},
\label{eq:b definition}
\end{equation}
where $B_+ \neq B_-$ are real numbers. We have denoted
this operator the complementary operator to
$\hat{A}$ since the eigenstates of $\hat{B}$ are equally
weighted superpositions of the eigenstates of $\hat{A}$.
Hence, if we prepare a state so that the outcome of a
measurement corresponding to $\hat{A}$ can be predicted
with certainty, then nothing can be predicted about the
measurement outcome corresponding to $\hat{B}$,
and vice versa. Since the eigenstates of $\hat{B}$ are
parameterized by $\varrho$ there exists a whole set of
complementary operators to $\hat{A}$. We note that
irrespective of
$\varrho$, the states
$\ket{B_+}$ and
$\ket{B_-}$ are the Hadamard transformations of the
$\ket{A_+}$ and
$\ket{A_-}$ states. This is no coincidence since the
fact that $\hat{B}$ is the complementary operator
to $\hat{A}$ makes the two corresponding bases
optimal for quantum cryptography. This brings in light
the intimate link \cite{Fuchs} between the
complementarity relations, quantum information and
quantum cryptography.
It should be noted that the observable $\hat{A}$ plays
no particular role vis-a-vis the observable $\hat{B}$
in the expression (\ref{eq:first complementarity}).
Assume that the state $\hat{\rho}$ remains invariant,
but that $\hat{A}$ and $\hat{B}$ are interchanged so that the
maximum likelihood estimation and the
generalized visibility measurement pertains to the
outcome of a measurement of
$\hat{B}$, and that new unitary transformations
are defined that have identically the same form as
(\ref{eq:ups}) and (\ref{eq:ubs}) if expressed in the
$\ket{B_+}$ and $\ket{B_-}$ basis. Then we find that
\begin{equation}
P_B = 2 \rho_{12} | \cos(\theta-\varrho) |
\label{eq:B predictability}
\end{equation}
(where we have labeled this predictability with an
index $B$ not to confuse it with the predictability of
estimating $\hat{A}$) and that the new ``visibility''
is given by
\begin{equation}
V_B = \sqrt{w_+^2 + w_-^2 - 2 w_+ w_- + 4 \rho_{12}^2
\sin^2(\theta -\varrho)} .
\label{eq:B visibility}
\end{equation}
Hence, although the likelihood of estimating $\hat{B}$
correctly in general is different from the likelihood
of estimating $\hat{A}$, relation (\ref{eq:first
complementarity}) still holds. We observe that it is
always possible to find an operator $\hat{B}$ for
which the predictability between the measurement
outcomes is zero. This represents the quantum erasure
measurement operator
\cite{Walther,Tan,Kwiat2,Bjork,Herzog,Scully,Kim}. However, the
sum
$P^2 + V^2$ remains invariant and depends only on the
state, not on the choice of complementary operators by
which one estimates and measures $P$ and $V$.
To explore the symmetry between the pairs $\hat{A}$ and
$\hat{B}$, we can see from
(\ref{eq:visibility expressed in rho}) and
(\ref{eq:B predictability}) that the
predictability
$P_B$ of a measurement of the {\em proper}
$\hat{B}$ operator is
\begin{equation}
P_B = 2 \rho_{12} = V
\label{eq:B and P relation}
\end{equation}
By the ``proper $\hat{B}$ operator'' we mean the
complementary operator to $\hat{A}$ that, for the state
$\hat{\rho}$, optimizes the visibility (or
minimizes the variance of $\hat{B}$, see below).
Therefore the operator is defined with
$\varrho=\theta$. Note that the word proper hence is
in reference to the measured state. In the same manner
we see from (\ref{eq:predictability}) and (\ref{eq:B
visibility}) that for the proper $\hat{B}$ operator
\begin{equation}
V_B = |w_+ - w_-| = P .
\label{eq:B and V relation}
\end{equation}
\section{The Heisenberg-Robertson uncertainty relation}
The commutator between $\hat{A}$ and $\hat{B}$ follows from
the definition of the operators and can be expressed:
\begin{equation}
[\hat{A},\hat{B}] = (A_+ - A_-)(B_+ - B_-)(\ket{A_+}\bra{A_-}
- \ket{A_-}\bra{A_+}) .
\label{eq:commutator}
\end{equation}
We see that the operators $\hat{A}$ and $\hat{B}$ are
non-commuting, non-canonical operators. Since the
operators are non-canonical they will be subject to a
generalized uncertainty inequality
\cite{Schrodinger,Robertson,Merzbacher,Puri,Trifonov}.
The uncertainty inequality reads
\begin{equation}
\langle (\Delta \hat{A})^2 \rangle \langle(\Delta \hat{B})^2
\rangle \geq {1 \over 4} \left ( \langle \hat{C}
\rangle^2 + \langle \hat{F} \rangle^2 \right ) ,
\label{eq:uncertainty relation}
\end{equation}
where the $\hat{C}$ is directly proportional
to the commutator and is defined as
\begin{equation}
\hat{C} = -i [\hat{A},\hat{B}] ,
\label{eq:C definition}
\end{equation}
and $\langle \hat{F} \rangle$ is the correlation between
the observables and is defined
\begin{equation}
\langle \hat{F} \rangle = \langle \hat{A} \hat{B} +
\hat{B} \hat{A}\rangle - 2 \langle \hat{A}
\rangle \langle \hat{B}\rangle .
\label{eq:F definition}
\end{equation}
The expectation value and the variance of the state
(\ref{eq:state definition}) can be computed to be
\begin{equation}
\langle \hat{A} \rangle = A_+ w_+ + A_- w_- ,
\label{eq:expectaton value of a}
\end{equation}
and
\begin{equation}
\langle (\Delta \hat{A})^2 \rangle = (A_+ - A_-)^2 w_+ w_-
.
\label{eq:variance of a}
\end{equation}
Using (\ref{eq:predictability}) we can write
\begin{equation}
{\langle (\Delta \hat{A})^2 \rangle \over (A_+ - A_-)^2} =
{1-P^2 \over 4}.
\label{eq:variance of a, alternative form}
\end{equation}
This equation shows the direct link between a
measurement of the operator $\hat{A}$ and the
predictability $P$. If e.g. $P$ is unity, then the
variance of $\hat{A}$ is zero.
To show that a similar relation holds between $\hat{B}$
and $V$ we compute:
\begin{eqnarray}
\langle \hat{B} \rangle & = & {B_+ \over 2} \, [ 1+ 2
\rho_{12} \cos(\theta-\varrho) ] \nonumber \\
& & + {B_- \over
2} \, [ 1- 2 \rho_{12} \cos(\theta-\varrho) ] ,
\label{eq:expectaton value of b}
\end{eqnarray}
and
\begin{equation}
\langle (\Delta \hat{B})^2 \rangle = (B_+ - B_-)^2 {1-4
\rho_{12}^2 \cos^2(\theta-\varrho)\over 4} .
\label{eq:variance of b}
\end{equation}
The variance of $\hat{B}$ is minimized for the
proper operator $\hat{B}$. We note that for
the proper complementary operator, the normalized and
dimensionless variance is given by
\begin{equation}
{\langle (\Delta \hat{B})^2 \rangle \over (B_+ - B_-)^2} =
{1-4 \rho_{12}^2\over 4} = {1-V^2 \over 4}.
\label{eq:normalized variance of b}
\end{equation}
The visibility is thus directly linked to the variance,
or uncertainty, of the proper complementary
operator to $\hat{A}$. As a final result we note that
the normalized uncertainty product can be written
\begin{eqnarray}
{\langle (\Delta \hat{A})^2 \rangle \langle (\Delta
\hat{B})^2 \rangle \over (A_+ - A_-)^2 (B_+ - B_-)^2} &
\geq & w_+ w_- {1-4 \rho_{12}^2 \over 4} = \nonumber \\
& & {(1-P^2)(1-V^2) \over 16} .
\label{eq:uncert prod, alternative form}
\end{eqnarray}
{\em Hence, there is a direct link between the
complementarity relation (\ref{eq:Complementarity
relation}) and the minimum uncertainty product.} This
is one of the main conclusions of this paper. The
result is true regardless if the state in question is
pure or mixed. However, for a pure state the right
hand side of (\ref{eq:uncert prod, alternative form})
can be further simplified to read
\begin{equation}
{\langle (\Delta \hat{A})^2 \rangle \langle (\Delta
\hat{B})^2 \rangle \over (A_+ - A_-)^2 (B_+ - B_-)^2}
\geq {V^2 P^2 \over 16} .
\label{eq:new uncertainty rel.}
\end{equation}
In Fig.
\ref{fig:Uncertainty product} the normalized minimum
uncertainty product (solid line) is plotted
versus the probability $w_+$. The dashed line
represents the maximum uncertainty product.
The Robertson intelligent states $\ket{\psi_{IS}}$ are
given by the eigenstate solution of the equation \cite{Trifonov}
\begin{equation}
(\hat{A} + i \lambda \hat{B}) \ket{\psi_{IS}} =
(\langle \hat{A} \rangle + i \lambda
\langle \hat{B} \rangle) \ket{\psi_{IS}} .
\label{eq:IS equation}
\end{equation}
The complex parameter $\lambda$ satisfies
$|\lambda|^2 = \langle (\Delta \hat{A})^2 \rangle/\langle (\Delta \hat{B})^2
\rangle$. We will here give the solutions for the cases where $\lambda$ is
either real or imaginary \cite{Puri}. For imaginary
$\lambda$ we find
\begin{equation}
\ket{\psi_{IS1}} = \sqrt{w_+} \ket{A_+} \pm e^{i
\varrho}
\sqrt{w_-} \ket{A_-} .
\label{eq. IS 1}
\end{equation}
As $|\lambda|$ increases from
zero towards infinity the two solutions evolve from
$w_+ = 1/2$ towards $w_+ \rightarrow 0$ and $w_+
\rightarrow 1$, respectively (from point A to points
B$_1$ and B$_2$ in Fig. \ref{fig:Uncertainty product}).
The minimum uncertainty states (which are the
intelligent states with the minimum uncertainty)
belong to this class of intelligent states with
$w_+ = 0$, 1/2, and 1, respectively. As expected these are the
eigenstates of
$\hat{A}$ and $\hat{B}$. The intelligent states
with $\lambda$ real are given by
\begin{equation}
\ket{\psi_{IS2}} = {1 \over \sqrt{2}}( \ket{A_+} \pm
e^{i (\varrho \pm \beta)}
\ket{A_-}) .
\label{eq. IS 2a}
\end{equation}
When $|\lambda|$ goes from 0 towards 1,
$\beta$ goes from 0 to $\pi/2$ (and the uncertainty
product goes along the dotted line from point A to
point C in Fig.
\ref{fig:Uncertainty product}). When subsequently
$1 \leq |\lambda|
\rightarrow \infty$, the second class of intelligent
states becomes
\begin{equation}
\ket{\psi_{IS2}} = \sqrt{w_+} \ket{A_+} \pm i e^{i
\varrho} \sqrt{w_-}
\ket{A_-} ,
\label{eq. IS 2b}
\end{equation}
where $w_+$ evolves from 1/2 towards 0 and from 1/2
towards 1, for the respective states. (The uncertainty
product goes along the dashed line from point C to
points B$_1$ and B$_2$.) The two sets of intelligent states considered
above are found to be the states with extreme uncertainty products.
\section{Complementary two-state operators}
In the preceding section we saw that there are
infinitely many complementary operators to
$\hat{A}$. {\em However, for any choice of $\varrho$ there
are only two sets of three mutually complementary
operators.} If we explicitly write out the operator
$\hat{B}$ as a function of $\varrho$, v.i.z.
$\hat{B}(\varrho)$, then the two sets are
$\hat{A}$, $\hat{B}(\varrho)$, $\hat{B}(\varrho + \pi/2)$ and
$\hat{A}$, $\hat{B}(\varrho)$, $\hat{B}(\varrho - \pi/2)$. To
clarify the meaning of this statement we note that the
operator pair $\hat{A}$ and $\hat{B}(\varrho)$, the
pair $\hat{A}$ and
$\hat{B}(\varrho+\pi/2)$ as well as the pair
$\hat{B}(\varrho)$ and $\hat{B}(\varrho+\pi/2)$ are all
complementary. This is the meaning of the term
``a set of mutually complementary operators''.
To identify such an abstract
set of operators with one (of many) specific
observables we can e.g. assume that
$\hat{A}$ corresponds to the spin
$z$ operator $\hat{\sigma}_z$ of a spin 1/2 particle,
with eigenvalues $\pm \hbar/2$. If we furthermore
assume that the eigenvalues of the operators
$\hat{B}(\varrho)$ and $\hat{B}(\varrho \pm \pi/2)$ are $B_1
= - B_2 = \hbar/2$, we readily recognize the other two
complementary operators as the spin
$\hat{\sigma}_u$ and
$\hat{\sigma}_v$ operators, where $u,v,z$ denotes a
right-handed (left-handed) orthogonal Euclidian vector
set for the choice $\varrho + \pi/2$ ($\varrho -
\pi/2$). With this choice the commutation relation
(\ref{eq:commutator}) reduces to the familiar spin
operator commutator
$[\hat{\sigma}_i,\hat{\sigma}_j] = i \hbar
\epsilon_{ijk} \hat{\sigma}_k$ with $i,j,k \rightarrow
u,v,z$, in this case.
Note, however, that orthogonal
spin operators is only one realization of a mutually
complementary set. One can construct infinitely many
such triplets of operators for every two-state system.
In \cite{Bruss,Gisin} such a triplet set was
constructed starting from two orthogonally polarized
single photon states. The three pairs of eigenstates
were subsequently used as bases in a six-state quantum
cryptography protocol. {\em We think that a
generalization of this idea of mutually complementary
operators in Hilbert-spaces of higher dimension have
an obvious application for the construction of
eavesdropping-safe quantum cryptographic protocols.}
\begin{figure}
\leavevmode
\epsfxsize=8cm
\epsfbox{fig2lanl.eps}
\vspace{0.4cm}
\caption{An example of complementary operators. The
single photon two-mode state prepared at left can be
measured either by a number-difference operator (top)
or by a phase-difference measurement (bottom). By
adjusting the phase shift $\varrho$ any operator
$\hat{B}(\varrho)$ can be implemented.}
\label{fig:Operator illustration}
\end{figure}
As a second example of complementary operator
pairs we consider a single particle in a well defined
spatio-temporal mode, impinging on a symmetric
Mach-Zehnder interferometer. The particle can take
either of the two paths. The corresponding
states can be written
$\ket{1}\otimes\ket{0}$ and $\ket{0} \otimes \ket{1}$.
Defining $\hat{A}$ to be the two-mode particle
number-difference operator
$(\hat{n}_1 -\hat{n}_2)/2$ with the
eigenvectors above and with eigenvalues
$\pm 1/2$, we find that the operator $\hat{B}$ corresponds
to the two-mode phase-difference operator (in the first
particle manifold) defined by Luis and S\'{a}nchez-Soto
\cite{Luis,Luis 2,Luis 3}:
\begin{equation}
\hat{\phi}_{12} = \theta |\phi_1\rangle\langle\phi_1| +
(\theta + \pi) |\phi_2\rangle\langle\phi_2|,
\label{eq:phase-difference states}
\end{equation}
where $\ket{\phi_1}=(\ket{1}\otimes\ket{0} +
e^{i \varrho} \ket{0} \otimes \ket{1})/\sqrt{2}$
and $\ket{\phi_2}=(\ket{1}\otimes\ket{0} - e^{i \varrho}
\ket{0} \otimes \ket{1})/\sqrt{2}$, and the
corresponding eigenvalues are $\theta$ and $\theta +
\pi$. Fig. \ref{fig:Operator illustration} provides a
schematic illustration of the implementation of the
state preparation (left) and the operator set (right).
{\em Note that the proper operators in this case,
where the eigenstates are discrete, do not correspond
to position and momentum operators}, as textbook
discussions of this specific interferometric duality
experiment often indicate. This has already been noted
and discussed by de Muynck \cite{Muynck}. In the same
vein Luis and S\'{a}nchez-Soto have used the
phase-difference operator to analyze the
mechanism which enforces complementarity, and
specifically loss of fringe visibility with increasing
distinguishability \cite{Luis 3}.
Identifying the operator $\hat{A}$ with any Hermitian
operator (for which it makes sense to have a
restricted two-state Hilbert-space) it is
always possible to write down a complementarity
relation for $\hat{A}$. Since the form
of the complementary operator $\hat{B}$ to $\hat{A}$ is
known, and it has a simple form, it is often possible
to identify this operator with some known observable.
Since complementarity follows directly from the
superposition principle, and therefore permeates all of
quantum mechanics, the term {\em welcher weg}
experiment which is intimately tied to complementarity
should perhaps be replaced by the term {\em welcher
zustand} experiment.
\section{Complementarity and uncertainty relations for
simultaneous measurements}
So far we have discussed the standard
Heisenberg-Robertson uncertainty relation which makes
a statement about the preparation of a state. E.g. the
variance of $\hat{A}$ computed in (\ref{eq:variance of a})
is the variance associated with a
sharp measurement of $\hat{A}$ on an ensemble of systems
all prepared in the state $\hat{\rho}$. The
measurement will either destroy the state or
collapse the state into an eigenstate of $\hat{A}$ so the
sharp measurement will preclude any meaningful
subsequent measurement of $\hat{B}$. However, it is also
interesting to see what the uncertainty product becomes
if one tries to {\em simultaneously} measure
$\hat{A}$ {\em and} $\hat{B}$. By ``simultaneous'' we
mean that we make (necessarily unsharp)
measurements of both $\hat{A}$ and $\hat{B}$ on each
individual system mode in the ensemble. In order to do
so we need to entangle the system with an auxiliary
meter mode, associated with an arbitrarily large
Hilbert-space
${\cal H}_m$, see Fig. \ref{fig:Simultaneous
measurement}. If the entanglement between the
system and meter modes is perfect (to be quantified
below), then a sharp measurement of the state of the
meter mode will collapse the system mode into an
eigenstate of e.g. $\hat{A}$. This is the principle of a
quantum non-demolition measurement. However, in order
to subsequently be able to say something about $\hat{B}$ of
the initial system state from the same copy of the
quantum state, the entanglement cannot be perfect, and
to be able to say something about
$\hat{A}$ it cannot be zero. Therefore both the
$\hat{A}$ and the $\hat{B}$ measurements need to be unsharp,
that is, associated with additional statistical
uncertainties than what follows from the preparation of
the state. This is well known for simultaneous
measurements \cite{Muynck,Arthurs,Arthurs
2,She,Stenholm,Appleby}.
\begin{figure}
\leavevmode
\epsfxsize=8cm
\epsfbox{fig3lanl.eps}
\vspace{0.3cm}
\caption{A schematic representation of a simultaneous
measurement of $\hat{A}$ and $\hat{B}$ on a single system.
The planes P and D denote the spatial points
where it is appropriate to measure $P$ or $V$, and
$D$ or $V_e$, respectively. If an unsharp measurement
is performed $D$ {\em and} $V_e$ can be measured at
plane D.}
\label{fig:Simultaneous measurement}
\end{figure}
In the following we will only treat the case where the
composite system is pure as this will set the lower
limit to our ability to simultaneous measure or predict
the values of the observables
$\hat{A}$ and $\hat{B}$. We shall assume that the
entanglement is accomplished through some unitary
operation which does not change the probabilities $w_+$
and $w_-$. (The more general case is discussed in
\cite{Bjork}). Such an entanglement should perhaps be
called quantum non-demolition measurement
type entanglement (where the word measurement refers
to $\hat{A}$), since a perfect entanglement of this type
will allow one to make a sharp QND measurement of
$\hat{A}$. This is not the most general form of
entanglement, but it is the type of entanglement
needed for our purposes, so a sufficiently
general pure state of the entangled system can be
written
\begin{eqnarray}
\ket{\psi_e} & = & \sqrt{w_+} \ket{A_+} \otimes
\ket{M_+} + e^{i \theta} \sqrt{w_-} \ket{A_-} \otimes
\ket{M_-}
\nonumber \\ & \equiv & \sqrt{w_+} \ket{A_+} \otimes
\ket{M_+} \nonumber \\
& & + e^{i \theta} \sqrt{w_-} \ket{A_-} \otimes (c
\ket{M_+} +
\sqrt{1-c^2} \ket{M_\perp} ) ,
\label{eq:entangled system}
\end{eqnarray}
where $c$ is real and positive and defined by
$|\bra{M_-}M_+\rangle|=c$, and
$\bra{M_+}M_\perp\rangle=0$. We see that
since
$\hat{A}$ has a binary measurement outcome, we need only
consider the reduced 2-dimensional meter mode
Hilbert-space ${\cal H}_r$ spanned by $\ket{M_+}$ and
$\ket{M_\perp}$. The density operator associated with
$\ket{\psi_e}$ will be denoted
$\hat{\rho}_e$. With the help of the parameter
$c$, which is a measure of the entanglement between the
modes, the distinguishability
$D$ based on an optimal measurement of the meter
mode, can be expressed
\begin{equation}
D \equiv ||\bra{A_+} \hat{\rho}_e \ket{A_+} - \bra{A_-}
\hat{\rho}_e
\ket{A_-}||_1 = \sqrt{1 - 4 c^2 w_+ w_-} .
\label{eq:distinguishability definition}
\end{equation}
(The notation $ ||\hat O||_1\equiv {\rm Tr}\{\sqrt{\hat
O^\dagger \hat O}\}$ denotes the trace-class norm of
the operator $\hat O$.) The distinguishability $D$ has
the same connection to the likelihood of guessing
correctly about the outcome of a sharp $\hat{A}$
measurement, that is $D = 2L-1$. However, $D$ is
associated with a factual {\em measurement} and not
simply by an estimate based on {\em a priori}
knowledge. It may be noted that $P \leq D$ always holds
\cite{Englert,Bjork} provided the QND-type of
entanglement is used.
The visibility $V_e$ of the entangled system is given
by
\begin{equation}
V_e = 2 |\bra{A_-} {\rm Tr}_r \{ \hat{\rho}_e \}\ket{A_+}| =
2 c \sqrt{w_+ w_-} ,
\label{eq:visibility definition}
\end{equation}
where the partial trace is taken over ${\cal H}_r$.
Since the state $\ket{\psi_e}$ is pure, the relation
\begin{equation}
D^2 + V_e^2 =1
\label{eq:D complementarity}
\end{equation}
holds \cite{Englert}.
A measurement of the meter mode will allow us to
deduce more or less about the initial state of the
system (\ref{eq:state definition}). If e.g. $c=0$,
then the wavefunction (\ref{eq:entangled system})
represents a maximally entangled state and a
measurement of the meter mode, using
$\ket{M_+}$ and
$\ket{M_\perp}$ as projection bases, will yield a
perfect correlation with a subsequent $\hat{A}$ measurement
of the system mode. However, as noted above, such a
measurement of the meter mode will preclude any
meaningful information to be reaped about the
complementary observable $\hat{B}$. Therefore, we will in
the following discuss how to perform an optimal
simultaneous measurement of
$\hat{A}$ and $\hat{B}$. With simultaneous we understand a
measurement where we try to obtain some information of
both $\hat{A}$ and $\hat{B}$ from a single copy of a
pure state (\ref{eq:state definition}) which after an
entangling interaction is transformed to
(\ref{eq:entangled system}).
To find the minimum uncertainty product of a
simultaneous measurement of $\hat{A}$ and $\hat{B}$ we
have assumed that the $\hat{A}$ information will be
obtained by making a sharp measurement of the meter
mode, and the
$\hat{B}$ information will be obtained by making a
subsequent sharp measurement on the system mode. (We
note that one could also have chosen to do the
opposite. However, if we do so, the system and meter
modes should be entangled in a different manner than
what has been assumed above. Still, if we choose the
inverse measurement procedure and do it optimally, the
final result, quantified by a simultaneous uncertainty
product, remains identical to the result below.) Let us
start with the
$\hat{B}$ measurement. The two pertinent projectors are
$|B_+\rangle \langle B_+|$ and $|B_-\rangle \langle
B_-|$. The associated probabilities are
\begin{equation}
P_{B_\pm} = \bra{B_\pm} {\rm Tr}_r \{ \hat{\rho}_e \}
\ket{B_\pm} = {1
\over 2} \pm c \sqrt{w_+ w_-} \cos(\theta-\varrho) .
\label{eq:PB+}
\end{equation}
If the two outcomes are associated with the
values
$B_+'$ and
$B_-'$, then the mean of the measurement, or rather
estimation, of
$\hat{B}$ (which we will denote $\hat{B}'$) will yield
\begin{equation}
\langle \hat{B}' \rangle = {B_+' + B_-' \over 2} + (B_+' -
B_-') c
\sqrt{w_+ w_-}
\cos(\theta-\varrho) .
\label{eq:Mean B estimation}
\end{equation}
We note that, without loss of generality, we can
assume a gauge such that the true eigenvalues $B_+$
and $B_+$ fulfill $B_+ = - B_- = B$, and hence $B_+' = -
B_-' = B'$. To make the estimated mean (\ref{eq:Mean B
estimation}) equal the true mean (\ref{eq:expectaton
value of b}) we set $B_\pm' = B_\pm/c$. We point out
that the choice is independent of the initial state of
the system mode (which is characterized by $w_+$ and
$\theta$), but depends on the degree of entanglement
$c$. The assumption of a ``true mean'' meter is
essential to what follows, and is also made by Arthurs,
Kelly and Goodman in their seminal papers on
simultaneous measurements \cite{Arthurs,Arthurs 2}.
From the assumption follows that the variance of the
estimate of $\hat{B}$ is
\begin{eqnarray}
\langle (\Delta \hat{B}')^2 \rangle & = & {(B_+'-B_-')^2
\over 4} \, [ 1 - 4 c^2 w_+ w_-
\cos^2 (\theta-\varrho) ] \nonumber \\
& = & B^2 \left( {1 \over c^2} - 4
w_+ w_-
\cos^2 (\theta-\varrho) \right) .
\label{eq:B estimation variance}
\end{eqnarray}
It is obvious from this expression that in order to
minimize the variance of the estimate of $\hat{B}$, one
should choose $\theta=\varrho$. We also see that when
$c =1$, that is, when the system and meter modes are
unentangled, the result reduces identically to
(\ref{eq:variance of b}). On the other hand, when
$c\rightarrow 0$, the variance diverges in spite
of the fact that
$\hat{B}$ has a finite number of finite eigenvalues.
This is a consequence of our requirement that the
mean of the estimate should equal the true mean of the
state.
If the (proper) choice $\theta=\varrho$ is made, then
we can express the normalized variance in the
visibility (\ref{eq:visibility definition}):
\begin{equation}
{\langle (\Delta \hat{B}')^2 \rangle \over (B_+'-B_-')^2}
= {1-V_e^2 \over 4}
\label{eq:reformulation}
\end{equation}
It is to be expected that such a relation holds,
because the visibility, as shown above, corresponds to
a sharp measurement of the uncertainty of operator
$\hat{B}$.
In order to best estimate the outcome of a
measurement of $\hat{A}$ of the initial system state from a
measurement on the meter mode, we need to find the
optimal projectors. Since the entangled state
(\ref{eq:entangled system}) can be expressed in a 2
$\times$ 2 Hilbert space, we need only construct two
meter mode projectors. The most general forms for the
projector-states are
\begin{equation}
\ket{M_1} = \cos(\gamma) \ket{M_+} + e^{i \kappa}
\sin(\gamma) \ket{M_\perp}
\label{eq:Def M1}
\end{equation}
and
\begin{equation}
\ket{M_2} = -\sin(\gamma) e^{i \kappa} \ket{M_+} +
\cos(\gamma) \ket{M_\perp} .
\label{eq:Def M2}
\end{equation}
However, it is immediately obvious that in order to
best estimate $\hat{A}$ the choice $\kappa=0$ (or
$\kappa=\pi$ which is equivalent to $\ket{M_1}
\leftrightarrow \ket{M_2}$) should be made since
$\ket{M_\perp}$ is defined such that $c$ is real.
Furthermore, again we shall assume that a gauge is
chosen so that $A_+ = -A_- = A$, and therefore the
measurement values associated with the two outcomes
$A_+'$ and $A_-'$ fulfill $A_+' = -A_-' = A'$. With this
choice, the expectation value of the estimate of
$\hat{A}$ (which we call $\hat{A}'$) becomes
\begin{eqnarray}
\langle \hat{A}' \rangle & = & A' \{ [\cos^2(\gamma)
- \sin^2(\gamma)] [ w_+ - w_-(1-2c^2)] \nonumber
\\ & & + 4 w_- c
\sqrt{1-c^2}
\cos(\gamma)
\sin (\gamma) \}
\label{eq:Mean A estimation}
\end{eqnarray}
We see that in order to make the estimated mean
correct, c.f. (\ref{eq:expectaton value of a}),
and independent of the initial system state, the
following two conditions must be met:
\begin{equation}
{\cos^2(\gamma) - \sin^2(\gamma) \over
2 \cos(\gamma) \sin(\gamma) } = -{\sqrt{1-c^2} \over
c}
\label{eq:Condition 1}
\end{equation}
and
\begin{equation}
A' = {A \over \cos^2(\gamma) - \sin^2(\gamma)} = {A
\over \sqrt{1-c^2}},
\label{eq:Condition 2}
\end{equation}
where the second equality in (\ref{eq:Condition 2})
follows from (\ref{eq:Condition 1}). Note that both
conditions are state independent. That is, $\langle
\hat{A}' \rangle = \langle \hat{A} \rangle$ irrespective of
$w_+$ and $\theta$ if the two conditions are met. The
variance of the estimate of
$\hat{A}$ can then be computed to be
\begin{equation}
\langle (\Delta \hat{A}')^2 \rangle = A^2 \left ( {c^2
\over 1-c^2} + 4 w_+ w_- \right ).
\label{eq:A estimation variance}
\end{equation}
We see that, as expected, the estimated variance equals
the true variance for $c=0$, that is, when the system
and meter modes are maximally entangled. The estimated
variance diverges when $c \rightarrow 1$. This, too,
is a consequence of requiring a correct estimated mean.
The normalized and dimensionless simultaneous
uncertainty product is hence
\begin{eqnarray}
\lefteqn{{\langle (\Delta \hat{A}')^2 \rangle \langle (\Delta
\hat{B}')^2 \rangle \over 16 A^2 B^2}} & & \nonumber \\
& & = \left ( {c^2 \over
4(1-c^2)} + w_+ w_- \right )\left ( {1 \over 4 c^2} -
w_+ w_-
\right ).
\label{eq:Uncertainty product}
\end{eqnarray}
As expected, the uncertainty product is larger
than (\ref{eq:uncert prod, alternative form}) and
depends both on the initial state and on the degree of
entanglement between the system mode and the meter
mode. The entanglement parameter $c$ can be used to
shift the measurement uncertainty, to some extent, from
one of the operators to the complementary one.
If, for each choice of
$w_+$, we optimize the entanglement to minimize the
uncertainty product, and this is the principle used
in \cite{Arthurs,Arthurs 2} to derive the
minimum uncertainty product for a simultaneous
measurement of two {\em canonical} non-commuting
operators, then the ensuing normalized minimum
uncertainty product is given by:
\end{multicols}
\vspace{-0.5cm}
\noindent\rule{0.5\textwidth}{0.4pt}\rule{0.4pt}{\baselineskip}
\widetext
\begin{equation}
{\langle (\Delta \hat{A})^2 \rangle \langle (\Delta
\hat{B})^2 \rangle \over 16 A^2 B^2} = {-16 w_+^2
w_-^2 (1-4w_+ w_-)^2+ [ 1 - 12 w_+ w_-(1-4w_+ w_-) ]
\sqrt{ w_+ w_- (1 - 4 w_+ w_-)} \over 16 [ -4 w_+
w_- (1-4w_+ w_-) + \sqrt{w_+ w_- (1 - 4 w_+
w_-)} ]}
\label{eq:Messy expression}
\end{equation}
\begin{multicols}{2}
\noindent
and this expression holds for
\begin{equation}
c = \sqrt{{-4 w_+ w_- + 2\sqrt{w_+ w_-(1-4w_+ w_-)}}
\over 1 - 8 w_+ w_- } .
\label{eq:optimal entanglement}
\end{equation}
This is the other major result of this paper,
where the physical implications of (\ref{eq:Messy
expression}) rather than the (rather messy) form
should be retained. Remember that this result is
contingent on {\em a priori} information about the
preparation of the state (i.e. to be able to make a
minimum uncertainty measurement of the state,
$w_+$ and
$\theta$ must be known). The result is plotted in Fig.
\ref{fig:Simultaneous uncertainty product},
solid line. This is the minimum uncertainty of a
simultaneous measurement of the two complementary and
non-canonical operators $\hat{A}$ and $\hat{B}$. The result
should be compared with the standard uncertainty
product (\ref{eq:uncert prod, alternative form}) of
a state in two-dimensional Hilbert-space.
The corresponding distinguishability and visibility
are plotted as dash-dotted and long dashed lines,
respectively, in Fig. \ref{fig:Simultaneous uncertainty
product}. We have assumed that the entanglement
parameter $c$ for each
$w_+$ is chosen according to (\ref{eq:optimal
entanglement}). By necessity the distinguishability
is higher than the predictability for the entangled
state while the visibility is correspondingly lower
than for the initial state (c.f. Fig.
\ref{fig:Uncertainty product}). We
also observe that due to the ``noise term'' in
(\ref{eq:A estimation variance}) there is no simple
connection between the uncertainties of $\hat{A}'$ and
the distinguishability (\ref{eq:distinguishability
definition}). Therefore one cannot find any general
relation between (\ref{eq:D complementarity}) and the
non-simultaneous uncertainty relation (\ref{eq:uncert
prod, alternative form}). This was noted already by
Englert who asserted that ``\ldots the duality
relation (\ref{eq:D complementarity}) is logically
independent of the uncertainty relation
\ldots .'' The reason is that whereas $\langle (\Delta
\hat{A})^2 \rangle$, $\langle (\Delta \hat{B})^2 \rangle$,
$P$, $V$ and $V_e$ all correspond to the uncertainties
of factual measurements, $D$ does not. Instead it
characterizes a ML estimate, which cannot be directly
related to a factual measurement of the
corresponding operator.
\begin{figure}
\leavevmode
\epsfxsize=8cm
\epsfbox{fig4lanl.eps}
\vspace{0.3cm}
\caption{The normalized dimensionless
minimum uncertainty product for a
simultaneous measurement of the operators $\hat{A}$ and
$\hat{B}$, solid line. The dot-dashed line represents the
distinguishability and the long dashed line the
visibility of the entangled state, provided that the
optimum entanglement parameter $c$ is chosen for each
value of $w_+$.}
\label{fig:Simultaneous uncertainty product}
\end{figure}
It is interesting to make a connection between
(\ref{eq:Messy expression}) and (\ref{eq:uncert prod,
alternative form}) via (\ref{eq:optimal
entanglement}). We have already noted that the two
uncertainty relations are different since they
represent physically different measurements on
physically different states. However, we have striven,
as far as possible, to try to design a ``fair''
procedure to simultaneous measure
$\hat{A}$ and $\hat{B}$ of the state (\ref{eq:state
definition}) through a measurement of the
state (\ref{eq:entangled system}). Specifically we
have required that the expectation values of
the corresponding measurements coincide. Using the fact
that the optimum entanglement given by (\ref{eq:optimal
entanglement}) can be expressed
\begin{equation}
c = \sqrt{{V(P-V) \over P^2-V^2}} ,
\label{eq:new optimal c}
\end{equation}
some somewhat tedious algebra will show that the
normalized simultaneous uncertainty product can
be expressed
\begin{equation}
{\langle (\Delta \hat{A}')^2 \rangle \langle (\Delta
\hat{B}')^2 \rangle \over 16 A^2 B^2} = {(1-VP)^2 \over
16}.
\label{eq:Uncertainty product 2}
\end{equation}
This rather remarkably simple result should be
compared to (\ref{eq:uncert prod, alternative
form}) and (\ref{eq:new uncertainty rel.}). {\em Hence,
we see that our requirement of correct expectation
values enforces a simultaneous uncertainty
product which is uniquely dictated by the preparation
of the state.} This is not wholly surprising as e.g.
the eigenstates of
$\hat{A}$ and
$\hat{B}$ could reasonably be expected to have the
smallest simultaneous uncertainty product.
Appleby \cite{Appleby} has argued that the
simultaneous uncertainty relation is less general than
the Heisenberg-Robertson uncertainty relation. We
agree, and this is demonstrated by our analysis. While
the Heisenberg-Robertson uncertainty relation is based
on sharp, non-simultaneous measurements of the two
complementary operators, and therefore operationally
well defined, the simultaneous uncertainty relation is
based on unsharp measurements. How should these unsharp
measurements be performed? More specifically, how
should e.g. the meter-state projection basis
(\ref{eq:Def M1}) and (\ref{eq:Def M2}) be chosen, and
how should the outcomes of the meter-system
measurement be interpreted? We have, following Arthurs
and Kelly, required that the expectation value of
the unsharp measurement equals the true mean. This
requirement enters our analysis through
(\ref{eq:Condition 1}) and (\ref{eq:Condition 2}). In
Appleby's terminology this defines a retrodictively
unbiased measurement. This is, however, not the only
reasonable choice. We can instead choose the
meter-state projection basis to optimize the
distinguishability for every value of the entanglement
parameter $c$. With this choice one cannot make a
state-independent retrodictively unbiased measurement.
None-the-less, the choice is reasonable but will
result in a different minimum uncertainty relation
than the one we derived. In our eyes, if the meter
projection basis is chosen to optimize the
distinguishability, it is more natural to make the
quantitative statement of complementarity in terms of
equation (\ref{eq:D complementarity}). The conclusion
is that any expression that makes a statement of a
simultaneous measurement of complementary observables
irrevocably must involve properties of the meter, in
addition to the properties of the state.
It is noteworthy that the simultaneous uncertainty
relation we derived differs from that derived by
Arthurs, Kelly, and Goodman \cite{Arthurs,Arthurs 2}.
They, and subsequent workers, implicitly or
explicitly assumed that the pertinent operators were
canonical, and in this case the uncertainty product
for a simultaneous measurement is simply four times
larger than the standard uncertainty product. In our
case the situation is more complex. Due to the fact
that $\hat{A}$ and $\hat{B}$ are non-canonical the
uncertainty product of a simultaneous measurement is
not simply scaled by a constant factor. Specifically,
the simultaneous uncertainty product of the
eigenstates to
$\hat{A}$ and $\hat{B}$ is non-zero due to the ``correct mean
assumption''. We believe that this is a general result
for any non-canonical observables.
\section{Conclusions}
In physics textbooks quantum complementarity is often
exemplified in terms of one ``particle'' passing
through a double-slit. The complementary observables
are usually taken to be the canonical position and
momentum operators, without much justification. We
have shown how complementarity is a natural
consequence of the superposition principle, and
explored the connection between complementarity and
the uncertainty relations. We have shown that for any
two-state system one can always formulate a
generalized complementarity relation, and that this
relation typically cannot be interpreted in terms of
position and momentum operators. We have also indicated
that for a system with a discrete number of
non-degenerate eigenstates, the corresponding
operators are not canonical. Never-the-less, for a
two-state system simple and rather intuitive relations
hold between expressions of complementarity and
uncertainty.
We have also shown that if a simultaneous
measurement of complementary operators is made on a
two-state system, a different uncertainty
relation arises than that derived by Arthurs
and Kelly. Finally, we have indicated some natural
connections between complementarity and quantum
information. This is to be expected since a two-state
system is a natural physical manifestation of a qubit,
irrespective of its particular physical implementation
(spin, excitation, charge, etc.).
\acknowledgments
This work was supported by grants from
the Swedish Technical Science Research Council, STINT,
the Royal Swedish Academy of Sciences and by INTAS
through Grant 167/96.
|
\section{Introduction}
M\,82 is the best studied nearby starburst galaxy
(${\rm D}\,=\,3.25\, {\rm Mpc}$). The central few hundred
parsecs of this galaxy are heavily obscured by dust and gas which
hides the central starburst region against direct observations at optical
wavelengths. Evidence for strong star--forming activity in the
central region comes from radio (e.g. Kronberg {\rm et\ts al.}\
\cite{kronberg81}) and infrared observations (e.g. Telesco {\rm et\ts al.}\
\cite{telesco91}) and also from the prominent bipolar outflow visible
in \ifmmode{{\rm H}\alpha}\else{${\rm H}\alpha$}\fi\ (e.g. Bland \& Tully \cite{bland88}, McKeith {\rm et\ts al.}\ \cite{mckeith95},
Shopbell \& Bland--Hawthorn \cite{shopbell98}) and in X-rays
(e.g. Bregman {\rm et\ts al.}\ \cite{bregman}).
The massive star formation (SF) is believed to be fuelled by the large
amount of molecular gas which is present in the centre of M\,82.
On the other hand, SF effects the distribution and kinematics of the
surrounding interstellar medium (ISM). Recent millimetre
continuum observations (Carlstrom \& Kronberg
\cite{carlstrom90}) suggested that the H{\small II} regions in M\,82
have swept up most of the surrounding neutral gas and dust into dense
shells. This is in agreement with the standard picture: shells are
created by young star--forming regions through strong stellar
winds of the most massive stars in a cluster and through subsequent
type--II supernovae (e.g.\ Tenorio--Tagle \& Bodenheimer \cite{tenorio88}).
These processes are thought to blow huge cavities filled with coronal
gas into their ambient ISM (e.g.\ Cox \& Smith \cite{cox74}, Weaver {\rm et\ts al.}\
\cite{weaver77}). This hot interior is then believed to drive the
expansion of the outer shell of swept--up material.
Once superbubbles reach sizes that are comparable to the thickness of
a galaxy's disk, the bubble will eventually break out into the
halo. This then leads to an outflow of the hot gas with
velocities much higher than the expansion of the shell within the
disk of the galaxy. In the following we present
evidence for a molecular superbubble in M\,82 which already broke out
of the disk and seems to contribute significantly to the
well--known prominent outflow of M\,82.
\section{Observations}
\subsection{Molecular Lines}
For our analysis we used the {\ifmmode{{^{12}{\rm CO}(J\!=\!1\! \to \!0)}}\else{{$^{12}{\rm CO}(J\!=\!1\! \to \!0)$}}\fi}\ data cube obtained by Shen \& Lo
(\cite{shen}) with the BIMA array (spatial resolution: $2.5''$) and
the \ifmmode{^{13}{\rm CO}(J\!=\!1\! \to \!0)}\else{$^{13}{\rm CO}(J\!=\!1\! \to \!0)$}\fi\ data cube from Neininger {\rm et\ts al.}\ (\cite{nico13}) observed
with the Platau de Bure interferometer (PdBI) (spatial resolution:
$4.2''$). In addition, we used unpublished PdBI data of the
\ifmmode{^{12}{\rm CO}(J\!=\!2\! \to \!1)}\else{$^{12}{\rm CO}(J\!=\!2\! \to \!1)$}\fi\ and \ifmmode{{\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)}\else{${\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)$}\fi\ transitions. These observations where carried out in
April 1997 in the CD configuration, resulting in a spatial resolution
of $1.4''\times 1.2''$ (\ifmmode{^{12}{\rm CO}(J\!=\!2\! \to \!1)}\else{$^{12}{\rm CO}(J\!=\!2\! \to \!1)$}\fi) and $3.7'' \times 3.5''$ (\ifmmode{{\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)}\else{${\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)$}\fi), and a velocity
resolution of 3.3 \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ and 6.8 \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi, respectively. In order to
increase the sensitivity to extended CO emission we combined the
\ifmmode{^{12}{\rm CO}(J\!=\!2\! \to \!1)}\else{$^{12}{\rm CO}(J\!=\!2\! \to \!1)$}\fi\ data cube with single dish measurements obtained with the IRAM 30m
telescope. These observations were carried out in May
1998. The combination is essential for this particular study
because the receding part of the superbubble is only marginally visible
in the mere interferometer maps. A full account of the data
reduction will be given elsewhere.
\subsection{Evidence for an Expanding Superbubble}
Fig.~\ref{12co10} shows the integrated $^{12}{\rm CO}$ line emission
published by Shen \& Lo (\cite{shen}). The cross corresponds to the
position of the supernova remnant 41.9+58 (SNR 41.9+58) which
is the strongest cm continuum point--source in M\,82 (Kronberg {\rm et\ts al.}\
\cite{kronberg81}). It is considered to be the aftermath of a
'hypernova', which exhibits a radio luminosity 50--100 times greater
than typical for type--II SNe (Wilkinson \& de Bruyn
\cite{wilkinson}). The line along the major axis indicates the
orientation of the position--velocity (pv) diagrams shown in
Fig.~\ref{pv}.
\begin{figure}[h]
\hspace*{0.5cm}
\resizebox{7cm}{!}{\includegraphics{Bb221-f1.eps}}
\caption{Integrated $^{12}{\rm CO}$ line emission from Shen \& Lo (1995).
The lines along the major and minor axis of M\,82 indicate the
orientations of the pv--diagrams shown in Figs.~\ref{pv} and
\ref{pv-minor}. The cross marks the position of SNR 41.9+58. }
\label{12co10}
\end{figure}
\noindent The pv--cut is orientated in such a way that the signature of
the expanding superbubble is visible most distinctly. The angular axis
of the pv diagrams correspond to the offset in arcseconds from SNR
41.9+58. Besides a constant velocity gradient (cf. Shen \& Lo
\cite{shen}) the pv--diagrams show an expanding ring--like feature
centred on SNR 41.9+58, with a central velocity
of ${\rm V}_{{\rm lsr}}\! \approx \! 150\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$. The approaching
velocity component of the ring is
clearly seen in all cubes and is centred at $100\, \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$. The receding
component is only marginally visible in the $^{13}{\rm CO}\\(J\!=\!1\! \to \!0)$
and \ifmmode{{\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)}\else{${\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)$}\fi\ lines; its
central velocity is about $190\, \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$. An enlargement of the pv--diagram
along the major axis is shown in Fig.~\ref{pv-minor} (left). From this
dia\-gram we estimate the radius of the ring to be ($65\,\pm\, 5$)~pc and
the expansion velocity ($45\, \pm \, 5$)~\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi. A pv--diagram along the minor
axis centred on SNR 41.9+58 is presented in Fig.~\ref{pv-minor}
(right). The orientation of the cut is shown in Fig.~\ref{12co10}.
\begin{figure}[h]
\vspace*{-0.4cm}
\hspace*{0.5cm}
\resizebox{6.5cm}{!}{\includegraphics{Bb221-f2.eps}}
\caption{The pv--diagrams along the major axis of M\,82 (orientation
shown in Fig.~\ref{12co10}) for the {\ifmmode{{^{12}{\rm CO}(J\!=\!1\! \to \!0)}}\else{{$^{12}{\rm CO}(J\!=\!1\! \to \!0)$}}\fi}, \ifmmode{^{12}{\rm CO}(J\!=\!2\! \to \!1)}\else{$^{12}{\rm CO}(J\!=\!2\! \to \!1)$}\fi, \ifmmode{^{13}{\rm CO}(J\!=\!1\! \to \!0)}\else{$^{13}{\rm CO}(J\!=\!1\! \to \!0)$}\fi\ and \ifmmode{{\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)}\else{${\rm C}^{18}{\rm O}\-(J\!=\!1\! \to \!0)$}\fi\ cubes.
The slice is centred on SNR 41.9+58.}
\label{pv}
\end{figure}
\noindent Fig.~\ref{pv-minor}\,(right) reveals two emission features
centred at the position of SNR 41.9+58, which correspond to the approaching
($v \! \approx \! 100\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$) and
the receding component ($v\! \approx \! 190\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$) of the expanding
superbubble. Hardly any CO emission with a velocity between
$v \! = \! 100\, \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$ and $v \! = \! 190\, \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$ is found south
and north of SNR 41.9+58. The pv--diagrams therefore show
that the expanding molecular shell has already broken out of the disk
and now only shows the signature of an expanding molecular ring.
It should be noted that the remaining molecular gas in M\,82
shows clear solid--body rotation.
\subsection{Other wavelengths}
\paragraph{Radio continuum observations:\\}
Radio continuum observations at 408 MHz by Wills {\rm et\ts al.}\
(\cite{wills97}) unveiled a prominent `hole' of approximately 100 pc
diameter around SNR 41.9+58 which they attribute to free--free
absorption. They propose that this feature is due to absorption by a
large H\,{\small{\sc II}} region that has been photoionized by a
cluster of early--type stars of which the progenitor of SNR 41.9+58
was originally a member. It should be noted, however, that SNR 41.9+58
is only about 50 years old (Wilkinson \& de Bruyn \cite{wilkinson})
and therefore cannot be the source that drives the expansion
of the superbubble. But its presence supports the scenario that the
expanding molecular superbubble is powered by an interior stellar cluster.
\vspace*{-0.3cm}
\begin{figure}[h]
\hspace*{-0.5cm}
\resizebox{8cm}{!}{\includegraphics{Bb221-f3.eps}}
\caption{The pv--diagrams for the {\ifmmode{{^{12}{\rm CO}(J\!=\!1\! \to \!0)}}\else{{$^{12}{\rm CO}(J\!=\!1\! \to \!0)$}}\fi}\ transition along the
major and minor axis of M\,82. The spatial axis for each
diagram is centred on SNR 41.9+58. The pv--diagram along the major
axis is an enlargement of Fig.~\ref{pv} (top).
The orientation of the cut along the minor axis is indicated in
Fig.~\ref{12co10}.}
\label{pv-minor}
\end{figure}
\vspace*{-1cm}
\paragraph{Neutral hydrogen (H{\small I}):\\}
H\,{\small{\sc I}}\ emission line studies show that the central kpc of M\,82
is dominated by absorption (Yun {\rm et\ts al.}\ \cite{yun93}), which makes
it difficult to study the
H\,{\small{\sc I}}\ kinematics in the central part of the galaxy. Recent H\,{\small{\sc I}}\
absorption studies against supernova remnants in M\,82 by Wills {\rm et\ts al.}\
(\cite{wills98}) disclosed two absorption features against SNR
41.9+58. The velocities of the components are $87 \pm 7\, \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ $ and
$200 \pm 7\, \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$. The H\,{\small{\sc I}}\ component at 87 \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ can be
attributed to the approaching component of the expanding superbubble.
The absorption feature at 200 \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi, if associated with the receding part,
seems to contradict the hypothesis that SNR 41.9+58 is located within
the expanding superbubble. However, the velocity resolution of the
absorption study was rather poor (26.4 \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi) and other explanations
cannot be ruled out at this point.
\paragraph{Optical observations:\\}
Optical observations of the centre of M\,82 suffer heavily
from light absorption by the prominent dust lanes, rendering an
optical analysis of this particular region impossible.
HST V-- and I--band images of the centre of M\,82 (O'Connell {\rm et\ts al.}\
\cite{connell95}) do not reveal any optical sources in the
vicinity of SNR 41.9+58. The prominent \ifmmode{{\rm H}\alpha}\else{${\rm H}\alpha$}\fi\ outflow is visible north
and south of the absorbing dust lane (Bland \& Tully \cite{bland88},
McKeith {\rm et\ts al.}\ \cite{mckeith95}, Shopbell \& Bland--Hawthorn
\cite{shopbell98}). The orientation
of its filaments suggests that the \ifmmode{{\rm H}\alpha}\else{${\rm H}\alpha$}\fi\ outflow emerges at least partly
from the location of the molecular superbubble.
\paragraph{X--ray observations:\\}
Fig.~\ref{xray} shows an overlay of archival ROSAT HRI X--ray data
(Bregman {\rm et\ts al.}\ \cite{bregman}) onto the CO emission. To emphasize
the diffuse, extended X--ray outflow we subtracted a model of the
brightest X--ray point source at $\alpha=9^h55^m50^s.4$,
$\delta=69^{\circ}40'47.5''$ (J2000.0) and smoothed the residual image
to a $6''$ resolution. The positions of the three point sources
in the field are marked with stars, SNR 41.9+58 is marked by a
cross (see also Bregman {\rm et\ts al.}). The overlay shows that most of the diffuse
X--ray emission arises from the vicinity of SNR 41.9+58. The shape of
the diffuse emission is elongated, extending several arcminutes along
the minor axis of the galaxy. Model calculations by Bregman {\rm et\ts al.}\ show
that the emission is consistent with an outflow of heated material from the
central region coincident with the position of the expanding molecular
superbubble.
\begin{figure}[h]
\hspace*{0.3cm}
\resizebox{8cm}{!}{\includegraphics{Bb221-f4.eps}}
\caption{Integrated {\ifmmode{{^{12}{\rm CO}(J\!=\!1\! \to \!0)}}\else{{$^{12}{\rm CO}(J\!=\!1\! \to \!0)$}}\fi}\ line emission of M\,82 (greyscale).
The contour map represents the diffuse X--ray emission without the
strongest point--source in the core of M\,82 as observed by the ROSAT
HRI. The three X-ray point--sources are marked with a star, the
position of SNR 41.9+58 by a cross; the centre of M\,82 ($2.2\,\ifmmode{\mu {\rm m}}\else{$\mu$m}\fi\ $
peak) is indicated by a filled box.}
\label{xray}
\end{figure}
\noindent We estimate an upper limit for the energy
still present in the hot gas emerging from the superbubble. For the
calculation we adopt a density ${\rm n}=0.2\,{\rm cm}^{-3}$ and
a temperature ${\rm kT}=1\,{\rm keV}$ for the central region
(Bregman {\rm et\ts al.}) as an upper limit for the mean values in the outflow.
We find an upper limit for the energy of
${\rm E}_{\rm gas} \le 2.7 \times 10^{53}\,$ergs present in a
cylindrical volume with $r=130\,{\rm pc}$ and $z=550\,{\rm pc}$ centred
on the superbubble. From the thickness of the molecular disk of M\,82
($\approx 90\,{\rm pc}$) and the expansion velocity of the superbubble
we estimate a time elapsed since outbreak of
$\tau\,\approx\,3 \times 10^5\,{\rm yr}$.
The required outflow velocity to reach
the observed height above the disk therefore is
${\rm v}\,\approx\,1000\,\ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi$. This value is comparable with
estimates based on particle aging as
derived from radio continuum observations
(Seaquist {\rm et\ts al.}\ \cite{seaquist85}) and measured outflow
velocities in \ifmmode{{\rm H}\alpha}\else{${\rm H}\alpha$}\fi\ (McKeith {\rm et\ts al.}\ \cite{mckeith95}).
\paragraph{Ionized gas:\\}
Most studies of the ionized gas in M\,82 unfortunately do not provide
sufficient spatial and spectral resolution to reveal details of the region
under study. A high--resolution study of the Ne\,{\small{\sc II}}
emission line has been published by Achtermann \& Lacy (\cite{achtermann}).
The integrated line emission in the south--western
region of M\,82 shows a similar double--peaked feature centred on SNR
41.9+58 as traced by the CO. The pv--cut along the major axis clearly shows
a disturbed velocity field in the vicinity of SNR 41.9+58.
The approaching side of the expanding superbubble at 95 \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ is clearly
visible while the receding side at 190 \ifmmode{{\rm \ts km\ts s}^{-1}}\else{\ts km\ts s$^{-1}$}\fi\ is only marginally seen.
The diameter of the Ne\,{\small{\sc II}} shell ($\approx$\,100\,pc)
seems to be somewhat smaller than the corresponding CO
feature, suggesting that the inner
part of the expanding superbubble is ionized. Similar features are
visible in the distribution and the kinematics of
the H41$\alpha$ emission line (Seaquist {\rm et\ts al.}\ \cite{seaquist96}).
The Ne\,{\small{\sc II}} distribution might represent the transition
from the hot coronal gas with a temperature of several $10^{6}\,$K
towards the cold molecular rim of the superbubble.
\section{Discussion}
The picture that emerges from the observations presented above is the
following: a major SF event at the centre of what today shows up as the
prominent expanding molecular superbubble created a cavity
filled with coronal gas by the combined effects of
strong stellar winds and SN explosions. The pressure of the hot--gas
interior drove the expansion of the shell of swept--up material
until it broke out of the disk. The subsequent outflow of hot
material today shows up as diffuse X--ray emission and contributes to
the prominent \ifmmode{{\rm H}\alpha}\else{${\rm H}\alpha$}\fi\ filaments.
From the observational parameters of the expanding superbubble
(expansion velocity and diameter) we estimate a kinematic age of the
superbubble of $1\times10^{6}$ years. The kinematic age is an
upper limit for the actual age since the superbubble is most probably
decelerating. We use Chevalier's equation (Chevalier \cite{chev74})
to derive the amount of energy needed to create the expanding superbubble.
Since Chevalier's equation applies to H\,{\small{\sc I}}\ shells only we corrected
the energy input by a factor of 2 to correct for the difference in
mass between \ifmmode{{\rm H}_2}\else{H$_2$}\fi\ and H. We estimate an ambient \ifmmode{{\rm H}_2}\else{H$_2$}\fi\ density prior to
the creation of the shell of order $120\,{\rm cm}^{-3}$. This is done
by converting the {\ifmmode{{^{12}{\rm CO}(J\!=\!1\! \to \!0)}}\else{{$^{12}{\rm CO}(J\!=\!1\! \to \!0)$}}\fi}\ line integral to \ifmmode{{\rm H}_2}\else{H$_2$}\fi\ column density using a
conversion ratio of N(\ifmmode{{\rm H}_2}\else{H$_2$}\fi)/W(CO) = $1.2 \times 10^{20}$ cm$^{-2}$
K$^{-1}$ km$^{-1}$s (Smith {\rm et\ts al.}\ \cite{smith}) and estimating the
volume of the region from which material was swept up to be
cylindrical with $z=100\,{\rm pc}$ and $r=65\,{\rm pc}$. We derive a
total energy of order $2\times 10^{54}\,$ergs, which corresponds to an
energy equivalent of approximately 1000 type--II SNe and the strong
stellar winds of their progenitors. With the estimate for the age of
the superbubble this leads to a SN rate of 0.001 SN yr$^{-1}$ for the
central stellar cluster. This is a reasonable number given the fact
that the SN rate in the central part of M\,82 was estimated to be
about 0.1 SN yr$^{-1}$ (Kronberg {\rm et\ts al.}\ \cite{kronberg81}). The
total \ifmmode{{\rm H}_2}\else{H$_2$}\fi\ mass of the superbubble is about $8\times 10^6\,\ifmmode{{\rm M}_{\odot}}\else{M$_{\odot}$}\fi$.
Less than 15\% of the total energy is still present in the hot gas
which emerges from the superbubble; the fraction of kinetic
energy of the molecular superbubble is about 10\%
(${\rm E}_{\rm kin}\,\approx\,1.6 \times 10^{53}\,$ergs).
It should be noted, however, that the numbers given above are only order of
magnitude estimates.
The wealth of observations presented above indicates that the region
under study represents an unusually active segment of the starburst in
M\,82. The finding of an expanding molecular superbubble in this
particular region supports this view and provides clear evidence for a
footprint of violent SF in the ISM of M\,82. Furthermore the analysis of the
molecular superbubble presents an alternative tool to investigate the
energy release of the central source which drives the X--ray and
contributes to the \ifmmode{{\rm H}\alpha}\else{${\rm H}\alpha$}\fi\ outflow.
\acknowledgements{A.W. and F.W. acknowledge DFG grant III GK--GRK 118/2.
We thank J. Shen and K.Y. Lo for making available their CO data,
J. Kerp for his help on the ROSAT data and the referee,
P. Kronberg, for helpful comments. We acknowledge the IRAM
staff for carrying out the observations and the help provided
during the data reduction.}
|
\section{Introduction}
We present numerical results for the effect of shear flow on the
spinodal decomposition of a two-dimensional binary fluid using lattice
Boltzmann simulations. We show how the lattice Boltzmann algorithm
can be generalised to allow the introduction of the Lees-Edwards
boundary conditions, which are commonly used in molecular dynamics
simulations to impose a shear flow without introducing walls. Results
are presented showing how the competition between the ordering effects
of the free energy and the disordering effects of the shear influences the
spinodal decomposition and phase ordering of the fluid.
For a recent review see Onuki \cite{onukiref}.
When a binary fluid consisting of an equal amount of two components,
A and B say, is rapidly cooled below the critical temperature
it phase separates into an A-rich and B-rich phase. Once
well-defined domains of each phase are formed the typical domain
size grows according to a power law
\begin{equation}
R(t) \sim t^\alpha
\end{equation}
where $\alpha$ is the growth exponent\cite{bray}.
$\alpha$ depends on the growth mechanism, which is dictated by the
surface tension, viscosity and diffusivity of the fluid, and the
time elapsed after the quench. In two-dimensional systems diffusive
Lifshitz-Slyozov growth gives $\alpha = \frac{1}{3}$ while
hydrodynamics can lead to faster growth with $\alpha = \frac{2}{3}$.
The most obvious effect of shear flow on the domain growth is that
the growing domains are elongated in the direction of the flow,
leading to an anisotropic morphology. Experiments in three dimensions have
shown that a string-like phase of thin domains oriented parallel to
the shear can be formed in strong shear\cite{hashimoto}. Such domains,
which would normally be expected to be unstable due to the
Rayleigh instability, appear to be stabilised by the shear, although
very recent experiments show that they can eventually break
up in strong shear \cite{matsuzaka}.
This apparent stabilization suggests the possibility of a dynamic
equilibrium when stretching and breaking of the domains as the result
of the shear is balanced by their growth due to the thermodynamic
driving force and to the coalescence of the domains, which
can itself be driven by the shear. This was first proposed by Ohta
and Nozaki \cite{otha} on the basis of two-dimensional simulations
using a cell dynamic approach. These simulations, however, did not
include hydrodynamics.
Simulations of phase separation under shear which include
hydrodynamics are limited. Rothman performed early work using lattice
gas cellular automata in two and three dimensions and was able to see
the anisotropy of the growth \cite{olson,rothman2}. Wu {\it et. al.}
undertook Langevin simulations in two and three dimensions and report
the eventual formation of a string phase in three dimensions \cite{skrdla}.
Padilla and Toxvaerd performed molecular dynamics simulations on a
two-dimensional Lennard-Jones system, again pointing out the
anisotropic nature of the domain growth \cite{padilla}. In the simulations a
peak was seen in the excess shear viscosity as a function of time
corresponding to the increase in the lengths of interfaces in the
system. However, there seems to be no evidence for a shear-induced
dynamic equilibrium.
Here we simulate phase separation under shear using a lattice
Boltzmann scheme in the same spirit as the model introduced by
Orlandini {\it et. al.}, which imposes phase separation by defining
the fluid equilibrium as the minimum of an input free energy
\cite{enzo,swift2}. This method has been very successful in obtaining
results for phase separation in the absence of shear \cite{a_scaling}.
A particular advantage of the approach is that the fluid viscosity and
diffusivity can be tuned, and this has allowed us to compare
simulations for parameter values where diffusive or hydrodynamic phase
separation dominates. We find either phases striped in the shear
direction, or a dynamic equilibrium where the length scales remain
approximately constant in time, depending on the relative strengths of
the shear and the ordering.
The lattice Boltzmann approach is described in \S 2. Because
this is a lattice rather than particulate simulation
method, it is not immediately obvious how to define Lees-Edwards
shear boundary conditions. An approach for doing this is given
in \S 3. In particular it is necessary to generalize the
normal definition of the lattice Boltzmann equilibrium distribution.
In \S 4 we define suitable measures to characterise the
anisotropic morphology of the spinodal decomposition patterns
when shear is applied. The results of our simulations are
contained in \S 5, where the effect of shear is compared for
different fluid viscosities. \S 6 summarises the
results and discusses outstanding questions.
\section{The Lattice Boltzmann Approach}
The starting point for lattice Boltzmann simulations\cite{doolen}
is the evolution
equation, discrete in space and time, for a set of distribution
functions, ${f_i}$, each associated with a velocity vector, ${\bf v}_i$. For
the sake of simplicity we consider a single relaxation time, the
so-called BGK approximation
\cite{quian}. The evolution equation for the
$\{f_i\}$ is
\begin{equation}
f_i({\bf x}+ {\bf v}_i \Delta t, t+\Delta t) -f_i({\bf x},t)=
\frac{\Delta t}{\tau_1} (f_i^0-f_i),
\label{LBE1}
\end{equation}
where ${\bf x}$ is a lattice point, $\Delta t$ is the time step, and
${\bf v_i} \Delta t$ is normally constrained to be a lattice
vector. The relaxation time is $\tau_1$ and $f_i^0$ is the
equilibrium distribution. For a two-component system a second,
equivalent equation is also needed
\begin{equation}
g_i({\bf x}+ {\bf v}_i \Delta t, t+\Delta t) -g_i({\bf x},t)=
\frac{\Delta t}{\tau_2} (g_i^0-g_i).
\label{LBE2}
\end{equation}
Physical quantities are defined as moments of the distribution
functions. To model the isothermal flow of a binary mixture of
components A and B, we choose
\begin{equation}
\sum_i f_i = n,\;\;\;\;\;
\sum_i f_i {\bf v}_i = n {\bf u},\;\;\;\;\;
\sum_i g_i = \varphi,
\label{eqn3}
\end{equation}
where $n$ is the total density field, ${\bf u}$ is the velocity field
and $\varphi$ is the field corresponding to the difference in the
density of components A and B.
We require mass conservation for both components and momentum
conservation for the bulk. This is equivalent to
constraining the equilibrium distributions to obey
\begin{equation}
\sum_i f_i^0 = n, \;\;\;\;\;
\sum_i g_i^0 = \varphi,\;\;\;\;\;
\sum_i f_i^0 {\bf v}_i = n {\bf u}.
\label{conserve_nu}
\end{equation}
We also need to define higher-order moments of the equilibrium
densities. The choice for these moments is within the free energy
lattice Boltzmann scheme used here\cite{enzo,swift2}
\begin{eqnarray}
\sum_i f_i^0 v_{i\alpha} v_{i\beta} &=& P_{\alpha\beta} + nu_\alpha
u_\beta, \label{cons1}\\
\sum_i g_i^0 v_{i\alpha} &=& \varphi u_\alpha,\\
\sum_i g_i^0 v_{i\alpha} v_{i\beta} &=& \Gamma \mu \delta_{\alpha\beta}
+ \varphi u_\alpha u_\beta, \label{cons3}
\end{eqnarray}
where $P_{\alpha \beta}$ is the pressure tensor, $\Gamma$ is a mobility
parameter, $\mu$ is the chemical potential for the density difference
and $\delta$ is the Kronecker delta. The physical motivation for these
constraints is twofold; firstly to ensure the correct form of the
macroscopic equations of motion and secondly to reproduce the correct
thermodynamics of the
binary mixture in equilibrium as discussed in more detail below.
Taylor-expanding the evolution equations (\ref{LBE1}) and (\ref{LBE2})
to second order in the derivatives gives the macroscopic equations of
motion for the binary fluid\cite{thesis}. These are the continuity
equation for the
total density
\begin{equation}
\partial_t n+\partial_\alpha n u_\alpha=0,
\label{cont_n}
\end{equation}
a convection-diffusion equation governing the evolution of the density
difference
\begin{equation}
\partial_t \varphi+\partial_\alpha (\varphi u_\alpha) =
\omega_2
\left( \Gamma \nabla^2 \mu- \partial_\beta \left(
\frac{\varphi}{n} \partial_\alpha P_{\alpha\beta} \right)\right),
\label{cdeq}
\end{equation}
and, in the incompressible limit,
the incompressible Navier Stokes equations for a non-ideal
system
\begin{equation}
n \partial_t u_\alpha+n u_\beta \partial_\beta u_\alpha =
-\partial_\beta P_{\alpha\beta}
+ \frac{n \omega_1}{3}
\nabla^2 u_\alpha
+ O(\partial^3)
\label{NS}
\end{equation}
where $\omega_{1,2}=\tau_{1,2}-\Delta t/2$ and
the viscosity is given by $\nu=n\omega_1/3$.
The thermodynamic fields entering the simulation are the
pressure tensor and the chemical potential which follow from the free
energy of the system.
We consider the free energy of a simple binary fluid.
A--A and B--B interactions are zero, but there is an A--B
repulsion $\lambda n_A n_B$ where $n_A$ and $n_B$ are the number
densities of A- and B-particles, respectively, and $\lambda$ is a
parameter describing the interaction strength. This system can be
described by the Landau free energy functional
\begin{equation}
\Psi = \int d{\bf r} \left\{ \psi(\varphi,n,T)
+ \frac{\kappa}{2} (\nabla \varphi)^2\right\} \label{totf}
\end{equation}
where $T$ is the temperature
and $\kappa$ is a measure of the excess interface free energy (surface
tension). The free energy density of the homogeneous system is
\cite{reichel}
\begin{eqnarray}
\psi(\varphi,n,T)&=&\frac{\lambda
n}{4}\left(1-\frac{\varphi^2}{n^2}\right) -Tn \nonumber\\
&&+\frac{T}{2}(n+\varphi) \ln\left(\frac{n+\varphi}{2}\right) \nonumber\\
&&+\frac{T}{2}(n-\varphi) \ln\left(\frac{n-\varphi}{2}\right). \label{bulkf}
\end{eqnarray}
For temperatures greater than a critical temperature $T_c=\lambda /2$
the system remains in a single phase. For
$T < T_c$ there is phase separation into two states with
$\varphi = \pm \varphi_0$.
From the free energy (\ref{totf}) we derive the local chemical
potential $\mu$ as the functional derivative of the total free energy
$\Psi$ with respect to the concentration difference field $\varphi({\bf
x})$
\begin{equation}
\mu({\bf x}) = \frac{\delta \Psi}{\delta \varphi({\bf x})}
= -\frac{\lambda}{2} \frac{\varphi}{n} + \frac{T}{2} \ln\left(
\frac{n+\varphi}{n-\varphi} \right) - \kappa \nabla^2 \varphi.
\end{equation}
Equilibrium corresponds to $\mu(\phi,n,T)=0$.
The derivation of the pressure tensor is slightly more involved and
is discussed in Appendix A\cite{yang}. We obtain
\begin{eqnarray}
P_{\alpha\beta}&=&
(n \partial_n \psi+ \varphi \partial_\varphi \psi -\psi)
\delta_{\alpha\beta} \nonumber\\
&&+ \kappa ( \partial_\alpha \varphi
\partial_\beta \varphi - \frac{1}{2} \partial_\gamma \varphi
\partial_\gamma \varphi \delta_{\alpha\beta} - \varphi \partial_\gamma
\partial_\gamma \varphi) \nonumber\\
&=&
(nT + \varphi \mu^0(x))\delta_{\alpha\beta}\nonumber\\
&&+ \kappa ( \partial_\alpha \varphi
\partial_\beta \varphi - \frac{1}{2} \partial_\gamma \varphi
\partial_\gamma \varphi \delta_{\alpha\beta} - \varphi \partial_\gamma
\partial_\gamma \varphi)\nonumber\\
&=& (nT + \varphi \mu(x)) \delta_{\alpha\beta}\nonumber\\
&&+\kappa (\partial_\alpha
\varphi \partial_\beta \varphi-\frac{1}{2} \partial_\gamma \varphi
\partial_\gamma \varphi \delta_{\alpha\beta})
\label{e:pressuretensor}
\end{eqnarray}
where the first term is the ideal gas pressure, the second term is the
osmotic pressure with $\mu^0=\partial_\phi \psi$ and the third term is
related to the surface tension. The osmotic
pressure was omitted in the original definition of the model
\cite{enzo,swift2}. The chemical potential and pressure tensor are
input to the lattice Boltzmann scheme through equations (\ref{cons1})
and (\ref{cons3}).
In equilibrium the simulated fluid minimises the free energy (\ref{totf}).
It remains only to define the equilibrium distributions $f_i^0$ and
$g_i^0$ introduced in the evolution equations (\ref{LBE1}) and
(\ref{LBE2}). Normally an
expansion to second order in the velocities is sufficient to reproduce
the constraints (\ref{conserve_nu}) -- (\ref{cons3})\cite{doolen}.
However, this ceases to be the
case when Lees-Edwards shear boundary conditions are introduced. In
the next section we discuss how the equilibrium distribution can be
defined to allow the use of Lees-Edwards boundary conditions.
\section{Shear boundary conditions}
Possibly the easiest way to introduce shear flow in a lattice Boltzmann
simulation is to include walls moving in a lattice direction. Even for
a wall with neutral wetting, however, phase separation is
strongly enhanced at the walls and the wall effects easily dominate
the phase separation process for all but the largest systems. The
effect of walls on phase separation is an interesting phenomenon
in its own right, but it is not the process we are interested in
studying here.
The problem caused by explicit walls can be overcome in a relatively
simple and efficient manner by introducing a Klein-bottle symmetry to
the lattice. This is done by forcing the fluid to have a given
velocity along one line in the direction of the shear flow. In a
one-component mixture this induces a linear velocity profile. For a
two-component mixture, however, the dynamics are influenced by the
V-shaped velocity profile at the forcing line because of the non-local
interactions. We used this algorithm to produce preliminary results
but it has no advantages over the method derived below.
A more regular shear flow can be produced by extending the idea of
Lees-Edwards boundary conditions, widely used in Molecular Dynamics
\cite{lee_edw}, to lattice Boltzmann simulations. Briefly, Lees and
Edwards simulated shear boundary conditions for a shear in the
$x$-direction in a simulation box of dimensions $(X,Y)$ by
introducing periodic boundary conditions in the
$y$-direction. Particles that left the box at the lower boundary for
position $(x,y=0)$ reappeared at the upper boundary at position $(x+ut
(\bmod X),y=Y)$ with a velocity that was changed by $v \rightarrow
v+u$.
To implement this idea for lattice Boltzmann simulations we are faced
with two difficulties. Firstly the densities are defined on a lattice
and the Lees-Edwards boundary conditions lead to densities defined between the
lattice points. Secondly we need to impose a Galilean transformation
for the densities which are streamed across the lattice.
The non-fitting of the lattice is relevant for both the streaming and
for the calculation of derivatives at $y=1$ and $y=Y$. We solve
this problem by a linear interpolation scheme. For any density we
define
\begin{eqnarray}
f[x,y=0]&=&(1-R(ut)) f[x+I(ut),y=Y]\nonumber\\
&& + R(ut) f[x+I(ut)+1,y=Y]
\end{eqnarray}
where $I(z)$ is the largest integer with $I(z)<z$ and
$R(z)=z-I(z)$. If we pass the break in the lattice from the other side
we define similarly
\begin{eqnarray}
f[x,y=Y+1]&=&(1-R(ut)) f[x-I(ut),y=1]\nonumber\\
&& + R(ut) f[x-I(ut)-1,y=1].
\end{eqnarray}
These formulae are used both for the streaming of the
Galilean-transformed Boltzmann densities, $f_i$, and for the calculation
of density gradients.
\begin{figure}
\centerline{\psfig{figure=ninevel.eps,width=5cm}}
\caption{Our numbering of the velocity vectors in a nine-velocity model.}
\label{fig:ninevel}
\end{figure}
It is rather more difficult to see how the Galilean
transformation should be defined. Let us consider the special case of
a two-dimensional,
nine-velocity model where the velocities are numbered as indicated
in Figure \ref{fig:ninevel}. We need to perform a Galilean transformation on
the $\{5,2,6\}$ and the $\{7,4,8\}$ velocities as these
will carry mass and momentum across the boundaries. To define
the transformation we demand mass and $y$-momentum conservation
\begin{equation}
n_p \equiv f_5+f_2+f_6=f_5'+f_2'+f_6'
\end{equation}
an appropriate change in the $x$-momentum
\begin{equation}
(f_5-f_6)-(f'_5-f'_6)=n_p u
\label{udef}
\end{equation}
and conservation of the local pressure
\begin{eqnarray}
P^p_{xx}&=&\sum{\frac{f_i}{{n_p}^2} (n{v_i}_x-n_p u_x)^2}\nonumber\\
&=&\frac{1}{{n_p}^2}\left((f_5(n_p-n_p u_x)^2+f_2(n_p
u_x)^2\right. \nonumber\\
&&\left. + f_6(-n_p-n_p u_x)^2\right)\nonumber\\
&=&\frac{1}{{n_p}^2}\left(f_5'(n_p-n_p u_x')^2+f_2'(n_p u_x')^2\right.
\nonumber\\
&&\left. + f_6'(-n_p-n_pu_x')^2 \right)
\end{eqnarray}
where the prime denotes the transformed quantities.
This system of equations can be solved to give a unique solution for
the Galilean-transformed densities $f_i'$
\begin{eqnarray}
f_2'&=& f_2 + 2 (f_5-f_6) u - n_p u^2,\label{cond1}\\
f_5'&=& f_5 + (-\frac{3}{2} f_5-\frac{1}{2}f_2+\frac{1}{2}f_6) u +
\frac{n_p}{2} u^2,\\
f_6'&=& f_6 + (-\frac{1}{2}f_5 + \frac{1}{2} f_2+
\frac{3}{2} f_6)u + \frac{n_p}{2} u^2.\label{cond3}
\end{eqnarray}
This definition can be extended to a Galilean transformation
for all densities and, equivalently, to a transformation in different lattice
directions.
In order for this transformation to make sense we need to make sure
that equation (\ref{udef}) is consistent with the definition of the
equilibrium distribution, $f_i^0$ in Eqn.~(\ref{LBE1}) {\it i.e.}
that an equilibrium distribution
for a velocity $u$ Galilean transformed by a velocity $\Delta u$ is
equal to the equilibrium distribution for velocity $u+\Delta u$. It is
conventional to define the equilibrium distribution as a polynomial in
second order in u. A generic expansion is
\begin{eqnarray}
f_i^0&=&A_\sigma n+B_\sigma n u_\alpha {v_i}_\alpha + C_\sigma n u^2
\nonumber\\
&&+
D_\sigma n u_\alpha u_\beta {v_i}_\alpha {v_i}_\beta+ G_{\sigma\alpha\beta}
n {v_i}_\alpha {v_i}_\beta
\label{equil}
\end{eqnarray}
where $A_\sigma,B_\sigma,C_\sigma,D_\sigma,G_{\sigma\alpha\beta}$ are
constants that have the absolute value of the corresponding velocity
vector $\sigma=|{\bf v}_i|$ as an index. However, substituting
(\ref{equil}) into (\ref{udef}) shows that this equation is not
satisfied in equilibrium. In practice this leads to a step in the
$u_x$ profile at the boundary.
There is, however, no {\it a priori} reason to use a second-order
expansion in the velocity for the equilibrium distribution. All that
is needed for a valid equilibrium distribution is that
(\ref{conserve_nu})--(\ref{cons3}) hold and that the distribution
obeys the conditions (\ref{cond1})--(\ref{cond3}).
Let $T_{\alpha\beta}=P_{\alpha\beta}/n$. Then, if we require,
\begin{equation}
\frac{f^0_1-f^0_3}{f^0_1+f^0_0+f^0_3}=u_x,\;\;\;\;\;\;\;\;\;\;\;
\frac{f^0_2-f^0_4}{f^0_2+f^0_0+f^0_4}=u_y
\label{ccond}
\end{equation}
\begin{eqnarray}
f^0_5+f^0_6-(f^0_5+f^0_6+f^0_2) (T_{xx}+u_x u_x)=0,\\
f^0_5+f^0_8-(f^0_5+f^0_8+f^0_1) (T_{yy}+u_y u_y)=0,\\
f^0_8+f^0_7-(f^0_8+f^0_7+f^0_4) (T_{xx}+u_x u_x)=0,\\
f^0_6+f^0_7-(f^0_6+f^0_7+f^0_3) (T_{yy}+u_y u_y)=0. \label{clast}
\end{eqnarray}
(\ref{ccond})-(\ref{clast}), together with (\ref{conserve_nu})--(\ref{cons3})
are a completely determined set of equations
with the solution
\begin{eqnarray}
f^0_0&=&n (1- T_{xx} -u_x^2)(1-T_{yy}-u_y^2),\nonumber\\
f^0_1&=&\frac{1}{2} n ( T_{xx}+u_x +u_x^2)(1-T_{yy}-u_y^2),\nonumber\\
f^0_2&=&\frac{1}{2} n ( T_{yy}+u_y +u_y^2)(1-T_{xx}-u_x^2),\nonumber\\
f^0_3&=&\frac{1}{2} n ( T_{xx}-u_x +u_x^2)(1-T_{yy}-u_y^2),\nonumber\\
f^0_4&=&\frac{1}{2} n ( T_{yy}-u_y +u_y^2)(1-T_{xx}-u_x^2),\nonumber\\
f^0_5&=&\frac{1}{4} n ( T_{xy} + T_{xx} T_{yy} +T_{yy} (u_x+u_x^2)
\nonumber\\&& +
T_{xx} (u_y+u_y^2)+ u_x u_y (1+u_x+u_y+u_x u_y)),\nonumber\\
f^0_6&=&\frac{1}{4} n (-T_{xy} + T_{xx} T_{yy} +T_{yy} (-u_x+u_x^2)
\nonumber\\&& +
T_{xx} (u_y+u_y^2)- u_x u_y (1-u_x+u_y-u_x u_y)),\nonumber\\
f^0_7&=&\frac{1}{4} n ( T_{xy} + T_{xx} T_{yy} +T_{yy} (-u_x+u_x^2)
\nonumber\\&& +
T_{xx} (-u_y+u_y^2)+ u_x u_y (1-u_x-u_y+u_x u_y)),\nonumber\\
f^0_8&=&\frac{1}{4} n (-T_{xy} + T_{xx} T_{yy} +T_{yy} (u_x+u_x^2)
\nonumber\\&& +
T_{xx} (-u_y+u_y^2)- u_x u_y (1+u_x-u_y-u_x u_y)).\nonumber
\end{eqnarray}
For this equilibrium distribution
\begin{equation}
\frac{f^0_5-f^0_6}{f^0_5+f^0_2+f^0_6}=u_x+\frac{T_{xy}}{T_{yy}+u_y+u_y^2}
\end{equation}
which is consistent with the Galilean transformation (\ref{udef}).
For a two-component system we similarly define the $g_i^0$ using
\begin{equation}
\frac{g^0_1-g^0_3}{g^0_1+g^0_0+g^0_3}=u_x,\;\;\;\;\;\;\;\;\;
\frac{g^0_2-g^0_4}{g^0_2+g^0_0+g^0_4}=u_y
\label{cg1}
\end{equation}
and imposing
\begin{equation}
g_0 =\varphi- \ell \Gamma \mu -\varphi (u_x^2+u_y^2)\label{cge}
\end{equation}
where $\ell$ is a free parameter that can be used to improve stability
(we choose $\ell=1$). Solving equations (\ref{cg1}) and (\ref{cge})
and (\ref{conserve_nu})--(\ref{cons3}) gives
\begin{eqnarray}
g^0_1 &=&1/2 \{(\ell-1-u_x)\Gamma\mu+(1+u_x-u_y^2)\varphi u_x\},\nonumber\\
g^0_2 &=&1/2 \{(\ell-1-u_y)\Gamma\mu+(1+u_y-u_x^2)\varphi u_y\},\nonumber\\
g^0_3 &=&1/2 \{(\ell-1+u_x)\Gamma\mu-(1-u_x-u_y^2)\varphi u_x\},\nonumber\\
g^0_4 &=&1/2 \{(\ell-1+u_y)\Gamma\mu-(1-u_y-u_x^2)\varphi u_y\},\nonumber\\
g^0_5 &=&1/4 \{(2-\ell+u_x+u_y)\Gamma\mu+(1+u_x+u_y) \varphi u_x u_y\},\nonumber\\
g^0_6 &=&1/4 \{(2-\ell-u_x+u_y)\Gamma\mu-(1-u_x+u_y) \varphi u_x u_y\},\nonumber\\
g^0_7 &=&1/4 \{(2-\ell-u_x-u_y)\Gamma\mu+(1-u_x-u_y) \varphi u_x u_y\},\nonumber\\
g^0_8 &=&1/4 \{(2-\ell+u_x-u_y)\Gamma\mu-(1+u_x-u_y) \varphi u_x u_y\}.\nonumber
\end{eqnarray}
The macroscopic flow equations are unaffected by the choice of the further constraints
(\ref{ccond})--(\ref{clast}) and (\ref{cg1})--(\ref{cge}) or by
the detailed structure of the equilibrium distributions. Therefore,
these alterations in the model can change the numerical stability and the
behaviour of quantities like the spurious velocities, but they
leave the evolution of the macroscopic quantities unaffected, at least
to second order in the derivatives.
\section{Measures for non-isotropic patterns}
To characterise the features of phase separation under shear it is
necessary to construct measures for the length scales of the
sheared systems which will in general be anisotropic.
Measures that are based on Fourier
transforms cannot be easily used for sheared systems because the
system is no longer periodic.
Length scales derived from derivatives do not require periodicity.
Derivatives need to be evaluated for the algorithm and are
readily available. We define a tensor
\begin{equation}
d_{\alpha \beta}
= \frac{\sum_{\bf x} \partial_\alpha^D \varphi({\bf x},t)
\partial_\beta^D \varphi({\bf x},t)}{
\sum_{\bf x} \varphi^2({\bf x},t)}\label{eqn:d}
\end{equation}
where $\partial_\alpha^D$ is the symmetric discrete derivative in
direction $\alpha$. Because the tensor is symmetric it can be
diagonalised to give two eigenvalues $\lambda_1,\lambda_2$ and an
angle $\theta^\star$
\begin{eqnarray}
\lambda_1 &=& \frac{d_{xx}+d_{yy}}{2} +
\sqrt{\frac{(d_{xx}-d_{yy})^2}{4}+d_{xy}^2},\\
\lambda_2 &=& \frac{d_{xx}+d_{yy}}{2} -
\sqrt{\frac{(d_{xx}-d_{yy})^2}{4}+d_{xy}^2},\\
\theta^\star &=& \tan^{-1}\left(\frac{d_{yy}}{d_{xy}-\lambda_2}\right).
\label{tetd}
\end{eqnarray}
The two eigenvalues give two orthogonal length
scales
\begin{equation}
R^{\star}_1(t)=\frac{1}{\lambda_1(t)L_w},\;\;\;\;\;\;\;\; \label{rd1}
R^{\star}_2(t)=\frac{1}{\lambda_2(t)L_w},
\end{equation}
where $L_w$ is the interface width. It appears because
$d_{\alpha\beta}$ scales inversely with the interface
width\cite{thesis}.
$L_w$, used as a constant here, could in principle be anisotropic. That this
anisotropy is not a strong effect can be seen by comparing these
length scales with scales that are explicitly independent of the
interface width.
One such measure is related to the lengths of the
interfaces in the system. The interface can be represented
by a set of contours. These contours consist of small line
segments $\vec{l}_i$ and the length of the interface can be written
\begin{equation}
L_I=\sum_i |\vec{l}_i|.
\end{equation}
In order to extract the preferred direction of the interface we define
the vector
\begin{equation}
\vec{D} = R^{-1}\left(\sum_i R(\vec{l}_i)\right).
\end{equation}
The operator $R$ is defined by
\begin{equation}
R(\vec{x})=|\vec{x}| \left(
\begin{array}{c}
\cos(2 \theta)\\
\sin(2\theta)
\end{array}\right)
\end{equation}
where $\theta$ is the angle between
the argument of R and the $x$-axis.
$\vec{D}$ is a vector that is zero for isotropic closed contours
and which points in the average direction of the interface for
non-isotropic closed contours. Two length scales and an angle
that correspond to the intuitive result for
oriented rectangular objects can be defined from these measures
\begin{equation}
R^{\circ}_1=\frac{XY}{L_I+|\vec{D}|} \label{rc1},\;\;\;\;\;\;\;\;
R^{\circ}_2=\frac{XY}{L_I-|\vec{D}|},
\end{equation}
\begin{equation}
\theta^{\circ} =
\cos^{-1}\left(\frac{\hat{\vec{x}}.\vec{D}}{|\vec{D}|}\right).
\label{tetl}
\end{equation}
Thus we have defined two independent sets of measures for the structure of
non-isotropic patterns that will now be used to examine spinodal
decomposition under shear.
\section{Simulation results}
For all the simulations we used a total density $n=2$, an
interaction parameter $\lambda=1.1$, which corresponds to a critical
temperature $T_c=0.55$, and a temperature $T=0.5$. The equilibrium
values of the order parameter were then $\varphi_0=\pm 1$. The mobility was
$\Gamma=2$, the relaxation time for the order parameter in
Eqn.~(\ref{LBE2})
was $\tau_2=1$ and the interface free energy parameter was
$\kappa=0.002$, which corresponds to an interface width of
approximately three lattice spacings. The relaxation parameter for the
total density Eqn.~(\ref{LBE1}), $\tau_1$, was varied: $\tau_1=100$
gave a high viscosity and $\tau_1=1$ an intermediate viscosity.
The shear transformation, S, is defined as
\begin{equation}
S \left( \begin{array}{c}x \\y \end{array} \right) =
\left( \begin{array}{r}x+\dot{\gamma} t y \\y \end{array} \right).
\end{equation}
Shear flow applied to a system undergoing spinodal decomposition
stretches the original pattern. This effect is only relevant once the
deformation caused by the flow is of the same order or larger than the
deformation caused by the coarsening process. This requires
\begin{equation}
\dot{\gamma} t > 1 \label{sh1}.
\end{equation}
We therefore expect to observe the effect of the shear flow
for $t>1/\dot{\gamma}$.
\begin{figure}
\begin{center}
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear0.eps,width=3cm}}
\begin{center} time=0 \end{center}
\end{minipage} \
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear1.eps,width=3cm}}
\begin{center} time=1 \end{center}
\end{minipage} \
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear2.eps,width=3cm}}
\begin{center} time=2 \end{center}
\end{minipage} \
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear3.eps,width=3cm}}
\begin{center} time=3 \end{center}
\end{minipage} \\
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear4.eps,width=3cm}}
\begin{center} time=4 \end{center}
\end{minipage} \
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear5.eps,width=3cm}}
\begin{center} time=5 \end{center}
\end{minipage} \
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear6.eps,width=3cm}}
\begin{center} time=6 \end{center}
\end{minipage} \
\begin{minipage}[t]{3cm}
\centerline{\psfig{figure=shear10.eps,width=3cm}}
\begin{center} time=10 \end{center}
\end{minipage}
\end{center}
\caption{Applying shear (with shear rate $\dot{\gamma}=1$) to a system
without internal dynamics leads to homogenization.}
\label{fig:shear}
\end{figure}
To help understand the effect of shear-flow on a phase-separating
system let us first consider a pattern without any internal dynamics
that undergoes a shear transformation. This
transformation is illustrated in Figure \ref{fig:shear}, where we
start from a frozen spinodal decomposition pattern and show successive
iterations of a shear transformation with $\dot{\gamma}=1$.
\begin{figure}
\begin{minipage}{8cm}
\centerline{\psfig{figure=small_sd128x256G0.004t100.eps,width=8cm}}
\begin{center} (a)
\end{center}
\end{minipage}
\begin{minipage}[c]{9.5cm}
\begin{minipage}[c]{4cm}
\begin{minipage}{4cm}
\begin{minipage}[c]{0.5cm}
\small
$\mbox{}\;\;\theta^*$\\ $\mbox{}\;\;\theta^\circ$\\
\vspace{2.5cm}
\end{minipage}\hspace{-0.3cm}\nolinebreak \hspace{0cm}
\begin{minipage}{3.5cm}
\centerline{\psfig{figure=sd128x256G0.004t100.eps,width=3.5cm}}
\end{minipage}
\end{minipage}
\vspace{-0.3cm}
\small \hfill time
\begin{center} (b) \end{center}
\end{minipage}
\begin{minipage}[c]{4cm}
\begin{minipage}{4cm}
\begin{minipage}[c]{0.5cm}
\small
$R^*_{1,2}$\\$R^\circ_{1,2}$\\
\vspace{2.5cm}
\end{minipage}\nolinebreak \hspace{0.1cm}
\begin{minipage}[c]{3.5cm}
\centerline{\psfig{figure=sdlen128x256G0.004t100.eps,width=3.5cm}}
\end{minipage}
\end{minipage}
\vspace{-0.3cm}
\small \hfill time
\begin{center} (c) \end{center}
\end{minipage}
\end{minipage}\\
\caption{(a) Spinodal decomposition under shear for a high viscosity
binary fluid $(\tau_1=100,L_x=256,L_y=128)$. The high viscosity
suppresses internal hydrodynamic degrees of freedom. The shear rate is
$\dot{\gamma}=0.004$ which corresponds to a shear time $t_s=250$. (b)
Variation of the orientation (in degrees) of the pattern with time. (c) Variation
of the length scales (in arbirtary units) with time.} \label{highvisc_highshear}
\end{figure}
The
structure develops an orientation that slowly aligns with the shear
direction while the stretching increases the length of the domains
along the shear. Once the width of the domains is smaller than the
original width of the interface the system is effectively a
homogeneous mixture.
This effect is known as shear-induced mixing. It can be observed in
the lattice Boltzmann fluids if the stretching
effect of the shear flow is much faster
than the growth of the domains via diffusion or flow. Numerically this can be
achieved by choosing a very low mobility and a high viscosity.
Phase separation is
suppressed because of the mixing properties of the shear flow
unless the phase-separating structure is aligned with the shear
direction. For finite lattices we sometimes observe at much later times a
nucleation of complete stripes that span the system and are
periodic in the shear direction. The time required to form these
stripes depends on the system size and it seems reasonable to assume
that this phenomenon does not occur in infinite systems.
We now consider a high viscosity fluid ($\tau_1=100$) in which
diffusive but not hydrodynamic modes are important. The internal
dynamics leads to domain coarsening and can also prevent a complete
mixing of the system. Figure \ref{highvisc_highshear} shows the spinodal decomposition
pattern of the high-viscosity binary mixture. For very short times
($t<300 \sim \dot{\gamma}^{-1}$) we observe the familiar spinodal
decomposition pattern. It is, however, coarsening in a new way via
shear flow-induced collisions of the domains. This process enhances
domains oriented in the collision direction. Then for $300<t<1000$ the
flow slowly turns the striped pattern and stretches it. At $t\sim
1000$ the rupturing of domains starts to be important and for
$1000<t<15000$ there is a continuous stretching and rupturing that
effectively stops the phase ordering process. At $t \sim 15000$ the system
developes stripes that span the system. Because periodic stripes
are unaffected by the shear flow if they are completely aligned with
it the
system now grows via the diffusion mechanism.
This evolution can be followed more quantitatively by measuring the
orientation angle and the length scales defined in Section IV. Figure
\ref{highvisc_highshear}b shows the angle of orientation to the
$x$-axis measured by $\theta^\star$ (Eqn. \ref{tetd}) and
$\theta^\circ$ (Eqn. \ref{tetl}). The two different measures for the
angle agree very well. The pattern tilts at very early times
($t<2000$) and then slowly aligns with the direction of the shear flow
as periodic stripes are created.
The graph in Figure \ref{highvisc_highshear}c shows the
length-scales $R_{1,2}^\star$ defined in Eqns.
(\ref{rd1}) and the length scales $R_{1,2}^\circ$ defined in
Eqns. (\ref{rc1}). We very clearly see a separation of
length scales and a good agreement of the two different measures. A
minimum of the larger length scale at $t\sim 17000$ indicates the
creation of periodic stripes spanning the system. After this time
the growth of domains is no longer hindered by the continual breaking
of stretched domains.
We now turn to consider a system with a lower viscosity that allows for a
hydrodynamic response of the domains to the shear flow. Results are
presented in Figure \ref{lowvisc_highshear}. It is
immediately obvious that the pattern differs
\begin{figure}
\begin{minipage}{8cm}
\centerline{\psfig{figure=phase_sd128x256G0.004t1.eps,width=8cm}}
\begin{center} (a)
\end{center}
\end{minipage}
\begin{minipage}[c]{9cm}
\begin{minipage}[c]{4cm}
\begin{minipage}{4cm}
\begin{minipage}[c]{0.5cm}
\small
$\mbox{}\;\;\;\theta^*$\\ $\mbox{}\;\;\;\theta^\circ$
\vspace{2.5cm}
\end{minipage}\nolinebreak \hspace{0cm}
\begin{minipage}{3.5cm}
\centerline{\psfig{figure=sd128x256G0.004t1.eps,width=3.5cm}}
\end{minipage}
\end{minipage}
\vspace{-0.3cm}
\small \hfill time
\begin{center} (b) \end{center}
\end{minipage}
\begin{minipage}[c]{4cm}
\begin{minipage}{4cm}
\begin{minipage}[c]{0.5cm}
\small
$R^*_{1,2}$\\$R^\circ_{1,2}$
\vspace{2.5cm}
\end{minipage}\nolinebreak \hspace{0.1cm}
\begin{minipage}[c]{3.5cm}
\centerline{\psfig{figure=small_sdlen128x256G0.004t1.eps,width=3.5cm}}
\end{minipage}
\end{minipage}
\vspace{-0.3cm}
\small \hfill time
\begin{center} (c) \end{center}
\end{minipage}
\end{minipage}\\
\caption{(a) Spinodal decomposition under shear for a medium
viscosity binary fluid $(\tau_1=1,L_x=256,L_y=128)$. The effect of internal
flow causes the domains
to remain at an angle to the shear direction. The shear rate is
$\dot{\gamma}=0.004$ which corresponds to a shear time $t_s=250$. (b)
Variation of the orientation (in degrees) of the pattern with time. (c) Variation
of the length scales (in arbitrary units) with time.} \label{lowvisc_highshear}
\end{figure}
\noindent from that in Figure
\ref{highvisc_highshear}. The final state does not simply
consist of periodic stripes, but of dynamic structures that are constantly
stretched, broken and deformed by the flow. At least on this time
scale a state of dynamic
equilibrium is reached where the ordering effects of the spinodal
decomposition balance the disordering effects of the shear.
The quantitative measures in Figures \ref{lowvisc_highshear}b and \ref{lowvisc_highshear}c
show that after initial fluctuations the orientation of
the pattern converges to a value that fluctuates about a
finite angle to the shear direction. This phenomenon is similar to the
behaviour of a single sheared drop that lies at a finite angle
to a shear flow\cite{stone}.
The graph of length scales again shows a very clear
distinction between the large and small length scales.
Strong oscillations are seen. These may be finite size effects
because the system is so
small and contains only a few domains.
However such oscillations have been seen in experiments \cite{matsuzaka} and ina model system \cite{corberi}.
\begin{figure}
\begin{minipage}{8cm}
\centerline{\psfig{figure=sd512G0.0001.eps,width=8cm}}
\begin{center} (a)
\end{center}
\end{minipage}
\begin{minipage}[c]{9cm}
\begin{minipage}[c]{4cm}
\begin{minipage}{4cm}
\begin{minipage}[c]{0.5cm}
\small
$\mbox{}\;\;\theta^*$\\ $\mbox{}\;\;\theta^\circ$\\
\vspace{2.5cm}
\end{minipage}\hspace{-0.3cm}\nolinebreak \hspace{0cm}
\begin{minipage}{3.5cm}
\centerline{\psfig{figure=sd512G0.0001.ang.eps,width=3.5cm}}
\end{minipage}
\end{minipage}
\vspace{-0.3cm}
\small \hfill time
\begin{center} (b) \end{center}
\end{minipage}
\begin{minipage}[c]{4cm}
\begin{minipage}{4cm}
\begin{minipage}[c]{0.5cm}
\small
$R^*_{1,2}$\\$R^\circ_{1,2}$\\
\vspace{2.5cm}
\end{minipage}\nolinebreak \hspace{0.1cm}
\begin{minipage}[c]{3.5cm}
\centerline{\psfig{figure=sd512G0.0001.lengr.eps,width=3.5cm}}
\end{minipage}
\end{minipage}
\vspace{-0.3cm}
\small \hfill time
\begin{center} (c) \end{center}
\end{minipage}
\end{minipage}\\
\caption{(a) Spinodal decomposition under shear for a high
viscosity binary fluid $(\tau_1=100,L_x=512,L_y=512)$.
The high viscosity suppresses internal
hydrodynamic degrees of freedom. The shear rate is
$\dot{\gamma}=0.0001$ which corresponds to a shear time $t_s=1000$. (b)
Variation of the orientation (in degrees) of the pattern with time. (c) Variation
of the length scales (in arbitrary units) with time.}\label{lowvisc_lowshear}
\end{figure}
We have, so far, considered strong shear flow. Let us now consider
the same viscosity, where both diffusive and hydrodynamic flow is
possible, but lower the shear rate so that the early time
spinodal decomposition is unaffected by the flow. In
Figure \ref{lowvisc_lowshear} the spinodal decomposition for a
shear rate $\dot{\gamma}=0.0001$ is shown for a system with
$\tau_1=1$. For times
$t<1/\dot{\gamma}=10000$ we see the typical spinodal decomposition
pattern for these viscosities.
Hydrodynamic flow leads to circular domains which then grow through
the slower diffusive mechanism. After this
time, the stretching of the domains dominates over the domain
growth and the pattern becomes non-isotropic. By $t\sim 10000$ the
pattern comprises large-stripe like domains together with the nested
pattern of drops within drops in the large domains.
As the
large domains are stretched, the drops inside them coalesce with the
walls and slowly the stripes are cleaned of the small
included drops.
These results also clearly show up in the measurements given in Figure
\ref{lowvisc_lowshear}.
After $t>10000$ the orientation slowly
converges towards a tilting angle $\theta\sim 7^\circ$, the long and
short length scales split and the $R^\star\sim R^\circ \sim R^1 \sim
t^\frac{2}{3}$ growth law breaks down. In the $R^\#$ measure
derived from the number of domains we see a slight increase from the
normal growth law corresponding to the process of shear cleaning the
stripes from drops.
\section{Conclusions} \label{s:outlook}
In this paper we have investigated the effects of shear flow on
systems undergoing spinodal decomposition. In order to study these
systems we introduced an extension to the lattice Boltzmann
algorithm that allows
simulation of shear flow problems with Lees-Edwards boundary
conditions. We find that the effect of shear flow on spinodal
decomposition depends strongly on the viscosity of the fluid. Systems
with a very high viscosity tend to order in the shear direction,
whereas systems with a lower viscosity arrive at a dynamic
stationary state where the domains lie at a finite angle to the shear
direction.
One of the problems in simulating spinodal decomposition
under shear is that the shear flow induces long-range
correlations much faster than for un-sheared systems so
that larger lattice sizes are required to examine long-time
behaviour.
Therefore there remain many unexplored problems concerning the structure of
spinodal decomposition under shear. For example,
it would be interesting to investigate the
transition between the sheared and non-sheared patterns for different
viscosities and to ask whether the
late-time decomposition patterns are statistically
independent of an initial shear.
\section*{Appendix A}
\label{c:pressuretensor}
We show how the full pressure tensor
(\ref{e:pressuretensor}) is derived.
The pressure of a homogeneous system is defined as the volume
derivative of the free energy. Writing the full volume dependence of
the densities $n=N/V$ and $\varphi=(N_A-N_B)/V$ explicitly we see that:
\begin{eqnarray}
P &=& -\partial_V \int_V
\psi\left(\frac{N}{V},\frac{N_A-N_B}{V}\right)
\nonumber\\
&=&-\partial_V \left( V \psi\left(\frac{N}{V},\frac{N_A-N_B}{V}\right)
\right)
\nonumber\\
&=&n \partial_n \psi +\varphi \partial_\varphi \psi - \psi.
\end{eqnarray}
For a non-homogeneous system the pressure is no longer a scalar but a
tensor. The correct form of the pressure tensor can be derived from a
Lagrangian expression for the free energy which is minimized in
equilibrium
\begin{eqnarray}
L &=& \int_V \left(\psi(n,\varphi)
+ \frac{\kappa}{2} \partial_\alpha \varphi
\partial_\alpha \varphi\right) \nonumber\\
&&+ \mu_\varphi (\int_V \varphi - (N_A-N_B)) +
\mu_n (\int_V n - N).
\end{eqnarray}
To obtain differential equations for the equilibrium we
evaluate the Euler-Lagrange equations and get
\begin{eqnarray}
\mu_\varphi &=& -\partial_\varphi \psi + \kappa \partial_\alpha
\partial_\alpha \varphi,\\
\mu_n &=& -\partial_n \psi.
\label{chempot}
\end{eqnarray}
We multiply these equations with $\partial_\beta \varphi$ and
$\partial_\beta n$, respectively and sum the
equations. Remembering that $\mu_\varphi$ and $\mu_n$ are constants,
this yields
\begin{eqnarray}
\partial_\beta (\varphi \mu_\varphi+ n \mu_n) &=& -\partial_\alpha ( \psi
\delta_{\alpha\beta}\nonumber\\&&
+ \kappa \{\partial_\alpha \varphi \partial_\beta
\varphi - \frac{1}{2} \partial_\gamma \varphi \partial_\gamma \varphi
\delta_{\alpha\beta})\}.
\label{chempot1}
\end{eqnarray}
We then substitute the expressions for the chemical
potentials (47) and (48)
into~(\ref{chempot1})
and subtract the right-hand side from the left-hand
side to derive a tensor $\sigma$ that has a zero divergence
\begin{eqnarray}
\partial_\alpha \sigma_{\alpha\beta} &=&
\partial_\alpha ((\varphi \partial_\varphi \psi+ n \partial_n \psi -
\psi) \delta_{\alpha\beta} \nonumber\\&&
+ \kappa ( \partial_\alpha \varphi
\partial_\beta \varphi - \frac{1}{2} \partial_\gamma \varphi
\partial_\gamma \varphi \delta_{\alpha\beta} - \varphi \partial_\gamma
\partial_\gamma \varphi \delta_{\alpha\beta})).
\end{eqnarray}
For a uniform system $\sigma_{\alpha\beta} = P
\delta_{\alpha\beta}$ reduces to the homogeneous pressure. The
divergence of the pressure tensor must vanish in equilibrium. We
therefore identify $\sigma_{\alpha\beta}$ with the pressure tensor
$P_{\alpha\beta}$.
\def\jour#1#2#3#4{{#1} {\bf #2}, #3 (#4)}.
\def\tbp#1{{\em #1}, to be published}.
\def\tit#1#2#3#4#5{{#1} {\bf #2}, #3 (#4)}
\defAdv. Phys.{Adv. Phys.}
\defEuro. Phys. Lett.{Euro. Phys. Lett.}
\defPhys. Rev. Lett.{Phys. Rev. Lett.}
\defPhys. Rev.{Phys. Rev.}
\defPhys. Rev. A{Phys. Rev. A}
\defPhys. Rev. B{Phys. Rev. B}
\defPhys. Rev. E{Phys. Rev. E}
\defPhysica A{Physica A}
\defPhysica Scripta{Physica Scripta}
\defZ. Phys. B{Z. Phys. B}
\defJ. Mod. Phys. C{J. Mod. Phys. C}
\defJ. Phys. C{J. Phys. C}
\defJ. Phys. Chem. Solids{J. Phys. Chem. Solids}
\defJ. Phys. Cond. Mat{J. Phys. Cond. Mat}
\defJ. Fluids{J. Fluids}
\defJ. Fluid Mech.{J. Fluid Mech.}
\defAnn. Rev. Fluid Mech.{Ann. Rev. Fluid Mech.}
\defProc. Roy. Soc.{Proc. Roy. Soc.}
\defRev. Mod. Phys.{Rev. Mod. Phys.}
\defJ. Stat. Phys.{J. Stat. Phys.}
\defPhys. Lett. A{Phys. Lett. A}
|
\section{Introduction}
This paper is a continuation of our series of papers on the
investigation of stars which are believed to be in the
post--Asymptotic Giant Branch (post--AGB) stage of evolution
(Klochkova~1995; Za\v{c}s et al.~1995, 1996; Klochkova \&
Panchuk~1996a, 1998; Klochkova et al.~1997a; Klochkova \& Mishenina~1998).
The post--AGB stars (hereafter also referred to as
proto--planetary nebulae -- PPNe) being in the transition phase
from AGB to planetary nebulae offer an opportunity to study in
detail a chemical composition which has undergone changes due to
nucleosynthesis and mixing processes in the course of the stars
evolution. Here we present new results for the peculiar
supergiant with a large infrared excess IRAS\,04296+3429.
On the 12/25/60 $\mu$m colour--colour diagram from the IRAS data
infrared source IRAS\,04296$+$3429 (hereafter IRAS\,04296),
associated with a faint carbon--rich star (Omont et al.~1993;
Loup et al.~1993) classified as type G0\,Ia by Hrivnak et
al.~(1994), is located in the region occupied by planetary
nebulae, non--variable OH/IR stars, and proto--planetary nebulae
(Iyengar \& Parthasarathy~1997). The object IRAS\,04296 belongs
to the small group of sources which show a spectral feature
around 21\,$\mu$m (Kwok et al.~1989). This feature is seen only
for some post--AGB objects and has not been detected either in
the preceding (AGB) nor in succeeding (PN) evolution stage. Note
that a search for new 21\,$\mu$m emitters by means of {\it ISO SWS}
observations among candidates selected by Henning et al.~(1996)
failed to give any detections (Henning, private communication).
Using a medium-resolution (3\,\r{AA}) optical spectrum, Hrivnak~(1995)
found that IRAS\,04296 is a strongly reddenned
(E(B-V)\,=\,1.3) G-star with features indicative of high
luminosity, with molecular absorption features
of ${\rm C_2}$ and rarely observed features of
${\rm C_3}$ {\it of circumstellar origin} and quite strong
absorption lines of s-process elements (Ba, Sr, Y) indicating
that outer layers of the atmosphere of IRAS\,04296 have been
enriched by products of nucleosynthesis. Therefore this star is
very well suited for the study of detailed chemical abundances
which have been changed by the third dredge--up.
Indeed, Decin et al.~(1998), using high resolution spectra,
obtained the chemical abundance pattern for this object and
concluded that its metal--deficient, carbon--rich atmosphere has
large overabundances of s--process elements.
In Sect.\,2 we describe our observational material for
IRAS\,04296 and discuss its molecular features, comparing them
with the corresponding spectrum of the Hale--Bopp comet.
Sect.\,3 is devoted to presentation of the main parameters and
detailed analysis of the chemical composition of IRAS\,04296
derived from our optical spectra. The next section presents
modelling of spectral energy distribution for this source with
the aim to get insight into its physical parameters (mainly to
determine the stellar effective temperature which is crucial for
the chemical composition estimation). Finally, in Sect.\,5 we
discuss the results obtained and compare them to the results for
related objects.
\section{Optical spectrum of IRAS\,04296}
\subsection{Observations and spectra reduction}
We have obtained spectra of IRAS\,04296 with the CCD
(1140\,{$\times$}\,1170 pixels) equipped echelle spectrometer
PFES mounted at the prime focus of the 6\,m telescope of SAO RAS
(Panchuk et al.~1998). The echelle-grating with ${\rm
75\,gr/mm}$ and with a blaze angle of ${\rm
\Theta\,=\,64.3\,^{o}}$ was used. A diffracton grating with
${\rm 300\,gr/mm}$ was used as the cross-disperser. The camera
has f\,=\,140\,mm. The projected angular size of the input slit is
0.54\,arc sec.
We observed IRAS\,04296 on October 07, 1996 (JD2450363.6) and
February 26, 1997 (JD2450506.3). The echelle frames with 25
echelle-orders cover the spectral region ${\rm
\lambda\lambda}$\,4420-8300\,\r{AA}. The average spectral
resolution was 0.4\,\r{AA}. The signal-to-noise ratio was in the
range 50--110 for different spectral orders.
All usual procedures needed for echelle-images reduction (bias
subtraction, cosmic ray removal, optimal order extraction,
rebinning) were made using the ECHELLE context of the MIDAS
system. A Gaussian function approximation was made for the
measurement of equivalent widths. The comparison spectrum source
was an argon-filled thorium hollow-cathode lamp.
The distinctive features of the optical spectrum of the source
IRAS\,04296 are a peculiar profile of the ${\rm H_{\alpha}}$
line (see Fig.\,\ref{HI}), molecular emission bands and very strong
absorption lines of ionized atoms of s--process elements
(Y, Zr, Ba, La, Ce, Pr, Nd). For example, the equivalent widths
of Ba\,II lines (6141 and 6496\,\r{AA}) exceed 0.6\,\r{AA}.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f1}}
\caption[]{
The IRAS\,04296+3429 spectrum near ${\rm H_{\alpha}}$.
}
\label{HI}
\end{figure}
\subsection{Emission molecular bands}
Absorptional bands of several molecules (${\rm C_2}$, CN, TiO,
etc.) are often present in the spectra of post-AGB stars (see,
for example, Hrivnak~1995; Bakker et al.~1997). However,
molecular {\it emission} features are only very rarely observed
in the optical spectra of PPNe. One such example is RAFGL\,2688
(the Egg Nebula) for which Crampton et al.~(1975) observed
emission features of the ${\rm C_2}$ molecule in a medium
resolution spectrum. On the other hand, it is well known that
cometary nuclei spectra show prominent Swan band emission.
In both spectra of IRAS\,04296 we have discovered strong {\it
emission} in the (0;0) and (0;1) bands of the Swan system of
the ${\rm C_2}$ molecule. On Figs.~\ref{Swan1}--\ref{Swan3} we
present a comparison between the spectrum of IRAS\,04296
(observed on February, 26, 1997) and that of the Hale-Bopp comet
(observed on March 30, 1997 with the same spectrometer) around
bands (0;1), (0;0) and (1;0), respectively. From
Figs.~\ref{Swan1}--\ref{Swan3} it is clear that emission band
(1;0) at 4735\,\r{AA}\ is absent in the spectrum of IRAS\,04296 while
the bands (0;1) at 5635\,\r{AA}\ and (0;0) at 5165\,\r{AA}\ are reliably
measured.
Hrivnak~(1995) obtained the spectrum of IRAS\,04296
inside the blue spectral region, 3872--4870\,\r{AA}\r{AA},
therefore he could not observe emission features of ${\rm C_2}$ at
5165 and 5635\,\r{AA}\r{AA}.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f2}}
\caption[]{
Comparison of the IRAS\,04296\,+\,3429
${\rm C_2}$ Swan band head (0;1) at ${\rm \lambda}$\,=\,5635\,\r{AA}
with that for the Hale-Bopp comet nuclei
}
\label{Swan1}
\end{figure}
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f3}}
\caption[]{
The same as Fig.\,2 but for ${\rm C_2}$
Swan band head (0;0) at ${\rm \lambda}$\,=\,5165\,\r{AA}
}
\label{Swan2}
\end{figure}
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f4}}
\caption[]{
The same as Fig.\,2 but for ${\rm C_2}$
Swan band head (1;0) at ${\rm \lambda}$\,=\,4735\,\r{AA}}
\label{Swan3}
\end{figure}
To understand the observed ratios between different bands, we
have estimated the temperature function for monochromatic
coefficient of absorption per molecule ($\sigma_{\lambda}$) for
the Swan bands in the ``just overlapping'' approximation (JOA,
Golden~1967). For the microturbulent velocity ${\rm
\xi_t\,=\,7\,km/sec}$ this approximation works well near the
band heads. Values of $\sigma_{\lambda}$ for band heads of (0;0)
at 5165\,\r{AA}\ and (1;0) at 4735\,\r{AA}\ coincide within 0.2\,dex
for the temperatures range 3000--7000\,K, while
$\sigma_{\lambda}$ for the band (0;1) at 5635\,\r{AA}\ is
systemically lower by about 0.6\,dex. Taking into account these
relations between $\sigma_{\lambda}$'s for different band heads
and since we do not observe the 4735\,\r{AA} band in RAFGL\,2688
and IRAS\,04296, we can conclude that it is impossible to
describe the intensity ratios of ${\rm C_2}$ emission bands for
these objects by means of an equilibrium vibrational
temperature in the 3000--7000\,K range.
To explain emission bands intensities for comets the mechanism
of resonance fluorescence has been proposed (Zanstra~1928, Swings~1941).
In that case population of vibration--rotational levels
for the molecule is described by the Boltzmann approximation,
however the value of {\it T} in the exponent no longer has the
meaning of equilibrium temperature but it is a distribution
parameter only. We suggest that the same mechanism could be
responsible for the observed emission bands of IRAS\,04296.
However, it is clear from Figs.~\ref{Swan1}--\ref{Swan3}
that there are significant differences in the equivalent widths
of the emission bands in the spectra of our supergiant and of
the Hale--Bopp comet nucleus. They could be explained by a
difference of radiation fluxes which illuminate ${\rm C_2}$
molecules in these objects. The temperature of IRAS\,04296
($\rm T_{eff}$ around\, 6300\,K) is sufficiently higher than that for
the Sun, therefore the band (1,0) at 4735\,\r{AA} for IRAS\,04296
should be stronger than that for the Hale-Bopp comet nuclei.
However, Fig.\,\ref{Swan3} shows the opposite behaviour. It
could mean that radiation field of IRAS\,04296 which excites the
${\rm C_2}$ molecules is strongly reddened by matter located
between its photosphere and the region which produces the ${\rm
C_2}$ emission.
Together with the emission bands of the Swan system (Klochkova
et al.~1997b) absorption bands of the Phillips system (1:0),
(2;0), (3;0) have been revealed in the spectrum of IRAS\,04296
(Bakker et al.~1997). Let us try to explain this phenomenon
within the resonance fluorescence mechanism ordinary used to
interpret comets' spectra. As a first approximation, we assume that
the vibrational distribution corresponds to the
effective temperature of the star illuminating a circumstellar
envelope if vibrational transitions in the low triplet state of
a homonuclear molecule are strictly forbidden. But even when
interpreting comets' spectra such an approach appears to be too
poor. The intensity distributions for different systems of bands
and for bands of individual systems of the resonance fluorescence
of the ${\rm C_2}$ molecule have been considered in papers by
Krishna Swamy \& O'Dell~(1977, 1979, 1981). The intensities of
bands have been calculated taking into account the excitation of
the Swan, Ballick-Ramsay and Fox-Herzberg triplet systems,
Phillips and Milliken singlet systems as well as singlet-triplet
transitions in low states. It has been shown, in particular,
that at the value of the moment of singlet-triplet transitions
${\rm |R_e|^{2}=10^{-5}}$ and at the heliocentric distance of a
comet d=1\,a.u. the ratios of intensities of sequences ${\rm
\Delta\nu=0,\,1,\,-1}$ in the Phillips system to the intensity
of sequence ${\rm \Delta\nu=0}$ of the Swan system is equal to
0.094, 0.11 and 0.04, correspondingly (Krishna Swamy \& O'Dell~1981).
This agrees well with results of measurement of comets'
spectra. Using these results of Krishna Swamy \& O'Dell~(1981),
we may suppose that the intensity of main bands of the Swan
system is ten times higher than that in the Phillips system.
Now consider the case of IRAS\,04296. Let us add such an emission
spectrum of the ${\rm C_2}$ on the stellar continuum.
In order to observe the emission bands of both the Swan and the
Phillips systems over the continuum in such a combined spectrum,
the stellar flux at ${\rm \lambda}$ = 5165\,\r{AA}
must be at least 10 times
higher than near ${\rm \lambda}$\,=\,7720\,\r{AA}.
From Kurucz's~(1979) tables it follows that the ratio of the fluxes
near these wavelenghts for the Sun (the emitter in the case of comets) is
equal to ${\rm F_{\lambda 5175} / F_{\lambda 7750} = 1.5}$.
For the model with $\rm T_{eff}$ = 6300\,K this ratio is equal to
${\rm F_{\lambda 5175} / F_{\lambda 7750} = 1.9}$.
From the real spectral energy distrubution observed for
IRAS\,04296 (Kwok~1993) the ratio of the fluxes is essentially smaller:
${\rm F_{\lambda 5165} / F_{\lambda 7720} = 0.1}$. Therefore,
the conditions to observe the absorption bands of the Phillips
system and the emission bands of the Swan system may arise
inside the circumstellar envelope of IRAS\,04296.
\subsection{Radial velocity of the IRAS\,04296}
We measured the radial velocity ${\rm V_r}$ using our best
spectrum, that obtained in February, 1997.
The average value of the radial velocity from numerous metal
lines (${\rm V_r\,=\,-56.0\pm 0.8\,km/s}$) and from the
$\rm H_{\alpha}$ absorption line (${\rm V_r\,=\,-54.5\,km/s}$) agree
within the accuracy of measurement.
This radial velocity is consistent with a membership
of an old population as suggested by the low
metallicity (see Table\,1). It should be noted also that the
value of radial velocity we derived agrees with the value of
${\rm V_r\,=\,-59\,km/s}$ which was given by Omont et al.~(1993)
from CO data and ${\rm V_r\,=\,-62\,km/s\pm 1.0\,km/s}$ by Decin
et al.~(1998) from optical spectrum. There is still no sign of
an essential temporal variability of ${\rm V_r}$ for the object.
\subsection{Absorption bands identified with DIBs}
From comparison of observed and synthetic spectra of IRAS\,04296
we discovered some strong absorptional features whose positions
coincide with known diffuse interstellar bands (DIBs)
(Jenniskens et al.~1994).
In Fig.\,\ref{DIB} we illustrate the presence of DIB's by
showing a spectral region of IRAS\,04296. We have calculated the
synthetic spectrum using the code STARSP (Tsymbal~1995) and the
atmospheric parameters and abundances of chemical elements we here
obtained. It should be noted that for such a comparison in the
spectral range near
${\rm \lambda}$ 6270-6310\,\r{AA} the telluric spectrum has been
removed from the observed spectrum.
In the following paper we plan to study in detail such
identified with DIB's absorptions, we have revealed in the
spectra of several related objects (IRAS\,04296,
IRAS\,23304+6147, IRAS222223+4327), here we limited ourself by
such short information.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f5}}
\caption[]
{The observed (line) and synthetic (dots)
spectra of the object IRAS\,04296\,+\,3429 near absorbtional features identified
with the DIBs. The telliric spectrum was removed from the observed spectrum of
the object}
\label{DIB}
\end{figure}
\section{Determination of atmospheric parameters and calculation of
chemical composition}
For understanding of an object at an advanced evolutionary
stage, it is very important to know its metallicity and detailed
chemical abundance pattern. Our echelle spectra provide such a
possibility due to their large wavelengths coverage.
To study the chemical composition, we have used the
plane-parallel homogeneous models generated by the MARCS program
(Gustafsson et al.\,1975). It should be noted, however, that
unstable and very extended atmospheres of supergiants probably
require more advanced model atmospheres. Therefore, our results
should be treated as only preliminary ones. For a chemical
composition calculation by the model atmosphere method, one
needs to know the values of the effective temperature ($\rm T_{eff}$),
surface gravity (log\,g) and microturbulent velocity (${\rm
\xi_t}$). Determination of $\rm T_{eff}$ is problematic even for
normal supergiants due to their extended atmospheres and
significant non-LTE effects. In the case of so peculiar a
supergiant as IRAS\,04296, for which the energy distribution is
strongly distorted by interstellar and circumstellar extinction,
determination of $\rm T_{eff}$ is the most difficult problem. We
cannot use for this purpose equivalent widths and profiles of
H\,I lines (well known criteria of atmospheric conditions for
normal supergiants), since these lines are strongly distorted in
the spectrum of IRAS\,04296 as seen in Fig.\,\ref{HI}.
Therefore, we have applied the spectroscopic method for
temperature determination of IRAS\,04296, forcing the abundance
derived for each line to be independent of the
lower excitation potential (EP). We have estimated
that $\rm T_{eff}$ = 6300\,K with an internal uncertainty
$\Delta$\,$\rm T_{eff}$ = 250\,K. To check the realiability of our
determination we have modelled the spectral energy distribution
for this source (see Sect.\,4) and got a very similar
temperature near 6500\,K. The surface gravity log\,g\,=\,0.0 was
estimated through the ionization balance of the Fe\,I and Fe\,II
abundances. The errors on the parameter log\,g is determined by forcing a
maximum difference between ${\rm \epsilon\,(Fe\,I)}$ and ${\rm
\epsilon\,(Fe\,II)}$ to be 0.1\,dex (where here and hereafter,
${\rm log\,\epsilon\,(X)\,=\,log\,N(X)-log\,N(H)}$). It should
be noted that the hydrogen abundance ${\rm log\,N(H)}$\,=\,12.
Such a difference is achieved by varying the log\,g value
by ${\rm \pm 0.2}$ keeping other parameters
($\rm T_{eff}$ and ${\rm \xi_t}$) constant.
The microturbulent velocity value based on equivalent
widths (W) of Fe\,I and Fe\,II lines is quite high, equal to
7\,km/s. This value is determined with an uncertainty of ${\rm
\pm 1.0\,km/s}$, which is typical for F, G--supergiants.
To illustrate the choice of model parameters for the object
IRAS\,04296 in the Fig.\,\ref{Fe} are shown the excitation
potential -- abundance diagram and the equivalent width --
abundance diagram for lines of neutral (dots) and ionized
(crosses) iron atoms. As follows from this figure, there are no
essential dependences for values considered. The large
dispersion is mainly explained by errors of measurement of
equivalent widths of weak absorption lines for such a faint
object as the IRAS\,04296 (see, for example, the similar
dispersion on the Fig.\,1 in the paper by Decin et al.~(1998)
for the brighter object IRAS\,22223+4327, V=9.7).
We have checked the determination of IRAS\,04296 model parameters
using weaker FeI and FeII lines and concluded that the
parameters are steady within the erorr box up to
${\rm W}$ = 100-150\,m\r{AA}. This can also be seen from Fig.\,\ref{Fe}.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f6a}}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f6b}}
\caption[]{
{\it Upper:} iron abundance FeI (circles) and FeII (crosses) calculated
for IRAS\,04296
with model parameters ${\rm T_{eff}=6300\,K}$, ${\rm log\,g=0.0}$ and
${\rm \xi_t=7.0\,km/sec}$ using lines with different EP of a low level;
{\it bottom:} the same as a function of the equivalent width W}
\label{Fe}
\end{figure}
It is well known that the plane-parallel static model atmosphere
method does not give correct abundances for high luminosity
stars (luminosity classes Ia, Ia+). The profiles of the spectral
lines observed are broadened by non-thermal mecha\-nisms whose
influence may be variable at different levels in the atmosphere.
Therefore, to obtain more reliable estimates of chemical element
abundances we use weak lines with ${\rm W < }$250\,m\r{AA}.
The average values of the equivalent widths ${\rm \overline{W}}$ we
used for the abundances calculations are also given in Table\,1.
Only the BaII abundance was calculated using 3 very strong lines:
${\rm W(\lambda}$ 5853.67)\,=\,464\,m\r{AA},
${\rm W(\lambda}$ 6141.71)\,=\,679\,\,m\r{AA}
and ${\rm W(\lambda}$ 6496.90)\,=\,738\,\,m\r{AA},
because the weaker lines of this element were not available.
In general, the weak lines formed in deeper atmospheric layers are
more correctly described by the standard static model. The limitation
of equivalent width of lines used to ${\rm W < }$250\,m\r{AA}
significantly reduces the influence of uncertainty in the choice
of ${\rm \xi_t}$. Note, however, that the main factor in the abundance
errors for most species remains the uncertainty of the $\rm T_{eff}$ value.
Therefore, we have checked our estimation of $\rm T_{eff}$ by modelling of
spectral energy distribution for IRAS\,04296.
Computed abundances of 26 chemical elements are presented in
Table\,1. In the head of the Table\,1 parameters of the adopted
model atmosphere are shown. The dependence of chemical
composition determination on uncertanties of the model
atmosphere parameters is discussed in Za\v{c}s et al.~(1995). In
the second column of Table\,1 derived abundances are given as
${\rm log\,\epsilon\,(X)}$, while in the third column estimated
uncertainties of ${\rm \sigma\,=\,\Delta\,log\,\epsilon\,(X)}$ are shown.
In the next column, the number of spectral lines used for chemical
composition calculation is indicated.
\begin{table*}
\caption[]{Model atmosphere parameters adopted and abundances of chemical
elements. Here, n is number of
lines used for calculation, ${\sigma}$ -
the standard deviation, ${\rm \overline{W}}$ - the average equivalent
width, in ${\rm m\r{AA}}$, of lines used for the content calculation}
\begin{tabular}{l|crrr|crrr}
\hline
&&&&&&&& \\
&\multicolumn{4}{c|}{IRAS\,04296+3429} &\multicolumn{4}{c}{${\rm\alpha\,Per}$}\\
&\multicolumn{4}{c|}{$\rm T_{eff}$\,=\,6300\,K, log\,g\,=\,0.0, $\xi_t$\,=\,7.0\,km/s}
&\multicolumn{4}{c}{$\rm T_{eff}$\,=\,6500\,K, log\,g\,=\,1.2,$\xi_t$\,=\,4.7\,km/s}\\[5pt]
\hline
&\multicolumn{4}{c|}{}& \multicolumn{4}{}{}\\
Element&${\rm log\,\epsilon (X)}$ & \hspace{0.6cm} ${\sigma}$
& n &${\rm \overline{W}}$& ${\rm log\,\epsilon (X)}$
& \hspace{0.6cm}
${\sigma}$&n &${\rm \overline{W}}$ \\[3pt]
\hline
Li\,I & ${\rm \ge2.70}$ & & 1 & 32 & & & & \\
C\,I & 8.55 & 0.46 & 21 & 69 &8.16&0.14& 13 & 47 \\
N\,I & 7.96 & 0.10 & 4 & 99 &8.35&0.10& 4 & 127\\
O\,I & 8.22 & 0.05 & 3 & 26 &8.35&0.06& 4 & 23 \\
Na\,I & 5.91 & 0.24 & 3 & 68 &6.48&0.06& 4 & 48\\
Mg\,I & & & & &7.83&0.03& 2 & 56\\
Mg\,II & 8.08 & 0.03 & 2 &254 & & & & \\
Al\,I & 6.66 & 0.14 & 3 & 68 &6.57&0.16& 4 & 32\\
Si\.I & 7.29 & 0.20 & 11 & 37 &7.68&0.16& 16 & 45\\
Si\,II & 6.97 & & 1 & 22 &7.81& & 1 & 278\\
S\,I & 6.80 & 0.21 & 7 & 30 &7.53&0.23& 2 & 187\\
Ca\,I & 5.71 & 0.30 & 19 & 98 &6.41&0.22& 14 & 122\\
Sc\,II & 2.51 & 0.28 & 10 &164 &2.72&0.07& 6 & 119\\
Ti\,II & 3.91 & 0.33 & 5 &184 &4.78&0.08& 4 & 47\\
V\,II & 3.26 & 0.28 & 4 & 26 &3.54&0.10& 4 & 22\\
Cr\,II & 4.94 & 0.28 & 10 &108 &5.54&0.12& 9 & 136\\
Mn\,I & & & & &5.25&0.09& 3 & 63\\
Fe\,I & 6.66 & 0.30 & 55 & 75 &7.48&0.21&111 & 59\\
Fe\,II & 6.65 & 0.22 & 19 &131 &7.51&0.09& 10 & 154\\
Cu\,I & 3.61 & & 1 & 38 &4.66& & 1 & 36\\
Zn\,I & 3.84 & & 1 & 9 & & & & \\
Y\,II & 2.60 & 0.14 & 2 &168 &2.20&0.40& 2 & 32\\
Zr\,I & & & & &3.38&0.10& 4 & 6\\
Zr\,II & 2.38 & & 1 &165 & & & & \\
Ba\,II & 3.78 & 0.47 & 3 &627 &2.06& & 1 & 212\\
La\,II & 1.55 & 0.44 & 6 &116 &1.04&0.08& 4 & 20\\
Ce\,II & 1.53 & 0.16 & 5 & 83 & & & & \\
Pr\,II & 0.61 & & 1 & 19 & & & & \\
Nd\,II & 1.73 & 0.31 & 12 &102 &0.84&0.08& 4 & 8\\
Eu\,II & 0.01 & 0.04 & 2 & 20 &0.44&0.09& 3 & 20\\
\hline
\end{tabular}
\end{table*}
A lot of absorption lines of different elements (CNO-elements,
light metals, iron group elements, Ce, Nd, Eu) have been
reliably measured in the spectrum of IRAS\,04296. It is
important that we have not found any dependence of the abundances of
these species on the equivalent width or on the excitation
potential. Therefore the microturbulent velocity does not vary
between different chemical elements.
The gf--values for most of the spectral lines used for the
abundance calculations were taken from the list used by Luck~(1991).
The S and CNO-abundances were determined by using the
gf--data from Waelkens et al.~(1991) and Giridhar et al.~(1994).
The list of lines with the adopted gf--values, excitation
potentials of the lower level and equivalent widths we measured
for the object IRAS\,04296 are available by e-mail
(<EMAIL>).
To verify the method of analysis we observed with the same
spectral device the normal supergiant ${\alpha}$\,Per.
The same procedures for processing and the same list of lines
were used for analysis of the ${\alpha}$\,Per spectrum.
This supergiant, whose parameters, ${\rm T_{eff}=6500\,K}$,
${\rm log\,g=1.2}$, ${\rm \xi_t=4.7\,km/sec}$ are very close to
the object studied, is very convenient as a standard for the method
testing because of its membership in the young open cluster
${\rm \alpha Per}$ which has solar chemical composition
(Klochkova \& Panchuk~1985; Boesgaard~1989). Using its membership
of this cluster, we may predict that
${\rm \alpha\,Per}$ also has normal solar chemical composition
(aside from the expected nonsolar CNO triad abundances relative
to iron). As it is shown in Table\,1 ${\rm \alpha Per}$ has indeed
the abundances of chemical elements close to solar ones, except for
CNO and several elements whose abundances are calculated with a
large uncertainity due to a small number of spectral lines used.
\section{Spectral energy distribution}
Details of the computer code used for solution of the radiative
transfer in dusty envelopes can be found in Szczerba et
al.~(1997). In brief: the frequency--dependent radiative transfer
equation is solved for a dust under assumption of
spherically--symmetric geometry for its distribution taking into
account particle size distribution and quantum heating effects
for the very small dust particles.
The modelled source is certainly C--rich (see Omont et al.~1995).
Therefore, for modelling of its spectral
energy distribution (SED) we assumed that dust is composed of:
policyclic aromatic hydrocarbons (PAH) for dust sizes $a$ between
5 and 10\,\r{AA} (see Szczerba et al.~(1997) for details
concerning PAH properties), amorphous carbon grains (of AC type
from Rouleau \& Martin~1991) for $a$\,$>$\,50\,\r{AA}, and dust
with an opacity obtained from averaging of the absorption
efficiences for PAH and AC grains according to the formula: $$
Q_{{\rm abs},\,\nu}\,=\,
{\rm f}\,\cdot\,Q_{{\rm abs},\,\nu}^{{\rm PAH}}(a)\,
+\,(1-{\rm f})\,\cdot\,Q_{{\rm abs},\,\nu}^{\rm AC}(a),$$ for
grain sizes between 10 and 50\,\r{AA}. Here: f\,=\,1 for $a$\,=\,
10\,\r{AA} and f\,=\,0 for $a$\,=\,50\,\r{AA}. Dust with opacity values
constructed in this way allow us to use a continous distribution
of dust grain sizes and fill the gap between properties of
carbon--bearing molecules and small carbon grains.
The ${\rm 21\,\mu}$m feature was approximated by a gaussian with
parameters determined from modelling of IRAS\,07134$+$1005
(centre wavelength equal to ${\rm 20.6\,\mu}$m, and width of
${\rm 1.5\,\mu}$m) which has the strongest
feature among the known ${\rm 21\,\mu}$m sources. In the case of
${\rm 30\,\mu}$m band we used the addition of two half--gausians
with the same strength and different width. Initial fit was done
to IRAS 22272$+$5435 and its parameters were: width for short
wavelength side $\sigma_{\rm L}\,=\,{\rm 4\,\mu}$m, width for
long wavelength side ${\sigma_{\rm R}\,=\,9\,\mu}$m and central
wavelength ${\rm 27.2\,\mu}$m (see Szczerba et al.~1997). For
modelling of IRAS\,04296 we have reduced the strength of this
feature by 50\%. Superposition of the 21 and ${\rm 30\,\mu}$m
features was added to the absorption properties of amorphous carbon
in order to construct an empirical opacity function (EOF).
In Fig.\,\ref{IRAS04296} the best fit obtained from the solution
of the radiative transfer problem including quantum heating
effects for the PAH grains is shown together with observational
data which will be described in detail elsewhere. Note, however,
that we present also two sets of photometry (from B to M band)
corrected for interstellar extinction (open symbols)
according to the average extinction law of Cardelli et
al.~(1989), assuming that total extinction at V is 1.0 or 2.0
magnitudes and plotting only the smallest and largest value of
corrected fluxes at given band. This estimate of the total extinction
range can be inferred from
the analysis of data presented by Burstein \& Heiles~(1982).
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{MS7673.f7}}
\caption[]
{Fit to the spectral energy distribution of IRAS\,04296\,+\,3429 obtained
with empirical opacity function, taking into account
quantum heating effects (heavy solid line) and assuming that
star radiates according to a model stellar atmosphere calculations for
log\,g\,=\,0.5 and $\rm T_{eff}$\,=\,6500 K (thin long--dashed line).
The heavy dashed line shows the fit obtained for star with
$\rm T_{eff}$\,=\,6000 K. The thin solid line underlying 21
and ${\rm 30\,\mu}$m features represents the estimated
continuum level}
\label{IRAS04296}
\end{figure}
\begin{table}
\caption[]{Model parameters for IRAS\,04296+3429.
Precise meaning of the symbols used can be found in Szczerba et
al.~(1997)}
\begin{tabular}{ l l }
\hline
\noalign{\smallskip}
parameter & value \\
\noalign{\smallskip}
\hline
\noalign{\smallskip}
$\rm T_{eff}$ & 6500\,\,K \\
log (L$_{\rm star}$\,[L$_{\odot}$]) & 3.92 \\
$d$ & 5.4\,\,kpc \\
& \\
$R_{\rm out}$ & 0.5\,\,pc \\
$V_{\rm exp}$ & 12\,km\,s$^{-1}$ \\
& \\
$R_{\rm in}$(hot dust shell) & 6.4\,10$^{-4}$\,\,pc \\
$\overline{T}_{\rm d}$[R$_{\rm in}$(hot dust shell)] & 870\,\,K \\
$\rho_{\rm gas}$ (hot dust shell) & $\sim$\,r$^{-2.0}$ \\
$\dot{M}_{\rm post-AGB}$ & 4.0\,10$^{-7}$\,M$_{\odot}\,{\rm yr}^{-1}$\\
& \\
$R_{\rm in}$(main shell) & 7.06\,10$^{-3}$\,\,pc \\
$\overline{T}_{\rm d}$[R$_{\rm in}$(main shell)] & 270\,\,K \\
$\rho_{\rm gas}$ (main shell) & $\sim$\,r$^{-2.6}$ \\
$\dot{M}_{\rm AGB}^{\rm min}$ & 1.70\,10$^{-5}$\,M$_{\odot}\,{\rm yr}^{-1}$ \\
$\dot{M}_{\rm AGB}^{\rm max}$ & 2.19\,10$^{-4}$\,M$_{\odot}\,{\rm yr}^{-1}$ \\
& \\
$a_{-}$ & 5\r{AA} \\
$a_{+}$ & 0.25\,$\mu$m \\
$p$ & 3.5 \\
& \\
$t_{\rm dyn}$ & 575\,\,yr \\
$M_{\rm dust}$ & 0.0071\,M$_{\odot}$ \\
\noalign{\smallskip}
\hline
\end{tabular}
\end{table}
The best fit to the spectral energy distribution of IRAS\,04296
is shown by heavy solid line (see Table\,2 for details
concerning parameters of the model). Our modelling procedure was
such that we tried to get fits to the SED which fall in between the
extinction corrected fluxes. In this way, we have taken into
account not only the effect of the circumstellar extinction but
also of interstellar extinction. The thin long--dashed line
represents the input energy distribution of the central star for
log\,g\,=\,0.5 and $\rm T_{eff}$\,=\,6500\,K according to model
atmosphere calculations of Kurucz (private communication). The
heavy short--dashed line shows the fit which was obtained with
the same assumptions but changing the effective temperature of
the star to 6000\,K. As one can immediately see in the IR range
of the spectrum the quality of the fits are very similar.
However, in the optical and ultraviolet (UV) part of the
spectrum the fit assuming $\rm T_{eff}$\,=\,6000\,K is not able to
explain extinction corrected data. In consequence, we are quite
convinced that our estimation of $\rm T_{eff}$ for IRAS\,04296 close
to 6500\,K is reasonable and, what is even more important,
agrees pretty well with the spectroscopic estimation (6300\,K).
Note that spectral type of this source was found to be G0~Ia
from the low resolution spectrum (Hrivnak~1995) which implies an
effective temperature of around 5500\,K for the star if we asume
that the same relationship applies for post--AGB supergiants as
for ``normal'' ones (see Schmidt--Kaler~1982). For such a low
temperature we were not able to fit even the reddenned data in
the UV.
The thin solid line in the wavelength range from about 18 to
${\rm 48\,\mu}$m represents the model continuum level found after
solution of radiative transfer equation for dust without EOF using
the parameters as in Tab.\,2 while keeping the dust
temperature (or probability distribution of dust temperature)
the same as for the case of dust with EOF. Taking into account
the estimated continuum level and assuming that 21\,$\mu$m feature
extends from 18 to 22\,$\mu$m we estimate the energy emitted in 21\,$\mu$m
band as about 5.7 \% of the total IR flux (251\,L$_{\odot}$ for
$\lambda$'s from 5 to 300\,$\mu$m assuming a distance to the source
of 1 kpc). With the dotted line for wavelengths longer than 18\,$\mu$m
we present the fit which was obtained using an opacity
function with the EOF for only 21\,$\mu$m component. It is clear
that such fit is not able to explain IRAS photometry at 25\,$\mu$m.
Our recent {\it ISO} observations show that this source is also a
30\,$\mu$m emitter. In the forthcoming paper (Szczerba et al.~1999)
we will discuss this finding in detail.
\section{Discussion}
As it is shown in Table\,1, the metallicity for IRAS\,04296
is significantly decreased relative to the solar value: the average
abundance for the elements of the iron-group with respect to the Sun
is ${\rm [(Ti,V,Cr,Fe)/H]_{\odot}\,=\,-0.9}$ with the standard
deviation ${\rm \sigma = 0.2}$.
Recently Decin et al.~(1998), using high resolution spectra and
model atmospheres method, calculated abundances of 14 chemical
elements in the IRAS\,04296 atmosphere. Their results are in
qualitative agreement with these ones presented here, but there
are some significant differences. Decin et al.~(1998) calculated
chemical composition of this object assuming $\rm T_{eff}$\,=\,7000\,K,
log\,g\,=\,1.0, $\xi_t$\,=\,4\,km/s, rather different from
the model atmospheres parameters found in this work. It should
be noted that we estimated the effective temperature by two
independent methods, and it is worth stressing that we have
obtained consistent values of the effective temperature:
$\rm T_{eff}$\,=\,6300\,K from numerous Fe\,I, Fe\,II spectral lines and
$\rm T_{eff}$\,=\,6500\,K from modelling of the spectral energy
distribution of this source. The difference in effective
temperature between Decin et al.~(1998) and our estimation
(${\Delta}$\,$\rm T_{eff}$\,=\,700\,K) is able to explain different
metallicities estimated by Decin et al.~(1998) and by us (${\rm
\Delta\,log\,\epsilon(Fe)\approx\,0.2}$). The same is true for
the case of the rare-earth element abundances: large
differences, about 1\,dex, in the values could be explained by
differences in model atmosphere parameters.
Let us consider now in more detail the peculiarities in the
chemical composition of the object. For this purpose, in
Table\,3 we present the logarithmic differences
$${{\rm [X/Fe]_{\odot}\,=\,[log\,\epsilon(X) - log\,\epsilon (Fe)]_{\star}
- [log\,\epsilon(X) - log \,\epsilon (Fe)]_{\odot}}}$$
between chemical compositions of different objects and the Sun
(solar abundaces from Grevesse et al.~(1996)): in the second
column for IRAS\,04296, in the third column for IRAS\,07134+1005
(hereafter IRAS\,07134 - asscociated with the peculiar F--type
supergiant HD\,56126) and in the fourth one for the star
ROA\,24. The objects are similar from point of view of their
atmospheric parameters ($\rm T_{eff}$, log\,g) and relative chemical
composition. It should be noted that the metal deficient
supergiant ROA\,24 (Fehrenbach's star) belongs to the globular
cluster ${\rm \omega\,Cen}$ and could be considered as a typical
{\it halo} object in the post-AGB evolution stage.
\begin{table}
\caption{Relative abundances of chemical
elements for IRAS\,04296+3429 in comparison to
related PPNe. The data by Grevesse \& Noels~(1996) are adopted
for solar abundances.}
\begin{tabular}{lcccc}
&&&& \\
\hline\LARGE
& IRAS\,04296+3429 & IRAS\,07134+1005$^a$ & ROA\,24$^b$ \\
&${\rm [Fe/H]_{\sun}\,=\,-0.84}$ &-1.00 &-1.77 \\[5pt]
\hline
Element &\multicolumn{3}{c}{${\rm [X/Fe]_{\odot}}$}\\[5pt]
\hline
Li\,I & ${\rm \ge +0.23 }$ & & \\
C\,I & +0.84 &+1.08 &+0.67 \\
N\,I & +0.83 &+1.03 &+1.02 \\
O\,I & +0.19 &+0.63 &+1.01 \\
Na\,I & +0.42 &+0.54 &+0.71 \\
Mg\,I & &+0.97 &+0.31 \\
Mg\,II & +1.34 & &+0.09 \\
Al\,I & +1.03 &+1.48 & \\
Si\,I & +0.58 &+0.95 &+0.80 \\
Si\,II & +0.26 & &+1.03 \\
S\,I & +0.43 &+0.63 & \\
Ca\,I & +0.19 &+0.45 &+0.60 \\
Sc\,II & +0.18 &-0.07 &-0.13 \\
Ti\,II & -0.27 & &+0.33 \\
V\,II & +0.10 &-0.03 &+0.15 \\
Cr\,II & +0.11 & &+0.65 \\
Cu\,I & +0.24 &+1.03 &-0.01 \\
Zn\,I & +0.08 & & \\
Y\,II & +1.20 &+1.70 &+0.37 \\
Zr\,II & +0.62 & & \\
Ba\,II & +2.49 &+0.99 &+0.96 \\
La\,II & +1.17 &+1.59 &+0.54 \\
Ce\,II & +0.82 & &+1.60 \\
Pr\,II & +0.74 & & \\
Nd\,II & +1.07 &+1.30 &+0.67 \\
Eu\,II & +0.34 &+1.06 &+0.25 \\
\hline
\multicolumn{4}{l}{a -- Klochkova~(1995),} \\
\multicolumn{4}{l}{b -- Gonzalez \& Wallerstein~(1992).}\\
\end{tabular}
\end{table}
The carbon overabundance ${{\rm [C/Fe]_\odot\,=\,+0.8}}$
(revealed from intensities of 21 absorption lines
with the standard deviation ${\rm \sigma = 0.46}$) and the
enhancement of nitrogen ${{\rm [N/Fe]_\odot\,=\,+0.8}}$ (from 4
lines, ${\rm \sigma = 0.10}$) suggest that IRAS\,04296 underwent the
third dredge--up episode.
The oxygen content based on intensity of 3 weak lines
near ${\rm \lambda \approx }$ 6155\,\r{AA} is determined with a small
internal error.
From the Fe--deficiency and CNO abundances (${\rm C/O\,>\,2}$)
we can conclude that IRAS\,04296 is a low mass object in
advanced stage of evolution. For an {\it unevolved}
metal--deficient object (with ${\rm [Fe/H]_{\odot} \approx -0.9}$)
the average value of ${\rm [C/Fe]_{\odot}}$ is only about
-0.2 (Tomkin et al.~1995), the average value of ${\rm [N/Fe]_{\odot}}$
is ${\rm \approx 0}$ (Wheeler et al.~1989, Timmes et al.~1995)
and the average value of ${\rm [O/Fe]_{\odot}}$ is ${\approx +0.5}$
(Wheeler et al.~1989, Timmes et al.~1995, Klochkova \& Panchuk~1996b).
The atmospheres of the post-AGB stars IRAS\,07134 and ROA\,24 are also
overabundant in both carbon and nitrogen. Note however, that for
most of the PPNe candidates studied, strong relative changes
between elements of the CNO--group are observed (Luck et
al.~1983; Lambert et al.~1988; Klochkova~1995; Za\v{c}s et
al.~1995, 1996; van Winckel et al.~1996a, 1996b; van Winckel~1997).
The abundances of some light metals (Na, Al, Mg, Si, Ca) are enhanced
for all three stars. The average value for these elements is
${\rm [X/Fe]_{\odot}\,=\,+0.6}$ for IRAS\,04296; +0.9 for IRAS\,07134 and
+0.6 for ROA\,24, with the standard deviations: ${\rm \sigma\,=\,0.4}$,
0.4 and 0.36, respectively.
We did not still include the K\,I abundance into our results,
since we suspect that the equivalent width of its line near
${\rm \lambda}$ 7699\,\r{AA} could be significantly distorted due
to circumstellar and interstellar components.
The iron--group element zinc is the most important for determination
of real (initial) value of the metallicity of a star since, firstly,
its abundance follows that of iron in a wide [Fe/H]$_{\odot}$ interval
(Sneden \& Crocker~1988; Wheeler et al.~1989, Sneden et al.~1991) and,
secondly, zinc having a low condensation temperature is not depleted by
selective separation processes onto dust grains (Bond~1992).
A close to solar abundance of Zn relative to iron
(${\rm [Zn/Fe]_{\odot}\,=\,+0.1}$) permits us to conclude about the
inefficiency of the selective separation processes in the IRAS\,04296
envelope. This conclusion is based also on an absence of overdeficiency
of light depleted elements (Ca, Sc).
Besides, the relative abundance (${\rm [S/Fe]_{\odot}\,=\,+0.4}$
with the standard deviation ${\rm \sigma = 0.21}$) of S, a chemical
element which is not depleted by dust--gas separation, for IRAS\,04296
is close to the value for unevolved metal-deficient dwarfs
(Fran\c{c}ois~1987, Timmes et al.~1995). This futher confirms
the lack of selective separation in the envelope of the object studied.
Individual abundances of the heavy s-process metals Y and Zr are
determined with a relatively large error because of the small
number of lines measured. However, the average value ${\rm
[X/Fe]_{\odot}\,=\,+0.9}$ for Y and Zr is sufficiently
reliable. In addition, the abundance of heavy s-process element
Ba (${\rm [Ba/H]_{\odot}\,=\,+2.5}$) derived from the
equivalent width of strong lines could be altered by a
systematic error due to the complexity of the outer regions of
the stellar atmopshere as discussed above. Nevertheless, we
conclude that there is a Ba excess.
The abundance of lanthanides (La, Ce, Pr, Nd) are strongly enhanced relative
to iron for the objects from Table\,3. For these heavy metals the
average value is ${{\rm [X/Fe]_{\odot}\,=\,+1.0,\,+1.4,\,+0.9}}$
for IRAS\,04296, IRAS\,07134 and ROA\,24, respectively, with the standard
deviations 0.2 and 0.6 for IRAS\,04296 and ROA\,24. Moreover,
for all these objects we see the overabundance of Eu which is
predominantly produced by the r--process.
Excess of s-process elements has been reliably found up to now
in three objects investigated at the 6\,m telescope:
IRAS\,04296+3429, IRAS\,07134+1005 and IRAS\,22272+5435.
Besides, similar conclusions have appeared for another four PPN
candidates (and for one object in common): HD\,158616 (van
Winckel et al.~1995); IRAS\,19500-1709\,=\,HD\,187885 (van
Winckel~1997); IRAS\,05341+0852 (Reddy et al.~1997);
IRAS\,22223+4327 and IRAS\,04296+3429 (Decin et al.~1998). In
atmospheres of most PPN candidates overdeficiency (with respect
to their metallicity) of heavy nuclei is generally observed
(Klochkova~1995; van Winckel et al.~1996a, 1996b; Klochkova \&
Panchuk~1996a; van Winckel~1997), whose existence in the
atmospheres of post--AGB low--mass supergiants has not yet found
a clear explanation.
In consequence, we can state that chemical abundances pattern
for the source IRAS\,04296 is related to its old galactic
population membership and dredge--up of matter enriched by the
nucleosynthesis products. It may be part of the old disk population.
As has been concluded already by Decin et al.~(1998) all the
post--AGB candidates mentioned above (only these, up to now,
show an s--process element enhancement!) belong to the small
group of PPNe (Kwok et al.~1989; Kwok et al.~1995) which have in
their IR spectrum an unidentified emission band at about ${\rm
21\,\mu}$m. This feature is neither found in the
spectra of their predecessors, AGB stars, nor in the spectra of
PNe. Note, once more, that the search by means of the {\it ISO} for the
new 21\,$\mu$m emitters among candidates selected by Henning et
al.~(1996) failed (Henning, private communication).
As has been stated in the
papers by Kwok et al.~(1989, 1995), the objects whose spectra
contain the ${\rm 21\,\mu}$m band are carbon-rich stars. Our
investigations based on the spectra from the 6\,m telescope, for
IRAS\,07134 (Klochkova~1995), IRAS\,22272+5435 (Za\v{c}s et
al.~1995) and IRAS\,04296 (Klochkova et al.~1997b), confirmed
that ${\rm C/O>1}$ for all of them. In this context, the
conclusion that the carrier of the 21\,$\mu$m band is related to
C is natural. For example, Buss et al.~(1990) have supposed that
this feature may be caused by polycyclic
aromatic hydrocarbons. On the other hand, Goebel~(1993) has
identified the ${\rm 21\,\mu}$m band with the vibrational band
of the SiS$_2$ molecule, the presence of which is consistent
with the temperature in the envelope.
Taking into account the available results on chemical
composition for subclass of PPNe with the 21\,$\mu$m feature:
IRAS\,07134+1005 (Parthasarathy et al.~1992, Klochkova~1995),
IRAS\,22272+5435 (Za\v{c}s et al.~1995), IRAS\,19500-1709 (van
Winckel et al.~1996a), IRAS\,05341+0852 (Reddy et al.~1997),
IRAS\,22223+4327 (Decin et al.~1998), and IRAS\,04296 (Decin et
al.~1998; Klochkova et al.~1997b; this paper) we see that the
carbon-rich atmospheres of these objects are also enriched by
s--process elements. It is evident that there is a strong
correlation between presence of the 21\,$\mu$m feature, ${\rm
C_2}$, ${\rm C_3}$ molecular bands, and excess of the s--process
elements. Decin et al.~(1998) were the first who pointed out
this relationship. What is even more important, an excess of
s--process elements was not found for a number of IRAS sources
with altered CNO-content but without the 21\,$\mu$m feature (some
of which are oxygen-rich stars rather than carbon-rich stars):
IRAS\,06338$+$5333 (Luck \& Bond~1984; Bond \& Luck~1987),
IRAS\,07331$+$0021 (Luck \& Bond~1989; Klochkova \& Panchuk~1996a),
IRAS\,09276$+$4454 (Klochkova \& Mishenina~1998),
IRAS\,12175$-$5338 (van Winckel~1997), IRAS 12538$-$2611 (Luck
et al.~1983; Klochkova \& Panchuk~1988b; Giridhar et al.~1997),
IRAS\,15039$-$4806 (van Winckel et al.~1996b),
IRAS\,17436$+$5003 (Klochkova \& Panchuk~1988a; Luck et
al.~1990; Klochkova~1998), IRAS\,18095$+$2704 (Klochkova~1995),
and IRAS\,19114$+$0002 (Za\v{c}s et al.~1996, Klochkova~1998).
Therefore, it seems that carrier of 21\,$\mu$m feature is {\it
strongly} related to the whole chemical composition pattern
typical for the third dredge--up (excess of s-process elements),
and not only to the C-richness of the photosphere.
That 21\,$\mu$m feature is not observed around AGB--stars showing
s--process elements could be explained by the physical
conditions which are inappropriate for the excitation of this
band, while its non--presence in planetary nebulae may be a
result of carrier destruction by the highly energetic photons.
\section{Conclusions}
We conlude that IRAS\,04296$+$3429 is a PPN candidate with a
chemical composition which coincides with theoretical
predictions for the post--AGB objects: very large excess of
carbon and nitrogen are revealed. Moreover, the real
excess (relative to iron) of heavy metals Y, Zr, Ba, La, Ce, Pr,
Nd synthesized by the neutronization process indicates an
effective third dredge--up and further confirms IRAS\,04296 to
be in the advanced post-AGB evolution stage.
The emission of ${\rm C_2}$ molecular lines discovered in the
spectra of IRAS\,04296 and its similarity to the emission of the
Hale--Bopp comet allow us to suggest that in both cases the same
mechanism (the resonance fluorescence) is responsible for the
observed features.
Several strong absorptional features whose positions coincide
with known diffuse interstellar bands (DIBs) are found in the
spectrum of IRAS\,04296.
In addition, from the SED modelling of the spectral energy
distribution we showed that 25\,$\mu$m flux cannont be explained
without assumption that the 30\,$\mu$m emission feature is
present in this source. Our recent {\it ISO} observation has detected
the 30\,$\mu$m band in this source.
\begin{acknowledgements}
We are much indebted to the
referee Hans van Winckel for critical reading the manuscript and
valuable advice.
This work has been supported by project 1.4.1.1 of the Russian
Federal Program ``Astronomy'' and grant
2.P03D.002.13 of the Polish State Committee for Scientific
Research. One of us (R.Sz.) gratefully acknowledges the support
from the Canadian Institute of Theoretical Astrophysics.
\end{acknowledgements}
|
\section{Introduction}
\label{sec:I}
Recently, a unified theory of antiferromagnetism (AF) and
superconductivity (SC) has been proposed for the high $T_c$
cuprates\cite{so5}. This theory is based on the $SO(5)$ symmetry
between AF and SC, and offers a unified description of the
global phase diagram for this class of materials.
While the theory was originally proposed
as a effective field theory description, it was soon realized
that the $SO(5)$ symmetry could be implemented exactly
at a microscopic level\cite{henley,rabello,burgess2,szh,eder2},
and it can also be checked numerically
in common strongly correlated models such as the $t-J$
model\cite{meixner,eder,hanke-review}.
While the phase diagram\cite{so5,assa,burgess1,hu_xiao} and collective
excitations\cite{resonance1,resonance2} in the
SC state derived from these $SO(5)$ models bear strong resemblance
with the high $T_c$ cuprates, and a number of novel experimental
predictions have been made\cite{junction1,vortex,junction2,goldbart1,goldbart2},
the Mott insulating behavior at
half-filling is a puzzling aspect which challenges the fundamental
validity of the $SO(5)$ models\cite{greiter,greiter-reply,anderson,sns}.
To be more precise, the exact
$SO(5)$ symmetry requires collective charge two excitation at
half-filling to have the same mass as the collective spin wave
excitations. This condition is clearly violated in a Mott insulating
system where all charge excitations measured with respect to a particle
hole symmetric point have a large energy gap of few $eV$, while
the spin wave excitations are massless. In the original $SO(5)$ proposal,
it was pointed out that this situation is analogous to a easy-axis
antiferromagnet in a external uniform field, and a $SO(5)$
symmetry breaking term at half-filling
was introduced in order to describe this
asymmetric behavior between spin and charge. The
chemical potential also introduces a $SO(5)$ symmetry breaking
term, however, it was shown that these two terms could
compensate each other\cite{so5,eder} so that the {\it static potential} governing
the $SO(5)$ superspin could still be $SO(5)$ symmetric.
Since the asymmetry between the charge and spin excitations at half-filling
is of the order of the Coulomb energy scale $U$, the $SO(5)$ symmetry
breaking terms must also be of that order. Since there are various types
of symmetry breaking terms, one might hope that their effects could
partially cancel each other to arrive at a qualitatively
correct picture. However, this type of cancellation is very delicate,
and approximate calculations could easily lead to erroneous
conclusions. In particular, one is interested in which physical
properties could exhibit $SO(5)$ symmetric properties in the limit
when the Coulomb gap is taken to infinity. For example, one could
ask the following questions:
1) One of the hallmarks of the $SO(5)$ symmetry is not only the degeneracy
between the AF and SC states at a given chemical potential, but the
approximate degeneracy among all mix-states interpolating between AF and SC,
{\it i.e.} the independence of the ground state energy on the superspin
angle. What is the potential barrier separating the AF and SC states at
their degeneracy point in the limit $U\rightarrow \infty$? If there is
a large energy barrier in this limit, one would argue that the concept
of $SO(5)$ symmetry is not a useful one, at least not for quantitative
calculations. On the other hand, if the potential barrier is finite and
small in the $U\rightarrow \infty$ limit, the concept of a approximate
$SO(5)$ symmetry would be a useful one.
2) Exact $SO(5)$ symmetry predicts four massless collective modes. In the
half-filled AF state, besides the two conventional massless spin wave
modes, the exact $SO(5)$ symmetry predicts a massless
doublet of $\pi^\pm$ modes, with charge $\pm 2$. However, a Mott
insulator has a large gap to all charge excitations. Therefore, it is
clear that one of the $\pi^\pm$ has to be projected out of the spectrum
in the limit $U\rightarrow \infty$, say the $\pi^+$ mode carrying
charge $+2$. What happens to the rest of the Goldstone modes, the
$\pi^-$ mode carrying charge $-2$ and the $\pi^\alpha$ triplet mode of the
SC state? In the $U\rightarrow \infty$ limit, can they all be
simultaneously massless at the transition point between AF and SC?
Since the pure SC state can only be reached with a finite doping
concentration, is it possible that the Gutzwiller projection does not
affect the $\pi^\alpha$ triplet mode of the SC state?
In order to address these questions, it is desirable to construct a
low energy effective theory without any parameters of the order of
the Coulomb scale $U$.
In this work, we construct a class of projected $SO(5)$ models which treat
the Gutzwiller constraint exactly and locally on every site.
We use this model to answer
the physical questions posed above and show that the answers are affirmative.
In the $U\rightarrow \infty$ limit, when the Gutzwiller constraint is
implemented exactly, the ground state energy can still be $SO(5)$ symmetric
and independent of the superspin direction. After projecting out the $\pi^+$
mode, all other Goldstone modes remain massless at the symmetric point.
The dispersion relation of the collective modes bear unique
signature of the projected $SO(5)$ symmetry. Furthermore, the
$\pi^\alpha$ triplet modes of the pure SC states are unaffected
by the Gutzwiller projection. These
properties define the concept of a projected $SO(5)$ symmetry ($pSO(5)$),
whose properties and consequences we shall explore in this paper.
The fundamental quantity in the $SO(5)$ theory is the locally
defined five component superspin vector $n_a(x)=(n_1,n_2,n_3,n_4,n_5)$
describing the local AF and SC order parameters respectively.
In the nonlinear $\sigma$ model formalism, these are treated
as mutually commuting coordinates
and their dynamics is given by their conjugate
momenta $p_a(x)=(p_1,p_2,p_3,p_4,p_5)$. The
charge operator is the angular momentum in the $n_1-n_5$ plane:
\begin{equation}
Q (x) = L_{15} = n_1 p_5 - n_5 p_1
\label{charge}
\end{equation}
Implementing the Gutzwiller constraint corresponds to requiring
\begin{equation}
Q(x) \le 0
\label{Gutzwiller}
\end{equation}
for every local $SO(5)$ rotor. From equations
(\ref{charge}) and (\ref{Gutzwiller}) and subsequent discussions,
we shall see that the Gutzwiller projection in the $SO(5)$ formalism
corresponds to going
from a fully symmetric $SO(5)$ rotor model to a {\it chiral}
$SO(5)$ rotor model, where both the static potential of the individual
rotors and the coupling between the rotors are still $SO(5)$
symmetric, but the rotors are constrained to rotate only in one
sense in the $n_1-n_5$ plane, consistent with (\ref{Gutzwiller}).
This observation reveals a deep connection
between the Gutzwiller projection and the Lowest-Landau-Level (LLL)
projection in the fractional quantum Hall effect\cite{qhe-book}.
To be more
precise, the Gutzwiller projection represented by equations
(\ref{charge}) and (\ref{Gutzwiller}) is
analogous to the LLL projection, where all states in the LLL
have a definite sign of angular momentum. The LLL projection
can be analytically implemented by separating the
cyclotron degrees of freedom
from the guiding center degrees of freedom, which amounts to
changing the commuting property between the $X$ and $Y$ coordinates
to a canonically conjugate commutation relation:
\begin{equation}
[X,Y] = i \l_0
\label{guiding-center}
\end{equation}
where $l_0$ is the Landau length.
Exploiting this analogy, we find that the original $SO(5)$ model can be
fully Gutzwiller projected without changing its form, if one imposes
the simple quantization condition between the superconducting components
of the superspin vector:
\begin{equation}
[n_1,n_5] = i/2
\label{non-commute}
\end{equation}
In the symmetric $SO(5)$ model, the wave function of the $SO(5)$ rotors
are functions of the local coordinates $n_1$ and $n_5$, while the
projected $SO(5)$ model only depends on their holomorphic combination
$z=n_1-i n_5$ and is independent of their anti-holomorphic
combination $\bar z=n_1+i n_5$.
This way, we arrive at a natural projection of the $SO(5)$ model
where the local Gutzwiller constraint is taken into account exactly,
and the resulting model is free of the large Coulomb $U$ parameter.
Because the functional form of the symmetric $SO(5)$ model remain
the same and only the quantization condition is modified upon projection,
many important properties associated with the $SO(5)$ symmetry remain.
The central hypothesis of the $SO(5)$ theory is that this projected
model is quantitatively accurate in describing both the static and dynamic
properties of the high $T_c$ cuprates, and we shall compare the properties
of this model with the phenomenology of the high $T_c$ systems.
\section{Construction of projected $SO(5)$ Models}
\label{sec:II}
We begin with the symmetric $SO(5)$ Hamiltonian defined on a lattice,
\begin{eqnarray}
H &=&
\Delta \sum_{ x } L_{ab}^2(x)
- J \sum_{ <xx'> } n_a(x) n_a(x')
\nonumber \\
&+& V \sum_{ <xx'> }
L_{ab}(x) L_{ab}(x')
\label{sigma}
\end{eqnarray}
where $n_a(x)$ denotes the five component superspin vector on a
given site, and $L_{ab}(x)$ is the $SO(5)$ symmetry generator,
\begin{equation}
L_{ab} = n_a p_b - n_b p_a
\end{equation}
expressed here in terms of the superspin vector $n_a$ and its
canonically conjugate momenta $p_a$,
\begin{equation}
[n_a, p_b] = i \delta_{ab}
\end{equation}
This lattice quantum non-linear $\sigma$ model can be rigorously derived
as the low energy limit of a microscopic $SO(5)$ ladder model\cite{szh,eder2}.
On the ladder, the rung $SO(5)$ singlet state is the vacuum
$|\Omega\rangle$, from which the lowest SO(5) multiplet
$| a\rangle= t_a^\dagger|\Omega\rangle$ is created by a quintet
of Bose creation operators, which satisfy
\begin{equation}
[t_a, t_b^\dagger] = \delta_{ab},~~~~~ t_a |\Omega\rangle=0
\end{equation}
Here, $a=2,3,4$
denote the triplet (magnon) states, and $a=1,5$ are the hole and particle pair states.
(see Fig. (\ref{fig1})).
The superspin coordinates are microscopically constructed using these lattice bosons,
\begin{equation}
n_a = \frac{1}{\sqrt{2}} (t_a + t_a^\dagger) \ \ \
p_a = \frac{1}{i\sqrt{2}} (t_a - t_a^\dagger)
\end{equation}\begin{figure*}[h]
\centerline{\epsfysize=4.0cm
\epsfbox{fig1.eps}
}
\caption{Schematic representation of the singlet state, the triplet
magnon states and the hole and particle pair states.}
\label{fig1}
\end{figure*}
Due to their microscopic origin, these bosonic states
are hard-core bosons, in the sense that one can not define two of them on the
same rung. The $\Delta$ term in equation (\ref{sigma})
describes the gap energy of the
magnon and the pair states, the $J$ term stands for the hopping and the
spontaneous creation/destruction process of these states, and the $V$
term describes their nearest-neighbor interaction.
This quantum nonlinear $\sigma$ model can in principle
also be derived in higher dimensions from a microscopic $SO(5)$
symmetric model\cite{henley,rabello,burgess2}, by
introducing a superspin vector as a
Hubbard-Stratonovich decoupling field, and integrate out the fermionic
degrees of freedom in a gradient expansion. However, we shall proceed
more heuristically here\cite{eder3}. For a two dimensional system, one can
imagine that the quantum $\sigma$ model Hamiltonian is obtained from a ``block spin"
type of coarse-graining of the microscopic electron Hamiltonian, and is defined
on a lattice with twice the lattice constant compared to the microscopic electron
model. (This doubled unit cell is the minimal size needed to define the
local AF and d wave SC order parameters).
Therefore, each site $x$ in the effective model correspond
to a plaquette of
the microscopic electron model. (On a ladder, this corresponds to going from the
lattice sites to ladder rungs). The $SO(5)$ singlet state $|\Omega\rangle$ corresponds
to a ``RVB" type of singlet state, while the five-fold states $t_a^\dagger|\Omega\rangle$
describe the triplet magnon states, and the $d$-wave hole and particle pair states
on a plaquette. Unlike the ladder case, the magnon and the $d$- wave pair states
could condense in the ground state to form AF and SC broken symmetry states.
In fact, Eder\cite{eder3} has recently shown that properties of the AF states
can be described by a coherent state of magnon condensation on top of a uniform spin
liquid state. Our model therefore describes competition among the ``RVB" type
of singlet vacuum and the two forms of broken symmetry order.
While it is reasonable to take $J$ and $V$ to be approximately equal for magnons and
pairs, the gap energy $\Delta$ for the neutral magnons and the charged pairs
are very different in the insulating state at half-filling. In fact, their difference
is of the order of the insulating gap $U$ at half-filling.
Taking into account the hard-core condition and neglecting the nearest-neighbor interaction
$V$ for now (it has higher powers of time and space derivatives in
the continuum limit), we can express the general
anisotropic $SO(5)$ model as
\begin{eqnarray}
H &=&
\Delta_s \sum_{ x } t_\alpha^\dagger t_\alpha(x) +
\Delta_c \sum_{ x } t_i^\dagger t_i(x) \nonumber \\
&-& J_s \sum_{ <xx'> } n_\alpha(x) n_\alpha(x') -
J_c \sum_{ <xx'> } n_i(x) n_i(x')
\label{anisotropic}
\end{eqnarray}
In this paper we shall use the convention where $a,b,..=1,2,3,4,5$ denote
the superspin indices, $\alpha,\beta,..=2,3,4$ denote the spin indices
and $i,j=1,5$ denote the charge indices, and repeated indices are summed over.
The main focus of our paper is to consider the limit where
$\Delta_c >> \Delta _s$.
Let us define the charge eigen-operators $t_h$ and $t_p$ as
\begin{equation}
t_1 = \frac{1}{\sqrt{2}} (t_h + t_p) \ \ \
t_5 = \frac{1}{i\sqrt{2}}(t_h - t_p)
\label{p-h}
\end{equation}
From this definition, it is clear that $t_h^\dagger$ is the
creation operator for a hole pair and $t_p^\dagger$ is the
creation operator for a particle pair. We can introduce a
chemical potential term
\begin{equation}
H_\mu = \mu \sum_x (t_p^\dagger t_p(x) - t_h^\dagger t_h(x))
\end{equation}
to describe the effects of doping. In the presence of this
chemical potential term, the gap energy of the hole
and particle pairs are $\Delta_c - \mu$ and $\Delta_c + \mu$
respectively. A chemical potential of the order of the charge
gap $\Delta_c$ is needed to induce a metal-insulator transition
in this system. Near such a transition point, the gap
energy of the hole pair
\begin{equation}
\tilde \Delta_c = \Delta_c - \mu
\end{equation}
can be comparable to the spin gap $\Delta_s$, while the
gap towards a particle pair excitation is of the order of twice
the charge gap, and needs to be projected out of the spectrum
in the low energy limit.
Therefore, within this formalism, the Gutzwiller projection is
equivalent to restricting ourselves to the projected Hilbert
space where
\begin{equation}
t_p(x) |\Psi\rangle = 0
\label{constraint1}
\end{equation}
at every site $x$. Within this projected Hilbert space, the
projected Hamiltonian takes the form
\begin{eqnarray}
H &=&
\Delta_s \sum_{ x } t_\alpha^\dagger t_\alpha(x) +
\tilde \Delta_c \sum_{ x } n_i(x) n_i(x) \nonumber \\
&-& J_s \sum_{ <xx'> } n_\alpha(x) n_\alpha(x') -
J_c \sum_{ <xx'> } n_i(x) n_i(x')
\label{Hamiltonian}
\end{eqnarray}
This Hamiltonian has no parameters of the order of $U$, and it is
reasonable to expect $\Delta_s\sim\tilde\Delta_c$ and $J_s\sim J_c$.
We see that the form of the Hamiltonian hardly changes from the
unprojected model, but the
definition of $n_1$ and $n_5$ is changed from
\begin{eqnarray}
n_1 &=& \frac{1}{\sqrt{2}} (t_1 + t_1^\dagger) =
\frac{1}{2} (t_h + t_p + t_h^\dagger + t_p^\dagger) \nonumber \\
n_5 &=& \frac{1}{\sqrt{2}} (t_5 + t_5^\dagger) =
\frac{1}{2i} (t_h - t_p - t_h^\dagger + t_p^\dagger)
\label{before}
\end{eqnarray}
to
\begin{equation}
n_1 = \frac{1}{2} (t_h + t_h^\dagger) \ \ \
n_5 = \frac{1}{2i} (t_h - t_h^\dagger)
\label{after}
\end{equation}
From equation (\ref{before}), we see that $n_1$ and $n_5$ commute
with each other before the projection. However, after the projection,
they acquire a nontrivial commutation relation, as can be seen from
equation (\ref{after}):
\begin{equation}
[n_1,n_5] = i/2
\end{equation}
Therefore, the Gutzwiller projection can be analytically implemented
in the $SO(5)$ theory by retaining the form of the Hamiltonian and
change only the quantization condition.
\section{Analogy with lowest-Landau-level projection}
\label{LLL}
The discussions outlined above reveal a deep connection between
the Gutzwiller projection within the $SO(5)$ formalism and the
projection onto the lowest Landau level (LLL) in the context of
the fractional quantum Hall effect. Consider the
problem of a charged particle
in a strong magnetic field $B$ and a rotationally symmetric potential
$V(X,Y)$. In the absence of a magnetic field, all eigenstates form
irreducible representations of the two dimensional rotation
group $O(2)$, characterized by integral eigenvalues of the angular
momentum operator
\begin{equation}
L_Z = X P_Y - Y P_X
\end{equation}
However, in the presence of a strong magnetic field and projected into
the LLL, only negative eigenvalues of $L_Z$ are realized. This is
analogous to the situation encountered here. The local charge operator
in the $SO(5)$ theory takes the form of the angular momentum in the
$n_1-n_5$ plane as given by equation (\ref{charge}). When doubly
occupied sites are locally projected out, the local charge operator, or
the angular momentum in the $n_1-n_5$ plane, takes only negative values.
Since the chemical potential couples directly the angular momentum in the
$n_1-n_5$ plane, it plays the role of a fictitious magnetic field
threading every $SO(5)$ rotor in the $n_1-n_5$ plane. The Landau level
spacing $\hbar \omega_c$ is analogous to the charge gap $\Delta_c$
encountered here, and both are taken to be infinity in the projected
models. After the projection, the Hamiltonian in the Landau level problem
retains its $O(2)$ symmetric form,
\begin{equation}
H = V(X,Y)
\end{equation}
although a new quantization condition is imposed between $X$ and $Y$, as
given by equation (\ref{guiding-center}). This is analogous to the
observation we made here that the Hamiltonian formally retains a
$SO(5)$ symmetric form after the projection (\ref{constraint1}), but the
quantum dynamics is changed due to the non-trivial commutator between
$n_1$ and $n_5$. In both cases only a part of the full symmetry multiplets
remain after the projection. However, the formal symmetry of the
Hamiltonian has direct physical manifestations despite the projection.
For example, in the LLL problems, semi-classical orbits of the guiding
center coordinates are still $O(2)$ symmetric. In our case, we shall see
that the static potential for the superspin vector can still be $SO(5)$
invariant despite the projection.
Perhaps the most explicit way to establish the precise connections between
these two problems is to consider the constraints on the wave function.
In the symmetric gauge of the LLL problem, the annihilation operator for
the cyclotron coordinates takes the form\cite{qhe-book}
\begin{equation}
a= \partial_{\bar z} + z/4
\end{equation}
where $z=X+iY$ and $\bar z=X-iY$. Projection onto LLL requires
\begin{equation}
a \Psi(z,\bar z) = 0
\label{constraint2}
\end{equation}
which determines the form of the LLL wave function to be
\begin{equation}
\Psi(z,\bar z) = f(z) e^{-z\bar z/4}
\end{equation}
where $f(z)$ is a holomorphic function of $z$ only.
This holomorphic condition also places strong constraints
in many-body systems and led to the celebrated Laughlin's
wave function. Our
no-double-occupancy constraint (\ref{constraint1}) is
analogous to the LLL constraint (\ref{constraint2}).
In fact from equations (\ref{p-h}) and (\ref{before}),
we obtain
\begin{equation}
t_p = \frac{1}{2} ( z + 2 \partial_{\bar z})
\end{equation}
where $z=n_1-in_5$ and $\bar z=n_1+in_5$. For a single
unprojected $SO(5)$ rotor, the wave function
$\Psi(n_a)$ is a function of the superspin
coordinates. However, the Gutzwiller projection
(\ref{constraint1}) restricts the wave function to be
\begin{equation}
\Psi(n_1,n_2,n_3,n_4,n_5) = f(z=n_1-in_5,n_2,n_3,n_4) e^{-z\bar z/2}
\end{equation}
where $f(z,n_2,n_3,n_4)$ is a holomorphic function of $z$.
For a collection of $SO(5)$ rotors, the superspin coordinates are
themselves functions of the lattice sites $x$, and
$\Psi[n_a(x)]$ is a functional is the superspin
coordinates at each site. For the projected $SO(5)$ models, this
functional is restricted to take the form
\begin{equation}
\Psi[n_a(x)] = f(z(x),n_\alpha(x)) \prod_{x} e^{-z\bar z(x)/2}
\label{functional}
\end{equation}
where $f(z(x),n_\alpha(x))$ is a holomorphic functional of
$z(x) = n_1(x)-in_5(x)$.
The formal but precise analogy between the two types of projection
allows us to introduce the concept of a chiral $SO(5)$ rotor.
This is a system of rotors with $SO(5)$ invariant potential and
coupling, however, the rotation within the $n_1-n_5$ plane is
chiral, {\it i.e.} only one sense of the rotation is allowed.
Such a system of chiral $SO(5)$ rotors is described by the
wave functional in equation (\ref{functional}).
\section{$SO(5)$ Symmetry of the ground state energy}
\label{ground-state}
Having discussed the general notions of the projected $SO(5)$ model,
we are now in a position to explore the phase diagram of this model.
As we commented earlier, the projected $SO(5)$ model describes the
competition and unification of the spin liquid, AF and the SC states.
In the original unprojected $SO(5)$ symmetric model, not only are the AF
and SC states degenerate in energy, but they are also degenerate with
all the intermediate coexistence states.
This points out a route from AF to SC with no potential barrier,
and introduces the concept that the metal-insulator transition in the
high $T_c$ systems can be viewed as a smooth rotation of the $SO(5)$
superspin. One of the key questions to be answered in this work is
what happens to the picture in the case of projected $SO(5)$ symmetry.
In anticipation of the competition of the states discussed above, we
construct a class of variational wave functions in the coherent
state representation:
\begin{equation}
|\Psi\rangle = \prod_x (\cos\theta(x)+\sin\theta(x)(m_\alpha(x)t^\dagger_\alpha(x)
+\Delta(x) t^\dagger_h(x))) |\Omega\rangle
\label{variational}
\end{equation}
Here $|\Omega(x)\rangle$ denotes a local singlet state defined by
$t_\alpha(x)|\Omega(x)\rangle=t_h(x)|\Omega(x)\rangle=0$ and $|\Omega\rangle$
is a product state of these local singlets, $|\Omega\rangle=\prod_x|\Omega(x)\rangle$.
$\theta(x)$ is a local variational parameter describing the competition
between long range order and quantum disorder.
For $\theta(x)=0$ our variational
wave function describe a spin singlet ground state, while a non-zero
value of $\theta$ describes a coherent state formed by the
local singlet, the magnon or the hole pair state.
This wave function is a generalization of the coherent state
description of a AF state in terms of a magnon condensate\cite{eder2,eder3}.
As we shall see from the following equation (\ref{superspin_vector}),
$sin2\theta$ stands for the
length of the $SO(5)$ superspin vector. $m_\alpha(x)$ and $\Delta(x)$
are general complex variational parameters describing the local
amplitude for magnons and hole pairs. We notice that this wave function
satisfies both the Gutzwiller constraint (\ref{constraint1}) and
the hard-core constraint for magnons and hole pairs exactly.
It is easy to see that
\begin{eqnarray}
\langle\Psi|n_\alpha(x)|\Psi\rangle &=& \frac{1}{\sqrt{2}} sin2\theta(x)
Re(m_\alpha(x))\nonumber \\
\langle\Psi|n_1(x)|\Psi\rangle &=& \frac{1}{2} sin2\theta(x)
Re(\Delta(x))\nonumber \\
\langle\Psi|n_5(x)|\Psi\rangle &=& \frac{1}{2} sin2\theta(x) Im(\Delta(x))
\label{superspin_vector}
\end{eqnarray}
where $Re$ and $Im$ denote the real and imaginary parts of a complex
number. The coupling terms in the projected $SO(5)$ Hamiltonian depend only
on $n_\alpha(x)$, $n_1(x)$ and $n_5(x)$. Therefore, the coupling energy
depends only on the real part of $m_\alpha(x)$ while it depends on both the
real and imaginary parts of $\Delta(x)$. Therefore, for discussing the ground state
wave functions, we can assume without
loss of generality that $m_\alpha(x)$ is real and $\Delta(x)=m_1(x)+im_5(x)$.
The normalization condition $\langle\Psi|\Psi\rangle=1$ can be implemented by
the constraint that
\begin{equation}
m_a^2(x) = (m_1^2 + m_5^2 + m_\alpha^2)(x) = 1
\label{normalization}
\end{equation}
Therefore, we see that although we have completely projected out the
particle pair states, the local degrees of freedom can still be represented
by a vector on a five dimensional sphere.
Uniform states are obtained by taking all parameters to be constant. For
$\Delta=0$ and $sin2\theta\neq 0$, our wave function $|\Psi\rangle$
describes a pure AF state with the following properties:
\begin{eqnarray}
\langle\Psi|Q|\Psi\rangle &=& \langle\Psi|\sum_x t^\dagger_h t_h(x)|\Psi\rangle =0 \nonumber \\
\langle\Psi|N_\alpha|\Psi\rangle &=& \langle\Psi|\sum_x n_\alpha(x)|\Psi\rangle
= N \frac{1}{\sqrt{2}} sin2\theta \ m_\alpha \nonumber \\
\langle\Psi|S_\alpha|\Psi\rangle &=& \langle\Psi|\sum_x i\epsilon^{\alpha\beta\gamma}
t^\dagger_\beta t_\gamma(x)|\Psi\rangle =0 \nonumber \\
\langle\Psi|S^2|\Psi\rangle &=& \langle\Psi|(\sum_x S(x))^2|\Psi\rangle
= N 2 \sin^2\theta +N \sin^4\theta(1-m^4_{\alpha})
\label{AF-state}
\end{eqnarray}
where $N$ is the number of lattice sites. Equation (\ref{AF-state}) describes a
half-filled state with a macroscopic Neel magnetization,
and vanishing uniform magnetization. Furthermore, this state is composed as a
linear superposition of eigenstates with different values of the total spin, and
the fluctuation of the total spin scales like
$\sqrt{\langle S^2\rangle} \propto \sqrt{N}$, just as one expects
from a standard Neel state.
On the other hand, for $m_\alpha=0$ and $sin2\theta\neq 0$, $|\Psi\rangle$
describes a pure SC state with the following properties:
\begin{eqnarray}
& & \langle\Psi|Q|\Psi\rangle = \langle\Psi|\sum_x t^\dagger_h t_h(x)|\Psi\rangle =
N \sin^2 \theta \nonumber \\
& & \langle\Psi|N_1 + i N_5|\Psi\rangle = \langle\Psi|\sum_x (n_1 + i n_5)(x)|\Psi\rangle
= N \frac{1}{2} sin2\theta \ (m_1+im_5) \nonumber \\
& & \langle\Psi|Q^2|\Psi\rangle - \langle\Psi|Q|\Psi\rangle^2 = N \sin^2\theta \cos^2 \theta
\label{SC-state}
\end{eqnarray}
Equation (\ref{SC-state}) describes a state with a finite doping density
and a finite SC order parameter. Just as in the standard BCS case,
this state is composed as a
linear superposition of eigenstates with different values of the total charge, and
the fluctuation of the total charge scales like
$\sqrt{\langle Q^2\rangle-\langle Q\rangle^2} \propto \sqrt{N}$,
just as one expects from a standard SC state.
However, besides these two {\it pure} states, there is a class of {\it mixed} states
which interpolates between the pure AF and SC states. Taking $m_1=\sin\alpha$ and
$m_2=\cos\alpha$, we see that the mixed states
have the following property:
\begin{eqnarray}
& & \langle\Psi|Q|\Psi\rangle = N \sin^2\theta \sin^2\alpha \nonumber \\
& & \langle\Psi|N_1 + i N_5|\Psi\rangle = N \frac{1}{2} sin2\theta \sin\alpha \nonumber \\
& & \langle\Psi|N_2|\Psi\rangle = N \frac{1}{\sqrt{2}} sin2\theta \cos\alpha
\label{mixed}
\end{eqnarray}
Therefore, we see that there is a continuous family of intermediate mixed states
interpolating between the pure AF state at half-filling and the pure SC
state with finite doping density. As the $SO(5)$ angle $\alpha$ rotates continuously
from a pure AF state with $\alpha=0$ to a pure SC state with $\alpha=\pi/2$,
the hole density of the mixed state interpolates
continuously between these two limits. Therefore, our wave function gives a unified
description of AF and SC and points out a precise route from AF to SC as
the doping level is varied. In order for this route, or small deviations from this
route, to be physically realized in the high $T_c$ superconductors, we have to
demonstrate that there is no large energy barrier for the intermediate mixed states,
or that the ground state energy is approximately independent of the $SO(5)$
mixing angle $\alpha$. In particular, we have to show that the energy barrier is
independent of the Hubbard energy $U$, in the limit of large $U$. In the following, we
shall investigate this question.
The energy functional $\langle\Psi|H|\Psi\rangle$
describes the coupling between these five dimensional vectors $m_\alpha(x)$, and
it is given by
\begin{eqnarray}
& & \langle\Psi|H|\Psi\rangle = E(\theta(x),m_a(x)) \nonumber \\
&=& -\frac{J_s}{2} \sum_{xx'} sin2\theta(x) sin2\theta(x') m_\alpha(x) m_\alpha(x')
- \frac{J_c}{4}\sum_{xx'} sin2\theta(x) sin2\theta(x') m_i(x) m_i(x') \nonumber \\
& & + \Delta_s \sum_x \sin^2\theta(x) m_\alpha^2(x)
+ \tilde \Delta_c \sum_x \sin^2\theta(x) m_i^2(x)
\label{energy}
\end{eqnarray}
This ground state energy functional describes a systems of coupled rotors
satisfying the constraint (\ref{normalization}). At the point $J_c=2J_s$
and $\tilde \Delta_c=\Delta_s$ in parameter space,
this rotor model is exactly $SO(5)$ symmetric. This is a central observation
of this work.
From this consideration we learn a very important lesson about the compatibility
of the Mott insulating gap and the idea of a smooth $SO(5)$ rotation from AF to SC.
As we have seen, the large asymmetry between the charge and spin gap at half-filling
necessitates the removal of the $Q(x)>0$ part of the $SO(5)$ multiplets, therefore,
the dynamics close to half-filling has to be modified. But the static potential
governing the transition from AF to SC can remain $SO(5)$ symmetric, and in particular,
the energy barrier separating these two states can remain small in the limit where the
Mott insulating gap tends to infinity.
We also observe a crucial difference between
the projected and the unprojected $SO(5)$ models. In the unprojected $SO(5)$ model
with the full $SO(5)$ symmetry, AF and SC states are degenerate at half-filling,
and the rotation between these two states can be continuously performed without changing
the density to going away from half-filling. This case is similar to the well-known
degeneracy between the CDW state and the $s$ wave SC state for the negative $U$
Hubbard model at half-filling. In the projected $SO(5)$ model, where all particle
pair states have been locally removed, the $SO(5)$ rotation from the AF to SC states
are accompanied by the continuous change of the hole density, and a pure SC state
can only be reached at a finite critical hole doping density $\rho_c=\sin^2\theta$.
While the unprojected $SO(5)$ symmetry is only valid at half-filling, the
projected $SO(5)$ symmetry can be valid for a range of doping concentration
$0<\rho<\rho_c$, since all these doping concentrations correspond to the same value
of the chemical potential $\mu=\mu_c$ at which the ground state energy (\ref{energy})
is $SO(5)$ symmetric.
However, it should be pointed out that the projected $SO(5)$ symmetry at
$\mu=\mu_c$ has only been demonstrated within the variational mean field
approximation. This corresponds to the semiclassical limit, and becomes
exact only in the large $s$ limit, where $s$ labels the representation
of the local $SO(5)$ group at a given site. Quantum fluctuations
can be systematically investigated
as a $1/s$ expansion. Assuming uniform ground states, we have studied the
effect of zero point fluctuations in Sec. \ref{fluctuations}
and found that at $\mu=\mu_c$,
the intermediate mixed states have slightly higher energy than the AF and
SC state. Therefore, quantum fluctuation leads to a slight breaking of the
projected $SO(5)$ symmetry. The important point here is that this symmetry
breaking effect can be systematically controlled in the semi-classical $1/s$
expansion, and certainly is independent of the Coulomb energy scale $U$.
This fluctuation would induce a first order transition and
predict phase separation of AF and SC states at
$\mu=\mu_c$. However, there are also other competing interactions such
as nearest-neighbor and next-nearest-neighbor interactions which tend to
reduce the barrier, and could also lead to non-uniform states like stripes.
Due to the complexity of the calculations, we
shall defer the detailed studies of these competing effects to future works.
\section{Phase Diagram}
\label{phase}
In this section we investigate the phase diagram of the projected $SO(5)$ model
within the framework of the variational wave function (\ref{variational}).
Taking uniform values of the variational parameters $\theta$,
$m_x=\cos\alpha$, $m_1=\sin\alpha$ and $m_y=m_z=m_5=0$, the variational energy
in (\ref{energy}) reduces to
\begin{equation}
\frac{E(\theta,\alpha)}{N} = -J_s \sin^2 2\theta \cos^2\alpha
-\frac{J_c}{2} \sin^2 2\theta \sin^2\alpha + \Delta_s \sin^2 \theta \cos^2\alpha
+\tilde\Delta_c \sin^2 \theta \sin^2\alpha
\end{equation}
In the following, we shall mainly study the $SO(5)$ symmetric case, and
take $J_c=2J_s=2J$. Defining $x=\sin^2\theta$ and $y=\cos^2\alpha$, and the
dimensionless coupling constants $\epsilon\equiv \frac{E(x,y)}{4JN}$,
$\delta_s\equiv \frac{\Delta_s}{4J}$ and
$\tilde\delta_c\equiv \frac{\Delta_c}{4J}$, we obtain
\begin{equation}
\epsilon(x,y) = x^2-x+\tilde\delta_c x+
(\delta_s-\tilde\delta_c)xy
\label{mean_energy}
\end{equation}
We shall minimize (\ref{mean_energy}) with respect to $x$ and $y$, subject to
the condition that $0\leq x,y\leq 1$.
The phase diagram can be plotted in the two dimensional parameter
space of $\tilde\delta_c$ and $\delta_s$. We
notice that $\epsilon(x,y)$ depends linearly on $y$, therefore,
for $\delta_s > \tilde\delta_c$ we obtain $y_{min}=0$ and
\begin{eqnarray}
& 0<x_{min}=\frac{1-\tilde\delta_c}{2}<1
& \ \ \ for \ \ \ -1<\tilde\delta_c < 1 \nonumber \\
& x_{min}=0 & \ \ \ for \ \ \ \tilde\delta_c > 1 \nonumber \\
& x_{min}=1 & \ \ \ for \ \ \ \tilde\delta_c < -1
\end{eqnarray}
Similarly, for $\delta_s < \tilde\delta_c$ we obtain $y_{min}=1$ and
\begin{eqnarray}
& 0<x_{min}=\frac{1-\delta_s}{2}<1
& \ \ \ for \ \ \ -1< \delta_s < 1 \nonumber \\
& x_{min}=0 & \ \ \ for \ \ \ \delta_s > 1 \nonumber \\
& x_{min}=1 & \ \ \ for \ \ \ \delta_s < -1
\end{eqnarray}
From these equation, we can determine the phase diagram as shown in Fig. (\ref{fig2}).
\begin{figure*}[h]
\centerline{\epsfysize=10.0cm
\epsfbox{fig2.eps}
}
\caption{Phase diagram of the projected $SO(5)$ model in the $\delta_s$ versus
$\tilde\delta_c$ plane. Phase boundaries are depicted by the solid lines.
Variation of the chemical potential traces out a
one dimensional trajectory as shown on the dotted line.}
\label{fig2}
\end{figure*}
There are seven different phases on this phase diagram.
$x_{min}=0$ corresponds to a quantum disordered
singlet state with no condensed bosons. $x_{min}=1$ and $y_{min}=0$ corresponds
to a quantum disordered state with completely filled hole pairs.
$x_{min}=1$ and $y_{min}=1$ corresponds
to a quantum disordered state with completely filled magnons.
$0<x_{min}<1$ and $y_{min}=1$ describe a pure AF phase, while
$0<x_{min}<1$ and $y_{min}=0$ describes a pure SC phase.
When $-1< \delta_s =\tilde\delta_c < 1$
a continuous family of mixed AF/SC states labeled by a free superspin angle
$0<\alpha<\pi/2$ is realized, while for
$\delta_s =\tilde\delta_c < -1$
a continuous family of quantum disordered states labeled by a free superspin angle
$0<\alpha<\pi/2$ is obtained.
The system traces out a one dimensional trajectory in this two dimensional
phase diagram as the chemical potential is increased, as depicted in
Fig. (\ref{fig2}).
Increasing the chemical potential decreases the $\tilde\delta_c$ parameter
while holding $\delta_s$ constant. $\delta_s$ describes the degree
of quantum spin fluctuations in the system, since in the AF phase, the
size of the Neel moment
\begin{equation}
m_{AF}=\sqrt{\frac{1}{2}(1-\delta_s^2)}
\label{moment}
\end{equation}
decreases with increasing $\delta_s$.
For $\delta_s<1$, the system goes through
a phase transition from AF to SC at $\mu=\mu_c=\Delta_c-\Delta_s$. At this
critical value of the chemical potential, the $\theta$ parameter remains
fixed, but the $\alpha$ parameter changes continuously from $0$ to $\pi/2$,
and correspondingly, the density changes from $0$ to
\begin{equation}
\rho_c\equiv \sin^2\theta=x_{min}=\frac{1-\delta_s}{2}
\label{rho_c}
\end{equation}
This behavior gives a density versus chemical potential diagram as shown in
Fig. (\ref{fig3}).
\begin{figure*}[h]
\centerline{\epsfysize=4.0cm
\epsfbox{fig3.eps}
}
\caption{Density versus chemical potential relation in the projected
$SO(5)$ model. Unlike the case of a generic first order transition,
the ground state in the density range $0<\rho<\rho_c$ is uniform,
rather than phase separated.}
\label{fig3}
\end{figure*}
As we see, for densities in the range $0<\rho<\rho_c$, the
system is infinitely compressible since $\partial\rho/\partial\mu=\infty$.
For $\mu>\mu_c$ or $\rho>\rho_c$, the system has a finite compressibility
of $\partial\rho/\partial\mu=1/8J$.
It is interesting to plot both the AF and the SC order parameters
$\langle n_1\rangle$ and $\langle n_2\rangle$ as a function of the density
for the whole range of $0<\rho<1$. We will restrict to the case of
$0<\delta_s<1$ where the undoped state is a AF state. We obtain the following
doping dependence of the SC order parameter:
\begin{equation}
\langle n_1\rangle = \left\{ \begin{array} {l}
\sqrt{\rho(1-\rho_c)} \ \ \ for \ \ \ \rho<\rho_c \\
\sqrt{\rho(1-\rho)} \ \ \ for \ \ \ \rho>\rho_c
\end{array} \right.
\end{equation}
and the doping dependence of the of the AF order parameter:
\begin{equation}
\langle n_2\rangle = \left\{ \begin{array} {l}
\sqrt{2(1-\rho_c)(\rho_c-\rho)} \ \ \ for \ \ \ \rho<\rho_c \\
0 \ \ \ for \ \ \ \rho>\rho_c
\end{array} \right.
\end{equation}
These behaviors are depicted in Fig. (\ref{fig4}).
\begin{figure*}[h]
\centerline{\epsfysize=6.0cm
\epsfbox{fig4.eps}
}
\caption{AF ($\langle n_2\rangle$) and SC ($\langle n_1\rangle$)
order parameters versus density in the
projected $SO(5)$ model.}
\label{fig4}
\end{figure*}
We note several interesting features of the phase diagram. First of all,
we can use the energy functional (\ref{energy}) as a starting point for
a finite temperature classical fluctuation analysis, and estimate the
transition temperature due to the classical fluctuations.
Within such a framework, the 3D AF ($T_N$) and SC ($T_c$) transition
temperatures are proportional to the stiffness of the spin and the
phase fluctuations, which are in turn proportional
to $\langle n_2\rangle^2$ and $\langle n_1\rangle^2$
respectively. Therefore, Fig. (\ref{fig4}) gives an approximate estimate of the
transition temperatures. For small $\rho_c$, we see that there is
a sharp drop of $T_N$ and a maximum of $T_c$ at $\rho=1/2$.
The system is a AF insulator at doping $\rho=0$ and a pure SC
state for $\rho>\rho_c$. The reason for a maximum of $T_c$ at
$\rho=1/2$ is due to the strong correlation of the charged bosons.
The charge bosons have a hard-core
interaction, therefore, they are insulating at both $\rho=0$ and
$\rho=1$ and have maximal charge stiffness at $\rho=1/2$. We can
perform a rough translation of this optimal doping value in our
effective model to the microscopic model. Since our effective
model is defined on a unit cell with twice the lattice spacing
of the microscopic model, $\rho=1/2$ therefore describes one
hole pair per four sites in the microscopic model, or a doping
of $x=1/4=25\%$ in the conventional language. This crude
argument tends to overestimate the value for optimal doping,
since it neglects the effects of unpaired electrons. However,
considering the crudeness of the estimate, it is still
reasonably close to the optimal doping $x=15\%$ observed in the LSCO
family of high $T_c$ superconductors.
In the regime of $0<\rho<\rho_c$, the system is a coherent mixture
of AF and SC order. For this
entire range of densities, the system has a projected $SO(5)$
symmetry within the variational approximation discussed above.
The projected $SO(5)$ symmetry manifests itself in terms of
a infinite compressibility in the region $0<\rho<\rho_c$
and, as we shall see in next section,
a charge mode with a dispersion relation
$\omega\sim k^2$. Since such a state is rather
unusual, and maybe highly susceptible to density fluctuations,
we would like to discuss more detailed physical properties in
this region.
First let us comment on the fact that there are familiar physical
systems whose uniform ground states are infinitely compressible.
The free boson model is certainly such an example, and the density
mode also has a $\omega\sim k^2$ dispersion relation. But the
infinite compressibility is due to the absence of the interaction,
which is not characteristic of the strongly interaction system
considered here. A less trivial example is the spin $1/2$ XXZ ferromagnetic
Heisenberg model, given by the Hamiltonian,
\begin{equation}
{\cal H} = J \sum_{i,j} (S^x_i S^x_j + S^y_i S^y_j + \Delta S^z_i S^z_j)
\end{equation}
where $J<0$ and the sum extends over nearest neighbor sites of a square lattice.
This model can be interpreted as quantum hard-core boson model, where
the fully polarized spin down state could be identified with the
vacuum of the bosons, the XY part of the Hamiltonian describes the
hopping of the bosons and the last term describes the nearest neighbor
attraction between the bosons if $\Delta>0$. When $\Delta>1$, the system
is in the Ising limit, and the spontaneous breaking of the $Z_2$ symmetry
implies phase separation of the bosons. On the other hand, when
$0<\Delta<1$, the system is in the XY limit, and the ground state is
a superfluid. Therefore, the anisotropy parameter describe the
competition between superfluidity and phase separation. At $\Delta=1$,
the system has a $SU(2)$ symmetry and the dispersion relation becomes
quadratic. Different directions of the
ferromagnetic polarizations are degenerate and
can be changed without any energy cost.
Since the $z$ component of the ferromagnetic polarization is identified
with the total density of the hard-core bosons, the $SU(2)$ vacuum
degeneracy implies infinite compressibility of the corresponding boson
system for the entire range of boson densities $0<\rho<1$.
These two examples illustrates that there is nothing intrinsically
pathological about having a system with infinite compressibility.
The second example is more generic, and shows that uniform states
with infinite compressibility can be obtained in systems on the
verge of phase separation, and the infinite compressibility can be
ensured by symmetry. Both of these properties are also shared by
the projected $SO(5)$ model. These models are on the verge of
phase separation into AF and SC phases, and the infinite compressibility
is a result of the projected $SO(5)$ symmetry.
It indicates that
small perturbations, such as next nearest neighbor interactions, quantum fluctuations,
and quenched disorder will be very important to determine the true ground state.
With such perturbations, the ground state is expected to be unstable toward
the experimentally reported textures
such as the spin glass, stripes and incommensurate spin density waves.
Next let us investigate the phenomenological consequence of this
remarkable property. One of the most puzzling
properties of the high $T_c$ superconductors is the constant
chemical potential in the underdoped samples. For LSCO systems,
where the doping level can be varied continuously by the $Sr$
concentration, this effect has been dramatically observed in
the ARPES experiments and the constant chemical potential
persists from the weakly doped insulator to optimally doped
superconductor\cite{fujimori}. In fact numerical calculations on the
Hubbard model also reveal similar divergent behavior of the
compressibility as the metal-insulator transition is approached from
the metallic side\cite{imada}.
The simplest explanation of the small chemical potential shift
is a two phase mixture with different densities at a first order
phase transition. If the system globally phase separates into
two different spatial regions with different charge densities
but the same free energy densities, the added charges only change
the proportion of mixture of the two phases and do not change
the energy, therefore, $\partial\mu/\partial\rho=0$. However,
this situation of global phase separation can certainly not
occur in a system with long ranged Coulomb interaction and is
ruled out in the real high $T_c$ system.
A phenomenon possibly related to the tendency of phase separation
is the formation of stripes\cite{zaanen,kivelson,scalapino2}.
A stripe state can be viewed as
microscopic phase separation of AF and SC into alternating regions,
where each region has different charge density and the same free energy
density. However, a crucial difference between the global phase
separation and this picture of microscopic phase separation is that
the stripe state has infinitely many surfaces between AF and SC,
and the surface energy makes a finite contribution to the total
energy in the thermodynamic limit. In this picture, doping can be
accomplished by converting AF stripes into SC stripes, thereby
creating more surfaces separating AF and SC regions. Finite
doping density therefore leads to a finite density of surfaces
and the accumulated surface energy would in general lead to
a shift of the chemical potential {\it under a generic situation}.
Additional physical conditions are needed to ensure the constant
chemical potential in the stripe phase.
Therefore, the absence of the chemical potential shift
places a very strong constraint on possible theoretical
explanations. The projected $SO(5)$ model proposed in this work
offers a possible explanation for the absence of chemical
potential shift. At a critical value of the chemical potential
where the AF and SC states have degenerate energy density,
we can have three situations, where
the intermediate mixed states have higher, lower or degenerate
energy compared to the AF and SC states. When the intermediate
states have higher energy, the system will go through a first
order phase transition at $\mu=\mu_c$
and this will lead to global phase separation into AF and SC regions.
On the other hand, if the intermediate states have lower energy, there
exists a range of chemical potential $\mu_{c1}<\mu<\mu_{c2}$ where
the mixed phase has a uniform and continuously varying density.
In this case, $\partial\mu/\partial\rho\ne 0$ is obtained.
When the region $\mu_{c1}<\mu<\mu_{c2}$ shrinks to zero, we obtain
the limiting $SO(5)$ symmetric case where the system is on the boundary
between a first order transition and two second order phase transitions.
In this case, $\partial\mu/\partial\rho=0$ for a range
of densities $0<\rho<\rho_c$. Therefore, if we restrict to
ground states where the density is not globally inhomogeneous,
the absence of the chemical shift directly implies
$SO(5)$ symmetry.
Within this model, we can therefore define a experimental procedure
to measure one of the most crucial parameter of the theory,
namely $\rho_c$. Using the experimental ARPES data for LSCO system
we would identify $\rho_c$ to be approximately the same as the
optimal doping density. We recall that $\rho_c$ is also a measure
of the degree of the quantum spin fluctuation in the system.
For the bi-layer materials such as YBCO and BISCO superconductors,
the quantum spin fluctuation are stronger due to the inter-layer
spin exchange, and we would predict that $\rho_c$ should be less
than the optimal doping value.
\section{Collective modes}
\label{modes}
Having discussed the ground state properties and the phase diagram of the model,
we are now in a position to study the collective excitations of the model. We have
argued that the ground state energy can remain $SO(5)$ symmetric despite the
projection. However, the projection does affect the collective
excitation spectrum near half-filling. Nonetheless, as we shall see, there remains
a unique signature of the projected $SO(5)$ symmetry in the collective excitation
spectra.
In principle, the collective excitation spectra can be obtained straightforwardly
by studying the quadratic fluctuations around the mean field minima. The resulting
quadratic boson Hamiltonian can be simply diagonalized. The main complication in the
procedure is the hard-core boson constraint, which requires
\begin{equation}
t^\dagger_\alpha t_\alpha (x) + t^\dagger_h t_h (x) \leq 1
\label{hard-core}
\end{equation}
for every site. There are several ways to implement this constraint rigorously.
One is to follow the mapping from the one-component hard-core
boson model to the $XY$ model and generalize it to a multi-component hard-core boson
model. One could also convert the above inequality constraint to an equality
constraint by introducing a boson creation and annihilation operator for the
singlet state. This approach will be implemented in the Appendix.
For simplicity of presentation, here we shall adopt a less rigorous approach and
introduce a on-site boson repulsion term
\begin{equation}
W \sum_x (t^\dagger_\alpha t_\alpha+ t^\dagger_h t_h)^2
\label{soft-core}
\end{equation}
to our Hamiltonian (\ref{Hamiltonian}) and
convert the hard-core constraint to a soft-core
constraint. We shall show later that all results can be expressed in terms
of the order parameter, which is implicitly dependent on $W$, but there is
no explicit dependence on $W$.
We have verified that all three methods give the same long wave length
spectra for the collective modes in the limit of low boson density.
In the next section, we shall present another
calculation based on the continuum effective Lagrangian method, which also reproduces
the same spectra.
To simplify presentation, we shall concentrate on the case where the ground
state energy functional (\ref{energy}) is $SO(5)$ symmetric, {\it i.e.} for coupling
constants $J_c=2J_s\equiv 2J$ and $\tilde\Delta_c=\Delta_s\equiv\Delta$. We choose
the direction of spontaneously broken symmetry to be
$\langle t_x\rangle = \langle t^\dagger_x\rangle = x$ and
$\langle t_h\rangle = \langle t^\dagger_h\rangle = y$. The extremal condition
can be easily determined to be
\begin{equation}
x^2+y^2 \equiv r^2 = \frac{4J-\Delta}{2W}
\label{extremal}
\end{equation}
The combination $x^2+y^2$ expresses the fact that the classical minimum is
$SO(5)$ symmetric. We can therefore write $x=r \cos\alpha$ and $y=r \sin\alpha$.
Expanding the boson operators as:
\begin{equation}
t_x=x+a_x, \ \ t_y=a_y, \ \ t_z=a_z, \ \ t_h=y+a_h
\label{expansion}
\end{equation}
we obtain the following quadratic Hamiltonian
\begin{eqnarray}
H &=& (\Delta +2Wr^2)\sum_x(a_{x}^{\dagger}a_x+a_{h}^{\dagger}a_h)
+ Wx^2\sum_x (a_{x}^{\dagger}+a_x)^2+ Wy^2\sum_x (a_{h}^{\dagger}+a_h)^2
\nonumber \\
&-& \frac{J}{2}\sum_{<x,x'>} (a_{x}^{\dagger}(x)+a_x(x))( a_{x}^{\dagger}(x')+a_x(x'))
-J\sum_{<x,x'>}(a_{h}^{\dagger}(x)a_h(x')+h.c.)
\nonumber \\
&+& 2Wxy \sum_{x}( a_{h}^{\dagger}(x)+a_h(x))( a_{x}^{\dagger}(x)+a_x(x))
\nonumber \\
&-& \frac{J}{2}\sum_{<x,x'>}( a_{y}^{\dagger}(x)+a_y(x))( a_{y}^{\dagger}(x')+a_y(x'))
- \frac{J}{2}\sum_{<x,x'>}( a_{z}^{\dagger}(x)+a_z(x))( a_{z}^{\dagger}(x')+a_z(x'))
\nonumber \\
&+& (\Delta +2Wr^2)\sum_x(a_{y}^{\dagger}a_y+a_{z}^{\dagger}a_z)
\label{quadratic}
\end{eqnarray}
We are in particular interested in the collective mode spectra for the
AF insulating state with $\alpha=0$, the mixed states with $0<\alpha<\pi/2$
and the SC state with $\alpha=\pi/2$ and how they connect to each other.
From this quadratic Hamiltonian we can learn a number of important features.
First we notice that the $a_y$ and $a_z$ modes
are decoupled for all ranges of $0\leq \alpha\leq \pi/2$, but most
importantly, their dispersion relations are independent of $\alpha$
and given by
\begin{equation}
\omega(k) = v_s k \ \ , \ \ v_s = 2J
\label{spin_wave}
\end{equation}
where $k\equiv a |\vec k|$ and $a$ is the lattice constant.
This is indeed a very remarkable property. At $\alpha=0$, $a_y$ and
$a_z$ modes are nothing but the transverse AF spin
wave modes. AF spin waves are usually viewed
as Goldstone modes and their existence is due to the AF long range order.
However, as $\alpha$ changes continuously from $0$ to $\pi/2$, the AF long
range order continuously diminishes until it vanishes at $\alpha=\pi/2$.
The reason that the properties of the $a_y$ and $a_z$ modes do not change
at all is due to the $SO(5)$ symmetry of this model, since the
diminishing AF order is compensated by the increasing SC order as the superspin
angle $\alpha$ is varied. As we shall see, the $a_x$ mode is the AF spin
amplitude mode at $\alpha=0$, but it becomes massless and degenerate with
the $a_y$ and $a_z$ modes at $\alpha=\pi/2$. These three modes form a massless
$\pi$ triplet mode whose existence is purely a consequence of the SC order.
Therefore, as $\alpha$ is continuously varied from $0$ to $\pi/2$, the
transverse AF spin wave modes gradually change their character to become
the $\pi$ triplet resonance of the SC state. As we shall see, for $\mu>\mu_c$,
the $\pi$ triplet mode becomes massive.
At the AF point $\alpha=0$, the spin amplitude mode $a_x$ is decoupled from the
SC mode $a_h$ and can be diagonalized separately. The dispersion for the spin
amplitude mode has the conventional massive relativistic form
for small $k$
\begin{equation}
\omega^2(k) = 16J^2 (k^2/4 + m_x^2) \ \ , \ \ m_{x}^{2} = \frac{Wx^2}{J} =
\frac{4J-\Delta}{2J}
\label{spin_amp}
\end{equation}
On the other hand, we have a massless SC Goldstone mode $a_h$ with the
following dispersion,
\begin{equation}
\omega(k) = J k^2
\label{SC-mode}
\end{equation}
This mode is an important new prediction of the $SO(5)$ theory. It is the
counterpart of the $\pi$ resonance in the AF state. In the unprojected
$SO(5)$ model, there are two such modes, with charge $\pm 2$, and they
represent gapless fluctuations from AF to SC at half-filling. In the
projected $SO(5)$ model, the charge $+2$ mode is projected out of the spectrum,
however, the charge $-2$ mode remain massless at $\mu=\mu_c$. It is
also a manifestation of the gapless fluctuation from AF to SC at half-filling,
but the SC fluctuation is hole-like, rather than both hole and particle-like
as in the unprojected case. We see again that a large Mott-Hubbard gap is
fully compatible with gapless SC fluctuation at half-filling.
Experimental detection of this mode could provide a important test of the
projected $SO(5)$ symmetry.
The $a_x$ and the $a_h$ modes also decouple in the pure SC state with
$\alpha=\pi/2$. However, their physical interpretation change.
The $a_x$ mode becomes gapless at this point with the same dispersion
as in (\ref{spin_wave}). Therefore, the three modes $a_x$, $a_y$ and $a_z$
form a gapless $\pi$ triplet mode of the pure SC state, and represent
the gapless fluctuation from SC to AF at $\mu=\mu_c$, but with a finite
hole density $\rho_c$ given in (\ref{rho_c}).
The dispersion for the $a_h$ mode is given by
\begin{equation}
\omega(k) = 2J m_h k \ \ , \ \ m_{h}^{2} = \frac{Wy^2}{J} =
\frac{4J-\Delta}{2J}
\label{phase_mode}
\end{equation}
and has the natural interpretation of a linearly dispersing phase Goldstone
mode of the SC state.
In the intermediate mixed state with $0<\alpha<\pi/2$, the $a_x$ and the
the $a_h$ modes are coupled. Diagonalization of these modes gives:
\begin{equation}
\frac{\omega^2(k)}{(4J)^2} = \left\{ \begin{array}{l}
(1+m_h^2)k^2/4+m_x^2 \\
(1+m_h^2/m_x^2)k^4/16
\end{array} \right. \ \ \
\left\{ \begin{array}{l}
m_x^2=\frac{4J-\Delta}{2J}\cos^2\alpha \\
m_h^2=\frac{4J-\Delta}{2J}\sin^2\alpha
\end{array} \right.
\label{mixed_modes}
\end{equation}
The upper massive mode has predominantly spin amplitude character,
and we see that the gap diminishes continuously until it reaches
zero at $\alpha=\pi/2$ to become the massless $\pi$ triplet mode.
The lower mode has predominantly SC fluctuation character, and has
a gapless $\omega\propto k^2$ dispersion. In the mixed region where
both the AF and the SC order parameters are non-zero, one would
naturally expect a gapless phase mode corresponding to the SC order.
However, in a interacting boson system, the phase mode is expected
to have linear dispersion on general ground. Therefore, what is
surpassing and new here is not the gapless nature of the SC mode,
but its {\it quadratic dispersion}. In order to locate the origin of the
quadratic dispersion, we have perturbed the model away from the
projected $SO(5)$ symmetric point so that a uniform
mixed state is stabilized as
a classical minimum. SC fluctuation around such a non-$SO(5)$ symmetric
point is gapless and has linear dispersion. A quadratic dispersion
is only realized at the $SO(5)$ symmetric point. Therefore, the
quadratic dispersion is a unique signature of the projected $SO(5)$
symmetry in the entire range of densities $0<\rho<\rho_c$!
To understand the physical origin of this
remarkable phenomenon, we notice that a boson system with gapless
quadratic dispersion generally has infinite compressibility. This
can be directly seen from the compressibility sum rule
\begin{equation}
\kappa \equiv \frac{\partial\rho}{\partial\mu}
= \frac{1}{2} \lim_{k\rightarrow 0}\lim_{\omega\rightarrow 0}
\chi(k,\omega) \sim
\lim_{k\rightarrow 0} \frac{k^2}{\omega^2(k)}
\label{compressibility}
\end{equation}
where $\chi(k,\omega)$ is the dynamical density correlation function.
Because of the quadratic dispersion relation, we can see explicitly that
$\chi(k)\propto 1/k^2$ for small $k$, therefore, a infinite compressibility
is obtained. On the other hand, a infinite compressibility implies that
$\frac{\partial\mu}{\partial\rho}=0$, {\it i.e.} the chemical potential
is independent of doping. But this is exactly the prediction of the
projected $SO(5)$ model! For $0<\rho<\rho_c$, the chemical potential is
pinned at the $SO(5)$ symmetric point $\mu=\mu_c$, where the
superspin vector can point in any direction. To accommodate a
long wave length fluctuation of the hole density, the system rotates
into another degenerate minimum with a different superspin angle $\alpha$
and a different hole density.
For this reason, the chemical potential does not change and the system
is infinitely compressible.
\section{Quantum corrections to the mean-field solution}
\label{fluctuations}
At the mean-field level,
the ground-state energy of the Hamiltonian
(\ref{Hamiltonian}) with (\ref{soft-core})
depends on the AF and SC order parameters $x$ and $y$ only via
their combination $r^2=x^2+y^2$ reflecting the SO(5) invariance of the
mean-field result.
However, the zero-point energy of the bosons in the
quadratic Hamiltonian gives a correction to the
ground-state energy due to quantum fluctuations. (Calculations of the quantum
fluctuation effects at the $SO(3)$ spin-flop transition have been
studied in ref. \cite{assa2}.
As we will show below, this correction turns out to depend on $x^2$
and $y^2$ separately.
For the soft-constraint case, $E_1$ can be expressed as
\begin{equation}
\label{egs}
E_1 = \frac12 \sum_k
\sum_{i} \omega_i(k) - 2 (\Delta + 2 W r^2)
\end{equation}
where $\omega_i(k)$ are the four collective modes described in
Sec. \ref{modes} (although extended to all values of $k$ in the BZ).
In a systematic $1/s$ expansion, this correction
to the ground state energy scales like $1/s$, and is therefore small
in the semi-classical limit. Alternatively,
$E_1$ can be seen as a small correction to the mean-field energy $E_0$
for small values of the parameter
$\epsilon \equiv \frac{r^2 W}{2 J}=1-\frac{\Delta}{4J}$.
However, contrary to $E_0$, $E_1$ also depends on the superspin angle
$\alpha$ and thus produces a small SO(5)-symmetry breaking.
We now evaluate the $\alpha$-dependent part of Eq. (\ref{egs}) in the
small-$\epsilon$ limit.
We thus parametrize
$\chi = 1- 2 \sin^2\alpha$ ($-1 \leq \chi \leq 1$),
differentiate
the expression (\ref{egs}) with respect to $\chi$, and expand it
up to second order in $\epsilon$.
By further transforming $(\cos k_x + \cos k_y)/2=1-q$ with $ 0\leq q\leq 2$,
the derivative of $E_1$ can be expressed as the sum of two terms
\begin{equation}
\frac{d}{d \chi} E_1 = E_{1A}' + E_{1B}' \;,
\end{equation}
where
\begin{equation}
\label{intg}
{E_{1A}'\atop E_{1B}'} = 2 J \int_0^2 d\ q \ {\cal D}( q)
{G_{A} \atop G_{B}} \;,
\end{equation}
with
\begin{equation}
\label{g1}
G_A \equiv
=
{\frac{\left( \left( {1-\sqrt{ q}} \right) \,
\epsilon \right) }{2\,{\sqrt{ q}}}} +O(\epsilon^2)\;,
\end{equation}
\begin{equation}
\label{g2}
G_B \equiv
-
{\frac{\left( {\sqrt{ q}-1} \right) \,
\left( \chi\,{\sqrt{ q}} -1 - \chi - {\sqrt{ q}}
\right) \,
\epsilon^2}{4\,
\left( 1 + {\sqrt{ q}} \right) \,
{{ q}^{{\frac{3}{2}}}}}} + O(\epsilon^3) \;,
\end{equation}
and the density of states ${\cal D}(q)$ is defined as
\begin{equation}
{\cal D}(q) = \frac{1}{4 \pi^2}
\int \ d k_x \ d k_y \ \delta\left[q-1+(\cos k_x + \cos
k_y)/2\right] \;.
\end{equation}
Unfortunately,
the terms of the $\epsilon$ expansion in $G_B$ diverge
when integrated over $ q$.
It is thus convenient to carry out the transformation $q= \epsilon z$ in
$G_B$ and {\it then} expand in powers of $\epsilon$.
We obtain
\begin{equation}
\label{g2u}
G_B = \frac{\sqrt{\epsilon}}{2} \left( \frac{1}{\sqrt{z + 1 + \chi }}
- \frac{1}{\sqrt{z}} \right) + O(\epsilon) \;.
\end{equation}
For small $\epsilon$,
the integral in $z$ can be extended
to $\infty$
and we obtain from Eq. (\ref{intg})
\begin{equation}
\label{e1bf}
E_{1B}' = 2J \
\epsilon \int_0^{\infty} d z\ {\cal D}(z \epsilon)\ G_B =
\frac{ -(\sqrt{1 + \chi }\ \epsilon^{3/2})}{\pi} +
O(\epsilon^2 \log \epsilon)\;.
\end{equation}
While in evaluating Eq. (\ref{e1bf}) we only need
the density of states at $q=0$,
${\cal D}(0) = \pi^{-1}$,
for the first term $E_{1A}'$ one needs ${\cal D}(q)$
in the whole domain $0 \leq q \leq 2$
However, $E_{1A}'$ is independent of $\chi$ and thus it merely fixes
the value of the critical chemical potential.
A numerical integration yields
\begin{equation}
\label{e1af}
E_{1A}' = 0.28 \ J \ \epsilon \;.
\end{equation}
By integrating over $\chi$ Eqs. (\ref{e1af}) and (\ref{e1bf}), we finally obtain the total contribution
to the ground-state energy correction.
\begin{equation}
\frac{E_1}{2 J} = 0.14\ \chi \ \epsilon
-\frac{2}{3 \pi} \left(\epsilon\ (1 + \chi)\right)^{3/2} + O(\epsilon^2 \log \epsilon) +
{\rm const.}\;.
\end{equation}
$E_1$ thus lifts the degeneracy as a function of $\chi$ and initially
favors
the pure superconducting phase ($\chi=-1$).
A small chemical potential term
$-\mu' y^2 = -\mu' r^2 (1-\chi)/2 $ with
$\frac{\mu' r^2}{2 J}= \frac{\mu'_c r^2}{2 J}= - 0.28 \ \epsilon +
\frac{2}{3 \pi} \ (2 \epsilon)^{3/2} $
restores the degeneracy between the
pure superconducting ($\chi=-1$) and the pure antiferromagnetic
($\chi=+1$) phases. However, due to the convexity of $E_1$ as a function of
$\chi$ there is always a barrier ($\propto J \epsilon^{3/2}$)
between the two phases since
the mixed phase always has a higher energy. This means that
at $\mu'=\mu'_c$ and for intermediate densities
the system prefers to phase separate
between the two pure phases
rather than choosing the mixed phase.
However, the important point is that this barrier remains small in the
limit $U\to\infty$.
In view of the symmetry breaking effects of the quantum fluctuations,
it would be interesting to see whether there is a limit when the
the wave function (\ref{variational}) and the projected $SO(5)$ symmetry
becomes exact. Rokhsar and Kotliar\cite{rokhsar}
have shown that this type of wave functions
are actually exact in the limit of infinite dimensions. Therefore,
besides the $1/s$ expansion, we could also use a $1/d$ expansion
(where $d$ is the space dimension) to
systematically control the $SO(5)$ symmetry breaking quantum effects.
Besides quantum fluctuations, there are also other symmetry breaking terms.
Nearest-neighbor and next-nearest-neighbor Coulomb interactions also break
the $SO(5)$ symmetry, however, their corrections to the ground state is
concave, {\it i.e.} the energy of the intermediate states are lowered.
Therefore,
they can also lead to uniform mixed states in some region of the
phase diagram. The detailed study of all these competing effects will
be carried out in subsequent works.
\section{Low energy effective Lagrangian}
\label{lagrangian}
While the projected $SO(5)$ model defined on a lattice enables us to make
some connection to the underlying microscopic physics, for most discussions
concerning the long wave length and low energy degrees of freedom, it is
desirable to have a effective continuum Lagrangian. Such a formulation can be directly
obtained by taking the long wave length limit of the projected bosonic model
discussed previously. However, in order to make the connection to the
unprojected $SO(5)$ model clearer, we shall motivate our discussion from
the original $SO(5)$ effective model.
The effective Lagrangian for a fully $SO(5)$ symmetric model takes the form of
\begin{equation}
{\cal L} = \frac{\chi}{2} (\partial_t n_a)^2 - \frac{\rho}{2} (\partial_k n_a)^2
- V(n)
\label{full_L}
\end{equation}
where $\chi$ measures the superspin susceptibility, $\rho$ measures the
superspin stiffness, $k=x,y$ denotes the spatial directions and $V(n)$
is a scalar function of the superspin magnitude $n_a^2$ only. There are
three important symmetry breaking effects connected with the presence of
a large Mott-Hubbard gap. First is an asymmetry in the scalar potential,
which can be described by a additional term
\begin{equation}
V_g(n) = + \frac{g}{2} (n_1^2 + n_5^2)
\label{V_g}
\end{equation}
which for positive $g$ favors AF at half-filling. Second is the asymmetry
between the spin ($\chi_s$) and the charge ($\chi_c$) susceptibilities, which
modifies the kinetic energy to
\begin{equation}
\frac{\chi_s}{2} (\partial_t n_\alpha)^2 + \frac{\chi_c}{2} (\partial_t n_i)^2
\end{equation}
The last symmetry breaking effect is due to the chemical potential $\mu$, which
enters the Lagrangian as a gauge coupling in the time direction, and modifies the
charge part of the kinetic energy to
\begin{equation}
\frac{\chi_c}{2} ((\partial_t n_1 + \mu n_5)^2 + (\partial_t n_5 - \mu n_1)^2)
\end{equation}
Combining these three symmetry breaking terms, we obtain
\begin{eqnarray}
{\cal L} &=& \frac{\chi_s}{2} (\partial_t n_\alpha)^2 +
\frac{\chi_c}{2} ((\partial_t n_1 + \mu n_5)^2 + (\partial_t n_5 - \mu n_1)^2) -
\frac{\rho}{2} (\partial_k n_a)^2
- V(n) - \frac{g}{2}n_i^2 \nonumber \\
&=& \frac{\chi_s}{2} (\partial_t n_\alpha)^2 + \frac{\chi_c}{2} (\partial_t n_i)^2
+ \mu\chi_c(n_5\partial_t n_1 - n_1\partial_t n_5) + \frac{\mu^2\chi_c-g}{2} n_i^2
-\frac{\rho}{2} (\partial_k n_a)^2 - V(n)
\end{eqnarray}
In the presence of a large Mott-Hubbard gap, all these three symmetry breaking
terms are of the order of $U$, {\it i.e.} $\chi_c^{-1} \sim g \sim \mu_c \sim U$.
Therefore, this Lagrangian contains high energy degrees of freedom of the order
of $U$. However, as already observed in \cite{so5}, at $\mu=\mu_c=\sqrt{g/\chi_c}$,
their effects cancel completely in the time-independent part of the
Lagrangian, and the static potential is $SO(5)$ symmetric just as in the original
unprojected model. We also observe that near the AF/SC transition point
where $\mu\sim \mu_c$, the first order time derivative term is of the order of one.
Furthermore, in the spirit of the low frequency and wave vector expansion, we
only need to retain the first order time derivative term in the charge sector
and can drop the second term in the above Lagrangian. Combining all these considerations,
we obtain the following low energy effective Lagrangian near the AF/SC transition
region, which is free of any parameters of the order of $U$:
\begin{equation}
{\cal L} = \frac{\chi_s}{2} (\partial_t n_\alpha)^2
+ (n_5\partial_t n_1 - n_1\partial_t n_5)
-\frac{\rho}{2} (\partial_k n_a)^2 - V(n)
\label{projected_L}
\end{equation}
This is exactly the Lagrangian counterpart of the $SO(5)$ projection
procedure discussed
previously in the Hamiltonian language. Dropping the second order time
derivative terms removes half of the (high energy) degrees of freedom,
and redefines the canonical conjugacy of the dynamical variables. In particular,
the conjugate variable of $n_1$ is nothing but $n_5$ itself, since
$p_1=\delta {\cal L}/\delta \partial_t n_1 = 2 n_5$. Standard quantization procedure
requires the canonical commutation relation $[n_1,p_1]=i$,
which in this case just reproduces equation (\ref{non-commute}). This confirms
the fact that the $SO(5)$ projection does not change the form of the interaction
potential, only the commutation relation between $n_1$ and $n_5$.
It is easy to see that the low energy effective Lagrangian
(\ref{projected_L}) produces exactly the same long wave length
collective mode spectrum as
the projected $SO(5)$ Hamiltonian (\ref{Hamiltonian}) defined
on a lattice. To facilitate the comparison, we take the
$SO(5)$ potential to be
\begin{equation}
V(n) = -\frac{\delta}{2} \sum_a n_a^2 + \frac{W}{4} n_a^4 \ \ , \ \ \delta>0
\end{equation}
Assuming broken symmetry in the $n_1$ and $n_2$ directions, we find that
the $n_3$ and $n_4$ modes always decouple, and they have a linear
spin wave dispersion relation with $v_s=\sqrt{\rho/\chi_s}$.
The Euler-Lagrangian equations of motion gives the following
dispersion relation for the $n_1$ and $n_2$ modes:
\begin{equation}
\omega^2 = \left\{ \begin{array} {l}
\frac{1}{4}\rho^2k^4 \\
\frac{\rho}{\chi_s}k^2 +\frac{2\delta}{\chi_s}
\end{array} \right.
\end{equation}
for the AF state with $\langle n_1\rangle =0, \langle n_2\rangle \neq 0$,
\begin{equation}
\omega^2 = \left\{ \begin{array} {l}
\frac{1}{2}\rho\delta k^2 \\
\frac{\rho}{\chi_s} k^2
\end{array} \right.
\end{equation}
for the SC state with $\langle n_1\rangle \neq 0, \langle n_2\rangle =0$, and
\begin{equation}
\omega^2 = \left\{ \begin{array} {l}
\frac{1}{4}(1+\sin^2\alpha/\cos^2\alpha)\rho^2k^4 \\
\frac{2\delta \cos^2\alpha}{\chi_s} +(\frac{1}{2}\delta \rho \sin^2\alpha+ \frac{\rho}{\chi_s}) k^2
\end{array} \right.
\end{equation}
for the mixed state with $\langle n_1^2\rangle = \frac{\delta}{W} \sin^2\alpha$ and
$\langle n_2^2\rangle = \frac{\delta}{W} \cos^2\alpha$.
These dispersion relations agree exactly with the lattice model results at the
projected $SO(5)$ symmetric point if we make the following identification
\begin{equation}
\rho = 2J \ \ , \ \
\chi_s = \frac{1}{2J} \ \ , \ \
\delta = 2(4J -\Delta)
\end{equation}
The effective Lagrangian can be easily used to discuss effects of $SO(5)$ symmetry
breaking. The simplest form of symmetry breaking is increasing the chemical potential
beyond the critical value $\mu_c$, so that a pure SC state is realized. The
chemical potential enters the effective Lagrangian through the gauge coupling
in the time direction via the following substitution
\begin{equation}
\partial_t n_1 \Rightarrow \partial_t n_1 + \delta \mu n_5 \ \ , \ \
\partial_t n_5 \Rightarrow \partial_t n_5 - \delta \mu n_5 \ \ , \ \
\end{equation}
where $\delta\mu\equiv \mu-\mu_c$ is the deviation of the chemical potential
away from the critical value. In this case, the spin triplet excitations
acquire a finite mass gap, with the following dispersion relation:
\begin{equation}
\omega^2(k) = \frac{\rho k^2}{\chi_s} + \frac{4(\mu-\mu_c)}{\chi_s}
\end{equation}
and the mass gap increases with increasing doping in the SC state.
We summarize the behavior of the collective modes obtained in the
previous two sections in Fig. (\ref{fig5}).
\begin{figure*}[h]
\centerline{\epsfysize=6.0cm
\epsfbox{fig5.eps}
}
\caption{Evolution of the collective mode spectra as a function of density
in the projected $SO(5)$ model. Fig. 5a shows the gap towards spin excitations.
Charge excitations are gapless for the entire region of density, however,
the dispersion relation changes from $\omega\sim k^2$ to $\omega\sim k$
at $\rho_c$, as indicated in Fig. 5b.}
\label{fig5}
\end{figure*}
We see that while there are significant modifications of the collective
mode spectra in the density regime $0<\rho<\rho_c$ from the unprojected
$SO(5)$ symmetry, the spectra beyond $\rho_c$ is essentially identical to
the behavior expected from the unprojected $SO(5)$ symmetry. This should
be expected from our general considerations about the Gutzwiller projection
without much detailed calculations. We argued that the only effect of the
Gutzwiller projection is to change the quantum commutation relation
between the SC components of the superspin $n_1$ and $n_5$. However, for
$\rho>\rho_c$, the system is in a pure SC phase where these components
acquire classical expectation values. In this case, the modification of
the quantum commutation relation does not have any significant effect.
This argument can also be illustrated by a simple picture of a {\it chiral}
$SO(5)$ sphere, as depicted in Fig. (\ref{fig6}).
\begin{figure*}[h]
\centerline{\epsfysize=8.0cm
\epsfbox{fig6.eps}
}
\caption{Pictorial representation of a chiral $SO(5)$ sphere. }
\label{fig6}
\end{figure*}
In this picture, the north and
south poles represent the three AF directions, and the equatorial plane represent
the SC directions. The sphere is perfectly $SO(5)$ symmetric. However, the
chemical potential along the pole direction acts like a fictitious magnetic
field which restricts the sense of the rotation in the SC plane.
Small oscillations
of a vector pointing close to the north pole enclose the fictitious magnetic
flux, and can only execute chiral rotations. This amounts to the projection of
the particle-pair states at half-filling. On the other hand, small oscillations
of a vector pointing anywhere along the equator does not enclose the fictitious
magnetic flux, and their dynamics is therefore unaffected by the projection.
Dynamics of a vector pointing anywhere between the north pole and the equator
is also partially affected by the projection, but the symmetry of the static
potential bears a unique signature.
\section{Conclusion}
\label{sec:VII}
The main purpose of this paper is to introduce the concept of
projected $SO(5)$ models and discuss the properties of this
model in connection with high $T_c$ superconductivity.
The projected $SO(5)$ model describes the low energy and
long distance bosonic degrees of the freedom near the
AF/SC transition. We showed that the Gutzwiller projection
can be implemented analytically on every site in the
$SO(5)$ theory. In the presence of a infinite Mott-Hubbard
gap, we show that static properties of the model can remain
$SO(5)$ symmetric, while the modification of the dynamics
can be completely cast into a non-trivial commutation
relation between the two SC components of the $SO(5)$
superspin. Unlike the unprojected $SO(5)$ models which
can only have the full dynamic $SO(5)$ symmetry at half-filling,
the projected $SO(5)$ model can have static $SO(5)$ symmetry
at a critical value of the chemical potential $\mu_c$ and for a finite
range of doping $0\le\rho\le\rho_c$. At $\rho=0$, the system has
a AF ground state and zero compressibility. In the intermediate regime
$0<\rho<\rho_c$, the system has mixed
AF and SC order and infinite compressibility. For $\rho>\rho_c$,
the system has a pure SC ground state, the SC order parameter
rises to a maximal value before it decreases with doping.
At the projected $SO(5)$ symmetric point, we can understand
precisely the evolution of the collective modes. On the
AF side, we have two gapless spin wave modes and a gapless
charge mode describing the gapless fluctuation from AF to SC.
In the intermediate density regime $0<\rho<\rho_c$, the physical
properties of the spin waves remain unchanged, while the massive
spin amplitude mode gradually decreases its energy and merges
with the two spin wave modes at $\rho=\rho_c$. The charge mode
in the intermediate density regime is gapless, but has quadratic
dispersion relation, which is a unique signature of the projected
$SO(5)$ symmetry. For $\rho>\rho_c$, the charge mode is gapless
with linear dispersion relation, and the $\pi$ triplet spin
mode becomes massive, and gradually increases its energy with
increasing doping. In this regime, the behavior of the collective
modes are identical to the unprojected $SO(5)$ model.
This very simple model can form the basis to understand many
novel and puzzling properties of the high $T_c$ superconductors
in a unified framework. It points out a route from AF to SC through
a gradual rotation of the superspin angle. At the projected $SO(5)$
symmetry point the mean field energy is independent of the superspin angle, and
therefore it offers an explanation
of the absence of the chemical potential shift in the underdoped
regime without global phase separation. It predicts a phase diagram
which is qualitatively consistent with the observed phase diagram
in the high $T_c$ materials. In the underdoped regime of the phase
diagram, the system have large AF and SC fluctuations, and these
fluctuations can be responsible for the pseudogap physics observed
in these materials.
There are many possible directions to carry out this line of research
in the future. The most important issue is to understand the precise
nature of the intermediate state in the regime $0<\rho<\rho_c$.
Since the system has infinite compressibility in this regime,
different small perturbation may select different ground states.
Such perturbing effects might include quantum fluctuations and
longer ranged interactions. In particular, we would like to
investigate the possibility that these perturbations might lead
to the formation of incommensurate order or stripes.
In this work, we have discussed extensively the collective
fluctuations in the long wave length limit. Due to the definitions
of our effective lattice model, the $k\rightarrow 0$ limit
corresponds to the $k\rightarrow 0$ limit in the SC correlation
functions and the $k\rightarrow (\pi,\pi)$ limit of the AF
spin correlation functions. Within the $SO(5)$ theory, the
$\pi$ resonance in the SC state is viewed as the $SO(5)$ symmetry
partner of the $k\rightarrow 0$ Goldstone mode of the SC
phase fluctuation. While the commensurate neutron resonance
mode is observed in both YBCO and BISCO superconductors,
all high $T_c$ systems also have incommensurate spin
fluctuations. How can these features be explained within the
current theoretical model?
The fact that LSCO and YBCO have very different Fermi surface
shapes and yet have similar incommensurate magnetic peaks
strongly suggests that the incommensurate peaks are not
sensitive to Fermi surface effects and should be explainable
within an effective bosonic model.
Let us recall that the collective mode of a superfluid boson
system consists of linearly dispersing phonon branch and
another roton branch with a minimum located at the inverse
inter-particle spacing. So far, we have only studied the
phonon branch of the charged bosons. By analogy, the roton
branch should also exist, with a wave vector determined by
the density of the charged bosons or doping. Within the
$SO(5)$ theory, while the commensurate neutron resonance
can be viewed as the $SO(5)$ partner of the SC phase mode,
the incommensurate magnetic peaks can be viewed as the
$SO(5)$ partner of the roton minimum of the charged bosons.
A detailed quantitative analysis of this picture will be carried
out in the future.
However, while the ground state in the doping range $0<\rho<\rho_c$
may depend sensitively on small perturbation effects,
at finite temperature, these perturbation effects
should be small and the system should display more universal
properties. We have shown that the projected $SO(5)$ symmetry
should be valid for the entire doping range $0<\rho<\rho_c$,
and we shall quantitatively study the manifestation of this
symmetry at finite temperature, and see if the projected
$SO(5)$ symmetry can give a universal explanation
of the pseudogap physics.
{\it Note Added:} After completing this work, we received a very interesting
paper by Coen van Duin\cite{coen}, in which he also observed the
``remnant $SO(5)$ behavior in the large $U$ limit".
\section*{Acknowledgments}
We would like to acknowledge useful discussions with D. Arovas,
J. Berlinsky, E. Demler, R. Eder, C. Kallin,
S. Kivelson and D. Scalapino.
SCZ and JPH are supported by the NSF under grant numbers DMR-9814289.
WH and EA are supported by FORSUPRA II, BMBF (05 605 WWA 6)
and by the Deutsche Forschungsgemeinschaft (AR 324/1-1)
and (HA 1537/17-1).
AA is supported by the Israel Science Foundation. AA,
EA and WH would like to acknowledge the support and hospitality of the
Stanford Physics department, where most of this work was carried out.
|
\section{Introduction}
A great deal of effort has been devoted in recent times to the dynamics of
scalar fields in the radiation and matter dominated era. The motivations are
manyfold: first, several theories of fundamental physics predict the
existence of scalar fields \cite{fri}\cite{fer}\cite{wet}; second, a slowly
rolling scalar field may mimic the behavior of a cosmological constant at
the present time, in agreement with a popular model of structure formation
and with the observation of an accelerated space expansion \cite{fri}\cite
{zla}\cite{cal}; third, the scalar field may alleviate the constraints on a
true cosmological constant \cite{cob}; fourth, the additional source of
fluctuations produced by the scalar field may give new observable effects on
the cosmic microwave background and on the structure formation \cite{fer}
\cite{perr}\cite{via}.
So far, most work focused on fields with minimal coupling to gravity \cite
{fri}\cite{fer}\cite{zla} \cite{cal}\cite{lid} \cite{cop}. In this case, the
dynamics in a homogeneous and isotropic space-time is completely determined
once one specifies the matter fluid equation of state and the field
potential. For the former, the obvious choices of interest are the equation
of state of a relativistic fluid and of a pressureless one. For the latter,
although there are no observations or fundamental principles to guide our
investigation, potentials like power-laws, exponential and a handful of
other cases have been selected, basing either on simplicity or on some
particle physics model. Among the infinite solutions of the system of
equations, the attractor solutions are of course of the greatest interest.
Among the attractors, those which have a power-law behavior, denoted also as
scaling solutions, are particularly simple to find and to study.
Consequently, the study of the scalar field dynamics has focussed on the
search of {\it scaling attractors}. To be interesting for cosmological
purposes, these attractor solutions must also lead to a energy density in
the scalar field which is a non-negligible fraction of the total energy
density. Finally, if we want to explain recent observations of the
large-scale geometry of the space-time \cite{per}\cite{rei}, the scalar
factor has to be accelerated at the present.
In minimal coupling theories the Lagrangian is the sum of the
Einstein-Hilbert gravity Lagrangian and of the scalar field sector. The
non-minimal coupling (NMC) adds a new term which, in its simplest form, may
be written as (for a more general form that includes derivatives see Ref.
\cite{ame})
\begin{equation}
f(\phi )R.
\end{equation}
For instance, Refs. \cite{acc}\cite{ame90}\cite{fut}\cite{kas} adopted $
f\sim \phi ^2$ discussing the model in the context of inflation. In Ref.
\cite{uza} several attractor scaling solutions in the matter dominated
regime with power-law and exponential potential have been found. Other forms
of $f(\phi )$ have been considered (see e.g. \cite{ste}).
A common feature of all these investigations, perhaps obviously, is the
choice of specific potentials and coupling functions. The purpose of this
paper is to show that it is possible to find attractor solutions in NMC
models in which both the coupling $f(\phi )$ and the potential $V(\phi )$
are left unspecified, and only their relation matters. In other words, we
will find a class of models in which the dynamics of the system is
independent of the coupling and of the potential, and depends only on their
relation. In particular, we will find attractor solutions for all models for
which we can write \cite{bel}
\begin{equation}
V(\phi )=Af(\phi )^M. \label{condp}
\end{equation}
This relation holds, for instance, when both $V$ and $f$ are power-law, or
exponential, but is also valid for much more complicated functions, like
products of power law and exponential. In the limit of strong coupling, the
dynamics of the cosmological solution will be shown to depend essentially
only on $M$ and on the fluid matter equation of state.
After performing a conformal rescaling of the metric, the NMC system is
written as a scalar field in pure General Relativity with an exponential
potential and an extra coupling to the ordinary matter (see e.g. \cite{wet95}
, in which however only the case $f\sim \phi ^2$ has been discussed). This
system shows a surprisingly rich phase space structure, with four different
attractors. Two of these are qualitatively similar to the attractors found
in the system without extra coupling. The other two however are new, and
have not been previously identified. Although we derived this system from a
class of NMC theories, we remark that it is interesting on its own, and many
cosmological properties of its trajectories have yet to be worked out. Here
we study it mainly to constrain the NMC model, and find that the constraint
on the variability of the gravitational constant rules out this class of
models as explanation for the accelerated expansion rate of the Universe.
In the next section we work out the field equations. In Section 3 we find
and discuss the attractor solutions, in Section 4 we discuss their
cosmological properties, and in the final Section we draw the conclusion and
point to new developments.
\section{Field equations}
Consider the Lagrangian of a NMC scalar field plus a perfect fluid matter
component ($\kappa ^2\equiv 8\pi M_p^{-2}$)
\begin{eqnarray}
L_{tot} &=&L(\phi ,R)+2\kappa ^2L_\phi +2\kappa ^2L_{matter}, \\
L(\phi ,R) &=&-f(\phi )R, \\
L_\phi &=&\frac 12\phi _{,\mu }\phi ^{,\mu }-V(\phi ).
\end{eqnarray}
Contrary to the usual notation, we found convenient to include into $f(\phi
) $ the constant that produces the Einstein-Hilbert term, so that our $f$ is
$1+\kappa ^2\xi f(\phi )$ in the notation of, e.g., Ref. \cite{acc}\cite
{ame90}\cite{fut}\cite{kas}). We will always assume $f>0$, since it acts as
an effective gravitational constant,
\begin{equation}
G_{eff}=(2\kappa ^2f)^{-1}.
\end{equation}
The Einstein equations are
\begin{equation}
G_{\mu \nu }=L_{,R}^{-1}\left[ \frac 12g_{\mu \nu }(L-L_{,R}R)-g_{\mu \nu
}\Box L_{,R}+(L_{,R})_{;\mu \nu }+\kappa ^2T_{\mu \nu \left( \phi \right)
}+\kappa ^2T_{\mu \nu \left( m\right) }\right] ,
\end{equation}
where $L_{,R}$ denotes here $dL/dR$, and where the scalar field
energy-momentum tensor is
\begin{equation}
T_{\mu \nu \left( \phi \right) }=\phi _{,\mu }\phi _{,\nu }-\frac 12g_{\mu
\upsilon }\phi _{,\alpha }\phi ^{,\alpha }+g_{\mu \upsilon }V(\phi ),
\end{equation}
and the fluid tensor is
\begin{equation}
T_{\mu \nu \left( m\right) }=(\rho +p)u_\mu u_\nu -g_{\mu \nu }p.
\end{equation}
Now, under the conformal transformation
\begin{equation}
\widetilde{g}_{\mu \upsilon }=e^{2\omega }g_{\mu \upsilon },
\end{equation}
the following transformations (see e .g. \cite{bar}\cite{sch}\cite{mae})
occur: the kinetic term
\begin{equation}
K_{\mu \nu \left( \phi \right) }=\phi _{,\mu }\phi _{,\nu }-\frac 12g_{\mu
\upsilon }\phi _{,\alpha }\phi ^{,\alpha },
\end{equation}
remains invaried ($K_{\mu \nu }=\widetilde{K}_{\mu \nu }$); the potential
term $g_{\mu \upsilon }V(\phi )$ becomes $e^{-2\omega }\widetilde{g}_{\mu
\upsilon }V(\phi )$; and the perfect fluid tensor becomes
\begin{equation}
T_{\mu \nu (m)}=e^{-2\omega }\widetilde{T}_{\mu \nu (m)}.
\end{equation}
Putting
\begin{equation}
2\omega =\log f,
\end{equation}
it follows that the equations in the rescaled metric (sometimes called
Einstein frame, while the old metric is the Jordan frame) are
\begin{equation}
\widetilde{G}_{\mu \nu }=\kappa ^2\left[ F^2(\phi )\widetilde{K}_{\mu \nu
\left( \phi \right) }+\widetilde{g}_{\mu \upsilon }e^{-4\omega }V(\phi
)+e^{-4\omega }\widetilde{T}_{\mu \nu \left( m\right) }\right] ,
\end{equation}
where
\begin{equation}
F^2(\phi )=\frac 1f+\left( \frac{f^{^{\prime }}}{cf}\right) ^2, \label{fphi}
\end{equation}
where $c^2=2\kappa ^2/3$ and where the prime denotes derivation with respect
to $\phi $. We can then define a new canonical field
\begin{equation}
\psi \equiv \int d\phi F(\phi ),
\end{equation}
a new potential
\begin{equation}
U(\psi )\equiv \frac{V(\phi )}{f(\phi )^2},
\end{equation}
and a new matter tensor
\begin{equation}
\widetilde{T}_{\mu \nu \left( m\right) }^{*}\equiv e^{-4\omega }\widetilde{T}
_{\mu \nu \left( m\right) }.
\end{equation}
Finally, all these definitions lead to the canonical equations in the new
metric $\widetilde{g}_{\mu \nu }$
\begin{equation}
\widetilde{G}_{\mu \nu }=\kappa ^2\left[ \widetilde{T}_{\mu \nu \left( \psi
\right) }+\widetilde{T}_{\mu \nu \left( m\right) }^{*}\right] .
\end{equation}
The new matter energy-momentum tensor can be written as
\begin{equation}
\widetilde{T}_{\mu \left( m\right) }^{*\nu }=diag(\rho e^{-4\omega
},-pe^{-4\omega },-pe^{-4\omega },-pe^{-4\omega })=diag(\rho
^{*},-p^{*},-p^{*},-p^{*}).
\end{equation}
As a last step, we rewrite the new metric in the Friedmannian form
\begin{equation}
\widetilde{g}_{\mu \nu }=diag(1,-\widetilde{a}^2-\widetilde{a}^2,-\widetilde{
a}^2),
\end{equation}
where the old time and the old scale factor are
\begin{equation}
t=\int e^{-\omega (\tilde{t})}d\widetilde{t}, \label{timerescal}
\end{equation}
and
\begin{equation}
a=e^{-\omega (\widetilde{t})}\widetilde{a}. \label{arescal}
\end{equation}
The equation of motion for the fields are obtained as the covariant
conservation laws of the energy tensors. In the old frame they read
\begin{eqnarray}
\Box \phi +V^{\prime }+f^{\prime }R/2\kappa ^2 &=&0, \nonumber \\
T_{\mu \nu (\phi )}^{;\mu } &=&0.
\end{eqnarray}
The transformation to the new frame is performed according to the rules
\begin{eqnarray}
R &=&e^{2\omega }(\widetilde{R}-6\widetilde{g}^{\alpha \beta }\omega
_{,\alpha }\omega _{,\beta }+6\widetilde{\Box }\omega ), \nonumber \\
\Box &=&e^{2\omega }(\widetilde{\Box }-2\widetilde{g}^{\alpha \beta }\omega
_{,\alpha }\nabla _\beta ).
\end{eqnarray}
From now on, we omit all the tilde, until we return to the original
quantities. Finally, the full set of equations in the Friedmann metric read:
\begin{eqnarray}
\ddot{\psi}+3H\dot{\psi}+U_{,\psi } &=&\frac 12W_{,\psi }(\rho ^{*}-3p^{*}),
\label{sys1} \\
\dot{\rho}^{*}+3H\left( \rho ^{*}+p^{*}\right) &=&-\frac 12W_{,\psi }\dot{
\psi}(\rho ^{*}-3p^{*}), \\
3H^2 &=&\kappa ^2\left( \rho ^{*}+\frac 12\dot{\psi}^2+U\right) , \\
-2\dot{H} &=&\kappa ^2\left( \rho ^{*}+p^{*}+\dot{\psi}^2\right) ,
\label{sys4}
\end{eqnarray}
where
\begin{equation}
W=\log f(\phi ),\quad W_{,\psi }=\frac{f^{\prime }}{fF}.
\end{equation}
As already remarked in the Introduction, the system (\ref{sys1}-\ref{sys4}),
here derived from a NMC model, is interesting on its own. Indeed, we can
regard either the Jordan or the Einstein frame as the physical one. In the
former case, we have to express the solutions of the above system back in
the original frame, and study its cosmological consequences in the original
frame, as we will do below. In the latter case, the solutions of the system
are the physical solutions, and their properties can be directly compared to
observations. In particular, the constraints from the variability of $G$ ,
which we will find to limit heavily the cosmological viability of our
solutions, apply only assuming the physical frame to be the original Jordan
one.
\section{Solutions}
The full dynamics of the system (\ref{sys1}-\ref{sys4}) is specified by the
potential $U$ and by the equation of state $p=(w-1)\rho $. In the following
we consider only $0\leq w\leq 2$. To write down the potential $U(\psi )$, we
have first to find the relation between $\psi $ and $\phi $. This is where
the possibility of a dynamics independent of the potential and of the
coupling function arises. In fact, if we assume that \cite{bel}
\begin{equation}
f^{\prime 2}\gg c^2f, \label{condf}
\end{equation}
then we can simplify Eq. (\ref{fphi}):
\begin{equation}
F^2(\phi )=\left( \frac{f^{^{\prime }}}{cf}\right) ^2. \label{fdef}
\end{equation}
It follows
\begin{equation}
c\psi =\int \frac{df}f=\log f, \label{psif}
\end{equation}
where the integration constant can be absorbed into a redefinition of $\psi $
. It follows that the conformal function $\omega $ equals $c\psi /2$.
Therefore, once we have the dynamics of $\psi $ in the transformed metric,
we can write down the solution in terms of the original metric without
having to specify $f(\phi ),$ provided we express also the potential $V(\phi
)$ as a function of $f(\phi )$. With the assumption (\ref{condf}) we get $
W_{,\psi }=c$ in the system (\ref{sys1}-\ref{sys4}), so that putting
\[
\beta =4-3w,
\]
the first two equations become
\begin{eqnarray}
\ddot{\psi}+3H\dot{\psi}+U_{,\psi } &=&\frac 12c\beta \rho ^{*}, \nonumber
\label{wet} \\
\dot{\rho}^{*}+3Hw\rho ^{*} &=&-\frac 12c\beta \dot{\psi}\rho ^{*}.
\label{wet}
\end{eqnarray}
The condition (\ref{condf}) holds true in several cases. For instance, it is
verified for large $\phi $ by any function $f(\phi )$ which grows faster
than quadratically, that is $\lim_{\phi \rightarrow \infty }f(\phi )/\phi
^2\rightarrow \infty $. In the often-studied quadratic case, $f=1+\kappa
^2\xi \phi ^2$, for large $\phi $ we can put $f(\phi )=\kappa ^2\xi \phi ^2$
. Then, instead of Eq. (\ref{condf}), one has $f^{\prime 2}=4\kappa ^2\xi f$
, and all that changes is that in Eq. (\ref{fdef}) and Eq. (\ref{psif}) $c^2
$ is replaced by $c^2/(1+1/6\xi )$. In this case, all the results found
below become {\it exact}. The weak coupling limit in which $\kappa ^2\xi
\phi ^2\ll 1$, i.e. $\xi \ll (\kappa ^2\phi ^2)^{-1}$, on the other hand, is
excluded in the present analysis. We could then label our case as the {\it
strong coupling} limit. In fact, it is easily seen that it corresponds to
the limit in which the Lagrangian can be approximated as $-f(\phi )R-2\kappa
^2V(\phi )$, neglecting the kinetic term $\frac 12\phi _{,\mu }\phi ^{,\mu }$
. Notice however that this does not imply that the scalar field kinetic
terms in the field equations are negligible, because the non-minimal
coupling itself introduces other kinetic terms.
Now, as anticipated, suppose we can write $V(\phi )=Af(\phi )^M$ .The
potential becomes then
\begin{equation}
U(\psi )=\frac{Af(\phi )^M}{f(\phi )^2}=Ae^{\sqrt{2/3}\mu \kappa \psi },
\end{equation}
where
\begin{equation}
\quad \mu \equiv M-2.
\end{equation}
Therefore, the potential can be written as an exponential, whatever the
shape of $V$ and of $f$, provided that the condition (\ref{condf}) and the
relation (\ref{condp}) are fulfilled. The sign of $\mu $ selects the
direction in which the field $\psi $, and thus the variable $f$, rolls. If $
\mu >0$, $\psi $ rolls toward $-\infty $, so that $f\rightarrow 0$, and the
effective gravitational constant $G_{eff}$ increases with time. In the
opposite case, $\mu <0$, we have that $G_{eff}$ decreases in the future. We
emphasize that if $f(\phi )$ is quadratic, then all results below remain
valid provided $\mu $ is replaced by $\mu _q=\mu /(1+1/6\xi )^{1/2}$ and $
\beta $ by $\beta _q=\beta /(1+1/6\xi )^{1/2}.$
The scalar field dynamics in NMC theories is then reduced to the scalar
field dynamics in pure general relativity with an exponential potential and
with a scalar field/matter coupling. In the radiation case in which $w=4/3$,
we have $\beta =0$, and the source terms decouple. The decoupling occurs
also when we can neglect the matter energy density $\Omega _{\rho
^{*}}=\kappa ^2\rho ^{*}/3H^2$ with respect to the scalar field energy
density $\Omega _\psi $. In these cases, the problem is identical to that
already solved in, e.g., Ref. \cite{fer}\cite{lid}\cite{cop}\cite{rat}. The
case $\beta \neq 0$ has been already discussed by Wetterich in \cite{wet95},
where some of its attractors have been identified. Here we extend the
analysis to the full classification of critical points and attractors
(finding two new attractors) and express the solutions in terms of the old
frame. We keep $\beta $ as an independent parameter as long as possible, and
proceed to replacing it by $4-3w$ only in the graphics, in order to narrow
the parameter space to two dimensions, namely $w$ and $\mu $. The formulas
apply however to the more general case, unless otherwise specified.
Following Copeland et al. \cite{cop} we define
\begin{equation}
x=\frac{\kappa \dot{\psi}}{\sqrt{6}H},\quad y=\frac{\kappa \sqrt{U}}{\sqrt{3}
H},
\end{equation}
and introduce the independent variable $\alpha =\log a(t)$. Notice that $x^2$
and $y^2$ give the fraction of total energy density carried by the scalar
field kinetic and potential energy, respectively . Then, we can rewrite the
system (\ref{sys1}-\ref{sys4}) as
\begin{eqnarray}
x^{\prime } &=&-3x+3x\left[ x^2+\frac 12w(1-x^2-y^2)\right] -\mu y^2+\frac 12
\beta (1-x^2-y^2), \nonumber \\
y^{\prime } &=&\mu xy+3y\left[ x^2+\frac 12w(1-x^2-y^2)\right] .
\label{syst2}
\end{eqnarray}
where the prime is here $d/d\alpha $. The system is invariant under the
change of sign of $y$ and of $\alpha $. Since it is also limited by the
condition $\rho ^{*}>0$ to the circle $x^2+y^2\leq 1$, we may study only the
unitary semicircle of positive $y$. The critical points, those that verify $
x^{\prime }=y^{\prime }=0$, are scaling solutions, on which the scalar field
equation of state is
\begin{equation}
w_\psi =\frac{2x^2}{x^2+y^2}=const,
\end{equation}
the scalar field total energy density is $\Omega _\psi =x^2+y^2$, and the
scale factor is
\begin{equation}
a\sim t^p,\quad p=\frac 2{3w}\left[ \frac w{w+\Omega _\psi (w_\psi -w)}
\right]
\end{equation}
(the slope $p$ is not to be confused with the pressure).
Copeland et al. \cite{cop} have shown that the system (\ref{syst2}) with $
\beta =0$ and an exponential potential has up to five critical points, that
can be classified according to the dominant energy density: one dominated by
the scalar field total energy density (let us label this point as solution $
a $ and refer to its coordinates as $x_a,y_a$ ), one in which the fractions
of energy density in the matter and in the field are both non-zero (labelled
$b$), one dominated by the matter field ($c$), and finally two dominated by
the kinetic energy of the scalar field, of which one at $x=-1$ ($d$) and one
at $x=+1$ ($e$).
The critical points on which the matter field becomes negligible reduce to
the $\beta =0$ case: therefore, the solutions $a,$ $d$, and $e$ remain the
same also for $\beta \neq 0.$ The points $b$ and $c$ are instead modified.
The solution $c$ is no longer matter dominated: rather, the scalar field
kinetic energy and the matter energy take up a constant fraction of the
total energy. In MDE, the scalar field kinetic energy amounts to $\Omega
_\psi =1/9.$ The critical points in the general case $\beta \neq 0$ are
listed in Tab. I, where we put $g(\beta ,w,\mu )\equiv \beta ^2+2\beta \mu
+18w.$
\[
\begin{tabular}{|c|c|c|c|c|c|}
\hline
& $x$ & $y$ & $\Omega _\psi $ & $p$ & $w_\psi $ \\ \hline
$a$ & $-\mu /3$ & $\left( 1-x_a^2\right) ^{1/2}$ & $1$ & $3/\mu ^2$ & $2\mu
^2/9$ \\ \hline
$b$ & $-\frac{3w}{2\mu +\beta }$ & $-x_b\left( \frac g{9w^2}-1\right) ^{1/2}$
& $\frac g{\left( \beta +2\mu \right) ^2}$ & $\frac 2{3w}\left( 1+\frac \beta
{2\mu }\right) $ & $\frac{18w^2}g$ \\ \hline
$c$ & $\frac \beta {6-3w}$ & $0$ & $\left( \frac \beta {6-3w}\right) ^2$ & $
\frac{6(2-w)}{\beta ^2+9(2-w)w}$ & $2$ \\ \hline
$d$ & $-1$ & $0$ & $1$ & $1/3$ & $2$ \\ \hline
$e$ & $+1$ & $0$ & $1$ & $1/3$ & 2 \\ \hline
\multicolumn{6}{|c|}{Tab. I} \\ \hline
\end{tabular}
\]
Although the number and position of the critical points is affected only
quantitatively by the extra coupling, their stability properties are
modified in a more radical way. In particular, while for $\beta =0$ only the
points $a$ and $b$ can be attractors, here we show that also $c$ and $d$ may
be stable. Only the point $e$ remains always unstable.
The stability analysis is performed as usual by linearization around the
critical points. The parametric regions in which the real part of both
eigenvalues of the linearization matrix is negative are regions of
stability. To simplify the discussion, we only consider the crucial property
of stability versus instability, paying no attention to the topography of
the critical point (whether it is a knot, spiral, or saddle). In the
following, we say that an attractor exists if it lies in the region $0\leq
x^2+y^2\leq 1$. The parameter spaces are plotted in Fig. 1.
{\bf Point }$a$ .
The solution $a$ exists for $\mu ^2<9$, and is an attractor only for $\mu
_{-}<\mu <\mu _{+}$ where
\begin{equation}
\mu _{\pm }=\frac 14\left( -\beta \pm \sqrt{\beta ^2+72w}\right) ,
\end{equation}
(e.g., $\mu _{-}=-2.39$ and $\mu _{+}=1.89$ for $w=1$). On this attractor we
have $w_\psi =2\mu ^2/9$ and
\begin{equation}
p_a=3/\mu ^2,
\end{equation}
inflationary if $|\mu |<\sqrt{3}$.
{\bf Point }$b$ .
The attractor $b$ exists and is stable in the region delimited by $\mu <\mu
_{-}$ and $\mu >\mu _{+}$ and the two branches of the curve
\begin{equation}
\mu _0=-\frac 1{2\beta }\left( \beta ^2+18w-9w^2\right) .
\end{equation}
The scale factor slope on the attractor is
\begin{equation}
p_b=\frac 2{3w}\left( 1+\frac \beta {2\mu }\right) .
\end{equation}
and, for $\beta =4-3w$, is inflationary within the two branches of the curve
\begin{equation}
\mu _i=\frac{4-3w}{3w-2}.
\end{equation}
It is remarkable that the inflationary region for the point $b$ includes
values smaller than $w\approx 0.91$, and therefore {\it excludes} the MDE
equation of state $w=1$. This conclusion is not changed by replacing $\mu $
and $\beta $ with their counterparts $\mu _q$ and $\beta _q$ in the case of
a quadratic coupling $f(\phi )$.
{\bf Point }$c$ .
This point exists for $w<5/3$, and is stable below the lower branch and
above the upper branch of $\mu _0$ $.$ The slope is
\[
p_c=\frac{6(2-w)}{\beta ^2+9(2-w)w},
\]
and it is never accelerated if $\beta =4-3w$. The point $c$ shares with $b$
the property that matter and scalar field have both a non-vanishing fraction
of the energy density.
{\bf Point }$d$ .
This point exists for all values of the parameters, and if $\beta =4-3w$ is
stable for $w>5/3$ and $\mu >3$. Its slope is always $p_d=1/3$.
{\bf Point }$e.$
This point exists and is unstable for all values of the parameters if $\beta
=4-3w$.
The complex structure of the parameter space is summarized in Fig. 2. Notice
that 1), for each value of the parameters $w,\mu $ there is one and only one
attractor; 2) for $w=1$ the points $a,b$ or $c$ can be stable, depending on $
\mu $; 3) these solutions are inflationary in the shaded region; 4) only the
point $a$ can be accelerated for $w=1$ or larger. In Fig. 3 we present four
phase spaces displaying in turn the four possible attractors. The parameters
correspond to the points marked with stars in Fig. 2. As already remarked,
attractors $c$ and $d$ have not been previously noticed. Also, it is
important to remark that the attractors are not only locally stable, but
extend their basin of attraction to all of the phase space. That is, {\it any
} possible initial condition lead to the attractor.
\section{Back to the Jordan frame}
Here we leave the dynamical analysis of the system in the rescaled frame and
get back to the original one. What the attractors look like in the Jordan
frame?
Reintroducing the tildes, we have along the attractors $a$ and $b$ (from now
on, quantities without tildes are expressed in the original metric)
\begin{equation}
c\psi =-\frac 2\mu \log |\widetilde{t}/\tau _{a,b}|, \label{tab}
\end{equation}
(for $\mu \neq 0$) where
\begin{equation}
\tau _{a,b}^{-1}=\sqrt{2A}\mu c\frac{x_{a,b}}{y_{a,b}}.
\end{equation}
On the attractors $c$ and $d$ , for which $y=0$, $\psi \rightarrow \infty $,
and the conformal transformation cannot be performed. Since $\psi $ is
proportional to $\log f$, the attractors $c$ and $d$ lead to an effective
gravitational constant that is either zero or infinite and are therefore to
be rejected as possible solutions in the Jordan frame. Of course,
trajectories that have not already reached the attractor cannot be excluded,
but these are not scaling solutions, and will not be further considered in
this paper.
Form Eq. (\ref{tab}) it follows (neglecting the subscripts)
\begin{equation}
e^{2\omega }=(\widetilde{t}/\tau )^{-2/\mu }.
\end{equation}
From the latter expression we can evaluate the relation between the old and
new time and scale factor, given by Eq. (\ref{timerescal}) and (\ref{arescal}
). We obtain (for $\mu \neq 0,-1$)
\begin{eqnarray}
\widetilde{t} &\sim &t^{\frac \mu {1+\mu }}, \nonumber \\
\widetilde{a} &\sim &t^{-1/(1+\mu )}a(t).
\end{eqnarray}
As can be seen, for $\mu \rightarrow \pm \infty $ the old and new metric
coincide; in this limit the scalar field vanishes on the attractor, and the
system reduces to the pure perfect fluid Friedmann case.
It follows that in the original variables the scale factor is again a power
law
\begin{equation}
a\sim t^{p^{\prime }},\quad p^{\prime }=\frac{1+\mu p}{1+\mu }.
\end{equation}
On the attractor $a$, $p_a=3/\mu ^2,$ which is inflationary (both in the
original and in the rescaled frame) if $\mu ^2<3$, that is $2-\sqrt{3}<M<2+
\sqrt{3}$ , we have
\begin{equation}
p_a^{\prime }=\frac{\mu +3}{\mu (1+\mu )}.
\end{equation}
Consider now some special cases. If $0>\mu >-1$ the scale factor follows a
{\it pole-like} inflation, $a\sim (t_0-t)^{p_a^{\prime }}$ with negative
exponent. For $\mu =0$ ,(i.e. $V\sim f^2$) the old and new metric coincide
(up to a constant), the field freezes to a constant and its energy drives a
deSitter expansion. The system reduces asymptotically to pure general
relativity with a cosmological constant. Finally, for $\mu =-1$ (i.e. $V\sim
f$), the scale factor is power law accelerated in the new frame, but maps
again to a deSitter expansion in the original frame. If $f$ is quadratic in $
\phi $, then the inflationary condition on the solution $a$ reads
\begin{equation}
\mu ^2<3(1+1/6\xi ). \label{acc}
\end{equation}
On the attractor $b$ , on the other hand, putting $p_b=2/(3w^{\prime })$
with $w^{\prime }=w\left( 1+\beta /2\mu \right) ^{-1}$ we obtain
\begin{equation}
p_b^{\prime }=\frac{3w^{\prime }+2\mu }{3w^{\prime }(1+\mu )}.
\end{equation}
Notice that $p_b^{\prime }\rightarrow 2/3w$ for $\mu \rightarrow \pm \infty $
, as expected. Since the property of being accelerated is conformally
invariant (for positive definite conformal factors), going back to the old
frame do not change qualitatively the attractors found so far. Also, it is
not difficult to check that $\Omega _{\rho ^{*}}=\kappa ^2\rho ^{*}/3
\widetilde{H}^2$ and $\Omega _\psi $ are invariant under conformal
transformation, so that $\Omega _\psi =\Omega _\phi $ .
It can be shown that the choice $V\sim f^M$ is the only one that allows
scaling attractors in both the old and the new metric. Other choices are
possible that allow scaling solutions either in the old or in the new
metric: for instance, $V\sim f^2\left( \log f\right) ^M$ gives scaling
attractors in the new metric but not in the old one.
\section{Cosmological properties}
Once we have the analytical expression of the attractors, we must consider
whether they are viable as cosmological solutions. The attractor solution $a$
is inflationary (accelerated) and the scalar field is asymptotically the
dominating component. As such, it may match the observations of an
accelerated expansion; for instance, the value $w_\psi \approx 0.4$
suggested in Ref. \cite{tur} implies
\[
M\approx 0.7\text{ or }3.3.
\]
On the other hand, since $\Omega _\phi \rightarrow 1$, in order to allow for
a substantial fraction in the ordinary matter component at the present, the
attractor has not to be already reached.
The solution $b$ has some drawbacks. First, is not accelerated at all for $
w=1$; second, the constraints from nucleosynthesis do not allow a large
fraction of energy density in the scalar field, so that it cannot provide
closure energy. However, as argued in \cite{fer}, models which reach this
attractor compare favorably with observations of large scale structure, and
may have a simple interpretation in terms of fundamental physics.
Both solutions $a$ and $b$ are heavily constrained by the upper limits on
the variability of the gravitational constant. We have
\begin{equation}
|\dot{G}/G|=|\dot{f}/f|=\frac 2{|1+\mu |}\frac 1t.
\end{equation}
Comparing with the observational constraint $|\dot{G}/G|<a10^{-10}yr^{-1}$,
and assuming $t\approx 10$ Gyr, we obtain the condition
\begin{equation}
\mu >\frac 2a-1. \label{cn}
\end{equation}
Current constraints (see e.g. \cite{gue} ) give $a\approx 0.1$ or smaller.
This implies $\mu >20$, too large for the attractor $a$ to exist. A similar
problem arises if $f$ is quadratic. Along the attractor $b$, the energy
density in the scalar field is a constant fraction of the total energy. In
MDE (and for $\beta =4-3w=1$) this is
\begin{equation}
\Omega _\phi =\frac{19+2\mu }{\left( 1+2\mu \right) ^2}.
\end{equation}
The constraint (\ref{cn}) gives
\begin{equation}
\Omega _\phi \leq 0.035,
\end{equation}
which confines the scalar field contribution to that of a minor component.
This constraint is three or four times stronger than that imposed by the
nucleosynthesis \cite{fer} on a minimally coupled field.
\section{Conclusions}
In this paper we have investigated a large class of NMC models in the limit
of strong coupling with a perfect fluid matter component, searching for
attractors that might provide a decaying cosmological constant. These models
include all the cases in which the potential $V(\phi )$ is a power of the
coupling function $f(\phi )$, regardless of their functional form. We have
shown that
\begin{enumerate}
\item The NMC system can be reduced to a scalar field with an exponential
potential, a minimal coupling to gravity, and an extra coupling to the
matter.
\item For each pair of the parameters $w,\mu $ there is one out of four
possible scaling attractors: one, $a$, scalar field dominated and possibly
accelerated; one, $b$, decelerated if $w\geq 0.91$ and with constant ratio
of scalar field total energy to matter; one, $c$, always decelerated and
with constant ratio of scalar field kinetic energy to matter; and finally
one, $d$, also always decelerated, and dominated by the field kinetic energy.
\item Attractors $c$ and $d$ are acceptable only in the rescaled frame; in
the original frame they lead to a gravitational constant either vanishing or
infinite.
\item This choice $V\sim f^{M\text{ }}$ is the only choice (in the strong
coupling regime) for which there is a scaling attractor both in the original
and in the rescaled metric.
\item The constraint on the time variability of $G$ rules out the
accelerated models, and only allows a very small fraction of the energy
density to be in the NMC scalar field.
\end{enumerate}
Clearly, this analysis is not yet conclusive. Viable solutions might exist
for which one or more of the following is true: $a$) the attractors are not
yet reached; $b$) $V$ does not equal $f^M$; $c$) the strong coupling regime
does not apply. For instance, assuming $f=1+\kappa ^2\xi \phi ^2$, and in
the limit of weak coupling , the $|\dot{G}/G|$ bound can be satisfied for
small $\xi $, and the solutions are cosmologically acceptable, although by
construction do not add much to the minimally coupled model.
\section{Acknowledgments}
I am indebted to Carlo Baccigalupi, Francesca Perrotta and Jean-Philippe
Uzan for useful discussions on the topic.
|
\section{FIGURE CAPTIONS}
\begin{figure}
\plotone{gmokinfig1.ps}
\caption[gmokinfig1.ps]{Plot of rest-frame $(U-V)_{AB}$ vs. $z$
for the CFRS sample, computed as in Lilly et al.\ (1995). Galaxies
which were observed in our kinematic study are shown as large symbols,
while the rest of the CFRS galaxies are shown as small crosses.
Objects for which we failed to detect an emission line due to
instrumental reasons have been omitted. Note that the sample shown
above includes the three mergers and three face-on galaxies which were
excluded from the analysis.}
\end{figure}
\newpage
\begin{figure}
\plotone{gmokinfig2.ps}
\caption[gmokinfig2.ps]{\footnotesize Two projections of the ($\sigma_{v}$,
$r_{1/2}$,\ $M_{B}$) space for a sample of local galaxies (top 8
panels, small symbols), for CFRS galaxies at $z<0.4$ (top 8 panels,
large symbols with error bars), and for CFRS galaxies at $z>0.45$
(bottom two panels). Gray symbols represent $S/N<20$ objects;
galaxies with rest-frame $(U-V)_{AB}\leq1.14$ are shown as
circles, and those with rest-frame $(U-V)_{AB}\geq1.14$ are shown
as squares; points with arrows represent 1-$\sigma$ upper-limits. The
dotted lines are fiducials (constant mass lines on the left-hand
panels, and log$\sigma = 0.134 M_{B} + C$ on the right-hand
panels). The dashed box in the upper eight panels indicates the
location of the $z>0.45$ CFRS galaxies as defined from the bottom two
panels.}
\end{figure}
\end{document}
|
\section{Introduction}
In recent years there have been a number of papers on gauge invariances
in systems with second class constraints. Basically this involves
unravelling, using the language of constaints, gauge symmetries hidden
in such systems. By doing so it has sometimes been possible to obtain
a deeper and more illuminating view of these systems.
In unravelling such hidden symmetries, the basic idea is that the
original system is a gauge fixed version of a certain gauge theory;
the latter reverts to the former under certain gauge fixing
conditions. The advantage in having a gauge theory lies in the fact
that other gauges can also be considered. Further it is possible to get
more than one gauge theory for the same second class constrained system.
Two methods have been suggested to make this conversion of second class
theories into gauge theories. One method, based originally on the idea
of Faddeev and Shatashvili \cite{FaSha}, is now called the Batalin -
Fradkin method \cite{BaFa} and is formulated in an enlarged phase space.
The other method, based on the idea of Mitra-Rajaraman \cite{RaPa} is
what we call {\em gauge unfixing} \cite{RaVy}, and does not use any
enlarged phase space. Rather it confines itself to the phase space of
the original second class system.
Even though these methods look very different in their formulations,
when they are applied to many systems like Chern - Simons theory,
chiral Schwinger model, etc., the results are basically the same,
implying that as far as these examples are concerned, the two methods
are essentially equivalent. In what follows we illustrate this equivalence
for three such examples. We compare the results of the two methods for
these systems. Any conclusions arising out of such a comparison might be
illuminating when the formal equivalence is considered. Such a formal
equivalence will be considered separately.
In Section 2 we review the two methods, and look at specific
systems in Section 3. We conclude in Section 4.
\section{The Formalisms}
We consider a finite dimensional system in phase space with
co-ordinates $ q^i $ and conjugate momenta $ p_i ~(i = 1,2,
\ldots N). $ The system has two second class constraints,
\begin{equation}
Q_1(q,p) \approx 0, \hspace{1.2in} Q_2(q,p) \approx 0
\end{equation}
defining a constraint surface $ \sum_2$. Due to their second
class nature, the $2 \times 2$ antisymmetric matrix $E$ whose
elements $E_{mn}$ are Poisson brackets among the $ Q$'s,
\begin{equation}
E_{ab}(q,p) = \{ Q_a, Q_b \}\hspace{0.8in}a,b=1,2
\end{equation}
is {\em invertible} everywhere, even on the surface $ \sum_2. $
The canonical Hamiltonian is $ H_c $ and the total Hamiltonian is
\begin{equation}
H = H_c + \mu_1Q_1 + \mu_2Q_2,
\end{equation}
where the multipliers $ \mu_1, \mu_2 $ are determined on the
surface $ \sum_2 $ by demanding consistency of the two constraints
with respect to $H$.
\subsection{Batalin - Fradkin (BF) method}
As mentioned earlier, the idea behind this method \cite{FaSha, BaFa}
is to enlarge the phase space by including new variables. Since we have
taken the number of second class constraints here to be two, we
introduce two variables $\Phi^a (a=1,2)$. The enlarged phase space
$(q,p,\Phi)$ has the basic Poisson brackets
\begin{equation}
\{ q^i, p_j \} = \delta^i_j \hspace{1.2in} \{\Phi^a, \Phi^b\} =
\omega^{ab}
\end{equation}
with all other Poisson brackets zero. The antisymmetric $ 2\times
2 $ matrix $ \omega^{ab} $ is a constant matrix, unspecified for
the present.
The first class constraints are obtained as functions in this {\em
extended} phase space. Since we had initially two second class
constraints, there will now be {\em two} first class constraints,
given in general by
\begin{eqnarray}
{\widetilde Q_a}(q^i,p_i,\Phi^a) & = & Q_a + \sum_{m = 1}^{
\infty} Q_a^{(m)}, \hspace{0.8in} Q_a^{(m)} \sim (\Phi^a)^{m} \\
{\widetilde Q_a}(q^i,p_i,0) & = & Q_a^{(0)} = Q_a \nonumber
\end{eqnarray}
where the second line gives the boundary condition. The terms of
various orders in the expansion for $ {\widetilde Q_a} $ are
obtained by demanding that the $ {\widetilde Q_a} $ are strongly
first class,
\begin{equation}
\{ {\widetilde Q_a}, {\widetilde Q_b} \} = 0.
\end{equation}
For instance for the first order this requirement gives
\begin{equation}
E_{ab} + X_{ac}\omega^{cd}X_{db} = 0
\end{equation}
which can be satisfied, using (4), if we write and substitute
\begin{equation}
Q_a^{(1)} = X_{ab}(q,p)\Phi^b,
\end{equation}
in (6) and consider terms upto first order. For many systems,
such as the ones we consider in the next section, the higher
order terms are all zero. It must be noted that there is an
inherent {\em arbitrariness} in the choice of the $ \Phi^a $
and the coefficients $ X_{ab}. $ This choice may be exploited
to advantage.
To get gauge invariant observables, we note that relevant
quantities of the original second class system in general are
not gauge invariant with respect to the new first class constraints.
They are made gauge invariant by modifying them in the extended
phase space. In particular the gauge invariant Hamiltonian \cite{BaFa}
can be written in general as
\begin{equation}
{\widetilde H} = H + \sum_{m=1}^{\infty} H^{(m)} \hspace{1in}
H^{(m)} \sim (\Phi^a)^{m}
\end{equation}
and the terms of various orders are obtained by demanding that
\begin{equation}
\{ {\tilde H}, {\widetilde Q_m} \} = 0.
\end{equation}
A similar procedure is followed to obtain other gauge invariant quantities
also \cite{BaFa}. We finally remark that eqn. (7) can always be written so
in the case of 2 constraints. For more than 2 second class constraints,
this has to be taken as an assumption, which {\em need not} hold in the
very general case. In a sense, the $ X_{ab} $ can be called the ``square
root'' of the matrix $E$ \cite{BaFa}.
\subsection{The Gauge Unfixing method}
This method \cite{RaPa,RaVy}, in contrast to the BF method, makes no
enlargement of the phase space. Rather, since the number of second
class constraints is even (we consider here only bosonic constraints),
only half of these constraints are chosen to form a first class subset,
and the other half as the corresponding gauge fixing subset. This latter
subset is discarded, retaining only the first class subset, and so we
have a gauge theory.
In a general system, getting a first class subset is a non-trivial
issue \cite{RaVy}; this might be possible only under certain conditions.
However in the case of only two second class constraints (which we
consider here) the first class constraint can always be chosen.
For instance, we can choose $ Q_1 $ as our first class constraint,
and $ Q_2 $ as its gauge fixing constraint. We redefine, using (2),
\begin{equation}
Q_1 \rightarrow \chi = E_{12}^{-1}Q_1 \hspace{1.5in} Q_2
= \psi
\end{equation}
and no longer consider the $ \psi $ as a constraint. The gauge invariant
Hamiltonian and other physical quantities are obtained by defining
the projection operator
\begin{equation}
I\!\!P \cong ~: e^{-\psi{\hat \chi}} : \hspace{1.2in} {\hat \chi}(A) \equiv
\{ \chi, A \}
\end{equation}
and operating $I\!\!P$ on any phase space function $A$. There is
an ordering prescribed for the action of $ I\!\!P; $ the $ \psi
$ is always outside the Poisson bracket in the expansion of the
exponential. The action of $ I\!\!P $ on relevant quantities gives
the gauge invariant quantities.
It must be noted that even in this method, there is an inherent {\em
arbitrariness}; either of the two second class constraints can be chosen
to be first class. The two choices define two different projection
operators, and the gauge theories so constructed will in general be
different. This arbitrariness can be exploited to advantage.
\section{Examples}
\subsection{Chiral Schwinger Model}
This well known anomalous gauge theory \cite{JaRa,Kim,Asv} involves
chiral fermions coupled to a $U(1) $ gauge field in $1+1$ dimensions.
Classically the theory has gauge invariance, but this is lost upon
quantisation. We look at its bosonised version, the advantage being
that the corresponding classical theory itself has no gauge invariance.
We have
\begin{equation}
{\cal L} = -\frac{1}{4}F_{\mu\nu}F^{\mu\nu} + \frac{1}{2}(\partial_{\mu}
\phi)^2 + e(g^{\mu\nu} - \epsilon^{\mu\nu})(\partial_{\mu}\phi)A_{\nu}
+ \frac{1}{2}e^2\alpha A_{\mu}^2
\end{equation}
where $g^{\mu\nu} $ = diag$(1,-1)$, ~$ \epsilon^{01} = - \epsilon^{10} =
1 $ and $ \alpha $ is the regularisation parameter. The theory is gauge
non-invariant for all values of $ \alpha. $ We consider the case $ \alpha
> 1.$
The canonical Hamiltonian density is
\begin{eqnarray}
{\cal H}_c & = & \frac{1}{2}\pi_1^2 + \frac{1}{2}\pi_{\phi}^2 + \frac{1}{2}
(\partial_{1}\phi)^2 + e(\partial_1\phi + \pi_{\phi})A_1 + \frac{1}{2}
e^2(\alpha + 1)A^2_{1} \nonumber \\
&& \rule{0mm}{8mm}\hspace{0.6in} - A_0\left [- \partial_1\pi_1 +
\frac{1}{2}e^2(\alpha - 1)A_0 + e(\partial_1\phi + \pi_{\phi}) + e^2A_1
\right ]
\end{eqnarray}
where $ ~\pi_1 = F^{01} = \partial^0A^1 - \partial^1A^0, $ and $ \pi_{\phi}
= \partial_0\phi + e(A_0 - A_1) $ are the momenta conjugate to $ A_1 $ and
$\phi$ respectively. The constraints are
\begin{eqnarray}
Q_1 & = & \pi_0 = 0\nonumber \\
Q_2 & = & - \partial_1\pi_1 + e^2(\alpha - 1)A_0 + e(\partial_1\phi +
\pi_{\phi}) + e^2A_1 = 0
\end{eqnarray}
defining a constraint surface $ \sum_2. $ These are of the second class,
\begin{equation}
E_{12} = \{ Q_1(x), Q_2(y) \} = -e^2(\alpha - 1)\delta(x - y).
\end{equation}
\vspace{2mm}
Following the BF method \cite{Kim}, the phase space is extended by
introducing two fields $ \Phi_1, \Phi_2 $, with Poisson bracket relations
of the form (4). To get the first class constraints (5) and (8), we recall
that there is a natural arbitrariness in choosing the matrices $
\omega^{ab} $ and $ X_{ab}. $ This allows the choice
\begin{eqnarray}
\omega = \left( \begin{array}{c}
0 ~~~ 1\\
-1 ~~ 0
\end{array} \right) \delta(x-y) \hspace{0.6in}
X (x,y) = e\sqrt{\alpha-1}\left( \begin{array}{c}
1 ~~ 0\\
0 ~~ 1
\end{array} \right) \delta(x-y)
\end{eqnarray}
\vspace{1mm}
\noindent This choice allows the two new fields to form a canonically
conjugate pair. The terms beyond the first order in the expansion (5)
are all zero. We then get
\begin{equation}
{\widetilde Q_m} = Q_m + e\sqrt{\alpha-1}\Phi_m, \hspace{0.8in}m = 1,2
\end{equation}
which, using (16) and (17), can be verified to be strongly first class.
Using similar arguments the gauge invariant Hamiltonian for the choice
(17) is
\begin{eqnarray}
{\widetilde H}_{BF}
& = & H_c + {\displaystyle \int}dx\left[ - \frac{\displaystyle e\pi_1 +
e(\alpha-1)\partial_1A_1}{\displaystyle \sqrt{\alpha-1}} \Phi_1 +
\frac{\displaystyle e^2}{\displaystyle
2(\alpha-1)}\Phi_1^2 \right.\nonumber \\
&& \hspace{1.6in}\left.+ \frac{1}{2}(\partial_1\Phi_1)^2 + \frac{1}{2}(
\Phi_2)^2 - \frac{{\widetilde Q_2}\Phi_2}{\displaystyle e\sqrt{\alpha-1}}
\right]
\end{eqnarray}
where $ H_c $ is the canonical Hamiltonian. This $ {\widetilde H_{BF}}
$ has zero PBs with the first class constraints (18).
\vspace{4mm}
Coming to the {\em Gauge Unfixing} (GU) method \cite{Asv}, we reiterate
that no new field need be introduced. The first class constraint is taken
to be just one of the two existing constraints. We choose, after a
rescaling
\begin{equation}
\chi = \frac{1}{\displaystyle e^2(\alpha-1)} Q_2 ~,
\end{equation}
so that the relevant constraint surface is $ \sum_1 $ defined by $ \chi
\cong 0.$ The gauge fixing-like constraint is $ \psi = 0,$ and is discarded
(that is {\em unfixed}). To get the gauge invariant Hamiltonian we
construct a projection operator $ I\!\!P $ of the form (12) and use it on
the canonical Hamiltonian $H_c$,
\begin{equation}
{\widetilde H}_{GU} = H_c + {\displaystyle \int}
dx\left[\frac{\displaystyle \pi_1 + (\alpha-1)\partial_1A_1}{\displaystyle
{\alpha-1}} Q_1 + \frac{1}{\displaystyle 2e^2(\alpha-1)}(\partial_1Q_1)^2
+ \frac{1}{\displaystyle 2(\alpha-1)^2}Q_1^2, \right]
\end{equation}
which is gauge invariant with respect to the $\chi$.
We see that, apart from the term $ {\displaystyle \int} dx \left
(\frac{\displaystyle \Phi_2^2}{2} - \frac{\displaystyle \Phi_2}{
\displaystyle 2\sqrt{\alpha - 1}}{\widetilde Q_2} \right), $ the $
{\widetilde H_{GF}}$ and the ${\widetilde H_{GU}}$ in (19) and (21) are
the same, if we make the identification $ \Phi_1 = - \frac{\displaystyle
Q_1}{\displaystyle e\sqrt{\alpha - 1}} $. We however emphasise the two
rather different paths used to get these Hamiltonians. One requires the
introduction of an extra (canonical) pair of fields, while the other
doesn't need this. In both cases extra terms are needed to make the
original Hamiltonian gauge invariant. For the $ {\widetilde H_{BF}} $
these terms had to be written down by introducing extra fields, whereas
in the $ {\widetilde H_{GU}} $ these terms involve a variable
{\em already present} in the original theory.
\vspace{4mm}
We look at the path integral quantisation for these two Hamiltonians. For
Hamiltonian $ {\widetilde H_{BF}} $ we have,
\begin{eqnarray}
{\cal Z}_{BF} & = & {\displaystyle \int}{\cal D}(\pi_{\mu}, A^{\mu},
\pi_{\phi}, \phi, \Phi_1, \Phi_2, \mu_1, \mu_2) ~e^{iS_{BF}}\\
S_{BF} & = & {\displaystyle
\int} dxdt \left [\pi_0{\dot A^0} + \pi_1{\dot A^1} + \pi_{\phi}{\dot
\phi} + \Phi_2{\dot \Phi_1} - {\widetilde{\cal H}}_{BF} - \mu_1{\widetilde
Q_1} - \mu_2{\widetilde Q_2} \right].\nonumber
\end{eqnarray}
Here $ \mu_1, \mu_2 $ are undetermined Lagrange multipliers
corresponding to the first class constraints $ {\widetilde Q_1}, {
\widetilde Q_2} $ respectively. If we make the transformations $ \mu_2
\rightarrow \mu_2^{\prime} = \mu_2 - \frac{\displaystyle \Phi_2}{
\displaystyle e(\alpha - 1)}, $
$ A_0 \rightarrow A_0^{\prime} = A_0 - \mu_2^{\prime}, ~\pi_1
\rightarrow \pi_1^{\prime} = \pi_1 + \partial_0A_1 - \partial_1A_0^{
\prime}, ~\pi_{\phi} \rightarrow \pi_{\phi}^{\prime} = \pi_{\phi} - {\dot
\phi} - e(A_0^{\prime} - A_1), $ and then integrate over $ \pi_i^{\prime},
{}~\pi_{\phi}^{\prime}, ~\Phi_2, $ we get
\begin{eqnarray}
{\cal Z}_{BF} & = & {\displaystyle \int} {\cal D}(A^{\mu}, \phi, \Phi_1)
{}~~e^{iS_{BF}} \nonumber \\
S_{BF} & = & {\displaystyle \int} dxdt \left( - \frac{1}{4}F_{\mu\nu}
F^{\mu\nu} + \frac{e^2\alpha}{2}A_{\mu}A^{\mu} + e(\eta^{\mu\nu} -
\epsilon^{\mu\nu})(\partial_{\mu}\phi)A_{\nu} +
\frac{1}{2}(\partial_{\mu}\phi)^2 \right.\\
&& \hspace{0.8in} \left. + \frac{1}{2}(\partial_{\mu}\Phi_1)^2 -
\frac{e}{\sqrt{\alpha-1}}\Phi_1[(\alpha-1)\eta^{\mu\nu}
+ \epsilon^{\mu\nu}](\partial_{\mu}A_{\nu})\right).
\nonumber
\end{eqnarray}
The action $ S_{BF} $ above is just the gauge invariant version for the
chiral Schwinger model. As is well known, this action was first obtained
by merely adding the (Schwinger) terms in the variable $ \Phi_1 $ to
the original bosonised action (13). It has also been obtained using
other arguments. In the Batalin-Fradkin approach, these Schwinger terms
and $ \Phi_1 $ come up due to the extension of the phase space.
\vspace{4mm}
Coming to the Hamiltonian $ {\widetilde H_{GU}} $ of the Gauge Unfixing
method, we have
\begin{eqnarray}
{\cal Z}_{GU} & = & {\displaystyle \int} {\cal D}(A^{\mu}, \pi_{\mu},
\phi, \pi_{\phi}, \lambda) ~e^{iS_{GU}}\nonumber\\
S_{GU} & = & {\displaystyle \int} dxdt \left [\pi_0{\dot A^0} +
\pi_1{\dot A^1} + \pi_{\phi}{\dot \phi} - {\cal {\widetilde H}}_{GU} -
\lambda\chi \right],
\end{eqnarray}
where $ \lambda $ is the arbtirary Lagrange multiplier. We make the
transformations $ A_0 \rightarrow A_0^{\prime} = A_0 - \frac{\lambda}{
e^2(\alpha-1)}, ~\pi_1 \rightarrow \pi_1^{\prime} = \pi_1 + \partial_0
A_1 - \partial_1A_0^{\prime} + \frac{\pi_0}{\alpha-1}, \pi_{\phi}
\rightarrow \pi_{\phi}^{\prime} = \pi_{\phi} - {\dot \phi} + eA_1 -
eA_0^{\prime}$ and $ \lambda \rightarrow \lambda^{\prime} = \lambda
+ \partial_0\pi_0. $ We then integrate over $ \pi_1^{\prime}, \pi_{\phi
}^{\prime}$ and $ \lambda^{\prime} $ in the path integral to get
\begin{eqnarray}
{\cal Z}_{GU} & = & {\displaystyle \int} {\cal D}(A^{\mu}, \phi, \pi_0)
{}~e^{iS_{GU}}\nonumber\\
S_{GU} & = & {\displaystyle \int} dxdt \left( - \frac{1}{4}F_{\mu\nu}
F^{\mu\nu} + \frac{e^2\alpha}{2}A_{\mu}A^{\mu} + e(\eta^{\mu\nu} -
\epsilon^{\mu\nu})(\partial_{\mu}\phi)A_{\nu} + \frac{1}{2}(\partial_{
\mu}\phi)^2 \right.\\
&& \hspace{1.2in} \left. + \frac{1}{2}\frac{\displaystyle (\partial_{\mu}
\pi_0)^2}{\displaystyle e^2(\alpha-1)} + \frac{\displaystyle
\pi_0}{\displaystyle \alpha-1}[(\alpha-1)\eta^{\mu\nu} +
\epsilon^{\mu\nu}](\partial_{\mu}A_{\nu})\right).\nonumber
\end{eqnarray}
We see that on making the replacement $ \pi_0 = - e\sqrt{\alpha-1}\Phi_1
$ in (25), we get the same result as in the Batalin-Fradkin case (23).
This is achieved here by introducing no extra fields. The extra field
of the BF method is found here {\em within} the original phase space.
Further the Schwinger terms are the same in both cases. We also note
that the extra term $ \int \left(\frac{(\Phi_2)^2}{2} + \ldots \right)$
{} which comes upon comparing the Hamiltonians in (19) and (21) have been
integrated away in the path integral (23).
\subsection{The abelian Proca model}
This $ (3+1) $ - dimensional theory is given by the Lagrangian \cite{Stu,
NbRb, AsvP}
\begin{equation}
{\cal L} = -\frac{1}{4}F_{\mu\nu}F^{\mu\nu} + \frac{m^2}{2}A^{\mu}A_{\mu}
\end{equation}
where $m$ is the mass of the $ A_{\mu} $ field, $ g_{\mu\nu} = $diag$ ~
(+,-,-,-)$
and $ F_{\mu\nu} = \partial_{\mu}A_{\nu} - \partial_{\nu}A_{\mu}. $ The
canonical Hamiltonian is given by
\begin{equation}
H_c = {\displaystyle \int d^3x ~{\cal H}_c} = {\displaystyle \int
d^3x \left ( \frac{1}{2}{\pi_i}{\pi_i}
+ {\frac{1}{4}}F_{ij}F_{ij} - \frac{m^2}{2}(A_0^2 - A_i^2) +
A_0({\partial_i}\pi_i)\right ),}
\end{equation}
with $ \pi_i = -F_{0\,i} $ the momenta conjugate to the $A^i$. The second
class constraints are
\begin{equation}
Q_1 = \pi_0(x) \approx 0,\hspace{1.2in}
Q_2 = (-{\partial_i}\pi_i + {m^2}A_0)(x) \approx 0,
\end{equation}
which together define the surface $ \sum_2 $ in the phase space. Their
second class nature is due to the mass term in the Lagrangian.
The matrix $E$ of eqn. (2) is here
\begin{equation}
E = \left( \begin{array}{c}
0\;\;\;{-m^2}\\
\!\!{m^2}\:\;\;\;0
\end{array}\right ) \delta(x-y),
\end{equation}
whose determinant is non-zero everywhere.
\vspace{3mm}
Using the Batalin - Fradkin method \cite{NbRb}, we introduce an extra
canonical pair of fields $ \theta $ and $ \pi_{\theta}, $ with $ \{
\theta(x), \pi_{\theta}(y) \} = \delta(x-y). $ As earlier, transformations
in this extended phase space modify the constraints (28) to
\begin{equation}
{\widetilde Q_1} = Q_1 + m^2 \theta \hspace{1.2in} {\widetilde Q_2} =
Q_2 + \pi_{\theta},
\end{equation}
which, using (29) can be seen to be strongly first class. The form for
these constraints corresponds to the choice of $ \theta $ and $
\pi_{\theta} $ as a canonically conjugate pair.
The corresponding gauge invariant Hamiltonian is
\begin{equation}
{\widetilde H_{BF}} = H_c + {\displaystyle \int} d^3x \left(\frac{
\displaystyle \pi_{\theta}^2}{\displaystyle 2m^2} + \frac{m^2}{2}
(\partial_i\theta)^2 - m^2\theta\partial_iA_i\right),
\end{equation}
with respect to which the first class constraints
satisfy
\begin{equation}
\{ {\widetilde Q_1}, {\widetilde H_{BF}} \} = {\widetilde Q_2}\hspace{1in}
\{ {\widetilde Q_2}, {\widetilde H_{BF}} \} = 0
\end{equation}
\vspace{4mm}
On the other hand, in the Gauge Unfixing method, there is only one first
class constraint, one of the two in (28). For our purposes we choose this
constraint to be
\begin{equation}
\chi = \frac{1}{m^2}(-\partial_i\pi_i + m^2A_0) \approx 0,
\end{equation}
and throw away the other $ \psi = \pi_0. $ The relevant constraint surface
is defined by $ \chi \approx 0. $ The $ H_c $ of (27) does not have zero
PB with this $ \chi $ on this new surface, and hence is not gauge
invariant. Using a projection operator of the form (12) on $H_c$ we get
the gauge invariant Hamiltonian
\begin{equation}
{\widetilde H_{GU}} = H_c + \int d^3x \left[\psi \partial_iA_i -
\frac{1}{2m^2} \psi\partial^2_i\psi\right].
\end{equation}
Note the similarity between the Hamiltonians $ {\widetilde H_{BF}} $ and
$ {\widetilde H_{GU}}. $ Indeed, apart from the term $ \int d^3x~ \frac{
\displaystyle \pi_{\theta}^2}{2m^2} $ in (31) the two Hamiltonians are
the same if we make the identification $ \psi = -m^2\theta. $
\vspace{4mm}
We look at path integral quantisations. For the $ {\widetilde H_{BF}} $,
we have
\begin{eqnarray}
{\cal Z}_{BF} & = & {\displaystyle \int} {\cal D}(\pi_o, A_0, \pi_i,
A^i, \pi_{\theta}, \theta, \mu_1, \mu_2) ~e^{iS_{BF}} \\
{S_{BF}} & = & {\displaystyle \int} d^4x \left[\pi_0{\dot A}_0 +
\pi_i{\dot A}^i + \pi_{\theta}{\dot \theta} - {\cal H}_c - \frac{
\displaystyle \pi_{\theta}^2}{2m^2} - \frac{m^2}{2}
(\partial_i\theta)^2 + m^2\theta\partial_iA_i - \mu_1{\widetilde Q_1}
- \mu_2{\widetilde Q_2}\right],\nonumber
\end{eqnarray}
where $ \mu_1 $ and $ \mu_2 $ are undetermined Lagrange multipliers.
After some redefinitions of fields and integration over momenta and the
$ \mu$'s, we get
\begin{equation}
{\cal Z}_{BF} = {\displaystyle \int} {\cal D}(A^{\mu}, \theta) ~
exp ~{i{\displaystyle \int} d^4x \left[ - \frac{1}{4}
F^2_{\mu\nu} + \frac{m^2}{2} A_{\mu}^2 + \frac{m^2}{2}\partial_{\mu}
\theta\partial^{\mu}\theta - m^2\theta\partial_{\mu}A^{\mu} \right ].}
\end{equation}
The last line gives just the St\"uckelberg gauge invariant action
\cite{Stu}. The $ \theta $ field is called the St\"uckelberg scalar,
which was originally introduced by St\"uckelberg directly into the
Proca Lagrangian to make it gauge invariant. Thus in the BF formalism,
the extra field introduced is the St\"uckelberg scalar.
\vspace{4mm}
On the other hand, the path integral for the gauge unfixed Hamiltonian
$ {\widetilde H_{GU}} $ is
\begin{equation}
{\cal Z}_{GU} = {\displaystyle \int} {\cal D}(A^{\mu}, \pi_{\mu},
\lambda)~exp ~{i\displaystyle \int} d^4x ~\left(\pi_0{\dot A_0} +
\pi_i{\dot A_i} - {\widetilde H_{GU}} - \lambda \chi\right),
\end{equation}
with no extra fields. After redefinition of $ A_0, \pi_i $ and the $
\lambda $ and integrating over $ \lambda^{\prime} $ and $ \pi_i^{\prime},
$ we get
\begin{equation}
{\cal Z}_{GU} = {\displaystyle \int}{\cal D}(A^{\mu}, \theta) ~
exp ~i{\displaystyle \int} d^4x ~\left[ - \frac{1}{4}F_{\mu\nu}^2
+ \frac{m^2}{2}A_{\mu}^{\mu} + \frac{m^2}{2}(\partial_{\mu}\theta)^2 -
m^2\theta\partial_{\mu}A^{\mu} \right],
\end{equation}
where $ \theta = - \frac{\displaystyle \pi_0}{\displaystyle m^2}. $ This
is just the path integral (37) obtained earlier using the Batalin-Fradkin
method.
No extra fields were introduced. Rather the extra co - ordinate field of
the BF method which was recognised earlier as the St\"uckelberg scalar
corresponds here to $ - \frac{\displaystyle \pi_0}{m^2}, $ which was
already present in the phase space of the original second class
constrained theory. This suggests that the extra field need not be
introduced at all.
\subsection{Abelian Chern-Simons Theory}
This $ 2+1 $ dimensional theory \cite{RaVy, RaBa} consists of a
complex field interacting with an abelian Chern-Simons field. The
theory is described by the Lagrangian density
\begin{equation}
{\cal L} = ({\cal D}_{\mu}\phi)^*({\cal D}^{\mu}\phi) +
\frac{\alpha}{4\pi}
\epsilon_{\mu\nu\lambda}A^{\mu}\partial^{\nu}A^{\lambda},
\end{equation}
with $ {\cal D}_{\mu}\phi = (\partial_{\mu} - iA_{\mu})\phi, ~g_{\mu\nu}$
= diag $(1,-1,-1). $ and $ \mu,\nu,\lambda = 0,1,2$. We also have
\begin{equation}
H_c = \int d^2x \left [({\vec \nabla} + i{\vec A})\phi^* \cdot ({\vec
\nabla} - i{\vec A})\phi + \pi_{\phi^*}\pi_{\phi} + A_0(j_0 -
\frac{\alpha}{2\pi}\epsilon_{ij}\partial_iA_j)\right ]
\end{equation}
with $ \pi_{\phi} = {\dot \phi}^* + iA_0\phi^*, ~\pi_{\phi^*} =
{\dot \phi} - iA_0\phi,$ and $ j_0 = i(\phi\pi_{\phi} -
\phi^*\pi_{\phi^*})$.
\vspace{1mm}
\noindent The constraints are
\begin{eqnarray}
\pi_0 \approx 0 &\hspace{0.1in}& Q_3 = (j_0 - \frac{\alpha}{2\pi}
\epsilon_{ij}\partial_iA_j) - \frac{\alpha}{2\pi}\partial_1Q_1 +
\partial_2Q_2 \approx 0\nonumber\\
Q_1 = -\frac{2\pi}{\alpha}(\pi_1 + \frac{\alpha}{4\pi}A_2) \approx 0
&\hspace{0.3in}& Q_2 = (\pi_2 - \frac{\alpha}{4\pi}A_1) \approx 0
\end{eqnarray}
with the first line showing the first class constraints. The
second line gives the second class constraints of the theory,
\begin{equation}
\{ Q_1(x), Q_2(y) \} = \delta(x-y).
\end{equation}
Instead of the canonical Hamiltonian (40), we will consider the total
Hamiltonian which guarantees the time consistency of $Q_1$ and $Q_2$ (on
the surface defined by both $Q_1$ and $Q_2$),
\begin{eqnarray}
H = {\displaystyle \int}d^2x \left \{{\cal H}_c + u_1Q_1 + u_2Q_2
\right\} & \hspace{0.21in} & u_1 = i\phi^*{\cal D}_2\phi -
i\phi({\cal D}_2\phi)^* + \frac{\alpha}{2\pi}\partial_1A_0
\nonumber\\
&& u_2 = \frac{2\pi}{\alpha} \left [i\phi^*{\cal D}_1\phi - i\phi({\cal
D}_1\phi)^* - \frac{\alpha}{2\pi}\partial_2A_0 \right ].
\end{eqnarray}
\vspace{3mm}
We now get the gauge theory using the Batalin-Fradkin method \cite{RaBa}.
The new variables $ \Phi^1, ~\Phi^2 $ serve to enlarge the phase space,
and have the Poisson brackets (4). In this enlarged phase space, we have
the strongly first class constraints (after appropriate choice of the $
\omega $ and the $ X $ matrices),
\begin{equation}
{\widetilde Q}_1 = Q_1 + \Phi^1 \hspace{1.4in} {\widetilde Q}_2 = Q_2
- \Phi^2.
\end{equation}
The corresponding Batalin-Fradkin gauge invariant Hamiltonian is
\begin{equation}
{\widetilde H}_{BF} = H + {\displaystyle \int} d^2x \left
\{\phi\phi^*(\Phi^1)^2 +
\left (\frac{2\pi}{\alpha}\right )^2 \phi\phi^*(\Phi^2)^2 - 2\phi\phi^*
\left [\Phi^1{\widetilde Q}_1 - \Phi^2{\widetilde Q}_2 \right]\right \}
\end{equation}
\vspace{2mm}
We apply the {\sf gauge unfixing} method \cite{RaVy} to this theory. We
redefine
\begin{equation}
\chi = -\frac{2\pi}{\alpha}(\pi_1 + \frac{\alpha}{4\pi}A_2)
\hspace{1.4in} \psi = (\pi_2 - \frac{\alpha}{4\pi}A_1),
\end{equation}
with $ \{ \chi(x), \psi(y) \} = \delta(x-y).$ As usual we choose the $
\chi $ as the first class constraint and
discard the $ \psi $. The gauge invariant Hamiltonian with respect to $
\chi $ is given by
\begin{equation}
{\widetilde H}_{GU} = H + {\displaystyle \int} d^2x \left [
(\frac{2\pi}{\alpha})^2 \phi\phi^* \psi^2 \right ].
\end{equation}
We see that apart from the term $ \int d^2x ~\phi\phi^*(\Phi^1)^2 $
(and those proportional to the ${\widetilde Q}_1$ and $ {\widetilde Q}_2$
in (45)), the Hamiltonians $ {\widetilde H}_{BF} $ and $ {\widetilde
H}_{GU} $ are the same. However this extra term can be easily introduced
in the Hamiltonian $ {\widetilde H}_{GU}, $ since it is basically
proportional to the gauge unfixing first class constraint $ \chi. $
By making use of these extra terms in $ {\widetilde H}_{GU}, $ we can
see this equivalence using the path integral too. After various
redefinitions and integrations, we get the final gauge invariant action
to be
\begin{eqnarray}
S & = & {\displaystyle \int} d^2x dt \left \{ ({\cal D}_{\mu}\phi)^*
{\cal D}^{\mu}\phi + \frac{\alpha}{4\pi}\epsilon_{\mu\nu\lambda}
A^{\mu}\partial^{\nu}A^{\lambda} - \left (\frac{2\pi}{\alpha}\right
)^2\phi\phi^*(\Phi^2)^2 \right.\nonumber\\
&& \hspace{2.0cm}\left. - (\frac{2\pi}{\alpha}\Phi^2)\left[i\phi^*{\cal
D}_1\phi - i\phi({\cal D}_1\phi)^* + \frac{\alpha}{2\pi} F_{02} \right]
\right.\\
&& \hspace{1cm}\left. + \frac{1}{4\phi\phi^*}\left[{\dot \Phi^2} +
\frac{\alpha}{2\pi}F_{01} - [i\phi^*{\cal D}_1\phi - i\phi({\cal
D}_1\phi)^* + \frac{\alpha}{2\pi} F_{02} \right]^2\right\}.\nonumber
\end{eqnarray}
We have deliberately omitted the subscripts BF and GU here, in order to
emphasize that the same result is obtained for both the methods. The BF
action is as given above, with the terms in the $ \Phi^2 $ as extra
terms in order to give the new gauge symmetry. In the GU result the
action is the same as in (48), with the $ \Phi^2 $ being replaced by the
$\psi$. It may also be noted that the action in its final form is not
manifestly Lorentz invariant.
\section{Conclusion}
In conclusion, we have seen that for the three systems above, the two
vastly different methods described in Section 2 unearth essentially the
same gauge theories. This is seen both classically and using the path
integral. In both methods, extra terms have to be introduced
in the Hamiltonian; in the ${\widetilde H}_{BF}
$ case these terms involve new fields, whereas in ${\widetilde H}_{GU}$
these terms are found in the original phase space. Thus, at least as far
as the above three systems are concerned, one need not really introduce
a totally new variable. In other words to get the hidden gauge symmetries
one need not look outside the original system, they are present within
the original system itself.
When their second class constraint structures are considered, the above
three systems are simple ones. The $E$ matrices of (2) involve only
constants, so that getting the gauge invariant Hamiltonians is quite easy;
the new Hamiltonians will have a finite number of terms. This situation
however need not be seen for other systems. Sometime the gauge invariant
Hamiltonians may be in series form, in which case it has to be seen if
closed form expressions are possible. It is to be seen if the two methods
are equivalent in such more general cases also. This question is being
currently looked into.
In our analysis above, the three systems involved only two second class
constraints. It was mentioned earlier that in the gauge unfixing case,
this does not pose any problem in choosing the first class constraint;
either one of the two can be chosen, to give more than one gauge theory.
However in the Batalin-Fradkin case, both constraints are to be converted
into first class; but even here this conversion is always possible, the
reason being the ``square root'' matrix $X$ of (8) can always be obtained
{}from the $E$ matrix.
On the other hand, in the case of more than two second class constraints,
we may have additional problems; in the GU method, the choice of the first
class subset becomes non-trivial, and in the BF method finding the
``square root'' $X$ matrix becomes non-trivial. But once the first class
subset (or $X$ matrix) is found, then the new gauge symmetry is defined. In
this regard we mention that in the GU method, a certain assumption regarding
the $E$ matrix of (2) has to be used to obtain the first class subset.
\newpage
{\noindent\large{\bf{Acknowledgements}}}
\vspace{3mm}
We wish to thank the Council for Scientific and Industrial Research,
New Delhi, India for financial assistance for this work. We also thank
Dr B A Kagali (BU) for constant encouragement, and Prof M N Anandaram
(BU) and Centre for Theoretical Studies (Indian Institute of Science,
Bangalore) for providing computer facilities.
|
\section{Introduction}
Crystallography is concerned with point sets in $\mathbb{R}^{n}$ that are
discrete and distributed more or less uniformly. In ``classical''
crystallography, periodicity was imposed as well, but this
restriction is not considered as fundamental in the ``modern'' era. With
the discovery of intermetallic quasicrystals \cite{AlMn} in the early
1980s, it became clear that there exist aperiodic point sets that
share a basic property with periodic point sets. This property should
really be associated with a distribution: in this case, the
distribution formed by placing a Dirac delta at each point of the set.
In these terms, the class of aperiodic sets singled out by
crystallography is characterized by the property that the Fourier
transform of the corresponding distribution has support on a lattice
\cite{Mermin}. The rank of this ``Fourier lattice'' equals the dimension
of space for periodic sets, and exceeds it (but is still finite), in
the case of \emph{quasiperiodic} sets. The vertex set of the Penrose
tiling of the plane \cite{Penrose} is a familiar example of a
quasiperiodic set.
The standard construction of quasiperiodic sets $\mathcal{A}$ begins by embedding
$\mathbb{R}^{n}=Y$ in a larger Euclidean space,
$\mathbb{R}^{m+n}=X\times Y$. Into $X\times Y$ one then immerses a
smooth $m$-manifold $\Surf$ that is (\textit{i}) transversal to $Y$,
and (\textit{ii}) invariant under the action of a lattice $\Lambda$
generated by $n+m$ linearly independent translations in $X\times Y$.
Point sets $\mathcal{A}\subset Y$ are obtained as sections (``cuts'') of
$\Surf$ by spaces parallel to $Y$. More formally, in terms of the
standard projections
\begin{equation}
\begin{align}
\pi_{X}&\colon \Surf\to X\\
\pi_{Y}&\colon \Surf\to Y,
\end{align}
\end{equation}
the section of $\Surf$ at $x\in X$ is the set
\begin{equation}\label{pointSets}
\mathcal{A}(x)=\pi_{Y}\circ{\pi_{X}}^{-1}(x).
\end{equation}
Periodicity or quasiperiodicity of $\mathcal{A}(x)$ is determined by the rank
of the lattice $\Lambda_{Y}=\Lambda\cap Y$. Since the generators of
$\Lambda$ were assumed to be linearly independent,
$\Rk{(\Lambda_{Y})}\leq n$. Quasiperiodicity corresponds to
$\Rk{(\Lambda_{Y})}<n$, with complete absence of periodicity
characterized by $\Lambda_{Y}=\{0\}$. An important motivation for
constructing quasiperiodic sets, in this context, is the fact that symmetry
groups that cannot be realized by periodic point sets in
$\mathbb{R}^{n}$, \emph{can}
be realized by periodic surfaces in $\mathbb{R}^{m+n}$.
Transversality and periodicity are relatively mild restrictions on
the manifold $\Surf$, called the ``atomic surface'' by physicists. A
further restriction, one which leads to point sets called ``model
sets'' \cite{Moody}, is to require that $\Surf$ is the $\Lambda$-orbit of
a polytope in $X$. The algorithm which constructs $\mathcal{A}(x)$ from such
$\Surf$ naturally leads to the terminology ``window'' or ``acceptance
domain'' for the corresponding polytopes. Model sets can always be
organized into finitely many tile shapes, and, because of this
simplicity, have dominated the study of quasiperiodic sets.
A different viewpoint on the construction of $\Surf$, pioneered by
Kalugin \cite{Kalugin} and Katz \cite{Katz}, emphasizes the continuity properties of $\mathcal{A}(x)$
with respect to $x$. Consider in more detail the construction of a
model set: $\Surf=\Poly+\Lambda$, where $\Poly\subset X$ is a
polytope. Now, if $x\in \Poly+\pi_{X}(\lambda)$ for some
$\lambda\in\Lambda$, then $y=\pi_{Y}(\lambda)\in \mathcal{A}(x)$. But now
consider what happens when $x$ crosses the boundary of
$\Poly+\pi_{X}(\lambda)$. As $x$ ``falls off the edge of the earth'',
the corresponding point $y$ in the point set $\mathcal{A}(x)$ disappears. By
the same process, of course, points can spontaneously appear ``out of
thin air''. To gain control over these processes, Kalugin \cite{Kalugin}
and Katz \cite{Katz} advocated a restriction on $\Poly$, in relation to
$\Lambda$, such that whenever $x$ falls off the edge of one polytope,
$\Poly+\pi_{X}(\lambda)$, it falls within another, say
$\Poly+\pi_{X}(\lambda')$. This restriction corresponds mathematically
to the statement that the boundaries of the disconnected components of
$\Surf=\Poly+\Lambda$ can be ``glued'' together to form a topological
manifold without boundary.
In the process of restoring transversality to the glued complex of
polytopes one encounters the problems addressed by singularity theory.
The map $\pi_{X}$ should now be a smooth (but not necessarily 1-to-1)
map of $m$-mainfolds.
In the trivial
situation, when $\pi_{X}$ has no singularities, $\Surf$ must be
diffeomorphic to a collection of hyperplanes. This is the situation explored by
Levitov \cite{Levitov} for point sets in two and three dimensions and various
symmetry groups.
When $\Surf$ is a generic 2-manifold, we have the classic result of
Whitney \cite{Whitney} that the stable singularities of smooth maps, such as
$\pi_{X}$, are folds and cusps, having respectively codimension one
and two. Because the cusp is always accompanied by two folds, the
locus of singular values of $\pi_{X}$ consists of curves. The space
$X$ is thus populated by singular curves such that whenever $x$
crosses a curve, a pair of points in $\mathcal{A}(x)$ merge and annihilate.
One motivation for the present work was the desire to eliminate this
point-merging singularity to the greatest extent possible.
By giving the 2-manifold $\Surf$ a complex structure, and identifying
$X$ with the complex plane, we impose additional regularity by
insisting that $\pi_{X}$ is locally holomorphic. The singularities of $\pi_{X}$
will then be isolated points. A construction that naturally
leads to a $\pi_{X}$ with this property is to let $\Surf$ be (locally)
the graph of a holomorphic function, $f\colon X\to Y$. Globally this
corresponds to a Riemann surface $\Surf$ immersed in $\mathbb{C}^{2}=X\times
Y$ and having an atlas of compatible charts in $X$. The other
ingredient needed by our construction is some way to guarantee
that $\Surf$ is invariant with respect to a lattice $\Lambda$. We meet
this challenge by using conformal maps between triangles to define a
fundamental graph of $\Surf$. Schwarz reflections in the triangle
edges extend this graph and generate the isometry group of $\Surf$.
For appropriate choices of triangles, the isometry group has a lattice
subgroup with the desired properties.
In the second half of this paper we classify a subset of all
Riemann surfaces generated by conformal maps of triangles. This
subset is characterized by the property that the conformal map is regular at one
vertex of the triangles and that the edges at this vertex make the
largest possible angle, $\pi/2$. With the only other restriction being
that the corresponding point sets $\mathcal{A}(x)$ are
discrete in $Y$, one arrives at a set of seven surfaces. Four of these are
quasiperiodic. The point set
obtained from a section of one of them is shown in Figure 1. Also
shown in Figure 1 is a much studied model set \cite{pent}: a tiling of
boats, stars, and jester's-caps (whose vertices coincide with a subset
of the Penrose-tiling vertex set). The point set determined by the
Riemann surface can be said to be approximated by the vertex set of
the tiling by a systematic process that renders the Riemann surface
piecewise flat. The other surfaces obtained in our partial
classification, when flattened, also produce familiar tilings (Fig.
3).
\begin{figure}
\centerline{\epsfbox{fig1.eps}}
\caption{Point set (small circles) given by the section of a Riemann
surface compared with a tiling of boats, stars, and jester's caps \cite{pent}.}
\end{figure}
\newpage
\section{Riemann surfaces generated by conformal maps}
\subsection{Immersed Riemann surfaces}
We consider Riemann surfaces as analytically continued holomorphic
functions, interpreted geometrically as surfaces immersed in
$\mathbb{C}^{2}$. Our treatment follows closely the notation and
terminology of Ahlfors \cite{Ahlfors}.
\begin{defn}
A \emph{function element} $F=(U,f)$ consists of a domain $U\subset
\mathbb{C}$ and a holomorphic function $f\colon U\to \mathbb{C}$.
\end{defn}
\begin{defn}
Function elements $F_{1}=(U_{1},f_{1})$ and $F_{2}=(U_{2},f_{2})$ are \emph{direct
analytic continuations} of each other iff $V=U_{1}\cap U_{2}\neq\emptyset$
and $f_{1}=f_{2}$ when restricted to $V$.
\end{defn}
\begin{defn}
The \emph{complete, global analytic function determined by function element}
$F_{0}=(U_{0},f_{0})$ is the maximal collection of function elements
$\F$
such that for any $F_{i}\in \F$ there exists a chain of
function elements $F_{0},\ldots,F_{i}$, all in $\F$,
with every link in the chain a direct analytic continuation.
\end{defn}
Up to this point the set of function elements comprising a complete,
global analytic function $\F$ only possesses the discrete topology,
where $F_{1}\cap F_{2}=\emptyset$ whenever $F_{1}$ and $F_{2}$ are
distinct elements of $\F$. By refining this topology we can identify
$\F$ with a surface and, ultimately, a Riemann surface.
Consider a pair of function elements in $\F$, $F_{1}=(U_{1},f_{1})$ and
$F_{2}=(U_{2},f_{2})$. In the refined topology we define the
intersection by
\begin{equation}
F_{1}\cap F_{2}=\left\{
\begin{array}{ll}
F_{3}=(U_{3},f_{3}) & \parbox[t]{1.75in}{if $F_{1}$ and $F_{2}$ are related by
direct analytic continuation,} \\
& \\
\emptyset & \parbox[t]{1.75in}{otherwise,}
\end{array}\right.
\end{equation}
where $U_{3}=U_{1}\cap U_{2}$ and $f_{3}$ is $f_{1}=f_{2}$ restricted
to $U_{3}$. It is straightforward to check that this defines a valid
topology. Moreover, the projection $\pi\colon \F\to \mathbb{C}$ given
by
\begin{equation}
\pi\colon (U,f)\mapsto U,
\end{equation}
provides the complex charts that identify $\F$ with a Riemann surface.
Throughout the rest of this paper we will mostly be interested in
Riemann surfaces immersed in $\mathbb{C}^{2}$.
\begin{defn}
Let $\F$ be a complete, global analytic function. The \emph{immersed
Riemann surface $\Surf$ corresponding to $\F$} is the image of the
immersion $\Psi\colon \F\to \mathbb{C}^{2}$ given by
\begin{equation}
\Psi\colon (U,f)\mapsto \{ (x,f(x))\colon x\in U \}.
\end{equation}
\end{defn}
\begin{notation} We denote the first component of $\mathbb{C}^{2}$ by
$X$, the second by $Y$.
\end{notation}
If
we restrict the immersion $\Psi$ to a single function
element, $F_{0}=(U_{0},f_{0})$, we obtain the \emph{graph}
\begin{equation}
\Surf_{0}=\{ (x,f_{0}(x))\in X\times Y\colon x\in U_{0}\}\label{graph}
\end{equation}
Thus $\Surf_{0}$ represents a piece of $\Surf$ and in fact determines all of
$\Surf$; $\Surf$ is
connected because every pair of function elements in a complete
global analytic function is related by a chain of direct analytic
continuations. $\Surf$ is the \emph{completion} of $\Surf_{0}$.
\begin{notation}
We write $[\Surf_{0}]$ to denote the completion of the graph $\Surf_{0}$.
\end{notation}
Since all subsequent references to ``Riemann surface'' will be as
a surface immersed in $X\times Y$, we drop the qualifier ``immersed'' below.
We also omit the term ``complete'', since the only instances of
incomplete surfaces, graphs, will always be identified as such. Given a
Riemann surface $\Surf$, we will frequently make use of the
projections
\begin{equation}
\begin{align}
\pi_{X}&\colon \Surf\to X,\\
\pi_{Y}&\colon \Surf\to Y.
\end{align}
\end{equation}
The historical construction of Riemann surfaces we have followed can
be criticized for its inequivalent treatment of the spaces $X$ and $Y$.
We can correct this fault by insisting that the functions $f$
appearing in the function elements $(U,f)$ are not just holomorphic
in their respective domains $U$, but \emph{conformal}
(holomorphic with holomorphic inverse).
The graph \eqref{graph} could then be equally written as
\begin{equation}
\Surf_{0}=\{ ({f_{0}}^{-1}(y),y)\in X\times Y\colon y\in V_{0}\},
\end{equation}
where $V_{0}=f_{0}(U_{0})$. If this ``inversion'', or interchange of
$X$ with $Y$, is to work for all function elements $(U,f)$,
then one must remove all points $x_{0}\in U$, where $f$ behaves
locally as $f(x)-f(x_{0})=c(x-x_{0})^{m}+\dotsb$, with $m>1$. These correspond to branch
points of the map $\pi_{Y}$. Conversely, had we begun with the
inverted function elements our Riemann surface would have \emph{included}
branch points of $\pi_{X}$, \textit{i.e.} points where $f$ is singular.
In keeping with tradition we augment our definition of a Riemann
surface $\Surf$
to \emph{include} all points $(x_{0},y_{0})$ where $\Surf$ behaves
locally like the algebraic curve $(y-y_{0})^{n}=c(x-x_{0})^{m}$,
where $m$ and $n$ are positive integers.
\begin{defn}
A point $(x,y)\in \Surf$ is \emph{regular} if the corresponding complete, global
analytic function
contains a function
element $(U,f)$, with $x\in U$ and $f$ conformal at $x$. A point which is
not regular is \emph{singular}.
\end{defn}
\subsection{Transformations}
Two transformations of Riemann surfaces will be needed in our
discussion of symmetry properties. These are defined in terms of
their action on the spaces $X$ and $Y$ and induce a transformation on
Riemann surfaces as subsets of $X\times Y$. Let $(x,y)$ be a
general point in $X\times Y$ and define
the following transformations:
\begin{align}
\tau(a,b;c,d)&: (x,y)\mapsto (a x+b ,c y+d )\\
\sigma &: (x,y)\mapsto (\bar{x},\bar{y})
\end{align}
Transformation $\tau$ (for complex constants $a$, $b$, $c$, and $d$)
is the general bilinear map while $\sigma$ corresponds
to Schwarz reflection (componentwise complex conjugation).
\begin{lem}
If $\Surf$ is a Riemann surface and $T$ is either of the transformations
$\tau$ or $\sigma$, then $T\Surf$ is again a Riemann surface.
\end{lem}
\begin{proof}
Write $T(x,y)=(T_{X}x, T_{Y}y)$ where $T_{X}$ and $T_{Y}$ are just
maps of the complex plane. Since $\Surf$ corresponds to a
complete global analytic function
$\F$, we need to verify
that $T\Surf$ corresponds to some other complete global analytic
function $\F_{T}$. From our definitions we see that $\F_{T}$ is
obtained from $\F$ by substituting each function element $F=(U,f)\in \F$
by $F_{T}=(T_{X}U,T_{Y}\circ f\circ T_{X}^{-1})$. It is
easily checked that $T_{X}$ is open and $T_{Y}\circ f\circ
T_{X}^{-1}$ is holomorphic for both of the transformations
being considered. Thus $F_{T}$ remains a valid function element. One
also verifies that the direct analytic continuation relationships
among function elements are unchanged by these transformations.
\end{proof}
\begin{cor}\label{moveTinside}
If $\Surf_{0}$ is a graph and $T$ is either of the transformations
$\tau$ or $\sigma$, then $[T\Surf_{0}]=T[\Surf_{0}]$.
\end{cor}
Two special transformations are rotations and translations, for
which we introduce the following notation:
\begin{align}
r(\theta,\phi)&=\tau(e^{i\theta},0;e^{i\phi},0)\\
t(u,v)&=\tau(0,u;0,v).
\end{align}
More generally, transformations $T\colon X\times Y\to X\times Y$
which act isometrically on the spaces $X$ and $Y$ are
just the products of Euclidean motions in $X$ and $Y$.
Isometries of Euclidean spaces normally include reflections; to
preserve the structure of the immersed Riemann surface, however, any
reflection in $X$ (complex conjugation) must be accompanied by a
reflection in $Y$.
\begin{defn}
The \emph{isometries of} $X\times Y$ is the group of transformations
generated by $\sigma$, $r(\theta,\phi)$, and $t(u,v)$.
\end{defn}
In what follows we use the term ``isometry'' only in this sense.
Isometries which preserve a Riemann surface $\Surf$ are called
isometries of $\Surf$ and form a group. The maximal group of isometries
is called the isometry group of $\Surf$.
\begin{defn}
The group of \emph{proper isometries} of $X\times Y$ is the normal subgroup of
isometries generated by $r(\theta,\phi)$ and $t(u,v)$. Any element of
the coset, $\sigma g$, where $g$ is a proper isometry, is called a
\emph{Schwarz reflection}.
\end{defn}
\subsection{Surfaces generated by conformal maps of triangles}\label{conformalMapTriangle}
We now focus on the class of Riemann surfaces determined
by graphs which solve a purely geometrical problem:
the conformal map between two bounded triangular regions, $P\subset X$ and
$Q\subset Y$.
The Riemann mapping theorem \cite{Ahlfors} asserts there is a three-parameter
family of conformal maps $f\colon P\to Q$ that extend to
homeomorphisms of the closures $\Bar{P}$ and $\Bar{Q}$. To fix these
parameters we require that the three vertices of $\Bar{P}$ map to
the vertices of $\Bar{Q}$. This defines the graph
\begin{equation}
P|Q=\{(x,f(x))\colon x\in P\},
\end{equation}
and a corresponding Riemann surface $[P|Q]$. The closure of $P|Q$
is defined analogously and is written $\Bar{P}|\Bar{Q}$. One of the
main benefits of using a conformal map of triangles to determine a
Riemann surface $\Surf$ is that its isometry group can be understood
simply in terms of its action on a partition of $\Surf$ into
\emph{tiles}.
Just as $\Bar{P}$ can be decomposed into an interior $P$, edges which
bound $P$, and vertices which bound each edge, there is a
corresponding cell decomposition of the graph $\Bar{P}|\Bar{Q}$. For
example, if $P_{1}$ is one vertex of $\Bar{P}$, and $f(P_{1})=Q_{1}$ is
its image in $\Bar{Q}$, then we use the symbol $P_{1}|Q_{1}$ to
represent the corresponding vertex of $\Bar{P}|\Bar{Q}$. Each vertex of
$\Bar{P}|\Bar{Q}$ is associated with two angles, a vertex angle of $P$
and the corresponding vertex angle of $Q$. Let the three angle pairs
be $\alpha_{i},\beta_{i}$, $i=1,2,3$. If $\alpha_{i}=\beta_{i}$ for
all $i$, then $P$ is similar to $Q$ and $f$ is just a linear map.
Because the corresponding Riemann surface would be trivial (a plane)
we exclude this case. It is impossible to have
$\alpha_{i}\neq\beta_{i}$ for just one $i$ since then the angle sum
could not be $\pi$ in both triangles. Thus we must have at least two vertices with
unequal angles. At these vertices $f$ fails to be conformal. Any
vertex of $\Bar{P}|\Bar{Q}$ where the corresponding angles in $P$ and
$Q$ are unequal will be called a \emph{singular vertex}. The singular
vertices of $\Bar{P}|\Bar{Q}$ are the only singular points of
$\Bar{P}|\Bar{Q}$.
The edges of $\Bar{P}|\Bar{Q}$ (associated with each pair of vertices
$ij=12,13,23$) are effectively the generators of the isometry group of
$[P|Q]$. By $P_{12}|Q_{12}$ we mean the graph given by the
restriction of $f$ to the edge $P_{12}$ of $\Bar{P}$ with image an edge
$Q_{12}$ of $\Bar{Q}$. Consider the triangles $P'$ and $Q'$ obtained
from $P$ and $Q$ by reflection in these edges. The graph $P'|Q'$,
determined by the conformal map $g\colon P'\to Q'$, is clearly related
to $P|Q$ by an isometry of $X\times Y$, a Schwarz reflection which we
call $\sigma_{12}$. Because $\sigma_{12}$ fixes every point of
$P_{12}|Q_{12}$, we have that $f(x)=g(x)$ for all $x\in P_{12}$. A
basic result from complex analysis then tells us that the function
elements $(P,f)$ and $(P',g)$ are related by analytic continuation.
Thus $[P|Q]=
[P'|Q']=[\sigma_{12}(P|Q)]=\sigma_{12}[P|Q]$, by Corollary \ref{moveTinside}.
\begin{defn}
The group $G$ generated by the
Schwarz reflections $\sigma_{ij}$ which fix the three edges of
a triangular graph $P|Q$ is called the \emph{edge group} of $P|Q$. The
edge group of $P|Q$ is a subgroup of the isometry group of $[P|Q]$.
\end{defn}
In order
to show that the edge group of a triangular graph is the \emph{maximal}
isometry group, we first need to
refine the sets on which these groups act.
\begin{notation}
The symbol $\Check{\Surf}$ corresponds to the Riemann surface $\Surf$
whose singular points have been removed.
\end{notation}
\begin{defn}
A \emph{real curve} of the Riemann surface $\Surf$ is any curve $\Gamma\subset
\Check{\Surf}$, homeomorphic to
$\mathbb{R}$, and pointwise invariant with respect to a Schwarz
reflection.
\end{defn}
Since both $\pi_{X}\colon\Check{\Surf}\to X$ and
$\pi_{Y}\colon\Check{\Surf}\to
Y$ are immersions, the map $\pi_{Y}\circ{\pi_{X}}^{-1}\colon
\pi_{X}(\Gamma)\to \pi_{Y}(\Gamma)$ is an immersion as well. Thus it
makes sense to use our graph notation, $\Gamma=\gamma|\delta$, for real
curves, where $\gamma=\pi_{X}(\Gamma)$ and $\delta=\pi_{Y}(\Gamma)$.
A real curve $\gamma|\delta$ is geometrically no different from the
edge of a triangular graph; the projections $\gamma$ and $\delta$ are
always straight lines. Any real curve is isometric with the graph of
a real analytic function.
The three real curves which bound the triangular graph $P|Q$ generate
a topological cell decomposition of $\Bar{P}|\Bar{Q}$ into vertices,
edges, and the graph $P|Q$ itself. The generators of the edge group,
$\sigma_{ij}$, acting on $\Bar{P}|\Bar{Q}$, generate three closed graphs, each
having one edge in common with $\Bar{P}|\Bar{Q}$. By continuing this
construction we obtain a cell decomposition of $[P|Q]$ into
2-cells isometric with $P|Q$, 1-cells isometric with one of the edges
of $\Bar{P}|\Bar{Q}$, and points. The cell complex as a whole defines
a tiling $\T$; the 2-cells by themselves form a set of tiles,
$\T_{2}$, and every element of $\T_{2}$ can be expressed as $g(P|Q)$,
where $g$ is an element of the edge group, $G$.
To show that $G$ is the maximal isometry group we first need to check
that the tiling $\T$ is \emph{primitive},
that is, there is no refinement of the tiles $\T_{2}$ by
additional real curves within $[P|Q]$ we may have missed. For this it
suffices to check that there are no real curves within $P|Q$. Before
we can prove this statement we need some basic properties of real
curves.
\begin{defn}
A real curve is \emph{complete} if it is not a proper subset of any other
real curve.
\end{defn}
\begin{lem}\label{singEndpoints}
The closure in $\Surf$ of a complete real curve
$\gamma|\delta\subset\Check{\Surf}$, if bounded, has singular endpoints.
\end{lem}
\begin{proof}
Without loss of generality let
$\gamma$ and $\delta$ lie on the real axes of, respectively,
$X$ and $Y$. The functions $f$ of the function elements $(U,f)$,
which represent $\Surf$
locally, will then have power series on the real axis (of $X$) with
real coefficients. Since a real power series when analytically
continued along the real axis continues to be real, we can continue
$\gamma|\delta$ until we encounter either a singularity of $f$
or a zero of $f'$ (\textit{i.e.} a singularity of $f^{-1}$ on the real axis of $Y$).
\end{proof}
The next Lemmas deal with the angles formed by intersecting real curves.
\begin{defn}
The \emph{angle between lines} $\gamma$ and $\gamma'$ (in $X$
or $Y$),
denoted $\angle (\gamma, \gamma')$, is
the smallest counterclockwise rotation required to make $\gamma$
parallel to $\gamma'$.
\end{defn}
\begin{lem}\label{sameAngle}
If real curves $\gamma|\delta\subset\Check{\Surf}$ and
$\gamma'|\delta'\subset\Check{\Surf}$ intersect, then
$\angle (\gamma, \gamma')=\angle (\delta, \delta')$.
\end{lem}
\begin{proof}
Near the point of intersection $\Check{\Surf}$ is represented by a
function element $(U,f)$ where $f$ is conformal. The equality of
angles, formed by a pair of lines in $X$ and their images by $f$ in
$Y$, is simply the
geometrical statement that $f$ is conformal.
\end{proof}
\begin{lem}\label{intersectionAngle}
Only a finite number $n>1$ of real curves can intersect at any point
of a nontrivial Riemann surface
and the angle formed by any pair must be a multiple of $\pi/n$.
\end{lem}
\begin{proof}
Suppose $\gamma|\delta$ and $\gamma'|\delta'$ intersect
with angle $\angle (\gamma, \gamma')=\angle (\delta,
\delta')=\alpha>0$ on a nontrivial Riemann surface $\Surf$;
for convenience, let $(0,0)$ be the point of intersection.
These curves are fixed by Schwarz
reflections $\sigma$ and $\sigma'$ respectively, and
$\sigma'\sigma=r(2\alpha,2\alpha)$ is an isometry of $\Surf$.
The neighborhood of the point of intersection is the graph
\begin{equation}
\Surf_{0}=\{(x,f(x))\colon x\in U\},
\end{equation}
where $U$ is a neighborhood of the origin in $X$, and $f$ is
conformal at $x=0$. The Taylor series for $f$ at the origin has the
form
\begin{equation}
f(x)=\sum_{k=1}^{\infty}a_{k}x^{k},
\end{equation}
where $a_{1}\neq 0$. A short calculation shows
\begin{equation}
r(2\alpha,2\alpha)\Surf_{0}=\{(x,f_{\alpha}(x))\colon x\in U_{\alpha}\},
\end{equation}
where $U_{\alpha}=e^{i 2\alpha}U$ is again a neighborhood of the
origin, and
\begin{equation}
f_{\alpha}(x)=\sum_{k=1}^{\infty}a_{k}e^{i 2\alpha(1-k)}x^{k}.
\end{equation}
Since $r(2\alpha,2\alpha)$ is an isometry, the Taylor
series for $f$ and $f_{\alpha}$ must agree, term by term. Now if
$\alpha=\pi \omega$ and $\omega$ is irrational, then $\omega(1-k)$
can be an integer only for $k=1$ (so that $e^{i 2\alpha(1-k)}=1$). But this
requires $a_{k}=0$ for $k>1$ which is impossible since $\Surf$ is
nontrivial. Thus we may assume $\omega=p/q$ where $p$ and $q$ are
relatively prime positive integers, $p<q$ (since $\alpha<\pi$), and
$a_{1+m q}\neq 0$ for some integer $m>0$.
Now let $\gamma''|\delta''$
be any real curve that intersects $\gamma|\delta$ at the origin; then
$\angle (\gamma, \gamma'')=\angle (\delta,
\delta'')=\pi(p'/q')$ by the argument just given, where $p'$ and $q'$ are
relatively prime positive integers, $p'<q'$. However, since $a_{1+m q}\neq
0$, we must have $e^{-i 2\pi(p'/q')m q}=1$, or that $q'$ divides the
product $m q=n$. This shows that $\angle (\gamma, \gamma'')$ is a
multiple of $\pi/n$, for some $n>1$.
\end{proof}
Clearly any triangular graph with a nontrivial isometry must be
``isoceles'' and fails to be primitive because it can be
decomposed into two isometric tiles. This is made precise by the
following Lemma.
\begin{lem}\label{noRealCurvesInside}
Let $P|Q$ be a nontrivial triangular graph with trivial isometry
group, then $P|Q$ contains no real curves.
\end{lem}
\begin{figure}
\centerline{\epsfbox{fig2.eps}}
\caption{Diagrams used in the proof of Lemma \ref{noRealCurvesInside}.}
\end{figure}
\begin{proof}
Suppose $P|Q$ contains a real curve and call its completion
$\gamma|\delta$. We recall that $\gamma$, and its closure in $X$,
$\bar{\gamma}$, are straight lines and $\bar{\gamma}$
cannot have an endpoint within $P$ (Lemma \ref{singEndpoints}).
The possible geometrical
relationships between $\bar{\gamma}$ and $P$ are diagrammed in Figure
2.
Since $P|Q$ is nontrivial, at least two vertices are singular and are
shown circled in each diagram. Either $\bar{\gamma}$ intersects two
edges of $P$, as in cases $A$ and $B$, or, it intersects an edge and the
opposite vertex which may be singular (case $C$) or possibly
regular (case $D$). The vertex labels on the diagram refer to our
notation for the vertex angles and edges. For example, $\alpha_{1}$
and $\beta_{1}$ are the angles in $P$ and $Q$, respectively, of
vertex 1; $P_{12}|Q_{12}$ is the edge (real curve) bounded by vertices
1 and 2, etc.
Case $A$ is easily disposed of using Lemma \ref{sameAngle}:
\begin{equation}
\begin{align}
\angle(\bar{\gamma},P_{12})=
\angle(\bar{\gamma},P_{13})+\alpha_{1}
=\angle(\bar{\delta},Q_{13})+\alpha_{1}
&=\angle(\bar{\delta},Q_{12})-\beta_{1}+\alpha_{1}\\
&=\angle(\bar{\gamma},P_{12})-\beta_{1}+\alpha_{1}.
\end{align}
\end{equation}
This is impossible because vertex 1 is singular
($\alpha_{1}\neq\beta_{1}$).
By using Schwarz reflection to imply the existence of additional real
curves, the remaining cases either reduce to case $A$ or imply the
existence of a singularity within $P|Q$ or one of its edges ---
neither of which is possible.
First consider case $B$. Let $\gamma$ intersect $P_{13}$ at
$x_{1}$ and $P_{23}$ at $x_{2}$, forming angles $\theta_{1}$ and
$\theta_{2}$ (see Fig. 2). Any other complete real curve with
projection $\gamma'$ which intersects $x_{1}$ makes a finite angle
with $\gamma$ by Lemma \ref{intersectionAngle}. Thus we may assume
the angles $\theta_{1}$ and $\theta_{2}$ are the smallest possible
(for a $\gamma$ that intersects
both $P_{13}$ and $P_{23}$). Since one of $\theta_{1}$ and
$\theta_{2}$ must be greater than $\pi/2$, we assume without loss of
generality it is $\theta_{1}$. If we now reflect $\gamma$ in $P_{13}$
we obtain a real curve with projection $\gamma'$ such that $\gamma'$
intersects $P_{13}$ but not $P_{23}$. Thus case $B$ always reduces to
cases $A$ or $C$.
In case $C$ we consider the sequence of real curves with
projections $\gamma_{k}$, where $\gamma_{0}=P_{23}$,
$\gamma_{1}=\gamma$, and $\gamma_{k+1}$ is the image of $\gamma_{k-1}$
under reflection in $\gamma_{k}$. Let $\theta_{k}$ be the angle formed
at vertex 2 in $P$ by $\gamma_{k}$. Clearly for some $k$
we arrive at a $\gamma'=\gamma_{k}$ such that
$\theta=\theta_{k}\leq\alpha_{2}/2$ (see Fig. 2).
This leads to three subcases: $C_{1}$, where
$\angle(\gamma',P_{13})\leq\pi/2$, $C_{2}$, where $\gamma'$ and
$P_{13}$ are perpendicular, and $C_{3}$, where
$\angle(\gamma',P_{13})\geq\pi/2$. In case $C_{2}$, $\theta<\alpha_{2}/2$ since
otherwise $P|Q$ would have a nontrivial isometry (reflection in
$\gamma'|\delta'$). All three subcases immediately lead to
contradictions. In $C_{1}$, reflecting $\gamma'$ in $P_{13}$ presents us
with a $\gamma''$ satisfying case $A$. In $C_{3}$, the image of vertex 1
under reflection in $\gamma'$ implies a singularity within $P$; in
$C_{2}$
the same reflection implies a singularity on $P_{13}$.
Case $D$: we either have $\angle(\gamma,P_{12})=\pi/2$, case $D_{1}$, or
$\angle(\gamma,P_{12})\neq\pi/2$, case $D_{2}$. Since $P$ has no
nontrivial isometry, a reflection in $\gamma$ in case $D_{1}$ would place the image
of either vertex 1 or 2 (both singular) somewhere on $P_{12}$. In
$D_{2}$, a
Schwarz reflection
of $\gamma$ leads to case $A$.
\end{proof}
\begin{thm}\label{maxIsometryGroup}
Let $P|Q$ be a nontrivial triangular graph with trivial isometry
group, then the maximal isometry group of $[P|Q]$ is the edge group
of $P|Q$.
\end{thm}
\begin{proof}
We use the real curves to decompose $[P|Q]$ into
a set of tiles (2-cells) $\T_{2}$. Lemma \ref{noRealCurvesInside}
tells us that $P|Q\in\T_{2}$, so that $\T_{2}=G(P|Q)$, where $G$ is
the edge group of $P|Q$. On the other hand,
if $h$ is an isometry of $[P|Q]$, then $h(P|Q)=P'|Q'\in\T_{2}$,
where $P'|Q'=g(P|Q)$ for some $g\in G$. But since $P|Q$ has no
nontrivial isometries, the map $h^{-1}g\colon P|Q\to P|Q$ must be the
identity
and $h=g$.
\end{proof}
We conclude this section with a formula for the topological genus of
a Riemann surface $[P|Q]$ compactified by the translation subgroup of
its isometry group, the \emph{lattice group} $\Lambda$ of $[P|Q]$.
\begin{defn}
The \emph{vertex groups}
$G_{i}$, $(i=1,2,3)$, of a triangular graph $P|Q$, are the subgroups of the
edge group of $P|Q$ generated by the adjacent edges of, respectively,
the three vertices of
$P|Q$.
\end{defn}
\begin{thm}
Let $P|Q$ be a nontrivial triangular graph with trivial isometry
group. Let $G$ be the isometry group of $[P|Q]$, $\Lambda$
its lattice group, and $G_{i}$, $(i=1,2,3)$, the three vertex groups
of $P|Q$. If $|G/\Lambda|$ is finite, the genus $g$ of the surface
$[P|Q]/\Lambda$ satisfies
\begin{equation}\label{genusFormula}
2-2g=|G/\Lambda|\left(\sum_{i=1}^{3}\frac{1}{|G_{i}|}-\frac{1}{2}\right).
\end{equation}
\end{thm}
\begin{proof}
If $|G/\Lambda|$ is finite we can view $[P|Q]/\Lambda$ as a
finite cell complex. We can relate the number of 0-cells, $N_{0}$, and
the number of 1-cells, $N_{1}$, in this complex to the number of
2-cells, $N_{2}$. Since every 2-cell is bounded by three 1-cells, each
of which bounds exactly one other 2-cell, $N_{1}=(3/2)N_{2}$.
Similarly, the boundary of each 2-cell contains three
0-cells (the vertices $i=1,2,3$), each of which belongs to the
boundary of a number of 2-cells equal to the order of the corresponding
vertex group, $|G_{i}|$. Thus
$N_{0}=(\sum_{i=1}^{3}|G_{i}|^{-1})N_{2}$. Finally, since $P|Q$ has
no nontrivial isometry, and $G/\Lambda$ acts transitively on the 2-cells
of
$[P|Q]/\Lambda$, $N_{2}=|G/\Lambda|$. The result
\eqref{genusFormula} follows from Euler's formula,
$2-2g=N_{2}-N_{1}+N_{0}$.
\end{proof}
\subsection{Discreteness and uniformity}
The whole point of immersing a Riemann surface $\Surf$ in
$X\times Y=\mathbb{C}^{2}$ is
that by forming sections of $\Surf$, \textit{i.e.} intersections with $X=\mathbb{C}$,
one obtains patterns of points. A very primitive
property of a point set, normally taken for granted in
crystallography, is discreteness.
\begin{defn}
The \emph{section} of the Riemann surface $\Surf$ at $x$ is the set
\begin{equation}
\mathcal{A}(x)=\pi_{Y}\circ {\pi_{X}}^{-1}(x).
\end{equation}
\end{defn}
The basic property of a holomorphic function, that its zeros form a
discrete set, translates to the statement that the preimages
${\pi_{X}}^{-1}(x)$ are discrete in a Riemann surface $\Surf$. We use the stronger
property that ${\pi_{X}}^{-1}(x)$ is
discrete in $X\times Y$ to define a \emph{discrete Riemann surface}.
This is equivalent to the following statement about sections:
\begin{defn}
A Riemann surface $\Surf$ is \emph{discrete} if its sections $\mathcal{A}(x)$
are discrete in
$Y$ for every $x\in X$.
\end{defn}
All the statements we can make about discreteness of a Riemann surface
$\Surf$ hinge upon properties of the lattice group of $\Surf$, $\Lambda$.
One property is the rank, $\Rk(\Lambda)$, given by
the cardinality of the generators of $\Lambda$.
The orbit of the origin of $X\times Y$, $\Lambda
(0,0)$, is called a lattice and is also represented by the symbol
$\Lambda$. When $\Rk(\Lambda)=4$, a second property is the
determinant of the lattice, $\det{\Lambda}$. If $\det{\Lambda}>0$, the
four generators of $\Lambda$ are linearly independent (as vectors in
$X\times Y$); if $\det{\Lambda}=0$ the generators are linearly
dependent and $\Lambda$ (as a lattice) is not
discrete in $X\times Y$. A lattice with $\Rk(\Lambda)>4$ is never
discrete in $X\times Y$.
\begin{notation}
The standard measure for a set $A$ is written $|A|$. If $A$ is a
region in $\mathbb{C}$ then $|A|$ is its area; if $A$ is a set of points,
then $|A|$ is its cardinality. Finally, if $\Lambda$ is a rank 4
lattice, then $|\Lambda|=\sqrt{\det{\Lambda}}$ is the volume in
$X\times Y$ of its fundamental region.
\end{notation}
The following Lemma provides a necessary condition for
discreteness:
\begin{lem}\label{notDiscrete}
The lattice group $\Lambda$, of a discrete, nontrivial Riemann
surface $\Surf$, is discrete (as a lattice) in $X\times Y$ and in
particular, $\Rk{\Lambda}\leq 4$.
\end{lem}
\begin{proof}
If $\Lambda$ is not discrete we can find a sequence of
$t(u,v)\in\Lambda$ such that both $\|u\|\to 0$ and $\|v\|\to 0$.
Let $(x_{0},y_{0})$ be a regular point of $\Surf$, then $y_{0}\in \mathcal{A}(x_{0})$.
Near $(x_{0},y_{0})$ we can represent
$\Surf$ by the graph
\begin{equation}
U|V=\{(x,f(x))\colon x\in U\},
\end{equation}
where $U$ is a neighborhood of $x_{0}$ in $X$, $f$ is conformal in
$U$, and
$f(x_{0})=y_{0}$. Since $t(u,v)$ is an isometry, $t(u,v)(U|V)\subset
\Surf$. From
\begin{equation}
t(u,v)(U|V)=\{(x+u,f(x)+v)\colon x\in U\},
\end{equation}
we see that $f(x_{0}-u)+v\in \mathcal{A}(x_{0})$, since $x_{0}-u\in U$ as
$\|u\|$ can be arbitrarily small.
Since $\Surf$ is discrete, $y_{0}$ is isolated in $Y$ and there must be a subsequence
$t(u',v')$ such that $f(x_{0}-u')+v'=f(x_{0})$. If, within the
sequence $t(u',v')$, there is a subsequence $t(u'',v'')$ with $u''=0$,
then $v''=f(x_{0})-f(x_{0}-u'')=0$ and
we have a contradiction. Thus there must be a subsequence with
$u''\neq 0$. Since $f$ is conformal at $x_{0}$,
\begin{equation}
\begin{align}
\lim_{u''\to 0}\frac{f(x_{0})-f(x_{0}-u'')}{u''}&=f'(x_{0})\\
&=\lim_{(u'',v'')\to (0,0)}\frac{v''}{u''}.
\end{align}
\end{equation}
But the second limit, above, is independent of $x_{0}$ so we are
forced to conclude that $f'$ is constant. This is
impossible because $\Surf$ is nontrivial.
\end{proof}
The point sets studied in crystallography normally are Delone sets
and have the property of being \emph{uniformly discrete} \cite{Moody}.
For a point set in $\mathbb{R}^{n}$ this means there exists a real
number $r>0$ such that a spherical neighborhood of radius $r$ about
any point of the set contains no other point of the set. For the sets
$\mathcal{A}(x)$ generated by Riemann surfaces this property is clearly too
strong: it is violated whenever $x$ is near a branch point of
$\pi_{X}$. We therefore adopt a weaker form of this property which is
nevertheless stronger than discreteness and useful in establishing the
existence of the density.
\begin{defn}
Let $B_{r}(y)\subset Y$ be an open disk of radius $r$ centered at $y$.
A Riemann surface $\Surf$, with sections $\mathcal{A}(x)$, is \emph{finitely discrete}
if for some $r>0$, $|\mathcal{A}(x)\cap B_{r}(y)|$ is uniformly bounded above for
all $x\in X$ and $y\in Y$.
\end{defn}
Since a disk of radius $r'$ can always be covered by finitely many
disks of radius $r$, the finitely discrete property holds for any
$r'$ once it has been established for a particular $r$.
We now introduce the class of Riemann surfaces which is the focus of
this study.
\begin{defn}
A Riemann surface
is \emph{crystallographic} if its lattice group $\Lambda$ has rank 4 and
$|\Lambda|>0$.
\end{defn}
\begin{defn}
Any isometry $g$ of $X\times Y$ can uniquely be expressed in the form
$g=g_{0}\lambda$, where $g_{0}$ fixes the origin and $\lambda$ is a
translation. The \emph{derived point group} of the isometry group $G$,
$\psi(G)$, is the image of $G$ by the homomorphism $\psi\colon g\mapsto
g_{0}$.
Since $\Ker{\psi}=\Lambda$, the lattice group of $G$, we have the
isomorphism $\psi(G)\simeq G/\Lambda$.
\end{defn}
\noindent
It is important to remember that $\psi(G)$ need not be a subgroup
of $G$; nevertheless, the lattice group of $G$ is always left
invariant by $\psi(G)$, just as it is invariant within $G$.
Furthermore, if a Riemann surface is crystallographic, then the action
(by conjugation) of $\psi(G)$ on its lattice group $\Lambda$ is a faithful
representation of $G/\Lambda$. Since the isometry group of a (finite
rank) lattice is finite, we have that $G/\Lambda$, for a
crystallographic Riemann surface, is always finite.
\begin{lem}\label{uniformlyDiscrete}
A Riemann surface determined by a triangular graph $P|Q$, if crystallographic,
is finitely discrete.
\end{lem}
\begin{proof}
Let $B_{r}(x,y)\subset X\times Y$ be an open ball of radius $r$
centered at an arbitrary point $(x,y)$. Consider the piece of the
Riemann surface within this ball,
$\Surf_{B}=[P|Q]\cap B_{r}(x,y)$, and the projection $\pi_{X}\colon
\Surf_{B}\to X$. $[P|Q]$ is finitely discrete if there is a
uniform upper bound on the number of preimages $\pi_{X}^{-1}(x)$.
$[P|Q]$ is covered by the orbit of closed graphs,
$G(\Bar{P}|\Bar{Q})$, where $G$ is the edge group of $P|Q$. Let
$\Lambda$ be the lattice group of $G$, then $G(\Bar{P}|\Bar{Q})$ is the union
of cosets, $H_{i}(\Bar{P}|\Bar{Q})$, $i=1,\ldots,N$, where
$N=|G/\Lambda|$ is finite because $[P|Q]$ is crystallographic. Again,
because $[P|Q]$ is crystallographic, all but finitely many graphs in
$H_{i}(\Bar{P}|\Bar{Q})$ have empty intersection with a ball of radius
$r$, in particular, $B_{r}(x,y)$. Thus we have a bound (independent of
$x$ and $y$) on the number of graphs in $G(\Bar{P}|\Bar{Q})$ which
intersect $B_{r}(x,y)$. But a graph can have at most one preimage of
$\pi_{X}$; hence $|\pi_{X}^{-1}(x)|$ is uniformly bounded above.
\end{proof}
The sections $\mathcal{A}(x)$ of a crystallographic Riemann surface also possess a
uniformity with
respect to the parameter $x$. Our handle on this property is provided,
in part, by the smooth behavior of $\mathcal{A}(x)$ with $x$. Before we can
proceed, however, we need to be aware of two point sets in $X$ which
create problems: branch points (of $\pi_{X}$) and crossing points.
\begin{defn}
A point $(x,y)\in\Surf$ is a \emph{self-intersection point} if in the
description of $\Surf$ as a complete global analytic function there exist
function elements $(U,f)$ and $(V,g)$, such that $x\in U\cap V$,
$f\neq g$ in $U\cap V$, and $f(x)=g(x)=y$. The point $x\in X$ is
called a \emph{crossing point}.
\begin{lem}
A Riemann surface determined by a triangular graph $P|Q$ has countably many
self-intersection points and $\pi_{X}$ has countably many branch
points.
\end{lem}
\begin{proof}
$[P|Q]$ is the union of countably many closed graphs
$\Bar{P}_{i}|\Bar{Q}_{i}$ given by the orbit of $\Bar{P}|\Bar{Q}$
under the action of the edge group. Since each graph has at most three
singular points, $\pi_{X}$ has countably many branch points. If there
were uncountably many self-intersection points then uncountably many
must arise from one pair of distinct graphs, say $\Bar{P}_{i}|\Bar{Q}_{i}$ and
$\Bar{P}_{j}|\Bar{Q}_{j}$. Let $f_{i}$ and $f_{j}$ be the
corresponding conformal maps; then $f_{i}(x)=f_{j}(x)$ would have
uncountably many solutions $x\in \Bar{P}_{i}\cap\Bar{P}_{j}$.
Thus either $f_{i}=f_{j}$, a contradiction,
or the zeroes of $f_{i}-f_{j}$ would not be isolated, another
impossibility.
\end{proof}
\end{defn}
\begin{lem}\label{uniformInX}
Let $[P|Q]$ be crystallographic, $X_{c}\subset X$ its crossing
points, and $X_{b}\subset X$ the
branch points of $\pi_{X}$; then for any pair $x,x'\in X\setminus
(X_{b}\cup X_{c})$, there exists a bijection of sections of $[P|Q]$,
\begin{equation}
\Psi\colon \mathcal{A}(x)\to\mathcal{A}(x'),
\end{equation}
such that $\|y-\Psi(y)\|$ is uniformly bounded above for $y\in \mathcal{A}(x)$.
\end{lem}
\begin{proof}
We arrive at $\Psi$ by composing bijections
\begin{equation}
\begin{align}
\Psi_{1}&\colon\mathcal{A}(x)\to\mathcal{A}(x''),\\
\Psi_{2}&\colon\mathcal{A}(x'')\to\mathcal{A}(x'),
\end{align}
\end{equation}
such that
$\|y-\Psi_{1}(y)\|$ and $\|y''-\Psi_{2}(y'')\|$ are (correspondingly)
uniformly bounded. The Lemma then
follows by application of the triangle inequality.
Since $[P|Q]$ is crystallographic, we can partition $X\times Y$
into translates of a bounded fundamental region, $V(0)$, of its
lattice $\Lambda$. Thus for any pair $x,x'\in X$ we can write
\begin{equation}
\begin{align}
(x,0)&\in V(0)+\lambda,\\
(x',0)&\in V(0)+\lambda' ,
\end{align}
\end{equation}
where $\lambda,\lambda'\in\Lambda$. Consider the point
\begin{equation}\label{xDoublePrime}
(x'',y'')=(x,0)+\lambda'-\lambda\in V(0)+\lambda'.
\end{equation}
Since
\begin{equation}
(x''-x',y'')=(x'',y'')-(x',0)\in V(0)-V(0),
\end{equation}
both $\|x''-x'\|$ and $\|y''\|$ have upper bounds independent of $x$ and $x'$.
Now
\begin{equation}
\mathcal{A}(x)=\{y\in Y\colon (x,y)\in [P|Q]\},
\end{equation}
and
\begin{equation}
\mathcal{A}(x)+y''=\{y'\in Y\colon (x,y'-y'')\in [P|Q]\}.
\end{equation}
But $(x,y'-y'')\in [P|Q]$ iff $(x,y'-y'')+\lambda''\in [P|Q]$, where
$\lambda''\in \Lambda$. Choosing
\begin{equation}
\lambda''=\lambda'-\lambda=(x''-x,y''),
\end{equation}
we obtain
\begin{equation}
\mathcal{A}(x)+y''=\{y'\in Y\colon (x'',y')\in [P|Q]\}=\mathcal{A}(x'').
\end{equation}
As our first bijection we take the translation $\Psi_{1}(y)=y+y''$,
where $\|y''\|$ is uniformly bounded from above.
The point of this intermediate step is that for $\Psi_{2}$ we need
consider only pairs of sections with bounded separation $\|x''-x'\|$.
In constructing $\Psi_{2}$ we avoid branch points and crossing points.
Since $x\in
X\setminus (X_{b}\cup X_{c})$, equation \eqref{xDoublePrime} implies $x''\in
X\setminus (X_{b}\cup X_{c})$. Let $\gamma\colon [0,1]\to X\setminus (X_{b}\cup X_{c})$
be a smooth rectifiable curve with $\gamma(0)=x''$
and $\gamma(1)=x'$. To show that $\gamma$ exists we recall that
$X_{b}\cup X_{c}$ is countable.
We can then find $\gamma$ in the uncountable family of
circular arcs with endpoints $x''$ and $x'$, since each point of
$X_{b}\cup X_{c}$ can eliminate at most one arc.
The curve $\gamma(t)$ generates a homotopy of the sections
$\mathcal{A}(x'')$ and $\mathcal{A}(x')$. At each point $y''\in \mathcal{A}(x'')$,
$\gamma(t)$ is lifted to a unique curve
$\gamma(t)|\delta(t)\subset[P|Q]$ with endpoint
$(\gamma(0),\delta(0))=(x'',y'')$ and we define our second
bijection by $\Psi_{2}(y'')=\delta(1)$. To finish the proof we need
to show that $\|\delta(1)-\delta(0)\|$ is uniformly bounded.
Let $\Check{P}$ be the closed subset of $\Bar{P}$ that is a
suitably small distance $r$ or greater from any of its vertices that are
branch points of $\pi_{X}$. Let $\Check{P}|\Check{Q}\subset \Bar{P}|\Bar{Q}$
be the corresponding graph. The orbit under the edge group,
$\Check{\Surf}=G(\Check{P}|\Check{Q})\subset
[P|Q]$, is a Riemann surface from which all the points of
ramification (of the map $\pi_{X}$) have been ``cut out''. The
complement, $\Hat{\Surf}=[P|Q]\setminus\Check{\Surf}$, is the disjoint union of the
branched neighborhoods of all the points of ramification. It is
possible to find curves $\gamma(t)$, such as the circular arcs
considered above, where
the branched neighborhoods $\Hat{\Surf}_{i}$ visited by
$\gamma(t)|\delta(t)$ are visited only once, for
$t\in T_{i}\subset [0,1]$. Also, because we can bound the length $L$
of $\gamma$ and there is a minimum distance between branch points (on
the branched covering of $X$), the number of such subintervals $T_{i}$
is bounded, \textit{i.e.} $i=1,\dots,N$. The bounds on $N$ and $L$
are uniform bounds, independent of the points $x$ and $x'$.
Two additional bounds are needed before we can proceed to bound
$\|\delta(1)-\delta(0)\|$. The first is an upper bound $D$ on
the diameter of the projection of a
branched neighborhood, $\pi_{Y}(\Hat{\Surf}_{i})$. This follows from
the fact that $\Hat{\Surf}_{i}$ is isometric with the branched neighborhood
$\Hat{\Surf}_{j}$ of a vertex of $\Bar{P}|\Bar{Q}$, and
$\Hat{\Surf}_{j}\subset G_{j}(P|Q)$, where $G_{j}$ is the
corresponding vertex group. Clearly the maximum diameter of
$\pi_{Y}(G_{j}(P|Q))$ is bounded because $Q$ is bounded.
The map $f\colon P\to Q$ (which defines
$P|Q$), when restricted to $\Check{P}$ is conformal
and $\|f'\|$ has a maximum value, $\mu$, since $\Check{P}$ is closed.
This means that if $\gamma(t)|\delta(t)\in \Check{P}|\Check{Q}$,
then
\begin{equation}
\left\|\frac{d\delta}{dt}\right\|=\left\|f'(\gamma)\frac{d\gamma}{dt}
\right\|\leq\mu\left\|\frac{d\gamma}{dt}\right\|.
\end{equation}
Because $\Check{\Surf}$ is generated from $\Check{P}|\Check{Q}$ by the
action of $G$, this bounds applies globally, for $\gamma(t)|\delta(t)\in
\Check{\Surf}$.
We are now ready to complete the proof:
\begin{equation}\label{mainInequality}
\|\delta(1)-\delta(0)\|=
\left\|\int_{[0,1]}\frac{d\delta}{dt}\right\|
\leq\left\|\int_{\cup_{i=1}^{N}T_{i}}\frac{d\delta}{dt}\right\|+
\left\|\int_{[0,1]\setminus\cup_{i=1}^{N}T_{i}}\frac{d\delta}{dt}\right\|.
\end{equation}
For each piece of the curve in a branched neighborhood we have
\begin{equation}
\left\|\int_{T_{i}}\frac{d\delta}{dt}\right\|\leq D,
\end{equation}
while in the complement ($\Check{\Surf}$),
\begin{equation}
\begin{align}
\left\|\int_{[0,1]\setminus\cup_{i=1}^{N}T_{i}}\frac{d\delta}{dt}\right\|&\leq
\int_{[0,1]\setminus\cup_{i=1}^{N}T_{i}}\left\|\frac{d\delta}{dt}\right\|\\
&\leq\mu\int_{[0,1]\setminus\cup_{i=1}^{N}T_{i}}\left\|\frac{d\gamma}{dt}\right\|\\
&\leq \mu L.
\end{align}
\end{equation}
Inequality \eqref{mainInequality} thus becomes
\begin{equation}
\|\delta(1)-\delta(0)\|\leq N D + \mu L.
\end{equation}
\end{proof}
For discrete Riemann surfaces with sufficiently uniform sections
$\mathcal{A}(x)$, one can define their \emph{density}.
\begin{defn}
Let $B_{R}(0)\subset Y$ be a disk of radius $R$ centered at the origin. The limit
\begin{equation}
\rho(x)=\lim_{R\to \infty}\frac{|B_{R}(0)\cap \mathcal{A}(x)|}{|B_{R}(0)|},
\end{equation}
if it exists and is finite, is the density of $\mathcal{A}(x)$.
\end{defn}
With the aid of Lemmas \ref{uniformlyDiscrete} and \ref{uniformInX}
we can show, that for a crystallographic Riemann surface generated by a triangular
graph,
$\rho(x)$ exists and is (essentially) independent of
$x$.
\begin{notation}
The standard volume form in $X$ is $\omega_{X}=dx\wedge d\bar{x}$, its
pullback on a Riemann surface $\Surf$ is written
${\pi_{X}}^{\ast}\omega_{X}$.
\end{notation}
\begin{thm}\label{density}
If $[P|Q]$ is crystallographic with lattice group $\Lambda$,
its sections $\mathcal{A}(x)$ have density
\begin{equation}\label{KaluginFormula}
\rho=\frac{1}{|\Lambda|}\int_{[P|Q]/\Lambda}{\pi_{X}}^{\ast}\omega_{X},
\end{equation}
independent of $x$, provided $x$ is not a crossing point or a
branch point of $\pi_{X}$.
If $G$ is the edge group of $P|Q$, then
\begin{equation}\label{rhoFormula}
\rho=\frac{|G/\Lambda| |P|}{|\Lambda|}.
\end{equation}
\end{thm}
\begin{proof}
\begin{notation}
The expression $c=\order{(1/R)}$ indicates there exist constants $c_{1}$
and $c_{2}$ (independent of $R$) such that for sufficiently large $R$,
$c_{1}/R<c<c_{2}/R$.
\end{notation}
Let $B_{R}(0)\subset Y$ be a disk of radius $R$ centered at the origin
and let
\begin{equation}
N_{R}(x)=|B_{R}(0)\cap \mathcal{A}(x)|.
\end{equation}
We first obtain a bound on the difference,
$N_{R}(x)-N_{R}(x')$, when neither $x$ nor $x'$
is a crossing point or a branch point of $\pi_{X}$.
By Lemma \ref{uniformInX} there exists
a bijection $\Psi\colon \mathcal{A}(x)\to\mathcal{A}(x')$ such that if $y\in B_{R}(0)\cap
\mathcal{A}(x)$, then $\Psi(y)\in B_{R+d}(0)\cap\mathcal{A}(x')$, where $d>0$ is a
constant independent of $R$, $x$, and $x'$. This shows
\begin{equation}
\begin{align}
N_{R}(x)&\leq |B_{R+d}(0)\cap\mathcal{A}(x')|\\
&=N_{R}(x')+|(B_{R+d}(0)\setminus B_{R}(0))\cap\mathcal{A}(x')|.
\end{align}
\end{equation}
We can cover the annulus $B_{R+d}(0)\setminus B_{R}(0)$ by $M_{R}$
disks $B_{r}(y')$ of a fixed radius $r>0$, where, for sufficiently
large $R$, $M_{R}<m R$ and $m$ is a constant independent of $R$. By
Lemma \ref{uniformlyDiscrete}, $|B_{r}(y')\cap\mathcal{A}(x')|<n$, where $n$ is
independent of $x'$ and $y'$. Thus $N_{R}(x)-N_{R}(x')<m n R$.
Combining this bound with the bound obtained by
interchanging $x$ and $x'$, we arrive at the statement
\begin{equation}\label{boundDeltaN}
N_{R}(x)-N_{R}(x')=|B_{R}(0)|\order{(1/R)}.
\end{equation}
We now introduce a disk $C_{R}(0)\subset X$ and consider the region
$W_{R}=C_{R}(0)\times B_{R}(0)\subset X\times Y$. Since the set of
branch points and crossing points is countable and has
zero measure in $X$, and $\pi_{X}$ is otherwise smooth,
\begin{equation}\label{forms1}
\int_{\pi_{X}([P|Q]\cap W_{R})}\omega_{X}=
\int_{[P|Q]\cap W_{R}}{\pi_{X}}^{\ast}\omega_{X}.
\end{equation}
Because $[P|Q]$ is crystallographic, we can partition $X\times
Y$ into translates of a bounded fundamental region of its lattice,
$V(0)$.
Let $\pi_{\Lambda}\colon [P|Q]\to [P|Q]/\Lambda$ be the standard
projection on the quotient. On $[P|Q]\cap V(0)$ the map
$\pi_{\Lambda}$ is 1-to-1 and
\begin{equation}\label{forms2}
\int_{[P|Q]\cap V(0)}{\pi_{X}}^{\ast}\omega_{X}=
\int_{[P|Q]/\Lambda}(\pi_{\Lambda}^{-1})^{\ast}{\pi_{X}}^{\ast}\omega_{X}=\rho |\Lambda|.
\end{equation}
This defines $\rho$, which we can make positive by appropriate choice
of orientation on $[P|Q]$.
Turning now to the region $W_{R}$, there is a maximal subset
$\Lambda_{-}\subset \Lambda$ such that $\Lambda_{-}+V(0)\subset W_{R}$ and a
smallest subset $\Lambda_{+}\subset \Lambda$ such that
$W_{R}\subset\Lambda_{+}+V(0)$. If $\lambda\in\Lambda_{+}\setminus\Lambda_{-}$,
then $V_{\lambda}=(\lambda+V(0))\cap W_{R}$ is a proper subset of a
fundamental region and
\begin{equation}
0<\int_{[P|Q]\cap
V_{\lambda}}{\pi_{X}}^{\ast}\omega_{X}<\rho|\Lambda|.
\end{equation}
From this it follows that
\begin{equation}
\rho|\Lambda_{-}||\Lambda|<
\int_{[P|Q]\cap W_{R}}{\pi_{X}}^{\ast}\omega_{X}<
\rho|\Lambda_{+}||\Lambda|,
\end{equation}
and, from straightforward estimates of $\Lambda_{+}$ and $\Lambda_{-}$, we
conclude
\begin{equation}\label{forms3}
\int_{[P|Q]\cap
W_{R}}{\pi_{X}}^{\ast}\omega_{X}=\rho|W_{R}|(1+\order{(1/R)}).
\end{equation}
The projection $\pi_{X}([P|Q]\cap W_{R})$ covers the disk
$C_{R}(0)$ multiple times, the multiplicity at the point $x\in
C_{R}(0)$ being the number $N_{R}(x)$ defined above. Thus
\begin{equation}\label{forms4}
\int_{\pi_{X}([P|Q]\cap W_{R})}\omega_{X}=
\int_{C_{R}(0)}N_{R}(x)\omega_{X}.
\end{equation}
We can again neglect the countable set of branch points $X_{b}$ and
crossing points $X_{c}$ to argue,
for $x_{0}\in X\setminus (X_{b}\cup X_{c})$ fixed,
\begin{equation}\label{forms5}
\begin{align}
\int_{C_{R}(0)}N_{R}(x)\omega_{X}&=
\int_{C_{R}(0)}N_{R}(x_{0})\omega_{X}+
\int_{C_{R}(0)}(N_{R}(x)-N_{R}(x_{0}))\omega_{X}\nonumber\\
&=N_{R}(x_{0})|C_{R}(0)|+|C_{R}(0)||B_{R}(0)|\order{(1/R)},
\end{align}
\end{equation}
where in the last step we used \eqref{boundDeltaN}. Combining
\eqref{forms1}, \eqref{forms3}, \eqref{forms4}, \eqref{forms5}, and
using $|W_{R}|=|B_{R}(0)||C_{R}(0)|$, we obtain
\begin{equation}
\frac{N_{R}(x_{0})}{|B_{R}(0)|}=\rho(1+\order{(1/R)}),
\end{equation}
and thus
\begin{equation}
\lim_{R\to\infty}\frac{N_{R}(x_{0})}{|B_{R}(0)|}=\rho.
\end{equation}
To evaluate $\rho$ from \eqref{forms2}, we regard $[P|Q]/\Lambda$
as $|G/\Lambda|$ equivalence classes of tiles, all isometric to $P|Q$.
The result \eqref{rhoFormula} follows because the integral of the
form ${\pi_{X}}^{\ast}\omega_{X}$ over $P|Q$ is just the volume of
$\pi_{X}(P|Q)=P$ in $X$.
\end{proof}
Formula \eqref{KaluginFormula} for the density was introduced by
Kalugin \cite{Kalugin} to extend the notion of stoichiometry to quasicrystals.
Because the 2-form ${\pi_{X}}^{\ast}\omega_{X}$ is closed, this
formula gives the same density for 2-manifolds homologous in the torus
$(X\times Y)/\Lambda$. One must remember, however, that this homology
invariant only corresponds to the true density when the map $\pi_{X}$
is orientation preserving (see equation \eqref{forms4}).
\newpage
\section{Classification of discrete Riemann surfaces generated by
conformal maps of right triangles}
\subsection{Conformal maps of right triangles}
The simplest nontrivial conformal maps of triangles, $f\colon P\to Q$,
are those where one of
the vertices of the corresponding graph $P|Q$ is regular. The
edges of $P|Q$ adjacent to this vertex are real
curves, and by Lemma \ref{intersectionAngle}, belong to a set of $n>1$ real
curves intersecting at the same vertex with minimum angle $\pi/n$.
This suggests that among those maps with one regular vertex,
the simplest case is $n=2$, \textit{i.e.} the conformal map of
right triangles with the right angle being the regular vertex.
Without loss of
generality, we give $P$ and $Q$ a standard scale, position, and
angular orientation as specified by the vertices $x_{i}|y_{i}$ of the
corresponding graph $P|Q$:
\begin{equation}\label{triangleVertices}
\begin{align}
x_{1}|y_{1}&=0|0\nonumber\\
x_{2}|y_{2}&=\cos{\alpha}|\cos{\beta}\\
x_{3}|y_{3}&=e^{i\alpha}|e^{i\beta},\nonumber
\end{align}
\end{equation}
where $0<\alpha<\pi/2$ and
$0<\beta<\pi/2$ are two free real parameters. Below we frequently use
the abbreviations $a=\cos{\alpha}$, $b=\cos{\beta}$.
$P$ and $Q$ have angles $\alpha$ and $\beta$, respectively, at vertex
1, and the corresponding complementary angles at vertex 3.
A nontrivial graph $P|Q$ has $\alpha\neq\beta$ with vertices 1 and 3
singular; vertex 2 is regular.
The Schwarz-Christoffel formula \cite{Ahlfors} gives the conformal map
$f$ as the composition $f=h\circ g^{-1}$ where $g$ and $h$ map the
upper half plane of $Z=\mathbb{C}$ conformally onto, respectively, $P$
and $Q$. Explicitly:
\begin{align}
g(z)&=A\int_{0}^{z}z^{\frac{\alpha}{\pi}-1}
(1-z)^{-\frac{1}{2}}dz,\label{map_{g}}\\
h(z)&=B\int_{0}^{z}z^{\frac{\beta}{\pi}-1}
(1-z)^{-\frac{1}{2}}dz.\label{map_{h}}
\end{align}
In both \eqref{map_{g}} and \eqref{map_{h}} the branches of the
fractional powers are chosen so that the integrands are real and
positive for
$z\in (0,1)$. The normalization factors $A$ and $B$ are positive real
numbers determined by the conditions $g(1)=a$ and $h(1)=b$. Further properties
of $g$ and $h$ are easily checked, in
particular, $g(\infty)=e^{i\alpha}$ and $h(\infty)=e^{i\beta}$.
\subsection{Isometry groups}
\begin{notation}
Let $K$ be a set of elements of a group $G$, and let $k\in G$ be some
element. We denote by $\langle K\rangle$
the subgroup generated by the elements of $K$, and by $\{k\}_{G}$ the
conjugacy class of $k$ in $G$.
\end{notation}
In the case of right triangles, $P|Q$ can have a nontrivial isometry only if
$\alpha=\pi/4$ and $\beta=\pi/4$. But this makes $P|Q$ trivial. From
Theorem \ref{maxIsometryGroup} we know that for nontrivial $P|Q$ the
isometry group $G$ is just the edge group generated by the three
Schwarz reflections:
\begin{equation}
\begin{align}
\sigma_{12}&=\sigma,\\
\sigma_{13}&=r(2\alpha,2\beta) \sigma,\\
\sigma_{23}&=t(2 a,2 b) r(\pi,\pi) \sigma.
\end{align}
\end{equation}
From the isometry group
\begin{equation}\label{generators}
G=\langle\sigma_{12},\sigma_{13},\sigma_{23}\rangle
=\langle\sigma,r(2\alpha,2\beta),t(2 a,2 b)r(\pi,\pi)\rangle,
\end{equation}
we wish to extract the lattice group $\Lambda$. Helpful in this
enterprise are the vertex group
\begin{equation}\label{G1}
G_{1}=\langle\sigma,r(2\alpha,2\beta)\rangle,
\end{equation}
and its cyclic subgroup,
\begin{equation}\label{R}
R=\langle r(2\alpha,2\beta)\rangle.
\end{equation}
Sets of translations invariant with respect to $G_{1}$, or \emph{stars}, play
a central role in the construction of $\Lambda$. In what follows we
will need two stars:
\begin{equation}
\begin{align}
\Sigma&=\{t(2 a,2 b)\}_{G_{1}},\label{Sigma}\\
\Sigma^{-1}&=r(\pi,\pi)\,\Sigma\,r(\pi,\pi)=\{t(-2 a,-2
b)\}_{G_{1}}.\label{Sigma-1}
\end{align}
\end{equation}
The main result is contained in the following Lemma:
\begin{lem}\label{groups}
Let $G$ be the isometry group of the Riemann surface $[P|Q]$
generated by the conformal map of the right triangles specified in
\eqref{triangleVertices} and let $R$ be defined by \eqref{R}, $G_{1}$
by \eqref{G1}, $\Sigma$ and $\Sigma^{-1}$ by \eqref{Sigma} and
\eqref{Sigma-1}. If $r(\pi,\pi)\in
R$, then $G$ has lattice group $\Lambda=\langle\Sigma\rangle$ and
$G=\Lambda G_{1}$; otherwise, $G$ has lattice group
$\Lambda=\langle\Sigma\Sigma^{-1}\rangle$ and $G=\Lambda G_{1}\, \cup\, t(2
a,2 b) r(\pi,\pi)\Lambda G_{1}$.
\end{lem}
\begin{proof}
First consider the case $r(\pi,\pi)\in R$; then
\begin{equation}
G=\langle\sigma,r(2\alpha,2\beta),t(2 a,2 b)\rangle.
\end{equation}
Now consider the group $H=\Lambda G_{1}$, where $\Lambda=\langle \Sigma\rangle$
is normal in $H$. Clearly $H\subset G$. Moreover, one easily verifies $gH=Hg=H$,
where $g$ is any
of the
three generators of $G$. These two facts together show $G=H$;
$\Lambda$ is clearly the lattice group of $G$.
Next consider the case $r(\pi,\pi)\notin R$. For the generators of
$G$ we must now use \eqref{generators}. Consider the group
$\Tilde{G}=\Tilde{\Lambda}G_{1}$, where $\Tilde{\Lambda}=\langle \Sigma
\Sigma^{-1}\rangle$ is normal in $\Tilde{G}$. Clearly
$\Tilde{G}\subset G$. In contrast to the previous case, we can now
only verify that $g\Tilde{G}g^{-1}=\Tilde{G}$, where $g$ is any of the
three generators in \eqref{generators}. Thus $\Tilde{G}$ is normal in
$G$. $\Tilde{G}$ has index at most two, since multiplication of
$\Tilde{G}$ by the generators of $G$ produces at most two, possibly
distinct, cosets: $\Tilde{G}$ and $\Tilde{G}'=t(2 a,2 b)
r(\pi,\pi)\Tilde{G}$. But if $\Tilde{G}'=\Tilde{G}$, then we would
have some $\tilde{\lambda}\in\Tilde{\Lambda}$ and some $g_{1}\in G_{1}$ such that
$\tilde{\lambda} g_{1}=t(2 a,2 b) r(\pi,\pi)$, or $g_{1}=\tilde{\lambda}^{-1} t(2 a,2
b) r(\pi,\pi)$. Since $g_{1}$ fixes the origin, $\tilde{\lambda}^{-1} t(2 a,2
b)$ must be the trivial translation and $g_{1}=r(\pi,\pi)$. This
contradicts our assumption $r(\pi,\pi)\notin R$ and we conclude that
$\Tilde{G}$ and $\Tilde{G}'$ are distinct. Let $\Lambda$ be the
lattice group of $G$. Clearly $\Tilde{\Lambda}\subset\Lambda$. Now suppose
$\lambda\in\Lambda$ but $\lambda\notin\Tilde{\Lambda}$. Since then
$\lambda\notin\Tilde{G}$, we must have $\lambda\in\Tilde{G}'$, that
is, $\lambda=t(2 a,2 b)r(\pi,\pi)g_{1}\tilde{\lambda}$ for some
$g_{1}\in G_{1}$ and $\tilde{\lambda}\in\Tilde{\Lambda}$. But this implies
$r(\pi,\pi)g_{1}=t(-2 a,-2 b)\lambda{\tilde{\lambda}}^{-1}$, a translation,
and we arrive at the contradiction $r(\pi,\pi)g_{1}=1$. Thus $\Lambda=\Tilde{\Lambda}$.
\end{proof}
\subsection{The discreteness restriction}
The requirement that $[P|Q]$ is discrete places strong constraints on
the angles $\alpha$ and $\beta$ of the right triangles $P$ and $Q$.
\begin{lem}\label{rationalAngle}
If $[P|Q]$ is discrete, then $\alpha$ and $\beta$ are
rational multiples of $\pi$.
\end{lem}
\begin{proof}
By Lemma \ref{notDiscrete}, $[P|Q]$ has a discrete lattice, \textit{i.e.}
the orbit $\Lambda\cdot (0,0)$ is discrete in $X\times Y$. For right
triangles, Lemma \ref{groups} gives us $\Lambda=\langle \Sigma\rangle$
if $r(\pi,\pi)\in R$,
$\Lambda=\langle\Sigma\Sigma^{-1}\rangle$ otherwise. Thus
discreteness of $[P|Q]$ implies discreteness of the star $\Sigma\cdot (0,0)$
in $X\times Y$. Since $\Sigma=\{t(2 a, 2 b)\}_{G_{1}}=\{t(2 a, 2
b)\}_{R}$, $\Sigma\cdot (0,0)$ is just the orbit of $t(2 a,2 b)\cdot
(0,0)$ under action of the group $R$ generated by $r(2\alpha,2\beta)$.
Clearly $\Sigma\cdot (0,0)$ lies in a 2-torus $S_{1}\times S_{1}$
embedded in $X\times Y$. Since $\Sigma\cdot (0,0)$ is discrete, there
is a disjoint union of neighborhoods, each containing just one element
of $\Sigma\cdot (0,0)$. Moreover, since $R$ is an isometry of $X\times
Y$ and acts transitively on $\Sigma\cdot (0,0)$, there is a uniform
lower bound on the volumes of these neighborhoods. This implies the
existence of disjoint neighborhoods in $S_{1}\times S_{1}$, again with
a uniform lower bound on their measure. Since $S_{1}\times S_{1}$ has
finite measure, this is only possible if $R$ has finite order.
\end{proof}
\begin{notation}
Given positive integers $m$ and $n$, $\GCD{(m,n)}$ is their greatest
common divisor, $\LCM{(m,n)}$ their least common multiple.
\end{notation}
Lemma \ref{rationalAngle} allows us to write $\alpha=\pi (i/k)$,
$\beta=\pi (j/l)$, where $i$, $j$, $k$ and $l$ are positive integers
and $\GCD{(i,k)}=\GCD{(j,l)}=1$. Since $R=\langle r(2\pi(i/k),2\pi(j/l))\rangle$,
we identify $n=\LCM{(k,l)}$ as the order of $R$. A more convenient
parameterization is given by
\begin{equation}
\alpha=\pi\frac{p}{n},\quad\quad\beta=\pi\frac{q}{n},\label{angles}
\end{equation}
where $p=i(n/k)$, $q=j(n/l)$ are positive integers with no common factors that
are also factors of $n$.
The transformation
\begin{equation}\label{complementaryAngles}
t(\sin{\alpha},\sin{\beta})\,r(-\pi/2,-\pi/2)\,t(-\cos{\alpha},-\cos{\beta})\,\sigma,
\end{equation}
has the effect of replacing the
angles $\alpha$ and $\beta$
by $\alpha'=\pi/2-\alpha$ and $\beta'=\pi/2-\beta$ in the definition
of the triangles $P$ and $Q$.
Without loss of
generality we may therefore take the smallest of the four
angles $\alpha$, $\alpha'$, $\beta$ and $\beta'$, and rename this
angle $\alpha$. Transformation \eqref{complementaryAngles},
as well as an interchange of the spaces
$X$ and $Y$, will then give all other cases of triangles $P$ and $Q$
(in their standard position, orientation and scale).
From $0<\alpha\leq \alpha'$ we obtain
\begin{equation}
0<p\leq\frac{n}{4}.\label{pInequality}
\end{equation}
The inequalities $\alpha\leq\beta$ and $\alpha\leq\beta'$, plus
the condition to avoid triviality, $\alpha\neq\beta$, then give
\begin{equation}
p<q\leq\frac{n}{2}-p.\label{qInequality}
\end{equation}
Together, inequalities \eqref{pInequality} and \eqref{qInequality}
have no solution unless
\begin{equation}
n\geq 6.\label{lowerBound}
\end{equation}
Since the vertex group $G_{1}$ is generated by $R$ and $\sigma$,
\begin{equation}\label{orderG1}
|G_{1}|=2 n.
\end{equation}
We can use transformation \eqref{complementaryAngles}, which has the
effect of interchanging the angles at vertices 1 and 3, to compute the
order of $G_{3}$. Let $n'$, $p'$ and $q'$ be the integers
parameterizing the angles at vertex 3 (as in \eqref{angles}); then
\begin{equation}\label{switch1and3}
n'=\frac{2 n}{\GCD{\left(2 n,n-2 p,n-2 q\right)}}
\end{equation}
\begin{equation}
\begin{align}
\frac{p}{n}+\frac{p'}{n'}&=\frac{1}{2}\nonumber\\
\frac{q}{n}+\frac{q'}{n'}&=\frac{1}{2},\nonumber
\end{align}
\end{equation}
and,
\begin{equation}\label{orderG3}
|G_{3}|=2 n'.
\end{equation}
Finally, since the group of the regular vertex
is generated by two real curves which intersect at right angles,
\begin{equation}\label{orderG2}
|G_{2}|=4.
\end{equation}
From
\eqref {generators} we find
\begin{equation}
\psi(G)=\langle G_{1}, r(\pi,\pi) \rangle=\langle \sigma, R,
r(\pi,\pi) \rangle
\end{equation}
for the derived point group of $G$.
Recognizing $n=|R|$ as the order of an element of the
isometry group of the \emph{lattice} $\Lambda$, we can use the
following theorem of Senechal \cite{Senechal} and Hiller \cite{Hiller}, and the
requirement of discreteness, to bound $n$ from above.
\begin{thm}[Senechal \cite{Senechal}, Hiller \cite{Hiller}]\label{smallestRank}
Let $N(n)$ be the smallest integer such that the group
$\GL{(N(n),\mathbb{Z})}$ has an element of order $n$, then
\begin{equation}
N(n)=\sum_{{p_{i}}^{m_{i}}\neq 2}\phi({p_{i}}^{m_{i}}),
\end{equation}
where ${p_{1}}^{m_{1}}\dotsm$ is the prime factorization of $n$ and
$\phi(k)$ is Euler's totient function: the number of positive
integers less than and relatively prime to $k$ (note: the prime $2$ is
included in the sum only if its exponent is greater than $1$).
\end{thm}
\begin{lem}\label{nValues}
If a Riemann surface $[P|Q]$ generated by the conformal map of
right triangles $P$ and $Q$ is discrete, then the angles of $P$ and $Q$ are given
by \eqref{angles} and either $n\leq 6$, or $n=8,10$ or $12$.
\end{lem}
\begin{proof}
Let $G$ be the isometry group of $[P|Q]$, $\psi(G)$ its derived
point group, and $\Lambda$ its lattice group.
Since conjugation by $g\in\psi(G)$ leaves $\Lambda$ invariant,
consider the automorphisms $\Phi_{g}\colon
\Lambda\to\Lambda$ given by $\Phi_{g}(\lambda)=g\lambda g^{-1}$. If
$\Lambda$ has rank $N$, then the homomorphism $\Psi\colon\psi(G)\to
\Aut(\Lambda)$, where $\Psi(g)=\Phi_{g}$, induces a representation of
$\psi(G)$ by integral $N\times N$ matrices of determinant $\pm 1$. In
fact, $\Psi$ is an isomorphism since for any of the generators $g$ of
$\psi(G)$ we can easily find a $\lambda\in \Lambda$ which is not fixed
by $\Phi_{g}$ (it suffices to look within the star $\Sigma$ or, if
$r(\pi,\pi)\notin R$, $\Sigma \Sigma^{-1}$). Since
$R\subset\psi(G)$ has order $n$, there must be an element of order $n$
in $\GL{(N,\mathbb{Z})}$. By Theorem \ref{smallestRank} we must have
$N\geq N(n)$. On the other hand, if $[P|Q]$ is discrete, then
$\Lambda$ must be discrete (as a lattice in $X\times Y$) which is
possible only if $N\leq 4$. Since $\phi(p^{m})=p^{m-1}(p-1)$, we need only
consider the values of $\phi$ for powers of small primes: $\phi(4)=2$,
$\phi(8)=4$, $\phi(3)=2$, $\phi(5)=4$ (all other primes and higher
powers yield values greater than $4$). From these facts we obtain
just the values of $n$ given in the statement of the Lemma.
\end{proof}
With Lemma \ref{nValues} and inequalities \eqref{pInequality} and
\eqref{qInequality}, the set of possible combinations of $n$, $p$ and
$q$ is already finite. Several of these combinations can be
eliminated by the following Lemma which provides a lower bound on
$\Rk(\Lambda)$ when either $p$ or $q$ is a nontrivial divisor of $n$.
\begin{lem}\label{rankBound}
Let $\Lambda$ be the lattice group of a discrete Riemann surface
$[P|Q]$ with $P$ and $Q$ defined by parameters $n$, $p$ and $q$
\eqref{angles}; then if $p>1$ and $p$ divides $n$,
$\Rk(\Lambda)\geq\phi(n/p)+\phi(n/\GCD{(n,q)})$ (and the
same statement with $p$ and $q$ interchanged).
\end{lem}
\begin{proof}
Suppose $p$ divides $n$ and $n/p=d>1$. We may assume that $p$ does not divide
$q$ since otherwise $n$, $p$ and $q$ would have $p$ as a common
divisor. Let
$r=r(2\alpha,2\beta)=r(2\pi/d,2\pi(q/n))$; then $R=\langle r\rangle$
and $R_{p}=\langle r^{d}\rangle$ is a subgroup of $R$ of order $p$. By
looking in $\Sigma$ (or $\Sigma\Sigma^{-1}$), we can find a translation
$t_{0}=t(u,v)\in\Lambda$ such that $u\neq 0$ and $v\neq 0$. Now
consider the two products of translations
\begin{equation}
\begin{align}
t_{1}&=t(u_{1},v_{1})=\prod_{s\in R_{p}}(s\:t_{0}\:s^{-1}),\\
t_{2}&=t(u_{2},v_{2})=\prod_{k=1}^{d}(s_{k}\:t_{0}\:{s_{k}}^{-1}),
\end{align}
\end{equation}
where $s_{k}$ is any element of the coset $r^{k}R_{p}$, and
$t_{2}$ depends on the particular choice of coset elements.
Evaluating the products we find
\begin{equation}
\begin{align}
u_{1}&=p\,u,\\
v_{1}&=\left(\sum_{m=1}^{p}e^{2\pi i(q m/p)}\right)v.\label{zeroSum}
\end{align}
\end{equation}
Equation \eqref{zeroSum} implies $e^{-2\pi
i(q/p)}v_{1}=v_{1}$, and, since $p$ does not divide $q$, we conclude
$v_{1}=0$. Thus $t_{1}=t(p u,0)$. Similarly, we find
\begin{equation}
u_{2}=\left(\sum_{k=1}^{d}e^{2\pi i(k/d)}\right)u=0.
\end{equation}
On the other hand, $v_{2}$ is changed just by making a
different choice for one coset element $s_{k}$ (again because $p$
does not divide $q$). Thus we can always make a choice such that
$t_{2}=t(0,v_{2})$, where $v_{2}\neq 0$.
Now $\langle\{t_{1}\}_{R}\rangle=\Lambda_{X}$ is a lattice in $X$ isomorphic
to the cyclotomic lattice $\mathbb{Z}[e^{2\pi i/d}]$, while
$\langle\{t_{2}\}_{R}\rangle=\Lambda_{Y}$ is a lattice in $Y$ isomorphic to
$\mathbb{Z}[e^{2\pi i(q/n)}]$. Since
$\Lambda\supset\Lambda_{X}\Lambda_{Y}$,
$\Rk(\Lambda)\geq\Rk(\Lambda_{X})+\Rk(\Lambda_{Y})$. The statement of
the Lemma follows from the well known formula for the rank of a
cyclotomic lattice.
\end{proof}
As an example of the application of Lemma \ref{rankBound}, consider
the case $n=8$, $p=1$ and $q=2$.
For these numbers Lemma \ref{rankBound} gives
$\Rk(\Lambda)\geq\phi(4)+\phi(8)=6$, and the corresponding Riemann
surface would not be discrete.
Together, Lemma \ref{nValues}, Lemma \ref{rankBound} and the inequalities
\eqref{pInequality} and \eqref{qInequality} limit the set of
possibilities for $n$, $p$, and $q$ to the combinations
$(n,p,q)=(6,1,2)$, $(8,1,3)$, $(10,1,3)$, $(10,1,4)$, $(12,1,5)$, and
$(12,2,3)$. All other combinations (consistent with the inequalities)
correspond to Riemann surfaces whose lattice groups have ranks
exceeding 4 and therefore cannot be discrete. We have no general
method to settle the discreteness of these remaining candidates and
therefore considered them case by case. The procedure was to obtain
generators $e_{i}$ of the finite sets $\Sigma$, or $\Sigma\Sigma^{-1}$
if $r(\pi,\pi)\notin R$, since these generate the lattice group
$\Lambda$. The computations were performed in \textit{Mathematica}
using the function \texttt{LatticeReduce}, which implements the
Lenstra-Lenstra-Lovasz (LLL) lattice reduction algorithm. With the
exception of the case $(n,p,q)=(10,1,4)$, which was found to have rank
8, all others proved to have rank 4 and nonvanishing determinant. To
these 5 cases of crystallographic Riemann surfaces, two more should be
added which simply correspond to an interchange of the spaces $X$ and
$Y$ (obtained by interchanging $p$ and $q$). Although the interchangeability of these spaces was assumed in
the derivation of inequalities \eqref{pInequality} and
\eqref{qInequality}, clearly the sections $\mathcal{A}(x)$ will detect a
difference. For example, there is no isometry which relates the
surfaces $(10,1,3)$ and $(10,3,1)$ (or their sections). On the other
hand, the use of transformation \eqref{complementaryAngles} shows that the
surface $(10,3,1)$ is isometric with the surface described by
$(n',p',q')=(5,1,2)$ (where now $p'<q'$). However, not all
interchanges of $p$ and $q$ produce a
new Riemann surface. For example, $(n,p,q)=(8,3,1)$
corresponds, by \eqref{switch1and3}, to $(n',p',q')=(8,1,3)$ (the
values before the interchange).
We summarize this discussion by our main Theorem:
\begin{thm}
Up to linear transformations, there are seven discrete Riemann surfaces
$[P|Q]$ generated by
conformal maps of right triangles $P$ and $Q$. Each of these surfaces
is crystallographic; their properties, in particular the integers
$n$, $p$ and $q$ which specify the angles of $P$ and $Q$, are given
in Table 1.
\end{thm}
\begin{proof}
The properties listed in Table 1 are simple consequences
of general results.
Since
\begin{equation}
G/\Lambda\simeq\psi(G)=\langle\sigma, R,
r(\pi,\pi)\rangle,
\end{equation}
we see that $G/\Lambda$ is isomorphic to an abstract group generated by
an element of order 2, $\tilde{\sigma}$, and an element $\tilde{r}$,
which has order $2|R|=2 n$, if $n=|R|$ is odd (and therefore
$r(\pi,\pi)\notin R$) or $n$ is even and just one of $p$ and $q$ is
odd (since then the element of order two in $R$ is either $r(\pi,0)$
or $r(0,\pi)$). Otherwise ($n$ even, $p$ and $q$ both odd),
$r(\pi,\pi)\in R$ and $\tilde{r}$ has order $|R|=n$. These generators have the
relation $\tilde{\sigma}\tilde{r}\tilde{\sigma}={\tilde{r}}^{-1}$ and
imply $G/\Lambda\simeq d_{n}$, the dihedral group of order $2 n$,
or $G/\Lambda\simeq d_{2 n}$.
To compute the genus we use the orders of the vertex groups,
\eqref{orderG1}, \eqref{orderG3} and \eqref{orderG2}, in formula \eqref{genusFormula}:
\begin{equation}
2-2g=|G/\Lambda|\left(\frac{1}{2n}+\frac{1}{2n'}-\frac{1}{4}\right).
\end{equation}
The geometry of $\Lambda$
is completely specified by the Gram matrix formed from its generators
$e_{k}$, $k=1,\ldots,4$:
\begin{equation}
(M_{kl})=e_{k}\cdot e_{l},
\end{equation}
where $\cdot$ is the standard inner product.
If we consider each $e_{k}$ as a vector in $\mathbb{R}^{4}$ and form the
$4\times 4$
matrix $E$ whose rows are $e_{k}$, then $M=E E^{\text{tr}}$ and
$|\Lambda|=|\det E|=\sqrt{\det{M}}$.
Using the LLL algorithm it was found that the generators of $\Sigma$
and $\Sigma\Sigma^{-1}$ (and hence $\Lambda$) could always be written
as, respectively
\begin{equation}
\begin{align}
e_{k}&=2\left(a e^{i 2 k\alpha},b e^{i 2 k\beta}\right)\label{basis1}\\
e_{k}&=2\left(a(e^{i 2\alpha}-1) e^{i 2 k\alpha},b (e^{i 2\beta}-1) e^{i
2 k\beta}\right).\label{basis2}
\end{align}
\end{equation}
To help identify the lattice geometry it was sometimes necessary to
define a new basis $E'=S E$, where $S\in\SL{(4,\mathbb{Z})}$. The new
Gram matrix is then given by $M'=S M S^{\text{tr}}$.
Details of this analysis, for the seven combinations of $(n,p,q)$ in Table
1, are provided in the appendix.
The only additional data needed in the density formula
\eqref{rhoFormula} is the triangle area $|P|=(1/4)\sin{2\pi(p/n)}$.
The last column of Table 1 identifies which surfaces have doubly
periodic sections. In general, since the kernel of the homomorphism
$\pi_{X}\colon\Lambda\to\pi_{X}(\Lambda)$ is given by the lattice
$\Lambda_{Y}=\Lambda\cap Y$,
\begin{equation}
\Rk{\Lambda_{Y}}=\Rk{\Lambda}-\Rk{\pi_{X}(\Lambda)}.
\end{equation}
Surfaces with doubly periodic sections have $\Rk{\Lambda_{Y}}=2$, while
$\Rk{\Lambda_{Y}}=0$
corresponds to completely quasiperiodic sections. These are the only
cases that occur, since (from
\eqref{basis1} and \eqref{basis2}), $\pi_{X}(\Lambda)\simeq\mathbb{Z}[e^{2\pi i(p/n)}]$, the
cyclotomic lattice with rank $\phi(n/\GCD{(n,p)})$.
\end{proof}
\begin{table}
\centering
{\renewcommand{\arraystretch}{2}
\begin{tabular}
{|c||c|c|c|c|c|}
\hline
$(n,p,q)$&$G/\Lambda$&$\Lambda$&$g$&$\rho$&
\begin{tabular}{c}periodic\\quasiperiodic\end{tabular}\\ \hline
$(5,1,2)$&$d_{10}$&$A_{4}$&$2$&$\frac{1}{5}\sqrt{2+\frac{2}{\sqrt{5}}}$&q\\
$(6,1,2)$&$d_{12}$&$A_{2}\times A_{2}$&$2$&$\sqrt{\frac{1}{27}}$&p\\
$(8,1,3)$&$d_{8}$&$D_{4}$&$2$&$\sqrt{\frac{1}{8}}$&q\\
$(10,1,3)$&$d_{10}$&$A_{4}$&$2$&$\frac{1}{5}\sqrt{2-\frac{2}{\sqrt{5}}}$&q\\
$(12,1,5)$&$d_{12}$&$A_{2}\times A_{2}$&$3$&$1$&q\\
$(12,2,3)$&$d_{24}$&$A_{2}\times Z_{2}$&$5$&$1$&p\\
$(12,3,4)$&$d_{24}$&$A_{2}\times Z_{2}$&$5$&$\sqrt{\frac{4}{3}}$&p\\
\hline
\end{tabular}
}
\caption{Properties of the seven discrete Riemann surfaces generated
by conformal maps of right triangles}
\end{table}
\subsection{Piecewise flat surfaces and model
sets}\label{flattenedSurfaces}
Suppose a Riemann surface $\Surf$ is deformed into another surface,
$\Tilde{\Surf}$, not necessarily representable locally by graphs of
holomorphic functions. As long as the deformation preserves the
isometry group and transversality with respect to $Y$, all the
crystallographically relevant properties of the point set $\mathcal{A}(x)$
will be maintained in the corresponding deformed point set
$\Tilde{\mathcal{A}}(x)$. Kalugin's formula \eqref{KaluginFormula}, for example, makes
this invariance explicit for the density.
When a Riemann
surface is generated by a triangular graph $P|Q$, deformations that preserve
isometry and transversality are easily specified by
the map defining the fundamental graph,
$\tilde{f}:P\to Q$. We recall that in the Riemann surface,
$\tilde{f}$ is
holomorphic and extends to a homeomorphism on the closure $\Bar{P}$. For the
deformed surface $\Tilde{\Surf}$ we
continue to use the edge group $G$, defined by the geometry of the
triangles $P$ and $Q$, to form the orbit of the fundamental graph, but
insist only that $\tilde{f}$ is a homeomorphism. While still
preserving isometry and transversality, we will even go a step further
and modify the \emph{topology} of $\Tilde{\Surf}$ by defining
$\Tilde{\Surf}$ as the orbit under $G$ of an \emph{open} graph.
Under these circumstances, when $\Tilde{\Surf}$ is a collection of
disconnected components, we are free to relax the condition that
$\tilde{f}$ is a homeomorphism. In fact, we will primarily be
interested in the case when $\tilde{f}$ is a constant map, thereby
making each
piece of $\Tilde{\Surf}$ flat.
Let $P|y$ represent the graph of the constant map, $\tilde{f}(P)=y$.
We will weigh the merits of various choices of $y\in Y$ as giving optimal
``approximations'' of the map $P|Q$ defined by the conformal map
of triangles. By choosing $y=Q_{i}$, a vertex of $Q$, we go the
furthest in restoring partial connectedness to $\Tilde{\Surf}$. This is
because the action of the vertex group $G_{i}$ on $P|Q_{i}$ generates
a flat polygon (possibly stellated) composed of $|G_{i}|$ triangles.
Thus $|G_{i}|$ surface pieces will have been ``aligned'' by this
choice of $y$.
With the
exception of the $n=5$ and $n=10$ surfaces, $|G_{1}|=|G_{3}|=2 n>|G_{2}|$,
suggesting that either one of the singular vertices is a good choice for
$y$.
There is another criterion, however, that applies uniformly to
all the surfaces and even distinguishes among the two singular
vertices.
A natural question to ask is: which value of $y\in Q$ ``occurs with
the highest frequency'' in the graph $P|Q$? To give this question a
proper probabilistic interpretation, we suppose that $x$ is
sampled uniformly in $P$ and ask for the probability that
$\|f(x)-y_{0}\|<\Delta$, where $f$ is the conformal map of triangles
and $\Delta$ is the radius of a small disk about $y_{0}\in Q$. Since
$f$ is conformal, this condition (for $\Delta\to 0$) is equivalent to
$\|x-x_{0}\|<\Delta/\|f'(x_{0})\|$, where $x_{0}=f^{-1}(y_{0})$. Thus
the probability of finding $y$ in a neighborhood of $y_{0}$ is
maximized by minimizing $\|f'(x_{0})\|$. At a singular point
$\|f'(x_{0})\|$ either vanishes or diverges,
and in our case vanishes for $x_{0}\to Q_{1}$ since we always have
$p<q$. We therefore choose $y=Q_{1}$ for all of our surfaces.
The result of flattening the seven Riemann surfaces in Table 1 by this
prescription is particularly simple. In all cases $\Poly
=G_{1}(P|Q_{1})$
is a regular $(n/p)$-gon covered $p$ times (we restore connectedness
to $\Poly$ by including edges incident to vertex 1).
This means that if each point of $\Tilde{\mathcal{A}}(x)$ is counted with
multiplicity $p$, then $\Tilde{\mathcal{A}}(x)$ and $\mathcal{A}(x)$ have the same
density. Using Lemma \ref{groups}, we have
\begin{equation}
\Tilde{\Surf}=\Lambda\cdot \Poly,\qquad\Lambda=\langle\Sigma\rangle,
\end{equation}
if $r(\pi,\pi)\in R$, otherwise,
\begin{equation}
\Tilde{\Surf}=\Lambda\:\cup\:\Lambda\, t(2 a,2 b)\,
r(\pi,\pi)\cdot \Poly, \qquad\Lambda=\langle\Sigma\Sigma^{-1}\rangle.
\end{equation}
With the exception of the surface $(5,1,2)$,
$r(\pi,\pi)\cdot \Poly=\Poly$, and we have the simpler description:
\begin{equation}
\Tilde{\Surf}=\langle\Sigma\rangle\,\Poly,
\end{equation}
since
\begin{equation}
\langle\Sigma\Sigma^{-1}\rangle\:\cup\:
\langle\Sigma\Sigma^{-1}\rangle\,t(2 a,2 b) =\langle\Sigma\rangle.
\end{equation}
Figure 1 compares the point sets $\mathcal{A}(x)$ and $\Tilde{\mathcal{A}}(x)$, given
by the flattening process just described, of the Riemann
surface $(5,1,2)$ of Table 1. Edges have been added to
$\Tilde{\mathcal{A}}(x)$ to aid in the visualization of three tile shapes: the
boat, star, and jester's cap. There is a one-to-one correspondence
between the two point sets, with most pairs having quite small
separations. In any case, we are guaranteed the separation of
corresponding points never exceeds 1, the diameter of triangle $Q$.
$\Tilde{\mathcal{A}}(x)$, the vertex set of a popular tiling model \cite{pent}, is a
Delone set. $\mathcal{A}(x)$ fails to be a Delone set because triples of
points appear with arbitrarily short separations. That always three
points coalesce in this way is a signature of the order of the branch
points of $\pi_{X}$ for this surface. $\mathcal{A}(x)$, on the other hand, has
a ``dynamical'' advantage over $\Tilde{\mathcal{A}}(x)$. Seen as atoms in a
crystal or quasicrystal, the positions $\mathcal{A}(x)$ evolve continuously (in
fact analytically) with $x$ (viewed as a parameter), while the atoms
in $\Tilde{\mathcal{A}}(x)$ experience discontinuous ``jumps'', and for the most
part never move at all. The singular loci of $x\in X$ for the two
point sets have different dimensionalities: 0 for $\mathcal{A}(x)$, 1 for
$\Tilde{\mathcal{A}}(x)$; the dynamics of $\mathcal{A}(x)$ is thus more regular also in
this sense. To emphasize this point, we note that if $\gamma(t)$ is
almost any curve in $X$, then $\mathcal{A}(\gamma(t))$ is a regular homotopy
(see Lemma \ref{uniformInX}).
\begin{figure}
\centerline{\epsfbox{fig3.eps}}
\caption{Point sets given by sections of flattened Riemann surfaces.
Lines have been added to help identify tiles.}
\end{figure}
The sets $\Tilde{\mathcal{A}}(x)$ are examples of \emph{model sets}, as defined by
Moody \cite{Moody}, and therefore belong to the larger family of \emph{Meyer
sets}. That the set $\Tilde{\mathcal{A}}(x)$ shown in Figure 1 can be
organized into a finite set of tile shapes, for example, is a general
property of model sets. Model sets obtained by flattening the other
six Riemann surfaces of Table 1 are shown in Figure 3. In the three
periodic cases, of course, even $\mathcal{A}(x)$ is a Meyer set and flattening
is not necessary if that is our only goal. We flattened these
surfaces because the sets $\Tilde{\mathcal{A}}(x)$ are then particularly
symmetric. In the three quasiperiodic cases the sets $\Tilde{\mathcal{A}}(x)$
again organize themselves into tilings that have been discussed in the
quasicrystal literature \cite{oct,dec,dodec}.
\section*{Acknowledgements}
I thank the Aspen Center for Physics, where a large part of this paper
was written. Noam Elkies provided a useful suggestion for the proof of
Lemma \ref{uniformInX}.
\newpage
|
\section{Introduction}
The biggest stumbling-block for attempts to confront theories of
cosmological structure formation with observations of galaxy
clustering is the uncertain and possibly biased relationship
between galaxies and the distribution of gravitating matter. The
idea that galaxy formation might be biased goes back to the
realization by Kaiser (1984) that the reason Abell clusters
display stronger correlations than galaxies at a given separation
is that these objects are selected to be particularly dense
concentrations of matter. As such, they are very rare events,
occurring in the tail of the distribution function of density
fluctuations. Under such conditions a ``high-peak'' bias prevails:
rare high peaks are much more strongly clustered than more typical
fluctuations (Bardeen et al. 1986). If the properties of a galaxy
(its morphology, color, luminosity) are influenced by the density
of its parent halo, for example, then differently-selected
galaxies are expected to a different bias (e.g. Dekel \& Rees
1987). Observations show that different kinds of galaxy do cluster
in different ways (e.g. Loveday et al. 1995; Hermit et al. 1996).
In {\em local bias} models, the propensity of a galaxy to form at
a point where the total (local) density of matter is $\rho$ is
taken to be some function $f(\rho)$ (Coles 1993, hereafter C93;
Fry \& Gaztanaga 1993, hereafter FG93). It is possible to place
stringent constraints on the effect this kind of bias can have on
galaxy clustering statistics without making any particular
assumption about the form of $f$. In this {\it Letter}, we
describe the results of a different approach to local bias models
that exploits new results from the theory of hierarchical
clustering in order to place stronger constraints on what a local
bias can do to galaxy clustering. We leave the technical details
to Munshi et al. (1999a,b) and Bernardeau \& Schaeffer (1999);
here we shall simply motivate and present the results and explain
their importance in a wider context.
\section{Hierarchical Clustering}
The fact that Newtonian gravity is scale-free suggests that the
$N$--point correlation functions of self-gravitating particles,
$\xi_N$, evolved into the large-fluctuation regime by the action
of gravity, should obey a scaling relation of the form
\begin{equation}
\xi_p( \lambda {\bf r}_1, \dots \lambda {\bf r}_p ) =
\lambda^{-\gamma(p-1)} \xi_p( {\bf r}_1, \dots {\bf r}_p )
\label{hierarchical}
\end{equation}
when the elements of a structure are scaled by a factor $\lambda$
(e.g. Balian \& Schaeffer 1989). Observations offer some support
for such an idea, in that the observed two-point correlation
function $\xi(r)$ of galaxies is reasonably well represented by a
power law over a large range of length scales,
\begin{equation} \xi({\bf r}) = \Big ( {r \over 5h^{-1} {\rm Mpc}}
\Big )^{-1.8} \end{equation} (Groth \& Peebles 1977; Davis \& Peebles
1977) for $r$ between, say, $100 h^{-1} {\rm kpc}$ and $
10h^{-1}~{\rm~ Mpc}$. The observed three point function, $\xi_3$, is
well-established to have a hierarchical form \begin{equation}
\xi_3({\bf x}_a, {\bf x}_b, {\bf x}_c) = Q[\xi_{ab}\xi_{bc} +
\xi_{ac}\xi_{ab} + \xi_{ac}\xi_{bc}], \end{equation} where
$\xi_{ab}=\xi ({\bf x}_a, {\bf x}_b)$, etc, and $Q$ is a constant
(Davis \& Peebles 1977; Groth \& Peebles 1977). The four-point
correlation function can be expressed as a combination of graphs with
two different topologies -- ``snake'' and ``star'' -- with
corresponding (constant) amplitudes $R_a$ and $R_b$ respectively:
\begin{equation} \xi_4({\bf x}_a, {\bf x}_b, {\bf x}_c, {\bf x}_d) =
R_a[\xi_{ab}\xi_{bc}\xi_{cd} + \dots ({\rm 12~terms})] +
R_b[\xi_{ab}\xi_{ac}\xi_{ad} + \dots ({\rm 4~terms})] \end{equation}
(e.g. Fry \& Peebles 1978; Fry 1984).
It is natural to guess that all p-point correlation functions can be
expressed as a sum over all possible p-tree graphs with (in general)
different amplitudes $Q_{p,\alpha}$ for each tree diagram topology
$\alpha$. If it is further assumed that there is no dependence of
these amplitudes upon the shape of the diagram, rather than its
topology, the correlation functions should obey the following
relation: \begin{equation} \xi_p( {\bf r}_1, \dots {\bf r}_p ) =
\sum_{\alpha, ~p-{\rm trees}} Q_{p,\alpha} \sum_{\rm labellings}
\prod_{\rm edges}^{(p-1)} \xi({\bf r}_i, {\bf r}_j). \end{equation} To
go further it is necessary to find a way of calculating $Q_p$. One
possibility, which appears remarkably successful when compared with
numerical experiments (Munshi et al. 1999b; Bernardeau \& Schaeffer
1999), is to calculate the amplitude for a given graph by simply
assigning a weight to each vertex of the diagram $\nu_n$, where $n$ is
the order of the vertex (the number of lines that come out of it),
regardless of the topology of the diagram in which it occurs. In this case
\begin{equation} Q_{p,\alpha}=\prod_{\rm vertices} \nu_n.
\end{equation} Averages of higher-order correlation functions can be
defined as \begin{equation} \bar{\xi}_p=\frac{1}{V^p} \int \ldots \int
\xi_p({\bf r}_1\ldots {\bf r_p}) dV_1\ldots dV_p. \label{eq:xibar}
\end{equation} Higher-order statistical properties of galaxy counts
are often described in terms of the scaling parameters $S_p$
constructed from the $\bar{\xi}_p$ via \begin{equation}
S_p=\frac{\bar{\xi}_p}{\bar{\xi}_2^{p-1}}.\label{eq:s_p}
\end{equation} It is a consequence of the particular class of hierarchical clustering models
defined by equations (5) \& (6) that {\it all} the $S_p$ should be
constant, independent of scale.
\section{Local Bias}
Using a generating function technique, originally developed by
Bernardeau \& Schaeffer (1992), it is possible to derive a series
expansion for the $m$-point count probability distribution
function of the objects $P_m(N_1,....N_m)$ (the joint probability
of finding $N_i$ objects in the $i$-th cell, where $i$ runs from $1$ to $m$)
from the $\nu_n$. The
hierarchical model outlined above is therefore statistically
complete. In principle, therefore, any statistical property of the
evolved distribution of matter can be calculated just as it can
for a Gaussian random field. This allows us to extend various
results concerning the effects of biasing on the initial
conditions into the nonlinear regime in a more elegant way than is
possible using other approaches to hierarchical clustering.
For example, let us consider the joint occupation probability
$P_2(N_1, N_2)$ for two cells to contain $N_1$ and $N_2$ particles
respectively. Using the generating-function approach outlined
above, it is quite easy to show that, at lowest order,
\begin{equation}
P_2(N_1,N_2)=P_1(N_1)P_1(N_2)+P_1(N_1)b(N_1)P_1(N_2)b(N_2)\xi_{12}(r_{12
}), \label{eq:2pt} \end{equation}
where the $P_1(N_i)$ are the
individual count probabilities of each volume separately and
$\xi_{12}$ is the underlying mass correlation function. The
function $b(N_i)$ we have introduced in (9) depends on the set of
$\nu_n$ appearing in equation (6); its precise form does not
matter in this context, but the structure of equation (9) is very
useful. We can use (9) to define
\begin{equation}
1+\xi_{N_1 N_2}(r_{12}) \equiv \frac{P(N_1, N_2)}{P_1(N_1)P_1(N_2)},
\end{equation}
where $\xi_{N_1 N_2}(r_{12})$ is the cross-correlation of
``cells'' of occupancy $N_1$ and $N_2$ respectively. From this definition
and equation (9) it follows that
\begin{equation} \xi_{N_1 N_2}(r)=b(N_1)b(N_2)\xi_{12}(r);
\label{eq:2bias}
\end{equation}
we have dropped the subscripts on $r$ for clarity from now on.
>From (11) we can obtain
\begin{equation}
b_{N}^2(r_{12}) = \frac{\xi_{N N}(r)}{\xi_{12}(r)}
\end{equation}
for the special case where $N_1=N_2=N$ which can be identified with the usual
definition of the bias parameter associated with the correlations among a
given set of objects $\xi_{\rm obj}(r)=b^{2}_{\rm obj} \xi_{\rm mass}(r)$.
Moreover, note that at this order (which is valid on large
scales), the correlation bias defined by equation (11)
factorizes into contributions $b_{N_i}$ from each individual cell (Bernardeau 1996;
Munshi et al. 1999b).
Coles (1993) proved, under weak conditions on the form of a local
bias $f(\rho)$ as discussed in the introduction, that the
large-scale biased correlation function would generally have a
leading order term proportional to $\xi_{12}(r_{12})$. In other
words, one cannot change the large-scale slope of the correlation
function of locally-biased galaxies with respect to that of the
mass. This ``theorem'' was proved for bias applied to Gaussian
fluctuations only and therefore does not obviously apply to galaxy
clustering, since even on large scales deviations from Gaussian
behaviour are significant. It also has a more minor loophole,
which is that for certain peculiar forms of $f$ the leading order
term is proportional to $\xi_{12}^2$, which falls off more sharply
than $\xi_{12}$ on large scales.
Steps towards the plugging of this gap began with FG93 who used an
expansion of $f$ in powers of $\delta$ and weakly non-linear
(perturbative) calculations of $\xi_{12}(r)$ to explore the
statistical consequences of biasing in more realistic (i.e.
non-Gaussian) fields. Based largely on these arguments, Scherrer
\& Weinberg (1998), hereafter SW98, confirmed the validity of the C93
result in the non-linear regime, and also showed explicitly that
non-linear evolution always guarantees the existence of a linear
leading-order term regardless of $f$, thus plugging the small gap
in the original C93 argument. These works have
a similar motivation to ours, and also exploit hierarchical
scaling arguments of the type discussed in \S 2 {\em en route} to
their conclusions. What is different about the approach we have
used in this paper is that the somewhat cumbersome simultaneous
expansion of $f$
and $\xi_{12}$ used by SW98 is not required in our calculation: we use the
generating functions to proceed directly to the joint probability
(9), while SW98 have to perform a complicated sum over moments of a
bivariate distribution. The factorization of the probability
distribution (9) is also a stronger result than that presented by SW98,
in that it leads almost trivially to the C93 ``theorem'' but also
generalizes to higher-order correlations than the two-point case
under discussion here.
Note that the density of a cell of given volume is simply
proportional to its occupation number $N$. The factorizability of
the dependence of $\xi_{N_1 N_2}(r_{12})$ upon $b(N_1)$ and
$b(N_2)$ in (11) means that applying a local bias $f(\rho)$ boils
down to applying some bias function $F(N)=f[b(N)]$ to each cell.
Integrating over all $N$ thus leads directly to the same
conclusion as C93, i.e. that the large-scale $\xi(r)$ of
locally-biased objects is proportional to the underlying matter
correlation function. This has also been confirmed by numerically
using $N$-body experiments (Mann et al. 1998; Narayanan et al.
1998).
\section{Halo Bias}
In hierarchical models, galaxy formation involves the following
three stages:
\begin{enumerate}
\item the formation of a dark matter halo;
\item the settling of gas into the halo potential;
\item the cooling and fragmentation of this gas into stars.
\end{enumerate}
Rather than attempting to model these stages in one go by a simple
function $f$ of the underlying density field it is interesting to see
how each of these selections might influence the resulting statistical
properties. Bardeen et al. (1986), inspired by Kaiser (1984),
pioneered this approach by calculating detailed statistical properties
of high-density regions in Gaussian fluctuations fields. Mo \& White
(1996) and Mo et al. (1997) went further along this road by using an
extension of the Press-Schechter (1974) theory to calculate the
correlation bias of halos, thus making an attempt to correct for the
dynamical evolution absent in the Bardeen et al. approach. The
extended Press-Schechter approach seem to be in good agreement with
numerical simulations, except for small halo masses (Jing 1998). It
forms the basis of many models for halo bias in the subsequent
literature (e.g. Moscardini et al. 1998; Tegmark \& Peebles 1998).
The hierarchical models furnish an elegant extension of this work that
incorporates both density-selection and non-linear dynamics in an
alternative to the Mo \& White (1996) approach. We exploit the properties of
equation (\ref{eq:2pt}) to construct the correlation function of
volumes where the occupation number exceeds some critical value. For
very high occupations these volumes should be in good correspondence
with collapsed objects.
The way of proceeding is to construct a tree graph for all the
points in both volumes. One then has to re-partition the elements of
this graph into internal lines (representing the correlations within
each cell) and external lines (representing inter-cell correlations).
Using this approach the distribution of high-density regions in a
field whose correlations are given by eq. (5) can be shown to be
itself described by a hierarchical model, but one in which the vertex
weights, say $M_n$, are different from the underlying weights $\nu_n$
(Bernardeau \& Schaeffer 1992, 1999; Munshi et al. 1999a,b).
First note that a density threshold is in fact a form of local
bias, so the effects of halo bias are governed by the same
strictures as described in the previous section. Many of the other
statistical properties of the distribution of dense regions can be
reduced to a dependence on a scaling parameter $x$, where
\begin{equation} x=N/N_c. \end{equation} In this definition
$N_c=\bar{N}\bar{\xi}_2$, where $\bar{N}$ is the mean number of
objects in the cell and $\bar{\xi}_2$ is defined by eq.
(\ref{eq:xibar}) with $p=2$. The scaling parameters $S_p$ can be
calculated as functions of $x$, but are generally rather messy (Munshi
et al. 1999a). The most interesting limit when $x\gg 1$ is, however,
rather simple. This is because the vertex weights describing the
distribution of halos depend only on the $\nu_n$ and this dependence
cancels in the ratio (\ref{eq:s_p}). In this regime, \begin{equation}
S_p=p^{(p-2)} \end{equation} for all possible hierarchical models. The
reader is referred to Munshi et al. (1999a) for details. This result
is also obtained in the corresponding limit for very massive halos by
Mo et al. (1997). The agreement between these two very different
calculations supports the inference that this is a robust prediction
for the bias inherent in dense regions of a distribution of objects
undergoing gravity-driven hierarchical clustering.
\section{Discussion and Conclusions}
The main purpose of this {\em Letter} has been to advertise the
importance of recent developments in the theory of
gravitational-driven hierarchical clustering. The model described in
equations (5) \& (6) provides a statistically-complete prescription
for a density field that has undergone hierarchical clustering. This
allows us to improve considerably upon biasing arguments based on an
underlying Gaussian field.
These methods allow a simpler proof of the result obtained by SW98
that strong non-linear evolution does not invalidate the local
bias theorem of C93. They also imply that the effect of bias on a
hierarchical density field is factorizable. A special case of this
is the bias induced by selecting regions above a density
threshold. The separability of bias predicted in this kind of
model could be put to the test if a population of objects could be
found whose observed characteristics (luminosity, morphology,
etc.) were known to be in one-to-one correspondence with the halo
mass. Likewise, the generic prediction of higher-order correlation
behaviour described by the behaviour of $S_p$ in equation
(\ref{eq:s_p}) can also be used to construct a test of this
particular form of bias.
Referring to the three stages of galaxy formation described in \S 4,
analytic theory has now developed to the point where it is fairly
convincing on (1) the formation of halos. Numerical experiments are
beginning now to handle (2) the behaviour of the gas component
(Blanton et al. 1998, 1999). But it is unlikely that much will be
learned about (3) by theoretical arguments in the near future as the
physics involved is poorly understood (though see Benson et al. 1999).
Arguments have already been advanced to suggest that bias might not be
a deterministic function of $\rho$, perhaps because of stochastic or
other hidden effects (Dekel \& Lahav 1998; Tegmark \& Bromley 1999).
It also remains possible that large-scale non-local bias might be
induced by environmental effects (Babul \& White 1991; Bower et al.
1993).
Before adopting these more complex models, however, it is
important to exclude the simplest ones, or at least deal with that
part of the bias that is attributable to known physics. At this stage
this means that the `minimal' bias model should be that based on the
selection of dark matter halos. Establishing the extent to which
observed galaxy biases can be explained in this minimal way is clearly
an important task.
\acknowledgements
ALM acknowledges the support of the NSF-EPSCoR program, and DM
acknowledges the receipt of an Alexander von Humboldt research
fellowship. We thank Bob Scherrer for interesting discussions;
Francis Bernardaeau and Richard Schaeffer for helpful comments; and the
referee, David Weinberg, for extremely helpful criticisms of an earlier
version of this paper.
|
\section{Introduction}
To compute a Boolean function $f : \{ 0, 1\}^N \rightarrow \{ -1, 1\}$
in the black-box model, the only way the computer can access the input is
to ask an oracle questions like: what is $x_i$.
The complexity measurement is the number of times the computer asks.
For example, to compute $f ( x_0, x_1, \ldots, x_{N-1} ) = x_0 \vee x_1 \vee \ldots \vee
x_{N-1}$, any classical computer needs to query $\Omega ( N )$ times, while surprisingly,
there exists a quantum algorithm that queries only $O ( \sqrt{N} )$ times \cite{Gr96}.
The speedup of quantum computers can even be exponential if promise problems are
considered \cite{DJ92, Si94}. We study quantum lower bounds
for error-bounded computation of total Boolean functions only.
Grover's algorithm was shown to be optimal by
\cite{BBHT96, Za97}, and
tight bounds for all symmetric Boolean functions,
in particular for some
familiar ones like $PARITY$, $MAJORITY$,
are shown in \cite{BBCMW98} by
polynomial method. The latter relates
two characterizations of Boolean function complexity, the lowest degree
of approximation polynomials and the block sensitivity,
to lower bounds of quantum black-box complexity. A
Boolean
function $f$ on $N$ Boolean variables
can be uniquely
represented as a multi-linear
real polynomial that takes $f(x)$ for $x \in \{ 0, 1\}^N$.
Let $\tilde{f}$ be a real polynomial that approximates
$f$ and has the least degree.
The block sensitivity of $f$, $BS_f$, is the maximum (over all possible
inputs) number of non-intersecting blocks of
variables in an input such that flipping all variables in one
block flips the function value.
\begin{theorem} \cite{BBCMW98}
\label{deg}
To compute a Boolean function $f ( x_0, x_1, \ldots, x_{N-1})$ with error probability
$\epsilon$,,
any quantum computer needs to query at least $\frac{1}{2}deg ( \tilde{f} )$ times.
\end{theorem}
\begin{theorem}\cite{BBCMW98}
To compute a Boolean function $f$ on $N$ variables with error probability $\frac{1}{3}$,
any quantum computer
needs to ask at least $\frac{1}{4}\sqrt{BS_f}$ times.
\end{theorem}
In the latter theorem, replace $BS_f$ by the sensibility $S_f$,
we get the best known lower bound by of sensitivity.
$S_f$ is just the maximum (over all possible inputs)
number of bits such that flipping
each flips the function value.
Just like the lowest approximating degree,
block sensitivity, and sensitivity,
the influence of variables ($Inf_i(f)$)
and average
sensitivity ($\bar{S}_f$), introduced by \cite{KKL88},
are yet other important characterizations of Boolean function complexity.
One can show that $\sum_i In_i(f) = \bar{S}_f$.
Our main result provides the first known quantum black-box complexity lower
bound by influence of variables ( or equivalently, by average sensitivity):
\begin{theorem}\label{main}
To compute a Boolean function $f : \{ 0, 1\}^N \rightarrow \{ -1, 1\}$ with error
probability $\epsilon$, any quantum algorithm needs to query at least
$\frac{1 - 2\sqrt{\epsilon}}{2}\rho_f N =
\frac{1 - 2\sqrt{\epsilon}}{2} \bar{S}_f$
times, where $\rho_f$ is the average influence of variables of $f$.
\end{theorem}
Actually we prove a stronger results:
\begin{theorem}\label{general}
To compute Boolean function $f : \{ 0, 1\}^N \rightarrow \{ -1, 1\}$ with error
probability $\epsilon$, any quantum algorithm needs to query at least
$\frac{1}{2} \cdot \biggl[ 1 - \sqrt[k]{ \frac{ 1 + 2\sqrt{\epsilon}}{2}
+ \frac{1 - 2\sqrt{\epsilon}}{2} \sum_s \hat{f}_s^2 ( 1 - 2 \lambda_s )^k
} \biggr] \cdot N$ times, for any positive
odd integer $k$,
and $\hat{f}_s = E_{x} \Bigl[ f(x) ( -1 )^{ x \cdot s} \Bigr]$.
\end{theorem}
The lowest degree of approximation polynomials for symmetric functions
has a tight characterization \cite{Pa92}. However, for asymmetric
functions, the best general lower bound known is:
\begin{theorem}\cite{NS92} \label{appdeg} For any Boolean function $f$ and its
approximation $\tilde{f}$,
$deg(\tilde{f}) \ge \sqrt{{BS_f}/6}.$
\end{theorem}
Our lower bound by influence can be better than this bound
for some functions:
\begin{theorem}\label{degree}
For any polynomial $\tilde{f}$
that approximates $f$ with error probability $\epsilon$,
$deg(\tilde{f}) \ge \frac{1}{4} ( 1 - \frac{3\epsilon}{1+\epsilon})^2 \rho_f N$.
\end{theorem}
Note that, Theorem \ref{degree} and Theorem \ref{deg}
together imply a lower bound of quantum black-box complexity
that is asymptotically the same as in Theorem \ref{main}.
However, the former gives a better constant.
Now we turn to our proof ideas.
For any oracle $x$, let $\phi(x)$ be the state of the quantum computer after
$T$ times of queries.
If $f(x) \ne f(y)$, then, in order to distinguish $x$
and $y$, $\|\phi(x) -\phi(y)\|^2$ must be large, meaning,
lower bounded by $2 - 4\sqrt{\epsilon}$. If we pick $x$ uniformly random from
$\{ 0, 1\}^N$, and pick a random bit $x_i$, then flip $x_i$ to
get $y$, then we make a hard time for the quantum algorithm: on one hand,
$E = E_{(x, y)}\Bigl[\|\phi(x) -\phi(y)\|^2\Bigr]$ is large (meaning, lower bounded by
$(2 - 4\sqrt{\epsilon}) \rho_f$), on the other hand, this expected value is
upper-bounded by $\frac{4T}{N}$, which then gives us Theorem \ref{main}.
Here is how we get the upper bound of $E$ by $\frac{4T}{N}$.
The key observation of \cite{BBCMW98} is that
with oracle $x$, after $T$ queries,
the state of the quantum computer can be expressed as:
$$\label{classical}\phi( x ) = \sum_{c} p_c(x) |c\rangle,$$
where $\{c\}$ is a set of orthonormal basis, and $p_c$'s are polynomials
over $x_i$'s of degree at most $T$.
Since $p_c$'s
can be represented in Fourier
basis $\{ L_s \}$, where $L_s(x) = ( -1)^{ s \cdot x}$,
we have:
$$\label{equation:rep}\phi( x ) =
\sum_{ s \in \{ 0, 1\}^N, |s| \le T} \hat{\phi}_s ( -1 )^{ x \cdot s},$$
where $\hat{\phi}_s$ are constant vectors depending on $s$ only.
The number of queries comes in in the maximum possible weight of $s$ with
$\hat{\phi}_s \ne 0$,
since $|s|$ is the degree of polynomial $ ( -1 )^{ x \cdot s}$.
This representation of $\phi$ is implicit in \cite{FGGS98},
and has a
natural interpretation that will be described later.
We can now write down $E$ explicitly
in terms of $\hat{\phi}_s$, and then get the desired upper bound by $\frac{4T}{N}$.
For the lower bound of $deg(\tilde{f})$ by average sensitivity, we use the same idea.
We pick the same random pair $( x, y)$, and examine the expected value of
$\| \tilde{f}(x) - \tilde{f}(y) \|^2$ and bound this expectation from both sides --
the upper-bound side by $deg(\tilde{f})$, and the lower-bound side by $\bar{S_f}$.
Section \ref{background} reviews standard notions and facts about Fourier transform and
influence of variables.
Section \ref{blackbox} looks at quantum black-box computation from the viewpoint
of Fourier transform, and section \ref{mainresults}
provides the proofs, which
is followed by open problems.
Some miscellaneous notations: for any string $s \in \{ 0, 1\}^N$, $|s|$ equals to
the number of $1$'s in $s$, and $\lambda_s = |s|/N$, $e_i \in \{ 0, 1\}^N$ is the string with
the only $1$ in the $i$'th position, for $0 \le i \le {N-1}$. All
probability distributions are uniformly random over the corresponding domain. $+$
usually is bitwise $XOR$, and $\cdot$ is usually inner product.
\section{Fourier transform of Boolean functions and influence of variables}
\label{background}
For any $s \in \{ 0, 1\}^N$, let $L_s : \{ 0, 1\}^N \rightarrow \{ -1, 1\}$ be
$L_s ( x ) = ( -1 )^{ s \cdot x }$ for any $x \in \{ 0, 1\}^N$. $\{ L_s \}$ form
a basis for all functions mapping $\{ 0, 1\}^N$ to ${C}$.
\begin{lemma}\label{fourier} (Well known)Any $f : \{ 0, 1\}^N \rightarrow {C}$ can
be uniquely represented as:
$f = \sum_s \hat{f}_s L_s$, where
$\hat{f}_s = E_x \bigl[ f(x)L_s(x) \bigr]$,
and $max_{ s, \hat{f}_s \ne 0} \{ |s| \} = deg ( f )$, and
$\sum_s \hat{f}_s \hat{f}_s^* = E_s [ f(s) f^*(s)]$.
\end{lemma}
\begin{definition}\cite{KKL88} The influence of variable $x_i$ on Boolean
function $f(x_0, x_1, \ldots, x_{N-1})$ is given by:
$Inf_i( f ) = Pr_{x} \Bigl[ f(x) \ne f( x+ e_i) \Bigr],$
the average influence (of variables) is $ \rho_f = E_i \Bigl[ Inf_i ( f ) \Bigr].$
\end{definition}
\begin{lemma}\label{average}\cite{CG88, KKL88} For any Boolean function $f$ on $N$ variables, if
$f = \sum_s \hat{f}_s L_s$, then
$ \rho_f = \sum_s \hat{f}_s^2 \lambda_s.$
\end{lemma}
\begin{definition}The sensitivity of $f$ on input $x$ is given by
$S_f ( x ) = | \{ i : f( x ) \ne f( x + e_i ) \} |,$
and the average sensitivity is $\bar{S}_f = E_x \Bigl[ S_f ( x ) \Bigr]$.
\end{definition}
It's easy to check:
\begin{lemma} $\bar{S}_f = \rho_f N$.
\end{lemma}
\section{Quantum black-box algorithm from a viewpoint of Fourier transform}
\label{blackbox}
To compute a Boolean function $f$ on $N$ variables,
a quantum computer uses four sets
of registers $|i\rangle_I |a\rangle_A |w\rangle_W |r\rangle_R$: $I$
(index)
has ${\lg{N}}$
bits, $A$ (answer) and $R$(result) are of one bit each, and $W$ (working)
of a fixed number of bits. The quantum computer is
working in the Hilbert space spanned by the base vectors
$\{|i\rangle_I |a\rangle_A |w\rangle_W |r\rangle_R\}$.An
algorithm $\phi$ that computes $f$ in $T$ queries with error probability at most
$\epsilon$ can be represented as a sequence
of $2T + 1$ unitary operators on the $H$:
$$\phi: U_0 \rightarrow O \rightarrow U_1 \rightarrow O \rightarrow \ldots
\rightarrow O \rightarrow U_T,$$
where $U_i$'s are arbitrary unitary transformations defined by the algorithm,
and $O$'s are the query gates:
$$O : | i \rangle_I | r \rangle_A |w\rangle_W |r\rangle_R
\rightarrow | i \rangle_I | r + x_i \rangle_A |w\rangle_W |r\rangle_R.$$
The quantum computer starts from the constant vector $| \vec{0} \rangle_{IAWR}$, apply
the sequence of unitary transformations, and then observing the $R$ register yields
$f(x)$ with error probability smaller than $\epsilon$.
Essentially, $\phi$ defines a function $\{ 0, 1\}^N
\rightarrow \{ \vec{v} \in H, \| \vec{v} \|^2 = 1 \}$.
A key observation in \cite{BBCMW98} is,
\begin{lemma}\cite{BBCMW98} On oracle $x$, the final state of the quantum computer can
be written as:
$$\phi ( x ) = \sum_c p_c(x) | c \rangle,$$
where $c$ is taking over all possible configurations of all the registers, and
$p_c( x )$ is a polynomial mapping $\{ 0, 1\}^N$ to ${C}$ with $deg ( p_c ) \le T$.
\end{lemma}
Therefore, we can think of $\phi$ as a polynomial with coefficients in $H$. Then, any quantum
black-box algorithm defines such a polynomial of degree at most $T$.
Now we introduce the Fourier transform viewpoint of quantum black-box algorithms.
\begin{lemma} Let $\phi$ be defined by a quantum black-box algorithm,
then $\phi$ can be represented as
$$\phi = \sum_{s, |s| \le T} \hat{\phi}_s L_s,$$
where
$\hat{\phi}_s = E_x \Bigl[ \phi(x) L_s(x) \Bigr]$, and
$\sum_s \langle\hat{\phi}_s | \hat{\phi}_s\rangle = 1$.
\end{lemma}
Here is a natural interpretation of $\hat{\phi}_s$.
Let's shift the basis for $H$ from $\{ | c \rangle \}$ to that spanned by eigenvectors of the
oracle gates. Denote $F| j \rangle_A$ by $|\zeta_j\rangle_A$, where $F$ is the Hadamard
transform on $Z^m_2$. The following lemma is easy to check.
\begin{lemma}
For a general oracle gate considered in literature, namely, the oracle $x$ is regarded
as a function $\{ 0, 1\}^n \rightarrow \{ 0, 1\}^m$, and the oracle gate behaves in this way:
$$\label{oldway}O_x | i \rangle_I | j \rangle_A
= | i \rangle_I | j + x(i) \rangle_A,$$
Then
$$\label{newway}O_x | i \rangle | \zeta_j\rangle | w \rangle
| r \rangle = ( -1 )^{ x \cdot e^i_j} | i \rangle | \zeta_j\rangle | w \rangle
| r \rangle,$$
where $e^i_j\in \{ 0, 1\}^{2^nm}$ and has $j$ in the $i$th block, and $0$ in
other blocks (each block's length is $m$).
\end{lemma}
Now let's see what's going on in a quantum black-box computation, following
directions of $\{ | i \rangle |\zeta_j\rangle\}$:
We start from the constant vector $|\vec 0\rangle$, make a unitary transformation $U_0$, then
we project into the subspace spanned by some $| i_1 \rangle |\zeta_{j_1}\rangle$, then we make
a query, resulting a sign-flip $( -1 )^{ x \cdot e^{i_1}_{j_1} }$, then we continue our
walk in the same manner: make the second unitary transformation $U_1$, then go into
subspace of some $| i_2 \rangle |\zeta_{j_2}\rangle$, ask the oracle to flip our sign by
$( -1 )^{ x \cdot e^{i_2}_{j_2} }$, and so on. One good thing about this walk is that every time
we project to a subspace or make a query, for different oracle, the only difference is the
sign, and this sign is given by
$$ L_x ( e^{i_1}_{j_1} + e^{i_2}_{j_2} + \cdots + e^{i_t}_{j_t}),$$
if we've gone through the path
$$ p = | i_1 \rangle |\zeta_{j_1}\rangle \rightarrow | i_2 \rangle |\zeta_{j_2}\rangle
\cdots \rightarrow | i_t \rangle |\zeta_{j_t}\rangle.$$
We use $\alpha_p$ to denote the vector reached by path $p$ with
oracle $0$, and $\pi_p = e^{i_1}_{j_1} + e^{i_2}_{j_2} + \cdots + e^{i_t}_{j_t}$, then it's easy
to check:
\begin{lemma}
$\hat{\phi}_s = \sum_{ p, \pi_p = s} \alpha_p.$
\end{lemma}
\section{Main result}
\label{mainresults}
\subsection{Lower bound of quantum black-box complexity}
We're given a Boolean function $f : \{ 0, 1\}^N \rightarrow \{ -1, 1\}$, which
can be represented as $f = \sum_{s \in \{ 0, 1\}^N} \hat{f}_s L_s$.
Let $\phi : \{ 0, 1\}^N \rightarrow H$ be defined by a quantum black-box algorithm that uses
$T$ queries to compute $f$ with error probability bounded by $\epsilon$,
and for any input
$x \in \{ 0, 1\}^N$, $\phi ( x ) = \sum_{a: |a| \le T} \hat{\phi}_a L_a ( x )$.
Now we take $x$ uniformly random from $\{ 0, 1\}^N$, and take $i_1, i_2, \dots, i_k$ uniformly
and independently random from $\{ 0, 1, \cdots, N-1\}$, for positive
odd integer $k$. Let
$\vec{i}= e_{i_1} + e_{i_2} + \cdots + e_{i_k}$.
We would like to bound
$$E = E_{ x, i_1, i_2, \cdots, i_k } \biggl[
\Arrowvert \phi ( x ) - \phi ( x + \vec{i} ) \Arrowvert^2
\bigg]$$
from both side -- the lower bound is in in terms of the average sensitivity,
and the upper bound by $T$, and then get the lower bound for $T$.
To bound $E$ from below, first note that when a pair of input $x, y \in \{ 0, 1\}^N$
have different function value, the corresponding vector $\phi ( x)$
and $\phi ( y )$ are far apart:
\begin{fact}
If $f ( x ) \ne f ( y )$, $ \| \phi ( x ) - \phi ( y ) \|^2 \ge 2 -
4\sqrt{\epsilon}$.
\end{fact}
Therefore the fraction such that $f ( x ) \ne f ( x + \vec{i} )$
bounds $E$ from below. The fraction is given by the following lemma:
\begin{lemma}
$Pr_{x, i_1, i_2, \cdots, i_k} \big[ f ( x ) \ne f ( x + \vec{i} ) \big]
=
\frac{1}{2} - \frac{1}{2} \sum_s \hat{f}_s^2 ( 1 - 2 \lambda_s )^k$.
\end{lemma}
\begin{proof}
\begin{eqnarray}
&&Pr_{x, i_1, i_2, \cdots, i_k} \big[ f ( x ) \ne f ( x + \vec{i} ) \big]
\\
& = & \frac{1}{2} E \big[ 1 - f ( x ) f ( x + \vec{i} ) \big]\\
& = & \frac{1}{2} - \frac{1}{2} \sum_{s_1, s_2} \hat{f}_{s_1} \hat{f}_{s_2} E \big[
L_{s_1} ( x ) L_{s_2} ( x + \vec{i} ) \big]\\
& = & \frac{1}{2} - \frac{1}{2} \sum_{s_1, s_2} \hat{f}_{s_1} \hat{f}_{s_2}
E \big[ ( -1 )^{ x \cdot ( s_1 + s_2 ) + \vec{i} \cdot s_2} \big]\\
& = & \frac{1}{2} - \frac{1}{2} \sum_{s} \hat{f}_s^2 \Big( E_{i\in\{ 0, \ldots, N-1\}}
\big[ ( -1 )^ {e_i \cdot s} \big] \Big)^k\\
& = & \frac{1}{2} - \frac{1}{2} \sum_s \hat{f}_s^2 ( 1 - 2 \lambda_s )^k
\end{eqnarray}
\end{proof}
When $k = 1$, this probability is exactly the average sensitivity.
Putting the fact and the lemma together we have:
\begin{lemma}
$E \ge ( 2 - 4\sqrt{\epsilon} ) \big( \frac{1}{2} - \frac{1}{2} \sum_s \hat{f}_s^2 ( 1 - 2 \lambda_s )^k
\big)$
\end{lemma}
Now let's bound $E$ from above, and we'll see how powerful the Fourier
representation of $\phi$ is.
\begin{lemma}
$ E \le 2 - 2( 1 - \lambda_T )^k.$
\end{lemma}
\begin{proof} ( $|a| \le T$.)
\begin{eqnarray}
E & = & E \biggl[ \Arrowvert \sum_{ a, | a | \le T} L_a ( x ) [ 1 - L_a ( \vec{i} ) ]
\hat{\phi}_a \Arrowvert^2 \biggr]\\
& = & E \biggl[ \sum_{ a, b} L_{ a + b} ( x ) [ 1 - L_a ( \vec{i} ) ] [ 1 - L_b ( \vec{i} ) ]
\langle \hat{\phi}_a | \hat{\phi}_b \rangle \biggr]\\
& = & \sum_a E_{ i_1, i_2, \ldots, i_k} \biggl[ ( 1 - L_a ( \vec{i} ) )^2 \biggr]
\| \hat{\phi}_a \|^2\\
& = & \sum_a w_a \| \hat{\phi}_a \|^2
\end{eqnarray}
Where $w_a = E_{ i_1, i_2, \ldots, i_k} \bigl[ ( 1 - L_a ( \vec{i} ) )^2 \bigr]$. Now
let's bound $w_a$ by $\lambda_T$, and we think of $a \in \{ -1, 1\}^N$, with
the standard interpretation of $-1$ as $1$ in the original $a$, and $1$ as $0$.
\begin{eqnarray}
w_a & = & 4 Pr \bigl[ a_{i_1} \cdot a_{i_2} \cdots a_{i_k} = -1 \bigr]\\
& = & 4 E \bigl[ \frac{ 1 - a_{i_1} \cdot a_{i_2} \cdots a_{i_k} }{2} \bigr]\\
& = & 2 - 2 ( E_i [ a_i ] )^k\\
& = & 2 - 2 ( 1 - 2 \lambda_a)^k\\
& \le & 2 - 2 ( 1 - 2 \lambda_T)^k
\end{eqnarray}
Note that $\sum_a \| \hat{\phi}_a \|^2 = 1$, therefore,
$$ E \le 2 - 2 ( 1 - 2 \lambda_T )^k.$$
\end{proof}
Now put the above two lemmas together, solve the inequality, then we get
Theorem \ref{general}. Theorem \ref{main} is obtained by Theorem \ref{general}
and Lemma \ref{average}.
At the time of writing, we don't know if we can make use of the theorem
for $k>1$. But for $k=1$, any lower bound for sum of influence
implies lower bound for quantum black-box complexity. Here are two examples:
When $f$ is a random function, $\rho_f = \frac{1}{2}$, therefore we have:
\begin{corollary} To compute
a random function $f$ with error probability $\epsilon$,
the expected number of queries of
any quantum computer is at least $\frac{1 - 2\sqrt{\epsilon}}{4} N$.
\end{corollary}
This matches \cite{Da98} up to a constant factor,
though slightly worse than the lower
bound of \cite{Am98}. Since the $PARITY$ function has average sensitivity $1$, we
have
\begin{corollary} To compute
$PARITY$ with error probability $\epsilon$,
any quantum computer needs to query at least
$\frac{1 - 2\sqrt{\epsilon}}{2} N$ times.
\end{corollary}
This is also a known result \cite{BBCMW98}, \cite{FGGS98}.
\subsection{Degree lower bound of approximating polynomials}
\begin{definition}\cite{NS92}
Let $f : \{ 0, 1\}^N \rightarrow \{0, 1\}$ be a Boolean function, we say
$\tilde{f} : \{ 0, 1\}^N \rightarrow R$ approximates $f$ with
error probability $\epsilon$ if for any $x \in \{ 0, 1\}^N$,
$ | f(x) - \tilde{f}(x) | \le \epsilon.$
\end{definition}
Let $\tilde{f} = \sum_s \hat{\tilde{f}}_s L_s$, and $d = deg( \tilde{f} )$.
Define
$$E' = E_{x, i} \Biggl[
| \tilde{f}(x) - \tilde{f}( x + e_i ) |^2 \Biggr].$$
Putting the following two lemmas together we get Theorem \ref{degree}.
\begin{lemma} $E' \ge ( 1- 2\epsilon)^2 \rho_f$.
\end{lemma}
\begin{proof}
$E' \ge Pr \bigl[ f(x) \ne f(x + e_i) \bigr] ( 1- 2\epsilon )^2
= ( 1- 2\epsilon)^2 \rho_f.$
\end{proof}
\begin{lemma}
$E' \le 4 ( 1 + \epsilon )^2 \frac{d}{N}.$
\end{lemma}
\begin{proof}
\begin{eqnarray}
E' & = & E \Biggl[ | \sum_s \hat{\tilde{f}}_s ( 1 - ( -1 )^{ s \cdot e_i } ) L_s(x) |^2
\Biggr]\\
& = & \sum_{s_1, s_2} \hat{\tilde{f}}_{s_1} \hat{\tilde{f}}_{s_2}
E_x \Biggl[ L_{s_1+s_2} ( x )\Biggr]
E_i\Biggl[( 1 - ( -1 )^{ s_1 \cdot e_i } )( 1- ( -1 )^{ s_2 \cdot e_i } )\Biggr]\\
& = & \sum_s \hat{\tilde{f}}_s^2 E_i\Biggl[( 1 - ( -1 )^{ s \cdot e_i } )^2\Biggr]\\
& = & 4\sum_s \hat{\tilde{f}}_s^2 \lambda_s\\
& \le & \frac{4d}{N}\sum_s \hat{\tilde{f}}_s^2\\
& \le & 4 ( 1+ \epsilon )^2 \frac{d}{N}.
\end{eqnarray}
\end{proof}
Since most functions have large average influence, our
lower bound by influence (Theorem \ref{deg})
is better than the bound by block sensitivity (Theorem \ref{appdeg}) for
most functions. An extreme example is $PARITY$. Here is an
example of asymmetric function: Let $f ( x_0, x_1, x_2, x_3) =
x_0 ( x_1 - x_2 )^2 + ( 1 - x_0 ) ( x_2 - x_3 )^2$, and
$f_k$ is obtained by iterate $f$ $k$ times:
$f_k ( X_0, X_1, X_2, X_3) = f ( f_{k-1}(X_0),
f_{k-1}(X_1), f_{k-1}(X_2), f_{k-1}(X_3))$, where $X_i$ is the $i$th block
of $4^{k-1}$ input variables. Then one can show that $BS_{f_k} = 3^k$,
$\bar{S}_{f_k} = 2.5^k > \sqrt{BS_{f_k}}$.
It is conceivable that our lower bound
is beneficial in proving degree lower bounds of
asymmetric functions
when the bound by block sensitivity is not good.
(Although, in our example, $\bar{S}_{f_k}$
is not a tight bound for $\tilde{f}$ either: one can show
that $deg(\tilde{f}) \ge 3^k$.)
\section{Open problems}
So far the lower bound of quantum black-box complexity by
the degree of approximation polynomials implies
any other (asymptotic) lower bounds.
Is it asymptotically optimal for {\it all} Boolean functions?
For polynomials that approximates symmetric Boolean functions,
\cite{Pa92} gives a tight characterization on the lowest degree. However,
we do not know much about the case of asymmetric functions.
Is there any general tight bound for the lowest degree of
approximation polynomials?
The relations of block sensitivity and sensitivity and their average cases have been
long-standing open problems \cite{NS92, Ru95, Be96, Ve98}
. Is there any lower bound of quantum black-box complexity
and lower degree of approximation polynomials by average block
sensitivity?
\section{Acknowledgment}
The author is grateful to Andy Yao for insightful discussions
and encouragement,
to Umesh Vazirani, Ashwin Nayak, and Andris Ambainis
for helpful discussions and warm-hearted hosting, and to Ronald deWolf
for his comments and pointing out some mistakes in the previous version.
|
\section{Introduction}
Numerous works, for example
\cite{Parisi}, \cite{SpinGlass}, \cite{SpinGlass1} discuss
application of ultrametrics to investigation of spin glasses.
The most important example of ultrametric space is the field of
$p$-adic numbers, for introduction to $p$-adic analysis see
\cite{VVZ}.
In the present paper we apply the methods of $p$-adic analysis
to investigate the spontaneous symmetry breaking in the models of spin
glasses. We obtain the following results:
1)\qquad
A $p$-adic expression for the replica matrix $Q_{ab}$ is found.
It has the form $Q_{ab}=q_k$ where $k=\log_p |l(a)-l(b)|_p$
where the notations is expressed below.
It is shown that the replica matrix in the Parisi form \cite{Parisi}
in the models of spontaneous breaking of the replica symmetry
in the simplest case have the form of the Vladimirov operator of
$p$-adic fractional differentiation \cite{VVZ}.
2)\qquad
The model of hierarchical diffusion that was used in \cite{OS} to describe
relaxation of spin glasses
in our approach takes the form of the model of $p$-adic diffusion.
For instance, we reproduce the results of the paper \cite{OS}
using the methods of $p$-adic analysis.
The models of spontaneous breaking of the replica symmetry are used
for investigations of spin glasses
\cite{Parisi}, \cite{SpinGlass}, \cite{SpinGlass1}.
The breaking of symmetry in such models is described by the replica
$n\times n$ matrix
${\bf Q}=\left(Q_{ab}\right)$ in the Parisi form \cite{Parisi}.
This matrix looks as follows. Let us consider the set of integer numbers
$m_i$, $i=1,\dots,N$, where $m_i/m_{i-1}$
are integers for $i>1$ and $n/m_i$ are integers.
The matrix element of the replica matrix \cite{Parisi} is defined as
follows \begin{equation}\label{mat_el} Q_{aa}=0,\quad Q_{ab}=q_i,\qquad
\left[\frac{a}{m_{i}}\right]\ne \left[\frac{b}{m_{i}}\right];\quad
\left[\frac{a} {m_{i+1}} \right]= \left[\frac{b}{m_{i+1}} \right].
\end{equation}
Here $[\cdot]$ is the function of integer part
(we understand the integer part $[x]$ as follows: $[x]-1\le x\le [x]$
where $[x]$ is integer),
$q_i$ are some
(real) parameters.
An example of the matrix of this kind for
$m_i/m_{i-1}=2$ and $n=2^{N}$ have the form
\begin{equation}\label{matrix}
{\bf Q}=\left(
\begin{array}{ccccccccc}
{0}&q_{1}&q_{2}&q_{2}&q_{3}&q_{3}&q_{3}&q_{3}&\dots\\
q_{1}&{0}&q_{2}&q_{2}&q_{3}&q_{3}&q_{3}&q_{3}&\dots\\
q_{2}&q_{2}&{0}&q_{1}&q_{3}&q_{3}&q_{3}&q_{3}&\dots\\
q_{2}&q_{2}&q_{1}&{0}&q_{3}&q_{3}&q_{3}&q_{3}&\dots\\
q_{3}&q_{3}&q_{3}&q_{3}&{0}&q_{1}&q_{2}&q_{2}&\dots\\
q_{3}&q_{3}&q_{3}&q_{3}&q_{1}&{0}&q_{2}&q_{2}&\dots\\
q_{3}&q_{3}&q_{3}&q_{3}&q_{2}&q_{2}&{0}&q_{1}&\dots\\
q_{3}&q_{3}&q_{3}&q_{3}&q_{2}&q_{2}&q_{1}&{0}&\dots\\
\dots
\end{array}
\right)
\end{equation}
In the present paper we discuss the replica matrix
(\ref{matrix}) (more precisely, the generalization of this example
for the case of $p^N\times p^N$ matrices) using the language of
$p$-adic analysis. This allows to give the natural interpretation
for (\ref{matrix}) as the operator that can be diagonalized by
the $p$-adic Fourier transform. In particular this gives the spectrum
of the matrix (\ref{matrix}).
In the limit of infinite breaking of the replica symmetry $N\to\infty$
the dimension $p^N$ of the replica matrix tends to infinity, but
the $p$-adic norm of the dimension $|p^N|_p=p^{-N}$ tends to zero.
The conjecture by Volovich \cite{volovich} is that
this phenomenon might explain the paradoxical fact that in the replica method
the dimension of the replica matrix in the limit of infinite breaking of
the replica symmetry tends to zero.
Let us make here a brief review of $p$-adic analysis.
The field $Q_p$ of $p$-adic numbers is the completion of the field of rational
numbers $Q$ with respect to the $p$-adic norm on $Q$.
This norm is defined in the following way. An arbitrary rational number
$x$ can be written in the form $x=p^{\gamma}\frac{m}{n}$ with $m$ and $n$
not divisible by $p$. The $p$-adic norm of the rational number
$x=p^{\gamma}\frac{m}{n}$ is equal to $|x|_p=p^{-\gamma}$.
The most interesting property of the field of $p$-adic numbers is
ultrametricity. This means that $Q_p$ obeys the strong triangle inequality
$$
|x+y|_p \le \max (|x|_p,|y|_p).
$$
We will consider disks in $Q_p$ of the form
$\{x\in Q_p: |x-x_0|_p\le p^{-k}\}$.
For example, the ring $Z_p$ of integer $p$-adic numbers
is the disk
$\{x\in Q_p: |x|_p\le 1\}$
which is the completion of integers with the $p$-adic norm.
The main properties of disks in
arbitrary ultrametric space are the following:
{\bf 1.}\qquad
Every point of a disk is the center of this disk.
{\bf 2.}\qquad
Two disks either do not intersect or one of these disks
contains the other.
The $p$-adic Fourier transform
$F$ of the function $f(x)$ is defined as follows
$$
F[f](\xi)=\widetilde{f}(\xi)=\int_{Q_p}\chi(\xi x)f(x)d\mu(x)
$$
Where $d\mu(x)$ is the Haar measure.
The inverse Fourier transform have the form
$$
F^{-1}[\widetilde{g}](x)=\int_{Q_p}\chi(-\xi x)\widetilde{g}(\xi)d\mu(\xi)
$$
Here $\chi(\xi x)=\exp(i\xi x)$ is the character of the field of
$p$-adic numbers. For example, the Fourier transform of the indicator function
$\Omega(x)$ of the disk of radius 1 with center in zero (this is a function
that equals to 1 on the disk and to 0 outside the disk) is the function of the
same type:
$$
\widetilde{\Omega}(\xi)=\Omega(\xi)
$$
In the present paper we use the following Vladimirov operator
$D^{\alpha}_x$ of the fractional $p$-adic differentiation, that is
defined \cite{VVZ} as
\begin{equation}\label{diff}
D^{\alpha}_x
f(x)=F^{-1}\circ|\xi|_p^{\alpha}\circ F [f](x)=
\frac{p^{\alpha}-1}{1-p^{-1-\alpha}}
\int_{Q_p}\frac{f(x)-f(y)}{|x-y|_p^{1+\alpha}}d\mu(y)
\end{equation}
Here $F$ is the ($p$-adic) Fourier transform,
the second equality holds for $\alpha>0$.
For further reading on the subject of $p$-adic analysis see
\cite{VVZ}.
\section{The replica matrix}
Let us describe the model of the replica symmetry breaking using
the language of $p$-adic analysis. We will show that
the replica matrix ${\bf Q}=\left(Q_{ab}\right)$
can be considered as an operator in the space of functions on the finite set
consisting of $p^{N}$ points with the structure of the ring $p^{-N}Z/Z$.
The ring $p^{-N}Z/Z$ can be described as a set with the elements
$$
x=\sum_{j=1}^{N} x_j p^{-j},\qquad 0\le x_j\le p-1
$$
with natural operations of addition and multiplication up to modulus 1.
Let us consider the $p$-adic norm on this ring
(the distance can take values $0,p,\dots,p^{N}$).
We consider the following construction.
We introduce one-to-one correspondence
$$
l: 1,\dots,p^{N} \to p^{-N}Z/Z
$$
$$
l^{-1}: \sum_{j=1}^{N} x_j p^{-j} \mapsto 1+p^{-1}\sum_{j=1}^{N} x_j
p^{j}, \quad 0\le x_j \le p-1
$$
The formula (\ref{mat_el}) takes the form
\begin{equation}\label{mat_el_1}
Q_{aa}=0,\quad Q_{ab}=q_i,\qquad
\left[\frac{a}{p^{i-1}}\right]\ne \left[\frac{b}{p^{i-1}}\right], \quad
\left[\frac{a} {p^{i}} \right]= \left[\frac{b}{p^{i}} \right].
\end{equation}
Let us prove the following theorem.
{\bf Theorem.}
{\qquad\sl
The matrix element $Q_{ab}$
defined by (\ref{mat_el_1}) depends only on
$p$-adic distance between $l(a)$ and $l(b)$:
$$
Q_{ab}=\rho(|l(a)-l(b)|_p),
$$
where $\rho(p^{k})=q_{k}$, $\rho(0)=0$.
}
{\it Proof}
The condition $\left[\frac{a} {p^{i}} \right]= \left[\frac{b}{p^{i}} \right]$
in our notions have a form
$$
\left[\frac{1+p^{-1}\sum_{j=1}^{N} a_j p^{j}} {p^{i}} \right]=
\left[\frac{1+p^{-1}\sum_{j=1}^{N} b_j p^{j}}{p^{i}} \right]
$$
This means that $a_j=b_j$ for $j>i$. The condition
$\left[\frac{a}{p^{i-1}}\right]\ne \left[\frac{b}{p^{i-1}}\right]$
means that $a_i\ne b_i$. But these two condition both mean that
$|l(a)-l(b)|_p=p^{-i}$.
We get that the matrix element of the replica matrix
$Q_{ab}$ depends only on the $p$-adic distance
$|l(a)-l(b)|_p$: if $|l(a)-l(b)|_p$ equals to $p^{-k}$
then $Q_{ab}=q_{k}$ and the statement of the theorem follows.
The replica matrix $\left(Q_{ab}\right)$ acts on functions on
$p^{-N}Z/Z$ as on vectors with matrix elements $f_b$
where $b=l(y)$, $b=1,\dots, p^N$.
The action of the replica matrix in the space of functions
on $p^{-N}Z/Z$ takes the form
\begin{equation}\label{action_discrete}
{\bf Q}f(x)=\int_{p^{-N}Z/Z}\rho(|x-y|_p)f(y) d\mu(y)
\end{equation}
where the measure $d\mu(y)$ of one point equals to 1
and $f_b=f(l(b))$
(because we can consider the index $b$ of the vector
as the index of the first column of the matrix $\left(Q_{ab}\right)$).
It is easy to see that operators of the form (\ref{action_discrete})
have the following properties:
{\bf 1.}\qquad
The operators (\ref{action_discrete}) commute with operators of shift.
This means that the operators (\ref{action_discrete})
can be diagonalized by the Fourier transform
(in our case this is the discrete Fourier transform).
{\bf 2.}\qquad
The function $\rho$ depends on $p$-adic norm of argument.
{\bf 3.}\qquad
$\rho(0)=0$.
The language of $p$-adic analysis allows us to describe
the natural generalization of the operator (\ref{action_discrete}).
This generalization have the form of operator
\begin{equation}\label{action}
{\bf Q} f(x)
= \int_{Q_p} \rho({|x-y|_p})f(y) d\mu(y)
\end{equation}
where the function $\rho$ obeys the properties {\bf 1.}-{\bf 2.}
(an analogue of the property {\bf 3.} will be considered later).
Here and further in the present paper we use the agreement that
we will use the same notions (without special comments) for
analogous values in the discrete and the continuous ($p$-adic) cases.
It is easy to see that the character $\chi(kx)$ is the generalized eigenvector
for the operator (\ref{action}), if $\rho({|x|_p})\in L^1(Q_p)$.
Thus the operator (\ref{action}) can be diagonalized by the
$p$-adic Fourier transform $F$:
$ {\bf Q} f(x)=F^{-1} \circ \gamma(\xi) \circ F [f] (x) $.
From the property {\bf 2.} follows that the function $\gamma$
depends only on the $p$-adic norm of the argument:
$\gamma=\gamma(|\xi|_p)$. Therefore we get
$$
{\bf Q}
f(x)=F^{-1} \circ \gamma(|\xi|_p) \circ F [f] (x)
$$
\section{The model of hierarchical diffusion}
Let us reproduce now (partially) the results of the paper \cite{OS}
using the methods of $p$-adic analysis.
In the paper \cite{OS} relaxation of spin glasses
was described using the following
model of hierarchical diffusion. Let us consider $2^{N}$
points (we will also consider more general case of $p^{N}$ points, $p>0$
is prime), separated by barriers of energy.
The barriers of energy have the following form.
Let us enumerate the points by integer numbers starting from
$0$ to $2^{N}-1$ (analogously, from $0$ to $p^{N}-1$).
Let us consider the increasing sequence of the barriers of energy
(nonnegative numbers)
$0=\Delta_{0}<\Delta_{1}<\Delta_{2}<\dots<\Delta_{k}<\dots$.
We define the barriers of energy on the set of $p^{N}$ points
according to the following rule:
if $a-b$ is divisible by $p^{k}$ then the barrier between $a$-th and $b$-th
points equals to $\Delta_{k}$.
The hierarchical diffusion will be described by the ensemble of particles
that jump over the described above set of $p^{N}$ points.
Let us define the probability $q_{i}$ of transition (or jump) over the barrier
of energy $\Delta_i$ in the following way
$q_{i}=\exp(-\Delta_i)$, $i=1,2,\dots$.
Then the matrix of probabilities of transitions will be equal
(up to additive constant)
to the matrix ${\bf Q}$ of the form (\ref{matrix}).
We denote the density of particles at the $a$-th point as $f_a(t)$
and vector with elements that equal to densities at all points as
${\bf f}(t)$. We define the dynamics of the model using the following
differential equation \cite{OS}:
\begin{equation}\label{equos}
\frac{d}{dt}{\bf f}(t) = ({\bf Q}-\lambda_0 {\bf
I}){\bf f}(t)
\end{equation}
where $N\times N$ matrix ${\bf Q}$ for $p=2$ have the form
(\ref{matrix}) of the replica matrix for the model of the replica
symmetry breaking, ${\bf I}$ is the unity matrix,
$\lambda_0$ is the eigenvalue of the matrix ${\bf Q}$
that corresponds to the eigenvector with equal matrix elements.
This choice of the transition probability matrix is defined by the law
of the conservation of number of particles
(that is an analogue of the property {\bf 3.}).
Application of the technique developed in the section 2 allows
us to write the equation (\ref{equos}) in the form
\begin{equation}\label{equos_int_discrete}
\frac{d}{dt}f(x,t)=\int_{p^{-N}Z/Z}(f(y,t)-f(x,t))\rho(|x-y|_p) d\mu(y)
\end{equation}
where $f_a(t)=f(l(a),t)$.
For example, for the considered above
$q_{i}=\exp(-\Delta_i)$, $i=1,2,\dots$ and for the linear dependence
of the barrier energy $\Delta_i=i(1+\alpha)\ln p$ on $i$
we get $\rho(|x|_p)=|x|_p^{-1-\alpha}$ and the equation
(\ref{equos_int_discrete}) gets the form
\begin{equation}\label{equos1}
\frac{d}{dt}f(x,t)=
\int_{p^{-N}Z/Z}
\frac{f(y,t)-f(x,t)}{|x-y|_p^{1+\alpha}} d\mu(y)
\end{equation}
At the right hand side of the equation (\ref{equos1})
we get the discretization of the Vladimirov operator
$D^{\alpha}_x$ (\ref{diff}) of the fractional $p$-adic differentiation,
see \cite{VVZ}.
In the paper \cite{OS} the Cauchy problem for the equation (\ref{equos1})
with initial condition $f(x,0)=\delta_{x0}$ was investigated.
The dependence on time of the value $P_0(t)$
that in $p$-adic notions have the form
$$
P_0(t)=f(0,t)=\int_{p^{-N}Z/Z}\delta_{y0}f(y,t)d\mu(y)
$$
was found. In the present paper we will calculate the value $P_0(t)$
using the methods of $p$-adic analysis.
The $p$-adic generalization of the equation (\ref{equos_int_discrete})
have the following form
\begin{equation}\label{equos_int}
\frac{d}{dt}f(x,t)=\int_{Q_p}(f(y,t)-f(x,t))\rho(|x-y|_p) d\mu(y)
\end{equation}
Let us describe how to get the spectrum of the operator ${\bf D}$ at
the right hand side (or simply RHS) of
(\ref{equos_int}) (or the spectrum of times of relaxation for the model
of hierarchical diffusion \cite{OS}, describing spin glasses).
We will use the $p$-adic Fourier transform. It is easy to see that
the character $\chi(kx)$ is the generalized eigenfunction for the operator
at the RHS of (\ref{equos_int}) if
$\rho({|x|_p})\in L^1(Q_p\backslash U_{\epsilon})$,
where $U_{\epsilon}$ is the arbitrary neighborhood of 0,
or equivalently if $\forall k\in Z$ the series
$\sum_{i=k}^{\infty} |\rho(p^i)| p^{-i}$ converges.
For instance, 1 is the eigenfunction for the eigenvalue that equals to 0.
The proof is as follows:
$$
{\bf D} \chi(kx) =
\int_{Q_p}
(\chi(ky)-\chi(kx))\rho({|x-y|_p}) d\mu(y)=
$$
$$
=\chi(kx)
\int_{Q_p}
(\chi(k(y-x))-1)\rho({|x-y|_p}) d\mu(y)=
\chi(kx)
\int_{Q_p}
(\chi(ky)-1)\rho({|y|_p}) d\mu(y)
$$
To finish the proof we note that
$\chi(ky)$ is locally constant function that equals to 1
in some neighborhood of 0.
Using that the integral
$\int_{|y|_p\le p^i}\chi(ky) d\mu(y) = p^{i}$, if $|k|_p\le p^{-i}$
and equals to zero if $|k|_p > p^{-i}$, we get
\begin{equation}\label{spectrum}
{\bf D} \chi(kx) =
\left(-\left(1-p^{-1}\right)\sum_{p^{i}\le |k|_p}p^{-i} \rho(p^{i})
-\frac{p^{-1}}{|k|_p}\rho(|k|_p)\right)\chi(kx)
\end{equation}
This relation shows the correspondence between the spectrum of relaxation
times and the elements of the replica matrix in the form (\ref{matrix})
(here $q_i=\rho(p^{i})$).
The relation (\ref{spectrum}) reproduces the result obtained in
\cite{OS}, where more complicated technique was used.
Let us describe how to get the operator at the right hand side
of the equation (\ref{equos_int_discrete})
using the analogous operator
(at the right hand side of the equation (\ref{equos_int})) on $Q_p$.
Consider the finite dimensional subspace $V_N\subset L^2(Q_p)$
of the following form. The subspace $V_N$ consists of functions
with zero average with support in $p^{-N}Z_p$
that are constants on disks of radius 1.
Therefore the dimension of the subspace $V_N$ equals to $p^{N}-1$.
The operator at (\ref{equos_int}) maps this space into itself.
At the subspace $V_N$ the operator at the RHS of the equation
(\ref{equos_int}) takes the form
$$
{\bf D} f(x)=\int_{p^{-N}Z/Z}(f(y)-f(x))\rho(|x-y|_p) d\mu(y)
$$
that looks exactly like the operator at the RHS of the equation
(\ref{equos_int_discrete}).
But we will not get in this way the equation (\ref{equos_int_discrete})
because the operator at the RHS of (\ref{equos_int_discrete})
acts in the space of dimension that is larger by 1 than $V_N$.
This space can be obtained from the space $V_N$ by adding to $V_N$
the function that equals to 1 on the ball $p^{-N}Z_p$ (and to 0 outside).
Thus the model \cite{OS}
(up to the comments made above)
corresponds to the action of the operator of $p$-adic
fractional differentiation at the subspace.
We investigate the following $p$-adic generalization
of the model \cite{OS}. Let us consider the Cauchy problem
for the $p$-adic generalization of the equation (\ref{equos1})
\begin{equation}\label{equ_final}
\frac{d}{dt}f(x,t)+ A
D_x^{\alpha} f(x,t)=0,
\end{equation}
that have the form of the equation of $p$-adic diffusion
that was investigated in the book \cite{VVZ}.
We take the initial equation for the equation
(\ref{equ_final}) of the form
\begin{equation}\label{init_cond}
f(x,0)=\delta(x),
\end{equation}
This means that we investigate the fundamental solution
of the equation (\ref{equ_final}).
After the Fourier transform the equation (\ref{equ_final}) takes the form
$$
\frac{d}{dt}\widetilde{f}(\xi,t)+
A |\xi|_p^{\alpha} \widetilde{f}(\xi,t)=0,
$$
The solution of this equation is
$\widetilde{f}(\xi,0)e^{-A |\xi|_p^{\alpha}t}$.
Because the Fourier transform of the $\delta$-function with support in zero
equals to 1, we get finally
$$
\widetilde{f}(\xi,t)=e^{-A |\xi|_p^{\alpha}t}
$$
\begin{equation}\label{solution}
f(x,t)=\int_{Q_p}\chi(-\xi x)e^{-A |\xi|_p^{\alpha}t}d\mu(\xi)
\end{equation}
As the $p$-adic generalization of $P_0(t)$ we consider the value
$$
P_0(t)=\int_{|x|_p\le 1}f(x,t) d\mu(x)=\int_{Q_p}\Omega(x)f(x,t) d\mu(x)
$$
(we use the same notion) that for the solution (\ref{solution})
takes the form
\begin{equation}\label{p-adic_answer}
\int_{Q_p}\Omega(\xi) e^{-A |\xi|_p^{\alpha}t}d\mu(\xi)=
\int_{|\xi|_p\le 1}e^{-A |\xi|_p^{\alpha}t}d\mu(\xi)=
\left(1-p^{-1}\right)\sum_{k=0}^{\infty}p^{-k}e^{-Ap^{-\alpha k}t}
\end{equation}
The answer we have got coincides with the answer obtained in \cite{OS}
for (\ref{equos_int_discrete}).
The value found in \cite{OS} (they use $p=2$) have a form
\begin{equation}\label{their_answer}
P_0(t)=\lim_{n\to\infty}
\left(2^{-n}+{1\over 2}\exp\left(\frac{R^{n+1}t}{1-R}\right)
\sum_{m=0}^{n-1}\exp\left(-m\ln 2 -\frac{2-R}{1-R}R^{m+1}t\right)\right)
\end{equation}
where $0<R<1$ is some constant. It is easy to see that
(\ref{p-adic_answer}) and (\ref{their_answer}) coincide for
$R=p^{-\alpha}$ and $A=\frac{2-R}{1-R}$.
We see that $p$-adic analysis allows us to investigate
the models of hierarchical diffusion using the simple and natural formalism.
\vspace{10mm}
{\bf Acknowledgements}
\vspace{3mm}
The authors are grateful to I.V.Volovich for discussion.
This work was partially supported by INTAS 96-0698,
RFFI 98-03-3353a, RFFI 96-15-97352 and RFFI 96-15-96131 grants.
|
\section{Introduction}
It is presently fashionable to consider spatially flat cosmological
models with a cosmological constant (Ostriker and Steinhardt 1995).
Several projects are underway to measure the magnitude of the
cosmological constant, usually based on the luminosity-distance
(Perlmutter {\it et al.}\ 1997), the angular-diameter distance (Pen 1997), or
volume-distance (Kochanek 1996). Unfortunately, these distances are
only expressible in terms of Elliptic functions (Eisenstein 1997). In
order to simplify the repeated computation of difficult transcendental
functions or numerical integrals,
we present a fitting formula with the following properties:
\begin{enumerate}
\item it is exact for $\Omega_0\longrightarrow 1^-$ and
$\Omega_0\longrightarrow 0^+$ at all redshifts.
\item The relative error tends to zero as $z \rightarrow \infty$
for any value of $\Omega_0$.
\item For the range $0.2\le\Omega_0\le 1$, the relative error is less
than $0.4\%$.
\item For any choice of parameters, the relative error is always less
than $4\%$.
\end{enumerate}
Without further ado, the luminosity distance is given as:
\begin{eqnarray}
d_L &=&
\frac{c}{H_0}(1+z)\left[\eta(1,\Omega_0)-\eta(\frac{1}{1+z},\Omega_0)\right]
\nonumber\\
\eta(a,\Omega_0) &=& 2\sqrt{s^3+1}
\left[\frac{1}{a^4}-0.1540\frac{s}{a^3}+0.4304\frac{s^2}{a^2}
+0.19097\frac{s^3}{a}+0.066941s^4\right]^{-\frac{1}{8}}
\nonumber\\
s^3&=&\frac{1-\Omega_0}{\Omega_0}.
\label{eqn:dl}
\end{eqnarray}
We have used the Hubble constant $H_0$, and the pressureless matter
content $\Omega_0$. We recall
that an object of luminosity $L$ has flux $F=L/(4\pi d_L^2)$.
\section{Approximation}
Spatial flatness allows us to rewrite the Friedman-Robertson-Walker
metric as a conformally flat spacetime
\begin{equation}
ds^2=a^2(-d\eta^2+dr^2+r^2d\Omega).
\end{equation}
$\eta$ is the conformal time, and $r$ is the comoving distance. We
have set the speed of light $c=1$. The
scale factor $a$ can be normalized in terms of the redshift $a\equiv
1/(1+z)$. The Friedman equations determine
\begin{equation}
\left(\frac{da}{d\eta}\right)^2=a\Omega_0+a^4\Omega_\Lambda
\label{eqn:frw}
\end{equation}
where spatial flatness requires $\Omega_0+\Omega_\Lambda=1$. A change
of variables to $u=as$ where $s^3=(1-\Omega_0)/\Omega_0$ allows us to
express (\ref{eqn:frw}) parameter free
\begin{equation}
\eta = \sqrt{\frac{s^3+1}{s}}\int_0^{as} \frac{du}{\sqrt{u^4+u}}.
\label{eqn:etau}
\end{equation}
We can asymptotically approximate (\ref{eqn:etau}) as
\begin{equation}
\eta_1=2\sqrt{\frac{s^3+1}{s}}
\left[u^{-n}+n(2X)^{2n+1}u^{-1}+(2X)^{2n}\right]^\frac{-1}{2n}
\label{eqn:eta1}
\end{equation}
where $X\equiv (\int_0^\infty du/\sqrt{u^4+u})^{-1}=3\sqrt{\pi}/
(\Gamma[\frac{1}{6}]\Gamma[\frac{1}{3}])\simeq 0.3566$. $n$
is a free parameter, and at this stage a choice of $n=3$ approximates
all distances to better than $14\%$ relative accuracy. Equation
(\ref{eqn:eta1}) satisfies the following conditions: 1. it converges
to (\ref{eqn:etau}) as $u\rightarrow 0$ and
$u\rightarrow\infty$. 2. Its derivative converges to the derivative of
$\eta$ as $u\rightarrow \infty$. To improve on (\ref{eqn:eta1}) we
consider a polynomial expansion of $u^{-1}$ in the denominator with
two free parameters by setting $n=4$:
\begin{equation}
\tilde\eta=2\sqrt{\frac{s^3+1}{s}}
\left[u^{-4}+c_1u^{-3}+c_2u^{-2}+4(2X)^{9}u^{-1}
+(2X)^{8}\right]^{-1/8}.
\label{eta2}
\end{equation}
We will now choose the coefficients $c_1,c_2$ to minimize the relative
error in the approximate luminosity distance $\tilde d_L$,
\begin{equation}
e_\infty \equiv \|(\tilde d_L-d_L)/d_L\|_\infty
\label{eqn:err}
\end{equation}
where the subscript $\infty$ indicates the infinity norm, i.e. the
maximal value over the domain.
The error in (\ref{eqn:err}) tends to be dominated by $z=0$ (see for
example Figure \ref{fig:err2d}), for which we can express
\begin{equation}
e_\infty = \left\| \sqrt{u^4+u}\frac{d\tilde\eta}{du} - 1 \right\|_\infty.
\label{eqn:erru}
\end{equation}
Globally optimizing (\ref{eqn:erru}) allows us to reduce the error to
about $2\%$. But we choose the following trade-off for current
cosmological parameters: We want to minimize (\ref{eqn:erru}) over
the range $0.2\le \Omega_0\le 1$, which covers the popularly
considered parameter space. In that range, we find through a
non-linear equation solver that $c_1=-0.1540$ and $c_2=0.4304$. This
allows us to trade the global error of $2\%$ for a global error of
$4\%$, while reducing the error in the range of interest to $0.4\%$.
The global error surface plot is shown in Figure \ref{fig:err2d}.
The error at $z\rightarrow 0$ is shown in Figure \ref{fig:z0}.
At small $z$, any errors in $\tilde\eta$ are amplified, even though the global
errors in $\eta(z)$ are generally significantly smaller.
Figure (\ref{fig:eta}) shows the fit for the conformal time $\tilde\eta$
and its residual, which is accurate to $0.2\%$ globally.
One can always express the luminosity distance as a power series
expansion around $z=0$ using Equations (\ref{eqn:dl}) and
(\ref{eqn:etau}). For a flat universe, one obtains
\begin{equation}
\frac{H_0}{c}d_L = z + \left(1-\frac{3}{4}\Omega_0\right) z^2
+(9\Omega_0-10)\frac{\Omega_0}{8}z^3 + \cdots
\label{eqn:series}
\end{equation}
The series converges only slowly for $z\sim 1$: even when expanded to
sixth order in $z$, the maximal relative
error in the interval $0<z<1$ and $0.2<\Omega_0<1$ using the series
expansion in (\ref{eqn:series}) is 37\%.
\section{Conclusion}
We have presented a simple algebraic approximation to the luminosity
distance $d_L$ and the proper angular diameter distance
$d_A=d_L/(1+z)^2$ in a flat universe with pressureless matter and a
cosmological constant.
|
\section{Introduction}
One of the expected ways of achieving the Quark Gluon Plasma phase of
hadronic
matter is to increase the matter density to 5-10 times that of
normal nuclear matter. It has been thought that these high densities
may be achieved in central collisions of heavy nuclei at AGS
energies, provided that the relative motion is stopped. The
question of the degree of stopping in head-on collisions is thus of central
importance at these energies. Direct information pertaining to this
issue may be obtained by studying the transverse mass spectra of protons,
the majority of which are primordial, over a wide rapidity range to assess
whether the observed rapidity distribution is consistent with the initial
momentum of the projectile being converted to isotropic emission from a
source at rest in the center-of-mass system. We find that complete stopping,
defined in this way, is not achieved for the central
Au-Au collisions at any of the energies studied in the E917 experiment.
\section{Experimental arrangement}
The experimental arrangement is illustrated in Fig.
\ref{Back_fig1}. Beams of $^{197}$Au
with momenta of 6.84, 8.86 and 11.69 GeV/c per nucleon corresponding to
kinetic energies of 6.0, 8.0 and 10.8 GeV per nucleon were obtained from the
AGS at Brookhaven National Laboratory and focused onto a
Au-target of 1 mm thickness, which corresponds to $\sim$~3$\%$ interaction
\begin{figure}[hb]
\centerline{
\epsfig{file=Back_fig1.eps,width=\textwidth}}
\caption{Experimental arrangement.}
\label{Back_fig1}
\end{figure}
probability in the target. The trajectory of each beam particle was
determined by a beam vertex detector consisting of four planes of
scintillating fibers read out by position sensitive
photo-multiplier tubes arranged in orthogonal pairs and located at 5.84 m
and
1.72 m upstream from the target position. Further beam characterization was
performed using beam-time-zero and halo counters also placed upstream from
the target. Triggers for beam interactions with the target were obtained
by requiring that a signal of less than 75\% of that expected for the
full energy loss of a
Au-beam nucleus was registered in a circular ``bulls-eye'' \v{C}erenkov
detector placed 11 m downstream from the target.
The centrality of each beam-target interaction
was derived from the multiplicity of
particles (mostly pions) registered in a multiplicity detector array
subtending a solid angle of about 6.85 sr around the target and/or by the
total energy of the projectile remnant measured in the zero degree calorimeter.
In order to determine the reaction plane orientation in peripheral collisions
a hodoscope consisting of two orthogonal planes of 1 cm wide plastic
scintillator slats was placed in front of the zero degree
calorimeter. The azimuthal angle of the reaction plane determined from
the average position of the charged
projectile remnants in the hodoscope, relative to the beam
axis will be used to study collective flow characteristics of peripheral
collisions and the reaction plane dependence of the apparent source size
obtained in a Hanbury-Brown Twiss analysis of pion
pairs. Such an analysis is the subject of B. Holzman's talk at this
workshop \cite{Holzman}.
Particle spectra were obtained by momentum analysis in a movable
magnetic spectrometer consisting of a 0.4-Tesla magnet (Henry Higgins) and a
number of multi-wire ionization chambers used to determine the straight line
trajectories of particles entering and exiting the magnetic field. In
addition, a plastic scintillator wall located behind the spectrometer provided
particle identification by time-of-flight measurements relative to the
\v{C}erenkov
start detector located in front of the target.
This arrangement is capable of identifying charged pions, kaons, protons,
anti-protons and heavier nuclei. The separation of pions and kaons is
effective up to an energy of $\sim$1.75 GeV. Protons are separated from
pions up to $\sim$3.4 GeV, although there is a negligible kaon
contamination above $\sim$2.9 GeV.
The acceptance of the spectrometer is also sufficient to detect correlated
decay products of $\Phi \rightarrow K^+K^-$, $\Lambda \rightarrow p\pi^-$,
and $\overline \Lambda \rightarrow \overline p\pi^+$. The analysis of
$\Phi$-mesons and $\overline \Lambda$ is the subject of W.-C. Chang's talk
at this workshop\cite{Chang}.
The present talk will concentrate on the proton spectra obtained at the
three beam kinetic energies of 6, 8, and 10.8 GeV/nucleon, and an analysis
to the resulting rapidity distributions and fitted inverse slopes obtained
from the measured $m_t$-spectra.
\begin{figure}[ht]
\centerline{
\epsfig{file=Back_fig2.eps,height=4.in}}
\caption{Proton
$m_t$-spectra for central ($<$5\%) Au-Au collisions at a
beam kinetic energy of 10.8 GeV/nucleon for different
rapidity bins (laboratory system). Curves are Boltzmann distributions fitted
to the data. Spectra for adjacent rapidity bins are offset by factors of ten
to avoid overlapping.}
\label{Back_fig2}
\end{figure}
\section{Results}
In Fig. \ref{Back_fig2}, spectra of the invariant probability for proton emission per
trigger event are plotted for the 10.8 GeV/nucleon beam energy
as a function of the transverse mass $m_t-m_0$ for the 5\% most central
collisions as determined from the energy deposition in the
zero-degree calorimeter. The spectra are shown for different rapidity bins
as indicated. Only statistical error bars are shown.
Note that adjacent spectra are offset by factors of ten
to avoid overlapping. The range in $m_t$ reflects the acceptance of the
spectrometer in the different rapidity bins ranging from backwards to near
mid-rapidity at $y_{pp}$ = 1.613. The curves represent the best fits to the
spectra using a Boltzmann distribution, {\it i.e.}
\begin{equation}
\frac{1}{2\pi m_t}\frac{d^2N}{dydm_t} = Cm_t\exp(-m_t/T),
\end{equation}
where $C$ is a normalization constant and $T$ is the inverse slope of the
spectrum, both of which are determined from the fit to the data. We observe
that the experimental $m_t$-spectra are in excellent agreement with this
shape although they could also be described almost equally well by a pure
exponential function.
From these fits we derive the total probability for proton emission per unit
of rapidity, $dN/dy$, which is plotted as a function of rapidity in the
center-of-mass frame, $y-y_{cm}$, in the left panels of Fig. \ref{Back_fig3}. The
derived inverse slopes $T$ are shown in the right hand panels. Data are
shown for 5\% central collisions at all three beam energies, where the
centrality for the 6 and 8 GeV/nucleon data are obtained from the
the multiplicity array at the target position. The measured
points are represented by solid circles, whereas reflection around
mid-rapidity results in the open points. Error bars on the fit parameters
are purely statistical and do not include possible systematic errors.
\begin{figure}[ht]
\centerline{
\epsfig{file=Back_fig3.eps,height=4.in}}
\caption{Proton
rapidity distributions in the center-of-mass system (left
panels) and inverse slopes (right panels) are compared to a simple thermal
source prediction for the three beam energies. The arrows indicate target and beam
rapidities.}
\label{Back_fig3}
\end{figure}
We note that all three $dN/dy$ distributions are quite flat over the
measured rapidity range. The inverse slopes, $T$, show, however, a distinct
peaking at mid-rapidity. The solid curves in Fig. \ref{Back_fig3}
represent the
expectation for isotropic emission from a thermal source with temperature,
$T_0$, at rest in the center-of-mass system.
For such a source one expects that the $y$-dependence of the inverse slope is:
\begin{equation}
T = T_0/\cosh y
\end{equation}
and a $dN/dy$ distribution of \cite{Schnedermann93}
\begin{equation}
\frac{dN}{dy} \propto
T_0\left(m_0^2+2m_0\frac{T_0}{\cosh y }+2\frac{T_0^2}{\cosh^2 y
}\right)\exp(-m_0\cosh y/T_0),
\end{equation}
where $m_0$ is the proton rest mass.
In the right hand panels of Fig. \ref{Back_fig3}
we compare the rapidity dependence of the
inverse slope, $T$, with those predicted by this model. The source
temperature, $T_0$, was adjusted to account for the observed inverse slope at
mid-rapidity. We note that this naive model gives a rather good
representation of the observed inverse slopes.
On the other hand, a comparison of the predicted distribution in rapidity
$dN/dy$ with the measurements (left hand panels in Fig. \ref{Back_fig3})
reveals a
discrepancy which clearly demonstrates that the observed proton spectra
are inconsistent with isotropic emission from a single source at rest in the
center-of-mass system. Rather, we note that the rapidity distributions for
protons are flat over a wide range of rapidities
indicating a significant degree of either incomplete stopping or longitudinal
expansion at all three beam energies.
Of course, the naive model shown here also disregards the possible effects of
radial expansion in the fireball. It has been shown\cite{Schnedermann93},
however, that, within a
wide range of parameters, there is a strong anti-correlation
between the radial
expansion velocity, $v_0$, and the source temperature,
$T_{source}$, such that
it is impossible to disentangle their relative values from fits to spectra
of a single particle species {\it e.g.} protons. In the present analysis we
have therefore chosen to use only a single parameter, namely the {\it apparent}
source temperature, $T_0$, keeping in mind that
its value does not necessarily represent the
true temperature of the source formed in a central Au-Au collision.
\begin{figure}[hb]
\centerline{
\epsfig{file=Back_fig4.eps,height=4.in}}
\caption{Proton rapidity distributions (top) and inverse slopes (bottom)
for Au-Au collisions at 10.8 GeV/nucleon beam kinetic energy
compared with previously
published results from the E866 and E877 experiments at the AGS. The arrows
indicate target and beam rapidities.}
\label{Back_fig4}
\end{figure}
In Fig. \ref{Back_fig4} we compare the
$dN/dy$ (top panel) and inverse slopes $T$ (bottom
panel) from our experiment to
results from the E866 \cite{Ahle98} and E877\cite{Lacasse96}
experiments at the AGS. At
mid-rapidity there is good agreement between
the present data for the $dN/dy$ distribution and that from the
E866 experiment, but a relatively small
discrepancy between the two data sets is apparent
at less central rapidities. The source of this discrepancy is presently
being investigated. The E877 data were measured in the extreme forward rapidity
region but overlap with the present (reflected) data in the rapidity
region $y-y_{cm}$ = 0.8 - 1.1. Here, there appear to be
substantial discrepancies between the data sets.
The general trend of the three data sets is, however, clear.
There is a wide range -0.7$ < y-y_{cm} <$ 0.7 around mid-rapidity, where
the $dn/dy$ distribution is essentially flat, followed by a monotonic
decrease on either side. We note that this observed range of essentially
constant $dN/dy$ is inconsistent with thermal emission from
a stopped source (solid curve), as well as early predictions of the
Relativistic Quantum Molecular Dynamics model\cite{Sorge89} (RQMD v1.08,
\cite{Lacasse96}) (solid histogram), both of which are too sharply peaked at
mid-rapidity.
The slopes, however, appear to be in reasonable agreement with the $1/\cosh y$
dependence expected from the thermal model (solid curve) although this may
be fortuitous since the rapidity distribution clearly show that there
is either a significant amount of transparency or longitudinal expansion
at these energies which violates
the assumption of a thermal source at rest in the center-of-mass system.
\section{Conclusion}
An analysis of $m_t$-spectra for protons emitted in central Au-Au
collisions at beam kinetic energies of
6, 8 and 10.8 GeV/nucleon in terms of Boltzmann distributions has been
carried out, and the resulting $dN/dy$-distributions and inverse slopes
derived. They are compared to a simple thermal model assuming
isotropic emission from a source at rest in the center-of-mass system
corresponding to complete stopping of the colliding Au nuclei in central
collisions. We find that the rapidity distributions $dN/dy$ are substantially
wider and essentially constant around mid-rapidity
although the inverse slopes exhibit
the expected $1/\cosh y$ dependence on rapidity. We interpret this as a
manifestation of incomplete stopping or longitudinal expansion
of the entrance channel momenta at all three beam energies.
\begin{acknowledgments}
This work was supported by the Department of Energy,
the National Science Foundation, (USA),
and KOSEF (Korea).
\end{acknowledgments}
\begin{chapthebibliography}{1}
\bibitem{Holzman}
Holzman, B. {\it et al.} (1999) Contribution to these proceedings.
\bibitem{Chang}
Chang, W.-C. {\it et al.} (1999) Contribution to these proceedings.
\bibitem{Schnedermann93}
Schnedermann, E. {\it et al.} (1993) Phys.\ Rev.\ {\bf C48}, 2462.
\bibitem{Ahle98}
Ahle, L. {\it et al.} (1999) Phys.\ Rev.\ {\bf C57}, R466.
\bibitem{Lacasse96}
Lacasse, R. {\it et al.} (1996) Nucl.\ Phys.\ {\bf A610}, 153c.
\bibitem{Sorge89}
Sorge, H. {\it et al.} (1989) Ann.\ Phys.\ (NY)\ {\bf 192}, 266.
\end{chapthebibliography}
\end{document}
All the Things that can be Done with Figure Captions]
{All the Things that can be Done\\ with Figure Captions}
\begin{figure}[ht]
\vskip.2in
\caption{Short caption.}
\end{figure}
\begin{figure}[ht]
\vskip2pt
\caption{\protect\inx{Oscillograph} for memory address access operations, showing 500 ps
address access time and $\alpha\beta\Gamma\Delta\sum_{123}^{345}$
\protect\inx{superimposed signals}%
\protect\inxx{address,superimposed
signals} of address access in 1 kbit
memory plane.}
\end{figure}
\begin{figure}[ht]
\dblcaption{This caption will go on the left side of
the page. It is the initial caption of two side-by-side captions.}
{This caption will go on the right side of
the page. It is the second of two side-by-side captions.}
\end{figure}
\begin{figure}[ht]
\contcaption{This is a continued caption.}
\end{figure}
\inxx{captions,figure}
\begin{figure}[ht]
\caption{This caption is not continued so it has a new caption number.}
\end{figure}
\begin{figure}[ht]
\narrowcaption{This is a narrow caption so that it can
be at the side of the illustration. This is a narrow caption.
This is a narrow caption. This is a narrow caption.}
\end{figure}
\begin{figure}[ht]
\narrowcontcaption{This is a narrow continued caption.
This is a narrow continued caption. This is a narrow continued caption.}
\end{figure}
\clearpage
\begin{figure}[ht]
\letteredcaption{a}{Lettered caption.}
\end{figure}
\inxx{captions,lettered}
\begin{figure}[ht]
\lettereddblcaption{b}{One caption.}
{c}{Two captions.}
\end{figure}
\section{Making Tables}\inxx{Making tables}
Notice that the caption should be at the top of the table. Use
a line above the table, under the column heads, and at the
end of the table. This form of the tabular command makes the
table spread out to the width of the page.
\begin{table}[ht]
\caption[Effects of the Two Types of Scaling Proposed by Dennard
and Co-Workers.$^{a,b}$]
{Effects of the Two Types of Scaling Proposed by \protect\inx{Dennard}
and\newline
Co-Workers.$^{a,b}$}
\begin{tabular*}{\textwidth}{@{\extracolsep{\fill}}lcc}
\hline
\it Parameter&\it $\kappa$ Scaling &\it $\kappa$, $\lambda$ Scaling\cr
\hline
Dimension&$\kappa^{-1}$&$\lambda^{-1}$\cr
Voltage&$\kappa^{-1}$&$\kappa^{-1}$\cr
Currant&$\kappa^{-1}$&$\lambda/\kappa^{2}$\cr
\inx{Dopant Concentration}&$\kappa$&$\lambda^2/\kappa$\cr
\hline
\end{tabular*}
\begin{tablenotes}
$^a$Refs.~19 and 20.
$^b\kappa, \lambda>1$.
\end{tablenotes}
\end{table}
\inxx{captions,table}
\begin{table}[ht]
\letteredcaption{a}{A small table with a lettered table caption.}
\centering
\begin{tabular}{lcr}\hline
\it $\alpha\beta\Gamma\Delta$ One&\it Two&\it Three\cr\hline
one&two&three\cr
one&two&three\cr\hline
\end{tabular}
\label{table1a}
\end{table}
\clearpage
\begin{table}[ht]
\caption{Here is a table caption.}
\begin{center}
\begin{tabular}{||c||c||l}
\savehline
\it Cell\vrule height 14pt width 0pt depth 4pt
&\it Time (sec.)&\cr
\savehline
\savehline
1&432.22\vrule height 12pt width0pt&\cr
2&\phantom{3}32.32&\cr
3&\phantom{33}2.32&\cr
\savehline
\end{tabular}
\end{center}
\end{table}
\begin{table}[ht]
\contcaption{This is a continued caption.}
\begin{center}
\begin{tabular}{||c||c||l}
\savehline
\it Cell\vrule height 14pt width 0pt depth 4pt
&\it \inx{Time} (sec.)&\cr
\savehline
\savehline
4&532.22\vrule height 12pt width0pt&\cr
5&\phantom{3}12.02&\cr
6&\phantom{33}4.44&\cr
\savehline
\end{tabular}
\end{center}
\end{table}
\section{Other environments}
\begin{quote}
This is a sample of extract or quotation.\inxx{quotation}%
\inxx{quotation,extract}
This is a sample of extract or quotation.
This is a sample of extract or quotation.
\end{quote}
\begin{enumerate}
\item
This is the first item in the numbered list.
\item
This is the second item in the numbered list.
This is the second item in the numbered list.
This is the second item in the numbered list.
\end{enumerate}
\begin{itemize}
\item
This is the first item in the itemized list.
\item
This is the first item in the itemized list.
This is the first item in the itemized list.
This is the first item in the itemized list.
\end{itemize}
\begin{itemize}
\item[]
This is how to get an indented paragraph without
an item marker.
\item[]
This is how to get an indented paragraph without
an item marker.
\end{itemize}
\section[Small Running Head]{Some Sample Algorithms}
When you want to demonstrate some programming code, these are
the commands to use. Lines will be preserved as you see them
on the screen, as will spaces at the beginning of the line.%
\inxx{algorithm,State transition}\inxx{algorithm}
A backslash followed with a space will indent the line.
Blank lines will be preserved.
Math and font changes may be used.
\begin{algorithm}
{\bf state\_transition algorithm} $\{$
\ for each neuron $j\in\{0,1,\ldots,M-1\}$
\ $\{$
\ calculate the weighted sum $S_j$ using Eq. (6);
\ if ($S_j>t_j$)
\ $\{$turn ON neuron; $Y_1=+1\}$
\ else if ($S_j<t_j$)
\ $\{$turn OFF neuron; $Y_1=-1\}$
\ else
\ $\{$no change in neuron state; $y_j$ remains %
unchanged;$\}$ .
\ $\}$
$\}$
\end{algorithm}
Here is another sample algorithm:
\begin{algorithm}
{\bit Evaluate-Single-FOE} ({\bf x$_f$, I$_0$, I$_1$}):
\ {\bf I}+ := {\bf I}$_1$;
\ ($\phi,\theta$) := (0,0);
\ {\it repeat}\note{/*usually only 1 interation required*/}
\ \ (s$_{opt}${\bf E}$_\eta$) := {\bit Optimal-Shift} ({\bf I$_0$,I$^+$,I$_0$,x$_f$});
\ \ ($\phi^+$, $\theta^+$) := {\bit Equivalent-Rotation} ({\bf s}$_{opt}$);
\ \ ($\phi$, $\theta$) := ($\phi$, $\theta$) + ($\phi^+$, $\theta^+$);
\ \ {\bf I}$^+$:= {\bit Derotate-Image} ({\bf I}$_1$, $\phi$, $\theta$);
\ \ {\it until} ($\|\phi^+\|\leq\phi_{max}$ \& $\|\theta^+\|\leq\theta_{max}$);
\ {\it return} ({\bf I}$^+$, $\phi$, $\theta$, E$_\eta$).
End pseudo-code.
\end{algorithm}
\inxx{code,Pseudo}
This is an example of `codesamp' with a `codebox' included. Notice
that `underline' will still work even though this is basically
a verbatim environment.\inxx{code,Sample}
\begin{codesamp}
sqrdc(a, n)(a, qraux)\string{
\underline{DARRAY float[180] a[180];}
float qraux[180], col[180], nrmxl,t;
DO(1=0, n)\string{
\underline{ALIGN*(i=1, n) col[i]=a[l][i];}
\begin{codebox}{2.3in}
init*\string{ nrmxl=0.0;\string}
DO*(i=l, n)\string{
nrmxl += col[i]*col[i];\string}
combine*\string{nrmxl;\string}
\end{codebox}
nmxl=sqrt(nrmxl);
if (nrmxl != 0.00)\string{
if (col[1]=1.0+col[1];
\end{codesamp}
\section{Summary}
This is a \inx{summary} of this article.
\chapappendix{This is a Chapter Appendix}
This is an appendix which is meant to appear in individual chapters
of the edited book, not at the end of the book.
\begin{figure}[ht]
\caption{This is an appendix figure caption.}
\end{figure}
\begin{table}[ht]
\caption{This is an appendix table caption.}
\centering
\begin{tabular}{ccc}
\hline
one&two&three\\
\hline
C&D&E\\
\hline
\end{tabular}
\end{table}
\begin{equation}
\alpha\beta\Gamma\Delta
\end{equation}
\chapappendix{}
This is a chapter appendix without a title
meant to appear in individual chapters
of the edited book, not at the end of the book.
\begin{equation}
e=mc^2
\end{equation}
|
\section{Introduction}
\label{sec:introduction}
This paper describes a new computer code which simulates a self-gravitating,
relativistic perfect fluid in spherical symmetry. The fluid model uses
an equation of state $P=(\Gamma -1)\rho$, where $P$ and $\rho$ are
the fluid pressure and total energy density, respectively, and $\Gamma$ is
a constant satisfying $1 \le \Gamma \le 2$.
The code has been optimized for {\em ultrarelativistic} fluid flows,
that is, for flows with Lorentz factors much larger than unity.
This optimization involves a novel definition of the fluid variables, the use of
a modern high-resolution shock-capturing scheme, and care in reconstruction of
the primitive fluid variables---such as pressure and velocity---from
the conserved quantities which are actually evolved by the code.
Our new code was specifically developed to study the critical gravitational
collapse of perfect fluids. Critical collapse has become an
interesting subfield in general relativity since its initial
discovery in the massless Klein-Gordon system~\cite{mwc93},
and the perfect fluid model has played an important role
in advancing our understanding of the critical phenomena which arise
at the threshold of black hole formation.
(For an excellent introduction to critical phenomena, see the
review by Gundlach~\cite{cg97}.)
While the critical solutions for perfect fluids in spherical
symmetry have been the subject
of recent study~\cite{cejc,kha,dm,hka,kha2,pbmc,goliath,carr1,carr2},
the precise nature of the critical solutions
for $\Gamma \gtrsim 1.89$ was not previously
known, and thus one of the chief goals of our investigation was a thorough
analysis of this regime. In the remainder of this paper we
describe the equations of motion which are solved, and the numerical
techniques which we use to solve them. A companion paper~\cite{dnmc2}
describes in detail the results we have generated with the code.
\section{Geometry and fluid model}
\label{sec:intro_equations}
The Einstein equations couple the spacetime geometry, encoded
in the Einstein tensor, $G_{ab}$, to the stress-energy tensor, $T_{ab}$,
associated with the matter content of the spacetime:
\begin{equation}
G_{ab} = 8\pi T_{ab},
\end{equation}
(here and throughout, we use units
in which the speed of light, and Newton's gravitation constant
are unity: $c=1$ and $G=1$, and Latin indices $a, b, c, \cdots$ take on
the spacetime values $0, 1, 2, 3$.)
A fluid is a continuum model for a large number of
particles that uses macroscopic properties of a thermodynamic
system, such as internal energy and pressure, as fundamental
dynamical variables.
A perfect fluid has no shear stresses or dissipative forces,
and has a stress-energy tensor
\begin{equation}
T_{ab} = (\rho + P)u_a u_b + Pg_{ab},
\end{equation}
where $\rho$ is the energy density, $P$ is the pressure,
$u_a$ is the fluid's four-velocity, and $g_{ab}$ is the spacetime metric.
The energy density $\rho$
contains {\it all} contributions to the total energy, which for a perfect
fluid include the rest mass energy density, $\rho_o$,
and the internal energy density
\begin{equation}
\rho = \rho_o + \rho_o\epsilon,
\end{equation}
where $\epsilon$ is the specific internal energy.
The fluid number density, $n$, is related to $\rho_o$ via
\begin{equation}
\rho_o = mn,
\end{equation}
where $m$ is the rest mass of a single fluid particle.
The basic equations of motion for the fluid can be derived from
local conservation of (a) the energy-momentum
\begin{equation}
\nabla_aT^{ab} = 0,
\end{equation}
and (b) the particle number
\begin{equation}
\nabla_a \left( n u^a \right) = 0,
\end{equation}
where $\nabla_a$ is the (covariant)
derivative operator compatible with $g_{ab}$.
To these conservation laws one must adjoin an
equation of state, $P = P(\rho_o,\epsilon)$, which,
further, must be consistent with the first law of thermodynamics.
\subsection{Equation of state}
\label{sec:eos}
The equation of state (EOS) closes the fluid equations by providing a
relationship between the pressure and (in our case) the rest energy density
and internal energy. The nature of this relationship provides
much of the physics for a given system.
As mentioned in the introduction, our primary motivation for exploring
ultrarelativistic fluid dynamics is
to study perfect fluid critical solutions.
We expect these solutions to be scale invariant (self-similar),
and we therefore choose an EOS compatible with this symmetry.
The EOS
\begin{equation}
P = (\Gamma - 1)\rho,
\label{eq:ureos}
\end{equation}
where $\Gamma$ is a constant,
is the {\em only} EOS of the form $P=P(\rho)$ which
is compatible with
self-similarity~\cite{mcat,aotp,ce93},
and is notable for the fact that it results in a sound speed, $c_s$, which
is independent of density:
\begin{equation}
c_s = \sqrt{\frac{dP}{d\rho}}
= \sqrt{\Gamma -1}.
\end{equation}
One can argue that
this EOS is particularly appropriate for ultrarelativistic
fluids, and hence we will refer
to \eref{eq:ureos} as the {\it ultrarelativistic equation
of state.}
We note that
the EOS for a ``radiation fluid'' corresponds to $\Gamma = \frac{4}{3}$,
while $\Gamma=1$ gives a pressureless fluid (dust).
We do not consider the case of dust collapse here; hence, in
what follows, $1 < \Gamma \le 2$.
Another important fluid model is the ideal gas with the equation of state
\begin{equation}
\label{eq:igeos}
P=(\Gamma -1)\rho_o\epsilon.
\end{equation}
In the ultrarelativistic
limit, the kinetic energy of the constituent particles of the fluid
(or internal energy of the fluid in a thermodynamic context)
is much larger than the mass energy, $\rho_o\epsilon \gg \rho_o$,
giving $\rho\approx \rho_o\epsilon$.
Thus, one can interpret the EOS \eref{eq:ureos} as
the ultrarelativistic limit of the ideal-gas EOS.
As discussed in~\cite{dnmc2}, the ideal-gas EOS, in the
ultrarelativistic limit, becomes, in a limiting sense, scale invariant.
As the critical solutions reside in this
ultrarelativistic limit, the critical solutions for fluids
with the ideal-gas EOS are reasonably expected to be identical to
the critical solutions computed using
\eref{eq:ureos}. For this reason we hereafter limit our attention to
the ultrarelativistic equation of state.
\subsection{Geometric equations of motion}
\label{sec:geometric_eqs}
We use the ADM 3+1 formalism (specialized to spherical symmetry) to integrate
the Einstein equations, and
choose polar-areal coordinates
for simplicity of the equations of motion and for singularity avoidance.
Specifically, adopting a polar-spherical coordinate system
$(t,r,\theta,\phi)$, we write the spacetime metric as
\begin{equation}
\rmd s^2 = -\alpha(r,t)^2\,\rmd t^2 + a(r,t)^2\,\rmd r^2
+ r^2\,\left( \rmd\theta^2
+ \sin^2\theta\,\rmd\phi^2 \right),
\label{eq:metric}
\end{equation}
wherein the radial coordinate, $r$, directly measures proper surface area.
In analogy with the usual Schwarzschild form of the static spherically symmetric
metric, it is also useful to define the mass aspect function
\begin{equation}
m(r,t) \equiv \frac{r}{2}\left( 1 - \frac{1}{a^2}\right) .
\end{equation}
The fluid's coordinate velocity, $v$, and
the associated Lorentz gamma function, $W$, are defined by
\begin{equation}
v(r,t) \equiv \frac{au^r}{\alpha u^t}, \qquad
W(r,t) \equiv \alpha u^t.
\end{equation}
Since the fluid four-velocity is a unit-length, time-like vector ($u^a u_a = -1$),
we then have the usual relation between $W$ and $v$:
\begin{equation}
W^2 = \frac{1}{1 - v^2} .
\end{equation}
We now introduce two {\em conservation} variables
\begin{eqnarray}
\label{eq:tsdef}
\tau(r,t) &\equiv (\rho + P)W^2 - P \nonumber \\
S(r,t) &\equiv (\rho + P)W^2 v,
\end{eqnarray}
so named because they allow the fluid equations of motion to be
written in {\em conservation form} (albeit with the addition of a source
term), as discussed in detail in \sref{sec:conservation}.
In contrast to the conservation variables, we refer to the
quantities $P$ and $v$ as {\em primitive} variables.
With the above definitions, the non-zero components of the
stress-energy tensor are given by
\begin{equation}
\begin{array}{lccl}
T^t{}_t = - \tau &&& T^r{}_r = Sv + P \\
T^t{}_r = \displaystyle\frac{a}{\alpha}S
&&& T^\theta{}_\theta = T^\phi{}_\phi = P.\\
\end{array}
\end{equation}
A sufficient set of Einstein equations for the geometric variables $a$ and
$\alpha$ are given by (a) the non-trivial component of the momentum constraint
(the notation $\partial_x f$ denotes partial
differentiation, i.e.\ $\partial_x f \equiv \partial f/\partial x$\/)
\begin{equation}
\partial_t a = - 4\pi r\alpha a^2 S,
\label{eq:momcon}
\end{equation}
and by (b) the polar slicing condition, which follows from the demand that
metric have the form~(\ref{eq:metric}) for all $t$:
\begin{equation}
\label{eq:polar_slicing}
\partial_r (\ln\alpha) = a^2 \left[ 4\pi r\left( Sv + P \right) + \frac{m}{r^2} \right].
\end{equation}
An additional equation for $a(r,t)$,
\begin{equation}
\label{eq:hamcon}
\partial_r a = a^3 \left( 4\pi r\tau - \frac{m}{r^2} \right) \, ,
\end{equation}
follows from the Hamiltonian constraint.
\subsection{Fluid equations of motion}
\label{sec:fluid_eom}
Given the ultrarelativistic EOS~(\ref{eq:ureos}), the time evolution of
our perfect fluid is completely determined by $\nabla_a T^{ab}=0$.
The derivation of the equations of motion---which can
can naturally be written in conservation form---is a straightforward
piece of analysis, and will not be given in detail here.
Instead, we will simply quote the results, and for convenience in discussing
the numerical method of solution, we adopt a
``state vector'' notation. We thus define two-component vectors
$\hat{\bfq}$ and ${\bf w}$, which are the conservation
and primitive variables, respectively
\begin{equation}
\hat{\bfq}
\equiv \left[ \begin{array}{cc} \tau \\ S \end{array} \right]\! , \qquad
{\bf w} \equiv \left[ \begin{array}{cc} P \\ v \end{array} \right] \!.
\end{equation}
We then define a ``flux vector,'' $\hat{\bff}$, and a ``source vector'' $\hat{\bpsi}$
\begin{equation}
\hat{\bff} \equiv \left[\begin{array}{cc} S \\ Sv + P \end{array}\right] \qquad
\hat{\bpsi} \equiv \left[ \begin{array}{cc} 0 \\ \Sigma \end{array} \right]\! .
\end{equation}
These variables have been introduced with a hat $(\,\hat{}\,)$
to distinguish them from the new variables defined in \sref{sec:newvars},
which are subsequently used in the actual numerical solution algorithm.
Further, to expedite the discretization of the equations of motion,
we decompose the source term, $\Sigma$, into two pieces, as follows:
\begin{equation}
\Sigma \equiv \Theta + \frac{2\alpha P}{ar},
\end{equation}
where
\begin{equation}
\Theta \equiv (Sv - \tau) \left( 8\pi\alpha a r P + \alpha a \frac{m}{r^2}\right)
+ \alpha a P\frac{m}{r^2}.
\end{equation}
We note that
in spherically symmetric Minkowski spacetime we have $\Theta= 0$ and
$\Sigma = 2P/r$.
With the above definitions, we can now write the
fluid equations of motion in the conservation form
\begin{equation}
\partial_t \hat{\bfq} +
\frac{1}{r^2}\partial_r \left( r^2 X\hat{\bff}\right) = \hat{\bpsi},
\label{eq:con_eom}
\end{equation}
where
\begin{equation}
X \equiv \frac{\alpha}{a}
\end{equation}
is a purely geometric quantity.
Written in the above form,
the fluid equations of motion \eref{eq:con_eom} contain a mixture of conservation
and primitive variables, and thus it is necessary to transform
between both sets of variables at each step in the integration
procedure.
The primitive variables ${\bf w}$ can be expressed in terms of
the conservation variables
$\hat{\bfq}$ by inverting the definitions~\eref{eq:tsdef}
of the conservation variables:
\begin{equation}
P = -2\beta\tau + \left[ 4\beta^2\tau^2
+ (\Gamma-1)(\tau^2 - S^2)\right]^{\frac{1}{2}}
\label{eq:pdef}
\end{equation}
\begin{equation}
v = \frac{S}{\tau + P},
\label{eq:vdef}
\end{equation}
where the non-negative constant $\beta$ is defined by
\begin{equation}
\beta \equiv \frac{1}{4}\left( 2 - \Gamma \right).
\end{equation}
The pressure equation~(\ref{eq:pdef}) comes from the solution of
a quadratic with a specific root
chosen to yield a physical (non-negative) pressure.
This demand ($P \ge 0$)
further requires that $\tau \ge |S|$. A second physical
requirement is that $v$ be bounded by the speed of light, $|v| \le 1$,
and from~(\ref{eq:vdef}) this will clearly be automatically satisfied when
$\tau \ge |S|$.
These physical restrictions on
the primitive variables can sometimes be violated in numerical
solutions of the fluid equations, and we discuss some numerical
techniques aimed at ameliorating such difficulties
in sections \ref{sec:floor} and \ref{sec:velocity}.
Finally, we note that the above transformation from $\tilde{\bfq}$ to ${\bf w}$
is particularly simple in that it can be expressed algebraically.
The corresponding transformations for the gamma-law gas EOS~\eref{eq:igeos}
involves a transcendental equation which, in a numerical
implementation, must be solved iteratively at each grid point.
\subsection{New conservative fluid variables}
\label{sec:newvars}
Using the conservation variables $\hat{\bfq}$ defined above,
and the numerical method described in sections \ref{sec:numerical_methods}
and \ref{sec:solving}, we developed a preliminary code to solve the
relativistic fluid equations. We then tested this code by considering
evolutions in Minkowski spacetime using slab and spherical symmetry.
The tests in slab symmetry were completely satisfactory, modulo the
convergence limitations of the numerical scheme.
However, in spherical symmetry, we found that our method frequently failed
for ``stiffer'' fluids ($\Gamma \gtrsim 1.9$), most notably in ``evacuation
regions'' where $\rho\rightarrow 0$. Additionally, the fluid in such
regions often became {\em extremely} relativistic, and the combination
of $\rho \rightarrow 0$ and $|v| \rightarrow 1$ proved particularly
difficult to simulate.
These problems that we encountered in spherical symmetry led us
to seek a new set of conservation variables, and to motivate
this change of variables, first consider the evolutions shown in \fref{fig:piphi}.
Here we begin with a time-symmetric, spherical shell of fluid,
which has a Gaussian energy density profile.
Due to the time-symmetry, as the evolution unfolds,
the shell naturally splits into two sub-shells---one in-going
and one out-going---and as the sub-shells separate,
a new evacuation region forms in the region where the fluid was originally
concentrated. Examination of the conservation variable profiles reveals that
$|S|\approx \tau$, and this observation
suggests that we adopt new variables
\begin{equation}
\Phi \equiv \tau - S, \qquad
\Pi \equiv \tau + S,
\end{equation}
which loosely represent the in-going ($\Phi$)
and out-going ($\Pi$) parts of the solution.
Thus our new state vector of conservation variables is
\begin{equation}
{\bf q} \equiv \left[ \begin{array}{cc} \Pi \\ \Phi \end{array} \right]\! .
\end{equation}
Not surprisingly, the numerical difficulties in evacuation regions
are not completely cured with this change of variables;
however, the new variables ${\bf q}$ provide a {\em significant}
improvement over $\tilde{\bfq}$
in evolutions of spherically symmetric fluids with $\Gamma \gtrsim 1.9$.
\begin{figure}
\epsfxsize = 10cm
\centerline{\epsffile{piphi.ps}}
\caption{These plots show various fluid quantities at four different instances
(equally spaced in time)
in a flat spacetime, slab-symmetric evolution with $\Gamma = 1.9$.
The initial configuration is a time-symmetric
Gaussian pulse.
The top frames show the evolution of the original conservation variables,
$\tau$ and $S$. As the evolution proceeds, the pulse separates
into left and right-moving halves, and a vacuum region ($\tau \to 0$)
develops between the two sub-pulses.
The bottom frames show the evolution of the
new conservation variables, $\Pi$ and $\Phi$, which are specifically
defined so as to avoid the formation of such vacuum regions.
The correspondence of the new variables to left and
right moving ``waves'' is also evident.
Note that the plots of $\tau$, $\Pi$ and $\Phi$ have the same
vertical scale, while
the vertical scale for $S$ is shown separately.
The horizontal (radial) scale is the same for all of the plots.
}
\label{fig:piphi}
\end{figure}
The equations of motion for the new variables ${\bf q}$
can be readily found by adding
and subtracting the two components of \eref{eq:con_eom}, giving
\begin{equation}
\partial_t {\bf q} + \frac{1}{r^2}\partial_r \left[ r^2 X {\bf f} \right] = \bpsi,
\label{eq:eom_new_vars}
\end{equation}
where the flux and source terms are now given by
\begin{equation}
{\bf f} \equiv \left[ \begin{array}{c}
\frac{1}{2}(\Pi - \Phi)(1+v) + P \\
\rule{0mm}{4mm}
\frac{1}{2}(\Pi - \Phi)(1-v) - P \end{array} \right],
\qquad
\bpsi \equiv \left[\begin{array}{r} \Sigma\\
\rule{0mm}{4mm} -\Sigma\end{array} \right]\! .
\end{equation}
The transformation from conservative to primitive variables can be found
by simply changing variables in \eref{eq:pdef} and \eref{eq:vdef}
\begin{equation}
P = -\beta(\Pi + \Phi) + \left[ \beta^2\left( \Pi + \Phi\right)^2
+ (\Gamma -1)\Pi\Phi\right]^\frac{1}{2},\\
\label{eq:ppiphi}
\end{equation}
\begin{equation}
v = \frac{\Pi - \Phi}{\Pi + \Phi + 2P}.
\label{eq:vpiphi}
\end{equation}
We note that, given $\tau > |S|$, the new variables ${\bf q}$ are strictly
positive: $\Pi > 0$, $\Phi > 0$.
\subsection{The perfect fluid as a scalar field}
\label{sf}
There is a well-known relation between an irrotational,
stiff ($\Gamma=2$) perfect fluid
and a massless Klein-Gordon scalar field. In this section we discuss
the relationship between scalar fields and perfect fluids for
$0 < \Gamma \le 2$.
The perfect fluid equations of motion
\begin{equation}
\nabla_aT^{ab} = 0,
\end{equation}
can be written in terms of $\rho$, $P$, and $u^a$ as
\begin{equation}
u^a \nabla_a + (\rho + P)\nabla_a u^a = 0,
\label{eq:I}
\end{equation}
\begin{equation}
(\rho + P)u^a \nabla_a u^b + \left( g^{ab} + u^a u^b\right)\nabla_a P = 0.
\label{eq:II}
\end{equation}
If we assume the ultrarelativistic equation of state,
$P=(\Gamma-1)\rho$,
then these equations become
\begin{equation}
\nabla \left( \rho^{1/\Gamma}u^a\right) = 0,
\end{equation}
\begin{equation}
u^a \nabla_a u^b
+ \frac{(\Gamma - 1)}{\Gamma} \left( g^{ab} + u^a u^b\right) \nabla_a \ln\rho = 0.
\end{equation}
We seek a specific combination of $\rho$ and $u^a$ that allows
the fluid equations to be written in terms of a single variable,
and therefore introduce the {\it ansatz}
\begin{equation}
w^a \equiv \rho^\mu u^a,
\end{equation}
where $\mu$ is a constant that will be determined below.
From elementary contractions
we can express both $\rho$ and $u^a$ in terms of $w^a$
\begin{equation}
\rho = \left( -w_a w^a \right)^{\frac{1}{2\mu}},
\label{eq:rho_w}
\end{equation}
\begin{equation}
u^a = \left( -w_b w^b\right)^{-\frac{1}{2}} w^a.
\label{eq:u_w}
\end{equation}
However, it remains to see if $\mu$ can be chosen such that
$w^a$ will satisfy the fluid equations of motion.
We substitute expressions \eref{eq:rho_w} and \eref{eq:u_w}
into the momentum equation \eref{eq:II}, and find that
this equation is satisfied provided that
\begin{equation}
\mu = \frac{\Gamma-1}{\Gamma},
\end{equation}
{\it and}
\begin{equation}
\nabla_{[a} w_{b]} = 0.
\label{eq:w_circ}
\end{equation}
This latter condition allows one to write $w^a$ as the gradient of a
scalar field
\begin{equation}
w_a = \nabla_a \varphi.
\end{equation}
The equation of motion for $\varphi$
is obtained from \eref{eq:I}
\begin{equation}
\nabla_a \left[ \left( -\nabla_c \varphi \nabla^c\varphi \right)^\nu
\nabla^a\varphi\right] = 0,
\end{equation}
where
\begin{equation}
\nu = \frac{2 - \Gamma}{2(\Gamma - 1)}.
\end{equation}
The condition \eref{eq:w_circ}, $\nabla_{[a}w_{b]}=0$,
reduces to the requirement that the fluid be irrotational
\begin{equation}
\nabla_{[a}u_{b]} = 0.
\end{equation}
Thus, the fluid equations for an ultrarelativistic, irrotational
fluid can be written in
terms of a nonlinear equation for a scalar field, $\varphi$.
For the stiff fluid ($\Gamma=2$), we find that the equation of motion
for $\varphi$ becomes the massless Klein-Gordon equation
\begin{equation}
\nabla_a\nabla^a \varphi = 0.\qquad\qquad (\Gamma=2)
\end{equation}
One typically places physically motivated conditions on the fluid
variables, such as $\rho > 0$ and $u^a u_a =-1$. Solutions of
the Klein-Gordon equation, however, have time-like, null, and
space-like gradients ($\nabla_a\varphi$). With the usual physical
constraints on the fluid, then only a subset of possible Klein-Gordon
solutions can be interpreted as $\Gamma=2$ perfect fluids, namely those with
$\nabla_a\varphi\nabla^a\varphi < 0$.
\section{Numerical methods for fluid equations}
\label{sec:numerical_methods}
An important consideration for numerical solutions of
compressible fluid flow is how the numerical method will respond to
the presence or formation of
shocks, i.e. discontinuities in the fluid variables.
These discontinuities often cause the
dramatic failure of na\"\i ve finite difference schemes, and as
shocks form {\em generically} from smooth initial data, many special
techniques have been developed for the numerical solution
of fluid equations.
One approach is to introduce an {\em artificial viscosity}
that adds extra dissipation in the vicinity of a shock,
spreading the would-be-discontinuity over a few grid points.
This technique has been widely used, and has the advantages
of simplicity of implementation and computational efficiency.
However, Norman and Winkler~\cite{nw} investigated the use of
artificial viscosity in relativistic flows, and showed that an {\em explicit}
numerical scheme treats the artificial viscosity term inconsistently
in relativistic fluid dynamics, leading to large
numerical errors in the ultrarelativistic limit, $W\gg 1$.
A second approach to solving the fluid equations with shocks
comes from methods developed specifically for conservation
laws. These methods, usually variations or extensions of
Godunov's original idea~\cite{godunov}
to use piece-wise solution of the Riemann problem,
have proven to be very reliable and robust.
LeVeque~\cite{rjlbook,rjl98} has written excellent
introductions to conservative methods,
and our presentation here is in the spirit of his work.
Furthermore,
the application of these methods to problems in relativistic astrophysics
has been recently reviewed by Ib\'a\~nez and Mart\'\i~\cite{ibanez_marti}.
However, for the sake of completeness, we first briefly define and
discuss conservation laws, and outline a
general approach for their solution. We then discuss a linear
Riemann solver and a cell reconstruction method that results in a
scheme which, for smooth flows, is second order accurate in the
mesh spacing.
\subsection{Conservation methods}
\label{sec:conservation}
Conservation laws greatly simplify the mathematical description of
physical systems by focusing on quantities $\cal Q$---where $\cal Q$ may
be a state vector with multiple components---that do not change with time
\begin{equation}
\partial_t \int_V \rmd {\cal Q} = 0.
\end{equation}
In this section we discuss the derivation of numerical schemes
for this specific and important case where
$\int \rmd \cal Q$ is conserved on the computational domain.
Our discussion will be general, and not specifically tailored for
the fluid PDEs derived in \sref{sec:newvars},
but for simplicity we restrict the discussion to one dimensional (in space)
systems.
While conservation laws are often written in
{\em differential} form (e.g.\ $\nabla_a T^{ab} = 0$) it is
useful to first consider an {\em integral} formulation, which is
often the more fundamental expression.
Consider an arbitrary volume or cell, ${\cal C}_i$, with
a domain $[x_1,x_2]$.
The quantity of $\cal Q$ within ${\cal C}_i$ is denoted ${\cal Q}_i$,
and we define a density function ${\bf q}$ such that
\begin{equation}
{\cal Q}_i = \int_{x_1}^{x_2} \rmd x\, {\bf q}.
\label{eq:con_dens}
\end{equation}
The change of ${\cal Q}_i$ with time can be calculated
from the flux, ${\bf f}({\bf q})$, of ${\bf q}$
through the cell boundaries.
This consideration thus yields our conservation law:
\begin{equation}
\frac{\rmd}{\rmd t} \int_{x_1}^{x_2} \rmd x \, {\bf q}(x,t)
= {\bf f}({\bf q}(x_1,t)) - {\bf f}({\bf q}(x_2,t)).
\label{eq:gen_con_law}
\end{equation}
The conservation law
can be written in {\em integral} form by integrating
\eref{eq:gen_con_law} from an initial time, $t_1$, to a
final time, $t_2$,
\begin{eqnarray}
& &\int_{x_1}^{x_2} \rmd x \, {\bf q}(x,t_2) = \nonumber \\
& &\qquad \int_{x_1}^{x_2} \rmd x \, {\bf q}(x,t_1)
+ \int_{t_1}^{t_2}\rmd t \, {\bf f}({\bf q}(x_1,t))
- \int_{t_1}^{t_2}\rmd t \, {\bf f}({\bf q}(x_2,t))
\label{eq:con_law_int}
\end{eqnarray}
and the differential form follows from further manipulation {\em if}
we assume that ${\bf q}$ is differentiable:
\begin{equation}
\partial_t {\bf q} + \partial_x {\bf f}({\bf q}) = 0.
\label{eq:con_law_diff}
\end{equation}
It should be emphasized that the integral formulation
should be viewed as {\em the} primary mathematical form
for a conservation principle,
because it is {\em not} dependent on an assumption of differentiability.
For example, at a shock front in a fluid system, ${\bf q}$ is not
differentiable, and the differential form of the conservation law fails,
while the integral formulation is still satisfied.
Discretizations of conservation equations via finite differences
rely on the differential form, and artificial viscosity must be
added near shock fronts, forcing ${\bf q}$ to be differentiable.
An alternate strategy is to develop numerical algorithms
based directly on the integral formulation of the conservation laws.
The Godunov method and its extensions are examples of this latter approach,
and are the topic of the next section.
\subsection{Godunov's Method}
\label{sec:godunov}
Numerical algorithms for conservation laws are developed
by discretizing the equations in their fundamental integral form.
These methods derive from a {\em control volume} discretization,
whereby the domain is divided into {\em computational} cells, $C_i$, now
defined to span the interval $[x-\triangle x/2,x+\triangle x/2] \equiv [x_{i-1/2},x_{i+1/2}]$,
where $\triangle x$
is the (local) spatial discretization scale.
Following the derivation of the integral conservation
law~\eref{eq:con_law_int} for the computation cell $C_i$,
we introduce the {\it averaged} quantities, $\bar{\bfq}_i^n$:
\begin{equation}
\bar{\bfq}_i^n = \frac{1}{\triangle x}\int_{x_{i-1/2}}^{x_{i+1/2}} \rmd x\, {\bf q}(x,t_n),
\label{eq:ave}
\end{equation}
with $t_n \equiv n\triangle t$, where $\triangle t$ is the temporal
discretization scale. We then obtain the discrete form
of the conservation law~\eref{eq:con_law_int}
\begin{equation}
\bar{\bfq}^{n+1}_i = \bar{\bfq}^n_i -\frac{\triangle t}{\triangle x}
\left( {\bf F}_{i+1/2} - {\bf F}_{i-1/2} \right),
\label{eq:con_gen_method}
\end{equation}
where the ``numerical flux'' is defined by
\begin{equation}
{\bf F}_{i+1/2} = \frac{1}{\triangle t}\int_{t_n}^{t_{n+1}} \rmd t \,
{\bf f}({\bf q}({x_{i+1/2}},t).
\label{eq:flux_int}
\end{equation}
At first blush, a numerical method based on a discretization of the
integral conservation law does not appear promising: the flux
integral~\eref{eq:flux_int} does not appear readily solvable,
and it generally is not.
However, in his seminal work,
Godunov~\cite{godunov} devised a technique to approximately
evaluate the flux integral by replacing the function ${\bf q}(x,t_n)$
with $\tilde{\bfq}(x,t_n)$,
where $\tilde{\bfq}(x,t_n)$ is a piece-wise constant function.
In this approach, the individual cells (``control volumes'')
are treated as a sequence of ``shock tubes'', and a separate Riemann
initial value problem is solved at each cell interface.
Provided that the waves from neighboring cells do not interact---a proviso
which gives a Courant-type condition on the time-step---each
Riemann problem can be solved exactly to
yield the local solution $\tilde{\bfq}(x,t)$ (for $t>t_n$) for each ``shock tube.''
Furthermore, since the solution of each of the local Riemann problems is
self-similar,
$\tilde{\bfq}(x_{i+1/2},t)$ is a constant in time, and the evaluation of
the integral~\eref{eq:flux_int} becomes trivial.
This then allows one to find explicit expressions for the cell averages
at the advanced time,
$\bar{\bfq}^{n+1}$, via~\eref{eq:con_gen_method}.
In summary, the Godunov method proceeds as follows:
(a) From the average $\bar{\bfq}_i^n$, one ``reconstructs'' a piece-wise constant
function $\tilde{\bfq}(x,t_n)$ to approximate the solution in $C_i$;
(b) the Riemann problem is solved at the interfaces between cells, giving
the solution $\tilde{\bfq}(x,t)$ for $t_n < t \le t_{n+1}$;
(c) the solution $\tilde{\bfq}(x,t_{n+1})$ is averaged over the cell $C_i$ to
obtain the average at the advanced time, $\bar{\bfq}^{n+1}_i$.
We note that methods for solving the Riemann problem exactly
for relativistic fluids have been given by Smoller and Temple~\cite{smoller}
for the ultrarelativistic EOS, and by
Mart\'\i\ and M\"uller~\cite{mm} for the ideal-gas EOS.
Godunov's method has many nice properties: in particular, it
is conservative
and allows for the stable evolution of strong shocks.
However, the original scheme {\em does} have some shortcomings:
convergence is only first order, and the exact solution of the Riemann
problem may be computationally expensive, especially for relativistic
fluids.
The convergence of the scheme can be improved by providing a
more sophisticated reconstruction $\tilde{\bfq}(x,t_n)$, giving what are
known as
{\it high-resolution shock-capturing} methods. One such procedure
is described in \sref{sec:cell_recon}, with details concerning
the scheme's convergence given in \sref{sec:tests}.
In order to address the issue of computational efficiency, approximate
Riemann solvers have been developed that relate the
problem-at-hand to a simpler system, for
which the Riemann problem is easier to solve.
Several approximate Riemann solvers have been developed for classical
fluid dynamics, and many of these approximate methods have been
extended to relativistic fluid systems. These include
relativistic two-shock solvers~\cite{balsara,dai_woodward},
a relativistic HLLE solver~\cite{schneider},
and,
as discussed in \sref{sec:linear_riemann},
various linearized solvers.
\subsection{Cell reconstruction}
\label{sec:cell_recon}
Godunov-type numerical methods are based on solutions of the
Riemann initial value problem at the interfaces between cells.
As discussed above, during an update step one introduces
functions $\tilde{\bfq}(x,t)$--- defined piece-wise on the
intervals $[x_{i-1/2},x_{i+1/2}]$---to approximate the solution in the
control volumes $C_i$.
These functions are created
from the cell averages $\bar{\bfq}^n_i$, and hence are called
{\em reconstructions}.
Consider the cell interface at $x_{i+1/2}$: the state of the
fluid immediately to the
right (left) is $\tilde{\bfq}^r_{i+1/2}$ ($\tilde{\bfq}^\ell_{i+1/2}$).
The simplest reconstruction is to assume that $\tilde{\bfq}$ is piece-wise constant
\begin{equation}
\tilde{\bfq}^\ell_{i+1/2} = \bar{\bfq}_i, \qquad
\tilde{\bfq}^r_{i+1/2} = \bar{\bfq}_{i+1},
\end{equation}
as used in the original Godunov method and, as already discussed,
this reconstruction results in a numerical scheme in which
the spatial derivatives (and hence the overall scheme)
have first order accuracy. The convergence
can be improved by using a higher-order reconstruction for $\tilde{\bfq}$,
but care must be exercised so that the reconstruction does not induce
spurious oscillations near discontinuities (see \fref{fig:lim}).
We have chosen to use a piece-wise {\it linear} reconstruction for $\tilde{\bfq}$,
which formally results in a scheme with second order convergence.
(The convergence properties are discussed in greater detail in
\sref{sec:tests}.) The $\tilde{\bfq}$ are reconstructed using
the total variation diminishing (TVD) minmod limiter introduced by
van Leer \cite{vl}.
The van Leer limiter forces $\tilde{\bfq}$ to be monotonic near discontinuities,
and this reduces the (local) accuracy of the scheme to first order.
The first step of the reconstruction algorithm
involves the computation of the slope (derivative
of the dynamical variable) centered at the cell boundaries
\begin{equation}
{\bf s}_{i+1/2} = \frac{\bar{\bfq}_{i+1} - \bar{\bfq}_{i}}{r_{i+1} - r_{i}}.
\end{equation}
A ``limited slope'', $\bsigma_i$, is then calculated via
\begin{equation}
\bsigma_i = \hbox{minmod}({\bf s}_{i-1/2},{\bf s}_{i+1/2}),
\end{equation}
where the minmod limiter is defined by
\begin{equation}
{\rm minmod}(a,b) = \cases{
a & if $|a| < |b|$ and $ab>0$\\
b & if $|b| < |a|$ and $ab > 0$\\
0 & if $ab < 0$.}
\end{equation}
Using the limited slopes, we evaluate $\bar{\bfq}$ at the cell interfaces
as follows:
\begin{equation}
\tilde{\bfq}^\ell_{i+1/2} = \bar{\bfq}_i + \bsigma_i(r_{i+1/2} - r_i)
\end{equation}
and
\begin{equation}
\tilde{\bfq}^r_{i+1/2} = \bar{\bfq}_{i+1} + \bsigma_{i+1}(r_{i+1/2} - r_{i+1}).
\end{equation}
Finally, if we are unable to calculate physical values for
$\tilde{\bfw}^\ell$ and $\tilde{\bfw}^r$ (a situation which {\em can} and {\em does}
occur owing to the finite-precision nature of our computations)
we revert to a piece-wise
constant reconstruction for $\tilde{\bfq}^\ell$ and $\tilde{\bfq}^r$.
\begin{figure}
\epsfxsize = 8cm
\centerline{\epsffile{lim.ps}}
\caption{The three frames of this plot show different ways
a discretized function can be reconstructed in a control-volume
numerical method.
The solid line represents a continuum (or ``analytic'') function
and the solid hexagons represent discrete, approximate values of the function
defined at grid points. Frame (a) represents
the piece-wise constant reconstruction. Frame (b) shows a na\"\i ve
piece-wise linear reconstruction of each cell using ${\bf s}_{i+1/2}$.
This reconstruction oscillates
near discontinuities in the function---such oscillations can easily lead
to instabilities. Frame (c) shows a piece-wise linear reconstruction performed
with the minmod limiter as described in the text. This reconstruction
produces a discrete representation of the dynamical variable which
remains well-behaved near discontinuities.}
\label{fig:lim}
\end{figure}
\subsection{The Roe linearized solver}
\label{sec:linear_riemann}
Perhaps the most popular approximate Riemann solver is the linearized
solver introduced by Roe \cite{roe}.
This solver (and subsequent variants)
has been used in a variety of applications involving general relativistic
fluids~\cite{ibanez,eulderink,romero,banyuls,brandt,font},
and has proven to be robust and efficient. (The efficiency comparison
is relative to solving either the exact Riemann problem for relativistic
fluids, or a nonlinear approximation, such as the two-shock solver.)
As the name suggests,
the linearized solver approximates the full nonlinear problem by replacing the
nonlinear equations by {\it linear} systems
defined at each cell interface. The associated
linear Riemann problems can then be solved exactly and cheaply, and the
resulting solutions can be pieced together to produce an approximation
to the solution of the original, nonlinear equations. Thus, in
order to understand the
Roe scheme, it is instructive to first consider linear conservation laws.
The linear, scalar advection equation
\begin{equation}
\partial_t q + \lambda\partial_x q = 0,
\end{equation}
has the well-known solution
$q(x,t) = q(x-\lambda t,0),$
where $\lambda$ is a constant and $q(x,0)$ specifies the initial state.
This scalar solution can be extended to linear systems of
conservation equations
\begin{equation}
\partial_t{\bf q} + A \partial_x {\bf q} = 0,
\label{eq:lin_sys}
\end{equation}
where $A$, an $M\times M$ {\it constant} matrix, is, by
assumption, diagonalizable, with
real eigenvalues, $\lambda_\mu$.
(Greek indices take the values $1, \ldots , M$.)
Let $R$ be the matrix of right eigenvectors, ${\bf r}_\mu$, of $A$:
\begin{equation}
R \equiv [{\bf r}_1|\ldots|{\bf r}_M],
\end{equation}
and let $\Lambda$ be the diagonal matrix:
\begin{equation}
\Lambda \equiv \hbox{diag}[\lambda_1,\ldots,\lambda_M].
\end{equation}
We then have
\begin{equation}
A = R\Lambda R^{-1},
\end{equation}
and the solution of the system may be obtained by introducing
``characteristic variables'', ${\bf v}$:
\begin{equation}
{\bf v} = R^{-1}{\bf q}.
\end{equation}
Using characteristic variables, the equations \eref{eq:lin_sys} decouple
into a set of scalar advection equations
\begin{equation}
\partial_t {\bf v} + \Lambda\partial_x\,{\bf v} = 0,
\end{equation}
which can be immediately solved via:
\begin{equation}
{\bf v}_\mu(x,t) = {\bf v}_\mu(x - \lambda_\mu t,0).
\end{equation}
Given ${\bf v}(x,t)$,
the transformation ${\bf q} = R {\bf v}$ then produces the solution
of~~\eref{eq:lin_sys} in terms of the original variables, ${\bf q}$.
Turning now to the nonlinear case, the key idea is to first write the
nonlinear system in {\em quasilinear} form
\begin{equation}
\partial_t {\bf q} + {\cal A}({\bf q})\,\partial_x {\bf q} = 0.
\label{eq:lincons}
\end{equation}
Here, $\cal A$ is an $M\times M$ matrix which is now a function of ${\bf q}$.
Roe~\cite{roe} gives three specific criteria for the construction of $\cal A$:
\begin{enumerate}
\item ${\cal A}(\tilde{\bfq}^\ell,\tilde{\bfq}^r)\left( \tilde{\bfq}^r - \tilde{\bfq}^\ell\right) =
{\bf f}(\tilde{\bfq}^r) - {\bf f}(\tilde{\bfq}^\ell)$;
\item ${\cal A}(\tilde{\bfq}^\ell,\tilde{\bfq}^r)$ is diagonalizable with real eigenvalues;
\item ${\cal A}(\tilde{\bfq}^\ell,\tilde{\bfq}^r) \to {\bf f}'({\bf q})$ smoothly as
$\tilde{\bfq}^\ell,\tilde{\bfq}^r \to {\bf q}$.
\end{enumerate}
The latter two criteria can generally be satisfied by letting $\cal A$ be
the Jacobian matrix evaluated using the arithmetic average of the
conservation variables at the interface:
\begin{equation}
{\cal A} = \frac{\partial {\bf f}({\bf q})}{\partial {\bf q}}\bigg\vert_{{\bf q}=\bar{\bfq}_{i+1/2}},
\label{eq:jacobian}
\end{equation}
where
\begin{equation}
\bar{\bfq}_{i+1/2} = \frac{1}{2}\left( \tilde{\bfq}^\ell_{i+1/2} + \tilde{\bfq}^r_{i+1/2} \right).
\end{equation}
While this construction does not generally satisfy the first criterion,
(\ref{eq:jacobian}) is often used
in relativistic fluid dynamics~(see for example~\cite{ibanez,romero,font})
on the basis of its relative simplicity, and we also adopt this
approach.
On the other hand, other authors~\cite{eulderink} have constructed a
linearized Riemann solver for relativistic fluids with true Roe averaging, and
we therefore refer to our scheme as a ``quasi-Roe'' method.
Having defined a specific linearization, the scheme proceeds by
evaluation of ${\cal A}(\bar{\bfq}_{i+1/2})$---which is now viewed as a
matrix with (piecewise) constant coefficients---followed by the solution
of the Riemann problem for the resulting
linear system.
Carrying through an analysis not given here~(see e.g.~\cite{rjlbook}),
the Roe flux can be defined as
\begin{equation}
{\bf F}_{i+1/2} = \frac 1 2 \left[ {\bf f}(\tilde{\bfq}^r_{i+1/2}) + {\bf f}(\tilde{\bfq}^\ell_{i+1/2})
- \sum_\mu |\lambda_\mu|\triangle\bomega_\mu {\bf r}_\mu \right].
\label{eq:roeflux}
\end{equation}
where, again, $\lambda_\mu$ and ${\bf r}_\mu$ are the eigenvalues and
(right) eigenvectors,
respectively, of ${\cal A}({\bf q}_{i+1/2})$.
The quantities $\triangle\bomega_\mu$ are defined in terms of the
the jumps in the fluid variables across the interface
\begin{equation}
\tilde{\bfq}^r_{i+1/2} - \tilde{\bfq}^\ell_{i+1/2} = \sum_\mu \triangle\bomega_\mu {\bf r}_\mu .
\end{equation}
For completeness,
we give explicit expressions for the eigenvectors and eigenvalues of
the ultrarelativistic fluid system~\eref{eq:eom_new_vars} in \ref{sec:eigen}.
Finally, it is important to remember that approximate Riemann
solvers produce approximate solutions, which, under certain conditions,
may diverge from the physical solutions.
For example,
concentrating on the Roe solver,
Quirk~\cite{quirk} has recently reviewed
several ``subtle flaws'' in approximate solvers.
Fortunately, the approximate solvers often fail
in different ways, and where one solver produces an
unphysical solution, another solver may give the physical solution.
Thus, it may be necessary to investigate a particular problem with
multiple approximate Riemann solvers. Therefore, we have also
implemented Marquina's solver~\cite{marquina}, an alternative linear solver
that has also found application in relativistic fluid
studies~\cite{donat,font}, as an option in our code.
In addition to using the quasi-Roe and Marquina
solvers to investigate the critical collapse of perfect fluids,
we also implemented the HLLE solver in an independent code.
We found that the quasi-Roe solver gave accurate solutions, and
provided the best combination
of resolution and efficiency for the critical collapse problem.
Consequently, the results presented in \cite{dnmc2}
were obtained with this solver.
\section{Solving the Einstein/fluid system}
\label{sec:solving}
This section deals with some details of our numerical
solution of the coupled Einstein/fluid equations, including
the incorporation of source terms into our conservation laws,
regularity and boundary conditions,
and methods for calculating physical values for ${\bf w}$ in
the ultrarelativistic regime.
In addition, we describe the initial data and mesh structure
we have used in our studies of critical phenomena in fluid
collapse. Finally, we conclude the section with some remarks
on how we have tested and validated our code.
\subsection{Time integration}
\label{sec:time}
In \sref{sec:newvars}
the fluid equations of motion were written essentially in
conservation form, except that a source term, $\bpsi$,
had to be included.
While this source term clearly
breaks the strict conservation form of the equations, it can
be self-consistently incorporated into our numerical scheme by
using the method of lines to discretize space and time
separately.
Specifically,
the discretized fluid equations become
\begin{equation}
\frac{d\bar{\bfq}_i}{dt} = -\frac{1}{r^2_i\triangle r}
\left[ \left( r^2 X {\bf F}\right)_{i+1/2}
- \left( r^2 X {\bf F}\right)_{i-1/2}\right] + \bpsi(\bar{\bfq}_i),
\label{eq:update}
\end{equation}
where $\bar{\bfq}_i$ is the cellular average of ${\bf q}$, ${\bf F}_{i\pm1/2}$
are the numerical
fluxes defined by~(\ref{eq:roeflux}), and $X=\alpha/a$, as previously.
These equations can be integrated in time using standard techniques
for ODEs. In particular, Shu and Osher~\cite{shuosher} have investigated
different
ODE integration methods, and have found that the modified Euler method
(or Huen's method) is the optimal second-order scheme
consistent with the Courant condition required for a stabile evolution.
We briefly digress to define this scheme for a general set of
differential equations of the form
\begin{equation}
\frac{\rmd{\bf q}}{\rmd t} = L({\bf q}),
\end{equation}
where $L$ is a spatial differential operator.
Let ${\bf q}^n$ be the {\it discretized} solution at time $t=n\triangle t$,
and $\hat L$ be the discretized differential operator.
The modified Euler method
is a predictor-corrector method, with predictor
\begin{equation}
{\bf q}^* = {\bf q}^n + \triangle t\, \hat L({\bf q}^n),
\end{equation}
and corrector
\begin{equation}
{\bf q}^{n+1} = \frac{1}{2}({\bf q}^n + {\bf q}^*) + \frac{1}{2}\triangle t\,
\hat L({\bf q}^*).
\end{equation}
Again, we note that $\triangle t$ is subject to a Courant (CFL) condition,
which can be deduced empirically or possibly from a linearized stability
analysis.
Particularly in comparison to the treatment of the fluid equations,
numerical solution of the equations governing the geometric
quantities $\alpha$ and $a$ is straightforward.
As discussed previously,
the lapse, $\alpha$, is fixed by the
polar slicing
condition \eref{eq:polar_slicing},
while $a$ can be found from either the Hamiltonian
\eref{eq:hamcon} or momentum \eref{eq:momcon} constraints.
We have used discrete, second-order, versions of both equations for
$a$, and have obtained satisfactory results in both cases (the polar
slicing equation is likewise solved using a second-order scheme.)
In general, however, (and
particularly on vector machines) solution via the momentum
constraint yields a far more efficient scheme, and we thus generally use
the momentum equation to update $a$.
Full
details of our numerical scheme are
presented in \ref{sec:implementation}.
\subsection{Regularity and boundary conditions}
\label{sec:regbound}
In the polar-areal coordinate system, the lapse ``collapses'' exponentially
near an apparent horizon, preventing the $t=\,$constant surfaces from
intersecting the physical singularity which must develop interior to
a black hole. As the slices ``avoid'' the singularity,
elementary flatness holds at the origin for all times in the
evolution, giving
\begin{equation}
a(0,t) = 1.
\end{equation}
At each instant of time,
the polar-slicing condition~(\ref{eq:polar_slicing}) determines the
lapse only up to an overall multiplicative constant, reflecting the
reparameterization invariance, $t \to {\tilde t}(t)$, of the polar
slices.
We chose to normalize the lapse function so that as $r\to\infty$,
coordinate time corresponds to proper time.
On a finite computational domain, and provided no matter out-fluxes
from the domain, this condition is approximated via
\begin{equation}
\alpha a\Big|_{r_{\rm{max}}} = 1.
\end{equation}
In spherical symmetry the fluid flows along radial lines, and given
that there are no sources or sinks at the origin, we have that
$v(0,t) = S(0,t) = 0$. Thus
\begin{equation}
\Pi(0,t) = \Phi(0,t) = \tau(0,t).
\end{equation}
Regularity at the origin further require $\tau, \Pi$ and $\Phi$ to
have even expansions in $r$ as $r\to 0$:
\begin{equation}
\tau(r,t) = \tau_0(t) + r^2 \tau_2(t) + \Or(r^4)
\end{equation}
\begin{equation}
\Pi(r,t) = \Pi_0(t) + r^2 \Pi_2(t) + \Or(r^4) =
\tau_0(t) + r^2 \Pi_2(t) + \Or(r^4)
\end{equation}
\begin{equation}
\Phi(r,t) = \Phi_0(t) + r^2 \Phi_2(t) + \Or(r^4) =
\tau_0(t) + r^2 \Phi_2(t) + \Or(r^4)
\label{eq:Phi_expansion}
\end{equation}
On our radial grid $r_i, \,\, i = 1, 2, \cdots N$,
we use these expansions to compute grid-function values defined at
$r=r_1=0$ in terms of
values defined at $r=r_2$ and $r=r_3$.
Specifically, once the values $\Phi_2$ and
$\Phi_3$ have been updated via the equations of motion, we compute
$\Phi_1$ using a ``quadratic fit'' based on the
expansion~(\ref{eq:Phi_expansion}):
\begin{equation}
\Phi_1 = \frac{\Phi_2 r_3{}^2 - \Phi_3 r_2{}^2}%
{r_3{}^2 - r_2{}^2}.
\end{equation}
We then set $\Pi_1 = \Phi_1$.
At the outer boundary we apply out-flow boundary conditions, which in
our case are simply first-order extrapolations for $\Pi$ and $\Phi$:
\begin{equation}
\Phi_N = \Phi_{N-1}\qquad\Pi_N = \Pi_{N-1}.
\end{equation}
In addition, two ghost cells ($r=r_{N+1}, r=r_{N+2}$)
are added at the outer edge of the grid for ease in coding the
cellular reconstruction algorithm~\cite{rjl98}.
These ghost cells are also updated with first-order extrapolation.
\subsection{Floor}
\label{sec:floor}
The fluid model
is a continuum approximation, and, at least na\"\i vely, the fluid equations
become singular as $\rho \rightarrow 0$.
In these evacuation regions, both
the momentum and mass density are very small, and therefore the
velocity---which loosely speaking is the quotient of the
two---is prone to fractionally large numerical errors.
These errors then often result in the computation of unphysical
values for the fluid variables, such as supraluminal velocities,
negative pressures or negative energies.
(In addition, of course, our code must contend with the usual
discretization and round-off errors common to any numerical solution
of a set of PDEs.) At least from the point of view of Eulerian
fluid dynamics, it seems fair to say that a completely satisfactory
resolution of the evacuation problem does not exist. In the
absence of a mathematically rigorous and physically acceptable
procedure, we adopt the {\em ad hoc} approach of demanding that
$\rho > 0$ everywhere
on the computational domain, i.e.\ we exclude the possibility
that vacuum regions can form on the grid.
In terms of our conservation
variables ${\bf q}$, this requirement becomes $\Pi > 0$ and $\Phi > 0$.
In a wide variety of situations, our numerical solutions of the fluid
equations naturally satisfy these constraints.
However, the critical
solutions for ``stiff'' equations of state ($\Gamma \gtrsim 1.9$)
develop extremely relativistic velocities ($W > 10^6$) in regions
where $\rho$ is small~\cite{dnmc2}, and we are unable to solve
the fluid PDEs in these cases without imposing
{\em floor} (or minimum) values on ${\bf q}$. Specifically, at each step
in the integration we require
\begin{equation}
\Pi \ge \delta,\quad \Phi \ge \delta,
\end{equation}
where the floor $\delta$ is chosen to be several orders of
magnitude smaller than the density associated with what we
feel are the physically relevant
features of the solution---a typical value is $\delta = 10^{-10}$.
The floor is often applied in regions where $\Pi$ and $\Phi$ differ
greatly in magnitude, and discretization errors can easily lead to
the calculation of a negative value for either function.
For example, the floor may be applied to the ``in-going'' function
in a region where the fluid is overwhelmingly ``out-going.''
In these cases, the effect of the
floor is dynamically unimportant. However, the floor may be
invoked in other cases, where its effect on the dynamics is less certain.
Given the {\em ad hoc} nature of this regularization procedure,
the crucial question
is whether the floor affects the computed solutions in a substantial way.
We investigated this question by comparing critical solutions
for $\Gamma=2$ (the most extreme case) which were
calculated with two distinct floor values:
$\delta = 10^{-8}$, and our usual $\delta = 10^{-10}$.
The two solutions appeared identical,
and identical mass-scaling exponents~\cite{dnmc2} were calculated.
However, we note that the use of a floor makes estimates of the
maximum Lorentz factor
attained in the critical solutions
unreliable because the largest velocities occur in regions where
the floor is enforced.
\subsection{Calculating the velocity}
\label{sec:velocity}
The simple expression~\eref{eq:vdef} for $v$ in terms of ${\bf q}$,
when used na\"\i vely with finite precision arithmetic,
can result in the computation of unphysical, supraluminal
velocities.
For example, when searching for critical solutions we routinely
calculate fluid flows with $W\gtrsim 10^3$.
Thus, when calculating $v$ from
the quotient~\eref{eq:vdef}, small numerical errors can easily
conspire to give $|v|>1$, rather than the correct $|v|\gtrsim 0.999999$.
On the other hand, the combination
\begin{equation}
\chi \equiv W^2 v
\end{equation}
is insensitive to small numerical errors, and provides a better
avenue for calculating $v$ from the conservation variables.
From the definition \eref{eq:tsdef} of $S$ we have
\begin{equation}
\chi = \frac{(\Gamma - 1)}{\Gamma}\frac{S}{P}.
\end{equation}
The velocity can then be calculated from $\chi$ using
\begin{equation}
\label{eq:vchi}
v = \frac{1}{2\chi}\left(\sqrt{1 + 4\chi^2} - 1\right) .
\end{equation}
To the limit of machine
precision, $v$ is then in the physical range $-1 < v < 1$.
When $\chi \ll 1$, we calculate $v$ from a Taylor expansion of \eref{eq:vchi},
although \eref{eq:vdef} could also be used.
We also use $\chi$ when calculating ${\bf w}$ from ${\bf q}$ for the
ideal-gas EOS~\eref{eq:igeos}.
\subsection{Grid}
\label{sec:grid}
The black-hole-threshold critical solutions---which are our primary focus---are
generically self-similar, and as such, require
essentially unbounded dynamical range for accurate simulation.
Thus some sort of adaptivity in the construction
of the computational domain is crucial, and, indeed, the
earliest studies of critical collapse~\cite{mwc93} used
Berger-Oliger adaptive mesh refinement~\cite{bergoliger} to great
advantage.
However, in contrast to the early work, we know (at least schematically)
the character of the critical solutions we seek, and thus we can,
and have, used
this information to construct a simple, yet effective, adaptive grid method.
(Our approach is similar in spirit to
that adopted by Garfinkle~\cite{garfinkle} in his study of scalar field
collapse.)
Specifically, at any time during the integration our spatial grid has
three distinct domains: the two regions near $r=0$ and $r=r_{\rm max}$ have
uniform grid spacings (but the spacing near $r=0$ is typically
{\em much} smaller than that near the outer edge of the computational
domain), and the intermediate region has grid points distributed uniformly
in $\log(r)$ (see \fref{fig:grid}).
As a near-critical solution propagates to smaller spatial scales,
additional grid points are added in order to maintain some given
number of grid points between $r=0$ and some identifiable
feature of the critical solution. For example, we typically require that at
least 300 or so grid points
lie between the origin and the maximum of the profile of the
metric function $a$.
The primary advantage of this gridding scheme is that it
is simple to implement, and yet allows us to resolve detail over
many length scales: the ratio of the grid
spacing at the outer edge to the spacing at the origin is typically
$10^{10}$--$10^{13}$ at the end of an evolution.
The primary disadvantage of this scheme is that it is
specialized for critical collapse, and cannot be used for more general
physical problems.
\begin{figure}
\epsfxsize = 7.5cm
\centerline{\epsffile{grid.ps}}
\caption{Illustration of the re-meshing algorithm used in
investigations of critical collapse. The grid spacing
$\triangle r$ is shown as a function of $r$ on
a log-log plot.
The solid line represents the initial grid, the dotted line shows
the grid spacing after the first addition of points near the origin,
and
the dashed line shows the grid spacing after the second regridding.
Note that the grid spacings near the origin, and near the outer
edge of the computational domain are uniform (horizontal lines).
At each regridding cycle, the grid spacing near
the origin is halved, and the new points are
matched smoothly onto the previous grid.
A critical evolution may involve more than 20 regriddings, although only
a small number of points (50--150) may be added at a time.
}
\label{fig:grid}
\end{figure}
\subsection{Initial data for critical solutions}
We expect that the critical solutions in fluid collapse
will be universal, in the sense that {\em any} family of initial
data which generates families that ``interpolate'' between
complete dispersal and black hole formation, should exhibit the
same solution at the black hole threshold.
We have thus focused attention
on a specific form of initial data, which generates
initially imploding (or imploding/exploding) shells of fluid.
Specifically, the energy density
in the shells has a Gaussian profile,
\begin{equation}
\tau = \tau_o \exp\left[ -(r - r_o)^2 / \Delta^2\right] + K,
\end{equation}
where the constant $K$---typically of magnitude $10^{-6}\tau_o$---represents
a constant ``background''.
It should be note that
this background is used only in setting the
initial data, and is not held fixed during the evolution---in particular
$K$ is {\it not} a floor as discussed in \sref{sec:floor}.
The shells are either time-symmetric, or have an initial inward velocity
which is proportional to $r$. Critical solutions were found by
fixing $r_o$ and $\Delta$, and then tuning the pulse amplitude $\tau_o$.
\subsection{Tests}
\label{sec:tests}
\begin{figure}
\epsfxsize = 10cm
\centerline{\epsffile{conv3.ps}}
\caption{
Illustration of some of
the convergence
properties of the solution algorithm discussed in the text.
Here
we evolved a time-symmetric shell of
fluid ($\Gamma = 1.3$)
using uniform grids with three different resolutions: $\triangle r = h, 2h$ and
$4h$.
Convergence is investigated by
comparing the solutions obtained using the three distinct discretization
scales.
In frame~(c), the solid line is $\left(\tau_{2h} - \tau_{4h}\right)$
and the dotted line is $4\left(\tau_h - \tau_{2h}\right)$,
where the subscript on $\tau$ indicates the grid spacing
for a particular solution.
When the convergence is second order, the two lines should
(roughly) coincide, while when the convergence is first order, the amplitude of
the dotted line should be twice that of the solid line.
As expected, we see that the convergence is not second order at the
shock. (Of course the whole notion of convergence at a discontinuity
fails, as the notion of Richardson expansion requires smooth
functions.) However,
we also can see that the convergence is only first order at the
extrema of ${\bf q}$---at these points, the slope changes sign, and the
minmod limiter produces a first-order reconstruction. Frame~(d) shows
a more detailed view of a portion of the data displayed in (c).
For context, we also show $\tau$ in frame~(a) and $v$ in frame~(b).
}
\label{fig:conv3}
\end{figure}
When developing a code such as the one described here, there are
a number of tests which should be
passed in order to provide confidence that the algorithm is
producing reliable results.
Perhaps most fundamental of these is the convergence test, which
generally demonstrates that the numerical method is consistent and
has been correctly implemented, but which also provides an
{\em intrinsic} method for estimating the level of error in
a given numerical solution.
For our high-resolution shock-capturing scheme,
a general rule-of-thumb is that the convergence
should be (apparently) second order where the flow is
smooth, and first order at discontinuities,
where the effects of the slope limiter become important.
In addition, we can also expect first order convergence
near extrema of $\bar{\bfq}$, since at these points, the slope, ${\bf s}$,
changes sign, and the minmod limiter gives a piece-wise constant
reconstruction for $\tilde{\bfq}$.
A convergence test where these
effects are apparent is shown in \fref{fig:conv3}.
After the numerical algorithm has been correctly implemented,
one often compares results from the code to known closed-form
solutions.
In the early stages of code development, we tested
the shock-capturing algorithm in this fashion by
solving initial data for a shock tube, and comparing the
results with the known solution of the Riemann problem.
While the shock-tube provides a good
test of the fluid solver, the test is done in Minkowski space with
slab symmetry,
and can probe neither the implementation of the geometric factors in the
fluid equations, nor the discretized Einstein equations.
A few general relativistic fluid systems {\em can} be solved exactly,
and have traditionally been used to test new codes. These include
static, spherical stars (Tolman-Oppenheimer-Volkoff), spherical
dust collapse (Oppenheimer-Synder) and ``cosmological'' tests with
a Robertson-Walker metric. In our companion paper~\cite{dnmc2}, we advocate
the use of perfect fluid critical solutions as an additional
test problem; one which involves both dynamic gravitational
fields and highly relativistic fluid flows.
Thus we consider the ultimate test of our full GR/fluid
code to be the dynamical calculation of self-similar
perfect fluid critical solutions, which can then be compared to
solutions computed directly (but also numerically!) from
a self-similar {\em ansatz}~\cite{dnmc2}.
|
\section{Introduction}
Although color-magnitude diagrams have now been obtained for a few
very close galaxies -- most of them dwarf --, allowing to study
directly their stellar populations (e.g. Schulte-Ladbeck \& Hopp
\cite{SLH}; Aparicio \cite{Aparicio}), the synthesis of the spectral
energy distribution (SED) of nearby galaxies remains the most
efficient and systematic method to trace their star formation history
up to $z=0$.
Early optical studies have shown a clear correlation between the
SED -- and hence the star formation history -- and the galaxy Hubble type.
However, the age-star formation timescale and age-metallicity
degeneracies make it highly desirable to extend the wavelength range
to the ultraviolet (UV) and the near-infrared. Whereas the ultraviolet
is related to the young stellar populations in currently star-forming galaxies,
the near-infrared is dominated by old giant stars and is a measure of the star
formation rate integrated from the beginning, i.e. of the stellar
mass of the galaxy. When compared to shorter wavelengths, the NIR
may moreover put constraints on the stellar metallicity, owing to the
high sensitivity of the effective temperature of red giants
to the abundance of heavy elements.
In combination with optical data, it finally provides clues on the amount and the
distribution of dust because of the different extinction at both wavelengths.
Observations in the UV and the NIR are however hampered by the atmosphere.
Though interesting to study the
detailed star formation history of a specific galaxy, the spectra
obtained from Space are too scarce to afford a statistical analysis
from which the general trends as a function of type or mass, or the
effect of the dust, may be derived.
In spite of the loss of spectral resolution,
broad-band colors in the $J$, $H$ and $K$ atmospheric windows
are more promising.
As the spectra, they are usually obtained in small apertures
not representative of the whole galaxy
because of the color gradients, especially for disk galaxies,
and have to be extrapolated to be compared to the optical data.
Such extrapolation is however much easier for colors than for spectra
and can be applied to hundreds of objects.
In the following, we analyze a catalog of infrared aperture magnitudes
(section~\ref{catalogue}) and combine it with optical catalogs to
derive NIR growth curves of the magnitude as a function of the
aperture (section~\ref{croissance}).
We then compute total and effective magnitudes and colors (section~\ref{magnitudes}).
The NIR and optical-to-NIR colors are analyzed statistically as a function of
type, luminosity and inclination in section~\ref{statistiques}. We finally discuss
our results in section~\ref{discussion}.
\section{The optical and infrared data}
\label{catalogue}
The near-infrared data used in this paper come from the CIO catalog
(Gezari et al. \cite{Gezari}) which is a compilation
of the NIR observations published before 1995.
They are given in a large variety of systems (magnitudes $m_{\lambda}$,
$F_{\lambda}$, $F_{\nu}$, etc.)
and units, and are
first converted to magnitudes assuming Vega has a magnitude of 0.03.
The data are
reduced from the wavelength $\lambda$
to the central wavelengths of the $J$ ($\lambda_{\mathrm{J}}=1.24\mu\mathrm{m}$),
$H$ ($\lambda_{\mathrm{H}}=1.65\mu\mathrm{m}$) and $K$ ($\lambda_{\mathrm{K}}=2.21\mu\mathrm{m}$) filters of
Bessel \& Brett (\cite{BB}), assuming the following color equations:
\[
m_{\mathrm{J}}=m_{\lambda}+(\lambda_{\mathrm{J}}-\lambda)\frac{<J-H>}{\lambda_{\mathrm{J}}-\lambda_{\mathrm{H}}},\]
\begin{eqnarray*}
m_{\mathrm{H}}&=&m_{\lambda}+(\lambda_{\mathrm{H}}-\lambda)\frac{<J-H>}{\lambda_{\mathrm{J}}-\lambda_{\mathrm{H}}}
\mathrm{\qquad{}if\;}\lambda<1.65\mu\mathrm{m},\\
&=&m_{\lambda}+(\lambda_{\mathrm{H}}-\lambda)\frac{<H-K>}{\lambda_{\mathrm{H}}-\lambda_{\mathrm{K}}}
\mathrm{\qquad{}if\;}\lambda>1.65\mu\mathrm{m},
\end{eqnarray*}
\[m_{\mathrm{K}}=m_{\lambda}+(\lambda_{\mathrm{K}}-\lambda)\frac{<H-K>}{\lambda_{\mathrm{H}}-\lambda_{\mathrm{K}}},\]
where $<J-H>=0.7$ and $<H-K>=0.2$ are the typical $J-H$ and $H-K$
colors of a normal galaxy and depend very little on the galaxy type
or the aperture.
The CIO catalog is cross-correlated with the RC3 catalog (de
Vaucouleurs et al. \cite{RC3}) to
correct for the Galactic extinction. The NIR extinction is computed
from $A_{\mathrm{B}}$ as follows: $A_{\mathrm{J}}=0.182 A_{\mathrm{B}},\;A_{\mathrm{H}}=0.098 A_{\mathrm{B}},\;A_{\mathrm{K}}=0.068 A_{\mathrm{B}}$.
The magnitudes are also corrected for the redshift $z$
as $m_{\lambda}(0)=m_{\lambda}(z)-k_{\lambda}(z)$.
Values of $k_{\lambda}(z)$ are given in Table~\ref{corrk}
for different types of galaxies
and have been computed using the \textsc{p\'egase} model of
spectral evolution (Fioc \& Rocca-Volmerange \cite{FRV}).
They should be used only at $z<0.1$.
The redshift is taken from the NED database.
For a few galaxies, $z$ is unknown and we compute the $k$-correction
(but not the absolute magnitudes)
assuming $z=0.01$, the mean redshift of our catalog.
\begin{table}
\begin{center}
\begin{tabular}{l|c|c|c|c}
type&$k_{\mathrm{B}}$ & $k_{\mathrm{J}}$ &$k_{\mathrm{H}}$ &$k_{\mathrm{K}}$ \\
\hline
E&$5.1z$ & $-0.3z$ &$-0.1z$ &$-2.8z$ \\
S0&$4.9z$ & $-0.3z$ &$-0.1z$ &$-2.8z$ \\
Sa&$4.0z$ & $-0.3z$ &$-0.1z$ &$-2.8z$ \\
Sb&$3.6z$ & $-0.3z$ &$-0.1z$ &$-2.8z$ \\
Sc&$2.6z$ & $-0.3z$ &$-0.1z$ &$-2.8z$ \\
Sd&$2.2z$ & $-0.5z$ &$-0.3z$ &$-2.8z$ \\
Im&$1.8z$ & $-0.7z$ &$-0.4z$ &$-2.8z$
\end{tabular}
\caption{$k$-corrections in $B$, $J$, $H$ and $K$ as a function of the
type.}
\label{corrk}
\end{center}
\end{table}
The RC3 provides us also with the morphological type $T$,
the ratio $R_{25}$ of the major axis ($D_{25}$) to the minor axis
of the ellipsis corresponding to the isophot $\mu=25\mathrm{mag/arcsec}^2$,
the total $B$-magnitude $B_{\mathrm{T}}$ extrapolated to an infinite radius
and the circular effective aperture $A_{\mathrm{e}}$
containing half the light emitted in the $B$-band.
When available, we prefer to take $B_{\mathrm{T}}$ and $A_{\mathrm{e}}$
from the more recent catalog Hypercat (Prugniel \& H\'eraudeau
\cite{PH}).
This catalog
also gives the photometric type $T_{\mathrm{p}}$ corresponding to the best-fitting
growth curve ${\cal B}(X,T_{\mathrm{p}})$ of the $B$-magnitude as a function of
the circular aperture $A$, where $X=\log_{10}(A/A_{\mathrm{e}})$ (see Appendix~A).
\section{The NIR growth curve}
\label{croissance}
The detailed 2D-fitting of the profile of galaxies (de Jong
\& van der Kruit \cite{dJvdK})
is certainly the best way to compute their total magnitudes, but
surface photometry in the NIR is available for too few
of them to perform a statistical analysis. The largest such study
is based on only 86 spiral galaxies of all types
(de Jong \cite{dJa,dJb,dJc}), to compare to
typically 100 galaxies {\em per type} in this paper.
We therefore prefer
to extrapolate aperture magnitudes with a growth curve.
The computation of total (i.e. asymptotic)
NIR magnitudes comparable to the optical
ones determined by, e.g., de Vaucouleurs et al. (\cite{RC3})
or Prugniel \& H\'eraudeau (\cite{PH}) is however made difficult by the
small apertures achieved at these wavelengths. Few galaxies
have enough data from small to large apertures to constrain
the shape of the growth curve and one has to use another method,
combining observations of different galaxies having presumably
the same curve (cf. Griersmith \cite{Griersmith}).
To this purpose, one needs to scale the apertures to a
characteristic length $A_{\mathrm{c}}$ for each galaxy. The NIR growth
curve
${\cal M}(A/A_{\mathrm{c}})$
is then built by plotting $m(A)-m(A_{\mathrm{c}})$ as a function of
$\log_{10}(A/A_{\mathrm{c}})$.
Usually (Gavazzi et al. \cite{Gavazzi}; Gavazzi \& Boselli \cite{GB};
Tormen \& Burstein \cite{TB}; Frogel et al. \cite{Frogel}; Aaronson et
al. \cite{Aaronson}), the $D_{25}$
diameter has been used as characteristic length. As noticed
however by de Vaucouleurs et al. (\cite{RC2}), the $D_{25}$ diameter depends not only
on the shape of the profile but also on the central surface
brightness. For this reason, the ratio of the effective aperture $A_{\mathrm{e}}$ to $D_{25}$
is not a constant.
We prefer to adopt here an other procedure. We may expect some
relation between the optical profile and the NIR profile.
For this reason, we decide to build a growth curve as a function
of the $B$ {\em photometric} type $T_{\mathrm{p}}$ taken from Hypercat and of
$X=\log_{10}(A/A_{\mathrm{e}})$. The infrared magnitude $m(A_{\mathrm{e}})$ in the $B$
effective aperture is computed by interpolation
or by a small extrapolation in $\log_{10}(A)$.
Rather than a straight line,
we have used the functions proposed to compute
$H$ magnitudes at $A=D_{25}$ by Gavazzi et al. (\cite{Gavazzi}):
\[{\cal
M}(X_{25})=a+b(-0.35X_{25}+1.68X_{25}^2+0.07X_{25}^3)\]
where $X_{25}=\log_{10}(A/D_{25})$,
and fitted $a$ and $b$ to the observations.
This takes into account the curvature of the growth curve
near $A_{\mathrm{e}}$ and improves the determination of $m(A_{\mathrm{e}})$.
Because they are polynomials, these functions are however not suitable
to extrapolate the magnitude to the infinity.
The $J$, $H$ and $K$ data have been combined to build the growth curve.
This is justified by the fact that only small $J-H$ and $H-K$ color gradients
are observed (Aaronson \cite{Aaronsonthese}; Frogel et al. \cite{Frogel}).
\begin{figure*}
\begin{center}
\resizebox{!}{21cm}{\rotatebox{0}{\includegraphics{profilB_JHKarticle.ps}}}
\caption{{\bf (a) to (g):} Growth curve of the NIR magnitude as a
function of the aperture for the different photometric types.
Hollow and filled triangles correspond respectively
to galaxies with $\log_{10} R_{25}<0.3$ and
$\log_{10} R_{25}>0.3$. The solid and dashed lines are the mean optical and
NIR growth curves. They are
normalized to the effective aperture in the $B$-band and to the
corresponding magnitude.
{\bf (h):} The circles and vertical segments are the mean value $s_0$
and the intrinsic scatter $\sigma_{s_0}$ of $s$ as a function of the
median value of $T_{\mathrm{p}}$ in each bin. The broken line
is the 3-slope fit to $s_0(T_{\mathrm{p}})$ given in the text.}
\label{magouv}
\end{center}
\end{figure*}
The growth curves are plotted in Fig.~\ref{magouv} for 7 bins of
photometric type. As evidenced by this graph, the NIR growth curve
is flatter than the optical one, especially for $T_{\mathrm{p}}$ corresponding
to intermediate spirals: the NIR emission of galaxies is
more concentrated than in the optical.
This behavior is expected because the bulge is redder than the
disk and is therefore more prominent in the NIR than in the optical.
The scatter is also higher for intermediate types, which
might be due to an inclination dependency of the growth curve (Christensen
\cite{Christensen}; Kodaira et al. \cite{Kodaira}). We have therefore
distinguished in Fig.~\ref{magouv} between galaxies with $R_{25}<0.3$
(rather face-on for disk galaxies) and $R_{25}>0.3$, but no obvious trend
is observed and we will neglect any inclination dependency in the following.
Because of the scatter and the lack of data at high aperture,
where the flattening of the growth curve puts constraints
on the curvature,
attempts to fit growth curves similar to those of
Prugniel \& H\'eraudeau (\cite{PH}) with 3 parameters
($m_{\mathrm{T}}$, $\log_{10}(A_{\mathrm{e}})$ and $T_{\mathrm{p}}$)
did not prove successful.
Plotting the NIR magnitudes versus the $B$ magnitudes
however reveals a nearly linear -- though scattered --
relation, between these quantities, suggesting
to adopt a NIR growth curve of this form:
\[{\cal M}(X,T_{\mathrm{p}})=s{\cal B}(X,T_{\mathrm{p}}).\]
This ensures that the extrapolation at infinite $A$
converges since
$\lim_{X\rightarrow\infty}{\cal B}(X,T_{\mathrm{p}})=0$.
Such relation has been fitted for each photometric type bin (Fig.~\ref{magouv},
dashed lines) at $X>-1.5$.
The dispersion is due both to the uncertainties in the
individual data and to the intrinsic scatter in the shape of the
NIR growth curves for a given $B$ photometric type.
The following convention is adopted hereafter:
{\em $x \sim {\cal N}(\mu,\sigma^2)$} means that {\em $x$ is distributed
according to (or the density probability of $x$ is) a Gaussian
with mean $\mu$ and variance $\sigma^2$}.
Let us assume that, for each galaxy, the distribution of individual
data around the best-fitting NIR growth curve
$s{\cal B}(X,T_{\mathrm{p}})$ is $\sim {\cal N}(0,\sigma^2_{m_0})$
and that the distribution of the parameter $s$ characterizing the
growth curve is $s\sim {\cal N}(s_0,\sigma^2_{s_0})$.
At any aperture,
\begin{eqnarray*}
m(X)-m(0)\sim {\cal N}\Bigl\lbrace\!\!&\!\!&\!\!s_0\bigl[{\cal B}(X,T_{\mathrm{p}})-{\cal B}(0,T_{\mathrm{p}})\bigr],\\
\!\!&\!\!&\!\! \sigma_{m_0}^2+\sigma_{s_0}^2\bigl[{\cal B}(X,T_{\mathrm{p}})-{\cal
B}(0,T_{\mathrm{p}})\bigr]^2
\Bigr\rbrace.
\end{eqnarray*}
A maximum likelihood estimation of these parameters yields
$\sigma_{m_0}\simeq 0.06$, nearly independent of $T_{\mathrm{p}}$,
$\sigma_{s_0}\simeq 0.10$ for early types ($T_{\mathrm{p}} < -1$) and
$\sigma_{s_0}\simeq 0.15$ at later types.
The value of $s_0$ we obtained is plotted as a
function of the median $T_{\mathrm{p}}$ of each bin in Fig.~\ref{magouv}
and may be approximated with the following formulae:
\begin{eqnarray*}
s_0&=&0.87\mathrm{\qquad{}if\;}T_{\mathrm{p}}<-4.5,\\
&=&0.74-0.029T_{\mathrm{p}}\mathrm{\qquad{}if\;}-4.5<T_{\mathrm{p}}<2.5,\\
&=&0.64+0.011T_{\mathrm{p}}\mathrm{\qquad{}if\;}T_{\mathrm{p}}>2.5.
\end{eqnarray*}
The mean value $s_0$ is less than one for all types, corresponding
as expected to a blue-outwards gradient. This gradient is low for
early-types, increases for spirals, peaking at $T_{\mathrm{p}}\in [2-4]$ ($\sim$~Sb)
in good agreement with what has been obtained at optical wavelengths
by de Vaucouleurs \& Corwin (\cite{dVC}) and remains constant or
slightly decreases for late types.
The typical blueing of the $B-\mathop{\mathrm{NIR}}$ color of spirals from the
effective aperture to the infinity is $-0.75(1-s)\sim -0.2$,
close to the median value ($-0.19$) in $B-H$ and $B-K$
we have computed from the profiles published by de Jong (\cite{dJa}).
\section{Effective and total near-infrared magnitudes}
\label{magnitudes}
\subsection{Apparent magnitudes and colors}
Effective and total near-infrared magnitudes have been computed
for all the galaxies with known effective aperture by fitting the growth curves $s{\cal B}(X,T_{\mathrm{p}})$
introduced in the previous section to
the observed data $(X_i, m_i, \sigma_i),\;i\in[1,n]$, where
$\sigma_i$ is the uncertainty on the magnitude $m_i$.
The asymptotic magnitude depends primarily on the shape of the growth curve
at large aperture. To avoid problems with small apertures suffering
from seeing or off-centering, or which are contaminated by nuclear
emission or a central starburst, we remove all the observations
at $X<-1$ from the fit.
The procedure to compute the
total and effective NIR magnitudes $m_{\mathrm{T}}$, $m_{\mathrm{e}}$
and the corresponding
uncertainties $\sigma(m_{\mathrm{T}})$, $\sigma(m_{\mathrm{e}})$
due to the uncertainties in the aperture magnitudes
is detailed in Appendix~B.
The global uncertainty $\varsigma_{\mathrm{m_{\mathrm{T}}}}$ on
$m_{\mathrm{T}}$ taking also into account the uncertainties
in $T_{\mathrm{p}}$ and $A_{\mathrm{e}}$ is finally computed from
\[\varsigma_{m_{\mathrm{T}}}^2=\sigma^2(m_{\mathrm{T}})+\left(\frac{\partial
m_{\mathrm{T}}}{\partial T_{\mathrm{p}}}\right)^2\sigma^2(T_{\mathrm{p}})+\left(\frac{\partial
m_{\mathrm{T}}}{\partial A_{\mathrm{e}}}\right)^2\sigma^2(A_{\mathrm{e}})\]
and the same for $\varsigma_{m_{\mathrm{e}}}$.
In many cases, the photometric type
is unknown and we estimate it from the morphological type
with the following relations derived by cross-correlating
Hypercat and the RC3:
\begin{eqnarray*}
T_{\mathrm{p}}&=&-2.3+0.5T\mathrm{\qquad{}if\;}T<-1,\\
&=&-1.32+1.48T\mathrm{\qquad{}if\;}-1<T<4,\\
&=&1.4+0.8T\mathrm{\qquad{}if\;}T>4.
\end{eqnarray*}
We then simply assume $\sigma(T_{\mathrm{p}})=3$.
The uncertainty on the colors $m_1-m_2$, where $(m_1,m_2)\in [J,H,K]^2$,
is computed as
\begin{eqnarray*}\varsigma^2_{m_1-m_2}=\sigma^2(m_1)+\sigma^2(m_2)
\!&\!+\!&\!\left[\frac{\partial
(m_1-m_2)}{\partial T_{\mathrm{p}}}\right]^2\sigma^2(T_{\mathrm{p}})\\\!&\!+\!&
\!\left[\frac{\partial
(m_1-m_2)}{\partial A_{\mathrm{e}}}\right]^2\sigma^2(A_{\mathrm{e}}).\end{eqnarray*}
However, this is clearly an overestimate because $J$, $H$ and $K$ data
have usually been observed simultaneously and their errors
are presumably highly positively correlated.
When $B$ is one of the band, e.g. $m_1=B_{\mathrm{T}}$,
the partial derivatives are unknown.
We then estimate $\varsigma_{m_1-m_2}$ as
\[\varsigma^2_{m_1-m_2}=\varsigma^2_{m_1}+\varsigma^2_{m_2}.\]
This will usually be a slight overestimate for the partial
derivatives of $B_{\mathrm{T}}$ and $m_2$ should have the same sign, but most
of the uncertainty comes actually from the NIR data themselves.
\subsection{Absolute magnitudes}
The absolute magnitude $M_{\mathrm{T}}$ is computed as
\[M_{\mathrm{T}}=m_{\mathrm{T}}-25-5\log_{10}(v_{\mathrm{V}}/H_0)\]
where $v_{\mathrm{V}}$ is the velocity of a galaxy derived
from the redshift $z$ and corrected for the movement of the
Sun in the restframe of the Virgo cluster according to the equations
(18), (19), (20), (31) and (32) from Paturel
et al. (\cite{Paturel}).
We assume $H_0=65\,\mathrm{km.s^{-1}.Mpc^{-1}}$.
The uncertainty on the absolute magnitude is computed
from
\[\varsigma^2_{M_{\mathrm{T}}}=\varsigma^2_{m_{\mathrm{T}}}+\left[\frac{5}{\ln(10)}\right]^2\frac{(\partial
v_{\mathrm{V}}/\partial z)^2\sigma^2(z)+v^2_{\mathrm{p}}}{v^2_{\mathrm{V}}}\]
where $v_{\mathrm{p}}=350\,\mathrm{km.s^{-1}}$ is the typical peculiar
velocity of galaxies in the Las Campanas Redshift Survey (Lin et
al. \cite{LCRS}). At low redshift, peculiar velocities perturb
the redshift-distance relation and increase the uncertainty on
the absolute magnitude.
All the absolute magnitudes of the galaxies with $[(\partial
v_{\mathrm{V}}/\partial
z)^2\sigma^2(z)+v^2_{\mathrm{p}}]^{1/2}>v_{\mathrm{V}}/3$ have been
discarded in the following.
$\sigma(z)$ has been computed according to the formulae (1) and (2) of
the Appendix A of Paturel et al. (\cite{Paturel}), taking into account
both the internal uncertainty on the values of the redshift and the
discrepancies between them.
Covariances between the colors and the absolute magnitudes have also
been computed.
\section{Statistics of optical-near infrared colors}
\label{statistiques}
\subsection{$J-H$, $H-K$ and $J-K$ colors}
The galaxies have been distributed in eight broad types
according to the morphological type $T$ taken from the RC3
(see Table~\ref{types}).
\begin{table}
\begin{center}
\begin{tabular}{l|r@{}l}
type & $T\;$ & (RC3)\\
\hline
E &&$T<-3.5$\\
S0 &$-3.5<\;$&$T<-0.5$\\
Sa &$-0.5<\;$&$T<1.5$\\
Sb &$1.5<\;$&$T<3.5$\\
Sbc &$3.5<\;$&$T<4.5$\\
Sc &$4.5<\;$&$T<5.5$\\
Sd &$5.5<\;$&$T<8.5$\\
Im &$8.5<\;$&$T$
\end{tabular}
\caption{Correspondence between the types used in this paper and the
RC3 types.}
\label{types}
\end{center}
\end{table}
The $J-H$, $H-K$ and $J-K$ total and effective colors have been
plotted
in figure~\ref{figTJHK}
as a function of the type.
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{T.J_H_K.ps}}}
\caption{Total and effective NIR colors as a function of type.
A random component in $[-0.5, 0.5]$ has been added to the type
to make the graph clearer. The diamonds are the median colors per type.}
\label{figTJHK}
\end{figure*}
Because the uncertainties computed for $J-H$, $H-K$ and $J-K$ are
overestimates, we simply give in table~\ref{tabTJHK} the median color for
each type and compute its uncertainty by a bootstrap with
replacement.
The total and effective colors show almost no difference.
Whereas $H-K$ seems to be nearly
independent of the type, $J-H$ and $J-K$ decrease clearly at the
late types, indicating that the $J$-band is contaminated by young
stars. The colors are also redder for early-type spirals, which is
difficult to understand with current models of stellar evolution and
atmospheres. This effect is observed in both the total and effective
colors. The extrapolation to compute the latter being small, this may
not come from a problem with the growth curve. The extinction,
negligible in the NIR, is also unlikely to cause this phenomenon. A
possible explanation is that we begin to see in $H$ and $K$ the tail
of the dust infrared emission at short wavelengths. We note in particular that
the $J$ and $K$ samples contain a non-negligible fraction of galaxies
with nuclear activity of which central colors are in some cases as red
as $J-H\sim H-K\sim 1$.
\begin{table*}
\begin{center}
\begin{tabular}{l||r|c|c||r|c|c||r|c|c}
type&$N$&$(J-H)_{\mathrm{T}}$&$(J-H)_{\mathrm{e}}$&$N$&$(H-K)_{\mathrm{T}}$&$(H-K)_{\mathrm{e}}$&$N$&$(J-K)_{\mathrm{T}}$&$(J-K)_{\mathrm{e}}$\\
&&median&median&&median&median&&median&median\\
\hline
E&140&$ 0.71\pm 0.01$&$ 0.71\pm 0.01$&141&$ 0.21\pm 0.01$&$ 0.22\pm 0.01$&140&$ 0.92\pm 0.01$&$ 0.92\pm 0.01$\\
S0&154&$ 0.72\pm 0.01$&$ 0.72\pm 0.01$&157&$ 0.23\pm 0.01$&$ 0.23\pm 0.01$&155&$ 0.95\pm 0.01$&$ 0.95\pm 0.01$\\
Sa& 96&$ 0.74\pm 0.01$&$ 0.75\pm 0.01$& 96&$ 0.28\pm 0.01$&$ 0.28\pm 0.01$& 97&$ 1.01\pm 0.01$&$ 1.02\pm 0.01$\\
Sb& 93&$ 0.78\pm 0.01$&$ 0.79\pm 0.01$& 95&$ 0.26\pm 0.01$&$ 0.27\pm 0.01$& 95&$ 1.03\pm 0.01$&$ 1.04\pm 0.01$\\
Sbc& 46&$ 0.77\pm 0.02$&$ 0.77\pm 0.02$& 47&$ 0.23\pm 0.02$&$ 0.24\pm 0.02$& 47&$ 1.01\pm 0.02$&$ 1.01\pm 0.01$\\
Sc& 46&$ 0.72\pm 0.02$&$ 0.73\pm 0.02$& 46&$ 0.22\pm 0.02$&$ 0.22\pm 0.02$& 46&$ 0.96\pm 0.02$&$ 0.96\pm 0.02$\\
Sd& 26&$ 0.72\pm 0.04$&$ 0.71\pm 0.04$& 24&$ 0.22\pm 0.04$&$ 0.22\pm 0.04$& 24&$ 0.87\pm 0.05$&$ 0.88\pm 0.04$\\
Im& 22&$ 0.57\pm 0.05$&$ 0.56\pm 0.03$& 20&$ 0.24\pm 0.05$&$ 0.25\pm 0.04$& 25&$ 0.73\pm 0.05$&$ 0.74\pm 0.06$\\
\end{tabular}
\caption{Median total and effective $J-H$, $H-K$ and $J-K$ colors
(with their $1\sigma$-uncertainties) as a function
of type.}
\label{tabTJHK}
\end{center}
\end{table*}
\subsection{$B-J$, $B-H$ and $B-K$ colors}
The catalog is large enough to perform a statistical analysis
of the colors as a function of the type, the luminosity and the
inclination or the shape ($R_{25}$) of the galaxy. An interesting
quantity is also the intrinsic scatter in the colors at a given type,
luminosity and $R_{25}$, for it is a measure of the variations of the
star formation history and of the effects of dust (e.g. Peletier \&
de Grijs \cite{PdG}; Shioya \& Bekki \cite{SB}).
Note however that the intrinsic scatter may depend on the
type-binning, especially if the colors evolve rapidly from type to type.
The $B-H$ sample is by far the largest (more than 900 galaxies in $B-H$ against
600 in $B-J$ and $B-K$). It is also the most accurate
(the uncertainty is typically 0.16 mag in $(B-H)_{\mathrm{T}}$ and 0.09 mag in $(B-H)_{\mathrm{e}}$)
and the most complete in type, whereas the $B-J$ and $B-K$ sample are strongly
deficient in the latest types. For these various reasons, we will
mainly focus our discussion in the following on the analysis
of the $B-H$ data.
We adopt here the formalism proposed by Akritas (\cite{Akritas}).
Let us assume that the true color $Y_i$ of the $i^{\mathrm{th}}$ galaxy depends
linearly on $p$ quantities $X_{ij}, j\in[1,p]$ (the $X_{ij}$ being, e.g.,
$\log_{10}(R_{25})$ or the absolute magnitude):
\[Y_i=\beta_0+\sum_{j=1}^p\beta_j(X_{ij}-\mu_{j})+\epsilon_i,\]
where $\mu_j$, the median of $X_{ij}$, is introduced merely to
identify $\beta_0$ with the typical color
of the sample, and $\epsilon_i$ is the deviation
from the relation due to the intrinsic scatter. We assume
that $\epsilon_i\sim {\cal N}(0,\sigma^2)$ independently of the other
parameters. Let us write $\beta_{p+1}=\sigma$ to remind that
the scatter is also to be determined.
The observed variables, $x_{ij}$ and $y_i$, are related to the
true variables by
\[x_{ij}=X_{ij}+\epsilon_{ij},\]
\[y_i=Y_i+\epsilon_{Yi},\]
where $\epsilon_{ij}\sim {\cal N}(0,\sigma_{ij}^2)$ and $\epsilon_{Yi}\sim {\cal N}(0,\sigma_{Yi}^2)$.
We obtain
\[y_i-\beta_0-\sum_{j=1}^p\beta_j(x_{ij}-\mu_j)=\epsilon_{Yi}-\sum_{j=1}^p
\beta_j\epsilon_{ij}+\epsilon_i.\]
Assuming that $\epsilon_i$ is not correlated with the $\epsilon_{ij}$
and $\epsilon_{Yi}$, and that the covariances between $\epsilon_{ij}$
and $\epsilon_{ik}$ and between $\epsilon_{ij}$
and $\epsilon_{ijY}$ are $\gamma_{ijk}$ and $\gamma_{ijY}$, respectively,
we obtain that $y_i-\beta_0-\sum_{j=1}^p\beta_j(x_{ij}-\mu_j)\sim {\cal N}(0,{\sigma_i}^2+\beta_{p+1}^2)$
with
\[{\sigma_i}^2=\sigma_{Yi}^2+\sum_{j=1}^p\beta_j^2\sigma_{ij}^2
-2\sum_{j=1}^p\beta_j\gamma_{ijY}-\sum_{j,k=1}^p\beta_j\beta_k\gamma_{ijk}.\]
We have tested two procedures to estimate the $\beta_j$ from our
sample.
The first one estimates the $\beta_j$ by maximizing the logarithm of
the likelihood $\Lambda$ with respect to the $\beta_j, j\in[0,p+1]$, where
\begin{eqnarray*}
\ln(\Lambda)=-\frac{1}{2}\sum_{i=1}^n\Biggl\lbrace\!\!&\!\!&\!\!
\displaystyle{\frac{[y_i-\beta_0-\sum_{j=1}^p\beta_j
(x_{ij}-\mu_j)]^2}{{\sigma_i}^2
+\beta_{p+1}^2}}\\
\!\!&\!\!&\!\!\mbox{}+\ln({\sigma_i}^2+\beta_{p+1}^2)\Biggr\rbrace
\end{eqnarray*}
according to the formulae given above. The covariance matrix of the
$\beta_j$ is computed by inverting the curvature matrix
$\left(\alpha_{ij}\right)$, $(i,j)\in[0,p+1]^2$, where
\[\alpha_{ij}=\frac{\partial^2\left[-\ln(\Lambda)\right]}{\partial\beta_i
\partial\beta_j}
\textrm{\qquad{}(Kendall \& Stuart \cite{KS}).}\]
Maximum likelihood (ML) estimators are often biased and we therefore
wish to compare to another method.
The second procedure (MCES estimators) has been developed
by Akritas \& Bershady (\cite{AB})
and Akritas (\cite{Akritas}). According to the authors, it yields unbiased
estimators.
However, the $\sigma_{Yi}$ are not used
in the expression of the $\beta_i, i\le p$, which means that they have
all the same weight. This is especially not satisfying if the
intrinsic scatter is small or comparable to the uncertainties
in the colors -- which
happens in particular for elliptical galaxies -- and may give an
excessive importance to outliers.
The authors also do not provide the intrinsic scatter
$\beta_{p+1}$ and we estimate it from
\[
\beta_{p+1}^2=
\frac{1}{n}\sum_{i=1}^n\Bigl\lbrace\bigl[y_i-\beta_0
-\sum_{j=1}^p\beta_j\left(x_{ij}-\mu_j\right)\bigr]^2
-{\sigma_i}^2\Bigr\rbrace
\]
but when the intrinsic scatter is smaller or of the same order than
the uncertainties in the observables, it is
underestimated\footnote{This is shown by the fact that the ML and
MCES estimators of the scatter in the effective colors
(which are less uncertain) are similar and are also close
to the ML estimator for the total colors, whereas the MCES intrinsic
scatter in the total colors is smaller.} and we may
even obtain a negative (and
meaningless) value. We then assume $\beta_{p+1}=0$.
Uncertainties on the $\beta_j$ have been computed by a bootstrap
with replacement on the $(x_{ij},y_i)$ rather than by using the
cumbersome formulae proposed by Akritas \& Bershady (\cite{AB})
and Akritas (\cite{Akritas}), which anyway are not available for $\beta_{p+1}$.
Note that the uncertainties determined either from the curvature matrix
or by bootstrap become themselves very uncertain when the number
of galaxies is less than about 20 to 30.
In the following, values determined by the ML and MCES estimators are
written in boldface and italic characters, respectively.
Because we are mainly interested in ``normal'' galaxies, we have to
reject a few galaxies with unusual colors (Buta et al. \cite{Buta}).
Most of them are extremely
blue and, according to their optical colors (when available),
are presumably starbursting galaxies. Some blue compact dwarfs
may also be misclassified from their morphology as ellipticals.
In the opposite,
our sample contains some galaxies with very red colors, especially in
the $K$-band and their NIR emission is dominated by an active nucleus.
To reject these galaxies, we apply an iterative method.
Rather than using the standard deviation around the best fit, which is
very sensitive to the outliers, we estimate the observational scatter
from the median of the absolute deviations.
At each step, we compute
\[\delta=1.48\mathop{\rm median}\limits_{i\in[1,n]}\biggl|y_i-\beta_0
-\sum_{j=1}^p\beta_j(x_{ij}-\mu_j)\biggr|\]
from the
sample (including the ``abnormal'' galaxies), the factor 1.48 ensuring that in the case of a perfectly
Gaussian distribution, the standard deviation is $\delta$;
we define ``normal'' galaxies as those having $\left|y_i-\beta_0
-\sum_{j=1}^p\beta_j(x_{ij}-\mu_j)\right|<3\delta$
and then determine the $\beta_j$ from them only
by either of the methods discussed above.
We iterate this procedure 4 times, but the convergence is usually
achieved at the third one.
\subsection{The type-color relation}
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{T.B_JHK.ps}}}
\caption{Total and effective optical-to-NIR colors as a function
of the type. The diamonds are the standard colors determined by ML.
The vertical segments show the intrinsic scatter and the
horizontal ones the width of the type-bin.}
\label{figTBJHK}
\end{figure*}
The total and effective $B-J$, $B-H$ and $B-K$ colors
have been plotted as a function of the morphological type
on Figure~\ref{figTBJHK}. A very impressive trend to bluer colors
with advancing type is observed. The mean $(B-H)_{\mathrm{T}}$
of irregular galaxies is 1.34 mag bluer
than the one of ellipticals, to be compared to 0.68 in $U-B$, 0.47 in
$B-V$, 0.46 in $V-I_{\mathrm{c}}$ and only 0.23 in $V-R_{\mathrm{c}}$
(Fioc \& Rocca-Volmerange, in preparation). A similar gap between the
$(B-H)_{\mathrm{T}}$ of ellipticals and irregulars was
obtained in $(B-H)_{\mathrm{e}}$ by Buta (\cite{Buta95})
from a sample of 225 galaxies.
The $(B-H)_{\mathrm{T}}$ colors determined by the ML estimator are
almost always systematically redder by $\sim 0.02$ mag
than those computed using the MCES estimator. This bias comes from the
fact that the uncertainties are smaller for brighter galaxies in the
NIR, which are also usually redder. The ML estimator
giving more weight to the data with smaller uncertainties, it is
biased to the red.
A significant scatter is also observed within each type.
Part of it is due to the observational uncertainties, but it comes
mainly from the {\em intrinsic} scatter in the colors.
The intrinsic scatter increases from $\mathbf{0.13}/\mathit{0.08}$, depending on the
estimator, for ellipticals to $\sim 0.4$ for Sb and remains nearly constant
at later types.
\begin{table*}
\begin{center}
\begin{tabular}{l|r||c|c||c|c}
type & $N$ & $\beta_0[(B-J)_{\mathrm{T}}]$& $\sigma$ &$\beta_0[(B-J)_{\mathrm{e}}]$ &$\sigma$
\\
\hline
E&140&$\mathbf{ 3.07\pm 0.02}$&$\mathbf{ 0.13\pm 0.02}$&$\mathbf{ 3.15\pm 0.01}$&$\mathbf{ 0.13\pm 0.01}$\\
&&$\mathit{ 3.02\pm 0.02}$&$\mathit{ 0.10\pm 0.03}$&$\mathit{ 3.13\pm 0.02}$&$\mathit{ 0.14\pm 0.02}$\\
S0&155&$\mathbf{ 3.01\pm 0.02}$&$\mathbf{ 0.17\pm 0.02}$&$\mathbf{ 3.10\pm 0.02}$&$\mathbf{ 0.18\pm 0.01}$\\
&&$\mathit{ 2.97\pm 0.02}$&$\mathit{ 0.15\pm 0.03}$&$\mathit{ 3.07\pm 0.02}$&$\mathit{ 0.24\pm 0.02}$\\
Sa& 97&$\mathbf{ 2.74\pm 0.04}$&$\mathbf{ 0.31\pm 0.03}$&$\mathbf{ 2.91\pm 0.04}$&$\mathbf{ 0.31\pm 0.03}$\\
&&$\mathit{ 2.74\pm 0.04}$&$\mathit{ 0.25\pm 0.05}$&$\mathit{ 2.93\pm 0.03}$&$\mathit{ 0.27\pm 0.03}$\\
Sb& 95&$\mathbf{ 2.63\pm 0.04}$&$\mathbf{ 0.36\pm 0.04}$&$\mathbf{ 2.79\pm 0.04}$&$\mathbf{ 0.37\pm 0.03}$\\
&&$\mathit{ 2.62\pm 0.05}$&$\mathit{ 0.33\pm 0.04}$&$\mathit{ 2.79\pm 0.04}$&$\mathit{ 0.36\pm 0.04}$\\
Sbc& 47&$\mathbf{ 2.48\pm 0.06}$&$\mathbf{ 0.35\pm 0.05}$&$\mathbf{ 2.67\pm 0.06}$&$\mathbf{ 0.35\pm 0.04}$\\
&&$\mathit{ 2.45\pm 0.06}$&$\mathit{ 0.17\pm 0.10}$&$\mathit{ 2.66\pm 0.05}$&$\mathit{ 0.26\pm 0.07}$\\
Sc& 46&$\mathbf{ 2.27\pm 0.09}$&$\mathbf{ 0.53\pm 0.07}$&$\mathbf{ 2.44\pm 0.10}$&$\mathbf{ 0.61\pm 0.07}$\\
&&$\mathit{ 2.24\pm 0.08}$&$\mathit{ 0.45\pm 0.10}$&$\mathit{ 2.43\pm 0.10}$&$\mathit{ 0.56\pm 0.08}$\\
Sd& 26&$\mathbf{ 1.96\pm 0.09}$&$\mathbf{ 0.37\pm 0.07}$&$\mathbf{ 2.16\pm 0.09}$&$\mathbf{ 0.38\pm 0.07}$\\
&&$\mathit{ 1.93\pm 0.08}$&$\mathit{ 0.23\pm 0.12}$&$\mathit{ 2.14\pm 0.08}$&$\mathit{ 0.32\pm 0.10}$\\
Im& 28&$\mathbf{ 1.61\pm 0.09}$&$\mathbf{ 0.39\pm 0.07}$&$\mathbf{ 1.74\pm 0.10}$&$\mathbf{ 0.50\pm 0.08}$\\
&&$\mathit{ 1.52\pm 0.10}$&$\mathit{ 0.41\pm 0.10}$&$\mathit{ 1.72\pm 0.10}$&$\mathit{ 0.47\pm 0.08}$\\
\end{tabular}
\caption{$(B-J)_{\mathrm{T}}$ and $(B-J)_{\mathrm{e}}$ colors per
type. The values computed with the ML and MCES estimators are written
respectively in boldface and italic characters. $\sigma$ is the
intrinsic scatter.}
\label{BmoinsJ}
\end{center}
\end{table*}
\begin{table*}
\begin{center}
\begin{tabular}{l|r||c|c||c|c}
type & $N$ & $\beta_0[(B-H)_{\mathrm{T}}]$& $\sigma$ &$\beta_0[(B-H)_{\mathrm{e}}]$ &$\sigma$
\\
\hline
E&142&$\mathbf{ 3.76\pm 0.02}$&$\mathbf{ 0.13\pm 0.01}$&$\mathbf{ 3.86\pm 0.02}$&$\mathbf{ 0.14\pm 0.01}$\\
&&$\mathit{ 3.73\pm 0.02}$&$\mathit{ 0.08\pm 0.04}$&$\mathit{ 3.84\pm 0.02}$&$\mathit{ 0.13\pm 0.02}$\\
S0&157&$\mathbf{ 3.72\pm 0.02}$&$\mathbf{ 0.17\pm 0.02}$&$\mathbf{ 3.85\pm 0.02}$&$\mathbf{ 0.17\pm 0.01}$\\
&&$\mathit{ 3.69\pm 0.02}$&$\mathit{ 0.14\pm 0.03}$&$\mathit{ 3.82\pm 0.02}$&$\mathit{ 0.19\pm 0.02}$\\
Sa&110&$\mathbf{ 3.57\pm 0.04}$&$\mathbf{ 0.32\pm 0.03}$&$\mathbf{ 3.73\pm 0.04}$&$\mathbf{ 0.34\pm 0.03}$\\
&&$\mathit{ 3.54\pm 0.04}$&$\mathit{ 0.31\pm 0.04}$&$\mathit{ 3.73\pm 0.04}$&$\mathit{ 0.33\pm 0.03}$\\
Sb&171&$\mathbf{ 3.53\pm 0.03}$&$\mathbf{ 0.41\pm 0.03}$&$\mathbf{ 3.73\pm 0.03}$&$\mathbf{ 0.39\pm 0.02}$\\
&&$\mathit{ 3.52\pm 0.03}$&$\mathit{ 0.40\pm 0.03}$&$\mathit{ 3.71\pm 0.03}$&$\mathit{ 0.40\pm 0.03}$\\
Sbc&110&$\mathbf{ 3.33\pm 0.04}$&$\mathbf{ 0.34\pm 0.03}$&$\mathbf{ 3.53\pm 0.04}$&$\mathbf{ 0.35\pm 0.03}$\\
&&$\mathit{ 3.30\pm 0.03}$&$\mathit{ 0.29\pm 0.04}$&$\mathit{ 3.51\pm 0.03}$&$\mathit{ 0.32\pm 0.03}$\\
Sc&125&$\mathbf{ 3.24\pm 0.04}$&$\mathbf{ 0.44\pm 0.03}$&$\mathbf{ 3.42\pm 0.04}$&$\mathbf{ 0.44\pm 0.03}$\\
&&$\mathit{ 3.20\pm 0.04}$&$\mathit{ 0.42\pm 0.04}$&$\mathit{ 3.39\pm 0.04}$&$\mathit{ 0.43\pm 0.03}$\\
Sd&135&$\mathbf{ 2.94\pm 0.04}$&$\mathbf{ 0.45\pm 0.03}$&$\mathbf{ 3.13\pm 0.04}$&$\mathbf{ 0.46\pm 0.03}$\\
&&$\mathit{ 2.93\pm 0.04}$&$\mathit{ 0.45\pm 0.03}$&$\mathit{ 3.12\pm 0.04}$&$\mathit{ 0.47\pm 0.03}$\\
Im& 42&$\mathbf{ 2.41\pm 0.08}$&$\mathbf{ 0.46\pm 0.06}$&$\mathbf{ 2.56\pm 0.08}$&$\mathbf{ 0.46\pm 0.06}$\\
&&$\mathit{ 2.38\pm 0.08}$&$\mathit{ 0.42\pm 0.07}$&$\mathit{ 2.54\pm 0.08}$&$\mathit{ 0.44\pm 0.06}$\\
\end{tabular}
\caption{$(B-H)_{\mathrm{T}}$ and $(B-H)_{\mathrm{e}}$ colors per type.}
\label{BmoinsH}
\end{center}
\end{table*}
\begin{table*}
\begin{center}
\begin{tabular}{l|r||c|c||c|c}
type & $N$ & $\beta_0[(B-K)_{\mathrm{T}}]$& $\sigma$ &$\beta_0[(B-K)_{\mathrm{e}}]$ &$\sigma$
\\
\hline
E&142&$\mathbf{ 3.98\pm 0.02}$&$\mathbf{ 0.13\pm 0.01}$&$\mathbf{ 4.08\pm 0.02}$&$\mathbf{ 0.15\pm 0.01}$\\
&&$\mathit{ 3.95\pm 0.02}$&$\mathit{ 0.09\pm 0.04}$&$\mathit{ 4.06\pm 0.02}$&$\mathit{ 0.14\pm 0.02}$\\
S0&159&$\mathbf{ 3.97\pm 0.02}$&$\mathbf{ 0.18\pm 0.02}$&$\mathbf{ 4.07\pm 0.02}$&$\mathbf{ 0.19\pm 0.01}$\\
&&$\mathit{ 3.95\pm 0.02}$&$\mathit{ 0.15\pm 0.03}$&$\mathit{ 4.06\pm 0.02}$&$\mathit{ 0.19\pm 0.02}$\\
Sa& 99&$\mathbf{ 3.79\pm 0.04}$&$\mathbf{ 0.37\pm 0.03}$&$\mathbf{ 3.99\pm 0.04}$&$\mathbf{ 0.36\pm 0.03}$\\
&&$\mathit{ 3.79\pm 0.04}$&$\mathit{ 0.33\pm 0.04}$&$\mathit{ 4.00\pm 0.04}$&$\mathit{ 0.35\pm 0.03}$\\
Sb&100&$\mathbf{ 3.64\pm 0.05}$&$\mathbf{ 0.41\pm 0.04}$&$\mathbf{ 3.84\pm 0.05}$&$\mathbf{ 0.42\pm 0.04}$\\
&&$\mathit{ 3.62\pm 0.05}$&$\mathit{ 0.42\pm 0.05}$&$\mathit{ 3.82\pm 0.05}$&$\mathit{ 0.43\pm 0.04}$\\
Sbc& 48&$\mathbf{ 3.47\pm 0.08}$&$\mathbf{ 0.46\pm 0.06}$&$\mathbf{ 3.64\pm 0.07}$&$\mathbf{ 0.40\pm 0.05}$\\
&&$\mathit{ 3.45\pm 0.07}$&$\mathit{ 0.32\pm 0.10}$&$\mathit{ 3.64\pm 0.06}$&$\mathit{ 0.32\pm 0.06}$\\
Sc& 46&$\mathbf{ 3.26\pm 0.09}$&$\mathbf{ 0.49\pm 0.07}$&$\mathbf{ 3.46\pm 0.09}$&$\mathbf{ 0.50\pm 0.06}$\\
&&$\mathit{ 3.22\pm 0.08}$&$\mathit{ 0.39\pm 0.10}$&$\mathit{ 3.43\pm 0.08}$&$\mathit{ 0.45\pm 0.07}$\\
Sd& 24&$\mathbf{ 2.89\pm 0.10}$&$\mathbf{ 0.39\pm 0.08}$&$\mathbf{ 3.09\pm 0.10}$&$\mathbf{ 0.43\pm 0.08}$\\
&&$\mathit{ 2.80\pm 0.08}$&$\mathit{ 0.21\pm 0.13}$&$\mathit{ 3.01\pm 0.09}$&$\mathit{ 0.31\pm 0.10}$\\
Im& 25&$\mathbf{ 2.23\pm 0.11}$&$\mathbf{ 0.49\pm 0.09}$&$\mathbf{ 2.41\pm 0.11}$&$\mathbf{ 0.50\pm 0.08}$\\
&&$\mathit{ 2.21\pm 0.11}$&$\mathit{ 0.45\pm 0.10}$&$\mathit{ 2.40\pm 0.11}$&$\mathit{ 0.50\pm 0.08}$\\
\end{tabular}
\caption{$(B-K)_{\mathrm{T}}$ and $(B-K)_{\mathrm{e}}$ colors per type.}
\label{BmoinsK}
\end{center}
\end{table*}
\subsection{The $R_{25}$-color relation}
\label{corrinclin}
The color-inclination relation is potentially a very powerful constraint on
the amount of dust and its distribution relatively to stars.
As a disk becomes more and more inclined, its optical depth increases
and the colors redden. This must be especially striking when
one of the band (e.g. $B$) is heavily extinguished whereas the dust
is almost transparent in the other one (e.g. $H$). For these reasons,
various studies have tried to determine a color-inclination relation,
regrouping the galaxies either as a function of type (e.g. Boselli \&
Gavazzi \cite{BG94}) or as a function
of their NIR absolute magnitude (Tully et al. \cite{Tully}).
The samples used in these studies
where however small (about 100 galaxies) and it is worth to
look at the relation once again with our much larger catalog.
For an oblate ellipsoid, which is a standard model for a galaxy
disk (Hubble \cite{Hubble}),
the inclination $i$ (face-on corresponds to $i=0$)
is related to the true ratio $q_0$ of the minor to the major axis
and to the apparent ratio $q\simeq 1/R_{25}$ by
\[\cos^2i=\frac{q^2-q_0^2}{1-q_0^2}.\]
Practically, we prefer to establish a relation between the colors
and $\log_{10}(R_{25})$, both because $R_{25}$ is the directly observed
quantity and because the linear estimators used here provide
a better fit when $\log_{10}(R_{25})$ is used rather than $i$ or $\cos(i)$.
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{R25.B_H.ps}}}
\caption{$(B-H)_{\mathrm{T}}$-$\log_{10}(R_{25})$ relation per type.
The solid and dashed lines correspond respectively to the ML and
MCES estimators.}
\label{figinclin}
\end{figure*}
The relations are plotted in Fig.~\ref{figinclin} and
statistical estimators of the slope and the color are given
in table~\ref{tabinclin} for $B-H$.
\begin{table*}
\begin{center}
\begin{tabular}{l|r|c|c|c|c}
type & $N$ &$\beta_0[(B-H)_{\mathrm{T}}]$&$\beta_1[\log_{10}(R_{25})]$&$\mu_1$&$\sigma$\\
\hline
E&138&$\mathbf{ 3.77\pm 0.02}$&$\mathbf{ -0.77\pm 0.19}$&0.10&$\mathbf{ 0.12\pm 0.01}$\\
&&$\mathit{ 3.75\pm 0.02}$&$\mathit{ -1.00\pm 0.27}$&&$\mathit{ (0.03\pm 0.03)}$\\
S0&153&$\mathbf{ 3.72\pm 0.02}$&$\mathbf{ 0.04\pm 0.11}$&0.20&$\mathbf{ 0.17\pm 0.02}$\\
&&$\mathit{ 3.69\pm 0.02}$&$\mathit{ 0.04\pm 0.12}$&&$\mathit{ 0.15\pm 0.03}$\\
Sa&110&$\mathbf{ 3.53\pm 0.04}$&$\mathbf{ 1.00\pm 0.22}$&0.19&$\mathbf{ 0.31\pm 0.03}$\\
&&$\mathit{ 3.53\pm 0.04}$&$\mathit{ 0.87\pm 0.19}$&&$\mathit{ 0.24\pm 0.04}$\\
Sb&171&$\mathbf{ 3.48\pm 0.03}$&$\mathbf{ 1.03\pm 0.14}$&0.27&$\mathbf{ 0.34\pm 0.02}$\\
&&$\mathit{ 3.45\pm 0.03}$&$\mathit{ 1.12\pm 0.14}$&&$\mathit{ 0.34\pm 0.03}$\\
Sbc&110&$\mathbf{ 3.29\pm 0.04}$&$\mathbf{ 0.73\pm 0.19}$&0.25&$\mathbf{ 0.34\pm 0.03}$\\
&&$\mathit{ 3.27\pm 0.03}$&$\mathit{ 0.72\pm 0.22}$&&$\mathit{ 0.28\pm 0.04}$\\
Sc&125&$\mathbf{ 3.14\pm 0.04}$&$\mathbf{ 0.89\pm 0.15}$&0.29&$\mathbf{ 0.37\pm 0.03}$\\
&&$\mathit{ 3.12\pm 0.04}$&$\mathit{ 0.95\pm 0.17}$&&$\mathit{ 0.32\pm 0.03}$\\
Sd&135&$\mathbf{ 2.89\pm 0.04}$&$\mathbf{ 0.76\pm 0.15}$&0.34&$\mathbf{ 0.40\pm 0.03}$\\
&&$\mathit{ 2.88\pm 0.04}$&$\mathit{ 0.82\pm 0.16}$&&$\mathit{ 0.38\pm 0.03}$\\
Im& 41&$\mathbf{ 2.41\pm 0.08}$&$\mathbf{ 0.72\pm 0.36}$&0.29&$\mathbf{ 0.40\pm 0.06}$\\
&&$\mathit{ 2.39\pm 0.07}$&$\mathit{ 0.70\pm 0.34}$&&$\mathit{ 0.35\pm 0.08}$\\
\end{tabular}
\caption{$(B-H)_{\mathrm{T}}$ color per type as a function of $X_{i1}=\log_{10}(R_{25})$.}
\label{tabinclin}
\end{center}
\end{table*}
The slopes are represented as a function of type in
figure~\ref{couleurs_R25}.
\begin{figure}
\resizebox{\hsize}{!}{\rotatebox{0}{\includegraphics{couleurs_R25.ps}}}.
\caption{Slopes of the $(B-H)_{\mathrm{T}}$-, $(B-V)_{\mathrm{T}}$-,
and $(U-B)_{\mathrm{T}}$-$\log_{10}(R_{25})$ relations as a function of type.
The black and white squares correspond respectively
to the ML and MCES estimators.}
\label{couleurs_R25}
\end{figure}
Also shown on this graph are the slopes derived from the larger
$B-V$ and $U-B$ samples.
The typical slope of the $(B-H)_{\mathrm{T}}$-$\log_{10}(R_{25})$ relation is about 1 for spiral
galaxies, to be compared to 0.2--0.3 in $B-V$ and $U-B$.
Surprisingly high is the value of the slope for Sa galaxies. However,
the slope of the $(B-H)_{\mathrm{T}}$-{\em inclination} relation
would be shallower
for Sa since $q_0$ is a decreasing function of type (Bottinelli et
al. \cite{Bottinelli}),
i.e. Sa galaxies are thicker than later types, as also indicated by their
smaller median $\log_{10}(R_{25})$.
A dip is observed for Sbc galaxies, which seems strange since the
maximum extinction is expected precisely for this type (Fioc \&
Rocca-Volmerange \cite{FRV}). Though not inconsistent with a constant
slope for all spirals, given the uncertainties,
a similar dip is also obvious in the
slopes of the $(B-V)_{\mathrm{T}}$- and $(U-B)_{\mathrm{T}}$-$\log_{10}(R_{25})$ relations, suggesting
that this is a real phenomenon. One should first remember that
the slope is not a measure of the total extinction but of the
{\em reddening relative to face-on}.
With a simple radiative transfer model where the stars and the dust
are distributed homogeneously in an infinite slab, we find that when
the face-on optical depth $\tau_{\mathrm{V}}$ in the $V$-band
increases,
the slope begins first to steepen (Fig.~\ref{inclinBmoinsH}b)
but when $\tau_{\mathrm{V}}$ becomes larger than 1--2, the slope
flattens although the colors go on reddening (Fig.~\ref{inclinBmoinsH}a).
\begin{figure}
\resizebox{\hsize}{!}{\rotatebox{0}{\includegraphics{inclinB_H.ps}}}.
\caption{$(B-H)$-$\log_{10}(R_{25})$ relations computed with the radiative
transfer code (see text) for different face-on $V$-band optical
depths. We assume $q_0=0.18$, a typical value for an Sb.
{\bf Top:} reddening relative to the dust-free case. {\bf Bottom:} reddening
relative to face-on.}
\label{inclinBmoinsH}
\end{figure}
Although this modeling is simplistic and we can not infer from it the
value of the optical depth,
this suggests that the extinction may indeed be higher for Sbc
and that the slope has begun to decrease.
This conclusion is reinforced by the fact that the slope
{\em decreases} when we consider {\em effective} $B-H$ colors
($\beta_1[\log_{10}(R_{25})]=\mathbf{0.60}/\mathit{0.55}$), which
are probably more extinguished than total colors.
We observe a positive slope for irregular galaxies in $B-H$, contrary
to $B-V$ and $U-B$ where none is detected. However, the
uncertainties are large and the null slope is within 2$\sigma$.
Moreover, our ``Im'' type regroups in fact not only Im galaxies ($T=10$)
but also Sm ($T=9$) which are more elongated and redder.
The small size of the ``Im'' sample makes it impossible to
decide whether this reddening is due to the dust or to intrinsically
redder populations in Sm galaxies.
No slope is observed for lenticular galaxies, which may indicate
that no significant amount of dust is present in S0 (see also
Sandage \& Visvanatan \cite{SV}). Another
possibility is that
the disk is overwhelmed by the bulge. No
inclination dependence of the colors would then be expected
if the bulge is spherically symmetric or if it is devoid of dust.
The most striking result is the negative slope observed for
ellipticals. This certainly does not come from the dust because
the slope would be positive. For ellipticals, $R_{25}$ is not directly
related to the inclination because their true shape is unknown.
In the mean, rounder ellipticals seem redder and are probably more
metal-rich (Terlevich et al. \cite{Terlevich}). They also tend
to be brighter (Tremblay \& Merritt \cite{TM}) which may link
the color-$\log_{10}(R_{25})$ relation of ellipticals to their
color-magnitude relation.
No such relation is observed in $B-V$,
which is anyway a poor indicator of the metallicity.
The $(U-B)_{\mathrm{T}}$-$\log_{10}(R_{25})$ is not very conclusive:
whereas the ML estimator provides a negative slope,
as in $B-H$, the MCES estimator does not detect any.
Since no obvious bias between the two estimators is observed
for the other types and the intrinsic scatter is very small compared
to the uncertainties on the colors for
ellipticals -- which is the worst case for the MCES estimators --,
we tend to trust more the result given by the ML.
\subsection{The color-magnitude relation}
\subsubsection{Global relation}
The color-magnitude (CM) relation is an important constraint on the
dynamical models of galaxy formation for the absolute magnitude
may be related to the mass of the galaxy via a mass-to-luminosity
ratio.
An important question is the choice of the reference band, optical
or NIR, used for the absolute magnitude. Usually, the NIR has been
adopted because it suffers little extinction. Another reason is that it is
produced mainly by old giants and is therefore
expected to be a better indicator of the mass than the optical,
which is more sensitive to the recent star formation.
We have plotted on Figs.~\ref{CMglobB} and \ref{CMglobNIR} the color-absolute
magnitude relation with either the optical or the NIR as reference
band. The corresponding estimators are given in Table~\ref{tabCMglob}.
The $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$-$\mathop{\mathrm{NIR}}_{\mathrm{abs}}$ relation is much steeper than the
$(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$-$B_{\mathrm{abs}}$ one.
If there was no intrinsic scatter, the slopes should be nearly the
same.
For example, if $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}=\alpha\mathop{\mathrm{NIR}}_{\mathrm{abs}}+\kappa$, then
we should have
$(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}=[\alpha/(1+\alpha)] (B_{\mathrm{abs}}+\kappa)$, which has
almost the same slope since $|\alpha|<<1$.
The scatter is moreover smaller when the NIR is the reference
band. To detect a color-magnitude relation, the choice of the NIR as
reference band is thus clearly favored.
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{Babs.B_JHK.ps}}}
\caption{Global $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$-absolute $B$ magnitude relations.}
\label{CMglobB}
\end{figure*}
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{JHKabs.B_JHK.ps}}}
\caption{Global $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$-absolute $\mathop{\mathrm{NIR}}$ magnitude relations.}
\label{CMglobNIR}
\end{figure*}
\begin{table*}
\begin{center}
\begin{tabular}{@{\ }l@{\ }|@{\ }r@{\ }||@{\ }c@{\ }|@{\ }c@{\ }|@{\ }c@{\ }|@{\ }c@{\ }||@{\ }c@{\ }|@{\ }c@{\ }|@{\ }c@{\ }|@{\ }c@{\ }}
$(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$ & $N$ & $\beta_0[(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}]$&$\beta_1(B_{\mathrm{abs}})$&$\mu_1$ &$\sigma$ &$\beta_0[(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}]$ &$\beta_1(\mathop{\mathrm{NIR}}_{\mathrm{abs}})$&$\mu_1$&$\sigma$\\
\hline
$(B-J)_{\mathrm{T}}$& 540&$\mathbf{ 2.68\pm 0.02}$&$\mathbf{ -0.03\pm 0.02}$&-20.67&$\mathbf{ 0.47\pm 0.02}$&$\mathbf{ 2.68\pm 0.02}$&$\mathbf{ -0.20\pm 0.02}$&-23.34&$\mathbf{ 0.41\pm 0.01}$\\
&&$\mathit{ 2.65\pm 0.02}$&$\mathit{ -0.05\pm 0.03}$&&$\mathit{ 0.46\pm 0.02}$&$\mathit{ 2.67\pm 0.02}$&$\mathit{ -0.19\pm 0.02}$&&$\mathit{ 0.40\pm 0.02}$\\
$(B-H)_{\mathrm{T}}$& 814&$\mathbf{ 3.41\pm 0.02}$&$\mathbf{ -0.05\pm 0.01}$&-20.46&$\mathbf{ 0.45\pm 0.01}$&$\mathbf{ 3.42\pm 0.02}$&$\mathbf{ -0.19\pm 0.01}$&-23.90&$\mathbf{ 0.40\pm 0.01}$\\
&&$\mathit{ 3.39\pm 0.02}$&$\mathit{ -0.06\pm 0.02}$&&$\mathit{ 0.45\pm 0.01}$&$\mathit{ 3.40\pm 0.01}$&$\mathit{ -0.18\pm 0.01}$&&$\mathit{ 0.40\pm 0.01}$\\
$(B-K)_{\mathrm{T}}$& 546&$\mathbf{ 3.68\pm 0.02}$&$\mathbf{ -0.08\pm 0.02}$&-20.67&$\mathbf{ 0.49\pm 0.02}$&$\mathbf{ 3.69\pm 0.02}$&$\mathbf{ -0.24\pm 0.01}$&-24.38&$\mathbf{ 0.41\pm 0.01}$\\
&&$\mathit{ 3.66\pm 0.02}$&$\mathit{ -0.10\pm 0.03}$&&$\mathit{ 0.49\pm 0.02}$&$\mathit{ 3.68\pm 0.02}$&$\mathit{ -0.23\pm 0.02}$&&$\mathit{ 0.40\pm 0.01}$\\
\end{tabular}
\caption{$(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$ vs. $B_{\mathrm{abs}}$ or $\mathop{\mathrm{NIR}}_{\mathrm{abs}}$
color-magnitude relations, where $\mathop{\mathrm{NIR}}=J$, $H$ or $K$.}
\label{tabCMglob}
\end{center}
\end{table*}
Though the overall agreement is very good, we note that the ML
estimator of the slope is slightly biased when
compared to the {\em a priori} unbiased MCES estimator. Its slope is
``more positive'' in the $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$-$B_{\mathrm{abs}}$ relation because of the
positive correlation of the errors in $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$ and $B_{\mathrm{abs}}$,
and ``more negative'' in the $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$-$\mathop{\mathrm{NIR}}_{\mathrm{abs}}$ relation because of the
negative correlation of the errors in $(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$ and $\mathop{\mathrm{NIR}}_{\mathrm{abs}}$.
The slope of the
$(B-\mathop{\mathrm{NIR}})_{\mathrm{T}}$-absolute magnitude relation
increases with the wavelength of the NIR band
for there is also a CM relation in the NIR colors.
This effect
is less apparent in $(B-H)_{\mathrm{T}}$ because the $B-H$ sample contains
many late-type galaxies which have a flatter slope:
a linear fit for all the galaxies together
is hence too simplistic.
\subsubsection{Relation per type}
The global color-magnitude relation discussed above is actually
dominated by star-forming galaxies
($\beta_1(H_{\mathrm{abs}})=\mathbf{-0.21}/\mathit{-0.20}$
against $\mathbf{-0.07}/\mathit{-0.05}$ for E/S0).
These values are close to the slopes of the
$(B_{\mathrm{j}}-K)[D_{25}]$-$K_{\mathrm{abs}}$ relations
($-0.24\pm 0.03$ for Sa--Sdm and $-0.09\pm
0.04$ for E/S0)
determined by Mobasher et al. (\cite{Mobasher}).
The slope for spiral galaxies is much steeper than the
relations obtained by Bershady (\cite{Bershady}) for its
spectral types bk, bm, am and fm.
This is however not surprising, because using {\em spectral}
types rather than {\em photometric} types bins galaxies as a function
of their colors and tends to suppress any color-magnitude relation.
An especially interesting question is whether the mass is the main driving
force of galaxy evolution and star formation, or if the galaxy
type has also to be considered. To answer this question, it is worth to
look at the color-magnitude relation per type more thoroughly.
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{Habs.B_H.ps}}}
\caption{$(B-H)_{\mathrm{T}}$-absolute $H$ magnitude
relations per type.}
\label{figCMtypeNIR}
\end{figure*}
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{Habs.B_H0.ps}}}
\caption{$(B-H)^0_{\mathrm{T}}$-absolute $H$ magnitude
relations per type.}
\label{figCMtypeNIRcorr}
\end{figure*}
\begin{table*}
\begin{center}
\begin{tabular}{l|r|c|c|c|c||c|c|c}
type & $N$ &$\beta_0[(B-H)_{\mathrm{T}}]$&$\beta_1(H_{\mathrm{abs}})$&$\mu_1$&$\sigma$&$\beta_0[(B-H)^0_{\mathrm{T}}]$&$\beta_1(H_{\mathrm{abs}})$&$\sigma$\\
\hline
E&119&$\mathbf{ 3.79\pm 0.02}$&$\mathbf{ -0.05\pm 0.01}$&-24.95&$\mathbf{ 0.12\pm 0.02}$&$\mathbf{ 3.79\pm 0.02}$&$\mathbf{ -0.05\pm 0.01}$&$\mathbf{ 0.12\pm 0.02}$\\
&&$\mathit{ 3.76\pm 0.02}$&$\mathit{ -0.02\pm 0.02}$&&$\mathit{ (0.06\pm 0.04)}$&$\mathit{ 3.76\pm 0.02}$&$\mathit{ -0.02\pm 0.02}$&$\mathit{ (0.06\pm 0.04)}$\\
S0&110&$\mathbf{ 3.70\pm 0.02}$&$\mathbf{ -0.13\pm 0.02}$&-24.00&$\mathbf{ 0.14\pm 0.02}$&$\mathbf{ 3.67\pm 0.02}$&$\mathbf{ -0.12\pm 0.02}$&$\mathbf{ 0.18\pm 0.02}$\\
&&$\mathit{ 3.67\pm 0.02}$&$\mathit{ -0.10\pm 0.03}$&&$\mathit{ 0.14\pm 0.03}$&$\mathit{ 3.64\pm 0.02}$&$\mathit{ -0.11\pm 0.03}$&$\mathit{ 0.16\pm 0.03}$\\
Sa& 94&$\mathbf{ 3.52\pm 0.04}$&$\mathbf{ -0.22\pm 0.04}$&-23.85&$\mathbf{ 0.37\pm 0.03}$&$\mathbf{ 3.35\pm 0.04}$&$\mathbf{ -0.20\pm 0.03}$&$\mathbf{ 0.29\pm 0.03}$\\
&&$\mathit{ 3.51\pm 0.04}$&$\mathit{ -0.22\pm 0.05}$&&$\mathit{ 0.36\pm 0.04}$&$\mathit{ 3.35\pm 0.04}$&$\mathit{ -0.20\pm 0.05}$&$\mathit{ 0.28\pm 0.04}$\\
Sb&140&$\mathbf{ 3.53\pm 0.04}$&$\mathbf{ -0.14\pm 0.04}$&-24.31&$\mathbf{ 0.37\pm 0.03}$&$\mathbf{ 3.18\pm 0.03}$&$\mathbf{ -0.17\pm 0.03}$&$\mathbf{ 0.30\pm 0.02}$\\
&&$\mathit{ 3.51\pm 0.04}$&$\mathit{ -0.12\pm 0.05}$&&$\mathit{ 0.36\pm 0.03}$&$\mathit{ 3.17\pm 0.03}$&$\mathit{ -0.14\pm 0.04}$&$\mathit{ 0.29\pm 0.03}$\\
Sbc& 91&$\mathbf{ 3.36\pm 0.04}$&$\mathbf{ -0.19\pm 0.04}$&-24.12&$\mathbf{ 0.35\pm 0.03}$&$\mathbf{ 3.14\pm 0.03}$&$\mathbf{ -0.24\pm 0.03}$&$\mathbf{ 0.26\pm 0.03}$\\
&&$\mathit{ 3.33\pm 0.04}$&$\mathit{ -0.12\pm 0.07}$&&$\mathit{ 0.32\pm 0.04}$&$\mathit{ 3.12\pm 0.03}$&$\mathit{ -0.18\pm 0.06}$&$\mathit{ 0.25\pm 0.03}$\\
Sc&102&$\mathbf{ 3.22\pm 0.04}$&$\mathbf{ -0.19\pm 0.04}$&-23.44&$\mathbf{ 0.39\pm 0.03}$&$\mathbf{ 2.87\pm 0.03}$&$\mathbf{ -0.23\pm 0.03}$&$\mathbf{ 0.27\pm 0.03}$\\
&&$\mathit{ 3.19\pm 0.04}$&$\mathit{ -0.17\pm 0.05}$&&$\mathit{ 0.38\pm 0.03}$&$\mathit{ 2.85\pm 0.03}$&$\mathit{ -0.21\pm 0.04}$&$\mathit{ 0.25\pm 0.03}$\\
Sd& 95&$\mathbf{ 2.98\pm 0.05}$&$\mathbf{ -0.08\pm 0.04}$&-22.11&$\mathbf{ 0.46\pm 0.04}$&$\mathbf{ 2.65\pm 0.04}$&$\mathbf{ -0.13\pm 0.03}$&$\mathbf{ 0.37\pm 0.03}$\\
&&$\mathit{ 2.95\pm 0.05}$&$\mathit{ -0.07\pm 0.05}$&&$\mathit{ 0.46\pm 0.04}$&$\mathit{ 2.64\pm 0.04}$&$\mathit{ -0.12\pm 0.04}$&$\mathit{ 0.34\pm 0.03}$\\
Im& 20&$\mathbf{ 2.55\pm 0.09}$&$\mathbf{ -0.07\pm 0.05}$&-21.60&$\mathbf{ 0.34\pm 0.07}$&$\mathbf{ 2.32\pm 0.08}$&$\mathbf{ -0.09\pm 0.04}$&$\mathbf{ 0.27\pm 0.06}$\\
&&$\mathit{ 2.54\pm 0.08}$&$\mathit{ -0.06\pm 0.07}$&&$\mathit{ 0.31\pm 0.09}$&$\mathit{ 2.32\pm 0.08}$&$\mathit{ -0.08\pm 0.06}$&$\mathit{ 0.24\pm 0.11}$\\
\end{tabular}
\caption{$(B-H)_{\mathrm{T}}$ and $(B-H)^0_{\mathrm{T}}$ per type as a function of $X_{i1}=H_{\mathrm{abs}}$.}
\label{tabCMtype}
\end{center}
\end{table*}
We have plotted the $(B-H)_{\mathrm{T}}$-$H_{\mathrm{abs}}$
relation as a function
of type in Fig.~\ref{figCMtypeNIR}. We also show in
Fig.~\ref{figCMtypeNIRcorr} the
$(B-H)^0_{\mathrm{T}}$-$H_{\mathrm{abs}}$
relations where $(B-H)^0_{\mathrm{T}}$ is the color corrected to
face-on using the $\beta_1[\log_{10}(R_{25})]$ determined
in~\ref{corrinclin}. We assume that the extinction is negligible in
the NIR and do not correct $H_{\mathrm{abs}}$. No correction is also
applied to elliptical colors because their color-$R_{25}$ relation is not
due to the extinction.
The values of the estimators are given in Table~\ref{tabCMtype}.
We note that the intrinsic scatter in the corrected colors
is significantly reduced.
A more picturesque comparison of the relations for the different types
is plotted on Fig.~\ref{segments}. Though the agreement between the estimators
is not as good as previously and the uncertainties are large, the
slope obviously depends on the type. We tentatively summarize the
following results:
\paragraph{Star-forming galaxies:}
the slope is much flatter for
Sd--Im than for Sa--Sc galaxies. Mobasher et al. (\cite{Mobasher})
probably did not detect this because the few Sd/Im galaxies
in their late-spirals sample were lost among Sbc and Sc galaxies.
At a given NIR absolute magnitude, the mean colors of the
types are different, indicating that the CM relation is not simply a
type-mass relation: the NIR intrinsic luminosity must
be completed by the type to characterize the colors (and the star
formation history) of star-forming galaxies.
A similar conclusion was drawn by Gavazzi (\cite{Gavazzi93})
but was later challenged by Gavazzi et al. (\cite{Gavazzi96b}).
\paragraph{E/S0:}
the color-magnitude relation of early-type galaxies is usually
explained by the galactic winds produced by super-novae
in a starbursting environment, which expel the gas and quench the
star formation (Matthews \& Baker \cite{MB}).
Massive (and NIR bright) galaxies
having a deeper gravitational well, they are able to retain
the gas longer and to prolong the star-forming phase. This results
in a higher mean stellar metallicity and thus in redder colors
(Faber \cite{Faber}).
Besides its interest to constrain the models of galaxy formation,
the CM relation has also been proposed as a distance indicator.
We find a different slope for ellipticals and lenticulars.
The slope of S0 seems
closer to that of early and intermediate spirals than to that of
ellipticals, which may give some clues on their evolution.
The MCES estimation does not detect any significant slope for
ellipticals. Because of the small intrinsic scatter, it is maybe not
the best estimator\footnote{Note that the slopes obtained for the
{\em effective}
color vs. absolute magnitude relation with the two estimators
($\mathbf{-0.03}/\mathit{-0.02}$)
are in good agreement, both between them and with the MCES slope
of the total color-magnitude relation.}, but even
the slope determined by ML is smaller than that of the
comparable $(V-K)[5h^{-1}\,\mathrm{kpc}]$-$V_{\mathrm{T}}$ relation
established by
Bower et al. (\cite{Bower}) in the Virgo and Coma clusters.
Our intrinsic scatter is also larger that theirs. A possible
explanation of this discrepancy is that our sample contains
not only cluster galaxies but also field ellipticals, which may have a
different star formation history and increase the scatter (Larson et
al. \cite{Larson}; Kauffmann \& Charlot \cite{KC}; Baugh et
al. \cite{Baugh}; see yet
Bernardi et al. \cite{Bernardi} and Schade et al. \cite{Schade}).
However, we also note that
the relation established by Bower et al. is based on {\em aperture}
colors. Their aperture is comparable to the mean effective aperture
of ellipticals in Virgo and samples only about half their luminosity
(cf. also Kodama et al. \cite{Kodama}).
Since the effective aperture of bright ellipticals is larger, only
their inner regions are observed in the Bower et al.'s aperture.
Because of the blue-outwards color gradient, their aperture color is redder
and the slope is higher than the one derived from total colors.
We therefore encourage the modellists of elliptical galaxies, either to
compare their relations to
total observed colors, or to predict aperture colors.
\begin{figure}
\resizebox{\hsize}{!}{\rotatebox{-90}{\includegraphics{typeHabs.B_H0.ps}}}
\caption{Analytical $(B-H)^0_{\mathrm{T}}$-absolute $H$ magnitude
relations. The segments correspond to the mean of the ML and MCES fits and
extend from $\mu_0(H_{\mathrm{abs}})+1.5$
to $\mu_0(H_{\mathrm{abs}})-1.5$ for each type.}
\label{segments}
\end{figure}
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{Babs.B_H.ps}}}
\caption{$(B-H)_{\mathrm{T}}$-absolute $B$ magnitude
relations per type.}
\label{CMtypeB}
\end{figure*}
The color-magnitude relations as a function of the $B$ absolute
magnitude are finally plotted for each type on figure~\ref{CMtypeB}.
The slopes are very different of those obtained when the
NIR is the reference band. For spiral galaxies especially,
they tend to have a positive sign, with brighter galaxies (in $B$)
being bluer! This behavior is typically expected if the
relation is dominated by the extinction.
If we assume that the extinction is negligible in the NIR, the
same correction applied to colors may be used to compute face-on
absolute magnitudes $B^0_{\mathrm{abs}}$.
The $(B-H)^0_{\mathrm{T}}$-$B^0_{\mathrm{abs}}$ relations per type show then
more reasonable slopes (Fig.~\ref{CMtypeBcorr}).
\begin{figure*}
\resizebox{!}{22cm}{\rotatebox{0}{\includegraphics{Babs0.B_H0.ps}}}
\caption{$(B-H)^0_{\mathrm{T}}$-absolute $B^0$ magnitude
relations per type.}
\label{CMtypeBcorr}
\end{figure*}
\begin{table*}
\begin{center}
\begin{tabular}{l|r|c|c|c|c|c|c}
type & $N$ &$\beta_0[(B-H)_{\mathrm{T}}]$&$\beta_1(H_{\mathrm{abs}})$&$\mu_1$&$\beta_2[\log_{10}(R_{25})]$&$\mu_2$&$\sigma$\\
\hline
E&115&$\mathbf{ 3.80\pm 0.02}$&$\mathbf{ -0.05\pm 0.01}$&-24.97&$\mathbf{ -0.67\pm 0.19}$& 0.10&$\mathbf{ 0.10\pm 0.02}$\\
&&$\mathit{ 3.76\pm 0.02}$&$\mathit{ -0.02\pm 0.02}$&&$\mathit{ -0.83\pm 0.35}$&&$\mathit{ (0.04\pm 0.03)}$\\
S0&106&$\mathbf{ 3.66\pm 0.02}$&$\mathbf{ -0.16\pm 0.02}$&-24.05&$\mathbf{ 0.37\pm 0.16}$& 0.19&$\mathbf{ 0.18\pm 0.02}$\\
&&$\mathit{ 3.66\pm 0.02}$&$\mathit{ -0.12\pm 0.04}$&&$\mathit{ 0.27\pm 0.17}$&&$\mathit{ 0.14\pm 0.03}$\\
Sa& 94&$\mathbf{ 3.47\pm 0.04}$&$\mathbf{ -0.23\pm 0.03}$&-23.85&$\mathbf{ 1.18\pm 0.23}$& 0.17&$\mathbf{ 0.30\pm 0.03}$\\
&&$\mathit{ 3.49\pm 0.04}$&$\mathit{ -0.21\pm 0.04}$&&$\mathit{ 1.21\pm 0.20}$&&$\mathit{ 0.27\pm 0.04}$\\
Sb&140&$\mathbf{ 3.47\pm 0.03}$&$\mathbf{ -0.17\pm 0.03}$&-24.31&$\mathbf{ 0.97\pm 0.13}$& 0.25&$\mathbf{ 0.30\pm 0.02}$\\
&&$\mathit{ 3.45\pm 0.03}$&$\mathit{ -0.14\pm 0.04}$&&$\mathit{ 0.99\pm 0.13}$&&$\mathit{ 0.29\pm 0.02}$\\
Sbc& 91&$\mathbf{ 3.30\pm 0.03}$&$\mathbf{ -0.28\pm 0.03}$&-24.12&$\mathbf{ 1.22\pm 0.16}$& 0.25&$\mathbf{ 0.23\pm 0.03}$\\
&&$\mathit{ 3.28\pm 0.03}$&$\mathit{ -0.22\pm 0.06}$&&$\mathit{ 1.19\pm 0.21}$&&$\mathit{ 0.24\pm 0.03}$\\
Sc&102&$\mathbf{ 3.13\pm 0.04}$&$\mathbf{ -0.24\pm 0.03}$&-23.44&$\mathbf{ 1.10\pm 0.13}$& 0.29&$\mathbf{ 0.27\pm 0.03}$\\
&&$\mathit{ 3.11\pm 0.03}$&$\mathit{ -0.22\pm 0.04}$&&$\mathit{ 1.17\pm 0.14}$&&$\mathit{ 0.24\pm 0.03}$\\
Sd& 95&$\mathbf{ 2.91\pm 0.04}$&$\mathbf{ -0.14\pm 0.03}$&-22.11&$\mathbf{ 1.15\pm 0.16}$& 0.32&$\mathbf{ 0.33\pm 0.03}$\\
&&$\mathit{ 2.88\pm 0.04}$&$\mathit{ -0.15\pm 0.04}$&&$\mathit{ 1.23\pm 0.16}$&&$\mathit{ 0.33\pm 0.03}$\\
Im& 20&$\mathbf{ 2.52\pm 0.07}$&$\mathbf{ -0.07\pm 0.04}$&-21.60&$\mathbf{ 1.11\pm 0.29}$& 0.25&$\mathbf{ 0.18\pm 0.05}$\\
&&$\mathit{ 2.53\pm 0.06}$&$\mathit{ -0.05\pm 0.05}$&&$\mathit{ 1.09\pm 0.29}$&&$\mathit{ (0.04\pm 0.06)}$\\
\end{tabular}
\caption{$(B-H)_{\mathrm{T}}$ color per type as a function of
$X_{i1}=H_{\mathrm{abs}}$ and $X_{i2}=\log_{10}(R_{25})$.}
\label{CMinclin}
\end{center}
\end{table*}
\subsection{The color vs. $R_{25}$ \& absolute magnitude relation}
We finally give in Table~\ref{CMinclin} the estimators for a
simultaneous fitting of the $(B-H)_{\mathrm{T}}$ colors as a function
of $\log_{10}(R_{25})$ and $H_{\mathrm{abs}}$ in each type.The overall
agreement with the slopes determined previously from the separate
fitting of the colors with respect to either $R_{25}$ or
$H_{\mathrm{abs}}$ is correct. The requirement that $H_{\mathrm{abs}}$
be known tends to select the brightest galaxies and may explain
the small discrepancies in the slope as as a function of $\log_{10}(R_{25})$,
as this one depends probably also on the mass ($\sim
H_{\mathrm{abs}}$) of the galaxy.
The very small scatter obtained for Im galaxies indicates that there
are too few of them for the simultaneous analysis.
\section{Discussion}
\label{discussion}
NIR growth curves of the magnitude as a function of the
aperture have been built from the Catalog of Infrared Observations (Gezari et
al. \cite{Gezari}) and optical catalogs. Using these, we have been able
to compute total NIR apparent and absolute magnitudes, NIR and optical-to-NIR colors, and to
estimate their uncertainties for a large sample. A statistical
analysis of the colors as a function of type, inclination and
luminosity, using estimators taking into account the uncertainties
in the variables and
the intrinsic scatter in the colors, highlights the interest of the NIR and
notably of the comparison of the optical to the NIR.
Optical-to-NIR colors show a well defined sequence with type: the
mean irregular is 1.3 magnitude bluer in $B-H$ than the mean elliptical.
The intrinsic scatter is higher for star-forming galaxies ($\sigma\sim
0.4$) than for
ellipticals or lenticulars ($\sigma\sim 0.1$--$0.2$).
Because of the small extinction in the NIR, the optical-to-NIR colors
of spiral galaxies redden considerably with increasing inclination,
putting therefore constraints on the amount of the dust and the
respective distributions of dust and stars. S0 colors
do not depend on the inclination. Rounder elliptical are redder than
more elongated ones.
The color-absolute magnitude relation is much steeper and tighter
when the NIR is used as the reference band rather than the optical.
Examination of the color vs. $B$ absolute magnitude relation for each type
suggests that this is due to the extinction which dims and reddens
the galaxies and counteracts the color-NIR relation.
A color-magnitude relation exists in each type, with brighter galaxies
in the NIR being redder. The slope is steeper for Sa--Sc than for
early-type galaxies or Sd--Im. The relation we obtain for ellipticals is shallower
than in other studies. A likely explanation
is that this is due to the effect of the color gradient
in small apertures, a lesser, i.e. redder, fraction
of the galaxy being observed in large bright ellipticals
than in small faint ones
-- a problem we do not have using our
total (or we expect so) colors. The slope for S0 galaxies
is closer to that of spirals than to that of ellipticals.
The color at a given absolute magnitude becomes
bluer with increasing type, indicating that both the mass and the type
must be used to characterize the colors and to describe
the star formation history of galaxies.
Once corrected for the inclination and the magnitude effect,
the intrinsic scatter in the colors of spiral types drops
significantly, although a few outliers with very blue colors typical of
starbursting galaxies remain.
However satisfying these results are, some problems still exist.
NIR observations are obtained in small apertures and
the extrapolation to total magnitudes and colors suffers from significant
uncertainties. Nevertheless, we do not believe that our results
strongly depend on this extrapolation since similar slopes
are obtained when we use effective colors -- which are interpolated or
only slightly extrapolated -- instead of total colors.
The selection of galaxies is a more crucial problem.
The only criterion used here has been to take whatever was available
in the NIR.
For this reason, the sample, especially in $J$ and $K$, contains
few late-type galaxies but many cluster galaxies, Seyferts
and other active nuclei.
We also lack of intrinsically faint galaxies. The
color-magnitude relations we have determined should therefore
not be extrapolated at $H_{\mathrm{abs}}$ fainter than $\sim -20$.
Large samples as the Sloan Digital Sky Survey in the optical
or DENIS and 2MASS in the NIR should solve this problem.
The last important problem is the bias between the two estimators
we use, especially in the slope of the color-magnitude relation.
The discrepancy tends to disappear however when we consider
effective colors which have smaller uncertainties. The MCES
estimators seem to be closer to the true slopes than the ML,
but the latter provides a better estimate of the intrinsic
scatter and has a lower variance.
Combining the NIR and the optical enlarges considerably our vision
of the stellar populations and the dust content of galaxies.
In our next paper, we will extend our analysis to the near
ultraviolet-optical wavelength range. We will publish ``color''
energy distributions of galaxies from the near-UV to the NIR
that will be used as templates to constrain the
star formation history of galaxies of different types and masses.
A further step will be the extension of these techniques
to the far-UV and the far-IR.
\begin{acknowledgements}
This research has made use of:
\begin{itemize}
\item
the Catalog of Infrared Observations operated at the NASA/Goddard
Space Flight Center (Gezari et al. \cite{Gezari}),
\item
the Hypercat catalog operated at the CRAL-Observatoire
de Lyon (Prugniel \& H\'eraudeau \cite{PH}),
\item
the Third Reference Catalog of Bright Galaxies
(de Vaucouleurs G. et al. \cite{RC3}),
\item
the NASA/IPAC Extragalactic Database
(NED) which is operated by the Jet Propulsion Laboratory,
California Institute of Technology,
under contract with the National
Aeronautics and Space Administration.
\end{itemize}
We also acknowledge:
\begin{itemize}
\item
the Statistical Consulting
Center for Astronomy operated at the Department of Statistics,
Penn State University, M.G. Akritas (Director),
\item
Numerical Recipes by Press et al. (\cite{NR}).
\end{itemize}
M. F. acknowledges support from the
National Research Council through the Resident Research Associateship Program.
\end{acknowledgements}
|
\section{Introduction}
\label{se:introduction}
Deuterium is an important clue to the physics of the Big Bang. Its
creation rate during primordial nucleosynthesis
depended strongly on the number ratio of photons to baryons, a quantity not at
all well-known but crucial for a correct description of the earliest events
(e.g., Wilson \& Rood 1994; Smith et al. 1993).
Ever since,
conditions have not been right to add further deuterium to the
primordial production; neither nuclear fusion processes nor the spallation of
heavier nuclei by energetic cosmic rays can augment the original abundance
(although stellar flares were suggested by Mullan \& Linsky
[1999] to produce deuterium).
The deuterium abundance in fact decreases continuously as deuterium is
burned up in stars.
The present day deuterium abundance in the interstellar medium
provides a lower limit to its
primordial value, and it
reflects the history of stellar reprocessing of the gas;
measurements of its spatial variation may shed light on
the star formation history in a given region (e.g., Tosi 1998,
Tosi et al. 1998).
Deuterium and D-bearing species were detected in giant planets
(Encrenaz et al. 1996; Feuchtgruber et al. 1997), in cosmic rays
(Beatty et al. 1985; Casuso \& Beckman 1997; Geiss \& Gloeckner 1998),
in the local
ISM (Linsky et al. 1993, 1995; Linsky 1998; Piskunov et
al. 1997; Helmich et al. 1996; Jacq et
al. 1993; Turner 1990), and in extragalactic sources (Songaila et
al. 1994; Tytler et al. 1996; Burles \& Tytler 1998). Through recent
\ion{D}{i} Ly$\alpha$ observations of the local ISM, McCullough
(1992), Linsky et al. (1993, 1995), Piskunov et al. (1997), and
Dring et al. (1997) derived
a deuterium abundance [D]/[H]~$\simeq 1.5 \times 10^{-5}$ along the
line of sight toward nearby
bright stars. These observations are hampered however
by the possible confusion with the \ion{H}{i} Ly$\alpha$ line, which
is only $\simeq$80~km~s$^{-1}$\ apart.
High spectral resolution UV absorption
observations of DI Lyman $\delta$ and $\epsilon$
were recently performed by Jenkins et al. (1999) and Sonneborn et al.
(in prep.) along three
lines of sight, which yield significant [D]/[H] variations between
$7.4\times 10^{-6}$ and $2.1\times 10^{-5}$.
Observations of the hyperfine transitions of \ion{D}{i}
and \ion{H}{i} at radio frequencies (Chengalur et al. 1997), or of the
H$_2$\ and HD molecules, provide further means to sample the
deuterium abundance.
The [HD]/[H$_2$] ratio is
subject to variations due to small differences in chemical
reaction rates in high temperature molecular gas
as in shocks (see Sect.~\ref{se:chemistry}) and photodissociation regions.
To date, rotationally excited HD has been
detected with ISO
from giant planets (Feuchtgruber et al. 1997),
and in the interstellar medium
through UV-observations by Wright \& Morton (1979).
Accounting for the three lowest rotational levels of HD
Wright \& Morton found
[HD]/[H$_2$]~$< 6 \times 10^{-7}$ in the cold molecular gas
toward $\zeta$ Oph. D-bearing species such as HDO, CH$_2$DOH,
DCO$^+$ or DCN (Jacq et al. in prep.), are
observed at radio-frequencies in the ISM, but due to fractionation the
deuterium abundance cannot be deduced accurately from these observations.
The infrared emissivity of HD has been modeled and predicted for
photodissociation regions (Sternberg 1990 - not including the
reaction $\rm H+HD\getsto H_2+D$) and magnetic (C-type)
shocks (Timmermann 1996), where latter calculations include deuterated
species in the chemical reaction network. But even
state-of-the-art IR-spec\-tro\-meters had been unsuccessful in detecting
HD emission in the ISM, because of the low HD abundance, and because the pure
rotational and rotation-vibrational lines appear at wavelengths that are
affected by strong telluric interference.
ISO now for the first time opened the skies to a successful search for
HD emission. Wright et al.~(1999) detected the emission of HD 0-0
R(0) toward the Orion Bar photodissociation region, and derive
[HD]/[H$_2$]$= (2.0\pm 0.6)\times 10^{-5}$. A measure of the
deuterium abundance in PDRs is complicated by the fact
that the HD dissociation front is located deeper into the molecular
cloud than that of H$_2$.
The average excitation of HD may therefore be lower than that
of H$_2$, making it difficult to derive column densities referring to
the same regions. In regions of pure collisional excitation such as
shocks, this problem should not occur.
We here report HD observations toward the brightest H$_2$\ emission
region in the sky, Peak~1 in the Orion
molecular outflow that surrounds several deeply embedded
far-infrared sources
(Genzel \& Stutzki 1989; Menten \& Reid 1995; Blake 1997;
van Dishoeck et al. 1998; Stolovy et al. 1998; Schultz et al. 1998).
Our observations, which we discuss in Sect.~2,
were part of a line survey of shocked
molecular gas, and of a program to investigate the oxygen-chemistry in
the warm molecular interstellar medium with the
Short-Wavelength-Spectrometer (SWS).
In Sect.~3 we derive an HD column density from
the flux of one detected HD line, making use also of extinction
and excitation measurements from a large number of H$_2$\
lines which we detected in related observations.
We then estimate the deuterium abundance in the
shocked, warm gas of the Orion outflow.
In Sect.~4 we summarize our results.
\section{Observations}
\label{se:observations}
\begin{figure*}
\begin{minipage}[t]{0.71\linewidth}
\centering
\includegraphics[width=11cm]{figure1.ps}
\end{minipage}
\begin{minipage}[t]{0.24\linewidth}
\centering
\vspace*{-10.5cm}
\caption{
The various SWS apertures of our ISO observations of the Orion outflow,
overlaid
on a NICMOS H$_2$\ 1--0 S(1) map kindly provided by E.~Erickson and A.~Schultz
(Schultz et al. 1998).
White dots and the central white patch are continuum ghost-images of
stars and of Orion-BN, respectively.
The ISO aperture was centered on
$\rm 05^h 32^m 46^s.27$, $-05\degr ~24\arcmin~ 02\arcsec$ (1950).
Its
size corresponds to $14\arcsec \times 20\arcsec$ for wavelengths
smaller 12 $\mu$m, $14\arcsec \times 27\arcsec$ at 12 to 27.5 $\mu$m,
$20\arcsec \times 27\arcsec$ at 27.5 to 29 $\mu$m, and $20\arcsec
\times 33\arcsec$ at 29 to 45.2 $\mu$m.}
\label{fig:peak1}
\end{minipage}
\end{figure*}
\begin{table*}
\caption[]{
Summary of ISO-SWS observations of HD lines at Orion Peak~1.
}
\begin{tabular}{crcrcrrrrr}
\hline\noalign{\smallskip}
\multicolumn{1}{c}{transition} &
\multicolumn{1}{c}{$\lambda^{~~a}$} &
\multicolumn{1}{c}{integration$^b$} &
\multicolumn{1}{c}{peak} &
\multicolumn{1}{c}{$\Delta v~^c$} &
\multicolumn{1}{c}{$A$} &
\multicolumn{1}{c}{$E_{vJ}$} &
\multicolumn{1}{c}{$I_{\rm obs}$} &
\multicolumn{1}{c}{$A_{\lambda}$} &
\multicolumn{1}{c}{$N_{vJ}$$^d$} \\
\multicolumn{1}{c}{ } &
\multicolumn{1}{c}{$\mu$m} &
\multicolumn{1}{c}{sec} &
\multicolumn{1}{c}{Jy} &
\multicolumn{1}{c}{km~s$^{-1}$} &
\multicolumn{1}{c}{s$^{-1}$} &
\multicolumn{1}{c}{K} &
\multicolumn{1}{c}{erg~s$^{-1}$cm$^{-2}$sr$^{-1}$} &
\multicolumn{1}{c}{mag} &
\multicolumn{1}{c}{cm$^{-2}$} \\
\noalign{\smallskip}
\hline\noalign{\smallskip}
0--0 R(2) & 37.7015 & & $<$ 192 & 300 & 1.72(--6)$^e\!\!\!$ & 765.9& $<$9.9(--4)& 0.09 & $<$ 1.5(17)\\
0--0 R(3) & 28.5020 & & $<$ 100 & 350 & 4.11(--6) & 1270.7 & $<$ 7.9(--4) & 0.30 & $<$ 4.6(16)\\
0--0 R(4) & 23.0338 & & $<$ 35 & 350 & 7.91(--6) & 1895.3 & $<$ 6.0(--4) & 0.51 & $<$ 1.8(16)\\
0--0 R(5) & 19.4305 & 3010 & 2.37 & 134 & 1.33(--5) & 2635.8 & $(1.84\pm0.4)$(--5) & 0.60 & $(3.0\pm 1.1)$(14)\\
0--0 R(6) & 16.8940 & & $<$ 8.5 & 230 & 2.03(--5) & 3487.5 & $<$ 1.3(--4) & 0.53 & $<$ 1.1(15)\\
0--0 R(7) & 15.2510 & 1110 & $<$ 5.0 & 185 & 2.88(--5) & 4445.3 & $<$ 6.8(--5) & 0.41 & $<$ 3.3(14)\\
0--0 R(8) & 13.5927 & & $<$ 10 & 215 & 3.87(--5) & 5503.8 & $<$ 1.8(--4) & 0.37 & $<$ 5.6(14)\\
0--0 R(9) & 12.4718 & & $<$ 8 & 230 & 4.97(--5) & 6657.5 & $<$ 1.7(--4) & 0.54 & $<$ 4.3(14)\\
1--0 P(4) & 3.0690 & & $<$ 0.2 & 280 & 7.37(--6) & 5958.5 & $<$ 2.8(--5) & 1.11 & $<$ 2.0(14)\\
1--0 P(3) & 2.9800 & & $<$ 0.1 & 270 & 1.09(--5) & 5593.5 & $<$ 1.4(--5) & 1.08 & $<$ 6.4(13)\\
1--0 P(2) & 2.8982 & & $<$ 0.25 & 300 & 1.64(--5) & 5348.7 & $<$ 3.9(--5) & 0.94 & $<$ 1.0(14)\\
1--0 P(1) & 2.8225 & & $<$ 0.4 & 305 & 3.23(--5) & 5225.7 & $<$ 6.6(--5) & 0.80 & $<$ 7.6(13)\\
1--0 R(0) & 2.6900 & & $<$ 0.2 & 330 & 1.72(--5) & 5348.7 & $<$ 3.7(--5) & 0.70 & $<$ 7.0(13)\\
1--0 R(1) & 2.6326 & & $<$ 0.3 & 330 & 2.51(--5) & 5593.5 & $<$ 5.7(--5) & 0.70 & $<$ 7.3(13)\\
1--0 R(2) & 2.5811 & & $<$ 0.4 & 245 & 3.22(--5) & 5958.5 & $<$ 5.8(--5) & 0.72 & $<$ 5.7(13)\\
1--0 R(3) & 2.5350 & 1200 & $<$ 0.2 & 215 & 3.91(--5) & 6441.1 & $<$ 2.6(--5) & 0.74 & $<$ 2.1(13)\\
1--0 R(4) & 2.4943 & & $<$ 0.35 & 235 & 4.60(--5) & 7037.7 & $<$ 5.0(--5) & 0.76 & $<$ 3.5(13)\\
\noalign{\smallskip} \hline
\noalign{\smallskip}
\end{tabular}
$^{a}$ expected wavelengths. 0-0 R(5) is observed at $19.4290~\mu$m. \\
$^{b}$ on target integration time for SWS 02 observations; the SWS~01
2.4 -- 40 $\mu$m scan took 6538 sec. \\
$^{c}$ FWHM of observed HD line, or of neighboring H$_2$ lines.\\
$^{d}$ upper level column, corrected for extinction.\\
$^{e}$ numbers in brackets denote powers of ten.\\
\label{tab:lines}
\end{table*}
Orion Peak~1 was observed in the SWS~01 (full grating scan) and SWS~07
(Fabry-P\'erot) modes of the short wavelength spectrometer (de Graauw
et al. 1996) on board ISO (Kessler et al. 1996) on October 3, 1997,
and in the SWS~02 ($\approx 0.01~\lambda$ range grating scan)
mode on September 20, 1997 and February 15, 1998.
Figure \ref{fig:peak1}
illustrates the various aperture orientations with respect to the H$_2$\
1--0 S(1) emission observed with NICMOS on the HST (Schultz et
al. 1998). The full 2.3 to 45~$\mu$m SWS~01 spectrum was recorded in
its slowest mode with the highest possible resolution. A preliminary
reduction of this spectrum was presented by Bertoldi (1997).
Table~\ref{tab:lines} summarizes the HD line observations. The H$_2$\
lines will be discussed in more detail in a forthcoming article
(Rosenthal et al. in prep.).
The data reduction was carried out using standard Off Line Processing
(OLP) routines up to the Standard Processed Data (SPD) stage within
the SWS Interactive Analysis (IA) system. Between the SPD and Auto
Analysis Result (AAR) stages, a combination of standard OLP and
in-house routines were used to extract the individual spectra. The
in-house routines included an interactive dark-current subtraction for
individual scans and detectors as well as for the removal of
fringes. The flux calibration errors range from 5\% at 2.4~$\mu$m\ to
30\% at 45~$\mu$m. (SWS-Instrument Data User Manual, version 3.1). The
statistical uncertainties derived from the line signal to noise ratio
are for most detected lines smaller than the systematic errors due to
flux calibration uncertainties.
\subsection{Line Fluxes}
\label{se:flux}
\begin{figure*}
\resizebox{\hsize}{!}{\includegraphics{figure2.ps}}
\vspace{-0.9cm}
\caption{
Spectra of HD non-detections.
Three of the spectra were obtained in the
SWS~02 mode with ISO, all others were taken in the SWS~01 mode.
Neighboring H$_2$\ and \ion{H}{I}
lines are identified at their expected positions.
}
\label{fig:hd_spec}
\end{figure*}
The spectra at the expected wavelengths of seventeen HD transition lines
are shown in Figs.~\ref{fig:hd_spec} and \ref{fig:00R5}.
The wavelengths
(Table \ref{tab:lines}) of the $v=0-0$ pure rotational HD transitions
were adopted from Ulivi et al. (1991), while those for the
rotation-vibrational lines were computed from the rotational
(Essenwanger \& Gush 1984) and vibrational
constants (Herzberg 1950). Only one of the seventeen lines
sought was detected, the 0--0 R(5) (i.e. $J=6\rightarrow 5$) dipole
transition at an expected wavelength of 19.4305~$\mu$m\footnote{
The experimental value for the R(5) transition wavelength is
$19.4305\pm0.0001\mu$m, whilst the calculated value is
$19.431002\pm0.000008\mu$m (Ulivi et al. 1991 and references therein).
The discrepancy is yet unexplained.
}
(observed at
19.4290 $\mu$m, see Fig.~\ref{fig:00R5}).
Although the detection may seem marginal, the line does
appear in two independent observations. It is apparent in the
fringed data, and after defringing,
it stands out well above the noise in the
central wavelength range of the coadded SWS~02 scans.
A beam-averaged flux of $(1.84 \pm 0.4) \times
10^{-5}$erg~s$^{-1}$cm$^{-2}$sr$^{-1}$\ was derived
from a continuum-subtracted integration over the feature;
the error derives from
the line's S/N$=4.5$ (the RMS noise being evaluated within
$\pm 500~{\rm km~s^{-1}}$)
plus an estimated 11\% flux calibration uncertainty.
The line width (FWHM)
is 134~km~s$^{-1}$, which agrees with the instrumentally expected width
for extended objects, 130~km~s$^{-1}$, and is not sufficient to resolve
the emission, which should have a velocity dispersion similar
to that observed in H$_2$ 1--0 S(1), which has a
FWHM of $\approx 50$ km~s$^{-1}$, with emission from -100 to +100 km s$^{-1}$
(Chrysostomou et al. 1997; Stolovy et al. 1998).
The line center is positioned at
$v_{\rm helio} \simeq -23$~km~s$^{-1}$, which is within the
range of line center
velocities, $v_{\rm helio} \simeq -38$ to +41~km~s$^{-1}$, observed in the
H$_2$\ 1--0 S(1) emission (Chrysostomou et al. 1997). The high
spectral resolution and sensitivity of the ISO SWS
around 19~$\mu$m made $0-0$ R(5) the
most promising line for a detection of HD in Peak 1. The stronger lines
from lower
rotational states suffer from the rapidly rising
continuum level at longer wavelengths and the resulting strong fringing.
\begin{figure}
\epsfig{file=figure3.ps,width=8.7cm,
clip=,bbllx=2pt,bblly=12pt,bburx=497pt,bbury=555pt}
\caption{Defringed AOT~02 spectra of two 0--0 R(5) observations performed at the
labeled dates, and the merged spectrum.
The line is visible even in the
original fringed data of both observations. Continuum levels
differed in both datasets (due to differing orientations of the aperture
with respect to Orion-BN), so that a constant was added to one before averaging
them, and in order to display them together.
The rising noise near the scan edges is expected and due to the lower
detector coverage. In the merged spectrum, the line stands out with
a peak flux density to RMS noise ratio of 4.5.
$v_{\rm helio}$ refers to the expected line wavelength of 19.4305 $\mu$m.}
\label{fig:00R5}
\end{figure}
The available
integration time was insufficient to detect two other HD lines
which we tried to observe in the SWS~02 mode.
We derived upper flux limits for a total of
sixteen HD lines from the noise level near the expected line
wavelengths (Table~\ref{tab:lines}). The peak-to-peak noise envelope
approximately corresponds to a 3$\sigma$ flux density dispersion, and
we assumed that a line with a peak flux density of 3$\sigma$ would
have stood out clearly enough to be detected. As upper flux limits we
therefore adopted the 3$\sigma$ flux density noise level
times the FWHM of neighboring H$_2$\ lines, except for 0--0 R(7), R(8),
R(9), for
which we adopted the linewidths expected from instrumental resolution.
\section{Extinction correction and H$_2$\ excitation}
\label{se:excitation}
The column density of molecules in a particular rotation-vibrational level
$(v,J)$ is computed from the observed line flux,
$I_{obs}(v,J\rightarrow v',J')$, of a transition to
a lower state $(v',J')$, times
the wavelength, $\lambda$, divided by the radiative transition rate
(Einstein) coefficient\footnote{
The Einstein coefficients for H$_2$\ and HD are taken
from Turner et al.~(1977), confirmed by Wolniewicz et
al. (1998), and Abgrall et al. (1982), respectively. The
transition energies for H$_2$\ were computed from the level
energies that were kindly provided by Roueff (1992, private
communication).}
for this transition, $A(v,J\rightarrow v',J')$, times a correction for
extinction along the line of sight:
\begin{equation}
N_{vJ} ~=~ { 4 \pi\over h c}~{ \lambda~ I_{obs}(v,J\rightarrow v',J')
\over A(v,J\rightarrow v',J') }
~~10^{0.4 A_{\lambda}}~,
\end{equation}
where $A_{\lambda}$ is the effective extinction at wavelength $\lambda$.
All H$_2$\ and HD transitions are optically thin.
Previous measurements of the Peak 1 molecular emission
estimated K band (2.12~$\mu$m) extinctions between 0.5 and 1 magnitude
(Everette et al. 1995).
To apply an extinction correction to the observed H$_2$\ and HD lines
requires knowledge of the ``extinction law" $A_\lambda/A_{\rm K}$ between
2.4 and 40~$\mu$m, which especially above $\sim 5$~$\mu$m
is observationally not well constrained
(Draine 1989). The near-IR extinction is usually approximated
to follow a power law
(see Fig.~\ref{fig:extinction}),
\begin{equation}
A_\lambda = A_{\rm K} \;
\left(~\lambda~/~2.12~ \mu{\rm m}~\right)^{-\epsilon}
~~~~~~~{\rm for}~~\lambda < 6~\mu{\rm m},
\label{eq:dust}
\end{equation}
with an increase in opacity beyond 6~$\mu$m due to stretch and bend mode
resonances in silicate grains (Draine \& Lee 1984), the shape and
depth of which are yet poorly understood. A recent
ISO study of the extinction toward the W51 \ion{H}{ii} region, e.g.,
finds $A_{19}/A_{10}\simeq 0.52 \pm 0.1$ and $A_{19}/A_{\rm K}\simeq
0.57\pm 0.1$
(Bertoldi et al. in prep.). Draine \& Lee (1984) suggested
$A_{18}/A_{9.7}\simeq 0.40$, and observations of circumstellar dust
emissivities suggest values for this ratio
between 0.35 (Pegourie \& Papoular 1985) and 0.5 (Volk \& Kwok 1988).
The effective extinction toward the emitting gas in Peak 1 may not
necessarily follow an average interstellar extinction law, since the
emitting and absorbing gas may be mixed. We therefore tried to
estimate the effective extinction as a function of wavelength toward
Peak 1 from the observed emission of Peak 1 itself. The differential
extinction between two wavelengths can be derived from a comparison
of H$_2$\ line fluxes of transitions arising either from the same upper
level, or from neighboring thermalized levels with a well determined
relative excitation. Since the warm HD and H$_2$\ are likely to be well
mixed, we will use the extinction toward the H$_2$\ to deredden also the
HD line intensities.
The extinction-corrected H$_2$\ column density distribution serves as a
``thermometer'' that probes the excitation conditions in the emitting
gas. Since we detected only one HD line, we cannot determine the HD
excitation directly --this would require at least two lines of
transitions arising from different upper levels. Instead, we rely on
the reasonable assumption that HD is subject to the same excitation
conditions as H$_2$. We can thereby derive both molecules' total column
densities in the warm, emitting gas, and from that, their abundance
ratio.
\begin{figure}
\epsfig{file=figure4.ps,width=8.7cm,
clip=,bbllx=8pt,bblly=28pt,bburx=490pt,bbury=485pt}
\caption{
Adopted extinction curve (solid line), including water ice and silicate
features.
Dashed lines show curves with different relative silicate features strengths,
$A_{19}/A_{10}= 0.35$ and 0.52. The dash-dotted curve assumes narrower
9.7~$\mu$m
feature. Wavelenghts of those
H$_2$ lines used to constrain the extinction curve are
marked -- but these are not datapoints!}
\label{fig:extinction}
\end{figure}
\subsection{2.4 -- 6~$\mu$m extinction}
\label{se:dust}
It is useful to plot the observed H$_2$\ or HD column densities, $N_{vJ}$,
divided by the state's statistical weight,\footnote{
The statistical weights are $g_{J}=
2J+1$ for para-H$_2$\ (even $J$) and $g_{J} = 3(2J+1)$ for
ortho-H$_2$ (odd $J$).
HD has non-identical nuclei (and a small dipole moment
$\sim 10^{-4}$~Debye). Therefore,
unlike H$_2$, which exists in the form
para-H$_2$\ and ortho-H$_2$, such a difference does not exist for
HD, and $g_J=2J+1$.}
$g_J$, against the state's energy, $E_{vJ}$. Examining this
``excitation diagram,'' which we first
plotted from the not yet extinction-corrected H$_2$\
line intensities, we found no indication up to $E_{vJ}/k\approx
40,000$~K of
fluorescent excitation, or for deviations from the statistical ortho-
to para-H$_2$\ abundance ratio of 3 (details will be given by
Rosenthal et al. in prep.). Instead, the
populations appear both in thermodynamic and statistical
equilibrium, which means, they depend only on the level energy
and degeneracy, not on the vibrational quantum number. As a consequence,
$N_{vJ}/g_J$ is a smooth function of the level energy.
In contrast, H$_2$\ that is fluorescently excited
generally displays level populations that are not
in vibrational {\it and} rotational LTE,
and also show ortho-to-para column density
ratios between neighboring states that are smaller than 3 (see e.g., Draine
\& Bertoldi 1996, Bertoldi 1997). Timmermann (1998) predicted
that in low-velocity shocks the ortho-to-para ratio can also be
significantly lower, which was confirmed in recent ISO-SWS
observations by Neufeld et al. (1998), who found a value of
$\simeq 1.2$ in the outflow HH54. Peak 1 however shows no deviations
from statistical equilibrium.
We were able to obtain the most reliable line fluxes
for the H$_2$\ $v=0-0$ and
$v=1-0$ lines between 2.4 and 6~$\mu$m that have upper level energies
$E_{vJ}/k = 5000 - 16,000$~K. Excluding lines in the water
ice absorption feature between 2.8 and 3.3~$\mu$m,
we selected thirty-two high S/N lines
in this wavelength and energy range to constrain the power law part of
the extinction curve.
Assuming that the H$_2$\ columns
$N_{vJ}/g_J$ vary smoothly with $E_{vJ}$, we fit a
second order polynomial to the observed column densities,
corrected for extinction
following Eq.~(1) with $A_{\rm K}$ and $\epsilon$ as free parameters. We
searched for values of $A_{\rm K}$ and $\epsilon$ which minimize the
dispersion of $N_{vJ}/g_J$.
Within the range $\epsilon= 1-2$,
$A_{\rm K}$ is well constrained to $1.0\pm 0.1$ magnitudes. It was not
possible to narrow down the value of $\epsilon$ further, so we adopted
$\epsilon=1.7$, which was suggested by previous observational
studies (see Draine 1989; Brand et al. 1988 used $A_{\rm K}=0.8$
and $\epsilon=1.5$).
The second order fit to $N_{vJ}/g_J$ versus $E_{vJ}/k\equiv 1000~ T_3$~K
for $A_{\rm K}=1.0$ and $\epsilon=1.7$ is (see Fig.~\ref{fig:h2_ex})
\begin{equation}
\log(N_{vJ}/g_J)~ =~ 18.88 - 0.402 \: T_3 + 0.0092 \: T_3^{2},
\label{eq:fit2}
\end{equation}
for $5800{\rm ~K}<E_{vJ}/k<17,000$~K, and has a fit quality $\chi^2=0.13$;
the uncorrected ($A_{\rm K}=0$) dispersion is $\chi^2=0.39$.
We should note that our extinction curve between 4 and 7 $\mu$m\ is not
constrained well enough to address the claim by Lutz et al. (1997)
that the extinction curve flattens and lacks the presumed minimum at
7 $\mu$m.
\begin{figure}
\psfig{file=figure5.ps,width=8.7cm,
clip=,bbllx=45pt,bblly=70pt,bburx=520pt,bbury=790pt}
\caption{
H$_2$\ excitation diagram: the dots, diamonds, and squares denote
dereddened and beam-averaged column densities of transitions
in the $v$ = 0--0, 1--0, and 2--1 bands, respectively. The solid line
represent the least squares fit, Eqs.~(\ref{eq:fit2}) and (\ref{eq:fit1}),
which apply for $E_{vJ}/k$ larger and smaller than 5800~K, respectively.}
\label{fig:h2_ex}
\end{figure}
\subsection{Mid-IR extinction}
\label{se:mid-IR}
Having constrained the extinction curve below 6~$\mu$m and the H$_2$\
excitation from 5800 -- 17,000~K, we now estimate the strength of the
10~$\mu$m silicate feature with the observed H$_2$\ $v=0$, $J\le 8$ column
densities.
The intensity of the 0--0 S(3) line at 9.66~$\mu$m gives a
column density of the $v=0$, $J=5$ level that is significantly
below that expected from an interpolation between the $J=3$
(0--0 S(1) 17.0~$\mu$m) and $J=8$ (0--0 S(6) 6.1~$\mu$m) levels.
The observed $J=3$ to $J=7$ level populations are all affected by silicate
absorption. We estimate the shape of the 9.7~$\mu$m and 18~$\mu$m
silicate features from the calculations of Draine \&
Lee (1984), and adjust the strengths of the features
individually (Fig.~\ref{fig:extinction}).
Assuming $A_{\rm K}=1.0$ mag and tuning $A_{9.7}$
such that the $J=5$ level column density smoothly
follows that of the $J=3-8$ levels (Fig.~\ref{fig:h2_ex} shows the
resulting dereddened columns) results in a possible range $A_{9.7} =
(1.3-1.5)$~mag, if we allow for a variation in
$A_{19}/A_{10}=0.35-0.52$, and a 15\% uncertainty of
the $J=5$ column due to the difference between two
measurements of the
0--0 S(3) line in different AOTs. When we narrow the 10~$\mu$m
feature below reasonable width estimates, the extinction correction for
the $J=4$ and $J=6$ levels decreases significantly,
thereby lowering the dereddened flux
in the 0--0 S(3) line and decreasing the required extinction
to $A_{9.7}\approx (1.1-1.3)$~mag.
Considering all uncertainties we estimate $A_{9.7}\approx
(1.35\pm 0.15)$~mag, and with $A_{19}/A_{10}\approx 0.35 - 0.52$, we
find $A_{19} \approx (0.61 \pm 0.15)$~mag. Our value for
$A_{19}/A_{\rm K} = (0.61 \pm 0.15)$ is in good agreement with the
$0.57 \pm 0.1$ found toward W51.
The extinction correction for the detected HD
line we then estimate as $10^{0.4 A_{19}} \approx 1.75 \pm 0.25$.
The dereddened column densities of the H$_2$\ $v=0$, $J\le 8$ levels are
fit by
\begin{equation}
\log\left(N_{vJ}/g_J\right) ~=~ 20.17 - 0.765 \: T_3 +
0.0344 \: T_3^{2}.
\label{eq:fit1}
\end{equation}
At upper level energies
$E_{Jv} / k = 0$, 2636~K (HD $J=6$) and 5800~K this corresponds to
excitation temperatures $T_{ex} \simeq 570$, 740 and 1190~K, respectively.
\subsection{Water ice feature}
\label{se:water}
The rotation-vibrational HD transitions 1--0 P(2), P(3), and P(4)
fall into the water ice absorption feature at $3.05\pm 0.25$~$\mu$m.
To correct these line flux limits for extinction
we estimated the depth and width of the feature from the apparently
enhanced (over the power law, Eq.~[\ref{eq:fit2}]) extinction of
five H$_2$\ lines
with $6000{\rm ~K}<E_u/k<\rm 16,000$~K and 2.8~$\mu$m~$<\lambda<$~3.3~$\mu$m.
The Gaussian
\begin{equation}
\Delta A_{ice}(\lambda)~ \approx ~0.58
~e^{-[(\lambda-3.05{\rm \mu m})/(\sqrt2\cdot 0.15{\rm \mu m})]^2}~
{\rm mag}
\end{equation}
approximately fits the additional
extinction noticeable for the five H$_2$\ lines, and was used
to correct the rovibrational HD line fluxes.
\begin{figure
\psfig{file=figure6.ps,width=8.7cm,
clip=,bbllx=45pt,bblly=60pt,bburx=525pt,bbury=780pt}
\caption{
HD excitation diagram. Pure rotational transitions are denoted by
dots, and $v$ = 1--0 transitions by diamonds. The
line represents the fit Eq.~(\ref{eq:fit3}).
The error of the 0--0 R(5) line
is computed from spectral noise (22\%) and
uncertainties in the flux calibration (11\%)
and the extinction at 19.4~$\mu$m ($\simeq$14\%).
}
\label{fig:hd_ex}
\end{figure
\subsection{Total column of $\rm H_2$ and {\rm HD}}
\label{se:LTE}
How can we derive the total HD column density with only one measured
high-excitation level? We can make use of the observed H$_2$\
excitation, and estimate that the excitation conditions -- i.e., the
fractions of the total gas column density that are at a particular
temperature and density -- are the same for HD and for H$_2$. We assume
that at least up to the excitation energy of HD $J=6$
($E_{06}/k=2636$~K), the H$_2$\ level populations are thermalized, and
thereby reflect the kinetic temperature distribution of the gas.
The assumption of thermalized level populations is
supported by the lack of
deviations from vibrational degeneracy or non-statistical
ortho-to-para ratios, and by detailed non-LTE
calculations of the H$_2$\ level populations, which show that deviations
from LTE are small at $J\le 8$ ($E/k\le 5800$~K)
for densities higher than $10^5~{\rm cm^{-3}}$ and
temperatures above 600~K (Draine \& Bertoldi, unpublished).
Because they are permitted dipole
transitions, the radiative decay rates
of HD are much higher than those of H$_2$\
levels at comparable energy.
This results in ``critical'' densities (i.e. densities
above which a level is
thermalized for a given temperature) which are higher for HD than for H$_2$.
If the gas density was high enough to thermalize the
HD levels, then HD would show the same level excitation as H$_2$:
normalizing
the H$_2$\ populations (\ref{eq:fit1}) with the measured HD $J=6$ level,
we would then expect the HD populations to follow
\begin{equation}
\log(N_{vJ}/g_J) ~=~ 15.14 - 0.765 ~ T_3 + 0.0344 ~ T_3^2~
\label{eq:fit3}
\end{equation}
for $T<6000$~K.
The measured upper limits to sixteen HD level column densities
are consistent with this distribution (Fig.~\ref{fig:hd_ex}).
However, deviations from LTE are important at the
expected density of $10^5-10^6~{\rm cm^{-3}}$ and temperatures of $600-1000$~K in the
shocked gas of the OMC-1 outflow.
Let $n_{vJ}/n_{{\rm LTE},vJ}$ be the actual non-LTE
population in level $(v,J)$, divided
by its LTE value.
The total warm
HD column density is given by the sum over all level column densities
\begin{equation}
N({\rm HD}) ~=~ \frac{n_{\rm LTE,06}}{n_{06}} ~
\sum_{vJ} \left(N_{vJ}\over g_J\right)
~(2J+1) ~ \frac{n_{vJ}}{n_{{\rm LTE},vJ}},
\label{eq:HDtot}
\end{equation}
where $N_{vJ}/g_J$ is given by Eq.~(\ref{eq:fit3}).
\begin{figure}
\epsfig{file=figure7.ps,width=8.7cm,
clip=,bbllx=35pt,bblly=25pt,bburx=440pt,bbury=465pt}
\caption{
Deviations from LTE HD level populations plotted against the rotation
quantum number $J$ for different densities and kinetic gas
temperatures. $n_{0J}$ and $n_{{\rm LTE},0J}$ refer to densities in an
individual $v=0$ state under non-LTE and LTE
conditions.
}
\label{fig:hd_pop}
\end{figure}
To compute the deviations from LTE of the HD level population we solved the
equations of statistical equilibrium (see Timmermann 1996), including
radiative decay and excitation/deexcitation through collisions with He,
H$_2$, and H. The abundance of H varies throughout the shock, and we
adopted [H]/[H$_2$]$=1$, which is typical for the dissociation fraction
in the hot layers of a partially dissociative shock.
In Fig.~\ref{fig:hd_pop} we compare the resulting level
population to that in LTE for gas volume densities of $3\times 10^5$ and
$3\times 10^6{\rm cm^{-3}}$, and three different kinetic temperatures that should
span the range typical of the observed, shocked gas.
HD collisional (de)excitation rate coefficients with H$_2$ were
computed properly only for pure rotational transitions up to $J=4$ and
temperatures up to 600 K and 300 K, respectively (Sch\"afer 1990).
The rate coefficients for higher level transitions and temperatures were
extrapolated from the lower ones. The H--HD and H$_2$--HD collision rate
coefficients were recently computed by Roueff \& Flower (1999) and
Roueff \& Zeippen (1999), respectively.
We find that for H$_2$\ densities below $10^6~{\rm cm^{-3}}$ the
HD level populations
deviate significantly from LTE. At a density of $3\times 10^5~{\rm cm^{-3}}$
and a temperature
of 600~K, the population of the $J=6$ level is about a factor two
below its LTE value.
\begin{figure}
\epsfig{file=figure8.ps,width=8.7cm,
clip=,bbllx=10pt,bblly=20pt,bburx=495pt,bbury=465pt}
\caption{Total HD column density computed from non-LTE level populations
as a function of H$_2$\ density for three different gas temperatures.
The H$_2$\ density in the preshock gas has been
estimated at $(1-3.5)\times 10^{5}\rm cm^{-3}$, and the
shocked gas temperatures at $\approx 600-1000$~K, so we estimate $N({\rm HD})
\approx (2.0\pm0.75)\times 10^{16}\rm cm^{-2}$.
}
\label{fig:NHD}
\end{figure}
To assess the non-LTE effects on the derived total HD column density,
we evaluated $N({\rm HD})$ from Eq.~(\ref{eq:HDtot})
and plot this in Fig.~\ref{fig:NHD} as a function
of the gas density for three different temperatures.
Only at H$_2$\ densities above $10^6~{\rm cm^{-3}}$
do the populations assume LTE, and the warm HD column is
\begin{equation}
N({\rm HD})_{\rm LTE} ~=~ (1.36\pm 0.38)\times 10^{16}\rm cm^{-2}~.
\end{equation}
The error derives from the uncertainty of the $J=6$ column
measured by the 0--0 R(5) line, and is a combination of
spectral noise (22\%), flux calibration uncertainty (11\%),
and the uncertainty of the extinction correction (14\%).
At lower densities, the total column density
much depends on the gas temperature,
and since H is the strongest collision partner of HD,
on the dissociation fraction.
The emission of the OMC-1 outflow
had been modeled previously with C-shocks that propagate at velocities
of order $35-40~{\rm km~s^{-1}}$ into gas with
densities of $n({\rm H_2})=(1-3.5)\times 10^5~{\rm cm^{-3}}$
(Draine \& Roberge 1982;
Chernoff et al.~1982; Kaufman \& Neufeld 1996).
The gas in
such shocks reaches temperatures of order 1000~K,
and remains at approximately constant density through the region where
most of the H$_2$ emission occurs.
The lowest H$_2$\ levels we observed
show excitation temperatures of 600--700~K, which probably
reflects the kinetic temperature of much of the warm, emitting gas.
Taking these temperatures and densities as estimates for the prevailing
excitation conditions, we estimate that the total observed
HD column density must be in the range
\begin{equation}
N({\rm HD}) ~=~ (2.0\pm 0.75)\times 10^{16}\rm ~ cm^{-2}~.
\end{equation}
If we had neglected H--HD collisions, the HD level would be less thermalized,
and our estimate for the total HD column would rise to
$(3.5\pm 1.4)\times 10^{16}\rm ~ cm^{-2}$. However, we believe that
much of the emission arises from partially dissociative shocks, so that
collisions with neutral hydrogen should be important.
We now compare the HD column to that of warm H$_2$.
Summing over the H$_2$\ level populations given by
the least squares fits, Eqs.~(\ref{eq:fit2}) and
(\ref{eq:fit1}), we find
\begin{equation}
N({\rm H_2})~=~(2.21\pm 0.24)\times 10^{21}\rm~ cm^{-2}~,
\end{equation}
where the error reflects a maximum flux calibration uncertainty of 11\%
in the 7 to 19.5~$\mu$m range. By summing from $J=0$, we
extrapolated the observed H$_2$\ level populations, $J\ge 3$, to the
unobserved levels $J=0-2$. We thereby account only for the
{\it warm} H$_2$, not for
the {\it total}~ H$_2$\ column along the line of sight,
which includes over
$10^{22}\rm ~cm^{-2}$ of cold gas in the molecular cloud that
embeds the outflow.
In this cold gas, which we are not concerned with, most H$_2$\ is
in its ground states $J=0$ and $J=1$, and does
not affect our analysis of the
warm outflow gas seen in the higher levels.
Dividing the HD and H$_2$\ column densities, we derive a first estimate of the
abundance ratio
\begin{equation}
{\rm [HD]/[H_2]}~=~(9.0\pm 3.5)\times 10^{-6}~.
\end{equation}
\subsection{Chemical depletion of HD in shocks}
\label{se:chemistry}
\begin{figure}
\psfig{file=figure9.ps,width=8.7cm, height=11.cm,
clip=,bbllx=45pt,bblly=90pt,bburx=535pt,bbury=710pt}
\vspace{-0.5cm}
\caption{Example of the density and temperature structure
in a planar magnetic C-type shock, which was
modeled with the code of Timmermann (1996, 1998). The
pre-shock densities $n({\rm H_2}) = 2.5 \times 10^5\rm ~cm^{-3}$ and $n({\rm
HD}) = 7.5$~cm$^{-3}$, the shock speed $u_{\rm s} =
35$~km~s$^{-1}$, and $B_0=700\mu$G.
The depletion of HD relative to H$_2$\ reaches a maximum
value of 2.4 behind the hottest layer of the shock.
}
\label{fig:hd_shock}
\end{figure}
The OMC-1 outflow emission arises from warm
molecular gas that is shock-heated by
high-velocity ejecta which originates from one of
the deeply embedded protostars in the vicinity of Orion IRc2.
The emission is
a mixture from fast, dissociative J-type shocks, in which the molecular emission
comes from where molecules reform, and from slower
($<50\rm ~km~s^{-1}$), partially- or non-dissociative C-shocks, in which
the molecules radiate in a magnetic precursor where
the temperature rises to its peak value (Fig. \ref{fig:hd_shock}).
In J-shocks that propagate into a medium with
density of order $10^5-10^6\rm cm^{-3}$, the fraction of the bulk motion
energy radiated away in H$_2$ lines is very small, of
order 0.1\% (Neufeld \& Dalgarno 1989).
In C-type shocks, however, about half the kinetic energy is converted
to H$_2$ emission. For J-type shocks to dominate the H$_2$ and HD
emission, about 1000 times more energy would have to be dissipated
in fast J-shocks than in the slower C-shocks. For an even
distribution of magnetic field strengths and
shock velocities up to $\sim 100~$km~s$^{-1}$, C-shocks therefore
vastly dominate the molecular emission.
The outflow emission may therefore be well modeled as arising from such
magnetic, partially-dissociative shocks.
Most of the deuterium is locked in HD in the dense pre-shock
molecular gas. In a high-velocity C-shock, both HD and H$_2$\ are
either dissociated through collisions with ions
that stream through the neutral gas at a speed of order the shock velocity,
or through collisions with warm H$_2$\ and H at temperatures in excess of
2500~K. In a partly dissociative C-shock,
HD is depleted more than H$_2$\ because atomic hydrogen
can efficiently destroy HD through the reaction
\begin{equation}
{\rm D + H_2}\getsto {\rm HD + H} + \Delta H_0,
\label{eq:hddest}
\end{equation}
where $\Delta H_0/k = 418$~K is the enthalpy
difference. Figure \ref{fig:hd_shock}
illustrates the density profiles for H$_2$\ and HD across such a C-shock.
In the warmest layers, HD is depleted by up to a factor 2.4
relative to H$_2$. In the cooling region, the atomic
deuterium reacts stronger again with H$_2$\ to form HD,
and the equilibrium of Eq.~(\ref{eq:hddest}) is shifted back
towards HD. Averaged over the region where the
temperature exceeds 400~K, from where
most of the observable emission arises,
HD is depleted relative to H$_2$\ by a factor 1.67 in the particular case
we display in Fig.~\ref{fig:hd_shock}.
In C-shocks propagating at velocities below 25~km~s$^{-1}$, H$_2$\ is not
dissociated significantly, and because of the low abundance
of atomic hydrogen, the chemical depletion of HD is negligible.
\subsection{Deuterium abundance}
\label{se:abundance}
To derive the deuterium abundance in the warm, shocked gas
we adopt the non-LTE level distribution of HD we computed above,
and further account for
the possibility of chemical depletion. With the column density ratio
derived for temperatures of 600--900~K and densities of
$(1-3.5)\times 10^5~{\rm cm^{-3}}$, we found
\begin{equation}
{\rm [D]/[H]= 0.5 [HD]/[H_2]} = (4.5 \pm 1.7)\times 10^{-6}~,
\end{equation}
and accounting for chemical depletion of HD by 1.67, the abundance
is raised to
\begin{equation}
{\rm [D]/[H]}= (7.6 \pm 2.9)\times 10^{-6}~,
\end{equation}
where no error was added for uncertainties in the chemical
depletion factor.
The main uncertainty of our value appears to arise from
the indirect measure of the HD excitation, and of the abundance of atomic
hydrogen in the warm shocked region. Less dissociative shocks
would require larger corrections for non-LTE level populations, which
would raise the implied HD abundance: neglecting all H--HD collisions,
we earlier derived $N({\rm HD})\approx (3.5\pm 1.4)\times 10^{16}\rm cm^{-2}$,
which yields [D]/[H]$=7.9\times 10^{-6}$. Now chemical depletion should not
occur in shocks with a low abundance of neutral hydrogen, so that this
values reflects the actual deuterium abundance. Interestingly, we find that
the effects of non-LTE and of
chemical depletion nearly cancel, so that the derived deuterium
abundance turns out to be not very sensitive on how dissociative the
shock is.
\section{Summary}
ISO for the first time enabled the detection in the interstellar
medium of an infrared transition in the electronic ground state of
deuterated hydrogen, HD. We here report the
discovery of the $v$=0--0 R(5) line at 19.4290 $\mu$m
with the ISO Short Wavelength Spectrometer in the warm,
shocked molecular gas of the Orion OMC-1 outflow, at the bright emission
``Peak 1.'' Upper flux limits for sixteen other HD lines were
measured, all of which appear consistent with expectations when
considering the observed 0--0 R(5) line flux.
A large number of H$_2$\ lines were detected (Rosenthal et al. in prep.) and
utilized to analyze the HD observations. The near- and mid-infrared
extinction toward the emitting region was derived by minimizing the
dispersion in the observed H$_2$\ level column densities with respect to
an LTE excitation model. Thereby we derive a near-infrared
(K band) extinction of $(1.0\pm0.1)$ magnitudes, a 9.7 $\mu$m
extinction of $(1.35\pm 0.15)$ mag, and from an estimated range of
$A_{19}/A_{9.7}= 0.35 - 0.52$ we correct the HD 0-0 R(5) flux for
extinction by $(0.61\pm0.15)$ mag, i.e. a factor $1.75\pm0.25$.
The dereddened H$_2$\ level populations served to estimate the
excitation conditions in the gas. While H$_2$\ was assumed to have
thermalized level populations, those for HD were
computed in detail by making use of the H$_2$\ excitation.
Due to non-LTE effects at $J>3$,
the total warm HD column density was found to
be sensitive to the gas density and temperature at the
densities estimated to
prevail in the shocked gas,
$n({\rm H_2})\approx(1-3.5)\times 10^5~{\rm cm^{-3}}$.
Our estimate for the observed warm HD column density is
$N({\rm HD})=(2.0\pm0.75 )\times 10^{16}~{\rm cm^{-2}}$, and for the warm
molecular hydrogen, $N({\rm H_2})=(2.21\pm0.24)\times 10^{21}~{\rm cm^{-2}}$.
Their relative abundance is therefore $\rm [HD]/[H_2]=(9.0\pm 3.5)\times
10^{-6}$.
We note that in high-velocity C-shocks, HD may be
depleted relative to H$_2$ because of an asymmetry (due to a small
binding energy difference) in the
deuterium-hydrogen exchange reaction ${\rm HD + H} \getsto {\rm D + H_2}$.
Estimating that this lowers the warm HD column
by about 40\%, we derive a deuterium abundance
in the warm shocked gas, [D]/[H]$=(7.6\pm2.9)\times 10^{-6}$.
If the emitting shocks were on average less dissociative than assumed, the chemical
depletion would be less pronounced.
But at the same time, the lower H-HD collision rate
would enhance the HD $J=6$ level population's deviation from LTE, to the
effect that our implied
total column of HD would increase. We estimate that the two effects
approximately cancel, and that [D]/[H] remains at $\approx 8\times 10^{-6}$,
independent of how dissociative the shock actually is.
The major uncertainties in our estimate for the deuterium abundance
arise from our indirect measure of
the HD excitation, and since we must therefore rely on non-LTE excitation
models, on the
uncertainty of the preshock density and the
abundance of neutral hydrogen in the
shock. Future ground-based near-IR observations of
ro-vibrational transition lines of HD could
constrain the effects of non-LTE and thereby narrow the error margins
of the deuterium abundance. Detailed shock models of the recently resolved,
multiple
bow-shaped emission in the Orion outflow (Schultz et al. 1998)
might also yield better
estimates for the pre-shock densities and shock velocities, and thereby
of the H$_2$ dissociation and of the
chemical depletion fraction of HD in such shocks.
The deuterium abundance we find, $(7.6\pm 2.9)\times 10^{-6}$,
is lower than that derived through most DI absorption measurements in
the local ISM, but it is consistent with
that found recently
by Jenkins et al. (1999) toward $\lambda$ Orionis, $(7.4^{+1.9}_{-1.3})
\times 10^{-6}$,
and by Wright et al. (1999) toward the Orion Bar, $(1.0\pm 0.3) \times 10^{-5}$.
This could indicate that toward Orion, the deuterium abundance is indeed
somewhat lower than on average.
We are thankful to J.~Lacy, B.~Draine, and M.~Walmsley
for valuable comments, to A.~Schultz for providing the NICMOS
image, and to the SWS Data Center at MPE, especially to H.~Feuchtgruber
and E.~Wieprecht,
for their support. SWS and ISODC at MPE are supported by DARA under grants
50QI86108 and 50QI94023. FB acknowledges support by the
Deutsche Forschungsgemeinschaft (DFG) through its ``Physics of
Star Formation'' program.
CMW acknowledges support by NFRA/NWO grant 781-76-015 and of an ARC Research
Fellowship.
|
\section*{Introduction and Theoretical Background}
The primary interest of E865 is the forbidden decay $K^+\rightarrow\pi^+\mu^+ e^-$.
Results from two independent analyses
\cite{db,sp} have yielded a limit of
2 $\times~10^{-10}$ for a preliminary 1995 data set,
data from 1996 and 1998 has a
statistical reach
of order $10^{-11}$. However, an important by-product of this experiment
has been that E865 has significantly increased world sample sizes
for several other $K^+$ decay modes
with 3 charged particles in the final state.
(Final states containing a $\pi^{0}$ are included, since the $\pi^{0}$ is
detected through
$\pi^{0} \rightarrow \gamma e^{+} e ^{-}$). A detailed understanding
of the K decays, among them $\pi^+e^+e^-$ and $\pi^+\mu^+\mu^-$, tests models of the
weak interaction and low energy QCD.
The focus of this paper is a new measurement of the branching fraction
for $K^+\rightarrow\pi^+\mu^+\mu^-$.
The previous best measurement, from E787 \cite{kmm787},
based on a small number
of fully reconstructed events and about 200 partially reconstructed
events, is
$(5.0 \pm 1.0)\times 10^{-8}$; they did not make a form factor
measurement, because of limited
acceptance and incomplete event information.
With specific assumptions about the form of the interaction
(typically, vector, with a linear q$^2$ dependence, as in Ke3),
the expected $\pi^+\mu^+\mu^- / \pi^+e^+e^-$ ratio can be
calculated from the E865 results, and from theory, and
compared with the experimental observation.
For the $\pi^+e^+e^-$ we use the most precise results,
a branching fraction of $(2.82\pm 0.04\pm 0.15)10^{-7}$
and $\lambda$ of $0.182 \pm 0.01 \pm 0.007$,
based on our 10000 events \cite{se}.
Short distance contributions \cite{short} to $K \rightarrow
\pi \it{l} \it{l}$ are only of order $10^{-9}$.
Long distance contributions \cite{bs,pl,eils,vain}
come close to the observed $\pi^+e^+e^-$ rate and give a $\pi^+\mu^+\mu^-$ to $\pi^+e^+e^-$
ratio of about 0.22-0.24.
Vector and a1 meson dominance \cite{pl}
is an example of a simple model with a
parameter free prediction for the branching fractions but small
form factor dependence on $q^2$, similar to Ke3.
Chiral Perturbation theory parameters
allow a range of predictions for the form factor. In $O(p^4)$
\cite{epr,gabbiani},
the form factor and its $q^2$ dependence
are tightly correlated with the branching ratio.
Using an $O(p^6)$ calculation of an explicit "pion loop term" \cite{dam}
with a polynomial of the expected Chiral Perturbation
behavior at $O(p^6)$
gives additional parameters and flexibility.
The E787
$\pi^+\mu^+\mu^- / \pi^+e^+e^-$ ratio, $ 0.18 \pm 0.04$, is
about 1.5 $\sigma$ below predictions.
\section*{Experimental Apparatus}
E865 at BNL \cite{hmavan} is a magnetic spectrometer
illuminated by an intense
unseparated 6 GeV/c beam, of
$\approx 10^{8} K^{+}$ and
$\approx 2 \times 10^{9} \pi^{+}$ per 1.6 sec AGS pulse.
Momentum measured in the spectrometer is compared with
energy deposit in the 600 module 15 r.l. deep "Shashlik" calorimeter.
Electron
and positron identification is done by two threshold
Cerenkov counters with $H_{2}$ on the left (primarily
negative particles) and $CH_{4}$ on the right (primarily
positive particles). A 24 plane proportional tube - iron plate
range stack identifies muons.
The $\pi^+e^+e^-$ data were taken parasitically in 1995 and 1996, and
$\pi^+\mu^+\mu^-$ data in a 1997 reduced intensity run.
\section*{Event Selection and Analysis}
The $\pi^+\mu^+\mu^-$ events are normalized
to the $\pi^+\pi^+\pi^-$ final state, with similar kinematics.
The trigger for all modes \cite{hmavan} requires
three particles in a kinematically plausible configuration
in hodoscope counters and the calorimeter.
The analysis required: a good reconstructed vertex; reasonable vector
momentum; and two electrons or two muons,
one negatively charged
on the left and one positively charged on the right.
For the $\pi^+e^+e^-$ events, Cerenkov counter light is required both in the
trigger and in the analysis; and for the $\pi^+\mu^+\mu^-$ events, muon chamber
signals were.
Cuts in track chisquare help eliminate
the primary background for the $\pi^+\mu^+\mu^-$ events,
secondary decays ($\pi \rightarrow \mu \nu$)
from $K^{+} \rightarrow \pi^+ \pi^+ \pi^-$ .
The information described
qualitatively above was combined quantitatively into an
"event likelihood".
The stability of the branching fraction as a function of
event likelihood
is shown in the plot on the left in Figure \ref{fig:stabil},
where the quantity R (proportional to the
branching fraction) is plotted against the event likelihood cut.
R is defined as: $r_{\mu}/r_{3\pi}$ where
$r_{\mu~ (or ~ 3 \pi)} = Ndata_{\mu (or~ 3 \pi)}/Nmc_{mu (or~ 3\pi)}$
and Ndata and Nmc are respectively the accepted data (simulated) events.
While R is relatively stable, the number of signal events (not shown
because of space constraints) drops from 700 to 200
as the event likelihood moves from -19 to -10. This drop
in signal is accompanied by a drop in background to signal
from $\approx 40 \%$ to $\approx 2\%$ , reflecting the
large admixture of background at the large negative
value of the event likehood, and loss of signal as the
event likelihood approaches -10. For our final result,
we use an event likelihood cut of -13,
which gives $\approx$ 400 signal events.
The effective mass of the $\pi^+\mu^+\mu^-$ final state is shown in
the right hand side of Figure \ref{fig:stabil}.
Background from $K^+ \rightarrow \pi^+\pi^+\pi^-$, with two
pions decaying to muons, is shown by the dark-shaded
curve at low effective masses.
\begin{figure}[htb]
\centerline{
\psfig{figure=ratio2.eps,height=2.3in,width=2.3in}
\psfig{figure=pmm_ms1.eps,height=2.3in,width=2.3in}
}
\caption[]{
Left: R as a function of the joint likelihood cut,
and track $\chi^2$ cut.
Right:$\pi^+\mu^+\mu^-$ invariant mass distribution for
joint likelihood cut of -13.
}
\label{fig:stabil}
\end{figure}
The systematic error is estimated as $\approx 7\%$,
dominated by $\approx 3\%$ from selection criteria
and background subtraction, and $\approx 4\%$ from
normalization uncertainties.
\section*{Results and Discussion}
The $\mu\mu$ effective mass distribution ($q^2$) and
the
$\cos\theta_{\pi\mu^+}$ distribution (where $\theta_{\pi\mu^+}$ is the
angle between the $\pi^+$ and $\mu^+$ in the $\mu\mu$ center of mass frame)
are not shown but are consistent with a vector interaction
in the decay, with a form factor $\approx 0.2$, as seen in the $\pi^+e^+e^-$ events
\cite{hmavan,se}.
Fit contours for the branching fraction and $\lambda$
(the form factor $q^2$ dependence) are shown in Figure \ref{fig:lambda},
assuming a linear $q^2$ dependence as in Ke3.
\begin{figure}[htb]
\centerline{
\psfig{figure=brlam.eps,height=2.3in,width=2.3in}
\psfig{figure=mutoee.eps,height=2.1in,width=2.1in}
}
\caption{
Left: $\chi^{2}$ contour of Br vs $\lambda$.
Each contour is one unit of $\chi^{2}$.
The solid point is the $\chi^{2}$ minimum,
and the open point with error bars
is the prediction from the $\pi^+e^+e^-$ measurement.
Right:Relative branching ratios of the $\pi^+\mu^+\mu^-$ and $\pi^+e^+e^-$ decay
modes vs $\lambda$, assuming a linear dependence of the
form factor on $q^2$. The previous result is shown as a band,
while the point at low $\lambda$ is from Lichard's meson dominance model,
as representative
of simple long distance models.
The point at high $\lambda$ is the result presented in this paper, using
the $\lambda$ and the $\pi^+e^+e^-$ branching fraction from
the thesis of Scott Eilerts, the most precise determination
of these parameters.
}
\label{fig:lambda}
\end{figure}
The $\pi^+\mu^+\mu^-$
branching fraction and $\lambda$ are larger than expected
in a simple meson dominance model but agree with
expectations from $\pi^+e^+e^-$.
Our present understanding is that our $\pi l l $ data (branching ratios, and
q$^2$ dependence,
taken together, Ref. \cite{hmavan,se} and from this paper)
are inconsistent
with Chiral Perturbation theory at $O(p^4)$
\cite {epr,gabbiani},
but, due to the additional parameters available,
are consistent with
an O($p^6$) calculation \cite{dam}.
Detailed
systematic studies and comparison with Chiral Perturbation
theory are in progress.
\section*{Acknowledgments}
We thank our home country research institutes
for financial support. The experiment
would not have been possible without the valiant and sustained
efforts of the AGS machine operators and technical support crew.
The Pitt group owes a large debt of gratitude to high school
teacher Ivan Ober and
many students who helped in design, construction and
commissioning of the Cerenkov counters, especially Elizabeth
Battiste, Rebecca Chapman, Amy Freedman, Tuan Lu, Cindy Miller,
Melinda Nickelson, Tim Stever, Paula Pomianowski, and Craig Valine.
\section*{References}
|
\section{Introduction}
Resonances are obtained in the scattering of two (or more) elementary
particles, and quasistationary states decay into a two (or many) particle
system with masses $m_{i}$ and spins $s_{i}$, $i=1,\,2\cdots$.
Relativistic resonances and
decaying states are therefore described in the direct product space of two
irreducible representation spaces of the Poincar\'e group
${\cal H}={\cal H}_{1}(m_{1},s_{1})\otimes{\cal H}_{2}(m_{2},s_{2})$.
Non-relativistic resonances and decaying states
have been described by Gamow vectors \cite{our}.
Gamow vectors are characterized by a
value of angular momentum $j$ in the center-of-mass frame and by a complex
energy $z_{R}=\left(E_{R}-i\frac{\Gamma}{2}\right)$, representing
resonance energy $E_{R}$ and lifetime $\frac{\hbar}{\Gamma}$. They
are generalized eigenvectors in a Rigged Hilbert Space
$\Phi\subset\H\subset\Phi^{\times}$ of the self-adjoint Hamiltonian
$H$ with complex eigenvalue $z_{R}$ \cite{our}.
Relativistic resonances and unstable particles
are characterized by their spin (total angular momentum in the
center-of-mass frame of the decay products) and the value ${\mathsf{s}}={\mathsf{s}}_{R}
\equiv \left(M_{R}-i\frac{\Gamma}{2}\right)^{2}$ of the invariant mass
squared ${\mathsf{s}}=(p_{1}+p_{2})^{2}=(E^{2}-\bbox{p}^{2})$
where $M_{R}$ is the resonance mass and $\frac{\hbar}{\Gamma_{R}}$
is its lifetime. We want to find relativistic Gamow vectors which
are generalized eigenvectors of the total mass operator
$M^{2}=P_{\mu}P^{\mu}=(P_{1\mu}+P_{2\mu})(P_{1}^{\mu}+P_{2}^{\mu})$
with complex eigenvalue ${\mathsf{s}}_{R}$
and with spin $j$. These must be obtained from the direct
product space ${\cal H}_{1}(m_{1},s_{1})\otimes{\cal H}_{2}(m_{2},s_{2})$.
Eigenspaces of $M^{2}$ with real values of invariant mass ${\mathsf{s}}$
and total angular momentum $j$ are obtained by the relativistic partial
wave analysis \cite{aW60,hJ62,aM62}
using the Wigner basis, i.e., using momentum eigenvectors
$|\bbox{p}_{i},s_{3i}(m_{i},s_{i})\>$ in the spaces
${\cal H}_{i}$ and eigenvectors $|\bbox{p},j_{3}({\mathsf{s}},j)\>$
of $P_{\mu}=P_{1\mu}+P_{2\mu}$ in the direct
product space $\cal H$.
In distinction to the non-relativistic
case, in the relativistic case Lorentz transformations intermingle energy
and momenta. If one wants to make an analytic continuation
of ${\mathsf{s}}$ from the values $(m_{1}+m_{2})^{2}\leq {\mathsf{s}}<\infty$
to the complex values ${\mathsf{s}}_{R}$ (of the pole position in the second
sheet of the relativistic $S$-matrix $S_{j}({\mathsf{s}})$) this will also
lead to complex momenta. To restrict the unwieldy
set of complex momentum representations \cite{cm} we want to construct
complex mass representations of the Poincar\'e group $\cal P$ whose
momenta are ``minimally complex''
in the sense that though $p_{\mu}$ and $m$ are complex, the
$4$-velocities $\hat{p}_{\mu}\equiv \frac{p_{\mu}}{m}$ remain real.
This can be carried out because, as explained
in section $2$, the $4$-velocity eigenvectors
$|\bbox{\hat{p}},j_{3}({\mathsf{s}},j)\>$
provide as valid basis vectors for the representation space
of $\cal P$ as the usual momentum eigenvectors.
Moreover, they are more useful for physical reasoning than the
momenta eigenvectors, because the $4$-velocities
seem to fulfill to rather good approximation ``velocity super-selection
rules'' which the momenta do not \cite{velocityvectors}.
Therefore we will use the velocity basis $|\bbox{\hat{p}}_{i},s_{3i}(m_{i},s_{i})\>$
for the relativistic partial wave analysis and obtain the Clebsch-Gordan
coefficients of the Poincar\'e group for the velocity basis. This is done
in section $3$ for $s_{1}=s_{2}=0$,
which applies to the case of $\pi^{+}\pi^{-}$
in the final state.
This gives the velocity eigenvectors $|\bbox{\hat{p}},j_{3} ({\mathsf{s}},j)\>$
of the direct product space ${\cal H}=\sum_{j=0}^{\infty}
\int_{(m_{1}+m_{2})^{2}}^{\infty}d\mu({\mathsf{s}}){\cal H}({\mathsf{s}},j)$
from which we obtain the
four-velocity scattering states $|\bbox{\hat{p}},j_{3} ({\mathsf{s}},j)^{\pm}\>$
using the Lippmann-Schwinger equation as e.g., done in
\cite{sW95}. The relativistic Gamow vectors
$|\bbox{\hat{p}},j_{3}({\mathsf{s}}_{R},j)^{\pm}\>$
will be obtained in a subsequent paper from the scattering states
by analytic continuation. In the Appendix, we derive the Clebsch-Gordan
coefficients for the velocity basis of $\cal P$ for the general case.
\begin{comment}
Relativistic quantum mechanical systems are described by
unitary representations of the Poincar{\'e} group $\cal P$.
Following Wigner \cite{eW39}, (see also \cite{sW95} Ch.2),
an elementary relativistic system,
an elementary particle with mass $m$ and spin $j$, is that
quantum system whose space of physical states is the
representation space of an unitary irreducible representation
(UIR) of the Poincar{\'e} group $\cal P$. More complicated
relativistic systems are described by direct sums of UIR
(for ``towers'' of elementary particles) or by direct
products of UIR (for combination of two or more elementary particles).
A direct product of UIR may be decomposed into a continuous
direct sum (integral) of irreducible representations.
UIR are characterized by three
invariants $(m^{2},\, j,\, {\rm sign}\,p_{0})$; where $j$ represents the
spin and where the real number $m$ (with $m^{2}>0$) represents
the mass of the elementary particle.
We restrict ourselves here to ${\rm sign}\,p_{0}=+1$, the positive
energy representations
of the restricted Poincar\'e
group, and leave aside parity ${\cal U}_{P}$ and time reversal
${\cal A}_{T}$ for the time being.
The UIR of the Poincar\'e group $\cal P$ describe stable elementary
particles (stationary systems). There is only a very small number of
truly stable particles in nature
and most relativistic (and also non-relativistic) quantum systems
are decaying states (weakly or electromagnetically) or hadron resonances with
an exponential decay law and a finite lifetime
$\tau_{R}=\frac{\hslash}{\Gamma}$ (in their rest frame) and a Breit-Wigner
energy (at rest) or mass distribution. The UIR of $\cal P$ therefore
describe only a few of the relativistic quantum systems in nature; for
the vast majority of elementary particles listed in the Particle
Physics Booklet \cite{particledata}, the UIR provide
only a more or less approximate description, because most
of these particles are decaying states or resonances with a
mass and width. This suggests to use complex mass representations
of the Poincar\'e group. Complex mass will also mean complex
momenta $p^{\mu}$ which could lead to large classes of
very complicated representations of the Poincar\'e group. We will
therefore find those complex momentum representations
which are ``minimally complex''
in the sense that though $p_{\mu}$ and $m$ are complex, the
$4$-velocities $\hat{p}_{\mu}\equiv \frac{p_{\mu}}{m}$ remain real.
This can be carried out because, as explained
in section $2$, the $4$-velocity eigenvectors
provide as valid basis vectors for the representation space
of $\cal P$ as the usual momentum eigenvectors.
Moreover, they are more useful for physical reasoning than the
momenta eigenvectors, because the $4$-velocities
seem to fulfill to rather good approximation ``velocity super-selection
rules'' which the momenta do not \cite{bR41}.
In section $3$ they will be used as a basis for the reduction of the
direct product of two UIR spaces into continuous direct sums.
As will be shown in a subsequent paper, this will put the $S$-matrix of
a two-particle elastic process in a form readily suitable for
analytic continuation leading to the relativistic Gamow
vectors.
\end{comment
\section{Velocity Basis of the Poincar\'e Group}
We denote the ten generators of the unitary representation
${\cal U} (a, \Lambda)$ of $(a, \Lambda)\in \cal{P}$, by
\begin{equation}
\label{generators}
P^{\mu},\, J^{\mu\nu}{\hspace{1cm}} \mu,\nu\,=\, 0,1,2,3\, .
\end{equation}
The standard choice of the invariant operators and of a complete
set of commuting observables (c.s.c.o.) is
\begin{eqnarray}
\nonumber
M^{2}=P_{\mu}P^{\mu}\,&,&\quad W=-w_{\mu}\,w^{\mu},\\
\label{csco}
P_{i}\,(i=1,2,3)\,&,&\quad S_{3}=M^{-1}
{\cal U}(L(p))\,w_{3}\,{\cal U}^{-1}(L(p))\, ,
\end{eqnarray}
here
\begin{equation}
\label{w}
w_{\mu}=\frac{1}{2}\,\epsilon_{\mu \nu \rho \sigma}\, P^{\nu}
J^{\rho \sigma}\, ,
\end{equation}
$M^{-1}$ is the inverse square root of the positive definite operator
$P^{\mu}P_{\mu}$,~and ${\cal U}(L(p))$ is the representation of the boost that
depends upon the parameters $p_{\mu}\,(\mu=0,1,2,3)$, which are
the eigenvalues of the operators
$P_{\mu}$. Only three of these parameters are independent
in an irreducible representation, because of
the relation $m^{2}=p_{\mu}p^{\mu}$. The standard boost
(``rotation free'') matrix $L^{\mu}_{.\, \nu}(p)$ is given by
\begin{equation}
\label{standardboost}
L^{\mu}_{.\,\nu}(p)=
\bordermatrix{ &\nu=0 &\nu=n \cr
\mu=0&\frac{p^{0}}{m} &-\frac{p_{n}}{m} \cr
\mu=m&\frac{p^{m}}{m} &\delta^{m}_{n}-
\displaystyle{\frac{\frac{p^{m}}{m}\,
\frac{p_{n}}{m}}
{1+\frac{p^{0}}{m}}}\cr }\, .
\end{equation}
Note that $p_{\mu}=\eta_{\mu\,\nu}p^{\nu}$ and we use the metric
$\eta_{\mu\,\nu}=\scriptstyle{\left(
\begin{array}{cccc}
1 & & & 0 \\
&-1 & & \\
& & -1 & \\
0 & & & -1
\end{array}\right)}\, $
\footnote{Some of the references we use here have different convention,
e.g., $\eta_{\mu\, \nu}\rightarrow -\eta_{\mu\, \nu}$ \cite{sW95},
and $L^{-1}\rightarrow L(p)$ \cite{hJ62}.}.
It has the property that
\begin{equation}
\label{standardmomentum}
L^{-1}(p)^{\mu}_{.\,\nu}p^{\nu}=
\left(\begin{array}{c}
m\\
0\\
0\\
0
\end{array}\right)\, .
\end{equation}
One feature shown in (\ref{standardboost})
which we want to make use of, is that the
boost $L^{\mu}_{.\,\nu}(p)$ does not depend upon $p$ but only upon the
4-velocity $\frac{p}{m}\equiv \hat{p}$. The complete basis system in the irreducible
representation space ${\cal H}(m^{2},j)$ which consists of eigenvectors
of the c.s.c.o.\ (\ref{csco}) is the Wigner basis usually denoted as
\begin{equation}
\label{wignerbasis}
|\bbox{p},j_{3}(m,j)\rangle \, .
\end{equation}
It has the transformation property under the translation $(a,{\rm I})$ and the
Lorentz transformation $(0,\Lambda)$ :
\begin{mathletters}
\label{poincaretransformations}
\begin{equation}
\label{translation}
{\cal U}(a,{\rm I})|{\bbox p},j_{3}\rangle=
e^{ip^{\mu}a_{\mu}}|{\bbox p},j_{3}
\rangle
\end{equation}
\begin{equation}
\label{lorentztransformation}
{\cal U}(0,\Lambda)|{\bbox p},\xi\rangle=\sum_{\xi^{'}}|
\bbox{\Lambda p},
\xi^{'}\rangle D_{\xi^{'}\xi}({\cal R}(\Lambda,p))\, ,
\end{equation}
where $\cal{R}$ is the Wigner rotation
\begin{equation}
\label{wignerrotation}
{\cal R}(\Lambda,p)=L^{-1}(\Lambda p)\Lambda L(p)\, .
\end{equation}
The Wigner rotation depends upon the
$10$ parameters of $\Lambda$ and upon the parameters
$\hat{p}^{\mu}=\frac{p^{\mu}}{m}\, .$
In an UIR there are $3$ independent $\hat{p}^{\mu}$ and :
\begin{equation}
\label{boost}
|p,j_{3}\rangle={\cal U}(L(p))|{\bbox p}={\bbox 0},
j_{3}\rangle \, ,
\end{equation}
\end{mathletters}
where we have omitted the fixed values $m\, j$ as we shall often do
in an UIR.
Every vector (of a dense subspace of physical states) of ${\cal H} (m,j)$
can be written according to Dirac's basis vector decomposition as
\begin{mathletters}
\label{diracbasisvectors}
\begin{equation}
\label{dirac}
\phi=\int \, d\mu\left(\bbox{p}\right)\sum_{\xi}|\bbox{p},\xi
\rangle
\langle {\bbox p},\xi\,|\,\phi\rangle \, ,
\end{equation}
where one has many arbitrary choices for the measure.
It is usually chosen to be given by
\begin{equation}
\label{measure}
d\mu\left(\bbox{p}\right)=\rho\left(\bbox{p}\right)d^{3}
\bbox{p}\, ,
\end{equation}
where one can choose any (measurable) function $\rho$, in particular
a smooth function. The choice of $\rho$ is connected to the
``normalization'' of the Dirac kets through :
\begin{equation}
\label{normalization}
\langle \xi^{'},\bbox{p'}\,|\,\bbox{p},\xi \rangle
=\frac{1}{\rho\left(\bbox{p}\right)}\, \delta^{3}
(\bbox{p}-{\bbox p'})\,
\delta_{\xi \xi'} \, .
\end{equation}
One convention\footnote{This is the convention of \cite{eW39,aW60,hJ62,aM62},
but not
of \cite{sW95}} for $\rho$
is the Lorentz invariant measure :
\begin{equation}
\label{invariantmeasure}
\rho\left(\bbox{p}\right)=\frac{1}{2E(\bbox{p})}
\, ,\quad\text{where }
E\left(\bbox{p}\right)=\sqrt{m^{2}+{\bbox{p}}^{\,2}}\, .
\end{equation}
\end{mathletters}
The mathematically precise form of the Dirac decomposition is
the Nuclear Spectral Theorem for the complete system of
commuting (essentially self-adjoint) operators. It is the same as
(\ref{diracbasisvectors}), however with well
defined mathematical quantities. The state vectors $\phi$
in (\ref{dirac}) must be elements of a
dense subspace $\Phi$ of the representation space $\H$ of an UIR :
\begin{equation}
\phi \in \Phi \subset \H(m,j)\,;
\end{equation}
and the basis vectors $|\bbox{p},\xi\rangle \in \Phi^{\times}$
are elements of the space of antilinear functionals
on $\Phi$ which fulfill the condition :
\begin{mathletters}
\begin{equation}
\label{generalizedeigenvector}
\langle P_{i} \psi \,|\, \bbox{p}, \xi \rangle=p_{i}\,
\langle \psi \,|\,
\bbox{p}, \xi
\rangle \quad\text{for every }\psi \in \Psi\, .
\end{equation}
This condition means the $|\bbox{p},\xi\>$ are generalized eigenvectors
of $P_{i}$, which is
also written as
\begin{equation}
P_{i}^{\times}|\bbox{p},\xi\>=p_{i}\,|\bbox{p},\xi\>\, ,
\end{equation}
\end{mathletters}
where $P_{i}^{\times}$ is an extension of $P_{i}^{\dagger}(=P_{i})$;
and the ``component of $\phi$ along the basis vector
$|\bbox{p},\xi\rangle$'', the
$\langle \bbox{p},\xi\,|\,\phi \rangle=\<\phi\,|\,\bbox{p},\xi\>^{*}
$, are antilinear
continuous functionals $F(\phi)=\langle \bbox{p},\xi|\phi\rangle^{*}$
on the space $\Phi$.
The space $\Phi$ is a dense nuclear subspace of $\cal{H}$ \cite{aB73}.
(E.g., $\Phi$
could be chosen to be the subspace of
differentiable vectors of $\cal{H}$ equipped with a nuclear topology
defined by the countable number of norms~:
$||\phi||_{p}=\sqrt{(\phi,(\Delta+1)^{p}\phi)}$, where
$\Delta=\sum_{\mu} P_{\mu}^{2}+\sum_{\mu \, \nu}\frac{1}{2}J^{2}_{\mu \, \nu}
$ is the Nelson operator \cite{eN59}. But it could also
be chosen as another dense nuclear subspace of $\cal{H}$.)
The three spaces form a Gel'fand triplet, or Rigged Hilbert Space
\begin{equation}
\label{RHS}
\Phi \, \subset \, {\cal H} \, \subset \, \Phi^{\times}
\end{equation}
and the bra-ket $< \, | \, >$ is an extension of the scalar product
$( \, , \, )$. The $\langle \bbox{p}, \xi\,|\,\phi \rangle
=\<\phi\,|\,\bbox{p}, \xi\>^{*}$ are the Wigner momentum
wavefunctions.
The Wigner kets (\ref{wignerbasis})
are not the only basis system of ${\cal H}(m,j)$
that one can use to expand every vector $\phi \in \Phi$. For every
different choice of c.s.c.o.\
in the enveloping algebra $\cal{E}({\cal P})$ (the algebra generated by
$P_{\mu}$, $J_{\mu \, \nu}$) one obtains a different system
of basis vectors; in this
way one can obtain e.g., Lorentz basis
(eigenvectors of the Casimir operators of
$SO(3,1)_{J_{\mu \nu}}$ \cite{hJ62,aB73}), or
the spinor basis (whose Fourier transforms are
the relativistic fields \cite{sW95}) etc.
We want to choose still another basis system,
which is similar to the Wigner basis except that it is
a basis of eigenvectors of the $4$-velocity operator
$\hat{P}_{\mu}\equiv P_{\mu}M^{-1}$ rather than the momentum operator
$P_{\mu}$.
With the 4-velocity operator, one defines the operators
\begin{equation}
\label{what}
\hat{w}_{\mu}=\frac{1}{2}\,\epsilon_{\mu \nu \rho \sigma}
\hat{P}^{\nu}J^{\rho \sigma}=w_{\mu}M^{-1} \, ,
\end{equation}
and the spin tensor
$$
\Sigma_{\mu \, \nu}=\epsilon_{\mu \nu \rho \sigma}\hat{P}^{\rho}
\hat{w}^{\sigma}\,.
$$
The c.s.c.o.\ is then given by
\begin{equation}
\label{cscohat}
\hat{P}_{m},\quad S_{3},\quad \hat{W}=-\hat{w}_{\mu} \hat{w}^{\mu}
=\frac{1}{2}\Sigma_{\mu \nu}\Sigma^{\mu \nu},\quad M^{2}\, ,
\end{equation}
and we denote its generalized eigenvectors by
\begin{equation}
\label{wignerbasishat}
|\bbox{\hat{p}},j_{3};{\mathsf{s}}=m^{2},j\>\, ,
\end{equation}
where $\hat{p}_{\mu}=\frac{p_{\mu}}{m}$ are the eigenvalues of $\hat{P}_{\mu}$.
The basis vector expansion for every $\phi \in \Phi$
with respect to the basis system (\ref{wignerbasishat}) is given by
\begin{mathletters}
\label{diracbasisvectorhat}
\begin{equation}
\label{dirachat}
\phi=\sum_{j_{3}}\int \, \frac{d^{3}\hat{p}}{2 \hat{p}^{0}}
\,|\bbox{\hat{p}}, j_{3}\rangle \langle j_{3}, \bbox{\hat{p}}
\,|\, \phi \rangle \, ,
\end{equation}
where we have chosen the invariant measure
\begin{eqnarray}
\label{measurehat}
d\mu(\bbox{\hat{p}}) &=& \frac{d^{3}\hat{p}}{2\hat{p}^{0}}
= {\frac{1}{m^{2}}} \, {\frac{d^{3}p}{2 E(\bbox{p})}}\\ \nonumber \hat{p}^{0} &=& \sqrt{1+\bbox{\hat{p}}^{2}} \, .
\end{eqnarray}
As a consequence of (\ref{measurehat}),
the $\delta$-function normalization of these velocity-basis vectors is
\begin{eqnarray}
\nonumber
\langle \xi , \bbox{\hat{p}}\,|\,\bbox{\hat{p}'}, \xi' \rangle
&=& 2 \hat{p}^{0} \delta^{3}(\bbox{\hat{p}}-\bbox{\hat{p}'})
\, \delta_{\xi \xi'}\\
\label{normalizationhat}
&=& 2 p^{0} m^{2} \delta^{3}
(\bbox{p}-\bbox{p'})\,
\delta_{\xi \xi'} \, .
\end{eqnarray}
\end{mathletters}
Mathematically, every c.s.c.o.\ is equally valid.
But, for a given physical problem one c.s.c.o.\ may be more useful than
another. For instance a c.s.c.o.\ that contains physically
distinguished observables
(e.g., observables whose eigenstates happen to appear predominantly in nature)
is more useful for calculations in physics than the c.s.c.o.\ whose
eigenvectors are very different from physical eigenstates.
Two different c.s.c.o.'s lead to different basis systems, whose vectors can be
expanded with respect to each other. But this expansion is usually very
complicated and intractable, for which reason the choice of the physically
right c.s.c.o.\ is very important for each particular physical problem. This
is the reason for which the Lorentz basis of the
Poincar\'e group is pretty useless for physics, because
the Casimir operators of $SO(3,1)$ are not important observables as compared
to the momentum. However, the two c.s.c.o.\ (\ref{csco}) and (\ref{cscohat})
are not even different in an irreducible representation of ${\cal P}$,
since its operators differ only by a factor of
the operator $M$, which is an invariant. The basis systems (\ref{wignerbasis})
and (\ref{wignerbasishat}) are therefore the same,
i.e., their values differ by a normalization-phase factor $N(p,j_{3})$
\begin{equation}
\label{basisrelation}
|\,\bbox{\hat{p}}, j_{3} \, (m,j)\rangle
=|\,{\bbox p},j_{3} \, (m,j) \rangle \, N(p, j_{3}).
\end{equation}
The Poincar\'e transformations (\ref{poincaretransformations})
act on the basis vectors (\ref{basisrelation}) in the following way
\begin{mathletters}
\label{poicaretranformationshat}
\begin{eqnarray}
&{\cal U}(a,{\rm I})|\bbox{\hat{p}},j_{3}\rangle=
e^{im\hat{p}^{\mu}a_{\mu}}|\bbox{\hat{p}}, j_{3}\rangle \label{translationhat}\\
&{\cal U}(L(\hat{p}))|\bbox{\hat{p}}=\bbox{0},j_{3}
\rangle=|\bbox{\hat{p}},j_{3}\rangle \, .
\label{boosthat}
\end{eqnarray}
\end{mathletters}
The distinction between the basis vectors $|\,\bbox{p}, \xi \rangle$
and $|\,\bbox{\hat{p}}, \xi \rangle$ becomes important if one does not have an
unitary irreducible representation of $\cal{P}$ but a representation with
many different values for $(m^{2},j)$, e.g., ${\cal H}=\sum_{m^{2},j}
\oplus {\cal H}(m,j)$. Then one has besides the observables
(\ref{generators}), additional observables
$X_{\alpha}$ (generators of an intrinsic symmetry group
or a spectrum generating group) and an additional system of
commuting observables :
\begin{equation}
\label{commutingobservables}
B=B_{1},B_{2}, \cdots , B_{N}
\end{equation}
whose eigenvalues, $b=(b_{1},b_{2}, \cdots ,b_{N})$,
characterize the elementary
particles described by ${\cal H}(m,j)={\cal H}^{b}(m,j)$
\footnote{The quantum numbers $b$ are called the particle species
numbers in \cite{sW95}.}.
In order that (\ref{csco}) and (\ref{commutingobservables}) combine
into a c.s.c.o., the operators $B$ have to commute with
$M^{2}, P_{\mu}, W \text{ and }S_{3}$.
If also the other observables $X_{\alpha}$,
which change the particle species number $b$, commute with
$M^{2}, P_{\mu}, W \text{ and }S_{3}$, then the combination of
(\ref{csco}) and (\ref{commutingobservables}) gives a useful
c.s.c.o. However, if the $X_{\alpha}$ do not commute with $M^{2}$
(i.e., the particle species number changing operators $X_{\alpha}$
transform also from one mass eigenstate to another mass eigenstate changing
also the mass $m_{b}$ into $m_{b'}$) then the $X_{\alpha}$ will also
not commute with $P_{\mu}$, $\left[ X_{\alpha}, P_{\mu}\right]
\neq 0$. In this case, it may still happen \cite{velocityvectors} that a
``velocity superselection rule'' holds :
\begin{equation}
\label{velocitysuperselectionrule}
\left[ X_{\alpha}, \hat{P}_{\mu} \right] =0\quad (\text{or at least }
\left[ X_{\alpha},\hat{P}_{\mu} \right]\approx 0)\, .
\end{equation}
Then combination of (\ref{commutingobservables}) with
(\ref{cscohat}), i.e., the
\begin{equation}
\label{cscohat2}
\hat{P}_{i},\, \hat{w}_{3},\, \hat{W},\, M^{2},\, B_{1},\cdots,\,B_{N}
\end{equation}
will form a useful c.s.c.o., but the combination of
(\ref{csco}) with (\ref{commutingobservables}) will not.
The generalized eigenvectors of (\ref{cscohat2}),
$|\bbox{\hat{p}},\xi,b,m,j
\rangle $, will then be a much more useful basis system for every
$\phi \in \Phi \subset {\cal H}=\sum \oplus {\cal H}^{b}(m,j)$ than the
corresponding momentum eigenvectors.
Using the eigenvectors of (\ref{cscohat2}), we have the Dirac basis vector
expansion :
\begin{equation}
\label{dirachat2}
\phi=\sum_{m,b}\sum_{j,\xi}\int \frac{d^{3}\hat{p}}{2\, \hat{p}^{0}}
|\,\bbox{\hat{p}},\xi,b,m,j\rangle
\langle j,m,b,\xi, \bbox{\hat{p}}\,|\,
\phi \rangle \text{ for every } \phi \in \Phi\, .
\end{equation}
The momentum eigenvectors $|\bbox{p},\xi,b \ldots \rangle$
may either not exist (if $\left[ B, P_{\mu} \right] \not= 0$),
or if they do exist, they are not useful because the $X_{\alpha}$
change the value of $p$, which then becomes a function of $b$,
$p=p_{b}$.
As a consequence, quantities like form factors depend upon
$b$ through $p$. In contrast, using the velocity eigenvectors
$|\bbox{\hat{p}},\xi,b,\cdots\rangle$ under the assumption
(\ref{velocitysuperselectionrule}) will lead to form factors with
universal (independent of $b$) dependence upon the four-velocity.
This was the original motivation for the introduction of
the velocity-basis vectors $|\bbox{\hat{p}},\xi,b,\cdots\rangle$
\cite{velocityvectors}.
The subject of the present work is the description of relativistic
decaying states by representations of the Poincar\'e group,
combining Wigner's idea \cite{eW39} of the description of
stable relativistic particles by an UIR of $\cal P$, with Gamow's
idea of describing decaying particles by eigenvectors with
complex energy. Therefore, we need in the rest frame basis vectors
with complex energy, i.e., the $m$ (and the ${\mathsf{s}}=m^{2}$) in
(\ref{wignerbasis}) or in (\ref{wignerbasishat})
has to be continued to complex values e.g., to ${\mathsf{s}}=(M_{R}-i\Gamma/2)^{2}$.
This will result in a continuation of the momenta $p_{\mu}$ to complex
values as well and can lead to an enormous complication of the
Poincar\'e group representations (see e.g., \cite{cm}).
We want to do this analytic continuation in the invariant mass ${\mathsf{s}}$
such that the $p_{\mu}$ are continued to complex values in such
a way that the $\hat{p}_{\mu}=\frac{p_{\mu}}{\sqrt{{\mathsf{s}}}}$ remain real.
Then, we obtain a smaller class of complex
mass representations of $\cal P$ which are as similar in property
as possible to Wigner's UIR $(m,j)$. These are the minimally
complex-mass representations which we shall denote
by $({\mathsf{s}},j)$.
For this minimal analytic continuation to be possible, it must be
compatible with the boost (\ref{boost}) and (\ref{boosthat}).
The crucial observation is that the boosts $L(p)$ are in fact, according to
(\ref{standardboost}) only
functions of $\hat{p}_{\mu}=\frac{p_{\mu}}{\sqrt{{\mathsf{s}}}}\,$; $L(p)=L(\hat{p})$.
As a consequence, the operators representing the boost
${\cal U}(L(p))={\cal U}(L(\hat{p}))$ are functions of the real parameters
$\hat{p}$ and not of complex parameters $p$.
This means they are the same operator functions in all the
subspaces of the
direct sum $\sum_{m_{b},j}\oplus{\cal H}(m_{b},j)$ and of the
continuous direct sum
\begin{equation}
\label{directsum}
\sum_{j,n}\int_{m_{0}^{2}}^{m_{1}^{2}}\oplus
{\cal H}^{n}({\mathsf{s}},j)d\mu({\mathsf{s}})
\end{equation}
of the irreducible representations
\begin{equation}
\label{irrep}
{\cal H}({\mathsf{s}},j),\qquad
{\mathsf{s}}=p_{\mu}p^{\mu}=E-\bbox{p}^{2}\,.
\end{equation}
If we consider in (\ref{directsum}) only
(continuous) direct sums with the same value
for $j=j_{R}$ then ${\cal U}(\Lambda)$ for any Lorentz transformation
$\Lambda$ is, according to (\ref{wignerrotation}),
the same operator function of
the $6$ parameters
which are given by the three $\hat{p}^{m}$ or the three $v^{m}$ :
\begin{equation}
\label{uandv}
\left(
\begin{array}{c}
\hat{p}^{0} \\
\hat{p}^{m}
\end{array}
\right)=\left(
\begin{array}{c}
\left( 1-\frac{\bbox{v}^{2}}{c^{2}} \right)^{-\frac{1}{2}} \\
\left( 1-\frac{\bbox{v}^{2}}{c^{2}} \right)^{-\frac{1}{2}}v^{m}
\end{array}\right)
\end{equation}
and the three rotation angles (e.g., Euler angles in the rest frame).
The analytic continuation in ${\mathsf{s}}$ can therefore be accomplished without
affecting the Lorentz transformations. The Lorentz transformations in
the minimally-complex mass representation are represented unitarily
by the same operators ${\cal U}(\Lambda)$ as in Wigner's UIR $(m,j_{R})$.
At rest, on $|\bbox{0},j_{3}\, ({\mathsf{s}},j_{R})\>$, only the time translations
of $\cal P$ will be represented non-unitarily for complex values of ${\mathsf{s}}$.
And using (\ref{boosthat}) only the label ${\mathsf{s}}$ in the velocity basis
$|\bbox{\hat{p}},j_{3}\, ({\mathsf{s}},j_{R})\>$ is complex.
The basis vector decomposition (\ref{dirachat2}) using the velocity basis,
\begin{equation}
\label{dirachat3}
\phi= \sum_{j_{3}}\int d\mu({\mathsf{s}})\int d\mu (\bbox{\hat{p}})
|\,\bbox{\hat{p}},j_{3}({\mathsf{s}},j)\rangle
\langle ({\mathsf{s}},j)j_{3},\bbox{\hat{p}}\,|\,
\phi\rangle
\quad\text{for } \phi \in \Phi \subset {\cal H}({\mathsf{s}},j)\, ,
\end{equation}
is therefore more suitable than (\ref{diracbasisvectors})
that uses the momentum basis,
because $\bbox{\hat{p}}$ is independent of ${\mathsf{s}}$ while
$\bbox{p}=\sqrt{{\mathsf{s}}}\bbox{\hat{p}}$ is not. If we deform the contour
of integration for ${\mathsf{s}}$ from the real axis as in (\ref{directsum}) into the
complex ${\mathsf{s}}$-plane then the integral over $d\mu(\bbox{\hat{p}})$ in (\ref{dirachat3})
remains unaffected.
\section{Relativistic Kinematics for (two-particle) Resonance Scattering}
Continuous direct sums like (\ref{directsum}) appear in the case
of scattering experiments of two relativistic particles like e.g.,
the process
\begin{mathletters}
\label{resonancescattering}
\begin{equation}
\label{twoelectronsresonance}
e^{+}\, e^{-} \rightarrow \rho^{0} \rightarrow
\pi^{+}\, \pi^{-} \, ,
\end{equation}
or the more theoretical process
\begin{equation}
\label{twopionsresonance}
\pi^{+}\,\pi^{-} \rightarrow \rho^{0} \rightarrow
\pi^{+}\, \pi^{-} \, .
\end{equation}
\end{mathletters}
These processes predominantly happen in the $j^{P}=1^{-}$ partial
amplitude if the
$\rho$-meson mass region is selected for the invariant mass square
\begin{equation}
\label{invariantmass}
{\mathsf{s}} =(p_{1}+p_{2})^{2}=E_{\rho}^{2}+\bbox {p}_{\rho}^{2}\, ,\,\,
E_{\rho}=E_{1}+E_{2}\, ,\,\,
\bbox{p}_{\rho}=\bbox{p_{1}}+\bbox{p_{2}}\, ,
\end{equation}
where $p_{1}$ and $p_{2}$ are the momenta
of the two pions $\pi^{+}$,$\pi^{-}$
\footnote{
Though our discussions apply with obvious modifications
to the general case of
$$1+2+3+\cdots \rightarrow R_{i}\rightarrow 1^{'}+2^{'}+3^{'}+\cdots$$
these generalizations lead to enormously more complicated equations.
For the sake of simplicity, we shall therefore
consider a resonance scattering process like (\ref{resonancescattering}).}.
The relativistic one particle states are given by an irreducible
representation space
${\cal H}^{n_{i}}(m_{i},s_{i})$ of the Poincar\'e group $\cal{P}$.
The independent, \emph{interaction-free}
two-particle states (or $n$ particle states)---
like the $\pi^{+}\, \pi^{-}$ system in (\ref{twopionsresonance})---
are given by the direct product of the irreducible representation
spaces ${\cal H}(m_{1},s_{1})$ and ${\cal H}(m_{2},s_{2})$ :
${\cal H}^{n_{1}}(m_{1},s_{1})
\otimes{\cal H}^{n_{2}}(m_{2},s_{2})\equiv {\cal H}$.
Empirical evidence suggests that the resonances in processes like
(\ref{resonancescattering}) appear in one partial amplitude
with a given value of resonance spin $j_{R}$ (e.g., $j_{\rho}^{P}=1^{-}$).
Therefore, the first problem is the reduction of the direct product
${\cal H}(m_{1},s_{1})\otimes{\cal H}(m_{2},s_{2})$ into a direct sum
of ${\cal H}^{n}({\mathsf{s}},j)$; the second problem is how to
go from the free two-particle system to the interacting two-particle
system.
The first problem has been solved in general
\cite{aW60,hJ62,aM62}
\begin{equation}
\label{reduction}
{\cal H}\equiv {\cal H}^{n_{1}}(m_{1},s_{1})\otimes
{\cal H}^{n_{2}}(m_{2},s_{2})
=\int_{(m_{1}+m_{2})^{2}}^{\infty}d\mu({\mathsf{s}})\sum_{nsl}\sum_{j}
\oplus{\cal H}^{nsl}({\mathsf{s}},j)\,.
\end{equation}
The sums in (\ref{reduction}) extend over
$$
j=
\begin{array}{ccccl}
0 & 1 & \cdots & \text{ if } & s_{1}+s_{2}=\text{ integer }\\
1/2 & 3/2 & \ldots & \text{ if } & s_{1}+s_{2}=\text{ half integer }
\end{array}\, ,
$$
and the degeneracy indices $(l,s)$ for a given $j$ are summed over
\begin{eqnarray}
\nonumber
s&=&s_{1}+s_{2}\, ,\, s_{1}+s_{2}-1\, ,\, \ldots |s_{1}-s_{2}|\\
\nonumber l&=&j+s\, ,\, j+s-1\, ,\, j+s-2\, ,\, \ldots j-s \, .
\end{eqnarray}
Here $j$ represents the total angular momentum of the combined
$\pi^{+}\pi^{-}$ system; one of these values will be the resonance spin
$j_{R}$. The degeneracy indices $(s,l)$ for each fixed value of $j$
are the total spin angular momentum and the total orbital angular
momentum of the two $\pi$, respectively.
The quantum number $n$ is summed over all channel numbers that can be obtained
by combining the species numbers $n_{1}$ and $n_{2}$ of the two $\pi$.
Instead of the invariant mass square ${\mathsf{s}}=p_{\mu}\,p^{\mu}=E^{2}-
\bbox{p}^{2}$ that we have used in
(\ref{reduction}) one often uses $w=\sqrt{{\mathsf{s}}}$, the invariant mass
or the energy in the center of mass system of the two particles $n_{1},
n_{2}$ \cite{aW60,hJ62,aM62}. The choice of the measure
\begin{equation}
\eqnum{\ref{reduction}$a$}
d\mu({\mathsf{s}})=\rho({\mathsf{s}})d{\mathsf{s}},\quad (\text{or if one uses $w$, of }
d\mu(w)=\rho(w)dw)
\end{equation}
depends upon the normalization of the system
of generalized basis vectors of (\ref{reduction}). We shall use
\begin{equation}
\label{choice}
\eqnum{\ref{reduction}$b$}
\rho({\mathsf{s}})=1,\text{ and then } \rho(w)=2w
\end{equation}
if we label the basis by $w$ so that we do not change
the ``normalization'' of the kets.
The resonance space will be related (but will not be identical)
to a subspace of (\ref{reduction}) with a definite value of
angular momentum $j$ (e.g., $j=j_{3}^{P}=1^{-}$
in case of the $\rho$-resonance
of (\ref{resonancescattering})). This is based on empirical evidence;
resonances appear in one particular partial amplitude with a particular
value of resonance spin $j=j_{R}$
(though it may happen that there are more than one
resonance in the same partial amplitude, but at different
resonance energy ${\mathsf{s}}_{R_{1}},\,{\mathsf{s}}_{R_{2}},\,\cdots$).
We will therefore single
out a particular subspace
\begin{equation}
\label{subspace}
{\cal H}^{nls}=\int_{(m_{1}+m_{2})^{2}}^{\infty}
d{\mathsf{s}} \oplus {\cal H}^{nls}({\mathsf{s}},j)
\end{equation}
with definite degeneracy or/and channel quantum numbers
$\eta=ls,$ $n$.
The reduction (\ref{reduction}) is usually done using the Wigner
momentum kets (\ref{wignerbasis}) in which the Clebsch-Gordan
coefficients are given by \cite{aW60,hJ62,aM62} :
\begin{equation}
\label{cg}
\langle\, p_{1}s_{13}\,p_{2}s_{23}\,[m_{1}s_{1}, m_{2}s_{2}]\,|\,
pj_{3}\,[wj],\eta\,\rangle,
\text{ where $\eta$ now denotes }\eta=n,l,s\,.
\end{equation}
For the reasons mentioned above we want to work
with the 4-velocity eigenkets
$|\hat{p},j_{3}\,[w,j],\eta\,\rangle$
which are eigenvectors of the operators
\begin{equation}
\label{operators}
\hat{P}_{\mu}=({P}^{(1)}_{\mu}+{P}^{(2)}_{\mu})M^{-1},\,\,
M^{2}=(P^{(1)}_{\mu}+P^{(2)}_{\mu})
(P^{(1)\mu}+P^{(2)\mu})
\end{equation}
with eigenvalues
\begin{equation}
\label{eigenvalues}
\hat{p}^{\mu}=
\left(
\begin{array}{c}
\hat{E}=\frac{p^{0}}{w}=\sqrt{1+\bbox{\hat{p}}^{2}}=\hat{p}^{0}\\
\bbox{\hat{p}}=\frac{\bbox{p}}{w}
\end{array} \right)
\text{ and eigenvalues }
w^{2}={\mathsf{s}}\, .
\end{equation}
In here $\hat{P}^{(i)}_{\mu}$ are the 4-velocity operators in the one
particle spaces ${\cal H}^{n_{i}}(m_{i},s_{i})$ with eigenvalues
$\hat{p}^{i}_{\mu}=\frac{p^{i}_{\mu}}{m_{i}}$.
The Clebsch-Gordan coefficients are the transition coefficients
$\langle\hat{p}_{1}\hat{p}_{2}\,s_{13}s_{23}\,[m_{1}s_{1},m_{2}s_{2}]\,
|\,\hat{p}j_{3}\,[wj],\eta\,\rangle$
between the direct product basis
\begin{equation}
\label{directproductbasis}
|\hat{p}_{1}s_{13}\,m_{1}s_{1}\rangle\otimes
|\hat{p}_{2}s_{23}\,m_{2}s_{2}\rangle
\equiv
|\hat{p}_{1}\hat{p}_{2}\,s_{13}s_{23}\,[m_{1}s_{1},m_{2}s_{2}]\,\rangle
\end{equation}
and the angular momentum basis
$|\hat{p} j_{3}\,[wj],\eta\,\>$.
To obtain the Clebsch-Gordan coefficients, one follows the same
procedure as given in the classic papers \cite{aW60,hJ62,aM62}
for the Clebsch-Gordan coefficients
(\ref{cg}). This will be done in the Appendix, where the general
case will be discussed. Here we shall restrict
ourselves to the special case $s_{1}=0,s_{2}=0$ to avoid the
inessential complications due to the $SO(3)$ Clebsch-Gordan coefficients
for the angular momentum couplings $s_{1}\otimes s_{2}\rightarrow s$,
$s\otimes l \rightarrow j$ and the occurrence of the Wigner
rotations $R(L^{-1}(\hat{p}),\hat{p}_{i})$ of the inverse
boost $L^{-1}(\hat{p})$ which will enter in (\ref{cg}).
Also for the process (\ref{twopionsresonance}) this is sufficient,
since $s_{{\pi}^{+}}=s_{{\pi}^{-}}=0$. There is no degeneracy
of the angular momentum basis vectors in this case and
$|\hat{p} j_{3}\,[wj]\,\>$ is given in terms of (\ref{directproductbasis})
by
\begin{eqnarray}
\label{expansionhat}
&|\,\hat{p}j_{3}\,[wj]\,\rangle=\int\frac{d^{3}\hat{p}_{1}}{2\hat{E}_{1}}
\frac{d^{3}\hat{p}_{2}}{2\hat{E}_{2}}
\,|\,\hat{p}_{1}\hat{p}_{2}[m_{1}m_{2}]\,\rangle\,\langle\,
\hat{p}_{1}\hat{p}_{2}[m_{1}m_{2}]\,|\,\hat{p}j_{3}\,[wj]\,\rangle\\
\nonumber
&\text{for any }(m_{1}+m_{2})^{2}\leq w^{2}<\infty\qquad j=0,1,\cdots
\end{eqnarray}
The choice of the measure $\frac{d^{3}\hat{p}_{i}}{2\hat{E}_{i}(\bbox{\hat{p}}_{i})}
=\frac{d^{3}p_{i}}{m_{i}^{2}2E_{i}}$ is the same as
(\ref{dirachat}).
From the 4-translation invariance (conservation of
$4$-momentum) it follows that the Clebsch-Gordan is
of the form
\begin{equation}
\label{deltafour}
\langle\,\hat{p}_{1}\hat{p}_{2}\,|\,\hat{p}j_{3}\,[wj]\,\rangle
=\delta^{4}(p-r)\langle\!\langle\,\hat{p}_{1}\hat{p}_{2}\,|\,\hat{p}j_{3}
\,[wj]\,\rangle\!\rangle\, ,\quad \text{where } r\equiv p_{1}+p_{2}\, .
\end{equation}
The reduced matrix element in the center-of-mass is in analogy to the
non-relativistic case given by \cite{aB93}
\begin{equation}
\label{reducedelement}
\langle\!\langle\,\hat{p}_{1}^{cm}\hat{p}_{2}^{cm}\,|\,
\bbox{0}j_{3}\,[wj]\,\rangle\!\rangle=Y_{jj_{3}}(\bbox{e})
\tilde\mu_{j}(w,m_{1},m_{2})\, ,
\end{equation}
where $\tilde{\mu}_{j}(w,m_{1},m_{2})$
is a function of $w$ (or ${\mathsf{s}}$) which depends
upon our choice of ``normalization'' for the basis vectors
$|\hat{p} j_{3}\,[wj]\,\>$ in (\ref{expansionhat}).
The equations (\ref{deltafour}) and (\ref{reducedelement})
are combined into
\begin{eqnarray}
\label{cgseries}
&\langle\,\hat{p}_{1}\hat{p}_{2}\,|\,\hat{p}j_{3}\,[wj]\,\rangle
=2\hat{E}(\bbox{\hat{p}})\delta^{3}(\bbox{p}-\bbox{r})
\delta(w-\epsilon)Y_{jj_{3}}(\bbox{e})\mu_{j}(w,m_{1},
m_{2})\\
\nonumber &\text{ with }\epsilon^{2}=r^{2}=(p_{1}+p_{2})^{2} \, ,
\end{eqnarray}
where again $\mu_{j}(w,m_{1},m_{2})$ is a function
that fixes the $\delta$-function ``normalization'' of
$|\hat{p} j_{3}\,[wj]\,\>$.
The unit vector $\bbox{e}$ in (\ref{reducedelement})
is chosen to be in the c.m.\ frame
the direction of $\bbox{\hat{p}}_{1}^{cm}=-\frac{m_{2}}{m_{1}}\,\bbox{\hat{p}}_{2}
^{cm}$. In general it is obtained from the relative ``4-momentum''
$q_{\mu}$ of Michel and Wightman \cite{aW60}
by $e_{i}=L^{-1}(p)^{.\,\mu}_{i}q_{\mu}$.
The $\mu_{j}(w,m_{1},m_{2})$ and $\tilde\mu_{j}(w,m_{1},m_{2})$
are some weight functions
which are determined from the required ``normalization'' of the
4-velocity kets (\ref{expansionhat}). Since for a fixed value
of $[wj]$ these generalized eigenvectors are the basis of the
irreducible representation space ${\cal H}(w,j)$ of the Poincar\'e group,
we want them to be normalized like (\ref{measurehat}),
which in (\ref{expansionhat}) has been already
assured by the choice of the invariant measure
$\frac{d^{3}\hat{p}_{i}}{2\hat{E}_{i}}$.
Therefore, in analogy to (\ref{normalizationhat}),
we take for the normalization
of the basis vectors (\ref{expansionhat}) to be
\begin{eqnarray}
\label{normalizationchoice}
\langle\,\hat{p}'j_{3}'\,[w'j']\,|\,\hat{p}j_{3}\,[wj]\,\rangle\,
=2\hat{E}(\bbox{\hat{p}})\delta^{3}(\bbox{\hat{p}'}
-\bbox{\hat{p}})\delta_{j_{3}'j_{3}}\delta_{j'j}\delta({\mathsf{s}}-{\mathsf{s}}')\, ,\\
\nonumber
\text{ where }\hat{E}(\bbox{\hat{p}})=\sqrt{1+\bbox{\hat{p}}^{2}}
=\frac{1}{w}\sqrt{w^{2}+\bbox{p}^{2}}\equiv\frac{1}{w}E(\bbox{p},w)\, .
\end{eqnarray}
The $\delta$-function normalization $\delta({\mathsf{s}}'-{\mathsf{s}})=\frac{1}
{2w}\delta(w-w')$ in (\ref{normalizationchoice})
is a consequence of the choice (\ref{choice}) for the measure.
After we have chosen the normalization as in
(\ref{normalizationchoice}), one determines
the weight function $\mu_{j}(w,m_{1},m_{2})$
using (\ref{expansionhat}). The result is :
\begin{equation}
\label{weight}
\left| \mu_{j}(w,m_{1},m_{2})
\right|^{2}=\frac{2m_{1}^{2}m_{2}^{2}w^{2}}
{\sqrt{\lambda(1,(\frac{m_{1}}{w})^{2},(\frac{m_{2}}{w})^{2})}}\, ,
\end{equation}
where $\lambda$ is defined by \cite{aW60}:
$$
\lambda(a,b,c)=a^{2}+b^{2}+c^{2}-2(ab+bc+ac)\,.
$$
Except for the normalization factor $\mu$, which follows from
our chosen normalization (\ref{normalizationchoice}),
the values of the Clebsch-Gordan
coefficients (\ref{cgseries}) is quite obvious
\footnote{
A formula like (\ref{cgseries}) is also given and explained in section
$3.7$ of \cite{sW95} which for $s=0,\, s_{1}=s_{2}=0$
agrees with (\ref{cgseries}) except for the normalization factor
(\ref{weight}). For $s\ne 0,\, s_{i}\ne 0$, see Appendix.}.
It expresses momentum conservation and the only factor
that one may be puzzled about is that it should be consistent with
the 4-velocity normalization expressed by the
$\delta^{3}(\bbox{\hat{p}'}-\bbox{\hat{p}})$,
$\bbox{\hat{p}}=\frac{\bbox{p}}{w}$ in
(\ref{normalizationchoice}).
Therewith, we have obtained by (\ref{expansionhat}) with
(\ref{directproductbasis}) and (\ref{cgseries})
a system of basis vectors for the space (\ref{reduction})
(with $s_{1}=s_{2}=0$) which is the representation space
of scattering processes like (\ref{twopionsresonance}).
As expected, the basis vectors are outside the Hilbert space;
$|\,\hat{p}j_{3}\,[wj]\,\rangle\in\Phi^{\times}\supset{\cal H}\supset\Phi$.
They have definite values of angular momentum $j$ and invariant
mass $w\equiv\sqrt{{\mathsf{s}}}$
\footnote{
Written in terms of Hilbert spaces, $d\mu({\mathsf{s}})$ means Lebesgue integrations.
However, within the RHS mathematics, one can choose for
$\langle\, \phi\,|\,\hat{p}j_{3}\,[wj]\,\rangle$ a smooth function
and use Riemann integration and assign to each vector a well defined value
$w$ (not just up to a set of measure zero)}; we shall define the Gamow vectors
(describing $\rho^{0}$) in terms of linear combinations of these
c.m.-energy eigenvectors with a definite value of $j$. However,
since the resonances form and decay under the influence of an
interaction and the $|\hat{p}j_{3}\,[wj]\,\rangle$ are interaction-free
eigenvectors of the ``free-particle'' Hamiltonian
\begin{equation}
\label{freehamiltonian}
K=P_{1}^{0}+P_{2}^{0}
\end{equation}
we have to go from the free-particle basis vectors
(\ref{expansionhat}) to the interaction-basis vectors.
This can be done in analogy to the non-relativistic case and may be
justified in two ways :\\
1) One assumes that the time translation generator for the interaction
system has two terms (\cite{sW95} Ch.3),
$H_{0}$ and the interaction $V$
\begin{equation}
\label{interactionhamiltonian}
H=H_{0}+V
\end{equation}
in such a way that to each eigenvector of $H_{0}$ with
eigenvalue $E=w\sqrt{1+{\bbox{\hat{p}}}^{2}}$,
\begin{equation}
\label{freeeigenvalue}
H_{0}\,|\,\hat{p}j_{3}\,[wj]\,\rangle=E\,|\,\hat{p}j_{3}\,[wj]\,\rangle\,,
\end{equation}
there correspond eigenvectors of $H$ with the same eigenvalue
\begin{equation}
\label{interactioneigenvalue}
H|\hat{p}j_{3}\,[wj]^{\pm\, int}\,\rangle=E\,|\,\hat{p}j_{3}\,[wj]^{\pm\, int }
\,\rangle \, .
\end{equation}
Since vectors are not completely defined by the requirement that
they be eigenvectors of an operator with a given eigenvalue
(but may differ by a phase factor (phase shifts) or unitary transformation
(S-matrix) in case of degeneracy) we have added the additional label int.
This additional specification of the eigenvectors can be chosen in a
variety of ways that are connected with the spaces $\Phi$ that one admits,
i.e., with initial and final boundary conditions (as explained for the
non-relativistic case in \cite{aB93}).
Since (\ref{interactionhamiltonian}) may be a questionable
hypothesis in relativistic physics a second justification does not
make use of the existence of the Hamiltonian splitting
(\ref{interactionhamiltonian}).\\
2) One assumes the existence of an $S$-operator and of
M{\o}ller operators $\Omega^{+}$ and $\Omega^{-}$.
$\Omega^{+}$ transforms non-interacting states $\phi^{in}$ which
are prepared by an apparatus far away from the interaction region
into exact state vectors $\phi^{+}$,
\begin{equation}
\label{freein}
\Omega^{+}\phi^{in}=\phi^{+}\, ,\quad
\phi^{+}(t)=e^{-iHt}\phi^{+}\, ,
\end{equation}
which evolve with the exact time-evolution operator $H$.
$\Omega^{-}$ transforms observables $|\psi^{out}\rangle\langle\psi
^{out}|$ registered by the detector placed far away from the
interaction region into the vectors $\psi^{-}$ which evolve with
the exact $H$ in the interaction region :
\begin{equation}
\label{freeout}
\Omega^{-}\psi^{out}=\psi^{-}\, ,
\quad
\psi^{-}(t)=e^{iHt}\psi^{+}\, ,
\end{equation}
where $t$ is the time in the c.m.\ frame.
The basis vectors for the free-particle space and the interaction-basis
vectors are then assumed to be related by
\footnote{In non-relativistic scattering off a fixed target one assumes
that the $|\bbox{p}^{+}\rangle$ related by (\ref{moeller})
to the $|\bbox{p}\rangle$ are not eigenvectors
of $\bbox{P}$ since $[V,P_{i}]\ne 0$.}
\begin{equation}
\label{moeller}
|\,\hat{p}j_{3}\,[wj]^{\pm}\,\rangle=\Omega^{\pm}|\,\hat{p}j_{3}\,[wj]\,\rangle
\, .
\end{equation}
If (\ref{interactionhamiltonian}) also holds then the symbol
$\Omega^{\pm}$ at the center-of-mass is given by the solution of the
Lippmann-Schwinger equation
\begin{equation}
\label{interactioninout}
|\bbox{0}j_{3}\,[wj]^{\pm}\,\rangle=\left(1+\frac{1}{w-H\pm i\epsilon}V\right)
|\bbox{0}j_{3}\,[wj]\,\rangle\, .
\end{equation}
The vectors $|\hat{p}j_{3}\,[wj]^{\pm}\,\rangle$ are obtained from the basis
vectors at rest $|\bbox{0}j_{3}\,[wj]^{\pm}\,\rangle$ by the boost (rotation-free
Lorentz transformation) ${\cal U}(L(\hat{p}))$ whose parameters
are the $\hat{p}^{m}$ and whose generators are the
interaction-incorporating observables
\begin{equation}
\label{interactionobservables}
P_{0}=H,\quad P^{m},\quad J_{\mu\nu} \, ,
\end{equation}
i.e., the exact generators of the Poincar\'e group (\cite{sW95} section $3.3$).
These vectors (\ref{interactioninout}), which for a fixed value of
$[wj]$ span an irreducible representation space of the Poincar\'e
group with the ``exact generators'', will be used for the definition
of the relativistic Gamow vectors. The values
of $j$ and ${\mathsf{s}}=w^{2}$ are $j=\text{ integer }$
(for $s_{1}=s_{2}=0$ otherwise also half integer) and
$(m_{1}+m_{2})^{2}\leq {\mathsf{s}} <\infty$.
The value of $j$ will be fixed and represents the resonance spin;
the same we do with parity and the degeneracy quantum numbers
($n,\,\eta$). The values of ${\mathsf{s}}$ we shall continue from the
physical values into the complex plane of the relativistic
$S$-matrix.
\acknowledgments
We are grateful for some helpful correspondence with L.~Michel.
Support from the Welch Foundation is gratefully acknowledged.
|
\section{Introduction:}
4U~2129+47 and EXO~0748-676 are two low-mass x-ray binaries which
undergo high/low transitions in their X-ray flux. Both are viewed
nearly edge on, and their eclipse lightcurves show evidence for an
extended x-ray emission region often called an accretion disk corona
(ADC). 4U~2129+47 is currently the only ADC source in a low state,
and was very well studied in the 1980s when it was in the high state.
EXO~0748-676 is currently in the high state, but was serendipitously
observed in a low state by the Einstein Observatory before it was
discovered as a bright transient source (Parmar et~al.\/ 1986).
The optical and x-ray lightcurves of 4U~2129+47 in the 1970s and 1980s
showed a v-shaped partial eclipse, which lead to the model of a
partial eclipse of an extended x-ray emission region (McClintock
et~al.\/ 1982). The inclination is believed to be high enough so that
the accreting neutron star is not directly visible, but is shielded
from our view by vertical structure at the outer edge of the accretion
disk. The x-rays we do observe are only the few percent of those
emitted from the vicinity of the neutron star which are scattered into
our line of sight by the ADC. EXOSAT observations in 1983 failed to
detect the source, and the optical modulation seen previously was also
found to be missing (Peitch et~al.\/ 1986). We previously reported on
observations of this source in quiescence with the ROSAT HRI, which,
although sufficient to detect the source at $L_x \sim 10^{33.5}$erg/s,
could not measure the spectrum or a detailed lightcurve (Garcia 1994).
EXO~0748-676 was discovered in outburst with EXOSAT (Parmar et~al.\/ 1986).
While it shows sharp (square wave) eclipses, there is a residual flux
of a few percent at the bottom of the eclipse. This residual flux is
interpreted as due to an accretion disk corona, which covers a
large geometric area and is therefore not fully eclipsed.
It was serendipitously observed with the Einstein IPC
before it entered the high, discovery state.
Because of their transient nature, 4U~2129+47 and EXO~0748-676 are
often referred to as soft x-ray transients (SXT). Several SXT have
been shown to have mass functions $>3M_\odot$, and on this basis are
considered black holes, or BH~SXT (van Paradijs and McClintock 1995,
Tanaka and Shibazaki 1996). Because they show Type~I x-ray bursts,
4U~2129+47 and EXO~0748-676 clearly contain neutron stars (NS), and
are therefore NS~SXT. Observations of BH~SXT provide good evidence
that they accrete via advection dominated accretion flows (ADAF,
Narayan, McClintock \& Yi 1996, Narayan, Barret \& McClintock 1997),
rather than by thin disks, when they are in quiescence. The
theoretical basis for ADAF applies equally well to NS~SXT in
quiescence (Yi et~al.\/ 1996). Detailed study of the accretion
including the effects of a rotating magnetic field certainly allow
that ADAFs exist in NS SXT (Menou et~al.\/ 1999).
Comparison of outburst and quiescent x-ray luminosities in SXT may
have provided evidence for the existence of event horizons in BH~SXT.
(Narayan, Garcia, \& McClintock 1997a, Garcia et~al.\/ 1998, Asia
et~al.\/ 1998). Extending the small sample of objects used in these
studies may help to prove (or disprove) this fundamental result; hence
these observations of two NS~SXT in quiescence are of interest.
In addition, these source are of interest because they offer
opportunity to study the way the structure of the accretion disk and
ADC are effected by wide variations in the x-ray flux of the central
source. Accretion disk models predict that the both the vertical
structure and the run of temperature with radius changes dramatically
under the influence of strong x-ray irradiation (Shakura \& Sunyaev
1973, Vrtilek et~al.\/ 1990), but there are few chances to confirm
these predictions over wide variations in luminosity. The ADC is formed
as a result of the strong x-ray irradiation (Begelman, McKee \&
Shields 1983), and once again there are few opportunities to study its
structure (in a single source) over very wide ranges in x-ray
luminosity.
\section{Observations:}
EXO~0748-676 was serendipitously observed in its low state during a
10~ks observation with the {\sl Einstein}\/ Observatory in May~1980.
The ROSAT PSPC observed 4U~2129+47 in the low state for a total of
$\sim$30~ks on 3~June~1994, and these data were processed with the
standard SASS pipeline. Both datasets were analyzed using IRAF/PROS
V2.3.1 and XSPEC V10.0. Poisson weighting of the errors (Gehrels 1986)
was used. The results from PROS and XSPEC were found to be
consistent. When necessary, we grouped the data in larger energy
bins in order to maintain the number of counts per bin ${_>\atop^{\sim}}$~9.
Previous studies have shown that under these conditions, the results
from Poisson weighting of $\chi^2$ fitting are consistent with those
from maximum likelyhood methods (Narayan, Barret \& McClintock 1997b)
In order to guide our extraction of the source counts from the images,
we first generated the azimuthally averaged radial profile of counts
centered on the sources. For 4U2129+47 we found that a $0.5'$ radius
extraction maximized the S/N. For EXO~0748-676, we found that a $2'$
radius extraction maximized the S/N. For both sources the background
was determined from a larger annulus around the source.
For both sources, we found that a variety of simple models
(Raymond-Smith thermal, bremsstrahlung, blackbody, power law) gave
equally acceptable fit results. As our main interest in the spectral
fitting is to derive quiescent luminosities, in order to allow
comparisons between NS and BH luminosity swings (i.e.,
Narayan et~al.\/ 1997a, Asai et~al.\/ 1998)
we limit ourselves below to
simple black-body fits and compute unabsorbed luminosities over the
0.5-10~keV range. We note that the computed luminosities are
insensitive to the exact form of the spectrum, for example, using a
bremsstrahlung spectrum results in only 10\% changes in the
luminosities.
\subsection{EXO~0748-676 Quiescent Spectrum}
The $\sim 70$~source counts extracted from the IPC image were fit to a
variety of simple spectral models. The best fit blackbody parameters
are ${\rm kT} = 0.14$~keV, ${\rm log(N_H)} = 21.9~$cm$^{-2}$. The very low
number of counts allow a wide range in acceptable parameters. In
order to limit the parameter space, we restrict ourselves to log(${\rm
N_H}) = 21.35$~cm$^{-2}$ (which is within the allowed range). This is
the value predicted by the relation of Predehl and Schmitt (1995),
given the optical reddening of E(B-V)$=0.42 \pm 0.03$ (Schoembs and
Soeschinger 1990). We then computed the confidence regions for
temperature and emitted (unabsorbed) luminosity, as shown in Figure~1.
The best fit blackbody temperature and 0.5-10.0~keV luminosity,
assuming a distance of 10~kpc (Parmar et~al.\/ 1986), are
$${\rm kT = 0.22^{+0.14}_{-0.10}~keV}$$
$${\rm L_x} = 1.0^{+0.5}_{-0.2} \times 10^{34} {\rm ergs~s^{-1}}$$
These results are fairly insensitive to the value of ${\rm N_H}$
assumed, in that a 50\% increase results in only a 15\% decrease in
best fit temperature and a 20\% increase in the emitted 0.5-10.0~keV
luminosity. This energy band contains ${_>\atop^{\sim}}$~70\% of the bolometric
luminosity at the best fit temperature. At the
lowest temperatures allowed by the 90\% confidence interval
(Figure~1), this band contains only 30\% of the bolometric luminosity.
The effective radius of this possible blackbody emitter (${\rm R = (
L_{bol}/4\pi\sigma T^4)^{1/2}}$ ) is ${\rm R} = 8^{+12}_{-5}$~km, comparable
to the radius of a neutron star.
\subsubsection{4U~2129+47 Quiescent Spectrum}
Various simple models (as above) give acceptably good fits to the
$\sim 200$ source counts extracted from the image, and the fit
statistics show no preference for one model over any other. The best
fitting black body model has ${\rm kT} = 0.18$~keV and ${\rm log(N_H)} =
21.5~$cm$^{-2}$. As with EXO~0748-676, we fix the absorption at the
optically determined value (which is within the fit range) in order to
reduce the allowed parameter range. The optical extinction has been
found to be $A_V = 0.9$ (Cowley \& Schmidtke 1990), which corresponds
to ${\rm log(N_H)} = 21.2$ (Predehl \& Schmitt 1995). We then
computed the confidence regions for temperature and emitted
(unabsorbed) luminosity, as shown in Figure~2. The best fit blackbody
temperature and 0.5-10.0~keV luminosity, assuming a distance of
6.3~kpc (Parmar et~al.\/ 1986) are
$${\rm kT = 0.21_{-0.03}^{+0.04} keV}$$
$${\rm L_x = 5.3^{+0.8}_{-1.0} \times 10^{32} ergs~s^{-1} }$$
We note that this ${\rm A_V}$ is measured to the F7IV counterpart of
4U2129+47, which is clearly not the secondary transferring mass to the
neutron star. In using this ${\rm A_V}$ and a spectroscopic distance of
6.3~kpc (Cowley \& Schmidtke 1990) we are implicitly assuming that
this star is in the physical proximity of the mass transferring binary
(Thorstensen et~al.\/ 1988, Garcia et~al.\/ 1989, Cowley \& Schmidtke
1990, van Paradijs \& McClintock 1995.)
Once again, these results are fairly insensitive to the value of ${\rm
N_H}$ assumed, in that a 50\% increase results in only a 10\% drop in
best fit temperature and a 40\% increase in the emitted bolometric
luminosity. The 0.5-10.0~keV band contains $\sim 80$\% of the
bolometric luminosity at the best fit temperature, and $\sim 70$\% at
the lowest allowed temperature.
The effective radius of this possible blackbody emitter (${\rm R =
(L_{bol}/4\pi\sigma T^4)^{1/2}}$ ) is $R = 1.7^{+0.5}_{-0.6}$~km,
substantially smaller than the radius of a neutron star.
We previously reported the quiescent ROSAT HRI flux (Garcia 1994),
assuming the spectral parameters measured in the high state. This
overestimates the flux by a factor of two. When the softer quiescent
spectrum determined above is used we calculate an emitted luminosity
(0.5-10.0~keV) of $9.7 \times 10^{32}$ergs~s$^{-1}$. The 68\%
confidence bounds on the PSPC spectrum correspond to a $\sim 20\%$
uncertainty in the calculation of the HRI flux. Thus it appears that
the source has faded by a factor of $\sim 2$ during the 2.5~year
interval between the HRI and PSPC observations.
\subsubsection{4U~2129+47 Quiescent Lightcurve}
We generated the quiescent lightcurve of 4U~2129+47 (Figure~3) by
binning the background subtracted PSPC data into 7 bins based on the
McClintock et~al.\ 1982 ephemeris. Two other lightcurves are plotted
in Figure~3: The scaled on-state lightcurve re-extracted from the IPC
CD-ROM archive, and a square-wave lightcurve with an eclipse width
of~0.1. The eclipse duration of this last lightcurve is $\sim 1/2$
the width of the high-state eclipse, which is what one expects if the
x-ray source was a point (rather than extended) source.
In order to determine if the observed low-state lightcurve was well
described by either the square-wave or the scaled on-state lightcurve,
we calculated the chi-squared for each of these. The trial
lightcurves were first artificially binned to match the sampling of
the observed (7 bin) lightcurve. For the scaled on-state lightcurve
we calculate a reduced $\chi^2$ for 6~degrees of freedom of
$\chi^2/\nu = 0.53$ (80\% random probability), and for the square wave
we find $\chi^2/\nu = 1.7$ (10\% random probability). Clearly either
trial lightcurve is an acceptable representation of the observations,
although the scaled on-state lightcurve may be somewhat favored.
Given the limited statistics, one might reasonably ask if any source
variability has been formally detected at all. Testing the 7~bin
lightcurve against a steady source we find $\chi^2/\nu = 1.2$, clearly
allowing that the source is steady. However, two other tests provide
evidence that some sort of an eclipse still occurs in the
quiescent state.
First, we have cross-correlated the observed 7~bin lightcurve against
the scaled on-state curve in order to determine the phase of minimum
light. The best fit phase and 68\% confidence limits are $0.0\pm 0.2$
on the ephemeris of McClintock et~al.\/ (1982). The accumulated error in the
ephemeris is $\pm 0.1$, so while the present data agrees with this
ephemeris it is not able to refine it. Given the large error in the
determination of phase zero, this provides only weak evidence for an
eclipse.
Second, we binned the data into 4~equal bins, one of which is centered
on the eclipse phase. Testing this more coarsely binned lightcurve
against a steady source we find $\chi^2/\nu = 3.3$, which has a random
probability of only $\sim 2\%$ for a steady source. In addition, the
minimum of this binned lightcurve occurs at the expected phase of
eclipse, which has a random probability of 1 in 4. This provides
somewhat stronger evidence that an eclipse may still be occurring at
phase zero.
\section{Discussion}
A glance at Figure~3 shows that the quiescent lightcurve appears
remarkably similar to a scaled version of the outburst lightcurve.
However, the statistics are poor, and the lightcurve binned into
7~phase intervals is formally consistent with a steady source, a
square wave (eclipse), or a scaled version of the on-state lightcurve.
However, when binned more coarsely the lightcurve provides tantalizing
evidence that there is an eclipse occurring at the
expected phase.
This result differs somewhat from what we found in the ROSAT HRI data
obtained $\sim 2.5$~years before these data. The HRI data excluded a
scaled version of the outburst lightcurve, and these PSPC data allow
it. Given that the source flux decreased by a factor of $\sim 2$
between the observations, it would not be surprising if the lightcurve
also changed shape. However, the statistical uncertainties of the HRI
lightcurve are large, and (like the PSPC lightcurve) they allow the
possibility that the source was constant.
The eclipse (square wave) lightcurve is what one expects based on the
standard models of the system. At this low luminosity, the ADC should
have collapsed (White \& Holt 1982, Begelman, McKee \& Shields
1983), and the secondary should eclipse the x-ray emission from the
neutron star for $\sim 10\%$ of the orbit. A smoothly modulated
lightcurve would be hard to understand, as it would imply that the
structure of the accretion disk has not changed despite the observed
large change in x-ray luminosity, and presumably mass transfer rate.
In the high state, the modulation of the x-ray flux was accompanied by
a strong modulation of the optical flux. What optical modulation
should we expect given the apparent modulation in the low-state x-ray
flux? In order to answer this question we need an estimate of the
quiescent disk V magnitude, which we make by comparing 4U~2129+47 to
Cen X-4. This SXT also contains a NS primary, and the quiescent disk
has been measured to have an absolute magnitude ${\rm M_V \sim 9}$
(McClintock \& Remillard 1990). The disk in Cen X-4 may be somewhat
bigger than that in 4U~2129+47 due to its longer period (15.1~h), and
also may appear somewhat brighter due to its more face-on viewing
angle ($i \sim 40^o$, Shahbaz, Naylor \& Charles 1993). Based
on the amplitude of the optical modulation in outburst
(McClintock, Remillard, \& Margon 1981)
and the correlation between modulation and inclination
(Van Paradijs, Van der Klis, \& Pedersen 1988), the
inclination of 4U~2129+47 is likely to be $i$~${_>\atop^{\sim}}$~80$^o$. The
differences in the period and inclination would
lower our estimate of the absolute magnitude of
the quiescent disk in 4U~2129+47 by several magnitudes, based
on the correlations found in cataclysmic variables (i.e., equations 2.63 and
3.3 from Warner 1995). Thus ${\rm M_V \sim 9}$ is a comfortable lower
limit to the quiescent disk in 4U~2129+47.
At the 6.3~kpc distance of 4U~2129+47, and with ${\rm A_V} =
1.5$, this disk would have ${\rm V} \sim 24.5$. The quiescent counterpart
has ${\rm V}=17.9$, so even if the disk were 100\% modulated we would expect
a fractional modulation of only $\sim 0.2\%$, which is well below the
observed limit of 1.2\% amplitude (99\% confidence, Thorstensen
et~al.\/ 1988).
\subsection{Comparison to Other Quiescent SXT}
The x-ray temperatures of $\sim 0.2$~keV which we find for
these two SXT in quiescence are similar to those found for other
NS~SXT (Verbunt et~al.\/ 1994. Asai et~al.\/ 1996, Asai et~al.\/ 1998
and references therein). The luminosity of 4U~2129+47 is also similar to that
found in these sources, but EXO~0748-676 is at the extreme high end of
the distribution (Narayan et~al.\/ 1997a). There is no obvious reason
for this, in particular the orbital period of EXO~0748-676 is typical,
so the average mass transfer rate should be typical as well (Menou
et~al.\/ 1999). Given
the observed variability of quiescent NS~SXT (Campana et~al.\/ 1998), it
is possible that the single quiescent measurement caught this system in
a particularly luminous state.
The emitting radius we compute assuming a black-body spectrum is
smaller than a NS in the case of 4U~2129+47. Black-body fits often
indicate emitting areas smaller than a NS surface, indicating either
that the accretion is channeled onto a small fraction of the NS
surface (Asai et~al.\/ 1996, Menou et~al.\/ 1999)
or that the
black-body spectral fits erroneously indicate a small surface area
(Rajagopal \& Romani 1996, Rutledge et~al.\/ 1999). EXO~0748-676 is
unusual in that the black-body radius is consistent with the entire
NS surface, owing to the larger than typical luminosity in
quiescence.
In summary, we note that the difference in the luminosity swings of
outbursting and quiescent SXT may indicate the existence of event
horizons in BH~SXT (Narayan et~al.\/ 1997a, Garcia et~al.\/ 1998, Asai
et~al.\/ 1998, Menou et~al.\/ 1999, but see Chen et~al.\/ 1998 for an
opposing view). Observations with AXAF and XMM will undoubtedly add
more SXT to the studied sample, and push upper limits lower,
helping to more definitively test this difference. XMM may allow an
accurate measurement of the quiescent light curve of the eclipsing
NS~SXT 4U~2129+47, therefore providing a spatially resolved picture of
an ADC in a quiescent system.
We thank Jeff McClintock for helpful comments on an earlier draft of
this paper. This work was supported in part by the Chanrda Science
Center through contract NAS 8-39073.
\clearpage
\newpage
|
\section{Introduction}
In most of the numerical analysis literature, complexity
and stability of numerical algorithms are usually estimated
in terms of the problem instance dimension and of a `condition number'.
\par
For instance, the complexity of solving an $n \times n$
linear system $Ax=b$ is usually estimated in terms of the
dimension $n$ (actually the input size is $n(n+1)$) and of
the condition number $\cond{} (A) = \norm{A} \norm{A^{-1}}$.
\par
There is a set of problems instances with $\cond{}(A) = \infty$,
and in most cases it makes no sense to attempt solving those
problem instances. There are also problem instances (in our case, matrices)
close to the locus of degenerate problem instances. Those will have a
large condition number, and will be said to be {\em
ill-conditioned}.
\par
It is usually accepted that ill-conditioned
problem instances are hard to solve. Thus, for complexity purposes a
problem instance with a large condition number should be considered
`large'. Therefore, when considering problems defined for
real inputs, a reasonable measure for the input size
would be (in our example): $n^2 \log_2 \cond{}(A)$.
(Compare to ~\cite{SMALE97} Formula~2.1 and paragraph below.
See also the discussion in ~\cite{BP}, Chapter~3, Section~1).
\medskip
\par
Another tradition, derived from classical complexity theory
and pervasive in several branches of literature (such as linear
programming), is to consider the subset of problems instances with integer
coefficients. Hence the input size is the number of coefficients
times the bit-size of the largest coefficient (in absolute value).
\par
In this paper, the following classical problems of numerical
analysis are considered:
\newcounter{numitems}
\begin{enumerate}
\item Solving a general $n \times n$ system of linear equations.
\item Minimal squares problem for a full-rank matrix.
\item Non-symmetric eigenvalue problem.
\item Solution of one univariate polynomial.
\item Solution of a non-degenerate system of $n$ polynomial equations
in $n$ variables.
\setcounter{numitems}{\value{enumi}}
\end{enumerate}
\par
All those problems share the feature mentioned above: there is a
degenerate locus, and problem instances with real coefficients can be as close
to the degenerate locus as wished. This implies that they can be
arbitrarily ill-conditioned.
\par
However, in Theorems~\ref{th1} to ~\ref{th5} below, we provide
bounds for the condition number of problems instances with
integer coefficients and not in the degenerate locus.
Those bounds depend on the dimension (size) of the problem instance and on the
bit-size of its coefficients.
\medskip
\par
In the analysis of iterative algorithms, one further considers a
certain quantity that can be used to bound the speed of convergence
and hence the number of iterations
to obtain a given approximation. For instance, for
power methods (or QR iteration without shift)
in the symmetric eigenvalue problem, one can bound
the number of steps in terms of the desired accuracy and of the
ratio between different eigenvalues. The farther this number is
from 1, the faster is the convergence.
\par
Once again, if input has real coefficients, this quantity can be
arbitrarily close to 1. However, explicit bounds for that quantity
will be given for inputs with integer coefficients for
\begin{enumerate}
\setcounter{enumi}{\value{numitems}}
\item QR iteration without shift for the Symmetric Eigenvalue Problem.
\item Graeffe iteration for solving univariate polynomials.
\end{enumerate}
\medskip
\par
The reader should be warned that the results herein are worst
case estimates, and are overly pessimistic for application purposes.
The main motivation for those results is to convert numerical
analysis estimates into `polynomial time' estimates, not the
opposite.
\section{Statement of main results}
\begin{notation} \norm{.} stands for the {\em 2-norm}: if
$x \in \mathbb R^n$ or $\mathbb C^n$, then
\[ \norm{x} = \sqrt{\sum_{i=1}^n |x_i|^2}
\punct{.}
\]
If $A$ is a matrix, then
\[
\norm{A} = \sup_{\norm{x} = 1} \norm{Ax}
\punct{.}
\]
\end{notation}
\subsection{Linear equation solving}
The first problem considered is linear equation solving:
given an $n \times n$ matrix $A$ and a vector $b \in
\mathbb R^n$, find $x \in \mathbb R^n$ such that
$ Ax = b $.
\par
Its condition number (with respect to the 2-norm) is
defined as
\[
\cond{}(A) = \norm{A} \norm{A^{-1}}
\punct{.}
\]
\par
Comprehensive treatment of the perturbation theory for this
problem can be found in the literature, such as~\cite{DEMMEL}
Section~2.2, ~\cite{HIGHAM} Chapter~7, ~\cite{TB} Lecture~12,
etc...
\begin{theorem}\label{th1}
Let $A$ be an $n \times n$ matrix with integer
coefficients. If $A$ is invertible, then
\[
\cond{}(A) \le n^{\frac{n}{2}+1}
\left(
\max_{i,j} |A_{ij}|
\right) ^n
\punct{.}
\]
\end{theorem}
\par
No originality is claimed for Theorem~\ref{th1}. This result is
included for completeness and because its proof is elementary,
yet illustrates the principle behind the other results.
\subsection{Minimal squares}
The second problem in the list is minimal squares fitting.
Let $A$ be an $m \times n$ matrix, $m \ge n$, with full rank,
and let $b \in \mathbb R^m$. One has to find $x$ to minimize
$\norm{A x - b}^2$.
\par
Let $r = Ax - b$ be the residual, we are minimizing $\norm{r}^2$.
Let
\[
\sin \theta = \frac{\norm{r}}{\norm{b}}
\punct{.}
\]
\par
According to~\cite{DEMMEL} p. 117
(Compare to ~\cite{TB} Lecture~18 and~\cite{HIGHAM}
Section~19.1),
the condition number of the linear least squares problem
is
\[
\cond{LS} (A,b)
=
\frac{2 \cond{}(A)}{\cos \theta} + \tan \theta \ \cond{}(A)^2
\punct{.}
\]
\par
Since we do not assume $A$ to be square, we need to give a
new definition for $\cond{}(A)$. Let $\sigma_{\text{\tiny MAX}}(A)$
and $\sigma_{\text{\tiny MIN}}(A)$ be respectively the largest and
the smallest singular values of $A$. Then set
\[
\cond{}(A) = \frac{\sigma_{\text{\tiny MAX}}(A)}{\sigma_{\text{\tiny MIN}}(A)}
\punct{.}
\]
When $m=n$, this definition is equal to the previous one.
\medskip
\par
The singular locus is now the set of pairs $(A,b)$ such that
$A$ does not have full rank (i.e. $\sigma_{\text{\tiny MIN}}(A)=0$)
or such that $\norm{r} = \norm{b}$ (i.e. $b$ is orthogonal
to the image of $A$).
\par
The result is:
\begin{theorem}\label{th2}
Let $A$ be an $m \times n$ matrix with integer coefficients,
and assume that $A$ has full rank. Let $b \in \mathbb Z^m$.
Set $H = \max _{i,j} \left( |A_{ij}|, |b_i| \right)$. Then
if $b$ is not orthogonal to the image of $A$, we have:
\[
\cond{LS} (A,b) \le
3n ^{\frac{n}{2}+1} m^{n+\frac{1}{2}}H^{2n+1}
\punct{.}
\]
\end{theorem}
\par
\subsection{Non-symmetric eigenvalue problem}
Let $A$ be an $n \times n$ matrix and let $\lambda$
be a single eigenvalue of $A$. The condition number
of $\lambda$ depends on the angle
between the left and right eigenvectors:
\par
Let $x$, $y$ be respectively right and left norm-1
eigenvectors of $A$ associated to $\lambda$:
$Ax = \lambda x$, $y^* A = \lambda y^*$, and
$\norm{x} = \norm{y} = 1$. Then
\[
\cond{NSE}(A, \lambda) = \sec(\widehat{x,y}) = \frac{1}{y^* x}
\punct{.}
\]
\par
See ~\cite{DEMMEL} Theorem~4.4 p.~149 for references.
\begin{theorem}\label{th3}
Let $A$ be an $n \times n$ matrix with integer
coefficients, and let $\lambda$ be a single
eigenvalue of $A$. Then
\[
\cond{NSE}(A, \lambda)
\le
n^{3n} 2^{2n} \left( 2 \sqrt{n} H(A) \right)^{2 n^3 - 2 n}
\punct{.}
\]
\end{theorem}
\subsection{Solving univariate polynomials}
The condition number (in affine space) for solving a univariate
polynomial $f(x) = \sum_{i=0}^d f_i x^i$ can
be defined~(\cite{BCSS} page 228) as:
\[
\mu(f) = \max_{\zeta \in \mathbb C: f(\zeta)=0} \mu(f,\zeta)
\punct{,}
\]
where
\[
\mu(f, \zeta) =
\frac { \left( \sum_{i=0}^d |\zeta|^{2i} \right)^{\frac{1}{2}}}
{ |f'(\zeta)|}
\punct{.}
\]
\par
The degenerate locus is the set of polynomials with a multiple
root or with a root at infinity.
\begin{theorem}\label{th4}
Let $f: x \mapsto \sum_{i=0}^d f_i x^i$ be a univariate polynomial with
integer coefficients, without multiple roots. Then
\[
\mu(f) \le 2^{2d^2 - 2} d^{2d} \left(\max |f_i| \right)^{2d^2}
\punct{.}
\]
\end{theorem}
\subsection{Solving systems of polynomials}
A similar condition number exists for systems of polynomials.
However, for the purpose of condition number theory, it is usually
convenient to homogenize the equations and to study the
perturbation theory of the `roots' in complex projective
space. This can also be seen as a change of metric, that
simplifies the formula of the condition number and of several
theorems (See~\cite{BCSS} Chapters~10, 12, 13).
\par
Let $f = (f_1,$ $\cdots,$ $f_n)$ be a system of polynomials in
variables $x_1, \cdots, x_n$. We homogenize the system by
multiplying each coefficient
$f_{iJ} x^J = f_{iJ} x_1^{J_1} x_2^{J_2} \cdots x_n^{J_n}$ of $f_i$ by
$x_0^{J_0}$, where we choose $J_0 = \deg f_i - (J_1 + \cdots J_n)$.
We obtain a system of homogeneous polynomials in $n+1$ variables,
that we call $F=(F_1, \cdots, F_n)$. The natural space for the
roots of $F$ is projective space $\mathbb P^n$, defined as the
space of all `rays'
\[
(x_0: \cdots : x_n) =
\{ (\lambda x_0, \cdots, \lambda x_n): \lambda \in \mathbb C\}
\punct{.}
\]
\par
where $x_0$, \dots, $x_n$ are not all equal to $0$.
\par
Every finite root $(x_1, \cdots, x_n)$ of $f$ corresponds to the
projective root of $F$ given by $(1: x_1: \cdots: x_n)$, and
projective roots of $F$ correspond either to a finite root
of $f$ or to a root `at infinity'.
\par
Suppose that the coefficients of $f$ (hence of $F$) are made
to depend upon a parameter $t$. The condition number
bounds the absolute speed of the roots of $F$ (in
projective space) with respect to the absolute speed
of the coefficients of $F$. Recall that the roots $\zeta$
of $F$ are in projective space, so their speed vector $\dot \zeta$
belongs to the tangent space $T_{\zeta} \mathbb P^n$.
\par
The condition number of $F$ at a root
turns out to be:
\[
\mu(F,\zeta) = \norm{F} \norm{ \left( \mathrm{D}F(\zeta)_{| T_\zeta}\right) ^{-1}
\left[ \begin{matrix}
\norm{\zeta}^{d_1 -1}\\
& \ddots\\
& & \norm{\zeta}^{d_n -1}\\
\end{matrix} \right] }
\]
where $\zeta \in \mathbb C^{n+1}$ is such that
$(\zeta_0 : \cdots : \zeta_n)$ is a root of $F$
(See Proposition 7c in Page 230 of~\cite{BCSS}).
We did not define the norm of a polynomial yet. Above,
$\norm{.}$ stands for the unitary invariant
norm~(See \cite{WEYL} Chapter III-7 or ~\cite{BCSS} Section~12.1),
that is the most reasonable
generalization of the 2-norm to spaces of
polynomials:
\begin{notation}
Let $G$ be a homogeneous degree $d$ polynomial in $n+1$ variables.
Then
\[
\norm{G} = \sqrt{ \sum_J \frac{ |G_J|^2 }{\binomial{d}{J}} }
\]
where $\binomial{d}{J}$ is $\frac{d!}{J_0 ! \cdots J_n!}$.
Let $F$ be a system of homogeneous polynomials. Then
\[
\norm{F} = \sqrt{ \sum \norm{F_i}^2 }
\punct{.}
\]
\end{notation}
\par
With these definitions, the number $\mu(F,\zeta)$ is invariant
under scalings of $F$, $\zeta$, and under the action of the
unitary group $U(n+1)$, where an element $Q \in SU(n+1)$ acts
by $Q:(F,\zeta) \mapsto (F \circ Q, Q \zeta)$.
\par
In order to define the condition number of a system of
$n$ equations in $n$ variables, we set:
\[
\mu(f) = \max_{\zeta} \mu(F,\zeta)
\]
where $\zeta$ ranges over the roots of $F$. (Another
possibility is to restrict $\zeta$ to the non-degenerate
roots of $F$. This would make no difference in this paper).
The following theorem is true if one restricts $\zeta$ to
any subset of the roots of $F$.
\begin{theorem}\label{th5}
Let $f$ be a system of $n$ polynomial equations in $n$ variables,
with integer coefficients. We write $H(f)$ for the maximum of the absolute
value of the coefficients of $f$, $S(f)$ for the number of
non-zero coefficients of $f$ and $D$ for $\max d_i$.
Assume that $\mu(f)$ is finite. Then
\[
\mu(f) \le ((n+1) S H(f))^{D^{cn}}
\]
where $c$ is an universal constant.
\end{theorem}
\subsection{Symmetric eigenvalue problem}
\label{sep}
\par
Let $A$ be an $n \times n$ real positive symmetric matrix, and let
$\lambda_1 \ge \lambda_2 \ge \cdots \lambda_n \ge 0$ be its
eigenvalues.
\par
Unlike the non-symmetric eigenvalue problem,
the symmetric eigenvalue problem has absolute condition number
always equal to 1 (See \cite{DEMMEL} Theorem~5.1. See also ~cite{PARLETT}
Fact~1.11 p.16).
\par
However, when using an iterative algorithm, the
ratio of eigenvalues
\[
\ratio(A) = \min_{j > i} \frac{\lambda_j}{\lambda_i}
\]
may play an important role for estimating convergence. For instance,
according to ~\cite{TB} Theorem~28.4, the QR algorithm without shift
converges linearly with speed $\frac{1}{\ratio(A)}$. Convergence may
get slower when $\ratio(A) \rightarrow 1$. Therefore one can bound
the speed of convergence by bounding
\[
\relgap(A) =
\ratio(A) - 1
=
\min_{j > i} \frac{\lambda_j - \lambda_i}{\lambda_i}
\]
above from zero. If $\relgap(A) > \delta_0$, then
$\ratio(A) > 1+\delta_0$.
After $k > \lceil \frac{1}{\delta_0} \rceil$ iterations,
one gets
\[
\ratio(A)^{k} > 1 + k \delta_0 \ge 2
\punct{.}
\]
\par
Thus it suffices to perform
$O( \frac{1}{\delta_0} \log_2 \frac{1}{\delta_1})$
iterations to obtain a result with accuracy $\delta_1$.
\par
Also, the quantity
$\relgap(A)^{-1}$ can also be interpreted as a condition number
for the eigenvectors (See ~\cite{DEMMEL} Theorem~5.7 p.~208).
We will show here that
\par
\begin{theorem}\label{th6}
Let $A$ be an $n \times n$ matrix with integer coefficients.
Then
\[
\relgap(A)^{-1} \ge 8^{-n} (4n)^{-n^2} \left( \max_{i,j} |A_{ij}| \right)^{-2n^2}
\punct{.}
\]
\end{theorem}
\subsection{Graeffe iteration}
Let $f: x \mapsto \sum_{i=0}^d f_i x^i$, $f_d=1$
be a monic univariate polynomial
with zeros $\zeta_1, \cdots, \zeta_d$. Those zeros can be ordered
such that
\[
|\zeta_1 | \ge |\zeta_2| \ge \cdots \ge |\zeta_d|
\punct{.}
\]
\par
The Graeffe operator maps the polynomial $f(x) = \prod_{i=1}^d (x-\zeta_i)$
into the polynomial $Gf(x) = (-1)^d f(\sqrt{x}) f(-\sqrt{x}) =
\prod_{i=1}^d (x-\zeta_i^2)$.
\par
In ~\cite{OSTROWSKI,OSTROWSKI2}, it is explained how to
recover the actual roots of $f$ after a certain number of
Graeffe iterations, with a good approximation. The number of required
iterations depends on the ratio:
\[
\ratio(f) = \max_{|\zeta_j| > |\zeta_i|} \frac{|\zeta_j|}{|\zeta_i|}
\punct{.}
\]
\par
Unlike in Section~\ref{sep}, we do not require here
that the roots have different absolute value.
We consider also the auxiliary quantity
\[
\relgap(f) = \ratio(f) - 1 = \max_{|\zeta_j| > |\zeta_i|}
\frac{|\zeta_j| - |\zeta_i|}{|\zeta_i|}
\punct{.}
\]
\par
By the above definitions, the `condition number'
$\relgap(f)^{-1}$ is always finite. In order to
recover the roots within relative precision
$\delta$, the number
of Graeffe iterations to perform is
\[
O(\log \relgap(f)^{-1} + \log d + \log \log \delta^{-1})
\punct{.}
\]
\par
For clarity of exposition, we will show that
bound under a special hypothesis: all the roots should be different
positive real numbers. For the general case, see ~\cite{MZ97} and
~\cite{MZ98}. Also, all estimates here are `up to the first order',
and quadratic error terms will be discarded.
\medskip
\par
After $k$ steps of Graeffe iteration one
obtains the polynomial
\[
g(x) = G^kf(x) =
\sum_{i=0}^d g_i x^i = \prod_{i=1}^d (x - \zeta_i^{2^k})
\]
with $\ratio(g) = \ratio(f)^{2^k}$.
\par
Expanding each $g_i$ as the $(d-i)$-th elementary symmetric function
of the $\zeta_i^{2^k}$, one obtains
\begin{eqnarray*}
g_0 &=& \sigma_d ( \zeta_1^{2^k}, \cdots , \zeta_d^{2^k} )\\
g_1 &=& \sigma_{d-1} ( \zeta_1^{2^k}, \cdots , \zeta_d^{2^k} )\\
&\vdots&\\
g_{d-1} &=& \sigma_1 ( \zeta_1^{2^k}, \cdots , \zeta_d^{2^k} )\\
g_d &=& 1
\end{eqnarray*}
\par
We can use the special hypothesis to bound
\[
\left( \zeta_1 \zeta_2 \cdots \zeta_{d-i} \right) ^{2^k} = g_i(1+\delta_i)
\]
with $|\delta_i| < \frac{2^d}{\ratio(g)} + \text{h.o.t.}$
Hence
\[
\zeta_i^{2^k} = \frac{g_{d-i+1}}{g_{d-i}} (1 + \delta_i')
\]
with $|\delta_i'| < \frac{2^{d+1}}{\ratio(g)} + \text{h.o.t.}$
\par
Since we assumed the $\zeta_i$ are all positive, we can recover
them by taking $2^k$-th roots
\[
\zeta_i = \left(\frac{g_{d-i+1}}{g_{d-i}}\right)^{2^{-k}} (1 + \delta_i'')
\punct{.}
\]
\par
with $|\delta_i''| \le \frac{2^{d+1-k}}{\ratio(g)} + \text{h.o.t.}$
\par
Now we can use the estimate on $\ratio(g) = \ratio(f)^{2^k}$ to
deduce that
$O(\log \relgap(f)^{-1} + \log \delta^{-1})$ steps are sufficient
to obtain a relative
precision $\delta$ in the roots. Indeed after $k_1 = \log_2 \relgap(f)^{-1}$
steps,
\[
\ratio(G^{k_1}f) = \ratio(f)^{2 ^{ \log_2 \relgap(f)^{-1}}} \ge 2
\punct{.}
\]
\par
After extra $k_2 = \log_2 (d +1+ \log_2 \delta^{-1})$ steps, one gets
\[
\ratio(G^{k_1+k_2}(f)) > 2^{2^{\log_2 (d+1+\log_2 \delta^{-1})}}
= 2^d \delta^{-1}
\punct{.}
\]
\par
So we can set $k = k_1 + k_2 + 1$, the last $1$ to get rid of the
high order terms, to deduce that $|\delta_i''| < \delta$.
\medskip
\par
\begin{theorem}\label{th7}
Let $f: x \mapsto \sum_{i=0}^d f_i x^i$ be a polynomial with
integer coefficients. Then
\[
\relgap(f)^{-1} > \left( 8 \max |f_i| \right)^{-2d}
\punct{.}
\]
\end{theorem}
\medskip
\par
This says that Graeffe iteration is `polynomial time', in the sense
that we can obtain relative accuracy $\delta$ of the roots after
\[
O( d \log \max |f_i| + \log \log \delta^{-1})
\]
steps.
\section{Background material}
The proof of Theorems~\ref{th3} to ~\ref{th7} will make
use of the {\em absolute multiplicative height function} $H$
to bound inequalities involving
algebraic numbers.
\par
The construction of the height function $H$ is quite standard in number
theory and we refer the reader to ~\cite{LANG} Chapter~II
or to~\cite{SILVERMAN}
pages 205--214. For applications to complexity theory, see~\cite{BCSS}
Chapter~7 and ~\cite{M98}.
\par
The height function is naturally defined in the projectivization
$\mathbb P^n(\qalg)$ of the algebraic numbers $\qalg$. It returns a
real number $\ge 1$. We can also extend it to complex projective space
$\mathbb P^n$ by setting $H(P) = \infty$ when $P \not \in \mathbb P^n(\qalg)$.
We will adopt this convention in order to simplify the notation of
domains and ranges.
\par
Also, if $x = (x_1, \cdots, x_n) \in \mathbb C^n$, we can define its
height as $H(x) = H(x_1: \cdots : x_n:1)$.
\par
We can also define the height of matrices, polynomials and systems of
polynomials as the height of the vector of all the coefficients.
\medskip
\par
The following properties of heights will be used in the sequel.
First of all, we can explicitly write the height of a vector
with integer coefficients as:
\begin{proposition}\label{integers}
If $u \in \mathbb Z^n$, then $H(u) = \max_{1\le i \le n} |u_i|$,
where $|\, . \,|$ is the standard absolute value.
\end{proposition}
Proposition~\ref{integers} follows from the construction of
the height function. One immediate consequence is that if
$v \in \mathbb Q^n$, then $H(v) = \max |m v_i|, |m|$ where
$m$ is the greatest common denominator of the $v_i$'s.
\par
We can use the following fact to bound the height of the
roots of an integral polynomial:
\begin{proposition}\label{roots}
If $f: x \mapsto f(x) = \sum_{i=0}^d f_i x^i$ is a non-zero
polynomial with integer coefficients, and if $x$ is a root of $f$,
then $H(x) \le 2 \max |f_i|$.
\end{proposition}
\par
Proposition~\ref{roots} is Theorem~5 in ~\cite{M98}. Compare
with Theorem~5.9 in~\cite{SILVERMAN}, where the coefficients of
$f$ are algebraic numbers.
\par
We can use a bound on the height to bound absolute values above and below:
\begin{proposition}\label{bound}
Let $K$ be an algebraic extension of $\mathbb Q$,
and let $x \in K$, $x \ne 0$. Then
\[
H(x)^{-\deg[ K:\mathbb Q]} \le |x| \le H(x)^{\deg[ K:\mathbb Q]}
\punct{.}
\]
\end{proposition}
The height of a vector and of its coordinates can be related by:
\begin{proposition}\label{coordinate}
\[
H(x_1) \le H(x_1, \cdots, x_n) \le H(x_1) H(x_2) \cdots H(x_n)
\punct{.}
\]
\end{proposition}
Propositions~\ref{bound} and~\ref{coordinate} follow immediately
from the construction of the height function.
The height function is invariant under permutation of coordinates,
and also:
\begin{proposition}\label{conjugate}
Let $K$ be an algebraic extension of $\mathbb Q$, and
let $g \in \text{Gal}_{[K:\mathbb Q]}$. Then for any $x \in K$,
$H(g(x)) = H(x)$
\end{proposition}
\par
Proposition~\ref{conjugate} is Lemma~5.10 in~\cite{SILVERMAN}.
\begin{proposition}\label{morphisms}
Let
\function{F=(F_0, \cdots, F_m)}
{\mathbb C^{n_1} \times \mathbb C^{n_2} \times \cdots \times \mathbb C^{n_k}}
{\mathbb C^{m+1}}
{P^1, \cdots , P^k}
{F(P^1, \cdots, P^k)}
be a system of multi-homogeneous polynomials with algebraic coefficients,
where each $F_i$ has degree $d_j$ in variables $P^j$.
Let the $P^j$ be algebraic. Then
\[ H(F(P)) \le (\max S(f_i)) H(F) H(P^1)^{d_1} \cdots H(P^k)^{d_k}
\punct{.}
\]
\end{proposition}
\par
In the case $k=1$, this is similar to Theorem~5.6 in~\cite{SILVERMAN}
(where $\max S(f_i)$ is not given explicitly). For the general case
see Theorem~4 in~\cite{M98}.
\begin{proposition}\label{polynomials}
Let
\function
{G=(G_1, \cdots, G_m)}
{\mathbb C^{n_1} \times \mathbb C^{n_2} \times \cdots
\times \mathbb C^{n_k}}
{\mathbb C^{m}}
{Q^1, \cdots , Q^k}
{G(Q^1, \cdots, Q^k)}
be a system of polynomials with algebraic
coefficients, where each $G_i$ has degree at most $d_j$ in
variables $Q^j$. Let the $Q^j$ be algebraic. Then
\[
H(G(Q)) \le (\max S(G_i)) H(G)
H(Q^1)^{d_1} \cdots H(Q^k)^{d_k}
\punct{.}
\]
\end{proposition}
This is Corollary~1 in~\cite{M98}. Some consequences of this
are that $H(\sum_{i=1}^n x_i) \le n \prod H(x_i)$ and that
$H(\prod_{i=1}^n x_i) \le \prod H(x_i)$.
The following fact follows also from the construction
of heights:
\begin{proposition}\label{square}
If $x$ is an algebraic number,
\[
H(x^2) = H(x)^2
\punct{.}
\]
\end{proposition}
\par
Also, it makes sense to bound the height of the roots of
a system of polynomials with respect to the height, size
and degree of the system. Corollary~6 in ~\cite{KP} is:
\begin{proposition}\label{krickpardo}
[Krick and Pardo]
Let $f_1, \cdots , f_r$ $r \le n$ be polynomials
in $\mathbb Z[x_1, \cdots$ , $x_r]$ of degree and height
bounded by $d \ge n$ and $\eta$, respectively,
and let $V$ denote the algebraic affine variety defined
by: $V = \{ x: f_1(x) = \cdots f_r(x) = 0 \}$.
\par
Then $V$ has at most $d^n$ isolated points, and their height verifies:
\[
\log_2 H(P) \in d^{O(n)} (\log_2 r + \log_2 \eta)
\punct{.}
\]
\end{proposition}
\section{Proof of Theorems}
\begin{notation}
If $A$ is a real (resp. complex) matrix, then $A^*$ is the
real (resp. complex) transpose of $A$, $(A^*)_{ij}= \bar{A_{ji}}$.
The same convention will be used for vectors.
\par
The vectors of the canonical basis will be denoted by $e_1 =
[1,0,0,\cdots]^*$, $e_2 = [0,1,0,\cdots]^*$, etc...
\end{notation}
\subsection{Proof of Theorem~\ref{th1}}
\begin{eqnarray*}
\norm{A} &=& \sup_{\norm{u}=1} \norm{A u} \text{\ by definition} \\
&\le& \sum_j |u_j| \norm{ [A_{1j}, \cdots, A_{nj}]^* }
\text{\ by triangular inequality}\\
&\le& \sqrt{n} \max_j \norm{ [A_{1j}, \cdots, A_{nj}]^* }
\text{\ since \norm{u}=1}\\
&\le& n \max_{ij} |A_{ij}|
\end{eqnarray*}
\par
Let $A(i, u)$ be the matrix obtained by replacing
the $i$-th column of $A$ by the vector $u$. Then if
$v = A^{-1} u$, Cramer's rule is:
\[
v_i = \frac{ \det A(i, u) } {\det A}
\punct{.}
\]
\par
Since $A$ has integer coefficients and
$\det A \ne 0$, one can always bound $|v_i| \le |\det A(i, u)|$.
By Hadamard inequality, this implies:
\begin{eqnarray*}
|v_i| &\le& \norm{u} \max_j \norm{ [A_{1j}, \cdots, A_{nj}]^* }^{n-1}\\
&\le& \norm{u} (\sqrt{n})^{n-1} \left(\max_{i,j} |A_{ij}|\right)^{n-1}
\end{eqnarray*}
Therefore,
\begin{eqnarray*}
\norm{A^{-1}} &=& \sup_{\norm{u}=1} \norm{A^{-1} u} \text{\ by definition} \\
&\le& n^{\frac{n}{2}} \left(\max_{i,j} |A_{ij}|\right)^{n-1} \\
\end{eqnarray*}
\par
Combining the bounds for \norm{A} and \norm{A^{-1}}, one obtains:
\[
\cond{}(A) \le n^{\frac{n}{2}+1} \left(\max_{i,j} |A_{ij}|\right)^{n}
\punct{.}
\]
\subsection{Proof of Theorem~\ref{th2}}
In order to estimate $\cond{}(A)$, we write
\begin{eqnarray*}
\cond{}(A) &=& \sqrt{ \cond{} (A^* A) } \\
&\le & n^{\frac{n}{4}+\frac{1}{2}} H^{n} m^{n/2}
\end{eqnarray*}
\par
In order to bound $\cos \theta$, we use the assumption that
$b$ is not orthogonal to the image of $A$. Hence $\norm{A^* b} \ge 1$
and the `normal equation' $ A^* A x = A^* b $ implies:
\[
\norm{A^* A x} \ge 1
\punct{.}
\]
Therefore,
\[
\cos \theta = \frac{\norm{A^* A x}}{\norm{b}} \ge \frac{1}{\norm{b}}
\ge \frac{1}{H \sqrt{m}}
\]
and $\frac{1}{\cos \theta}$ and $\tan \theta$ are bounded above by
$H \sqrt{m}$. Putting all together,
\begin{eqnarray*}
\cond{LS} (A,b) &\le&
2 n ^{\frac{n}{4}+\frac{1}{2}} m^{\frac{n}{2}+\frac{1}{2}} H^{n+1}
+
n ^{\frac{n}{2}+1} m^{n+\frac{1}{2}} H^{2n+1}
\\
&\le&
3 n ^{\frac{n}{2}+1} m^{n+\frac{1}{2}} H^{2n+1}
\end{eqnarray*}
\subsection{Proof of Theorem~\ref{th3}}
\begin{lemma}\label{lemma1}
Let $B$ be an $n \times n$ matrix with integer
coefficients. Let $p(t) = \det (B - tI) = \sum p_i t^i$.
Then
\[
\max |p_i| \le \left( 2 \sqrt{n} \max_{i,j} |B_{ij}| \right)^n
\punct{.}
\]
\end{lemma}
\begin{proof}[Proof of Lemma~\ref{lemma1}]
\[
p_i = \sum_C \pm \det C
\]
where $C$ ranges over the $(n-i)\times(n-i)$ sub-matrices of $B$ of
the form $C_{kl}=B{s_ks_l}$ for some $1 \le s_1 < \cdots < s_{n-1} \le n$.
Hence
\begin{eqnarray*}
|p_i| &\le& \binomial{n}{i} \max_C |\det C| \\
&\le& \binomial{n}{i} \left( \sqrt{n-i} \max_{ij} |C_{ij}| \right)^{n-i}\\
&\le& \left(2 \sqrt{n} \max_{i,j} |B_{ij}| \right)^n
\end{eqnarray*}
\end{proof}
\begin{lemma}\label{lemma2}
Let $A$ be an $n \times n$ matrix with integer coefficients and
let $\lambda$ be an eigenvalue of $A$. Then
\[
H(\lambda) \le 2 \left(2 \sqrt{n} \max_{i,j}|A_{ij}| \right)^n
\punct{.}
\]
\end{lemma}
\begin{proof}[Proof of Lemma~\ref{lemma2}]
Apply Proposition~\ref{roots} to the polynomial $p(t)$
from Lemma~\ref{lemma1}.
\end{proof}
\begin{lemma}\label{lemma3}
Let $B$ be an $n \times n$ matrix with integer coefficients.
Let $q(t) = \det (B-tI + t e_n^* e_n) = \sum q_i t^i$. Then
\[
\max |q_i| \le 2 \left( 2 \sqrt{n} \max_{i,j} |B_{ij}| \right)^n
\punct{.}
\]
\end{lemma}
\begin{proof}[Proof of Lemma~\ref{lemma3}]
Let $p(t) = \det (B-tI)$ and let $r(t) = \det (\tilde B - t I)$
where $\tilde B$ is the $(n-1)\times(n-1)$ matrix obtained by deleting
the $n$-th row and the $n$-th column of $B$. Then, by multi-linearity
of the determinant,
\[
p(t) = q(t) \pm t\ r(t)
\punct{,}
\]
hence
\[
q(t) = p(t) \pm t\ r(t)
\punct{.}
\]
\par
Therefore,
\begin{eqnarray*}
\max |q_i| &\le& \max |p_i| + \max |r_i| \\
&\le&
\left(2 \sqrt{n} \max_{i,j} |B_{ij}| \right)^n
+
\left(2 \sqrt{n-1} \max_{i,j} |B_{ij}| \right)^{n-1}
\text{\ (Lemma~\ref{lemma1})}
\\
&\le&
2 \left(2 \sqrt{n} \max_{i,j} |B_{ij}| \right)^n
\end{eqnarray*}
\end{proof}
\begin{lemma}\label{lemma4}
Let $A$ be an $n\times n$ matrix with integer coefficients.
Let $\lambda$ be an isolated eigenvalue of $A$ and let
$x$ be an eigenvector associated to $\lambda$, $Ax = \lambda x$.
Then
\[
H(x_1 : \cdots : x_n) \le
n 2^n \left( 2 \sqrt {n} \max_{i,j}|A_{ij}| \right)^{n^2 -1}
\punct{.}
\]
\end{lemma}
\begin{proof}[Proof of Lemma~\ref{lemma4}]
Assume without loss of generality that the first $n-1$ lines
of $A-\lambda I$ are independent. Let $M_1$, \dots, $M_i$, \dots, $M_n$
be the sub-matrices obtained from $A-\lambda I$ by deleting the
last line and the $i$-th column. Then we can scale $x$ in such a
way that
\[
x_i = \pm \det M_i
\punct{.}
\]
\par
We have $M_n = B_n - \lambda I$. By reordering rows and columns, we
obtain for each $i<n$ that $M_i$ is of the form:
\[
M_i = B_i - \lambda I + \lambda e_{n-1}^* e_{n-1}
\]
where $B_i$ is the sub-matrix of $A$ obtained by deleting the last
line and the $i$-th column.
\par
Set $q^{(i)} (\lambda) = \det M_i = \sum q^{(i)}_j \lambda^j$. Now by
Lemma~\ref{lemma3},
\[
\max |q^{(i)}_j| \le 2 \left( 2 \sqrt{n-1} \max_{k,l} |A_{kl}| \right)^{n-1}
\punct{.}
\]
\medskip
\par
We consider now the morphism:
\function{q}{\mathbb P}{\mathbb P^n}
{(\lambda:1)} { (q^{(1)}(\lambda): \cdots : q^{(n)}(\lambda)) }
\par
Then $x=q(\lambda)$ and
\begin{eqnarray*}
H(x) &=& H(q(\lambda)) \\
&\le& n H(q) H(\lambda)^{n-1}
\\
&\le&
n 2 \left( 2 \sqrt{n-1} \max_{i,j} |A_{ij}|\right)^{n-1}
2^{n-1} \left( 2 \sqrt{n} \max_{i,j} |A_{ij}|\right)^{n(n-1)}
\\
&\le&
n 2^{n} \left( 2 \sqrt{n} \max_{i,j} |A_{ij}|\right)^{n^2 -1}
\end{eqnarray*}
the first inequality because of Proposition~\ref{morphisms},
and the second because of Lemma~\ref{lemma2}.
\end{proof}
\medskip
\par
\begin{proof}[End of the Proof of Theorem~\ref{th3}]
Proposition~\ref{polynomials} implies
\[
H(y^* x) \le n H(x) H(y)
\punct{.}
\]
\par
We claim that $\deg [ \mathbb Q[y^* x] : \mathbb Q] \le n$. Indeed,
$x$ and $y$ can be obtained by solving systems of linear equations
with coefficients in $\mathbb Q[\lambda]$, thus
$x_i, y_i \in \mathbb Q[\lambda]$. Also, $\overline{y_i}
\in \mathbb Q[\bar{\lambda}] = \mathbb Q[\lambda]$ so
$\deg [ \mathbb Q[y^* x] : \mathbb Q] \le n$ as claimed.
\par
By hypothesis $y^* x \ne 0$. Hence, by Proposition~\ref{bound},
\begin{eqnarray*}
|y^* x| &\ge& \left(n H(x) H(y) \right)^{-n} \\
&\ge& n^{-3n} 2^{-2n} \left( 2 \sqrt{n} \max_{i,j} |A_{ij}| \right)^{-2n^3+2n}
\end{eqnarray*}
\end{proof}
\subsection{Proof of Theorem~\ref{th4}}
\begin{proof}
According to Proposition~\ref{roots},
\[
H(\zeta) \le 2 \max |f_i|
\punct{.}
\]
\par
Also,
\[
H(f'_i) \le d H(f) = d \max |f_i|
\]
and according to Proposition~\ref{polynomials}
\begin{eqnarray*}
H(f'(\zeta)) &\le& d H(f') H(\zeta)^{d-1} \\
&\le& d^2 2^{d-1} H(f)^d \\
\end{eqnarray*}
\par
and hence $|f'(\zeta)| \ge d^{-2d} H(f)^{-d^2}
2^{-d(d-1)}$. On the other hand,
\begin{eqnarray*}
\left(
\sum_{i=0}^d |\zeta|^{2i}
\right)^\frac{1}{2}
&\le&
\sqrt{d+1}
\max(1,|\zeta|^d)
\\
&\le&
\sqrt{d+1}
H(\zeta)^{d^2}
\\
&\le&
\sqrt{d+1}
2^{d^2} H(f)^{d^2}
\end{eqnarray*}
\par
Hence
\[
\mu(f,\zeta) \le 2^{2d^2 -2} H(f)^{2 d^2} d^{2d}
\punct{.}
\]
\end{proof}
\subsection{Proof of Theorem~\ref{th5}}
\begin{lemma}\label{lemma5}
Let $A$ be an $n \times n$ invertible matrix with algebraic
coefficients. Then
\[
H(A^{-1}) \le n H(A)^n
\punct{.}
\]
\end{lemma}
\begin{proof}[Proof of Lemma~\ref{lemma5}]
Let $A(i,j)$ be the sub-matrix of $A$ obtained by deleting
the $i$-th row and the $j$-th column. By Cramer's rule,
$\left(A^{-1}\right)_{ji} = \frac{\det A(i,j)}{\det A}$. Therefore we
should define the degree $n$ morphism:
\function{\varphi}{\mathbb P^{n^2}}{\mathbb P^{n^2}}
{\begin{array}{l}\left(A_{11}: A_{12}: \cdots \right.\\
\hspace{1cm} \left. \cdots : A_{nn} : 1\right)\end{array}}
{\begin{array}{l}\left( \det A(1,1): \det A(1,2): \cdots \right. \\
\hspace{1cm} \left. \cdots : \det A(n,n) : \det A\right)\end{array}}
\par
Then by Proposition~\ref{morphisms},
\begin{eqnarray*}
H( \varphi(A) ) &\le& n! H(A)^n H(\varphi) \\
&\le& n! H(A)^n
\end{eqnarray*}
\end{proof}
Let us fix the notations
\[
M
=
\left[ \begin{matrix}
\norm{\zeta}^{1-d_1}\\
& \ddots\\
& & \norm{\zeta}^{1-d_n}\\
& & & \norm{\zeta}^{-1}\\
\end{matrix} \right]
\left[
\begin{matrix}
\mathrm{D}F(\zeta) \\
\zeta^*
\end{matrix}
\right]
\]
and
\[
C =
\mathrm{D}F(\zeta)_{| T_\zeta} ^{-1}
\left[ \begin{matrix}
\norm{\zeta}^{d_1-1}\\
& \ddots\\
& & \norm{\zeta}^{d_n-1}\\
\end{matrix} \right]
\punct{.}
\]
\par
Let $\zeta \in \mathbb C^{n+1}$ be a fixed
representative for a root of $F$.
Any $u \in T_{\zeta} \mathbb P^n$ can be written
as a vector in $\mathbb C^{n+1}$, orthogonal to
$\zeta$. Computing $u = Cv$ is the same as
solving $Mu = \left[ \begin{matrix}v\\0 \end{matrix}\right]$.
The operator $C$
is the same as $(M^{-1})_{|x_{n+1}=0}$. Therefore,
\[
\norm{C} \le \norm{M^{-1}}
\punct{.}
\]
\begin{lemma} \label{lemma6}
In the conditions of Theorem~\ref{th5},
\[
H \left( M
\right)
\le S \sqrt{n+1}^{D-1} H(\zeta)^{2 D - 2} D H(f)
\punct{.}
\]
\end{lemma}
\begin{proof}[Proof of Lemma~\ref{lemma6}]
We apply Proposition~\ref{polynomials} to the
system:
\[
\zeta, N \mapsto
\left( \cdots, N^{d_i -1} \frac{\partial F_i}{\partial x_j} (\zeta),
\cdots, \bar \zeta_j, \cdots \right)
\]
with $N = \norm{\zeta}^{-1}$ to obtain
\[
H(M) \le
S H(\zeta)^{D-1} H(N)^{D-1} H(\mathrm{D}F)
\punct{.}
\]
\par
We can bound $H(\mathrm{D}F) \le D H(F)$ and $H(N) = H( \norm{\zeta})
= \sqrt{H(\sum |\zeta_i|^2)}$. We can apply Proposition~\ref{morphisms}
to the map
\function{\varphi}{\mathbb C^{n+1}}{\mathbb C}{\zeta, \bar \zeta}
{\sum \zeta_i, \bar \zeta_i}
to get $H(N^2) \le (n+1) H(\zeta) H(\bar \zeta)$. Proposition~\ref{conjugate}
implies $H(\zeta) = H(\bar \zeta)$, hence:
\[
H(N^2) \le (n+1) H(\zeta)^2
\]
and by Proposition~\ref{square},
\[
H(N) \le \sqrt{n+1} H(\zeta)
\punct{.}
\]
Thus, we can estimate that
\[
H(M) \le S \sqrt{n+1}^{D-1} H(\zeta)^{2D-2} D H(f)
\punct{.}
\]
\end{proof}
\begin{proof}[End of the Proof of Theorem~\ref{th5}]
By definition of the norm, $\norm{f} \le S H(f)$. By
Lemma~\ref{lemma5} and Lemma~\ref{lemma6},
we have:
\begin{eqnarray*}
H(C)
&\le&
(n+1) H(M) ^n \\
&\le&
(n+1) S^n \sqrt{n+1}^{nD-n} H(\zeta)^{2 nD - 2n} D^n H(f)^n
\\
\end{eqnarray*}
\par
Knowing that $\deg \left[\mathbb Q[\zeta] : \mathbb Q \right] \le D^n$,
we can use Proposition~\ref{bound} to deduce that
\begin{eqnarray*}
\norm{C} &\le& (n+1) H(C) \\
&\le& (n+1)
\left(
(n+1) S^n \sqrt{n+1}^{nD-n} H(\zeta)^{2 nD - 2n} D^n H(f)^n
\right)^{D^n}
\end{eqnarray*}
According to Proposition~\ref{krickpardo},
\[
H(\zeta) \le n H(f)^{D^{c'n}}
\]
\par
where $c'$ is a universal constant. Thus,
\begin{eqnarray*}
\mu(f,\zeta)
&\le&
S H(f)
(n+1)\left( (n+1) S^n \sqrt{n+1}^{nD-n}
n^{2 nD -2n} \right. \\
& & \hspace{4cm} \left. H(f)^{D^{c'n} (2 nD - 2n)} D^n H(f)^n\right)^{D^n} \\
&\le&
((n+1) S H(f))^{D^{cn}}
\end{eqnarray*}
\par
where $c$ is a universal constant.
\end{proof}
\subsection{Proof of Theorem~\ref{th6}}
Lemma~\ref{lemma2} implies:
\[
H(\lambda) \le 2 \left( 2 \sqrt{n} \max_{i,j} |A_{ij} \right) ^n
\punct{.}
\]
Hence (Proposition~\ref{polynomials}),
\[
H\left( \frac{\lambda_j}{\lambda_i} - 1\right)
\le
8 (4n)^n \left( \max_{i,j} |A_{ij}| \right) ^{2n}
\punct{.}
\]
\par
Thus, by Proposition~\ref{bound},
\[
\left| \frac{\lambda_j}{\lambda_i} - 1\right|
\le
8^n (4n)^{n^2} \left( \max_{i,j} |A_{ij}| \right) ^{2n^2}
\punct{.}
\]
\subsection{Proof of Theorem~\ref{th7}}
According to Proposition~\ref{roots},
\begin{eqnarray*}
H(\zeta_i) &\le& 2 H(f) \\
H(\zeta_j) &\le& 2 H(f)
\end{eqnarray*}
\par
Moreover, $H(|\zeta_i|) \le H(\zeta_i)$ because $|\zeta_i|^2 = \zeta_i
\bar{\zeta_i}$ and $H(\zeta_i) = H(\bar{\zeta_i})$
(Propositions~\ref{conjugate},~\ref{polynomials} and~\ref{square}). Thus,
\begin{eqnarray*}
H \left( \frac{ |\zeta_j| } {|\zeta_i|} -1 \right)
&\le& 2 H(|\zeta_i|) H(|\zeta_j|) \\
&\le& 8 H(f)^2
\end{eqnarray*}
\par
It follows that
\[
\relgap(f)^{-1} \ge \left( 8 H(f) \right) ^{-2\deg [\mathbb Q[|\zeta_i|,
|\zeta_j|]: \mathbb Q]} \ge \left( 8 H(f) \right) ^{-2d}
\punct{.}
\]
\section{Further comments}
As mentioned before, a reasonable definition for the
`real complexity' input size is the number
of coefficients of a given problem instance, times the logarithm
of its condition number.
\par
Theorems~\ref{th1} to~\ref{th4} show that the `real complexity' input
size is no worse than a polynomial of the `classical complexity' input
size, for problem instances with integer coefficients. Theorem~\ref{th5}
also, if one considers $D^n$ as part of the input size. It may be
possible to replace $D^n$ by the B\'ezout number $\prod d_i$,
that is the number of solutions of a generic system
of polynomials.
\par
Since the `real complexity' of the problems considered can be bound by
common numerical analysis techniques, those Theorems provide a scheme
to convert `real complexity' bounds into `classical complexity' bounds.
\medskip
\par
The same idea is behind Theorems ~\ref{th6} and ~\ref{th7}. In the case
of the iterative algorithms considered, the number of iterations for
obtaining a certain approximation can also be bounded in terms of
a `condition number'. In the case of problem instances with integer
coefficients, the `condition number' is also polynomially bounded in
terms of the input size.
\medskip
\par
Those Theorems have many features in common, and this is not a
coincidence. A more general approach is to interpret the condition number
as the inverse of the distance to the degenerate locus. This can
be bounded in terms of the height of the problem instance, and in
terms of the degenerate locus (degree, dimension, height). However,
bound obtained this way will be no sharper and possibly worse than
the direct bounds obtained by using the exact expression for the
condition number.
\medskip
\par
This paper was written while the author was visiting MSRI at Berkeley.
He wishes to thank MSRI for its generous support.
Thanks to Bernard Deconinck, Jennifer Roveno, Paul Gross, Raquel,
and very special thanks to Paulo Ney de Souza and family.
\bibliographystyle{plain}
|
\section{Krichever map}
Consider the following geometric datas:
\begin{enumerate}
\item $C$ is a reduced irreducible complete algebraic curve defined over
a field $k$.
\item $p$ is a smooth $k$-rational point.
\item ${\cal F}$ is a torsion free coherent sheaf of ${\cal O}_C$-modules on $C$ of rank
$2$.
\item $t_p$ is a local parameter of point $p$, i.~e.
$\hat{{\cal O}_p} = k[[t_p]]$, where $\hat{{\cal O}_p}$ is the completion of local
ring ${\cal O}_p$ along the maximal ideal.
\item $e_p$ ia a basis of rank $2$ free module $\hat{{\cal F}_p}$
over the ring $\hat{{\cal O}_p}$, where $\hat{{\cal F}_p} \stackrel{\rm def}{=} {\cal F} \otimes_{{\cal O}_C}
\hat{{\cal O}_p} $.
\end{enumerate}
In the sequel we will {\em call} such collection $(C,p,{\cal F},t_p,e_p)$
a {\em quintet}. If the Euler characteristic $\chi({\cal F})= \mu$,
then we will say, that the quintet have an index $\mu$.
On the other hand, consider $V= k((z)) \oplus k((z))$,
$V_0 = k[[z]] \oplus k[[z]]$. Then {\em define} the infinite Grassmanian
$Gr^{\mu}(V)$ as the set of $k$-vector subspaces $W$ of $V$,
which are Fredholm of index $\mu$, i.~e.,
$\mathop {\rm dim}_k V / (V_0 + W ) < \infty $, $\mathop {\rm dim}_k W \cap V_0 < \infty$, and \\
$\mathop {\rm dim}_k W \cap V_0 - \mathop {\rm dim}_k V / (V_0 + W ) = \mu $.
There exists {\em the Krichever map} $K$ from the set of quintets of index
$\mu$ to $Gr^{\mu}(V)$, which can be shortly defined as the following
chain of
evident maps:
$$
(C,p,{\cal F},t_p, e_p) \to
H^0(C \backslash p, {\cal F})
\hookrightarrow
H^0 ( \mathop {\rm Spec} {\cal O}_p
\backslash p,{\cal F})
\hookrightarrow
H^0 ( \mathop {\rm Spec} \hat{{\cal O}_p}
\backslash p,{\cal F}) = \qquad \qquad
$$
\begin{equation} \label{chain}
\qquad \qquad \qquad \qquad \qquad
=
\hat{{\cal F}_p} \otimes_{\hat{{\cal O}_p}} \hat{K_p}
\stackrel{e_p}{\to}
(\hat{{\cal O}_p} \oplus \hat{{\cal O}_p})
\otimes_{\hat{{\cal O}_p}} \hat{K_p}
\stackrel{t_p}{\to} k((z)) \oplus k((z)) \; \mbox{,}
\end{equation}
where $\hat{K_p}$ is the fraction field of $\hat{{\cal O}_p}$.
The Krichever map $K$ and its basic properties is considered in details by
many authors. See, for example, \cite{M} for an algebraic description.
(But note that
the definition in~\cite{M} is slightly different from our definition.)
And see~\cite{PW} for an analytic description.
For any $W \in Gr^{\mu}(V) $ define the ring
$$A_W \stackrel{\rm def}{=} \{f \in k((z)) : fW \subset W \} $$
\begin{lemma}[see~\cite{M}] \label{odin}
For any $W \in Gr^{\mu}(V) $
and any subring $A \subset A_W$ such that $k \subset A$ we have
1) $A \cap k[[z]] = k$ and
2) if $A \ne k$, then the ring $A$ has dimension $1$ over $k$.
\end{lemma}
{\bf Proof\ }.
The first statement follows from $\mathop {\rm dim}_k W \cap V_0 < \infty$.
For the proof of the second statement consider the following subgroup of
${\mbox{\dbl Z}}$:
$$
\left\{ n \in {\mbox{\dbl Z}} \; \, : \; \, n= \nu(a) \, , \; a \in \mathop {\rm Frac} A
\right\} \mbox{,}
$$
where $\nu$ is the discrete valuation of $k((z))$
and $\mathop {\rm Frac} A$ is the fraction field of the ring $A$.
Then this subgroup is $r {\mbox{\dbl Z}}$ for some integer $r$.
Therefore
1) there exist $f, g \in A$ such that $r = \nu(f) - \nu(g)$
2) for any integer $n$
\begin{equation} \label{zvezda}
\mathop {\rm dim}\nolimits_k A \cap z^{-nr}k[[z]] / A \cap z^{(-n+1)r}k[[z]] \le 1
\end{equation}
Now from~(\ref{zvezda}), the first statement of this lemma,
and $r = G.C.D. (\nu(f), \nu(g))$
we have
$$
\mathop {\rm dim}\nolimits_k A / k[f,g] < \infty \quad \mbox{.}
$$
Thus the dimension of the ring $A$ over $k$ is at most two.
Now let us assume that $f$ and $g$ are algebraically independent
elements over $k$. Then for all pairs of integers $n_1 > 0$
and $n_2 > 0$ elements $f^{n_1} g^{n_2}$
are included in one basis of vector space $A$ over $k$.
But it contradicts to~(\ref{zvezda}) and this contradiction
concludes the proof of lemma. \vspace{5pt}
Notice also the following well-known statement.
\begin{lemma} \label{Shurik}
$W \in Gr^{\mu}(V) $
is in the image of the Krichever map
if and only if $\mathop {\rm rank} A_W =1$, i.e.,
there exist two elements $f, g \in A_W$ such that $\nu(f)$
and $\nu(g)$ are relatively prime.
\end{lemma}
{\bf Proof\ }.
Note that the condition $\mathop {\rm rank} A_W = 1$ is equivalent to
\begin{equation} \label{zvzvezda}
\mathop {\rm dim}\nolimits_k k((z)) / (A_W + k[[z]]) < \infty \quad \mbox{.}
\end{equation}
Let $W= K(C,p,{\cal F},t_p,e_p)$. By $A_C$ denote the image of the ring
$H^0 (C \backslash p, {\cal O}_C)$ in $k((z))$:
$$
H^0 (C \backslash p, {\cal O}_C)
\hookrightarrow
\hat{K_p}
\stackrel{t_p}{\to}
k((z)) \quad \mbox{.}
$$
From the Riemann-Roch theorem we have
$$
\mathop {\rm dim}\nolimits_k k((z)) / (A_C + k[[z]]) < \infty \quad \mbox{.}
$$
From $A_C \subset A_W$ it follows~(\ref{zvzvezda}).
Now let $\mathop {\rm rank} A_W = 1$.
Fix any subring $A \subset A_W$ such that $k \subset A$
and $\mathop {\rm dim}\nolimits_k A_W / A < \infty$.
By lemma~\ref{odin} the ring $A$ has dimension $1$ over $k$.
Then from graded $k$-algebra
$$
A_* \stackrel{\rm def}{=} \bigoplus_{i=0}^{\infty} A \cap z^{-i}k[[z]]
$$
we construct the complete irreducible reduced curve
$C = \mathop {\rm Proj} A_*$.
And from~(\ref{zvzvezda}) we see
that the graded $A_*$-module
$$
W_* \stackrel{\rm def}{=}
\bigoplus_{i=-\infty}^{\infty} W \cap z^{-i}V_0
$$
determines the rank 2 torsion free coherent sheaf $W_* \,\tilde{}$ on $C$.
Moreover, we have $C = \mathop {\rm Spec} A \cup p$,
where $p$ is a smooth $k$-rational point, $\hat{{\cal O}_p}= k[[z]]$,
${\cal F} \mid_{C \backslash p} = W \, \tilde{}$,
$\hat{{\cal F}_p} = V_0 $,
$t_p = z $,
$e_p = {(1,0), (0,1) \in V}$.
And
$$
\bigoplus_{i=0}^{\infty} H^0(C, {\cal O}(ip)) \simeq A_*
\qquad
\mbox{,}
\qquad
\bigoplus_{i=-\infty}^{\infty} H^0(C, {\cal F}(ip)) \simeq W_* \quad \mbox{,
and}
$$
$$
H^0(C, {\cal F}) \simeq W \cap V_0
\qquad
\mbox{,}
\qquad
H^1(C, {\cal F}) \simeq V / (W + V_0) \quad \mbox{.}
$$
This lemma is proved. \vspace{7pt}
\noindent {\bf Remark.} If we fix a triplet $(C, p, t_p)$
and identify $(C,p,{\cal F}, t_p, e_p)$
with $(C, p, {\cal F}', t_p, e_p')$
for every sheaf isomorphism $\alpha : {\cal F} \to {\cal F}'$
such that $\alpha (e_p) = e_p'$;
then the Krichever map is an injective map.
In this case $W \in Gr^{\mu}(V)$ is in the image of such map if and only
if $A_C W \subset W$. \\[4pt]
Now let $W =
K(C,p,{\cal F},t_p,e_p) \in
Gr^{\mu}(V) $.
Then with every subsheaf ${\cal G}$ of ${\cal F}$
we can associate the $k((z))$-vector subspace
$L_{{\cal G}} \subset V$ such that $L_{{\cal G}} \cap W \ne 0$ :
$$
H^0(C \backslash p, {\cal G}) \hookrightarrow H^0(C \backslash p, {\cal F})
= W \subset Gr^{\mu}(V)
$$
\begin{equation} \label{zv}
\qquad
{\cal G} \longmapsto H^0(C \backslash p, {\cal G}) \cdot k((z)) \subset{V}
\end{equation}
We have the following proposition.
\begin{prop} \label{p1}
Let $W= K(C,p,{\cal F},t_p,e_p) $,
$
R({\cal F})$ be the set of all rank 1 coherent subsheaves $ {\cal G} \subset {\cal F}$
such that the sheaf $ {\cal F}/{\cal G} $ is a free torsion coherent
sheaf,
$
R(W) $ be the set of all 1-dimensional $k((z))$-vector subspaces $L \subset
V$
such that $L \cap W \ne 0$.
Then map~(\ref{zv}) is an one-to-one correspondence between
$R({\cal F})$ and $R(W)$.\\
Moreover, if ${\cal G} \in R({\cal F}) \mapsto L_{{\cal G}} \in R(W)$,
then
\begin{equation} \label{ech}
\chi(C, {\cal G}) = \mathop {\rm dim}\nolimits_k W \cap L_{{\cal G}} \cap V_0
- \mathop {\rm dim}\nolimits_k L_{{\cal G}}/ (W \cap L_{{\cal G}} + L_{{\cal G}}
\cap V_0 )
\quad \mbox{.}
\end{equation}
\end{prop}
{\bf Proof\ }.
It is not difficult to see that
the set $R({\cal F})$ is in an one-to-one correspondence with $1$-dimensional
$k(\eta)$-vector subspaces of ${\cal F}_{\eta} \stackrel{\rm def}{=} H^0(\mathop {\rm Spec} k(\eta),{\cal F}) $,
where $\eta \hookrightarrow C$ is the general point.
We have the canonical imbedding of ${\cal F}_{\eta}$
into $V$ as the third term $H^0 (\mathop {\rm Spec} {\cal O}_p \backslash p, {\cal F}) = {\cal F}_{\eta}$
of chain~(\ref{chain}). And this imbedding gives us a one-to-one
correspondence between $R({\cal F})$ and $R (W)$.
Moreover, we have the following explicit construction. Let $L \in R(W)$.
Then the graded $(A_C)_*$-module $(L \cap W)_* \stackrel{\rm def}{=} \bigoplus\limits_{i =
-\infty}^{\infty} (L \cap W) \cap z^{-i} V_0$
determines the rank $1$
torsion free sheaf ${\cal G}_L = (L \cap W)_* \, \tilde{}$
on $C = \mathop {\rm Proj} (A_C)_*$. Then $L \to {\cal G}_L$
inverts
map~(\ref{zv})
and
$$H^0(C, {\cal G}_L) \simeq W \cap L \cap V_0 \quad \mbox{,}$$
$$H^1 (C, {\cal G}_L) \simeq L / (W \cap L + L \cap V_0) \quad \mbox{.}$$
\section{Determinant bundle}
As was explicitly described in~\cite{AMP},
$ Gr^{\mu}(V) $ admits a structure of an algebraic scheme,
which represents a functor, which generalizes the usual finite-dimensional
grassmanian. There exists an open covering of $ Gr^{\mu}(V) $
by means of $ Gr^{\mu}_A(V) $, where $A$ is a commensurable with $V_0$
$k$-vector subspaces in $V$, i.~e., $\mathop {\rm dim}_k (A+V_0)/ (A \cap V_0) < \infty
$
and
$$ Gr^{\mu}_A(V) \stackrel{\rm def}{=} \{ W \in \ Gr^{\mu}(V) :
V = A \oplus W \}
$$
is an infinite-dimensional affine space which is isomorphic
to the spectrum of a polynomial ring of infinitely many variables.
One can define a linear bundle (determinant bundle) $Det$
on $Gr^{\mu}(V)$ such that for any $W \in Gr^{\mu}(V)$
$$
Det_W = \bigwedge^{max} (W \cap V_0) \otimes \bigwedge^{max} (V / W + V_0)^*
\qquad \mbox{.}
$$
Now let us describe the determinant bundle $Det$ more explicitly
("in coordinats"). For this goal we
identify the $k$-vector spaces $V$ and $k((t))$
by means of the following continuous isomorphism:
$$
V \longrightarrow k((t))
$$
\begin{equation} \label{diez}
(\sum a_i z^i ) \oplus (\sum b_i z^i) \longmapsto \sum_i (a_i t^{2i}
+ b_i t^{2i+1})
\end{equation}
Note that $V_0 \to k[[t]]$.
Let $S_{\mu}$ be the set of sequences $\{ s_{-\mu +1}, s_{-\mu +2},
\ldots \}$ of integers such that:
1) these sequences are
strictly decreasing;
2) $s_k = -k$ for $k \gg 0$.
Note that for every $\mu \in {\mbox{\dbl Z}} $ the set $S_{\mu}$
is in an one-to-one correspondence with the Young diagramms.
For every $S = \{ s_{-\mu +1}, s_{-\mu +2},
\ldots \} \in S_{\mu}$ let
$$H_S \in Gr^{\mu}(V) $$
be the $k$-vector space generated by $\{ t^{s_k} \}$,
$$A_S \subset k((t))$$
be the smallest closed $k$-vector space generated by
$\{ t^q \mid q \in {\mbox{\dbl Z}} \backslash S \} $.
Then
\begin{equation} \label{dbldiez}
Gr^{\mu}(V) = \bigcup_{S \in S_{\mu}} Gr_{A_S}^{\mu}(V) \quad \mbox{.}
\end{equation}
For any $S \in S_{\mu}$ let $p_S : k((t)) \to H_S$
be the projection from the decomposition $ H_S \oplus A_S = k((t))$.
By $0(\mu) \in S_{\mu}$ denote $\{ \mu-1, \: \mu -2, \: \mu -3, \: \ldots \}$.
We are now going to describe the determinant bundle $Det$
in other words (see~\cite[~\S 7.7]{PS}). Given $W \in Gr^{\mu}(V)$
let us define an {\em admissible} isomorphism to be an isomorphism
$$
w: H_{0(\mu)} \longrightarrow W \qquad \qquad (w(t^{\mu-1}) \in W,
\;
w(t^{\mu-2}) \in W, \ldots )
$$
such that
$$p_{0(\mu)} \cdot w - Id \quad : \quad H_{0(\mu)} \to H_{0(\mu)}$$
has finite rank. (Note that for any $W \in Gr^{\mu}(V)$ we can always
find the corresponding admissible isomorphism.)
Then the determinant bundle $Det$ consists of
$$ \{ [\lambda, w] : \lambda \in k,\; w \; \mbox{is an admissible isomorphism}
\} \quad \mbox{,}$$
which is identified with respect to the following equivalence relation:
$$
[\lambda, w] \sim [\lambda \cdot \det (w'^{-1} \cdot w), w'] \quad \mbox{,}
$$
where $w'$ and $w$ are two admissible isomorphisms for $W \in Gr^{\mu}(V)$.
(Note that
$$w'^{-1} \cdot w - Id \quad : \quad H_{0(\mu)} \to H_{0(\mu)} $$
has finite rank as an endomorphism of $H_{0(\mu)}$, and
hence $\det (w'^{-1} \cdot w)$ is well defined with respect to the basis:
$\{ t^{\mu -1}, t^{\mu-2}, t^{\mu-3}, \ldots \}$.)
Let $Det^*$ be the dual vector bundle to $Det$. For every
$S \in S_{\mu}$ we can construct the global section
$$
\pi_S \in H^0(Gr^{\mu}(V), Det^*) \qquad : \quad Det \longrightarrow k
$$
$$
[\lambda,w] \stackrel{\pi_S}{\longrightarrow} \lambda \det(p_S \cdot w)
\in k \quad \mbox{,}
$$
where $ p_S \cdot w - Id \; : \; H_0 \to H_S$ has finite rank.
Therefore we can compute $\det (p_S \cdot w)$ with respect to
basises
$$
\{t^{\mu -1}, \; t^{\mu -2}, \; \ldots \} \in H_{0(\mu)}
\qquad \mbox{and} \qquad \{t^{s_{-\mu +1}},\; t^{s_{-\mu+2}},\; \ldots \}
\in H_S \quad \mbox{.}
$$
And as proven in~\cite{AMP} and~\cite{PS},
we have a closed imbedding ("Pl\"ucker imbedding"):
$$
Gr^{\mu}(V) \hookrightarrow {\mbox{\dbl P}} (\Pi (S_{\mu})^*) \stackrel{\rm def}{=} \mathop {\rm Proj} (Sym(\Pi
(S_{\mu})))
$$
$$
W \longmapsto \left\{ \pi_S \to \pi_S (w) \stackrel{\rm def}{=} \det (p_S \cdot w) \right\}
\quad \mbox{,}
$$
where $\Pi (S_{\mu}) \subset H^0(Gr^{\mu}(V), Det^*)$
is the $k$-vector subspace generated by the global sections
$\{ \pi_S \; : \; S \in S_{\mu} \}$,
and with every $k$-point $W \in Gr^{\mu}(V) $
we associate the $1$-dimensional $k$-vector subspace
$k \cdot \left\{ \pi_S \mapsto \pi_S(w) \, : \, S \in S_{\mu} \right\}
\subset \Pi (S_{\mu})^* $,
which is the same
for all admissible isomorphisms $w$ of $W$.
Note also that $\pi_S (w) \ne 0 $ if and only if
$W \in Gr^{\mu}_{A_S}(V)$, i.~e. $W \oplus A_S = V$. \\[4pt]
Consider the group
$$GL (V,V_0) \stackrel{\rm def}{=} \{ g \in Aut_k(V) \;:\; \mathop {\rm dim}\nolimits_k (gV_0 + V_0) /
(gV_0 \cap V_0) < \infty \} \quad \mbox{.}$$
Note that "in coordinats" $GL (V,V_0) $
corresponds to the group of invertible elements
of the algebra of matrices $(A_{i j})_{ i,j \in {\mbox{\sdbl Z}}}$
such that for every integer $l$ the number of non-zero $A_{i j}$
with $j \ge l$, $i \le l$ is finite.
And the action of $(A_{i j})_{i,j \in {\mbox{\sdbl Z}}}$ on $k((t))$
is $E_{i j} (t^j) = t^i$. (Here $E_{i j}$ is the matrix with a $1$
on the $(i,j)$-th entry and zeros elsewhere.)
Define the subgroup
$$GL_0 (V, V_0) \stackrel{\rm def}{=}
\{ g \in GL(V,V_0) \; : \;
\mathop {\rm dim}\nolimits_k gV_0 / (gV_0 \cap V_0) = \mathop {\rm dim}\nolimits_k V_0 / (gV_0 \cap V_0)
\} \quad \mbox{.}
$$
Then we have the obvious action of $GL_0 (V, V_0)$
on $Gr^{\mu}(V)$.
Let the group $\hat{GL_0} (V, V_0)$
be the central extension of the group $GL_0 (V, V_0)$
by means of $k^*$, which is the group of all
automorphisms of linear bundle $Det$
on $Gr^0(V)$
which cover actions of elements of group $GL_0 (V, V_0)$
on $Gr^0(V)$.
There exists an explicit description of
$\hat{GL_0} (V, V_0)$
and its action on $Det \mid_{Gr^{\mu}(V)}$ (see~\cite{PS}).
Let $H$ be the group
$$
H = \left\{ (g, E) \in GL_0 (V, V_0) \times GL_k (t^{-1}k[t^{-1}]) \quad :
\quad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad
\right. $$
$$
\left.
\qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad
g_{--} \, - E \; : \; t^{-1}k[t^{-1}] \to t^{-1}k[t^{-1}]
\quad \mbox{has finite rank}
\right\} \mbox{,}
$$
where $g_{--} \stackrel{\rm def}{=} (g)_{i,j \le -1}$. Let $N \subset H$
be the normal subgroup defined as $N = \{ (1,E) \in H \: : \:
\det E = 1 \}$.
Then $\hat{GL_0} (V, V_0) = H/N$,
and the projection on the first factor gives us the map onto
$GL_0 (V, V_0) $.
Now describe the action of $\hat{GL_0} (V, V_0) $
on $Det \mid_{Gr^{\mu}(V)}$. \\[6pt]
\underline{$\mu = 0$} : for $(g,E) \in H$
and $[\lambda, w] \in Det \mid_{Gr^0 (V)}$
\begin{equation} \label{f}
(g,E) [\lambda, w] \stackrel{\rm def}{=} [\lambda, gwE^{-1}]
\end{equation}
This expression gives us the correct action of
$\hat{GL_0} (V, V_0) $
on $Det \mid_{Gr^0 (V)}$. \\[6pt]
\underline{$\mu \ne 0$}.
Define the action $a \in \hat{GL_0} (V, V_0) $
on $Det \mid_{Gr^{\mu} (V)}$
as action of
\begin{equation} \label{ff}
\sigma^{- \mu} \cdot \tilde{\sigma}^{\mu} (a) \cdot \sigma^{\mu}
\quad \mbox{,}
\end{equation}
where $\sigma: Det \mid_{Gr^{\mu} (V)} \to Det \mid_{Gr^{\mu-1} (V)} $
is defined by the formula $\sigma \cdot [\lambda, w] = [\lambda,
t^{-1} \circ w]$; $t^{-1} \circ w \: : \: H_{0(\mu-1)} \to t^{-1}W$,
$t^{-1} \circ w (x) = t^{-1} w (tx)$ is admissible, $x \in H_{0(\mu-1)}$;
and the automorphism $\tilde{\sigma} \: : \:
\hat{GL_0} (V, V_0) \to \hat{GL_0} (V, V_0) $
is induced by the following endomorphism of $H$:
$$
(g, E) \longmapsto (t^{-1} g t, E_{\sigma}) \quad \mbox{, where}
$$
$$
E_{\sigma} \mid_{t^{-2}k[t^{-1}]} = t^{-1} E t
\qquad
\mbox{and}
\qquad
E_{\sigma}(t^{-1})=1 \quad \mbox{.}
$$ \\[1pt]
Now consider the group $\Gamma = GL(2, k[[z]])$,
which is the subgroup of $GL_0 (V, V_0)$.
\begin{lemma} \label{zvezd}
The central extension $\hat{GL_0} (V, V_0)$
splits over $\Gamma$.
\end{lemma}
{\bf Proof\ }.
Define a group homomorphism
\begin{equation} \label{fff}
h \; : \; \Gamma \longrightarrow H \quad \mbox{by} \quad h(\gamma)
= (\gamma ,\gamma_{--}) \quad \mbox{for} \quad \gamma \in \Gamma \;
\mbox{.}
\end{equation}
(Observe that $\Gamma$ keeps $V_0$ stable and hence
$\gamma_{--} \in GL_{k} (k[t^{-1}]))$.
Then the group homomorphism
$$
\Gamma \stackrel{h}{\longrightarrow} H \to H/N = \hat{GL_0} (V, V_0)
$$
splits the central extension over $\Gamma$.
\\[10pt]
In $V = k((z)) \oplus k((z))$ consider two lines:
$$
l_1= k((z)) \oplus 0 \qquad \mbox{and} \qquad l_2= 0 \oplus k((z)) \quad
\mbox{,}
$$
which correspond to $k((t^2))$ and $t k((t^2))$ in $k((t))$.
For any $S \in S_{\mu}$ let
$$
n_i(S) \stackrel{\rm def}{=} \mathop {\rm dim}\nolimits_k H_S \cap V_0 \cap l_i -
\mathop {\rm dim}\nolimits_k l_i / (H_S \cap l_i
+ l_i \cap V_0) \; \mbox{,} \quad i=1,2
$$
Note that
\begin{equation} \label{twodiez}
n_1(S) + n_2(S) = \mu \quad \mbox{.}
\end{equation}
Let $
\left(
\begin{array}{cc}
\alpha^{a_1} & 0 \\
0 & \nu^{a_2}
\end{array}
\right)
\in \Gamma$,
$\alpha, \: \nu \in k^*$ $a_i \in {\mbox{\dbl Z}} $.
By lemma~\ref{zvezd} we can consider an action of
$
\left(
\begin{array}{cc}
\alpha^{a_1} & 0 \\
0 & \nu^{a_2}
\end{array}
\right)
$ on the determinant bundle $Det$.
\begin{prop} \label{dddd}
Let $S \in S_{\mu}$, let $\pi_S$ be the corresponding global section of
$Det^{*} \mid_{Gr^{\mu}(V)}$.
Let
$
\left(
\begin{array}{cc}
\alpha^{a_1} & 0 \\
0 & \nu^{a_2}
\end{array}
\right)^*
$
be the action of
$
\left(
\begin{array}{cc}
\alpha^{a_1} & 0 \\
0 & \nu^{a_2}
\end{array}
\right)
$
on $H^0 (Gr^{\mu}(V), Det^*)$.
Then
$$
\left(
\begin{array}{cc}
\alpha^{a_1} & 0 \\
0 & \nu^{a_2}
\end{array}
\right)^*
(\pi_S)=
\alpha^{a_1 \cdot n_1(s)}
\nu^{a_2 \cdot n_2(s)}
\pi_S \quad \mbox{.}
$$
\end{prop}
{\bf Proof\ } follows from
an explicit description of actions of
$
\left(
\begin{array}{cc}
\alpha & 0 \\
0 & 1
\end{array}
\right)
$
and
$
\left(
\begin{array}{cc}
1 & 0 \\
0 & \nu
\end{array}
\right)
$
on the determinant bundle.
We get this description by means of direct
calculations from formulas~(\ref{f}), (\ref{ff}) and~(\ref{fff}).\\[10pt]
For any $S \in S_{\mu}$ let $n(S) \stackrel{\rm def}{=} n_1(S) - n_2(S)$.
\begin{lemma} \label{ch}
Let $W \in Gr^{\mu} (V)$.
Let $S(W) = \{S \in S_{\mu} \; : \; \pi_S(W) \ne 0 \}$.
Then we have
\begin{enumerate}
\item
if $\min\limits_{S \in S(W)} n(S) > -\infty$, then $l_1 \cap W \ne 0$;
\item if $\mathop {\rm rank} A_W = 1$, $l_1 \cap W \ne 0$,
then $\min\limits_{S \in S(W)} n(S) > - \infty$.
\end{enumerate}
\end{lemma}
{\bf Proof\ }.
Remind that
\begin{equation} \label{kr}
\pi_S (W) \ne 0 \quad \mbox{if and only if}
\quad
W \oplus A_S = V
\end{equation}
Consider the second statement of this lemma.
If $l_1 \cap W \ne 0$, then from $\mathop {\rm dim}_k W \cap V_0 < \infty $
we have
$\mathop {\rm dim}_k W \cap l_1 \cap V_0 < \infty$.
And from $\mathop {\rm rank} A_W = 1$
we have
$$
\mathop {\rm dim}\nolimits_k k((z)) / (A_W + k[[z]]) < \infty
\quad \mbox{, therefore}
$$
$$
\mathop {\rm dim}\nolimits_k l_1 / (W \cap l_1 + V_0 \cap l_1) < \infty
\quad
\mbox{.}
$$
Now let some $S' \in S(W)$, then
$W \oplus A_{S'} = V$.
Hence $W \cap A_{S'} = 0$. Hence
\begin{equation} \label{dvezv}
(W \cap l_1) \cap (A_{S'} \cap l) = 0 \quad \mbox{.}
\end{equation}
From~(\ref{dvezv}) we obtain
$$
n_1(S') \ge \mathop {\rm dim}\nolimits_k W \cap l_1 \cap V_0 -
\mathop {\rm dim}\nolimits_k l_1 /( W \cap l_1 + V_0 + l_1) \quad \mbox{.}
$$
From~(\ref{twodiez}) we have
$
n(S') = 2 n_1(S') - \mu
$. Therefore
\begin{equation} \label{gg}
n(S') \ge 2 \left( \mathop {\rm dim}\nolimits_k W \cap l_1 \cap V_0
- \mathop {\rm dim}\nolimits_k l_1 / (W \cap l_1 + V_0 \cap l_1)
\right) - \mu \quad \mbox{.}
\end{equation}
The second statement of this lemma is proved.
Now consider the first statement of this lemma.
From
$\min\limits_{S \in S(W)} n(S) > - \infty$
we see that
$\min\limits_{S \in S(W)} n_1(S) > - \infty$.
Let
$$
\tau = \min(0,\min\limits_{S \in S(W)} n_1(S)) -1 \quad \mbox{.}
$$
For any integer $m>0$ define
$$
B_m = z^{\tau}k[[z]] \oplus z^{m + \mu} k[[z]] \subset V
\quad \mbox{.}
$$
Fix some integer $m_0 > \max (0, -\mu, -\tau)$ such that
$$
W \cap (z^{m_0 + \mu} V_0) = 0
\qquad
\mbox{and}
\qquad
V / (W + z^{-m_0} V_0) = 0 \; \mbox{.}
$$
We will show that for every $m > m_0$
\begin{equation} \label{zvzv}
W \cap B_m \ne 0 \; \mbox{.}
\end{equation}
For this goal we consider the images $\tilde{W}$
of $W \cap (z^{-m}V_0)$
and $\tilde{l_1}$ of $l_1 \cap (z^{-m} V_0)$
in the finite-dimensional Grassmanian manifold
$$Gr (2m+\mu, z^{-m}V_0 / z^{m+\mu} V_0) \; \mbox{.} $$
Now~(\ref{zvzv}) is equivalent to
$$
(\tilde{W} \cap \tilde{l_1})
\cap
((z^{\tau}V_0 / z^{m+\mu} V_0 ) \cap \tilde{l_1}) \ne 0 \; \mbox{.}
$$
And the last formula follows from comparison
of dimensions of finite-dimensional vector spaces:
$$
\mathop {\rm dim}\nolimits_k \tilde{l_1} = 2m + \mu
$$
$$
\mathop {\rm dim}\nolimits_k (z^{\tau} V_0 / z^{m+\mu} V_0) \cap \tilde{l_1} =
m+\mu -\tau
$$
\begin{equation} \label{rrr}
\mathop {\rm dim}\nolimits_k
\tilde{W} \cap \tilde{l_1} >
m + \tau \quad \mbox{,}
\end{equation}
where~(\ref{rrr}) is true by the following reason. \\
If
\begin{equation} \label{bec}
\mathop {\rm dim}\nolimits_k \tilde{W} \cap \tilde{l_1} \le m + \tau \quad \mbox{,}
\end{equation}
then we find a $k$-vector subspace $K \subset \tilde{l_1}$
such that $K$
is the span of $\{ z^{i_k} \}$
for some integers $i_{k}$
and
$\tilde{l_1} = (\tilde{W} \cap \tilde{l_1}) \oplus K $.
Then we can find $\tilde{L} \in
Gr(2m+\mu, z^{-m}V_0 / z^{m+\mu} V_0) $
such that $\tilde{L}$
is the span of $\{ z^{i_l} \}$ for some integers $i_l$,
$\tilde{L} \cap \tilde{l_1} = K$,
and
$$z^{-m}V_0 / z^{m+\mu}V_0 = \tilde{W} \oplus \tilde{L} \quad \mbox{.}$$
Let $L$ be the pre-image of $\tilde{L}$
in $z^{-m}V_0$.
We can find the Pl\"ucker coordinate
$S' \in S_{\mu}$ such that $L = A_{S'}$. Then, by construction,
\begin{equation} \label{krkr}
V = W \oplus A_{S'} \quad \mbox{,}
\end{equation}
and from~(\ref{bec}) $n_1(S') \le \tau < \min\limits_{S \in S(W)} n_1(S)$.
The last formula contradicts~(\ref{krkr}).
Therefore formula~(\ref{rrr}) is proved. Hence formula~(\ref{zvzv})
is true.
Now from the Fredholm condition on $W$ we
have $\mathop {\rm dim}_k W \cap (z^{\tau} V_0) < \infty$.
From~(\ref{zvzv}) we have non-empty filtration
$ W \cap B_m $
for all sufficiently large integers m.
The last one is equivalent to
$$
(W \cap (z^{\tau} V_0)) \cap l_1 \ne 0 \quad \mbox{.}
$$
Indeed, assume the converse. Then choose
a finite $k$-basis $e_1, \ldots, e_n$
of $ W \cap z^{\tau}V_0 $
such that $\nu_2(e_i) \ne \nu_2(e_j)$
for all pairs $i,j$. (Here $\nu_2(e_k) = p$ if $e_k \in B_p$
but $e_k \ne B_{p+1}$.)
Then for any $m > \max\limits_i \nu_2(e_i) $
we have $W \cap B_m = 0$.
This contradiction concludes the proof of lemma~\ref{ch}. \\[3pt]
\begin{lemma} \label{chch}
Let $W \in Gr^{\mu}(V)$,
$\mathop {\rm rank} A_W = 1$,
and condition~1 or condition~2
of the previous lemma is true,
then
\begin{equation} \label{chk}
\min_{S \in S(W)} n(S) =
2 \left( \mathop {\rm dim}\nolimits_k W \cap l_1 \cap V_0
- \mathop {\rm dim}\nolimits_k l_1/(W \cap l_1 + V_0 \cap l_1)
\right) - \mu
\end{equation}
\end{lemma}
{\bf Proof\ }.
From the proof of expression~(\ref{gg})
we have that the left hand side of expression~(\ref{chk})
is greater or equal to the right hand side of~(\ref{chk}).
Now fix some integer $m > 0$ such that
$$
W \cap (z^{m + \mu} V_0) = 0
\qquad
\mbox{and}
\qquad
V / (W + z^{-m} V_0) = 0 \; \mbox{.}
$$
Consider the images $\tilde{W}$
of $W \cap (z^{-m}V_0)$
and $\tilde{l_1}$ of $l_1 \cap (z^{-m} V_0)$
in
$ z^{-m}V_0 / z^{m+\mu} V_0$.
Then find a $k$-vector subspace $K \subset \tilde{l_1}$
such that $K$
is the span of $\{ z^{i_k} \}$
for some integers $i_{k}$
and
\begin{equation} \label{l}
\tilde{l_1} = ( \tilde{W} \cap \tilde{l_1}) \oplus K \quad \mbox{.}
\end{equation}
Then we can find $\tilde{L} \subset
z^{-m}V_0 / z^{m+\mu} V_0 $
such that $\tilde{L}$
is the span of $\{ z^{i_l} \}$ for some integers $i_l$,
$\tilde{L} \cap \tilde{l_1} = K$,
and
\begin{equation} \label{ll}
z^{-m}V_0 / z^{m+\mu}V_0 = \tilde{W} \oplus \tilde{L} \quad \mbox{.}
\end{equation}
Denote by $L$
the pre-image of $\tilde{L}$
in $z^{-m} V_0$.
Then there exists some $S' \in S_{\mu}$
such that $L = A_{S'}$ and $V = W \oplus A_{S'}$.
From~(\ref{l}) and~(\ref{ll})
we have $n(S') = n(W)$. Lemma~\ref{chch} is proved.
\section{Stable and non-stable points}
For torsion free sheaves on $C$ one can define the notion
of "stability" and "semistability". (See~\cite{N}.)
Remind these definitions.
\begin{defin}
A rank $2$ torsion free sheaf ${\cal F}$ on a curve $C$
is called a semistable sheaf if for any subsheaf ${\cal G} \in R({\cal F})$
$$
2 \chi({\cal G}) \le \chi ({\cal F}) \quad \mbox{.}
$$
\end{defin}
\begin{defin}
A rank $2$ torsion free sheaf ${\cal F}$ on a curve $C$
is called a stable sheaf if for any subsheaf ${\cal G} \in R({\cal F})$
$$
2 \chi({\cal G}) < \chi ({\cal F}) \quad \mbox{.}
$$
\end{defin} \vspace{3pt}
\begin{th} \label{10vkv}
Let $W = K (C, p, {\cal F}, t_p, e_p)$.
Then the sheaf ${\cal F}$ is not a (semi)stable sheaf if and only if
there exists some $g \in SL (2, k[[z]])$
such that for any $S \in S(gW)$ $n(S) \ge 0$ $(>0)$.
\end{th}
{\bf Proof\ } follows from proposition~\ref{p1},
lemmas~\ref{Shurik}, \ref{ch} and~\ref{chch}, and facts that
1) group $SL(2, k[[z]])$ acts transitively on the set of all
$1$-dimensional over $k((z))$ vector subspaces in $V$;
2) the group $SL(2, k[[z]])$ keeps $V_0$ stable. \\[4pt]
\noindent
{\bf Remark}.
From lemma~\ref{chch} and formula~(\ref{ech}) we can also
obtain numerical datas of nonstability of the sheaf. \\[5pt]
\noindent {\bf Example 1.} (See also~\cite[ prop.~3.8]{M}.)
Let $\mu = 2\nu$ for some integer $\nu$,
and $W = K(C, p, {\cal F}, t_p, e_p)$
is a point of the big cell of $Gr^{\mu}_V$, i.~e.,
\begin{equation} \label{33}
\pi_{0(\mu)} (W) \ne 0 \quad \mbox{,}
\end{equation}
then the sheaf ${\cal F}$ is semistable.
Indeed, if condition~(\ref{33}) is true,
then $W \oplus z^{\nu} V_0 = V$
and for any $g \in SL(2, k[[z]])$,
$g (z^{\nu} V_0) = z^{\nu} V_0$,
$gW \oplus z^{\nu} V_0 = V $.
Therefore
$\pi_{0(\mu)} (gW) \ne 0 $ . But $n (0(\mu)) = 0$,
and we apply theorem~\ref{10vkv}. \\[2pt]
We can generalize example 1 in the following way. \\[1pt]
\noindent {\bf Example 2.}
With every $W \in Gr^{\mu}(V)$
one can associate (see~\cite{P}):
1) the function $T_W : {\mbox{\dbl Z}} \to \{ 0,1,2 \}$
such that $T_W(i) = 2$ for $i \ll 0 $, $T_W(i)= 0 $ for
$i \gg 0$, and $T_{gW} = T_W$ for any $g \in SL(2, k[[z]])$
$$
T_W (i) \stackrel{\rm def}{=} \mathop {\rm dim}\nolimits_k (W \cap z^i V_0)/ (W \cap z^{i+1} V_0) \;
\mbox{,}
$$
2) the point
\begin{equation} \label{resh}
x_W = (x_{W,i_1}, \ldots , x_{W, i_{l_W}})
\in ({\mbox{\dbl P}}^1_k)^{l_W}
\quad \mbox{,}
\end{equation}
where $i_1, \ldots, i_{l_W}$
are all the integers such that
$ T_W(i_j)=1 $, and the point $x_{W,i_j} \in {\mbox{\dbl P}}^1_k $
is the $1$-dimensional $k$-vector space
$(W \cap z^{i_j} V_0)/(W \cap z^{i_j+1} V_0)$
in the $2$-dimensional $k$-vector space $z^{i_j}V_0 / z^{i_j+1}V_0$.
Note that all ${\mbox{\dbl P}}^1_k$ from~(\ref{resh})
have the canonical homogeneous coordinates,
which are induced by
$z^{i_j} \oplus 0 \in V$
and $0 \oplus z^{i_j} \in V$
for the corresponding $j$.
By means of these coordinates we can identify
all ${\mbox{\dbl P}}^1_k$ from~(\ref{resh})
and consider $x_W$ as $l_W$
ordered points on a line.
There exists the natural diagonal action of the group
$SL(2,k)$ on $({\mbox{\dbl P}}^1_k)^{l_W}$.
Consider the Segre imbedding
$({\mbox{\dbl P}}^1_k)^{l_W}
\hookrightarrow
{\mbox{\dbl P}}^{2l_W -1}_k
$.
There is a unique linear action of
$SL(2,k)$ on
${\mbox{\dbl P}}^{2l_W -1}_k
$
which makes the
Segre imbedding into a $SL(2,k)$-morphism
(the $l_W$-fold tensor represantation of $SL(2,k)$).
Also, $SL(2,k)$ is a reductive group.
Therefore G.I.T. determines (semi)stable
points on $({\mbox{\dbl P}}^1_k)^{l_W}$ with respect to the $SL(2,k)$ action
and the Segre imbedding.
From~\cite{N} we have the following explicit description:
$x \in ({\mbox{\dbl P}}^1_k)^{l_W}$ is a (semi)stable point if and only if
no point of ${\mbox{\dbl P}}^1_k$ occurs as a component of
$x \ge l_W/2$ $(> l_W/2)$ times.
Now we can formulate the following proposition.
\begin{prop} \label{den}
Let $W = K (C, p, {\cal F}, t_p, e_p)$.
Then
1) If $1 \notin {\rm Im}\: T_W$, then ${\cal F}$ is a semistable sheaf.
2) If $x_W \in ({\mbox{\dbl P}}^1_k)^{l_W}$
is a (semi)stable point with respect to the $SL(2,k)$ action
and the Segre imbedding, then ${\cal F}$ is a (semi)stable sheaf.
\end{prop}
{\bf Proof\ }.
Indeed, if
$1 \notin {\rm Im}\: T_W$,
then let $M \subset V$
be the span of $\Bigl\{ \{z^i \oplus 0 \}, \{0 \oplus z^i \} \: : \:
T_W(i) = 2
\Bigr\} $.
We can find $S' \in S_{\mu}$
such that $M = H_{S'}$.
By construction, $W \oplus A_{S'} = V$. Hence $\pi_{S'} (W) \ne 0$.
For any $g \in SL(2, k[[z]])$
we have $T_{gW}=T_W$. Therefore
$\pi_{S'} (gW) \ne 0$.
From theorem~\ref{10vkv} and $n(S')=0$
we see that ${\cal F}$ is a semistable sheaf.
Now consider the case 2. Let
$
x_W = (x_{W,i_1}, \ldots , x_{W, i_{l_W}})
\in ({\mbox{\dbl P}}^1_k)^{l_W}
$.
For every $k \in \{i_1, \ldots, i_{l_W} \}$
we can choose $z_k \in V$
equal to $\{ z^k \oplus 0 \}$
or $\{ 0 \oplus z^k \}$
such that the image of the $k$-line $k \cdot z_k$
in $z^k V_0 / z^{k+1} V_0$
does not coincide with $x_{W,k}$.
Let $M \subset V$
be the span of $\Bigl\{ z_k : k \in \{i_1, \ldots, i_{l_W} \}
\Bigr\} $
and $\Bigl\{ \{z^i \oplus 0 \}, \{ 0 \oplus z^i \} \:
: \: T_W(i)=2 \Bigr\} $.
Then there exists $S' \in S_{\mu}$
such that $M = H_{S'}$
and $W \oplus A_{S'} = V$,
i.~e., $\pi_{S'} (W) \ne 0$.
Moreover, if $x_W \in ({\mbox{\dbl P}}^1_k)^{l_W} $
is a (semi)stable point,
then from the explicit description of
such points we can choose $z_k \in V$
such that
\begin{equation} \label{dom}
n(S') < 0 \; (\le 0) \quad \mbox{.}
\end{equation}
The action
of $SL(2,k)$ on $({\mbox{\dbl P}}^1_k)^{l_W} $
is induced by the action of $SL(2,k[[z]])$
on $V$.
Therefore $x_{gW} \in ({\mbox{\dbl P}}^1_k)^{l_{gW}} $
is a (semi)stable point for any $g \in SL(2, k[[z]])$.
Now from this fact, formula~(\ref{dom}),
and theorem~\ref{10vkv}
we get that ${\cal F}$
is a (semi)stable sheaf.
\\[4pt]
\noindent
{\bf Remark}.
Suppose that
$x \in
({\mbox{\dbl P}}^1_k)^{l} $
is not a semistable point with
respect to the $SL(2,k)$ action and the Segre imbedding.
Then it is easy to construct quintets
$(C, p, {\cal F}, t_p, e_p)$
with stable, semistable, and nonstable sheaves ${\cal F}$
such that for
$W = K(C, p, {\cal F}, t_p, e_p)$
we have $l_W = l$ and $x_W = x$.
\\[4pt]
\noindent
{\bf Remark}.
One can introduce the cellular decomposition of $Gr^{\mu}(V)$
indexed by the set $S_{\mu}$ (see~\cite{PS}):
$W \in Gr^{\mu}(V)$
belongs to the cell
$S = \{ s_{-\mu +1}, s_{-\mu +2}, \ldots \}$
if $\pi_S (W) \ne 0$
while $\pi_{S'}(W) = 0$
unless $s'_{-\mu +1} \ge s_{-\mu +1},
s'_{-\mu +2 } \ge s_{-\mu +2}, \ldots $.
(Such $S \in S_{\mu}$
is uniquely defined for each $W \in Gr^{\mu}(V)$.)
It is not difficult to see that the case~1 of proposition~\ref{den}
corresponds to the following statement:
if $W = K (C, p, {\cal F}, t_p, e_p)$
belongs to the cell $S \in S_{\mu}$
such that $H_S \cap l_1 = H_S \cap l_2$,
then ${\cal F}$ is a semistable sheaf.\\[5pt]
Now note that the action of the group $SL(2, k[[z]])$
on $Gr^{\mu}(V)$ keeps
$W = K(C, p, {\cal F}, t_p, e_p) \in Gr^{\mu} (V)$
in the image of the Krichever map and does not change
the curve $C$, the point $p$,
the local parameter $t_p$, and the sheaf ${\cal F}$.
So this action is an action
only on "basises $e_p$".
Therefore if we want to
get something like "moduli spaces"
of sheafs on $C$,
we would need "to kill"
the action of $SL(2, k[[z]])$
on $Gr^{\mu} (V)$.
On the other hand, by lemma~\ref{zvezd},
the group $SL(2, k[[z]])$
acts on $Det \mid_{Gr^{\mu} (V)}$.
Moreover, from this action we obtain
an
infinite-dimensional
representation
of $SL(2, k[[z]])$
in the $k$-vector space $\Pi (S_{\mu})$.
(In general, this representation is the restriction of representation of
$\hat{GL_0} (V, V_0) $
in $\Pi (S_{\mu})$, which can be got in the same way and is usually
called the infinite wedge-representation (see~\cite{KP}).)
Hence we have a linear action of
$SL(2, k[[z]])$ on ${\mbox{\dbl P}} (\Pi (S_{\mu})^*)$.
And by construction, this action
is compatible with the action of
$SL(2, k[[z]])$
on $Gr^{\mu} (V)$
via the Pl\"ucker imbedding.
These arguments lead us to the following definitions.
For any $W \in Gr^{\mu} (V)$
denote by $\hat{W}$
some nonequal to $0$ element from $ \Pi (S_{\mu})^* $
such that the image of $\hat{W}$
in ${\mbox{\dbl P}} (\Pi (S_{\mu})^*)$
coincides with the image of $W$
via the Pl\"ucker imbedding.
Let
$$
T = \left\{
\left(
\begin{array}{cc}
\lambda & 0 \\
0 & \lambda^{-1}
\end{array}
\right)
\in
SL(2,k[[z]]),
\: \lambda \in k^*
\right\}
$$
be a 1-parametric subgroup:
$
k^* \to SL(2, k[[z]])
$.
For any $g \in SL(2, k[[z]] $
define the group ${}^gT = gTg^{-1}$.
Let ${}^gT(\lambda) = g
\left(
\begin{array}{cc}
\lambda & 0 \\
0 & \lambda^{-1}
\end{array}
\right)
g^{-1}$ be an element from ${}^gT$.
It will be convinient
to identify $k^*$
with a subset of
${\mbox{\dbl P}}^1_k$ in the obvious manner.
In fact, for any $\lambda \in k$,
we identify $\lambda$ with the point $(1, \lambda)$
in ${\mbox{\dbl P}}^1_k$,
and write $\infty$ for the extra point $(0,1)$.
Note that the morphism $k^* \to \Pi (S_{\mu})^* \; : \;
\lambda \mapsto {}^gT(\lambda) \hat{W}$
may or may not extend to the points $0$, $\infty$,
but if it does extend to either or both of these
points the extension is unique.
Thus we can define the expressions
$
\mathop {\rm lim}\limits_{\lambda \to 0}
{}^gT(\lambda) \hat{W}
$
and
$
\mathop {\rm lim}\limits_{\lambda \to \infty}
{}^gT(\lambda) \hat{W}
$ in an obvious way.
\begin{defin}
A closed point
$W \in Gr^{\mu} (V)$
is a semistable point with respect
to the 1-parametric subgroup ${}^gT$
if for $x = 0$ and $x = \infty $
$
\lim\limits_{\lambda \to x}
{}^gT(\lambda) \hat{W}
$
does not exist
or
$
\mathop {\rm lim}\limits_{\lambda \to x}
{}^gT(\lambda) \hat{W} \ne 0
$.
\end{defin}
\begin{defin}
A closed point
$W \in Gr^{\mu} (V)$
is a stable point with respect
to the 1-parametric subgroup ${}^gT$
if for $x = 0$ and $x = \infty $
$
\mathop {\rm lim}\limits_{\lambda \to x}
{}^gT(\lambda) \hat{W}
$
does not exist.
\end{defin}
Now using these definitions and proposition~\ref{dddd}
we can reformulate theorem~\ref{10vkv}
in the following way.(Compare~\cite{N}.)
\begin{th} \label{th1}
Let $(C,p,{\cal F}, t_p, e_p)$
be a quintet,
$\chi ({\cal F}) = \mu$.
Then the sheaf ${\cal F}$
is a (semi)stable sheaf on the curve $C$
if and only if the point
$K(C,p,{\cal F}, t_p, e_p) \in Gr^{\mu}(V)$
is a (semi)stable point
with respect to the 1-parametric
subgroups ${}^gT$ for all $g\in SL(2, k[[z]])$.
\end{th}
\noindent {\bf Acknowledgments} \\
This note was done during my visit to the Abdus Salam ICTP
in the end of 1998. I am very much grateful to Professor M.~S.~Narasimhan
for the invitation me to the ICTP and for the excellent working
conditions.
I am deeply grateful also to my scientific advisor Professor
A.~N.~Parshin for the permanent attention and help.
|
\section{Introduction}
The solution of the many body problem beyond the meanfield level is not a
very well settled problem. Though the meanfield approach for all kinds of
many body problems is quite uniquely defined, the determination of the
higher order correlation functions is not. Besides the usual partial
resummation of Feynman graphs ( e.g. ring summation in RPA) there also exist
variational ans\"atze such as those introduced by Jastrow, Gutzwiller,
together with the Resonating Valence Bond approach, etc. \cite{Blai86}.
However only in
rare cases these variational approaches can be worked through to the end by
minimizing the groundstate energy so that any new route can have interesting
perspectives. In most cases there remains the additional problem of how to
determine the excited states. One of the attractive features of the
Raleigh-Ritz variational Hartree-Fock
(HF) theory is indeed that it yields, consistently within the same theory,
groundstate and excited states (quasiparticle excitations).
Since some time we have elaborated on a theory for two body correlations
functions which in a certain sense can be considered as an extension of HF
theory to two body clusters. We for instance obtain selfconsistent
nonlinear equations for the correlation functions which simultaneously
determine the correlated groundstate energy and the spectrum of excitations.
We named this approach Self - Consistent Random Phase Approximation (SCRPA),
since it is a consistent generalization of the standard linear RPA
approach \cite{Schu73,Duk90,Duk96,Duk98a,Duk98b}.
This formalism was also developed independently by a second group of authors
which coined for it the name Cluster Hartree-Fock (CHF) which seems also
very appropriate \cite{Rop95}.
This type of theory took its roots several decades back starting with the work
of Hara \cite{Har64}. Considerable progress was achieved by D. Rowe using the
equation of motion method which is summarized in \cite{Row68a}. Some years
later the theory was rederived using the method of many body Green functions
\cite{Schu71,Schu73}. Since that time not much progress was made on the formal
aspect of the theory until the more recent works cited above.
The SCRPA has
lately
given a series of interesting results for various many-body problems
\cite{Kru94,Duk96,Duk98a}.
Nevertheless some
open problems persisted in the past with this formalism concerning for
instance the consistent evaluation of single particle quantities such as the
single particle density matrix or the occupation numbers. An approximation
which lately came very much in use in relating these quantities back to
SCRPA (or to its poorer but numerically easier variant the so called
Renormalized RPA (RRPA) \cite{Har64}) is based on the particle number
method which long time ago already was advocated by D. J.Rowe
\cite{Row68}. Very recently we have proposed
and applied a different method which calculates these quantities via the
single particle Green's function with a mass operator coupling back to the
SCRPA \cite{Duk96,Duk98b}. In those works, however, neither a detailed
derivation nor an assessment of its quality was given . On the other hand it
has been pointed out that certain consistency relations are indeed
fulfilled.
The purpose of the present paper is therefore to give a quite detailed
derivation and to make a systematic investigation in a model case of the
Green's function approach and to contrast it with other methods.
The paper is organized as follows: In section II the SCRPA equations are
deduced, their coupling with the single particle Green's functions is
presented in section III, the application to the Lipkin model is developed
in section IV, the numerical results in section V and the conclusions
are given in section VI.
\section{Outline of the problem}
Self Consistent RPA can be derived in various ways. The method which
probably exhibits most clearly the analogy with ordinary HF theory is the
one due to Baranger \cite{Bar70}. Let us first rederive in this way
single particle HF.
To this end we define a mean single particle energy in the following way
\begin{equation}
\epsilon _{\mu }=\frac{\sum\limits_{\nu ,k}\left\{ \left( E_{\nu
}^{N+1}-E_{0}^{N}\right) \left| \left\langle N,0\right| \varphi _{k}^{\mu
}a_{k}\left| N+1,\nu \right\rangle \right| ^{2}+\left( E_{0}^{N}-E_{\nu
}^{N-1}\right) \left| \left\langle N,0\right| \varphi _{k}^{\mu
*}a_{k}^{\dagger }\left| N-1,\nu \right\rangle \right| ^{2}\right\} }{%
\sum\limits_{\nu ,k}\left\{ \left| \left\langle N,0\right| \varphi
_{k}^{\mu
}a_{k}\left| N+1,\nu \right\rangle \right| ^{2}+\left| \left\langle
N,0\right| \varphi _{k}^{\mu *}a_{k}^{\dagger }\left| N-1,\nu \right\rangle
\right| ^{2}\right\} } \label{e}
\end{equation}
where $E_{\nu }^{N}$ and $\left| N,\nu \right\rangle $ are in principle
exact eigenenergies and eigenstates of the Hamiltonian for a system with
N particles. For the groundstate
we have $\nu =0$ and $a_{k}^{\dagger }$ is a single particle creation
operator. Minimizing (\ref{e}) with respect to the amplitudes $\varphi
_{k}^{\mu }$ and $\varphi _{k}^{\mu *}$ leads directly to the following
eigenvalue problem
\begin{equation}
\sum_{k^{\prime }}\left\langle 0\right| \left\{ a_{k},\left[ H,a_{k^{\prime
}}^{\dagger }\right] \right\} \left| 0\right\rangle \ \varphi _{k^{\prime
}}^{\nu }=\varepsilon _{\nu }\ \varphi _{k}^{\nu } \label{HF}
\end{equation}
where $\left\{ ..., ...\right\}$ is the anticommutator.
It is easy to verify that (\ref{HF}) is just one of the forms of the usual
single particle HF equations, once $| 0 \rangle$ is chosen to be a
Slater determinant.
Let us now in the same way find equations which describe another form of
elementary excitations of the system such as density vibrations. To this
purpose we define in analogy to (\ref{e}) a mean excitation energy:
\begin{equation}
E_{\mu } =
\frac{\sum\limits_{\nu ,
k > k^\prime
}\left\{ \left(E_{\nu
}^{N}-E_{0}^{N}\right) \left| \left\langle N,0\right| X_{kk^{\prime }}^{\mu
}a_{k}^{\dagger }a_{k^{\prime }}\left| N,\nu \right\rangle \right|
^{2}-\left( E_{\nu }^{N}-E_{0}^{N}\right) \left| \left\langle N,0\right|
Y_{kk^{\prime }}^{\mu }a_{k^{\prime }}^{\dagger }a_{k}\left| N,\nu
\right\rangle \right| ^{2}\right\} }
{\sum\limits_{\nu ,k > k^{^{\prime }}}
\left\{ \left| \left\langle N,0\right| X_{kk^{\prime }}^{\mu
}a_{k}^{\dagger }a_{k^{\prime }}\left| N,\nu \right\rangle \right|
^{2}-\left| \left\langle N,0\right| Y_{kk^{\prime }}^{\mu }a_{k^{\prime
}}^{\dagger }a_{k}\left| N,\nu \right\rangle \right| ^{2}\right\} }
\label{E}
\end{equation}
Minimization with respect to the amplitudes $X_{kk^{\prime }}^{\mu },
Y_{kk^{\prime }}^{\mu }$ leads to
\begin{equation}
\left\langle 0\right| \left[ \delta Q,\left[ H,Q_{\nu }^{\dagger }\right]
\right] \left| 0\right\rangle =E_{\nu }\ \left\langle 0\right| \left[ \delta
Q,Q_{\nu }^{\dagger }\right] \left| 0\right\rangle \label{RPA}
\end{equation}
where
\begin{equation}
Q_{\nu }^{\dagger } =
\sum_{k > k^{\prime }}
\left( X_{kk^{\prime }}^{\nu }a_{k}^{\dagger }a_{k^{\prime}} -
Y_{kk^{\prime }}^{\nu }a_{k^{\prime
}}^{\dagger }a_{k}\right) \label{Q}
\end{equation}
and $\delta Q$ is a variation (with respect to $X$ or $Y$) of $Q^{\dagger }$%
. Equation (\ref{RPA}) constitutes the SCRPA equations which are described
in great detail elsewhere \cite{Duk90,Duk96,Duk98a,Duk98b}. Explicitly
\begin{equation}
\left(
\begin{array}{ll}
A & B \\
-B & -A
\end{array}
\right) \left(
\begin{array}{l}
X^{\nu } \\
Y^{\nu }
\end{array}
\right) =E_{\nu }\ {\cal N\ }\left(
\begin{array}{l}
X^{\nu } \\
Y^{\nu }
\end{array}
\right) \label{AB}
\end{equation}
where the matrices $A$ and $B$ are double commutators coming from the left
hand side of (\ref{RPA}) and ${\cal N}$ is the norm matrix to be discussed
in the following section. They lead to a nonlinear eigenvalue problem for
the amplitudes $X$ and $Y$ which therefore have to be determined iteratively
very much like the HF eqs. (\ref{HF}).
Equations (\ref{RPA},\ref{Q},\ref{AB}) are equivalent to
\begin{eqnarray}
\left\langle 0\right| \left[ Q_\nu,\left[ H,Q_{\nu' }^{\dagger }\right]
\right] \left| 0\right\rangle = E_{\nu } ~\delta_{\nu, \nu'} \label{QHQ}
\\
\left\langle 0\right| \left[ Q_\nu^{\dagger },\left[ H,Q_{\nu'
}^{\dagger }\right] \right] \left| 0\right\rangle = 0 \label{QHQt}
\end{eqnarray}
This form is interesting since these equations have exactly the same
structure as any mean field Hartree-Fock-Bogoliubov equations, be it for
single Fermions or Bosons or, as here, for Fermion pairs.
For a hamiltonian with two body
interactions one verifies easily that (\ref{RPA}) contains at most one and
two body density matrices. Roughly speaking the two body density matrices
can be expressed as quadratic forms of the amplitudes $X$ and $Y$ (for more
details see \cite{Duk90,Duk96,Duk98a,Duk98b}).
An important point is to realize that (\ref{Q}) is {\bf not} restricted to
the particle-hole (ph) and (hp) subspaces as is of common use in
the nuclear literature on the subject \cite{Blai86,Row68,Rin80}. Here the
only restriction in (\ref{Q}) is that it should not contain any diagonal
(i.e. Hermitian) components. Therefore in
$Q_{\nu }^{\dagger } = \sum_{k \neq k^{\prime }}
\chi_{kk^{\prime }}^{\nu }a_{k}^{\dagger }a_{k^{\prime}}$ the matrix
is not Hermitian. The single particle basis in which
(\ref{RPA},\ref{Q},\ref{AB}) shall be solved is obtained from
\begin{equation}
\left\langle 0\right| \left[ H,Q_{\nu}^{\dagger } \right]
\left| 0\right\rangle = \left\langle 0\right| \left[ H,Q_{\nu}
\right] \left| 0\right\rangle = 0 \label{HQ}
\end{equation}
One can show that (\ref{HQ}) is obtained from the minimization of the
SCRPA ground state energy with respect to the basis
\cite{Duk90,Duk96,Duk98b} but one also
directly realizes that (\ref{HQ}) is consistent with the equations of
motion (\ref{QHQ},\ref{QHQt}).
The matrix $B$ contains the pair potential of the two fermion pairs whereas
the matrix $A$ contains the normal selfconsistent potential for Fermion
pairs. Qualitatively we can represent the selfconsistent equations (\ref{AB}%
) as in Fig. 1 \cite{Schu73}
Figure 1
\noindent
where the
wiggly line stands for quantum fluctuations. Such a selfconsistent mean
field potential for density fluctuations as shown in Fig. 1 seems quite
natural, since the groundstate of an interacting Fermi system can be
considered as a gas of quantal fluctuations. The presence of fluctuations
also has a feedback on the single particle motion, an issue which we
mainly want to consider in this paper. For example, to couple back
consistently the
single particle density matrix $\rho _{kk^{\prime }}=\left\langle 0\right|
a_{k}^{\dagger }a_{k^{\prime }}\left| 0\right\rangle $ to the amplitudes $X$
and $Y$ in order to close the system of equations, has been a matter of
debate in the past\cite{Cat96}. It should be noted that, depending on the
problem at
hand, it also can happen that certain elements of the two body density
matrix can not directly be expressed via $X$ and $Y$ amplitudes. In the
Lipkin model which we will study below we will see that indeed a particular
matrix element of the two body density matrix falls into this category. We
will, however, demonstrate that once we have a method at hand that allows to
calculate the single particle density matrix we will also find a reliable
method of how to evaluate the missing two body elements.
\section{Coupling the single particle Green's function to the Self
Consistent RPA}
The eigenvalue problem (\ref{RPA}) has as usual a corresponding Green's
Function (GF) formulation. For the following it is useful to also briefly
outline this approach which, of course, is completely equivalent
to the eigenvalue problem (\ref{RPA}).
Let us therefore define the two time chronological Green's function at
zero temperature which describe density fluctuations
\begin{equation}
G_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime }}}^{t-t^{\prime
}}=-i\left\langle 0\right| T\left( a_{k_{2}}^{\dagger }a_{k_{1}}\right)
_{t}\left( a_{k_{1}^{^{\prime }}}^{\dagger }a_{k_{2}^{^{\prime }}}\right)
_{t^{\prime }}\left| 0\right\rangle \label{G1}
\end{equation}
where $T$ is the chronological operator and
\begin{equation}
O_{t}=e^{iHt}Oe^{-iHt} \label{G2}
\end{equation}
with $H$ the full Hamiltonian operator.
In principle in (\ref{G1}) one should take only the fluctuating operator
$a^\dagger a - \langle 0 |a^\dagger a | 0 \rangle$ but since in the
equations of motion (\ref{RPA},\ref{Q}) any c-number drops out we will
stay with the definition given in (\ref{G1}).
The Dyson equation for (\ref{G1})
corresponding to (\ref{RPA}) reads after Fourier transformation in the
approximation of the instantaneous mass operator \cite{Duk90,Duk98a}:
\begin{equation}
\omega G_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime }}}^{\omega }={\cal N}%
_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime
}}}+\sum_{p_{1}p_{2}k_{1}k_{2}p_{1}p_{2}}{\cal H}^{\left( 0\right) }\
G_{p_{1}p_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime }}}^{\omega } \label{G3}
\end{equation}
with
\begin{equation}
{\cal N}_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime }}}=\left\langle
0\right| \left[ a_{k_{2}}^{\dagger }a_{k_{1},}a_{k_{1}^{^{\prime
}}}^{\dagger }a_{k_{2}^{^{\prime }}}\right] \left| 0\right\rangle
\label{G4}
\end{equation}
and
\begin{equation}
{\cal H}_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime }}}^{\left( 0\right)
}=\sum_{p_{1}p_{2}}\left\langle 0\right| \left[ a_{k_{2}}^{\dagger
}a_{k_{1},}\left[ H,a_{p_{1}}^{\dagger }a_{p_{2}}\right] \right] \left|
0\right\rangle \ {\cal N}_{p_{1}p_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime
}}}^{-1} \label{G5}
\end{equation}
One easily recognizes from (\ref{G3}-\ref{G5}) the equivalence with (\ref
{RPA}). Since the Eqs. (\ref{G3},\ref{G4},\ref{G5}) have been derived at
length in several preceding articles \cite{Duk90,Duk98a} we will not
represent them
here.
For the coupling with the single particle Green's function it is useful to
define a SCRPA T-matrix from (\ref{G3}) in the following way
\begin{equation}
G_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime }}}^{\omega
}=G_{k_{1}k_{2}k_{1}^{\prime }k_{2}^{\prime }}^{0}\
+G_{k_{1}k_{2}p_{1}p_{2}}^{0}T_{p_{1}p_{2}p_{2}^{\prime }p_{2}^{\prime
}}^{SCRPA}\ G_{p_{1}^{\prime }p_{2}^{\prime }k_{1}^{\prime }k_{2}^{\prime
}}^{0} \label{G6}
\end{equation}
with
\begin{equation}
G_{k_{1}k_{2}k_{1}^{\prime }k_{2}^{\prime }}^{0}=\frac{n_{k_{2}}-n_{k_{1}}}{%
\omega -\varepsilon _{1}+\varepsilon _{2}}\delta _{k_{1}k_{1}^{\prime
}}\delta _{k_{2}k_{2}^{\prime }} \label{G60}
\end{equation}
where $n_{k}=\left\langle 0| a_{k}^{\dagger }a_{k}|0 \right\rangle $ and
$ \varepsilon _{k}=\frac{k^{2}}{2m}+\sum_{k^{^{\prime }}}\overline{v}%
_{kk^{^{\prime }}kk^{^{\prime }}}\ n_{k^{^{\prime }}}$ are the occupation
numbers and generalized single particle energies which we assumed without
loss of generality to be diagonal and $\overline{v}_{k_{1}k_{2}k_{3}k_{4}}$
is the antisymmetrised matrix element of the two body interaction. With (\ref
{G3}-\ref{G5}) the T-matrix in (\ref{G6}) is uniquely defined. Since this is
quite standard procedure we do not further elaborate on the form of the
T-matrix. A form equivalent to (\ref{G6}) is given by (we use summation
convention)
\begin{equation}
G_{k_{1}k_{2}k_{1}^{\prime }k_{2}^{\prime }}=G_{k_{1}k_{2}k_{1}^{\prime
}k_{2}^{\prime }}^{0}\
+G_{k_{1}k_{2}p_{1}p_{2}}^{0}K_{p_{1}p_{2}p_{2}^{\prime }p_{2}^{\prime
}}^{SCRPA}\ G_{p_{1}^{\prime }p_{2}^{\prime }k_{1}^{\prime }k_{2}^{\prime }}
\label{G61}
\end{equation}
with
\begin{equation}
K_{k_{1}k_{2}k_{1}^{\prime }k_{2}^{\prime }}^{SCRPA}={\cal H}%
_{k_{1}k_{2}k_{1}^{\prime }k_{2}^{\prime }}^{\left( 0\right) }-\left(
\varepsilon _{k_{1}}-\varepsilon _{k_{2}}\right) \delta _{k_{1}k_{1}^{\prime
}}\delta _{k_{2}k_{2}^{\prime }} \label{G62}
\end{equation}
From (\ref{G61}-\ref{G62}) we also read off the equality
\begin{equation}
\sum_{k_{3}k_{4}}K_{k_{1}k_{2}k_{3}k_{4}}^{SCRPA}G_{k_{3}k_{4}k_{1}^{\prime
}k_{2}^{\prime }}=\sum_{k_{3}k_{4}}T_{k_{1}k_{2}k_{3}k_{4}}^{SCRPA}\
G_{k_{3}k_{4}k_{1}^{\prime }k_{2}^{\prime }}^{0} \label{G63}
\end{equation}
The important point to recognize is that the mass operator of the single
particle Dyson equation
\begin{equation}
\left( \omega -\varepsilon _{k}\right) \ G_{kk^{^{\prime }}}^{\omega
}=\delta _{kk^{^{\prime }}}+\sum_{p}{\cal M}_{kp}^{\omega }\ G_{pk^{^{\prime
}}}^{\omega } \label{G7}
\end{equation}
has a well known representation in terms of the full two body T-matrix
\cite{Duk96}. For better visibility we present the relation graphically
in figure 2.
Figure 2
At this point it has now become obvious what our interrelation of single
particle GF and SCRPA shall be: we have to replace in Fig. 2 the full
T-matrix by the approximate $T^{SCRPA}\left( \omega \right) $ defined in (%
\ref{G6}). In addition to this obvious construct there also exists a direct
and strong consistency requirement. It stems from the fact that we have now
two ways of calculating the correlation energy: the first uses the well
known relation between the single particle GF and the ground state energy
\cite{Duk98a,Jan91,Mig67}
\begin{equation}
E_{0}=-\frac{i}{2}\ \lim_{t^{\prime }-t\rightarrow 0^{+}}Tr\left( i\frac{%
\partial }{\partial t}+\varepsilon _{k}\right) \ G_{kk}^{t-t^{\prime }}\quad
\label{G9}
\end{equation}
The second expresses the correlation energy density via the two body GF (\ref
{G1}) :
\begin{equation}
E_{corr}=
{\frac i 4} \lim_{t'-t \rightarrow 0^+}
Tr\left[ \overline{v}_{k_{1}k_{2}^{\prime}k_{2}k_{1}^{\prime }}
\left( G_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime}}}^{t-t' } -
G_{k_{1}k_{2}k_{1}^{^{\prime }}k_{2}^{^{\prime }}}^{(0) t-t'}
\right) \right] \label{G10}
\end{equation}
where again $\overline{v}_{k_{1}k_{2}^{\prime}k_{2}k_{1}^{\prime }}$ is
the antisymmetrised two-body matrix element entering in the Hamiltonian
$H$.
The requirement is now that both expressions for the correlation energy,
that is, the one deduced from (\ref{G9}) and (\ref{G10}),
agree. This is equivalent to the Hugenholtz-van Hove theorem which states
that the chemical potential $\mu $ calculated via the single particle GF
must be equal (at equilibrium ) to the energy per particle when calculated
from (\ref{G10}). It turns out that this only is achieved when expanding the
GF in (\ref{G7}) to first order in the mass operator
\begin{equation}
G_{k}=G_{k}^{0}+G_{k}^{0}\ {\cal M}_{k}^{\omega }\ G_{k}^{0} \label{G11}
\end{equation}
with
\begin{equation}
\left( \omega -\varepsilon _{k}\right) \ G_{k}^{0}=1 \label{G12}
\end{equation}
Of course one can use the iterated solution of the Dyson equation, $i.e.$ $%
G_{k}=\left( \omega -\varepsilon _{k}-{\cal M}_{k}^{\omega }\right) ^{-1}$
but for consistency then the particle-hole propagators of the SCRPA equation
must be redefined accordingly. This has been discussed in \cite{Duk98a}
and may be elaborated in the future but for the moment we keep with the more
restrictive consistency relation (\ref{G11}) together with (\ref{G3}-\ref{G6}%
).
For space reasons we have been relatively short in this general section. We
will, however, work out in some detail the model case of the next section so
that the reader, by analogy, shall be able to reconstruct details also in
the general case quite easily.
\section{Application to the Lipkin Model}
The Hamiltonian of the Lipkin \cite{Lip65} model is given by
\begin{equation}
H=\varepsilon J_{0}-{\frac{V}{2}}(J_{+}^{2}+J_{-}^{2}) \label{lip1}
\end{equation}
\noindent with
\begin{eqnarray}
J_{0} &=&{\frac{1}{2}}\sum_{m=1}^{\Omega }(c_{1m}^{\dagger
}c_{1m}-c_{0m}^{\dagger }c_{0m})~~~, \nonumber \\
J_{+} &=&\sum_{m=1}^{\Omega }c_{1m}^{\dagger
}c_{0m}~,~~~~~~~~~~~~~J_{-}=\sum_{m=1}^{\Omega }c_{0m}^{\dagger }c_{1m}~~~~
\label{lip2}
\end{eqnarray}
\noindent The indices 0 and 1 denote the lower and upper levels
respectively, separated by an energy $\varepsilon $, and $m$ is the angular
momentum projection in each shell with degeneracy $\Omega $.
The commutation relations between these three operators, which are the
generators of the SU(2) group, are
\begin{equation}
\lbrack J_{+},J_{-}]=2J_{0},~~~~~~~~[J_{0},J_{\pm }]=\pm J_{\pm }~~~.
\label{lip3}
\end{equation}
In the Lipkin model the number of particles is exactly that needed to
completely fill the lower shell, i.e. $N=\Omega $.
\subsection{SCRPA equations}
The SCRPA solutions are built with the operators (we stay in the normal
phase)
\begin{equation}
Q^{\dagger }={\frac{1}{\sqrt{-2 \langle J_0 \rangle }}}[X J_{+} -
Y J_{-}]~,~~~~~
Q={\frac{1}{\sqrt{-2 \langle J_0 \rangle }}} [X J_{-}-Y J_{+}]
\label{q}
\end{equation}
\noindent acting over a correlated vacuum $|0\rangle $,which is defined by
the equation
\begin{equation}
Q|RPA\rangle =0 \label{vac}
\end{equation}
to yield the excited state
\begin{equation}
\left| 1\right\rangle =Q^{\dagger }|RPA\rangle ~~.
\end{equation}
The SCRPA equations (\ref{RPA}) then take the following form
\begin{equation}
\left(
\begin{array}{cc}
A & B \\
-B & -A
\end{array}
\right) \left(
\begin{array}{c}
X \\
Y
\end{array}
\right) = E \left(
\begin{array}{c}
X \\
Y
\end{array}
\right) \label{SCRPA}
\end{equation}
\noindent with the matrix elements $A$ and $B$ defined by \cite{Row68,Par68}
\begin{eqnarray}
A &=&\left\langle [J_{-},[H,J_{+\langle }]]\right\rangle /\left\langle
[J_{-},J_{+}]\right\rangle \nonumber \\
B &=&\left\langle [J_{+},[H,J_{+}]]\right\rangle /\left\langle
[J_{-},J_{+}]\right\rangle \label{a,b}
\end{eqnarray}
where we used $\left\langle \cdots \right\rangle $ for $\left\langle
RPA\right| \cdots \left| RPA\right\rangle $ .
The normalization of the excited state $Q^{\dagger }|RPA\rangle $ is given by
\begin{equation}
\langle QQ^{\dagger }\rangle =\langle [Q,Q^{\dagger }]\rangle
= X^{2}-Y^{2} = 1 \label{norm}
\end{equation}
With (\ref{norm}) the inversion of (\ref{q}) yields $J_{+}=\sqrt{-2
\langle J_{0}\rangle}\left(
XQ^{\dagger }+YQ\right) $ and the matrix elements of the SCRPA matrices read
\[
A_{SCRPA}=\varepsilon +2VXY~~~
\]
\begin{equation}
B_{SCRPA}=2V{\frac{\langle J_{0}^{2}\rangle }{\langle J_{0}\rangle }}%
+V(X^{2}+Y^{2}) \label{ab}
\end{equation}
From (\ref{ab}) we see that we face exactly the problem discussed in Sect.
3. The single particle occupation $\left\langle J_{0}\right\rangle $ and
the square $\left\langle J_{0}^{2}\right\rangle $ can not directly be
expressed
in terms of $X$ and $Y$ . Of course as is well known in the present simple
model it is possible to calculate the RPA groundstate via (\ref{vac})
explicitly \cite{Lip65,Par68} :
\begin{equation}
|RPA\rangle =\sum_{l=0}^{\Omega /2}{\frac{(\Omega -2l)!}{(\Omega /2-l)!l!}}%
~\left( {\frac{Y}{X}}\right) ^{2}~J_{+}^{2l}~|HF\rangle \label{rpavac}
\end{equation}
and therefore also $\left\langle J_{0}\right\rangle $ and $\left\langle
J_{0}^{2}\right\rangle $ can explicitly be calculated \cite{Duk90}.
However, this is not the usual situation and in general it will be very
difficult if not impossible to solve the vacuum condition (\ref{vac}). We
have therefore to develop other methods to get access to these quantities
independently. As a word of caution we should mention again (see
Section II) that it is not
possible to include $J_{0}$ as a further component into the definition of
the RPA excitation operator $Q^{\dagger }$ (\ref{q}) since $J_{0}$ is
hermitian and it is then impossible to define the norm of the RPA excited
state.
In the next section we will therefore elaborate on the evaluation of $%
\left\langle J_{0}\right\rangle $ via the single particle GF the way we
have
outlined it in the general section 3.
For later use we also introduce here the renormalized RPA (RRPA) matrix
elements
\begin{equation}
A_{RRPA} = \varepsilon ,~~~~~~~
B_{RRPA} = 2 V \langle J_{0}\rangle ~~~~~~~~~~.
\end{equation}
They will be used below when we will compare the results of the SCRPA not
only to the exact solution but also to RRPA.
\subsection{SCRPA and the single particle Green's function}
\bigskip As outlined above we have to construct a mass operator for the s.
p. GF such that it yields exactly the same groundstate energy via eq.
(\ref{G9}) as
when calculated directly from the two body GF (\ref{G10}). In order to
explain the
principle we first want to exemplify the procedure with standard RPA. In
this case we have to put in eq. (\ref{ab}) $X=1$ , $Y=0$ and $\left\langle
J_{0}\right\rangle = -\frac{\Omega }{2}$, $\left\langle
J_{0}^{2}\right\rangle =\frac{\Omega ^{2}}{4}$ . Let us for example
consider
the interaction energy to RPA order
\begin{equation}
E_{pot}=-\frac{V}{2}\left( \left\langle J_{+}^{2}\right\rangle
+\left\langle J_{-}^{2}\right\rangle \right) \Rightarrow E_{pot}^{RPA}=-V\
\Omega \ X\ Y \label{epot}
\end{equation}
Using one of the RPA equations (dropping $1/\Omega $ corrections):
\begin{equation}
V\ \Omega \ X=\left( E +\varepsilon \right) \ Y \label{xx}
\end{equation}
and multiplying this equation with $X$ we obtain for $E_{pot\text{ }}$:
\begin{equation}
E_{pot}^{RPA}=-V\ \Omega \ \frac{X^{2}}{E+\varepsilon }\ V\ \Omega
\label{eprpa}
\end{equation}
Expression (\ref{eprpa}) can be identified with the evaluation of the
Feynman graph shown in Figure 3
Figure 3
\noindent
where the wiggly line represents the RPA phonon with energy $E$ . The
particle-hole bubble has energy $\varepsilon _{ph}=\varepsilon $ and
together with the phonon the vertical cut has energy $E +\varepsilon $
what corresponds to the energy denominator in (\ref{eprpa}). The amplitude
of the phonon is $X^{2}$ and the two dots of the graph represents the
interaction squared. As usual we can obtain the mass operator from the
groundstate graph in cutting open the hole line. Therefore we obtain e. g.
for the GF of the upper level ( $G_{1m}^{t-t^{\prime }}=-i\left\langle
T\left( a_{1m}\left( t\right) a_{1m}^{\dagger }\left( t^{\prime }\right)
\right) \right\rangle $ ) in the approximation of eq. (\ref{G11})
\begin{equation}
G_{1m}^{\omega }=\frac{1}{\omega -\frac{\varepsilon }{2}}+\frac{1}{\omega -%
\frac{\varepsilon }{2}}\ V\ \frac{\Omega ^{2}X^{2}}{\omega +E+\frac{%
\varepsilon }{2}}\ V\ \frac{1}{\omega -\frac{\varepsilon }{2}} \label{H14}
\end{equation}
where the mass operator has the obvious graphical representation of Fig. 4
Figure 4
Using the (exact) relation (what is just a variant of (\ref{G9})):
\begin{equation}
\frac{i}{2}\lim_{t^{\prime }-t\rightarrow 0^{+}}\left( i\frac{\partial }{%
\partial t}-\frac{\varepsilon }{2}\right) G_{1m}^{t-t^{\prime }}=-\frac{V}{2}%
\left\langle J_{+}^{2}\right\rangle \label{H15}
\end{equation}
and inserting into the $lhs$ expression (\ref{H14}) we obtain
\begin{equation}
-i\lim_{t^{\prime }-t\rightarrow 0^{+}}\left( i\frac{\partial }{\partial t}-%
\frac{\varepsilon }{2}\right) G_{1m}^{t-t^{\prime }}=-\frac{1}{2}\left(
V\Omega \right) ^{2}\frac{X^{2}}{E+\varepsilon } \label{H16}
\end{equation}
This is just half the potential energy (\ref{eprpa}) in the standard RPA
approach. Proceeding analogously with $%
G_{0m}$ and adding to (\ref{H16}) the corresponding expression yields the
missing
factor $2$ . This demonstrates that our construction of the mass operator in
(\ref{H14}) is consistent with the RPA groundstate energy.
Since we now have the s. p. GF at hand it is straightforward to calculate
the occupation numbers via
\begin{equation}
\left\langle a_{1m}^{\dagger }a_{1m}\right\rangle =
-i\lim_{t^{\prime}-t\rightarrow 0^{+}}
G_{1m}^{t-t^{\prime }} \label{H17}
\end{equation}
Inserting into (\ref{H17}) the $rhs$ of (\ref{H14}) yields
\begin{equation}
\sum_{m}\left\langle a_{1m}^{\dagger }a_{1m}\right\rangle =\left( V\Omega
\right) ^{2}\frac{X^{2}}{\left( \varepsilon +E\right) ^{2}}=Y^{2}
\label{H18}
\end{equation}
where in the last equality we again made use of the RPA equations. It is now
easy to restore the value for $\left\langle J_{0}\right\rangle $ since $%
\sum_{m}\left\langle a_{0m}^{\dagger }a_{0m}\right\rangle =\Omega
-\sum_{m}\left\langle a_{1m}^{\dagger }a_{1m}\right\rangle $ and therefore
\begin{equation}
\left\langle J_{0}\right\rangle =-\frac{\Omega }{2}+Y^{2} \label{H19}
\end{equation}
It is interesting to realize that (\ref{H19}) corresponds to the Holstein
Primakoff boson expansion of $\left\langle J_{0}\right\rangle $
\cite{Rin80}, a result which of course is consistent with RPA theory.
Let us now repeat the same procedure but with SCRPA. Using the SCRPA
equations, in analogy to the steps above, we can write for $E_{pot}$ :
\begin{equation}
E_{pot}^{SCRPA}=-2\left\langle J_{0}\right\rangle V\frac{\widetilde{A\ }%
XY+B\ X^{2}}{E+\varepsilon } \label{H20}
\end{equation}
where $\widetilde{A}=A-\varepsilon $ and $A,B$ are determined in (\ref{ab}).
Again in cutting open the hole line we now find in analogy with (\ref{H14})
for the mass operator according to (\ref{G11})
\begin{equation}
G_{1m}^{\omega }=\frac{1}{\omega -\frac{\varepsilon }{2}}-\frac{1}{\omega -%
\frac{\varepsilon }{2}}\ V\frac{\left\langle 0\right| J_{-}\left|
1\right\rangle }{\omega +E+\frac{\varepsilon }{2}}\left[ \left\langle
0\right| J_{-}\left| 1\right\rangle \widetilde{A}+\left\langle 1\right|
J_{+}\left| 0\right\rangle B\right] \ \frac{1}{\omega -\frac{\varepsilon
}{2}%
} \label{H21}
\end{equation}
where $\left| 1\right\rangle $ again is the excited state $Q^{\dagger
}\left| 0\right\rangle $ . In the RPA limit we obtain (\ref{H14}). We
immediately check that indeed we get back from $G_{1m}$ ( and $G_{0m}$ ) the
correct expression (\ref{H20}) for $E_{pot}^{SCRPA}$ inserting (\ref{H21})
into the $lhs$ of (\ref{H15}) ( and similar for $G_{0m}$ ).
Since we now have a consistent SCRPA expression for the single particle GF
at hand we proceed, as this was our goal, to the calculation of $%
\left\langle J_{0}\right\rangle $. Inserting (\ref{H21}) into the $rhs$ of
(%
\ref{H17}) one directly obtains
\begin{eqnarray}
\left\langle J_{0}\right\rangle = &-\frac{\frac{\Omega }{2}}{1-2\left[
\widetilde{A\ }XY+B\ X^{2}\right] \frac{V}{\left( \varepsilon
+E \right) ^{2}} } \label{H22}\\
= &-\frac{\frac{\Omega }{2}}{1 + 2 XY \frac{V}{\left(
\varepsilon +E \right) } } \nonumber
\end{eqnarray}
Of course this is still an implicit equation for $\left\langle
J_{0}\right\rangle $, since the SCRPA eigenvalue $E$ depends on it.
Before proceeding it is
interesting to study several limits of (\ref{H22}). Of course for the
interaction going to zero we recover the free gas limit $\left\langle
J_{0}\right\rangle $ $=-\frac{\Omega }{2}$ . We already checked that (\ref
{H21}) goes over into the RPA limit (\ref{H14}) when $\widetilde{A}$, $B$
and the transition amplitudes are replaced by their RPA expressions.
Therefore we also recover the boson expansion result.
One should note that in order to obtain the correct RPA result one must
not make the mistake to
go over to the RPA limit, i. e. $X=1$ , $Y=0$ , $\left\langle
J_{0}^{2}\right\rangle $ $=\left\langle J_{0}\right\rangle
^{2}=\frac{\Omega
^{2}}{4}$ only in (\ref{H22}) because to get (\ref{H22}) already the
assumption has been used that $\left\langle J_{0}\right\rangle $ $\neq -%
\frac{\Omega }{2}$ on the $rhs$ of (\ref{H21}) what would not be consistent
with the RPA groundstate energy then. Now if we nevertheless take the RPA
limit, using directly (\ref{H22}), one obtains
\begin{equation}
\left\langle J_{0}\right\rangle =-\frac{\frac{\Omega
}{2}}{1+\frac{2}{\Omega
}Y^{2}} \label{H23}~~~~~~~.
\end{equation}
This result is interesting because it is precisely the lowest order result
which one obtains with the number operator method \cite{Cat96,Cat94}. In
the light of our theory this formula (\ref{H23}) seems to be inconsistent
because if on the $rhs$ of (\ref{H21}) one keeps $\left\langle
J_{0}\right\rangle $ $\neq -\frac{\Omega }{2}$ , there is no reason to
drop
all the other terms going beyond standard RPA. So in this light the pure
lowest order boson result (\ref{H19}) seems to be more consistent than the
partially resummed series (\ref{H23}). We will see later that this is indeed
confirmed by numerical results.
\subsection{Determination of $\left\langle J_{0}^{2}\right\rangle $}
In principle we are still short of the expectation value of the square of
the occupation number.
Eventually we could try to establish an analogous expression to what has
been found for $\left\langle J_{0}\right\rangle $ (\ref{H21}). However, at
least in the present model the factorization relation
\begin{equation}
\left\langle J_{0}^{2}\right\rangle \cong \left\langle J_{0}\right\rangle
^{2} \label{H24}
\end{equation}
seems to be extremely well fulfilled for the whole range of the interaction
strength considered (see next section). Of course this may be a
particularity of the model but we suppose that, as long as the operator $%
J_{0}$ or analogous operators in other problems are sufficiently
collective,
equation (\ref{H24}) should work quite reasonably. In order to check this we
present the ratio $r = - \sqrt{\left\langle J_{0}^{2}\right\rangle}
/\left\langle J_{0}\right\rangle $
for the fixed interaction strength $%
\chi = V\left( \Omega -1\right) /\varepsilon = 1.$ (i. e. at the meanfield
transition point where fluctuations are expected to be maximal) as a
function of $\Omega $ in Figure 5. The exact results are represented by
full squares, those obtained using the exact RPA vacuum (\ref{rpavac}) by
a full line (SCRPA) and a dotted line (RRPA).
Figure 5
Only for $\Omega $ values lower than 4 one can see a significant
deviation from unity. So definitely s-wave shells are difficult
candidates.
On the other hand should there be no degeneracy at all like in a rotating
nucleus or in an electron system in a magnetic field there is no need to
know the occupation number square since we have anyway
\begin{equation}
\left\langle a_{k}^{\dagger }a_{k}a_{k}^{\dagger }a_{k}\right\rangle
=\left\langle a_{k}^{\dagger }a_{k}\right\rangle \label{H25}
\end{equation}
So unless there is appearance of two fold degenerate levels in a problem one
is probably well off with the factorization (\ref{H24}). In the former case
a perturbative expansion of square operators in terms of linear operators as
proposed in \cite{Duk90} using RPA excited states as intermediate
states should adequately improve on (\ref{H24}) which represents the zero
order approximation. This approximation is based on expanding the
expectation value of any two body operator by inserting a complete set of
RPA states. Specifically for the Lipkin model we have \cite{Duk90}
\begin{equation}
\left\langle J_{0}^{2}\right\rangle =\sum_{l}\frac{\left| \left\langle
J_{0}Q^{\dagger 2l}\right\rangle \right| }{\left\langle Q^{2l}Q^{\dagger
2l}\right\rangle } \label{H251}
\end{equation}
Truncating to first order and evaluating the expectation values using the
vacuum condition we finally arrive to
\begin{equation}
\left\langle J_{0}^{2}\right\rangle =\left\langle J_{0}\right\rangle
^{2}+%
\frac{4XY\left\langle J_{0}\right\rangle ^{2}}{2\left\langle
J_{0}^{2}\right\rangle +\left( X^{2}+Y^{2}\right) \left\langle
J_{0}\right\rangle } \label{H252}
\end{equation}
a relation which expresses $\left\langle J_{0}^{2}\right\rangle $ in terms
of $\left\langle J_{0}\right\rangle .$
Let us next study the numerical results as they follow from our SCRPA
theory described above.
\section{Numerical results}
In this section we mostly will present results for $\Omega = 14$. We will
begin
in first place to investigate the quality of the results for the correlation
part of the groundstate energy, i. e. the correlation energy
\begin{equation}
E_{corr}=\left\langle H\right\rangle -\varepsilon \frac{N}{2} \label{I1}
\end{equation}
with
\begin{equation}
\left\langle H\right\rangle =\left\langle J_{0}\right\rangle \left[
\varepsilon -V\ X\ Y\right] \label{I2}
\end{equation}
Figure 6
We show $E_{corr}$ as a function of $\chi = V\left( \Omega
-1\right) /\varepsilon $ in Fig. 6 for the RPA (dashed line) ,
RRPA (small dots), SCRPA (full line) and the exact solution (full
squares). It is a very
well known fact that RPA due to the quasiboson approximation, i. e. the
violation of the Pauli principle, overestimates in general quite strongly
the correlations and in fact overbinds in the groundstate energy. This the
more, the closer one comes to the phase transition point where RPA
collapses. This strong overbinding of the RPA was for example also found in
a recent calculation \cite{Gue96} of the electronic binding
energy of a metallic cluster. When compared with the exact results SCRPA
performs extremely well for $E_{corr}$ up to and even beyond the mean
field phase transition point $\chi = 1$ whereas RRPA starts to deviate
strongly from the exact result at $\chi\cong 1$ .
Since it is not possible to distinguish in Fig. 6 that the SCRPA values
of $E_{corr}$ stays consistently above the exact ones, we also present the
results in Table 1.
Table 1
Another interesting quantity is the excitation energy. We show $%
E$ as a function of $\chi $ in Fig. 7. A similar scenario as in the
previous figure prevails: SCRPA yields
by far the best agreement with the exact results though the differences for $%
\chi \gtrsim 1$ are now more pronounced. It is also true
here
that the SCRPA
excitation energy stays consistently above the exact results as can be
seen from Table 2.
Figure 7
One
could
conclude from that that the SCRPA also leads to an upper bound for
the excitation energy. This conjecture may be backed from the fact that we
actually derived in Section 2 the SCRPA equations from a minimization
within respect an average excitation energy. However, before drawing any
definite conclusion in this respect, a more general model with more levels
must be studied.
Table 2
Let us now come to the investigation of the quality of the different
expressions for $\left\langle J_{0}\right\rangle $. There are essentially
three: the one which we prefer on theoretical grounds is the one from the
Green's function approach (\ref{H22}), since it is the only one which
fulfills a strong consistency relation with SCRPA equations (i. e. the
Hugenhotlz-van Hove theorem). The second is the quasiboson approximation (%
\ref{H19}) which represents the lowest order correction in $1/\Omega $ to
the free gas results. The third comes from the so-called number operator
expression (\ref{H23}) which has recently become very popular in the nuclear
physics literature \cite{Cat96,Cat94,Toi95}. We have shown that it is as
well obtainable from the GF approach in operating additional approximations to
(\ref{H22}), and that those approximations are not consistent among
them.
Figure 8
In Fig. 8 we show the quantity $\Omega /2+\left\langle
J_{0}\right\rangle $ as a function of $\chi $ for the three approximations
to $\left\langle J_{0}\right\rangle $ when used in the SCRPA equations
(of course only the one corresponding to GF method corresponds to our
definition proper of SCRPA). In addition we show in Fig. 8 also the exact
result (full squares). The solution of the GF method (and therefore the
SCRPA proper) is shown
by the full line. The quasiboson approximation is shown by the broken line
and the number operator method by the dotted line. Not unexpectedly the GF
results are closest to the exact ones. Somewhat a
surprise is that the number operator method works no better than the
quasiboson approximation. However in the light of our discussion in
section 4 where we
argue that one passes from the GF expression (\ref{H22}) in an essentially
uncontrolled way to the number operator expression (\ref{H23}) this outcome
may seem less astonishing.
We should also say that the injection of $\left\langle J_{0}\right\rangle
$
and $\left\langle J_{0}^{2}\right\rangle $ as expressed with the RPA
groundstate wavefunction (\ref{rpavac}) into the SCRPA equation
still improves the results in Fig. 8 with respect to GF. However, we do
not show this result in order not to overload the figure and because it
corresponds to a situation which in general is not realizable.
In Figure 9 $\Omega /2+\left\langle J_{0}\right\rangle $ is shown not as
a function of $\chi$ for fixed $\Omega $ but for
fixed $\chi = 1.$ as a function of $\Omega $ .
Figure 9
Again we see that $\Omega =2$ appears as the worst case. It is, however,
interesting to see that for this case the differences between the various
approximations are also largely enhanced without, however, inverting their
respective order.
One last interesting quantity is the ratio $Y/X$ as a function of $\chi $,
shown in Fig. 10.
It is well known that this ratio goes to 1 when approaching the phase
transition point in RPA (as seen in th broken line) while the value of
$X$ and $Y$ tend to $\infty $ individually. This then makes any RPA result
close to a phase transition
meaningless. On the other hand in SCRPA this ratio still stays of the order $%
1/2$ around the transition point and also $X$ and $Y$ remain within very
reasonable limits ($X=1.156, Y=0.580$ at $\chi = 1.$)
Figure 10
A word of caution is worth here. While the energetics and the occupation
numbers obtained with the SCRPA are very close to the exact ones, the
wave functions around and beyond $\chi =1$ (the value at which
standard RPA collapses), being far better than those obtained with RPA
or RRPA, can nontheless have an overlap with the exact wave function of
less than 50\% \cite{Hir99}. In this case the SCRPA
must be extended to the deformed basis \cite{Duk90}.
\section{Conclusions}
In this work we addressed the question of how to close the SCRPA equations
in a consistent way and, in particular, of how to calculate single
particle quantities such as occupation numbers in this formalism. We showed
in detail how to couple back SCRPA into the single particle propagator
consistently. The consistency criterion was based on the fulfillment of
the Hugenholtz-van Hove theorem which states that the chemical potential
obtained from the single particle propagator must be equal (at
equilibrium) to the energy per particle when directly calculated via the
correlation
function. For some problems (for instance in such schematic models as
considered here) there may also be correlation functions which involve the
expectation value of the square of the occupation number operator, which
fall out of the SCRPA space. We, however, showed that in general it seems to
be an excellent approximation to replace the expectation values of these
operators squared by the product of expectation values of the individual
operators. Only for the very special case of $\Omega =2$ we found that
some
caution has to prevail, though a perturbative expansion has been already
proposed (Eq. (\ref{H251})) to improve this approximation when needed.
Concerning the numerical results we found that SCRPA yields for this model
case excellent results (besides $\Omega =2$, see above). For instance we
found that groundstate as well as excited energies are always close but
consistently above the exact values. We also calculated the occupation
numbers from the
proposed form of the single particle propagator and found that they are
closest to the exact values in comparison with other proposed approximate
forms for the occupation numbers. Somewhat as a surprise comes the fact that
the so called number operator method yields results not better
than the quasiboson approximation. We give reasons which may back that this
is in fact a generic feature. One should say, however, that the numerical
differences for the occupation numbers using the different methods are, at
least for the model considered, not very pronounced.
We also should mention that it is not very difficult to obtain good
results for the Lipkin model in incorporating groundstate correlations in
one way or the other. However, at comparable numerical complexity, the
SCRPA equations do at least equally well, if not better, than any other
theory on the market. In this respect we refer the reader to our earlier
study of ref. \cite{Duk90}.
A more severe test would be to apply the present SCRPA
scheme to other more realistic models like for example the multilevel
pairing model for which, in the superfluid phase, the number operator
approximation is not anymore valid. Such studies shall be presented in
future work.
\smallskip
{\large Acknowledgments}
This work was supported in part by Conacyt (M\'{e}xico) and by the DIGICYT
(Spain) under contract No PB95/0123. J.G.H. thanks the warm hospitality
received during his visit at the IEM, CSIC, Madrid, where part of this work
was done.
|
\section{Introduction}
\setcounter{equation}{0}
At present there exists a special interest to the so-called M-
and F-theories \cite{HTW,Sc,Du,Vafa}.
These theories are ``supermembrane'' analogues of
superstring models \cite{GrSW} in $D=11,12$ etc. The low-energy limit of
these theories leads to models governed by the action
\beq{2.1i}
S = \int_{M} d^{D}z \sqrt{|g|} \{ {R}[g] - 2\Lambda
- h_{\alpha\beta}\;
g^{MN} \partial_{M} \varphi^\alpha \partial_{N} \varphi^\beta
- \sum_{a \in \Delta}
\frac{\theta_a}{n_a!} \exp[ 2 \lambda_{a} (\varphi) ] (F^a)^2_g \},
\end{equation}
where $g = g_{MN} dz^{M} \otimes dz^{N}$ is a metric,
$\varphi=(\varphi^\alpha)\in \mbox{\rm I$\!$R} ^l$
is a vector from dilatonic scalar fields,
$(h_{\alpha\beta})$ is a positively-defined symmetric
$l\times l$ matrix ($l\in \mbox{\rm I$\!$N} $), $\theta_a = \pm 1$,
\beq{2.2i}
F^a = dA^a =
\frac{1}{n_a!} F^a_{M_1 \ldots M_{n_a}}
dz^{M_1} \wedge \ldots \wedge dz^{M_{n_a}}
\end{equation}
is a $n_a$-form ($n_a \geq 1$) on a $D$-dimensional manifold $M$,
$D > 2$, $\Lambda$ is cosmological constant and $\lambda_{a}$ is a
$1$-form on $ \mbox{\rm I$\!$R} ^l$: $\lambda_{a} (\varphi) =\lambda_{a \alpha}
\varphi^\alpha$, $a \in \Delta$, $\alpha=1,\ldots,l$. In (\ref{2.1i}) we
denote $|g| = |\det (g_{MN})|$, \beq{2.3i} (F^a)^2_g = F^a_{M_1 \ldots
M_{n_a}} F^a_{N_1 \ldots N_{n_a}} g^{M_1 N_1} \ldots g^{M_{n_a} N_{n_a}},
\end{equation}
$a \in \Delta$, where $\Delta$ is some finite (non-empty) set. In
models with one time all $\theta_a = 1$ when the signature of the metric
is $(-1,+1, \ldots, +1)$.
In \cite{IMC} it was shown that after dimensional reduction on the
manifold
\beq{1.1}
M_{*}\times M_1\times\dots\times M_n
\end{equation}
with $M_i$ being Einstein space ($i = 1, \ldots, n$)
and when the composite
$p$-brane ansatz is considered
(for review see, for example, \cite{IMC,DKL,St,BREJS,AV,AR})
the problem is reduced to the gravitating
self-interacting $\sigma$-model with certain constraints imposed. For
electric $p$-branes see also \cite{IM0,IMO,IMR}
(in \cite{IMR} the composite electric case was considered).
In cosmological (or spherically symmetric) case $M_{*} = \mbox{\rm I$\!$R} $ and
the problem is effectively reduced to a
Toda-like system with the Lagrangian \cite{IMJ}
\beq{1.2}
L=\frac12 \bar G_{AB}\dot x^A\dot x^B-
\sum_{s\in S_{*}} A_s \exp(2U^s_A x^A),
\end{equation}
and the zero-energy constraint $E = 0$ imposed, where
$(\bar G_{AB})$ is a non-degenerate symmetric
$N \times N$ matrix ($N = n + l$),
$A_s \neq 0$, $x = (x^A) \in \mbox{\rm I$\!$R} ^N$, $U^s = (U^s_A) \in \mbox{\rm I$\!$R} ^N$,
$s \in S_{*}$.
The considered cosmological model contains some stringy cosmological
models (see for example \cite{LMPX}). It may be obtained (at a classical
level) from a multidimensional cosmological model with a perfect fluid
\cite{IM3,GIM} as a special case.
The integrability of the Lagrange equations corresponding to
(\ref{1.2}) crucially depends upon the scalar
products $(U^{s_1},U^{s_2})$, $s_1, s_2 \in S_{*}$,
where
\beq{1.3}
(U,U')=\bar G^{AB} U_A U'_B,
\end{equation}
$U,U'\in \mbox{\rm I$\!$R} ^N$, where $(\bar G^{AB})=(\bar G_{AB})^{-1}$.
In the orthogonal case
\beq{1.4}
(U^s,U^{s'})=0
\end{equation}
$s, s' \in S_{*}$,
a class of cosmological and spherically-symmetric solutions was obtained
in \cite{IMJ}. Special cases were also considered in
\cite{LPX,BGIM,BIM,GrIM}. Recently the ``orthogonal''
solutions were generalized to so-called ``block orthogonal''
case \cite{Br1,IMJ1}.
This paper is devoted to the investigation of the possible
oscillating (and probably stochastic) behaviour near the
singularity (see \cite{Mi}-\cite{IM} and references therein)
for cosmological models with $p$-branes.
We remind that
near the singularity one can have an oscillating behavior like in
the well-known mixmaster (Bianchi-IX) model \cite{Mi}-\cite{MTW}
(see also \cite{P}-\cite{M1}).
Multidimensional generalizations and analogues of this model
were considered by many authors (see, for
example, \cite{BK}-\cite{CDDQ}). In \cite{IKM1,IKM,IM} a billiard
representation for a multidimensional cosmological models near the
singularity was considered and a criterion for a volume of the
billiard to be finite was established in terms of illumination of the unit
sphere by point-like sources. For perfect-fluid this was considered in
detail in \cite{IM}. Some topics related to general
(non-homogeneous) situation were considered in \cite{KM}.
Here we apply the billiard approach
suggested in \cite{IKM1,IKM,IM} to a
$p$-brane cosmology. The cosmological model with $p$-branes
may be considered as a special case of a cosmological
model for multicomponent perfect fluid
with the equations of state for ``brane'' components:
$p_i^s = - \rho^s$ or
$p_i^s = \rho^s$, when brane $s$ ``lives'' or ``does not live''
in the space $M_i$ respectively, $i = 1, \ldots, n$.
The paper is organized as follows. In Sec. 2 the cosmological
model with $p$-branes is considered. Sec. 3 deals with the
Lagrange representation to equations of motion and the diagonalization
of the Lagrangian. In Sec. 4 a billiard approach in the
multidimensional cosmology with $p$-branes is developed.
A necessary condition for
the existence of oscillating (e.g. stochastic behaviour)
near the singularity is established:
\beq{1.5}
m \geq n+l,
\end{equation}
where $m$ is the number of $p$-branes.
In Sections 5 and 6 some examples of billiards
(e.g. in truncated $D=11$ supergravity etc.) are considered.
\section{\bf The model}
\setcounter{equation}{0}
Equations of motion corresponding to (\ref{2.1i}) have the following
form
\bear{2.4i}
R_{MN} - \frac{1}{2} g_{MN} R = T_{MN},
\\
\label{2.5i}
{\bigtriangleup}[g] \varphi^\alpha -
\sum_{a \in \Delta} \theta_a \frac{\lambda^{\alpha}_a}{n_a!}
e^{2 \lambda_{a}(\varphi)} (F^a)^2_g = 0,
\\
\label{2.6i}
\nabla_{M_1}[g] (e^{2 \lambda_{a}(\varphi)}
F^{a, M_1 \ldots M_{n_a}}) = 0,
\end{eqnarray}
$a \in \Delta$; $\alpha=1,\ldots,l$.
In (\ref{2.5i}) $\lambda^{\alpha}_{a} = h^{\alpha \beta}
\lambda_{a \beta }$, where $(h^{\alpha \beta})$
is a matrix inverse to $(h_{\alpha \beta})$.
In (\ref{2.4i})
\bear{2.7i}
T_{MN} = T_{MN}[\varphi,g]
+ \sum_{a\in\Delta} \theta_a e^{2 \lambda_{a}(\varphi)} T_{MN}[F^a,g],
\end{eqnarray}
where
\bear{2.8i}
T_{MN}[\varphi,g] =
h_{\alpha\beta}\left(\partial_{M} \varphi^\alpha \partial_{N} \varphi^\beta -
\frac{1}{2} g_{MN} \partial_{P} \varphi^\alpha \partial^{P} \varphi^\beta\right),
\\
T_{MN}[F^a,g] = \frac{1}{n_{a}!}\left[ - \frac{1}{2} g_{MN} (F^{a})^{2}_{g}
+ n_{a} F^{a}_{M M_2 \ldots M_{n_a}} F_{N}^{a, M_2 \ldots M_{n_a}}\right].
\label{2.9i}
\end{eqnarray}
In (\ref{2.5i}), (\ref{2.6i}) ${\bigtriangleup}[g]$ and ${\bigtriangledown}[g]$
are the Laplace-Beltrami and covariant derivative operators respectively
corresponding to $g$.
Let us consider a manifold
\beq{2.10g}
M = \mbox{\rm I$\!$R} \times M_{1} \times \ldots \times M_{n}
\end{equation}
with the metric
\beq{2.11g}
g= w \mbox{\rm e} ^{2{\gamma}(t)} dt \otimes dt +
\sum_{i=1}^{n} \mbox{\rm e} ^{2\phi^i(t)} g^i ,
\end{equation}
where $w=\pm 1$, $t$ is a distinguished coordinate which, by
convention, will be called ``time'';
$g^i = g^i_{m_{i} n_{i}}(y_i) dy_i^{m_{i}} \otimes dy_i^{n_{i}}$
is a metric on $M_{i}$ satisfying the equation
\beq{2.12g}
R_{m_{i}n_{i}}[g^i ] = \xi_{i} g^i_{m_{i}n_{i}},
\end{equation}
$m_{i},n_{i}=1,\ldots,d_{i}$; $d_{i} = \dim M_i$, $\xi_i= \mathop{\rm const}\nolimits$,
$i=1,\dots,n$; $n \in \mbox{\rm I$\!$N} $. Thus, $(M_i,g^i)$ are Einstein spaces.
The functions $\gamma,\phi^i$: $(t_-,t_+)\to \mbox{\rm I$\!$R} $ are smooth.
{\bf Remark.}
{\em It is more correct to write in (\ref{2.11g})
$\hat{g}^{i}$ instead of $g^{i}$, where $\hat{g}^{i} = p_{i}^{*} g^{i}$
is the pullback of the metric $g^{i}$ to the manifold $M$ by the
canonical projection: $p_{i} : M \rightarrow M_{i}$, $i = 1,
\ldots, n$. In what follows we omit "hats" for simplicity.}
Each manifold $M_i$ is assumed to be oriented and connected,
$i = 1,\ldots,n$. Then the volume $d_i$-form
\beq{2.13g}
\tau_i = \sqrt{|g^i(y_i)|}
\ dy_i^{1} \wedge \ldots \wedge dy_i^{d_i},
\end{equation}
and the signature parameter
\beq{2.14g}
\varepsilon (i) = \mbox{\rm sign} \det (g^i_{m_{i}n_{i}}) = \pm 1
\end{equation}
are correctly defined for all $i=1,\ldots,n$.
Let
\beq{2.15g}
\Omega_0 = \{ \emptyset, \{ 1 \}, \ldots, \{ n \},
\{ 1, 2 \}, \ldots, \{ 1, \ldots, n \} \}
\end{equation}
be a set of all subsets of
\beq{2.25n}
I_0\equiv\{ 1, \ldots, n \}.
\end{equation}
Let $I = \{ i_1, \ldots, i_k \} \in \Omega_0$, $i_1 < \ldots < i_k$.
We define a form
\beq{2.17i}
\tau(I) \equiv \tau_{i_1} \wedge \ldots \wedge \tau_{i_k},
\end{equation}
of rank
\beq{2.19i}
d(I) \equiv \sum_{i \in I} d_i,
\end{equation}
and a corresponding $p$-brane submanifold
\beq{2.18i}
M_{I} \equiv M_{i_1} \times \ldots \times M_{i_k},
\end{equation}
where $p=d(I)-1$ (${\rm dim M_{I}} = d(I)$).
We also define $ \varepsilon $-symbol
\beq{2.19e}
\varepsilon (I) \equiv \varepsilon (i_1) \ldots \varepsilon (i_k).
\end{equation}
For $I = \emptyset$ we put $\tau(\emptyset) = \varepsilon (\emptyset) = 1$,
$d(\emptyset) = 0$.
For fields of forms we adopt the following "composite electro-magnetic"
ansatz
\beq{2.27n}
F^a=
\sum_{I\in\Omega_{a,e}}{\cal F}^{(a,e,I)}+\sum_{J\in\Omega_{a,m}}{\cal F}^{(a,m,J)},
\end{equation}
where
\bear{2.28n}
{\cal F}^{(a,e,I)}=d\Phi^{(a,e,I)}\wedge\tau(I), \\ \label{2.29n}
{\cal F}^{(a,m,J)}= \mbox{\rm e} ^{-2\lambda_a(\varphi)}
*\left(d\Phi^{(a,m,J)}\wedge\tau(J)\right),
\end{eqnarray}
$a\in\Delta$, $I\in\Omega_{a,e}$, $J\in\Omega_{a,m}$ and
\beq{2.29nn}
\Omega_{a,e},\Omega_{a,m}\subset \Omega_0.
\end{equation}
(For empty $\Omega_{a,v}=\emptyset$, $v=e,m$, we put $\sum\limits_\emptyset=0$ in
(\ref{2.27n})). In (\ref{2.29n}) $*=*[g]$ is the Hodge operator on $(M,g)$.
For potentials in (\ref{2.28n}), (\ref{2.29n}) we put
\beq{2.28nn}
\Phi^s=\Phi^s(t),
\end{equation}
$s\in S$, where
\beq{6.39i}
S=S_e \sqcup S_m, \qquad
S_v \equiv \sqcup_{a\in\Delta}\{a\}\times\{v\}\times\Omega_{a,v},
\end{equation}
$v=e,m$. Here $\sqcup$ means the union of non-intersecting sets.
The set $S$ consists of elements
$s=(a_s,v_s,I_s)$, where $a_s \in \Delta$, $v_s = e,m$ and
$I_s \in \Omega_{a,v_s}$ are ``color'', ``electro-magnetic'' and
``brane'' indices, respectively.
For dilatonic scalar fields we put
\beq{2.30n}
\varphi^\alpha=\varphi^\alpha(t),
\end{equation}
$\alpha=1,\dots,l$.
From (\ref{2.28n}) and (\ref{2.29n}) we obtain
the relations between dimensions of $p$-brane
worldsheets and ranks of forms
\bear{2.d1}
d(I) = n_a - 1, \quad I \in \Omega_{a,e},
\\ \label{2.d2}
d(J) = D - n_a - 1, \quad J \in \Omega_{a,m},
\end{eqnarray}
in electric and magnetic cases respectively.
\section{\bf Lagrange representation}
\setcounter{equation}{0}
Here, like in \cite{IMJ}, we impose a restriction on
$p$-brane configurations, or, equivalently, on $\Omega_{a,v}$.
We assume that the energy momentum tensor $(T_{MN})$ has a block-diagonal
structure (as it takes place for $(g_{MN})$).
Sufficient restrictions on $\Omega_{a,v}$ that guarantee
a block-diagonality of $(T_{MN})$ are the following ones:
{\bf 1.} for any $a \in \Delta$ and $v=e,m$ there are no
$I, J \in \Omega_{a,v}$ such that
\beq{B.2}
I= \{i\}\sqcup(I\cap J), \ J=\{j\}\sqcup(I\cap J),
\quad d_i = d_j = 1, \ i \neq j;
\end{equation}
{\bf 2.} for any $a \in \Delta$ there are no
$I \in \Omega_{a,m}$ and $J \in \Omega_{a,e}$
such that
\beq{B.1}
\bar I=\{j\}\sqcup J, \quad d_j = 1,
\end{equation}
$i,j=1, \ldots, n$.
In (\ref{B.1})
\beq{B.28i}
\bar I \equiv I_0 \setminus I
\end{equation}
is a ``dual'' set ($I_0$ is defined in (\ref{2.25n})).
The restrictions (\ref{B.2}) and (\ref{B.1}) are
trivially satisfied when $n_1\le1$ and $n_1=0$ respectively, where
$n_1$ is the number of $1$-dimensional manifolds among $M_i$.
It follows from \cite{IMC} (see Proposition 2 in \cite{IMC}) that the
equations of motion (\ref{2.4i})--(\ref{2.6i}) and the Bianchi
identities
\beq{2.b}
d{\cal F}^s=0, \quad s\in S,
\end{equation}
for the field configuration (\ref{2.11g}), (\ref{2.27n})--(\ref{2.29n}),
(\ref{2.28nn}), (\ref{2.30n}) with the restrictions (\ref{B.2}),
(\ref{B.1}) imposed are equivalent to equations of motion for
Lagrange system with the Lagrangian
\bear{2.25gn}
L = \frac{1}{2}
{\cal N}^{-1} \biggl\{G_{ij}\dot\phi^i\dot\phi^j
+h_{\alpha\beta}\dot\varphi^{\alpha}\dot\varphi^{\beta}
+ \sum_{s\in S} \varepsilon _s\exp[-2U^s(\phi,\varphi)](\dot\Phi^s)^2
-2{\cal N}^{2}V(\phi)\biggr\},
\end{eqnarray}
where $\dot x\equiv dx/dt$,
\beq{2.27gn}
V = {V}(\phi) =
(-w\Lambda) \mbox{\rm e} ^{2{\gamma_0}(\phi)} +
\frac w2\sum_{i =1}^{n} \xi_i d_i
\mbox{\rm e} ^{-2 \phi^i + 2 {\gamma_0}(\phi)}
\end{equation}
is a potential with
\beq{2.24gn}
\gamma_0 = \gamma_0(\phi)
\equiv\sum_{i=1}^nd_i\phi^i, \label{2.32g}
\end{equation}
and
\beq{2.24gn1}
{\cal N}=\exp(\gamma-\gamma_0)>0
\end{equation}
is the lapse function,
\bear{2.u}
U^s = U^s(\phi,\varphi)= -\chi_s\lambda_{a_s}(\varphi) +
\sum_{i\in I_s}d_i\phi^i, \\ \label{2.e}
\varepsilon _s=(- \varepsilon [g])^{(1-\chi_s)/2} \varepsilon (I_s)\theta_{a_s}
\end{eqnarray}
for $s=(a_s,v_s,I_s)\in S$, $ \varepsilon [g]= \mbox{\rm sign} \det (g_{MN})$,
(more explicitly, (\ref{2.e}) reads: $ \varepsilon _s= \varepsilon (I_s) \theta_{a_s}$ for
$v_s = e$ and $ \varepsilon _s=- \varepsilon [g] \varepsilon (I_s) \theta_{a_s}$ for
$v_s = m$)
\bear{2.x1}
\chi_s=+1, \quad v_s=e; \\ \label{2.x2}
\chi_s=-1, \quad v_s=m,
\end{eqnarray}
and
\beq{2.c}
G_{ij}=d_i\delta_{ij}-d_id_j
\end{equation}
are components of a ``pure cosmological'' minisupermetric;
$i,j=1,\dots,n$ \cite{IMZ}.
Let $x=(x^A)=(\phi^i,\varphi^\alpha)$,
\bear{2.35n}
\bar G=\bar G_{AB}dx^A\otimes dx^B=G_{ij}d\phi^i\otimes d\phi^j+
h_{\alpha\beta}d\varphi^\alpha\otimes d\varphi^\beta, \\ \label{2.36n}
(\bar G_{AB})=\left(\begin{array}{cc}
G_{ij}&0\\
0&h_{\alpha\beta}
\end{array}\right),
\end{eqnarray}
$U^s(x)=U_A^sx^A$ is defined in (\ref{2.u}) and
\beq{2.38n}
(U_A^s)=(d_i\delta_{iI_s},-\chi_s\lambda_{a_s\alpha}).
\end{equation}
Here
\beq{2.39n}
\delta_{iI}\equiv\sum_{j\in I}\delta_{ij}=\begin{array}{ll}
1,&i\in I\\
0,&i\notin I
\end{array}
\end{equation}
is an indicator of $i$ belonging to $I$. The potential (\ref{2.27gn})
reads
\beq{2.40n}
V= (-w\Lambda) \mbox{\rm e} ^{2U^\Lambda(x)} +
\sum_{j=1}^n\frac w2\xi_jd_j \mbox{\rm e} ^{2U^j(x)},
\end{equation}
where
\bear{2.41n}
U^j(x)=U_A^jx^A=-\phi^j+\gamma_0(\phi),
\\ \label{2.42n}
U^\Lambda(x)=U_A^\Lambda x^A=\gamma_0(\phi),
\\ \label{2.43n}
(U_A^j)=(-\delta_i^j+d_i,0)
\\ \label{2.44n}
(U_A^\Lambda)=(d_i,0).
\end{eqnarray}
The integrability of the Lagrange system (\ref{2.25gn}) depends
upon the scalar products of the co-vectors $U^{\Lambda}$, $U^j$, $U^s$
corresponding to $\bar G$:
\beq{2.45n}
(U,U')=\bar G^{AB}U_AU'_B,
\end{equation}
where
\beq{2.46n}
(\bar G^{AB})=\left(\begin{array}{cc}
G^{ij}&0\\
0&h^{\alpha\beta}
\end{array}\right)
\end{equation}
is a matrix inverse to (\ref{2.36n}). Here (as in \cite{IMZ})
\beq{2.47n}
G^{ij}=\frac{\delta^{ij}}{d_i} + \frac1{2-D},
\end{equation}
$i,j=1,\dots,n$. These products have the following form
\cite{IMC}
\bear{2.48n}
(U^i,U^j)=\frac{\delta_{ij}}{d_j}-1,
\\ \label{2.50n}
(U^\Lambda,U^\Lambda)=-\frac{D-1}{D-2},
\\ \label{2.51n}
(U^s,U^{s'})=
d(I_s \cap I_{s'})+\frac{d(I_s)d(I_{s'})}{2-D}
+\chi_s \chi_{s'} \lambda_{a\alpha}\lambda_{b\beta}
h^{\alpha\beta},
\\ \label{2.52n}
(U^s,U^i)=-\delta_{iI_s},
\\ \label{2.53n}
(U^{\Lambda}, U^i)=- 1,
\\ \label{2.54n}
(U^{\Lambda}, U^s)= \frac{d(I_s)}{2-D},
\end{eqnarray}
where $s=(a_s,v_s,I_s)$, $s'=(a_{s'},v_{s'},I_{s'})\in S$.
First we integrate the ``Maxwell equations'' (for $s\in S_e$)
and Bianchi identities (for $s\in S_m$):
\bear{5.29n}
\frac d{dt}\left(\exp(-2U^s)\dot\Phi^s\right)=0
\Longleftrightarrow
\dot\Phi^s=Q_s \exp(2U^s),
\end{eqnarray}
where $Q_s$ are constants, $s \in S$. We put
\bear{5.30n}
Q_s \ne 0,
\end{eqnarray}
for all $s \in S$.
For fixed charges
$Q=(Q_s, s \in S)$ Lagrange equations for the Lagrangian
(\ref{2.25gn}) corresponding to $(x^A)=(\phi^i,\varphi^\alpha)$,
(when relations (\ref{5.29n})
are substituted) are equivalent to
Lagrange equations for the Lagrangian \cite{IMJ}
\beq{5.31n}
L_Q=\frac12 {\cal N}^{-1} \bar G_{AB} \dot x^A\dot x^B-
{\cal N} V_Q.
\end{equation}
where
\beq{5.32n}
V_Q=V+\frac12\sum_{s\in S} \varepsilon _sQ_s^2\exp[2U^s(x)],
\end{equation}
$(\bar G_{AB})$ and $V$ are defined in (\ref{2.36n}) and (\ref{2.40n})
respectively.
\subsection{ Diagonalization of the Lagrangian}
The minisuperspace metric (\ref{2.35n})
has a pseudo-Euclidean signature $(-,+, \ldots ,+)$,
since the matrix $(G_{ij})$ has the pseudo-Euclidean
signature, and $(h_{\alpha \beta})$ has the Euclidean one.
Hence there exists
a linear transformation
\beq{2.21o}
z^{a}=e^{a}_{A}x^{A},
\end{equation}
diagonalizing the minisuperspace metric (\ref{2.35n})
\beq{2.22o}
\bar G= \eta_{ab}dz^{a} \otimes dz^{b}=
-dz^{0} \otimes dz^{0} + \sum_{k=1}^{N-1}dz^{k}\otimes dz^{k},
\end{equation}
where
\beq{2.23o}
(\eta_{ab})=(\eta^{ab}) \equiv diag(-1,+1, \ldots ,+1),
\end{equation}
and here and in what follows
$a,b = 0, \ldots ,N-1$; $N =n+l$.
The matrix of linear transformation
$(e^{a}_{A})$ satisfies the relation
\beq{2.24o}
\eta_{ab} e^{a}_{A} e^{b}_{B} = \bar{G}_{AB}
\end{equation}
or, equivalently,
\beq{2.25o}
\eta^{ab} = e^{a}_{A}\bar{G}^{AB} e^{b}_{B} = (e^{a},e^{b}),
\end{equation}
where $e^a = (e^a_A)$.
Inverting the map (\ref{2.21o}) we get
\beq{2.28o}
x^{A} = e_{a}^{A} z^{a},
\end{equation}
where for components of the inverse matrix
$(e_{a}^{A}) = (e^{a}_{A})^{-1}$ we obtain from (\ref{2.25o})
\beq{2.29o}
e_{a}^{A} = \bar{G}^{AB} e^{b}_{B} \eta_{ba}.
\end{equation}
Like in \cite{IM} we put
\beq{2.30o}
e^{0} = q^{-1} U^{\Lambda}, \qquad
q = [(D-1)/(D-2)]^{1/2}= [- (U^\Lambda,U^\Lambda)]^{1/2}.
\end{equation}
and hence
\beq{2.31o}
z^0 = e^{0}_{A} x^A = \sum_{i=1}^{n} q^{-1} d_i x^i.
\end{equation}
In $z$-coordinates (\ref{2.21o}) with $z^0$ from
(\ref{2.31o}) the Lagrangian (\ref{5.31n}) reads
\beq{2.32o}
L_Q = {L}_Q(z^{a}, \dot{z}^{a}, {\cal N})
= \frac{1}{2} {\cal N}^{-1}
\eta_{ab} \dot{z}^{a} \dot{z}^{b} - {\cal N} {V}(z),
\end{equation}
where
\beq{2.34o}
{V}(z) =
\sum_{r \in S_{*}} A_{r} \exp(2 U^{r}_a z^a)
\end{equation}
is a potential,
\beq{2.34oa}
S_{*} = \{ \Lambda \} \cup \{ 1, \ldots, n \} \cup S
\end{equation}
is an index set and
\beq{2.34ob}
A_{\Lambda} = -w \Lambda, \quad A_j = \frac{w}{2} \xi_j d_j, \quad
A_s = \frac{1}{2} \varepsilon _s Q_s^2,
\end{equation}
$j = 1, \ldots, n$; $s \in S$.
Here we denote
\beq{2.35o}
U^{r}_a = e_{a}^{A} U^{r}_A =
(U^{r}, e^{b}) \eta_{ba},
\end{equation}
$a = 0, \ldots , N-1$; $r \in S_{*}$ (see (\ref{2.29o})).
From (\ref{2.45n})-(\ref{2.47n}), (\ref{2.30o}) and (\ref{2.35o})
we deduce
\beq{2.36o}
U^{r}_0 = - (U^{r}, e^{0}) =
(\sum_{i=1}^{n} U^{r}_i ) / q(D-2),
\end{equation}
$r \in S_*$.
For the $\Lambda$-term and curvature-term components
we obtain from (\ref{2.30o}) and (\ref{2.36o})
\beq{2.37o}
U^{\Lambda}_0 = q > 0 , \qquad U^{j}_0 = 1/q > 0,
\end{equation}
$j= 1,\ldots,n$.
For ``brane'' components we get
from (\ref{2.38n}), (\ref{2.30o}) and (\ref{2.36o})
\beq{2.38o}
U^{s}_0 = d(I_s)/\sqrt{(D-2)(D-1)} > 0.
\end{equation}
We remind that (see (\ref{2.48n}) and (\ref{2.50n}))
that
\beq{2.39o}
(U^{\Lambda},U^{\Lambda}) = (D-1)/(2-D) < 0, \qquad (U^{j},U^{j}) =
(\frac{1}{d_j} - 1) < 0,
\end{equation}
for $d_j > 1$, $j= 1,\ldots,n$. For $d_j = 1$ we have $\xi^{j}
= A_{j} = 0$.
\section{Billiard representation}
\setcounter{equation}{0}
Here we put the following restrictions on parameters of the model:
\bear{4.1n}
{\bf (i)} \qquad \qquad \varepsilon _s = + 1,
\\ \label{4.2n}
{\bf (ii)} \quad d(I_s) < D-2,
\end{eqnarray}
$s \in S$.
For $\theta_a = 1$, $a \in \Delta$, and $ \varepsilon [g] = -1$,
the first restriction means that all $ \varepsilon (I_s) = 1$, $s \in S$, i.e.
all $p$-branes are either Euclidean or contain even number of
``times''. Restriction ${\bf (ii)}$ implies
\beq{4.3}
(U^s,U^{s})=
d(I_s ) \left(1 +\frac{d(I_s)}{2-D} \right)
+ \lambda_{a_s}^2 >0,
\end{equation}
where $\lambda^2 = \lambda_{\alpha} \lambda_{\beta}
h^{\alpha\beta}$, $D>2$. As we shall see below, both restrictions
are necessary for a formation of potential walls
in the Lobachevsky space when a certain asymptotic in time variable
is considered.
Here we consider a behaviour of the dynamical system, described
by the Lagrangian (\ref{2.32o}) with the potential (\ref{2.34o})
for $N \geq 3$ in the limit
\beq{3.1o}
z^0 \rightarrow -\infty, \qquad z =(z^0, \vec{z}) \in {\cal V}_{-},
\end{equation}
where ${\cal V}_{-}
\equiv \{(z^0, \vec{z}) \in \mbox{\rm I$\!$R} ^N | z^0 < - |\vec{z}| \}$
is the lower light cone. For the volume scale factor
\beq{3.2o}
v = \exp(\sum_{i=1}^{n} d_{i}x^{i}) = \exp(qz^0)
\end{equation}
($q > 0$) we have in this limit $v \rightarrow + 0$.
Under certain additional assumptions the limit (\ref{3.1o}) describes
an approaching to the singularity.
Due to relations
(\ref{2.34ob}), (\ref{2.37o})-(\ref{2.39o}), (\ref{4.1n}) and (\ref{4.3})
the parameters $U^r$ in the potential (\ref{2.34o})
obey the following restrictions:
\bear{3.3o}
&& {\bf 1.} \ A_r > 0 \
{\rm for} \ (U^r)^2 = -(U^r_0)^2 + (\vec{U}^r)^2 > 0;
\\ \label{3.4o}
&& {\bf 2.} \ U^r_0 > 0 \ {\rm for \ all} \ r \in S_{*}.
\end{eqnarray}
Now we restrict the Lagrange system (\ref{2.32o}) on ${\cal V}_{-}$, i.e.
we consider the Lagrangian
\beq{3.5o}
L_{-} \equiv L|_{TM_{-}} , \qquad M_{-} = {\cal V}_{-} \times \mbox{\rm I$\!$R} _{+},
\end{equation}
where $ TM_{-}$ is a tangent vector bundle over $M_{-}$ and
$ \mbox{\rm I$\!$R} _{+} \equiv \{ {\cal N} > 0 \}$. (Here $F|_{A}$ means a restriction
of function $F$ on $A$.)
Introducing an analogue of the Misner-Chitre coordinates in
$\cal{V}_{-}$ [36,37] which reduce the problem to a unit disk
\bear{3.6o}
&&z^0 = - \exp(-y^0) \frac{1 + \vec{y}^2}{1 - \vec{y}^2}, \\
&&\vec{z} = - 2 \exp(-y^0) \frac{ \vec{y}}{1 - \vec{y}^2},
\end{eqnarray}
$|\vec{y}| < 1$, we get for the Lagrangian (\ref{2.32o})
\beq{3.8o}
L_{-} = \frac{1}{2} {\cal N}^{-1} e^{- 2 y^0}
[- (\dot{y}^{0})^2 + \bar{h}_{ij}(\vec{y}) \dot{y}^{i} \dot{y}^{j}]
- {\cal N} V.
\end{equation}
Here
\beq{3.9o}
\bar{h}_{ij}(\vec{y}) = 4 \delta_{ij} (1 - \vec{y}^2)^{-2},
\end{equation}
$i,j =1, \ldots , N-1$, and
\beq{3.10o}
V = {V}(y) =
\sum_{r \in S_*} A_r \exp[\bar{\Phi}(y,2U^r)],
\end{equation}
where
\beq{3.11o}
{\bar{\Phi}}(y,u) \equiv - e^{-y^0}(1 - \vec{y}^2)^{-1}
[u_0 (1 + \vec{y}^2) + 2 \vec{u}\vec{y}],
\end{equation}
We note that the $(N-1)$-dimensional open unit disk (ball)
\beq{3.12o}
D^{N-1} \equiv \{ \vec{y}= (y^1, \ldots, y^{N - 1})| |\vec{y}| < 1 \}
\subset \mbox{\rm I$\!$R} ^{N-1}
\end{equation}
with the metric $\bar{h} = \bar{h}_{ij}(\vec{y})
dy^i \otimes dy^j $ is one
of the realization of the $(N-1)$-dimensional Lobachevsky space
$H^{N-1}$.
We fix the gauge
\beq{3.13o}
{\cal N} = \exp(- 2y^0) = - z^2.
\end{equation}
Then, it is not difficult to verify that the
Lagrange equations for the Lagrangian (\ref{3.8o}) with the gauge fixing
(\ref{3.3o}) are equivalent to the Lagrange equations for the
Lagrangian
\beq{3.14o}
L_{*} = - \frac{1}{2} (\dot{y}^{0})^2 + \frac{1}{2}
\bar{h}_{ij}(\vec{y}) \dot{y}^{i} \dot{y}^{j} - V_{*}
\end{equation}
with the energy constraint imposed
\beq{3.15o}
E_{*}
= - \frac{1}{2} (\dot{y}^{0})^2 + \frac{1}{2}
\bar{h}_{ij}(\vec{y}) \dot{y}^{i} \dot{y}^{j} + V_{*} = 0.
\end{equation}
Here
\beq{3.16o}
V_{*} = e^{-2y^0} V =
\sum_{r \in S_*} A_{\alpha} \exp[\Phi(y,2U^r)],
\end{equation}
where
\beq{3.17o}
{\Phi}(y,u) = - 2y^0 + \bar{\Phi}(y,u).
\end{equation}
Now we are interested in a behavior of the dynamical system
in the limit $y^0 \rightarrow - \infty$ (or, equivalently, in
the limit $z^2 = -(z^0)^2 + (\vec{z})^2
\rightarrow - \infty$, $z^0 < 0$) implying (\ref{3.1o}).
Using the relations ($U_0 \neq 0$ )
\bear{3.18o}
&&{\Phi}(y,2U) = - 2 U_0 \exp(-y^0)
\frac{{A}(\vec{y}, -\vec{U}/U_0)}{1 - \vec{y}^2} - 2y^0, \\
&& {A}(\vec{y}, \vec{v}) \equiv
(\vec{y} - \vec{v})^2 -\vec{v}^2 + 1,
\end{eqnarray}
we get
\beq{3.20o}
\lim_{y^0 \rightarrow - \infty} \exp {\Phi}(y, 2U) = 0
\end{equation}
for $U^2 = - (U_0)^2 + (\vec{U})^2 \leq 0$, $U_0 > 0$ and
\beq{3.21o}
\lim_{y^0 \rightarrow - \infty} \exp {\Phi}(y,U) =
{\theta_{\infty}}(-{A}(\vec{y}, - \vec{U}/U_0))
\end{equation}
for $U^2 > 0$, $U_0 > 0$. In (\ref{3.21o}) we denote
\bear{3.22o}
{\theta_{\infty}}(x) \equiv + &\infty, &x \geq 0, \nonumber \\
& 0 , &x < 0.
\end{eqnarray}
Using relations (\ref{3.3o}), (\ref{3.4o}) and relations (\ref{3.16o}),
(\ref{3.20o}), (\ref{3.21o}) we obtain
\beq{3.23o}
V_{\infty}(\vec{y}) \equiv
\lim_{y^0 \rightarrow - \infty} {V_{*}}(y^0, \vec{y}) = \sum_{s \in
S_{*}} {\theta_{\infty}}(-{A}(\vec{y}, -\vec{U^s}/U_0^s)).
\end{equation}
The potential $V_{\infty}$ may be written as follows
\bear{3.25o}
{V_{\infty}}(\vec{y}) =
{V}(\vec{y},B) \equiv &0, &\vec{y} \in B,
\nonumber \\
&+ \infty, &\vec{y}
\in D^{N-1} \setminus B,
\end{eqnarray}
where
\beq{4.4}
B = \bigcap_{s \in S} {B}_{s} \subset D^{N-1},
\end{equation}
\beq{4.5}
{B}_{s} = \{ \vec{y} \in D^{N-1} |
|\vec{y} - \vec{v}^{s}| > r_{s} \},
\end{equation}
where
\beq{4.6}
\vec{v}^{s} = - \vec{U}^s/U^s_0,
\end{equation}
($|\vec{v^s}| > 1$) and
\beq{4.7}
r_s = \sqrt{(\vec{v}^s)^2 - 1},
\end{equation}
$s \in S$. Remind that $(U^s_a) = (U^s_0, \vec{U}^s)$ is defined by
relation (\ref{2.35o}).
$B$ is an open domain.
Its boundary $\partial B = \bar{B} \setminus B$ is formed by
certain parts of $m = |S|$ $(N-2)$-dimensional
spheres with the centers in the points $\vec{v}^s$
($|\vec{v^{\alpha}}| > 1$) and radii $r_s$, $s \in S$.
So, in the limit $y^{0} \rightarrow - \infty$ we are led to the
dynamical system
\bear{3.30o}
&L_{\infty} = - \frac{1}{2} (\dot{y}^{0})^2 + \frac{1}{2}
\bar{h}_{ij}(\vec{y}) \dot{y}^{i} \dot{y}^{j} - {V_{\infty}}(\vec{y}), \\
&E_{\infty} = - \frac{1}{2} (\dot{y}^{0})^2 + \frac{1}{2}
\bar{h}_{ij}(\vec{y}) \dot{y}^{i} \dot{y}^{j} + {V_{\infty}}(\vec{y}) = 0,
\end{eqnarray}
which after the separating of $y^0$ variable
\beq{3.32o}
y^0 = \omega (t - t_0),
\end{equation}
($\omega \neq 0$ , $t_0$ are constants) is reduced to the Lagrange
system with the Lagrangian
\beq{3.33o}
L_{B} = \frac{1}{2} \bar{h}_{ij}(\vec{y})
\dot{y}^{i} \dot{y}^{j} - {V}(\vec{y},B).
\end{equation}
Due to (\ref{3.32o})
\beq{3.34o}
E_{B} = \frac{1}{2}
\bar{h}_{ij}(\vec{y}) \dot{y}^{i} \dot{y}^{j} + {V}(\vec{y},B) =
\frac{\omega^2}{2}.
\end{equation}
We put $\omega > 0$, then the limit $t \rightarrow - \infty$ describes
an approach to the singularity.
When $S \neq \emptyset$ the Lagrangian
(\ref{3.33o}) describes a motion of a particle of unit mass,
moving in the ($N-1$)-dimensional billiard $B \subset D^{N-1}$ (see
(\ref{4.4})). The geodesic motion in $B$
corresponds to a ``Kasner epoch'' and the reflection from the boundary
corresponds to the change of Kasner epochs \cite{IM}.
Let the billiard $B$ has an infinite volume: ${\rm vol} B = +\infty$ and
there are open zones at the infinite sphere
$|\vec{y}| =1$. After a finite number of reflections from the boundary
a particle moves towards one of these open zones.
In this case for a
corresponding cosmological model we get the ``Kasner-like''
behavior in the limit $t \rightarrow - \infty$ and the absence
of a stochastic behaviour.
Let ${\rm vol} B < + \infty$. There are two possibilities
in this case: i) the
closure of the billiard $\bar{B}$ is
compact (in the topology of $D^{N-1}$) ; ii) $\bar{B}$ is non-compact.
In these two cases the motion of a particle is oscillating.
In \cite{IM} we proposed the simple geometric criterion for
finiteness of volume of $B$ and compactness of $\bar{B}$
in terms of the positions of the
points (\ref{4.6}) with respect to
the ($N-2$)-dimensional unit sphere $S^{N-2}$
($N \geq 3$).
{\bf Definitions.}
{\em A point $\vec{n} \in S^{k}$ ($k \geq 1$) is
illuminated by a point-like source located at a point
$\vec{v} \in \mbox{\rm I$\!$R} ^{k+1}$, $|\vec{v}| > 1$,
if and only if $|\vec{n} - \vec{v}| \leq \sqrt{|\vec{v}|^2 -1}$.
A point $\vec{n} \in S^{k}$ is
strongly illuminated by a point-like source located at the point
$\vec{v} \in \mbox{\rm I$\!$R} ^{k+1}$, $|\vec{v}| > 1$, if and only if $|\vec{n} -
\vec{v}| < \sqrt{|\vec{v}|^2 -1}$. The subset $P \subset S^{k}$ is
called (strongly) illuminated by a point-like sources at $\{
\vec{v}^{\alpha}, \alpha \in A \}$ if and only if any point from $P$ is
(strongly) illuminated by some source at $\vec{v}^{\alpha}$, $\alpha \in A
$. }
{\bf Proposition \cite{IM}.}
{\em The billiard $B$ (\ref{4.4}) has a finite
volume if and only if point-like
sources of light located at the points $\vec{v}^{\alpha}$ (\ref{4.6})
illuminate the unit sphere $S^{N-2}$.
The closure of a billiard $\bar{B}$ is compact
(in the topology of $D^{N-1} \simeq H^{N-1}$) if and only if
sources at points (\ref{4.6}) strongly illuminate $S^{N-2}$.}
We remind, that the problem of illumination of a convex body in a
multidimensional vector space by point-like sources for the first time was
considered in \cite{Sol,BG}. For the case of $S^{N-2}$ this problem is
equivalent to the problem of covering spheres with spheres
\cite{Fe,R}.
There exists a topological bound on a number of point-like
sources $m$ illuminating the sphere $S^{N-2}$
\cite{BG}:
\beq{3.40o}
m \geq N.
\end{equation}
Thus, we obtain the restriction (\ref{1.5}).
According to this restriction the number of $p$-branes $m =|S|$
should at least exceed the critical value $N = n+l$ for
the existence of oscillating (e.g. stochastic) behaviour
near the singularity.
We remind that Kasner-like solutions
have the following form
\bear{4.8}
&&g = w d\tau \otimes d\tau + \sum_{i=1}^{n} A_i \tau^{2 \alpha^i} g^i,
\\ \label{4.9}
&&\varphi^{\beta} = \alpha^{\beta} \ln \tau + \varphi^{\beta}_0,
\\ \label{4.10}
&& \sum_{i=1}^{n} d_i \alpha^i =
\sum_{i=1}^{n} d_i (\alpha^i)^2 +
\alpha^{\beta} \alpha^{\gamma} h_{\beta \gamma}= 1,
\\ \label{4.11}
&& F^a= 0
\end{eqnarray}
where $A_i > 0$, $\varphi^{\beta}_0$ are constants
$i = 1, \ldots, n$; $\beta, \gamma = 1, \ldots, l$;
$a \in \Delta$. These solutions correspond to zero
$p$-brane charges.
If the vector of Kasner parameters $\alpha = (\alpha^{A}) =
(\alpha^{i}, \alpha^{\gamma})$ obeys the relations
\beq{4.12}
U^s(\alpha) = U_A^{s} \alpha^A = \sum_{i\in I_s}d_i\alpha^i
-\chi_s\lambda_{a_s \gamma}\alpha^{\gamma} > 0,
\end{equation}
$s \in S$, then the field configuration (\ref{4.8})-(\ref{4.11})
is the asymptotical (attractor) solution for a family of (exact)
solutions with non-zero charges: $Q_s \neq 0$, when $\tau \to +0$.
Relations (\ref{4.12}) may be easily
understood using relations (\ref{5.29n}).
Indeed, from (\ref{5.29n}) and zero value limits for forms
$F^a$, $a \in \Delta$, we get
\beq{4.12b}
\exp[2U^s(x)] = C_s \tau^{2U^s(\alpha)} \to 0, \quad C_s \neq 0,
\end{equation}
for $\tau \to +0$. These relations imply (\ref{4.12}).
Now we give a rigorous explanation of (\ref{4.12}).
Let us denote by $K $ a
set of Kasner vector parameters $\alpha =(\alpha^{A}) \in \mbox{\rm I$\!$R} ^{N}$
satisfying (\ref{4.10}). $K$ is an ellipsoid isomorphic to $S^{N-2}$.
The isomorphism is defined by the relations
\beq{4.13}
\alpha^A = e^A_a n^a / q, \quad (n^a) = (1, \vec{n}),
\quad \vec{n} \in S^{N-2}.
\end{equation}
Here we use the diagonalizing matrix $(e^A_a)$ and the parameter $q$
defined in subsection 3.1 (see (\ref{2.30o})).
{\bf Proposition 1.} {\em Let us consider a point-like source
located at a point $\vec{v}^s$ from (\ref{4.6}).
A point $\vec{n} \in S^{N-2}$ is illuminated
by this source if and only if the corresponding Kasner vector
$\alpha =(\alpha^{A}) \in K$ defined by (\ref{4.13})
satisfies the relation
\beq{4.14}
U^s(\alpha) = U_A^{s} \alpha^A =
\sum_{i\in I_s}d_i \alpha^i
-\chi_s \lambda_{a_s \gamma} \alpha^{\gamma} \leq 0.
\end{equation}
}
{\bf Proof.}
A point $\vec{n} \in S^{N-2}$ is illuminated
by a point-like source of light located at a point $\vec{v}^s$,
$|\vec{v}^s| > 1$, if and only if
\beq{4.15}
\vec{n} \vec{v}^s \geq 1,
\end{equation}
$s \in S$.
For $ \vec{v}^s = - \vec{U}^s/U_{0}^s$, $U^s_{0} >0$, this inequality
may be rewritten as
\beq{4.16}
n^a U_a^s = q e_A^a \alpha^A U_a^s = q \alpha^A U_A^s \leq 0,
\end{equation}
$s \in S$. Thus, we obtain the relation (\ref{4.14}).
{\bf Corollary.}
{\em A point $\vec{n} \in S^{N-2}$ is not illuminated
by a source at a point $\vec{v}^s$ from (\ref{4.6})
if and only the relation (\ref{4.12}) is satisfied.}
A small modification of the Proposition 1
is the following one.
{\bf Proposition 1a.}
{\em A point $\vec{n} \in S^{N-2}$ is
strongly illuminated by a source at $\vec{v}^s$
if and only if $U^s(\alpha) < 0$.}
Due to Proposition 1 the criterion of the finiteness of
a billiard volume (see Proposition)
may be reformulated
in terms of inequalities on Kasner-like parameters \cite{IM}.
{\bf Proposition 2.}
{\em Billiard $B$ (\ref{4.4}) has a finite volume if and only if
there are no $\alpha$ satisfying the
relations (\ref{4.10}) and (\ref{4.12}).}
The positions of sources are defined (up to $O(N-1)$-rotation)
by scalar products
\bear{4.8b}
\vec{v}^{s} \vec{v}^{s'} =
\frac{\vec{U}^{s} \vec{U}^{s'}}{ U^{s}_0 U^{s'}_0} = \qquad
\\ \nonumber
= 1 + \frac{(D-2)(D-1)}{d(I_s) d(I_{s'})} \left[
d(I_s \cap I_{s'})+\frac{d(I_s)d(I_{s'})}{2-D}
+\chi_s \chi_{s'} \lambda_{a\alpha} \lambda_{b\beta}
h^{\alpha\beta} \right].
\end{eqnarray}
Thus, we obtained a billiard representation
for the model under consideration when the
restrictions (\ref{4.1n}) and (\ref{4.2n}) are
imposed.
Now we relax the first restriction,
i.e. we put
\beq{4.1m}
\varepsilon _s = -1,
\end{equation}
for some $s \in S$. Relation (\ref{4.1m}) occurs when
spherically symmetric solutions with p-branes are considered
\cite{IMJ,IMJ1}. In this case we may obtain ``waterfall
potentials'' with $V = - \infty$ instead of $V = + \infty$
inside of ``walls''. The ``waterfall potentials'' prevent
the oscillating behaviour near the singularity but meanwhile
do not forbid the existence of solutions with Kasner-like
asymptotical behaviour (if there are open shadow zones).
Let us consider the following example:
\beq{4.1r}
1 \in I_s, \qquad d_1 =1,
\end{equation}
for all $s \in S$, i.e. all ``branes'' overlap the one-dimensional
space $M_1$. In this case the Kasner set
\beq{4.1o}
\alpha^1 = 1, \quad \alpha^i = \alpha^{\beta} = 0,
\end{equation}
$i =2, \ldots, n$; $\beta =1, \ldots, l$, satisfies
(\ref{4.10}) and (\ref{4.12}).
For any $ \varepsilon _s = \pm 1$, $s \in S$,
we get a family of solutions with asymptotical
Kasner-like behaviour with parameters $\alpha=(\alpha^A)$
belonging to an open neighbourhood of the (Milne) point
from (\ref{4.1o}).
\section{Examples of two-dimensional billiards}
In this section we give several examples of two-dimensional
billiards with finite areas
that occur in the models under consideration.
\subsection{Billiard is a square}
Here we consider a model defined on the manifold
\beq{4.8a}
M = \mbox{\rm I$\!$R} \times M_{1} \times M_{2}
\end{equation}
governed by the Lagrangian
\beq{4.9a}
{\cal L}= {R}[g] - 2\Lambda -
g^{MN} \partial_{M} \varphi \partial_{N} \varphi
- \frac{1}{n_1!} \exp[ 2 \lambda_{1} \varphi ] (F^1)^2_g
- \frac{1}{n_2!} \exp[ 2 \lambda_{2} \varphi ] (F^2)^2_g,
\end{equation}
where $d_{1} = \dim M_1 = d$, $d_{2} = \dim M_2 = d$, $D = 1+ 2d$,
$n_1 = d+1$, $n_2 = d$, $d \geq 2$,
$w =-1$ and $ \varepsilon (1) = \varepsilon (2) = 1$.
Let
\beq{4.9b}
s_1 = (1,e,\{1 \}),\quad s_2 = (1,e, \{2 \}), \quad
s_3 = (2,m,\{1 \}), \quad s_4 = (2,m,\{2 \}),
\end{equation}
i.e. we have two electric branes corresponding to the form $F^1$,
and two magnetic branes corresponding to $F^2$. Branes
$s_1$ and $s_3$ ``live'' in $M_1$ and branes
$s_2$ and $s_4$ ``live'' in $M_2$.
We put
\beq{4.10b}
\lambda_{1} = \lambda_{2} = \lambda, \qquad \lambda^2 = d/2.
\end{equation}
Then from (\ref{4.8b}) we get
\beq{4.10c}
(\vec{v}^{s_i})^2 = - \vec{v}^{s_1} \vec{v}^{s_4} =
- \vec{v}^{s_2} \vec{v}^{s_3} = 2 (2d-1),
\end{equation}
$i = 1,2,3,4$,
and
\beq{4.11a}
\vec{v}^{s_1} \vec{v}^{s_2} =
\vec{v}^{s_1} \vec{v}^{s_3} =
\vec{v}^{s_2} \vec{v}^{s_4} =
\vec{v}^{s_3} \vec{v}^{s_4} = 0.
\end{equation}
\begin{figure}[cht]
\begin{center}
\epsfig{file=pic1.eps,height=9cm,width=9cm}
\end{center}
\caption{
Square billiard in the 5-dimensional model with two internal spaces,
scalar field and four ``branes'' (two electric and two magnetic).}
\end{figure}
This means that the points $\vec{v}^{s_i}$, $i = 1,2,3,4$,
form a square in $ \mbox{\rm I$\!$R} ^2$ containing $S^1$ ($\vec{v}^{s_1} = -
\vec{v}^{s_4}$, $\vec{v}^{s_2} = - \vec{v}^{s_3}$), i.e. all
points of $S^1$ are illuminated by these four points. The billiard
$B$ (\ref{4.4}) is a sub-compact square
($\bar{B}$ is compact) in the Lobachevsky space.
For $d = 2$ ($D=5$) it is depicted on Fig. 1.
\subsection{Billiard is a triangle}
Let us consider a model defined on
\beq{4.12a}
M = \mbox{\rm I$\!$R} \times M_{1} \times M_{2} \times M_{3}
\end{equation}
and governed by the Lagrangian
\beq{4.13a}
{\cal L}= {R}[g] - 2\Lambda
- \frac{1}{n_1!} (F^1)^2_g ,
\end{equation}
where $d_{i} = \dim M_i = d$,
$n_1 = d+1$, $d \geq 2$,
$w =-1$, $ \varepsilon (i) = 1$, $i =1, 2, 3$, and $D = 1+ 3d$.
Let
\beq{4.14a}
s_1 = (1,e,\{1 \}),\quad s_2 = (1,e, \{2 \}), \quad
s_3 = (1,e,\{3 \}),
\end{equation}
i.e. we have three electric branes corresponding to
the form $F^1$. The brane $s_i$ ``lives'' in $M_i$, $i =1, 2, 3$.
From (\ref{4.8b}) we get
\bear{4.15a}
(\vec{v}^{s_i})^2 = 6d -2,
\\ \label{4.16a}
\vec{v}^{s_i} \vec{v}^{s_j} = 1 - 3d,
\end{eqnarray}
$i \neq j$; $i,j = 1,2,3$.
For $d \geq 2$ the points $\vec{v}^{s_i}$, $i = 1,2,3$,
form a triangle in $ \mbox{\rm I$\!$R} ^2$ containing $S^1$ and all
points of $S^1$ are illuminated by these three points. The billiard
(\ref{4.4}) is a sub-compact triangle in the Lobachevsky space $D^2$.
For $d = 2$ it is depicted on Fig. 2.
\begin{figure}[cht]
\begin{center}
\epsfig{file=pic2.eps,height=9cm,width=9cm}
\end{center}
\caption{
Triangle billiard in the 7-dimensional model
with three internal spaces and three electric ``branes''.}
\end{figure}
For $d = 1$ we obtain (at least formally) the billiard
$B$ depicted on Fig. 3.
The closure of $B$
is not compact but the area of $B$ is finite.
This billiard appears in the well-known Bianchi-IX model
(for a review, see \cite{P,K1,IM}).
For $d = 1$ the restriction on composite $p$-branes (\ref{B.2})
is not satisfied but to avoid this obstacle we may consider non-composite
case when the Lagrangian (\ref{4.13a}) is replaced
by the following Lagrangian in the dimension $D =4$
with three 2-forms $F^a$, $a=1,2,3$,
\beq{4.13b}
{\cal L}= {R}[g] - 2\Lambda
- \sum_{a=1}^{3} \frac{1}{2!} (F^a)^2_g ,
\end{equation}
and the relation (\ref{4.14a})
is replaced by its non-composite analogue:
\beq{4.14b}
s_1 = (1,e,\{1 \}),\quad s_2 = (2,e, \{2 \}), \quad
s_3 = (3,e,\{3 \}).
\end{equation}
\begin{figure}[cht]
\begin{center}
\epsfig{file=pic3.eps,height=9cm,width=9cm}
\end{center}
\caption{
Triangle billiard for $D=4$ coinciding with that of Bianchi-IX
model.}
\end{figure}
\section{$D =11$ supergravity}
\subsection{``Truncated'' $D =11$ supergravity}
Now we consider a ``truncated'' bosonic sector of
$D = 11$ supergravity governed by the action
(``truncated'' means without Chern-Simons terms)
\beq{6.8}
S_{tr} =
\int d^{11}z \sqrt{|g|} \biggl\{R[g] - \frac{1}{4!} F^2_g \biggr\}
\end{equation}
where $F = d A = F^1$ is a 4-form. In this case we have
electric 2-branes ($d(I_s) = 3$) and magnetic
5-branes ($d(I_s) = 6$).
From (\ref{4.8b}) we get
\bear{6.9}
(\vec{v}^{s})^2 = 21, \qquad v_s = e,
\\ \label{6.10}
(\vec{v}^{s})^2 = 6, \qquad v_s = m,
\end{eqnarray}
and
\bear{6.11}
\vec{v}^{s} \vec{v}^{s'} =
&&= 1 + 10 [ d(I_s \cap I_{s'}) - 1],
\qquad v_s = v_{s'} = e,
\\ \label{6.12}
&&= 1 + \frac{5}{2} [ d(I_s \cap I_{s'}) - 4],
\qquad v_s = v_{s'} = m,
\\ \label{6.13}
&&= 1 + 5 [ d(I_s \cap I_{s'}) - 2],
\qquad v_s = e, \ v_{s'} = m.
\end{eqnarray}
Scalar products (\ref{6.9})-(\ref{6.13}) are (in some sense)
``building blocks'' for constructing billiards for the model
under consideration.
Now we suggest an example of a billiard with a finite volume
that occurs in the truncated $D = 11$ supergravity.
Let us consider the metric (\ref{2.11g})
defined on the manifold
\beq{6.14}
M = \mbox{\rm I$\!$R} \times M_{1} \times \ldots \times M_{5},
\end{equation}
where all $(M_i, g^i)$, $i = 1, \ldots, 5$,
are 2-dimensional Einstein manifolds of the Euclidean signature
and $w = -1$.
We consider ten magnetic $5$-branes wrapped
on six-dimensional ``submanifolds''
$M_i \times M_j \times M_k$, $1 \leq i < j < k \leq 5$.
Thus, the $5$-branes are labeled by indices
\beq{6.15}
s = s(i,j,k) = (1,m, \{ i,j,k \}),
\end{equation}
$1 \leq i < j < k \leq 5$.
It follows from (\ref{6.10}) and (\ref{6.12}) that
all vectors $\vec{v}^{s}$ have the same length:
$|\vec{v}^{s}| = \sqrt{6}$ and $\vec{v}^{s} \vec{v}^{s'} = 1, -4$
for dimensions of intersections $d(I_s \cap I_{s'}) = 4, 2$
respectively.
Now we prove that in our case the set of sources located at points
$\vec{v}^{s} \in \mbox{\rm I$\!$R} ^4$, $s \in S = S_m$
``illuminates'' the 3-dimensional (Kasner) sphere $S^3$
and hence (according to the Proposition from Section 4)
the 4-dimensional billiard
(\ref{4.4}) $B \subset D^{4}$ has a finite volume.
Due to Proposition 2 from Section 4 it is sufficient to
prove the following proposition.
{\bf Proposition 3.}
{\em There are no $\alpha = (\alpha^1, \ldots, \alpha^5) \in
\mbox{\rm I$\!$R} ^5$ satisfying the relations
\beq{6.16}
\sum_{i=1}^{5} \alpha^i =
\sum_{i=1}^{5} (\alpha^i)^2 = \frac{1}{2},
\end{equation}
and
\beq{6.17}
\alpha^i + \alpha^j + \alpha^k > 0,
\end{equation}
for all $1 \leq i < j < k \leq 5$.}
This proposition is a consequence of the following statement.
{\bf Proposition 4.}
{\em Let $\alpha = (\alpha^1, \ldots, \alpha^5) \in
\mbox{\rm I$\!$R} ^5$ satisfy the relations (\ref{6.16}) and
$\alpha_1 \leq \alpha_2 \leq \alpha_3 \leq \alpha_4 \leq \alpha_5$.
Then
\beq{6.18}
\alpha^1 + \alpha^2 + \alpha^3 \leq 0,
\end{equation}
and $\alpha^1 + \alpha^2 + \alpha^3 = 0$ only if
$\alpha = \alpha_1$, where
\beq{6.18a}
\alpha_1 = \left(-\frac{1}{2}, \frac{1}{4},
\frac{1}{4}, \frac{1}{4}, \frac{1}{4} \right).
\end{equation}
}
{\bf Proof.} Let us consider the set $K$ of
Kasner vector parameters
$\alpha = (\alpha^1, \ldots, \alpha^5) \in \mbox{\rm I$\!$R} ^5$
satisfying (\ref{6.16}). Let
\beq{6.19}
G = \{ \alpha \in K| \alpha^1 < \alpha^2 <
\alpha^3 < \alpha^4 < \alpha^5 \}
\end{equation}
and
\beq{6.20}
\bar{G} = \{ \alpha \in K| \alpha^1 \leq \alpha^2
\leq \alpha^3 \leq \alpha^4 \leq \alpha^5 \}.
\end{equation}
$G$ is an open submanifold of the 3-dimensional ``Kasner''
manifold $K$ and $\bar{G}$ is a closure of $G$.
$\bar{G}$ is compact subset of $K$. Let us
consider a smooth function $f : \mbox{\rm I$\!$R} ^5 \rightarrow \mbox{\rm I$\!$R} $ defined by
the relation
\beq{6.21}
f(\alpha) = \alpha^1 + \alpha^2 + \alpha^3.
\end{equation}
Let $f_{|} =f_{| \bar{G}}$ be a restriction of $f$ on $\bar{G}$.
$f_{|}$ is a continuous function reaching an (absolute) maximum
on the compact (topological) space $\bar{G}$:
\beq{6.22}
C = {\rm max} f_{|} = f(\alpha_{max}),
\end{equation}
$\alpha_{max} \in \bar{G}$. It is clear that $C \geq 0$, since
$C \geq f(\alpha_1) =0$,
where $\alpha_1$ is defined in (\ref{6.18a}).
To prove the proposition it is sufficient to prove that the point
of maximum $\alpha_{max}$ is unique and
\beq{6.23}
\alpha_{max} = \alpha_1.
\end{equation}
Let us prove the relation (\ref{6.23}).
The point $\alpha_{max}$
does not belong to $G$. Indeed, if we suppose that
$\alpha_{max} \in G$ we get the conditional extremum relation
\beq{6.24}
d \Phi (\alpha) = 0
\end{equation}
at $\alpha = \alpha_{max} $, where
\beq{6.25}
\Phi(\alpha) = \Phi_0(\alpha) \equiv
f(\alpha) + \lambda_1 \sum_{i=1}^{5} \alpha^i +
\lambda_2 \sum_{i=1}^{5} (\alpha^i)^2,
\end{equation}
and $\lambda_1$ and $\lambda_2$ are Langrange multipliers.
It follows from (\ref{6.24}) and (\ref{6.25}) that
\bear{6.26}
\partial_i \Phi = 1 + \lambda_1 + 2 \lambda_2 \alpha^i = 0, \quad i= 1,2,3,
\\ \label{6.27}
\partial_j \Phi = \lambda_1 + 2 \lambda_2 \alpha^j = 0, \quad j= 4,5,
\end{eqnarray}
at $\alpha = \alpha_{max} $. Relations (\ref{6.26}) and (\ref{6.27})
imply $\lambda_2 \neq 0$ and $\alpha^4 = \alpha^5$. The latter
contradicts the inequality $\alpha^4 < \alpha^5$ for points
in $G$. Thus, $\alpha_{max} \in \bar{G} \setminus G$.
The set $\partial G = \bar{G} \setminus G$ is a border of
the curved tetrahedron $\bar{G}$.
It is a union of four faces
\bear{6.28}
\alpha^1=\alpha^2 < \alpha^3 < \alpha^4< \alpha^5 \quad (\Gamma_1),
\\ \label{6.29}
\alpha^1< \alpha^2 = \alpha^3 < \alpha^4< \alpha^5 \quad (\Gamma_2),
\\ \label{6.30}
\alpha^1< \alpha^2 < \alpha^3 = \alpha^4< \alpha^5 \quad (\Gamma_3),
\\ \label{6.31}
\alpha^1< \alpha^2 < \alpha^3 < \alpha^4 = \alpha^5 \quad (\Gamma_4),
\end{eqnarray}
six edges
\bear{6.32}
\alpha^1=\alpha^2 = \alpha^3 < \alpha^4< \alpha^5 \quad (E_{12}),
\\ \label{6.33}
\alpha^1= \alpha^2 < \alpha^3 = \alpha^4< \alpha^5 \quad (E_{13}),
\\ \label{6.34}
\alpha^1= \alpha^2 < \alpha^3 < \alpha^4= \alpha^5 \quad (E_{14}),
\\ \label{6.35}
\alpha^1<\alpha^2 = \alpha^3 = \alpha^4< \alpha^5 \quad (E_{23}),
\\ \label{6.36}
\alpha^1< \alpha^2 = \alpha^3 < \alpha^4 = \alpha^5 \quad (E_{24}),
\\ \label{6.37}
\alpha^1< \alpha^2 < \alpha^3 = \alpha^4= \alpha^5 \quad (E_{34}),
\end{eqnarray}
and four vertices
\bear{6.38}
\alpha^1 < \alpha^2 = \alpha^3 = \alpha^4= \alpha^5 \quad (V_1),
\\ \label{6.39}
\alpha^1 = \alpha^2 < \alpha^3 = \alpha^4= \alpha^5 \quad (V_2),
\\ \label{6.40}
\alpha^1 = \alpha^2 = \alpha^3 < \alpha^4= \alpha^5 \quad (V_3),
\\ \label{6.41}
\alpha^1 = \alpha^2 = \alpha^3 = \alpha^4< \alpha^5 \quad (V_4).
\end{eqnarray}
The point of maximum $\alpha_{max}$ does not belong to any
face $\Gamma_a$, $a=1,2,3,4$. Indeed, if we suppose that
$\alpha_{max}$ belongs to some face $\Gamma_a$
we get the conditional extremum relation (\ref{6.24})
with modified $\Phi$:
\beq{6.42}
\Phi = \Phi_0 + \lambda_3 (\alpha^a - \alpha^{a+1}),
\end{equation}
$a=1,2,3,4$. But one may verify that
there are no solutions of the relation (\ref{6.24}) in this case.
Analogous arguments lead us to a non-existence of
points of maximum among the points of edges
$E_{12}, E_{13}, E_{14}, E_{23}, E_{24}, E_{34}$. Here the only
difference is that there exist (only) two extremal points
belonging to $E_{13}$ and $E_{23}$ respectively:
\bear{6.33a}
\alpha^1= \alpha^2 = \frac{1}{10} \left(1 - \frac{12}{\sqrt{14}} \right),
\alpha^3 = \alpha^4 =
\frac{1}{10} \left(1 + \frac{3}{\sqrt{14}} \right),
\alpha^5 =
\frac{1}{10} \left(1 + \frac{18}{\sqrt{14}} \right),
\\ \label{6.35a}
\alpha^1
= \frac{1}{10} \left(1 - \frac{3}{2} \sqrt{6} \right),
\alpha^2 = \alpha^3 = \alpha^4 =
\frac{1}{10} \left(1 - \frac{1}{4}\sqrt{6} \right),
\alpha^5 =
\frac{1}{10} \left(1 + \frac{9}{4}\sqrt{6} \right),
\end{eqnarray}
but
$f(\alpha) < 0$ in these points and hence they are not
the points of maximum. Thus, $\alpha_{max}$ belongs to the
set of vertices:
\bear{6.38a}
\alpha^1 = - \frac{1}{2}, \quad
\alpha^2 = \alpha^3 = \alpha^4= \alpha^5 = \frac{1}{4} \quad (V_1),
\\ \label{6.39a}
\alpha^1 = \alpha^2 =
\frac{1}{10} \left(1 - \frac{3}{2} \sqrt{6} \right), \quad
\alpha^3 = \alpha^4= \alpha^5 =
\frac{1}{10} \left(1 + \sqrt{6} \right) \quad (V_2),
\\ \label{6.40a}
\alpha^1 = \alpha^2 = \alpha^3 =
\frac{1}{10} \left(1 - \sqrt{6} \right), \quad
\alpha^4= \alpha^5 =
\frac{1}{10} \left(1 + \frac{3}{2} \sqrt{6} \right) \quad (V_3),
\\ \label{6.41a}
\alpha^1 = \alpha^2 = \alpha^3 = \alpha^4 = - \frac{1}{20}, \quad
\alpha^5 = \frac{7}{10} \quad (V_4).
\end{eqnarray}
The calculation gives us $f(\alpha) < 0$ for
$\alpha = V_i$, $i =2,3,4$. Thus, the first vertex $V_1 = \alpha_1$
is the point of maximum. The Proposition 4 is proved.
We also proved the Proposition 3 (as a consequence
of the Proposition 4) and
due to Proposition 2 the following proposition.
{\bf Proposition 5.} {\em For the model (\ref{6.8})-(\ref{6.15})
the Kasner sphere $S^3$ is illuminated
by the set of ten sources located at points
$\vec{v}^{s} \in \mbox{\rm I$\!$R} ^4$ from (\ref{4.6}) , with $s = s(i,j,k)$,
$1 \leq i < j < k \leq 5$, defined in (\ref{6.15}),
and hence the billiard $B$ (\ref{4.4}) has a finite volume.}
Moreover, this proposition may be strengthen as follows.
{\bf Proposition 5a.} {\em In terms of Proposition 5
all points of the Kasner $S^3$ sphere
except five points $\vec{n}_1, \ldots, \vec{n}_5
\in S^3$ corresponding to the Kasner sets
\beq{6.18b}
\alpha_1 =
\left(-\frac{1}{2}, \frac{1}{4}, \frac{1}{4}, \frac{1}{4}, \frac{1}{4}
\right), \ldots, \alpha_5 = \left( \frac{1}{4}, \frac{1}{4}, \frac{1}{4},
\frac{1}{4}, -\frac{1}{2} \right) \in K,
\end{equation}
respectively
(see (\ref{4.13})) are strongly illuminated by the set of ten sources
$\vec{v}^{s} \in \mbox{\rm I$\!$R} ^4$, $s \in S$.}
{\bf Proof.} The set $K$ of Kasner parameters
satisfying (\ref{6.16}) is a union of $5! = 120$ ``sectors''
\beq{6.18c}
K = K_{12345} \cup K_{21345} \cup \ldots \cup K_{54321},
\end{equation}
where
\beq{6.18d}
K_{i_1 i_2 i_3 i_4 i_5} =
\bar{G} = \{ \alpha \in K| \alpha^{i_1} \leq \alpha^{i_2}
\leq \alpha^{i_3} \leq \alpha^{i_4} \leq \alpha^{i_5} \}
\end{equation}
and $(i_1, i_2, i_3, i_4, i_5)$ is a permutation of $(1, 2, 3, 4, 5)$.
Due to Proposition 1a and Proposition 4
any point $\vec{n} = \vec{n}(\alpha) \in S^3$ corresponding to
$\alpha \in K_{12345} \setminus \{ \alpha_1 \}$ is strongly
illuminated by the source located at point
$\vec{v}^{s}$, with $s = s(1,2,3)$.
Analogously, any point $\vec{n} =
\vec{n}(\alpha) \in S^3$ corresponding to
$\alpha \in K_{i_1 i_2 i_3 i_4 i_5} \setminus \{ \alpha_{i_1} \}$ is
strongly illuminated by the source located at point $\vec{v}^{s}$, with
$s = s(i_1,i_2,i_3)$, where
$(i_1, i_2, i_3, i_4, i_5)$ is a permutation of $(1, 2, 3, 4, 5)$.
The proposition is proved.
The points $\vec{n}_1, \ldots, \vec{n}_5$ from Proposition 5a
are not strongly illuminated (see Proposition 1a). Thus,
Proposition 5, Proposition 5a and Proposition (from Section 4)
imply that the billiard $B$ (\ref{4.4}) has a finite volume
but its closure $\bar{B}$ is not compact.
Points $\vec{n}_1, \ldots, \vec{n}_5 \in S^3$ are
ending points of five ``horns'' of $B$.
(These ``horns'' look similar to potential energy
``valleys'' that occur in some toy models, e.g.
related to M(atrix) theory \cite{AMRV}).
Using relations (\ref{4.13}) one
may verify that
\beq{6.8b}
\vec{n}_i \vec{n}_j = - \frac{1}{4}, \quad i \neq j,
\end{equation}
$i,j = 1, \ldots, 5$. This means that
$\vec{n}_1, \ldots, \vec{n}_5$ are vertices
of a 4-dimensional simplex.
Here we considered the billiard $B$
generated by ten sources, corresponding to
non-zero charges :
$Q_s \neq 0$, $s \in S$. If some charges are zero:
$Q_s = 0$, $s \in S_0$,
then the corresponding points $\vec{v}^{s}$,
$s \in S_0$ ($S_0 \neq \emptyset$)
are ``switched off'' and billiard is
generated by $m \leq 9$ point-like sources. In this
case we have the following proposition.
{\bf Proposition 5b.} {\em In terms of Proposition 5
any subset of sources $\vec{v}^s$, $s \in S \setminus S_0$,
with $S_0 \neq \emptyset$, does not illuminate
the Kasner sphere $S^3$ and hence the billiard $B$
generated by this subset has an infinite volume.}
{\bf Proof.} Without a loss of generality
let us suppose that $S_0$ contains $s(1,2,3)$,
i.e. at least the source at $\vec{v}^s$, $s=s(1,2,3)$,
is ``switched off''. Then, the point $\vec{n}$ corresponding to
the set $\alpha = (\alpha^i)$ from
(\ref{6.40a}) belongs to the shadow, since
\beq{6.17a}
\alpha^i + \alpha^j + \alpha^k > 0,
\end{equation}
for all $1 \leq i < j < k \leq 5$, $(i,j,k) \neq (1,2,3)$.
Thus, the shadow set is non-empty. (This set
is open since it is an intersection
of a finite number of open shadow sets corresponding to sources of light.)
The proposition is proved.
\subsection{Inclusion of Chern-Simons term}
Now we consider the total bosonic sector action
for $D =11$ supergravity with the Chern-Simons term included:
\beq{6.8a}
S = S_{tr} + c \int_{M} A \wedge F \wedge F
\end{equation}
where $S_{tr}$ is defined in (\ref{6.8}),
$c = {\rm const}$, ($F = d A$).
Since the second term in (\ref{6.8a}) (Chern-Simons term)
does not depend upon a metric the Einstein equations
are not changed. The only modification of equations of motion
is related to ``Maxwell'' equations
\beq{6.43}
d*F = {\rm const} \ F \wedge F.
\end{equation}
Due to (\ref{6.43}) solutions
to field equations corresponding to the truncated
model (\ref{6.8}) with the trivial Chern-Simons term
\beq{6.44}
F \wedge F = 0.
\end{equation}
are also solutions for $D =11$ supergravity.
Now, we are interested in existence of solutions
for the truncated model with non-zero charges $Q_s \neq 0$
satisfying (\ref{6.44}).
Calculating the Hodge dual in (\ref{2.29n})
and using the solution (\ref{5.29n}) we get
\beq{6.45}
F = - e^f \sum_{s \in S} Q_s \tau(\bar{I}_s), \quad
f = 2 \sum_{i=1}^{5}x^i - \gamma,
\end{equation}
where $\bar{I}_s = \{1,2,3,4,5 \} \setminus I_s$, $s \in S =S_m$,
and $S$ is the index set defined in (\ref{6.15}).
For the Chern-Simons term we obtain
\beq{6.46}
e ^{-2f} F \wedge F = 2 \sum_{i = 1}^{5} P_i \tau(\bar{\{i \}}),
\end{equation}
where
\beq{6.46b}
2P_i = \sum_{s,s' \in S} Q_s Q_{s'} \delta(I_s \cap I_{s'}, \{i \}),
\end{equation}
$i =1,2,3,4,5$. Here $\delta(A,B) =1$ for $A = B$,
$\delta(A,B) = 0$ otherwise.
{\bf Proposition 6.} {\em Let all charges be non-zero:
$Q_s \neq 0$ for all $s \in S$. Then
\beq{6.44b}
(F \wedge F)(z) \neq 0,
\end{equation}
in all points $z \in M$.}
{\bf Proof.} Let us suppose that the Chern-Simons term vanishes
in some point (i.e. relation (\ref{6.44}) in some point is satisfied).
Since forms $\tau(\bar{\{i \}})$
are linearly independent at any point,
we obtain $P_i = 0$, $i =1,2,3,4,5$,
or explicitly
\bear{6.45a}
P_1 = Q_{123} Q_{145} + Q_{124}Q_{135}+ Q_{125}Q_{134} = 0,
\\ \label{6.46a}
P_2 = Q_{123} Q_{245} + Q_{124}Q_{235}+ Q_{125}Q_{234} = 0,
\\ \label{6.47a}
P_3 = Q_{123} Q_{345} + Q_{134}Q_{235}+ Q_{135}Q_{234} = 0,
\\ \label{6.48a}
P_4 = Q_{124} Q_{345} + Q_{134}Q_{245}+ Q_{145}Q_{234} = 0,
\\ \label{6.49a}
P_5 = Q_{125} Q_{345} + Q_{135}Q_{245}+ Q_{145}Q_{235} = 0,
\end{eqnarray}
where we denote $Q_{ijk} = Q_s$ for $s = s(i,j,k)$,
$1 \leq i < j < k \leq 5$. Let us denote
$k_2 = Q_{345}$, $k_3 = Q_{245}$, $k_4 = Q_{235}$, $k_5 = Q_{234}$
and $a = Q_{145}$, $b = Q_{135}$. From (\ref{6.49a}) we
get
\beq{6.50}
Q_{125} = - \frac{k_3}{k_2} b - \frac{k_4}{k_2} a.
\end{equation}
From the relation $k_3 P_3 +k_4 P_4 +k_5 P_5 = 0$ we get
\beq{6.51}
Q_{134} = - \frac{k_5}{k_4} b - \frac{k_5}{k_3} a.
\end{equation}
From (\ref{6.47a})and (\ref{6.48a}) we
deduce
\beq{6.52}
Q_{123} = \frac{k_4 k_5}{k_2 k_3} a, \quad
Q_{124} = \frac{k_3 k_5}{k_2 k_4} b.
\end{equation}
Substituting (\ref{6.50})-(\ref{6.52}) into eq. (\ref{6.45a})
we get
\beq{6.53}
(k_4 a)^2 + (k_3 b)^2 + (k_4 a + k_3 b)^2 = 0.
\end{equation}
Hence $a=b =0$. But this contradicts our supposition, that
$Q_s \neq 0$. So, the proposition is proved.
Thus, it follows from Proposition 5b and Proposition 6,
that the inclusion of the Chern-Simons term leads us to
the billiard $B$ of infinite volume. In this case some
Kasner (shadow) zones are opened and we have the Kasner-like
behaviour near the singularity.
\section{Discussions}
In this paper we considered the behaviour near the
singularity of the multidimensional model describing the cosmological
evolution of several Einstein spaces in the theory with scalar fields
and fields of forms. Using the results from \cite{IKM1,IKM,IM} we
obtained the billiard representation on multidimensional Lobachevsky
space for the cosmological model near the singularity. We suggested
and studied
examples with oscillating behaviour near the singularity in the model with
four $p$-branes and a square billiard and in the model with three
$p$-branes, when the billiard is a triangle (like it takes place in
Bianchi-IX model). A 4-dimensional billiard with a finite volume in the
``truncated'' $D = 11$ supergravity was also considered. It was shown
that the inclusion of the Chern-Simons term leads to the destruction
of some confining walls of the billiard.
\bigskip
\noindent
{\bf Acknowledgments.}
This work was supported in part
by the Russian Ministry for
Science and Technology and Russian Fund for Basic Research,
project N 98-02-16414. V.N.M. is grateful
to the Dept. Math., University of Aegean for the kind
hospitality during his stay in Karlovassi, Greece in December
1998 - January 1999.
\small
|
\section{Introduction}
The last decade has seen significant experimental progress in charm baryon
physics. Most of the ground state baryons containing one c-quark have now been
established \cite{PDG}. Their classification in the standard quark model is
based on three different spin configurations: the lowest lying configuration
has $J^P=\frac{1}{2}^+$ ($\Lambda_c$, $\Xi_c$) with the two light
quarks in the antisymmetric spin 0 configuration, the next higher
configuration $J^P=\frac{1}{2}^+$ ($\Sigma_c$, $\Xi'_c$, $\Omega_c$) and
the highest configuration $J^P=\frac{3}{2}^+$ ($\Sigma^*_c$, $\Xi^*_c$,
$\Omega^*_c$) where the two light quarks are in the symmetric spin 1
configuration, coupling down and up to $\frac{1}{2}^+$ and $\frac{3}{2}^+$,
respectively, with the heavy quark.
Some of their decay modes have been seen and have been used for the
determination of their masses (see \cite{PDG}).
The CLEO Coll. \cite{CLEO1} has presented
evidence for a pair of excited charm baryons, one decaying into
$\Lambda_c^+\pi^+$ with a mass difference $M(\Lambda_c^+\pi^+)-M(\Lambda_c^+)$
of $234.5\pm 1.1\pm 0.8$ MeV and a width of $17.9^{+3.8}_{-3.2}\pm 4.0$ MeV,
and the other into $\Lambda_c^+\pi^-$ with a mass difference
$M(\Lambda_c^+\pi^-)-M(\Lambda_c^+)$ of $232\pm 1.0\pm 0.8$ MeV and a width
of $13.0^{+3.7}_{-3.0}\pm 4.0$ MeV. The CLEO Coll. interpreted these data
as evidence for the $\Sigma^{*++}_c$ and $\Sigma^{*0}_c$,
the spin $\frac{3}{2}^+$ excitations of the $\Sigma_c$ baryons.
Earlier, the CLEO Coll. \cite{CLEO2} has also reported
evidence for a pair of excited charmed baryons, one decaying into
$\Xi_c^+\pi^-$ with a mass difference $M(\Xi_c^+\pi^-)-M(\Xi_c^+)$
of $178.2\pm 0.5\pm 1.0$ MeV and a width of $<5.5$ MeV,
and the other \cite{CLEO3} into $\Xi_c^0\pi^+$ with a mass difference
$M(\Xi_c^0\pi^+)-M(\Xi_c^0)$ of $174.3 \pm 0.5\pm 1.0$ MeV and a width
of $<3.1$ MeV. They interpreted these data as evidence
for the $\Xi^{*0}_c$ and $\Xi^{*+}_c$, the spin $\frac{3}{2}^+$
excitations of the $\Xi_c$ baryons.
Recently, the CLEO Coll. \cite{CLEO4} has reported on the
observation of two narrow states in the decay modes $\Xi_c^+\gamma$ and
$\Xi_c^0\gamma$. They have been interpreted as the $\Xi_c'$ states.
The mass differences $M(\Xi_c^{+\prime})-M(\Xi_c^{+})$ and
$M(\Xi_c^{0\prime})-M(\Xi_c^{0})$ have been measured to be
$107.8\pm 1.7\pm 2.5$ MeV and $107.0\pm 1.4\pm 2.5$ MeV, respectively.
Except for the $\Omega_c^*$ and some of the isospin partners of the
charm baryon ground states most of the ground state charm baryons are now
well established.
Three collaborations (ARGUS \cite{ARGUS}, E687 \cite{E687} and
CLEO \cite{CLEO5}) have seen a doublet of particles decaying into
$\Lambda_c^+\pi^+\pi^-$. They have been interpreted as the lowest lying
orbitally excited states of the $\Lambda_c^+$: $\Lambda_{c1}^+(2593)$
with $J^P=\frac{1}{2}^-$ and $\Lambda_{c1}^{*+}(2625)$ with
$J^P=\frac{3}{2}^-$. The PDG average for the decay width of the
$\Lambda_{c1}^+(2593)$ is $3.6^{+2.0}_{-1.3}$ MeV. The branching fractions
into the $\Lambda_c^+\pi^+\pi^-$ and the $\Sigma_c^{++}(2455)\pi^-$
and $\Sigma_c^{0}(2455)\pi^+$ modes are estimated to be $\approx 67\%$,
$24\pm 7\%$ and $18\pm 7\%$, respectively. For the decay width of the
$\Lambda_c^+(2625)$ there exists only an upper limit ($<1.9$ MeV) and branching
ratios \cite{PDG}, \cite{ARGUS}-\cite{CLEO5}.
The CLEO Coll. \cite{CLEO6} has reported preliminary evidence for a new
charm baryon decaying into $\Xi_c^+\pi^+\pi^-$ via the intermediate
$\Xi_c^{*0}$-state. The measured mass difference is given by
$M(\Xi_c^+\pi^+\pi^-)-M(\Xi_c^+)=349.4\pm 0.7\pm 1.0$ MeV, and its width is
$\Gamma<2.4$ MeV. This new particle was interpreted as the $\Xi^{*+}_{c1}$
with $J^P=\frac{3}{2}^-$, the charm-strange partner
of the $\Lambda_{c1}^{*+}(2625)$.
The classification and the decay properties of ground and excited state
charm and bottom baryons have been reviewed in \cite{KKP}. The analysis
was based on the heavy quark limit given by the leading order in the
$1/m_Q$-expansion. In the heavy quark limit the dynamics of the heavy and
light quarks decouple leading to a number of model independent relations
between various decay modes of the heavy baryons.
Further relations between physical observables can be obtained if one
assumes additional symmetries for the light quark system.
For instance, by using a constituent quark model picture for the light
diquark system with an underlying $SU(2N_f)\otimes O(3)$ symmetry,
a number of further relations were derived in \cite{HKLT} for the form
factors in semileptonic $b-c$ transitions.
The subject of this paper mostly concerns the one-pion and one-photon decays
of ground state charm baryons as well as those of the lowest-lying p-wave
states. We shall also determine the one-photon decays of
the orbital excitations of bottom baryons. The analysis of these transitions
provides an excellent laboratory for tests of heavy quark symmetry predictions
on the one hand and tests of the soft dynamics of the light-side one-pion
(photon) diquark transitions on the other hand. In the heavy
quark limit the pion and photon are emitted from the light diquark system
while the heavy quark is unaffected by the emission process.
Single pion transitions of charm baryons were analyzed before in \cite{KKP,HKT}
by again using the constituent quark model picture for the light
diquark system with its underlying $SU(2N_f)\otimes O(3)$ symmetry.
Using this symmetry one significantly reduces the number of
independent coupling factors \cite{KKP,HKT}.
A similar constituent quark model approach has been employed in \cite{PY}
to establish relationships
among the coupling constants characterizing the decays of s-wave and p-wave
heavy baryons. The results of the work \cite{PY} are
the same as in \cite{HKT} since both approaches are based on the same
constituent quark model picture.
Light-front (LF) quark model functions with a factorized
harmonic-oscillator transverse momentum component and a longitudinal
component given by a $\delta$-function have been constructed in \cite{TOK}.
They have been employed to calculate the strong couplings for
$\Sigma_c\to\Lambda_c\pi$,
$\Lambda_{c1}\to\Sigma_c\pi$, and $\Lambda_{c1}^*\to\Sigma_c\pi$ decay modes
which correspond to P-wave, S-wave and D-wave transitions, respectively.
The flavor-spin symmetry of the heavy quarks and the spontaneously broken
$SU(3)_L\otimes SU(3)_R$ chiral symmetry of the light quarks were exploited
to describe the interactions of heavy mesons and heavy baryons with
the $\pi$, $K$, and $\eta$ mesons considered as the Goldstone bosons
\cite{Yan}. This approach contains three free parameters and was applied
to strong and semileptonic decays of heavy hadrons. The radiative decays of
heavy mesons and heavy baryons were also studied within this formalism in
\cite{Cheng1}. The two orbital excitations of the $\Lambda_c$ have been
analyzed within this approach which is referred to as the
heavy hadron chiral perturbation theory (HHCPT) by Cho \cite{Cho}
(see also \cite{PY}).
Information on one of the three free parameters of HHCPT in these decays was
obtained from the radiative decay $\Xi_c^{\prime * 0}\to \Xi_c^0\gamma$
by Cheng \cite{Cheng2}. Pion transitions of the ground states and some excited
p-wave states of heavy baryons
were studied within HHCPT in \cite{HDH}. The coupling constants were
estimated using the experimental data on the pion decays of strange
$\Lambda$ and $\Sigma$ baryons treating the $s$-quark as static.
The radiative decays of some excited heavy baryon states have been calculated
in the so-called bound state picture \cite{Chow} motivated by the large
$N_c$ limit. It was shown that the $\Lambda_{c1}(2593)\to\Lambda_c\gamma$
and $\Lambda_{c1}^*(2625)\to\Lambda_c\gamma$ decays are severely suppressed
whereas the $\Lambda_{b1}^*(5900)\to\Lambda_b\gamma$ mode possibly
dominates over the strong decay mode.
Some aspects of the phenomenology of new baryons with charm and strangeness
were discussed in \cite{CF}. The authors of Ref. \cite{CF} estimated the
expected width of the excited state $\Xi_{c1}(\frac{3}{2})^+$
of the $\Xi_c$ family using the experimental value for the width of
the $\Lambda_{c1}(\frac{1}{2})^+$ within the $SU(3)$ flavor symmetry limit.
Also it was pointed out that one should search for the $\Omega_{c2}^*$ baryon
in the $\Xi_c K$ decay channel.
The ratio of electric quadrupole (E2) and magnetic dipole (M1) components
for the decay $\Sigma_c^*\to\Lambda_c\gamma$ has been computed by
Savage \cite{Savage} within the HHCPT approach. The result was that the
$1/m_c$ suppression of the E2 amplitude is compensated for by a small energy
denominator arising from the infrared behavior of pion loop graphs in chiral
perturbation theory. This leads to a $E2/M1$ ratio of the order of a few
percent depending among others on
the $\Sigma_c^* - \Sigma_c$ spin symmetry breaking mass difference.
The corresponding ratio in the bottom baryon sector is smaller by a factor
of $\sim M_c/M_b$.
The couplings in the HHCPT Lagrangian have been estimated from QCD sum rules
in an external axial field \cite {GY}.
The radiative decays of heavy baryons were studied with the light cone QCD
sum rules in the leading order of HQET in \cite{ZD}.
All approaches considered above exploit the symmetry of the heavy quark limit
with some additional assumptions on the structure of the light quark
system without employing any dynamical scheme for the composite structure of
hadrons.
A relativistic quark model of hadrons \cite{EI}-\cite{ILL} has been developed
some time ago to describe physical observables in the low-energy domain.
The model is based on an effective hadron-quark Lagrangian and the
compositeness condition $Z_H=0$ where $Z_H$ is a wave function renormalization
constant of the hadron \cite{SW}.
This approach has been employed to give a unified description of
leptonic and semileptonic decays of heavy mesons in \cite{IS}.
Recently, this model called the Relativistic Three-Quark Model (RTQM) has
been applied to calculate physical observables in the decays of heavy
baryons \cite{ILKK}-\cite{IKL}. All results of the RTQM model are expressed
through a few model parameters: the masses of the light quarks, the mass
differences of the heavy baryon and heavy quark, and the size of the Gaussian
distribution of constituents inside the hadron.
The exclusive semileptonic and nonleptonic decays of
charmed and bottom baryons have been considered within the RTQM
\cite{ILKK,IKLR1}. Preliminary results for the strong and radiative decays
of heavy baryons have been reported in \cite{IKLR2,IKL}.
In this paper we extend this application by reporting the simultaneous
calculation of a range of one-pion and one-photon transitions
between heavy baryons.
Our article is divided into five sections with a single appendix.
In Sec. II we give some necessary background material on the RTQM and discuss
features of gauging the interaction Lagrangian in this approach.
The calculation of the matrix elements of the one-pion and one-photon
transitions between heavy baryons is given Sec. III.
In Sec. IV we present our numerical results on the
heavy baryon observables in their strong and radiative transitions.
We compare our results with available experimental data
and with the results of some other model approaches.
We make some concluding remarks in Sec. V. The Appendix contains a detailed
discussion of the gauge invariance of radiative transitions in the RTQM.
\section{Relativistic Three-Quark Model}
A detailed description of the Relativistic Three-Quark Model can be found in
Refs. \cite{ILL,ILKK}. Here we will only give the necessary background
material
needed for the description of two-body strong and electromagnetic decays
of heavy flavored baryons. We will discuss the interaction Lagrangian
describing the coupling of heavy baryons with their constituent quarks and
its gauging. We also specify the form of the light and heavy quark propagators.
The Lagrangian describing the coupling of a heavy baryon to its
constituent light and heavy quarks considerably simplifies in the heavy quark
limit (see \cite{ILKK}). One has
\begin{eqnarray}\label{LHB_int}
{\cal L}_{B_Q}^{\rm int}(x)&=&g_{B_Q}\bar B_Q(x)
\Gamma_1 Q^a(x)\int\hspace*{-0.1cm} d\xi_1\hspace*{-0.1cm}
\int\hspace*{-0.1cm} d\xi_2
F_B(\xi_1^2+\xi_2^2)\\ \label{int_bar}
&\times&q^b(x+3\xi_1-\sqrt{3}\xi_2)C\Gamma_2\lambda_{B_Q}
q^c(x+3\xi_1+\sqrt{3}\xi_2)\varepsilon^{abc}+{\rm h.c.}\nonumber\\
F_B(\xi_1^2+\xi_2^2)&=&
\int\hspace*{-0.1cm}\frac{d^4k_1}{(2\pi)^4} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{(2\pi)^4} \hspace*{0.1cm}
e^{ik_1\xi_1+ik_2\xi_2}
\tilde F_B\biggl\{\frac{[k_1^2+k_2^2]}{\Lambda_B^2}\biggr\}
\nonumber
\end{eqnarray}
The interaction Lagrangian of the pion with its constituent light quarks is
given by
\begin{eqnarray}\label{int_pi}
{\cal L}_\pi^{\rm int}(x)=\frac{ig_\pi}{\sqrt{2}}\vec\pi(x)\int
\hspace*{-0.1cm} d\xi F_\pi(\xi^2)\bar q(x+\xi/2)
\gamma^5\vec\lambda_\pi q(x-\xi/2)
\end{eqnarray}
where
\begin{eqnarray}
F_\pi(\xi^2)=\int\hspace*{-0.1cm}\frac{d^4k}{(2\pi)^4}
e^{ik\xi} \tilde F_\pi\biggl\{\frac{k^2}{\Lambda_\pi^2}\biggr\}
\nonumber
\end{eqnarray}
Here $\Gamma_i$ and $\lambda_{B_Q}$ are spinor and flavor matrices
which define the quantum numbers of the relevant three-quark currents.
They are listed in Table I. The square brackets $[...]$ and curly brackets
$\{...\}$ denote antisymmetric and symmetric flavor and spin combinations of
the light degrees of freedom\footnote{We use the following
notation for excited heavy $P$-wave baryons: \\
i) $\Lambda_{c1; S}$, $\Lambda_{c1; S}^*$,
$\Sigma_{c0; S}$ and
$\Xi_{c1; S}^*$ are symmetric under the exchange of the momenta
of the two light quarks
(called $K$-states in \cite{KKP,HKLT,HKT});
ii) $\Sigma_{c1; A}$ and $\Sigma_{c1; A}^*$ are antisymmetric under the
exchange of the momenta of the two light quarks
(called $k$-states in \cite{KKP,HKLT,HKT}).}.
The coupling strength of the respective
hadrons with their constituent quarks are denoted by the coupling constants
$g_{B_Q}$ and $g_\pi$. The parameters
$\Lambda_B$ and $\Lambda_\pi$ define the size of the distributions of light
quarks inside the heavy baryon and pion, respectively.
The baryon parameter $\Lambda_B$ is chosen to be
the same for charm and bottom baryons to provide the correct
normalization of the baryonic Isgur-Wise function in the heavy
quark limit \cite{ILKK}.
The gauging of the nonlocal interaction Lagrangian Eq.(\ref{int_bar})
can be done by using a path-independent formalism based on the
path-independent definition \cite{Mandelstam} of the derivative of the path
integral. The details may be found in \cite{Terning} and also
in \cite{ILL} where this formalism was employed to calculate
the nucleon electromagnetic form factors in the quark model in a gauge
invariant way. In order to make the Lagrangian
Eq.(\ref{LHB_int}) gauge invariant in the presence of an electromagnetic
field $A_\mu(x)$ the time-ordered P-exponent
\begin{equation}\label{exp}
P\exp\biggl\{ieQ\int\limits_x^y dz_\mu A^\mu(z)\biggr\}
\end{equation}
($Q=\rm diag\{2/3,-1/3,-1/3\}$ is the charge matrix)
is attached to each light quark field $q(y)$. We then have
\begin{eqnarray}\label{Lagr_GI}
{\cal L}_{B_Q}^{\rm int, e.m.}(x)&=&g_{B_Q}\bar B_Q(x)\Gamma_1 Q^a(x)
\int d^4\xi_1\int d^4\xi_2F_B(\xi_1^2+\xi_2^2)\\
&\times&\exp(ieQ\int\limits_x^{x+3\xi_1-\sqrt{3}\xi_2} dz_\mu A^\mu(z))
q^b(x+3\xi_1-\sqrt{3}\xi_2 )C\Gamma_2\lambda_{B_Q}\nonumber\\
&\times&\exp(ieQ\int\limits_x^{x+3\xi_1+\sqrt{3}\xi_2} dz_\mu A^\mu(z))
q^c(x+3\xi_1+\sqrt{3}\xi_2)\varepsilon^{abc}+{\rm h.c.}\nonumber
\end{eqnarray}
The Lagrangian Eq. (\ref{Lagr_GI}) generates nonlocal vertices which involve
the heavy baryons, photons and the light and heavy quarks. The gauge
formalism of Mandelstam is based on the definition of the derivative
of the path integral
\begin{equation}\label{path}
I(y,x,P)=\int\limits_x^y dz_\mu A^\mu(z)
\end{equation}
where P is the path taken from $x$ to $y$. When calculating Feynman diagrams
the derivative of $I(y,x,P)$ is defined such that
$$
\frac{\partial}{\partial y^\mu}I(y,x,P)=A_\mu(y).
$$
This means that the field $A_\mu(y)$ does not depend on the path used
in the definition of the line integral. The Mandelstam prescription makes
all matrix elements involving a photon gauge-invariant and path-independent.
The free quark Lagrangian and the interaction Lagrangian for transitions
involving the orbital excited states (P-states, etc.) are gauged by
the standard minimal substitution, i.e. for the derivative coupling appearing
in the $P$-states one has to replace $\partial_\mu\to\partial_\mu+ieA_\mu$.
Now we discuss the implementation of gauge invariance in the one-photon
transitions of heavy baryons.
There are several relevant diagrams:
i) the triangle diagrams in Fig. 2a and 2b with a photon emitted
by the heavy and light quark, respectively
ii) the contact interaction-type diagrams in Fig. 3 with a photon emitted
by one of the two nonlocal baryon-quark vertices
and
iii) the pole diagrams in Fig. 4 with a photon emitted by one of the two heavy
baryons.
The contributions coming from the triangle diagram in Fig. 2a
and the pole diagrams in Fig. 4 are nonleading in $1/m_Q$ and vanish
in the heavy quark limit.
The contact interaction-type diagrams in Fig. 3 are important to reproduce
the Ward-Takahashi identities for the connected Green functions
(see, for example, Ref. \cite{ILL}). Their contributions
to the matrix element of the one-photon transitions of ground state charm
baryons are nonleading in the heavy mass expansion,
at least when the photon is on its mass shell $(q^2=0)$.
However, they contribute to the matrix elements
of one-photon transitions of the orbitally excited charm baryon states
$B_{c1}\to B_c\gamma$ and $B_{c1}^*\to B_c\gamma$ where
$B_c=\Lambda_c$ or $\Xi_c$. Formally the contact interaction type
contributions in this case are of the order
$O(qv)\approx O(0)$ ($qv=(m_i^2-m_f^2)/(2m_i) \approx 0$) and one would
naively assume that they vanish in the heavy quark limit.
However, contrary to the ground state transitions, the excited state
transitions $B_{c1}\to B_c\gamma$ and $B_{c1}^*\to B_c\gamma$ are of the
same order $Q(qv)$ and thus the contact interaction-type contribution must
be kept in this case. In the Appendix we provide a detailed discussion of
the gauge-invariance structure of the matrix
elements for one-photon transitions of excited charm
baryons\footnote{In our previous analysis of we neglected the contact
interaction-type diagrams of the $P$-wave heavy baryon
transitions \cite{IKL}. Their
contribution to the invariant matrix element is very small of order 3-4\%.
We demonstrate in the Appendix that the contact interaction-type
contribution is needed to reproduce the correct gauge invariance structure
of the photon transitions.}.
For radiative transitions involving P-wave states there are additional contact
interaction-type diagrams which contribute to the leading order in the
heavy quark expansion. They result from the minimal substitution prescription
for the derivatives acting on the excited
heavy baryon fields in the heavy baryon-three-quark interaction
Lagrangian (\ref{LHB_int}). In the following we will refer to such
diagrams as "local contact" diagrams in order to distinguish them from the
contact diagrams generated by the gauging of the nonlocal heavy
baryon-three-quark vertex ("nonlocal contact" diagrams).
Thus, in the heavy quark limit the radiative decay of a ground state heavy
baryon is described by the solely triangle diagram in Fig. 2b with the
photon emitted by the light quark whereas there are additional contributions
coming from both the ``nonlocal contact'' in Fig. 3 and ``local contact''
in Fig. 4 diagrams for the radiative decays of heavy excited baryon states.
For the one-pion transitions one only needs the only triangle diagram
depicted in Fig. 1.
Let us now specify our model parameters. For the light quark propagator with
a constituent mass $m_q$ we shall use the standard form of the free fermion
propagator
\begin{equation}\label{light}
S_q(k)=\frac{1}{m_q-\not\! k}.
\end{equation}
For the heavy quark propagator we shall use the leading term
$S_v(k,\bar\Lambda_{\{q_1q_2\}})$ in the inverse mass expansion of
the free fermion propagator:
\begin{eqnarray}
S_Q(p+k)&=&\frac{1}{m_Q-(\not\! p+\not\! k) }=
S_v(k,\bar\Lambda_{\{q_1q_2\}})
+O\biggl(\frac{1}{m_Q}\biggr),
\nonumber\\
&&\nonumber\\
S_v(k,\bar\Lambda_{\{q_1q_2\}})&=&-\frac{1+\not\! v}
{2(vk+\bar\Lambda_{\{q_1q_2}\})}. \label{heavy}
\end{eqnarray}
where we introduce the mass difference parameter
$\bar\Lambda_{\{q_1q_2\}}=M_{\{Qq_1q_2\}}-m_Q$ which is the difference
between the heavy baryon mass $M_{\{Qq_1q_2\}}\equiv M_{B_Q}$ and the heavy
quark mass $m_Q$. The four-velocity of the heavy quark is denoted by $v$ as
usual. We shall neglect a possible mass difference between the constituent
$u-$ and $d-$quark and thus take $\bar\Lambda : = \bar\Lambda_{uu}=
\bar\Lambda_{ud}= \bar\Lambda_{dd}$, $ \bar\Lambda_s : = \bar\Lambda_{us}=
\bar\Lambda_{ds}$ meaning that there are altogether three independent
mass difference parameters: $ \bar\Lambda$, $ \bar\Lambda_s$, and
$\bar\Lambda_{ss}$.
The vertex functions $F_B$ and $F_\pi$ are arbitrary functions except
that they should fall off sufficiently fast in the ultraviolet region to
render the Feynman diagrams ultraviolet finite.
In principle, their functional forms are calculable from the solutions
of the Bethe-Salpeter equations for the baryon and pion bound states.
For example, in \cite{IKMR} semileptonic meson transitions have been
analyzed using the heavy-quark limit of the Dyson-Schwinger
equation. The results were found to be quite insensitive to the details of
the functional form of the heavy-meson Bethe-Salpeter amplitude.
We will use this observation as a guiding principle and choose simple
Gaussian forms for the vertices $F_B$ and $F_\pi$.
Their Fourier transforms read
\begin{equation}\label{BS}
\tilde F_B(k_1^2+k_2^2)=\exp\left(\frac{k_1^2+k_2^2}{\Lambda_B^2}\right),
\hspace{1cm}
\tilde F_\pi(k^2)=\exp\left(\frac{k^2}{\Lambda_\pi^2}\right).
\end{equation}
where $\Lambda_B$ and $\Lambda_\pi$ characterize the size of the
distributions of the light quarks inside a baryon and pion, respectively.
The coupling constants $g_{B_Q}$ and $g_\pi$ in Eqs. (1) and (2) are calculated
from {\it the compositeness condition} (see, ref.~\cite{EI,SW}), which means
that the renormalization constant of the hadron wave function is set equal to
zero:
$Z_{B_Q}=1-g_{B_Q}^2\Sigma^\prime_{B_Q}(M_{B_Q})=0$ where $\Sigma_{B_Q}$ is
a baryon mass operator, and similarly for the pion
$Z_\pi=1-g_\pi^2\Pi^\prime_\pi(m^2_\pi)=0$ where $\Pi_\pi$ is
a pion mass operator.
A drawback of our approach is the lack of confinement. To avoid
the appearance of unphysical imaginary parts in the Feynman
diagrams, we shall require that $M_{B_Q}<m_Q+m_{q_1}+m_{q_2}$ which
implies that the mass difference parameter $\bar\Lambda_{q_1q_2}$ is bounded
from above by the requirement
$\bar\Lambda_{q_1q_2}<m_{q_1}+m_{q_2}$.
As mentioned before the masses of the $u$ and the $d$ quarks are set equal
($m_u=m_d=m_q$). The value of $m_q$ is determined from an analysis of
nucleon data: $m_q$=420 MeV \cite{ILL}.
The pion Gaussian size parameter $\Lambda_\pi=$ 1 GeV
is fixed from the description of low-energy pion observables (the coupling
constants $f_\pi$ and $g_{\pi\gamma\gamma}$, and the electromagnetic radii
and form factors of the transitions $\pi\to\pi\gamma$ and
$\pi\to\gamma\gamma^*$)\cite{AIKL}-\cite{IL}.
The parameters $\Lambda_B$, $m_s$, $\bar\Lambda$ are taken from
an analysis of the $\Lambda^+_c\to\Lambda^0+e^+ +\nu_e$ decay data.
A good description of the present average value of the branching ratio
$B(\Lambda_c^+\to\Lambda e^+ \nu_e$) = 2.2 $\%$ can be achieved using
the following values of the parameters $\Lambda_B$, $m_s$ and $\bar\Lambda$:
$\Lambda_B$=1.8 GeV, $m_s$=570 MeV and $\bar\Lambda$=600 MeV.
The values of the parameters $\bar\Lambda_s$ and $\bar\Lambda_{ss}$
are determined from the heuristic relations
$\bar\Lambda_s = \bar\Lambda + (m_s - m)$
and $\bar\Lambda_{ss} = \bar\Lambda + 2(m_s - m)$, which give
$\bar\Lambda_s$ = 750 MeV and $\bar\Lambda_{ss}$ = 900 MeV. We mention that,
using the same values of $\Lambda_{B_Q}$=1.8 GeV and $\bar\Lambda$=600 MeV,
one obtains a width of $5.4\times 10^{10}s^{-1}$ and a value of
$\rho^2 = 1.4$ for the slope of the Isgur-Wise function in the decay
$\Lambda_b^0\to\Lambda_c^+ e^- \bar\nu_e$ \cite{IKLR2}.
Finally, the mass values of the charm baryon states including current
experimental errors are listed in TABLE I \cite{PDG,CLEO4}.
For the pion masses we take
$m_{\pi^\pm} = $ 139.6 MeV and $m_{\pi^0}$ = 135 MeV~\cite{PDG}.
All dimensional parameters entering the Feynman diagrams are expressed
in units of $\Lambda_B$. The integrals are calculated in the Euclidean
region both for internal and external momenta. The final results are obtained
by analytic continuation of the external momenta to the physical region
after the internal momenta have been integrated out.
\section{Matrix elements of one-pion and one-photon transitions}
We begin our discussion with the one-pion and
one-photon transitions of the ground state baryons. As discussed in Sec. II
in the heavy quark limit they are described by the triangle two-loop diagrams
Fig. 1 and Fig. 2b.
The contribution of the triangle diagram (Fig. 1 and Fig. 2b) to the matrix
element of the one-pion (one-photon) transition
$B^i_Q(p)\to B^f_Q(p^\prime)+X(q)$ has the following form in
the heavy quark limit
\begin{eqnarray}\label{vertex2}
M_{inv, \Delta}^X(B_Q^i\to B_Q^fX)=g_Xg^i_{\rm eff} g^f_{\rm eff}
\cdot \bar u(v^\prime)\Gamma_1^f \frac{(1+\not\! v)}{2}\Gamma_1^iu(v)
\cdot I_{q_1q_2, \Delta}^{if}(v,q)
\end{eqnarray}
where the symbol $X$ denotes a pion or a photon in the final state,
i.e. $X=\pi$ or $\gamma$
and $g_X=g_\pi/\sqrt{2}$ (pion-quark-antiquark coupling) or
$e$ (electron charge);
$g_{\rm eff}=g_{B_Q} \Lambda_B^2/ \sqrt{6}/(8\pi^2)$.
$I_{q_1q_2, \Delta}^{if}(v,q)$ is a two-loop quark integral the form
of which depends on whether one is computing pion or photon transitions.
For the pion one has
\begin{eqnarray}\label{int_pion}
\hspace*{-1cm}I_{q_1q_2, \Delta}^{if}(v,q)|_{X=\pi}&=&
\int\hspace*{-0.1cm}\frac{d^4k_1}{\pi^2i} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{4\pi^2i} \hspace*{0.1cm}
\frac{\tilde F_B(k_1,k_2,q)\tilde F_B(k_1,k_2,0)}
{[-k_1v-\bar\Lambda_{q_1q_2}]}
\Pi_{q_1q_2, \Delta}^\pi(k_1,k_2,q)\\[2mm]
\tilde F_B(k_1,k_2,q)&\equiv&
\tilde F_B\biggl\{-6\biggl[(k_1+q)^2+(k_2-q)^2+(k_1+k_2)^2\biggr]\biggr\}
\nonumber
\end{eqnarray}
Here
$\Pi_{q_1q_2, \Delta}^\pi(k_1,k_2,q)$ is the structure integral
corresponding to the light quark loop:
\begin{eqnarray}\label{Pi1}
\Pi_{q_1q_2, \Delta}^\pi(k_1,k_2,q)=C_{\rm flavor}
\tilde F_\pi\biggl\{-\biggl(k_2-\frac{q}{2}\biggr)^2\biggr\}
{\rm tr}\biggl[\Gamma_2^i S_{q_2}(k_1+k_2)\Gamma_2^f
S_{q_1}(k_2-q)\gamma^5 S_{q_1}(k_2)\biggr]
\end{eqnarray}
where $C_{\rm flavor}={\rm tr}[\lambda_\pi \lambda_{B^i} \lambda_{B^f}]$ is
a trace of flavor matrices and $\Gamma_{1(2)}^i$ and $\Gamma_{1(2)}^f$
are the Dirac matrices of the initial and the final baryons, respectively.
For the photon transition one has
\begin{eqnarray}\label{int_el}
\hspace*{-1cm}I_{q_1q_2, \Delta}^{if}(v,q)|_{X=\gamma}&=&
\varepsilon^*_\mu(q)
\int\hspace*{-0.1cm}\frac{d^4k_1}{\pi^2i} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{4\pi^2i} \hspace*{0.1cm}
\frac{\tilde F_B(k_1,k_2,q)\tilde F_B(k_1,k_2,0)}
{[-k_1v-\bar\Lambda_{q_1q_2}]}
\Pi_{q_1q_2, \Delta}^{\gamma;\mu}(k_1,k_2,q)\\[2mm]
\tilde F_B(k_1,k_2,q)&\equiv&
\tilde F_B\biggl\{-6\biggl[(k_1+q)^2+(k_2-q)^2+(k_1+k_2)^2\biggr]\biggr\}
\nonumber
\end{eqnarray}
\begin{eqnarray}\label{Pi2}
\Pi_{q_1q_2, \Delta}^{\gamma;\mu}(k_1,k_2,q)&=&
Q_{q_2q_2}{\rm tr}\biggl[\Gamma_2^i S_{q_1}(k_1+k_2)
\Gamma_2^fS_{q_2}(k_2-q)\gamma^\mu S_{q_2}(k_2)\biggr]\\
&-&Q_{q_1q_1}{\rm tr}\biggl[\Gamma_2^f S_{q_2}(-k_1-k_2)\Gamma_2^i
S_{q_1}(-k_2)\gamma^\mu S_{q_1}(-k_2+q)\biggr]
\nonumber
\end{eqnarray}
where $\varepsilon^*_\mu(q)$ is the polarization vector of the photon.
Radiative transitions of excited heavy baryon states involve additional
contributions from the "local contact" and "nonlocal contact"
diagrams. In our recent analysis \cite{IKL} we have neglected
the "nonlocal contact" diagrams because their contribution to the
corresponding invariant matrix elements is quite small (of order 3-4\%).
Here we present a complete analysis of radiative transitions of heavy baryons
by taking into account all possible diagrams contributing to these processes in
the heavy quark limit as is necessary to obtain a gauge invariant result.
In the heavy quark limit the contributions of the
"local contact" diagrams (lcd) and "nonlocal contact" diagrams (ncd) to
the matrix element of the excited heavy baryon transition
$B^i_{Q1}(p)\to B^f_Q(p^\prime)+\gamma$ simplify to
\begin{eqnarray}\label{vertex_add1}
M_{\rm inv, \rm lcd}^\gamma(B_{Q1}^i\to B_Q^f\gamma)=
eg^i_{\rm eff} g^f_{\rm eff}
\cdot \bar u(v^\prime)\Gamma_1^f \frac{(1+\not\! v^\prime)}{2}
\Gamma_1^iu_\nu(v)\cdot I_{q_1q_2, {\rm lcd}}^{if, \nu}(v,q)
\end{eqnarray}
\begin{eqnarray}\label{int_pion10}
\hspace*{-1cm}I_{q_1q_2, {\rm lcd}}^{if, \nu}(v,q)&=& - \varepsilon^*_\nu(q)
\int\hspace*{-0.1cm}\frac{d^4k_1}{\pi^2i} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{4\pi^2i} \hspace*{0.1cm}
\frac{\tilde F_B(k_1,k_2,0)\tilde F_B(k_1,k_2,-q)}
{[-k_1v^\prime-\bar\Lambda_{q_1q_2}]}
\Pi_{q_1q_2}^\gamma(k_1,k_2)
\nonumber
\end{eqnarray}
where
\begin{eqnarray}
\Pi_{q_1q_2}^{\gamma}(k_1,k_2)=
Q_{q_2q_2}{\rm tr}\biggl[\Gamma_2^i S_{q_2}(k_1+k_2)
\Gamma_2^f S_{q_2}(k_2)\biggr]
+Q_{q_1q_1}{\rm tr}\biggl[\Gamma_2^f S_{q_2}(k_1+k_2)\Gamma_2^i
S_{q_1}(k_2)\biggr]
\nonumber
\end{eqnarray}
and
\begin{eqnarray}\label{vertex_add2}
M_{\rm inv, ncd}^\gamma(B_{Q1}^i\to B_Q^f\gamma)&=&
eg^i_{\rm eff} g^f_{\rm eff}
\cdot \bar u(v^\prime)\Gamma_1^f \biggl[\frac{(1+\not\! v)}{2}
I_{q_1q_2, {\rm ncd}}^{if, \nu; L}(v,q) \nonumber\\
&+&\frac{(1+\not\! v^\prime)}{2}
I_{q_1q_2, {\rm ncd}}^{if, \nu; R}(v,q)\biggr]\Gamma_1^i u_\nu(v)
\end{eqnarray}
\begin{eqnarray}\label{int_pion20}
\hspace*{-1cm}I_{q_1q_2, {\rm ncd}}^{if, \nu; L}(v,q)&=&\varepsilon^*_\mu(q)
\int\hspace*{-0.1cm}\frac{d^4k_1}{\pi^2i} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{4\pi^2i} \hspace*{0.1cm}
(k_1-k_2+q)^\mu k_1^\nu \frac{\tilde F_B(k_1,k_2,0)}
{[-k_1v-\bar\Lambda_{q_1q_2}]}\nonumber\\
&\times&\frac{\tilde F_B(k_1,k_2,0)-\tilde F_B(k_1,k_2,q)}
{(k_1-k_2)q+q^2}\Pi_{q_1q_2}^\gamma(k_1,k_2)\\[2mm]
\nonumber\\
\hspace*{-1cm}I_{q_1q_2, {\rm ncd}}^{if, \nu; R}(v,q)&=&\varepsilon^*_\mu(q)
\int\hspace*{-0.1cm}\frac{d^4k_1}{\pi^2i} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{4\pi^2i} \hspace*{0.1cm}
(k_1-k_2-q)^\mu k_1^\nu \frac{\tilde F_B(k_1,k_2,0)}
{[-k_1v^\prime-\bar\Lambda_{q_1q_2}]}\nonumber\\
&\times&\frac{\tilde F_B(k_1,k_2,0)-\tilde F_B(k_1,k_2,-q)}
{(k_2-k_1)q+q^2}\Pi_{q_1q_2}^\gamma(k_1,k_2)\\[2mm]
\nonumber
\end{eqnarray}
As an illustration of our calculational procedure we evaluate a typical
integral corresponding to a one-photon transition (the same integral
for the one-pion transition can be found in Ref. \cite{IKLR2})
\begin{eqnarray}
R^{if; \mu}_{q_1q_2}(v,q) =
\int\hspace*{-0.1cm}\frac{d^4k_1}{\pi^2i} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{\pi^2i}
\hspace*{0.1cm}\frac{\tilde F_B(k_1,k_2,q)\tilde F_B(k_1,k_2,0)}
{[-k_1v-\bar\Lambda_{q_1q_2}]}
{\rm tr}[\Gamma_2^iS_{q_2}(k_1+k_2)\Gamma_2^fS_{q_1}(k_2-q)\gamma^\mu
S_{q_1}(k_2)]\nonumber
\end{eqnarray}
We have
\begin{eqnarray}
\hspace*{-1cm}& &R^{if; \mu}_{q_1q_2}(v,q)=
\int\limits_0^\infty \hspace*{-0.1cm}ds_1 \tilde F_B^L(6s_1)
\int\limits_0^\infty \hspace*{-0.1cm}ds_2 \tilde F_B^L(6s_2)e^{2s_2q^2}
\int\limits_0^\infty \hspace*{-0.1cm}d^4 \alpha
e^{\alpha_3\bar\Lambda-(\alpha_1+\alpha_4)m_{q_1}^2-\alpha_2m_{q_2}^2}
\nonumber\\
\hspace*{-1cm}&\times&{\rm tr}\biggl[\Gamma_2^i
\biggl(m_{q_2}-\frac{\not\!\partial_1+\not\!\partial_2}{2}\biggr)\Gamma_2^f
\biggl(m_{q_1}-\frac{\not\!\partial_2}{2}-\not\! q\biggr)\gamma^\mu
\biggl(m_{q_1}-\frac{\not\!\partial_2}{2}\biggr)\biggr]\nonumber
\int\hspace*{-0.1cm}\frac{d^4k_1}{\pi^2i} \hspace*{-0.1cm}
\int\hspace*{-0.1cm}\frac{d^4k_2}{\pi^2i} \hspace*{0.1cm}
e^{kAk-2kB}\nonumber
\end{eqnarray}
where the matrices A and B are defined by
\[A_{ij}=\left(
\begin{array}{ll}
\mbox{$2(s_1+s_2)+\alpha_2$} & \hspace*{.5cm} \mbox{$s_1+s_2+\alpha_2$}\\
\mbox{$s_1+s_2+\alpha_2$} & \hspace*{.5cm}
\mbox{$2(s_1+s_2)+\alpha_1+\alpha_2+\alpha_4$}
\end{array}
\right) \]
\[B_{i}=\left(
\begin{array}{l}
\mbox{$-s_2q-\alpha_3v/2$} \\
\mbox{$(s_2+\alpha_1)q$}
\end{array}
\right) \]
The integration over $k_1$ and $k_2$ results in
\begin{eqnarray}
\hspace*{-1cm}& &R^{if; \mu}_{q_1q_2}(v,q)=
\int\limits_0^\infty \hspace*{-0.1cm}ds_1 \tilde F_B^L(6s_1)
\int\limits_0^\infty \hspace*{-0.1cm}ds_2 \tilde F_B^L(6s_2)e^{2s_2q^2}
\int\limits_0^\infty \hspace*{-0.1cm}d^3\alpha
e^{\alpha_3\bar\Lambda-(\alpha_1+\alpha_4)m_{q_1}^2-\alpha_2m_{q_2}^2}
\nonumber\\
\hspace*{-1cm}&\times&{\rm tr}\biggl[\Gamma_2^i
\biggl(m_{q_2}-\frac{\not\!\partial_1+\not\!\partial_2}{2}\biggr)\Gamma_2^f
\biggl(m_{q_1}-\frac{\not\!\partial_2}{2}-\not\! q\biggr)\gamma^\mu
\biggl(m_{q_1}-\frac{\not\!\partial_2}{2}\biggr)\biggr]
\frac{e^{-BA^{-1}B}}{|A|^2}
\nonumber
\end{eqnarray}
Let us, as an example, choose
$\Gamma_2^i=\gamma^\nu$ and $\Gamma_2^f=\gamma^5$.
In the limit $qv=(m_i^2-m_f^2)/(2m_i) \approx 0$ where $m_i$
and $m_f$ are the masses of the initial and the final baryons, respectively,
we find
$$R^{{\rm VP}; \mu\nu}_{q_1q_2}(v,q)=
4i\varepsilon^{\mu\nu\alpha\beta}q^\alpha v^\beta
\cdot\int\limits_0^\infty\hspace*{-0.1cm}
\frac{d^3\alpha\alpha_1\alpha_3}{2|A|^2}
\tilde F_B^2(6z)\{m_{q_1}(A_{11}^{-1}+A_{12}^{-1})-m_{q_2}A_{12}^{-1}\}
$$
\[A_{ij}=\left(
\begin{array}{ll}
\mbox{$2+\alpha_2$} & \hspace*{.2cm} \mbox{$ 1+\alpha_2$}\\
\mbox{$ 1+\alpha_2$} & \hspace*{.2cm} \mbox{$2+\alpha_1+\alpha_2$}
\end{array}
\right),
\hspace*{.5cm}
A^{-1}_{ij}=\frac{1}{|A|}\left(
\begin{array}{ll}
\mbox{$2+\alpha_1+\alpha_2$} & \hspace*{.2cm} \mbox{$-(1+\alpha_2)$}\\
\mbox{$-(1+\alpha_2)$} & \hspace*{.2cm} \mbox{$2+\alpha_2$}
\end{array}
\right) \]
The evaluation of the other remaining matrix elements proceeds along similar
lines.
In the case of one-pion transitions the general expansion of the transition
matrix elements $M_{\rm inv}^\pi(B_Q^i\to B_Q^f\pi)$
into a minimal set of covariants reads
\begin{eqnarray}\label{matrix_el1}
& &\underline{\mbox{One-pion transitions}}\\
M_{inv}^\pi(\Sigma_c\to\Lambda_c\pi) &=&
\frac{1}{\sqrt{3}}g_{\Sigma_c\Lambda_c\pi}
I_1 \bar u(v^\prime) \not\! q \gamma_5 u(v)
\hspace*{.5cm}
\underline{\mbox{p-wave transition}}\nonumber \\
M_{inv}^\pi(\Sigma_c^*\to\Lambda_c\pi) &=&
g_{\Sigma^{*}_c\Lambda_c\pi} I_1
\bar{u}(v^\prime)q_{\mu}u^{\mu}(v)
\hspace*{.5cm}\underline{\mbox{p-wave transition}}\nonumber \\
M_{inv}^\pi(\Lambda_{c1; S}\to\Sigma_c\pi) &=&
f_{\Lambda_{c1; S}\Sigma_c\pi} I_3
\bar{u}(v^{\prime})u(v)
\hspace*{.5cm}\underline{\mbox{s-wave transition}}\nonumber \\
M_{inv}^\pi(\Lambda_{c1; S}^*\to\Sigma_c\pi) &=& \frac{1}{\sqrt{3}}
f_{\Lambda_{c1; S}^{*}\Sigma_c\pi} I_3
\bar{u}(v^{\prime})\gamma_5\not\! q q_{\mu}u^{\mu}(v)
\hspace*{.5cm}
\underline{\mbox{d-wave transition}}
\nonumber
\end{eqnarray}
where the $I_1$ and $I_3$ are the flavor factors
which are directly connected to the flavor coefficients
$C_{\rm flavor}$ (see Eq. (\ref{Pi1})) via the relations
$I_i = f_i \cdot C_{\rm flavor}$, $i=1$ or $3$. The sets of $I_i$ and $f_i$
are given in Table II.
We have also indicated the orbital angular momentum of the pion in
Eq. (\ref{matrix_el1}).
For the one-photon transitions one similarly has
\begin{eqnarray}\label{matrix_el}
& &\underline{\mbox{One-photon transitions}}\\
M_{inv}^\gamma(\Sigma_c\to\Lambda_c\gamma) &=&
\frac{2i}{\sqrt{3}}f_{\Sigma_c\Lambda_c\gamma}
\bar u(v^\prime) \not\! q \not\!\varepsilon^*(q) u(v)
\hspace*{.5cm}
\underline{\mbox{M1 transition}}\nonumber \\
M_{inv}^\gamma(\Sigma_c^*\to\Lambda_c\gamma) &=&
2f_{\Sigma_c^*\Lambda_c\gamma}
\bar u(v^\prime)\epsilon(\mu \varepsilon^* v q) u^\mu(v)
\hspace*{.5cm}
\underline{\mbox{M1 transition}}\nonumber \\
M_{inv}^\gamma(\Sigma_c^*\to\Sigma_c\gamma) &=&
2if_{\Sigma_c^*\Sigma_c\gamma}
\bar u(v^\prime) \frac{\gamma^\mu \gamma^5}{\sqrt{3}}
u^\nu(v) (q_\mu \varepsilon^*_\nu(q) - q_\nu \varepsilon^*_\mu(q))
\hspace*{.5cm}
\underline{\mbox{M1 transition}}\nonumber \\
M_{inv}^\gamma(\Lambda_{c1; S}\to\Lambda_c\gamma) &=&
\bar u(v^\prime) F_{\Lambda_{c1; S}\Lambda_c\gamma}
[g^{\mu\nu} vq - v^{\mu}q^{\nu}] \frac{\gamma^\nu\gamma^5}{\sqrt 3}
u(v)\varepsilon^*_\mu(q)
\hspace*{.5cm}
\underline{\mbox{E1 transition}}\nonumber \\
M_{inv}^\gamma(\Lambda_{c1; S}^*\to\Lambda_c\gamma) &=&
\bar u(v^\prime) F_{\Lambda_{c1; S}\Lambda_c\gamma}^*
[g^{\mu\nu} vq - v^{\mu}q^{\nu}] u^\nu(v)\varepsilon^*_\mu(q)
\hspace*{.5cm}
\underline{\mbox{E1 transition}}\nonumber
\end{eqnarray}
where the couplings are manifestly gauge invariant.
In writing down the one-photon coupling structure in Eq. (\ref{matrix_el})
we have omitted the E2 coupling in the $\Sigma_c^*\to\Lambda_c\gamma$
transition and the M2 coupling in the $\Lambda_{c1; S}^*\to\Lambda_c\gamma$
transitions as predicted by heavy quark symmetry \cite{KKP}. Since heavy quark
symmetry is manifest in our model calculation the coupling structure
(\ref{matrix_el}) is sufficient for our purposes.
In the heavy quark limit the coupling constants in
Eqs. (\ref{matrix_el1}) and (\ref{matrix_el}) become flavor independent.
Also some of the above coupling constants become related in the heavy quark
limit. The relations read \cite{KKP,Yan,Cho}
\begin{eqnarray}\label{rel_con}
g_{\Sigma_c\Lambda_c\pi}=g_{\Sigma^{*}_c\Lambda_c\pi}=g,
\hspace*{1cm}
f_{\Sigma_c\Lambda_c\gamma}=f_{\Sigma^*_c\Lambda_c\gamma}=f,
\hspace*{1cm}
F_{\Lambda_{c1; S}\Lambda_c\gamma}=F_{\Lambda^{*}_{c1; S}\Lambda_c\gamma}=F
\end{eqnarray}
Returning to our model calculation the coupling constant $f$ can be
represented as
\begin{eqnarray}
f&=&(\mu_1-\mu_2)
\frac{R_{\Sigma_Q\Lambda_Q\gamma}}{\sqrt{R_{\Lambda_Q}}\sqrt{R_{\Sigma_Q}}}
\\[2mm]
R_{\Sigma_Q\Lambda_Q\gamma}&=&\frac{1}{4}
\int\limits_0^\infty\hspace*{-.1cm}d^3\alpha\alpha_3(\alpha_1+\alpha_2)
\tilde F^2_B(6z)\frac{A_{11}^{-1}}{|A|^2}
\nonumber\\[2mm]
R_{B_Q}&=&
\int\limits_0^\infty\hspace*{-.1cm}d^3\alpha\alpha_3
\frac{\tilde F^2_B(6z)}{|A|^2}
\biggl\{1+d_{B_Q}\frac{\alpha_3}{m^2_q}\frac{\partial z}{\partial\alpha_3}
-\frac{\alpha_3^2}{4m^2_q}A_{12}^{-1}(A_{11}^{-1}+A_{12}^{-1})\biggr\}
\nonumber
\end{eqnarray}
where $\mu_i=e_i/(2m_q)$ is the magnetic moment of the i-th light quark.
Here
$$
z=\frac{\alpha_3^2}{4}A_{11}^{-1}+m^2_q(\alpha_1+\alpha_2)
-\bar\Lambda\alpha_3,
\hspace*{1cm}
d_{B_Q}=\left\{
\begin{array}{ll}
1 & \,\,\, \mbox{for} \,\,\, B_Q=\Lambda_{Q}\\
\frac{1}{2} & \,\,\, \mbox{for} \,\,\, B_Q=\Sigma_{Q}\\
\end{array}
\right.
$$
The calculation of the other coupling factors proceeds along similar lines.
In addition to the relations (\ref{rel_con}) there is the identity
between $f_{\Sigma_c^*\Lambda_c\gamma}$ and $f_{\Sigma_c^*\Sigma_c\gamma}$
couplings obtained in the constituent quark model \cite{HKLT,HKT}:
$$f_{\Sigma_c^*\Lambda_c\gamma}/f_{\Sigma_c^*\Sigma_c\gamma} =
(\mu_1-\mu_2)/(\mu_1+\mu_2)=3$$
In our model the light diquark current in $\Sigma_c$-baryon
is different than that in $\Lambda_c$-baryon (see TABLE I).
However numerically the latter relation are reproduced with an
accuracy 1$\%$.
One can then go on and calculate the one-pion and one-photon decay rates
using the general formula
\begin{eqnarray}\label{rate1}
\Gamma (B_Q^i\to B_Q^fX)
= \frac{1}{2J+1} \quad \frac{ \mid \vec{q} \mid}{8 \pi
M_{B_Q}^{2}}\sum_{spins} \mid M^{\pi}_{inv} (B_Q^i\to B_Q^fX) \mid^{2}
\end{eqnarray}
where $\mid \vec{q} \mid $ is the pion (photon) momentum in the rest frame
of the decaying baryon. In terms of the coupling constants (\ref{rel_con})
one obtains
\begin{eqnarray}\label{all_rates}
\Gamma\left( \Sigma_c \rightarrow \Lambda_c \pi \right)&=& g^2I_1^2
\frac{{\mid \vec{q} \mid}^3}{6\pi} \frac{M_{\Lambda_c}}{M_{\Sigma_c}}
\nonumber\\
\Gamma\left( \Sigma^*_c \rightarrow \Lambda_c \pi \right)&=& g^2I_1^2
\frac{{\mid\vec{q}\mid}^3}{6\pi} \frac{M_{\Lambda_c}}{M_{\Sigma^*_c}}
\nonumber\\
\Gamma\left(\Lambda_{c1; S}\rightarrow\Sigma_c\pi \right)
&=&f^2_{\Lambda_{c1; S}\Sigma\pi} I_3^2\frac{\mid \vec{q} \mid}{2\pi}
\frac{M_{\Sigma_c}}{M_{\Lambda_{c1; S}}}\nonumber\\
\Gamma\left(\Lambda^{*}_{c1; S}\rightarrow\Sigma_c\pi\right)
&=& f^2_{\Lambda^{*}_{c1; S}\Sigma\pi} I_3^2
\frac{{\mid\vec{q}\mid}^5}{18\pi}\frac{M_{\Sigma_c}}{M_{\Lambda^*_{c1; S}}}
\\
\Gamma\left(\Sigma_c\rightarrow \Lambda_c \gamma \right)&=&
\frac{4}{3\pi} f^2 {\mid \vec{q} \mid}^3 \frac{M_{\Lambda_c}}{M_{\Sigma_c}}
\nonumber\\
\Gamma\left(\Sigma^*_c\rightarrow \Lambda_c\gamma\right)&=&
\frac{4}{3\pi} f^2 {\mid \vec{q} \mid}^3 \frac{M_{\Lambda_c}}{M_{\Sigma_c^*}}
\nonumber\\
\Gamma\left(\Sigma^*_c\rightarrow \Sigma_c\gamma\right)&=&
\frac{4}{9\pi} f^2_{\Sigma_c^*\Sigma_c\gamma}
{\mid \vec{q} \mid}^3 \frac{M_{\Sigma_c}}{M_{\Sigma_c^*}}
\nonumber\\
\Gamma\left(\Lambda_{c1; S}\rightarrow\Lambda_c\gamma \right)&=&
\frac{1}{3\pi} F^2 {\mid \vec{q} \mid}^3
\frac{M_{\Lambda_c}}{M_{\Lambda_{c1; S}}}
\nonumber\\
\Gamma\left(\Lambda_{c1; S}^*\rightarrow\Lambda_c\gamma \right)&=&
\frac{1}{3\pi} F^2 {\mid \vec{q} \mid}^3
\frac{M_{\Lambda_c}}{M_{\Lambda_{c1; S}^*}}
\nonumber
\end{eqnarray}
The unknown masses of the excited bottom baryons
$\Lambda_{b1; S}$ and $\Lambda_{b1; S}^*$
are estimated from the heuristic relation:
$m_{\Lambda_{b1; S}^{(*)}} = m_{\Lambda_{c1; S}^{(*)}} +
(m_{\Lambda_{b}^{0}} - m_{\Lambda_{c}^{+}})$.
\section{Results}
We now present our numerical results for
the strong and radiative transitions of heavy baryons.
In Table III we list our results for the one-pion coupling constants.
For comparison we also give the results of the Light-Front (LF) Quark Model
\cite{TOK} where the masses of light quarks are varied in the range
$m_{u(d)}=280\pm 60$ MeV and $m_s=m_{u(d)}+150$ MeV $=430\pm 60$ MeV.
Our predictions for the one-pion coupling constants
are in qualitative agreement with the Light-Front quark model prediction.
As in the Light-Front
quark model \cite{TOK} the strength of single pion couplings of the
charmed baryon ground-state to the antisymmetric P wave states are
suppressed with respect to those of the symmetric multiplet.
The coupling values calculated in the Light-Front approach show a
strong dependence on the masses of the light quarks.
The results of Ref. \cite{TOK} are quite close to our results for
the mass choice $m_{u(d)}=220$ MeV and $m_s=370$ MeV. However, such low values
of the constituent quark masses are excluded from the analysis of the magnetic
moments and charge radii of nucleons.
The smaller value of $f_{\Lambda_{c1; S}^{*}\Sigma_c\pi}$
is welcome when one compares the results for exclusive one-pion rates of the
$\Lambda_{c1; S}^{*}$ (see, Table IV) with experimental
data \cite{PDG}. It is seen that our predictions are consistent with current
experimental estimates whereas the Light-Front model results lie above
the experimental rates. Also our predictions for $d$-wave
transitions can be seen to be consistent with the bound on the total rate of
the $\Lambda_{c1; S}^*$ by summing up the three exclusive one-pion
rates of the $\Lambda_{c1; S}^{*}$ and comparing the sums
to the total experimental rate $\Gamma({\Lambda^{*}_{c1; S}}) <1.9$ MeV.
From the results of our model
calculation we obtain $\Gamma(\Lambda_{c1; S}^{*}) > 0.25 \pm 0.03 $ MeV
consistent with the experimental results.
All our results for the one-pion decay rates
of charm baryons are collected in Table IV.
The uncertainties for the calculated rates
reflect the experimental errors in the charm baryon masses (see, Table I). For
comparison we have also listed the predictions of the Light-Front quark
model \cite{TOK} and experimental results, where available.
More precise data on the one-pion transitions of the
excited $\Lambda_{c1; S}$ baryon states to the ground states will allow for
a more detailed comparison with the model
predictions of dynamical models such as described in our approach and
in the Light-Front quark model.
A few comments should be done concerning the relations of RTQM results to
other approaches. The first comment concerns the baryon one-pion
coupling constants appearing in Eq. (18). The chiral formalism employed in
\cite{Yan} implies that all such constants are proportional to the factor
$1/f_\pi$ associated with the pion field. In RTQM approach they are described
by Eq.(9) and are proportional to the pion-quark coupling $g_\pi$ which is
determined by the compositeness condition discussed in Sec.II.
However, it appears that the Goldberger-Treiman relation $g_\pi=2m_q/f_\pi$ is
valid with an accuracy of a few percent. Hence the pion constant $f_\pi$
effectively appears as a dimensional parameter in our expressions for
the baryon one-pion coupling constants.
The second comment concerns the relation of the RTQM approach to
the constituent quark model based on the $SU(4)\otimes O(3)$ symmetry for
the light diquark system \cite{HKLT,HKT}.
Exploiting this symmetry one can reduce the number of the effective coupling
constants both for one-pion and one-photon transitions \cite{HKLT,HKT}.
For example, the ratio of the two coupling constants
$f_{\Lambda_{c1; S}\Sigma_c\pi}$ and $f_{\Sigma_{c0; S}\Lambda_c\pi}$
that govern the $s$-wave transitions
$\Lambda_{c1; S}\to\Sigma_c\pi$ and $\Sigma_{c0; S}\to\Lambda_c\pi$ is
predicted to be
$f_{\Lambda_{c1; S}\Sigma_c\pi}/f_{\Sigma_{c0; S}\Lambda_c\pi}=-1/\sqrt{3}$
in the $SU(4)\otimes O(3)$ model. We emphasize that
the $SU(4)\otimes O(3)$ symmetry is not realized in the RTQM approach.
For instance, differing from the $SU(4)\otimes O(3)$ approach the light
diquark current in $\Sigma_{c0; S}$-baryon
is not related to that in $\Lambda_{c1; S}$ (see TABLE I).
Yet, numerically the $SU(4)\otimes O(3)$ relations are reproduced with an
amazing accuracy of around 1$\%$. For the two above processes
$\Lambda_{c1; S}\to\Sigma_c\pi$ and $\Sigma_{c0; S}\to\Lambda_c\pi$ we find
numerically
$f_{\Lambda_{c1; S}\Sigma_c\pi}/f_{\Sigma_{c0; S}\Lambda_c\pi}=-0.58$ almost
identical to the ratio $-1/\sqrt{3}$ in the $SU(4)\otimes O(3)$ approach.
Note that the $SU(4)\otimes O(3)$ symmetry can be implemented explicitly
in the RTQM by replacing the light quark propagator by a constant value and
by modifying the spinor structure of the baryon-quark vertices
as was demonstrated in \cite{IKLR1}.
Numerical results for the one-photon decay rates are listed in Table V.
As for one-pion rates the errors in our rate values reflect the experimental
errors in the charm baryon masses \cite{PDG,CLEO4} (see Table I).
For the sake of comparison we also list the results of the model
calculations \cite{Cho}-\cite{Chow} mentioned earlier on. Our results are
quite close to the results of the nonrelativistic quark model \cite{Cheng1}.
In \cite{Cho} the coupling strengths were parameterized in terms of the
unknown effective coupling parameters $c_{RS}$ and $c_{RT}$.
A first rough estimate of the
unknown coupling parameters can be obtained by setting them equal to 1 on
dimensional grounds \cite{Cho}. As is evident from Table V such an estimate
is basically supported by our dynamical calculation. We do not agree with
the predictions on the charm and bottom p-wave decay rates of \cite{Chow}
except for the $\Lambda^*_{b1; S}\to \Lambda_b^0\gamma$ rate where we are
closer to the rate calculated in \cite{Chow}.
Recently the radiative decays of bottom baryons were studied with the use
of the light-cone QCD sum rules \cite{ZD} in the leading order of heavy
quark effective theory. For the decay rates of the $\Sigma_b$ and
$\Sigma_b^*$ baryons to $\Lambda_b^0\gamma$ the authors of \cite{ZD}
obtained
$$
\Gamma(\Sigma_b\to\Lambda_b\gamma)=\alpha_{\rm eff}{\mid \vec{q} \mid}^3
\,\,\,\,\, \mbox{and}\,\,\,\,\,
\Gamma(\Sigma_b^*\to\Lambda_b\gamma)=\alpha^*_{\rm eff}{\mid \vec{q}\mid}^3
$$
where the couplings $\alpha_{\rm eff}$ and $\alpha^*_{\rm eff}$ are
approximately equal to each other. The authors of \cite{ZD} quote
$\alpha_{\rm eff}\approx\alpha^*_{\rm eff}\approx 0.03 $ GeV~$^{-2}$.
In order to compare our model results with the results in \cite{ZD} we set
$M_{\Lambda_Q}=M_{\Sigma_Q}$ in Eq. (\ref{all_rates}). We then obtain
$\alpha_{\rm eff} = 4f^2/(3\pi) \approx 0.015 $GeV$^{-2}$ which is one-half of
the prediction of Ref.~\cite{ZD}.
One has to remark that the rate of the $\Sigma_c^*\to\Sigma_c\gamma$ decay is
suppressed in comparison with the rate of
$\Sigma_c^*\to\Lambda_c\gamma$ decay as
\begin{eqnarray}
\frac{\Gamma\left(\Sigma^*_c\rightarrow \Sigma_c\gamma\right)}
{\Gamma\left(\Sigma^*_c\rightarrow \Lambda_c\gamma\right)}
= \frac{1}{3}
\biggl(\frac{f_{\Sigma_c^*\Sigma_c\gamma}}{f_{\Sigma_c^*\Lambda_c\gamma}}
\biggr)^2 \frac{M_{\Sigma_c}}{M_{\Lambda_c}}
\biggl(\frac{M^2_{\Sigma_c^*}-M^2_{\Sigma_c}}
{M^2_{\Sigma_c^*}-M^2_{\Lambda_c}}\biggr)^3 \approx 10^{-3}
\end{eqnarray}
This estimate coincides with the prediction of the constituent quark
model \cite{HKLT,HKT}.
\section{Conclusion}
Using the same Relativistic Three-Quark Model (RTQM) employed before
successfully in studies of semileptonic decays of heavy baryons \cite{ILKK},
we have analyzed strong (one-pion) and radiative decays of heavy baryons.
We have obtained predictions for the values of couplings of charmed
baryons with pions and for the rates of the one-pion transitions
$B^i_c(p)\to B^f_c(p^\prime)+\pi(q)$. We have compared our results with
those obtained in Light-Front Quark-Model \cite{TOK}.
We found that as in the Light-Front
quark model \cite{TOK} the strength of single pion couplings of the
charmed baryon ground-state to the antisymmetric P wave states are
suppressed with respect to those of the symmetric multiplet.
We have also obtained predictions for the rates of the one-photon transitions
$B^i_Q(p)\to B^f_Q(p^\prime)+\gamma(q)$. We have compared our results with
the results of other model calculations~\cite{Cho}-\cite{Chow}. Unfortunately,
there is no data on the one-photon transitions yet to compare our results
with. For the one-photon decays
from the $p$-wave states $\Lambda_{c1; S}\to\Lambda_c+\gamma$ and
$\Lambda_{c1; S}^*\to\Lambda_c+\gamma$ our predicted rates are one order of
magnitude below the upper limits given by the experiments calling for an
one-order of magnitude improvement on the experimental upper limits.
Although the $\Xi_c^\prime \to\Xi_c + \gamma$ one-photon decays have now been
seen \cite{CLEO4} it will be close to impossible to obtain rate values for
these decays because the $\Xi_c^\prime$-states are far too narrow to be
experimentally resolvable. The total
widths of the $\Sigma_c$, $\Sigma_c^*$ and $\Xi_c^*$ states are larger
because of their strong decays via one-pion emission. In fact the widths of
the $\Sigma_c^{*++}$ and $\Sigma_c^{*0}$ have been determined \cite{PDG}. One
can hope that one-pion branching ratios can be experimentally determined for
the $\Sigma_c$, $\Sigma_c^*$ and $\Xi_c^*$ one-photon decay modes in the
near future. We are looking forward to compare the predictions of
the Relativistic Three-Quark Model for the calculated rates with
future experimental data.
\section*{Acknowledgments}
\noindent
M.A.I, V.E.L and A.G.R thank Mainz University for hospitality
where a part of this work was completed. This work was supported
in part by the Heisenberg-Landau Program
and by the BMBF (Germany) under contract 06MZ865. J.G.K. acknowledges
partial support by the BMBF (Germany) under contract 06MZ865.
|
\section{Introduction}
The Galactic Center (hereafter GC) region has
been observed at different spatial scales and energy bands
by many X--ray missions.
The main results in the 0.5--10 keV range have been obtained with
the Einstein Observatory (Watson et al., 1981), ROSAT (Predehl \& Trumper, 1994)
and, more recently, with ASCA (Maeda et al. 1996, 1998; Koyama et al. 1996).
At higher energies,
the telescopes ART--P (Pavlinsky, Grebenev \& Sunyaev 1994)
and SIGMA (Goldwurm et al. 1994) on--board the GRANAT satellite performed a monitoring
of the GC region leading to the discovery
of many new sources.
These observations
have shown that the GC X--ray emission comes
both from point sources
and from a diffuse component. Within a few degrees
from the direction of the GC region
there is a strong concentration of X--ray sources. The zoo
of these objects
is very rich and comprises both transient and
permanent sources, both ``standard" and very peculiar
objects, like e.g. the ``bursting pulsar'' GRO~J1744--28
(Lewin et al. 1996), the ``microquasar'' 1E~1740.7--2942 (Mirabel et al. 1992),
and the 2 ms pulsar SAX J1808.4--3658 (in't Zand et al. 1998a,
Wijnands \& van der Klis 1998).
The region
($|l|<2^\circ$)$\times$($|b|<0.5^\circ$)
around the GC
was observed with the $BeppoSAX$ satellite
during 1996--1998 for a total of $\sim$140 hours of
effective exposure time (Sidoli et al., 1998a, 1999).
In this paper we report on the observations of
the X--ray point sources in the GC region obtained with the $BeppoSAX$ Narrow
Field Instruments
(for the $BeppoSAX$ Wide Field Camera results
see, e.g., Ubertini et al. (1999)).
Although our observations cannot equal the spatial resolution
obtained with ROSAT at lower energies, they provide good images with
adequate spectral resolution above a few keV.
The sensitivity is similar to that obtained with ASCA, that
thanks to its solid state detectors has a
higher spectral resolution.
On the other hand, the BeppoSAX mirrors provide a narrower and
more regular point response
function with respect to that of the ASCA telescopes.
This allows an accurate analysis of the spatial morphology
of this complex sky region and
reduces the problem of stray light contamination from strong
sources outside the field of view.
The detailed results on
the X--ray diffuse emission from SgrA East and West
and from the molecular
clouds in this region (SgrB2, SgrC, SgrD) will be
reported in a separate paper.
\section{Observations}
The $BeppoSAX$ satellite (Boella et al. 1997a) carries a complement of
several imaging and non-imaging X--ray detectors, covering a broad
energy range from 0.1 keV to 300 keV.
Most of the results presented here have been obtained with the imaging
instruments placed in the focal planes of four identical X--ray Mirror Units:
the Medium--Energy Concentrator Spectrometer
(MECS, Boella et al. 1997b) and the
Low--Energy Concentrator Spectrometer (LECS, Parmar et al. 1997).
Each Mirror Unit consists of 30 nested, confocal mirrors with a double
cone approximation to the Wolter I geometry and a focal length of 185 cm.
The MECS instrument is based on three position--sensitive
gas--scintillation proportional counters, providing 1.3--10 keV images
over a field of view with $\sim28'$ radius.
The MECS is characterized by a total
(three telescopes) effective area of $\sim$150~cm$^2$ at 6 keV,
a relatively good angular resolution
(50\% power radius of $\sim75''$ at 6 keV, on-axis) and
a moderate energy resolution (FWHM $\sim$8.5$\sqrt{6/{\rm E_{keV}}}$\%).
One (M1) of the three nearly identical units composing the MECS had a failure
in May 1997; all the observations performed after that date were carried out
only with the remaining two units (M2 and M3).
The fourth Mirror Unit is associated with the LECS instrument.
This consists of a position--sensitive
gas--scintillation proportional counter utilizing an
ultra thin, organic detector entrance window (1.25 $\mu$m) and a driftless configuration
to extend the low energy response down to 0.1 keV.
Its spatial and energy resolutions are similar to those of the MECS,
but the field of view is slightly smaller ($\sim18'$ radius).
Due to UV contamination leaking through the entrance window during
sunlit periods,
the LECS is operated only at satellite night time,
resulting in effective exposure times smaller than those of the other
$BeppoSAX$ instruments (typically $\sim$50\% of the MECS values).
For the analysis of 1E~1740.7--2942, which has a hard spectrum
extending above the MECS energy range
we have also used the Phoswich Detector System
instrument
(PDS, Frontera et al. 1997).
This non-imaging instrument points in the same direction of the
MECS and LECS telescopes and
covers the 15--300 keV energy range.
It consists of a square array of
four indipendent NaI(Tl)/CsI(Na) phoswich scintillation
detectors equipped with collimators defining a field of view
of $\sim$1.3$^\circ$ (FWHM).
There are two separate collimators, one for each pair of detectors,
that can be independently rocked off-axis to monitor the background.
During our observations the default rocking mode with offset angles of $\pm210'$
and dwell time 96 s was used.
The PDS provides an energy resolution of
$\sim$15\% at 60 keV (FWHM).
The log of our $BeppoSAX$ observations of the GC region
performed in 1996--98 is shown in Table~1.
The pointing directions were chosen to cover a $\sim4^\circ$ long region
at b=0$^\circ$,
as well as a few known sources slightly outside the galactic plane.
We chose to
partially overlap adjacent pointings, in order to continuously
cover the galactic plane with the central
part of the MECS field of view ($\sim8'$ radius),
where the instrument sensitivity and angular resolution are best.
As shown in Table 1, a few pointing directions were observed twice,
at time intervals of 6 or 12 months. This was partly planned,
to study the long term variability of the sources, and partly
resulted from operational constraints.
The exposure time achieved in the different regions is not uniform,
ranging from a minimum of 7 ks to $\sim$340 ks at the position of 1E~1743.1--2843.
An exposure-corrected mosaic of the MECS images in the 2--10 keV
energy range is shown in Fig.~1.
\placetable{tbl-1}
\section{Data Analysis}
Except where noted, the spectral analysis of all the point sources
reported here has been carried out as follows:
the LECS and MECS counts have been extracted from a circular region
with $4'$ radius centered at the position of the source and rebinned
in order to have at least 20 counts per energy bin.
The $4'$ radius corresponds to an encircled energy of 90\% .
In several cases we used a smaller extraction radius
($2'$, $\sim$60\% enclosed energy)
in order to reduce the problems due to source crowding or
unfavourable locations in the MECS field of view.
In any case the response matrices
appropriate to the adopted extraction radius and source position on the
detector have been used.
While these matrices include the elements
to properly derive the correct spectral shape of the sources observed off-axis,
the resulting fluxes can be affected by an uncertainty
in their absolute normalization that in some cases could reach $\sim$40\%.
From each spectrum we have subtracted a local background,
extracted from a source free region of the same observation:
in fact, the standard $BeppoSAX$ background spectra
obtained from blank field observations
underestimate the actual background present in the GC region, which is dominated
by the galactic diffuse emission.
The PDS data were reduced using the
SAXDAS software package (Version 1.2.0). The time intervals before and
after Earth occultations were excluded and the spurious spikes (Guainazzi \&
Matteuzzi 1997) filtered out. The background subtraction was performed
with the standard rocking collimator technique.
The background-subtracted spectra were rebinned in order to
have at least 100 counts for each energy channel and then analyzed using the
response matrix released on August 31, 1997.
The resulting LECS/MECS/PDS spectra have been
fitted with different models implemented in XSPEC (Version 10.00).
We took into account the known intercalibration normalizations
between the
different instruments on--board $BeppoSAX$ (Fiore, Guainazzi \& Grandi 1999),
using different relative normalizations
A$_{LECS}\sim$0.85, A$_{MECS}$=1 and A$_{PDS}\sim$0.8.
MECS spectra were extracted in the 2--10 keV energy range, while for the
LECS we used only the data below 2 keV
(we verified that using the LECS data above this
energy did not significantly
improve the results obtained with the MECS).
\section{Results}
Several previously known sources have been observed in our
$BeppoSAX$ pointings:
1E~1743.1--2843 (Watson et al. 1981),
the persistent black hole candidate 1E~1740.7--2942
(Sunyaev et al. 1991a),
the X--ray bursters SLX~1744--299, SLX~1744--300, A~1742--294, KS~1741--293
(Skinner et al. 1990; Pavlinsky, Grebenev \& Sunyaev 1994; Kawai et al. 1988),
and the source(s) at the GC position
(Watson et al. 1981).
We also detected two X--ray transients discovered very recently:
SAX~J1747.0--2853 (in't Zand et al. 1998b) and XTE~J1748--288
(Smith, Levine \& Wood 1998), and
discovered a new X--ray source that we identified as the plerionic
emission from the composite supernova remnant G0.9+0.1 (Mereghetti,
Sidoli \& Israel 1998).
All the sources observed in our survey are listed in Table~2.
Note that in Table~2 we have reported the best source positions
and corresponding uncertainties
available in the literature.
The positions derived from our analysis have typical
uncertainties of the order of $\sim1'$
(95\% confidence level) and are in agreement with those
of Table~2.
We have listed in Table 2 also the positions of a few other
objects which are relevant in the following discussion
of the sources near SgrA*.
The results for the individual sources are reported in the following
sections, while in Table~3 we have summarized
the parameters of the best fits.
All the luminosities are given for an assumed
distance of 8.5 kpc, unless stated differently.
\placetable{tbl-2}
\placetable{tbl-3}
\subsection{The source(s) at the Galactic Center position (1E~1742.5--2859)}
The GC field was observed for 100 ks in August 1997 (Obs. n.3).
Strong X--ray emission peaked at the GC position
is clearly detected with the LECS and MECS instruments.
The central part of the MECS image is shown in Fig.~2, where the positions
of the X--ray sources previously detected with other satellites have been
indicated.
The background subtracted radial profiles of the X--ray emission in the
soft (2--5 keV) and hard (5--10 keV) energy ranges
are shown in Fig.~3.
The data points represent the surface brightness measured in concentric
rings centered at the position of SgrA*, while the solid
lines show for comparison
the profiles expected from a single unresolved point source.
The expected profiles have
been normalized to yield the same number of
counts within 10$'$ as the measured data.
It is clear that the shape of the observed profiles
is incompatible with
a single point--like source.
The observed radial profile results from the contribution of
a diffuse component and one (or possibly more) sources close
to the SgrA* position.
\placefigure{fig2}
\placefigure{fig3}
The X--ray images of this region with the best angular resolution available so far are those
obtained with the ROSAT PSPC instrument
(Predehl \& Trumper, 1994) which, in addition to the diffuse emission,
showed the presence of three different
sources. Their error regions (20$''$ radius) are indicated in Fig.~2 with the
small circles labelled PT6, PT7 and PT8.
One of these sources (PT7)
is highly absorbed and located within $10''$ from Sgr A*, while
PT6 has been interpreted as a foreground star, due to its smaller
column density.
It is evident that the emission detected with $BeppoSAX$ is peaked
at the position of PT7, although we cannot exclude that also the other
ROSAT sources (and AX~J1745.6--2901, see below) contribute to some
of the detected X--rays.
Though the limited angular resolution hampers a detailed analysis
and introduces unavoidable uncertainties, we proceeded as follows to
estimate the spectrum and flux to be ascribed to
the GC point source(s).
A spectrum was extracted from a circular region with $2'$ radius
centered at the position of the X--ray peak flux (see Fig.~2).
The background was taken from an external annular region ($6'-8'$), in order
to subtract the contribution from the diffuse emission.
A power law
does not give an appropriate fit to the resulting spectrum, which clearly
shows evidence for emission lines at 6.7 keV and also at lower energies,
indicative
of a possible thermal origin of the emission.
The sulfur line at $\sim2.4$ keV is particularly bright.
A power law with a gaussian line at 6.7 keV gives a better result, with
a photon index
$\Gamma\sim2.6$ and a column density
$N_H\sim8\times 10^{22}$cm$^{-2}$ ($\chi^2=1.1$).
When the centroid energy and width of the gaussian are let free in the fit,
we obtain the values E=6.7 keV, $\sigma$=13 eV and an equivalent width
EW$\sim1.2$ keV.
The same results were found fixing the $\sigma$ of
the iron line at 0. No evidence for a 6.4 keV line of
fluorescent origin is present, contrary to what is found in the
surrounding diffuse emission (a detailed study of which is deferred to a
separate paper).
The 2--10 keV flux corrected for the absorption is
$\sim4\times 10^{-11}$ ergs cm$^{-2}$ s$^{-1}$ and the
luminosity is $\sim3\times 10^{35}$ ergs s$^{-1}$ at a distance of 8.5 kpc.
\placefigure{fig4}
We also fitted the data with a thermal plasma model (MEKAL in XSPEC, Mewe,
Gronenschild \& van den Oord 1985). It properly
accounts for the 6.7 keV iron emission line, but leaves some
residuals around 2.4 keV (see Fig.~4), possibly indicating the presence of
a multitemperature
plasma.
In order to properly account for the emission lines at lower
energies, we fitted the spectrum with the sum of two MEKAL components with
different temperatures (kT$_{M1}\sim0.8$ keV and kT$_{M2}\sim5$ keV),
obtaining a higher N$_{H}$.
Also a single MEKAL model (kT$_{M}\sim1.3$ keV) plus a power law with
a gaussian added at 6.7 keV fit well the data.
All the spectral results are summarized in Table~3.
\subsection{AX J1745.6--2901}
During observations of the GC region performed
with the ASCA satellite in 1993 and 1994
a possibly new source was discovered $\sim1.3'$ SW
of SgrA* (Maeda et al. 1996).
The spectrum of this source, later named AX~J1745.6--2901 (Maeda et al. 1998),
was harder than that of the X--ray emission
coming from the position of Sgr~A*.
AX~J1745.6--2901 varied by a factor $\sim$5 in flux between
the two oservations separated by one year, and it showed an X--ray burst
and periodic (8.4 hr) eclipses (Maeda et al. 1996).
These authors proposed that AX~J1745.6--2901 could be the
quiescent counterpart of the bright soft X--ray transient
A1742-289 observed in outburst in 1975 (Eyles et al. 1975,
Branduardi et al. 1976).
However, the absence of eclipses at the 8.4 hr period in the
Ariel V archival data of the A1742-289 outburst (Kennea \& Skinner 1996),
makes this association uncertain. It is therefore likely that
AX~J1745.6--2901 is a different neutron star LMXRB.
In order to search for the possible presence of AX~J1745.6--2901
in our observations, we compared the MECS images accumulated in the
soft (2--6 keV) and hard (6--10 keV) energy ranges.
This was done independently for the two MECS units (M2 and M3)
to obtain the best angular resolution by avoiding possible small
misalignement errors.
Furthermore, we selected only time intervals in which the $BeppoSAX$
attitude was determined with two star trackers
(indeed small deviations $\lsim2'$ may be present
when no star tracker is active or when only one is in use).
\placefigure{fig5}
The resulting images for the M3 unit are shown in Fig.~5.
In the high energy image, an excess emission in the SW
direction is clearly resolved from the softer (and partly diffuse)
emission due to the 1E~1742.5--2859 (GC) source(s).
The position of this excess is consistent with
that of AX~J1745.6--2901.
It is visible only
in the higher energy channels of M3: in fact the higher background
due to misplaced events from the iron calibration sources prevents
its detection also in M2 (Chiappetti et al. 1998).
The weakness of the source and the presence of the nearby much brighter
X--ray emission hampers
an accurate determimation of its flux and spectrum,
which, in any case, is consistent with the lower value observed
with ASCA in 1993 (Maeda et al. 1996).
\subsection{1E~1740.7--2942}
We obtained two observations (Obs. n. 1 and 6) pointed on the
``microquasar'' source 1E~1740.7--2942 (Mirabel et al. 1992).
Our spectral analysis is based on the longer observation performed
in September 1997.
As shown by previous observations with coded mask imaging
instruments in the hard X--ray band (Skinner et al. 1987,
Goldwurm et al. 1994), 1E~1740.7--2942 is the strongest persistent source at
energies greater than $\sim$30 keV within a few degrees from the
GC. We therefore analyzed also the PDS data, in which
the source was clearly detected up to $\sim$150 keV. Due to the crowding of
this region of the sky, particular care was
devoted to the analysis of this non-imaging detector.
The PDS field of view includes the positions of the
persistent LMXRB A~1742--294 and of the transient KS~1741--293.
As shown by our simultaneous MECS data, the latter source was not active
at the time of observation.
A~1742--294 lies at $\sim30'$ from 1E~1740.7--2942 and it has
a similar flux in the 2--10 keV range. However,
it has a much softer spectrum (see Table~3).
Using our best fit thermal bremsstrahlung results and taking into account
the reduced off-axis response of the PDS collimator, we estimate
that A~1742--294 contributes less than 10\% to the observed
counts above 40 keV.
A further problem arised by the contamination of one of
the two offset collimator positions used in the background determination
by the bright X--ray source GX~3+1.
We therefore estimated the PDS background using only one of the two
off-axis collimator positions.
A single power law provides an adequate fit to the
LECS and MECS spectrum below 10 keV (photon index
$\Gamma\sim$1.5, $N_{H}\sim$1.5$\times$10$^{23}$ cm$^{-2}$).
We searched in the MECS data for the possible presence of iron lines
obtaining a negative result: for lines in the range 6.4--6.7 keV
we
can give a 90\% upper limit to the equivalent width of $\sim30$ eV
(with a line width $\sigma$ fixed at 0). This upper limit
increases up to $\sim 50$ eV for $\sigma$ = 500 eV. These results are
similar to the findings of Sakano et al. (1999) who
recently reanalyzed all the ASCA data on 1E~1740.7--2942.
However, a single power law cannot provide an adequate
fit over the broad energy range covered by BeppoSAX. In fact the
inclusion of the PDS data clearly shows the presence of a spectral
turn over at higher energies (see Fig. 6).
Good fits could be obtained with a power law with an exponential cut-off
or with Comptonization models (see Table~3).
1E 1740.7-2942 is of particular interest since,
beside being one of the few galactic objects with persistent hard
X-ray emission extending above $\sim$100 keV,
it has a peculiar radio counterpart with double sided relativistic jets
(Mirabel et al. 1992).
The observation of a transient broad spectral feature
at $\sim$480 keV (Bouchet et al. 1991),
suggested that 1E 1740.7-2942
might be related to the 511 keV line observed from the GC region. On the
basis of these properties and of its hard X-ray spectrum similar to that of Cyg ~X-1,
1E 1740.7-2942 is generally considered a black hole candidate.
The broad band spectrum measured with BeppoSAX is consistent with
this interpretation.
\placefigure{fig6}
\subsection{A~1742--294}
A~1742--294, although sometimes referred to as a transient source,
has always been detected in all the GC observations
since 1975 (see references in van Paradijs 1995).
It was observed twice in the
outer part of the MECS field of view (Obs. n. 8 and 9).
In the 1997 observation A~1742--294 was very close to the
region of the field of view used for the on board calibration
sources. So we extracted the counts for the spectral
analysis from a small circular region with $2'$ radius.
In both observations acceptable fits were
obtained both with a power law and with
a thermal bremsstrahlung (see Table~3). In March 1998
the source was softer and brighter (of
roughly a factor of two). Moreover, some variability
on a time scale of hours was observed during
both observations (see the light curves in Fig. 7).
We detected three type I X--ray bursts from A~1742--294:
one on September 16 (20:04:04 UT) and two on March 31, 1998
(9:44:58 UT and 15:01:20 UT).
The poor statistics hampers a detailed
study of the neutron star luminosity, radius and temperature
variations during the bursts.
\placefigure{fig7}
\subsection{SLX~1744--299 and SLX~1744--300}
SLX~1744--299, first discovered with the Spartan 1 experiment
(Kawai et al. 1988), was subsequently resolved into two distinct
objects separated by $\sim2.5'$
with the SL2-XRT coded mask telescope
(Skinner et al. 1987, 1990).
We obtained a single observation pointed on these two sources
in September 1997 (Obs. n.7). For the spectral analysis we used extraction
radii of only $2'$, and the same background estimated from a circular
corona surrounding both sources. Though the results have to be taken
with some caution due to the unavoidable cross-contamination of the
two spectra, we can safely conclude that SLX~1744--300 is fainter
and has a slightly softer spectrum than SLX~1744--299.
However, contrary to Predehl et al. (1995),
we cannot claim a significant difference in their absorption column
densities (see results in Table~3).
Though both SLX~1744--299 and SLX~1744--300 have been reported to
emit X--ray bursts (Skinner et al. 1987, 1990; Sunyaev et al. 1991b), none
was detected during our observation.
\subsection{XTE~J1748--288}
This transient source was discovered with the RossiXTE satellite in June
1998 (Smith, Levine \& Wood 1998). Its hard X--ray spectrum and its
association
with a variable radio source with evidence for the presence of jets
establish XTE~J1748--288 as a possible black hole
candidate
(Strohmayer \& Marshall 1998, Hjellming et al. 1998, Fender \& Stappers
1998).
We detected XTE~J1748--288 in the August 1988 observation (Obs. n.11).
Its spectrum, extracted from a $2'$ radius region to avoid the
problems due to the poorly known additional absorption from
the strongback support of the MECS detector entrance window,
could be fit equally well by a hard power law
(photon index $\sim$1.5) or by a thermal bremsstrahlung with kT$\sim$40
keV.
The flux was $\sim$10$^{-10}$ erg~cm$^{-2}$ s$^{-1}$
(2--10 keV, corrected for the absorption).
The light curve of the XTE~J1748--288 outburst
obtained with the RossiXTE ASM is shown in Fig.~8.
The date of our observation and the measured flux
are indicated by the dashed lines, showing that we obtained a
positive detection of XTE~J1748--288 when it was well below the ASM
sensitivity threshold.
It is interesting to note that the flux measured with
$BeppoSAX$ is consistent with the extrapolation of
the decaying light curve observed with the ASM
until about mid-July.
This suggests that the nearly exponential decay, with e-folding
time $\sim$20 days, lasted until the date of our observation
(August 26th).
\placefigure{fig8}
\subsection{SAX~J1747.0--2853}
The source SAX~J1747.0--2853,
positionally consistent with the transient GX~0.2--0.2
observed in 1976 (Proctor, Skinner \& Willmore 1978), was discovered
with the $BeppoSAX$ Wide Field Camera (WFC) instrument
during an outburst in March 1998 (in't Zand et
al. 1998b, Bazzano et al. 1998).
\placefigure{fig9}
During our observation pointed on 1E~1743.1--2843 (Obs. n.10),
SAX~J1747.0--2853 was detected at an off-axis angle of $\sim13'$.
During this observation, performed about 20 days
after the WFC ones,
a Type-I X--ray burst from this source was seen (Sidoli et al. 1998b).
The burst, starting at 1:40 UT of 1998 April 15,
was also visible at energies above 10 keV with the PDS instrument.
The burst light curve in different energy ranges is shown in Fig.~9,
where we also present the variations of the spectral fit parameters
during the burst (see Sidoli et al. 1998b for details).
The spectral softening clearly indicates that the burst is of Type I,
confirming the LMXRB nature of this source.
Assuming an Eddington luminosity at the burst
peak, we obtain a distance of $\sim$10 kpc,
and a blackbody radius consistent with the expectations for a neutron star.
\subsection{KS~1741--293}
During our observation pointed on the molecular cloud SgrC performed
in March 1998 (Obs. n.9), we detected a source positionally coincident with
the transient burster KS1741-293 (in't Zand et al. 1998c).
Since the source was located close to the strongback support
of the MECS detector entrance window, we used a small
extraction radius ($2'$) for the spectral analysis.
The background was estimated locally from a source free region
of the same observation.
The spectrum is relatively soft and could be described equally well
by a power law and a thermal bremsstrahlung (see Table~3),
giving in both cases a high column density
$N_{H}\sim$2$\times10^{23}$ cm$^{-2}$, and a
luminosity L$_{X}\sim10^{36}$ erg~s$^{-1}$
(2--10 keV, corrected for absorption).
Slightly lower column densities and luminosity were obtained with
a blackbody spectrum (see Table~3).
The same region of sky was observed by $BeppoSAX$ in September 1997 (Obs. n.8)
and KS~1741--293 was not detected, with an upper limit of
L$_{X}$ $<$ 10$^{35}$ erg~s$^{-1}$.
\subsection{1E~1743.1--2843}
This bright and highly absorbed source has been repeatedly observed
in several of our pointings. The best fit results
reported in Table~3 refer to the
observation performed in April 1998 (Obs. n.11), in which 1E~1743.1--2843
was on-axis.
This source is probably a persistent LMXRB, though no X--ray
bursts have been detected so far.
For a complete discussion of all the $BeppoSAX$ and
ASCA observations of 1E~1743.1--2843 see Cremonesi et al. (1999).
\subsection{The supernova remnant G0.9+0.1}
The discovery of X--ray emission from the center of the radio
supernova remnant G0.9+0.1 has been reported by Mereghetti, Sidoli
\& Israel (1998).
The location of G0.9+0.1 was imaged in April and September 1997, during
observations pointed on the Sgr~B2 molecular cloud. The angular
resolution of the MECS at the off--axis location of G0.9+0.1 ($\sim14'$)
only allowed to establish that the X--ray emission is associated with the
plerionic radio core of the supernova remnant and not with the surrounding
shell ($\sim8'$ diameter, Helfand \& Becker 1987). However, it was not
possible to discriminate between a point source (i.e. a neutron star) and
a synchrotron nebula of a few arcmin extent.
We have performed a new spectral analysis based on the improved
calibration results that are now available.
The derived spectral parameters are within the
uncertainties of those previously reported
(Mereghetti, Sidoli \& Israel 1998), but they
are more precise.
The best fit power--law spectrum has a photon index
$\Gamma=2.5$, $N_H=2.5\times 10^{23}$cm$^{-2}$, and flux $F = 2\times
10^{-11}$ erg cm$^{-2}$ s$^{-1}$ (2--10 keV, corrected for the
absorption). Equally good fits were also obtained with blackbody
and thermal bremsstrahlung spectra (see Table~3), while a
Raymond-Smith thermal plasma model with abundances fixed at the solar
values gave a worse result.
The $BeppoSAX$ discovery of X--ray emission from the central region of
G0.9+0.1 confirms its plerionic morphology derived from the radio
observations. The presence of a young, energetic pulsar in G0.9+0.1 could
also explain part of the high energy gamma-ray emission observed with
EGRET from the region of the Galactic Canter (Mayer--Hasselwander et al.
1998).
\subsection{G359.23--0.92 (The Mouse)}
The radio source G359.23--0.92, also known as the Mouse,
belongs to a small class of
radio nebulae characterized by an axially symmetric morphology
(Yusef-Zadeh \& Bally 1987). It is believed that these sources are produced by
relativistic particles ejected by a compact object (either a
pulsar or a binary system) moving with high speed along the axis
of symmetry (Shaver et al. 1985, Helfand \& Becker 1985).
Predehl \& Kulkarni (1995) using the ROSAT PSPC instrument
discovered X--ray emission associated with the ``head'' of the Mouse,
located only $2.3'$ northeast of SLX~1744--299.
Since the MECS data of SLX~1744--299/300 showed some evidence
for an elongation in the direction of G359.23--0.92,
we performed the same procedure described
in section 4.2 for AX~J1745.6--2901.
The resulting 6--10 keV image, shown in Fig.~10, confirms the
MECS detection of G359.23--0.92.
\placefigure{fig10}
To estimate the spectral parameters of G359.23--0.92 we used a small extraction radius ($2'$) and
measured the background (which is dominated by the counts of the
nearby sources) from a similar region at the same distance from SLX~1744--299.
A power law fit gave $\Gamma\sim$ 1.9--2.3, a
column density of $4-6\times$10$^{22}$~cm$^{-2}$ and an unabsorbed flux in the 2--10 keV
band of the order of $3\times$10$^{-11}$ erg~cm$^{-2}$ s$^{-1}$.
This flux is higher than the value expected based on the ROSAT results
at energies below 2.4 keV (Predehl \& Kulkarni 1995), however this discrepancy
might be due to different reasons other than source variability.
First, the flux
reported by these authors was based on the assumption of a value of only
$2\times10^{22}$~cm$^{-2}$ for the column density: we found that for our N$_{H}$ value
the MECS and ROSAT PSPC count rate are consistent with the same flux level.
Another possibility is that we underestimated the contamination from SLX~1744--299/300.
In any case, the detection at energies greater than 6 keV is statistically significant and supports
a non-thermal origin for the X--ray emission from the head of the ``Mouse''.
\section{Discussion}
Our $BeppoSAX$ survey of the inner 4 degrees of the galactic
plane shows that a few persistently bright sources characterize
the constellation of
X--ray sources near the GC. Their properties indicate
that they consist of compact accreting objects similar to the classical
accretion powered sources found elsewhere in the Galaxy.
Both systems containing neutron stars (e.g. SLX~1744--299/300, A~1742--294)
and likely black holes (1E~1740.7--2942 ) are present.
However, these persistently bright objects represent only a small fraction of
the X--ray sources in this region. Three transient
sources (SAX~J1747.0--2853, KS~1741--293 and XTE~J1748-288) were
detected in our survey, but it is known from previous observations
that many more are present.
We therefore confirm
the clustering of X--ray sources towards the GC
already noted by several authors.
It is likely that on the large scale ($\sim$ degrees) this simply
reflect the general mass distribution enhancement, as found
by Grebenev, Pavlinsky \& Sunyaev (1996) who considered only the
hard X--ray sources detected with the ART-P coded mask telescope.
On the other hand, considering the distribution of all the sources
reported within several arcminutes from SgrA* there is
marginal evidence of an additional source population
(Skinner 1993). However, this conclusion might be biased by the
different sensitivity and angular resolution of the many observations
of this field. Though it is obvious that the regions closer to the
center have been observed more deeply, the effect of this bias is
difficult to quantify.
We finally note that the case of the three sources in the direction of
SLX J1744--299/300 (at b = --1.5$^\circ$ ) is maybe more surprising that the presence of 4-5
sources close to SgrA*. While G359.23--0.92 is very likely a foreground
object, the two X--ray bursters SLX J1744--299 and SLX J1744--300
could be at the same distance and therefore somehow related
(e.g. members of a star cluster). Observations in X-rays with better
angular resolution and deep searches in the infrared are needed
to study these objects.
\placefigure{fig11}
In Fig.~11 we have plotted the column density and power law photon index
(in the 2--10 keV range) for all the sources that we have observed
(to compare the spectra of the different sources,
we have used a power law also in a few cases in which other models
provided better fits).
It is clear that sources containing neutron stars, as indicated
by the presence of Type I bursts, have softer spectra than the
black hole candidates. Based on this spectral dichotomy,
it is likely that also 1E~1743.1-2843
contain a neutron star.
Fig.~11 also shows that the column densities of these sources
vary over a factor $\sim$4 range.
This is probably related to
the clumpiness of the molecular clouds in this region rather than to
a corresponding spread in the source distances.
\subsection{Sgr A* and the galactic center X--ray source(s)}
At least four point sources (Predehl \& Trumper 1994, Maeda et al. 1996),
maybe five (Kennea \& Skinner 1996) and possibly more could
contribute to the X--ray flux detected with $BeppoSAX$
when pointing at the GC,
in addition to the diffuse emission
that permeates the region (Maeda \& Koyama 1996).
However, as shown in Fig~2, the MECS source position is closer to
the ROSAT counterpart
of SgrA* than to the other two ROSAT sources (PT6 and PT8).
The emission lines detected in the MECS spectrum within $2'$ from the GC
clearly indicate a substantial
contribution from the thermal diffuse emission associated to the
SgrA West and SgrA East regions (Yusef-Zadeh et al. 1997).
Unfortunately, this contribution is difficult to quantify, and the limited angular
resolution does not allow to measure the flux to be ascribed to SgrA* itself.
We can place a very conservative upper limit to the luminosity of SgrA* by assuming
that it is the only point source and that the diffuse emission at its position
has the same surface brightness of the annular region ($6'-8'$)
described above. In this case, using the flux reported in Table~3,
we obtain a luminosity of $\sim3\times10^{35}$~ergs~s$^{-1}$
(2--10 keV, corrected for the absorption).
However, this is probably an overestimate of the true luminosity of SgrA* for the fact
that the diffuse emission has a surface brightness distribution that increases
toward the GC and because
there might be a contribution from the other point sources.
A more realistic upper limit to the X--ray emission from SgrA*
can be placed if we consider an X--ray background rising towards the GC:
we measured the surface brightness from three concentric
annular regions ($2'-4'$, $4'-6'$ and $6'-8'$)
and extrapolated it at our source position.
In this case we find a new background value, four times higher than the previous one
and the
new flux for the point source is
$\sim1.5\times 10^{-11}$ ergs cm$^{-2}$ s$^{-1}$, that translates into
a luminosity at 8.5 kpc of $\sim10^{35}$~ergs~s$^{-1}$.
This upper limit is similar to the value reported
by the ASCA team (Maeda et al., 1998) and is
well below the Eddington luminosity
for a super--massive black hole.
Several authors tried to explain this high energy
underluminosity of SgrA* with different models for the accretion,
from a spherical accretion
(Melia 1992)
to the advection dominated accretion flow
(Narayan, Yi \& Mahadevan 1995; Narayan et al. 1998). These models
have still some problems in the radio and in the $\gamma$--ray bands (EGRET source
2EG~1746--2852), where substantial
excesses are present. Several solution has been
proposed: the inclusion of an emission process
associated with protons (Mahadevan et al. 1998) is able to solve the problem
of the radio excess, while the EGRET source is considered as a gamma--ray
upper limit due to the poor spatial resolution. Another possible explanation is
that the EGRET source is not related with SgrA*, but with the
SgrA East supernova remnant shell (Melia et al. 1998).
The XMM and AXAF observatories with their superior
imaging capabilities will substantially contribute
to solve the problem of the high energy emission from SgrA* and
put more stringent contraints to the accretion models.
\acknowledgments
We thank Silvano Molendi for useful discussions and help with the data analysis.
\clearpage
\begin{deluxetable}{cclcl}
\footnotesize
\tablecaption{BeppoSAX observations summary. \label{tbl-1}}
\tablewidth{0pc}
\tablehead{
\colhead{Obs.} & \colhead{Pointing Direction} & \colhead{Observation Date} & \colhead{MECS Exposure} &
\colhead{Main} \nl
\colhead{n.} & \colhead{R.A. and Dec. (J2000)} &
\colhead{Start \& Stop time (UT)} & \colhead{Time (ks)} & \colhead{target}}
\startdata
1 & 17 43 54.8 --29 44 42 & 1996 Sep 03 10:56 - 03 16:30 & 7 & 1E 1740.7--2942 \nl
2 & 17 47 20.0 --28 24 02 & 1997 Apr 05 08:57 - 06 09:33 & 49 & Sgr B2 \nl
3 & 17 45 40.3 --29 00 23 & 1997 Aug 24 06:01 - 26 08:41 & 100 & Sgr A* \nl
4 & 17 47 20.0 --28 24 02 & 1997 Sep 03 05:18 - 04 09:36 & 51 & Sgr B2 \nl
5 & 17 48 39.2 --28 05 56 & 1997 Sep 04 22:45 - 06 03:22 & 49 & Sgr D \nl
6 & 17 43 54.8 --29 44 42 & 1997 Sep 07 05:45 - 08 06:57 & 43 & 1E 1740.7--2942 \nl
7 & 17 47 25.8 --30 01 01 & 1997 Sep 10 14:07 - 11 18:36 & 21 & SLX 1744--299 \nl
8 & 17 44 41.3 --29 28 14 & 1997 Sep 16 11:30 - 16 20:47 & 15 & Sgr C \nl
9 & 17 44 41.3 --29 28 14 & 1998 Mar 31 06:47 - 31 20:54 & 25 & Sgr C \nl
10 & 17 46 19.9 --28 44 00 & 1998 Apr 13 13:36 - 15 05:51 & 71 & 1E 1743.1--2843 \nl
11 & 17 46 50.2 --28 33 04 & 1998 Aug 26 16:34 - 28 16:30 & 79 & Sgr B1 \nl
\enddata
\end{deluxetable}
\clearpage
\begin{deluxetable}{llllll}
\footnotesize
\tablecaption{X--ray sources in the observed region. \label{tbl-2}}
\tablewidth{0pc}
\tablehead{
\colhead{Source} & \colhead{RA (J2000)} & \colhead{Dec (J2000)} & \colhead{Error} & \colhead{Notes} & \colhead{Ref.\tablenotemark{a} } }
\startdata
XTE~J1748--288 &17 48 08 &-28 28 48 & $1'$ & Transient &1 \nl
SLX~1744--299 &17 47 27 &-29 59 55 & $1'$ & Burster &2 \nl
SLX~1744--300 &17 47 27 &-30 02 15 & $1'$ & Burster &2 \nl
G0.9+0.1 &17 47 21 &-28 09 22 & & SNR &3 \nl
G359.2--0.8 (The Mouse) &17 47 15.2 &-29 58 00 & $8''$ & &4 \nl
SAX~J1747.0--2853 &17 47 0 &-28 52 12 & $1'$ & Transient/Burster &5 \nl
1E~1743.1--2843 &17 46 19.5 &-28 53 43 & $1'$ & &6 \nl
A~1742--294 &17 46 05 &-29 30 54 & $1'$ & Burster &7 \nl
RXJ~1745.7--2858 &17 45 45.5 &-28 58 17 & $20''$& Star? (PT6) & 8 \nl
1E~1742.5--2859 &17 45 40.7 &-29 00 10 & $1'$ & SgrA*? & 6 \nl
RXJ~1745.6--2900 &17 45 40.4 &-29 00 23 & $20''$& SgrA*? (PT7) &8 \nl
AX~J1745.6--2901 &17 45 36 &-29 01 34 & $25''$ & Transient/Burster? & 9 \nl
RXJ~1745.5--2859 &17 45 32 &-28 59 29 & $20''$& PT8 & 8 \nl
A~1742--289 &17 45 37.3 &-29 01 04.8 & $2.8''$& Transient & 10 \nl
KS~1741--293 &17 44 49.2 &-29 21 06.3 & $1'$ & Transient/Burster &11 \nl
1E~1740.7-2942 &17 43 54.8 &-29 44 42.8 & $1''$ & Black hole candidate &12 \nl
\tablenotetext{a}{References are: 1=Strohmayer \& Marshall 1998, 2=Skinner et al. 1990,
3=Helfand \& Becker 1987, 4=Predehl \& Kulkarni 1995, 5=Bazzano et al. 1998, 6=Watson et al. 1981,
7=Pavlinsky, Grebenev \& Sunyaev 1994, 8=Predehl \& Trumper 1994, 9=Maeda et al. 1996, 10=Davies et al. 1976,
11=In't Zand et al. 1990, 12=Mirabel et al. 1992}
\enddata
\end{deluxetable}
\clearpage
\begin{deluxetable}{llclcl}
\footnotesize
\tablewidth{0pc}
\tablecaption{Results of the Spectral Fits (errors are 90\% confidence level). \label{tbl-3}}
\tablehead{
\colhead{Source} & \colhead{Model\tablenotemark{a}} &
\colhead{Column density} & \colhead{Parameter\tablenotemark{b}} &
\colhead{Red.$\chi^2$ (d.o.f.)} & \colhead{Flux\tablenotemark{c}} \nl
\colhead{(data) } & \colhead{ } & \colhead{($10^{22}$ cm$^{-2}$)} &
\colhead{ } & \colhead{ } &
\colhead{ }}
\startdata
Galactic center &PL+line(6.7) & $8.3\pm{0.5}$ & $\Gamma=2.6\pm{0.1}$ & $1.12$ (284) & $4.0\pm{0.2}$ \nl
(MECS) & & & $EW(6.7)=1.2 $ & & \nl
&Brems+line(6.7) & $6.6\pm{0.3}$ & $T = 5.1\pm{0.5}$ & $1.20$ (284) & $3.3\pm{0.2}$ \nl
& & & $EW(6.7)=1.2 $ & & \nl
&M & $7.0\pm{0.3}$ & $T = 4.1\pm{0.3}$ & $1.16$ (288) & $3.5\pm{0.2}$ \nl
&M+M & $11.3\pm{1.2}$ & $T_{M1} = 0.80^{+0.18}_{-0.08} $& $1.00$ (285)& $6.0^{+0.6}_{-1.0} $\nl
& & & $T_{M2} =4.9\pm{0.4}$& & \nl
&M+PL+line(6.7) & $10\pm{1}$ & $T_{M}=1.3^{+0.2}_{-0.6}$ &$0.97$ (281)& $5.5\pm{1}$ \nl
& & & $\Gamma=1.7\pm{1}$& & \nl
& & & $EW(6.7)=0.7-1 $ & & \nl
\hline
1E~1740.7--2942 &PL & $14.7\pm{0.4}$ & $\Gamma=1.52\pm{0.04}$& $1.10$ (180) & $49.5\pm{1.0}$ \nl
(MECS) &Brems & $14.1\pm{0.3}$ & $T=34.5\pm{6.0}$ & $1.12$ (180) & $47.4\pm{0.4}$ \nl
\cline{2-6}
(LECS+MECS &PL & $14.5\pm{0.3} $ & $\Gamma=1.45\pm{0.3}$ & $2.84$ (283) & $49\pm{1}$ \nl
+PDS) &Brems & $12.9\pm{0.2}$ & $T=174\pm{6}$ & $1.80$ (283) & $45\pm{0.5}$ \nl
&CompST & $14.6\pm{0.2}$ & $T=24\pm{1}$ & $1.30$ (282) & $49\pm{0.5}$ \nl
& & & $\tau=5.5\pm{0.1}$ & & \nl
&PL+E$_{c}$ & $13.6\pm{0.2}$ & $\Gamma=1.36\pm{0.02}$ & $1.24$ (281) & $46.7\pm{0.5}$ \nl
& & & E$_{c}=52\pm{4}$ & & \nl
& & & E$_{fold}=105\pm{10}$ & & \nl
\hline
A~1742--294 &PL & $6.5\pm{0.5}$ & $\Gamma=1.72\pm{0.11}$& $0.995$ (238) & $30\pm{1}$ \nl
(MECS-Obs.9) &Brems & $5.9\pm{0.4}$ & $T=16.3^{+5.1}_{-3.6}$& $0.986$ (238) & $29\pm{1}$ \nl
\cline{2-6}
&PL & $6.9\pm{0.3}$ & $\Gamma=1.97\pm{0.05}$& $0.99$ (323) & $63\pm{1}$ \nl
(MECS-Obs.10) &Brems & $6.0\pm{0.2}$ & $T=10.3\pm{1.0}$ & $1.01$ (323) & $58\pm{1}$ \nl
\hline
SLX~1744--299 &PL & $5.1\pm{0.2}$ & $\Gamma=2.1\pm{0.1}$ & $0.92$ (364) & $20\pm{0.4}$ \nl
(MECS) &Brems & $4.2\pm{0.2}$ & $T=8.9\pm{0.6}$ & $0.99$ (364) & $18\pm{0.4}$ \nl
\hline
SLX~1744--300 &PL & $5.3\pm{0.2}$ & $\Gamma=2.2\pm{0.1}$ & $1.06$ (342) & $12\pm{0.3}$ \nl
(MECS) &Brems & $4.2\pm{0.2}$ & $T=7.7\pm{0.5}$ & $1.09$ (342) & $11\pm{0.3}$ \nl
\hline
XTE~J1748--288 &PL & $7.3\pm{0.5}$ & $\Gamma=1.46\pm{0.1}$ & $0.922$ (163) & $11.2\pm{0.3}$ \nl
(MECS) &Brems & $6.9\pm{0.4}$ & $T=40^{+24}_{-11} $ & $0.93$ (163) & $10.9\pm{0.3}$ \nl
\hline
SAX~J1747.0--2853&Brems & $8.3^{+0.6}_{-0.3}$ & $T=6.1^{+0.9}_{-0.7}$ & $1.01$ (138) & $4.0^{+0.2}_{-0.3}$ \nl
(MECS) & & & & & \nl
\hline
KS~1741--293 &PL & $20\pm{2}$ & $\Gamma=2\pm{0.2}$ & $0.992$ (213) & $14.5^{+2.0}_{-0.6}$ \nl
(MECS) &Brems & $18\pm{1}$ & $T = 11\pm{3}$ & $0.917$ (213) & $12.5^{+1.2}_{-0.5}$ \nl
&BB & $12\pm{2}$ & $T = 1.8\pm{3}$ & $0.966$ (213) & $8.5\pm{0.4}$ \nl
\hline
1E~1743.1--2843 &BB & $13\pm{1}$ & $T = 1.8\pm{1}$& $1.01$ (178)& $16.5\pm{0.5}$ \nl
(MECS) & & & & & \nl
\hline
G0.9+0.1 &PL & $25^{+17}_{-10}$ & $\Gamma=2.5^{+1.5}_{-0.8}$& $0.72$ (28) & $2.0^{+4.0}_{-1.0}$ \nl
(MECS) &Brems & $23^{+12}_{-8}$ & $T=6^{+13}_{-3} $ & $0.72$ (28) & $1.5^{+0.2}_{-0.5}$ \nl
&BB & $16^{+10}_{-10}$ & $T=1.6^{+0.6}_{-0.5} $ & $0.72$ (28) & $0.9^{+0.2}_{-0.4}$ \nl
\hline
G359.23-0.92 &PL & $4-6$ & $\Gamma=1.9-2.3$& & $\sim3$ \nl
(MECS) & & & & & \nl
\tablenotetext{a}{PL=Power Law,PL+E$_{c}$= Power law plus high energy cutoff (E$_{c}$=cutoff energy in keV;
E$_{fold}$=e-folding energy in keV), BB=Blackbody,
Brems=Thermal Bremsstrahlung, CompST=Comptonization model
(Sunyaev \& Titarchuk 1980), line=gaussian line, M=MEKAL model in XSPEC.} \nl
\tablenotetext{b}{All the temperatures and energies are in keV, $\Gamma$=photon index.} \nl
\tablenotetext{c}{Unabsorbed fluxes (2--10 keV) are in units of $10^{-11}$ ergs cm$^{-2}$ s$^{-1}$}
\enddata
\end{deluxetable}
|
\section{OVERALL PICTURE}
In the Newtonian approximation,
half of the gravitational energy
is released in an extended disk during accretion onto a neutron star
with a weak magnetic field.
If the star rotates slowly,
the other half of the energy
must be released in the immediate vicinity of the star
as the matter transfers from the disk onto the surface.
This energy release is generally thought to occur
in a narrow
(both in radius and in latitude)
boundary layer of the disk
which is formed near the contact boundary
between the disk and the star.
The boundary layer is significant
not only due to contribution to the luminosity of the compact object.
It is important since the boundary-layer area is small
compared to that of the disk,
the layer's emission is harder
and must be observed in a different spectral range
relative to the disk emission.
Accretion onto a neutron star differs in many ways
from accretion onto a white dwarf.
First, the percentage of energy
which is released in the boundary layer
can exceed appreciably $50\%$
because of the general-relativity effects (Sunyaev and Shakura 1986).
The boundary-layer luminosity $L_{SL}$
is greater than the disk luminosity $L_d$
even for rapidly rotating neutron stars,
but the ratio $L_{SL}/L_d$ decreases
with the increasing angular velocity of the star (Sibgatullin and Sunyaev 1998).
Second, at luminosities
$L > L_{edd}/100 \sim 10^{36}$ erg s$^{-1},$
the local flux of radiative energy
which is emitted from the boundary layer
is of the order of the Eddington flux,
i.e., the force of radiation pressure on an electron
is of the order of the gravitational force acted on a proton.
There are also other dissimilarities.
Most of the authors
who considered disk accretion onto stars
(Shakura and Sunyaev 1973;
Lynden-Bell and Pringle 1974;
Pringle and Savonije 1979;
Papaloizou and Stanley 1986;
Popham et al. 1993),
white dwarfs
(Tylenda 1981;
Meyer and Meyer-Hofmeister 1989;
Popham and Narayan 1995),
and neutron stars
(Shakura and Sunyaev 1988; Bisnovatyi-Kogan 1994)
suggest that the accreting-plasma velocity
decreases from the Keplerian velocity
to the stellar rotation velocity
through the turbulent friction
between the differentially rotating layers
of accreting matter
within a thin boundary layer of the disk (see Fig. 1).
In this case, $d\omega/dr > 0$ in the boundary layer,
the boundary-layer meridional extent is
$R\theta_{bl}= R \sqrt{2 T_{bl}/m_p}/v_{\varphi}^k,$
and its height is
$h_{bl}\simeq R (2T_{bl}/m_p)/(v_{\varphi}^k)^2,$
where $R$ is the stellar radius,
$T_{bl}$ is the boundary-layer temperature
given in energy units,
$m_p$ is the proton mass,
and $v_{\varphi}^k$ is the Keplerian velocity at the radius $R$
(Shakura and Sunyaev 1988).
The meaning of the geometrical quantities $h_{bl}$ and $\theta_{bl}$
is clear from Fig. 1.
We see that in this approach,
$h_{bl}$ is determined by the scale height
in a gravity field
with a free fall acceleration
$g_0 = \gamma M/R^2 = $ $(v_{\varphi}^k)^2/R.$
The boundary-layer meridional extent, $R\theta_{bl},$
is given by the same formula that is used to calculate the disk thickness,
i.e., only the tangential component of the gravity force
is taken into account.
The thickness, $h_{bl},$ and the meridional extent turn out to be independent
of latitude and accretion rate, respectively.
\begin{figure}
\epsfxsize=17cm
\epsffile{r1.ps}
\caption{(a) Shape of the boundary layer in the standard approach.
The notation is as follows:
$e$ is the equatorial plane,
$D$ is the disk,
and $S$ is the stellar surface.
(b) Angular velocity $\omega = v_{\varphi}/r$ versus radius,
where $v_{\varphi}$ is the rotation velocity;
1 -- the Keplerian dependence:
$\omega\propto r^{ - 3/2};$
the angular velocity $\omega_n$ in the neck
is at a maximum
and approximately equal to the Keplerian velocity,
$f_n =$
$\omega_n/2\pi \approx$
$ v_{\varphi}^k / 2 \pi R =$
$1.84 (M/M_{\odot})^{1/2}/R_6^{3/2}$ kHz,
$R_6 = R/(10$ km);
inside the star from the center up to its surface $r \leq R,$
the rotation is rigid: $\omega \equiv \omega_S.$}
\end{figure}
\begin{figure}[tbh]
\epsfxsize=10cm
\epsfbox[-300 87 400 753]{r2.ps}
\caption{Rotation of the matter in the disk
and on the stellar surface
in the proposed model;
$S$ is the stellar surface,
$p$ is the pole,
and $e$ is the equator.}
\end{figure}
In this paper,
we take a completely different approach to the problem.
The deceleration of rotation is considered simultaneously
with the spread of accreted matter (see Figs. 2-4).
Discarding popular beliefs,
we assume that the deceleration of rotation
due to the gripping on
(hooking, catching on)
the slowly moving dense matter
beneath the underlying surface, $S,$
is the principal friction mechanism.
This gripping is due to the turbulent viscosity.
This problem has much in common with the well-studied
(both theoretically and experimentally)
problem of the deceleration of a subsonic or supersonic flow
above a solid surface
{\it (see Subsec. 2.5 for a detailed discussion
of the turbulence description
accepted in this paper).}
\begin{figure}[tbh]
\epsfxsize=16cm
\epsffile{r3.ps}
\caption{Proposed model:
(a) the star is the sphere $S,$
the one-dimensional spread layer is $1D_{bl},$
the two-dimensional transition between the disk and the spread layer is $2D,$
and the one-dimensional disk is $1D_d;$
(b) the linear rotation velocity $v_{\varphi}$ versus latitude.
The rotation is "localized" in the latitude belt
$0 < \theta < \theta_{\star}$ --
for a nonrotating star,
the matter outside this belt is essentially nonrotating;
(c) local radiative flux $q$ versus latitude.
The arrows $a$ and $b$ "fence" the ring belt of enhanced brightness.
In Fig. 3a, this belt lies between the latitudes $a$ and $b.$}
\end{figure}
\begin{figure}[tbh]
\epsfxsize=10cm
\epsfbox[-300 197 400 650]{r4.ps}
\caption{Spread of the rotating plasma from the disk, $D,$
over the neutron-star surface, $S.$
Here, $I$ is the intermediate zone near the disk neck,
$0<\theta<\theta^{\star}$ is the hot belt,
and
$\theta > \theta^{\star}$
is the cold part of the spread layer.
The rotation velocity $v_{\varphi}$ (filled circle) is directed along the normal to the plane
of the figure.
The slowly circulating (in $\varphi$ and $\theta)$
dense underlying layers of matter beneath the spread layer
are indicated by the dashes.}
\end{figure}
Here, we ignore the magnetic field, $H,$ of the neutron star.
As we show in Subsec. 4.8,
this assumption is valid for
$H < 2.4\times 10^9 (L_{SL}/L_{edd})^{0.57}$ gauss,
where $L_{SL}$ and $L_{edd}$ are
the spread-layer luminosity and the Eddington critical luminosity,
respectively.
In addition, when deriving the equations of motion of the accreting flow
over the stellar surface,
we disregard its rotation and general-relativity effects.
An appreciable fraction of the X-ray bursters are neutron stars
with a rotation frequency
$f_{ns} = \omega_{ns}/2\pi$
of the order of 300 Hz
see, e.g., a review by Van der Klis 1998).
At the same time, the Keplerian rotation frequency
$f_k = \omega_k/2\pi$
at the disk-star boundary in the problem under consideration
is about 2000 Hz.
Thus,
the effects of stellar rotation,
which are of the order of the ratio $(f_{ns}/f_k)^2,$
are small,
and the computations
which are performed below
also yield an acceptable first approximation
to the problem of accretion onto a rotating neutron star.
\subsection{Comparison of the Standard and Proposed Approaches}
The most important points (I-IV)
that pertain to our comparison of the approaches
are listed below
(I) Let's consider the relationship between the radial
(inflow to the star, $v_r < 0)$
and meridional $(v_{\theta})$
velocities.
The meridional velocity is commonly ignored $(|v_r| \gg v_{\theta}).$
The one-dimensional $(1D)$ approach to the accreting disk
also extends to the boundary layer.
In any case, $v_{\theta}$ is generally omitted from the equations.
The boundary layer turns out to be an extension of the disk
in the sense that the $1D$ description of the disk $(1D_d)$
also applies to the boundary layer.
The disk thickness $h$ passes through a minimum near the neck (see Fig. 1a).
After the neck, the flow broadens
as it approaches the stellar surface
and 'digs itself' into the star.
Meanwhile, a flow with an approximately $1D$ disk,
with a $2D$ region of a sharp turn of the flow near the neck,
and with an approximately $1D$ (hydraulic) spread of the boundary layer
over the neutron-star surface
seems more natural (see Figs. 1, 3, 4 and Subsec. 2.1).
In this case, $v_{\theta} \gg |v_r|$ in the boundary layer.
We specify inflow conditions
at the interface between the boundary layer and the $2D$ region
when calculating the former,
because we do not consider here an exact $2D$ problem
of the flow in the intermediate zone.
In this formulation,
it would be more correct to talk not about the {\it boundary layer of the disk}
but about the {\it spread layer.}
The dynamics of the latter bears no relation to the disk.
Here, the problem of the deceleration and spread
of the rapidly rotating matter
arises.
(II) We consider the deceleration of rotation
and the structure of isolines of rotation velocity $v_{\varphi}.$
In the standard and proposed approaches,
we have
$v_{\varphi}=v_{\varphi}(r)$
and $v_{\varphi}=v_{\varphi}(\theta)$ in the boundary layer, respectively.
The lines of constant $v_{\varphi}$ (isolines)
are perpendicular to the equatorial plane in the standard model
and perpendicular to the surface $S$ in the proposed model.
In both cases, the rotation is Keplerian in the disk at $r > r_n,$
where $r_n$ is the neck radius.
The $v_{\varphi}$ isolines are indicated in Fig. 3a by the dots.
They are roughly perpendicular to the surface $S$ inside the boundary layer.
The isolines make a sharp turn near $S$
and then run in a dense bundle beside $S$ for a star at rest.
This implies that the rotation velocity
$v_{\varphi} = \omega r$ abruptly decreases
near the interface between the spread layer and the underlying surface.
The deceleration near the surface $S$ is discussed in Subsec. 2.5.
(III) In the standard models,
the boundary layer and main energy release
concentrate near the equator (see Fig. 1).
In the proposed model,
the width $\theta_{\star}$ of the rotating belt is variable.
It increases with increasing accretion rate, $\dot M.$
As $\dot M \to \dot M_{pole},$
the rotation extends to the entire stellar surface:
$\theta_{\star}\to 90^0.$
Calculations show that the spread-layer luminosity
which corresponds to this accretion rate
is
$L_{pole} \approx 0.9 L_{edd,}$
where
$$
L_{edd}= 4\pi R^2 q_0 = 1.26 \times 10^{38} \frac{M}{M_{\odot}},
\; \mbox{erg/s},
$$
$$
q_0=
\frac{m_p \, g_0 \, c}{\sigma_T}=
\Sigma_T \, g_0 \, c =
1.00\times 10^{25}\frac{(M/M_{\odot})}{R_6^2},
\; \mbox{erg/s cm}^2,
\eqno (1.1)
$$
where $g_0 =\gamma M/R^2 =(v^k_{\varphi})^2/R,$
$v^k_{\varphi}$ is the Keplerian velocity at the stellar surface,
$\Sigma_T =m_p/\sigma_T = 2.5$ g/cm$^2$
defines a column of matter with a unit optical depth for Thomson scattering,
$m_p$ is the proton mass (the plasma is assumed to be hydrogenic),
$\sigma_T = (8\pi/3)(e^2/m_e c^2)^2$
is the Thomson scattering cross section,
and $R_6 = R/(10$ km).
Below, we talk only about the spread-layer luminosity
at the stellar surface normalized to the critical Eddington luminosity.
Clearly, the accretion disk makes a comparable contribution
to the system's total luminosity.
(IV) We consider the latitude dependence of the intensity of radiation $q$
from the belt.
In the standart approach,
the function $q(\theta)$ has a maximum on the equator.
In the approach under discussion,
$q(\theta)$ turns out to have a minimum at $\theta = 0^0.$
Moreover, $q(\theta = 0) \approx 0.$
The distribution of $q$ (see Fig. 3c) resembles the letter M,
if the plot is complemented with the half
which is symmetric around the equator.
\subsection{Characteristic Features of the Model}
The problem of the spread of rapidly rotating accreting matter
over a neutron-star surface
has several important features.
(I) The centrifugal force
which acts on the accreting matter in the vicinity of the equator
essentially offsets the gravitation,
because the accreting matter in this vicinity
has a velocity that is approximately equal to the Keplerian velocity.
As one recedes toward the pole,
the centrifugal force decreases due to the deceleration of rotation.
(II) In the radiating belt,
the radiation pressure works against the difference
between the gravitational component
and the normal component of the centrifugal force.
Because of the rapid rotation,
the local Eddington luminosity
at which the radiation-pressure force equals the pressing force
is appreciably lower than the luminosity $q_0$
calculated from the standard formula (1.1).
(III) The dissipation of rotational kinetic energy
causes a strong energy release near the lower boundary of the spread layer.
As a result, powerful radiation,
which diffuses through the spreading plasma
and forms the observed spectrum, is produced in this sublayer.
The normal component of the gradient in the radiation pressure
(or the normal component of the radiation-pressure force)
is opposite the direction of the gravitational force.
Its direction coincides with the direction
of the normal component of the centrifugal force.
Within the layer,
the normal component of the radiation-pressure gradient
(together with the centrifugal force)
decreases the free fall acceleration,
decreases the density of the matter in the layer,
and increases its thickness.
An important point is that the increase in the height of the column
(in the spread-layer thickness)
and the drop in the plasma density
cause the efficiency of the turbulent friction
of the layer against the stellar surface
to decrease.
As a result, the deceleration of Keplerian rotation
occurs on larger meridional scales
than that for a small radiation-pressure force.
(IV) The numerical solution of the equations of radiation hydrodynamics,
which is presented below,
gives the following picture.
The local radiative flux, $q,$ at the equator
is much lower than the Eddington critical value (1.1).
This is attributable to the action of the centrifugal force
and to the fact that the rotation velocity $v_{\varphi}$
near the equator is close to the Keplerian velocity.
A ring zone
in which the energy release increases
due to the decrease in the centrifugal force
is formed at a distance from the equator along the meridian.
Up to $70\%$ of the layer luminosity is released
in two such latitude rings (see Figs. 3a and 3c).
In this case, the local flux $q$
never reaches the Eddington limit (1.1),
because the finite rotation velocities $v_{\varphi}$
are required for the energy release
through friction against the underlying layers,
while the allowance for the centrifugal force turns out to be important.
The flux $q$ at maximum depends on accretion rate
and accounts for $70-97\%$ of the limit (1.1).
(V) The spreading plasma within the bright belt is radiation-dominated
$(p_r \gg p_{pl}).$
The high saturation of the radiating belt with radiation
causes the speed of sound, $c_s,$ in the layer,
which proves to be much larger
than the speed of sound in plasma without radiation,
to increase significantly.
As a result,
the flow rotation velocity, $v_{\varphi},$ in the layer
is moderately supersonic:
$\mathop{\rm Ma}\nolimits_{\varphi} = v_{\varphi}/c_s \sim 5.$
The meridional spread is subsonic,
which is important in choosing a friction model.
If the Mach numbers, $\mathop{\rm Ma}\nolimits_{\varphi},$ were calculated
from the plasma speed of sound alone,
then the flux would be hypersonic with $\mathop{\rm Ma}\nolimits_{\varphi} \sim 100;$
moreover, the flux in $v_{\theta}$ would be transonic.
(VI) At $\theta > \theta_{\star},$
the velocity, $v_{\varphi},$ drops virtually to zero,
and the layer contracts to a thin, cold weakly radiating film
('dark' layer),
which slowly moves toward the poles.
Having lost the excess (relative to the star) angular velocity
in the radiating belt,
the matter becomes cold and dense.
The intensity of the radiation from its surface
decreases by several orders of magnitude.
It moves relative to the star with a velocity that is considerably lower
than the speed of sound.
Of course, this matter is not accumulated near the poles.
There is a slow flow
in which it spreads under the hot layers over the entire stellar surface.
This matter forms the surface
against which the radiation-dominated layer rubs.
(VII) If the local radiative flux were the limiting Eddington flux (1.1),
the width of the radiating belt
(let us denote this width by $\theta^{\star}_0$
to distinguish it from $\theta_{\star})$
could be estimated from the energy balance
$$
(1/2)\,\dot M \,( v_{\varphi}^k )^2 =
q_0 \, 4\pi R^2 \,\sin \theta^{\star}_0,\;\;\;\;\;
\theta_0^{\star} = \arcsin ( L_{SL}/L_{edd} ), \eqno (1.1)'
$$
where $L_{SL}$ is the spread-layer luminosity.
As will be seen from the analysis in the main body of this study,
$(1.1)'$ underestimates appreciably $\theta_{\star}.$
This is attributable to the action
of the radial component of the centrifugal acceleration.
As follows from the calculations which are given below,
the approximate balance is
$$
\dot M\,\frac{(v_{\varphi}^k)^2}{4} =
2 \pi R^2 \, \int_0^{ \theta_{\star} }
\, q_{eff} (\theta) \; d\sin \theta, \eqno (1.2)
$$
$$
q_{eff} (\theta) = \Sigma_T \, g_{eff}(\theta) \, c, \;\;\;
g_{eff}(\theta) = g_0 \, G_{eff} (\theta),\;\;\;
G_{eff} = 1 - U^2 - W^2 \approx 1-W^2, \eqno (1.2)'
$$
where $U = v_{\theta}/v_{\varphi}^k,$
$W = v_{\varphi}/v_{\varphi}^k,$
$v_{\varphi}^k$ is the Keplerian velocity.
In $(1.2)',$ we assumed
that the centrifugal force
is almost completely determined by the rotation velocity, $v_{\varphi},$
because the centrifugal force related to the spread velocity is small:
$v_{\theta}^2/R \ll g_0.$
The latitude dependence of rotation velocity
$v_{\varphi}( \theta )$ is required to calculate $\theta_{\star}$ using (1.2).
The approximate validity of relations (1.2) and $(1.2)'$
is attributable to the fact
that the Eddington radiative flux $q_r$ in the radiating belt
offsets the difference $g_{eff}$
between the weight and the radial component of the centrifugal force
with a high precision [see below Subsec. 1.3 (I)].
The appearance of rings of enhanced luminosity follows precisely from this.
Indeed, it follows from $(1.2)'$
that $q( \theta )$ increases with $\theta,$
because $v_{\varphi}( \theta )$ decreases with latitude.
(VIII) We carried out this study
in an effort to explain the observed spectra of bursters
in a wide range of their luminosities.
Several variable sources that are X-ray bursters exhibit variations
in the X-ray luminosity by hundreds of times
(see Campana et al. 1998a, 1998b, Gilfanov et al. 1998).
The theory,
which we construct here,
claims to describe some of the spectral features in the light curve
of a neutron star as the luminosity changes by two orders of magnitude.
(IX) The atmosphere under consideration
is characterized by the delicate balance of three forces:
gravitation, centrifugal force, and radiation-pressure force.
Under these conditions
(just as in the atmospheres of massive hot stars),
a strong wind
which flows away from the zone of enhanced brightness
is inevitably formed.
As a first approximation,
we assume below
that this wind affects only slightly the transport characteristics
of the spread layer and its optical depth.
(X) It follows from the qualitative estimates
and the detailed calculations
which are presented below
(see Subsecs. 4.3 and 4.5)
that the derived solutions in the radiating belt
lie near the limit of validity of the shallow-water approximation.
The ratios of the effective layer thickness,
$h_{eff} = [\,|\, d \ln \rho/dr \,|\, ]^{-1},$
in this region
(the derivative is calculated near the base of the spread layer)
to the stellar radius, $R,$ are typically $\sim 0.1.$
\begin{figure}[tbh]
\epsfxsize=10cm
\epsffile{r5.ps}
\caption{Profiles 1, 2, 3 and 4 of the spread-layer (S.L.) effective thickness
$h_{eff}(\theta)$
for various accretion rates $\dot M.$
The relative luminosities, $L_{SL}/L_{edd},$
that correspond to labels 1, 2, 3 and 4 are 0.01, 0.04, 0.2 and 0.8, respectively.
The stellar mass and radius are
$1.4 M_{\odot}$
and 12 km.
(a) Linear scales, the shape of the radiating belt is seen;
(b) logarithmic scales, the dark layer and the spread layer
are seen better
than in the upper panel
(especially at small $\dot M;$
the scale $R \theta$ covers the entire distance from the equator to the pole).}
\end{figure}
Nevertheless, the hydraulic approximation
(Schlichting 1965; Chugaev 1982)
describes satisfactorily the experimental situation
at the values of $h_{eff}/R.$
The geometrical characteristics of the spread layer are shown in Fig. 5.
The spread layer is divided into two distinctly different parts
at $\theta_{\star}:$
the radiating belt (r.b.) and the dark layer (d.l.).
Their thicknesses, $h_{rb}$ and $h_{dl},$ differ markedly.
The thickness $h_{dl}$
is of the order of the scale height
in a gravity field
with $g_0=\gamma M/R^2.$
The slope $\tan \beta$ is $dh_{eff}/R d\theta.$
This slope changes only slightly inside the radiating belt (see Fig. 5a).
The angle $\beta$ is very small in the dark layer.
The angle $\beta$ in the radiating belt
increases as the accretion rate, $\dot M,$ decreases.
Therefore flows with low $\dot M$ are described worse
by the $1D$ approximation.
In Figs. 5a and 5b, profiles 1-4 refer to the different values of $\dot M.$
Profiles 1-4 from Fig. 5a are repeated in Fig. 5b on a logarithmic scale.
This is necessary
because the dark-layer thickness, $h_{dl},$
is so small
that it cannot be compared with the radiating-belt thickness $h_{rb}$
on a linear scale.
In addition, the flow separation into two characteristic regions
is more clearly seen in Fig. 5b.
The latitude $\theta_{\star}$
corresponds to the thin transition zone of a rapid change in thickness.
In turn, the linear coordinates in Fig. 5a are required
to show the shape of the radiating belt.
(XI) The flows in the disk and in the spread layer are closely related.
The accreting flow is furnished through the disk,
while irradiation of the disk from the stellar surface
affects its characteristics.
This issue is analyzed in Subsec. 2.1,
where the disk thickness in the zone of its contact
with the spread layer is estimated.
(XII) In a radiation-dominated layer,
the hydrodynamic meridional transfer of radiative enthalpy toward the poles
plays a major role.
This enthalpy per unit mass
is of the order of the gravitational energy $\gamma M/R.$
\subsection{Main Results}
We derive the system of equations of radiation hydrodynamics
that describes the deceleration of rotation of the accreting matter
and its meridional spread.
The derivation is based on the assumption
that Thomson scattering contributes mainly to the opacity
and that the energy release concentrates in a relatively thin sublayer
at the base of the spread flow.
The derived equations {\it differ markedly}
from the standard shallow-water equations.
A qualitative analysis of the derived system of equations
and their solutions (Sec. 3) allows us,
first, to develop a procedure for the effective numerical integration
of the equations
and,
second,
to make a clear interpretation of the calculations.
We carried out a numerical analysis of the system.
The problems of the numerical analysis,
without which an independent verification of the data would be impossible,
are outlined in Secs. 3 and 4.
It is important to set out the procedure for the solution,
because if this procedure is clear
the results can be reproduced.
We thus would like to draw the reader's attention
to the following subsections:
(I) smallness of the coefficients,
rigidity of the system,
steep and gentle segments (Subsec. 3.5);
(II) multiplicity of the solutions,
selection criteria (Subsecs. 3.6 and 4.1-4.4);
(III) circulation of the cooled accreting matter
which lost the rotational velocity component,
its spread and settling (Subsecs. 4.2-4.4);
and (IV) modification of the solutions
as the friction coefficient and the accretion rate are varied
(Subsecs. 4.5 and 4.7).
Below, we dwell on the most important results.
(I) {\bf Rings of enhanced brightness.}
Typical latitude profiles of the flux $q$ are shown in Figs. 6-8.
The radiating belt broadens as the accretion rate increases.
The latitude $\theta_{max}$
at which the bright rings lie
increases with increasing $\dot M$ and $L_{SL}/L_{edd}.$
The bright rings are not narrow;
their width is $\Delta \theta \simeq \theta_{\star}.$
At a high accretion rate $\dot M \simeq \dot M_{pole}$
[see Subsec. 1.1 (III)],
the entire stellar surface,
except the small equatorial and polar zones,
emits almost uniformly.
In this case, the bright ring spreads out.
Profile 4 in Fig. 7 refers to this situation.
\begin{figure}[h]
\epsfxsize=7.cm
\epsfbox[-250 180 430 630]{r6.ps}
\caption{Local radiative flux $q(\theta)$ (solid lines)
and effective free fall acceleration (including the centrifugal acceleration)
$g_{eff} (\theta)$
(dashed lines) versus latitude.
An important point is that
$q/q_0\approx g_{eff}/g_0$
in the radiating belt [see (1.2) and (1.2)'].
Labels 1 and 3 refer to
$L_{SL}/L_{edd} = 0.01$ and $0.2,$
respectively.}
\end{figure}
\begin{figure}[h]
\epsfxsize=7.5cm
\epsfbox[-250 180 430 590]{r7.ps}
\caption{Changes in the profiles of the local radiative flux $q$
(solid lines)
with $\dot M.$
The radiating belt broadens with increasing $\dot M.$
The boundary
$\theta_{\star}$
lies at the right edge of the belt in the zone of an abrupt decrease
in $q(\theta).$
The $g_{eff}/g_0$ profile at $L_{SL}/L_{edd}=0.8$
is indicated by the dashed line.
The scale $R \theta$ terminates at the pole.
Labels 2, 3, and 4 refer to
$L_{SL}/L_{edd} = 0.04, 0.2,$ and 0.8, respectively.}
\end{figure}
\begin{figure}[tbh]
\epsfxsize=7.2cm
\epsfbox[-250 180 430 665]{r8.ps}
\caption{$q( \theta )$ profiles in a wide range of $\dot M.$
Labels 1, 2, 3, and 4
refer to $L_{SL}/L_{edd} = 0.01, 0.04, 0.2,$ and $0.8,$
respectively.}
\end{figure}
The function $q(\theta),$
which gives the radiative flux,
has the maximum at $\theta_{max}$
with which the ring of enhanced brightness is associated.
This ring is marked by arrows $a$ and $b$ in Fig. 3c.
The ring bounded by latitudes $a$ and $b$
which correspond to these arrows
is shown in Fig. 3a.
To the right of $\theta_{max},$
the rotation velocity $v_{\varphi}$ vanishes (see Figs. 3b and 3c).
For this reason,
$q( \theta )$ rapidly decreases to the right of $\theta_{max}.$
The increase in $q( \theta )$ to the left of $\theta_{max}$
(see Fig. 3c)
with increasing $\theta$ is caused by the decrease in $v_{\varphi}$
(see Fig. 3b).
Indeed, the decrease in $v^2_{\varphi}$
leads to an increase in the acceleration
$g_{eff} \approx g_0 \, (1-W^2),$
where $W=v_{\varphi}/v_{\varphi}^k,$
and, consequently, to an increase in $q,$
$\; q \propto g_{eff}$ [see formulas (1.2) and $(1.2)'].$
\begin{figure}[tbh]
\epsfxsize=8cm
\epsfbox[-250 180 360 650]{r9.ps}
\caption{Profiles of the meridional velocity $v_{\theta}$
and of the spread-layer surface density $\Sigma_S$ for
$L_{SL}/L_{edd}=0.2.$
At $\theta_{\star},$
the flow abruptly slows down,
and its density increases.
In the upper corner, the Keplerian velocity
$v_{\varphi}^k$ is shown for
$M = 1.4 M_{\odot}$
and $R = 12$ km.}
\end{figure}
(II) {\bf Surface density, optical depth, and spectral hardness.}
In the steady-state case
in the absence of sources and sinks of matter in the spread layer,
the flux of mass is conserved
when spread along the meridian.
We ignore the settling of matter to the bottom within this layer
(Subsecs. 4.3 and 4.4).
The meridional velocity is thus $v_{\theta} \propto 1/\Sigma_S,$
where $\Sigma_S = \int \rho dr$ is the surface density,
the integral is taken over the spread-layer thickness.
Accordingly, $\Sigma_S$ reaches a minimum at the maximum of $v_{\theta}.$
These profiles are illustrated in Fig. 9.
The spread layer consists of the radiating belt and the dark layer
which border at $\theta_{\star}.$
The maximum of $v_{\theta}$ and the minimum of $\Sigma_S$
lie near the edge of the radiating belt.
In the dark layer, $\Sigma_S$ abruptly increases,
while $v_{\theta}$ abruptly decreases.
This is clearly seen in Figs. 9 and 10.
The dark layer terminates at the sonic point.
This important problem is analyzed in Subsecs. 4.1-4.4.
Figure 10 shows
how the spread-layer surface density $\Sigma_S$ depends on accretion rate.
The spread-layer optical depth $\tau_T = \Sigma_S/\Sigma_T$
for Thomson scattering
rapidly decreases with decreasing $\dot M.$
At $L_{SL}/L_{edd}\sim 10^{-2},$ we have $\tau_T \sim 3.$
No blackbody radiation can be produced
in the medium at such a small optical depth;
the protons and electrons have temperatures much higher than 1 keV.
The low-frequency photons emitted by the dense underlying layers
are comptonized by hot electrons of the spread layer
and form the experimentally observed hard power-law tails in the spectra
(Barret et al.1992, Campana et al. 1998a, 1998b, Gilfanov et al. 1998).
In turn, some of the hard photons penetrate deep into the underlying layer
and heat it up
[see Sunyaev and Titarchuk (1980) for a discussion of this problem].
\begin{figure}[tbh]
\epsfxsize=8cm
\epsfbox[-220 180 400 630]{r10.ps}
\caption{Shape of the $\Sigma_S( \theta )$ profile
at different spread-layer luminosities.
At a fixed $\dot M,$
the surface density rises with increasing $\theta$
at $\theta_{\star},$
after which an optically thick weakly radiating layer begins.
Labels 1, 2, 3, and 4
refer to $L_{SL}/L_{edd} = 0.01, 0.04, 0.2,$ and $0.8,$ respectively.}
\end{figure}
At a high spread-layer luminosity $L_{SL} > 0.1 L_{edd},$
we have a different picture.
In this case, the spectrum is much softer and exhibits no pronounced hard tail.
The optical depth for Thomson scattering is great $(\tau_T \sim 1000).$
The layer optical depth for effective absorption
$\sqrt{\tau_T \; \tau_{ff} (\nu, T_e)}$
(Shakura 1972; Felten and Rees 1972; Shakura and Sunyaev 1973;
Illarionov and Sunyaev 1972)
turns out to be also large.
Deep in the layer,
the most important physical process is saturated comptonization.
The parameter $y = (T_e/m_e c^2) \, \tau_T^2,$
which characterizes comptonization in a layer with an optical depth $\tau_T,$
is large even at $\tau_T > \sqrt{m_e c^2/T_e} \sim 20.$
At $T_e \sim T \sim \tau_T^{1/4}$ (see Subsec. 2.2)
and $\tau_T \sim 10^3,$
we have $y \sim (T_S/m_e \, c^2)\, \tau_T^{9/4} \sim 10^5$ deep in the layer.
The dimensionless frequency,
near which the rate of photon absorption due to the free-free processes
equals the rate of photon removal upward along the frequency axis
via comptonization,
is
$x_0 = h \nu_0/T_e =3 \times 10^5\, n_e^{1/2}/T_e^{9/4}.$
This frequency,
deep in the layer,
is high.
Here, the electron density is in cm$^{ - 3},$
and the electron temperature is in Kelvins.
As can be seen from the plots in Illarionov and Sunyaev (1975),
comptonization produces the Wien spectrum deep in the layer at $y>10,$
while at $y > 10^5$ the spectrum is similar to a blackbody
due to the comptonization of low-frequency photons.
Consequently,
the combined effect of bremsstrahlung and comptonization
produces a blackbody spectrum deep in the layer.
The emission from the dense underlying layers
is yet another source of soft photons for comptonization.
Nevertheless, the emergent spectrum differs markedly from a blackbody spectrum,
because at each frequency we see a portion of the spectrum
that is produced at its own depth
[see Illarionov and Sunyaev (1972) for a discussion of this problem].
{\bf Integrated spectrum of the spread layer.}
The spread-layer spectrum is given by the integral
of the emergent spectra at each point of the layer over its surface.
Here, it is important
that the spectrum and the local radiative flux depend on latitude.
The integrated spectrum is not presented
for the following two reasons.
(a) A considerable fraction of the spread-layer radiation
falls on the disk surface (see Fig. 2).
This fraction is partially reflected from the disk
and partially reprocessed by it into softer radiation.
The intensity of the disk-reprocessed radiation
depends on the binary's inclination (see Lapidus and Sunyaev 1985).
Allowance for these effects requires a special study.
(b) Because of the shielding by the disk,
the observer cannot see the lower half of the neutron star.
Part of its surface is located in the shadow region
and does not contribute to the direct rays viewed by the observer
(here, we disregard the general-relativity effects).
In Fig. 11,
the flux of "direct" photons from the stellar surface is plotted
against the cosine of the binary's inclination angle, $i,$
which is measured from the star's polar axis (see Figs. 1-4).
In these calculations,
we assumed the indicatrix of emission per unit area to be
$\varphi (\mu) = 1 + 2.06 \, \mu
\;(\mu = \cos i);$
this is the first approximation for the angular dependence
of the intensity of the emergent radiation from a scattering atmosphere
(Sobolev 1949; Chandrasekhar 1960).
In the calculations, we took into account the profiles $q( \theta )$
(see Figs. 6-8).
\begin{figure}[tbh]
\epsfxsize=9cm
\epsfbox[-220 170 392 700]{r11.ps}
\caption{Functions $I(\mu)/I(0)$
$ [I$(erg s$^{ - 1}$ steradian$^{ - 1},$ $\;\mu = \cos i]$
which give the intensity of the radiation from the visible neutron-star surface.
The binary's inclination, $i,$ is measured from the polar axis.
Labels 1, 2, 3, and 4 refer to
$L_{SL}/L_{edd} = 0.01, 0.04, 0.2,$ and $0.8,$ respectively.
The functions are normalized to the intensity of the radiation from the equator $I(0).$}
\end{figure}
Clearly, the presence of a "shadow" and the reflection by the disk
must lead to an appreciable polarization of the emission
from an accreting neutron star.
At low luminosities,
at which the radiating belt is bound by the equatorial zone,
the flux toward an observer in the equatorial plane $(\mu = 0)$
exceeds the flux toward the polar axis $(\mu = 1).$
Curves 1-3 in Fig. 11 correspond to these luminosities.
At high luminosities, the emission extends to the entire stellar surface
(curve 4 in Fig. 7).
In this case,
the flux toward the polar axis is greater
than the flux toward the equatorial plane
(curve 4 in Fig. 11).
Indeed, the ratio $I(1)/I(0) = 2$ for a uniform emission
of the stellar surface,
because the disk occults the lower stellar hemisphere.
(III) {\bf Deceleration efficiency.}
The meridional distributions of rotation velocity $v_{\varphi}$
are shown in Fig. 12.
We performed the calculations for $\delta = 10^{-2},$
where $\delta = 1 - W_0 = (v^k_{\varphi} - v^0_{\varphi})/v^k_{\varphi},$
where $v^0_{\varphi}$ is the initial rotation velocity in the spread layer
when passing from the disk to the layer,
and $v^k_{\varphi}$ is the Keplerian velocity (see Subsecs. 3.4 and 4.6).
Similar results are also obtained for different values of $\delta \ll 1.$
As $\dot M$ increases,
the deceleration of rotation requires an increasingly large area
of contact
between the radiating belt and the stellar surface.
The boundary of this belt is $\theta_{\star}.$
\begin{figure}[tbh]
\epsfxsize=11cm
\epsfbox[-100 180 450 665]{r12.ps}
\caption{Meridional profile of the rotation velocity.
The radiating belt is simultaneously the rotating belt (cf. Figs. 6-8).
Labels 1, 2, 3, and 4 refer to
$L_{SL}/L_{edd} = 0.01, 0.04, 0.2,$ and $0.8,$ respectively.}
\end{figure}
An interesting characteristic of the deceleration efficiency
is the number of rotations
a test particle makes around the star
before it looses its rotation velocity
and reaches the latitude $\theta_{\star}.$
\vspace{.2cm}
\centerline{{Table.} Dependencies of
$\theta_{\star},$
$N_{rot}$
and $q/Q^+$ on $L_{SL}/L_{edd}.$}
\vspace{.2cm}
\hspace{4.3cm}
\begin{tabular}{|c|c|c|c|c|} \hline
$L_{SL}/L_{edd} $ & 0.01 & 0.04 & 0.2 & 0.8 \\ \hline
$R\theta_{\star},$ km & 0.43 & 1.36 & 4.5 & 16.4 \\ \hline
$N_{rot} $ & 0.3 & 1.1 & 5.4 & 36 \\ \hline
$(q/Q^+)_{max} $ & 1.01 & 1.05 & 1.3 & 3.8 \\ \hline
\end{tabular}
\vspace{.2cm}
\noindent The table shows
how the width of the belt $\theta_{\star}$
and the number of rotations,
$$
N_{rot} = \Delta \varphi/2\pi,\;\;\;\;\;
\Delta\varphi = \int_{\theta_0}^{\pi/2} \,\omega \,dt \approx
\int_0^{\theta_0} \frac{v_{\varphi}}{v_{\theta}}\;
\frac{d \theta}{\cos \theta},
$$
of its rotating plasma
depend on the spread-layer relative luminosity $L_{SL}/L_{edd.}$
As we see from the table, at a low spread-layer luminosity $(0.01 L_{edd}),$
a test particle does not make even one rotation around the star,
rising by 430 m along the meridian and covering 23 km in latitude.
Note that the meridional boundary-layer extent (see Fig. 1)
$R \theta_{bl}= R \sqrt{2 T_{bl}/m_p}/v_{\varphi}^k$
is a mere 70 m in the standard approach (at $T_{bl} = 3$ keV).
At a high luminosity, the deceleration efficiency is much lower:
at $L_{SL} = 0.8 L_{edd},$
test particles make 36 rotations around the star
before they slow down and reach $\theta_{\star}.$
(IV) {\bf Advection of radiative energy.}
The spreading matter in an optically thick layer transports radiative energy
along the meridian from the equatorial region
into the region of the bright rings.
The energy transfer is described by the energy equation
(see Subsecs. 2.7 and 3.2).
Our calculations (see Fig. 13) show that the energy flux $q$
emitted per unit area of the spread layer in the equatorial region
is much smaller than the local energy release
through the turbulent friction in a column
$Q^+ = \int \, \dot \epsilon_t \, dy,$
where the integral is taken over the spread-layer thickness
[Subsec. 2.5, (2.17)].
On the other hand,
the flux $q$ at higher latitudes for larger luminosities exceeds appreciably
$Q^+$
[see the table,
which gives the dependence of maximum $(q/Q^+)_{\max}$ in the spread layer
on $L_{SL}/L_{edd}].$
This is only to be expected,
because the spread-layer optical depth for Thomson scattering
is large at high luminosities.
As a result, the characteristic diffusion time $t_d$ of the photons
from the base of the hot layer
turns out to be of the order of the time of hydrodynamic motion $t_h$
through the radiating belt,
where $t_d\sim (h/c)\,\tau_T$
and $t_h \sim R \theta_{\star}/v_{\theta}.$
The advection of radiative energy favors
the formation of bright latitude rings at the stellar surface.
\begin{figure}[tbh]
\epsfxsize=9cm
\epsfbox[-210 279 400 650]{r13.ps}
\caption{Spread dynamics is determined in many ways
by the hydrodynamic transfer of radiative energy.
The ratio of the energy flux, $q,$
emitted per unit area of the radiating layer
to the frictional energy release $Q^+$
per unit area of the contact surface
between the spread layer and the star.
The energy is transferred from zone A into zone B
through meridional advection.
This calculation is for $L_{SL}/L_{edd} = 0.2.$}
\end{figure}
\subsection{Problems That Require Further Studies}
(I) {\bf Possible Instability of the Spread Layer.}
The energy-producing sublayer props up the spread flow from below,
and the radiation pressure dominates.
This may be the cause of convective instability
with floating 'photon bubbles'.
Clearly, this requires that the spread layer be optically thick.
The fluctuations which are produced by the bubbles
may be responsible for the observed low-frequency noise
and even for quasi-periodic oscillations in accreting neutron stars.
Although, on the other hand,
the rotation velocity, $v_{\varphi},$ inside the spread layer
outside the energy-producing sublayer,
along with the normal component of the centrifugal force,
slightly increases with height.
In addition,
the tail\footnote{
The frictional heat release
concentrates mainly in the energy-producing sublayer;
see Subsec. 2.5.}
of the volume density of dissipative energy production
$\dot \epsilon_t$ erg s$^{ - 1}$ cm$^{ - 3}$
extends from the energy-producing sublayer into the spread layer.
As a result,
the flux $q$ and, hence, the Eddington force slightly increase
with distance from the surface.
These two effects play a stabilizing role
with respect to the convective instability.
The instability is also stabilized
by turbulent viscosity and turbulent diffusion,
which are attributable to the presence of a wall
(the stellar surface from below)
and of shear turbulence.
This turbulence causes, first, the mixing of density fluctuations
(diffusive mixing of photon bubbles)
and, second, the momentum exchange between the bubbles and the ambient medium
(viscous deceleration).
The problem of the vertical stability of hydrostatic quasi-equilibrium,
indubitably,
requires further study.
(II) {\bf Gap between last stable orbit and surface.}
Here, we do not consider the case
where the stellar radius $R$ is smaller
than the radius of the marginally stable Keplerian orbit
$R_c = 3 R_g.$
In this case,
part of the gravitational energy is released in the zone of contact
between the disk and the star
due to an appreciable radial velocity
(it increases with the increasing $R_c - R)$
of the accreting matter on the stellar surface.
The energy release at the equator causes the layer to swell in this zone
because of the radiation pressure,
and the $1D$ approximations appear to become inapplicable.
(III) {\bf Rapid rotation.}
The case of rapid rotation of a star,
where its angular velocity, $\omega_S,$
accounts for an appreciable fraction
of the Keplerian angular velocity, $\omega_k,$
requires a special analysis.
\subsection{Structure of the Paper}
Here, we derive and numerically solve the system of equations of motion
of the matter over a neutron-star surface in the $1D$ approximation.
The vertical (along $r)$
and horizontal (along the meridian $\theta)$ scales
are separated in this approximation.
The gradients in $r$ are assumed to be large
compared to the gradients in the polar angle $\theta.$
For this reason,
the vertical profile is constructed
from the conditions of hydrostatic quasi-equilibrium.
The profile is given by simple analytic expressions.
The averaging of the quantities
over the vertical structure
with which the derivation begins
allows us to obtain a system of equations
for these quantities
that contains only gradients in $\theta.$
These are the equations of radiation-dominated shallow water
that describe the dynamics of the high-velocity layer
in which the newly accreted matter flows over the surface
of the slowly moving high-density underlying layers (Sec. 2).
This completes the part of the paper
in which we formulate the problem.
The part
in which we present the results
is devoted to the steady-state solutions of our model.
In this part, we describe the physics of adaptation:
effects of the friction model,
Eddington swelling of the boundary layer,
regulation of the deceleration by the swelling, etc. (Secs. 3 and 4).
We study the mathematical properties of radiation-dominated shallow water:
the structure of the solutions,
the limiting set of steady-state solutions,
its one-parameter structure,
the critical solution, etc. (Sec. 3).
We analyze the dependences of the boundary-layer characteristics
on $\dot M$
and on model parameters (Secs. 3 and 4).
Note that it follows from the solutions
that the bulk of the radiating belt
is roughly in {\it hydrostatic equilibrium along the meridian,}
as was previously suggested by Shakura and Sunyaev (1988).
In this case,
the "pressing" force to the equator is the tangential component
of the centrifugal force,
which counteracts the meridional component of the pressure gradient.
The interaction of the spreading flow with the underlying matter
is described in Subsec. 2.1.
Here, we also estimate the effects of disk heating
by the spread-layer radiation
and determine the minimum possible disk thickness near the neck.
The neck forms a kind of base of the disk
at the location of its interaction with the spread layer.
In Subsec. 2.2, we determine the structure
of a radiation-dominated atmosphere
in which the scattering by free electrons is a major contributor
to the opacity.
The force and energy characteristics of this atmosphere
which are required to calculate its energy and transport properties
are derived in Subsec. 2.3.
A simple formula for radiative losses in the spread layer
is written out in Subsec. 2.4.
Subsec. 2.5 deals with the friction model.
In Subsecs. 2.6 and 2.7, we derive the equations of "radiation" shallow water.
Transformations of this system of equations are made in Subsecs. 3.1-3.3.
In Subsecs. 3.4 and 3.5,
we study the phase space of the system
and introduce an important surface -- the "levitation" surface.
The dependence of the spread velocity on the Mach number
is analyzed in Subsec. 3.6.
The solutions of the derived system of equations
are analytically and numerically analyzed in Subsecs. 4.1 and 4.2.
Subsec. 4.3 investigates the critical regime
in which the spread velocity reaches the speed of sound
for the spread of matter at the edge of the radiating belt.
The dynamics of the matter spread outside the radiating belt
is analyzed in Subsec. 4.4.
The effect of friction on the results is considered in Subsec. 4.5.
Such an analysis is necessitated by the fact that,
first, the friction coefficient is determined only approximately
and, second, this coefficient decreases by a factor of 1.5 to 2
in the supersonic regime (see Subsec. 2.5).
We construct the dependence $q(\theta)$ in Subsec. 4.6
and study how the spread flow is modified
as the accretion rate $\dot M$ is varied in Subsec. 4.7.
At large $\dot M,$
the "squeezing" of the flow due to the convergence of meridians to the pole
becomes important.
In addition, the spread flow in the radiating belt
is strongly subsonic at high $\dot M$
and transonic at low $\dot M.$
In Subsec. 4.8, we estimate the limiting magnetic field strength
that can still be ignored
when the deceleration and spread are studied.
The results are summarized in the conclusion.
\section{HYDRAULIC APPROXIMATION}
\subsection{Spread Scheme}
The spread flow is illustrated in Fig. 4.
Here, $e$ is the equatorial plane,
$D$ is the disk,
$I$ is the transition region
between the disk and the boundary layer
(the region of $2D$ flow),
and $S$ is the stellar surface.
Let $h(\theta)$ be the boundary-layer thickness.
For the $1D$ approximation to be valid, it is necessary that
$$
h\ll R, \;\;\;\;\;\;\; \;\; |dh/d l| \ll 1, \eqno (2.1)
$$
where $l = R \theta$ is the arc length along the meridian.
Note, incidentally, that it follows from the calculations (see below)
that the angle $\arctan |dh/dl|$ is constant
in order of magnitude at $0 < \theta < \theta_{\star}.$
The latitudinally and meridionally moving spread layer
draws (grips) the thin underlying layers into motion
via turbulent friction.
The underlying layer is indicated by the dashed line in Fig. 4.
The $v_{\varphi}$ distribution inside the boundary layer is roughly uniform
in height (see Subsec. 2.5 for a discussion).
Almost all of the drop in $v_{\varphi}$ takes place near the surface, $S,$
which is the bottom of the boundary-layer,
$r = r_S$ or $y = y_S.$
The main turbulent energy release occurs near the bottom.
Since the problem is a steady-state problem,
the temperature $T$ is constant at $y < y_S$
(these layers were heated earlier,
and the energy release in them is small),
while the vertical radiative flux $q$
is approximately constant above $y_S$
and equal to zero below.
Because of the abrupt change in $q$ at the surface $S,$
the plasma parameters also change when this surface is crossed.
The underlying-layer density $\rho_2$ exceeds the boundary-layer density
in the hot subregion
$0 < \theta < \theta_{\star}.$
In the region $\theta > \theta_{\star},$
the density of the underlying layers rapidly increases with depth,
because they are isothermal.
For this reason,
the velocity in the underlying layers rapidly decreases with depth.
{\bf Disk heating by spread-layer radiation.}
Let us find out
how the disk thickness near a neutron-star surface will change
under spread-layer radiation.
The angular rotation velocity $\omega$ of the accreting matter
is known to reach a maximum in the transition zone
between the disk and the spread layer (Shakura and Sunyaev 1973).
Accordingly,
$ d\omega (r=r_n)/dr=0.$
This point in the flow is called the neck (see Figs. 1, 3, and 4).
The dissipative production of thermal energy vanishes at $r_n,$
because the derivative of $\omega (r)$ at this point changes sign
and because the heat release,
which is proportional to $(d\omega/dr)^2,$
is small in the neck zone.
The surface density of viscous energy release in the disk
(in Newton's approximation)
behaves as
$Q^+=$
$(3 \gamma M \dot M/4\pi r^3)(1 - \sqrt{R/r}).$
At $r\gg R,$ the energy release $Q^+$ is a factor of 3 greater
than the local gravitational energy release
$\gamma M \dot M/4\pi r^3,$
which is attributable to the transfer of mechanical energy
by viscous forces over the disk from regions with smaller $r$
to regions with larger $r.$
At the same time, $Q^+ \to 0$ as $r\to R.$
Therefore, the disk half-thickness,
$h=$
$(3/8\pi)(\sigma_T \,\dot M/m_p \, c)(1 - \sqrt{R/r}),$
also decreases as one approaches the neck (Shakura and Sunyaev 1973).
This consideration ignores disk heating by the spread-layer radiation
and the pressure of external radiative flux.
Let us estimate
how the heating by external radiation
affects the disk half-thickness near the neck.
Let $T_s,$ $T_c,$ and $T_r$ be the disk-surface temperature,
the temperature in the disk equatorial plane,
and the external-radiation temperature,
respectively.
Since the flux from the spread layer
accounts for a fraction of the Eddington flux,
its temperature is $T_r = 1-1.5$ keV.
The disk albedo is assumed to be 0.3-0.5.
Consequently, the disk surface near the neck
will be heated to temperatures of the order of 1 keV.
Clearly, the surface temperature
under quasi-steady-state conditions
(the radial velocity of the matter near the neck is low)
is less than or equal to $T_c.$
Using a hydrostatic estimate of the half-thickness,
$h = R c_s/v_{\varphi}^k,$
and assuming that $c_s \sim \sqrt{T_c/m_p},$
we obtain
$h = 30 R_6^{3/2}\, \sqrt{T_c}/\sqrt{M/M_{\odot}}$ m,
where $T_c$ is in keV,
and $R_6 = R/(10$ km).
It thus follows that the disk angular half-thickness, $\theta_0,$ is finite.
The conclusion about the finite but small disk half-thickness near the neck
is crucial for the validity of our approach
to the problem of accretion onto a neutron star.
Here, we do not consider the advection solution for the flow in the disk
(Narayan and Yi 1995)
for two reasons.
First, the neutron star has a solid surface,
and it is, therefore, difficult to imagine accretion
without energy release on the surface.
Second, the flux from the spread layer is so large
that it will cool the plasma flow
at a distance of several tens of neutron-star radii
through comptonization.
As a result,
an accretion pattern with a relatively thin (geometrically) accretion disk
is most likely realized near the star.
\subsection{Structure of the Spread Layer (Atmosphere)}
We make the following assumptions:
(I) The energy is mainly released in a thin sublayer
near the boundary between the spreading and underlying layers.
(II) Thomson scattering makes a major contribution to the opacity.
(III) We define the free fall acceleration $g_{eff}$
within the spreading layer
with allowance for the contribution of the centrifugal acceleration
as the difference
$g_{eff} = \gamma M/R^2 - v_{\varphi}^2/R - v_{\theta}^2/R$
[see (1.1) and (1.1)'].
For completeness, we added the centrifugal acceleration
due to the meridional motion.
It is small compared to the acceleration $\gamma M/R^2.$
The rotation velocity, $v_{\varphi},$
is assumed to be constant within the atmosphere
(if $\theta$ is fixed).
Under these assumptions,
the atmospheric structure is described by the system of equations
$$
p=g_{eff} \; \Sigma, \eqno (2.2)
$$$$
q =\frac{c\,\Sigma_T}{3}\;\frac{d\epsilon}{d\Sigma}, \eqno (2.3)
$$$$
p=2\;\frac{\rho\, T}{m_p}+\frac{\epsilon}{3}, \eqno (2.4)
$$$$
\epsilon=a \; T^4. \eqno (2.5)
$$
The system of equations of hydrostatics (2.2),
radiative heat conduction (2.3),
thermodynamic state (2.4),
and Stefan-Boltzmann's law (2.5)
relates the unknown functions $p, \epsilon, \rho,$ and $T$
of the vertical column density $\Sigma,$
which is a Lagrangian coordinate $(d \Sigma/dy = \rho).$
The $y$ and $\Sigma$ axes are directed downward.
Accordingly, the pressure in (2.2) increases with $\Sigma.$
The $q$ is the absolute flux.
We assume that the layer borders vacuum $(p = 0)$
at $y = 0, \Sigma = 0.$
The boundary-layer bottom
$y = y_S = h,$
$\;\Sigma = \Sigma_S$
is the surface $S$ (see Fig. 4).
The system of equations (2.2)-(2.5)
is written for an arbitrary fixed latitude $\theta.$
Due to the approximate assumption\footnote{
$v_{\varphi}$ is assumed to depend on latitude alone.
By $v_{\varphi}$ we mean the $r$-averaged velocity in the boundary layer.}
of constant $v_{\varphi}(\theta)$ in the boundary layer $0 < y < h,$
the acceleration $g_{eff}$ in (2.2) is constant
as $\Sigma$ increases deep into the layer.
Eq. (2.3) holds for radiative transfer in an optically thick plasma
with $\Sigma_S \gg \Sigma_T$ with dominating Thomson scattering.
We assume that the turbulent dissipation is localized near the surface $S.$
It thus follows that the heat flux $q$ in the spread layer $0 < y < h,$
which is transferred via radiative heat conductance,
is approximately constant.
The total pressure consists of the plasma and radiation pressures.
The system of equations (2.2)-(2.5) can be easily solved.
The solution that satisfies the boundary condition
$$
p(0)=0, \;\;\;
\epsilon(0)=0, \;\;\;
\rho(0)=0, \;\;\;
T(0)=0, \eqno (2.6)
$$
is
$$
p = g_{eff} \Sigma,\;\;\;
\epsilon = \frac{3q}{c}\,\tau_T,\;\;\;
\rho = \frac{m_p\,g_{wr}\,\Sigma_T}{2}\;
\left( \frac{a c}{3q}\;\tau_T^3 \right)^{1/4},\;\;\;
T = \left( \frac{3q}{a c}\;\tau_T \right)^{1/4}, \eqno (2.7)
$$$$
g_{wr}=g_0\, G_{wr},\;\;\;\;
G_{wr} \equiv \Delta =G_{eff}-G_r,\;\;\;\;
G_r=\frac{g_r}{g_0},\;\;\;\;
g_r=\frac{q}{c \,\Sigma_T}, \eqno (2.8)
$$
where $G_{eff}=g_{eff}/g_0$ [see Subsec. 2.2 (III)].
The difference, $\Delta,$ of the gravitational acceleration,
the component of the centrifugal acceleration normal to $S,$
and the radiation acceleration $g_r$ (2.8)
is particularly important for the subsequent discussion.
Let us derive the dependence on $y.$
The Lagrangian and Eulerian differentials
are related by $d\Sigma = \rho \, dy.$
Substituting the solutions (2.7) and (2.8) into this relation
and integrating it using the conditions (2.6),
we obtain
$$
\Sigma = \frac{\beta \; \Delta^4}{3\times 2^{12}\, q}\, y^4 =
\frac{y\,\rho}{4}, \;\;\;
\rho = \frac{\beta\;\Delta^4}{3\times 2^{10}\, q}\,y^3,\;\;\;
\beta = m_p^4\;g_0^4\;a\,c\,\Sigma_T, \eqno (2.9)
$$$$
p_{pl} = g_{wr}\;\Sigma,\;\;\;
p_r=g_r\,\Sigma,\;\;\;
p=p_{pl}+p_r = g_{eff}\,\Sigma,\;\;\;
T=\frac{m_p\;g_{wr}}{8}\,y. \eqno (2.10)
$$
These solutions are very simple.
The atmosphere under consideration is similar in structure
to the atmospheres of X-ray bursters during outbursts
and to the atmospheres of supermassive stars.
We describe these solutions in some length,
because the calculations of transport characteristics
are based on them.
\subsection{Lateral Force and the Surface Energy Density}
Here we study the dynamics of a layer of shallow water or a fluid film
that flows over the underlying surface $S$ (see Fig. 4).
The thickness, $h,$ of the layer is assumed to be small
compared to its horizontal extent $R \theta$ [see (2.1)].
When the equations with gradients in $\theta$ are derived,
the layer is broken up into the differential elements
$R \,\delta \theta.$
The force interaction between adjacent elements
is effected by the lateral force
$\int_0^h\, p\,dy.$
Substituting (2.9) and (2.10), we obtain
$$
\int_0^h\;p \,dy =
\frac{8}{5}\;\frac{G_{eff}}{\Delta}\;\frac{T_S\; \Sigma_S}{m_p},
\eqno (2.11)
$$
where, as above,
the subscript $S$ denotes values near the bottom of the spread layer.
It is important that the force (2.11)
contains the large dimensionless factor $G_{eff}/\Delta$
at $\Delta \ll 1.$
The layer transports mass, momentum, angular momentum, and energy
over the spherical surface $S.$
In particular, the advection of radiative energy takes place.
The expressions for the surface densities of mass,
momentum, and angular momentum are obvious:
$\Sigma_S,$
$\;v_{\theta}\, \Sigma_S,$
$R\,\cos\theta \, v_{\varphi}\,\Sigma_S,$
where $v_{\theta}(\theta)$ is the $r$-averaged meridional velocity
in the layer,
and $R \cos \theta$ is the arm relative to the rotation axis,
which the polar axis is.
Let us calculate the surface energy density (erg cm$^{ - 2 })$
using the distributions (2.9) and (2.10).
The total energy
of the layer consists of the kinetic energy
$
K = (v_{\theta}^2 + v_{\varphi}^2) \, \Sigma_S/2,
$
the gravitational energy $(E_g),$
and the internal energy $(E_{int}).$
In turn, the latter consists of the plasma $(E_{pl})$ and radiative $(E_r)$
contributions.
The surface density of gravitational energy is
$$
E_g = \int_0^h\;\rho\,g_{eff}\, (h-y) dy=
\frac{8}{5}\;\frac{G_{eff}}{\Delta}\;\frac{T_S\;\Sigma_S}{m_p}. \eqno (2.12)
$$
Interestingly, the expressions (2.11) and (2.12) are identical.
The energy (2.12) is measured
from the bottom of the layer $y = h.$
Since the volume density of plasma internal energy is
$3 \rho T/m_p = (3/2)p_{pl},$
we have
$E_{pl}=(3/2) \int_0^h \; p_{pl}\, dy.$
Calculating this integral and the integral
$$
E_r = \int_0^h\; \epsilon\,dy =
(24/5)\,(G_r/\Delta)\,T_S \, \Sigma_S/m_p,
$$
we obtain
$$
E_{int}=\frac{12}{5}\;\frac{G_{eff}+G_r}{\Delta}\;
\frac{T_S\;\Sigma_S}{m_p}. \eqno (2.13)
$$
All expressions (2.11)-(2.13)
that appear in the formulas for horizontal interactions
in the layer
increase as the layer swells with $\Delta\to 0.$
\subsection{Radiative Cooling}
The energy losses by the layer surface are determined by the flux
which comes from the inside
and can be readily expressed in terms of the layer surface temperature, $T_v,$
on the side of vacuum
$(q = a\, c\, T_v^4/4).$
This is the temperature at an optical depth of the order of unity
$(\Sigma_v \simeq \Sigma_T).$
The solutions (2.7) and (2.8),
in which the integration constant
for the equation of radiative heat transfer (2.3)
was discarded,
is valid at large $\Sigma$ $\;(\Sigma \gg \Sigma_T).$
Formula (2.7) for $T$ allows the local flux, $q,$ to be expressed in terms
of the local $\Sigma$ and $T$
$\;(q = a\,c\,T^4/3\,\tau_T).$
In particular, inside the spread layer near its bottom,
we obtain
$$
q = a \, c \, T_S^4 \; \Sigma_T/3 \, \Sigma_S. \eqno (2.14)
$$
\subsection{Turbulent Deceleration of Rotation and Viscous Energy Release}
{\bf Profiles of average and fluctuation velocities.}
Let us dwell on friction and dissipation
and analyze the friction of the boundary layer against the star.
The turbulence via the friction
it produced
decelerates and heats up the layer.
In this case, the rotation and meridional spread of the matter in the layer
draws the underlying layers into motion
through turbulent viscous gripping.
This issue is discussed in Subsec. 2.1.
Since the source of heat concentrates near the boundary $S,$
the temperature under the surface $S$ is constant,
while the entropy $s$ rapidly decreases with depth.
Stratification with a large reserve of gravitational stability arises
(the Richardson number is great in the underlying layers).
It is difficult for the high-entropy layer,
which moves from above,
to "grip" the low-entropy underlying layer.
This situation roughly corresponds to a wind above a smooth surface.
{\it Turbulence in a gaseous layer}
that flows with a subsonic velocity above a fixed boundary
has been well studied and described by
Loitsyanskii (1973),
Schlichting (1965),
Chugaev (1982),
Landau and Lifshitz (1986),
and, recently, by Lin et al. (1997).
For low viscosity,
the height distribution of the mean velocity
is described by Prandtl-Karman's universal logarithmic profile
$$
\langle v \rangle \approx \frac{5}{2} \, v_{\star}\,
\ln (7.7 \;v_{\star}\; \Delta y/\nu),\;\;\;\;\;\;\;
\tau = \rho v_{\star}^2, \eqno (2.15)
$$
where $\Delta y$ is {\it measured from the surface} $S,$
and $\tau$ is the tangential stress
which does not depend on $\Delta y.$
The parameter $v_{\star}$ is important.
It determines the turbulent velocity and pressure pulsations.
The universal profile is such
that $v_{\star}$ is velocity pulsations about the mean in all directions:
$v_{\star}\simeq$
$\sqrt{\, \langle (v_{\theta}- \langle v_{\theta} \rangle )^2 \rangle }\simeq$
$\sqrt{\langle v_r^2 \rangle }\simeq$
$\sqrt{\, \langle (v_{\varphi}- \langle v_{\varphi} \rangle )^2 \rangle}$
or
$$
v_{\star}=
\sqrt{
\,\langle\,(\vec v-
\langle v_{\varphi}\rangle \, \vec e_z -
\langle v_{\theta}\rangle \, \vec e_x \,)^2 \,\rangle\,/3 },
$$
where $\vec v = \{v_{\theta}, v_r, v_{\varphi}\},$
and $x, z$ is the wall plane.
This implies that the pulsations are isotropic
at a given distance, $\Delta y,$ from the wall
in a comoving frame of reference
which moves with the mean velocity $\langle v \rangle$ of the flow
at this distance.
The order-of-magnitude amplitude of the pressure pulsations about the mean
is $\rho v_{\star}^2.$
An important point is that
the velocity $v_{\star}$ and the tangential stress $\tau$
in the region of turbulent mixing
do not vary as the distance to the wall $\Delta y$ is varied.
Actually, the law (2.15) follows from the condition
of the invariance of $\tau.$
When this law is deduced,
the stress is written in Newton's form
$\tau = \rho \, \nu_t \, d \langle v \rangle /dy,$
and the mixing-length theory is used,
in which the turbulent viscosity $\nu_t = l_t\, v_{\star}$
is determined by the scale of the wall vortices
$l_t = \kappa \, \Delta y.$
The coefficient $\kappa\approx 0.4$ is called Karman's coefficient.
The integration constant in Newton's friction law
$\langle v\rangle = \int dy \,\tau \dots,$
which appears in (2.5) under the logarithm,
is chosen by joining the viscous and turbulent solutions
in the transition region between the viscous sublayer and the turbulent region.
Formula (2.15) holds outside the viscous sublayer,
when $\Delta y > l_{\nu},$
$\; v_{\star} \, l_{\nu}/\nu \approx 12,$
where $l_{\nu}$ is the thickness of the viscous sublayer
which is determined by the molecular and radiative viscosities.
The disk-accretion theory commonly uses an expression
for the kinematic viscosity $\nu_t$
of the form $\nu_t = \alpha \, h_d \,(c_s)_d,$
where $h_d$ and $(c_s)_d$ are the disk thickness and the speed of sound
in the disk, respectively.
For the Keplerian dependence of angular velocity
$\omega_d$ on $r,$
this expression reduces to a simple formula for the stress
$\tau = \alpha p,$
where $p$ is the pressure in the disk (Shakura and Sunyaev 1973).
Let us compare the formulas
$ \nu_t= \alpha \, h_d \, (c_s)_d$
and $\nu_t= \kappa \, \Delta y \, v_{\star}.$
We see that the wall turbulence differs from the disk turbulence,
first, by the form in which the mixing length is written
$(\kappa \, \Delta y$ instead of $h_d)$
and, second, by the formula for turbulent velocity pulsations
$[v_{\star}$ instead of $\alpha \, (c_s)_d].$
{\bf Friction coefficient.}
Let us calculate the stress $\tau.$
For this purpose, we extend the distribution (2.15),
which is weakly dependent on $\Delta y,$
to the height $\Delta y = h.$
To obtain an estimate,
we assume that at this height
the mean velocity of flow $\langle v \rangle$
is equal to the Keplerian velocity.
The ion viscosity is
$\nu_i=$
$2.2 \times 10^{-15} \; T^{2.5}/\rho_S \; \ln \Lambda$
$=1.3 \times 10^4$ cm$^2$ s$^{-1},$
where,
$T$ is in degrees,
and $\rho_S$ is in g cm$^{-3}$ (Spitzer 1962).
We also assume the following:
$h = 100$ m,
$\Sigma_S = 300$ g cm$^{-2},$ $T = 3$ keV,
and the Coulomb logarithm $\ln \Lambda = 10.$
This viscosity is small compared to the {\it radiative viscosity}
$$
\nu_r = \frac{4}{15} \; \frac{a \,T^4\, \Sigma_T}{c\,\rho^2}
\simeq \frac{\rho_r}{\rho} \, l_T \; c
$$
(Weinberg 1972);
here, $\rho_r = a T^4/c^2$ is the photon "density",
$l_T = 1/n \sigma_T$ is the photon mean free path,
and $n = \rho /m_p.$
It follows from the numerical calculations
that the radiation density $\rho_r$
accounts for a small fraction $(\sim 10^{-2})$
of the plasma density $\rho_S.$
For the above values of $\rho$ and $T,$
we obtain $\nu_r = 1.7 \times 10^7$ cm$^2$ s$^{-1}.$
The mixing is governed by the radiative viscosity:
$\nu_r \gg \nu_i.$
Substituting $\nu_r$ in (2.15)
and setting $M = 1.4 M_{\odot},$
and $R = 12$ km
when calculating the Keplerian velocity,
we arrive at the equation
$4.2 \times 10^9 = v_{\star} \ln (v_{\star}/220),$
where the unknown is in cm s$^{-1}.$
The estimates are given in the approximation of Newton's potential.
Solving this equation,
we obtain
$v_{\star} \approx 3\times 10^8$ cm s$^{-1}.$
In this case, the viscous-layer thickness
$l_{\nu} \approx$
$12 \nu/v_{\star} \approx$
$0.7$ cm is much smaller than the scale height $h.$
The Reynolds number is
Re$ = h v_{\varphi}^k/\nu \sim 10^{7}.$
Let us write
$$
\tau = \alpha_b \; \rho \; v^2. \eqno (2.16)
$$
The coefficient of friction against the bottom
is $\alpha_b = v_{\star}^2/v^2$
[cf. (2.15) and (2.16)].
In our example,
$\alpha_b \approx (v_{\star}/v_{\varphi}^k)^2 \approx 0.81 \times 10^{-3}.$
The value of $\alpha_b$ changes only slightly
as the parameters are varied.
In Subsec. 3.1, we compare the terms
which are associated with $\alpha_b$ (wall turbulence)
and $\alpha$
(a coefficient that is widely used in the disk-accretion theory).
{\bf Dissipation and the height distribution of energy release.}
The volume power density (erg s$^{-1}$ cm$^{-3})$
of the heat source $\dot \epsilon_t$ attributable to turbulent friction
is given by
$$
\dot \epsilon_t (\Delta y) \approx
(5/2) \; v_{\star} \; \tau / \Delta y \eqno (2.17)
$$
(Landau and Lifshitz 1986).
It follows from (2.17) that the surface density
of the total viscous energy release is
$$
Q^+ = \int_0^h \, \dot \epsilon_t (\Delta y) \;d (\Delta y)\approx
\int_{l_{\nu}}^h \, \dot \epsilon_t \;d (\Delta y) \approx
2.5 \,\tau \,v_{\star}\; \ln \frac{h}{l_{\nu}}\approx \tau v. \eqno (2.18)
$$
Here, $v$ is the height-averaged velocity in the layer.
In the case under consideration,
where the flow moves latitudinally at velocity $v_{\varphi}$
and meridionally at velocity $v_{\theta},$
the total velocity
$\sqrt{ v_{\theta}^2 + v_{\varphi}^2 }$
must be substituted in (2.18).
The energy release (2.17)
takes place mainly near the bottom of the boundary layer.
For example,
$$
[\, \int_{0.1 h}^h \, \dot \epsilon_t \;d (\Delta y)\,] / Q^+
\approx 0.2.
$$
An analysis of the calculations
which are given below
shows that $v_{\varphi}$ is several times faster
than the speed of sound $c_s$ at $\theta < \theta_{\star}$ --
{\it the flow is moderately supersonic.}
In this case, the expressions (2.15)-(2.18) {\it remain valid}
(Loitsyanskii 1973; Schlichting 1965).
In a supersonic case with the Mach number
$\mathop{\rm Ma}\nolimits \simeq 5,$
$\alpha_b$ in (2.16) decreases by a factor of 1.5 to 2.
Above, we described the surface characteristics (Subsec. 2.3),
radiative cooling (Subsec. 2.4),
and dissipative heating (Subsec. 2.5).
Now, we can proceed to a derivation of the balance equations
that relates to all these physical processes.
\subsection{Equations of mass, momentum, and angular momentum transfer}
Let us derive the first three equations
of the system of four $1D$ equations
which represent the laws of conservation of mass
(equation $\Sigma),$
momentum (or angular momentum along the meridian) in $\theta$
(equation $v_{\theta}),$
angular momentum (equation $v_{\varphi}),$
and energy (equation $T).$
We take a film annulus
with the area $dS = 2 \pi R C \, R \delta\theta$
on the sphere $S,$
where $C = \cos\theta.$
Its mass is $dS \;\Sigma_S.$
The mass flux through the $\theta = \mathop{\rm const}\nolimits$ conic surface
(annulus boundary)
is
$2 \pi R C v_{\theta} \Sigma_S.$
Accordingly, the equation $(\Sigma)$ is
$$
R C (\Sigma_S)'_t = - (C \Sigma_S \; v_{\theta})'_{\theta}. \eqno (2.19)
$$
{\bf Equation $(v_{\theta}).$}
The momentum flux is
$2 \pi R C v_{\theta} \Sigma_S \; v_{\theta},$
the lateral force is
$2 \pi R C \int_0^h p dy,$
and the component of the centrifugal force
which is tangential to the sphere is
$(dS\;\Sigma_S\; v_{\varphi}^2/RC)\sin \theta.$
The meridional friction force $\tau_{\theta} \; dS$
is
given by the relations
$$
\tau=\alpha_b \;\rho_S \; v^2,\;
v=\sqrt{v_{\theta}^2+v_{\varphi}^2},\;\,
\tau_{\theta} = \tau \sin \beta,\;
\sin\beta=v_{\theta}/v,\;
\tau_{\theta}=\alpha_b \;\rho_S \; v_{\theta} \,v,
$$
where $\tau_{\theta}$ is the meridional component
of the total shear stress,
the vector $\vec \tau$ is directed opposite
to the total flow velocity vector
$v_{\theta} \vec e_{\theta} + v_{\varphi} \vec e_{\varphi},$
$\rho_S$ is the density at the bottom of the spread layer,
and $\beta$ is the angle
between the total velocity vector and the latitude.
These relations follow from (2.16).
Substituting (2.11) into the momentum equation, we obtain
$$
R C (\Sigma_S\; v_{\theta})'_t = - (C \Sigma_S\; v_{\theta}^2)_{\theta} -
\frac{8}{5} C \left(\frac{G_{eff}}{\Delta}\frac{T_S\Sigma_S}{m_p}\right)'_{\theta}
- R C \tau_{\theta} -
\sin\theta \Sigma_S \; v_{\varphi}^2. \eqno (2.20)
$$
The bottom density $\rho_S$ is related to the sought-for functions
$v_{\theta}, v_{\varphi}$ and $T_S$
(below, we eliminate $\Sigma_S$ by using the mass integral)
by
$$
\rho_S=\frac{4\,\Sigma_S}{h}=\frac{m_p \, g_{wr} \,\Sigma_S}{2 \,T_S},
\;\;\;\;\;\;\;\;\;
h=\frac{8 T_S}{m_p\, g_0 \,\Delta},
\eqno (2.21)
$$
where $q_{wr}$ is given by (2.8).
We see that at the same surface density of the matter,
the {\it layer thickness,} $h,$ is inversely proportional to $\Delta.$
The smaller $\Delta,$ the higher the accuracy
with which the radiation-pressure force offsets the difference
between the gravitational and centrifugal forces.
The subsystem (2.19), (2.20) resembles the $1D$ hydrodynamic equations
$\rho'_t = (\rho u)'_x,$
$\;\rho u'_t = - \rho u u'_x - (\rho T/m_p)'_x.$
Note the "pressure" enhancement in (2.20)
compared to the formula for an "ideal gas" $T_S\;\Sigma_S$
because of the factor $G_{eff}/\Delta$
(see similar remarks in Subsec. 2.3).
{\bf Equation $(v_{\varphi}).$}
The rate of change in the angular momentum of the boundary-layer annulus
is
$dS\,(\Sigma_S\;v_{\varphi})'_t \,R C.$
Writing the angular momentum flux
$ 2 \pi R \, C v_{\theta} \, \Sigma_S \; v_{\varphi} R \, C$
and the friction torque about the polar axis as
$$
\tau_{\varphi} \, dS \, R\, C, \;\;\;\;\;\;\;
\tau_{\varphi} = \alpha_b \;\rho_S \; v_{\varphi} \, \sqrt{v_{\theta}^2+v_{\varphi}^2},
$$
we obtain
$$
R\, C^2 \,(\Sigma_S\; v_{\varphi})'_t = - (C^2 \;\Sigma_S\; v_{\theta} \,v_{\varphi})'_{\theta} -
R \,C^2 \;\alpha_b \;\rho_S \;v_{\varphi} \,\sqrt{v_{\theta}^2+v_{\varphi}^2}. \eqno (2.22)
$$
In (2.22), in order to determine the latitudinal friction $\tau_{\varphi},$
we projected the total stress $\tau$ onto the latitude,
much as was done above
when the meridional friction was calculated.
\subsection{Energy Transfer and Emission by the Layer}
Let us write out the {\it energy equation} $(T).$
In this equation the formulas
for the rate of change in the total energy of the annulus,
for the enthalpy flux
(the total energy flux and the power of lateral forces),
and for the radiative losses
are used.
They are
$$
dS \;\left(K+E_{sum}\right)'_t,\;\;\;\;\;
E_{sum} = E_g + E_{int} =
\frac{4 T_S\;\Sigma_S}{5 m_p}\;\;\frac{5 G_{eff}+3 G_r}{\Delta},
$$
$$
2 \pi R C v_{\theta} \, (K+H),\;\;\;\;\;
H = E_{sum} + \int_0^h \, p dy = \frac{4 T_S\;\Sigma_S}{5 m_p}\;
\frac{7 G_{eff}+3 G_r}{\Delta}
$$
$$
q\, dS,\;\;\;\;\;\;\;K=(1/2) \, \Sigma_S\;(v_{\theta}^2+v_{\varphi}^2).
$$
Adding up these expressions,
we obtain the sought-for equation for the total energy
$$
R C \left(E_{sum}+ K\right)'_t =
-\left[C v_{\theta} \left(H+K\right)\right]'_{\theta}-R C q. \eqno (2.23)
$$
This equation allows for the energy (including radiative energy) advection
and radiative cooling.
\section{TRANSFORMATIONS AND ANALYSIS OF THE SYSTEM}
\subsection{Steady-State Spread}
In the steady-state case, Eq. (2.19) is integrable.
From the condition of {\it mass conservation,} we have
$$
\frac{1}{2}\;\dot M= 2\pi R C v_{\theta} \Sigma_S \eqno (3.1)
$$
i.e., the matter spreads out over the two hemispheres.
Since $C \,v_{\theta} \,\Sigma_S,$ is constant,
from the system (2.20), (2.22), and (2.23) we derive
$$
C\, \Sigma_S \; v_{\theta}\; v_{\theta}'+\frac{4}{5}\, C \left(\frac{G_{eff}}{\Delta}\,
\frac{2 \,T_S}{m_p}\,
\Sigma_S \right)' =
- R \,C \,\tau_{\theta} - \sin \theta \,\Sigma_S \; v_{\varphi}^2,
\eqno (3.2)
$$$$
\tau_{\theta}=\alpha_b \;\rho_S \;v_{\theta} \,\sqrt{v_{\theta}^2+v_{\varphi}^2},
$$$$
C \,\Sigma_S \; v_{\theta} \,(C \,v_{\varphi})' = - R\, C^2 \,\tau_{\varphi},\;\;\;\;\;\;\;
\tau_{\varphi}=\alpha_b \;\rho_S\; v_{\varphi} \,\sqrt{v_{\theta}^2+v_{\varphi}^2}, \eqno (3.3)
$$$$
C\, \Sigma_S \; v_{\theta} \left( \frac{v_{\theta}^2+v_{\varphi}^2}{2}+
\frac{2}{5}\;\frac{2 T_S}{m_p}\;\frac{7G_{eff}+3G_r}{\Delta} \right)'
= - R \,C \,q. \eqno (3.4)
$$
The prime denotes differentiation with respect to $\theta.$
Because of the integral (3.1),
the number of unknowns decreases.
Eqs. (2.23) and (3.4) reflect the absence of radiative flux
from the layer to the star.
In addition,
we ignore the mechanical work between the layer and the star
$(\tau v_{\varphi}^{star}\approx 0),$
because the star is massive
$(\rho^{star}$ is great even in the underlying layer)
and because the velocity of its surface layers is low
$(v_{\varphi}^{star}\ll v_{\varphi};$
see Subsecs. 2.1 and 2.5).
{\bf Comparison with $\alpha-$friction.}
{\it We disregard the friction between adjacent annuli}
(or differential elements $\delta \theta)$
in comparison with the friction against the bottom.
Indeed, in the theory of disk accretion (Shakura and Sunyaev 1973)
with $"\alpha$-friction",
the turbulent stress $\tau$
between the annuli is
$\alpha \, h \, c_s \, \rho \; (dv_{\varphi} / dx),$
where $h$ is the layer thickness (see Subsec. 2.1),
$h c_s$ gives an estimate
of the maximum possible hydrodynamic interaction,
$\alpha$ is a coefficient that is small compared to unity,
and $x$ is the arc length along the meridian.
If we take into account this interaction,
instead of Eq. (3.3) we obtain,
for example,
$$
- C \,v_{\theta} \,\Sigma_S\;(C\, v_{\varphi})' -
R\, C^2 \;\tau_{\varphi} +
(\alpha/ R) \,(C^2 \,h^2\, \rho\, c_s \, v_{\varphi}')' = 0.
$$
This equation contains the second-order derivative
with respect to $v_{\varphi}.$
The ratio of the third and second terms is
$$
(\alpha/\alpha_b)\,(c_s/v_{\varphi})\,(h/R\theta_{\star})^2.
$$
Neglect of the friction between adjacent annuli
is justified by the smallness of
$(h/R\theta_{\star})^2$
in the shallow-water approximation (2.1).
Note that the term with
$\tau_{\varphi}$
drops out in the angular momentum equation for an accretion disk --
there is no friction against the bottom.
If we set $C\approx 1$ in Eq. (3.4),
discard $v_{\theta}^2$
compared to
$v_{\varphi}^2,$
differentiate $v_{\varphi}^2,$
use (3.3) to eliminate $v_{\varphi}',$
and ignore radiative losses $q,$
then we obtain
$(v_{\theta} \,H)' = R \, \tau_{\varphi} \, v_{\varphi}.$
This equation together with Eq. (3.3)
describe the transformation of rotational energy into enthalpy.
\subsection{Natural Scales}
Let us rewrite the system (3.2)-(3.4) in dimensionless form.
In units of the stellar radius $R = 1$
and the Keplerian velocity at the stellar equator $v_{\varphi}^k=1,$
it takes the form
$$
U^2 U'+ \frac{4}{5} \,C\,U^2 \left(\frac{G_{eff}}{\Delta}\;\frac{\hat T}{C\,U}\right)'=
- F_{\theta} - F_{cf}, \;\;\;\;\; C\equiv \cos \theta\eqno (3.5)
$$$$
U W W' = - F_{\varphi} + F_{cf}, \eqno (3.6)
$$$$
U^2 U'+\frac{2}{5} U \left(\frac{7G_{eff}+3G_r}{\Delta} \hat T \right)' =
F_{\varphi} - F_{cf} - \dot E_{rad}. \eqno (3.7)
$$
In what follows,
$U=v_{\theta}/v_{\varphi}^k,$
$W=v_{\varphi}/v_{\varphi}^k,$
and
$\hat T=2T_S/m_p \,(v_{\varphi}^k)^2$
are the dimensionless sought-for functions.
Eq. (3.5) can be derived from (3.2) by multiplying it by
$v_{\theta}.$
Below, we give the functions and the notation
that are used to write the system.
The function
$$
\Delta=\Delta (U,W,\hat T;\theta)= 1-U^2-W^2-G_r,\;\;\;\;\;\;\;
G_r = B C U \hat T^4, \eqno (3.8)
$$$$
B=\frac{4\pi}{3}R^2\;\frac{a (m_p (v_{\varphi}^k)^2/2)^4}{\dot M v_{\varphi}^k}=
\frac{0.65\times 10^{18}(M/M_{\odot})^{7/2} }{
R_6^{5/2}\,(L_{SL}/L_{edd})} \eqno (3.9)
$$
[see (1.2), (2.8), (2.14), and (3.1)].
For completeness,
we included the centrifugal acceleration
which is associated with the meridional motion in expression (3.8) for
$\Delta.$
In the radiating layer,
it is small compared to the acceleration
which is associated with the latitudinal motion:
$ U^2 \ll W^2.$
In the dark layer, the centrifugal forces are insignificant.
In (3.9) and below
$\dot M = 10^{17} \, \dot M_{17}$ g s$^{-1}.$
The terms
$F_{\theta},$ $F_{\varphi},$ $F_{cf},$ and $\dot E_{rad}$
which are related to the friction,
the centrifugal force (cf),
and the radiation are given by
$$
F_{\theta}=\frac{\alpha_b U^2 V \Delta}{\hat T},\,\;\;\;
F_{\varphi}=\frac{\alpha_b W^2 V \Delta}{\hat T},\,\;\;\;
F_{cf}=\frac{S}{C} U W^2, \,\;\;\;
\dot E_{rad} = A C^2 U^2 \hat T^4. \eqno (3.10)
$$$$
A=\frac{1}{2}\; \frac{B}{(L_{SL}/L_{edd})},\;\;\;\;\;
S\equiv\sin\theta,\;\;\;\;C\equiv\cos\theta,\;\;\;\;
V=\sqrt{U^2+W^2}.
\eqno (3.11)
$$
We see that the decrease in $\Delta$
leads to a decrease in $F_{\theta}$ and $F_{\varphi}$
[see also (2.20)-(2.22)].
Physically, this is caused by the drop in density (2.21)
at the bottom due to an increase in the boundary-layer thickness.
The system of equations (3.5)-(3.7)
for the sought-for functions $U, W,$ and $\hat T$
includes three parameters, $L_{SL}/L_{edd},$ $B,$ and $\alpha_b.$
\subsection{Transformation to an Explicit Form}
The system (3.5)-(3.7) relates the derivatives of the unknown functions
$U, W,$ and $\hat T.$
For this system to be numerically integrated,
it must be resolved for the derivatives.
The functions $U, W,$ and $\hat T$
enter into the pressure and the enthalpy
which are differentiated in Eqs. (3.5) and (3.7)
in a complicated way via the functions
$\Delta, G_{eff,}$ and $G_r.$
Since the transformation of these equations to a form
that contains only the derivatives of the sought-for functions
turns out to be cumbersome,
it makes sense to give the final results of this transformation.
We have
$$
U' =( e_{\tau} u_{00}-u_{\tau} e_{00})/d, \eqno (3.12)
$$ $$
W'=\omega,\;\;\;\;\;\;\; \omega= ( - F_{\varphi} + F_{cf} )/U W, \eqno (3.13)
$$ $$
\hat T'=( - e_{u } u_{00}+u_{u } e_{00})/d, \eqno (3.14)
$$
where
$$
u_u=U^2-\frac{4}{5}\;
\frac{ G_{eff}^2 - 2 G_{eff} G_r - 2 U^2 G_r }{\Delta^2}
\;\hat T,\;\;\;\;\;\;\;
u_{\tau}=\frac{4}{5}\frac{G_{eff} (G_{eff}+3G_r)}{\Delta^2}U,
$$$$
u_w=\frac{8 G_r U W \hat T}{5\Delta^2},\;\;\;\;\;\;\;
u_0= - \frac{4}{5}\frac{\sin\theta}{C}
\frac{G_{eff} ( - G_{eff} + 2 G_r) }{\Delta^2}U\hat T,
$$$$
e_u=U^2 + 4 \frac{ G_r (G_{eff} + 2 U^2) }{\Delta^2}\hat T,\;\;\;\;\;\;\;
e_{\tau}=\frac{2}{5}\;\,\frac{ 7 G_{eff}^2 + 36 G_{eff} G_r - 3 G_r^2 }{\Delta^2} U,
$$$$
e_w= 8 \frac{G_r}{\Delta^2} U W \hat T,\;\;\;\;\;\;\;
e_0= - 4\frac{\sin\theta}{C}\frac{G_{eff} G_r}{\Delta^2} U\hat T,
$$$$
d=u_u e_{\tau}-u_{\tau}e_u, \eqno (3.15)
$$$$
u_{00}=-u_0-u_w\omega-F_{\theta}-F_{cf},\;\;\;\;\;\;
e_{00}=-e_0-e_w\omega+F_w-F_{cf}-\dot E_{rad}.
$$
Here, $u_u, u_w, u_{\tau}, e_u, e_w,$ and $e_{\tau}$
are the coefficients at the derivatives
$U', W',$ and $\hat T'$ in Eqs. (3.5) $(u)$
and (3.7) $(e),$
respectively.
We represented Eqs. (3.5) and (3.7) as
$u_u U' + u_{\tau} \hat T' = u_{00}$
and
$e_u U' + e_{\tau} \hat T' = e_{00}$
and then solved this linear system for $U'$ and $T'.$
\subsection{Phase Space and its Section by the 'Levitation' Surface
$(\Delta = 0)$}
The four-dimensional
$(\theta, U, W, \hat T)$
phase space of the system
(3.12)-(3.14) is filled with the integral curves
$U(\theta), W(\theta), \hat T(\theta).$
Each "trajectory" (along $\theta)$ of this stream of integral curves
represents a steady flow of the accreting plasma
over the stellar surface $S$ (Figs. 2-4)
which locally satisfies
the differential balances of mass, momentum, angular momentum, and energy
(2.19), (2.20), (2.22), and (2.23).
It is necessary to study and classify all these flows.
An analysis shows
that the numerical integration is stable only in the direction of increasing $\theta.$
The initial data
$\theta_0$ and $U_0=U(\theta_0), W_0=W(\theta_0), \hat T_0=\hat T(\theta_0)$
at the interface between the disk and the spread layer are therefore required.
There are many initial points in the four-dimensional set
$\theta_0, U_0, W_0,$ and $\hat T_0.$
Here, $\theta_0$ is the angular width of the disk
in the zone of its contact with the layer;
$ U_0$ and $W_0$ are the initial spread and rotation velocities, respectively;
and $\hat T_0$ is the initial temperature.
The first narrowing of the exhaustive search
for
$\theta_0, U_0, W_0,$ and $\hat T_0$
is ensured by the fact that $\theta_0$
turns out to be insignificant if
$\theta_0\ll\theta_{\star}.$
The second narrowing is associated with the separation
of
$W_0$ from the set of initial data.
At the disk-layer interface, $W$ is close to the Keplerian velocity.
Accordingly, we set $W_0 = 1 - \delta$
and perform three series of calculations
with $\delta = 10^{ - 1}, 10^{ - 2},$ and $10^{ - 3.}$
In each series, the initial points lie on the $U_0, T_0$
(spread velocity, temperature at the layer bottom)
plane.
A special place in the classification analysis is occupied by the
$\Delta (\theta, U, W, \hat T) = 0$
"levitation" surface of the accretion layer,
where $\Delta$ is a combination
of the gravitational and centrifugal forces and the radiation-pressure force
[see (2.8) and (3.8)].
Let us consider the intersection
of this surface
with $(U, \hat T)$ planes,
in particular,
with the $(U_0, \hat T_0)$ plane.
In each section, the $\Delta = 0$ curve is a hyperbola,
$$
U \hat T^4=\mathop{\rm const}\nolimits \eqno (3.16)
$$
(we assume that $\theta$ is small
and that $C = \cos \theta \approx 1);$ see (2.8) and (3.8).
There are the regions $\Delta<0$ and $\Delta>0$
above and under the hyperbola (3.16).
Clearly, only the region of initial data under the hyperbola is physically meaningful,
because in this case,
the total acceleration $g_{wr}$ is directed downward
and because the gravitation presses the layer against the stellar surface.
\subsection{Attracting or Limiting Surface and One-Parameter Structure}
Let us consider the "triangular" area $\Delta > 0$
that is bound by the $U_0$ and $\hat T_0$ axes
and by the hyperbola (3.16).
This area contains the boundary $(\Delta \ll 1)$
and distant $(\Delta \sim G_{eff})$ subregions.
These subregions differ greatly in area.
Accordingly, the trajectory of the common position
starts from the subregion $\Delta \sim G_{eff}.$
Consider Eqs. (3.5)-(3.7).
An important point is that $U_0$ and $\hat T_0$ are small.
This is the reason why
$|d\ln\hat T/d\theta|$$\gg $$1$
in the distant subregion
and
$|d\ln\hat T/d\theta|$$\sim $$1$
in the boundary subregion,
implying that the "hard" or steep segments of the trajectories
lie in the subregion
which is farther from the $\Delta = 0$ surface (deep in the "triangular" area)
and the "soft" or gentle segments lie in the boundary subregion.
Let us study the general trajectory
on which $\Delta \sim G_{eff}$ at the starting point.
In the immediate vicinity of the starting point
(i.e., on a steep segment),
we can assume for a rough estimate that the basic equation
is the energy equation (3.7).
$\Delta$ changes most rapidly during the displacement in $\theta,$
because it contains the fourth power of $\hat T.$
The remaining variables $(U, W, \hat T)$ change more slowly.
We disregard their change and factor them outside the differentiation.
Thus, only one unknown function $\hat T^4$ remains.
Assume that $C \approx 1$ and $G_r \simeq G_{eff}.$
In the right part of the energy equation,
we discard the small power $F_{cf}$ of the centrifugal force
compared to the power $F_{\varphi}$
which is associated with the latitudinal friction stress
$\tau_{\varphi}$
and radiative losses.
Taking
$\Delta\approx 1-W^2-B U \hat T^4$
as the unknown function
instead of $\hat T^4$ and rewriting the equation for this new unknown function,
we obtain
$$
\frac{\Delta'}{\Delta^2}= - \frac{\alpha W^3/\hat T+(A/B) U}{4 G_{eff} \hat T U}\Delta+
\frac{(A/B)}{4\hat T}. \eqno (3.17)
$$
We substitute the following typical values:
$ A/B \sim 10,$
$\, U \sim 10^{ - 5},$
$\, \hat T \sim 10^{-4},$
$\, W=1-\delta,$
$\, 10^{-3}<\delta <10^{-1},$
$\, \alpha_b\sim 10^{-3}.$
It follows from the estimate
that the first term (friction) dominates in the numerator of the fraction
before $\Delta$ in (3.17).
For $\Delta \sim G_{eff},$
the first term dominates in the right part of (3.17).
We see that $\Delta$ rapidly decreases on the steep segment.
This decrease causes $\Delta$ to drop to such a small value
that the first and second terms in the right part of (3.17) become equal.
\begin{figure}[tbh]
\epsfxsize=8cm
\epsfbox[-200 185 400 656]{r14.ps}
\caption{Qualitative structure of the field of integral curves.
We see that the calculations in the direction of increasing $\theta$ are stable
because of the attraction to the limiting surface $L.$
The surface $L$ is a separatrix -- it separates the colliding streams of integral curves.
The stream from the $\Delta = 0$ surface repulses the curves
which come from the "thickness"
(from the distant region; see the text)
to this surface.
Since the ratios $v_{\theta}/c$ and $\sqrt{2T_S/m_p}/c$ are small,
the relative gap
(for example, $\Delta \hat T/\hat T)$ between the $L$ and $\Delta = 0$ surfaces is small.}
\end{figure}
Thus, any trajectory consists of steep and gentle segments
(see Fig. 14).
The steep and gentle segments lie in the distant
and boundary subregions,
respectively.
On the steep segment,
the first and second of the three terms in Eq. (3.17) are important.
On the gentle segment, all three terms are comparable.
On the steep segment, $\Delta$ rapidly decreases with increasing $\theta.$
As one makes a transition from the steep segment to the gentle one,
the decrease in $\Delta$ becomes saturated.
The derivative $|d\ln\Delta/d\theta|$
is small on the gentle segment,
implying that the radiative losses are roughly offset by the frictional heating.
On the steep segment, the energy balance,
if by the energy balance we mean the equality of the right part of Eq. (3.17) to zero,
is not maintained --
the radiative losses are small compared to the heating.
This implies that the left part of Eq. (3.17) plays an important role here.
Note that the term in the left part corresponds to the advective transfer
of radiative enthalpy
[cf. Subsec. 1.3 (IV)].
The extent of the steep segment with large (in magnitude) derivatives
is small:
$\Delta\theta\ll\theta_{\star},$
$\; \Delta\theta = \theta_{tr}-\theta_0,$
where $\theta_{tr}$ refers to the transition
from the steep segment to the gentle one on a given trajectory (see Fig. 14).
When the complete system is calculated numerically,
the short steep segment is traversed in small steps
(a fraction of $\Delta \theta),$
while the extended gentle segment is traversed with large steps
(a fraction of $\theta_{\star}).$
At $\theta \approx \theta_{tr},$
the trajectory makes a sharp turn.
As can be seen,
the trajectories are attracted or caught in the boundary subregion,
which lies near\footnote{
Accordingly,
in this part of spread flows,
the effective or total pressing is small,
and the layer is thick.}
the $\Delta = 0$ surface
and do not leave this subregion.
For this reason,
the "levitation" surface is called the attracting or limiting surface,
hence the stability of the calculations
and the relative "crudeness" of the results
with respect to $U_0$ and $\hat T_0$
and, consequently, with respect to the disk parameters\footnote{
Thus, of all the disk parameters,
only the rate of mass inflow $\dot M$
is important for the flow in the spread layer.
The regulation (selection) within the one-parameter $(1d)$ family
(see below)
of the flow
that is actually realized
is associated with the conditions in the spreading flow
at the interface between the radiating and dark layers
rather than with the disk parameters.}.
Thus,
the steep segments along with the distant subregion turn out to be unimportant.
Only the boundary subregion remains.
Accordingly, we can immediately take the initial data near the curve (3.16).
In this way we further narrow down
(for the third time, see above)
the set of initial data
$ (\theta_0, U_0, W_0, \hat T_0).$
Let us introduce the polar (angle/radius) coordinates
in the $(U_0, \hat T_0)$ plane
instead of the $U_0$ and $\hat T_0$ coordinates.
Let the angle be related to the ratio
$$
i=\sqrt{\hat T_0}/U_0. \eqno (3.18)
$$
We use the deviation $\delta\Delta$ from the hyperbola (3.16)
instead of the radius.
One\footnote{
We obtain one value of $i$ instead of the pair of $U_0$ and $\hat T_0.$
The initial points lie in a narrow band near the hyperbola.}
parameter $i$ (3.18)
now runs the set of initial data,
because the deviation $\delta \Delta$ is insignificant.
A {\it certain trajectory}
(steady accretion flow)
corresponds to {\it each} value of $i.$
We thus arrive at a one-parameter family of solutions,
which we will call the $1d$ family.
In phase space,
the integral curves from this family cover the surface $L$
near the $\Delta = 0$ surface (see Fig. 14).
\subsection{Within the One-Parameter Family. The Dependence on Latitude.
Critical Regime}
Consider an arbitrary steady flow from the family of solutions
for a given $\dot M$
but with different $i.$
It represents the $\theta-$distribution of $U, W,$ and $\hat T.$
Let us study this flow.
There are two characteristic regions above and below $\theta_{\star}.$
The lower region is the radiating belt.
In this belt,
$ [W(\theta)]^2 $ is great;
the loss of rotational energy maintains the radiative flux $q(\theta)$
at such a level
that $\Delta \ll 1$
(or $q \approx q_{eff};$ see Sec. 1).
The rotational energy is exhausted inside the belt.
As a result, the luminosity $q(\theta)$ abruptly decreases
in the transition zone at $\theta \approx \theta_{\star}$ --
the accretion-flow surface darkens.
Concurrently, the function $\Delta (\theta)$ increases
$[ \Delta (\theta) \approx 1$
for $\theta > \theta_{\star} ],$
$ g_{wr}$ is large
$ (g_{wr} \approx g_0),$
and the layer under the strong press contracts
and becomes very thin outside the radiating belt
$ (h$ is a few centimeters).
The lower (in latitude) region is of greatest interest.
Since the radiation pressure in this region is large,
the speed of sound
$
c_s\simeq\sqrt{p/\rho}
$
is high compared to the meridional spread velocity
$v_{\theta}.$
For this reason,
the meridional distributions inside the radiating belt become hydrostatic.
In any solution from the family of solutions, the Mach number,
$$
\mathop{\rm Ma}\nolimits_{\theta}=\frac{v_{\theta}}{c_s}, \;\;\;\;\;\;\;
c_s^2 = \frac{c^2}{3}\,
\left(1 + \frac{3}{4}\,\frac{\rho_S\, c^2}{a T_S^4}\right)^{-1}
+ \frac{10}{3} \,\frac{T}{m_p},
\eqno (3.19)
$$
which was calculated for a radiation-dominated plasma (Weinberg 1972),
must increase with $\theta$
(the pressure decreases, and the flow accelerates).
Indeed, the quantity (2.11) decreases with increasing $\theta,$
because its gradient
counteracts the tangential component of the centrifugal force,
which tends to return the plasma to the equator
[see Eq. (2.20)].
Let us compare the $\theta$ dependences for various values of $i$ that run the family.
The Mach number $\mathop{\rm Ma}\nolimits_{\theta}$ (3.19) at the interface
between the disk and the spread layer
$ [ \mathop{\rm Ma}\nolimits_{\theta} (\theta = 0) ]$
is smaller,
if $i$ is larger
$[i$ is the reciprocal of the Mach number
without a factor
$\simeq \sqrt{ G_{eff}/\Delta };$
cf. the definitions (3.18) and (3.19)].
We take $\mathop{\rm Ma}\nolimits_{\theta} (\theta_0)$
at the interface between the disk and the spread layer
and assume that the disk is thin
$ (\theta_0 \ll \theta_{\star}).$
We can thus write
$ \mathop{\rm Ma}\nolimits_{\theta}(\theta = \theta_0) = \mathop{\rm Ma}\nolimits_{\theta}(\theta=0).$
At a fixed $i,$
the function $\mathop{\rm Ma}\nolimits_{\theta} (\theta)$
monotonically increases with $\theta$
in the interval $0 < \theta < \theta_{\star}.$
Consequently,
$ \mathop{\rm Ma}\nolimits_{\theta} (\theta) $
reaches a maximum in this interval at its edge.
An analysis of the effect of $i$
shows that
$\mathop{\rm Ma}\nolimits_{\theta} (\theta_{\star}),$
together with $\mathop{\rm Ma}\nolimits_{\theta} (0),$
increases with decreasing $i.$
On the $i$ axis,
the family fills the semiaxis $i_{\star} < i < \infty.$
The edge $i_{\star}$ corresponds to the critical spread regime,
in which the {\it sonic point} $\theta_c$
lies at the boundary of the radiating belt
$[\, \theta_c = \theta_{\star}, \; \mathop{\rm Ma}\nolimits_{\theta} (\theta_{\star}) \simeq 1\,].$
$\mathop{\rm Ma}\nolimits_{\theta} (\theta)$ also monotonically increases with $\theta$
for $i < i_{\star}.$
If $i < i_{\star},$
then the rotational energy at $\theta_c,$
at which $ \mathop{\rm Ma}\nolimits_{\theta} (\theta_c) \simeq 1,$
has not yet been exhausted:
$ v_{\varphi} (\theta_c) > 0, $
i.e., the latitude $\theta_c$
is inside the radiating belt.
At the sonic point $\theta_c,$
the solution becomes two-valued ('tips over').
Accordingly, we reject the solutions with $i < i_{\star}$ as nonphysical.
{\it The critical regime} occupies a special place.
It restricts the $1d$ family of subsonic
(for $\theta < \theta_{\star})$
flows.
In this regime,
the plasma is ejected with the speed of sound from the radiating belt.
Accordingly,
the region at $\theta > \theta_{\star}$
has no effect on the flow in the radiating belt.
At $i > i_{\star},$
the regions
$\theta < \theta_{\star}$
and $\theta > \theta_{\star}$ are coupled.
For example, the action of lateral forces
(and friction $\tau_{\theta})$
between elements $\delta \theta$
extends from the region $\theta > \theta_{\star}$
into the region $\theta < \theta_{\star}.$
\section{PHYSICS OF DECELERATION AND SPREAD}
\subsection{Numerical Simulations}
The system (3.12)-(3.14) was integrated numerically.
In doing so, we approximated the derivatives by finite differences.
We specified $i$
to calculate the initial point
and found the intersection point
with the coordinates $(U_{01}, \hat T_{01})$
of the curves
$i = i(U, \hat T)$
and $\Delta (U, \hat T) = 0.$
We took $U_0 = 0.9 U_{01}$
and $\hat T_0 = 0.9\hat T_{01}$ as the initial values.
The values of $\theta_0$ and $W_0$ were varied
$ (10^{ - 3} < \theta_0 < 10^{ - 2},$
$ 10^{ - 3} < 1-W_0 < 10^{ - 1}).$
The steep segment was traversed with the step
$ \sim 10^{-6} \Delta\theta.$
We took $\theta, U, W,$ and $\hat T$ at its end as the initial values
when we calculated the gentle segment with a step
$\sim 10^{-6} \theta_{\star}.$
The round-off error was $10^{ - 16}$ (double accuracy).
A preliminary estimate of $\theta_{\star}$
was obtained from formula (1.1)' for the Eddington scale.
The family consists of subsonic (subcritical)
and supersonic (supercritical) subfamilies
in the radiating belt.
In the solutions from the supercritical subfamily,
the sonic point lies inside the radiating belt
(see Subsec. 3.6).
As was stated, these solutions are discarded.
When we say below that the flow is subsonic,
we mean that it is subsonic in the radiating belt.
The determinant $d( \theta )$ (3.15)
was calculated along an integral curve.
The first zero of the function $d(\theta),$
when running in $\theta$ from $\theta_0,$
is called the sonic point $\theta_c.$
At this point, the integral curve makes a turn,
and the solution becomes two-valued.
Since the determinant $d$ appears in the denominator
of the derivatives (3.12) and (3.14),
the derivatives $U'$ and $\hat T'$ have a singularity at $\theta_c.$
Thus, the solution exists in the interval
$\theta_0 < \theta < \theta_c.$
\subsection{One-Parameter Family as a Whole
and the Passage Through the Critical Point}
As was shown in Subsecs. 3.4-3.6,
the integral curves with general initial data
form a one-parameter (or $1d)$ family.
An example of a family is given in Fig. 15
$(\mathop{\rm Ma}\nolimits_{\theta}$ profiles, transcritical and supercritical solutions, $100<i<160)$
and Fig. 16
$(\Sigma_S$ profiles in the same family, subcritical and supercritical solutions,
$100 < i < 500).$
At a fixed $\theta,$ $\;\mathop{\rm Ma}\nolimits_{\theta}$ increases,
while $\Sigma_S$ decreases with decreasing $i.$
In Fig. 15, the upper and lower solutions refer to $i = 100$ and $160,$
respectively.
In Fig. 16,
the upper and lower solutions refer to $i = 500$ and $100,$
respectively.
The supercritical solutions terminate at $\theta_c < \theta_{\star}.$
The subcritical solutions extend into the region $\theta > \theta_{\star},$
i.e., their termination point $\theta_c$
lies to the right of $\theta_{\star}.$
In Figs. 15 and 16, the two upper (Fig. 15) and two lower (Fig. 16) solutions
are supercritical.
The arrow in Figs. 15 and 16 marks the critical solution
with $i = i_{\star}$ and $\theta_c = \theta_{\star}.$
In Fig. 15,
the subcritical solutions with $i > i_{\star}$ run below (above in Fig. 16)
this solution.
The solutions
for which $i$ lies in the vicinity of the critical value $i_{\star}$
$\; ( |i - i_{\star}| \ll i_{\star} ).$
are called transcritical.
\begin{figure}
\epsfxsize=16cm
\epsffile{r15.ps}
\caption{One-parameter family of steady spread flows $(100<i<160).$
The dependences $\log [\mathop{\rm Ma}\nolimits_{\theta} (R\theta)]$
are for
$\dot M=4\cdot 10^{17}$ g s$^{ - 1},$
$1-W_0=10^{-2},$
$\; \theta_0=10^{-3},$
and
$\alpha_b = 10^{-3}.$
$R \theta$ is the arc length along the meridian in km,
$R$ is the stellar radius.
Here $R = 10$ km and $M=1.4 M_{ \odot }.$
The calculations were performed using Newton's approximation.
Recall that the curves terminate at the sonic point
$\theta_c,$
at which the determinant is $d(\theta_c) = 0$
[see Eq. (3.15) and Subsec. 4.1],
or are bounded by the computational interval.}
\end{figure}
\begin{figure}
\epsfxsize=16cm
\epsffile{r16.ps}
\caption{$\Sigma_S$ profiles for various values of $i.$
The four upper profiles rest on the end of the computational interval.
The remaining profiles terminate at the sonic point.
The parameters of the variant are given in the caption to Fig. 15.}
\end{figure}
Let us discuss the transcritical behavior.
The function $\theta_c(i)$ increases
with increasing $i$ (see Fig. 17 and Figs. 15 and 16).
This function appears to have a discontinuity at $i_{\star}.$
In our example,
$ i_{\star} \approx 151.404.$
At least in our example at the accuracy we chose $(1.5-$mm step),
the discontinuity cannot be reduced by refining $i$
$\; (i_- = 151.403, \;i_+ = 151.405).$
The discontinuity boundaries in $R \theta$ are:
$R \theta_- = 4.1365$ and $R \theta_+ = 5.8448$ km;
the relative width is $\Delta \hat \theta =$
$ 2 ( \theta_+ - \theta_- )/( \theta_+ + \theta_- )=0.34.$
\begin{figure}[tbh]
\epsfxsize=8cm
\epsfbox[-200 185 400 656]{r17.ps}
\caption{Passage of the sonic point $\theta_c$
through latitude
$\theta_{\star}$ as $i$ is varied (3.18).
The latitude $\theta_c$ is given in degrees.
The parameters of the variant are given in the caption to Fig. 15.}
\end{figure}
The latitude $\theta_{\star}$ divides the flow into two characteristic regions
(Sec. 1 and Subsec. 3.6).
It separates the rotating (separation in $W),$
hot (separation in $\hat T),$
levitating (separation in $\Delta$ and $h),$
and radiating (separation in $q)$ region
from the nonrotating, cold, pressed and dark region.
The separation into these regions is clearly seen
on the subsonic profiles in Figs. 15 and 16
(see the arrows $\theta_{\star}).$
The transition between them is sharp.
We can also determine $\theta_{\star}$
from the left boundary of the discontinuity in Figs. 15 and 16
(see also Fig. 17).
It is marked by the arrow in Figs. 15-17.
The discontinuity in $\theta_c(i)$
is produced by the passage of the sonic point through the belt boundary
$\theta_{\star}.$
In the momentum equation (3.5),
the momentum flux in the subsonic flow is small
compared to the pressure gradient.
The gradient is balanced either by the component of the centrifugal force $F_{cf}$
or by the meridional friction $F_{\theta}.$
Calculations show that the meridional friction\footnote{
Of course, the latitudinal friction $F_{\varphi}$
in the equations of angular momentum (3.3), (3.6) and energy (3.4), (3.7)
cannot be ignored in the rotating belt.}
$ F_{\theta}$
and the force $F_{cf}$
can be disregarded inside and outside the radiating and rotating belt,
respectively.
This is yet another characteristic
in which there is a clear separation at $\theta_{\star}.$
{\bf Dark part of the layer. The sonic point and its position.}
In the dark region, the system (3.5)-(3.7) simplifies greatly.
Here, we may assume that $\Delta = G_{eff} = 1$
and ignore the rotation of matter.
The flow as a whole
(the radiating and dark layers)
is numerically analyzed below.
The numerical analysis must be preceded by a qualitative study.
Since the variability of $\cos\theta$ in the equations
is of little importance for a qualitative analysis,
we set $\cos\theta = 1.$
The qualitative analysis leads us to conclude
that the flow terminates at the sonic point,
as confirmed below by the numerical calculations
which are free from the assumed simplifications.
This conclusion is important in constructing the ensuing theory
of slow spread and settling of matter over a neutron-star surface,
because in this way the radiative-frictional mechanism of spread deceleration
is revealed.
As a result of the simplifications above,
we obtain
$$
(5U^2-4\hat T)U' + 4 U\,\hat T' = -5\alpha_b U^3/\hat T, \eqno (4.1)
$$$$
5 U U' + 14\hat T'= - 5 A\, U\,\hat T^4. \eqno (4.2)
$$
Resolving the system (4.1), (4.2) for the derivatives,
we obtain
$$
U' = 10 U^2 (2 A \hat T^5-7\alpha_b U)/\hat T d_1, \eqno (4.3)
$$$$
\hat T' = 5 U [\, (4\hat T-5 U^2)A\hat T^5 + 5\alpha_b U^3]/\hat T d_1, \eqno (4.4)
$$$$
d_1=50 U^2 - 56\hat T. \eqno (4.5)
$$
Eliminating $\theta$ from the system (4.3), (4.4),
we arrive at the equation
$$
\frac{d\hat T}{dU} =
\frac{(4\hat T-5U^2)(A/\alpha_b)\hat T^5 + 5 U^3 }{
2 U\, [\,2(A/\alpha_b)\hat T^5 - 7 U\,]} \eqno (4.6)
$$
in the phase plane,
which allows an effective qualitative analysis,
because it is two-dimensional.
Let us estimate the ratio of the frictional
$(\alpha_b U)$
and radiative
$(A\hat T^5)$ terms in Eq. (4.6).
For
$\dot M_{17}\sim 1,$
$\; \alpha_b\sim 10^{-3},$
$\; T\simeq 0.5$ keV,
$ U\sim \sqrt{\hat T}$
they are comparable.
We offset the large $(\sim 10^{23})$ constant $A/\alpha_b$
by changing the scale
$$
U\to \sqrt{\lambda} \hat U,\;\;\;\;\;\;\;
\hat T\to\lambda\tilde T,\;\;\;\;\;\;\;
\lambda=(\alpha_b/A)^{2/9}=
2.7\cdot 10^{-5} \,\alpha_b^{2/9} \,\dot M_{17}^{4/9},
$$
[see (3.9) and (3.11)].
Note that the velocity and temperature at $\hat U = \tilde T = 1$
are
$v_{\theta}=330 \,\alpha_3^{1/9} \,\dot M_{17}^{2/9}$ km s$^{ - 1}$
and
$\,T=0.57\,\alpha_3^{2/9} \, \dot M_{17}^{4/9}$ keV,
where we set $\alpha_b = 10^{ - 3} \alpha_3.$
Replace $\hat U, \tilde T$ by $\mathop{\rm Ma}\nolimits_{\theta}, \tilde T.$
In the dark flow,
we have
$ \mathop{\rm Ma}\nolimits_{\theta}^2 = (3/5) \hat U^2/\tilde T$
[see (3.19)].
Making the replacements
$X=\sqrt{10/9} \mathop{\rm Ma}\nolimits_{\theta}$
and
$Y= \tilde T^{9/2},$
we can smooth out the fourth power of temperature from Stefan-Boltzmann's law,
which complicates the analysis.
In these variables,
Eq. (4.6) takes the form
$$
\frac{dY}{dX}=
9\,\frac{Y}{X}\;
\frac{ \sqrt{2/3} (8-15X^2)Y + 15 X^3 }
{ \sqrt{2/3} (8+15X^2)Y-56X-15X^3}. \eqno (4.7)
$$
A phase portrait of Eq. (4.7) is shown in Fig. 18.
The function
$[X]_c (i) = X [\theta_c (i)] = X(i)$
gives the Mach number at the end of the integral curve (at the sonic point).
It is constant
$[X (i)\equiv \sqrt{56/75}]$
at
$i > i_{\star},$
i.e., when the sonic point lies in the dark region.
In the radiating region,
in which $i < i_{\star}$
and $\theta_c < \theta_{\star},$
this function slightly changes with $i.$
For $100 < i < i_{\star},$
$\;X(i)$ is smaller than $\sqrt{56/75}$ by $5\%$
and abruptly increases to this value at $i_{\star}.$
The vertical line 4
$\; (X = \sqrt{56/75})$
in Fig. 18 is a locus of sonic points $\theta_c.$
On this line, the determinants $d$ (3.15)
and $d_1 = \lambda \, \tilde T (75X^2 - 56)\,$ (4.5)
vanish.
It is easy to show
that the vertical line 4 and the isoclines of zeros (curve 2)
and infinities (curve 1) intersect at a single point.
\begin{figure}[tbh]
\epsfxsize=12cm
\epsfbox[-100 207 500 633]{r18.ps}
\caption{Phase plane of Eq. (4.7):
1 -- the isocline of infinities,
the arrows indicate the orientation of integral curves on the isoclines;
2 -- the isocline of zeros;
3 -- its asymptotic limit for $Y \to \infty,$
4 -- the sonic point
$(d_1 = 0,$ $\;\mathop{\rm Ma}\nolimits_{\theta} = \sqrt{56/75}).$
The integral curves 5
were taken from the calculations of the complete system
which are shown in Figs. 15-17
(the subsonic subfamily,
$i > i_{\star}).$}
\end{figure}
It follows from the qualitative analysis of Eq. (4.7)
that the integral curves
which start from the subsonic region
must necessarily intersect the vertical line 4.
Indeed, the segment of the integral curve at
$\theta>\theta_{\star}$
starts from the subsonic region.
The derivative $dY/dX$ (4.7) is positive
above the isocline of infinities at
$X<\sqrt{8/15},$
while the derivative $dY/d\theta$ is negative [see Eq. (4.4)].
Hence, on the integral curve in the $X, Y$ plane,
the motion occurs in the direction of decreasing $Y$ and $X$
as $\theta$ increases.
The $Y$ semiaxis cannot be intersected.
The integral curve 5
must then necessarily intersect the isocline of infinities
(see Fig. 18).
Below the isocline of infinities, $dY/dX$ is negative.
This time the $X$ semiaxis cannot be intersected.
The isocline of zeros 2 lies behind the "triangle"
that is formed by the segments of the $X$ semiaxis,
the vertical line 4,
and the isocline of infinities 1.
Therefore,
the stream of integral curves 5 arrives at the vertical line 4.
The curves crowd together near the limiting curve that corresponds to $i_{\star}.$
The initial value of $X$ (or $\mathop{\rm Ma}\nolimits_{\theta})$
decreases with increasing $i.$
Thus,
we established the causes of the "tipping"
and determined the position
of the sonic point in the dark region by analyzing the simplified system
(4.1) and (4.2).
\subsection{Preference of the Critical Regime. The Pedestal Under
the Radiating Belt}
In Figs. 1-4
we assumed that the surface $S$ is equipotential ("horizontal").
The equipotential surface of a nonrotating star is spherical.
Let us find out whether the underlying surface can be spherical.
Calculations show that the surface density $\Sigma_S (\theta)$ of the plasma
which constitutes the spreading layer is higher in the dark region (Fig. 10).
In addition, the acceleration
$g_{eff} = [\,1 - U^2 (\theta) - W^2 (\theta)\,]\, g_0$
in a differentially rotating radiating belt is smaller
than that in the dark region.
Consequently, the weight
$g_{eff} (\theta)\,\Sigma_S (\theta)$
of the spreading layer per unit area is larger in the dark region;
the pressure of the spreading layer on the underlying layers is also higher.
We ignore the motion in the underlying layer
beneath the surface $S.$
Since the motion in this layer is strongly subsonic,
the hydrostatic contribution to the pressure is more important
than the dynamical contribution.
Because of the difference in pressure on the surface $S$
from the spreading layer under the radiating region,
this surface rises to form a pedestal.
Let
$h_{US}(\theta) = R_S(\theta)-R_S(90^0),$
where
$R_S$ is the radius of this surface.
We have
$$
(h_{US})_{hs}=\frac{p_d-p_b}{\rho_{US} \; g_0}\simeq
\frac{p_d/\rho_{US}}{g_0}\simeq \frac{(c_s)_{US}^2}{g_0}
\simeq 17\,T_{US},
$$
because there is no rotation under $S$
and because the total pressing force $g_0$ is in action.
Here, $p_d$ and $p_b$ are the pressures on the surface $S$
under the dark and radiating regions, respectively;
$\rho_{US}$ is the density of the matter under $S;$
$ (c_s)_{US}$ and $T_{US}$ are the speed of sound and the temperature
under $S;$
and $h_{US}$ and $T_{US}$ are given in cm and keV, respectively.
In addition, the radiating belt is considerably hotter
with
$T_{US}(\theta) = T_S(\theta)$ (see Subsec. 2.1).
As noted above,
the radiating belt heated up the underlying layers,
causing the pedestal height to increase by $(h_{US})_T \simeq (c_s)^2_S/g_0.$
Let us compare the height
$(h_{US})_{hs} + (h_{US})_T$
with the thickness
$$
h_{eff}(\theta)=\frac{1}{| (d\ln \rho/dy)_S \,|}=
\frac{4}{3}\;\frac{R\;\hat T}{\Delta}
$$
of the spreading layer,
(see Fig. 5).
We see that the thickness of the spreading flow in the radiating belt is much larger
than the pedestal height.
This is because $g_{wr}$ and $g_0$ differ.
Consequently, the relief of the surface $S$ is not important
for the flow in the radiating belt.
At the same time,
the height $h_{US}(0)$ is of the order of the thickness
of the spreading dark layer,
implying that the dark flow cannot affect the flow in the radiating belt.
The flow in the belt must therefore be transcritical.
\subsection{Spread of Dark Matter}
The dark flow is specified by the functions $v_{\theta}$ and $T_S.$
Its segment
$\theta_{\star}<\theta<\theta_c$
terminates at the sonic point
(see Subsec. 4.2).
The separation
$\theta_c - \theta_{\star}$
increases with a decreasing initial Mach number,
for example,
$\mathop{\rm Ma}\nolimits_{\theta} (\theta_{\star}).$
In order to reach the pole
$(\theta_c>90^0),$
it is necessary that
$\mathop{\rm Ma}\nolimits_{\theta} (\theta_{\star})$
be sufficiently small.
In the model (3.2)-(3.4),
the surface $S$ is impenetrable
[integral (3.1)].
Above this surface,
the newly accreted matter is transported
from the source at the equator to the sink.
Let us consider the sink.
The dark layer is thin compared to the stellar radius,
$h \ll R.$
It may
mix with the underlying fluid
when traversing a long path,
$(s_{mix}-s_{\star})/h \gg 1;$
the boundary $S$ terminates in the mixing zone.
In another case, the mixing is ineffective,
and the dark flow runs up to the pole.
In this case,
we have a "bath-filling" flow with a filling "tap"
near the boundary $\theta_{\star}.$
The mixing zone then is located not far from the tap:
$s_{mix}-s_{\star}$
is of the order of $(10-100) h.$
In the "bath",
the matter is essentially in hydrostatic equilibrium
and, naturally, moves at velocities much lower than the speed of sound.
This matter flows under the radiating belt
(especially since the pressure under the radiating layer
is lower than the pressure under the dark layer; see Subsec. 4.3).
The boundary $S$ is preserved under this belt.
The "bath" filling is very slow,
and the flow is strongly subsonic.
Accordingly, the weight contribution $(h_{US})_{hs}$ to the pedestal vanishes,
and only the temperature contribution $(h_{US})_T$ remains.
Note the possibility of subsonic circulation under $S$
not only due to the viscous entrainment
(gripping) by the high-velocity layer
(see Subsecs. 1 and 2.1)
but also
due to the difference in the heights of equal pressure.
In the Earth's atmosphere,
this leads to trade-wind circulation
which is attributable to solar heating of the equatorial zone
(Palmen and Newton 1969; Bubnov and Golitsyn 1995).
The filling causes the fluid to rise at the velocity
$v_a$
or
to settle at the same velocity.
This is the radial component of the total velocity.
It is easy to see that
$v_a\simeq (1/2) (h/R) (\rho_S/\rho_{US})\, v_{\theta}$
is small
compared to the spread velocity $v_{\theta}$ in the radiating belt --
the matter traverses the belt rapidly but settles slowly.
At the parameters of the problem under consideration,
$ v_a \sim 10$ m s$^{ - 1}.$
Such velocity ensures a uniform increase in the surface density
over the entire stellar surface.
A particular regime of the motion of the dark fluid
affects the dynamics of a loose belt only slightly.
Accordingly, the approach (3.1)-(3.4) is acceptable
for describing the belt all the way to the pole.
Of course,
flows which run up to the pole require a special analysis.
It is necessary to introduce the parameter $v_a,$
change the boundary condition downstream,
and add the equation for $\Sigma_S$ to the system (3.2)-(3.4)
instead of the integral (3.1).
Note also
that the allowance for the moderate penetrability of the surface $S$
under the radiating and rotating belt
results in certain quantitative changes,
although qualitative conclusions remain valid.
This can be seen from the results (see below)
on the effect of variations in the accretion rate $\dot M$ on the belt structure.
\subsection{Effect of Friction}
Since the intensity of radiation cooling cannot exceed the Eddington limit,
the area of the emitting surface must be large enough
to remove the released heat [see (1.1), (1.1)', (1.2), (1.2)'].
The system under study has excess energy,
if the emitting area is larger than the area of the disk base.
The disk accretion in the disk-layer transition zone
then gives way to the spread of matter over the stellar surface.
The flow boundary
( in the plane
that passes through the polar axis)
turns $90^0$ in the transition zone
(see Figs. 1-4 and Sec. 1).
In this case,
the radial flow
(disk, $|v_r| \gg v_{\theta})$
transforms to the meridional flow
(spread, $|v_r| \ll v_{\theta}).$
Because of the delay in the deceleration,
a reservoir filled with fluid
which retains its rotation
carried from the disk
is formed.
Its surface emits radiation (radiating belt).
To assess the role of the friction model,
we varied the coefficient $\alpha_b$ in (2.16).
The cases with $\alpha_b = 10^{ - 3}$ and $10^{ - 3}/3$
are compared in Fig. 19.
The transcritical profiles,
which refer to the discontinuity boundaries
$i_-$ and $i_+,$
are shown in this figure
(see Subsec. 4.2).
Let us compare the transcritical profiles for these values of $\alpha_b.$
The critical value of $i_{\star}$
increases roughly proportional to $1/\alpha_b$
(by a factor of 2.7).
The functions
$\mathop{\rm Ma}\nolimits_{\theta} (\theta)$
and
$ U(\theta)$
decrease,
while
$\Sigma_S(\theta)$ increases
$\propto\alpha_b$
and
$\propto 1/\alpha_b,$
respectively.
The value of $\theta_{\star}$ increases by $6\%.$
The functions
$W(\theta)$
are virtually equal,
differing by only fractions of a percent.
Consequently, the $q(\theta)$ profile [see (1.2)] is invariant to $\alpha_b.$
The bottom temperature $T_S$ rises
$\propto (1/\alpha_b)^{1/4}$
due to the increase in $\Sigma_S$ [see (2.14)].
\begin{figure}[tbh]
\epsfxsize=10cm
\epsfbox[-180 128 403 750]{r19.ps}
\caption{Response of the function $q(\theta)$
to variations in $i$ and $\alpha_b.$
Curves 1, 2, and 3 refer to moderately supersonic
(1), slightly subsonic (2), and moderately subsonic (3)
spread regimes for $\alpha_b = 10^{ - 3.}$
Curve 4 gives a subsonic transcritical profile
for $\alpha_b = 10^{ - 3}/3.$
We see that variations in $\alpha_b$ affect the $q$ profile only slightly.}
\end{figure}
The degree of levitation is
$\Delta \simeq 10^{ - 3}-10^{ - 2},$
since it responds to a variation in $\alpha_b$
in the same way $T_S$ does.
The height $h_{eff}$
(see Subsec. 4.3)
is invariant to $\alpha_b$ variation
with the same accuracy as $W.$
As can be seen,
there are parameters
which change significantly as $\alpha_b$ is varied
(these include $i_{\star},$ $U,$ $\mathop{\rm Ma}\nolimits_{\theta},$ $\Sigma_S)$
and parameters
which change weakly $(T_S, \Delta, \theta_{\star})$
or only slightly
$(h_{eff}, W, q).$
Note that in view of the behavior noted above,
the tangential stress (2.16) belongs to the parameters
which respond only slightly,
because
$\rho_S=(4/3)\Sigma_S/h_{eff}\propto 1/\alpha_b.$
This also follows from Eqs. (2.22) and (3.3)
and from the invariance of $W(\theta).$
The width of the discontinuity
in $\theta_c$ at $i_{\star}$ $\;(\Delta \hat\theta,$ see Subsec. 4.2)
depends on $\dot M$ and $\alpha_b.$
\subsection{Effect of Initial Parameters}
It is clear from the arguments
which lead to formula (1.2)
that the equatorial part of the rotating belt radiates weakly,
because the normal component of the centrifugal force is large here.
Thus,
there are two bright belts
at latitudes $\theta_{\star}$ and $ - \theta_{\star}.$
The greatest brightness is reached near the edge of the radiating belt.
The maximum is appreciably lower than unity,
because
where $W \simeq 1,$
the subtraction of the centrifugal component is significant
and where $W$ is small,
there is nothing to radiate.
It follows from the smallness of $\Delta$
compared to $G_{eff}$
that $q \approx q_{eff}$
[see Sec. 1 for a discussion of formula (1.2)].
Let us analyze the effect of the initial parameters $W_0$ and $\theta_0,$
associated with the disk,
on the results.
We compare the cases with $1-W_0 = 10^{ - 2}$ and $10^{ - 1.}$
They correspond to a change
in the deviation of the rotation velocity from the Keplerian velocity
by an order of magnitude.
For this change in $W_0,$
the kinetic energy increases by $2\%,$
$i_{\star}$ increases by $42\%,$
the belt width $\theta_{\star}$ increases by $3\%,$
$\mathop{\rm Ma}\nolimits_{\theta}$
inside the belt decreases by $12\%,$
the surface density $\Sigma_S$ increases by $7\%,$
the rotation velocity $W$ increases by $1.4\%,$
the temperature rises by $0.8\%,$
the degree of levitation $\Delta$ decreases by $13\%,$
the height $h_{eff}$ increases by $14\%,$
and $q$ decreases by $3\%.$
Let us increase the meridional extent of the disk base by a factor of 10:
$\theta_0=10^{-3}$
is replaced by $10^{ - 2.}$
The critical value of $i_{\star}$
increases by $8\%.$
The distributions of
$\mathop{\rm Ma}\nolimits_{\theta}, \Sigma_S, W$
and other variables for disks of thickness
$2 R \theta_0 = 20$ and 200 m differ only slightly.
The distribution in $\theta$ seems to shift
by an amount of the order of the difference between the values of $\theta_0.$
Thus, for $1-W_0 \ll 1$ and $\theta_0 \ll \theta_{\star},$
variations in $W_0$ and $\theta_0$ have a marginal effect on the results,
in particular, on the $q (\theta)$ profile.
\subsection{Dependence on Accretion Rate}
Above, we discussed the main points
regarding variations in $\alpha_b,$
$W_0,$
and $\theta_0.$
Let us now study the effect of $\dot M.$
The critical value $i_{\star}(\dot M)$
and the belt width $\theta_{\star}(\dot M)$
are plotted versus the accretion rate in Fig. 20.
In the calculations,
we approached the critical value by compressing the transcritical vicinity
by the fork method (see Subsecs. 3.6, 4.2, 4.3).
The width
$\theta_{\star}(\dot M)$
was determined from the last calculated supercritical profile.
We took the position of the sonic point in this profile as the belt width.
An analysis of the effect of variations in $\dot M$
on the spread-flow structure
shows that the relative discontinuity width
$\Delta\hat \theta$ (Subsec. 4.2)
decreases with increasing $\dot M.$
At $\dot M \approx \dot M_{pole},$
the discontinuity disappears,
and the dependence on $i$ becomes smooth;
the edge of the radiating belt comes close to the pole.
Under these conditions, the edge of the rotating belt
was taken as
$\theta_{\star}(\dot M).$
We chose the smallest value of $i$
at which this edge lies at a distance from the pole
and took it as $i_{\star}.$
\begin{figure}[tbh]
\epsfxsize=9cm
\epsfbox[-180 128 403 750]{r20.ps}
\caption{Critical value of $i_{\star},$
radiating-belt width
$R \theta_{\star},$
and surface density $(\Sigma_S)_{\star}$
at the belt boundary versus $\dot M.$}
\end{figure}
In Fig. 20,
we also plotted $(\Sigma_S)_{\star}$ versus $\dot M.$
The value of $(\Sigma_S)_{\star}$
is
$\Sigma_S (\theta_{\star}),$
where $\theta_{\star}$ was determined from the last calculated supercritical profile.
Near the point of stoppage at the pole
for $\dot M \approx M_{pole},$
we took the edge of the rotating belt as
$\theta_{\star}.$
The function
$(\Sigma_S)_{\star} (\dot M)$
begins to rapidly increase near the pole.
It is physically clear
that after the point of stoppage of the loose radiating belt at the pole,
the spread of matter slows down greatly (Subsec. 4.4).
The slowdown causes the surface density $\Sigma_S$ to increase proportionally.
The power law fitting the dependences are
$i_{\star} \approx (\Sigma_S)_{\star}$ [g cm$^{ - 3}]$
$\approx 35 \dot M_{17}^{1.14},$
$ \; R \theta_{\star} = s_{\star}$ [km] $\approx 1.4 \dot M_{17}^{0.8}.$
Let us define the contrast
as the ratio of $\Sigma_S$ in the central zone
to
$(\Sigma_S)_{\star}.$
The contrast increases with increasing $\dot M.$
The fact that it is greater than unity
is attributable to the existence of a strongly subsonic central zone.
For this reason,
the spreading layer stretches when it approaches the edge of the radiating belt.
While at large $\dot M$ there are a subsonic center and a transonic edge,
at small $\dot M$
the entire radiating belt is transonic.
\subsection{Limiting Magnetic Field}
Let us estimate the magnetic-field strength
that still has no effect on the pattern of deceleration and spread outlined above.
Compare the magnetic pressure
$H^2/8\pi$
and the pressure $p_S = g_{eff}\,\Sigma_S$ (2.2)
at the bottom of the spread layer.
To obtain an estimate,
we assume that $g_{eff} \sim g_0.$
We take
$(\Sigma_S)_{\star}$
as a typical value of
$\Sigma_S$
and use the fitting dependence on $\dot M$
which we derived in Subsec. 4.7.
As a result,
we obtain $H < H_{max},$
$\; H_{max} = 4\times 10^8 (\dot M_{17})^{0.57}$ gauss.
A weaker field appears to "get buried" under the layers of accreting plasma.
X-ray bursters give an example of neutron stars with a weak magnetic field.
\section*{CONCLUSION}
The spread of matter
during disk accretion was considered.
We found a radiating belt to be formed in a certain range of $\dot M.$
The belt width
$\theta_{\star}$
depends on $\dot M.$
At
$L_{SL}/L_{edd}\simeq 3 \times 10^{-3},$
the width
$\theta_{\star}$
is of the order of the thin-disk thickness
$\theta_0=10^{-3}\div 10^{-2}.$
The belt disappears at
$L_{SL}/L_{edd} \approx L_{pole}/L_{edd} \approx 0.9,$
because the entire stellar surface radiates.
The belt rotates.
Its rotation velocity depends on latitude.
This belt levitates in the sense
that the pressing acceleration $g_{wr}$ is small
compared to the gravitational acceleration $g_0.$
A delay in the deceleration of rotation is responsible for the appearance of the belt.
This delay in the deceleration is attributable to a great decrease in the bottom density
due to the levitation and loosening of the matter.
\section*{ACKNOWLEDGMENTS}
We wish to thank G.S. Golitsyn and N.I. Shakura for helpful discussions.
We also wish to thank N.R. Sibgatullin, a referee of the paper,
for a careful reading of the manuscript,
for the checking of the equations,
and for valuable remarks.
\section*{REFERENCES}
\noindent Barret, D., Bouchet, L., Mandrou, P., et al., Astrophys. J., 1992, vol. 394, p. 615.
\noindent Bisnovatyi-Kogan, G.S., Mon. Not. R. Astron. Soc., 1994, vol. 269, p. 557.
\noindent Boubnov, B.M. and Golitsin, G.S.,
Convection in Rotating Fluids,
Fluid Mech. and its Appl., vol. 29,
Dordrecht: Kluwer, 1995.
\noindent Campana, S., Stella, L., Mereghetti, S., et al., astro-ph/9803303, 26 Mar, 1998a.
\noindent Campana, S., Colpi, M., Mereghetti, S., et al., astro-ph/9805079, 6 May, 1998b.
\noindent Chandrasekhar, S., Radiative Transfer, New York: Dover, 1960.
\noindent Chugaev, R.R., Gidravlika (Hydralics), Leningrad: Energoizdat, 1982.
\noindent Felten, J.E. and Rees, M.J., Astron. Astrophys., 1972, vol. 17, p. 226.
\noindent Gilfanov, M., Revnivtsev, M., Sunyaev, R., and Churazov, E., Astron. Astrophys., 1998,
vol. 338, p. L83.
\noindent Illarionov, A.F. and Sunyaev, R.A., Astrophys. Space Sci., 1972, vol. 19, p. 61.
\noindent Illarionov, A.F. and Sunyaev, R.A., Soviet Astron., 1975,
vol. 18(4), p. 413.
\noindent Landau, L.D. and Lifshitz, E.M., Gidrodinamika (Hydrodynamics), Moscow: Nauka, 1986.
\noindent Lapidus, I.I. and Sunyaev, R.A., Mon. Not. R. Astron. Soc., 1985, v. 217, p. 291.
\noindent Lin, C.-L., Moeng, C.-H., and Sullivan, P.P., Phys. Fluids, 1997, vol. 9(11), p. 3235.
\noindent Loitsyanskii, L.G., Mekhanika zhidkosti i gaza (Fluid and Gas Mechanics),
Moscow: Nauka, 1973.
\noindent Lynden-Bell, D. and Pringle, J.E., Mon. Not. R. Astron. Soc., 1974, vol. 168, p. 603.
\noindent Meyer, F. and Meyer-Hofmeister, E., Astron. Astrophys., 1989, vol. 221, p. 36.
\noindent Narayan, R. and Yi I., Astrophys. J., 1995, vol. 452, p. 710.
\noindent Palmen, E. and Newton, W., Atmospheric Circulation Systems,
N.Y. and London: Academic Press, 1969.
\noindent Papaloizou, J.C.B. and Stanley, G.Q.G., Mon. Not. R. Astron. Soc., 1986, vol. 220, p. 593.
\noindent Popham, R. and Narayan, R., Astrophys. J., 1995, vol. 442, p. 337.
\noindent Popham, R., Narayan, R., Hartmann, L., and Kenyon, S., Astrophys. J., 1993, vol. 415,
p. L127.
\noindent Pringle, J.E. and Savonije, G.J., Mon. Not. R. Astron. Soc., 1979, vol. 187, p. 777.
\noindent Schlichting, H., Grenzschicht--Theorie, Karlsruhe: G. Braun, 1965.
\noindent Shakura, N.I. and Sunyaev, R.A., Adv. Space Res., 1988, vol. 8, p. 135.
\noindent Shakura, N.I. and Sunyaev, R.A., Astron. Astrophys., 1973, vol. 24, p. 337.
\noindent Shakura, N.I., Soviet Astron, 1972, vol. 16(3), p. 532.
\noindent Sibgatullin, R.N. and Sunyaev, R.A., Astron. Lett., 1998,
vol. 24(6), p. 774. (Astro-ph/9811028).
\noindent Sobolev, V.V., Uchenye Zapiski Leningr. Univ., Ser. Matem. Nauk, 1949, vol. 18, issue 116, p. 3.
\noindent Spitzer, L., Physics of Fully Ionized Gases, N.Y.: Interscience,
1962.
\noindent Sunyaev, G.A. and Shakura, N.I., Soviet Astron. Lett., 1986,
vol. 12(2), p. 117.
\noindent Sunyaev, R.A. and Titarchuk, L.G., Astron. Astrophys., 1980, vol. 86, p. 121.
\noindent Tylenda, R., Acta Astron., 1981, vol. 31, p. 267.
\noindent Van der Klis, M., astro-ph/9812395, 22 Dec, 1998.
\noindent Weinberg, S., Gravitation and Cosmology: Principles and Applications of
the General Theory of Relativity, N.Y. etc.: John Wiley and Sons, 1972.
\end{document} |
\section{Introduction}
Quantum anomalies both in the Riemannian and in the
Riemann-Cartan spacetimes were calculated previously
using different methods,
see e.g. \cite{torsfree,Yajima}. However,
recently \cite{ChandiaZ}
the completeness of these earlier calculations have been questioned
which all demonstrated that the Nieh--Yan four-form \cite{NY} is irrelevant to the axial anomaly.
For the axial anomaly, we have a couple of distinguished
features. Most prominent is its relation with the Atiyah--Singer
index theorem. But also from the viewpoint of perturbative
{\em quantum} field theory (QFT), the chiral anomaly has
some features which signal its conceptual importance.
For all topological field theories like
BF-theories, Chern--Simons, and for all
topological effects like the anomaly,
the remarkable fact holds that the relevant
invariants do not renormalize --- higher
order loop corrections do not alter the one-loop value
of the anomaly, for example. The fact
that the anomaly is stable against radiative corrections
guarantees that it can be given a topological interpretation.
For the anomaly,
this is the Adler--Bardeen theorem, while other topological field
theories are carefully designed to have,
amongst other properties, vanishing beta functions.
Another feature is finiteness: in any approach, the chiral anomaly
as a topological invariant is a finite quantity.
In a spacetime with torsion,
Chandia and Zanelli \cite{ChandiaZ} argue that the Nieh--Yan (NY)
four-form $d\,^*\! A$ will add to this
quantity. As usual, they confront the fact that such a term,
if it is generated at all, is ill-defined, independent of the
regularization. In their case, they use a Fujikawa-type approach
and propose to absorb the regulator mass in a rescaled vierbein.
However, there is a severe misunderstanding in the Ref. \cite{ChandiaZ}. While there is no doubt that
the NY term can be possibly generated, as demonstrated previously
\cite{Yuri1,Yajima},
this is not the end of the argument. In order
to obtain a finite quantity, the tetrads have to be rescaled. While
this might look as an innocent manipulation, this is not so.
In rescaling the tetrad, the authors of Ref. \cite{ChandiaZ} ignore
the presence of renormalization conditions and the generation
of a scale upon renormalization. Rescaling the tetrad would
ultimately change the wave function renormalization $Z$-factor.
This factor creeps into the definition of the NY term at the
quantum level, and thus a rescaling of the tetrad does not
achieve the desired goals. This is not to be surprised: QFT
demands a new $Z$-factor for the NY term,
in sharp contrast to proper topological invariants at the
quantum level, which remain unchanged under renormalization.
With no renormalization condition available for the NY term,
and other methods obtaining it as zero, we can only conclude that
the response function of the quantum field theory to a gauge
variation (this is the anomaly) delivers no NY term.
Or, saying it differently, its finite value is zero after
renormalization.
\section{Gravitational Chern--Simons and Pontrjagin terms}
In our notation, Clifford--algebra valued
exterior forms \cite{MMM96},
the constant Dirac matrices $\gamma_\alpha$ obeying
$\gamma_\alpha\,\gamma_\beta +\gamma_\beta\gamma_\alpha=2o_{\alpha\beta}$
are saturating the index of the orthonomal coframe one--form $\vartheta^\alpha$ and its
Hodge dual $\eta^\alpha:={}^\ast\vartheta^\alpha$ via
$\gamma:=\gamma_\alpha\vartheta^\alpha$ and
${}^\ast\gamma=\gamma^\alpha\eta_\alpha\,.$
In terms of the {\em connection}
$\Gamma := {i\over 4} \Gamma^{\alpha\beta}\,\sigma_{\alpha\beta}$, the
$SL(2,C)$--covariant
exterior derivative is given by $D=d+ \Gamma\wedge$,
where
${\sigma}_{\alpha\beta}= \frac{i}{2}
(\gamma_\alpha\gamma_\beta-\gamma_\beta\gamma_\alpha)$ are
the Lorentz generators entering also in the
Clifford-algebra valued two-form
$\sigma:={i\over 2}\gamma\wedge\gamma = {1\over 2}\,{\sigma}_{\alpha\beta}
\,\vartheta^\alpha\wedge\vartheta^\beta$.
Differentiation of these independent
variables leads to the Clifford
algebra--valued two--forms of {\em torsion}
$\Theta :=D\gamma =T^{\alpha}\gamma_{\alpha}$
and {\em curvature}
$\Omega := d\Gamma +\Gamma\wedge \Gamma =
{i\over 4}R^{\alpha\beta}\,\sigma_{\alpha\beta}$
of Riemann--Cartan (RC) geometry.
The {\em Chern--Simons} (CS) term \cite{PRs} for the Lorentz
connection $C_{\rm RR} :=$ $- Tr\, \big( {\Gamma}\wedge {\Omega}
- {1\over 3} {\Gamma}\wedge {\Gamma}\wedge {\Gamma}\big)$ and
its corresponding Pontrjagin term
$dC_{\rm RR} = - Tr\, \left( {\Omega}\wedge {\Omega}\right)=
{1\over 2}\,R^{\alpha\beta} \wedge R_{\alpha\beta}$ have the familiar form.
Since the coframe is the translational part of the
Cartan connection \cite{PRs}, there arises also the
{\em translational} CS term \cite{Mi92}
\begin{equation}
C_{\rm TT} := {1\over{8\ell^2}} Tr\, ( {\gamma} \wedge {\Theta} )=
\frac{1}{2}
\left( C_{\rm RR} -\hat C_{\rm RR}\right)
\label{CTT}
\end{equation}
which is related to the Nieh--Yan four--form \cite{NY}:
\begin{equation}
dC_{\rm TT}
={2\over{\ell^2}} \left(T^\alpha\wedge
T_\alpha+R_{\alpha\beta}\wedge\vartheta^\alpha\wedge\vartheta^\beta\right) \,.
\label{eq:NY}
\end{equation}
A fundamental length $\ell$ unavoidably occurs here for
dimensional reasons. This can be also motivated
by a de Sitter type \cite{GH} approach, in which the $sl(5,R)$--valued
connection $\hat\Gamma =\Gamma +(1/\ell)(\vartheta^\alpha L^4{}_\alpha +
\vartheta_\beta L^\beta{}_4{})$ is expanded into the dimensionless linear connection
$\Gamma$ plus the coframe $\vartheta^\alpha= e_i{}^\alpha\, dx^i$
with canonical dimension $[length]$.
The corresponding CS term $\hat C_{\rm RR}$
splits via
$\hat C_{\rm RR} =C_{\rm RR} -2 C_{\rm TT}$
into the linear one
and that of translations, see
the footnote 31 of Ref. \cite{PRs}. This relation has recently been
``recovered" by Chandia and Zanelli \cite{ChandiaZ}.
The one--form of {\em axial vector torsion}
\begin{equation}
A:={1\over 4}\,Tr\left(\check{\gamma}\rfloor
{^*\Theta}\right)= {1\over 4}\,^*Tr(\gamma\wedge\Theta) =
\,^*(\vartheta^\alpha\wedge T_\alpha) \label{axitor}
\end{equation}
is a {\em conformal invariant} under the combined transformation of
{\em classical} Weyl rescalings of the coframe, in contrast to
$\,^*\!A=-2\ell^2C_{\rm TT}$, cf. Eqs. (3.14.1,9) of Ref. \cite{PRs}.
\section{Dirac fields in Riemann--Cartan spacetime}
The Dirac Lagrangian is given by the manifestly {\em Hermitian}
four--form
\begin{eqnarray}
L_{\rm D}(\gamma,\psi,D\psi)&=&
{i\over 2}\left\{\overline{\psi}\,{^*\gamma}\wedge D\psi
+\overline{D\psi}\wedge{^*\gamma}\,\psi\right\}+{^* m}\,
\overline{\psi}\psi\nonumber\\
&=&L(\gamma,\psi,D^{\{\}}\psi)
-{1\over 4}\,A \wedge\overline{\psi}\gamma_5\,^* \gamma\psi\,,
\label{decldirac}
\end{eqnarray}
for which $\overline{\psi}:=\psi^\dagger\gamma_0$ is the Dirac adjoint
and $^* m=m\eta$ the mass term, cf. \cite{MMM96}.
The decomposed Lagrangian (\ref{decldirac}) leads to the
following form of the Dirac equation
\begin{equation}
i \,^*\gamma\wedge \stackrel{\smile}{D}\psi + \,^*m\psi =
i \,^*\gamma\wedge \left[D^{\{\}} +{i\over 4}\,m\gamma +
{i\over 4}\, A\gamma_5\right]\psi= 0
\label{tt}
\end{equation}
in terms of the Riemannian connection $\Gamma^{\{\}}$ with $D^{\{\}}\gamma=0$ and the {\em irreducible} piece
(\ref{axitor}) of the torsion.
Hence, in a RC spacetime a Dirac spinor does
only feel the {\em axial torsion} one--form $A$. This can also be seen from
the identity (3.6.13) of Ref. \cite{PRs} which specializes here
to the ``on shell" commutation relation
\begin{equation}
[\stackrel{\smile}{D}\, ,\stackrel{\smile}{D}]= \Omega^{\{\}} +
{i\over 4}\, \gamma_5 dA- {i\over 8}\, m^2\sigma\,. \label{comrel}
\end{equation}
In contrast to Ref. \cite{ChandiaZ}, Eq. (27), there arise in (\ref{comrel})
no tensor or vector pieces of the torsion,
because our operator $\stackrel{\smile}{D}$
in (\ref{tt}) is the only possible result from the Lagrangian
(\ref{decldirac}), which is {\em Hermitian} as required by QFT.
{}From the Dirac equation (\ref{tt}) and its adjoint one can readily
deduce the well--known ``classical axial anomaly"
$ dj_5 = d ({1\over 3} \overline{\psi}\sigma\wedge\gamma \psi)=
2miP =2m i\overline{\psi}\gamma_5\psi$
for {\em massive} Dirac fields also
in a RC spacetime. If we restore chiral symmetry in the limit
$m\rightarrow 0$, this leads to {\em classical conservation law} of the
axial current for massless Weyl spinors, or since $dj =0$,
equivalently, for the {\em chiral current}
$ j_\pm :={1\over 2}
\overline{\psi}(1 \pm\gamma_5)\,^*\gamma\psi =
\overline{\psi}_{\rm L,R}\,^*\gamma\psi_{\rm L,R}$.
The Einstein--Cartan--Dirac (ECD) theory of a gravitationally
coupled spin
$\frac{1}{2}$ fermion field provides a {\em dynamical} understanding
of the axial anomaly on a classical (i.e., not quantized) level.
{}From Einstein's equations
$-(1/2)\,
\eta_{{\alpha}{\beta}{\gamma}}\wedge R^{{\beta}{\gamma}}=
\ell^{2}\,\Sigma_{\alpha} $
and the purely algebraic {\em Cartan relation}
$-(1/2) \eta_{\alpha\beta\gamma}\wedge T^{\gamma}=\ell^{2}
\tau_{\alpha\beta} =
-(\ell^2/4)\,\eta_{\alpha\beta\gamma\delta}\,\overline{\Psi}
\gamma_5\gamma^{\delta}\Psi\eta^{\gamma} $
one finds \cite{MMM96,MieKr}
\begin{equation}
dj_5 \cong 4 dC_{\rm TT}
={2\over{\ell^2}} \left(T^\alpha\wedge
T_\alpha+R_{\alpha\beta}\wedge\vartheta^\alpha\wedge\vartheta^\beta\right)
\label{eq:classan}
\end{equation}
which establishes a link
to the NY four form \cite{NY}, but only for {\em massive} fields
\cite{MieKr}.
However, if we restore chiral invariance for the Dirac fields
in the limit $m\rightarrow 0$, we find
within the dynamical framework of ECD theory that the
NY four--form tends to zero ``on shell", i.e.
$dC_{\rm TT}\cong (1/4) dj_5 \rightarrow 0$.
This is consistent with the fact that a
Weyl spinor does not couple to torsion at all, because the remaining axial torsion
$A$ becomes a {\em lightlike} covector, i.e.
$A_\alpha A^\alpha\eta =A\wedge\,^* \!A \cong (\ell^4/4)\,^*j_5\wedge j_5 =0$.
Here we implicitly assume that the light-cone structure
of the axial covector $\,^*j_5$ is not spoiled by quantum corrections, i.e. that no
``Lorentz anomaly" occurs as in $n=4k +2$ dimensions \cite{Leut}.
\section{Chiral anomaly in QFT}
When quantum field theory (QFT) is involved, other boundary terms
may arise in the {\em chiral anomaly} due to the
non--conservation of the axial current, cf. \cite{zum,Hir}.
Now, to approach the anomaly in the context of spacetime with torsion,
we will proceed by switching off the curvature and concentrate
on the last term in the decomposed Dirac Lagrangian (\ref{decldirac}).
Then, this term can be
regarded as an {\em external} axial covector $A$ (in view of (\ref{comrel})
without Lorentz or ``internal" indices) coupled to
the axial current $j_5$ of the Dirac field in an {\em initially flat}
spacetime. By applying
the result (11--225) of Itzykson and Zuber \cite{IZ}, we find
that only the term
$dA\wedge dA$ arises in the
axial anomaly, but {\em not} the NY type term
$d\,^*A\sim dC_{\rm TT}$ as was
recently claimed \cite{ChandiaZ}.
After switching on the curved spacetime of Riemannian geometry, we finally
obtain for the axial anomaly
\begin{equation}
\langle dj_5\rangle= 2m \langle iP \rangle+ \frac{1}{24\pi^2}
\left[ Tr\!\left(\Omega^{\{\}}\wedge\Omega^{\{\}}\right) -
\frac{1}{4} dA\!\wedge\!dA\right].\label{anom} \end{equation}
Besides other perturbative methods as point-splitting, there is
the further option to use
{\em dimensional regularization}.
If one adopts the $\gamma_5$ scheme of Ref. \cite{DK},
one immediately concludes that only the result
(\ref{anom}) can appear.
The only effect of the $\gamma_5$ problem is the replacement
of the usual trace by a non-cyclic linear functional.
The anomaly appears as the sole effect of this non-cyclicity.
There is no
room for other sources of non-cyclicity apart from the very fermion
loops which produce the result (\ref{anom}).
The whole effect of non-cyclicity is to have an operator $\Delta$,
$\Delta^2=0$, and the anomaly is in the image modulo
the kernel of $\Delta$, which summarizes the fact that in this
$\gamma_5$ scheme no other anomalous contributions are possible
beside (\ref{anom}).
But at this stage we have not discussed the
possibility of a contorted spacetime which {\em cannot} be {\em adiabatically}
deformed to the torsion-free case. In such a case it has been argued
\cite{ChandiaZ} that the boundary
term $dC_{\rm TT}$ occurs,
multiplied by a factor $M^2$. This factor $M^2$ corresponds to a
regulator mass in a Fujikawa type approach.
For instance, in the heat kernel approach, the first nontrivial terms
\cite{Yuri1,Yajima}, which potentially
could contribute to the axial anomaly, read
\begin{eqnarray}
Tr(\gamma_5 K_2) &=& -d\,^*\! A\,, \qquad
\qquad {\cal K}=\,^*\!\stackrel{\smile}{D}\wedge^*\!\stackrel{\smile}{D}{}^*\!A \label{ano2} \\
Tr(\gamma_5 K_4) &=& {1\over 6}
\left[Tr\left(\Omega^{\{\}}\wedge \Omega^{\{\}}\right) -
{1\over 4} dA\wedge dA +d{\cal K}\right].\nonumber
\end{eqnarray}
However, there is an essential difference in the physical dimensionality
of the
terms $K_2$ and
$K_4$. Whereas in $n=4$ dimensions the Pontrjagin type term $K_4$ is
dimensionless, the term
$K_2\sim 2\ell^2 dC_{\rm TT}$ carries dimensions.
It can be consistently absorbed
in a counterterm, and thus {\em discarded} from the final result for the
anomaly.
This is also in agreement with the analysis in \cite{bell} where,
in the framework of string theory, the chiral anomaly
in the presence of torsion
had a smooth adiabatic limit to the case of vanishing torsion.
In contrast, in Ref. \cite{ChandiaZ} it is argued that such contributions
can be maintained by absorbing the divergent factor in a rescaled
coframe $\widetilde\vartheta^\alpha:= M\vartheta^\alpha$ and propose
to consider the Wigner--In\"on\"u contraction $M\rightarrow \infty$ in the de
Sitter gauge approach \cite{GH}, with
$M\ell$ fixed.
Apart from the fact that
this would change also the dimension of $\psi$, in
order to retain the physical dimension $[\hbar]$ of the Dirac action,
there are several points
which seem unsatisfactory in such an argument:
1. As the difference (\ref{CTT}) of two Pontrjagin
classes, the term $dC_{\rm TT}$ is a topological invariant after all.
Now, it is actually
{\em not} this term
which appears as the torsion-dependent extra contribution to the
anomaly, but more precisely $-d\,^*\!A= 2
\ell^2 dC_{\rm TT}$. Thus, measuring its proportion in units of the
topological invariant $dC_{\rm TT}$, we find that it vanishes when we consider
the proposed limit $M\to\infty$, keeping $M\ell$ constant.
2. Instead of rescaling the vierbein, it is consistent
to compensate the ill-defined term by a counterterm.
This implies
that consistently a renormalization condition can be imposed which
guarantees the anomaly to have the
value (\ref{anom}).
Even if one renders this extra term finite by a rescaling as in Ref.
\cite{ChandiaZ}, one has to confront the fact that a
(finite) renormalization condition can be imposed
which settles the anomaly at this value.
Further, if one were to keep this extra finite term, it would
be undetermined, and is thus not related to the anomaly at all.
Also, on-shell renormalization conditions adjust the
wave function renormalization of the fermion propagator to have
unit residuum at the physical mass. Any rescaling of the tetrad
cannot dispense for the fact that the NY term needs
renormalization by itself,
as it is proven by the very calculation of \cite{ChandiaZ}.
3. From a {\em renormalization group} point of view, it is the scaling of the
coupling which determines
the scaling of the anomaly (regarded as a Green's function), a property which is
desperately needed to maintain the validity, e.g., of
the proof of the Adler--Bardeen theorem.
Or, to put it otherwise, an anomaly is stable against radiative corrections for
the reason that such
corrections are compensated by a renormalization of the coupling.
While, on the other hand, the topological invariant of
Ref.\cite{ChandiaZ} has no such property,
its interpretation as an anomaly seems dubious to us.
The only consistent way out is to impose a renormalization
condition
which adjust the finite value of the NY term to be zero, which is
patently stable under radiative corrections.
4. Finally, it is well-known that usually the appearance of a chiral
anomaly is deeply connected with the presence of a {\em conformal anomaly}
\cite{Salam,Ellis,Kae}. This makes sense: usually,
conformal invariance is lost due to the (dynamical) generation of a scale.
But this is the very mechanism which destroys chiral invariance as well.
Thus, one would expect any argument to fail
trying to combine strict conformal
invariance with a chiral anomaly.
\section{Conclusions}
Our conclusion is that the NY term $dC_{\rm TT}$ does NOT
contribute to the {\em chiral anomaly} in $n=4$ dimensions, neither classically nor in quantum
field theory, in sharp contrast to Ref. \cite{ChandiaZ}.
We once more stress the interrelation between the scale and chiral invariance
\cite{Salam,Ellis,Kae}.
Since renormalization amounts to a continous scale deformation,
only the Riemannian part of the Pontrjagin term contributes
to the topology of the chiral anomaly.
The result of Chandia et al. is very different in spirit from a
typical anomaly, where, in pQFT, the relevant Green's function is
UV-finite and only implicitly depends on the scale via the
coupling. This contrasts the fact that the Chandia et al. term
would depend {\em explicitly} on that scale.
Since the $A$ is {\em not} a gauge field, one can also legitimately
absorb the contribution from the axial torsion covector $A$ into the redefined
{\em gauge--invariant} physical current
$\widehat j_5:=
j_5 +(1/96\pi^2) A\wedge dA + M^2 \,^*\!A$,
where the last term depends explicitly on the regulator mass $M$.
It may arise from the counterterm $M^2\, A\wedge^*\!A$ to
the Dirac Lagrangian (\ref{tt}).
Then, due to
\begin{eqnarray} \langle d\widehat j_5\rangle&=&
\langle dj_5 \rangle+(1/96\pi^2) dA\wedge dA + M^2 d\,^*\!A \\ \label{anomriem}
&=& 2m \langle iP \rangle+ (1/24\pi^2)
Tr\!\left(\Omega^{\{\}}\wedge \Omega^{\{\}}\right),\nonumber
\end{eqnarray}
only the {\em Riemannian} contribution remains for the axial anomaly of
this new {\em physical} current. A consistent way to avoid regularization
problems for $M\rightarrow \infty$ is to assume that the ``photon" $A$ is
is transverse, i.e. $ d\,^*\!A=0$, which is just the
vanishing of the NY term.
On would surmize that
in $n=2$ dimensional models only the term $d\,^* \!A$ survives in
the heat kernel expansion (\ref{ano2}), since it then has the correct
dimensions. However, it is well-known \cite{PRs} that in 2D the axial
torsion $A$ vanishes identically.
In general, the Pontrjagin topological invariant, as an integral over a {\em closed}
four-form, depends also upon the second fundamental form of
the embedding of the boundary $\partial M$ into $M$. In Ref.
\cite{PW99} this is generalized to spaces
with torsion supporting our view that the index shall be independent
of regulator masses, hence excluding contributions from
the NY term.
Since the integral $\int_{M} d\,^* \!A\equiv \int_{\partial M}\,^* \!A$
vanishes {\em identically} for
manifolds without boundary, the NY invariant would occur only
for non-closed manifolds, anyhow.
A situation where torsion
is indeed realized in a {\em discontinous} manner arises
for the cosmic string solution within
the EC theory \cite{SO,A94,TO}. We have shown in detail in Ref.
\cite{MieKr} that
the NY term (\ref{eq:NY}) {\em vanishes identically} for this
example of a spinning cosmic
string exhibiting a {\em torsion line defect}.
The instantons of Ref. \cite{ChandiaZ} with non-vanishing NY term are clearly
detached from an Einstein-type dynamics, and also the recent analysis in
Ref. \cite{ChopinSoo} fails to substantiate the presence of the NY term.
\acknowledgments
We would like to thank Alfredo Mac\'{\i}as, Friedrich W. Hehl, and
Yuri Obukhov
for useful comments and the referee
for pointing out Ref. \cite{PW99}.
This work was partially supported by CONACyT, grant No. 28339E, and the
joint German--Mexican project DLR--Conacyt
E130--2924 and MXI 009/98 INF.
One of us (D.K.) acknowledges support by a Heisenberg fellowship of the DFG
and thanks the I.H.E.S. (Bures-sur--Yvette) for hospitality. E.W.M. thanks Noelia
M\'endez C\'ordova for encouragement.
|
\section{Introduction}\label{intro}
\neu{intro-setup}
For an arbitrary polytope ${\Delta}$, Brion has introduced in \cite{Br} certain
invariants $h^{{p},{q}}({\Delta})$.
These are defined as dimensions of cohomology groups $H^{{p},{q}}$ of
complexes which are associated directly to the polytope ${\Delta}$. In case that ${\Delta}$ is a rational polytope, these invariants are exactly the Hodge numbers
$\dim H^{p}(X,\Omega_{X}^{q})$
of the corresponding projective toric variety $X=I\!\!P({\Delta})$.
\par
In this paper, we focus on the ${K}$-vector spaces ${D}^{k}({\Delta}):=H^{{k},1}$
(${k} \geq 2$) and ${D}^1({\Delta}):=H^{1,1}/{K}$. The notation is suggested by a
second interpretation of these vector spaces: In \S \ref{deform} we will show that
there is a close relation between ${D}^{k}({\Delta})$ and the vector
spaces $T^{k}$ describing the deformation theory of the toric Gorenstein
singularity $X_{\mbox{\rm\footnotesize cone}({\Delta})}$ associated to the lattice
polytope ${\Delta}$.\\
Our main result is a vanishing theorem for ${D}^2({\Delta})$ for a certain
class polytopes. An important special case is
\par
{\bf Theorem} (cf.\ \zitat{van}{pyramid})
{\em
Let ${\Delta}$ be an $n$-dimensional, compact, convex polytope such that every
three-dimensional face is a pyramid.
If no vertex is contained in more than $(n-3)$ two-dimensional,
non-triangular faces, then ${D}^2({\Delta})=0$.
}
\par
There is a natural class of polytopes that arises from {\em quivers} (see \cite{AH})
to which these seemingly strange conditions apply. Special examples of such quiver polytopes
appeared in \cite{BCKvS} as a description of toric degenerations of Grassmannians
and partial flag manifolds that appeared in the works of Strumfels \cite{S} and
Lakshmibai \cite{L}. In a forthcoming paper \cite{AvS}, we will apply the above vanishing result to show that the Gorenstein singularities provided by so-called flag like quivers are
{\em unobstructed} and {\em smoothable} in codimension three.
\par
\neu{intro-contents}
The paper is organized as follows:
\vspace{1ex}\\
In \S \ref{hodge} we recall some notions of homological and
cohomological systems on polyhedral complexes. For the special case
of simplicial sets, these can be found in \cite{GM} or \cite{EMS}. We quote the definition of the
polyhedral Hodge numbers and review their basic properties.
\vspace{1ex}\\
In \S \ref{Tinv}, we introduce the ${D}$-invariants from a slightly different point of view
as above and show their relation to the polyhedral Hodge numbers. We present some examples
as well as elementary
properties, such as the relation of ${D}^1({\Delta})$ to the Minkowski
decomposition of polytopes.
\vspace{1ex}\\
The paragraphs \S \ref{van} and \S \ref{proof} contain the vanishing theorem for ${D}^2$
and its proof. The result is obtained from a spectral sequence
relating the ${D}$-invariants of a polytope to those of its faces;
${D}^2({\Delta})$ is represented as the kernel of some differential on the
$E_2$-level. In Theorem \zitat{van}{cumbersome}, this description is transformed
into an explicit set of equations describing ${D}^2({\Delta})$.
\vspace{1ex}\\
The final \S \ref{deform} deals with the relations of
the ${D}$-invariants to deformation theory that was mentioned before.
In the paper \cite{AQ}, a combinatorial description of the cotangent cohomology modules $T^{k}(X_{\mbox{\rm\footnotesize cone}({\Delta})})$ was given. From this description it appears that the
$T^{{k}}$ are very sensitive to the interaction of the polytope ${\Delta}$ with the lattice structure of the ambient space. As a consequence, these invariants are often very difficult to calculate explicitly.\\
On the other hand, the invariants ${D}^{k}({\Delta})$ are rather coarse;
they only
depend on the polytope ${\Delta}$ {\em up to projective equivalence}, and the lattice structure is
not involved at all. Nevertheless, in Theorem \zitat{deform}{geq2} we formulate a
sufficient conditions on ${\Delta}$ ensuring that the $T^{k}$ {\em are} determined by
${D}^{k}({\Delta})$.
In particular, the vanishing Theorem \zitat{van}{pyramid} yields
a vanishing theorem for $T^2(X_{\mbox{\rm\footnotesize cone}({\Delta})})$ as well.
\par
\neu{intro-ack}
{\em Acknowledgement}: We would like to thank M.~Brion and P.~McMullen for
valuable comments and discussions.
\par
\section{Hodge numbers for polytopes}\label{hodge}
\neu{hodge-cohom}
Let ${\Sigma}=\cup_{{k}\geq0}{\Sigma}_{k}$ be a finite, {\em polyhedral complex} in a
${K}$-vector space ${V}$ (${I\!\!\!\!Q}\subseteq {K} \subseteq{I\!\!R}$),
i.e.\ a set of polyhedra in ${V}$ that is closed under the face operation
and with the additional property
that for any two ${\sigma},{\tau}\in{\Sigma}$, the intersection
${\sigma}\cap{\tau}$ is either empty or a common face of both polyhedra.
Here ${\Sigma}_{k}$ denotes the subset of ${k}$-dimensional elements of ${\Sigma}$.
Examples for such polyhedral complexes
are {\em simplicial sets} as well as {\em fans}.
\par
{\bf Definition:}
A {\em cohomological system} ${\CF}$ on ${\Sigma}$ is a covariant functor
from ${\Sigma}$ to the category ${\CA\!\CB}$ of abelian groups (or to any other
abelian category ${\cal A}$).
\par
Here ${\Sigma}$ becomes a small category by declaring
the face relations ``${\tau}\leq{\sigma}$'' to be the morphisms. So a cohomological
system is nothing else than a collection of abelian groups ${\CF}({\sigma})$
for ${\sigma}\in{\Sigma}$ together
with compatible face maps ${\CF}({\tau})\to {\CF}({\sigma})$.\\
Similarly, a {\em homological system} is defined as a contravariant functor
from ${\Sigma}$ to ${\CA\!\CB}$.
\par
We fix for each polyhedron ${\sigma}\in{\Sigma}$ an orientation. This enables us to
introduce for each pair $({\tau},{\sigma})$ of elements of ${\Sigma}$ a number
$\varepsilon({\tau},{\sigma})$ as follows:
\begin{itemize}\vspace{-2ex}
\item
If ${\tau}$ is a facet (i.e.\ a codimension-one face) of ${\sigma}$, then we may
compare the original orientation of ${\tau}$ with that inherited from
${\sigma}$. Depending on the result, we define
$\varepsilon({\tau},{\sigma}):=\pm 1$.
\item
If ${\tau}$ is {\em not} a facet of ${\sigma}$, then we simply set
$\varepsilon({\tau},{\sigma}):=0$.
\vspace{-1ex}\end{itemize}
\par
Each cohomological system ${\CF}$ on ${\Sigma}$ gives rise to a
complex $\cxCF{{\kss \bullet}}$ of abelian groups:
\[
\cxCF{{k}}:= \oplus_{{\sigma}\in{\Sigma}_{k}} \,{\CF}({\sigma})
= \oplus_{[{\sigma}\in{\Sigma};\, \dim {\sigma}={k}]} \,{\CF}({\sigma})\,.
\]
The differential $d:\cxCF{{k}}\to\cxCF{{k}+1}$ is defined in the obvious way,
using the $\varepsilon({\tau},{\sigma})$ introduced above. The associated
cohomology is denoted by
\[
\cxHF{{k}}:= H^{k}\big(\cxCF{{\kss \bullet}}\big).
\]
Note that there is an analogous construction for homological systems.
\par
\neu{hodge-spectral}
The cohomology groups of a cohomological system ${\CF}$
can sometimes be computed
using certain subcomplexes of ${\Sigma}$.
To be more precise,
let $\cxM{i}\subseteq {\Sigma}$ be subcomplexes with $\cup_i \cxM{i}={\Sigma}$.
The {\em nerve} ${\CM}$ of this covering is the simplicial set defined as
\[
{\CM}_p:= \{i_0\leq\dots\leqi_p\,|\;
\cxM{i_0}\cap\dots\cap \cxM{i_p} \neq \emptyset\}.
\]
We obtain cohomological systems
${\cal H}^q({\CF})$ on ${\CM}$ via
\[
{\cal H}^q({\CF}): (i_0\leq\dots\leqi_p) \,\mapsto \,
H^q\big(\cxM{i_0}\cap\dots\cap \cxM{i_p},\, {\CF}\big)\,.
\]
\par
{\bf Proposition:}
{\em
There is a degenerating spectral sequence
$E_2^{p,q}=H^p\big({\CM},\,{\cal H}^q({\CF})\big) \Rightarrow \cxHF{p+q}$
with differentials $d_r: E_r^{p,q}\to E_r^{p+r,q-r+1}$.
}
\par
{\bf Proof:}
Consider the double complex
\[
C^{p,q}:=\oplus_{{\sigma}\in [\cxM{i_0}\cap\dots\cap \cxM{i_p}]_q}
\,{\CF}({\sigma}) \; \mbox{ with }\;
d_I:C^{p,q}\to C^{p+1,q}\,,\; d_{II}:C^{p,q}\to C^{p,q+1}\,.
\]
The first spectral sequence yields
$E_2^{p,q}=H^p_I H^q_{II}(C^{{\kss \bullet},{\kss \bullet}})=H^p\big({\CM},\,{\cal H}^q({\CF})\big)$;
the other one provides the complex $\cxCF{{\kss \bullet}}$ at the $E_1$-level, i.e.\
$\cxHF{{\kss \bullet}}$ is the cohomology of the total complex.
\hfill$\Box$
\par
\neu{hodge-def}
Now assume that ${\cx}$ is a {\em fan} in the ${\cxD}$-dimensional
vector space ${\cxV}$, i.e.\ its
elements are polyhedral cones with $0$ as their common vertex. Note that the
intersection of cones from ${\cx}$ is always non-empty.
Another special feature of fans is that they come with an important
cohomological
system for free: ${\cxF}({\cxa}):=\span_k({\cxa})$.
From this cohomological system ``$\span$'' one derives various other systems
like $\raisebox{0.7ex}{${\cxV}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}$ and its exterior powers.
These give rise
to the so called {\em Hodge spaces} of ${\cx}$, a notion
which is due to Brion:
\[
H^{{p},{q}}({\cx}):= H^{{\cxD}-{p}}\Big({\cx},\,\Lambda^{q}
\raisebox{0.7ex}{${\cxV}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$} \Big)^\ast\,.
\]
For rational fans, Danilov has shown in \S 12 of \cite{Da}
that
$H^{{p},{q}}({\cx})$
is $H^{p}(X,\Omega_X^{q})$ where $X=X_{\cx}$ denotes the toric variety
induced by ${\cx}$, and $\Omega_X^{q}$ is the reflexive hull of the K\"ahler
${q}$-differentials on $X_{\cx}$.
For general fans ${\cx}$, Brion has obtained the following vanishing results:
\par
{\bf Proposition:} (cf.\ \S 1 of \cite{Br})
{\em
\begin{itemize}\vspace{-2ex}
\item[(i)]
$H^{{p},{q}}({\cx})=0$ for ${p}<{q}$.
\item[(ii)]
If $|{\cx}|:=\cup_{
{\cxa}\in
{\cx}}{\cxa}$
is not contained in any hyperplane, then $H^{{\cxD},{q}}({\cx})=0$
for ${q}<{\cxD}$, and $H^{{\cxD},{\cxD}}({\cx})$ is isomorphic to ${K}$.
\item[(iii)]
If $|{\cx}|={\cxV}$, and if $e$ is a positive integer such that
cones with dimension at most $e$ are simplicial, then
$H^{{p},{q}}({\cx})=0$ for ${p}-{q}>{\cxD}-e$.
\item[(iv)]
Assume that $|{\cx}|={\cxV}$ and that any two non-simplicial cones in ${\cx}$
intersect only at the origin. Then $H^{2,1}({\cx})=0$.
\vspace{-1ex}\end{itemize}
}
Note that the assumption of (iv) implies that any $({\cxD}-1)$-dimensional cone
in ${\cx}$ is simplicial. Hence, by (iii), it follows that
$H^{{p},1}({\cx})=0$ for ${p}\geq 3$.
\par
\neu{hodge-polytope}
Let ${\Delta}\subseteq{K}^n$ be a compact, convex polytope. It gives rise to the
(inner) {\em normal fan} ${\fn(\kP)}$ in the dual space
$({K}^n)^\ast\cong{K}^n$.
Brion has shown
that the diagonal Hodge spaces $H^{{p},{p}}({\fn(\kP)})$ have then a special
combinatorial meaning; they coincide with the
spaces of the so-called Minkowski ${p}$-weights of ${\Delta}$.\\
In this paper we will focus on the spaces
$H^{{p},1}({\fn(\kP)})$ which sit close to the boundary of the Hodge diamond.
\par
\section{The ${D}$-invariants}\label{Tinv}
\neu{Tinv-setup}
Let ${\Delta}\subseteq{K}^n$ be a {\em compact, convex polytope}; the cone over
it, denoted by $\mbox{\rm cone}({\Delta})$, generates a (non-complete)
fan ${\big\langle\cone(\kP)\big\rangle}$ in ${K}^{n+1}$. This gives rise to the following
invariants of the polytope ${\Delta}$:
\[
{D}^{k}({\Delta}):= H^{k}\Big({\big\langle\cone(\kP)\big\rangle},\,
\raisebox{0.7ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}
\Big) = H^{{k}+1} \big({\big\langle\cone(\kP)\big\rangle},\,\span\big)\,.
\]
The equality is a result of
the exactness of the complex $C^{\kss \bullet}({\big\langle\cone(\kP)\big\rangle},\,{K}^{n+1})$
sitting in the middle
of the short exact sequence of cohomological systems
\[
0\to \span \longrightarrow {K}^{n+1} \longrightarrow
\raisebox{0.7ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}
\to 0\,.
\]
Up to isomorphisms, the vector spaces ${D}^{k}({\Delta})$ depend only
on the {\em projective equivalence class} of the given polytope ${\Delta}$.
However, as examples from McMullen and Smilansky show, they are not
combinatorial invariants of ${\Delta}$.
\par
{\bf
From now on, we will always assume that ${\Delta}\subseteq{K}^n$ has the full
dimension $n$.}
\par
{\bf Lemma:}
{\em
Denote by ${\Delta}^{\scriptscriptstyle\vee}$ the polytope that is {\em polar} to ${\Delta}$, i.e.\ the face
lattice of ${\Delta}^{\scriptscriptstyle\vee}$ is opposite to that of ${\Delta}$, and the cones
$\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee})$ and $\mbox{\rm cone}({\Delta})$ are mutually dual. Then, there is a
perfect pairing
\[
{D}^{k}({\Delta}^{\scriptscriptstyle\vee}) \,\times\, {D}^{n-{k}}({\Delta}) \longrightarrow {K}\,.
\vspace{-3ex}
\]
}
\par
{\bf Proof:}
If ${\cxa}\leq\mbox{\rm cone}({\Delta})$ is an $(n+1-{k})$-dimensional face, then
$\big[{\cxa}^\bot\cap\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee})\big]\leq \mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee})$ is a face of
dimension ${k}$ with
$\span\big[{\cxa}^\bot\cap\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee})\big]={\cxa}^\bot$.
Moreover, all faces of $\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee})$ arise in this way. Hence,
\[
\renewcommand{\arraystretch}{1.5}
\begin{array}{rcl}
{D}^{k}({\Delta}^{\scriptscriptstyle\vee})
&=&
H^{k}\Big({\big\langle\cone(\kP^\veee)\big\rangle},\,
\raisebox{0.7ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$} \Big)
\;=\;
H^{k}\Big({\big\langle\cone(\kP^\veee)\big\rangle},\, \big(({\kss \bullet})^\bot\big)^\ast\Big)\\
&=&
H_{k}\big({\big\langle\cone(\kP^\veee)\big\rangle},\, ({\kss \bullet})^\bot\big)^\ast
\;=\;
H^{n+1-{k}}\big({\big\langle\cone(\kP)\big\rangle},\span\big)^\ast
\;=\;
{D}^{n-{k}}({\Delta})^\ast\,.
\vspace{-3ex}
\end{array}
\]
\hspace*{\fill}$\Box$
\par
\neu{Tinv-ex}
The following remarks are intended to obtain a better feeling for the
meaning of the invariants ${D}^{k}({\Delta})$.
\par
\begin{itemize}\vspace{-2ex}
\item[(i)]
For ${\Delta}=\emptyset$ we define $\mbox{\rm cone}(\emptyset):=0$, hence
${D}^{k}(\emptyset)=0$ for every ${k}\in{Z\!\!\!Z}$.
\item[(ii)]
If ${\Delta}$ is a point, then $\mbox{\rm cone}({\Delta})={K}_{\geq 0}$. In particular,
${D}^0(\mbox{point})={K}$ is the only non-trivial ${D}$-space.
\item[(iii)]
Let $\dim({\Delta})\geq 1$. Then, the defining complex for the ${D}^{k}({\Delta})$
looks like
\[
\begin{array}{ccccccccccccc}
0 & \longrightarrow & C^0 & \longrightarrow & C^1 & \longrightarrow &
\dots & \longrightarrow & C^n & \longrightarrow &
C^{n+1}& \longrightarrow & 0\\
&& || && || &&&& || && ||&&\\
&& {K}^{n+1} &&
\makebox[3em]{$\oplus_{a\in{\Delta}}
\raisebox{0.5ex}{${K}^{n+1}$}\hspace{-0.3em}\big/\hspace{-0.3em}
\raisebox{-0.5ex}{${K}\cdot a$}$}
&&&&
\makebox[3em]{$\oplus_{f<{\Delta}}
\raisebox{0.5ex}{${K}^{n+1}$}\hspace{-0.3em}\big/\hspace{-0.2em}
\raisebox{-0.5ex}{$\span\,f$}$}
&& 0&&
\end{array}
\]
with $a\in{\Delta}$ and $f<{\Delta}$ running through the vertices and facets of ${\Delta}$,
respectively. In particular, the injectivity of $C^0\hookrightarrow C^1$
implies ${D}^0({\Delta})=0$ and, by the previous lemma,
${D}^n({\Delta})={D}^0({\Delta}^{\scriptscriptstyle\vee})^\ast=0$.\\
Hence, ${D}^1({\Delta}),\dots,{D}^{n-1}({\Delta})$ are the only non-trivial
${D}$-invariants of a polytope ${\Delta}\subseteq{K}^n$.
\vspace{-1ex}\end{itemize}
Denote by $f_j({\Delta})$ the number of $j$-dimensional faces of ${\Delta}$
with $f_{-1}:=1$, i.e.\ the Euler equation says $\sum_{j=-1}^n (-1)^j f_j=0$.
Then $\dim C^{k}=(n+1-{k})\cdot f_{{k}-1}$.
\vspace{1ex}\\
{\em $n =2$}:\\
The only non-trivial invariant is ${D}^1$ with
$\;\dim {D}^1({\Delta}) = -\dim C^0 + \dim C^1 - \dim C^2 = f_0({\Delta})-3$.
\vspace{1ex}\\
{\em $n =3$}:\\
${D}^1$ and ${D}^2$ may be non-trivial with
$\;\dim {D}^2({\Delta}) - \dim {D}^1({\Delta}) = \sum_{k} (-1)^{k} \dim C^{k} =
f_2({\Delta})-f_0({\Delta})$.
\par
\neu{Tinv-brion}
We would like to compare the ${D}$-invariants with Brion's Hodge
spaces. First, there are the straightforward relations
\[
{D}^{k}({\Delta})= H^{k}\big({\big\langle\cone(\kP)\big\rangle},\,
\raisebox{0.5ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.5ex}{$\span$} \big)
= H^{n+1-{k},1}\big({\big\langle\cone(\kP)\big\rangle}\big)^\ast
\]
and
\[
{D}^{k}({\Delta}) = {D}^{n-{k}}({\Delta}^{\scriptscriptstyle\vee})^\ast = H^{{k}+1,1}\big({\big\langle\cone(\kP^\veee)\big\rangle}\big)\,.
\]
The ${D}$-invariants have also a direct description in terms of
the normal fan ${\fn(\kP)}$ of ${\Delta}$.
\par
{\bf Proposition:}
{\em
Let ${\Delta}\subseteq{K}^n$ be a compact, convex polytope of dimension $n$
and denote by ${\fn(\kP)}$ its inner normal fan.
Then, there is an exact sequence
\[
0\to {K} \longrightarrow H^{1,1}\big({\fn(\kP)}\big)
\longrightarrow {D}^1({\Delta})\to 0\,.
\]
For the remaining indices ${k}\neq 1$ we have
$H^{{k},1}\big({\fn(\kP)}\big)={D}^{k}({\Delta})$.
}
\par
{\bf Proof:}
Assume that both ${\Delta}$ and ${\Delta}^{\scriptscriptstyle\vee}$ contain the origin as an interior point.
Then, the projection ${K}^{n+1}\rightarrow\hspace{-0.8em}\rightarrow{K}^n$ induces an isomorphism of fans
$\pi:\partial\,\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee})\to{K}_{\geq 0}\cdot\partial{\Delta}^{\scriptscriptstyle\vee}\cong{\fn(\kP)}$.
Moreover, we obtain the following diagram of cohomological systems:
\[
\dgARROWLENGTH=0.8em
\begin{diagram}
\node[4]{0}
\arrow{s}
\node{0}
\arrow{s}\\
\node[4]{{K}}
\arrow{e,t}{\sim}
\arrow{s}
\node{{K}}
\arrow{s}\\
\node{\mbox{on the fan}\;\partial\,\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee}):\hspace{3em}}
\node{0}
\arrow{e}
\node{\span}
\arrow{e}
\arrow{s,r}{\sim}
\node{{K}^{n+1}}
\arrow{e}
\arrow{s}
\node{\raisebox{0.4ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}}
\arrow{e}
\arrow{s}
\node{0}\\
\node{\mbox{on the fan}\;{\fn(\kP)}:\hspace{5em}}
\node{0}
\arrow{e}
\node{\span}
\arrow{e}
\node{{K}^{n}}
\arrow{e}
\arrow{s}
\node{\raisebox{0.4ex}{${K}^{n}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}}
\arrow{e}
\arrow{s}
\node{0}\\
\node[4]{0}
\node{0}
\end{diagram}
\]
Since $H^{{k},1}\big({\fn(\kP)}\big)=H^{n-{k}}\big({\fn(\kP)},
\raisebox{0.4ex}{${K}^{n}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}\big)^\ast$
and
\[
{D}^{k}({\Delta})={D}^{n-{k}}({\Delta}^{\scriptscriptstyle\vee})^\ast=
H^{n-{k}}\big({\big\langle\cone(\kP^\veee)\big\rangle},
\raisebox{0.4ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}\big)^\ast=
H^{n-{k}}\big(\partial\,\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee}),
\raisebox{0.4ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}\big)^\ast,
\]
the last column of the above diagram implies the long exact sequence
\[
\dots\to H^{n-{k}+1}(\partial\,\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee}),{K})^\ast \to
H^{{k},1}\big({\fn(\kP)}\big) \to {D}^{k}({\Delta})
\to H^{n-{k}}(\partial\,\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee}),{K})^\ast \to\dots\;.
\]
On the other hand, by comparison with the cohomology groups
$H^{{\kss \bullet}}({\big\langle\cone(\kP^\veee)\big\rangle},{K})=0$, we obtain that
$H^{{\kss \bullet}}(\partial\,\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee}),{K})$ is also trivial -- with the only
exception
$H^{n}(\partial\,\mbox{\rm cone}({\Delta}^{\scriptscriptstyle\vee}),{K})={K}$.
\hfill$\Box$
\par
\neu{Tinv-T1}
It is well-known that the vector space $H^{1,1}\big({\fn(\kP)}\big)$ of
Minkowski $1$-weights is generated (as an abelian group) by the
semi-group of Minkowski summands of ${K}_{\geq 0}$-multiples of ${\Delta}$.
It is useful to see this fact directly:
\par
$H^{1,1}\big({\fn(\kP)}\big) = H^{n-1}\big({\fn(\kP)},
\raisebox{0.4ex}{${K}^{n}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.7ex}{$\span$}\big)^\ast
= H_{n-1}\big({\fn(\kP)},({\kss \bullet})^\bot\big)$
equals the kernel
\[
\ker\Big[\oplus_{[\dim {\cxa}=n-1]}{\cxa}^\bot \longrightarrow
\oplus_{[\dim {\cxb}=n-2]}{\cxb}^\bot\Big]
=
\ker\Big[\oplus_{d<{\Delta}}{K}\cdot d \longrightarrow
\oplus_{[f<{\Delta}, \,\dim f=2]} \,\span\, f \Big]
\]
with $d\in{K}^n$ running through the edges of ${\Delta}$. The latter space
encodes Minkowski summands of ${K}_{\geq 0}\cdot{\Delta}$ just by
keeping track of the dilatation
factors of the ${\Delta}$-edges, cf.\ \cite{versal}, Lemma (2.2).
\vspace{1ex}\\
Note that the trivial Minkowski summand ${\Delta}$ itself induces the element
$\underline{1}$ in $H^{1,1}\big({\fn(\kP)}\big)\subseteq\oplus_{d<{\Delta}}{K}\cdot d$.
It is exactly this element which is killed in the projection
$H^{1,1}\big({\fn(\kP)}\big)\rightarrow\hspace{-0.8em}\rightarrow{D}^1({\Delta})$ from the previous proposition.
\par
{\bf Corollary:}
{\em
Polytopes ${\Delta}$ with only triangles as two-dimensional faces have a trivial
${D}^1({\Delta})$.
In particular, for simplicial three-dimensional polytopes,
the only non-trivial ${D}$-invariant is ${D}^2({\Delta})$; it has dimension
$f_0({\Delta})-4$.
}
\par
{\bf Proof:}
The first claim is clear. The dimension of ${D}^2({\Delta})$ for three-dimensional
polytopes follows from \zitat{Tinv}{ex}, the Euler equation,
and the fact that
$\,3\,f_2({\Delta})=2\,f_1({\Delta})$ if ${\Delta}$ is simplicial.
\hfill$\Box$
\par
{\bf Examples:} 1) Since the icosahedron $I$ is simplicial, one obtains
${D}^1(I)=0$ and $\,\dim {D}^2(I)=8$.
\vspace{0.5ex}\\
2) Consider three-dimensional pyramids $P^m$ and
double pyramids ${\Diamond}^m$ over an $m$-gon; in both cases we have a trivial
${D}^1$ focusing again the interest on ${D}^2$. Whereas
${D}^2(P^m)$ is also trivial, we do have $\,\dim {D}^2({\Diamond}^m)=m-2$.
\par
\neu{Tinv-double}
We finish this chapter by an extension of the previous example.
We denote by ${\Diamond}({\Delta})\subseteq{K}^{n+1}$ the
{\em double pyramid} over the polytope ${\Delta}\subseteq{K}^n$.
On the polar
level, this means that ${\Diamond}({\Delta})^{\scriptscriptstyle\vee}={\Delta}^{\scriptscriptstyle\vee}\times I$ with
$I:=[0,1]\subseteq{K}^1$.
\par
{\bf Proposition:}
{\em
The natural inclusion
${D}^1({\Delta}^{\scriptscriptstyle\vee})\hookrightarrow{D}^1({\Delta}^{\scriptscriptstyle\vee}\times I)$ has a
one-dimensional cokernel. For ${k}\geq 2$, there are isomorphisms
${D}^{k}({\Delta}^{\scriptscriptstyle\vee})\stackrel{\sim}{\longrightarrow}{D}^{k}({\Delta}^{\scriptscriptstyle\vee}\times I)$.
\vspace{0.5ex}\\
Thus, the ${D}$-invariants of a double pyramid depend on those of the base
via
\[
{D}^{k}\big({\Diamond}({\Delta})\big)={D}^{{k}-1}({\Delta})
\hspace{0.5em}\mbox{for}\hspace{0.5em} {k}\neq n
\hspace{1em} \mbox{ and}\hspace{1em}
\dim\, {D}^n\big({\Diamond}({\Delta})\big)=\dim\,{D}^{n-1}({\Delta})+1\,.
\vspace{-3ex}
\]
}
\par
{\bf Proof:}
Just to impress the reader,
we are going to use the language of triangulated categories.
The normal fan ${\cf}({\Delta}^{\scriptscriptstyle\vee}\times I)$ can be easily expressed by
${\cf}({\Delta}^{\scriptscriptstyle\vee})$; if $N,S\in{K}^{n+1}$ denote the ``poles''
$\pm e^{n+1}$, then
\[
{\cf}({\Delta}^{\scriptscriptstyle\vee}\times I)=
{\cf}({\Delta}^{\scriptscriptstyle\vee}) \sqcup {\cf}^N({\Delta}^{\scriptscriptstyle\vee}) \sqcup {\cf}^S({\Delta}^{\scriptscriptstyle\vee})
\;\mbox{ with }\;
{\cf}^{N/S}({\Delta}^{\scriptscriptstyle\vee}):=\big\{\langle{\cxa},N/S\rangle\subseteq{K}^{n+1}
\,|\; {\cxa}\in{\cf}({\Delta}^{\scriptscriptstyle\vee})\big\}.
\]
Since the complex
$C^{\kss \bullet}\big({\cf}({\Delta}^{\scriptscriptstyle\vee}\times I),\,
\raisebox{0.5ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.4ex}{$\span$} \big)$
is isomorphic to the shifted mapping cone $C^{\kss \bullet}_{(\pi,\pi)}[-1]$ with
\[
\pi\,:\;
C^{\kss \bullet}\Big({\cf}({\Delta}^{\scriptscriptstyle\vee}),\,
\raisebox{0.6ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.6ex}{$\span$} \Big)
\rightarrow\hspace{-0.8em}\rightarrow
C^{\kss \bullet}\Big({\cf}({\Delta}^{\scriptscriptstyle\vee}),\,
\raisebox{0.6ex}{${K}^{n}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.6ex}{$\span$} \Big)
\]
and $(\pi,\pi):\bullet\to\bullet\oplus \bullet$,
we obtain that
$\,C^{\kss \bullet}\big({\cf}({\Delta}^{\scriptscriptstyle\vee}\times I),\,
\raisebox{0.5ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.4ex}{$\span$} \big)[1]$
and
$\,C^{\kss \bullet}\big({\cf}({\Delta}^{\scriptscriptstyle\vee}),\,{K}\big)[1]$ are on top of the distinguished
triangles over the maps $(\pi,\pi)$ and $\pi$, respectively.
Hence, the octahedral axiom for triangulated categories yields a new
distinguished triangle
\[
\dgARROWLENGTH=0.1em
\begin{diagram}
\node[2]{\makebox[3em]{$C^{\kss \bullet}\Big({\cf}({\Delta}^{\scriptscriptstyle\vee}),\,
\raisebox{0.6ex}{${K}^{n}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.6ex}{$\span$} \Big)$}}
\arrow{sw,t}{[1]}\\
\node{\,C^{\kss \bullet}\Big({\cf}({\Delta}^{\scriptscriptstyle\vee}),\,{K}\Big)[1]}
\arrow[2]{e}
\node[2]{\,C^{\kss \bullet}\Big({\cf}({\Delta}^{\scriptscriptstyle\vee}\times I),\,
\raisebox{0.5ex}{${K}^{n+1}$}\hspace{-0.4em}\big/\hspace{-0.2em}
\raisebox{-0.4ex}{$\span$} \Big)[1]}
\arrow{nw}
\end{diagram}
\]
inducing the long exact sequence
\[
\dots\to
H^{{k},1}\big({\cf}({\Delta}^{\scriptscriptstyle\vee}\times I)\big)^\ast
\to
H^{{k},1}\big({\cf}({\Delta}^{\scriptscriptstyle\vee})\big)^\ast
\to
H^{n-{k}+2}\big({\cf}({\Delta}^{\scriptscriptstyle\vee}),\,{K}\big)
\to
H^{{k}-1,1}\big({\cf}({\Delta}^{\scriptscriptstyle\vee}\times I)\big)^\ast
\to\dots\;.
\]
As already mentioned at the end of the proof of Proposition
\zitat{Tinv}{brion}, the spaces
$H^{n-{k}+2}\big({\cf}({\Delta}^{\scriptscriptstyle\vee}),\,{K}\big)$ vanish unless ${k}=2$.
Thus, it remains to use Proposition \zitat{Tinv}{brion} itself, and to
remark that the injection
${D}^1({\Delta}^{\scriptscriptstyle\vee})\hookrightarrow{D}^1({\Delta}^{\scriptscriptstyle\vee}\times I)$
cannot be an isomorphism.
\hfill$\Box$
\par
{\bf Example:}
Let ${\Delta}$ be the three-dimensional cuboctahedron obtained by cutting the eight
corners of a cube.
\begin{center}
\unitlength=0.7pt
\begin{picture}(284.00,224.00)(80.00,559.00)
\thinlines
\linethickness{0.1pt}
\put(80.00,560.00){\line(1,0){160.00}}
\put(80.00,720.00){\line(1,0){160.00}}
\put(200.00,780.00){\line(1,0){160.00}}
\put(200.00,620.00){\line(1,0){160.00}}
\put(240.00,560.00){\line(2,1){120.00}}
\put(240.00,720.00){\line(2,1){120.00}}
\put(80.00,720.00){\line(2,1){120.00}}
\put(80.00,560.00){\line(2,1){120.00}}
\put(360.00,620.00){\line(0,1){160.00}}
\put(80.00,560.00){\line(0,1){160.00}}
\put(240.00,560.00){\line(0,1){160.00}}
\put(200.00,620.00){\line(0,1){160.00}}
\thicklines
\put(161.00,560.00){\line(5,1){136.00}}
\put(297.00,588.00){\line(-6,5){56.00}}
\put(241.00,635.00){\line(-1,-1){76.00}}
\put(360.00,698.00){\line(-3,-5){65.00}}
\put(81.00,640.00){\line(1,-1){81.00}}
\put(82.00,639.00){\line(1,1){78.00}}
\put(160.00,717.00){\line(-3,5){20.00}}
\put(140.00,750.00){\line(-1,-2){54.00}}
\put(161.00,720.00){\line(5,1){137.00}}
\put(298.00,748.00){\line(-1,-2){56.00}}
\put(242.00,636.00){\line(-1,1){79.00}}
\put(298.00,748.00){\line(5,-4){61.00}}
\put(359.00,700.00){\line(-1,1){79.00}}
\put(280.00,779.00){\line(3,-5){18.00}}
\put(142.00,749.00){\line(5,1){139.00}}
\put(163.00,560.00){\circle*{10}}
\put(296.00,588.00){\circle*{10}}
\put(296.00,748.00){\circle*{10}}
\put(140.00,750.00){\circle*{10}}
\put(160.00,720.00){\circle*{10}}
\put(81.00,640.00){\circle*{10}}
\put(240.00,637.00){\circle*{10}}
\put(360.00,700.00){\circle*{10}}
\put(280.00,780.00){\circle*{10}}
\end{picture}
\end{center}
Then, ${\Delta}$ may be decomposed into a Minkowski sum of two tetrahedra --
the summands are formed by the triangles in every other corner. In
particular, $\dim\,{D}^1({\Delta})=1$. Moreover, since
$f_0({\Delta})=12$, $f_1({\Delta})=24$, $f_2({\Delta})=14$,
Example \zitat{Tinv}{ex} implies $\dim\,{D}^2({\Delta})=3$.\\
The four-dimensional double pyramid ${\Diamond}({\Delta})$ has two kinds of facets:
16 tetrahedra and 12 pyramids over squares.
From the previous proposition, we obtain
${D}^1\big({\Diamond}({\Delta})\big)=0$,
$\dim\,{D}^2\big({\Diamond}({\Delta})\big)=1$, and
$\dim\,{D}^3\big({\Diamond}({\Delta})\big)=4$.
\par
\section{The vanishing theorem}\label{van}
\neu{van-brion}
Let ${\Delta}\subseteq{K}^n$ be an $n$-dimensional, compact, convex
polytope with $n\geq 1$.
We would like to find conditions under which some of the spaces ${D}^{k}({\Delta})$
vanish.\\
The first idea is to check Brion's properties (iii) and (iv) of
\zitat{hodge}{def} for this purpose. However, we do not
find surprising results in this way -- for instance the first two claims of the
following proposition are already contained in \cite{Br}. The third assertion
generalizes the observation made in Corollary \zitat{Tinv}{T1}.
\par
\begin{minipage}{\textwidth}
{\bf Proposition:}
{\em
\begin{itemize}\vspace{-2ex}
\item[(1)]
If ${\Delta}$ is a simple polytope, then ${D}^{k}({\Delta})=0$ for ${k}\geq 2$. In
particular, for the space of Minkowski summands we obtain
$\,\dim {D}^1({\Delta})=\sum_j (-1)^{j+1}\,j\cdot f_j({\Delta})$.
\vspace{-1ex}\end{itemize}
}
\end{minipage}
{\em
\begin{itemize}\vspace{-2ex}
\item[(2)]
If each face of ${\Delta}$ contains at most one non-simple vertex, then we have
still ${D}^2({\Delta})=0$.
\item[(3)]
If any ${\ell}$-face of ${\Delta}$ is a simplex, then ${D}^{k}({\Delta})=0$ for ${k}<{\ell}$.
\vspace{-1ex}\end{itemize}
}
(A vertex of ${\Delta}$ is called {\em simple} if it sits in exactly $n=\dim{\Delta}$
different facets. Moreover, the whole polytope is said to be
simple if every vertex is.)
\par
{\bf Proof:}
The first two claims exploit the fact that
${D}^{k}({\Delta})=H^{{k},1}\big({\fn(\kP)}\big)$
for ${k} \geq 2$. The dimension of ${D}^1({\Delta})$ follows from
$\,\dim {D}^1({\Delta})=\sum_{k} (-1)^{{k}+1}\dim {D}^{k} ({\Delta})=
\sum_{k} (-1)^{{k}+1}(n+1-{k})\cdot f_{{k}-1}$
as in \zitat{Tinv}{ex}.\\
On the other hand, the proof of the third assertion uses
\zitat{hodge}{def}(iii) for the ``dual'' fan ${\cf}({\Delta}^{\scriptscriptstyle\vee})$: \\
${D}^{k}({\Delta})={D}^{n-{k}}({\Delta}^{\scriptscriptstyle\vee})^\ast=
H^{n-{k},1}\big({\cf}({\Delta}^{\scriptscriptstyle\vee})\big)^\ast=0\,$ if $\,(n-{k})-1>n-({\ell}+1)$.
\hfill$\Box$
\par
\neu{van-spectral}
More interesting results can be obtained from working with the spectral sequence
introduced in \zitat{hodge}{spectral}. When applied to the ``affine'' fan
${\cx}:={\big\langle\cone(\kP)\big\rangle}$, it provides us with a nice tool for studying the spaces ${D}^{k}({\Delta})$
up to a certain bound ${k}<{\ell}$:
\par
Write $\kPM{i}<{\Delta}$ for
the ${\ell}$-dimensional faces and denote
by ${\Delta}^{({\ell})}:=\cup_i\kPM{i}$ the ${\ell}$-skeleton of our polytope ${\Delta}$.
Then, the simplicial complex ${\CM}$ with
${\CM}_p:= \{i_0\leq\dots\leqi_p\}$
carries the cohomological system
\[
{\cal \kT}^q: (i_0\leq\dots\leqi_p) \,\mapsto \,
{D}^q\big(\kPM{i_0}\cap\dots\cap \kPM{i_p}\big)\,.
\]
\par
{\bf Proposition:}
{\em
There is a degenerating spectral sequence
with differentials $d_r: E_r^{p,q}\to E_r^{p+r,q-r+1}$
such that
$E_2^{p,q}=H^p\big({\CM},\,{\cal \kT}^q\big) \Rightarrow {D}^{p+q}({\Delta})$ for
$p+q<{\ell}$.
}
\par
{\bf Proof:}
Let ${\cx}:={\big\langle\cone(\kP)\big\rangle}^{({\ell}+1)}$
be the union of cones with dimension at most $({\ell}+1)$. Then,
using the cohomological system
\[
{\cal H}^q(\span\,\mbox{\rm cone}):
(i_0\leq\dots\leqi_p) \,\mapsto \,
H^q\big(\mbox{\rm cone}\,\kPM{i_0}\cap\dots\cap \mbox{\rm cone}\,\kPM{i_p},\,\span\big)\,,
\]
section \zitat{hodge}{spectral} yields a degenerating spectral sequence with
\[
E_2^{p,q}=H^p\big({\CM},\,{\cal H}^q(\span\,\mbox{\rm cone})\big) \Rightarrow
H^{p+q}\big({\cx},\,\span\big).
\]
On the other hand, by \zitat{Tinv}{setup} we have
$H^q\big(\mbox{\rm cone}\,\kPM{i_0}\cap\dots\cap \mbox{\rm cone}\,\kPM{i_p},\,\span\big)=
{D}^{q-1}\big(\kPM{i_0}\cap\dots\cap \kPM{i_p}\big)$ and,
for $p+q<{\ell}+1$,
$H^{p+q}\big({\cx},\,\span\big)=H^{p+q}\big({\big\langle\cone(\kP)\big\rangle},\,\span\big)=
{D}^{p+q-1}({\Delta})$. Hence, an index shift by one completes the proof.
\hfill$\Box$
\par
\neu{van-supp}
The cohomology groups $E_2^{p,q}=H^p\big({\CM},\,{\cal \kT}^q\big)$ remain unchanged
when we build the complex $C^{\kss \bullet}\big({\CM},\,{\cal \kT}^q\big)$ only from
the strict tuples $(i_0<\dots<i_p)$.
\vspace{0.5ex}
In particular, besides $E_2^{0,0}=0$, we obtain at the first glance that
$E_2^{p,q}=0$ for $q\geq {\ell}$ or $p\geq 1\,,q={\ell}-1$.
However, since the previous proposition restricts us to the region
$p+q<\ell$ anyway, these first vanishings do not help.
\ktrash{\begin{center}
\unitlength=1.200000pt
\begin{picture}(230.00,80.00)(-60.00,0.00)
\put(148.00,68.00){\makebox(0.00,0.00){$d_2$}}
\put(128.00,68.00){\vector(2,-1){40.00}}
\put(-60.00,38.00){\makebox(0.00,0.00){
$\begin{array}{c}E_2^{{\kss \bullet},{\kss \bullet}}\\(\mbox{for}\, {\ell}=3)\end{array}$}}
\put(100.00,0){\makebox(0.00,0.00){$p$}}
\put(3.00,80.00){\makebox(0.00,0.00){$q$}}
\put(-5.00,8.00){\makebox(0.00,0.00){${D}^0$}}
\put(-5.00,28.00){\makebox(0.00,0.00){${D}^1$}}
\put(-5.00,48.00){\makebox(0.00,0.00){${D}^2$}}
\put(-5.00,68.00){\makebox(0.00,0.00){${D}^3$}}
\multiput(2,64)(14,-14){5}{\line(1,-1){9.00}}
\put(8.00,8.00){\circle*{2.00}}
\put(28.00,8.00){\circle*{2.00}}
\put(48.00,8.00){\circle*{2.00}}
\put(68.00,8.00){\circle*{2.00}}
\put(88.00,8.00){\circle*{2.00}}
\put(8.00,28.00){\circle*{2.00}}
\put(28.00,28.00){\circle*{2.00}}
\put(48.00,28.00){\circle*{2.00}}
\put(68.00,28.00){\circle*{2.00}}
\put(88.00,28.00){\circle*{2.00}}
\put(8.00,48.00){\circle*{2.00}}
\put(28.00,48.00){\circle*{2.00}}
\put(48.00,48.00){\circle*{2.00}}
\put(68.00,48.00){\circle*{2.00}}
\put(88.00,48.00){\circle*{2.00}}
\put(8.00,68.00){\circle*{2.00}}
\put(28.00,68.00){\circle*{2.00}}
\put(48.00,68.00){\circle*{2.00}}
\put(68.00,68.00){\circle*{2.00}}
\put(88.00,68.00){\circle*{2.00}}
\put(8.00,8.00){\line(1,0){90.00}}
\put(8.00,78.00){\line(0,-1){70.00}}
\put(8.00,8.00){\circle{5.00}}
\put(28.00,48.00){\circle{5.00}}
\put(48.00,48.00){\circle{5.00}}
\put(68.00,48.00){\circle{5.00}}
\put(88.00,48.00){\circle{5.00}}
\put(8.00,68.00){\circle{5.00}}
\put(28.00,68.00){\circle{5.00}}
\put(48.00,68.00){\circle{5.00}}
\put(68.00,68.00){\circle{5.00}}
\put(88.00,68.00){\circle{5.00}}
\end{picture}}
\end{center}
}
\par
For vertices $a\in{\Delta}$ we denote by ${\Delta}(a)$ the corresponding vertex figure;
it is the polytope obtained by cutting ${\Delta}$ with a hyperplane
sufficiently close to $a$. The faces of ${\Delta}(a)$ are exactly the vertex
figures of those ${\Delta}$-faces containing $a$.
\par
{\bf Lemma:}
{\em
Unless $p={\ell}$, the vector spaces $E_2^{p,0}$ on the buttom row vanish.
The remaining one may be expressed by singular cohomology groups with
values in ${K}$ as
\[
E_2^{{\ell},0}=\oplus_{a\in{\Delta}\,\mbox{\rm \small vertex}}
\,H^{\ell}\big({\Delta}(a), \,{\Delta}(a)^{({\ell}-1)}\big) =
\oplus_{a\in{\Delta}\,\mbox{\rm \small vertex}}
\,\widetilde{H}^{{\ell}-1}\big({\Delta}(a)^{({\ell}-1)}\big)
\]
with ${\Delta}(a)^{({\ell}-1)}$ denoting the $({\ell}-1)$-skeleton of the
vertex figure ${\Delta}(a)$.
}
\par
{\bf Proof:}
According to the remark at the beginning of the present section
\zitat{van}{supp}, the vector space
$E_2^{p,0}=H^p\big({\CM},{\cal \kT}^0\big)$ is the $p$-th cohomology of the complex
\[
0\longrightarrow
\oplus_{i_0}{D}^0(\kPM{i_0})
\longrightarrow
\oplus_{i_0<i_1}{D}^0(\kPM{i_0}\cap\kPM{i_1})
\longrightarrow
\oplus_{i_0<i_1<i_2}{D}^0(\kPM{i_0}\cap\kPM{i_1}\cap\kPM{i_2})
\longrightarrow\dots
\]
which will also be denoted by ${\cal \kT}^0$ (creating a slight abuse of
notation).
On the other hand,
for any vertex $a\in{\Delta}$ we call ${\cal \kT}^0(a)$ the complex built similarly as
${\cal \kT}^0$, but using only those faces $\kPM{i}<{\Delta}$ containing $a$.
Since ${D}^0$ is trivial unless its argument is a point, the canonical
projection
\[
{\cal \kT}^0\rightarrow\hspace{-0.8em}\rightarrow\oplus_{a\in{\Delta}}{\cal \kT}^0(a)
\]
yields an isomorphism of complexes.\\
This splitting enables us to
fix an arbitrary vertex $a\in{\Delta}$ and to forget
about faces $\kPM{i}$ which do not contain $a$. Then, using the
vertex figures $\kPM{i}(a)<{\Delta}(a)$, the whole story may be translated
into singular cohomology with values in ${K}$ via
\[
{D}^0\big(\kPM{i_0}\cap\dots\cap\kPM{i_p}\big)
=H^0\big({\Delta}(a),\, \kPM{i_0}(a)\cap\dots\cap\kPM{i_p}(a)\big).
\]
Denoting by $C_q({\kss \bullet})$ the singular $q$-chains,
the Mayer-Vietoris spectral sequence yields
\[
\begin{array}{r@{\;}c@{\;}l}
H^p\big({\cal \kT}^0(a)\big)= H^p\Big(\Big[
\raisebox{0.5ex}{$C_{\kss \bullet}\big({\Delta}(a)\big)$}
\hspace{-0.1em}\Big/\hspace{-0.1em}
\raisebox{-0.7ex}{$\sum_i C_{\kss \bullet}\big(\kPM{i}(a)\big)$}
\Big]^\ast\Big)
&=&H^p\big({\Delta}(a),\,\cup_i \kPM{i}(a)\big)\\
&=&H^p\big({\Delta}(a),\,{\Delta}(a)^{({\ell}-1)}\big)
= \widetilde{H}^{p-1}\big({\Delta}(a)^{({\ell}-1)}\big),
\end{array}
\]
cf.\ \zitat{proof}{sing} for more details.
Now, the observation that
the latter groups vanish unless $p={\ell}$ finishes the proof.
\hfill$\Box$
\par
{\bf Corollary:}
{\em
(1) Let ${\ell}\geq 2$. If any at most ${\ell}$-dimensional face $\kPM{}<{\Delta}$
satisfies ${D}^{k}(\kPM{})=0$ for $0<{k}<{\ell}$, then
so does the polytope ${\Delta}$ itself.
\vspace{0.5ex}\\
(2) If there is an ${\ell}\geq 2$ such that ${D}^1(\kPM{})=0$ for every
${\ell}$-face $\kPM{}<{\Delta}$, then ${D}^1({\Delta})=0$.
}
\par
{\bf Proof:}
(1) This generalization of the last claim of
Proposition \zitat{van}{brion}
follows directly from the spectral sequence \zitat{van}{spectral}: The
assumption means that $E_2^{p,q}=0$ for $p+q<{\ell},\, q\neq 0$, and the
previous lemma takes care of the case $q=0$.
\vspace{0.5ex}\\
(2) Here, the assumption translates into the vanishing $E_2^{0,1}=0$.
\hfill$\Box$
\par
\neu{van-T1T2}
For the rest of this chapter, we focus on the case ${\ell}=3$, i.e.\ we would like
to investigate ${D}^1({\Delta})$ and ${D}^2({\Delta})$ by studying the 3-dimensional faces
of ${\Delta}$.
Here comes the actual situation of the second layer of our spectral
sequence (the big circles stand for the vanishing of the corresponding
$E_2$-term):
\vspace{-2ex}
\begin{center}
\unitlength=1.200000pt
\begin{picture}(230.00,80.00)(-60.00,0.00)
\put(148.00,68.00){\makebox(0.00,0.00){$d_2$}}
\put(128.00,68.00){\vector(2,-1){40.00}}
\put(-60.00,33.00){\makebox(0.00,0.00){$E_2^{{\kss \bullet},{\kss \bullet}}\;\mbox{for}\;{\ell}=3$}}
\put(100.00,0){\makebox(0.00,0.00){$p$}}
\put(3.00,80.00){\makebox(0.00,0.00){$q$}}
\put(-5.00,8.00){\makebox(0.00,0.00){${D}^0$}}
\put(-5.00,28.00){\makebox(0.00,0.00){${D}^1$}}
\put(-5.00,48.00){\makebox(0.00,0.00){${D}^2$}}
\put(-5.00,68.00){\makebox(0.00,0.00){${D}^3$}}
\multiput(2,64)(14,-14){5}{\line(1,-1){9.00}}
\put(8.00,8.00){\circle*{2.00}}
\put(28.00,8.00){\circle*{2.00}}
\put(48.00,8.00){\circle*{2.00}}
\put(68.00,8.00){\circle*{2.00}}
\put(88.00,8.00){\circle*{2.00}}
\put(8.00,28.00){\circle*{2.00}}
\put(28.00,28.00){\circle*{2.00}}
\put(48.00,28.00){\circle*{2.00}}
\put(68.00,28.00){\circle*{2.00}}
\put(88.00,28.00){\circle*{2.00}}
\put(8.00,48.00){\circle*{2.00}}
\put(28.00,48.00){\circle*{2.00}}
\put(48.00,48.00){\circle*{2.00}}
\put(68.00,48.00){\circle*{2.00}}
\put(88.00,48.00){\circle*{2.00}}
\put(8.00,68.00){\circle*{2.00}}
\put(28.00,68.00){\circle*{2.00}}
\put(48.00,68.00){\circle*{2.00}}
\put(68.00,68.00){\circle*{2.00}}
\put(88.00,68.00){\circle*{2.00}}
\put(8.00,8.00){\line(1,0){90.00}}
\put(8.00,78.00){\line(0,-1){70.00}}
\put(8.00,8.00){\circle{5.00}}
\put(28.00,8.00){\circle{5.00}}
\put(48.00,8.00){\circle{5.00}}
\put(88.00,8.00){\circle{5.00}}
\put(28.00,48.00){\circle{5.00}}
\put(48.00,48.00){\circle{5.00}}
\put(68.00,48.00){\circle{5.00}}
\put(88.00,48.00){\circle{5.00}}
\put(8.00,68.00){\circle{5.00}}
\put(28.00,68.00){\circle{5.00}}
\put(48.00,68.00){\circle{5.00}}
\put(68.00,68.00){\circle{5.00}}
\put(88.00,68.00){\circle{5.00}}
\end{picture}}
\end{center}
\par
{\bf Proposition:}
{\em
Denote by $\kPM{i}<{\Delta}$
the three-dimensional faces of ${\Delta}$. Then
\vspace{-1ex}
\begin{itemize}\vspace{-2ex}
\item[(1)]
$\,{D}^1({\Delta})=\ker\Big[
\oplus_i {D}^1(\kPM{i})\longrightarrow
\oplus_{i<j} {D}^1(\kPM{i}\cap\kPM{j})\Big]\,$ and
\vspace{-1ex}
\item[(2)]
if $\,{D}^2(\kPM{i})=0$ for every $i$, then
$\,{D}^2({\Delta})=\ker\Big[ d_2:E_2^{1,1}\longrightarrow E_2^{3,0}\Big]\,$.
\vspace{-1ex}\end{itemize}
}
\par
{\bf Proof:}
The claims follow from
${D}^1({\Delta})=E_{\infty}^{0,1}=E_2^{0,1}=H^0\big({\CM},{\cal \kT}^1\big)$
and, since $E_2^{0,2}=0$ in (2), from
${D}^2({\Delta})=E_{\infty}^{1,1}=E_3^{1,1}$.
\hfill$\Box$
\par
\neu{van-cumbersome}
We are going to apply the previous properties to obtain an explicit
description of
${D}^2({\Delta})$ by equations. In the following we will use the symbols
$a$, $\kPV{}$, $\kPP{}{}$, ${F}$ to denote ${\Delta}$-faces of dimension 0, 2, 3,
and 4, respectively.
\par
{\bf Notation:}
Whenever $(\kPV{},{F})$ is a flag with dimension vector $(2,4)$, then we
denote by $\kPP{(\kPV{},{F})}{}$ and $\kPP{}{(\kPV{},{F})}$
the two unique three-dimensional faces
sitting in between. Their order depends on the orientation of the whole
configuration.\\
For any two-dimensional face $\kPV{}\leq{\Delta}$ we fix some three-dimensional
face $\kPP{}{}(\kPV{})$ containing $\kPV{}$.
\par
{\bf Theorem:}
{\em
Assume that ${D}^1(\kPP{}{})={D}^2(\kPP{}{})=0$ for every three-dimensional
face $\kPP{}{}\leq{\Delta}$. Then,
${D}^2({\Delta})\subseteq {K}^{\#\{(0,2,3)-\mbox{\rm\footnotesize flags}\}}$ is given by the
following equations in the variables called
$\kts{a}{\kPP{}{}}{\kPV{}}$:
\begin{itemize}\vspace{-2ex}
\item[(1)]
If $(a,{F})$ is a flag with dimension $(0,4)$, then
\[
\sum_{a\in\kPV{}\subseteq{F}}
\Big[\kts{a}{\kPP{(\kPV{},{F})}{}}{\kPV{}} -
\kts{a}{\kPP{}{(\kPV{},{F})}}{\kPV{}}\Big]
=0\,.
\makebox[1em][l]{$\hspace*{4em} (1)_{(a,{F})}$}
\]
\item[(2)]
For every flag $(\kPV{},\kPP{}{})$, the coordinates
$\kts{{\kss \bullet}}{\kPP{}{}}{\kPV{}}$ provide an affine relation among the
vertices of $\kPV{}$, i.e.\
\[
\sum_{a\in\kPV{}} \kts{a}{\kPP{}{}}{\kPV{}}\cdot [a,1]=0\,.
\makebox[1em][l]{$\hspace*{8.6em} (2)_{(\kPV{},\kPP{}{})}$}
\]
\item[(3)]
Finally, for each $(0,2)$-flag $(a,\kPV{})$, we simply have
\[
\kts{a}{\kPP{}{}(\kPV{})}{\kPV{}} =0\,.
\makebox[1em][l]{$\hspace*{10.2em} (3)_{(a,\kPV{})}$}
\vspace{-3ex}
\]
\vspace{-1ex}\end{itemize}
}
\par
Note that the equations $(2)_{(\kPV{},\kPP{}{})}$ imply that we can
completly forget
about the triangular faces $\kPV{}$; they provide only trivial coordinates
$\kts{{\kss \bullet}}{{\kss \bullet}}{\kPV{}}=0$.
\par
The {\em proof of the previous theorem} consists of a detailed, but
straightforward analysis of the differential map
$d_2:E_2^{1,1}\to E_2^{3,0}$. Since it is quite long and technical,
we postpone these calculations to their own section \S \ref{proof}.
In the rest of \S \ref{van}, we continue with a discussion of the
consequences and applications.
\par
\neu{van-dim4}
{\bf Corollary:}
{\em
If the polytope ${\Delta}$ is four-dimensional, then
${D}^2({\Delta})\subseteq {K}^{\#\{(0,2)-\mbox{\rm\footnotesize flags}\}}$
is given by the easier equations
\begin{itemize}\vspace{-2ex}
\item[(1)]
$\;\sum_{\kPV{}\ni a} s(a,\kPV{})=0\,$
for every vertex $a\in{\Delta}$, and
\item[(2)]
$\;\sum_{a\in\kPV{}} s(a,\kPV{})\cdot [a,1]=0\,$ for the two-dimensional faces
$\kPV{}\leq{\Delta}$.
\vspace{-1ex}\end{itemize}
}
\par
{\bf Proof:}
Since ${F}={\Delta}$, we may just set
$\kPP{}{}(\kPV{}):=\kPP{}{(\kPV{},{\Delta})}$ and
$s(a,\kPV{}):= \kts{a}{\kPP{(\kPV{},{\Delta})}{}}{\kPV{}}$.
\hfill$\Box$
\par
{\bf Example:}
Consider the double pyramid ${\Diamond}({\Delta})$ of Example \zitat{Tinv}{double}.
A non-trivial element of the one-dimensional ${D}^2\big({\Diamond}({\Delta})\big)$
may be obtained by assigning $\pm 1$ to the vertices of each rectangle
such that adjacent vertices obtain opposite signs.
\par
\neu{van-pyramid}
The main point of the present paper is to provide a vanishing theorem for
${D}^2({\Delta})$ for polytopes
whose three-dimensional faces are not assumed to be simplices.
\par
{\bf Definition:}
We define an inductive process of {\em``cleaning'' vertices and
two-dimensional faces} of ${\Delta}$. At the beginning, all faces are
assumed to be ``contaminated'', but then one may repeatedly apply the
following rules (i) and (ii) in an arbitrary order:
\begin{itemize}\vspace{-2ex}
\item[(i)]
A two-dimensional ${m}$-gon $\kPV{}<{\Delta}$ is said to be clean if at least
$({m}-3)$ of its vertices are so. (In particular, every triangle is
automatically clean.)
\item[(ii)]
A vertex of ${\Delta}$ is declaired to be clean if it is contained in no more than
$(n-3)$ two-dimensional faces that are not cleaned yet.
\vspace{-1ex}\end{itemize}
{\bf Examples:}
(1)
If no vertex of ${\Delta}$ is contained in more than $(n-3)$ two-dimensional,
non-triangular faces, then every vertex and every two-dimensional face
may be cleaned.
\vspace{1ex}\\
(2)
Each vertex of the four-dimensional double pyramid ${\Diamond}({\Delta})$ shown in
Example \zitat{Tinv}{double} sits in exactly $2=(n-3)+1$
quadrangular, two-dimensional faces. In particular, it is not possible
to clean any of them at all.
\par
{\bf Theorem:}
{\em
Let ${\Delta}$ be an $n$-dimensional, compact, convex polytope such that
every three-dimensional face is a pyramid.
If every vertex
(or, equivalently, every two-dimensional face)
may be cleaned in the sense of the previous definition,
then ${D}^2({\Delta})=0$.
}
\par
{\bf Remarks:}
(1)
Pyramids are the easiest three-dimensional solids with trivial
${D}$-invariants. Moreover,
polytopes with only pyramids as three-dimensional faces do naturally arise
from quivers, cf.\ \cite{AvS} for more details.
\vspace{1ex}\\
(2)
The double pyramid ${\Diamond}({\Delta})$ from Example \zitat{Tinv}{double}
has a non-trivial ${D}^2$. This shows that the assumption concerning the cleaning of vertices
cannot be dropped.
\par
{\bf Proof:}
Using the dictionary
\vspace{-2ex}
\begin{center}
\begin{tabular}{r@{\hspace{1em}$\longleftrightarrow$\hspace{1.2em}}l}
``the vertex $a$ is clean''
&
$\kts{a}{\kPP{}{}}{\kPV{}}=0$ for every $\kPV{},\kPP{}{}$\\
``the 2-face $\kPV{}$ is clean''
&
$\kts{a}{\kPP{}{}}{\kPV{}}=0$ for every $a,\kPP{}{}$,
\end{tabular}
\vspace{-2ex}
\end{center}
the vanishing of ${D}^2({\Delta})$ is a consequence of Theorem
\zitat{van}{cumbersome} and the following two facts:
\vspace{1ex}\\
(i) If $\kPV{}$ is an ${m}$-gon, then, for any $\kPP{}{}$, the
equation $(2)_{(\kPV{},\kPP{}{})}$ of Theorem \zitat{van}{cumbersome}
says that the coordinates
$\kts{{\kss \bullet}}{\kPP{}{}}{\kPV{}}$ describe an $({m}-3)$-dimensional vector
space. Hence, if $({m}-3)$ of them vanish, then they do all.
In particular, as already mentioned in \zitat{van}{cumbersome}, we do not
have to care about triangular faces $\kPV{}$.
\vspace{1ex}\\
(ii) Assume that $\kPP{A}{}, \kPP{B}{}$ are two pyramids
with common facet $\kPV{}<{\Delta}$.\\
We denote by ${\Delta}(\kPV{})$ the $(n-3)$-dimensional vertex figure
of a slice of ${\Delta}$ transversal to $\kPV{}$.
In particular, the faces of ${\Delta}(\kPV{})$ correspond to those of ${\Delta}$
containing $\kPV{}$:
While $\bar{\kPV{}}:=\kPV{}(\kPV{})=\emptyset$, the two pyramids
turn into vertices
$\kPPb{A}{}:=\kPP{A}{}(\kPV{})$ and
$\kPPb{B}{}:=\kPP{B}{}(\kPV{})$. Moreover, any four-dimensional face
${F}<{\Delta}$ containing $\kPV{}$ corresponds to an edge ${\bar{\kPF}}$ in
${\Delta}(\kPV{})$.\\
The important feature about pyramids as three-dimensional faces is
the following:
Any two non-triangular, two-dimensional faces of ${\Delta}$ span an at least
four-dimensional space.
Hence, for any two-dimensional $\kPV{\prime}$, different from
$\kPV{}$, there is {\em at most one}
four-dimensional ${F}^\prime<{\Delta}$ containing both
$\kPV{}$ and $\kPV{\prime}$.\\
Thus, if there are given $(n-4)$
(contaminated) faces
$\kPV{{k}}$ additional to $\kPV{}$, then they induce at most $(n-4)$
four-dimensional faces ${F}^{k}$
in this way.
Since $\dim\,{\Delta}(\kPV{})=n-3$, this means that it is possible
to find a path along the edges of ${\Delta}(\kPV{})$ connecting the vertices
$\kPPb{A}{}$ and $\kPPb{B}{}$, but avoiding $\bar{{F}}^{k}$
(${k}=1,\dots,n-4$).\\
Let us, w.l.o.g., assume that $\kPPb{A}{}$ and
$\kPPb{B}{}$ are directly connected via an edge $\bar{{F}}$ with
${F}$ not containing the $(n-4)$ faces $\kPV{{k}}\neq\kPV{}$. Hence,
$\kPPb{A}{}=\kPP{(\kPV{},{F})}{}$,
$\kPPb{B}{}=\kPP{}{(\kPV{},{F})}$, and
in the equation $(1)_{(a,{F})}$ of Theorem \zitat{van}{cumbersome}
\[
\sum_{a\in{\kss \bullet}\subseteq{F}}
\Big[\kts{a}{\kPP{({\kss \bullet},{F})}{}}{{\kss \bullet}} -
\kts{a}{\kPP{}{({\kss \bullet},{F})}}{{\kss \bullet}}\Big]
=0\,,
\]
we automatically
sum only over $\kPV{}$ itself and, additionally, over two-dimensional faces
which are already clean.
\hspace*{\fill}$\Box$
\par
\section{The proof of the ${D}^2$-equations}\label{proof}
Here, we present the proof of Theorem \zitat{van}{cumbersome}.
It consists of a detailed, but
straightforward analysis of the differential map
$d_2:E_2^{1,1}\to E_2^{3,0}$.
\par
\neu{proof-e11}
{\em Describing $E_2^{1,1}$:}\\
According to the remark at the beginning of section \zitat{van}{supp},
the vector space
$E_2^{1,1}=H^1\big({\CM},{\cal \kT}^1\big)$ equals the kernel
\[
E_2^{1,1}=\ker\Big[
\oplus_{i_0<i_1}{D}^1(\kPM{i_0}\cap\kPM{i_1})
\longrightarrow \oplus_{i_0<i_1<i_2}
{D}^1(\kPM{i_0}\cap\kPM{i_1}\cap\kPM{i_2})\Big]\,.
\]
Denote by $\kPV{1},\dots,\kPV{{M}}$ the two-dimensional faces of ${\Delta}$ which
are no triangles;
each $\kPV{{k}}$ is contained in some three-dimensional faces
$\kPP{0}{{k}},\dots,\kPP{N_{k}}{{k}}$.
Note that certain $\kPP{}{}$'s might occur in more
than one of these lists. Nevertheless,
\[
E_2^{1,1}\;\cong\;
\oplus_{{k}=1}^{{M}} {D}^1\big(\kPV{{k}}\big)^{\oplusN_{k}}
\]
with the $i$-th summand in ${D}^1\big(\kPV{{k}}\big)^{\oplusN_{k}}$ being
identified with ${D}^1(\kPP{0}{{k}}\cap\kPP{i}{{k}})$; the remaining entries
in ${D}^1(\kPP{i_0}{{k}}\cap\kPP{i_1}{{k}})$ may be obtained
in the usual way as differences
from those of ${D}^1(\kPP{0}{{k}}\cap\kPP{i_1}{{k}})$ and
${D}^1(\kPP{0}{{k}}\cap\kPP{i_0}{{k}})$. On the other hand, if the
intersection $\kPM{i_0}\cap\kPM{i_1}$ is less than two-dimensional,
then ${D}^1(\kPM{i_0}\cap\kPM{i_1})=0$, anyway.\\
We choose the special three-dimensional face $\kPP{}{}(\kPV{{k}})$
mentioned in \zitat{van}{cumbersome} to be $\kPP{0}{{k}}$.
\par
\neu{proof-d2}
{\em Describing $d_2$:}\\
From \zitat{hodge}{spectral} we recall that
the double complex inducing the spectral sequence we are dealing with,
looks as follows:
\[
C^{p,q}=\oplus_{A\in [\kPM{i_0}\cap\dots\cap \kPM{i_p}]_q}
\,\span\big(\mbox{\rm cone}(A)\big) \; \mbox{ with }\;
d_I:C^{p,q}\to C^{p+1,q}\,,\; d_{II}:C^{p,q}\to C^{p,q+1}\,.
\]
We fix one of the two-dimensional faces $\kPV{{k}}$ and call it
$\kPV{}$. During \zitat{proof}{d2}, we abbreviate the three-dimensional faces
$\kPP{0}{{k}},\dots,\kPP{N_{k}}{{k}}$ containing $\kPV{{k}}=\kPV{}$
by $\kPP{0}{},\dots,\kPP{N}{}$. The index $i$ will be reserved for
these $\kPP{i}{}$, while $j\notin\{0,\dots,N\}$
points to those three-dimensional faces $\kPP{j}{}<{\Delta}$
belonging {\em not} to this list.\\
Assume that $\kPV{}$ is an ${m}$-gon with vertices $a^{\nu}$
(${\nu}\in{Z\!\!\!Z}/{m}{Z\!\!\!Z}$). Then, by \zitat{Tinv}{T1},
an element of ${D}^1(\kPV{})$ may be represented
as an ${m}$-tuple $(t_1,\dots,t_{m})\in{K}^{m}$ with $t_{\nu}$ being the
dilatation factor assigned to the edge $\overline{a^{\nu} a^{{\nu}+1}}<\kPV{}$.
In particular, we may start our tour through the double complex with
an
\[
x=(t^1,\dots,t^N)\in{D}^1\big(\kPV{}\big)^{\oplusN}\subseteq E_2^{1,1}
\;\mbox{ with each $t^i$ represented as }\;
t^i=(t^i_1,\dots,t^i_{m})\in{K}^{m}.
\]
The corresponding element $x\in C^{1,1}$ looks like
\[
x_{i_0i_1}\big(\overline{a^{\nu} a^{{\nu}+1}}\big)=
(t^{i_1}_{\nu}-t^{i_0}_{\nu})\cdot \kov{30}{10}{a^{\nu} a^{{\nu}+1}}
\in
\span(a^{\nu}, a^{{\nu}+1})
\hspace{1em}\mbox{ with }\;t^0_{\nu}:=0\,,
\]
and we have to walk through $C^{{\kss \bullet},{\kss \bullet}}$ along the following path:
\begin{center}
\unitlength=1.200000pt
\begin{picture}(200.00,60.00)(-40.00,0.00)
\put(-65.00,40.00){\makebox(0.00,0.00){Differential $d_2$:}}
\put(160.00,30.00){\makebox(0.00,0.00){$d_{II}$}}
\put(160.00,65.00){\makebox(0.00,0.00){$d_{I}$}}
\put(150.00,20.00){\vector(0,1){20.00}}
\put(150.00,60.00){\vector(1,0){20.00}}
\put(50.00,20.00){\makebox(0.00,0.00){$\mapsto$}}
\put(40.00,30.00){\makebox(0.00,0.00){$\uparrow$}}
\put(20.00,50.00){\makebox(0.00,0.00){$\uparrow$}}
\put(30.00,40.00){\makebox(0.00,0.00){$\mapsto$}}
\put(92.00,20.00){\makebox(0.00,0.00){$d_2(x)\in C^{3,0}$}}
\put(40.00,10.00){\makebox(0.00,0.00){$y$}}
\put(-2.00,40.00){\makebox(0.00,0.00){$x\in C^{1,1}$}}
\put(20.00,60.00){\circle{5.00}}
\put(20.00,20.00){\circle*{2.50}}
\put(40.00,20.00){\circle*{4.00}}
\put(60.00,20.00){\circle*{4.00}}
\put(20.00,40.00){\circle*{4.00}}
\put(40.00,40.00){\circle*{4.00}}
\put(60.00,40.00){\circle*{2.50}}
\put(20.00,60.00){\circle*{2.50}}
\put(40.00,60.00){\circle*{2.50}}
\put(60.00,60.00){\circle*{2.50}}
\end{picture}}
\vspace{-2ex}
\end{center}
The components of the image $d_{I}(x)\in C^{2,1}$ vanish unless exactly two
of the three indices belong to faces $\kPP{i}{}$ containing $\kPV{}$.
In this case, we obtain
\[
\renewcommand{\arraystretch}{1.3}
\begin{array}{rcl}
d_{I}(x)_{i_0i_1j_2}\big(\overline{a^{\nu} a^{{\nu}+1}}\big)&=&
x_{i_1j_2}\big(\overline{a^{\nu} a^{{\nu}+1}}\big)
-x_{i_0j_2}\big(\overline{a^{\nu} a^{{\nu}+1}}\big)
+x_{i_0i_1}\big(\overline{a^{\nu} a^{{\nu}+1}}\big)\\
&=& x_{i_0i_1}\big(\overline{a^{\nu} a^{{\nu}+1}}\big)\\
&=&(t^{i_1}_{\nu}-t^{i_0}_{\nu})\cdot \kov{30}{10}{a^{\nu} a^{{\nu}+1}}
\end{array}
\]
if $(\kPP{i_0}{}\cap \kPP{i_1}{})\cap\kPP{j_2}{}=
\kPV{}\cap\kPP{j_2}{}=\overline{a^{\nu} a^{{\nu}+1}}$.\\
Now, we lift this result to an element $y\in C^{2,0}$, i.e.\ we solve
the equation $d_{II}(y)=d_{I}(x)$. Obviously, the following $y$ does the job:
\[
\renewcommand{\arraystretch}{1.3}
y_{i_0i_1j_2}(a^{\nu}):=
\left\{\begin{array}{rl}
(t^{i_1}_{{\nu}-1}-t^{i_0}_{{\nu}-1}) \cdot a^{\nu} &
\mbox{if } \kPV{}\cap\kPP{j_2}{}=\overline{a^{{\nu}-1} a^{\nu}}\\
(t^{i_1}_{\nu}-t^{i_0}_{\nu}) \cdot a^{\nu} &
\mbox{if } \kPV{}\cap\kPP{j_2}{}=\overline{a^{\nu} a^{{\nu}+1}}
\end{array}\right.
\]
and $y_{\dots}({\kss \bullet}):=0$ for any other constellation. Its image
$d_{I}(y)\in C^{3,0}$ asks for quadrupels $(i_0,i_1,j_2,j_3)$
with still exactly two indices belonging to $\kPV{}$-solids.
Up to antisymmetric permutation of the four indices, we have
\[
\renewcommand{\arraystretch}{1.3}
d_{I}(y)_{i_0i_1j_2j_3}(a^{\nu})=
\left\{\begin{array}{ll}
(t^{i_1}_{{\nu}-1}-t^{i_0}_{{\nu}-1}-t^{i_1}_{\nu}+t^{i_0}_{\nu}) \cdot a^{\nu}
& \mbox{if } \kPV{}\cap\kPP{j_2}{}=\overline{a^{\nu} a^{{\nu}+1}}\,,\;
\kPV{}\cap\kPP{j_3}{}=\overline{a^{{\nu}-1}a^{\nu}}\\
(t^{i_1}_{{\nu}-1}-t^{i_0}_{{\nu}-1}) \cdot a^{\nu}
& \mbox{if } \kPV{}\cap\kPP{j_2}{}=\{a^{\nu}\}\,,\;
\kPV{}\cap\kPP{j_3}{}=\overline{a^{{\nu}-1}a^{\nu}}\\
-(t^{i_1}_{\nu}-t^{i_0}_{\nu}) \cdot a^{\nu}
& \mbox{if } \kPV{}\cap\kPP{j_2}{}=\overline{a^{\nu} a^{{\nu}+1}}\,,\;
\kPV{}\cap\kPP{j_3}{}=\{a^{\nu}\}
\end{array}\right.
\]
and zero for any other constellation.
The element $d_{I}(y)$ represents the cohomology class
\[
d_2(x)\in E_2^{3,0}=H^3\big({\cal \kT}^0\big)=
\oplus_{a\in{\Delta}}H^3\big({\cal \kT}^0(a)\big)\,.
\]
Hence, the only non-trivial components $d_2(x)(a\in{\Delta})$ occur for
$a=a^{\nu}\in\kPV{}$, and they look like $d_{I}(y)(a^{\nu})$ in
the formula above.
\par
\neu{proof-sing}
{\em Transfer from $H^3\big({\cal \kT}^0(a)\big)$
to singular cohomology:}\\
Let $a\in{\Delta}$ be an arbitrary vertex.
As already indicated in the proof of Lemma \zitat{van}{supp}, we have to
use the Mayer-Vietoris spectral sequence to describe the isomorphism
\[
H_2\big({\Delta}(a)^{(2)}\big)=H_3\big({\Delta}(a),{\Delta}(a)^{(2)}\big)
\stackrel{\sim}{\longrightarrow}
H_3\big({\cal \kT}_0(a)\big)
\]
with ${\cal \kT}_0(a)$ meaning
the complex built by the same recipe as
${\cal \kT}^0(a)$ in \zitat{van}{supp}, but using homology
\[
{D}_0\big(\kPM{i_0}\cap\dots\cap\kPM{i_p}\big)
:=H_0\big({\Delta}(a),\, \kPM{i_0}(a)\cap\dots\cap\kPM{i_p}(a)\big)
\]
instead of ${D}^0$.
Denoting by $C_q({\kss \bullet})$ the singular $q$-chains and abbreviating the vertex
figures $\kPM{i}(a)<{\Delta}(a)$ simply by $\kPMb{i}<\bar{{\Delta}}$,
we define
\[
K_{p,q}:=\oplus_{i_0\leq\dots\leqi_p}
\raisebox{0.5ex}{$C_q(\bar{{\Delta}})$}
\hspace{-0.1em}\Big/\hspace{-0.1em}
\raisebox{-0.7ex}{$C_q(\kPMb{i_0}\cap\dots\cap\kPMb{i_p})$}
\hspace{1em}\mbox{with}\hspace{1em}
\begin{array}[t]{rl}
d_{I}:&K_{p,q}\to K_{p-1,q}\\
d_{II}:&K_{p,q}\to K_{p,q-1}\,.
\end{array}
\]
The spectral sequence obtained by taking the vertical homology first
yields the complex ${\cal \kT}_0(a)$ as $E^1_{{\kss \bullet},0}$ and zero elsewhere.
The other one, beginning with the horizontal homology, has
$E^1_{0,{\kss \bullet}}$ as the only entries at the first level. They form
the complex
$\raisebox{0.3ex}{$C_{\kss \bullet}(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$\sum_i C_{\kss \bullet}(\bar{\kPM{i}})$}$
which is quasi\-isomorphic to
$\raisebox{0.3ex}{$C_{\kss \bullet}(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_{\kss \bullet}(\cup_i\kPMb{i})$}$.\\
Hence, the existence of the isomorphism promised above is clear. However,
we have to understand what the isomorphism really does with
$[{\bar{\kPF}}]\in H_3(\bar{{\Delta}},\cup_i\kPMb{i})$. To see this
we chase $[{\bar{\kPF}}]$ along the following diagram:
{\small
\[
\begin{array}{@{}c@{}c@{}c@{}c@{}c@{}c@{}c@{}c@{}}
H_3(\bar{{\Delta}},\cup_i\kPMb{i})\\
|\\
\raisebox{0.3ex}{$C_3(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$\Sigma_i C_3(\bar{\kPM{i}})$}
&\leftarrow&
\oplus_{i_0}\raisebox{0.3ex}{$C_3(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_3(\kPMb{i_0})$}\\
\downarrow&&\downarrow\\
\dots&\leftarrow&
\oplus_{i_0}\raisebox{0.3ex}{$C_2(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_2(\kPMb{i_0})$}
&\leftarrow&
\oplus_{i_0\leqi_1}\raisebox{0.3ex}{$C_2(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_2(\kPMb{i_0}\cap\kPMb{i_1})$}\\
&&\downarrow&&\downarrow\\
&&\dots&\leftarrow&
\oplus_{i_0\leqi_1}\raisebox{0.3ex}{$C_1(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_1(\kPMb{i_0}\cap\kPMb{i_1})$}
&\leftarrow&
\oplus_{i_0\leqi_1\leqi_2}\raisebox{0.3ex}{$C_1(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_1(\kPMb{\underline{i}})$}\\
&&&&\downarrow&&\downarrow\\
&&&&\dots&\leftarrow&
\oplus_{i_0\leqi_1\leqi_2}\raisebox{0.3ex}{$C_0(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_0(\kPMb{\underline{i}})$}
&\leftarrow\dots\\
&&&&&&\downarrow\\
&&&&&&\dots&\leftarrow\dots
\end{array}
\vspace{-2ex}
\]
}
\par
Fixing an arbitrary index $i_0:=0$, we arrive at the third row with
$[\partial{\bar{\kPF}}]\in \raisebox{0.3ex}{$C_2(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}\raisebox{-0.3ex}{$C_2(\kPMb{0})$}$.
Let $\partial{\bar{\kPF}}=\kPMb{1}\cup\dots\cup\kPMb{l}$
and assume that the orientation of the $\kPMb{i}$ is inherited from some
orientation of ${\bar{\kPF}}$. Then, a possible lift to the right is
\[
-[\kPMb{i}]\,\in\; \raisebox{0.5ex}{$C_2(\bar{{\Delta}})$}
\hspace{-0.1em}\Big/\hspace{-0.1em}
\raisebox{-0.7ex}{$C_2(\kPMb{0}\cap\kPMb{i})$}\,,
\quadi=1,\dots,l\,.
\]
Applying the vertical boundary operator and lifting again to the right,
we obtain
\[
[\kPMb{i}\cap\kPMb{j}]\,\in\;
\raisebox{0.5ex}{$C_1(\bar{{\Delta}})$}
\hspace{-0.1em}\Big/\hspace{-0.1em}
\raisebox{-0.7ex}{$C_1(\kPMb{0}\cap\kPMb{i}\cap\kPMb{j})$}
\]
with $(\kPMb{i},\kPMb{j})$ running through the pairs of
mutually adjacent faces of ${\bar{\kPF}}$ with $i<j$.
Our convention is that the edges $[\kPMb{i}\cap\kPMb{j}]$
inherit their orientation from the first argument, i.e.\
$[\kPMb{i}\cap\kPMb{j}]=-[\kPMb{j}\cap\kPMb{i}]$.
\vspace{1ex}
\[
\begin{array}{@{}c@{}c@{}c@{}c@{}c@{}}
\dots\leftarrow&
\oplus_{i_0\leqi_1\leqi_2}\raisebox{0.3ex}{$C_1(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_1(\kPMb{\underline{i}})$}\\
&\downarrow\\
\dots\leftarrow&
\oplus_{i_0\leqi_1\leqi_2}\raisebox{0.3ex}{$C_0(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_0(\kPMb{\underline{i}})$}
&\leftarrow&
\oplus_{i_0\leqi_1\leqi_2\leqi_3}\raisebox{0.3ex}{$C_0(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_0(\kPMb{\underline{i}})$}\\
&\downarrow&&\downarrow\\
&\dots&\leftarrow&
\oplus_{i_0\leqi_1\leqi_2\leqi_3} H_0(\bar{{\Delta}},\,\kPMb{\underline{i}})&
-\!\!\!- H_3\big[\dots\to
\oplus_{\underline{i}} H_0(\bar{{\Delta}},\,\kPMb{\underline{i}})\to\dots\big]
\end{array}
\]
It is easy to apply the boundary operator to the edges
$[\kPMb{i}\cap\kPMb{j}]$; but for doing the last lifting to
$\oplus_{i_0\leqi_1\leqi_2\leqi_3}\raisebox{0.3ex}{$C_0(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_0(\kPMb{\underline{i}})$}$,
we have to
introduce for each vertex ${\bar{\kPX}}\in{\bar{\kPF}}$ an auxillary function $\varphi_{X}$.
Its arguments are triples of ${\bar{\kPF}}$-facets containing ${\bar{\kPX}}$, and it is
determined by the following properties:
\begin{itemize}\vspace{-2ex}
\item[(i)]
$\varphi_{X}(\kPMb{i},\kPMb{j},\kPMb{k})$ is antisymmetric in its
arguments.
\item[(ii)]
If any two of the arguments
intersect only in $\{{\bar{\kPX}}\}$,
then $\varphi_{X}(\kPMb{i},\kPMb{j},\kPMb{k}):=0$.
\begin{center}
\unitlength=0.5pt
\begin{picture}(154.00,151.00)(29.00,673.00)
\put(105.00,753.00){\line(-4,5){46.00}}
\put(105.00,753.00){\line(3,-5){46.00}}
\put(105.00,753.00){\line(-1,0){76.00}}
\put(105.00,753.00){\line(-1,-3){27.00}}
\put(105.00,753.00){\line(3,4){52.00}}
\put(105.00,753.00){\line(4,-1){78.00}}
\put(98.00,762.00){\line(5,1){16.00}}
\put(91.00,771.00){\line(5,1){32.00}}
\put(84.00,780.00){\line(5,1){48.00}}
\put(77.00,789.00){\line(5,1){64.00}}
\put(121.00,759.00){\line(-3,-5){7.00}}
\put(132.00,746.00){\line(-3,-5){11.50}}
\put(143.00,743.00){\line(-3,-5){16.00}}
\put(154.00,740.00){\line(-3,-5){20.50}}
\put(88.00,753.00){\line(3,-5){11.00}}
\put(77.00,753.00){\line(3,-5){18.00}}
\put(66.00,753.00){\line(3,-5){25.00}}
\put(55.00,753.00){\line(3,-5){32.00}}
\put(102.00,812.00){\makebox(0,0)[cc]{$\kPMb{k}$}}
\put(161.00,711.00){\makebox(0,0)[cc]{$\kPMb{j}$}}
\put(50.00,715.00){\makebox(0,0)[cc]{$\kPMb{i}$}}
\end{picture}
\end{center}
\item[(iii)]
Denote by $u({\bar{\kPX}})$ the number of two-dimensional ${\bar{\kPF}}$-facets meeting in ${\bar{\kPX}}$.
Then, depending on $u({\bar{\kPX}})$ and on the fact if there are ``isolated''
arguments or not, we distinguish between three cases:
\begin{center}
\hspace*{1em}
\unitlength=0.6pt
\begin{picture}(151.00,203.00)(61.00,620.00)
\put(135.00,750.00){\line(-2,1){94.00}}
\put(135.00,750.00){\line(1,1){76.00}}
\put(135.00,750.00){\line(0,-1){86.00}}
\put(83.00,776.00){\line(3,1){116.00}}
\put(99.00,768.00){\line(3,1){80.00}}
\put(115.00,760.00){\line(3,1){44.00}}
\put(83.00,776.00){\line(2,-3){52.00}}
\put(109.00,762.00){\line(2,-3){26.00}}
\put(162.00,777.00){\line(-1,-2){27.00}}
\put(189.00,804.00){\line(-1,-2){54.00}}
\put(81.00,732.00){\makebox(0,0)[cc]{$\kPMb{i}$}}
\put(176.00,727.00){\makebox(0,0)[cc]{$\kPMb{j}$}}
\put(133.00,824.00){\makebox(0,0)[cc]{$\kPMb{k}$}}
\put(135.00,643.00){\makebox(0,0)[cc]{$u({\bar{\kPX}})=3$}}
\put(135.00,615.00){\makebox(0,0)[cc]
{$\varphi_{X}(\kPMb{i},\kPMb{j},\kPMb{k}):=1$}}
\end{picture}
\hfill
\begin{picture}(151.00,203.00)(61.00,620.00)
\put(105.00,753.00){\line(-4,5){46.00}}
\put(105.00,753.00){\line(3,-5){46.00}}
\put(105.00,753.00){\line(-1,0){76.00}}
\put(105.00,753.00){\line(-1,-3){27.00}}
\put(105.00,753.00){\line(3,4){52.00}}
\put(105.00,753.00){\line(4,-1){78.00}}
\put(121.00,759.00){\line(-3,-5){7.00}}
\put(132.00,746.00){\line(-3,-5){11.50}}
\put(143.00,743.00){\line(-3,-5){16.00}}
\put(154.00,740.00){\line(-3,-5){20.50}}
\put(88.00,753.00){\line(3,-5){11.00}}
\put(77.00,753.00){\line(3,-5){18.00}}
\put(66.00,753.00){\line(3,-5){25.00}}
\put(55.00,753.00){\line(3,-5){32.00}}
\put(100.00,739.00){\line(5,2){10.00}}
\put(97.00,729.00){\line(5,2){17.00}}
\put(94.00,719.00){\line(5,2){24.00}}
\put(91.00,709.00){\line(5,2){31.50}}
\put(88.00,699.00){\line(5,2){39.00}}
\put(161.00,711.00){\makebox(0,0)[cc]{$\kPMb{k}$}}
\put(50.00,715.00){\makebox(0,0)[cc]{$\kPMb{i}$}}
\put(110.00,685.00){\makebox(0,0)[cc]{$\kPMb{j}$}}
\put(235.00,643.00){\makebox(0,0)[cc]{$u({\bar{\kPX}})\geq 4$}}
\put(125.00,615.00){\makebox(0,0)[cc]
{$\varphi_{X}(\kPMb{i},\kPMb{j},\kPMb{k}):=2/u({\bar{\kPX}})$}}
\end{picture}
\hfill
\begin{picture}(151.00,203.00)(61.00,620.00)
\put(105.00,753.00){\line(-4,5){46.00}}
\put(105.00,753.00){\line(3,-5){46.00}}
\put(105.00,753.00){\line(-1,0){76.00}}
\put(105.00,753.00){\line(-1,-3){27.00}}
\put(105.00,753.00){\line(3,4){52.00}}
\put(105.00,753.00){\line(4,-1){78.00}}
\put(98.00,762.00){\line(5,1){16.00}}
\put(91.00,771.00){\line(5,1){32.00}}
\put(84.00,780.00){\line(5,1){48.00}}
\put(77.00,789.00){\line(5,1){64.00}}
\put(88.00,753.00){\line(3,-5){11.00}}
\put(77.00,753.00){\line(3,-5){18.00}}
\put(66.00,753.00){\line(3,-5){25.00}}
\put(55.00,753.00){\line(3,-5){32.00}}
\put(100.00,739.00){\line(5,2){10.00}}
\put(97.00,729.00){\line(5,2){17.00}}
\put(94.00,719.00){\line(5,2){24.00}}
\put(91.00,709.00){\line(5,2){31.50}}
\put(88.00,699.00){\line(5,2){39.00}}
\put(102.00,812.00){\makebox(0,0)[cc]{$\kPMb{k}$}}
\put(50.00,715.00){\makebox(0,0)[cc]{$\kPMb{i}$}}
\put(110.00,685.00){\makebox(0,0)[cc]{$\kPMb{j}$}}
\put(125.00,615.00){\makebox(0,0)[cc]
{$\varphi_{X}(\kPMb{i},\kPMb{j},\kPMb{k}):=1/u({\bar{\kPX}})$}}
\end{picture}
\hspace*{1em}
\end{center}
\vspace{-1ex}\end{itemize}
Now, it is not difficult to check that a possible lifting of the tuple
$\big(\partial[\kPMb{i}\cap\kPMb{j}]\big)_{i<j}$
to the vector space
$\oplus_{i_0\leq\dots\leqi_3}\raisebox{0.3ex}{$C_0(\bar{{\Delta}})$}
\hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.3ex}{$C_0(\kPMb{\underline{i}})$}$
is given by
\[
\varphi_{X}(\kPMb{i},\kPMb{j},\kPMb{k})\cdot [{\bar{\kPX}}]
\;\in\;
\raisebox{0.7ex}{$C_0(\bar{{\Delta}})$} \hspace{-0.1em}\big/\hspace{-0.1em}
\raisebox{-0.7ex}{$C_0(\kPMb{0}\cap\kPMb{i}\cap\kPMb{j}\cap\kPMb{k})$}
\]
with $(\kPMb{i},\kPMb{j},\kPMb{k})$ running through all triples of
facets of ${\bar{\kPF}}$ with $\kPMb{i}\cap\kPMb{j}\cap\kPMb{k}=\{{\bar{\kPX}}\}$
and $i<j<k$. The projection to
$H_0(\bar{{\Delta}},\,\kPMb{0}\cap\kPMb{i}\cap\kPMb{j}\cap\kPMb{k})$
does not change the shape of the element
$\varphi_{X}(\kPMb{i},\kPMb{j},\kPMb{k})\cdot [{\bar{\kPX}}]$. However, if
${\bar{\kPX}}\in\kPMb{0}$, then the element $[{\bar{\kPX}}]$ spanning the whole homology
group vanishes, anyway.\\
Note that the final result of the previous construction cannot depend on
the choice of $\kPMb{0}$ made in the very beginning.
In particular, one might exploit this freedom to take for $\kPMb{0}$ one
of the ${\bar{\kPF}}$-faces, or to do exactly the opposite.
\par
\neu{proof-interpret}
{\em Interpreting $d_2(x)(a^{\nu})$ inside
$H^3\big({\Delta}(a^{\nu}),{\Delta}(a^{\nu})^{(2)}\big) $:}\\
We apply the previous calculation to show how $d_2(x)(a^{\nu})$ acts on
a homology class
$[{\bar{\kPF}}]=[{F}(a^{\nu})]\in H_3\big({\Delta}(a^{\nu}),{\Delta}(a^{\nu})^{(2)}\big)$ induced
by a four-dimensional face ${F}<{\Delta}$ containing $a^{\nu}$:
Unless $\kPV{}\subseteq{F}$, we have
$\big\langle d_2(x)(a^{\nu}), [{\bar{\kPF}}]\big\rangle=0$. However,
assuming $\kPV{}\subseteq{F}$, then ${F}$ contains exactly two faces
$\kPP{i}{},\kPP{k}{}$ with common facet $\kPV{}$, and the result is
\[
\big\langle d_2(x)(a^{\nu}), [{\bar{\kPF}}]\big\rangle \;=\;
\big(t^{k}_{{\nu}-1}-t^{k}_{\nu}\big) -
\big(t^{i}_{{\nu}-1}-t^{i}_{\nu}\big)\,.
\]
{\em Proof:}
Using the notation
of \zitat{proof}{d2}, we select the first $\kPV{}$-solid
$\kPP{0}{}=\kPP{}{}(\kPV{})$ as the face inducing the vertex figure
$\kPMb{0}<\bar{{\Delta}}={\Delta}(a^{\nu})$ being fixed in \zitat{proof}{sing}.
Then, the only quadrupels having a chance to produce a non-trivial entry
in both steps are
\[
[0,i,j_2,j_3] \hspace{1em} \mbox{with }
\renewcommand{\arraystretch}{1.3}
\begin{array}[t]{l@{\hspace{0.5em}}l}
\bullet&
i\in\{0,\dots,N\}
\hspace{0.8em}\mbox{and}\hspace{0.8em}
\kPP{i}{},\,\kPP{j_2}{},\,\kPP{j_3}{}<{F} \\
\bullet&
\kPV{}\cap\kPP{j_2}{},\, \kPV{}\cap\kPP{j_3}{}=
\{a^{\nu}\},\; \overline{a^{{\nu}-1}a^{\nu}},\; \mbox{ or }\; \overline{a^{{\nu}}a^{{\nu}+1}}\\
\bullet&
\kPV{}\cap\kPP{j_2}{}\cap\kPP{j_3}{}=\{a^{\nu}\}\,.
\end{array}
\]
Focusing on the vertex figures
$\bar{\kPV{}}\leq\kPPb{i}{}\leq{\bar{\kPF}}$ at $a^{\nu}$, we see that
$\kPPb{i}{}$ is a polygon with
$\bar{\kPV{}}=\overline{{a}^{{\nu}-1}{a}^{{\nu}+1}}$ as one of its edges.
While $\bar{\kPV{}}\cap\kPPb{j_2}{}\cap\kPPb{j_3}{}=\emptyset$, the
result of \zitat{proof}{sing} implies that the
intersection $\kPPb{i}{}\cap\kPPb{j_2}{}\cap\kPPb{j_3}{}$ has to be
some point ${\bar{\kPX}}\neq\,\overline{a}^{{\nu}-1},\,\overline{a}^{{\nu}+1}$.
\begin{center}
\unitlength=0.6pt
\begin{picture}(220.00,160.00)(54.00,670.00)
\put(155.00,755.00){\line(-5,-1){98.00}}
\put(155.00,755.00){\line(-3,-5){49.00}}
\put(155.00,755.00){\line(2,-3){54.00}}
\put(155.00,755.00){\line(5,-1){83.00}}
\put(239.00,738.00){\line(3,5){35.00}}
\put(274.00,797.00){\line(-5,4){51.00}}
\put(223.00,838.00){\line(-4,-5){68.00}}
\put(56.00,735.00){\line(4,-5){49.50}}
\put(104.50,673.00){\line(1,0){104.50}}
\put(95.00,656.00){\makebox(0,0)[cc]{$\bar{a}^{{\nu}-1}$}}
\put(213.00,659.00){\makebox(0,0)[cc]{$\bar{a}^{{\nu}+1}$}}
\put(140.00,775.00){\makebox(0,0)[cc]{${\bar{\kPX}}$}}
\put(220.00,779.00){\makebox(0,0)[cc]{$\kPPb{j_2}{}$}}
\put(104.00,720.00){\makebox(0,0)[cc]{$\kPPb{j_3}{}$}}
\put(159.00,694.00){\makebox(0,0)[cc]{$\kPPb{i}{}$}}
\end{picture}
\end{center}
Hence, fixing $\kPPb{i}{}$ and
choosing the ordering of the ${\Delta}$-faces and their corresponding indices
well, we obtain contributions to
$\big\langle d_2(x)(a^{\nu}), [{\bar{\kPF}}]\big\rangle$ only from the arguments
$\kPPb{j_2}{}$ and $\kPPb{j_3}{}$ running through the two-dimensional
${\bar{\kPF}}$-faces fitting in one of the following cases:
\begin{itemize}\vspace{-2ex}
\item[(i)]
$\kPPb{j_2}{}$ has a common edge $\overline{a^{{\nu}+1}{X}}$ with $\kPPb{i}{}$.
Then, if $\kPPb{j_3}{}\ni{\bar{\kPX}}$ is adjacent to one of them, we obtain twice
\[
d_2(x)_{0ij_2j_3}(a^{\nu})\cdot
\varphi_{X}(\kPMb{i},\kPMb{j_2},\kPMb{j_3})\;=\;
-t^{i}_{\nu} \cdot 2/u({\bar{\kPX}})\,.
\]
Moreover, there are $\big(u({\bar{\kPX}})-4\big)$ possibilities such that
$\kPPb{j_3}{}\ni{\bar{\kPX}}$ is ``isolated''. Each constellation yields
the contribution
$-t^{i}_{\nu}\cdot 1/u({\bar{\kPX}})$.
\item[(ii)]
$\kPPb{j_3}{}$ has a common edge $\overline{{X} a^{{\nu}-1}}$ with $\kPPb{i}{}$.
Then, as in (i), we obtain twice $t^{i}_{{\nu}-1}\cdot 2/u({\bar{\kPX}})$
and $\big(u({\bar{\kPX}})-4\big)$-times $t^{i}_{{\nu}-1}\cdot 1/u({\bar{\kPX}})$.
\item[(iii)]
If both $\overline{a}^{{\nu}+1}\notin\kPPb{j_2}{}$ and
$\overline{a}^{{\nu}-1}\notin\kPPb{j_3}{}$, then the result of
\zitat{proof}{d2} shows that this case contributes
nothing to $\big\langle d_2(x)(a^{\nu}), [{\bar{\kPF}}]\big\rangle$.
\vspace{-1ex}\end{itemize}
Altogether, this adds up to $(t^{i}_{{\nu}-1}-t^{i}_{\nu})$,
and we should finally remark that the exceptional cases ``$u({\bar{\kPX}})=3$''
and ``$\kPPb{i}{}$ is a triangle'' yield the same result. In the latter
situation, the cases (i) and (ii) might overlap.
\hfill$(\Box)$
\par
Finally, we should remark that it is exactly the differences
$t^{i}_{{\nu}-1}-t^{i}_{\nu}$ which are called
$\kts{a^{\nu}}{\kPP{i}{}}{\kPV{}}$ in \zitat{van}{cumbersome}.
The equations $(2)_{(\kPV{},\kPP{}{})}$ of the theorem say nothing
else than that these $s$-variables come from some $t$'s satisfying the
equations for Minkowski summands of $\kPV{}$ as mentioned in \zitat{Tinv}{T1}.
\hspace*{\fill}$\Box$
\par
\section{Applications to deformation theory}\label{deform}
\neu{deform-setup}
Let $N,M$ be two finitely generated, free abelian groups which are mutually
dual;
denote by $N_{I\!\!R}, M_{I\!\!R}$ the vector spaces obtained by extending the scalars.
Each polyhedral, rational cone $\sigma\subseteq N_{I\!\!R}$ with apex in $0$
gives rise to an affine toric variety $X_\sigma:=\mbox{\rm Spec}\,{\,I\!\!\!\!C}[\sigma^{\scriptscriptstyle\vee}\cap M]$.
It comes with an action of the torus
$N_{\,I\!\!\!\!C}\otimes_{\,I\!\!\!\!C}\C^\ast=\mbox{\rm Spec}\, {\,I\!\!\!\!C}[M]$, which leads to a stratification into orbits which are parametrized by the
faces of $\sigma$. We refer to \cite{Da} for more details.\\
In particular, the trivial face $\sigma\leq\sigma$ corresponds to a unique fixed point
$\mbox{\rm orb}(\sigma)=0$ of
the torus action. It is the ``most singular'' point of $X_\sigma$, and we are
going to study the deformation theory of the germ $(X_\sigma,0)$.
\par
The point that makes toric varieties so exciting is the fact that many
algebro-geometric properties of $X_\sigma$ (or its non-affine
generalizations) translate directly into combinatorial properties of cones
and their relation to the lattice structure $N\subseteq N_{I\!\!R}$. A first
example of such a translation can be seen in \zitat{hodge}{def}. We will need in the future
the following two further examples of such translations:
\begin{itemize}\vspace{-2ex}
\item
$X_\sigma$ is Gorenstein if and only if
$\sigma$ is the cone over a compact, convex
lattice polytope ${\Delta}\subseteq{I\!\!R}^n$ sitting in an affine hyperplane of
height one.
This means that $N={Z\!\!\!Z}^n\times{Z\!\!\!Z}$, and
${\Delta}$ is a polytope with vertices in ${Z\!\!\!Z}^n\times\{1\}$.
\item
$X_\sigma$ is, additionally, smooth in codimension two iff the edges of ${\Delta}$
do not contain any interior lattice points.
\vspace{-1ex}\end{itemize}
\par
\neu{deform-Ti}
If $X=\mbox{\rm Spec}\,A$ is an affine algebraic variety, then the cohomology of the cotangent
complex produces $A$-modules $T^{k}_X$ which play an important role in deformation
theory: $T^0_X$ describes infinitesimal automorphisms, $T^1$ describes infinitesimal
deformations, and $T^2_X$ contains the obstructions for extending
deformations to larger base spaces.
See \cite{BeCh} for a nice survey, or \cite{Loday} for the details.\\
In the case that $X=X_\sigma$ is a toric variety, the ring
$A={\,I\!\!\!\!C}\,[\sigma^{\scriptscriptstyle\vee}\cap M]$ as well as the modules $T^{k}_X$ are
$M$-graded. It is possible to obtain combinatorial formulas for the
homogeneous pieces $T^{k}_X(-R)$ with $R\in M$. This has been done in
\cite{AQ}, and we recall the result:
\par
Assume we are given a rational, polyhedral cone
$\sigma=\langle a^1,\dots,a^{m}\rangle \subseteq N_{{I\!\!R}}$
with $a^1,\dots,a^{m}\in N$ denoting its {\em primitive
fundamental generators},
i.e.\ none of the $a^{\nu}$ is a proper multiple of an element of $N$.
The dual cone is
$\sigma^{{\scriptscriptstyle\vee}}:= \{ r\in M_{{I\!\!R}}\,|\; \langle \sigma,\,r\rangle \geq 0\}
\subseteq M_{{I\!\!R}}$. For any degree $R\in M$ and
face $\tau\leq\sigma$ we introduce a special subset of lattice points of $\sigma^{{\scriptscriptstyle\vee}}$:
\[
K_\tau^R:= \sigma^{\scriptscriptstyle\vee} \cap \big(R-\mbox{\rm int}\, \tau^{\scriptscriptstyle\vee}\big)\cap M
\subseteq (\sigma^{\scriptscriptstyle\vee}\cap M)\,.
\]
In particular, $K_0^R=\sigma^{\scriptscriptstyle\vee}\cap M$, whereas $K_\sigma^R$ consists of a finite
set of lattice points.
For an arbitrary subset $K \subset M$ we set:
\[
\ko{\mbox{\rm Hom}}\,(K,{\,I\!\!\!\!C}):= \big\{f:K\to{\,I\!\!\!\!C}\,\big|\;
f(r)+f(s)=f(r+s) \;\mbox{ if }\; r,s,r+s\in K\big\}\;.
\]
For each given $R \in M$, these sets give rise to a cohomological system
$\ko{\mbox{\rm Hom}}\,\big(K^R_{\kss \bullet},{\,I\!\!\!\!C}\big)$ on the ``affine'' fan ${\langle\sigma\rangle}$.
\par
{\bf Theorem:}
(cf.\ \cite{AQ}, (5.3))\quad
{\em
For ${k}\leq 2$, one has
$$T^{k}_X(-R)= H^{k}\big({\langle\sigma\rangle},\,\ko{\mbox{\rm Hom}}\,(K^R_{\kss \bullet},{\,I\!\!\!\!C})\big)\;.$$
Moreover, if either ${k} \leq 1$, or if ${k}=2$ and $X_\sigma$ is Gorenstein in
codimension two, then
$$T^{k}_X(-R)=H^{k}\big({\langle\sigma\rangle},\,\span_{\,I\!\!\!\!C}(K^R_{\kss \bullet})^\ast\big)\;.$$}
\par
{\bf Remarks:}
1) There is always a natural homomorphism of cohomological systems
$\,\span_{\,I\!\!\!\!C}(K^R_{\kss \bullet})^\ast \to \ko{\mbox{\rm Hom}}\,(K^R_{\kss \bullet},{\,I\!\!\!\!C})$, but in general their
cohomology groups are different. The second part of the theorem thus gives a
condition under which we can replace the complicated system $\ko{\mbox{\rm Hom}}\,(K^R_{\kss \bullet},{\,I\!\!\!\!C})\big)$
by a slightly simpler system of {\em vector spaces}.
\vspace{1ex}\\
2) The module structure of $T^{k}$ is the natural one: If
$x^s\in{\,I\!\!\!\!C}[\sigma^{\scriptscriptstyle\vee}\cap M]$, then the multiplication with $x^s$ is
obtained from the map $T^{k}_X(-R)\to T^{k}_X(-R+s)$ provided by the inclusions
$K^{R-s}_\tau\subseteq K^R_\tau$.
\vspace{1ex}\\
3) The property {\em Gorenstein in codimension two} translates into the following
condition for the cone: For every two-dimensional face
$\langle a^{\nu},a^{\mu}\rangle<\sigma$ there is an $R_{{\nu}{\mu}}\in M$
with $\langle a^{\nu},R_{{\nu}{\mu}}\rangle = \langle a^{\mu},R_{{\nu}{\mu}}\rangle =1$.
\par
\neu{deform-leq1}
Let ${\Delta}\subseteq{I\!\!R}^n$ be a lattice polytope; via $\sigma:=\mbox{\rm cone}({\Delta})$
it gives rise to a {\em toric Gorenstein singularity} $X:=X_{\Delta}$.
For this special case, we are going to
explain the relations between the vector spaces $T^{k}_X(-R)$ and the
coarse ${D}$-invariants ${D}^{k}$ defined in \S \ref{Tinv}.
\vspace{1ex}\\
If $a^1,\dots,a^{m}\in{Z\!\!\!Z}^n$ denote the vertices of ${\Delta}$,
then $a^{\nu}:=(a^{\nu},1)\in N$ are the fundamental generators of $\sigma$.
Moreover, there is a special degree $R^\ast:=[\underline{0},1]\in M$; it recovers
the polytope from the cone via ${\Delta}=\sigma\cap [R^\ast=1]$.
\par
{\bf Proposition:}
{\em
Let ${\Delta}$ and $X:=X_{\Delta}$ be as before. If $R\in M$ is a degree such that
$R\leq 1$ holds everywhere on ${\Delta}$, then
${\Delta}\cap [R=1]$ is a face of ${\Delta}$ and
\[
T^{k}_X(-R)= {D}^{k}\Big({\Delta}\cap [R=1]\Big)
\hspace{1em}\mbox{for}\hspace{0.8em} {k}\leq 2.
\vspace{-3ex}
\]
}
\par
{\bf Proof:}
The reader should convince her/himself from the fact that
the property $R \leq 1$ in ${\Delta}$ imlpies
\[
\span_{\,I\!\!\!\!C}(K^R_\tau)=\left\{\begin{array}{ll}
\tau^\bot & \mbox{if } \tau\leq \mbox{\rm cone}({\Delta}\cap [R=1])\\
0 & \mbox{otherwise}\,.
\end{array}\right.
\vspace{-3ex}
\]
The claim then follows from Theorem \zitat{deform}{Ti}. \hspace*{\fill}$\Box$
\par
\neu{deform-loc}
It is possible to describe $T^1_X(-R)$
in the Gorenstein case also for degrees with $R\not\leq 1$ on ${\Delta}$.
However, in the following three sections of the present paper,
we look for sufficient
conditions forcing $T^1_X(-R)$ and $T^2_X(-R)$ to vanish for those $R$.
\par
Assume that $\sigma=\langle a^1,\dots,a^{m}\rangle \subseteq N_{I\!\!R}$
is a rational, polyhedral cone as in \zitat{deform}{Ti}.
For any degree $R\in M$, we define another
homological system $V^R_{\kss \bullet} \supseteq\span_{\,I\!\!\!\!C}(K^R_{\kss \bullet})$ on ${\langle\sigma\rangle}$ by
\[
\renewcommand{\arraystretch}{1.1}
V^R_\tau:=\bigcap_{a^{\nu}\in\tau} V^R_{a^{\nu}}
\hspace{2em}\mbox{with}\hspace{1.5em}
V^R_{a^{\nu}}:=\span_{\,I\!\!\!\!C}(K^R_{a^{\nu}})=
\;\left\{
\begin{array}{@{}ll}
M_{\,I\!\!\!\!C} & \mbox{if } \,\langle a^{\nu},R\rangle\geq 2\\
(a^{\nu})^\bot & \mbox{if } \,\langle a^{\nu},R\rangle=1\\
0 & \mbox{if } \,\langle a^{\nu},R\rangle\leq 0\,.
\end{array}\right.
\vspace{-2ex}
\]
\par
Let $X_\sigma$ be {\em smooth in codimension two}, i.e.\
whenever $\langle a^{\nu},a^{\mu}\rangle<\sigma$ is a two-dimensional face,
then the set $\{a^{\nu},a^{\mu}\}$ may be extended to a ${Z\!\!\!Z}$-basis of the
whole lattice $N$. In particular, for any $R\in M$, we have
$V^R_{\langle a^{\nu},a^{\mu}\rangle} =
\span_{\,I\!\!\!\!C}(K^R_{\langle a^{\nu},a^{\mu}\rangle})$
for these faces. Hence, for $X_{sigma}$ smooth in codimension two one has
$$T^1_X(-R)=H^1\big({\langle\sigma\rangle}, \,(V^R_{\kss \bullet})^\ast\big) .$$
\par
{\bf Definition:}
If $X_\sigma$ is smooth in codimension two, then we define the
{\em local contribution} of
a three-dimensional face $\tau\leq\sigma$ to $T^2_X(-R)$ as
\[
T^2_{\tau,\,\mbox{\footnotesize loc}}(-R)
\;:=\;
\left(
\raisebox{1ex}{
$V^R_\tau$
}\hspace{-0.5em}\Big/\hspace{-0.5em} \raisebox{-1ex}{
$\span_{\,I\!\!\!\!C} K^R_\tau$}\right)^\ast
=
\left(
\raisebox{1ex}{
$\bigcap_{a^{\nu}\in\tau}\big(\span_{\,I\!\!\!\!C} K^R_{a^{\nu}}\big)$
}\hspace{-0.5em}\Big/\hspace{-0.5em} \raisebox{-1ex}{
$\span_{\,I\!\!\!\!C} \big(\bigcap_{a^{\nu}\in\tau}K^R_{a^{\nu}}\big)$}
\right)^\ast
\,.
\]
\par
If $\dim \sigma=3$ itself, then Theorem \zitat{deform}{Ti} tells us that
$T^2_X(-R)=T^2_{\sigma,\,\mbox{\footnotesize loc}}(-R)$. Moreover,
for the general case, we obtain the straightforward
\par
{\bf Proposition:}
{\em
Let $X_\sigma$ be smooth in codimension two. If there are no local
contributions from three-dimensional faces to $T^2_X(-R)$
(i.e.\ if $\,T^2_X$ sits in codimension at least four), then
$$T^2_X(-R)=H^2\big({\langle\sigma\rangle},\,(V^R_{\kss \bullet})^\ast\big)\; .$$
}
\par
{\bf Application:}
If the three-dimensional faces of $\sigma$ are either smooth (generated by a
part of a ${Z\!\!\!Z}$-basis of $N$) or isomorphic to cones over unit squares in
${Z\!\!\!Z}^2$, then $X_\sigma$ is a {\em conifold} in codimension three,
i.e.\ it is smooth in codimension two
and has at most $A_1$-singularities in codimension three. In particular,
for thoses cones,
the assumption of the previous proposition is satisfied for
every multidegree $R\in M$.
\par
{\bf Example:}
To get some familiarity with the sets $K^R_\tau$, we explain the vanishing of the local
contributions for conifolds on the combinatorial level.. Let $\tau$ be the cone over
a unit square.
Unless $R$ is positive at the four vertices of this square, the space
$\bigcap_{a^{\nu}\in\tau}\big(\span_{\,I\!\!\!\!C} K^R_{a^{\nu}}\big)$ vanishes, anyway.
Now, focusing on these four positive
values, there are only the following possibilities:
\begin{center}
\hspace*{\fill}
\unitlength=0.4pt
\begin{picture}(80.00,80.00)(30.00,730.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(20.00,820.00){\makebox(0,0)[cc]{${\scriptstyle \langle a^4,R\rangle=1}$}}
\put(122.00,820.00){\makebox(0,0)[cc]{${\scriptstyle\langle a^3,R\rangle=1}$}}
\put(20.00,720.00){\makebox(0,0)[cc]{${\scriptstyle \langle a^1,R\rangle=1}$}}
\put(122.00,720.00){\makebox(0,0)[cc]{${\scriptstyle\langle a^2,R\rangle=1}$}}
\put(70.00,770.00){\makebox(0,0)[cc]{$\tau$}}
\end{picture}
\hspace*{\fill}
\unitlength=0.4pt
\begin{picture}(80.00,80.00)(30.00,730.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$1$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(70.00,770.00){\makebox(0,0)[cc]{$\tau$}}
\end{picture}
\hspace*{\fill}
\unitlength=0.4pt
\begin{picture}(80.00,80.00)(30.00,730.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(70.00,770.00){\makebox(0,0)[cc]{$\tau$}}
\end{picture}
\hspace*{\fill}
\unitlength=0.4pt
\begin{picture}(80.00,80.00)(30.00,730.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$\geq 2$}}
\put(70.00,770.00){\makebox(0,0)[cc]{$\tau$}}
\end{picture}
\hspace*{\fill}
\end{center}
For these four cases we get:
\begin{itemize}\vspace{-2ex}
\item
$\bigcap_{a^{\nu}\in\tau}\big(\span_{\,I\!\!\!\!C} K^R_{a^{\nu}}\big)=\tau^\bot=
\span_{\,I\!\!\!\!C} \big(\bigcap_{a^{\nu}\in\tau} K^R_{a^{\nu}}\big)$,
\item
$\bigcap_{a^{\nu}\in\tau}\big(\span_{\,I\!\!\!\!C} K^R_{a^{\nu}}\big)=
(a^3,a^4)^\bot=\tau^\bot+
\unitlength=0.3pt
\begin{picture}(80.00,60.00)(30.00,760.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$1$}}
\end{picture}
=\span_{\,I\!\!\!\!C} \big(\bigcap_{a^{\nu}\in\tau} K^R_{a^{\nu}}\big)$,
\item
$\bigcap_{a^{\nu}\in\tau}\big(\span_{\,I\!\!\!\!C} K^R_{a^{\nu}}\big)=
(a^4)^\bot=\tau^\bot+
\unitlength=0.3pt
\begin{picture}(80.00,90.00)(30.00,760.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$1$}}
\end{picture}
\,+\,
\begin{picture}(80.00,90.00)(30.00,760.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$1$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$0$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$1$}}
\end{picture}
=\span_{\,I\!\!\!\!C} \big(\bigcap_{a^{\nu}\in\tau} K^R_{a^{\nu}}\big)$, and
\item
$\bigcap_{a^{\nu}\in\tau} K^R_{a^{\nu}}$ contains
\unitlength=0.3pt
\begin{picture}(80.00,90.00)(30.00,760.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$1$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$0$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$0$}}
\end{picture}
\hspace{0.5em},\hspace{0.5em}
\begin{picture}(80.00,90.00)(30.00,760.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$0$}}
\end{picture}\hspace{0.5em},\hspace{0.5em}and\hspace{0.5em}
\begin{picture}(80.00,90.00)(30.00,760.00)
\put(40.00,800.00){\line(1,0){60.00}}
\put(100.00,800.00){\line(0,-1){60.00}}
\put(100.00,740.00){\line(-1,0){60.00}}
\put(40.00,740.00){\line(0,1){60.00}}
\put(40.00,800.00){\circle*{6}}
\put(100.00,800.00){\circle*{6}}
\put(100.00,740.00){\circle*{6}}
\put(40.00,740.00){\circle*{6}}
\put(30.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(110.00,820.00){\makebox(0,0)[cc]{$0$}}
\put(30.00,720.00){\makebox(0,0)[cc]{$1$}}
\put(110.00,720.00){\makebox(0,0)[cc]{$1$}}
\end{picture}
\hspace{0.1em};\hspace{0.5em}hence
$\span_{\,I\!\!\!\!C} \big(\bigcap_{a^{\nu}\in\tau} K^R_{a^{\nu}}\big)=M_{\,I\!\!\!\!C}$.
\vspace{2ex}
\vspace{-1ex}\end{itemize}
So indeed $T^2_{\tau,\,\mbox{\footnotesize loc}}(-R)=0$ for all $R$.
\par
\neu{deform-contract}
Since we have related $T^{k}_X(-R)$ to the cohomolgy groups of
$C_{\kss \bullet}\big({\langle\sigma\rangle},V^R_{\kss \bullet}\big)$, we are going to show the exactness of this
complex for the degrees in question. Let us begin with
a topological lemma stating the
contractibility of certain subcomplexes of polytopes.
\par
{\bf Lemma:}
{\em
Let ${\Delta}\subseteq{I\!\!R}^n$ be a polytope.
For a hyperplane $H\subseteq{I\!\!R}^{n+1}$ and any subfan
${\cal C}\subseteq\big\{\tau\leq\mbox{\rm cone}({\Delta})\,\big|\; \tau\subseteq H\big\}$,
we define
\[
{\big\langle\cone(\kP)\big\rangle}^{H,{\cal C}}:=\big\{\tau\leq\mbox{\rm cone}({\Delta})\,\big|\; \tau\subseteq H^+\,
\mbox{\rm and }\, \tau\cap H\in{\cal C}\big\}
\]
with $H^+$ denoting a closed half space corresponding to $H$.
Then, if ${\Delta}\cap \mbox{\rm int}(H^+)$ is non-empty, the constant cohomological
system is acyclic, i.e.\
\[
H^{\kss \bullet}\Big({\big\langle\cone(\kP)\big\rangle}^{H,{\cal C}},\,{Z\!\!\!Z}\Big)=0\,.
\vspace{-3ex}
\]
}
\par
{\bf Proof:}
We have to check that the corresponding polyhedral subcomplex
${\Delta}^{H,{\cal C}}\subseteq{\Delta}$ is contractible. But this is a consequence
of the following two points:
\begin{itemize}\vspace{-2ex}
\item[(i)]
${\Delta}_{H,{\cal C}}:={\Delta}\cap\big[\mbox{\rm int}( H^+) \cup |{\cal C}|\big]\,\subseteq {\Delta}$
is star shaped, hence contractible.
\item[(ii)]
We use the general fact that, if $Q$ is a polytope and $\widetilde{H}^+$
is a subset of the closed halfspace $H^+$ containing
$\mbox{\rm int}(H^+)$ with $Q\not\subseteq \widetilde{H}^+$,
then $\partial Q\cap \widetilde{H}^+$ is a deformation retract of
$Q\cap \widetilde{H}^+$.
This enables us to successively get rid
of ``damaged'' ${\Delta}$-faces contained in ${\Delta}_{H,{\cal C}}$. In the end we get
that ${\Delta}^{H,{\cal C}}$ is a deformation retract of ${\Delta}_{H,{\cal C}}$.
\vspace{-5ex}
\vspace{-1ex}\end{itemize}
\hspace*{\fill}$\Box$
\par
\neu{deform-geq2}
We return to the situation of \zitat{deform}{leq1} and
\zitat{deform}{loc}, i.e.\
${\Delta}\subseteq{I\!\!R}^n$ is a lattice polytope giving rise to the Gorenstein cone
$\sigma:=\mbox{\rm cone}({\Delta})\subseteq N_{I\!\!R}={I\!\!R}^{n+1}$.
\par
{\bf Proposition:}
{\em
If $R\in M$ is a degree such that $R\not\leq 1$ on ${\Delta}$, then the
complex induced by the homological system $V^R_{\kss \bullet}$ is exact.
}
\par
{\bf Proof:}
The degree $R\in M$ induces a subfan
\[
{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}:=\{ \tau\leq\mbox{\rm cone}({\Delta})\,|\;
\langle a^{\nu},R\rangle\geq 1\;\mbox{ for every }a^{\nu}\in\tau\}\;\subseteq{\big\langle\cone(\kP)\big\rangle}.
\]
For every $\tau\in{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}$, we write $\tilde{\tau}\leq\tau$ for the
face spanned only by those generators $a^{\nu}\in\tau$ satisfying
$\langle a^{\nu},R\rangle=1$. The homological system $V^R_{\kss \bullet}$ can more conveniently
be described as
\[
V^R_\tau=\left\{\begin{array}{ll}
\tilde{\tau}^\bot\subseteq M_{\,I\!\!\!\!C} & \mbox{if } \tau\in {\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}\\
0 & \mbox{otherwise}.
\end{array}\right.
\]
We construct a homotopy between $0$ and the identical map
$\mbox{id}:(V^R_{\kss \bullet})^\ast \to (V^R_{\kss \bullet})^\ast$.
Hence, denoting by ${Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_i\big]$ the free abelian group
generated by the ${k}$-dimensional cones, it remains to construct a homotopy
\[
\dgARROWLENGTH=2em
\begin{diagram}
\node{{Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_{{k}+1}\big]}
\arrow{e,t}{\partial}
\arrow{s,l}{\mbox{\footnotesize id}}
\node{{Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_{{k}}\big]}
\arrow{e,t}{\partial}
\arrow{s,l}{\mbox{\footnotesize id}}
\arrow{sw,t}{D^{k}\hspace{-0.7em}}
\node{{Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_{{k}-1}\big]}
\arrow{s,l}{\mbox{\footnotesize id}}
\arrow{sw,t}{D^{{k}-1}\hspace{-0.8em}}\\
\node{{Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_{{k}+1}\big]}
\arrow{e,t}{\partial}
\node{{Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_{{k}}\big]}
\arrow{e,t}{\partial}
\node{{Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_{{k}-1}\big]}
\end{diagram}
\]
such that $D^{k}(\tau)\in {Z\!\!\!Z}\big[{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}_{{k}+1}\big]$
may be written as
$D^{k}(\tau)=\sum_v\lambda_v\phi_v$ with $\lambda_v\in{Z\!\!\!Z}$ and
$\phi_v\in{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}$ such that $\tilde{\phi}\subseteq\tilde{\tau}$:\\
Assume that $D^{{k}-1}$ has been already constructed. If
$\tau\in{\big\langle\cone(\kP)\big\rangle}^{[R\geq 1]}$ is an ${k}$-dimensional cone, then we can
apply Lemma \zitat{deform}{contract} with $H:=[R=1]$ and
${\cal C}$ being the fan consisting of $\tilde{\tau}$ and its faces.
Since $\big(\tau - D^{{k}-1}(\partial\tau)\big)\in{\big\langle\cone(\kP)\big\rangle}^{H,{\cal C}}$ and
\[
\partial\big(\tau - D^{{k}-1}(\partial\tau)\big)
\;=\;
\partial\tau - (\partial\circ D^{{k}-1})(\partial\tau)
\;=\;
(D^{{k}-2}\circ\partial)(\partial\tau)
\;= 0,
\]
there exists an element $D^{k}(\tau)\in{\big\langle\cone(\kP)\big\rangle}^{H,{\cal C}}_{{k}+1}$ such that
$\partial D^{k}(\tau)= \tau - D^{{k}-1}(\partial\tau)$.
\hfill$\Box$
\par
As a straightforward consequence of the Propositions
\zitat{deform}{leq1}, \zitat{deform}{loc}, \zitat{deform}{geq2}
and of Theorem \zitat{van}{pyramid}
we obtain the following
\par
{\bf Theorem:}
{\em
Assume that the two-dimensional faces of ${\Delta}$ are either squares or
triangles with area $1$ and $1/2$, respectively, i.e.\
$X_{\Delta}$ is a conifold in codimension three. Then, if $R\in M$ is any
degree, we have for ${k}\leq 2$
\[
\renewcommand{\arraystretch}{1.2}
T^{k}_X(-R)=\left\{\begin{array}{@{}cl}
{D}^{k}\big({\Delta}\cap [R=1]\big) & \mbox{if } R\leq 1 \mbox{ on }{\Delta}\\
0 & \mbox{otherwise}.
\end{array}\right.
\]
In particular, if ${\Delta}$ additionally
satisfies the cleaning condition \zitat{van}{pyramid}
and contains only pyramids as three-dimensional faces, then $T^2_X=0$.
}
\par
{\small
|
\section{Introduction}
The merger hypothesis for elliptical galaxy formation, as put forth by
Toomre \& Toomre (1972; see also \cite{Toomre77}), posits that two
spiral galaxies can fall together under their mutual gravitational
attraction, eventually evolving into an elliptical-like remnant. An
early objection to this hypothesis was that the cores of ellipticals
are too dense to result from the dissipationless merging of two
spirals (\cite{Carlberg86,Gunn87,HSH93}). An obvious solution to this
objection is to include a dissipative (gaseous) component in the
progenitors (for other solutions, see
e.g.~\cite{VW85,Barnes88,Lake89}). Numerical experiments including
such a component readily showed that gaseous dissipation can
efficiently drive large amounts of material into the central regions
(\cite{Negroponte83,Noguchi86,BarnesH91}). Indeed, it was considered
a great success for early hydrodynamical work to be able to reproduce
dense knots of gas within the central regions of simulated merger
remnants, similar to the gas concentrations observed in IR
luminous mergers (\cite{BarnesH91,San88b}). Since the inferred
central gas mass densities in the observed gas knots are comparable to
the stellar mass density seen in the cores of normal ellipticals
($\sim 10^2 M_\odot$ pc$^{-3}$), it seems natural that they
could be the seed of a high surface brightness remnant
(\cite{Kormendy92}).
Subsequent numerical work by Mihos \& Hernquist (1994; hereafter MH94)
finds that the dissipative response of the simulated gas component is
so efficient that the resulting mass profiles of the simulated
remnants are unlike those seen in normal ellipticals. In particular,
ensuing starformation leaves behind a dense stellar core whose surface
density profile does not join smoothly onto the de Vaucoleurs
$r^{1/4}$ profile of the pre-existing stellar population. Instead,
the profiles exhibit a ``spike'' at small radii, with a suggested
increase in surface brightness by factors $\sim$100. While the
predicted break in the mass density profile occurs at spatial scales
comparable to the gravitational softening length of the simulations,
making the precise slope somewhat questionable, the conclusion that
the profiles should exhibit a clear break was considered firm
(MH94). This prediction, if confirmed, offers a means to constrain the
frequency of highly dissipative mergers in the past by searching for
their fossil remnants in the cores of nearby ellipticals. However, the
numerical formalisms used in MH94 to model gaseous dissipation, star
formation, and energy injection back into the ISM from massive stars
and SNe (``feedback") are necessarily ad hoc in nature. As such,
these predictions should be viewed as preliminary (as Mihos \&
Hernquist themselves note).
We investigate this question observationally by converting the
observed gas column densities of on-going or late-stage mergers into
optical surface brightness by assuming a stellar mass-to-light ratio
appropriate for an evolved population. This light is added to the
observed luminosity profile after allowing it to age passively. The
resulting luminosity profile is examined for anomalous features such
as the sharp break predicted by MH94.
We conduct this experiment with the late-stage mergers NGC 3921 and
NGC 7252 and with the ultraluminous infrared (ULIR) merger Arp 220.
These systems were chosen because they have been observed in the
CO(1-0) molecular line transition with resolutions (full width at half
maximum) of $2^{\prime\prime}-2.5^{\prime\prime}$
(\cite{YH99a,Wang92,Sco97}). The resulting spatial radial resolution
(300--400 pc; $H_o$= 75 km s$^{-1}$ Mpc$^{-1}$) is similar to the
hydrodynamical smoothing length used in MH94 ($\sim$ 350
pc)\footnote{Assuming scaling parameters appropriate for a
MilkyWay-like progenitor.}, indicating that the molecular line
observations have sufficient resolution to resolve the types of mass
concentrations found in the simulations. The molecular gas surface
densities of each of these systems are plotted in
Figure~\ref{fig:radplotA}, converted from CO fluxes by adopting a
conversion factor of $N_{H_2} /I_{CO} = 3\times10^{20} \, {\rm
cm^{-2}\,(K \, km \, s^{-1})^{-1}}$ (\cite{Young91}). Detailed
studies on each of these systems, which fully discuss their status as
late stage mergers, can be found in Schweizer (1996) and Hibbard \&
van Gorkom (1996) for NGC 3921; Schweizer (1982) and Hibbard et
al.~(1994) for NGC 7252; and Scoville et al.~(1997)
for Arp 220.
\section{Results}
\subsection{NGC 3921 \& NGC 7252}
For these moderately evolved merger remnants (ages of $\sim$ 0.5--1
Gyr since their tidal tails were launched, \cite{Hib94,HvG96}), the
observed gas and luminosity profiles are used to predict the expected
luminosity profile of a 2 Gyr old remnant. We assume that all of the
molecular gas is turned into stars at the same radii, adopting an
exponentially declining starformation history. The present luminosity
profile is allowed to fade due to passive aging effects, and the final
luminosity profile is the sum of these two populations. The molecular
gas profiles plotted in Fig.~\ref{fig:radplotA} are converted into gas
mass densities by multiplying by a factor of 1.36 to take into
consideration the expected contribution of Helium. Optical luminosity
profiles have been obtained by Schweizer (1982, 1996), Whitmore et
al.~(1993), and Hibbard et al.~(1994), showing them to be well fitted
by an $r^{1/4}$ profile over all radii, with no apparent luminosity
spikes.
The gas surface densities ($\Sigma_{gas}$ in $M_\odot\,{\rm pc}^{-2}$)
are converted to optical surface brightnesses ($\mu_B$) by dividing by
the stellar mass-to-light ratio ($M_*/L_B$) expected for a 2 Gyr old
population. We adopt the stellar mass-to-light ratios given by de Jong
(1995, Table 1 of ch.~4), which were derived from the population
synthesis models of Bruzual \& Charlot (1993) for an exponentially
declining star formation history, a Salpeter IMF, and Solar metallicity
($M_*/L_B = 0.82 \, M_\odot L_\odot^{-1}$ at 2 Gyr). Noting that 1
$L_\odot \,{\rm pc}^{-2}$ corresponds to $\mu_B$ = 27.06 mag
arcsec$^{-2}$ (adopting $M_{B,\odot}=+5.48$), the conversion from gas
surface density to optical surface brightness is given by $\mu_B(r) =
27.06\,{\rm mag \, arcsec}^{-2} - 2.5\times log[\Sigma_{gas}(r) /
(M_*/L_B) ]$.
The luminosity profiles of the evolved remnants are estimated from the
observed $B$-band profiles (\cite{Schwe96a,Hib94}), allowing for a
fading of +1 mag arcsec$^{-2}$ in the $B$-band over the next 2 Gyr
(\cite{BC93,Schwe96a}), and adding in the expected contribution of the
population formed from the molecular gas, calculated as above. We
emphasize that this should favor the production of a luminous
post-merger population, since it assumes the that none of the
molecular gas is lost to stellar winds or SNe and the adopted IMF
favors the production of many long-lived low-mass stars.
The results of this exercise are plotted in Figure~\ref{fig:radplotB}.
This plot shows that the observed gas densities in NGC 3921 and NGC
7252, although high, are not high enough to significantly affect the
present luminosity profiles. The profile of NGC 3921 is basically
indistinguishable from an $r^{1/4}$ profile. The profile of NGC 7252
does show a slight rise at small radii, but not the clear break
predicted by MH94. Therefore the resulting luminosity profiles of
these remnants are now and should remain fairly typical of normal
elliptical galaxies, and the conclusions of MH94 are not applicable to
{\it all} mergers of gas-rich galaxies. Since both of these systems
also obey the Faber-Jackson relationship (\cite{Lake86}) and NGC 7252
falls upon the fundamental plane defined by normal ellipticals
(\cite{Hib94,Hib95}), we conclude that at least some mergers of
gas-rich systems can evolve into normal elliptical galaxies as far as
their optical properties are concerned.
\subsection{Arp 220}
Since Arp 220 is an extremely dusty object, its optical luminosity
profile is poorly suited for a similar analysis. Instead, we use a
luminosity profile measured in the near-infrared, where the dust
obscuration is an order of magnitude less severe. Arp 220 was recently
observed with camera 2 of NICMOS aboard the HST (Scoville et
al.~1998), and we use the resulting $K-$band luminosity profile,
kindly made available by N. Scoville. Since Arp 220 is presently
undergoing a massive starburst, the fading factor is much less certain
than for the already evolved systems treated above, and depends
sensitively on what fraction of the current light is contributed by
recently formed stars. We adopt a situation biased towards the
production of a discrepant luminosity profile by assuming that the
entire population was pre-existing, converting the observed $K-$band
profile to an evolved $B-$band profile by adopting a $B-K$ color of 4,
appropriate for a 10 Gyr old population (\cite{deJong}). The
contribution due to the population formed from the molecular disk
is calculated exactly as before. The resulting profile is shown in
Figure~\ref{fig:radplotB}.
This figure shows that Arp 220 is predicted to evolve a luminosity
profile with a noticeable rise at small radii. This is due to the peak
in the molecular gas surface density at radii less than 0.5 kpc
(Fig.~\ref{fig:radplotA}). We conclude that Arp~220 has the potential
to evolve a similar feature in its luminosity profile, {\it if} indeed
all of the current molecular gas is converted into stars. However,
the expected rise of $\sim$2 mag arcsec$^2$ in surface brightness (a
factor of $\sim$6) is considerably lower than the two orders of
magnitude increase predicted by the simulations (see Fig.~1 of MH94).
\section{Discussion}
From the above exercise, we conclude that neither NGC 3921 nor NGC
7252 are expected to show a significant deviation in their luminosity
profiles, and that the maximum rise expected for Arp 220 is
considerably lower than the two orders of magnitude increase predicted by
the simulations of MH94. We conclude that the numerical formalisms
adopted in the simulations to treat the gas and star formation are
incomplete. Mihos \& Hernquist enumerate various possible shortcomings
of their code. For example, their star-formation criterion is
extrapolated from studies of quiescent disk galaxies, and may not
apply to violent starbursts. Perhaps most importantly, their
simulations fail to reproduce the gas outflows seen in ULIR galaxies
(``superwinds" e.g.~\cite{HAM90,HLA93}), suggesting that the numerical
treatment of feedback is inadequate.
In spite of these results, it is still interesting that under some
conditions there might be an observational signature of a past merging
event in the light profile of the remnant. The question is for which
mergers might this be the case? Since $\Sigma_{H_2}$ is tightly
correlated with IR luminosity (\cite{YH99b}), we infer that only the
ultraluminous IR galaxies retain the possibility to evolve into
ellipticals with a central rise in their luminosity profiles. While
such profiles are not typical of ellipticals in general, they are not
unheard of. For example, $\sim$ 10\% of the the Nuker sample profiles
presented by Byun et al.~(1996) show such anomalous cores (e.g., NGC
1331, NGC 4239). It is therefore possible that such systems evolve
from ultraluminous IR galaxies. This can be tested by careful
``galactic archaeology'' in such systems to search for signatures of a
past merger event (e.g., \cite{SS92,Malin97}).
However, it is not a foregone conclusion that systems like Arp 220
will evolve anomalous profiles. This system presently hosts a very powerful
expanding ``superwind'' (\cite{Heckman96}), which may be able to eject
a significant fraction of the cold gas in a ``mass-loaded flow''
(e.g. \cite{Heckman99}). Such winds are common in ULIR galaxies
(\cite{HAM90,HLA93}). Another related possibility is that the IMF may
be biased towards massive stars (i.e.~``top heavy'',
\cite{Young86,Sco91b}). Such mass functions will leave far fewer stellar
remnants than the IMF adopted here. A third possibility is that the
standard Galactic CO-to-H$_2$ conversion factor is inappropriate for
ULIR galaxies, and that the high gas surface densities derived from CO
observations (and thus the resulting stellar luminosity profile) may
be over-estimated (see \cite{Downes93,Bryant96}).
Some support for the idea that central gas cores may be depleted by
the starburst is given by a population synthesis model of NGC~7252,
which suggests that it experienced an IR luminous phase
(\cite{Uta94}). While the current radial distribution of molecular
gas in NGC~7252 is flat and lacks the central core seen in Arp~220, it
appears to connect smoothly with that of Arp~220 in
Fig.~\ref{fig:radplotA}. Therefore one may speculate that NGC~7252
did indeed have a radial gas density profile much like Arp~220 but has
since lost the high density gas core as a result of prodigious massive star
formation and/or superwind blowout. However, the burst parameters are
not strongly constrained by the available observations, and a weaker
burst spread over a longer period may also be allowed
(\cite{Uta94}). Further insight into this question could be obtained
by constraining the past star formation history in other evolved
merger remnants.
In conclusion, a comparison of the peak molecular column densities and
optical surface brightnesses in NGC~3921 and NGC~7252 suggests that
some mergers between gas-rich disks {\it will} evolve into
elliptical-like remnants with typical luminosity profiles, even
considering their present central gas supply. For ULIR galaxies like
Arp~220 the case is less clear. Such systems will either produce an
excess of light at small radii, as seen in a small number of
ellipticals, or require some process such as mass-loaded galactic
winds or a top-heavy IMF to deplete the central gas supply without
leaving too many evolved stars. If the latter possibility can be
excluded, then the frequency of such profiles may be used to constrain
the number of early type systems formed via ULIR mergers.\footnote{ We
note that any subsequent {\it dissipationless} merging of these cores
with other stellar system will tend to smooth out these profiles.}
\section{Summary}
\begin{itemize}
\item{Even under assumptions that favor the production of a
luminous post-merger population, the dense molecular gas complexes
found in the centers of NGC 3291 and NGC 7252 should not significantly
alter their luminosity profiles, which are already typical of
elliptical galaxies (\cite{Schwe82,Schwe96a}). Since these systems
also obey the Faber-Jackson relationship (\cite{Lake86}) and NGC 7252
falls upon the fundamental plane defined by normal ellipticals
(\cite{Hib94,Hib95}), it appears that at least some mergers of
gas-rich systems can evolve into normal elliptical galaxies as far as
their optical properties are concerned.}
\item{The dense molecular gas complex found in the center of
Arp 220 may result in a moderate rise in the remnants' luminosity
profile at small radii. Since the molecular gas column density is a
tight function of IR luminosity (\cite{YH99b}), we conclude that this
condition may apply to all of the ultraluminous infrared
galaxies. However, this does not preclude a merger origin for
elliptical galaxies since (1) About 10\% of the Nuker sample
ellipticals (Byun et al. 1996) show such rises in their radial light
profiles, and (2) it is possible that much of the gas in such systems
is blown into intergalactic space by the mass-loaded superwinds found
emanating from such objects (\cite{Heckman99,HLA93}).}
\item{The maximum expected rise in the luminosity profiles are
considerably lower than the orders of magnitude increase predicted
by the simulations (MH94). We therefore suggest that the
numerical formalisms adopted in the simulations to treat the gas
and star formation are incomplete.}
\end{itemize}
\acknowledgements
The authors thank N. Scoville for kindly providing the NICMOS K-band
profile for Arp 220. We thank F. Schweizer and J. van Gorkom for
comments on an earlier version of this paper, and R. Bender and
C. Mihos for useful discussions, and the referee, J. Barnes, for a
thorough report.
|
\section{INTRODUCTION}
This is the third part of our eight presentations in which we consider
applications of methods from wavelet analysis to nonlinear accelerator
physics problems. This is a continuation of our results from [1]-[8],
which is based on our approach to investigation
of nonlinear problems -- general, with additional structures (Hamiltonian,
symplectic or quasicomplex), chaotic, quasiclassical, quantum, which are
considered in the framework of local (nonlinear) Fourier analysis, or wavelet
analysis. Wavelet analysis is a relatively novel set of mathematical
methods, which gives us a possibility to work with well-localized bases in
functional spaces and with the general type of operators (differential,
integral, pseudodifferential) in such bases.
In this part we consider effects of insertion devices (section 2)
on beam dynamics.
In section 3 we consider generalization of our variational approach
for the case of variable coefficients.
In section 4 we consider more powerful variational approach
which is based on ideas of para-products and approximation
for multiresolution approach, which gives us possibility
for computations in each scale separately.
\section{Effects of Insertion Devices on Beam Dynamics}
Assuming a sinusoidal field variation, we may consider according to [9]
the analytical treatment of the effects of insertion devices on beam dynamics.
One of the major detrimental aspects of the installation of insertion devices
is the resulting reduction of dynamic aperture. Introduction of non-linearities
leads to enhancement of the amplitude-dependent tune shifts and distortion of
phase space. The nonlinear fields will produce significant effects at large
betatron amplitudes.
The components of the insertion device magnetic field used for the derivation of
equations of motion are as follows:
\begin{eqnarray}
B_x&=&\frac{k_x}{k_y}\cdot B_0 \sinh(k_xx)\sinh(k_yy)\cos(kz)\nonumber\\
B_y&=&B_0\cosh(k_xx)\cosh(k_yy)\cos(kz)\\
B_z&=&-\frac{k}{k_y}B_0\cosh(k_xx)\sinh(k_yy)\sin(kz),\nonumber
\end{eqnarray}
with $k_x^2+k_y^2=k^2=(2\pi/\lambda)^2$,
where $\lambda$ is the period length of the insertion device, $B_0$ is
its magnetic field, $\rho$ is the radius of the curvature in the field $B_0$.
After a canonical transformation to change to
betatron variables, the Hamiltonian is averaged over the period of
the insertion device and hyperbolic functions are expanded to the fourth
order in $x$ and $y$ (or arbitrary order).
Then we have the following Hamiltonian:
\begin{eqnarray}
H&=&\frac{1}{2}[p_x^2+p_y^2]+\frac{1}{4k^2\rho^2}[k_x^2x^2+k_y^2y^2]\nonumber\\
&+&\frac{1}{12k^2\rho^2}[k_x^4x^4+k_y^4y^4+3k_x^2k^2x^2y^2]\\
&-&\frac{\sin(ks)}{2k\rho}[p_x(k_x^2x^2+k_y^2y^2)-2k_xp_yxy]\nonumber
\end{eqnarray}
We have in this case also nonlinear (polynomial with degree 3) dynamical system with variable
(periodic) coefficients. As related cases we may consider wiggler and undulator
magnets. We have in horizontal $x-s$ plane the following equations
\begin{eqnarray}
\ddot{x}&=&-\dot{s}\frac{e}{m\gamma}{B_z(s)}\\
\ddot{s}&=&\dot{x}\frac{e}{m\gamma}B_z(s),\nonumber
\end{eqnarray}
where magnetic field has periodic dependence on $s$ and hyperbolic on $z$.
\section{Variable Coefficients}
In the case when we have situation when our problem is described by a system of
nonlinear (polynomial)differential equations, we need to consider
extension of our previous approach which can take into account
any type of variable coefficients (periodic, regular or singular).
We can produce such approach if we add in our construction additional
refinement equation, which encoded all information about variable
coefficients [10].
According to our variational approach we need to compute integrals of
the form
\begin{equation}\label{eq:var1}
\int_Db_{ij}(t)(\varphi_1)^{d_1}(2^m t-k_1)(\varphi_2)^{d_2}
(2^m t-k_2)\mathrm{d} x,
\end{equation}
where now $b_{ij}(t)$ are arbitrary functions of time, where trial
functions $\varphi_1,\varphi_2$ satisfy a refinement equations:
\begin{equation}
\varphi_i(t)=\sum_{k\in{\bf Z}}a_{ik}\varphi_i(2t-k)
\end{equation}
If we consider all computations in the class of compactly supported wavelets
then only a finite number of coefficients do not vanish. To approximate
the non-constant coefficients, we need choose a different refinable function
$\varphi_3$ along with some local approximation scheme
\begin{equation}
(B_\ell f)(x):=\sum_{\alpha\in{\bf Z}}F_{\ell,k}(f)\varphi_3(2^\ell t-k),
\end{equation}
where $F_{\ell,k}$ are suitable functionals supported in a small neighborhood
of $2^{-\ell}k$ and then replace $b_{ij}$ in (\ref{eq:var1}) by
$B_\ell b_{ij}(t)$. In particular case one can take a characteristic function
and can thus approximate non-smooth coefficients locally. To guarantee
sufficient accuracy of the resulting approximation to (\ref{eq:var1})
it is important to have the flexibility of choosing $\varphi_3$ different
from $\varphi_1, \varphi_2$. In the case when D is some domain, we
can write
\begin{equation}
b_{ij}(t)\mid_D=\sum_{0\leq k\leq 2^\ell}b_{ij}(t)\chi_D(2^\ell t-k),
\end{equation}
where $\chi_D$ is characteristic function of D. So, if we take
$\varphi_4=\chi_D$, which is again a refinable function, then the problem of
computation of (\ref{eq:var1}) is reduced to the problem of calculation of
integral
\begin{eqnarray}
&&H(k_1,k_2,k_3,k_4)=H(k)=\nonumber\\
&&\int_{{\bf R}^s}\varphi_4(2^j t-k_1)\varphi_3(2^\ell t-k_2)\times\\
&&\varphi_1^{d_1}(2^r t-k_3)
\varphi_2^{d_2}(2^st-k_4)\mathrm{d} x\nonumber
\end{eqnarray}
The key point is that these integrals also satisfy some sort of refinement
equation:
\begin{equation}
2^{-|\mu|}H(k)=\sum_{\ell\in{\bf Z}}b_{2k-\ell}H(\ell),\qquad \mu=d_1+d_2.
\end{equation}
This equation can be interpreted as the problem of computing an eigenvector.
Thus, we reduced the problem of extension of our method to the case of
variable coefficients to the same standard algebraical problem as in
the preceding sections.
So, the general scheme is the same one and we
have only one more additional
linear algebraic problem by which we in the same way can parameterize the
solutions of corresponding problem.
As example we demonstrate on Fig.~1 a simple model of (local) intersection
and the corresponding multiresolution representation (MRA).
\begin{figure}[ht]
\centering
\epsfig{file=tha136a.eps,width=60.5mm, bb= 0 200 599 590, clip}
\caption{Simple insertion.}
\end{figure}
\begin{figure}[ht]
\centering
\epsfig{file=tha136b.eps, width=82.5mm, bb= 0 200 599 600, clip}
\caption{MRA representations.}
\end{figure}
\section{Evaluation of Nonlinearities Scale by Scale}
We consider scheme of modification of our variational
approach in the case when we consider different scales separately. For this reason
we need to compute errors of approximations. The main problems come of course
from nonlinear terms. We follow the approach from [11].
Let $P_j$ be projection operators on the subspaces $V_j, j\in{\bf Z}$:
\begin{eqnarray}
P_j &:& L^2({\bf R}) \to V_j\\
(P_j f)(x)&=&\sum_k <f,\varphi_{j,k}> \varphi_{j,k}(x)\nonumber
\end{eqnarray}
and $Q_j$ are projection operators on the subspaces $W_j$:
\begin{eqnarray}
Q_j=P_{j-1}-P_j
\end{eqnarray}
So, for $u\in L^2({\bf R})$ we have $u_j=P_ju\quad$ and $u_j\in V_j$,
where $\{V_j\}, j\in{\bf Z}$ is a multiresolution analysis of $L^2({\bf R})$.
It is obviously that we can represent $u_0^2$ in the following form:
\begin{equation}\label{eq:form1}
u_0^2=2\sum^n_{j=1}(P_ju)(Q_ju)+\sum^n_{j=1}(Q_ju)(Q_ju)+u_n^2
\end{equation}
In this formula there is no interaction between different scales.
We may consider each term of (\ref{eq:form1}) as a bilinear mappings:
\begin{eqnarray}\label{eq:form2}
\displaystyle
M_{VW}^j : V_j\times W_j\to L^2({\bf R})=
V_j{\oplus_{j'\geq j}W_{j'}}
\end{eqnarray}
\begin{eqnarray}\label{eq:form3}
M_{WW}^j : W_j\times W_j\to L^2({\bf R})=V_j\oplus_{j'\geq j}W_{j'}
\end{eqnarray}
For numerical purposes we need formula (\ref{eq:form1}) with a finite number of
scales, but when we consider limits $j\to\infty$ we have
\begin{equation}
u^2=\sum_{j\in {\bf Z}}(2P_ju+Q_ju)(Q_ju),
\end{equation}
which is para-product of Bony, Coifman and Meyer.
Now we need to expand (\ref{eq:form1}) into the wavelet bases. To expand
each term in (\ref{eq:form1}) into wavelet basis, we need to consider
the integrals of the products of the basis functions, e.g.
\begin{equation}
M^{j,j'}_{WWW}(k,k',\ell)=\int^\infty_{-\infty}\psi^j_k(x)
\psi^j_{k'}(x)\psi^{j'}_\ell(x)\mathrm{d} x,
\end{equation}
where $j'>j$ and
\begin{equation}
\psi^j_k(x)=2^{-j/2}\psi(2^{-j}x-k)
\end{equation}
are the basis functions.
If we consider compactly supported wavelets then
\begin{equation}
M_{WWW}^{j,j'}(k,k',\ell)\equiv 0\quad \mbox{for}\quad |k-k'|>k_0,
\end{equation}
where $k_0$ depends on the overlap of the supports of the basis functions
and
\begin{equation}\label{eq:form4}
|M_{WWW}^r(k-k',2^rk-\ell)|\leq C\cdot 2^{-r\lambda M}
\end{equation}
Let us define $j_0$ as the distance between scales such that for a given
$\varepsilon$ all the coefficients in (\ref{eq:form4}) with labels
$r=j-j'$, $r>j_0$ have absolute values less than $\varepsilon$. For the purposes of
computing with accuracy $\varepsilon$ we replace the mappings in
(\ref{eq:form2}), (\ref{eq:form3}) by
\begin{equation}\label{eq:z1}
M_{VW}^j : V_j\times W_j\to V_j\oplus_{j\leq j'\leq j_0}W_{j'}
\end{equation}
\begin{equation}\label{eq:z2}
M_{WW}^j : W_j\times W_j\to V_j\oplus_{J\leq j'\leq j_0}W_{j'}\nonumber
\end{equation}
Since
\begin{equation}
V_j\oplus_{j\leq j'\leq j_0}W_{j'}=V_{j_0-1}
\end{equation}
and
\begin{equation}
V_j\subset V_{j_0-1},\qquad W_j\subset V_{j_0-1}
\end{equation}
we may consider bilinear mappings (\ref{eq:z1}), (\ref{eq:z2}) on
$V_{j_0-1}\times V_{j_0-1}$.
For the evaluation of (\ref{eq:z1}), (\ref{eq:z2}) as mappings
$V_{j_0-1}\times V_{j_0-1}\to V_{j_0-1}$
we need significantly fewer coefficients than for
mappings (\ref{eq:z1}), (\ref{eq:z2}). It is enough to consider only
coefficients
\begin{equation}
M(k,k',\ell)=2^{-j/2}\int^\infty_\infty\varphi(x-k)\varphi(x-k')\varphi(x-\ell)\mathrm{d}
x,
\end{equation}
where $\varphi(x)$ is scale function. Also we have
\begin{equation}
M(k,k',\ell)=2^{-j/2}M_0(k-\ell,k'-\ell),
\end{equation}
where
\begin{equation}
M_0(p,q)=\int\varphi(x-p)\varphi(x-q)\varphi(x)\mathrm{d} x
\end{equation}
Now as in section (3) we may derive and solve a system of linear equations to
find $M_0(p,q)$ and obtain explicit representation for solution.
We are very grateful to M.~Cornacchia (SLAC),
W.~He\-rrmannsfeldt (SLAC),
Mrs. M.~Laraneta (UCLA), J.~Ko-\\
no (LBL)
for their permanent encouragement.
|
\section{Introduction}
One of the important properties of the behavior of the nervous system,
discovered early on \cite{Adrian50}, that has attracted
significant amount of attention, is the synchronization
of neuronal discharges. In recent years the advent of improved
experimental techniques has provided vast amounts of new synchronization
data. Concomitantly, there has been a resurgence in interest and
controversy concerning the functional relevance of synchronization.
It has been established that {\it in vivo} cortical neurons have noisy
spike trains \cite{Softky93} (but see in contrast \cite{Gur97}),
and that groups of neurons discharge coherently as found in population
recordings (such as EEGs, or by arrays of extracellular electrodes, for
a review see \cite{Aertsen93}). These two facts have sparked major
controversies. Firstly, does noise (or precise timing) in neuronal spike
trains contain information \cite{Rieke97,Softky95}, or is information merely
due to noisy processing of an average firing rate
\cite{Shadlen94,Shadlen95,Shadlen98}? Secondly, is synchronization
functionally (or even statistically) significant\cite{Laurent97,Abeles98},
or just an epiphenomenon \cite{Shadlen98}?
In this paper we focus on two different aspects of synchronization
that have received little attention so far. Can realistic neuronal networks
synchronize under the biological conditions of variable intrinsic
neuronal properties, and the noise-induced neuronal unreliability?
What kind of synchronization can be obtained, and what are
its pertinent statistical properties? It is necessary
to resolve these two questions to properly formulate the issues
to be studied in experiment, and to analyze different
ways of probing the experimental
data. Here we focus our attention on the extensively-studied synchronous
gamma oscillations in hippocampus
\cite{traub,Traub98a,whittington,traub2,buhl98,JM99}.
Theoretical and computational work has shown that mutual inhibition
is capable of synchronizing neuronal networks \cite{vreeswijk94,Wang93}.
Subsequent in vitro experiments have convincingly established
the role of GABA-ergic hippocampal interneurons in gamma oscillations
\cite{whittington,traub}. Wang \& Buzs\'aki \cite{wang} studied the effect of
current heterogeneity and partial connectivity on the synchronization
of the hippocampal network. They only found strong synchronization
in the gamma frequency range when the current heterogeneities
were small \cite{wang,white}. In strong synchronization all neurons
in a local circuit spike within a short interval of each other.
This suggests that strong synchronization can only be obtained when
the intrinsic properties of the neurons are not too different.
According to Ref.\cite{wang} this would mean a less than $10\%$ difference in
current drive, or average firing rate. It has been hard to pinpoint the
amount of variability in intrinsic properties in the {\em in vitro}
and {\it in vivo} preparations of different brain areas.
It is however not unreasonable to assume the presence of more than
$10\%$ variability in these preparations. Strong
synchronization is also not robust against noise \cite{CNS}.
It would therefore seem unlikely for strong synchronization
to be present in hippocampus under physiological conditions.
Indeed, here we show that stochastic weak synchronization (SWS)
is more prevalent in parameter space, and is also robust against
neuronal heterogeneities and synaptic noise. We conjecture that
as a consequence it is much more likely to occur in neuronal systems.
In SWS, neurons spike within a short interval from each other,
but not necessarily at each period \cite{Hakim98,Amit97}.
The synchronization is called stochastic, because
the particular cycle in which the neurons fire is random.
This makes the properties of this state different
from the well known cluster states studied by previous authors
\cite{Golomb92,GolombRinzel93,GolombRinzel94,Kopell94}.
There each neuron always fires at the same cycle with the same
cluster. Both strong and stochastic weak synchronization
yield periodic population oscillations. The difference
can then only be ascertained using multi-unit recordings.
We use cross correlation analysis to show that noise and heterogeneity
affect the synchronization properties of our network in very different
ways. Large enough noise and heterogeneity will, however, stop
strongly synchronized oscillations. We demonstrate that neither
adding a periodic drive nor increasing synaptic coupling can significantly
increase robustness of strong synchronization. Finally we determine
for what parameters robust self-induced 40Hz synchronous oscillations
can be obtained.
\section{Methods}
\subsection{Single neuron model}
Our aim here is to establish physiological
criteria for robust synchronization in the gamma frequency range.
The use of a biophysically realistic model is therefore of pivotal importance.
At the same time it is also important to balance the amount of complexity
versus practical simplicity \cite{Rall95}.
We have therefore not attempted to use the latest available data to construct
a detailed multi-compartmental model. The computer requirements
to sample the full relevant parameter space, and perform our type of
analysis, would be extremely demanding even using very fast
computers.
It has been shown, nonetheless, that one and two compartmental models can
accurately generate spike trains of the right shape and frequency
\cite{PinskyRinzel94,Mainen94,MainenSejn98}.
Multi-compartmental models may be necessary to assess the synaptic
integration of inputs located on different parts of the dendritic
tree. This is currently an intensely studied area in
electrophysiology \cite{Yuste96,Koch97,Sakmann99}.
Here we study a model previously introduced by others\cite{wang}. The model
has been shown to reproduce the salient features of the dynamics
of hippocampal interneurons.
The neurons are modeled as a single compartment
with Hodgkin-Huxley type sodium and potassium channels.
In this work all the neurons are connected to all others and
themselves (ALL to ALL connectivity) via
inhibitory GABA$_A$-synapses.
The equation for the membrane potential of a neuron is
(the index $i$ of the neuron is omitted)
\begin{equation}
C_m \frac{dV}{dt}=-I_{Na}-I_K-I_L-I_{syn}+I+C_m\xi.
\label{SINGNEUR}
\end{equation}
Here we use: the leak current
\(
I_L=g_L (V-E_L),
\)
the sodium current
\(
I_{Na}=g_{Na} m_{\infty}^3 h (V-E_{Na}),
\)
the potassium current:
\(
I_K=g_K n^4 (V-E_{K}),
\) and the synaptic current:
\(
I_{syn}=g_{syn}s(V-E_{syn})
\).
The Gaussian noise is denoted as $\xi$ (see below),
and $I$ is the tonic drive.
The channel kinetics are given in terms of $m$, $n$, and $h$.
They satisfy the following first order kinetics:
\begin{equation}
\frac{dx}{dt}=\phi (\alpha_x (1-x)-\beta_x x).
\end{equation}
Here $x$ labels the different kinetic variables $m$, $n$, and $h$, and
$\phi=5$ is a dimensionless time-scale that
can be used to tune the temperature-dependent
speed with which the channels open
or close. The rate constants
are \cite{wang},
\begin{eqnarray}
\alpha_m&=& \frac{-0.1(V+35)}{\exp(-0.1(V+35))-1}, \nonumber\\
\beta_m&=& 4\exp(-(V+60)/18), \nonumber\\
\alpha_h&=&0.07\exp(-(V+58)/20), \nonumber\\
\beta_h&=&\frac{1}{\exp(-0.1(V+28))+1}, \nonumber\\
\alpha_n&=&\frac{-0.01 (V+34)}{\exp(-0.1(V+34))-1}, \nonumber\\
\beta_n&=&0.125~\exp(-(V+44)/80). \nonumber
\end{eqnarray}
We make the approximation that $m$ follows the asymptotic value
$m_{\infty}(V(t)))=\alpha_m/(\alpha_m+\beta_m)$ instantaneously.
The synaptic gating variable $s$ obeys the following equation
\cite{Perkel81,wangrinzel93,wang}:
\begin{equation}
\frac{ds}{dt}=\alpha F(V_{p})(1-s)-\beta s,
\label{SYNNEUR}
\end{equation}
with $\alpha=12~ ms^{-1}$, $\beta=1/\tau_{syn}$,
$F(V_p)=1/(\exp(-V_p/2)+1)$, and $V_p$ is the presynaptic potential.
The function $F(V_p)$ is chosen such that when the presynaptic
neuron fires, $V_p>0$, the synaptic channel opens.
The decay time of the postsynaptic hyperpolarization
is chosen as $\tau_{syn}=1/\beta= 10~ms$ (or $20$ ms in some instances).
We use a reversal potential of
$E_{syn}=-75~mV$ for the inhibitory (GABA$_A$) synapses \cite{Buhl95}.
The standard set of values for the conductances used in this work is
$g_{Na}=35$, $g_K=9$, $g_L=0.1$, and $g_{syn}=0.1$ (in $mS/cm^2$),
and we have taken
$E_{Na}=55~mV$, $E_K=-90~ mV$, and $E_L=-65~ mV$. The membrane
capacitance is $C_m=1\mu F/cm^2$.
Unless stated otherwise we will use the standard set
of parameters listed above. When no current
value is specified we use $I=1\,\mu A/cm^2$. The network will then
spike at approximately $39$ Hz.\\
We chose the initial values for the membrane potential at the start of
the simulations uniformly random between $-70$ and $-50~mV$.
The kinetic variables $m$, $n$, $h$, and $s$ are set
to their asymptotic stationary values
corresponding to that starting value of the membrane potential.
The resulting equations with noise are integrated using an adapted
second order Runge-Kutta method \cite{greenside}, with time step $dt=$0.01 ms.
The accuracy of this integration method was checked for
the dynamical equations without noise ($D=0$) by
varying $dt$ and comparing the result to
the one obtained with the standard 4th order Runge-Kutta method \cite{Press}
with a time-step $dt$ of $0.05~ms$.
We normalize all quantities by the surface area of the neuron.
This leads to the following system of units: the membrane
potential $V$ in $mV$, time $t$ in ms, firing rate $f$ in Hz,
membrane capacitance $C_m$ in
$\mu F/cm^2$, conductance $g_x$ in $mS/cm^2$, voltage noise $\xi$ in $mV/ms$,
strength of neuroelectric noise $D$ in $mV^2/ms$,
the rate constants $\alpha_x$ and $\beta_x$ in $ms^{-1}$, and the current $I$
in $\mu A/cm^2$. The kinetic variables $m$, $n$, $h$, $s$, and the
time-scale $\phi$ are dimensionless. Results in our paper are expressed in
this system of units.
\subsection{Heterogeneity and synaptic noise}
We have included heterogeneity in the applied current.
For each run we draw the applied current for each neuron
from a uniform distribution. The average of the current distribution
is $I$ and the variance is $\sigma^2_I$.
The current heterogeneity represents the variation
in the intrinsic properties of the neurons in the hippocampus.
Experimental measurements of quantities like the input resistance $R_{in}$,
the membrane time-scale, the spontaneous spiking rate,
the shape of the somatic action potential (amplitude, width, rise and
fall time), and the afterhyperpolarization, show considerable variance
\cite{Lacaille87,Lacaille90,Morin96}.
It is hard to determine how much of the variance is due to measurement
errors, and how much is actually due to intrinsic neuronal variability.
Here we assume the main effect of the variability is to
change the intrinsic frequency of the neurons (which can be varied
using the current drive in our model).
Another source of heterogeneity in {\it in vitro} experiments is
the glutamate pressure
ejection method \cite{whittington}. It can lead to an inhomogeneous
activation of metabotropic glutamate receptors, and thus to a variable current.
In this paper we will consider $\sigma_I$ as a free parameter.
At least three sources of noise can be identified \cite{JohnstonWubook}:
random inhibitory postsynaptic potentials (IPSP) and excitatory postsynaptic
potentials (EPSP), stochasticity of the synaptic transmission,
and the stochasticity of the channel dynamics.
Here we assume that the variability in the neuronal discharge is
mainly due to synaptic noise \cite{Calvin67}.
We have compared the effects
of Poisson distributed spike trains of EPSPs and IPSPs to that of a Gaussian
noise current on interspike interval (ISI) variability. Poisson and Gaussian noises,
do not yield identical results.
The statistics obtained from both models, however, are similar in the parameter
regime studied \cite{CNS98}. For
the purpose of our studies we consider that Poisson and Gaussian
distributions
are two alternate ways of producing noisy spike trains with particular
statistics. Therefore, the synaptic
noise is only implemented as a Gaussian distributed,
white noise current in neuron $i$, with $\langle \xi_i(t) \rangle=0$, and
$\langle \xi_i(t) \xi_j(t') \rangle = 2D \delta(t-t')\delta_{ij}$.
The noise currents in different neurons are assumed independent.
\subsection{Measured quantities}
From our simulations we obtain the time trace for the
membrane potential $V_i(t)$ of each neuron. We determine
the spike-trace $X_i$ from $V_i$ as follows: $X_i(t)=1$ when $V_i(t)$
crosses $0\,mV$ (i.e. $V(t^-)\le 0<V(t^+)$), and it is zero elsewhere.
From $X_i$ we obtain $X(t)=\sum_i X_i(t)$. $X$ is proportional
to the instantaneous firing rate of the network. We also calculate
the correlations function $\kappa$ \cite{wang}:
\begin{equation}
\kappa\equiv \sum_{i\ne j} \frac{\langle
{\hat X}_i(n\tau){\hat X}_j(n\tau)\rangle }
{\sqrt{\langle {\hat X}_i(n\tau)\rangle \langle {\hat X}_j(n\tau)\rangle}}.
\label{calefaccion}
\end{equation}
This function measures the amount of strong synchronization,
and depends on the bin size $\tau$ of the time discretization
\begin{equation}
{\hat X}_i(n\tau)=\int_{(n-1)\tau}^{n\tau} ds~ X_i(s).
\end{equation}
We use $\tau=200~dt=2~ms$ for oscillations in the gamma-frequency
range, or $T/10$ for periodic drives with period $T$.
We also evaluate other measures that yield further
detailed quantitative characterization of the network behavior.
We calculate the time autocorrelation function:
\begin{equation}
g_x(t)= \frac{\langle x(t) x(0) \rangle-\langle x \rangle^2}{
(\langle x^2\rangle-\langle x \rangle^2)},
\end{equation}
and the cross correlation function:
\begin{equation}
g_{xy}(t)= \frac{\langle x(t) y(0) \rangle-\langle x(t)
\rangle \langle y(0) \rangle}
{\sqrt{\langle x(t)^2\rangle \langle y(0)^2\rangle}}.
\end{equation}
Here $x$ and $y$ can be any of the variables
$X_i$, $V_i$, and $X$,
and $\langle \rangle$ is a shorthand notation for the time-average.
We also consider the more conventional interspike interval
histogram (ISIH) \cite{Gerstein62}, averaged over all network neurons.
From the ISIH one can obtain two statistics: the average ISI,
$\tau_{ISI}$, and the standard deviation of the ISI, $\sigma_{ISI}$. The
ratio $\sigma_{ISI}/\tau_{ISI}$ is known as the coefficient of variation (CV).
The average firing rate is $f=1/\tau_{ISI}$, and the population standard
deviation of $f$ is $\sigma_f$.
\begin{equation}
\sigma_f=\sum_j f_j^2 - f^2,
\end{equation}
where $f_j=1/\tau_{ISI}^j$ is the average firing rate of the $j$th neuron.
In addition we plot rastergrams, with the action potential of each neuron
plotted as a filled circle, with the y-coordinate given by the neuron index
and the x-coordinate by the spiking time.
To analyze the stochastic weak synchronization network
dynamics we need to apply a different method.
The population period $\tau_n$ is different from the population
averaged ISI, and to estimate it we proceed as follows.
First we determine the firing rate ${\hat X}(t)$ as before
with $1$ ms bins. In the stochastic weak synchronization
state ${\hat X}(t)$ will consist of a
number of approximately equidistant peaks of finite width
(see Fig.$~$\ref{EXTRA5}f). We use the position of the first
maximum of the Fourier transform at nonzero frequency as an estimate $T$ for the period
$\tau_n$. We calculated the weight $\langle {\hat X}(t) \rangle$,
the average position $t_c^i=\langle t{\hat X}(t)\rangle$,
and the width $\sigma_c^i=\sqrt{\langle t^2 {\hat X}(t)\rangle-(t_c^i)^2}$
of the $i$th peak. The time average is taken over a range
$[-0.35 T, 0.35 T]$ about the estimated position
$t_c^{i-1}+T$ of the peak. We calculated the number of spikes that fall outside
this region. If the average number of missed spikes is more than one
per cycle we reject the cluster state. The cycle length (time between
two consecutive cluster firings) is defined as $\tau_c^i=t_c^i-t_c^{i-1}$.
We determine the average cluster size $N_c=\langle N_c^i\rangle$, its
$CV(N_c)=\sqrt{\langle (N_c^i)^2\rangle-N_c^2}/N_c$, the average cycle
length $\tau_n=\langle \tau_c^i \rangle$,
its $CV(\tau_n)=\sqrt{\langle (\tau_c^i)^2\rangle-\tau_n^2}/\tau_n$,
and the average width $\sigma_c=\langle \sigma_c^i\rangle$. Here the
average $\langle \cdot \rangle$ is given by the sum
over all cycles in the run (after discarding a transient).
We characterize the strength of the synchronization using a
modified $\kappa_W$ and $CV_W$. In the SWS state the ISIH has multiple
peaks. The CV of the ISI receives contributions
from the variance within each peak, but also of the variance between
the multiple peaks. We are only interested in the former, and the
conventional CV is thus an overestimate. Instead we use
$CV_W=\sigma_c/\tau_n$ which is related to the average width of one peak
in the ISIH. The coherence $\kappa$ measures the number of
coincident spikes between two spike trains. Consider two neurons that do
not spike at each cycle, but when they
both do, the spikes are coincident (that is in
the same bin).
If the probability of spiking in a cycle is $p=N_c/N$, and both neurons
fire statistically independent, we obtain $\kappa=p$. These neurons
can be considered synchronous
and we want $\kappa_{W}=1$. We therefore normalize $\kappa$ by $p$.
There is a subtlety in the calculation of the average firing rate. In the
deterministic noiseless case one ISI is enough to determine the average value
(after discarding the transient).
[Note that counting the number of spikes in a fixed interval is not an
efficient way to determine the exact firing rate.] In the
presence of noise, however, you need at least $10$
ISIs to accurately determine the average. In networks with large
current heterogeneities there are neurons with high and very low firing
rates (Fig.$~$\ref{EXTRA5}). The average ISI for the low firing rate is
less accurate than for the high firing rate neurons in the network.
However, it carries equal weight in the conventional average $\tau_{ISI}=\sum_j
\tau_{ISI}^j$. We have therefore used a weighted average
$\tau_{ISI}=\sum_j n_j \tau_{ISI}^j/\sum_j n_j$
(here $n_j$ is the number of intervals over which
$\tau_{ISI}^j$ is calculated), and the approximate
identity $N_c/\tau_n\approx N_s/\tau_{ISI}$ can be used as a check.
$N_s$ is the number of active neurons, defined as the neurons that
have more than two ISIs after the transient.
\section{Results}
\subsection{Non robustness of strong synchronization}
In this section we describe the results of our simulations
for a network of $N=100$ interneurons, connected all to all, with either
synaptic noise (SN), or current heterogeneities (CH).
In Fig.$~$\ref{RFIG5} we plot coherence parameter $\kappa$ (defined in Eq.$~$(\ref{calefaccion}))
versus the strength of the synaptic noise D, and versus the standard deviation
of the current heterogeneities $\sigma_I$. We find that strong
synchronization is lost for approximately $D>0.10~mV^2/ms$ and
$\sigma_I > 0.1~\mu A/cm^2$ (with the
standard set of parameters listed in Methods).
The mechanism by which strong synchronization is lost, however, is different
in the CH case compared to the mechanism with SN. This difference shows up only
if we studies the whole state of the network using cross
correlation functions, instead of the average quantities shown
in Fig.$~$\ref{RFIG5}. Next we compare these
two mechanisms. Wang\&Buzsaki (WB) \cite{wang} have already analyzed
the case with current heterogeneity. We have reproduced part of their
work, and we will refer to their corresponding figures.
In both CH and SN cases the neuronal firing rate decreases when the network
desynchronizes. We have plotted the time-trace of the synaptic
drive $s(t)$ in Figs.$~$\ref{RFIG5}c and d. The phasic part decreases,
and the tonic part of $s(t)$ increases with increasing $D$ and $\sigma_I$.
The increased tonic part is responsible for the lower average firing rate.
The firing rate of the CH neurons saturates
(when averaged over enough realizations of the current
heterogeneities\footnote{This assumes
\[ \int_{I_{av}-\sigma_I\sqrt{12}}^{I_{av}+\sigma_I\sqrt{12}}
f(I) dI\approx f(I_{av})\]}), whereas for the SN it increases
steadily as a function of $D$ for large values of $D$. This is because
the single
neuron firing rate increases with $D$ \cite{CNS98}, but tonic
inhibition saturates to its highest value in the asynchronous network.
The dispersion $\sigma_f$ (see Methods)
with CH is larger than the one in SN (not shown).
In SN all the neurons have identical intrinsic properties, and
the expectation value for the average frequency of each neuron
is the same. The dispersion $\sigma_f$ in this case represents the
fluctuations in the average ISI due to the finite averaging time. With CH
the neurons have different intrinsic frequencies,
and the dispersion $\sigma_f$ increases with $\sigma_I$ (and does not go
to zero after a long averaging time, see WB Fig.$~$5B).
In Fig$~$\ref{RFIG6}I and II we compare the correlation functions for the
SN and CH case, respectively. In (a) we have the strongly
synchronous network, in (c) the asynchronous network, and in (b) a
transition state. The difference between SN and CH becomes clear
when one considers the cross correlation functions. With CH the number of pairs
that are phase locked drops gradually (see WB Fig.$~$8E). The pairs that
are phase locked, are tightly phase locked (Fig.$~$\ref{RFIG6}Ib1,4,5,6), and
there is no dispersion in the cross correlations,
only a relative phase. Even in the asynchronous state the autocorrelation
function $g_{X_i}$ for a single neuron is sharp, i.e. the neuron fires
regularly with a fixed frequency (Fig.$~$\ref{RFIG6}Ic1 and c13).
The population average of $g_{X_i}$, however,
is disordered (Fig.$~$\ref{RFIG6}Ic12), since each of the neurons has a different firing rate.
In the SN case there is already dispersion due to the noise-induced jitter in the spike time, in the autocorrelation Fig.$~$\ref{RFIG6}IIa11,
and in the cross correlations (Fig.$~$\ref{RFIG6}IIa2-a10).
The dispersion increases gradually with $D$.
The difference between the CH and SN cases is also evident in the distribution
of $\kappa$ values for each pair in the network (Fig.$~$\ref{EXTRA1},
WB Fig.$~$8E). For SN there is a well defined peak, with the average
shifting to lower values as $D$
increases, (Fig.$~$\ref{EXTRA1}a-c), whereas for CH there is a broad
distribution for small $\sigma_I$
(Fig.$~$\ref{EXTRA1}d), a peak
at low values of $\kappa$ combined with a broad distribution for
moderate values
of $\sigma_I$ (Fig.$~$\ref{EXTRA1}e). For higher values of $\sigma_I$
the network is in an asynchronous regime,
and only the peak for low $\kappa$ values is present (Fig.$~$\ref{EXTRA1}f).
We have compared the ISIH for a network neuron to the ISIH of
an isolated neuron (Figs.$~$\ref{RFIG7}c-e), and also
the values of
$\tau_{ISI}$ and $\sigma_{ISI}$ (Figs.$~$\ref{RFIG7}a and b).
The CV of the network neuron is higher
than the CV of an isolated neuron which in turn is higher than
the CV of an isolated neuron with autosynaptic feedback. The inhibitory
coupling in the network increases the effect of the noise
compared to uncoupled neurons: the jitter in the spike times reduces
the phasic component of $s(t)$ (Fig.$~$\ref{RFIG5}c). This effect does not
take place in a neuron with autosynaptic feedback: the size of the
phasic component
does not decrease with $D$, only the timing deteriorates.\\
\subsection{Weak enhancement of robustness due to resonance effect}
In this section we drive the tonically active neurons
with an external periodic drive.
This drive may represent the effects of putative pacemaker
neurons, similar to the ones that were recently found in the striate cortex
of cats \cite{Gray96}. There is no compelling evidence for having the $40$Hz pacemaker
neurons projecting to hippocampal interneurons.
Here we model the pacemaker as an
excitatory synapse driven by a periodic pulse train, and investigate
its effects on the robustness of synchronization.
When a single neuron is driven by a periodic drive
it will entrain, or phase lock, when the drive frequency
is close to the natural frequency (or in some cases close to a
rational fraction) \cite{Cowan98}. Here we vary the natural frequency
via the current. In Fig.$~$\ref{RFIG8}c we show the
coefficient of variation (CV) of an isolated neuron. When the neuron
is entrained the CV drops to zero. The ISIH then consists
of a single peak. The entrainment occurs for a range of current values,
$I=0.87$-$1.0$. Since the firing rate is constant, the $f-I$
has a flat step (not shown).
Outside the range of entrainment the ISIH
has more structure (Fig.$~$\ref{RFIG8}d and f).
When there is noise present
in the neuron, the CV will increase. For weak noise
the CV in the entrainment regime will still be lower compared to
the CV when the neuron is not entrained. For the network
the CV is also lower in the entrainment regime (Fig.$~$\ref{RFIG8}b),
though the CV increases
faster with the noise strength compared to the CV of the isolated neuron.
The synchronization of the network,
measured by $\kappa$ (Fig.$~$\ref{RFIG8}c),
is significantly enhanced in the region of entrainment
(for $D=0.004$-$0.04$). The enhanced synchronization
disappears for higher values $D>0.2$.
\subsection{Effect of synaptic coupling strength on robustness of strong synchronization}
We have also studied the effect of varying the synaptic coupling strength $g_{syn}$.
For large enough $D$ the network will be asynchronous.
We find that the network frequency in that case decreases with increasing
values of $g_{syn}$ (Fig.$~$\ref{RFIG9}).
For an asynchronous network the synaptic drive has a constant
tonic hyperpolarizing conductance, decreasing the firing rate.
The stronger the coupling the larger the decrease. The
synchronization measured by the
parameter $\kappa$ displays a different
behavior. In Fig.$~$\ref{RFIG9}c we plot the
$\kappa$ versus $g_{syn}$ for one specific value $D=0.02$ (and $I$ chosen
such that the firing rate is approximately $39$ Hz).
It is interesting to note that stronger coupling does not
necessarily mean a higher value of $\kappa$.
The coherence $\kappa$ has a local maximum for $g_{syn}=0.1$,
for higher values of $g_{syn}$, $\kappa$ decreases (see WB Fig.$~$12B).
For $g_{syn}>0.3$, $\kappa$ starts increasing again. We have studied
the underlying dynamics of this non monotonous behavior.
In Fig.$~$\ref{RFIG10}c-f we plot the ISIH for different
values of $g_{syn}$. For $g_{syn}\ge 0.2$ one finds more than
one peak. In the rastergrams (Figs.$~$\ref{RFIG10}a and
(b)), one can see that the dynamics corresponds
to a population that has a well defined frequency,
but individual neurons sometimes miss, or skip, a period. Despite
this small asynchrony when the neuron fires, it does so
in synchrony with the others. As a consequence the rastergrams looks
much more ordered compared to the one for $g_{syn}=0.1$ at the same
noise strength $D=0.002$.
\subsection{Larger $g_{syn}$ leads to robust stochastic weak synchronization}
In this subsection we discuss the robust $40$ Hz rhythms found for higher
$g_{syn}$ values. We have doubled the synaptic decay constant to
$\tau_{syn}=20$. Here we will use the modified $\kappa_W$ and $CV_W$
as mentioned in the Methods section.
In Fig.$~$\ref{EXTRA2}
we vary $g_{syn}$ from $0.05$ to $2.5$ with a spacing of $0.05$.
The neuron number is kept equal to $N=100$, and we use $I=2.0$,
and $\sigma_I=0$.
For $D=0.0$ and $\sigma_I=0$ the network is in a strongly synchronized
state, with the network
frequency $f_n$ the same as the single neuron firing rate $f$. The frequency
is exactly the same as the one for a single neuron with autosynaptic feedback, as one
would expect. This coherent state can be arrived at from
many different random initial conditions.
For weak noise, $D=0.008$,
the network stays in a strongly synchronized state for $g_{syn}<0.25$.
For higher $g_{syn}$ skipping starts to occur, the fractional cluster size
decreases from values close to one to values below one-half at $g_{syn}=1.2$.
At that point the network is in a real (albeit stochastic) cluster state,
on average the neuron only fires once every two cycles. We will refer to
all states for which certain active neurons do not fire at each cycle as
a stochastic weak synchronized (SWS)
network. The network frequency,
$f_n$, and the single neuron firing rate $f$ both decrease with
increasing $g_{syn}$.
When the network settles in the SWS state $f$ starts to differ considerably
from its value at the $D=0$ state. The strength of synchronization
increases with $g_{syn}$, that is $CV_W$ decreases and $\kappa_W$ increases.
For values $g_{syn}>2.0$, $CV_W$ and $\kappa_W$ slowly saturate.
For stronger noise $D=0.04$ and $D=0.20$ the network is asynchronous
for low values of $g_{syn}$. We have therefore excluded these points based
on the criterium discussed in the Methods section. The network frequency starts out
at a higher value, and the neurons fire at a lower rate compared
to the $D=0.008$ case. The strength of synchronization, $\kappa_W$ and $CV_W$,
is reduced compared to the one for $D=0.008$, but still increases with $g_{syn}$.
Note that all the neurons in the network still have a nonzero firing rate.
It is thus possible to obtain
weakly synchronized oscillations
in a network consisting of $100$ neurons
in the frequency range between $20$ and $40$ Hz.
We find that noise is necessary to obtain SWS.
We have studied SWS in the presence of weak current heterogeneities,
say for $\sigma_I=0.02$. Without noise ($D=0$) the network is in an
strongly synchronized state, and
$\kappa$ displays a maximum as a function of $g_{syn}$ (WB Fig12B).
One also clearly notices the effect of suppression \cite{white}:
for larger $g_{syn}$ the inhibition of faster spiking neurons stops
the firing of neurons driven by a smaller current. As a result the total
number $N_s$
of active neurons gradually drops (Fig.$~$\ref{EXTRA3}f).
For a small amount of noise, $D=0.008$, the situation changes dramatically.
A SWS state is obtained, and all neurons remain active ($N_s/N=1$), while
$\kappa$ saturates for $g_{syn}>0.5$, and the value of $\kappa$ for
$g_{syn}>0.8$ is even
higher than without noise. Thus noise may actually improve the coincidence.
Of course noise does increase the width, $CV_W$, of the peaks in the
instantaneous
firing rate. The single neuron firing rate decreases significantly compared
to that for $D=0$, whereas the network frequency is only weakly affected.
We have performed numerical simulations
for system sizes
$N=10$, $20$, $50$, $100$, $200$, and $1000$. We have used
the following parameters values: $I=5.0$, $D=0.2$, $g_{syn}=1$, $\tau_{syn}=20$,
and $\sigma_I=0.1$.
The network frequency increases with system size, whereas the firing rate
stays approximately constant with a dip around $N=50$. The measures
for coherence, $\kappa_W$ and $CV_W$, are also only weakly dependent on
system size.
The cycle-to-cycle fluctuations in cluster size vary approximately
as $\sqrt{N}$ (Fig.$~$\ref{EXTRA4}g).
The strength of the inhibition is determined by the
number of neurons that fired in the previous cluster, and
in turn it determines at what time the first neurons
become disinhibited. One therefore expects cluster size fluctuations
and cycle length fluctuations to be intimately related.
Indeed, the standard deviation of cycle length varies as $1/\sqrt{N}$ with
$N$ the number of neurons (Fig.$~$\ref{EXTRA4}h).
This means that larger networks are better at generating a precise cycle
length,
whereas size does not matter as much for the coincidence of spikes
measured by $\kappa_W$ and $CV_W$.
In each simulation we randomly draw a set of driving currents $I_j$ for
each neuron $j$ from a uniform probability distribution. The results one obtains may
critically depend on the
particular realization of driving currents. One expects that for
larger systems this is less of a problem. The population distribution
of $I$ is more likely to approach the original ensemble distribution
of currents for a given neuron.
Here we have studied the range of values for the measured quantities
($f_n$, $f$, and so on)
for ten different realizations. We find that for most
quantities (for these
parameter values) the range of values decreases with $N$,
and for $N\ge 500$ one realization will give a result close
to the expectation value.
We now vary $\sigma_I$ and $D$ for the following fixed parameter set
$N=1000$, $g_{syn}=2$, $\tau_{syn}=20$, and $I=3.5$.
For $D=0$ and $\sigma_I=0$ the network is strongly synchronized
at $20$ Hz. The instantaneous firing rate consists of a sequence
of regularly spaced delta functions (Fig.$~$\ref{EXTRA5}e),
the ISIH has a single delta peak at $50$ ms (Fig.$~$\ref{EXTRA5}b), and
all neurons spike at the same frequency (Fig.$~$\ref{EXTRA5}a).
Increasing $D$ increases the network frequency, but decreases the
single neuron firing rate (Fig.$~$\ref{RFIG15}). The population activity is
still periodic (Fig.$~$\ref{EXTRA5}f), but the peaks have a finite width
(as well as the ISIH), and the ISIH becomes multimodal. This process
continues with ISIH spreading out more and more, with the $CV_W$ increasing,
and $\kappa_W$ decreasing.
As mentioned before we need some noise to generate an SWS state.
Here we use $D=0.2$, while at the same time varying $\sigma_I$. For
finite $\sigma_I$
there is still a coherent population activity (Fig.$~$\ref{EXTRA5}g),
despite the fact that neurons have different firing rates (Fig.$~$\ref{EXTRA5}c).
Increasing $\sigma_I$ will reduce coherence, $\kappa_W$ decreases and
$CV_W$ increases. At the same time both $f_n$ and $f$ increase
(Fig.$~$\ref{RFIG15}d).
This is different from the effect of increasing $D$. Higher
$\sigma_I$ leads to
suppression, with fast spiking neurons preventing slower ones from firing, and as
a result part of the inhibition disappears, while further increasing
the firing rate
and its average (calculated from the active neurons).
On the other hand noise increases the tonic inhibition for each
neuron, and thus leads to
a reduced firing rate. Also the progression of the asynchronous state
is different.
The first peak in the ISIH becomes broader, and the higher order
ones have a reduced prominence (Fig.$~$\ref{EXTRA5}d). For increasing $D$ the
peaks just wash out.
\subsection{Comparison to externally induced stochastic weak synchronization}
We find that
robustness can be enhanced
by getting a network in a SWS state.
The single neuron discharge in an SWS state is very similar
to that obtained in
Stochastic Resonance \cite{Wies95} using a subthreshold {\it external} drive.
In this subsection we compare the previous self-induced SWS state
to the one induced by an external drive.
We drive the system by a sinusoidal current of amplitude $1.2$ (admittedly
large). The average value of the driving current is zero.
For weak noise the spikes of a single neuron are spaced many
cycles apart (Fig.$~$\ref{RFIG12}a). With increasing noise
the ISIH starts to look more like the ones in Fig.$~$\ref{EXTRA5}b. The
population activity
also is periodic (Fig.$~$\ref{RFIG12}e). The distribution
of $\tau_{ISI}$ in Fig.$~$\ref{RFIG12}a looks similar to the one
in Fig.$~$\ref{EXTRA5}a. For high values of $D$ the neuron
can spike more than once during one cycle, as a result the
number of coincident spikes is reduced.
The addition of current heterogeneity does not seem to affect
the network behavior (as long as
all the neurons are still below threshold). The different ISIH
in Fig.$~$\ref{RFIG12}d look very similar. However the corresponding
range of $\tau_{ISI}$
values in Fig.$~$\ref{EXTRA5}c does increase with $\sigma_I$.
\section{Discussion}
Previous authors have recognized that strong synchronization
is only moderately robust against neuronal heterogeneity \cite{wang,white}.
We have previously shown that the same holds including synaptic noise \cite{CNS}.
The basic premise of synchronization by mutual inhibition is
almost trivial, since the network consists of intrinsically periodically
spiking neurons. Their output produces a periodic synaptic drive,
which in turn is fed back into the network. Inhibition thus allows
a phase lock at zero phase with this drive. Heterogeneity
and noise reduces the phasic, and increases the tonic part
of the synaptic drive, leading to a reduction in synchronization, and
eventually leading to an asynchronous state (Fig.$~$\ref{RFIG5}c,d).
The synchronization
behavior of networks of physiological realistic neurons, however, is by no means fully
understood.
In this work we showed that the loss of synchronization
proceeds via {\bf different} mechanisms in the presence of synaptic noise
compared to the presence of current heterogeneity.
This is evident from the cross correlations shown
in Figs.$~$\ref{RFIG6} and $~$\ref{EXTRA1}.
We also found that the noise-induced precision loss
in the uncoupled neuron is exacerbated by the inhibitory coupling.
All of these could seem obvious based on previous
work on heterogeneity \cite{wang,white}. However, its consequences for real
life biological networks had not been fully appreciated.
Our results, combined with previous results, show that there
is a problem with strong synchronization by mutual inhibition, since
it is unlikely to occur in {\it in vitro} or {\it in vivo} systems.
[There are exceptions such as for example the pacemaker nucleus
in electric fish \cite{Moortgat98}, where the neurons are coupled via
gap junctions.]
The aim of this paper was to treat this problem: how can
one obtain robust synchronization in the presence of
synaptic noise and neuronal heterogeneity?
Our results are twofold. First, methods to increase
the robustness of strong synchronization have been ineffective.
Second, we showed that robust stochastic weak synchronization (SWS) can be obtained
for biophysically realistic parameter values. SWS is consistent with
previous experimental data.
In what follows we discuss these two important results
in more detail.
We believe that strong synchronization
is not robust enough. To make sure that we do
not prematurely discard strong synchronization
by mutual inhibition we have to make an effort to
increase robustness.
In this paper we discussed two simple methods
to increase robustness of strong synchronization.
One method was to increase the synaptic coupling $g_{syn}$.
Inhibition is responsible for synchronization.
It is then quite natural to expect that increasing the strength
of inhibition increases robustness. The fact that this does not
happen is surprising. For current heterogeneity this
is in part due to suppression \cite{white,wang}.
We have studied this effect for synaptic noise in more detail.
We found that neurons skip periods for higher values of $g_{syn}$
(see Fig.$~$\ref{RFIG10}).
In other words the strongly synchronized state becomes unstable,
and a weakly synchronized state emerges. This
weakly synchronized state looked more coherent (Fig.$~$\ref{RFIG10}a,b), and
provided the impetus to further study the robustness of the SWS states.
We also added a periodic drive to the neuron.
Recent experimental work shows that the CV of neurons
on an entrainment step is reduced compared to the CV outside
the step \cite{Cowan98}.
A clear physiological correlate in hippocampus of this drive is lacking
at present.
Since we try to reject our conjecture this lack of physiological realism
is not a problem. We found a moderate increase in robustness
(see Fig.$~$\ref{RFIG8}).
The periodic drive is not as effective as one would have intuited, however.
In fact the inhibitory connections reduce the increase in robustness compared
to the increase in the single uncoupled neuron (see Fig.$~$\ref{RFIG8}).
If we add a subthreshold periodic drive with
noise to a quiescent neuron we obtain weak synchronization. This is
known as Stochastic Resonance in excitable systems \cite{Wies95}.
To summarize: our attempts to significantly increase robustness of strong
synchronization failed. Instead we found weak
synchronization. Thus it is easier to find weak synchronization in parameter
space than it is to find strong synchronization.
If one accepts the fact, however, that strong synchronization
is not robust against noise and heterogeneity, and that periodic
population oscillations are found in experiments, then one has to
carefully consider the possible relevance of weak synchronization.
Weak synchronization as well as strong synchronization lead to a periodic
population discharge,
and specifically to an inhibitory synaptic drive indistinguishable from the one found
in pyramidal neurons in \cite{whittington}.
Moreover, the clusters that form in stochastic weak synchronization
bear a resemblance to the neuronal assemblies found
in some experiments \cite{Laurent96a}, and that are thought to play
a role in putative binding \cite{binding3}.
The question then is, is SWS more robust, and can it be found
for the gamma frequency range for biophysically realistic parameters?
What is needed is a higher
total synaptic conductance, and noise. The necessary amount of noise
is very small, $D>0.004$ is sufficient. The noise prevents the occurrence
of suppression (Fig.$~$\ref{EXTRA3}). In suppression the faster neurons
prevent the slower ones from firing. This reduces the inhibition of the
faster ones, and allows them to fire at different frequencies and at
random relative phases. Suppression is thus detrimental to synchronization.
We obtained SWS for different system sizes (we studied
networks from $10$ to $1000$ neurons). The coincidence properties
($\kappa_W$, and $CV_W$) did not vary much with size. The
temporal precision of the population oscillation, however, increases
approximately as $1/\sqrt{N}$ (Fig.$~$\ref{EXTRA4}h). Large networks can thus
produce
precise pacemaker rhythms. In addition the statistical quantities
in small networks show more variation with different realizations of
the current drive.
It is of considerable interest to understand why weak synchronization
is so much more robust and prevalent compared to strong synchronization.
In strong synchronization one requires an equal firing rate for each neuron,
and coincident spikes, while weak synchronization requires only coincident
spikes. By definition, then, weak synchronization is easier to generate.
In fact strong synchronization is intrinsically less robust, for it requires
phase locking between neuron pairs. This is only possible
(depending on intrinsic properties) for a small difference in driving
currents. There is a price to pay, there will be a small phase difference
between the firings of each neuron. Pairs with a large phase difference are
less stable against the influence of noise. It is instructive to consider this
problem as the nonlinear dynamics of a single neuron driven by
a periodic drive.
The neuron can be entrained on different n:m steps.
On these steps in the $f$-$I$ plot the neuron generates $n$
action potentials during
$m$ cycles of the periodic drive for a range of values of $I$
(or intrinsic frequency).
Noise induces jitter in the spike time, but the CV on the steps is reduced
compared to the CV outside the steps (and steps with higher $m$ values).
The entrainment produces a phase difference between the firing time
and the crest of the drive. The size of the phase difference depends on the
intrinsic frequency of the neuron. Robustness is less for current values
close to the edge of the step (with unfavorable phase differences).
Therefore noise will reduce the width of the phase-locking step (not shown).
Here we found network parameters for which coincidence could be maintained
despite highly variable firing rates of the neuronal populations.
Our work, and also a recent study \cite{white},
is to a large extent based on the recent contributions by Wang\&Buzsaki
\cite{wang}.
It is therefore important to briefly reiterate, and spell out how our work
extends the work of Wang \& Buzsaki,
and how it differs \cite{white}.
We included the effect of synaptic noise. We have shown that for the
purposes of our modeling work a Gaussian white noise current
can adequately reproduce experimental ranges of CV \cite{CNS98}.
Biophysically realistic amounts of noise do affect the synchronization
we have studied. The noise effects are also different from those of current
heterogeneity as was discussed before. The issue of noise was not addressed
by Wang \& Buzsaki.
Another important difference is that previous works \cite{wang,white}
studied only strong synchronization. Here we have proposed that stochastic
weak synchronization underlies the synchronized population oscillations
in the hippocampus. For this reason our
robust 40Hz population rhythms were obtained for
different values of the coupling parameters $g_{syn}$, $\tau_{syn}$,
and the driving current $I$, and the system size $N$
compared to previous work\cite{wang}.
In our computational work we actually needed a small amount of noise to obtain
weak synchronization.
The synchronization properties of
large networks
may be of some mathematical interest. Our networks are small,
and probably the behavior can change
quantitatively when increasing the network size significantly.
However, in this paper we only addressed the
question as to whether networks of physiologically realistic size and connectivity
can robustly synchronize. Recent experimental work suggests that
interneurons contact
on the order of $60$ other interneurons \cite{Sik95}. For that reason we
only vary our network size between $N=10$ and $1000$.
In the introduction we mentioned recent {\it in vivo} work
and the controversies on the functional role of synchronization.
Our work obviously does not contribute to the understanding of the function
of synchronization. An important question is what kind of synchronization
can be sustained in biophysically realistic networks.
Traub and coworkers \cite{traub,traub2,whittington,Traub98a} looked for
physiological correlates of the gamma rhythms using {\it in vitro}
and computational experiments. Their results show the crucial role of
inhibition,
and have provided much of the impetus for our work.
The nervous system produces, for some unknown reason, periodic population
activity using circuitry consisting of noisy and heterogeneous neurons.
Our results establish that it
is possible for inhibitory neurons to be the driving
force for synchronization under these conditions.
\section{Acknowledgments}
This work was partially funded by the Northeastern University CIRCS fund,
and the Sloan Center for Theoretical Neurobiology (PT).
We thank W-J Rappel for help during the initial stage of this
work, and TJ Sejnowski for useful suggestions.
Part of the calculations were performed at Northeastern University
High Performance Computer Center.
|
\section{Introduction}
Phase transitions in the early universe can give rise to a variety of topological defects which interest cosmologists as possible sources for the primordial density perturbations which seeded galaxy formation through gravitational instability \cite{BSV}. Defects are discontinuities in the vacuum and can be classified according to the topology of the vacuum manifold of the field theory model being used. Thus disconnected manifolds give domain walls, non-simply connected manifolds, strings, and vacuum manifolds with non-trivial $\Pi_2$ and $\Pi_3$ homotopy groups give monopoles and textures respectively. Defects are either local if the symmetry broken is gauged or global if it is not. The gravitational effects of both local strings \cite{LSTR} and monopoles \cite{LMON} and global domain walls \cite{GDW}, strings \cite{GSTR} and monopoles \cite{GM} have been studied extensively.
In this paper we will be concerned with texture defects, which were first proposed as possible seeds for large-scale structure formation by Turok \cite{TUROK1}. In the case of a local texture the gauge field can everywhere absorb the gradient of the texture scalar field and the defect becomes merely another vacuum configuration. Hence it is only textures arising from a broken global symmetry that will interest us. The scenario for texture formation and evolution can be put simply as follows. A non-Abelian global symmetry is broken as the early universe cools through a temperature of order the GUT-scale. This leads to a random distribution of knots, regions where the texture field winds around the vacuum manifold non-trivially. As the knots come within the cosmological horizon, the unstable texture field collapses to a very small size and unwinds, emitting a shell of Goldstone boson radiation. During their evolution, textures interact gravitationally with the cosmic fluid, inducing geometrical perturbations.
The gravitational effects of such defects were first considered by Turok and Spergel \cite{TUROK2} and N\"{o}tzold \cite{NOTZ} in the weak-field approximation. The strong gravity of textures was then investigated by Durrer \textit{et al} \cite{DUR} and Barriola and Vachaspati \cite{BV}. Guided by the fact that an exact self-similar solution to the texture field equation exists in flat space, Durrer \textit{et al} found self-similar solutions to the coupled Einstein field equations describing the metric and scalar field of a texture in curved space. The equations derived by Barriola and Vachaspati were related to those of Durrer \textit{et al} by a complicated co-ordinate transformation. Subsequently, the microwave background anisotropies induced in the texture model of structure formation have been the subject of much study (for example, see \cite{MICRO}). Over the next few years, as high resolution maps of the cosmic microwave sky become available, it should be possible to test this and other models of structure formation to high precision.
Of course, most work on defects has been performed within the context of general relativity. However, at sufficiently large energy scales it seems likely that gravity is not described solely by the Einstein action, but becomes modified, for example, by superstring terms which are scalar-tensor in nature \cite{LESG}. This low-energy superstring gravity is reminiscent of the scalar-tensor theories of Jordan, Brans and Dicke \cite{JBD}, and is in fact equivalent to Brans-Dicke theory for particular parameter values. The implications of scalar-tensor gravity for topological defects, be it Brans-Dicke or low-energy superstring theory, have been explored for local strings \cite{LSTRD}, global strings \cite{GSTRD,GSTRDUS} and global monopoles \cite{GMD,GMDUS}.
Here we will consider the implications of low-energy string gravity for global textures. We take a general form for the interaction of matter with the superstring dilaton - that is, we consider an arbitrary coupling $e^{2a\phi}{\mathcal{L}}$ between the texture lagrangian and the dilaton $\phi$. We will be concerned in particular with the questions of whether a non-singular spacetime exists for the texture, and of how the metric properties of the defect are affected by the dilaton mass. The layout of the paper is as follows. After briefly reviewing the work of Durrer \textit{et al} on textures in Einstein gravity, we present an analysis for the texture in superstring gravity, for both massless and massive dilatons. We then compare our results with those found for other global defects and present some conclusions.
\section{Textures in Einstein gravity}
In this section we will review the metric properties of textures in Einstein gravity (further details can be found in \cite{DUR}). A simple model giving rise to texture solutions is described by the Lagrangian
\begin{equation}
{\mathcal{L}}(\hat{\psi}^{i})=\frac{1}{2} \nabla_{\mu} \hat{\psi}^{i} \nabla^{\mu} \hat{\psi}^{i} - \frac{\lambda}{4} (\hat{\psi}^{i} \hat{\psi}^{i} - \eta^{2})^{2} \label{FullLagr}
\end{equation}
where $\hat{\psi^{i}}$ is a set of real scalar fields, ${i=1 \ldots 4}$. We will consider the behaviour of a single texture knot and put $\hat{\psi}=\eta \psi$. Note also that since we have not fixed $G=1$, we are free to set $\lambda \eta^2=1$ without loss of generality, which corresponds to using string rather than Planck units (now G is some small number of ${\mathcal{O}}(10^{-6})$). Then
\begin{equation}
{\mathcal{L}}(\psi^{i})=\eta^2 \left ( \frac{1}{2} \nabla_{\mu} \psi^{i} \nabla^{\mu} \psi^{i} - \frac{1}{4} (\psi^{i} \psi^{i} - 1)^{2} \right ) \label{ELagr}
\end{equation}
This model has a global $O(4)$ symmetry and the vacuum manifold is the three sphere $S^3$. The symmetry is spontaneously broken to $O(3)$ when the field $\psi^{i}$ acquires a vacuum expectation value. For grand-unified theories, $\eta$, the magnitude of this vacuum expectation value, is typically of the order $10^{16}$ GeV. Note that unlike other topological defects, the texture can remain in the vacuum manifold throughout space. In fact, the field will only leave the manifold when it has collapsed to a small enough size that its gradient energy is great enough for the texture to unwind. Hence, in order to make calculations more tractable, it is a good approximation to replace the potential term in (\ref{ELagr}) by the constraint
\begin{equation}
\psi^{i} \psi^{i} = 1
\end{equation}
The dynamics of the texture field are then determined by the Lagrangian
\begin{equation}
{\mathcal{L}_{\beta}}(\psi^{i})=\eta^2 \left ( \frac{1}{2} \nabla_{\mu} \psi^{i} \nabla^{\mu} \psi^{i} - \beta (\psi^{i} \psi^{i} - 1) \right )
\end{equation}
where $\beta$ is a Lagrange multiplier, giving the equation of motion
\begin{equation}
\nabla_{\mu} \nabla^{\mu} \psi^{i} + \left ( \nabla_{\mu} \psi^{j} \nabla^{\mu} \psi^{j} \right ) \psi^{i} = 0
\end{equation}
Field configurations which have a well-defined limit at spatial infinity fall into classes of the homotopy group $\Pi_3 (S^3)= Z\hskip -3mm Z $, the integers of this group corresponding to winding number.
We will look for spherically-symmetric configurations describing the texture and thus make the hedgehog ansatz
\begin{equation}
\psi^{i}= \left ( \cos \chi \left (r,t \right ),\sin {\chi \left (r,t \right )} \sin \theta \cos \xi ,\sin \chi \left (r,t \right ) \sin \theta \sin \xi ,\sin \chi \left (r,t \right ) \cos \theta \right ) \label{hedgehog}
\end{equation}
There are a number of equivalent gauges in which we could discuss solutions. We will choose to write the metric
\begin{equation}
ds^2=e^{2\gamma (r,t)} \left ( dt^2-dr^2 \right ) -r^2 \omega (r,t)^2 \left (d\theta^2+\sin^2 \theta d \xi^2 \right ) \label{metric}
\end{equation}
In terms of the new variable $\chi$ the Lagrangian becomes
\begin{equation}
{\mathcal{L}}=\eta^2 \left ( \textstyle{1\over2} e^{-2\gamma} \left ( \dot{\chi}^2 - {\chi ' }^2 \right ) - \displaystyle{\frac{\sin ^2 \chi}{r^2 \omega^2}} \right )
\end{equation}
where the dot and dash denote differentiation with respect to $t$ and $r$. The equation of motion for $\chi$ is
\begin{equation}
r^2 \left ( \omega^2 \dot{\chi} \right )\dot{ } - \left (r^2 \omega^2 \chi ' \right )' + e^{2\gamma} \sin 2\chi = 0
\end{equation}
In flat space this reduces to
\begin{equation}
\ddot{\chi}- \chi '' - \frac{2}{r}\chi ' + \frac{\sin 2 \chi}{r^2} = 0
\end{equation}
which has the exact self-similar solution for $t<0$
\begin{equation}
\chi=2 \arctan (-r/t) \label{flat}
\end{equation}
describing texture collapse. Note that as $-r/t \rightarrow \infty$, $\chi \rightarrow \pi$ and $\psi$ has the constant limit $\psi=(-1,\bf{0})$ corresponding to winding number one.
The texture couples to the metric via its energy-momentum tensor
\begin{equation}
G_{\mu\nu}=8 \pi G T_{\mu\nu}
\end{equation}
Rescaling the energy-momentum tensor, $\hat{T}_{\mu\nu}=T_{\mu\nu}/\eta^2$, we have
\begin{eqnarray}
\hat{T}_t^t & = & \textstyle{1\over2} e^{-2\gamma} \left ( \dot{\chi}^2 + {\chi '}^2 \right ) + \displaystyle{\frac{\sin^2 \chi}{r^2 \omega^2}} \nonumber \\
\hat{T}_t^r & = & -e^{-2\gamma} \dot{\chi} \chi ' \nonumber \\
\hat{T}_r^r & = & - \textstyle{1\over2} e^{-2\gamma} \left ( \dot{\chi}^2 + {\chi '}^2 \right ) + \displaystyle{\frac{\sin^2 \chi}{r^2 \omega^2}} \nonumber \\
\hat{T}_{\theta}^{\theta} & = & - \textstyle{1\over2} e^{-2\gamma} \left ( \dot{\chi}^2 - {\chi '}^2 \right )
\end{eqnarray}
After a little manipulation, Einstein's equations reduce to
\begin{eqnarray}
\omega \left ( \ddot{\omega} - \omega '' \right ) - \frac{4\omega \omega '}{r} + \frac{1}{r^2} \left ( e^{2\gamma}- \omega^2 \right ) + \dot{\omega}^2 - {\omega '}^2 & = & \frac{\epsilon e^{2\gamma} \sin^2 \chi}{r^2} \label{EinstOne} \\
{( r \omega )' } \, \dot{ } - \gamma' {(r \omega)}\dot{ }- \dot{\gamma} (r \omega)' & = & - \textstyle{1\over2} \epsilon (r \omega) \dot{\chi} \chi ' \label{EinstTwo} \\
\frac{1}{r \omega} \left ( (r \omega)'' - (r \omega)\dot{ } \, \dot{ } \right ) + \gamma'' - \ddot{\gamma} & = & \textstyle{1\over2} \epsilon \left ( \dot{\chi}^2 - {\chi '}^2 \right ) \label{EinstThree}
\end{eqnarray}
where $\epsilon=8 \pi G \eta^2$ measures the gravitational strength of the texture. In a GUT scenario $\epsilon \approx 10^{-5}$.
Guided by the fact that a self-similar solution to the texture field equation exists in flat space-time, we make the self-similar ansatz
\begin{equation}
\chi (r,t)=\chi (x), \ \gamma(r,t)=\gamma(x), \ \omega (r,t)=\omega (x)
\end{equation}
where $x=-r/t$. If a dash now denotes differentiation with respect to $x$, the gravitational field equations become
\begin{eqnarray}
{(x\omega)'}^2 - (x^2\omega ')^2 + (x\omega)(x\omega)''(1-x^2) & = & e^{2\gamma} (1-\epsilon \sin^2 \chi) \label{SEinstOne} \\
(x\omega)'' -\gamma' (2x\omega' + \omega) & = & -\textstyle{1\over2} \epsilon (x\omega) {\chi '}^2 \label{SEinstTwo} \\
\frac{(x \omega)''}{x \omega} + \frac{(\gamma' (1-x^2))'}{1-x^2} & = & - \textstyle{1\over2} \epsilon \chi'^2 \label{SEinstThree}
\end{eqnarray}
and the equation of motion for $\chi$ is
\begin{equation}
(x^2-1) (x^2\omega^2 \chi')' + e^{2\gamma} \sin 2\chi = 0
\end{equation}
Note however that (\ref{SEinstTwo}) and (\ref{SEinstThree}) imply
\begin{equation}
\frac{(\gamma'(1-x^2))'}{\gamma'(1-x^2)} + \frac{2\omega'}{\omega} + \frac{1}{x} = 0
\end{equation}
which can be immediately integrated to give
\begin{equation}
\gamma'= \frac{b}{x(1-x^2)\omega^2}
\end{equation}
where $b$ is some constant. If we make the assumption that $\gamma'$ should be finite at $x=0$ (i.e. when $r=0$ or as $t \rightarrow -\infty$) since we do not expect strong curvature effects here except at the instant the texture unwinds, we must have $b=0$ and so $\gamma=const$. Without loss of generality, we set $\gamma=0$. Hence we finally obtain the coupled equations relating the metric and texture fields $\omega$ and $\chi$
\begin{eqnarray}
(x \omega)'' + \textstyle{1\over2} \epsilon (x \omega) {\chi '}^2 & = & 0 \label{OmegaEq} \\
(x^2-1) (x^2 \omega^2 \chi')' + \sin 2\chi & = & 0 \label{ChiEq}
\end{eqnarray}
together with the first-order constraint equation
\begin{equation}
(x\omega)'^2 - (x^2\omega')^2 + \textstyle{1\over2} \epsilon (x^2-1) (x\omega \chi')^2 = 1- \epsilon \sin^2 \chi \label{EinsteinConstraint}
\end{equation}
We can easily obtain a linearised solution to these equations for the collapsing texture by expanding around the self-similar flat-space solution
\begin{equation}
\chi(x) = 2 \arctan x
\end{equation}
giving, to $O(\epsilon)$
\begin{equation}
\omega (x) \approx 1+ \epsilon \left ( \frac{\arctan (x)}{x} - 1 \right )
\end{equation}
with the boundary condition $\omega \rightarrow 1$ as $t \rightarrow -\infty$. To construct numerical solutions to the field equations we need the behaviour near $x=0$. We must have $\chi (0)=0$ or $\psi$ would be singular at the origin. Then requiring that $\omega (0)$ and $\omega '' (0)$ are finite we obtain $\omega' (0)=0$ from (\ref{OmegaEq}) and $\omega (0)=1$ from the constraint equation (\ref{EinsteinConstraint}). The field equations then imply the expansions
\begin{eqnarray}
\omega (x) & = & 1+ l x^2 + m x^4 + O (x^5) \nonumber \\
\chi (x) & = & \chi ' (0) x + n x^3 + O(x^5)
\end{eqnarray}
where
\begin{eqnarray}
l & = & - \frac{1}{12} \epsilon \chi ' (0) ^2 \nonumber \\
m & = & -\frac{1}{100} \epsilon \chi ' (0) ^ 2 \left ( 3 + \left ( \frac{19 \epsilon}{24} -2 \right ) \chi ' (0) ^2 \right ) \\
n & = & \frac{1}{15} \chi ' (0) ( 3+ (\epsilon -2) \chi ' (0)^2 ) \nonumber
\end{eqnarray}
The only parameter in the expansions is $\chi'(0)$ which is determined as follows. The equation of motion for $\chi$ (\ref{ChiEq}) has a critical point at $x=1$ and $\chi ''$ remains finite here only if $\sin 2\chi = 0$. Hence for any given $\epsilon$ the non-singular solution is found by determining $\chi'(0)$ such that $\chi (1)=\pi/2$.
Figure \ref{fig:pic1} shows the amplitude of $\psi$ for the collapsing texture when $\epsilon=10^{-1}$ together with the flat-space solution $\chi (x)= 2 \arctan x$. Note that this and all subsequent figures are plotted for $0<x=-r/t<1$ and $1<y=2+t/r<2$, and hence the whole of the collapsing texture solution $0<-r/t<\infty$ is shown. Even for the unreasonably large value of $\epsilon=10^{-1}$, $\chi$ remains close to the flat-space solution. For any $\epsilon >0$ the asymptotic value $\chi(\infty)$ always slightly overshoots the flat-space value $\pi$ and hence $\psi$ no longer has a well-defined limit at spatial infinity. The metric function $\omega$ is shown in Figure \ref{fig:pic2} for $\epsilon=10^{-1}$, $10^{-3}$ and $10^{-5}$ and is close to its linearised approximation for all values of $\epsilon$. Figure \ref{fig:pic3} shows the time-independent quantity
\begin{equation}
\hat{R}^2_{\mu \nu \rho \sigma}=t^4 R^2_{\mu \nu \rho \sigma} \propto \left ( \frac{1-{(x \omega)'}^2+ (x^2 {\omega}')^2}{x^2 \omega^2} \right ) ^2 + 2 \left ( \frac{(1-x^2)(x\omega)''}{x\omega} \right ) ^2 \label{Curve}
\end{equation}
for $\epsilon=10^{-1}$.
The solutions describe a collapsing texture which unwinds at $t=0$. The spacetime geometry has a simple interpretation. $\omega$ changes rapidly around the light-cone $x=1$, and is almost constant, $\omega=1$ for $x \ll 1$ and $\omega \approx 1-\epsilon$ for $x \gg 1$. Hence we have a shell of rapidly changing $\omega$ moving inwards at the speed of light. Inside the shell space is nearly flat, and outside it can be approximated by a space with constant deficit solid angle $\Delta \approx \epsilon$. During the actual unwinding event $(r,|t|) \le (1/\eta,1/\eta)$, the texture leaves the vacuum manifold, and hence the sigma-model approximation is inappropriate. However, the solutions after collapse ($t>0$) can be approximated by patching together sigma-model solutions such that the winding number vanishes after collapse (see \cite{DUR}), and $\omega$ now describes the curvature effects of an expanding shell of Goldstone boson radiation.
\section{Textures in dilaton gravity}
We are interested in the behaviour of the texture metric when gravitational interactions take a form typical of low-energy string theory. In its minimal form, string gravity replaces the gravitational constant $G$ by a scalar field, the dilaton. To account for the unknown coupling of the dilaton to matter, we choose the action
\begin{equation}
S=\int d^4 x \sqrt{-\hat{g}} \left [ e^{-2\phi} \left ( -\hat{R} - 4 (\hat{\nabla} \phi)^2 - \hat{V} (\phi) \right ) + e^{2a\phi} {\mathcal{L}} \right ]
\end{equation}
where ${\mathcal{L}}$ is as in (\ref{ELagr}) and $a$ is an arbitrary constant. The dilaton potential $\hat{V}(\phi)$ is for the moment assumed general. Note that this action is written in terms of the string metric which appears in the string sigma model. To facilitate comparison with the previous section we instead choose to write the action in terms of the `Einstein' metric
\begin{equation}
g_{\mu \nu}=e^{-2\phi} \hat{g}_{\mu \nu}
\end{equation}
in which the gravitational part of the action appears in the normal Einstein form
\begin{equation}
S= \int d^4 x \sqrt{-g} \left [ -R + 2 (\nabla \phi) ^2 -V(\phi) + e^{2(a+2)\phi} {\mathcal{L}} (\psi^i,e^{2\phi} g) \right ]
\end{equation}
where $V(\phi)=e^{2\phi} \hat{V} (\phi)$. As before, we expect the texture will only be forced to leave the vacuum manifold when it has collapsed to a very small size, and so impose $\psi^i \psi^i =1$. The dynamics of the texture field are then determined by the Lagrangian
\begin{equation}
{\mathcal{L}}_{\beta}= \textstyle{1\over2} e^{2(a+1)\phi} {\nabla}_{\mu} {\psi}^i {\nabla}^{\mu} {\psi^i} - \beta ({\psi}^i {\psi}^i-1)
\end{equation}
giving the equation of motion
\begin{equation}
\nabla_{\mu} \nabla^{\mu} \psi^i + 2(a+1) \nabla_{\mu} \phi \nabla^{\mu} \psi^i + (\nabla_{\mu} \psi^j \nabla^{\mu} \psi^j) \psi^i=0
\end{equation}
Einstein's equation becomes
\begin{equation}
G_{\mu \nu}=\textstyle{1\over2} e^{2(a+2)\phi} T_{\mu \nu} + S_{\mu \nu}
\end{equation}
where the energy-momentum tensor is now
\begin{equation}
T_{\mu \nu}= 2 \frac{\delta {\mathcal{L}}(\psi^i , e^{2\phi}g)}{\delta g^{\mu \nu}} - g_{\mu \nu} {\mathcal{L}}(\psi^i , e^{2\phi}g) = e^{-2\phi} \nabla_{\mu} \psi^i \nabla_{\nu} \psi^i - g_{\mu \nu} {\mathcal{L}}
\end{equation}
and
\begin{equation}
S_{\mu \nu} = 2 \nabla_{\mu} \phi \nabla_{\nu} \phi + \textstyle{1\over2} g_{\mu \nu} V(\phi) - g_{\mu \nu} (\nabla \phi)^2
\end{equation}
is the energy-momentum tensor of the dilaton, which has the equation of motion
\begin{equation}
- \Box \phi = \frac{1}{4} \frac{\partial V}{\partial \phi} - \frac{a+2}{2} e^{2(a+2)\phi} {\mathcal{L}} (\psi^i,e^{2\phi}g) + \frac{1}{4} e^{2(a+1)\phi} g^{\mu \nu} \nabla_{\mu} \psi^i \nabla_{\nu} \psi^i
\end{equation}
As before, we will look for spherically-symmetric configurations describing the texture (\ref{hedgehog}) and choose the metric to take the form (\ref{metric}). Then the Lagrangian is
\begin{equation}
{\mathcal{L}} = \eta^2 e^{-2\phi} \left ( \textstyle{1\over2} e^{-2\gamma} (\dot{\chi}^2-{\chi ' }^2) -\displaystyle{\frac{\sin^2 \chi}{r^2 \omega ^2}} \right )
\end{equation}
The rescaled, modified energy-momentum tensor, $\hat{T}_{\mu \nu}=e^{2(a+2)\phi}T_{\mu \nu}/\eta^2$, is given by
\begin{eqnarray}
\hat{T}_t^t & = & e^{2(a+1)\phi} \left (\textstyle{1\over2} e^{-2\gamma} \left ( \dot{\chi}^2 + {\chi '}^2 \right ) + \displaystyle{\frac{\sin^2 \chi}{r^2 \omega^2}} \right ) \nonumber \\
\hat{T}_t^r & = & - e^{2(a+1)\phi} e^{-2\gamma} \dot{\chi} \chi ' \\
\hat{T}_r^r & = & - e^{2(a+1)\phi} \left ( \textstyle{1\over2} e^{-2\gamma} \left ( \dot{\chi}^2 + {\chi '}^2 \right ) - \displaystyle{\frac{\sin^2 \chi}{r^2 \omega^2}} \right ) \nonumber \\
\hat{T}_{\theta}^{\theta} & = & - \textstyle{1\over2} e^{2(a+1)\phi} e^{-2\gamma} \left ( \dot{\chi}^2 - {\chi '}^2 \right )
\end{eqnarray}
and the Einstein equations may be conveniently written
\begin{eqnarray}
\omega \left ( \ddot{\omega} - \omega '' \right ) - \frac{4\omega \omega '}{r} + \frac{1}{r^2} \left ( e^{2\gamma}- \omega^2 \right ) + \dot{\omega}^2 - {\omega '}^2 & = & \frac{\epsilon e^{2(a+1)\phi} e^{2\gamma} \sin^2 \chi}{r^2} + \textstyle{1\over2} e^{2\gamma} \omega^2 V(\phi) \label{DilEinstOne} \\
{( r \omega )' }^{.} - \gamma' {(r \omega)}^{.}- \dot{\gamma} (r \omega)' & = & - \textstyle{1\over2} (r \omega ) \left (\epsilon e^{2(a+1)\phi} \dot{\chi} \chi ' + 2 \dot{\phi} \phi ' \right ) \label{DilEinstTwo} \\
\frac{1}{r \omega} \left ( (r \omega)'' - (r \omega)^{..} \right ) + \gamma'' - \ddot{\gamma} & = & \textstyle{1\over2} \epsilon e^{2(a+1)\phi} \left ( \dot{\chi}^2 - {\chi '}^2 \right ) \nonumber \\
& & + \dot{\phi}^2 - {\phi '}^2 - \textstyle{1\over2} e^{2\gamma} V(\phi) \label{DilEinstThree}
\end{eqnarray}
where $\epsilon=\eta^2/2$. The equation of motion for the texture field $\chi$ is
\begin{equation}
r^2 (\omega ^2 \dot{\chi})\dot{ } - (r^2 \omega^2 \chi ')' + 2(a+1)r^2 \omega^2 (\dot{\phi} \dot{\chi} - \phi ' \chi ') +e^{2\gamma} \sin 2\chi = 0
\end{equation}
and the dilaton equation is
\begin{equation}
r^2 (\omega ^2 \dot{\phi} )\dot{} - (r^2 \omega^2 \phi')' + \frac{1}{4} e^{2\gamma}r^2 \omega^2 \frac{\partial V}{\partial \phi} -\epsilon (a+1) e^{2(a+1)\phi} \left ( \textstyle{1\over2} r^2 \omega^2 (\dot{\chi}^2-{\chi'}^2) -e^{2\gamma} \sin^2 \chi \right ) = 0 \label{FullDil}
\end{equation}
We will consider massless and massive dilatons in turn.
\subsection{Massless dilatonic gravity}
For the massless dilaton $V(\phi)=0$. As in the Einstein case, it is consistent to make the self-similar ansatz
\begin{equation}
\chi (r,t)=\chi (x), \ \gamma(r,t)=\gamma(x), \ \omega (r,t)=\omega (x), \ \phi (r,t)=\phi (x)
\end{equation}
and rewrite the equations in terms of the new variable $x=-r/t$. Again, it is simple to show that on the assumption that $\gamma$ is regular at $x=0$, we must have $\gamma'=0$ where $'$ denotes differentiation with respect to $x$. Hence the coupled equations for the metric, texture and dilaton fields reduce to
\begin{eqnarray}
(x\omega)'' + (x\omega) (\textstyle{1\over2} \epsilon e^{2(a+1)\phi} {\chi'}^2 + {\phi '}^2) & = & 0 \label{MasslessDilaton1} \\
(x^2-1) (x^2 \omega^2 \chi ')' +2(a+1) x^2 \omega^2 \phi' \chi ' (x^2-1) + \sin 2\chi & = & 0 \label{MasslessDilaton2} \\
(x^2-1) (x^2 \omega^2 \phi' )' +\epsilon (a+1) e^{2(a+1)\phi} \left ( -\textstyle{1\over2} x^2 \omega^2 {\chi'}^2 (x^2-1) +\sin^2 \chi \right ) & = & 0 \label{MasslessDilaton3}
\end{eqnarray}
together with the first-order constraint equation
\begin{equation}
{(x \omega)'}^2 - (x^2 \omega')^2 + (x^2-1)(x\omega)^2 (\textstyle{1\over2} \epsilon e^{2(a+1)\phi} {\chi'}^2 +{\phi'}^2) = 1-\epsilon e^{2(a+1)\phi} \sin^2 \chi \label{MasslessConstraint}
\end{equation}
We can find linearised solutions to the coupled field equations by expanding around the flat-space, self-similar solution. To $O(\epsilon)$
\begin{eqnarray}
\omega (x) & \approx & 1 + \epsilon e^{2(a+1)\phi_0} \left ( \frac{\arctan x}{x} -1 \right ) \\
\phi (x) & \approx & \phi_0 + \epsilon \left ( c_1 + \frac{c_2}{x} + (a+1)e^{2(a+1)\phi_0} \times \right . \nonumber \\
& & \left . \left ( \ln \sqrt{ \left | \frac{1+x^2}{1-x^2} \right | } - \frac{1}{x} \left ( \arctan x + \ln \sqrt{\left | \frac{1+x}{1-x} \right | } \right ) \right ) \right )
\end{eqnarray}
where $\phi_0$, $c_1$ and $c_2$ are constants. We expect space to be flat before texture collapse $t \rightarrow -\infty$, and during collapse at $r=0$. Hence we will impose $\phi=0$ when $x=0$. This fixes the constants $\phi_0$ and $c_i$ and we obtain
\begin{eqnarray}
\omega(x) & \approx & 1 +\epsilon \left ( \frac{\arctan x}{x} - 1\right ) \label{LinDilOmega} \\
\phi(x) & \approx & \epsilon (a+1) \left ( 2+ \ln \sqrt{\left | \frac{1+x^2}{1-x^2} \right |} - \frac{1}{x} \left ( \arctan x + \ln \sqrt{\left | \frac{1+x}{1-x} \right |} \right ) \right ) \label{LinDilDil}
\end{eqnarray}
Note that if $a=-1$ then $\phi=0$ to first order. Although (\ref{MasslessDilaton1}) includes the dilaton field even if $a=-1$, in this case a trivial dilaton $\phi=0$ satisfies the boundary conditions and the equations are reduced to those seen in Einstein gravity. This is in contrast to the behaviour of other global topological defects in string gravity (see \cite{GSTRDUS,GMDUS}) where the dilaton is non-trivial for $a=-1$, and is due to the texture field being constrained to remain in the vacuum manifold. Note also that although $\phi$ is continuous in the linearised approximation, $\phi'$ diverges at $x=1$. This is an indication of qualitatively different behaviour that will make the solution of the full field equations slightly more complicated than in the Einstein case.
We attempt to find numerical solutions as before. We first need the behaviour of the fields near $x=0$. Taking $\phi(0)=0$, again we must have $\chi(0)=0$ or $\psi$ would be singular at the origin. Then requiring that $\omega(0)$, $\omega''(0)$ and $\phi'(0)$ are finite implies $\omega'(0)=0$ from (\ref{MasslessDilaton1}). Thus from the constraint equation (\ref{MasslessConstraint}), $\omega(0)=1$ and we obtain the expansions
\begin{eqnarray}
\omega (x) & = & 1 + l x^2 + m x^4 + O(x^5) \nonumber \\
\chi (x) & = & \chi' (0) x + n x^3 + O(x^5) \\
\phi (x) & = & p x^2 + q x^4 + O(x^5) \nonumber
\end{eqnarray}
where
\begin{eqnarray}
l & = & -\frac{1}{12} \epsilon \chi'(0)^2 \nonumber \\
m & = & -\frac{1}{100} \epsilon \chi'(0)^2 \left ( 3+ \left ( \epsilon \left ( k^2 + \frac{19}{24} \right ) -2 \right ) \chi'(0)^2 \right ) \nonumber \\
n & = & \frac{1}{15} \chi'(0) \left ( 3+ \left ( \epsilon \left ( 1-\frac{3}{2} k^2 \right ) -2 \right ) \chi'(0)^2 \right ) \\
p & = & \frac{1}{4} \epsilon k \chi'(0)^2 \nonumber \\
q & = & \frac{1}{80} \epsilon k \chi'(0)^2 \left (8+ \left ( \epsilon \left ( k^2 + \frac{8}{3} \right ) -4 \right ) \chi'(0)^2 \right ) \nonumber
\end{eqnarray}
and $k=a+1$. As before, the only free parameter in the expansions is $\chi'(0)$ and the equation of motion for $\chi$ (\ref{MasslessDilaton2}) has a critical point at $x=1$. Note however that the dilaton equation (\ref{MasslessDilaton3}) also has a critical point here and we cannot simply set $\chi(1)=\pi/2$ to ensure the continuity of all derivatives at $x=1$. In fact, as can be seen from the field equations, only one of $\chi$ and $\phi$ can be differentiable at $x=1$. We choose to adjust $\chi'(0)$ so that $\chi'$ remains finite to facilitate comparisons with the Einstein solutions. Then $\phi'$ will not remain finite as $x\rightarrow 1$ and, as can be seen from (\ref{MasslessDilaton1}), neither will $\omega''$. Simple numerical integration will fail at $x=1$.
Instead, we integrate from $x=0$ to $x^{-}$ close to $x=1$. Using $\omega''(x^{-})$ and $\chi''(x^{-})$ obtained from the field equations, we can estimate the values of $\chi$ and $\omega$ and their derivatives and $\phi$ at $x=1$, remembering that $\phi'$ is not finite here. Again, we adjust $\chi'(0)$ so that $\chi(1) = \pi/2$. Rewriting the field equations in terms of $v=1/x$,
\begin{eqnarray}
\ddot{\omega} + \omega \left ( \textstyle{1\over2} \epsilon e^{2(a+1)\phi} {\dot{\chi}}^2 + {\dot{\phi}}^2 \right ) & = & 0 \\
(\omega^2 \dot{\chi}) \dot{\ } + 2(a+1) \omega^2 \dot{\phi} \dot{\chi} + \frac{\sin 2\chi}{1-v^2} & = & 0 \\
{(\omega^2 \dot{\phi})} \dot{\ } + \epsilon (a+1) e^{2(a+1)\phi} \left ( -\textstyle{1\over2} \omega^2 \dot{\chi}^2 + \displaystyle{\frac{\sin^2 \chi}{1-v^2}} \right ) & = & 0
\end{eqnarray}
where a dot indicates differentiation with respect to $v$, we can integrate from $v=0$ to $v^{-}=1/x^{-}$ and estimate the values of the variables at $v=1$. Then ensuring that the values obtained from both integrations for $\omega$, $\omega'$, $\chi$, $\chi'$ and $\phi$ match at $x=v=1$ we should have obtained complete solutions to the field equations such that $\chi$ and $\omega$ are continuous and once-differentiable and $\phi$ is continuous but not differentiable at $x=1$. Of course, some error is involved in the estimates we have made, but the results are reasonably robust to changes in $x^{-}$.
Figures \ref{fig:pic4} and \ref{fig:pic5} show respectively the amplitude of $\psi$ and the metric function $\omega$ for the collapsing texture for $\epsilon=10^{-3}$ and $|a+1|=5,10$ and $15$ together with the solutions found in Einstein gravity ($a=-1$). Figure \ref{fig:pic6} shows the modulus of the dilaton field $|\phi|$ for $\epsilon=10^{-3}$ and $|a+1|=5,10$ and $15$, together with the linearised approximation (\ref{LinDilDil}). Finally, Figure \ref{fig:pic7} shows the time-independent quantity $\hat{R}^2_{\mu \nu \rho \sigma}=t^4 R^2_{\mu \nu \rho \sigma}$ for the same solutions. Under the sign change $a+1 \rightarrow -(a+1)$, the solutions for $\chi$ and $\omega$ are unaltered, but $\phi \rightarrow -\phi$.
Note that as $|a+1|$ increases, the gradient of $\chi$ at $x=1$ progressively decreases and that $\hat{R}^2_{\mu \nu \rho \sigma}$ develops a `kink' here, an indication that curvature effects are increasing on the light-cone. In fact, beyond a certain value of $|a+1|$, our numerical procedure fails and such solutions can no longer be found. We will now show that for any $\epsilon$, there is a value of $|a+1|$ beyond which non-singular texture solutions do not exist in massless dilatonic gravity.
Consider the equation of motion for the texture field $\chi$ (\ref{MasslessDilaton2}) which we rewrite as
\begin{equation}
\frac{(x^2 \omega^2 {\chi}')'}{x^2 \omega^2 \chi'} = -2(a+1){\phi}' + \frac{\sin 2\chi}{x^2 (1-x^2) \omega^2 \chi'}
\end{equation}
For some small $\kappa$, with $1 \ge x > \kappa$, we can integrate this to give
\begin{equation}
\ln(x^2 \omega^2 \chi') - \ln({\kappa}^2 \omega (\kappa)^2 \chi'(\kappa)) = -2(a+1)(\phi (x)-\phi (\kappa)) + \int_{\kappa}^{x} \frac{\sin 2\chi}{\hat{x}^2 (1-\hat{x}^2) \omega^2 \chi'} d\hat{x}
\end{equation}
Then taking exponentials and letting $\kappa \rightarrow 0$ gives
\begin{equation}
\chi'(x) = \frac{1}{x^2 \omega^2} \exp \left ( -2(a+1)\phi + \int_{0}^{x} \frac{\sin 2\chi}{\hat{x}^2 (1-\hat{x}^2) \omega^2 \chi'} d\hat{x} \right ) \label{TextureInt}
\end{equation}
Suppose that $\chi'$ decreases to zero at some $x_0 \le 1$. Then given that the integrand in (\ref{TextureInt}) is positive on the interval $[0,x_0]$, we must either have $\exp(2(a+1)\phi)\rightarrow \infty$ or $|\omega| \rightarrow \infty$ as $x \rightarrow x_0$. Now if $\omega \rightarrow 0$ on $[0,x_0]$, then $\exp(2(a+1)\phi) \rightarrow \infty$ for $\chi'$ to remain finite. But (\ref{MasslessDilaton1}) implies $\omega \le 1$, so we definitely must have $\exp(2(a+1)\phi) \rightarrow \infty$ as $x \rightarrow x_0$. Then consider the constraint equation (\ref{MasslessConstraint}) which we write as
\begin{equation}
-\frac{1 - {(x\omega)'}^2 + (x^2 \omega')^2}{x^2\omega^2}+ \frac{(1-x^2) (x\omega)''}{x\omega} = - \frac{\epsilon e^{2(a+1)\phi} \sin^2 \chi}{x^2 \omega^2}
\end{equation}
Since we have $0 < \chi(x_0) \le \pi/2$, the RHS tends to $- \infty$ as $x \rightarrow x_0$. So at least one of the terms on the LHS must tend to $- \infty$. In either case, the quantity $t^4 R^2_{abcd}$ (\ref{Curve}) becomes infinite as $x \rightarrow x_0$ and the spacetime is singular.
It remains to be shown that as we increase $|a+1|$ we necessarily approach a solution for which $\chi'(1)=0$. By integrating the dilaton equation (\ref{MasslessDilaton3}) we obtain :
\begin{equation}
(a+1)\phi'(x) = \frac{\epsilon (a+1)^2}{x^2 \omega (x)^2} \int_0^{x} e^{2(a+1)\phi} \left ( \textstyle{1\over2} \hat{x}^2 \omega^2 {\chi'}^2 + \displaystyle{\frac{\sin^2 \chi}{1-\hat{x}^2}} \right ) d\hat{x} \label{DilatonInt}
\end{equation}
Hence for any $x \in [0,1]$, $(a+1)\phi'(x) > 0$. Now consider the equation of motion for $\chi$ which we integrate to give :
\begin{equation}
\omega(1)^2 \chi'(1) = -2(a+1) \int_0^1 x^2 \omega^2 \phi' \chi' dx + \int_0^1 \displaystyle{\frac{\sin 2\chi}{1-x^2}} dx \label{Chi1}
\end{equation}
Clearly, the first term on the RHS is negative, whilst the second is positive. Then note that (\ref{DilatonInt}) also shows that
\begin{equation}
(a+1) x^2 \omega^2 \phi' > \textstyle{1\over2} \epsilon (a+1)^2 \displaystyle{\int_0^x} e^{2(a+1)\phi} \hat{x}^2 \omega^2 {\chi'}^2 d\hat{x}
\end{equation}
since the other term in the integrand in (\ref{DilatonInt}) is always positive on $[0,1]$. But (\ref{TextureInt}) gives
\begin{equation}
x^2 \omega^2 \chi' = \exp \left ( -2 (a+1) \phi + \int_0^x \frac{\sin 2\chi}{\hat{x} (1-\hat{x}^2) \omega^2 \chi'} d\hat{x} \right ) > e^{-2(a+1)\phi}
\end{equation}
implying
\begin{eqnarray}
(a+1) x^2 \omega^2 \phi' & > & \textstyle{1\over2} \epsilon (a+1)^2 \displaystyle{\int_0^x} e^{2(a+1)\phi} \hat{x}^2 \omega^2 {\chi'}^2 d\hat{x} \nonumber \\
& > & \textstyle{1\over2} \epsilon (a+1)^2 \displaystyle{\int_0^{x}} \chi' d\hat{x} = \textstyle{1\over2} \epsilon (a+1)^2 \chi (x)
\end{eqnarray}
Hence
\begin{equation}
2(a+1) \int_0^1 x^2 \omega^2 \phi' \chi' dx > \epsilon (a+1)^2 \int_0^1 \chi \chi' dx = \textstyle{1\over2} \epsilon (a+1)^2 [\chi^2]_0^1
\end{equation}
and since we require $\chi(1)=\pi/2$, we have
\begin{equation}
2(a+1) \int_0^1 x^2 \omega^2 \phi' \chi' dx > \frac{\epsilon (a+1)^2 \pi^2}{8}
\end{equation}
Now consider the other term on the RHS of (\ref{Chi1})
\begin{equation}
\int_0^1 \frac{\sin 2\chi}{1-x^2} dx = \textstyle{1\over2} \displaystyle{\int_0^1} \displaystyle{\frac{\sin 2\chi}{1+x}} dx + \textstyle{1\over2} \displaystyle{\int_0^1} \displaystyle{\frac{\sin 2\chi}{1-x}} dx
\end{equation}
For this to grow without bound as $|a+1|$ increases we would certainly need, for example
\begin{equation}
\frac{\sin 2\chi}{1-x} > \frac{1}{\sqrt{1-x}}
\end{equation}
in a neighbourhood of $x=1$, since the integral $\int_0^1 (1-x)^{-1/2}=2$ shows no such extreme behaviour. But then $\sin 2\chi > \sqrt{1-x}$ in this region, requiring
\begin{equation}
|\chi' \cos 2\chi| > \frac{1}{4\sqrt{1-x}}
\end{equation}
As $\cos 2\chi \approx -1$, this is clearly not the case since we are looking for solutions for which $\chi'(1)$ remains finite. Hence for any value of $\epsilon$, there must be a critical value of $c=|a+1|$ above which
\begin{equation}
2(a+1) \int_0^1 x^2 \omega^2 \phi' \chi' dx > \frac{\epsilon (a+1)^2 \pi^2}{8} > \int_0^1 \frac{\sin 2\chi}{1-x^2} dx
\end{equation}
For $|a+1|>c$, we must have $\chi'(x_0)=0$ for some $x_0 \le 1$ and the texture spacetime is singular.
We can obtain a rough estimate of the critical value of $|a+1|$ by using the flat-space solution for the texture, $\chi_0=2 \arctan x$, for which
\begin{equation}
\displaystyle{\int_0^1} \displaystyle{\frac{\sin 2\chi_0}{1-x^2}} dx = 1
\end{equation}
Then
\begin{equation}
|a+1| \approx \frac{2}{\pi} \sqrt{\frac{2}{\epsilon}}
\end{equation}
In fact the actual critical value is considerably lower than this. As we approach this value, numerical investigation shows
\begin{equation}
\displaystyle{\int_0^1} \displaystyle{\frac{\sin 2\chi}{1-x^2}} dx \approx 1/2 \Rightarrow |a+1| \approx \frac{2}{\pi} \sqrt{\frac{1}{\epsilon}}
\end{equation}
agreeing well with the values we have found.
For $\epsilon=10^{-3}$, the critical value of $|a+1|$ is
\begin{equation}
|a+1| \approx \frac{2}{\pi}\sqrt{\frac{1}{10^{-3}}} \approx 20
\end{equation}
Figures \ref{fig:pic4} - \ref{fig:pic7} indicate that as we increase $|a+1|$ towards the critical value we see progressively larger deviations from the Einstein theory solutions. The asymptotic values of both $\chi$ and $\omega$ are reduced from their values in Einstein theory. The geometrical interpretation of solutions is as before; however, as $|a+1|$ increases, $\omega$ changes more rapidly near the light cone, and the solid deficit angle of the space outside this region is progressively greater. Note also from Figure \ref{fig:pic6} that as $|a+1|$ increases, the deviations of $\phi$ from the linearised solution are large. It is unclear how far to trust the solutions obtained for $|a+1|$ near the critical value, as errors introduced by our numerical procedure may become significant.
For the non-singular solutions we have found with $|a+1|$ less
than the critical value, $R^2_{\mu \nu \rho \sigma}$ remains finite
even though $\phi'$ and $\omega''$ diverge as $x \rightarrow 1$. Thus
the null hypersurface $x=1$ is not a scalar polynomial curvature
singularity; that is, any polynomial constructed from the scalars $R^2_{\mu \nu \rho \sigma}$, $R^2_{\mu \nu}$ and $R$ remains finite here. However, we should note that the possibility remains that this hypersurface is a non-scalar polynomial curvature singularity.
Indeed, it seems that the `non-singular' behaviour observed
is confined to the Einstein frame. If we make the inverse transformation
$\hat{g}_{\mu \nu}=e^{2\phi} g_{\mu \nu}$ back to the string frame, the
properties of the conformal factor are included in the curvature. Even
though the metric function $\hat{\omega}=e^{\phi}\omega$ remains continuous, its
derivative diverges. Examining $R^2_{\mu \nu \rho \sigma}$ in the string frame, we see that the hypersurface $x=1$ has become a scalar polynomial curvature singularity. By examining the geodesic equations in a linearised approximation in the string frame, one can show that null and timelike geodesics reach the null hypersurface $x=1$ at finite affine parameter. We are thus able to conclude that there are no non-singular texture solutions in the string frame unless $a=-1$, at least if we impose the self-similar ansatz.
Finally, it is instructive to note that we can obtain the linearised solution (\ref{LinDilDil}) directly from (\ref{FullDil}) by a method other than direct integration of the self-similar field equations. Expanding in powers of $\epsilon$ about the flat space solution $\chi=\arctan(-r/t)$ we obtain the following equation for $\hat{\phi_1}=r \phi_1$
\begin{equation}
\ddot{\hat{\phi_1}}- {\hat{\phi_1}}''= 2(a+1)r \frac{r^2-3t^2}{(r^2+t^2)^2}
\end{equation}
This is simply the inhomogenous wave equation. The Green's function appropriate for the boundary conditions we require, that is $\hat{\phi_1} \rightarrow 0$ as $t \rightarrow -\infty$ and at $r=0$, is
\begin{equation}
G(r,t;r',t')=- \textstyle{1\over2} H[(r-r')^2 - (t-t')^2]
\end{equation}
where $H$ is the Heaviside step-function. Hence we can write down the solution for $\phi_1$ as
\begin{eqnarray}
\phi_1 & = & - \frac{2(a+1)}{r} \int_0^{\infty} dr' \int_{-\infty}^{\infty} dt' \ G(r,t;r',t') r' \frac{r'^2-3t'^2}{(r'^2+t'^2)^2} \nonumber \\
& = & - \frac{a+1}{r} \int_0^r dr' \int_{t-(r-r')}^{t+(r-r')} dt' \ r' \frac{r'^2-3t'^2}{(r'^2+t'^2)^2}
\end{eqnarray}
This expression, valid for both $r<|t|$ and $r>|t|$, indeed gives (\ref{LinDilDil}) after some calculation. This will enable us to write down a solution in the case of the massive dilaton, to which we now turn.
\subsection{Massive dilatonic gravity}
For the massive dilaton, we take $V(\phi)=2m^2 \phi^2$ where the mass $m$ is measured in units of the Higgs mass. We do not expect this to be the exact form of the potential, but for small perturbations of the dilaton away from its vacuum value, we might expect a quadratic form to be a good approximation. We take $10^{-11} \le m \le 1$, representing a range for the unknown dilaton mass of 1 TeV - $10^{15}$ GeV. The gravitational field equations are
\begin{eqnarray}
\omega \left ( \ddot{\omega} - \omega '' \right ) - \frac{4\omega \omega '}{r} + \frac{1}{r^2}
\left ( e^{2\gamma}- \omega^2 \right ) & + & \dot{\omega}^2 - {\omega '}^2 =
\frac{\epsilon e^{2(a+1)\phi} e^{2\gamma} \sin^2 \chi}{r^2} + m^2 e^{2\gamma} \omega^2 \phi^2
\label{MassDilEinstOne} \\
{( r \omega )' }^{.} - \gamma' {(r \omega)}^{.}- \dot{\gamma} (r \omega)' & = & - \textstyle{1\over2} (r \omega)
\left (\epsilon e^{2(a+1)\phi} \dot{\chi} \chi ' + 2 \dot{\phi} \phi ' \right )
\label{MassDilEinstTwo} \\
\frac{1}{r \omega} \left ( (r \omega)'' - (r \omega)^{..} \right ) + \gamma'' - \ddot{\gamma} & = &
\textstyle{1\over2} \epsilon e^{2(a+1)\phi} \left ( \dot{\chi}^2 - {\chi '}^2 \right ) + \dot{\phi}^2 -
{\phi '}^2 - m^2 e^{2\gamma} \phi^2 \label{MassDilEinstThree}
\end{eqnarray}
and the dilaton equation is
\begin{equation}
r^2 (\omega ^2 \dot{\phi} )^{.} - (r^2 \omega^2 \phi')' + m^2 e^{2\gamma} r^2
\omega^2 \phi -\epsilon (a+1) e^{2(a+1)\phi} \left ( \textstyle{1\over2} r^2 \omega^2
(\dot{\chi}^2-{\chi'}^2) -e^{2\gamma} \sin^2 \chi \right ) = 0
\end{equation}
The presence of the $m^2 e^{2\gamma} r^2 \omega^2 \phi$ term means we can no longer reduce the equations to a system of ordinary differential equations by writing them in terms of the self-similar variable $x=-r/t$. Instead, we expand the equations about flat space as they are. Then to $O(\epsilon)$, the dilaton equation gives
\begin{equation}
\ddot{\hat{\phi}_1} - {\hat{\phi}_1}'' + m^2 \hat{\phi}_1 = 2 (a+1) r \displaystyle{\frac{r^2-3t^2}{(r^2+t^2)^2}} \label{Klein}
\end{equation}
where $\phi=\epsilon \phi_1+\ldots$ and $\hat{\phi}_1=r\phi_1$. (\ref{Klein}) is the inhomogenous Klein-Gordon equation. From the analysis above for the massless dilaton, we know the prescription that will give the appropriate boundary conditions as $t \rightarrow -\infty$ and at $r=0$. That is, the correct Green's function is
\begin{equation}
G(r,t;r',t') = - \textstyle{1\over2} H[(r-r')^2-(t-t')^2] J_0 [m\sqrt{(r-r')^2-(t-t')^2}]
\end{equation}
where $J_0$ is the Bessel function of order zero. Hence the solution for $\phi_1$ is
\begin{equation}
\phi_1 = -\frac{a+1}{r} \int_0^r dr' \int_{t-(r-r')}^{t+(r-r')} dt' \ J_0 [m\sqrt{(r-r')^2-(t-t')^2}] \ r' \frac{r'^2-3t'^2}{(r'^2+t'^2)^2} \label{MassiveSol}
\end{equation}
Note that for small argument $y$, $J_0 (y) \approx 1 - y^2/4$. Hence if $mr \ll 1$, (\ref{MassiveSol}) gives
\begin{equation}
\phi_1 \approx - \frac{a+1}{r} \int_0^r dr' \int_{t-(r-r')}^{t+(r-r')} dt' \ \left ( 1-\frac{m^2}{4} \left ( (r-r')^2 -(t-t')^2 \right ) \right ) r' \frac{r'^2-3t'^2}{(r'^2+t'^2)^2}
\end{equation}
If we further impose $r \ll t$, then to next to leading order we obtain
\begin{eqnarray}
\phi_1 & \approx & \frac{3(a+1)}{r} \int_0^r dr' \int_{t-(r-r')}^{t+(r-r')} dt' \ r' \left ( \frac{1}{t'^2} + \frac{m^2}{4} \left ( \frac{t}{t'}-1 \right )^2 \right ) \nonumber \\
& = & 3 (a+1) \left ( \left ( 2- \ln (1-x^2) - \frac{1}{x} \ln \left ( \frac{1+x}{1-x} \right ) \right ) \right . \nonumber \\
& & \left . + \frac{m^2t^2}{4} \left ( 4+ \frac{x^2}{3} -3 \ln (1-x^2) -\frac{1}{x} (2+x^2) \ln \left ( \frac{1+x}{1-x} \right ) \right ) \right ) \label{SmallRSol}
\end{eqnarray}
where $x=-r/t$. For the values of $r$ and $t$ under consideration, (\ref{SmallRSol}) remains very close to the massless solution (\ref{LinDilDil}). For $r \gg t$, where we expect $\ddot{\phi_1}, \ \phi_1'' \approx 0$, we obtain from (\ref{Klein})
\begin{equation}
\phi_1 \approx \frac{2(a+1)}{m^2 r^2}
\end{equation}
Thus the texture supports a diffuse dilaton cloud. This power law fall-off of a massive scalar field seems counterintuitive until one remembers that the dilaton is in fact part of the gravitational sector of the theory, and therefore couples to the energy-momentum of the texture. The slow fall-off of this energy-momentum is what supports the diffuse dilaton cloud.
Finally, note that in the case of other global topological defects \cite{GSTRDUS,GMDUS}, it was found that whilst spacetimes were generically singular for the massless dilaton, for the massive dilaton they were similar to those found in Einstein gravity. Here, the presence of the dilaton seems less destructive, presumably because we have constrained the texture field to remain in the vacuum manifold, and we have shown that non-singular solutions do exist for the massless dilaton, at least for some parameter values. Hence, it might be expected that although the texture and metric fields for the massive dilaton will show some complicated non-self-similar behaviour, they should be well approximated by the Einstein theory solutions.
\section{Discussion}
We have studied the metric and dilaton fields of a global texture in low-energy string gravity for an arbitrary coupling of the texture Lagrangian to the dilaton : $e^{2a\phi}{\mathcal{L}}$. For the massless dilaton, we were able to reduce the partial differential equations describing the metric, texture and dilaton fields to ordinary differential equations in the self-similar variable $x=-r/t$. We found that in contrast to the behaviour of other global topological defects in dilaton gravity, non-singular spacetimes do exist, at least for certain parameter values. However, we have shown that for any value of $\epsilon$, which measures the gravitational strength of the texture, there exists a critical value of $|a+1|$ beyond which non-singular solutions do not exist. This critical value is approximately given by
\begin{equation}
|a+1| \approx \frac{2}{\pi} \sqrt{\frac{1}{\epsilon}}
\end{equation}
For values of $|a+1|$ for which non-singular solutions do exist, the behaviour of the metric
and texture fields is broadly similar to Einstein gravity, but the far-field spacetime shows
an increased solid deficit angle. We have noted that this `non-singular'
behaviour seems to be confined to the Einstein frame. In the string frame, the null hypersurface $x=1$ is a scalar polynomial curvature singularity and null and timelike geodesics reach this hypersurface at finite affine parameter. Thus the texture spacetime is singular in the string frame if $a\ne -1$.
For $a=-1$, the boundary conditions are satisfied by a trivial dilaton
$\phi=0$, and the field equations reduce to those found in Einstein
gravity. This is in contrast to other global topological defects \cite{GSTRDUS,GMDUS} where $\phi$ remains non-trivial when $a=-1$, and is due to
the fact that in the sigma-model approximation we have constrained the texture field to remain in the
vacuum manifold.
For the massive dilaton, the partial differential equations governing the fields cannot be simplified as they were in the massless case. However, using the appropriate Green's function, we were able to write down an integral for the dilaton in the linearised approximation that satisfies the appropriate boundary conditions. For $r \ll t$ and $mr \ll 1$ the dilaton was well approximated by the massless solution (\ref{LinDilDil}). For $r \gg t$, we found the asymptotic behaviour $\phi \approx 2 \epsilon (a+1) / m^2 r^2$. By contrasting the results obtained for the texture in massless dilatonic gravity with those for other global topological defects, we are led to suspect that the presence of a massive dilaton will do little to alter the behaviour of the texture from that found in Einstein gravity. Astrophysical bounds \cite{MICRO} on global textures obtained from their metric field will hence be unchanged by the dilaton; the conclusion that global texture models of structure formation are not favored by current observations of CMB anisotropies (see for example \cite{PENSEL}) still holds. Moreover, since the dilaton fall-off is a power law in $mr$, we expect that the Damour-Vilenkin bound \cite{DAMVIL} will hold, and global textures will be inconsistent with a low (TeV) mass dilaton.
In conclusion, we have found that in low-energy string gravity, textures show qualitatively different behaviour than in Einstein gravity, but perhaps to a lesser extent than was found for other global topological defects.
\section*{Acknowledgements}
O.D. is supported by a PPARC studentship. I would like to thank Ruth Gregory for many helpful discussions; I would also like to thank the referee for their insightful comments.
|
\section{Introduction} \label{sec:introduction}
For at least the next 6 years the Fermilab Tevatron will remain the
highest energy collider in the world. The Tevatron upgrade will
provide an exciting opportunity for discovering physics beyond the
Standard Model. The hadronic environment at the Tevatron presents a
number of challenges and extracting new physics signals can be
difficult. In this respect, signatures with low Standard Model
(especially QCD) backgrounds are extremely valuable, as they may
provide our best opportunity for finding new physics before the LHC
turns on.
Supersymmetry (SUSY) \cite{original} has been fascinating particle
physicists for more than 25 years. It seems an intrinsic component of
theories unifying gravity and gauge interactions such as string
theory, M-theory or supergravity, and has played an important role in
the `second string revolution' of the last few years. The minimal
supersymmetric extension of the Standard Model (MSSM) is a
well-defined, renormalizable and calculable model, which offers a
technical solution to the hierarchy problem \cite{hierarchy}, if the
masses of the superpartners of the standard model particles are of
order the weak scale. Our belief that supersymmetry might be relevant
at energy scales accessible at present colliders is reinforced by the
successful gauge coupling unification \cite{gauge unification}. Also,
due to decoupling the MSSM is generally in agreement with precision
data \cite{precision fits}. In addition, a generic prediction of the
MSSM is the existence of a light Higgs boson \cite{light
Higgs, Higgs constraints}, which is preferred by fits to data
\cite{Higgs mass fit}. In summary, the MSSM is a well-motivated
extension of the Standard Model, which has a very rich and interesting
phenomenology \cite{MSSM review}.
Because of the relatively low (compared to the LHC or NLC)
center of mass energy and integrated luminosity in Run II,
the Tevatron is able to explore only the low end of the
superpartner spectrum. Searches for colored superpartners
(squarks and gluinos) are done in jetty channels, which suffer
from relatively large backgrounds. On the other hand, $SU(2)$-gaugino
pair production leads to a unique clean trilepton signature
\cite{3L,Mrenna}, which has been considered a `gold-plated'
mode for SUSY discovery at the Tevatron. Both CDF and D0 have
already performed Run I trilepton analyses \cite{3L-exp}.
In light of the importance of this channel in Run II, it is clamant to
\begin{itemize}
\item have a reliable estimate of both signal and background rates.
We would prefer to determine background rates from data, but until Run
II we primarily rely on Monte Carlo simulations. ISAJET and PYTHIA
have been two of the most commonly used event generators in SUSY
analyses. While there is a reasonable agreement for the signal,
PYTHIA-based studies \cite{Mrenna,LM} have obtained larger values for
the trilepton backgrounds (mainly $WZ$ and $ZZ$) than ISAJET-based
analyses \cite{3L,BPT,BK}. This discrepancy was noticed and discussed
in the TeV 2000 Report \cite{TeV2000}, but was attributed to the
different lepton rapidity cuts used in the various analyses.
\item use an optimized set of cuts, which will maximize the Tevatron
reach. A first step in this direction was taken in Refs.~\cite{BK},
where softer lepton $p_T$ cuts have been proposed, thus enhancing
signal over background throughout a large part of parameter space.
\item include next-to-leading order (NLO) corrections to the
production cross sections. The corrections to diboson production
\cite{VV_corr}, $t\bar{t}$ production \cite{tt_corr} and Drell-Yan
\cite{DY_corr}, have been known for some time, and the corrections
will soon be available for chargino-neutralino production as well
\cite{Chi_corr}.
\item identify regions of parameter space where the reach via the
trilepton signature is diminished and try to find an alternative
search strategy in those regions. An example of this sort is the large
$\tan\beta$ region with light sleptons, where one often finds that
both the chargino and the neutralino decay predominantly to tau
leptons. Then the trilepton signal has a very small branching ratio
{\em and} the leptons are quite soft, which can make it unobservable
at Run II, even for chargino masses as low as 100 GeV. In this case it
is possible to recover sensitivity by considering alternative
signatures with tau jets \cite{LM}. Another alternative to the
trilepton signature is the inclusive like-sign dilepton channel
\cite{JN}, where the signal acceptance is increased by not requiring
the odd-sign lepton in the event.
\end{itemize}
In this paper, we shall try to address most of these issues. We
perform detailed Monte Carlo simulations of signal and background
using both PYTHIA and ISAJET and explain the cause for the largest
background discrepancy. We also determine the maximum reach by
applying an optimal set of cuts at each point in the supersymmetric
parameter space.
Also, we make use of the presence of a sharp edge in the dilepton
invariant mass distribution of the signal by applying a more
restrictive invariant mass cut, thus reducing the $WZ$ and $ZZ$
backgrounds. As the NLO corrections to gaugino production are not yet
available, we conservatively use leading order cross sections for all
processes. Preliminary results \cite{Chi_corr} show that the
$k$-factor is roughly the same for both signal and background. Hence,
we expect that the Tevatron reach will be improved once NLO
corrections are incorporated.
We show our results for the discovery potential of the upgraded
Tevatron in the so-called minimal supergravity model (mSUGRA)
\cite{SUGRA}. This model has universal soft parameter boundary
conditions at the grand unification scale, and its spectrum displays
characteristic properties. For example, the imposition of electroweak
symmetry breaking results in\footnote{$\mu$ is the Higgsino mass
parameter and $M_2$ is the soft supersymmetry breaking SU(2) gaugino
mass.} $|\mu|>M_2$, so that the lightest chargino and lightest two
neutralinos are gaugino-like. Also, the squark and slepton masses are
generation independent, except at large $\tan\beta$ where the third
generation masses can be lighter. This model has five input
parameters: the scalar mass $M_0$, the gaugino mass $M_{1/2}$, the
$A$-term $A_0$, the ratio of vacuum expectation values $\tan\beta$,
and the sign of the $\mu$ term. We show results for $\mu>0$ and
$A_0=0$.
We adopt a signature driven approach by comparing and contrasting
three of the cleanest channels for Run II -- the trilepton (3L)
\cite{3L, Mrenna}, like-sign dilepton (2L) \cite{JN} and dilepton plus
tau jet (2L1T) \cite{LM} channels. The 3L channel is the long studied
``gold-plated'' channel. The 2L channel has larger signal acceptance
compared to 3L, but it is not {\it a priori} clear whether this
advantage will be spoiled by the concomitant increase in the
background. The 2L1T channel is known to be important at large
$\tan\beta$, where the right-handed tau-slepton is lighter than the
first two generation sleptons. Here we will discuss this channel at
small $\tan\beta$ as well.
We describe in detail our numerical analysis in
Section~\ref{sec:analysis}, where we also describe the cuts we
consider for each signature. We discuss all non-negligible
backgrounds and their evaluation in Section~\ref{sec:backgrounds}.
Then, in Section~\ref{sec:SUGRA}, we map our results for the Tevatron
reach onto the parameter space of the mSUGRA model.
We reserve Section~\ref{sec:conclusions} for our conclusions.
\section{Analysis} \label{sec:analysis}
In this section we describe our numerical analysis. We use
PYTHIA v. 6.115 \cite{PYTHIA} and ISAJET v. 7.42 \cite{ISAJET} for
event generation, and the SHW v. 2.2 detector simulation package
\cite{SHW}, which mimics an average of the CDF and D0 Run II detector
performance. We use PYTHIA for the background determinations, together
with TAUOLA \cite{tauola} to account for the correct (on average) tau
polarization in tau decays. We have made several modifications in
SHW, which are appropriate for our purposes:
\begin{enumerate}
\item We extend the tracking coverage to $|\eta|<2.0$, which increases
the electron and muon acceptance, as is expected in Run II \cite{TDR}.
For muons with $1.5<|\eta|<2.0$, we apply the same fiducial efficiency
as for $1.0<|\eta|<1.5$. However,
we still require that tau jets are reconstructed only up to $|\eta|<1.5$.
\item We retain the existing electron isolation requirement and add a
muon isolation requirement $I<2$ GeV, where $I$ is the total
transverse energy contained in a cone of size $\Delta
R=\sqrt{\Delta\phi^2+\Delta\eta^2}=0.4$ around the muon.
\item We increase the jet cluster $E_T$ cut to 15 GeV and correct the
jet energy for muons. We also add a simple electron/photon rejection cut
$E_{em}/E_{had}<10$ to the jet reconstruction algorithm, where
$E_{em}$ ($E_{had}$) is the cluster energy from the
electromagnetic (hadronic) calorimeter.
\item We correct the calorimeter $\rlap{\,/}E_T$ for muons.
\item We account for an incorrect assignment of neutralino particle
id's in the ISAJET translation of STDHEP v. 4.05 \cite{STDHEP}
\footnote{The assignment has been corrected in STDHEP v. 4.06.}.
\end{enumerate}
The addition of the muon isolation cut and the jet $E_{em}/E_{had}$
cut allows us to uniquely resolve the ambiguity arising in SHW v. 2.2,
when a lepton and a jet are very close.
As we mentioned in the introduction, we show results for three of the
cleanest SUSY channels in Run II at the Tevatron -- trileptons,
inclusive like-sign dileptons and dileptons plus a tau jet. In our
analysis we consider both channel specific and channel independent
cuts. In most of those cases, we use several alternative values for
the cut on a particular variable. For example, we try several $\rlap{\,/}E_T$
cuts, several sets of $p_T$ cuts, etc. We employ a parameter space
dependent cut optimization: at each point in SUSY parameter space, we
consider all possible combinations of cuts, and determine the best
combination by maximizing $S/\sqrt{B}$. In contrast to superior
neural network analyses, the additional CPU requirements when
employing this simple optimization are negligible. We concede that it
may not be possible to perform an identical analysis with real data,
particularly due to trigger issues. Even so, it is useful and
interesting to see which cuts work best in the different parts of
parameter space, and to see how much one can gain by choosing optimal
cuts.
We first list the channel-independent cuts, which in general are designed
to suppress backgrounds common to all three channels.
\begin{enumerate}
\item Four $\rlap{\,/}E_T$ cuts: $\rlap{\,/}E_T>\{15,20,25\}$ GeV or no cut.
\item Six high-end invariant mass cuts for any pair of opposite sign,
same flavor leptons. The event is discarded if: $|M_Z -
m_{\ell^+\ell^-}|<\{10,15\}$ GeV; or $m_{\ell^+\ell^-}>\{50,60,70,80\}$ GeV.
\item Four azimuthal angle cuts on opposite sign, same flavor leptons:
two cuts on the difference of the azimuthal angle of the two highest
$p_T$ leptons, $|\Delta\varphi|<\{2.5,2.97\}$, one cut
$|\Delta\varphi|<2.5$ for {\em any pair} leptons, and no cut.
\item An optional jet veto (JV) on QCD jets in the event.
\end{enumerate}
We list the channel-specific $p_T$ cuts in Table~\ref{pt cuts}. In
the 3L channel, the first four $p_T$ cuts in the table also require a
central lepton with $p_T>11$ GeV and $|\eta|<1.0$ or 1.5.
\begin{table}[t!]
\centering
\renewcommand{\arraystretch}{1.5}
\begin{tabular}{||c||c|c|c||}
\hline\hline
Channel & \multicolumn{3}{c||}{$p_T$ cuts} \\
\hline\hline
3L & $p_T(\ell_1)$ & $p_T(\ell_2)$ & $p_T(\ell_3)$ \\ \hline
1 & 11 & 5 & 5 \\ \hline
2 & 11 & 7 & 5 \\ \hline
3 & 11 & 7 & 7 \\ \hline
4 & 11 & 11 & 11 \\ \hline
5 & 20 & 15 & 10 \\ \hline
\hline
\end{tabular}
\vspace{5mm}
\begin{tabular}{||c||c|c||}
\hline\hline
2L & $p_T(\ell_1)$ & $p_T(\ell_2)$ \\ \hline
\hline
1 & 11 & 9 \\ \hline
2 & 11 & 11 \\ \hline
3 & 13 & 13 \\ \hline
4 & 15 & 15 \\ \hline
5 & 20 & 20 \\ \hline
\hline
\end{tabular}
\vspace{5mm}
\begin{tabular}{||c||c|c|c||}
\hline
\hline
2L1T & $p_T(\ell_1)$ & $p_T(\ell_2)$ & $p_T(\tau)$\\ \hline
\hline
1 & 8 & 5 & 10 \\ \hline
2 & 8 & 5 & 15 \\ \hline
3 & 11 & 5 & 10 \\ \hline
4 & 11 & 5 & 15 \\ \hline
\hline
\end{tabular}
\parbox{5.5in}{
\caption{Channel-specific sets of $p_T$ cuts. \label{pt cuts}}}
\end{table}
For all channels we impose a low-end invariant mass cut on any pair of
opposite sign, same flavor leptons, $m_{\ell^+\ell^-}>11$ GeV. This
cut is designed to suppress a number of backgrounds, e.g. Drell-Yan,
$b\bar{b}$, $c\bar{c}$, and the contribution from $W\gamma^*$ (see
below). It is common for lepton analyses to include a cut on the
$(\Delta\eta,\Delta\varphi)$ distance between any two leptons $\Delta
R>0.4$, which suppresses background from $b\bar{b}$, $c\bar{c}$, and
anomalously reconstructed cosmics. We choose not to include this cut,
since we do not simulate those backgrounds. (Monte Carlo simulations
do not reliably estimate the backgrounds from $b\bar{b}$ and
$c\bar{c}$ production.) We have checked, however, that the effect of
the $\Delta R$ cut on signal and background is negligible.
In the next section we briefly discuss the main backgrounds for the
three channels. This will also motivate the choice of some of the cuts
above.
\section{Backgrounds}\label{sec:backgrounds}
We simulate the following background processes (with the generated
number of events in parentheses): $ZZ$ ($10^6$), $WZ$ ($10^6$), $WW$
($10^6$), $t\bar{t}$ ($10^6$), $Z+$jets ($8\cdot10^6$) and $W+$jets
($8\cdot10^6$).
\subsection{Backgrounds to the trilepton channel}
We start with the $WZ$ background, which is known to be the major
source of background for the 3L channel. The total $WZ$ cross section
at Run II will be $\sim 2.6$ pb. Folding in the branching ratios of
$W$ and $Z$ to leptons, we get a 3L WZ background cross section
production of 46 fb. It has a reducible and an irreducible
component. The irreducible component ($\sim 3$ fb) is due to
$Z\rightarrow\tau^+\tau^-\rightarrow \ell^+\ell^-$ decays. The
invariant mass of the resulting lepton pair from the tau decays is
usually far from the $Z$-mass, in a region which is typical of the
signal. Hence there is no obvious cut which can substantially reduce
this part of the $WZ$ background without at the same time reducing
signal-to-noise. On the other hand, the remaining background ($\sim
43$ fb) is reducible, since it arises from $Z\rightarrow \ell^+\ell^-$
decays. In this case the invariant mass $m_{\ell^+\ell^-}$ of the
resulting lepton pair is equal to $p_Z^2 = (p_Z)^\mu (p_Z)_\mu$, where
$(p_Z)_\mu$ is the 4-momentum of the parent boson. Most of the time
the $Z$ is produced nearly on-shell, $p_Z^2 \approx M_Z^2$. Hence, the
invariant mass cut $|m_{\ell^+\ell^-}-M_Z|>10$ GeV is very efficient
in removing this source of background. However, the parent can also be
off-shell, due to either the $Z$-width, or to $WZ$/$W\gamma$
interference. In PYTHIA, where only the first effect is modeled, the
lepton pair invariant mass distribution follows a Breit-Wigner
shape. We find that roughly 10\% of the 3L background events pass the
dilepton invariant mass cut, thus bringing the reducible background
cross section down to about 4.3 fb. This is almost a factor of two
larger than the corresponding irreducible background cross section
(compare to 2.6 fb). Since ISAJET does not incorporate either $Z$-tail
effect, we find essentially no reducible background cross section from
ISAJET after the dilepton invariant mass cut is
applied\footnote{Energy smearing in the detector simulation produces a
very small background which survives the 10 GeV $Z$-mass window cut.}.
This difference between ISAJET and PYTHIA largely accounts for
the discrepancy in the backgrounds found in Refs.~\cite{Mrenna,LM} and
Refs.~\cite{3L,BPT,BK}. In what follows we use PYTHIA for our
background estimate.
We illustrate the above discussion in Fig.~\ref{inv mass}
\begin{figure}[t]
\centerline{\psfig{figure=fig_im.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small The invariant mass distribution of any pair
opposite sign, same flavor leptons for the signal events (with
$M_0=700$ GeV, $M_{1/2}=160$ GeV, $\tan\beta=5$) and the PYTHIA $WZ$
background. We impose a set of cuts from Ref.~\cite{BK}:
$p_T(\ell)>\{11,7,5\}$ GeV, central lepton with $p_T>11$ GeV and
$|\eta|<1.0$, $\rlap{\,/}E_T>25$ GeV and $|m_{\ell^+\ell^-}-M_Z|>10$ GeV. Each
histogram is normalized to its cross section.
\label{inv mass}}}
\end{center}
\end{figure}
where we show the invariant mass distribution of {\em any} pair of
opposite sign, same flavor leptons for both signal and $WZ$
background. The signal point has $M_0=700$ GeV, $M_{1/2}=160$ GeV and
$\tan\beta=5$, which results in $m_{\tilde\chi_1^{\pm}}\simeq
m_{\tilde\chi_2^0}\simeq122$ GeV. The leptons are required to pass the
set of cuts from Ref.~\cite{BK}: $p_T(\ell)>\{11,7,5\}$ GeV, central
lepton with $p_T>11$ GeV and $|\eta|<1.0$, $\rlap{\,/}E_T>25$ GeV and
$|m_{\ell^+\ell^-}-M_Z|>10$ GeV. The histograms are normalized to the
respective cross section.
Even with the modeling of the $Z$-width effect, we caution that the
$WZ$ simulation in PYTHIA is still not realistic, since the
$W\gamma^*$ contribution is neglected. It results in a peak at low
invariant mass, and the resulting distribution is markedly different
from the result shown in Fig.~\ref{inv mass}. We anticipate that it
will be necessary to cut away {\em all} events on the low-end of the
dilepton invariant mass distribution. We therefore always apply the
cut $|m_{\ell^+\ell^-}|>11$ GeV for all three channels. This cut also
helps in eliminating background lepton pairs from Drell-Yan, as well
as $J/\psi$ and $\Upsilon$ decays.
Since recent trilepton analyses of the Tevatron reach \cite{BPT,BK}
use ISAJET for the simulation, they underestimate the trilepton
background. As a result, we find a significantly reduced Tevatron
reach. However, in some cases we can employ an invariant mass cut
which will substantially reduce the $WZ$ background with no
significant loss of signal. The signal distribution in Fig.~\ref{inv
mass} has a sharp kinematic cut-off at around
$m_{\tilde\chi_2^0}-m_{\tilde\chi_1^0}\sim
m_{\tilde\chi^{\pm}_1}/2\sim 60$ GeV, and we can exploit this feature
to increase signal-to-noise. Indeed, by applying the more stringent
cut $m_{\ell^+\ell^-}>m_{\tilde\chi^{\pm}_1}/2 \sim 60$ GeV, we can
eliminate most of the off-shell $Z$ events, at almost no cost to
signal. This is why in addition to standard $Z$-mass window cuts we
consider four dilepton mass cuts which eliminate {\em all} events
above a given invariant mass value. Also notice that only a very small
fraction of signal events have invariant dilepton masses between 0 and
11 GeV, so that the low-end invariant mass cut $m_{\ell^+\ell^-}>11$
GeV is quite efficient in improving $S/\sqrt{B}$. Our discussion of
the $WZ$ background can be similarly applied to the less serious, but
nevertheless non-negligible $ZZ$ background.
The second largest background to the 3L channel is from dilepton
$t\bar{t}$ events, where there happens to be a third isolated lepton
from a $b$-jet. This background is most easily suppressed by a jet
veto.
Finally, the remaining 3L background one should worry about is $Z$-jet
production. In this case part of the background is due to events where
a jet fakes a lepton. Monte Carlo simulations, especially with a
simplified detector simulation like SHW, cannot give a reliable
estimate of this background. In order to estimate the fake rate one
has to understand the details of the detector response as well as the
jet fragmentation. Only with Run II data will one be able to obtain a
good estimate. For our study we follow a procedure which makes use of
Run I data. It was used in Ref.~\cite{JN} to study the $W+{\rm jets}$
background to the 2L channel (see the next subsection).
\subsection{Backgrounds to the like-sign dilepton channel}\label{sec:2L}
We now discuss the backgrounds to the 2L channel. Ref.~\cite{JN}
observed that it can be advantageous to not require the odd-sign
lepton in the 3L events due to the gain in signal acceptance. At the
same time, events with two like-sign leptons are still quite rare at
the Tevatron, so the 2L channel was suggested as a possible
alternative to 3L for SUSY searches in Run I. However, the rates for
the relevant diboson backgrounds were somewhat underestimated, since
ISAJET was used for the simulation. Here we are interested in
determining whether the 2L channel will be useful in the larger
background environment of Run II.
We first do a back-of-the-envelope comparison of the $WZ$ backgrounds
for the 3L and 2L channels. After not requiring the odd-sign lepton,
one is left with the choice of {\em vetoing} that lepton. In the case
of a veto, we find for the relative size of the two backgrounds
\beq
{\sigma_{WZ}({\rm 2L}) \over \sigma_{WZ}({\rm 3L})}\simeq
{2\varepsilon_l (1-\varepsilon_l) +0.35\varepsilon_\tau \left[
0.65+0.35(1-\varepsilon_\tau) \right] \over
2\varepsilon_l^2\varepsilon_Z + 0.35^2\varepsilon^2_\tau} \sim
{1-\varepsilon_l \over\varepsilon_l\varepsilon_Z},
\label{2L/3L veto}
\eeq
where $\varepsilon_l$ is the acceptance for leptons coming directly
from $W$ or $Z$ decays, $\varepsilon_\tau$ is the acceptance for the
(usually softer) leptons coming from leptonic tau decays, and
$\varepsilon_Z$ is the efficiency of the $Z$-window invariant mass
cut: $|m_{\ell^+\ell^-}-M_Z|>10$ GeV (we have neglected its effect on
the {\em irreducible} background component). We find from Monte Carlo
that the typical Run II values for these efficiencies in $WZ$
production are $\varepsilon_l\sim 0.60$, $\varepsilon_\tau\sim 0.46$
and $\varepsilon_Z\sim 0.10$. Plugging into Eq.~(\ref{2L/3L veto}),
we find for the ratio 6.3, which agrees reasonably well with the
result 5.8 from our full Monte Carlo simulation. For the signal point
shown in Fig.~\ref{inv mass} we find $\varepsilon_l=0.62$ and
$\varepsilon_Z=0.99$ and the corresponding ratio is \beq
{\sigma_{signal}({\rm 2L}) \over \sigma_{signal}({\rm 3L})}\simeq
{1-\varepsilon_l \over\varepsilon_l\varepsilon_Z}\sim 0.6. \eeq This
reveals that vetoing the third lepton is definitely not a good idea.
In comparison to the 3L channel, the signal goes down, while the major
background component is increased almost 6 times!
We therefore only consider the {\em inclusive} 2L channel, where we do
not have any requirements on the third (odd-sign) lepton, just as in
Ref.~\cite{JN}. In this case, the signal acceptance is definitely
increased. Unfortunately, the corresponding increase in the background
is even larger than before: \beq {\sigma_{WZ}({\rm 2L}) \over
\sigma_{WZ}({\rm 3L})}\simeq {2\varepsilon_l
(1-\varepsilon_l+\varepsilon_l\varepsilon_Z) +0.35\varepsilon_\tau
\over 2\varepsilon_l^2\varepsilon_Z + 0.35^2\varepsilon^2_\tau} \sim
7.3
\label{2L/3L}
\eeq for the typical values of the efficiencies. We can see
immediately that the 2L channel can compete with the 3L on the basis
of $S/\sqrt{B}$ only if the lepton acceptance for the signal is less
than $1/\sqrt{7.3}\sim 37\%$. However, for typical values of the SUSY
model parameters the lepton acceptance is much higher.
To make matters worse, the 2L channel suffers from a potentially large
new source of background: $W+$jet production where the jet fakes a
lepton. Although the rate for a jet faking a lepton is quite small,
on the order of $10^{-4}$, the large $W+$jet cross section results in
a major background for the 2L channel. As we mentioned earlier, the
best way to estimate this background is from data, since Monte Carlo
simulations are not reliable for fakes. In our analysis we shall
follow the procedure of Ref.~\cite{JN}, where the rate for observing
an isolated track which would otherwise pass the lepton cuts was
measured in the Run I $Z+$jet event sample. This rate was then
multiplied by the probability that, given an isolated track, it would
fake a lepton. This probability was measured in Run I minimum bias
events to be $\sim 1.5\%$, independent of $p_T$ \cite{JN}. In our
study we first simulate with Monte Carlo the $p_T$ distribution of
isolated tracks in $W$ and $Z$ production, which is shown in
Fig.~\ref{pT_iso}. In this figure we plot the isolated track $p_T$
distribution corresponding to the 2L background, i.e. we combine a
real lepton with $p_T>11$ GeV and $|\eta|<2$ with a same sign isolated
track with $|\eta|<2$. Hence, the 2L background cross section is
obtained by multiplying the cross section from Fig.~\ref{pT_iso} by
the probability that an isolated track will fake a lepton. We see that
the background from fakes falls extremely fast with $p_T$, so a larger
$p_T$ requirement will substantially suppress it.
\begin{figure}[t]
\centerline{\psfig{figure=fig_pt_iso.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small $p_T$ distribution of isolated tracks in (a)
$W+$jet production and (b) $Z+$jet production. The isolated tracks
have $I<2$ GeV and are outside the $\Delta R=0.7$ cone around any
jet. The events in the distributions potentially contribute to the 2L
background: they have one real lepton with $p_T>11$ and $|\eta|<2$ and
one same sign isolated track with $|\eta|<2$.
\label{pT_iso}}}
\end{center}
\end{figure}
We normalize the isolated track rate to data. Using the measured 1.5\%
fake rate per isolated track, we find 1.5 fb of cross section when
running the simulation at $\sqrt{s}=1800$ GeV and using the set of
cuts from Ref.~\cite{JN}. This is half the cross section found in
Ref.~\cite{JN}. Hence, to match the data to PYTHIA/SHW we need to
double the isolated track rate obtained from Monte Carlo.
\subsection{Backgrounds to the dilepton plus tau jet channel}
The largest background to the 2L1T channel is Drell-Yan \cite{LM},
where the tau jet is a fake. As it turns out, SHW does quite a good
job in simulating the fake tau rate \cite{fake taus}, so we can safely
rely on the Monte Carlo for this background. We shall not further
discuss the backgrounds to the 2L1T channel; we refer the interested
reader to Refs.~\cite{LM,Workshop report}. We find a marginal increase
in the background rate here, due to the larger $\eta$ coverage used in
this analysis.
\subsection{Summary and discussion}\label{sec:summary}
In Table~\ref{Table_BG} we summarize our results for the
different backgrounds to the three channels.
\begin{table}[t!]
\centering
\renewcommand{\arraystretch}{1.5}
\begin{tabular}{||c||c|c||c|c||c||}
\hline\hline
Process & 3L & \cite{BK}
& 2L & \cite{JN}
& 2L1T \\ \hline
& &
& $E=2$ TeV & $E=1.8$ TeV
& \\
& &
& $|\eta(\ell)|<2$ & $|\eta(\ell)|<1$
& \\ \hline\hline
$ZZ$ & 0.21 $\pm$ 0.01 & 0.04
& 1.836 $\pm$ 0.006 & 0.1 $\pm$ 0.01
& 0.372 $\pm$ 0.003 \\ \hline
$WZ$ & 1.39 $\pm$ 0.01 & 0.40
& 8.79 $\pm$ 0.03 & 1.1 $\pm$ 0.02
& 1.36 $\pm$ 0.01 \\ \hline
$WW$ & 0.009 $\pm$ 0.003& ---
& 0.002 $\pm$ 0.001 & 0 $+$ 0.02
& 0.54 $\pm$ 0.02 \\ \hline
$t\bar{t}$ & 0.33 $\pm$ 0.01 & 0.14
& 0.21 $\pm$ 0.01 & 0 $+$ 0.02
& 1.64 $\pm$ 0.03 \\ \hline
$Z+$jet & 0.13 $\pm$ 0.01 & ---
& 3.58 $\pm$ 0.05 & 1.1 $\pm$ 0.1
& 11.3 $\pm$ 0.6 \\ \hline
$W+$jet & --- & ---
& 10.5 $\pm$ 0.2 & 3.0
& --- \\ \hline\hline
total & 2.07 $\pm$ 0.02 & 0.58
& 24.9 $\pm$ 0.2 & 5.6
& 15.2 $\pm$ 0.6 \\ \hline\hline
\end{tabular}
\parbox{5.5in}{
\caption{ Background cross sections after cuts (in fb) for the three
channels, each with a set of cuts described in the text. All errors
are statistical. We also list the 3L background found in
Ref.~\cite{BK} and the 2L background found in Ref.~\cite{JN}.
\label{Table_BG}}}
\end{table}
The results are presented for a standard choice of cuts in each case.
For the 3L channel, we pick the set of cuts from Ref.~\cite{BK}:
$p_T(\ell)>\{11,7,5\}$ GeV, central lepton with $p_T>11$ GeV and
$|\eta|<1.0$, $\rlap{\,/}E_T>25$ GeV, $|m_{\ell^+\ell^-}-M_Z|>10$ GeV, and no
$|\Delta\varphi|$ or JV cuts. For the 2L channel we use the following
cuts from Ref.~\cite{JN}: $p_T(\ell)>\{11,11\}$ GeV,
$|m_{\ell^+\ell^-}-M_Z|>10$ GeV, and no $\rlap{\,/}E_T,\ |\Delta\varphi|$ or JV
cuts. Here we require both leptons to have $|\eta|<2.0$, whereas
Ref.~\cite{JN} requires $|\eta|\lower.7ex\hbox{$\;\stackrel{\textstyle<}{\sim}\;$} 1$. Finally, for the 2L1T channel
we use the cuts $p_T(\tau)>15$ GeV, $|\eta(\tau)|<1.5$,
$p_T(\ell)>\{8,5\}$ GeV, $|\eta(\ell)|<2.0$,
$|m_{\ell^+\ell^-}-M_Z|>10$ GeV, $\rlap{\,/}E_T>20$ GeV, and no JV, from
Ref.~\cite{LM}. All errors in the table are statistical.
Comparing to Refs.~\cite{BK,JN} we see a significant increase in the
$WZ$ and $ZZ$ backgrounds to both the 3L and 2L channels. In the 2L
case this is in part due to the larger $\eta$ coverage that we use, as
well as the higher center-of-mass energy. The remaining difference is
due to the lack of $Z$-tail effects in the ISAJET simulation. The $WZ$
background is almost as large as $W+$jet, which is not surprising
based on the estimate in Eq.~(\ref{2L/3L}) (compare the $WZ$
backgrounds to 3L versus 2L). We remind the reader that our 2L
$W+$jet background has been normalized to the 2L $W+$jet background
from Ref.~\cite{JN}, so the difference in the $W+$jet background
seen in the table is due solely to the different tracking coverage
and center of mass energy.
In the 3L case, we find significantly larger backgrounds than
Ref.~\cite{BK}. In fact, we surmise that the lepton efficiency is
smaller in SHW than in the ISAJET detector simulation, and if one were
to take this into account the background differences would be even
larger. Part of the difference in the $WZ$ and $ZZ$ backgrounds is due
to the fact that PYTHIA gives a $\sim15\%$ larger diboson cross
section than ISAJET. However, the largest part of the difference in
the diboson background rates is due to the $Z$-width. Notice that with
the fake rate procedure discussed in Sec.~\ref{sec:2L} we are able to
obtain an estimate of the $Z+$jet trilepton background (where the jet
fakes a lepton). This background has not been taken into account in
previous studies.
We do not trust the Monte Carlo simulation to provide a reliable
estimate for the $W+$jet and $Z+$jet backgrounds where the jet gives
rise to a real isolated lepton. We expect these backgrounds to be
small, and we ignore them.
In the next section, we present results for the Tevatron reach in the
three channels in the minimal gravity-mediated SUSY breaking model.
In our simulations we use PYTHIA for the background estimates and
ISAJET for the signal.
\section{Discovery reach for gravity-mediated models} \label{sec:SUGRA}
We next discuss the discovery potential of the upgraded Tevatron in
the mSUGRA model. There are various SUSY production processes which
give rise to the lepton signals under consideration. The dominant
source of signal in most regions of parameter space is
$\tilde\chi_1^+\tilde\chi_2^0$ production. Other chargino/neutralino
and slepton production processes also contribute. Typically these
processes constitute a small fraction of the total signal cross
section, but in some regions of parameter space (e.g. large
$\tan\beta$, small $M_0$) they can dominate. We ignore the possibility
of small contributions from squark and gluino production processes.
We first review which parts of the mSUGRA parameter space are
accessible in Run II via the 3L signature. The
$\tilde\chi_1^+\tilde\chi_2^0$ production cross section depends
primarily on the chargino mass, and scales roughly as
$M_{1/2}^{-5.5}$. Therefore, at the Tevatron we can only explore
regions with small $M_{1/2}$, where the cross section is large. The
$\tilde\chi_1^+\tilde\chi_2^0$ production proceeds predominantly via
$s$-channel $W$-boson exchange. There is also destructive interference
from $t$-channel squark exchange. As a result, the
$\tilde\chi_1^+\tilde\chi_2^0$ cross section is slightly enhanced in
case of heavy squarks (i.e. at large values of $M_0$).
At large values of $M_0$ ($M_0\gsim700$ GeV) the sleptons are heavy
and the gauginos decay to three-body final states dominantly via real
(at large $M_{1/2}$) or virtual (small $M_{1/2}$) $W$- or $Z$-boson
exchange. In this case the reach is determined solely by the signal
production cross section. At smaller values of $M_0$ the off-shell
slepton mediated decays destructively interfere with the gauge
mediated decays. At small values of $M_0$ ($M_0\lower.7ex\hbox{$\;\stackrel{\textstyle<}{\sim}\;$} M_{1/2})$
sleptons become lighter than the $\tilde\chi_1^+/\tilde\chi_2^0$ and
two-body decays to lepton final states open up. This enhances the
leptonic branching ratios of the gauginos and improves the Tevatron
reach. We illustrate these features in Fig.~\ref{br_frac_lowtb} where
we show the branching ratios\footnote{When we use the term `branching
ratios' in relation to a pair of particles, we mean a sum of products
of branching ratios. For example, if the two-body decays of
$\tilde\chi_1^+$ and $\tilde\chi_2^0$ are closed,
BR$(\tilde\chi_1^+\tilde\chi_2^0 \rightarrow 3L) \equiv \left[{\rm BR}
(\tilde \chi_1^+ \rightarrow \tilde \chi_1^0\ell^+\nu_\ell) + {\rm BR}
(\tilde \chi_1^+ \rightarrow \tilde \chi_1^0 \tau^ + \nu_\tau
\rightarrow \tilde\chi_1^0\ell^+ \nu_\ell\bar \nu_\tau \nu_\tau
)\right]\left[{\rm BR} (\tilde \chi_2^0 \rightarrow \tilde
\chi_1^0\ell^+\ell^-) + {\rm BR} (\tilde \chi_2^0 \rightarrow \tilde
\chi_1^0\tau^+\tau^- \rightarrow\right.$$ \tilde \chi_1^0 $
$\ell^+ \ell^- \bar\nu_\ell$ $\nu_\ell\bar\nu_\tau$
$\left.\nu_\tau)\right]$.} of a chargino-neutralino pair into the 3L,
2L and 2L1T channels at an mSUGRA model point with $\tan\beta=5$,
$\mu>0$, $A_0=0$ and (a) $M_{1/2}=175$ GeV or (b) $M_{1/2}=250$ GeV.
We plot versus $m_{\tilde \tau_1}$ (bottom axis) or $M_0$ (top axis).
\begin{figure}[t]
\centerline{\psfig{figure=fig_br_tb5.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small Branching ratios of a chargino-neutralino pair into
the 3L (solid), 2L (dash) and 2L1T (dot dot dash) channels versus the
lightest stau mass (bottom axis), or alternatively, versus $M_0$ (top
axis), with mSUGRA model parameters $\tan\beta=5$, $\mu>0$, $A_0=0$
and (a) $M_{1/2}=175$ GeV or (b) $M_{1/2}=250$ GeV. The arrows
indicate the chargino threshold $m_{\tilde\chi_1^\pm}=m_{\tilde
\tau_1}$. For the range of $M_0$ values shown the chargino mass
varies from 118 to 139 GeV in (a), and from 187 to 201 GeV in (b).
\label{br_frac_lowtb}}}
\end{center}
\end{figure}
Notice that because of the rather small value of $\tan\beta$, all
slepton flavors are practically degenerate.
In Fig.~\ref{br_frac_lowtb}(a) the chargino mass
$m_{\tilde\chi_1^\pm}$ is only about 120 GeV, so the two-body decay to
the $W$-boson is closed. If $M_0>92$ GeV the chargino decays are
three-body and the 2L and 3L channels have a larger branching ratio
than the 2L1T, roughly by a factor of two. The dip in the leptonic
branching ratios near $m_{\tilde \tau_1}\sim 260$ GeV is due to
destructive interference between the $Z$ and $\tilde\ell$-mediated
graphs \cite{BR interference}. In the region $M_0<92$ GeV the two-body
decays to sleptons are open and quickly become dominant. But notice
that while $\tilde\chi_2^0$ can decay to all slepton flavors through
its bino component, $\tilde\chi_1^\pm$ decays dominantly to a stau,
since the chargino couplings to the right-handed sleptons are
proportional to the corresponding lepton masses. Hence, the branching
fraction of the 2L1T channel dominates the region $40\lower.7ex\hbox{$\;\stackrel{\textstyle<}{\sim}\;$} M_0\lsim90$
GeV. For $M_0\lsim40$ GeV the 2L1T branching ratio rapidly drops while
the 2L and 3L branching ratios increase and become dominant again, due
to two-body decays of $\tilde\chi^\pm_1$ to sneutrinos.
In Fig.~\ref{br_frac_lowtb}(b) the mass of $\tilde\chi_1^\pm$ is near
190 GeV and it is always above the $W$-boson threshold. If $M_0$ is
greater than about 400 GeV the $\tilde\chi_2^0$ is above the $Z$-boson
threshold, and in this region the branching ratios for the three
signatures are determined by the $W$- and $Z$-boson branching ratios
to leptons. In the region $M_0<150$ GeV the sleptons are lighter than
the chargino, and the two-body decay modes to sleptons greatly enhance
the leptonic branching fractions. The 2L1T channel does not become as
dominant because the chargino decays to quarks via an on-shell
$W$-boson. The decrease in the leptonic branching ratios below
$M_0\simeq70$ GeV is as before due to two-body decays to sneutrinos.
In Fig.~\ref{br_frac_hightb} we show similar plots of the
branching fractions, but for $\tan\beta=35$.
\begin{figure}[t]
\centerline{\psfig{figure=fig_br_tb35.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small The same as Fig.~\ref{br_frac_lowtb}, but for
$\tan\beta=35$. This time the chargino mass varies from
126 to 140 GeV (in (a)), and from 192 to 201 GeV (in (b)).
\label{br_frac_hightb}}}
\end{center}
\end{figure}
Many of the features seen in Fig.~\ref{br_frac_lowtb} are present here
as well. For example, in Fig.~\ref{br_frac_hightb}(a) we see the broad
region ($150\lower.7ex\hbox{$\;\stackrel{\textstyle<}{\sim}\;$} m_{\tilde \tau_1}\lower.7ex\hbox{$\;\stackrel{\textstyle<}{\sim}\;$} 400$ GeV) where the
destructive $Z$-$\tilde\ell$ interference severely diminishes the
branching ratios. And again we observe the $\tilde\chi_2^0\rightarrow
\tilde\chi_1^0 Z$ threshold in Fig.~\ref{br_frac_hightb}(b), near
$m_{\tilde \tau_1}=260$ GeV. However, in this case the tau slepton is
significantly lighter than the first two generation sleptons. Hence,
the branching ratio to the three tau final state quickly approaches
100\% below the stau threshold, and the 2L1T branching ratio is large.
We also see that the 2L channel is competitive at small values of
$M_0$, and is preferred over 3L on the basis of branching ratio by a
factor of 2.8.
To summarize, we see that depending on the values of the mSUGRA
parameters, any of the three channels offers some promise to be
observed in Run II. In the rest of this section, we shall do a
comparative study of the three channels, accounting for all relevant
background processes and using a realistic detector simulation. We
shall not only update the existing 3L analyses with improved estimates
of the $WZ$, $ZZ$ and Drell-Yan background rates, but foremost we are
interested in evaluating the different prospects each channel can
offer. For example, we would like to see whether the 2L channel offers
reach beyond the 3L channel, or can cover regions where the 3L channel
is suppressed. Also, we want to determine the region of parameter
space where the 2L1T channel can offer an independent check of a
signal in one of the other channels.
The cut optimization procedure is an important ingredient in our
analysis. We show results for the reach with the optimal set of cuts,
determined independently at each point in SUSY parameter space. The
optimal set of cuts maximizes $S/\sqrt{B}$. We require the observation
of at least 5 signal events, and present our results as $3\sigma$
exclusion contours in the $M_0-M_{1/2}$ plane, for two representative
values of $\tan\beta$ -- 5 and 35. We fix $\mu>0$ and $A_0=0$.
In Fig.~\ref{m0mh_3L} we show the Tevatron reach in the 3L channel
\begin{figure}[t]
\centerline{\psfig{figure=fig_m0mh_3l.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small Tevatron reach in the 3L channel for mSUGRA models
with $\mu>0$, $A_0=0$, and (a) $\tan\beta=5$ or (b) $\tan\beta=35$. We
show the reach with both a standard set of soft cuts \cite{BK}
(dashed, for large $M_0$ only), as well as with the optimal set of
cuts (solid) (see text). The reach is shown for 30 ${\rm fb}^{-1}$,
10 ${\rm fb}^{-1}$ and 2 ${\rm fb}^{-1}$ total integrated luminosity
(from top to bottom). The cross-hatched region is excluded by current
limits on the superpartner masses. The dot-dashed lines correspond to
the projected LEP-II reach for the chargino and the lightest Higgs
masses. In Fig.~(a) the left dotted line shows where
$m_{\tilde\nu_\tau}=m_{\tilde\chi_1^\pm}$ and the right dotted line
indicates $m_{\tilde\tau_1}=m_{\tilde\chi_1^\pm}$. In Fig.~(b) the
dotted lines show where $m_{\tilde e_R}=m_{\tilde\chi_1^\pm}$ (left)
and $m_{\tilde\tau_1}=m_{\tilde\chi_1^\pm}$ (right).
\label{m0mh_3L}}}
\end{center}
\end{figure}
with a standard set of soft cuts \cite{BK} (dashed lines, for large
$M_0$ only to prevent crowding), as well as with the optimal set of
cuts (solid). We show the expected reach for 2, 10 and 30 ${\rm
fb}^{-1}$ total integrated luminosity. The cross-hatched region is
excluded by current limits on the superpartner masses and the
dot-dashed lines indicate the projected LEP-II reach for the chargino
and the lightest Higgs mass\footnote{It should be kept in mind that at
small $\tan\beta$ the Higgs mass is a sensitive function of
$\tan\beta$. For example, if the Higgs-boson is not discovered at
LEP-II, $\tan\beta=3$ will be excluded. The reach contours are much
less sensitive to $\tan\beta$.}.
In Fig.~\ref{m0mh_3L}(a) the left (right) dotted line marks the
$\tilde\nu_\tau/\tilde\chi_1^\pm$ ($\tilde\tau_1/\tilde\chi_1^\pm$)
mass threshold, while in Fig.~\ref{m0mh_3L}(b) the left (right) dotted
line marks the $\tilde e_R/\tilde\chi_1^\pm$
($\tilde\tau_1/\tilde\chi_1^\pm$) mass threshold.
Comparing to the results of Refs.~\cite{BPT,BK}, we see the Tevatron
reach indicated in Fig.~\ref{m0mh_3L} is greatly reduced, due to the
larger backgrounds found in our analysis. Even with optimized cuts our
results show a much smaller observable region. For example, using the
set of cuts of Ref.~\cite{BK} at $\tan\beta=35$ and $M_0=1$ TeV, our
results indicate that the region bounded by the 2 fb$^{-1}$ 3-$\sigma$
contour extends to $M_{1/2}=123$ GeV (or 136 GeV with optimal cuts),
whereas in Ref.~\cite{BK} the corresponding 5-$\sigma$ region extends
to 180 GeV. Similarly, their 30 fb$^{-1}$ 3-$\sigma$ contour extends
to $M_{1/2}=250$ GeV, while ours extends to $M_{1/2}=186$ GeV (198 GeV
with optimal cuts).
As expected, there is respectable reach beyond LEP-II at small values
of $M_0$ and $\tan\beta$, where the chargino and neutralino 2-body
decays to sleptons are open. As expected from
Fig.~\ref{br_frac_lowtb}(a), the Tevatron has no sensitivity beyond
LEP-II in the region $200\lower.7ex\hbox{$\;\stackrel{\textstyle<}{\sim}\;$} M_0\lsim400$ GeV. For large values of
$M_0$, where the leptonic decays of the gauginos are three-body, we
find some sensitivity for both values of $\tan\beta$. In this region
there is a clear benefit to using optimized cuts. At Run II (2
fb$^{-1}$), with the default set of cuts, only the region with small
$\tan\beta$, small $M_0$, and small $M_{1/2}$ can be explored beyond
LEP-II. With optimal cuts, we see at large $\tan\beta$ a
non-negligible region can be excluded beyond LEP-II. Looking beyond
Run II, we notice that at TeV33 (30 fb$^{-1}$) optimization can prove
equivalent to doubling and sometimes even tripling the total
integrated luminosity! In Fig.~\ref{advantage} we
\begin{figure}[t]
\centerline{\psfig{figure=fig_del_lum.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small The difference $\Delta L$ (in ${\rm fb}^{-1}$)
between the required total integrated luminosity for the optimal set
of cuts and the default set of cuts, for the 3L signal. The three
(solid) contours correspond to (from top to bottom) $\Delta L = 30,
10$ and 2 ${\rm fb}^{-1}$. The dotted lines indicate the optimal-cut
30 ${\rm fb}^{-1}$ reach contours from Fig.~\ref{m0mh_3L}.
\label{advantage}}}
\end{center}
\end{figure}
show the difference in the required luminosity when using the fixed
set of cuts from \cite{BK} and the optimized cuts. As a guideline, we
also show the 30 ${\rm fb}^{-1}$ optimal-cut contours (dotted lines) from
Fig.~\ref{m0mh_3L}.
It is instructive to examine the optimized sets of cuts. In
Fig.~\ref{best cuts 1} (Fig.~\ref{best cuts 2}) we show the optimal
sets of cuts for the 3L channel in the $M_0$, $M_{1/2}$ plane, for
$\tan\beta=5$ ($\tan\beta=35$).
\begin{figure}[t]
\centerline{\psfig{figure=fig_bc_3t5.ps,height=4.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small The optimal sets of 3L cuts in the $M_0,\ M_{1/2}$
plane, for $\tan\beta=5$. The key indicates which symbols correspond
to which cuts (central lepton rapidity, $\rlap{\,/}E_T$, invariant mass,
$\Delta\varphi$, and $P_T$ cuts) (see text). The dotted line indicates
the optimal-cut 30 ${\rm fb}^{-1}$ reach contours from
Fig.~\ref{m0mh_3L}.
\label{best cuts 1}}}
\end{center}
\end{figure}
\begin{figure}[t]
\centerline{\psfig{figure=fig_bc_3t35.ps,height=4.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small Same as Fig.~\ref{best cuts 1}, with $\tan\beta=35$.
\label{best cuts 2}}}
\end{center}
\end{figure}
We use the following notation to describe the set of cuts at each
point. The central symbol is the number of the $p_T$ cut according to
Table~\ref{pt cuts}. The left superscript indicates which central
lepton $\eta$ cut was chosen, either $|\eta|<1.0$ (labeled ``10'') or
$|\eta|<1.5$ (``15''). A left subscript denotes the type of
$|\Delta\varphi|$ cut: for the two highest $p_T$ opposite sign, same
flavor leptons the $|\Delta\varphi|<2.5$ cut is indicated by ``2'',
the $|\Delta\varphi|<3.0$ cut by ``3'', and no symbol indicates no
$|\Delta\varphi|$ cut. The cut $|\Delta\varphi|<2.5$ for {\em any}
pair of opposite sign, same flavor leptons is indicated by ``a''. The
right superscript shows the $\rlap{\,/}E_T$ cut: $\rlap{\,/}E_T>\{15,20,25\}$ GeV
(``15'',``20'',``25''), or no cut (no symbol). A right subscript
denotes the dilepton invariant mass cut:
$|m_{\ell^+\ell^-}-M_Z|>\{10,15\}$ GeV (``10'',``15'') or
$|m_{\ell^+\ell^-}|<\{50,60,70,80\}$ GeV
(``50'',``60'',``70'',``80''). And finally, a tilde over the central
symbol indicates that the luminosity limit came from requiring 5
signal events rather than $3\sigma$ exclusion.
The jet veto cut is not indicated on the figures. However, except for
one point at large $\tan\beta$ the jet veto was never selected as an
optimal cut. Indeed, the major 3L background events (from WZ) are just
as likely to contain extra jets as the signal.
There are several lessons to be drawn from Figs.~\ref{best cuts 1} and
\ref{best cuts 2}. As expected, in those cases where the signal is
strong and dominates over the background (typically for small
$M_{1/2}$ or small $M_0$), softer cuts are beneficial. Indeed, the
majority of the points which can be discovered with 5 signal events,
have selected $p_T$ cut ``1'', which is the most lenient set of $p_T$
cuts. For larger values of $M_{1/2}$, where the background is more
important, harder cuts on the lepton $p_T$'s are preferred. In fact,
we see that in the region of interest for TeV 33 at large $M_0$, the
hardest $p_T$ cuts we consider (from Ref.~\cite{BPT}, indicated by
``5''), often work best.
Figs.~\ref{best cuts 1} and \ref{best cuts 2} clearly indicate the
advantage of the invariant mass cuts introduced in
Section~\ref{sec:analysis}. Of all the points on the plots where the
background is an issue, there are only a few at which a conventional
$Z$-mass window cut is optimal. In all other cases it is advantageous
to cut all events with invariant dilepton masses above a certain
threshold. Notice also how the value of the threshold tends to
increase with increasing $M_{1/2}$. This is expected because the sharp
edge in the signal distribution is roughly at $0.4\cdot M_{1/2}$
(see Fig.~\ref{inv mass}).
We also see from the figures that a missing $E_T$ cut is preferred
essentially everywhere in parameter space. However, a soft $\rlap{\,/}E_T$ cut
($\rlap{\,/}E_T>15$ GeV) often gives the better reach. The latest trilepton
analyses \cite{BPT,BK} have chosen to {\em increase} the $\rlap{\,/}E_T$ cut
from its nominal Run I value of 20 GeV to 25 GeV. Our results suggest
that in the off-line analysis one is better off with a lower $\rlap{\,/}E_T$
cut. However, more work is needed to conclusively determine what the
best $\rlap{\,/}E_T$ cut will be in the actual analysis. For example, triggering
and energy mismeasurement issues will have to be carefully taken into
account.
Lastly, in the great majority of parameter space it is better not to
require a $\varphi$ cut.
Looking back at Fig.~\ref{m0mh_3L}(a), we observe that near the
slepton-chargino mass thresholds the reach becomes diluted. This
effect can easily be overlooked, since it only shows up very close to
threshold. We find that it is entirely due to the suppressed signal
acceptance. Indeed, although the branching ratio is increased
immediately below threshold, the lepton resulting from the
$\tilde\chi_2^0\rightarrow \tilde \ell^\pm \ell^\mp$ decay tends to be
very soft and it can fail the analysis cuts.
We next discuss the prospects for the 2L channel. In
Fig.~\ref{m0mh_2L} we show the 2L channel reach for the Tevatron.
\begin{figure}[t]
\centerline{\psfig{figure=fig_m0mh_2l.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small The same as Fig.~\ref{m0mh_3L}, for the 2L channel.
The dashed lines correspond to a set of cuts from Ref.~\cite{JN}.
The 2, 10 and 30 fb$^{-1}$ reach contours are plotted.
\label{m0mh_2L}}}
\end{center}
\end{figure}
Due to the much larger background the reach in the 2L channel is not
as great as in the 3L channel.
There is one important exception, however -- the 2L channel does not
lose sensitivity near the chargino-charged slepton threshold. Indeed,
because of the Majorana nature of the neutralinos, in 50\% of the
events the lost soft lepton in the $\tilde\chi_2^0\rightarrow \tilde
\ell^\pm \ell^\mp$ decay is of the {\em opposite} sign as the charged
lepton from the chargino decay. The remaining two hard leptons are
then of the same sign and they are readily reconstructed. Therefore,
near the $\tilde\ell/\tilde\chi_1^\pm$ mass threshold, the 2L
channel may prove to be a valuable alternative to 3L.
Fig.~\ref{m0mh_3L}(b) reveals that at large $\tan\beta$ one starts to
lose 3L sensitivity in the small $M_0$ region, since the decays to tau
final states dominate. In fact, for $M_0\lower.7ex\hbox{$\;\stackrel{\textstyle<}{\sim}\;$} 300$ GeV and
$\tan\beta=35$, we find no 3L reach in Run II beyond LEP-II. Only with
multiple years of running and collecting soft lepton events will the
Tevatron be able to start improving on the LEP mSUGRA bounds. One can
improve this situation by considering alternative signatures with tau
jets \cite{LM}. Of those, the 2L1T channel is singled out on the basis
of both sensitivity and statistical importance.
In Fig.~\ref{m0mh_2L1T} we show the Tevatron reach in the 2L1T channel.
\begin{figure}[t]
\centerline{\psfig{figure=fig_m0mh_2l1t.ps,height=3.5in}}
\begin{center}
\parbox{5.5in}{
\caption[] {\small The same as Fig.~\ref{m0mh_3L}, for the 2L1T
channel. In this case the optimal cuts yield very little improvement
relative to a default set of cuts from Ref.~\cite{LM} (listed in
Sec.~\ref{sec:summary}), so the contours corresponding to the default
cuts are not shown.
\label{m0mh_2L1T}}}
\end{center}
\end{figure}
In this case we compare the optimal reach to a set of cuts from
Ref.~\cite{LM} (listed in Sec.~\ref{sec:summary}). We find very little
difference between this fixed set of cuts and the optimal set, so we
do not show the fixed cut lines in the figure.
The 2L1T channel has no reach in the large $M_0$ region. However, when
the two-body decays to staus open up the 2L1T branching fraction is
larger than the other two channels, leading to some sensitivity. While
the region accessible in the 2L1T channel at small $\tan\beta$ is not
competitive with the 3L and 2L reach, it can improve the statistical
significance in case of exclusion, or it can serve as an important
confirmation and provide unique information about the model
parameters in case of discovery. This channel offers the greatest
reach at large $\tan\beta$ (together with the related much cleaner
signature of two like-sign leptons plus a tau jet \cite{LM}) in the
small $M_0$ region.
\section{Conclusions} \label{sec:conclusions}
In this paper we studied three of the cleanest and most promising
channels for SUSY discovery at the Tevatron in Run II. We revisited
the trilepton and like-sign dilepton analyses, improving them in
several key aspects. For example, we used a more realistic simulation
of the major backgrounds. We found larger backgrounds than previous
analyses. We used a procedure relying on Run I data to estimate
backgrounds involving fake leptons. And we introduced an invariant
mass cut which took advantage of a sharp edge in the signal dilepton
invariant mass distribution. This cut was generally more effective in
increasing signal-to-noise than the standard invariant mass cuts.
Also, we varied the cuts at each point in the supersymmetric parameter
space, and determined at each point the set of cuts which yields the
largest reach. We found that this cut optimization can significantly
enhance the Tevatron reach. Lastly, we analyzed the reach of the 2L1T
channel.
If nature is supersymmetric at low energies, and the superpartners (in
particular the gauginos) are light, there is a good chance that the
Tevatron will discover them in its upcoming runs. However, the 3L
signature has limited reach, and we can improve the reach by
optimizing cuts and considering alternative clean signatures.
\section*{Acknowledgments}
We thank J.~Campbell, R.K.~Ellis, J.~Lykken, J.~Nachtman and F.~Paige
for useful discussions. K.T.M. (D.M.P.) is supported by Department of
Energy contract DE-AC02-76CH03000 (DE-AC02-98CH10886).
{\em Note added:} We thank H.~Baer for providing us with a preliminary
version of a preprint by H.~Baer, M.~Drees, F.~Paige, P.~Quintana and
X.~Tata. This paper is an update of their previous study \cite{BPT} of
the reach of the Tevatron in the 3L channel. Part of the improvement
is the inclusion of the $Z$-width in the $WZ$ background determination.
\newpage
|
\section{Introduction}\label{sec:intro}
As well known, the larger colour charge of gluons ($C_A=\Nc=3$)
compared to quarks ($\CF=(\Nc^2-1)/2\Nc=4/3$) leads to various
distinctive differences between the two types of jets, for recent articles see e.g.~\cite{jetscales} and the review~\cite{khozeochs}. Thus, a detailed comparison
of the properties of quark and gluon jets provides one of the most
instructive tests of the basic ideas of QCD. An experimental
verification of these differences has been a subject of quite
intensive investigations, especially in the last years,
e.g.~\cite{hamacher}. However, obtaining of the
theoretically adequate information about the properties of the gluon
jet appears to be not an easy task. Recall that the analytical QCD
results address the comparison between the energetic gluon and quark
jets emerging from the point-like colourless sources, and that (unlike
the \pair q case) the pure high energy gg events at present are not
available experimentally.\footnote{In principle, it is possible to
create a pure source of the colour singlet gg events at a future
linear \ee collider through the process $\gamma\gamma\rightarrow
\mrm{gg}$~\protect{\cite{khozegg}}.}
So far, most studies of the structure of gluon jets have been
performed in three-jet events of \ee annihilation. As a rule, these
rely on a jet finding procedure both for selection of the \pair qg
events and for a separation between the jets in an event. Without
special care, such an analysis is inherently ambiguous and may suffer
from the lack of the direct correspondence to the underlying
theory. Recently some more sophisticated approaches have been
exploited (see e.g.~[3, 5-8]) which allow better theoretical
significance. There are still a number of issues which are frequently
overlooked in the present gluon jet analyses and some further
theoretical efforts are required. First of all, this concerns particle
multiplicity distributions in the jets. Clarification of these issues
is the main aim of this paper. More detailed description of the
theoretical framework can be found in ref.~\cite{jetscales}.
In particular the following problems are addressed.
1. Different approaches to the three-jet studies employ different
definitions of the \pair qg kinematics. In particular, this concerns
such a key variable as a transverse momentum scale of the gluon,
$p_\perp$. Our first issue here is to discuss an exact definition of
this quantity, which governs radiation from the gluon.
2. The definition of the three-jet topology with the gluon registered
at a given $p_\perp$ imposes an obvious requirement that there are no
other subjets in the event with the transverse momentum exceeding
$p_\perp$. We have to investigate quantitatively the impact of this
requirement on the jet sample.
3. To calculate predictions from perturbative QCD, using the
assumption of local parton hadron duality (LPHD)~\cite{LPHD}, a cutoff
is needed for the infrared singularities. As discussed in detail in ref.~\cite{jetscales} such a cutoff depends on the
soft hadronization process and can {\em not} be uniquely specified
from perturbative QCD alone. Thus, the result is necessarily model
dependent.
In what follows we discuss these three issues successively in sections
2, 3 and 4, and in section 5 we study their effect on analyses of
3-jet events in $e^+e^-$-annihilation.
\section{Definition of $p_\perp$}\label{sec:ptdef}
In the simplest case of soft radiation, $p_\perp$ can be easily
defined, as the quark and antiquark specify a unique direction. For
large $p_\perp$ gluons, however, the q and \anti q get recoils such
that there is no obvious direction against which the transverse
momentum should be measured. To have well defined expressions such a
direction has to be specified. In the Lund dipole formalism~[10-16]
$p_\perp$ has been defined according to (subscript $\Lund$ for Lund)
\eqbe
p_{\perp \Lund}^2 \equiv \frac{s\srm{qg}s\srm{g\ol q}}{s}, \label{e:ptLdef}
\eqen
where $s\srm{qg}$ denotes the squared
mass of the quark-gluon system etc. In this particular frame the
gluon rapidity is given by the expression
\eqbe
y=\frac{1}{2}\ln(\frac{s\srm{qg}}{s\srm{g\ol q}}).\label{e:yLdef}
\eqen
The kinematically allowed region is given by
\eqbe
p_{\perp\Lund} < \frac{\sqrt s}2;~~|y| <\ln\left(\frac{\sqrt s}{p_{\perp\Lund}}\right) \equiv \frac{1}{2} (L -{\kappa_\Lund});~~~L\equiv\ln(\frac{s}{\Lambda^2}),~\kappa_\Lund \equiv \ln(\frac{p_{\perp\Lund}^2}{\Lambda^2}). \label{e:yptranges}
\eqen
These variables have
the advantage that the phase space element usually expressed in the
scaled energy variables $ x\srm q$ and $ x\srm{\ol q}$ is exactly
given by the simple relation
\eqbe s \mrm dx\srm q\mrm dx\srm{\ol q} = \mrm dp_{\perp \Lund}^2\mrm dy.\label{e:PSelement}
\eqen
As discussed in section~\ref{sec:qqg}, $p_{\perp\Lund}$ may also work
well as a scale parameter in the QCD cascade.
An alternative definition has also been used in the literature, e.g.\ by the Leningrad group~\cite{ADKT,khoze}
\eqbe p_{\perp \SP}^2 \equiv \frac{s\srm{qg}s\srm{g\ol q}}{s\srm{q\ol q}}. \label{e:ptSPdef} \eqen
This definition corresponds to the gluon transverse momentum in the $\mrm{q\ol q}$ cms (with respect to the $\mrm{q\ol q}$ direction). It is notable that in this frame the gluon rapidity is also exactly given by the expression in Eq~(\ref{e:yLdef}). The two $p_\perp$-definitions agree for soft gluons, but deviate for harder gluons. While $p_{\perp \Lund}$ is always bounded by $\sqrt s/2$, $p_{\perp \SP}$ has no kinematic upper limit in the massless case.
\section{Bias from restrictions on subjet transverse momenta}\label{sec:bias}
The effect of a cutoff in $p_\perp$ has been discussed previously~\cite{Lundbias,Catanibias}. Here we give a brief review of the results, in order to end the section with an investigation of the numerical importance of subleading terms. These are essential for a correct analysis of three-jet events, which will be discussed in section~\ref{sec:qqg}.
To see the qualitative features of the bias we first study \ee$\rightarrow$ \pair q events within the Leading Log approximation (LLA). The quark and antiquark emit gluons according to the well-known radiation pattern
\eqbe \mrm dn\srm g \approx \CF\frac{\alpha_s}{\pi} \frac{\mrm dx\srm q \mrm dx\srm{\ol q}}{(1-x\srm q)(1-x\srm{\ol q})} = \CF\frac{\alpha_s(p_\perp^2)}{\pi} \frac{\mrm d
p_\perp^2}{p_\perp^2} \mrm dy \equiv \CF \frac{\alpha_s(\kappa)}{\pi} \mrm d\kappa
\mrm dy;~~~~\kappa \equiv \ln(p_\perp^2/\Lambda^2). \label{e:dng} \eqen
We have here used Eq~(\ref{e:PSelement}), and in the following we define $p_\perp$ and $y$ according to Eqs~(\ref{e:ptLdef}) and~(\ref{e:yLdef}), unless otherwise stated.
Due to colour coherence the hadronic multiplicity $\N g(\kappa)$ in a
gluon jet depends on the $p_\perp$ of the gluon and not on its energy (see, e.g., refs~\cite{ADKT,khoze}).
Summing up the contributions from all gluons in a cascade we arrive at the
average multiplicity $\Nqq(L=\ln(s/\Lambda^2))$ in the original $\mrm{q\ol q}$ system~[13, 15-18]
(Refs~[15-18]
include also nonleading terms.)
\eqbe
\Nqq(L) \approx \int_{\kappa_0}^{L} \mrm d\kappa \int_{-\frac{1}{2} (L -
\kappa)}^{\frac{1}{2} (L -
\kappa)} \mrm dy \CF \frac{\alpha_s(\kappa)}{\pi} \N g(\kappa) =
\int_{\kappa_0}^{L} \mrm d\kappa (L - \kappa)\CF \frac{\alpha_s(\kappa)}{\pi}
\N g(\kappa). \label{e:Nq} \eqen
(We have here introduced a lower cutoff $\kappa_0$ for the integral over transverse momentum. This point will be discussed in section~\ref{sec:cutoff}.)
Taking the derivative with respect to $L$ we find
\eqbe
N^\prime\srm {q\ol q}(L) \approx \int_{\kappa_0}^{L} \mrm d\kappa \CF \frac{\alpha_s(\kappa)}{\pi}
\N g(\kappa). \label{e:Nqprime} \eqen
Consider now a sample of events selected in such a way that there are {\em no} subjets with $p_\perp>p_{\perp\mrm{cut}}$. (Within a $k_\perp$-based cluster scheme with a resolution parameter $p_{\perp\mrm{cut}}$, this means that there are only two primary q and \anti q jets.)
To obtain the multiplicity $\Nqq(L,\kappa\srm{cut})$ in this biased sample, we
must restrict the $\kappa$ integral in Eq~(\ref{e:Nq}) to the region $\kappa < \kappa\srm{cut}$. We then find~\cite{Lundbias}
\eabe
\Nqq(L,\kappa\srm{cut}) & \approx &
\Nqq(\kappa\srm{cut}) +
(L - \kappa\srm{cut}) N^\prime\srm{q\ol q}(\kappa\srm{cut}).
\label{e:Nqtwoscale} \eaen
The first term corresponds to two cones around the q and \anti q jet directions. Here the $p_\perp$ of the emissions is limited by the kinematical constraint in Eq~(\ref{e:yptranges}) rather than by $\kappa\srm{cut}$. It also corresponds exactly to an unbiased $\mrm{q\ol q} $ system with cms energy $p_{\perp\mrm{cut}} $. The second term describes a central rapidity plateau of width $(L - \kappa\srm{cut})$, in which the limit for gluon emission is given by the constraint $\kappa\srm{cut}$. This expression for a two-jet event can be generalized for a biased multi-jet configuration, and a similar discussion applies also to the multiplicity variance, cf.\ ref~\cite{Lundbias}. (Similar equations for biased two-jet and three-jet events were later discussed also in ref~\cite{Catanibias}.)
The average particle multiplicity in the selected two-jet sample is smaller than in an unbiased sample. The modification due to the bias is similar to the suppression from a Sudakov form factor. It is formally ${\cal O}(\alpha_s)$, but it also contains a factor $\ln^2(s/p_{\perp}^2)$. Thus, it is small for large $p_\perp$-values but it becomes significant for smaller $p_\perp$. This clearly demonstrates that the multiplicity in this restricted case depends on {\em two} scales, $\sqrt s$ and $p_{\perp\mrm{cut}}$. The $p_\perp$ of an emitted gluon is related to the virtual mass of the radiating parent quark. Therefore, the two scales $\sqrt s/2$ and $p_{\perp\mrm{cut}}$ represent the energy and virtuality of the quark and antiquark initiating the jets.
\begin{figure}[tb]
\begin{center}
\mbox{\psfig{figure=nqqratio.ps,width=0.5\figwidth}}
\end{center}
\caption{
{\em The effect from the bias due to a constraint $p_{\perp\mrm{cut}}$ on emitted subjets, at 90GeV energy. The figure shows the ratio of biased over unbiased multiplicities as a function of $p_{\perp\mrm{cut}}$. The results for LLA and MLLA relations (Eqs~(\ref{e:Nqtwoscale}) and~(\ref{e:NqtwoscaleMLLA}), respectively) differ significantly from each other. The result of the MLLA relation in Eq~(\ref{e:NqtwoscaleMLLA}), using the $p\srm{\perp}$ definition in Eq~(\ref{e:ptLdef}), is in good agreement with \textsc{Ariadne} MC and Durham cluster algorithm results.}
}
\label{f:twoscale}
\end{figure}
Though the LLA result in Eq~(\ref{e:Nqtwoscale}) describes the qualitative features of the bias, subleading corrections are needed for a quantitative analysis. Within the Modified Leading Log approximation (MLLA)~\cite{MLLA}, subleading terms are included, which affect the prediction for the unbiased multiplicities and, thus, implicitly also the biased multiplicity in Eq~(\ref{e:Nqtwoscale}). Furthermore, it is in~\cite{jetscales} shown that the expression in Eq~(\ref{e:Nqtwoscale}) for the biased multiplicity is explicitly changed when MLLA corrections are considered.
An unbiased system should be restored when $p\srm{\perp cut}$ approaches the kinematical limit $\sqrt s/2$, but the r.h.s.\ of Eq~(\ref{e:Nqtwoscale}) equals the unbiased quantity $\Nqq(L)$ only when $p\srm{\perp cut} = \sqrt s$. The relation consistent with the MLLA is~\cite{jetscales}
\eabe
\Nqq(L,\kappa\srm{cut}) & \approx &
\Nqq(\kappa\srm{cut}+c\srm q) +
(L - \kappa\srm{cut}-c\srm q) N^\prime\srm{q\ol q}(\kappa\srm{cut}+c\srm q);~~~c\srm q = \frac32.
\label{e:NqtwoscaleMLLA} \eaen
The bias is illustrated in Fig~\ref{f:twoscale}. The dotted line shows results from the \textsc{Ariadne} MC~\cite{ariadne}, when the Durham cluster algorithm~\cite{Durham} is used to define a biased sample of events classified as two-jet events with a $y\srm{cut}$ equal to $p^2_{\perp\mrm{cut}}/s$. The MC results agree well with the prediction of Eq~(\ref{e:NqtwoscaleMLLA}), where for $p_{\perp\mrm{cut}}$ we have used the $p_\perp$-definition in Eq~(\ref{e:ptLdef}) (solid line). The predicted effect is below 5\% for $p_{\perp\mrm{cut}} > 20$GeV, but increases rapidly for smaller $p_{\perp\mrm{cut}}$.
Fig~\ref{f:twoscale} presents also the result using the LLA relation in Eq~(\ref{e:Nqtwoscale}) (dashed line). To elucidate the effect of the differences between Eq~(\ref{e:Nqtwoscale}) and~(\ref{e:NqtwoscaleMLLA}), we have used the same expression for the unbiased quantities $\Nqq$ and $N^\prime\srm{q\ol q}$. (These are obtained by a simple fit to \textsc{Ariadne} MC results, which are in good agreement with the MLLA\@.) As seen, the subleading terms are important; the LLA relation significantly overestimates the effect.
To our knowledge experimental data for this bias have not been presented. Such data should be obtainable in a rather straightforward analysis, which thus readily could test the accuracy of the MC result or the MLLA relation.
\section{Infrared cutoffs}\label{sec:cutoff}
Gluon radiation diverges for collinear and soft emissions. Therefore, to
estimate the hadronic multiplicity from the assumption of LPHD~\cite{LPHD}, a cutoff is needed.
Naturally, the cutoff must be Lorentz invariant.
For collinear emissions a single Feynman
diagram dominates, and there are two possibilities, the virtual mass,
$\mu$, of the emitting parent parton or the transverse momentum, $p_\perp$,
of the emitted gluon measured relative to the parent parton direction.
These quantities are connected by the relation
\eqbe p_\perp^2 = \mu^2 z(1-z), \eqen
where $z$ equals the light cone momentum fraction taken by the emitted gluon.
The transverse momentum is directly related to the
formation time, and, therefore, we regard this as the most natural choice
for a cutoff. (For a further discussion see ref~\cite{jetscales}.)
For soft emissions no obvious cutoff is available, however. As several
Feynman diagrams contribute and interfere, there is no unique parent
parton. Consequently $\mu^2$ or $p_\perp^2$ cannot be uniquely specified and,
therefore, cannot be directly used. (Obviously a cut in energy is not
possible, as this is not Lorentz invariant.)
For soft emissions from a single $\mrm{q\ol q}$ colour dipole a cutoff in $p_\perp$
is still the natural choice if measured in the cms, where the $\mrm q$ and
$\mrm{\ol q}$ move back to back. For emissions from a more complicated state the
situation simplifies greatly in the large-$\N c$ limit, as many
interference
terms disappear. In this limit the emission corresponds to a set of {\em
independent} colour dipoles~\cite{dip,earlyLund}.
The natural choice for the cutoff is then
$p_\perp$ in the cms of the emitting dipole (measured with respect to the
dipole direction). We note that this implies that the soft gluons
connect the hard partons in exactly the same way as the string in the
string fragmentation model~\cite{string}, which illustrates the connection between perturbative QCD and the string model~\cite{ADKT}.
For the physical case with 3 colours, extra interference terms appear
with relative magnitude $1/\N c^2$~\cite{ADKT,DKT...}. Here nonplanar
Feynman diagrams contribute, and it is impossible to uniquely specify
a parent parton or a relevant $p_\perp$. Thus, a more fundamental
understanding of confinement is needed to specify the cutoff, which
cannot be determined from perturbative QCD alone~\cite{jetscales}. In
hadronization models the $1/\N c^2$ interference terms correspond to
the problem of ``colour reconnection'', and different models have been
proposed~\cite{reconnection}. None of these can be motivated from
first principles, and only experimental data can differentiate among
the various models.
In spite of the formal uncertainties, the success of current Monte
Carlo programs~\cite{MC,ariadne} indicate that the colour suppressed
interference terms do not have a very large effect. This is also
supported by recent searches by OPAL of the reconnection effects in
hadronic $Z$ events~\cite{OPAL}. In most parton cascade formalisms,
a cascade cutoff motivated in the large-$\Nc$ limit is used also for
finite $\Nc$. The colour interference effects are accounted for by
reducing the colour factor from $\Nc/2$ to $\CF$ in regions collinear
with quarks and antiquarks, and, due to colour coherence, also in some
parts of the central rapidity region. We note, however, that some
subtle interference phenomena, as a matter of principle, cannot be
absorbed into a probabilistic scheme, see~\cite{DKT...} for
details. These are still awaiting a thorough experimental test.
\section{Formalism for three-jet events}\label{sec:qqg}
After these general discussions we are now ready to consider three-jet
\pair qg systems. To simplify the discussion we first study the
large-$\Nc$ limit. The emission of softer gluons from a
$\mrm{q\ol{q}g}$ system corresponds then to two dipoles which emit gluons independently. If a gluon jet is resolved
with transverse momentum $p_\perp$, this imposes a constraint on the
emission of subjets from the two dipoles. Thus, the contribution from
each dipole is determined by an expression like
Eq~(\ref{e:NqtwoscaleMLLA}). For relatively soft primary gluons the
constraint should be given by $p_{\perp\mrm{cut}} =
p_{\perp\mrm{g}}$. For hard gluons $p_{\perp\Lund}$ is of
the same order as its parent quark virtuality, and in ref~\cite{conny}
it is shown that ${\cal O}(\alpha_s^2)$ matrix elements are well described if $p_{\perp\Lund}$ is used as an ordering
parameter for the perturbative cascade. This is also indicated by the successful applications of the \textsc{Ariadne} MC\@. We will, therefore, assume that the constraint on further
emissions is well described by the identification $p_{\perp\mrm{cut}}
= p_{\perp\Lund}$. The multiplicity in a qg dipole
with an upper limit on $p_\perp$ can, just as for the \pair q case discussed in
section~\ref{sec:bias}, be described as two forward jet regions and a
central plateau.
We note that if the three-jet events were selected using a
cluster algorithm with a {\em fixed} resolution scale
, then the constraint on subjet transverse momenta, $p_{\perp\mrm{cut}}$, would be
smaller than the $p_\perp$ of the
gluon jet (as the gluon jet was resolved). In this case most jet definitions give three jets which are all biased~\cite{Lundbias,Catanibias}. We will, however, here focus
on three-jet configurations obtained by iterative clustering until
exactly three jets remain, {\em without} a specified resolution scale, where hence the constraint on subjet $p_\perp$ is
described by $p_{\perp\mrm{cut}} = p_{\perp\Lund}$. As we will see, this implies that the bias on the gluon jet is negligible, which makes this selection procedure suitable for an investigation of unbiased gluon jets.
For finite $\Nc$ the different dipoles in a multi-parton configuration can not be completely independent of each other. However, encouraged by the success of MC programs, let us assume that the main effect of finite $\Nc$ is that the colour factor, which determines
softer gluon emission, is reduced from $\Nc/2$ to $\CF$ in the domains
where the emission is dominated by radiation from the quark or the
antiquark leg. Let us assume that a rapidity range $Y\srm q$ in the $\mrm{qg}$
dipole is similar to a corresponding range in a $\mrm{q\ol q}$ dipole,
while the remaining range $L\srm{qg} - Y\srm q$ is similar to a range in
one half of a $\mrm{gg}$ system. The corresponding ranges in the
$\mrm{g\ol q}$ dipole are $Y\srm{\ol q}$ and $L\srm{g\ol q} -
Y\srm{\ol q}$. This implies that the total multiplicity in the \pair qg event corresponds to the expression
\eqbe \N{q\ol qg} = \Nqq(\YY,\kappa_{\Lund}) + \frac1 2\Ngg(L\srm{qg} + L\srm{g\ol q} - Y\srm{ q}-Y\srm{\ol q},\kappa_{\Lund}). \label{e:3scales} \eqen
For the constraint $p_{\perp\mrm{cut}}$ we have here written $\kappa_\Lund$, which is appropriate for the selection procedure discussed above.
As discussed in section 4, the size of $Y\srm{q}$ and $Y\srm{\ol q}$ cannot be uniquely determined within perturbative QCD. Possibly the most natural choice is to assume that the quantity $\YY$ corresponds to the energy in the \pair q subsystem~\cite{khoze}, which implies
\eqabcA {\YY \approx \ln(s\srm{q\ol q}/\Lambda^2) \equiv L\srm{q\ol q}. \label{e:YqqSP}}{3ex}
The relation in Eq~(\ref{e:YqqSP}a) can be regarded as an educated guess, but a finite shift cannot be excluded. In ref~\cite{GG} it is assumed that
\eqabcB {\YY \approx \ln (s/\Lambda^2) = L, \label{e:YqqL}}{3ex}b
which agrees with Eq~(\ref{e:YqqSP}a) to leading order. For relatively soft gluons we have $s\srm{q\ol q} \approx s$, and in this case Eqs~(\ref{e:YqqSP}a) and~(\ref{e:YqqL}b) are approximately equivalent. The assumption in Eq~(\ref{e:YqqSP}a) implies that the energy scale for the gluon term is given by $L\srm{qg} + L\srm{g\ol q} - L\srm{q\ol q} = \kappa_{\SP}$. Similarly we get from Eq~(\ref{e:YqqL}b) the corresponding gluonic energy scale $\kappa_{\srm{Lu}}$.
The effect of the $p_\perp$ constraint is rather different in the two terms in Eq~(\ref{e:3scales}). For the gluon term the energy scale is in general only slightly larger than the bias scale $\kappa_\Lund$. This implies that in most cases the bias can be disregarded in this term. Inserting the different assumptions in Eqs~(\ref{e:YqqSP}a) and~(\ref{e:YqqL}b)
into Eq~(\ref{e:3scales}) then gives
\eqabcA {\N{q\ol qg} \approx \N {q\ol q}(L\srm{q\ol q},\kappa_\Lund) + \frac{1}{2}\N {gg}( \kappa_{\SP}), \label{e:Nqqg}}{3.3ex}
\vspace{-4ex}
\eqabcB{\N{q\ol qg} \approx \Nqq(L,\kappa_\Lund) + \frac1 2\Ngg(\kappa_\Lund).\label{e:NqqgL}}{3.3ex}b
We note that the consistency between Eqs~(\ref{e:Nqqg}a) and~(\ref{e:NqqgL}b) follows from the fact that the total rapidity range in the two dipoles, $L\srm{qg} + L\srm{g\ol q}$, can be expressed in two different ways by the equalities $L\srm{qg} + L\srm{g\ol q} = L\srm{q\overline q} + \kappa_\Le = L+ \kappa_\Lund $. In particular, we see from these equalities that the argument in $\Ngg$ has to be $p_{\perp\SP}^2$ in Eq~(\ref{e:Nqqg}a) and $p_{\perp\Lund}^2$ in Eq~(\ref{e:NqqgL}b), and not e.g.\ $(2p_\perp)^2$.
\begin{figure}[tb]
\parbox{0.47\textwidth}{
\mbox{\psfig{figure=nqqgbias.ps,width=0.45\textwidth}}
\caption{ {\em $\N{q\ol qg}$ as a function of $p_{\perp\Lund}$ for $\sqrt{s\srm{q\ol q}} = $ 60 GeV. The different predictions from Eqs~(\ref{e:Nqqg}a,b) and~(\ref{e:NqqgSP}) illustrates the importance of the bias at moderate $p_\perp$.}}
\vspace{6ex}
\label{f:Nqqg}
}
\hspace{0.03\textwidth}
\parbox{0.47\textwidth}{
\mbox{\psfig{figure=nggbias.ps,width=0.45\textwidth}}
\caption{{\em The prediction for $\Ngg$, obtained by subtracting from $\N{q\ol qg}$ the quark contribution $\Nqq$, changes significantly if the bias in the \pair q-term is neglected. The figure shows the effect for $\sqrt{s\srm{q\ol q}}=60$GeV, with $\N{q\ol qg}$ given by Eq~(\ref{e:Nqqg}a).}}
\label{f:Ngg}
}
\end{figure}
The leading effect of a finite shift in $ \YY$ is colour-suppressed, and therefore not expected to be large. However, subleading corrections introduce a difference between the results of Eqs~(\ref{e:Nqqg}a) and~(\ref{e:NqqgL}b). This is seen in
Fig~\ref{f:Nqqg}, where the difference is approximately 1 particle for $\sqrt{s\srm{q\ol q}}=60$GeV.
In the calculations of $\N{q\ol qg}$ in Fig~\ref{f:Nqqg}, we have used the expressions in~\cite{jetscales} for the multiplicities $\Nqq$ and $\Ngg$. These include MLLA corrections and recoil effects, which implies that $\Ngg < 2\Nqq$ for accessible energies. Consequently, the result for $\N{q\ol qg}$ grows with the assumed value of $\YY$.
While the bias is not serious for the gluon term in
Eq~(\ref{e:3scales}), it is more important for the \pair q term. Focusing
on events with comparatively large values of $p_{\perp }$, where
the bias is less essential, and using the assumption in
Eq~(\ref{e:YqqSP}a), we arrive at the result of ref~\cite{khoze}:
\eqbe \N{q\ol qg}(s,p_{\perp \SP}^2) = [\N{q\ol q}(s\srm{q\ol q})+ \frac1 2\Ngg(p_{\perp \SP}^2)](1+ {\cal O}(\alpha_s)).\label{e:NqqgSP} \eqen
The bias is formally of order $\alpha_s$, and is here taken into account by the factor $(1+{\cal O} (\alpha_s))$. The result of this expression, neglecting the ${\cal O} (\alpha_s)$ term, is also shown in Fig~\ref{f:Nqqg}. The effect of the bias corresponds to less than one charged particle for $p_{\perp \mrm{cut}}$ larger than $\sim 10$GeV, but becomes much more important for smaller $p_{\perp \mrm{cut}}$-values.
An alternative way to express this result is the effect on extracting $\N{gg}$ from data for $\N{q\ol q g}$, as illustrated in Fig~\ref{f:Ngg}. $\Ngg$ can be extracted by subtracting the biased quark multiplicity $\Nqq(L\srm{q\ol q},\kappa_\Lund)$ from $\N{q\ol qg}$, here assumed to be described by Eq~(\ref{e:Nqqg}a). Neglecting the bias in the subtracted $\Nqq$ term gives a significantly different result. The relative effect of the bias is in this case larger, and it exceeds 20\% for $p_\perp < 15$GeV. Furthermore, to get a reliable result for $\Ngg$, the relevance of subleading terms in the biased quark multiplicity needs to well understood. For the solid line in Fig~\ref{f:Ngg}, the MLLA relation in Eq~(\ref{e:NqtwoscaleMLLA}) is used to subtract the \pair q contribution from the total multiplicity. Instead using the LLA relation in Eq~(\ref{e:Nqtwoscale}) would give a prediction for $\Ngg$ which is about three charged particles higher for most values of $p\srm{\perp cut}$.
Although the effect of the bias is very important for small $p_\perp$, we also see from Figs~\ref{f:Nqqg} and~\ref{f:Ngg} that it can be neglected for large $p_\perp$-values, where, thus, the results in ref~\cite{khoze} and Eq~(\ref{e:NqqgSP}) can be safely used. This implies e.g.\ that the bias is negligible in gluon systems defined as the hemisphere opposite to two quasi-collinear quark jets, thoroughly investigated by OPAL~\cite{JWGary,OPAL}.
It would be very interesting to compare the results in Figs~\ref{f:Nqqg} and~\ref{f:Ngg} to experiments. Experimental data on $\N{q\ol qg}$ can be directly compared to the Monte Carlo or MLLA results in Fig~\ref{f:Nqqg}, Data on the difference $\N{q\ol qg} - \Nqq$ can be compared either to the predictions in Fig~\ref{f:Ngg} or to experimental results for $\Ngg$ obtained through one of the methods described in ref~\cite{jetscales}. We have compared the results in Fig~\ref{f:Nqqg} with MC simulations, where the $p_\perp$ scale is determined by the Durham cluster algorithm. The MC results (not shown) indicate that an analysis based on jet reconstruction is accurate enough to illustrate the effects the bias, but perhaps not to distinguish between the assumptions in Eq~(\ref{e:Nqqg}a) and~(\ref{e:NqqgL}b). We also note that the effects described here may have a phenomenological impact on the recent analysis of $\N{q\ol qg}$~\cite{DELPHI}, which employs the two-scale dependence.
\section{Conclusion}
A series of subtle effects influence an analysis of the difference
between quark and gluon jets in a real life experiment. In this letter
we discuss and clarify effects associated with
\begin{itemize}
\item the definition of $p_\perp$,
\item the bias from restrictions on subjet $p_\perp$,
\item the problem that infrared cutoffs cannot be uniquely defined from perturbative QCD.
\end{itemize}
We also demonstrate the impact of these effects on the analysis of
three-jet events in $e^+e^-$-annihilation.
\subsubsection*{Acknowledgments}
We thank K. Hamacher, W. Ochs, R. Orava and T. Sj\"ostrand for useful discussions.
This work was supported in part by the EU Fourth Framework Programme `Training and Mobility of Researchers',
Network `Quantum Chromodynamics and the Deep Structure of Elementary Particles', contract FMRX-CT98-0194
(DG 12 - MIHT).
|
\section{Introduction}
Although the first mid-infrared (5-20\,$\micron$) spectra of late-type
stars were taken over thirty years ago, no systematic comparison with
more recent spectra of identical sources has ever been made. In the
late 1960's and early 1970's, Merrill, Stein, and collaborators
obtained a homogeneous set of mid-infrared spectra on a large set of
oxygen-rich and carbon-rich asymptotic giant branch (AGB) stars. In
the 1980's, the Infrared Astronomical Satellite (IRAS) provided an
atlas of 8-22\,$\micron$ spectra, but on the question of comparison
with previously observed sources the IRAS Explanatory Supplement
merely stated: ``The comparison showed satisfactory results
(\cite{iras88}, pp. IX-20).'' Recently, we published mid-infrared
spectra of red giants and supergiants, many of which show long-period
variability, with increased temporal coverage to resolve spectral
changes taking place on a pulsational time scale (\cite{mgd98}=MGD98).
MGD98 identified stars that showed significant spectral shape
variations during a pulsational cycle and characterized the observed
changes. Some observers in the past have interpreted spectral changes as
reflecting long-term changes rather than pulsational ones (e.g.,
\cite{fgs75}), but their data did not adequately sample a sufficiently
long time interval to test this interpretation. Cognizant of these
issues, we now compare these three homogeneous sets of data to search
for significant long-term changes in spectral shape.
The 8-13\,$\micron$ spectra of most of these stars contain significant
emission from circumstellar dust grains. If this dust is forming in
steady state at a typical condensation temperature of $\sim$1200\,K,
then the inner dust radius is at $\sim$3-4~R$_\star$, corresponding to
about 10\,AU for Mira variables. Most of the mid-infrared dust
emission comes from this region containing the hottest dust. If the
dust is moving at typical velocities for molecules observed in the
outflows ($\sim 10 \km\secd^{-1}$) then the emitting dust within $\sim$20~AU is
replenished every 10~years. Hence, we expect an investigation of
mid-infrared spectra over a 25~year time span to effectively probe the
temporal uniformity of the mass-loss around a given star.
In this paper we first introduce and describe the data sets used for
spectral comparison. Care was taken to understand the particular
calibration procedures and most of the spectra were extracted from
large homogeneous sets. Also by comparing these sets of data
as a whole, systematic biases in any given data set may be discovered
through comparison with the other two. Indeed, in our comparison of the data
sets we have discovered a subset of miscalibrated LRS spectra in the IRAS
database. Lastly, we discuss the implications of our results for the
understanding of mass-loss processes on the AGB and single out a few stars
for special comment.
\section{The Data Sets}
This paper makes use entirely of published mid-infrared spectra, and
we refer the interested reader to the original articles for complete
observing details. For most stars considered here, three epochs of
data exist in the literature: one from the late 1960's or early 1970's
(Epoch I), one from the IRAS-LRS Atlas observed in 1983 (Epoch II),
and, lastly, UKIRT data from the mid-1990's (Epoch III). See Table~1
for additional information on the target stars. We will now describe
the data set used for each epoch and the calibration issues unique to
each.
Merrill \& Stein (1976abc) published the most extensive set of
mid-infrared spectra during the late 1960's and 1970's and we have
used their data when available. This strategy makes the comparison of
spectra taken decades apart less susceptible to varying calibration
techniques of the observers. Since all spectra by a given observer
may suffer from the same systematic errors, these errors may be
identified by comparison with a more recent, homogeneous set of
spectra. In addition, spectra from Gillett, Low \& Stein (1968),
Hackwell (1972), Forrest, Gillett \& Stein (1975), and Giguere, Woolf,
\& Webber (1976) were used to supplement the 3-epoch data set
available for analysis. All these authors calibrated their spectra by
observing standard stars and assuming blackbody temperatures (e.g.,
$\alpha$~CMa as a 10000\,K blackbody), although the calibration star
used for each particular spectrum was rarely documented.
Occasionally $\alpha$~Boo (K1.5III) and $\alpha$~Tau (K5III) were
used, being approximated as 4000\,K blackbodies
(\cite{gf73}), a problematic assumption due to absorption in the
fundamental band of SiO (\cite{cohen95}). However, recent work
(\cite{cww92}) has found that the spectrum of $\alpha$~Tau
can be fit very well by a 3800\,K blackbody between 8 and
20\,$\micron$, and hence these calibrations made by Merrill, Stein,
and others are quite satisfactory in this wavelength regime.
The InfraRed Astronomical Satellite (IRAS) was launched in January
1983 and operated for nearly a year. It was outfitted with the
Low-Resolution Spectrometer (LRS) which collected thousands of spectra
at wavelengths between 8 and 22\,$\micron$. The spectrometer measured
two overlapping wavelength ranges: one from 7.7 to 13.4\,$\micron$ and
another from 11.0 to 22.6\,$\micron$. The spectral shape was
calibrated by assuming the intrinsic $\alpha$~Tau spectrum was equal
to a 10000\,K blackbody; this and additional information can be found
in the IRAS Explanatory Supplement (1988). Cohen, Walker \& Witteborn
(1992) measured the mid-infrared spectrum of $\alpha$~Tau
independently and found systematic deviations from the ideal Planck
spectrum, and their correction factors were applied to all IRAS-LRS
data before comparison in this paper. The IRAS-LRS Atlas data were
obtained from an internet repository maintained by Kevin Volk at the
University of Calgary
(http://iras3.iras.ucalgary.ca/\verb+~+volk/getlrs\_plot.html) and
further description of the data set, including scans not originally
included in the IRAS-LRS Atlas, can be found elsewhere (\cite{vc89};
\cite{volk91}; \cite{kvb97}).
The most recent epoch of mid-infrared spectrophotometry considered
here was carried out at the United Kingdom Infra-Red Telescope
(UKIRT) from 1994 to 1997 (\cite{mgd98}). All standard stars used for
flux calibration were of spectral type K0 or earlier, so that
absorption in the fundamental band of SiO, which is very prominent in
late K and M giants and supergiants, is minimal in the ratioed spectra
and does not affect the shapes of the reduced spectra.
In this paper we do not concern ourselves with the absolute flux level of
the mid-infrared spectra, because of the difficulty in reconciling the
broad-band filter bandpasses used by IRAS and the early workers, which
would be required to precisely compare the absolute
photometry. In addition, previous workers sampled the stellar spectrum at
incongruous pulsational phases. For example, Merrill \& Stein (1976ab)
presented relative spectrophotometry accompanied by a table of broad-band
fluxes; unfortunately, spectra obtained at various and undocumented phases
during the pulsational cycle of variable stars were averaged together.
This is also true for the IRAS-LRS spectra; the published Atlas
spectra result from an averaging of multiple scans taken at different times.
Therefore, we present only relative spectra, normalized to the average
broadband 8-13\,$\micron$ flux. Only the recent UKIRT data
(\cite{mgd98}) sampled the stellar spectra throughout the pulsational
cycle, and we include spectra nearest to maximum and minimum for all stars
when possible. Although adding some clutter to the figures, this
allows one to determine if spectral shape differences are likely due
to pulsation effects, rather than long-term evolution of the dust
shell properties.
Observers during Epochs~I and III generally used photometric apertures
about 5\arcsec\, in size, while IRAS had a larger one
(6\arcmin$\times$15\arcmin). The Egg Nebula is the only source in
this paper whose mid-infrared emission is known to be extended on
the former scale. The Infrared Spatial Interferometer has surveyed a
number of these bright mid-infrared sources (e.g., \cite{danchi94})
and showed that nearly all of the mid-infrared emission occurs on
scales smaller than 5\arcsec. However, direct imaging of faint
emission around these stars has rarely been done. In a few cases,
some emission has been detected outside of the small aperture used by
Epoch I and III observers. In particular, Sloan and collaborators
have detected distant silicate material around $\alpha$~Ori
(\cite{sloan93}) and carbon grains around IRC~+10216 (\cite{sloan95}).
While only contributing a small fraction of the total flux, it is
possible that such missing flux can contribute significantly to
spectral features in the mid-infrared region in some cases. Hence some
small differences observed between Epoch II (IRAS) and the other data sets
may be attributable to such an effect.
\section{Results}
\subsection{Categorization Procedure}
In order to avoid the ambiguity associated with subjective
classifications, a quantitative measure of spectral variability was
developed to identify sources exhibiting statistically significant
long-term evolution of their mid-infrared spectral shapes. A simple
$\chi^2$-based statistic, defined in the Appendix, allows robust
identification of stars exhibiting long-term changes in their spectra.
This statistic takes into account the pulsation-related spectral changes observed in MGD98 as well as measurement error in determining the likelihood of
long-term spectral variability.
Using the procedure outlined in the Appendix, two relevant statistics,
$\chi^2_{\rm{Epoch\,I}}$ and $\chi^2_{\rm{IRAS}}$ have been calculated
for all sources, and these results appear in Table\,1. These values
are indicative of the spectral changes occurring between Epoch I and
Epoch III and between Epoch II (IRAS) and Epoch III, respectively.
Epoch I and Epoch II data were not quantitatively compared
to each other because the spectra did not temporally sample the
pulsational cycle of the sources.
The stars in this survey showed a continuum of variability; some stars
evinced virtually no variability, while others showed obvious spectral
changes. In order to divide our sample, we have adopted the cutoff
value of $\chi^2 = 1.5$ to indicate significant variability. Based on
these quantities, the stars have been divided into three categories
and presented in Figures~1-3. Figure~1 contains the spectra of
6~stars which are consistent with the hypothesis of no spectral
evolution between all three epochs. Figure~2 contains spectra of
9~stars that show good agreement between Epoch~I and Epoch~III, but
in which the IRAS spectra appear discrepant. Finally, Figure~3
presents 14~stars whose spectral shapes are significantly
different between Epoch I and Epoch III, independent of the IRAS data.
\subsection{Constant Stars}
Figure~1 contains spectra of stars whose spectral shapes have not
changed, within known uncertainties, between all three epochs under
study. Quantitatively, the selection criterion for inclusion in this
figure is $\chi^2_{\rm Epoch~I}\leq 1.50$ and $\chi^2_{\rm IRAS}\leq
1.50$; in addition, three stars with no available Epoch I data are
included here. Many of these stars are calibrators because of
their general lack of variability and relatively little dust emission
(e.g., $\alpha$~Boo, $\alpha$~Tau, $\delta$~Oph). The excellent
agreement found for these stars provides confidence that changes in
spectral shape described in the next two sections are indeed real and
not the result of the differing calibration procedures of the
individual observers. Note that CIT~3 is highly variable, and
the large spectral shape changes occurring during its pulsational cycle
would mask all but the most dramatic long-term spectral evolution.
\begin{figure}
\begin{center}
\centerline{\epsfxsize=\columnwidth{\epsfbox{figure1.eps}}}
\caption{Normalized mid-infrared spectra of late-type stars that showed no
apparent changes in spectral shape using data from Epochs I, II, and III.
Epoch III spectra near maximum and minimum flux are shown.
Spectra references are for
Gillett, Low, \& Stein (1968=GLS68), Hackwell (1972=Hackwell72),
Forrest, Gillett, \& Stein (1975=FGS75),
Merrill \& Stein (1976=MS76abc), IRAS Science Team (1986=IRAS86),
and Monnier, Geballe, \& Danchi (1998=MGD98).
}
\end{center}
\end{figure}
\subsection{IRAS data}
Figure~2 presents spectra of stars that show excellent agreement
between the earliest and the most recent epochs ($\chi^2_{\rm Epoch~I}
\leq 1.5$), but whose IRAS measurements show systematic disagreement
($\chi^2_{\rm IRAS} > 1.5$) with recent data. In a few obvious
cases, the miscalibration of one of the IRAS-LRS detectors appears
responsible (e.g., $\alpha$~Her) for this discrepancy. The red and
blue detectors of IRAS-LRS share a common wavelength region between
11.1 and 13.4\,\micron, and should report self-consistent spectral
fluxes. In the Appendix, we define the statistic $\sigma_{\rm
overlap}$ which quantifies how well the observed spectral slopes in
the two LRS detectors agree in the overlap region. In short, a
straight line is fitted to the ratio of the blue and red data in the
overlap region. For self-consistent data, the slope should be zero
and $\sigma_{\rm overlap}$ is the number of standard deviations away
from zero of the best-fit slope.
Hence, values above 3 are very
unlikely ($<1$\%) to occur by chance for well-calibrated data and
Gaussian random noise. This statistic has been calculated for all
IRAS data in this paper and the results can be found in Table~1.
\begin{figure}
\begin{center}
\figurenum{2}
\centerline{\epsfxsize=\columnwidth{\epsfbox{figure2_a.eps}}}
\caption{Normalized mid-infrared spectra of late-type stars that showed no
apparent changes in spectral shape between Epochs I and III, but whose
IRAS spectra (Epoch II) appear discrepant.
Refer to Figure~1 for spectral references.
}
\end{center}
\end{figure}
\begin{figure}
\figurenum{2}
\begin{center}
\centerline{\epsfxsize=\columnwidth{\epsfbox{figure2_b.eps}}}
\caption{continued}
\end{center}
\end{figure}
The spectra in this figure are ordered by increasing $\sigma_{\rm
overlap}$; hence, spectra at the end of the figure show this
miscalibration effect most strongly. There is no evidence for
miscalibration problems in the overlap region ($\sigma_{\rm
overlap}\simle2.0$) for the first three spectra (SW~Vir, RX~Boo, Egg
Nebula). The large angular size of the Egg Nebula is the most likely
explanation for the different spectral shapes observed for that star
(see previous discussion at the end of \S2), However, RX~Boo and
SW~Vir are not known to be extended, and the apparent spectral
variability may indicate the presence of transient
strengthening/weakening of the silicate feature around these two
Semi-regular variable stars. Note that only one Epoch III spectrum exists
for SW~Vir, hence the importance of pulsation-related
variability on this star's spectral shape is not known. In fact, the formal
$\chi^2_{\rm Epoch~I}$ for SW~Vir is larger than 1.5, but was included
in this figure because differences between the Epoch I and Epoch III
data are likely due to pulsation effects not quantified by the
statistic.
By ordering the spectra by $\sigma_{\rm overlap}$, we can see a
systematic spectral ``tilt'' associated with the spectral
slope-mismatch. All the IRAS spectra in Figure~2 with large
$\sigma_{\rm overlap}$ also evince an overall spectral shape which is
{\em bluer} than the Epoch I and III data, easily seen as excess
emission between 8 and 9\,\micron. Hence, it is reasonable to conclude
that most of the IRAS-LRS data with high $\sigma_{\rm overlap}$ values
have some residual miscalibration, not previously identified, and
appear slightly bluer than in reality. Inspection of Table~1 reveals
that 11 of the 29 IRAS-LRS spectra considered here have $\sigma_{\rm
overlap}\geq 3.0$. An understanding and future correction of
this miscalibration may be useful.
\subsection{Evidence for long-term changes}
The spectra of the remaining 14~stars appear in Figure~3a-c, and all show
significant changes between Epochs~I and III. The Epoch~II spectra of
some of these stars are flawed in the same manner as discussed in
\S3.3, while others appear to be largely consistent with one epoch or
the other, or intermediate between the two. All spectra here have
$\chi^2_{\rm Epoch~I}>1.5$ and the $\chi^2_{\rm IRAS}$ statistic was
not used in determining membership in this class.
In addition, these 14 stars have been further subdivided based on the type of
spectral variability observed.
\subsubsection{Strong silicate emission features}
Figure 3a contains the spectra of 7 stars whose strong silicate features
show long term variability.
Because the dust grains around these
stars have such strongly wavelength-dependent
opacities near 10~$\micron$, the profile of the silicate
feature and its equivalent width are highly sensitive to changes in the
spectral emission characteristics, temperature, and optical depth structure
of the dust shells. Many of these stars were identified in MGD98 as
possessing silicate features that change shape even during a pulsational
cycle. However, the observed changes between Epoch~I and III lie outside the
envelope of variation previously observed.
\begin{figure}
\begin{center}
\figurenum{3}
\centerline{\epsfxsize=\columnwidth{\epsfbox{figure3a_a.eps}}}
\caption{Normalized mid-infrared spectra of late-type stars that showed
significant changes in spectral shape between Epochs I and III, independent
of the IRAS data. Refer to Figure~1 for spectral references.
{\em a.} Stars with strong silicate emission features. }
\end{center}
\end{figure}
\begin{figure}[t]
\begin{center}
\figurenum{3a}
\centerline{\epsfxsize=\columnwidth{\epsfbox{figure3a_b.eps}}}
\caption{
continued}
\end{center}
\end{figure}
The changes observed in the emission spectra of the dust could be due to
long-term variations in the properties of the dust particles
themselves. Dust optical constants can be affected by changing
chemical abundances or other physical conditions in the stellar
atmospheres (see MGD98) as the star evolves, perhaps linked to thermal
pulses or other nuclear shell burning phenomena. Changes in the
elemental composition may be the only way to explain the dramatic
change in the VY~CMa spectrum (see below).
Alternatively, these differences may be due to episodic mass-loss
events, the observed variations in the
spectra being consistent with an overall cooling
of the circumstellar dust shell. However, interferometric
observations of these dust shells in the infrared indicate the
presence of dust close to the expected condensation radius
(\cite{dyck84}; \cite{danchi94}), making it difficult to understand
why the mean dust shell temperatures should all be decreasing in time.
See MGD98 for more detailed discussions on the dust temperature's
effect on the mid-infrared spectra. In short, the observed changes in
the spectral
shape require a large temperature change which would imply a
dust-star separation incompatible with interferometer measurements.
Regardless of the mechanism for producing the observed changes, the
fact that few stars showed a decrease in the silicate feature peak
relative to the continuum suggests that these stars spend most of
their time building up their silicate peaks. Assuming the silicate
strengthening has been gradual and continuous over the last 25 years,
the typical time between episodes of silicate feature weakening must
be much longer than that probed here. Considering
that 5 Mira variables with the strongest silicate emission
features all showed the same
systematic change in spectral shape, we estimate that this time
scale is greater than about 125~years. Because of the fewer number of
supergiant candidates, we can only say the time scale relevant for
these stars is likely to be greater than $\sim$40~years.
\subsubsection{Spectral slope changes}
Figure 3b shows the spectra of 4 stars whose Epoch I and Epoch III
spectra differ in slope by an amount well outside that expected from the
statistical error in the measurements.
In all cases here, the spectra have become redder, consistent with a
long-term cooling of the dust shell. However, the slopes of the Epoch~II
spectra are not always intermediate
to the Epoch I and III
spectra, and hence the spectral tilt may have different origin (e.g.,
R~Leo). Variation in the dust production on time scale of 10-20 years
could produce such changes, and would be in general agreement with
conclusions based on interferometry (e.g.,
\cite{danchi94}). However, we note that a slight spectral
miscalibration across the wavelength band can also mimic this
effect.
\begin{figure}
\figurenum{3}
\begin{center}
\centerline{\epsfxsize=\columnwidth{\epsfbox{figure3b.eps}}}
\caption{{\em b.} Stars evincing long-term spectral tilts.
}
\end{center}
\end{figure}
\subsubsection{Excess variability of the carbon stars}
The spectra of the 3 carbon stars in our sample are found in Figure 3c.
MGD98 noted that the spectra from the carbon stars showed more
apparent variability on a pulsation time scale than oxygen-rich
stars with weak silicate features. The long-term variability reported here
is also fairly small, but statistically significant. Variability of the complex dust distributions recently observed around a few such
stars using interferometric techniques
(\cite{hb98}; \cite{weigelt98}; \cite{tmd99}) may play a role in this observed
enhanced variability.
\begin{figure}[h]
\figurenum{3}
\begin{center}
\centerline{\epsfxsize=\columnwidth{\epsfbox{figure3c.eps}}}
\caption{
{\em c.} Carbon stars, displaying enhanced variability.
}
\end{center}
\end{figure}
\subsection{Comments on individual sources}
{\em $\alpha$~Ori:}
The silicate feature has shifted toward slightly shorter wavelengths
now. While not consistent with a generally cooling of the dust shell,
this may be caused by slight replenishing of inner dust shell as
recently observed by Bester et al. (1996). Although poor calibration could
explain the difference, changes in the optical properties of the dust
as it ages may be important here.
{\em Egg Nebula:}
Observers during Epochs~I and III generally used
photometric apertures about 5\arcsec\, in size, while IRAS had a
larger one (6\arcmin$\times$15\arcmin). Because this nebula is
extended on the same scale as the smaller, ground-based apertures,
direct comparison of the data sets is difficult.
{\em IRC~+10420:}
The Epoch I data for this source is rather noisy and was included only to
demonstrate that no gross changes in the silicate feature have
occurred (see VY~CMa comments below). The high noise level prevents
a detailed comparison between Epoch I and II data sets.
{\em R~Hya:}
Peculiarly, the IRAS-LRS Atlas spectra from the blue and red detectors
agree well in slope but are noticable offset from one another (see Figure~2).
{\em U Her:}
Because MGD98 only published a single observation of this
source, the identification of this source as experiencing a long-term
spectral variation is suspect. However, the Epoch I observation
appears visually to be outside the typically envelope of
pulsation-related variability seen in the other spectra of Figure~3a.
New observations of this source during a pulsational cycle are
important to resolve this ambiguity.
{\em VY~CMa:}
The mid-infrared spectrum of the red supergiant VY~CMa has undergone a
remarkable change over the last 25 years (\cite{mgd98}). What was once one
of the strongest silicate features observed by Merrill \& Stein (1976a) is
now nearly flat and featureless. Although the spectral slopes of the
IRAS-LRS red and blue detectors are significantly different, it appears that
the strength of the feature was intermediate in 1983 to that observed before
and after and suggests the spectral change did not occur ``overnight,'' but
instead evolved over a number of years. A significant increase in
optical depth, sufficient to cause self-absorption in the silicate
feature, could explain most of the flattening of the spectrum.
However, a self-absorbed silicate feature usually shows more residual structure
(e.g., NML~Cyg, CIT~3) than apparent here, implying the emission
properties of the dust grains themselves have undergone a relatively rapid
transformation.
It is not known what physical mechanism
could cause such a dramatic change in the dust properties, but perhaps could
be due to a change in the chemical abundances. This may be related to
the ``remarkable change'' in the infrared polarization between 1970 and 1974
observed by Maihara et al. (1976). See Monnier et al. (1999) for recent
high-resolution observations which sheds some light on this mystery.
\section{Conclusions}
We have compiled a multi-epoch set of mid-infrared spectra on 29~stars
spanning a $\sim$25~year observing period. This comparison has
established the high quality of published ground-based spectra and
allowed an investigation into the long-term stability of dust shells
and their properties.
The mid-infrared spectral shapes of about half of the target
stars have remained nearly constant over the entire period of time.
This suggests that the mass-loss rates have not changed significantly
during the last 25~years. This constancy has allowed us to show
that IRAS-LRS spectra may be somewhat miscalibrated when the slopes of the
red and blue spectra do not match in their shared wavelength region.
Most of the stars showing strong silicate features have experienced an
increase in the strength of the silicate peak relative to the
continuum. This is naturally explained if an oxygen-rich star generally
experiences long periods of gradual silicate feature strengthening,
punctuated by relatively rare periods when the feature weakens. The
physical conditions regulating dust formation may be varying on this
long time scale, affecting the optical properties of the dust
particles in the outflow. Alternatively, general cooling of the dust
shells over the last 25~years can explain this effect, although this
hypothesis is at odds with interferometric observations of hot dust
around many of these stars. The dramatic near-disappearance of the silicate
emission feature around VY~CMa deserves attention, and may lead to new insights
into mass-loss processes around evolved stars.
This rudimentary analysis highlights the not surprising fact that the
circumstellar environments of late-type stars are not always simple, and
that assumptions of steady-state dust production on time
scales of decades can not always be justified. These data provide further
evidence that mass-loss characteristics and dust signatures change not only
on the time scale of the large amplitude pulsations, but on
significantly longer ones as well.
\acknowledgments
{The authors would like to thank Kevin Volk for providing IRAS-LRS
spectra via his internet repository and for valuable discussion of its
contents. JDM thanks C. D. Matzner and A. P. Zemgulys for
numerous discussions.
This research has made productive use of NASA's Astrophysics Data
System Abstract Service.
This analysis was supported by the National
Science Foundation (Grants AST-9315485 \& AST-9731625)
and by the Office of Naval Research (OCNR N00014-89-J-1583 \&
FDN0014-96-1-0737).
}
|
\section{Introduction}
In the supersymmetric Standard Model (SM), the quadratically divergent
corrections to the Higgs (mass)$^2$ cancel due to supersymmetry (SUSY).
The remaining corrections are logarithmically divergent, proportional
to the SUSY breaking masses of the sparticles (the superpartners of the
SM particles) and result in
a negative Higgs (mass)$^2$ due to the large top quark Yukawa coupling.
Thus, the superpartners of the SM particles
must have masses $\stackrel{<}{\sim} 1$ TeV
in order for
SUSY to solve the gauge hierarchy problem and lead to natural electroweak
symmetry breaking.
With
the sparticle masses at the weak scale,
these new particles
(especially gluinos and squarks)
will be produced in significant amounts at the
LHC.
After the initial discovery of the sparticles,
the focus will be on precision measurements
of their masses and mixings just as, for
example, the next step after the discovery of the heavy quarks
was the measurement of their detailed properties.
In this paper, a relatively clean signal at the LHC for
detecting
the mixing angle between the
scalar partners of the charged leptons (the sleptons) is presented.
A flavor-violating signal is obtained from the production of
real sleptons, followed by their oscillation into a different
flavored slepton, and subsequent decay to a lepton.
Some formulae for these oscillations are given in section \ref{formulae}.
At a $e e$ linear collider, the production of slepton pairs
can then give $e \mu$ events with missing energy. This
was studied in \cite{krasnikov0,nima1}. Dilepton flavor {\em and} CP violating
signals at the LHC and NLC were studied in \cite{nima2}.
At a hadron collider (the LHC), sleptons can be pair-produced by the
Drell-Yan process
giving the same signal. This was studied in \cite{krasnikov2,krasnikov1},
and is a promising signal for large flavor mixing angles and when the
SUSY background is known to be small.
Real sleptons can also be produced at the LHC
in the
decays of the next-to-lightest neutralino $(\chi^0 _2)$, which
are mainly produced in the
cascade decays of gluinos and squarks.
In section \ref{cascade2l}, flavor violating
dilepton events from $\chi_2^0$ decays are briefly
considered.
The production of $\chi^0_2$ {\em pairs} can
give rise to events with 4 leptons, with
the dramatic flavor violating
signal identified by a $(3e+\mu)$ or $(3\mu +e)$ lepton signature,
hard jets, no $b-$jets, and of course
missing energy. This is discussed in section \ref{cascade4l}.
Conditions on the supersymmetric
spectrum that are favorable for the suppression of the
dominant supersymmetric background,
occuring from
{\em heavier} neutralino/chargino and stop decays,
are identified.
Ideas for determining the remaining
dominant supersymmetric
background occuring from $\tau$ decays are also presented.
These are all conveniently
summarized in the Conclusion.
In section \ref{quick}, a brief estimate of the expected
$4$-lepton signal at
a generic point in SUSY parameter space is given.
Next, in section \ref{LHCpoint},
a particular LHC Point \cite{snowmass,ian}
is considered. It is found that at this Point, a $5\sigma$ discovery
($2\sigma$ exclusion) is obtained for a right-handed (RH)
first and second generation
mixing angle $\sin \theta_R >0.13$ $(\sin \theta_R >0.08)$
with an integrated luminosity of $100$ fb$^{-1}$ at {\em low luminosity}.
The discovery potential at high luminosity is still optimistic
though less quantitative, due to uncertainties
in $\tau-$jet detection efficiencies and larger $b-$jet mistagging rates.
In any case,
the values for the mass splitting
(between $\tilde{e}$ and $\tilde{\mu}$)
that are favorable for the discovery
of a signal satisfy the $\mu \rightarrow e \gamma$ bound even
for a maximal
mixing angle. Thus the LHC has the opportunity of probing mixing angles
that are beyond the reach of the current $\mu \rightarrow e \gamma$
limit.
\subsection{Lepton Flavor Violation due to Slepton Mass Mixing}
\label{formulae}
To begin consider the lepton-slepton-neutral gaugino vertex with the leptons
and sleptons in the gauge basis:
\begin{equation}
\tilde{l}_{a}^{\ast \; gauge} \; l^{gauge}_{a}
\chi^0,
\label{basis1}
\end{equation}
where $a =1,2,3$ is a flavor index.
Next perform a
unitary transformation, $V$, on both $l_g$ {\em and} $\tilde{l}_g$ to
go to the mass eigenstate
basis for the $l$'s:
\begin{equation}
\tilde{l}_{\alpha}^{\ast} l_{\alpha} \chi^0.
\label{basis2}
\end{equation}
In this basis the coupling remains diagonal in
flavor space (now denoted
by $\alpha$).
In general,
however, the slepton and lepton mass
matrices are {\em not} related so that
the same unitary matrix, $V$, may not diagonalize them {\em both}.
In this general case,
the slepton (mass)$^2$ matrix in the basis $\tilde{l}_{\alpha}$ is
{\em not} diagonal. For example, even if the slepton (mass)$^2$
matrix in the gauge basis $\tilde{l}_a^{gauge}$ of Eqn.(\ref{basis1})
is diagonal but not
$\propto$ {\bf 1}, it will have off-diagonal elements in the basis
$\tilde{l}_{\alpha}$ of Eqn.(\ref{basis2}).
So,
a further unitary transformation, $W$, is needed
to rotate to the slepton mass basis. In this basis the
slepton-lepton-gaugino vertex is:
\begin{equation}
\tilde{l}_{i}^{\ast} \; l_{\alpha} W_{i \alpha} \chi^0.
\end{equation}
So, in the mass basis for leptons {\em and} sleptons
($l_{\alpha}$ and $\tilde{l}_i$)
a mixing matrix $W \neq$ {\bf 1} in general
appears at the
neutral gaugino-lepton-slepton vertex.
This means that there is a coupling between, for example,
$\tilde{e}$ (in the mass basis),
$\mu$ and $\chi ^0$ -- this will be referred to as SUSY
lepton flavor violation.
The focus of this paper is the detection of
this SUSY lepton flavor violation at the LHC.
The theoretical expectations for $W$ are varied. In
models with broken flavor symmetries, it is expected that
$W \sim V_{KM}$. In such cases a Cabibbo-like mixing angle for
the first two generations and a $\Delta m/m $ close to the
$\mu \rightarrow e \gamma$ bound is expected \cite{flavor}.
In contrast, in models of
gauge-mediated supersymmetry breaking
the dominant contribution
to the soft masses is universal and it naively appears that
there is no interesting flavor physics. There is, however, a subdominant
flavor non-universal supergravity contribution. This likely results in
{\em large} mixing angles \cite{nima2}. The magnitude of $\Delta m/m$ depends
on the supersymmetry breaking scale and while clearly model--dependent,
could easily be $\sim \Gamma /m$ or larger,
which is needed to give an observable flavor-violating signal at the LHC
(this is discussed later in this section).
For simplicity, the
case of $1-2$ mixing with mixing angle $\theta$ is discussed.
In this case there are strong limits on the mixing angle and
the $\tilde{e} - \tilde{\mu}$ mass splitting from lepton
flavor changing processes. For example, $\mu \rightarrow e \gamma$
gives an important constraint. For degenerate left-handed sleptons,
and with the LSP $(\chi^0 _1)$ approximately bino-like $(\tilde{B}^0)$,
the constraint on $\sin 2 \theta _R$ and the mass splitting $\Delta m$
between
the right-handed sleptons is
approximately
\begin{equation}
\sin 2\theta_R (\Delta m) /m \stackrel{<}{\sim}
0.01 \times \sqrt{\frac{BR( \mu \rightarrow e \gamma)}
{4.9 \times 10^{-11}}}.
\label{muegamma}
\end{equation}
(A more proper
formula is given in Section \ref{LHCpoint}).
Suppose a real
selectron is produced in the basis of Eqn.(\ref{basis2}) (say
in association with an electron). Since $\tilde{e} \; (\alpha =1)$
is not a mass eigenstate, there is a probability that as it
propagates it will convert to a $\tilde{\mu} \; (\alpha =2)$
and hence decay into a $\mu$ \cite{nima1,nima2}:
\begin{equation}
P(\tilde{e}_{\alpha = 1} \rightarrow \mu \chi_1^0 ) = 2 \sin ^2 \theta
\cos ^2 \theta \; x,
\label{x}
\label{osc}
\end{equation}
where $x = (\Delta m)^2 / \left( (\Delta m)^2 + \Gamma ^2 \right)$ is
the quantum
interference factor and
assuming $BR ( \tilde{l}
\rightarrow l \chi_1^0 ) = 1$.
Here $\Gamma$ is the decay width of the slepton.
Note that for $\Delta m \stackrel{>}{\sim} \Gamma$ the interference effect can
be neglected so that $x \sim 1$. In this case the oscillation
probability can be large.
Typically, $\Gamma \sim \alpha_{em} m \sim 0.01 m$ so that $x \sim 1$
if $(\Delta m)/m \stackrel{>}{\sim} 0.01.$
This is close to the upper bound from the $\mu \rightarrow e \gamma$ limit,
so there could be a suppression due to either $x$ or $\sin \theta$
\cite{nima1,nima2}.
It is possible, however, that for a specific SUSY spectrum the decay
width could be much smaller than this naive estimate, allowing for a larger
range of $\Delta m /m$ consistent with the rare decay limit (even
for large
mixing angles) {\em and}
$x \sim 1$ so that
the oscillation
signal is not suppressed \footnote{In fact, this occurs at the LHC Point
discussed
in Section \ref{LHCpoint}.}.
Similarly, a neutralino can decay into
$e^+ \; \mu^-$ or $e^- \; \mu^+$ through an intermediate slepton:
\begin{equation}
\chi_2^0 \rightarrow \tilde{l}^+ l^- \hbox{, } \tilde{l}^- l^+
\rightarrow l^+ l^- \chi_1^0.
\label{chi2a}
\end{equation}
Using Eqn.(\ref{osc}) the rate for a flavor violating decay is
\begin{eqnarray}
BR ( \chi_2^0 \rightarrow e^+ \mu^- \chi_1^0)
&=& 2 \;
\sin ^2 \theta \cos ^2 \theta \; x
\times BR ( \chi_2^0 \rightarrow
\tilde{e}^- e^+, \tilde{\mu}^+ \mu^-).
\label{chi2bi}
\end{eqnarray}
Here to simplify notation $BR(\chi_2^0 \rightarrow
\tilde{e}^- e^+, \tilde{\mu}^+ \mu^-) \equiv
BR(\chi_2^0 \rightarrow
\tilde{e}^- e^+)+BR(\chi_2^0 \rightarrow\tilde{\mu}^+ \mu^-)$. This
notation will be used throughout the paper.
Also, the
BR on the right-side of Eqn.(\ref{chi2bi}) is in the absence of any mixing.
In the case of interest here of small mass splittings, $\Delta m \ll m$,
the neutralino decay rate into selectrons or smuons are equal
in the absence of any mixing. Next, in the absence of mixing,
\begin{equation}
BR(\chi_2^0 \rightarrow e^+ e^- \chi^0 _1) =2 \;
BR(\chi_2^0 \rightarrow \tilde{e}^+ e^-) .
\end{equation}
The factor of two occurs since
$\chi_2^0$ may decay to $\tilde{e}$'s of both charges. This
result and Eqn.(\ref{chi2bi}) relates the flavor-violating
and flavor-conserving decays:
\begin{eqnarray}
BR ( \chi_2^0 \rightarrow e^+ \mu^- \chi_1^0) =& 2 \;
\sin ^2 \theta \cos ^2 \theta \; x
\times BR ( \chi_2^0 \rightarrow l^+ l^- \chi^0 _1),
\label{chi2b}
\end{eqnarray}
where the BR on the right-side of the above equation is in the
absence of mixing.
Here $l$ is either $e$ or $\mu$.
This result applies for $\chi_2^0$ decays to real sleptons, {\it i.e.,}
for $m_{\chi_2^0} > m_{\tilde{l}}$.
For $m_{\chi_2^0} < m_{\tilde{l}}$, there is an
additional suppression of $(\Delta m)/m$ in
the decay amplitude due to the supersymmetric analog of the
Glashow-Iliopoulos-Maiani (GIM)
cancellation as in the case of $\mu \rightarrow e \gamma$,
resulting in negligible $e \mu$ signal. So an observable $e \mu $
signal requires the production of real sleptons \footnote{Alignment
models with $\Delta m \sim m$ are not considered here since $\sin \theta
\sim O(10^{-2})$.}.
\section{Slepton Production by Drell-Yan Process}
\label{dy}
One way to produce sleptons at a hadron collider is through the
Drell-Yan process:
\begin{equation}
p \; p (\hbox{or} \; \bar{p})
\stackrel{\gamma,Z}{\rightarrow} \tilde{l}^{\ast} \tilde{l} \rightarrow
l^+ l^- \chi_1^0 \chi^0_1.
\end{equation}
Thus the production of sleptons is identified by
events with no jets,
2 hard isolated leptons and $ \not \! p_T$, assuming that
$\chi^0 _1$ is stable or decays outside the detector.
These events will be referred to as ``flavor conserving''
dilepton events.
There is a SM background to the signal
from $W^+ W^-$ and $\bar{t} t$ production. These backgrounds are known,
in principle. In \cite{baer} a set of
kinematic cuts on the leptons, as well as a jet--veto, are found which
sufficiently reduce these backgrounds.
These cuts reduce the signal as well -- of course, the reduction is much more
for the background.
There is also a
SUSY background from $p p \rightarrow
\chi ^ + \chi ^ -
\rightarrow W^+ W^- \chi _1 ^0 \chi _1 ^0$.
This background
depends
on the model--dependent $\chi^+ \chi^-$ production cross section.
But, for supergravity motivated parameter choices with
$m_{\tilde{q}} \approx m_{\tilde{g}}$, this background can
be sufficiently reduced by using the same cuts used to remove the SM
background \cite{baer}.
For example, from the analysis of \cite{baer} (see Table III of \cite{baer})
with $10$ (fb)$^{-1}$ and for a
slepton mass $\sim 100$ GeV
there are $\sim 20$
signal events with no background events remaining after the cuts.
Actually, a clever method \cite{krasnikov1}
for detecting the sleptons is to form the
asymmetry $A_F=N(e^+e^- + \mu^+ \mu^-)-N(e^+\mu^- + e^-\mu ^+)$.
The background does not contribute to $A_F$, so a non-zero
value would provide evidence for slepton production.
In the lepton flavor mixing case the pair production of
sleptons will produce
$e \mu$ events with $ \not \! p_T$ -- these
events will be referred to as ``flavor
violating'' dilepton events.
The background to this signal is from the same sources as for the
flavor conserving dilepton signal (with the
same rate) as well as from $\tilde{\tau} \tilde{\tau}^{\ast}$ production
followed by leptonic decays of $\tau$s.
The
detection of SUSY lepton flavor violation
using the above
flavor violating dilepton events for the CMS detector at the LHC
was studied in references \cite{krasnikov2,krasnikov1} for
the case of maximal mixing $(\theta = \pi/4)$.
With the mixing angle being maximal, the flavor violating dilepton
signal rate is high; see Eqn.(\ref{osc}) (assuming $x \sim 1$).
In fact, the number of flavor conserving and flavor
violating events from slepton production in this case are the same and
each is equal
to one half the signal in the zero mixing case so that
$A_{F} \approx 0$ (unlike the case of zero or non-maximal
mixing). In the case where
the production cross--sections for
staus and the lightest charginos are comparable to that of the sleptons,
the production rate for the
SUSY background to $e \mu$ events
is $\sim 4 \%$ of the total
flavor conserving signal (in the absence of
mixing). \footnote{Here, it is assumed that $BR \left( \chi^+
\rightarrow W^+ \chi_1^0 \right) \approx 100 \%$ so that
the leptonic BRs of $\chi^+$ are the same as for $W$.
If the left-handed sleptons
are lighter than $\chi^+$, then the leptonic BRs of $\chi^+$ may be enhanced
substantially, in turn increasing the SUSY background.}
Thus, the chargino and stau backgrounds
are
much smaller.
The high signal and low SUSY background rate (compared to the signal)
for maximal mixing enables detection of a $
5\sigma$ flavor violating signal
for sleptons masses up to 250 GeV and LSP masses $m_{\chi^0 _1} <
0.4 m_{\tilde{e}_R}$ with an integrated luminosity of
$100$ fb$^{-1}$.
There are some objections to the generality
of this result, though. A more general spectrum could
result in a larger chargino or stau background. For example,
there is no reason to expect the
chargino production cross--section to be related to the slepton
production cross-section. However,
as mentioned above,
the kinematics of slepton
production and decay are different enough from that of
the chargino background
that an appropriate set of kinematic cuts
could distinguish the two, at least for supergravity motivated
parameter choices with comparable
squark and gluino masses \cite{baer}.
Next, the stau background
is sensitive to the stau mass, which is likely to differ from the
selectron and smuon masses
\footnote{The rare decays
$\tau \rightarrow e \gamma$, $\tau \rightarrow \mu \gamma$ and
$\mu \rightarrow e \gamma$ allow for $O(1)$ splitting between the
third and first two generation scalars for $CKM-$like
mixing angles.}.
The stau background has softer leptons, so
a cut on the $p_T$ of the leptons may help distinguish this background from
the signal. The success of this
may require large statistics and knowledge of the stau
production cross--section.
Thus, in general, the SUSY background may not be small.
Next, detection of flavor violation for smaller mixing angles is discussed.
Since the signal is
$\propto \sin^2 \theta$, it is significantly
smaller for say Cabibbo-like mixing angles. In this case, it is crucial
to know the SUSY background more precisely
since it is comparable to the
signal (assuming similar cross sections for sleptons and charginos).
While the quantity $A_F$ ($>
0$ for non-maximal mixing)
is, up to statistical fluctuations,
background--free as far as slepton {\em detection} is concerned, it
is not useful for providing evidence
for slepton {\em flavor violation} since
the chargino background
would need to be determined first.
This is
because the deviations in the values of $A_F$ and $N(e \mu)$
from the SM
for a non-zero mixing angle could
be reproduced, in the case of zero mixing angle, with
a lower slepton production cross--section
and a higher chargino production cross--section.
Even if the SUSY background can be reduced sufficiently by an
appropriate set of cuts,
since the signal is suppressed by the small mixing angle (there will
also be a reduction of the signal due to these cuts),
it may not possible to probe Cabibbo-like mixing angles. For example,
in the case of no mixing,
Table 4 of reference
\cite{krasnikov1} gives 195 dilepton signal events
for the set of cuts labeled 1
with
$L=10$fb$^{-1}$ and a slepton mass of $100$
GeV. The number of signal events in the case of
mixing for $L=100$fb$^{-1}$
is then
$1950 \times
2\times \sin^2 \theta \cos ^2 \theta$ (assuming $x \sim 1$).
The SM background from $WW$ production is 9920
for the same set of cuts. Thus a $5\sigma$ signal (requiring
$S/\sqrt{B} >5$) is possible only for $\sin \theta \stackrel{>}{\sim}
0.4$.
Since this signal was obtained for a 24 GeV LSP, only larger
angles will be probed for larger LSP masses (since the leptons
will be softer in that case). For sleptons heavier
than 100 GeV the prospects for detecting small mixing
angles are clearly worse.
Thus, in the situation where
the SUSY background is {\em known} to be small, {\em e.g.} if an
appropriate set of cuts for a more general spectrum
can separate the chargino background
from the signal, then the
flavor violating dilepton events from Drell--Yan production of
sleptons is a promising signal for the
detection of flavor violation in the case of {\em large} mixing angles.
Otherwise, it is important to look for
other discovery channels for slepton flavor violation.
\section{Slepton Production in Cascade Decays}
The other way to produce sleptons is through the cascade decays
of gluinos and squarks. At the LHC,
the production cross sections of squarks and gluinos
are much
larger than the Drell-Yan production of sleptons,
neutralinos, and charginos.
So, a larger production of sleptons (if they are light)
is expected in the cascade decays than
from direct Drell-Yan production.
In a generic SUSY event, the production of
two real (or virtual from gluino
decay) squarks will be followed by their cascade decays ultimately
to the LSP through intermediate electroweak sparticles
(sleptons, charginos, neutralinos).
Assuming for simplicity
that the spectrum is gaugino-like,
$i.e.$,
$\chi_2^0 \approx \tilde{W}_3$, $\chi_1^+ \approx \tilde{W}^+$ and
$\chi_1^0 \approx \tilde{B}$, the following squark decays
are obtained:
\begin{eqnarray}
BR ( \tilde{q}_R \rightarrow q \chi_1^0 ) & \approx & 1 ,\nonumber \\
BR ( \tilde{q}_L \rightarrow q \chi_2^0 ) & \approx &
\frac{1}{3}, \nonumber \\
BR ( \tilde{q}_L \rightarrow q^{\prime} \chi_1^{+,-}) & \approx &
\frac{2}{3}.
\end{eqnarray}
Thus, a typical SUSY event is:
\begin{eqnarray}
p p & \rightarrow & \tilde{g} \tilde{g}, \; \tilde{g} \tilde{q}
\rightarrow
\tilde{q} \tilde{q} \nonumber \\
& \rightarrow & \chi_{EW} \chi'_{EW} \; + \; X,
\end{eqnarray}
with $\chi_{EW}$, $\chi'_{EW}$ one of $\chi^0_{1,2}$, $\chi^{+,-}_1$.
\subsection{Dilepton Events}
\label{cascade2l}
If
one of the squarks decays to $\chi_2^0$ followed
by the decay of $\chi_2^0$ to a slepton (if
$BR( \chi_2^0 \rightarrow \tilde{l} l)$ is significant)
a large number of
$e \mu$ events in the presence of
lepton flavor mixing (see Eqns.(\ref{chi2a}) and (\ref{chi2b}))
is obtained.
These events also have
at least $2$ high $p_T$ jets and large
$ \not \! p_T$.
There is no background from $W^+W^-$ production since this background
contains no hard jets (assuming jet detection is good).
There is a background from
$t \bar{t}$ production followed by leptonic decays
of the $W$'s from the top quarks. This can be reduced by
rejecting events with $b$-jets
or using a high
$\not \! p_T$ cut.
There is a SUSY background from
the decays of both
squarks to charginos,
followed by chargino decays
to $W^+$, $W^-$ or $\tilde{l}$, $\tilde{l}^{\ast}$. This background is
distinguishable from the signal though.
The invariant mass distribution of the 2 leptons from the $\chi_2^0$ decay
has a sharp edge (which is a
function of the neutralino and slepton masses) \cite{snowmass,ian}
unlike the case of
the $2$ leptons from $\chi^+ \chi^-$ decays.
Also,
the angle between the 2 leptons from the decay of $\chi_2^0$ is likely to
be smaller than in the case of 2 leptons from $\chi^+$ and $\chi^-$.
Such kinematic cuts on the invariant mass of the dileptons and
the angle between them
easily reduce the number of background events
sufficiently if we are interested in
detecting flavor {\it conserving} dileptons from
$\chi_2^0$ decays.
But, in the case of the flavor {\it violating} dilepton events,
(as in Section \ref{dy}) since the signal
is suppressed by the mixing angle (while the background is the same),
the number of background events that survive (relative to the signal)
after cuts
depends crucially on
the model--dependent cross sections for producing
$\chi^+ \chi^-$ vs. $\chi_2^0$
\footnote{
For example, the ratio of the number of events with
$\chi^+ \chi^-$ to those with
(at least) $1 \; \chi_2^0$
is larger for
$s$-channel $\tilde{q} \tilde{q}^{\ast}$
production than for gluino pair production which is seen as follows.
For the $\tilde{g} \tilde{g}$ case, the probability of getting
$2 \; \tilde{q}_L$ is $1/4$ compared to a probability of
$3/4$ for getting at least one $\tilde{q}_L$
whereas for $s$-channel $\tilde{q} \tilde{q}^{\ast}$ production the
probabilities are the same. {\em Same} sign
chargino events are also obtained from $\tilde{g} \tilde{g}$
production whereas $s$-channel $\tilde{q} \tilde{q}^{\ast}$ production
can give only opposite-sign chargino pairs.
Thus, if the $s$-channel $\tilde{q} \tilde{q}^{\ast}$ production
is larger, the number of $\chi^+ \chi^-$ events
relative to $\chi _2 ^0$ events increases.}. So in general
it is difficult to be sure that the cuts have reduced the
background sufficiently.
\footnote{
There is also a
SUSY background from
$\chi_2^0$ decays to $\tilde{\tau} \tau$
followed by leptonic decays of the $\tau$'s. A
cut on the dilepton invariant
mass
can reduce this: the leptons from the $\tau$ decays are softer
than those from the $\tilde{e} / \tilde{\mu} / \chi_2^0$
decays and so have a smaller invariant mass.
But, since, in general,
$BR \left( \chi_2^0 \rightarrow \tilde{\tau} \tau \right)$
is {\em not} related to $BR \left( \chi_2^0 \rightarrow \tilde{e} e \right)$,
as for the chargino background, we cannot be sure that the $\tilde{\tau}
\tau$ background has been sufficiently reduced (by the cuts)
since this
background is unknown.}
In the circumstance that $\chi^+ \chi^-$ are
dominantly produced from $\tilde{g} \tilde{g}$
cascade decays,
the $\chi^+ \chi^-$ flavor violating
background can be estimated as follows.
An equal number of like-sign and
unlike-sign chargino pairs are expected
since $\tilde{g}$ is a Majorana particle.
The like-sign chargino pairs produce like-sign dileptons so that
the opposite-sign chargino $e \mu$ background
can be estimated from the number of like-sign $ee$ and
$\mu \mu$ events.
Unfortunately, in the more general case
the $\chi^+ \chi^+$ and $\chi^+ \chi^-$
production cross sections are
not related since the chargino pairs do not
always come from gluino pair decays.
For example,
$p p \rightarrow \tilde{q}_L \tilde{q}^{\ast}_L$ can lead to $\chi^+
\chi^-$, but
not to $\chi^+ \chi^+$.
It might be possible to estimate the $\chi^+ \chi^-$
background by analysing the (observed) (signal $+$ background)
distribution of the invariant mass of
the flavor violating dileptons \cite{paige}. As
mentioned earlier, the dilepton invariant mass distribution for
$\chi_2^0$ decay has a
sharp edge unlike the case of
the background. The position of this
edge (denoted by $M_{ll}$)
can be easily found by looking at the distribution
of the invariant mass of flavor
{\em conserving} dileptons (where the
$\chi^+ \chi^-$
background is very small) \cite{snowmass, ian}.
The existence of
an
edge in the (observed) {\em opposite}--flavor dilepton distribution
{\em at $M_{ll}$} would then be
indication of flavor violation. However, since the flavor violating dilepton
signal is suppressed by (small) mixing relative to the
flavor conserving dilepton signal (whereas the $\chi^+ \chi^-$
background is the same for both kinds of dileptons), the edge at
$M_{ll}$ in the opposite flavor
dilepton case might not be as sharp as for the same flavor dilepton
case -- this depends
on the model-dependent cross sections for producing $\chi^+ \chi^-$ vs.
$\chi_2^0$.
Next, in the distribution of the invariant mass of
the flavor {\em violating} dileptons, the events beyond
$M_{ll}$
(this value can be obtained from the same flavor dilepton distribution if
the edge is not so sharp in the opposite--flavor
dilepton distribution)
are mostly from the $\chi^+ \chi^-$
background \cite{paige}.
Extrapolating (assuming say a flat distribution for the
$\chi^+ \chi^-$
background) from the data in this region,
the
$\chi^+ \chi^-$
background in the region
with invariant mass less than $M_{ll}$
can be estimated. An excess of $e \mu$ events
(with invariant mass between zero and $M_{ll}$)
over this estimate
will be a signal for flavor violation.
\footnote{The invariant mass
of the leptons from the $\tilde{\tau} \tau$ decays (from $\chi_2^0$) is less
than $M_{ll}$ and so this background cannot be estimated this way.}
This extrapolation may
not be reliable
for invariant masses much smaller than $M_{ll}$
since the
distribution of the $\chi^+ \chi^-$
background in this region is not known.
A detailed simulation is required
to know this distribution (it is known only
that it does not have an edge at $M_{ll}$).
Near $M_{ll}$ the extrapolation should
be more
reliable and that is the region where the signal is peaked (since
the flavor violating dileptons from $\chi_2^0$ decay also have
a sharp edge at $M_{ll}$). An excess in this region
(rather than the whole region between zero mass and $M_{ll}$)
might thus
be a better signal for flavor violation \cite{paige} -- as
mentioned earlier, the
distribution will have a edge (or a ``step'') at $M_{ll}$.
Also, the $\tilde{\tau} \tau$ background in the region near $M_{ll}$
is negligible since the leptons from these decays are softer \cite{paige}.
However,
statistics
are larger if the region from zero mass to $M_{ll}$ is used.
The chargino background can also be eliminated in
considering a flavor violating {\em and} $CP$ violating
dilepton signal \cite{nima2}. The presence of non--trivial phases in the
slepton mixing matrix $W$ breaks $CP$, and results in a
non-vanishing asymmetry:
$N( e^+ \mu^- -
e^- \mu^+) \neq 0$.
In this case, the $\chi^+ \chi^-$ background
is not important since it is CP symmetric.
To summarize, if the
{\em number}
of $e \mu$
events (that pass certain
cuts) from either Drell-Yan or cascade production is used
to detect flavor violation, the SUSY
background from $\chi^+ \chi^-$ pairs (which passes the same cuts)
is difficult to estimate, in general, and may be too large.
The possibility of using the {\em observed}
opposite--flavor dilepton mass {\em
distribution} (in the case of cascade decays) to estimate the
chargino background
is interesting, though, and warrants further study \cite{paige}.
\subsection{Events
with
$4$ leptons}
\label{cascade4l}
A dramatic flavor violating signal is obtained through the
pair production of {\it two} $\chi^0 _2$s, followed by the
decays of {\em both} $\chi_2^0$s to slepton and lepton pairs.
Such an event contains 4 leptons and occurs
if both squarks in a SUSY event decay into $\chi_2^0$.
If one of the $\chi_2^0$s has a
flavor violating decay: $\chi_2^0 \rightarrow \tilde{l} l \rightarrow
e \mu$, then events containing
$3 e \; 1 \mu$, or $3 \mu \; 1 e$ will be produced. A typical
decay chain is then:
\begin{eqnarray}
\tilde{q}_L \tilde{q}' _L
&\rightarrow& \chi^0 _2 q + \chi^0 _2 q' \nonumber \\
\chi^0_2 &\rightarrow& \tilde{l} l \rightarrow
e^+ e^- \chi^0 _1 \nonumber \\
\chi^0_2 &\rightarrow& \tilde{l}' l' \rightarrow
\mu ^+ e^- \chi^0 _1 \;.
\end{eqnarray}
These events are identified by 4 isolated leptons (with the 3+1 flavor
structure), at least
$2$ high
$p_T$ jets, $ \not \! p_T$, and concentrating on only those
events produced from the decays of first two generation squarks,
no $b-$jets. These events will be referred
to as ``flavor violating'' 4 lepton events.
The absence of $b-$jets is important in distinguishing the signal
from other SUSY and SM backgrounds (see below).
The backgrounds to these events arise from both SM and SUSY
sources.
The dominant SM
background occurs from $t \bar{t}$ production
with semileptonic decays of
the $b$s (or $t \bar{t} \gamma$
production with $2$ leptons
from $\gamma$)
and leptonic decays of the $W$s. In this case, however,
the leptons from $b$ decays
will not be isolated (or
the invariant mass of $2$ of the leptons will be zero
in the case of $t \bar{t} \gamma$). Also,
these events have $2$ $b$ quarks and
can be rejected using $b$-jet veto. Double gauge
boson production can give 4 lepton events, but none of these
events have the 3+1 flavor structure.
Triple gauge boson production ($WWZ$ or $WW\gamma$)
can give events with
$4$ leptons and the correct flavor asymmetry,
but some initial state gluon radiation is
needed to give the
$2$ hard jets. The production cross-section for such events
is small. Also,
events of this kind can also be rejected since the invariant mass
of $2$ of the leptons will either be zero
or $m_Z$.
One important obstacle in identifying flavor-violating {\it dilepton}
events was the potentially large background from $\chi^+ \chi^-$
production. In the $4$ lepton signal, however, there is no $\chi^+ \chi^-$
background
from the squark decays since this gives only $2$ leptons.
The weak decay
$\tilde{q} \rightarrow W \tilde{q}'$, if kinematically allowed,
can lead to a possible background. For example, the process
\begin{eqnarray}
\tilde{q}_L \tilde{q}'_L &\rightarrow& W^-
\tilde{q}'' _L + \chi^+ _1 q
\nonumber \\
W^- &\rightarrow& e^- \bar{\nu} \nonumber \\
\tilde{q}'' _L &\rightarrow& \chi^0_2 q ''
\rightarrow \mu^{+} \mu^{-} + \cdots
\nonumber \\
\chi^+ _1 &\rightarrow& \mu^+ \nu \chi^0_1
\end{eqnarray}
is a potential background. For the first two generation squarks, however,
the decay $\tilde{q} \rightarrow W \tilde{q}'$ is kinematically forbidden.
This is because the
mass splitting in an electroweak doublet occurs from the electroweak
$D-$terms
and is less than
$ m^2_W /m_{\tilde{q}} < m_W$. This process is allowed for
the top and bottom squarks, but such an event contains 2 $b$-jets
and this background can be reduced with a $b$-jet veto.
There is a
SUSY background to the flavor violating
$4$ lepton events
from production of heavier neutralinos or chargino
in the cascade decays of
squarks.
For example,
\begin{eqnarray}
\tilde{q}_L \tilde{q} & \rightarrow & \chi^0_3 \chi_2^0 + \cdots,
\nonumber \\
\chi_3^0 \rightarrow W^+ \chi^- & \rightarrow & e \mu + \cdots, \nonumber \\
\chi_2^0 \rightarrow \tilde{l} l & \rightarrow & ee \;
(\hbox{or} \; \mu \mu) \chi^0_1.
\end{eqnarray}
This background is small in the so-called
gaugino-like region. In this region
there is very little gaugino-Higgsino mixing. Then,
the heavier chargino and the two heaviest
neutralinos are dominantly Higgsinos and the two lightest neutralinos
and the lighter chargino are mainly gauginos; this turns out to be
typical of the
SUSY parameter space still allowed by experimental data.
Thus, the decays of the
first two generation squarks into the heavier
neutralinos or chargino
are highly suppressed by the first two generation
Yukawa couplings,
small gaugino-Higgsino
mixing,
and also by phase space.
Another potentially large background can also occur from the production of
the {\em heavier} sleptons (say, the left-handed) and/or sneutrinos.
Sleptons
can decay
to $\chi^0 _2 l$ and $\tilde{\nu}$ to
$\chi^{\pm}_1 l$ if kinematically
allowed.
If the neutralino and chargino decay to leptons, then
this decay chain can give $3$ (or $2$) leptons. With $1$ (or $2$) leptons
from another decay of this kind (or some other decay chain), this
can mimic
the flavor violating $4$-lepton signal.
If
the left-handed sleptons are paired produced through the Drell-Yan
mechanism, then these events do not contain any hard jets and may be rejected.
Thus, the only source for a background from heavier sleptons is
their
production in the decays
of gluinos and squarks. Such a decay does not occur directly, but only
through the decays of gluinos and squarks to the {\em heavier} neutralino
and chargino. The {\em heavier} neutralinos and chargino can then decay
to the left-handed sleptons. As argued in the previous paragraph though,
in the gaugino-like region, the heavier neutralinos/chargino are dominantly
Higgsinos so that their decays to the sleptons are suppressed
by the lepton Yukawa couplings and small gaugino/Higgsino mass
mixing angles. So
this background is negligible.
However, top squarks
(and bottom squarks
for large $\tan \beta$) will have significant decay branching fractions
into heavier neutralinos or chargino even if they are
purely Higgsinos since the third generation Yukawa couplings (and hence
couplings of the squarks to Higgsinos) are large. Further, as mentioned
earlier,
$W$s may be produced in the direct decay of stops or sbottoms.
Also,
top {\em quarks} from stop or sbottom decays produce $W$s. Both of these
processes give
additional isolated leptons. This leads to a potential background
even if stops or sbottoms
decay only to the lighter chargino and neutralinos. For example,
the following decay chain is a possible background:
\begin{eqnarray}
\tilde{t} \tilde{t}^{\ast} & \rightarrow & b \chi^- +
t \chi_2^0, \nonumber \\
t & \rightarrow & W^+ b \rightarrow e^+ \; b + \cdots,
\nonumber \\
\chi^- & \rightarrow & W^- \chi^0_1 \rightarrow \mu^- + \cdots,
\nonumber \\
\chi_2^0 & \rightarrow & e^+ e^- \chi_1^0.
\label{stopdecay}
\end{eqnarray}
These backgrounds to flavor violating
$4$ lepton events can be reduced
by
rejecting any $4$ lepton event that contains at least one $1$ $b$-jet.
Note that the top or bottom squark background
has at least $2$ $b$ quarks.
The efficiency for rejecting this
background is discussed in a later section where a specific
spectrum is considered.
There is also an important
SUSY background from decays of taus and staus
produced from the decays of two $\chi _2^0$s. That is,
\begin{eqnarray}
\chi _2^0 & \rightarrow & \tilde{\tau} \tau \rightarrow e \mu
\chi^0_1 + ...,
\nonumber \\
\chi _2^0 & \rightarrow & \tilde{l} l \rightarrow ee \; (\hbox{or}
\; \mu \mu) \chi^0_1.
\end{eqnarray}
This background can be estimated/measured as follows.
In the above decay chain, if one $\tau$ decays hadronically instead of
leptonically, the result is
$3 e \; 1 \tau$-jet events.
If a lower bound on the $\tau$-jet detection efficiency is known,
an upper bound on the number of
$3 e \; 1 \mu$ events coming from $\tau$ decays is obtained
by using the number
of $3 e \; 1 \tau$-jet events. An
excess of $3 e \;
1 \mu$ events
over this background is a signature of lepton flavor violation.
Lastly, the following $\chi_2^0$
decay chains can also give flavor violating dileptons:
\begin{eqnarray}
\chi_2^0 & \rightarrow & h \; (\hbox{or}) \; Z \; \chi_1^0 \nonumber \\
h \; (\hbox{or}) \; Z & \rightarrow & \tau \tau \rightarrow e \mu.
\end{eqnarray}
In combination with another $\chi_2^0$ decay to $ee$ or $\mu \mu$,
these decay chains can give flavor violating $4$-lepton events.
In the gaugino region, the decay $\chi_2^0 \rightarrow Z \chi_1^0$ is
suppressed since there is no vertex with $Z$ and $2$ neutral gauginos.
In any case, an {\em effective} $BR \left( \chi_2^0 \rightarrow
\tau \tau \right)$ can be defined to include these two decay chains
in addition to the $\chi_2^0 \rightarrow \tilde{\tau} \tau$ decay.
It will be shown in section \ref{LHCpoint} that this
(in general unknown) BR does not affect
the estimate of the (effective) $\tau$ background
obtained by using the $3 e \; 1 \tau$-jet events.
\subsubsection{A quick estimate of number of $4$ lepton events}
\label{quick}
A typical value for the total SUSY production cross section
(gluinos and squarks) at the LHC is:
\begin{equation}
\sigma_{SUSY} \sim 10 \;
\frac{\pi \alpha_S^2}{\hat{s}} \sim 100 \; \hbox{pb}
\end{equation}
with $\sqrt{\hat{s}} \sim 1$ TeV, $\alpha_S \sim 0.1$ and summed over
colors and generations (the factor of 10).
Assuming that the probability to get a $\tilde{q}_L$
is $1/2$ and $BR ( \tilde{q}_L \rightarrow
\chi_2^0 \; q) = 1/3$, this gives
\begin{equation}
\sigma_{\chi^0_2 \chi^0_2} \sim \sigma_{SUSY} \left( \frac{1}{2} \right)^2
\left( \frac{1}{3} \right)^2 \sim 3 \; \hbox{pb}.
\end{equation}
If $BR ( \chi _2^0 \rightarrow \chi^0 _1 l^+ l^-)
\sim 0.16$ (for each of $l = e$, $\mu$) and
for $\sim$ one year of running at low
luminosity which gives an integrated luminosity of
$L \sim 10$ (fb)$^{-1}$,
the expected number of events is :
\begin{eqnarray}
N \left( \chi_2^0 \chi_2^0 \right) && \sim 30,000, \nonumber \\
N \left( 4 \; l \; \hbox{where} \; l = e, \mu
\right)&=&(2 \times 0.16 )^2
N\left( \chi_2^0 \chi_2^0 \right) \sim \; 3300, \nonumber \\
N \left( 3 l + l' \right) &=&
4 \sin^2 \theta \cos^2 \theta \; x \; N(4 l) \sim \; 550,
\end{eqnarray}
for $\sin \theta \approx 0.2$ and $x\sim 1$.
To be clear,
\begin{equation}
S_{FV} \equiv N(3l+l')=\left(N(e^+ \mu^- \mu^+ \mu^-)+
(+ \leftrightarrow -)\right)+(\mu \leftrightarrow e),
\end{equation}
and
\begin{equation}
N(4l)=N(e^+ e^- e^+ e^-)+N(e^+ e^- \mu^+ \mu^-)
+N(\mu^+ \mu^- \mu^+ \mu^-)
+ S_{FV}.
\label{all4l}
\end{equation}
In the next section this definition for $N(4l)$ is trivially extended
to include
leptons produced by the decay of $\tau$s.
Thus, typically, a large number of $4$ lepton
flavor violating events is expected from the cascade decays of squarks
\footnote{{\em Both} $\chi_2^0$s decaying to flavor violating dileptons
gives $(e^+ \mu^-)(e^+ \mu^-)$ and $(e^+ \mu^-)(\mu^+ e^-)$
events. The latter cannot be distinguished from the events where
one $\chi_2^0$ decays to $e^+e^-$ and the other to $\mu^+ \mu^-$.
The former events can also be used as a signal of flavor violation, but
the number of these events is expected to be very small since they require
both $\chi_2^0$s to decay into flavor violating dileptons.
For simplicity these events were not included in Eqn.(\ref{all4l}).}.
\subsubsection{Detailed estimates at Point $5$ (Point $A$) of LHC studies}
\label{LHCpoint}
One Point of the LHC supersymmetry studies \cite{snowmass,ian} contains
a spectrum that
is favorable for the detection of a flavor-violating 4 lepton signal.
The minimal supergravity input parameters for this point are:
\begin{eqnarray}
m_0 = 100 \hbox{ GeV,} \; & M_{1/2} = 300 \hbox{ GeV, }
\;& A_0 = 300 \hbox{ GeV,}
\nonumber \\
\tan \beta = 2.1 \;,& \; \hbox{sgn}(\mu) = + \;, & \;
m_{top} = 170 \hbox{ GeV.}
\end{eqnarray}
Renormalization group evolution of these input parameters to the weak scale
results in a mass spectrum which is given in Table \ref{spectrum}.
Note that $m_{\chi_2^0} \approx 230 \;
\hbox{GeV} \; > m_{\tilde{l}_R} \approx 160$ GeV so that the decay of
$\chi^0 _2$ into real sleptons is allowed.
\begin{table}
\begin{center}
\begin{tabular}{llllllll} \hline
$\tilde{g}$ & 770 & $\tilde{q}_L$ &685 & $\tilde{q}_R$ & 660 &$h$ & 100
\\ \hline
$\tilde{t}_1$ & 500 & $\tilde{t}_2$ &715 & $\tilde{b}_1$ &635
& $\tilde{b}_2$ &660 \\ \hline
$\tilde{l}_L $ &240 & $\tilde{l}_R$ & 160 & $\chi^{\pm} _1$ &230 &
$\chi^{\pm} _2$ & 500 \\ \hline
$\chi^0 _1$ & 120 & $\chi^0 _2$ & 230 &
$\chi^0 _3$ &480 & $\chi^0 _4$ & 505 \\ \hline
\end{tabular}
\end{center}
\caption{Mass spectrum in GeV at LHC Point \protect\cite{snowmass,ian}.
Here $\tilde{q}= \tilde{u}$, $\tilde{d}$, $\tilde{c}$, $\tilde{s}$, and
$\tilde{l}= \tilde{e}$, $\tilde{\mu}$, $\tilde{\tau}$.}
\label{spectrum}
\end{table}
\begin{table}
\begin{center}
\begin{tabular}{cccccc} \hline
$\tilde{g} \tilde{g}$ & 1750 & $\tilde{g} \tilde{q} \;, \;
\tilde{g} \tilde{q}^{\ast}$ & 8300 &
$\tilde{q} \tilde{q}^{\ast}$ & 2380 \\ \hline
$\tilde{q} \tilde{q}'$ & 2820 &
$\tilde{b} \tilde{b}^{\ast}$ & 300
& $\tilde{t} \tilde{t}^{\ast}$ & 700 \\ \hline
\end{tabular}
\end{center}
\caption{The production cross-sections in fb for different SUSY particles
at the LHC Point \protect\cite{snowmass,ian}. Here all flavors
$\tilde{q}_H=\tilde{u}$, $\tilde{d}$, $\tilde{c}$,
$\tilde{s}$
and $H=L$, $R$ are summed over.}
\label{cross-section}
\end{table}
The production cross-section for SUSY particles is presented in
Table \ref{cross-section}, and is dominated by $\tilde{g} \tilde{q}$
production.
In total
$\sigma_{SUSY} \approx 16$ pb. To estimate the number of
signal and background events, the branching fractions of the sparticles
are needed. These are given in Table \ref{br}. Note that at this Point
$BR(\chi^0 _2 \rightarrow \tilde{l}_i l_i) \sim 0.12$ and is
reduced due to the large branching fraction
$BR(\chi^0 _2 \rightarrow h \chi^0 _1)$.
\begin{table}
\begin{center}
\begin{tabular}{cccccc} \hline
$\tilde{g} \rightarrow q \tilde{q}_L $ & 30 &
$\tilde{g} \rightarrow q \tilde{q}_R $ & 30 &
$\tilde{g} \rightarrow \tilde{t}_1 t $ & 14 \\ \hline
$\tilde{g} \rightarrow \tilde{b}_L b $ & 15 &
$\tilde{g} \rightarrow \tilde{b}_R b $ & 10 &
$\tilde{t}_2 \rightarrow Z_0 \tilde{t}_1$ &26 \\ \hline
$\tilde{t}_2 \rightarrow \chi^0 _4 t$ &21 &
$\tilde{t}_2 \rightarrow \chi^{\pm} _2 b$ &18 &
$\tilde{t}_2 (\tilde{t}_1) \rightarrow \chi^{\pm} _1 b$ &15 (63)
\\ \hline
$\tilde{t}_2 (\tilde{t}_1) \rightarrow \chi^0 _2 t $ & 8 (17) &
$\tilde{t}_2 \rightarrow \chi^0 _3 t $ & 6 &
$\tilde{t}_2 (\tilde{t}_1)\rightarrow \chi^0 _1 t $ & 6 (20) \\ \hline
$\tilde{q}_L \rightarrow q \chi^0 _2 $ & 32 &
$\tilde{q}_L \rightarrow q \chi^{\pm} _1 $ & 64 &
$\tilde{q}_L \rightarrow q \chi^0 _1$ & 1.5 \\ \hline
$\tilde{q}_L \rightarrow q \chi^{\pm} _2$ & 1.5 &
$\tilde{q}_L \rightarrow q \chi^0_4$ & 1 &
$\tilde{q}_R\rightarrow q \chi^0 _1$ & 99 \\ \hline
$\chi^0 _2 \rightarrow \tilde{l}_R l $ & 36 &
$\chi^0 _2 \rightarrow h \chi^0 _1$ & 63 &
$\tilde{l}_R \rightarrow l \chi^0 _1 $ &100 \\ \hline
$\tilde{l}_L \rightarrow \chi^0 _1 e $ & 90 &
$\chi^+_1 \rightarrow W^+ \chi_1^0 $ & 98 &
$h \rightarrow \tau \tau$ & 5 \\ \hline
\end{tabular}
\end{center}
\caption{Branching fractions (in percent) for sparticles at LHC Point
\protect\cite{ian}.
Here $\tilde{q}=\tilde{u}$, $\tilde{d}$, $\tilde{c}$,
$\tilde{s}$, and $\tilde{l} =\tilde{e}$, $\tilde{\mu}$, $\tilde{\tau}$.}
\label{br}
\end{table}
This gives from
decays of first two generation
squarks
the number of $\chi^0 _2$ pairs {\it produced}:
\begin{equation}
N \left( \chi_2^0 \chi_2^0 \right) =(0.32)^2 \times
(\sigma_{\tilde{g} \tilde{g}}
\times(0.3)^2 + \frac{1}{2}
\sigma_{\tilde{g} \tilde{q}} \times 0.3
+\frac{1}{4}
\sigma_{\tilde{q} \tilde{q}'}
+\frac{1}{2}\sigma_{\tilde{q}^{\ast} \tilde{q}}) L
\approx 3400.
\end{equation}
(The factors of $1/2$ and $1/4$ are easy to understand: $1/2$ of all
$\tilde{q} \tilde{q}^{\ast}$ produced from s-channel gluon and
4--point contact interaction \footnote{It is assumed that
all of the $\tilde{q} \tilde{q}^{\ast}$ production is by this channel.
This is reasonable since most of the hard collisions at LHC energies are
likely to be gluon-gluon.},
and $1/4$ of all
$\tilde{q} \tilde{q}'$s produced (from t-channel gluino exchange)
are left-handed pairs.)
This is for one year of running at low luminosity ($L =
10$ fb$^{-1}$) and for one
detector. Hereafter estimates of event numbers will use this
integrated luminosity.
A realistic detection
efficiency of
$90 \%$ for single $e$, $\mu$,
and
$90 \%$ for the single-prong decay $\tau \rightarrow \pi \nu \;
(BR \approx 0.11)$
will be used. These are needed to determine the
number of 4-lepton and 3-lepton $+ \tau$--jet events that are detected.
Later, a comment on a more realistic $\tau$-jet
detection will be made.
Next the 4--lepton signal and background are estimated.
Due to the decay chain
\begin{eqnarray}
\chi_2^0 & \rightarrow & h \chi_1^0 \nonumber \\
h & \rightarrow & \tau \tau
\end{eqnarray}
the {\em effective} $BR \left( \chi_2^0 \rightarrow \tau \tau
\chi^0_1 \right)
\equiv R_{\tau}$ is
\begin{eqnarray}
R_{\tau} & = & BR \left( \chi_2^0 \rightarrow \tilde{\tau}_R \tau \right)
+ BR \left( \chi_2^0 \rightarrow h \chi_1^0 \right)
\times BR \left( h \rightarrow \tau \tau \right) \nonumber \\
& = & 0.15.
\end{eqnarray}
Using the above BR and
$BR \left( \chi_2^0 \rightarrow \tilde{l}_R l \right) = 0.12$
for each of $l = e$, $\mu$ and
$BR \left( \tau \rightarrow e \nu\right) \approx BR \left(
\tau \rightarrow \mu \nu \right)
\approx 1/2 \times 0.35$,
\begin{eqnarray}
BR \left( \chi_2^0 \rightarrow e e \chi^0 _1,\mu \mu \chi^0 _1 \right)
&=&
2 \times 0.12 \times \left( 1 - 2 \sin^2 \theta \cos ^2 \theta x \right)
\nonumber \\
& &+ R_{\tau}
\times (0.35)^2 \times \frac{1}{2}, \nonumber \\
BR \left( \chi_2^0 \rightarrow e \mu \chi^0 _1 \right) &=&
2 \times 0.12 \times 2 \sin^2 \theta \cos ^2 \theta x +
R_{\tau} \times (0.35)^2 \times \frac{1}{2},
\end{eqnarray}
where the first terms in each equation are from decays of
$\tilde{e}$ and $\tilde{\mu}$ and the second terms are from
$\tau$ decays.
Then,
the total number of $4$-lepton events expected from $\chi_2^0$
pair decays (including detection efficiencies, but parameterizing
the acceptance cut as $\varepsilon_{CUT}$
-- see later \footnote{Detection efficiency
refers to the probability that the lepton (or $\tau$-jet in a later case)
will be detected {\em given} that it passes the acceptance cuts. })
is
\begin{eqnarray}
N(4l) & = & N \left( \chi_2^0 \chi^0_2 \right) \times
\left(BR \left( \chi_2^0 \rightarrow e e ,\mu \mu , e \mu \right) \right) ^2
(0.9)^4 \times \varepsilon_{CUT} \nonumber \\
& = & 3400 \times \left( 0.24 + R_{\tau}
\times (0.35)^2 \right)^2
\times (0.9)^4 \times \varepsilon_{CUT} \nonumber \\
& \approx & 149 \times \varepsilon_{CUT}.
\label{4l}
\end{eqnarray}
To get $3e \; 1 \mu + 3 \mu \; 1 e$ events, one $\chi_2^0$ has to decay
into $e e / \mu \mu$ and the other to $e \mu$.
Thus, the number of $3e \; 1 \mu + 3 \mu \; 1 e$
events from flavor-mixing for $\sin \theta = 0.2$ and $x \sim 1$ (it is
shown later that these values are consistent with the $\mu \rightarrow
e \gamma$
limit)
is
\begin{eqnarray}
S_{FV} & = & N \left( \chi_2^0 \chi^0_2 \right)
\times BR
\left( \chi_2^0 \rightarrow e e \chi^0 _1, \mu \mu \chi^0 _1 \right)
\times (0.9)^4 \times \varepsilon_{CUT} \nonumber \\
& & \times 2 \times 0.24
\times
2 \sin^2 \theta \cos ^2 \theta x
\nonumber \\
& \approx & 20 \times \varepsilon_{CUT} .
\label{mixing}
\end{eqnarray}
There is an extra factor of $2$ since either $\chi_2^0$ can
decay to flavor violating dileptons. Next, the
number of $3e \; 1 \mu + 3 \mu \; 1 e$ events from
leptonic decays of $\tau$s produced from
$\chi_2^0$ is
\begin{eqnarray}
B_{FV} & = & N \left( \chi_2^0 \chi^0_2 \right) \times 2 \times
\left( R_{\tau} \times (0.35)^2 \times \frac{1}{2} \right)
\nonumber \\
& &\times
BR
\left( \chi_2^0 \rightarrow e e \chi^0 _1, \mu \mu \chi^0 _1 \right)
\times (0.9)^4 \times \varepsilon_{4l} \times \varepsilon_{CUT} \nonumber \\
& \approx & 9 \times \varepsilon_{CUT} .
\label{tau4l}
\end{eqnarray}
Here, $\varepsilon_{4l}$ is the acceptance for $4$ leptons
with $2$ of them coming from the decay chain $\chi_2^0
\rightarrow \tau \tau$ {\em relative} to that
for all $4$ leptons coming from $\chi_2^0 \rightarrow \tilde{e} e$
or $\mu \tilde{\mu}$.
Since the leptons from the $\tau$ decay
are softer, it is expected that $\varepsilon_{4l} \stackrel{<}{\sim} 1$.
\footnote{Strictly speaking, the factor
$\varepsilon_{4l}$ should be included in determining
$N(4l)$ and $S_{FV}$ as well. But since the number of events in these samples
from $\tau$ decays is very
small, it is a good approximation to
assume $\varepsilon_{4l} \approx
1$ in those numbers.}
To get the number in the last line above, $\varepsilon_{4l} \approx
1$ has been assumed.
Finally, the above $2$ numbers are from the leptonic decays
of $2$
$\chi_2^0$s from the decays of first two generation
squarks only. As mentioned before,
stop/sbottom decays to $W$, $\chi_3^0$ etc.
can give a background
to the flavor violating $4$-lepton signal (see Eqn.(\ref{stopdecay})).
To reject these events, a $b$-jet veto is used. This implies that
events with $4$ leptons coming from
$2$ $\chi_2^0$ decays with (at least) one $\chi_2^0$
coming from a stop/sbottom decay will also be rejected; this is the reason
for not including the $\chi_2^0$ pairs from stop/sbottom decays in
the numbers above.
Measuring the background
from $\chi_2^0 \rightarrow \tau \tau$ decays is discussed next.
As mentioned earlier, the idea
is to measure the number of
$(3 e \; \tau -\hbox{jet})+ \; (2 e \; 1 \mu \;\tau -\hbox{jet})+ \cdots$
events where $\tau$-jet refers to the
hadronic decay of $\tau$. At this
LHC Point
the number of these events (including detection efficiencies)
is
\begin{eqnarray}
N(3l + \tau-\hbox{jet}) & = & N \left( \chi_2^0 \chi^0_2 \right) \times
2 \times \left( R_{\tau} \times 2 \times
0.35 \times \varepsilon_{\tau} \right) \times \varepsilon_{CUT}
\times \nonumber \\
& & BR
\left( \chi_2^0 \rightarrow e e \chi^0 _1, \mu \mu \chi^0 _1,
e \mu \chi^0 _1 \right) \times (0.9)^3 \times \varepsilon_{3l}.
\label{taujet}
\end{eqnarray}
A factor of $2$ is due to {\em either} $\tau$ decaying to a jet.
Here, $\varepsilon_{\tau}$ includes
$BR
\left( \tau \rightarrow \hbox{hadron} \right)$ {\em and} the
efficiency for detecting
a hadronic decay of $\tau$. The variable $\varepsilon_{3l}$ is
the acceptance
for $(3 + 1 \tau-\hbox{jet})$ {\em relative}
to that for $4$ leptons all of which come from the decay chain $\chi_2^0
\rightarrow \tilde{e} e$, $\tilde{\mu} \mu$. It is expected
that $1 \stackrel{>}{\sim}
\varepsilon_{3l} \stackrel{>}{\sim} \varepsilon_{4l}$
since the lepton from the $\tau$ decay is softer than the $\tau$-jet
and since the $\tau$ decay products
(both lepton and jet) are softer than the leptons from the decay chain
$\chi_2^0
\rightarrow \tilde{e} e$, $\tilde{\mu} \mu$.
From Eqns.(\ref{tau4l}) and (\ref{taujet})
and assuming $BR \left( \chi_2^0 \rightarrow e e \chi^0 _1, \mu \mu
\chi^0 _1 \right)
\approx BR \left( \chi_2^0 \rightarrow e e \chi^0 _1, \mu \mu \chi^0 _1,
e \mu \chi^0 _1 \right)$,
the following relation is
obtained
\begin{eqnarray}
B_{FV} & \approx & N(3l + \tau-\hbox{jet})
\times \frac{0.9 \times \frac{0.35}{2} \times
\varepsilon_{4l}}{2 \times \varepsilon_{3l} \times \varepsilon_{\tau}}.
\label{4l3ljetrel}
\end{eqnarray}
Note that $R_{\tau}$ {\em cancels} in the ratio.
Thus,
using
the $\left( 3l + \tau-\hbox{jet} \right)$
detection together with an understanding
of the $\tau$ detection efficiency ($\varepsilon_{\tau}$),
as well
as the acceptance for 4 leptons (with $2$ of them from $\tau$ decays)
versus (3 leptons $+ \tau$-jet) ($\varepsilon_{4l} / \varepsilon_{3l}$),
the number of $(3e\; 1\mu+ 3 \mu \; 1e)$ events
from $\tau$ decay
(Eqn.(\ref{tau4l}))
contained in the full 4-lepton sample can be obtained
from the above relation.
This is important, as it means that
the $\chi^0 _2 \rightarrow \tau \tau$ background
to the flavor-violating signal
can be determined without knowing the relative branching fraction
of $\chi^0 _2$ to $h$,
$\tilde{l}l$, or $\tilde{\tau} \tau$.
Assuming that
the detection efficiency for the decay
$\tau \rightarrow \pi \nu$ (which has a BR of 0.11)
is $0.9$ so that
$\varepsilon_{\tau} \approx 0.9 \times 0.11$,
and assuming $\varepsilon_{3l} \approx 1$ gives
\begin{equation}
N(3l + \tau-\hbox{jet}) \approx 11 \times \varepsilon_{CUT}.
\end{equation}
Independent of this, it is worth remarking that with enough
stastistics it might be possible to measure
$BR(\chi^0 _2 \rightarrow h \chi^0 _1)$,
$BR(\chi^0 _2 \rightarrow \tilde{e} e, \tilde{\mu} \mu)$ and
$BR(\chi^0 _2 \rightarrow \tilde{\tau} \tau)$ {\em assuming} that
these are the dominant decay modes of $\chi_2^0$. The decay chain
$\chi^0 _2 \rightarrow h \chi^0_1 \rightarrow b \bar{b} \chi^0_1$
(where $\chi_2^0$ is from cascade decays of
squarks as usual) gives $b \bar{b}$ events with high
$p_T$ jets and $\not \! p_T$. Comparing these to
the number of dilepton events from $\chi^0 _2$ decays gives
\begin{equation}
\frac{N(b \bar{b})}{N(2l)} \propto
\frac{BR(\chi^0 _2 \rightarrow h \chi^0 _1)}{BR(\chi^0 _2 \rightarrow
\tilde{e} e \chi^0 _1, \tilde{\mu} \mu \chi^0 _1)}.
\end{equation}
Similarly, the number of
$3l+1\tau$--jet events compared to 4--lepton events is
\begin{equation}
\frac{N(3l+\tau- \hbox{jet})}{N(4l)} \propto
\frac{R_{\tau}}
{BR(\chi^0 _2 \rightarrow
\tilde{e} e \chi^0 _1, \tilde{\mu} \mu \chi^0 _1)}.
\end{equation}
All the events in the above two equations have in addition
high $p_T$ jets and $\not \! p_T$ to make sure that these are from
cascade decays of squarks.
From these two measurements and the assumption that $\sum BRs=1$ the
above--mentioned branching ratios can be obtained. This could provide
complementary information to the flavor violating signal discussed here.
Returning to the main subject of this section,
an observation of an
excess
of the `flavor violating'
4--leptons events
over those from $\tau$ decay (Eqn.(\ref{tau4l}))
would be a strong evidence for lepton
flavor violation.
But, before concluding that SUSY lepton
flavor violation has been detected, the background to the
flavor violating $4$ lepton events from stop/sbottom
production (see Eqn.({\ref{stop}) below) must be removed,
and also
the $\tau$-jet detection efficiency $\varepsilon_{\tau}$ must be known.
These two issues are discussed next.
The $\tau$ hadronic decays from $Z \rightarrow \tau \tau$ at the
LHC were
simulated for the ATLAS detector
in \cite{atlas} \footnote{There is also a study of
detecting $\tau$-jets from heavy SUSY Higgs decay for the CMS detector
\cite{cms}.}. This study
shows that
a detection efficiency $\epsilon_{\tau}$ for
a hadronic $\tau$ decay (including the multi-prong decays, {\it i.e,}
a total $\tau$ decay BR of 0.65) of $\approx 40 \times 0.65 \%$
with a rejection factor of 15 for non-$\tau$ jets
can be achieved. This is possible since
$\tau$-jets have lower particle multiplicity, narrower profile and
smaller invariant mass than the QCD jets \cite{atlas}.
A similar detection efficiency (or even better detection
efficiency and rejection of non-$\tau$ jets
if the strategy is optimized for this case)
for $\tau$-jets from sparticle decays could be expected.
The important point about this though is that it
suffices to know a {\em lower} limit on the $\tau$-jet detection efficiency
to get an {\em upper} limit on the number of $\left( 3 e \; 1 \mu \right) +
\left(3 \mu \; 1 e
\right)$ events from tau decays using
the $\left( 3 e \; \tau -\hbox{jet} \right)$
events (see Eqn.(\ref{4l3ljetrel})). Similarly, since
$\varepsilon_{4l} \stackrel{<}{\sim} \varepsilon_{3l}$,
an {\em upper} limit on $B_{FV}$ can be obtained even
though these $\varepsilon$'s may not be known precisely.
Also, if the $\tau$-jet detection (and QCD jet rejection) is good,
there will be large number of events with
$2$ leptons and $2$ $\tau$-jets from $2$ $\chi_2^0$ decays.
These can be used in addition to the $3$ lepton $1 \tau$-jet
events to estimate the background to flavor violating $4$ lepton events
from $\tilde{\tau}/\tau$ decays.
To reduce the stop and sbottom backgrounds a $b$-jet veto
can be used. Before using this veto, the number of
expected $3e \; 1\mu+3 \mu \; 1e$ events from decays of
$\tilde{t}$ or $\tilde{b}$ to $W$, $\chi^0 _3$, $\chi^{\pm}_1$ etc.
(in the absence of any flavor mixing)
can be estimated using the production cross-sections and
branching fractions from Tables \ref{cross-section} and \ref{br}.
The result is, including lepton detection efficiencies:
\begin{equation}
N(\tilde{t} \; \hbox{or} \; \tilde{b})\approx 50 \times \varepsilon_{CUT}.
\label{stop}
\end{equation}
Each of these events has at least
$2$ $b$ quarks. So with a $b$-detection
efficiency of
$60 \%$ (and rejection factor of $200$
against non-$b$ jets at low luminosity
\cite{tp}), the number of $3 e \; 1 \mu + \cdots$ events
from stop/sbottom decays after the $b$-jet veto
goes down to $8$. This can be further reduced
by using a $b$-tagging efficiency of $90 \%$
with a mistag rate of $25 \%$ ({\it i.e.}, rejection factor
of $4$ against non-$b$ jets) at low luminosity \cite{ian,tp}; this
will reduce the signal by a bit. This
strategy can be optimized
depending on the luminosity \cite{tp}.
Lastly, to get actual number of
events, the cuts used to select these events must also
be taken into account. The effect of these
cuts on the signal and background rates is buried in the
fudge factor $\varepsilon_{CUT}$.
For example, $p_T \stackrel{>}{\sim} 10$ GeV and $\mid \eta \mid
\stackrel{<}{\sim} 2.5$ is required
to be able to detect $e$ or $\mu$.
Also, to reduce any
remaining small SM background,
{\it i.e.,} to make sure that these {\em are} SUSY events, various
cuts on
$\not \! p_T$, $p_T$ of jets, a variable $M_{eff}$ \cite{snowmass,ian}
related
to $\not \! \! p_T$, $p_T$ of jets, can be imposed.
Analysis of
the events simulated in
\cite{ian} showed that there were $\sim 40$
events with $4$ leptons with no $b$-jets
that pass all the cuts mentioned
above compared to the estimate of $\sim 149$ from
cross sections and BRs, Eqn.(\ref{4l}): there is an
acceptance factor of $\varepsilon_{CUT} \sim 1/4$
from the various kinematic cuts.
We have also checked that almost all of these
(simulated) events have $2$ $\chi_2^0$s as expected.
\footnote{The information about whether an event in the simulation
has
$\chi_2^0$s,
$\tilde{t}$s, $\chi_3^0$s etc. is from the event generator.}
There are very few events in this sample (from the simulation)
with heavier neutralinos/chargino in agreement with the
expectation from the very small
BRs of the first two generation squarks to these sparticles at
this point in the SUSY parameter space \cite{ian} (see Table \ref{br}).
The number of events (from the simulation)
with at least $1$ $b$ quark and $4$ leptons
is also in rough agreement (up to the acceptance factor) with the
number of $4$ lepton events with
at least $1$ stop/sbottom
expected from the cross section and branching fraction
estimates. \footnote{We have also checked that
these simulated events {\em do}
have at least $1$ stop/sbottom.
There are very few events in this sample with {\em no}
stops/sbottoms but with $b$-jets from initial state
gluon radiation.}
Including an acceptance factor of $\varepsilon_{CUT} \sim 1/4$ for both
background and signal, a
$b-$jet detection efficiency of 60$\%$ (which was not included in
Eqn.(\ref{stop}))
and
detection efficiency of 90$\%$ for
the decay $\tau \rightarrow \pi \nu$, and a 66$\%$ 4-lepton detection
efficiency (the $\tau$ and lepton detection efficiencies were
included in the previous estimates of $S_{FV}$ etc.),
a summary of the expected number of
events at {\it low luminosity} is :
\begin{eqnarray}
N (4l) & \approx & 37 \times \frac{L}{\hbox{10 fb}^{-1}} \nonumber \\
S_{FV} & \approx & 5 \times \frac{\sin^2 2 \theta}{0.15} x \times
\frac{L}{\hbox{10 fb}^{-1}} \nonumber \\
B_{FV} & \approx & 2 \times \frac{L}{\hbox{10 fb}^{-1}} \nonumber \\
N(3l+\tau-\hbox{jet}) & \approx
& 3 \times \frac{L}{\hbox{10 fb}^{-1}} \nonumber \\
N(\tilde{t} \hbox{ or }\tilde{b}) & \approx
& 2 \times \frac{L}{\hbox{10 fb}^{-1}} .
\end{eqnarray}
While these numbers may be a little small for
one detector and
one year of running at low luminosity $(L=10$fb$^{-1}$), there is cause for
optimism.
More {\em integrated} luminosity $L$ from $>1$ year
of running and/or $2$ detectors can significantly increase the statistics.
Further, a
larger $BR(\chi^0_2 \rightarrow
\tilde{l} l)$ would give more statistics.
This could occur at a point in the SUSY parameter space with a heavier
Higgs boson, and thus a lower
$BR(\chi^0_2 \rightarrow h \chi^0_1)$.
To illustrate the discovery or exlcusion significance of these
results, an integrated luminosity of $L=100$ fb$^{-1}$ is
considered.
This could occur for
5 years of running at low luminosity for two detectors
\footnote{One year of running at high luminosity is also possible. In this
case however, the $b-$jet mistag rate increases to 1 in 6 for a
$b-$tagging efficiency of 80$\%$ \cite{tp}. Since most of the signal events
occur from $\tilde{q} \tilde{g}$ production and so contain at least
three
hard jets, approximately 40$\%$ of the signal could be rejected.
In this case the discovery (and exclusion)
limits on $\sin \theta_R$ increase by about
$25\%$. In addition, the tau-jet detection efficiency at high
luminosity is not known since a low luminosity was used in
the ATLAS study.}.
For this integrated luminosity there are 22 4--lepton flavor
violating events from the $\tilde{\tau}/\tau$ background, and
30 3--lepton $\tau-$jet events. There will also be
125 4--lepton flavor
violating events {\it before the $b-$jet veto} from the
$\tilde{t}/ \tilde{b}$ background. Next,
the $b-$tagging efficiency is optimized so that
the $\tilde{t}/ \tilde{b}$ background is (less than or)
equal
to the $1\sigma$ statistical error in $\tilde{\tau}/\tau$ background
while at the same time the reduction of the signal due to mistagging
is small.
This is achieved with a $b-$tagging efficiency of
$80 \%$, rather than the $60 \%$ of before. At this higher tagging
efficiency there is a mistag rate of 1 in 50, so there is very little
reduction of the signal.
With an $80 \%$ $b-$tagging efficiency, 5
$\tilde{t}/ \tilde{b}$ background events remain since each event has
at least 2 $b-$jets. Then the background is dominated by the
$\tilde{\tau}/\tau$ decays.
A $5\sigma$ $(2\sigma)$ discovery
(exclusion) requires that
$S/\sqrt{B} >5 $ $(S/\sqrt{B} > 2)$, and this requires
$>$23 ($>9$) signal events. So a $5\sigma$ discovery is obtained for
\begin{equation}
\sqrt{x} \sin 2 \theta_R > 0.26 \; (5 \sigma \; \hbox{discovery})
\; \hbox{ or }
\sin \theta_R > 0.13 \; \hbox{ for } x\sim 1.
\end{equation}
If no signal is observed then the $2 \sigma$ exclusion limit is
\begin{equation}
\sqrt{x} \sin 2 \theta_R > 0.16 \; (2 \sigma \; \hbox{exclusion})
\; \hbox{ or }
\sin \theta_R > 0.08 \; \hbox{ for } x\sim 1.
\end{equation}
To end this Section, these values of $\sin 2 \theta_R$ and $\Delta m/m$ that
may be probed by the LHC are compared to the constraints on these
parameters obtained from
$\mu \rightarrow
e \gamma$.
The LHC signal is proportional to
$\sin ^2 2 \theta _R \; x$, with $x \sim 1$ if $\Delta m \stackrel{>}{\sim}
\Gamma$ and
$x \ll 1$ if $\Delta m \ll \Gamma$. The decay $\mu \rightarrow
e \gamma$
places an upper limit on
$\sin 2 \theta _R \Delta m /m$ (Eqn.(\ref{muegamma}))
so that there is competition between the two probes of flavor
violation.
Thus, in order for the signal at the LHC to be significant in
the region of the
$\left( \sin 2 \theta _R, \; \Delta m /m \right)$ plane {\em beyond}
the reach of the
$\mu \rightarrow e \gamma$ limit,
there should be a range of $\Delta m /m$
where $\Delta m \stackrel{>}{\sim} \Gamma$
so that $x \sim 1$ {\em and} $\Delta m/m$ is small enough
(for a given value of $\sin 2 \theta_R$)
so that $\mu \rightarrow
e \gamma$ is suppressed.
It will be seen that for
$\Delta m/m \sim \Gamma /m$ (so that
$x \sim 1$),
at this LHC Point,
$\sin 2 \theta_R$ is unconstrained by the $\mu \rightarrow e \gamma$
limit, affording the LHC the opportunity to either detect a signal
or extend the limit.
At this Point $\chi^0 _1 \approx \tilde{B}^0$. A computation
of the one-loop
$\tilde{B}^0$ contribution gives
\begin{equation}
\sin 2 \theta _R \frac{\Delta m^2_R}{\tilde{m}_R ^2}
\left(\frac{\hbox{100 GeV } \tilde{m}_R}{M^2 _{\chi^0 _1}} \right)^2
20 F(\alpha_L, \alpha_R,t) <
0.013 \times
\sqrt{\frac{BR(\mu \rightarrow e \gamma)}{4.9 \times 10^{-11}}}.
\label{ueg}
\end{equation}
Here $\alpha_H=\tilde{m}^2 _H /M^2 _{\chi^0 _1}$ $(H=L,R)$,
$t=(A+\mu \tan \beta )/ \tilde{m}_R$,
\begin{equation}
F(\alpha_L, \alpha_R,t)=H(\alpha_R)+t \alpha^{1/2} _R
\frac{\partial K}{\partial \alpha_R}(\alpha_L,\alpha_R),
\end{equation}
with
\begin{equation}
K(x,y)=\frac{g(x)-g(y)}{x-y}, \; g(x)=\frac{1+2 x \log x- x^2}{2 (x-1)^3},
\end{equation}
and
\begin{equation}
H(x)=\frac{-x^3+9 x^2+9x-17-(6+18 x) \log x}{6 (x-1)^5}.
\end{equation}
Two useful facts are $H(1)=\frac{\partial}{\partial y} K(1,1)=-1/20$;
hence the factor of 20 on the
left side of Eqn.(\ref{ueg}).
At this LHC Point, $m_{\tilde{l}_R} \sim $160 GeV,
$m_{\chi^0_1} \sim$ 120 GeV,
and $m_{\tilde{l}_L} \sim$ 240 GeV.
Inputing these masses into the above formula simplifies it to:
\begin{equation}
\frac{\sin 2 \theta_R}{0.39}
\times \frac{\Delta m_R}{\tilde{m}_R} \times(1 +0.48 t) < 0.03 \times
\sqrt{\frac{BR(\mu \rightarrow e \gamma)}{4.9 \times 10^{-11}}} .
\label{nmue}
\end{equation}
At this Point $t \approx 5-10$. However, a
larger variation in $t$ is allowed without affecting the
flavor violating signal, since both
$A$ and sgn$(\mu)$ do not qualitatively affect
the 4--lepton event rate \footnote{It is important to maintain
the relation $m_{\chi^0 _2}> m_{\tilde{e}_R}$ though.}.
In any case, the values $\sin 2 \theta \approx 0.39$
and
$\Delta m_R \sim \Gamma$
(so that $x \sim 1$) with a typical
value of $\Gamma \sim \alpha_{em} m \sim 10^{-2} \times \tilde{m}_R$
are consistent with
$\mu \rightarrow e \gamma$ -- recall that $\sin
2 \theta \approx 0.39$ and $x \sim 1$ was assumed to
obtain the estimate of $S_{FV}$ in Eqn.(\ref{mixing}).
In fact, at this LHC Point
$\Gamma \sim 125$ MeV \cite{ian}, so that $\Gamma / m \sim 8 \times
10^{-4}$ which is smaller than $\alpha_{em}$.
So for $\Delta m/m \stackrel{>}{\sim}
2 \; \Gamma/ m \approx 1.6 \times 10^{-3}$,
it follows that $x \sim 1$. From Eqn.(\ref{nmue}) and for maximal mixing
($\sin 2 \theta _R=1$),
$\Delta m/ m < 0.39 \times 0.03 /(1+0.48 t) \approx 4 \times 10^{-3}$ (for
$t \approx 5$). Thus
for $1.6 \times 10^{-3} \stackrel{<}{\sim}
\Delta m/ m \stackrel{<}{\sim} 4 \times 10^{-3}$
and $\sin 2\theta_R =1$, $\mu \rightarrow e \gamma$ is satisfied
and $x \sim 1$.
So at this
Point even for maximal mixing
there is a large range of $\Delta m/m$ for which
$x \sim 1$ and $\mu \rightarrow e \gamma$ is safe. Of course, smaller
mixing could be probed by the LHC, in which case the upper
bound on $\Delta m/m$ allowed by $\mu \rightarrow
e \gamma$ is larger. In this case
for a given $\sin \theta _R$ there is a larger range of
$\Delta m /m$
for which $x \sim 1$ (so that there is no
suppression of the LHC signal)
{\em and} $\mu \rightarrow e \gamma$ is
safe.
\section{Conclusions}
We believe that it is possible
to detect SUSY lepton flavor violation
at the LHC using events with $4$
leptons from the cascade decays of squarks provided the following
conditions are satisfied:
0. Either $R-$parity is conserved or $\chi^0 _1$ $(LSP)$ decays
outside the detector,
1. $\chi_2^0$ pair production in cascade decays of squarks is large and
$\chi_2^0$ has a large decay branching fraction
to $\tilde{l}^{\ast} l$ (to get
enough statistics),
2. Hadronic decays of $\tau$s can be detected with a known efficiency
so that the background from the $\chi_2^0 \rightarrow
\tau \tau$ decay can be estimated,
3. The $b$-jet detection
efficiency is good so that the background from
events with stop/sbottom
can be rejected,
4.
The stop/sbottom production rate, either
direct or in gluino decays, is not {\em too much} larger than the
production of
first two generation squarks,
5. The first two generation squarks
decay largely to $\chi_2^0$, $\chi_1^0$ and
$\chi_1^{\pm}$, so that the background to flavor violating $4$
lepton events from decays of heavier neutralinos/chargino
to $W$s, lighter chargino, {\em heavier}
sleptons etc. is small.
This condition can be realised in the so-called gaugino-like region,
6. The mass splitting is $\Delta m \sim \Gamma$ or {\em larger},
so there is no
suppression of the signal due to the quantum interference effect.
The arguments presented here are clearly semi-quantitative, and
further study requiring a detailed
simulation of these processes is required.
This is beyond the scope of this work.
\section{Acknowledgements}
We thank Nima Arkani-Hamed and Mahiko Suzuki
for discussions, and Ian Hinchliffe in particular for discussions
and allowing us to use
the simulations
of Point 5 of
the LHC study. This work was
supported in part by the U.S. Department of Energy under Contract
DE-AC03-76SF00098.
KA thanks
Frank Paige for
discussions. MG thanks the Natural Sciences and Engineering Research
Council of Canada for their support.
|
\section{Weighted projective spaces}
We will be working with weighted projective spaces, which are certain
(singular) quotients of usual projective space. Alternatively, they may be
described as quotients of $\fC^{n+1}$ by a $\fC^*$-action. We assume the
weights $(w_0,\ldots, w_n)$ are given, let ${\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}_{w_i}$ denote the group
of $w_i$th roots of unity, and consider the action of ${\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}:=
{\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}_{w_1}\times \cdots \times {\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}_{w_n}$ on $\fP^n$ as follows. Let
$g=(g_0,\ldots, g_n) \in {\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}$, and consider for $(z_0:\ldots: z_n)$
homogenous coordinates on $\fP^n$ the action
\[ (g,(z_0:\ldots: z_n)) \mapsto (g_0z_0:\ldots: g_nz_n).\]
Alternatively, consider the action of $\fC^*$ on $\fC^{n+1}$ given by
\[ (t,(z_0,\ldots, z_n)) \mapsto (t^{w_0}z_0,\ldots, t^{w_n}z_n).\]
In both cases, the resulting quotient is the weighted projective space,
which we will denote by $\fP_{(w_0,\ldots, w_n)}$. General references for
weighted projective spaces are \cite{dolg} and \cite{Y}. A {\it weighted
hypersurface} is the zero locus of a weighted homogenous polynomial $p$.
We will assume the weights are {\it normalized} in the sense that no $n$ of
the $n+1$ weights have a common divisor $>1$. Both for the weighted
projective spaces as well as for the weighted hypersurfaces this assumption
is no restriction (cf.\ \cite{dolg} 1.3.1 and \cite{Y}, pp.\ 185-186). We
will write such isomorphisms in the sequel without further comment, for
example $\fP_{(2,3,6)}\cong \fP_{(2,1,2)}\cong \fP_{(1,1,1)}=\fP^2$, where
the first equality is because the last two weights are divisible by 3, the
second while the first and last are divisible by 2.
We will use the notation $\fP_{(w_0,\ldots, w_n)}[d]$ to denote either a
certain weighted hypersurface of degree $d$, or to denote the whole family
of such (the context will make the usage clear). In the particular case
that the weighted polynomial $p$ is of Fermat type, then there is a useful
fact, corresponding to the above normalizations. For example, in
$\fP_{(2,3,6)}$ consider the weighted hypersurface $x_0^6+x_1^4 +x_2^2=0$.
Then the isomorphism $\fP_{(2,3,6)}\cong \fP_{(2,1,2)}$ above is given by
the introduction of a new variable $(x_0')=x_0^3,$ which is in spite of
appearances a one to one coordinate transformation (becuase of admissible
rescalings), and the Fermat polynomial becomes $(x_0')^2+x_1^4+x_2^2=0$.
Again, the isomorphism $\fP_{(2,1,2)}\cong \fP_{(1,1,1)}$ is given by
setting $(x_1')=x_1^2$, and the Fermat polynomial becomes
$(x_0')^2+(x_1')^2+x_2^2=0$, which is a quadric in the projective plane. We
{\it denote} this process by the symbolic expressions
\[ \fP_{(2,3,6)}[12]\cong \fP_{(2,1,2)}[4] \cong \fP_{(1,1,1)}[2].\]
It is well-known how to resolve the weighted projective space
$\fP_{(w_0,\ldots, w_n)}$. For this, one takes the following vectors in
$\fR^n$,
\begin{equation}\label{vectors}
v_0 = {1\over w_0} \left(\begin{array}{c} -1 \\ -1 \\ \vdots \\ -1
\end{array}\right),\quad v_1={1\over w_1}\left(\begin{array}{c} 1 \\ 0
\\ \vdots \\ 0
\end{array}\right), \quad \ldots, \quad v_n={1\over w_n}
\left(\begin{array}{c} 0 \\ \vdots \\ 0 \\ 1
\end{array}\right),
\end{equation}
and considers the lattice $\cL=\fZ v_0+\fZ v_1+\cdots + \fZ v_n$ in
$\fR^n$. The $n+1$ vectors $v_0,\ldots, v_n$ give a cone decomposition of
$\fR^n$, and this decomposition is {\it refined} until the resulting
decomposition satisfies: each cone has volume 1, where the volume is
normalized in such a way that the standard simplex in $\fR^n$ has volume
$\prod w_i$. This is equivalent to: any set of $n$ vectors spanning one of the
cones of the decomposition form a $\fZ$-basis of the lattice $\cL$.
Now suppose we are given a weighted hypersurface of degree $d$ in
$\fP_{(w_0,\ldots, w_n)}$, such that $d=\sum w_i$. Then, as is well-known,
this is a sufficient condition for the variety to be Calabi-Yau, i.e., the
dualizing sheaf is trivial. Supposing moreover that the hypersurface is
quasi-smooth, then in dimensions 2 and 3, by work of Roan and Yau
(\cite{GRY}, section 3),
there is a resolution of singularities such that the smooth
variety is still Calabi-Yau. In this case, the resolution (described in
\cite{RY}) is easier than of the ambient projective spaces themselves. The
reason is that it effectively reduces to a question of cones in one
dimension less. In particular, in the case of Calabi-Yau threefolds, the
resolution is described in terms of a simplicial decomposition of a
triangle (which is the {\it face} of one of the cones mentioned above). Let
$X \subset}\def\gk{\kappa}\def\cL{{\cal L} \fP_{\bf w}$ denote the singular weighted hypersurface. Then under
the assumption that $X$ is quasi-smooth, the singularities are all quotient
singularities by abelian groups. Locally they can be written as quotients
of $\fC^3$ by the following transformations
\begin{eqnarray*} \psi: \fC^3 & \longrightarrow}\def\cF{{\cal F} & \fC^3 \\
(z_1,z_2,z_3) & \mapsto & \left(\exp({a\over d})z_1,\exp({b\over d})z_2,
\exp({c\over d})z_3\right),
\end{eqnarray*}
which describes an action of the group of $d$th roots of unity
${\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}_d$ on $\fC^3$. Let $e^i,\ i=1,2,3$ denote the standard unit
vectors in $\fR^3$, and let
\[ v=\left( \begin{array}{c} a/d \\ b/d \\ c/d
\end{array} \right). \]
Let $\cL$ denote the lattice in $\fR^3$ spanned by the $e^i$ and
$v$. Finally, let $\sigma$ denote the cone
\[ \sigma=\left\{ \sum _{i=1} ^3 x_i e^i \in \fR^3 | x_i \geq 0,
i=1,2,3\right\}. \]
Then this affine cone determines a toric variety, which is a neighborhood
of the singularity in question. Next one uses the fact that, since $X$ is
Calabi-Yau, $a+b+c$ is divisible by $d$, and this in turn implies that the
integral vectors we require to decompose the cone to get a smooth cone
decomposition all lie in a hypersurface. This is given by
\[ \cF = \sigma \cap \left\{ \sum x_i =1 \right\}.\]
This is a triangle, and we need to determine the number of vertices, edges
and two-simplices of the simplicial decomposition of $\cF$
to determine the number of resolution divisors, the
number of intersection curves and intersection points.
First of all, since the area of $\cF$ is equal to $d$, there must be a
total of $s=d$ simplices in the decomposition.
\begin{lemma}\label{formula} Define integers $d_i$ as follows.
\[ d_1=gcd(a,d),\quad d_2=gcd(b,d), \quad d_3=gcd(c,d).\]
Then the number $v$ of vertices and $e$ of egdes in a smooth decomposition is
\[ v = {d+2 + ( d_1+d_2+d_3) \over 2},\quad e= {3d +(d_1+d_2+d_3) \over
2}.\]
Here, $d$ is the order of the group acting, and is arbitrary (not
necessarily odd as in \cite{RY}).
\end{lemma}
{\bf Remark:} This formula is different than that given in \cite{RY}. In
that paper, the authors only consider cases in which $d_i=1$ all $i$, which
is the same thing as only having {\it isolated} singular points, and in
these cases, our formula does agree with theirs.
\noindent{\bf Proof:} Note that if the singular point is not isolated, then
there are singular curves meeting at the point; in such a case, if the
singularity along the curve is $\fZ/e\fZ$, then $e-1$ divisors are
introduced to resolve the curve. This corresponds to $e-1$ vertices of the
decomposition, which lie on one of the edges. Let $y$ denote the number of
vertices which lie on the boundary of $\cF$, i.e., on one of the
edges. This number is determined exactly as in \cite{RY}, and is $y+3 =
d_1+d_2+d_3$, where the 3 are the original vertices of the triangle. Next,
it is easy to see that we may assume that all $y$ of these lie on one edge,
as in the following picture; the number of vertices, edges and simplices
remains contant:
\[ \epsfbox{ryau.eps}\]
Now it is easy to count the number of simplices (which is $d$) in terms of
$x=$ the number of inner vertices, and $y$. As in the following picture,
one gets the equation
\[ x+1 + x+y = d.\]
\[ \epsfbox{ryau2.eps} \]
Furthermore, the total number of vertices is $v=3+x+y$, and from these two
equations we get the formula for $v$. Since the triangle has Euler number =
$1=v-e+d$, the number of edges $e$ follows from this. \hfill $\Box$
\section{The twist map}
Let $V_1,\ V_2$ be weighted hypersurfaces
defined as follows.
\begin{equation}\label{8} \parbox{10cm}{$\displaystyle V_1 = \{x_0^{\ell}
+p(x_1,\ldots,x_n) =0 \} \subset}\def\gk{\kappa}\def\cL{{\cal L} \fP_{(w_0,w_1,\ldots, w_n)} \\ V_2 = \{
y_0^{\ell} + q(y_1,\ldots, y_m) =0 \} \subset}\def\gk{\kappa}\def\cL{{\cal L} \fP_{(v_0,v_1,\ldots,v_m)}$},
\end{equation}
where we assume both $p$ and $q$ are quasi-smooth. The degrees of these
hypersurfaces are
\[ \nu=\deg(V_1) = \ell\cdot w_0,\quad \mu=\deg(V_2) = \ell\cdot v_0.\]
We then consider the hypersurface
\begin{equation}\label{9} X:= \{ p(z_1,\ldots, z_n) -q(t_1,\ldots, t_m) =0
\} \subset}\def\gk{\kappa}\def\cL{{\cal L} \fP_{(v_0 w_1, \ldots, v_0 w_n, w_0 v_1, \ldots, w_0 v_m )}.
\end{equation}
Note that the degree of $X$ is $v_0\cdot \deg(p)=w_0\cdot \deg(q) =
v_0w_0\ell$.
The following was shown in part I of this paper.
The rational map
\begin{eqnarray*} \Phi: \fP_{(w_0,w_1,\ldots, w_n)} \times
\fP_{(v_0,v_1,\ldots,v_m)} & \longrightarrow}\def\cF{{\cal F} & \fP_{(v_0 w_1, \ldots, v_0 w_n, w_0
v_1, \ldots, w_0 v_m )} \\ ((x_0,\ldots, x_n),(y_0,\ldots, y_m)) &
\mapsto & (y_0^{w_1/w_0}\cdot x_1, \ldots, y_0^{w_n/w_0}\cdot x_n,
x_0^{v_1/v_0}\cdot y_1, \ldots, x_0^{v_m/v_0}\cdot y_m)
\end{eqnarray*}
restricts to $V_1\times V_2$ to give a rational generically finite
map onto $X$. Under the assumption that $w_0, v_0$ and $\ell$ have no
non-trivial common divisor, this map is generically $\ell$ to
one and $V_1\times V_2 \longrightarrow}\def\cF{{\cal F} X$ is the projection onto the quotient of
$V_1\times V_2$ by ${\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}_{\ell}$, which acts on the product $V_1\times
V_2$. Moreover, assuming that $X$ is Calabi-Yau, $V_2$ is Calabi-Yau and
$w_0 > 1$,
there is a resolution of singularities $\tilde{X}$ of $X$ which possesses a
fibration onto a resolution $Y$ of $V_1/{\bgm}} \def\bgls{\hbox{\scriptsize\boldmath$\lambda$}_{\ell}$ (\cite{I}, Lemma 3.4).
\section{Kodaira's
singular elliptic fibers with torsion Monodromy}
Before turning to degenerations of K3 surfaces we show how to rederive
Kodaira's classification of singular fibers of elliptic surfaces as an
application of the twist map. So pretend we had no idea about this
classification. We will characterize singular fibers in terms of the
relations which their monodromy matrices must fulfill. In aftermath, using
the fact that the monodromy matrices are elements of $SL(2,\fZ)$, we could,
using the known properties of $SL(2,\fZ)$, derive the classification given
by Kodaira.
These singular fibers were classified by Kodaira; his method was to
construct these fibers as quotients of smooth families of elliptic curves
of the form $D\times E$, where $D$ is a disc and $E$ is an elliptic curve
with an automorphism, i.e., of modulus either $i$ or $\varrho}\def\fP{{\bf P}}\def\gD{\Delta$. His
construction was hence in terms of a {\it local} group action. We will show
how this can be easily derived upon application of our twist map, which
displays things in terms of {\it global} quotients. We
consider the following K3 surfaces which are images under the twist map of
products of a curve $C$ and an elliptic curve $E$.
\begin{table}
\[ \!\begin{array}{|c|c|c|c|c|c|c|c|} \hline
& (w_0,w_1,w_2) & (v_0,v_1,v_2) & \ell & (k_1,k_2,k_3,k_4) & d &
\hbox{singular fibers} & Monodromy \\ \hline \hline
1& (2,1,1) & (1,1,1) & 3 & (1,1,2,2) & 6 & 6\times \IV & A \\ \hline
2& & (1,1,2) & 4 & (1,1,2,4) & 8 & 8\times \III & B \\ \hline
3& & (1,2,3) & 6 & (1,1,4,6) & 12 & 12\times \rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I & C , C^2=A
\\ \hline
4& (3,1,2) & (1,1,2) & 4 & (1,2,3,6) & 12 & 6\times \III, 1\times \I_0^*
& B,\ B^{-2}=-1 \\ \hline
5& & (1,2,3) & 6 & (1,2,6,9) & 18 & 9\times \rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I, 1\times \I_0^* & C,\
C^{-3}=-1 \\ \hline
6& (4,1,3) & (1,1,1) & 3 & (1,3,4,4) & 12 & 4\times \IV, 1\times \IV^* & A,\
A^{-1} \\ \hline
7& & (1,2,3) & 6 & (1,3,8,12)& 24 & 8\times \rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I, 1\times \IV^* & C,\
C^{-2}=A^{-1} \\ \hline
8& (5,1,4) & (1,1,2) & 4 & (1,4,5,10)& 20 & 5\times \III, 1\times \III^* & B,\
B^{-1} \\ \hline
9& (7,1,6) & (1,2,3) & 6 & (1,6,14,21)&42 & 7\times \rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I, 1\times \rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I^* & C,\
C^{-1} \\ \hline
10& (5,2,3) & (1,2,3) & 6 & (2,3,10,15)&30 & 5\times \rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I, 1\times \IV^*,
1\times \I_0^* & C,\ C^{-2}, C^{-3}=-1 \\ \hline
11& (11,5,6) & (1,2,3) & 6 & (5,6,22,33) & 66 & 2 \times \rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I,\ 2\times
\rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I^* & C,\ C^{-1} \\ \hline
\end{array}\]
\caption{K3 surfaces with constant modulus elliptic fibrations}
\end{table}
In Table 1, the K3 surfaces of Fermat type with the named weights are
described as images under the twist map of products $C\times E$. There are
three elliptic curves which occur, namely:
\begin{eqnarray*} E_1 & = & \{ y_0^3 +y_1^3 +y_2^3 = 0 \} \subset}\def\gk{\kappa}\def\cL{{\cal L}
\fP_{(1,1,1)}=\fP^2. \\ E_2 & = & \{y_0^4+y_1^4+y_2^2 = 0 \} \subset}\def\gk{\kappa}\def\cL{{\cal L}
\fP_{(1,1,2)}. \\ E_3 & = & \{y_0^6+y_1^3+y_2^2=0\} \subset}\def\gk{\kappa}\def\cL{{\cal L} \fP_{(1,2,3)}.
\end{eqnarray*}
and the curves $C$ are those curves of
degree $w_0\ell$ in $\fP_{(w_0,w_1,w_2)}$, which we take to be of Fermat
type (except for case 11).
In the last columns we list the Monodromy matrices, without using
what they look like. For example, in the fourth case, we have six singular
fibers of type III (this is seen by finding the number of zeros of the
polynomial $x_1^{12}+x_2^6=0\subset \fP_{(1,2)}$, which is six), and each
has monodromy matrix $B$. Recall that the monodromy gives a representation
of $\pi_1(B-\gD)$, where $B$ is the base of the fibration and $\gD$ is the
ramification locus. Since in our case $B=\fP^1$, it is known what this
fundamental group is: $\pi_1(\fP^1-\{n \hbox{ points}\}) = < \alpha}\def\fC{{\bf C}}\def\gS{\Sigma}\def\fZ{{\bf Z}_1,\ldots,
\alpha}\def\fC{{\bf C}}\def\gS{\Sigma}\def\fZ{{\bf Z}_n | \prod_i\alpha}\def\fC{{\bf C}}\def\gS{\Sigma}\def\fZ{{\bf Z}_i =1 >$. In case 4, it follows that since we have six
singular fibers of type III, the remaining monodromy matrix (it is easily
verified that there is just one) is given by a matrix $M$ satisfying the
relation $B^6\cdot M =1$, and since $B^4=1$, it follows that
$M=B^{-2}$. But since $B^4=1$ it follows that $(B^2)^2=(B^{-2})^2=1$ and hence
$M^2=1$. Similar considerations apply in all other cases.
To determine the structure of the degenerate fibers, note that for the first
three examples, we have (let us use $(z_1,z_2,z_3,z_4)$ as weighted
homogenous coordinates on the image weighted projective three-space) that
$z_1$ and $z_2$ are both non-vanishing, while for the sum of $d$th powers
we have $z_1^d
+z_2^d=0$. It is known that for weighted projective spaces for weights of
the form $(1,k_2,k_3,k_4)$, the affine open subset $z_1\neq 0$ is
really just a $\fC^3$. It follows that we may view, for each pair
$(z_1,z_2)$ such that $z_1^d
+z_2^d=0$, the corresponding fiber of the K3 as the curve given by the {\it
affine} equation in $\fC^2$
which results. In the three cases of interest, these
affine equations are
\[ \begin{array}{|c|c|c|} \hbox{affine equation} & \hbox{picture} &
\hbox{Monodromy matrix} \\ \hline
x^3+y^2=0 & \epsfysize=1cm \epsfbox{II.eps} & C,\ C^6=1 \\ \hline
x^4+y^2=0 & \epsfysize=1cm \epsfbox{III.eps} & B,\ B^4=1 \\ \hline
x^3+y^3=0 & \epsfysize=1cm \epsfbox{IV.eps} & A,\ A^3=1 \\ \hline
\end{array}
\]
To describe the other singular fibers which occur, one must consider now
the resolution of the singular weighted projective space. Then additional
singular fibers occur if: one of the ``special'' sections
$C_0:=\{z_2=0\}\cap X$ or $C_{\infty}:=\{z_1=0\}\cap X$ is {\it not} a
smooth plane cubic, an elliptic curve.
We will describe this for cases giving rise to the
singular fibers of types I$_0^*$, II$^*$, III$^*$ and IV$^*$,
respectively. These are the cases 4 (or 5), 6, 8 and 9 above. The monodromy
matrices of these fibers are $-1,\ C^{-1},\ B^{-1}$ and $A^{-1}$,
respectively, as we have explained above.
{\it Case 4: $\fP_{(1,2,3,6)}$} To see whether the curves $C_0$ and
$C_{\infty}$ are indeed rational, we use the following compact method
explained above:
\[ C_0 = \fP_{(1,3,6)}[12] \cong \fP_{(1,1,2)}[4],\]
which is the original
elliptic curve we started with.
\[ C_{\infty} = \fP_{(2,3,6)}[12]\cong \fP_{(2,1,2)}[4]\cong \fP_{(1,1,1)}[2],\]
which means this curve is isomorphic to a quadric in the usual projective
plane, hence {\it rational}. Next we need the singular locus of the
ambient space. This is\footnote{the notation $\{\ldots\}_{\fZ_k}$ indicates
that $\{\ldots\}$ is fixed under a $\fZ/k\fZ$ stabilizer}
$\gS=\{z_1=z_2=0\}_{\fZ_3} \cup \{z_1=z_3=0\}_{\fZ_2} =
\gS_2\cup \gS_1$, and the equation of the hypersurface is
$X=\{z_1^{12}+z_2^6+z_3^4+z_4^2=0\}$,
so for the intersections we have $\gS_1\cap X
=\fP_{(2,6)}[12]\cong \fP_{(1,3)}[6]\cong \fP_{(1,1)}[2] = 2$ points, while for
$\gS_2\cap X=\fP_{(3,6)}[12]\cong \fP_{(1,2)}[4]\cong \fP_{(1,1)}[2]=2$ points. Note
that all four points lie on the curve $C_{\infty}$, while {\it two} of them
lie on the curve $C_0$. In particular, the two curves meet in two
points. Now resolve the singularities of the ambient space; since we have
two $\fZ_2$ points and two $\fZ_3$ points, we get a total of 6 exceptional
curves, and as already mentioned, there are two ``fibers'', one rational
curve and one elliptic curve. The elliptic curve $C_0$ is clearly a smooth
fiber; it intersects two of the singular points, which has only one
explanation: there are two sections meeting it, which are components of the
exceptional locus. A picture will make this clearer:
\[ \unitlength1cm
\begin{picture}(10,7.5)
\put(-2,0){ \epsfbox{fig1.eps} }
\put(-2.1,7.2){$C_{\infty}$}
\put(.5,6.1){$C_0$}
\put(-1,6.3){$\fZ_3$}
\put(-.2,5.7){$\fZ_3$}
\put(-2,6.4){$\fZ_2$}
\put(0,4.5){$\fZ_2$}
\put(4.1,6.8){$C_{\infty}$}
\put(4.8,3.5){$C_0$}
\put(3.5,.65){$F_{\infty} = {\bf I}_0^*$}
\put(0.3,.75){$C_0$}
\put(7,2){The extended Dynkin diagram of $D_4$}
\put(9.5,1){\large{\bf I}$_0^*$}
\end{picture}
\]
We see easily the smooth fiber $C_0$ and the fiber of type $\I_0^*$,
$F_{\infty}$, consisting of the proper transform of $C_{\infty}$ and four
exceptional $\fP^1$'s introduced in the resolution.
Since we already deduced above that the monodromy matrix fulfills $M^2 =1$,
it follows that we have derived the structure of the singular
fiber of type I$_0^*$.
{\it Case 6: $\fP_{(1,3,4,4)}$}
The Fermat hypersurface is $X=\{z_1^{12}+z_2^4+z_3^3+z_4^3=0\}$ and the
singular locus consists of a single component
$\gS=\{z_1=z_2=0\}_{\fZ_4}$. The intersection with $X$ is
$\fP_{(4,4)}[12]\cong \fP_{(1,1)}[3]=3$ points. Note that at these three points
the two curves $C_{\infty}$ and $C_0$ intersect. They are
$C_{\infty}=\fP_{(3,4,4)}[12]\cong \fP_{(3,1,1)}[3]$ which is rational and
$C_0=\fP_{(1,4,4)}[12]\cong \fP_{(1,1,1)}[3]$, which is the smooth elliptic
curve. There are on the intersection three $\fZ_4$ points, resolving them
gives, in addition to the three sections, one smooth fiber and one fiber of
type IV$^*$. The picture is as follows
\[\unitlength1cm
\begin{picture}(10,6.5) \put(-2,0){\epsfbox{case6.eps}}
\put(-2.5,6){$C_{\infty}$}
\put(-.7,1.5){$C_0$}
\put(-1,5.5){$\fZ_4$}
\put(-1,4.3){$\fZ_4$}
\put(-1,3.3){$\fZ_4$}
\put(3.5,6){$C_{\infty}$}
\put(6.8,6){$C_0$}
\put(7.5,2){The extended Dynkin diagram of $E_6$}
\put(9.8,1){\large{\bf IV}$^*$}
\end{picture}
\]
{\it Case 8: $\fP_{(1,4,5,10)}[20]$}
The singular locus here is $\gS_1=\{z_1=z_3=0\}_{\fZ_2}$ and
$\gS_2=\{z_1=z_2=0\}_{\fZ_5}$. The curve $C_0$ is $\fP_{(1,5,10)}[20]\cong
\fP_{(1,1,2)}[4]$, which is elliptic, and $C_{\infty}=\fP_{(4,5,10)}[20]\cong
\fP_{(4,1,2)}[4]\cong \fP_{(2,1,1)}[2]$, which is rational. The two curves $C_0$
and $C_{\infty}$ meet at $X\cap \gS_2 =\fP_{(5,10)}[20] \cong \fP_{(1,2)}[4]
\cong \fP_{(1,1)}[2]=2$ points. Hence there are two sections in the exceptional
locus. There are $X\cap \gS_1 = \fP_{(4,10)}[20] \cong \fP_{(2,5)}[10]
\cong \fP_{(1,1)}[1] =1$ point more singularities. Resolving singularities we
easily find a smooth fiber, a fiber of type III$^*$ and two sections of the
fibration.
\[\unitlength1cm
\begin{picture}(10,3.6) \put(-3,0){\epsfbox{fig2.eps}}
\put(-2.6,1.3){$C_0$}
\put(-.4,.4){$C_{\infty}$}
\put(2.5,.5){$C_0$}
\put(5.5,.5){$C_{\infty}$}
\put(-2,3.3){$\fZ_5$}
\put(-1.4,1.6){$\fZ_5$}
\put(-2.1,2.3){$\fZ_2$}
\put(6.8,.5){The extended Dynkin diagram of $E_7$}
\put(9.2,0){\large{\bf III}$^*$}
\end{picture}
\]
Finally, the most interesting case is the one giving rise to the II$^*$
type fiber.
\noindent
{\it Case 9: $\fP_{(1,6,14,21)}$}
The Fermat hypersurface is given by
$$X=\{ z_1^{42}+z_2^7+t_1^3+t_2^2 =0 \} \subset \fP_{(1,6,14,21)}.$$
The two special curves are given by setting $z_1$ and $z_2$ to be zero:
$$C_{\infty}=\{z_1=0\} \cap X =\{ z_2^7+t_1^3+t_2^2=0\} \subset
\fP_{(6,14,21)} = \{ (z_2')^1+(t_1')^1+(t_2')^1 =0 \} \subset
\fP_{(1,1,1)}$$
which is clearly just a linear $\fP^1$, and
$$C_0=\{ z_1^{42} +t_1^3+t_2^2 =0\} \subset \fP_{(1,14,21)} \cong \{ (z_1')^6
+t_1^3+t_2^2 =0\} \subset \fP_{(1,2,3)},$$
which is clearly just our elliptic curve. The singular locus of the ambient
space is $\gS_1=\{z_1=z_2=0\}_{\fZ_7}$ and $\gS_2=\{z_1=z_3=0\}_{\fZ_3}$
and $\gS_3=\{z_1=z_4=0\}_{\fZ_2}$. Furthermore, $X\cap \gS_1
=\fP_{(14,21)}[42] \cong \fP_{(2,3)}[6]\cong \fP_{(1,1)}[1]=1 $ point, $X\cap \gS_2=
\fP_{(6,21)}[42]\cong \fP_{(2,7)}[14]\cong \fP_{(1,1)}[1]=1$ point, $X\cap
\gS_3=\fP_{(6,14)}[42] \cong \fP_{(3,7)}[21]\cong \fP_{(1,1)}[1]= 1$ point. All these
points are on the curve $C_{\infty}$, one of them is the intersection with
$C_0$. This is described in the following picture
\[\unitlength1cm
\begin{picture}(10,4)
\put(-3,0){\epsfbox{case8.eps} }
\put(-3,3){$C_{\infty}$}
\put(-.5,.9){$C_0$}
\put(2.2,2.85){$C_{\infty}$}
\put(6.8,-.3){$C_0$}
\put(-2.7,2.4){$\fZ_3$}
\put(-1.4,2.4){$\fZ_2$}
\put(0.2,2.3){$\fZ_7$}
\put(7.5,1.5){The extended Dynkin diagram of $E_8$}
\put(10,.5){\large{\bf II}$^*$}
\end{picture}
\]
For completeness we discuss briefly the last two cases in the table.
\smallskip
{\it Case 10: $\fP_{(2,3,10,15)}$}.
First we have the fixed points given by $x_0=0$.
We are looking for solutions of
$$ \{ x_1^{15}+x_2^{10} =0 \} \subset {\bf P}_{(2,3)},$$
which is the same as $ \{ (x_1')^5+(x_2')^5 =0 \}
\subset {\bf P}_{(1,1)}$, of which
there are obviously only five solutions. So we have five singular fibers of
type II. The singular locus of the ambient space is
$\gS_1=\{z_1=z_2=0\}_{\fZ_5}$, $\gS_2=\{z_1=z_3=0\}_{\fZ_3}$ and
$\gS_3=\{z_2=z_4=0\}_{\fZ_2}$. We have for the intersections $X\cap \gS_1 =
\fP_{(10,15)}[30]\cong \fP_{(2,3)}[6]\cong \fP_{(1,1)}[1]=1$ point, $X\cap
\gS_2=\fP_{(3,15)}[30] \cong \fP_{(1,5)}[10]\cong \fP_{(1,1)}[2]=2$ points, and $\gS_3
= \fP_{(2,10)}[30]\cong \fP_{(1,5)}[15]\cong \fP_{(1,1)}[3]=3$ points. The two curves
$C_0$ and $C_{\infty}$ meet in a single point (the $\fZ_5$ point), hence
there is a single section in the exceptional locus. There are three $\fZ_2$
points on $C_0$, while there are two $\fZ_3$ points on the $C_{\infty}$, in
addition to the common $\fZ_5$ point. For the curves $C_0$ and $C_{\infty}$
we have $C_0=\fP_{(2,10,15)}[30]\cong \fP_{(1,5,15)}[15]\cong \fP_{(1,1,3)}[3]$,
which is a rational curve, and
$C_{\infty}=\fP_{(3,10,15)}[30]\cong \fP_{(1,10,5)}[10] \cong \fP_{(1,2,1)}[2]$,
which is also a rational curve. We have the picture:
\[\unitlength1cm
\begin{picture}(10,8)
\put(0,0){\epsfbox{fig3.eps}}
\put(0,3.3){$C_{\infty}$}
\put(2.9,4.5){$C_0$}
\put(4.3,7.5){$C_0$}
\put(5,5.2){$C_{\infty}$}
\put(-.8,6.6){$\fZ/2\fZ$}
\put(2.4,3.7){$\fZ/3\fZ$}
\put(.8,5.9){$\fZ/5\fZ$}
\put(5.1,2.8){$C_0$}
\put(9.8,3){$C_{\infty}$}
\put(5.1,0){I$_0^*$}
\put(9.8,.3){IV$^*$}
\end{picture}\]
\smallskip
{\it Case 11:} $\fP_{(5,6,22,33)}$
For the weights in this case, a Fermat hypersurface is
not possible. We consider instead the following polynomial:
\[ \{z_0^{12}z_1+z_1^{11}+z_2^3+z_3^2=0\} \subset}\def\gk{\kappa}\def\cL{{\cal L} \fP_{(5,6,22,33)}.\]
We see without difficulty that this is the image under the twist map
\begin{eqnarray*} \fP_{(11,5,6)}\times \fP_{(1,2,3)} & \longrightarrow}\def\cF{{\cal F} &
\fP_{(5,6,22,33)}\\((x_0:x_1:x_2),(y_0:y_1:y_2)) & \mapsto &
(y_0^{5/11}x_1: y_0^{6/11}x_2:x_0^2y_1:x_0^3y_2 )
\end{eqnarray*}
of the product
$ \{ x_0^6+x_1^{12}x_2 +x_2^{11}=0\} \times \{y_0^6+y_1^3+y_2^2=0\}$.
As was explained in part I, this is a K3 surface, and we now describe the
singular fibers. The singular locus has the following components:
$\gS_1=\{z_1=z_2=0\}_{\fZ_{11}}$, $\gS_2=\{z_1=z_3=0\}_{\fZ_3}$,
$\gS_3=\{z_1=z_4=0\}_{\fZ_2}$, $\gS_4=(\gS_2\cap \gS_3)_{\fZ_6}$ and
$\gS_5=\{(1,0,0,0)\}_{\fZ_5}$. The intersections with $X$ are as
follows. $\gS_1\cap X=\fP_{(22,33)}[66]$= 1 point, $\gS_2\cap X =
\fP_{(6,33)}[66]\cong \fP_{(6,3)}[6]\cong \fP_{(2,1)}[2] =$ 1 point, $\gS_3 \cap X =
\fP_{(6,22)}[66] \cong \fP_{(6,2)}[6] \cong \fP_{(3,1)}[3]=$ 1 point. At the same
time the fibers $C_0$ and $C_{\infty}$ are as follows: $C_0 = \{ z_2=0\} =
\{ z_2^3+z_4^2=0\}$, which is a cusp, and $C_{\infty} =\{z_1=0\} =
\fP_{(6,22,33)}[66] \cong \fP_{(6,2,3)}[6] \cong \fP_{(1,1,1)}[1]$, which is a
rational curve. Note that the $\fZ_5$ point is at the cusp of $C_0$. We
have the following picture:
\[\unitlength1cm
\begin{picture}(10,8)
\put(-4,0){\epsfbox{figcase11.eps}}
\put(0.6,7.2){$C_{\infty}$}
\put(3,0){$C_0$}
\put(-1.8,2.5){section}
\put(11.5,4.5){section of }
\put(11.5,4){the fibration}
\put(6.5,2.5){II$^*$}
\put(8.9,2.5){II$^*$}
\end{picture}
\]
The $\fZ/5\fZ$ point at the cusp of $C_0$ is resolved in the usual manner,
being replaced by a chain of length four. The proper transform of $C_0$ is
the ``center'' curve of the resulting configuration, which we have drawn in
the picture seperately for clarity. Thus we get the two fibers of type
$\rm II} \def\III{\rm III} \def\IV{\rm IV} \def\I{\rm I^*$.
\section{Singular K3 fibers with torsion monodromy}
In this section we wish to do the same as above, but now for a set of
K3-fibrations.
\subsection{Fibers analogous to Kodaira's type II, III and IV}
First we have the analogy to the simple Kodaira fibers. Once again, we have
some unknown monodromy matrix, of which we know only the order. In these
cases, just as above, we have an affine surface as the singular
fiber. These are listed in Table 2.
\subsection{Fibers analogous to Kodaira's type I$_0^*$, II$^*$, III$^*$ and IV$^*$}
\begin{table}
\begin{tabular}{|c|c|c|c|c|c|c|} \hline
fiber & affine equation & $\mu$ & Euler \# & fiber & Monodromy & relation \\ \hline\hline
IV$_1$ & $z^6+x^3 + y^3 =0$ & 20 & 4 & \epsfysize=2cm \epsfbox{plot1.ps}
& M(IV$_1$) & M(IV$_1$)$^6$=1 \\ \hline
III$_1$ & $z^8+x^4+y^2=0$ & 21 & 3 & \epsfysize=2cm \epsfbox{plot2.ps} &
M(III$_1$) & M(III$_1$)$^8$=1 \\ \hline
II$_1$ & $z^{12}+x^3+y^2=0$ & 22 & 2 & \epsfysize=2cm \epsfbox{plot3.ps} &
M(II$_1$) & M(II$_1$)$^{12}$ =1 \\ \hline
IX$_1$ & $z^6+x^4+y^2=0$ & 15 & 9 & \epsfysize=2cm \epsfbox{plot4.ps} &
M(IX$_1$) & M(IX$_1$)$^{12}$=1 \\ \hline
VIII$_1$ & $z^9+x^3+y^2=0$ & 16 & 8 & \epsfysize=2cm\epsfbox{plot5.ps} &
M(VIII$_1$) & M(VIII$_1$)$^{18}$=1 \\ \hline
XII$_1$ & $z^4+x^3+y^3=0$ & 12 & 12 & \epsfysize=2cm \epsfbox{plot6.ps} &
M(XII$_1$) & M(XII$_1$)$^{12}$=1 \\ \hline
X$_1$ & $z^8+x^3+y^2=0$ & 14 & 10 & \epsfysize=2cm \epsfbox{plot7.ps} &
M(X$_1$) & M(X$_1$)$^{24}$=1 \\ \hline
XII$_2$ & $z^5+x^4+y^2=0$ & 12 & 12 & \epsfysize=2cm \epsfbox{plot8.ps}&
M(XII$_2$) & M(XII$_2$)$^{20}$=1 \\ \hline
XII$_3$ & $z^7+x^3+y^2=0$ & 12 & 12 & \epsfysize=2cm \epsfbox{plot9.ps} &
M(XII$_3$) & M(XII$_3$)$^{42}$=1 \\ \hline
VI$_1$ & $z^{10}+x^3+y^2=0$ & 18 & 6 & \epsfysize=2cm \epsfbox{plot10.ps}
& M(VI$_1$) & M(VI$_1$)$^{15}$=1 \\ \hline
\end{tabular}
\caption{List of singular K3-fibers with nilpotent Monodromy}
\end{table}
Here we repeat the analysis above, this time applied to weighted
hypersurfaces which are K3-fibrations. We consider the cases listed in
Table 3.
\begin{table}
\[ \begin{array}{|l|l| c | l | c|c|c|} \hline (w_0,w_1,w_2)
& (v_0,v_1,v_2,v_2) & \ell & (k_1,k_2,k_3,k_4,k_5) & d & Euler \# &
singular fibers \\ \hline \hline
(2,1,1) & (1,1,2,2) & 6 & (1,1,2,4,4) & 12 & -192 & 12\times \hbox{IV$_1$} \\
& (1,2,3,6) & 12 & (1,1,4,6,12) & 24 & -312 & 24\times \hbox{IX}_1
\\
& (1,6,14,21) & 42 & (1,1,12,28,42) & 84 & -960 & 84
\times \hbox{XII}_3 \\ \hline
(3,1,2) & (1,1,2,2) & 6 & (1,2,3,6,6) & 18 & -144 & 9\times \hbox{IV}_1,
1\times \hbox{IV}_1^* \\
& (1,2,3,6) & 12 & (1,2,6,9,18) & 36 & -228 & 18\times
\hbox{IX}_1, 1\times \hbox{IX}_1^* \\
& (1,6,14,21) & 42 & (1,2,18,42,63) & 126 & -720 & 63\times
\hbox{XII}_3, 1\times \hbox{XII}_3^* \\ \hline
(4,1,3) & (1,1,2,2) & 6 & (1,3,4,8,8) & 24 & -120 & 8\times \hbox{IV}_1,\
1\times \hbox{IV}_1^{**} \\
& (1,2,3,6) & 12 & (1,3,8,12,24) & 48 & -192 & 16\times
\hbox{IX}_1,\ 1\times \hbox{IX}_1^{**} \\
& (1,6,14,21) & 42 & (1,3,24,56,84) & 168 & -624 & 56\times
\hbox{XII}_3, 1\times \hbox{XII}_3^{**} \\ \hline
(5,1,4) & (1,2,3,6) & 12 & (1,4,10,15,30) & 60 & -168 & 15\times
\hbox{IX}_1, 1\times \hbox{IX}_1^{3*} \\ \hline
(7,1,6) & (1,2,3,6) & 12 & (1,6,14,21,42) & 84 & -132 & 14\times
\hbox{IX}_1,\ 1\times \hbox{IX}_1^{4*} \\
& (1,6,14,21) & 42 & (1,6,42,98,147) & 294 & -480 & 49\times
\hbox{XII}_3,\ 1\times \hbox{XII}_3^{3*} \\ \hline
(5,2,3) & (1,1,2,2) & 6 & (2,3,5,10,10) & 30 & -72 & 5\times \hbox{IV}_1,\
1\times {\rm IV}_1^{*}, 1\times {\rm IV}_1^{**}\\
& (1,2,3,6) & 12 & (2,3,10,15,30) & 60 & -108 &10\times
\hbox{IX}_1, 1\times {\rm IX}_1^*, 1\times {\rm IX}^{**}_1 \\
& (1,6,14,21) & 42 & (2,3,30,70,105) & 210 & -384 & 35\times
\hbox{XII}_3, 1\times {\rm XII}_3^*, 1\times {\rm XII}_3^{**} \\ \hline
\end{array}
\]\caption{K3-fibered Calabi-Yau weighted hypersurfaces which are
also elliptic fibered, have constant modulus and are of Fermat type}
\end{table}
Just as in the case of the elliptic fibrations we have the surfaces
$C_0:=\{z_2=0\}\cap X$ and $C_{\infty}:=\{z_1=0\}\cap X$. In the first
three cases it is easy to see that both of these surfaces are just smooth
fibers, hence the only singular fibers which occur are those for which the
affine surface listed in Table 2 of types IV$_1$, IX$_1$ and XII$_3$,
respectively, describe the singular fibers. Those of interest to us here
occur in the remaining cases. We begin by discussing the singular fibers
denoted IV$_1^*$, IX$_1^*$ and XII$_3^*$ in Table 3.
\subsection{Fibers analogous to I$_0^*$}
In this section we consider the cases 4-6 in the table above, in which the
monodromy matrix of the singular fiber fulfills $M^2=1$.
In this respect, each of
these is an analog of Kodaira's I$_0^*$ type fiber.
\smallskip\noindent{\it Case }IV$_1^*$
We describe this example in more detail as a description of the general
proceedure to be used in the sequel.
The bad fiber is $C_{\infty} \cong \fP_{(2,3,6,6)}[18] \cong \fP_{(2,1,2,2)}[6]
\cong \fP_{(1,1,1,1)}[3]$, a cubic surface. (This is a del-Pezzo surface of
degree 3, which is $\fP^2$ blown up in six points). The singular locus of
the ambient space is $\gS_1=\{z_1=z_2=0\}_{\fZ_3},\
\gS_2=\{z_1=z_3=0\}_{\fZ_2}$ and their intersection is $(\gS_1\cap
\gS_2)_{\fZ_6}$. Note that $\gS_1\cap X =\fP_{(3,6,6)}[18]\cong \fP_{(1,2,2)}[6]
\cong \fP_{(1,1,1)}[3]$, which is a cubic curve, which is elliptic. The other
intersection is $\gS_2\cap X
=\fP_{(2,6,6)}[18]\cong \fP_{(1,3,3)}[9]\cong \fP_{(1,1,1)}[3]$, which is again an
elliptic curve. They meet in the three $\fZ_6$-points on $X$
($\fP_{(6,6)}[18]\cong \fP_{(1,1)}[3] $= three points). We have the following
picture:
\[\unitlength1cm
\begin{picture}(10,3)(0,.5)\put(0,0){\epsfbox{IVstar.eps}}
\put(.9,1.5){$C_{\infty}$}\put(1.5,.5){$C_0$}\put(-.3,2.9){The curves
$\gS_i$}
\end{picture}
\]
Each of the dots represents a $\fZ_6$--point, $\gS_1\cap X$ is the intersection
of the two surfaces $C_0$ and $C_{\infty}$.
The curve $\gS_1$ is the {\it base locus} of the K3-fibration, i.e., every
fiber $X_s$ passes through $\gS_1$; we think of the base $\fP^1$ of the
fibration as the exceptional $\fP^1$ of directions through $\gS_1$. Note
the general fiber is $\fP_{(1,3,6,6)}[18]\cong \fP_{(1,1,2,2)}[6]$, which
has three $\fZ_2$ points ($\{z_1=z_2=0\}_{\fZ_2} \cap X = \fP_{(2,2)}[6]
\cong \fP_{(1,1)}[3]$), whose resolutions are in fact sections of the
elliptic fibration of the fiber. This exceptional $\fP^1$ in each fiber is
the intersection of the fiber with the exceptional
divisor ($\cong \fP_{(1,2,3)}$ described below) at each $\fZ_6$-point.
The $\fZ_6$-singular points are
of the type ${1\over 6}(1,2,3)$, which is the usual shorthand for the
quotient of $\fC_3$ by the action $(z_1,z_2,z_3)\mapsto (e^{2\pi i\over6}z_1,
e^{2\pi i \cdot 2\over 6}z_2,e^{2\pi i \cdot 3\over 6}z_3)$.
To describe the resolution of
the singularities, we refer to the paper \cite{aspin}, in which these
$\fZ_6$-points have been resolved. The resolution is described by a cone
decomposition of the triangle with vertices $\gk_1:=(6,-2,-3),\
\gk_2:=(0,1,0),\ \gk_3:=(0,0,1)$\footnote{this is a different, but
equivalent, description of what we described above},
which is the face of a three-dimensional
cone. Here this decomposition looks as follows\footnote{in this and
following diagrams, vertices which are circled belong to divisors $E$
which are sections of the fibration (i.e., $E\cap X_s$ is a curve for all
$s\in \fP^1$), hence do not belong to the singular fiber.}:
\[ \epsfbox{triangle1.eps} \]
in which the two (respectively one) vertices on the edge
$\overline{\gk_1,\gk_2}$ (respectively $\overline{\gk_1\gk_3}$) correspond
to the two (respectively one) exceptional divisors over $\gS_i$. One of
the two divisors over $\gS_1$ is a curve-section of the fibration: every
fiber passes through $\gS_1$, hence the intersection of one of the divisors
with each fiber is a curve, in this case isomorphic to the curve $\gS_1$,
which is elliptic. Thus, only the
other two components over the singular loci $\gS_1$ and $\gS_2$
belong to the singular fiber. The vertex in the middle of the
triangle corresponds to an additional exceptional divisor, which one easily
sees is just a copy of $\fP_{(1,2,3)}$, which is then resolved when the
singular curves $\gS_i$ are. There are six cones (triangles) decomposing
the big one, corresponding to the fact that the lattice is of index
six. See also \cite{RY} for details on these matters. Altogether there are
at each $\fZ_6$-point a total of four exceptional divisors; three of these
are the exceptional divisors over the $\gS_i$, the additional one at each
point is the $\fP_{(1,2,3)}$ just mentioned. Note that these latter
exceptional surfaces are also $\fP^1$-sections of the fibration, as they
lie on $\gS_1\cap X$, hence meet all fibers.
After resolution of singularities, we have the following divisors which
were introduced:
\begin{enumerate}\item Over $\gS_1$, two elliptic ruled surfaces
$\Theta_{1,1}$ and $\Theta_{1,2}$, which intersect each other in a
section of the ruling. \item Over $\gS_2$, an elliptic ruled surface,
$\Theta_{2,1}$. \item Over each $\fZ_6$-point, a (resolution of a) copy of
$\fP_{(1,2,3)}$; this intersects the fiber $F_{\infty}$ in the union of
four rational curves, and intersects the other fibers in an exceptional
$\fP^1$. Let $\Theta_3,\ \Theta_4,\ \Theta_5$ denote the three
exceptional divisors introduced over the three points. \item The proper
transform $\Theta$ of $C_{\infty}$.
\end{enumerate}
The proper transform $[C_{\infty}]$ is the cubic surface
$\fP_{(2,3,6,6)}[18] \cong \fP_{(1,1,1,1)}[3]$ blown up at three disjoint
points. Let $\Theta$ denote this surface; it is a rational elliptic surface
with $e(\Theta)=12,\ K_{\Theta}^2=0$. The singular fiber $F_{\infty}$ is
\[ F_{\infty} = \Theta\cup \Theta_{1,1}\cup \Theta_{2,1}.\]
The following divisors are sections of the fibration (hence do not belong
to the fiber $F_{\infty}$): $\Theta_{1,2}$, $\Theta_3$, $\Theta_4$,
and $\Theta_5$. The fiber $F_{\infty}$ is depicted in Figure 1.
\begin{figure}
\[\unitlength1cm
\begin{picture}(10,5) \put(0,0){ \epsfbox{fiber1.eps}}
\put(.8,1.3){$\Theta$}\put(1.5,2.7){$\Theta_2$}\put(6,3.7){$\Theta_1$}
\put(6.7,.7){Three exceptional $\fP^1$'s on the }
\put(6.7,.2){blow-up of the cubic surface}
\end{picture}
\]
\caption{\label{figure1}This is the singular fiber of type ${\rm VI}^*_1$.
It consists of
three components, the image of the K3 surface itself, here denoted
$\Theta$, which is the proper transform of
a copy of $\fP_{(1,1,1,1)}[3]$, a cubic surface. The exceptional
divisor over $\gS_i$ is denoted $\Theta_i$, and is an elliptic ruled
surface, while $\Theta$ is a rational elliptic surface (it is $\fP^2$
blown up at nine points), and each of the intersections
$\Theta \cap \Theta_i$ is an elliptic curve.}
\end{figure}
We can also determine the properties of the monodromy matrix which
determines this bad fiber: since the fibration has 9 singular fibers of
type IV$_1$, and M(IV$_1$)$^6=1$, it follows that the monodromy matrix $M$
here must fulfill M(IV$_1$)$^9 \cdot M =1$, i.e., $M=$M(IV$_1$)$^{-3}$ and
hence $M^2=1$.
So this fiber is in a sense an analoge of the Kodaira type I$_0^*$.
Moreover, an easy calculation gives the Euler number of this fiber. Indeed,
since the fibered threefold has Euler number $-144$, and there are nine
fibers of type IV$_1$ and a single fiber of type IV$^*_1$, we get from the
formula
\[ e(X) = 24\cdot(2-(9+1)) + 9 \cdot e({\rm IV}_1) + 1\cdot e({\rm IV}_1^*)
= -144,\]
that the Euler number of the bad fiber is $12$. We can check this, by
calculating the Euler number of our bad fiber: it is
\[ e({\rm IV}_1^*) = e(\Theta)+e(\Theta_1)+e(\Theta_2) -e(\Theta\cap
\Theta_1) - e(\Theta\cap\Theta_2) = 12 + 0 + 0 - 0 - 0,\]
since elliptic ruled surfaces, as well as elliptic
curves, have Euler number 0.
In the sequel we will not go into such detail.
\smallskip\noindent{\it Case} IX$_1^*$
The bad fiber is again $C_{\infty} \cong \fP_{(2,6,9,18)}[36] \cong
\fP_{(2,2,3,6)}[12] \cong \fP_{(1,1,3,3)}[6]$.
The singular locus of the ambient space is $\gS_1=\{z_1=z_2=0\}_{\fZ_3}$,
$\gS_2= \{z_1=z_4=0\}_{\fZ_2}$ and the lower-dimensional parts are given by
$\gS_3=\{z_1=z_2=z_3=0\}_{\fZ_9}$ and $\gS_4=\{z_1=z_2=x_4=0\}_{\fZ_6} =
\gS_1\cap \gS_2$. The intersections with $X$ give $\gS_1\cap X
=\fP_{(6,9,18)}[36] \cong \fP_{(2,3,6)}[12] \cong \fP_{(2,1,2)}[4]\cong \fP_{(1,1,1)}[2]$,
which is a rational curve, and $\gS_2\cap X =\fP_{(2,6,18)}[36]
\cong \fP_{(1,3,9)}[18] \cong \fP_{(1,1,3)}[6]$ which is a curve of
genus 2 (it is a double cover of $\fP^1$ branched along a sextic, or
alternatively, a degree six curve on the Hirzebruch surface
$\fP_{(1,1,3)}$).
These two intersect in $\gS_1\cap \gS_2\cap X =\fP_{(6,18)}[36]
\cong \fP_{(1,3)}[6] \cong \fP_{(1,1)}[2]$ which is two points. This is the locus
$\gS_4\cap X$ and consists of two points. The $\fZ_9$--locus yields
$\fP_{(9,18)}[36] \cong \fP_{(1,2)}[4] \cong \fP_{(1,1)}[2]$ which is also two
points. The configuration then looks like
\[\unitlength1cm
\begin{picture}(10,4)(0,-.5)\put(0.2,-.6){\epsfbox{IXstar.eps}}
\put(.9,1.5){$C_{\infty}$}\put(1.5,.5){$C_0$}\put(-.3,2.9){The curves
$\gS_i$}
\end{picture}
\]
in which the $\fZ_6$ points are the filled circles, the $\fZ_9$-points the
filled squares. The resolution of the curves $\gS_i,\ i=1,2$ is the same as
above, hence the resolution of the $\fZ_6$ points is precisely as above. To
describe the $\fZ_9$ points, we note that since they lie on a $\fZ_3$
curve, one of the fractions is $3/9$, hence we have ${1\over
9}(1,3,5)$. The resolution of this follows the same pattern as above. We
now find the following cone decomposition:
\[ \epsfbox{triangle2.eps} \]
Using Lemma \ref{formula},
we see that in our case of ${1\over 9}(1,3,5)$ we have $d_1=d_3=1,\
d_2=3,\ d_1+d_2+d_3=5$ and hence the number of vertices (including the
three corners of the original triangle) is 8, the number of edges is 16,
and the cone decomposition is as given above. The two vertices on the edge
of the triangle of course correspond to the two exceptional divisors over
the curve $\gS_1$, on which the $\fZ_9$-points lie. There are at each
$\fZ_9$-point three additional exceptional divisors. Note that again, one
of these three is a $\fP^1$-section of the fibration, while the other two
are components of the bad fiber. We have circled the vertices corresponding
to the exceptional divisors which are $\fP^1$-sections of the fibration.
After the resolution of singularities, our fiber will look as in Figure
\ref{figure2}.
\begin{figure}
\[\unitlength1cm
\begin{picture}(10,5)(2,0) \put(0,0){ \epsfbox{fiber2.eps}}
\put(.8,1.3){$\Theta$}\put(1.5,2.7){$\Theta_1$}\put(5.1,4){$\Theta_2$}
\put(5,3){$\Theta_{1,1}$}\put(4.5,2.7){$\Theta_{1,2}$}
\put(6.5,2.5){$\Theta_{2,1}$}\put(6,2.2){$\Theta_{2,2}$}
\put(10,5){$\Theta$ is the quotient surface}
\put(10,4.5){$\Theta_1$ is the resolution over $\gS_1$, a}
\put(10,4){rational ruled surface}
\put(10,3.5){$\Theta_2$ is the resolution over $\gS_2$, a}
\put(10,3){$g=2$ ruled surface}
\put(10,2){$\Theta_{i,j}$ are the exceptional divisors }
\put(10,1.5){from resolution of the $\fZ_9$ points}
\end{picture}\]
\caption{\label{figure2}--- The singular fiber of type IX$_1^*$ ---}
\end{figure}
Once again the monodromy is easily seen to satisfy $M^2=1$,
and the Euler number of
this singular fiber can be calculated as above; it is 18.
\smallskip\noindent{\it Case} XII$_3^*$
The ambient space is $\fP_{(1,2,18,42,63)}$, and the singular locus in this
space is
\begin{itemize} \item $\gS_1=\{z_1=z_2=0\}_{\fZ_3} \cong \fP_{(18,42,63)}
\cong \fP_{(1,1,3)}$. \item $\gS_2 =\{ z_1=z_5=0\}_{\fZ_2} \cong
\fP_{(2,18,42)} \cong \fP_{(1,3,7)}$. \item $\gS_3=\gS_1\cap \gS_2
=\{z_1=z_2=z_5=0\}_{\fZ_6} \cong \fP^1$.
\item $\gS_4=\{z_1=z_2=z_3=0\}_{\fZ_{21}} \cong \fP^1$.
\end{itemize}
The bad fiber is $C_{\infty}\cong \fP_{(2,18,42,63)}[126] \cong
\fP_{(1,3,7,21)}[21]$; it contains the two curves $X\cap \gS_1$ and $X\cap
\gS_2$, and these two curves intersect in just one point. There is a
further singular point $\gS_4\cap X$, which also lies on $\gS_1$ but not on
$\gS_2$. Hence we have the picture
\[\unitlength1cm
\begin{picture}(10,4)(0,.5)\put(0,0){\epsfbox{XII3star.eps}}
\put(3,4){$C_{\infty}$}\put(3,1){$C_0$}
\put(7,3.7){$\gS_2$, a $\fZ_2$ curve}
\put(6.2,2.5){$\gS_1$, a $\fZ_3$ curve}
\put(2.2,2.5){The $\fZ_6$ singular point}
\put(-2,3.4){The $\fZ_{21}$ singular point}
\end{picture}
\]
The $\fZ_6$-point is resolved just as above, with a single exceptional
divisor which is a $\fP^1$-section of the fibration. For the $\fZ_{21}$
point, we deduce from the fact that it lies on a $\fZ_3$ curve, that it is
of type ${1\over 21}(1,2,18)$. This yields $d_1=d_2=1,\ d_3=3$ and hence,
by the formula above, $v={23+5 \over 2} = 14$, of which five are on the
boundary of the triangle. As we have already mentioned, any decomposition
of the cone with this many vertices yields a smooth resolution of the
singular point. As described above, since the singular point
is on $\gS_1$, it meets every fiber, so one of the components is a
curve-section of the fibration. We choose the following decomposition:
\[ \epsfbox{triangle3.eps} \]
We have nine exceptional divisors, of which one is a section of the
fibration, the other eight belong to the singular fiber. After resolution
of singularities, our singular fiber thus looks as in Figure \ref{figure3}.
\begin{figure}
\[\unitlength1cm \begin{picture}(10,4.3)(2,-.2)
\put(0,0){\epsfbox{XII3stres.eps}}
\put(1.4,.65){$\Theta_{1,1}$}\put(.5,.75){$\Theta_{1,2}$}
\put(2.3,.2){$\Theta$} \put(.7,3){$C_0$} \put(5,3.6){$\Theta_9$}
\put(8,3.5){$\Theta_1-\Theta_8$}
\put(7.5,.5){$\Theta_{2,1}$}
\put(10,4){The fiber is the sum }
\put(10,3.5){$\Theta + \Theta_1+\ldots,
+\Theta_8+\Theta_{1,1}+\Theta_{2,1}$}
\put(10,3){where $\Theta$ is the quotient fiber}
\put(10,2.5){ $\cong
\fP_{(1,3,7,21)}[21]$, $\Theta_i,\ i=1,\ldots, 8$ }
\put(10,2){are the resolution divisors of }
\put(10,1.5){the $\fZ_{21}$ point, $\Theta_9$ the one which}
\put(10,1){is a section, $\Theta_{1,1}$ is rational ruled}
\put(10,.5){$\Theta_{2,1}$ is $g=6$ ruled.}
\end{picture}
\]
\caption{\label{figure3} --- The singular fiber of type XII$^*_3$ ---}
\end{figure}
Note the surface $\Theta_9$, which is a section of the fibration, not a
component of the singular fiber. This component of the resolution
corresponds to the circled vertex above, and meets all the other eight
components, while each of the other eight meets only two others, as drawn.
\subsection{Fibers analogous to Kodaira's type IV$^*$}
In this section we consider the cases 7-9 in our table. In all these cases,
the monodromy matrix is an element of order 3, so in a sense an analog of
Kodaira's type IV$^*$ type fiber.
\smallskip\noindent{\it Case } IV$^{**}_1$:
The ambient space is $\fP_{(1,3,4,8,8)}$, which has the following
singularities:
\begin{itemize}\item $\gS_1 = \{z_1=z_4=0\}_{\fZ_4}$, \item
$\gS_2=\{z_1=z_2=z_3=0\}_{\fZ_8}$.
\end{itemize}
For the intersections we have $\gS_1\cap X = \fP_{(4,8,8)}[24] \cong
\fP_{(1,2,2)}[6]\cong \fP_{(1,1,1)}[3]$, which is an elliptic curve. $\gS_2
\cap X = \fP_{(8,8)}[24]\cong \fP_{(1,1)}[3]$, which consists of three
points. So the singular locus of $X$ is a curve and three additional points
on that curve. The bad fiber is $C_{\infty} = \fP_{(3,4,8,8)} \cong
\fP_{(1,2,3,2)}[6]$, which is a double cover of the space $\fP_{(1,2,3)}$
branched over a sextic curve. Since the $\fZ_8$ points lie on a $\fZ_4$
curve, their type is ${1\over 8}(1,3,4)$, and we have $d_1=d_2=1,\ d_3=4$
so by our formula above we get $v={10+6 \over 2} =8$, of which 3 lie on the
edge (corresponding to the exceptional divisors of the $\fZ_4$-curve),
and we take the following cone decomposition:
\[ \epsfbox{triangle4.eps} \]
This means that at each of the three $\fZ_8$ points, we have two
exceptional divisors, one of which is a section of the fibration. In the
picture of the cone decomposition we have circled the two vertices which
correspond to exceptional divisors which are sections of the fibration. The
others then belong to the singular fiber, so we have a total of five
components in addition to the image of the K3 itself. After resolution of
singularities, we get the fiber in Figure \ref{figure4}.
\begin{figure}
\[ \unitlength1cm \begin{picture}(10,4)(2,0)
\put(-.5,0){
\put(0,0){\epsfbox{IVstarstar.eps}}
\put(8,2.5){$\Theta$}\put(6.5,1.2){$\Theta_{1,1}$}
\put(5.3,1.7){$\Theta_{1,2}$} \put(1.2,1.5){$\Theta_1$}
\put(2.8,1.3){$\Theta_2$} \put(4.3,1.2){$\Theta_3$}
}
\put(10,4){The fiber is the sum }
\put(10,3.5){$\Theta+\Theta_1+\Theta_2+\Theta_3+\Theta_{11}+\Theta_{1,2}$}
\put(10,3){where $\Theta$ is the quotient fiber }
\put(10,2.5){$\cong \fP_{(1,2,2,3)}[6]$, $\Theta_i,\ i=1,\ldots, 3$}
\put(10,2){are the resolution divisors of the }
\put(10,1.5){three $\fZ_8$ points, $\Theta_{1,i},\ i=1,2$}
\put(10,1){are the resolution divisors}
\put(10,.5){over the singular curve.}
\end{picture}
\]
\caption{\label{figure4} --- The singular fiber of type IV$^{**}_1$ ---}
\end{figure}
\bigskip\noindent{\it Case} IX$^{**}_1$:
The ambient space is $\fP_{(1,3,8,12,24)}$, which has as singular locus:
\begin{itemize}\item $\gS_1=\{z_1=z_2=0\}_{\fZ_4}$, \item
$\gS_2=\{z_1=z_3=0\}_{\fZ_3}$, \item $\gS_3=\gS_1\cap \gS_2
=\{z_1=z_2=z_3=0\}_{\fZ_{12}}$, \item $\gS_4=\{z_1=z_2=z_4=0\}_{\fZ_8}$.
\end{itemize}
The intersections with $X$ are $\gS_1\cap X = \fP_{(8,12,24)}[48]\cong
\fP_{(1,1,1)}[2]$, a rational curve, $\gS_2\cap X =\fP_{(3,12,24)}[48]\cong
\fP_{(1,1,2)}[4]$, an elliptic curve, $\gS_3\cap X = \fP_{(12,24)}[48]\cong
\fP_{(1,1)}[2]$, consisting of two points, and $\gS_4\cap X =
\fP_{(8,24)}[48]\cong \fP_{(1,1)}[2]$, again two points. Hence, before
resolution, our fiber $C_{\infty} = \fP_{(3,8,12,24)}[48] \cong
\fP_{(1,1,2,2)}[3]$ looks as follows:
\[ \unitlength1cm\begin{picture}(10,3.5)
\put(0,0){\epsfbox{IXstarstar.eps}}
\put(2,1){$C_0$}\put(2.5,2){$C_{\infty}$}
\put(0,2.5){$\gS_1 : \fZ_4$}
\put(1.1,3.3){$\gS_2 : \fZ_3$}
\put(2.8,2.7){$\fZ_{12}$}\put(6.2,2.6){$\fZ_{12}$}
\put(3.9,3.4){$\fZ_8$}\put(5.3,3.4){$\fZ_8$}
\end{picture}
\]
The situation at each of the $\fZ_8$ points is just as in the previous
example, as each lies again on a $\fZ_4$ curve. Hence the resolution of
these points introduces two exceptional divisors each, one of which is a
component of the singular fiber. Looking at the $\fZ_{12}$ points, we see
that as they are again on $\fZ_3$ curves, they are of type ${1\over
12}(1,2,9)$. Hence we have $d_1=1,\ d_2=2,\ d_3=3$ and the formula for the
number of vertices of a cone decomposition yields $v= {14+6\over 2}=10$
vertices, of which there are three on the boundary corresponding to the $\fZ_4$
curve, and two others on another boundary, corresponding to the $\fZ_3$
curve. Hence there are two vertices in the interior, and we have the cone
decomposition:
\[\epsfbox{triangle5.eps}\]
In total, in addition to the proper transform $\Theta$ of the singular
fiber $C_{\infty}$, we have two divisors over $\gS_1$ (the third is a
section of the fibration and not a component of the fiber), two over
$\gS_2$, and over each of the two $\fZ_8$ and two $\fZ_{12}$ points we also
have two exceptional divisors, one of which is not a section, hence a
component of the fiber. After resolution of singularities, we have the
picture of Figure \ref{figure5}.
\begin{figure}
\[\unitlength1cm\begin{picture}(10,4)(.7,0)
\put(-2,-.6){\epsfbox{IXstarstarres.eps}}
\put(6,2.5){$\Theta$}\put(4.7,2){$\Theta_{2,1}$}
\put(5,3.4){$\Theta_{2,2}$}\put(4,1.8){$\Theta_1$}
\put(1.5,1.8){$\Theta_3$}\put(.7,1.8){$\Theta_2$}
\put(2.5,1.9){$\Theta_4$}
\put(2.7,.6){$\Theta_{1,1}$}
\put(-1,1){$\Theta_{1,2}$}
\put(5.7,1.5){$\gS_1$}\put(3.7,.1){$\gS_2$}
\put(8,4){The singular fiber is the sum}
\put(8,3.5){$\Theta+\Theta_1+\Theta_2+\Theta_3+\Theta_4+$}
\put(8,3){$\Theta_{1,1}+\Theta_{1,2}
+\Theta_{2,1}+\Theta_{2,2}$}
\put(8,2.5){where $\Theta$ is the proper transform of}
\put(8,2){the surface $C_{\infty}$ and the $\Theta_i$}
\put(8,1.5){$i=1,\ldots,4$ are the resolution surfaces of }
\put(8,1){the two $\fZ_8$ points and the }
\put(8,.5){the two $\fZ_{12}$ points; the $\Theta_{1,i},\ i=1,2$}
\put(8,0){are rational ruled, the $\Theta_{2,i},\ i=1,2$}
\put(8,-.5){are elliptic ruled.}
\end{picture}
\]
\caption{\label{figure5}--- The singular fiber of type IX$^{**}_1$ ---}
\end{figure}
\bigskip\noindent{\it Case} XII$^{**}_3$:
The ambient space is $\fP_{(1,3,24,56,84)}$, which has the following
singular locus:
\begin{itemize}\item $\gS_1=\{z_1=z_2=0\}_{\fZ_4}$, \item
$\gS_2=\{z_1=z_4=0\}_{\fZ_3}$, \item $\gS_3=\{z_1=z_2=z_5=0\}_{\fZ_8}$,
\item $\gS_4=\{z_1=z_2=z_3=0\}_{\fZ_{28}}$, \item $\gS_5=\gS_1\cap \gS_2
=\{z_1=z_2=z_4=0\}_{\fZ_{12}}$.
\end{itemize}
For the intersections with $X$ we get $\gS_1\cap X\cong\fP_{(1,1,1)}[1]$ is
rational, $\gS_2\cap X\cong \fP_{(1,2,7)}[14]$, a $g=3$ curve, $\gS_3\cap
X$, $\gS_4\cap X$ and $\gS_5\cap X$ all consist of just a single point. The
singular fiber is $C_{\infty}\cong \fP_{(1,2,7,14)}[14]$, a rational
surface. This looks as follows:
\[ \unitlength1cm \begin{picture}(10,4)
\put(0,0){\epsfbox{XIIstarstar.eps}}
\put(6,3.9){$C_{\infty}$}\put(6.5,.5){$C_0$}
\put(3.7,3.6){$\fZ_8$}\put(6.5,2.3){$\fZ_{28}$}
\put(5,2.6){$\fZ_{12}$} \put(7.5,2.3){$\gS_1 : \fZ_4$}
\put(8.5,3.3){$\gS_2 : \fZ_3$}
\end{picture}
\]
The $\fZ_{12}$ point is just the same as above, yielding upon resolution
one additional component to the fiber. Similarly, the $\fZ_8$ point is the
same as above, yielding also one additional component to the fiber. It
remains to resolve the $\fZ_{28}$ point. Note that as it lies on a $\fZ_4$
curve, it is of type ${1\over 28}(1,3,24)$. We have $d_1=d_2=1,\ d_3=4$ and
for the number of vertices we get $v={30+6\over 2} =18$, of which only six
are on the boundary. Hence we must insert 12 vertices in the interior. We
choose the following cone decomposition:
\[ \epsfbox{triangle6.eps} \]
Once again, the two components of the resolution which give rise to
sections of the fibration instead of components of the singular fiber are
circled. We again see in the middle the eight components giving rise to an
$A_7$ configuration, but this time the component which meets all
eight is a component of the singular fiber instead of a section. In
addition, we have two more components, each of which meets the special
component and four of the components of the $A_7$ chain. After
resolution, the fiber looks as in Figure \ref{figure6}.
\begin{figure}
\[\unitlength1cm \begin{picture}(10,5)(3.5,0)
\put(0,0){\epsfbox{XIIstarstarres.eps}}
\put(2.7,1){$\Theta$}\put(1.5,1.2){$\Theta_{1,1}$}
\put(.9,1.8){$\Theta_{1,2}$}\put(7.5,1.3){$\Theta_{2,1}$}
\put(8.6,2.2){$\Theta_{2,2}$} \put(7.2,4.2){$\Theta_9$}
\put(3.7,2){$\Theta_{10}$}\put(5,2){$\Theta_{11}$}
\put(2.7,.25){$\Theta_{12}$}\put(3.2,1.5){$\Theta_{13}$}
\put(10,4){The singular fiber is the sum}
\put(10,3.5){$\Theta+\Theta_1+\cdots+\Theta_{13}+\Theta_{1,1}+ \Theta_{1,2}
+\Theta_{2,1} +\Theta_{2,2}$}
\put(10,3){where $\Theta$ is the proper transform of $C_{\infty}$}
\put(10,2.5){and $\Theta_i,\ i=1,\ldots, 11$ are the divisors}
\put(10,2){of the resolution of the $\fZ_{28}$ point}
\put(10,1.5){$\Theta_{12}$ is from the $\fZ_{12}$ point}
\put(10,1){$\Theta_{13}$ is from the $\fZ_{8}$ point}
\put(10,.5){$\Theta_{1,i},\ i=1,2$ are rational ruled}
\put(10,0){$\Theta_{2,i}\ i=1,2$ are $g=3$ ruled. }
\end{picture}
\]
\caption{\label{figure6} --- The singular fiber of type XII$^{**}_3$ --- }
\end{figure}
\subsection{An analog of Kodaira's type III$^*$ fiber}
We consider now the tenth case in the table. The monodromy matrix has order
four, so in this sense this is an analog of Kodaira's type III$^*$ fiber.
\smallskip\noindent{\it Case} IX$^{***}_1$:
The ambient space is $\fP_{(1,4,10,15,30)}$ with singular locus
\begin{itemize}\item $\gS_1=\{z_1=z_2=0\}_{\fZ_5}$, \item
$\gS_2=\{z_1=z_4=0\}_{\fZ_2}$, \item $\gS_3=\gS_1\cap \gS_2 =
\{z_1=z_2=z_4=0\}_{\fZ_{10}}$, \item $\gS_4 = \{
z_1=z_2=z_3=0\}_{\fZ_{15}}$.
\end{itemize}
For the intersections with $X$ we have $\gS_1\cap X \cong
\fP_{(1,1,1)}[2]$, a rational curve, $\gS_2\cap X \cong \fP_{(1,2,3)}[6]$,
an elliptic curve, and the two intersections $X\cap \gS_3$ and $X\cap
\gS_4$ both consist of two points. Hence the picture is just as in the case
IX$^{**}_1$ above, with the $\fZ_{12}$ points now replaced by $\fZ_{10}$
points, and the $\fZ_{8}$ points there replaced by $\fZ_{15}$ points
here. The resolution of the $\fZ_2$ curve is an elliptic ruled surface, a
component of the singular fiber, and the resolution of the $\fZ_5$ curve is
a union of four rational ruled surfaces, of which one is a section of the
fibration, while the other three give components of the singular fiber. So
we just have to resolve the $\fZ_{10}$ and $\fZ_{15}$ points. Note that
since both lie on a $\fZ_5$ curve, they are of types ${1\over 10}(1,4,5)$
and ${1\over 15}(1,4,10)$ (or $(2,3,x)$, it doesn't matter). Hence we have
$d_1=1,\ d_2=2,\ d_3=5$ in the first and $d_1=d_2=1,\ d_3=5$ in the second
case. Hence we have $v={12+8\over 2}=10$ vertices in the first case and
$v={17+7 \over 2}=12$ vertices in the second. We take the two following
cone decompositions:
\[\epsfbox{triangle7.eps}\hspace*{1cm}\]
From the singular curves we get $1+4=5$ components, from each $\fZ_{10}$
point an additional one and at each $\fZ_{15}$ point, we get four further
compoenents. Hence the singular fiber has a total of 15 components, in
addition to $\Theta$, the proper transform of $C_{\infty}$.
After resolution of singularities, the singular fiber looks as in Figure
\ref{figure7}.
\begin{figure}
\[ \unitlength1cm\begin{picture}(10,9.2)
\put(-3,0){\epsfbox{IXstarstarstarres.eps}}
\put(3,9.2){$\Theta_{1,1}-\Theta_{1,4}$}
\put(3.6,4.5){$\Theta_{3,1}-\Theta_{3,4}$}
\put(4.6,5.3){$\Theta_{4,1}-\Theta_{4,4}$}
\put(2.5,.5){$\Theta_1$}\put(7.5,5.2){$\Theta_2$}
\put(4,2.8){$\Theta_{2,1}$} \put(2.5,2){$\Theta$}
\put(4.5,8.5){The singular fiber is the union of 16 components}
\put(4.5,8){$\Theta+\Theta_{1,1}+\cdots+\Theta_{1,4}+\Theta_{2,1}
+\Theta_1+\Theta_2+$}
\put(5,7.5){$\Theta_{3,1}+\cdots +\Theta_{3,4}+\Theta_{4,1}+\cdots +
\Theta_{4,4}$}
\put(6,7){where $\Theta$ is the proper transform of $C_{\infty}$, and}
\put(7.5,6.5){$\Theta_{1,i},\ i=1,\ldots, 4$ are the}
\put(9,6){components resolving the}
\put(9.5,5.5){$\fZ_5$ curve, $\Theta_{2,1}$ resolving}
\put(9.5,5){the $\fZ_2$ curve}
\put(9.5,4.5){$\Theta_{i,j}, i=3,4,\ j=1,\ldots, 4$}
\put(9.5,4){are the divisors }
\put(9,3.5){resolving the $\fZ_{15}$ points,}
\put(8.5,3){$\Theta_i,\ i=1,2$}
\put(8,2.5){resolving the $\fZ_{10}$ points.}
\end{picture}
\]
\caption{\label{figure7} --- The singular fiber of type IX$^{***}_1$ ---}
\end{figure}
\subsection{Analogs of Kodaira's type II$^*$ fiber}
{\it Case} IX$^{****}_1$:
The ambient space is $\fP_{(1,6,14,21,42)}$, which has the following
singular locus:
\begin{itemize}\item $\gS_1=\{z_1=z_2=0\}_{\fZ_7}$, \item
$\gS_2=\{z_1=z_3=0\}_{\fZ_3}$, \item $\gS_3=\{z_1=z_4=0\}_{\fZ_2}$,
\item $\gS_4=(\gS_1\cap \gS_2)_{\fZ_{21}}$, \item $\gS_5=(\gS_1\cap
\gS_3)_{\fZ_{14}}$, \item $\gS_6=(\gS_2\cap \gS_3)_{\fZ_6}$.
\end{itemize}
The intersections $\gS_i\cap X$ are all rational curves, which meet two at
a time. We have the following configuration in $C_{\infty}$:
\[\unitlength1cm
\begin{picture}(10,3.5)
\put(0,0){ \epsfbox{IVstarstarstar.eps} }
\put(1.8,2.3){$\fZ_6$}\put(3.3,2.3){$\fZ_6$}
\put(1.2,3.3){$\gS_2 -\ \fZ_3$} \put(3,-.3){$\gS_3 -\ \fZ_2$}
\put(7,-.3){$\gS_1 -\ \fZ_7$}
\put(7.5,1){$C_{\infty}$} \put(5.4,1.4){$\fZ_{14}$}
\put(6.1,.8){$\fZ_{14}$}
\put(3.7,2.8){$\fZ_{21}$} \put(5,2.3){$\fZ_{21}$}
\end{picture}
\]
\begin{figure}
\[ \unitlength1cm \begin{picture}(10,5) \put(-3,0){\epsfbox{IX4starres.eps}}
\put(2.7,1){$\Theta_{5,2}$}\put(4.5,1){$\Theta$}
\put(1.85,1.4){$\Theta_{5,1}$} \put(.5,1){$\Theta_{5,4}$}
\put(.7,1.7){$\Theta_{5,3}$} \put(0,1.8){$\Theta_{3,1}$}
\put(-.7,2.2){$\Theta_{4,1}$} \put(-1.4,2.7){$\Theta_{4,2}$}
\put(1.6,3.9){$\Theta_{2,1}$} \put(.7,4.1){$\Theta_{2,2}$}
\put(3.8,2.2){$\Theta_{4,1}-\Theta_{4,5}$}
\put(2.5,5.7){$\Theta_{4,6}-\Theta_{4,10}$}
\put(7.9,1){$\Theta_{1,1}-\Theta_{1,6}$}
\put(7.9,.5){$\Theta_{1,6}$ is a section, not a component}
\end{picture}
\]
\caption{ {\label{figure8}--- The singular fiber of type IX$^{****}_1$ ---} }
\medskip
{The singular
fiber is a sum of 26 components. $\Theta$ is the proper transform of
$C_{\infty}$, and for the exceptional divisors we have choosen the
notation so that the divisors $\Theta_{i,j}$ resolve the singular locus
$\gS_j$. There are then $5+2+1+10+4+2 = 25$ components of the various loci. }
\end{figure}
\smallskip
The $\fZ_6$ points are resolved precisely as in the cases above; there is
one exceptional divisor, which this time is a component of the singular
fiber, as it is not contained in all fibers, but only in $C_{\infty}$. It
remains to resolve the $\fZ_{14}$ and $\fZ_{21}$ points. They are of types
${1\over 14}(1,6,7)$ and ${1\over 21}(1,6,14)$, and we have $d_1=1,\
d_2=2,\ d_3=7$ in the first case and $d_1=1,\ d_2=3,\ d_3=7$ in the second
case. By our formula above, this means we have $v=13$ and $v=17$,
respectively, leading to 3 and 6 inner vertices, respectively. We choose
the following cone decompositions:
\[ \epsfbox{triangle8.eps} \]
Again the vertices corresponding to the sections of the fibration are
circled. Just as above, from this we can without difficulty derive the
singular fiber. It will look as in Figure \ref{figure8}.
\smallskip\noindent{\it Case} XII$^{***}_3$:
The ambient projective space is $\fP_{(1,6,42,98,147)}$ with singular locus
\begin{itemize}\item $\gS_1=\{z_1=z_2=0\}_{\fZ_7}$, \item
$\gS_2=\{z_1=z_4=0\}_{\fZ_3}$, \item $\gS_3=(\gS_1\cap \gS_2)_{\fZ_{21}}$,
\item $\gS_4=\{z_1=z_2=z_3=0\}_{\fZ_{49}}$.
\end{itemize}
The intersections $\gS_i\cap X,\ i=1,2$ are both rational curves, which
meet in a single point. Furthermore, $\gS_4\cap X$ consists also of a
single point. One sees easily that the $\fZ_{21}$ point is resolved exactly
as above in the case IX$^{***}_1$. It remains to resolve the $\fZ_{49}$
point. Since this lies on a $\fZ_7$ curve, the singularity is of the type
${1\over 49}(1,6,42)$, and we have $d_1=d_2=1,\ d_3=7$. The number of
vertices is then ${51+9\over 2} = 30$, of which $3+6$ are on the
boundary. It follows that we have to include $21$ inside vertices. We take
the cone decomposition of Figure \ref{figurecone}.
\begin{figure}
\[\epsfbox{triangle9.eps} \]
\caption{\label{figurecone} The cone decomposition for the case
XII$^{***}_3$}
\end{figure}
The singular fiber will look quite a bit like that of type XII$_3^{**}$,
but will have nine additional components. This will look as displayed in
Figure \ref{figure9}.
\begin{figure}
\[
\unitlength1cm \begin{picture}(10,8.5)
\put(-2,0){\epsfbox{XII3starres.eps}}
\put(2,.5){$\Theta$}
\put(1.5,1.3){$\Theta_{1,1}$}
\put(.8,2){$\Theta_{1,2}$}
\put(.7,2.8){$\Theta_{1,3}$}
\put(0,3.5){$\Theta_{1,4}$}
\put(0,4.4){$\Theta_{1,5}$}
\put(0,5.8){$\Theta_{1,6}$ (a section)}
\put(8.5,5){$\Theta_{4,9}$}
\put(3,7.5){$\Theta_{4,1}-\Theta_{4,8}$}
\put(8,3.8){$\Theta_{4,12}-\Theta_{4,14}$}
\put(8,3.3){and $\Theta_{4,15}-\Theta_{4,17}$}
\put(6.4,4.6){$\Theta_{4,18}$}
\put(4.35,.5){$\Theta_{3,1}$}
\put(5.1,.7){$\Theta_{3,2}$}
\put(5.8,.3){$\Theta_{3,3}$}
\put(6.4,.5){$\Theta_{3,4}$}
\put(7.4,1.6){$\Theta_{3,5}$}
\put(3.8,3.3){$\Theta_{4,10}$}
\put(5.3,3.3){$\Theta_{4,11}$}
\put(8.7,1.1){$\Theta_{2,1}$}
\put(10.3,2.5){$\Theta_{2,2}$}
\end{picture}
\]
\caption{\label{figure9}--- The singular fiber of type XII$_3^{***}$ ---}
\medskip
The singular fiber is the union of 33 components. $\Theta$ is the proper
transform of $C_{\infty}$, and again the components $\Theta_{i,j}$ are the
exceptional divisors resolving $\gS_i$. Of the 20 components resolving the
$\fZ_{49}$ point, we have not drawn two of them, which correspond to the
vertices labeled with a square in the above cone decomposition, as they
would have cluttered up the picture too much.
\end{figure}
\subsection{The other cases}
We now consider the last three cases in Table 3. The difference between
these and the above cases is that there are now two bad fibers of $^*$
type. These are the fibers which are the total transforms of the surfaces
$C_{\infty}$ and $C_0$. The fiber $C_0$ is now also singular because of the
fact that the first weight is no longer unity. A detailed analysis is not
necessary. Consider case 13, i.e., the projective space
$\fP_{(2,3,5,10,10)}$. The surface $C_{\infty}$ is
$\fP_{(2,5,10,10)}[30]\cong\fP_{(1,1,1,1)}[3]$, a cubic surface, and the
two singular curves on it are elliptic and meet in three points. Without
difficulty we recognize the singular fiber of type IV$^{*}_1$. The surface
$C_0$ is $\fP_{(3,5,10,10)}[30]\cong \fP_{(1,2,2,3)}[6]$, and we recognize
the singular fiber of type IV$^{**}_1$. Note that the singular curve
$\gS_1-\{z_1=z_2=0\}$ is $\fZ_5$, and yields upon resolution four components,
one of which is a section of the fibration. The other three split; one of
the components belongs to the fiber at $\infty$, while the other belongs to
the fiber of type IV$^{**}_1$, and indeed, the first has one, the second
has two such components. Similarly, one can check that the exceptional
divisors which lie over the singular points split, one component is a
section, the others belong to one or the other fiber.
The same methods apply to the remaining cases, and the results are listed
in Table 3.
|
\section{Quaternion Quantum Mechanics}
After six decades, quaternion quantum mechanics is coming out of age. The
earliest known reference on a possible generalization of quantum theory with
respect to the background field dates back to 1934 with the paper by Jordan et
all \cite{Jordan}. The use of quaternions properly was proposed by Birkoff
and von Neumann in 1936, later developed by Finkelstein \cite{Fink}, and more
recently by Adler and others \cite{Adler}.
The original motivation for quaternion quantum mechanics was formal:
The propositional calculus implies that it is possible to represent the pure
states of a quantum system by rays on a Hilbert defined on any associative
division algebra. This includes the quaternion algebra as the most general
case. Contrasting with this, the next division (but non associative) algebra,
the octonion algebra, has been always associated with physical arguments,
notably in connection with the $SU(3)$ gauge symmetry and strong interactions
\cite{Pais,Tiomno,Oliveira,Brumby}.
In its essence, quaternion quantum mechanics is a modification of the
complex quantum theory, in which the wave functions belong to a Hilbert
space defined over the quaternion field. As a physical theory, it should prove
to be effective at some high energy level, exhibiting experimental evidence
which would distinguish it from the complex theory \cite{Peres}.
As in all Clifford algebras, it is possible to represent the quaternion algebra as a tensor
product of two independent complex algebras. Therefore, in principle we could
write quaternion functions with complex components and use two
independent expressions of complex analyticity. However this does not mean
that complex concept of analyticity extends trivially to quaternions.
This paper has two purposes. One of them is to show
by a simple argument that the combined spin and isospin states
leads to quaternion quantum mechanics, and vice versa, that the
algebraic consistency of quaternion quantum mechanics necessarily implies in
the existence of an intrinsic angular momentum representing the combined spin
and isospin at the level of an unbroken $SU(2)$ gauge theory.
The second and more formal topic is a consequence of the first:
The emergence of a quaternionic Hilbert space requires
the solution of differential equations involving functions of a quaternionic
variable, quaternionic Fourier expansions and quaternionic phase
transformations.
We will see that the traditional complex notion of analyticity characterized
by direction independent derivatives does not necessarily apply, so that
quaternionic quantum states may exhibit different behavior along different
directions in space.
\section{Isospin and Quaternions}
The standard textbook explanation on why quantum mechanics should be defined
over the complex field is based on the double slit experiment, together
with the complex phase difference for the wave functions.
However, this can also
be explained by a real quantum theory provided a special operator
such that $J^{2}=1$ and $J^{T}=-J$ is introduced \cite{Fink}.
Although most people would agree that this is equivalent to quantum mechanics
over the complex field, a more definitive argument for the complex algebra
comes from the spin. The existence and classification of the spinor
representations of the rotation subgroup of the Lorentz group demands a
solution of quadratic algebraic equations for the eigenvalues of the
invariant operators. This can be guaranteed only within the complex field.
The bottom line is, complex quantum mechanics is a requirement
of the spin and its spinor structure \cite{Chevalley}.
Following a similar argument we may convince ourselves
that quaternion quantum mechanics is an algebraic requirement of the spin
together with the isospin and the associated spinor structures. In fact, while the
spin is associated with the spinor representations of the $SO(3)$ subgroup
of the Lorentz group, the
isospin is given by a representation of the gauge group $SU(2)$.
Actually the two groups are isomorphic and they have
equivalent representations given by the Pauli matrices (they are
identical representations, distinguished only by
different notations), acting on independent spinor spaces.
In the case of the unbroken $SU(2)$ gauge theory,
spin and isospin are present in a combined symmetry scheme, so that
the total spinor space is a direct sum of the spinor space
and the isospinor spaces: $K=I\oplus J $.
The spinor space is represented as
a complex plane (a Gauss plane) generated by the basis
\[
\mbox{I= (spinor space) }: \left\{
1= \left( \begin{array}{l} 1\\ 0 \end{array} \right),
i= \left( \begin{array}{l} 0\\ 1 \end{array} \right)
\right\}
\]
and the isospinor space represented by another complex
plane generated by
\[
\mbox{J=(isopinor space) }: \left\{
1= \left( \begin{array}{l} 1\\ 0 \end{array} \right),
j= \left( \begin{array}{l} 0\\ 1 \end{array} \right)
\right\}
\]
Thus, $I$ and $J$ can be taken as two independent Gauss planes sharing the
same real unit $1$, but with different and independent imaginary units $i$ and $j$
respectively.
As long as the isospin symmetry and the spinor
representation of the Lorentz group remain combined, the corresponding
angular momenta add up to generate a total angular momentum
represented in the direct sum of the two spinor spaces.
On the other hand, it has been argued that when this combined symmetry is broken
the $SU(2)$ degree of freedom reappears as a spin degree of freedom
\cite{Jackiw}. Following these ideas a fermionic state is derived from
a magnetic monopole model in four dimensions associated with an $SU(2)$
soliton \cite{Hooft,Vachaspati,Singleton,Emch}.
The question we address here concerns with the algebraic consistency of
the combined spinor-isospinor symmetry as it remains unbroken. In this case,
we end up with the total spinor space generated by the hypercomplex basis
$\{ 1,i,j\} $. According to Hamilton this direct sum is
algebraically consistent only if a third imaginary unit $k$ such that
$k=ij$ is introduced. That is, we can only close the algebra if
a third complex plane
\[\mbox{K=( new spinor space)} :\left\{
1= \left( \begin{array}{l} 1\\ 0 \end{array} \right),
k= \left( \begin{array}{l} 0\\ 1 \end{array} \right)
\right\}
\]
is introduced. We conclude that an additional spin-half field should also be
present. This new spinor may be the
generator of the Jackiw-Rebbi spin degree of freedom after the symmetry is
broken. However, this is not all. When the combined symmetry remains
unbroken the three spinors produce a full quaternion algebra with basis
$\{1,i,j,k\}$, such that its group of automorphisms
carry the combined symmetry. If we add to the space of these quaternion
wave functions a Hilbert product compatible with the quaternion algebra we
obtain a quaternionic Hilbert space.
This space should reduce to the usual Hilbert space of
complex quantum mechanics with separate spin, isospin plus one extra spinor
degree of freedom at the level of the combined symmetry breaking.
In conclusion, quaternion quantum mechanics appears as consistent condition
of the combined spin and isospin symmetries.
A possible relation between quaternions and the isotopic spin was suggested
by C. N. Yang \cite{Fink} and by E. J. Schremp \cite{Schremp}.
However, their basic arguments are distinct from the ones based on the
combination of symmetries.
When the two spinor spaces are taken together, they give way to a quaternionic
spin operator given by a $2\times 2$
matrix representation of the quaternion algebra given by the Pauli matrices
\begin{eqnarray*}
\sigma^{0}=\left(
\begin{array}{cc}
1 & 0\\
0 & 1
\end{array}
\right),\;\;
\sigma^{1}=\left(
\begin{array}{cc}
0 & 1\\
1 & 0
\end{array}
\right),
\\
\sigma^{2}=\left(
\begin{array}{cc}
0 & -i\\
i & 0
\end{array}
\right),\;\;
\sigma^{3}=\left(
\begin{array}{cc}
1 & 0\\
0 & -1
\end{array}
\right)
\end{eqnarray*}
This has a one to one correspondence with the quaternion
algebra\footnote{ Greek indices are space-time indices and they run from 0 to 3. Small case
Latin indices run from 1 to 3. Capital Latin indices are spinor indices running
from 1 to 2. The quaternion multiplication table is taken to be
\[
e_{i}e_{j} = -\delta_{ij} +\sum\epsilon_{ijk}e_{k},\;\;\;
e_{i}e_{0} = e_{0}e_{i} =e_{i}
\]
The conjugate of a quaternion $X$ is $\bar{X}$, with
$\bar{e}_{i}=-e_{i},\;\; \bar{e}_{0}=e_{0}$. The quaternion norm is
$|X|^{2}=\sum (X^{\alpha})^{2}$ and the inverse of $X$ is
$X^{-1}=\bar{X}/ |X| $.}
whose elements are the quaternions
\begin{equation}
\Psi = \sum_{\alpha=0}^{3} \Psi_{\alpha}\sigma^{\alpha}
=\left(
\begin{array}{cc}
\Psi_{0}+\Psi_{3} & \Psi_{1} - i\Psi_{2}\\
\Psi_{1}+ i\Psi_{2} & \Psi_{0}-\Psi_{3}
\end{array}
\right), \label{eq:Q}
\end{equation}
As it is well known, the above matrix corresponds to a
vector field in space-time, associated to a pair of two component
spinors $\Psi^{\alpha} =\sigma^{\alpha}_{AB}\xi^{A}{\bar{\zeta}}^{B}$.
A possible interpretation for this pair of spinors is
given by the spin and isospin states in a $SU(2)$ model
\cite{Hooft,Vachaspati,Singleton,Emch}.
Since the inner automorphism of the quaternion algebra
correspond to the isometries of space-time, the existence of a combined
spin-isospin structure also have implications on the classical notion of derivative
of a function. In ordinary complex analysis
the derivative of a function does not depend on the direction in the complex
plane along which the limit is taken. This has to be so because the complex
plane is generated by only one real and one imaginary direction, leading to
the Cauchy-Riemann conditions for analyticity.
On the other hand, in the the quaternionic case there are three imaginary
directions generating a space that is isomorphic to $I\!\!R^{3}$,
where in principle there is no reason for the derivatives to be all equal.
This means that the properties of differentiable equations
involving quaternion functions of a quaternionic variable does not necessarily
coincide with those of ordinary complex quantum theory. Fortunately this can
be examined through the methods of classical analysis.
\section{Analysis of Quaternion Fields}
The earliest known study on the analysis of quaternion functions
using the same concepts of complex analysis
was made by Fueter in 1932, finding very restrictive generalizations of
the Cauchy-Riemann conditions \cite{Fueter}. Some alternative
criteria for defining quaternion analyticity have been suggested
\cite{Ketchum,Ferraro,Maia,Nash,Gursey,Khaled}, but to date there is not a
consensus on what is meant by quaternion analyticity.
To understand the nature of the difficulties we need to start from the basic
principles.
Denoting a generic quaternion function by $ f(X)=\sum
U_{\alpha}(X)e^{\alpha}$, and $\Delta f= [f(X+\Delta X) -f(X)]$ we may define
its left derivative as
\[
f'(X) = \mbox{lim}_{\Delta X\rightarrow 0}\delta f(X) (\Delta X)^{-1},
\]
and the right derivative as
\[
'f(X) = \mbox{lim}_{\Delta X\rightarrow 0} (\Delta X)^{-1}\Delta f(X)
\]
where the limits are taken with $|\Delta X|\rightarrow 0$
along the direction of the four-vector $\Delta X$ which depends on the
3-dimensional vector $\vec{\Delta X}$.
To compare with the complex case,
we may use the exponential form of a quaternion:
A vector of $I\!\!R^{3}$, $\xi =\sum X_{i}e^{i}$
associate a quaternion $\xi$ with norm
$\vert \xi\vert^{2}=\xi\bar{\xi}=\sum X_{i}^{2}$ and square $\xi^{2}
=-\xi\xi=-\sum X_{i}^{2}$.
Defining the unit quaternion (iota)
$\iota ={\xi}/{\sqrt{\xi\bar{\xi}}} $ such that $\iota^{2}=-1$,
we may construct a Gauss plane generated by $\iota$ and the quaternion unit
$e^{0}$. In this plane a quaternion $X=X_{0}e^{0}+\sum X_{i}e^{i}$ can be
expressed as $ X=\vert X\vert (cos\gamma +\iota\sin\gamma) $,
with $tan\; \gamma = {X_{0}}/{\sqrt{\sum X_{i}^{2}}}$.
Therefore, if we define the quaternion
exponential as
\[
exp(\iota\gamma)= e^{\iota\gamma} =cos\gamma +\iota sin\gamma,
\]
then we may express
$\Delta X=\vert\Delta X\vert e^{\iota\gamma}\;\;\mbox{and}\;\;\Delta
X^{-1}={ e^{-\iota\gamma}}/{\vert\Delta X\vert}$, where the three dimensional
direction is included in the definition of $\iota$.
Contrasting with the complex case, we cannot neglect the phase factor
$e^{\iota \gamma}$ during the limiting process because we have a functions
depending on three variables. The independence of the
limit with the phase is a privilege (or rather a limitation) of complex
theory. Furthermore, here we have the added complication that the left and
right derivatives do not necessarily coincide.
To see the consequences of this, consider the derivatives of a quaternion
function $f(X)$ along a fixed direction $\Delta X =\Delta X_{\beta} e^{\beta}
$ (no sum), indicated by the index within parenthesis
\begin{eqnarray}
f'(X)_{(\beta)} = \frac{\partial U_{0}}{\partial X_{\beta}}e^{0}
(e^{\beta})^{-1} +
\sum_{i}\frac{\partial U_{i}}{\partial X_{\beta}}e^{i}(e^{\beta})^{-1}\\
'f(X)_{(\beta)} = \frac{\partial U_{0}}{\partial X_{\beta}}
(e^{\beta})^{-1}e^{0} +
\sum_{i}\frac{\partial U_{i}}{\partial X_{\beta}}(e^{\beta})^{-1}e^{i}
\end{eqnarray}
From direct calculation we find that
\begin{eqnarray*}
f'(X)_{(0)}& =& \, 'f(X)_{(0)}\\
f'(X)_{(j)}&=& \, 'f(X)_{(j)}
-2\sum_{i,k}\epsilon^{ijk}\frac{\partial U_{i}}{\partial X_{j}} e^{k}
\end{eqnarray*}
Imposing that these derivatives are selectively equal,
four basic classes of complex-like analytic functions are obtained
\begin{description}
\item[ Class A)]
Right analytic functions
\begin{equation}
\left.
\begin{array}{ll}
\! f'(X)_{(0)} \! = f'(X)_{(i)}\\
\! f'(X)_{(i)} \! = f'(X)_{(j)}
\end{array}
\right \}
\Rightarrow
\left\{
\begin{array}{ll}
\frac{\partial U_{\alpha}}{\partial X_{\alpha}} = \frac{\partial U_{\beta}}{\partial X_{\beta}}\\
\frac{\partial U_{i}}{\partial X_{0}} = -\frac{\partial U_{0}}{\partial X_{i}}\\
\frac{\partial U_{i}}{\partial X_{j}} = \sum\epsilon^{ijk}\frac{\partial U_{k}}{\partial X_{0}}
\end{array}
\right.
\label{eq:A}
\end{equation}
\item[Class B)] Left analytic functions
\begin{equation}
\left.
\begin{array}{ll}
\! 'f(X)_{(0)}\! = \, 'f(X)_{(i)}\\
\! 'f(X)_{(i)}\! =\, 'f(X)_{(j)}
\end{array}
\right\}
\Rightarrow
\left\{
\begin{array}{ll}
\frac{\partial U_{\alpha}}{\partial X_{\alpha}}=\frac{\partial U_{\beta}}{\partial X_{\beta}}\\
\frac{\partial U_{i}}{\partial X_{0}}=-\frac{\partial U_{0}}{\partial X_{i}}\\
\frac{\partial U_{i}}{\partial X_{j}}=-\sum\epsilon^{ijk}\frac{\partial U_{k}}{\partial X_{0}}
\end{array}\right.
\label{eq:B}
\end{equation}
\item[Class C)] Left-right analytic functions
\begin{equation}
f'(X)_{(\alpha)}= \, 'f(X)_{(\alpha)}
\Rightarrow
\left\{
\begin{array}{ll}
\frac{\partial U_{\alpha}}{\partial X_{\alpha}}=\frac{\partial
U_{\beta}}{\partial X_{\beta}}\\
\frac{\partial U_{i}}{\partial X_{j}}=-\frac{\partial U_{j}}{\partial X_{i}}\\
\frac{\partial U_{i}}{\partial X_{0}}=-\frac{\partial U_{0}}{\partial X_{i}}\;\;\;\;
\end{array}
\right.
\label{eq:C}
\end{equation}
\item[Class D)] The total analytic functions
\begin{equation}
\left.
\begin{array}{lll}
\! f'(X)_{(\alpha)}\! =\! f'(X)_{(\beta)}\\
\! 'f(X)_{(\beta)}\! =\! \, 'f(X)_{(\alpha)}\\
\! 'f(X)_{(\alpha)}\!=\! f'(X)_{(\alpha)}
\end{array}
\right\}
\Rightarrow
\left\{
\begin{array}{ll}
\frac{\partial U_{\alpha}}{\partial X_{\alpha}}=\frac{\partial U_{\beta}}{\partial X_{\beta}}\\
\frac{\partial U_{\alpha}}{\partial X_{\beta}}= 0\;\; \alpha\neq \beta
\end{array}
\right.
\label{eq:D}
\end{equation}
\end{description}
As we can see, these conditions are very restrictive, specially when we
consider the applications to quantum mechanics, suggesting
the adoption of different criteria for analyticity.
\section{ Harmonicity}
Consider the operator
$\partial\!\!\! \slash =\sum e^{\alpha}\partial_{\alpha}=\sum e^{\alpha}{\partial}/{ \partial X_{\alpha}}$
acting on the right and on the left of a function $f(X)$
\begin{eqnarray}
\partial\!\!\! \slash f(X) & = & (\frac{ \partial U_{0}}{ \partial X_{0}} +
\sum \frac{\partial U_{i}}{\partial X_{0}}e^{i})\nonumber\\
& + & \sum[\frac{\partial U_{0}}{\partial X_{i}}e^{i}
- \sum \frac{\partial U_{i}}{\partial X_{j}}(\delta^{ij}
-\epsilon^{ijk}e^{k})] \nonumber\\
& = & 'f(X)_{(0)} + \sum_{j}\, 'f(X)_{(j)}\vspace{3mm}\\
f(x)\partial\!\!\! \slash & = & (\frac{\partial U_{0}}{\partial X_{0}} +\sum \frac{\partial
U_{i}}{\partial X_{0}}e^{i}) \nonumber\\
& + & \sum[\frac{\partial U_{0}}{\partial X_{i}}e_{i}
-\sum \frac{\partial U_{i}}{\partial X_{j}
}(\delta^{ij}+ \epsilon^{ijk}e^{k})]\nonumber\\
& = & f'(X)_{(0)} - \sum_{j} f'(X)_{(j)}
\end{eqnarray}
Using these results four new classes of quaternion functions can be defined:
\begin{description}
\item[Class E)]
The functions such that
\begin{equation}
\partial\!\!\! \slash f(X)= f(X)\partial\!\!\! \slash \Rightarrow
\left\{
\begin{array}{ll}
\frac{\partial U_{i}}{\partial X_{j}}= \frac{\partial U_{j}}{\partial X_{i}}
\end{array}
\right.
\label{eq:E}
\end{equation}
\item[Class F)] The left harmonic functions
\begin{equation}
\partial\!\!\! \slash f(X)\!=\! 0 \Rightarrow
\left\{
\begin{array}{ll}
\frac{\partial U_{0}}{\partial X_{0}}=\sum_{i} \frac{\partial U_{i}}{\partial X_{i}}\\
\frac{\partial U_{k}}{\partial X_{0}} +\frac{\partial U_{0}}{\partial
X_{k}}\!=\! \sum_{ij}\epsilon^{ijk}\frac{\partial U_{i}}{\partial X_{j}}
\end{array}\right.
\label{eq:F}
\end{equation}
\item[Class G)] The right harmonic functions
\begin{equation}
f(X)\partial\!\!\! \slash \!=0\! \Rightarrow\!
\left\{
\begin{array}{ll}
\frac{\partial U_{0}}{\partial X_{0}}=\sum_{i} \frac{\partial U_{i}}{\partial X_{i}}\\
\frac{\partial U_{k}}{\partial X_{0}} +\frac{\partial U_{0}}{\partial
X_{k}}\!=\!-\!\sum_{ij}\epsilon^{ijk}\frac{\partial U_{i}}{\partial X_{j}}
\;\;\; \end{array}\right.
\label{eq:G}
\end{equation}
\item[Class H)] The left and right harmonic functions
\begin{equation}
\partial\!\!\! \slash f(X)\!=\! 0\;\mbox{and}\; f(X)\partial\!\!\! \slash \! =\!0\! \Rightarrow
\left\{
\begin{array}{ll}
\frac{\partial U_{0}}{\partial X_{0}}=\! \sum_{i}\frac{\partial U_{i}}{\partial X_{i}}\\
\frac{\partial U_{i}}{\partial X_{0}}= \! -\!\frac{\partial U_{0}}{\partial X_{i}}\\
\frac{\partial U_{i}}{\partial X_{j}}= \frac{\partial U_{j}}{\partial X_{i}}\;\;\;\;\;
\end{array}
\right. \label{eq:H}
\end{equation}
\end{description}
Notice that for the classes F, G and H we have
\[
\sum\delta^{ij}\frac{\partial^{2} U_{0}}{\partial X_{i}\partial
X_{j}}+\frac{\partial^{2} U_{0}}{\partial X_{0}^{2}}=\Box^{2} U_{0}=0
\]
where $\Box^{2}=\partial\!\!\! \slash \bar{\partial\!\!\! \slash }$. Similarly, $\Box^{2}U_{k}=0$,
so that those classes describe harmonic functions in the sense that
$\Box^{2} f(X)=0$.
A non trivial example of class H quaternion function is given by a
instantons field expressed in terms of quaternions \cite{Atiah}. The
connection of an anti self dual $SU(2)$ gauge field is given by the form
\begin{equation}
\omega =\sum_{\alpha}A_{\alpha(X)}dx^{\alpha} \label{eq:INSTANTON}
\end{equation}
where $A_{0}=\sum U_{k}e^{k}$ and $A_{k}=U_{0}e^{k}
-\epsilon_{ijk}U_{i}e^{j} $ and where
\[
U_{0}= \frac{\frac{1}{2} X_{0}}{1+|X|^{2}},\;\;\;
U_{i}=\frac{-\frac{1}{2}X_{i}}{1+|X|^{2} }
\]
are the components of the quaternion function $f(X)=U_{\alpha}e^{\alpha}$.
We can easily see that $f(X)$ satisfy the conditions \rf{H}
in the region of space-time defined by $\sum X_{i}^{2}=-2X_{0}$. In fact, this
is a particular case of a wider class of quaternion functions with
components $U_{\alpha} =g_{\alpha}(X)/(1 +|X|^{2})$, where $g_{\alpha}(X)$ are
some real functions. The case of instantons correspond to the choice
$g_{0}=\frac{-1}{2}
\sum X_{i}^{2}$ and $g_{i} =\frac{\partial g_{0}}{\partial X_{i}}$.
It is also interesting to notice that the anti instantons do not belong to
the same class of analyticity as the instantons.
\section{Integral Theorems}
Given a quaternion function $f(X)$ defined on a orientable 3-dimensional
hypersurface
$S$ with, unit normal vector $\eta$ we may define two integrals
\[
\int_{S} f(X) dS_{\eta},\;\,\mbox{and}\,\;\int_{S} dS_{\eta} f(X)
\]
where $dS_{\eta}=\sum dS_{i}e^{i}$ denotes the quaternion hypersurface
element with components
\begin{eqnarray*}
dS_{0}=dX_{1}dX_{2}dX_{3},\;\; dS_{1}=dX_{0}dX_{2}dX_{3},\\
dS_{2}=dX_{0}dX_{1}dX_{3}, \;\; dS_{3}=dX_{0}dX_{1}dX_{2}.
\end{eqnarray*}
On the other hand,
denoting by $dv=dX_{0}dX_{1}dX_{2}dX_{3}$ the 4-dimensional
volume element in a region $\Omega$ bounded by $S$, we obtain
after integrating in one of the variables we obtain
\begin{eqnarray*}
& &\int_{\Omega} \partial\!\!\! \slash f(X)dv=
\int_{\Omega} e^{\alpha}\partial_{\alpha} e^{\beta}U_{\beta} dv =\\
&&\int_{\Omega}[(\partial_{0}U_{0}\!-\!\sum_{i}\partial_{i}U_{i}) \!+\!
\sum_{i} (\partial_{0}U_{i}+\partial_{i}U_{0}) e^{i} \! +\!\epsilon^{ijk}\partial_{i}U_{j}e^{k}
] dv
\end{eqnarray*}
and noting that
\begin{eqnarray*}
\int_{\Omega}\partial_{0}U_{0}dv =\int_{S}U_{0}dS_{0},\,
&&\int_{\Omega}\partial_{0}U_{i}dv = \int_{S}U_{i}dS_{0},\\
\int_{\Omega}\partial_{i}U_{0}dv =\int_{S}U_{0}dS_{i},\;\;
&&\int_{\Omega}\partial_{i}U_{j}dv =\int_{S}U_{j}dS_{i}
\end{eqnarray*}
it follows that
\begin{eqnarray*}
\int_{\Omega} \partial\!\!\! \slash f(X)dv &=&
\int_{S}
[( U_{0}dS_{0}-\sum\delta^{ij} U_{i}dS_{j})e^{0}\\
& +&\sum(U_{i}dS_{0}
+U_{0}dS_{i} )e^{i} -\sum \epsilon^{ijk} U_{i}dS_{j}e^{k} ]
\end{eqnarray*}
It is a simple matter to see that this is exactly the same expression
of the surface integral
\[\int_{S}dS_{\eta}f(X)=\sum \int_{S} U_{\alpha} dS_{\beta} e^{\beta}e^{\alpha}
\]
Therefore, we obtain the result
\begin{equation}
\int_{\Omega} \partial\!\!\! \slash f(X)dv=\int_{S} dS_{\eta} f(X) \label{eq:RDIV}
\end{equation}
and similarly we obtain for the left hypersurface integral
\begin{equation}
\int_{\Omega}f(X)\partial\!\!\! \slash dv=\int_{S} f(X)dS_{\eta}\label{eq:LDIV}
\end{equation}
The above integrals hold for any of the previously defined classes of
functions and they difference is
\[
\sum\epsilon^{ijk}e^{k}\!\!\int_{S}(U_{i}dS_{j}\!-\! U_{j}dS_{i})\!=\!
\!-\!\sum\epsilon^{ijk}e^{k}\!\!
\int_{\Omega}(\frac{\partial U_{i}}{\partial X_{j}} + \frac{\partial
U_{j}}{\partial X_{i}})dv
\]
which vanish on account of Green's theorem in the $(i,j)$ plane.
The following result extends the first Cauchy's Theorem for quaternion
functions:
{\em If $f(X)$ is of H class in the interior of a region $\Omega$
bounded by a hypersurface $S$ then }
\begin{equation}
\int_{S} f(X) dS_{\eta} =\int_{S}dS_{\eta} f(X)=0 \label{eq:CAUCHY1}
\end{equation}
This follows immediately from eqns. \rf{RDIV}, \rf{LDIV} and the condition
for a class $H$ function \rf{H} where $\partial\!\!\! \slash f(X)= 0$ and $f(X)\partial\!\!\! \slash =0$.
The second Cauchy's theorem is also true only for class H functions
{\em If $f(X)$ satisfy the conditions of class $H$, in a region bounded by a
simple closed 3-dimensional hypersurface $S$, then for $P \in S $,}
\begin{equation}
f(P)=\frac{1}{\pi^{2}}\int_{S} f(X)(X-P)^{-3}dS_{\eta} \label{eq:CAUCHY2}
\end{equation}
In fact, the integrand does not satisfy the class $H$ conditions in $\Omega$
as it is not defined at $P$ and consequently
the previous theorem does not apply. However this point may be isolated by a
sphere with surface $S_{0}$ with center at $P$ and radius $\epsilon$
such that it is completely inside $\Omega$.
Applying the previous theorem in the region bounded by $S$ and $S_{0}$
we obtain
\[
\int_{S} f(X)(X-P)^{-3}dS_{\eta} +\int_{S_{0}}f(X)(X-P)^{-3}dS_{\eta} =0
\]
Now, the primary condition for a function belonging to class E through H is
that its components are regular so that we may calculate their Taylor
series around $P$:
$
U_{\alpha}(X) = U_{\alpha}(P) +\epsilon^{\beta}\frac{\partial
U_{\alpha}}{\partial x^{\beta}}\rfloor_{P} +\cdots .
$
Using this expansion in the integral over $S_{0}$ and taking the limit
$\epsilon\rightarrow 0$, it follows that
\begin{equation}
f(P)\! =\!\left( \int_{S} f(X)(X\!-\! P)^{-3}dS_{\eta}\right) \left(
\int_{S_{0}} (X\!-\! P)^{-3}dS_{\eta}\right)^{-1} \label{eq:fp}
\end{equation}
In order to calculate the integral over the sphere it is convenient to use four
dimensional spherical coordinates
$(r,\theta, \phi,\gamma) $, such that
$X_{0}=r sin\gamma $, $X_{1}=r cos\gamma\, sin\theta\,cos\phi $,
$X_{2}=r cos\gamma\, sin\theta\,sin\phi $ and $X_{3}=r cos\gamma\, cos\theta$
where $\theta\in(0,\pi)$, $ \phi\in(0,2\pi)$, $ \gamma\in(-\pi/2 , \pi/2)$.
With this, the coordinates $X_{0},X_{1},X_{2},X_{3}$ correspond to the
coordinates of a space-time point with quaternion norm $|X|^{2}=r^{2}$, while
$\gamma $ span values from the past to the future.
Then the volume element is $dv=J dr d\theta d\phi d\gamma$ where
$J=-\epsilon^{3}cos^{2}\gamma\sin\theta$ is the Jacobian determinant.
Using the polar form, the unit normal to the sphere centered at $P$ can be
written as $\eta=e^{\iota\gamma}$ and $X-P =\epsilon e^{\iota\gamma}=\epsilon
\eta$, so that
\[
\int_{S_{0}} (X-P)^{-3}dS_{\eta}=\int_{S_{0}}
e^{-2\iota\gamma}sin^{2}\gamma sin\theta\, d\theta\, d\phi\, d\gamma=\pi^{2}
\]
After replacing in \rf{fp} we obtain the result \rf{CAUCHY2}.
Notice that the power $(-3)$ in \rf{CAUCHY2} is not accidental
as it is the right power required to cancel the Jacobian
determinant as $\epsilon\rightarrow 0$.
\section{Power Series}
To conclude, consider the particular function $f(X)=(1-X)^{-3}$, with
$\vert X\vert<1$.
It is a simple matter to see that it can be expanded as
\begin{equation}
(1\!-\! X)^{-3}\!=\!\sum_{1}^{\infty}\frac{n(n+1)}{2}X^{n-1}\! =\!
\sum_{m=0}^{\infty}\frac{(m\!+\!1)(m\!+\!2)}{2}X^{m}\label{eq:example}
\end{equation}
Using this particular case we may prove the following general result
for quaternion functions:
{\em Let $f(X)$ be of class H inside a region $\Omega$ bounded by a surface
$S$. Then for all $X$ inside $\Omega$ there are coefficients $a_{n}$ such
that }
\begin{equation}
f(X) =\sum_{0}^{\infty} a_{n}(X-Q)^{n} \label{eq:TH3}
\end{equation}
The proof is a straightforward adaptation from the similar complex theorem.
If $S_{0}$ is the largest sphere in $\Omega$ centered at $Q$,
the integral \rf{CAUCHY2} for a point $P=X$ inside $\Omega$ gives
\[
f(X)\! =\! \frac{1}{\pi^{2}}\int_{S}\!
f(X')(X'\!-\! Q)^{-3}[1\!-\! (X'\!-\!Q)^{-1}(X\!-\!Q)]^{-3}dS'_{\eta}
\]
Assuming that $|X-Q| < |X'-Q|$ and using \rf{example}, the integrand is
equivalent to
\begin{eqnarray*}
&&[1-(X'-Q)^{-1}(X-Q)]^{-3}\\
&=&\sum_{0}^{m=\infty}\frac{(m+1)(m+2)}{2}(X'-Q)^{-m}(X-Q)^{m}
\end{eqnarray*}
so that
\begin{eqnarray}
&&f(X)=\frac{1}{\pi^{2}}\sum_{m=0}^{\infty}\frac{(m+1)(m+2)}{2}\times\nonumber\\
&&\times\int_{S_{0}}f(X')(X'-Q)^{-3-m} (X-Q)^{m} dS'_{\eta} \label{eq:FP}
\end{eqnarray}
Now we may write
$(X-Q)^{m}=\epsilon^{m}e^{m\iota\gamma}$ and $dS'_{\eta}=e^{\iota\gamma}dS' $,
and it follows that
\[
(X\!-\!Q)^{m}dS'_{\eta}\!-\!dS'_{\eta}(X\!-\!Q)^{m}=\epsilon dS'(e^{m\iota\gamma}
e^{\iota\gamma}\!-\!e^{\iota\gamma}e^{m\iota\gamma})\! =\! 0
\]
Therefore \rf{FP} is equivalent to
\begin{eqnarray*}
&&f(X)=\frac{1}{\pi^{2}}\sum_{m=0}^{\infty}\frac{(m+1)(m+2)}{2}\times\\
&&\times \int_{S_{0}}f(X')(X'-Q)^{-3-m}dS_{\eta}\, (X-Q)^{m}
\end{eqnarray*}
or, after defining the coefficients
\begin{equation}
a_{m}\!=\! \frac{1}{\pi^{2}}\frac{(m\!+\!1)(m\!+\! 2)}{2}
\int_{S_{0}}f(X')(X'\!-\! Q)^{-3-m} dS'_{\eta} \label{eq:COE1}
\end{equation}
we obtain \rf{TH3}.
This important result shows that class $H$ functions can be expressed as a
convergent positive power series.
Therefore the class H or the latter property could be taken to represent a class of
analyticity for quaternion functions, in the same sense of the real and
complex analyticity. However, unlike the complex case from \rf{H} we see that
their derivatives depend on the direction in which the limit is taken.
\section{Discussion}
At the level of an unbroken $SU(2)$ gauge theory
the algebraic properties of the spinors predicts a combined spin-isospin
angular momentum here called the k-spin (after i-spin for complex and j-spin
for
isospin). The three resulting spinors give way to a full quaternionic
spinor operator obtained from the linear combination of the Pauli matrices
in a specific representation. In this way, we conclude that
quaternion quantum mechanics may be effective at the level of the combined
spinor symmetry. The quaternionic spinor operator naturally associates a
vector in space-time whose physical interpretation depends on the $SU(2)$
model of gauge field considered. We have suggested the
't Hooft-Poliakov monopole as a possible interpretation of that vector
field.
The emergence of quaternionic spinor fields requires a
proper analysis of quaternion functions of a quaternionic variable.
We have shown that there is a class of analytic quaternion functions
which can be represented by a positive power series, a
property which is shared with the other two associative division algebras (the
real and complex functions). Outside the conditions for class H
the the power expansions would also have negative powers and the associated
poles as points in space-time and their corresponding residues \cite{Maia}.
The harmonic property implicit in class H implies in the possibility that the
quaternion quantum fields and states can be represented in terms of
quaternionic Fourier expansions, something that is required to represent the
quaternion wave packets.
As it has been noted, quaternion analyticity does not
imply that the derivatives are independent of direction in space
and in this respect complex analysis and the
corresponding quantum theory may be considered to be somewhat limited as compared with
quaternion analysis. The direction dependent property should be detectable at
the level of the combined symmetry
It is conceivable that the characterization of analyticity either by class H
or more generally by positive power series expansions will
not hold at higher energy levels,
where the wave functions are subjected to fast variation at a sort time. In
this case, the best we may hope that these functions remain
differentiable and any appeal to analyticity in the sense of a converging
power series may be regarded as an unduly luxury.
In this respect, the above results may hold
for quaternions quantum mechanics at an intermediate energy theory, where
the combined symmetry includes the $SU(2)$ group.
For higher energies we would expect the emergence of the $SU(3)$ group
and the octonion algebra.
|
\section{Introduction}
In this paper, we determine the stability of magnetic
(or Abelian Higgs) vortices.
These are certain critical points of the energy functional
\begin{equation}
\label{eq:ac}
E(\psi,A) = \frac{1}{2} \int_{{\bf R}^2} \left\{ |\nabla_A \psi|^2
+ (\nabla \times A)^2 + \frac{\lambda}{4}
(|\psi|^2-1)^2 \right\}
\end{equation}
for the fields
\[ A : {\bf R}^2 \rightarrow {\bf R}^2 \;\;\;\;\;
\mbox{ and } \;\;\;\;\;
\psi : {\bf R}^2 \rightarrow {\bf C}. \]
Here
$\nabla_A = \nabla - iA$
is the covariant gradient, and $\lambda > 0$ is a coupling
constant. For a vector, $A$, $\nabla \times A$ is the
scalar $\partial_1 A_2 - \partial_2 A_1$, and for a scalar $\xi$,
$\nabla \times \xi$ is the vector $(-\partial_2 \xi, \partial_1 \xi)$.
Critical points of $E(\psi,A)$ satisfy the {\em Ginzburg-Landau}
(GL) equations
\begin{equation}
\label{eq:eq1}
-\Delta_A\psi +
\frac{\lambda}{2}(|\psi|^2-1)\psi = 0
\end{equation}
\begin{equation}
\label{eq:eq2}
\nabla \times \nabla \times A
- \Im(\bar{\psi} \nabla_A \psi) = 0
\end{equation}
where $\Delta_A = \nabla_A \cdot \nabla_A$.
Physically, the functional $E(\psi,A)$ gives the
difference in free energy between the superconducting and normal states
near the transition temperature in the Ginzburg-Landau theory.
$A$ is the vector potential ($\nabla \times A$ is the induced
magnetic field), and $\psi$ is an {\em order parameter}. The modulus
of $\psi$ is interpreted as describing the local density of
superconducting Cooper pairs of electrons.
The functional $E(\psi,A)$ also gives the energy of a
static configuration in the
Yang-Mills-Higgs classical gauge theory on ${\bf R}^2$, with abelian
gauge group $U(1)$. In this case $A$ is a connection
on the principal $U(1)$- bundle ${\bf R}^2 \times U(1)$,
and $\psi$ is the {\em Higgs field} (see \cite{jt} for details).
A central feature of the functional $E(\psi,A)$ (and the GL equations)
is its infinite-dimensional symmetry group. Specifically,
$E(\psi,A)$ is invariant under $U(1)$ {\em gauge transformations},
\begin{equation}
\label{eq:g1}
\psi \mapsto e^{i\gamma}\psi
\end{equation}
\begin{equation}
\label{eq:g2}
A \mapsto A + \nabla \gamma
\end{equation}
for any smooth $\gamma : {\bf R}^2 \rightarrow {\bf R}$.
In addition, $E(\psi,A)$ is invariant under coordinate
translations, and under the coordinate rotation transformation
\begin{equation}
\label{eq:rot}
\psi(x) \mapsto \psi(g^{-1}x) \;\;\;\;\;\;\;\;\;\;
A(x) \mapsto gA(g^{-1}x)
\end{equation}
for $g \in SO(2)$.
Finite energy field configurations satisfy
\begin{equation}
\label{eq:bc}
|\psi| \rightarrow 1 \;\;\;\;\; \mbox{ as }
\;\;\;\;\; |x| \rightarrow \infty
\end{equation}
which leads to the definition of the {\em topological degree},
$\mbox{deg}(\psi)$, of such a configuration:
\[
\mbox{deg}(\psi) = \mbox{deg} \left(
\left. \frac{\psi}{|\psi|} \right|_{|x| = R} :
{\bf S}^1 \rightarrow {\bf S}^1 \right)
\]
($R$ sufficiently large). The degree is related to the phenomenon
of flux quantization. Indeed, an application of Stokes' theorem shows that
a finite-energy configuration satisfies
\[
\mbox{deg}(\psi) = \frac{1}{2\pi} \int_{{\bf R}^2} (\nabla \times A).
\]
We study, in particular, ``radially-symmetric'' or ``equivariant''
fields of the form
\begin{equation}
\label{eq:psiA}
\psi^{(n)}(x) = f_n(r)e^{in\theta} \;\;\;\;\;\;\;\;\;\;
A^{(n)}(x) = n\frac{a_n(r)}{r} \hat{x}^{\perp}
\end{equation}
where
$(r,\theta)$ are polar coordinates on ${\bf R}^2$,
$\hat{x}^{\perp} = \frac{1}{r} (-x_2,x_1)^t$,
$n$ is an integer, and
\[
f_n, a_n : [0,\infty) \rightarrow {\bf R}.
\]
It is easily checked that such configurations
(if they satisfy~(\ref{eq:bc})) have degree $n$.
The existence of critical points of this form is well-known
(see section~\ref{subsec:vort}). They are called {\em $n$-vortices}.
Our main results concern the stability of these
$n$-vortex solutions. Let
\[
L^{(n)} = \mbox{ Hess } E (\psi^{(n)}, A^{(n)})
\]
be the linearized operator for GL around the $n$-vortex, acting on
the space
\[
X = L^2({\bf R}^2,{\bf C}) \oplus L^2({\bf R}^2,{\bf R}^2).
\]
The symmetry group of $E(\psi,A)$ gives rise to an infinite-dimensional
subspace of $\ker(L^{(n)}) \subset X$ (see section~\ref{subsec:zeromodes}),
which we denote here by $Z_{sym}$.
We say the $n$-vortex is (linearly) {\em stable} if for some $c > 0$,
\[
L^{(n)}|_{Z_{sym}^{\perp}} \geq c,
\]
and {\em unstable} if $L^{(n)}$ has a negative eigenvalue.
The basic result of this paper is
the following linearized stability statement:
\begin{thm}
\label{thm:main}
\begin{enumerate}
\item (Stability of fundamental vortices) \\
For all $\lambda > 0$, the $\pm1$-vortex is stable.
\item (Stability/instability of higher-degree vortices) \\
For $|n| \geq 2$, the $n$-vortex is
\[
\left\{ \begin{array}{cc}
\mbox{ stable } & \mbox{ for } \lambda < 1 \\
\mbox{ unstable } & \mbox{ for } \lambda > 1.
\end{array} \right.
\]
\end{enumerate}
\end{thm}
Theorem~\ref{thm:main} is the basic ingredient in a proof of the nonlinear
dynamical stability/instability of the $n$-vortex for certain
dynamical versions of the GL equations.
These include the GL gradient flow equations,
the Abelian Higgs (Lorentz-invariant) equations, and
the Maxwell equations coupled to a nonlinear Schr\"odinger equation.
These dynamical stability results are established in a
companion paper~(\cite{g2}).
The statement of theorem~\ref{thm:main} was conjectured in~\cite{jt}
on the basis of numerical observations (see~\cite{jr}).
Bogomolnyi (\cite{bog}) gave an argument for instability of vortices
for $\lambda > 1$, $|n| \geq 2$. Our result
rigorously establishes this property.
The solutions of (\ref{eq:eq1}-\ref{eq:eq2}) are well-understood
in the case of {\em critical coupling}, $\lambda = 1$.
In this case, the {\em Bogomolnyi method} (\cite{bog}) gives a pair
of first-order equations whose solutions are global minimizers
of $E(\psi,A)$ among fields of fixed degree (and hence solutions of the
the GL equations). Taubes (\cite{t1,t2})
has shown that all solutions of GL with $\lambda = 1$ are solutions
of these first-order equations, and that for a given degree $n$,
the gauge-inequivalent solutions
form a $2|n|$-parameter family.
The $2|n|$ parameters describe the locations of the zeros of the scalar
field. This is discussed in more detail in \cite{jt}
(see also \cite{bgp}) and section~\ref{sec:crit}.
We remark that for $\lambda = 1$, an $n$-vortex solution~(\ref{eq:psiA})
corresponds to the case when all $|n|$ zeros of the
scalar field lie at the origin.
The remainder of this paper is organized as follows.
In section~\ref{sec:vort} we describe in detail various properties
of the $n$-vortex. In particular, we establish an important estimate
on the $n$-vortex profiles which differentiates between the cases
$\lambda < 1$ and $\lambda > 1$.
In section~\ref{sec:lin}, we introduce the linearized operator, fix the
gauge on the space of perturbations, and identify the zero-modes due to
symmetry-breaking.
Sections~\ref{sec:block} through \ref{sec:high} comprise a proof of
theorem~\ref{thm:main}. A block-decomposition for the linearized operator is
described in section~\ref{sec:block}. This approach is similar to that
used to study the stability of non-magnetic vortices in~\cite{os1}
and~\cite{g1}. In section~\ref{sec:fund},
we establish the positivity of certain blocks
(those corresponding to the radially-symmetric variational problem,
and those containing the translational zero-modes) for all $\lambda$,
which completes the stability proof for the $\pm 1$-vortices. The basic
techniques are the characterization of symmetry-breaking in terms of
zero-modes of the Hessian (or linearized operator), and
a Perron-Frobenius type argument,
based on a version of the maximum principle for systems
(proposition~\ref{prop:mainmp}), which shows that
the translational zero-modes correspond to the bottom of the spectrum
of the linearized operator.
A more careful analysis is needed for $|n| \geq 2$.
This requires us to review some aspects of the critical case
($\lambda = 1$) in section~\ref{sec:crit}.
The stability/instability proof for $|n| \geq 2$ is completed in
section~\ref{sec:high}. We use an extension of Bogomolnyi's instability
argument, and another application of the Perron-Frobenius theory.
{\em Acknowledgment: the first author would like to thank
the Courant institute for its hospitality during part of the
preparation of this paper, and especially J. Shatah for some
helpful discussions. Part of this work is toward fulfillment of the
requirements of the first author's PhD at the University of Toronto.
The second author thanks Yu. N. Ovchinnikov
for many fruitful discussions.}
\section{The $n$-vortex}
\label{sec:vort}
In this section we discuss the existence, and properties,
of $n$-vortex solutions.
\subsection{Vortex solutions}
\label{subsec:vort}
The existence of solutions of (GL) of the form~(\ref{eq:psiA})
is well-known:
\begin{thm}[Vortex Existence; \cite{p,bc}]
\label{thm:exist}
For every integer $n$, there is a solution
\begin{equation}
\label{eq:nvortex}
\psi^{(n)}(x) = f_n(r)e^{in\theta} \;\;\;\;\;\;\;\;\;\;
A^{(n)}(x) = n\frac{a_n(r)}{r} \hat{x}^{\perp}
\end{equation}
of the variational equations (\ref{eq:eq1})-(\ref{eq:eq2}).
In particular, the radial functions ($f_n$, $a_n$)
minimize the radial energy functional
\begin{equation}
\label{eq:erad}
E^{(n)}_r(f,a) = \frac{1}{2} \int_0^{\infty}
\left\{ (f')^2 + n^2\frac{(1-a)^2 f^2}{r^2}
+ n^2\frac{(a')^2}{r^2} + \frac{\lambda}{4}(f^2-1)^2 \right\} rdr
\end{equation}
(which is the full energy functional~(\ref{eq:ac}) restricted
to fields of the form~(\ref{eq:psiA}))
in the class
\[
\{ f,a : [0,\infty) \rightarrow {\bf R} \;\; | \;\;
1-f \in H_1(rdr), \frac{a}{r} \in L^2_{loc}(rdr),
\frac{a'}{r} \in L^2(rdr) \}.
\]
The functions $f_n$, $a_n$ are smooth, and have the
following properties (for $n \not= 0$):
\begin{enumerate}
\item $0 < f_n < 1$, $0 < a_n < 1$ on $(0,\infty)$
\item $f_n', a_n' > 0$
\item $f_n \sim cr^n$, $a_n \sim dr^2$,
as $r \rightarrow 0$ ($c>0$ and $d>0$ are constants)
\item $1-f_n$, $1-a_n \rightarrow 0$
as $r \rightarrow \infty$, with an exponential rate of decay.
\end{enumerate}
\end{thm}
We call ($\psi^{(n)}$, $A^{(n)}$) an {\em $n$-vortex}
(centred at the origin).
It follows immediately that the functions $f_n$ and $a_n$
satisfy the ODEs
\begin{equation}
\label{eq:ode1}
-\Delta_r f_n + \frac{n^2(1-a_n)^2}{r^2}f_n +
\frac{\lambda}{2}(f_n^2-1)f_n = 0
\end{equation}
and
\begin{equation}
\label{eq:ode2}
-a_n'' + \frac{a_n'}{r} - f_n^2(1-a_n) = 0.
\end{equation}
\begin{rem}
To our knowledge, it is not known if solutions of the form~(\ref{eq:psiA})
are unique. In the appendix, we show that for $\lambda \geq 2n^2$,
any such solution minimizes $E_r^{(n)}$.
\end{rem}
\begin{rem}
The functions $f_n$ and $a_n$ also depend on $\lambda$, but
we suppress this dependence for ease of notation.
When it will cause no confusion, we will also drop the subscript $n$.
\end{rem}
\begin{rem}
The discrete symmetry $\psi \mapsto \bar{\psi}$, $A \mapsto -A$ of (GL)
interchanges $(\psi^{(n)}, A^{(n)})$ and $(\psi^{(-n)}, A^{(-n)})$.
Thus, we can assume $n \geq 0$.
\end{rem}
\subsection{An estimate on the vortex profiles}
The following inequality, relating the exponentially decaying
quantities $f'$ and $1-a$, plays a crucial role in the
stability/instability proof.
\begin{prop}
We have
\begin{equation}
\label{eq:ineq}
\left\{ \begin{array}{cc}
f'(r) > \frac{n(1-a(r))}{r}f(r) & \;\;\;
\mbox{ for } \;\;\; \lambda < 1 \\
f'(r) < \frac{n(1-a(r))}{r}f(r) & \;\;\;
\mbox{ for } \;\;\; \lambda > 1
\end{array} \right.
\end{equation}
\end{prop}
{\em Proof:}
Define $e(r) \equiv f'(r) - \frac{n(1-a(r))}{r}f(r)$.
The properties listed in theorem~\ref{thm:exist} imply that
$e(r) \rightarrow 0$ as $r \rightarrow 0$ and as $r \rightarrow \infty$.
Using the ODEs ((\ref{eq:ode1})-(\ref{eq:ode2})) we
can derive the equation
\[
(-\Delta_r + \alpha)e + \frac{e}{f}e' = (1-\lambda)f^2f'
\]
where
\[
\alpha(r) = \frac{1+n(1-a)}{r^2}(1 + \frac{rf'}{f})
+ f^2 + \frac{na'}{r} > 0
\]
and the result follows from the maximum principle. $\Box$
\section{The linearized operator}
\label{sec:lin}
In this section, we introduce the linearized operator (or Hessian)
around the $n$-vortex, and identify its symmetry zero-modes.
\subsection{Definition of the linearized operator}
\label{section:linop}
We work on the real Hilbert space
\[
X = L^2({\bf R}^2;{\bf C}) \oplus L^2({\bf R}^2;{\bf R}^2)
\]
with inner-product
\[
< (\xi,B), (\eta,C) >_X = \int_{{\bf R}^2}
\{ \Re (\bar{\xi}\eta) + B \cdot C \}.
\]
We define the linearized operator, $L_{\psi,A}$
(= the Hessian of $E(\psi,A)$) at a solution
$(\psi,A)$ of (\ref{eq:eq1})-(\ref{eq:eq2})
through the quadratic form
\[
\frac{\partial^2}{\partial \epsilon \partial \delta}
E(\psi + \epsilon\xi + \delta\eta, A + \epsilon B + \delta C)
|_{\epsilon = \delta = 0} =
\langle (\eta, C) L_{\psi,A} (\xi, B) \rangle_X
\]
for all $(\xi, B)$, $(\eta, C)$, $\in X$.
The result is
\[
L_{\psi,A} \left( \begin{array}{c} \xi \\ B \end{array} \right)
= \left( \begin{array}{c}
[-\Delta_A + \frac{\lambda}{2}(2|\psi|^2-1)]\xi +
\frac{\lambda}{2}\psi^2\bar{\xi}
+ i[2\nabla_A \psi + \psi\nabla]\cdot B \\
\Im( [\bar{\nabla_A\psi}-\bar{\psi}\nabla_A] \xi)
+ (-\Delta + \nabla\nabla + |\psi|^2) \cdot B
\end{array} \right).
\]
\subsection{Symmetry zero-modes}
\label{subsec:zeromodes}
We identify the part of the kernel of the operator
\[
L^{(n)} \equiv L_{\psi^{(n)},A^{(n)}}
\]
which is due to the symmetry group.
\begin{prop}
\label{prop:modes}
We have
\begin{enumerate}
\item
\begin{equation}
\label{eq:gmode}
L^{(n)} \left( \begin{array}{c}
i\gamma \psi^{(n)} \\ \nabla \gamma
\end{array} \right) = 0
\end{equation}
for any $\gamma : {\bf R}^2 \rightarrow {\bf R}$
\item
\begin{equation}
\label{eq:tmode}
L^{(n)} \left( \begin{array}{c}
\partial_j \psi^{(n)} \\ \partial_j A^{(n)}
\end{array} \right) = 0
\end{equation}
for $j=1,2$.
\end{enumerate}
\end{prop}
{\em Proof:}
We use the basic result that the generator of a one-parameter group
of symmetries of $E(\psi,A)$, applied to the $n$-vortex, lies in the kernel
of $L^{(n)}$.
The vector in~(\ref{eq:gmode}) is easily seen to be the generator
of a one-parameter family of gauge transformations~(\ref{eq:g1}-\ref{eq:g2})
applied to the $n$-vortex. Similarly, the vector in~(\ref{eq:tmode})
is the generator of coordinate
translations applied to the $n$-vortex. $\Box$
\begin{rem}
Applying the generator of the coordinate rotational symmetry~(\ref{eq:rot})
to the $n$-vortex gives us nothing new, it is contained in the
gauge-symmetry case.
\end{rem}
We define $Z_{sym}$ to be the subspace of $X$ spanned by
the $L^2$ zero-modes described in proposition~\ref{prop:modes}.
We recall that the $n$-vortex is called {\em stable}
if there is a constant $c > 0$ such that
\begin{equation}
\label{eq:stabdef}
L^{(n)} |_{Z_{sym}^{\perp}} \geq c,
\end{equation}
and {\em unstable} if $L^{(n)}$ has a negative eigenvalue.
\subsection{Gauge fixing}
\label{subsec:gauge}
In order to remove the infinite dimensional kernel of
$L^{(n)}$ arising from gauge symmetry, we restrict
the class of perturbations. Specifically, we restrict
$L^{(n)}$ to the space of those perturbations
$(\xi, B) \in X$
which are orthogonal to the $L^2$ gauge zero-modes~(\ref{eq:gmode}).
That is,
\[
\left\langle \left( \begin{array}{c}
i\gamma \psi^{(n)} \\ \nabla \gamma
\end{array} \right), \left( \begin{array}{c}
\xi \\ B \end{array} \right) \right\rangle_X = 0
\]
for all $\gamma$. Integration by parts gives the gauge condition
\begin{equation}
\label{eq:gchoice}
\Im(\overline{\psi^{(n)}}\xi) = \nabla \cdot B.
\end{equation}
As is done in \cite{s}, we consider a modified quadratic form
$\tilde{L}^{(n)}$, defined by
\[
< \alpha, \tilde{L}^{(n)} \alpha > =
< \alpha, L^{(n)} \alpha >
+ \int (\Im(\overline{\psi^{(n)}} \xi) - \nabla \cdot B)^2
\]
for $\alpha = (\xi, B) \in X$.
Clearly, $\tilde{L}^{(n)}$ agrees with $L^{(n)}$ on the subspace
of $X$ specified by the gauge condition~(\ref{eq:gchoice}).
This modification has the important effect of shifting the essential
spectrum away from zero (see~(\ref{eq:contspec})).
A straightforward computation gives the following expression for
$\tilde{L}^{(n)}$:
\[
\tilde{L}^{(n)} \left( \begin{array}{c} \xi \\ B \end{array} \right) =
\left( \begin{array}{c}
[-\Delta_A + \frac{\lambda}{2}(2|\psi|^2-1) + \frac{1}{2}|\psi|^2]\xi
+ \frac{1}{2}(\lambda-1)\psi^2\bar{\xi} + 2i\nabla_A \psi \cdot B \\
2\Im[\bar{\nabla_A \psi}\xi] + [-\Delta + |\psi|^2]B
\end{array} \right).
\]
To establish theorem~\ref{thm:main},
it suffices to prove that $\tilde{L}^{(n)} \geq c > 0$
on the subspace of $X$
orthogonal to the translational zero-modes~(\ref{eq:tmode}).
$\tilde{L}^{(n)}$ is a real-linear operator on $X$.
It is convenient to identify $L^2({\bf R}^2;{\bf R}^2)$ with
$L^2({\bf R}^2;{\bf C})$ through the correspondence
\begin{equation}
\label{eq:complex}
B = \left( \begin{array}{c} B_1 \\ B_2 \end{array} \right)
\leftrightarrow B^c \equiv B_1 - iB_2,
\end{equation}
and then to complexify the
space $X \mapsto \tilde{X} = [L^2({\bf R}^2;{\bf C})]^4$ via
\begin{equation}
\label{eq:comp}
(\xi, B) \mapsto
(\xi, \bar{\xi}, B^c, \bar{B}^c).
\end{equation}
As a result, $\tilde{L}^{(n)}$ is replaced by the
complex-linear operator
\[
\tilde{\tilde{L}}^{(n)} =
\mbox{ diag } \{ -\Delta_A, -\overline{\Delta_A}, -\Delta, -\Delta \}
+ V^{(n)}
\]
where
\[
V^{(n)} = \left( \begin{array}{cccc}
\frac{\lambda}{2}(2|\psi|^2-1) + \frac{1}{2}|\psi|^2 &
\frac{1}{2}(\lambda-1)\psi^2 & -i(\partial_A^* \psi) &
i(\partial_A \psi) \\ \frac{1}{2}(\lambda - 1)\bar{\psi}^2 &
\frac{\lambda}{2}(2|\psi|^2-1) + \frac{1}{2}|\psi|^2 &
-i(\bar{\partial_A \psi}) & i(\bar{\partial_A^* \psi}) \\
i(\bar{\partial_A^* \psi}) & i(\partial_A \psi) &
|\psi|^2 & 0 \\ -i(\bar{\partial_A \psi}) &
-i(\partial_A^* \psi) & 0 & |\psi|^2 \end{array} \right).
\]
Here we have used the notation
\[
\partial_A \equiv \partial_z - iA
\]
where $\partial_z = \partial_1 - i\partial_2$ (and the superscript c has
been dropped from the complex function $A$ obtained from the vector-field
$A$ via~(\ref{eq:complex})).
The components of $V^{(n)}$ are bounded, and it follows from standard
results (\cite{rs2}) that
$\tilde{\tilde{L}}^{(n)}$ is a self-adjoint operator on $\tilde{X}$,
with domain
\[
D(\tilde{\tilde{L}}^{(n)}) = [H_2({\bf R}^2;{\bf C})]^4
\]
\section{Block decomposition}
\label{sec:block}
We write functions on ${\bf R}^2$ in polar coordinates. Precisely,
\begin{equation}
\label{eq:polar}
\tilde{X} = [L^2({\bf R}^2; {\bf C})]^4 =
[L^2_{rad} \otimes
L^2({\bf S}^1; {\bf C})]^4
\end{equation}
where $L^2_{rad} \equiv L^2({\bf R}^{+}, rdr)$.
Let $\rho_n : U(1) \rightarrow Aut([L^2({\bf S}^1; {\bf C})]^4)$
be the representation whose action is given by
\[
\rho_n(e^{i\theta})
(\xi, \eta, B, C)(x) =
(e^{in\theta}\xi, e^{-in\theta}\eta, e^{-i\theta}B,
e^{i\theta}C) (R_{-\theta}x)
\]
where $R_{\alpha}$ is a counter-clockwise rotation in ${\bf R}^2$
through the angle $\alpha$.
It is easily checked that the linearized operator $\tilde{\tilde{L}}^{(n)}$
commutes with $\rho_n(g)$ for any $g \in U(1)$.
It follows that $\tilde{\tilde{L}}^{(n)}$ leaves invariant
the eigenspaces of $d\rho_n(s)$ for any $s \in i{\bf R} = Lie(U(1))$.
The resulting block decomposition
of $\tilde{\tilde{L}}^{(n)}$, which is described in this section,
is essential to our analysis. In particular, the translational zero-modes
each lie within a single subspace of this decomposition.
\subsection{The decomposition of $L^{(n)}$}
\label{section:blocks}
In what follows, we define, for convenience,
$b(r) = \frac{n(1-a(r))}{r}$.
\begin{prop}
\label{prop:decomp}
There is an orthogonal decomposition
\begin{equation}
\label{eq:orth}
\tilde{X} = \bigoplus_{m \in {\bf Z}}
( e^{i(m+n)\theta} L^2_{rad} \oplus
e^{i(m-n)\theta} L^2_{rad} \oplus
-i e^{i(m-1)\theta} L^2_{rad} \oplus
i e^{i(m+1)\theta} L^2_{rad} ),
\end{equation}
under which
the linearized operator around the vortex, $\tilde{\tilde{L}}^{(n)}$,
decomposes as
\[
\tilde{\tilde{L}}^{(n)} = \bigoplus_{m \in {\bf Z}} \hat{L}_m^{(n)}
\]
where
\begin{equation}
\label{eq:lm2}
\hat{L}_m^{(n)} = -\Delta_r(Id) + \hat{V}_m^{(n)}
\end{equation}
with
\[
\hat{V}_m^{(n)} = \frac{1}{r^2}
\mbox{ diag } \{ [m + n(1-a)]^2, [m - n(1-a)]^2, [m-1]^2, [m+1]^2 \}
+ V'
\]
and
\[
V' = \left( \begin{array}{cccc}
\frac{\lambda}{2}(2f^2-1) + \frac{1}{2}f^2
& \frac{1}{2}(\lambda - 1)f^2 &
f' - bf & -[f' + bf] \\
\frac{1}{2}(\lambda-1)f^2 &
\frac{\lambda}{2}(2f^2-1) + \frac{1}{2}f^2 & -[f' + bf]
& f' - bf \\ f' - bf &
-[f' + bf] & f^2 & 0 \\
-[f' + bf] & f' - bf & 0 &
f^2 \end{array} \right).
\]
\end{prop}
{\em Proof:}
The decomposition~(\ref{eq:orth})
of $\tilde{X}$ follows from the usual Fourier decomposition
of $L^2({\bf S}^1;{\bf C})$, and the relation~(\ref{eq:polar}).
An easy computation shows that
$\tilde{\tilde{L}}^{(n)}$
preserves the space of vectors of the form
\begin{equation}
\label{eq:eig}
(\xi e^{i(m+n)\theta},
\eta e^{i(m-n)\theta}, -i\alpha e^{i(m-1)\theta},
i\beta e^{i(m+1)\theta} )
\end{equation}
and that it acts on such vectors via~(\ref{eq:lm2}).
$\Box$
It follows that $\hat{L}_m^{(n)}$ is self-adjoint on $[L^2_{rad}]^4$.
It will also be convenient to work with a rotated version
of the operator $\hat{L}_m^{(n)}$,
\[
L_m^{(n)} \equiv \left\{ \begin{array}{cc}
R \hat{L}_m^{(n)} R^T & m \geq 0 \\
R' \hat{L}_m^{(n)} (R')^T & m < 0
\end{array} \right.
\]
where
\[
R = \frac{1}{\sqrt{2}} \left( \begin{array}{cccc}
1 & 1 & 0 & 0 \\
-1 & 1 & 0 & 0 \\
0 & 0 & 1 & 1 \\
0 & 0 & 1 & -1 \end{array} \right), \;\;\;\;\;\;\;\;\;\;
R' = \frac{1}{\sqrt{2}} \left( \begin{array}{cccc}
1 & 1 & 0 & 0 \\
1 & -1 & 0 & 0 \\
0 & 0 & 1 & 1 \\
0 & 0 & 1 & -1 \end{array} \right).
\]
We have
\begin{equation}
\label{eq:lm1}
L_m^{(n)} = -\Delta_r(Id) + V_m^{(n)}
\end{equation}
where
\[
V_m^{(n)} = \left( \begin{array}{cccc}
\frac{m^2}{r^2} + b^2 + \frac{\lambda}{2}(3f^2-1) &
-2|m|\frac{b}{r} & -2bf & 0 \\
-2|m|\frac{b}{r} &
\frac{m^2}{r^2} + b^2 + \frac{\lambda}{2}(f^2-1) + f^2 &
0 & -2f' \\ -2bf & 0 & \frac{m^2+1}{r^2} + f^2 &
-2\frac{|m|}{r^2} \\ 0 & -2f' &
-2\frac{|m|}{r^2} & \frac{m^2+1}{r^2} + f^2 \end{array} \right).
\]
\subsection{Properties of $L_m^{(n)}$}
\label{subsec:gen}
\begin{prop}
\label{prop:props}
We have the following:
\begin{enumerate}
\item
\begin{equation}
\label{eq:flip}
L_m^{(n)} = L_{-m}^{(n)}
\end{equation}
\item
\begin{equation}
\label{eq:contspec}
\sigma_{ess}(L_m^{(n)}) = [\min(1,\lambda), \infty)
\end{equation}
\item
For $|n|=1$ and $m \geq 2$,
\begin{equation}
\label{eq:mon}
L_m^{(n)} - L_1^{(n)} \geq 0
\end{equation}
with no zero-eigenvalue.
\end{enumerate}
\end{prop}
{\em Proof:}
The first statement is obvious.
The second statement follows in a standard way from the fact that
\[
\lim_{r \rightarrow \infty} V_m^{(n)}(r) =
\mbox{ diag } \{\lambda, 1, 1, 1\}
\]
To prove the third statement, we compute
\[
\hat{L}_m^{(n)} - \hat{L}_1^{(n)} =
\frac{m-1}{r^2} \mbox{ diag }
\{ m+2n(1-a), \; m-2n(1-a), \; m-1, \; m+3 \}
\]
which is non-negative, with no zero-eigenvalue
for $m \geq 2$, $n=1$. $\Box$
\begin{rem}
In light of~(\ref{eq:flip}), we can assume from now on that $m \geq 0$.
This degeneracy is a result of the complexification~(\ref{eq:comp})
of the space of perturbations.
\end{rem}
\subsection{Translational zero-modes}
\label{section:zeros}
The gauge fixing (section~\ref{subsec:gauge})
has eliminated the zero-modes arising from
gauge symmetry. The translational zero-modes remain.
As written in~(\ref{eq:tmode}), the translational zero-modes fail
to satisfy the gauge condition~(\ref{eq:gchoice}). Further, they do
not lie in $L^2$.
A straightforward computation shows that
if we adjust the vectors in~(\ref{eq:tmode}) by gauge zero-modes
given by~(\ref{eq:gmode}) with $\gamma = -A_j$, $j=1,2$, we obtain
\[
T_1 = \left( \begin{array}{c}
(\nabla_{A} \psi)_1 \\
(\nabla \times A) e_2
\end{array} \right)
\;\;\;\;\;\;
T_2 = \left( \begin{array}{c}
(\nabla_{A} \psi)_2 \\
-(\nabla \times A) e_1
\end{array} \right)
\]
where $e_1 = (1, 0)$ and $e_2 = (0, 1)$.
$T_1$ and $T_2$ satisfy~(\ref{eq:gchoice}),
and are zero-modes of the
linearized operator. Note also that $T_{\pm 1}$ decay exponentially
as $|x| \rightarrow \infty$, and hence lie in $L^2$.
It is easily
checked that $T_1 \pm iT_2$ lie in the $m=\pm1$ blocks for $\hat{L}_m^{(n)}$.
After rotation by $R$, we have
\[
L_{\pm1}^{(n)} T = 0
\]
where
\[
T = (f', b f, n\frac{a'}{r}, n\frac{a'}{r}).
\]
\section{Stability of the fundamental vortices}
\label{sec:fund}
In this section we prove the first part of theorem~\ref{thm:main}.
Specifically, we show that for some $c > 0$, $L_m^{(\pm 1)} \geq c$
for $m \not= 1$, and $L_1^{(\pm1)} |_{T^{\perp}} \geq c$.
In light of the discussions in sections~\ref{subsec:gauge},
\ref{section:blocks}, and \ref{section:zeros}, this will establish the
stability of the $\pm1$-vortices.
\subsection{Non-negativity of $L_0^{(n)}$ and radial minimization}
\label{subsection:min}
\begin{prop}
\label{prop:rad}
$L_0^{(n)} \geq 0$ for all $\lambda$.
\end{prop}
{\em Proof:}
From the expression~(\ref{eq:lm1}) we see that $L_0^{(n)}$
breaks up:
\begin{equation}
\label{eq:split}
L_0^{(n)} = N_0 \oplus M_0
\end{equation}
(abusing notation slightly) where
\[
M_0 = -\Delta_r(Id) + W_0
\]
with
\[
W_0 = \left( \begin{array}{cc}
b^2 + \frac{\lambda}{2}(3f_n^2-1) & -2bf \\
-2bf & \frac{1}{r^2} + f^2
\end{array} \right)
\]
and
\[
N_0 = \left( \begin{array}{cc}
-\Delta_r + b^2 + \frac{\lambda}{2}(f^2-1) + f^2 &
-2f' \\ -2f' & -\Delta_r + \frac{1}{r^2} + f^2.
\end{array} \right)
\]
An easy computation shows that
$M_0$ is precisely the Hessian of the radial energy,
$Hess E_r^{(n)}$~(see (\ref{eq:erad})).
Since the $n$-vortex minimizes $E_r^{(n)}$,
we have $M_0 \geq 0$. It remains to show $N_0 \geq 0$. We establish
the stronger result, $N_0 > 0$. Note that
\[
N_0 = G_0^* G_0
\]
where
\[
G_0 = \left( \begin{array}{cc}
\partial_r - f'/f & f \\
f & \partial_r + 1/r
\end{array} \right)
\]
In fact, $G_0$ has no zero-eigenvalue. To see this, we first
remark that $G_0$ is a relatively compact perturbation of
$G_0|_{\lambda = 1}$, due to the exponential decay of the field
components. It follows from an index-theoretic calculation done
in~\cite{w,s}, that $G_0|_{\lambda = 1}$ is Fredholm, with index $0$.
We conclude that the same is true of $G_0$ (for any $\lambda$).
Finally, it is a simple matter to check that $G_0^*$ has trivial kernel. If
\[
G_0^* \left( \begin{array}{c} \xi \\ \beta \end{array} \right) = 0
\]
it follows that
\[
(-\Delta_r + f^2)\beta = 0
\]
and hence that $\beta = 0$, and so $\xi = 0$.
The relation $N_0 > 0$ follows from this, and the the fact that
$\sigma_{ess}(N_0) = [1,\infty)$. $\Box$
\subsection{A maximum principle argument}
Removing the equality in proposition~\ref{prop:rad} requires more work.
First, we establish an extension of the maximum
principle to systems (see, eg, \cite{lm,pa} for related results).
We will use this
also in the proof that the the translational zero-mode is the
ground state of $L_1^{(n)}$ (section~\ref{subsec:L1}).
\begin{prop}
\label{prop:mainmp}
Let $L$ be a self-adjoint operator on $L^2({\bf R}^n;{\bf R}^d)$ of the form
\[
L = -\Delta (Id) + V
\]
where $V$ is a $d \times d$ matrix-multiplication operator with
smooth entries. Suppose that $L \geq 0$ and that
for $i \not= j$, $V_{ij}(x) \leq 0$ for all $x$.
Further, suppose $V$ is irreducible in the sense that for any splitting
of the set $\{1,\ldots,d\}$ into disjoint sets $S_1$ and $S_2$, there is
an $i \in S_1$ and a $j \in S_2$ with $V_{ij}(x) < 0$ for all $x$.
Finally, suppose that
$L \xi = \eta \in L^2$ with $\eta \geq 0$ component-wise, and
$\xi \not\equiv 0$.
Then either
\begin{enumerate}
\item $\xi > 0$ or
\item $\eta \equiv 0$ and $\xi < 0$.
\end{enumerate}
\end{prop}
{\em Proof:}
We write $\xi = \xi^+ - \xi^-$ with $\xi^+, \xi^- \geq 0$ component-wise,
and compute
\[
0 \;\; \leq \;\; <\xi^-, L \xi^-> \;\; = \;\; <\xi^-, L \xi^+>
- <\xi^-, L \xi>.
\]
Since $\xi_j^+$ and $\xi_j^-$ have disjoint support, we have
\[
r.h.s = \sum_{j \not= k} <\xi_j^-, V_{jk} \xi_k^+>
- <\xi^-,\eta> \;\; \leq 0.
\]
Thus we have
\begin{enumerate}
\item $0 = \;\; <\xi^-, L\xi^->$
\item $0 = \;\; <\xi_j^-, V_{jk} \xi_k^+>$ for all $j \not= k$
\end{enumerate}
Since $L \geq 0$, the first of these implies $L \xi^- = 0$
and hence $L \xi^+ = \eta$. So if $\eta \not\equiv 0$, then
$\xi^+ \not\equiv 0$. If $\eta \equiv 0$ and $\xi^+ \equiv 0$,
replace $\xi$ with $-\xi$ in what follows.
An application of the strong maximum principle
(eg. \cite{gt}, Thm. 8.19) to each component of the equation
\[
L \xi^+ = \eta
\]
now allows us to conclude that for each $k$, either $\xi_k^+ > 0$
or $\xi_k^+ \equiv 0$.
We know that for some $k$, $\xi_k^+ > 0$.
Looking back at the second listed equation
above, and using the irreducibility of $V$,
we then see that $\xi_j^- \equiv 0$ for all $j$.
Finally, we can easily rule out the possibility $\xi_k \equiv 0$
for some $k$, by looking back at the equation satisfied by $\xi_k$.
Thus we have $\xi > 0$. $\Box$
\subsection{Positivity of $L_0^{(n)}$}
\label{subsection:l0}
Now we apply proposition~\ref{prop:mainmp} to show
$M_0 > 0$. The trick here is to find a function $\xi$ which satisfies
$M_0 \xi \geq 0$. This allows us to rule out the existence of a
zero-eigenvector, which would be positive by proposition~\ref{prop:mainmp}.
To obtain such a $\xi$, we differentiate the vortex with respect to the
parameter $\lambda$. Specifically,
differentiation of the Ginzburg-Landau
equations with respect to $\lambda$ results in
\begin{equation}
\label{eq:diff}
M_0 \xi = \eta
\end{equation}
where
\[
\xi = \left( \begin{array}{cc}
\partial_{\lambda} f \\ n \partial_{\lambda} a/r
\end{array} \right)
\]
and
\[
\eta = \left( \begin{array}{cc}
\frac{1}{2}(1-f^2)f \\ 0 \end{array} \right) \geq 0.
\]
We can now establish
\begin{prop}
\label{prop:l02}
For all $\lambda$, $L_0^{(n)} \geq c > 0$.
\end{prop}
{\em Proof:}
We have already shown in the proof of proposition~\ref{prop:rad}, that
$N_0 > 0$ and $M_0 \geq 0$. Hence, due to~(\ref{eq:split})
and~(\ref{eq:contspec}),
it suffices to show that $Null(M_0) = \{0\}$.
Suppose $M_0 \zeta = 0$, $\zeta \not\equiv 0$.
Proposition~\ref{prop:mainmp} then implies $\zeta > 0$
(or else take $-\zeta$). Now
\[
0 = \;\; <M_0\zeta, \xi> \;\; = \;\; <\zeta, M_0\xi> \;\;
= \;\; <\zeta, \eta> \;\; > \;\; 0
\]
gives a contradiction. $\Box$
\begin{rem}
Proposition~\ref{prop:mainmp} applied to equation~(\ref{eq:diff}) also
gives $\xi > 0$. That is, the vortex profiles increase
monotonically with $\lambda$. This can be used to show that the rescaled
vortex $(f_n(r/\sqrt{\lambda}), a_n(r/\sqrt{\lambda}))$ converges as
$\lambda \rightarrow \infty$ to $(f^*, 0)$, where $f^*$ is the (profile
of) the $n$-vortex solution of the ordinary GL equation:
$-\Delta_r f^* + n^2 f^*/r^2 + ({f^*}^2-1)f^* = 0$. This result was
established by different means in~\cite{abg}.
\end{rem}
\subsection{Positivity of $L_1^{(\pm1)}$}
\label{subsec:L1}
\begin{prop}
\label{prop:l1}
$ L_1^{(\pm1)} \geq 0 $
with non-degenerate zero-eigenvalue given by $T$.
\end{prop}
{\em Proof:}
Let $\mu = inf spec L_1^{(\pm 1)} \leq 0$, which is an
eigenvalue by~(\ref{eq:contspec}).
Suppose $L_1^{(\pm1)} S = \mu S$. Applying proposition~\ref{prop:mainmp}
to $L_1^{(\pm 1)} - \mu$ (note that $V_1^1$ satisfies the irreducibility
requirement) gives
$S > 0$ (or $S < 0$). Further, $\mu$ is
non-degenerate, as if $\mu$ were degenerate, we would have two
strictly positive eigenfunctions which are orthogonal, an impossibility.
Now if $\mu < 0$, we have
$<S,T> = 0$, which is also impossible. Thus $S$ is a multiple of $T$,
and $\mu = 0$. $\Box$
\subsection{Completion of stability proof for $n=\pm1$}
\label{subsec:mon}
We are now in a position to complete the proof of the first
statement of theorem~\ref{thm:main}.
By proposition~\ref{prop:l02},
$L_0^{(\pm1)} \geq c > 0$.
By proposition~\ref{prop:l1} and~(\ref{eq:contspec}),
$L_1^{(\pm1)}|_{T^{\perp}} \geq \tilde{c} > 0$.
Finally, by~(\ref{eq:mon}), $L_m^{(\pm1)} \geq c' > 0$
for $|m| \geq 2$. It follows from proposition~\ref{prop:decomp} that
$\tilde{L}^{(n)} \geq c > 0$ on the subspace of $X$ orthogonal
to the translational zero-modes. By the discussion of
section~\ref{subsec:gauge}, this gives theorem~\ref{thm:main} for $n = \pm1$.
$\Box$
\section{The critical case, $\lambda = 1$}
\label{sec:crit}
In order to prove the remainder of theorem~\ref{thm:main},
we exploit some results from the $\lambda = 1$ case.
\subsection{The first-order equations}
Following \cite{bog}, we use an integration by parts
to rewrite the energy~(\ref{eq:ac}) as
\begin{equation}
\label{eq:bog}
E(\psi,A) = \frac{1}{2} \int_{{\bf R}^2} \{ |\partial_A \psi|^2
+ [\nabla \times A + \frac{1}{2}(|\psi|^2-1)]^2 +
\frac{1}{4}(\lambda - 1)(|\psi|^2-1)^2 \}
+ \pi \mbox{deg}(\psi)
\end{equation}
(recall, since we work in dimension two, $\nabla \times A$ is a scalar)
where $\mbox{deg}(\psi)$ is the topological degree of $\psi$,
defined in the introduction.
We assume, without loss of generality, that $deg(\psi) \geq 0$.
Clearly, when $\lambda = 1$, a solution of the first-order equations
\begin{equation}
\label{eq:first1}
\partial_A \psi = 0
\end{equation}
\begin{equation}
\label{eq:first2}
\nabla \times A + \frac{1}{2}(|\psi|^2-1) = 0
\end{equation}
minimizes the energy within a fixed topological
sector, $\deg(\psi) = n$, and hence is stable.
Note that we have identified the vector-field $A$ with a complex
field as in~(\ref{eq:complex}).
The $n$-vortices~(\ref{eq:nvortex})
are solutions of these equations (when $\lambda = 1$).
Specifically,
\begin{equation}
\label{eq:firstode1}
n\frac{a'}{r} = \frac{1}{2}(1-f^2)
\end{equation}
and
\begin{equation}
\label{eq:firstode2}
f' = n\frac{(1-a)f}{r}.
\end{equation}
In fact, it is shown in~\cite{t2} that for $\lambda = 1$,
any solution of the variational equations solves the first-
order equations~(\ref{eq:first1}-\ref{eq:first2}).
Beginning from expression~(\ref{eq:bog})
for the energy, the variational
equations (previously written
as~(\ref{eq:eq1}-\ref{eq:eq2}))
can be written as
\begin{equation}
\label{eq:eq21}
\partial_A^*[\partial_A \psi] + \psi[\nabla \times A
+ \frac{1}{2}(|\psi|^2-1)] + \frac{1}{2}(\lambda - 1)
(|\psi|^2-1)\psi = 0
\end{equation}
\begin{equation}
\label{eq:eq22}
i\bar{\psi}[\partial_A \psi] - i\partial_{\bar{z}}[\nabla \times A
+ \frac{1}{2}(|\psi|^2-1)] = 0
\end{equation}
(here $\partial_A^* \equiv -\partial_z + iA$ is the
adjoint of $\partial_A$).
\subsection{First-order linearized operator}
We show that the linearized operator at $\lambda = 1$ is the
square of the linearized operator for the first-order equations.
Linearizing the first-order equations~(\ref{eq:first1}-\ref{eq:first2})
about a solution,
$(\psi, A)$ (of the first-order equations)
results in the following equations for the perturbation,
$\alpha \equiv (\xi, B)$:
\[
\partial_A\xi - iB\psi = 0
\]
\[
\nabla \times B + \Re(\bar{\psi}\xi) = 0.
\]
Now using $-i\partial_zB = \nabla \times B - i(\nabla \cdot B)$,
and adding in the gauge condition~(\ref{eq:gchoice}),
we can rewrite this as
\begin{equation}
\label{eq:1lin}
L_1 \alpha = 0
\end{equation}
where
\[
L_1 = \left( \begin{array}{cc}
\partial_A & -i\psi \\
\bar{\psi} & -i\partial_z
\end{array} \right).
\]
If we linearize the full (second order) variational
equations (in the form~(\ref{eq:eq21}-\ref{eq:eq22}))
around $(\psi, A)$, we obtain
\[
\partial_A^*[\partial_A\xi - iB\psi] + i\bar{B}[\partial_A \psi]
+ \psi[\nabla \times B + \Re(\bar{\psi}\xi)]
\]
\[
+ \xi[\nabla \times A + \frac{1}{2}(|\psi|^2-1)]
+ \frac{1}{2}(\lambda - 1)[(|\psi|^2-1)\xi + 2\psi\Re(\bar{\psi}\xi)] = 0
\]
and
\[
i\bar{\psi}[\partial_A\xi - iB\psi] + i\bar{\xi}[\partial_A \psi]
-i\partial_{\bar{z}}[\nabla \times B + \Re(\bar{\psi}\xi)] = 0.
\]
\begin{prop}
When $\lambda = 1$,
these linearized equations can also be written
\[
L_1^* L_1 \alpha = 0
\]
\end{prop}
{\em Proof:}
This is a simple computation using the fact that
the first-order equations~(\ref{eq:first1}-\ref{eq:first2}) hold. $\Box$
This relation holds also on the level of the blocks.
A straightforward computation gives
\[
L_m^{(n)}|_{\lambda = 1} = F_m^* F_m
\]
where
\[
F_m = \left( \begin{array}{cccc}
\partial_r - b & \frac{m}{r} & 0 & f \\
\frac{m}{r} & \partial_r - b & -f & 0 \\
0 & -f & \partial_r + 1/r & \frac{m}{r} \\
f & 0 & \frac{m}{r} & \partial_r + 1/r
\end{array} \right)
\]
\subsection{Zero-modes for $\lambda = 1$}
It was predicted in \cite{w} (and proved rigorously in \cite{s}) that
for $\lambda = 1$, the linearized operator around any degree-$n$
solution of the first-order equations has a $2|n|$-dimensional
kernel (modulo gauge transformations). This kernel arises because the
Taubes solutions form a $2|n|$-parameter family, and all have the
same energy. The zero-eigenvalues
are identified in \cite{bog}, and we describe them here.
Let $\chi_m$ be the unique solution of
\[
(-\Delta_r + \frac{m^2}{r^2} + f^2) \chi_m = 0
\]
on $(0,\infty)$ with
\[
\chi_m \sim r^{-m} \;\;\;\; \mbox{ as } \;\;\;\; r \rightarrow 0
\]
and
\[
\chi_m \rightarrow 0 \;\;\;\; \mbox{ as } \;\;\;\; r \rightarrow \infty
\]
for $m = 1,2,\ldots,n$. Then it is easy to check that
\begin{equation}
\label{eq:bogmodes}
F_{\pm m} W_m = 0
\end{equation}
where
\[
W_m = \left( \begin{array}{c}
f\chi_m \\ f\chi_m \\ -(\chi_m' + m\chi_m/r) \\
-(\chi_m' + m\chi_m/r)
\end{array} \right).
\]
We remark that
\[
\chi_1 = \frac{1-a}{r}
\]
and it is easily verified that for $\lambda = 1$,
$W_{\pm 1} = T$ are the translational zero-modes.
\section{The (in)stability proof for $|n| \geq 2$}
\label{sec:high}
Here we complete the proof of theorem~\ref{thm:main}.
The idea is to decompose $L_m^{(n)}$ into a sum of two
terms, each of which has the same (translational) zero-mode
(for $m=1$) as $L_m^{(n)}$. One term is manifestly
positive, and the other satisfies restrictions of
Perron-Frobenius theory.
We begin by modifying $F_m$, and defining, for any $\lambda$,
\[
\tilde{F}_m \equiv
\left( \begin{array}{cccc}
(\partial_r - \frac{f'}{f}) \cdot q & \frac{m}{r} & 0 & f \\
\frac{m}{r}q & \partial_r - \frac{f'}{f} & -f & 0 \\
0 & -f & \partial_r + 1/r & \frac{m}{r} \\
fq & 0 & -\frac{m}{r} & \partial_r + 1/r
\end{array} \right)
\]
where we have defined
\begin{equation}
\label{eq:q}
q(r) \equiv \frac{n(1-a)f}{rf'}
\end{equation}
and $\partial_r \cdot q$ denotes an operator composition.
By~(\ref{eq:firstode2}), we have
$q \equiv 1$ for $\lambda = 1$.
We also set, for $m=1, \ldots, n$,
\[
\tilde{W}_m = \left( \begin{array}{c}
q^{-1} f \chi_m \\ f \chi_m \\
-( \chi_m' + m \frac{\chi_m}{r} ) \\
-( \chi_m' + m \frac{\chi_m}{r} )
\end{array} \right)
\]
Now $\tilde{W_m}$ has the following properties:
\begin{enumerate}
\item $\tilde{W}_{\pm 1}$ is the translational zero-mode
$T$ for all $\lambda$
\item when $\lambda = 1$, $\tilde{W}_m = W_m$, $m = \pm 1, \ldots, \pm n$,
give the $2n$ zero-modes~(\ref{eq:bogmodes}) of the linearized operator.
\end{enumerate}
These $W_m$ were chosen in \cite{bog} as candidates
for directions of energy decrease (for $|m| \geq 2$)
when $\lambda > 1$.
Intuitively, we think of $\tilde{W}_m$ as a perturbation that tends
to break the $n$-vortex into separate vortices of lower degree.
Now, $\tilde{F}_m$ was designed to have the following properties:
\begin{enumerate}
\item $\tilde{F}_m = F_m$ when $\lambda = 1$ (this is clear)
\item $\tilde{F}_m \tilde{W}_m = 0$ for all $m$ and $\lambda$
(this is easily checked).
\end{enumerate}
A straightforward computation gives
\begin{equation}
\label{eq:keysplit}
L_m^{(n)} = \tilde{F}_m^* \tilde{F}_m
+ J M_m
\end{equation}
where $J = \mbox{diag} \{ 1, 0, 0, 0 \}$ and
\[
M_m = l_m - ql_mq + (\lambda - q^2)f^2
\]
with
\[
l_m = -\Delta_r + \frac{m^2}{r^2} + b^2 + \frac{\lambda}{2}(f^2-1).
\]
By construction, when $m=1$, the second term in the
decomposition~(\ref{eq:keysplit}) must have a zero-mode corresponding to the
original translational zero-mode. In fact, one can easily
check that $M_1 f' = 0$.
\begin{prop}
\label{prop:mp}
For $|n| \geq 2$,
$M_1$ has a non-degenerate zero-eigenvalue corresponding to $f'$,
and
\[
\left \{ \begin{array}{cc}
M_1 \geq 0 & \lambda < 1 \\
M_1 \leq 0 & \lambda > 1
\end{array} \right.
\]
on $L^2_{rad}$.
\end{prop}
{\em Proof:}
We recall inequality~(\ref{eq:ineq}), which implies that
for $\lambda < 1$, $q < 1$, and for $\lambda > 1$, $q > 1$.
The operator $M_1$ is of the form
\begin{equation}
\label{eq:firstterm}
M_1 = (1-q^2)(-\Delta_r) + \mbox{ first order } + \mbox{ multiplication. }
\end{equation}
One can show that
$M_1$ is bounded from below (resp. above) for $\lambda < 1$
(resp. $\lambda > 1$).
We stick with the case $\lambda < 1$ for concreteness.
Suppose $M_1 \eta = \mu \eta$ with $\mu = infspec M_1 \leq 0$.
Applying the maximum principle (eg proposition~\ref{prop:mainmp} for $d=1$)
to~(\ref{eq:firstterm}), we conclude
that $\eta > 0$. If $\mu < 0$, we have
$<\eta, f'> = 0$, a contradiction. Thus $\mu = 0$, and is non-degenerate
by a similar argument.
$\Box$
We also have
\begin{lemma}
\label{lemma:mon}
For $m \geq 2$, $M_m - M_1$
is non-negative for $\lambda < 1$, non-positive for $\lambda > 1$,
and has no zero-eigenvalue.
\end{lemma}
{\em Proof:} This follows from the equation
\[
M_m - M_1 = (1-q^2)\frac{m^2-1}{r^2}. \;\;\;\;\;\;\;\; \Box
\]
{\em Completion of the proof of theorem~\ref{thm:main}:}
Suppose now $\lambda < 1$.
Since $\tilde{F}_m^* \tilde{F}_m$ is manifestly non-negative,
and $M_m > M_1$ for $m \geq 2$, we have
$L_m^{(n)} \geq 0$ for $m \geq 1$ (with only the
translational $0$-mode). Combined with~(\ref{eq:contspec}) and
propositions~\ref{prop:l02} and~\ref{prop:decomp}, this gives stability
of the $n$-vortex for $\lambda < 1$.
Now suppose $\lambda > 1$. By~(\ref{eq:keysplit}), proposition~\ref{prop:mp}
and lemma~\ref{lemma:mon},
we have for $m = \pm 2, \ldots \pm n$,
\[
<\tilde{W}_m, L_m^{(n)} \tilde{W}_m > \;\;\; < \;\; 0.
\]
We remark that $\tilde{W}_m$ corresponds to an element of the un-complexified
space $X$, and so $L^{(n)}$ has negative eigenvalues.
This establishes the instability of the $n$-vortex
for $|n| \geq 2$, $\lambda > 1$,
and completes the proof of theorem~\ref{thm:main}.
$\Box$
\section{Appendix: vortex solutions are radial minimizers}
\begin{prop}
For $\lambda \geq 2n^2$,
a solution of the equations~(\ref{eq:ode1}-\ref{eq:ode2})
minimizes $E_r^{(n)}$.
\end{prop}
{\em Proof:}
It suffices then to show $M_0 = Hess E_r^{(n)} > 0$ (see
section~\ref{subsection:min}).
We write $M_0 = L_0 + Z_0$ where
\[
L_0 = diag\{ l, -\Delta_r \}
\]
with $l = -\Delta_r + b^2 + \frac{\lambda}{2}(f^2-1)$ and
\[
Z_0 = \left( \begin{array}{cc}
2\lambda f^2 & -2bf \\
-2bf & \frac{1}{r^2} + f^2 \end{array} \right).
\]
We note that $l f = 0$ (one of the GL equations). It follows from the
fact that $f > 0$ and a Perron-Frobenius type argument (see \cite{os1})
that $l \geq 0$ with no zero-eigenvalue. It suffices to show $Z_0 \geq 0$.
Clearly $tr(Z_0) > 0$, and
\[
\det(Z_0) = 2\lambda f^4 + \frac{2f^2}{r^2}[\lambda - 2n^2(1-a)^2]
\]
is strictly positive for $\lambda \geq 2n^2$. $\Box$
|
\section{Introduction}
The one-dimensional (1D) half-filled Kondo lattice is a simple model for a
group of compounds called ``the Kondo insulators''\cite{Aeppli}. They exhibit
the high-temperature behavior of usual Kondo systems, such as the
Curie-Weiss-like magnetic susceptibility, but
at lower temperatures evolve into a semiconducting
phase with small gaps. At low-energy, a
Kondo insulator is a typical realization of spin-charge
separation. This aspect manifests itself in the difference in size between the
charge gap and spin gap.
Since its discovery, the indirect
Ruderman-Kittel-Kasuya-Yosida (RKKY) interaction between localized magnetic
impurities embedded in a host metal has played an important role in the theory
of magnetism. In 1D the RKKY oscillation amplitude displays only a
very slow decay $\sim 1/x$. The conduction electrons are subject to the
resulting exchange field which oscillates spatially
with the Fermi wavelength. It produces Kondo localization. This conclusion
was obtained by an exact diagonalization study\cite{Tsunetsugu}. The result
has been confirmed with the density matrix renormalization group
method\cite{Yu}, and later supported by the
mapping to a non-linear sigma model\cite{Tsvelik} and by the bosonization
approach\cite{Zachar,KLH}. The charge gap varies linearly with the Kondo
coupling\cite{Shibata}. In this paper another important question is examined:
given the RKKY oscillation, what are the theoretical consequences for
the low-temperature spin properties?
The RKKY interaction (as we shall see later) cannot break the spin
rotational
symmetry due to Kondo localization. Fluctuations in
the local spin configurations lead to
a paramagnetic fluid. Since the model
for the local spin degrees of freedom is not purely integrable, no
exact solution has been proposed to date. In fact, by considering an
RKKY-like spin interaction, a topological term
{\it \`a la Haldane}\cite{Haldane0} is not
justified so that gapless magnons with S=1 should be the
prevalent excitations over the spin array. In consequence,
the spin effective model we shall study may be viewed
at intermediate distances as an
S=1/2 Heisenberg chain interacting with a `quasi-ordered' spin
array (with non-topological configurations). Conventionally, the low-lying
excitations (often called
spinons) of the S=1/2 Heisenberg spin chain are represented by one gapless
bosonic mode. In the language of the theory of critical phenomena, this means
that the model is described by a central charge which is equal to C=1.
For the present Kondo system, we will show how
the frustration imposed by the localized spins may play its tricks and the
bosonic mode or spinon decouples into two modes of Majorana fermions (with
C=1/2 each) having different spectra. The spinon splitting phenomenon is at
the basis of the new chiral fixed point presented here and also
occurs in spin ladders which retain the local
symmetry of the Kagom\'e net\cite{Azaria}. Note, however, that
in our problem including the
dynamics of localized spins will make the chiral fixed point unstable in the
far Infra Red (IR) limit. In
agreement with previous numerical results\cite{Tsunetsugu,Shibata}, the
system should flow to a disordered spin liquid phase with a finite
antiferromagnetic (AFM) length scale $\xi_{AFM}$ close to that in the
single impurity Kondo model. In the following, we shall introduce the Kondo
spin liquid phase as a direct consequence of the chiral fixed point
instability in the presence of strong fluctuations.
\vskip 0.1cm
The starting point is the Hamiltonian:
\begin{equation}
{\cal H}=-t\sum_{\langle i,j\rangle, \alpha}
c^{\dag}_{i,\alpha}c_{j,\alpha}+h.c. + J_K\sum_{i,\alpha,\beta}
c^{\dag}_{i,\alpha}\vec{\sigma}_{\alpha,\beta}c_{i,\beta}.\vec{S}_i
\end{equation}
The first term represents the electron hopping between nearest-neighbor sites
$i$ and $j$. The second term is the Kondo coupling $(J_K\ll t)$ between the
localized
spin $\vec{S}_i$ and the mobile electron at the same site. Adding a direct
exchange $J_H>0$ between the nearest {\it core} spins gives quite
different physics. It ensures the presence of local AFM fluctuations
leading to topological configurations for the localized spins. If the relation
$J_H\gg J_{RKKY}\sim {J_K}^2/t$ is satisfied, the spin theory at low
temperature
is described in terms of an O(3) nonlinear $\sigma$ model where the
topological term has no contribution\cite{Tsvelik,Betouras}. That produces
disordered Kondo fluids with quite short AFM correlation lengths
$\xi_{AFM}\simeq \exp(\pi S)$. Excitations of the O(3) nonlinear $\sigma$
model are S=1 triplets. In the extreme limit $J_H\gg J_K$, the physics
becomes similar to the two-leg spin ladder\cite{Rice}; spinons confine to
form both S=1 triplet and singlet excitations with gaps $m_t$ and $m_s$
$(m_t,m_s\propto J_K^2$ and $\xi_{AFM}\sim m_t^{-1}$)\cite{White,KLH2}.
Here, we mainly focus on the interesting case $J_H\rightarrow 0$.
\section{Basis of our formalism}
The approach followed in this paper is based on bosonization techniques
for both charge
and spin degrees of freedom of the conduction electrons. For complete
reviews, see refs.\cite{Haldane,affleck}.
For $J_K=0$, the model for the
conduction band is gapless and is characterized by the separation of spin
and charge. Its low-energy spin properties belong to the same universality
class as the Heisenberg model. The low-temperature behavior is then
described by the level-1 SU(2) Wess-Zumino-Witten (WZW) conformal field theory
(CFT)\cite{Tsvelik2}.
The physical particles (or spinons=spin 1/2 excitations)
are included through the primary
fields $\Phi^{(1/2)}$ and $\Phi^{(1/2)\dag}$ from the representation of the
$SU(2)$ group\cite{Bernard}. The WZW action taking into account the dynamics of
the spinon
objects is explicitly given by:
\begin{eqnarray}
& &S_{WZW}=-\frac{1}{16\pi}\int d^2 x
Tr(\partial_{\mu}\Phi^{(1/2)\dag}\partial_{\mu}\Phi^{(1/2)})\\ \nonumber
&+&\frac{i}{24\pi}\int_0^{\infty} d\epsilon\int d^2 x
\epsilon^{\alpha\beta\gamma} Tr({\cal A}_{\alpha}{\cal A}_{\beta}{\cal A}_
{\gamma})\\
\end{eqnarray}
and, ${\cal A}_{\alpha}=\Phi^{(1/2)\dag}\partial_{\alpha}\Phi^{(1/2)}$.
This description, which has its origin in the structure of the Haldane-Shastry
spin chain with $1/x^2$ exchange\cite{HS}, stresses the fact that the
fundamental
fields in this theory (or spinon fields) may be viewed as free fields apart
from purely statistical (in this case: semionic\cite{Bouwknegt}) interactions
that may be
taken into account by a rule generalizing the Pauli principle.
The electronic spin density is represented as
$\vec{S}_c(x)=\vec{J}_c(x)+e^{i2k_Fx}\vec{n}_c(x)$, where
$\vec{J}_c=\vec{J}_{cR}+\vec{J}_{cL}$ and
\begin{equation}
\vec{n}_c=\frac{1}{2\pi
a}\hbox{Tr}\{\vec{\sigma}(\Phi^{(1/2)}+\Phi^{(1/2)\dag})\}\cos{\sqrt{2\pi}
\Phi_c}
\end{equation}
are,
respectively, the smooth and staggered parts of the magnetization.
The precise relationship between the chiral spin currents and the
spinon fields has been discussed in detail in ref.\cite{affleck}.
The lattice
step $a$ defines the required short-distance cut-off.
The charge sector
is similarly described in terms of a U(1) scalar field $\Phi_c$ leading to a
CFT
central charge C=1 for holons\cite{BA} as well. In
the charge sector, the free action
is given by:
\begin{equation}
S_{g;v_{\rho}}=\frac{1}{2}\int dxd\tau\
\frac{1}{v_{\rho}g}(\partial_{\tau}\Phi_c)^2-\frac{v_{\rho}}{g}(\partial_x
\Phi_c)^2
\end{equation}
Neutral excitations are 1D acoustic
plasmons which propagate
with velocity $v_{\rho}$ and are characterized by the Luttinger parameter
$g=v_F/v_{\rho}$. For a 1D free electron gas $g$ is equal to one and
$v_F=2t\sin(k_Fa)$ is the velocity for the charge and spin degrees of
freedom. In the continuum limit the Kondo interaction takes the form:
\begin{eqnarray}
\label{zero}
{\cal H}_{int}&=&\lambda_2(\vec{J}_{cL}+\vec{J}_{cR})\vec{S}_j\\ \nonumber
&+&\frac{\lambda_3}{2\pi a}e^{i(2k_F-\pi)x}\ \hbox{Tr}(\vec{\sigma}
\Phi^{(1/2)}).\cos\sqrt{2\pi}{\Phi}_c\vec{S}_j+h.c.
\end{eqnarray}
where $\lambda_{2,3}\propto J_K$. Note that $\delta=2k_F-\pi$ measures the
deviation
from half-filling. Here, we mainly consider the half-filled case
$\delta\rightarrow 0$. The configuration of localized spins can be also
parameterized as
\begin{equation}
\vec{S}_j=S(a\vec{L}_j+(-1)^x
[1-(a{L}_j)^2]^{1/2}\vec{n}_j)
\end{equation}
where $j$ enumerates the sites and
$\vec{L}_j.\vec{n}_j=0$. Finally, to quantify correctly the local spins, we
must use the path integral representation. It leads to an extra term in the
action (the so-called Berry phase)\cite{Fradkin,Tsvelik}:
\begin{equation}
S_{B}=iS\int dxd\tau\ \vec{L}.(\vec{n}\times \partial_{\tau}\vec{n})
\end{equation}
As already mentioned in the introduction, the RKKY interaction between local
moments should play a crucial role, for $J_K\ll t$.
\section{RKKY interaction as a staggering field}
The standard treatment of the
RKKY interaction corresponds to a calculation of the correlation
function
\begin{equation}
C(x)=\frac{\cos(2k_Fx)}{2\pi a}\langle S_0^z
\hbox{Tr}{\sigma}^z\Phi^{(1/2)}(x)\cos\sqrt{2\pi}{\Phi}_c(x)\rangle
\end{equation}
The
angular brackets
indicate a thermal average. This function describes the spatial correlation of
the electron-spin density with the impurity spin on $j=0$. Another impurity
spin located at $j=x$ would see this correlation, and lowest-order perturbation
theory in $\lambda_3$ constitutes an exact derivation of the RKKY law. For $
x<\xi_T=v_F/\hbox{k}_B T$, this yields: $H_{RKKY}=\lambda_{RKKY}\sum_{\langle
0,x\rangle}\vec{S}_0\vec{S}_x$
with
\begin{equation}
\label{trois}
\lambda_{RKKY}=\frac{\lambda_3}{2\pi a}C(x)=\frac{-\lambda_3^2}{2\pi v_F}
\frac{\cos(2k_Fx)}{x^g}a^{-1}
\end{equation}
In the range of temperature $x<\xi_T$, the RKKY oscillation amplitude
displays only a very slow algebraic decay. In the noninteracting case $g=1$,
the usual $x^{-1}$ decay is recovered.
The RKKY law, prominent at
high temperatures,
should then favor the following configuration for the localized
spins: $\vec{n}_j=\frac{1}{a}\vec{n}$ where $\vec{n}$ is a {\it fixed} unit
vector and
$1-(a{L}_j)^2\sim 1$. Using the Hamiltonian (\ref{zero}), we find that it
generates a static staggering magnetic field
$h_s=S\lambda_3$ on conduction electrons:
\begin{eqnarray}
{\cal H}_{int}&=&h_s\vec{n}_c\cdot\vec{n}\\ \nonumber
&=& \frac{h_s}{2\pi a^2}\cos\sqrt{2\pi}{\Phi}_c\hbox{Tr}(\vec{\sigma}
\Phi^{(1/2)})\cdot\vec{n}+h.c.
\end{eqnarray}
At high temperatures, the conduction electrons are subject to a perfectly
static potential as if there was a finite staggered spin moment.
Now, we expand the partition function ${\cal Z}={\cal Z}_0+\delta{\cal Z}
[{h_s\neq 0}]$
to the second order in $h_s$. We find that
$\delta {\cal Z}/{\cal Z}_0$ is equal to\footnote{To simplify the
expression, we have taken $v_{\rho}\sim v_F=1$}:
\begin{equation}
\hbox{\Large(}\frac{hs}{2\pi}\hbox{\Large)}^2
\int \frac{d\tau_1 d\tau_2 dx_1 dx_2}{a^4}\ \hbox{\Large(}
\frac{(x_1-x_2)^2+(\tau_1-\tau_2)^2}{a^2}\hbox{\Large)}^{-\frac{1+g}{2}}
\end{equation}
${\cal Z}_0$ is the partition function of the free system with $J_K=0$.
To make the (total) partition function
invariant under the cut-off transformation $a\rightarrow a'=ae^{d\ln L}$,
$h_s$ has to obey:
\begin{equation}
\frac{dh_s}{d\ln L}=(2-\frac{1}{2}-\frac{g}{2})h_s
\end{equation}
Note that ${\cal Z}_0$ is not affected by such a rescaling.
Starting with a free electron gas
$(g\rightarrow 1)$, we confirm that it produces
the localization of conduction electrons at an energy scale $\Delta_c\propto
h_s=J_K/2$\cite{Tsvelik,Shibata}, defined such as $h_s(\Delta_c)\sim 1$.
It should be noted that
spin flip events do not contribute to the perturbative
result (\ref{trois}), to this order. Of course, the exponent $g$ is
also affected by the Kondo interaction. We find:
\begin{equation}
\frac{dg}{d\ln L}=-2\pi g^2\lambda_3^2 J_o(\delta(L)a)
\end{equation}
$J_o$ is the Bessel function. Since we restrict our arguments to
the case of half-filling, we put $\delta=0$ and $J_o(0)=1$.
Correlations of $\cos\sqrt{2\pi}{\Phi}_c$ show a long-range
order at zero temperature because the renormalized exponent $g^*$ goes
to zero. We can average
$\langle\cos\sqrt{2\pi}{\Phi}_c(x)\rangle\sim\sqrt{\Delta_c}$. The charge
motion is frozen and the leading order parameter in the
charge sector is the so-called
$4k_F$ charge density wave (CDW): $\rho_{4k_F}\propto \cos
\sqrt{8\pi}{\Phi}_c(x)$. As for the
occurrence of the Mott-Hubbard gap due to Umklapps\cite{Haldane},
the (Mott)-Kondo
insulating state occurs due to the pinning of the $4k_F$ CDW\cite{remark}:
\begin{equation}
\langle \rho_{4k_F}(x)\rho_{4k_F}(0)\rangle \propto x^{-4g^*}\sim
\hbox{constant}
\end{equation}
The charge field, on the other hand, could also be affected by
the disorder. As shown in ref.\cite{KLH}, a strongly disordered 1D Kondo array
will crossover to an Anderson localization state. Thus, we now study
spin properties of such an insulating (Mott)-Kondo state with very small
randomness.
Since $\lambda_3$ violates the separation of charge and spin one
can expect massive triplet modes with a spectral gap
$J_KS$. At quite short distance, one can treat $\vec{n}$ as a constant
vector. Writing $\Theta=i\vec{\sigma}.\vec{n}$, the result is a uniaxial Kondo
effect, leading to an Ising-like solution $\Phi^{(1/2)}=\Phi^{(cl)}$.
Using properties of Pauli matrices\cite{Betouras,Shelton}:
\begin{equation}
\hbox{Tr}\vec{\sigma}\Phi^{(cl)\dag}\hbox{Tr}
\vec{\sigma}\Theta+h.c.=\hbox{Tr}\Phi^{(1)},\
\Phi^{(1)}=\Phi^{(cl)\dag}\Theta
\end{equation}
we check that spins of
conduction and localized electrons confine to form triplets (or magnons
represented by the `composite' field $\Phi^{(1)}$) with a mass $m_t=
\Delta_c$ . Such a classical solution is ruled by:
\begin{equation}
\vec{n}_c=
\hbox{Tr}\vec{\sigma}\Phi^{(cl)\dag}=-\hbox{Tr}\vec{\sigma}\Theta=-\vec{n}
\end{equation}
When $x\simeq {\cal L}_{loc}$ with ${\cal L}_{loc}=(J_K S)^{-1}$, the
so-called spin density glass state aims to arise\cite{KLH}: conduction
electrons feel a staggered potential within this scale and open a
quasiparticle gap due to Bragg scattering.
However, starting with an isotropic system, it cannot be longer stabilized for
temperatures much lower than $\Delta_c$. Indeed, from Eq. (\ref{trois}), we
find the following recursion law for the RKKY exchange in the
delocalized phase:
\begin{equation}
\frac{d\lambda_{RKKY}}{d\ln L}=\frac{1}{2\pi v_F}{\lambda_3}^2
\end{equation}
It cannot provide an energy scale more relevant
than $\Delta_c$ since it involves processes in ${\lambda_3}^2$ (
the electron gas localization occurs due to scattering events $\propto
\lambda_3$). Then, the conduction electrons
effectively `screen away' the internal field before a true magnetic transition
(with breaking SU(2) symmetry) can occur.
\section{SU(2) symmetry not broken: consequences}
The uniaxial solution
$\Theta(x,\tau)=\Theta$ and
$\Phi^{(1/2)}=\Phi^{(cl)}$ is not available at long distances: {\it there are
certainly states in the gap}.
For $J_H\gg J_{RKKY}$, it is important to include a topological term (for
the local moments) first derived by Haldane\cite{Haldane0}:
\begin{equation}
\label{to}
S_{top}=\frac{i}{8\pi}\int dxd\tau \epsilon_{\mu\nu}\hbox{\large(}\vec{n}
\cdot[
\partial_{\mu}\vec{n}\times\partial_{\nu}\vec{n}]\hbox{\large)}
\end{equation}
Then, the model becomes solvable in the semi-classical limit
(`large-S expansion'), by including the Berry phase and integrating out
both conduction electron variables and fast modes $\vec{L}$. The result is
an 0(3) nonlinear $\sigma$ model built out
the field $\vec{n}(x,\tau)$ (with
triplet excitations). The resulting topological term has no
contribution and then
the gapless ordered state of the isotropic
sigma model is marginally unstable (due to the implicit breaking
of conformal invariance) and it opens a gap (`the so-called Haldane gap')\cite
{Haldane0,Tsvelik}:
\begin{equation}
m\sim J_K S\exp -\pi S
\end{equation}
which is not so far from $\Delta_c$.
The spin gap $m$ has a topological origin, and
excitations in the interval between
$m$ and $J_K S$ are known to
be massive spin polarons formed due to an interaction
between electrons and kinks of the unit vector
$\vec{n}(x,\tau)$\cite{Tsvelik}. To my knowledge, such Kondo insulator
(ruled by two different energy scales in charge and spin sectors)
was the first realization of so-called spin-charge separation.
On the other hand, a
direct exchange between localized moments is crucial for the occurrence of
massive spin polarons\cite{Betouras} and such a topological description
is not expected
to remain available when $J_H=0$.
Numerical results\cite{Tsunetsugu,Shibata} predict in
contrast a pure Kondo ground state with a very large AFM length
$\xi_{AFM}$ driven by
spin flip processes or the Kondo term $\lambda_2$, which
has not been taken into account in the semi-classical approach of
ref.\cite{Tsvelik}.
To obtain precisely low-energy spin excitations in this limit we adopt
the following scheme.
For $J_H=0$, there is no mathematical justification
to include a topological term of the form of Eq.(\ref{to}) for
local moments since the model is not integrable. Then we
do not introduce it. In consequence, we start
with a spin array which is
described by usual spin waves with spin ${\Delta
S}^z=\pm 1$. Since a true magnetic transition does not arise,
ferromagnetic local fluctuations may occur in the
far IR limit leading to {\it free} triplet particles in the spectrum.
The forward Kondo scattering term should then play a crucial role
at very long distances. The effect of
fast variables $L_j$ will be considered later
when $x\sim \xi_{AFM}$ (i.e. when the presumed quasi long-range order
is completely destroyed). Using the so-called operator
product expansions of ref.\cite{Fujimoto}, we can check that
$\lambda_2$ is not renormalized by a term like $g^*{\lambda_3}^2/2\pi v_F$
due to the {\it insulating} nature of the ground sate ($g^*\rightarrow
0$). Ferromagnetic fluctuations are well decoupled if we
neglect the Berry phase (which is only reponsible for the
correct quantization of local spins).
Before, let us show that spinons in the electron gas
cannot {\it vanish} (confine) completely when the spin array
develops only a {\it quasi-order} (with isotropic
correlation functions): the spin density glass state becomes
less stable and critical Ising modes can survive in the spectrum.
\subsection{Refermionization and critical Ising modes}
Let us start with a spin array which satisfies the following quasi-order
condition:
\begin{equation}
\langle
n^z(x)n^z(0)\rangle=\langle n^{+}(x)n^{-}(0)\rangle\propto x^{-2\alpha}
\end{equation}
It means that the local magnetization operator $n^i(x,\tau)$ with $i=x,y,z$ has
the scaling dimension $\alpha$. Since the
RKKY interaction between local moments yields only
a very slow algebraic decay in 1D, we deduce:
$\alpha\ll 1/2$, 1/2 being the scaling dimension of the staggered magnetization
operator in the Heisenberg chain with nearest neighbor exchange\cite{affleck}).
To solve the problem at intermediate
distances (when
${\cal L}_{loc}\ll x\ll \xi_{AFM}$), we assume that the parameter
$\alpha$ is too small that we can replace in first
approximation local moment operators
by effective expectation values
$\langle n^z(x,\tau)\rangle\sim\langle n^{\perp}(x,\tau)\rangle=
\gamma\neq 0$,
obtained by averaging in time and space slow fluctuations in
the spin array. The electron gas is submitted to an SU(2)-invariant
{\it quasi-static} staggering potential.
Then, it is not difficult to integrate out
local spin degrees of freedom.
To treat correctly the electron gas, it is now convenient to use the
Abelian representation\cite{KLH2},
\begin{equation}
\vec{n}_c\sim(\cos\sqrt{2\pi}\Theta_s,\ -\sin\sqrt{2\pi}\Theta_s,\
\sin\sqrt{2\pi}\Phi_s)
\end{equation}
where $\Theta_s$ is the field dual to $\Phi_s$. Then,
at long distances, the term
$\lambda_3$ can be symmetrized as\cite{KLH}:
\begin{eqnarray}
\label{deux}
{\cal H}_{int}&\sim&\frac{1}{2\pi a}\Delta_c^{3/2}(\sin\sqrt{2\pi}\Phi_s(x)+
\sin\sqrt{2\pi}\Theta_s(x))\\ \nonumber
&\sim& \frac{\Delta_c}{2\pi a}(\cos\sqrt{4\pi}\Phi_s(x)+\cos\sqrt{4\pi}
\Theta_s(x))
\end{eqnarray}
We have replaced the charge operator by its expectation value and
used the definition of $\Delta_c=J_K S$.
In view to respect conventional notations we have changed $\Phi_s\rightarrow
\Phi_s+\sqrt{\pi}/4$ and $\Theta_s\rightarrow \Theta_s+\sqrt{\pi}/4$ in the
second equation. In ref.\cite{KLH}, we did not solve the Hamiltonian in
the isotropic case. To solve it, we require an effective fermionic theory
similar to Hubbard\cite{Schulz} or spin ladder\cite{Shelton} models, and
carbon nanotube problems\cite{Egger}.
The refermionization technique is defined, as follows.
Let us define new effective fermion
operators for the
spin channel. Right- and left-moving components $(q=\pm=R,L)$ can be written
in terms of the bosonic phase field,
\begin{equation}
\psi_{sq}(x)=\frac{\eta_q}{\sqrt{2\pi a}}\exp
\large\{-i \sqrt{\pi}(q\Theta_s+\Phi_s)(x)\large\}
\end{equation}
Then we have
\begin{eqnarray}
\label{un}
\frac{1}{\pi a}\cos\sqrt{4\pi}\Phi_s&=&\eta_R\eta_L(\psi_{sR}^{\dag}\psi_{sL}-
\psi_{sL}^{\dag}\psi_{sR}),\\ \nonumber
\frac{1}{\pi a}\cos\sqrt{4\pi}\Theta_s&=&-\eta_R\eta_L(\psi_{sR}^{\dag}
\psi_{sL}^{\dag}-\psi_{sL}\psi_{sR}).
\end{eqnarray}
Klein factors have been chosen as $\eta_R\eta_L=i$.
Apart from the usual mass bilinear term (which favors the pinning
of the spin density wave towards the easy-axis), the Hamiltonian
${\cal H}_{int}$ also contains a `Cooper-pairing' term originating from the
cosine of the dual field, which
guarantees the presence of quantum fluctuations at
zero temperature. Now, we introduce two Majorana fields
\begin{equation}
\xi_{1\nu}=\frac{\psi_{s\nu}+\psi_{s\nu}^{\dag}}{\sqrt{2}},\hskip 0.5cm
\xi_{2\nu}=\frac{\psi_{s\nu}-\psi_{s\nu}^{\dag}}{\sqrt{2}i},\hskip 0.5cm
(\nu=R,L)
\end{equation}
and by using (\ref{un}) we find\cite{Shelton,Egger}
\begin{eqnarray}
\frac{1}{\pi a}\cos\sqrt{4\pi}\Phi_s&=&i(\xi_{1R}\xi_{1L}+
\xi_{2R}\xi_{2L}),\\ \nonumber
\frac{1}{\pi a}\cos\sqrt{4\pi}\Theta_s&=&-i(\xi_{1R}\xi_{1L}-\xi_{2R}\xi_{2L}).
\end{eqnarray}
Refermionization of the spin sector then yields
\begin{eqnarray}
{\cal H}(s)&=&\frac{-i v_F}{2}\sum_{j=1}^2\int dx\
(\xi_{jR}\partial_x\xi_{jR}-\xi_{jL}\partial_x\xi_{jL})\\ \nonumber
&+&i\Delta_c\int dx\ \xi_{2R}\xi_{2L}
\end{eqnarray}
Of course, for $m_2=-\Delta_c$\footnote{Including the nonuniversal
number $\gamma<1$ (since ${\vec{n}}^2\sim 2{\gamma}^2=1$) in the Eq.(23),
the mass $m_2$ will be slightely rescaled as:
$m_2^*=m_2\gamma=-\Delta_c\gamma$.}, the model flows to strong
couplings rendering
the Majorana field $\xi_2$ massive. The Majorana fermion
$\xi_1$ remains {\it massless}.
The Hamiltonian ${\cal H}(s)$ shows
explicitly that
the bosonic mode $\Phi_s$ (or $\Phi^{(1/2)}$ in the non-Abelian language)
decouples into two modes of real (Majorana) fermions (or half-spinons) having
different spectra. We have a splitting of the
spinon field and it should lead to $\langle\hbox{Tr}\vec{\sigma}\Phi^{(1/2)}
\rangle\not=\hbox{constant}$.
The spin-singlet real fermions $\xi_1$ remain
free. Let us stress that it contributes to a {\it new} chiral fixed point.
The specific heat is still linear at low
temperatures, $T\ll \Delta_c$ and comes from gapless spin excitations. Using
the general formula $C_V=\pi CT/3v$\cite{aff2}, we find
$C_V=\pi T/6v_F$. A free
Hamiltonian of Majorana fermions is ruled by a central charge $C=1/2$
\cite{Tsvelik2}.
\subsection{Remnant of electronic spin fluctuations}
To compute spin correlation functions, it is
accurate to exploit the well-known correspondence between the 2D Ising model
and 1D Majorana fermions. Both are described by $C=1/2$. We
can define two decoupled Ising models as follows\cite{Itzykson},
\begin{eqnarray}
\label{quatre}
\cos\sqrt{\pi}\Phi_s&=&\mu_1\mu_2,\qquad \sin\sqrt{\pi}\Phi_s=\sigma_1\sigma_2
\\ \nonumber
\cos\sqrt{\pi}\Theta_s&=&\sigma_1\mu_2,\qquad \sin\sqrt{\pi}\Theta_s=\mu_1
\sigma_2
\end{eqnarray}
Bosonic exponents are expressed in terms of the order ($\sigma$) and disorder
($\mu$) parameters of two Ising models. On the other hand, a theory of a
massive Majorana fermion field describes long-distance properties of the
two-dimensional Ising model, the fermionic mass being proportional
to $m\sim (T-T_c)/T_c$. We conclude that ${\cal H}(s)$ is equivalent to two
decoupled
2D Ising models. The Ising model $(\sigma_1,\mu_1)$ related to $\xi_1$ will be
critical ($T=T_c$), while the Ising model $(\sigma_2,\mu_2)$ related to
$\xi_2$ will be {\it under} criticality (since $m_2=-\Delta_c<0$ we have
$T<T_c$). To perform the complete correspondence, using Eq. (\ref{quatre})
one obtains: $\xi_1\sim\cos\sqrt{\pi}(\Phi_s+\Theta_s)\sim \sigma_1\mu_1$ and
$\xi_2\sim\sin\sqrt{\pi}(\Phi_s+\Theta_s)\sim \sigma_2\mu_2$.
Since the second Ising model is under criticality, the average of the
disorder operator
is zero, $\langle\mu_2\rangle=0$. The order operator $\langle\sigma_2\rangle$
has then a finite value, and correlation functions of $\cos\sqrt{\pi}\Phi_s$
and $\cos\sqrt{\pi}\Theta_s$ decay exponentially. From the exact solution of
the 2D Ising model one immediately obtains\cite{Itzykson},
\begin{equation}
\langle
\sin\sqrt{\pi}\Phi_s(x)\sin\sqrt{\pi}\Phi_s(x')\rangle\sim\left|x-x'\right|
^{-1/4}
\end{equation}
with the same result for the $\sin\sqrt{\pi}\Theta_s$ operator. The above
proves that the fields $\Phi_s$ and $\Theta_s$ are not pinned: values of
$\sin\sqrt{\pi}\Phi_s$ and $\sin\sqrt{\pi}\Theta_s$ are not fixed to 1. By
virtue of the Heisenberg uncertainty relation, we have simply shown that it
is impossible to completely pin a self-dual field. Note that for
$\lambda_3=0$, scaling dimensions of all these bosonic operators is 1/4. For
$\lambda_3\not =0$, it rescales $1/4\rightarrow 1/8$. More generally, the
operators $\sin\sqrt{2\pi}\Phi_s$ and $\sin\sqrt{2\pi}\Theta_s$ acquire
a `halved' scaling dimension: $\eta=\beta^2/8\pi$ with $\beta=\sqrt{2\pi}$
(in respect to the ground
state $\cos\sqrt{2\pi}\Theta_s$ ($\cos\sqrt{2\pi}\Phi_s$) tends to zero).
The uniform part of the correlation functions decays exponentially. The nature
of this new chiral fixed point mainly manifests itself in {\it staggered}
{\it fusion rules} between spins of conduction electrons:
\begin{equation}
\hskip -0.15cm \hbox{Tr}\sigma^a\Phi^{(1/2)\dag}({\bf
x})\hbox{Tr}\sigma^{b}\Phi^{(1/2)}({\bf
0})\sim\frac{\delta_{ab}}{z^{1/2}}+\epsilon^{abc}(\frac{z}{\bar{z}})^{1/2}J_c^c
\hskip -0.15cm
\end{equation}
with ${\bf x}=(x,\tau)$, ${\bf 0}=(0,0)$ and $z,\bar{z}=x\pm iv_F\tau$. The
occurrence of the factor $(z/\bar{z})^{1/2}$ is usual and confirms that
braiding properties of spinons are those of
semions\cite{Bouwknegt}. On the other hand, we emphasize that the halved
scaled exponent 1/2 in the first term characterizes a new universality class
for the
S=1/2 Heisenberg chain which is finally equivalent to a C=1/2 critical
theory. In this new chiral fixed point half of the spinon field still
fluctuates. Regarding simply a spinon (spin 1/2 topological object)
as a domain wall
$(...\downarrow\uparrow\downarrow\downarrow\uparrow\downarrow...)$, the fact
that the electronic
$\xi_2$ modes acquire a mass (in our language) means physically
that the other half participates in bound states with local moments.
Taking into account local moment degrees of freedom,
observable massive excitations are triplets only. Then, we have:
\begin{equation}
\Phi^{(1)}+\xi_1=\Phi^{(1/2)\dag}\Theta(x,\tau)
\end{equation}
$\Phi^{(1)}$ being defined in Eq.(16). This is the main result of
the present work\cite{exp}.
\vskip 0.1cm
It
can lead to interesting experimental predictions. A related quantity of
direct experimental relevance is the NMR relaxation rate
${\cal T}_1$. For large $J_K$, the RKKY regime fails $(C(x)\rightarrow 0)$
and the
main contribution comes from the staggered spin-spin correlation functions in
the electron gas. The corresponding susceptibility has the temperature
dependence
\begin{equation}
\chi(2k_F)\sim T^{2\eta -2}\ \hbox{with}\ 2\eta=1/2
\end{equation}
Then, we predict
an NMR relaxation rate, $1/{\cal T}_1\propto T^{-1/2}$. This $T$-dependence
is quite unusual because the underlying magnetic order between electrons
changes the effective field seen by the nuclei. But a related behavior has
also been
reported for the so-called dimerized, frustrated and ladder models when the
magnetic field reaches the critical field $H_{c}$ (typically the spin
gap)\cite{Chitra}. The resulting ${\cal T}_1$ is different from that in the
single impurity case. There,
Friedel oscillations in the unitary limit lead to an NMR rate which in
contrast increases with the distance\cite{Eggert,Egger1}.
\section{Fixed point and conclusion}
Now, let us turn to the stability of the chiral phase
when one moves away from $(aL)^2=0$ in the far IR limit, where strong
fluctuations are permitted. There,
$1-(a\vec{L})^2\sim{\gamma}^2
\rightarrow 0$: the $\lambda_3$ Kondo exchange becomes
$2k_F$ oscillating, leading to free interchain S=1
spins in the spectrum $(m_2^*\rightarrow 0)$. However, the
charge sector should not be
affected because the localization length is sufficiently large, ${\cal
L}_{loc}\gg a$. Integrating out the remnant of massive $\xi_2$ modes,
the leading far IR behavior of the $q=0$ electronic spin component coincides
again with the one in the effective S=1/2 spin chain.
At small enough
$\lambda_2$, the forward Kondo exchange can be viewed as a weak perturbation
to the chiral fixed point. Starting from order ${\lambda_2}^2$ on, spin flips
contribute:
\begin{equation}
\frac{d\lambda_2}{d\ln L}=\frac{1}{2\pi v_F}{\lambda_2}^2
\end{equation}
The system flows to a purely Kondo state at an energy scale:
\begin{equation}
m\sim J_K S\exp-2\pi v_F/\lambda_2\ll\Delta_c
\end{equation}
typically the single-site Kondo temperature.
Both $\xi_1$ and $\xi_2$ acquire a mass. We do not predict a particular
lattice enhancement effect\cite{Tsunetsugu}. A similar conclusion has
been obtained in ref.\cite{Fujimoto} (but there is an important mistake in
the recursion law for the Kondo coupling $\lambda_3$). All the spin
correlation functions now decay exponentially.
Such a
Kondo liquid is another theoretical example of spin-charge
separation (occurring in the 1D Kondo lattice model) in
agreement with experiments on three dimensional
Kondo insulators\cite{Aeppli} and previous
numerical results\cite{Shibata}. This can be shown by the
absence of a field-induced metal-insulator transition\cite{Carruzzo}.
The system remains insulating even when a finite magnetization is induced by
the external field because
$m\ll\Delta_c$. The most typical case is the large-U limit of the 1D
Hubbard model. The charge gap is of order U while magnetic excitations are
gapless. During the magnetization process (occurring for a small field $\sim
t^2/U$), it does not change.
In summary, there exists an intermediate but still low-energy region where the
chiral fixed point presented here is dominant. It brings new physics in
Kondo ladder systems. Indeed, a spin liquid with both massive
triplet and critical Ising modes takes place. It should be noted that similar
spin spectra have been occurred in other coupled spin chain systems, like
the so-called Kagom\'e lattice
antifferomagnet stripped to its basis\cite{Azaria} or the
two-leg spin ladder with a dimer four-spin interaction\cite{Ners}.
On the other hand,
strong fluctuations among localized spins in the far IR produce
gapless triplet excitations so that the chiral phase becomes unstable. Driven
by the dynamics of localized spins, the system flows to a typical Kondo
fluid. Based on previous numerical results, the gap value is similar to the
single-impurity Kondo temperature.
The author thanks B. Dou\c cot, M. Fabrizio and T. Jolicoeur for fruitful
remarks and constructive criticisms, and the theoretical group of ETH for
stimulating discussions.
|
\section{Introduction}
Recent observations of the infra-red (IR) and sub-millimeter (sub-mm)
extra-galactic background (Puget et al 1996; Fixsen et al 1998;
Burigana \& Popa 1998; Schlegel, Finkbeiner \& Davis 1998; Hauser et
al 1998; Dwek et al 1998; Biller et al 1998) provide powerful
constraints on models of galaxy evolution, since a large fraction
($\sim 2/3$) of stellar energy release, after reprocessing by dust, is
emitted in this part of the spectrum. Various number count surveys are
also being pursued across the waveband, from SCUBA at $850\:{\rm \mu
m}$ (Blain et al 1999; Smail et al 1998), to ISO at $175\:{\rm \mu m}$
(Puget et al 1998) and $15\:{\rm \mu m}$ (Aussel et al 1998; Altieri
at al 1998). All of the evidence points to strong evolution in the
galaxy population as we look out to high redshift. We would like to
try and understand this behavior.
Several groups have attempted to model the IR and sub-mm emission from
galaxies, using forward evolution models (Blain \& Longair 1993;
Franceschini et al 1994; Guiderdoni et al 1998; Jimenez \& Kashlinsky
1998; Trentham, Blain and Goldader 1999). Guiderdoni's model is one of
the most detailed. They model structure formation in the Cold Dark
Matter (CDM) cosmological scenario, and introduce star formation with
various recipes for ``quiet-disk'', ``burst'' and ``ultra-luminous''
sources. The relative fractions of these populations and the
timescales of star formation are free parameters in their model. By
tuning these they are able to reproduce current observations, but
there is a lack of physical motivation in many of their assumptions
concerning the number and luminosity evolution of their populations.
In contrast to these models, Malkan \& Stecker (1998) present a
simple, empirically based calculation of the IR background. Using
observed luminosity functions and spectral energy distributions, they
predict reasonable values for the background flux by implementing
various schemes of galaxy evolution. However, these prescriptions are
still without a sound physical basis.
Pei, Fall \& Hauser (1998) present yet a different approach. They
consider the evolution of the global stellar, gaseous, chemical and
radiation contents of the Universe. As inputs they use observations of
the extra-galactic background, interstellar gas density history from
damped Ly$\alpha$ surveys and the rest frame ultra-violet emissivity
history from optical galaxy surveys. They derive cosmic histories of
star formation, metallicity and radiation from stars and dust. Their
global approach cannot distinguish between the different types of star
formation processes in galaxies, such as in disks and starbursts, and
thus comparison to galaxy number count surveys is difficult.
This work approaches the problem in a semi-empirical manner, somewhere
in between the above extremes. The model makes use of reliable
empirical results where they exist. For example, the spectra and
luminosity functions of galaxies in the IR have been well determined
by the IRAS satellite (Saunders et al 1990; Malkan \& Stecker
1998). The star formation rate (SFR) history of the Milky Way's disk
is empirically derived from local observations, and then applied
globally to all disks, assuming our galaxy is typical of these
systems. We present a simple physical model relating SFR history to
the IR luminosity, including dust, gas and spectral evolution. Apart
from disks, observations of IR sources in the local universe, also
reveal many high-luminosity, interacting systems, which appear to be
undergoing an intense burst of star formation (Sanders \& Mirabel
1996). These systems have been termed starbursts. We develop a
physical model for the evolution of this population, assuming they
result from mergers and strong tidal interactions between gas rich
systems. The final theoretical input to the model is the evolution of
the number density and average mass of the disk systems. This is
obtained from the model of collision-induced galaxy formation
(Balland, Silk \& Schaeffer 1998, hereafter BSS98), which attributes
galactic morphology to the number of collisions and tidal interactions
suffered by a particular galaxy.
We evolve the present day populations of disks and starbursts
backwards in time, out to high redshift. This
approach results in relatively few free parameters. Without fine
tuning, we predict number counts in various wavebands and the IR and
sub-mm extra-galactic background, which agree remarkably well with
observations. We present the details of the model in \S 2, our
results in \S 3 and our conclusions in \S 4.
\section{The Model}
We model the evolution of two distinct populations which contribute
significant flux in the IR and sub-mm: disk galaxies and
starbursts. The latter we define to be systems undergoing a
violent merger or strong tidal interaction, which leads to high rates
of heavily obscured star formation and the creation of a spheroidal
stellar system or component. An Einstein-de Sitter cosmology
has been assumed with $H_{0}=\:50\:{\rm km\:s^{-1}\:Mpc^{-1}}$. Thus
$t_{0}=13.0$ Gyr. A summary of the model parameters, which are
described below, is shown in Table \ref{tab:param}.
\subsection{Disk galaxies}
We take the present day IR luminosity function of disk galaxies to be
well described by the IRAS far-IR (FIR) luminosity function (Saunders et al
1990) below a certain threshold luminosity, $L_{cut}$:
\begin{equation}
\label{PhiLspi}
\Phi^{disk}(L)=\cases {0 & $L<L_{min}$\cr
C^{disk}(z)(L/L_{*}^{disk}(z))^{1-\alpha}\exp\left[-\frac{1}{2\sigma^{2}}\log^{2}_{10}\left(1+\frac{L}{L_{*}^{disk}(z)}\right)\right] & $L_{min}<L<L_{cut}$\cr
0 & $L>L_{cut}$, \cr}
\end{equation}
where $C^{disk}_{0}=2.6\times 10^{-2}\:h^{3} \:{\rm Mpc^{-3}}$,
$L_{*}^{disk}=10^{8.77}\:h^{-2}\:L_{\odot}$ and $\alpha = 1.09$. We
take $L_{min}=10^{5}\:h^{-2}\:L_{\odot}$. Observationally, the
luminosity function is not well constrained at these low luminosities,
but the extra-galactic background and, for the flux range of interest,
the number counts are insensitive to the choice of $L_{min}$. We take
the upper limit to be $L_{cut}\sim5.6\times10^{10}\:h^{-2}\:L_{\odot}$
(Sanders \& Mirabel 1996). This defines the difference between
disk-like systems and starbursts, which are modeled in \S 2.2. The
effect of the precise choice of $L_{cut}$ is examined in \S 3. Note,
the above numerical values of $L_{min}$, $L_{*}^{disk}$ and $L_{cut}$
are defined for $z=0$.
Following Malkan \& Stecker (1998), we divide the IRAS luminosity
function into seven different spectral classes, ranging from $10^{-3}$
to $10^{3}\:L_{*}$. The spectra of these classes are derived
empirically from their IRAS colors. These spectra are then
analytically extended beyond 400 ${\rm \mu m}$ into the sub-mm regime,
assuming a dust emissivity dependence of $\nu^{\beta}$, with
$\beta=1.5$ (Franceschini, Andreani, \& Danese 1998; Roche \& Chandler
1993). Results for $\lambda\lesssim1000\:{\rm \mu m}$ are quite
insensitive to the precise choice of $\beta$, since most of the flux
detected in this part of the spectrum is emitted from sources at
$z\gtrsim1$, from their rest frame infra-red. At shorter wavelengths,
the spectra average over the line and bump features seen around
$10\:{\rm \mu m}$. This may be important for some of the details of
the $15\:{\rm \mu m}$ ISO CAM source counts. As the luminosities of
the galaxies evolve, the relative proportion of sources in each
spectral class changes, and in this way we take account of spectral
evolution.
Disk galaxies are observed to be undergoing relatively steady star
formation over many Gyr. Their IR luminosity is due to dust heating
from both young stars, which spend most of their lives in dusty
star-forming regions, and older stars, which contribute to the general
interstellar radiation field (ISRF). The present-day fraction of the
total from young star heating is $f_{0}^{young}\sim0.65$ (Mayya \& Rengarajan
1997; Devereux et al 1994; Xu \& Helou 1996; Walterbos \& Greenawalt
1996). We consider the luminosity evolution due to these two heating
sources separately.
The IR luminosity of disks due to young stars, $L^{ir,young}$, is
expected to evolve with redshift in the following manner:
\begin{equation}
\label{Lirsfr}
\frac{L^{ir,young}(z)}{L^{ir,young}_{0}}=\frac{\phi(z)}{\phi_{0}}\frac{(1-e^{-\tau^{uv}(z)})}{(1-e^{-\tau^{uv}_{0}})}\frac{M^{gal}(z)}{M^{gal}_{0}},
\end{equation}
where $\phi(z)$ is the local star formation rate (SFR) per unit disk
mass of disk galaxies, $\tau^{uv}(z)$ is the disk optical depth to
young stellar energy release and $M^{gal}(z)$ is the mean baryonic
mass of disk galaxies. These are all redshift dependent.
We derive $\phi$ from the metallicity distributions, $dN/dZ$, of local
G-type stars (Rocha-Pinto \& Maciel 1996; Wyse \& Gilmore 1995),
(Figure \ref{fig:nummetals}), and the age-metallicity relationship,
$dZ/dt$, of local F-type stars (Edvardsson et al 1993), (Figure
\ref{fig:diskmodel}a), using
\begin{equation}
\label{philocal}
\phi(t)\propto\frac{dN}{dZ}\frac{dZ}{dt}.
\end{equation}
$\phi(t)$ is further constrained by having to produce a local
present-day stellar disk surface density of $\sim 40\:{\rm
M_{\odot}\:pc^{-2}}$ (Sackett 1997), assuming the returned mass
fraction is small. From a sample of nearby white dwarfs and
consideration of their cooling curves, Oswald et al (1996) find the
local disk of the Milky Way has an age of $\sim10\pm1$ Gyr. Using a
similar method, Knox, Hawkins \& Hambly (1999) derive an age of
$10_{-1}^{+3}$ Gyr. However, because we wish to describe the galaxy
averaged evolution of disks, we take the mean age to be $\sim12$
Gyr, since it is expected the inner disk will start forming at earlier
times. For simplicity we use this formation time to complete the
normalization of $\phi(t)$, which we now take to approximate the
evolution of the total Galactic SFR history. An age of 12 Gyr
corresponds to a disk formation time, $t_{f}\approx1$ Gyr, and
redshift, $z_{f}\approx5.5$. At these high redshifts, the model
predictions of the source counts and backgrounds are quite insensitive
to the precise choice of $z_{f}$. The major uncertainty in this
procedure is in fitting the age-metallicity relation, because of the
large scatter in the data. We fit the data for $dN/dZ$, which show
less scatter, with a Gaussian. Only the fit at solar metallicities and
below affects the derived SFR history. $dZ/dt$ is fit with a function
which rises asymptotically from low metallicities at $t\sim t_{f}$,
and then levels off to solar metallicity at $t\sim t_{0}$, with the
additional constraint that the integrated SFR history (Figure
\ref{fig:diskmodel}b) matches the observed stellar disk surface
density. The SFR rises rapidly at early times, peaks at a level
roughly ten times that of today, and then decays exponentially.
Assuming the Milky Way is typical of disk galaxies, we apply $\phi$ to
the entire population of disks. This method enables us to empirically
probe the evolution of disk systems separately from the other
components of the Universe.
With a redshift independent dust-to-metals mass ratio (Pei et al 1998,
their Figure 1), $\tau$ is proportional to the metallicity, $Z$, and
the local gas density in the disk, $\rho^{gas}$. Thus,
\begin{equation}
\label{tau}
\frac{\tau(z)}{\tau_{0}}=\frac{Z(z)}{Z_{0}}\frac{\rho^{gas}(z)}{\rho_{0}^{gas}}=\frac{Z(z)}{Z_{0}}\left(\frac{\phi(z)}{\phi_{0}}\right)^{2/3},
\end{equation}
where we have used the Schmidt law, $\phi\propto(\rho^{gas})^{n}$,
with $n=1.5$, to relate gas density to the rate of star
formation. This relation is observed to hold over a large range of gas
densities relevant to both disks and starbursts (Kennicutt 1998). The
gas density history is shown in Figure \ref{fig:diskmodel}c. We know
the metallicity history directly from the age-metallicity relationship
(Figure \ref{fig:diskmodel}a). The optical depth history is shown in
Figure \ref{fig:diskmodel}d.
$M^{gal}(z)$ is derived from the model of collision-induced galaxy
formation of BSS98 and is shown in Figure \ref{fig:diskmodel}e. The
mean baryonic mass in a galaxy is controlled by the competition
between merging and cooling. It results from consideration of the
Press-Schechter mass function, together with an applied cooling
constraint, which gives an upper limit to galaxy masses. There is
little evolution in the mean mass out to $z\sim2$. By $z\sim 5$ it has
fallen by a factor of a few.
The fraction of the baryonic mass bound to a galaxy that is in the
form of gas, $f^{gas}$, can be calculated, given $\phi$ and
$f^{gas}_{0}$. This gas is not necessarily in the disk of the galaxy,
but may be in the process of falling in from the halo. It is
available to form stars in the event of a merger or close encounter
with another galaxy, as tidal forces will channel the gas (or some
portion of it) into the central starbursting regions. Thus knowledge
of the evolution of $f^{gas}$ is necessary for modeling of the
starbursts in \S 2.2. From observations of local disk galaxies, we
take $f^{gas}_{0}\sim 0.1$ (Young \& Scoville 1991; Young et al
1995). The effect of varying this choice is investigated in \S3. Its
evolution is shown in Figure \ref{fig:diskmodel}f.
We describe the IR luminosity due to dust heating by the older stellar
population as follows:
\begin{equation}
\label{Lirold}
\frac{L^{ir,old}(z)}{L^{ir,old}_{0}}=\frac{\int_{t_{f}}^{t_{z}}\phi(t)\:dt}{\int_{t_{f}}^{t_{0}}\phi(t)\:dt}\frac{(1-e^{-\tau^{opt}(z)})}{(1-e^{-\tau_{0}^{opt}})}\frac{M^{disk}(z)}{M^{disk}_{0}}.
\end{equation}
For simplicity it has been assumed that the intensity of the ISRF of
disks in the optical scales as the integrated SFR history. The
combination of $L^{ir,young}$ and $L^{ir,old}$ is a simple
representation of the true situation in which there is a continuously
varying contribution to dust heating over the whole mass spectrum and
environments of stars present in a galaxy.
From Wolfire et al (1999) we set $\tau^{uv}_{0}\sim 0.7$. We assume
$\tau^{opt}_{0}\sim 0.1$, placing present-day disks in the optically
thin limit, with respect to the general ISRF. The calculation of the
IR and sub-mm number counts and background is relatively insensitive
to the choices of $\tau^{uv}_{0}$ and $\tau^{opt}_{0}$, provided we
are in the optically semi-thick and thin regimes respectively, since,
in this model, it is the scaling of the luminosity function, with
respect to today's, which is important.
We implement luminosity evolution of the present-day characteristic
luminosity, $L^{disk}_{*}$, using equations (\ref{Lirsfr}) and
(\ref{Lirold}), each contributing the appropriate fraction,
$f_{0}^{young}$ and $1-f_{0}^{young}$ respectively to the total. This
is shown in Figure \ref{fig:diskevol}a.
We obtain number density evolution (Figure \ref{fig:diskevol}b) from
the models of BSS98 for field (i.e. non-cluster) galaxies, and apply
this scaling to the normalization constant, $C^{disk}(z)$, of the
luminosity function. In their model disk galaxies form from clouds
which experience few, if any, collisions between the formation time
and the epoch under consideration. Strong collisions will tend to
prevent the gas from settling into a disk, allow for tidal exchanges
which average out angular momentum, and lead to the formation of
ellipticals.
\subsection{Starbursts}
We model starburst galaxies as gas rich systems which are undergoing a
merger or a strong tidal interaction. We assume the starburst number
density scales with the rate of such interactions, while the mean
luminosity scales as the average galactic mass, $M^{gal}$, and the gas
fraction of baryonic matter, $f^{gas}$, available for star
formation. We treat the star formation as being completely obscured by
dust. Observations support this approximation (Sanders \& Mirabel
1996), and it is imagined that even for metal free systems, the
initial burst-induced star formation acts very quickly to pollute the
ISM and so obscure the vast majority of the stellar energy release.
We model the local ($z\lesssim0.2$) starburst population with the high
luminosity end of the IRAS FIR luminosity function:
\begin{equation}
\label{PhiSB}
\Phi^{sb}(L)=\cases {0 & $L<L_{cut}$\cr
C^{sb}(z)(L/L_{*,0}^{sb}(z))^{1-\alpha}\exp\left[-\frac{1}{2\sigma^{2}}\log^{2}_{10}\left(1+\frac{L}{L_{*,0}^{sb}(z)}\right)\right] & $L_{cut}<L<L_{max}$\cr
0 & $L>L_{max}$,\cr}
\end{equation}
where $C^{sb}_{0}=2.6\times 10^{-2}\:h^{3} \:{\rm Mpc^{-3}}$,
$L_{*}^{sb}=10^{8.77}\:h^{-2}\:L_{\odot}$, $\alpha = 1.09$ (Saunders
et al 1990) and $L_{cut}\sim5.6\times10^{10}\:h^{-2}\:L_{\odot}$
(Sanders \& Mirabel 1996). We take
$L_{max}=10^{15}\:h^{-2}\:L_{\odot}$. The results are insensitive to
this choice because of the steepness of the luminosity function,
equivalent to $L^{-2.35}$, (Kim \& Saunders 1998). Note, $L_{*}^{sb}$
is not a characteristic starburst luminosity. It is simply used to
parameterize the luminosity function.
We use the same spectral luminosity classes as for the disk
model. Starbursts, with higher FIR luminosities than disks, have a
stronger component of warm dust emission.
$L_{*}^{sb}$ is modeled to evolve as follows:
\begin{equation}
\label{LstarSB}
\frac{L_{*}^{sb}(z)}{L_{*,0}^{sb}}=\frac{M^{gal}(z)}{M_{0}^{gal}}\frac{f^{gas}(z)}{f_{0}^{gas}}.
\end{equation}
Since all the energy released by star formation is reradiated in the
far-IR, $L_{*}^{sb}$ is proportional to the average galactic mass and
gas fraction of the merging systems.
To model the number density evolution, we assume $C^{sb}$ is
proportional to the rate of collisions between gas rich systems,
$\Gamma^{coll}$. The collision rate of any given galaxy is
\begin{equation}
\label{gamma1}
\Gamma^{coll}_{1}=\frac{v}{\lambda_{mfp}}\propto n^{gal}R^{2}v\propto n^{gal}_{c}(1+z)^{3}M^{gal}\frac{v_{0}}{(1+z)^{1/2}},
\end{equation}
where $v$ is the mean peculiar velocity of galaxies, $\lambda_{mfp}$
is the mean free path before a collision or strong interaction occurs,
$R$ is the mean galactic linear size, $n^{gal}$ is the true number density of
(gas rich) galaxies and $n^{gal}_{c}$ is the corresponding comoving
number density. We assume that galaxies are disk-like so that
$M^{gal}=k R^{2}$ for some constant $k$, which we take to be redshift
independent. The redshift dependence of $v=v_{0}(1+z)^{-1/2}$ results
from consideration of structure formation in the linear regime. The
total collision rate is thus:
\begin{equation}
\label{gammaSB}
\frac{\Gamma^{coll}}{\Gamma^{coll}_{0}}=\frac{C^{sb}}{C^{sb}_{0}}=(1+z)^{5/2}\left(\frac{n_{c}^{gal}}{n_{c,0}^{gal}}\right)^{2}\frac{M^{gal}}{M^{gal}_{0}}.
\end{equation}
We obtain $n_{c}^{gal}(z)$ from BSS98, by summing their disk and
irregular morphological types. These dominate the galaxy population at
any particular epoch. The luminosity and number density evolutions of
the starburst population are shown in Figure \ref{fig:SBevol}. The luminosity evolution depends on the value of $f_{0}^{gas}$ and the effect of its variation is shown.
\section{Results}
For our IR and sub-mm sources we predict number counts, redshift
distributions, the intensity of the extra-galactic background and the
global SFR history. The values of our model's parameters are based
directly on observations (Table \ref{tab:param}) and so we do not
attempt to fine tune them to obtain ``perfect'' fits to all the
observations. Rather, we demonstrate that this simple model of galaxy
evolution is consistent with all the available data, and examine the
effects of varying our two most sensitive model parameters, $L_{cut}$
and $f_{0}^{gas}$.
The disk, starburst and total integrated number counts are shown in
Figure \ref{fig:numcounts}. The model agrees well with observations
made in the IRAS $12 \:{\rm \mu m}$ (Rush et al 1993), ISO CAM $15
\:{\rm \mu m}$ (Aussel et al 1998; Altieri et al 1998), IRAS $60
\:{\rm \mu m}$ (Lonsdale et al 1990), ISO PHOT $175 \:{\rm \mu m}$
(Puget et al 1998) and SCUBA $850 \:{\rm \mu m}$ (Blain et al 1999;
Smail et al 1998 and references therein) passbands. At 15 ${\rm \mu
m}$ disks dominate the counts at high fluxes, while the starburst
contribution becomes comparable between $10^{-3}$ to $10^{-4}$ Jy. At
lower fluxes, the starburst counts flatten off, as we are probing the
limit of the distribution. At 60 ${\rm \mu m}$ the disks again
dominate at the high flux end, with starbursts becoming more important
at around $10^{-2}$ Jy. At 175 ${\rm \mu m}$ the starburst counts rise
steeply to dominate the total between $\sim 1$ and $10^{-2}$ Jy, which
is the region probed by current ISO PHOT observations. At 850 ${\rm
\mu m}$ starbursts completely dominate at all fluxes above $10^{-3}$
Jy.
Figure \ref{fig:numcountszoom} shows the total counts in expanded
regions of the flux-number count diagrams relevant to the latest
observations. The ISO 15 ${\rm \mu m}$, ISO 175 ${\rm \mu m}$ and
SCUBA 850 ${\rm \mu m}$ plots also show the effect of varying
$f_{0}^{gas}$ and $L_{cut}$ from the fiducial values. Varying
$f_{0}^{gas}$ affects the starburst population while $L_{cut}$
dictates the relative contribution of disks to starbursts. The IRAS 60
${\rm \mu m}$ data, sampling the low redshift population, are very
insensitive to these parameters.
Figure \ref{fig:zdist} shows the predicted redshift distributions of
the source populations observed by ISO CAM, IRAS, ISO PHOT and
SCUBA. The discontinuities in the distributions are artifacts
resulting from our simplistic method of dividing the luminosity
function into disk and starburst sources and discrete spectral
classes. The observed disks are at lower redshifts relative to the
bulk of the observed starbursts. In the case of the SCUBA sources
detected with $S>0.63$ mJy, about $2/3$ are predicted to be
starbursting systems, and the rest normal disks. Most of these
sources are predicted to be at redshifts greater than one, with many
seen out to redshift five and beyond. This is due to the steep slope
of the Rayleigh-Jeans portion of the modified blackbody spectra of the
sources. As more redshift determinations are made of these source
samples, it will be interesting to compare the inferred distributions,
corrected for clustering effects and incompleteness, to the model.
The predictions of the extra-galactic background in the IR to sub-mm
are shown in Figure \ref{fig:ebl}. Recent independent estimates of the
flux from the FIRAS residuals between 150 - 5000 ${\rm \mu m}$ (Puget
et al 1996; Fixsen et al 1998) agree on the spectrum and amplitude of
the background, although the sizes of the systematic errors are not
well determined. Strong upper limit constraints are also being
reported at shorter wavelengths from observations of TeV $\gamma$-rays
(Stanev \& Franceschini 1998; Biller et at 1998), while lower limits
can be placed from the summed flux predicted by number count
surveys. Our fiducial model is consistent with the existing data,
although at longer wavelengths it predicts a flux about twice as high
as the mean of the FIRAS amplitudes. This discrepancy is within the
bounds of model and observational uncertainties, as illustrated in the
lower panel of Figure \ref{fig:ebl}.
Finally we derive the global SFR history (Figure \ref{fig:globalSFR})
by assuming a conversion factor of $2\times10^{-10}\:{\rm
M_{\odot}\:yr^{-1}\:L^{FIR}_{\odot}}$. This is in agreement with
recent calibrations ($1 - 3\times10^{-10}\:{\rm
M_{\odot}\:yr^{-1}\:L^{FIR}_{\odot}}$) from starburst synthesis models
(Leitherer \& Heckman 1995; Lehnert \& Heckman 1996; Meurer et al
1997). This conversion factor is applied to the far-IR luminosity due
to dust heating by young stars in disks, accounting for the evolving
optical depth, as well as to the optically thick starburst population.
The results for the disk contribution are thus relatively sensitive to
our choice of $\tau_{0}^{uv}\sim0.7$. However, with this fiducial
value, the model predictions agree well with the low redshift
observations, given the uncertainties in the SFR to far-IR luminosity
conversion factor. At high redshifts the model is consistent with
recent results which suggest a relatively flat SFR history out to
$z\sim4$ (Steidel et al 1998).
If we assume that our starbursts lead to the formation of spheroidal
stellar systems, such as ellipticals and the bulges and halos of disk
galaxies, then we can predict the mass densities of the disk and
spheroidal components that exist in the present-day
Universe. Neglecting mass returned to the inter-stellar and
inter-galactic media, we find $\Omega_{disk}\approx5.1\times10^{-3}$
and $\Omega_{sph}\approx4.4\times10^{-3}$. These estimates are
sensitive to $L_{cut}$ and are uncertain by factors of a
few. Observationally, the spheroidal component is dominated by the
bulges and halos of disk galaxies, due to the paucity of ellipticals
in the field (Binggeli, Sandage \& Tammann 1988). Our results suggest
the mass in spheroids is comparable to the mass in the disks,
consistent with the analysis of Schechter \& Dressler (1987).
Our model indicates that there is a broad peak of spheroid formation
at $z\sim3$ (Figure \ref{fig:globalSFR}). Recently, evolved
ellipticals have been seen in deep NICMOS images (Benitez et al 1998),
requiring a very high ($z\gtrsim5$) redshift of formation.
Extrapolating the global SFR predictions of our model to these high
redshifts, whilst uncertain, indicates that the SFR's are high enough
to account for some very early elliptical formation. Note, however,
within the framework of our model, the contribution to the
extra-galactic background and the number counts from these early times
($t<1$ Gyr) is negligible. In other words, almost all the star
formation and associated energy release necessary to account for the
observed extra-galactic background occurs at redshifts $\lesssim 5$
and is accessible to current observations.
\section{Conclusions}
We have presented a simple model for the evolution of disk galaxies
and starbursts, tied as closely as possible to observations. Despite
its simplicity, the model takes account of dust, gas and spectral
evolution in a self-consistent manner, and is able to predict source
counts and the extra-galactic background in the IR to sub-mm
consistent with observations, without recourse to fine tuning of
parameters. A disk-only model, with $L_{cut}=L_{max}$, gives only
marginally greater counts and fluxes than the disk component of the
fiducial model, thus failing to account for observations. This
demonstrates the significant role of the starburst population. The
fiducial model we present is based on the best estimates and
observations of our various model parameters (Table
\ref{tab:param}). We have not attempted to vary these to obtain the
best fit to the, often uncertain, high redshift IR and sub-mm data.
The predicted global SFR history agrees with recent observations,
corrected for dust extinction, indicating that SFR's remain high from
$z\sim1$ back to $z\gtrsim4$. At high redshifts the SFR is dominated
by starbursts and is driven by galaxy-galaxy interactions. Over the
history of the Universe, the total star formation occurring in
starbursts is comparable to that in disks. This suggests the
baryonic mass in bulges, formed from starbursts, is similar to that in
disks. The bulges form at high redshift, and are then thought to act
as the seeds for disk formation as gas infalls.
The energy release associated with the entire star formation history
to $z\sim5$ is enough to account for all of the observed
extra-galactic background. This implies that the vast majority of star
formation in the Universe occurs over this period.
The majority ($\sim 2/3$) of the sources recently detected with SCUBA
are identified as starbursts. Typical redshifts are $1\lesssim z
\lesssim 5$. The direct effect of these sources, along with other
foregrounds, on the angular power spectrum of the microwave background
has been examined by Gawiser et al (1999). The effect of
cluster-induced lensing of the sources on the background has been
investigated by Scannapieco, Silk \& Tan (1999) and found to be small.
Future possible improvements include a more sophisticated method of
distinguishing disks and starbursts in the infra-red luminosity
function, a more detailed treatment of the opacity of starbursts and
the inclusion of active galactic nuclei. Extending the model to the
ultra-violet, optical and radio, will allow additional observations to
help constrain the galaxy evolution. Consideration of alternative
cosmologies is left to a future date. The model will be considerably
refined once results are available from future observations with
FIRST, SIRTF and the NGST. Model results are available electronically
at http://astro.berkeley.edu/~jt/irmodel.html.
\acknowledgements We are grateful to Matt Malkan for kindly providing
his data for the IR galaxy spectra in electronic form, and to Herve
Aussel for his ISO CAM 15 ${\rm \mu m}$ number count data. We thank
R. Bouwens, J. Puget, E. Scannapieco, A. Cumming and E. Gawiser for
helpful discussions.
|
\section{Introduction}
Let ${\mathfrak{A}}$ be a simple separable C*-algebra, and let $\pi_1, \pi_2$ be
representations of ${\mathfrak{A}}$ on Hilbert spaces ${\mathcal H}_1,{\mathcal H}_2$. The
representations
$\pi_1,\pi_2$ are said to be algebraically equivalent if $\pi_1({\mathfrak{A}})''$ and
$\pi_2({\mathfrak{A}})''$ are isomorphic von Neumann algebras. If there is an
automorphism $\alpha$ of ${\mathfrak{A}}$ such that $\pi_1$ and $\pi_2\circ \alpha$ are
quasi-equivalent, then $\pi_1,\pi_2$ are clearly algebraically equivalent.
Powers proved in \cite{Pow67} that if ${\mathfrak{A}}$ is a UHF algebra the converse is
true. His method extends readily to the case that ${\mathfrak{A}}$ is an AF-algebra,
\cite{Bra72}.
See also section~12.3 in \cite{KR86}. In the special case that $\pi_1$ (and
therefore $\pi_2)$ is irreducible, Kadison's transitivity theorem
therefore implies that if ${\mathfrak{A}}$ is a simple AF algebra and if
$\omega_1$ and $\omega_2$ are pure states on ${\mathfrak{A}}$, there exists an
automorphism $\alpha$ of ${\mathfrak{A}}$ such that $\omega_1=\omega_2\circ
\alpha$. To our knowledge, this question has only been settled in the
affirmative when
${\mathfrak{A}}$ is an AF-algebra. As a beginning of a possible resolution of the question
for purely infinite algebras, we here prove the statements in the abstract.
Recall from \cite{Cun77} that the Cuntz algebra
${\mathcal O}_d$ is the C*-algebra generated by $d$ operators $s_1,\ldots,s_d$
satisfying
\begin{eqnarray*}
&&s_j^\ast s_i=\delta_{ij}{\openone} \\
&&\sum_{i=1}^d s_is_i^\ast={\openone}
\end{eqnarray*}
There is an action $\gamma$ of the group $U(d)$ of unitary $d\times d$
matrices on
${\mathcal O}_d$ given by
$$
\gamma_g(s_i)=\sum_{j=1}^d g_{ji}s_j
$$
for $g=[g_{ij}]_{i,j=1}^d$ in $U(d)$. In particular the {\em gauge action\/}
$\tau=\gamma|_{{\bf T}}$ is defined by
$$
\tau_z(s_i)=zs_i\;,\qquad z\in{\bf T}\subset{\bf C}\;.
$$
If ${\rm UHF}_d$ is the fixed point subalgebra under the gauge action, then
${\rm UHF}_d$ is the closure of the linear span of all Wick ordered polynomials of
the form
$$
s_{i_1}\ldots s_{i_k} s_{j_k}^\ast\ldots s_{j_1}^\ast
$$
${\rm UHF}_d$ is isomorphic to the UHF algebra of Glimm type $d^\infty$:
$$
{\rm UHF}_d\cong M_{d^\infty}=\bigotimes_1^\infty M_d
$$
in such a way that the isomorphism carries the Wick ordered polynomial above
into the matrix element
$$
e_{i_1 j_1}^{(1)}\otimes e_{i_2 j_2}^{(2)}\otimes\cdots\otimes
e_{i_k j_k}^{(k)}\otimes{\openone}\otimes{\openone}\otimes\cdots\;.
$$
In the case that $d$ is a power of a prime, the gauge action $\tau$ is
in fact characterized by the fact that its fixed
point algebra is isomorphic to ${\rm UHF}_d$, i.e. if $\alpha$ is another faithful
action of ${\bf T}$ on ${\mathcal O}_d$ such that the fixed point algebra
${\mathcal O}_d^\alpha$ is isomorphic to ${\rm UHF}_d$, then either
$z\mapsto\alpha_z$ or $z\mapsto\alpha_z^{-1}$ is conjugate to $\tau$. This
follows from \cite[Corollary~4.1]{BK99}. (Since ${\rm UHF}_d$ is simple and
$\alpha$ is faithful, the crossed product ${\mathcal O}_d\times_\alpha {\bf T}$ is
stably isomorphic to ${\rm UHF}_d$, \cite{KT78}, and in particular it is simple.
Since
$$
{\mathcal O}_d^\alpha\cong P_\alpha(0)({\mathcal O}_d
\times_\alpha{\bf T})P_\alpha(0)\;,
$$
\noindent
$[P_\alpha(0)]$ is just $[{\openone}]$ when $K_0({\mathcal O}_d\times_\alpha{\bf T})$ is
identified with $K_0({\mathcal O}_d^\alpha)$. By the Pimsner-Voiculescu exact
sequence it follows that
$\widehat{\alpha}_\ast$ on
$K_0({\mathcal O}_d\times_\alpha{\bf T})={\bf Z}[\frac{1}{d}]$ is multiplication by $d$
or $1/d$. For this last argument it is important that $d$ is a power of a
prime, as seen from the example $d=6$ and $\widehat{\alpha}_\ast$
equal to multiplication by 4/9 on ${\bf Z}[\frac{1}{6}]$) Because of this, our main result Theorem~5 can be given the
following more universal form:
\begin{corollary}
Assume that $d$ is a power of a prime. Let $\varphi_1$ and
$\varphi_2$ be pure states on ${\mathcal O}_d$, and
assume that there exist actions $\alpha_i$ of ${\bf T}$ on ${\mathcal O}_d$
such that
${\mathcal O}_d^{\alpha_i}\cong{\rm UHF}_d$ and
$\varphi_i\circ\alpha_i=\varphi_i$ for $i=1,2$. Then there exists an
automorphism $\beta$ of ${\mathcal O}_d$ such that
$$
\varphi_1=\varphi_2\circ\beta
$$
\end{corollary}
The question whether any pure state on ${\mathcal O}_d$ is invariant under a
gauge action is left open.
The restriction of $\gamma_g$ to ${\rm UHF}_d$
is carried into the action
$$
\mathop{\rm Ad}\nolimits(g)\otimes\mathop{\rm Ad}\nolimits(g)\otimes\cdots
$$
on $\bigotimes\limits_1^\infty M_d$. We define the canonical endomorphism
$\lambda$ on ${\rm UHF}_d$ (or on ${\mathcal O}_d$) by
$$
\lambda(x)=\sum_{j=1}^d s_j x s_j^\ast
$$
and the isomorphism carries $\lambda$ over into the one-sided shift
$$
x_1\otimes x_2\otimes x_3\otimes \cdots \to {\openone}\otimes x_1\otimes
x_2\otimes\cdots
$$
on $\bigotimes\limits_1^\infty M_d$.
If $\eta_1,\ldots,\eta_d$ are complex scalars with
$\sum\limits_{j=1}^d|\eta_j|^2=1$, we can define a state on ${\mathcal O}_d$
by
$$
\varphi_\eta(s_{i_1}\ldots s_{i_k}\ s_{j_\ell}^\ast\ldots s_{j_1}^\ast)
=\eta_{i_1}\ldots \eta_{i_k}\ \overline{\eta_{j_\ell}}\ldots
\overline{\eta_{j_1}}
$$
\cite{Cun77}, \cite{Eva80}, \cite{BJP96}, \cite{BJ97},
\cite{BJKW}.
This state is pure, and non-gauge invariant, and the $U(d)$ action is
transitive on these states, which are called Cuntz states. The restriction of
$\varphi_\eta$ to ${\rm UHF}_d$ identifies with the pure product state given by
infinitely many copies of the vector state defined by the vector
$(\eta_1,\ldots,\eta_d)$ on
$M_d$.
In this paper we will also consider the one-one correspondence between the
set
${\mathcal U}({\mathcal O}_d)$ of unitaries in ${\mathcal O}_d$ and the set
${\rm End}({\mathcal O}_d)$ of unital endomorphisms of ${\mathcal O}_d$. If
$u\in{\mathcal U}({\mathcal O}_d)$ then $\alpha_u(s_i)=us_i$ defines an
endomorphism, and if $\alpha\in{\rm End}({\mathcal O}_d)$ the corresponding
unitary is $u=\sum\limits_{i=1}^d \alpha(s_i)s_i^\ast$. It has been proved by
R\o{}rdam that
$$
{\mathcal U}_i=\{u\in{\mathcal U}({\mathcal O}_d)|\alpha_u\
\mbox{ is an inner automorphism}\}
$$
is a dense subset of ${\mathcal U}({\mathcal O}_d)$,
[R\o{}r93].
We give a
shorter proof of this, and also show that
$$
{\mathcal U}_a=\{u\in{\mathcal U}({\mathcal O}_d)|\alpha_u\
\mbox{ is an automorphism}\}
$$
is a dense $G_\delta$ subset of ${\mathcal U}({\mathcal O}_d)$ such that the
complement ${\mathcal U}({\mathcal O}_d)\setminus {\mathcal U}_a$ is also
dense.
By using the above correspondence between ${\mathcal U}({\mathcal O}_d)$ and
$\mathop{\rm End}\nolimits({\mathcal O}_d)$, it follows (see the proof of Proposition~8) that if
$\omega$ is a pure state and $\varphi_0$ a Cuntz state there exists an
endomorphism $\alpha$ of ${\mathcal O}_d$ such that
$\varphi_0=\omega\circ\alpha$. Although the automorphism group is dense in
$\mathop{\rm End}\nolimits({\mathcal O}_d)$ (in the topology of pointwise convergence), the
question whether $\alpha$ can be chosen to be an automorphism is left open (in
this approach).
\section{Transitivity of the automorphism group on the pure gauge-invariant
states}
\setcounter{equation}{0}
In this section we prove the first main result mentioned in the abstract.
Let ${\rm UHF}_d$ be the UHF algebra of type $d^\infty$ and let $(A_n)$ be an
increasing sequence of C*-subalgebras of ${\rm UHF}_d$ such that
${\rm UHF}_d=\overline{\cup A_n}$ and $A_n\cong M_{d^n}$. We first use Power's
transitivity on ${\rm UHF}_d$ to find an approximate factorization for any pure
state on ${\rm UHF}_d$:
\begin{lemma}
Let $\varphi$ be a pure state of ${\rm UHF}_d$ and $\varepsilon>0$. Then there
exists a pure state $\varphi'$ of ${\rm UHF}_d$, an increasing sequence $\{B_n\}$
of finite type I subfactors of ${\rm UHF}_d$, and an increasing subsequence
$\{k_n\}$ in
${\bf N}$ such that $\varphi'|B_n$ is a pure state of $B_n$ and $A_{k_n}\subset
B_n\subset A_{k_{n+1}}$ for every $n$, and
$$
\|\varphi-\varphi'\|<\varepsilon\;.
$$
\end{lemma}
\begin{proof}
Since the automorphism group $\mathop{\rm Aut}\nolimits({\rm UHF}_d)$ of ${\rm UHF}_d$ acts transitively
on the set of pure states of ${\rm UHF}_d$, \cite{Pow67}, there exists an
increasing sequence
$\{D_n\}$ of finite type I subfactors of ${\rm UHF}_d$ such that $D_n\cong
M_{d^n}$ and
$\varphi|D_n$ is pure for every $n$. Then we can find sequences $\{u_n\}$ and
$\{v_n\}$ of unitaries in ${\rm UHF}_d$ and increasing sequences $\{k_n\}$ and
$\{\ell_n\}$ in ${\bf N}$ such that
\begin{eqnarray*}
&&A_{k_1}\subset\mathop{\rm Ad}\nolimits(v_1u_1)(D_{\ell_1})\subset A_{k_2}\subset
\mathop{\rm Ad}\nolimits(v_2u_2v_1u_1)(D_{\ell_2})\subset A_{k_3}\subset \cdots \\
&&u_n\in {\rm UHF}_d\cap\mathop{\rm Ad}\nolimits(v_{n-1}u_{n-1}\ldots v_1u_1)(D_{\ell_{n-1}})' \\
&&v_n\in {\rm UHF}_d\cap A_{k_n}' \\
&&\|u_n-1\|< \varepsilon/2^{n+2}\qquad
\|v_n-1\|<\varepsilon/2^{n+2}
\end{eqnarray*}
where $D_0=\C1$. (Let $k_1=1$. Then we choose $u_1$ and $\ell_1$ such that
$A_{k_1}\subset\mathop{\rm Ad}\nolimits u_1(D_{\ell_1})$ and $\|u_1-1\|<\varepsilon/8$. Further we
choose $k_2$ and $v_1$ such that $v_1\in {\rm UHF}_d\cap A_{k_1}'$,
$\|v_1-1\|<\varepsilon/8$, and, $\mathop{\rm Ad}\nolimits(v_1u_1)(D_{\ell_1})\subset A_{k_2}$. We
just repeat this process.) Then the limit $w=\lim v_n u_n\ldots v_1u_1$ exists
and is a unitary such that $\|w-1\|<\varepsilon/2$ and
$$
A_{k_1}\subset\mathop{\rm Ad}\nolimits w(D_{\ell_1})\subset A_{k_2}\subset \mathop{\rm Ad}\nolimits
w(D_{\ell_2})\subset\cdots
$$
Let $\varphi'=\varphi\circ\mathop{\rm Ad}\nolimits w^\ast$. Then $\varphi'$ is a pure state with
$\|\varphi-\varphi'\|<\varepsilon$ and $\varphi'|\mathop{\rm Ad}\nolimits w(D_{\ell_n})$ is a pure
state for every $n$. Put $B_n=\mathop{\rm Ad}\nolimits w(D_{\ell_n})$.
\end{proof}
We next show that for any pair of pure states $\varphi_1,\varphi_2$ on
${\rm UHF}_d$, there is a tensor product decomposition of ${\rm UHF}_d$ such that
$\varphi_1,\varphi_2$ have approximate factorizations with respect to certain
sub-decompositions (necessarily different for $\varphi_1$ and $\varphi_2$):
\begin{lemma}
Let $\varphi_1$ and $\varphi_2$ be pure states of ${\rm UHF}_d$ and let
$\varepsilon>0$. Then there exist pure states $\varphi_1',\varphi_2'$, and
$\psi$ of ${\rm UHF}_d$, an increasing sequence $\{k_n\}$ in ${\bf N}$ and an increasing
sequence $\{B_n\}$ of finite type I subfactors of $A$ such that
\begin{eqnarray*}
&&\|\varphi_i-\varphi_i'\|<\varepsilon \\
&&\varphi_1'|B_{2n+1}\quad \mbox{is pure} \\
&&\varphi_2'|B_{2n}\quad \mbox{is pure} \\
&&\psi|B_{6k-1}\cap B_{6k-3}'=\varphi_1'|B_{6k-1}\cap B_{6k-3}' \\
&&\psi|B_{6k+2}\cap B_{6k}'=\varphi_2'|B_{6k+2}\cap B_{6k}' \\
&&\psi|B_{6k}\cap B_{6k-1}'\quad \mbox{is pure}, \\
&&\psi|B_{6k-3}\cap B_{6k-4}'\quad \mbox{is pure}, \\
&&k_{n+1}-k_n\to\infty \\
&&A_{k_1}\subset B_1\subset A_{k_2}\subset B_2\subset A_{k_3}\subset
B_3\subset\cdots
\end{eqnarray*}
\end{lemma}
\begin{proof}
It follows from the previous lemma that there exist pure states $\varphi_i'$,
increasing sequences $\{B_{in}\}$ of finite type I subfactors of $A$, and an
increasing sequence $\{k_n\}$ in ${\bf N}$ such that
\begin{eqnarray*}
&&\|\varphi_i-\varphi_i'\|<\varepsilon\;, \\
&&\varphi_i|B_{in}\quad \mbox{is pure for}\ i=1,2\;, \\
&&A_{k_1}\subset B_{i1}\subset A_{k_2}\subset B_{i2}\subset
A_{k_3}\subset\cdots
\end{eqnarray*}
By passing to subsequences of $\{k_n\}$ and $\{B_{in}\}$ and setting
$B_n=B_{1n}$ if $n$ is odd and $B_n=B_{2n}$ if $n$ is even, we may assume that
\begin{eqnarray*}
&&\varphi_1'|B_{2n+1}\quad \mbox{is pure} \\
&&\varphi_2'|B_{2n}\quad \mbox{is pure} \\
&&k_{n+1}-k_n\to\infty \\
&&A_{k_1}\subset B_1\subset A_{k_2}\subset B_2\subset A_{k_3}\subset\cdots \\
\end{eqnarray*}
Then $\varphi_1'$ has a tensor product decomposition into pure states on the
matrix subalgebras $B_{2n+1}\cap B_{2n-1}'$, and $\varphi_2'$ likewise on the
subalgebras $B_{2n}\cap B_{2n-2}'$. Thus we can define a pure state $\psi$ by
requiring that it decomposes under the tensor product decomposition
\begin{eqnarray*}
\lefteqn{\hspace*{-2em}\ldots\otimes(B_{6k-4}\cap B_{6k-6}')\otimes
(B_{6k-3}\cap B_{6k-4}')\otimes(B_{6k-1}\cap B_{6k-3}')} \\
&&\otimes(B_{6k}\cap B_{6k-1}')\otimes(B_{6k+2}\cap B_{6k}')\otimes\cdots
\end{eqnarray*}
into states given by:
\begin{eqnarray*}
&&\psi|B_{6k-1}\cap B_{6k-3}'=\varphi_1'|B_{6k-1}\cap B_{6k-3}'\;, \\
&&\psi|B_{6k+2}\cap B_{6k}'=\varphi_2'|B_{6k+2}\cap B_{6k}'\;, \\
&&\psi|B_{6k}\cap B_{6k-1}'\quad \mbox{is an arbitrary pure state,} \\
&&\psi|B_{6k-3}\cap B_{6k-4}'\quad \mbox{is an arbitrary pure state}.
\end{eqnarray*}
\end{proof}
Recall that $\tau$ is the gauge action of ${\bf T}$ on ${\mathcal O}_d$, i.e.,
$$
\tau_z(s_i)=zs_i\;,\qquad z\in{\bf T}\;.
$$
Let $\varepsilon$ be the conditional expectation of ${\mathcal O}_d$ onto
${\rm UHF}_d$ defined by
$$
\varepsilon(x)=\int_{\bf T} \tau_z(x)\frac{|dz|}{2\pi}\;,\qquad
x\in{\mathcal O}_d\;.
$$
Note that if $\varphi$ is a gauge-invariant state of ${\mathcal O}_d$, then
$$
\varphi=\varphi|_{{\rm UHF}_d}\circ\varepsilon\;.
$$
Recall that $\lambda$ is canonical endomorphism of ${\mathcal O}_d$:
$\lambda(x)=\sum\limits_{i=1}^d s_ixs_i^\ast$, $x\in{\mathcal O}_d$, and
that the restriction of $\lambda$ to ${\rm UHF}_d$ is the one-sided shift $\sigma$.
\begin{lemma}
If $\varphi$ is a gauge-invariant state on ${\mathcal O}_d$ then the
following conditions are equivalent:
{\rm (i)}\quad $\varphi$ is pure
{\rm (ii)} \
$\varphi|_{{\rm UHF}_d}$ is pure and
\qquad $\varphi|_{{\rm UHF}_d}\circ\sigma^n$ is disjoint from $\varphi$ for
$\;n=1,2,\ldots$
\end{lemma}
\begin{proof}
(i)$\Rightarrow$(ii). Since $\varphi$ is pure, and gauge-invariant, it
follows that $\varphi|_{{\rm UHF}_d}$ is pure. Let $p$ be the support projection
of $\varphi$ in ${\mathcal O}_d^{\ast\ast}$. Since $p$ is minimal, and
$\varphi$ is gauge-invariant, it follows that for any $a\in{\rm UHF}_d$ and
any multi-index $I=(i_1,i_2,\ldots,i_n)$ with $|I|=n\geq 1$,
$$
pas_I p=\varphi(as_I)p=0\;,
$$
where $s_I=s_{i_1}s_{i_2}\ldots s_{i_n}$. Thus we obtain that
$$
p({\rm UHF}_d)\lambda^n(p)=0\;,
$$
which implies that $\varphi|_{{\rm UHF}_d}\circ \sigma^n$ is disjoint from
$\varphi$.
\bigskip
(ii)$\Rightarrow$(i). Let $p$ be the support projection of
$\varphi|_{{\rm UHF}_d}$ in
${\rm UHF}_d^{\ast\ast}\subset{\mathcal O}_d^{\ast\ast}$. It suffices to show that
for any multi-indices $I, J$
$$
ps_Is_J^\ast p\in{\bf C} p
$$
since the linear span of $s_Is_J^\ast$ is dense in ${\mathcal O}_d$. If
$|I|\not=[J|$, we have that $ps_Is_J^\ast p=0$ by using the fact that
$\varphi|_{{\rm UHF}_d}\circ\sigma^n$ is disjoint from $\varphi$ for
$n=\Big||I|-|J|\Big|$. If $[I|=|J|$, we have that $ps_Is_J^\ast
p=\varphi(s_Is_J^\ast)p$ since $\varphi|_{{\rm UHF}_d}$ is pure.
\end{proof}
\begin{lemma}
Let $\varphi_1$ and $\varphi_2$ be gauge-invariant pure states of ${\mathcal
O}_d$ such that all
\newline
$\varphi_i|_{{\rm UHF}_d}\circ\sigma^n$, $i=1,2$,
$n=0,1,2,\ldots$ are mutually disjoint. Then there exists an automorphism
$\alpha$ of ${\mathcal O}_d$ such that
$\alpha\circ\tau_z=\tau_z\circ\alpha$, $z\in{\bf T}$ and
$\varphi_1=\varphi_2\circ\alpha$.
\end{lemma}
\begin{proof}
By Lemma~4, $\psi_1=\varphi_1|_{{\rm UHF}_d}$ and $\psi_2=\varphi_2|_{{\rm UHF}_d}$ are
pure states on ${\rm UHF}_d$. Applying Lemma~3 on $\psi_1,\psi_2$ in lieu of
$\varphi_1,\varphi_2$, with $\varepsilon=1$, we obtain pure states
$\psi_1',\psi_2'$ and $\psi$ of ${\rm UHF}_d$ with the properties given there.
Since $\psi_i$ is equivalent to $\psi_i'$,
$\varphi_i'=\psi_i'\circ\varepsilon$ is a pure state of ${\mathcal O}_d$ by
Lemma~4 and this state is equivalent to $\varphi_i=\psi_i\circ\varepsilon$.
By Kadison's transitivity theorem we have a unitary $u\in{\rm UHF}_d$ such that
$\psi_i'=\psi_i\circ\mathop{\rm Ad}\nolimits u$; it follows that $\varphi_i'=\varphi_i\circ\mathop{\rm Ad}\nolimits u$.
It is not automatical that $\psi$ satisfies the condition that all
$\psi\circ\sigma^n$, $n=0,1,2,\ldots$ are mutually disjoint and are disjoint
from $\psi_i'\circ\sigma^n$. But using the freedom in constructing
$\psi|_{B_{6k}\cap B_{6k-1}'}$ and $\psi|_{B_{6k-3}\cap B_{6k-4}'}$
successively, we can certainly impose this condition.
Thus we obtain three pure states $\psi_1',\psi_2',\psi$ of ${\rm UHF}_d$ such that
all $\psi_i'\circ\sigma^n$, $\psi\circ\sigma^n$ are mutually disjoint and
$\psi_i'$ and
$\psi$ are spotwise asymptotically equal as specified in Lemma~3. It now
suffices to prove the lemma for the pairs
$(\psi_1'\circ\varepsilon,\psi\circ\varepsilon)$ and
$(\psi_2'\circ\varepsilon, \psi\circ\varepsilon)$. Thus replacing
$\varphi_1,\varphi_2$ by one of these pairs, we may assume the lemma satisfy
the additional condition that there exists an increasing sequence $\{k_n\}$ in
${\bf N}$ and an increasing sequence $\{B_n\}$ of finite type I subfactors of
${\rm UHF}_d$ such that
\begin{eqnarray*}
&&A_{k_1}\subset B_1\subset A_{k_2}\subset B_2\subset A_{k_3}
\subset B_3\subset \\
&&\varphi_i|_{B_{3n+1}}\quad \mbox{is pure}\;, \\
&&\varphi_1|_{B_{3n+3}\cap B_{3n+1}'}=\varphi_2|_{B_{3n+3}\cap
B_{3n+1}'}\quad \mbox{is pure} \\
&&k_{3n+3}-k_{3n+2}\to\infty\;.
\end{eqnarray*}
We shall construct a sequence $\{v_n\}$ of unitaries in ${\rm UHF}_d$ such that
\newline
$\alpha=\lim\limits_{n\to\infty}\mathop{\rm Ad}\nolimits(v_n v_{n-1}\ldots v_1)$ defines an
automorphism of ${\mathcal O}_d$ with $\varphi_1=\varphi_2\circ\alpha$. To
ensure the existence of the limit we choose the unitaries such that they
mutually commute and $\sum\|\lambda(v_n)-v_n\|<\infty$. Since $\alpha$
commutes with the gauge action $\tau$, this will complete the proof.
We fix a large $N\in{\bf N}$. We choose $n_1$ so large that the support
projections $e_i^{(1)}=\mathop{\rm supp}\nolimits(\varphi_i|_{B_{3n_1+1}})$ are almost orthogonal
and
$k_{3n_1+3}-k_{3n_1+2}>2^{2(N+1)}$. Let $w_1$ be a partial isometry in
$B_{3n_1+1}$ with $w_1^\ast w_1=e_1^{(1)}$, $w_1w_1^\ast=e_2^{(1)}$. By the
polar decomposition of the approximate unitary
$$
w_1+(1-e_2^{(1)})w_1^\ast(1-e_1^{(1)})+(1-e_2^{(1)})(1-e_1^{(1)})\;,
$$
we obtain a unitary $v_1\in B_{3n_1+1}$ such that
$$
v_1 e_1^{(1)}=w_1
e_1^{(1)}=e_2^{(1)}w_1=e_2^{(1)}v_1\in B_{3n_1+1}
$$
and
$v_1(1-e_2^{(1)})(1-e_1^{(1)})\approx (1-e_2^{(1)})(1-e_1^{(1)})$.
We next choose $n_2>n_1$ so large that
$$
\sigma^n\circ\mathop{\rm supp}\nolimits(\varphi_i|_{B_{3n_2+1}\cap B_{3n_1+3}'}),\quad
i\!=\!1,2,\;\,
n\!=\!-2^{N-1},-2^{-N+1}+1,\ldots,0,\ldots,2^{N+1}
$$
are almost orthogonal and
$k_{3n_2+2}-k_{3n_1+1}>2^{2(N+2)}$. (Though $\sigma$ is an endomorphism,
$\sigma^{-n}$ on $B_{3n_2+1}\cap B_{3n_1+3}'$ is well defined for
$n=1,2,\ldots,k_{3n_1+2}$.) Let $w_2$ be a partial isometry in
$B_{3n_2+1}\cap B_{3n_2+3}'$ such that
$$
w_2^\ast w_2=e_1^{(2)}=\mathop{\rm supp}\nolimits(\varphi_1|_{B_{3n_2+1}\cap B_{3n_1+3}'})
$$
and
$$
w_2 w_2^\ast=e_2^{(2)}=\mathop{\rm supp}\nolimits(\varphi_2|_{B_{3n_2+1}\cap B_{3n_1+3}'})\;,
$$
and let $\zeta$ be a partial isometry in $A_{k_{3n_2+2}+1}\cap
A_{k_{3n_1+3}}'$ such that $\zeta^\ast\zeta=e_1^{(2)}$ and
$\zeta\zeta^\ast=\sigma(e_1^{(2)})$.
Assume for the moment that $\sigma^\ell(e_i^{(2)})$, $i=1,2$;
$\ell=-2^{N+1},-2^{N+1}+1,\ldots,2^{N+1}$ are all orthogonal and set
$$
e_{ij}=
\left\{\!\!
\begin{array}{ll}
\sigma^{i-1}(\zeta)\sigma^{i-2}(\zeta)\ldots \sigma^{j}(\zeta)
& \quad i>j \\ [.5ex]
\sigma^i(e_1^{(2)}) & \quad i=j \\ [.5ex]
\sigma^i(\zeta^\ast)\sigma^{i+1}(\zeta^\ast)\ldots
\sigma^{j-1}(\zeta^\ast) & \quad i<j\end{array}\right.
$$
for $i,j=-2^{-N+1},\ldots,2^{N+1}$. Then $(e_{ij})$ is a family of matrix
units such that $\sigma(e_{ij})=e_{i+1,j+1}$ when $|i|, |i+1|,|j|, |j+1|\leq
2^{N+1}$. Let
\begin{eqnarray*}
\lefteqn{E=e_1^{(2)}+
\sum_{\ell=1}^{2^{N+1}-1}(1-e_1^{(2)})
\bigg\{ \frac{2^{N+1}-\ell}{2^{N+1}}e_{\ell,\ell}
+\frac{\ell}{2^{N+1}}e_{\ell-2^{N+1},\ell-2^{N+1}}} \\
&&+\frac{1}{2^{N+1}}
\sqrt{(2^{N+1}-\ell)\ell}\;
(e_{\ell,\ell-2^{-N+1}}+e_{\ell-2^{-N+1},\ell})\bigg\} (1-e_1^{(2)})
\end{eqnarray*}
as in \cite{Kis95}. Then $E$ is a projection in $D_2=A_{(k_{3n_2+2}+2^{N+1})}
\cap A_{(k_{3n_1+3}-2^{N+1})}'$ and satisfies
$$
\|\sigma(E)-E\|\sim \frac{1}{2^{\frac{N+1}{2}}}\;.
$$
Let $w=w_2+(1-e_2^{(2)})
\Big( \sum\limits_{\ell=1}^{2^{N+1}}(\sigma^\ell(w_2)+
\sigma^{-\ell}(w_2))\Big)(1-e_1^{(2)})$ and
$$
v=wE+(1-F)w^\ast(1-E)+(1-F)(1-E)
$$
where $F=wEw^\ast$.
By the orthogonality assumption on $\sigma^{\ell}(e_i^{(2)})$,
$v$ is a unitary in
$D_2$ and satisfies
\begin{eqnarray*}
&&\|\sigma(v)-v\|\approx\|\sigma(E)-E\|\;, \\
&&v e_1^{(2)}=w_2e_1^{(2)}=e_2^{(2)}w_2=e_2^{(2)}v\;.
\end{eqnarray*}
Note also that $v$ commutes with $v_1$ and $e_i^{(1)}$.
Now, the projections $\sigma^{\ell}(e_i^{(2)})$, $i=1,2$,
$\ell=-2^{N+1},\ldots,2^{N+1}$ are not actually orthogonal but
choosing $n_2$ so large that they are very close to being orthogonal,
we may obtain a unitary $v_2$ in $D_2$ by polar decomposition of $v$
such that $v_2$ satisfies the same conditions as above, i.e.,
\begin{eqnarray*}
&&v_2e_1^{(2)}=w_2e_1^{(2)}=e_2^{(2)}w_2=e_2^{(2)}v_2\in B_{3n_2+1}\cap
B_{3n_1+3}'\;, \\
&&\|\lambda(v_2)-v_2\|\sim 2\raise.5ex\hbox{$^{-\frac{N+1}{2}}$}
\end{eqnarray*}
and $v_2\in D_2$.
Since
\begin{eqnarray*}
\lefteqn{
\mathop{\rm supp}\nolimits(\varphi_1|_{B_{3n_2+1}})} \\
&&=\mathop{\rm supp}\nolimits(\varphi_1|_{B_{3n_1+1}})
\mathop{\rm supp}\nolimits(\varphi_1|_{B_{3n_1+3}\cap B_{3n_1+1}'})
\mathop{\rm supp}\nolimits(\varphi_1|_{B_{3n_2+1}\cap B_{3n_1+3}'}) \\
&&=
e_1^{(1)}pe_1^{(2)}
\end{eqnarray*}
with $p=\mathop{\rm supp}\nolimits(\varphi_1|_{B_{3n_1+3}\cap B_{3n_1+1}'})=
\mathop{\rm supp}\nolimits(\varphi_2|_{B_{3n_1+3}\cap B_{3n_1+1}'})$, and since the operators
$v_1e_1^{(1)}=e_2^{(1)}v_1,p$, and $v_2e_1^{(2)}=e_2^{(2)}v_2$ commute, we
obtain that
\begin{eqnarray*}
\lefteqn{v_1v_2\cdot\mathop{\rm supp}\nolimits(\varphi_1|_{B_{3n_2+1}})=v_1v_2e_1^{(1)}
pe_1^{(2)}} \\
&&=v_1e_1^{(1)}v_2e_1^{(2)}p \\
&&=e_2^{(1)}v_1e_2^{(2)}v_2p \\
&&=pe_2^{(1)}e_2^{(2)}v_1v_2=\mathop{\rm supp}\nolimits(\varphi_2|_{B_{3n_2+1}})v_1v_2\;.
\end{eqnarray*}
Here we have also used the fact that $v_1$ commutes with $e_2^{(2)}$.
We repeat this procedure. Thus we obtain an increasing sequence $\{n_k\}$ in
${\bf N}$ and a sequence $\{v_k\}$ of mutually commuting unitaries such that
\begin{eqnarray*}
&&\|\lambda(v_k)-v_k\|\sim
2\raise.5ex\hbox{$^{-\frac{N+k}{2}}$}\;, \\
&&v_k e_1^{(k)}=e_2^{(k)}v_k\in{\mathcal B}_{3n_k+1}\cap
{\mathcal B}_{3n_{k-1}+3}'
\end{eqnarray*}
where
$$
e_i^{(k)}=\mathop{\rm supp}\nolimits(\varphi_i|_{{\mathcal B}_{3n_k+1}\cap
{\mathcal B}_{3n_{k-1}+3}'})\;,
$$
and such that $\mathop{\rm Ad}\nolimits(v_k\ldots v_1)$ maps
$\mathop{\rm supp}\nolimits(\varphi_1|_{{\mathcal B}_{3n_k+1}})$ into
$\mathop{\rm supp}\nolimits(\varphi_2|_{{\mathcal B}_{3n_k+1}})$. Then the limit
$\alpha=\lim\limits_k\mathop{\rm Ad}\nolimits(v_k\ldots v_1)$ defines the desired automorphism.
\end{proof}
\begin{theorem}
Let $\varphi_1$ and $\varphi_2$ be gauge-invariant pure states of ${\mathcal
O}_d$. Then there exists an automorphism $\alpha$ of ${\mathcal O}_d$ such
that $\varphi_1=\varphi_2\circ\alpha$.
\end{theorem}
\begin{proof}
If $\varphi_1$ is disjoint from $\varphi_2$, then it follows that
$(\varphi_i|_{{\rm UHF}_d})\circ \sigma^n=\varphi_i\circ\lambda^n|_{{\rm UHF}_d}$,
$i=1,2$,
$n=0,1,2,\ldots$ are mutually disjoint (by Lemma~4); thus the assertion
follows from Lemma~5. If
$\varphi_1$ is equivalent to
$\varphi_2$, there is a unitary $u\in{\mathcal O}_d$ such that
$\varphi_1=\varphi_2\mathop{\rm Ad}\nolimits u$ (by Kadison's transitivity).
\end{proof}
\section{Pure states mapped into Cuntz states by endomorphisms}
\setcounter{equation}{0}
There is a one-to-one correspondence between the set ${\mathcal U}({\mathcal
O}_d)$ of unitaries of ${\mathcal O}_d$ and the set $\mathop{\rm End}\nolimits({\mathcal O}_d)$ of
unital endomorphisms of ${\mathcal O}_d$; if $u\in {\mathcal U}({\mathcal
O}_d)$, the endomorphism $\alpha_u$ is defined by $\alpha_u(s_i)=us_i$ and if
$\alpha\in\mathop{\rm End}\nolimits({\mathcal O}_d)$, $\alpha$ corresponds to the unitary $u$
defined by $u=\sum\limits_{i=1}^d \alpha(s_i)s_i^\ast$. Define
\begin{eqnarray*}
&&{\mathcal U}_i=\{u\in {\mathcal U}({\mathcal O}_d)|\quad \alpha_u\;\,
\mbox{is an inner automorphism}\} \\
&&{\mathcal U}_a=\{u\in {\mathcal U}({\mathcal O}_d)|\quad \alpha_u\;\,
\mbox{is an automorphism}\} \\
&&{\mathcal U}_s={\mathcal U}({\mathcal O}_d)\setminus {\mathcal U}_a\;.
\end{eqnarray*}
\begin{proposition}
Let ${\mathcal U}_i,{\mathcal U}_a,{\mathcal U}_s$ be as above.
(i)\quad
${\mathcal U}_i$ is a dense subset of ${\mathcal U}({\mathcal O}_d)$.
(ii)\quad
${\mathcal U}_a$ is a dense $G_\delta$ subset of ${\mathcal U}({\mathcal
O}_d)$.
(iii)
$\;{\mathcal U}_s$ is a dense $F_\sigma$ subset of ${\mathcal U}({\mathcal
O}_d)$.
\end{proposition}
\begin{proof}
M.~R\o{}rdam proved (i) in [R\o{}r93] and the other statements are more or
less known.
We shall give a proof of (i). We again denote by $\lambda$ the canonical
endomorphism of ${\mathcal O}_d:\lambda(x)=\sum\limits_{i=1}^d s_ixs_i^\ast$,
$x\in{\mathcal O}_d$. Since the unitary corresponding to $\mathop{\rm Ad}\nolimits v$ is
$v\lambda(v^\ast)$, it suffices to show that $v\lambda(v^\ast)$, $v\in
{\mathcal U}({\mathcal O}_d)$, is dense in ${\mathcal U}({\mathcal O}_d)$. If
${\rm UHF}_d$ denotes the C*-subalgebra generated by $s_{i_1}s_{i_2}\ldots
s_{i_n}s_{j_n}^\ast\ldots s_{j_1}^\ast$, then we mentioned in the
introduction that ${\rm UHF}_d$ is isomorphic to the UHF algebra
$\bigotimes\limits_{{\bf N}}M_d$ and $\lambda|{\rm UHF}_d$ corresponds to the one-sided
shift on $\bigotimes\limits_{{\bf N}}M_d$. Thus $\lambda|{\rm UHF}_d$ satisfies the
Rohlin property, \cite{BKRS93}, \cite{Kis95}. In particular for any $n$ and
$\varepsilon>0$ there is an orthogonal family $e_0,e_1,\ldots,e_{n-1}$ of
projections in
${\rm UHF}_d$ such that
\begin{eqnarray*}
&&\sum_{i=0}^{d^n-1} e_i=1 \\
&&\|\lambda(e_i)-e_{i+1}\|<\varepsilon
\end{eqnarray*}
with $e_{d^n}=e_0$. The similar properties hold for $\mathop{\rm Ad}\nolimits u\circ\lambda$,
i.e., if ${\rm UHF}_d^u$ denotes the C*-subalgebra generated by
$us_{i_1}us_{i_2}\ldots us_{i_n}s_{j_n}^\ast u^\ast\ldots
s_{j_1}^\ast u^\ast$, then $\mathop{\rm Ad}\nolimits u\circ\lambda|{\rm UHF}_d^u$ corresponds to the
one-sided shift on $\bigotimes\limits_{{\bf N}}M_d$. Hence for any $n$ and
$\varepsilon>0$ there is an orthogonal family $f_0,f_1,\ldots,f_{d^n-1}$ of
projections in ${\rm UHF}_d^u$ such that
\begin{eqnarray*}
&&\sum_{i=0}^{d^n-1}f_i=1 \\
&&\|\mathop{\rm Ad}\nolimits u\circ\lambda(f_i)-f_{i+1}\|<\varepsilon
\end{eqnarray*}
with $f_{d^n}=f_0$. Suppose we have chosen such projections $e_i,f_i$ for
the same $n$. Since $K_0({\mathcal O}_d)={\bf Z}/(d-1){\bf Z}$, we have that
$[e_0]=1=[f_0]$ in $K_0({\mathcal O}_d)$ and so obtain a partial isometry
$w\in{\mathcal O}_d$ such that $w^\ast w=e_0$, $ww^\ast=f_0$. We find
unitaries $v_1,v_2\in{\mathcal O}_d$ such that $\mathop{\rm Ad}\nolimits
v_1\lambda(e_i)=e_{i+1}$, $\mathop{\rm Ad}\nolimits v_2\mathop{\rm Ad}\nolimits u\lambda(f_i)=f_{i+1}$, and
$\|v_1-1\|\approx 0$, $\|v_2-1\|\approx 0$ (depending on $\varepsilon$). Let
$$
z=w^\ast(L_{v_2u}R_{v_1^\ast}\lambda)^{d^n}(w)
$$
where $R_{v_1^\ast}$ is the right multiplication by $v_1^\ast$ and
$L_{v_2u}$ is the left multiplication by $v_2u$. Since
$(L_{v_2u}R_{v_i^\ast}\lambda)^i(w)$ is a partial isometry with initial
projection $e_i$ and final projection $f_i$, $z$ is a unitary in
$e_0{\mathcal O}_d e_0$. Since $K_1({\mathcal O}_d)=0$ and ${\mathcal O}_d$
has real rank zero, we find a sequence
$z_0,z_1,\ldots,z_{d^n-1}$ of unitaries in $e_0{\mathcal O}_d e_0$ such that
$z_0=z$, $z_{d^n-1}=1$,
$$
\|z_i-z_{i+1}\|<4/d^n\;.
$$
Define a unitary $v$ by
$$
v=\sum_{i=0}^{d^n-1}(L_{v_2u}R_{v_1^\ast}\lambda)^i(wz_i)
$$
Then since
\begin{eqnarray*}
\lefteqn{v-(L_{v_2u}R_{v_1^\ast}\lambda)(v)} \\
&&=\sum_{i=1}^{d^n-1}(L_{v_2u}R_{v_1^\ast}\lambda)^i(wz_i-wz_{i-1})
+wz_0-(L_{v_2u}R_{v_1^\ast}\lambda)^{d^n}(w)\;,
\end{eqnarray*}
it follows that
$$
\|v-L_{v_2u}R_{v_1^\ast}\lambda(v)\|<4/d^n
$$
or
$$
\|v-u\lambda(v)\|\lesssim 4/d^n\;.
$$
This completes the proof of (i).
Since ${\mathcal U}_a\supset {\mathcal U}_i$, ${\mathcal U}_a$ is dense. That
${\mathcal U}_a$ is a
$G_\delta$ set follows from
$$
{\mathcal U}_a=\bigcap_n \bigcap_j \bigcup_i \bigg\{
u\in {\mathcal U}({\mathcal O}_d);
\|\alpha_u(x_i)-x_j\|<\frac{1}{n}\bigg\}
$$
where $\{x_i\}$ is a dense sequence in ${\mathcal O}_d$.
If ${\mathcal U}_a$ contains a non-empty open set, then it follows that
${\mathcal U}_a={\mathcal U}({\mathcal O}_d)$ or ${\mathcal U}_s=\emptyset$.
Because for any unitaries $u,w$ of ${\mathcal O}_d$ we find a unitary $v$
such that
$w\lambda(v)\approx vu$. (Apply the previous argument for the endomorphism
$\mathop{\rm Ad}\nolimits u\circ\lambda$ instead of
$\lambda$ and the unitary $wu^\ast$.) Since
$v{\mathcal U}_a\lambda(v^\ast)={\mathcal U}_a$ for any unitary
$v\in{\mathcal O}_d$, the above fact implies that ${\mathcal U}_a$ contains an
arbitrary unitary. But we know that
${\mathcal U}_s\not=\emptyset$. For example if
$u=\sum s_is_js_i^\ast s_j^\ast$, then $\alpha_u=\lambda$ and
$\lambda({\mathcal O}_d)'\simeq M_d$. Thus we obtain that ${\mathcal U}_s$ is
dense.
\end{proof}
For a unit vector $\xi\in{\bf C}^d$ we have defined the Cuntz state $f_\xi$ of
${\mathcal O}_d$ by
$$
f_\xi(s_{i_1}\ldots s_{i_m}s_{j_n}^\ast\ldots
s_{j_1}^\ast)=\xi_{i_1}\ldots
\xi_{i_m} \overline{\xi_{j_n}}\ldots\overline{\xi_{j_1}}
$$
It follows that $f_\xi$ is a unique pure state of ${\mathcal O}_d$ satisfying
$$
f_\xi\bigg(\sum_{i=1}^d\overline{\xi_i}s_i\bigg)=1\;.
$$
Let $F$ be the linear span of $s_is_j^\ast$, $i,j=1,\ldots,d$. Then $F$ is
isomorphic to $M_d$ and each unitary $u$ in $F$ defines an automorphism
$\alpha_u$ of ${\mathcal O}_d$. This group of automorphisms acts transitively
on the compact set of Cuntz states.
We denote by $f_0$ the Cuntz state $f_\xi$ with $\xi=(1,0,\ldots,0)$.
\begin{proposition}
If $\varphi$ is a pure state of ${\mathcal O}_d$, there is a unital
endomorphism $\alpha$ of ${\mathcal O}_d$ such that $\varphi\circ\alpha=f_0$,
where $f_0$ is the Cuntz state defined above. Furthermore $\alpha$ may be
chosen so that $\pi_\varphi\circ\alpha({\mathcal O}_d)''$ contains the
one-dimensional projection onto ${\bf C}\Omega_\varphi$.
\end{proposition}
\begin{proof}
It suffices to show that if $\varphi$ is a pure state there is a unitary
$u\in{\mathcal O}_d$ such that
$$
\varphi(us_1)=1\;.
$$
Since ${\mathcal O}_d$ has real rank zero, there is a decreasing sequence
$(e_n)$ of projections in ${\mathcal O}_d$ such that $\varphi$ is the unique
state satisfying $\varphi(e_n)=1$ for $n=1,2,\ldots$, i.e., $(e_n)$ converges
to the support projection of $\varphi$ in ${\mathcal O}_d^{\ast\ast}$. We may
further assume that $[e_n]=0$ in $K_0({\mathcal O}_d)$.
Pick up a projection $e=e_n$ such that $\varphi(e)=1$ and $e<1$. Then
$es_1^\ast$ is a partial isometry with initial projection $s_1es_1^\ast$ and
final projection $e$. Let $w$ be a partial isometry such that $w^\ast
w=1-s_1es_1^\ast$ and $ww^\ast=1-e$. Then
$u=es_1^\ast+w$ is a unitary in ${\mathcal O}_d$ such that
$$
us_1e=(es_1^\ast+w)s_1e=e\;.
$$
Thus we have that $\varphi(us_1)=1$.
To prove the last statement we shall modify $u$ so that $\varphi$ is the
unique state satisfying
$$
\varphi(us_1)=1\;.
$$
We have chosen $e=e_n$. We let
$$
h=\sum_{k=1}^\infty 2^{-k}e_{n+k}\;.
$$
Then $h$ is self-adjoint with $0\leq h\leq 1$ and $\varphi$ is the only state
satisfying $\varphi(h)=1$. Let
$$
u_1=e^{2\pi ih}u\;.
$$
Then $u_1 s_1e=e^{2\pi ih}e$ and the assertion follows.
\end{proof}
|
\section{Introduction}
The coupling between copropagating optical pulses in a nonlinear medium has led to many
important applications in optical fiber systems such as optical switching and
soliton-dragging logic gates \cite{Islam}. The governing equation for the propagation
of two orthogonally polarized pulses in a monomode birefringent fiber is given by the
coupled nonlinear Schr\"{o}dinger(NLS) equation, where the nonlinear coupling terms
are determined by the third-order susceptibility tensor $\chi^{(3)} $ of the fiber.
In an isotropic medium, the tensor $\chi^{(3)} $ has three independent components
$\chi^{(3)}_{xxyy}, \chi^{(3)}_{xyxy} $ and $ \chi^{(3)}_{xyyx}$
and the nonlinear polarization components which account for the nonlinear coupling terms
take the form
\begin{eqnarray}
P_{x}&=& {3\epsilon_{0} \over
2}\Big[ [(\chi^{(3)}_{xxyy}+\chi^{(3)}_{xyxy}+\chi^{(3)}_{xyyx})|E_{x}|^2
+(\chi^{(3)}_{xxyy}+\chi^{(3)}_{xyxy})|E_{y}|^2 ]E_{x}
+\chi^{(3)}_{xyyx}E_{y}^2 E_{x}^{*} \Big], \nonumber \\
P_{y}&=& {3\epsilon_{0} \over
2}\Big[ [ (\chi^{(3)}_{xxyy}+\chi^{(3)}_{xyxy}+\chi^{(3)}_{xyyx})|E_{y}|^2
+(\chi^{(3)}_{xxyy}+\chi^{(3)}_{xyxy})|E_{x}|^2 ]E_{y}
+\chi^{(3)}_{xyyx}E_{x}^2 E_{y}^{*} \Big].
\label{Pnl}
\end{eqnarray}
In the case of silicar fibers, $\chi^{(3)}_{xxyy}\approx \chi^{(3)}_{xyxy} \approx
\chi^{(3)}_{xyyx}$ and the nonlinear terms above have a ratio of
3;2;1. However, when the fiber is elliptically birefringent with the ellipticity angle
$\theta \approx 35^{o}$,
and also the beat length due to birefrigence is much smaller than the typical propagation
distances, the coupled NLS equation takes the form of the vector NLS equation whose
nonlinear terms have a ratio of 1;1;0 \cite{Agr},
which is known to be integrable via the inverse scattering method \cite{ZaSha}\cite{zak}.
In general, the coupled NLS equations with arbitrary coefficients are not integrable.
Mathematically, there exists a systematic way of generalizing the NLS equation to the
multi-component cases \cite{fordy} and to the higher-order cases \cite{kps}
using group theory which preserves the integrabiiity structure.
This gives rise to various integrable, coupled NLS equations among $N$ scalar
fields $\psi_i ; i = 1,..., N$ with specific set of coupling parameters.
For $N=2$, the vector NLS equation is the only nontrivial integrable equation in the
group theoretic construction.
However, it is not known whether there can be other cases of the integrable coupled NLS
equation for $N=2$ with nonlinear coupling terms as in Eq. (\ref{Pnl}) except for the
vector NLS equation.
In this paper, using the Painlev\'{e} analysis we investigate the integrability properties
of the coupled NLS equation relevant to the propagation of orthogonally polarized
optical waves in an isotropic medium. Motivated by Eq. (\ref{Pnl}), we consider the general
form of the coupled NLS equation such that
\begin{eqnarray}
i \bar{\partial } q_1 &=& \partial ^2 q_1 + q_1 (\gamma _1 |q_1|^2 + \gamma _2 |q_2|^2 ) + \gamma _3 q_1^*
q_2^2 + \gamma _4 q_1^2 q_2^*, \nonumber \\
i \bar{\partial } q_2 &=& \beta \partial ^2 q_2 + q_2 (\gamma _2 |q_1|^2 + \gamma _1 |q_2|^2 ) + \gamma _3 q_2^*
q_1^2 + \gamma _4 q_2^2 q_1^* ,
\label{nlseqns}
\end{eqnarray}
where $\beta = \pm 1$ signify the relative sign of the group-velocity dispersion terms and we
use the notation $\partial = \partial /\partial z , \bar{\partial } =\partial / \partial \bar{z} $.
We find that the system passes the Painlev\'{e} test whenever the parameters belong to one of
the following four classes; $(i) ~\beta =1, \gamma_1 = \gamma_2 ,
\gamma_3 = \gamma_4 =0 $, $(ii) ~\beta =1 , ~ \gamma_2 = 2 \gamma_1 , \gamma_3 = -\gamma_1 ,
\gamma_4 \mbox{ arbitrary }$, $(iii)~ \beta =1, \gamma_2 = 2 \gamma_1 ,
\gamma_3 = \gamma_1 , \gamma_4 = 0$ and $(iv) ~\beta = -1, \gamma_1 =- \gamma_2 ,
\gamma_3 = \gamma_4 =0 $.
Case (i)( and (iv)) is the well-known vector NLS equation. The integrability of cases (i) and (iv) have
been demonstrated by Zakharov and Schulman by deriving an appropriate inverse
scattering formalism \cite{zak}\cite{zak1}. However, cases (ii) and (iii) are new as far as we know. In particular,
case (ii) corresponds to the propagation in the isotropic nonlinear medium with
the property that $\chi^{(3)}_{xxyy}+\chi^{(3)}_{xyxy}=-2\chi^{(3)}_{xyyx}$.
We find the Hirota bilinearization and the B\"{a}cklund transformation
of cases (ii) and (iii), and compute soliton solutions. As for the integrability of
cases (ii) and (iii), we prove that they are essentially identical to two independent
NLS equations. This implies that in the case (ii), there is no physical interactions between
two optical pulses with opposite circular polarizations.
We also show that our Painlev\'{e} analysis is consistent with the group theoretical method of
generalizing the integrable NLS equations when the group theoretical method is combined with
the reduction procedure.
\section{Painlev\'{e} analysis of the coupled NLS equation }
The Painlev\'{e} analysis for a partial differential equation was first introduced by
Weiss, Tabor, and Carnevale\cite{weiss} who defined that a partial differential equation
has the Painlev\'{e} property if its general solution is single-valued about the movable
singularity manifold. This method is to seek a solution of a given differential equation
in a series expansion in terms of $\phi (z,\bar{z} ) = z-\psi (\bar{z} )$, where $\psi (\bar{z} )$ is an
arbitrary
analytic function of $\bar{z} $ and $\phi =0$ defines a non-characteristic movable singularity
manifold. Then, the equation has the Painlev\'{e} property, thus becomes integrable,
if there exists a sufficient number of arbitrary functions in the series solution.
For $\beta =1$, we postulate a solution of the form,
\begin{eqnarray}
q_1 &=& \sum_{m \ge 0} R_m (\bar z) (z-\psi )^{m-\sigma}, \nonumber \\
q_1^* &=& \sum_{m \ge 0} S_m (\bar z) (z-\psi )^{m-\sigma}, \nonumber \\
q_2 &=& \sum_{m \ge 0} T_m (\bar z) (z-\psi )^{m-\sigma}, \nonumber \\
q_2^* &=& \sum_{m \ge 0} U_m (\bar z) (z-\psi )^{m-\sigma}.
\label{subst}
\end{eqnarray}
Substituting these ans\"{a}tze into Eq. (\ref{nlseqns}) and looking at the leading order behavior,
we find that $\sigma=1$ and the following equations should be satisfied:
\begin{eqnarray}
\gamma _1 U_0^2 T_0 + \gamma _2 R_0 S_0 U_0 + \gamma _3 S_0^2 T_0 + \gamma _4 U_0^2 R_0 + 2 U_0 &=&
0, \nonumber \\
\gamma _1 T_0^2 U_0 + \gamma _2 R_0 S_0 T_0 + \gamma _3 R_0^2 U_0 + \gamma _4 T_0^2 S_0 + 2 T_0 &=&
0, \nonumber \\
\gamma _1 R_0^2 S_0 + \gamma _2 R_0 T_0 U_0 + \gamma _3 T_0^2 S_0 + \gamma _4 R_0^2 U_0 + 2 R_0 &=&
0, \nonumber \\
\gamma _1 S_0^2 R_0 + \gamma _2 S_0 T_0 U_0 + \gamma _3 U_0^2 R_0 + \gamma _4 S_0^2 T_0 + 2 S_0 &=&
0.
\label{init}
\end{eqnarray}
In order to facilitate solving Eq. (\ref{init}), we define
$x \equiv U_0 R_0,~ y \equiv T_0 S_0, ~t \equiv R_0 S_0,~ s \equiv
U_0 T_0$ so that the first two equations in Eq. (\ref{init}) can be written as
\begin{eqnarray}
\gamma _1 s + \gamma _2 t +2 + \gamma _4 x = -\gamma _3 {ty \over x}, \nonumber \\
\gamma _1 s + \gamma _2 t +2 + \gamma _4 y = -\gamma _3 {tx \over y},
\label{red1}
\end{eqnarray}
while the last two as
\begin{eqnarray}
\gamma _1 t + \gamma _2 s +2 + \gamma _4 x = -\gamma _3 {sy \over x}, \nonumber \\
\gamma _1 t + \gamma _2 s +2 + \gamma _4 y = -\gamma _3 {sx \over y}.
\label{red2}
\end{eqnarray}
Each pair can be combined to give $(x-y)(\gamma _4 - \gamma _3 t {x+y \over xy})=0$,
and $(x-y)(\gamma _4 - \gamma _3 s {x+y \over xy})=0$. One can readily check that solutions
of these equations can be classified in seven different cases,
\vglue .2in
\noindent
(case 1) $x=y, \gamma_{1}=\gamma_{2}+\gamma_{3}$, \\
(case 2) $x=y, t = x $, \\
(case 3) $x=y, t=-x $, \\
(case 4) $t=s, \gamma _4 = \gamma _3 t {x+y \over xy}$, \\
(case 5) $x=-y, \gamma_{4}=0, \gamma_{2}=\gamma_{1}+\gamma_{3}, t+s=-2/\gamma_{1} $
\\
(case 6) $\gamma _3=\gamma _4=0, s=t=-2/(\gamma_{1}+\gamma_{2})$.
\\
(case 7) $\gamma_{3}=\gamma_{4}=0, \gamma_{1}=\gamma_{2}, t+s=-2$
\vskip .2in
For each cases, we check the powers, so called resonances, at which the arbitrary functions
can arise in the series solution. Equating coefficients of the $(z-\psi)^{j-3}$ term in Eq.
(\ref{nlseqns}) with the ans\"{a}tze in Eq. (\ref{subst}), we obtain a system of four linear
algebraic equations in ($R_j, S_j, T_j, U_j$) which are given in a matrix form by,
\begin{equation}
Q_{j} \pmatrix{ R_j \cr S_j \cr T_j \cr U_j } =\pmatrix{ F_j \cr G_j \cr H_j \cr K_j }.
\label{recursion}
\end{equation}
The $4 \times 4$ matrix $Q_{j}=(j-1)(j-2) I_{4 \times 4} +
\pmatrix{Q^{(1)}_{j} & Q^{(2)}_{j} \cr
Q^{(3)}_{j} & Q^{(4)}_{j} }$ has block components:
\begin{eqnarray}
Q^{(1)}_{j} &=&
\pmatrix{ 2 \gamma _1 R_0 S_0 + \gamma _2 T_0 U_0 +2 \gamma _4 R_0 U_0
&
\gamma _1 R_0^2 + \gamma _3 T_0 ^2 \cr
\gamma _1 S_0^2 + \gamma _3 U_0 ^2 &
2 \gamma _1 R_0 S_0 + \gamma _2 T_0 U_0 +2 \gamma _4 S_0 T_0 }
,
\nonumber \\
Q^{(2)}_{j} &=&
\pmatrix{\gamma _2 R_0 U_0 + 2 \gamma _3 T_0 S_0
&
\gamma _2 R_0 T_0 + \gamma _4 R_0 ^2 \cr
\gamma _2 S_0 U_0 + \gamma _4 S_0 ^2
&
\gamma _2 S_0 T_0 + 2 \gamma _3 U_0 R_0 } ,
\nonumber \\
Q^{(3)}_{j} &=&
\pmatrix{ \gamma _2 T_0 S_0 + 2 \gamma _3 R_0 U_0
&
\gamma _2 R_0 T_0 + \gamma _4 T_0 ^2
\cr
\gamma _2 U_0 S_0 + \gamma _4 U_0 ^2
&
\gamma _2 R_0 U_0 + 2 \gamma _3 T_0 S_0 } ,
\nonumber \\
Q^{(4)}_{j} &=&
\pmatrix{ 2 \gamma _1 T_0 U_0 + \gamma _2 R_0 S_0 +2 \gamma _4 S_0 T_0
&
\gamma _1 T_0^2 + \gamma _3 R_0 ^2
\cr
\gamma _1 U_0^2 +\gamma _3 S_0 ^2
&
2 \gamma _1 T_0 U_0 + \gamma _2 R_0 S_0 +2 \gamma _4 U_0 R_0} ,
\end{eqnarray}
and
\begin{eqnarray}
F_j &\equiv&
-\sum^{l+m+n=j }_{0 \le l , m, n <j } (\gamma _1 R_l R_m S_n + \gamma _2 R_l T_m U_n
+\gamma _3 S_l T_m T_n + \gamma _4 R_l R_m U_n) + i R_{j-2} ^{'} - i (j-2) \psi^{'} R_{j-
1} , \nonumber \\
G_j &\equiv&
-\sum^{l+m+n=j }_{0 \le l , m, n <j } (\gamma _1 S_l S_m R_n + \gamma _2 S_l U_m T_n
+\gamma _3 R_l U_m U_n + \gamma _4 S_l S_m T_n) - i S_{j-2} ^{'} + i (j-2) \psi^{'} S_{j-
1} , \nonumber \\
H_j &\equiv&
-\sum^{l+m+n=j }_{0 \le l , m, n <j } (\gamma _1 T_l T_m U_n + \gamma _2 T_l R_m S_n
+\gamma _3 U_l R_m R_n + \gamma _4 T_l T_m S_n) + i T_{j-2} ^{'} - i (j-2) \psi^{'} T_{j-
1} , \nonumber \\
K_j &\equiv&
-\sum^{l+m+n=j }_{0 \le l , m, n <j } (\gamma _1 U_l U_m T_n + \gamma _2 U_l S_m R_n
+\gamma _3 T_l S_m S_n + \gamma _4 U_l U_m R_n) - i U_{j-2} ^{'} +i (j-2) \psi^{'} U_{j-
1} .
\end{eqnarray}
The resonances occur when $\det Q_{j} = 0$. Now, we compute the resonance values and check
the Painlev\'{e} property of Eq. (\ref{nlseqns}) for each seven cases as introduced above.
\vskip .2in
\begin{center}
{\bf Case 1. $\bf x=y; ~ \gamma_{1}=\gamma_{2}+\gamma_{3}$}
\end{center}
\vskip .2in
In this case, we can solve for $T_{0} , R_{0}$ such that
\begin{equation}
T_0={-2 U_0 \over \gamma _1 (S_0^2 + U_0^2) +\gamma _4 S_0 U_0}, \ \
R_0={-2 S_0 \over \gamma _1 (S_0^2 + U_0^2) +\gamma _4 S_0 U_0} .
\label{case1a1}
\end{equation}
When we substitute these solutions into the resonance condition, $\det Q_{j} = 0$, we find that
the resonances do not occur at the integer values of $j$. Therefore, this case does not pass
the Painlev\'{e} test for integrability.
\vskip .2in
\begin{center}
{\bf Case 2 and Case 3. $\bf x=y;~~ t=\pm x$ }
\end{center}
\vskip .2in
We have solutions,
\begin{equation}
S_0=\pm U_0={-2 \over \gamma _1 + \gamma _2 +\gamma _3 \pm \gamma _4} {1 \over R_0},\ \ T_0= \pm R_0,
\end{equation}
where $+$ and $-$ sign correspond to the Case 2 and the Case 3 respectively.
Substituting these solutions into the resonance condition $\det Q_{j} = 0$, we
find that the resonance values $j=-1, 0, 1, 1, 2, 2, 3, 4$ occur when $\gamma _2=2 \gamma _1,
\gamma _3= \pm \gamma _4+\gamma _1$. The resonance $j=-1$ is related with the arbitrariness of
$\psi$, while the resonance $j=0$ is related with the arbitrariness of $R_0$. The recursion
relation in Eq. (\ref{recursion}) determines $R_1, S_1, T_1, U_1$ in terms of
$R_0, S_0, T_0, U_0, \psi$. The degree of multiplicity of the resonance $j=1$ is two
and it turns out that there exist two arbitrary functions consistently only if $\gamma_{4}=0$.
Therefore, the case where $\gamma_{2} = 2 \gamma_{1}, \gamma_{3} = \gamma_{1} $
and $\gamma_{4}=0$ passes the Painlev\'{e} test.
\vskip .2in
\begin{center}
{\bf Case 4. $\bf t=s; ~ \gamma_{4}xy=\gamma_{3}t(x+y) $}
\end{center}
\vskip .2in
Eq. (\ref{init}) together with the condition $t=s, ~ \gamma_{4}xy=\gamma_{3}t(x+y) $
results in
\begin{equation}
t = {-2 \over \gamma _1 + \gamma _2 -\gamma _3 + (\gamma _4 ^2 / \gamma _3) },\ \
x = ({\gamma _4 \over 2 \gamma _3 } \pm \sqrt{(\gamma _4/2 \gamma _3)^2 -1}) t ,
\end{equation}
and
\begin{equation}
S_0 = {t^2 \over x} {1 \over T_0},\ \ U_0 = {t \over T_0},\ \ R_0 = {x \over t}
T_0.
\label{zerosol}
\end{equation}
When we substitute these solutions into the resonance condition $\det Q_{j} = 0$, we obtain
\begin{equation}
(j-4)(j-3)j(j+1)(j^2-3j+2 {\gamma _4 ^2 -4 \gamma _3 ^2 \over \gamma _4 ^2 -\gamma _3 ^2 + \gamma _2 \gamma _3
+ \gamma _3 \gamma_1 })(j^2-3j+2 {\gamma _4 ^2 -2 \gamma _3 ^2 +2 \gamma _2 \gamma_3 -2\gamma_1
\gamma _3 \over \gamma _4 ^2 -\gamma _3 ^2 + \gamma _2 \gamma _3 + \gamma_1 \gamma _3})=0.
\label{deter}
\end{equation}
Note that the Painlev\'{e} test requires the resonances $j$ to be integers and
the degeneracy of resonance at $j=0$ to be one since there is only one arbitrary function
$T_0$ as in Eq. (\ref{zerosol}). This requirement leads to the result,
$ \gamma _2=2 \gamma_{1} , \gamma _3= -\gamma_{1} $ and $\gamma_{4}$ arbitrary, so that
resonances are $j=-1,0,1,1,2,2,3,4$. The
recursion relation in Eq. (\ref{recursion}) determines $T_1, U_1, T_2, U_2$ such as
\begin{eqnarray}
T_1 &=& {1 \over 4} (\sqrt{\gamma _4 ^2 -4}-\gamma _4 )(2 R_1 + i\sqrt{\gamma _4 ^2 -4} T_0
\psi_x ), \nonumber \\
U_1 &=& -{1 \over 2} (\sqrt{\gamma _4 ^2 -4}+\gamma _4 )( S_1 + {i\over \sqrt{\gamma _4 ^2 -4}}
{\psi_x \over T_0} ), \nonumber \\
T_2 &=& {1 \over 2} (\sqrt{\gamma _4 ^2 -4}-\gamma _4 )(R_2 + {\sqrt{\gamma _4 ^2 -4} \over 12}
( T_0 \psi_x ^2 + 2 i{\partial T_0 \over \partial x} ) ), \nonumber \\
U_2 &=& {1 \over 12} (\sqrt{\gamma _4 ^2 -4}+\gamma _4 )(-6 S_2 + {i\over \sqrt{\gamma _4 ^2 -
4}} ({\psi_x ^2 \over T_0} +2 i{1 \over T_0 ^2 } {\partial T_0 \over \partial x})).
\end{eqnarray}
Similarly, $R_3, T_3, U_3$ are determined in terms of $\psi, T_0, R_1, S_1, R_2, S_2$.
In the same way, we can check that there exists one arbitrary function at the j=4
resonance and no more arbitrary functions in higher lebels. All these facts have been
confirmed with the symbolic manipulation program Macsyma.
Thus, the system passes the Painlev\'{e} test
when $ \gamma _2=2 \gamma_{1} , \gamma _3= -\gamma_{1} $ and $\gamma_{4}$ arbitrary.
We show that this case is indeed integrable in Sec. 3.
\vskip .2in
\begin{center}
{\bf Case 5. $\bf x=-y; ~ \gamma_{4}=0; \gamma_{2}=\gamma_{1}+\gamma_{3}; t+s=-2/\gamma_{1} $}
\end{center}
\vskip .2in
In this case, the resonances are
at $j=-1, 0, 0, 3, 3, 4, {3 \over 2} \pm \sqrt{9 +16 \gamma _3
/ \gamma _1 }$, which in turn requires that $\gamma _1 = -2 \gamma _3,\ \ \gamma _2 = -\gamma _3.$ But
inconsistency among the four equations in Eq. (\ref{recursion}) arises at the $j=2$
level, so that the Painlev\'{e} test fails.
\begin{center}
{\bf Case 6. $\bf \gamma_{3}=\gamma_{4}=0; s=t=-2/(\gamma_{1}+\gamma_{2})$}
\end{center}
\vskip .2in
The resonance condition, $det Q_{j}=0$, leads to the following solutions;
\begin{equation}
j = -1, 0, 0, 3, 3, 4, {3 \over 2} \pm {1 \over 2(\gamma _1 + \gamma _2)} \sqrt{25 \gamma _1 ^2
+ 18 \gamma _1 \gamma _2 -7 \gamma _2 ^2}.
\end{equation}
The integer resonances occur if (i) $\gamma _2 = 3 \gamma _1$, or (ii) $\gamma _2 = - \gamma _1$.
The first case (i) leads to inconsistencies among four equations in Eq. (\ref{recursion})
at j=2, while the second case (ii) similarly leads to
inconsistency at $j=0$. Therefore, the Painlev\'{e} test fails in this case.
\vskip .2in
\begin{center}
{\bf Case 7. $\bf \gamma_{3}=\gamma_{4}=0; ~ \gamma_{1}=\gamma_{2} ; ~ t+s = -2 $ }
\end{center}
\vskip .2in
This case corresponds to the well-known integrable vector NLS equation considered by
Zakharov and Schulman \cite{zak}. Together with the parameters;
$\gamma _1 = \gamma _2, \gamma _3 = \gamma _4 =0$, Eq. (\ref{init}) reduces to
\begin{equation}
2 + \gamma _1 ( T_0 U_0 + R_0 S_0 ) = 0.
\end{equation}
The resonances are $j=-1, 0, 0, 0, 3, 3, 3, 4$, and it has been checked that
the proper number of arbitrary functions exist. Thus, this case passes the Painlev\'{e} test.
\vskip .2in
So far, we have considered the case where $\beta =1$ in Eq. (\ref{nlseqns}).
For $\beta = -1$,
using the notion of the degenerate dispersion law, Zahkarov and Schulmann found another
integrable theory with anomalous dispersive term \cite{zak1}.
The Painlev\'{e} analysis for the $\beta = -1$ case can be done in the same way as for the
$\beta =1$ case. Thus, we suppress the details of analysis and simply state the results.
The leading order equation is given by
\begin{eqnarray}
\gamma _1 s + \gamma _2 t -2 + \gamma _4 x = -\gamma _3 {ty \over x}, \nonumber \\
\gamma _1 s + \gamma _2 t -2 + \gamma _4 y = -\gamma _3 {tx \over y}, \nonumber \\
\gamma _1 t + \gamma _2 s +2 + \gamma _4 x = -\gamma _3 {sy \over x} \nonumber \\
\gamma _1 t + \gamma _2 s +2 + \gamma _4 y = -\gamma _3 {sx \over y},
\label{red3}
\end{eqnarray}
whose solutions can be grouped into five distinct cases;
\vglue .2in
\noindent
(case 1) $x=y$ \\
(case 2) $\gamma _4 =0, x=-y, \gamma _3=\gamma _1 +\gamma _2$ \\
(case 3) $\gamma _4 =0, x=-y, t=-s $ \\
(case 4) $\gamma _3 = \gamma _4 =0, \gamma_{1} = -\gamma_{2} $ \\
(case 5) $\gamma _3 = \gamma _4 =0, t =-s$ .
\vskip .2in
Here, only the case 4 passes the Painlev\'{e} test. In this case,
$S = (T_0 U_0 -2 )/ R_0 $ and resonances are $j=-1, 0, 0, 0, 3,
3, 3, 4$. This is the integrable system found by Zakharov and Schulmann \cite{zak}.
All other cases lead to inconsistencies at $j=1$ level thus failing
the Painlev\'{e} test.
\section{Hirota bilinearization and solitons}
One of the main result of the Painlev\'{e} test is to find a new case of coupled
NLS equation in Eq. (\ref{nlseqns}) with parameters given by
$\gamma_2 =2 \gamma_1 , \gamma_3 = - \gamma_1 $ and
$\gamma_4 $ arbitrary. With an appropriate scaling, we can always set the nonzero
$\gamma_1 $ to one. Also, as we show in Sec. 4, we can set $\gamma_4 $ to zero.
From now on, we restrict
to this case ($\beta =1, \gamma_1 =1, \gamma_2 =2, \gamma_3 = -1, \gamma_4 =0$)
and analyze its solution and integrability structures.
It is well known that the Painlev\'{e} analysis in the preceding section can be related
to the B\"{a}cklund transformation (BT). In order to derive the BT, we truncate the
series in Eq. (\ref{subst}) up to a constant level term and substitute $(z-\psi)$ by
an arbitrary function $\phi (z, \bar{z})$ to be determined later. Then, the corresponding BT
is given by
\begin{equation}
q_1 = {R_0 \over \phi} + R_1,\ \ q_2 = {T_0 \over \phi} + T_1,
\label{BT}
\end{equation}
where the set $(R_1,T_1)$ is a known solution of the coupled NLS equations,
which we assume to be the trivial solution $R_1=T_1=0$. In order for the new set $(q_1 , q_2 )$
to be also a solution, the following equations should hold \cite{sa}
\begin{eqnarray}
i \phi \bar{D} R_0 \cdot \phi = i \phi D ^2 R_0 \cdot \phi - R_0 D ^2 \phi \cdot \phi + R_0^2
R_0^* +2 R_0 T_0 T_0^* - R_0^* T_0^2 \nonumber \\
i \phi \bar{D} T_0 \cdot \phi = i \phi D ^2 T_0 \cdot \phi - T_0 D ^2 \phi \cdot \phi + T_0^2
T_0^* +2 T_0 R_0 R_0^* - T_0^* R_0^2, \nonumber \\
\label{hirota}
\end{eqnarray}
Here, the Hirota's bilinears $D$ and $\bar{D}$ are defined by
\begin{equation}
\bar{D} ^n D ^m f \cdot g = \Big({\partial \over \partial \bar{z} }-{\partial \over \partial \bar{z} ' }\Big) ^n
\Big({\partial \over \partial z}-{\partial \over \partial z'}\Big) ^m
f(z,\bar{z} ) g(z^{'},\bar{z}^{'})\Big|_{z=z^{'} \atop \bar{z} =\bar{z}^{'}}.
\end{equation}
Equation (\ref{hirota}) can be decoupled as
\begin{eqnarray}
R_0 D ^2 \phi \cdot \phi -( \gamma _1 R_0^2 R_0^* + \gamma _2 R_0 T_0 T_0^* + \gamma _3 R_0^* T_0^2 ) &=& \lambda_1 R_0
\phi \cdot \phi, \nonumber \\
T_0 D ^2 \phi \cdot \phi -( \gamma _1 T_0^2 T_0^* + \gamma _2 T_0 R_0 R_0^* + \gamma _3 T_0^* R_0^2 ) &=& \lambda_2 T_0
\phi \cdot \phi, \nonumber \\
i \bar{D} R_0 \cdot \phi &=& D ^2 R_0 \cdot \phi - \lambda_1 R_0 \cdot \phi , \nonumber \\
i \bar{D} T_0 \cdot \phi &=& D ^2 T_0 \cdot \phi - \lambda_2 T_0 \cdot \phi.
\label{HIR}
\end{eqnarray}
Now, explicit N-solitons can be constructed in the usual way by solving $\phi, R_0, T_0$
in terms of power series.
\subsection{one-soliton}
For one soliton solution, we choose $\lambda_{1}=\lambda_{2}=0$ and assume solutions in a series
form in $\epsilon $ such that $\phi =1 + \epsilon^2 h, ~ R_0=\epsilon R, ~ T_0 = \epsilon T$.
Then, by equating the coefficients of the polynomials to zero in Eq. (\ref{HIR}) and solving
them explicitly, we obtain
\begin{equation}
R = \alpha \exp\Big( i (a^2-b^2) \bar{z} + 2ab\bar{z} + i a z + b z \Big) ,\ \ T =\beta \exp\Big(
i (a^2-b^2) \bar{z} + 2ab \bar{z} +i a z + b z \Big),
\end{equation}
where $\alpha, \beta$ are arbitrary complex numbers while $a,b$ are arbitrary real numbers.
$h$ is also obtained by solving the third order equation such that
\begin{equation}
h={1 \over 8 b^2} (|\alpha|^2 + 2 |\beta|^2 - {\alpha^* \beta^2 \over \alpha}) \exp(2 b z+ 4 ab \bar{z}
).
\end{equation}
Consistency requires that phases of the complex numbers $\alpha$ and $ \beta$
should be either the same, or differ by $\pi /2$.
In the case of the same phase, we parameterize $\alpha $ and $\beta$ by
\begin{equation}
\alpha = \sqrt{8} b \cos k e^{\Delta +i \theta},\ \ \beta = \sqrt{8} b \sin k e^{\Delta +i
\theta},
\end{equation}
in terms of arbitrary real numbers $k, \theta, \Delta$.
Then, the final form of the one soliton solution is given by
substituting $\epsilon =1$ in Eq. (\ref{BT}) such that
\begin{eqnarray}
q_1 &=& \sqrt{2} b \cos k e^{i (a^2-b^2)\bar{z} + i a z +i \theta} {\rm sech}
(bz+2ab\bar{z} +\Delta), \nonumber \\
q_2 &=& \sqrt{2} b \sin k e^{i (a^2-b^2)\bar{z} + i a z+i \theta} {\rm sech}
(bz+2ab\bar{z}+\Delta).
\end{eqnarray}
In the case where phases differ by $\pi /2$, $\alpha $ and $\beta$ are given by
\begin{equation}
\beta = \pm i \alpha = \pm i \sqrt{8} b e^{\Delta +i \theta}.
\end{equation}
Then, the corresponding one-soliton solution is
\begin{eqnarray}
q_1 &=& \sqrt{2} b e^{i (a^2-b^2)\bar{z} + i a z +i \theta}
{\rm sech} (b z+2ab \bar{z}+\Delta) , \nonumber \\
q_2 &=& \pm i \sqrt{2} b e^{i (a^2-b^2)\bar{z} + i a z +i \theta} {\rm sech}
(b z+2ab \bar{z}+\Delta).
\end{eqnarray}
\subsection{two-soliton}
The two-soliton solution can be obtained using the series expansion $\phi=1 + \epsilon^2
h_1 + \epsilon^4 h_2,\ \ R_0=\epsilon \rho_1 + \epsilon^3 \rho_2,\ \ T_0=\epsilon \tau_1 + \epsilon^3 \tau_2$.
Inserting these ans\"{a}tze into Eq. (\ref{HIR}), we obtain solutions
\begin{eqnarray}
\rho_1 &=& f+g , ~ \tau_1 = i \rho_1 ; ~~
f \equiv e^{-i k^2 \bar{z} + k z +\eta_f} , ~ g \equiv e^{-i l^2 \bar{z} + l z +\eta_g
},
\nonumber \\
h_1 &=& 2 \left({f f^* \over (k+k^* )^2} + {f g^* \over (k+l^* )^2} + {g f^* \over
(l+k^* )^2} + {g g^* \over (l+l^* )^2} \right),
\end{eqnarray}
where $k, l, \eta_f, \eta_g$ are arbitrary complex numbers.
Also, after a lengthy but straightforward calculation we obtain
\begin{eqnarray}
\rho_2 &=& 2 (l-k)^2 \left( {f f^* g \over (k+k^*)^2 (l+k^*)^2} + {f g g^* \over
(k+l^*)^2 (l+l^*)^2} \right)
, ~~
\tau_2 = i \rho_2, \nonumber \\
h_2 &=& {4 (l-k)^2 (l^*-k^*)^2 f f^* g g^* \over (k+k^*)^2 (l+k^*)^2 (k+l^*)^2
(l+l^*)^2}.
\end{eqnarray}
Finally, the two-soliton solution is obtained by taking $\epsilon =1$ in
the BT equation $q_1 = {R_0 \over \phi}, q_2 = { T_0 \over \phi } $.
Surprisingly, there exists a different type two-soliton solution which can be
obtained by a simple linear superposition of the left-polarized one-soliton with
the right-polarized one-soliton;
\begin{equation}
q_1 = {f \over \phi_1} + {g \over \phi_2}, \ \ q_2 = i{f \over \phi_1} -i {g \over \phi_2},
\end{equation}
where $\phi_1 = 1 + 2 f f^* / (k+k^* )^2, \phi_2 = 1 + 2 g g^* / (l+l^* )^2 $.
The reason underlying the existence of such a linear superposition
is explained in the following section.
\section{Integrability }
The Painlev\'{e} test in Sec. 2 suggests new integrable cases of coupled NLS equations.
As we have shown in the preceding section, the coupled NLS equation with
$\gamma_1 =1, \gamma_2 =2 , \gamma_3 = -1, \gamma_4 = 0 $ possesses exact soliton solutions, which reflects
the integrability of the equation. Before proving the integrability by deriving the
corresponding Lax pair, we first note that taking $\gamma_4 =0$ is not essential.
Make a change of variables such that
\begin{equation}
Q_1 = x q_1 + y q_2,~~ Q_2 = y q_1 + x q_2.
\label{cil}
\end{equation}
If $(Q_1 , Q_2 )$ satisfy the coupled NLS equation in Eq. (\ref{nlseqns}) with
$\gamma_1 =1, \gamma_2 =2 , \gamma_3 = -1, \gamma_4 = 0 $ ,
then $(q_1 , q_2 )$ satisfy Eq. (\ref{nlseqns}) but with parameters
$\gamma_1 =1, \gamma_2 =2 , \gamma_3 = -1, \gamma _4={4xy \over x^2+y^2}$. Thus, we set $\gamma_4 $ to zero
without loss of generality.
The integrability and the Lax pair of the coupled NLS equation in Eq. (\ref{nlseqns}) with
$\gamma_1 =1, \gamma_2 =2 , \gamma_3 = \pm 1 , \gamma_4 =0$ follows from the observation that these equations
can be embeded in the integrable coupled NLS equation based on the symmetric space $Sp(2)/U(2)$
given by\footnote{The generalization of NLS equation using symmetric spaces and the concept of
matrix potential can be found in \cite{fordy,kps,bps,park1,park2}}
\begin{eqnarray}
i \bar{\partial } \psi_{1} &=& \Big[ \partial ^2 \psi_{1}+ 2 \psi
_{1}^{2}\psi^{*}_{1}+4 \psi_{1}\psi_{2}\psi^{*}_{2}
+2 \psi_{2}^{2}\psi^{*}_{3} \Big],
\nonumber \\
i \bar{\partial } \psi_{2} &=& \Big[ \partial ^2 \psi_{2}+2 \psi_{2}\psi_{1}\psi^{*}_{1}
+2 \psi_{2}^{2} \psi^{*}_{2}+2 \psi_{3}\psi_{1}\psi^{*}_{2}+
2 \psi_{3}\psi_{2}\psi^{*}_{3}
\Big],
\nonumber \\
i \bar{\partial } \psi_{3} &=& \Big[ \partial ^2 \psi_{3}+2 \psi_{3}^{2}\psi^{*}_{3}+4 \psi_{3}\psi_{2}\psi^{*}_{2}
+2 \psi_{2}^{2}\psi^{*}_{1} \Big].
\label{sp}
\end{eqnarray}
Consistent reductions can be made if we take $\psi_{1} = \pm \psi_{3} $, which are precisely the
cases $\gamma_1 =2, \gamma_2 =4 , \gamma_3 = \pm 2, \gamma_4 = 0 $ in Eq. (\ref{nlseqns}). Furthermore,
Eq. (\ref{sp}) arises from the Lax pair
\begin{equation}
L_{z} \Psi \equiv \Big[ \partial + E + \lambda T \Big] \Psi =0, ~~
L_{\bar{z}} \Psi \equiv \Big[ \bar{\partial } + ({1 \over 2}[ E, ~ \tilde{E} ] - \partial \tilde{E} )
- \lambda E - \lambda^2 T \Big] \Psi =0
\label{Lax1}
\end{equation}
where the $4\times 4$ matrices $E$ and $T$ are
\begin{equation}
E = \pmatrix{0 & 0 & \psi_{1} & \psi_{2} \cr
0 & 0 & \psi_{2} & \psi_{3} \cr
-\psi_{1}^{*} & -\psi_{2}^{*} & 0 & 0 \cr
-\psi_{2}^{*} & -\psi_{3}^{*} & 0 & 0 } , ~
T = \pmatrix{i/2 & 0 & 0 & 0 \cr
0 & i/2 & 0 & 0 \cr
0 & 0 & -i/2 & 0 \cr
0 & 0 & 0 & -i/2 }
\label{emat}
\end{equation}
By taking $\psi_{1} = \pm \psi_{3}$ in Eqs. (\ref{Lax1}) and (\ref{emat}), we obtain the
Lax pair for the coupled NLS equation in Eq. (\ref{nlseqns}) with
$\gamma_1 =2, \gamma_2 =4 , \gamma_3 = \pm 2 , \gamma_4 =0$.
More directly, the integrability can be shown by mapping the coupled NLS equation into two
independent (decoupled) NLS equations as follows;
if we substitute
\begin{equation}
\Psi_1 = q_1 + i q_2, \ \ \Psi_2= q_1 - i q_2,
\label{lt}
\end{equation}
in the two independent NLS equations, $i \bar{\partial } {\Psi_k} = \partial ^2 \Psi_k +2 |\Psi_k|^2
\Psi_k ; k=1,2$, we recover Eq. (\ref{nlseqns}) with
$\gamma _1=2, \gamma _2=4, \gamma _3=-2, \gamma _4=0$. Similarly
using the substitution $\Psi_1 = q_1 + q_2, \Psi_2= q_1 - q_2$, we
obtain Eq. (\ref{nlseqns}) with $\gamma _1=2, \gamma _2=4, \gamma _3=2, \gamma _4=0$.
This explains why the linear superposition of two solitons was possible in the previous
section. The decomposition of the coupled NLS equation into two independent NLS equations
implies that the linear combination of solutions according to Eq. (\ref{lt}) becomes a solution
of the coupled NLS equation. Group theoretically, such a decomposition corresponds to
the embedding of symmetric spaces, $(SU(2)/U(1)) \times (SU(2)/U(1)) \subset Sp(2)/U(2)$.
According to the group theoretic construction of the NLS equation using Hermitian symmetric
spaces \cite{helgason}, the above embedding results in two decoupled NLS equations.
It is interesting to see that this decoupling behavior is also reflected in the Painlev\'{e}
analysis. Besides the solution of the leading order equation (\ref{init}) (the case 4 in Sec. 2)
which enables the present coupled NLS equation to pass the Painlev\'{e} test,
for the set of parameters $\gamma_2 = 2 \gamma_1 , \gamma_3 = -\gamma_1 $, we have another set of solutions of
the leading order equation (\ref{init}),
\begin{equation}
U_0 = {-2 T_0 \over T_0^2 + R_0^2},\ \ S_0 ={-2 R_0 \over T_0^2 +
R_0^2}.
\end{equation}
This has resonances at $j=-1, -1, 0, 0, 3, 3, 4, 4$. This solution also passes the test.
Note that all resonances are double poles and each poles are precisely those of the NLS equation.
This suggests that the systems under consideration are indeed two independent NLS systems.
So far, we have restricted to the case $\beta =1$. For $\beta =-1$, our Painlev\'{e} analysis showed
that the only integrable case is the vector NLS equation considered by
Zakharov and Schulmann,
\begin{equation}
i \bar{\partial } \Psi = \partial ^2 \Psi + \Psi \xi \Psi,\ \ \
-i \bar{\partial } \xi = \partial ^2 \xi + \xi \Psi \xi,
\end{equation}
where $\Psi =(\psi_1, \psi_2) $ and $ \xi =(\chi_1, \chi_2)$.
Using the reduction $\xi = \Psi^* A$ with $A= \pmatrix{-1 & 0 \cr 0 & 1 }$
and substituting $q_1 = \psi_1, q_2 =\psi_2 ^*$, one can recover the vector
NLS equation as in Eq. (\ref{nlseqns}) with $\beta =-1, \gamma_1 = -\gamma_2 , \gamma_3 =\gamma_4 =0$.
In a similar vein, we construct a new integrable equation with $\beta = -1$
which resembles the previous decoupling NLS equation with $\beta =1$.
We take
\begin{equation}
M = \pmatrix{ \chi_1 & \chi_2 \cr \chi_2 & -\chi_1 },
~~N = \pmatrix{ -\chi_1^* & \chi_2 ^* \cr \chi_2 ^* & \chi_1 ^*},
\label{zs1}
\end{equation}
and define the coupled NLS equation by
\begin{eqnarray}
i \bar{\partial } M &=& \partial ^2 M -2MNM, \nonumber \\
-i \bar{\partial } N &=& \partial ^2 N - 2NMN.
\label{zs2}
\end{eqnarray}
We find that Eq. (\ref{zs2}) arises from the Lax pair ($[L_z, L_{\bar z}]=0$)
\begin{eqnarray}
L_z &=& \partial + \pmatrix{0 & M \cr N & 0} +i {\lambda \over 2}
\pmatrix{I_{2 \times 2} & 0 \cr 0 & -I_{2 \times 2}}, \nonumber \\
L_{\bar z} &=& \bar{\partial } -i \pmatrix{0 & \partial M \cr -\partial N & 0} -i \pmatrix{MN & 0 \cr 0 & -NM}
-\lambda \pmatrix{0 & M \cr N & 0} -{i \over 2} \lambda^2
\pmatrix{I_{2 \times 2} & 0 \cr 0 & -I_{2 \times 2}}.
\end{eqnarray}
If we substitute $q_1 = \chi_1, q_2 = \chi_2 ^*$, we have an integrable
equation with anomalous dispersion term and asymmetric coupling,
\begin{eqnarray}
i \bar{\partial } q_1 &=& \partial ^2 q_1 + 2( |q_1|^2 q_1 -2 |q_2|^2 q_1 -q_2^{*2} q_1^*), \nonumber \\
i \bar{\partial } q_2 &=& -\partial ^2 q_2 + 2( |q_2|^2 q_2 -2 |q_1|^2 q_2 -q_1^{*2} q_2^*).
\end{eqnarray}
This equation does not belong to the coupled NLS equation in Eq. (\ref{nlseqns})
which has been Painlev\'{e} tested.
\section{Discussion}
In this paper, we have performed a Painlev\'{e} analysis for coupled NLS
equations with coherent coupling terms as given in Eq. (\ref{nlseqns}).
Besides the well-known vector NLS equation ($\beta=\pm 1; \gamma_2 =\pm \gamma_1 , \gamma_3 = \gamma_4 =0)$,
we have found new integrable cases which are defined by the set of parameters
with $\beta =1$; (i)~ $\gamma_2 = 2 \gamma_1 , \gamma_3 = -\gamma_1 , \gamma_4 $ arbitrary ,
or (ii) ~ $\gamma_2 = 2 \gamma_1 , \gamma_3 = \gamma_1 , \gamma_4 =0$. Painlev\'{e} analysis
shows that these are the only integrable cases except the vector
NLS equation. We have shown that these new equations are essentially identical
to two independent sets of NLS equations. Physically, the first case describes the propagation of optical pulses in
an isotropic nonlinear medium in which the third-order susceptibility tensor satisfies that
$\chi^{(3)}_{xxyy}+\chi^{(3)}_{xyxy}=-2\chi^{(3)}_{xyyx}$, while the second case
does not have a similar interpretation.
The linear transformation in Eq. (\ref{lt}), which decouples the interacting
NLS equation (case (i)) into two independent NLS equations, also maps two
orthogonal linearly polarized lights into the left and the right circularly polarized lights.
Thus, in such an isotropic medium, left and right circularly polarized lights
do not interact each other thereby preserving circular polarizations.
This case may be compared with a polarization preserving fiber where only one
particular polarization direction is preserved.
It would be interesting to know whether there exists nonlinear isotropic
materials possessing this property.
\vglue .3in
{\bf ACKNOWLEDGEMENT}
\vglue .2in
This work was supported in part by Korea Science and Engineering Foundation
(KOSEF) 971-0201-004-2/97-07-02-02-01-3, and by the program of Basic Science Research, Ministry
of Education BSRI-97-2442, and by KOSEF through CTP/SNU.
\vglue .3in
|
\section{Acknowledgments}
This work is supported by the U.S. Department of Energy
under contracts No.~DE-FG05-86ER40273 (SC), No.~W-7405-ENC-36 (PRP)
and the Florida State University Supercomputer Computations Research
Institute which is partially funded by the Department of Energy through
contract No.~DE-FC05-85ER25000.
|
\section{Introduction}
Spiral galaxies contain a widespread layer of ionized Hydrogen, known
as diffuse ionized gas (DIG, also called WIM for warm ionized
medium). The properties of this component of the interstellar medium
(ISM) are important for many aspects of galactic research, such as the
influence of massive stars on the ISM, the porosity of the ISM, and
the disk-halo connection. The Reynolds layer, as DIG in the Milky Way
is known, has a filling factor of at least 0.2, and accounts for
nearly all of the mass of ionized gas, equal to about 30\% of the HI
mass. DIG is less dense than HII regions ($n_e$ $\sim$ 0.2 cm$^{-3}$,
compared to 10$^2$ to 10$^4$ cm$^{-3}$ in HII regions) but has a
similar temperature ( $\sim$ 8000K). An excellent review of the
properties of the Reynolds layer can be found in Reynolds
(\markcite{re90}1990).
The [SII] 6717+6731\AA\space to H$\alpha$~ ratio is higher in the DIG
relative to HII regions, while [OIII] 5007\AA\space to H$\alpha$~ is
lower. In the Reynolds layer and in M31 the [NII] 6548+6584\AA\space
to H$\alpha$~ ratio is about the same as in HII regions
(\markcite{re89}Reynolds 1989; \markcite{gwb97}Greenawalt, Walterbos,
\& Braun 1997), while for NGC 891, the best studied edge-on galaxy, it
appears to be higher in the DIG (\markcite{ra97a}Rand 1997a). These
ratios can result from photoionization by a radiation field with a low
ionization parameter $U$, the ratio of the photon density to the gas
density (\markcite{m86}Mathis 1986; \markcite{dm94}Domg\"orgen \&
Mathis 1994). However, transporting ionizing photons from HII regions
to the DIG is still a problem, because of the long pathlengths that
ionizing photons have to travel. In order to remain ionized, the
Reynolds layer requires at least 15\% of the Lyman continuum photons
from OB stars in the Galaxy, equal to the amount of energy produced in
supernovae. Studies of external galaxies suggest the requirement is
even higher, consistently 30 to 50\% (\markcite{wb94}Walterbos \&
Braun 1994;\markcite{hwg96} Hoopes, Walterbos, \& Greenawalt
1996;\markcite{fwgh96} Ferguson {\it{et al.}}~ 1996; \markcite{gwth98}
Greenawalt {\it{et al.}}~ 1998). Whether this energy is leaking out of
density-bounded HII regions, or can be provided by field OB stars is
still an open question. Recent spectroscopic observations have
challenged the photoionization models, including lower HeI
5876\AA\space than predicted (\markcite{rt95}Reynolds \& Tufte
1995,\markcite{ra97b} Rand 1997b) indicating a softer ionizing
spectrum than expected, higher [OIII] 5007\AA\space in some galaxies
than in others (\markcite{whl97}Wang, Heckman, \& Lehnert 1997;
\markcite{m97}Martin 1997) and rising [OIII]/H$\alpha$~ ratio with height in
the halo of NGC 891 (\markcite{ra98}Rand 1998). Shock ionization by
supernovae is another source which likely plays a role at some level
and may explain the anomalous [OIII] ratios, but it cannot provide
enough energy to ionize the bulk of the DIG.
The Reynolds layer is vertically extended, with a scale height of 900
pc (\markcite{re90}Reynolds 1990). This distribution is quite
different from that of OB stars, which have a scale height closer
to 100 pc. H$\alpha$~ imaging of external edge-on galaxies has revealed the
presence of DIG with varying properties. The first galaxy to be
studied was NGC 891, which has a very thick DIG layer that has been
traced as far as z=3.5 kpc in imaging (\markcite{rkh90}Rand, Kulkarni,
\& Hester 1990, hereafter RKH; \markcite{d90}Dettmar
1990; \markcite{pbs94}Pildis, Bregman, \& Schombert 1994), and z=5 kpc
with spectroscopy (\markcite{ra97b}Rand 1997b). Several other
galaxies possess smaller DIG layers, and others show very little
extraplanar emission (\markcite{rkh92}Rand, Kulkarni, \& Hester 1992;
\markcite{ra96}Rand 1996; \markcite{w91}Walterbos 1991). Filamentary
structure is visible in the extraplanar emission of several galaxies,
indicating an active disk-halo connection and illustrating a probable
relation between DIG and star formation.
Here we present deep H$\alpha$~ images of 5 edge-on galaxies. Some properties
of these galaxies are given in Table 1. We attempt to address several
questions with these images. First we would like to know the extent of
the DIG layers in galaxies down to levels fainter than previously
studied. The DIG in NGC 891 may extend even farther into the halo at
very low surface brightness, and galaxies that do not show a bright
DIG layer may in fact have faint extraplanar emission. This is
important for understanding the structure of galaxies and their
gaseous halos. Second, the correlation of DIG properties with other
parameters of galaxies, such as star-formation rate or Hubble type,
can be a clue toward the ionization source of the DIG, as well as the
mechanism for cycling gas into the halo. The strengths of other
emission lines can also provide information on the ionization source
through comparison with models ({\it{i.e.}}~ \markcite{m86}Mathis 1986;
\markcite{dm94}Domg\"orgen \& Mathis 1994). For this we obtained [SII]
and [OIII] images NGC 4631.
Another goal of this project is to investigate the possibility of
detecting H$\alpha$~ emission from the outer disks of galaxies. Such emission
is expected to occur when neutral Hydrogen in the outer disk and halo
is ionized by the metagalactic radiation field (\markcite{ss76}Silk \&
Sunyaev 1976). The observation of a sharp cutoff in the HI disks of
several galaxies (\markcite{vg91}NGC 3198 by Van Gorkom 1991;
\markcite{css89}M33 by Corbelli, Schneider, \& Salpeter 1989) provides
indirect evidence that the outer disk gas may be ionized. Maloney
(\markcite{m93}1993), Dove \& Shull
(\markcite{ds94}1994) and Corbelli \& Salpeter (\markcite{cs93}1993)
have modeled the situation, and they predict the emission to be very
faint, from 0.2 pc cm$^{-6}$\space in emission measure down to 0.05
pc cm$^{-6}$\space (\markcite{m93}Maloney 1993). Fabry-Perot observations have
pushed the observational limits close to the theoretical predictions
(Bland-Hawthorn, Freeman, \& Quinn 1997, Vogel {\it{et al.}}~ 1995). We
obtained very deep images of two edge-ons, NGC 3003 and UGC 9242, for
the purpose of searching for emission from the outer disk. Direct
detection of ionized gas would allow the determination of the strength
of the metagalactic radiation field at wavelengths below 912 \AA, and
would have important implications for cosmological models, for
prospects of measuring rotation curves of galaxies at large radii, and
for our knowledge of the structure of spiral galaxies.
The layout of this paper is as follows. In section 2 we detail the
observations and data reduction techniques. In section 3 we describe
the H$\alpha$~ morphology of the galaxies in our sample. In section 4 we
examine the DIG in these galaxies, including the vertical extent, and
contribution to the total H$\alpha$~ luminosity. In section 5 we discuss the
[OIII] and [SII] images of NGC 4631. In section 6 we discuss the outer
disk emission in NGC 3003 and UGC 9242. Section 7 contains a
discussion of the implications of our results.
\section{The Data}
\subsection{Observations and Data Reduction}
The images discussed in this paper were obtained during several
different observing runs at KPNO. A log of the observations is given
in table 2. For all of the datasets we removed the bias level and bias
structure using standard methods in IRAF. Twilight flatfields were
combined into a super flatfield which was used to remove gain
variations. In some cases (noted below) the galaxies were observed
using the shift and stare technique, where the telescope is moved
between exposures to place the galaxy on different regions of the
chip. This allows the use of the image frames to produce a night sky
flatfield, if the galaxy is small enough in the field of view of the
telescope. We constructed a night sky flatfield by median combining
the galaxy images after editing out the galaxy and foreground
stars. The resulting image was heavily smoothed and then applied to
the original images after flattening with twilight flatfields.
NGC 891 was observed with the Burrell-Schmidt telescope at KPNO in
both H$\alpha$~ and narrow-band continuum filters. Four pointings were
observed and combined into a mosaic. The large field of view of the
Schmidt (1$^{\circ}$.15) allowed us to make a night sky flatfield even
though NGC 891 is a relatively large galaxy (12$^{\prime}$). The final
flatfielding accuracy is about 1\% across the images. NGC 891 was also
observed with the 0.9 meter at KPNO, in H$\alpha$~ and narrow-band
continuum. Two pointings (out of 4 planned) were observed. NGC 891 is
too big in the 23$^{\prime}$ field of view of the 0.9 meter to use the
images to make a night sky flatfield, but five blank sky images were
taken through the H$\alpha$~ filter, with the telescope moved slightly after
each exposure. These were combined to make a night sky flatfield which
was then applied to the H$\alpha$~ images. The continuum image was already
flat to better than 1\% without the use of a night sky flatfield.
We obtained H$\alpha$~ and narrow-band continuum images of NGC 3003 and UGC
9242 using the 0.9m telescope at KPNO. All of the images used were
taken during photometric conditions. The final mosaics consist of nine
pointings for each galaxy. A few of the images of UGC 9242 had to be
discarded due to large diffraction spikes from a bright star off of
the image. The 23$^{\prime}$ field of view is large enough compared to
these galaxies that the object images could be used to make a night
sky flatfield, which provided a flatfielding accuracy for NGC 3003 of
about 0.5\%, and probably much better than this in the region the
galaxy covers. The night-sky flat alone did not produce satisfactory
results for the images of UGC 9242, but a second night sky flatfield
made from observations of blank sky worked well on the remaining
structure. The flatfielding uncertainty after this correction was
about 0.5\%, and as with NGC 3003 it is better over smaller scales.
NGC 4244 was observed at the KPNO 0.9 meter, during non-photometric
conditions. Three 30 minute exposures in the same position were
combined to produce the final image. NGC 4631 was also observed at
the KPNO 0.9 meter in non-photometric conditions. A different detector
was used, which provided a 6.6$^{\prime}$ field of view. Three
pointings were required to cover the 14.3$^{\prime}$ length of NGC
4631, and these were combined into a mosaic during the reduction. The
small field of view and necessity of mosaicing make it difficult to
match background levels, hence flatfielding uncertainties may be
larger than for the other targets. Each pointing consists of
approximately 1.4 hours, with two pointings overlapping in the central
field (roughly 7 arcminutes). Images of the central field were also
obtained in [OIII] and [SII] filters.
The 0.9 meter observations of NGC 891, NGC 3003, and UGC 9242 were
calibrated using observations of spectrophotometric standard
stars. The Schmidt observation of NGC 891 and the 0.9 meter
observations of NGC 4244 and NGC 4631 were calibrated using the
R-magnitude of the galaxy. This was scaled by the response and
transmission of the continuum filter and used to calibrate the
continuum image. Then the stars in the line image were scaled to the
continuum, taking into account the differences in the response of the
filters. The [OIII] image of NGC 4631 was calibrated in a similar
manner using the V magnitude of the galaxy. For convenience, all of
the H$\alpha$~ images were converted to emission measure as if no [NII] was
transmitted by the filter. The wide filter used on NGC 4631 contains
the [NII] lines, but the narrow filters used for the other galaxies
transmit very little [NII]. The [NII]/H$\alpha$~ ratio has been observed to
vary from 0.6 to 1.1 in the DIG of NGC 891 (\markcite{ra97a}Rand
1997a). At the highest value, [NII] transmitted by the narrow filters
contributes only 10$-$20\% of the observed H$\alpha$~ flux. For NGC 4631 we
correct for [NII] where it is relevant.
\subsection{Continuum Subtraction}
One of the most uncertain procedures in analyzing emission line
images is the removal of the underlying stellar continuum.
Subtracting too little continuum will leave a faint background that
can be interpreted as diffuse line emission, while subtracting too
much continuum can remove a layer of real diffuse emission. These
concerns can be minimized by observing the continuum close in
wavelength to the line, but varying and unknown H$\alpha$~ absorption
produced by the stars in the galaxy makes a perfect subtraction
difficult to achieve.
As a first determination of the continuum scale factor, we measured
the fluxes of foreground stars in both images, and computed the factor
needed to make them equal (for galaxies that were calibrated using the
R magnitude, the fluxes are made equal during calibration). This
relies on the foreground stars being of similar spectral type as the
stars in the galaxy, which is not necessarily the case. We then
visually inspected the images to be sure that there are no negative
regions, and that there is no obvious component of the stellar
continuum left in the image. For all of the galaxies the scale factor
derived from foreground stars was satisfactory, except the [OIII]
image of NGC 4631. For this image we adjusted the scale factor to
achieve the best subtraction. To quantify the dependence of our
analysis on the continuum subtraction, we vary the scale factor by
$\pm$ 3\%. At this level it is usually obvious that the continuum is
incorrectly subtracted. Our results throughout this paper will include
the variation in the continuum subtraction in the total uncertainty.
\subsection{Scattered Light}
Another potential problem with analyzing faint emission is the
possibility of scattered light. Telescope optics produce halos around
point sources, which could be mistaken for diffuse emission. In
addition, dust within the galaxy itself may scatter light from HII
regions, producing a halo of scattered light around HII
regions. Walterbos \& Braun (\markcite{wb94}1994) concluded that
scattered light from within the galaxy is not a major component, based
on the spectrum of the DIG. The [SII] to H$\alpha$~ ratio is high in the DIG
compared to HII regions, which would not be expected if the DIG were
actually scattered light since the lines are very close in
wavelength. Ferrara {\it{et al.}}~ (\markcite{fbdg96}1996) modeled the
scattering of HII region light by dust in the halo of NGC 891, and
found it contributed only 10\% of the DIG emission at 600 pc off the
plane.
Methods of determining the contribution of scattered light from optics
have been explored by Walterbos \& Braun (\markcite{wb94}1994) and
Hoopes {\it{et al.}}~ (\markcite{hwg96}1996), and methods
for correcting for scattered light have been used by
RKH\markcite{rkh90}. Following these methods, we determined the
possible contribution of scattered light from bright foreground stars
in all of our images. We determined the radius in which 95\% of the
energy from the star is encircled, shown in table 2. Scattered light
from HII regions will appear as halos around the regions to this
radius, with only 5\% of the light scattered further. We find that DIG
extends further than this in all 5 galaxies, at a much higher level
than 5\%.
Scattered light can also affect the determination of the scale height
of emission above the plane. To correct for this we deconvolved the
images with the stellar point spread function (PSF). The PSF was
measured using 10 stars in both the line and continuum images (except
NGC 4631, where only 5 stars were suitable). The images were then
deconvolved with the PSF using the Lucy-Richardson deconvolution
algorithm with the task LUCY in IRAF. This was done before calibration
and continuum subtraction. Vertical profile fitting was done on the
deconvolved images (see section 4.1).
\section{Results for Individual Galaxies}
The H$\alpha$~ images are shown in figures 1$-$5. The H$\alpha$~ luminosities of
the galaxies are listed in table 3. The correction for Galactic
extinction (\markcite{bh84}Burstein \& Heiles 1984) is negligible for
all of the galaxies except NGC891. The value listed in the table has
been corrected; the luminosity before correcting for extinction is 2.8
$\times$10$^{40}$ erg s$^{-1}$. No correction for internal extinction
was made. The uncertainty given in the table is due to varying the
continuum subtraction by $\pm$ 3\%. Our H$\alpha$~ luminosity for NGC 891
agrees well with Rand {\it{et al.}}~ (\markcite{rkh92}1992). Our flux for NGC
4631 is about 6\% higher than that given in Rand {\it{et al.}}~
(\markcite{rkh92}1992), but our image contains [NII] emission as well,
so our H$\alpha$~ flux is actually lower. There are no published luminosities
of NGC 3003, NGC 4244, and UGC 9242 with which to compare. NGC 4631
and NGC 4244 were observed during non-photometric conditions.
In this section we discuss the general appearance of the H$\alpha$~ images.
The H$\alpha$~ morphology of several of the galaxies in our sample have been
discussed in great detail elsewhere, so we will keep our description
brief. These include NGC 891 (\markcite{rkh90}RKH;\markcite{d90}
Dettmar 1990;\markcite{pbs94} Pildis, Bregman, \& Schombert 1994) and
NGC 4631 (\markcite{rkh92}Rand {\it{et al.}}~ 1992;
\markcite{gdd96}Golla, Dettmar, \& Domg\"orgen 1996). NGC 4244 was
discussed briefly by Walterbos (\markcite{w91}1991) and Walterbos \&
Braun (\markcite{wb96}1996). NGC 3003 and UGC 9242 have not previously
been imaged in H$\alpha$.
\subsection{NGC 891}
In figure 1 we show both the Schmidt and the 0.9 meter continuum
subtracted H$\alpha$~ images of NGC 891. Vertically extended emission is
clearly present. The DIG layer can be traced to more than 3 kpc away
from the plane in these figures. The emission seems fairly uniform,
but does appear brighter and more extended near two large HII regions
on the east side of the disk. In the higher resolution 0.9 meter image
the extraplanar emission begins to show filamentary structure,
although it is not as pronounced as in NGC 4631 (see below). There is
a correlation with the brightest star forming regions, with the
brightest DIG near bright HII regions in the inner disk. Rand
(\markcite{ra97a}1997a) noted that the bright filaments almost always
connect to an HII region in the disk.
\subsection{NGC 3003}
The morphology of this galaxy (figure 2) suggests that it may be
disturbed, in that the spiral arms appear asymmetric. There is a small
dwarf galaxy (or possibly a background galaxy) at the lower right edge
of figure 2, but other than this there are no close companions visible
in our images, which cover an area of 300 $\times$ 300 kpc at the
distance of NGC 3003. However, just outside of our field of view is
the small spiral NGC 3021. These two galaxies are separated by about
200 kpc in projected distance, and about 50 km s$^{-1}$ in velocity
(Tully 1988). The total H$\alpha$~ luminosity of 1.18 $\times$ 10$^{41}$ erg
s$^{-1}$ indicates that active star formation is ongoing. This H$\alpha$~
luminosity is about an order of magnitude higher than for the
starbursts NGC 253 and M82. In fact, of the galaxies observed by Young
{\it{et al.}}~ (\markcite{y96}1996), one of the largest sample of galaxies
observed in H$\alpha$~ in the literature, only six of the 120 spirals have
higher observed H$\alpha$~ luminosities than NGC 3003 (not corrected for
extinction). The H$\alpha$~ image shows a bright nucleus, and many bright HII
regions, including a very bright region in the outer western part of
the disk. A projected spiral arm protrudes below the galaxy in the
image, with a bright HII region at the end. The H$\alpha$~ emitting disk is
about 35 kpc across. The galaxy is not quite edge-on, so it is
difficult to make any statements about the vertical extent of the DIG,
though there is pervasive diffuse emission in the disk.
\subsection{NGC 4244}
This is a nearby galaxy (3.1 Mpc), so we can resolve a great deal of
detail in the ionized disk. The HII regions distribution is extended
in the vertical direction, a result perhaps of a weaker disk potential
(\markcite{o96}Olling 1996). There are a few HII regions as far as 700
pc above the midplane. DIG is clearly visible in the disk, spread
between several bright HII regions. It is immediately obvious,
however, that an extensive DIG layer such as in NGC 891 is absent in
this galaxy. Some short filaments exist in the central regions of the
galaxy, but overall the DIG layer is confined to the disk. The galaxy
appears quiescent in H$\alpha$, and the low H$\alpha$~ luminosity and low FIR
surface brightness imply that there is relatively little star
formation occurring. Most of the HII regions are small and faint,
except for two bright complexes on either end of the disk. The
absence of long bright filaments, such as those seen in NGC 891
and NGC 4631, imply that the galaxy also lacks any visible disk-halo
interaction.
\subsection{NGC 4631}
The H$\alpha$~+[NII] image of this galaxy is shown in figure 4. An overlay of
H$\alpha$~ and x-ray emission for this galaxy, based on our data, can be
found in Wang {\it{et al.}}~ (\markcite{wwsnb95}1995). There appear to be
significant flatfielding uncertainties on the south side of the disk,
near the bottom edge of the image, so we restrict our analysis to the
north side. The image shows a disturbed disk, most likely due to an
interaction with its companion galaxies NGC 4627 and NGC 4656. The
disk appears to be actively forming stars, and the disk-halo interface
is also very active in this galaxy. Rand {\it{et al.}}~ (\markcite{rkh92}1992)
pointed out two bright vertical ``worms'' of emission east of the
nucleus on the north side of the disk. We note that these worms are
connected to longer, fainter filaments, and the western worm appears
to curve back around toward the disk. There is an even larger loop of
emission which surrounds the two worms (indicated by the arrows in
figure 4). It begins just east of the easternmost worm and extends 3.5
kpc into the halo. Unfortunately there is a seam in the image where
the loop might reconnect to the disk. The surface brightness at the
top of the loop is about 12 pc cm$^{-6}$. The large loop is barely visible in
the H$\alpha$~ image presented by Rand {\it{et al.}}~ (\markcite{rkh92}1992, their
figure 5). What appears to be another filament about 3 kpc east of the
giant loop in our image is actually another seam. There is significant
extraplanar emission in the form of discrete features such as the
loops and worms, and a smoother component can be traced up to about 2
kpc on the north side. Donahue, Aldering, \& Stocke
(\markcite{das95}1995) detected a faint halo extending 16 kpc from the
plane of NGC 4631, with a maximum brightness of 0.69 pc cm$^{-6}$\space per
square arcsecond. We could not confirm the detection because of the
smaller field of view of our observations. This emission is probably
related to the high star formation and disturbed nature of this galaxy
resulting from the tidal interaction, as it is too bright to be caused
by the metagalactic ionizing radiation field.
\subsection{UGC 9242}
This galaxy appears to be close to exactly edge on (figure 5). The
continuum image shows a very thin and symmetric disk. The H$\alpha$~ emitting
disk is about 35 kpc across. It has a bright nucleus in H$\alpha$~ and
several bright star forming regions. There are two plumes visible on
the north side of the nucleus. They can be traced about 1.8 kpc from
the midplane, with typical emission measures ranging from 20 just
above the disk to about 5 at the highest point. These are probably
associated with star formation in the nucleus, but it is not
known whether these are two separate outflows, or whether they are the
brightened edges of a conical outflow such as that seen in NGC 253
(\markcite{ham90}Heckman, Armus, \& Miley 1990). If it is a conical
outflow, it is 1200 pc wide at the top. There is also some structure
on the other side of the nucleus, which also may be 2 spurs of
emission, shorter and weaker than the counterparts on the north
side. This suggests a double sided outflow, with a morphology similar
to the ``H'' shaped filaments seen in NGC 3079
(\markcite{hkrd90}Hester {\it{et al.}}~ 1990; \markcite{vcb95}Veilleux, Cecil,
\& Bland-Hawthorn 1995) and NGC 4013 (\markcite{ra96}Rand 1996). The
bright nucleus is about 1 kpc across, with an H$\alpha$~ luminosity in a
12$^{\prime\prime}$ (about 1 kpc) diameter aperture of 4.2 $\times$ 10$^{39}$ erg
s$^{-1}$, comparable to the brightest HII regions in most galaxies
(\markcite{k88}Kennicutt 1988), and in fact equal to the total H$\alpha$~
luminosity of NGC 4244. Except for these filaments there is very
little obvious extraplanar emission, and there is no evidence for
disk-halo interaction beyond the nucleus, but like all of the galaxies
in our sample the disk is filled with DIG.
\section{The Diffuse Ionized Gas}
\subsection{Vertical Extent}
Walterbos \& Braun (\markcite{wb96}1996) compared the appearance of
the DIG layers in three of the galaxies in our sample, NGC 891, NGC
4244, and NGC 4631. Figure 6 is a more exact comparison, incorporating
UGC 9242. The images are shown on the same spatial scale, brightness
scale, and logarithmic stretch. The comparison shows the range of DIG
morphologies. The most prominent example is the smooth, bright,
extended layer in NGC 891. The patchy, filamentary layer in NGC 4631
may be related to recently enhanced star formation as a result of an
encounter. The most common appearance of the DIG may be more similar
to the weaker layers in NGC 4244 and UGC 9242. The inclination of NGC
3003 hinders our analysis of the extraplanar emission.
Following \markcite{rkh90}RKH and Rand (\markcite{ra96}1996), we have
attempted to characterize the DIG as an exponential layer. We fit the
vertical profile averaged over the central 10 kpc to increase the
signal to noise. In order to avoid including light from HII regions in
the fit, we excluded emission from $|z| \le$ 300 pc. NGC 4244 has a
thicker disk of HII regions, so we excluded emission from $|z| \le$
500 pc. We tried fits using a single exponential function as well as
fits using two exponentials. The parameters of the best fits for three
of the galaxies are given in table 4. The vertical profile of NGC 3003
suggests that it may also possess faint extraplanar emission, but the
lower inclination of this galaxy make the detection uncertain, as such
emission may arise from the outer part of the disk. We do not address it
further here, nor do we attempt the analysis for NGC 4631 due to its
disturbed nature and the poor flatfielding of the image. Note that the
scale heights given are for the surface brightness (emission
measure). The electron scale height is twice the emission measure
scale height, assuming the emission is only H$\alpha$. If much [NII]
is included in the filter, a rising [NII]/H$\alpha$~ ratio in the halo such
as those observed in NGC 891 (\markcite{ra98}Rand 1998) and NGC 4631
(\markcite{gdd96}Golla, Dettmar, \& Domg\"orgen 1996) could make the
scale heights appear larger than they really are. The electron scale
height for the Reynolds layer is 900 pc (\markcite{re90}Reynolds
1990).
Previous imaging of NGC 891 (\markcite{rkh90}RKH) has traced the
extraplanar DIG to as far as 3.5 kpc from the plane of the
galaxy. Spectroscopy (\markcite{ra97b}Rand 1997b) has traced the
emission even further, to at least 5 kpc. Figure 7 shows a vertical
profile of the deconvolved Schmidt H$\alpha$~ image of NGC 891, averaged over
the central 10 kpc. The emission can be traced as far as 5 kpc off the
plane in both directions. Thus we confirm the spectroscopic
detection. There is a suggestion that the emission continues even
further, to as far as 7 kpc, on the west side of the disk. However,
foreground stars in this part of the image make this
uncertain. Although the two fits look quite similar in the linear
scaling, the logarithmically scaled plot shows that two exponential
functions better fit the faint component of the profile. For
comparison, the stellar thin disk of NGC 891 has a scale height of 425
pc, and the stellar thick disk has a scale height of about 1.9 kpc
from surface photometry (\markcite{m99}Morrison 1999;
\markcite{mmhsb97}Morrison {\it{et al.}}~ 1997).
Even averaged over 10 kpc the extraplanar emission in NGC 4244 is very
weak (figure 8). Using two exponential functions, we find one
component with a very small scale height, about a factor of 2$-$3
lower than that of the Galaxy, and a fainter, broader component. The
faint component may be due to a flatfielding problem, such as low
level vignetting in the image. If we fit only one exponential the
scale heights are closer to that of the Galaxy (450 pc). The
two-exponential fit is statistically better, but both functions
describe the emission fairly well. NGC 4244 does not appear to possess
a thick stellar disk (\markcite{m99}Morrison 1999).
Figure 9 shows that the extraplanar emission in UGC 9242 cannot be
described by a single exponential layer. Using two exponential
functions, the bright component of the vertical profile is well
described by a relatively low scale-height exponential, as expected
from the appearance of the H$\alpha$~ image. However, there is a faint tail
of emission which is clearly visible out to 3$-$4 kpc. The scale
height of this component is similar to and perhaps even larger than
the scale height of the Reynolds layer. As noted, UGC 9242 has bright
filaments expending from the nucleus which may hamper the fit. In
figure 10 we show the vertical profile of the central 20 kpc with the
central 3 kpc excluded so as to remove the contribution of these
filaments. The parameters of the model fits are shown in table 4. A
single exponential still cannot fit the emission. The scale heights
for the bright component are similar to the 10 kpc fit, but the faint
components are even more extended. The high-z tail appears to extend
well past 5 kpc.
In figure 11 we show the profiles of NGC 891, UGC 9242, and NGC 4244
overplotted with the best fit model for each. Although NGC 891 has a
much brighter DIG layer than UGC 9242, extraplanar emission in both
galaxies actually reaches comparable distances above the disk. NGC
4244 clearly has no extraplanar emission comparable to the other two
galaxies. The logarithmically scaled plots clearly show that two
exponential components make up the extraplanar DIG in NGC 891 and UGC
9242, and possibly also in NGC 4244.
Table 5 shows some parameters derived from the model fits. It is
necessary to know the diameter of the DIG cylinder to derive these
properties, so we estimate it from the H$\alpha$~ images. The emission
measure in a column perpendicular to the disk is given for each of the
two components. We call the lower scale height component ``thick
disk'' DIG (so as not to be confused with the HII region thin disk),
and the higher scale height component ``halo'' DIG. The two components
contribute nearly equally to the total perpendicular emission measure
in NGC 891, while the halo components is weaker in the other two
galaxies. The Galaxy resembles UGC 9242 in terms of total
perpendicular emission measure and surface density, but keep in mind
that the Galactic DIG parameters are derived for the solar
neighborhood, while for the external galaxies these DIG parameters
apply closer to the center where it is brighter. Following
RKH\markcite{rkh90}, a constant filling factor of $\phi$=0.25 is
assumed when calculating the surface density, although it may rise in
the halo (\markcite{kh1988}Kulkarni \& Heiles 1988). The surface
densities include helium at solar abundance.
\subsection{Diffuse Fractions}
Another way to compare the DIG in different galaxies is through the
diffuse fraction, which is the contribution of the DIG to the total
H$\alpha$~ luminosity. We us the same method used by Hoopes {\it{et al.}}~
(\markcite{hwg96}1996) for isolating diffuse emission from
HII region emission. This technique compensates for the varying
brightness of the DIG layer with radius in the galaxy and galaxy
inclination (the reasons a simple isophotal cut at a given surface
brightness fails). The method involves subtracting smoothed version of
the image from the original to remove the diffuse component, then
making a mask on the resulting image which removes pixels greater than
a certain value and replaces them with zero. The mask is applied to
the original image to remove HII regions emission. The technique was
tested by applying it the M31 images of Walterbos \& Braun
\markcite{wb92}(1992, \markcite{wb94}1994). We chose a scale of
900 pc for the median box used to smooth the image, and a cut level of
50 pc cm$^{-6}$\space to make the mask, because these parameters resulted in
diffuse fractions similar to those that were found by manually
cataloging and removing the HII regions by hand in M31. Note that the
cut level of 50 pc cm$^{-6}$\space in the image after subtracting the
smoothed version does not correspond to the same level in the original
image.
We applied this technique to the galaxies in our sample. The measured
diffuse fractions are given in table 3. Our H$\alpha$~ flux NGC 4631 is low
compared to Rand {\it{et al.}}~ (\markcite{rkh92}1992), so to
test the effects of a possible calibration error we multiplied the
image by 1.3 and recalculated the diffuse fraction using the same
method, which gave 37 $-$ 41\%, very similar to the original
value. The diffuse fractions for NGC 3003, NGC 4244, NGC 4631, and UGC
9242 all fall within the 30 to 50\% range found for face on galaxies
(\markcite{hwg96}Hoopes {\it{et al.}}~ 1996;
\markcite{gwth98}Greenawalt {\it{et al.}}~ 1998). NGC 891, however, shows a
much higher fraction, 83$-$86\%. The diffuse fraction for NGC 891 was
measured on the 0.9 meter image. Applying the technique to the lower
resolution Schmidt image gives a fraction of 90\%. The difference
between the fractions measured on the Schmidt and 0.9 meter images of
NGC 891 is most likely due to unresolved HII regions in the lower
resolution Schmidt image being counted as DIG.
Comparison of the diffuse fractions found in edge-on galaxies to those
found in face-on systems is not straightforward. Since the HII region
layer is confined to the midplane where the dust density is the
highest, one might expect to measure a higher diffuse fraction in edge
on galaxies, as the HII regions would be affected more by the dust
disk than they would in face-on galaxies. The DIG extends well out of
the dust disk, especially in NGC 891, so much of it is less
obscured. This may explain the high ratio in NGC 891. As the only Sb
galaxy in our sample, it may have a higher metal content than the
later type spirals, and the higher abundance of dust may obscure more
of the disk HII region emission. A comparison of the H$\alpha$~ surface
brightness with the FIR surface brightness supports the idea that NGC
891 contains more dust than the other galaxies in our sample. The FIR
luminosity is an upper limit to the star formation rate, as an unknown
fraction arises from dust heated by older stars. The H$\alpha$~ luminosity
gives a lower limit to the star formation rate, since it is affected
by extinction. Table 3 shows that the L$_{H\alpha}$/L$_{FIR}$ ratio
for NGC 891 is low relative to the other galaxies in the sample,
implying higher extinction. Interestingly, NGC 4244 has a high value
for this ratio, implying that it may be dust poor.
\section{Line Ratios in NGC 4631}
Figure 12 shows a subsection of the H$\alpha$~ image of NGC 4631 and the
[OIII]/H$\alpha$~ and [SII]/H$\alpha$~ ratio images. The figure shows a portion of
the disk east of the bulge of the galaxy. In the [OIII]/H$\alpha$~ image the
cores of HII regions show a high ratio, while the DIG shows a lower
ratio. Not all HII regions have high [OIII]/H$\alpha$~ ratios, however, and
in those that do only the central core has an elevated ratio, except
for one region described below. The [SII]/H$\alpha$~ image shows the opposite
trend, with the DIG showing a higher [SII]/H$\alpha$~ ratio than HII regions.
Applying the mask made on the H$\alpha$~ image of NGC 4631 to the [OIII] and
[SII] images allows us to investigate the line strengths in the DIG
and in HII regions. The continuum in the bulge did not subtract well
from the [OIII] image, so we omitted that region and determined the
ratio in the disk east and west of the bulge. The global [SII]/H$\alpha$~
and [OIII]/H$\alpha$~ ratios are given in table 6. The [SII]/(H$\alpha$~+[NII])
ratio is elevated in the DIG, while [OIII]/(H$\alpha$~+[NII]) is lower. Note
that the H$\alpha$~ filter also contains a contribution from the nearby [NII]
lines, which must be taken into account when comparing with other
measurements. In the disk [NII]/H$\alpha$~ varies from 0.1 to 0.2 in HII
regions, and 0.3 to 0.5 in the DIG (\markcite{gdd96}Golla, Dettmar, \&
Domg\"orgen 1996). Thus when comparing to measurements made without
[NII], the observed [OIII]/H$\alpha$~ and [SII]/H$\alpha$~ ratios for NGC 4631 could
be 1.1 to 1.2 times higher in HII regions, and 1.3 to 1.5 times higher
in the DIG. Another important point is that the images are not
corrected for internal extinction. Greenawalt {\it{et al.}}~
(\markcite{gwb97}1997) found that the extinction in HII regions is
higher than that in the DIG in M31. If this is also true in NGC 4631
then correcting for extinction would reduce the [OIII]/H$\alpha$~ ratio in
HII regions more than in the DIG. The large uncertainties at high z
distance, due to low signal in the [OIII] and [SII] images, prevent us
from investigating the behavior of the line ratios away from the
plane.
There is an extended region on the SE side of the disk where the
[OIII]/H$\alpha$~ ratio reaches over 1.0 (near the bottom of the images in
figure 12). This is comparable to the ratios seen in the cores of HII
regions, but the high [OIII] gas appears more extended than other HII
regions, and lies in a region where the H$\alpha$~ emission is about 180
pc cm$^{-6}$. By contrast, the cores of HII regions which show high [OIII]
have emission measures of about 1500 pc cm$^{-6}$. This is the same location
in the galaxy where Rand \& van der Hulst (\markcite{rvdh93}1993)
found a large HI supershell. The shell has been modeled as a collision
of a high velocity cloud with the disk of NGC 4631
(\markcite{rs96}Rand \& Stone 1996). The region of high [OIII]
emission is about 650 pc in diameter, but borders on a large HII
region which also shows high [OIII]/H$\alpha$~, so the true extent is
difficult to measure. The ratios in a 370 pc wide vertical slice
through the region are shown in figure 13. The location of this slice
is marked in figure 12. The [SII]/H$\alpha$~ ratio in this region is about
0.2, close to the value in HII regions. The gas here may be shock
ionized, which can produce a high [OIII]/H$\alpha$~ ratio
(\markcite{sm79}Shull \& McKee 1979;
\markcite{ds95}Dopita \& Sutherland 1995). Shock-ionized gas can also
have low [SII]/H$\alpha$~ if the density is high enough to
collisionally de-excite S$^+$ (\markcite{ds95}Dopita \& Sutherland
(1995). This region was not included in the determination of the
global [OIII]/H$\alpha$~ and [SII]/H$\alpha$~ ratios.
\section{Limits on Emission from the Outer Disk}
One of the goals of this project was to detect or set limits on H$\alpha$~
emission from the outer disk. These observations were driven by the
recent discovery of sharp edges to the HI disks in nearby galaxies
such as NGC 3198 (\markcite{vg91}van Gorkom 1991). A possible
explanation for these edges is that the outer Hydrogen is ionized,
making it undetectable to 21 cm observations, as suggested by Silk \&
Sunyaev (\markcite{ss76}1976). There are no ionizing stars at these large
radii, and since the HI typically extends several kpc past the optical
or H$\alpha$~ disk any ionizing radiation from within the galaxy would surely
be absorbed by the intervening HI. Therefore the ionizing source must
be extragalactic, and is thought to be the metagalactic ionizing
radiation field produced by quasars and AGN.
This idea has been modeled (\markcite{m93}Maloney 1993;\markcite{cs93}
Corbelli \& Salpeter 1993;\markcite{ds94} Dove \& Shull 1994) and the
observed cutoff can be reproduced. A crucial test, however, is to
directly detect the ionized outer disk. The models indicate that the
emission would be extremely faint, as low 0.05 pc cm$^{-6}$, but possibly as
high as 0.2 pc cm$^{-6}$, depending on the gas density, clumpiness, and the
strength of the ionizing radiation field. This is fainter than imaging
studies have reached in the past. Ionized gas was detected in the
outer disk of NGC 253 (\markcite{bfq97}Bland-Hawthorn {\it{et al.}}~ 1997) at a
level of 0.23 pc cm$^{-6}$, using very sensitive Fabry-Perot
observations. The emission would imply a metagalactic radiation field
stronger than the current upper limit of 8$\times$10$^{-23}$ergs
cm$^{-2}$ s$^{-1}$ Hz$^{-1}$ sr$^{-1}$ (Vogel {\it{et al.}}~
\markcite{vwrh95}1995), and the authors argue that the gas is
photoionized by disk OB stars which can see the warped outer disk.
With extreme care in flatfielding, it may be possible to detect this
level of emission through deep narrow-band imaging. Recently Donahue
{\it{et al.}}~ (\markcite{das95}1995) detected a very faint halo around NGC
4631. The surface brightness of this halo is as high as 0.69 pc cm$^{-6}$,
much brighter than that expected from the metagalactic radiation
field, so the emission is most likely a result of the star formation
activity in the galaxy or related to the gravitational interaction
with its neighbors. We used similar imaging techniques in order to
optimize the detection of faint emission. Our target galaxies (NGC
3003 and UGC 9242) were chosen to be at high galactic latitude so that
they were not affected by many bright foreground stars or emission
from the Reynolds layer. The sensitivity of the NGC 3003 image is
about 2.0 pc cm$^{-6}$\space per square arcsecond, and about 2.7 pc cm$^{-6}$\space
per square arcsecond for UGC 9242, much brighter than the strongest
expected emission of 0.25 pc cm$^{-6}$\space found by Maloney (1993). In
order to reach fainter levels, we spatially averaged over an ever
increasing area to lower the noise until the limit of the flatfielding
accuracy was reached and no further increase in S/N with smoothing was
apparent. The NGC 3003 image was binned into 25$\times$25 pixel
(17.3$^{\prime\prime}$$\times$17.3$^{\prime\prime}$) boxes, which increased the sensitivity by
a factor of 17.3 to 0.13 pc cm$^{-6}$. The UGC 9242 could be averaged over
30$\times$30 pixel (20.7$^{\prime\prime}$$\times$20.7$^{\prime\prime}$) regions, increasing
the sensitivity to 0.13 pc cm$^{-6}$. The rms intensity deviation between the
background levels in boxes in flat regions of the image was equal to
or less than this limit, even for boxes separated by large distances.
The spatial scale for these limits are 2 kpc for NGC 3003 and 2.6 kpc
for UGC 9242.
The binned images show no obvious outer disk emission. In figure 14 we
show the major axis profiles of the two galaxies from the binned
images. UGC 9242 has a distinct hole on the west side (positive major
axis distance in figure 14). This appears to be a flatfielding error
at the edge of one of the images which went into the final
mosaic. Aside from this, any outer disk emission is below the
limits of our flatfielding accuracy. We also smoothed using a median
filter, using the same size median box as the average box used
above. The NGC 3003 image shows no evidence for outer disk emission,
but the UGC 9242 image has some interesting features (see figure
15). Faint extraplanar emission is visible on both sides of the
disk. On the east side of the disk there is a bright star, which may
account for the emission in this region. On the west side, however,
the stars are fainter and less likely to contribute much scattered
light. On this side the emission ranges from about 2 pc cm$^{-6}$\space at 2
kpc above the midplane to our limit of 0.13 pc cm$^{-6}$\space at 8.5 pc
above the midplane. It appears that the emission is centered towards
the nucleus, which may imply a connection to the central starburst,
but might also just reflect the distribution of halo gas.
The H$\alpha$~ disk also appears more extended in the median smoothed
image. The H$\alpha$~ emission extends about 8 kpc past the southern edge of
the optical disk (the left side of figure 15). The optical edge is at
about -21 Mpc in figure 14. Several factors lead us to question the
validity of this feature. The emission ranges in brightness from 0.16
to 0.20 pc cm$^{-6}$, just barely above our detection limit, making this a
1$-$2$\sigma$ detection at best, and is not confirmed in the binned
image. There is a background galaxy and some faint foreground stars
which may affect the flatfielding in this region. There are other
variations of similar magnitude near stars in the image. An example of
this can be seen just below the extended disk in figure 15, where
scattered light from a background galaxy and a group of several small
stars created a spot in the smoothed image. Although stars are removed
more cleanly during continuum subtraction in the deconvolved images
than in the original images, there are still residuals due to effects
such as a changing PSF across the field, leading to this further
source of uncertainty for faint emission. If this emission were real,
it would contradict the limits set by the NGC 3003 image, as well as
other established upper limits on the metagalactic ionizing flux (see
section 7.2).
\section{Discussion}
\subsection{Halo Emission}
We have detected the DIG layer in NGC 891 out to at least 5 kpc from
the plane, and possibly as far as 7 kpc. NGC 891 has the brightest and
largest DIG layer known, and it is not surprising to detect emission
so far from the plane in a deep image. What is surprising is that UGC
9242, which shows much less extraplanar emission, still has a halo
extending as far as 3 to 4 kpc, though significantly fainter than the
NGC 891 halo. The image of UGC 9242 is of very high sensitivity, so it
is possible that similar faint halos might be seen around other
edge-ons even if they do not possess a bright DIG layer. We also
confirm the existence of extraplanar emission in NGC 4631, although we
could not have verified the halo claimed for NGC 4631 by Donahue {\it{et al.}}~
(\markcite{das95}1995). For the other galaxies in our sample, even in
our very sensitive images of NGC 3003 and UGC 9242, we do not detect
such a halo.
Excluding NGC 3003, which has too low an inclination to properly study
the extraplanar emission, NGC 4631 and NGC 891 are the brightest in
H$\alpha$, and they also show the most extraplanar emission of the
sample. Although directly inferred star formation rates from the H$\alpha$~
luminosity can be very unreliable for edge-on galaxies due to
extinction, in a relative sense NGC 891 and NGC 4631 would appear to
be the most actively star forming galaxies of the sample (again
excluding NGC 3003). The extreme difference between these galaxies and
NGC 4244, with its low H$\alpha$~ luminosity and weak DIG layer, point to a
link between active star formation and extraplanar emission, as
discussed previously by Rand (\markcite{ra96}1996). This conclusion
is supported by the FIR luminosity (\markcite{ra96}Rand 1996), which
traces star formation more accurately than H$\alpha$~ in edge-on galaxies,
although some fraction of the FIR emission may stem from dust heated
by general starlight, not OB stars. In table 3 we list the FIR
luminosity normalized by the square of the disk diameter (D$_{25}$),
following Rand (\markcite{ra96}1996). The galaxies with the highest
current star formation have the most prominent extraplanar
emission. This does not carry over to the diffuse fractions,
however. In fact the diffuse fractions seem relatively constant with
the exception of NGC 891, and this may be due to higher extinction in
that galaxy. NGC 891 is the only Sb in the sample and may have a
higher dust content than the rest of the sample (all Sc galaxies). The
constancy of the diffuse fraction further reinforces the connection
with star formation, since it essentially means that the DIG
luminosity scales with the HII region luminosity. It is also
interesting that there is not a more pronounced difference between
face-on and edge-on diffuse fractions. This might imply that we are
seeing most of the disk H$\alpha$~ emission even in edge-on galaxies, with
the exception of NGC 891. Diffuse fractions for more edge-on spirals
are necessary to test this idea further.
The vertical profiles of NGC 891, UGC 9242, and possibly NGC 4244 are
best described by two distinct exponential components, raising the
possibility that more than one mechanism is responsible for creating
H$\alpha$~ halos. NGC 891 possesses a thick stellar disk, while NGC 4244 does
not, which may suggest a connection between the mechanism which
creates a thick disk and that which is responsible for the H$\alpha$~
halo. However, a counter-example can be found in NGC 4565, which does
possess a thick stellar disk (\markcite{m99}Morrison 1999), but has
little extraplanar H$\alpha$~ emission (\markcite{rkh92}Rand {\it{et al.}}~ 1992). An
x-ray observation of NGC 891 (\markcite{bp94}Bregman \& Pildis 1994;
\markcite{bh97}Bregman \& Houck 1997) revealed a halo of 10$^6$ K gas,
with a distribution similar to the H$\alpha$~ emission. The idea that hot
supernovae-heated gas vented into the halo through chimneys is
responsible for this emission led the authors to calculate whether the
cooling of this hot gas could be the source of the H$\alpha$~ emission. They
found that the cooled gas mass was an order of magnitude too low to
account for the mass of H$\alpha$~ emitting gas. Rand (\markcite{ra97b}1997b)
suggested that cooling gas may explain the more-extended halo
component of the H$\alpha$~ emission, while gas ionized by photons from OB
stars leaking out of the disk is responsible for the brighter,
less-extended component. The H$\alpha$~ image of the DIG in UGC 9242 revealed
a very extended halo component, but the scale height of the disk
component is much lower than that of the similar component in NGC 891.
Perhaps cooling gas could be responsible for a faint halo such as in
UGC 9242, but some mechanism prevents the gas which makes up the disk
component from reaching as high as in NGC 891. There is indication of
an outflow into the halo from the nuclear filaments on both sides of
the disk, which might provide a source of hot halo gas. Table 6 shows
that the halo components in both UGC 9242 and NGC 4244 are much less
prominent relative to the disk components than is true for NGC 891,
where the two components are nearly equal. If cooling gas is
responsible for the halo component, NGC 891 must have a much more
active chimney mode than UGC 9242 and NGC 4244.
Recently several galaxies have been observed to have high [OIII]/H$\alpha$~
ratios in the DIG, including NGC 891 (\markcite{ra98}Rand 1998). The
implication is that a single ionization source cannot be responsible
for both the low [OIII] and high [OIII] emitting gas, so another
source of ionization is necessary. As Rand (\markcite{ra98}1998) and
Wang \& Heckman (\markcite{whl97}1997) point out, shock ionized gas
can have high [OIII]/H$\alpha$, so shocks may be an important
mechanism for ionizing this component of the DIG. In NGC 891 the high
[OIII] values are measured in the very high-z gas, suggesting that
shock ionization or some other mechanism is important in the upper
halo. In NGC 4631 we measure [OIII]/H$\alpha$~+[NII] ratios consistent with
the DIG in several external galaxies (\markcite{gwb97}Greenawalt {\it{et al.}}~
1997; \markcite{whl97}Wang, Heckman, \& Lehnert 1997), but higher than
the Milky Way ratios. Photoionization models ({\it{i.e.}}~
\markcite{dm94}Domg\"orgen \& Mathis 1994) have been constructed which
reproduce the Galactic values, so they underpredict the [OIII] in NGC
4631. The [OIII]/H$\alpha$~ ratio seen in the DIG of M31 (Greenawalt {\it{et al.}}~
1997\markcite{gwb97}) is similar to those we find for NGC 4631. Those
authors found that a model adjusted so that the ionizing radiation is
less diluted on average could reproduce the observed ratios. A similar
situation may exist in NGC 4631, where the overall star formation is
enhanced in the disk. It then appears that photoionization is the only
ionization mechanism necessary for NGC 4631, in the disk at least. The
sensitivity at high z in the [OIII] and [SII] images is not good
enough to determine the ratio, so a trend such as that in NGC 891 may
well exist.
\subsection{Outer Disk Emission}
Our efforts to detect the outer disk H$\alpha$~ emission allow us to set an
upper limit on the metagalactic ionizing radiation field. Maloney
(\markcite{m93}1993) constructed a model of the ionization of the
outer disk of NGC 3198 by the metagalactic radiation field. The model
predicts a range of expected H$\alpha$~ surface brightness for different
values of the ionizing flux. An important result from the model was
that the predictions are insensitive to galaxy parameters, so we can
apply the results for NGC 3198 to the galaxies in our sample. The
models assume that the gas is smoothly distributed; if the gas is
clumpy the emission will be brighter.
Comparison of our upper limit on the H$\alpha$~ emission of 0.13 pc cm$^{-6}$\space
for the outer disk of NGC 3003 with the predicted emission measure in
\markcite{m93}Maloney (1993, see their figure 14) shows that the upper
limit on the metagalactic ionizing flux J$_{\nu}$ is between 8 and 12
$\times$10$^{-23}$ergs cm$^{-2}$ s$^{-1}$ Hz$^{-1}$ sr$^{-1}$. The
calculations in
\markcite{m93}Maloney (1993) are based on the observed sharp cutoff in
the HI disk at large radii. The scenario is one in which the HI disk dips
below a critical column density where it becomes optically thin to the
metagalactic ionizing flux, going from a mostly neutral disk to a
mostly ionized disk. Vogel
{\it{et al.}}~ (\markcite{vwrh95}1995) set a limit on extragalactic H$\alpha$~
emission using Fabry-Perot observations of an intergalactic Hydrogen
cloud. They placed a 2$\sigma$ limit on the metagalactic ionizing flux
to 8$\times$10$^{-23}$ergs cm$^{-2}$ s$^{-1}$ Hz$^{-1}$ sr$^{-1}$, in
agreement with the result presented here. Donahue {\it{et al.}}~
(\markcite{das95}1995) found that
J$_{\nu}$$<$3.3$\times$10$^{-23}$ergs cm$^{-2}$ s$^{-1}$ Hz$^{-1}$
sr$^{-1}$, by imaging intergalactic Hydrogen clouds in H$\alpha$, more
stringent than our limit.
A simple check of these numbers can be made by calculating the number
of ionizing photons required to produce the observed H$\alpha$~ emission. An
emission measure of 0.13 pc cm$^{-6}$\space requires 8$\times$10$^{4}$
photons cm$^{-2}$ s$^{-1}$, assuming the HI is optically thick to
ionizing photons and that 45\% of ionizing photons result in H$\alpha$~
photons (Case B). The current limit is 6$\times$10$^{4}$ photons
cm$^{-2}$ s$^{-1}$ for two-sided incident ionizing flux (Vogel {\it{et al.}}~
\markcite{vwrh95}1995). This would imply that our limit is a factor of
1.3 higher, so J$_{\nu}$ $<$ 11 $\times$10$^{-23}$ergs cm$^{-2}$
s$^{-1}$ Hz$^{-1}$ sr$^{-1}$. We have not quite reached the limits set
by Fabry-Perot observations nor that set by H$\alpha$~ imaging of Hydrogen
clouds, so it is not surprising that we do not detect any outer disk
emission from NGC 3003. It is important to remember that the
conversion from H$\alpha$~ emission measure to metagalactic ionizing flux
assumes that no [NII] is present. The filter used here may transmit
[NII] at the 10$-$20\% level. This is normally a small percentage of
H$\alpha$~, but it could be significant if the [NII]/H$\alpha$~ ratio is very high
in the outer disk, as was found by Bland-Hawthorn {\it{et al.}}~
(\markcite{bfq97}1997) for NGC 253.
In UGC 9242 there is an indication of outer disk H$\alpha$~ emission at the
0.16 pc cm$^{-6}$\space level. If this emission were a result of the
metagalactic ionizing radiation field, it would require that
J$_{\nu}$=17.0$\times$10$^{-23}$ergs cm$^{-2}$ s$^{-1}$ Hz$^{-1}$
sr$^{-1}$, higher than the Vogel {\it{et al.}}~ (\markcite{vwrh95}1995) upper
limits, and higher than the limit set by NGC 3003. The presence of
several foreground stars and a background galaxy at this location in
the image lead us to question whether this feature is real. It could
be either the result of flatfielding errors, or it could be real
emission but powered by some source other than the metagalactic
ionizing radiation field.
\acknowledgments
We thank the KPNO staff for their help during these observing runs. We
also thank B. Greenawalt for observing NGC 891, and M. F. Steakley for
reducing the NGC 4244 and NGC 4631 images. We thank the referee, James
Schombert, for useful comments which improved the presentation of the
results. This research was supported by the NSF through grant
AST-9617014, and by a Cottrell Scholar Award from Research
Corporation. C.G.H. was supported by a grant from the New Mexico Space
Grant Consortium.
|
\section{Introduction}
The thermodynamics and dynamical properties of vortex matter in high
temperature superconductors have received the interest of a large
experimental and theoretical community\cite{Blatter}. One of the most
salient findings, already predicted in 1985\cite{Nelson}, is the first order
melting transition of the vortex lattice. Large thermal fluctuation in a
high $T_{c}$ material induce the melting transition first observed in
transport measurements\cite{Safar} and confirmed by a number of equilibrium
properties like direct measurements of the magnetization jump \cite{Zeld} and
latent heat\cite{Schilling}.
It is now well accepted that in clean Bi$_2$Sr$_2$CaCu$_2$O$_8$ (BSCCO) and
untwined YBa$_2$Cu$_3$O$_7$ (YBCO) samples
the melting of the vortex lattice for an external magnetic field parallel to
the $c$-axis of the samples is first order and that the superconducting
coherence is lost simultaneously in all directions at the melting
temperature. Despite of the great amount of work devoted to the subject, the
nature of the melting transition of the vortex lattice is still
controversial.
The first theories for melting of the vortex lattice were based on the
Lindemann criterion\cite{Lindemann} and describe a situation in which the
vortex lattice melts into a liquid of entangled vortex lines. However the
behavior of the entropy change at the melting transition suggest that at
the first order transition a simultaneous loss of the triangular crystal
structure and a decoupling of the
planes take place. In this last scenario, the liquid would be a liquid of
pancake vortices. Many authors have analyzed the experimental results
obtaining good fittings with this picture. The precise nature of the
transition and the structure of the liquid phase at the melting line is
still an open problem.
More recently, some striking regularities -weak universalities- observed
in the transport properties at the melting temperature were interpreted in
terms of the Hansen-Verlet\cite{Verlet},\cite{Rojo} freezing criterion. This
criterion, that has been proved to work very well in the case of classical
liquids, states that the liquid freezes when the first peak of the structure
factor $S(k)$ reaches a critical value. However, for the case of vortex
matter, the validity of this criterion has not been analyzed.
In this context it is interesting to revisit the density functional theory
for vortices in the highly anisotropic high $T_{c}$ superconductors. This
theory describes interacting pancake vortices in layered superconductors.
The interactions are of electromagnetic nature, Josephson interactions
involving charge transfer are neglected, and consequently it is a good
starting point to describe highly anisotropic systems. However, as we
discuss below, it may be also appropriate to describe even the
less anisotropic high $T_{c}$ cuprates. The theory allows to calculate the
melting temperature, thermodynamic properties at the melting line as well as
the validity of the phenomenological criteria for melting and freezing.
In what follows we present the model for the inter-vortex interaction and
the method for the calculus of the melting line. We present the results for
the phase diagram and compare them with experimental data. We analyze the
validity of the Lindemann and Hansen and Verlet criteria for melting and
freezing.
\section{Description of the Model}
Our starting point is the functional density theory for pancake vortices in
a layered superconductor in a magnetic field perpendicular to the layers. In
a layered system in the limit of infinite effective mass perpendicular to
the layers (zero Josephson coupling) the pancake vortices can be treated as
point like classical particles restricted to move in the planes. The
vortex-vortex interaction is given by a three-dimensional (3D) anisotropic
pair potential\cite{Feig} which in Fourier space is given by:
\begin{equation}
\beta V({\bf k})=\frac{{\Gamma \lambda }^{{2}}{[k}_{\perp }^{{2}}{+(4/d}^{{2}%
}{)sin}^{{2}}{(k}_{{z}}{d/2)]}}{{k}_{{\perp }}^{{2}}{[1+\lambda }^{{2}}{k}_{{%
\perp }}^{{2}}{+4(\lambda }^{{2}}{/d}^{{2}}{)sin}^{{2}}{(k}_{{z}}{d/2)]}},
\label{interk}
\end{equation}
here, ${\bf k}=({\bf {k}_{\perp }},{k}_{z})$ is the wave vector, $\lambda $ is
the planar London penetration depth and ${\Gamma =\beta d\Phi }_{{0}}^{{2}}{%
/4\pi \lambda }^{{2}}$ is a dimensionless strength parameter where ${\Phi }_{%
{0}}$ is the flux quantum{, }${d}${\ is the distance between planes and }$%
\beta =(k_{B}T)^{-1}$ is the inverse temperature.
Following the pioneering work of Ramakrishnan and Yussouff (RY) \cite
{Krishna1} and the extensions made to describe the vortex matter \cite
{Krishna3},\cite{Krishna2}, the functional describing the free energy
difference between the solid and liquid phases is given by:
\begin{eqnarray}
\frac{\Delta \Omega }{k_{B}T} &=&\sum_{n}\int {d^{2}r_{\perp }}\left\{ {\rho
_{n}({\bf r}_{\perp }{\,})}\left[ {\ln \left( \frac{\rho _{n}({\bf r}_{\perp
}{\,})}{\rho _{\ell }}\right) -1}\right] {+\rho _{\ell }}\newline
\right\} \nonumber \\
&&-\frac{1}{2}\sum_{n,n^{\prime }}\int {d^{2}r_{\perp }\,d^{2}r_{\perp
}^{\prime }\,c}_{\left| n-n^{\prime }\right| }{(|{\bf r}_{\perp }^{\prime }\,%
}{-{\bf r}_{\perp }\,|)\times } \nonumber \\
&&{\left[ \rho _{n}({\bf r}_{\perp })-\rho _{\ell }\right] \left[ \rho _{%
{n}^{\prime }}({{\bf r}_{\perp }^{\prime }})-\rho _{\ell }\right] }.
\label{fren}
\end{eqnarray}
here, ${\bf r}$ $\equiv ({\bf r}_{\perp },nd)$ where ${\bf r}_{\perp }$ is
the in-plane coordinate and $n$ is an integer (the plane index). The quantity
$\rho _{n}({\bf r%
}_{\perp })=<\sum_{i}\delta ({\bf r}_{\perp }-{\bf r}_{\perp i}^{n})>$ is
the vortex density with ${\bf r}_{\perp i}^{n}$ the 2D coordinate of the
i-th. particle at plane $n$, $\rho _{\ell }$ is the mean areal density of
the liquid and ${c}_{\left| n-n^{\prime }\right| }{(|{\bf r}_{\perp
}^{\prime }\,-{\bf r}_{\perp }\,|)}$ is the direct correlation function in
the liquid phase. The first integral in the right hand side of Eq. \ref{fren}
describes the non-interacting contribution to the free energy, the second
integral incorporates the effect of the interactions up to second order in
the difference $({\rho _{{\sl n}}({{\bf r}_{\perp }})-\rho _{\ell })}$.To
evaluate this energy difference it is necessary to calculate the direct
correlation function of the liquid defined below.
The pair distribution function is defined as:
\begin{equation}
g({\bf r},{\bf r}^{\prime })=\frac{\rho _{n,n^{\prime }}({\bf r}_{\perp },%
{\bf r}_{\perp }^{\prime })}{\rho _{n}({\bf r}_{\perp })\rho _{{\sl n}%
^{\prime }}({\bf r}_{\perp }^{\prime })}.
\end{equation}
where:
\begin{equation}
\rho _{n,n^{\prime }}({\bf r}_{\perp },{\bf r}_{\perp }^{\prime
})=<\sum_{i\neq j}\delta ({\bf r}_{\perp }-{\bf r}_{\perp i}^{n})\delta (%
{\bf r}_{\perp }^{\prime }-{\bf r}_{\perp j}^{\,n^{\prime }})>
\end{equation}
here $\rho _{n,n^{\prime }}({\bf r}_{\perp },{\bf r}_{\perp }^{\prime })$ is
the probability density of finding two particles at ${\bf r=}({\bf r}_{\perp
},nd)$ and ${\bf r}^{\prime }{\bf =}({\bf r}_{\perp }^{\prime },n^{\prime
}d) $ respectively. Due to the symmetry of the liquid $g({\bf r},{\bf r}%
^{\prime })=g_{\left| n-n^{\prime }\right| }(\left| {\bf r}_{\perp }-{\bf r}%
_{\perp }^{\prime }\right| ).$
The direct correlation function $c_{n}({\bf r}_{\perp })$ is defined by
means of Ornstein-Zernike equation \cite{ozie}:
\begin{equation}
h_{n}(r_{\perp })=c_{n}(r_{\perp })+\rho _{\ell }\sum_{n^{\prime }}\int
d^{2}r_{\perp }^{\prime }c_{\left| n-n^{\prime }\right| }(|r_{\perp
}-r_{\perp }^{\prime }|)h_{n^{\prime }}(r_{\perp }^{\prime }). \label{OZ}
\end{equation}
where $h_{n}(r_{\perp })=g_{n}(r_{\perp })-1$ is the pair correlation
function$.$
This integral equation for the direct correlation function $c_{n}(r_{\perp
}) $ can be solved in the hipernetted chain approximation, and inserting the
solution in Eq.(\ref{fren}), the free energy difference between the solid
and liquid phase is given only as a functional of the density $\rho _{n}(%
{\bf r}_{\perp }{\,})$. At this point, we can expand the density in a set of
trial reciprocal-lattice vectors:
\begin{equation}\label{dens}
\rho _{n}({\bf r}_{\perp }{\,})/\rho _{\ell}=1+\eta +\sum_{{\bf G\neq 0}}\rho
_{%
{\bf G}}{{e}^{i{\bf G\cdot r}}}
\end{equation}
The quantities $\eta $ and $\rho _{{\bf G}}$, may be viewed as dimensionless
order parameters, $\eta $ is the fractional density difference between the
liquid and the solid phases and the $\rho _{{\bf G}}$ are the Fourier components
of the density in the solid phase. Minimizing the free energy (\ref{fren})
with respect to these parameters, we obtain the following equation:
\begin{equation}
1+\eta +\sum_{{\bf G\neq 0}}\rho _{{\bf G}}{e}^{i{\bf G\cdot r}}=\exp ({\rho
_{\ell }c_{0}\eta +\rho _{\ell }\sum_{{\bf G\neq 0}}\rho _{{\bf G}}c_{{\bf G}%
}{e}^{i{\bf G\cdot r}}}), \label{auto}
\end{equation}
where $c_{{\bf G}}$ is the Fourier transform of the direct correlation
function and $c_{0}\equiv c_{{\bf G=0}}$.
Due to the small compressibility of the system the fractional density change
is small ($\eta \ll 1$), and as a first approximation can be neglected in
the first term of (\ref{auto}). Note that the product ${c_{0}\eta }$ is not
necessarily small and the first term in the exponential of the right hand
side of (\ref{auto}) has to be retained. Transforming Fourier Eq. (\ref{auto})
we obtain after some algebra a set of coupled equations for the $\rho _{
{\bf G}}$:
\begin{equation}
\rho _{{\bf G}}=\frac{\sum_{n}\int_{A_{c}}d^{2}r_{\perp }e^{-i{\bf G\cdot r}
}\exp ({\rho _{\ell }\sum_{{\bf G\neq 0}}\rho _{{\bf G}}c_{{\bf G}}{e}^{i
{\bf G\cdot r}})}}{\sum_{n}\int_{A_{c}}d^{2}r_{\perp }\exp ({\rho _{\ell
}\sum_{{\bf G\neq 0}}\rho _{{\bf G}}c_{{\bf G}}{e}^{i{\bf G\cdot r}})}}.
\label{equrhog}
\end{equation}
here $A_{c}$ is the area of the in-plane unit cell. Given the Fourier
components $c_{{\bf G}}$ of the direct correlation function, these equations
can be solved for \{$\rho _{{\bf G}}$\}. Before presenting the numerical
results, the following points deserve a comment:
$i)$ A uniform liquid is always a solution of (\ref{equrhog}) \{$\rho _{{\bf
G}}=0$\}. As the $c_{{\bf G}}$ increase, non-uniform solutions may suddenly
appear (with $\rho _{{\bf G}}\sim 0.5$) and correspond to the solid phase
with a spatially periodic density.
$ii)$ At the transition the free energy difference $\Delta \Omega $ defined
in (\ref{fren}) is equal to zero. To calculate the transition line in the $
B-T$ phase diagram, we have to calculate the direct correlation function as
a function of the magnetic induction and temperature and use it as an
input in the RY density functional theory. The transition is obtained when a
set of order parameters \{$\rho _{{\bf G}}\neq 0$\} given by (\ref{equrhog})
satisfy the condition $\Delta \Omega =0$.
$iii)$ As discussed in previous works, the numerical calculation can be
simplified by noting that in the liquid phase, the Fourier transform $c_{
{\bf G}}$ of the direct correlation function rapidly decay with increasing $
\left| {\bf G}\right| $, it is then a good approximation to truncate the
series ${\sum_{{\bf G\neq 0}}\rho _{{\bf G}}c_{{\bf G}}{e}^{i{\bf G\cdot r}}}
$ in (\ref{equrhog}) and take only a few terms with $c_{{\bf G}}\neq 0$. The
number of terms retained defines the number of self consistent order
parameters \{$\rho _{{\bf G}}$\} included in the theory. In what follows we
consider only two order parameters corresponding to the first two peaks of
the in-plane direct correlation function as first suggested by Ramakrishnan
\cite{Krishnad} for a 2D system.
$iv)$ With the resulting set of order parameters that characterize the solid
phase at the melting point in the limit $\eta \ll 1$, the fractional density
change can be estimated. However in the RY approximation used here the
results for $\eta $ are not accurate, in particular it is ease to show that $
\eta >0$ in contradiction to what is expected for the case of vortices.
Rather than improving the estimate of $\eta $ by including higher order
terms in the functional $\Delta \Omega $ of Eq. (\ref{fren}) -which are
given by not available three particle correlation functions- we estimate the
entropy change at the transition and calculate $\eta $ by means of the
Clausius-Clapeyron relation. The higher order corrections to $\Delta \Omega $
are not strongly coupled to the order parameters \{$\rho _{{\bf G}}$\} and
the present approximation is known to give accurate results when the
crystal structure of the solid phase is given.
\section{Correlation functions}
In this section we present results for the pair correlations in the liquid
phase, we use the hipernetted chain approximation (HNC) which is well
described in the literature. For the sake of completeness we only give here
the basic formulas of classical liquids theory relevant for our case.
The pair correlation function can be written in the form:
\begin{equation}
g_{n}(r_{\perp })={e}^{[g_{n}(r_{\perp })-1-\beta V_{n}(r_{\perp
})-c_{n}(r_{\perp })+B_{n}(r_{\perp })]}, \label{clausu}
\end{equation}
where $B_{n}(r_{\perp })$ is known as the bridge function and its closed
expression is not known for an arbitrary potential. In the HNC approximation
we have $B_{n}(r_{\perp })\equiv 0$ which is an excellent approximation for
long ranged soft potentials as the one we are dealing with \cite{ozie}. In the
HNC approximation Eq. (\ref{clausu}) can be rewritten in the following form:
\begin{equation}
c_{n}(r_{\perp })={e}^{[-\beta V_{n}(r_{\perp })+Y_{n}(r_{\perp
})-1-Y_{n}(r_{\perp })]}, \label{HNC}
\end{equation}
where $Y_n(r_\perp)=h_n(r_\perp)-c_n(r_\perp)$ . This equation together with
the Ornstein Zernike
relation (Eq.(\ref{OZ})) are the basic equations that allows to evaluate the
pair correlation functions. The asymptotic behavior of the direct
correlation function as $r_{\perp }\to \infty $ and $n\to \infty $ is known
to be $c_{n}(r_{\perp })=-V_{n}(r_{\perp })/k_{B}T$. For the
numerical calculation it is then convenient to divide the correlation
functions in short and long ranged contributions according to:
\begin{equation}
c_{n}(r_{\perp })=c_{nS}(r_{\perp })+c_{nL}(r_{\perp }),
\end{equation}
with
\begin{equation}
c_{nL}(r_{\perp })=-\beta V_{n}(r_{\perp })
\end{equation}
and
\begin{equation}
Y_{n}(r_{\perp })=Y_{nS}(r_{\perp })+Y_{nL}(r_{\perp }),
\end{equation}
with
\begin{equation}
Y_{nL}(r_{\perp })=+\beta V_{n}(r_{\perp }).
\end{equation}
Due to the highly anisotropic nature of the pair potential, as a first
approximation we take the short ranged contribution different from zero only
for the in-plane correlations:
\begin{equation}
c_{nS}(r_{\perp })={\delta }_{n,0}c_{0S}(r_{\perp })
\end{equation}
and the problem reduces to calculate a single function ($c_{0S}(r_{\perp })$
) given by the following set of integral equations:
\begin{equation}
c_{0S}(r_{\perp })=\exp (Y_{0S}(r_{\perp }))-1-Y_{0S}(r_{\perp }),
\label{OZ_lay}
\end{equation}
\begin{equation}
Y_{0S}(k_{\perp })=\frac{d}{2\pi }\int_{-\frac{\pi }{d}}^{\frac{\pi }{d}
}dk_{z}\frac{c_{0S}(k_{\perp })-\beta V({\bf k})}{1-\rho_{\ell}[c_{0S}(k_{\perp
})-\beta V({\bf k})]}-c_{0S}(k_{\perp }), \label{HNC_lay}
\end{equation}
the last equation is the Fourier transformed Ornztein-Zernike equation. This
expression can be further simplified by an analytic integration due to the
particular form of the pair potential.
Finally the out of plane short ranged
correlation functions in a second order approximation are given by the
following equation:
\begin{equation}
Y_{nS}(k_{\perp })=\frac{d}{2\pi }\int_{-\frac{\pi }{d}}^{\frac{\pi }{d}
}dk_{z}{e}^{-ik_{z}nd}\frac{c_{0S}(k_{\perp })-\beta V({\bf k})}{
1-\rho_{\ell}[c_{0S}(k_{\perp })-\beta V({\bf k})]}. \label{HNC_layn}
\end{equation}
In Fig. \ref{paird} we present results for the pair distribution
function where the lengths are in units of the characteristic distance $a$
given by $\pi a^2 \rho_\ell =1$. The first peak of the in plane pair
distribution function occurs
at a finite value of the $r_{\perp }$which is given by the mean
inter-particle distance $(\sim 2a)$. For the out of plane distribution function
the
maximum occurs at $r_{\perp }=0$ which shows the tendency of pancake
vortices to form stacks. In Fig. \ref{decay} the behavior of $h(r_{\perp
}=0,n\neq 0)$ is presented and shows an exponential decay as a function of $
n $. This behavior allows as to define a correlation length $\Lambda $
through the following relation: $h(r_{\perp }=0,n)\propto \exp (-dn/\Lambda
).$ In Fig. \ref{lambda} we present the behavior of $\Lambda $ as a function
of the magnetic induction $B$, that can be fitted with $\Lambda \propto 1/
\sqrt{B}$. Other authors \cite{Krishna3} had an excellent fit of their data
in a smaller range of fields with a function of the form $\Lambda
=A_{1}+A_{2}/B.$
It is important to note that the characteristic length $\Lambda $ describes
the long distance behavior of the correlation functions. The short distance
behavior is given by $h(r_{\perp }=0,n)$ with small $n$ which in general
decays very fast. As can be seen in Fig.\ref{decay}, $h(r_{\perp }=0,n=1)$ is
very
small ($\simeq 10^{-2}$), and although its value depends on field and
temperature, in the liquid phase, it always remains much smaller than $1$.
As the temperature decreases the system develops stronger correlations. This
is clearly observed in the structure factor, defined by:
\begin{equation}
S({\bf k}_{\perp },k_{z})=1+\rho _{\ell }\sum_{n}\int d^{2}r_{\perp
}\,h_{n}(r_{\perp })\,{e}^{i({\bf k}_{\perp }\cdot {\bf r}_{\perp
}+k_{z}nd)}.
\end{equation}
In Fig. \ref{strfac} the structure factor is shown for two different
temperatures, the lower the temperature, the higher the first peak in $S(
{\bf k})$.
In the next section we use these results to calculate the melting line.
\section{Phase diagram}
We used the correlation functions calculated above as an input to the RY
density-functional theory of freezing. In Fig. \ref{pdiag1} we show the
results obtained for the phase diagram with parameters corresponding to
BSCCO but without temperature dependence: $\lambda(T)=\lambda(0)=1500$\AA ,
$d=15$\AA $\,$ and $T_{c}=90K$. To
understand these results we resort to a simple model obtained by mapping the
vortex system onto a 2-D boson system. In this model, the free energy of a
system of vortices is given by\cite{Nelsone}:
\begin{equation}
{\rm F}[r(z)]=\int_{0}^{L}\,dz\left\{ \sum_{i}\frac{1}{2}g\left( \frac{
dr_{i}(z)}{dz}\right) ^{2}+\sum_{i<j}V(r_{ij}(z))\right\} ,
\end{equation}
where $r_{i}(z)$ determines the position and shape of the i-th. vortex, $g$
is the elastic energy and the last term describes the interaction between
vortices, which is taken to be logarithmic, $r_{ij}(z)$ is the distance
between vortices. In a solid phase, of lattice constant $a_0$, we can use an
harmonic approximation for the pair interaction
potential. Assuming a characteristic length $l$ which defines the
correlation along the field direction, the free energy after a thermodynamic
average can be approximated by:
\begin{equation}
{{\rm F}^{\prime }}={\rm F}/N\simeq \frac{<r^{2}>}{2}\left\{ \frac{g\,}{l}
+kl\right\} , \label{fren3}
\end{equation}
where ${{\rm F}^{\prime }}$ is the mean free energy per characteristic vortex
segment of length $l$, $N$ is the number of independent segments in the
sample, $<r^{2}>$ is the mean square displacement of the vortices and $
k\propto \epsilon _{0}/a_{0}^{2}$ is the curvature of the logarithmic pair
potential at the mean inter-particle distance. Using the Lindemann melting
criterion $<r^{2}>=u_{L}^{2}a_{0}^{2}$, with $u_{L}\sim 0.1-0.3$ along the
transition line and the equipartition theorem that states that ${{\rm F}
^{\prime }\sim k}_{B}T$, the expression above reduces to:
\[
{k}_{B}T_{m}=\frac{u_{L}^{2}a_{0}^{2}}{2}\left( \frac{g}{l}+kl\right) .
\]
If the characteristic length $l$ is taken as the distance between planes
which should be a good approximation at least in the high field regime we
obtain:
\begin{equation}
k_{B}T_{m}\simeq \frac{u_{L}^{2}\epsilon _{0}d}{2}\left\{ 1+\frac{g\Phi _{0}
}{d^{2}\epsilon _{0}B_{m}}\right\} . \label{meltB}
\end{equation}
If the inter-plane correlations were relevant at the melting point, the
characteristic length $l$ should be obtained minimizing the free energy (\ref
{fren3}). In this case the melting line is given by:
\begin{equation}
k_{B}T_{m}=u_{L}^{2}\sqrt{g\epsilon _{0}}\left( \frac{\Phi _{0}}{B_{m}}
\right) ^{\frac{1}{2}}. \label{meltA}
\end{equation}
The density functional data for the melting line were fitted with the
following expression $T_{m}=T_{m}^{2D}+A_{1}/B_{m}$ according to (\ref{meltB}).
This type of behavior is expected for a highly anisotropic system at high
fields, when the liquid state at the melting temperature presents very short
correlations along the $c$ axis{\bf . }Although the fit is less accurate for
small fields we never obtain a behavior of the type given by (\ref{meltA}).
This suggests that in our field and temperature range, the system
simultaneously melts and the planes decouple to form a liquid of pancake
vortices. The less accurate fit for the low field data could indicate that
we are entering a cross-over to a line melting regime. In the limit of high
fields the melting temperature goes asymptotically to the 2D melting
temperature $T_{m\text{ }}^{2D}$. This is due to the fact that at high
fields the inter-plane correlations become irrelevant and the system behaves
as a collection of independent planes. The 2D melting temperature, obtained
with a functional density theory in the RY approximation, is lower than
expected from a Kosterlitz-Thouless dislocation unbinding theory.
This is presumably due to the fact that the HNC approximation underestimates
the correlations in the 2D case as was discussed by other authors\cite
{Krishna3}.
In order to compare our results with the experimental data, we included the
temperature dependence of the penetration length\cite{Leel}: $\lambda(T)
=\lambda
(0)/[1-(T/T_{c})^{4}]^{1/2}$. In Fig. \ref{zeldov} we present our
results with $\lambda (0)=1500$\AA $\,$ together with the experimental
results of Zeldov et al. In the experimental data the first order transition
ends at a critical point presumably due to point-like pinning centers, an
effect not included in this functional density theory. In order to make this
comparison, $\lambda (0)$ was used as a fitting parameter. A more realistic
value of $\lambda (0)$ for BSCCO is $\sim 2000$\AA . If the $\lambda (0)$
is increased, the melting line moves to lower fields due to the decrease of
the intensity of the pancake-pancake interactions, this could be compensated
by the inclusion of a small Josephson coupling between planes in a more
realistic treatment.
\section{Entropy and magnetic induction jump at $T_{m}$}
The entropy jump can be calculated as the temperature derivative of the free
energy difference:
\begin{equation}
\Delta S=-\left( \frac{\partial \Delta \Omega }{\partial T}\right) _{V,H}
\label{entropia1}
\end{equation}
Using Ecs. \ref{dens} and \ref{auto} the free energy difference of
Ec.\ref{fren} can be written as:
\begin{eqnarray}
\frac{\Delta \Omega }{k_{B}T} &=&-\ln {\left[ \frac{1}{v_{c}}{
\int_{v_{c}}d^{3}r\exp {\left( \sum_{{\bf G}}\rho _{{\bf G}}c_{
{\bf G}}e^{i{\bf G}\cdot {\bf r}}\right) }}\right] } \nonumber \\
&&+\frac{1}{2}\sum_{{\bf G}}{c_{{\bf G}}}{\rho _{{\bf G}}^{2}}\,,
\label{difenfour}
\end{eqnarray}
To evaluate the entropy jump we must calculate the temperature derivatives
of the Fourier transforms of the liquid direct correlation functions ${c_{
{\bf G}}}$ at constant $H$ and $V$, that are not known. Following reference
\cite{Krishnad}, this temperature derivatives can be estimated on the solid
phase using the fluctuation-dissipation theorem
in the classical limit. This procedure gives for the structure factor $S({\bf
G})$ a
linear dependance on temperature. Using this result we obtain the following
expression for the entropy jump:
\begin{equation}
\Delta s/k_{B}=3[\rho _{1}^{2}(1-c_{1})+\rho _{2}^{2}(1-c_{2})],
\end{equation}
in our two order parameter calculation.
In Fig. \ref{entrop} we present the density functional results for the
entropy change. Its average value $\sim 0.3k_{B}$ is consistent with
experimental data\cite{Zeld}. With the entropy jump result and the
Clausius-Clapeyron relation we can calculate the magnetic induction jump at
the transition line:
\begin{equation}
\frac{\Delta B}{B_{m}}=\left( \frac{d\Phi _{0}}{4\pi \Delta s}\frac{dB_{m}}{
dT}\right) ^{-1},
\end{equation}
where we have made the following approximation:
\begin{equation}
\frac{dH_{m}}{dT}\sim \frac{dB_{m}}{dT}.
\end{equation}
The results for the relative magnetic induction jump are plotted in Fig. \ref
{bjump} with the Zeldov et al. experimental data digitalized. The parameters
used to obtain this data were the same as those used to calculate the phase
diagram of Fig. \ref{zeldov}. There is a very good agreement in the
magnitude and temperature dependance with this direct experimental
measurement.
\section{Melting and freezing criteria}
In this section we analyze the Lindemann melting criterion and the
Hansen-Verlet freezing criterion for the vortex system. The Lindemann
criteria states that the solid melts when the square root of the mean square
displacement of the particles is an order of magnitude smaller than the
lattice parameter i.e. $<u^{2}>/a_{0}^{2}=u_{L}^{2}$, with $u_{L}\sim 0.1$.
This criteria has been widely used to estimate the phase diagram and
different properties of the system at the melting line. In general it is
accepted that for the melting of vortex mater the Lindemann parameter $u_{L}$
is larger than for other classical 3D solids and its value is usually taken
in the range $0.2-0.3$.
We estimated the mean square displacement as:
\begin{equation}
<u^{2}>\simeq \frac{1}{A_{c}}\int_{A_{c}}d^{2}r_{\perp }\,r_{\perp }^{2}
\frac{\rho _{n}({\bf r}_{\perp }\,)}{\rho _{\ell }},
\end{equation}
the density at the melting line is obtained using (\ref{auto}) and the
results for the Lindemann parameter are shown in Fig. \ref{lindemann}. For
low fields, numerical results show that $u_{L}$ $\simeq $ $0.2$ in good
agreement with previous estimates, as the field is increased (the melting
temperature decreases) the Lindemann parameter increases. This is due to the
fact that, as discussed above, in the present model as the field increases
the interplane coupling decreases and the transition becomes essentially 2D.
For a 2D transition, the Lindemann parameter is larger than for a 3D one.
The Hansen-Verlet freezing criteria is the counterpart of the Lindemann
criteria and states that a liquid freezes when the first peak of the
structure factor reaches the value $2.85$ in a 3D classical system and of
the order of $5.5$ for a 2D plasma \cite{plasma2d}. Having evaluated the
structure factor at the melting transition within the functional density
theory, we can show the validity of this criteria for the case of freezing
of the vortex liquid. The results for $S(k_{\max })$ are shown in Fig. \ref
{lindemann} and as in the case of the Lindemann criteria the Hansen-Verlet
criteria is valid for low fields with a value $S(k_{\max })\sim 6.5$, much
larger than that obtained for isotropic 3D system. As the melting
temperature decreases $S(k_{\max })$ towards its two dimensional value. In
our case the value of $S(k_{\max })$ at small fields is larger than the 2D
value due to the following reason: the idea behind the Hansen-Verlet
criteria is that when the liquid develops correlations, that overcame some
critical value, it freezes. At high temperatures the first peak in $S(k)$
dominates and is a good measure of the correlations, as the temperature
decreases, the liquid develops stronger correlations for large ${\bf G}$ and
both the first and second peaks in $S(k)$ became relevant at the freezing
point. Consequently, it is not necessary for $S(k_{\max })$ to become too
large to indicate that the liquid has developed critical correlations. This
can be viewed as a relative increase of the order parameter $\rho _{2}$ with
respect to $\rho _{1}$at the transition as the melting temperature
decreases. The results indicate that for low fields the ratio $\rho
_{1}/\rho _{2}$ is larger than $4$, while at high fields is of the order of $
2$.
\section{Summary and discussion}
In this work we study the melting transition for the vortex system in
layered superconductors using the RY functional density theory. Using
reasonable set of parameters for BSCCO we obtained qualitative and
quantitative agreement with the available experimental data.
The functional density theory allows an estimation of the entropy jump at
the transition and using the Clausius-Clapeyron relation we estimated the
jump in magnetic induction. The same parameters used to fit the phase
diagram, give results for the magnetic induction jump that are in
quantitative agreement with the experimental results. Finally we analyzed
the validity of the melting and freezing criteria and showed that in the
region of interest -the low field region where the first order transitions
occurs- the Lindemann and the Hansen-Verlet criteria are valid within a
error of $10\%$ which is typical for this type of criteria.
The model predicts the melting of the vortex lattice into a liquid of almost
decoupled pancake vortices even in the low field region. In this sense we
may interpret our results as a simultaneous decoupling and melting of the
vortex lattice\cite{Glazman} rather than a melting into a liquid of vortex
lines. This picture may be appropriate for a highly anisotropic system like
BSCCO. However, in a recent paper Kitazawa et al.\cite{Kitazawa}
successfully used a scaling suitable for simultaneous decoupling and melting
not only for BSCCO but also for less anisotropic systems like YBCO. This
scaling predicts that the melting line is given by:
\begin{equation}
B_{m}=C\left( \frac{T_{c}}{T_{m}}-1\right) \label{scale}
\end{equation}
where the prefactor depends on the anisotropy and interplane separation of
the system. Both the experimental results of Zeldov et al. for BSCCO and our
numerical results for low fields are in good agreement with this expression
as shown in Fig. \ref{sublim}. In the figure we present results corresponding
to a zero temperature penetration depth $\lambda (0)=1500$\AA\ which gives
very good agreement with the results of Zeldov et al. for the low field
region where the first order transition is experimentally observed. As
discussed in section IV, a more realistic value for $\lambda (0)$ is $2000$
\AA , which gives a melting line laying below the experimental points.
However, in any real system there is some Josephson coupling between the
planes that should compensate the lowering of the electromagnetic coupling
due to the increase of $\lambda (0)$. In this sense, we take $\lambda (0)$
as an effective parameter that measures the strength of the coupling between
planes. A change in $\lambda (0)$ does not change the temperature dependance
of $B_{m}$ for low fields, but only the coefficient $C$ in Eq.(\ref{scale}).
If experimental results confirm the fact that even for less anisotropic
system, the first order melting transition correspond to a simultaneous
decoupling and melting of the vortex lattice, the present theory could be
used to describe these systems provided that $\lambda (0)$ or the interplane
distance are chosen as effective parameters. In particular the validity of
the melting and freezing criteria obtained for the highly anisotropic
laminar BSCCO could be extended to other less anisotropic systems like YBCO.
|
\section{Introduction and Results}
\label{intro}
Many useful materials arising in the domain of chemical engineering are
liquid-crystalline, frequently lamellar. They are generally processed
under conditions of shear flow, and often structured by the addition of
particles \cite{shouche} . The complex dynamic modulus $\g*=G'(\omega) + i
G''(\omega)$ of these materials as a function of the
angular frequency $\omega$ (=2$\pi f$) presents many puzzles. For example, the lamellar
L$_{\alpha}$ phase, although in principle a one-dimensional stack of
two-dimensional fluid layers and hence incapable, in the ideal, perfectly
ordered state, of sustaining a static
shear stress, generally displays \cite{larson3} in practice a modest
storage modulus $G'(\omega)$ at small $\omega$. In addition, it has an anomalously
large low-frequency loss modulus $G''(\omega)$. Since many soft solids,
lamellar or otherwise, share these features\cite{cates}, it is clear
that rheometry alone cannot discriminate between various models for these
materials.
Since the observed rheometric properties of lamellar phases (for some
recent studies see \cite{diat,larson2,meyer}) are presumably
a consequence of topological defects and textures, one must first
correlate the rheometry to direct visual observations. One could
then hope to build a theory in two stages, by explaining first how
the defects produce the rheology, and second why flow or sample
preparation produces the defects to begin with.
A major motivation for the present study, in particular, is the
observation \cite{shouche} of a large enhancement of the shear
modulus of L$_{\beta}$ gels upon the addition of a very small
concentration of particles.
Accordingly, this paper reports parallel studies of the viscoelastic
properties and defect structure of the lamellar
phase, and
the effect thereon of shear-treatment and particulate additives.
Our main results are as follows:
\begin{itemize}
\item {\it Pure lamellar phase without shear-treatment}: The
L$_{\alpha}$ samples have $G'(\omega) \sim 10^3$ Pa and $G''(\omega) \sim 10^2$ Pa at
$\omega = 1$ rad/sec. Both $G'(\omega)$ and $G''(\omega)$ depend only weakly on $\omega$,
with $G'(\omega)$ nearly flat. Between crossed polars, the system appears dense
with defects.
\item {\it Shear-treatment without added particles}: If the system is
subjected to steady shear at a low rate $\dot{\gamma_s}$ ( 1 to 50
s$^{-1}$) for a specified duration $t_s$, the small-amplitude
frequency-dependent moduli measured after switching off the steady shear
depend strongly on both $\dot{\gamma_s}$ and $t_s$. In general, the moduli thus
measured decrease with increasing $t_s$ for fixed $\dot{\gamma_s}$, but appear
roughly to level off at a value which is an increasing function of
$\dot{\gamma_s}$. The $G'(\omega),\,G''(\omega)$ reported here seem to depend weakly on the
shear history. Observations between crossed polars show that the
shear-treated samples consist of macroscopically well-oriented lamellae
parallel to the applied velocity, and normal to the velocity gradient,
threaded by a rather beautiful network of oily-streak defects
\cite{kleman,klemanbook} (see
section \ref{defnopart} for background on these defects), as
seen recently in cholesterics \cite{zap}. It is
thus clear that the dominant contribution to the rheology comes from
this defect network in an otherwise well-oriented phase \cite{zap}, not
from a polydomain averaging of the smectic elasticity, as suggested
by \cite{kawa}.
\item {\it Effect of particulate additives}: Suspending micron size
particles had a dramatic effect on the structure and rheology. Prior to
shear treatment both $G '$ and $G ''$ are appreciably larger. Under
shear-treatment, the oily-streak network survives much longer, with the
spheres located at its nodes, and the measured $G'$ and $G ''$ persist
longer. This picture of a particle-stabilised defect network is
consistent with the observations of \cite{zap} on colloid-doped
cholesterics. It is worth adding that careful observations under the
polarising microscope reveal unambiguously that the alignment at the
particle surface is homeotropic.
\end{itemize}
The relevance of our work to that of Shouche {\it et al.} \cite{shouche}
should be emphasised here. It was conjectured in \cite{shouche} that
the enhancement of elasticity when particles are added to an L$_{\beta}$
gel arose from interparticle bridges formed by surfactant molecules.
Our study suggests strongly that oily-streak defects, not surfactant
molecules, form the bridges, and we are now turning our attention to
L$_{\beta}$ phases to see if this is so.
The remainder of this paper is organised as follows. In section II we
specify the samples used and describe our experimental setup. In section
III we present in detail our visual observations of the defect structure
and its evolution under shear, with and without added particles. The
steady-shear and small-amplitude rheometry of these systems are
discussed in Section IV. We close in section V with a tentative
interpretation and analysis of our results.
\section{Experimental}
\label{exp}
\subsection{Sample and sample preparation}
\label{prep}
The lamellar (L$_{\alpha}$) sample is a commercially available (Galaxy)
anionic surfactant, sodium dodecyl ether sulphate or SLES (73.2 wt.\%)
in water with a substantial but unspecified concentration of ionic
impurities. For some of the experiments we have diluted this sample with
distilled water and homogenized the mixture in the oven at $\sim$
70$^\circ$C for a few hours. We have dispersed 9.55$\pm$0.44 $\mu$m
polystyrene + 30\% polybutylmethacrylate (initially suspended in water, Bangs Lab) spheres at
nominal volume fractions of $\sim$ 0.5 \% in the L$_{\alpha}$ sample.
The mixtures were made by manually mixing in the particles with a glass
rod or spatula. The samples were centrifuged briefly to remove air
bubbles.
\subsection{Imaging setup}
\label{imsetup}
Our experimental studies of the flow properties of particle laden
lyotropic smectic liquid crystals are in two parts. While the
rheometry is carried out in a commercial rheometer (details below),
the visualisation of the defect structure under flow is done in
a custom-built flow cell. The design is based on one
available in the literature \cite{larson}, and was executed by Holmarc
Instruments.
\begin{figure}
\begin{center}
\epsfig{file=exset.eps,width=11cm,bbllx=0,bblly=0,bburx=477,bbury=415}
\end{center}
\vspace{1cm}
\caption{Schematic diagram of the rheooptics setup.}
\label{fig:imagesetup}
\end{figure}
Two horizontal glass plates (Borosilicate, BK7) lying parallel to the
$xy$ plane form the walls of the channel, and the sample is sheared
between these plates, providing a linear (plane Couette) flow with
velocity along $x$ and gradient along $z$. The stationary lower glass
plate is fixed to a metallic plate mounted on two micrometer screws,
which are used to ensure that the lower glass plate is parallel to the
upper plate. The upper plate is attached to the moving arm of a
translation stage, and the vertical ($z$) separation between the plates
adjusted using a micrometer (least count = 10$\,\mu$m) fixed to the
moving arm of the stage. The moving arm is mechanically linked to a
linear stepper motor via a lead screw (Holmarc Instruments), allowing a
minimum step size 0.005$\,$mm. The motor is controlled by a computer
using an ISA interface card (Semica Data Systems). This translation
stage is mounted on the microscope stage as shown in Fig.
\ref{fig:imagesetup}.
The sample is viewed through crossed polars using a Nikon Optiphot2-Pol
polarising microscope fitted with a CCD camera (Sony TK-S300) whose
output goes simultaneously to a video recorder and a computer for image
analysis (NI IMAQ 1408 image grabbing card). The line of sight is in the
direction of the velocity gradient. Note that a sample aligned
homeotropically, i.e., with layers parallel to the plates, should appear
dark through crossed polars.
The glass slides were cleaned thoroughly before the sample was loaded
and were not pretreated for any preferential alignment. The rheooptic
experiments were conducted at room temperature (25--28$^{\circ}$C).
Our rheometric studies have been carried out on a Rheolyst AR1000N (TA
Instruments) stress contolled rheometer. For all the experiments quoted
in this paper we have used a 4$\,$cm parallel plate geometry and the
temperature was maintained at 25 $^{\circ}$C.
\section{Defect structure of the lamellar phase, and its
evolution under shear}
\label{defects}
\subsection{Defect network of the lamellar phase without particles}
\label{defnopart}
The sample when loaded presents a bright appearance because it is full
of defects. The texture is that of a typical lyotropic lamellar liquid
crystal. We then subject it to large-amplitude oscillatory shear of
frequency 0.1$\,$Hz and end-to-end amplitude 1$\,$mm. The sample
thickness $d$ is 100$\,\mu$m for the case we are reporting in this
paper, with a triangle-wave strain, so that in each phase of its motion
there is a constant shear-rate $\dot{\gamma} = 2$ s$^{-1}$. The
structure observed between crossed polars evolves steadily from its
initial highly defect-ridden state, Fig. \ref{fig:images}(a), to one
with large dark (and hence homeotropic) regions, with a striking,
sparse, sample-spanning network of linear oily-streak \cite{kleman}
defects (Fig. \ref{fig:images}b), and finally to an almost totally
homeotropically oriented lamellar state with a weak background of
defects (Fig. \ref{fig:images}c). That the defects span the sample is
clear from the video because different portions of a typical defect line
move at different speeds under shear, indicating that they are at
different depths. This network structure is very similar in appearance
to that seen in \cite{zap}, and the behaviours of the two systems are
broadly alike.
We remind the reader that oily streaks \cite{klemanbook} are formed in
lamellar phases upon the close approach of a pair of
parallel dislocation lines with Burgers vectors with magnitude
$b_1$ and $b_2$ and opposite sign. Instead of annihilating partly
to yield a simple dislocation with Burgers vector of magnitude
$|b_1 - b_2|$, they display a complex internal structure, whose
nature depends on material properties such as the splay and saddle-splay
rigidity moduli of the lamellae. Oily streaks have been
discussed in detail in \cite{schn} and \cite{kleman}, and the latter paper
provides an explanation of the variety of inner structures observed
in these defects. The two main possibilities considered are: (i) the
dislocation lines retain a core structure consisting of a pair of
disclination lines, with modulations runing along their length,
and (ii) they nucleate an array of focal conic domains if the
saddle-splay modulus favours large negative Gaussian curvature.
Our system shows focal domains, and we will therefore use the model
of \cite{kleman} to estimate the line energy of the streaks in
section \ref{summary}.
The following important differences between our
experiments and those of \cite{zap} on cholesteric liquid crystals
remain to be understood: (i) The defect network in our study, unlike
that in \cite{zap}, shows no sign of coarsening in the absence of shear
treatment. Indeed, we observe no spontaneous annealing of the initially
defect-ridden structure unless sheared. Possibly the much larger ratio
of sample thickness to layer spacing in our study is responsible for
this difference; the absence of surface treatment of our plates could
also play a role. (ii) Our network under shear evolves mainly by the
{\em thinning} and eventual {\em disappearance} of lines; we almost never
\footnote{Our video footage contains one possible candidate for a
detachment-and-retraction event.} observe the {\em detachment} and {\em
retraction} events reported in \cite{zap}. It remains unclear whether
detachment events are taking place at scales unresolved by our imaging,
but thinning is clearly an important component of the process.
(iii) The terminal defect density is higher if the imposed shear-rate
is higher. (iv) The
defect line segments are not as straight as those in \cite{zap} appear
to be, and flex rather than rotate rigidly when sheared. This could have
implications for the elasticity of the network. (v) As the sample is
continuously sheared, regions which had initially become homeotropic
display the onset of a square grid like defect structure. A
similar type of shear induced defect structure has been seen by Larson
{\it et al.}\cite{larson2} in a thermotropic liquid crystal. This is due to
disruption of the monodomains in the form of undulation instabilities
\cite{os}. Lastly, we do not see, over the range of shear rate
examined, the multilamellar vesicles (``onions'' \cite{diat}) found in
shear studies of more dilute lamellar phases.
\vspace{1cm}
\begin{figure}
\begin{center}
\epsfig{file=images2.ps,width=14cm}
\end{center}
\caption{Images of the lamellar liquid crystal (concentration of SLES=70w/w \%) with and without particles, as a function of shear time ($t_s$), $\dot{\gamma_s}$=2 s$^{-1}$. Without particles,
$t_s$=(a) 0, (b) 140 and (c) 350 sec. With particles (diameter 9.55
$\mu$m): $t_s$= (d) 0, (e) 140 and (f)= 350 sec. The magnification is
$\times$ 100. The white scale bar in (a) is 100 $\mu$m. The particles
are aggregated at the network nodes in (d) and (f). The dark patch at
the lower right hand corner of (d) is an air bubble.}
\label{fig:images}
\end{figure}
\subsection{The effect of suspended particles on the defect network}
\label{defpart}
The effect of a small concentration (nominally 0.5 \%) of polystyrene +
30\% polybutylmethacrylate spheres of diameter 9.55 $\mu$m on the
behaviour of the defect network was quite dramatic. Note first that the
initial configuration shows a clearly homeotropic anchoring of the
lamellar phase onto the particles. This can be seen explicitly, by
noting the colour variation as a function of angle around the particle
(colour pictures will be sent by the authors upon request, and may be
viewed at http://144.16.75.130/lcrheol/) when it is viewed with a
$\lambda$ (530$\,$nm) plate inserted with its vibration directions at
45$^\circ$ with respect to the extinction position of the crossed
polarisers, and using the colour charts provided in standard books on
polarising microscopy, e.g. \cite{hart}. Before the application of
shear, the particles appear to be well dispersed in the sample.
If we shear at 2 s$^{-1}$ as in the undoped case an oily streak network
appears. Now we find that the network after 140 seconds (Fig.\
\ref{fig:images}e), is much denser than in Fig.\ \ref{fig:images}(b)
with particles aggregated at its nodes. After 350 seconds of shearing
the network is still substantially present (Fig.\ \ref{fig:images}f),
suggesting that the particles are {\it stabilising} it against dissolution. As
the sample is being sheared there is a tendency for the particles to
aggregate and at much longer times, $\sim $900$\,$s, the defect
structure anneals out leaving an almost perfectly aligned sample in
particle-free regions. Similar behaviour was observed for a system doped
with 19$\,\mu$m silica particles. The detailed mechanism for the decay
of the network is unclear, as remarked above, and so therefore is the
reason for its stabilisation. We will, however, offer some speculations
on this subject below.
\section{Viscoelastic properties of the sheared lamellar phase,
and the effect thereon of particulate additives}
\label{viscoel}
\subsection{Shear-treatment and rheometry without particles}
\label{rheonopart}
Our aim is to correlate the visualisation studies of section III with
the mechanical properties of the lamellar phase. Accordingly, we try to
reproduce as far as possible the shear-treatment conditions of section
III and study its effects on the rheometry. The rheometry was carried
out in a parallel-plate geometry\footnote{While this geometry has a
nonuniform shear rate, we prefer it to the (viscometric) cone-and-plate
geometry because the wedge-shaped cross-section of the latter generates
a tilt grain boundary. In addition, there is the danger of particles
getting stuck in the narrow gap near the cone center.} with a gap of
100$\,\mu$m and plate diameter 4$\,$cm. $\dot{\gamma_s}$, the shear-rate of
shear-treatment referred to in all our rheometric studies, is
defined at the outer rim of the parallel-plate geometry.
In the absence of added particles, we subject the lamellar-phase samples
to steady shear for a time $t_s$, stop and measure the
small-amplitude\footnote{strain amplitude of 5 $\times$ 10$^{-3}$; data
for smaller amplitudes were too noisy} dynamic moduli over a wide
frequency range, resume shearing, and repeat this procedure for a total
shearing time of upto an hour and a half. In some cases, we sheared at 1
s$^{-1}$ for half an hour before increasing the shear-rate to 25
s$^{-1}$, measuring dynamic moduli as before. (While we observed
variation of up to 10\% in the dynamic modulii between samples, the
qualitative features and trends remained unchanged.) The results of this
procedure, summarised in Figs. \ref{fig:rheo1} to \ref{fig:rheo4}, are
as follows: (i) The samples initially had $G'(\omega)$, $G''(\omega)$ of about 1500 Pa
and 400 Pa respectively at $\omega = 2 \pi $ rad/sec. At any given
shear-rate (ranging from 1 to 50 s$^{-1}$), the values of $G'$ and $G''$
at fixed $\omega$ show a general tendency to decrease with time,
settling down to a steady-state value at long times, as seen in Figs.\
\ref{fig:rheo1} and \ref{fig:rheo1g2}.\vspace{1cm}
\begin{figure}
\begin{center}
\psfrag{xlabel}{{\small $t_s(sec)$}}
\psfrag{ylabel}{$G'$(Pa)}
\epsfig{file=G1.eps,width=11cm,bbllx=90,bblly=246,
bburx =540,bbury=560}
\end{center}
\vspace{1cm}
\caption{Variation of $G'$ as a function of shear time $t_s$ for various shear rates.
Frequency, $f$=1Hz ($\omega$=2$\pi$ rad/sec), T=25 $^{\circ}$C. }
\label{fig:rheo1}
\end{figure}
The decrease is not monotone, nor
is the steady state truly steady; there are slow oscillations or even a
tendency for the moduli to increase at long times. We suspect the origin
of the oscillations is in the small sample thickness ($d=100\mu$m), or
in irregularities introduced by flow startup. The long-time increase may
be a result of the working-in of defects by shear. As mentioned in
section III, the video clips do show the formation of a grid-like
pattern in the initially homeotropic region as the sample is being
sheared. This may contribute to the long-time increase in the moduli.
\vspace{1cm}
\begin{figure}
\begin{center}
\psfrag{xlabel}{$t_s(sec)$}
\psfrag{ylabel}{$G''$(Pa)}
\epsfig{file=G2.eps,width=11cm,bbllx=90,bblly=246,
bburx =540,bbury=560}
\end{center}
\vspace{1.2cm}
\caption{Variation of $G''$ as a function of shear time $t_s$ for various shear rates.
Frequency, $f$=1Hz, T=25 $^{\circ}$C. }
\label{fig:rheo1g2}
\end{figure}
(ii) The lower the imposed shear-rate, the lower the steady-state values
of $G'$ and $G''$ (at the reference frequency of 2$\pi$ rad/s). (iii)
The overall {\em shapes} of the rheometric spectra $G'(\omega)$ and $G''(\omega)$ (Figs.
\ref{fig:rheo2} and \ref{fig:rheo2g2})
remain invariant under shear-treatment, although the magnitudes of the
moduli, as well as the frequency locations of specific features depend
on the duration and shear-rate of the shear-treatment. The $G''(\omega)$
spectrum has roughly a $\omega^{1/2}$ form at large $\omega$ and is
rather flat at small $\omega$, suggesting the presence of a very large
intrinsic time scale in the system \cite{barnes}. The origin of this
long time scale, as in many soft and ill-characterised solids, remains
unclear \cite{kawa,cates,weitzetal}. A similar variation in the
frequency response of $G'$ under shear treatment has been observed in
block copolymers by Riise {\it et al.}\ \cite{riise}.
\begin{figure}
\begin{center}
\psfrag{xlabel}{$\omega$ (rad/sec)}
\psfrag{ylabel}{$G'$ (Pa)}
\epsfig{file=sles1g1.eps,width=11cm,bbllx=90,bblly=246,bburx =540,bbury=560}
\end{center}
\vspace{1cm}
\caption{$G'(\omega)$ at different shear times $t_s$ for $\dot{\gamma_s}$=1s$^{-1}$, T=25$^\circ$C.
}
\label{fig:rheo2}
\end{figure}
\vspace{1cm}
\begin{figure}
\begin{center}
\psfrag{xlabel}{$\omega$(rad/sec)}
\psfrag{ylabel}{$G''$(Pa)}
\epsfig{file=sles1g2.eps,width=11cm,bbllx=90,bblly=246,bburx =540,bbury=560}
\end{center}
\vspace{1cm}
\caption{$G''(\omega)$ at different shear times $t_s$ for $\dot{\gamma_s}$=1s$^{-1}$, T=25$^\circ$C.
}
\label{fig:rheo2g2}
\end{figure}
(iv) The $G'(\omega)$ curves for different durations of shear-treatment can be
made to collapse onto the $t_s=0$ (no shear-treatment) curve (Fig.
\ref{fig:rheo3}) if log($G'(\omega)/G'_o$) is plotted against
log($\omega/\omega_o$), where $G'_o$ and $\omega_o$ are, respectively, a
reference modulus and a characteristic frequency that shift. We find
that both $G'_o$ and $\omega_o$ decrease monotonically (Fig.
\ref{fig:go}) as $t_s$ increases. This is pleasingly consistent with the
idea of an elasticity arising from a coarsening defect network (see the
discussion in section \ref{summary}). We emphasize that the functions
$G'_o(t_s)$ and $\omega_o(t_s)$ are unique only upto arbitrary
multiplicative factors, as the data collapse can be achieved if the $G'(\omega)$
and $\omega$ for no shear-treatment are multiplied by the same
respective factors. For reasons we do not understand, however, the data
on $G''(\omega)$ do not show a satisfactory collapse.
\vspace{1cm}
\begin{figure}
\psfrag{xlabel}{log($ \omega/\omega_o$)}
\psfrag{ylabel}{log$(G'/G'_o)$}
\begin{center}
\epsfig{file=gcal.eps,width=11cm,bbllx=90,bblly=246,bburx =540,bbury=560}
\end{center}
\vspace{1cm}
\caption{The $G'(\omega)$ data of Fig. \ref{fig:rheo2} for various shear
times collapsed onto the zero shear treatment ($t_s=0$) curve.
}
\label{fig:rheo3}
\end{figure}
\vspace{1cm}
\begin{figure}
\begin{center}
\epsfig{file=go.eps,width=10cm}
\end{center}
\caption{Variation of the scale factors $G'_o$ and $\omega_o$ as functions
of shear time.
}
\label{fig:go}
\end{figure}
(v) After shearing at 1
s$^{-1}$ for about 30 minutes, reaching a quasisteady $G'$ of about 400
Pa, the shear-rate was increased abruptly to 25 s$^{-1}$. It can be seen
from Fig.\ \ref{fig:rheo4} that $G'$ picks up rapidly, showing a
tendency to climb back to the value that it would have had if sheared
directly at 25 s$^{-1}$. Thus, shear on the one hand reduces an
initially high modulus, but is equally capable of increasing an
initially {\em low} modulus. Despite some apparent history-dependence,
we speculate that shear-treatment imposed for a long enough time does
produce a structure with a well-defined dynamic-modulus spectrum
depending only on the value of the shear-rate. We propose to study the
rheological response to small oscillations about the steady flow, not by
stopping the flow, which inevitably introduces artifacts. Linear
response about {\em steady} flow is the true measure of the properties
of the nonequilibrium steady state we are studying, namely the sheared
lamellar phase.
\vspace{1cm}
\begin{figure}
\begin{center}
\psfrag{ylabel}{$t_s$(sec)}
\psfrag{xlabel}{$G'$(Pa)}
\epsfig{file=g251.eps,width=11cm,bbllx=90,bblly=246,bburx =540,bbury=560}
\end{center}
\vspace{2cm}
\caption{ $G^{\prime}$ as a function of shear time $t_s$, $f$=1Hz, T=25 $^{\circ}$C. Note the jump in the modulus when $\dot{\gamma_s}$ was changed abruptly from
1 s$^{-1}$ to 25 s$^{-1}$.}
\label{fig:rheo4}
\end{figure}
\vspace{2cm}
To compare the rheometry with the light microscopy we have measured the
transmitted light intensity as a function of shear time $t_s$
(Fig.\ \ref{fig:slesig}).
Since homeotropically aligned regions appear dark,
this measures the density of defects. As seen in the figure,
the decay of the transmitted intensity is in clear qualitative agreement
with the decrease in the storage modulus under
similar conditions of shear treatment ($d = 100 \mu$m , $\dot{\gamma_s}
= 2$ s$^{-1}$), supporting the conjecture that the elasticity arises
primarily from the defect network. A similar correlation has been noted
by Larson {\it et al.} \cite{larson2} in thermotropics.
\vspace{1cm}
\begin{figure}
\begin{center}
\epsfig{file=slesig.eps, width=10cm}
\end{center}
\caption{ Transmitted intensity and $G'$ of the lamellar sample
(SLES=73.2 w/w \%) as functions of shear time $t_s$, $\dot{\gamma_s}$=2 s$^{-1}$,
$d=100 \mu$m.
}
\label{fig:slesig}
\end{figure}
\vspace{1.5cm}
\subsection{Rheometry with particles}
\label{rheopart}
The effect of particulate additives on the viscoelastic properties was
studied by adding monodisperse polystyrene + 30\% polybutylmethacrylate
spheres (9.55 $\pm$ 0.44 $\mu$m diameter, 0.5 \% volume fraction ) to
the lamellar phase sample. The storage modulus is consistently higher in
the presence of particles (Fig.\ \ref{fig:g1p10_30}) but approaches that
of a particle-free sample at the longest times ($\sim$ 10 min). A
similar trend is observed for $G''$ (Fig. \ref{fig:g2p10_30}), although
the data has more scatter. As mentioned in section III, the particles
tend to coagulate at very long times, and the anchoring of the defect
network is lost; this too is in keeping with the ultimate decrease of
the moduli to those of the particle-free sample. The variation in the
dynamic modulii between samples was greater in the presence of particles
than without, roughly 20\%, though part of it could be attributed to the
difficulty in maintaining a constant particle concentration in all
samples, but all the trends were reproducible.
\vspace{1cm}
\begin{figure}
\begin{center}
\psfrag{xlabel}{$t_s$(sec)}
\psfrag{ylabel}{$G'$(Pa)}
\epsfig{file=g1p10_30.eps,width=11cm,bbllx=90,bblly=246,bburx =540,bbury=560}
\end{center}
\vspace{1.5cm}
\caption{Variation of $G'$ of the lamellar sample (SLES=70w/w \%)
as a function of shear-time $t_s$, with and without particles, $\dot{\gamma_s}$=2 s$^{-1}$, $f$=1 Hz and T=25 $^\circ$C.
}
\label{fig:g1p10_30}
\end{figure}
\begin{figure}
\begin{center}
\psfrag{xlabel}{$t_s$(sec)}
\psfrag{ylabel}{$G''$(Pa)}
\epsfig{file=g2p10_30.eps,width=11cm,bbllx=90,bblly=246,bburx =540,bbury=560}
\end{center}
\vspace{2cm}
\caption{Variation of $G''$ of the lamellar sample as a function of
shear-time $t_s$, with and without particles,
$\ \dot{\gamma_s}$=2 s$^{-1}$, $f$=1 Hz and T=25 $^\circ$C.
}
\label{fig:g2p10_30}
\end{figure}
\section{Summary and analysis}
\label{summary}
Our video microscopy studies have shown that sheared lyotropic lamellar
phases display a striking defect network structure made up of oily
streaks. These defects anneal away slowly under shear treatment, by a
route which appears to be the thinning and disappearance of lines. The
addition of a small concentration of suspended particulate impurities
greatly retards the decay of this defect network. Previous studies of
particulate additives in a lyotropic gel system showed a large increase
in the rigidity modulus \cite{shouche}. It was conjectured there that
surfactant molecules adsorb on the particles, resulting in direct
interparticle bridges and stress transmission. The oily streak network
we observe is a far more likely candidate for such a stress-transmitting
structure.
We present a rough theoretical estimate of the rigidity of the
network. As in \cite{zap}. we argue that a network of oily-streak lines with
mesh size $\ell$ and tension $\Gamma$ should have a shear modulus
$G' \sim \Gamma / \ell^2$. The value of $\Gamma$ depends on
the internal structure of the oily streaks.
Our images of the oily-streak network (Fig. \ref{fig:images}a, for example,
or other images not presented in the paper)
clearly show the presence of focal domain arrays. We thus estimate the
line tension using the model of \cite{kleman} rather than the
striated dislocation line pairs of \cite{schn}, and find reasonable
agreement with our measurements.
In the focal domain model \cite{kleman}, there are two contributions
to the line energy: a mean-curvature energy piece logarithmic in the width
of the streak, and a major contribution from layer compressions in a
narrow region interpolating between the domains and the bulk undistorted
layering. The curvature term turns out to be negligibly small here,
and we shall ignore it. For an isolated oily streak in a sample of
thickness $h$, the compression energy is found \cite{kleman} to be of
the form
\begin{equation}
\label{streakenergy}
{\cal E} = \bar{B} h^2 \left[{e \over {\sqrt{1 - e^2}}} -
{a \over h} \right]^5
\end{equation}
times a geometrical factor of order unity. Here $\bar{B}$ is the
layer compression modulus at constant chemical
potential of surfactant, and $e$ and $a$ are respectively the eccentricity
and major axis of the ellipses constituting the domains.
In a {\em network} of oily streaks, as distinct from an individual streak,
we expect this dependence on the thickness $h$ to be screened at the scale of
the mesh size $\ell$, i.e., $h$ should be replaced by $\ell$ in
(\ref{streakenergy}). Our observations suggest $e \simeq 0.5$ to $0.6$,
$a \simeq 5$ to $ 10 \mu$m, and $\ell \simeq 50 \mu$m, giving an energy
per unit length ${\cal E} \sim (10^{-3}$ to $10^{-2}) \bar{B} \ell^2$ and hence a
contribution to the shear modulus which scales with but is much smaller in magnitude than the layer compression modulus:
$G' \sim (10^{-3}$ to $10^{-2})\bar{B}$. For $\bar{B} \sim 10^5$ Pa,
this gives $G' \sim 10^2$ to $10^3$ Pa, which is in the right range.
The fifth power in (\ref{streakenergy}) does of course make this estimate
rather sensitive to the precise values of the parameters involved.
While the modulus of an initially defect-ridden state decreases
substantially with shear-treatment, the data for the frequency-dependent
storage modulus after different amounts of shear-treatment can be scaled
onto a single curve, Fig. \ref{fig:rheo3}. This data collapse can be rationalised as
follows: At each stage of the shear-treatment, the network has a
mesh size $\ell$, a characteristic inverse timescale $\omega_o$
(the relaxation rate of structures at the scale $\ell$),
and a characteristic modulus $G'_o$ (the shear modulus of an elementary
cell of the network). It is natural to assume that $\omega_o$
decreases as $\ell$ increases, since a coarser mesh
should relax more slowly, and that $G'_o = \Gamma / \ell^2$,
where $\Gamma$ is the line tension.
This suggests a frequency-dependent shear modulus
$G'(\omega) = G'_o F(\omega/\omega_o)$, which should account
for the data collapse provided $\Gamma$ either does
not evolve under coarsening or else evolves in a
manner determined entirely by $\ell$.
The scaling is thus quite easily understood in the focal domains
model \cite{kleman}, in which $\Gamma$ is independent of the
thickness of the lines for a coarse network (see our estimates
in the previous paragraph)\footnote{In the dislocation model \cite{schn,kleman},
however, $\Gamma$ depends on the Burgers-vector content of the line,
i.e., on its thickness, and will thus decrease as $\ell$ increases,
since our oily streaks appear to thin under shear
treatment. We do not consider this case here since our images
show the focal domains of \cite{kleman}.}.
Although the shear modulus of an initially highly rigid (and therefore
presumably very defect-ridden) state decreases gradually with time as
the state is sheared steadily, an initially low-modulus state (produced by
shear-treatment at low rates) becomes more rigid upon shearing at a high
rate. This tells us that the observed steady-state defect structure is
the result of a competition between the working-in and the working-out
of defects by shear. We conjecture that there is a unique structural and
rheological state associated with a given shear-rate rather than a given
shear-history. The data does show some history-dependence, but we
suspect that this is a transient. Indeed, it would be most appropriate
to study the properties of the sheared lamellar phase not by stopping
the flow but rather by treating it as a nonequilibrium steady state, and
measuring its linear rheological response to a small oscillatory
component superposed on the steady shear. We are currently pursuing such
studies.
\section*{Acknowledgement}
\label{ack}
We thank Prof A. K. Sood for extensive access to imaging
facilities, and R. Adhikari for useful discussions.
Partial funding for this project, including a
Project Associateship for GB, came from Unilever Research India.
|
\section{Introduction}
\label{sec:intr}
It is generally believed that supersymmetric ${\rm SU}(N)$ matrix models
in $d=9$
dimensions admit exactly one normalizable zero-energy solution for each $N>1$,
while they admit none for all other dimensions for which the models can be
formulated, i.e., for $d=2,3,5$. For various approaches to this problem see
e.g. \cite{c1a}--\cite{c1l}.
In this article, we would like to summarize (and slightly modify/extend) what
is known about the behaviour of ${\rm SU}(2)$ zero--energy solutions far out at
infinity in (and near) the space of configurations where the bosonic potential
(the trace of all commutator--squares) vanishes. Based on some early 'negative'
result concerning $N=2,\, d=2$ (that used rather different
techniques/arguments; see \cite{c1a, c4a})
we started our investigation of the asymptotic behaviour, in the fall of 1997,
with a Hamiltonian Born--Oppenheimer analysis of that $N=2,\, d=2$ case.
Some months later,
we realized that the rather complicated Hamiltonian analysis (Halpern
and Schwartz \cite{c1h} had, in the meantime, derived the form of the wave
function for $d=9$ near $\infty$, by Hamiltonian Born--Oppenheimer methods)
can be replaced by a
simple first order analysis, using only the first order operators $Q$,
and first order perturbation theory. One finds that asymptotically
normalizable, ${\rm SU}(2)$ and ${\rm SO}(d)$ invariant, wave functions do
not exist for
$d=2,3$, and $5$, in contrast to $d=9$, where there is exactly one.
We close these introductory words by recalling that the models discussed below
arise in at least 3 somewhat different ways: As supersymmetric extensions of
regulated membrane theories in $d+2$ space--time dimensions
\cite{c2a, c4a},
as reductions (to $0+1$ dimension) of $d+1$ dimensional Super Yang Mills
theories \cite{c3a}--\cite{c3c},
and, for $d=9$, as a description of the dynamics of D--0 branes in superstring
theory, \cite{c5a, c5b}. In this physical interpretation, the existence of a
normalizable zero--energy solution is an important consistency requirement.
The paper is organized as follows. In Section \ref{sec:mod} we recall the
definition of the models, and in Section \ref{sec:res} we state our main
result about zero--modes. The proof is given in Section \ref{sec:pr} and
Appendix 1. We suggest to skip Subsection \ref{sec:pr5} and Appendix 1 at a
first reading. As a warm--up the
reader is advised to read Appendix 2, where a simpler model is
treated by the same method.
\section{The models}
\label{sec:mod}
The configuration space of the bosonic degrees of freedom is $X=\mathbb R^{3d}$
with coordinates
\begin{equation}
q=(\vec{q_1},\ldots,\vec{q_d})=(q_{sA})_{\substack{s=1,\ldots, d\\A=1,2,3}}\;.
\nonumber
\end{equation}
To describe the fermionic degrees of freedom let, as a preliminary,
\begin{equation}
\gamma^i=(\gamma^i_{\alpha\beta})_{\alpha,\beta=1,\ldots,s_d}\;,\qquad
(i=1,\ldots,d)\;,
\label{e1.1aa}
\end{equation}
be the {\it real} representation of smallest dimension, called $s_d$, of the
Clifford algebra with $d$ generators:
$\{\gamma^s,\gamma^t\}=2\delta^{st}{\sf 1}\mkern-5.0mu{\rm I}$. On the representation space,
${\rm Spin}(d)$ is realized through matrices
$R\in{\rm SO}(s_d)$, so that we may view
\begin{equation}
{\rm Spin}(d)\hookrightarrow{\rm SO}(s_d)\;,
\label{e1.1a}
\end{equation}
as a simply connected subgroup. We recall that
\begin{equation*}
s_d=\cases 2^{[d/2]}\;,& d=0,1,2 \mod 8\;,\\
2^{[d/2]+1}\;,& \hbox{otherwise}\;,
\endcases
\end{equation*}
where $[\cdot]$ denotes the integer part. We then consider the Clifford
algebra with $s_d$ generators and its irreducible representation on
$\mathcal C=\mathbb C^{2^{s_d/2}}$. On
$\mathcal C^{\otimes 3}$ the Clifford generators
\begin{equation}
(\vec{\Theta}_1,\ldots,\vec{\Theta}_{s_d})=
(\Theta_{\alpha A})_{\substack{\alpha=1,\ldots,s_d\\A=1,2,3}}
\nonumber
\end{equation}
are defined, satisfying
$\{\Theta_{\alpha A} ,\Theta_{\beta B}\}=\delta_{\alpha\beta}\,\delta_{AB}$.
The Hilbert space, finally, is
\begin{equation}
\mathcal H={\rm L}^2(X,\mathcal C^{\otimes 3})\;.
\label{e1.0}
\end{equation}
There is a natural representation of ${\rm SU}(2)\times{\rm Spin}(d)\ni(U,R)$
on $\mathcal H$. In fact, the group acts naturally on $X$ through its
representation ${\rm SO}(3)\times{\rm SO}(d)$ (which we also denote by
$(U,R)$). On $\mathcal C^{\otimes 3}$
we have the representation $\mathcal R$ of ${\rm Spin}(s_d)\ni R$
\begin{equation}
\mathcal R(R)^*\Theta_{\alpha A}\mathcal R(R)=\widetilde{R}_{\alpha\beta}
\Theta_{\beta A}\;,
\label{e1.2}
\end{equation}
where $\widetilde{R}=\widetilde{R}(R)$ is its ${\rm SO}(s_d)$ representation.
Through ${\rm SO}(s_d)={\rm Spin}(s_d)/\mathbb Z_2$ and (\ref{e1.1a}) we have
\begin{equation}
{\rm Spin}(d)\hookrightarrow{\rm Spin}(s_d)\;,
\label{e1.2a}
\end{equation}
and thus a representation $\mathcal R$ of ${\rm Spin}(d)$. The representation
$\mathcal U$ of ${\rm SU}(2)\ni U$ on $\mathcal C^{\otimes 3}$ is
characterized by
$\mathcal U(U)^*\Theta_{\alpha A}\mathcal U(U)=U_{AB}\Theta_{\alpha B}$.
We shall now restrict to $d=2,\,3,\,5,\,9$, where $s_d=2,\,4,\,8,\,16$, the
reason being that in these cases
\begin{equation}
s_d=2(d-1)\;,
\label{e1.1}
\end{equation}
whereas $s_d$ is strictly larger otherwise. Eq. (\ref{e1.1}) is essential
for the algebra (\ref{eq2}) below \cite{c3c}.
The supercharges, acting on $\mathcal H$, are given by the $s_d$ hermitian
operators
\begin{equation}
Q_\beta= \vec{\Theta}_\alpha \cdot \bigl(
-{\rm i} \gamma^t_{\alpha\beta} \vec{\nabla}_t + \frac{1}{2} \,
\vec{q}_s \times \vec{q}_t\,\gamma^{st}_{\beta\alpha} \bigr)\;,
\qquad(\beta=1,\ldots, s_d)\;,
\nonumber
\end{equation}
where $\gamma^{st}=(1/2)(\gamma^s \gamma^t -\gamma^t \gamma^s)$. These
supercharges transform as scalars under
${\rm SU}(2)$ transformations generated by
\begin{equation}
J_{AB}=-{\rm i}(q_{sA}\partial_{sB}-q_{sB}\partial_{sA})-\frac{{\rm i}}{2}
(\Theta_{\alpha A}\Theta_{\alpha B}-\Theta_{\alpha B}\Theta_{\alpha A})
\equiv L_{AB}+M_{AB}\;,
\nonumber
\end{equation}
resp. as vectors in $\mathbb R^{s_d}$ under ${\rm Spin}(d)$ transformation
generated by
\begin{equation}
J_{st} = -{\rm i}(\vec{q}_s\cdot\vec{\nabla}_t -
\vec{q}_t\cdot\vec{\nabla}_s)-\frac{{\rm i}}{4}
\,\vec{\Theta}_\alpha \gamma^{st}_{\alpha\beta} \vec{\Theta}_\beta
\equiv L_{st}+M_{st}\;.
\nonumber
\end{equation}
The anticommutation relations of the supercharges are
\begin{equation}
\bigl\{Q_\alpha, Q_\beta\bigr\}=\delta_{\alpha\beta}H+
\gamma^t_{\alpha\beta}q_{tA}\varepsilon_{ABC}J_{BC}\;.
\label{eq2}
\end{equation}
Here, $H$ is the Hamiltonian
\begin{equation}
H = - \sum_{s=1}^9 \, \vec{\nabla}_s^2 + \sum_{s<t}
\bigl( \vec{q}_s \times \vec{q}_t \bigr)^2 +
{\rm i} \vec{q}_s \cdot\bigl(\vec{\Theta}_\alpha \times \vec{\Theta}_\beta\bigr)
\, \gamma^s_{\alpha\beta} \; ,
\label{e1.4}
\end{equation}
which commutes with both $J_{AB}$ and $J_{st}$. The question we address is the
possibility of a normalizable state $\psi\in\mathcal H$ with zero energy,
i.e., with $H\psi=0$, which is a singlet w.r.t. both ${\rm SU}(2)$ and
${\rm Spin}(d)$. Note that on ${\rm SU}(2)$ invariant states
$H=2Q_\beta^2\ge 0$ and in fact the energy spectrum is
(\cite{c4b}) $\sigma(H)=[0,\infty)$. Equivalently, we look for zero-modes
\begin{equation}
Q_\beta\psi=0\;,\qquad(\beta=1,\ldots, s_d)\;.
\nonumber
\end{equation}
\section{Results}
\label{sec:res}
The potential $\sum_{s<t}( \vec{q}_s \times \vec{q}_t)^2$ vanishes on the
manifold
\begin{equation}
\vec{q}_s = r \vec{e} E_s
\nonumber
\end{equation}
with $r>0$ and $\vec{e}{\,}^2=\sum_s E_s^2 = 1$. The dimension of the manifold is
$1+2+(d-1)=3d-2(d-1)$. Points in a conical neighborhood
of the manifold can be expressed in terms of tubular (or ``end--point'')
coordinates \cite{b2}
\begin{equation}
\vec{q}_s = r \vec{e} E_s + r^{-1/2}\vec{y}_s
\label{e2.00}
\end{equation}
with
\begin{equation}
\vec{y}_s\cdot \vec{e} = 0\;,\qquad\vec{y}_s E_s =\vec{0}\;.
\label{e2.1}
\end{equation}
A prefactor
has been put explicitely in front of the transversal coordinates $\vec{y}_s$,
so as to anticipate the length scale $r^{-1/2}$ of the ground state.
The change
\begin{equation}
(\vec{e},E, y)\mapsto(-\vec{e},-E, y)
\label{e2.0}
\end{equation}
does not affect $\vec{q}_s$. Rather than identifying the two coordinates for
$\vec{q}_s$,
we shall look for states which are even under the antipode map
(\ref{e2.0}).
\smallskip
We can now describe the structure of a putative ground state.
\smallskip\noindent
{\bf Theorem} {\it Consider the equations $Q_\beta\psi=0$ for a formal power
series solution near $r=\infty$ of the form
\begin{equation}
\psi=r^{-\kappa}\sum_{k=0}^\infty r^{-\frac{3}{2}k}\psi_k\;,
\label{e2.4}
\end{equation}
where: $\psi_k=\psi_k(\vec{e},E,y)$ is square integrable w.r.t.
${\rm d} e\,{\rm d} E\,{\rm d} y$;
\newline
\hphantom{where:} $\psi_k$ is ${\rm SU}(2)\times{\rm Spin}(d)$
invariant;\newline
\hphantom{where:} $\psi_0\neq 0$.\newline
Then, up to linear combinations,
\begin{itemize}
\item{d=9}: The solution is unique, and $\kappa=6$;
\item{d=5}: There are three solutions with $\kappa=-1$ and one
with $\kappa=3$;
\item{d=3}: There are two solutions with $\kappa=0$;
\item{d=2}: There are no solutions.
\end{itemize}
All solutions are even under the antipode map (\ref{e2.0}),
\begin{equation*}
\psi_k(\vec{e},E,y)=\psi_k(-\vec{e},-E,y)\;,
\end{equation*}
except for the state $d=5,\,\kappa=3$, which is odd.}
\medskip\noindent
{\bf Remarks} 1. The equation $Q_\beta\psi=0$ can be viewed as an ordinary
differential equation in $z=r^{3/2}$ for a function taking values in
${\rm L}^2({\rm d} e\,{\rm d} E\,{\rm d} y,\mathcal C^{\otimes 3})$ (see eq. (\ref{e3.00})
below). It turns out that
$z=\infty$ is a singular point of the second kind \cite{b1}. In such a
situation the series (\ref{e2.4}) is typically asymptotic to a true solution,
but not convergent.
\par\noindent
2. The integration measure is
${\rm d} q={\rm d} r\cdot r^2{\rm d} e\cdot r^{d-1}{\rm d} E\cdot r^{-\frac{1}{2}\cdot 2(d-1)}{\rm d} y=
r^2{\rm d} r\,{\rm d} e\,{\rm d} E\,{\rm d} y$. The wave function (\ref{e2.4}) is square
integrable
at infinity if
$\int^\infty{\rm d} r\,r^2(r^{-\kappa})^2<\infty$, i.e., if $\kappa>3/2$. The
theorem is consistent with the statement according to which {\bf only} for
$d=9$ a (unique) normalizable ground state for (\ref{e1.4}) (which would
have to be even) is possible.
\par\noindent
3. Note that the connection of matrix models with supergravity requires the
zero--energy solutions to be ${\rm Spin}(d)$ singlets only for $d=9$.
\bigskip
The case $d=2$ can be dealt with immediately. We may assume
$\gamma^2 =\sigma_3, \,
\gamma^1 = \sigma_1$ (Pauli matrices), so that
\begin{equation*}
M_{12}=\frac{{\rm i}}{2}\Theta_{1A}\Theta_{2A}\;,
\end{equation*}
with commuting terms. Since, for each $A=1,2,3,\, (\Theta_{1A}\Theta_{2A})^2
=-1/4$, we see that $M_{12}$ has spectrum in $\mathbb Z/2 +1/4$. Given that
$L_{12}$ has spectrum $\mathbb Z$, no state with $J_{12}\psi=0$ is possible.
We mention \cite{c1a} that, more generally, for $d=2$ no
normalizable ${\rm SU}(2)$ invariant ground state exists.
The proof of the theorem will thus deal with $d=9,5,3$ only.
\section{Proof}
\label{sec:pr}
We shall first derive the power series expansion of the supercharges
$Q_\beta$. To this end we note that
\begin{eqnarray}
\frac{\partial}{\partial q_{tA}}&=&
r^{1/2}(\delta_{st}-E_sE_t)(\delta_{AB}-e_Ae_B)
\frac{\partial}{\partial y_{sB}}
\label{e3.000}\\
&&+r^{-1}[e_AE_t(r\frac{\partial}{\partial r}+
\frac{1}{2}y_{sB}\frac{\partial}{\partial y_{sB}})+
{\rm i} e_BE_tL_{BA}+{\rm i} e_AE_sL_{st}]
+{\rm O}(r^{-5/2})\;,\nonumber
\end{eqnarray}
with the remainder not containing derivatives w.r.t. $r$ (see Appendix 1 for
derivation). This yields
\begin{equation}
Q_\beta=r^{1/2}Q_\beta^0+
r^{-1}(\widehat{Q}_\beta^1r\frac{\partial}{\partial r}+Q_\beta^1)
+r^{-5/2}Q_\beta^2+\ldots
\label{e3.00}
\end{equation}
with $r$--independent operators
\begin{eqnarray*}
Q_\beta^0&=&-{\rm i}\Theta_{\alpha A}\gamma^t_{\alpha\beta}(\delta_{st}-E_sE_t)
(\delta_{AB}-e_Ae_B)\frac{\partial}{\partial y_{sB}}+
\vec{\Theta}_\alpha
\cdot(\vec{e}\times\vec{y}_t)E_s\gamma^{st}_{\beta\alpha}\;,\\
\widehat{Q}_\beta^1&=&
-{\rm i} (\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^t_{\alpha\beta}E_t\;,\\
Q_\beta^1&=&\Theta_{\alpha A}\gamma^t_{\alpha\beta}
\bigl(e_BE_tL_{BA}+
e_AE_sL_{st}-\frac{{\rm i}}{2}\,e_AE_ty_{sB}\frac{\partial}{\partial y_{sB}})+
\frac{1}{2}\vec{\Theta}_\alpha\cdot(\vec{y}_s\times\vec{y}_t)
\gamma^{st}_{\beta\alpha}\;.
\end{eqnarray*}
The explicit expressions of $Q_\beta^n,\, (n\ge 2)$ will not be needed. We
then equate coefficients of powers of $r^{-3/2}$ in the equation
$Q_\beta\psi=0$ with the result
\begin{eqnarray}
Q_\beta^0\psi_n+
\bigl(-(\kappa+\frac{3}{2}(n-1))\widehat{Q}_\beta^1+Q_\beta^1\bigr)
\psi_{n-1}+Q_\beta^2\psi_{n-2}+\ldots+Q_\beta^n\psi_0&=&0\;,\nonumber\\
\qquad(n&=&0,1,\ldots)\;.
\label{e3.0}
\end{eqnarray}
\subsection{The equation at $n=0$}
\label{sec:pr1}
The equation at $n=0$,
\begin{equation}
Q_\beta^0\psi_0=0\;,
\label{e3.1}
\end{equation}
admits precisely the (not necessarily ${\rm SU}(2)\times{\rm Spin}(d)$
invariant) solutions
\begin{equation}
\psi_0(\vec{e},E, y)={\rm e}^{-\sum_s\vec{y}_s{\,}^2/2}|F(E,\vec{e})\rangle\;,
\label{e3.2}
\end{equation}
(with $\vec{y}$ restricted to (\ref{e2.1})), where the fermionic states
$|F(E,\vec{e})\rangle$ can be described as follows:
Let $\vec{n}_\pm$ be two complex vectors satisfying
$\vec{n}_+\cdot\vec{n}_-=1,\,\vec{e}\times \vec{n}_\pm=\mp{\rm i}\vec{n}_\pm$ (and
hence
$\vec{n}_\pm\cdot\vec{n}_\pm=0$, $\vec{n}_+\times\vec{n}_-=-{\rm i}\vec{e}$\,). For
any vector $v\in\mathbb R^{s_d}$
we may introduce $\vec{\Theta}(v)=\vec{\Theta}_\alpha v_\alpha$, as well as
fermionic operators $\vec{\Theta}(v)\cdot\vec{n}_\pm$ satisfying
canonical anticommutation relations:
\begin{equation}
\bigl\{\vec{\Theta}(u)\cdot\vec{n}_+,\vec{\Theta}(v)\cdot\vec{n}_-\bigr\}
=u_\alpha v_\alpha\;,\qquad
\bigl\{\vec{\Theta}(u)\cdot\vec{n}_\pm,\vec{\Theta}(v)\cdot\vec{n}_\pm\bigr\}
=0\;.
\nonumber
\end{equation}
Then, $|F(E,\vec{e})\rangle$ is required to obey
\begin{equation}
\vec{\Theta}(v)\cdot\vec{n}_\pm|F(E,\vec{e})\rangle=0
\qquad\hbox{for}\qquad
E_s\gamma^sv=\pm v\;.
\label{e3.4}
\end{equation}
To prove the above, let us note that
\begin{eqnarray}
&&\hskip 1cm\bigl\{Q_\alpha^0,Q_\beta^0\bigr\}=
\delta_{\alpha\beta}H^0+
\gamma^t_{\alpha\beta}E_t\varepsilon_{ABC}M_{AB}e_C\;,
\label{e3.4a}\\
H^0&=&\bigl[-(\delta_{st}-E_sE_t)(\delta_{AB}-e_Ae_B)
\frac{\partial}{\partial y_{sA}}\frac{\partial}{\partial y_{tB}}
+\sum_s\vec{y}_s^2\bigr]+
{\rm i} E_s\gamma^s_{\alpha\beta}\vec{e}\cdot
\bigl(\vec{\Theta}_\alpha \times \vec{\Theta}_\beta\bigr)\nonumber\\
&\equiv& H^0_B+H^0_F\;.\nonumber
\end{eqnarray}
By contracting eq. (\ref{e3.4a}) against $\delta_{\alpha\beta}$, resp.
$\gamma^t_{\alpha\beta}E_t$ we see that the equations (\ref{e3.1}) are
equivalent to the pair of equations
\begin{equation}
H^0\psi_0=0\;,\qquad \varepsilon_{ABC}M_{AB}e_C\psi_0=0\;.
\label{e3.4b}
\end{equation}
Here, $H^0_B$ is a harmonic oscillator in $2(d-1)$ degrees of
freedom, with orbital ground state wave function
${\rm e}^{-\sum_s\vec{y}_s^2/2}$ and energy $2(d-1)$. On the other hand,
\begin{eqnarray}
H^0_F&=&-E_s\gamma^s_{\alpha\beta}
\bigl((\vec{\Theta}_\alpha\cdot\vec{n}_+)(\vec{\Theta}_\beta\cdot\vec{n}_-)-
(\vec{\Theta}_\alpha\cdot\vec{n}_-)(\vec{\Theta}_\beta\cdot\vec{n}_+)\bigr)
\nonumber\\
&=&-s_d+
2P^+_{\alpha\beta}(\vec{\Theta}_\alpha\cdot\vec{n}_-)
(\vec{\Theta}_\beta\cdot\vec{n}_+)
+2P^-_{\alpha\beta}(\vec{\Theta}_\alpha\cdot\vec{n}_+)
(\vec{\Theta}_\beta\cdot\vec{n}_-)\;,
\label{e3.5}
\end{eqnarray}
where we used the spectral decomposition $E_s\gamma^s=P^+-P^-$. In view of
(\ref{e1.1}), the equation $H^0\psi_0=0$ is fulfilled iff the fermionic
state is annihilated by the last two positive terms in (\ref{e3.5}), i.e., if
(\ref{e3.4}) holds. The second equation (\ref{e3.4b}) is now also satisfied,
since
\begin{eqnarray}
\frac{1}{2}\varepsilon_{ABC}M_{AB}e_C&=&
-\frac{{\rm i}}{2}\vec{e}\cdot
\bigl(\vec{\Theta}_\alpha \times \vec{\Theta}_\alpha \bigr)\nonumber\\
&=&\frac{1}{2}
\bigl((\vec{\Theta}_\alpha\cdot\vec{n}_+)(\vec{\Theta}_\alpha\cdot\vec{n}_-)-
(\vec{\Theta}_\alpha\cdot\vec{n}_-)(\vec{\Theta}_\alpha\cdot\vec{n}_+)\bigr)
\nonumber\\
&=&P^-_{\alpha\beta}(\vec{\Theta}_\alpha\cdot\vec{n}_+)
(\vec{\Theta}_\beta\cdot\vec{n}_-)-
P^+_{\alpha\beta}(\vec{\Theta}_\alpha\cdot\vec{n}_-)
(\vec{\Theta}_\beta\cdot\vec{n}_+)
\label{e3.5c}
\end{eqnarray}
annihilates $|F(E,\vec{e})\rangle$.
\par
\subsection{${\rm SU}(2)\times{\rm Spin}(d)$ invariant states}
\label{sec:pr2}
We recall that the representation $\mathcal R[\cdot]$ of ${\rm Spin}(d)$ on
$\mathcal H$ is
$(\mathcal R[R]\psi)(q)=\mathcal R(R)(\psi(R^{-1}q))$, where $\mathcal R(R)$
acts on $\mathcal C^{\otimes 3}$. Similarly for ${\rm SU}(2)$.
The invariant solutions among (\ref{e3.2}) are thus
those which satisfy
\begin{equation}
\mathcal U(U)|F(E,\vec{e})\rangle =|F(E,U\vec{e})\rangle\;,\qquad
\mathcal R(R)|F(E,\vec{e})\rangle =|F(RE,\vec{e})\rangle\;,
\label{e3.5a}
\end{equation}
for $(U,R)\in{\rm SU}(2)\times{\rm Spin}(d)$. These states are in bijective
correspondence to states invariant under the `little group'
$(U,R)\in{\rm U}(1)\times{\rm Spin}(d-1)$, i.e., to states
$|F(E,\vec{e})\rangle$ satisfying
\begin{equation}
\mathcal U(U)|F(E,\vec{e})\rangle =|F(E,\vec{e})\rangle\;,\qquad
\mathcal R(R)|F(E,\vec{e})\rangle =|F(E,\vec{e})\rangle\;,
\label{e3.5b}
\end{equation}
for some arbitrary but fixed $(E,\vec{e})$ and all $U,\,R$ with
$U\vec{e}=\vec{e},\,RE=E$. The first relation holds on all of (\ref{e3.4}). In
fact the generator (\ref{e3.5c}) of the group $\mathcal U(U)$ of rotations
$U$ about $\vec{e}$ annihilates $|F(E,\vec{e})\rangle$, as we just saw. To
discuss the second relation
(\ref{e3.5b}) we note that the generators of ${\rm Spin}(d-1)$ (i.e., of the
fermionic rotations about $E$), are $M_{st}U_sV_t$ with $U_sE_s=V_sE_s=0$. We
write
$M_{st}=M_{st}^{\perp}+M_{st}^{\parallel}$,
where
\begin{equation}
M_{st}^{\perp}=-({\rm i}/2)(\vec{\Theta}_\alpha\cdot\vec{n}_+)
\gamma^{st}_{\alpha\beta}(\vec{\Theta}_\beta\cdot\vec{n}_-)\;,\qquad
M_{st}^{\parallel}=-({\rm i}/4)(\vec{\Theta}_\alpha\cdot\vec{e}\,)
\gamma^{st}_{\alpha\beta}(\vec{\Theta}_\beta\cdot\vec{e}\,)\;,
\label{e3.8b}
\end{equation}
and remark that, by a computation similar to (\ref{e3.5c}),
$M_{st}^{\perp}U_sV_t$ annihilates $|F(E,\vec{e})\rangle$. As a result, we may
study the representation $\mathcal R$ of the group ${\rm Spin}(d-1)$ through
its embedding in the Clifford algebra generated by the
$\vec{\Theta}_\alpha\cdot\vec{e}$.
\par
The operators $\vec{\Theta}_\alpha\cdot\vec{e}$ leave the space (\ref{e3.4})
invariant and act irreducibly on it. That space is thus isomorphic to
$\mathcal C$, and ${\rm Spin}(s_d)$ acts according to (\ref{e1.2}) (with
$\Theta_{\alpha A}$ replaced by $\vec{\Theta}_\alpha\cdot\vec{e}$). This
representation decomposes (see e.g. \cite{b3}) as
\begin{equation}
\mathcal C=(2^{(s_d/2)-1})_+\oplus (2^{(s_d/2)-1})_-
\label{e3.2a}
\end{equation}
w.r.t. the subspaces where
$\Theta\equiv
2^{s_d/2}\prod_{\alpha=1}^{s_d}\vec{\Theta}_\alpha\cdot\vec{e}=+1$,
resp. $-1$. The embedding (\ref{e1.2a}) and the corresponding branching of the
representation (but not the statement of the theorem!) depend on the choice
of the $\gamma$--matrices. In order to select a definite embedding, let
\begin{equation}
\gamma^d =\left(
\begin{array}{cc} {\sf 1}\mkern-5.0mu{\rm I} & \ 0 \\ 0 & - {\sf 1}\mkern-5.0mu{\rm I} \end{array} \right) \; , \quad
\gamma^{d-1} = \left(
\begin{array}{cc} \ 0 &\ {\sf 1}\mkern-5.0mu{\rm I} \\ \ {\sf 1}\mkern-5.0mu{\rm I} & \ 0\end{array} \right) \; , \quad
\gamma^j =\left(
\begin{array}{cc} 0 & {\rm i} \Gamma^j\\ -{\rm i} \Gamma^j& 0 \end{array} \right) \;
\label{e3.3}
\end{equation}
with $\Gamma^j,\,(j= 1,\ldots, d-2)$ purely imaginary, antisymmetric, and
$\{ \Gamma^j, \Gamma^k \} = 2 \delta_{jk} {\sf 1}\mkern-5.0mu{\rm I}_{s_d/2}$. Then
(\ref{e3.2a}) branches as (see \cite{b4}, resp. \cite{c1k, c1l})
\begin{equation}
\mathcal C=\cases (44\oplus 84)\oplus 128\;,&\qquad (d=9)\;,\\
(5\oplus 1\oplus 1\oplus 1)\oplus (4 \oplus 4)\;,&\qquad (d=5)\;,\\
2\oplus (1\oplus 1) \;,&\qquad (d=3)\;,
\endcases
\label{e3.6}
\end{equation}
when viewed as a representation of ${\rm Spin}(d)$.
(The choice
$\widetilde{\gamma}^i_{\alpha\beta}=
\widetilde{R}_{\alpha'\alpha}\gamma^i_{\alpha'\beta'}
\widetilde{R}_{\beta'\beta}$ with
$\widetilde{R}\in {\rm O}(s_d),\,\det\widetilde{R}=-1$ would have inverted the
branching of the representations on the r.h.s. of (\ref{e3.2a})). The
case $d=3$ deserves a remark, as there are additional inequivalent embeddings
${\rm Spin}(d=3)\hookrightarrow{\rm Spin}(s_d=4)$, and one has to consider the
one appropriate to (\ref{e1.2a}). In fact $R\in{\rm Spin}(3)={\rm SU}(2)$ acts
in the fundamental representation on $\mathbb C^2$, the irreducible
representation space of the complex Clifford algebra with $3$ generators. The
real representation (\ref{e3.3}) is obtained by joining two complex
representations, followed by an appropriate change $T$ of basis. The embedding
(\ref{e1.2a}) is thus realized through $R \mapsto
T^{-1}(R\otimes{\sf 1}\mkern-5.0mu{\rm I}_2)T$ and the embedding
${\rm su(2)}_{\mathbb C}\hookrightarrow{\rm so(4)}_{\mathbb C}
={\rm su(2)}_{\mathbb C}\oplus{\rm su(2)}_{\mathbb C}$ is equivalent to
$u\mapsto(u,0)$.
\par
The further branching
${\rm Spin}(d)\hookleftarrow{\rm Spin}(d-1)$ yields
\begin{equation}
\mathcal C=\cases (1\oplus 8_{\rm v}\oplus 35_{\rm v})
\oplus(28\oplus 56_{\rm v})
\oplus(8_{\rm s}\oplus 8_{\rm c}\oplus 56_{\rm s}\oplus 56_{\rm c})
\;,&\qquad (d-1=8)\;,\\
1\oplus 1\oplus 1\oplus (1\oplus 4)\oplus
(2_+\oplus 2_-)\oplus(2_+\oplus 2_-)
\;,&\qquad (d-1=4)\;,\\
(1_1\oplus 1_{-1})\oplus 1_0\oplus 1_0 \;,&\qquad (d-1=2)\;.
\endcases
\label{e3.7}
\end{equation}
The content of invariant states stated in the theorem is now manifest. One
should notice that for
$d=3$ the little group ${\rm U}(1)$ is abelian and
the singlets $1_{\pm 1}$ do not correspond to invariant states. For later
use we also retain the fermionic ${\rm Spin}(d)$ representation to which the
remaining singlets are associated,
\begin{equation}
44\quad(d=9)\;;\qquad 1, 1, 1, 5\quad(d=5)\;;\qquad 1,1\quad(d=3)\;,
\label{e3.8}
\end{equation}
together with the corresponding eigenvalue of $\Theta$:
\begin{equation}
\Theta=\quad 1\quad(d=9)\;;\qquad 1, 1, 1, 1\quad(d=5)\;;
\qquad -1,-1\quad(d=3)\;.
\label{e3.8a}
\end{equation}
\subsection{Even states}
\label{sec:pr3}
It remains to check which of these states satisfy $|F(-E,-\vec{e})\rangle
=|F(E,\vec{e})\rangle$. Let us begin by noting that by (\ref{e3.5a})
\begin{equation*}
|F(-E,-\vec{e})\rangle={\rm e}^{{\rm i} M_{AB}e_Au_B\pi}{\rm e}^{{\rm i} M_{st}E_sU_t\pi}
|F(E,\vec{e})\rangle\;,
\end{equation*}
where $\vec{u}\in\mathbb R^3$, resp. $U\in\mathbb R^d$ are unit vectors
orthogonal to $\vec{e}$, resp. $E$. The ${\rm Spin}(d)$ rotation can be
factorized as
${\rm e}^{{\rm i} M_{st}E_sU_t\pi}=
{\rm e}^{{\rm i} M_{st}^{\perp}E_sU_t\pi}{\rm e}^{{\rm i} M_{st}^{\parallel}E_sU_t\pi}$.
We claim that
${\rm e}^{{\rm i} M_{st}^{\parallel}E_sU_t\pi}$ $|F(E,\vec{e})\rangle=
\sigma|F(E,\vec{e})\rangle$
with
\begin{equation}
\sigma=\quad 1\quad(d=9)\;;\qquad 1, 1, 1, -1\quad(d=5)\;;
\qquad 1,1\quad(d=3)\;.
\label{e3.9}
\end{equation}
The operator represents a rotation
$R\in{\rm Spin}(d)$ with $RE=-E$ in the representation (\ref{e3.8}). For $d=9$
the latter can be
realized on symmetric traceless tensors $T_{ij},\,(i,j=1,\ldots, 9)$, where
the ${\rm Spin}(8)$--singlet is $E_iE_j-(1/9)\delta_{ij}$, implying
$\sigma=1$. For $d=5$, the last representation (\ref{e3.8}) is just the vector
representation, where $\sigma=-1$. As the remaining cases are evident, eq.
(\ref{e3.9}) is proven. A computation using (\ref{e3.3}) and, without loss
$E=(0,\ldots,0,1),\,U=(0,\ldots,1,0)$ shows
\begin{eqnarray*}
{\rm e}^{{\rm i} M_{d, d-1}^{\perp}\pi}|F(E,\vec{e})\rangle&=&
\prod_{\alpha=1}^{s_d/2}{\rm e}^{[
(\vec{\Theta}_\alpha\cdot\vec{n}_+)
(\vec{\Theta}_{\alpha+s_d/2}\cdot\vec{n}_-)-
(\vec{\Theta}_{\alpha+s_d/2}\cdot\vec{n}_+)
(\vec{\Theta}_\alpha\cdot\vec{n}_-)]\pi/2}|F(E,\vec{e})\rangle\\
&=&\prod_{\alpha=1}^{s_d/2}
(\vec{\Theta}_{\alpha+s_d/2}\cdot\vec{n}_+)
(\vec{\Theta}_\alpha\cdot\vec{n}_-)|F(E,\vec{e})\rangle\equiv
|\overline{F}(E,\vec{e})\rangle\;,\\
{\rm e}^{{\rm i} M_{AB}e_Au_B\pi}|\overline{F}(E,\vec{e})\rangle&=&
\prod_{\alpha=1}^{s_d}{\rm e}^{
(\vec{\Theta}_\alpha\cdot\vec{e})(\vec{\Theta}_\alpha\cdot\vec{u})\pi}
|\overline{F}(E,\vec{e})\rangle\\
&=&(-1)^{s_d/4}\Theta\prod_{\alpha=1}^{s_d/2}
(\vec{\Theta}_\alpha\cdot\vec{n}_+)
(\vec{\Theta}_{\alpha+s_d/2}\cdot\vec{n}_-)|\overline{F}(E,\vec{e})\rangle=
|F(E,\vec{e})\rangle\;,
\end{eqnarray*}
where we used (\ref{e3.8a}) in the last step. Together with (\ref{e3.9}) this
proves the statement of theorem
concerning the invariance under (\ref{e2.0}).
\subsection{The equation at $n>0$}
\label{sec:pr4}
We next discuss the equations (\ref{e3.0})${}_n$ with $n\ge 1$. Let $P_0$ be
the orthogonal projection onto the states (\ref{e3.2}), i.e., onto the null
space of $Q_\beta^0$. We replace them with an equivalent pair of
equations, obtained by multiplication of (\ref{e3.0})${}_{n+1}$ with $P_0$,
resp. of (\ref{e3.0})${}_n$ with $Q_\beta^0$, which
is injective on the range of the complementary projection
$\overline{P}_0=1-P_0$:
\begin{eqnarray}
P_0\bigl(-(\kappa+\frac{3}{2}n))\widehat{Q}_\beta^1+Q_\beta^1\bigr)P_0\psi_n
=-P_0\bigl(Q_\beta^1\overline{P}_0\psi_n
+Q_\beta^2\psi_{n-1}+\ldots+Q_\beta^{n+1}\psi_0\bigr)\;,&&\nonumber\\
\qquad (n=0,1,\ldots)\;,&&
\label{e3.10}\\
(Q_\beta^0)^2\psi_n
=-Q_\beta^0\Bigl(
\bigl(-(\kappa+\frac{3}{2}(n-1))\widehat{Q}_\beta^1+
Q_\beta^1\bigr)\psi_{n-1}
+Q_\beta^2\psi_{n-2}+\ldots+Q_\beta^n\psi_0\Bigr)\;,&&\nonumber\\
\qquad (n=1,2,\ldots)&&
\label{e3.11}
\end{eqnarray}
(we used $P_0\widehat{Q}_\beta^1\overline{P}_0=0$). Here, and until the end of
this subsection, no summation over $\beta$ is understood. The equation
(\ref{e3.10}) at $n=0$ reads
\begin{equation}
P_0Q_\beta^1\psi_0=\kappa P_0\widehat{Q}_\beta^1\psi_0
\;(=\kappa\widehat{Q}_\beta^1\psi_0)\;.
\label{e3.13}
\end{equation}
\par
We shall verify this by explicit computation later on. Since a similar issue
will show up in solving the equation (\ref{e3.10}) at $n>0$, let us also
present a more general statement, whose proof is postponed to the next
subsection.
\smallskip\noindent
{\bf Lemma} {\it Let $T_\beta$ be linear operators on the range of $P_0$, which
transform as real spinors of ${\rm Spin}(d)$ and commute with the antipode
map. Then, for each invariant state we have
\begin{equation}
T_\beta\psi_0=\kappa\widehat{Q}_\beta^1\psi_0\;,
\label{e3.14}
\end{equation}
with $\kappa$ depending only on the associated representation (\ref{e3.8}).}
\medskip\noindent
We now assume having solved the equations (\ref{e3.10}, \ref{e3.11}) up to
$n-1$ for ${\rm Spin}(d)$ invariant
$\psi_1,\ldots\psi_{n-1}$ (which is true for $n-1=0$), and claim the
same is possible for $n$. Since $Q_\beta^0$ is
invertible on the range of $\overline{P}_0$, eq.~(\ref{e3.11})${}_n$
determines $\overline{P}_0\psi_n$ uniquely. The fact that
the solution so obtained is independent of $\beta$ and is ${\rm Spin}(d)$
invariant may deserve a comment, because the equivalence of the equations
$Q_\beta\psi=0$ and $(Q_\beta)^2\psi=0$, which holds on (\ref{e1.0}), does not
apply in the sense of formal power series (\ref{e2.4}). Consider the expansion
(\ref{e3.00}), i.e.,
\begin{equation*}
Q_\beta = r^{1/2}\sum_{k=0}^\infty r^{-\frac{3}{2}k}[Q_\beta]_k\;, \qquad
[Q_\beta]_k=
Q_\beta^k+\delta_{1k}\widehat{Q}_\beta^1r{\partial\over\partial r}\;,
\end{equation*}
as well as its formal square
\begin{equation*}
(Q_\beta)^2 = r\sum_{k=0}^\infty r^{-\frac{3}{2}k}[(Q_\beta)^2]_k\;.
\end{equation*}
Notice that $(Q_\beta)^2$ is, by (\ref{eq2}), independent of $\beta$ and
${\rm Spin}(d)$ invariant as an operator on ${\rm SU}(2)$ invariant power
series. Similarly, let $[Q_\beta\psi]_k$ (given by
the l.h.s. of (\ref{e3.0})) and $[(Q_\beta)^2\psi]_k$ be the
coefficients of the
corresponding series. By induction assumption we have $[Q_\beta\psi]_k=0$ for
$k=0,\ldots,\,n-1$. Since $Q_\beta(Q_\beta\psi)=(Q_\beta)^2\psi$,
we obtain
\begin{eqnarray*}
[(Q_\beta)^2\psi]_n&=&\sum_{k=0}^n Q_\beta^k[Q_\beta\psi]_{n-k}
-(\kappa+\frac{3}{2}n-2)\widehat{Q}_\beta^1[Q_\beta\psi]_{n-1}
=Q_\beta^0[Q_\beta\psi]_n\;,\\{}
[(Q_\beta)^2\psi]_n&=&(Q_\beta^0)^2\psi_n+\widetilde{\psi}_{n-1}\;,
\end{eqnarray*}
where $\widetilde{\psi}_{n-1}$ (determined by $\psi_0,\ldots\psi_{n-1}$) has
the desired properties. The equation (\ref{e3.11})${}_n$, i.e.,
$Q_\beta^0[Q_\beta\psi]_n=0$ is thus equivalent to
$(Q_\beta^0)^2\psi_n=-\widetilde{\psi}_{n-1}$, which exhibits the claim.
On the other hand, invariance requires
$P_0\psi_n$ to be a linear combination of invariant singlets. For the ansatz
$P_0\psi_n=\lambda_n\psi_0$,
eq. (\ref{e3.10})${}_n$ reads
\begin{equation}
\frac{3}{2}n\lambda_n\widehat{Q}_\beta^1\psi_0=
-P_0\bigl(Q_\beta^1\overline{P}_0\psi_n
+Q_\beta^2\psi_{n-1}+\ldots+Q_\beta^{n+1}\psi_0\bigr)\;,
\nonumber
\end{equation}
because of (\ref{e3.13}). Again, by the lemma, this holds true for suitable
$\lambda_n$. Indeed, this solution for $P_0\psi_n$ is the only one.
\subsection{Proof of the lemma}
\label{sec:pr5}
The vectors $T_\beta\psi_0,\, (\beta=1,\ldots, s_d)$ transform under
${\rm Spin}(d)$ as real spinors,
although they might be linearly dependent. By reducing matters to the little
group as before, any representation of that sort is specified by the values
$|F^\beta(E,\vec{e})\rangle$ of its states (see (\ref{e3.2}))
at one point $(E,\vec{e})$, which are required to satisfy
\begin{equation*}
\widetilde{R}_{\beta\alpha}(R)|F^\alpha(E,\vec{e})\rangle=
\mathcal R(R)|F^\beta(E,\vec{e})\rangle
\end{equation*}
for $R$ with $RE=E$. Pretending the states $|F^\beta(E,\vec{e})\rangle$ to be
linearly independent, the branching
${\rm Spin}(d)\hookleftarrow{\rm Spin}(d-1)$ yields
\begin{eqnarray*}
16=8_{\rm s}\oplus 8_{\rm c}\quad(d=9)\;;&&\qquad
4\oplus 4=(2_+\oplus 2_-)\oplus(2_+\oplus 2_-)\quad(d=5)\;;\\
2\oplus 2&=&(1_1\oplus 1_{-1})\oplus(1_1\oplus 1_{-1})\quad(d=3)\;.
\end{eqnarray*}
For $d=9,5$ each term on the r.h.s. occurs as often as in (\ref{e3.7}), and
$\psi_0$ can indeed be chosen so that the $s_d$ vectors
$\widehat{Q}_\beta^1\psi_0$
are independent. Not so in the last case, where the vectors $T_\beta\psi_0$
just belong to $1_1\oplus 1_{-1}$. We
continue the discussion for different values of $d$ separately.
\par
$\bullet\,d=9$. Any linear transformation $K$ commuting with a
${\rm Spin}(9)$ representation as above is thus of the form
$K=\kappa_{\rm s}\oplus \kappa_{\rm c}$.
If $K$ also commutes with the antipode map, then
$\kappa_{\rm s}=\kappa_{\rm c}\equiv\kappa$. Applying this to the
representation $\widehat{Q}_\beta^1\psi_0$ and to the map
$K:\,\widehat{Q}_\beta^1\psi_0\mapsto T_\beta\psi_0$ yields the
claim.
$\bullet\,d=5$. Let us regroup
$(2_+\oplus 2_-)\oplus(2_+\oplus 2_-)\cong
(2_+\otimes{\sf 1}\mkern-5.0mu{\rm I}_2)\oplus(2_-\otimes{\sf 1}\mkern-5.0mu{\rm I}_2)$. Then any map $K$ commuting
with the representation is of the form
\begin{equation*}
K=({\sf 1}\mkern-5.0mu{\rm I}\otimes K_+)\oplus({\sf 1}\mkern-5.0mu{\rm I}\otimes K_-)\;,
\end{equation*}
where $K_-$ is conjugate to $K_+$ if $K$ commutes with the
antipode map. This allows for a four dimensional space of such maps $K$.
To proceed further we shall again assume that $E=(0,\ldots,0,1)$ and introduce
creation operators
\begin{equation*}
a_\alpha^*=\frac{1}{\sqrt{2}}[(\vec{\Theta}_\alpha\cdot\vec{e})+
{\rm i}(\vec{\Theta}_{\alpha+4}\cdot\vec{e})]\;,\qquad(\alpha=1,\ldots 4)
\end{equation*}
which then define a vacuum through $a_\alpha|0\rangle=0$. We next choose an
orthonormal basis $\{\psi_0^1,\ldots, \psi_0^4\}$ for the 4-dimensional
subspace of singlets in the range of $P_0$ by specifying the values of the
corresponding fermionic parts (see (\ref{e3.2})) at
$(E,\vec{e})$:
\begin{eqnarray*}
|F_0^4(E,\vec{e})\rangle&=&
\frac{1}{\sqrt{2}}(|0\rangle-a_1^*a_2^*a_3^*a_4^*|0\rangle)\;,\\
|F_0^i(E,\vec{e})\rangle&=&
\frac{1}{2\sqrt{2}}\widetilde{\Gamma}^i_{\alpha\beta}
a_\alpha^*a_\beta^*|0\rangle=
\frac{{\rm i}}{4}(\gamma^4\widetilde{\gamma}^i)_{\alpha\beta}
(\vec{\Theta}_\alpha\cdot\vec{e})(\vec{\Theta}_\beta\cdot\vec{e})
|F_0^4(E,\vec{e})\rangle\;,
\qquad(i=1,2,3)\;,\\
\end{eqnarray*}
where
\begin{equation*}
\widetilde{\gamma}^i=\left(
\begin{array}{cc} 0 & {\rm i} \widetilde{\Gamma}^i\\
-{\rm i} \widetilde{\Gamma}^i& 0 \end{array} \right)
=\sigma^{-1}\gamma^i\sigma\;,
\qquad
\sigma=\left(
\begin{array}{cc} \Sigma & 0 \\ 0 & \Sigma\end{array} \right)
\end{equation*}
with $\Sigma\in{\rm O}(4)$ and $\det\Sigma=-1$.
Note that $\psi_0^4$ is the singlet belonging to the 5--dimensional
fermionic representation of ${\rm Spin}(5)$. One can verify that the four maps
\begin{equation*}
K^i:\,\widehat{Q}_\beta^1\psi_0^1\mapsto
\cases
\widehat{Q}_\beta^1\psi_0^i\;,&(i=1,2,3)\;,\\
\gamma^t_{\beta\alpha}E_t \widehat{Q}_\alpha^1\psi_0^4\;,&(i=4)\;,\\
\endcases
\end{equation*}
besides being of the kind just discussed, are linearly independent. Therefore
any map $K$ of the above form is a linear combination thereof. In particular
this applies,
for any $(\underline{x}, x_4)\in\mathbb R^{3+1}$, to the map
$K:\,\widehat{Q}_\beta^1\psi_0^1\mapsto
x_iT_\beta\psi_0^i+x_4\gamma^t_{\beta\alpha}E_t T_\alpha\psi_0^4$,
hence
\begin{equation*}
x_iT_\beta\psi_0^i+x_4\gamma^t_{\beta\alpha}E_t T_\alpha\psi_0^4
=y_i\widehat{Q}_\beta^1\psi_0^i+
y_4\gamma^t_{\beta\alpha}E_t\widehat{Q}_\alpha^1\psi_0^4\;.
\end{equation*}
This defines a linear map
$\kappa: (\underline{x},x_4)\mapsto(\underline{y},y_4)$
on $\mathbb R^{3+1}$. We claim that
\begin{equation}
\kappa: (R\underline{x},x_4)\mapsto(R\underline{y},y_4)
\label{e3.13a}
\end{equation}
for
$R\in {\rm SO}(3)$, which implies
$\kappa={\rm diag}(\kappa_1=\kappa_2=\kappa_3, \kappa_4)$ and hence
(\ref{e3.14}). Eq. (\ref{e3.13a}) can be proven using
$R_{ij}\psi_0^i=\mathcal R\psi_0^j$ for $\mathcal R\in {\rm Spin}(8)$
projecting to $R\in{\rm Spin}(3)\subset{\rm Spin}(5)
\hookrightarrow{\rm SO}(8)$. This
in turn follows from (\ref{e1.2}) and from $\mathcal R\psi_0^4=\psi_0^4$.
$\bullet\,d=3$. Analogously to $d=9$.
\subsection{Determination of $\kappa$}
\label{sec:pr6}
Since $J_{AB}\psi_0=J_{st}\psi_0=0$ we may replace $Q_\beta^1$ by
\begin{equation}
Q_\beta^1=\Theta_{\alpha A}\gamma^t_{\alpha\beta}
\bigl(-e_BE_tM_{BA}-e_AE_sM_{st}
-\frac{{\rm i}}{2}\,e_AE_ty_{sB}\frac{\partial}{\partial y_{sB}})+
\frac{1}{2}\vec{\Theta}_\alpha\cdot(\vec{y}_s\times\vec{y}_t)
\gamma^{st}_{\beta\alpha}\;.
\label{e4.1}
\end{equation}
We discuss the contributions to (\ref{e3.13}) of these four terms
separately.
\par
i) With
\begin{equation}
e_BM_{BA}=-\frac{{\rm i}}{2}\bigl((\vec{\Theta}_\beta\cdot\vec{e})\Theta_{\beta A}-
\Theta_{\beta A}(\vec{\Theta}_\beta\cdot\vec{e})\bigr)
\nonumber
\end{equation}
we find
\begin{eqnarray*}
\Theta_{\alpha A}e_BM_{BA}&=&{\rm i}\bigl(
(\vec{\Theta}_\alpha \cdot\vec{n}_+)(\vec{\Theta}_\beta\cdot\vec{n}_-)+
(\vec{\Theta}_\alpha \cdot\vec{n}_-)(\vec{\Theta}_\beta\cdot\vec{n}_+)
\bigr)(\vec{\Theta}_\beta\cdot\vec{e}\,)\;,\\
P_0\Theta_{\alpha A}e_BM_{BA}\psi_0&=&
{\rm i}(\vec{\Theta}_\alpha\cdot\vec{e}\,)\psi_0\;,
\end{eqnarray*}
since only the term with $\beta=\alpha$ survives the projection $P_0$. Hence
\begin{equation}
-P_0\Theta_{\alpha A}\gamma^t_{\alpha\beta}e_BE_tM_{BA}\psi_0
=\widehat{Q}_\beta^1\psi_0
\label{e4.3}
\end{equation}
contributes 1 to $\kappa$.
\par
ii) Similarly,
\begin{equation}
-P_0(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^t_{\alpha\beta}E_sM_{st}\psi_0=
-(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^t_{\alpha\beta}E_s
M_{st}^{\parallel}\psi_0\;,
\nonumber
\end{equation}
where $M_{st}^{\parallel}$ is given in (\ref{e3.8a}). For the r.h.s. we then
claim
\begin{equation}
-(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^t_{\alpha\beta}E_s
M_{st}^{\parallel}\psi_0=\kappa'\widehat{Q}_\beta^1\psi_0
\label{e4.3a}
\end{equation}
with
\begin{equation}
\kappa'=\cases 9\,,&\qquad(d=9)\;,\\
0,0,0,4\,,&\qquad(d=5)\;,\\
0,0\,,&\qquad(d=3)\;.
\endcases
\label{e4.3b}
\end{equation}
This is clear in the cases where the representation in (\ref{e3.8}) is
already a singlet, i.e., when $\kappa'=0$. To prove the two
remaining cases we first establish
\begin{equation}
-(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^t_{\alpha\beta}E_s
M_{st}^{\parallel}\psi_0
=-\frac{{\rm i}}{2}\gamma^s_{\alpha\beta}E_s[\vec{\Theta}_\alpha\cdot\vec{e}\,,
M_{ut}^{\parallel}M_{ut}^{\parallel}]\psi_0
-{\rm i}\frac{d^2-d}{8}(\vec{\Theta}_\alpha\cdot\vec{e}\,)
\gamma^s_{\alpha\beta}E_s\psi_0\;,
\label{e4.4}
\end{equation}
or the equivalent equation obtained by multiplication from the right
with $E_u\gamma^u$:
\begin{equation}
-(\vec{\Theta}_\alpha\cdot\vec{e}\,)(\gamma^t\gamma^u)_{\alpha\beta}
E_uE_sM_{st}^{\parallel}\psi_0=
-\frac{{\rm i}}{2}[\vec{\Theta}_\beta\cdot\vec{e}\,,
M_{ut}^{\parallel}M_{ut}^{\parallel}]\psi_0
-{\rm i}\frac{d^2-d}{8}(\vec{\Theta}_\beta\cdot\vec{e}\,)\psi_0\;.
\label{e4.5}
\end{equation}
To this end
we note that, by the invariance of $\psi_0$, its
fermionic part $|F(E,\vec{e})\rangle$ at $E\in S^{d-1}$ is invariant under
rotations of ${\rm Spin}(d) $ leaving $E$ fixed:
$(\delta_{us}-E_uE_s)M_{sv}^{\parallel}(\delta_{vt}-E_vE_t)\psi_0=0$,
i.e.,
\begin{equation}
(M_{st}^{\parallel}E_uE_s+M_{uv}^{\parallel}E_vE_t)\psi_0=
M_{ut}^{\parallel}\psi_0\;.
\label{e4.5a}
\end{equation}
Using
$\gamma^t\gamma^u=-\gamma^{ut}+\delta^{ut}{\sf 1}\mkern-5.0mu{\rm I}$ and the observation just made
we rewrite the l.h.s. of (\ref{e4.5}) as
\begin{eqnarray*}
-(\vec{\Theta}_\alpha\cdot\vec{e}\,)(\gamma^t\gamma^u)_{\alpha\beta}
E_uE_sM_{st}^{\parallel}\psi_0&=&
(\vec{\Theta}_\alpha\cdot\vec{e}\,)
\gamma^{ut}_{\alpha\beta}E_uE_sM_{st}^{\parallel}\psi_0\cr
&=&\frac{1}{2}(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^{ut}_{\alpha\beta}
(E_uE_sM_{st}^{\parallel}-E_tE_sM_{su}^{\parallel})\psi_0\cr
&=&\frac{1}{2}(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^{ut}_{\alpha\beta}
M_{ut}^{\parallel}\psi_0\;.
\end{eqnarray*}
The commutation relation
\begin{equation*}
{\rm i}[\vec{\Theta}_\alpha\cdot\vec{e}\, ,M_{ut}^{\parallel}]=
\frac{1}{2}\gamma^{ut}_{\alpha\beta}(\vec{\Theta}_\beta\cdot\vec{e}\,)
\end{equation*}
follows from (\ref{e1.2}) or by direct computation. It implies
\begin{eqnarray*}
{\rm i}[\vec{\Theta}_\alpha\cdot\vec{e}\, ,M_{ut}^{\parallel}M_{ut}^{\parallel}]&=&
\frac{1}{2}\gamma^{ut}_{\alpha\beta}\{
\vec{\Theta}_\beta\cdot\vec{e}\, ,M_{ut}^{\parallel}\}=
\gamma^{ut}_{\alpha\beta}(\vec{\Theta}_\beta\cdot\vec{e}\,)M_{ut}^{\parallel}
-\frac{1}{2}\gamma^{ut}_{\alpha\beta}
[\vec{\Theta}_\beta\cdot\vec{e}\,,M_{ut}^{\parallel}]\cr
&=&
\gamma^{ut}_{\alpha\beta}(\vec{\Theta}_\beta\cdot\vec{e}\,) M_{ut}^{\parallel}
-{\rm i}\frac{d^2-d}{4}\,\vec{\Theta}_\alpha\cdot\vec{e}\,\;.
\end{eqnarray*}
Solving for the first term on the r.h.s. proves (\ref{e4.5}) and hence
(\ref{e4.4}). Let us now note that for $d=9$ the fermionic part of
$\psi_0$, resp. of $(\vec{\Theta}_\alpha\cdot\vec{e}\,)\psi_0$ belongs to the
$44$, resp. $128$ representation of ${\rm Spin}(9)$ (see (\ref{e3.6})).
Eq. (\ref{e4.4}) then implies
\begin{equation*}
-(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^t_{\alpha\beta}E_s
M_{st}^{\parallel}\psi_0=
(C(44)-C(128)+9)\widehat{Q}_\beta^1\psi_0=9\widehat{Q}_\beta^1\psi_0\;,
\end{equation*}
where we used the values \cite{b4} of the Casimir: $C(44)=C(128)=18$. In the case
$d=5$ the
fermionic part of $\psi_0$, resp. of
$(\vec{\Theta}_\alpha\cdot\vec{e}\,)\psi_0$ belongs to the representation
$5$, resp. $4\oplus 4$. We conclude that
\begin{equation*}
-(\vec{\Theta}_\alpha\cdot\vec{e}\,)\gamma^t_{\alpha\beta}E_s
M_{st}^{\parallel}\psi_0=
(C(5)-C(4)+\frac{5}{2})\widehat{Q}_\beta^1\psi_0=4\widehat{Q}_\beta^1\psi_0\;,
\end{equation*}
given that $C(5)=4,\,C(4)=5/2$.
\par
We remark that the proof of (\ref{e4.3b}) can be shortened by using the
lemma, according to
which (\ref{e4.3a}) holds true for some $\kappa'$.
Thus, contracting with
$\widehat{Q}_\beta^1\psi_0$ and summing over $\beta$, we find
\begin{eqnarray*}
-\kappa'(\psi_0,\widehat{Q}_\beta^1\widehat{Q}_\beta^1\psi_0)&=&
-{\rm i}(\psi_0,(\vec{\Theta}_\gamma\cdot\vec{e})\gamma^u_{\gamma\beta}E_u
(\vec{\Theta}_\alpha\cdot\vec{e})\gamma^t_{\alpha\beta}E_s
M_{st}^{\parallel}\psi_0)\\
&=&4(\psi_0,E_uM_{ut}^{\parallel}M_{st}^{\parallel}E_s\psi_0)\\
&=&2(\psi_0,M_{ut}^{\parallel}
(M_{st}^{\parallel}E_uE_s+M_{uv}^{\parallel}E_vE_t)\psi_0)
=2(\psi_0,M_{ut}^{\parallel}M_{ut}^{\parallel}\psi_0)\;.
\end{eqnarray*}
In the step before last we relabeled indices in half the expression; in the
last one we used (\ref{e4.5a}).
Using $\widehat{Q}_\beta^1\widehat{Q}_\beta^1=-s_d/2$ we obtain
$(s_d/2)\kappa'=2\cdot 2\cdot C$, i.e., $\kappa'=8C/s_d$,
where $C$ is the Casimir
in the representation (\ref{e3.8}). The above values of $C(44)\,(d=9)$ and of
$C(5)\,(d=5)$ yield again (\ref{e4.3b}).
\par
iii) Using ${\rm d}{\rm e}^{-y^2/2}/{\rm d} y=
-y{\rm e}^{-y^2/2}$ we get
\begin{equation}
\frac{1}{2}y_{sB}\frac{\partial}{\partial y_{sB}}\psi_0=
-\frac{1}{2}y_{sB}y_{sB}\psi_0=
-\frac{1}{2}\sum_{sB}(y_{sB}^2-\frac{1}{2})\psi_0-
\frac{1}{4}\cdot 2(d-1)\psi_0\;,
\label{e4.9}
\end{equation}
where the sum, consisting of second Hermite functions, is annihilated by
$P_0$.
\par
iv) The last term in (\ref{e4.1}), when acting on $\psi_0$, is similarly
annihilated by $P_0$.
\smallskip
Collecting terms (\ref{e4.3}, \ref{e4.3b}, \ref{e4.9}) we find
\begin{equation*}
\kappa=1+\kappa'-\frac{1}{2}(d-1)=\cases 6\,,&\qquad(d=9)\;,\\
-1,-1,-1,3\,,&\qquad(d=5)\;,\\
0, 0\,,&\qquad(d=3)\;.
\endcases
\end{equation*}
\setcounter{section}{0}
\section*{Appendix 1}
To prove (\ref{e3.000}) we shall compute the partial derivatives in
\begin{equation}
\frac{\partial}{\partial q_{tA}}=
\frac{\partial r}{\partial q_{tA}}\frac{\partial}{\partial r}+
\frac{\partial e_B}{\partial q_{tA}}\frac{\partial}{\partial e_B}+
\frac{\partial E_s}{\partial q_{tA}}\frac{\partial}{\partial E_s}+
\frac{\partial y_{sB}}{\partial q_{tA}}\frac{\partial}{\partial y_{sB}}\;.
\label{eq:a0}
\end{equation}
We regard $r,\,\vec{e},\,E,\,y$ as functions of $q$ defined by
$\vec{e}{\,}^2=\sum_s E_s^2 = 1$ and (\ref{e2.00}, \ref{e2.1})
and solve for their differentials by taking different contractions of
\begin{equation*}
{\rm d} q_{tA}=(e_AE_t-\frac{1}{2}r^{-3/2}y_{tA}){\rm d} r+
rE_t{\rm d} e_A+re_A{\rm d} E_t+r^{-1/2}{\rm d} y_{tA}\;.
\end{equation*}
Using that
\begin{eqnarray*}
e_A{\rm d} y_{tA}+y_{tA}{\rm d} e_A=0\;,&\qquad E_t{\rm d} y_{tA}+y_{tA}{\rm d} E_t=0\;,&\\
e_A{\rm d} e_A=0\;,&\qquad E_t{\rm d} E_t=0\;,&
\end{eqnarray*}
the contractions are:
\begin{eqnarray}
e_AE_t{\rm d} q_{tA}&\!=\!&{\rm d} r\;,\nonumber\\
(\delta_{BA}-e_Be_A)E_t{\rm d} q_{tA}&\!=\!&r{\rm d} e_B-r^{-1/2}y_{tA}{\rm d} E_t\;,
\label{eq:a1}\\
e_A(\delta_{st}-E_sE_t){\rm d} q_{tA}&\!=\!&r{\rm d} E_s-r^{-1/2}y_{sA}{\rm d} e_A\;,
\label{eq:a2}\\
(\delta_{BA}-e_Be_A)(\delta_{st}-E_sE_t){\rm d} q_{tA}&\!=\!&
-\frac{1}{2}r^{-3/2}y_{sB}{\rm d} r+
r^{-1/2}({\rm d} y_{sB}+e_By_{sA}{\rm d} e_A+E_sy_{tB}{\rm d} E_t)\;.
\nonumbe
\end{eqnarray}
We solve (\ref{eq:a1}, \ref{eq:a2}) for ${\rm d} e_B,\, {\rm d} E_s$:
\begin{eqnarray*}
{\rm d} r&=&e_AE_t{\rm d} q_{tA}\;,\\
{\rm d} e_B&=&(m^{-1})_{BC}(r^{-1}(\delta_{CA}-e_Ce_A)E_t+r^{-5/2}y_{tC}e_A)
{\rm d} q_{tA}\\
&=&(r^{-1}(\delta_{BA}-e_Be_A)E_t+{\rm O}(r^{-5/2})){\rm d} q_{tA}\;,\\
{\rm d} E_s&=&(M^{-1})_{su}(r^{-1}(\delta_{ut}-E_uE_t)e_A+r^{-5/2}y_{sA}E_t)
{\rm d} q_{tA}\\
&=&(r^{-1}(\delta_{st}-E_sE_t)e_A+{\rm O}(r^{-5/2})){\rm d} q_{tA}\;,\\
{\rm d} y_{sB}&=&[r^{1/2}(\delta_{BA}-e_Be_A)(\delta_{st}-E_sE_t)+
\frac{1}{2}r^{-1}e_AE_ty_{sB}]{\rm d} q_{tA}
-e_By_{sA}{\rm d} e_A-E_sy_{tB}{\rm d} E_t\;,
\end{eqnarray*}
where $m,\, M$ are the matrices
\begin{equation*}
m_{AB}=\delta_{AB}-r^{-3}y_{tA}y_{tB}\;,\qquad
M_{st}=\delta_{st}-r^{-3}y_{sA}y_{tA}\;.
\end{equation*}
We can now read off the partial derivatives appearing in (\ref{eq:a0}) and
obtain
\begin{eqnarray}
\frac{\partial}{\partial q_{tA}}&=&
r^{1/2}(\delta_{st}-E_sE_t)(\delta_{AB}-e_Ae_B)
\frac{\partial}{\partial y_{sB}}
+r^{-1}[e_AE_t(r\frac{\partial}{\partial r}+
\frac{1}{2}y_{sB}\frac{\partial}{\partial y_{sB}})]
\nonumber\\
&&+r^{-1}(\delta_{AC}-e_Ae_C)E_t
(\delta_{CB}\frac{\partial}{\partial e_B}-e_By_{sC}\frac{\partial}{\partial
y_{sB}}) \nonumber\\
&&+r^{-1}(\delta_{ut}-E_uE_t)e_A
(\delta_{us}\frac{\partial}{\partial E_s}-E_sy_{uB}\frac{\partial}{\partial
y_{sB}})
+{\rm O}(r^{-5/2})\;,\label{eq:a4}
\end{eqnarray}
with the remainder not containing derivatives w.r.t. $r$. Finally, we insert
this expression into
\begin{eqnarray*}
{\rm i} L_{BA}&=&
q_{sB}\frac{\partial}{\partial q_{sA}}-q_{sA}\frac{\partial}{\partial q_{sB}}\\
&=&[(\delta_{AC}-e_Ae_C)y_{sB}-(\delta_{BC}-e_Be_C)y_{sA}]
\frac{\partial}{\partial y_{sC}}\\
&&+e_B(\delta_{AC}\frac{\partial}{\partial e_C}-
e_Cy_{sA}\frac{\partial}{\partial y_{sC}})
-e_A(\delta_{BC}\frac{\partial}{\partial e_C}-
e_Cy_{sB}\frac{\partial}{\partial y_{sC}})\;,
\end{eqnarray*}
(with no higher order corrections, as $L_{AB}$ is of exact order ${\rm O}(r^0)$) and
then into
\begin{equation*}
{\rm i} r^{-1} e_BE_tL_{BA}=r^{-1}(\delta_{AC}-e_Ae_C)E_t
(\delta_{CB}\frac{\partial}{\partial e_B}-e_By_{sC}\frac{\partial}{\partial
y_{sB}})\;.
\end{equation*}
Similarly, we have
\begin{equation*}
{\rm i} r^{-1} e_AE_sL_{st}=r^{-1}(\delta_{ut}-E_uE_t)e_A
(\delta_{us}\frac{\partial}{\partial E_s}-E_sy_{uB}\frac{\partial}{\partial
y_{sB}})\;.
\end{equation*}
Together with (\ref{eq:a4}), this proves (\ref{e3.000}).
\section*{Appendix 2}
Consider
\begin{equation} \label{xy-Hamiltonian}
H = ( -{\partial_x}^2 -{\partial_y}^2 + x^2y^2 ) {\sf 1}\mkern-5.0mu{\rm I} +
\left(
\begin{array}{cc}
x & -y \\
-y & -x
\end{array}
\right)\;,
\end{equation}
which is the square of
\begin{equation*}
Q = {\rm i}
\left(
\begin{array}{cc}
\partial_x & \partial_y + xy \\
\partial_y - xy & - \partial_x
\end{array}
\right) .
\end{equation*}
Just as in (\ref{e1.4}), the bosonic potential $V$ ($=x^2y^2$) is
non-negative,
but vanishing in regions of the configuration space that extend to infinity
(causing the classical partition function to diverge). Quantum--mechanically,
just as in (\ref{e1.4}), the bosonic system is stabilized by the zero point
energy of fluctuations transverse to the flat directions; the fermionic matrix
part in (\ref{xy-Hamiltonian}) exactly cancels this effect, causing the
spectrum to cover the whole positive real axis \cite{c4b}. As simple as it is,
it has remained an open question (for now more than 10 years) whether
(\ref{xy-Hamiltonian}) admits a normalizable zero energy solution, or
not. The argument, derived in a few lines below, gives `no' as an answer
and provides the simplest illustration of our method: as $x \rightarrow
+ \infty, \ Q \Psi = 0$ has two approximate solutions,
\begin{equation} \label{xy-solutions}
\Psi_{+} = e^{-\frac{xy^2}{2}} \left( \begin{array}{c} 0 \\ 1 \end{array}
\right)
\quad \mathrm{and} \quad
\Psi_{-} = e^{+\frac{xy^2}{2}} \left( \begin{array}{c} 1 \\ 0 \end{array}
\right)\;,
\end{equation}
the first of which should be chosen for $\Psi_{0}$ in the
asymptotic expansions
\begin{equation} \label{xy-expansion}
\Psi = x^{-\kappa} ( \Psi_{0} + \Psi_{1} + ... )\; .
\end{equation}
In this simple example, the sum $Q = \sum_{n=0}^{\infty} Q^{(n)}$ terminates
after the first two terms, and
\begin{equation*}
0 \stackrel{\mathrm{!}}{=} Q \Psi =
\left( \left(
\begin{array}{cc}
0 & \partial_y + xy \\
\partial_y -xy & 0
\end{array}
\right)
+
\left(
\begin{array}{cc}
\partial_x & 0 \\
0 & - \partial_x
\end{array}
\right) \right)
\left( x^{- \kappa} ( \Psi_0 + \Psi_1 + ... ) \right)\;,
\end{equation*}
yields (as already anticipated, cp. (\ref{xy-solutions}))
\begin{equation*}
\left(
\begin{array}{cc}
0 & \partial_y + xy \\
\partial_y -xy & 0
\end{array}
\right)
\Psi_0 = 0
\end{equation*}
and
\begin{equation} \label{xy-eqn2}
\left(
\begin{array}{cc}
0 & \partial_y + xy \\
\partial_y -xy & 0
\end{array}
\right) \Psi_n +
x^{\kappa}
\left(
\begin{array}{cc}
\partial_x & 0 \\
0 & - \partial_x
\end{array}
\right) x^{-\kappa} \Psi_{n-1} = 0\;, \qquad n = 1, 2, ... \; .
\end{equation}
Multiplying (\ref{xy-eqn2}) by $\Psi_{0}^{\dagger}$ and integrating over $y$
one
sees that
\begin{equation*}
\int_{- \infty}^{+ \infty} e^{-\frac{xy^2}{2}} x^{\kappa}( 0 , -
\partial_x) x^{-\kappa} \Psi_{n-1} dy
\end{equation*}
has to vanish, implying in particular
\begin{eqnarray*}
0 & = & \int_{- \infty}^{+ \infty} \left( \frac{y^2}{2} +
\frac{\kappa}{x} \right)
e^{- x y^2}
dy\;,
\\ \nonumber
\kappa & = & - \frac{1}{4} \; ,
\end{eqnarray*}
which proves that (\ref{xy-Hamiltonian}) does not admit any
square--integrable solution of the form (\ref{xy-expansion}). A different
approach has recently been undertaken by Avramidi \cite{avr}. Finally note
that, calculating the $\Psi_{n > 0}$
from
(\ref{xy-eqn2}), yields the asymptotic expansion, $x \rightarrow +
\infty $,
\begin{equation*}
\Psi(x,y) =
x^{\frac{1}{4}} e^{-\frac{xy^2}{2}} \sum_{n=0}^{\infty} x^{-
\frac{3n}{2}}
\left(
\begin{array}{c}
\frac{y}{4x} f_n(xy^2) \\
g_n(xy^2)
\end{array}
\right)\;,
\end{equation*}
where $f_0 = 1 = g_0,\, f_1=0=g_1$, and the $f_n (s), \,g_{n}(s)$ are the
(unique) polynomial solutions
\begin{equation*}
f_n (s) = \sum_{i = 0}^n f_{n,i} s^i\; , \qquad
g_n(s) = \sum_{ i = 0}^n g_{n,i} s^i
\end{equation*}
of
\begin{eqnarray*}
2s f'_n + ( 1 - 2s ) f_n & = & \left( 1 - 2s - 6n \right) g_n + 4s g'_n\;,
\\ \nonumber
8 g'_{n+2} & = & \left( \frac{3}{4} + \frac{s}{2} + \frac{3n}{2} \right)
f_n - s f'_n\; .
\end{eqnarray*}
\medskip
\noindent {\bf Acknowledgments.\/} We thank A. Alekseev, I. Avramidi, V. Bach,
F. Finster, H. Nicolai, C. Schweigert, R. Suter, P. Yi for useful discussions.
We also
thank the following institutions for support: the Albert Einstein Institute,
the Fields Institute,
the Erwin Schr\"odinger Institute, the Institute for Theoretical
Physics of ETH, the Mathematics Department of Harvard University, the Deutsche
Forschungsgemeinschaft.
|
\section{Introduction}
As is well known, the existence of consistent classical
actions for extended supersymmetric objects of spatial dimension
$p$ is restricted to certain dimensions $D$ of spacetime.
This is \emph{e.g.}\ the case of the $p$-branes of the minimal or `old'
branescan \cite{Ach.Eva.Tow.Wil:87}, which restricts the actions
to certain values $(D,p)$ for which there exists a Wess-Zumino (WZ)
term. This is needed for the $\kappa$-symmetry of the full action
that matches the physical bosonic and fermionic degrees
of freedom on the worldvolume $W$, and the WZ term is given
by a closed $(p+2)$-form which can be interpreted \cite{Azc.Tow:89}
as a Chevalley-Eilenberg \cite{Che.Eil:48} (CE)
$(p+2)$-cocycle on superspace.
The first classification of $p$-branes \cite{Ach.Eva.Tow.Wil:87}
was restricted to fields forming a scalar supermultiplet on
$W$, consisting of scalars and spinors after gauging the
$\kappa$-symmetry.
The restriction to superspace coordinates $x^\mu,\theta^\alpha$ on
$W$ was later removed with the addition of higher spin fields, vectors
or antisymmetric tensors, forming vector or antisymmetric tensor
supermultiplets on $W$. This, together with the Bose-Fermi
matching conditions led to an enlargement of the possibilities
(see \cite{Duf.Khu.Lu:95,Duf:96b} and earlier
references therein) for the classically allowed supermembranes.
Recently, $p$-branes including an abelian vector gauge
field on $W$ have been interpreted as (Dirichlet)
$D$-branes \cite{Pol:95} (see \cite{Pol.Cha.Joh:96} for a review).
Their kinetic term is described by a Born-Infeld
type Lagrangian which replaces the usual Nambu-Goto
one to accommodate the vector potential;
in similarity with the $p$-branes in \cite{Ach.Eva.Tow.Wil:87},
there also exists a $\kappa$-symmetric
worldvolume action \cite{Aga.Pop.Sch:97} for them.
The introduction of other objects such as $L$-branes (which
have linear supermultiplets on $W$) \cite{How.Rae.Rud.Sez:98}
etc., have enlarged the number and types of $p$-branes.
Finally,
the emergence of a web of dualities among the five consistent
$10$-dimensional string theories, all presumably
subsumed, together with $D=11$
supergravity, in the eleven dimensional $M$-theory (see, \emph{e.g.}\
\cite{Sch:97b})
has led to the `second superstring revolution' and to a
change of the conventional views of supersymmetry.
One version of the $M$-theory, $M$-atrix theory
\cite{Ban.Fis.She.Sus:97}, even
reinterprets spacetime coordinates as non-commuting matrices.
The existence of various
extended objects for which there is no unified description
suggests that, in the same way Minkowski space was enlarged
to the superspace $\Sigma$ to treat bosons and fermions
simultaneously, it may be necessary to extend $\Sigma$ further
to accommodate in a unified point of view a number of
the physical aspects mentioned above. In particular,
one might hope to remove the need of defining fields
{\it directly} on $W$ if an extended superspace ${\tilde \Sigma}$
is introduced, as it will be seen to be the case. This
extension of $\Sigma$
is tantamount to enlarging the $D$-dimensional superPoincar\'e
$sP$ to $\widetilde{sP}$ and to defining the extended superspace
$\tilde{\Sigma}$ by
the quotient $\widetilde{sP}/Spin(1,D-1)$.
Endowing $\tilde{\Sigma}$ with a supergroup structure means that
there must exist new superalgebras going beyond the
ordinary supersymmetry algebra, and several of them have
been discussed in various contexts
\cite{Azc.Gau.Izq.Tow:89,Gre:89,Ber.Sez:89,
Azc.Izq.Tow:91, Ber.Sez:95, Bar:96,
Sor.Tow:97,Sez:97,Der.Gal:97,Bar:97,Sak:98}.
Our point of view, however, will be to assume that fermions
(in the form of odd abelian spinor translations)
are the only basic ({\it i.e.}, initial) entities.
We shall then look for the most
general superspace groups that are allowed by group
extension theory and discuss their consequences for
a unified picture of superbranes.
We find this path rather natural, but it is
not the only one. Another possibility is to take
the worldvolume supersymmetry of the $p$-branes into account
by elevating the target superspace coordinates to worldvolume
superfields \cite{Ban.Pas.Sor.Ton.Vol:95,How.Sez:97}
(see also \cite{How.Rae.Rud.Sez:98}
and references therein), but we shall not follow this
superembedding or `double supersymmetry' approach.
Stated as above, the problem is first a mathematical one.
Much in the same way that rigid superspace is itself a group
extension, and hence supersymmetry is the result of
the non-trivial cohomology of a certain odd superstranslation
group $\text{sTr}_D$ \cite{Ald.Azc:85,Azc.Izq:95}, it is worth
looking for all the possible group extensions of
the various $\text{sTr}_D$ ({\it i.e.}, $\text{sTr}_D$=
\{$N$=1, $\text{sTr}_{11}$, $\text{sTr}_{10}$, IIA, IIB, etc\})
to explore their r\^ole in more general theories.
At the algebra level, the possible supersymmetry algebras were already
investigated in \cite{Haa.Lop.Soh:75} and, allowing tensor
`central' charges, in \cite{Hol.Pro:82} (tensorial charges
were also considered in \cite{DAu.Fre:82,Ziz:84b,Ziz:84}).
But there is also a physical reason behind the mathematical
extension problem.
It is known that the quasi-invariance of a Lagrangian under a symmetry
indicates that the (second) cohomology group is non-trivial,
and that the symmetry group may hence be extended.\footnote{
For a detailed account of quasi-invariance, Noether currents,
cohomology and
extensions, see \cite{Azc.Izq:95,Azc:92}.}
This was exploited in \cite{Azc.Gau.Izq.Tow:89} to extend the supersymmetry
algebra for the supersymmetric extended objects by topological charges.
For Lagrangians containing a quasi-invariant piece $\phi^*(b)$ constructed
from a form $b$ on a group
$\Sigma$ by pulling it back to a manifold $W$ by $\phi^*$, where
$\phi:W\to\Sigma$, the extension may allow us, as it is the case
with the WZ terms $b$ of supersymmetric objects, to obtain
manifestly invariant terms $\tilde b$
\footnote{This is not always possible.
When the group $G$ is simple, the extension appears only at the
loop algebra (of charge densities) level,
as for the $su(2)$ Kac-Moody algebra for a WZW model, and
disappears for the
algebra of charges, as required by Whitehead's lemma.}
by defining them on an \emph{extended} group manifold $\tilde \Sigma$.
In fact, it was shown in \cite{Ber.Sez:95}
that to every free differential algebra in
\cite{Azc.Tow:89} corresponds a new spacetime superalgebra, from which
invariant forms can be found to define new WZ terms
\footnote{Although it may be argued that these invariant
terms should no longer be called WZ terms, we shall retain this
name for them.}. We shall take the analysis of
\cite{Ber.Sez:95, Azc.Tow:89,Sez:97} further
(see also \cite{Sor.Tow:97}) by considering various
superbrane types and by emphasizing the supergroup manifold
point of view. Thus, we shall look for and introduce extended
superspace groups $\tilde \Sigma$ in a systematic way
(we restrict our attention here to rigid superspaces). The
additional variables in these will determine symmetries to which
(topological) charges may correspond via the standard Noether
theorem. For the branes of the old branescan, these new
variables will appear only in the WZ term and as a total
differential. This will be different for the D-branes,
for which we will obtain, nevertheless, that it is also possible
to find an action defined on an extended superspace (thus
removing the necessity of introducing directly worldvolume
fields) with a WZ term given by a CE $(p+2)$-cocycle.
By showing that all these structures and extended superspaces
${\tilde \Sigma}$ follow from a basic odd translation group
$\text{sTr}_D$
defined by the Grassmann spinors of the specific theory,
we may conclude that the $\tilde \Sigma$'s (and
the corresponding extended superPoincar\'e groups $\widetilde{sP}$)
are in a way as fundamental as the standard one, and
necessary for a proper description of the physics involved around
$M$-theory and its six weak coupling limit corners.
The new variables may be relevant in the search for superbrane
actions, in the description of
dualities or in the quantisation process.
This paper is organised as follows. Sec. 2 contains
all central extensions of $\text{sTr}_D$, including ordinary
superspace, for various dimensions, and its
results are summarised in a table. Sec. \ref{NCESA} considers in
general the inclusion of additional non-central generators.
Sec. \ref{Nns} is devoted to the structure of
the new superspaces ${\tilde \Sigma}$ and provides a compact
expression for the contribution to the Noether
charges coming from the WZ terms of the various possible actions,
once they are formulated on ${\tilde \Sigma}$. Sec.
\ref{p12} shows how the simplest $D=10,11$ extended
superspaces are relevant to construct a manifestly invariant
WZ term, both for the Green-Schwarz superstring \cite{Sie:94,Gre:89}
(which we will complete with an additional contribution),
and for the supermembrane. We shall recover there the results of
\cite{Ber.Sez:95} and compute the topological charges which, in
our approach, correspond to the new group variables.
The question of the linearity of the group action is seen in Sec.
\ref{css} to be associated with a coboundary election.
Sections \ref{casedbr} and \ref{m5b} show how the case of the IIA
D$p$-branes and $M$5-brane may also be treated within the same
framework \emph{i.e.}, how branes containing vector and tensor
fields on $W$ may be defined directly on suitably extended
superspaces. We shall argue,
in fact, that the picture is general and that suitable
target superspaces exist on which to define all the fields
appearing in the $p$-brane actions, including the various
vector (see also~\cite{Sak:98}), tensor, etc. worldvolume fields.
This is tantamount to establishing a general fields/extended
superspace democracy in which the worldvolume fields and the
extended superspace
variables are on the same footing, as it was already the
case for the minimal branescan. Indeed this correspondence
between coordinates and fields has occasionally been
discussed in the past in other contexts
(see \cite{Ber:79,Sch:80,Gay.Rom.Sch:81,Ald.Azc:83}).
Sec. \ref{2bn} contains a brief discussion of the origin
of the contributions to the Noether charges in the D-branes
case \cite{Ham:98} in our approach. Finally, an Appendix
complements the general theory of non-central extensions
of superspace in Sec. \ref{NCESA} and gives the proof of some
needed $\Gamma$-matrix identities.
\section{Central extensions and their superspaces}
\label{CESA}
\subsection{Standard superspace as a central extension}
\label{Ssce}
Let $\theta$ be an arbitrary Grassmann spinor
in a $D$-dimensional spacetime. Its components
$\theta^\alpha$ ($2^{[D/2]}$ where $[D/2]$ denotes the
integer part of $D/2$, or $2^{(D/2)-1}$ in the Weyl case)
determine an abelian group of supertranslations, generically
denoted $\text{sTr}_D$, with group composition law
\begin{equation}
\theta''^\alpha = \theta'^\alpha + \theta^\alpha
\quad.
\label{st}
\end{equation}
When the Lorentz part is considered explicitly, there is an action
$\rho$ of $Spin(1,D-1)$ on $\text{sTr}_D$ and the relevant
group becomes $\text{sTr}_D \circ Spin(1,D-1)$, where $\circ$
indicates semidirect product. Then~(\ref{st}) is replaced by
\begin{equation}
\theta'' = \theta' + \rho(A) \theta \quad, \quad \quad
A'' = A'A \quad,
\label{stlor}
\end{equation}
where $A \in Spin(1,D-1)$ and $\rho(A)$ is the appropriate
spin representation.
The spinor $\theta$ is
often restricted to be of some specific type, usually minimal
(\emph{e.g.}, Majorana (M), Weyl (W) or Majorana-Weyl (MW), when possible);
it may carry an
additional index $i=1,\dots,N$ if there is more
than one supersymmetry. Associated with~(\ref{st}) is the abelian Lie
superalgebra $\{D_\alpha,D_\beta\}=0$~\footnote{
Since we shall be considering left-invariant (LI) generators and forms,
we shall use here $D$'s (rather than $Q$'s) to denote the
generators of the right translations (the $Q$'s being realized
as the right-invariant (RI) generators
of the left translations). This distinction is of course
irrelevant for an abelian group such as
(\ref{st}) but it is not so when non-abelian parts are added
(nevertheless, the corresponding structure constants differ only
in a sign). Furthermore, LI and RI generators commute, $\{Q,D\}=0$.
We may look at the $D$'s as covariant
derivatives and at the $Q$'s as the generators of the (left)
supersymmetry transformations; see Sec. \ref{Nns}.},
which can also be described in terms of the left-invariant
(LI) one-forms $\Pi^\alpha=d\theta^\alpha$ and the trivial
Maurer-Cartan (MC) equation
\begin{equation}
d \Pi^\alpha = 0 \quad.
\label{2.2}
\end{equation}
\noindent Extending $\text{sTr}_D$ by the Minkowski
translations $x^\mu$ , $\mu=0,\dots , D-1$ leads to standard
(rigid) superspace~\cite{Ald.Azc:85,Azc.Izq:95}.
Let us adopt the free differential algebra
(FDA)\footnote{The term refers to an algebra generated
by differential forms which is closed under the action of
$d$~\cite{Sul:77}. For early physical applications of
FDA in supersymmetry see \cite{DAu.Fre.Reg:80,PvN:83} and
references therein.} point of view to discuss the extension problem,
since forms are especially convenient in the construction of
actions for extended objects.
Let $\theta^\alpha$ be Majorana, and
consider the two-form
$(C \Gamma^\mu)_{\alpha\beta} \Pi^\alpha\wedge \Pi^\beta$ on
$\text{sTr}_D$. It defines a
non trivial CE two-cocycle on the superalgebra of
$\text{sTr}_D$, \emph{i.e.} \ it is left-invariant (LI),
closed and not given by the differential of a LI one-form.
Since by construction
the two-cocycle transforms as a Lorentz
vector, it is consistent to extend the FDA~(\ref{2.2}) by a one-form
$\Pi^\mu$ such that
\begin{equation}
d \Pi^\mu = {1\over 2} (C \Gamma^\mu)_{\alpha\beta}
\Pi^\alpha \Pi^\beta
\label{2.3}
\end{equation}
(we omit the wedge product henceforth).
The above extension immediately implies
$\{D_\alpha,D_\beta\}=(C \Gamma^\mu)_{\alpha\beta} X_\mu$, with
$X_\mu$ central.
One still has to relate the newly introduced one-form to the
coordinate $x^\mu$. We define
\begin{equation}
\Pi^\mu = dx^\mu + {1\over 2} (C \Gamma^\mu)_{\alpha \beta}
\theta^\alpha d\theta^\beta
\quad
\label{2.3b}
\end{equation}
and choose the transformation law for $x^\mu$ so that
$\Pi^\mu$ is LI
\begin{equation}
x''^\mu = x'^\mu +x^\mu -{1 \over 2} (C \Gamma^\mu)_{\alpha\beta}
\theta'^\alpha \theta^\beta
\quad.
\label{xgcl}
\end{equation}
This gives rigid
superspace $\Sigma$, parametrized by $(\theta^\alpha,x^\mu)$ and
with group law given by (\ref{st}) and (\ref{xgcl}).
The above simple example
exhibits already the key features of the extension algorithm.
Given a particular FDA to be extended, one identifies in general
a non-trivial two-cocycle of a desired Lorentz covariant nature and
introduces a new LI one-form, the differential of which is given by
the cocycle. The new form (here, (\ref{2.3b})) together with the
MC equations (here, eqns. (\ref{2.2}) and (\ref{2.3}))
automatically define by duality an extended Lie algebra.
The new LI one-form is given by the sum of the
differential of the new group parameter and the potential one-form
of the CE two-cocycle on $\text{sTr}_D$, which is {\it not} LI.
Finally, the transformation properties of the
new coordinate are fixed so as to guarantee the left
invariance of the new one-form, while those of the original
manifold are unmodified. The additional one-form can be made LI
only if it is defined on
the \emph{extended} superspace manifold ${\tilde \Sigma}$.
The new (central) generator, associated with translations
along the new coordinates, modifies the r.h.s. of the
original commutators of the algebra.
Since adding (\ref{2.3}) to (\ref{2.2}) involves a \emph{central}
extension, we could have introduced a dimensionful constant\footnote{
The value of this constant determines the specific element in
cohomology space that characterizes the central extension.} as a
factor in the r.h.s. of (\ref{2.3}).
By not doing so, the dimensions of $\Pi^\mu$ are fixed to be
$[\Pi^\alpha]^2=[\theta^\alpha]^2$.
We shall, as usual, take
$[\theta^\alpha]=L^{1\over 2}$ so that $[\Pi^\mu]=L$.
If we add the Lorentz group, the result must reflect the action
$\sigma$ of $Spin(1,D-1)$ on the extension
cocyle (see, \emph{e.g.}\ \cite{Azc.Izq:95},
Sec. 5.3), but we shall not consider explicitly the effect of
the simple part of the algebra which, apart from
extracting the various tensor-valued second cohomology groups from
that of the trivial ($\sigma$=0) action $H_0^2(\text{sTr}_D)$
(see below), plays no essential r\^ole in our discussion once only
Lorentz covariant objects are used. Thus, \emph{central} means,
where appropriate, central up to Lorentz transformations.
The extension procedure described above can be applied more than
once ---there are two
basic patterns one may follow in this case. One can start in
each step with the same original manifold $\text{sTr}_D$,
and keep adding two-cocycles and
central generators or, in each step, one can consider
the result of the previous extension as the starting manifold.
In the first case (Sec.~\ref{Mces}) all new
generators remain central and appear only at the
r.h.s. of the original $\{D,D\}$ anticommutator. In the second case,
a richer structure emerges since the generators introduced
at each step can, in principle, modify all previous commutators.
We shall give the details of this second construction in
Sec.~\ref{NCESA}.
\subsection{Maximal central extensions of superspace}
\label{Mces}
Let $\theta^\alpha$ be Majorana.
We may obtain additional Lorentz tensors, leading to new central
charges, by considering
\begin{equation}
d\Pi^{\mu_1\dots \mu_p} \equiv {1\over 2}
(C \Gamma^{\mu_1\dots \mu_p})_{\alpha\beta} \Pi^\alpha \Pi^\beta
\, ,\qquad (\Gamma^{\mu_1\dots \mu_p} =
\Gamma^{[\mu_1} \Gamma^{\mu_2} \cdots
\Gamma^{\mu_p]}
\equiv {1 \over p!} \epsilon^{\mu_1 \dots \mu_p}_{\nu_1 \dots
\nu_p} \Gamma^{\nu_1} \dots \Gamma^{\nu_p}) \, ,
\label{2.4}
\end{equation}
where\footnote{%
We adopt $C\equiv C_{-}$ for simplicity.
By not considering $C_{+} \Gamma^{\mu} C_{+}^{-1} = {\Gamma^{\mu}}^T$
we rule out, \emph{e.g.}, the pseudoMajorana spinors that exist for $D=8,9$
$[\text{mod}\, 8]$ (see \cite{Nie:84}).%
}
$C \Gamma^{\mu} C^{-1} = - {\Gamma^{\mu}}^T$.
The antisymmetry in the Lorentz indices is needed to rule out trivial
dependences coming from the fact that
$\{\Gamma^\mu,\Gamma^\nu\}=2\eta^{\mu \nu}$.
The left invariance of the new forms in~(\ref{2.4})
\emph{requires} new group parameters
$\varphi^{\mu_1 \dots \mu_p}$ so that (\emph{cf.} (\ref{2.3b}))
\begin{equation}
\Pi^{\mu_1\dots \mu_p} =
d \varphi^{\mu_1 \dots \mu_p} +
{1\over 2}(C \Gamma^{\mu_1\dots \mu_p})_{\alpha \beta}
\theta^\alpha \Pi^\beta
\quad.
\label{2.8}
\end{equation}
The superalgebra generator (LI vector field) $Z_{\mu_1\dots \mu_p}$,
corresponding to $\Pi^{\mu_1\dots \mu_p}$, is realized
by $Z_{\mu_1\dots \mu_p}= \partial / \partial
\varphi^{\mu_1\dots \mu_p}$ on the extended group manifold.
At this stage there are no restrictions coming from the
Jacobi identity, equivalent to
$d(d \Pi^{\mu_1\dots \mu_p})= 0$, which follows
trivially from $d \Pi^\alpha =0$. This is an
alternative way of stating that the $p$-tensor-valued
mapping on $\text{sTr}_D \otimes \text{sTr}_D$,
\begin{equation}
\xi^{\mu_1\dots \mu_p} (\theta',\theta)=
\theta'^\alpha (C \Gamma^{\mu_1\dots \mu_p})_{\alpha \beta}
\theta^\beta\quad,
\label{jcocycle}
\end{equation}
satisfies trivially the two-cocycle condition
\begin{equation}
\xi(\theta,\theta') +
\xi(\theta + \theta' ,\theta'') =
\xi(\theta , \theta' + \theta'') +
\xi(\theta' ,\theta'')
\, .
\label{cocycle}
\end{equation}
The symmetry of $(C\Gamma ^{\mu_1 \dots \mu_p})_{\alpha \beta}$
is needed to prevent
the two-cocycle (\ref{jcocycle}) from being trivial (\emph{i.e.}, a
two-coboundary), since $\eta(\theta)$=
$\theta^\alpha (C \Gamma^{\mu_1\dots
\mu_p})_{\alpha \beta} \theta^\beta$ on $\text{sTr}_D$, which might
generate $\xi$ through $\xi_{cob}(\theta',
\theta)\equiv \eta(\theta'+\theta) - \eta(\theta')
-\eta(\theta)$, is identically zero.
Thus, (\ref{jcocycle}) defines a non-trivial extension.
For a given spacetime dimension $D$, the symmetry condition
restricts the rank of the tensors that are allowed in (\ref{2.4}).
Hence, the problem of finding all \emph{central}
extensions of the algebra $\{D_\alpha,D_\beta\}=0$
(or of the Lie FDA (\ref{2.2})) is reduced
to finding a basis of the symmetric space $\Pi^{(\alpha}\otimes
\Pi^{\beta)}$ in terms of tensors
$(C \Gamma^{\mu_1\dots \mu_p})_{\alpha\beta}$ symmetric in
$\alpha,\beta$; they define the Lie algebra CE two-cocycles
$\Pi^\alpha (C \Gamma^{\mu_1\dots \mu_p})_{\alpha \beta}\Pi^\beta$.
When $D$ is \emph{even}, the space of matrices with
indices $(\alpha \beta)$ is $2^D$-dimensional. Since
$2^{D}= \sum_{p=0}^D {D \choose p}$, a basis for this
space is provided by the $(2^D-1)$ matrices given by the Lorentz tensors
$\Gamma^{\mu_1\dots \mu_p}$ of rank $1\le p \le D$ plus the unit matrix.
For $D$ \emph{odd}, the spinors have dimension $2^{(D-1)/2}$ and, since
$2^{D-1} = \sum_{p=0}^{(D-1)/2} {D \choose p}$, a basis is
provided by the $(2^{D-1} -1)$ matrices given by the
tensors $\Gamma^{\mu_1\dots \mu_p}$
of rank $1\le p \le (D-1)/2$ plus the unit matrix.
The difference is a consequence of the fact that, for any $D$,
\begin{equation}
\Gamma^{\mu_1\dots \mu_p} \Gamma^{D+1} \propto
\epsilon^{\mu_1\dots \mu_D} \Gamma_{\mu_{p+1}\dots \mu_D}
\quad,
\label{2.6}
\end{equation}
where $\Gamma^{D+1}$ is the chirality matrix.
For $D$ odd, $\Gamma^{D+1}\propto 1$,
and only the tensors of rank $0\le p \le (D-1)/2$
are linearly independent.
For $D$ \emph{even}, $C \Gamma^{\mu_1\dots \mu_p}$
satisfies (see \emph{e.g.}, \cite{Wes:98})
\begin{equation}
\begin{array}{c}
\displaystyle
(C \Gamma^{\mu_1\dots \mu_p}) = \epsilon (-1)^{(p-1)(p-2)/2}
(C\Gamma^{\mu_1\dots \mu_p})^T \quad,\quad
\mu=0,1,\dots,D-1
\\[0.3cm]
\displaystyle
\epsilon = -\sqrt{2} \cos {\pi\over 4} (D+1)
\quad.
\end{array}
\label{2.9}
\end{equation}
Thus, $\epsilon$=$1$ $(-1)$ for $D$=$2,4\ (6,8)\ [\text{mod}\,8]$
so that $(C \Gamma^{\mu_1\dots \mu_p})_{\alpha\beta}$ is
symmetric for $p$= $1,2\ [\text{mod}\,4]$
if $D=2,4\ [\text{mod}\,8]$ and for $p=3,4$ if
$D=6,8\ [\text{mod}\,8]$.
For $D$ \emph{odd}, it turns out that the same condition,
$\epsilon (-1)^{(p-1)(p-2)/2}$=$1$, holds for
$D$=$3\ [\text{mod}\, 4]$ with $\epsilon$ =
$\displaystyle -\sqrt{2} \cos {\pi\over 4}D$.
We have excluded here (somewhat arbitrarily) the $D$=$5, 9$ cases
because in these dimensions no $C$ such that
$C \Gamma^{\mu} C^{-1} = - {\Gamma^{\mu}}^T$ exists.
The number of cohomology spaces
$H_{\sigma}^2(\text{sTr}_D \circ Spin(1,D-1))$
for various $\text{sTr}_D$ groups is given in the table.
As a two-form, the various CE two--cocycles
are given by $d\theta (C\Gamma^{\mu_1 \dots \mu_p}) d\theta$.
The corresponding new generators $Z_{\mu_1 \dots \mu_p}$ are
all central, as is $X_{\mu}$ itself. They are on the same footing
and may thought of as generalised momenta. Each of
the resulting extensions {\it defines} an extended superspace
group; we will denote them generically by $\tilde{\Sigma}$.
The table also includes the cases in which the spinor is
Majorana-Weyl or complex (Dirac and Weyl).
If the spinor is complex the independent tensors $\Gamma^0
\Gamma^{\mu_1\dots \mu_p}$ may appear. The effect of
considering Weyl spinors is taken into account by introducing
a chiral projector ($\mathcal{P}_+$, say).
The different extended supersymmetry algebras can be easily
found from the results in the table. We shall only give below
two examples which contain formulae that will be explicitly
used later on.
To avoid cumbersome
factorials, we use a normalization of the generators which is
tantamount to defining the duality relations by
$\Pi^{\mu_1\dots\mu_p}(Z_{\nu_1\dots\nu_p})=
\frac{1}{p!}\epsilon^{\mu_1\dots\mu_p}_{\nu_1\dots\nu_p}$ so that
$\Pi^{\mu_1\dots\mu_p}(C\Gamma^{\nu_1\dots\nu_p}Z_{\nu_1\dots\nu_p})=
C\Gamma^{\mu_1\dots\mu_p}$.
\begin{figure}
\begin{turn}{90}
\begin{minipage}{\textheight}
\scalebox{.9}
{
\begin{tabular}{|c|c|c|lcl|c|c|c|c|cc|}
\hline
\bf{1}
&
\bf{2}
&
\bf{3}
&
\,
&
\bf{4}
&
\,
&
\bf{5}
&
\bf{6}
&
\bf{7}
&
\bf{8}
&
\multicolumn{2}{c|}
{\bf{9}}
\\
\hline
$D$ &
$n$=$2^{[D/2]}$ & ${n(n+1)\over 2}$ &
\multicolumn{3}{l|}{
\begin{minipage}{0.1\textheight} real $n'$ [complex] dim. of spinor
\end{minipage}
}
&
\begin{minipage}{0.15\textheight}
rank $p$ of symmetric $(C \Gamma^{\mu_1\dots \mu_p})_{\alpha\beta}$
$\{(C \Gamma^{\mu_1\dots \mu_p}\mathcal{P}_+)_{\alpha\beta}\}$
\end{minipage}
&
\multicolumn{1}{l|}
{\begin{minipage}{0.18\textheight}
$\dim (C \Gamma^{\mu_1\dots \mu_p})$
$\{\dim (C \Gamma^{\mu_1\dots \mu_p}\mathcal{P}_+)\}$\, ;
total real [complex] dimension
\end{minipage}}
&
\multicolumn{1}{l|}{\begin{minipage}{0.1\textheight}
rank of
$(\Gamma^0 \Gamma^{\mu_1\dots \mu_p})$
$\{(\Gamma^0 \Gamma^{\mu_1\dots \mu_p}\mathcal{P}_+)\}$
\end{minipage}}
&
\begin{minipage}{0.13\textheight}
$\dim(\Gamma^0 \Gamma^{\mu_1\dots \mu_p})$
$\{\dim(\Gamma^0 \Gamma^{\mu_1\dots \mu_p}\mathcal{P}_+)\}$
\end{minipage}
&
\multicolumn{2}{l|}{
\begin{minipage}{0.14\textheight}
real $H^2_\sigma(\text{sTr}_D\circ Spin(1,D-1))$ spaces
$(\dim H^2_0(\text{sTr}_D))$
\end{minipage}
}
\\
\hline
2 & 2 & 3 & 1 & & {\bfseries MW} & \{1\} & \{$\frac 12$ 2 ; 1 \} &
\multicolumn{2}{c|}{} & 1 & (1)
\\
2 & 2 & 3 & 2 & & M & 1,2 & 2,1; 3 &
\multicolumn{2}{c|}{do not contribute further} &
2 & (3)
\\
3 & 2 & 3 & 2 & & {\bfseries M} & 1 & 3; 3 &
\multicolumn{2}{c|}{in the real case}
& 1 & (3)
\\
4 & 4 & 10 & 4 & & {\bfseries M} & 1,2 & 4,6; 10 &
\multicolumn{2}{c|}{} & 2 & (10)
\\
\hline
6 & 8 & 36 & 8 & [4] & {\bfseries W} & \{3\} & \{$\frac 12$ 20 ; 20 [10]\}
& \{1,3,5\} & \{6, $\frac 12$ 20 ; 16\} & 3 & (36)
\\
6 & 8 & 36 & 16 & [8] & D & 0,3,4 & 1,20,15 ; 72 [36] &
0,1,2,3,4,5,6 & 1,6,15,20,15,6,1 ; 64 & 10 & (136)
\\
7 & 8 & 36 & 16 & [8] & {\bfseries D} & 0,3 & 1,35 ; 72 [36] &
0,1,2,3, & 1,7,21,35 ; 64 & 6 & (136)
\\
8 & 16 & 136 & 16 & [8] & {\bfseries W} & \{0,4,8\} &
\{1, $\frac 12$ 70 ; 72 [36]\} & \{1,3,5,7\} &
\{8,56 ; 64\} & 4 & (136)
\\
8 & 16 & 136 & 32 & [16] & D & 0,3,4,7,8 &
1,56,70,8,1; 272 [136] & 0,1,2,3,4,5,6,7,8 & 1,8,28,56,70,56,28,8,1 ; 256 &
14 & (528)
\\
\hline
10 & 16 & 136 & 16 & & {\bfseries MW} & \{1,5,9\}
& \{10, $\frac 12$ 252; 136\} &
\multicolumn{2}{c|}{} & 2 & (136)
\\
10 & 32 & 528 & 32 & & M & 1,2,5,6,9,10 & 10,45,252,210,10,1; 528 &
\multicolumn{2}{c|}{do not contribute further} &
6 & (528)
\\
11 & 32 & 528 & 32 & & {\bfseries M} & 1,2,5 & 11,55,462; 528 &
\multicolumn{2}{c|}{in the real case}
& 3 & (528)
\\
12 & 64 & 2080 & 64 & & {\bfseries M}
& 1,2,5,6,9,10 & 12,66,792,924,220,66; 2080 &
\multicolumn{2}{c|}{} & 6 & (2080)
\\
\hline
\end{tabular}
}
\vspace*{0.5cm}
\newline
\begin{small}
\emph{Some Lie algebra second cohomology groups for
$\text{sTr}_D$ (minimal spinors are in boldface)}.
$n$ is the complex dimension of a Dirac spinor,
equal to the real dimension of Majorana spinors for
$D=2,3,4\ [\text{mod}\,8]$.
The fourth column gives the dimension of the spinor indicated.
The fifth and sixth column give the ranks for which
$(C \Gamma^{\mu_1\dots \mu_p})_{\alpha\beta}$
(or $(C \Gamma^{\mu_1\dots \mu_p} \mathcal{P}_+)_{\alpha\beta}$)
are symmetric (as deduced from
(\ref{2.9})) and the dimension of these Lorentz tensors; $C$ itself is
symmetric in $D=6,7,8\ [\text{mod}\,8]$.
The seventh and eigth columns do the same for the additional tensors
$(\Gamma^0 \Gamma^{\mu_1\dots \mu_p})$
$(\Gamma^0 \Gamma^{\mu_1\dots \mu_p}\mathcal{P}_+)$ appearing in
the complex spinor case. These hermitian (adding $i$ when needed)
tensors are limited by duality (eqn.~(\ref{2.6})) in the odd
($D=7$) case and to odd rank by the presence of $\mathcal{P}_+$
in the Weyl case. The $\frac 12$ indicates
halving due to self-duality.
The number of real cohomology groups
$H_\sigma^2(\text{sTr}_D\circ Spin(1,D-1))$
is given by the first number in the last column.
These spaces are the relevant (\emph{i.e.} tensorial) ones,
once the Lorentz symmetry is considered since in this case
$\text{sTr}_D\circ Spin(1,D-1)$
(rather than $\text{sTr}_D$)
is the group to be extended.
The action $\sigma$ of $Spin(1,D-1)$ on the extension cocycles
is automatically taken
into account by using only Lorentz covariant objects for them.
The bracketed number in the last column ignores the Lorentz
part and, as a result,
$\dim H^2_0(\text{sTr}_D) ={n' \choose 2} + n'$
since the elements of $\text{sTr}_D$ are odd
(for an ordinary $n'$-dimensional abelian group $\dim H_0^2 =
{n' \choose 2}$).
The number ${n' \choose 2} + n'$ is given nevertheless since it
serves as a check on the degrees of freedom: it is equal to the sum of the
total real dimensions in the sixth and eigth columns.
\end{small}
\end{minipage}
\end{turn}
\end{figure}
\subsection{Applications}
\subsubsection{N=1 theory extended superspace}
\label{sectene1}
For $D$ even, the basic spinors in (\ref{st}) may be reduced to
$2^{D/2-1}$-dimensional Weyl spinors,
and the discussion of the possible
$H_\sigma^2(\text{sTr}_{D}\circ Spin(1,D-1))$ spaces must take
this into account.
Let $D=2\ [\text{mod}\, 8]$ and let $\theta^\alpha$ be
MW. The symmetry of
$(C\Gamma^{\mu_1\dots \mu_p}\mathcal{P}_\pm)_{\alpha\beta}$
is now achieved if \emph{both}
$C\Gamma^{\mu_1\dots \mu_p}$ and $C\Gamma^{\mu_1\dots
\mu_p}\Gamma^{D+1}$
(or, on account of (\ref{2.6}), $C\Gamma^{\mu_1\dots \mu_{D-p}}$)
are symmetric .
Hence, there are central charges for $D=2,\ p=1$
and $D=10,\ p=1,5,9$ (\emph{i.e.}, $p$=1 $[\text{mod}\, 4]$).
As a result, the \emph{$D$=10,$\,$N=1 extended superspace algebra}
has the form
\begin{equation}
\{D_\alpha^+,D_\beta^+\}=
(C\Gamma^\mu \mathcal{P}_+)_{\alpha\beta} X_\mu
+ (C\Gamma^{\mu_1\dots \mu_5} \mathcal{P}_+)_{\alpha\beta}
Z_{\mu_1\dots \mu_5}
+ (C\Gamma^{\mu_1\dots \mu_9} \mathcal{P}_+)_{\alpha\beta}
Z_{\mu_1\dots \mu_9}
\quad.
\label{ene1}
\end{equation}
Due to $\mathcal{P}_+$ and to
(\ref{2.6}), the first and last term in the r.h.s
may be grouped into a single one,
$(C\Gamma^\mu \mathcal{P}_+)(X_\mu+Z_\mu)$;
classically, $Z_\mu$ may be absorbed by redefining $X_\mu$ and
the previous analysis shows that the vector--valued cohomology
space is one--dimensional. The second term may be rewritten as
$(C\Gamma^{\mu_1\dots \mu_5}) Z^+_{\mu_1\dots \mu_5}$ where
$Z^+_{\mu_1\dots \mu_5}$ is a self-dual 5-tensor \cite{Hol.Pro:82},
$\displaystyle Z^+_{\mu_1\dots \mu_5} =
{1\over 2}(Z_{\mu_1\dots \mu_5} +
{\epsilon_{\mu\dots\mu_5}}^{\mu_6\dots \mu_{10}}
Z_{\mu_6\dots \mu_{10}}$),
with half the number of components of $Z_{\mu_1\dots \mu_5}$.
As a result, the degrees of freedom in eqn. (\ref{ene1})
match:
${16\choose 2} + 16 = 136 = 10 + \frac{1}{2}
{10\choose 5}$ (see table). But in general $Z_\mu$
cannot be reabsorbed, since the Green-Schwarz action for the
heterotic superstring produces such a contribution to the
algebra \cite{Azc.Gau.Izq.Tow:89}, of an origin different
from that of $X_\mu$. Mathematically, this corresponds to
the fact that the group parameters are different for $X_\mu$
(translations $x^\mu$) and $Z_\mu$ ($\varphi^\mu$);
they are locally equivalent, much in the same way $\mathbb{R}\sim S^1$
locally, but they are different globally.
We may, however, achieve the symmetry under the exchange
of $X_9$
and $Z_9$ (say) when the 9-direction is a circle of radius $R$.
Then the spectra of $X_9$ and $Z_9$ are isomorphic under the
T-duality exchange $R\to 1/R$ (see \cite{Tow:97,Giv.Por.Rab:94}).
The FDA form of the $D$=10, $N$=1 superalgebra (\ref{ene1}) is
given by the MC relations
\begin{equation}
d\Pi^\alpha =0\, , \qquad
d\Pi^\mu = {1\over 2} (C\Gamma^\mu)_{\alpha\beta} \Pi^\alpha \Pi^\beta
\, ,
\qquad d\Pi^{\mu_1\dots \mu_5} =
{1\over 2} (C\Gamma^{\mu_1\dots \mu_5})_{\alpha\beta} \Pi^\alpha \Pi^\beta
\, ,
\label{MCene1}
\end{equation}
where $\Pi^\alpha$, $\Pi^\mu$ and $\Pi^{\mu_1\dots \mu_5}$ are
defined as ($\alpha=1,\dots,32$)
\begin{equation}
\Pi^\alpha = \mathcal{P}_+ d \theta^\alpha
\, , \qquad
\Pi^\mu = dx^\mu + {1\over2} (C\Gamma^\mu)_{\alpha\beta} \theta^\alpha
\Pi^\beta
\, , \qquad
\Pi^{\mu_1\dots \mu_5} = d \varphi^{\mu_1\dots \mu_5} +
{1\over2} (C\Gamma^{\mu_1\dots \mu_5})_{\alpha\beta} \theta^\alpha
\Pi^\beta
\, .
\end{equation}
If $Z_\mu$ is included separately, this
introduces a further extension which requires adding a new LI
form associated with it, $\Pi_\mu^{(\varphi)}$,
$d\Pi_\mu^{(\varphi)}={1 \over 2} (C \Gamma_{\mu}
\mathcal{P}_+)_{\alpha \beta} d\theta^\alpha
d\theta^\beta$ (we have written the index down for consistency
with later notation, as in Sec. \ref{css}).
At the group level this means that the MW
translations generate two types of transformations
\emph{i.e.}, one has to distinguish between the translations $x^\mu$ and
the $\varphi_\mu$, some of which may be compact, in which case
the corresponding group law expression should be understood locally.
\subsubsection{IIA theory centrally extended superspace}
Let us consider now the
$H_\sigma^2(\text{sTr}_{10}\circ Spin(1,9),\text{(IIA)})$ spaces.
The IIA superalgebra is the $D=10$ algebra associated with
two $16$-dimensional spinors of opposite chiralities which may be
combined into a Majorana spinor. Then (see table),
the \emph{IIA theory maximally extended algebra} \cite{Hol.Pro:82}
is found to be
\begin{eqnarray}
\{D_\alpha,D_\beta\} & = &
(C \Gamma^{\mu})_{\alpha\beta} X_\mu
+
(C \Gamma^{\mu_1 \mu_2})_{\alpha\beta} Z_{\mu_1 \mu_2}
+
(C \Gamma^{\mu_1 \dots \mu_5})_{\alpha\beta} Z_{\mu_1 \dots \mu_5}
\nonumber \\
& & \mbox{} +
(C \Gamma^{11})_{\alpha \beta} Z
+
(C \Gamma^{\mu}\Gamma^{11})_{\alpha \beta}Z_{\mu}
+
(C \Gamma^{\mu_1 \dots \mu_4} \Gamma^{11} )_{\alpha \beta}
Z_{\mu_1 \dots \mu_4}
\quad,
\label{2.14}
\end{eqnarray}
since the tensor spaces
\{$\Gamma^{\mu_1\dots \mu_p}$\} and \{$\Gamma^{\mu_1\dots
\mu_{D-p}}\Gamma^{11}$\} are isomorphic by eqn.~(\ref{2.6}).
Notice that $X_\mu$ and $Z_\mu$ belong to different cohomology classes
(their corresponding two-cocycles are not cohomologous
due to the presence of $\Gamma^{11}$ in the $Z$ term).
The associated IIA Lie FDA, involving the LI one-forms dual to
the generators in (\ref{2.14}), is given by
\begin{equation}
\addtolength{\arraycolsep}{-1ex}
\begin{array}{rclcrclcrcl}
d \Pi^\alpha & = & 0
\, , & &
d \Pi^\mu & = & {1\over 2}
(C \Gamma^{\mu})_{\alpha\beta} \Pi^\alpha \Pi^\beta
\, , & &
d \Pi^\mu_{(z)} & = & {1\over 2} (C \Gamma^{\mu}\Gamma^{11})_{\alpha
\beta} \Pi^\alpha\Pi^\beta
\\
d \Pi^{\mu_1 \mu_2} & = & {1\over 2} (C \Gamma^{\mu_1 \mu_2})_{\alpha
\beta} \Pi^\alpha \Pi^\beta
\, , & &
d \Pi^{\mu_1 \dots \mu_5} & = & {1\over 2}
(C \Gamma^{\mu_1 \dots \mu_5})_{\alpha\beta} \Pi^\alpha \Pi^\beta
\, , & &
& &
\\
d \Pi^{\mu_1 \dots \mu_4} & = & {1\over 2}
(C \Gamma^{\mu_1 \dots \mu_4} \Gamma^{11})_{\alpha \beta}
\Pi^\alpha \Pi^\beta
\, , & &
d \Pi & = & {1\over 2} (C \Gamma^{11})_{\alpha \beta} \Pi^\alpha
\Pi^\beta
\, .
\end{array}
\label{2.15}
\end{equation}
The new group parameters define the \emph{IIA theory centrally extended
superspace}, parametrized by the coordinates
$(\theta^\alpha,x^\mu,\varphi^\mu,\varphi^{\mu_1 \mu_2},
\varphi^{\mu_1 \dots \mu_5},
\varphi^{\mu_1 \dots \mu_4},\varphi)$.
The IIB case with $\Pi^{\alpha\,i} \equiv \mathcal{P}_+
d\theta^{\alpha\,i}$
$(i=1,2)$ is treated similarly by noticing that the presence of
$\epsilon_{ij}$ allows for $C\Gamma^{\mu_1\mu_2\mu_3}\mathcal{P}_+$,
which is skew-symmetric.
\section{Non-central extensions and their superspaces}
\label{NCESA}
We start now from standard rigid superspace, eqns.~(\ref{2.2}),
(\ref{2.3}) for real, odd translations.
To keep the discussion as general as possible, we rescale
$\Pi^\mu$, $\Pi_{\mu_1\dots \mu_p}$ by an arbitrary dimensionless
constant $a_s$, so
that~(\ref{2.3}), (\ref{2.4}) become
\begin{equation}
d \Pi^\mu = a_s (C \Gamma^\mu)_{\alpha\beta}
\Pi^\alpha \Pi^\beta
\quad,\quad
d\Pi_{\mu_1\dots \mu_p} \equiv a_0
(C \Gamma_{\mu_1\dots \mu_p})_{\alpha\beta} \Pi^\alpha \Pi^\beta
\quad.
\label{dPimu}
\end{equation}
Let us fix $p$ and consider the
resulting extended superspace, parametrized by
$(\theta^\alpha, x^\mu, \varphi_{\mu_1 \dots \mu_p})$, as our
starting group manifold. We look for a non-trivial CE two-cocycle
with $p$ indices on the above extended superspace.
This may now involve any of the LI forms available,
$\Pi^\mu, \Pi^\alpha$ or $\Pi_{\mu_1 \dots \mu_p}$. Inspection of the
possible Lorentz tensors shows that the external Lorentz
indices of this two-cocycle have to be of the type
$(\mu_1 \dots \mu_{p-1}
\alpha_1)$ and, hence, the only available LI two-forms are
\begin{equation}
\rho^{(1)}_{\mu_1 \dots \mu_{p-1} \alpha_1}=(C \Gamma_{\nu \mu_1 \dots
\mu_{p-1}})_{\beta \alpha_1} \Pi^\nu \Pi^\beta
\quad, \quad \quad
\rho^{(2)}_{\mu_1 \dots \mu_{p-1} \alpha_1}= (C \Gamma^\nu)_{\beta
\alpha_1} \Pi_{\nu \mu_1 \dots \mu_{p-1}} \Pi^\beta
\label{cand1}
\quad .
\end{equation}
For $p=1$, both are closed. For $p \geq 2$,
$d (\rho^{(1)} + \lambda_2 \rho^{(2)})=0$ gives
$\displaystyle \lambda_2={a_s \over a_0}$
provided\footnote{
Primed indices are understood to be symmetrised (with unit weight).}
\begin{equation}
(C \Gamma^\nu)_{\alpha' \beta'} (C \Gamma_{\nu \mu_1 \dots
\mu_{p-1}})_{\gamma' \delta'}=0
\quad,
\label{Gammaprop}
\end{equation}
which holds only for certain values of
$(D,p)$~\cite{Ach.Eva.Tow.Wil:87} (for $p=1$, $D=3,4,10$ and, with
the appropriate modifications for complex spinors, $D=6$). The
existence of such a constraint, on both $D$ and $p$, is a new
feature---the non-triviality of (\ref{jcocycle})
only restricted $p$. We introduce now a new one--form
$\Pi_{\mu_1 \dots \mu_{p-1} \alpha_1}$ with
\begin{equation}
d\Pi_{\mu_1 \dots \mu_{p-1} \alpha_1} = a_1 \left((C \Gamma_{\nu
\mu_1 \dots \mu_{p-1}})_{\beta \alpha_1} \Pi^\nu \Pi^\beta
+ {a_s \over a_0} (C\Gamma^\nu)_{\beta \alpha_1} \Pi_{\nu \mu_1
\dots \mu_{p-1}} \Pi^\beta)\right)
\label{dPim1a1}
\end{equation}
(for $p=1$ the coefficient of the second term can be arbitrary,
see Sec.~\ref{css}).\footnote{
For $p=1$, the one-form in the l.h.s. of~(\ref{dPim1a1}) becomes
$\Pi_\alpha$ -- notice that this is unrelated to $\Pi^\alpha$ (so
that \emph{e.g.} \ $d\Pi_\alpha$ is non-zero). In general, we will not
raise or lower the Lorentz indices of forms, their position being
used to distinguish between different types of them as in
\cite{Sez:97}.}
The above MC equation implies that both $\left[D,X\right]$
and $\left[D,Z^{\mu_1 \dots \mu_p}\right]$ are
modified by a
term proportional to $Z^{\mu_1 \dots \mu_{p-1} \alpha_1}$, the
latter being the only central generator at this stage ($Z^{\mu_1
\dots \mu_{p-1} \alpha_1}$ is central because,
by construction, $\Pi_{\mu_1 \dots\mu_{p-1} \alpha_1}$
cannot appear at the r.h.s. of a MC equation
expressing the differential of a LI form).
This is a general feature of the extension scheme in this section:
at any stage in the chain of extensions, the only central
generator present is the last one introduced. Thus, each extension
is central, but the resulting algebra/group is not a central
extension of superspace: all generators but the last one
have non-zero commutators as a consequence of the
subsequent extensions. A second feature here is that
successive extensions substitute one spinorial index for a vectorial
one, preserving the total number of indices. The chain
ends with the introduction of a generator
with $p$ spinorial indices.
Repeating the above procedure, one finds that the next three
extensions are in some sense exceptional (see~(\ref{dPim1a24})
below), while the one introducing five spinorial indices and
all others after it follow a pattern which can be used to
derive a recursion formula. Skipping the somewhat involved
algebra (see Appendix A), we list first the results for the next
three extensions
\begin{eqnarray}
d \Pi_{\mu_1 \dots \mu_{p-2} \alpha_1 \alpha_2} & = &
a_2 \left( (C \Gamma_{\nu \rho \mu_1 \dots \mu_{p-2}})_{ \alpha_1
\alpha_2} \Pi^\nu \Pi^\rho
+ {a_s \over a_0} (C \Gamma^\nu)_{\alpha_1 \alpha_2}
\Pi_{\nu \rho \mu_1 \dots \mu_{p-2}} \Pi^\rho \right. \nonumber \\
& & - {a_s \over a_1} \left. (C \Gamma^\nu)_{\alpha_1 \alpha_2}
\Pi_{\nu \mu_1 \dots \mu_{p-2} \beta} \Pi^\beta
-8 {a_s \over a_1} (C \Gamma^\nu)_{\alpha'_1 \beta}
\Pi_{\nu \mu_1 \dots \mu_{p-2} \alpha'_2} \Pi^\beta \right)
\quad,
\nonumber \\
d \Pi_{\mu_1 \dots \mu_{p-3} \alpha_1 \alpha_2 \alpha_3} & = &
a_3 \left( (C \Gamma^\nu)_{\alpha'_1 \alpha'_2}
\Pi_{\nu \rho \mu_1 \dots \mu_{p-3} \alpha'_3} \Pi^\rho \right.
+ {5a_1 \over 4a_2} (C \Gamma^\nu)_{\alpha'_1 \beta}
\Pi_{\nu \mu_1 \dots \mu_{p-3} \alpha'_2 \alpha'_3} \Pi^\beta
\nonumber \\
& & + {a_1 \over 4a_2} \left. (C \Gamma^\nu)_{\alpha'_1 \alpha'_2}
\Pi_{\nu \mu_1 \dots \mu_{p-3} \beta \alpha'_3} \Pi^\beta \right)
\quad,
\nonumber \\
d \Pi_{\mu_1 \dots \mu_{p-4} \alpha_1 \alpha_2 \alpha_3
\alpha_4} & = &
a_4 \left( (C \Gamma^\nu)_{\alpha'_1 \alpha'_2}
\Pi_{\nu \rho \mu_1 \dots \mu_{p-4} \alpha'_3 \alpha'_4} \Pi^\rho
\right.
- {48a_sa_2 \over 5a_1 a_3} (C \Gamma^\nu)_{\alpha'_1 \beta}
\Pi_{\nu \mu_1 \dots \mu_{p-4} \alpha'_2 \alpha'_3 \alpha'_4}
\Pi^\beta \nonumber \\
& & - {12a_s a_2\over 5a_1 a_3} \left.
(C \Gamma^\nu)_{\alpha'_1 \alpha'_2}
\Pi_{\nu \mu_1 \dots \mu_{p-4} \beta \alpha'_3 \alpha'_4} \Pi^\beta
\right)
\label{dPim1a24}
\end{eqnarray}
(the $a_k$'s in the r.h.s. normalise the $\Pi$'s with $k$
spinorial indices). For the remaining extensions, which introduce
one-forms with five or more spinorial indices, one
establishes the following recursion formula
\begin{eqnarray}
d \Pi_{\mu_1 \dots \mu_{p-(k+2)} \alpha_1 \dots \alpha_{k+2}} & = &
a_{k+2} \left\{ (C \Gamma^\nu)_{\alpha'_1 \alpha'_2}
\Pi_{\nu \rho \mu_1 \dots \mu_{p-(k+2)} \alpha'_3 \dots
\alpha'_{k+2}} \Pi^\rho \right. \nonumber \\
& &
+ \lambda^{(k+2)}_2 (C \Gamma^\nu)_{\alpha'_1 \beta}
\Pi_{\nu \mu_1 \dots \mu_{p-(k+2)} \alpha'_2 \dots \alpha'_{k+2}}
\Pi^\beta \nonumber \\
& &
\left. + \lambda^{(k+2)}_3 (C \Gamma^\nu)_{\alpha'_1 \alpha'_2}
\Pi_{\nu \mu_1 \dots \mu_{p-(k+2)} \beta \alpha'_3 \dots
\alpha'_{k+2}}
\Pi^\beta \right\}
\quad,
\label{dPim1ak2}
\end{eqnarray}
where
\begin{equation}
\lambda^{(k+2)}_2 = -{a_s \over a_{k+1}}\left( {2 \over
\lambda^{(k+1)}_2} + {k \over \lambda^{(k+1)}_3}\right) \quad ,\quad
\lambda^{(k+2)}_3 = -{a_s \over a_{k+1}}{k+1 \over
\lambda^{(k+1)}_2}
\quad.
\label{la23}
\end{equation}
Notice that the above recursion starts at $k=3$, which implies $p
\geq 5$. On the other hand, the maximum value of $p$ (of
interest to us) for
which~(\ref{Gammaprop}) holds true is $p=5$,
\emph{i.e.}~(\ref{dPim1ak2}) is relevant here only for $k=3$, $p=5$.
It is easily checked that $\left[ \Pi_{\mu_1 \dots \mu_{p-l} \alpha_1
\dots \alpha_l} \right]=L^{1+{l \over 2}}$.
We give related explicit results, for $p=1,2$, in Sec.~\ref{p12}.
\section{Structure of the new superspaces and Noether currents}
\label{Nns}
\subsection{Fibre bundle structure}
\label{fbs}
All extended superspaces have a natural
bundle structure, in which the basis is the group to be extended
and the fibre is the group by which we extend. For instance,
superspace $\Sigma$ itself and the various extensions
$\tilde \Sigma$ in Sec. \ref{Mces} may be considered as the
total spaces of principal bundles over the $\text{sTr}_D$'s
of the specific theory. The two-forms which define the
extensions are curvatures of invariant
connections valued on the central algebras by
which $\text{sTr}_D$ is extended. The $D$'s are then the
horizontal lifts of the vector fields
${\partial \over\partial \theta^\alpha}$ on the specific
base manifold $\text{sTr}_D$; this justifies the `covariant
derivative' name which may be given to the $D$'s in
the algebra of the $\tilde\Sigma$'s.
Similar considerations apply at any step in the chain of extensions
in Sec. \ref{NCESA}. From this point of view, after the last
step, one has a
bundle structure with the last coordinate in the fibre and all
the rest in the base. As we show below, there is also
another relevant
bundle structure with $\Sigma$ in the
base and all new coordinates in the fibre.
Let us now discuss the general case treated in
Sec.~\ref{NCESA}. $\tilde{\Sigma}$ is parametrised by the
coordinates $(\theta^\alpha, x^\mu, \varphi_{\mu_1 \dots
\mu_p},\varphi_{\mu_1 \dots\mu_{p-1}\alpha_1},\dots,
\varphi_{\alpha_1 \dots \alpha_p})$. We will denote
them collectively by the row vector $\phi = (z^a, \varphi_A)$ where $z$
parametrizes the base (superspace $\Sigma$) and $\varphi$ the fibre (the
space of all new coordinates). Referring to
this block form, we will say that the superspace part is
`two--dimensional' while the fibre part has `dimension' $p+1$.
The LI
one--forms that reduce to the differentials of these coordinates
at the identity will be denoted by $\Pi=(H^a \, , \, \, \Theta_A)$
and the
dual LI vector fields by the column vector $Z=(D_a \, , \, \, Y^A)^t$
($t$
denotes matrix transposition). The corresponding RI objects will
carry an additional hat.
Under a right group transformation, $g
\rightarrow gg'$, $Z$ transforms like $Z \rightarrow T'^t Z$,
$T'$
being a matrix of (primed) functions on the group, called the
adjoint representation. It holds
\begin{equation}
T''=T'T \quad,
\quad \quad \quad
Z \cdot T \vert_e = \rho_{\hbox{\small adj}}(Z)
\quad,
\label{Tprop}
\end{equation}
$\rho_{\mbox{\small adj}}(Z)$ being the adjoint representation of $Z$,
given by the structure constants. Inspection of the MC equations
then reveals that $T$ is a lower triangular matrix with units along
the diagonal. We put accordingly
\begin{equation}
T^t =\left( \begin{array}{cc} A & C \\ 0 & B \end{array} \right)
\quad,
\quad \quad \quad
(T^t)^{-1} =\left( \begin{array}{cc} A^{-1} & -A^{-1} C B^{-1} \\
0 & B^{-1} \end{array} \right)
\label{Tblock}
\end{equation}
with $A$, $B$ upper triangular matrices. The dimensions of $A$,
$B$, $C$ (in block form) are $2 \times 2$, $(p+1) \times (p+1)$,
$2 \times (p+1)$ respectively. (\ref{Tblock}) shows that the
fibre is a subgroup, with adjoint representation given by $B^t$.
In this notation, the LI vector fields transform like
\begin{equation}
\left( \begin{array}{c} D \\Y \end{array} \right)
\rightarrow
\left( \begin{array}{c} A' \, D + C' \, Y \\
B' \,Y
\end{array} \right)
\quad .
\label{Ztrans}
\end{equation}
The LI forms similarly transform
according to $\Pi \rightarrow \Pi (T'^t)^{-1}$, \emph{i.e.}
\begin{equation}
( \begin{array}{cc} H \, , & \Theta \end{array} )
\rightarrow
( \begin{array}{cc} H \, A'^{-1} \, , & -H \, A'^{-1} C' B'^{-1} +
\Theta \, B'^{-1} \end{array} )
\quad .
\label{Pitrans}
\end{equation}
For the RI objects it holds $\hat{Z}=(T^t)^{-1} \, Z$,
$\hat{\Pi}= \Pi \, T^t$ \emph{i.e.}
\begin{equation}
\left( \begin{array}{c} \hat{D} \\ \hat{Y} \end{array} \right)
=
\left( \begin{array}{c} A^{-1} \, D - A^{-1}CB^{-1} \, Y \\
B^{-1} \, Y \end{array} \right)
\quad, \quad \quad \quad
( \begin{array}{cc} \hat{H} \, , & \hat{\Theta} \end{array} )
=
( \begin{array}{cc} H \, A \, , & H \, C + \Theta \, B \end{array})
\quad.
\label{ZPihat}
\end{equation}
The Lie algebra valued one--form $\omega = \Theta \, Y$ serves as a
connection in the bundle. Indeed, one easily verifies that $\omega$ is
invariant under~(\ref{Ztrans}),~(\ref{Pitrans}) when $T$ is
restricted to the subgroup of the fibre ($A=I$, $C=0$). The horizontal
subspace is spanned by the kernel of $\omega$, \emph{i.e.} \ by
the components of $D$, the latter being the horizontal lifts of the
standard superspace generators $D^{(s)}_\alpha ={\partial \over
\partial \theta^\alpha}$, $X^{(s)}_\mu = {\partial \over \partial
x^\mu} + {1 \over 2} (C \Gamma_\mu)_{\alpha\beta} \theta^\beta {\partial \over \partial
\theta^\alpha}$.
In later applications, in section~\ref{p12}, the
explicit form of the matrix $B^{-1}$ is needed -- we present here
a few remarks that facilitate its computation. Inspection of the
r.h.s. of the MC equations for the new one--forms, in
section~\ref{NCESA}, shows that they always contain one new
one--form, multiplied by a $\Pi^\alpha$ or $\Pi^\mu$. For the
dual Lie algebra this implies that the new generators commute
among themselves and only have, in general, non--zero commutators
with the superspace generators $D_\alpha$, $X_\mu$. In other
words, the group by which we extend $\Sigma$ is {\em abelian}
(to begin with) and its generators
acquire, as a result of the extension, non--zero commutators only
with the superspace generators.
The structure of the resulting Lie algebra is, in symbolic form,
\begin{equation}
[D,D] \sim D+Y
\quad, \quad \quad \quad
[D,\, Y] \sim Y
\quad, \quad \quad \quad
[Y,\, Y] = 0
\quad,
\label{LAsymb}
\end{equation}
in the notation introduced earlier. For $T$, the expression
\begin{equation}
T = e^{\phi^A \, \rho_{adj}(Z_A)} \equiv e^{z^a \, \rho_{adj}(D_a)
+ \varphi_A \, \rho_{adj}(Y^A)}
\label{Texpr}
\end{equation}
is well known. From the third of~(\ref{LAsymb}) though, we infer that
the restriction of $\rho_{adj} \, (Y)$ to the fibre (\emph{i.e.} \
to the sub-block corresponding to $B^t$) is zero.
Denoting this sub-block by $\rho^{(f)}_{adj} \, (Y)$ we find for
$B$
\begin{equation}
B = e^{z^a \, \, \rho^{(f)}_{adj} \, (D_a)^t} \equiv e^{\theta^\alpha
\, \, \rho^{(f)}_{adj} \, (D_\alpha)^t + x^\mu \, \,
\rho^{(f)}_{adj} \, (X_\mu)^t}
\quad,
\label{Bexpr}
\end{equation}
where the matrices $\rho^{(f)}_{adj} \, (D_a)$ are given by
the structure constants that appear in (the explicit form of) the
second of~(\ref{LAsymb}). The interesting point here is that $B$
depends on $(\theta,x)$ only -- the new variables enter in $T$
only through $A$, $C$.
\subsection{Invariant actions for the minimal branescan}
\label{seWZt}
As already mentioned, part of
the motivation for studying superspace extensions comes from
their relevance in the construction of manifestly invariant
$p$--brane actions. For the branes of the old branescan,
WZ terms on $\Sigma$ have the form
\begin{equation}
S_{WZ} = \int_W d^{p+1}\xi \, {\cal L}_{WZ} = \lambda \int_W
\phi^{*} (b)
\quad,
\label{3.1}
\end{equation}
where $b$ is defined\footnote{For an explicit form of the
quasi-invariant $b$ on $\Sigma$ see \cite{Eva:88}.}
as the potential of the closed $(p+2)$--form $h$ on superspace
\begin{equation}
h = (C \Gamma_{\mu_1\dots \mu_p})_{\alpha\beta} \Pi^{\mu_1}\dots
\Pi^{\mu_p}
\Pi^\alpha \Pi^\beta \quad,\quad db=h
\quad.
\label{3.2}
\end{equation}
$W$ in~(\ref{3.1}) is the $(p+1)$--dimensional worldvolume
swept out by the $p$--brane, parametrized by
$\{\xi^i\} =(\tau, \sigma^1, \dots ,\sigma^p), i=0,1\dots p$
and $\phi^{*}$ is the pullback of the embedding $\phi: W
\rightarrow \Sigma$.
The constant $\lambda$ is fixed by the requirement of
$\kappa$--invariance of the total action \cite{Ach.Eva.Tow.Wil:87}
(we will ignore $\lambda$ henceforth).
As is well known \cite{Ach.Eva.Tow.Wil:87} (see also~\cite{Azc.Tow:89}),
the closure of $h$ is equivalent to the condition (\ref{Gammaprop})
which we have seen to guarantee the existence of the non--central
extensions of Sec.~\ref{NCESA}. Using the new LI one--forms available
we may obtain a LI potential $\tilde{b}$ for $h$ on $\tilde{\Sigma}$.
Its general form is
\begin{equation}
\tilde{b} = \sum_{k=0}^p \Pi_{\mu_1\dots \mu_{p-k} \alpha_1 \dots
\alpha_k} ( b_{k}
\Pi^{\mu_1}\dots \Pi^{\mu _{p-k}} \Pi^{\alpha_1} \dots \Pi^{\alpha_k}
) \equiv \Theta_C \Lambda^C
\quad,
\label{bdef}
\end{equation}
where the last equation uses the notation of the previous section and
defines $\Lambda^C$.
The $b_k$'s are numerical constants, determined by the second eqn.
in ~(\ref{3.2}). We check that $[ \tilde{b}]=
L^{1+{k \over 2}} L^{p-k} L^{k \over 2} = L^{p+1}$.
We compute now the explicit form for an invariant ${\cal L}_{WZ}$
in~(\ref{3.1}) (\emph{i.e.}, with $\tilde{b}$ instead of $b$). For a general
one--form $\Pi$ we put $\phi^{*}(\Pi)
\equiv \Pi_i d\xi^i$, so that
\begin{equation}
\Pi^\alpha_i = \partial_i{\theta^\alpha}, \quad \quad \quad
\Pi^\mu_i = \partial_i{x^\mu} + a_s (C \Gamma^\mu)_{\alpha \beta}
\theta^\alpha \partial_i{\theta^\beta}
\quad.
\label{Pialmui}
\end{equation}
Eqns.~(\ref{bdef}) and~(\ref{3.1}) give for the Lagrangian
\begin{equation}
{\cal L}_{WZ} = \Theta_{Ci} \Lambda^{Ci}
\quad,
\label{Lexpr2}
\end{equation}
where, expanding multi-indices and using the above notation,
\begin{equation}
\Lambda^{\mu_1 \dots \mu_{p-k} \alpha_1 \dots \alpha_k i} \equiv
b_k \epsilon^{ij_1 \dots j_p}
\Pi^{\mu_1}_{j_1} \dots \Pi^{\mu_{p-k}}_{j_{p-k}}
\Pi^{\alpha_1}_{j_{p-k+1}} \dots \Pi^{\alpha_k}_{j_p}
\quad.
\label{LCidef}
\end{equation}
\subsection{Noether currents for the new symmetries}
\label{noecur}
The invariance under translations of the
action of the supersymmetric objects implies
the existence of conserved currents. The integrals of the charge
densities over a spacelike section of the worldvolume give
constants of the motion for the
$p$--brane. When the action contains the standard WZ term $b$, the
Noether current includes a term $\Delta$ coming from the
quasi-invariance of $b$, $\delta b=d \Delta$. This was used
in~\cite{Azc.Gau.Izq.Tow:89} to
find topological extensions of the supersymmetry algebra. When $b$ is
replaced by the invariant $\tilde{b}$ in~(\ref{bdef}), $\Delta$ is no
longer present. However, the Noether current receives now a
contribution from the additional fields $\phi^{*} (\varphi_{A})$,
which leads to the same result.
One can derive general expressions for the currents
associated with the new generators.
In the present case,
where the relevant part of the Lagrangian is obtained
by pulling back to $W$ forms initially defined on
$\tilde{\Sigma}$, it is convenient to work
on the extended superspace, where quantities have
a direct geometrical interpretation, and to pull the
result back to $W$ at the end.
To keep the discussion general, consider a manifold
$M$ which can serve as worldvolume and a target space $N$,
of dimensions $m$, $n$ respectively ($m < n$) and an
embedding $\phi$ of $M$ into $N$, $\phi: x \mapsto y(x)$ where
$\{x^i\}$ ($\{y^j\}$) are local coordinates on $M$ ($N$). Consider
furthermore an action $S$ given by
\begin{equation}
S = \int_M \phi^*(\alpha)
\quad,
\label{Sdef}
\end{equation}
where $\alpha$ is a $k$--form on $N$ and $\phi^*$ is the pullback
map associated with the embedding. We assume that the
submanifolds $x^0=x^0_{init}$, $x^0=x^0_{fin}$ of $M$ form its
boundary
$\partial M$ -- their embeddings in $N$ are the initial and final
configuration respectively. The equations of motion are
\begin{equation}
\delta_Y S= \int_M \phi^*( L_Y \alpha)=0
\quad,
\label{deltaY}
\end{equation}
where $Y$ is an arbitrary vector field on $N$
which vanishes on $\phi(\partial M)$ -- we denote their solutions
generically by $\phi_{cl}$.
Proceeding along the lines of the standard derivation, with inner
derivations taking up the role of the partials
$\frac{\partial}{\partial \phi_{,i}}$, one finds the equations of
motion in the form
\begin{equation}
\phi^*(i_Y d \alpha)=0 \quad ,\label{eom3}
\end{equation}
where now $Y$ is an arbitrary vector field, not necessarily
vanishing on $\partial M$.
For a symmetry generated by $Y_0$ one obtains
\begin{equation}
d \left( \phi^*_{cl} (J_{(Y_0)}) \right) = 0
\, , \qquad \qquad
J_{(Y_0)} \equiv i_{Y_0} \alpha
\, ,
\label{Jcons}
\end{equation}
which is the current conservation equation.
For a quasi-invariant Lagrangian, $\phi^*(L_{Y_0}
\alpha)=\phi^*(d\Delta)$, the conserved current picks up
a term in
$\Delta$, $\phi^*(J_{(Y_0)})=\phi^*(i_{Y_0} \alpha -
\Delta)$.
In the present case, $(M,N,\alpha)$ correspond to
$(W,\tilde{\Sigma},{\tilde b})$. The variation of the total action
from that of
the new coordinates comes only from ${\cal L}_{WZ}$ so
that~(\ref{eom3}) gives
\begin{equation}
0= \phi^*(i_{Y}d \tilde{b}) = \phi^*(i_{Y} h)
\quad,
\label{eomS1}
\end{equation}
where $Y$ is an arbitrary vector field along the fibre. Since $h$
is horizontal, the above equations of motion obtained from
variations of the $\varphi_A$'s, are satisfied trivially,
consistent with the appearence of the
$\varphi_A$'s in the Lagrangian through exact differentials. For
the Noether currents associated to translations along the new
coordinates we have $Y_0 \rightarrow \hat{Y}^A$ and~(\ref{Jcons})
gives
\begin{equation}
d \left( \phi^*_{cl} (J^A) \right) = 0
\quad, \quad \quad \quad
J^A = i_{\hat{Y}^A} \tilde{b}
\quad.
\label{JconsS}
\end{equation}
With $\tilde{b}$ as in~(\ref{bdef}) and $\hat{Y}^A={(B^{-1})^A}_C
Y^C$ (see~(\ref{ZPihat})),
the second of~(\ref{JconsS}) gives for $J^A$
\begin{equation}
J^A = {(B^{-1})^A}_D i_{Y^D} \Pi_C \Lambda^C
= {(B^{-1})^A}_C \Lambda^C
\quad,
\label{JAS}
\end{equation}
since $i_{Y^D} \Pi_C= {\delta^D}_C$. Notice that $J^A$ is, in
this case, a form on $\Sigma$ (rather than $\tilde{\Sigma}$).
Effecting explicitly the pullback in the first of~(\ref{JconsS}) we
find\footnote{Eqn. (\ref{jAi}) for $j^{Ai}$ also follows from
the standard expression for the current associated with an
`internal' symmetry of a Lagrangian ${\cal L}$,
$j^{Ai}=\delta^{A}\varphi(\xi) {\partial{\cal L} \over \partial_i\varphi
(\xi)}$. However, for the currents considered here the relevant
part of ${\cal L}$ is just ${\cal L}_{WZ}$. Since
${\cal L}_{WZ}$ is obtained from a form on ${\tilde \Sigma}$,
the above derivation allows us to exploit the geometry
of ${\tilde\Sigma}$. The above expression for the current also
makes clear that, within the canonical formalism, the integrated
charge operators will reproduce the original symmetry algebra.}
\begin{equation}
\partial_i j^{Ai}=0
\quad,
\quad \quad \quad
j^{Ai}(\xi) \equiv {(B^{-1})^A}_C(\xi) \, \Lambda^{Ci}(\xi)
\quad.
\label{jAi}
\end{equation}
Finally, the conserved charges
$Q^A$ are given by (expanding multi-indices)
\begin{equation}
Q^{\mu_1 \dots \mu_{p-k} \alpha_1 \dots \alpha_k} =
\int_{W_\tau} d \sigma^1 \dots d \sigma^p\sum^p_{m=0}
(B^{-1})^{\mu_1 \dots \mu_{p-k} \alpha_1 \dots \alpha_k}_
{\, \, \, \nu_1 \dots \nu_{p-m} \beta_1 \dots \beta_m}
b_m \epsilon^{0j_1 \dots j_p} \Pi^{\nu_1}_{j_1} \dots
\Pi^{\nu_{p-m}}_{j_{p-m}}
\Pi^{\beta_1}_{j_{p-m+1}} \dots
\Pi^{\beta_m}_{j_p}
\quad,
\label{jAexpr}
\end{equation}
where $W_\tau$ is a hypersurface of constant $\tau$. Notice that
since $B=B(\theta,\, x)$, the integrand above involves only
superspace variables.
\section{Applications: $p$=1,2}
\label{p12}
\subsection{$D$=10, N=1 and the Green--Schwarz superstring}
\label{css}
The case of the superstring is somewhat special, from the
point of view of the extension algorithm of Sec.~\ref{NCESA}:
the first additional generator to be introduced,
$Z^\mu$, is a vector, as $X_\mu$.
We shall keep it here separate and denote by $\varphi_\mu$
the associated parameter. Fixing
$(a_s,a_0,a_1)=({1 \over 2},{1 \over 2},1)$
in~(\ref{dPimu}),(\ref{dPim1a1}),
we find for the FDA\footnote{In Sec. \ref{css} all
spinors are Majorana-Weyl ($\theta^\alpha\equiv
(\mathcal{P}_+\theta)^\alpha$, \em{etc.}).}
\begin{equation}
\begin{array}{rclcrcl}
d\Pi^\alpha & = & 0
\, ,
& \qquad &
d \Pi^\mu & = & {1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \Pi^\alpha
\Pi^\beta
\, ,
\\
d \Pi_\alpha & = & (C \Gamma_\mu)_{\alpha\beta}
\Pi^\mu \Pi^\beta
+ (C \Gamma^\mu)_{\alpha\beta}\Pi_\mu \Pi^\beta
\, ,
& \qquad &
d \Pi_\mu^{(\varphi)} & = & {1 \over 2} (C \Gamma_\mu)_{\alpha\beta}
\Pi^\alpha \Pi^\beta
\, ,
\label{ssFDA}
\end{array}
\end{equation}
where $\mu=0, \dots , 9$.
Notice that $d(d\Pi_\alpha)=0$ is implied by
(\ref{Gammaprop}) for $p=1$, $(C \Gamma^\mu)_{\alpha'\beta'}
(C \Gamma_\mu)_{\gamma'\delta'}=0$.
$\Pi_\mu^{(\varphi)}$ in the above equation is the one obtained
from the second of
(\ref{dPimu}) for $p=1$ -- we have added a superscript to
avoid any confusion with $\Pi^\mu$, since they have similar
differentials; recall also that $\Pi_\alpha$ and $\Pi^\alpha$ are
unrelated.
As mentioned in Sec.~\ref{NCESA}, the two terms in the r.h.s.
of the last of~(\ref{ssFDA}) are individually closed and hence
their relative normalization cannot be fixed by requiring
$d(d\Pi_\alpha)=0$. We have nevertheless chosen the above
symmetric normalization
for convenience -- the results that follow, and in
particular~(\ref{ssJAS}) that involves cancellations, do not
depend essentially on this choice.
The corresponding Lie algebra is given by
\begin{equation}
\{ D_\alpha, D_\beta\} = (C \Gamma^\mu)_{\alpha\beta} X_\mu + (C \Gamma_\mu)_{\alpha\beta} Z^\mu
\, , \qquad
\left[D_\alpha, X_\mu\right] = (C \Gamma_\mu)_{\alpha\beta} Z^\beta
\, , \qquad
\left[D_\alpha, Z^\mu\right] = (C \Gamma^\mu)_{\alpha\beta} Z^\beta
\, , \qquad
\label{ssLa}
\end{equation}
which reduces to the Green algebra~\cite{Gre:89} if one omits
$Z^\mu$.
The associated group manifold (extended superspace)
$\tilde{\Sigma}$ is parametrized by
$(\theta^\alpha,x^\mu,\varphi_\mu,\varphi_\alpha)$ via
\begin{equation}
g(\theta^\alpha,x^\mu,\varphi_\mu,\varphi_\alpha)
= e^{\theta^\alpha D_\alpha + x^\mu X_\mu +
\varphi_\mu Z^\mu + \varphi_\alpha Z^\alpha}
\quad.
\label{ssge}
\end{equation}
Making use of the Baker-Cambell-Hausdorff (BCH) formula, where,
for the algebra~(\ref{ssLa}),
terms of order four and higher vanish,
we find the $\tilde{\Sigma}$ group law
\begin{equation}
\addtolength{\arraycolsep}{-.6ex}
\begin{array}{rclcrcl}
\displaystyle
\theta''^\alpha &=& \displaystyle \theta'^\alpha + \theta^\alpha
\, ,
& \qquad &
\displaystyle
x''^\mu & = & \displaystyle x'^\mu + x^\mu - {1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \theta'^\alpha
\theta^\beta
\\[1.5ex]
\displaystyle
\varphi''_\alpha &=& \displaystyle \varphi'_\alpha +\varphi_\alpha
+{1 \over 2} (C \Gamma_\mu)_{\alpha\beta} \theta'^\beta x^\mu
- {1 \over 2} (C \Gamma_\mu)_{\alpha\beta} x'^\mu \theta^\beta
& \qquad &
\displaystyle
\varphi''_\mu &=& \displaystyle \varphi'_\mu + \varphi_\mu - {1 \over 2}
(C \Gamma_\mu)_{\alpha\beta} \theta'^\alpha \theta^\beta
\, .
\\[1.5ex]
& & \displaystyle \mbox{}+{1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \theta'^\beta \varphi_\mu
-{1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \varphi'_\mu \theta^\beta
& \qquad &
& &
\\[1.5ex]
& & \displaystyle \mbox{}+{1 \over 6} (C \Gamma_\mu)_{\alpha\beta} (C \Gamma^\mu)_{\gamma\delta}
(\theta'^\gamma \theta'^\beta \theta^\delta
+\theta'^\delta \theta^\gamma \theta^\beta)
\, ,
& \qquad &
& &
\label{ssgcl}
\end{array}
\end{equation}
The bilinear terms in the expression for
$\varphi''_\alpha$ are the ones that give rise to the fourth of the MC
equations~(\ref{ssFDA})---the trilinear terms are required by
the associativity of the group law. Their sum gives
the spinor valued two-cocycle $\xi_\alpha$ associated with the
central extension of ${\tilde \Sigma} (\theta^\alpha,
x^\mu, \varphi_\mu)$ by $\varphi_\alpha$.
One can now relate the LI one-forms
to the coordinate differentials
\begin{equation}
\addtolength{\arraycolsep}{-.5ex}
\begin{array}{rclcrcl}
\displaystyle
\Pi^\alpha &=& \displaystyle d \theta^\alpha
\, ,
& \quad &
\displaystyle
\Pi^\mu & = & \displaystyle d x^\mu +{1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \theta^\alpha d
\theta^\beta
\\[1.5ex]
\displaystyle
\Pi_\alpha &=& \displaystyle d \varphi_\alpha
-{1 \over 2} (C \Gamma_\mu)_{\alpha\beta} \theta^\beta
dx^\mu
-{1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \theta^\beta d\varphi_\mu
& \quad &
\displaystyle
\Pi_\mu^{(\varphi)} &=& \displaystyle d \varphi_\mu +{1 \over 2} (C \Gamma_\mu)_{\alpha\beta}
\theta^\alpha
\, .
\\[1.5ex]
& &
\displaystyle
\mbox{} +{1 \over 2} (C \Gamma_\mu)_{\alpha\beta} x^\mu d \theta^\beta
+{1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \varphi_\mu d \theta^\beta
& \quad &
& &
\\[1.5ex]
& &
\displaystyle
\mbox{} +{1 \over 3} (C \Gamma_\mu)_{\alpha\beta} (C \Gamma^\mu)_{\gamma\delta} \theta^\gamma
\, , & \quad &
& &
\label{ssLIf}
\end{array}
\end{equation}
(see also~\cite{Der.Gal:97} although, omitting
$\Pi_\mu^{(\varphi)}$, we disagree with the corresponding
expressions
there). One may check that the LI forms in (\ref{ssLIf})
satisfy the FDA (\ref{ssFDA}). From the group composition law, one
can compute the LI vector fields dual to (\ref{ssLIf})
satisfying (\ref{ssLa})
\begin{eqnarray}
D_\alpha &=& \partial_\alpha
+ {1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \theta^\beta
\partial_\mu
+{1 \over 2} (C \Gamma_\mu)_{\alpha\beta} \theta^\beta \partial^\mu
\nonumber \\
& &
-{1 \over 2} (C \Gamma_\mu)_{\alpha\beta} x^\mu \partial^\beta
-{1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \varphi_\mu \partial^\beta
+{1 \over 6} (C \Gamma^\mu)_{\alpha\beta} (C \Gamma_\mu)_{\gamma\delta} \theta^\beta \theta^\delta
\partial^\gamma
\nonumber \\
X_\mu &=& \partial_\mu + {1 \over 2} (C \Gamma_\mu)_{\alpha\beta} \theta^\beta
\partial^\alpha
\nonumber \\
Z^\mu &=& \partial^\mu + {1 \over 2} (C \Gamma^\mu)_{\alpha\beta} \theta^\beta
\partial^\alpha
\nonumber \\
Z^\alpha &=& \partial^\alpha\qquad (\text{note \ } \partial^\mu \equiv {\partial
\over\partial\varphi_\mu}\ , \, \, \, \partial^\alpha \equiv {\partial \over
\partial\varphi_\alpha} )\quad .
\label{ssLIvf}
\end{eqnarray}
Effecting a right group
translation in~(\ref{ssLIf}) and reexpressing the result in terms
of the $\Pi$'s, or computing the exponential in~(\ref{Texpr}),
one finds for $(T^t)^{-1}$
{\small
\begin{equation}
(T^t)^{-1} = \left( \begin{array}{cc|cc}
{\delta_\gamma}^\alpha
& - (C \Gamma^\mu)_{\gamma \delta} \theta^\delta
& - (C \Gamma_\kappa)_{\gamma \delta} \theta^\delta \, \,
& \, \, (C \Gamma_\lambda)_{\gamma \beta} x^\lambda
+ (C \Gamma^\nu)_{\gamma \beta} \varphi_\nu
+ (C \Gamma_\lambda)_{\beta \delta} (C \Gamma^\lambda)_{\gamma
\epsilon} \theta^\epsilon \theta^\delta
\\
0
& {\delta_\nu}^\mu
& 0
& - (C \Gamma_\nu)_{\beta \delta} \theta^\delta
\\
\hline
0 & 0
& {\delta^\rho}_\kappa
& - (C \Gamma^\rho)_{\beta \delta} \theta^\delta
\\
0 & 0 & 0
& {\delta^\zeta}_\beta
\end{array}
\right)
\label{ssTti}
\end{equation}}
\noindent \mbox{}(the matrix indices are $\,_{\gamma}
\,_{\nu} \,^{\rho} \,^{\zeta}$ for the rows and $\,^{\alpha}
\,^{\mu} \,_{\kappa} \,_{\beta}$ for the columns).
When eqns.~(\ref{ssgcl}) are linearized in the primed variables
(viewed as the parameters of the transformation),
they provide a realization of the algebra~(\ref{ssLa}),
acting on the coordinate (unprimed) variables.
For our particular choice of parametrization~(\ref{ssge}), this
action is rendered non-linear by the
last term in the expression for $\varphi''_\alpha$.
This term can be eliminated by modifying the two-cocycle by a
two-coboundary, the latter being generated by a suitable
spinor-valued function
$\eta_\alpha$ on ${\tilde\Sigma}(\theta^\alpha,x^\mu,\varphi_\mu)$.
Indeed,
with $\eta_\alpha \equiv {1 \over 6} (C \Gamma_\mu)_{\alpha\beta} x^\mu \theta^\beta
+{1 \over 6} (C \Gamma^\mu)_{\alpha\beta} \varphi_\mu \theta^\beta$ we find
for the coboundary $\xi^{cob}_\alpha (g',g) \equiv \eta_\alpha(g'g)
-\eta_\alpha(g') -\eta_\alpha(g)$, the expression
\begin{eqnarray}
\xi^{cob}_\alpha &=&
{1 \over 6} (C \Gamma_\mu)_{\alpha\beta} (x'^\mu \theta^\beta + \theta'^\beta x^\mu)
+ {1 \over 6} (C \Gamma^\mu)_{\alpha\beta} (\varphi'_\mu \theta^\beta +\theta'^\beta
\varphi_\mu)
\nonumber \\
& & + {1 \over 6} (C \Gamma_\mu)_{\alpha\beta} (C \Gamma^\mu)_{\gamma\delta} (\theta'^\gamma
\theta'^\beta \theta^\delta - \theta'^\delta
\theta^\gamma \theta^\beta)
\quad.
\label{xicob}
\end{eqnarray}
The new cocycle $\bar{\xi}_{\alpha}= \xi_{\alpha}+\xi^{cob}_{\alpha}$
modifies the last eqn. in~(\ref{ssgcl}) to read
\begin{eqnarray}
\bar{\varphi}''_\alpha &=&
\bar{\varphi}'_\alpha + \bar{\varphi}_\alpha
+ (C \Gamma_\mu)_{\alpha\beta} ({2 \over 3} \theta'^\beta x^\mu
- {1 \over 3} x'^\mu \theta^\beta)
+ (C \Gamma^\mu)_{\alpha\beta} ({2 \over 3} \theta'^\beta \varphi_\mu
- {1 \over 3} \varphi'_\mu \theta^\beta)
\nonumber \\
& & + {1 \over 3} (C \Gamma_\mu)_{\alpha\beta} (C \Gamma^\mu)_{\gamma\delta} \theta'^\gamma
\theta'^\beta \theta^\delta
\quad,
\label{bvpgcl}
\end{eqnarray}
which is linear in the unprimed variables
although the definite symmetry
properties under exchange of primed and unprimed variables are
now lost. The terms in~(\ref{bvpgcl}) linear in the primed
coordinates agree (omitting $\varphi_\mu$) with the (first-order)
results
of~\cite{Der.Gal:97}, where the (equivalent) coordinate
redefinition $\varphi_\alpha \rightarrow \bar{\varphi}_\alpha
\equiv\varphi_\alpha + \eta_\alpha$ is given. For the LI one-form
associated with the new coordinate we find
\begin{eqnarray}
\Pi_{(\bar{\varphi}_\alpha)} &=& d \bar{\varphi}_\alpha
- {2 \over 3} (C \Gamma_\mu)_{\alpha\beta} \theta^\beta d x^\mu
- {2 \over 3} (C \Gamma^\mu)_{\alpha\beta} \theta^\beta d \varphi_\mu
\nonumber \\
& & + {1 \over 3} \left( (C \Gamma_\mu)_{\alpha\beta} x^\mu
+ (C \Gamma^\mu)_{\alpha\beta} \varphi_\mu
+(C \Gamma_\mu)_{\alpha \delta} (C \Gamma^\mu)_{\beta \gamma}
\theta^\gamma \theta^\delta \right) d \theta^\beta
\quad.
\label{newPia}
\end{eqnarray}
The manifestly invariant WZ term for the
superstring action is given by
\begin{equation}
S_{WZ} =\int_W \phi^{*}(\tilde{b}) =
\int_W \phi^{*}(\Pi_\mu^{(\varphi)}
\Pi^\mu + {1 \over 2} \Pi_\alpha \Pi^\alpha) \quad,
\label{ssWZt}
\end{equation}
which differs from the one in~\cite{Sie:94} by the term
in $\Pi_\mu^{(\varphi)} \Pi^\mu$. It is immediately checked,
using~(\ref{ssFDA}), (\ref{ssLIf}), that $d
\tilde{b}=db= (C \Gamma_\mu)_{\alpha\beta} \Pi^\mu \Pi^\alpha \Pi^\beta$ and hence that
$\phi^*(\tilde{b})$ and the standard WZ term $\phi^*(b)$ are
equivalent, differing only by an exact differential.
Since the string tension $T$ has dimensions [$T$]=$ML^{-1}$ and
[$\Pi_\mu^{(\varphi)}$]=$L$, [$\Pi_\alpha$]=$L^{3/2}$, the
products $T \Pi_\alpha \Pi^\alpha$,
$T\Pi_\mu^{(\varphi)}\Pi^\mu$, have the dimensions $ML$ of an action.
To compute the conserved Noether currents, we start
from~(\ref{JAS}) which gives the closed forms $J^A$ on
$\tilde{\Sigma}$. With
$B^{-1}$ given by the lower right block of $(T^t)^{-1}$
(see~(\ref{ssTti}), (\ref{Tblock})) we get
\begin{equation}
\addtolength{\arraycolsep}{-.6ex}
\begin{array}{rcl}
J^\mu &
=
& {(B^{-1})^\mu}_\nu \Pi^\nu
+ {1 \over 2} {(B^{-1})^\mu}_\beta \Pi^\beta
\, = \, \Pi^\mu - {1 \over 2} (C \Gamma^\mu)_{\beta \gamma}
\theta^\gamma d \theta^\beta \, = \, dx^\mu
\nonumber \\[1.5ex]
J^\alpha &=& {1 \over 2} {(B^{-1})^\alpha}_\beta \Pi^\beta
\, = \, {1 \over 2} d \theta^\alpha
\, ,
\end{array}
\label{ssJAS}
\end{equation}
which, after pulling back on the worldvolume, gives for
the charge $Q^\mu$ (see~(\ref{jAexpr}))
\begin{equation}
Q^\mu = \int^{2\pi}_0 d \sigma \, {\partial x^\mu \over \partial \sigma}
\quad,
\label{sscc}
\end{equation}
$Q^\alpha$ being zero because we assume that $\theta$ is periodic
in $\sigma$. The integral~(\ref{sscc}) may lead to a non-zero
result if the topology is nontrivial~\cite{Azc.Gau.Izq.Tow:89}.
\subsection{$D=11$ and the case of the supermembrane}
\label{csm}
Our starting point is the FDA of Sec.~\ref{NCESA} with $p=2$.
We fix the normalisation of the forms by setting
$(a_s,a_0,a_1,a_2)=({1 \over 2},{1 \over 2},1,-{1 \over 2})$
so that the dual Lie algebra becomes
\begin{eqnarray}
\{ D_\alpha,D_\beta\} &=& (C \Gamma^\mu)_{\alpha\beta} X_\mu +
(C \Gamma_{\mu\nu})_{\alpha\beta}
Z^{\mu\nu}
\nonumber \\
\left[ X_\mu,D_\alpha \right] &=& -(C \Gamma_{\mu\nu})_{\alpha\beta} Z^{\nu\beta}
\nonumber \\
\left[ X_\mu,X_\nu \right] &=& (C \Gamma_{\mu\nu})_{\alpha\beta} Z^{\alpha\beta}
\nonumber \\
\left[ X_\mu, Z^{\lambda\tau} \right] &=& {1 \over 2}
\delta^{\left[ \lambda \right.}_\mu
(C \Gamma^{\left. \tau \right]})_{\alpha\beta} Z^{\alpha\beta}
\nonumber \\
\left[ D_\alpha,Z^{\mu\nu} \right] &=& (C \Gamma^{\left[
\mu \right.})_{\alpha\beta} Z^{\left.\nu \right] \beta}
\nonumber \\
\{ D_\alpha, Z^{\nu\beta} \} &=&
({1 \over 4}(C \Gamma^\nu)_{\gamma\delta}
\delta^\beta_\alpha +2(C \Gamma^\nu)_{\gamma\alpha}
\delta^\beta_\delta) Z^{\gamma\delta}
\quad,
\label{smLa}
\end{eqnarray}
coinciding with that given in~\cite{Ber.Sez:95}.
The associated extended superspace group
manifold is parametrized by the coordinates
$(\theta^\alpha,x^\mu,\varphi_{\mu\nu},\varphi_{\mu\alpha},\varphi_{\alpha\beta})$ via\footnote{We use a
parametrization different from the one in~\cite{Ber.Sez:95}.}
\begin{equation}
g = e^{\theta^\alpha D_\alpha + x^\mu X_\mu + \varphi_{\mu\nu} Z^{\mu\nu} + \varphi_{\mu\alpha} Z^{\mu\alpha} + \varphi_{\alpha\beta} Z^{\alpha\beta}}
\quad,
\label{smge}
\end{equation}
where $\varphi_{\mu\nu}$, $\varphi_{\alpha\beta}$ are antisymmetric and symmetric,
respectively, in their indices.
Application (with the help of FORM) of the BCH formula,
where now terms of order five and higher
vanish, results in the following group law
\begin{eqnarray}
{\theta^{\alpha}}^{''} & = & {\theta^{\alpha}}^{'}
+ {\theta^{\alpha}}^{}
\\
{x^\mu}'' &=& {x^{\mu}}^{'} + {x^{\mu}}^{}
- 1/2\, \, {\theta^{\alpha_{2}}}^{'}\,
{\theta^{\alpha_{3}}}^{}\,
(C \Gamma^{\mu})_{\alpha_{2} \alpha_{3}}
\\
\varphi''_{\mu_1 \mu_2}
&=& \varphi^{'}_{\mu_{1} \mu_{2}}+
\varphi^{}_{\mu_{1} \mu_{2}}
- 1/2\, \, {\theta^{\alpha_{3}}}^{'}\,
{\theta^{\alpha_{4}}}\,
(C \Gamma_{\mu_{1} \mu_{2}})_{\alpha_{3} \alpha_{4}}
\\
\varphi''_{\mu_{1} \alpha_{2}}
&=& \varphi^{'}_{\mu_1 \alpha_2}- 1/2\, \,
{\theta^{\alpha_{3}}}^{'}\,
(C \Gamma^{\mu_{4}})_{\alpha_{3} \alpha_{2}}\,
\varphi_{\mu_{1} \mu_{4}}
\nonumber \\
& & - 1/2\, \, {\theta^{\alpha_{3}}}^{'}\,
(C \Gamma_{\mu_{1}
\mu_{4}})_{\alpha_{3} \alpha_{2}}\, {x^{\mu_{4}}}
\nonumber \\
& & + 1/12\, \, {\theta^{\alpha_{3}}}^{'}\,
{\theta^{\alpha_{4}}}^{'}\, {\theta^{\alpha_{5}}}\,
(C \Gamma^{\mu_{6}})_{\alpha_{3} \alpha_{2}}\,
(C \Gamma_{\mu_{1} \mu_{6}})_{\alpha_{4} \alpha_{5}}
\nonumber \\
& & + 1/12\, \, {\theta^{\alpha_{3}}}^{'}\,
{\theta^{\alpha_{4}}}^{'}\, {\theta^{\alpha_{5}}}\,
(C \Gamma^{\mu_{6}})_{\alpha_{4} \alpha_{5}}\,
(C \Gamma_{\mu_{1} \mu_{6}})_{\alpha_{3} \alpha_{2}}
\pm ((1) \leftrightarrow (2))
\\
\varphi''_{\alpha_{1} \alpha_{2}}
&=& \varphi^{'}_{\alpha_{1} \alpha_{2}}
- 1/4\, \, (C \Gamma^{\mu_{3}})_{\alpha_{1} \alpha_{2}}\, \,
{x^{\mu_{4}}}^{'}\, \, \varphi_{\mu_{3} \mu_{4}}
\nonumber \\
& & + 1/2\, \, (C \Gamma_{\mu_{3} \mu_{4}})_{\alpha_{1}
\alpha_{2}}\, \, {x^{\mu_{3}}}^{'}\, \, {x^{\mu_{4}}}
\nonumber \\
& & + 1/6\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}^{'}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{3}
\alpha_{1}}\, \,
(C \Gamma^{\mu_{6}})_{\alpha_{4} \alpha_{2}}\, \,
\varphi_{\mu_{5}
\mu_{6}}
\nonumber \\
& & + 1/6\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}^{'}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{3}
\alpha_{1}}\, \,
(C \Gamma_{\mu_{5} \mu_{6}})_{\alpha_{4} \alpha_{2}}\, \,
{x^{\mu_{6}}}
\nonumber \\
& & - 1/24\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}^{'}\, \, {\theta^{\alpha_{5}}}\, \,
{\theta^{\alpha_{6}}}\, \, (C \Gamma^{\mu_{7}})_{\alpha_{3}
\alpha_{1}}\, \,
(C \Gamma^{\mu_{8}})_{\alpha_{4} \alpha_{6}}\, \,
(C \Gamma_{\mu_{7} \mu_{8}})_{\alpha_{5} \alpha_{2}}
\nonumber \\
& & - 1/24\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}^{'}\, \, {\theta^{\alpha_{5}}}\, \,
{\theta^{\alpha_{6}}}\, \, (C \Gamma^{\mu_{7}})_{\alpha_{3}
\alpha_{1}}\, \,
(C \Gamma^{\mu_{8}})_{\alpha_{5} \alpha_{2}}\, \,
(C \Gamma_{\mu_{7} \mu_{8}})_{\alpha_{4} \alpha_{6}}
\nonumber \\
& & - 1/24\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{1}
\alpha_{2}}\, \,
(C \Gamma^{\mu_{6}})_{\alpha_{3} \alpha_{4}}\, \,
\varphi^{'}_{\mu_{5} \mu_{6}}
\nonumber \\
& & + 1/48\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{1}
\alpha_{2}}\, \,
(C \Gamma_{\mu_{5} \mu_{6}})_{\alpha_{3} \alpha_{4}}\, \,
{x^{\mu_{6}}}^{'}
\nonumber \\
& & - 1/6\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{3}
\alpha_{1}}\, \,
(C \Gamma^{\mu_{6}})_{\alpha_{4} \alpha_{2}}\, \,
\varphi^{'}_{\mu_{5} \mu_{6}}
\nonumber \\
& & - 1/6\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{3}
\alpha_{1}}\, \,
(C \Gamma_{\mu_{5} \mu_{6}})_{\alpha_{4} \alpha_{2}}\, \,
{x^{\mu_{6}}}^{'}
\nonumber \\
& & + 1/48\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{3}
\alpha_{4}}\, \,
(C \Gamma^{\mu_{6}})_{\alpha_{1} \alpha_{2}}\, \,
\varphi^{'}_{\mu_{5} \mu_{6}}
\nonumber \\
& & + 1/12\, \, {\theta^{\alpha_{3}}}^{'}\, \,
{\theta^{\alpha_{4}}}\, \, (C \Gamma^{\mu_{5}})_{\alpha_{3}
\alpha_{4}}\, \,
(C \Gamma_{\mu_{5} \mu_{6}})_{\alpha_{1} \alpha_{2}}\, \,
{x^{\mu_{6}}}^{'}
\nonumber \\
& & - {\theta^{\alpha_{3}}}^{'}\, \, \varphi_{\mu_{4}
\alpha_{1}}\, \, (C \Gamma^{\mu_{4}})_{\alpha_{3} \alpha_{2}}
\nonumber \\
& & - 1/8\, \, {\theta^{\alpha_{3}}}^{'}\, \, \varphi_{\mu_{4}
\alpha_{3}}\, \, (C \Gamma^{\mu_{4}})_{\alpha_{1} \alpha_{2}}
\pm ((1) \leftrightarrow (2))
\quad
\label{smgcl}
\end{eqnarray}
(in the expression for $\varphi_{\alpha_1 \alpha_2}$, the r.h.s.
is assumed to be
symmetrised, with unit weight, in $\alpha_1$, $\alpha_2$).
The $\pm ((1) \leftrightarrow (2))$ instruction in the last two
expressions
means that for each term displayed, one has to add (subtract) a
similar term, with the primed and unprimed variables
exchanged (taking into account
statistics), if the order of the term is odd (even) in the
coordinates (\emph{e.g.} \ $\theta'^\alpha \varphi_{\mu \beta} \pm ((1)
\leftrightarrow (2)) = \theta'^\alpha \varphi_{\mu \beta} - (-
\varphi'_{\mu \beta} \theta^\alpha)$). This symmetry property
can be seen to hold in general:
from the BCH formula $e^{A'} e^A=e^{f(A',A)}$
it follows that $f(-A,-A')$=$-f(A',A)$ and hence, terms of
order $n$ in the coordinates are symmetric ($n$ odd) or
antisymmetric ($n$ even) under the above exchange of the two
spaces. The linearized (in the primed variables) form of the
previous expressions has been given in~\cite{Der.Gal:97}.
Starting from~(\ref{smgcl}), we find for
the LI vector fields of~(\ref{smLa})
\begin{eqnarray}
D_{\alpha_1} &=&
\partial_{\alpha_{1}}
+ 1/12\, \, {\theta^{\alpha_{2}}}^{}\, \,
{\theta^{\alpha_{3}}}^{}\, \, \left((C \Gamma^{\mu_{4}})_{\alpha_{2}
\alpha_{5}}\, \, (C \Gamma_{\mu_{6} \mu_{4}})_{\alpha_{3}
\alpha_{1}} + (C \Gamma_{\mu_{4}\mu_{6}})_{\alpha_{2}
\alpha_{5}}\, \, (C\Gamma^{\mu_{4}})_{\alpha_{3} \alpha_{1}}
\right)\, \, \partial^{\mu_{6} \alpha_{5}}
\nonumber \\
& & + 1/12\, \, {\theta^{\alpha_{2}}}^{}\, \,
\varphi^{}_{\mu_{3} \mu_{4}}\, \, \left((C
\Gamma^{\mu_{3}})_{\alpha_{2} \alpha_{5}}\, \, (C
\Gamma^{\mu_{4}})_{\alpha_{1} \alpha_{6}}+
(C\Gamma^{\mu_{3}})_{\alpha_{5} \alpha_{6}}\, \,
(C\Gamma^{\mu_{4}})_{\alpha_{2} \alpha_{1}} \right)\, \,
\partial^{\alpha_{5} \alpha_{6}}
\nonumber \\
& & - 1/3\, \, {\theta^{\alpha_{2}}}^{}\, \,
{x^{\mu_{3}}}^{}\, \, (C \Gamma^{\mu_{4}})_{\alpha_{2}
\alpha_{5}}\, \, (C \Gamma_{\mu_{3} \mu_{4}})_{\alpha_{6}
\alpha_{1}}\, \, \partial^{\alpha_{5} \alpha_{6}}
\nonumber \\ & & + 1/8\, \, {\theta^{\alpha_{2}}}^{}\, \,
{x^{\mu_{3}}}^{}\, \, (C \Gamma^{\mu_{4}})_{\alpha_{5}
\alpha_{6}}\, \, (C \Gamma_{\mu_{3} \mu_{4}})_{\alpha_{2}
\alpha_{1}}\, \, \partial^{\alpha_{5} \alpha_{6}}
\nonumber \\
& & + 1/2\, \, {\theta^{\alpha_{2}}}^{}\, \, (C
\Gamma^{\mu_{3}})_{\alpha_{2} \alpha_{1}}\, \, \partial_{\mu_{3}}
\nonumber \\
& & + {\theta^{\alpha_{2}}}^{}\, \,
(C \Gamma_{\mu_{3} \mu_{4}})_{\alpha_{2} \alpha_{1}}\, \,
\partial^{\mu_{3} \mu_{4}}
\nonumber \\
& & - 1/2\, \, {x^{\mu_{2}}}^{}\, \,
(C \Gamma_{\mu_{2} \mu_{3}})_{\alpha_{1} \alpha_{4}}\, \,
\partial^{\mu_{3} \alpha_{4}}
\nonumber \\ & & - 1/2\, \, \varphi^{}_{\mu_{2} \mu_{3}}\, \, (C
\Gamma^{\mu_{2}})_{\alpha_{1} \alpha_{4}}\, \, \partial^{\mu_{3} \alpha_{4}}
\nonumber \\ & & + 1/8\, \, \varphi^{}_{\mu_{2} \alpha_{1}}\, \, (C
\Gamma^{\mu_{2}})_{\alpha_{3} \alpha_{4}}\, \, \partial^{\alpha_{3} \alpha_{4}}
\nonumber \\ & & + \varphi^{}_{\mu_{2} \alpha_{3}}\, \, (C
\Gamma^{\mu_{2}})_{\alpha_{1} \alpha_{4}}\, \, \partial^{\alpha_{4} \alpha_{3}}
\\
X_{\mu_1} &=&
\partial_{\mu_{1}}
- 1/6\, \, {\theta^{\alpha_{2}}}^{}\, \,
{\theta^{\alpha_{3}}}^{}\, \, (C
\Gamma^{\mu_{4}})_{\alpha_{2} \alpha_{5}}\, \, (C \Gamma_{\mu_{1}
\mu_{4}})_{\alpha_{6} \alpha_{3}}\, \, \partial^{\alpha_{5} \alpha_{6}}
\nonumber \\ & & + 1/2\, \, {\theta^{\alpha_{2}}}^{}\, \, (C
\Gamma_{\mu_{1} \mu_{3}})_{\alpha_{2} \alpha_{4}}\, \, \partial^{\mu_{3} \alpha_{4}}
\nonumber \\ & & + 1/2\, \, {x^{\mu_{2}}}^{}\, \, (C \Gamma_{\mu_{2}
\mu_{1}})_{\alpha_{3} \alpha_{4}}\, \, \partial^{\alpha_{3} \alpha_{4}}
\nonumber \\ & & - 1/4\, \, \varphi^{}_{\mu_{1} \mu_{2}}\, \, (C
\Gamma^{\mu_{2}})_{\alpha_{3} \alpha_{4}}\, \, \partial^{\alpha_{3} \alpha_{4}}
\\
Z^{\mu_1 \mu_2} &=&
\partial^{\mu_{1} \mu_{2}}
+ 1/12\, \, {\theta^{\alpha_{3}}}^{}\, \,
{\theta^{\alpha_{4}}}^{}\, \, (C \Gamma^{\mu_{1}})_{\alpha_{3}
\alpha_{5}}\, \, (C \Gamma^{\mu_{2}})_{\alpha_{4} \alpha_{6}}\,
\, \partial^{\alpha_{5} \alpha_{6}}
\nonumber \\ & & - 1/12\, \, {\theta^{\alpha_{3}}}^{}\, \,
{\theta^{\alpha_{4}}}^{}\, \, (C \Gamma^{\mu_{2}})_{\alpha_{3}
\alpha_{5}}\, \, (C \Gamma^{\mu_{1}})_{\alpha_{4} \alpha_{6}}\,
\, \partial^{\alpha_{5} \alpha_{6}}
\nonumber \\ & & + 1/4\, \, {\theta^{\alpha_{3}}}^{}\, \, (C
\Gamma^{\mu_{1}})_{\alpha_{3} \alpha_{4}}\, \, \partial^{\mu_{2} \alpha_{4}}
\nonumber \\ & & - 1/4\, \, {\theta^{\alpha_{3}}}^{}\, \, (C
\Gamma^{\mu_{2}})_{\alpha_{3} \alpha_{4}}\, \, \partial^{\mu_{1} \alpha_{4}}
\nonumber \\ & & + 1/4\, \, {x^{\mu_{1}}}^{}\, \, (C
\Gamma^{\mu_{2}})_{\alpha_{3} \alpha_{4}}\, \, \partial^{\alpha_{3} \alpha_{4}}
\\
Z^{\mu_1 \alpha_2} &=&
\partial^{\mu_{1} \alpha_{2}}
+ 1/8\, \, {\theta^{\alpha_{2}}}^{}\, \,
(C \Gamma^{\mu_{1}})_{\alpha_{3} \alpha_{4}}\, \,
\partial^{\alpha_{3} \alpha_{4}}
+ {\theta^{\alpha_{3}}}^{}\, \,
(C \Gamma^{\mu_{1}})_{\alpha_{3} \alpha_{4}}\, \,
\partial^{\alpha_{4} \alpha_{2}}
\\
Z^{\alpha_1 \alpha_2} &=& \partial^{\alpha_{1} \alpha_{2}}
\quad
\label{smLIvf}
\end{eqnarray}
(antisymmetrisation with unit weight in $\mu_1$, $\mu_2$ is
understood in the r.h.s. of the expression for $Z^{\mu_1 \mu_2}$).
The manifestly invariant WZ term on the extended superspace
$\tilde{\Sigma}$ is given by~\cite{Ber.Sez:95}
\begin{equation}
\tilde{b}={2 \over 3} \Pi_{\mu \nu} \Pi^\mu \Pi^\nu
- {3 \over 5} \Pi_{\mu \alpha} \Pi^\mu \Pi^\alpha
- {2 \over 15} \Pi_{\alpha \beta} \Pi^\alpha \Pi^\beta
\label{smbt}
\end{equation}
([${\tilde b}$]=$L^3$, [$Tb$]=$ML$).
It depends on the additional $\varphi$ variables through total
differentials
since $d \tilde{b}=db=h= (C \Gamma_{\mu
\nu})_{\alpha\beta} \Pi^\mu \Pi^\nu \Pi^\alpha \Pi^\beta$.
The computation of the full $T$ matrix for the supermembrane is
rather tedious. For the
Noether currents though (corresponding to the new variables) we
only need $B^{-1}$. Reading off the relevant structure constants
from~(\ref{smLa}), we find
\begin{eqnarray}
\rho^{(f)}_{adj} \, (D_\alpha) &=&
\left( \begin{array}{ccc}
0 & 0 & 0
\\
(C \Gamma^{[\mu_1})_{\alpha \alpha_2} {\delta_{\kappa_2}}^{\nu_1]}
& 0 & 0
\\
0 & {1 \over 4} (C \Gamma^{\kappa_1})_{\beta_2 \gamma_2}
{\delta_\alpha}^{\alpha_1} + 2 (C \Gamma^{\kappa_1})_{\beta'_2
\alpha} {\delta_{\gamma'_2}}^{\alpha_1} & 0
\end{array} \right)
\quad,
\nonumber \\
\rho^{(f)}_{adj} \, (X_\rho) &=&
\left( \begin{array}{ccc}
0 & 0 & 0
\\
0 & 0 & 0
\\
{1 \over 2} {\delta_\rho}^{[\mu_1} (C \Gamma^{\nu_1]})_{\beta_2
\gamma_2} & 0 & 0
\end{array} \right)
\quad.
\label{DXadj}
\end{eqnarray}
Using these in the exponential in~(\ref{Bexpr}) we get for
$B^{-1}$
\begin{equation}
B^{-1}= \left( \begin{array}{ccc}
\delta^{\mu_1 \nu_1}_{\mu_2 \nu_2}
&
- \theta^\alpha (C \Gamma^{[\mu_1})_{\alpha \alpha_2}
{\delta^{\nu_1]}}_{\kappa_2}
&
- {1 \over 2} x^{[\mu_1} (C \Gamma^{\nu_1]})_{\beta_2 \gamma_2}
+ \theta^\alpha \theta^\beta (C \Gamma^{[\mu_1})_{\alpha
\gamma'_2} (C \Gamma^{\nu_1]})_{\beta'_2 \beta}
\\
0 & \delta^{\kappa_1 \alpha_1}_{\kappa_2 \alpha_2}
& -{1 \over 4} \theta^{\alpha_1} (C \Gamma^{\kappa_1})_{\beta_2
\gamma_2} -2 \theta^{\alpha} (C \Gamma^{\kappa_1})_{\beta'_2
\alpha} {\delta^{\alpha_1}}_{\gamma'_2}
\\
0 & 0 & \delta^{\beta_1 \gamma_1}_{\beta_2 \gamma_2}
\end{array} \right)
\quad
\label{smBi}
\end{equation}
(the external indices are $\,^{\mu_1 \nu_1} \,\,^{\kappa_1
\alpha_1} \,\,^{\beta_1 \gamma_1}$ for the rows and $\,_{\mu_2
\nu_2} \,\,_{\kappa_2 \alpha_2} \,\,_{\beta_2 \gamma_2}$ for the
columns).
Substituting now in~(\ref{JAS}) we find for the forms $J^A$ on $\Sigma$
\begin{eqnarray}
J^{\mu_1 \nu_1} &=& {2 \over 3} dx^{\mu_1} dx^{\nu_1}
+ {1 \over 15} \theta^\alpha (C \Gamma^{[\nu_1})_{\alpha \beta}
dx^{\mu_1]} d\theta^\beta
+ {1 \over 15} x^{[\mu_1} (C \Gamma^{\nu_1]})_{\alpha \beta}
d \theta^\alpha d \theta^\beta
\nonumber \\
&=& d \left( {2 \over 3} x^{[\mu_1} dx^{\nu_1]}
+ {1 \over 15} \theta^\alpha x^{[\mu_1}
(C \Gamma^{\nu_1]})_{\alpha \beta} d\theta^\beta \right)
\quad,
\nonumber \\
J^{\kappa_1 \alpha_1} &=& -{3 \over 5} dx^{\kappa_1}
d\theta^{\alpha_1}
- {1 \over 30} (C \Gamma^{\kappa_1})_{\alpha_3 \alpha_2}
\theta^{\alpha_3} d \theta^{\alpha_2} d \theta^{\alpha_1}
+ {1 \over 30} (C \Gamma^{\kappa_1})_{\alpha_3 \alpha_2}
\theta^{\alpha_1} d \theta^{\alpha_3} d \theta^{\alpha_2}
\nonumber \\
&=& d \left({3 \over 5} dx^{\kappa_1} \theta^{\alpha_1}
- {1 \over 30} (C \Gamma^{\kappa_1})_{\alpha_3 \alpha_2}
\theta^{\alpha_3} \theta^{\alpha_1} d
\theta^{\alpha_2} \right)
\quad,
\nonumber \\
J^{\beta_1 \gamma_1} &=& -{2 \over 15} d \theta^{\beta_1}
d \theta^{\gamma_1}
= d (-{2 \over 15}\theta^{\beta_1}
d\theta^{\gamma_1})
\quad.
\label{JAsm}
\end{eqnarray}
The above locally exact expressions result from rather
non-trivial cancellations. The currents
are obtained by pulling back to $W$ the forms (\ref{JAsm}). For
periodic $\theta$'s the charges
$Q^{\kappa_1 \alpha_1}$, $Q^{\beta_1 \gamma_1}$ (but not
$Q^{\mu_1\nu_1}$
for a non-trivial two-cycle \cite{Azc.Gau.Izq.Tow:89})
turn out to be zero. Thus, this case
provides a realization of the algebra (\ref{smLa}) where only the
$Q^{\mu_1\nu_1}$ term is non-zero.
\section{The case of D-branes}
\label{casedbr}
Let us consider first a bosonic background for which all forms of the
R$\otimes$R sector and the dilaton vanish so that
the two-form ${\cal F}\equiv F-B$, where $F$=$dA$ and $B$ is the
NS$\otimes$NS two-form, reduces to $F$; $A$ is the Born-Infeld
one-form on the worldvolume $A(\xi)=A_i(\xi)d\xi^i$.
Then the action of the D$p$-brane reduces to
\begin{equation}
I= \int d^{p+1} \xi \sqrt{-\det(\partial_i x^\mu \partial_j x_\mu
+F_{ij})}
\quad .
\label{borninfeld}
\end{equation}
Let us look for a manifestly supersymmetric generalisation
of this action on a suitable extension of flat superspace
(we shall consider here only $D=10$, IIA D-branes).
For the ordinary $p$-branes the supersymmetrisation is
achieved by substituting first $\Pi^\mu_i$ for $\partial_ix^\mu$
and then by adding a WZ term $b$, $db=h$, with $h$ characterised
\cite{Azc.Tow:89} by being a non-trivial CE cohomology
$(p+2)$-cocycle on superspace $\Sigma$.
It was shown in the previous sections how to make these
WZ terms manifestly invariant. We shall extend this
to the D$p$-branes case
by showing first that the WZ terms
may be characterised and classified by CE-cocycles as well, and
then by finding manifestly supersymmetric potentials ${\tilde b}$
on the superspaces $\tilde \Sigma$ which are obtained by the
techniques of Sec. \ref{NCESA} or by dimensional reduction from
these. We shall restrict ourselves here to the D$2$-brane case,
and hence to its associated $\tilde \Sigma$
parametrised by $(\theta^\alpha,x^\mu,\varphi_\mu,\varphi_\alpha,
\varphi_{\mu\nu},\varphi_{\mu\alpha},\varphi_{\alpha\beta},
\varphi)$.
\subsection{CE-cocycle classification of D-branes}
\label{CEdB}
The new feature in the D$p$-brane case is the field $A_i(\xi)$
directly defined on the worldvolume. The one-form $A$ transforms
under supersymmetry in such a way that $ {\cal F}\equiv dA-B$
is invariant, where $B$ is a two-form on superspace such
that
\begin{equation}
dB=-\Pi^\mu(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}\Pi^\alpha\Pi^\beta
\quad . \label{6.3}
\end{equation}
Let us now consider $A$ as an abstract form. In our approach, the
possible WZ terms will be some non-trivial $(p+2)$-cocycles
of the cohomology of a certain FDA
(here, of IIA type). This FDA is
generated by the supersymmetric invariant $\Pi^\alpha$, $\Pi^\mu$ and $\cal
F$ and defined by the structure relations
\begin{equation}
d\Pi^\alpha = 0
\, , \qquad
d\Pi^\mu = \frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}
\Pi^\alpha\Pi^\beta
\, , \qquad
d{\cal F} = \Pi^\mu(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}
\Pi^\alpha\Pi^\beta
\, .
\label{freedifferential}
\end{equation}
Note that $dd\equiv 0$ because the identity
$(C\Gamma^\mu\Gamma_{11})_{\alpha{'}\beta{'}}
(C\Gamma_\mu)_{\delta{'}\epsilon{'}}=0$ is valid in $D=10$.
The non-trivial $(p+2)$-cocycles are given by
closed ($p+2$)-forms $h$ constructed from $\Pi^\mu$,
$\Pi^\alpha$, ${\cal F}$ that cannot be written as the differential of a
($p+1$)-form constructed from them, and
with the same dimensions as the kinetic
Lagrangian, {\it i.e.} $[h]=L^{p+1}$. This second
requirement is necessary to avoid introducing dimensionful constants in the
action other than the tension, [$T$]=$ML^{-p}$.
Since ${\cal F}$ is a two-form, $h$ can be expanded in powers of
${\cal F}$ as
\begin{equation}
h=\sum^{\left[\frac{p+2}{2}\right]}_{n=0}\frac{1}{n!}h^{(p+2-2n)}(\Pi^\mu,
\Pi^\alpha){\cal F}^n \quad ,
\label{expansionh}
\end{equation}
where $h^{(k)}$ is a form of order $k\equiv p+2-2n$ and, since $D=10$,
$p\leq 8$ excluding the degenerate case $p+1=10$.
Moreover, since $[h^{(p+2-2n)}]=L^{p+1-2n}$, the $k$-forms $h^{(k)}$,
$[h^{(k)}]=L^{k-1}$, must be
\begin{equation}
h^{(k)}=a^{(k)}\Pi^{\mu_1}\dots \Pi^{\mu_{k-2}}(C\Gamma_{\mu_1\dots
\mu_{k-2}}\Gamma)_{\alpha\beta}\Pi^\alpha\Pi^\beta\, ,\qquad k=2,\dots,p+2
\, , \label{generalform}
\end{equation}
where $\Gamma={\bf 1}$ or $\Gamma_{11}$ so that
$(C\Gamma_{\mu_1\dots \mu_{k-2}}\Gamma)_{\alpha\beta}$
is symmetric (\emph{i.e}, $k-2=1,2,5,6,9,10$ for ${\bf 1}$ and
$0,1,4,5,8,9$ for $\Gamma_{11}$). Since
\begin{equation}
dh=\sum_{n=0}^{\left[\frac{p+2}{2}\right]}\frac{1}{n!}dh^{(p+2-2n)}
{\cal F}^n
+\sum_{n=1}^{\left[\frac{p+2}{2}\right]}\frac{1}{(n-1)!}
(-1)^p h^{(p+2-2n)}(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}\Pi^\mu
\Pi^\alpha\Pi^\beta {\cal F}^{n-1}\quad , \label{difh}
\end{equation}
the required
closure of $h$ is equivalent to the following
set of equations
\begin{equation}
\begin{array}{rclcl}
dh^{(p-2[p/2])} &=& 0 \, ,
& \quad &
\text{for}\; n=\left[\frac{p+2}{2}\right]
\\[1.3ex]
dh^{(p+2-2n)}+(-1)^p h^{(p-2n)}\Pi^\mu(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}
\Pi^\alpha\Pi^\beta &=& 0 \, ,
& \quad &
\text{for}\; n=\left[\frac{p}{2}\right],\dots, 0 \, .
\end{array}
\label{recgeneral}
\end{equation}
At this point it is convenient to examine separately the $p$ odd
and $p$ even cases.
\medskip
\noindent {\it a) $p$ even}
\medskip
The first eqn. in~(\ref{recgeneral}) gives
$dh^{(0)}=0$. This means
$h^{(0)}=0$ because $h^{(0)}\neq 0$ would imply by (\ref{expansionh})
having an additional dimensionful constant,
$[h^{(0)}]=L^{-1}$ ([${\cal F}]= L^2$).
For $n=\frac{p}{2}$ the second of~(\ref{recgeneral}) gives $dh^{(2)}=0$. Inserting
$h^{(2)}$ from eq. (\ref{generalform}), we obtain an identity, so
$a^{(2)}$ is arbitrary.\footnote{We note in passing that in the heterotic case, for
which $N=1$ ($\Pi^\alpha$ is MW), $a^{(2)}=0$ because
$(C\Gamma_{11})_{\alpha\beta}\Pi^\alpha
\Pi^\beta=C_{\alpha\beta}\Pi^\alpha\Pi^\beta=0$ since
${(\Gamma_{11})^\alpha}_\beta\Pi^\beta=\Pi^\alpha$.
So the chain of equations that follows does
not appear and there are no non-trivial WZ terms.
This shows, as expected, that there are
no D-branes in the heterotic case.}
The remaining equations (for $n=\frac{p-2}{2}$, etc.)
are equivalent, by factoring
out products of forms $\Pi^\mu$ and $\Pi^\alpha$, to
\begin{eqnarray}
a^{(4)}(C\Gamma^{\mu_2})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1\mu_2})_{\delta{'}
\epsilon{'}}-a^{(2)}(C\Gamma_{11})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0
\nonumber \\
2a^{(6)}(C\Gamma^{\mu_4})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1\dots\mu_4} \Gamma_{11})_{\delta{'}
\epsilon{'}}-a^{(4)}(C\Gamma_{[\mu_1\mu_2})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_3]}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0
\nonumber \\
3a ^{(8)}(C\Gamma^{\mu_6})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1\dots\mu_6})_{\delta{'}
\epsilon{'}}-a^{(6)}(C\Gamma_{[\mu_1\dots \mu_4}
\Gamma_{11})_{\alpha{'} \beta{'}}(C\Gamma_{\mu_5]}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0
\nonumber \\
4a^{(10)}(C\Gamma^{\mu_8})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1\dots\mu_8}\Gamma_{11})_{\delta{'}
\epsilon{'}}-a^{(8)}(C\Gamma_{[\mu_1\dots \mu_6})_{\alpha{'}
\beta{'}}(C\Gamma_{\mu_7]}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0
\, .
\label{equationsc}
\end{eqnarray}
Note that the
number of identities from (\ref{equationsc}) that are necessary
to show that $h$ is closed depends on the values of $p$ since
$2\leq k\leq p+2$, $k$ even. Specifically, for the
D$2$-brane only the first identity is used, for the D$4$-brane the
first two identities are relevant and the first three (four)
identities are required for the existence of the D$6$-(D$8$-)branes.
Equations (\ref{equationsc}) have to be identically satisfied for
certain values of the
$a$'s to be determined. To find them, one may multiply
the equations by $(\Gamma^\nu C^{-1})^{\beta\delta}$ (although the
resulting system is not equivalent to the original one).
This procedure gives some
equalities between gamma matrices that are only satisfied if $a^{(2)}=
-a^{(4)}$, $a^{(4)}=-6a^{(6)}$, $a^{(6)}=-15a^{(8)}$ and
$a^{(8)}=-28a^{(10)}$.
These values are the solution of eqs. (\ref{equationsc}) and determine
closed forms $h$ by (\ref{expansionh}), (\ref{generalform})
provided that the following identities are satisfied
\begin{eqnarray}
(C\Gamma^{\mu_2})_{\alpha{'}\beta{'}}(C\Gamma_{\mu_1\mu_2})_{\delta{'}
\epsilon{'}}+(C\Gamma_{11})_{\alpha{'}\beta{'}}(C\Gamma_{\mu_1}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0
\nonumber \\
(C\Gamma^{\mu_4})_{\alpha{'}\beta{'}}(C\Gamma_{\mu_1\dots\mu_4}
\Gamma_{11})_{\delta{'}
\epsilon{'}}+3(C\Gamma_{[\mu_1\mu_2})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_3]}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0
\nonumber \\
(C\Gamma^{\mu_6})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1\dots\mu_6})_{\delta{'}
\epsilon{'}}+5(C\Gamma_{[\mu_1\dots \mu_4}
\Gamma_{11})_{\alpha{'} \beta{'}}(C\Gamma_{\mu_5]}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0
\nonumber \\
(C\Gamma^{\mu_8})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1\dots\mu_8}\Gamma_{11})_{\delta{'}
\epsilon{'}}+7(C\Gamma_{[\mu_1\dots \mu_6})_{\alpha{'}
\beta{'}}(C\Gamma_{\mu_7]}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0 \, ,
\label{Identities}
\end{eqnarray}
as it is indeed the case (see Appendix B).
Therefore we have shown that, for $p$ even, there exist
closed $(p+2)$-forms $h$ with the required dimensions for all
even values of $p$, $p\leq 8$.
To prove that the $h$'s obtained from eq. (\ref{generalform}) for
the appropriate values of $a^{(k)}$ are not CE-trivial, it is
sufficient to note that if there were a potential form
$b(\Pi^\mu,\Pi^\alpha,{\cal F})$,
$db=h$, then this form would be a Lorentz-invariant ($p+1$)-form with physical
dimensions $L^{p+1}$, which does not exist for $p+1<10$ since $p$ is even.
\medskip
\noindent{\it b) $p$ odd}
\medskip
In this case, the first eqn. in~(\ref{recgeneral}) gives
$dh^{(1)}=0$. Again, this means
$h^{(1)}=0$ because obviously there are no Lorentz-scalar
one-forms that can be constructed from $\Pi^\mu$, $\Pi^\alpha$
and $\cal F$. On the other hand, $k$ in $h^{(k)}$ has
now the range $3\leq k\leq p+2$, $k$ odd. Of these $h^{(k)}$,
those corresponding to $k=5$ and $k=9$ vanish independently of the
type of the matrix $\Gamma$ (see table). This leaves us with $h^{(3)}$
and $h^{(7)}$ and the second of (\ref{recgeneral}) leads to
\begin{equation}
\begin{array}{rclcrcl}
\displaystyle
a^{(3)}(C\Gamma^{\mu_1})_{\alpha{'}\beta{'}}(C\Gamma_{\mu_1}\Gamma)_{\delta{'}
\epsilon{'}} &=& 0 \, ,
& \qquad &
a^{(3)}(C\Gamma_{[\mu_1}\Gamma)_{\alpha{'}\beta{'}}(C\Gamma_{\mu_2]}\Gamma_{11}
)_{\delta{'}\epsilon{'}} &=& 0
\\[1.3ex]
\displaystyle
a^{(7)}(C\Gamma^{\mu_7})_{\alpha{'}\beta{'}}(C\Gamma_{\mu_1\dots \mu_5}\Gamma
)_{\delta{'}\epsilon{'}} &= & 0 \, ,
& \qquad &
a^{(7)}(C\Gamma_{[\mu_1\dots\mu_5}\Gamma)_{\alpha{'}\beta{'}}(C\Gamma_{\mu_6]}
\Gamma_{11})_{\delta{'}\epsilon{'}} &=& 0 \, .
\end{array}
\label{oddcase}
\end{equation}
In the $a^{(3)}$ equations, $\Gamma$ has to be $\Gamma_{11}$ (the other
possibility, ${\bf 1}_{32}$, may be shown to be inconsistent
by multiplying the second expression by
$(\Gamma^\nu C^{-1})^{\beta\delta}$). Multiplying the fourth equation
by $(\Gamma^\nu C^{-1})^{\beta\delta}$ shows that
$a^{(7)}=0$ for both $\Gamma$=${\bf 1}$, $\Gamma_{11}$. Thus we
have shown that the only candidate for a WZ term in the
odd $p$ case is obtained from $h^{(3)}$ \emph{i.e.}, from
\begin{equation}
h=\Pi^\mu(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}\Pi^\alpha\Pi^\beta
{\cal F}^{\frac{p-1}{2}}\quad ,\quad p\geq 1\quad . \label{triviala}
\end{equation}
But then $h=d(\frac{2}{p+1}{\cal F}^{(p+1)/2})$
by (\ref{freedifferential}),
and hence is a trivial CE cocycle. Therefore in the $D=10$,
IIA theory there are no non-trivial WZ terms for the D-branes with
$p$ odd. The other values for $p$ found in our discussion are
precisely those for which D-branes of type IIA are known
to exist. Thus the \emph{IIA D-branes are, as the scalar $p$-branes}
\cite{Azc.Tow:89}, \emph{characterized by non-trivial CE cocycles}.
\bigskip
\subsection{D-branes defined on extended superspace}
\label{Dbext}
As we saw, one reason for considering superspace extensions
associated with extended objects is that it is possible to
find on $\tilde{\Sigma}$ manifestly invariant WZ terms since
then $h$ may be expressed as the differential of a LI form
${\tilde b}$. We shall now show that this is also possible
for the D$2$-brane. The starting point is now the FDA given in
eq. (\ref{2.15}) with the generators with more than two vector
indices absent,
plus the equation for $d{\cal F}$ {\it i.e.}
\begin{equation}
\addtolength{\arraycolsep}{-.5ex}
\begin{array}{rclcrcl}
\displaystyle
d\Pi^\alpha&=&0
& \qquad \quad &
\displaystyle
d\Pi^\mu &=& \frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}\Pi^\alpha\Pi^\beta
\\[1.3ex]
\displaystyle
d\Pi &=& \frac{1}{2}(C\Gamma_{11})_{\alpha\beta}\Pi^\alpha\Pi^\beta
& \qquad \quad &
\displaystyle
d\Pi_{\mu\nu} &=& \frac{1}{2}(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^\alpha
\Pi^\beta
\\[1.3ex]
\displaystyle
d\Pi_\mu^{(z)} &=& \frac{1}{2}(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}
\Pi^\alpha\Pi^\beta
& \qquad \quad &
\displaystyle
d{\cal F}&=& (C\Gamma_\mu\Gamma_{11})_{\alpha\beta}\Pi^\mu\Pi^\alpha
\Pi^\beta \, .
\label{freedbrane}
\end{array}
\end{equation}
The reason one should start from (\ref{freedbrane})
is that the dual of the algebra defined
by the first five equations is the one obtained
when one computes the algebra of the Noether charges associated with
the supertranslations in the case of the type IIA D$2$-brane (see
\cite{Ham:98}).
The next step, as was done in Sec. \ref{NCESA}, is extending this algebra
with the generators obtained by replacing vector indices by
spinorial ones. In the case of the D$2$-brane this is not difficult
to do because, apart from the equation for $d{\cal F}$,
the free differential algebra one starts with is actually the dimensional
reduction of the $11$-dimensional one
\begin{equation}
d\Pi^{\tilde\mu}=\frac{1}{2}(C\Gamma^{\tilde\mu})_{\alpha\beta}\Pi^\alpha
\Pi^\beta
\, ,\qquad \quad
d\Pi_{\tilde\mu\tilde\nu}=
\frac{1}{2}(C\Gamma_{\tilde\mu\tilde\nu})_{\alpha\beta}\Pi^\alpha
\Pi^\beta \, ,
\label{freeeleven}
\end{equation}
in which one sets $\Pi^{\tilde\mu}\equiv(\Pi^\mu,\Pi^{10}=\Pi)$,
$\Pi_{\tilde\mu\tilde\nu}
\equiv(\Pi_{\mu\nu},\Pi_{\mu 10}=\Pi_\mu^{(z)})$. Since
this $D=11$ algebra has already been extended
recursively in Sec. \ref{NCESA} by the new one-forms
$\Pi_{\tilde\mu\alpha}$ and $\Pi_{\alpha\beta}$, the extended
algebra in $D=10$ is simply its dimensional
reduction, for which $\Pi_{\tilde\mu\alpha}\equiv(\Pi_{\mu\alpha},
\Pi_\alpha^{(z)})$. The result is given by eqs. (\ref{freedbrane}) plus
\begin{eqnarray}
d\Pi_{\mu\alpha}&=&(C\Gamma_{\nu\mu})_{\alpha\beta}\Pi^\nu\Pi^\beta+
(C\Gamma_{11}\Gamma_\mu)_{\alpha\beta}\Pi\Pi^\beta+
(C\Gamma^\nu)_{\alpha\beta}\Pi_{\nu\mu}\Pi^\beta-(C\Gamma_{11})_{\alpha\beta}
\Pi_\mu^{(z)}\Pi^\beta\quad ,
\nonumber
\\
d\Pi_\alpha^{(z)}&=&(C\Gamma_\nu\Gamma_{11})_{\alpha\beta}\Pi^\nu\Pi^\beta
+(C\Gamma^\nu)_{\alpha\beta}\Pi_\nu^{(z)}\Pi^\beta\quad ,
\nonumber
\\
d\Pi_{\alpha\beta}&=&-\frac{1}{2}(C\Gamma_{\mu\nu})_{\alpha\beta}
\Pi^\mu\Pi^\nu-
(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}\Pi^\mu\Pi-\frac{1}{2}
(C\Gamma^\mu)_{\alpha\beta} \Pi_{\mu\nu}\Pi^\nu
\nonumber
\\
& & \mbox{} +\frac{1}{2}(C\Gamma_{11})_{\alpha\beta}
\Pi_{\mu}^{(z)}\Pi^\mu-
\frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}
\Pi_\mu^{(z)}\Pi
+\frac{1}{4}(C\Gamma^\mu)_{\alpha\beta}\Pi_{\mu\delta}\Pi^\delta
\nonumber
\\
& & \mbox{}+ \frac{1}{4}(C\Gamma_{11})_{\alpha\beta}\Pi_\delta^{(z)}\Pi^\delta
+2(C\Gamma^\mu)_{\delta\alpha{'}}\Pi_{\mu\beta{'}}\Pi^\delta+
2(C\Gamma_{11})_{\delta\alpha{'}}\Pi_{\beta{'}}^{(z)}\Pi^\delta
\quad ,
\label{freedbraneext}
\end{eqnarray}
which, apart from the $d{\cal F}$ equation, corresponds to
the dimensional reduction of (\ref{smLa}) (with $Z^\mu=2Z^{\mu 10}$)
\begin{eqnarray}
\{ D_\alpha,D_\beta\} &=& (C\Gamma^\mu)_{\alpha\beta} X_\mu +
(C \Gamma_{\mu\nu})_{\alpha\beta}
Z^{\mu\nu}+(C\Gamma_{11})_{\alpha\beta}Z+(C\Gamma_\mu
\Gamma_{11})_{\alpha\beta}Z^\mu
\nonumber \\
\left[ X_\mu,D_\alpha \right] &=& -
(C\Gamma^{\mu\nu})_{\alpha\beta} Z^{\nu\beta}-
(C\Gamma_\mu
\Gamma_{11})_{\alpha\beta} Z^\beta
\nonumber \\
\left [Z,D_\alpha\right] &=& (C\Gamma_{11}\Gamma_\mu)_{\alpha\beta}Z^{\beta\mu}
\nonumber \\
\left[ X_\mu,X_\nu \right] &=& (C\Gamma^{\mu\nu})_{\alpha\beta} Z^{\alpha\beta}
\nonumber \\
\left[Z,X_\mu\right] &=& (C\Gamma_{11}\Gamma_\mu)_{\alpha\beta}Z^{\alpha\beta}
\nonumber \\
\left[ X_\mu, Z^{\lambda\tau} \right] &=& {1 \over 2}
\delta^{\left[ \lambda \right.}_\mu
(C \Gamma^{\left. \tau \right]})_{\alpha\beta}Z^{\alpha\beta}
\nonumber \\
\left[ X_\mu,Z^\nu\right] &=& \frac{1}{2}\delta^\nu_\mu(C\Gamma_{11})_{\alpha\beta}
Z^{\alpha\beta}
\nonumber \\
\left[ Z,Z^\mu\right] &=& -\frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}Z^{\alpha\beta}
\nonumber \\
\left[ D_\alpha,Z^{\mu\nu} \right] &=& (C \Gamma^{\left[
\mu \right.})_{\alpha\beta} Z^{\left.\nu \right] \beta}
\nonumber \\
\left[ D_\alpha,Z^\mu\right] &=& -(C\Gamma_{11})_{\alpha\beta}Z^{\mu\beta}+
(C\Gamma^\mu)_{\alpha\beta}Z^\beta
\nonumber \\
\{ D_\alpha, Z^{\nu\beta}\} &=&
({1 \over 4}(C \Gamma^\nu)_{\gamma\delta}
\delta^\beta_\alpha +2(C \Gamma^\nu)_{\gamma\alpha}
\delta^\beta_\delta) Z^{\gamma\delta}
\nonumber \\
\{ D_\alpha, Z^\beta\} &=& ({1 \over 4}(C
\Gamma_{11})_{\gamma\delta}\delta^\beta_\alpha+2
(C
\Gamma_{11})_{\gamma\alpha}\delta^\beta_\delta) Z^{\gamma\delta}
\quad .
\label{smLaext}
\end{eqnarray}
We can now show, using the new forms in (\ref{freedbraneext}), that
it is possible to find an invariant
WZ term ${\tilde b}$, $h=d{\tilde b}$, on
the extended superspace. In our case $h$ is given by (eqs. (\ref{expansionh}),
(\ref{generalform}); $k=2,4$)
\begin{equation}
h=(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^\mu\Pi^\nu\Pi^\alpha\Pi^\beta
-(C\Gamma_{11})_{\alpha\beta}\Pi^\alpha\Pi^\beta {\cal F}
\quad . \label{hagain}
\end{equation}
Again, it is possible to expand ${\tilde b}$ as
${\tilde b}=b^{(3)}+b^{(1)}{\cal F}$.
Using this expression in $h=d{\tilde b}$, and identifying the result with $h$, yields
$db^{(1)}=-(C\Gamma_{11})_{\alpha\beta}\Pi^\alpha\Pi^\beta$, from which follows
that $b^{(1)}=-2\Pi$. Similarly,
\begin{equation}
db^{(3)}=-2(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}\Pi\Pi^\mu\Pi^\alpha
\Pi^\beta +(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^\mu\Pi^\nu\Pi^\alpha
\Pi^\beta=(C\Gamma_{\tilde\mu\tilde\nu})_{\alpha\beta}
\Pi^{\tilde\mu}\Pi^{\tilde\nu}\Pi^\alpha\Pi^\beta
\label{wztrivial}
\end{equation}
where in the last equality we have rewritten the expression using the
eleven-dimensional notation. This has the advantage that the expression
for $b^{(3)}$ in $D=11$ was given in (\ref{smbt}),
\begin{equation}
b^{(3)}=\frac{2}{3}\Pi_{\tilde\mu\tilde\nu}\Pi^{\tilde\mu}\Pi^{\tilde
\nu}-\frac{2}{15}\Pi_{\alpha\beta}\Pi^\alpha\Pi^\beta-
\frac{3}{5}\Pi_{\tilde\mu\alpha}\Pi^{\tilde\mu}\Pi^\alpha \quad.
\label{bersez}
\end{equation}
Reducing (\ref{bersez}) to $D=10$ and adding $b^{(1)}{\cal F}=
-2\Pi {\cal F}$ we find the invariant WZ term,
\begin{equation}
{\tilde b}=
\frac{2}{3}\Pi_{\mu\nu}\Pi^\mu\Pi^\nu+\frac{4}{3}\Pi_\mu^{(z)}\Pi^\mu
\Pi-\frac{2}{15}\Pi_{\alpha\beta}\Pi^\alpha\Pi^\beta
-\frac{3}{5}\Pi_{\mu\alpha}\Pi^\mu\Pi^\alpha-\frac{3}{5}\Pi_\alpha^{(z)}
\Pi\Pi^\alpha-2\Pi {\cal F}\quad .
\label{bdbrane}
\end{equation}
This shows that on the extended superspace corresponding to
eqs. (\ref{freedbrane}) and (\ref{freedbraneext}), the WZ term
of the type IIA D$2$-brane can be made invariant,
as was the case for the ordinary $p$-branes.
We expect that this result holds for the other values of $p$.
In contrast with the case of ordinary $p$-branes, the extended
free differential algebra is not the dual of a Lie algebra
because of the equation for the two-form $d{\cal F}$.
However, it is easy to check
that
\begin{equation}
d(\frac{1}{2}\Pi^\alpha\Pi_\alpha^{(z)}-\Pi^\mu\Pi_\mu^{(z)})=
(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}\Pi^\mu\Pi^\alpha
\Pi^\beta \label{6.22}
\end{equation}
so that, on the extended superspace,
we may set
\begin{equation}
{\cal F}=
\frac{1}{2}\Pi^\alpha\Pi_\alpha^{(z)}-\Pi^\mu\Pi_\mu^{(z)}\quad,
\label{fonsuper}
\end{equation}
in accordance with (\ref{freedbrane}), (\ref{freedbraneext}).
This is not a surprising fact since from (\ref{freedbrane})
we see that $d{\cal F}$ is equal to the $h$ corresponding
to the WZ term of the type IIA superstring on a flat background.
So it has to be possible to write it as the differential of an
invariant form ${\tilde b}=
{\tilde b}(\Pi^\mu,\Pi^\alpha,\Pi_\mu^{(z)},\Pi_{\alpha}^{(z)})$
on the fully extended superspace ${\tilde \Sigma}$ of the IIA
superstring. Since ${\cal F}=dA-B$ and $B$ is defined on $\Sigma$,
$dA$ may be written on ${\tilde \Sigma}$.
Making use of its LI forms \emph{i.e.}, of
\begin{equation}
\addtolength{\arraycolsep}{-0.8ex}
\begin{array}{rclcrcl}
\displaystyle
\Pi^\alpha &=& \displaystyle d\theta^\alpha
\, , & \quad &
\displaystyle
\Pi^\mu&=& \displaystyle dx^\mu+\frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}\theta^\alpha
d\theta^\beta
\\[1.3ex]
\displaystyle
\Pi_\mu^{(z)}&=& \displaystyle d\varphi_\mu+\frac{1}{2}
(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}
\theta^\alpha d\theta^\beta
\, , & \quad &
\displaystyle
\Pi_\alpha^{(z)}&=& \displaystyle d\varphi_\alpha-(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}
dx^\mu\theta^\beta-(C\Gamma^\mu)_{\alpha\beta}d\varphi_\mu\theta^\beta
\\[1.3ex]
\displaystyle
& & & & & & \displaystyle -\frac{1}{6}\left[ (C\Gamma_\mu\Gamma_{11})_{\alpha\beta}(C\Gamma^\mu)_{\delta
\epsilon} \! + \! (C\Gamma_\mu\Gamma_{11})_{\delta\epsilon}
(C\Gamma^\mu)_{\alpha
\beta} \right]\theta^\beta \theta^\delta d\theta^\epsilon
\, ,
\end{array}
\end{equation}
it is easy to identify $A$ as the one-form on $\tilde\Sigma$
\begin{equation}
A=\varphi_\mu dx^\mu+\frac{1}{2}\varphi_\alpha d\theta^\alpha \quad .
\label{newA}
\end{equation}
In this way, {\it the customary Born-Infeld worldvolume field $A_i(\xi)$
becomes here $\phi^*(A)$}, with $A$ on $\tilde\Sigma$
given by (\ref{newA}), and its existence may be looked at as
a consequence of extended supersymmetry.
The previous discussion shows that it is natural
to rewrite the action of the D-branes on a flat background
by using \emph{only} objects that are initially defined on the
appropriately extended superspace. We show now that
the Euler Lagrange (EL) equations are still
the same (provided a rather natural condition is met) and that the gauge
transformations of $A_i(\xi)$ can be reinterpreted in the new language. So at this
point it seems that the geometric difference between the ordinary $p$-branes and the
D$p$-branes is that while the action of the former may be defined
from forms on ordinary superspace $\Sigma$, the action of the latter
requires the extended superspace of the IIA superstring if one
whishes to avoid objects that only have a meaning on
the worldvolume. In the IIA superstring case
the extended superspace was also considered, but the
new variables appeared only in the WZ term and as total derivatives
(Sec. \ref{css}) and thus had trivial EL equations. In the D-brane
case, in contrast, these variables have nontrivial EL equations.
Let us now see how the EL equations change by making the substitution
$A\rightarrow \varphi_\mu dx^\mu+\frac{1}{2}\varphi_\alpha d\theta^\alpha$.
Let $I[x^\mu,\theta^\alpha,A_i]$ be the action before making the
subtitution, where $A_i$ are the coordinates of the form
$A$ ($A=A_i d\xi^i$). The EL equations are
\begin{equation}
\frac{\delta I}{\delta x^\mu}=0
\, ,\qquad
\frac{\delta I}{\delta \theta^\alpha}=0
\, ,\qquad
\frac{\delta I}{\delta A_j}=0 \, .
\label{ELbefore}
\end{equation}
When the substitution is made, there are new terms in the equations, and they
read
\begin{equation}
\label{elr}
\begin{array}{rclcrcll}
\displaystyle
\int d\xi{'}^{p+1}\frac{\delta I}{\delta A_j(\xi{'})}
\frac{\delta A_j(\xi{'})}{\delta x^\mu(\xi)}+\frac{\delta I}{\delta
x^\mu} & = & 0
\, , & \qquad &
\displaystyle
\frac{\delta I}{\delta\varphi_\mu} =
\frac{\delta I}{\delta A_j}\partial_j x^\mu & = & 0
\\[2.6ex]
\displaystyle
\int d\xi{'}^{p+1}\frac{\delta I}{\delta A_j(\xi{'})}
\frac{\delta A_j(\xi{'})}{\delta \theta^\alpha(\xi)}+
\frac{\delta I}{\delta \theta^\alpha} &= & 0
\, , & \qquad &
\displaystyle
\frac{\delta I}{\delta\varphi_\alpha}=
\frac{1}{2} \frac{\delta I}{\delta A_j}\partial_j
\theta^\alpha & = & 0 \, ,
\end{array}
\end{equation}
where the additional contributions come from the partial functional
derivative terms. If
$\frac{\delta I}{\delta A_j}\partial_j x^\mu$ and
$\frac{\delta I}{\delta A_j}\partial_j \theta^\alpha$ were zero
without $\frac{\delta I}{\delta A_j}$ being zero, this
would imply the collapse of one worldvolume dimension. Thus, we
must have $\frac{\delta I}{\delta A_j}=0$ which
in the first equation of each set in (\ref{elr})
implies eqs. (\ref{ELbefore}). Hence both actions are equivalent.
In fact, it may be shown that there is an additional
gauge freedom which accounts for the difference of
degrees of freedom between $A$ (eqn. (\ref{newA})) and $A_i(\xi)$,
but we shall not discuss this here\footnote{We thank P. Townsend
for discussions on this point.}.
This seems to indicate that, when
the action on $W$ is obtained from entities on
${\tilde \Sigma}$, there is an additional gauge freedom
which in our formulation plays a r\^ole complementing
that of $\kappa$-symmetry.
Finally, the $U(1)$ gauge field $A_i(\xi)$ on $W$
has a gauge transformation $\delta A_i(\xi)=\partial_i
\Lambda(\xi)$. The question now is what is the gauge transformation
of the component fields if one writes $A_id\xi^i$ as $\phi^*(A)$.
In other words, for a
given $\Lambda(\xi)$, there should be
a transformation of $\varphi_\mu$ and $\varphi_\alpha$ in (\ref{newA})
reproducing $\delta_i\Lambda$. This may be obtained by
means of a superfield $\lambda$ such that
$\phi^*\lambda(x^\mu,
\theta^\alpha)=\Lambda(\xi)$. Then if under a
gauge transformation one defines
$\delta \varphi_\mu= \partial_\mu \lambda$ and $\delta \varphi_\alpha=
2\partial_\alpha \lambda$, $\phi^*(A)$ behaves as expected since then
$\delta\phi^* [\varphi_\mu dx^\mu+
\frac{1}{2}\varphi_\alpha d\theta^\alpha]=\partial_i \Lambda$.
The fact that when the
supersymmetry transformations of a field close only
modulus a gauge transformation one obtains
an extension of a FDA is not restricted to D$p$-branes.
In fact, one may achieve manifest invariance by
introducing an electromagnetic potential on
the worldsheet in the Green-Schwarz superstring action,
in which case the string tension is the circulation of
the potential around the string \cite{Tow:92}
(see also \cite{Azc.Izq.Tow:92})
and a similar result applies to the other scalar
$p$-branes ~\cite{Ber.Lon.Tow:92}. Clearly, the
worldvolume fields introduced there could be defined
on our appropriate extended superspaces as well. As for the
IIB D$p$-brane, an analysis similar to that in this
section for the IIA case would classify them by first
showing that WZ terms exist for odd $p$. In a second stage,
the worldvolume gauge field $A$ may be expressed as the pull-back
of a IIB superspace one-form. In fact, this last point for the $A$
in the $p$=1 IIB D-string case has been discussed very
recently in \cite{Sak:98}
by introducing an appropriate extended group manifold.
We may conclude, then, that the different worldvolume fields
may be expressed in terms of forms defined on suitably extended
superspaces.
\section{Noether charges and D-brane actions}
\label{2bn}
It follows from the discussion of Sec. \ref{casedbr} that the
worldvolume field $A(\xi)$ that appears in the D$2$-brane action
may be written on the superstring extended superspace
parametrized by $(x^\mu,\theta^\alpha,\varphi_\mu, \varphi_\alpha)$.
On the other hand, the D2-WZ term, which is quasi-invariant in
these coordinates, can be made strictly invariant by further extending the
previous superspace to
${\tilde \Sigma}=(\theta^\alpha,x^\mu,\varphi_\mu,
\varphi_\alpha$, $\varphi_{\mu\nu},\varphi_{\mu\alpha},\varphi_{\alpha\beta},
\varphi)$. In this way, the whole action is invariant.
If one now computes the canonical commutators (or Poisson brackets)
of the charges
corresponding to
the symmetries of the action, the resulting algebra is exactly
the RI version of the Lie algebra dual to (\ref{freedbrane})
(removing its last line) plus (\ref{freedbraneext}),
given in (\ref{smLaext}).
The RI generator algebra ($\{Q,Q\}$, etc.) is
the same as (\ref{smLaext}) with an
additional minus sign on the r.h.s.
Let us concentrate on the $\{ Q_\alpha, Q_\beta\}$ commutator,
\begin{equation}
\{ Q_\alpha,Q_\beta\}= (C\Gamma^\mu)_{\alpha\beta}P_\mu +
(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}{\hat Z}^\mu +
(C\Gamma_{\mu\nu})_{\alpha\beta}{\hat Z}^{\mu\nu}
+ (C\Gamma_{11})_{\alpha\beta}{\hat Z}\quad
\label{QQ}
\end{equation}
(there has been a redefinition of the generators so that $\{ Q , Q \}
= + C\Gamma^\mu P_\mu$ etc).
Let us assume that we had written the action, as it is
customary, in terms of $(x^\mu,\theta^\alpha,A)$ alone,
with $A=A(\xi)$ directly defined on $W$. In this case,
the $C\Gamma_{\mu\nu}$ and $C\Gamma_{11}$ contributions
would come from the quasi-invariance of the WZ Lagrangian, while
$C\Gamma_{11}\Gamma_\mu$ would be the result of the contribution of the
$A(\xi)$ field to the Noether current \cite{Ham:98} (see also
\cite{Ber.Tow:98}).
This follows easily from the appropriate definition of the
conserved Noether currents and charges (see, \emph{e.g.}, \cite{Azc.Izq:95})
which include an additional piece when the Lagrangian is
quasi-invariant, a common feature of the conventional actions
for $p$-branes \cite{Azc.Gau.Izq.Tow:89}. In the present
D-branes case, there is an additional contribution due to the
\emph{worldvolume} field $A(\xi)$ since its transformation
properties, $\delta A=\Delta$, are \emph{postulated} to
compensate for those of the composite object $B$,
$\delta B= d\Delta$,
so that ${\cal F}= dA-B$ is invariant. As a result, the
supersymmetry transformations {\it do not close on $A$}, and this
produces an additional term by a mechanism similar to the one in
the standard quasi-invariance case.
These modifications become evident in our context \emph{i.e.},
by formulating the action on the extended superspace. Let us
consider the D2-brane Lagrangian with the quasi-invariant WZ term
$b=b(x^\mu,\theta^\alpha,\varphi_\mu,\varphi_\alpha)$. The conserved
Noether currents then have to include the quasi-invariance
piece. If we wrongly ignored this additional term, the
(canonical formalism) algebra of the corresponding (non-conserved,
non-Noether) charges would be the algebra of the symmetries
$x^\mu,\theta^\alpha,\varphi_\mu,\varphi_\alpha$ of the
Lagrangian, \emph{i.e.}
\begin{equation}
\{ Q_\alpha,Q_\beta\}= (C\Gamma^\mu)_{\alpha\beta}P_\mu +
(C\Gamma_\mu\Gamma_{11})_{\alpha\beta}{\hat Z}^\mu \quad .
\label{2Db}
\end{equation}
The algebra of the conserved Noether charges is not (\ref{2Db}),
however, because these must include the quasi-invariance
contribution. We may find the correct algebra immediatley
by replacing the quasi-invariant WZ term $b$, by
${\tilde b}={\tilde b}(x^\mu,\theta^\alpha,\varphi_\mu,\varphi_\alpha,
\varphi_{\mu\nu},\varphi_{\mu\alpha},\varphi_{\alpha\beta},\varphi)$,
which is manifestly invariant since the transformation properties
of the additional variables $(\varphi_{\mu\nu},
\varphi_{\mu\alpha},\varphi_{\alpha\beta},\varphi)$ remove the
quasi-invariance of $b$. By definition, the transformation
properties of all the coordinates obviously close into the group
law or algebra. Hence, it follows that the algebra of charges
computed using the canonical formalism reproduces (\ref{QQ}), and
that the contributions to ${\hat Z}^{\mu\nu}$ and $\hat Z$ are
entirely due to the WZ term $\tilde b$ (or to the quasi-invariance
of $b(x^\mu,\theta^\alpha,\varphi_\mu,\varphi_\alpha)$ if we
used $b$ instead).
\section{Branes with higher order tensors: the case of the
M$5$-brane}
\label{m5b}
We shall now show that the previous analysis can be applied also
to other extended objects that are neither ordinary $p$-branes
nor D-branes. We shall consider here the case of the $D=11$
$M$5-brane, which contains a worldvolume two-form field $A$
in the action (see \cite{Ban.Lec:97}). The action in a flat bosonic
background depends on $A$ trough $H=dA-C$, where $C$ is a
background three-form. We shall take as our starting point the case with
$C=0$ and with all other forms of rank higher than one in that action
vanishing. We do not need to worry about the (generalized)
self-duality condition for A on the worldvolume, since this
condition may arise as a field equation for an auxiliary field
(see \cite{Ban.Lec:97, Aga.Par:97}). The supersymmetric
action of the M$5$-brane is obtained in two steps. First,
one substitutes $H=dA-C$ for $dA$, where $C$ is a form on
ordinary flat superspace such that
\begin{equation}
dC=-(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^\mu\Pi^\nu\Pi^\alpha\Pi^\beta
\quad , \label{11a}
\end{equation}
and the transformation properties of the worldvolume field $A$ are fixed so
that $H$ is invariant. Secondly, a WZ term is added to
obtain $\kappa$ symmetry.
Let us now find the WZ term in our framework. It should be
obtained from the FDA generated by the abstract invariant forms
$\Pi^\alpha$, $\Pi^\mu$, $H$,
\begin{equation}
d\Pi^\alpha=0
\, ,
\qquad
d\Pi^\mu=\frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}\Pi^\alpha\Pi^\beta\, ,
\qquad
dH=(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^\mu\Pi^\nu\Pi^\alpha\Pi^\beta
\, ,
\label{11b}
\end{equation}
and be given by a CE-non-trivial potential $b$ of a closed form
$h(\Pi^\alpha,\Pi^\mu,H)$. Thus, we have
to solve the problem of finding nontrivial $(p+2)$-cocycles of the FDA
(\ref{11b}). We shall find that there is no solution unless $p=5$.
A general $(p+2)$-form on (\ref{11b}) can be written as
\begin{equation}
h=h^{(p+2)}+h^{(p-1)}H\quad ; \label{11c}
\end{equation}
there are no further powers of $H$ since $H^2\equiv H\wedge
H=0$. The closure of $h$ gives
\begin{equation}
dh^{(p+2)}+dh^{(p-1)}H-(-1)^ph^{(p-1)}
(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^\mu\Pi^\nu\Pi^\alpha\Pi^\beta=0\quad ,
\label{11ca}
\end{equation}
which is equivalent to
\begin{equation}
\addtolength{\arraycolsep}{.3ex}
\begin{array}{l}
\displaystyle
dh^{(p-1)}=0
\\[1.1ex]
\displaystyle
dh^{(p+2)}=(-1)^ph^{(p-1)}(C\Gamma_{\mu\nu})_{\alpha\beta}
\Pi^\mu\Pi^\nu\Pi^\alpha\Pi^\beta \, .
\end{array} \label{11d}
\end{equation}
Now, since $[h]=L^{p+1}$ and $[H]=L^3$,
\begin{equation}
\addtolength{\arraycolsep}{.3ex}
\begin{array}{l}
\displaystyle
h^{(p+2)}=a^{(p+2)}(C\Gamma_{\mu_1\dots\mu_p})_{\alpha\beta}
\Pi^{\mu_1}\dots\Pi^{\mu_p}\Pi^\alpha\Pi^\beta
\\[1.1ex]
\displaystyle
h^{(p-1)}=a^{(p-1)}(C\Gamma_{\mu_1\dots\mu_{p-3}})_{\alpha\beta}
\Pi^{\mu_1}\dots\Pi^{\mu_{p-3}}\Pi^\alpha\Pi^\beta\, ,
\end{array} \label{11e}
\end{equation}
for some constants $a^{(p+2)}$ and $a^{(p-1)}$. The first equation in
(\ref{11d}) requires $a^{(p-1)}=0$ unless $p-3=2$, in which case
$a^{(p-1)}$ is arbitrary due to the identity
$(C\Gamma_{\mu\nu})_{\alpha{'}\beta{'}}(C\Gamma^\nu)_{\delta{'}\epsilon{'}}=0$,
valid in $D$=$4,5,7,11$.
If $p\neq 5$, $a^{(p-1)}=0$ and $h^{(p-1)}=0$, and the second equation of
(\ref{11d}) gives $dh^{(p+2)}=0$, which again implies
$a^{(p+2)}=0$ unless $p=2$. But if $p=2$, we have $h=h^{(2+2)}\propto
(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^{\mu}\Pi^{\nu}\Pi^\alpha\Pi^\beta=dH$,
in which case $h$ is the differential of a LI form
and hence CE-trivial. Thus we are just left with the case $p=5$,
$a^{(5-1)}$ arbitrary. Inserting (\ref{11e}) in (\ref{11d}) gives,
factoring out the $\Pi^\alpha$'s and $\Pi^\mu$'s,
\begin{equation}
\frac{5}{2}a^{(5+2)}(C\Gamma^{\mu_5})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_1\dots\mu_5})_{\delta{'}\epsilon{'}}+
a^{(5-1)}(C\Gamma_{[\mu_1\mu_2})_{\alpha{'}\beta{'}}
(C\Gamma_{\mu_3\mu_4]})_{\delta{'}\epsilon{'}}=0\quad . \label{11f}
\end{equation}
The second identity in (\ref{11identities}) gives
$a^{(7)}=-\frac{2}{15} a^{(4)}$. The resulting closed form
\begin{equation}
h\propto (C\Gamma_{\mu_1\dots\mu_5})_{\alpha\beta}
\Pi^{\mu_1}\dots\Pi^{\mu_5}\Pi^\alpha\Pi^\beta -\frac{15}{2}
(C\Gamma_{\mu_1\mu_2})_{\alpha\beta}
\Pi^{\mu_1}\Pi^{\mu_2}\Pi^\alpha\Pi^\beta H \label{11g}
\end{equation}
is not CE-exact, as may be seen by an argument analogous to that used in
the IIA D-branes case: a LI potential form $b$ would have to be a
scalar six-form depending on
$\Pi^\alpha$, $\Pi^\mu$ and $H$ with dimensions $L^6$, which does not
exist.
It is possible to see, however, that a LI expression for $H$ exists
on the appropriate extended superspace. Since $H$ is a three-form,
it has formally the same properties as the invariant WZ term $\tilde b$
of the M$2$-brane, the extended superspace of which is the
one corresponding to the Lie FDA obtained by the methods of
Sec. \ref{NCESA}. Eqs. (\ref{2.2}), (\ref{dPimu}),
(\ref{dPim1a1}) and
the first of (\ref{dPim1a24}) with $(a_s,a_0,
a_1,a_2)=(\frac{1}{2},\frac{1}{2},1,-\frac{1}{2})$ give,
respectively,
\begin{equation}
\addtolength{\arraycolsep}{-.6ex}
\begin{array}{rclcrcl}
\displaystyle
d\Pi^\alpha &=& \displaystyle 0
\, , & \qquad &
\displaystyle
d\Pi^\mu & = & \displaystyle \frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}\Pi^\alpha\Pi^\beta
\\[1.3ex]
\displaystyle
d\Pi_{\mu\nu} &=& \displaystyle \frac{1}{2}(C\Gamma_{\mu\nu})_{\alpha\beta}\Pi^\alpha
\Pi^\beta
\, , & \qquad &
\displaystyle
d\Pi_{\alpha\beta} &=&
\displaystyle \mbox{}-\frac{1}{2}(C\Gamma_{\mu\nu})_{\alpha\beta}
\Pi^\mu
\Pi^\nu-\frac{1}{2}(C\Gamma^\mu)_{\alpha\beta}
\Pi_{\mu\nu}\Pi^\nu
\\[1.3ex]
\displaystyle
d\Pi_{\mu\alpha} & = &
\displaystyle (C\Gamma_{\nu\mu})_{\alpha\beta}\Pi^\nu\Pi^\beta+
(C\Gamma^\nu)_{\alpha\beta}\Pi_{\nu\mu}\Pi^\beta
\, , & \qquad &
& &
\displaystyle
\mbox{} + \frac{1}{4}(C\Gamma^\mu)_{\alpha\beta}
\Pi_{\mu\delta}\Pi^\delta+ 2(C\Gamma^\mu)_{\delta\alpha{'}}
\Pi_{\mu\beta{'}}\Pi^\delta
\, ,
\label{11h}
\end{array}
\end{equation}
\emph{i.e.}, the dual of (\ref{smLa})). Using again (\ref{smbt}) we may then write
\begin{equation}
H=\frac{2}{3}\Pi^\mu\Pi^\nu\Pi_{\mu\nu}+\frac{3}{5}\Pi^\mu\Pi^\alpha
\Pi_{\mu\alpha}-
\frac{2}{15}\Pi_{\alpha\beta}\Pi^\alpha \Pi^\beta
\quad . \label{11i}
\end{equation}
We might now go on and show that there exists a LI $\tilde b$ such that
$h=d{\tilde b}$ on a suitably extended superspace;
we shall omit its expression~\cite{Sez:97} since it is not needed
below. What we wish to show is that now we may
use (\ref{11i}) to replace the worldvolume two-form $A(\xi)$ by
the pull-back of the two-form $A$ on extended superspace given by
\begin{eqnarray}
A &=& \frac{2}{3}\varphi_{\mu\nu}dx^\mu dx^\nu-\frac{3}{5}\varphi_{\mu\alpha}
dx^\mu d\theta^\alpha-
\frac{2}{15}\varphi_{\alpha\beta}d\theta^\alpha d\theta^\beta
\nonumber \\
& & \mbox{} +\frac{1}{30}\varphi_{\mu\nu}x^\mu(C\Gamma^\nu)_{\alpha\beta}
d\theta^\alpha d\theta^\beta+
\frac{11}{30}\varphi_{\mu\nu}dx^\mu(C\Gamma^\nu)_{\alpha\beta}
\theta^\alpha d\theta^\beta-\frac{13}{180}\varphi_{\mu\nu}
(C\Gamma^\mu)_{\alpha\beta}(C\Gamma^\nu)_{\delta\epsilon}
\theta^\alpha d\theta^\beta \theta^\delta d\theta^\epsilon
\nonumber \\
& & \mbox{} + \frac{1}{10}\varphi_{\mu\alpha}(C\Gamma^\mu)_{\delta\epsilon}
\theta^\delta d\theta^\epsilon d\theta^\alpha
+ \frac{1}{20}\varphi_{\mu\alpha}(C\Gamma^\mu)_{\delta\epsilon}
d\theta^\delta d\theta^\epsilon \theta^\alpha \quad .
\label{11j}
\end{eqnarray}
Again, this expression may also be used to find the gauge
transformation $\delta A(\xi)=d\Lambda(\xi)$. This is achieved by the
one-form on
superspace $\lambda=\lambda_\mu dx^\mu+\lambda_\alpha d\theta^\alpha$,
$\phi^*(\lambda)=\Lambda(\xi)$. Then, if
\begin{eqnarray}
\delta\varphi_{\mu\nu}&=&\frac{3}{2}
\partial_{[\mu}\lambda_{\nu]}\quad ,
\nonumber \\
\delta\varphi_{\mu\alpha}&=&\mbox{} -\frac{5}{3}
(\partial_\mu\lambda_\alpha+\partial_\alpha\lambda_\mu)+\frac{11}{12}
\partial_{[\mu}\lambda_{\nu]}(C\Gamma^\nu)_{\alpha\beta}\theta^\beta\quad ,
\nonumber \\
\delta\varphi_{\alpha\beta}&=&\mbox{} -\frac{15}{2}\partial_{\alpha{'}}
\lambda_{\beta{'}}+\frac{15}{21}(C\Gamma^\mu)_{\alpha\beta}\theta^\delta
(\partial_\mu\lambda_\delta+\partial_\delta\lambda_\mu)
+\frac{15}{12}(C\Gamma^\mu)_{\delta\alpha{'}}\theta^\delta
(\partial_\mu\lambda_{\beta{'}}+\partial_{\beta{'}}\lambda_\mu)
\nonumber \\
& & \mbox{}- \frac{139}{240}(C\Gamma^\mu)_{\delta\beta}(C\Gamma^\nu)_{\alpha\epsilon}
\theta^\delta\theta^\epsilon \partial_{[\mu}\lambda_{\nu]}+
\frac{1}{20}x^\mu(C\Gamma^\nu)_{\alpha\beta}\partial_{[\mu}\lambda_{\nu]}
\label{gaugemem}
\end{eqnarray}
one obtains $\delta\phi^*(A)=d\Lambda(\xi)$.
As in the previous D-brane case, the EL equations derived from the
action $I[x^\mu,\theta^\alpha,A_{ij}]$,
\begin{equation}
\frac{\delta I}{\delta x^\mu}=0\quad ,
\quad \quad \quad
\frac{\delta I}{\delta \theta^\alpha}=0\quad ,
\quad \quad \quad
\frac{\delta I}{\delta A_{ij}}=0\, , \label{11k}
\end{equation}
are equivalent to the ones corresponding to the new action in which
$A(\xi)$ is the pull-back of (\ref{11j}).
Indeed, the equation for $\varphi_{\alpha\beta}$
gives $\frac{\delta I}{\delta A_{ij}}\partial_i
\theta^\alpha\partial_j\theta^\beta=0$, and substituting it into that of
$\varphi_{\mu\alpha}$,
\begin{equation}
\frac{\delta I}{\delta A_{ij}}
\left(-\frac{3}{5}\partial_ix^\mu\partial_j
\theta^\alpha+\frac{1}{10}(C\Gamma^\mu)_{\delta\epsilon}\theta^\delta
\partial_i\theta^\epsilon\partial_j \theta^\alpha+ \frac{1}{20}
(C\Gamma^\mu)_{\delta\epsilon}\theta^\alpha \partial_i\theta^\delta
\partial_j\theta^\epsilon\right)=0
\quad,
\label{onemoreeqn}
\end{equation}
gives $\frac{\delta I}{\delta A_{ij}}\partial_i x^\mu\partial_j
\theta^\alpha=0$ and so on. Therefore one obtains
\[
\addtolength{\arraycolsep}{-.5ex}
\begin{array}{rclcrclcrcl}
\displaystyle
\int d\xi{'}^{p+1}\frac{1}{2}\frac{\delta I}{\delta A_{ij}(\xi{'})}
\frac{\delta A_{ij}(\xi{'})}{\delta x^\mu(\xi)}+\frac{\delta I}{\delta
x^\mu}
& = &
\displaystyle
0
\, , & \qquad &
\displaystyle
\frac{\delta I}{\delta A_{ij}} \partial_i x^\mu \partial_j x^\nu
& = &
0
\, , & \qquad &
\displaystyle
& &
\\[3ex]
\displaystyle
\int d\xi{'}^{p+1}\frac{1}{2}\frac{\delta I}{\delta A_{ij}(\xi{'})}
\frac{\delta A_{ij}(\xi{'})}{\delta \theta^\alpha(\xi)}+
\frac{\delta I}{\delta \theta^\alpha}
& = &
0
\, , & \qquad &
\displaystyle
\frac{\delta I}{\delta A_{ij}}\partial_i x^\mu\partial_j \theta^\alpha
& = &
0
\, , & \qquad &
\displaystyle
\frac{\delta I}{\delta A_{ij}}\partial_i \theta^\alpha\partial_j\theta^\beta
&=&
0
\, .
\end{array}
\label{11l}
\]
The second equation
implies $\frac{\delta I}{\delta A_{ij}} \partial_i x^\mu=0$
for all $\mu$ if one wants to avoid the possibility of one
dimension of the object collapsing. This in turn implies
$\frac{\delta I}{\delta A_{ij}}=0$ for the same reason, and
inserting this equation into (\ref{11l}) gives (\ref{11k}).
\section{Conclusions}
We have provided in this paper a unified
approach to the study of various $p$-branes by defining
them on suitably extended superspaces.
All of these are supergroup manifolds, extensions of the basic odd abelian
groups ${\text{sTr}}_D$ determined by the spinors of the
specific theory considered. The extension algorithms in
Secs. ~\ref{Mces} and ~\ref{NCESA} show how they depend,
when they do, on specific identities for $\Gamma$-matrices.
The central extensions do not need any
$\Gamma$-identities, but the non-central ones require the identities
(\ref{Gammaprop}), precisely the ones needed to define the WZ
terms of the old branescan.
The centrally extended superspaces are associated with (topological)
charges, but the introduction of manifestly supersymmetric WZ terms
requires the addition of non-central variables, already
for the branes of the old branescan. When the procedure is
applied to D$p$-branes, it is seen that all the fields in their
action may also be defined by pullbacks of entities on the
previously introduced superspaces. In the language of FDA's, our
results show that all the FDA's involved in the formulation
of the $p$-branes considered here become {\it Lie} FDA's on
suitably extended superspaces. We conjecture, in view of the
previous discussion, that this is the case in general and that
there exists an extended superspace definition for all fields
appearing in the action of the various $p$-branes. In other
words, there exists a kind of field/extended superspaces democracy
by which all brane worldvolume fields are pullbacks from some target
superspace $\tilde\Sigma$.
The appropriate ${\tilde \Sigma}$ of the theory is given by an
extension of a certain $\text{sTr}_D$ and,
using ${\tilde \Sigma}$, the action can be defined in a manifestly
invariant form. In fact, in this field/extended superspace
democracy context, the invariance properties
seem to characterize essentially the superbrane actions.
It should not come then as a suprise
that $\kappa$-symmetric actions may also be introduced
for D$p$-branes, as in \cite{Aga.Pop.Sch:97} for D-branes
with rigid IIA and IIB superPoincar\'e symmetry. As
is the case for ordinary $p$-branes, $\kappa$-symmetry is
achieved when the relative coefficient of the
kinetic and WZ-like part is such that the Bogomol'nyi
bound is saturated.
Our extensions provide at the same time a connection between
the CE cocycles and the mechanism of partial breaking of
supersymmetry. The CE ($p$+2)-cocycles lead to
(extended) loop-type or worldvolume current algebras
(see, {\it e.g.} \cite{Azc.Gau.Izq.Tow:89,Azc.Izq.Tow:91,
Ber.Del.Sok:91,Ber.How.Pop.Sez.Sok:91,Ber.Per.Sez.Ste.Tow:93})
and the two-cocycle to the corresponding
algebra of charges defining the extended superspace algebra. The
new variables in the extended superspaces are also essential
to define (invariant) actions. They may also be relevant
in the problem of quantisation, the formulation of dualities
(see Sec.~\ref{sectene1}) and in the formulation of
the additional gauge freedom hidden in the definition
of some superbrane fields, the worldvolume definition of which
reflects an election of gauge. We suspect that the mathematical
existence of the extensions considered here has a deeper
meaning beyond the aspects discussed in this paper.
\vskip .5cm
\noindent
\emph{Note added}. After completion of this paper an
article \cite{Abe.Hat.Kam.Tok:99} has appeared in hep-th, where
an approach similar to that in Sec. \ref{Dbext} for the field
$A$ is given for the IIB D-brane case.
\vskip .5cm
\noindent
{\bf Acknowledgements}:
This research has been partially supported by the research grant
PB96-0756 from the DGICYT, Spain. C.C. wishes to thank the Spanish
Ministry of Education and Culture for a post-doctoral fellowship
and J.M.I acknowledges a grant from the Junta de Castilla y
Le\'on. J.C.P.B. thanks the CSIC and the Ministerio
de Educaci\'on for an FPI grant. Finally, the authors
wish to thank Paul Townsend for helpful discussions.
\vskip .8cm
\noindent |
\section{Introduction}
During an inflationary stage driven by the potential energy of a classical
scalar field, the temperature of the universe was redshifted to almost zero.
Thus a mechanism to raise the temperature of the universe at the end of the
inflationary stage is required to match this scenario with the standard hot
Big Bang cosmology \cite{Kolb90,Lin90}.
Currently it is assumed that soon after the end of inflation, the scalar field
began to oscillate near the minimum of its effective potential. Decay of this
scalar field through particle production and thermalization of its energy with
a huge increase in temperature and large entropy production, occurred during
the reheating period. It started with violent particle production through
parametric resonance, also called the preheating process, and lead to a highly
nonequilibrium distribution of the produced particles, subsequently relaxing
to an equilibrium state \cite{KoLiSta1,Boya1,Shta}.
In this paper we will consider the decay of the massive homogeneous inflaton
scalar field $\phi$ coupled with light bosons $\chi$, as it arises in chaotic
inflationary models. A detailed quantum mechanical analysis is quite
complicated and involves taking into account nonlinear processes, backreaction
effects, rescattering, etc \cite{Lin97}. Further, since particles created
during the preheating stage are far from equilibrium, their thermalization is
achieved via collisional relaxation. This process of thermalization has not
been understood yet from first principles \cite{Hu97a}. Here,
we follow an alternative description
of the physics, and instead of both fields and
their interaction we consider an effective, phenomenological model in
terms of an out-of-equilibrium mixture of two reacting fluids within the
framework of relativistic irreversible thermodynamics \cite{Isr76,Ger90}. The
difference in the equations of state of the fluid components and the
simultaneous deviation from the detailed balance in the rate equations for the
interfluid reactions give rise to an entropy production effect that manifests
itself as a reactive bulk pressure \cite{Zim97b}. This two-fluid model can be
described in the effective one-fluid picture, based on a single temperature
that we identify with the temperature of the reheating process. One of the
advantages of this approach is that it is possible to treat selfconsistently
the dynamics of geometry and matter \cite{ZPM}.
In this phenomenological approach we ignore microscopic details and
concentrate on most relevant macroscopic magnitudes like particle number
density, temperature, entropy production. The net effect of the interaction
term between both fields is collected in the particle number rate of change,
and the finite temperature takes into account the backreaction effect of the
produced particles. A phenomenological approach was initially proposed in
\cite{ZPM}. The methods used in that paper for the solution of the evolution
equations lead to a vanishing bulk viscous contribution to the speed of sound.
However their model depends strongly on this contribution so that their
procedure of calculation needs to be improved. In this direction we introduce
a physically founded perturbative scheme that yields more satisfactory
results.
The plan of this paper is as follows. Section 2 introduces the two-fluid
magnitudes and section 3 presents main equations of the model in the
one-temperature picture. In section 4 we introduce an expansion in powers of
a parameter that measures de departure from perfect fluid behavior. This
allows us to solve the system of equations order by order. In section 5 the
model with a linear ansatz for the decay rate is solved, and we show that it
reaches the reheating temperature only for very large times. We propose in
section 6 a quadratic ansatz for the decay rate that is better motivated both
from microphysical and geometrical point of view. We investigate thoroughly
the evolution of the physical magnitudes and we find that the behavior of this
model is very interesting. Finally, conclusions are stated in section 7.
\section{Two-fluid picture}
For the two-fluid model we assume that
the energy-momentum tensor $T ^{ik}$ splits into two perfect fluid parts,
\begin{equation}
T ^{ik} = T ^{ik}_{_{1 }}
+ T ^{ik}_{_{2 }} {\mbox{ , }}
\label{1}
\end{equation}
\noindent
with ($A = 1, 2$)
\begin{equation}
T^{ik}_{_{A }} = \rho_{_{A }} u^{i}u^{k}
+ p_{_{A }} h^{ik}{\mbox{ .}} \label{2}
\end{equation}
\noindent $\rho_{_{A }}$ is the energy density and $p_{_{A }}$ is the
equilibrium pressure of species $A$. For simplicity we assume that both
components share the same $4$-velocity $u^{i}$. The quantity $h^{ik}$ is the
projection tensor $h^{ik} =g^{ik} + u^{i}u^{k}$. The particle flow vector
$N_{_{A }}^{i}$ of species $A$ is defined as
\begin{equation}
N_{_{A }}^{i} = n_{_{A }}u^{i}{\mbox{ , }} \label{3}
\end{equation}
\noindent where $n _{_{A }}$ is the particle number density. We are interested
in situations where neither the particle numbers nor the energy-momenta of the
components are separately conserved. The
balance laws for the particle numbers are
\begin{equation}
N _{_{A } ;i}^{i} = \dot{n}_{_{A }}
+ \Theta n_{_{A }} = n _{_{A }} \Gamma _{_{A}}
{\mbox{ , }}
\label{4}
\end{equation}
\noindent where $\Theta \equiv u^{i}_{;i}$ is the fluid expansion and $\Gamma
_{_{A}}$ is the rate of change of the number of particles of species $A$.
There is particle production for $\Gamma _{_{A}} > 0$ and particle decay for
$\Gamma _{_{A}} < 0$, respectively. For $\Gamma _{_{A}} = 0$, we have separate
particle number conservation.
Interactions between the fluid components amount to the mutual exchange of
energy and momentum. Consequently, there will be no local energy-momentum
conservation for the subsystems separately. Only the energy-momentum tensor of
the system as a whole is conserved.
Denoting the loss- and source-terms in the separate balances by $t ^{i}_{_{A
}}$, we write
\begin{equation}
T ^{ik}_{_{A } ;k} = - t _{_{A }}^{i} {\mbox{ , }}
\label{9}
\end{equation}
\noindent
implying
\begin{equation}
\dot{\rho}_{_{A }}
+ \Theta\left(\rho_{_{A }}
+ p_{_{A }}\right)
= u _{a} t _{_{A }}^{a}
{\mbox{ , }}
\label{10}
\end{equation}
\noindent
and
\begin{equation}
\left(\rho _{_{A }}
+ p _{_{A}}\right) \dot{u}^{a}
+ p _{_{A } ,k}h ^{ak}
= - h ^{a}_{i}t ^{i}_{_{A }} {\mbox{ .}}
\label{11}
\end{equation}
\noindent All the considerations to follow will be independent of the specific
structure of the $t _{_{A }}^{i}$.
Each component is governed by a separate Gibbs equation
\begin{equation}
T _{_{A }} \mbox{d} s _{_{A }}
= \mbox{d} \frac{\rho _{_{A }}}{n _{_{A }}}
+ p _{_{A }}
\mbox{d} \frac{1}{n _{_{A }}} {\mbox{ , }}
\label{12}
\end{equation}
\noindent where $s _{_{A }}$ is the entropy per particle of species $A$. Using
eqs.(\ref{4}) and (\ref{10}) one finds for the time behaviour of the entropy
per particle
\begin{equation}
n _{_{A }} T _{_{A }}
\dot{s}_{_{A }} = u _{a} t _{_{A }}^{a}
- \left(\rho _{_{A }}
+ p _{_{A }}\right) \Gamma _{_{A}} {\mbox{ .}}
\label{13}
\end{equation}
\noindent With nonvanishing source terms in the balances for $n _{_{A }}$ and
$\rho _{_{A }}$, the change in the entropy per particle is different from zero
in general.
The equations of state of the fluid components are assumed to have the general
form
\begin{equation}
p_{_{A }} = p_{_{A }}
\left(n_{_{A }}, T_{_{A }}\right) \label{14}
\end{equation}
\noindent
and
\begin{equation}
\rho_{_{A }} =
\rho_{_{A }} \left(n_{_{A }},
T_{_{A }}\right){\mbox{ , }}
\label{15}
\end{equation}
\noindent i.e., particle number densities $n_{_{A }}$ and temperatures $T_{_{A
}}$ are regarded as the basic thermodynamical variables. The temperatures of
the fluids are different in general.
Differentiating relation (\ref{15}), using the balances (\ref{4}) and
(\ref{10}) as well as the general relation
\begin{equation}
\frac{\partial \rho_{_{A }}}{\partial n_{_{A }}} =
\frac{\rho_{_{A }} + p_{_{A }}}
{n_{_{A }}}
- \frac{T_{_{A }}}{n_{_{A }}}
\frac{\partial p_{_{A }}}
{\partial T_{_{A }}} {\mbox{ , }}
\label{16}
\end{equation}
\noindent that follows from the requirement that the entropy is a state
function, we find the following expression for the temperature behaviour
\cite{Calv,LiGer,ZPRD}:
\begin{equation}
\dot{T}_{_{A }} = - T_{_{A}} \left(\Theta -
\Gamma _{_{A}} \right)
\frac{\partial p_{_{A}}/\partial T_{_{A}}}{\partial \rho_{_{A}}/
\partial T_{_{A}}}
+ \frac{u _{a} t _{_{A}}^{a} - \Gamma _{_{A}} \left(\rho _{_{A}}
+ p _{_{A}}\right)}
{\partial \rho _{_{A}}/ \partial T _{_{A}}}
{\mbox{ .}}
\label{17}
\end{equation}
The entropy flow vector $S _{_{A}}^{a}$ is defined by
\begin{equation}
S _{_{A}}^{a} = n _{_{A}} s _{_{A}} u ^{a} {\mbox{ , }}
\label{18}
\end{equation}
\noindent and the contribution of component $A$ to the entropy production
density becomes
\begin{eqnarray}
S _{_{A} ;a}^{a} &=& n _{_{A}} s _{_{A}} \Gamma _{_{A}} + n _{_{A}}
\dot{s}_{_{A}}\nonumber\\
&=& \left(s _{_{A}} - \frac{\rho _{_{A}}
+ p _{_{A}}}{n _{_{A}}T _{_{A}}}
\right)
n _{_{A}}\Gamma _{_{A}} + \frac{u _{a} t _{_{A}}^{a}}{T _{_{A}}}
{\mbox{ , }}
\label{19}
\end{eqnarray}
\noindent
where relation (\ref{13}) has been used.
According to eq.(\ref{9}) the condition of energy-momentum conservation for
the system as a whole,
\begin{equation}
\left(T _{_{1}}^{ik} + T _{_{2}}^{ik}\right)_{;k} = 0 {\mbox{ , }}
\label{20}
\end{equation}
\noindent
implies
\begin{equation}
t _{_{1}}^{a} = - t _{_{2}}^{a} {\mbox{ .}}
\label{21}
\end{equation}
\noindent There is no corresponding condition, however, for the particle
number balance as a whole. Defining the integral particle number density $n$
as
\begin{equation}
n = n _{_{1}} + n _{_{2}} {\mbox{ , }}
\label{22}
\end{equation}
\noindent
we have
\begin{equation}
\dot{n} + \Theta n = n \Gamma {\mbox{ , }}
\label{23}
\end{equation}
\noindent
where
\begin{equation}
n \Gamma = n _{_{1}}\Gamma _{_{1}} + n _{_{2}}\Gamma _{_{2}}
{\mbox{ .}}
\label{24}
\end{equation}
\noindent $\Gamma$ is the rate by which the total particle number $n$ changes.
The entropy per particle is \cite{Groot}
\begin{equation}
s _{_{A}} = \frac{\rho _{_{A}} + p _{_{A}}}{n _{_{A}}
T _{_{A}}} - \frac{\mu _{_{A}}}
{T _{_{A}}}
{\mbox{ , }}
\label{25}
\end{equation}
\noindent where $\mu _{_{A}}$ is the chemical potential of species $A$.
Introducing the last expression into eq.(\ref{19}) yields
\begin{equation}
S ^{a}_{_{A} ;a} = - \frac{\mu _{_{A}}}{T _{_{A}}}
n _{_{A}}\Gamma _{_{A}}
+ \frac{u _{a}t ^{a}_{_{A}}}{T _{_{A}}} {\mbox{ .}}
\label{26}
\end{equation}
For the total entropy production density
\begin{equation}
S ^{a}_{;a} = S ^{a}_{_{1} ;a} + S ^{a}_{_{2} ;a}
\label{27}
\end{equation}
\noindent
we obtain
\begin{equation}
S ^{a}_{;a} = - \frac{\mu _{_{2}}}{T _{_{2}}}n \Gamma
- \left(\frac{\mu _{_{1}}}{T _{_{1}}} - \frac{\mu _{_{2}}}{T _{_{2}}}
\right)
n _{_{1}}\Gamma _{_{1}}
+ \left(\frac{1}{T _{_{1}}} - \frac{1}{T _{_{2}}}\right)u _{a}
t _{_{1}}^{a} {\mbox{ .}}
\label{28}
\end{equation}
We assume that the entropy per particle of
each of the components is preserved. The particles decay or are produced
with a fixed entropy $s _{_{A}}$. This `isentropy' condition amounts to the
assumption that the particles at any stage are amenable to a perfect fluid
description. When $\dot{s}_{_{A}} = 0$, we get from (\ref{13})
\begin{equation}
u _{a} t ^{a}_{_{A}} = \left(\rho _{_{A}} + p _{_{A}}\right)
\Gamma _{_{A}}
{\mbox{ , }}
\label{30}
\end{equation}
\noindent
and combining eqs.(\ref{21}) and (\ref{30}), one has
\begin{equation}
u _{a} t ^{a}_{_{1}} = \left(\rho _{_{1}} + p _{_{1}}\right)
\Gamma _{_{1}}
= - u _{a} t ^{a}_{_{2}} = - \left(\rho _{_{2}} + p _{_{2}}\right)
\Gamma _{_{2}}
{\mbox{ , }}
\label{31}
\end{equation}
\noindent
which provides us with a relation between the rates
$\Gamma _{_{1}}$ and
$\Gamma _{_{2}}$:
\begin{equation}
\Gamma _{_{2}} = - \frac{\rho _{_{1}} + p _{_{1}}}
{\rho _{_{2}} + p _{_{2}}}
\Gamma _{_{1}}
{\mbox{ .}}
\label{32}
\end{equation}
\noindent Inserting the last relation into equation (\ref{24}) yields
\begin{equation}
n \Gamma = n _{_{1}}\Gamma _{_{1}} h _{_{1}}
\left(\frac{1}{h _{_{1}}}
- \frac{1}{h _{_{2}}}\right) {\mbox{ .}}
\label{33}
\end{equation}
\noindent The quantities $h _{_{A}} \equiv \left(\rho _{_{A}} + p _{_{A}}
\right)/ n _{_{A}}$ are the enthalpies per particle.
With the relations (\ref{30}) and (\ref{33}) the entropy production
density (\ref{28}) becomes
\begin{equation}
S ^{a}_{;a} = \left(\rho _{_{1}} + p _{_{1}}\right)
\left[\frac{n _{_{1}} s _{_{1}}}{\rho _{_{1}} + p _{_{1}}}
- \frac{n _{_{2}} s _{_{2}}}{\rho _{_{2}} + p _{_{2}}}\right]
\Gamma _{_{1}}
= n _{_{1}}\Gamma _{_{1}} h _{_{1}}
\left[\frac{s _{_{1}}}{h _{_{1}} }
- \frac{ s _{_{2}}}{h _{_{2}}}\right]
{\mbox{ .}}
\label{34}
\end{equation}
\noindent We emphasize that according to the equations of state (\ref{14}) and
(\ref{15}) the quantities $\rho _{1}$, $p _{_{1}}$ and $s _{_{1}}$ depend on
$T _{_{1}}$, while $\rho _{_{2}}$, $p _{_{2}}$ and $s _{_{2}}$ depend on $T
_{_{2}}$. In general, we have $T _{_{1}} \neq T _{_{2}}$.
To complete the two-fluid model we give the thermodynamical properties of both
fluids. If we ignore effects associated with particle creation, after
inflation the field $\phi$ oscillates near the point $\phi=0$ at a frequency
given by its mass $m$. Its oscillation amplitude $\Phi$ falls off as $t^{-1}$
and its mean energy density $\rho_{\phi}= m^2\Phi^2/2$ decreases as $a^{-3}$,
in the same way as pressureless dust \cite{Bel85}. Hence we represent the
massive inflaton field as a fluid, named $1$, of nonrelativistic particles
with mass $m$, energy density $\rho_1=\rho_\phi$, pressure $p_1\ll \rho_1$ and
number density $n_1\simeq\rho_1/m$. On the other hand, the decay products of
the scalar field $\phi$ are ultrarelativistic for $m \gg m_\chi$, and we model
them as an ideal relativistic fluid of massless particles, with energy density
$\rho_2$, equilibrium pressure $p_2=\rho_2/3$ and particle density $n_2$. We
also assume that both fluids are classical and ideal and that each species
remain in thermal equilibrium along the reheating process. That is, we neglect
dissipative effects for each of the fluids in comparison with particle
production effects.
\section{Effective one-temperature picture}
We have also the effective one-temperature
alternative description of a two-component fluid that is
based on a single Gibbs-equation for the system as a whole:
\begin{equation}
T \mbox{d}s = \mbox{d} \frac{\rho}{n} + p \mbox{d} \frac{1}{n}
- \left(\mu _{_{1}} - \mu _{_{2}}\right) \mbox{d} \frac{n _{_{1}}}{n}
{\mbox{ , }}
\label{35}
\end{equation}
\noindent where $p$ is the equilibrium pressure, $\rho$ is the energy density
and $s$ is the entropy per particle. The temperature $T$ is the equilibrium
temperature of the whole system and is defined by \cite{UI,Zim96a}
\begin{equation}
\rho_{_{1}}\left(n_{_{1}},T_{_{1}}\right)
+ \rho_{_{2}}\left(n_{_{2}},T_{_{2}}\right)
= \rho \left(n, n _{_{1}}, T\right)
{\mbox{ .}}
\label{38}
\end{equation}
\noindent
Furthermore, we assume that the cosmic fluid as a whole is
characterized by the equations of state
\begin{equation}
p = p\left(n, n _{_{1}}, T\right)
\label{36}
\end{equation}
\noindent
and
\begin{equation}
\rho = \rho \left(n, n _{_{1}}, T\right){\mbox{ , }}
\label{37}
\end{equation}
The temperatures $T _{_{1}}$ and $T _{_{2}}$ do not appear as variables in the
effective one-temperature description. An (approximate) equilibrium
for the entire system is assumed to be established through the interactions
between the subsystems on the right-hand side of the balances (\ref{10}) and
(\ref{11}).
In this picture the cosmic
fluid splits into two effective fluid components whose equations of state are
assumed to have the same form as those of the "real" fluids, but shear this
common temperature. So, these two effective fluids are in a (formal) thermal
equilibrium between them while the "real" fluids are not.
If we set $T _{_{1}}=T _{_{2}}=T$
in (\ref{25}) and make the splitting for the energy density of the system
\begin{equation} \label{rho}
\rho \left(T\right) = \rho _{_{1}}\left(T\right) + \rho
_{_{2}}\left(T\right)
\end{equation}
\noindent
and for its equilibrium pressure
\begin{equation}\label{p}
p\left(T\right) = p _{_{1}}\left(T\right)
+ p _{_{2}}\left(T\right),
\end{equation}
\noindent
the description based on relation
(\ref{35}) is consistent with the description relying on the relations
(\ref{12}) for $ns \left(T\right) = n _{_{1}} s _{_{1}}\left(T\right) + n
_{_{2}} s _{_{2}}\left(T\right)$.
As long as the pressures are
those for classical gases, $p _{_{A}} = n _{_{A}} T$, the equilibrium
pressure $p$ of the system as a whole depends on $n = n _{_{1}} + n _{_{2}}$
only and the separate dependence on $n _{_{1}}$ on the right-hand side of
eq.(\ref{36}) may be omitted.
For components out of equilibrium, even when $\Gamma =
\Gamma _{_{1}} = \Gamma _{_{2}} = 0$, the total pressure is generally
different to the sum of the partial pressures. In this case we have
\cite{Zim96a}
\begin{equation}
p_{_{1}}\left(n_{_{1}},T_{_{1}}\right)
+ p_{_{2}}\left(n_{_{2}},T_{_{2}}\right)
= p\left(n, T\right)+\pi_d
{\mbox{ .}} \label{39}
\end{equation}
\noindent instead of (\ref{p}). Here $\pi_d$ is the viscous pressure arising
from differential temperature variation rate between both fluids
\cite{Zim96a}. From eq.(\ref{17}) the cooling rate $\dot{T}_{_{1}}/T _{_{1}}$
is different from $\dot{T}_{_{2}}/T _{_{2}}$ even for $\Gamma _{_{1}} = \Gamma
_{_{2}} = 0$ if the subsystems are governed by different equations of state.
The expansion of the Universe tends to increase the difference between $T
_{_{1}}$ and $T _{_{2}}$.
On the other hand, deviations from detailed balance, i.e., $\Gamma
_{_{A}} \neq 0$, leading to $\Gamma \neq 0$ in general, generate an
effective `reactive' bulk pressure $\pi$. For the corresponding
energy-momentum tensor of the system as a whole we write
\begin{equation}
T ^{ik} = \rho u ^{i}u ^{k} + \left(p + \pi \right) h ^{ik}
{\mbox{ .}}
\label{40}
\end{equation}
\noindent The reactive bulk pressure is determined by the consistency of the
expression for the entropy production density obtained in the two-temperature
picture (\ref{34}) and the expression obtained in the one-temperature picture
\begin{equation} \label{dS}
S^\mu_{;\mu}= s n \Gamma + n\dot s
\end{equation}
\noindent The last one arises from the entropy flow vector $S^\mu=snu^\mu$
\cite{Isr76}, where $s$ is the entropy per particle and $u^\mu$ is the
four-velocity of the fluid. It was estimated that $\pi_d$ is one order of
magnitude smaller than $\pi$ \cite{Zim97b}.
From the Gibbs-equation (\ref{35}) one finds for the change
in the entropy per particle
\begin{equation}
n \dot{s} = - \frac{\Theta }{T}\pi - \frac{\rho +
p}{T}\Gamma
- \frac{n _{_{1}} n _{_{2}}}{n}\left(\frac{\mu _{_{1}}
- \mu _{_{2}}}{T}\right)
\left(\Gamma _{_{1}} - \Gamma _{_{2}}\right) {\mbox{ .}}
\label{42}
\end{equation}
\noindent
Even for $\dot{s}_{_{1}} = \dot{s}_{_{2}} = 0$ we have $\dot{s} \neq 0$ in
general. However, during the preheating stage, particle production dominates
over per particle entropy change as the main entropy production source.
The equilibrium temperature of the system $T$ is the key magnitude of our
model. On the one hand we identified it with the reheating temperature, and on the
other hand its definition (\ref{38}) provides a mean to account for
backreaction effects of the created relativistic particles on their massive
counterparts. Its evolution law is given by \cite{Zim97b}
\begin{equation}\label{dT}
\frac{\dot{T}}{T} = \left(\Gamma-\Theta\right)
\frac{\partial p/\partial T}{\partial{\rho }/\partial T}
- \frac{\Theta \pi }{T \partial\rho/\partial T} \,
\end{equation}
The Einstein field equations for a spatially-flat
Robertson-Walker spacetime are
\begin{equation} \label{ZPM28}
3H^{2} = \kappa\rho
\ ,
\end{equation}
\begin{equation}
\dot{H} = - {\kappa\over 2}\left(\rho + p + \pi\right)
\ , \label{ZPM29}
\end{equation}
\noindent where $\kappa$ is Einstein's gravitational constant and $H=
\dot{a}/a$, $a$ is the cosmic scale factor. We use units such that $c=1$,
$k_B=1$ and $\hbar=1$, then $\kappa=8\pi/M_P^2$, where $M_P$ is the Planck
mass. Finally a dot denotes derivative with respect to comoving time.
Conservation of (\ref{40}) leads to
\begin{equation} \label{drho}
\dot\rho+3H\left(\gamma\rho+\pi\right)=0
\end{equation}
\noindent
where
\begin{equation} \label{gamma}
\gamma(t) \equiv 1 + {p(t)\over\rho(t)}
\end{equation}
\noindent is the time dependent polytropic index. Using (\ref{ZPM28}) and
(\ref{ZPM29}) we get
\begin{equation} \label{pi}
\kappa\pi=-3\gamma H^2-2\dot H
\end{equation}
When $m \gg T $, the equations of state for the nonrelativistic fluid
$1$ are:
\begin{equation}
\rho _{1} = n _{1} m + {\textstyle{3\over2}}
n _{1} T \ , \quad
p _{1} = n _{1} T
\label{ZPM36}
\end{equation}
\noindent
while for the relativistic fluid $2$ are:
\begin{equation}
\rho _{2} = 3 n _{2}T \ ,\quad
p _{2} = n _{2} T
\label{ZPM37}
\end{equation}
\noindent
For these fluids we have
\begin{equation} \label{gamma2}
\gamma=1+\frac{nT}{n_1m+\left(\frac{3}{2}n_1+3n_2\right)T}
\end{equation}
To calculate the evolution of the temperature we need also
$\Gamma$. Using (\ref{33}) and the equations of state (\ref{ZPM36}) and
(\ref{ZPM37}), we obtain
\begin{equation} \label{Gamma}
\Gamma=\frac{n_1 m}{4nT}Q
\end{equation}
\noindent where the decay rate $Q=|\Gamma_1|$ is an input to the model that
must be chosen from a fundamental microphysical theory or from
phenomenological considerations.
\section{The quasiperfect expansion}
\noindent We assume that the viscous
effects, are small. If $\tau$ is the mean
interaction time of the particles of the fluid, we have that $\nu=\left(\tau
H\right)^{-1}$ is the number of interactions in an expansion time.
Perfect fluid behavior occurs in the limit $\nu\to\infty$. Small departures from
this behavior occur for large $\nu$, and a consistent hydrodynamical description
of the fluids requires $\nu>1$. Thus we are lead to assume that $\tau H$ is
small and we propose the following "quasiperfect" expansion in powers of
$\nu^{-1}$
\begin{equation} \label{H1a}
H=H_0\left(1+\frac{h_1}{\nu} +\cdots\right)
\end{equation}
\begin{equation} \label{pi1}
\pi=\frac{\pi_1}{\nu}+\frac{\pi_2}{\nu^2}+\cdots
\end{equation}
\noindent
where $\pi_i$, will be fixed by the thermodynamical theory adopted.
Inserting (\ref{H1a}) and (\ref{pi1}) in (\ref{pi}), we find the equations that
determine the coefficients of the expansion up to first order in
$\nu^{-1}$
\begin{equation} \label{dH0}
\dot H_0+(3/2)\gamma H_0^2=0
\end{equation}
\begin{equation} \label{dH1}
\dot h_1=-\frac{\pi_1}{2H_0}
\end{equation}
\noindent
where we have assumed that $\dot\tau\ll\tau H$.
Solving these equations we get
\begin{equation} \label{H0}
H_0(t)=\frac{2}{3\int dt \gamma(t)}
\end{equation}
\noindent
and
\begin{equation} \label{H2}
H=H_0\left(1-\frac{\kappa}{2\nu}\int dt\, \frac{\pi_1}{H_0}
+\cdots\right)
\end{equation}
In the particular case we choose the truncated transport equation of Causal
Irreversible Thermodynamics for the bulk
viscosity pressure
\begin{equation}
\pi + \tau\dot{\pi} = - 3\zeta H .
\label{ZPM25}
\end{equation}
\noindent where we identify $\tau$ with the relaxation time,
the bulk viscosity coefficient $\zeta$ is given by \cite{Maar96}
\begin{equation}
\frac{\zeta}{\tau} = c_b^2 (\rho + p),
\label{ZPM26}
\end{equation}
\noindent
$c_b$ is the bulk viscous contribution to the speed
of sound $v$, $v^2=c_s^2+c_b^2\leq 1$, and
$c_s$ is the adiabatic sound speed, we find
\begin{equation} \label{pi2}
\kappa\pi\approx-\frac{9}{\nu} c_b^2 \gamma H_0^2
\end{equation}
\noindent
Also it can be easily seen that $H_0$ is an exact solution of equations
(\ref{ZPM28}), (\ref{ZPM29}), (\ref{gamma}), (\ref{ZPM25}) and (\ref{ZPM26})
provided $c_b=0$.
To get an estimation of the physical parameters in the quasiperfect regime it
is enough to keep calculations at zero order in $\nu^{-1}$ and we can neglect
the viscous terms in (\ref{drho}), (\ref{dT}) and (\ref{gamma2}). In this order of
approximation the results are independent of the form of the transport
equation. Nevertheless this approach will allow us to give a reasonable
description of the reheating process.
Then (\ref{drho}) becomes
\begin{equation} \label{drho0}
\dot \rho+3H_0 \gamma\rho\approx 0
\end{equation}
\noindent
whose solution is
\begin{equation} \label{rho0}
\rho(t)\approx \frac{\rho_0}{\left(\int dt\gamma(t)\right)^2}
\end{equation}
\noindent
where $\rho_0$ is a positive integration constant. Also (\ref{dT}) becomes
\begin{equation} \label{dT0}
\frac{\dot T}{T}\approx
\frac{2n}{3\left(n_1+2n_2\right)}
\left(\frac{n_1m}{4nT}Q-3H_0\right)
\end{equation}
\noindent
Then, when the decay rate $Q$ is large enough,
it may overcompensate the adiabatic term during the initial `preheating'
stage and make the temperature rise.
\section{Linear reheating}
Following \cite{ZPM} we assume first a linear ansatz $Q=\beta H$, with an
adimensional constant $\beta>0$.
To solve the coupled system of equations (\ref{23}), (\ref{4}),
(\ref{dT0}) and (\ref{gamma2})
for the evolution for $n$, $n_1$,
$n_2$, $T$ and $\gamma$, we use an iterative scheme that starts from a fully
nonrelativistic stage: $n_1=n$, $\gamma=1$. In this regime
\begin{equation} \label{H1}
H_0\approx \frac{2}{3t}
\end{equation}
\noindent
Then (\ref{dT0}) reduces to
\begin{equation} \label{dT1}
\frac{\dot T}{T}\simeq
\frac{4}{9t}
\left(\frac{m\beta}{4T}-3\right)
\end{equation}
\noindent
whose solution is
\begin{equation} \label{T1}
T(t)=T_1\left[1-\left(\frac{t_0}{t}\right)^{4/3}\right]
\end{equation}
\noindent where $t_0$ is an arbitrary integration constant with dimension of
time, and
$T_1=m\beta/12$. In order to describe the reheating scenario we need to choose
$t_0>0$. As in the previous inflationary stage the Universe "supercools"
\cite{Kolb90}, we assume that the inflation period ends at $t\approx t_0$ when
$T\approx 0$, and we take solution (\ref{T1}) for $t>t_0$. Then the
temperature grows monotonically and approaches asymptotically its maximum
value $T_1$, which we could call the reheating temperature of the linear
model, for $t\to\infty$.
We see from (\ref{T1}) that the temperature rises quite slowly. It takes a
time over $30 t_0$ for the temperature to approach at $1\%$ of $T_1$. For
this reason, we will not pursue further the consequences of the linear ansatz
and in a following section we will propose an improved ansatz for $Q$ which
leads a much faster rise in temperature.
\section{Quadratic reheating}
In the previous section, we have seen that a linear ansatz for $Q$ leads to a
very slow process of reheating. A microscopic interaction term like
$g\phi^2\chi^2$, suggests an effective quadratic decay rate $\Gamma\sim
\Phi^2\sim t^{-2}$. Here we look for an improved ansatz that takes into
account the scalar nature of $\Gamma$. In this direction we propose
$Q=\bar\beta R$, where $R=6\left(\dot H+2 H^2\right)$ is the curvature scalar
and $\bar\beta$ is a positive constant with dimension of time.
At zero order in $\nu^{-1}$ we obtain
\begin{equation} \label{Q2}
Q=6\bar\beta\left(2-\frac{3\gamma}{2}\right)H_0^2\equiv 6\tau_1 H_0^2
\end{equation}
\noindent
This expression shows that $Q\ge 0$ for $1\le\gamma\le 4/3$ and vanishes in a
radiation dominated era, when $\Phi$
is very small. Inserting (\ref{Q2}) in (\ref{Gamma})
and using (\ref{dT0}) we obtain at zero order in $\nu^{-1}$
\begin{equation} \label{dT2}
\frac{\dot T}{T}\simeq\frac{nH_0}{n_1+2n_2}\left(\frac{n_1m\tau_1}{nT}H_0-2\right)
\end{equation}
\noindent
and we solve it using the iterative scheme.
\subsection{Nonrelativistic regime}
In this regime when the fluid is dominated by massive particles,
equation (\ref{dT2}) becomes
\begin{equation} \label{dT3}
\dot T+2H_0 T\simeq m\tau_1 H_0^2
\end{equation}
\noindent
whose general solution is
\begin{equation} \label{T4}
T(t)=\frac{4m\tau_1}{3t}\left[1-\left(\frac{t_0}{t}\right)^{1/3}\right]
\end{equation}
\noindent As in the linear model, the reheating scenario corresponds to
$t_0>0$, and we take the solution (\ref{T4}) for $t>t_0$. This solution also
starts from $T=0$, but this time it rises violently to a maximum temperature
of reheating
\begin{equation} \label{Tr}
T_r=\frac{9\tau_1}{64 t_0}m
\end{equation}
\noindent in a period $t_r-t_0=37 t_0/27$, where $t_r$ is the time when
maximum temperature occurs. Assuming that inflation ends at $t_0=\delta/ m$,
where $\delta$ is a numeric constant, this rising period is $1.37 \delta/m$.
This is a time of the same order as the period of oscillation of the inflaton
field $2\pi/m$. After that, the temperature begins to fall. Though these
calculations are carried out in the approximation $T\ll m$, and this
implies that ( \ref{Tr} ) is strictly valid only for $\tau_1\ll t_0$, it also
suggests that $T_r$ may be of order $m$ when $\tau_1$ is large enough.
Inserting (\ref{H1}) and (\ref{Q2}) in (\ref{4}) and solving the resulting
equation for the number density of massive particles we find
\begin{equation} \label{n_1}
n_1(t)\simeq n_{10} \frac{e^{8\tau_1/\left(3t\right)}}{t^2}
\end{equation}
\noindent where $n_{10}$ is a positive integration constant, so that $n_1(t)$ is a
decreasing function.
Using (\ref{23}), (\ref{Gamma}), (\ref{T4}) and (\ref{n_1}), we find the
total number particle density in the first departure from the nonrelativistic
initial state
\begin{equation} \label{n}
n(t)\simeq\frac{n_0}{t^2}\left[1+\frac{n_{10}}{2n_0}\int dx
\frac{e^{8\tau_1/\left(3t_0 x\right)}}{x-x^{2/3}}\right]
\end{equation}
\noindent
where $x=t/t_0$, and $n_0$ is a positive constant. Assuming that the
reheating temperature is much smaller than $m$ we obtain the approximated
expression
\begin{equation} \label{n2}
n(t)\simeq \frac{n_0}{t^2}\left[1+\frac{3n_{10}}{2n_0}
\ln\left(x^{1/3}-1\right)\right]
\end{equation}
The expression (\ref{n_1}) for the density of nonrelativistic particles is a
decreasing function, while the expression (\ref{n2}) for the total number
density has a peak. The consistency requirement $n_1\leq n$ is satisfied after
the time $t_p$ when $n_1(t_p)=n(t_p)\equiv n_p$. This time is given by
\begin{equation} \label{xp}
x_p=\left[1+\exp\left(\frac{2\left(\alpha-1\right)}{3\alpha}\right)\right]^3
\end{equation}
\noindent
where $\alpha=n_{10}/n_0$, and
marks in our model the starting point of the preheating stage.
We demand the temperature
\begin{equation} \label{Tp}
T(t_p) \approx 4\left(\frac{4}{3}\right)^3 \left(x_p^{1/3}-1\right) T_r
\end{equation}
\noindent to be very low and this occurs when $x_p$ is close to $1$, so
that $\alpha\ll 1$. This condition corresponds to an explosive
production of light bosons. In effect, from
( \ref{22} ), ( \ref{n2} ) and ( \ref{n_1} ) we get\begin{equation} \label{n_2}
n_2(t)\approx \frac{3 n_{10}}{2t^2}\ln\left(
\frac{x^{1/3}-1}{x_p^{1/3}-1}\right)
\end{equation}
\noindent and just after $t_p$ this particle density grows linearly with an
exponentially large slope
\begin{equation} \label{n_22}
n_2(t)\approx \frac{n_{10} e^{2/\left(3\alpha\right)}}{2t_0^2}
\left(x-x_p\right), \qquad x-x_p\ll 1
\end{equation}
\noindent reaching a large peak density $n_p/\alpha$ in a time $x_m\approx
1+3\alpha/4$. After that, both $n$ and $n_2$ begin to fall rapidly as the
effect of expansion dominates over particle production.
In the regime of large particle production
we can neglect in (\ref{dS}) the change in $s$, so that
the entropy production density is approximately given by
\begin{equation}
S ^{a}_{;a} \approx sn\Gamma=\frac{sn_{10}\,
e^{8\tau_1/\left(3t\right)}}
{2t^2\left(t-t_0^{1/3}t^{2/3}\right)}
\label{S}
\end{equation}
\noindent
and we obtain the following lower bound for the entropy density
\begin{equation} \label{S1}
S >\frac{s n_{10}}{2t_0^2}\int_{x_p}^{x_r}
\frac{dx}{x^2\left(x-x^{2/3}\right)} \approx
-\frac{3s n_{10}}{2t_0^2}\ln\left(x_p^{1/3}-1\right)
\approx \frac{sn_0}{t_0^2}
\end{equation}
\noindent
Thus we find that the generated entropy per massive particle
$S/n(t_p)>s/\alpha$ is very high.
\subsection{Intermediate regime}
Here $\rho_1\approx\rho_2$, i.e. $ n_1 m \approx 3 n_2 T$ so that $n_2\gg n_1$
and $\gamma\approx{7\over6}$. At the microscopic level this corresponds to the
regime when the backreaction of the produced particles ceases to be
negligible. We can estimate the time $t_i$ when this regime begins from the
equation $\rho(t_i)=2\rho_1(t_i)$, using (\ref{ZPM36}), (\ref{rho0}),
(\ref{T4}) and (\ref{n_1}). Therefore $t_i$ becomes a function of $K\equiv
m\tau_1/\delta$ and the intermediate regime is reached provided $K>3\ln 2/8$.
For lower values of $K$ particle production rate is not large enough to
compensate the faster decrease of relativistic matter energy density due to
cosmic expansion. On the other hand, $K$ cannot be much larger than this lower
bound if $T_r\ll m$. For instance, we get $x_i\approx 41$ and $T_r/m\approx
0.038$ for $K=0.27$ . We
show in Fig. 1 the evolution of $\rho_1$ and $\rho_2$ for this value
of $K$, from fully nonrelativistic to intermediate regime.
In this intermediate regime holds
\begin{equation}
H_0(t) \approx \frac{4}{7t}
\label{H10}
\end{equation}
\begin{equation} \label{T10}
T(t)\approx T_0^{(i)}\frac{e^{-24\tau_1/\left(49 t\right)}}{t^{4/7}}
\end{equation}
\begin{equation} \label{n_110}
n_1(t)\approx n_{10}^{\left(i\right)}
\frac{e^{96\tau_1/\left(49 t\right)}}{t^{12/7}}
\end{equation}
\begin{equation} \label{n10}
n(t)\approx n_0^{\left(i\right)}
\frac{e^{-72\tau_1/\left(49 t\right)}}{t^{12/7}}
\end{equation}
\noindent where $T_0^{(i)}$, $n_{10}^{(i)}$ and $n_0^{(i)}$ are positive
integration constants fixed by matching at $t_i$ expressions (\ref{T10}),
(\ref{n_110}) and (\ref{n10}) with their nonrelativistic counterparts
(\ref{T4}), (\ref{n_1}) and (\ref{n}). After the initial stage of large
ultrarelativistic particle creation in the nonrelativistic regime the
nonrelativistic particle decay process slows down. All particle densities and
the temperature decrease, though the dilution and cooling rates are smaller
because of the slower cosmic expansion. The whole evolution of the
temperature and particle number densities along the nonrelativistic and
intermediate regimes is plotted in Figs. 2-4 for $K=0.27$.
\subsection{Ultra-relativistic regime}
In this last stage of reheating, the energy density
becomes dominated by
the radiation fluid, i.e. $n_2T \gg n_1m$,
$\gamma \approx {4\over3}$.
Therefore we have
\begin{equation}
H_0 \approx \frac{1}{2t}
\label{H20}
\end{equation}
\begin{equation} \label{T20}
T(t)\approx\frac{T_0^{\left(u\right)}}{t^{1/2}}
\end{equation}
\begin{equation} \label{n20}
n(t)\approx\frac{n_0^{\left(u\right)}}{t^{3/2}},
\qquad n_1(t)\approx\frac{n_{10}^{\left(u\right)}}{t^{3/2}}
\end{equation}
\noindent where $T_0^{(u)}$, $n_{10}^{(u)}$ and $n_0^{(u)}$ are positive
integration constants. This stage marks the end of reheating and the cosmic
medium approaches a perfect, relativistic fluid with vanishing viscosity and
particle production. The temperature and particle densities continue falling
at about the adiabatic rate. This is smaller than the intermediate stage rate
as the radiation-dominated universe expands even slower. A small remnant of
nonrelativistic particles may remain.
\section{Conclusions}
With the aim of understanding the process of reheating we have developed a
phenomenological model of two interacting fluids. In this model the
oscillating inflaton field is modeled after a nonrelativistic fluid of massive
particles that decay into an ultrarelativistic fluid of massless particles.
We have carried out a ``quasiperfect'' perturbative expansion to solve the
Einstein equations. This corresponds to a small viscous pressure regime. To
the lowest order in this expansion, the thermodynamical magnitudes do not
depend on the transport equation for the bulk viscosity.
A particle decay rate, linear in the expansion rate, does not yield a suitable
model as the maximum reheating temperature is reached only for very long
times. However an alternative quadratic ansatz for the decay rate gives a good
behavior of the thermodynamical magnitudes. The last one was investigated in
detail along its evolution from the initial nonrelativistic stage to the final
domination by ultrarelativistic particles. The equilibrium temperature of the
whole system and the produced particle number density rise violently in a time
comparable to the oscillation period of the inflaton field. Also, large
amounts of entropy are produced in this initial stage of reheating. The time
when the energy density of the decay products becomes comparable with the
nonrelativistic particles, is similar to the microscopic calculations for the
time when backreaction effects become important.
In a future paper we will investigate phenomenological reheating models far
from the perfect fluid regime.
\noindent
{\bf Acknowledgements}\\
This work has been partially supported by the Universidad de Buenos Aires
under Grant TX-93.
|
\section{Introduction}
\label{sect:intro}
The flux of atmospheric neutrinos and muons at very high energies, above 1
TeV, passes from being originated in the decays of pions and kaons to being
predominantly generated in semileptonic decays of charmed particles (see for
example \cite{g1}). This flux is of importance for large area detectors of
high energy cosmic neutrinos. Future ${\rm {km}^{3}}$ arrays would be able
to observe muons and neutrinos with energies that may reach $10^{12}$ GeV.
Atmospheric muons and neutrinos would be one of the most important
backgrounds, limiting the sensitivity of any ``neutrino telescope'' to
astrophysical signals. Besides, they might be used for detector calibration
and perhaps, more interestingly, be exploited to do physics, e.g. study
neutrino masses.
Present experimental attempts to detect atmospheric muons from charm are
spoiled by systematic errors. Theoretical predictions depend strongly on the
reliability of the model adopted for charm production and decay and differ
by orders of magnitude, due to the necessity of extrapolating present
accelerator data on open charm production in fixed target experiments, at
laboratory energies of about 200 GeV, to the larger energies needed for
atmospheric neutrinos, from 10$^{3}$ to ${10^{8}}$ GeV (at about $10^{8}$
GeV the rates become too small for a ${\rm {km^{3}}}$ detector). These
energies, from 40 GeV to 14 TeV in the center of mass, are comparable to the
energies of the future RHIC at Brookhaven, 200 GeV, and LHC at CERN, 7 TeV.
The theoretically preferred model, perturbative QCD (pQCD), was thought to
be inadequate because it could not account for several aspects of some of
the early data on open charm production (in conflict with each other, on the
other hand \cite{kvd}), and because of a sensitivity of the leading-order
(LO) calculation, the only existing until recently, to the charm quark mass,
to the low partonic momentum fraction, $x$, behavior of the parton
distributions and to higher order corrections. So, even if some now-obsolete
pQCD calculations have appeared \cite{ik,zhv}, the models for charm
production traditionally favored in studies of atmospheric fluxes have been
non-perturbative: for example, besides semi-empirical parametrizations of
the cross section, the quark-gluon string model (QGSM, a.k.a. dual parton
model), based on Regge asymptotics, and the recombination quark-parton model
(RQPM), incorporating the assumption of an intrinsic charm component in the
nucleon (see \cite{br}).
Today, however, pQCD predictions and experimental data are known to be
compatible \cite{a, alves1, alves2, ada, fmnr}: charm production experiments
form a consistent set of data, and the inclusion of next-to-leading order
(NLO) terms has been a major improvement over the leading-order treatment.
Quoting from Appel \cite{a}, ``the success of these calculations has removed
the impetus to look for unconventional sources of charm production beyond
the basic QCD".
A study based on pQCD was therefore performed in Ref. \cite{TIG} (called TIG
from now on). CLEO and HERA results were incorporated, but for simplicity
the LO charm production cross section was adopted, multiplied by a constant $
K$ factor of 2 to bring it in line with the next-to-leading order values,
and supplemented by parton shower evolution and hadronization according to
the Lund model. The neutrino and muon fluxes from charm were found to be
lower than the lowest previous prediction, namely a factor of 20 below the
RQPM \cite{bnsz}, of 5 below the QGSM \cite{c, gghv}, and of 3 below the
lowest curve in Ref. \cite{zhv}.
Here we use the same treatment of TIG, except for the very important
difference of using the actual next-to-leading order pQCD calculations of
Mangano, Nason and Ridolfi \cite{MNR} (called MNR from now on), as contained
in the program we obtained from them (see also \cite{nde}), to compute the
charm production cross sections. These are the same calculations used
currently to compare pQCD predictions with experimental data in accelerator
experiments. The main goal of this paper is to compare the fluxes obtained
with the NLO and with the LO, i.e. we will compute the $K$ factor for the
neutrino and muon fluxes. This $K$ factor is necessarily different from the $
K$ factor for charm production (which can be found in the literature),
because only the forward going leptons contribute significantly to the
atmospheric fluxes.
A similar comparison was very recently made in \cite{PRS}, using the
approximate analytical solutions introduced by TIG to the cascade equations in
the atmosphere. We make instead a full simulation of the cascades, using the
combined MNR and PYTHIA programs. These two treatments of the problem are
complementary. For comparison, we include results obtained with the CTEQ 3M
gluon structure function used in Ref.~\cite{PRS}. We find our CTEQ 3M results
to be close to those of the PRS study, in spite of the very different
approaches used in the two calculations.
Addressing right away a concern that has been expressed to us several times,
about the applicability of perturbative QCD calculations, mostly done for
accelerator physics, to the different kinematic domain of cosmic rays, we would
like to point out that, since the characteristic charm momentum in our
simulations is of the order of the charm mass, $k\simeq O(m_{c})$, we do not
have here
the uncertainty present in the differential cross sections \cite{MNR}, when $
k_{T}$ is much larger than $m_{c}$ (as is the case in accelerators), due to
the presence of large logarithms of $(k_{T}^{2}+m_{c}^{2})/m_{c}^{2}$.
Depending on the steepness of the gluon structure function we take, we do
have, however, large logarithms, known as ``ln(1/x)'' terms, where $x\simeq
\sqrt{4m_{c}^{2}/s}$ ($s$ is the hadronic center of mass energy squared) is
the average value of the hadron energy fraction needed to produce the $c\bar{
c}$ pair. These should not be important for steep enough gluon structure
functions (namely for values of $\lambda$ in Eq. (9) not very close to
zero), but we have not made any attempt to deal with this issue.
In the next section of this paper we explain our normalization of the NLO
charm production cross section in the MNR program. In Sect.~\ref
{sect:simulation} we describe the computer simulations used to calculate the
neutrino and muon fluxes. In Sect.~\ref{sect:fluxes} we show the results of
our simulations, we discuss the differences between a NLO and a LO approach
and we make a comparison with the fluxes of the TIG model.
In this paper we consider only vertical showers for simplicity (the same was
done by TIG).
\section{Charm production in perturbative QCD}
\label{sect:NLO}
In this section, we show evidence that perturbative QCD gives a fair
description of the present accelerator data on open charm production in the
kinematic region most important for cosmic ray collisions in the atmosphere.
There are still not many experiments on open charm production with good
enough statistics, despite the recent improvements, but many are expected in
the near future.
We use a NLO approach which is based on the MNR calculation, for which we
have obtained the computer code. The NLO cross section for charm production
depends on the choice of the parton distribution functions (PDFs) and on
three parameters: the charm quark mass $m_{c}$, the renormalization scale $
\mu_{R}$, and the factorization scale $\mu_{F}$.
\subsection{Choice of $m_{c}, \protect\mu_{R}, \protect\mu_{F}$}
MNR have two default choices of $m_{c}$, $\mu_{R}$ and $\mu_{F}$: for total
cross sections they choose $m_{c}$ = 1.5 GeV, $\mu_{R}= m_{c}$, $\mu_{F} = 2
m_{c}$; for differential cross sections they choose instead $m_{c}$ = 1.5
GeV, $\mu_{R}= m_{T}$, $\mu_{F} = 2 m_{T}$, where $m_{T} = \sqrt{k_{T}^{2} +
m_{c}^{2}}$ is the transverse mass.
The current procedure to reproduce the measured differential cross sections
\cite{alves2, ada, fmnr} is to use the MNR default choices for these three
parameters and multiply the result by the global factor of about 2 or 3
necessary to match the predicted and measured total inclusive cross
sections. Although this procedure might be acceptable in face of the
uncertainties in the pQCD predictions, we find it unsatisfactory from a
theoretical point of view. We prefer to fit the differential and total cross
sections with one and the same combination of $m_{c}$, $\mu_{R}$, and $
\mu_{F}$.
We make separate fits of $m_{c}$, $\mu_{R}$, and $\mu_{F}$ for each of the
following sets of PDFs: MRS R1, MRS R2 \cite{MRS1}, CTEQ 3M \cite{CTEQ3} and
CTEQ 4M \cite{CTEQ} (see the next subsection for details).
We are aware that several choices of $m_{c}$, $\mu_{R}$ and $\mu_{F}$ may
work equally well. In fact the cross sections increase by decreasing $\mu_{F}
$, $\mu_{R}$ or $m_{c}$, so changes in the three variables can be played
against each other to obtain practically the same results. We present here
just one such choice.
We choose $\mu_{R} = m_T$, $\mu_{F} = 2 m_T$ for all sets, and
\begin{eqnarray}
m_{c}&=&1.185{\rm ~GeV} \quad\hbox{for MRS R1,} \label{eq:1} \\
m_{c}&=&1.31{\rm ~GeV} \quad\hbox{for MRS R2,} \label{eq:1b} \\
m_{c}&=&1.24{\rm ~GeV} \quad\hbox{for CTEQ 3M,} \label{eq:1c} \\
m_{c}&=&1.27{\rm ~GeV} \quad\hbox{for CTEQ 4M.} \label{eq:1d}
\end{eqnarray}
We fit $m_{c}$, $\mu_{R}$, and $\mu_{F}$ to the latest available data on
charm production \cite{alves1, alves2, ada, fmnr} in proton-nucleon and
pion-nucleon collisions. We use mainly the data on $pN$ collisions, which
are more relevant to us, but examine also the $\pi N$ data to see how well
our choice of parameters works there.
The MNR program calculates the total cross section for $c\bar c$ pair
production, $\sigma_{c \bar{c}}$. We converted the experimental data on $
D^{+}$ or $D^{-}$ production $\sigma(D^{+},D^{-})$, $D^{0}$ or ${\bar{D}}^{0}
$ production $\sigma (D^{0},{\bar{D}}^{0})$, or the same cross sections just
for $x_{F} > 0$ ($x_{F}$ is the Feynman $x$), $\sigma_{+} (D^{+},D^{-})$ and
$\sigma_{+}(D^{0},\bar D^{0})$, into $\sigma_{c\bar c}$ values following
\cite{fmnr}.
The data we used for the `calibration' of the MNR program are shown in
Table~\ref{table:1} and Table~\ref{table:2} \cite{alves1, alves2, ada, fmnr}.
These tables also present a comparison of experimental data on total inclusive
D-production cross sections (converted to $\sigma_{c\bar c}$ total cross
sections) with those calculated with the MNR program.
For the data of Table~\ref{table:1}, for $pN$ collisions, the conversion is
done using
\begin{equation}
\sigma_{c\bar{c}}=1.5\times{\frac{1}{2}}\times\lbrack{\sigma(D^{+},D^{-})+
\sigma(D^{0},\bar{D}^{0})]} \label{eq:2}
\end{equation}
if cross sections are measured for any $x_{F}$, or
\begin{equation}
\sigma_{c\bar{c}}=1.5\times2\times{\frac{1}{2}}~~[{\sigma_{+}(D^{+},D^{-})+
\sigma_{+}(D^{0},\bar{D}^{0})]}~, \label{eq:3}
\end{equation}
if experimental data are given for $x_{F}>0$ only. The explanation of the
factors in Eqs. (\ref{eq:2}),(\ref{eq:3}) is as follows. The $\frac{1}{2}$
factors convert single $D$ inclusive into ${D\bar{D}}$ pair inclusive cross
sections. The 1.5 factors are required to take into account the production
of $D_{S}$ and $\Lambda_{c}$ (which is included in $\sigma_{c\bar{c}})$
through the ratios \cite{fmnr}
\begin{equation}
{\frac{\sigma(D_{S})}{{\sigma(D^{+},D^{0})}}}\simeq0.2,\qquad{\frac
{\sigma(\Lambda_{c})}{{\sigma(D^{+},D^{0})}}}\simeq0.3, \label{eq:4}
\end{equation}
(the same relation also for antiparticles). The factor 2 in Eq. (\ref{eq:3})
converts from $x_{F}>0$ to all $x_{F}$ (i.e. it is $\sigma_{c\bar{c}
}/\sigma_{c\bar{c}}(x_{F}>0)$ for the $pN$ case).
In the case of $\pi N$ collisions (Table~\ref{table:2}) the factor 2 in
equation (\ref{eq:3}) is replaced by 1.6, which is the value of $
\sigma_{c\bar c}/\sigma_{c\bar c}(x_{F} > 0)$ when a pion beam is used.
Table~\ref{table:1} explains our choice of $m_{c}$ values. The $m_{c}$
values in Eqs.(\ref{eq:1}),(\ref{eq:1b}),(\ref{eq:1c}) and (\ref{eq:1d})
reproduce well the central values of the $pN$ charm inclusive total cross
sections \cite{alves1}, using the program with the four different PDFs.
In Table~\ref{table:2} we also present a similar analysis for $\pi N$
collisions, using only MRS R1 for simplicity. In this case slightly higher
values of $m_{c}$ fit the $\pi N$ data \cite{alves1, fmnr} a bit better,
while $m_{c} = 1.185$ GeV, the value we take with the MRS R1 PDF, fits the $
pN$ data \cite{alves1, alves2, fmnr} a bit better. Notice that for the pions
we used a different PDF, SMR2 \cite{harri}, the same used in Refs. \cite
{alves1, alves2} (obviously not used in our calculations of atmospheric
fluxes). We present the $\pi N$ data just for completeness, to show that
they too are reasonably well fitted with our choice of parameters. These
other values of $m_{c}$ in Table~\ref{table:2} well reproduce the $\pi^{\pm}N
$ data at 250 GeV \cite{alves1} and the $\pi^{-}N$ data at 350 GeV \cite{ada}
(which seem a bit too low with respect to the data at 250 GeV). Even if each
value of $m_{c}$ reproduces best each total cross section, all three provide
reasonable fits to all data, as can be seen also in the Figs.~1--3.
In Figs.~1--3 we present total and differential cross sections calculated
with the MNR program and compared to the experimental data. As a way of
example, we describe our fits for MRS R1 only.
Fig.~1a shows the fit to $pN$ total cross sections (converted into $
\sigma_{c\bar c}$ values as described above). In addition to the
experimental value of Table~\ref{table:1} --- which is the fundamental one,
since it's the experiment whose differential cross sections we want also to
fit --- we added other experimental points coming from previous experiments
(for details see \cite{fmnr}). For $pN$ the $m_{c}= 1.185$ GeV is the best
choice.
Fig.~1b shows the same for $\pi N$ collisions. Here, as explained before,
values of $m_{c}= 1.25$ GeV or $m_{c}= 1.31$ GeV are a better choice. Again
we added here for completeness other experimental points coming from
previous experiments \cite{fmnr}.
Fig.~2ab shows fits to D-inclusive differential cross sections. In this
figure the theoretically obtained $d\sigma_{c\bar c}/dx_{F}$ and $
d\sigma_{c\bar c}/dp_{T}^{2}$ were converted into D-cross sections, with no
extra factors. Fig.~2ab presents the data of the E769 collaboration \cite
{alves2} for $pN$ and $\pi N$ at 250 GeV. In these cases the differential $
\sigma_{c\bar c}$ cross sections are converted into single inclusive ones
(by a factor of 2) and then into cross sections for production of $D^{\pm},
D^{0}, \bar D^{0}$ and $D_{S}^{\pm}$ (by a factor of 1.2/1.5, see Eq.~(\ref
{eq:4})) for the E769 data. For example,
\begin{equation}
{\frac{d\sigma}{dx_{F}}}(D^{\pm}, D^{0},\bar{D^{0}}, D_{S}^{\pm}) ~\simeq{
\frac{1.2}{1.5}} \times2 \times{\frac{d\sigma_{c\bar c}}{dx_{F}}}
\label{eq:5}
\end{equation}
for Fig.~2a (and similar factors for $d\sigma/dp_{T}^{2}$ for Fig.~2b). The
fit to the $d\sigma/dp_{T}^{2}~~ pN$ data in Fig.~2b seems to be a bit too
low, but it is not very different from the fit shown in Fig.~2 of reference
\cite{alves2}. The predicted $d\sigma/dp_{T}^{2}$ are not sensitive to
differences in $m_{c}$ that are instead more noticeable in $d\sigma/dx_{F}$.
Fig.~3ab presents the $\pi N$ data at 350 GeV of the WA92 collaboration \cite
{ada} in a way similar to Fig.~2ab. In these cases the differential $
\sigma_{c\bar c}$ cross sections are converted into a single inclusive ones
(by a factor of 2) and then into cross sections for production of $D^{\pm}$,
$D^{0}$ and $\bar D^{0}$ only (by a factor of 1.0/1.5, see Eq.~(\ref{eq:4}))
for the WA92 data. Similar conclusions can be drawn: for pions $m_{c}= 1.31$
GeV is the best choice in this case.
We have performed the same analysis with MRS R2, CTEQ 4M and CTEQ 3M, even
if we do not show here any of the fits. The results for total and
differential cross sections were similar to those shown for the MRS R1, the
only difference being the choice of $m_{c}$.
In conclusion, we obtain good fits to all data on charm production with one
choice of $\mu_{R},\mu_{F}$ and $m_{c}$ for each PDF, without other
normalizations.
\subsection{Choice of PDFs}
Consider the collision of a cosmic ray nucleus of energy $E$ per nucleon,
with a nucleus of the atmosphere in which charm quarks of energy $E_{c}$ are
produced, which decay into leptons of energy $E_{l}$ (in the lab. frame,
namely the atmosphere rest frame). Due to the steep decrease with increasing
energy of the incoming flux of cosmic rays, only the most energetic charm
quarks produced count for the final lepton flux, and these $c$ quarks come
from the interactions of projectile partons carrying a large fraction of the
incoming nucleon momentum. Thus, the characteristic $x$ of the projectile
parton, that we call $x_{1}$, is large. It is $x_{1}\simeq O(10^{-1})$. We
can, then, immediately understand that very small parton momentum fractions
are needed in our calculation, because typical partonic center of mass
energies $\sqrt{\hat{s}}$ are close to the $c\bar{c}$ threshold, $
2m_{c}\simeq 2$ GeV (since the differential cross section decreases with
increasing ${\hat{s}}$), while the total center of mass energy squared is $
s=2m_{N}E$ (with $m_{N}$ the nucleon mass, $m_{N}\simeq 1$ GeV). Calling $
x_{2}$ the momentum fraction of the target parton (in the nuclei of the
atmosphere), then, $x_{1}x_{2}\equiv \hat{s}/s=4m_{c}^{2}/(2m_{N}E)\simeq $
GeV/$E$. Thus, $x_{2}\simeq O$(GeV/0.1 $E$), where $E$ is the energy per
nucleon of the incoming cosmic ray in the lab. frame. The characteristic
energy $E_{c}$ of the charm quark and the dominant leptonic energy $E_{l}$
in the fluxes are $E_{l}\simeq E_{c}\simeq 0.1E$, thus $x_{2}\simeq O$(GeV/ $
E_{l}$), as mentioned above.
For $x>10^{-5}$ ($E\lesssim10^3$ TeV), PDFs are available from global
analyses of existing data. We use four sets of PDFs. MRS R1, MRS R2 \cite
{MRS1} and CTEQ 4M \cite{CTEQ}, incorporate most of the latest HERA data and
cover the range of parton momentum fractions $x\geq10^{-5}$ and momentum
transfers $Q^{2}\geq1.25-2.56$ GeV$^{2}$. MRS R1 and MRS R2 differ only in
the value of the strong coupling constant $\alpha_{s}$ at the Z boson mass:
in MRS R1 $\alpha_{s}(M_{Z}^{2})=0.113$, and in MRS R2 $
\alpha_{s}(M_{Z}^{2})=0.120$. The former value is suggested by ``deep
inelastic scattering'' experiments, and the latter by LEP measurements. This
difference leads to different values of the PDF parameters at the reference
momentum $Q_{0}^{2}=1.25\ {\rm GeV}^{2}$ where the QCD evolution of the MRS
R1 and R2 PDFs is started. The CTEQ 4M is the standard choice in the $
\overline{MS}$ scheme in the most recent group of PDFs from the CTEQ group ($
\alpha_{s}(M_{Z}^{2})=0.116$ for CTEQ 4M). We also use an older PDF by the
CTEQ group, namely the CTEQ 3M \cite{CTEQ3}, only for comparisons with \cite
{PRS}, where it is used as the main PDF.
For $x<10^{-5}$ ($E\gsim 10^3$ TeV), we need to extrapolate the available
PDFs. For $x\ll1$, all these PDFs go as
\begin{equation}
xf_{i}(x,Q^{2})\simeq A_{i}x^{-\lambda_{i}(Q^{2})}, \label{eq:7}
\end{equation}
where $i$ denotes valence quarks $u_{v},d_{v}$, sea quarks $S$, or gluons $g$
. The PDFs we used (except the older CTEQ 3M) have $\lambda_{S}(Q_{0}^{2})
\not =\lambda_{g}(Q_{0}^{2})$, in contrast to older sets of PDFs which
assumed an equality. As $x$ decreases the density of gluons grows rapidly.
At $x\simeq0.3$ it is comparable to the quark densities but, as $x$
decreases it increasingly dominates over the quark densities, which become
negligible at $x\lsim 10^{-3}$.
We need, therefore, to extrapolate the gluon PDFs to $x<10^{-5}$.
Extrapolations based on Regge analysis usually propose $xg(x)\sim
x^{-\lambda }$ with $\lambda \simeq 0.08$ \cite{lowx}, while evolution
equations used to resum the large logarithms $\alpha _{s}\ln (1/x)$
mentioned above, such as the BFKL (Balitsky, Fadin, Kuraev, Lipatov \cite
{BFKL}) find also $xg(x)\sim x^{-\lambda }$ but with $\lambda \simeq 0.5\ $
\cite{lowx}. A detailed analysis of the dependence of the neutrino fluxes on
the low $x$ behavior of the PDFs will be given in another publication \cite
{GGVlambda}. As mentioned above, in the present paper our goal is to compare
NLO to BORN simulations, for which we use a simplified extrapolation at low $
x$ of the gluon PDF, which is somewhat in between the two extreme
theoretical behaviors described above. For MRS R1-R2 and CTEQ 4M we take a
linear extrapolation of $\ln g(x)$ as a function of $\ln x$, in which we
took $\ln g(x)=-(\lambda _{g}(Q^{2})+1)\ln x+\ln A_{g}$, where $\lambda
_{g}(Q^{2})$ was taken as its value at $x=10^{-5}$, the smallest $x$ for
which the PDFs are provided; for the CTEQ 3M we used a polynomial
approximation which is included in the PDF package.
\section{Simulation of particle cascades in the atmosphere}
\label{sect:simulation}
We simulate the charm production process in the atmosphere and the subsequent
particle
cascades, by modifying and combining together two different programs: the
MNR routines \cite{MNR} and PYTHIA 6.115 \cite{jetset}.
The MNR program was modified to become an event generator for charm
production at different heights in the atmosphere and for different energies
of the incoming primary cosmic rays.
The charm quarks (and antiquarks) generated by this first stage of the
program are then fed into a second part which handles quark showering,
fragmentation and the interactions and decays of the particles down to the
final leptons. The cascade evolution is therefore followed throughout the
atmosphere: the muon and neutrino fluxes at sea level are the final output
of the process.
In this section we give a brief description of the main parts of the
simulation. Even if our program is completely different from the one used by
TIG, because it is constructed around the MNR main routines, nevertheless we
keep the same modeling of the atmosphere and of the primary cosmic ray flux
as in TIG and the same treatment of particle interactions and decays in the
cascade.
Our main improvement is the inclusion of a true NLO contribution for charm
production (and updated PDFs), so we keep all other assumptions of the TIG
model in order to make our results comparable to those of TIG. We study the
effect of modifying some of their other assumptions elsewhere \cite{GGVlambda}.
\subsection{The model for the atmosphere}
We assume a simple isothermal model for the atmosphere. Its density at
vertical height $h$ is
\begin{equation}
\rho(h)={\frac{X_{0}}{h_{0}}}\,e^{-h/h_{0}}, \label{eq:8}
\end{equation}
where the scale height $h_{0}=6.4\ {\rm km}$ and the column density $
X_{0}=1300\ {\rm g/cm^{2}}$ at $h=0$ are chosen as in TIG, to fit the actual
density in the range $3\ {\rm km}<h<40\ {\rm km,}$ important for cosmic ray
interactions. Along the vertical direction, the amount of atmosphere traversed
by a particle, the depth $X$, is related to the height $h$
simply by
\begin{equation}
X=\int_{h}^{\infty}\rho(h^{\prime})dh^{\prime}=X_{0}e^{-h/h_{0}}.
\label{eq:9}
\end{equation}
The atmospheric composition at the important heights is approximately
constant: 78.4\% nitrogen, 21.1\% oxygen and 0.5\% argon with average atomic
number $\langle A\rangle$ = 14.5.
\subsection{The primary cosmic ray flux}
Following TIG \cite{TIG}, we neglect the detailed cosmic ray composition and
consider all primaries to be nucleons with energy spectrum
\begin{equation}
\phi_N (E, 0) \left [\rm{nucleons}\over {cm^2~ s ~sr ~GeV~/A} \right ]
=
\begin{cases}
{1.7 (E/{\rm GeV})^{-2.7} & for $ E < 5~10^6$ GeV\cr
174 (E/{\rm GeV})^{-3} & for $ E > 5~10^6$ GeV \cr}
\end{cases}
\label{eq:10}
\end{equation}
The primary flux is attenuated as it penetrates into the atmosphere by
collisions against the air nuclei. An approximate expression for the
intensity of the primary flux at a depth $X$ is (see \cite{TIG} again)
\begin{equation}
\phi_{N}(E,X)=e^{-X/\Lambda_{N}(E)}~\phi_{N}(E,0)~. \label{eq:11}
\end{equation}
The nuclear attenuation length $\Lambda_{N}$, defined as
\begin{equation}
\Lambda_{N}(E)={\frac{\lambda_{N}(E)}{1-Z_{NN}(E)}}~, \label{eq:12}
\end{equation}
has a mild energy dependence through $Z_{NN}$ and $\lambda_{N}$. Here $Z_{NN}$
is the spectrum-weighted moment for nucleon regeneration in nucleon-nucleon
collisions, for which we use the values in Fig.~4 of Ref. \cite{TIG}. And
$\lambda_{N}$ is the interaction thickness
\begin{equation}
\lambda_{N}(E,h)={\frac{\rho(h)}{\sum_{A}\sigma_{NA}(E)n_{A}(h)}}~,
\label{eq:13}
\end{equation}
where $n_{A}(h)$ is the number density of air nuclei of atomic weight $A$ at
height $h$ and $\sigma_{NA}(E)$ is the total inelastic cross section for
collisions of a nucleon $N$ with a nucleus $A$.\footnote{We recall that the
elastic cross section contributes negligibly to the primary flux attenuation
because the average elastic energy loss is very small, less than 1 GeV at the
high energies we consider. This can be seen using the differential elastic
cross section $d\sigma_{el}/dQ^{2}=(d \sigma_{el}/dQ^{2})_{Q^{2}=0}\exp
(-bQ^{2})$ with $b=[7.9+0.9\ln p_{lab}]{\rm GeV}^{-2}$, with $p_{lab}$ in GeV
\cite{goulianos}. Here $Q$ is the momentum transfer of the colliding proton
of incoming momentum $p_{lab}$ and mass $M$. The mean energy loss is the mean
value of $Q^{2}/2M$ (here $M$ is the target proton mass) namely
$(1/2Mb)=67{\rm MeV}/(1+0.1\ln(p_{lab}/{\rm GeV}))$. This is 46 MeV at
$E=100{\rm GeV}$, and smaller at higher energies.} This cross section scales
essentially as $A^{2/3}$, since for the large nucleon-nucleon cross sections we
deal with, the projectiles do not penetrate the nucleus. So we set
$\sigma_{NA}(E)=A^{2/3}\sigma_{NN}(E)$. For $\sigma_{NN}(E)$ we use the fit to
the available data in Ref. \cite{RPP96}. Using our height independent
atmospheric composition, we simplify Eq.~(\ref{eq:13}) as follows,
\begin{equation}
\lambda_{N}(E,h)={\frac{\langle A\rangle}{\langle A^{2/3}\rangle}}\,{\frac{
{\rm u}}{\sigma_{NN}(E)}}=2.44\,{\frac{{\rm u}}{\sigma_{NN}(E)}}~.
\label{eq:14}
\end{equation}
Here $\langle~\rangle$ denotes average and u is the atomic mass unit, that
we write as
\begin{equation}
{\rm u}=1660.54{\rm ~mb~g/cm^{2}}. \label{eq:14b}
\end{equation}
We therefore find that in our approximations $\lambda_{N}(E)$ is independent
of height.
\subsection{Charm production with MNR routines}
As we remarked before, the modified MNR routines are the first stage of our
simulation. For a given energy $E$ of a primary incoming proton in the lab
system, i.e.\ in the atmosphere reference frame, we generate a collision with a
nuclear target at rest in the atmosphere, activating the MNR routines (primary
event, $pN$ collision, with $N=(p+n)/2$) .
These routines generate total and differential cross sections through a
VEGAS integration, which creates a large number of `subevents', each one
with a particular weight, which in the original MNR program are summed
together to calculate the final cross sections.
It is easy to modify the program so that each of these subevents (together with
its weight) can represent the production of a charm $c$ (or of a $c \bar{c}$
pair, or $c \bar{c}$ gluon, etc.) with given kinematics in any particular
reference frame of interest. The original MNR routines can calculate single
differential cross sections, in which the kinematics of only one final $c$
quark is available, and double differential cross sections, in which the full
kinematics of the $c \bar{c}$ pair (plus an additional parton in NLO processes)
becomes available, for each subprocess. We have used both these possibilities.
We will refer to them as `single' and `double' modes. The `single' is the mode
we use to obtain all our results. We use the `double' mode only to compare the
results of the independent fragmentation model used in the evolution of
cascades in the `single' mode, with the more reliable string fragmentation
model, which can only be used in the `double' mode, as we explain below.
The MNR program \cite{MNR, nde} contains all BORN and NLO processes. In the
`single' mode we can generate the following processes, with only the kinematics
of the $c$ quark available,
\begin{equation}
gg\rightarrow cX;\;\;q\bar{q}\rightarrow cX\;({\rm BORN})\quad gg\rightarrow
cX;\;\;q\bar{q}\rightarrow cX;\;\;qg\rightarrow cX\;({\rm NLO}) \label{eq:15}
\end{equation}
where q represents any light quark or antiquark. In the `double' mode we have the following
processes
\begin{equation}
gg\rightarrow c\bar{c};\;\;q\bar{q}\rightarrow c\bar{c}\;({\rm BORN})\quad
gg\rightarrow c\bar{c}g;\;\;q\bar{q}\rightarrow c\bar{c}g;\;\;qg\rightarrow c
\bar{c}q\;({\rm NLO}) \label{eq:16}
\end{equation}
for which the kinematics of all the outgoing partons is fully determined for
each `subevent'.
All the kinematical variables of the partons in the final state constitute
the input for the next stage of the program, described in the next
subsection.
An important characteristic of the first stage is that, besides $m_c$, $\mu_R$,
and $\mu_F$, we can select any desired PDF to be used with the charm production
routines. We have updated the set of PDFs in the original MNR program.
According to the discussion of Sect.~\ref{sect:NLO}, we use the MRS R1, MRS R2,
CTEQ 3M and CTEQ 4M parton distribution functions, together with the values of
$m_c$, $\mu_R$, and $\mu_F$ in Eqs.~(\ref{eq:1}--\ref{eq:1d}).
As a concrete example of the integrals performed in our program, here we
write the differential flux $\phi_{\mu}$ of muons (namely of $\mu^{+}+\mu^{-}
$) with energy $E_{\mu}$ \ ($\mu$ stands here for $\mu^{+}$ or $\mu ^{-}$)
in the `single' mode ($\phi_{\mu}$ has units cm$^{-2}$ s$^{-1}$ sr$^{-1}$ GeV
$^{-1}$)
\begin{eqnarray}
\phi_\mu(E_\mu) & = &\int_{E_\mu}^{\infty} dE \int_0^{\infty} dh\
\phi_N(E,X(h)) \sum_A n_A(h) \times \nonumber \\ &&\int_{E_\mu}^{E} dE_c
\left [{d \sigma(p A \to c Y;E,E_c) \over d E_c} \right]_{MNR} \left [
{dN_\mu (c \to \mu; E_c, E_\mu, h) \over d E_{\mu} } \right]_{PYTHIA}
\nonumber \\ &+& (c \to \bar{c}).
\label{eq:flux1}
\end{eqnarray}
Here $n_{A}(h)$ is the number density of nuclei of atomic number A in the
atmosphere, $E$ is the energy of the primary cosmic ray proton, $E_{c}$ the
energy of the charm produced in the collision $pA\rightarrow cY$ \ ($Y$ here
stands for anything else). Using the relation ${d\sigma(pA\rightarrow
cY)/dE_{c}}=A\ {d\sigma(pN\rightarrow cY)/dE_{c}}$, the sum over $A$ becomes
$\sum_{A}n_{A}(h)A=\rho(h)/u$. Using $dX=-\rho(h)dh$, Eq. (\ref{eq:11}), and
normalizing to one the distribution in depth $X$, $\phi_{\mu}$ becomes
\begin{equation}
\phi_{\mu}(E_{\mu})=\int_{E_{\mu}}^{\infty}dE\int_{X_{0}}^{\infty}dX\ \phi
_{N}(E,X\mathord{=}0)\ {\frac{e^{{-X/\Lambda}_{N}{(E)}}}{\Lambda_{N}(E)}\ }
\left[ {\frac{f(h)\Lambda_{N}(E)}{u}}\right] , \label{eq:flux2}
\end{equation}
where, from Eqs.(\ref{eq:12}) and (\ref{eq:14}), $\Lambda_{N}/u=2.44[\sigma
_{NN}(1-Z_{NN})]^{-1}$ and
\begin{equation}
f(h)=2\int_{E_{\mu}}^{E}dE_{c}\left[{\frac{d\sigma(pN\rightarrow cY;E,E_{c})
}{dE_{c}}}\right]_{MNR}\ \left[\frac{dN_{\mu}(c\rightarrow
\mu;E_{c},E_{\mu},h)}{dE_{\mu}}\right]_{PYTHIA}
\label{eq:flux3}
\end{equation}
Here the factor of 2 accounts for the muons produced by ${\bar{c}}$ (only c
quarks are used in the program for simplicity); the $pN$ inclusive charm
production cross section is computed with the MNR program (here are the
integrations over the PDFs and partonic cross sections) and the last square
bracket is the number of muons of energy $E_{\mu}$ which reach sea level,
produced in the cascades simulated by PYTHIA. Each cascade is initiated by a
$c$ quark (in the `single' case) of energy $E_{c}$ and momentum $k$
(provided by the MNR routines) at a height $h$ chosen through a random
number $R$ homogeneously distributed between 0 and 1, which gives the value
of the $X$ probability distribution in Eq. (\ref{eq:flux2}), namely $R=e^{{
-X/\Lambda }_{N}{(E)}}$.
\subsection{Cascade evolution with PYTHIA routines}
The parton $c$ (or partons in the `double' case) generated by the first
stage, namely by the MNR routines, are entered in the event list of PYTHIA
and they become the starting point of the cascade generation.
PYTHIA first fragments the $c$ quark (in the `single' mode, or all the
partons in the `double' mode) into hadrons, after showering, which can be
optionally shut off. The charm quarks hadronize into $D^{0}$, $\bar{D}^{0}$,
$D^{\pm}$, $D_{s}^{\pm}$ and $\Lambda_{c}$. We used here the Peterson
fragmentation function option. For each hadron produced, a simple routine
added to PYTHIA decides if the hadron interacts in the atmosphere (loosing
some energy) or decays. This is the same approach as in TIG. PYTHIA follows
in this way the cascade in the atmosphere and populates the histograms of
muons and neutrinos as a function of their different energies. We mention
here a few important technical details. The `single' and `double' modes
described before use different fragmentation models. In the `single' mode
only one $c$ quark is available and is entered at the beginning of the event
list (with its energy and momentum in the partonic CM reference frame). In
this case PYTHIA uses the `independent fragmentation' model (see \cite
{jetset} for details). We only include $c$ quarks and at the end multiply
the result by a factor of two to account for initial $\bar{c}$ quarks.
In the `double' mode, instead, which we only use at the LO, we start with
two ($c\bar{c}$) partons in the event list. In this case we opt to use the
`string fragmentation' model (Lund model, \cite{jetset}). This model
generally gives better results than the independent fragmentation, in which
energy and momentum conservation have to be imposed a posteriori and whose
results depend on the reference frame used, which empirically is chosen to
be the partonic CM frame. To impose energy and momentum conservation in the
independent fragmentation, we used the option (MSTJ(3)=1, see again \cite
{jetset}) in which particles share momentum imbalance compensation according
to their energy (roughly equivalent to boosting events to CM frame) but we
have convinced ourselves that the results do not depend much on the way of
imposing energy and/or momentum conservation, because trial runs with
different options have given similar results for the fluxes.
Even if independent fragmentation is in general less desirable than string
fragmentation, we use the `single' mode as our main choice. The main reason
to use the `single' mode is that the simulations run in acceptably short
times (a few days) on the SUN computers we use, while giving results
practically identical to the `double' mode in the comparisons we have made
(see Fig.~6c). The simulation of the cascades in the `double' mode takes
between five and ten times longer. We tested the goodness of the
independent fragmentation by comparing the outcome of fluxes computed at the
Born level, in which the charm fluxes at production are identical (we put
one $c$ in the atmosphere and multiply the outcome by two to account for the
${\bar c}$ in one case, and we put $c {\bar c}$ in the atmosphere, instead,
in the second case) and the sole difference in both modes is due to the
different fragmentation models used. The results were extremely close (at
Born level the difference is less than $5\%$, at energies above $10^5 GeV$),
as can be seen in Fig.~6c.
Apart from the mentioned differences between the `single' and `double'
modes, the simulations then proceed basically in the same way in both modes.
For each of the `subevents', i.e. for each set of initial parton(s) put in
the event list, a certain height in the atmosphere is randomly chosen as
explained above, this being the position at which the partons are generated
from the initial proton-nucleon collision. This random height $h$ is
generated in a way similar to TIG (see Ref. \cite{TIG}), but different,
because we include a correction for nucleon regeneration in nucleon-nucleon
collisions by using $\Lambda_{N}$, the nuclear attenuation length, in Eq. (\ref{eq:11}) instead of $\lambda_{N}$ , the interaction thickness (see Eqs.~(\ref{eq:12}),(\ref{eq:13}) and (\ref{eq:14})).The only difference compared
to TIG (see Eq. (15) in the last paper of Ref. \cite{TIG}) is the inclusion
of the $(1-Z_{NN})$ correction term. This was done because we could not
include regenerated protons directly in our simulation of the cascades,
since events and subevents are now created by the MNR routines and not by
PYTHIA, as it was in TIG.
When parton showering is included at the beginning of the cascade simulation
performed by PYTHIA, some double counting is present. The double counting
appears when a LO diagram, for example $gg \to c \bar{c}$, with a subsequent
splitting contained in PYTHIA, for example $c \to g c$ is summed to NLO
diagram, $gg \to g c \bar{c}$ with the same topology, as if both diagram
were independent, when actually the NLO contains the first contribution when
the intermediate $c$ quark on mass shell. We have not tried to correct this
double counting but have instead confronted the results obtained including
showering (our standard option) with those excluding showering (in which
case there is no double counting) and found very similar leptonic fluxes
(see Fig.~6b).
The particles generated after the initial hadronization are then followed
throughout the atmosphere and PYTHIA evolves the cascade with the same
treatment of interactions and decays proposed by TIG. The final number of
muons and neutrinos at sea level is therefore calculated considering all the
`subevents', each with its respective weight $W_{i}$ from the MNR program,
which produce the final particles through all the possible decay channels of
charmed particles decaying into prompt leptons. Since only the decay modes
of charmed hadrons going into $\mu$ or $\nu_{\mu}$ or $\nu_{e}$ are left
open in the simulation, and there are essentially just 2 modes for each
charmed particle (for example: $D^{+}\rightarrow e^{+\ }\nu_{e}\ +anything$
, with branching ratio $=0.172$; $D^{+}\rightarrow\mu^{+\ }\nu_{\mu }\
+anything$, with branching ratio $=0.172$; all other channels closed), the
branching ratios for each of these modes is fictitiously taken by PYTHIA to
be 1/2 and need to be normalized by multiplying by the actual branching
ratio ($0.172$ for the example above) and dividing by $1/2$. Besides, since
not all events are accepted by PYTHIA to generate a complete cascade, the
result is normalized by dividing by the sum of all the weights of accepted
events and multiplying it by the total c inclusive cross section.
\subsection{Summary}
To summarize, our computation of the final fluxes is organized as follows.
$\bullet$ An external loop over the primary energy $E$ generates an
integration over $E$ in the range $10^{1} - 10^{11} GeV$.
$\bullet$ For each primary energy $E$, the MNR routines generate `subevents'
with weight $W_{i}$, for all the LO and NLO processes.
$\bullet$ Each subevent is assigned a random height (so that implicitly an
integration over $h$ is performed) and all this is passed to PYTHIA as a
definite set of parton(s) to be put at the beginning of the event list.
$\bullet$ For each of these `subevents', PYTHIA treats showering
(in our standard option), hadronization and evolution of the cascade in the
atmosphere, and generates the final leptons.
$\bullet$ For each decay channel of interest, the produced leptons are
weighted with $W_{i}$ and then summed into the final fluxes.
\section{Neutrino and muon fluxes}
\label{sect:fluxes}
Figs.~4-6 show the results of our simulations. Fig.~4 shows the total
inclusive charm-anticharm production cross sections $\sigma_{c\bar c}$, and the
$K$ factor for $c$ production, namely the ratio between the NLO and Born cross
sections, $K_{c} = \sigma_{c\bar c}^{NLO} / \sigma_{c\bar c}^{Born}$, for the
four PDFs we consider and for TIG. Fig.~5 shows our main results obtained with
our default choice of options: a `single' mode calculation including the
contributions from all processes in Eq.~(\ref{eq:15}) and with parton showering
included in the cascade simulation performed by PYTHIA). Finally Fig.~6 shows
the relative importance of the processes included in the fluxes and a
comparison of the `single' and `double' modes and of the `on' and `off'
showering options.
In Fig.~4a, the total inclusive charm-anticharm production cross sections
$\sigma_{c\bar c}$ are plotted over the energy range needed by our program, $E
\leq 10^{11}$ GeV, for our four different PDFs. They were calculated using the
MNR program, with the `calibration' described in Sect.~\ref{sect:NLO}, up to
the NLO contribution. For comparison, we also show the cross section used by
TIG and the Born (LO) contribution for one of the PDFs, MRS R1. We see in the
figure that all our cross sections agree at low energies, as expected due to
our `calibration' at 250 GeV, and are very similar for energies up to
$10^{6}-10^{7}$ GeV. At higher energies they diverge, differing by at most 50\%
at the highest energy we use, $10^{11}$ GeV. In fact, at energies beyond
$10^{7}$ GeV, the CTEQ 3M cross section becomes progressively larger than the
CTEQ 4M and MRS R2 cross sections, which are very close to each other. The MRS
R1 becomes on the contrary progressively lower than the other three.
We see in Fig.~4a that for energies above $10^{4}$ GeV our cross sections are
considerably higher than the one used by TIG. This difference can be traced in
part to the use by TIG of an option of PYTHIA by which the gluon PDF is
extrapolated to $x \leq 10^{-4}$ with $\lambda=0.08$, while all the PDFs we use
have a higher value of $\lambda\simeq0.2-0.3$. And in part to TIG scaling the
LO cross sections obtained with PYTHIA by a constant $K$ factor of 2, while at
large energies the $K$ factor is actually larger than 2 by about 10-15\% (see
Fig.~4b).
In Fig.~4b we explicitly show the $K$ factor for $c$ production, namely the
ratio between the NLO and Born cross sections, $K_{c} = \sigma_{c\bar
c}^{NLO} / \sigma_{c\bar c}^{Born}$, for our PDFs and for TIG. All the $
K_{c}$ values are around the usually cited value of $2$ for most of the
intermediate energies, but are larger at the lowest energies and also at the
highest energies (except for CTEQ 3M), and they all are within about 15\% of
each other.
Fig.~5 contains three sets of figures, one for each lepton: $\mu$, $\nu_{\mu} $
and $\nu_e$. The left figure of each set shows the $E^{3}$-weighted vertical
prompt fluxes, for all our PDFs up to NLO (labelled `NLO') and, as an example,
the LO (labelled `BORN') for MRS R1, together with the total fluxes up to NLO
of TIG, both from prompt and conventional sources (dotted lines). The right
part of each set shows the corresponding $K_{l}$ value (where
$l=\mu,\nu_{\mu},\nu_e$), i.e. the ratio of the total NLO flux to the Born flux
of the figure on the left. The figures show that our fluxes are higher than
those of TIG for $E > 10^{3}~{\rm GeV}$. Leaving apart differences in the two
simulations that cannot be easily quantified, this discrepancy can largely be
explained by the different cross sections used by TIG and us: the TIG cross
section is lower than ours for $E > 10^{4}$ GeV. Using a value of $\lambda$
similar to TIG ($\lambda \simeq 0$) at small $x$, we obtain fluxes similar to
those of TIG at energies above $10^6$ GeV \cite {GGVlambda}.
In particular, our fluxes are all larger than TIG by factors of 3 to 10 at the
highest energies, what puts our fluxes in the bulk-part of previous estimates
(see Refs. \cite{bnsz,c,gghv,zhv}). There is an evident dependence of the
fluxes on the choice of PDF. It is remarkable that MRS R2 and CTEQ 4M give very
similar results. Those of the MRS R1 become lower and those of the older CTEQ
3M PDF become higher as the energy increases (both differing by about 30-50\%
at the highest energies with respect to the MRS R2-CTEQ 4M fluxes). This is due
to the intrinsic differences of the PDF packages used and the consequent
different extrapolated values of $\lambda$ at small $x$ or high energies.
The CTEQ 3M fluxes were included to compare our results with those of Ref.
\cite{PRS}. We find our CTEQ 3M results to be close to those of Ref. \cite
{PRS}, in spite of the very different approaches used in the two calculations.
Our fluxes lie between the two curves for CTEQ 3M shown in Fig.8 of Ref.
\cite{PRS}, corresponding to different choices of renormalization and
factorization scales. Our fluxes are lower (by 30-40\% at $10^{7}{\rm GeV}$),
than the main CTEQ 3M choice of Ref.~\cite{PRS} (solid line
of their Fig.8), which is calculated using values of $\mu_R$, $\mu_F$ and $m_c$
similar to ours. Our cross section for charm production, for the CTEQ 3M case,
is essentially equal to the one used in Ref. \cite{PRS} (shown in their Fig.
2), so the discrepancies in the final fluxes are to be explained in terms of
the differences in the cascade treatment. It is very difficult to trace the
reasons for these differences.
We also see in the figures that, for each PDF, the fluxes for the
different leptons are very similar: those for $\nu _{\mu }$ neutrinos and $
\nu _{e}$ are essentially the same, those for muons are only slightly lower
(about $10\%$ less at the energies of interest). Also the $K_{l}$'s don't
differ much for the three leptons, apart from some unphysical fluctuations
especially evident at the highest energies. Even if they differ
the PDF, they all show a similar energy dependence, namely they
increase at low energies and sometimes at high energies also. This behavior
is also similar to that of the $K_{c}$ factors in Fig.~4b, but with a weaker
overall energy dependence, as expected, since the leptons of a given energy
result from $c$ quarks with a range of higher energies.
The $K_l$ factors are all within the range $2.1-2.5$: they are approximately
$2.2$ for MRS R1, $2.4$ for MRS R2 and CTEQ 4M, and $2.3$ for CTEQ 3M. Thus,
our analysis shows that evaluating the lepton fluxes only at the Born level,
and multiplying them by an overall $K_{l}$ factor of about $2.2-2.4$ (i.e. 10
to 20\% larger than the value of $2$ used by TIG\footnote{ We note that in the
original TIG model there is no distinction between $K_c$ and $K_l$ factors
since only the Born level is considered. Their $K=2$ factor is just a
multiplicative constant which can be considered either a $ K_c$ or a
$K_l$.}), can be good enough to evaluate the NLO fluxes within about 10\%.
Thus we find the approach used by TIG, who multiplied the LO fluxes obtained
with PYTHIA by two, essentially correct, except for their relatively low $K$
factor and the discrepancies existing even at Born level between our fluxes and
those of TIG. In fact, as we mentioned previously, the differences between our
final results and those of TIG depend mostly on the different total inclusive
$c$ cross sections, which can be traced to the extrapolation of the gluon PDF
at small $x$ rather than to the K factor. Possible causes of the different
results due to the intrinsic differences of the computer simulations cannot be
easily quantified.
In Fig.~6 we address three issues. First, we show that the fluxes can be
obtained within about 30\% with just the gluon-gluon process. This would
speed up the simulations and, when using the MNR program, would give
(contrary to intuition) higher fluxes than those actually derived from all
processes. Secondly, we show that the fluxes obtained including or excluding
showering in the simulation made by PYTHIA (we included showering in our
standard options) do not differ significantly. The third issue we deal with
is the difference between the `single' and `double' modes described in Sect.
3. We show that at LO the results from a `double' mode calculation coincide
with those of the much shorter `single' mode, that we use in all our
calculations. Let us deal with these three issues in turn.
In Fig.~6a we show, for a given PDF, the MRS R1, the relative importance of the
different processes contributing to the final fluxes. The solid line is the
total flux obtained as the sum of all the processes of Eq.~(\ref{eq:15}) and
the dotted line shows the result of only gluon-gluon fusion ($gg$), the sum of
Born ($gg$) and pure NLO (excluding Born) $gg$ processes. Also shown are the
separate contributions only at the Born and at the NLO (excluding LO) of both
$gg$ and quark-antiquark ($q {\bar q}$) fusion, what clearly shows that $gg$
dominates. This is to be expected because the gluon PDF is either much larger
than (for $x < 0.1$) or comparable to (for $x \simeq {\cal O}(0.1)$) the quark
PDFs. The figure plots the absolute value of the quark-gluon ($qg$) terms
because, for the values of the factorization scale that we employ in our
calculations, these terms are negative. This is due to the way the original MNR
calculation is subdivided into processes. In fact, in the MNR program, a part
of the quark-gluon contribution to the cross sections is already contained in
other processes, and must be subtracted in the processes labelled as $qg$. The
amount subtracted depends on the factorization scale $\mu_{F}$ and may drive
the $qg$ contribution negative. Roughly speaking, if $\mu_{F}$ is small the
$qg$ term is positive,
otherwise (as in our case) the term is negative. The absolute value of the $
qg$ term is in between the $q\bar{q}$ and the $gg$ terms, what makes
negative the sum of all the processes different from $gg$. Thus, gluon-gluon
processes alone give a result slightly larger than the total, by about 30\%.
In Fig.~6b we check the effect of shutting off the showering option
available in PYTHIA. We study only one specific case, the MRS R1. The
overall effect is minimal: the exclusion of showering slightly increases the
energy of the parent charmed hadrons and therefore causes the final fluxes
of lepton daughters to move towards higher energies; the effect is barely
noticeable and just slightly more important for the Born fluxes (the
overall difference is about $5\%$). When showering is included some double
counting occurs, whose effect must be smaller than the difference between
the results with showering on and off (since in this case no double counting
occurs).
Finally in Fig.~6c we confront the `single' and `double' modes of the
program, for just one PDF, MRS R1, at Born level. At this level, the
calculation of the charm flux at production is identical (we obtain the
fluxes from $c$ and multiply by two at the end to account for the ${\bar c}$
in one case, and we obtain the fluxes directly from $c {\bar c}$ in the
other). So, what is actually compared in the two modes at the Born level is
the fragmentation model: independent fragmentation in the `single' mode and
string (Lund) fragmentation in the `double' mode. The results from both
modes at the Born level are almost identical: as already remarked the
difference is less than $5\%$ for energies above $10^6~{\rm GeV}$.
\section{Conclusions}
\label{sect:conclusions}
We have used the actual next-to-leading order perturbative QCD calculations
of charm production cross sections, together with a full simulation of the
atmospheric cascades, to obtain the vertical prompt fluxes of neutrinos and
muons.
Our treatment is similar to the one used by TIG, except for the very
important difference of including the true NLO contribution, while TIG used
the LO charm production cross section multiplied by a constant $K$ factor of
2 to bring it in line with the next-to-leading order values. The main goal
of this paper is to examine the validity of TIG's procedure by computing the
ratio of the fluxes obtained with the NLO charm production cross section
versus those obtained with the LO cross section.
These ratios, the $K_l$ factors are between 2.1 and 2.5 for the different gluon
PDFs in the energy range from $10^2$ to $10^9$ GeV (see Fig.~5). Consequently,
our analysis shows that evaluating the lepton fluxes only at the Born level,
and multiplying them by an overall factor of about $ 2.2-2.4$, slightly
dependent on the PDF, can be good enough to evaluate the NLO fluxes within
about 10\%. Therefore, we find the approach used by TIG (i.e.\ multiplying the
LO fluxes by two) essentially correct, except for their relatively low $K$
factor. We find different lepton fluxes than TIG, but this is mostly due to the
discrepancies, even at Born level, between our charm production cross sections
and TIG's.
In fact, the prompt neutrino and muon fluxes found by TIG were lower than
the lowest previous prediction. We find here instead fluxes in the bulk part
of those previous predictions. This difference can be traced largely to the
use by TIG of an option of PYTHIA by which the gluon PDF is extrapolated for
$x \leq 10^{-4}$ with $\lambda=0.08$, while all the PDFs in this paper have
a higher value of $\lambda\simeq0.2-0.3$. Using a value of $\lambda$ similar
to TIG ($\lambda \simeq 0$) we obtain fluxes similar to those of TIG, at
energies above $10^6$ GeV \cite{GGVlambda}.
\bigskip\bigskip\bigskip\bigskip
{\large \noindent \textbf{Acknowledgements}}
\bigskip The authors would like to thank the Aspen Center For Physics, where
this work was initiated, for hospitality, and M. Mangano and P. Nason for
the MNR program and helpful discussions. This research was supported in part
by the US Department of Energy under grant DE-FG03-91ER40662 Task C.
\newpage
|
\section{Introduction}
Neutron stars contain matter in one of the densest forms found in the
universe. Their central density ranges from a few time the density of
normal nuclear matter to about one order of magnitude higher,
depending on the star's mass and the equation of state (EOS). Neutron
stars therefore provide us with a powerful tool for exploring the
properties of such dense matter. In the last decades, this tool was
applied to, among other topics, the determination of the EOS of dense,
charge neutral, $\beta$-equilibrated matter by means of comparing the
theoretical predicted properties with observations of neutron
stars. This was attempted by studying, for example, the maximum stable
star mass \cite{Kerkwijk95a}, the minimum rotation period
\cite{Friedman86a,Weber92a}, or the thermal behaviour (Tsuruta 1966
\nocite{Tsuruta66}, Schaab et al. 1996 \nocite{Schaab95a}, Page 1997
\nocite{Page97a}; see Balberg, Lichtenstadt \& Cook 1998
\nocite{Balberg98a} for a recent review).
Recently, Strohmayer et al. \shortcite{Strohmayer96a} and Van der Klis
et al. \shortcite{VanDerKlis96a} discovered with the Rossi X-ray
Timing Explorer (RXTE) kilohertz quasi-periodic oscillations (QPOs) in
the X-ray brightness of low-mass X-ray binaries (LMXBs, see van der
Klis 1997 \nocite{VanDerKlis97a} for a recent review). In subsequent
observations, three QPOs were often detected simultaneously in a given
source. The frequency separation between both QPOs is almost constant,
although the frequencies of the two QPOs themselves vary by several
hundred Hertz. Up to now, the only exceptions are the Atoll sources
4U\,1608-52 and 4U\,1735-44 and the Z-source Scorpius X-1 in which the
frequency separation varies with the luminosity by roughly $\pm
15$\,\% \cite{Mendez98b,VanDerKlis97b,Ford98c}. Psaltis et
al. \shortcite{Psaltis98a} found that the QPO data of the other LMXBs
are consistent both with being constant and with varying by a similar
fraction as in the two sources quoted before. The frequency of
oscillations during type I X-ray bursts detected in some sources are
consistent with the frequency separation of the two QPOs or with its
first overtone. Only in the source 4U\,1636-536 the averaged frequency
separation, $\Delta\nu=251$~Hz, and the half of the frequency in type
I bursts, $\nu_{\rm burst}=581$~Hz, differ by approximately 15\,\%
\cite{Mendez98c}. In the power spectrum of bursts of the same source
Miller \shortcite{Miller98c} found a second signal at 290~Hz~$\sim 1/2
\nu_{\rm burst}$. These observations strongly favour the beat
frequency model where the frequency separation between the two QPOs
originates from the stellar spin, whereas the higher QPO is produced by
accreting gas in a stable, nearly circular orbit around the neutron
star. Though, it has to be clarified how the slightly varying
frequency separation and the small deviation of the frequency
separation from the burst frequency can be incorporated into this
model.
Beside the two kilohertz QPOs and the burst oscillations, also QPOs
with frequencies of a few tens of Hertz were detected in some
sources. Their frequencies correlate with the high frequency kilohertz
QPOs. These low frequency QPOs were interpreted by Stella \& Vietri
\shortcite{Stella97a} to originate from the Lense-Thirring precession of the
accreting disc due to the frame dragging effect of the rapidly
spinning neutron star \cite{Lense18a}.
Both, the identification of the high frequency kilohertz QPOs with the
orbital frequency of a stable circular orbit and the low frequency
QPOs with the frame dragging frequency of the same orbit allow to
constrain the mass of the neutron star and also the EOS of neutron
star matter \cite{Lamb97a,Stella97a}. If the evidence for the
detection of QPOs of the innermost stable circular orbit in the
sources 4U\,1608-52, 4U\,1636-536 \cite{Kaaret97a}, and 4U\,1820-30
\cite{Zhang98c} can be confirmed by future observations, the
constraints are rather severe, allowing only a few stiff EOSs.
The frequency separation of the two kilohertz QPOs show that the
neutron star in LMXBs are rapidly rotating with periods ranging from
2.5~ms to 4~ms. It can therefore be expected that the geometry of the
neutron star and its exterior space time is non-spherical. Since the
innermost stable orbit is located at only a few kilometres above the
star's surface, the deviation from the Kerr space time are
large and should not be neglected. In order to compare theoretical
neutron star models with QPO observations, a completely general
relativistic calculation of the rotating neutron star structure and
space time geometry is therefore necessary.
In order to study the impact and the discrimination power of the QPO
data in greater detail, we select a broad collection of modern EOSs,
which were obtained utilising numerous assumptions about the dynamics
and composition of super dense matter. To mention several, these are:
the many-body technique used to determine the EOS; the
model for the nucleon-nucleon interaction; description of electrically
charge neutral neutron star matter in terms of either only neutrons
and protons in generalised chemical equilibrium ($\beta$ equilibrium)
with electrons and muons, or nucleons, hyperons and more massive
baryon states in $\beta$ equilibrium with leptons; hyperon coupling
strengths in matter; inclusion of meson ($\pi$, $K$) condensation;
treatment of the transition of confined hadronic matter into quark
matter; and assumptions about the true ground state of strongly
interacting matter (i.e., absolute stability of strange quark matter
relative to baryon matter).
The paper is organised as follows. In Sect. \ref{sec:equations} we
summarise the equations which govern the space time structure and
compare the approximate values of the orbital frequencies, the radius
of the innermost stable orbit, and the Lense-Thirring precession
frequencies with the respective values from the exact numerical
solution of Einstein's equations. The physics of the EOSs is discussed
in Sect. \ref{sec:eos}. The high frequency kilohertz QPOs and their
interpretation in combination with their compatibility with the
different EOSs are discussed in Sect. \ref{sec:constr1}. The
implications of the identification of the low frequency QPOs with
Lense-Thirring precession are presented in Sect. \ref{sec:constr2}. We
summarise our results, the constraints to the neutron star masses, and
the conclusions concerning the neutron star EOS in
Sect. \ref{sec:concl}.
\section{Space Time Around Rapidly Rotating Neutron Stars}
\label{sec:equations}
\subsection{Einstein Equations}
The stationary, axis-symmetric, and asymptotic flat metric in quasi
isotropic coordinates reads
\begin{equation} \label{eq:cool.metric}
{\rm d} s^2 = -{\rm e}^{2\nu}{\rm d} t^2+{\rm e}^{2\phi}({\rm d}\varphi-N^\varphi {\rm d} t)^2
+{\rm e}^{2\omega}({\rm d} r^2+r^2{\rm d}\theta^2),
\end{equation}
where the metric coefficients $g_{\mu\nu}=g_{\mu\nu}(r,\theta)$ are
functions of $r$ and $\theta$ only. The metric coefficients are
determined by the Einstein equation ($c=G=1$)
\begin{equation}
{\bf G} = 8\pi{\bf T},
\end{equation}
and the energy-momentum conservation
\begin{equation}
{\bf\nabla}\cdot{\bf T}=0,
\end{equation}
where ${\bf T}= (e+p){\bf u}\otimes{\bf u}+p\,{\bf g}$ is the
stress-energy tensor of an ideal fluid with the 4-velocity
\begin{equation} \label{eq:def.u}
{\bf u} = \Gamma{\rm e}^{-\nu}{\bf e}_t + \Omega\Gamma{\rm e}^{-\nu}{\bf e}_\varphi,
\end{equation}
the energy density $e$, and the pressure $p$. The Lorentz factor
$\Gamma$ is given by ${\bf u}\cdot{\bf u}=-1$, hence
\begin{equation} \label{eq:gamma}
\Gamma =
\left(1-{\rm e}^{2(\phi-\nu)}\left(\Omega-N^\varphi\right)^2\right)^{-\tfrac{1}{2}}.
\end{equation}
$\Omega=u^\varphi / u^t$ is the angular velocity of the fluid with
respect to an observer at infinity. The proper velocity $U$ of the
fluid with respect to the local Eulerian Observer $\mathfrak{O}_0$
\cite{Smarr78a} is given by the equation
\begin{equation} \label{eq:proper.vel}
U = \frac{1}{g_{\varphi\varphi}^{1/2}\Gamma}{\bf e}_\varphi\cdot{\bf u}
= {\rm e}^{\phi-\nu}\left(\Omega-N^\varphi\right).
\end{equation}
Note that if the fluid were at rest with respect to the Observer
$\mathfrak{O}_0$, $U=0$, it would not necessarily be at rest for an inertial
observer at infinity, since $\Omega=N^\varphi\neq 0$. This phenomena
of \emph{dragging of the inertial frame} was first studied by Lense
\& Thirring \shortcite{Lense18a} and turns out to be important for the
investigation of the Lense-Thirring precession of the accreting disc
(see Sect. \ref{sec:constr2}).
The four non-trivial Einstein equations together with the
energy-momentum conservation are solved via a finite difference scheme
\cite{Schaab97a}. We follow Bonazzola et al. \shortcite{Bonazzola93a} in compactifying
the outer space to a finite region by using the transformation $r
\rightarrow 1/r$. The boundary condition of approximate flatness can
then be exactly fulfilled. The neutron star model is uniquely
determined by fixing one of the parameters: central density,
gravitational mass, baryon number, Kepler orbiting frequency at the
star's equator or at the marginally stable radius, as well as one of
the parameters: angular velocity, angular momentum, or stability
parameter $t=T/|W|$. The models of maximum mass and/or maximum
rotation velocity can also be calculated.
\subsection{Stable Circular Orbits} \label{ssec:stable}
Since ${\bf e}_t$ and ${\bf e}_\varphi$ are Killing vectors, the
components $p_t$ and $p_\varphi$ of the 4-momentum of a freely falling
particle are constants of motion and can be identified with the
negative of the specific energy $E$ (in units of the rest mass $m_0$)
and the specific angular momentum $L$, respectively. For a particle
motion confined to the equatorial plane the geodesic equation
$p_\mu^\mu=-1$ yields:
\begin{equation}
{\rm e}^{-2\omega}p_r^2 = -1+{\rm e}^{-2\nu}(E-N^\varphi L)^2-{\rm e}^{-2\phi}L^2
= {\rm e}^{-2\nu}V(E,L,r),
\end{equation}
where $V$ is the effective potential for the particle motion
\cite[pp. 655]{Misner73}. A circular orbit, $p_r=0$, is given by the
expression \eqref{eq:def.u} for the 4-velocity of the fluid inside the
star. Then, the specific energy $E$ and angular momentum $L$ can be
expressed in terms of the Lorentz factor $\Gamma$ (Eq. \ref{eq:gamma})
and the proper velocity $U$ (Eq. \ref{eq:proper.vel}):
\begin{align}
E &= \Gamma\left({\rm e}^\nu+{\rm e}^\phi N^\varphi U\right) \\
L &= {\rm e}^\phi\Gamma U.
\end{align}
The circular orbit exists if $V_{,r}=0$, thus\footnote{We use the usual
convention, that $\cdot_{,\mu}$ represents the partial differential
$\partial/ \partial x^\mu$.}
\begin{equation}
-\phi_{,r}U^2+{\rm e}^{\phi-\nu}N^\varphi_{,r}U+\nu_{,r} = 0.
\end{equation}
The proper velocity $U$ of a particle corotating with the star is given by
\begin{equation}
U = \frac{{\rm e}^{\phi-\nu}N^\varphi_{,r}
+\sqrt{{\rm e}^{2(\phi-\nu)}(N^\varphi_{,r})^2+4\phi_{,r}\nu_{,r}}}
{2\phi_{,r}}.
\end{equation}
The circular orbit is stable, if $V_{,rr}<0$ with (see also Cook,
Shapiro \& Teukolksky 1992 \nocite{Cook92a} and Datta, Thampan \&
Bombacci 1998 \nocite{Datta98a})
\begin{multline}
V_{,rr} = 2{\rm e}^{2\nu}\Gamma^2\left({\rm e}^{\phi-\nu}UN^\varphi_{,rr}
-{\rm e}^{2(\phi-\nu)}U^2(N^\varphi_{,r})^2+\nu_{,rr} \right. \\
\left. +2(\nu_{,r})^2
-U^2\phi_{,rr}+2U^2(\phi_{,r})^2-4\nu_{,r}\phi_{,r}U^2\right).
\end{multline}
The zero of $V_{,rr}$ determines the radius $r^{\rm ms}$ of the
innermost stable or marginally stable circular orbit. The Kepler
frequency $\nu_{\rm K}$ of a circular orbit as measured by a distant
observer is given by the proper velocity $U$ through the expression
\begin{equation}
\nu_{\rm K} = \frac{1}{2\pi}\left({\rm e}^{\nu-\phi}U+N^\varphi\right),
\end{equation}
and the Lense-Thirring precession frequency by
\begin{equation}
\nu_{\rm LT} = \frac{1}{2\pi}N^\varphi.
\end{equation}
The respective values at the innermost stable orbit with radius
$r^{\rm ms}$ are denoted by $\nu_{\rm K}^{\rm ms}$ and $\nu_{\rm
LT}^{\rm ms}$, respectively. The circumferential radius $r_{\rm circ}$
of the object is linked to the equatorial coordinate radius $r$ by the
expression
\begin{equation}
r_{\rm circ} = {\rm e}^\phi r.
\end{equation}
\subsection{Approximate Expressions}
Though we shall use the exact expressions for $\nu_{\rm K}$, $\nu_{\rm
LT}$, $\nu_{\rm K}^{\rm ms}$, and $\nu_{\rm LT}^{\rm ms}$ in our
investigation, we give also, for the purpose of comparison, the
approximate expressions valid to first order of the dimensionless
angular momentum $j=J/M^2$, where $J$ and $M$ denotes the angular
momentum and gravitational mass, respectively. The approximate
expressions have the advantage to constrain the mass and the angular
momentum independently of the EOS. To first order in $j$ the
corresponding expression are \cite{Miller98a}:
\begin{align}
\nu_{\rm k} &= \frac{1}{2\pi}\left(\frac{M}{r_{\rm circ}^3}\right)^{\tfrac{1}{2}}
\left(1-\left(\frac{M}{r_{\rm circ}}\right)^{\tfrac{3}{2}}j\right), \\
\nu_{\rm LT} &= \frac{2}{2\pi}\frac{M^2}{r_{\rm circ}^3}j, \\
r_{\rm circ}^{\rm ms} &=
6M\left(1-\left(\tfrac{2}{3}\right)^{\tfrac{3}{2}}j\right), \\
\nu_{\rm K}^{\rm ms} &= \frac{6^{-\tfrac{3}{2}}}{2\pi}\frac{1}{M}
\left(1+11\times 6^{-\tfrac{3}{2}}j\right) \label{eq:appr.Kms}, \\
\nu_{\rm LT}^{\rm ms} &= \frac{6^{-3}}{\pi}\frac{j}{M}.
\end{align}
\section{Equations of State} \label{sec:eos}
\subsection{Neutron Star Matter}
The EOS of neutron star matter is the basic input quantity to the
structure equations discussed in Sect. \ref{sec:equations}. Its
knowledge over a wide range of densities is necessary. Whereas the
density at the star's surface corresponds to the density of iron, the
density in the centre of a very massive star can reach 15 times the
density of normal nuclear matter. Since neutron star matter in
chemical equilibrium (i.e.\ generalized $\beta$-equilibrium) is highly
isospin-asymmetric and may carry net-strangeness its properties cannot
be explored in laboratory experiments. Therefore, one is left with
models for the EOS which depend on theoretically motivated assumptions
and/or speculations about the behaviour of super dense matter. One
main source of uncertainty is the competition between non-relativistic
versus relativistic models. Though both treatments give satisfactory
results for normal nuclear densities, they provide quite different
results if one extrapolates to higher densities (see, e.g, Huber et
al. 1998\nocite{Huber97a}). Moreover, one encounters strong
differences in the high density regime depending on the underlying
dynamics, the many-body approximation and the assumptions about the
composition. The simplest approach describes neutron star matter by
pure neutron matter. Since pure neutron matter is certainly not the
true ground state of neutron star matter, it will quickly decay by
means of the weak interaction into chemically equilibrated neutron
star matter, whose fundamental constituents -- besides neutrons -- are
protons, hyperons and possibly more massive baryons. Even a transition
to quark matter (so-called hybrid stars) and the occurrence of pion-
or kaon-condensates is possible.
The cross section of a neutron star can be split roughly into three
distinct regimes \cite{Boerner73b,Weber91}. The first one is the
star's outer crust, which consists of a lattice of atomic nuclei and a
Fermi liquid of relativistic, degenerate electrons. The inner crust
extends from neutron drip density, $\rho=4.3\times 10^{11}\mathrm{\,g\,cm}^{-3}$, to a
transition density of about $\rho_{\rm tr}=1.7\times 10^{14}\mathrm{\,g\,cm}^{-3}$
\cite{Pethick95}. Beyond this transition density $\rho_{\rm tr}$
one enters the star's third regime, that is, its core where all atomic
nuclei have dissolved into their constituents, protons and
neutrons. Furthermore, as outlined just above, due to the high Fermi
pressure the core will contain hyperons, eventually more massive
baryon resonances, and possibly a gas of free up, down and strange
quarks.
The EOS of the outer and inner crust has been studied in
several investigations and is rather well known. We shall adopt for
these regions the models derived by Haensel \& Pichon
\shortcite{Haensel94a} and by Negele \& Vautherin \shortcite{Negele73},
respectively. The models for the EOS of the star's core
are discussed in detail in Schaab et al. \shortcite{Schaab95a}.
An overview of the collection of EOSs used in this paper
for the core region is given in Tab. \ref{tab:eos}. We have chosen a
representative collection of different models in order to check which
EOSs are compatible with the QPOs and their theoretical
interpretation.
\begin{table*}
\centering \caption{Dynamics and approximation schemes for EOSs
derived for the cores of neutron stars \label{tab:eos}} \smallskip
\scriptsize
\begin{tabular}{ccccc}
\hline
\hline
EOS & Composition & Interaction & Many body approach & Reference \\
\hline
UV$_{14}$+UVII
& p, n, e$^-$, $\mu^-$
& Urbana V$_{14}$ and Urbana VII
& NRV
& Wiringa, Fiks \& Fabrocini \shortcite{Wiringa88} \\
UV$_{14}$+TNI
& p, n, e$^-$, $\mu^-$
& Urbana V$_{14}$ and TNI
& NRV
& Wiringa et al. \shortcite{Wiringa88} \\
TF
& p, n, e$^-$, $\mu^-$
& TF96
& Thomas-Fermi model
& Strobel et al. \shortcite{Strobel96a} \\
HV$_{\rm pn}$
& p, n, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Weber \& Weigel \shortcite{Weber89} \\
HV
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Weber \& Weigel \shortcite{Weber89} \\
RH1
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, $\Delta^{-,0,+,++}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\pi$, $\rho$
& RBHF+RH
& Huber et al. \shortcite{Huber97a} \\
RHF$_{\rm pn}$
& p, n, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\pi$, $\rho$
& RBHF+RH
& Huber et al. \shortcite{Huber97a} \\
RHF1
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, $\Delta^{-,0,+,++}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\pi$, $\rho$
& RBHF+RHF
& Huber et al. \shortcite{Huber97a} \\
RHF8
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, $\Delta^{-,0,+,++}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\pi$, $\rho$
& RBHF+RHF
& Huber et al. \shortcite{Huber97a} \\
G$_{300}$
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Glendenning \shortcite{Glendenning89} \\
G$^\pi_{300}$
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$, $\pi$-condensation
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Glendenning \shortcite{Glendenning89} \\
G$^{\rm K240}_{\rm pn}$
& p, n, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Glendenning \shortcite{Glendenning95b} \\
G$^{\rm K240}_{\rm M78}$
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Glendenning \shortcite{Glendenning97c} \\
G$^{\rm K240}_{\rm M78u}$
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Glendenning \shortcite{Glendenning95b} \\
G$^{\rm K240}_{\rm B180}$ & p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$, quarks
& exchange of $\sigma$, $\omega$, $\rho$
& RH
& Glendenning \shortcite{Glendenning97c} \\
TM1-m1
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$, $\phi$
& RH
& Schaffner \& Mishustin \shortcite{Schaffner95a} \\
TM1-m2
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$, $\sigma^*$, $\phi$
& RH
& Schaffner \& Mishustin \shortcite{Schaffner95a} \\
NLSH1
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$, $\phi$
& RH
& Schaffner \& Mishustin \shortcite{Schaffner95a} \\
NLSH2
& p, n, $\Lambda$, $\Sigma^{\pm,0}$, $\Xi^{0,-}$, e$^-$, $\mu^-$
& exchange of $\sigma$, $\omega$, $\rho$, $\sigma^*$, $\phi$
& RH
& Schaffner \& Mishustin \shortcite{Schaffner95a} \\
SM$_{\rm B145}$
& u, d, s
& Bag model
& Fermi gas
& Farhi \& Jaffe \shortcite{Farhi84} \\
SM$_{\rm B160}$
& u, d, s
& Bag model
& Fermi gas
& Farhi \& Jaffe \shortcite{Farhi84} \\
\hline\hline
\end{tabular}
\comments{Abbreviations: NRV=non-relativistic variational method,
RH=relativistic Hartree approximation, RHF=relativistic Hartree-Fock
approximation, RBHF=relativistic Br\"uckner-Hartree-Fock
approximation.}
\end{table*}
An important characteristic of relativistic models is the appearance
of a new saturation mechanism. In relativistic theories the repulsive
force is caused by the exchange of vector mesons coupled to the baryon
densities, whereas the attraction is coupled to the scalar densities
by means of the scalar mesons. Hence, the repulsion increases with
increasing density with respect to the attraction. Since
non-relativistic treatments do not distinguish both kinds of densities,
one has to introduce adhoc forces depending explicitly on density in
order to reproduce properties of saturated nuclear matter.
The potential parameters and coupling constants in the pure nucleonic
sector are adjusted to nucleon--nucleon scattering data and properties
of the deuteron in the case of the non-relativistic microscopic models
and the relativistic Br\"uckner-Hartree-Fock models. In this sense,
these models are called \emph{parameter free}. For the Hartree- and
the Hartree-Fock approximation, as well as for the Thomas-Fermi model
such an adjustment is not possible, since the free force parameters
would not yield saturation of nuclear matter. In such more
phenomenological approximations the coupling constants are adjusted to
properties of saturated nuclear matter or atomic nuclei
\cite{Weber89,Schaffner95a,Myers96a}. In a more recent and more elaborate
investigation \cite{Huber97a}, the adjustment was performed with
respect to properties of neutron star matter at 2--3 times nuclear
density known from relativistic Br\"uckner-Hartree-Fock
calculations. Relativistic Br\"uckner-Hartree-Fock calculations cannot
be performed over the whole density range at present, since an
inclusion of hyperons leads to rather complex equation systems. Even
for pure nucleonic matter, the self consistent method is numerically
stable up to 2--3 times nuclear density only \cite{Huber97a}.
Since the coupling constants of hyperons are not obtainable from
properties of normal nuclei and data of hypernuclei are relatively
scarce, one has a greater freedom in the selection of parameter sets
in the hyperonic sector. From a theoretical standpoint one may first
utilise the SU(6) symmetry of the quark model for the vector coupling
constants and adjust the $\sigma$-coupling from the binding energy of
the $\Lambda$ hyperon in nuclear matter. It turns out, however, that
this procedure gives different $\sigma$-couplings depending on the
chosen many-body approximation. Therefore, it seems reasonable to vary
both couplings with the constraint of the $\Lambda$ binding energy in
nuclear matter in such phenomenological many-body theories. A further
constraint in this procedure is the compatibility with the allowed
neutron star masses (for more details, see Huber et al. 1998
\nocite{Huber97a} and Huber 1998 \nocite{Huber98a}). The coupling
constants for the strange mesons can be obtained to a certain extent
from the data of double $\Lambda$-hypernuclei
\cite{Schaffner95a}. Batty et al. \shortcite{Batty94a} claim some doubts on
the existence of $\Sigma$ hyperons in neutron star matter. However,
the disappearance of $\Sigma$ hyperons would only slightly change the
EOS
\cite{Balberg96a}. A further open question is still the existence of
$\Delta$ isobars in the interior of neutron stars. In relativistic
Hartree treatments, the isospin unfavoured $\Delta^-$ isobar does not
appear because of the rather large $\rho$-coupling, which is necessary
to reproduce the empirical symmetry coefficient
\cite{Huber97a}. $\Delta$ isobars are therefore often neglected a
priori in relativistic Hartree treatments
\cite{Schaffner95a}. However, in relativistic Hartree-Fock
approximations, the $\rho$-coupling becomes smaller due to the
exchange contributions. For that reason, the charge favoured
$\Delta^-$ may now enter the composition depending on the behaviour of
the effective $\Delta$-mass in neutron star matter, which is quite
uncertain \cite{Huber97a}.
The possibility of a transition of confined hadronic matter into quark
matter at high densities is included in the EOS labelled
G$^{\rm K240}_{\rm B180}$~ \cite{Glendenning95b} (so-called hybrid stars). The transition
was determined for a bag constant of $B^{1/4}=180$ MeV which places
the energy per baryon of strange quark matter at 1100 MeV, well above
the energy per nucleon in saturated nuclear matter as well as in the
most stable nucleus, $^{56}{\rm Fe}$ ($E/A\approx$ 930 MeV). This
model predicts a transition to a mixed phase consisting of quark
matter and hadronic matter at a density as low as $1.6\,n_0$. The pure
quark-matter phase is reached at a density of about $10 \, n_0$, which
is close to the central density of the heaviest and thus most compact
star constructed from such an EOS. A phase transition to pion
condensation is accounted for in G$^\pi_{300}$. This EOS predicts a phase
transition at $n\approx 1.3\,n_0$. The possibility of absolutely
stable strange quark matter will be considered in the following
section.
\begin{figure} \centering
\psfig{figure=fig1.ps,height=\linewidth,angle=-90}
\caption{Pressure-density relation for several EOSs}
\label{fig:eos}
\end{figure}
The stiffness of the EOS depends strongly on the internal degrees of
freedom. Generally, the more degrees of freedom are taken into account
the softer the EOS becomes (s. Fig. \ref{fig:eos}). A softer EOS, in
turn, leads to lower maximum neutron star masses and, for fixed mass,
to higher central densities.
\subsection{Strange Stars} \label{ssec:strangestars}
The hypothesis that strange quark matter may be the absolute ground
state of the strong interaction (not $^{56}{\rm Fe}$) has been raised
independently by Bodmer \shortcite{Bodmer71} and Witten \shortcite{Witten84}.
If the hypothesis is true, then a separate class of compact stars
could exist, which range from dense strange stars to strange dwarfs to
strange planets \cite{Weber94,Glendenning94a,Glendenning95c}. In
principle both strange and neutron stars could exist. However if
strange stars exist, the galaxy is likely to be contaminated by
strange quark nuggets which would convert via impact ``conventional''
neutron stars into strange stars
\cite{Glendenning90,Madsen91,Caldwell91}. This would imply that the
objects known to astronomers as pulsars are probably rotating strange
matter stars and not neutron matter stars, as it is usually assumed.
Presently there is no sound scientific basis on which one can either
confirm or reject the hypothesis, so that it remains a serious
possibility of fundamental significance for compact stars
\cite{Weber96a,Weber96b,Madsen98a} plus various other physical
phenomena \cite{Aarhus91}. Below we will explore the implications of
the existence of strange stars with respect to the QPO phenomena. This
enables one to test the possible existence of strange stars and thus
draw definitive conclusions about the true ground state of strongly
interacting matter.
As pointed out by Alcock et al. \shortcite{Alcock86}, a strange star can
carry a solid nuclear crust whose density is strictly limited by
neutron drip. This possibility is caused by the displacement of
electrons at the surface of the strange matter core, which generates an
electric dipole layer there. It can be sufficiently strong to
produce a gap between ordinary atomic matter (crust) and the
quark-matter surface, which prevents a conversion of ordinary atomic
matter into the assumed lower-lying ground state of strange matter.
Obviously, free neutrons, being electrically charge neutral, cannot
exist in the crust, because they do not feel the Coulomb barrier and
thus would gravitate toward the strange-quark matter core, where they
are converted by hypothesis into strange matter. Consequently, the
density at the base of the crust (inner crust density) must always be
smaller than neutron drip density. The situation is illustrated in
Fig. \ref{fig:sm:eos1} where the EOS of a strange star
with crust is shown.
\begin{figure} \centering
\psfig{figure=fig2.ps,height=\linewidth,angle=-90}
\caption{Pressure-density relation for strange stars with crust. The bag
constant is $B^{1/4}=145$~MeV or 158~MeV, respectively, the mass of
the strange quark is $m_s=100$~MeV. }
\label{fig:sm:eos1}
\end{figure}
The strange-star models presented in the following are constructed for
an EOS of strange matter derived by Farhi \& Jaffe
\shortcite{Farhi84}. We shall study models with a fixed strange quark mass,
$m_{\rm s}=100$~MeV, and the two bag constants, $B^{1/4}=145$~MeV and
$B^{1/4}=158$~MeV.
\section{Kilohertz Quasi Periodic Oscillations} \label{sec:constr1}
\begin{table*}
\centering \caption{Observational data of kilohertz QPOs
\label{tab:qpo.data}}
\smallskip
\begin{tabular}{ccccccc}
\hline\hline
Source & Type & $\nu_{\rm QPO1}$ [Hz] & $\nu_{\rm QPO2}$ [Hz]
& $\Delta\nu_{\rm QPO1}$ [Hz] & $\nu_{\rm Burst}$ [Hz]
& References \\
\hline
4U\,0614+091 & Atoll & 750--$1145\pm 10$ & 480--800
& $323\pm 4$ & 328\dag
& 1 \\
4U\,1608-52\ddag & Atoll & 940--1125 & 650--890
& $233\pm 12$--$293\pm 7$&
& 2,3,4,5,6 \\
4U\,1636-536\ddag & Atoll & 1147--1228 & 830--940
& $257\pm 20$--$276\pm 10$& 581
& 6,7,8 \\
4U\,1728-34 & Atoll & 988--$1058\pm 12$ & 638--716
& $355\pm 5$ & 363
& 7,9 \\
KS 1731-260 & Atoll & 1176--$1197\pm 10$ & 900
& $260\pm 10$ & $523.92\pm 0.05$
& 10,11 \\
4U\,1735-44 & Atoll & 982--$1161\pm 1$ & 632--729
& $296\pm 12$--$341\pm 7$ &
& 12,13 \\
4U\,1820-30\ddag & Atoll & 640-$1060\pm 20$ & 400--800
& $275\pm 8$ &
& 14 \\
Aql X-1 & Atoll & 740--830 &
& & 549
& 15 \\
GX 3+1 & Atoll & &
& &
& 16 \\
4U\,1705-44 & Atoll & $1074\pm 10$ & 776--867
& $298\pm 11$ &
& 17 \\
Sco X-1 & Z & 872--1115 & 565--890
& 310--230 &
& 18 \\
GX 5-1 & Z & 567--895 & 325--448
& $298\pm 11$ &
& 19 \\
GX 17+2 & Z & 645--$1087\pm 12$ & 480--781
& $294\pm 8$ &
& 20 \\
Cyg X-2 & Z & 731--$1007\pm 12$ & 490--780
& $346\pm 29$ &
& 21 \\
GX 349+2 & Z & $978\pm 9$ & 712
& $266\pm 13$ &
& 22 \\
GX 340+0 & Z & 567--$820\pm 19$ & 247--625
& $325\pm 10$ &
& 23 \\
\hline\hline
\end{tabular} \comments{\dag: QPO is statistical not significant,
\ddag: $\nu_{\rm QPO1}^{\rm max}$ is independent of
luminosity. References 1: Ford et al. \shortcite{Ford97a}, 2:
M{\'e}ndez et al. \shortcite{Mendez97a}, 3: M{\'e}ndez et
al. \shortcite{Mendez98a}, 4: Yu et al. \shortcite{Yu97a}, 5: Yu et
al. \shortcite{Yu97b}, 6: Kaaret et al. \shortcite{Kaaret97a}, 7:
van der Klis \shortcite{VanDerKlis97a}, 8: Strohmayer et
al. \shortcite{Strohmayer98b}, 9: Strohmayer et
al. \shortcite{Strohmayer96a}, 10: Wijnands \& van der Klis
\shortcite{Wijnands97a}, 11: Smith, Morgan \& Bradt
\shortcite{Smith97a}, 12: Wijnands et al. \shortcite{Wijnands98b},
13: Ford et al. \shortcite{Ford98c}, 14: Zhang et
al. \shortcite{Zhang98c}, 15: Zhang et al. \shortcite{Zhang98a}, 16:
Strohmayer \shortcite{Strohmayer98a}, 17: Ford, van der Klis \&
Kaaret \shortcite{Ford98a}, 18: van der Klis \& Wijnands
\shortcite{VanDerKlis97b}, 19: Wijnands et
al. \shortcite{Wijnands98d}, 20: Wijnands et
al. \shortcite{Wijnands97b}, 21: Wijnands et
al. \shortcite{Wijnands98a}, 22: Zhang, Strohmayer \& Swank
\shortcite{Zhang98b}, 23: Jonker et al. \shortcite{Jonker98a}}
\end{table*}
In Table \ref{tab:qpo.data} the observations of kilohertz QPOs from 16
sources are summarised. The sources can be classified into two
classes, the Atoll- and the Z-sources, depending on the form of their
colour-colour diagram. Almost all sources show simultaneously two
kilohertz QPOs. Although the frequencies of both QPOs vary by several
hundred Hertz with the accretion rate, the variation of the frequency
separation is only small.
Therefore the observations strongly favour some kind of beat-frequency
model. In such models the upper kilohertz QPO corresponds to the
Kepler rotation at the inner edge of the accretion disk. The lower QPO
corresponds to the beat frequency between the upper QPO and the spin
of the neutron star. This interpretation is supported by the fact that
the frequency separation is equal to the QPO frequency (or half of it)
in type I X-ray bursts observed in some of the Atoll sources (see
Tab. \ref{tab:qpo.data}). The question remains however where the upper
QPO is generated. Strohmayer et al. \shortcite{Strohmayer96a}
suggested that this radius is identical to the magnetospheric radius,
whereas Miller et al. \shortcite{Miller96a} propose that it is
identical to the sonic radius (see van der Klis 1997
\nocite{VanDerKlis97a} for a critical discussion of these models). In
both cases, the orbital radius has to be larger than the radius of the
innermost stable orbit (see Sect. \ref{ssec:stable}), or equivalently,
the frequency $\nu_{\rm QPO1}$ of the upper QPO has to be smaller than
the Kepler frequency at the innermost stable orbit:
\begin{equation}
\nu_{\rm QPO1} \leq \nu_{\rm K}^{\rm ms}
\approx 2200 \frac{M_\odot}{M}~{\rm Hz},
\end{equation}
where the approximate expression \eqref{eq:appr.Kms} was used. The
exact expression depends on the EOS and on the spin period of the
neutron star. This inequality sets an upper limit to the mass of the
star. Obviously, the orbital radius has to be larger than the star's
radius, which is larger than the radius of the innermost stable orbit
for low star masses. This constraint sets a lower limit to the star
mass which again depends on the spin period and the EOS.
\begin{figure} \centering
\psfig{figure=fig3.ps,height=\linewidth,angle=-90}
\caption{Maximally allowed Kepler frequency $\nu_{\rm K}$ as function
of the neutron star's mass $M$ for the two EOSs RHF$_{\rm pn}$\ and RHF8
and two different spin frequencies $\nu_{\rm s}=275$ and 355~Hz.}
\label{fig:qpo3}
\end{figure}
Figure \ref{fig:qpo3} shows the maximally allowed Kepler frequency as
function of the gravitational mass for two EOSs, RHF$_{\rm pn}$\ and RHF8. If
the radius of the innermost stable orbit is smaller than the star's
radius the maximally allowed Kepler frequency is given by the Kepler
frequency at the star's equator. This kind of solution is represented
by the rising branch of $\nu_{\rm K}(M)$, since $\nu_{\rm K}$
increases with increasing gravitational mass. If the innermost stable
orbit is located outside the star the maximally allowed Kepler
frequency decreases again with increasing mass. The two EOSs differ
for $M\gtrsim 1.0\,M_\odot$ in the composition. In RHF8 hyperons are
included, whereas RHF$_{\rm pn}$\ assumes pure nucleonic matter. RHF8 is
therefore softer than RHF$_{\rm pn}$\ at high densities. This means also, that
RHF8 can only support smaller masses than RHF$_{\rm pn}$. The maximally
allowed Kepler frequency depends only slightly on the neutron star's
spin frequency, as long as the spin frequency is not larger than about
ten percent of the star's Kepler frequency. If one compares the
$\nu_{\rm K}$--curves for example with the frequency $\nu_{\rm
QPO1}^{\rm max}=1060$ of the highest kilohertz QPO observed in
4U\,1820-30 ($\nu_{\rm s}=275$~Hz), one obtains that the mass of
the neutron star has to be within the range between $M=1.0\,M_\odot$
and $2.25\,M_\odot$ (for RHF$_{\rm pn}$) and $M_{\rm max}=1.65\,M_\odot$ (for
RHF8), respectively.
If one adopts the interpretation of the sonic-point model one expects
that the inner boundary of the accretion disc is close to the innermost
stable orbit. Indeed, the observations of the sources 4U\,1820-30, 4U
1608-52, and 4U\,1636-536 seem to confirm this interpretation, since
the QPO frequency remains constant for high mass accretion rates
\cite{Kaaret97a,Zhang98c}. Therefore, only the right intersection
point of the $\nu_{\rm K}(M)$-curve with the line $\nu_{\rm
K}=\nu_{\rm QPO1}^{\rm max}$ is allowed by the
observations\footnote{The left intersection point is unlikely, since
it is expected that the lower and the upper frequency QPO cannot be
observed at the same time if the accretion disk would extend to the
neutron star surface \cite{Zhang97a}. This would be contrary to the
simultanous observation of both QPOs in several sources.}. The mass of
the neutron star in the binary system 4U\,1820-30 can now be
determined. For
RHF$_{\rm pn}$\, the mass is equal to $M=2.25M_\odot$ whereas RHF8 would be
excluded.
\begin{figure} \centering
\psfig{figure=fig4.ps,height=\linewidth} \caption{Range
of masses for which the Kepler frequency at the equator or at the
innermost stable orbit is larger than the highest observed QPO
frequency $\nu_{\rm QPO1}^{\rm max}=1228$~Hz in the source 4U
1636-536 ($\nu_{\rm s}=290$~Hz). The dashed bars refer to the
EOSs for which models with $\nu_{\rm K}^{\rm ms}=\nu_{\rm QPO1}^{\rm
max}$ do not exist. The approximate value $M_{\rm appr}=1.80M_\odot$
is also shown.} \label{fig:qpo2a}
\end{figure}
\begin{figure} \centering
\psfig{figure=fig5.ps,height=\linewidth} \caption{Same as
in Fig. \ref{fig:qpo2a} but for the source 4U\,1728-34 ($\nu_{\rm
QPO1}^{\rm max}=1100$~Hz, $\nu_{\rm s}=355$~Hz, $M_{\rm
appr}=2.01M_\odot$).} \label{fig:qpo2b}
\end{figure}
\begin{figure} \centering
\psfig{figure=fig6.ps,height=\linewidth} \caption{Same as
in Fig. \ref{fig:qpo2a} but for the source 4U\,1820-30 ($\nu_{\rm
QPO1}^{\rm max}=1060$~Hz, $\nu_{\rm s}=275$~Hz, $M_{\rm
appr}=2.07M_\odot$).} \label{fig:qpo2c}
\end{figure}
The figures \ref{fig:qpo2a}--\ref{fig:qpo2c} show the respective
ranges of masses for which the Kepler frequency at the equator or at
the innermost stable orbit is larger than the highest observed QPO
frequency $\nu_{\rm QPO1}^{\rm max}$ for all EOSs considered here (see
Tab. \ref{tab:eos}). The vertical line refers to the approximate mass,
which is obtained by setting $j=0$ in Eq. \eqref{eq:appr.Kms}. This
expression underestimated the upper limit of the mass. The dashed bars
refer to the EOSs for which models with $\nu_{\rm K}^{\rm ms}=\nu_{\rm
QPO1}^{\rm max}$ do not exist, i.e. whose maximally stable mass is to
small. If the interpretation of the highest observed QPO frequency in
the three sources 4U\,1820-30, 4U\,1608-52, and 4U\,1636-536 is
confirmed, the mass of the neutron star is constrained to the
respective right end of the solid bars. The EOSs, for which models
with $\nu_{\rm K}^{\rm ms}=\nu_{\rm QPO1}^{\rm max}$ do not exist
(dashed bars) are excluded.
Especially, the observation of 4U\,1820-30 agrees only with the two
nucleonic EOSs UV$_{14}$+UVII\ and RHF$_{\rm pn}$. If the interpretation of the highest
observed QPO frequency in this source is confirmed by further
observations, the existence of hyperons, meson condensates, quark
condensates, and strange stars can be rejected. The neutron star
matter consists thus only of nucleons and leptons. At least, the
hyperon fraction (i.e. strangeness) is much smaller than predicted by
actual relativistic calculations.
\begin{figure} \centering
\psfig{figure=fig7.ps,height=\linewidth} \caption{Same as
in Fig. \ref{fig:qpo2a} but for different sources and the fixed EOS
RHF$_{\rm pn}$.} \label{fig:qpo6a}
\end{figure}
\begin{figure} \centering
\psfig{figure=fig8.ps,height=\linewidth} \caption{Same as
in Fig. \ref{fig:qpo6a} but for RHF8.} \label{fig:qpo6b}
\end{figure}
Figures \ref{fig:qpo6a} and \ref{fig:qpo6b} show the allowed ranges of
mass of the neutron stars in the respective binaries for the EOSs
RHF$_{\rm pn}$\ and RHF8, respectively. For RHF8, the masses are limited by
the maximally stable star mass $M_{\rm max}\sim 1.65\,M_\odot$, which
depends on the spin period. All sources are consistent with a
canonical star mass of $M=1.4M_\odot$. Only if one assumes that the
highest QPO frequency is equal to the Kepler frequency at the
innermost stable orbit, the masses lie between $M=2.0\,M_\odot$ and
$2.5\,M_\odot$ (for RHF$_{\rm pn}$).
\section{Constraints on Equation of State by Lense-Thirring Precession}
\label{sec:constr2}
\begin{table*}
\centering \caption{Observational data of QPOs in the 10~Hz range. \label{tab:qpo}}
\smallskip
\begin{tabular}{ccccc}
\hline\hline
Source & $\nu_{\rm s}$ [Hz] & $\nu_{\rm K}$ [Hz]
& $\nu_{\rm QPO3}$ [HZ] & References \\
\hline
4U\,0641+91 & 323 & 900 & 22 & Stella \& Vietri \shortcite{Stella97a} \\
\hline
4U\,1608-52 & 233 & 800 & 20 & Yu et al. \shortcite{Yu97a} \\
\hline
& & 355 & 9.0 & \\
4U\,1728-34 & 355 & 551 & 14.1 & Ford \& van der Klis \shortcite{Ford98b} \\
& & 699 & 26.5 & \\
& & 1122 & 41.5 & \\
\hline
KS 1731-260 & 262 & 1197 & 27 & Wijnands \& van der Klis \shortcite{Wijnands97a} \\
\hline
4U\,1735-44 & 326 & 1149 & 29 & Wijnands et al. \shortcite{Wijnands98b} \\
\hline
& & 1050 & 5 & \\
Sco X-1 & 247 & 1101 & 9 & van der Klis \shortcite{VanDerKlis96a} \\
& & 1135 & 13 & \\
\hline
& & 644 & 5 & \\
GX 17+2 & 294 & 832 & 9 & Wijnands et al. \shortcite{Wijnands97b} \\
& & 1042 & 13 & \\
\hline
& & 731 & 5 & \\
Cyg X-2 & 346 & 856 & 9 & Wijnands et al. \shortcite{Wijnands98a} \\
& & 1007 & 13 & \\
\hline
& & 569 & 5 & \\
GX 340+0 & 325 & 730 & 9 & Jonker et al. \shortcite{Jonker98a} \\
& & 823 & 13 & \\
\hline
& & 506 & 5 & \\
GX 5-1 & 298 & 685 & 9 & Wijnands et al. \shortcite{Wijnands98d} \\
& & 889 & 13 & \\
\hline\hline
\end{tabular}\end{table*}
Stella \& Vietri \shortcite{Stella97a} suggested that a third oscillation
with a frequency $\nu_{\rm QPO3}$ around $\sim 10$~Hz is also produced
at the inner border of the accretion disk by Lense-Thirring
precession. The observed frequencies $\nu_{\rm QPO3}$
(s. Tab. \ref{tab:qpo}) can again be compared with theoretical neutron
star models. The neutron star models are constructed for a given spin
frequency $\nu_{\rm s}$ and a fixed, unfortunately unknown, star mass
$M$. The obtained monotone relations $\nu_{\rm K}(r,\theta=\pi/2)$ and
$\nu_{\rm LT}(r,\theta=\pi/2)$ are transformed into the relation
$\nu_{\rm LT}(\nu_{\rm K})$. Since the Lense-Thirring precession
frequency $\nu_{\rm LT}$ is, in first approximation, proportional to
$\nu_{\rm s}$ the relation $\nu_{\rm LT}/\nu_{\rm s}$ as function of
$\nu_{\rm K}$ is shown in Fig. \ref{fig:rot.qpo} for all EOSs. The
observations are shown as circles. The curves depend only weakly on
the neutron star mass. One can therefore draw conclusions from the
comparison of the theoretical curves with the observations, although
the star masses are unknown. Since both $\nu_{\rm LT}/\nu_{\rm s}$ and
$\nu_{\rm K}$ do not depend on $\nu_{\rm s}$ in first approximation,
the calculation of the theoretical curves for one specific spin
frequency, $\nu_{\rm s}=363$~Hz (this corresponds to the spin
frequency of the neutron star in 4U1728-34), is sufficient.
\begin{figure*} \centering
\psfig{figure=fig9.ps,height=\linewidth,angle=-90} \caption{Ratio
$\nu_{\rm LT}/\nu_{\rm s}$ of the Lense-Thirring precession
frequency $\nu_{\rm LT}$ ant the spin frequency $\nu_{\rm s}\equiv
363$~Hz as function of the Kepler frequency $\nu_{\rm K}$. The used
mass is $M=1.4M_\odot$. The first group of curves contains the EOSs:
NLSH1, NLSH2, TM1-m2, TM1-m1, {\bfRHF$_{\rm pn}$}, G$_{300}$, HV$_{\rm pn}$, and HV
(from the left to the right; the bold typed label corresponds to the
bold curve). Group 2: G$^{\rm K240}_{\rm pn}$, {\bf RHF8}, RH1, G$^\pi_{300}$, G$^{\rm K240}_{\rm M78u}$, and
G$^{\rm K240}_{\rm M78}$. Group 3: SM$_{\rm B145}$, TF, {\bf UV$_{14}$+UVII}, UV$_{14}$+TNI, and G$^{\rm K240}_{\rm B180}$. Group 4: {\bf
RHF1} and SM$_{\rm B160}$. The full (open) circles correspond to (half of)
the observations of the Atoll-sources in Tab. \ref{tab:qpo}.}
\label{fig:rot.qpo}
\end{figure*}
As can be seen in Fig. \ref{fig:rot.qpo}, the observations of
KS\,1731-260 and 4U\,1735-44 agree with the theoretical curves of
group 1, which contains the stiffest EOSs of our sample. This is in
agreement with the results of Stella \& Vietri
\shortcite{Stella97a}. However, the other observations, including the
observations of the Z-sources which are not shown, lie above all
theoretical curves. The observed frequency $\nu_{\rm QPO3}$ is thus
higher than the Lense-Thirring precession frequency even for the
stiffest EOSs. If one assume that the observed frequencies correspond
to the first overtone of the Lense-Thirring precession, the
Atoll-source observations are in the range of the theoretical curves.
In the case of the Z-sources, the detected frequencies $\nu_{\rm
QPO3}$ are not only too large in most sources, but additionally the
slope of the relation $\nu_{\rm QPO3}(\nu_{\rm QPO1})$ of the Z-source
observations is much higher than the slope of $\nu_{\rm LT}(\nu_{\rm
K})$ for the theoretically determined models. Even the assumption,
that the first overtone of the Lense-Thirring precession frequency
$\nu_{\rm LT}$ is detected \cite{Stella97a} cannot explain these
discrepancies.
\section{Conclusions} \label{sec:concl}
In this paper, we derived models of rapidly rotating neutron stars and
strange stars by solving the general relativistic structure equations
for a broad collection of modern EOSs. We compared the space time
geometry of these models with recently discovered QPOs in the X-ray
brightness of LMXBs.
If one follows the general beat-frequency interpretation of the
kilohertz-QPOs, i.e. that the higher frequency QPO originates at a
stable circular orbit, one can constrain the mass of the neutron star
to a range which depends on the EOS. This mass range is for all sources
and for all EOSs consistent with a canonical mass, $M=1.4\,M_\odot$. As
it was stated by Miller et al. \shortcite{Miller98b} and Thampan et
al. \shortcite{Thampan98a} the exact lower and upper limits of the neutron
star's mass can only be determined by using fully relativistic models
of rapidly rotating neutron stars. The exact limits differ from the
approximations with $j=0$ by roughly 10~\%.
As it was shown by Kaaret et al. \shortcite{Kaaret97a} and Zhang et
al. \shortcite{Zhang98c}, the observation of a maximum frequency $\nu_{\rm
QPO1}^{\rm max}$ of the high frequency QPO in the sources 4U\,1820-30,
4U1608-52, and 4U\,1636-536 favour the interpretation that this QPO
originates at the innermost stable orbit. If this interpretation is
correct, the mass of the neutron star can be exactly (within the
observational errors) determined for a given EOS. The approximately
obtained mass, $M=2.07\,M_\odot$, of the source 4U\,1820-30 is larger
than the maximum mass of most of the considered EOSs. This conclusion
is even strengthened if the observed maximum frequency $\nu_{\rm
QPO1}^{\rm max}$ is compared with the exact neutron star models. The
only allowed EOSs of our broad collection are then the UV$_{14}$+UVII\ and
RHF$_{\rm pn}$, which both describe neutron star matter as consisting of
nucleons and leptons only. The derived masses of the three sources lie
in the narrow range between $M=1.92\,M_\odot$ and $2.25\,M_\odot$.
This result is also of some importance for the nature of the object
left in the supernova 1987A. During the first ten seconds after the
supernova, neutrinos were detected \cite{Chevalier97a}. This means
that a protoneutron star was formed in the supernova. The fact that up
to now no pulsar emission could be detected was interpreted by Bethe
\& Brown \shortcite{Bethe95a} that the protoneutron star collapsed to a
black hole when the star became transparent to neutrinos after roughly
10~s. The estimated value of the baryonic mass $M_{\rm B}\sim
1.63-1.76\,M_\odot$ \cite{Bethe95a,Thielemann94a} gives thus an upper
limit to the maximum gravitational mass of a neutron star: $M_{\rm
max}\lesssim 1.6\,M_\odot$. This limit is in clear contradiction to
the derived mass of, e.g. 4U\,1820-30.
It is generally believed that neutron stars are born with a mass about
$1.4-1.5\,M_\odot$. Neutron stars in LMXBs would therefore accrete
$0.4-0.8\,M_\odot$ during their lifetime. It seems reasonable to
assume that some of the neutron stars in LMXBs accrete enough matter
to reach the maximally stable mass ($M_{\rm max}=2.20\,M_\odot$ for
UV$_{14}$+UVII, $M_{\rm max}=2.44\,M_\odot$ for RHF$_{\rm pn}$). The neutron star would
then collapse to a black hole.
The interpretation of the QPO with frequencies $\nu_{\rm QPO3}$ about
10~Hz as Lense-Thirring precession frequency of the accretion disk
seems not to be consistent with the theoretical star models, unless
one assumes that the first overtone of the precession is observed. In
the case of Z-sources however, the necessary ratio of the frame
dragging frequency and spin frequency, and thus the moment of inertia,
of the models if too small compared to the observed frequency
$\nu_{\rm QPO3}$ or half of it.
Our general conclusion is that the observations of kilohertz QPOs in
LMXBs provide us another powerful tool for probing the interior of
neutron stars. Compared to the other tools like observations of the
maximum mass \cite{Kerkwijk95a}, the limiting spin period
\cite{Friedman86a,Weber92a}, and cooling simulations
\cite{Tsuruta66,Schaab95a,Page97a}, the derived constraints,
especially in the sonic-point-interpretation, are rather strong. The
lower limit $M_{\rm max}\gtrsim 2.15\,M_\odot$ is only consistent with
two EOSs, UV$_{14}$+UVII\ and RHF$_{\rm pn}$, which are relatively stiff at high
densities. Their stiff behaviour at high densities seems to be only
possible if the neutron star matter consists of neutrons, protons, and
leptons only. At the most, a small admixture of hyperons may be
allowed. However, one has to admit that such a composition somehow
contradicts our conception of neutron star matter, since
fieldtheoretical models lead more or less inevitably to a more
complex composition at high density. If the given interpretation is
correct, one has therefore to reinvestigate the problem of super
dense neutron star matter by dropping, for instance, the standard
assumptions about the coupling of the hyperons in such regimes.
This specific conclusion depends however on the interpretation of the
maximum frequency of the kilohertz QPO within the
sonic-point-model. As a kind of beat-frequency model, this model
predicts a constant frequency separation $\Delta\nu=\nu_{\rm
QPO1}-\nu_{\rm QPO2}$. Recently two further examples, where this
constancy is not observed, has been discovered: 4U\,1608-52 and
4U\,1735-44 \cite{Mendez98b,Ford98c}. Moreover, the frequency
separation in the Atoll source 4U\,1636-536 seems not to be consistent
with the half of the frequency of the QPO in type I bursts
\cite{Mendez98c}. Though the beat-frequency models are the most
hopeful candidates in explaining the QPO-phenomenology it has to be
awaited how the variation of the frequency separation and deviation
from the QPO frequency in bursts can be incorporated to these models.
\section*{Acknowledgements}
One of us (Ch.~S.) gratefully acknowledges the Bavarian State for
financial support. We would like to thank Norman Glendenning and
J{\"u}rgen Schaffner-Bielich for providing us tables of their EOSs.
|
\section*{Introduction}
Since their discovery, active galactic nuclei (AGN) have stood out as
uniquely luminous objects in the Universe. We are fairly confident
that their ultimate power source is the release of gravitational energy
sustained by an accretion disk, which feeds matter directly onto a
supermassive black hole. Evidence for the accretion mechanism is
found in X-ray-bright Seyfert galaxies, which have broad Fe K$\alpha$ lines
indicative of radiatively efficient, geometrically thin accretion disks
extending down to the radius of marginal stability\cite{nandra97a,tanaka95}.
However, the mechanisms to produce electromagnetic radiation from
the accreting material and the manner in which this material
actually reaches the black hole are still unclear. Understanding the
accretion mechanism is important because nuclear activity
in galaxies is common (perhaps more common than previously thought) and
therefore accretion is likely to play a fundamental role in the
energetics of the Universe.
The Universe is populated with many classes of AGN,
the most powerful of which are
QSOs ($L_X \sim 10^{46-48}$ erg s$^{-1}$).
Their high luminosities make them observable at great distances,
thereby providing ways to obtain
fundamental information about the formation and
subsequent evolution of galaxies. As we look toward the early Universe,
studies of QSOs can help answer these
important questions:\hfill\break
$\bullet$ Do all galaxies contain massive black holes and
what roles do AGN play in the formation and evolution
of galaxies?\hfill\break
$\bullet$ How does the accreting material
make its way into the surroundings of the black hole,
and how is this material fed directly into the black hole?\hfill\break
$\bullet$ Is there a change
in the accretion efficiency or accretion rate with z?\hfill\break
$\bullet$ Are we seeing the flaring of short lived QSO events in
many nuclei or a slow decline in a few nuclei that have been
QSOs from the start?\hfill\break
Or, in observational terms:
{\it In what ways do AGN exhibit spectral evolution?}\hfill\break
X-ray observations are of particular value because they provide
a powerful diagnostic of the environs of the accretion flow and
a powerful means for tracing evolution. Variability studies show
that the X-ray continuum emission in AGN originates on the spatial
scales we are most interested in --
close to the black hole. X-rays can also
penetrate large amounts of gas and
dust in which some active nuclei are embedded. Moreover,
X-ray emission appears to be a universal property of QSOs\cite{at86},
which allows us to trace their properties out to high redshift.
The major limitation of X-ray studies is the need for
different instruments to cover the entire X-ray continuum range.
Unbiased and statistically valid X-ray samples of QSOs have been
difficult to obtain, but this will change with the
next generation of X-ray observatories,
beginning with the launch of $\it Chandra$ and $XMM$ in 1999.
This article evaluates the current status of X-ray spectroscopy studies of
QSOs, with emphasis on objects whose spectra appear to be dominated
by accretion mechanisms rather than jet/beaming mechanisms.
\section*{X-ray Continuum Properties of QSOs}
\begin{table}[b!]
\caption{The X-ray Spectral Properties of QSOs}
\label{table1}
\begin{tabular}{lccccccrr}
Radio & Sample& z$^a$ &Energy& Instr.&$<\Gamma_{\rm X}>$& Intrinsic$^b$
&Comments$^c$&Ref.\\
class & size & & [keV] & & & N$_{\rm H}$? & & \\
\tableline
RQQ &42 &$0.12\pm0.05$& $0.1-2.4$ & $Rosat$ &$2.56^{+0.10}_{-0.11}$&
no & $^d$ & \cite{swft96}\\
& 9 &$0.3\pm0.03$ & $0.1-2.4$ & $Rosat$ &$2.47\pm0.33$ & no &
$^d$ & \cite{swft96} \\
&19 &$0.19\pm0.08$& $0.1-2.4$ & $Rosat$ &$2.72\pm0.09$ & no &
$^d$ & \cite{l97} \\
&390&$0\rightarrow2.5$ & $0.1-2.4$ & $Rosat$ &$2.58\rightarrow2.22$
& ---& $^d$ & \cite{ybsv98}\\
&16 &$0.18\pm0.21$& $0.3-3.5$ &$Einstein$&$1.91^{+0.67}_{-0.36}$&
no & s. excess (7) & \cite{we87}\\
&12 &$0.076\pm0.04$& $0.1-10$ &$Exosat$ &$2.18\pm0.35$ & no & s.
excess (5) & \cite{comastri92}\\
& 9 &$0.54\pm0.47$&$0.5-10$ &$ASCA$ &$1.93\pm0.06$ & yes
(3) &Fe K (5) $^d$ & \cite{reeves97} \\
& 5 &$2.1\pm0.13$ & $2-10$ &$ASCA$ &$1.68\pm0.09$ & no &
FeK (1) $^d$ & \cite{v98} \\
& 7 &$0.13\pm0.08$& $2-20$ &$Ginga$ &$1.90\pm0.38$ & ---&
& \cite{w92} \\
RLQ &65 &$0.08\rightarrow2.3$& $0.1-2.4$ &$Rosat$
&$2.52\rightarrow1.87$ & no & $^d$ & \cite{swft96} \\
& 4 &$0.27\pm0.11$& $0.1-2.4$ &$Rosat$ &$2.15\pm0.14$ & no &
$^d$ & \cite{l97}\\
& 4 &$3.16\pm0.23$& $0.1-2.4$ &$Rosat$ &$1.71\pm0.08$ &yes (3)
& $^d$ & \cite{e94} \\
& 9 &$2.56\pm0.8$ & $0.1-2.4$ &$Rosat$ &$1.53\pm0.06$ &yes (2) &
$^d$ & \cite{c97}\\
&17 &$0.34\pm0.19$& $0.3-4.5$ &$Einstein$& $1.48^{+0.63}_{-0.36}$ &
no & & \cite{we87} \\
& 5 &$0.27\pm0.22$& $0.1-10$ &$Exosat$ & $1.79\pm0.19$ & no & s.
excess (1) & \cite{comastri92} \\
& 3 &$2.3\pm0.83$ & $0.5-10$ &$ASCA$ &$1.67\pm0.20$ &yes
(1) & $^d$ & \cite{s96} \\
&15 &$2.42\pm1.29$& $0.5-10$ &$ASCA$ &$1.63\pm0.04$ &yes
(9)&Fe K (2) $^d$ & \cite{reeves97}\\
& 9 &$2.56\pm0.8$ & $0.5-10$ &$ASCA$ &$1.61\pm0.04$ &yes
(6) & $^d$ & \cite{c97} \\
& 6 &$0.38\pm0.3$ & $2-20$ &$Ginga$ & $1.71\pm0.16$ & ---&
$^d$ & \cite{w92}\\
\end{tabular}
Notes: $^a$Mean redshift and standard deviation except for cases with
arrows, which indicate the range of z. $^b$Indicates
whether absorption in excess of the Galactic value is present.
Parenthesis contain the number of objects.
$^c$Indicates cases for which excess soft X-ray emission or
Fe K$\alpha$ emission is detected.
Parenthesis contain the number of objects.
$^d$Denotes data that are plotted in Figure 1.
\end{table}
The X-ray spectral characteristics of QSOs
for redshifts up to z $\sim3$ are summarized in Table 1. All
results derive from spectral modeling techniques, which assume that the
QSO spectrum in a given energy band can be modeled with an absorbed power
law having a photon index $\Gamma$ ($N(E)\propto$ E$^{-\Gamma}$). For cases
where the line-of-sight absorption is consistent with that due to
our Galaxy ($N_{\rm Hgal}$),
the fits have $N_{\rm H}$ fixed at $N_{\rm Hgal}$. For cases where
$N_{\rm H}$ is significantly larger than $N_{\rm Hgal}$, $N_{\rm H}$ is
left as a free parameter. In this article, the term ``soft X-ray''
is loosely applied to photon energies between 0.1 and 2.4 keV and the
term ``hard X-ray'' is loosely applied to
photon energies between 2 and 10 keV.
The energy bands of the experiments listed in Table 1
overlap and so only the
photon indices that are strongly weighted toward soft or hard energies
are discussed.
When photon index is plotted against redshift (Figure~\ref{firstfigure}),
it is apparent that $\Gamma$ decreases with increasing z.
For RQQs, $\Gamma$(soft) ranges from $\sim2.6$ at low z to
$\sim2.2$ at high z ($\Delta\Gamma=0.4$ from $z=0.1\rightarrow2$)
while $\Gamma$(hard) changes only slightly with z, from $\sim1.9$ to
$\sim1.7$. For RLQs, $\Gamma$(soft) decreases by a larger amount from
$\sim2.5$ at low z to $\sim1.7$ at high z ($\Delta\Gamma=0.8$
from $z=0.1\rightarrow3$), while $\Gamma$(hard) remains
fairly constant with z. In addition, the soft X-ray index
is related to
radio loudness in the sense that RLQs have systematically
smaller values of $\Gamma$(soft) than RQQs.
At high energies, the spectral shapes are similar
implying a common emission mechanism and minimal
spectral evolution.
What about possible selection biases?
The set of observations that most likely contain
a selection bias is the $ASCA$ sample of high-z
RQQs\cite{v98}.
For a given optical luminosity, RQQs are $\sim3$ times less
luminous in X-rays than RLQs\cite{z81} and so only the most
luminous RQQs have reliable hard X-ray data, especially at high z.
For a given distribution of $\Gamma$(hard), $ASCA$ may be
biased towards detecting objects with small $\Gamma$(hard).
On the other hand, these objects still
possess fairly steep soft X-ray slopes\cite{ybsv98}. Since
these objects have both steep low-energy spectra {\it and}
flat high-energy spectra,
the trend seen with $ASCA$ probably does represent
RQQs at high z.
The dichotomy of spectral
indices is robust and provides strong evidence for two distinct
emission mechanisms, one at low energies
that dominates the spectra up to z $= 1-2$, and one at high
energies that is approximately independent of z.
A distinct energy for this spectral
``break'' toward low energies (obviously a larger effect
in RQQs than in RLQs) has not
been found, but it is probably somewhere between 0.5 and 1 keV
for low-z objects\cite{comastri92}.
To explain the spectral changes with z,
RLQs must have their soft
component shifted out of the $Rosat$ band by z = 2, beyond
which $\Gamma$ reaches its redshift-independent
value\cite{bys97}. RQQs, on the other hand,
have not yet displayed the point at which the soft component is
shifted out of the $Rosat$ band, and so the spectral break must
occur at a higher energy compared to RLQs (Figure 1).
Depending on how the soft and hard X-ray
components are normalized,
either the soft X-ray emission is enhanced
in RQQs relative to RLQs or the
hard component is enhanced in RLQs relative to RQQs.
The fact that RLQs are the stronger X-ray sources
implies the latter.
\section*{Photoelectric Absorption}
\begin{table}[b!]
\caption{X-ray Properties of a Representative Sample of QSOs at z $>2$}
\label{table2}
\begin{tabular}{lcccccccc}
Quasar & z & $\Gamma_x$ & N$_{\rm H}^a$ & N$_{\rm Hgal}$ & log
$L_x$
& Radio & Fe K$\alpha$ EW$^b$ & Ref. \\
& & & [$10^{20}$ cm$^{-2}$] & [$10^{20}$ cm$^{-2}$] &[ergs s$^{-1}$]
& Class & [eV] & \\
\tableline
S5 $0014+81$ & 3.38 & $1.7\pm0.07$ & $27.4\pm4$ &
13.9
& 47.8 & RLQ & $<70$ & \cite{c97} \\
& & & $554^{+196}_{-170}$ (qf) & & & & & \\
$0040+0034$ & 2.00 & $1.79^{+0.15}_{-0.14}$ & $9.7^{+5.5}_{-5.2}$&
2.45
& 46.4 & RQQ & --- & \cite{v98} \\
PKS $0237-233$ & 2.22 & $1.68\pm0.06$ & $122^{+62}_{-54}$ (qf) &
2.39
& 46.8 & RLQ & $<32.2$ & \cite{reeves97} \\
$0300-4342$ & 2.30 & $1.69^{+0.24}_{-0.13}$ & $<6.8$ &
1.83
& 46.0 & RQQ & --- & \cite{v98} \\
Q $0420-388$ & 3.12 & $2.24\pm0.48$ & $34\pm33$ (qf)
& 1.91
& 46.9 & RLQ & --- & \cite{e94} \\
PKS $0438-436$& 2.85 & $1.5\pm0.1$ & $5.8^{+2.6}_{-1.4}$&
1.47
& 47.0 & RLQ &$<240$ & \cite{c97} \\
& & & $72^{+44}_{-22}$ (qf) & & & & & \\
PKS $0528+134$ & 2.07 & $2.64\pm0.07$ & $420\pm90$ (qf) &
23.0
& 47.1 & RLQ & $119\pm58$ & \cite{reeves97} \\
PKS $0537-286$& 3.11 & $1.4\pm0.1$ &$2.8^{+1.0}_{-0.7}$ &
1.95
& 47.3 & RLQ &$<139$ & \cite{c97} \\
& & & $16.7^{+19.3}_{-14.6}$ (qf) & & & & & \\
S5 $0836+71^c$ & 2.17 & $\sim1.5$ & $\sim3.3\pm0.7$ &
2.78
& 47.5 & RLQ &$<110$ & \cite{c97} \\
$1101-264$ & 2.15 & $2.19^{+0.58}_{-0.47}$ &$19.5^{+19.5}_{-
15.5}$& 5.68
& 45.8 & RQQ &$690\pm560$ & \cite{v98} \\
$1255+3536$ & 2.04 & $1.59^{+0.12}_{-0.11}$ & $<6.6$ &
1.22
& 46.3 & RQQ & --- & \cite{v98} \\
$1352-2242$ & 2.00 & $1.66^{+0.25}_{-0.23}$ & $<17.9$ &
5.88
& 46.0 & RQQ & --- & \cite{v98} \\
$1422+231$ & 3.62 & $1.68\pm0.14$ & $<151$ (qf) &
2.9
& 47.0 & RLQ & $<263$ & \cite{reeves97} \\
RXJ $1430.3+4203$&4.72 & $1.29\pm0.05$ & $<5.3$ &
1.4
& 47.1 & RLQ & $<100$ & \cite{fabian98} \\
Q $1508+571$ & 4.30 & $1.43\pm0.08$ & $<477$ (qf) &
1.34
& 47.2 & RLQ & $<156$ & \cite{reeves97} \\
PKS $1614+051$ & 3.21 & $1.43\pm0.14$ & $<250$ (qf) &
5.0
& 46.6 & RLQ &$<132$ & \cite{reeves97} \\
Q $1745+624$ & 3.89 & $1.68\pm0.25$ & $21^{+15}_{-14}$ &
3.4
& 47.0 & RLQ & $<180$ & \cite{k97} \\
& & & $61^{+108}_{-59}$ (qf) & & & & & \\
PKS $2126-158$ & 3.28 & $\sim1.6\pm0.1$ & $10.4^{+3.1}_{-2.3}$&
4.85
& 48.0 & RLQ &$<107$ & \cite{c97} \\
& & & $120^{+80}_{-50}$ (qf) & & & & & \\
PKS $2149-306$ & 2.35 & $1.54\pm0.05$ & $8.3^{+1.8}_{-2.2}$&
1.91
& 47.8 & RLQ & $<85$ & \cite{c97} \\
PKS $2351-154^d$ & 2.67 & 1.92(fixed) & $12.7^{+3.7}_{-
2.9}$& 2.18
& --- & RLQ & --- & \cite{schartel97} \\
& & & $200^{+90}_{-60}$ (qf) & & & & & \\
\end{tabular}
Notes:
$^a$(qf) absorption in the QSO frame
as opposed to the observer's frame (all others).\\
$^b$Assumes a narrow Gaussian line at 6.4 keV in the
QSO rest frame.\\
$^c$Absorption changes by $\Delta$N$_{\rm H}\sim8\times10^{20}$ in 0.8 yr.\\
$^d$Absorption changes on timescale of $<0.41$ yr in the QSO frame.\\
\end{table}
Characteristics for QSOs at z $>2$ that have
high-quality X-ray data are listed in
Table 2\footnote{Not necessarily a complete sample.}.
More than 1/2 of the RLQs possess absorption
in excess of the Galactic value.\footnote{A strong case
for absorption is made by Fiore et al. (1998) \cite{fiore98}.
An alternative but less favored explanation
is downward curvature in
the intrinsic spectrum, such as that resulting from emission
dominated by synchrotron losses.}
In contrast, RQQs and low z quasars lack significant
amounts of intrinsic absorption, with
a handful of RQQs, such as PG 1114+445 \cite{george97}
showing evidence for ionized absorption.
The absorption in RLQs can be as much as
$\sim 5 \times10^{22}$ cm$^{-2}$ in the quasar frame,
which is similar to that seen in intermediate-type
Seyfert galaxies such as NGC 4151. However, the
physical properties of the absorbers are not well known
because the X-ray data are ambiguous.
In some cases, data at other wavebands support
the contention that the absorption is
physically associated with the quasar\cite{e98}, while
in other cases, the data favor absorption at low
z \cite{s94}. If the absorber is intrinsic to the
quasar, it could be nuclear material as in low-z, low-$L_x$
objects or it could exist on a
larger scale such as the host galaxy (or protogalaxy).
Because of the uncertainties associated
with the properties of the absorbing material in
high-z RLQs, some general spectral trends are not clear.
For example, does the shape of the intrinsic spectrum
depend on the luminosity of the source?
Other QSO studies have not found
a significant correlation between $\Gamma$ and L$_X$
except for the general trend that
RLQs have smaller X-ray indices than
RQQs\cite{reeves97,w92,c97}. Table 2 also
shows no correlation between $\Gamma$ and L$_X$, but
$\Gamma$ depends on how the absorption is
handled in the spectral modeling.
Having $N_{\rm H}$ ``wrong'' can
make $\Gamma$ artificially large or small and
so the true values of $\Gamma$ are somewhat uncertain.
The data also do not require that {\it only}
high-z RLQs possess intrinsic absorption.
Indeed, if only the hard X-ray component is
absorbed, this underlying absorption
could remain ``hidden'' in other QSOs. In RQQs, the
stronger soft component might easily swamp the
underlying absorption, regardless
of z. For RLQs, where the soft component
is not as prominent, column densities of
a few $\times10^{22}$ cm$^{-2}$ would flatten the
observed spectrum at $\sim1-2$ keV in the quasar frame.
Such flattening could go undetected in low-z quasars in
the following ways. For $Rosat$, the
flattening falls too near the upper energy range
of the detector. For $ASCA$, which covers a larger
bandpass, flux from the soft component
may average out the flattening so that it is not detected.
On the other hand, for z = $1-2$ the
contaminating soft component is
shifted out of the $ASCA/Rosat$ band and
the absorption cutoff is shifted to around
$\sim0.3-0.6$ keV in the observer's frame, into a region
where it is more easily detectable.
Higher quality spectra at low X-ray
energies are needed to address this question.
It is interesting that QSOs do not show the same evidence for
spectral flattening at energies $>10$ keV \cite{v98,c97}
that is common in Seyfert
galaxies\cite{np94} and is the signature of
Compton reflection.
Fe K$\alpha$ emission is also weaker and/or
much less common in quasars and QSOs than in
Seyfert galaxies and nearby radio-loud
objects like 3C 120 \cite{g97}.
The equivalent width of
the Fe K$\alpha$ line decreases with increasing
luminosity from log $L_X = 41.7-47.2$ \cite{it95}. This
trend continues to z = 2 \cite{nandra97b} with
RLQs at high z rarely showing
Fe K$\alpha$ emission\cite{c97,reeves97}.
The lack of reflection and Fe K$\alpha$ emission
in QSOs suggests that they possess a different structure in their
accretion flow compared to lower-luminosity galaxies.
The difference may relate to the high luminosities and
high degree of ionization in the their inner regions.
A lack of Fe K$\alpha$ emission may signify complete ionization
of iron atoms in the accretion disk or that the inner parts
of the disk have been completely blown away. High ionization would
also allow the Compton reflection to suffer much less
absorption in the disk. In this case, reflection is
still present but it would have a shape almost indistinguishable
from the continuum source.
For RLQs, the X-ray emission associated with
the jet may simply dominate the spectrum,
rendering spectral features from the accretion flow undetectable.
\section*{Intrinsic X-ray Emission Mechanisms}
The source of the continuum emission in AGN is thought to be
ultimately linked to an accreting supermassive black hole.
Within the vicinity of the black hole, X-rays
can be produced via Compton upscattering of soft photons
off either electron-positron pairs\cite{zdz90},
a population of hot electrons\cite{tl95},
or bulk motion in the accretion
flow\cite{tmk96,tmk97}. The lack of annihilation lines
in Seyfert spectra causes some problems for pair
models\cite{haardt97} and so this discussion focuses
on the latter two mechanisms for X-ray production.
The emitting region is envisioned as
lying somewhere just above the accretion disk, often
represented as a smooth extended corona above
the disk or small-scale flaring regions on the disk surface.
Thermal Comptonization or bulk-motion Comptonization
predict that the distinct observational signature
of a black hole is an emergent power law with a
spectral turnover between $50-500$ keV.
If both processes are occurring, then the changes/differences
in photon index can be explained as a trade-off between
the two. If conditions
are such that bulk motion Comptonization dominates, then
for an accretion disk that orbits a Schwarzschild black
hole with an inner radius extending to the
innermost stable orbit,
$\Gamma$ approaches 2.5 for mass accretion rates $m_{\odot} >> 1$,
where $m_{\odot} = {M_{\odot}}/{(M_{e})_{\odot}}$ and $(M_e)_{\odot}$
is the Eddington rate. If thermal Comptonization dominates,
the spectra are harder, with $\Gamma \ge 1.5$.
This model has been successfully used to explain the spectra of
Galactic black-hole candidates, which show
$\Gamma$ = $1.3-1.9$ in their low (hard)
state and $\Gamma$ = 2.5
in their high (soft) state \cite{ebisawa96}.
For RQQs, where we think accretion mechanisms dominate
the X-ray spectrum, those
with $\Gamma\ge2$ may be cases where we see
X-rays from bulk motion Comptonization without modifications
by ionized absorbers or reflection, i.e., the raw
emergent spectrum from the accreting black hole.
Within the context of the accretion model, the
hard X-ray component in RQQs
can result from thermal Comptonization.
Different mechanisms dominate the spectra of RLQs.
RLQs show a significant correlation between their $0.1-2.4$
X-ray and total radio luminosity at 5 GHz \cite{b95}
and $\Gamma$ decreases with increasing radio loudness \cite{we87}
in the sense that flat spectrum radio quasars (FSRQs), which
are core-dominated,
have flatter X-ray spectra than their steep spectrum radio
quasar (SSRQ) counterparts\cite{bys97,wor87}.
This implies the presence of two emission mechanisms and
has led to the two-component beaming model, which
explains the difference in
observed properties as caused by the relative orientation of the
source axis to our line of sight \cite{bm87}.
In FSRQs, the highly beamed component points toward us
and we see mostly upscattering of low-energy photons off
relativistic electrons in the jet, while in
SSRQs we see mostly the isotropic emission.
\section*{Unification and Evolution}
Unified schemes explain the diversity of
AGN classes as resulting from inclination plus obscuration effects,
e.g., Seyfert type 2 (narrow-line) galaxies are
edge-on Seyfert 1 (broad-line) galaxies.
These models succeed when applied to specific
subsamples of AGN, but a global unification scheme
has proved elusive.
For radio-loud objects, two schemes seem to be required,
one for low luminosity objects and one for high luminosity
objects \cite{browne92}. For RQQs, broad absorption line
(BAL) QSOs can be unified
with nonBAL QSOs through axis orientation,
but there are many arguments against RQQs being
edge-on RLQs, including differences in their radio morphologies and
underlying galaxy hosts \cite{barthel92}. What we
are left with is a fragmented ``unification'' scheme.
It may be more promising to examine evolution
scenarios, i.e., are low luminosity Seyferts
connected to distant and more powerful AGN?
Seyfert galaxies have composite X-ray spectra that consist of an
underlying power law with $\Gamma\sim1.9$,
a Compton reflection tail above $\sim10$ keV, Fe
K$\alpha$ emission\cite{np94}, and significant photoelectric
absorption from neutral and/or ionized gas\cite{r97}.
High luminosity AGN lack significant evidence
of X-ray reprocessing and many lack significant
absorption. An evolutionary scenario that accounts
for the difference requires
QSOs to have much less cold material in their cores
or for the material to be very highly ionized.
As the source becomes less luminous with time,
the ionization decreases and/or more cold material
collects in the galaxy core.
QSOs also differ from Seyferts by (apparently) having
spectral components from two physically
distinct emission mechanisms.
One component is associated with a relativistic jet
while the other is most likely related to the
accretion mechanism. Since processes associated
with accretion are also thought to
produce the continuum emission
in Seyfert galaxies, an evolutionary connection
between RQQs and Seyferts is plausible.
The soft X-ray spectra of RQQs are significantly steeper
than Seyfert galaxies, but are also more
consistent with the predictions of bulk-motion Comptonization.
This would suggest that bulk motion Comptonization is
the more important physical mechanism
in RQQs while thermal Comptonization
of soft photons by a hot corona is the more
important physical mechanism in Seyfert galaxies.
\section*{The total energy output of the Universe}
Until recently, little attention has been given to sources of
energy in the universe that are not directly visible at optical-UV
wavelengths. It now seems probable that most AGN are
heavily absorbed, and that their central engines are primarily visible
via hard X-rays. The energy density of the X-ray background peaks at
$\sim30$ keV. Less than $15\%$ of this total energy density can be accounted for
by the $Rosat$ AGN, which dominate the soft X-ray background. If AGN
comprise the hard X-ray background, then most must have huge absorbing
columns (N$_{\rm H} \sim10^{22} - 10^{25}$ cm${-2}$)\cite{mgf94}.
The importance of the hard ($>10$ keV) X-ray band
for studying the total energy output for these objects cannot
be overemphasized.
A significant fraction of the energy in the Universe may
reside in absorbed AGN \cite{fabian98b} and the total
accretion energy released by these AGN may be comparable to
the energy generated by nuclear burning by the total stellar
population. If most of the accretion in the
Universe is highly obscured, then the emitted power per
galaxy based on optical, UV, or soft X-ray quasar luminosity functions
will be underestimated.
\section*{Summary and Future Prospects}
This article summarizes our current knowledge of the
X-ray spectral properties of QSOs.
X-ray observations from
z = 0.1 to $\sim3$ keV indicate two distinct X-ray
components in their spectra. One component is
soft with $\Gamma\sim2.0-2.5$
and the other is hard with $\Gamma\sim1.5-1.9$.
In the soft X-ray band the spectra flatten with increasing z
while in the hard X-ray band the spectra show
little change with z. RLQs have flatter
X-ray spectra than RQQs with the exception
of high energies at high z, where both have
similar spectral shapes. Accretion mechanisms such
as thermal Comptonization or bulk-motion
Comptonization are the most
likely source of the continuum in RQQs while
jet/beaming mechanisms dominate RLQs.
Significant absorption
is observed in RLQs at high z but the physical properties of
the absorbing material are uncertain.
More sensitive data will place stringent limits on
the absorption cutoffs and properties of the absorbing
gas, the spectral breaks between the soft and hard
X-ray continuum components, and the
signatures of X-ray reprocessing. Considering an evolutionary
scenario, RQQs and Seyfert galaxies are possibly
connected, with a different accretion mechanism
dominating in each.
Results from X-ray experiments such as $ASCA$ for low-z
AGN suggest that much
of the accretion in the Universe is highly obscured.
So far, X-ray observations at intermediate to
high z have shed little light on this question.
Surveys by $XMM$, ABRIXAS, and $Chandra$ (limited to $E<10$ keV)
will probe columns up to a few times 10$^{23}$
cm$^{-2}$; while future missions such as $Constellation-X$
(Valinia, this volume) will probe columns up to 10$^{25}$
cm$^{-2}$ for fluxes as low as
$\sim1\times10^{-14}$ ergs cm$^{-2}$ s$^{-1}$
in reasonable exposure times.
|
\section{Introduction}
The past few years have seen a dramatic evolution in our knowledge of the
history of star formation in optically-selected galaxies covering
the redshift range from $z=0$ to $z\sim 6$ (see, e.g.,
\markcite{madau98}Madau, Pozzetti, \& Dickinson 1998 and references therein;
\markcite{dey98}Dey et al.\ 1998;
\markcite{hu98}Hu, Cowie, \& McMahon 1998;
\markcite{weymann98}Weymann et al.\ 1998;
\markcite{cowie99}Cowie, Songaila, \& Barger 1999).
However, it has long been suspected that much of the
early star formation in galaxies takes place inside shrouds of dust
(e.g.\ \markcite{pp67}Partridge \& Peebles 1967).
While it is possible to attempt to correct the optical star formation
history for the effects of dust (\markcite{heckman98}Heckman et al.\ 1998;
\markcite{pettini98}Pettini et al.\ 1998),
the corrections are far from certain.
However, the importance of absorption and reradiation of light by dust
in the history of galaxy formation and evolution is evident: the
submm extragalactic background light (EBL)
(\markcite{puget96}Puget et al.\ 1996;
\markcite{guider97}Guiderdoni et al.\ 1997;
\markcite{schlegel98}Schlegel et al.\ 1998;
\markcite{fixsen98}Fixsen et al.\ 1998;
\markcite{hauser98}Hauser et al.\ 1998)
has approximately the same integrated energy density as the optical EBL.
Thus, an accurate reconstruction of the global star formation history
versus redshift requires an understanding of both the
rest-frame ultraviolet and the rest-frame far-infrared properties of the
galaxy populations.
Reradiation of stellar light by dust into the
far-infrared produces a thermal emission peak at $\lambda\sim 60-100$-$\mu$m
that is redshifted into the submm for galaxies at $z>1$.
Submm observations are unique in that the strong negative
K-corrections at 850-$\mu$m nearly compensate for
cosmological dimming beyond $z\simeq 1$, thereby making dusty galaxies
almost as easy to detect at $z\simeq 10$ as at $z\simeq 1$
(\markcite{bl93}Blain \& Longair 1993;
\markcite{mcmahon94}McMahon et al.\ 1994).
The revolutionary new camera, SCUBA (Submillimeter Common User
Bolometer Array; \markcite{holland98}Holland et al.\ 1998),
on the 15-m James Clerk Maxwell Telescope
(JCMT)\altaffilmark{2}\altaffiltext{2}{The JCMT is operated by
the Joint Astronomy Centre on behalf of the parent organizations
the Particle Physics and Astronomy Research Council in the United
Kingdom, the National Research Council of Canada, and The Netherlands
Organization for Scientific Research.}
on Mauna Kea resolves the submm EBL into its individual components
through a combination of unparalleled sensitivity and large field-of-view.
Recent SCUBA surveys
(\markcite{smail97}Smail, Ivison, \& Blain 1997;
\markcite{smail98}Smail et al.\ 1998;
\markcite{barger98}Barger et al.\ 1998;
\markcite{hughes98}Hughes et al.\ 1998;
\markcite{eales99}Eales et al.\ 1999;
\markcite{lilly99}Lilly et al.\ 1999;
\markcite{blain99}Blain et al.\ 1999)
have uncovered a substantial population of dusty galaxies
with properties similar to those expected for the distant counterparts to
the most luminous, merging systems observed locally, the ultraluminous
infrared galaxies (ULIGs) (\markcite{sanders96}Sanders \& Mirabel 1996).
In this paper we present our extensive new wide-area SCUBA surveys
of blank regions of sky. We analyze our source counts both empirically
and in terms of semi-analytic models.
\begin{deluxetable}{lcrrccrccc}
\tablewidth{450pt}
\small
\tablenum{1}
\tablecaption{Source Catalog\label{tab1}}
\tablehead{
\colhead{Name} & \multicolumn{3}{c}{RA(2000)} &
\multicolumn{3}{c}{Dec(2000)} & \colhead{$S_{850\mu{\rm m}}$\ (mJy)} &
\colhead{$N_{850\mu{\rm m}}$\ (mJy)} & \colhead{S/N}
}
\startdata
Lockman Hole & 10 & 34 & 2.1 & 57 & 46 & 25 & 5.1 & 0.69 & 7.4 \\
& 10 & 33 & 56.5 & 57 & 47 & 32 & 2.7 & 0.80 & 3.3 \\
SSA13-deep & 13 & 12 & 32.1 & 42 & 44 & 30 & 3.8 & 0.80 & 4.7 \\
& 13 & 12 & 28.0 & 42 & 44 & 58 & 2.3 & 0.61 & 3.8 \\
& 13 & 12 & 25.7 & 42 & 43 & 50 & 2.4 & 0.74 & 3.2 \\
& 13 & 12 & 28.9 & 42 & 45 & 59 & 3.5 & 1.06 & 3.3 \\
SSA13-wide & 13 & 12 & 19.8 & 42 & 38 & 28 & 5.0 & 1.51 & 3.3 \\
& 13 & 12 & 24.9 & 42 & 39 & 56 & 6.1 & 1.83 & 3.4 \\
& 13 & 12 & 26.3 & 42 & 41 & 46 & 6.3 & 1.91 & 3.3 \\
& 13 & 11 & 59.9 & 42 & 40 & 30 & 10.3 & 2.97 & 3.5 \\
& 13 & 12 & 5.1 & 42 & 44 & 34 & 7.8 & 2.27 & 3.4 \\
SSA17 & 17 & 6 & 25.0 & 43 & 57 & 39 & 8.7 & 2.10 & 4.2 \\
& 17 & 6 & 26.0 & 43 & 54 & 40 & 5.2 & 1.32 & 3.9 \\
& 17 & 6 & 37.2 & 43 & 55 & 31 & 4.8 & 1.29 & 3.7 \\
& 17 & 6 & 33.1 & 43 & 54 & 8 & 5.3 & 1.47 & 3.6 \\
& 17 & 6 & 20.3 & 43 & 54 & 4 & 9.6 & 3.09 & 3.1 \\
SSA22 & 22 & 17 & 34.1 & 00 & 13 & 54 & 8.2 & 1.19 & 6.9 \\
& 22 & 17 & 35.2 & 00 & 15 & 32 & 4.9 & 0.93 & 5.3 \\
& 22 & 17 & 36.1 & 00 & 15 & 54 & 4.0 & 1.02 & 4.0 \\
& 22 & 17 & 34.0 & 00 & 15 & 42 & 3.3 & 0.92 & 3.6 \\
& 22 & 17 & 42.0 & 00 & 16 & 2 & 5.0 & 1.62 & 3.1 \\
\enddata
\end{deluxetable}
\section{Observations}
\label{sec:obs}
In \markcite{barger98}Barger et al.\ (1998) we presented our deepest two
SCUBA maps of the Lockman Hole and the Hawaii Deep Survey field SSA13,
the observational details of which can be found in that paper. We
note, however, that due to a bug in the SCUBA observing software
during the time these data were taken (discovered and fixed
27 July 1998), the chopper was started
at the updated (Az,El) components corresponding to the requested (RA,Dec)
chop throw at the beginning of each new 30\ min integration but then
did not update for the remainder of the integration. Consequently, the
off-beams appear smeared in arclets from the initial positions.
Our wide-area SCUBA maps of the Hawaii Deep Survey fields
SSA17 and SSA22 were taken as mosaics during two runs in August 1998
and September 1998 and also during an earlier observing shift in
October 1997 (totaling 16 observing shifts).
Our wide-area map of SSA13 was taken during a run in February 1999
(7 observing shifts).
The maps were dithered to prevent any regions of the sky from repeatedly
falling on bad bolometers. The chop throw during the October 1997 run
was azimuthal. During the runs in 1998 and 1999 the chop throw was fixed at a
position angle of 90\ deg so that the negative beams would appear
45\ arcsec on either side East-West of the positive beam.
The negative images can be restored, thereby increasing the effective
exposure times on most of the field. Every few hours we did a `skydip'
(\markcite{manual}Lightfoot et al.\ 1998) to measure the zenith
atmospheric opacities at 450 and 850-$\mu$m, and we monitored the
225\ GHz sky opacity at all times to check for sky stability.
Fully sampled wavelength maps at 850-$\mu$m were completed
at median optical depths of 0.38 during
the SSA17 and SSA22 runs and 0.15 during the SSA13 run.
We performed pointing checks every hour during the observations
on the blazars 1633+382, 2223-052, or 1308+326.
We calibrated our data maps using
30\ arcsec diameter aperture measurements of the positive beam in
once or twice-nightly beam maps of the primary calibration sources Uranus
and Mars.
We reduced all our data in a standard
and consistent way using the dedicated SCUBA User Reduction Facility
(SURF; \markcite{surf}Jenness \& Lightfoot 1998).
Due to the variation in the density of bolometer samples across the maps,
there is a rapid increase in the noise levels at the very edges;
thus, we clipped the low exposure edges from our images. Our wide-area SCUBA
maps and multiwavelength follow-up data will be presented elsewhere
(Barger \& Cowie, in preparation).
\begin{figure*}[tbh]
\centerline{\psfig{figure=barger.fig1.ps,angle=90,width=6.5in}}
\figurenum{1}
\figcaption[barger.fig1.ps]{
Our 850-$\mu$m source counts (solid squares) with $1\sigma$ error limits
(jagged solid lines) are well described by the power-law
parameterization in Eq.~1 with $a=0$, $\alpha=3.2$, and
$N_0=3.0\times 10^4$\ deg$^{-2}$\ mJy$^{-1}$ (solid line).
The dashed curve shows a smooth extrapolation of our fit to EBL measurements
using the value $a=0.5$. Counts from Blain et al.\ (1999)
(open circles), Hughes et al.\ (1998) (open triangles), and
Eales et al.\ (1999) (open squares) are in good agreement with our data
and the empirical fit. Semi-analytic models A and E from Guiderdoni et al.\
(1998) are shown by the dotted curves.
\label{fig1}}
\end{figure*}
\section{Source Extraction and Completeness}
\label{extract}
The SURF reduction routines arbitrarily normalize all the data maps in
a reduction sequence to the central pixel of the first map; thus, the
noise levels in a combined image are determined relative to the
quality of the central pixel in the first map. To find the
absolute noise levels, we first eliminated the $\ge 3\sigma$
real sources in each field through subtraction of an appropriately
normalized version of the beam profile. We then iteratively adjusted
the noise normalization
until the dispersion of the signal to noise ratio
measured at random positions was approximately one.
This procedure should provide a conservative noise estimate since
it includes both fainter sources and correlated noise.
We scanned the maps at an array of positions separated by
6$''$ to locate all $\ge 2.8\sigma$ (to overselect) sources.
We note that the position determinations in the deep maps are not
affected by off-beam smearing because it is a symmetrical effect.
Using the peaked-up positions, we extracted only the $\ge 3\sigma$ sources.
We subtracted the brighter sources from the maps before extracting the
fainter sources. We calibrated the extracted fluxes, including both positive
and negative beams, to the 30\ arcsec diameter aperture fluxes of the
brightest sources; the measured fluxes are therefore unaffected by the
off-beam smearing.
Our source catalog is given in Table~\ref{tab1}. The noise levels for the
two deep maps are slightly lower than our initial analysis
(\markcite{barger98}Barger et al.\ 1998) because of improvements in
both the SURF reduction routines and our own extraction routines.
These improvements result in an increase in the number of measured
sources for these fields over our previously published results.
To determine the completeness of our source recovery, we assumed the shape
of the \markcite{guider98}Guiderdoni et al.\ (1998) `Scenario E' model
(see \S\ref{disc}) for the counts and simulated the expected sources
by adding appropriately renormalized calibration sources at random
positions to the final maps with the real sources removed.
We then reran our extraction procedure on these images.
Any sources detected at the $\ge 3\sigma$ level were considered
to be recovered.
Using this criterion, the output counts matched the
input counts; thus, no incompleteness correction is required.
We also investigated the completeness using the functional form in Eq.~1
(below) that flattens just below the 2\ mJy region and obtained the
same result.
At the $3\sigma$ level, noise may migrate sources above and below
the threshold, but empirically we have found that the net result,
given the shape of the counts versus flux,
is to leave the overall count determinations unchanged.
Because of non-uniformity across the maps, the flux sensitivity levels and
area coverages are related. For example, at $3\sigma$ limiting flux levels
of 20, 10, 8, 5, 4, 3, and 2.3\ mJy, our area coverages, after combining all
fields, are 141, 122, 104, 43, 31, 18, and 7.7\ arcmin$^2$, respectively. For
comparison, at a 2.1\ mJy $4\sigma$ flux limit, Hughes et al.\ (1998) have
an area coverage of 5.6\ arcmin$^2$.
The combined data of Eales et al.\ (1999) at $3\sigma$ flux limits of
2.8, 2.3, and 1.7\ mJy have areas 21.7, 6.7, and 3.4\ arcmin$^2$,
respectively (private communication from S.~Eales and S.~Lilly); the actual
$1\sigma$ measurement uncertainties given for the 10 and 14-hour sources
in Table~1 of Eales et al.\ (1999) are mostly in the range 0.9--1.1\ mJy.
We present our cumulative 850-$\mu$m source counts per square deg (solid
squares) with the $1\sigma$ errors (jagged solid lines) in Fig.~1.
The cumulative counts are the sum of the inverse areas of all the sources
brighter than flux $S$. We can model the differential counts
as a function of flux in a phenomenological way by fitting
\begin{equation}
n(S)=N_0/(a+S^{\alpha})
\end{equation}
to the data, where $S$ is measured in\ mJy.
Because the data are restricted to the
$>2$\ mJy range, the $a$ parameter makes essentially no difference
to a fit of observed number versus flux, and we take $a=0$ here.
A chi-square fit gives an optimal index $\alpha=3.2$, with
a 95\ per cent confidence range from 2.6 to 3.9, and a normalization
$N_0=3.0\times 10^4\ {\rm deg}^{-2}\ {\rm mJy}^{-1}$.
This fit, illustrated in Fig.~1 with a straight solid line, is
consistent with the absence of $\gg 10$\ mJy sources in our sample.
Over the measured $2-10$\ mJy range, the contribution to
the EBL is $9.3\times 10^3$\ mJy\ {\rm deg}$^{-2}$, which corresponds to
20 to 30\ per cent of the total EBL, depending on whether we adopt the
850-$\mu$m EBL measurement of $4.4\times 10^4$\ mJy\ {\rm deg}$^{-2}$ from
\markcite{fixsen98}Fixsen et al.\ (1998) or
$3.1\times 10^4$\ mJy\ {\rm deg}$^{-2}$
from \markcite{puget96}Puget et al.\ (1996).
Assuming only that the above parameterization provides an appropriately
smooth continuation to fluxes below 2\ mJy,
we can now fit the EBL with a suitable choice of $a$;
for $a$ in the range $0.4-1.0$ the integral of the
differential counts over all fluxes matches the Fixsen et al.\ and
Puget et al.\ EBL measurements.
The shape for $a=0.5$ is shown by the dashed line in Fig.~1.
The open circles with error bars show the cumulative counts
per square degree from the gravitationally-lensed sample of
\markcite{blain99}Blain et al.\ (1999). The Blain et al.\
sample has only one source below 2\ mJy (see last two columns of
their Table~1); at 0.5\ mJy we plot their 2$\sigma$ upper limit.
The open triangles with error bars show the blank field data from
\markcite{hughes98}Hughes et al.\ (1998).
The open squares show the blank field results from
\markcite{eales99}Eales et al.\ (1999); the statistical
uncertainty is only indicated on their brightest flux point since the
remainder of their data have similar uncertainties to those on our data.
The agreement between all the data sets is excellent and follows the
empirical fit obtained from our data and the EBL constraint.
\section{Discussion and Conclusions}\label{disc}
The IRAS satellite made the first observations of dust emission from
external galaxies when it measured the sky in four bands
between 12 and 100-$\mu$m. A wide variety of infrared luminosities were
observed, from normal spirals to ULIGs. The spectral energy
distributions (SEDs) of ULIGs are found to provide reasonable fits to the
SEDs of submm sources (\markcite{ivison98}Ivison et al.\ 1998).
Later in this discussion we will assume that the SED of the `prototypical'
ULIG Arp~220, the spectral shape of which can be represented by a
modified blackbody with emissivity $\lambda^{-1}$ and a dust temperature
of $T=47$\ K, is appropriate for the submm sources.
It has long been debated whether the dust-enshrouded local ULIGs are
powered by massive bursts of star formation induced by violent
galaxy-galaxy collisions or by AGN activity. A recent mid-infrared
spectroscopic survey of 15 ULIGs by
\markcite{genzel98}Genzel et al.\ (1998) found that $70-80$\ per cent
of the sample are dominantly powered by star formation and
$20-30$\ per cent by a central AGN.
Similarly, spectroscopic follow-up studies of a gravitationally
lensed submm sample (\markcite{barger99}Barger et al.\ 1999;
\markcite{ivison98}Ivison et al.\ 1998) indicate that at least
20\ per cent of the sample show some AGN activity. We assume in the
following discussion that a substantial fraction of the submm light
arises from star formation.
A variety of models have been proposed to predict the submm galaxy
population (e.g.\ \markcite{guider98}Guiderdoni et al.\ 1998;
\markcite{blain98}Blain et al.\ 1999;
\markcite{eales99}Eales et al.\ 1999;
\markcite{trentham99}Trentham, Blain, \& Goldader 1999).
\markcite{guider98}Guiderdoni et al.\ (1998) designed a family of
evolutionary scenarios that ranged from submm source populations due
solely to the presence of dust in optical galaxies (Scenario A) to
those due a rapidly increasing fraction of ULIGs with redshift (Scenario E).
In Fig.~1 we compare our blank field 850-$\mu$m cumulative
source counts with these Guiderdoni et al.\ models (dotted lines).
Scenario A
substantially underpredicts the observed 850-$\mu$m source counts, whereas
Scenario E overpredicts the counts by a smaller factor.
Because submm fluxes are approximately independent of redshift, the
predicted number distributions versus flux are primarily dependent on the
input luminosity functions. A downward renormalization of Scenario E to
describe our data would necessitate a steepening of the luminosity
function at the faint end in order to match the submm EBL.
Because of the unique aspects of the submm counts, it
is also possible to interpret the data in an empirical way.
With the empirical $a=0.5$ parameterization of
\S~\ref{extract}, we find that 60\ per cent of the 850-$\mu$m background
light arises from sources lying between just 0.5 and 2.0\ mJy.
The mean source flux averaged over the whole population is 0.7\ mJy.
Because of the steep divergence at low
flux levels of the empirical fit to the number counts, the turnover
of the $N(>S)$ distribution will
occur in essentially the same place independent
of the regularization of the divergence by specific $a$-parameter choices;
hence, our conclusion that the dominant counts occur in the 1\ mJy region
is not sensitive to the relative calibration of the EBL and
SCUBA observations. We estimate the
cumulative source density in the $1-10$\ mJy range to be
$10^4\ {\rm deg}^{-2}$, which is consistent with the measurement
($0.79\pm 0.3)\times 10^4\ {\rm deg}^{-2}$ by Blain et al.\ (1999);
our predicted density for the $0.5-2.0$\ mJy range is
$2.6\times 10^4\ {\rm deg}^{-2}$.
Limited and still rather uncertain spectroscopic data suggest that most
of the submm sources lie in the $z=1-3$
redshift range (\markcite{barger99}Barger et al.\ 1999;
\markcite{lilly99}Lilly et al.\ 1999).
At these redshifts the far-infrared (FIR) luminosity
is approximately independent of the redshift.
Thus, if we assume an Arp~220-like spectrum with
$T=47$\ K (e.g.\ \markcite{barger98}Barger et al.\ 1998),
the FIR luminosity of a characteristic $\sim 1$\ mJy source is in the range
$4-5\times 10^{11}\ {\rm h_{65}^{-2}\ L_\odot}$
for a $q_0=0.5$ cosmology ($7-15\times 10^{11}$ for $q_0=0.02$).
The FIR luminosity provides a measure of the current star formation rate
(SFR) of massive stars (\markcite{sy83}Scoville \& Young 1983;
\markcite{tt86}Thronson \& Telesco 1986),
${\rm SFR}\sim 1.5\ \times 10^{-10}\ (L_{FIR}/{\rm L_\odot})\
{\rm M_\odot\ yr^{-1}}$; a 1\ mJy source would
therefore have a star formation rate
of $\sim 70\ {\rm h_{65}^{-2}\ M_\odot\ yr^{-1}}$\ for $q_0=0.5$, placing
the `typical' submm source at the high end of SFRs in
optically-selected galaxies
(\markcite{pettini98}Pettini et al.\ 1998). If we were to allow the dust
temperature to go as low as 30\ K, $L_{FIR}$ and the corresponding SFR would
be $\sim 4$ times smaller.
For a cumulative source density of $4.0\times 10^{4}\ {\rm deg}^{-2}$
required to reproduce the EBL with 1\ mJy sources
($\langle N\rangle={\rm EBL}/\langle S\rangle$ with
$\langle S\rangle\sim 1$\ mJy)
and redshifts in the $1-3$\ range, the average space density is
$5\times 10^{-3}\ {\rm h_{65}^{3}\ Mpc^{-3}}$
for a $q_0=0.5$ cosmology ($10^{-3}$ for $q_0=0.02$).
This space density is rather insensitive to the upper cut-off on the
redshift distribution, dropping by only a factor of $\sim 2$ or 3
if we extend the volume calculation to $z=5$.
For comparison, the space density of present-day ellipticals
is about $10^{-3}\ {\rm h_{65}^{3}\ Mpc^{-3}}$
(\markcite{marzke94}Marzke et al.\ 1994;
\markcite{yoshii88}Yoshii \& Takahara 1988).
Within the still substantial uncertainty posed by the dust temperatures,
the estimated star formation rate from submm sources is
$\sim0.3\ {\rm h_{65}\ M_\odot\ yr^{-1}\ Mpc^{-3}}$
for $q_0=0.5$, which is nearly an order
of magnitude higher than that observed in the optical,
$\sim0.04\ {\rm h_{65}\ M_\odot\ yr^{-1}\ Mpc^{-3}}$
(e.g.\ \markcite{steidel}Steidel et al.\ 1999).
Several groups (\markcite{smail98}Smail et al.\ 1998;
\markcite{eales99}Eales et al.\ 1999;
\markcite{lilly99}Lilly et al.\ 1999;
\markcite{trentham99}Trentham, Blain, \& Goldader 1999)
have suggested that the submm sources are associated with
major merger events giving rise to the formation of spheroidal galaxies.
The approximate equality of the optical and submm backgrounds supports this
hypothesis; present-day spheroidal and disk populations have roughly
comparable amounts of metal density, and thus their formation is expected
to produce comparable amounts of light (\markcite{cowie88}Cowie\ 1988).
A star formation rate of
$70\ {\rm h_{65}^{-2}\ M_\odot\ yr^{-1}\ Mpc^{-3}}$ for $q_0=0.5$
could produce a spheroid of mass
$6\times 10^{10}\ {\rm h_{65}^{-2}\ M_\odot}$ in 0.8\ Gyr.
Recent CO observations of two high redshift submm sources by
\markcite{frayer98}\markcite{frayer99}Frayer et al.\ (1998, 1999)
show that each source has enough molecular gas to form an $L^*$ galaxy.
Thus, with a duty cycle that self-consistently matches the galaxy
number density, the submm population could plausibly evolve into
the present-day spheroidal population.
In summary, we have reported new 850-$\mu$m source counts from wide-area SCUBA
imaging data of the Hawaii Deep Survey fields SSA13, SSA17, and SSA22
and improved counts from deep SCUBA imaging data of SSA13 and the
Lockman Hole. Our data rule out models that assume the submm flux is due
only to dust in optical galaxies because such models vastly underpredict the
observed counts. We find that our cumulative counts, $N(>S)$, above a
$3\sigma$ detection threshold of 2\ mJy follow a simple power-law behavior.
By introducing an additional parameter constrained by the EBL,
we extrapolate our cumulative counts below 2\ mJy and infer
that the bulk of the submm EBL resides in sources near 1\ mJy, in
agreement with a fluctuation analysis by
\markcite{hughes98}Hughes et al.\ (1998) and lensed source
counts from \markcite{blain99}Blain et al.\ (1999).
The submm sources are plausible progenitors of the present-day
spheroidal population.
\smallskip
\acknowledgments
We thank an anonymous referee for helpful suggestions, and we thank
Stephen Eales and Simon Lilly for communicating to us their
flux sensitivities and area coverages.
AJB acknowledges support from NASA through contract number P423274 from the
University of Arizona, under NASA grant NAG5-3042.
|
\section{INTRODUCTION}
Modeling ultra-relativistic fluid flows is a great challenge for any
relativistic hydro code. Numerical difficulties arise from strong
relativistic shocks and from narrow physical structures (Norman \& Winkler
1986). Typical examples in astrophysics for such extreme conditions are
proto-stellar jets and blast waves of supernovae explosions. In recent years,
the development of numerical algorithms for relativistic fluid dynamics went
mainly along two different lines. First, there are the so called High
Resolution Shock Capturing (HRSC) methods, which allow to obtain numerically
discontinuous solutions of the relativistic hydrodynamic equations by solving
local Riemann problems between adjacent numerical cells. Some recently
developed HRSC techniques are those of Font et al.~ (1994), Falle \& Komissarov
(1996), Romero et al.~ (1996), Banyuls et al.~ (1997), Wen, Panaitescu, \& Laguna
(1997), Pons et al.~ (1998), and Komissarov (1999). However, employing analytic
solutions of the Riemann shock tube problem, it is not surprising that these
HRSC codes produce almost exact numerical solutions of flow structures
including discontinuities. In the second type of algorithms, shocks are
treated numerically not as fluid discontinuities, but are rather spread over
some small length with the help of an artificial viscosity. These algorithms
either solve the dynamical equations of relativistic hydrodynamics on an
Eulerian grid such as the finite difference schemes of Hawley, Smarr, \&
Wilson (1984b) and Norman \& Winkler (1986) or by using computational nodes
following the fluid motion such as the Lagrangian Smoothed Particle
Hydrodynamics (SPH) methods of Kheyfets, Miller, \& Zurek (1990) and Laguna,
Miller, \& Zurek (1993). All the latter methods seem to be
limited even for mildly relativistic flows containing shocks. Norman \&
Winkler (1986) suggest that the appearance of numerical inaccuracies and
instabilities is due to the time derivative of the relativistic
$\gamma$-factor in the energy equation of these numerical schemes. This
additional time derivative of a hydrodynamic variable renders the system of
evolution equations non-conservative. For non-relativistic hydrodynamics, it
is well-known that a numerical method based on non-conservative equations can
produce a solution which looks reasonable but is entirely wrong if shocks or
other discontinuities are involved (see LeVeque 1997 for a corresponding
analysis of Burger's equation).
The main intention of this paper is to present a conservation formulation of
the relativistic hydrodynamic equations that is designed for numerical
methods. This particular formulation allows hydro codes to resolve
ultra-relativistic shocks numerically with relativistic $\gamma$-factors of
even $1000$ by means of an artificial viscosity rather than using a Riemann
solver. The conservation form of the relativistic equations is obtained by
choosing suitable Lagrangian variables. Unfortunately, these Lagrangian
dynamical variables can be expressed in terms of the generic fluid variables
rest-mass density, 3-velocity, and thermodynamic pressure only through a set
of nonlinear algebraic equations. However, we show that these equations can
be solved analytically in a unique way.
For numerical work we discretize our set of partial differential equations by
the Smoothed Particle Hydrodynamics (SPH) method introduced by Lucy (1977)
and Gingold \& Monaghan (1977). In recent years, SPH became a popular
computational tool for numerically modeling complex three-dimensional fluid
flows in astrophysics. This is primarily due to its computational simplicity
and the absence of a computational grid. Furthermore, SPH is adaptive in the
sense that its computational nodes follow the fluid.
SPH has been already applied to relativistic fluid flows. Kheyfets et al.~
(1990) developed a relativistically covariant version of the SPH technique
by modeling the contact interactions with spatial smoothing functions in the
local comoving frame of the fluid. Laguna et al.~ (1993) applied the SPH method
to the so called ADM formalism of general relativity due to Arnowitt, Deser,
\& Misner (1962). In their relativistic SPH formulation they modified the
flat space kernels of the Newtonian SPH method which can become anisotropic
and are then no longer invariant under translations which leads to additional
terms in the SPH equations. In contrast to these methods, we have developed a
fully three-dimensional general relativistic SPH code on the basis of our set
of Lagrangian conservation equations. This code employs the standard SPH
approach as used in Newtonian theory with spherically symmetric kernels for
all particles. Our code is restricted to ideal fluids with artificial
viscosity using the ideal-gas equation of state. The influence of the fluid
on the spacetime metric is neglected, therefore, we consider only a
background spacetime with a given but otherwise arbitrary metric. In
addition, we neglect the fluid's self-gravity and do not account for
radiation or electromagnetic effects.
The layout of this paper is as follows. In section \ref{hydroeq} we derive
the formulation of the relativistic hydrodynamic equations in Lagrangian
conservative form and present the equations for calculating the generic fluid
properties rest-mass density, 3-velocity, and thermodynamic pressure from our
suitably chosen Lagrangian variables. Section \ref{spheq} gives a review of
the standard SPH method that we use, including a prescription of its
application to relativistic fluid flows in curved spacetimes and the
implementation of an artificial viscosity. Numerical test calculations of
one-dimensional ultra-relativistic flows with discontinuities are presented
in section \ref{codetests}. Throughout this paper we set the speed of light
$c$ and the Boltzmann's constant $k$ to unity, i.e., $c=k=1$. Latin indices
$\{i,j,\dots\}$ run from $1$ to $3$, Greek indices $\{\mu,\nu,\dots\}$ from
$0$ to $3$; the signature of the metric is $(-,+,+,+)$.
\section{\label{hydroeq}THE LAGRANGIAN CONSERVATION FORMULATION OF THE
GENERAL RELATIVISTIC IDEAL FLUID EQUATIONS}
Our starting point is the covariant formulation of the equations of
relativistic hydrodynamics for a perfect fluid with artificial viscosity: the
local conservation of baryon number
\begin{equation}
\label{covconti}
\left(\rho u^\mu\right)_{;\mu} = 0
\end{equation}
and the local conservation of energy-momentum
\begin{equation}
\label{covenmom}
T^{\mu\nu}_{\enspace\enspace;\nu} = 0 \enspace
\end{equation}
with the stress-energy tensor of a perfect fluid
\begin{equation}
\label{stressenergy}
T^{\mu\nu}=(\rho w+q)u^\mu u^\nu +(p+q)g^{\mu\nu} \enspace .
\end{equation}
The semicolon denotes the covariant derivative and the Einstein summation
convention is used. Here, $\rho$ is the rest-mass density, $u^\mu$ the
4-velocity of the fluid, $p$ the isotropic thermodynamic pressure,
$w=1+\varepsilon+p/\rho$ the relativistic specific enthalpy with specific
internal energy $\varepsilon$, $q$ the artificial viscous pressure, and
$g^{\mu\nu}$ the spacetime metric. All thermodynamic quantities in the
stress-energy tensor (\ref{stressenergy}) are measured in the local rest
frame of the fluid. The artificial viscous pressure $q$, which appears in
equation (\ref{stressenergy}), can be introduced into the stress-energy
tensor either by the substitution $p\rightarrow p+q$ or, equivalently, from
the stress-energy tensor of a viscous fluid (Misner, Thorne, \& Wheeler 1973;
Landau \& Lifshitz 1991) by ignoring the shear viscosity and heat conduction
and replacing the bulk viscous pressure by $q$. An explicit expression for
$q$ in terms of velocity gradients will be given in the subsequent section.
For a Lagrangian formulation of relativistic hydrodynamics suitable for SPH
one has to break the unity of time and space inherent in the covariant
formulation. This can be accomplished by applying the ADM formalism of
Arnowitt et al.~ (1962), where the spacetime is decomposed into an infinite
foliation of spatial hypersurfaces $\Sigma_t$ of constant coordinate time $t$
by writing the line element as
\begin{displaymath}
ds^2 = g_{\mu\nu} dx^\mu dx^\nu = -\left(\alpha^2 - \beta^i\beta_i\right)dt^2
+ 2\beta_i dx^i + \eta_{ij} dx^i dx^j \enspace .
\end{displaymath}
Here, $\alpha$ is the {\it lapse} function, $\beta^i$ the {\it shift} vector,
and $\eta_{ij}$ the spatial metric induced on $\Sigma_t$ with $\beta_i =
\eta_{ij}\beta^j$ and $\eta_{il}\eta^{lj} = \delta_{i}^{~j}$. In this paper
we consider only given background spacetimes, i.e., we do not solve the
Einstein equations. Thus, $g_{\mu\nu}$, $\alpha$, $\beta^i$, and $\eta_{ij}$
are given analytic functions of both space and time. From the definitions of
$\alpha$ and $\beta^i$ the basis vector field $\mbox{\boldmath$\partial$}_t$
of the coordinate basis
$\{\mbox{\boldmath$\partial$}_t,\mbox{\boldmath$\partial$}_i\}$ can be
decomposed into normal and parallel components with respect to the
hypersurfaces $\Sigma_t$
\begin{equation}
\label{partialt}
\mbox{\boldmath$\partial$}_t = \alpha \mbox{\boldmath$n$} + \beta^i
\mbox{\boldmath$\partial$}_i \enspace ,
\end{equation}
where $\mbox{\boldmath$n$}$ is the unit time-like vector field normal to the
slices $\Sigma_t$, i.e., $\mbox{\boldmath$n\cdot\partial$}_i=0$. Observers
having $\mbox{\boldmath$n$}$ as 4-velocity are at rest in the slices
$\Sigma_t$ --- they are called {\it Eulerian observers}. In the basis
$\{\mbox{\boldmath$n$},\mbox{\boldmath$\partial$}_i\}$ the 4-velocity of a
fluid has the representation
\begin{displaymath}
\mbox{\boldmath$u$} = \gamma \left( \mbox{\boldmath$n$} + \bar v^i
\mbox{\boldmath$\partial$}_i \right) \enspace ,
\end{displaymath}
whereas in the coordinate basis
$\{\mbox{\boldmath$\partial$}_t,\mbox{\boldmath$\partial$}_i\}$ it follows from
equation (\ref{partialt})
\begin{displaymath}
\mbox{\boldmath$u$} = u^\mu\mbox{\boldmath$\partial_\mu$} =
\frac{\gamma}{\alpha} \left( \mbox{\boldmath$\partial$}_t + v^i
\mbox{\boldmath$\partial$}_i \right)
\end{displaymath}
with
\begin{displaymath}
v^i = \alpha \bar v^i - \beta^i \enspace .
\end{displaymath}
From the normalization condition of the 4-velocity $u^\mu u_\mu=-1$
the relativistic $\gamma$-factor is given by
\begin{displaymath}
\gamma = \frac{1}{\sqrt{1-\eta_{ij}{\bar v^i}{\bar v^j}}} \enspace .
\end{displaymath}
In the following, we use both, the 3-velocity of the fluid ${\bar v^i}$
measured by Eulerian observers and the 3-velocity $v^i$ in the coordinate
basis.
We now derive the Lagrangian equations of relativistic hydrodynamics where
the Lagrangian or total time derivative $d/dt$ is defined as
\begin{displaymath}
\frac{d}{dt} = \frac{\alpha}{\gamma} u^\mu \partial_\mu =
\partial_t+v^i\partial_i \enspace .
\end{displaymath}
From the law of baryon-number conservation~(\ref{covconti}) we obtain
\begin{eqnarray}
\label{lagconti}
0 & = & \partial_\mu\left(\sqrt{-g}\rho
u^\mu\right)=\partial_t\left(\sqrt{-g}\rho\frac{\gamma}{\alpha}\right) +
\partial_i\left(\sqrt{-g}\rho \frac{\gamma}{\alpha}v^i\right)\nonumber\\
& = & \frac{d D^*}{dt} + D^*\partial_iv^i \enspace ,
\end{eqnarray}
where $g$ is the determinant of the spacetime metric $g^{\mu\nu}$, and
the relativistic rest-mass density $D^*$ is defined by
\begin{equation}
\label{densityvar}
D^* = \sqrt{-g}\frac{\gamma}{\alpha}\rho = \sqrt{\eta}\gamma\rho
\enspace .
\end{equation}
Here, $\eta$ is the determinant of the spatial metric $\eta_{ij}$ obeying
$\sqrt{\eta} = \sqrt{-g}/\alpha$. With the above definition of $D^*$ the
relativistic continuity equation (\ref{lagconti}) has the same form as in the
non-relativistic case. Note, however, that in equation (\ref{lagconti}) the
expression for $\partial_i v^i$ is not the divergence of a 3-vector on
$\Sigma_t$ but rather a sum of partial derivatives. Eulerian observers
measure the relativistic rest-mass density $D = -\rho\mbox{\boldmath$u\cdot
n$} = \rho\gamma$. By rewriting equation (\ref{lagconti}) for $D$, we obtain
the relativistic continuity equation
\begin{displaymath}
0 = \frac{dD}{dt} + D\partial_iv^i + D\frac{d}{dt}\ln{\sqrt{\eta}}
\end{displaymath}
which, in contrast to equation (\ref{lagconti}), contains an additional
source term. In this paper we will use both definitions of a relativistic
rest-mass density, i.e., $D$ and $D^*=\sqrt{\eta}D$.
In order to derive our relativistic expressions for the energy and momentum
equations, we first re-express the conservation law of energy-momentum
(\ref{covenmom}) by using the continuity equation (\ref{covconti})
\begin{equation}
\label{covenmom2}
0={T_\mu^{\enspace\nu}}_{;\nu} = \rho
u^\nu\left[\left(w+\frac{q}{\rho}\right)u_\mu\right]_{;\nu} +
\partial_\mu(p+q)\enspace .
\end{equation}
One can easily show that the covariant derivative in the first term of
equation (\ref{covenmom2}) can be written as
\begin{displaymath}
u^\nu \left[\left(w+\frac{q}{\rho}\right)u_\mu\right]_{;\nu}
=\frac{\gamma}{\alpha}\frac{d}{dt}\left[\left(w+\frac{q}{\rho}\right)u_\mu
\right] - \frac{1}{2}\left(w+\frac{q}{\rho}\right)g_{\alpha\beta,\mu}
u^\alpha u^\beta \enspace .
\end{displaymath}
Thus, equation (\ref{covenmom2}) becomes
\begin{eqnarray}
\label{covenmom3}
\frac{d}{dt}\left[\left(w+\frac{q}{\rho}\right)u_\mu\right]& = &
-\frac{\alpha}{\rho\gamma}\left[\partial_\mu(p + q) - \frac{1}{2}(\rho
w+q)u^\alpha u^\beta g_{\alpha\beta,\mu}\right]\nonumber\\
& = & -\frac{1}{D^*}\left[\partial_\mu\left[\sqrt{-g}(p+q)\right] -
\frac{\sqrt{-g}}{2}T^{\alpha\beta}g_{\alpha\beta,\mu}\right] \enspace ,
\end{eqnarray}
where we have used the identity
\begin{displaymath}
\frac{1}{\sqrt{-g}}\partial_\mu\left(\sqrt{-g}\right) = \frac{1}{2}
g^{\alpha\beta}g_{\alpha\beta,\mu} \enspace .
\end{displaymath}
Taking the spatial components of equation (\ref{covenmom3}), i.e.,
$\mu=i$, and using
\begin{displaymath}
u_i = g_{i\mu}u^\mu = \beta_i\frac{\gamma}{\alpha} + \eta_{ij}\frac{\gamma}
{\alpha}v^j = \gamma\eta_{ij}\bar v^j \enspace ,
\end{displaymath}
we get the momentum equation
\begin{equation}
\label{lagmomentum}
\frac{d}{dt}S_i = -\frac{1}{D^*}\left[\partial_i\left[\sqrt{-g}(p+q)\right] -
\frac{\sqrt{-g}}{2}T^{\alpha\beta}\partial_i g_{\alpha\beta}\right] \enspace ,
\end{equation}
with the relativistic specific momentum $S_i$ defined by
\begin{equation}
\label{momentumvar}
S_i = \left(w+\frac{q}{\rho}\right)\gamma\eta_{ij}\bar v^j \enspace .
\end{equation}
The component $DS_i = - \mbox{\boldmath$T$} (\mbox{\boldmath$n$} ,
\mbox{\boldmath$\partial$}_i)$ of the stress-energy tensor field
$\mbox{\boldmath$T$}$ is the relativistic momentum density in the
$i$-direction measured by Eulerian observers. The momentum equation
(\ref{lagmomentum}), which, in a similar form, was also used by Laguna et al.~
(1993), can be applied immediately to the SPH method because it contains only
spatial derivatives on its right hand side. In the Eulerian formulation an
expression similar to equation (\ref{lagmomentum}) without artificial
viscosity is given by Hawley, Smarr, \& Wilson (1984a) for the momentum
variable $DS_i$, which is more convenient for a Eulerian description. The
Newtonian limit yielding the non-relativistic Euler equation is obvious from
equation (\ref{lagmomentum}).
Taking the $\mu=0$--component of equation (\ref{covenmom3}), we obtain the
relativistic energy equation
\begin{equation}
\label{covenmom4}
\frac{d}{dt}\left[\left(w + \frac{q}{\rho}\right)u_0\right] =
-\frac{1}{D^*}\left[\partial_t\left[\sqrt{-g}(p + q)\right] -
\frac{\sqrt{-g}}{2} T^{\alpha\beta}\partial_t g_{\alpha\beta}\right]
\enspace ,
\end{equation}
which, unfortunately, has time derivatives of hydrodynamic variables on both
sides. Re-expressing the left hand side of equation (\ref{covenmom4}) as
\begin{displaymath}
\left(w + \frac{q}{\rho}\right)u_0 = -\alpha\left(w +
\frac{q}{\rho}\right)\gamma + \beta^iS_i
\end{displaymath}
with
\begin{displaymath}
u_0 = g_{0\mu}u^\mu = \left(\beta^i\beta_i-\alpha^2\right)
\frac{\gamma}{\alpha} + \beta_i\frac{\gamma}{\alpha} v^i
= \gamma\left(\eta_{ij}\beta^i\bar v^j - \alpha\right)
\enspace ,
\end{displaymath}
and rewriting the first term on the right hand side of equation
(\ref{covenmom4}) using equation (\ref{lagconti})\,,
\begin{eqnarray*}
\frac{1}{D^*} \partial_t\left[\sqrt{-g}(p + q)\right] & = &
\frac{1}{D^*}\frac{d}{dt}\left[\sqrt{-g}(p + q)\right] -
\frac{v^i}{D^*}\partial_i\left[\sqrt{-g}(p + q)\right]\nonumber\\
& = & \frac{d}{dt}\left(\sqrt{-g}\frac{p+q}{D^*}\right)
- \frac{1}{D^*}\partial_i\left[\sqrt{-g}(p + q)v^i\right]
\enspace ,
\end{eqnarray*}
we obtain the relativistic energy equation
\begin{equation}
\label{lagenergy}
\frac{d}{dt}\left[\alpha E - \beta^iS_i\right] =
-\frac{1}{D^*}\left[\partial_i\left[\sqrt{-g}(p + q)v^i\right] +
\frac{\sqrt{-g}}{2}T^{\alpha\beta}\partial_tg_{\alpha\beta}\right] \enspace ,
\end{equation}
with the total relativistic specific energy $E$ defined as
\begin{equation}
\label{energyvar}
E = \left(w + \frac{q}{\rho}\right)\gamma - \frac{p + q}{D} \enspace .
\end{equation}
The component $DE = \mbox{\boldmath$T$} (\mbox{\boldmath$n$},
\mbox{\boldmath$n$})$ of the stress-energy tensor field $\mbox{\boldmath$T$}$
is the total relativistic energy density measured by Eulerian observers.
As in the case of the relativistic momentum equation (\ref{lagmomentum}) the
right hand side of equation (\ref{lagenergy}) contains no time derivatives of
hydrodynamic variables. It is, therefore, well suited for the SPH method in
contrast to the energy equation used by Hawley et al.~ (1984a, 1984b) and
Laguna et al.~ (1993) which has a non-conservative form containing two total
time derivatives of hydrodynamical variables separately on both sides of the
equation. Norman \& Winkler (1986) suggest that the additional time
derivative of the relativistic $\gamma$-factor in the energy equation gives
rise to numerical inaccuracies and instabilities. Note that for the special
relativistic case equation (\ref{lagenergy}) without an artificial viscosity
can also be found in Monaghan (1992). In the non-relativistic case, equation
(\ref{lagenergy}) yields the energy equation for the total non-relativistic
specific energy $|\mbox{\boldmath$v$}|^2/2+\varepsilon_N$ written in a form
similar to our expression (\ref{lagenergy})
\begin{displaymath}
\frac{d}{dt}\left[\frac{1}{2} |\mbox{\boldmath$v$}|^2 + \left(w_N +
\frac{q}{\rho_N}\right) - \frac{p + q}{\rho_N}\right] =
-\frac{1}{\rho_N}\partial_i\left[(p + q)v^i\right] \enspace ,
\end{displaymath}
where the index $N$ denotes Newtonian quantities and $w_N = \varepsilon_N +
p/\rho_N$ is the non-relativistic specific enthalpy.
To close our system of hydrodynamical equations (\ref{lagconti}),
(\ref{lagmomentum}), and (\ref{lagenergy}), we have to add an equation of
state of the form $p=p(\rho,\varepsilon)$, which relates the thermodynamic
pressure $p$ to the rest-mass density $\rho$ and the specific internal energy
$\varepsilon$. We restrict ourselves to the ideal-gas equation of state given
by
\begin{equation}
\label{idealstateeq}
p = (\Gamma - 1)\rho\varepsilon
\end{equation}
with the ideal-gas adiabatic constant $\Gamma$.
Our system of ideal fluid equations is now complete. We have derived the
relativistic hydrodynamic equations (\ref{lagconti}), (\ref{lagmomentum}),
and (\ref{lagenergy}) in Lagrangian conservative form similar to their
Newtonian counterparts. This was achieved by choosing suitable hydrodynamic
variables $D^*$, $S_i$, and $E$ defined in equations (\ref{densityvar}),
(\ref{momentumvar}), and (\ref{energyvar}). Because of the equation of state
(\ref{idealstateeq}) and the use of the 3-velocity for moving particles in
the Lagrangian numerical methods, we now need to calculate the generic
hydrodynamic quantities $\rho$, $\bar v^i$, and $p$ from these variables by
solving a highly nonlinear system of equations. If an artificial viscous
pressure $q$ is included, a severe difficulty arises from $q$ being usually
expressed in terms of velocity gradients. Since there is no time evolution
equation for $q$, the character of the dynamic equations is thus changed
which is the usual situation in the hydrodynamics of viscous flows. Due to
the coupling of $\bar v^i$ and $q$, the suitably chosen dynamic variables
actually have to be solved iteratively for the generic hydrodynamic
variables. However, artificial viscosity operates only in the vicinity of
shock transitions and is zero everywhere else. For simplicity one can
calculate the generic variables in a single iteration taking $q$ from the
previous time step. Using the expression $w=1+p/(G\rho)$ with $G=1-1/\Gamma$
for the total relativistic specific enthalpy, the specific energy $E$ from
equation (\ref{energyvar}) can be written as
\begin{eqnarray}
\label{energyvar2}
E & = & w\gamma - (w-1)\frac{G}{\gamma} + \left(\gamma -
\frac{1}{\gamma}\right)\frac{q}{\rho}\nonumber\\
& = & \left(\gamma - \frac{G}{\gamma}\right)w + \frac{G}{\gamma} +
\left(\gamma - \frac{1}{\gamma}\right)\frac{q}{\rho}\enspace .
\end{eqnarray}
Solving equation (\ref{energyvar2}) for the relativistic specific enthalpy
$w$ and adding the term $q/\rho$, we obtain
\begin{equation}
\label{hqrho}
w + \frac{q}{\rho} = \frac{\tilde E\gamma - G}{\gamma^2 - G} \enspace ,
\end{equation}
where the variable $\tilde E$ is given by
\begin{displaymath}
\tilde E = E + \frac{q}{\Gamma D}\enspace .
\end{displaymath}
Using
\begin{equation}
\label{ssquare}
S^2 = \eta^{ij}S_iS_j = \left(w + \frac{q}{\rho}\right)^2 \left(\gamma^2 -
1\right)
\end{equation}
and inserting expression (\ref{hqrho}), the relativistic $\gamma$-factor can
be determined explicitly from
\begin{equation}
\label{gamzeroeq}
0 = \left(S^2 - \tilde E^2\right) \gamma^4 + 2G\tilde E\gamma^3+\left(\tilde
E^2 - 2GS^2 - G^2\right)\gamma^2 - 2G\tilde E\gamma +
G^2\left(1+S^2\right)
\end{equation}
as the root of a polynomial of degree four. In the appendix we show that a
solution of equation (\ref{gamzeroeq}) exists and that it is unique for all
allowed values of $G$, $\tilde E$, and $S^2$. With the value of $\gamma$
known, one can calculate first the rest-mass density $\rho$ from equation
(\ref{densityvar}), then the thermodynamic pressure $p$ from equation
(\ref{hqrho}) and the equation of state (\ref{idealstateeq}), and finally the
velocity $\bar v^i$ from equation (\ref{momentumvar}) using $\eta^{ij}S_j=(w
+ q/\rho)\gamma\bar v^i$.
\section{\label{spheq}THE SPH EQUATIONS}
In this section we derive the SPH formalism for our set of relativistic
hydrodynamic equations (\ref{lagconti}), (\ref{lagmomentum}),
(\ref{lagenergy}), and (\ref{idealstateeq}). Since in the previous section
we have obtained the dynamic equations in an appropriate Lagrangian form, we
can proceed in a way that is completely analogous to the Newtonian case.
Before we derive our relativistic SPH equations, we give a brief introduction
into the standard SPH formalism, which was invented independently by Lucy
(1977) and Gingold \& Monaghan (1977). For a review of the SPH method see
Monaghan (1992).
SPH is a numerical method which discretizes the dynamic equations on a set of
nodes, called particles, moving with the fluid. The final discrete equations
are obtained in two separate steps. First, all hydrodynamic functions on
$\Sigma_t$ are smoothed over a certain volume with the help of a smoothing
kernel $W(|\mbox{\boldmath $r$} - \mbox{\boldmath$r$}^\prime |,h)$, i.e., for
a continuous function $f(\mbox{\boldmath$r$})$ one has
\begin{equation}
\label{intintpol}
f(\mbox{\boldmath$r$}) = \int f(\mbox{\boldmath$r$}')\,
W(|\mbox{\boldmath$r$}-\mbox{\boldmath$r$}^\prime|,h)\, d\mbox{\boldmath$r$}'
+ O(h^2)~,
\end{equation}
where $\mbox{\boldmath$r$}=\{x^i\}$ is the set of spatial coordinates $x^i$.
The kernel $W$ is a smooth (differentiable) function with compact support of
size $h$, the so called smoothing length, it is normalized to unity
\begin{displaymath}
\int W(|\mbox{\boldmath$r$}-\mbox{\boldmath$r$}^\prime|,h)\,
d\mbox{\boldmath$r$}' = 1~,
\end{displaymath}
and consequently, the smoothing error in equation (\ref{intintpol}) is
$O(h^2)$. The above integrals extend over a region on $\Sigma_t$ around
$\mbox{\boldmath$r$}$ that contains the support of $W$. An example for $W$ is
the cubic-spline kernel of Monaghan \& Lattanzio (1985), and in our
simulations we use this kernel throughout. The second step consists of an
approximate evaluation of the above integral (\ref{intintpol}) at the
particle positions $\mbox{\boldmath$r$}_a$
\begin{equation}
\label{sumintpol}
\int f(\mbox{\boldmath$r$}')\,
W(|\mbox{\boldmath$r$}_a-\mbox{\boldmath$r$}^\prime|,h)\,
d\mbox{\boldmath$r$}' \approx \sum_b \frac{f_b}{n_b}W_{ab} =:
\left<f\right>_a \enspace .
\end{equation}
Here, $a$ and $b$ label the particles which are distributed in space with
number density $n(\mbox{\boldmath$r$})$. We define $f_a =
f(\mbox{\boldmath$r$}_a)$ for any function $f$ and $W_{ab} =
W(|\mbox{\boldmath$r$}_a - \mbox{\boldmath$r$}_b|,h)$. The important point
is that, applying the smoothing and discretization operations
(\ref{intintpol}), (\ref{sumintpol}), it is possible to derive approximate
expressions for derivatives
\begin{equation}
\label{sphsumgrad}
\left<\mbox{\boldmath$\nabla$}f\right>_a\ = \sum_b \frac{f_b}{n_b}
\mbox{\boldmath$\nabla$}_aW_{ab}~~,~~~~ \left<\partial_if^i\right>_a\ =
\sum_b \frac{1}{n_b}{\mbox{\boldmath$f$}_b}
\mbox{\boldmath$\cdot\nabla$}_aW_{ab}
\enspace ,
\end{equation}
where $\mbox{\boldmath$f$}={\{f^i\}}$,
$\mbox{\boldmath$\nabla$}=\{\partial_i\}$, and
$\mbox{\boldmath$\nabla$}_aW_{ab}$ is the gradient of the kernel
$W(|\mbox{\boldmath$r$}_a - \mbox{\boldmath$r$}_b|,h)$ taken with respect to
the coordinates of particle $a$.
With the discretization method described above we are now in the position to
derive the SPH equations for our system of relativistic hydrodynamic
equations (\ref{lagconti}), (\ref{lagmomentum}), (\ref{lagenergy}), and
(\ref{idealstateeq}), and nothing has to be changed compared to the standard
(non-relativistic) SPH method. In the equations (\ref{sumintpol}) and
(\ref{sphsumgrad}) we replace the number density $n_a$ by the mass per
particle $m_a$, i.e., $n_a = D^*_a/m_a$. Dropping the angle brackets
$\left<~\right>$ from now on, equation (\ref{sumintpol}) leads to the SPH
representation for the relativistic rest-mass density
\begin{equation}
\label{sphdensity}
D^*_a = \sum_b m_b W_{ab} \enspace .
\end{equation}
From equation (\ref{lagconti}) and using
\begin{displaymath}
D^*\mbox{\boldmath$\nabla\cdot v$} =
\mbox{\boldmath$\nabla\cdot$}(D^*\mbox{\boldmath$v$}) -
\mbox{\boldmath$v\cdot\nabla$}D^*\enspace ,
\end{displaymath}
we obtain the SPH form of the relativistic continuity equation
\begin{equation}
\label{sphconti}
\frac{d}{dt}D^*_a = - \sum_b m_b(\mbox{\boldmath$v$}_b -
\mbox{\boldmath$v$}_a)\mbox{\boldmath$\cdot\nabla$}_aW_{ab} \enspace .
\end{equation}
Here, $\mbox{\boldmath$v$}=\{v^i\}$ is just a shorthand notation for the set
of components $v^i$, whereas $\mbox{\boldmath$\nabla\cdot v$}$ stands for
$\partial_i v^i$ and is not the divergence of a vector field
$\mbox{\boldmath$v$}$. As a consequence, the total mass is conserved exactly.
Applying the Lagrangian time derivative to equation (\ref{sphdensity}) with
the smoothing length being constant in space and time, one can show that the
SPH expression of the relativistic rest-mass density $D^*$ in equation
(\ref{sphdensity}) automatically satisfies the relativistic continuity
equation. Thus, in SPH the relativistic rest-mass density $D^*$ can be
computed either from equation (\ref{sphdensity}) or equation
(\ref{sphconti}). If, however, the density $D = \gamma \rho$ is used instead
of $D^*$, a modification of the standard SPH method is required. Therefore,
Laguna et al. (1993) multiply the flat-space kernel $W(|\mbox{\boldmath$r$} -
\mbox{\boldmath$r$}'|,h)$ of Newtonian SPH with
$1/\sqrt{\eta(\mbox{\boldmath$r$}')}$. In a curved spacetime this factor
makes the kernel anisotropic, violates its translation invariance, and leads
to additional terms in the SPH approximation of derivatives. Applying the
relation $D^* = \sqrt{\eta}D$, the relativistic SPH formulation of Laguna
et al.~ (1993) can be cast into the standard SPH scheme which is considerably
simpler. We therefore suggest that there is no need to modify SPH for given
background spacetimes if the continuity equation is used in the conservative
form (\ref{lagconti}) without source terms.
In order to derive the SPH form of the relativistic momentum equation we
start from equation (\ref{lagmomentum}). Rewriting the pressure gradient as
\begin{displaymath}
\frac{1}{D^*}\mbox{\boldmath$\nabla$}\left[\sqrt{-g}(p+q)\right] =
\sqrt{-g}\left[\mbox{\boldmath$\nabla$}\left(\frac{p+q}{D^*}\right) +
\frac{p+q}{D^{*2}}\mbox{\boldmath$\nabla$}D^*\right] +
\frac{p+q}{D^*}\mbox{\boldmath$\nabla$}\sqrt{-g} \enspace ,
\end{displaymath}
we obtain
\begin{eqnarray}
\label{sphmomentumeq}
\frac{d}{dt}\mbox{\boldmath$S$}_a & = &-\sqrt{-g_a} \sum_b
m_b\left(\frac{p_a+q_a}{D^{*2}_a} +
\frac{p_b+q_b}{D^{*2}_b}\right)\mbox{\boldmath$\nabla$}_aW_{ab}\nonumber\\
&&{}-\frac{\sqrt{-g_a}}{D^*_a}\left[(p_a+q_a)\mbox{\boldmath$\nabla$}_a
\left(\ln{\sqrt{-g}}\right)_a-\frac{1}{2}T^{\alpha\beta}_a
\mbox{\boldmath$\nabla$}_a\left(g_{\alpha\beta}\right)_a\right]
\enspace ,
\end{eqnarray}
where $\mbox{\boldmath$S$}_a={\{S_i\}}_a$, and the metric gradients
$\mbox{\boldmath$\nabla$}\ln\sqrt{-g}$ and
$\mbox{\boldmath$\nabla$}g_{\alpha\beta}$ can be calculated analytically.
Thus, only the pressure gradient term in equation (\ref{sphmomentumeq}) has
to be smoothed, and it has been symmetrized to conserve linear and angular
momentum in SPH. Next, we proceed with the relativistic energy equation
(\ref{lagenergy}). In the expression
\begin{displaymath}
\frac{1}{D^*} \mbox{\boldmath$\nabla\cdot$}
\left[\sqrt{-g}(p+q)\mbox{\boldmath$v$}\right] = \frac{\sqrt{-g}}{D^*}
\mbox{\boldmath$\nabla\cdot$} \left[(p+q)\mbox{\boldmath$v$}\right] +
\frac{p+q}{D^*} \mbox{\boldmath$v\cdot\nabla$}\sqrt{-g}
\end{displaymath}
we replace the first term on the right hand side by
\begin{eqnarray*}
\frac{1}{D^*} \mbox{\boldmath$\nabla\cdot$} \left[(p+q)
\mbox{\boldmath$v$}\right] & = & \frac{1}{D^*} \mbox{\boldmath$v\cdot\nabla$}
(p+q) + \frac{p+q}{D^*} \mbox{\boldmath$\nabla\cdot v$}
\nonumber\\
& = &\mbox{\boldmath$v\cdot$} \left[ \mbox{\boldmath$\nabla$}
\left(\frac{p+q}{D^*} \right) + \frac{p+q}{D^{*^2}} \mbox{\boldmath$\nabla$}
D^* \right] \nonumber\\
&&{}+\frac{1}{2} \left[ \mbox{\boldmath$\nabla\cdot$} \left(\frac{p+q}{D^*}
\mbox{\boldmath$v$} \right) - \mbox{\boldmath$v\cdot\nabla$}
\left(\frac{p+q}{D^*}\right) \right. \nonumber\\
&&\left.{}+\frac{p+q}{D^{*2}} \left[ \mbox{\boldmath$\nabla\cdot$}
(D^* \mbox{\boldmath$v$}) - \mbox{\boldmath$v\cdot\nabla$} D^* \right]
\right] \enspace .
\end{eqnarray*}
With this combination of terms we obtain the SPH representation of the
relativistic energy equation
\begin{eqnarray}
\label{sphenergyeq}
\frac{d}{dt} {\left[ \alpha E - \beta^i S_i \right]}_a & = &
-\frac{\sqrt{-g_a}}{2} \sum_b m_b\left(\frac{p_a+q_a}{D^{*2}_a} +
\frac{p_b+q_b}{D^{*2}_b}\right) (\mbox{\boldmath$v$}_a +
\mbox{\boldmath$v$}_b) \mbox{\boldmath$\cdot\nabla$}_a W_{ab} \nonumber\\
&&{}-\frac{\sqrt{-g_a}}{D^*_a} \left[ (p_a+q_a) \mbox{\boldmath$v$}_a
{\mbox{\boldmath$\cdot\nabla$}_a \left(\ln{\sqrt{-g}}\right)}_a + \frac{1}{2}
T^{\alpha\beta}_a {\left(g_{\alpha\beta,t}\right)}_a \right]
\enspace .
\end{eqnarray}
Note that equation (\ref{sphenergyeq}) and the momentum equation
(\ref{sphmomentumeq}) contain identical symmetric factors in front of the
kernel gradients. As outlined in section \ref{hydroeq}, this form of the
relativistic energy equation is well suited for SPH because it contains no
time derivatives of hydrodynamic variables on its right hand side. Again,
there is no need to smooth the derivatives of the metric
$\mbox{\boldmath$\nabla$}\ln\sqrt{-g}$ and $g_{\alpha\beta,t}$. The last
equation that needs to be considered is the equation of state
(\ref{idealstateeq}), and its SPH formulation reads
\begin{equation}
\label{sphstateeq}
p_a = (\Gamma - 1)\rho_a\varepsilon_a \enspace .
\end{equation}
This completes our set of relativistic SPH equations (\ref{sphdensity}),
(\ref{sphmomentumeq}), (\ref{sphenergyeq}), and (\ref{sphstateeq}) for
computing general relativistic fluid flows in given arbitrary background
spacetimes.
We now focus our attention to the implementation of an artificial viscosity
term which is necessary to handle shock fronts. For the use of an artificial
viscosity in the standard SPH method see Monaghan \& Gingold (1983). In our
simulations, we have used the following artificial viscous pressure
\begin{equation}
\label{artvisc}
q_a = \left\{\begin{array}{ll} \rho_a w_a \left[ - \tilde\alpha c_a h_a
{\left( \mbox{\boldmath$\nabla\cdot v$} \right)}_a + \tilde\beta h_a^2
{\left( \mbox{\boldmath$\nabla\cdot v$} \right)}_a^2 \right] &
\mbox{if}\enspace {\left( \mbox{\boldmath$\nabla\cdot v$} \right)}_a < 0
\\ 0 & \mbox{otherwise}\end{array}\right.
\end{equation}
with
\begin{equation}
\label{sphdivv}
{\left( \mbox{\boldmath$\nabla\cdot v$} \right)}_a \approx
\frac{\mbox{\boldmath$v$}_{ab} \mbox{\boldmath$\cdot r$}_{ab}}
{|\mbox{\boldmath$r$}_{ab}|^2 + \tilde\varepsilon \bar h^2_{ab}} \enspace ,
\end{equation}
where $c_a=\sqrt{\Gamma p_a/(\rho_aw_a)}$ is the relativistic sound velocity
measured in the rest frame of the fluid, $\tilde\alpha$, $\tilde\beta$ and
$\tilde\varepsilon$ are numerical parameters, $\mbox{\boldmath$v$}_{ab}$,
$\mbox{\boldmath$r$}_{ab}$ stand for the differences
$\mbox{\boldmath$v$}_{ab} = \mbox{\boldmath$v$}_a - \mbox{\boldmath$v$}_b$,
$\mbox{\boldmath$r$}_{ab} = \mbox{\boldmath$r$}_a - \mbox{\boldmath$r$}_b$,
and $\bar h_{ab}=(h_a + h_b)/2$ is the mean value of the smoothing lengths of
particles $a$ and $b$. Without the enthalpy $w_a$, the expression
(\ref{artvisc}) for $q$ is almost equivalent to the standard SPH form of the
artificial viscosity invented by Monaghan \& Gingold (1983). Including $w_a$
into equation (\ref{artvisc}), the parameters $\tilde\alpha$ and
$\tilde\beta$ can be chosen to be of order unity even for shocks with
ultra-relativistic values of $\gamma$. The $\tilde\alpha$-term, which is
linear in the velocity differences, is similar to a physical shear and bulk
viscosity, and the quadratic $\tilde\beta$-term is the standard von
Neumann-Richtmyer (1950) artificial viscosity used in finite difference
methods for handling high Mach-number shocks. According to Monaghan \&
Gingold (1983), the un-smoothed representation of the velocity divergence in
equation (\ref{sphdivv}) acts more directly on the relative motion of
particle pairs and leads to a damping of irregular oscillations in shock
transitions. In equation (\ref{sphdivv}), the parameter $\tilde\varepsilon$
has been introduced to avoid singularities, and a typical value is
$\tilde\varepsilon=0.1$.
Since the expression (\ref{artvisc}) for the artificial viscosity is not
Lorentz invariant, we also performed calculations with a relativistically
covariant formulation of the artificial viscous pressure
\begin{equation}
\label{artviscrel}
q_a = \left\{\begin{array}{ll} \rho_a\left[-\tilde\alpha c_ah_a\theta_a +
\tilde\beta h_a^2\theta_a^2\right] & \mbox{if}\enspace \theta_a<0\\ 0 &
\mbox{otherwise}\end{array}\right.
\end{equation}
where
\begin{displaymath}
\theta_a = {\left({u^\mu}_{;\mu}\right)}_a = \frac{1}{\alpha_a} \left[
\frac{d\gamma_a}{dt} + \gamma_a \left( {\left( \mbox{\boldmath$\nabla\cdot
v$} \right)}_a + \frac{d}{dt} {\left(\ln\sqrt{\eta}\right)}_a \right) \right]
\enspace .
\end{displaymath}
However, this expression contains a time derivative of the $\gamma$-factor
which destroys the explicit nature of the Lagrangian form of the hydrodynamic
equations. One possibility to circumvent this problem is to take the backward
time difference approximation $[\gamma_a(t) - \gamma_a(t-\Delta t)]/\Delta t$
with time step $\Delta t$ for the time derivative $d\gamma_a/dt$. Since the
non-covariant form (\ref{artvisc}) turned out to be quite appropriate for the
resolution of shock structures, we used this approach in all our simulations
and did not pursue the covariant relation (\ref{artviscrel}).
To improve the local resolution of SPH, we allow the smoothing length $h$ to
vary in space according to the relativistic rest-mass density $D^*$ via
\begin{equation}
\label{varhsml}
h_a = \left(h_0\right)_a\left[\frac{\left(D^*_0\right)_a}{D^*_a}\right]^{1/d}
\enspace ,
\end{equation}
where $d$ denotes the dimension of the spatial slices $\Sigma_t$. For
particles of equal mass the scaling law (\ref{varhsml}) indicates that the
number of particles within the support of the kernel $W$ is approximately
kept constant in time. Thus, in regions where the gas is compressed, the
smoothing length is increased while it is decreased in rarefaction zones.
Since the smoothing length is now a function of the density $D^*$, equation
(\ref{sphdensity}) (or eq.~[\ref{sphconti}]) is now a nonlinear implicit
relation for the density. We solve this approximately by inserting $D^*_a$
from the previous time step into equation (\ref{varhsml}) to obtain an
estimate for $h_a$. Since the smoothing length is now a function of position
and time, the SPH form of the continuity equation contains additional terms
(Monaghan 1992). However, no such terms appear if the relativistic rest-mass
density $D^*_a$ is calculated from the computationally simpler equation
(\ref{sphdensity}).
\section{\label{codetests}NUMERICAL TESTS}
Although we have developed a fully three-dimensional general relativistic SPH
code, we restrict ourselves in this paper to the standard analytic test bed
of one-dimensional special relativistic shock problems: the shock tube and
the wall shock. For each simulation we show four diagrams of the numerical
results together with the analytic solution for the fluid variables
3-velocity $v:=\bar v^1=v^1$, rest-mass density $\rho$, thermodynamic
pressure $p$, and specific internal energy $\varepsilon$. These variables are
functions of the coordinate $x:=x^1 \in [0,100]$. The quality of the
simulations is measured in terms of the relative error with respect to the
analytic solution, i.e., for each hydrodynamic function $f$ an error $\Delta
f$ is calculated from
\begin{displaymath}
\Delta f = \frac{1}{Nf^0_{\rm max}}\sum^N_{b=1}|f_b - f^0_b| \enspace ,
\end{displaymath}
where the sum is over all $N$ particles and $f^0$ stands for the analytic
solution which has the maximum value $f^0_{\rm max}$.
\subsubsection*{a) Shock Tube Tests}
First we consider the shock tube problem, where initially a fluid at rest is
divided by a diaphragm into two regions of different densities and internal
energies. When the diaphragm is removed, a rarefaction wave travels into the
warm and dense medium and a compression wave into the cold and lower density
fluid. Between the two media a so called contact discontinuity is present.
For the analytic solution of the relativistic shock tube problem we refer to
Taub (1948), McKee \& Colgate (1973), Hawley et al.~ (1984a), and Marti \&
M\"uller (1994).
\begin{figure}[t]
\begin{center}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_1v.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_1d.ps}
\end{minipage}
\\
\vspace{10mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_1p.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_1e.ps}
\end{minipage}
\end{center}
\caption{\label{fig1} Numerical result of an SPH calculation with $1000$
particles (open circles) and analytic solution (solid line) for the
non-relativistic shock tube problem (note that the units are arbitrary except
for the velocity which is measured in units of $c$). The intermediate
pressure, velocity, and relativistic $\gamma$-factor are $p_m=0.303$,
$v_m=2.93\times10^{-3}$, and $\gamma_m=1.000004$, the positions of the shock,
the contact discontinuity, and the head and tail of the rarefaction wave are
$x_s=83.2$, $x_c=67.6$, $x_{h}=27.6$, and $x_{t}=48.7$, respectively, and the
velocity and relativistic $\gamma$-factor of the shock are
$v_s=5.54\times10^{-3}$ and $\gamma_s=1.000015$.}
\end{figure}
As any relativistic hydro code has to be tested for the Newtonian limit, we
start our series of shock tube simulations with a non-relativistic test case.
Figure \ref{fig1} shows the corresponding numerical and analytic solution
for a gas with $\Gamma=1.4$ at time $t=6000$ with the initial conditions
$\rho=10^5$, $\varepsilon=2.5\times10^{-5}$ for $x < 50$, and
$\rho=0.125\times10^5$, $\varepsilon=2\times10^{-5}$ for $x > 50$ (note that
the units are arbitrary except for the velocity which is measured in units of
$c$). The numerical calculation was performed with 1000 particles, initial
smoothing length $h=0.6$ ($\approx10$ interactions per particle), and
artificial viscosity parameters $\tilde\alpha=1$ and $\tilde\beta=2$. In the
initial distribution, particles were placed on a uniform grid, and the
particles at $x>50$ have a mass ten times smaller than the particles at
$x<50$. In the calculation of Figure \ref{fig1} the largest relative error
occurs in the fluid velocity where $\Delta v=1.1\%$.
\begin{figure}[t]
\begin{center}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_2v.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_2d.ps}
\end{minipage}
\\
\vspace{10mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_2p.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_2e.ps}
\end{minipage}
\end{center}
\caption{\label{fig2} Numerical result of an SPH calculation with $1000$
particles (open circles) and analytic solution (solid line) for the
relativistic shock tube problem. The intermediate pressure, velocity, and
relativistic $\gamma$-factor are $p_m=1.45$, $v_m=0.714$, and $\gamma_m=1.4$,
the positions of the shock, the contact discontinuity, and the head and tail
of the rarefaction wave are $x_s=87.3$, $x_c=82.1$, $x_{h}=17.8$, and
$x_{t}=57.5$, respectively, and the velocity and relativistic $\gamma$-factor
of the shock are $v_s=0.828$ and $\gamma_s=1.8$.}
\end{figure}
\newpage
Next, a mildly relativistic shock tube is investigated with initial
conditions $\rho=10$, $\varepsilon=2$ ($x<50$), and $\rho=1$,
$\varepsilon=10^{-6}$ ($x>50$). Figure \ref{fig2} shows the numerical result
and the analytic solution for a gas with $\Gamma=5/3$ at time $t=45$. We used
the same numerical parameters as in the non-relativistic shock tube problem,
i.e., 1000 particles, $h=0.6$, $\tilde\alpha=1$, and $\tilde\beta=2$. The
largest relative error is $\Delta v=1.0\%$. As can be seen in Figure
\ref{fig2}, the velocity profile of a relativistic rarefaction wave is no
longer linear as in the Newtonian case because of the relativistic velocity
addition formula. Comparing our results with the simulations of Hawley et al.~
(1984b) and Laguna et al.~ (1993) for the same $\gamma=1.4$ shock tube, we note
that numerical inaccuracies due to the non-conservative formulation of
their energy equation clearly show up in their results for the relativistic
rest-mass density and specific internal energy. To investigate the
convergence properties of our numerical method, we performed a calculation of
the $\gamma=1.4$ shock tube with $10000$ particles and initial smoothing
length $h=0.1$ ($\approx20$ interactions per particle). As can be seen in
Figure \ref{fig3}, the numerical calculation covers the analytic solution
almost exactly, and the largest relative error is reduced to $\Delta
v=0.2\%$.
\begin{figure}[t]
\begin{center}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_3v.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_3d.ps}
\end{minipage}
\\
\vspace{10mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_3p.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{tube_3e.ps}
\end{minipage}
\end{center}
\caption{\label{fig3} Numerical result of an SPH calculation with $10000$
particles (points) and analytic solution (solid line) for the relativistic
shock tube problem. The intermediate pressure, velocity, and relativistic
$\gamma$-factor are $p_m=1.45$, $v_m=0.714$, and $\gamma_m=1.4$, the
positions of the shock, the contact discontinuity, and the head and tail of
the rarefaction wave are $x_s=87.3$, $x_c=82.1$, $x_{h}=17.8$, and
$x_{t}=57.5$, respectively, and the velocity and relativistic $\gamma$-factor
of the shock are $v_s=0.828$ and $\gamma_s=1.8$.}
\end{figure}
When the relativistic $\gamma$-factor of the shock tube is increased by
increasing the initial specific internal energy ratio, the region between the
leading shock front and the trailing contact discontinuity becomes extremely
thin and dense. Thus, without specially designed adaptive methods, it will be
impossible to resolve these Lorentz contracted shells of matter, which are
typical for relativistic fluid flows. In addition, the specific internal
energy at the contact discontinuity may become negative in the SPH
simulations. In order to avoid these difficulties, we consider now a
simplified problem of a single shock front without the presence of a contact
discontinuity and a nonlinear rarefaction wave.
\subsubsection*{b) Wall Shock Tests}
\begin{figure}[t]
\begin{center}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_1v.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_1d.ps}
\end{minipage}
\\
\vspace{10mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_1p.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_1e.ps}
\end{minipage}
\end{center}
\caption{\label{fig4} Numerical result of an SPH calculation with about $250$
particles (open circles) and analytic solution (solid line) for the
relativistic $\gamma_i=1.8$ wall shock problem. The post shock properties are
$\rho_p=5.66$, $p_p=1.52$, and $\varepsilon_p=0.803$, and the position and
velocity of the shock front are $x_s=64.3$ and $v_s=-0.178$ with
$\gamma_s=1.02$.}
\end{figure}
As a second test of our numerical method we modeled the wall shock problem of
a cold relativistically moving fluid flowing towards a solid wall. As the
fluid hits the wall, a shock front forms, which then travels upstream against
the incoming fluid producing a hot and dense post shock region of zero
velocity.
\begin{figure}[t]
\begin{center}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_2v.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_2d.ps}
\end{minipage}
\\
\vspace{10mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_2p.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_2e.ps}
\end{minipage}
\end{center}
\caption{\label{fig5} Numerical result of an SPH calculation with about
$250$ particles (open circles) and analytic solution (solid line) for the
ultra-relativistic $\gamma_i=1000$ wall shock problem. The post shock
properties are $\rho_p\approx4$, $p_p\approx4/3\times10^3$, and
$\varepsilon_p\approx1000$, and the position and velocity of the shock front
are $x_s\approx33.3$ and $v_s\approx-1/3$ with $\gamma_s\approx1.06$.}
\end{figure}
Figure \ref{fig4} shows the numerical result and the analytic solution of a
mildly relativistic wall shock for a gas with $\Gamma=4/3$ at time $t=200$
moving to the right with the reflecting wall at $x=100$. The uniform initial
fluid properties are $D_i=1$, $v_i=0.832$ with $\gamma_i=1.8$, and
$\varepsilon_i=10^{-5}$. Initially, particles of equal mass are uniformly
distributed in the simulation domain. At the location of the solid wall we
use reflecting boundary conditions. The simulation of Figure \ref{fig4}
showing about $250$ particles was performed with the initial smoothing length
$h=3$ ($\approx10$ interactions per particle) and artificial viscosity
parameters $\tilde\alpha=0.25$ and $\tilde\beta=0.5$. The rest-mass density
$\rho$, the thermodynamic pressure $p$, and the specific internal energy
$\varepsilon$ show a spike-like feature (see Fig.~\ref{fig4}), which is
known in the literature as ``wall heating'' (Norman \& Winkler 1986). The
largest relative error appears in the specific internal energy where
$\Delta\varepsilon=1.1\%$.
The results of an ultra-relativistic shock simulation are shown in Figure
\ref{fig5}. The initial velocity $v_i$ is increased to $v_i=0.9999995$ which
corresponds to a relativistic $\gamma$-factor of $\gamma_i=1000$. All other
parameters are identical with those of the previous wall shock calculation.
Neglecting the pre-shock specific internal energy $\varepsilon_i$, one can
show that in the ultra-relativistic limit $v_i\to1$ the post shock properties
$\rho_p$, $p_p$, $\varepsilon_p$, and the shock velocity $v_s$ are given by
$\rho_p=D_i\Gamma/(\Gamma -1)$, $p_p = \Gamma D_i\gamma_i$,
$\varepsilon_p=\gamma_i$, and $v_s=-(\Gamma-1)$. For the wall shock problem
of Figure \ref{fig5} we thus obtain $\rho_p\approx4$,
$p_p\approx4/3\times10^3$, $\varepsilon_p\approx1000$, and $v_s\approx-1/3$.
At the time $t=200$ the shock front has moved the distance
$|v_s|t\approx66.7$ to the left reaching the spatial position
$x_s\approx33.3$. The largest relative error in this case is $\Delta
\rho=1.1\%$.
\begin{figure}[t]
\begin{center}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_3v.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_3d.ps}
\end{minipage}
\\
\vspace{10mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_3p.ps}
\end{minipage}
\hspace{15mm}
\begin{minipage}[t]{0.41\textwidth}
\epsfxsize=0.89\textwidth
\epsfbox{wall_3e.ps}
\end{minipage}
\end{center}
\caption{\label{fig6} Numerical result of a finite difference calculation
with a grid of $250$ zones (open circles) and analytic solution (solid line)
for the ultra-relativistic $\gamma_i=1000$ wall shock problem. The post shock
properties are $\rho_p\approx4$, $p_p\approx4/3\times10^3$, and
$\varepsilon_p\approx1000$, and the position and velocity of the shock front
are $x_s\approx33.3$ and $v_s\approx-1/3$ with $\gamma_s\approx1.06$.}
\end{figure}
A calculation of a mildly relativistic wall shock problem (with
$\gamma_i=2.24$) using artificial viscosity was also performed by Hawley
et al.~ (1984b) with their time-explicit Eulerian finite difference code. They
tried several formulations of their non-conservative energy equation by
omitting and including various artificial viscous pressure terms. They found
that the $q\partial_t\gamma$ term in their energy equation is unstable even
for mildly relativistic wall shocks. This is verified by the calculations of
Norman \& Winkler (1986) using an implicit adaptive-mesh finite difference
method, which shows the requirement of an implicit technique to handle the
$q\partial_t\gamma$ term. The fact that our method is able to treat even
ultra-relativistic shocks without problems is entirely based on our
conservation formulation of the relativistic hydrodynamic equations where no
additional time derivatives of hydrodynamic variables appear such as the
$q\partial_t\gamma$ term. In order to verify that this numerical stability is
not restricted to the SPH method, we also performed numerical simulations of
the wall shock problem with a simple explicit finite difference upwind
scheme. Figure \ref{fig6} shows the numerical result and the analytic
solution for the $\gamma_i=1000$ wall shock at time $t=200$ using a grid of
$250$ zones, and the maximum relative error occurs for the specific internal
energy where $\Delta\varepsilon=2.2\%$.
\section{\label{summary}SUMMARY}
We have derived a fully Lagrangian conservative form of the general
relativistic equations of hydrodynamics for a perfect fluid with artificial
viscosity in a given arbitrary background spacetime. This has been achieved
by choosing suitable Lagrangian time evolution variables. These variables are
connected to the generic fluid variables rest-mass density $\rho$, 3-velocity
$\bar v^i$, and thermodynamic pressure $p$ through a set of nonlinear
algebraic equations. For an ideal gas, we have shown that these equations can
be reduced to a single fourth-order equation with a unique solution which can
be explicitly calculated in terms of various roots. For more complex
equations of state the solution of the above algebraic equations has to be
performed numerically, and the question of uniqueness is more complicated.
Using our Lagrangian formulation, we have developed a three-dimensional
general relativistic SPH code based on the standard SPH approach of
non-relativistic fluid dynamics. The important point is that all metric
factors from covariant derivatives have been absorbed in the definition of
the Lagrangian variables and thus no longer appear in the fluid equations. As
a result, the SPH kernels remain spherically symmetric and are the same for
all particles. This is an essential difference to the covariant SPH approach
of Kheyfets et al.~ (1990) using kernels defined in the local comoving frame of
the fluid, and to the relativistic SPH method of Laguna et al.~ (1993) where,
in curved spacetime, the kernels are anisotropic and have no translational
symmetry which leads to additional terms in the SPH equations. The
relativistic continuity equation (\ref{lagconti}), for example, contains no
source term, hence we can identify the relativistic rest-mass density $D^* =
\sqrt{\eta} \gamma \rho$ as the appropriate SPH density for smoothing the
equations.
In this paper we have restricted ourselves to the numerical simulation of two
different one-dimensional examples, i.e., the special relativistic shock tube
and the wall shock. An empirical error estimate is obtained from a comparison
of the numerical results with the corresponding analytic solutions. The SPH
calculations with $1000$ particles show a typical maximum relative error of
about $1\%$, and an increase of the particle number with a decreasing
smoothing length reduces this error in a uniform way. The wall-shock problem
can be solved without any numerical difficulties for very large
$\gamma$-factors (at least $\gamma = 1000$). The shock-tube case suffers from
a resolution problem if $\gamma$ is large because a thin shell of matter
builds up between the shock front and the contact discontinuity. This can
only be resolved with the help of additional adaptive methods which increase
the number of particles in this zone.
An important ingredient in our SPH formulation is the treatment of shock
structures by an artificial viscosity rather than using a Riemann solver.
This is very easy to implement even for the two- or three-dimensional case.
Introducing an artificial viscous pressure $q$ is considered as a purely
numerical tool that acts as a filter to smear out steep gradients in the
hydrodynamic functions and suppresses unphysical oscillations. Thus, there is
no need to use a covariant expression for $q$ in terms of velocity
derivatives as long as the time evolution and the jump conditions of shocks
are represented correctly. Since the jump conditions follow from the
conservation of mass, momentum, and energy across the shock, it is obvious
that a conservative form of the relativistic hydrodynamic equations is
important for handling discontinuities in relativistic fluid dynamics
numerically. The simulation of relativistic flows with large $\gamma$-factors
including shocks is the traditional domain of HRSC methods. However, with our
formulation of the general relativistic hydrodynamic equations it is possible
to model such flows using an artificial viscosity independent of the
underlying numerical method, i.e., SPH or finite difference schemes.
\acknowledgments
\section*{ACKNOWLEDGEMENTS}
We want to thank P. Laguna for helpful discussions and for encouraging us to
publish our results. This work is part of the project A7 of the
Sonderforschungsbereich SFB~382 funded by the Deutsche Forschungsgemeinschaft
(DFG).
\newpage
|
\section{Introduction}
The anisotropic $S=\frac{1}{2}$ Heisenberg model, or XXZ chain, is one
of the most studied quantum spin system in statistical mechanics. Since
its exact solution by Yang and Yang\,\cite{1} this model has been
considered to be the classic example of the success of the Bethe ansatz.
In this paper, considering the XXZ chain as a prototype we introduce a
new set of quantum chains which preserve the property of integrability.
The XXZ Hamiltonian on an $L$-site chain is given by
\begin{eqnarray}
H(\Delta) = -\frac{1}{2}\sum_{i=1}^L(\sigma_i^x\sigma_{i+1}^x +
\sigma_i^y\sigma_{i+1}^y + \Delta\sigma_i^z\sigma_{i+1}^z) \,,
\end{eqnarray}
where $\sigma^x,\sigma^y,\sigma^z$ are spin-$\frac{1}{2}$ Pauli matrices,
$\Delta$ is the anisotropy parameter
and we assume periodic boundary conditions.
If we identify, in the $\sigma^z$-basis, a spin up or spin
down at a given site $i$ as a particle or hole at this site,
the XXZ Hamiltonian can be written iquivalently as
\begin{eqnarray}
H(\Delta) = - \sum_{i+1}^L P_{0}(c_i^+c_{i+1} + c_{i+1}^+c_i +
\Delta n_in_{i+1})P_0 \,,
\end{eqnarray}
where $c_i^+$ and $c_i$ are spinless creation and annihilation
operators and $n_i = c_i^+c_i$ is the density of particles at site $i$.
The projector $P_o$ in (1) forbids two particles to occupy the same
position in space (fermion like). The models under consideration are
obtained by a generalization of the operator $P_0$ and are given by
\begin{eqnarray}
H(\Delta) = - \sum_{i=1}^L P_t(c_i^+c_{i+1} + c_{i+1}^+c_i +
\Delta n_in_{i+ t +1})P_t\,, \hspace{1cm} t = 0,1,2,...
\end{eqnarray}
where $P_t$ now forbids two particles at distance less or equal to
$t$ $(0,1,2,...)$. The static interaction, controlled by $\Delta$,
acts only when two particles are at the closest possible positions
$(i,i+t+1)$. The case $t=0$ recovers the standard XXZ chain (2) or (1).
In terms of Pauli matrices we can also write the equivalent Hamiltonian
\begin{eqnarray}
H_{t}(\Delta) = -\frac{1}{2}\sum_{i=1}^LP_t(\sigma_i^x\sigma_{i+1}^x +
\sigma_i^y\sigma_{i+1}^y + \Delta\sigma_i^z\sigma_{i+t+1}^z)P_t \,,
\end{eqnarray}
where now $P_t$ projects out from the Hilbert space the states
(in the $\sigma^z$-basis) where two up spins $(\sigma_t =1)$ are at distance
smaller or equal to $t$. In coming from (1) to (2) or from (3) to (4)
we neglect a term proportional to a magnetic field. This term can be
trivially added in the above Hamiltonian since the magnetization
$\sum_{i=1}^L \sigma_i^z$ or the total number of particles
$n = \sum_{i=1}^zn_i$ are good quantum numbers. In terms of Pauli
matrices the operator $P_0 = 1$ and for $t>0$ it is non-local and can
be written in the $\sigma^z$-basis as
\begin{eqnarray}
P_t = \prod_i \left[\ffrac{1}{2}(1-\sigma_i^z) + \ffrac{1}{2}(1 +
\sigma_i^z) \prod_{l=1}^t \ffrac{1}{2}(1 -\sigma_{i+l}^z) \right]\,.
\end{eqnarray}
The Hamiltonians (2), although looking artificial, may describe interesting
physical situations where the states with up spins are associated with larger
excluded volume than those of down spins. This is precisely the
case when we consider the time fluctuations due to diffusion
of molecules with size $t+1$ (in units of the lattice spacing parameter)
on a lattice. Interpreting the master equation associated to this
problem as a Schr\"odinger equation in Euclidean time,\,\cite{2} the
associated Hamiltonian is given by the isotropic ferromagnetic version
of (3), namely $H_t(\Delta = +1)$. The anisotropic ferromagnetic version
$(\Delta>1)$ is related to the asymmetric diffusion of the molecules of
size $(t+1)$.\,\cite{3}
\section{The Bethe ansatz equations}
It is important to observe that similarly to the XXZ chain the
$z$-magnetization $m = \sum_i^L\sigma_i^z$, or the total number of up
spins $n$ (or the number of particles in (2) and (3)) is a conserved
quantity for the Hamiltonian (4). Consequently the Hilbert space can be
separated into block-disjoint sectors labelled by $n =0, 1, \ldots, L/(t+1)$.
In a given sector $n$ the wave functions of (4) can be written
\begin{eqnarray}
\Psi = \sum_{x_{1}=1}^{\bar L}\ldots
\sum_{x_{i+1}=x_i +t+1}^{\bar L +i(t+1)}\ldots\sum_{x_n=x_{n-1}+t+1}^{ L}
a(x_1,x_2,\dots,x_n)|x_1,x_2,\dots,x_n \rangle\,,
\end{eqnarray}
where $\bar{L} = L-n+1-t(n-1)$ and $a(x_1,...,x_n)$ are the amplitudes
of the spin configurations $|x_1,x_2,...,x_n\rangle$, having spin down at
the coordinates $x_i (i=1,2,...,n)$. We can calculate the above
amplitudes and the corresponding eigenenergy of (4) by applying the
Bethe ansatz
\begin{eqnarray}
a(x_1,x_2,...,x_n) = \sum_{P} A_{k_{p_1},k_{p_2},...,k_{p_n}}
e^{i(k_{p_1}x_1 + ... + k_{p_n}x_n)} \,,
\end{eqnarray}
where the summation is over all permutations $P = (p_1,p_2,...,p_n)$
and $A_{\{k\}}$ as well the wave numbers $\{k\} = {k_1,k_2,...,k_n}$
should be calculated from the eigenvalue equation for (4). The solution
follows the standard coordinate Bethe ansatz.\,\cite{4}
The validity of (7) for the case $n=1$ is immediate due to the
translational invariance of (4), and in
this case the energies are $E_k = -2\cos k - \frac{\Delta}{2}(L-4)$,
with $k = 2\pi p/L$ $(p =0,1,2,...,L-1)$. The case of general $n$ can be
verified as in the standard XXZ chain $(t=0)$. The energies and
momentum are given by
\begin{eqnarray}
E_{{k}} = -2\sum_{i=1}^n \cos 2k_i - \frac{\Delta}{2}(L - 4n)\,, \quad
P = \sum_{i=1}^n k_i \,.
\end{eqnarray}
The wave-numbers $\{k_i\}$ are obtained by solving the Bethe ansatz
equations
\begin{eqnarray}
e^{ik_j L} = (-1)^{n-1}\prod_{l=1}^n e^{it(k_j-k_l)}
\frac{1-2\Delta e^{ik_j}+e^{i(k_j+k_l)}}{1-2\Delta e^{ik_l}+e^{i(k_j+k_l)}}\,.
\end{eqnarray}
These equations at $t=0$ give us the well-known Bethe ansatz equations
for the XXZ chain. For $t>0$ these equations in a given sector
specified by $n$ up spins and momentum $P=\frac{2\pi}{L}p$ $(p=0,1,...,L-1)$
are the same as those of an XXZ chain also with $n$ up spins but
lattice size $L'=L-tn$, momentum $P' = \frac {2\pi}{L-tn}p$
$(\mbox{mod} 2\pi)$ and twisted boundary condition\,\cite{5}
\begin{eqnarray}
\sigma_{L+1}^x\pm i\sigma_{L+1}^y = e^{i\Phi}(\sigma_1^x \pm i\sigma_1^y)\,,
\quad \sigma_{L+1}^z = \sigma_1^z\,, \quad \Phi = \frac{2\pi}{L}p\,.
\end{eqnarray}
We will now study the phase diagram of these models. Calculating the
eigenspectra of (4) numerically by
brute-force diagonalization on small lattices, we see that in
absence of an external magnetic field the groundstate, for finite
chains will belong to a sector where the magnetization $M=2n-L$
depends on the lattice size $L$, exclusion parameter $t$ and
interaction $\Delta$, as long as $t>0$. This is distinct from the
standard XXZ case $(t=0)$, where the groundstate is always in
the half-filled sector $n=\frac{L}{2}$, with zero magnetization
for all values of $L$ and $\Delta < 1$. In order to have a well defined
thermodynamic limit for given values of $t$ and $\Delta$ it is
necessary to work with fixed values of the magnetization per site
$m=\frac{M}{L}=\frac{2n}{L}-1$, or equivalently for fixed values of
the density of up spins $\rho =\frac{n}{L}$.
We can force the groundstate to belong to such a sector with
$n=\rho L$ by adding in (4) an external magnetic field
$-h\sum_{i=1}^{L}\sigma_i^z$, which will depend on the values of
$t$, $\Delta$ and $\rho$.
It is important to remark that even when $\Delta =0$ we do not
have a completely free system for $t>0$, since in (9) a phase
connecting different values of ${k_j}$ is always present.
Nevertheless the Bethe-ansatz equations can be solved exactly
for finite lattices, and is then instructive to calculate the
eigenspectra at $\Delta = 0$.
\section{The operator content at $\Delta=0$}
Let us consider $H_t(\Delta =0,h)$, with a given value of the magnetic
field $h=h(\rho)$ that produces a fixed density
$\rho =\frac{n}{L}$ $(0\leq\rho\leq\frac{1}{t+1})$ of up spins.
Taking the logarithm of both sides of (9) we obtain
\begin{eqnarray}
Lk_j = \pi(\rho L-1) + \sum_{l=1}^{\rho L}t(k_j-k_l) + 2\pi m_j\,,\quad
j =1,2,\ldots \,,
\end{eqnarray}
where $m_j$ $(j=1,2,...,n)$ are integers. The solution of these
equations are given by the set $\{k_j\}$,
\begin{eqnarray}
k_j = \Theta + \frac{2\pi}{L(1-t\rho)}m_j\,, \quad
\Theta = \frac{\pi(\rho L- 1)-tP}{L(1-\rho t)}\,, \quad j = 1,2,\ldots,n\,,
\end{eqnarray}
where $\{m_j\}$ are distinct integers and $P$ is the momentum
associated to the state
\begin{eqnarray}
P = \sum_{j=1}^{\rho L}k_j = \frac{2\pi}{L}\tilde p \; (\mbox{mod} 2\pi)\,,
\quad \tilde p \in Z \,.
\end{eqnarray}
The groundstate has zero momentum and corresponds to the set $\{k_j\}$,
where
\begin{eqnarray}
k_j = \frac{2\pi}{L(1-t\rho)}j\,, \quad j = -\frac{\rho L -1}{2},
-\frac{\rho L -1}{2} +1, \ldots ,\frac{\rho L -1}{2} \,.
\end{eqnarray}
Using (14) in (8) we obtain the groundstate energy of (4), in
the presence of an external magnetic field,
\begin{eqnarray}
E_0(L,\rho) = -2\sin\left(\frac{\pi\rho}{1-t\rho}\right)
/
\sin\left(\frac{\pi}{L(1-t\rho)}\right) - h(2\rho-1)L\,.
\end{eqnarray}
The magnetic field $h=h(\rho)$ is to be calculated by imposing
that when $L \to \infty$, $e_{\infty} = \frac{E_0(L,\rho)}{L}$
will be a minimum as a function of $\rho$, i.e.
\begin{eqnarray}
h =h(\rho) = \frac{t}{\pi}\sin\left(\frac{\pi\rho}{1-t\rho}\right)-
\frac{1}{1-t\rho}\cos\left(\frac{\pi \rho}{1-t\rho}\right) \,.
\end{eqnarray}
This result also tells us that in the absence of a magnetic field the
groundstate should belong to the sector with magnetization
$m_0 = \frac{M_0}{L} = 2\rho_0 - 1$, where $h(\rho_0) = 0$.
In the case $t=0$, $\rho_0 =\frac{1}{2}$ and $m_0 =0$. For $t>0$
we should solve $h(\rho_0) = 0$ numerically; for example,
$\rho_0 = 0.3008443$, $\rho_0 = 0.2193388$ and $\rho_0 = 0.07354512$
for $t=1,2$ and $10$, respectively.
As we can see these values cannot be predicted by symmetry
arguments on (4) or (9).
The leading finite-size corrections of (15) are given by
\begin{eqnarray}
\frac{E^0 (L,\rho)}{L} = e_{\infty} - \frac{\pi c}{6L^2}\xi
+ o(L^{-2})\,,
\end{eqnarray}
where
\begin{eqnarray}
e_{\infty} = -2 \,\frac{1-\rho t}{\pi}\,\sin\left(\frac{\pi \rho}
{1 -\rho t}\right) - h(2\rho-1)\,, \quad
c\,\xi = \frac{2}{1-t\rho}\sin\left(\frac{\pi\rho}{1-\rho t}\right)\,.
\end{eqnarray}
These relations are a clear indication that the model is critical
for arbitrary values of $t$ and $\rho$, since (17) is the
expected behavior of a massless conformally invariant critical
point with $c$ and $\xi$ being the conformal anomaly and sound
velocity, respectively.\,\cite{6} The sound velocity can be
calculated from the energy momentum dispersion relation.
The lowest eigenenergy in the sector with $n = \rho L$ spins up,
having momentum $\frac{2\pi}{L}\tilde p$ $(\tilde p = 1,2,...)$ can be
obtained by choosing the particular solution of (11),
\begin{eqnarray}
k_j &=& \frac{2\pi}{L(1-t\rho)} j - \frac {2\pi t}{(1-t\rho)L^2}\,,
\nonumber \\
j &=& -\frac{\rho L-1}{2}, -\frac{\rho L-1}{2} + 1, \ldots,
\frac{\rho L-3}{2}, \frac{\rho L-1}{2} + \tilde p \,,
\end{eqnarray}
and the corresponding energy is given by
\begin{eqnarray}
\lefteqn{
E^{(\tilde p)}(L,\rho) = E^0(L,\rho)\; +} \nonumber\\
&&4\,\sin\left(\frac{\pi(\rho L + \tilde p - 1)}{(1+t\rho)L}\right)
\sin\left(\frac{\pi\tilde p}{L(1-t\rho)}\right) + o(L^{-1}) \,.
\end{eqnarray}
This gives us, for large $L$, a linear dispersion relation
with sound velocity
\begin{eqnarray}
\xi = \frac{2}{1-t\rho}\sin\left(\frac{\pi\rho}{1-t\rho}\right),
\end{eqnarray}
and consequently the conformal anomaly is $c=1$ for
all values of $t$ and $\rho$.
The conformal invariance of the infinite critical system,
beyong relation (17), also predicts the finite-size
corrections of excited states.\,\cite{7} For each primary
operator, with dimension $x_{\phi}$ and spin $s_{\phi}$,
in the operator algebra of the critical system there
exists an infinite tower of eigenstates of the Hamiltonian,
whose energy $E_{m,m'}^{\phi}$ and momentum $P_{m,m'}^{\phi}$,
in a periodic chain are given by
\begin{eqnarray}
E_{m,m'}^{\phi}(L) &=& E^0 + \frac{2\pi\xi}{L}(x_{\phi}+m+m')+
o(L^{-1})\,,\nonumber\\
P_{m,m'}^{\phi} &=& (s_{\phi}+m-m')\frac{2\pi}{L}\,,
\end{eqnarray}
where $m,m' = 0,1,2, \ldots $.
The groundstate corresponds to the identity operator
$x_{I}=0, s_{I}=0$ and the tower of states
$E_{m,m'}^{(\phi =I)} = \frac{2\pi}{L}(m-m')$
in our model are obtained by decreasing (increasing) the
values of the quasi-momenta $k_i$ in the set $\{k_i\}$ given in
(14). After the calculation of the eigenenergies
corresponding to many of these descendants we convince ourselves that
the degeneracy of such levels are given by
$d_{m,m'}^{\phi =I} = P(m)P(m')$ where $P(m)$ is the
number of integer partitions of $m$.
Another conformal dimension in the sector $n=\rho L$
can be obtained by taking negative (positive)
values of $k_j$ in the groundstate root
configuration (14) and replacing them by positive
(negative) values. For example, the lowest of these
eigenenergies corresponds to the solution
\begin{eqnarray}
k_j &=& \frac{2\pi}{(1-\rho t)L}(j - \rho t)\,, \nonumber \\
j &=& -\frac{\rho L-1}{2} +1, -\frac{\rho L-1}{2} + 2,\ldots ,
\frac{\rho L-1}{2}+1\,,
\end{eqnarray}
and has the value
\begin{eqnarray}
E = -2\cos\left(\frac{2\pi}{L}\right)\sin\left(\frac{\pi\rho}{1-\pi\rho}
\right)/ \sin\left(\frac{\pi}{(1-\rho t) L}\right)
\end{eqnarray}
and a macroscopic momentum $P=2\pi\rho$. Comparing
the finite-size corrections of (24) with (22) we see
that this energy corresponds to an operator with dimension
$x_{0,1}$ and spin $s_{0,1}$ given by
\begin{eqnarray}
x_{0,\bar m} = (1-\rho t)^2 {\bar m}^2\,, \quad
s_{0,\bar m} = n\bar m = \rho L\bar m\,,
\end{eqnarray}
respectively. The dimensions $x_{0,\bar m}$, with
$\bar m$ positive (negative) integer are obtained by taking
$\bar m$ of the lowest (highest) values of $k_j$ in (14)
and replacing them by the lower (higher) available
positive (negative) values. As in the
case of the identity operator the descendants of the above
operators, with dimension $x_{\bar m}+m+m'$ $(m,m'\in Z)$
are obtained by increasing (decreasing) the positive
(negative) values of $\{k_j\}$ in the configuration that produces
$x_{0,\bar m}$. As before we convince ourselves that the degeneracy
of these states is $P(m) P(m')$.
There exist also dimensions associated to eigenstates produced by the
addition $(\bar n = -1,-2,\ldots)$ or subtraction $(\bar n =1,2,\ldots)$
of spins up in the groundstate
sector. In this case $n = \rho L + \bar n$ and the lowest eigenenergy
is obtained from the set $\{k_j\}$ with
\begin{eqnarray}
k_j = \frac{2\pi}{L(1-\rho t -\bar n)}j\,, \quad j = -j_{\rm max},
-j_{\rm max}+1,\ldots, j_{\rm max}-1, j_{\rm max}\,,
\end{eqnarray}
where $j_{\rm max}=\frac{\rho L +\bar n -1}{2}$. These are
zero-momentum eigenstates with energies
\begin{eqnarray}
E^{\bar n} &=& -\;2\sin\left(\frac{\pi(\rho L +\bar n)}
{L(1-\rho t) -\bar n}\right)/
\sin\left(\frac {\pi}{L(1-\rho t) -\bar n}\right)\nonumber\\
&\phantom{=}& -\; h(2(\rho L+\bar n)-L)\,.
\end{eqnarray}
The finite-size corrections, when compared with (22) give us
the conformal dimensions
\begin{eqnarray}
x_{\bar n, 0} = \frac{\bar n^2}{4(1-t\rho)^2}\,.
\end{eqnarray}
In the general case, where we combine $\{k_j\}$ configurations
which give (25) and (26) we obtain the conformal dimensions
\begin{eqnarray}
x_{\bar n,\bar m} = \bar n^2 X_p + \frac{\bar m^2}{4X_p}\,,\quad
X_p = \frac{1}{4(1-t\rho)^2}\,\quad \bar n, \bar m \in Z
\end{eqnarray}
of primary operators with spin $s_{\bar n,\bar m} =
(\rho L + \bar n)\bar m$. If we calculate the eigenenergies
which correspond to the descendants with dimensions
$x_{\bar n,\bar m}^{m,m'} = x_{\bar n,\bar m}+m+m'$ we
obtain their degeneracy $ d_{\bar n,\bar m}^{m,m'} =
P(m) P(m')$. The dimensions (29) are the same as the dimensions
$x_{\bar n,\bar m}^G = \frac{\bar n^2}{4\pi K} + \pi K\bar m^2$
of operators with vorticity $\bar m$ and spin-wave number $\bar n$
in a Gaussian model\,\cite{8} with coupling constant
$K = (1-\rho t)^2/\pi$. The degeneracy $d_{\bar n,\bar m}^{m,m'}$
imply that the operator content of $H_t(\Delta = 0)$ is given
in terms of U(1) Kac-Moody characters. The situation is completely
analogous with that of the standard XXZ chain $(t=0)$ in its
massless regime.\,\cite{9} However, expression (29) shows different
dependences on the density $\rho$ (or magnetization),
depending on the value of $t$. In particular for
$\Delta = 0$ the dimensions are $\rho$-independent only in the
free-fermion case $t=\Delta = 0$.
The result (29) shows that the net effect of the parameter $t >0$ at
$\Delta =0$ is similar to that of the anisotropy parameter $\Delta \neq 0$
in the standard XXZ chain $(t = 0)$. In both cases the exponent $X_p$ changes
continuously with the density $\rho$ of up spins.
It is interesting to remark that when $\rho \rightarrow 0$ $X_p = 1/4$, for
all values of $t$. This independence ot $t$ arises since the restrictions
imposed by $t >0$ are not important due to the small amount of up spins in
this limit. The other limit of $X_p$ at high density,
$\rho \rightarrow 1/(t+1)$, can also be understood as follows. If we start
with the maximum density $\rho = 1/(t+1)$ and we take out one spin up,
the ``hole" produced will have the same effective
motion as that produced by adding $t+1$ ``particles" in the low-density
limit. Then according to (28) we should obtain the limiting value
$X_p = (t+1)^2/4$.
\section{The operator content at $\Delta \neq 0$}
Let us consider the case where $\Delta\ne0$. The Bethe ansatz
equations (9) supplemented by numerical calculations give the
same phase diagram for all values of $t$. For $\Delta>1$
the groundstate is $(t+2)$-degenerate, where $(t+1)$ states
are in the sector with maximum number of up spins
$(n = \frac{L}{1+t})$ and the last in the sector with no spins
up $(n=0)$. For $\Delta < 1$ we expect a massless behavior
similar to that in the XXZ chain. In fact, exploring the
similarity of our equations for arbitrary $t$, with those of
the XXZ chain $(t=0)$, we can apply the method of Ref.~10
to extract the finite-size corrections for arbitrary values of
$\Delta$ and $\rho$. The dimensions we obtained are given by
(29), where $X_p$ is now given by
\begin{eqnarray}
X_p = \frac{1}{4}(1-t\rho)^{-2}{\eta }^{-2}(U_0)\,,
\end{eqnarray}
where $U_0$ and $\eta(U)$ are obtained by solving some integral
equations. These equations for $-1\leq \Delta=-\cos\gamma \leq 1$ are
given by
\begin{eqnarray}
Q(U) &=& \frac{1}{2\pi}\frac{\sin\gamma}{\cos U-\cos\gamma}
-\frac{1}{2\pi}\int_{-U_0}^{U_0}
\frac{\sin(2\gamma) Q(U')}{\cosh(U-U')-\cos(2\gamma)}dU'\,, \\
1 &=& \eta(U) + \frac{1}{2\pi}\int_{-U_0}^{U_0}
\frac{\sin (2\gamma)\eta(U')}{\cosh(U-U')-\cos (2\gamma)}dU'\,,
\end{eqnarray}
while for $\Delta = -\cosh\gamma < -1$ they are given by
\begin{eqnarray}
Q(U) &=& \frac{1}{2\pi}\frac{\sinh\lambda}{\cosh\lambda - \cos U}
-\frac{1}{2\pi}\int_{-U_0}^{U_0}
\frac{\sinh (2\lambda) Q(U')}{\cosh (2\lambda) -\cos(U-U')}dU'\,, \\
1 &=& \eta(U) + \frac{1}{2\pi}\int_{-U_0}^{U_0}
\frac{\sinh (2\lambda) \eta(U')}{\cosh (2\lambda) -\cos(U-U')}dU'\,.
\end{eqnarray}
In Eqs.~(31)-(34) $U_0$ is calculated by imposing
\begin{eqnarray}
\int_{-U_0}^{U_0}Q(U) = \cases{
\frac{\rho}{1-t\rho}, & $0 \leq \rho \leq \frac{1}{2+\rho}$\,, \cr
\frac{1-(t+1)\rho}{1-t\rho}, &$\frac{1}{2+\rho}\leq \rho \leq
\frac{1}{1+\rho}$\,.\cr}
\end{eqnarray}
The value of $X_p$ is obtained by numerically solving Eqs.~(31)-(35).
In Fig.~1 we plot $X_p$ at $\Delta = -\frac{\sqrt2}{2}$ for
$t = 0,1,2$ and 3, while in Fig.~2 $X_p$ is shown for $t =2$ and some
values of $\Delta$ $(\-1 \leq \Delta \leq 1)$.
\begin{figure}[htbp]
\begin {center}
\mbox{\psfig{figure=fig1-xxz.ps,height=2.0in,width=2.0in,angle=-90}}
\end{center}
\caption{
The value of $X_p$ as a function of $\rho =\frac{n}{L}$ at
$\Delta = -\frac{\sqrt 2}{2}$, for $t=0$ (a), $t=1$ (b), $t=2$ (c)
and $t=3$ (d). }
\label{FIG1}
\end{figure}
\begin{figure}[htbp]
\begin {center}
\mbox{\psfig{figure=fig2-xxz.ps,height=2.0in,width=2.0in,angle=-90}}
\end{center}
\caption{The value of $X_p$ as a function of $\rho =\frac{n}{L}$ for $t=2$
and anisotropy $-1 \leq \Delta = -\cos \gamma \leq 1$, with
(a) $\gamma = \frac{\pi}{10}$, (b) $\gamma=\frac{\pi}{4}$, (c)
$\gamma = \frac{\pi}{3}, (d) \gamma = \frac{\pi}{2}$ and (e)
$\gamma = \frac{2\pi}{3}$.}
\label{FIG2}
\end{figure}
These results coincide with those calculated earlier for the standard
XXZ chain $(t =0)$.\,\cite{10} At low density $(\rho \rightarrow 0)$
and high density $(\rho (t + 1) \rightarrow 1)$ the exponent
$X_p = 1/4$ and $X_p = (t + 1)^2/4$, respectively. These results are
the same as those of the $\Delta =0$ case, since in these limits the
distribution of ``particles" at $\rho \rightarrow 0$ or ``holes"
at $\rho \rightarrow 1/(t + 1)$ are so diluted as to render the effect
of the interaction term $\Delta$ negligible. From (30)-(34) it is
possible to relate $X_p(t,\Delta)$ for different values of $t$ and $\rho$.
In particular at the point $\rho = 1/(2 + t)$, which corresponds to the
``half-filled" density for these models, we can show that
\begin{equation}
x_p(t,\frac{\rho}{2+t}) = \ffrac{1}{4}(t+2)^2 X_p(t=0,\rho=\ffrac{1}{2}) =
\ffrac{1}{8}(t+2)^2(\pi -\gamma)\,,
\end{equation}
for $-1 \leq \Delta =-\cos{\gamma} \leq 1$, where we have used the known
results\,\cite{5} for the XXZ chain in the absence of an external
magnetic field.
In Fig.~3 we plot $X_p$ for $t = 3$ and for some values of $\Delta \leq -1$,
while in Fig.~4 $X_p$ is shown at $\Delta = -2$ for $t =0,1,2$ and 3.
\begin{figure}[htbp]
\begin {center}
\mbox{\psfig{figure=fig3-xxz.ps,height=2.0in,width=2.0in,angle=-90}}
\end{center}
\caption{The curves $X_p(\rho)$ for $t=3$ and (a) $\Delta = -20$, (b)
$\Delta = -4$, (c) $\Delta = -2$. The expected curve at
$\Delta \rightarrow -\infty$ $X_p = 1/[4(1-4\rho)^2]$ is also shown.}
\label{FIG3}
\end{figure}
\begin{figure}[htbp]
\begin {center}
\mbox{\psfig{figure=fig4-xxz.ps,height=2.0in,width=2.0in,angle=-90}}
\end{center}
\caption{The curves $X_p(\rho)$ at $\Delta = -2$ for (a) $t=0$, (b) $t=1$,
(c) $t=2$ and (d) $t=3$.}
\label{FIG4}
\end{figure}
As before in the low density $(\rho \rightarrow 0)$ and high density $(\rho
\rightarrow 1/(t+1))$ limits we recover the $\Delta =0$ case. In Fig.~3 we
also notice that at the ``half-filling" density $\rho = 1/(2+t)$ the
exponent $X_p=(t+2)^2/4$, independently of the value of $\Delta$. A physical
explanation for this behavior is the following. Strictly speaking at the
``half-filling" point $\rho = 1/(t+2)$ the model has a gap for $\Delta < -1$,
as long as the magnetic field is zero. When the magnetic field increases,
reaching a critical value $h = h_c(\Delta,t)$ this gap vanishes. Therefore
the main physical effect at $\rho = 1/(2+t)$ is just the vanishing of the gap
and not the interaction $\Delta$
itself. In the limit $\Delta \rightarrow -\infty$
and $\rho < 1/(2+t)$ the model with a given value of $t$ behaves effectively
as a model with $t' = t+1$ and $\Delta =0$, since the spins up have infinite
repulsion and will be located in positions with no interactions among them
(distances greater than $t+1$). This can also be verifyed directly by setting
$\Delta \rightarrow -\infty$ in the Bethe ansatz equations (9).
Consequently for $\Delta \rightarrow -\infty$ and $\rho < 1/(2+t)$ we should
have $X_p = 1/[4(1-\rho(t+1))^2]$. In particular at the ``half-filling"
density $\rho = 1/(2+t)$ we obtain the value $X_p = (t+2)^2/4$. In Fig.~3
the limiting curve $(\Delta \rightarrow -\infty)$, $X_p = 1/4(1-4\rho)^2$,
for $t=3$, can be compared with that of $\Delta =-20$, obtained by solving \
(30) and (33)-(35). For $\Delta \rightarrow -\infty$, but $\rho >1/(2+t)$
this argument does not apply and we should recover the value $X_p = (t+1)^2/4$
at $\rho =1/(t+1)$. We believe the results reported here in the $\Delta <-1$
region are unknown even for the standard XXZ chain $(t = 0)$.
\section{Generalizations}
Before concludiing let us discuss an extension
of our results. We only considered the cases where $t\geq 0$
but what should be the quantum chain that would correspond
to $t=-1$? In terms of particles, like in (2) this
corresponds to the situation where we can have an arbitrary
number of particles on a site. In fact we were able to find an
exactly integrable quantum chain in this class, namely
\begin{eqnarray}
H_{-1}(\Delta) &=& -\sum_{j=1}^L\{\sum_{n=0}^{\infty}
\sum_{m=n+1}^{\infty}(E_j^{n,m}E_{j+1}^{m,n} +
E_j^{m,n}E_{j+1}^{n,m})\nonumber \\ & &+ \sum_{n=0}^{\infty}\Delta
(n-1)E_j^n E_{j+1}^n\} \,,
\end{eqnarray}
where $E^{n,m}$ are $\infty$-dimensional matrices located at
the lattice sites with elements $E^{n,m})_{i,j} =
\delta_{n,i}\delta_{m,j}$. The Bethe ansatz equations
are given by (9) with $t=-1$. This
Hamiltonian commutes with the operator
\begin{eqnarray}
\hat N =\sum_{j=1}^L\sum_{n=1}^{\infty}nE_j^{n,n}\,,
\end{eqnarray}
which gives the total number of particles. The Hamiltonian (37)
in a given sector with $N$ particles can be written in terms of
spin-$S$ SU(2) matrices with $S=(N-1)/2$.
\section*{Acknowledgments}
This work was supported in part by CNPq and FINEP -- Brazil,
and also by the Russian Foundation of Fundamental
Investigations under Grant No.RFFI 97--02--16146.
\section*{References}
|
\section{Introduction}
In \cite{K}, Kontsevich proved a classification theorem for deformation quantizations of $C^{\infty}(M)$ where $M$ is a smooth manifold. This theorem asserts that the set of isomorphism classes of deformations of $C^{\infty}(M)$ is in one-to-one correspondence with the set of equivalence classes of formal Poisson structures on $M$. This theorem follows from a more general theorem: the differential graded algebra ${\frak g}^{\bullet}_S(M)$ of multi-vector fields on $M$ is equivalent to the differential graded algebra ${\frak g}^{\bullet}_G(C^{\infty}(M))$ of Hochschild cochains of $C^{\infty}(M)$ (we recall exact definitions and statements from \cite{K} and references thereof in section \ref{s:foth}). In other words, the algebra of Hochschild cochains is {\it {formal}} (equivalent to its cohomology).
In this paper, we state a formality conjecture about the Hochschild and cyclic chain complexes of the algebra $C^{\infty}(M)$. It is well known that for any algebra $A$ the Hochschild chain complex $C_{\bullet}(A,A)$ and the negative cyclic complex $CC^-_{\bullet}(A)$ are modules over the Lie algebra of Hochschild cochains ${\frak g}^{\bullet}_G(A)$. Therefore, by virtue of the Kontsevich formality theorem, both the Hochschild (resp. negative cyclic) complex and the graded space of differential forms (resp. de Rham complex) are strong homotopy modules over ${\frak g}^{\bullet}_G(C^{\infty}(M))$. We conjecture that those modules are equivalent in an appropriate sense (cf. section \ref{s:foc}), or, in other words, that Connes' quasi-isomorphism from \cite{C}, \cite{L} is, in the right sense, ${\frak g}^{\bullet}_G(C^{\infty}(M))$-equivariant. As in \cite{K}, the correct language for stating our conjectures is that of homotopical algebra of Stasheff.
We derive several consequences from the above conjecture in section \ref{hocyde}. First, we compute the Hochschild and cyclic homology of deformed algebras given by the classification theorem of Kontsevich. In particular, we compute the space of traces on such a deformed algebra. Then, in subsection \ref{ss:ahat}, we show how to construct the $\widehat{A}$ class of an arbitrary Poisson manifold. In the case of a regular Poisson structure, this class is, conjecturally, the $\widehat{A}$ class of the tangent bundle to the foliation of symplectic leaves.
Finally, in section \ref{s:ga} we outline a possible proof of our conjectures, as well as their generalization, along the lines of a recent work of Tamarkin \cite{T}.
I am thankful to P. Bressler, G. Halbout, M. Kontsevich, R. Nest, J. Stasheff, D. Tamarkin, and S. Voronov for helpful discussions.
\section{Formality theorem of Kontsevich} \label{s:foth}
\subsection{Classification of star products} \label{ss:clasta}
Let $A_0$ be an associative unital algebra over a commutative unital ring $k$. {\it {A deformation}} \cite{G} of $A_0$ is a formal power series
$$a*b=ab+\sum_{m=1}^{\infty}t^m P_m(a,b)$$
where $P_m: A_0\times A_0\rightarrow A_0$ are $k$-bilinear forms such that
a) The product $*$ is associative
b) $1*f=f*1=f, \; \; f\in A_0$.
{\it {An isomorphism}} of two deformations $*$, $*'$ is a formal power series $T(a)=a+\sum_{m=1}^{\infty}t^m T_m(a)$ such $T(a*b)=T(a)*'T(b)$ for $a,b$ in $A_0$.
Let $M$ be a $C^{\infty}$ manifold. {\it {A deformation quantization}} of $C^{\infty}(M)$, or {\it {a star product}}, is a deformation of $A_0= C^{\infty}(M)$ such that $P_m$ are bidifferential operators \cite{BFFLS}. {\it {An isomorphism}} of two star products is an isomorphism of corresponding deformations such that the operators $T_m$ are differential.
Given a star product on a $C^{\infty}$ manifold $M$, one defines a Poisson bracket on $C^{\infty}(M)$ by
\begin{equation}
\{f,g\}=P_1(f,g)-P_1(g,f) \label{eq:pbr}
\end{equation}
For a bivector field $\pi$, put
\begin{equation}
\{f,g\}_{\pi}=<\pi, df \wedge dg> \label{eq:pbrpi}
\end{equation}
It follows from associativity of $*$ modulo $t^2$ that the Poisson bracket (\ref{eq:pbr}) is necessarily of the form $\{f,g\}_{\pi_0}$ for some bivector field $\pi_0$.
Recall that for a bivector field $\pi$ there exists unique trivector field $[\pi,\pi]_S$ such that
\begin{equation} \label{eq:bras}
\{f,\{g,h\}_{\pi}\}_{\pi} + \{g,\{h,f\}_{\pi}\}_{\pi}+\{h,\{f,g\}_{\pi}\}_{\pi} = <[\pi,\pi]_S, df \wedge dg \wedge dh>
\end{equation}
for any smooth functions $f$, $g$, and $h$. The expression $[\pi,\pi]_S$ is quadratic in $\pi$. The polarization of $[\pi,\pi]_S$ is a symmetric bilinear form $[\pi,\psi]_S$ with values in the space $\Gamma(M,\wedge^3T)$ of trivector fields. A bivector field $\pi$ such that $[\pi,\pi]_S=0$ is by definition {\it {a Poisson structure}} on $M$. It follows from associativity of $*$ modulo $t^3$ that for any star product $*$, the bivector field $\pi_0$ (formula (\ref{eq:pbrpi})) is a Poisson structure.
{\it {A formal Poisson structure}} is by definition a formal power series $\pi = \sum_{m=0}^{\infty}t^{m+1} \pi _m$ such that $[\pi,\pi]_S = 0$ in $\Gamma(M,\wedge^3T)[[t]]$.Every Poisson structure $\pi$ defines a formal Poisson structure $t\pi$. Two formal Poisson structures $\pi$ and $\pi '$ are {\it {equivalent}} if there is a formal power series $X=\sum_{m=1}^{\infty}t^m X_m$ such that $\pi ' = \operatorname{exp} (L_X) \pi$.
\begin{thm} \label{thm:konts}
(Kontsevich, \cite{K}). There is a bijection, natural with respect to diffeomorphisms, between the set of equivalence classes of formal Poisson structures on $M$ and the set of isomorphism classes of deformation quantizations of $C^{\infty}(M)$.
\end{thm}
If $\pi = \sum_{m=0}^{\infty}t^{m+1} \pi _m$ is a formal Poisson structure, we will denote by $*_{\pi}$ a star product from the equivalence class corresponding to $\pi$ by the above theorem. The Poisson structure associated to $*_{\pi}$ by formulas (\ref{eq:pbr}) and (\ref{eq:pbrpi}) will then be equal to $\pi_0$.
\subsection{Hochschild cochains} \label{ss:coch}
For a unital algebra $A$ over a commutative unital ring $k$, for $n\geq 0$ let
\begin{equation} \label{eq:hoco}
\tilde{C}^n(A,A) = Hom (A^{\otimes n}, A)
\end{equation}
If $A=C^{\infty}(M)$, we require that $\tilde{C}^n(A,A)$ consist of those maps
from $A^{\otimes n}$ to $A$ which are multi-differential.
Define {\it {the Gerstenhaber bracket}} (cf. \cite{G})
$$[\;,\;]_S: \tilde{C}^n(A,A) \otimes \tilde{C}^m(A,A) \rightarrow \tilde{C}^{n+m-1}(A,A)$$
as follows. For $D \in \tilde{C}^n(A,A)$ and $E \in \tilde{C}^m(A,A)$ put
\begin{equation} \label{eq:circ}
\gathered
(D\circ E)(a_1,\dots,a_{n+m-1})=\\
=\sum_{j = 0}^{n-1}
(-1)^{(m-1)j}
D(a_1,\dots,a_j,
E(a_{j+1},\dots,a_{j+m}),\dots);
\endgathered
\end{equation}
\begin{equation} \label{eq:lbra}
[D, \; E]_S= D\circ E - (-1)^{(n-1)(m-1)}E\circ D
\end{equation}
The bracket $[\;,\;]_S$ turns $\tilde{C}^{\bullet +1}(A,A)$ into a graded Lie algebra (\cite{G}). Put
\begin{equation} \label{eq:m}
m(a,b)=ab
\end{equation}
for $a,b \in A$.
One has $[m,m]=0$ (this is equivalent to $m$ being associative), so the operator \begin{equation} \label{eq:del}
\delta = [m,?]: \tilde{C}^{\bullet}(A,A) \rightarrow \tilde{C}^{\bullet +1}(A,A)
\end{equation}
satisfies $\delta ^2 = 0$. The complex $(\tilde{C}^{\bullet}(A,A), \delta)$ is called the unnormalized Hochschild cochain complex of $A$ with coefficients in the bimodule $A$. The cohomology of this complex is denoted by $H^\bullet (A,A)$ (the Hochschild cohomology).
Define the (normalized) {\it {Hochschild cochain complex}} of $A$ with coefficients in $A$ by
\begin{equation} \label{eq:hocon}
{C}^n(A,A) = Hom (\overline{A}^{\otimes n}, A)
\end{equation}
where $\overline{A} = A / k 1$. It is easy to see that $C^{\bullet} (A,A)$ is a subcomplex and a graded Lie subalgebra of $\tilde{C}^{\bullet} (A,A)$. It is well known that the embedding of $C^{\bullet} (A,A)$ into $\tilde{C}^{\bullet }(A,A)$ is a quasi-isomorphism (\cite{CE}).
Put
\begin{equation} \label{eq:allr}
{\frak g}^{\bullet}_G(A) = C^{\bullet + 1}(A,A)
\end{equation}
The differential $\delta$ and the bracket $[,]_S$ make ${\frak g}^{\bullet}_G(A)$ a differential graded Lie algebra.
\begin{remark} \label{remark:cup}
Formula
\begin{equation} \label{eq:cup}
\gathered
(D \smile E)(a_1, \dots, a_{n+m}) = \\
=(-1)^{nm}D(a_1, \dots, a_n)E(a_{n+1}, \dots, a_{n+m})
\endgathered
\end{equation}
defines an associative product on $C^{\bullet}(A,A)$. This product is compatible with the differential $\delta$, therefore $C^{\bullet}(A,A)$ is a differential graded algebra.
\end{remark}
The following theorem is essentially contained in \cite{HKR}
\begin{thm} \label{thm:quisco}
The formula
$$ D_{\pi}(a_1, \dots, a_n) = <\pi, da_1 \dots da_n>$$
defines a quasi-isomorphism
$$(\Gamma(T, \wedge^{\bullet} T),0) \rightarrow C^{\bullet}(C^{\infty}(M), C^{\infty}(M))$$ Under the isomorphism
$$\Gamma(T, \wedge^{\bullet} T) \rightarrow H^{\bullet}(C^{\infty}(M), C^{\infty}(M))$$
which is induced by this map on cohomology, the bracket induced by $[\;,\;]_G$ becomes the Schouten bracket $[\;,\;]_S$ (\ref{eq:sch}) and the product induced by the cup product becomes the wedge product.
\end{thm}
\subsection{$L_{\infty}$ algebras} \label{ss:lin}
{\it {An $L_{\infty}$ algebra}} is a graded vector space ${\frak g} ^{\bullet}$ together with a coderivation $\partial$ of the free cocommutative coalgebra $S({\frak g} ^{\bullet}[1])$ such that degree of $\partial$ is $1$ and $\partial ^2 = 0$.
Put
\begin{equation} \label{eq:vnesh}
\wedge^{m}({\frak g}^{\bullet})=S^{m}({\frak g}^{\bullet}[1])[-m]
\end{equation}
One has
\begin{equation} \label{eq:vnesh1}
\wedge^{\bullet}({\frak g}^{\bullet})=T({\frak g}^{\bullet})/I
\end{equation}
where $I$ is the two-sided ideal generated by all the elements $$xy-(-1)^{(|x|-1)(|y|-1)}yx$$ where $x, \;y$ are homogeneous elements of ${\frak g}^{\bullet}$.
A coderivation $\partial$ of degree $1$ is uniquely determined by its composition with the projection $S({\frak g}^{\bullet}[1]) \rightarrow {\frak g}^{\bullet}[1]$ which is a sequence of maps
\begin{equation} \label{eq:holie}
[,\dots,]_n : \wedge^n {\frak g}^{\bullet} \rightarrow {\frak g}^{\bullet}
\end{equation}
of degree $2-n$, $n=1,2,\dots$
The condition $\partial ^2 = 0$ is equivalent to the following: for any homogeneous elements $x_{1}, \dots, x_{n}$ of ${\frak g}^{\bullet}$,
\begin{equation} \label{eq:holie1}
\sum \pm [[x_{i_1}, \dots, x_{i_p}]_p, x_{j_1}, \dots, x_{j_q}]_{q+1} = 0
\end{equation}
where the sum is taken over all ${i_1} < \dots < {i_p}$, ${j_1} < \dots < {j_q}$ such that $\{1, \dots, n\}$ is the disjoint union of $\{{i_1}, \dots, {i_p}\}$ and $\{{j_1} , \dots ,{j_q}\}$. The signs $\pm$ are computed by the following rule: whenever a transposition of $x$ and $y$ appears, the result is multiplied by the sign $(-1)^{(|x|-1)(|y|-1)}$.
In particular, $\delta x = [x]_1$ is a differential on ${\frak g} ^{\bullet}$ and the bracket $[\;,\;]_2$ induces a graded Lie algebra structure on the cohomology of the complex $({\frak g} ^{\bullet}, \delta)$.
Any differential graded algebra $({\frak g} ^{\bullet}, [\;,\;], \delta)$ is an $L_{\infty}$ algebra if one puts $[\;]_1 = \delta$, $[\;,\;]_2 = [\;,\;]$ and $[\;,\dots,\;]_n=0$ for $n>2$.
Given two $L_{\infty}$ algebras ${\frak g} ^{\bullet}$ and ${\frak{h}} ^{\bullet}$, {\it {an $L_{\infty}$ morphism}} $f: {\frak g} ^{\bullet} \rightarrow {\frak{h}} ^{\bullet}$ is a morphism of differential graded coalgebras $S({\frak g} ^{\bullet}[1]) \rightarrow S({\frak{h}} ^{\bullet}[1])$. A morphism of graded coalgebras, without assuming that it commutes with differentials, is uniquely determined by its composition with the projection $S({\frak{h}} ^{\bullet}[1]) \rightarrow {\frak{h} }^{\bullet}[1]$, which is a sequence of linear maps $f_n : \wedge ^n {\frak g} ^{\bullet} \rightarrow {\frak{h}} ^{\bullet}$ of degree $1-n$. The condition that these maps define a morphism of differential coalgebras is equivalent to the following: for any homogeneous elements $x_1, \dots, x_n$ of ${\frak g}^{\bullet}$,
\begin{equation} \label{eq:molal}
\gathered
\sum \pm f_{q+1}([x_{i_1}, \dots, x_{i_p}]_p, x_{j_1}, \dots, x_{j_q}) = \\
\sum \pm \frac{1}{k!}[f_{n_1}(x_{i_{11}}, \dots, x_{i_{1n_1}}), \dots,f_{n_k}(x_{i_{k1}}, \dots, x_{i_{kn_k}})]_k
\endgathered
\end{equation}
The signs $\pm$ in (\ref{eq:molal}), and in the sum in the left hand side, are as in (\ref{eq:holie}). The sum in the right hand side is taken over all $k \geq 1$ and all ${i_{r1}}< \dots <{i_{rn_r}}$, $1\leq r \leq k$, such that $\{1,\dots, n\}$ is a disjoint union of $\{{i_{r1}}, \dots ,{i_{rn_r}}\}$, $1\leq r \leq k$.
In particular, $f_1$ is a morphism of complexes. We say that f is an $L_{\infty}$ quasi-isomorphism if $f_1$ is a quasi-isomorphism.
\subsection{Formality theorem} \label{ss:fothm}
Recall that for a manifold $M$ one can define the Schouten-Nijenhujs bracket
\begin{equation} \label{eq:sch}
[\;,\;]_S: \Gamma(M, \wedge^n T) \otimes \Gamma(M, \wedge^m T) \rightarrow \Gamma(M, \wedge^{n+m-1} T)
\end{equation} as the unique bilinear operation satisfying the following conditions:
\begin{enumerate}
\item for $X \in \Gamma(M, T)$, $[X, \pi]_S = L_X \pi$
\item for $f,g \in \Gamma(M, \wedge^0 T)$, $[f,g]_S = 0$
\item the bracket $[\;,\;]_S$ turns $\Gamma(M, \wedge^{\bullet +1} T)$ into a graded Lie algebra
\item for any $\pi, \varphi, \psi \in \Gamma(M, \wedge^{\bullet} T)$,
$$[\pi, \varphi \wedge \psi]_S = [\pi, \varphi]_S \wedge \psi + (-1)^{|\pi|(|\varphi|+1)}\varphi \wedge [\pi, \psi]_S$$
\end{enumerate}
(for $\pi \in \Gamma(M, \wedge^n T)$, we write $|\pi|=n-1$). When $n=m=2$, the above bracket coincides with the polarized bracket from (\ref{eq:bras}).
Denote by ${\frak g}^{\bullet}_S(M)$ the differential graded Lie algebra $\Gamma(M, \wedge^{\bullet +1} T)$ with the bracket $[\;,\;]_S$ and the differential $\delta = 0$.
\begin{thm} \label{thm:fothm}
(Kontsevich, \cite{K}). There exists natural $L_{\infty}$ quasi-isomorphism
$$K: {\frak g}_S^{\bullet}(M) \rightarrow {\frak g}_G^{\bullet}(C^{\infty}(M))$$
The component $K_1$ of $K$ coincides with the quasi-isomorphism from Theorem \ref{thm:quisco}
\end{thm}
{\bf {Proof of Theorem \ref{thm:konts}}.} For any differential graded Lie algebra $({\frak g} ^{\bullet}, [\;,\;], \delta)$, define
\begin{equation} \label{eq:mc}
MC({\frak g} ^{\bullet}) = \{ \pi \in t{\frak g} ^1[[t]]\;\; |\;\; \delta \pi +\frac{1}{2} [\pi, \pi] = 0\}
\end{equation}
The group $\operatorname{exp}(t{\frak g} ^0[[t]])$ acts on $MC({\frak g}^{\bullet})$ by
$$\delta + \operatorname{exp}(X)\pi = \operatorname{exp}(\operatorname{ad} (X))(\delta + \pi)$$
Put
\begin{equation} \label{eq:mod}
M({\frak g} ^{\bullet})=MC({\frak g} ^{\bullet}) / \operatorname{exp}(t{\frak g} ^{0} [[t]])
\end{equation}
For an $L_{\infty}$ homomorphism $f: {\frak g} ^{\bullet} \rightarrow
{\frak{h}} ^{\bullet}$, there is a well-defined map
$$M({\frak g} ^{\bullet}) \rightarrow M({\frak{h}} ^{\bullet})$$
induced by
\begin{equation} \label{eq:mapmod}
\pi \mapsto \sum_{n=1}^{\infty} \frac{1}{n!}f_n(\pi, \dots, \pi)
\end{equation}
If $f$ is an $L_{\infty}$ quasi-isomorphism then the above map is a bijection. Finally, note that $M({\frak g}^{\bullet}_S(M))$ is the set of equivalence classes of formal Poisson structures on $M$ and $M({\frak g} ^{\bullet}_G(A_0))$ is the set of isomorphism classes of deformations of $A_0$ for any algebra $A_0$. Indeed, the equation $\delta \Pi + \frac{1}{2}[\Pi, \Pi] = 0$ is equivalent to $[m+\Pi, m+\Pi]=0$, which is equivalent to the product $a*b = ab + \Pi(a,b)$ being associative.
\begin{remark} \label{rmk:modlinf}
One can easily define spaces $MC({\frak g} ^{\bullet})$ and $M({\frak g} ^{\bullet})$ for any $L_{\infty}$ algebra ${\frak g} ^{\bullet}$. For example, the Maurer-Cartan equation from (\ref{eq:mc}) becomes
$$\sum_{n=1}^{\infty}\frac{1}{n!}[\pi, \dots, \pi]_n = 0$$
\end{remark}
\section{Formality conjectures for Hochschild and cyclic chains} \label{s:foc}
\subsection{Hochschild and cyclic chain complexes} \label{ss:hocy}
For an algebra $A$ over $k$, define for $n \geq 0$
$$C_n(A,A)=A \otimes \overline{A} ^{\otimes n}$$
(recall that $\overline{A} = A / k1$);
define $b:C_n(A,A) \rightarrow C_{n-1}(A,A)$ by
\begin{equation} \label{eq:b}
\gathered
b(a_0 \otimes \dots \otimes a_n) = (-1)^n a_na_0 \otimes a_1 \otimes \dots \otimes a_{n-1} + \\
\sum _{i=0}^{n-1}(-1)^i a_0 \otimes \dots \otimes a_ia_{i+1} \otimes \dots a_n
\endgathered
\end{equation}
One has $b^2 = 0$. The complex $(C_{\bullet}(A,A), b)$ is called the Hochschild chain complex of $A$ with coefficients in the bimodule $A$. The homology of this complex is denoted by $H_{\bullet}(A,A)$, or $HH_{\bullet}(A)$, and is called {\it {the Hochschild homology}} of $A$.
\begin{remark} \label{rmk:tens}
If $A=C^{\infty}(M)$, one has to use one of the following three definitions of tensor powers of $A$:
\begin{enumerate}
\item $A^{\otimes n+1} = C^{\infty}(M^{n+1})$
\item $A^{\otimes n+1} = \text{germs}_{\Delta}C^{\infty}(M^{n+1})$
\item $A^{\otimes n+1} = \text{jets}_{\Delta}C^{\infty}(M^{n+1})$
\end{enumerate}
where $\Delta$ is the diagonal in $M^{n+1}$.
One defines $A\otimes \overline{A}^{\otimes n}$ accordingly. All the definitions above lead to the same answer for the Hochschild cohomology: the map
\begin{equation} \label{eq:mu}
\mu:( C_{\bullet} (C^{\infty}(M), C^{\infty}(M)),b) \rightarrow (\Omega^{\bullet}(M), 0)
\end{equation}
defined by
\begin{equation} \label{eq:mu1}
\mu (a_0 \otimes \dots \otimes a_n)=\frac{1}{n!}a_0 da_1 \dots da_n
\end{equation}
is a quasi-isomorphism of complexes (cf. \cite{C}).
\end{remark}
For $D \in C^d(A,A)$ and $a_0, \dots, a_n \in A$, define
\begin{equation} \label{eq:ld}
\gathered
L_D(a_0, \dots, a_n)=\sum_{i=0}^{n-d}(-1)^{(d-1)(i+1)}a_0 \otimes \dots a_i \otimes D(a_{i+1},\dots, a_{i+d}) \otimes \dots a_n + \\
\sum_{j=n-d}^{n}(-1)^{n(j+1)}D(a_{j+1}, \dots, a_0, \dots) \otimes a_{d+j-n} \dots \otimes a_j
\endgathered
\end{equation}
One has
$$ [L_D, L_E]=L_{[D,E]_G}$$
and
$$b=L_m$$
Thus the operators $L_D$ define an action of the differential graded Lie algebra ${\frak g} _G ^{\bullet}(A)$ on the complex $C_{\bullet}(A,A)$.
Recall that {\it {the cyclic differential}}
\begin{equation} \label{eq:bcap}
B: C_n(A,A) \rightarrow C_{n+1}(A,A)
\end{equation}
is defined by
\begin{equation} \label{eq:bcap1}
B(a_0 \otimes \dots a_n) = \sum_{i=0}^{n}(-1)^{ni}1\otimes a_i \otimes \dots \otimes a_0 \otimes \dots \otimes a_{i-1}
\end{equation}
One has $b^2 = Bb+bB = B^2 = 0$, as well as
$$[B, L_D]=0$$
Following Getzler's notation, put
\begin{equation} \label{eq:ccmin}
CC^-_{\bullet}(A) = (C_{\bullet}(A,A)[[u]],b+uB)
\end{equation}
where $u$ is a formal variable of degree $-2$. One sees that $CC^-_{\bullet}(A)$
is a differential graded module over the differential graded algebra ${\frak g} ^{\bullet}_G (A)$ for any algebra $A$.
Now consider the space $\Omega^{\bullet}(M)$ of differential forms on a manifold $M$. For a multivector field $\pi \in \Gamma (M, \wedge ^d T)$,
put
\begin{equation} \label{eq:lpi}
L_{\pi}=[d,i_{\pi}]
\end{equation}
where $i_{\pi}$ is the contraction
\begin{equation} \label{eq:ipi}
i_{\pi}: \Omega^{\bullet}(M) \rightarrow \Omega^{\bullet - d}(M)
\end{equation}
It is well known that
\begin{equation} \label{eq:lipi}
L_{[\pi,\psi]_S}=[L_{\pi},L_{\psi}]
\end{equation}
Therefore operators $L_{\pi}$ define an action of the algebra ${\frak g}_S^{\bullet}(M)$ on the graded space $\Omega^{\bullet}(M)$, as well as on the complex $(\Omega^{\bullet}(M)[[u]], ud)$.
\begin{thm} \label{thm:mucyc}
(Connes, \cite{C}). The formula
$$\mu(a_0 \otimes \dots \otimes a_n) = \frac{1}{n!}a_0 da_1 \dots d a_n$$
defines a functorial ${\Bbb{C}}[[u]]$-linear quasi-isomorphism
\begin{equation} \label{eq:mucyc}
CC_{\bullet}^-(C^{\infty}(M)) \rightarrow (\Omega^{\bullet}(M)[[u]],ud)
\end{equation}
\end{thm}
Under the homomorphisms
\begin{equation} \label{eq:klipi}
H_{\bullet}(C^{\infty}(M),C^{\infty}(M)) \rightarrow \Omega^{\bullet}(M)
\end{equation}
\begin{equation} \label{eq:piipi}
HC_{\bullet}^-(C^{\infty}(M)) \rightarrow H^{\bullet}(M)[[u]]
\end{equation}
which are induced by $\mu$ on cohomology, the operators induced by $L_D$ become $L_{\pi}$ where $D$ is a Hochschild cocycle and $\pi$ is the image of the cohomology class of $D$ under the isomorphism from Theorem \ref{thm:quisco}.
\subsection{$L_{\infty}$ modules} \label{ss:limo}
{\it {An $L_{\infty}$ module}} over an $L_{\infty}$ algebra ${\frak g}^{\bullet}$ is a graded vector space $M^{\bullet}$ together with a coderivation $\partial_M$ of degree $1$ of the free differential graded comodule $S({\frak g}^{\bullet} [1])\otimes M^{\bullet}$ such that $\partial_M ^2 = 0$. A coderivation $\partial_M$, without the condition $\partial_M ^2 = 0$, is uniquely determined by its composition with the projection of $S({\frak g}^{\bullet} [1])\otimes M^{\bullet}$ onto $M^{\bullet}$. This composition is a sequence of linear maps
\begin{equation} \label{eq:fi}
\phi _n : \wedge^n {\frak g}^{\bullet} \otimes M^{\bullet} \rightarrow M^{\bullet}
\end{equation}
of degree $1-n$. The condition $\partial_M ^2 =0$ is equivalent to the following: for any homogeneous elements $x_1, \dots, x_n$ of ${\frak g}^{\bullet}$ and $m$ of $M^{\bullet}$,
\begin{equation} \label{eq:linfmodu}
\gathered
\sum \pm \phi_{p+1}(x_{i_1}, \dots, x_{i_p},\phi_q( x_{j_1}, \dots, x_{j_q},m))+\\
+\sum \pm \phi_{q+1}([x_{i_1}, \dots, x_{i_p}]_p, x_{j_1}, \dots, x_{j_q},m) = 0
\endgathered
\end{equation}
where the signs are computed and the sum is taken as in (\ref{eq:holie}).
In particular,
$$\delta_M = \phi_0$$ is a differential on $M^{\bullet}$.
The maps $\phi _n$, $n \geq 0$, define an $L_{\infty}$ module structure on $M^{\bullet}$ if and only if the maps $\phi _n$, $n\geq 1$, define an $L_{\infty}$ morphism ${\frak g}^{\bullet} \rightarrow \text{Hom}(M^{\bullet},M^{\bullet})$ where the right hand side is viewed as a differential graded Lie algebra with the differential $[\delta_M, ?]$.
A morphism of $L_{\infty}$ modules over ${\frak g}^{\bullet}, \;$ $\varphi : M^{\bullet} \rightarrow N^{\bullet}$ is by definition a morphism of differential graded comodules $S({\frak g}^{\bullet} [1])\otimes M^{\bullet} \rightarrow S({\frak g}^{\bullet} [1])\otimes N^{\bullet}$. Such a morphism is uniquely determined by maps
$$\varphi _n : \wedge ^n {\frak g}^{\bullet} \otimes M^{\bullet} \rightarrow N^{\bullet}$$ of degree $-n$, $n \geq 0$, satisfying
\begin{equation} \label{eq:linfimohomo}
\gathered
\sum \pm \varphi _{q+1}([x_{i_1}, \dots, x_{i_p}]_p, x_{j_1}, \dots, x_{j_q},m) +\\
+\sum \pm \varphi_{p+1}(x_{i_1}, \dots, x_{i_p},\phi_q( x_{j_1}, \dots, x_{j_q},m))= \\ =
\sum \pm \phi_{p+1}(x_{i_1}, \dots, x_{i_p},\varphi_q( x_{j_1}, \dots, x_{j_q},m))
\endgathered
\end{equation}
\begin{remark} \label{remark:3edef}
An equivalent definition of an $L_{\infty}$ module structure on $M^{\bullet}$ is the following. Let ${\frak g}^{\bullet}$ be an $L_{\infty}$ algebra and $M^{\bullet}$ a graded vector space. On the graded space ${\frak g}^{\bullet} \oplus M^{\bullet}$ consider another grading in which ${\frak g}^{\bullet}$ is of degree zero and $M^{\bullet}$ is of degree one. Consider an $L_{\infty}$ algebra structure on ${\frak g}^{\bullet} \oplus M^{\bullet}$ such that:
\begin{enumerate}
\item ${\frak g}^{\bullet}$ is an $L_{\infty}$ subalgebra of ${\frak g}^{\bullet} \oplus M^{\bullet}$
\item all the operations $[\;,\dots,\;]_n$ are of degree zero with respect to the second grading on ${\frak g}^{\bullet} \oplus M^{\bullet}$
\item $[m_1, m_2, \dots]_n = 0$ for any $m_1, \; m_2 \in M^{\bullet}$.
\end{enumerate}
Similarly, one can define a morphism of $L_{\infty}$ modules as an $L_{\infty}$ morphism $f: {\frak g}^{\bullet} \oplus M^{\bullet} \rightarrow {\frak g}^{\bullet} \oplus N^{\bullet}$ which is of degree zero with respect to the second grading and for which $f_n(m_1, m_2, \dots ) = 0$ for any $m_1, \; m_2 \in M^{\bullet}$.
\end{remark}
\subsection{Formality conjecture for chains} \label{focha}
Let $K$ be the $L_{\infty}$ quasi-isomorphism from Theorem \ref{thm:fothm}. Via $K$, the differential graded Lie algebra ${\frak g}^{\bullet}_S(M)$ acts on $C_{\bullet}(C^{\infty}(M), C^{\infty}(M))$ and $CC_{\bullet}^-(C^{\infty}(M))$ as on $L_{\infty}$ modules.
\begin{conj} \label{conj:fococh}
There exists a natural quasi-isomorphism of $L_{\infty}$ modules
$$\phi: C_{\bullet}(C^{\infty}(M),C^{\infty}(M)) \rightarrow (\Omega^{\bullet}(M),0)$$
such that $\phi _0 $ is the quasi-isomorphism $\mu$ of Connes.
\end{conj}
This conjecture extends to the following
\begin{conj} \label{conj:focycoch}
There exists a natural ${\Bbb{C}}[[u]]$-linear quasi-isomorphism of $L_{\infty}$ modules
$$\phi: CC_{\bullet}^-(C^{\infty}(M)) \rightarrow (\Omega^{\bullet}(M)[[u]],ud)$$ such that $\phi_0$ is the Connes quasi-isomorphism $\mu$.
\end{conj}
\section{Hochschild and cyclic complexes of deformed algebras} \label{hocyde}
Let $\pi$ be a
formal Poisson structure on a manifold $M$. The isomorphism from Theorem \ref{thm:konts} provides a star product $*_{\pi}$ on $C^{\infty}(M)$. Put
\begin{equation} \label{eq:baba}
A(\pi) = (C^{\infty}(M), *_{\pi})
\end{equation}
This is an algebra over $k = {\Bbb{C}}[[t]]$. By $A(\pi)^{\otimes _k (n+1)}$ we will denote $C^{\infty}(M)^{\otimes (n+1)}[[t]]$ (cf. Remark \ref{rmk:tens}); similarly for $A(\pi) \otimes \overline{A(\pi)}^{\otimes _k n}$). If Conjecture \ref{conj:focycoch} is true, then one gets
\begin{cor} \label{cor:ccdef}
There exists a quasi-isomorphism
\begin{equation} \label{eq:cdef}
\mu^{\pi}: C_{\bullet}(A(\pi),A(\pi))\rightarrow (\Omega^{\bullet}(M)[[t]], L_{\pi})
\end{equation}
which extends to a ${\Bbb{C}}[[u]]$-linear quasi-isomorphism
\begin{equation} \label{eq:ccdef}
\mu^{\pi} :CC^-_{\bullet}(A(\pi)) \rightarrow (\Omega^{\bullet}(M)[[u]][[t]], L_{\pi}+ud)
\end{equation}
\end{cor}
\begin{pf} Let $K$ be the quasi-isomorphism from Theorem \ref{thm:fothm}. One checks that the formula
\begin{equation} \label{eq:kapi}
\mu^{\pi}(c) = \sum _{n=0}^{\infty}\frac{1}{n!}K_{n}(\pi, \dots, \pi , c)
\end{equation}
defines a quasi-isomorphism of complexes.
\end{pf}
\begin{remark}
The complexes in the right hand side of formulas (\ref{eq:cdef}, \ref{eq:ccdef}) were studied by Brylinski as quasi-classical approximations to the left hand sides. If $\pi = t \pi _0$ where $\pi _ 0$ is a Poisson structure, then filtration by powers of $t$ defines a spectral sequence with the $E^1$ term equal to the right hand side; this spectral sequence converges to the left hand side (\cite{Bryl}). The above Corollary implies that this spectral sequence degenerates at $E^1$
\end{remark}
\begin{cor} \label{cor:tr}
The space of continuous ${\Bbb {C}}[[t]]$-valued traces on $A(\pi)$ is isomorphic to the space of continuous ${\Bbb {C}}[[t]]$-linear ${\Bbb {C}}[[t]]$-valued functionals on $C^{\infty}(M)$ which annihilate all Poisson brackets $\{f,g\}_{\pi}$\end{cor}
(compare with \cite{CFS}, \cite{Fe}, \cite{NT3} for the symplectic case).
\subsection{The $\widehat{A}$ class of a Poisson manifold} \label{ss:ahat}
Consider a $C^{\infty}$ manifold M with a Poisson structure $\pi _0$. Assuming that Conjecture
\ref{conj:focycoch} is true, define the cohomology class $\widehat{A}(\pi_0)$ in $H^{\text{ev}}(M, {\Bbb{C}}[[t]])$ as folows.
Let $\pi = t \pi _0$. Recall that for any $k$-algebra $A$ the periodic cyclic complex of $A$ is defined by
\begin{equation} \label{eq:ccper}
CC^{per}_{\bullet}(A) = (C_{\bullet}(A,A)[u^{-1}, u]], b+uB)
\end{equation}
Localizing the $L_{\infty}$ quasi-isomorphism $\mu^{\pi}$ from Corollary \ref{cor:ccdef} with respect to $u$, one gets an $L_{\infty}$ quasi-isomorphism
\begin{equation} \label{eq:ccpdef}
\mu^{\pi} :CC^{per}_{\bullet}(A(\pi)) \rightarrow (\Omega^{\bullet}(M)[u^{-1},u]][[t]], L_{\pi}+ud)
\end{equation}
Now note that the complex in the right hand side of (\ref{eq:ccpdef}) is isomorphic to the complex $(\Omega^{\bullet}(M)[u^{-1},u]][[t]],ud)$. Indeed, $L_{\pi} = [d, i_{\pi}]$ and the desired isomorphism is given by $\operatorname{exp}(-u^{-1}ti_{\pi})$. One gets a quasi-isomorphism
\begin{equation} \label{eq:ccpdef0}
CC^{per}_{\bullet}(A(\pi)) \rightarrow (\Omega^{\bullet}(M)[u^{-1},u]][[t]], ud)
\end{equation}
If one views $1$ as an element of $C_0(A(\pi), A(\pi))$, and thus of $CC^{per}_0(A(\pi))$, then the value of the quasi-isomorphism (\ref{eq:ccpdef0}) at $1$ is an element of degree zero in $H^{\bullet}(M, {\Bbb{C}}[[t]])[u^{-1},u]]$, so it can be viewed as an element $\widehat{A}(\pi _0)$ of $H^{\text{ev}}(M, {\Bbb{C}}[[t]])$. Conjecturally, this class does not depend on $t$.
Consider the situation when $\pi _0$ is a regular Poisson structure. In this case, the symplectic leaves of $\pi_0$ form a foliation ${\cal{F}}$. The tangent bundle $T{\cal{F}}$ of this foliation is an $Sp(2n)$-bundle, and one can reduce its structure group to the maximal compact subgroup $U(n)$. Let $\widehat{A}(T{\cal{F}})$ be the $\widehat{A}$ class of the resulting $U(n)$-bundle.
More generally, suppose that $\pi _0$ comes from a symplectic Lie algebroid $({\cal{E}}, \omega) $ (\cite{MK}, \cite{BB}; cf.below for the definitions). This generality was suggested to us by Weinstein.
\begin{conj} \label{conj:ahat}
If $\pi _0$ comes from a symplectic Lie algebroid $({\cal{E}}, \omega)$ then
$$\widehat{A}(\pi_0) = \widehat{A} ({\cal{E}})$$
\end{conj}
Recall that {\it {a Lie algebroid}} is a vector bundle ${\cal{E}}$ whose sections form a sheaf of Lie algebras, together with a morphism of sheaves of Lie algebras ({\it {the anchor map}} )
\begin{equation} \label{eq:anch}
\rho : \Gamma({\cal{E}}) \rightarrow \Gamma(T)
\end{equation}
such that
\begin{equation} \label{eq:ancho}
[\xi, f\eta] = L_{\rho(\xi)}(f)\eta + f[\xi, \eta]
\end{equation}
where $\xi, \; \eta$ are local sections of ${\cal{E}}$ and $f$ is a local function. Recall that for a Lie algebroid ${\cal{E}}$, the de Rham complex is defined:
\begin{equation} \label{eq:dre}
\gathered
^{\cal{E}}\Omega^{\bullet}(M) = \Gamma (M, \wedge ^{\bullet}{\cal{E}}^*)\\
^ {\cal{E}}\Omega^{\bullet}(M) \stackrel{d}{\rightarrow}{^ {\cal{E}}\Omega^{\bullet + 1}(M)} \\
(d\omega)(\xi _1, \dots, \xi _{m+1}) = \sum_{i=1}^{m+1}(-1)^{i-1}\rho(\xi _i)\omega(\xi _1, \dots, \widehat{\xi _i}, \dots, \xi_{m+1}) + \\
\\ +\sum _{i<j}(-1)^{i+j} \omega([\xi_i, \xi_j], \dots, \widehat{\xi _i}, \dots, \widehat{\xi _j}, \dots)
\endgathered
\end{equation}
The algebra of {\it {${\cal{E}}$-differential operators}} $^{\cal{E}} D_M$ is the quotient of the enveloping algebra $U(\Gamma ({\cal{E}})$ by the ideal generated by the elements
$$ \xi f\eta - L_{\rho(\xi)}(f)\eta - f\xi \eta $$
where $\xi, \; \eta$ are local sections of ${\cal{E}}$ and $f$ is a local function.
{\it {A symplectic Lie algebroid}} is a Lie algebroid ${\cal{E}}$ together with a non-degenerate closed ${\cal{E}}$-form $\omega \in { ^ {\cal{E}}\Omega^{\bullet}(M)}$. We denote by $\pi _0 \in \Gamma (M, \wedge ^2{\cal{E}})$ the image of $\omega$ under the isomorphism $\wedge ^2{\cal{E}}^* \rightarrow \wedge ^2{\cal{E}}$ induced by $\omega$. The bivector field $(\wedge ^2 \rho)(\pi _0)$ is a Poisson structure. We will denote this Poisson structure also by $\pi _0$.
Let us outline the reasoning for which the above conjecture should be true. Corollary \ref{cor:ccdef} is true for $\pi = t \pi _0$ where $\pi _0$ is a regular Poisson structure or, more generally, when $\pi _0$ is given by a symplectic Lie algebroid $({\cal{E}}, \omega) $.The cohomology of the above complex will be denoted by $^{\cal{E}}H^{\bullet}(M)$.
When ${\cal{F}}$ is a foliation with a leafwise symplectic form $\omega$ and ${\cal{E}}$ is the tangent bundle of this foliation, then ${^ {\cal{E}}\Omega^{\bullet}(M)}$ is the de Rham complex of leafwise forms. The anchor map extends to a morphism of complexes $\Omega^{\bullet}(M) \rightarrow {^ {\cal{E}}\Omega^{\bullet}(M)}$.
In \cite{NT1}, we studied star products on $C^{\infty}(M)$ for which the corresponding Poisson structure $\pi _0$ is given by a symplectic Lie algebroid $({\cal{E}}, \omega)$ and the operators $P_m$ are ${\cal{E}}$-bidifferential (call them {\it {$ {\cal{E}}$-deformations}}). We regard two ${\cal{E}}$-deformations as equivalent if there is an equivalence of star products $T=\text{Id}+\sum_{m=1}^{\infty}t^mT_m$ for which all $T_m$ are ${\cal{E}}$-differential operators. We showed that Fedosov's methods from \cite{F} are applicable in this situation. In particular, to any ${\cal{E}}$-deformation one can associate a characteristic class
\begin{equation} \label{eq:teta}
\theta \in \frac{1}{t}\omega + {^ {\cal{E}}H^2(M, {\Bbb{C}})}[[t]]
\end{equation}
which defines a bijection between the set of equivalence classes of ${\cal{E}}$-deformations and $\frac{1}{t}\omega + {^{\cal{E}}H^2(M, {\Bbb{C}})}[[t]]$.
Suppose given an ${\cal{E}}$-deformation $A = (C^{\infty}(M), *)$ with the characteristic class $\theta$. In \cite{NT1} and \cite{BNT}, we constructed a ${\Bbb{C}}[[u]]$-linear trace density map
\begin{equation} \label{eq:muh}
\mu^t: CC_{\bullet}^-(A) \rightarrow (^ {\cal{E}}\Omega^{2n-\bullet}(M)((t))[[u]],d)
\end{equation}
whose localization with respect to $u$ provides a ${\Bbb{C}}[u^{-1},u]]$-linear morphism
\begin{equation} \label{eq:muhp}
\mu^t: CC_{\bullet}^{per}(A) \rightarrow (^ {\cal{E}}\Omega^{2n-\bullet}(M)((t))[u^{-1},u]],d)
\end{equation}
We proved
\begin{thm} \label{thm:rrbnt}
$$\mu ^t (1) = \sum_{p\geq 0}\widehat{A}({\cal{E}})_{2p}u^{n-p}$$
\end{thm}
Now assume for simplicity that $\theta = \frac{1}{t}\omega$.
Let $N=(-1)^{n}t^{k-n}$ on $^{\cal{E}}\Omega^{k}(M)$. Let $*: ^ {\cal{E}}\Omega^{\bullet}(M) \rightarrow ^ {\cal{E}}\Omega^{2n-\bullet}(M)$ be the symplectic star operator. Consider the sequence of morphisms of complexes \begin{equation} \label{eq:mnogo}
\gathered
(^ {\cal{E}}\Omega^{2n-\bullet}(M)((t))((u)),d) \stackrel{N}{\longrightarrow}( ^ {\cal{E}}\Omega^{2n-\bullet}(M)((t))((u)), td) \\
\stackrel {\operatorname{exp}(ut^{-1}i_{\pi_0})}{\longrightarrow} (^ {\cal{E}}\Omega^{2n-\bullet}(M)((t))((u)), td + uL_{\pi_0}) \\
\stackrel{*}{\longrightarrow} (^ {\cal{E}}\Omega^{\bullet}(M)((t))((u)), tL_{\pi_0}+ud) \\
\stackrel{\operatorname{exp}(-tu^{-1}i_{\pi_0})}{\longrightarrow} (^{\cal{E}}\Omega^{\bullet}(M)((t))((u)), ud)
\endgathered
\end{equation}
Denote the composition of the above maps by $\nu$.
\begin{conj} \label{conj:whocares}
The quasi-isomorphism (\ref{eq:ccpdef0}), composed with
\begin{equation} \label{eq:omeom}
\Omega^{\bullet}(M)((t))((u)) \rightarrow ^ {\cal{E}}\Omega^{\bullet}(M)((t))((u)),
\end{equation}
is equal to $\nu \mu ^t$.
\end{conj}
Compose $\nu$ with the isomorphism
$$^ {\cal{E}}\Omega^{\bullet}(M)((t))((u)) \rightarrow ^ {\cal{E}}\Omega^{\bullet}(M)((t))((u))$$
which is equal to $u^{n-k}$ on $^{\cal{E}} \Omega ^k$.
\begin{remark}
Note the symmetry between the formal variables $u$ and $t$ in the above calculations.
\end{remark}
Denote the resulting composition by $\nu _0$. We claim that $\nu _0$ is equal to the operator of multiplication by $\operatorname{exp}(-\frac{\omega}{ut})$. Indeed, one checks that $\nu _0 (1)$ is equal to $\operatorname{exp}(-\frac{\omega}{ut})$ (we use the fact that for any $z$
$$\operatorname{exp}(zi_{\pi})\frac{1}{n!}\omega^{n}=z^n\operatorname{exp}(z^{-1}\omega)).$$
But $\nu _0$ is a $C^{\infty}(M)$-linear endomorphism of $^ {\cal{E}}\Omega^{2n-\bullet}(M)((t))((u))$, thus it is the operator of multiplication by $\nu _0 (1)$. From Theorem \ref{thm:rrbnt} we see that
\begin{equation} \label{eq:numu1}
(\nu_0\mu^t)(1)=u^{n}\sum \widehat{A}({\cal{E}})_{2p}u^{-p}
\end{equation}
Therefore
\begin{equation} \label{eq:numu2}
(\nu_0\mu^t)(1)=\sum \widehat{A}({\cal{E}})_{2p}u^{p}
\end{equation}
Combining the above formula with Conjecture \ref{conj:whocares}, one sees that the composition of maps (\ref{eq:omeom}) and (\ref{eq:ccpdef0}) evaluated at $1$ is equal to $\sum \widehat{A}({\cal{E}})_{2p}u^{p}$.
\section{Homotopy Gerstenhaber algebras and modules} \label{s:ga}
In this section, we will outline a possible proof of the conjectures above, as well as their generalizations. This proof will follow the lines of the recent proof of the Kontsevich formality theorem, due to Tamarkin \cite{T}.
\subsection{Definitions} \label{ss:gade}
Recall that a graded space $V^{\bullet}$ is {\it {a Gerstenhaber algebra}} if it is a graded commutative associative algebra, $V^{\bullet}[1]$ is a graded Lie algebra, and the two operations on $V^{\bullet}$ satisfy the Leibnitz identity
\begin{equation} \label{eq:le}
[a,bc]=[a,b]c+(-1)^{(|a|-1)|b|}b[a,c]
\end{equation}
The Hochschild cohomology $H^{\bullet}(A,A)$ of any associative algebra $A$ is a Gerstenhaber algebra on which the product and the bracket are induced by the cup product and the Gerstenhaber bracket respectively ((\ref{eq:cup}), (\ref{eq:lbra})).
Let us recall a definition of a $G_{\infty}$ algebra, or a strong homotopy Gerstenhaber algebra. For a graded vector space $V^{\bullet}$, consider the free Lie coalgebra $\text{coLie}(V^{\bullet}[1])$ and the free cocommutative coalgebra $S(\text{coLie}(V^{\bullet}[1]))$. The latter graded space has a structure of a Lie coalgebra which is dual to the Berezin-Kirillov-Kostant Lie algebra structure (in the dual language, $S(\text{Lie}(V^{\bullet}[1]))$ is a Poisson algebra).
A structure of a $G_{\infty}$ algebra on $V^{\bullet}$ is by definition a map $\partial : S(\text{coLie}(V^{\bullet}[1])) \rightarrow S(\text{coLie}(V^{\bullet}[1]))$ of degree $1$ which is a coderivation with respect to both coalgebra structures and such that $\partial ^2 = 0$. Any Gerstenhaber algebra is a $G_{\infty}$ algebra. For any $G_{\infty}$ algebra $V^{\bullet}$, $V^{\bullet}[1]$ is an $L_{\infty}$ algebra.
For any $G_{\infty}$ algebra $V^{\bullet}$, one can define the cochain complex of coderivations of $S(\text{coLie}(V^{\bullet}[1]))$, with the differential $[\partial, ?]$. The cohomology of this complex is denoted by $H^{\bullet}(V^{\bullet}, V^{\bullet})$. This is a $G_{\infty}$ analogue of Hochschild cohomology.
Define a $G_{\infty}$ morphism $f:V^{\bullet} \rightarrow W^{\bullet}$ to be a morphism of differential graded Poisson coalgebras $S(\text{coLie}(V^{\bullet}[1])) \rightarrow S(\text{coLie}(W^{\bullet}[1]))$. We say that a $G_{\infty}$ algebra $V^{\bullet}$ is formal if there is a $G_{\infty}$ quasi-isomorphism
\begin{equation} \label{eq:forgest}
H^{\bullet}(V^{\bullet}, V^{\bullet}) \rightarrow V^{\bullet}
\end{equation}
A standard argument from homological algebra shows that obstructions to formality of a $G_{\infty}$ algebra $V^{\bullet}$ lie in $H^{\bullet}(V^{\bullet}, V^{\bullet})$.
\subsection{Tamarkin's proof} \label{ss:tap}
In \cite{T}, Tamarkin proves the following
\begin{thm} \label{thm:tam}
For any associative algebra $A$, the Hochschild cochain complex $C^{\bullet}(A,A)$ has a structure of a $G_{\infty}$ algebra whose underlying $L_{\infty}$ algebra is ${\frak g}^{\bullet}_G(A)$
\end{thm}
\begin{thm} \label{thm:tam1}
Let $A={\Bbb{C}}[[x_1, \dots, x_n]]$ or $A=C^{\infty}({\Bbb {R}}^n)$. The obstructions to formality of the $G_{\infty}$ algebra $C^{\bullet}(A,A)$ are equal to zero.
\end{thm}
This shows that the above algebras are formal as $G_{\infty}$ algebras. From this, using an argument with Gelfand-Fuks cohomology as in \cite{K}, one deduces
\begin{thm}
Let $A=C^{\infty}(M)$. Then $C^{\bullet}(A,A)$ is formal as a $G_{\infty}$ algebra.
\end{thm}
\subsection{Generalized formality conjecture for chains}
Conjecture \ref{conj:fococh} can be generalized along the lines of the previous subsection as follows. First, one can define $G_{\infty}$ modules and their homomorphisms as one did in the $L_{\infty}$ case (following any of the above definitions, for example the one from Remark \ref{remark:3edef}).
The problem with this definition is that, for example, $\Omega ^{\bullet}(M)$ is not a Gerstenhaber module over the Gerstenhaber algebra $\Gamma (\wedge ^{\bullet}(T))$. To correct this, for any Gerstenhaber algebra $V^{\bullet}$ define a new Gerstenhaber algebra $V^{\bullet}[\epsilon]$ by
\begin{equation} \label{epsilon}
(a+\epsilon b)(c+\epsilon d)=ac+\epsilon(bc + (-1)^{|a|}ad + (-1)^{|a|}[a,c])
\end{equation}
\begin{equation} \label{epsilonlie}
[a+\epsilon b,\;c+\epsilon d]=[a,c]+\epsilon([b,c] + (-1)^{|a|+1}[a,d ])
\end{equation}
(This is a deformation of $V^{\bullet}$ with an odd parameter along the Poisson bracket; the specifics of the graded case is that the deformed algebra remains commutative. Note that an isomorphism of this deformation to the trivial one is precisely a BV operator). Using Tamarkin's methods, one can prove that for any algebra $A$ there is a $G_{\infty}$ structure on $C_{\bullet}(A,A)[\epsilon]$ which induces the structure (\ref{epsilon}), (\ref{epsilonlie}) on $H_{\bullet}(A,A)[\epsilon]$.
\begin{conj} \label{conj:ger}
For any associative algebra $A$, the Hochschild chain complex $C_{\bullet}(A,A)$ is a $G_{\infty}$ module over the $G_{\infty}$ algebra
$C^{\bullet}(A,A)[\epsilon]$.
The underlying structure of an $L_{\infty}$ module over $C^{\bullet}(A,A)$ on $C_{\bullet}(A,A)$ is given by the action of $C^{\bullet}(A,A)$ by operators $L_D$ (\ref{eq:ld}).
\end{conj}
If the above conjecture is true then, by virtue of Theorem \ref{thm:tam}, both
$C_{\bullet}(C^{\infty}(M),C^{\infty}(M))$ and $\Omega ^{\bullet}(M)$ are $G_{\infty}$ modules over
$\Gamma(M, \wedge ^{\bullet}T)[\epsilon]$.
\begin{conj} \label{conj:fogerts}
There is a quasi-isomorphism of $G_{\infty}$ modules
$$C_{\bullet}(C^{\infty}(M),C^{\infty}(M)) \rightarrow \Omega ^{\bullet}(M)$$
\end{conj}
To generalize Conjecture \ref{conj:focycoch}, first note that the operator $\frac{\partial}{\partial \epsilon}$ is a BV operator on the algebra $V^{\bullet}[\epsilon]$ (\ref{epsilon}), (\ref{epsilonlie}). Conjecturally, in an appropriate sense, $C_{\bullet}^-(A, A)[\epsilon]$ is a homotopy BV algebra and $CC_{\bullet}^-(A)[\epsilon]$ is a homotopy BV module over it.
Let us finish with a partial case of the statement before Conjecture \ref{conj:ger} which can be obtained by explicit computation.
\begin{thm} \label{thm:tsydal}
\cite{DT}. On the Hochschild chain complex $C_{\bullet}(A,A)$, there is a structure of an $L_{\infty}$ module over
${\frak g}_G^{\bullet}(A)[\epsilon]$ whose restriction to ${\frak g}_G^{\bullet}(A)$ is given by the operators $L_D$.
\end{thm}
|
\section{Introduction}
In astronomy, the collection of data is often limited by the presence
of background noise. Various methods are used to filter the noise
while retaining as much ``useful'' information as possible. In recent
years, wavelets have played an increasing role in astrophysical data
analysis. It provides for a general parameter-free procedure to look
for objects of varying size scales. In the case of the Cosmic
Microwave Background (CMB) one is interested in the non-Gaussian
component in the presence of Gaussian noise and signal. An
application of wavelets is presented by Tenorio et al (1999). This
paper generalizes their analysis beyond the thresholding
approximation. X-ray images are also frequently noise dominated,
caused by instrumental and cosmic background. Successful wavelet
reconstructions were achieved by Damiani et al (1997a,b).
At times generic tests for non-Gaussianity are desired. Inflationary
theories predict, for example, that the intrinsic fluctuations in the
CMB are Gaussian, while topological defect theories predict
non-Gaussianity. A full test for non-Gaussianity requires measuring
all N-point distributions, which is computationally not tractable for
realistic CMB maps. Hobson et al (1998) have shown that wavelets are
a more sensitive discriminant between cosmic string and inflationary
theories if one examines only the one point distribution function of
basis coefficients.
For Gaussian random processes, Fourier modes are statistically
independent. Current theories of structure formation start from
an initially linear Gaussian random field which grows non-linear
through gravitational instability. Non-linearity occurs through
processes local in real space. Wavelets provide a natural basis which
compromise between locality in real and Fourier space. Pando \& Fang
(1996) have applied the wavelet decomposition in this spirit to the
high redshift $L_\alpha$ systems which are in the transition from
linear to non-linear regimes, and are thus well analyzed by the wavelet
decomposition.
We will concentrate in this paper on the specific case of data layed
out on a two dimensional grid, where each grid point is called a {\it
pixel}. Such images are typically obtained through various imaging
instruments, including CCD arrays on optical telescopes,
photomultiplier arrays on X-ray telescopes, differential radiometry
measurements using bolometers in the radio band, etc. In many
instances, the images are dominated by noise. In the optical, the sky
noise from atmospheric scatter, zodiacal light, and extragalactic
backgrounds, sets a constant flux background to any observation. CCD
detectors essentially count photons, and are limited by the Poissonian
discreteness of their arrival. A deep exposure is dominated by sky
background, which is subtracted from the image to obtain the features
and objects of interest. Since the intensity of the sky noise is
constant, it has a Poissonian error with standard deviation $e\propto
n^{1/2}$, where $n$ is the photon count per pixel. After subtracting
the sky average, this fluctuating component remains as white noise in
the image. For large modern telescopes, images are exposed to near
the CCD saturation limit, with typical values of $n\sim 10^4$. The
Poisson noise is well described by Gaussian statistics in this limit.
We would like to pose the problem of filtering out as much of the
noise as possible, while maximally retaining the data. In certain
instances, optimal methods are possible. If we know the data to
consist of astronomical point objects, which have a shape on the grid
given by the atmospheric spreading or telescope optics, we can test
the likelihood at each pixel that a point source was centered there.
The iterative application of this procedure is implemented in the
routine {\it clean} of the Astronomical Image Processing Software
(AIPS) (Cornwell \& Braun 1989).
If the sources are not point-like, or the atmospheric point spread
function varies significantly across the field, {\it clean} is no
longer optimal. In this paper we examine an approach to implement a
generic noise filter using a wavelet basis. In section
\ref{sec:classical} we first review two popular filtering techniques,
thresholding and Wiener. In section \ref{sec:opt} we generalize
Wiener filtering to inherit the advantages of thresholding. A
Bayesian approach to image reconstruction (Vidakovic 1998) is used,
where we use the data itself to estimate the prior distribution of
wavelet coefficients. We recover Wiener filtering for Gaussian data.
Some concrete examples are shown in section \ref{sec:example}.
\section{Classical Filters}
\label{sec:classical}
\subsection{Thresholding}
A common approach to supressing noise is known as thresholding. If
the amplitude of the noise is known, one picks a specific threshold
value, for example $\tau=3\sigma_{\rm noise}$ to set a cutoff at three
times the standard deviation of the noise. All pixels less than this
threshold are set to zero. This approach is useful if we wish to
minimize false detections, and if all sources of signal occupy only a
single pixel. It is clearly not optimal for extended sources, but
often used due to its simplicity. The basic shortcoming is its
neglect of correlated signals which covers many pixels. The choice of
threshold also needs to be determined heuristically. We will attempt
to quantify this procedure.
\subsection{Wiener Filtering}
\begin{figure}
\resizebox{\textwidth}{!}{
\includegraphics{pkwien.eps}
}
\caption{The power spectrum of figure \protect{\ref{plate:org}}. The
dashed line is the power spectrum measured from the noisy data. The
horizontal line is the noise. The lower solid line is the difference
between the measured spectrum and the noise. We see that the
measurement of the difference becomes noise limited at large $k$.
}
\label{fig:pkwien}
\end{figure}
In the specific case that both the signal and the noise are Gaussian
random fields, an optimal filter can be constructed which minimizes
the impact of the noise. If the noise and signal are stationary
Gaussian processes, Fourier space is the optimal basis where all modes
are uncorrelated. In other geometries, one needs to expand in
signal-to-noise eigenmodes (see e.g. Vogeley and Szalay 1996). One
needs to know both the power spectrum of the data, and the power
spectrum of the noise. We use the least square norm as a measure of
goodness of reconstruction. Let $E$ be the reconstructed image, $U$
the original image and $N$ the noise. The noisy image is called
$D=U+N$. We want to minimize the error $e=\langle (E-U)^2\rangle$.
For a linear process, $E=\alpha D$. For our stationary Gaussian
random field, different Fourier modes are independent, and the optimal
solution is $\alpha=\langle U^2\rangle/\langle D^2\rangle$. $U^2$ is
the intrinsic power spectrum. Usually, $D^2$ can be estimated from
the data, and if the noise power spectrum is known, the difference can
be estimated subject to measurement scatter as shown in figure
\ref{fig:pkwien}. Often, the powerspectrum decays with increasing wave
number (decreasing length scale): $\langle U^2\rangle =c k^{-n}$. For
white noise with unit variance, we then obtain $\alpha= c/(k^n+c)$,
which tends to one for small $k$ and zero for large $k$. We really
only need to know the parameters $c,n$ in the crossover region
$E^2\sim 1$. In section \ref{sec:example} we will illustrate a worked
example.
Wiener filtering is very different from thresholding, since modes are
scaled by a constant factor independent of the actual amplitude of the
mode. If a particular mode is an outlier far above the noise,
the algorithm would still force it to be scaled back. This can
clearly be disadvantageous for highly non-Gaussian distributions. If
the data is localized in space, but sparse, the Fourier modes dilute
the signal into the noise, thus reducing signal significantly as is
seen in the examples in section \ref{sec:example}. Furthermore,
choosing $\alpha$ independent of $D$ is only optimal for Gaussian
distributions. One can generalize as follows:
\section{Non-Gaussian Filtering}
\label{sec:opt}
We can extend Wiener filtering to Non-Gaussian Probability Density
Functions (PDFs) if the PDF is
known and the modes are still statistically independent. We will
denote the PDF for a given mode as $\Theta(u)$ which
describes a random variable $U$. When Gaussian white noise with unit variance
${\cal N}(0,1)$ is added, we obtain a new random variable $D=U+{\cal N}(0,1)$
with PDF
$f(d)=(2\pi)^{-1/2}\int \Theta(u) \exp(-(u-d)^2/2) du$. We can
calculate the conditional probability $P(U|D)=P(D|U)P(U)/P(D)$ using
Bayes' theorem. For the posterior conditional expectation value we obtain
\begin{eqnarray}
\langle U|D=d \rangle &=& \frac{1}{\sqrt{2\pi}f(d)}
\int \exp[-(u-d)^2/2] \Theta(u) u du \nonumber\\
&=& D + \frac{1}{\sqrt{2\pi}f(d)}
\partial_d \int \exp[-(u-d)^2/2] \Theta(u) du \nonumber \\
&=& D+ (\ln f)'(d).
\label{eqn:bayes}
\end{eqnarray}
Similarly, we can calculate the posterior variance
\begin{equation}
\left\langle (U-\bar{U})^2 | D=d \right\rangle = 1+(\ln f)''(d).
\label{eqn:var}
\end{equation}
For a Gaussian prior with variance $\sigma$, equation
(\ref{eqn:bayes}) reduces to Wiener filtering. We have a generalized
form for $\alpha=1+(\ln f)'/D$. For distributions with
long tails, $(\ln f)' \sim 0$, $\alpha \sim 1$, and we leave the
outliers alone, just as thresholding would
suggest.
For real data, we have two challenges:
1. estimating the prior distribution $\Theta$.
2. finding a basis in which $\Theta$ is most non-Gaussian.
\subsection{Estimating Prior $\Theta$}
The general non-Gaussian PDF on a grid is a function of $N$ variables,
where $N$ is the number of pixels. It is generally not possible to
obtain a complete description of this large dimensional space
(D. Field, this proceedings). It is often possible, however, to make
simplifying assumptions. We consider two descriptions: Fourier space
and wavelet space. We will assume that the one point distributions of
modes are non-Gaussian, but that they are still statistically
independent. In that case, one only needs to specify the PDF for each
mode. In a hierarchical basis, where different basis functions sample
characteristic length scales, we further assume a scaling form of the
prior PDF $\Theta_l(u)=l^{-\beta}\Theta(u/l^\beta)$. Here $l\sim 1/k$
is the characteristic length scale, for example the inverse wave
number in the case of Fourier modes.
For images, we often have $\beta
\sim 1$.
Wavelets still have a characteristic scale, and we can similarly
assume scaling of the PDF. In analogy with Wiener filtering, we first
determine the scale dependence. For computational simplicity, we use
Cartesian product wavelets (Meyer 1992). Each basis function has two
scales, call them $2^i,2^j$. The real space support of each wavelet
has area $A\propto 2^{i+j}$, and we find empirically that the variance
depends strongly on that area. The scaling relation does not directly
apply for $i\ne j$, and we introduce a lowest order correction using
$\ln(\sigma) = c_1 (i+j)+c_2(i-j)^2+c_3$. We then determine the best
fit parameters $c_i$ from the data. The actual PDF may depend on the
length scale $i+j$ and the elongation $i-j$ of the wavelet basis. One
could parameterize the PDF, and solve for this dependence (Vidakovic
1998), or bin all scales together to measure a non-parametric scale
averaged PDF. We will pursue the latter.
The observed variance is the
intrinsic variance of $\Theta$ plus the noise variance of ${\cal N}$, so
the variance $\sigma_{\rm
intrinsic}^2=\sigma_{\rm obs}^2-\sigma_{\rm noise}^2$ has error
$\propto \sigma_{\rm obs^2}/n$ where $n$ is the number of coefficients
at the same length scale. We weigh the data accordingly.
Because most wavelet modes are at short scales, most of the weight
will come near the noise threshold, which is what we desire. We now
proceed to estimate $f(d)$. Our Ansatz now assumes
$\Theta_{ij}(u)\propto\Theta(u/\exp[c_1(i+j)+c_2(i-j)^2+c_3])$ where
$\Theta(u)$ has unit variance. We can only directly measure $f_{ij}$.
We sort these in descending order of variance, $f_m$. Again,
typically the largest scale modes will have the largest variance. In
the images explored here, we find typical values of $c_1$ between 1
and 2, while $c_2 \sim -0.2$.
For the largest variance modes, noise is least important. From the
data, we directly estimate a binned PDF for the largest scale modes.
By hypothesis, $D=U/l^\beta+{\cal N}(0,\sigma_{l,\rm
noise})$.
We reduce the larger scale PDF
by convolving it with the difference of noise levels to
obtain an initial guess for the smaller scale PDF:
\begin{equation}
f'_{l'}(d) = \frac{(l'/l)^\beta}{\sqrt{\pi}}
\int_{-\infty}^\infty f_l\left[u (l/l')^\beta\right]
\exp\left[-\frac{(u-d)^2}{2(\sigma_{l',\rm noise}^2-\sigma_{l,\rm noise}^2) }
\right] du.
\end{equation}
To this we add the actual histogram of wavelets coefficients at the
smaller scale.
We continue this hierarchy to obtain an increasingly better estimate
of the PDF, having used the information from each scale.
In figure \ref{fig:alph} we show the optimal weighting function
obtained for the examples in section \ref{sec:example}.
\begin{figure}
\resizebox{\textwidth}{!}{
\includegraphics{palpha.ps}
}
\caption{The optimal filter function $\alpha$ for the non-Gaussian
wavelet model at $\sigma_{\rm noise}=\sigma_{\rm data}$. $U$ is given
in units of the total standard deviation $\sigma_{\rm
abs}^2=\sigma_{\rm noise}+\sigma_{\rm data}$.
At small
amplitudes, it is similar to a Wiener filter $\alpha=1/2$, but limits
to 1 for large outliers.}
\label{fig:alph}
\end{figure}
On the largest scales, the PDF will be poorly defined because
relatively few wavelets lie in that regime. The current
implementation performs no filtering, i.e. sets $\alpha=1$ for the
largest scales. A potential improvement could be implemented: Within
the scaling hypothesis, we can deconvolve the noisy $f(D)$ obtained
from small scales to estimate the PDF on large scales. The errors in
the PDF estimation are themselves Poissonian, and in the limit that we
have many points per PDF bin, we can treat those as Gaussian. The
deconvolution can then be optimally filtered to maximize the use of
the large number of small scale wavelets to infer the PDF of large
scale wavelets. Of course, the non-Gaussian wavelet analysis could
then be recursively applied to estimate the PDF. Instead of the
Bayesian prior PDF, we would then specify a prior for the prior. This
possibility will be explored in future work.
\subsection{Maximizing Non-Gaussianity using Wavelets}
\label{subsec:nongauss}
Errors are smallest if a large number of coefficients are near zero,
and when the modes are close to being statistically independent. Let
us consider several extreme cases and their optimal strategies.
Imagine that we have an image consisting of true uncorrelated point
sources, and each point source only occupies one pixel. Further
assume that only a very small fraction $\epsilon$ of possible pixels
are occupied, but when a point source is present, it has a constant
luminosity $L$. And then add a uniform white noise background with
unit variance. In Fourier space, each mode has unit variance from the
noise, and variance $L^2 \epsilon$ from the point sources. We easily
see that it will be impossible to distinguish signal from noise if
$L^2 \epsilon < 1$. In real space, white noise is also uncorrelated,
so we are justified to treat each pixel separately. Now we can easily
distinguish signal from noise if $L > \sqrt{-\ln(\epsilon)}$. If
$L=10$ and $\epsilon=0.001$, we have a situation where the signal is
easy to detect in real space and difficult in Fourier space, and in
fact the optimal filter (\ref{eqn:bayes}) is optimal in real space
where the points are statistically independent. In Fourier space,
even though the covariance between modes is zero, modes are not
independent.
Now consider the more realistic case that objects occupy more than one
pixel, but are still localized in space, and only have a small
covering fraction. This is the case of intermittent information. The
optimal basis will depend on the actual shape of the objects, but it
is clear that we want basis functions which are localized. Wavelets
are a very general basis to achieve this, which sample objects of any
size scale, and are able to effectively excise large empty regions.
We expect PDFs to be more strongly non-Gaussian in wavelet space than
either real or Fourier space.
In this formulation, we obtain not only a filtered image, but also an
estimate of the residual noise, and a noise map. For each wavelet
coefficient we find its posterior variance using (\ref{eqn:var}). The
inverse wavelet transform then constructs a noise variance map on the
image grid.
\section{Examples}
\label{sec:example}
In order to be able to compare the performance of the filtering
algorithm, we use as example an image to which the noise is added by
hand. The de-noised result can then be compared to the 'truth'.
We have taken a random image from the Hubble Space telescope,
in this case the 100,000th image (PI: C. Steidel). The original
picture is shown in figure \ref{plate:org}. The gray scale is from 0
to 255. At the top are two bright stars with the telescope support
structure diffraction spikes clearly showing. The extended objects
are galaxies. We then add noise with variance 128, which is shown in
figure \ref{plate:noise}. The mean signal to noise ratio of the image
is 1/4. We can tell by eye that a small number of regions still
protrude from the noise.
The power spectrum of the noisy image is shown in figure
\ref{fig:pkwien}. We use the known expectation value of the noise
variance. The subtraction of the noise can be performed even when the
noise substantially dominates over the signal, as can be seen in the
image. In most astronomical applications, noise is instrumentally
induced and the distribution of the noise is very well documented.
Blank field exposures, for example, often provide an empirical
measurement.
\begin{figure}
\resizebox{\textwidth}{!}{
\includegraphics{100KObs.ps}
}
\caption{The original image, taken from the Space Telescope web page
{\tt www.stsci.edu}. It is the 100,000th image taken with HST for
C. Steidel of Caltech.}
\label{plate:org}
\end{figure}
\begin{figure}
\resizebox{\textwidth}{!}{
\includegraphics{noisy.ps}
}
\caption{Figure \protect\ref{plate:org} with substantial noise added.}
\label{plate:noise}
\end{figure}
\begin{figure}
\resizebox{\textwidth}{!}{
\includegraphics{out.100kwien.ps}
}
\caption{Wiener filtered. We see that the linear scaling function
substantially reduces the flux in the bright stars.}
\label{plate:wiener}
\end{figure}
We first apply a Wiener filter, with the result shown in figure
\ref{plate:wiener}. We notice immediately the key feature: all
amplitudes are scaled by the noise, so the bright stars have been down
scaled significantly. The noise on the star was less than unity, but
each Fourier mode gets contributions from the star as well as the
global noise of the image. The situation worsens if the filling
factor of the signal regions is small. The mean intensity of the
image is stored in the $k=0$ mode, which is not significantly affected
by noise. While total flux is approximately conserved, the flux on
each of the objects is non-locally scattered over the whole image by
the Wiener filter process.
\begin{figure}
\resizebox{\textwidth}{!}{
\includegraphics{out.100kd12.ps}
}
\caption{Non-Gaussian wavelet filtered.
Several of the features that had been lost in the Wiener filtering
process are recovered here.}
\label{plate:dwt}
\end{figure}
The optimal Bayesian wavelet filter is shown in figure \ref{plate:dwt}.
A Daubechies 12 wavelet was used, and the prior PDF reconstructed
using the scaling assumption described in section \ref{sec:opt}.
We see immediately that the amplitudes on the bright objects are much
more accurate. We see also that the faint vertical edge-on spiral on
the lower right just above the bright elliptical is clearly visible in
this image, while it had almost disappeared in the Wiener filter.
\begin{figure}
\resizebox{\textwidth}{!}{
\includegraphics{error.100kd12.ps}
}
\caption{Error map. Plotted is the posterior Bayesian variance.
We see that some features, for example the small
dot in the upper left, have large errors associated with them, and are
therefore artefacts.}
\label{plate:err}
\end{figure}
The Bayesian approach allows us to estimate the error in the
reconstruction using Equation (\ref{eqn:var}). We show the result in
figure \ref{plate:err}. We can immediately see that some features in
the reconstructed map, for example the second faint dot above the
bright star on the upper left, have large errors associated with them,
and are indeed artefacts of reconstruction. Additionally, certain
wavelets experience large random errors. These appear as checkered
'wavelet' patterns on both the reconstructed image and the error map.
\section{Discussion}
Fourier space has the advantage that for translation invariant
processes, different modes are pairwise uncorrelated. If modes were
truly independent, the optimal filter for each $k$ mode would also be
globally optimal. As we have seen from the example in section
\ref{sec:opt}\ref{subsec:nongauss}, processes which are local in real
space are not optimally processed in Fourier space, since different
Fourier modes are not independent. Wavelet modes are not independent,
either. For typical data, the correlations are relatively sparse. In
the astronomical images under consideration, the stars and galaxies
are relatively uncorrelated with each other. Wavelets with compact
support sample a limited region in space, and wavelets which do not
overlap on the same objects on the grid will be close to independent.
Even for Gaussian random fields, wavelets are close to optimal since
they are relatively local in Fourier space. Their overlap in Fourier
space leads to residual correlations which are neglected. We see that
wavelets are typically close to optimal, even though they are never
truly optimal. But in the absence of a full prior, they allow us to
work with generic data sets and usually outperform Wiener filtering.
In our analysis, we have used Cartesian product Daubechies wavelets.
These are preferentially aligned along the grid axes. In the wavelet
filtered map (figure \ref{plate:dwt}) we see residuals aligned with the
coordinate axes. Recent work by Kingsbury (this proceedings) using
complex wavelets would probably alleviate this problem. The complex
wavelets have a factor of two redundancy, which is used in part to
sample spatial translations and rotational directions more
homogeneously and isotropically.
\section{Conclusions}
We have presented a generalized noise filtering algorithm. Using the
Ansatz that the PDF of mode or pixel coefficients is scale invariant,
we can use the observed data set to estimate the PDF. By application
of Bayes' theorem, we reconstruct the filter map and noise map. The
noise map gives us an estimate of the error, which tells us the
performance of the particular basis used and the confidence level of
each reconstructed feature. Based on comparison with controlled data,
we find that the error estimates typically overestimate the true error
by about a factor of two.
We argued that wavelet bases are advantageous for data with a small
duty cycle that is localized in real space. This covers a large class
of astronomical images, and images where the salient information is
intermittently present.
\begin{acknowledgements}
I would like to thank Iain Johnstone, David Donoho and Robert
Crittenden for helpful discussions. I am most grateful to the Bernard
Silvermann and the Royal Society for organizing this discussion meeting.
\end{acknowledgements}
|
\section*{Acknowledgements}
I thank G. Steigman for helpful comments.
|
\section{#1}}
\renewcommand{\theequation}{\thesection.\arabic{equation}}
\newcommand{\subsect}[1]{\setcounter{equation}{0}\subsection{#1}}
\renewcommand{\theequation}{\thesection.\arabic{equation}}
\def{\bf 1}{{\bf 1}}
\newcommand{\begin{equation}}{\begin{equation}}
\newcommand{\end{equation}}{\end{equation}}
\newcommand{\begin{array}}{\begin{array}}
\newcommand{\end{array}}{\end{array}}
\newcommand{\begin{eqnarray}}{\begin{eqnarray}}
\newcommand{\end{eqnarray}}{\end{eqnarray}}
\newcommand{\begin{eqnarray*}}{\begin{eqnarray*}}
\newcommand{\end{eqnarray*}}{\end{eqnarray*}}
\newtheorem{prop}{Proposition}
\newtheorem{lemma}{Lemma}
\newtheorem{theorem}{Theorem}
\newtheorem{corollary}{Corollary}
\title{The Geometry of the Quantum Euclidean Space}
\author{Gaetano Fiore$^{1,2}$, \ John Madore$^{3,4}$
\and
$\strut^1$Dip. di Matematica e Applicazioni, Fac. di Ingegneria\\
Universit\`a di Napoli, V. Claudio 21, 80125 Napoli
\and
$\strut^2$I.N.F.N., Sezione di Napoli,\\
Mostra d'Oltremare, Pad. 19, 80125 Napoli
\and
$\strut^3$Max-Planck-Institut f\"ur Physik
(Werner-Heisenberg-Institut)\\
F\"ohringer Ring 6, D-80805 M\"unchen
\and
$\strut^4$Laboratoire de Physique Th\'eorique et Hautes Energies\\
Universit\'e de Paris-Sud, B\^atiment 211, F-91405 Orsay
}
\date{}
\maketitle
\abstract{A detailed study is made of the noncommutative geometry of
$\b{R}^3_q$, the quantum space covariant under the quantum group
$SO_q(3)$. For each of its two $SO_q(3)$-covariant differential calculi
we find its metric, the corresponding frame and two
torsion-free covariant derivatives that are metric compatible up to a
conformal factor and which yield both a vanishing linear curvature. A
discussion is given of various ways of imposing reality conditions. The
delicate issue of the commutative limit is discussed
at the formal algebraic level.
Two rather different ways of taking the limit are suggested, yielding
respectively $S^2\times \b{R}$ and $\b{R}^3$ as the limit Riemannian
manifold.}
\vfill
\noindent
Preprint LMU-TPW 97-23
\noindent
Preprint 98-63, Dip. Matematica e Applicazioni, Universit\`a di Napoli
\medskip
\eject
\parskip 4pt plus2pt minus2pt
\sect{Introduction and motivation}
Already in 1947 Snyder, in an attempt to remove the divergences from
electrodynamics, suggested~\cite{Sny47} the possibility that the
micro-structure of space-time at the Planck level might be better
described using a noncommutative geometry. A few years later, in 1954,
Pauli suggested that the gravitational field might be considered as a
universal regularizer for all quantum-field divergences. An obvious way
to reconciliate these two points of view is to try to define a
gravitational field within the context of noncommutative geometry and
see if any connection between the commutation relations and gravity can
be found. It is our conjecture that in some yet-to-be-found sense the
gravitational field is a classical manifestation of the noncommutative
micro-structure of space-time. Following the usage in ordinary geometry
we shall identify the gravitational field as a metric-compatible,
torsion-free linear connection. We shall use the expression
`connection' and `covariant derivative' synonymously.
In Section~2 we give a brief overview of a particular version of
noncommutative geometry which can be considered as a noncommutative
extension of the moving-frame formalism if E. Cartan. More details can
be found elsewhere~\cite{Mad95}. There is not as yet a completely
satisfactory definition of either a linear connection or a metric within
the context of noncommutative geometry but there are definitions which
seem to work in certain cases. In the present article we add to the
list of examples which seem to lend weight to one particular
definition~\cite{DubMadMasMou96}. We refer to a recent review
article~\cite{Mad97} for a list of some other examples and references to
alternative definitions. More details of one alternative version can be
found in the book by Landi~\cite{Lan97}; for a general introduction to
more mathematical aspects of the subject we refer to the book by
Connes~\cite{Con94}.
The example we consider here is the 3-dim ``quantum Euclidean space''
~\cite{frt}, which has been previously studied by one of the authors.
This is the quantum space covariant under the quantum group
$SO_q(3)$. In Section~3
we give a review of this space. In Section~4 we recall the construction
of the $SO_q(3)$-covariant differential calculi~\cite{cawa} over them,
more specifically the two $SO_q(3)$-covariant calculi determined in the
$\hat R$-matrix formalism \cite{WesZum90}
and having the de Rham calculus as the commutative limit.
More details are to be found
in the thesis of one of the authors~\cite{fiothesis}. We shall find that
in order to introduce a `frame' as defined in Section~2 a new generator
for dilatation is needed, as in the construction of inhomogeneous
quantum groups from homogeneous ones, as well as the inverse of some
generator of the quantum Euclidean space. In Section~5 we define a
metric and two torsion-free linear connection on the quantum Euclidean
space, yielding vanishing linear curvatures. In Section~6 we consider
the commutative limit $q\to 1$ in order to determine the Riemannian
manifold which remains as a `shadow' of the noncommutative structure.
We suggest two rather different prescriptions for taking the limit: the
limit manifold is $S^2\times \b{R}$ according to the first (and simpler)
prescription, $\b{R}^3$ according to the second (and more sophisticated)
one. The initial generators $x^i$ of the quantum space in the former
tend to their natural cartesian limits, in the latter are to be
`renormalized' before taking the limit. For $q\neq 1$ $x^i$ appear as a
non-commutative analog of general (i.e. non-cartesian) coordinates. In
Section~7 we discuss the various ways one can construct a real
differential calculus from a combination of the two canonical ones. In
the last section we present our conclusions and we compare our results
with alternative definitions of linear connections. We compare also our
results with those found in the case of the Manin plane~\cite{DimMad96}
as well as similar results found~\cite{CerHinMadWes98} in the case of
the quantum Euclidean plane of dimension one.
\sect{Linear connections in the frame formalism} \label{preli}
The starting point is a noncommutative algebra $\c{A}$ which has as
commutative limit the algebra of functions on some manifold and over
$\c{A}$ a differential calculus~\cite{Con94} $\Omega^*(\c{A})$ which
has as corresponding limit the ordinary de~Rham differential
calculus. We recall that a differential calculus is completely
determined by the left and right module structure of the
$\c{A}$-module of 1-forms $\Omega^1(\c{A})$. We shall restrict our
attention to the case where this module is free of rank $n$ as a left
or right module and possesses a special basis $\theta^a$,
$1\leq a \leq n$, which commutes with the elements $f$ of the algebra:
\begin{equation}
[f, \theta^a] = 0. \label{fund}
\end{equation}
In particular the limit manifold must be parallelizable. We shall refer
to the $\theta^a$ as a frame or Stehbein. The integer $n$ plays the
role of `dimension'; it can be greater than the dimension of the limit
manifold but in this case the frame will have a singular limit. We
suppose further~\cite{DimMad96} that the basis is dual to a set of inner
derivations $e_a = \mbox{ad}\, \lambda_a$. This means that the differential is
given by the expression
\begin{equation}
df = e_a f \theta^a = [\lambda_a, f] \theta^a. \label{defdiff}
\end{equation}
One can rewrite this equation as
\begin{equation}
df = -[\theta,f], \label{extra}
\end{equation}
if one introduces~\cite{Con94} the `Dirac operator'
\begin{equation}
\theta = - \lambda_a \theta^a. \label{dirac}
\end{equation}
There is a bimodule map $\pi$ of the space
$\Omega^1(\c{A}) \otimes_\c{A} \Omega^1(\c{A})$ onto the space
$\Omega^2(\c{A})$ of 2-forms and we can write
\begin{equation}
\theta^a \theta^b = P^{ab}{}_{cd} \theta^c \otimes \theta^d \label{proj}
\end{equation}
where, because of (\ref{fund}), the $P^{ab}{}_{cd}$ belong to the center
$\c{Z}(\c{A})$ of $\c{A}$. We shall suppose that the center is
trivial, $\c{Z}(\c{A}) = \b{C}$, and therefore the
components $P^{ab}{}_{cd}$ are real numbers. Define the Maurer-Cartan
elements $C^a{}_{bc} \in \c{A}$ by the equation
\begin{equation}
d\theta^a = - {1\over 2} C^a{}_{bc} \theta^b \theta^c. \label{rot-coef}
\end{equation}
Because of~(\ref{proj}) we can suppose that
$C^a{}_{bc} P^{bc}{}_{de} = C^a{}_{de}$. It follows from the equation
$d(\theta^a f - f \theta^a) = 0$ that there exist elements
$F^a{}_{bc}$ of the center such that
\begin{equation}
C^a{}_{bc} = F^a{}_{bc} - 2 \lambda_e P^{(ae)}{}_{bc} \label{consis1}
\end{equation}
where $(ab)$ indicates symmetrization of the indices $a$ and $b$.
If on the other hand we define $K_{ab}$ by the equation
\begin{equation}
d\theta + \theta^2 = {1\over 2} K_{ab} \theta^a \theta^b, \label{consis2}
\end{equation}
then if follows from (2.3) and the identity $d^2 = 0$ that the $K_{ab}$
must belong to the center. Finally it can be shown~\cite{DimMad96,
MadMou98} that in order that~(\ref{consis1}) and (\ref{consis2}) be
consistent with one another the original $\lambda_a$ must satisfy the
condition
\begin{equation}
2 \lambda_c \lambda_d P^{cd}{}_{ab} -
\lambda_c F^c{}_{ab} - K_{ab} = 0. \label{manca}
\end{equation}
This gives to the set of $\lambda_a$ the structure of a twisted Lie
algebra with a central extension.
We propose as definition of a linear connection a map~\cite{Kos60}
\begin{equation}
\Omega^1(\c{A}) \buildrel D \over \longrightarrow
\Omega^1(\c{A}) \otimes_\c{A} \Omega^1(\c{A}) \label{2.2.4}
\end{equation}
which satisfies both a left Leibniz rule
\begin{equation}
D (f \xi) = df \otimes \xi + f D\xi \label{2.2.2}
\end{equation}
and a right Leibniz rule~\cite{DubMadMasMou96}
\begin{equation}
D(\xi f) = \sigma (\xi \otimes df) + (D\xi) f
\end{equation}
for arbitrary $f \in \c{A}$ and $\xi \in \Omega^1(\c{A})$.
We have here introduced a generalized flip
\begin{equation}
\Omega^1(\c{A}) \otimes_\c{A} \Omega^1(\c{A})
\buildrel \sigma \over \longrightarrow
\Omega^1(\c{A}) \otimes_\c{A} \Omega^1(\c{A}) \label{2.2.5}
\end{equation}
in order to define a right Leibniz rule which is consistent with the
left one. It is necessarily bilinear. A linear connection is therefore
a couple $(D, \sigma)$. It can be shown that a necessary as well as
sufficient condition for torsion to be right-linear is that $\sigma$
satisfy the consistency condition
\begin{equation}
\pi \circ (\sigma + 1) = 0. \label{2.2.6}
\end{equation}
Using the fact that $\pi$ is a projection one sees that the most general
solution to this equation is given by
\begin{equation}
1 + \sigma = ( 1 - \pi) \circ \tau \label{2.2.7}
\end{equation}
where $\tau$ is an arbitrary bilinear map
\begin{equation}
\Omega^1(\c{A}) \otimes \Omega^1(\c{A})
\buildrel \tau \over \longrightarrow
\Omega^1(\c{A}) \otimes \Omega^1(\c{A}). \label{2.2.8}
\end{equation}
If we choose $\tau = 2$ then we find $\sigma = 1 - 2 \pi$ and
$\sigma^2 = 1$. The eigenvalues of $\sigma$ are then equal to $\pm 1$.
This general formalism can be applied in particular to differential
calculi with a frame. Since $\Omega^1(\c{A})$ is a free module the
maps $\sigma$ and $\tau$ can be defined by their action on the basis
elements:
\begin{equation}
\sigma (\theta^a \otimes \theta^b) =
S^{ab}{}_{cd} \theta^c \otimes \theta^d, \qquad
\tau (\theta^a \otimes \theta^b) =
T^{ab}{}_{cd} \theta^c \otimes \theta^d. \label{2.2.9}
\end{equation}
By the sequence of identities
\begin{equation}
f S^{ab}{}_{cd} \theta^c \otimes \theta^d =
\sigma (f \theta^a \otimes \theta^b) =
\sigma (\theta^a \otimes \theta^b f) =
S^{ab}{}_{cd} f \theta^c \otimes \theta^d \label{2.2.10}
\end{equation}
and the corresponding ones for $T^{ab}{}_{cd}$ we conclude that the
coefficients $S^{ab}{}_{cd}$ and $T^{ab}{}_{cd}$ must lie in
$\c{Z}(\c{A})$. From~(\ref{2.2.7}) the most general form for
$S^{ab}{}_{cd}$ is
\begin{equation}
S^{ab}{}_{cd} =
T^{ab}{}_{ef} (\delta^e_c \delta^f_d - P^{ef}{}_{cd})
- \delta^a_c \delta^b_d. \label{2.2.11}
\end{equation}
A covariant derivative can be defined also by its action on the basis
elements:
\begin{equation}
D\theta^a = - \omega^a{}_{bc} \theta^b \otimes \theta^c. \label{2.2.12}
\end{equation}
The coefficients here are elements of the algebra. They are restricted
by (2.1) and the the two Leibniz rules. The torsion 2-form is defined as
usual as
\begin{equation}
\Theta^a = d \theta^a - \pi \circ D \theta^a. \label{deftorsion}
\end{equation}
If $F^a{}_{bc} = 0$ then it is easy to check~\cite{DubMadMasMou96}
that
\begin{equation}
D_{(0)} \theta^a = - \theta \otimes \theta^a +
\sigma (\theta^a \otimes \theta) \label{2.2.14}
\end{equation}
defines a torsion-free covariant derivative.
The most general $D$ for fixed $\sigma$ is of the form
\begin{equation}
D = D_{(0)} + \chi \label{2.2.15}
\end{equation}
where $\chi$ is an arbitrary bimodule morphism
\begin{equation}
\Omega^1(\c{A}) \buildrel \chi \over \longrightarrow
\Omega^1(\c{A}) \otimes \Omega^1(\c{A}). \label{2.2.16}
\end{equation}
If we write
\begin{equation}
\chi (\theta^a) = - \chi^a{}_{bc} \theta^b \otimes \theta^c \label{2.2.17}
\end{equation}
we conclude that $\chi^a{}_{bc} \in \c{Z}(\c{A})$.
In general a covariant derivative is torsion-free provided the condition
\begin{equation}
\omega^a{}_{de} P^{de}{}_{bc} = {1\over 2}C^a{}_{bc} \label{2.2.19}
\end{equation}
is satisfied. The covariant derivative~(\ref{2.2.15}) is torsion free
if and only
if
\begin{equation}
\pi \circ \chi = 0. \label{2.2.20}
\end{equation}
One can define a metric by the condition
\begin{equation}
g(\theta^a \otimes \theta^b) = g^{ab} \label{2.2.21}
\end{equation}
where the coefficients $g^{ab}$ are elements of $\c{A}$. To be well
defined on all elements of the tensor product $\Omega^1(\c{A})
\otimes_\c{A} \Omega^1(\c{A})$ the metric must be bilinear and by
the sequence of identities
\begin{equation}
f g^{ab} = g(f \theta^a \otimes \theta^b)
= g(\theta^a \otimes \theta^b f) = g^{ab} f \label{2.2.22}
\end{equation}
one concludes that the coefficients must lie in $\c{Z}(\c{A})$.
We define the metric to be symmetric if
\begin{equation}
g \circ \sigma \propto g. \label{symm}
\end{equation}
This is a natural generalization of the situation in ordinary
differential geometry where symmetry is respect to the flip which
defines the forms. If $g^{ab} = g^{ba}$ then by a linear transformation
of the original $\lambda_a$ one can make $g^{ab}$ the components of the
Euclidean (or Minkowski) metric in dimension $n$. It will not
necessarily then be symmetric in the sense that we have just used the
word.
Introduce the standard notation $\sigma_{12} = \sigma \otimes 1$,
$\sigma_{23} = 1 \otimes \sigma$, to extend to three factors of a module
any operator $\sigma$ defined on a tensor product of two factors.
Then there is a natural continuation of the map (\ref{2.2.4}) to the
tensor product $\Omega^1(\c{A}) \otimes_\c{A} \Omega^1(\c{A})$
given by the map
\begin{equation}
D_2(\xi \otimes \eta) = D\xi \otimes \eta +
\sigma_{12} (\xi \otimes D\eta). \label{2.2.4e}
\end{equation}
The map $D_2 \circ D$ has no nice properties but if one introduces the
notation $\pi_{12} = \pi \otimes 1$ then by analogy with the commutative
case one can set
\begin{equation}
D^2 = \pi_{12} \circ D_2 \circ D
\end{equation}
and formally define the curvature as the map
\begin{equation}
\mbox{Curv} = D^2. \label{curv}
\end{equation}
Because of the condition (\ref{2.2.6}) Curv is left linear. It can
be written out in terms of the frame as
\begin{equation}
\mbox{Curv} (\theta^a) =
- {1 \over 2} R^a{}_{bcd} \theta^c \theta^d \otimes \theta^b \label{2.16}
\end{equation}
Similarly one can define a Ricci map
\begin{equation}
\mbox{Ric} (\theta^a) =
{1 \over 2} R^a{}_{bcd} \theta^c g(\theta^d \otimes \theta^b). \label{2.17}
\end{equation}
It is given by
\begin{equation}
\mbox{Ric} \, (\theta^a) = R^a{}_b \theta^b. \label{2.18}
\end{equation}
The above definition of curvature is not satisfactory in the
noncommutative case. For a discussion of this point we refer to the
article by Dubois-Violette {\it et al.}~\cite{DubMadMasMou96}.
The curvature of the covariant derivative $D_{(0)}$ defined in
(\ref{2.2.14}) can be readily calculated. One finds after a short
calculation that it is given by the expression
\begin{eqnarray}
\mbox{Curv} (\theta^a)\hspace{-17pt}
&& =\theta^2 \otimes \theta^a -
\pi_{12} \sigma_{23} \sigma_{12}
(\theta^a \otimes \theta \otimes \theta) \nonumber\\[4pt]
&& = \theta^2 \otimes \theta^a +
\pi_{12} \sigma_{12}\sigma_{23} \sigma_{12}
(\theta^a \otimes \theta \otimes \theta). \label{2.19}
\end{eqnarray}
If $\xi = \xi_a \theta^a$ is a general 1-form then since Curv is left
linear one can write
\begin{eqnarray}
\mbox{Curv} (\xi ) \hspace{-17pt}
&& = \xi_a \theta^2 \otimes \theta^a -
\pi_{12} \sigma_{23} \sigma_{12}
(\xi \otimes \theta \otimes \theta) \nonumber\\[4pt]
&& = \xi_a \theta^2 \otimes \theta^a +
\pi_{12} \sigma_{12}\sigma_{23} \sigma_{12}
(\xi \otimes \theta \otimes \theta). \label{2.20}
\end{eqnarray}
We shall use this latter expression in Section~\ref{metrics}.
The compatibility of the covariant derivative~(\ref{2.2.12})
with the metric is expressed by the condition~\cite{DubMadMasMou95}
\begin{equation}
g_{23}\circ D_2= d\circ g . \label{met-comp}
\end{equation}
The covariant derivative~(\ref{2.2.12}) is compatible with the metric if
and only if~\cite{DimMad96}
\begin{equation}
\omega^a{}_{bc} + \omega_{cd}{}^e S^{ad}{}_{be} = 0. \label{2.2.23}
\end{equation}
This is a `twisted' form of the usual condition that
$g_{ad}\omega^d{}_{bc}$ be antisymmetric in the two indices $a$ and $c$
which in turn expresses the fact that for fixed $b$ the
$\omega^a{}_{bc}$ form a representation of the Lie algebra of the
Euclidean group $SO(N)$ (or the Lorentz group). When $F^a{}_{bc} = 0$
the condition that~(\ref{2.2.12}) be metric compatible can be
written as
\begin{equation}
S^{ae}{}_{df} g^{fg} S^{bc}{}_{eg} = g^{ab} \delta^c_d. \label{2.2.24}
\end{equation}
The algebra we shall consider is a $*$-algebra and we require that the
differential calculus be such that the reality condition
\begin{equation}
(df)^* = df^* \label{d-real}
\end{equation}
holds. Neither of the two differential calculi we shall introduce in
Section~4 satisfies this condition. In Section~7 we shall discuss how to
construct a real calculus $\Omega_R^*(\c{A})$ by taking a subalgebra
of the tensor product of the two calculi. We shall require that for
arbitrary $f \in \c{A}$ and $\xi \in \Omega_R^1(\c{A})$ one has
\begin{equation}
(f \xi)^* = \xi^* f^*, \qquad (\xi f)^* = f^* \xi^*. \label{sola}
\end{equation}
We shall suppose~\cite{DubMadMasMou95} that the extension of the
involution to the tensor product is given by
\begin{equation}
(\xi \otimes \eta)^* = \sigma(\eta^* \otimes \xi^*). \label{TPI}
\end{equation}
A change in $\sigma$ therefore implies a change in the definition of
an hermitian tensor. The reality condition for the metric becomes
then~\cite{FioMad98}
\begin{equation}
g\circ \sigma (\eta^* \otimes \xi^*) =
(g(\xi \otimes \eta))^* \label{reality}
\end{equation}
There is also a reality condition on the covariant derivative and the
curvature~\cite{FioMad98} which imply that the generalized flip
$\sigma$ must satisfy the braid equation.
\sect{The quantum Euclidean space}
The $R$-matrix or braid matrix $\hat R\equiv\Vert \hat R^{ij}_{hk}\Vert$
of $SO_q(3)$ \cite{frt} is a $3^2\times 3^2$ matrix satisfying
the braid equation
\begin{equation}
\hat R_{12}\,\hat R_{23}\,\hat R_{12} =\hat R_{23}\,\hat R_{12}\,
\hat R_{23}. \label{braid1}
\end{equation}
Here we have used the conventional tensor notation
$\hat R_{12} = \hat R \otimes 1$ used above for $\sigma$.
By repeated application of the Equation~(\ref{braid1}) one finds
\begin{equation}
f(\hat R_{12})\,\hat R_{23}\,\hat R_{12}
=\hat R_{23}\,\hat R_{12}\,f(\hat R_{23}) \label{braid2}
\end{equation}
for any polynomial function $f(t)$ in one variable. The
Equations~(\ref{braid1}) and~(\ref{braid2}) are evidently satisfied also
after the replacement $\hat R\rightarrow\hat R^{-1}$. The braid matrix
admits the projector decomposition
\begin{equation}
\hat R = q\c{P}_s - q^{-1}\c{P}_a + q^{-2}\c{P}_t \label{decompo}
\end{equation}
with
\begin{equation}
\c{P}_{\mu}\c{P}_{\nu} = \c{P}_\mu \delta_{\mu\nu}, \qquad
\sum_{\mu}\c{P}_{\mu} = 1, \quad \mu,\nu = s,t,a.
\end{equation}
The $\c{P}_s$, $\c{P}_a$, $\c{P}_t$ are $SO_q(3)$-covariant
$q$-deformations of respectively the symmetric trace-free,
antisymmetric and trace projectors. The trace projector is
1-dimensional and its matrix elements can be written in the form
\begin{equation}
\c{P}_t{}_{kl}^{ij} = (g^{mn}g_{mn})^{-1} g^{ij}g_{kl}, \label{bibi}
\end{equation}
where~\cite{frt} the `metric matrix' $g= (g_{ij})$ is a
$SO_q(3)$-isotropic tensor, a deformation of the metric matrix on the
classical Euclidean space. The $\hat R$ and $g$ satisfy the relations
\begin{equation}
g_{il}\,\hat R^{\pm 1}{}^{lh}_{jk} =
\hat R^{\mp 1}{}^{hl}_{ij}\,g_{lk}, \qquad
g^{il}\,\hat R^{\pm 1}{}_{lh}^{jk} =
\hat R^{\mp 1}{}_{hl}^{ij}\,g^{lk}. \label{utile1}
\end{equation}
The lower-case Latin indices $i,j,\dots$ will take
the values $(-, 0, +)$. The quantum Euclidean space is the formal
(associative) algebra $\c{A}$ with generators $x^i = (x^-, x^0, x^+)$
and relations
\begin{equation}
\c{P}_a{}_{kl}^{ij} x^k x^l=0 \label{xxcr}
\end{equation}
for all $i,j$. We introduce a grading in $\c{A}$ by requiring that the
degree of $(x^-, x^0, x^+)$ be respectively equal to $(-1, 0, +1)$. The
matrix elements of $\hat R$ and therefore of all the projectors fulfill
the condition
\begin{equation}
\hat R^{ij}_{kl} = 0 \qquad \mbox{if} \quad i+j \neq k+l.
\end{equation}
Consequently, all the terms appearing on the left-hand side of
Equation~(\ref{xxcr}) have the same total degree $i\!+\!j$.
If we use the explicit expression~\cite{frt} for $\c{P}_a$ and set
\begin{equation}
h = \sqrt q - 1/\sqrt q,
\end{equation}
then the relations~(\ref{xxcr}) can be written in the form
\begin{equation}
\begin{array}{l}
x^- x^0 = q\, x^0 x^-,\\[2pt]
x^+ x^0 = q^{-1} x^0 x^+,\\[2pt]
[x^+, x^-] = h (x^0)^2.
\end{array} \label{xxcr3}
\end{equation}
The first two equations define two copies of the $q$-quantum plane with
$q$ and $q^{-1}$ as deformation parameter and a common generator $x^0$.
The metric matrix is given by
$g_{ij} = g^{ij}$ with
\begin{equation}
g_{ij} = \left(
\begin{array}{ccc}
0 &0 & 1/\sqrt{q} \\
0 &1 & 0 \\
\sqrt{q} &0 & 0 \\
\end{array}\right). \label{metric}
\end{equation}
When $q \in \b{R}^+$ one obtains the real quantum
Euclidean space by giving $\c{A}$ the structure of a $*$-algebra with
\begin{equation}
(x^-)^* = \sqrt q x^+, \qquad (x^0)^* = x^0, \qquad
(x^+)^* = {1\over\sqrt q} x^-.
\end{equation}
This can be written in terms of the metric as
\begin{equation}
(x^i)^* = x^j g_{ji}. \label{real}
\end{equation}
We can use the summation convention if we consider the involution to
lower (or raise) an index. The condition (\ref{real}) is an
$SO_q(3,\b{R})$-covariant condition and three linearly independent,
hermitian `coordinates' can be obtained as combinations of the
$x^i$. We define
\begin{equation}
x^r = \Lambda^r_i x^i,
\qquad \Lambda^r_i:= {1 \over \sqrt2} \left(
\begin{array}{ccc}
1 &0 &\sqrt q\\
0 &\sqrt 2 &0\\
i &0 &-i\sqrt q
\end{array}\right). \label{realmat}
\end{equation}
With respect to the new `coordinates' the metric is given by
$$
g^{rs} = g^{ij} \Lambda^r_i \Lambda^s_j = {1 \over 2} \left(
\begin{array}{ccc}
q+1 &0 &i(q-1)\\
0 &2 &0\\
-i(q-1) &0 &q+1
\end{array}\right).
$$
The metric is hermitian but no longer real. In the limit $q\to 1$ one
sees that $g^{rs} \to \delta^{rs}$. It is more convenient to remain
with the original `coordinates' and a real metric.
The `length' squared
\begin{equation}
r^2 := g_{ij} x^i x^j
\end{equation}
is an $SO_q(3,\b{R})$-invariant, real and generates the center
$\c{Z}(\c{A})$ of $\c{A}$. Using~(\ref{real}) and (\ref{xxcr3}) it can
be written also in the forms
\begin{equation}
r^2 = (x^i)^* x^i = (\sqrt{q}+\frac 1{\sqrt{q}})x^-x^++q
(x^0)^2 = (\sqrt{q}+\frac 1{\sqrt{q}})x^+x^-+q^{-1} (x^0)^2.
\label{squarelenght}
\end{equation}
The commutation relation between $x^+$ and
$x^-$ can also be written in the form
\begin{equation}
q x^+ x^- - q^{-1} x^- x^+ = h r^2.
\end{equation}
There is obviously no obstruction to extending $\c{A}$ by adding to it
the inverse $r^{-2}$ of $r^2$, and also the square roots $r, r^{-1}$ of
these two elements. We shall further extend the algebra $\c{A}$ by
adding the inverse $(x^0)^{-1}$ of $x^0$ as a new generator with the
obvious extra relations.
Finally, we observe that if $|q|=1$ a compatible involution is defined
rather by $(x^i)^*= x^i$. The algebra describes a quantum space
covariant under the noncompact section $SO_q(2,1)$ of $SO_q(3,\b{C})$.
\sect{The $q$-deformed Euclidean differential calculi} \label{calculi}
There are two $SO_q(3)$-covariant differential calculi~\cite{cawa}
$\Omega^*(\c{A})$ and $\bar \Omega^*(\c{A})$ over $\c{A}$ neither of
which satisfy the reality condition~(\ref{d-real}). In this section we
shall write them in a form which will permit us in Section~7 to look
for a sensible definition of a real differential calculus. Let $d$ and
$\bar d$ be the respective differentials and set $\xi^i = dx^i$ and
$\bar \xi^i = \bar dx^i$. The calculi are determined respectively by
the commutation relations
\begin{equation}
x^i \xi^j = q\,\hat R^{ij}_{kl} \xi^k x^l \label{xxicr}
\end{equation}
for $\Omega^1(\c{A})$ and
\begin{equation}
x^i \bar \xi^j =
q^{-1}\,\hat R^{-1}{}^{ij}_{kl} \bar\xi^k x^l \label{xxicr'}
\end{equation}
for $\bar\Omega^1(\c{A})$. Using formulae~(\ref{utile1}) and the fact
that we would like to require that the calculus satisfy~(\ref{d-real}),
it is easy to see~\cite{olezu} that, if $q\in \b{R}^+$,
it is not possible to extend the
involution~(\ref{real}) to either calculus. This is possible only if
$|q|=1$, and will be considered elsewhere. On the other hand, if
$q\in \b{R}^+$ the involution can be extended to the direct sum
$\Omega^1(\c{A})\oplus\bar\Omega^1(\c{A})$ by setting, as well as
(\ref{sola}),
\begin{equation}
(\xi^i)^* = \bar\xi^j g_{ji}, \label{barstern}
\end{equation}
since this exchanges relations~(\ref{xxicr}) and~(\ref{xxicr'}).
This will be equivalent to
\begin{equation}
(df)^* = \bar d f^*. \label{real-cond}
\end{equation}
If in the limit $q\to 1$ we identify $\bar\xi^i=\xi^i$,
we recover the standard involution and the
differential is real.
By means of Equations~(\ref{xxicr}) and~(\ref{utile1}) it is
straightforward to check that the commutation relations
\begin{equation}
r^2 \xi^i = q^2\,\xi^i r^2. \label{balla}
\end{equation}
are satisfied. The commutation relations between the $\xi^i$ are
derived by taking the differential of~(\ref{xxicr}):
\begin{equation}
\c{P}_s{}_{kl}^{ij} \xi^k \xi^l = 0, \qquad\qquad
\c{P}_t{}_{kl}^{ij} \xi^k \xi^l = 0. \label{xixicr}
\end{equation}
If we extend the grading of $\c{A}$ to $\Omega^1(\c{A})$ by setting
$\mbox{deg}(\xi^i)= \mbox{deg}(x^i)$, we find that each term in
(\ref{xxicr}), (\ref{xixicr}) must have the same total degree.
Written out explicitly the Equations~(\ref{xxicr}) become
\begin{eqnarray}
&&x^- \xi^- = q^2\, \xi^- x^-, \nonumber\\[2pt]
&&x^0 \xi^- = q\, \xi^- x^0, \\[2pt]
&&x^+ \xi^- = \xi^- x^+, \nonumber\\[5pt]
&&x^- \xi^0 = q\,\xi^0 x^- +(q^2-1)\xi^-x^0, \nonumber\\[2pt]
&&x^0 \xi^0 = q\,\xi^0 x^0 - h (q+1) \xi^- x^+, \\[2pt]
&&x^+ \xi^0 = q\,\xi^0 x^+, \nonumber\\[5pt]
&&x^- \xi^+ = \xi^+ x^- - h (q+1) \xi^0 x^0 +
h^2 (q+1) \xi^- x^+, \nonumber\\[2pt]
&&x^0 \xi^+ = q\, \xi^+ x^0 + (q^2-1)\xi^0 x^+,\\[2pt]
&&x^+ \xi^+ = q^2\, \xi^+ x^+ \nonumber
\end{eqnarray}
and the Equations~(\ref{xixicr}) become~\cite{fiodet}
\begin{eqnarray}
&&(\xi^{\pm})^2 = 0,\nonumber\\[2pt]
&&(\xi^0)^2 = h \xi^- \xi^+,\nonumber\\[2pt]
&&\xi^- \xi^0 = - q^{-1}\xi^0 \xi^-,\\[2pt]
&&\xi^0 \xi^+ = - q^{-1}\xi^+ \xi^0,\nonumber\\[2pt]
&&\xi^- \xi^+ = - \xi^+ \xi^-.\nonumber
\end{eqnarray}
Consider the $SO_q(3)$-invariant 1-form
\begin{equation}
\eta := g_{ij} x^i \xi^j = q^{-1}\, g_{ij} \xi^j x^i.
\end{equation}
Using the projector decomposition of the $\hat R$-matrix, the
relations~(\ref{utile1}) between the $\hat R$-matrix and its inverse
as well as the relations~(\ref{xxcr}) which define the algebra one can
easily verify that
\begin{equation}
[\eta, x^i] = - q^{-2} (q-1) r^2\xi^i.
\end{equation}
Hence we conclude that
\begin{equation}
\theta := (q-1)^{-1} q^2 r^{-2} \eta \label{deftheta}
\end{equation}
is the `Dirac operator'~(\ref{dirac}) of $\Omega^1(\c{A})$. It
satisfies the conditions
\begin{equation}
d\theta = 0, \quad \theta^2 = 0. \label{fifa}
\end{equation}
It is of interest to note that
\begin{equation}
dr^2 = (1-q^{-2})\, r^2\,\theta. \label{diff-r}
\end{equation}
Since $r \in \c{Z}(\c{A})$, from relations (\ref{diff-r}) one
concludes immediately that the differential calculus cannot be based
on derivations as outlined in the first section. That is, there can
exist no decompositions of $\theta$ as in~(\ref{dirac}). This fact is
not necessarily a defect but we shall change it anyway by adding an
extra element, the `dilatator', to the algebra since at the same time
we can reduce $\c{Z}(\c{A})$ to $\b{C}$.
The most general Ansatz for the frame can be written in the form
$\theta^a := \theta^a_i \xi^i$ where the $\theta^a_i$ are elements of
$\c{A}$. We shall let the lower-case Latin indices $a,b,...$ take the
same values $(+,0,-)$ as $i,j,...$. The condition~(\ref{fund}) implies
that
\[
(r^2\theta^a_i - q^{-2} \theta^a_ir^2) \xi^i = 0
\]
from which we can conclude that
\[
r^2\theta^a_i - q^{-2} \theta^a_i r^2= 0.
\]
This equation has obviously no solution since $r^2 \in \c{Z}(\c{A})$.
To remedy this problem we extend the algebra $\c{A}$ by adding an
extra generator $\Lambda$, the `dilatator', and its inverse
$\Lambda^{-1}$, chosen such that
\begin{equation}
x^i \Lambda = q\, \Lambda x^i . \label{lambda}
\end{equation}
The introduction of a new generator $\Lambda$ is necessary also in a
different context, namely in the inhomogeneous extension of the
homogeneous quantum groups $SL_q(N)$, $SO_q(N)$ and
$q$-Lorentz~\cite{wessvari, munich, maj2}; more precisely, $\Lambda$
appears in the coproduct of the translation part generators. We do
not know if this is a coincidence or there is some more fundamental
link between the two phenomena.
We extend the original algebra $\c{A}$ defined by
Equations~(\ref{xxcr3}) by the addition not only of
$(x^0)^{-1}, r^{\pm 1}$ but also of $\Lambda^{\pm 1}$ as new
generators.
Since
$$
r \Lambda = q\, \Lambda r,
$$
clearly the center of $\c{A}$ is now trivial: $\c{Z}(\c{A}) = \b{C}$.
It is natural to extend the grading by setting $\mbox{deg}(\Lambda) = 0$.
We shall choose $\Lambda$ to be
unitary
\begin{equation}
\Lambda^* = \Lambda^{-1}.
\end{equation}
To within a normalization this is the only choice compatible
with the commutation relations.
We can now consider the previous Ansatz for the frame but with
coefficients in the algebra $\c{A}$. We must assume a linear
dependence on the generator $\Lambda$ and write
\begin{equation}
\theta^a := \Lambda^{-1}\, \theta^a_j \xi^j \label{ok}
\end{equation}
where the $\theta^a_j$ are elements of $\c{A}$ which do not depend on
$\Lambda$. The $\Lambda$-dependence is here dictated by the condition
$[r,\theta^a] = 0$. The condition $[x^i, \theta^a] = 0$ becomes
\begin{equation}
x^i \theta^a_j = \hat R^{-1}{}^{ki}_{lj}\theta^a_k x^l. \label{xdcr}
\end{equation}
Written out explicitly these Equations become
\begin{eqnarray}
&&x^- \theta^a_- = q^{-1} \theta^a_- x^- -
q^{-1} (q^2-1) \theta^a_0 x^0 + h^2 (1+q) \theta^a_+ x^+,\nonumber\\[2pt]
&&x^0 \theta^a_- = \theta^a_- x^0 + h(1+q) \theta^a_0 x^+, \label{1st}\\[2pt]
&&x^+ \theta^a_- = q\,\theta^a_- x^+,\nonumber\\[5pt]
&&x^- \theta^a_0 = \theta^a_0 x^- + h(1+q) \theta^a_+ x^0,\nonumber\\[2pt]
&&x^0\theta^a_0 = \theta^a_0 x^0 -
q^{-1} (q^2-1) \theta^a_+ x^+, \label{2nd}\\[2pt]
&&x^+ \theta^a_0 = \theta^a_0 x^+,\nonumber\\[5pt]
&&x^- \theta^a_+ = q\, \theta^a_+ x^-,\nonumber\\[2pt]
&&x^0 \theta^a_+ = \theta^a_+ x^0, \label{3rd}\\[2pt]
&&x^+ \theta^a_+ = q^{-1} \theta^a_+ x^+.\nonumber
\end{eqnarray}
The Equations~(\ref{3rd}) contain only $\theta^a_+$ and admit, to within
an arbitrary factor in the center of $\c{A}$, only two independent
solutions. We choose
\begin{eqnarray*}
&&\theta^-_+ = 0, \\[2pt]
&&\theta^0_+ = 0, \\[2pt]
&&\theta^+_+ = r^{-2} x^0,
\end{eqnarray*}
so that $\theta^+$ contains a term proportional to $\xi^+$.
We replace this result in Equations~(\ref{2nd})
which become then equations for $\theta^a_0$ and we find the solutions
\begin{eqnarray*}
&&\theta^-_0 = 0, \\[2pt]
&&\theta^0_0 = r^{-1}, \\[2pt]
&&\theta^+_0 = - (q+1)r^{-2} x^+.
\end{eqnarray*}
Finally, the Equations~(\ref{1st}) can be solved for $\theta^a_-$
yielding
\begin{eqnarray*}
&&\theta^-_- = (x^0)^{-1}, \\[2pt]
&&\theta^0_- = \sqrt q\, (q+1) r^{-1} (x^0)^{-1} x^+, \\[2pt]
&&\theta^+_- = - \sqrt q\, q (q+1) r^{-2} (x^0)^{-1} (x^+)^2.
\end{eqnarray*}
If we place these coefficients into Equation~(\ref{ok}) we completely
determine the frame.
We shall include a rescaling by a normalization factor
$\alpha$, which we leave free for the moment:
\begin{eqnarray}
&&\theta^- = \alpha^{-1}\Lambda^{-1}\, (x^0)^{-1} \xi^-, \nonumber\\[4pt]
&&\theta^0 = \alpha^{-1}\Lambda^{-1}\, r^{-1}(\sqrt q\,
(q+1) (x^0)^{-1} x^+ \xi^- + \xi^0), \label{stehbein3}\\[4pt]
&&\theta^+ = \alpha^{-1}\Lambda^{-1}\, r^{-2}(-\sqrt q\,
q (q+1) (x^0)^{-1} (x^+)^2 \xi^- - (q+1) x^+ \xi^0 + x^0 \xi^+).\nonumber
\end{eqnarray}
Note that we have enumerated the $\theta^a$ so that
$\mbox{deg}(\theta^a)=(-1,0,1)$, exactly as for the $x^i$ and $\xi^i$.
The above $\theta^a$ are determined up to linear transformations
with coefficients depending on $r$.
The commutation rules between $\Lambda$ and $\theta^a$
will depend on their $r$-normalization
as well as the commutation rules between $\Lambda$
and $\xi^i$, which we have not specified yet.
By differentiating Equation~(\ref{lambda}) we obtain the condition
\begin{equation}
\xi^i \Lambda + x^i d\Lambda = q d\Lambda x^i + q \Lambda \xi^i
\end{equation}
on the differential $d\Lambda$. A possible solution is given by the two
conditions
\begin{equation}
x^i\, d\Lambda = q d\Lambda x^i, \qquad
\xi^i \Lambda = q \Lambda \xi^i. \label{zero}
\end{equation}
In particular one can consistently set $d\Lambda = 0$. We shall do in
the sequel although it means that the condition $df = 0$ does not imply
that $f \in \b{C}$. This is not entirely satisfactory since one
would like the only `functions' with vanishing exterior derivative to be
the `constant functions'. It could be remedied by considering
a more general solution to Equation~(\ref{zero}). This would however
complicate our calculations since it would increase the number of
independent forms by one. A necessary condition for $d\Lambda=0$ is
that $[\Lambda, \lambda_a] = 0$. The $\lambda_a$ which we give below in
Equation~(\ref{lambda'}) satisfy the latter because they are homogeneous
functions of $x^i$ of degree zero. The condition that $\Lambda$ commutes
with the $\theta^a$ fixes the normalization factor $\alpha$ in
(\ref{stehbein3}) to be proportional to 1. In Section~\ref{climit} we
shall choose $\alpha$ in $\b{R}^+$, first as a fixed scale and then as a
suitable function of $h$, so as to adjust the commutative limit of the
frame. Another possibility would be to impose a commutation relation
between $\xi$ and $\Lambda$ of the type
\begin{equation}
\Lambda \xi^i= \xi^i\Lambda. \label{zero'}
\end{equation}
This follows, for example, from the condition~\cite{oleg2}
$\Lambda df = q d \Lambda f$ which in turn would be equivalent to
considering $\Lambda$ a suitable element of the associated Heisenberg
algebra\footnote{Such an element has a simple realization in terms of
$x^i$ and $SO_q(3)$-covariant twisted derivatives~\cite{oleg2}}. The
condition that $\Lambda$ commutes with the $\theta^a$ would now fixes
the normalization factor $\alpha$ in (\ref{stehbein3}) to be
proportional to $r^{-1}$ and will thus change the metric by a
conformal factor $r^2$. After imposing either commutation rule
between $\Lambda$ and $\theta^a$ the frame will be determined up to a
linear transformation with coefficients in $\b{C}$.
A direct and lengthy calculation\footnote{Performed using the program
for symbolic computations REDUCE.} with the explicit expression for the
matrix $\hat R^{ab}_{ij}$~\cite{frt} shows that the $\theta^a_i$ satisfy
the relations
\begin{equation}
\hat R^{ab}_{cd}\, \theta^d_j\theta^c_i =
\theta^b_l\theta^a_k\, \hat R^{kl}_{ij}. \label{basic}
\end{equation}
These are $3^4=81$ equations. However, since both
$R^{kl}_{ij}\equiv \hat R^{lk}_{ij}$ and $\theta^a_i$ are
lower-triangular matrices, $3^2(3^2-1)/2 = 36$ of them are trivial
identities. The proof actually consists then in checking `only' 45
equations. By repeated application of relations (\ref{basic}) it
immediately follows that the same relations hold also after replacing
$\hat R$ by any polynomial $f(\hat R)$,
\begin{equation}
f(\hat R)^{ab}_{cd}\, \theta^d_j\theta^c_i =
\theta^b_l\theta^a_k\, f(\hat R)^{kl}_{ij}. \label{basic1}
\end{equation}
In particular we can choose $f(\hat R) = \c{P}_t,\c{P}_s$. With the
help of relations~(\ref{fund}) and~(\ref{ok}) we find as a consequence
that the $\theta^a$ have the same commutation relations as the $\xi^i$:
\begin{equation}
\c{P}_t{}^{ab}_{cd} \theta^c \theta^d = 0 \qquad\qquad
\c{P}_s{}^{ab}_{cd} \theta^c \theta^d = 0. \label{ththcr}
\end{equation}
Therefore the $P$ of Equation~(\ref{manca}) is given by
\begin{equation}
P = \c{P}_a. \label{whynot}
\end{equation}
It is the $SO_q(3)$-covariant deformation of the antisymmetric projector.
Note also that for $f(\hat R)=\c{P}_t$ Equation~(\ref{basic1})
is equivalent to the relations
\begin{equation}
g_{cd}\theta^d_j\theta^c_i = \kappa g_{ij}\qquad\qquad
g^{ij}\theta^b_j\theta^a_i = \kappa g^{ab} \label{mamam}
\end{equation}
with
$$
\kappa = (g_{kl}g^{kl})^{-1} g_{ab} \theta^b_i \theta^a_j g^{ji}
= r^{-2}\alpha^{-2}
$$
because of Equation~(\ref{bibi}).
Consider the elements $\lambda_a \in \c{A}$ with
\begin{eqnarray}
&&\lambda_- = h^{-1}q \Lambda(x^0)^{-1}\alpha x^+, \nonumber\\[2pt]
&&\lambda_0 =
- h^{-1}\sqrt q \Lambda\alpha (x^0)^{-1} r, \label{lambda'}\\[2pt]
&&\lambda_+ = - h^{-1}\Lambda\alpha (x^0)^{-1} x^-. \nonumber
\end{eqnarray}
By direct calculation one verifies that
$\theta = - \lambda_a \theta^a$ is given by~(\ref{deftheta}).
Since $\Lambda$ is unitary the hermitian adjoints $\lambda_a^*$
are given by
\begin{equation}
\lambda_\pm^* = - \Lambda^{-2} g^{\pm b}\lambda_b, \qquad
\lambda_0^* = \Lambda^{-2} g^{0b}\lambda_b. \label{stira}
\end{equation}
The fact that the $\lambda_a$ are not anti-hermitian is related to the
fact that the differential $d$ is not real. We have chosen this rather
odd normalization to have the commutation relations (\ref{lambda-com})
below. A straightforward calculation yields the commutation relations
$[\lambda_a, x^i] = q\,\Lambda e_a^i$ with
\begin{equation}
(e_a^i) = \,\alpha\,\left(
\begin{array}{ccc}
+ x^0 & 0 & 0 \\
- (\sqrt q + 1/\sqrt q) x^+ & r & 0\\
-\sqrt q\, (q+1) (x^0)^{-1} (x^+)^2 & (q+1) r (x^0)^{-1} x^+
& r^2 (x^0)^{-1}
\end{array}\right).
\end{equation}
The $e_a^i$ is the inverse matrix of $\theta^a_i$:
\begin{equation}
e_a^i\theta^a_j = \delta^i_j\qquad\qquad
\theta^a_je^j_b=\delta^a_b. \label{inverse}
\end{equation}
Relations (\ref{basic}), (\ref{mamam}) imply that the matrix elements
$e^i_a$ satisfy also the `$RTT$-relations' \cite{frt}
\begin{equation}
\hat R_{kl}^{ij}\, e^k_a e^l_b
= e^i_c e^j_d \,\hat R_{ab}^{cd} \label{RTT}
\end{equation}
as well as the `$gTT$-relations' \cite{frt}
\begin{equation}
g^{ab} e^i_a e^j_b = r^2 \alpha^2 g^{ij}\qquad\qquad
g_{ij} e^i_a e^j_b = r^2 \alpha^2 g_{ab}. \label{gTT}
\end{equation}
Thus $(\alpha r)^{-1}e^i_a$ fulfill the same commutation relations of
the generators $T^i_a$ of $SO_q(3)$ and in this sense may be seen as a
`local' realization of $T^i_a$. However they do not satisfy the same
$*$-relations, nor is there an analog for the coproduct of $SO_q(3)$.
The $\lambda_a$ satisfy the commutation relations
\begin{eqnarray}
&&\lambda_- \lambda_0 = q\, \lambda_0 \lambda_-, \nonumber\\[2pt]
&&\lambda_+ \lambda_0 = q^{-1} \lambda_0 \lambda_+,\label{lambda-com}\\[2pt]
&&[\lambda_+, \lambda_-] = h\, (\lambda_0)^2. \nonumber
\end{eqnarray}
These are the same commutation relations as those satisfied
by the $x^i$. This is a remarkable fact and it underlines how weak a
constraint the commutation relations are on the algebra. The
$\lambda_a$ are related in fact to the $x^i$ by a rather complicated
nonlinear relation (\ref{lambda'}). They differ however from the
$x^a$ in that they commute with $\Lambda$.
The Equations~(\ref{lambda-com}) can be rewritten more compactly in the
form
\begin{equation}
\c{P}_a{}^{ab}_{cd} \lambda_a\lambda_b = 0, \label{lambdacr1}
\end{equation}
so that the constants $F^a{}_{bc}$ and $K_{ab}$ in Equation~(\ref{manca})
vanish. Hence Equation~(\ref{rot-coef}) is satisfied with
\begin{equation}
C^a{}_{bc} = - 2 \lambda_d \c{P}_a{}^{(da)}{}_{bc}
\end{equation}
an expression which is consistent with (\ref{consis1}) because of
(\ref{whynot}). Finally, it is easy to check that
\begin{equation}
g^{ab} \lambda_a \lambda_b = q h^{-2} (\Lambda\alpha)^2. \label{lambdacr2}
\end{equation}
One can easily find the relation between the $SO_q(3)$-covariant
derivatives $\partial_i$ introduced in Ref. \cite{cawa}, which fulfil the
modified Leibniz rule
\begin{equation}
\partial_ix^j=\delta_i^j+q\hat R^{jh}_{ik}x^k\partial_h ,
\end{equation}
and the $e_a$.
From the decomposition $d=\xi^i\partial_i=\theta^ae_a$ it is evident that
\begin{equation}
\partial_i= \Lambda^{-1}\tilde\theta^a_ie_a, \label{covder}
\end{equation}
where we have now decompesed $\theta^a$ in the form
$\theta^a=\xi^i\Lambda^{-1}\tilde\theta^a_i$.
We conclude this section by listing the basic formulae which, besides
(\ref{xxicr'}), characterize the barred differential calculus
$\bar\Omega^1(\c{A})$. The analogs of (\ref{xixicr}) are obtained by
taking the differential of~(\ref{xxicr'}):
\begin{equation}
\c{P}_s{}_{kl}^{ij} \bar\xi^k \bar\xi^l = 0, \qquad\qquad
\c{P}_t{}_{kl}^{ij} \bar\xi^k \bar\xi^l = 0. \label{barxibarxicr}
\end{equation}
All relations are compatible with the grading of $\c{A}$ extended by
setting $\mbox{deg}(\bar\xi^i)=\mbox{deg}(x^i)$. The
$SO_q(3)$-invariant 1-form
\begin{equation}
\bar\eta := g_{ij} x^i \bar\xi^j = q\,g_{ij} \bar\xi^j x^i
\end{equation}
fulfills in $\bar\Omega^1(\c{A})$ the commutation relations
\begin{equation}
[\bar\eta, x^i] = - q^2 (q^{-1}-1) r^2 \bar\xi^i.
\end{equation}
Hence
\begin{equation}
\bar\theta := (q^{-1} -1)^{-1} q^{-2} r^{-2} \bar\eta \label{defbartheta}
\end{equation}
is the `Dirac operator'~(\ref{dirac}) of $\bar\Omega^1(\c{A})$, and
\begin{equation}
\bar d\bar\theta = 0, \quad \bar\theta^2 = 0.
\end{equation}
From the definitions of $\theta$, $\bar\theta$ and the involution
it follows that in $\Omega^1(\c{A})\oplus\bar\Omega^1(\c{A})$
\begin{equation}
\theta^* = - \bar\theta, \qquad (\bar\theta)^* = - \theta. \label{stern}
\end{equation}
It is straightforward to show that
\begin{equation}
\bar dr^2 = (1-q^2)\, r^2\,\bar\theta.
\end{equation}
There exist a frame $\bar\theta^a$ for the differential calculus
$\bar\Omega^*(\c{A})$ in the form
\begin{equation}
\bar\theta^a := \Lambda \bar\theta^a_j\bar\xi^i, \label{barok}
\end{equation}
where the $\bar\theta^a_j$ are elements of $\c{A}$ which do not
depend on $\Lambda$. The $\Lambda$-dependence is here again dictated
by the condition $[r,\bar \theta^a] = 0$. The condition
$[x^i, \bar \theta^a] = 0$ becomes
\begin{equation}
\bar \theta^a_i x^j =
\hat R^{jk}_{il} x^l \bar \theta^a_k. \label{barxdcr}
\end{equation}
The elements $\bar\lambda_a\in\c{A}$ are introduced
through the decomposition
\begin{equation}
\bar\theta = - \bar\lambda_a \bar\theta^a \label{bardirac}
\end{equation}
and are dual to $\bar\theta^a$. As above, $\bar \theta^a$ and the
corresponding $\bar\lambda_a$ are determined up to a linear
transformation with coefficients in $\b{C}$. Now note that from $0 =
[\c{A},\theta^a]^* = [(\theta^a)^*,\c{A}]$ it follows that for
$q$ real positive
$(\theta^a)^*$ is a combination of $\bar\theta^b$. We choose the
second basis so that
\begin{equation}
(\theta^a)^*= \bar\theta^b g_{ba}. \label{bar-theta}
\end{equation}
This will automatically yield $\mbox{deg}(\bar\theta^a)=(-1,0,1)$
and the same commutation relations as for the $\theta^a$,
because of relations (\ref{utile1}):
\begin{equation}
\c{P}_t{}^{ab}_{cd} \bar\theta^c \bar\theta^d = 0 \qquad\qquad
\c{P}_s{}^{ab}_{cd} \bar\theta^c \bar\theta^d = 0. \label{bth-bthcr}
\end{equation}
The explicit expressions for $\bar \theta^a$ becomes
\begin{eqnarray}
&&\bar\theta^- = (\alpha q)^{-1} \, \Lambda \, r^{-2}(x^0 \bar\xi^- -
(q^{-1}+1) x^- \bar\xi^0 -\frac1{\sqrt q}\, q^{-1}(q^{-1}+1)
(x^0)^{-1} (x^-)^2 \bar\xi^+), \nonumber\\[4pt]
&&\bar\theta^0 = (\alpha q)^{-1}\, \Lambda \, r^{-1} (\bar\xi^0 +
\frac 1{\sqrt q}\,
(q^{-1}+1)(x^0)^{-1} x^- \bar\xi^+), \label{barstehbein3}\\[4pt]
&&\bar\theta^+ = (\alpha q)^{-1}\, \Lambda \, (x^0)^{-1} \bar \xi^+.
\nonumber
\end{eqnarray}
The corresponding $\bar\lambda_a$ are given by
$$
\bar\lambda_\pm = \Lambda^{-2} \lambda_\pm, \qquad
\bar\lambda_0 = - \Lambda^{-2} \lambda_0
$$
and therefore the involution on the $\lambda_a$ becomes
\begin{equation}
\lambda_a^* = - g^{ab}\bar\lambda_b.
\end{equation}
This is to be compared with~(\ref{real}).
\sect{Metrics and linear connections on the \\
quantum Euclidean space} \label{metrics}
We now look for metrics, generalized permutations and covariant
derivatives corresponding to each of the above differential
calculi. They will be essential ingredients do determine the correct
correspondence between mathematical objects and physical observables.
We shall see that it is not possible to satisfy all the requirements of
Section~\ref{preli}. Nevertheless, leaving the problem of reality aside
for the moment, to be treated in Section~\ref{involu}, we show that if
we allow a conformal factor in the metric then a unique linear
connection exists which is metric compatible and automatically
$SO_q(3)$-covariant.
We define covariant derivatives $D$ on $\Omega^1(\c{A})$ and $\bar D$ on
$\bar\Omega^1(\c{A})$ as maps
\begin{eqnarray*}
&&\Omega^1(\c{A}) \buildrel D \over \longrightarrow
\Omega^1(\c{A}) \otimes \Omega^1(\c{A}), \\[2pt]
&&\bar\Omega^1(\c{A}) \buildrel \bar D \over \longrightarrow
\bar\Omega^1(\c{A}) \otimes \bar\Omega^1(\c{A})
\end{eqnarray*}
which satisfy left and right Leibniz rules.
In accord with Equation~(\ref{2.2.9}), we look
for a generalized permutation $\sigma$ by starting with the Ansatz
\begin{equation}
\sigma (\theta^a\otimes \theta^b) =
S^{ab}{}_{cd}\,\theta^c\otimes \theta^d \label{gigio2}
\end{equation}
with $S^{ab}{}_{cd}$ complex numbers and we impose the
condition (\ref{2.2.6}). The Equation (\ref{ththcr})
implies that
\begin{equation}
S = C_s\c{P}_s - \c{P}_a + C_t\c{P}_t, \label{gala}
\end{equation}
where $C_s$ and $C_t$ are complex $3^2\times 3^2$ matrices.
Similarly, according to (\ref{2.2.21}) to define a
metric $g$ we start with the Ansatz
\begin{equation}
g(\theta^a\otimes\theta^b) = g^{a b} \label{fine`}
\end{equation}
with $g^{ab}$ again complex numbers. In view of Equation~(\ref{utile1})
a necessary condition for the metric-compatibility
condition~(\ref{2.2.24}) is that $g_{ab}$ be proportional to the matrix
defined in (\ref{metric}) and either $S = \hat R$ or $S = \hat R^{-1}$.
Without loss of generality we can set the proportionality factor equal
to 1, since this amounts to a redefinition of the factor $\alpha$ in
(\ref{stehbein3}). But because of (\ref{decompo}) these forms for $S$
are clearly not compatible with (\ref{gala}).
The best one can do is to weaken (\ref{2.2.24}) to a condition of
proportionality. To fulfill the latter it is sufficient that $S$ be
proportional to $\hat R$ or $\hat R^{-1}$. Then the double requirement
admits the two solutions
\begin{equation}
S = q\hat R, \qquad\qquad S = (q\hat R)^{-1}, \label{double}
\end{equation}
corresponding respectively to $C_s = q^2$, $C_t=q^{-1}$ or
$C_s = q^{-2}$, $C_t = q$. Therefore we have respectively
$$
S {}^{ae}{}_{df} g^{fg} S {}^{cb}{}_{eg}
= q^{\pm 2} g^{ac} \delta^b_d.
$$
In both cases $\sigma$ fulfills the braid Equation~(\ref{braid1}). If
we compare the above equation with (\ref{2.2.24}) we see that the
metric~(\ref{fine`}) is not compatible with the linear connection. We
shall show however below in Equation~(\ref{line-element}) that with a
conformal factor it is so; the (flat) linear connection is equal to the
Levi-Civita connection of a (flat) metric conformally equivalent to the
one which we have found. The symmetry condition (\ref{symm}) for the
metric follows from (\ref{utile1}) and from (\ref{decompo}). Since the
matrix $g^{ab}$ is not symmetric in the ordinary sense there can exist
no linear transformation $\theta^{\prime a} = \Lambda^a_b \theta^b$ such
that $g^{\prime ab} = \delta^{ab}$.
{From} Equations~(\ref{ok}), (\ref{basic1}), and (\ref{gigio2}) it is
possible to determine the action of $\sigma$ and $g$ on the
basis $\xi^i$. One finds that
\begin{equation}
\sigma (\xi^i\otimes \xi^j) =
S^{ij}{}_{hk}\,\xi^h\otimes \xi^k \label{gigio0}
\end{equation}
and
\begin{equation}
g(\xi^i\otimes \xi^j) =
g^{ij}\, \alpha^2 q^{-1}\,r^2\Lambda^2. \label{final}
\end{equation}
Equation~(\ref{gigio0}) his is the same formula as (\ref{gigio2});
Equation~(\ref{final}) is to be compared with the
Equation~(\ref{fine`}). From these results it is manifest that
$\sigma$ and $g$ are $SO_q(3)$-covariant; under the $SO_q(3)$ coaction
$\sigma(\xi^i\otimes \xi^j)$ and $g(\xi^i\otimes \xi^j)$ transform as
$\xi^i\otimes \xi^j$. Relations (\ref{gigio0}), (\ref{final}) of course
could have been obtained also by a direct calculation in the $\xi^i$
basis.
For either choice of $\sigma$ the covariant derivative
\begin{equation}
D_{(0)} \xi =
-\theta \otimes \xi + \sigma (\xi \otimes \theta), \label{conn}
\end{equation}
defined in (\ref{2.2.14}) is manifestly $SO_q(3)$-invariant, because
$\theta$ is. The most general $D$ is given by
Equation~(\ref{2.2.15}). But because of Equations~(\ref{ththcr}) we have
$$
\pi\circ \chi(\theta^a) = - \chi^a{}_{bc} \theta^b \theta^c =
- \chi^a{}_{bc} (\c{P}_s +\c{P}_a +\c{P}_t)^{bc}_{de} \theta^d \theta^e =
- \chi^a{}_{bc} \c{P}_a{}^{bc}_{de} \theta^d \theta^e
$$
and because of Equations~(\ref{whynot}) this extra term must vanish if
we require the torsion to vanish. Finally, one can show that a term
$\chi^a{}_{bc} (\c{P}_s +\c{P}_t)^{bc}_{de} \theta^d \otimes\theta^e$
is forbidden by $SO_q(3)$-covariance. The most general torsion-free and
$SO_q(3)$-covariant linear connection is given then by
\begin{equation}
D = D_{(0)}. \label{g-conn}
\end{equation}
The linear curvature tensor is given by Equation~(\ref{2.20}); the first
term is absent, because $\theta^2 = 0$. The result that the curvature
map $\mbox{Curv}$ is left-linear but in general not right-linear is
particularly evident from this formula. In fact it is easy to see that
$\mbox{Curv} = 0$. This relies only on the fact that $S$ fulfills the
braid equation (\ref{braid1}):
\begin{eqnarray}
\mbox{Curv}(\xi) \hspace{-17pt}&& = \pi_{12}\sigma_0{}_{12}
\sigma_0{}_{23}\sigma_0{}_{12}
(\theta^a\otimes \theta^b\otimes \theta^c)\:\xi_a\lambda_b\lambda_c
\nonumber\\[4pt]
&& = \pi_{12}(S_{12}S_{23}S_{12})^{abc}{}_{def}
(\theta^d\otimes \theta^e\otimes \theta^f)\:\xi_a\lambda_b\lambda_c
\nonumber\\[4pt]
&& = (S_{12}S_{23}S_{12})^{abc}{}_{def}
(\theta^d\theta^e\otimes \theta^f)\:\xi_a\lambda_b\lambda_c
\nonumber\\[4pt]
&& = -(S_{12}S_{23}\c{P}_a{}_{12})^{abc}{}_{def}
(\theta^d\theta^e\otimes \theta^f)\:\xi_a\lambda_b\lambda_c
\nonumber\\[4pt]
&& = -(\c{P}_a{}_{23}S_{12}S_{23})^{abc}{}_{def}
(\theta^d\theta^e\otimes \theta^f)\:\xi_a\lambda_b\lambda_c
\nonumber\\[4pt]
&& = 0. \nonumber
\end{eqnarray}
The last three equalities follow from respectively
Equations~(\ref{2.2.6}), (\ref{braid2}) and (\ref{lambdacr1}).
It is interesting to compute the result of the two flat connections on
the differentials. On one hand, if we choose $S = (q\hat R)^{-1}$ then
using Formulae~(\ref{xxicr}) and (\ref{utile1}) it is straightforward to
show that
\begin{equation}
D\xi^i=0. \label{cicca}
\end{equation}
This means that the `coordinates' are adapted to the zero-curvature
condition. On the other hand, it is straightforward to show that if
$S = q\hat R$ then
\begin{eqnarray}
D\xi^i \hspace{-17pt}&&=
(q^2-1)\theta \otimes \xi^i -(1+q^{-3})r^{-2}x^i [q^2\xi^l\otimes \xi^m
g_{lm}-q^4 g_{hj}\hat R^{ji}_{lm}x^h \xi^l\otimes \xi^m] \nonumber\\[4pt]
&&= (q^2-1)[\theta \otimes \xi^i +q^{-2}\xi^i\otimes \theta]
-q^2(1+q^{-1})r^{-2}x^i\xi^l\otimes \xi^mg_{lm}. \label{explicit}
\end{eqnarray}
This is different from zero. In this case the `coordinates' are not well
adapted.
We can repeat the same construction for the barred calculus
$\bar\Omega^1(\c{A})$. We simply list the results. There are
essentially a unique metric $g$ and two generalized permutations
$\sigma$ which fulfill the metric compatibility weakened by a conformal
factor. They are defined by
\begin{eqnarray}
&&g(\bar\theta^a\otimes\bar\theta^b) = g^{ab} \label{bbfine}\\[2pt]
&&\sigma (\bar\theta^a\otimes \bar\theta^b) =
\bar S^{ab}{}_{cd}\,\bar\theta^c\otimes \bar\theta^d, \label{bbgigio2}
\end{eqnarray}
with either
\begin{equation}
\bar S = q\hat R, \qquad\mbox{or}\qquad \bar S = (q\hat R)^{-1}.
\label{bbdouble}
\end{equation}
Correspondingly
$$
\bar S {}^{ae}{}_{df} g^{fg} \bar S {}^{cb}{}_{eg}
= q^{\pm 2} g^{ac} \delta^b_d.
$$
In the $\bar\xi^i$ basis,
\begin{eqnarray}
&&\sigma (\bar \xi^i\otimes \bar \xi^j) =
\bar S^{ij}{}_{hk}\,\bar\xi^h\otimes \bar\xi^k, \label{bbgigio0} \\[2pt]
&&g(\bar \xi^i\otimes \bar\xi^j) =
g^{ij}\, \alpha^2 q\, r^2\Lambda^{-2}.
\label{bbfinal}
\end{eqnarray}
The most general torsion-free $SO_q(3)$-covariant and (up to a conformal
factor) metric compatible covariant derivatives are
\begin{equation}
\bar D \bar \xi = -\bar \theta \otimes \bar \xi +
\sigma (\bar \xi \otimes \bar \theta) \label{bconn}
\end{equation}
with either choice of $\sigma$. They are manifestly
$SO_q(3)$-invariant. The associated linear curvatures
$\overline{\mbox{Curv}}$ vanish.
If one extends the involution to the second tensor power of the 1-form
algebra using (\ref{TPI}), then it is easy to verify that neither $D$
nor $\bar D$ are real. If $\bar S = S^{-1}$, they fulfill however
instead the relation
\begin{equation}
(D \xi)^* = \bar D \xi^*. \label{Dstar}
\end{equation}
\sect{The commutative limit} \label{climit}
The set $(x^i, e_a)$ generates the algebra $\c{D}_h$ of observables of
phase space; the $x^i$ generate the subalgebra of position observables,
whereas the $e_a$ the subalgebra of momenta observables. One has to look
for a complete set of three (the dimension of the classical underlying
manifold) independent commuting observables within the whole $\c{D}_h$,
and not just within the position space subalgebra, since the latter is
not abelian. Of course this is exactly what makes noncommutative
geometry interesting, in that it makes completely localized states
impossible. Here we shall restrict our consideration especially to
position space observables. It should be the theory which indicates
what is the correct identification of the operators $x^i$ in commutative
limit. By theory we mean the algebra, that is, the theory proper, as
well as the representation or the state. Below we find two possible and
rather different identifications of the $x^i$. A physically
satisfactory understanding should involve a Hilbert space representation
theory of the algebra such that the spectrum of the position observables
becomes `dense' in the limit manifold. Since the present analysis is
meant to be preliminary and heuristic in character, we shall
restrict our attention to the algebra.
In the commutative limit $\Lambda$ commutes both with coordinates and
derivations, and therefore must be equal to a constant. Since we have
already normalized $\Lambda$ so as to be unitary, the constant must be a
pure phase. One can always absorb the latter in the same normalization,
so that
\begin{equation}
\lim_{q\to 1} \Lambda =1. \label{L-limit}
\end{equation}
In the first identification we pick a fixed real $\alpha$ in
Equation~(\ref{stehbein3}) and suppose that the generators of the
algebra tend to their naive natural limit as (complex) coordinates on
a real manifold and that the frame tends to the corresponding limit of
a moving frame on this manifold.
{From} (\ref{final}) we find that in the commutative limit the metric is
given by the line element \begin{equation} ds^2 = (\alpha r)^{-2} \delta_{i,-j} dx^i
dx^j = (\alpha r)^{-2} (dx^2 + dy^2 + dz^2). \label{line-element} \end{equation}
The second expression is in terms of the real coordinates
(\ref{realmat}). If one uses spherical polar coordinates then one sees
immediately that the Riemannian space is $S^2 \times \b{R}$ with
$\log (\alpha r)$ the preferred coordinate along the line. The radius
of the sphere is equal to $\alpha^{-1}$. An interesting feature of this
identification is that neither the covariant derivative (\ref{g-conn})
nor (\ref{bconn}) is compatible with this non-flat metric. This was of
course to be expected since none of the $\sigma$ we have used satisfies
the metric compatibility condition (\ref{2.2.24}). The problem we are
considering here lies in fact a little outside the range of the general
theory of Section~2 because of the element $\Lambda$ which is not in the
center but which has nevertheless a vanishing differential.
Alternatively if $\alpha$ is proportional rather to $r^{-1}$ the
conformal factor in (\ref{line-element}) would disappear and the metric
would be the ordinary flat metric of $\b{R}^3$. We observed in
Section~4 that this would follow from the commutation relation
(\ref{zero'}).
In the commutative limit the frame~(\ref{stehbein3})
becomes a moving frame in the sense of Cartan. The singularity at
$x^0 = 0$ is to be expected , since there can be no global frame on a
non-parallelizable manifold like a sphere; for positive and negative
values of $x^0$ one has two different local sections of the frame bundle
on the two charts corresponding respectively to the upper and lower
hemisphere. According to (\ref{bar-theta}), the frame is however not
real. To find a real frame we first try the same linear transformation
(\ref{realmat}) used for the coordinates
$$
\theta^1 = {1 \over \sqrt 2} (\theta^- + \theta^+), \qquad
\theta^2 = \theta^0, \qquad
\theta^3 = {i \over \sqrt 2} (\theta^- - \theta^+).
$$
A short calculation yields
\begin{eqnarray}
&&\theta^1 = (\alpha x^0 r)^{-1} (rdx - xdr + i z dr), \nonumber\\[2pt]
&&\theta^2 = (\alpha x^0 r)^{-1} (rdr - i x dz + i z dx), \label{clf} \\[2pt]
&&\theta^3 = (\alpha x^0 r)^{-1} (rdz - i xdr - z dr).
\nonumber
\end{eqnarray}
Although in the commutative limit the differential is real, we see that
the frame is not. Equivalently, from (\ref{bar-theta}) we see that the
frames for the two differential calculi do not coincide in the
commutative limit, although the two differential calculi themselves do.
It is often found in specific calculations in General Relativity that it
is more convenient to use a complex frame to calculate real curvature
invariants. We can see however no property of the frame (\ref{clf})
which makes it particularly adapted to study the space $S^2 \times \b{R}$,
or the space $\b{R}^3$ in the case that $\alpha\propto r^{-1}$.
A more complete analysis should involve at this point the study of the
$*$-representa\-tions of the algebra of observables $\c{D}_h$, studied
e.g. in Ref.~\cite{fioijmpa}. We shall not enter it here, but simply
recall that the definition of the commutative limit of these
representations is rather delicate.
We shall now propose a different identification of the $x^i$ in the
commutative limit. The limit manifold is the desired $\b{R}^3$ and
for finite $h$ the $x^i$ will be a sort of general coordinates on
$\b{R}^3$. They are related to cartesian coordinates by a
transformation which becomes singular in the commutative limit. The
singularity can be removed by performing a renormalization procedure
(with diverging renormalization constant $\alpha^{-1}$
\footnote{Or with
a function $\alpha^{-1}=\alpha'r$, with $\alpha'$ a diverging constant,
in the case (\ref{zero'})})
before taking
the limit. We thus obtain cartesian coordinates $y^i$ on $\b{R}^3$.
We try to solve algebraically the problems arising from the first
identification. We reconsider the search of a real nondegenerate frame
in the commutative limit. If the condition
$\lim\limits_{h\rightarrow 0}x=0=\lim\limits_{h\rightarrow 0} z$, or
equivalently
\begin{equation}
x^{\pm}_{(0)}:=\lim\limits_{h\rightarrow 0}x^{\pm}=0, \label{problem2}
\end{equation}
were fulfilled on all points of the classical manifold, the frame
(\ref{clf}) would be real in the limit; apparently the latter would
also be degenerate ($\theta^1 = 0 = \theta^3$), but the latter
consequence can be avoided by choosing a diverging normalization
factor $\alpha^{-1}$ in (\ref{stehbein3}) and related definitions.
Let us first show by a one-dimensional example how such a
renormalization may change the situation. In the sequel we shall append
a suffix ${(0)}$ to any quantity to denote its commutative limit, for
example $r_{(0)}=\lim_{h\rightarrow 0} r$. Let $y\in\b{R}$ and
consider the change of coordinates
\begin{equation}
x=f(\alpha y)
\end{equation}
with some $\alpha$ going to zero when $h$ does and $f$ invertible, which
we shall assume normalized in such a way that $f'(0)=1$. Although
$x_{(0)} = f(0) = \mbox{constant}$, for example, $f(0)=0$ we have
$$
dx_{(0)}=0, \qquad \frac{\partial}{\partial x}|_{h=0}=\infty.
$$
Nevertheless one can extract the original nontrivial coordinate,
differential and derivative by taking a difference and performing a
rescaling before taking the limit:
\begin{equation}
y=\lim\limits_{h\rightarrow 0}\frac{x-x_{(0)}}{\alpha} \qquad
\theta^1\equiv dy=\lim\limits_{h\rightarrow
0}\frac{dx-dx_{(0)}}{\alpha}\qquad
\frac{\partial}{\partial y}=\lim\limits_{h\rightarrow 0}
\alpha \frac{\partial}{\partial x}. \label{rescale}
\end{equation}
Note that the above convergences are not uniform in $y$, namely are
slower as $y$ gets larger. This example suggests that we may solve the
degeneracy problem by taking the normalization factor $\alpha$ in
formula (\ref{stehbein3}) as a suitable infinitesimal in $h$.
Now we return to the the algebra which interests us here. We aim to show
that, upon assuming (\ref{problem2}), and by choosing an infinitesimal
$\alpha$ such that the commutative limit of $\alpha^{-1} x^{\pm}$ is
finite, the commutative limit of $\alpha^{-1}
x^-,\alpha^{-1}(x^0-1),\alpha^{-1}x^+$ is in fact a set of (complex)
cartesian coordinates $y^-,y^0,y^+$ of $\b{R}^3$. In addition, we shall
also sketch some basic ideas for a correspondence principle between the
deformed and the undeformed theory for finite $h$.
As a first step, we show that one
can construct objects $y^i$ and $\frac{\partial}{\partial y^i}$, having
the commutation relations of a set of classical coordinates and their
derivatives, as `functions' of $x^i, e_a, h$ and a free parameter
$\alpha$. The `functions' are in fact power series in $h$. The zero
degree term for $y^i$ is essentially fixed by (\ref{problem2}). We
stress in advance that the manipulations reported below are heuristic
and formal; these manipulations will have to be justified in the proper
representation theory framework.
The algebra of observables $\c{D}_h$ is a deformed Heisenberg algebra of
dimension 3 ($h$ is the deformation parameter). In fact, clearly the
commutation relations between the $x^i$ are such that
\[
[x^i,x^j]=O(h).
\]
Moreover, the change of generators (\ref{covder})
is independent of $\alpha$ (the dependences in $\theta^a_i$
and $\lambda_a$ cancel), invertible, and gives
\begin{eqnarray}
&&[\partial_i,x^j]=\delta^i_j +O(h), \nonumber\\
&&[\partial_i,\partial_j]=O(h).\nonumber
\end{eqnarray}
It is known (see e.g. ~\cite{pillin} and
references therein) that for any deformation $\c{D}_h$ of a Heisenberg
algebra $\c{D}$ there exists an isomorphism
$\c{D}_h\leftrightarrow\c{D}[[h]]$ of algebras over $\b{C}[[h]]$.
Applied to the present case, this implies that there exists a formal
realization of the canonical generators of $\c{D}$ in terms of
$x^i,\partial_j$, namely a set of formal power series in $h$
\begin{eqnarray}
&& y^i=f^i(x,\partial,h)= x^i+ O(h)\nonumber\\[2pt]
&& \frac{\partial}{\partial y^i},
=g_i(x,\partial,h)=\partial_i+ O(h),\nonumber
\end{eqnarray}
with coefficients equal to polynomials in $x^i,\partial_j$ and
reducing to the identity for $h=0$, such that the objects
$y^i, \frac{\partial}{\partial y^i}$ fulfill the canonical
commutation relations
\begin{equation}
[y^i,y^j]=0, \qquad
[\frac{\partial}{\partial y^i},\frac{\partial}{\partial y^j}]=0
\qquad[\frac{\partial}{\partial y^i},y^j]=\delta^i_j. \label{ccr}
\end{equation}
When $h=0$ the above transformation is the identity, so also for
$h\neq 0$ one can formally invert it to obtain a realization of the
`deformed generators' $x^i,\partial_i$ in the form of formal power
series in $h$ with coefficients equal to polynomials in
$y^i,\frac{\partial}{\partial y^i}$. (As a consequence, also $e_a$ can
be expressed in the same form; one can show the same also for
$\Lambda$). The above transformation is not unique.
It is defined up an infinitesimal inner automorphism of the algebra
\begin{equation}
x^i\rightarrow u\, x^i\, u^{-1} \qquad
\partial_i\rightarrow u\,
\partial_i\, u^{-1}, \label{innerauto}
\end{equation}
where $u=1+O(h)\in\c{D}_h[[h]]$. Since the commutation
relations (\ref{ccr}) are preserved upon
rescaling and translations of the $y^i$ by some constants
$\alpha,c^i\in\b{C}$, $y^i\rightarrow c^i+\alpha y^i$,
if we relax the condition that the transformation
be the identity at zero order in $h$
a larger family of realizations will be given by
\begin{eqnarray}
&& y^i=
\alpha^{-1}[f^i(x,\partial,h)- c^i]=
\alpha^{-1}[x^i-c^i+ O(h)]\label{puppi1} \\
&& \frac{\partial}{\partial y^i}=\alpha\, g_i(x,\partial,h)=
\alpha[\partial_i+ O(h)].
\label{puppi2}
\end{eqnarray}
Next we shall fix the parameters $\alpha, c^i$ through some basic
requirements on the commutative limit. As for
the remaining indeterminacy (\ref{innerauto}),
it should be fixed by geometrical requirements for
finite (though small) $h$, e.g. the ones suggested at the end of
the section.
In the first identification, namely (\ref{line-element}),
we have picked $c^i=0$ and $\alpha$ finite,
e.g. $\alpha=1$; consequently $y^i=\lim_{h\to 0} x^i$.
In the second identification we choose $c^{\pm}=0$ and $\alpha$ a suitable
infinitesimal in $h$ so that the limit (\ref{problem2}) is fulfilled;
consequently $y^{\pm}$ will be recovered as the limit
$\lim_{h\to0} (x^{\pm}/\alpha)$. In order that formulae
(\ref{puppi1}-\ref{puppi2})
contain a unique expansion parameter $\alpha$, instead of two
$\alpha,h$, we choose the infinitesimal as
$\alpha=O(h^{1/p})$, with some $p\in\b{N}$ to be
determined. We also choose $c^0\neq 0$
in order that $x^0$ does not
become singular in the commutative
limit. If for simplicity we normalize $c^0$
to 1, we will find $x^0_{(0)}=1$. Together with
(\ref{problem2}) and (\ref{squarelenght}), this implies
$r_{(0)}=1$.
Note that any polynomial (or power series) of the $x^i,r$ with
finite coefficients has a constant (i.e. independent of $y^i$)
commutative limit, because $\alpha\to 0$.
We shall formally identify the three $y^i$ thus obtained as a
set of (local) classical coordinates on the limit manifold, and
$\frac{\partial}{\partial y^i}$ as the derivatives with respect to
$y^i$. Thus the limit manifold will have dimension three.
In the commutative limit
\begin{equation}
r\rightarrow 1,\qquad (\alpha r)^{-1} dx^i\rightarrow dy^i,
\end{equation}
and the line element (\ref{line-element}) becomes
\begin{equation}
ds^2 \to \sum_idy^i dy^{-i}. \label{limit-line-element}
\end{equation}
This yields a vanishing Riemannian curvature and is consistent with
the vanishing (for all $q$) linear curvature computed in
Sect. \ref{metrics}. Assuming a trivial topology, the limit
Riemannian manifold will be thus Euclidean space, and $y^i$ will be
(complex) cartesian coordinates on $\b{R}^3$.
{From} $r_{(0)}=x_{(0)}=\Lambda_{(0)}=1$ one also easily finds
$$
\lim\limits_{\alpha\rightarrow 0}
\alpha^{-1}q\Lambda e^i_a=\delta^i_a, \qquad
\lim\limits_{\alpha\rightarrow 0}
\alpha\Lambda^{-1} \theta^a_i=\delta^a_i,
$$
whence
\begin{eqnarray}
&&\lim\limits_{\alpha\rightarrow 0} e_a
=\lim\limits_{\alpha\rightarrow 0}q\Lambda e^i_a\partial_i=
\lim\limits_{\alpha\rightarrow 0}\alpha^{-1}q\Lambda e^i_a
\frac{\partial}{\partial y^i}=\frac{\partial}{\partial y^a},\\[4pt]
&&\lim\limits_{\alpha\rightarrow 0} \theta^a
=\lim\limits_{\alpha\rightarrow 0}\Lambda^{-1} \theta^a_i dx^i=
\lim\limits_{\alpha\rightarrow 0}\Lambda^{-1} \theta^a_i
\alpha\alpha^{-1}dx^i=d y^a. \label{dopo}
\end{eqnarray}
This is consistent with the metric, stating that the set $\{\theta^a\}$
(and, in the dual formulation, the set $\{e_a\}$) is orthonormal. Let
us determine possible values for $\alpha$. Plugging (\ref{puppi1}) in
the commutation relations (\ref{xxcr3}), it is easy to see that at
lowest order in $\alpha$ the third relation can be satisfied only if
$p\ge 3$. We shall later exhibit a transformation
(\ref{puppi1}-\ref{puppi2}) based indeed on $\alpha=O(h^{1\over 3})$.
To understand the physical meaning of $\alpha$ one would have to
consider the $*$-represen\-tations of $\c{D}$. We shall see elsewhere
that it is related to the Planck lenght. Moreover in the commutative
limit $y^0$ will `inherit' from the $*$-representation studied
previouly~\cite{fioijmpa} a spectrum dense in the whole $\b{R}$, in
agreement with the fact that $y^0$ becomes a real cartesian coordinate.
Similarly, $y_{\bot}^2:=y^+y^-$ `inherits' a spectrum dense in the whole
$\b{R}^+$, in agreement with the fact that it becomes the square
distance from the $y^0$ axis.
So far we have left the realization (\ref{puppi1}-\ref{puppi2}) still
largely undetermined by the inner automorphism freedom
(\ref{innerauto}). We may restrict this freedom by imposing further
requirements on $x^i,e_a$, so that the physical meaning of the latter
at the
representation theoretic level become more manifest; for instance, since
they commute and are real, we may require that at all orders in $\alpha$
$x^0,r$ depend only on $\vec{y}$, so that their meaning of position
observables be manifest. This will be discussed in detail
elsewhere. Once determined, the realization (\ref{puppi1}-\ref{puppi2})
will be an essential ingredient of the {\it correspondence principle}
between the deformed and undeformed theory
for finite $h$. As an example, a formal realization of the
algebra fulfilling all previous requirements and such that the classical
involution $(y^i)^*=y^{-i}$ realizes also the involution $*$ of $\c{A}$
is
\begin{eqnarray}
&& x^0= e^{\alpha y^0-\frac{\alpha^2}2} \label{special1} \\
&& \Lambda= e^{-\alpha^2\frac{\partial}{\partial y^0}}\\
&& x^+=\sqrt{\frac{e^{\alpha^2 y^+y^-}-1}{q^2(q+1)}}
e^{\alpha y^0 +i\varphi} e^{-\alpha^2\frac{\partial}{\partial y^0}
+ 2\alpha\frac{\partial}{\partial (y^+ y^-)}} \\
&& x^- =(x^+)^*\sqrt{q}
\end{eqnarray}
with $e^{\alpha^3}=q$ and $e^{2i\varphi}=y^+/y^-$.
Its derivation will be given elsewhere.
As a consequence
\begin{equation}
r^2=e^{-2\alpha^3+\alpha^2y^+y^-+2\alpha y^0} \label{special2}
\qquad
(\frac{x^0}{r})^2=e^{\alpha^3-\alpha^2y^+y^-}.
\end{equation}
Thus, the surfaces $r=const$ are paraboloids with axis $y^0$, the
surfaces $x^0=const$ are planes perpendicular to $y^0$, the surfaces
$x^0/r=const$ are cylinders with axis $y^0$, and the lines $x^0=const$,
$r=const$ are circles perpendicular to and with center on the axis
$y^0$.
The exponential relation between $x^0$ and $y^0$ is analogous to the one
found \cite{CerHinMadWes98} for a 1-dimensional $q$-deformed model. From
the spectrum of $x^0$ \cite{fioijmpa} $y^0$ will inherit equidistant
eigenvalues of both signs, suggesting a `uniform' structure of space.
This will solve the problems \cite{fioijmpa} arising from the physical
identification of $x^i$ as cartesian coordinates.
Summing up, the resulting geometry is now flat
$\b{R}^3$, instead of the $S^2\times \b{R}$ found in the first
identification (the one with finite $\alpha$). $x^0, r$
may be adopted together with $\varphi$
as global curvilinear coordinates, but
only for finite $\alpha\approx h^{1/p}$,
when they are to be identified with the coordinates (\ref{special1}) and
(\ref{special2}); for $\alpha\to 0$ they become singular, and coordinates
$y^i$ may be used at their place.
\sect{The involution and the real calculus} \label{involu}
Formula (\ref{barstern}) gives an involution of
$\Omega^1(\c{A})\oplus\bar\Omega^1(\c{A})$. Unfortunately, the latter
has rank 6 as a $\c{A}$-bimodule, instead of 3, and is not generated
from $\c{A}$ through the action of a real differential.
The problem of constructing a differential calculus of rank 3 closed
under involution has
been considered by Ogievetsky and Zumino~\cite{olezu}, who expressed
$\bar\xi^i$ as functions not only of $x^i$, $\xi^i$, but also of
suitably $q$-deformed derivations $\partial_i$. A much more economical
construction is the following. Note that one can define an algebra
isomorphism $\varphi:\Omega^*(\c{A})\rightarrow \bar\Omega^*(\c{A})$
acting as the identity on $\c{A}$ and on the 1-forms through
\begin{equation}
\varphi(\theta^a)=\bar\theta^a,
\end{equation}
since the commutation relations among the $\theta^a$'s
and the $\bar\theta^a$'s are the same. Hence
$\star= *\circ \varphi$ is an involution of $\Omega^*(\c{A})$
acting as
\begin{equation}
(\theta^a)^{\star} = \theta^b g_{ba}.
\end{equation}
This implies that $(\xi^i)^{\star}$ can be expressed as combinations
of $\xi^j$ with coefficients in (the extended)
$\c{A}$. One can easily check that $\star$ has the
correct classical limit provided the commutative
limit fulfills (\ref{problem2}).
Nevertheless, since $\varphi(d)\neq \bar d$,
of course $d$ is not real under the involution.
It would be natural to define a real exterior derivative by
\[
d_r := (d + \bar d)
\]
and a rank 3 $\c{A}$-bimodule
$\Omega_r^1(\c{A})\subset\Omega^1(\c{A})\oplus\bar\Omega^1(\c{A})$
closed under $*$ from the generators
\[
\xi^i_r=\xi^i+\bar\xi^i.
\]
But this is impossible since
$\xi^i_r$'s do not close commutation relations with
the $x^j$'s, as evident from (\ref{xxicr}), (\ref{xxicr'}).
To generate the exterior algebra through a real differential
we need to double either the number of the coordinates or
the number of the 1-forms.
In general, consider an algebra $\c{A}$ with involution over which
there are two differential calculi $(\Omega^*(\c{A}), d)$ and
$(\bar\Omega^*(\c{A}), \bar d)$ neither of which is necessarily
real. Consider the product algebra $\tilde \c{A} = \c{A} \times \c{A}$
and over $\tilde \c{A}$ the differential calculus
\begin{equation}
\tilde \Omega^*(\tilde \c{A}) =
\Omega^*(\c{A}) \times \bar\Omega^*(\c{A}). \label{prod-cal}
\end{equation}
It has a natural differential given by $\tilde d = (d, \bar d)$. The
embedding
$$
\c{A} \hookrightarrow \tilde \c{A}
$$
given by $f \mapsto (f, f)$ is well defined and compatible with the
involution
\begin{equation}
(f,g)^* = (g^*,f^*) \label{alg-invol}
\end{equation}
on $\tilde \c{A}$. Suppose there exists a frame $\theta^a$ for
$\Omega^*(\c{A})$ and a frame $\bar \theta^a$ for $\bar
\Omega^*(\c{A})$ and a relation of the form (\ref{bar-theta}) between
them which extends the involution (\ref{alg-invol}). We can define
a real module $\Omega_r^1(\c{A})$ to be the
$\c{A}$-bimodule generated by the frame
$$
\theta_r^a = (\theta^a, \bar\theta^b g_{ba}). \label{realframe}
$$
This does not necessarily contain however the image of
$d_r = (d, \bar d)$. In fact in the case $\c{A} = \b{R}_q^3$ if we
require this to be the case then $\Omega_r^1(\c{A})$ is an
$\tilde \c{A}$-bimodule and $\Omega_r^*(\c{A}) = \tilde\Omega^*(\c{A})$.
Alternatively one can consider a larger `function' algebra
containing two copies of $\c{A}$, whose generators we denote by
$x^i,\bar x^i$, with cross commutation relations of the form
\begin{equation}
x^i\bar x^j = \hat R^{\pm 1}{}{ij}_{hk} \bar x^h x^k.
\end{equation}
We can define an involution by setting
\begin{equation}
(x^i)^*=\bar x^j g_{ji} \label{invol}
\end{equation}
and introduce an $SO_q(3)$-covariant differential calculus
$\Omega(\tilde \c{A})$ with a real differential $d_r$ by setting
\begin{equation}
d_r x^i = \xi^i,\qquad d_r\bar x^i = \bar\xi^i,
\end{equation}
with the commutation relations
\begin{equation}
\begin{array}{ll}
x^i\xi^j = q\hat R^{ij}_{hk}\xi^hx^k,
&\bar x^i\bar \xi^j = q^{-1}\hat R^{-1}{}^{ij}_{hk}\bar\xi^h\bar x^k,\\[6pt]
x^i\bar\xi^j = q^{-1}\hat R^{\pm 1}{}^{ij}_{hk}\bar \xi^hx^k,
&\bar x^i\xi^j = q\hat R^{\mp 1}{}^{ij}_{hk} \xi^h\bar x^k.
\end{array}
\end{equation}
(In the last two Equations we have picked a specific normalization).
These relations are compatible with (\ref{invol}) and $(d_r f)^*=d_r f^*$.
The commutation relations among the $\xi^i$ and
the $\bar\xi^i$ follow and are the same relations (\ref{xixicr}),
(\ref{barxibarxicr}) found in Section \ref{calculi}
for the analogous elements in the conjugate calculi,
whereas the cross relations are
\begin{equation}
\xi^i\bar\xi^j = -q^{-1}\hat R^{\pm 1}{}^{ij}_{hk}\bar \xi^h\xi^k.
\label{barxixicr}
\end{equation}
Thus one ends up with a real calculus with 6 independent coordinates
and 6 independent 1-forms.
But one can check that in this scheme there is no frame.
Finally, if we just let both the conjugate differentials $d,\bar d$
act on one copy of $\c{A}$ we generate a $\c{A}$-bimodule
of rank 6 closed under the involution $*$. An
enlarged exterior algebra is generated by $\xi^i,\bar\xi^i$ and
the generators of $\c{A}$; we need to add to the commutation
relations (\ref{xxicr}), (\ref{xixicr}), (\ref{xxicr'}), (\ref{barxibarxicr})
compatible commutation relations between $\xi^i$ and $\bar\xi^j$.
It is easy to show that the latter can only be of the form
$\xi^i\bar\xi^j =\gamma \hat R^ {-1}{}^{ij}_{hk}\bar\xi^h\xi^k$.
This is compatible also with this algebra being a $\Omega^*(\c{A})$
and a $\bar\Omega^*(\c{A})$ bimodule.
If we extend $d,\bar d$ by requiring that
\begin{equation}
d\bar\xi^i = 0, \qquad\qquad \bar d\xi^i=0, \label{barnilpo'}
\end{equation}
$\gamma$ will be fixed to be $-q^{-1}$, and we find again
(\ref{barxixicr}).
It is immediate to show that $\theta$, $\bar\theta$ are the
Dirac operators of $d,\bar d$ also in the enlarged algebra.
A direct and lengthy calculation\footnote{We have performed it using
the program for symbolical computations REDUCE.} shows that the
$\theta^a_i$, $\bar\theta^a_j$ introduced in Sect.
\ref{calculi} satisfy the commutation relations
\begin{equation}
\hat R^{ab}_{cd}\, \bar\theta^d_j\theta^c_i=
\theta^b_l\bar\theta^a_k\, \hat R^{kl}_{ij}. \label{basic'}
\end{equation}
As a consequence of (\ref{barxixicr}), (\ref{basic'}),
in the enlarged exterior algebra we find
\begin{equation}
\theta^a\bar\theta^b =-q\hat R^{-1}{}^{ab}_{cd}\, \bar\theta^c\theta^d .
\label{th-bthcr}
\end{equation}
We can also readily extend the generalized permutation $\sigma$
and the metric $g$ of Section \ref{metrics}
to $\Omega^1(\c{A})\oplus\bar\Omega^1(\c{A})$.
For simplicity we consider an Ansatz where $\sigma$
does not mix the $\theta^a$'s and
the $\bar\theta^a$'s:
\begin{eqnarray}
\sigma (\theta^a\otimes \bar\theta^b) & = &
V^{ab}{}_{cd}\,\bar\theta^c\otimes \theta^d, \label{bgigio2}\\
\sigma (\bar\theta^a\otimes \theta^b) & = &
\bar V^{ab}{}_{cd}\,\theta^c\otimes \bar\theta^d. \nonumber
\end{eqnarray}
Imposing conditions (\ref{2.2.6}), (\ref{th-bthcr}) we find
the numerical matrices $V,\bar V$:
\begin{equation}
V=q\hat R^{-1}\qquad \qquad\bar V=q^{-1}\hat R.
\end{equation}
Together with (\ref{gigio2}), (\ref{bbgigio2}), these relations
completely define $\sigma$. If we set
\begin{eqnarray}
&& g(\theta^a\otimes\bar\theta^b) = g^{a b}, \\
&& g(\bar\theta^a\otimes\theta^b) = \bar g^{a b},
\end{eqnarray}
the metric compatibility condition (\ref{met-comp})
on $\Omega^1(\c{A})\otimes\bar\Omega^1(\c{A})$ and
$\bar\Omega^1(\c{A})\otimes\Omega^1(\c{A})$ reads
\begin{equation}
\begin{array}{l}
\bar V {}^{ae}{}_{df} g^{f g} S {}^{cb}{}_{eg}
= g^{a c} \delta^b_d, \\
\bar V {}^{ae}{}_{df} \bar g^{f g} S {}^{cb}{}_{eg}
= \bar g^{ac} \delta^b_d.
\end{array}
\label{brrr}
\end{equation}
As well as its barred counterpart,
it can be satisfied only if the numerical matrices
$g^{ab}$, $\bar g^{ab}$ are proportional to the
matrix (\ref{metric}) and we restrict
(\ref{double}) to the choice
\begin{equation}
S=q\hat R, \qquad\qquad \bar S= (q\hat R)^{-1}. \label{unique}
\end{equation}
Note that in (\ref{brrr}) no conformal factor appears.
The connections defined by (\ref{conn}),
(\ref{bconn}) are now automatically extended to
the exterior algebra of rank 6, and it is easy to check that also
on this larger domain their linear curvatures are zero. Moreover, they
are still related by (\ref{Dstar}). We conclude by noting that
\[
ds^2:=\theta^a\otimes\bar\theta^b g_{ab} + \mbox{h. c.}
\]
is real and annihilated by the action of both $D,\bar D$
(in other words it is ``invariant under parallel transport'').
Therefore it
might be considered as a candidate for the square displacement.
\sect{Conclusions}
The fundamental open problem of the noncommutative theory of gravity
concerns of course the relation it might have to a future quantum theory
of gravity either directly or via the theory of strings and membranes.
But there are more immediate technical problems which have not received
a satisfactory answer. The most important ones concern the definition of
the curvature. It is not certain that the ordinary definition of
curvature taken directly from differential geometry is the quantity
which is most useful in the noncommutative theory. The main interest of
curvature in the case of a smooth manifold definition of space-time is
the fact that it is local. We have defined Riemann curvature as a map
Curv which takes $\Omega^1(\c{C}(V))$ into
$\Omega^2(\c{C}(V)) \otimes_{\c{C}(V)} \Omega^1(\c{C}(V))$.
If $\xi \in \Omega^1(\c{C}(V))$ then $\mbox{Curv}(\xi)$ at a given
point depends only on the value of $\xi$ at that point. This can be
expressed as a bilinearity condition; the above map is a
$\c{C}(V)$-bimodule map. If $f \in \c{C}(V)$ then
\begin{equation}
f \mbox{Curv}(\xi) = \mbox{Curv}(f\xi), \qquad
\mbox{Curv}(\xi f) = \mbox{Curv}(\xi) f.
\end{equation}
One would like to insure also that one is dealing with the
noncommutative version of a real manifold. This reality condition
exchanges left and right linearity and adds weight to the argument that
a bilinear curvature map is necessary. In the noncommutative case
bilinearity is therefore the natural and only possible expression of
locality. It has not yet been possible to enforce it in a satisfactory
manner~\cite{DubMadMasMou96}.
We have argued here in favour of one possible extension to
noncommutative geometry of the definition of a linear connection which
relies essentially on the above expression of reality and locality.
Alternative extensions have however been given. In fact one definition
has been proposed~\cite{CunQui95} which is indeed local in the sense we
have defined locality but which is valid only in the noncommutative case
and becomes singular in the commutative limit. The main difference with
alternative definitions and the one we use lies in the fact that the
locality condition is relaxed and the Leibniz rule is applied only from
one side; the module of 1-forms is considered only as a left (or right)
module. This means in fact that gravity is considered as a Yang-Mills
field with the general linear group or some $q$-deformation of it, as
structure group. One of the authors has given elsewhere~\cite{Mad97} a
partial list of these alternative proposals. We mention here only three
which are especially relevant to the example of quantum Euclidean space.
Ref. \cite{HeckSchmue95} is the closest to ours.
A quantum group ($SL_q(N),O_q(N)$ or $Sp_q(N)$) function algebra,
rather than a quantum space function algebra, is
chosen as algebra $\c{A}$. General definitions
of a metric, covariant derivative, torsion, curvature, metric-compatible
covariant derivative on $\c{A}$ are given; in particular, the quantum
group invariant/bicovariant ones are determined. The definitions
differ from ours essentially in that
$\c{A}$-bilinearity is weakened to a left- (or right-) linearity.
The existence of frames is not investigated. In Ref.~\cite{Cas95} a
$q$-Poincar\'e group function algebra is chosen as an algebra $\c{A}$; the
definition of differential calculus, linear connection, curvature, mimic
the group-geometric ones on the ordinary Poincar\'e group manifold.
The underlying Hopf algebra is triangular, what essentially means
$\hat R^2=1$ (in this respect the situation is simpler
than the one considered in the present work). The existence of metric,
torsion and frames is not investigated. The $q$-Minkowski space,
its differential calculus, linear connection, curvature, are meant to be
derived by projecting out the $q$-Lorentz group.
Other
authors~\cite{Dur98, maj97} have abandoned the definition of a
connection as a covariant derivative and proposed a $q$-deformed version
of a principle bundle to define a noncommutative `Riemannian geometry',
using for example the universal calculus. It is interesting to note
that according to the definition we use here the unique linear
connection associated with any universal calculus is necessarily
trivial~\cite{DubMadMasMou95}; the covariant derivative coincides with
the ordinary exterior derivative.
Finally, we would like to compare the results found here for the quantum
Euclidean space with the results found in~\cite{DimMad96} for the Manin
quantum plane. We start by noting that all commutation relations in both
cases are homogeneous separately in $x^i$ and $\xi^i$. A Dirac operator
$\theta$ (\ref{fund}) for a quantum group covariant differential
calculus must be of degree 1 in $\xi^i$, degree -1 in $x^j$ and quantum
group invariant. There can be no such object for the Manin plane
because, to mimic the construction (\ref{deftheta}), we would have to
use the isotropic tensor $\varepsilon_{ij}$ of $SL_q(2)$, the
$q$-deformed $\varepsilon$-tensor in the place of the isotropic tensor
$g_{ij}$ of $SO_q(3)$, but the analog $x^ix^j\varepsilon_{ij}$ of $r^2$
vanishes, by the $q$-commutation relations of the Manin plane. The
absence of such a $\theta$ prevents the construction of a torsion-free
linear connection by means of formula (\ref{2.2.14}) which, by means of
Equation~(\ref{2.2.19}), could eventually yield a bilinear zero
curvature. In Reference~\cite{DimMad96} a Stehbein was constructed
using the inverses of both Manin plane coordinates and no dilatator,
whereas we have needed here the inverse of one coordinate $x^0$ and the
introduction of the dilatator. Using the Stehbein, also a metric has
been found. However, it is not a quantum group covariant metric, as
found here. One could define a $SL_q(2)$-covariant metric $g_0$ on the
Manin plane by introducing a dilatator $\Lambda$, in the same way as in
Equation~(\ref{lambda}), (\ref{zero}) and by setting $g_0(\xi^i\otimes
\xi^j)=\Lambda^{-3}\varepsilon^{ij}$ (See Equation (\ref{final}).
This would be a symplectic form in the limit $q=1$.
An analog of equation (\ref{brrr}), but in the $\xi^i$ basis, would hold.
\section*{Acknowledgment}
One of the authors (JM) would like to thank the J. Wess for his
hospitality and the Max-Planck-Institut f\"ur Physik in M\"unchen for
financial support.
|
\section*{Figure Captions}
\subsection*{Figure 1}
\noindent (a) The Border Collision Circuit.
\newline \noindent (b) $i \leq i_{crit}$.
\newline \noindent (c) $i > i_{crit}$.
\subsection*{Figure 2}
\noindent (a) and (b) Bifurcation diagrams for the two-dimensional time-1 map of
the circuit
depicted in Fig. 1
with $Q_{bias} = -1.0$, $\rho = 0.10742$, $\Omega = 1.0642$ and $\Gamma =
1.2040$. In creating the bifurcation diagrams (a) and (b) small noise was
inserted
leading to two different realizations.
\noindent (c) and (d) Bifurcation diagrams for the equivalent canonical form,
Eq. (3), near criticality
with $\tau_< = 0.7$, $\tau_> = -2.0$ and $ d = 0.3$. In creating (c) and (d),
small noise was inserted leading to two different realizations.
\subsection*{Figure 3}
\noindent Two-dimensional period plot in parameter space for the canonical form,
Eq. (3),
with $0.5 < \tau_< < 0.9$, $-2.2 < \tau_> < -1.8$, $d = 0.3$, $\mu = 1$, and
initial conditions $(x_0, y_0) = (7.0, 2.0)$ and $(x_0, y_0) = (1.2, 0.0)$.
\subsection*{Figure 4}
\noindent Probability $b(\mu>\mu_2)$, plotted logarithmically,
as a function of $1/v$, based on 1,000,000 experiments for every value of $1/v$.
Solid lines and plus symbols represent {\em square} noise, while dashed lines
and asterisks represent {\em circular} noise.
\end{document}.
|
\section{Introduction}
The structure of the impurity band of semiconductors has been widely studied
during the last two decades both theoretically and experimentally
(See monograph\cite{mon} ). In the early theoretical studies the spin
structure of the impurity band was completely ignored. Recent experiments
suggest that the spin structure is very important for the variable range
hopping conductivity, especially near the metal-non-metal
transition\cite{shlim,sar,sim,ish,shl,krav}.
The exchange interaction must be the main mechanism of the spin-spin
interaction in the impurity band.
It appears as a result of the overlap
of the wave functions of different states. The scale of this interaction
decreases exponentially with increasing distance between the states. Thus,
this interaction becomes the most important one near the metal-non-metal
transition. In this region the scale of the interaction is of the order of the
binding energy of a single impurity.
In this paper we study the spin structure of the impurity band created by
Coulomb impurities in both two and three dimensional cases in the limit of low
density of impurities.
In the two-dimensional case the impurities may be located
either outside or inside the plane of electron gas. We assume that all
impurities are occupied by one electron. In this case we can consider the
coordinates of the occupied centers as random variables without any
correlations.
Our study of the spin structure is based upon the Heisenberg Hamiltonian
which takes into account the spin-spin interaction of the electrons localized
at different randomly distributed impurities.
\begin{equation}
H=\sum_{i\neq k}J_{ik}(I/2+{\bf s_i\cdot s_k}),
\label{ham}
\end{equation}
where ${\bf s}$ is spin 1/2 operator, $I$ is a unit matrix, and $i,k$ denote
different impurity atoms. The sum is over all pairs of impurities. The density of impurities
is assumed to be small.
This problem has a long story\cite{rosso,bhattc,lee,andres,bhatt,thomas,fisher,hirsch}.
The following important results have been obtained:\\
1. The ground state of the system consists of local singlets.\\
2. Rosso\cite{rosso}, Thomas and Rosso\cite{thomas}, Andres {\it et al.}\cite{andres} used
different selfoconsistent approaches to get the distribution of the
excitation energies of the singlet-triplet transitions.\\
3. Bhatt and Lee\cite{bhatt,lee} worked out a computational scaling approach which is exact
at small density of impurities. They have also mentioned a drastic difference
between the Heisenberg and
Ising models.\\
4. As far as we know all previous authors used simplified versions for the function $J(R)$.
Our paper pursues the following goals:\\
1. We analyze the existing methods to find the
distribution of excitation energies and propose a new modification for
one-, two- and three-dimensional cases.
Our approach is exact at low densities and it allows to get an
approximate analytical expression for this distribution.\\
2.
To get an estimate for the energy of spin ordering one needs a reliable
calculation of the coefficients $J_{ik}$ which are defined here as
$1/2$ of the singlet-triplet splitting for the two
states corresponding to the impurities $i$ and $k$. We have performed these
calculations for a pair of the Coulomb centers at a distance
$R_{ik}$. The result of the computations is a function $J(R)$ which is
reliable at all distances from zero to infinity.
The paper is organized as follows.
In Section 2 we consider Hamiltonian Eq.(\ref{ham}) in the case of small
impurity density. We show that the ground state mostly
consists of independent singlets. We show that the problem of finding these
singlets can be reduced to a non-trivial geometric problem. We solve it in
a mean field approximation and by computer modeling. The solution of this
problem gives the distribution function $F(E)$ for the energies of the
singlet-triplet transitions for a given function $J(R)$.
In Section 3 we calculate $J(R)$ and its inverse function $R(J)$.
For the $3D$-case we present interpolated formulae which are based upon the
results of well-known calculations for two hydrogen atoms. These calculations
include analytical results for large distances\cite{Gor},
numerical calculations at intermediate distances, and known results for the
singlet-triplet splitting
of the He atom. Similar interpolated formulae are presented for the $2D$-case.
They are based upon our original calculations given in the appendices.
We present an analytical expression for $J(R)$
at large distances, a numerical result for $J(a_B)$,
and variational calculations for a ``two-dimensional He atom''.
In the Conclusion we discuss the distribution function of singlet-triplet
splittings $F(\ln\varepsilon)$, where $\varepsilon=J/J(0)$,
and the density of free spins $\rho (T)$ at finite
temperature $T$. These two functions are the final results of our paper.
\section{Ground state and excited states of the Heisenberg Hamiltonian in the
impurity band}
\subsection{The structure of the ground state}
We find the ground state and excited states of the Hamiltonian Eq.(\ref{ham})
using the following properties of $J_{ik}$.
\begin{itemize}
\item All $J_{ik}>0$, which means an antiferromagnetic interaction.
\item The density of impurities $n$ is assumed to be small, so that the
average distance between them is larger than the characteristic length of the
exponential decay of $J_{ik}$. This means
there is
a very large dispersion of $J_{ik}$.
In fact we shall assume that if
$J_{ik}>J_{lm}$, then $J_{ik}\gg J_{lm}$. Thus, we ignore the cases when the
distance $R_{ik}$ is very close to the distance $R_{lm}$, assuming that these two pairs
are not very far from each other.
\end{itemize}
To understand the physics of the problem it is very helpful to consider the
Hamiltonian (\ref{ham}) with four impurities only (Fig. 1a). From a general
principle one can conclude\cite{lan} that the energy spectrum consists of
six levels, one level with spin $S=2$, three levels with $S=1$, and two levels
with $S=0$. Let us assume that $J_{12}$ is much larger than all other
$J_{ik}$ in this problem. Then the ground state wave function
describes two singlets at sites (1,2) and (3,4). It is easy to write the
energy of the ground state and the first excited state assuming
\begin{equation}
J^{\prime}=\sum^{\prime}J_{ik}\ll J_{12},
\label{in}
\end{equation}
where the sum includes all $J_{ik}$ except $J_{12}$ and $J_{34}$.
The ground state energy $E_0$ and the energy of the first excited state $E_1$ are given by
the equations
\begin{equation}
E_0=-J_{12}-J_{34}+J^{\prime}/2;\quad E_1=-J_{12}+J_{34}+J^{\prime}/2.
\label{en}
\end{equation}
The physical meaning of Eq. (\ref{en}) is simple. Two singlets (1,2)
and (3,4) do not
interact with each other if condition (\ref{in}) is fulfilled. The $J^{\prime}/2$
terms come from the first term in the Hamiltonian (\ref{ham}).
In this approximation the excitation energy is $E=2J_{34}$. The ground state
has a total spin $S=0$ while the first excited state has $S=1$.
Bhatt and Lee\cite{bhatt,lee} take into account the next approximation for the
excitation energy
\begin{equation}
E=2J_{34}+(J_{13}-J_{23})(J_{24}-J_{14})/J_{12}.
\label{bh}
\end{equation}
Since $J_{12}$ is the largest term, the second term should be small. It looks
like it can change the ground state from singlet to triplet if $J_{34}$ is
unusually small. However, such configurations are extremely rare. It happens
because in the case of small $J_{34}$ one should consider Fig.1b with 6 spins
rather than Fig.1a. Indeed, very small $J_{34}$ means a long distance between
impurities 3 and 4.
It is more likely that in this situation some other strong singlet (5,6) is
the nearest neighbor of the impurity 4 rather than the singlet (1,2).
In this 6-spin system we have 2 strongly connected groups
of spins, namely 1,2,3 and 4,5,6. Assume that $J_{12}$ and $J_{56}$ provide
the strongest bonds in each group. Suppose there is no interaction between the
groups. Then, the ground state in each of them is a degenerate
doublet. Altogether the system is 4-fold degenerate. If one takes into
account $J_{34}$ the degeneracy of the ground state will be lifted. One gets
a singlet and a triplet with the energy splitting $2J_{34}$. On the other hand,
the general 6-spin problem can be solved assuming that both $J_{12}$ and
$J_{56}$ are infinite. In this approximation one gets the same result:
the ground state is a singlet and the excitation energy $E=2J_{34}$. It
follows that the other bonds connecting the two groups, like $J_{35}$ may
contribute to the excitation energy only in the second order of perturbation
theory. This contribution will contain a small dimensionless coefficient like
$J_{35}/J_{12}$ and it may be neglected. Thus, it is
not necessary to take into account the renormalization of the weak bonds due to
their strong neighbors in the limit of small density. Bhatt and Lee also
mention\cite{lee} that their
computations show the triplet ground state in very rare cases.
Thus, we assume that the ground state energy of any even number
of impurities has $S=0$ and the system can be split into localized singlets.
To find the pairs of impurities which form the singlet in the ground state we
propose the following geometric problem.\\
1. For every impurity in the system find its nearest neighbor.\\
2. Take the pair with the smallest distance. Generally, the nearest neighbor
of a site A does not have site A as its nearest neighbor. But for the
closest pair this is the case.\\
3. This closest pair forms a singlet with the largest binding energy. To find
all other singlets remove both sites of the first pair. Go
to point 1 and continue until all the singlets will be found.\\
The same geometric problem has been proposed by Thomas and Rosso\cite{thomas} for
three-dimensional case.
Assuming that all neighboring $J_{ik}$ are very different one can write
the total energy of the lowest state in the form
\begin{equation}
E_0=-\sum_s J_{ik}+{1\over 2}\sum_{other} J_{ik},
\label{gr}
\end{equation}
where the first sum includes all pairs which form singlets and the second one
includes all other pairs.
One can prove that the distribution of singlets, obtained as a solution of the
problem above, gives the minimum of total energy. Suppose, for
example, that the solution prescribes the configuration of singlets
(1,2), (3,4) and (5,6), for impurities with the numbers from 1 to 6. One can
show that any other location of singlets at the same impurities, like (1,3),
(2,5) and (4,6), has larger energy.
We mention first that the contribution to the energy from all other
impurities like 7,8... is the same at all configurations of singlets of six
chosen impurities. Suppose now that $J_{12}\gg J_{34},J_{56}$. Then
all other $J_{ik}$ connecting the six impurities
are also less than $J_{12}$. Indeed, if one of them were larger, it would be used to
form a singlet instead of $J_{12}$. Thus, any rearrangement of the pairs
within 6 impurities that destroys singlet (1,2) increases the total energy.
In the same way one can show that rearrangement of singlets in the system of
four impurities 3,4,5,6 also increases the total energy. The same
consideration can be done for any even number of impurities. Thus, the solution
of the above geometric problem gives the ground state of the system.
\subsection{Solution of geometric problem and distribution function of excitation
energies}
We start with the simplest mean field approximation.
Suppose we are at the stage where all pairs with distance less than $R$ are
removed and we want to find the residual impurity density $n(R)$. The crucial point of the mean
field approximation is that we neglect correlations in the positions of the remaining impurities
except that they cannot be closer to one another than $R$.
We start with the two-dimensional case. Let us draw a circle around each
impurity with the radius $R$. There will be no other impurities inside the
circles. Now increase the radii from $R$ to $R+dR$ and calculate how many
impurities occur in the rings between $R$ and
$R+dR$. The total number of these impurities gives the decrease of $N(R)$, where
$N(R)=S n(R)$ and $S$ is the total area of the system.
Thus, one gets the equation
\begin{equation}
dN(R)=-N(R)2\pi R \tilde{n}(R)dR.
\end{equation}
Here $\tilde{n}(R)$ is the density of the impurities outside the circles.
It is slightly larger than $n(R)$ (see below), but in the simplest mean field approximation
we ignore this difference.
It is convenient to introduce the dimensionless coordinate $X=\sqrt{\pi n_0}R$
and the normalized number of particles (or density) $\rho(X)=n/n_0\equiv N/N_0$.
Here $n_0=n(0)$ is the initial concentration of particles. The differential
equation for $\rho(X)$ at $\tilde{n}=n$ has the form
\begin{equation}
d\rho=-2X \rho^2dX.
\end{equation}
The solution of this equation with the condition $\rho(0)=1$ is
\begin{equation}
\rho_2(X)={1\over {1+X^2}},
\label{n}
\end{equation}
where $\rho_2(X)$ is the two-dimensional density.
Similar calculations for the three- and one-dimensional cases give
\begin{equation}
\rho_d(X)={1\over {1+X^d}}.
\label{n3}
\end{equation}
Here
$X=(4\pi/3n_0)^{1/3}R$ at $d=3$ and $X=2n_0 R$ at $d=1$. This distribution has been obtained
by Rosso\cite{rosso} for $d=3$. One can show that at small $X$ the above results are exact,
including
$X^d$-corrections. Bhatt\cite{bhattc} has pointed out that it is not exact at large $X$.
We believe that the exact distribution has a following form at large $X$
\begin{equation}
\rho_d(X)={1\over b_d X^d},
\label{b}
\end{equation}
where the coefficient $b_d \neq 1$ and it depends on the dimensionality of space $d$.
It follows from Eq.(\ref{b}) that the average density $n(R)$ is
independent of $n_0$ at large values of $R$
and it is of the order of $R^{-d}$. This is because the average distance between impurities
cannot be smaller than $R$ by definition, and there are no reasons for it to be
substantially larger than $R$. That is why we believe that Eq.(\ref{b}) is exact at large $X$.
Our computer modeling confirms this point and it gives us the values $b_d$.
We propose an improved mean field approach which takes into account the fact that
the density
$\tilde{n}$ outside the circles is slightly
larger than the average density $n(R)$, because there are no impurities
inside the circles. For example, at $d=2$, one gets
\begin{equation}
\tilde{n}=\frac{N}{S-N\pi R^2\alpha},
\label{nimpr}
\end{equation}
where $N\pi R^2\alpha$ is the excluded area inside $N$ circles.
We have introduced a free parameter $\alpha<1$, which takes into account
the overlap of the circles.
Its value can be extracted from comparison with numerical computations.
Eq. (\ref{nimpr}) can be generalized for any $d$ to get a differential equation in
$\rho_d$
\begin{equation}
\frac{d\rho_d}{dX^d}=-\frac{\rho_d^2}{1-\alpha_d\rho_d X^d}.
\end{equation}
The solution is given by the following transcendental equation:
\begin{equation}
X^d=\frac{1-\rho_d^{b_d}}{b_d\rho_d}
\label{corr}
\end{equation}
with $b_d=\alpha_d+1$.
It is worth mentioning that if we would neglect the ``circles''
overlapping ($b_d\equiv 2$),
then the solution of Eq. (\ref{corr}) is
\begin{equation}
\rho_d(X)=\frac{1}{X^d+\sqrt{1+X^{2d}}},
\label{1stappr}
\end{equation}
which is an underestimate for large distances.
In the general case, for $1< b_d<2$,
the analytical solution of the transcendental equation (\ref{corr}) can be obtained only
for large and small values of $X$.
\begin{equation}
\rho_d\approx\left\{
\begin{array}{ll}
1-X^d +\frac{3-b_d}{2}X^{2d}+\cdots, & X\ll 1\\
\frac{1}{b_d X^d}\left(1-\left(\frac{1}{b_dX^d}\right)^{b_d}+\cdots
\right), & X\gg 1
\end{array}\right.
\end{equation}
We performed computer simulations of this problem for the one-, two- and
three-dimensional cases. The results are shown in Fig. 2 (a) together with the simple
mean field approximation of Eq. (\ref{n3}).
We found that fitting our numerical data using Eq. (\ref{corr})
shows excellent agreement if we choose $b_1=1.67,\ b_2=1.49,\ b_3=1.15$. It would be natural
to think that the simple mean field approach with $\alpha =0$ becomes exact for large
values of $d$.
Unfortunately, Eq. (\ref{corr}) does not have an analytical solution for all $X$ and so
it is not convenient for our purpose.
We found that the simple interpolated formula
\begin{equation}
\rho_d(X)=\frac{1}{X^d+\sqrt{1+(b_d-1)^2 X^{2d}}},
\label{inrho}
\end{equation}
which resembles Eq.(\ref{1stappr}),
describes the residual density well for the whole range of distances.
The comparison of this formula with
the results of computer modeling is shown in Fig. 2 (b),(c),(d) for d= 1,2,3. Below we use only Eq. (\ref {inrho})
with the values of $b_d$ obtained above.
\section{Calculation of $J(R)$}
\subsection{Three-dimensional case}
The spin-spin interaction constant is the splitting energy between
the ground states for total spin $S=1$ and $S=0$
$$2J=E_g^{S=1}-E_g^{S=0}\equiv ^3\Sigma_u^+ - ^1\Sigma_g^+$$
for hydrogen-like molecule, where nuclei are represented by two impurities.
Hereafter we use effective atomic units (a.u.) which means that all distances are
measured in units of the effective Bohr radius $a_B=\hbar^2\epsilon/m^*e^2$, and
energies in units of $m^*e^4/\hbar^2\epsilon^2$, where $m^*$ is the effective
carrier mass, and $\epsilon$ is the dielectric constant.
We propose a simple interpolated formula for the exchange constant based on
the most accurate numerical calculations of the hydrogen molecule \cite{vol}
and the following asymptotic expression\cite{Gor} for large $R$:
\begin{equation}
2J(R)\approx 1.636R^{5/2}\exp(-2R).
\label{as3d}
\end{equation}
We found $J(0)$ from the data for the
singlet-triplet splitting of the helium atom\cite{Rad}.
$$2J(0)=0.770\ \mbox{a.u.}$$
The numerical data\cite{vol} show that the behavior of the logarithm of the exchange
constant for small $R$ is well described by a second order polynomial.
To obtain the interpolated formula we match the second derivative of
$\ln\left(J(R)\right)$.
In two regions it has the following behavior
\begin{equation}\label{lnlim}
\frac{\partial^2\ln(J)}{\partial R^2}\approx\left\{
\begin{array}{ll}
-2\tilde{A}, & R\leq 1\\
-\frac{5}{2R^2}, & R\gg 1,
\end{array}\right.
\end{equation}
where $\tilde{A}$ is the matching constant. The simplest formula that satisfies
both conditions is
\begin{equation}
\frac{\partial^2\ln(J)}{\partial R^2}=-\frac{2\tilde{A}}{1+4/5\tilde{A}R^2}
\end{equation}
After integrating twice we obtain
\begin{eqnarray}\label{int3D}
\ln(J) &=&\ln(J(0))-\gamma R-\frac{5}{2}AR\arctan(AR)+\frac{5}{4}\ln(1+A^2R^2),
\end{eqnarray}
where $A$ and $\gamma$ are connected by equation
\begin{equation}
A=\sqrt{4\tilde{A}/5}=\frac{4(2-\gamma)}{5\pi}.
\nonumber
\end{equation}
This interpolated formula has one fitting parameter $\gamma$ and
the correct asymptotic behavior.
The parameter $\gamma$ has to be chosen to match small distances in an
optimal way. The least square method gives $\gamma=0.1$. The final equation is
\begin{equation}
\label{int3Dm}
2J_3(R)= 0.770\left(1+0.23R^2\right)^{5/4}\exp
\left(-0.1 R-1.210 R\arctan\left(0.484 R\right)\right)\nonumber\\
\end{equation}
For further calculations we need the inverse function
$R(J)$ as well. Because of the exponential character of the exchange
constant, the inverse function depends on energy logarithmically.
Therefore, we performed interpolation for the function
$R(x)$, where $x=\ln(J(0)/J)$.
The result is
\begin{equation}\label{R3d}
R_3=\frac{x}{2}+\frac{3.5x}{1+\frac{3.5x}{1.69+0.68\ln(1+x)}}
\end{equation}
The interpolated curves and all available data are shown in Fig. 3a.
\subsection{Two-dimensional case with in-plane impurities.}
We are unaware of any calculations of $J(R)$ for the two-dimensional case.
We have considered a general problem when the motion of
the electrons is confined to a plane, but the Coulomb impurities are at distances
$h_1$ and $h_2$ outside the plane. However, in this paper only the
calculations for in-plane impurities ($h_1=h_2=0$) are presented. The results for the general
case will be published elsewhere\cite{EFP}.
The case of the in-plane impurities corresponds to a $2D$ hydrogen-like
molecule with the Hamiltonian
\begin{equation}\label{H1}
\hat{H}=-\frac{\Delta_1}{2}-\frac{\Delta_2}{2}
-\sum_{j,i=1}^2\frac{1}{\sqrt{(x_i\pm a)^2 +y_i^2}} +
\frac{1}{\sqrt{(x_1-x_2)^2+(y_1-y_2)^2}}+\frac{1}{R}
\end{equation}
When $R\gg 1$ the singlet-triplet splitting constant
is calculated by making use the semiclassical approach
\cite{Gor,Smirnov,Flam99} (see Appendix \ref{ApC}).
We obtained the following result:
\begin{equation}\label{2dH_2}
2J(R)=30.413\, R^{7/4}\exp(-4R)
\end{equation}
To provide the point $R=0$ we performed variational calculations
for the two-dimensional helium atom. We found that (See Appendix \ref{ApHel})
\begin{eqnarray}\label{2dhel}
E(^{1}S)&=&-11.635\ \mbox{a.u.}\nonumber\\
E(^{3}S)&=& -8.193\ \mbox{a.u.}\nonumber\\
2J(0) &=&3.567\, (\pm 1\%)\ \mbox{a.u.}
\end{eqnarray}
Finally, we performed numerical calculations based on the method described in
\cite{Flam99} for the point $R=1$.
Using the same method as in the $3D$-case we get the following interpolated
formulas for $J(R)$ and $R\left(\ln(J_0/J)\right)$:
\begin{eqnarray}
2J_2(R) &=& 3.567\left(1+1.81R^2\right)^{7/8}\exp
\left(-0.3 R-2.355R\arctan\left(1.346R\right)\right)
\label{int2D}\\
R_2 &=& \frac{x}{4}+\frac{3x}{1+\frac{3x}{0.50+0.28\ln(1+x)}},
\label{R2d}\\
x &=& \ln\left(J(0)/J)\right).\nonumber
\end{eqnarray}
These results are shown in Fig. 3b.
\section{Conclusion}
We obtained an analytical expression
(\ref{inrho}) for the dimensionless density
of impurities which form singlet pairs with a distance larger than $R$.
We have also calculated the strength of the spin-spin interaction $J$ and obtained
analytical expressions for the function $R(J)$ for the three-dimensional (\ref{R3d})
and the two-dimensional (\ref{R2d}) cases. Combining Eq. (\ref{inrho}) with
Eqs. (\ref{R3d}) or (\ref{R2d}) one can calculate
an analytical expression for the density of singlet pairs $n(E)/2$ that has a
singlet-triplet energy splitting smaller
than $E=2J$. At finite temperature $T$ the pairs with $E<T$ are destroyed by a thermal
motion. Therefore, at a given temperature the function $n(E)$ at $E=T$ gives the
density of free spins in the system which contribute to the Curie susceptibility.
Thus, we obtain an analytical expression for the density of free spins $n(T)$.
We have also calculated the distribution function of excitation energy in a
logarithmic scale. It is defined as follows
\begin{equation}\label{Flog}
F(\ln(\varepsilon))=\frac{1}{2}\frac{d\,n}{d\,\ln(\varepsilon)}
\equiv \frac{n_0}{2}\frac{d\,\rho}{d\,R}\frac{d\,R}{d\,\ln(\varepsilon)},
\end{equation}
where $\varepsilon =E/(2J(0))\equiv J/J(0)$.
The analytical expression for $F(\ln(\varepsilon))$ based on Eqs. (\ref{inrho}),
(\ref{R3d}), (\ref{R2d}) is quite cumbersome.
In the two limits of large and small
energies (or small and large distances) the behavior of $F$ in the
leading order is
\begin{equation}\label{F2D}
\frac{F_{2D}}{n_0}\approx\left\{\begin{array}{ll}
10.56\pi n_0 a_B^2\ln(1/\varepsilon), &\qquad
\varepsilon\longrightarrow 1\\
\frac{16}{1.49\pi n_0 a_B^2}\left[\ln(1/\varepsilon)\right]^{-3}, &\qquad \varepsilon
\longrightarrow 0;
\end{array}\right.
\end{equation}
\begin{equation}\label{F3D}
\frac{F_{3D}}{n_0}\approx\left\{\begin{array}{ll}
128\pi n_0 a_B^3\left[\ln(1/\varepsilon)\right]^2, &\qquad
\varepsilon\longrightarrow 1\\
\frac{9}{1.15\pi n_0 a_B^3}\left[\ln(1/\varepsilon)\right]^{-4}, &\qquad \varepsilon
\longrightarrow 0.
\end{array}\right.
\end{equation}
Our results for $\rho (T)=n(T)/n_0$ are shown by the full lines in Fig. 4 (a), (b)
for the two-dimensional and the three-dimensional cases. We choose two different
dimensionless densities $n_0$ for each
case. They are $\pi n_0 a_B^2=0.1$ and 0.025 for $2D$ and $4\pi n_0 a_B^3/3=0.004$ and 0.016
for $3D$.
The dependence of the dimensionless distribution function $F/n_0$ for the two-dimensional and the
three-dimensional cases for the same two donor densities $n_0$ are presented on Fig. 4
(c), (d) by the full lines.
The most important features of both functions are the long logarithmic tails in the regions
of low temperature and low energy. Similar behavior has been obtained by Bhatt and Lee\cite{lee}.
Note that $\rho(T)$ decreases with increasing density $n_0$. This is not the case for the
distribution function. Larger density corresponds to larger distribution function $F$ at large
energies. This is because the derivative $n_0^{-1} dn/dR$ is larger for larger density $n_0$ at
small $R$. However, the dimensionless distribution function $F(\ln \varepsilon )$
is normalized to 1/2. That is why the functions for different densities
cross each other at some energy.
Thus, at
small energies the larger distribution function corresponds to smaller density.
To clarify the role of the functions $J(R)$,
which have been found here, we have calculated the density of free spins $\rho (T)$ and the
distribution function
$F(\ln(\varepsilon))$ for a simplified function $J_s(R)$ used in Ref.\cite{lee}.
\begin{eqnarray}\label{epslee}
J_s(R)&=& J(0)\exp(-2R/a_B) \qquad \mbox{for 3D,}\nonumber\\
J_s(R)&=& J(0)\exp(-4R/a_B) \qquad \mbox{for 2D.}
\end{eqnarray}
In these calculations we used our residual density $\rho(R)$. The results are shown
in Fig. 4 by the dashed lines. One can see that the difference is large. The
distribution function for our more accurate form of the exchange constant became narrower and
the decline in the beginning is steeper. This comes from the different behaviors
at small distances.
It is interesting to compare the numerical scaling calculations by Bhatt and Lee\cite{lee}
with our method of calculation $n(R)$. We have found that at the smallest density used
in Ref. \cite{lee} both methods give
similar results, but for larger densities there is a small deviation. We think that both
methods are exact in the limit of small densities, but the method of Bhatt and Lee
works in a wider range, because they take into
account the renormalization of weak bonds caused by their strong neighbors.
However, the great advantage of our method
is that it gives an analytical expression for $n(E)$.
\acknowledgments
This work is supported by the Australian Research
Council and by the Seed Grant of the University of Utah. A. L. Efros and V. V. Flambaum are
grateful to the Center for Theoretical Physics in Trieste for hospitality during
the work on this
project. A. L. Efros is grateful to R. Bhatt, D. Mattis, and B. Sutherland for helpful
discussions. I. Ponomarev acknowledge fruitful discussions with M. Kuchiev and G. Gribakin.
|
\section{Introduction}
\label{intro}
The non-equilibrium character of the dynamics of 3d spin glasses
below the zero field phase transition temperature has been extensively
studied by both experimentalists and theorists \cite{sgrf}. Two
different main tracks have been used to describe the aging and
non-equilibrium dynamics that is characteristic of spin glasses. On
the one hand phase space pictures \cite{hierarki}, which originate
from mean field theory \cite{mpv} and prescribe a hierarchical
arrangement of metastable states, and on the other hand real space
droplet scaling models which have been developed from renormalisation
group arguments \cite{FH,Henk}. Independently, a theoretical
description (which we shall not discuss here) of aging effects
\cite{CuKu1} and temperature variation effects \cite{CuKu2} has been
given by the direct solution of the dynamical equations of mean-field
like models.
When a spin glass is quenched from a high temperature (above $T_{g}$)
to a temperature $T_{1}$ below $T_{g}$, a wait time dependence of the
dynamic magnetic response is observed. This aging behaviour
\cite{aging,Sitges} corresponds to a slow evolution of the spin
configuration towards equilibrium. The `magnetic aging' observed
in spin glasses resembles the `physical aging' observed in the
mechanical properties of glassy polymers \cite{polymer,cavaille}, or
the `dielectric' aging found in supercooled liquids \cite{nagel}
and dielectric crystals \cite{KTNKLT}. However, a more detailed
comparison of aging in these various systems would show interesting
differences \cite{nagel}, particularly with respect to the effect of
the cooling rate \cite{uppsalaSaclay}.
Remarkable influences of slight temperature variations on the aging
process in spin glasses have been evidenced in a wide set of earlier
experiments \cite{hierarki,Sitges,UppsDT,DjurCuMn}. These influences
were further elucidated through the memory phenomena recently reported
in \cite{uppsalaSaclay}. The results have been interpreted both from
`phase space' and `real space' points of view.
In phase space models, aging is pictured as a random walk among the
metastable states. At a given temperature $T_{1}$, the system samples
the valleys of a fixed free-energy landscape. On the basis of the
experimental observations, it has been proposed \cite{hierarki} that
the landscape at $T_1$ corresponds to a specific level of a
hierarchical tree. When lowering the temperature to $T_{2}<T_1$, the
observed restart of aging is explained as a subdivision of the free
energy valleys into new ones at a lower level of the tree. The system
now has to search for equilibrium in a new, unexplored landscape, and
therefore acts at $T_2$ as if it had been quenched from a high
temperature. On the other hand, the experiments show that when
heating back from $T_{2}$ to $T_{1}$ the memory of the previous aging
at $T_{1}$ is recovered. In the hierarchical picture, this is
produced by the $T_2$-valleys merging back to re-build the
$T_1$-landscape.
In real space droplet pictures \cite{FH,Henk}, the aging behaviour at
constant temperature is associated with a growth of spin glass ordered
regions of two types (related by time-reversal symmetry). This is
combined with a chaotic behaviour as a function of temperature,
\cite{braymoore}, i.e. the equilibrium spin configuration at one
temperature is different from the equilibrium configuration at another
temperature. However, there is also an overlap between the
equilibrium spin structures at two different temperatures, $T$ and
$T\pm\Delta T$, on length scales shorter than the overlap length,
$L_{\Delta T}$. In this picture, chaos implies that if the spin glass
has been allowed to age a time, $t_{w}$, at a certain temperature, the
aging process is re-initialized after a large enough temperature
change. Intuitively, a growth of compact domains may not allow a
memory of a high temperature spin configuration to remain imprinted in
the system while the system ages at lower temperatures. However, as
suggested in \cite{uppsalaSaclay} and developed in \cite{jonssonetal},
a phenomenology based upon fractal domains and droplet excitations can
be able to incorporate the observed memory behaviour in a
real space droplet picture. The possibility of a fractal
(non-compact) geometry of the domains has been evoked in the past in
various theoretical contexts\cite{fractaldiv}, and also in close
connection with the aging phenomena \cite{fractclust,jpbdean,mfnoneq}.
In this paper, we report new results on the memory phenomenon observed
in low frequency ac-susceptibility of spin glasses. We first recall
and demonstrate an undisturbed memory phenomenon, and then show that
such a memory can be erased not only by heating the sample to a
temperature above the temperature where the memory is imprinted, but
also by waiting a long enough time below this temperature.
\section{Experimental}
\label{expe}
The experiments were performed on the insulating spin glass
$CdCr_{1.7}In_{0.3}S_{4}$ \cite{mtrl}
($T_{g}=16.7 K$), in a Cryogenic Ltd S600 SQUID magnetometer at Saclay. The ac
field used in the experiments had a peak magnitude of 0.3 Oe and
frequency $\omega/2\pi$=0.04 Hz. This low frequency makes the
relaxation of the susceptibility at a constant temperature in the spin
glass phase clearly visible (at the laboratory time scale of $10^1$ to
$10^5$ s).
\begin{figure}
\resizebox{0.47\textwidth}{!}{%
\includegraphics{fig1kjev.eps} }
\caption{ The measurement procedure in the `double memory experiments'
of Fig.2, 4 and 5.}
\label{un}
\end{figure}
The basic experimental procedure is illustrated in Fig. 1 and is as follows:
(i) Cooling: The experiments are always started at 20 K, a temperature
well above the spin glass temperature $T_{g}$=16.7 K. The
ac-susceptibility is first recorded as a function of decreasing
temperature. The sample is continuously cooled, but is additionally
kept at constant temperature at two intermittent temperatures $T_{1}$
and $T_{2}$ for wait times $t_{w1}$ and $t_{w2}$, respectively
($T_{1}<T_{2}<T_{g}$).
(ii) Heating: When the lowest temperature has been reached, the system
is immediately continuously re-heated and the ac-susceptibility is
recorded as a function of increasing temperature.
Except at $T_{1}$ and $T_{2}$ when decreasing the temperature, the
cooling and heating rates are constant ($\sim 0.1$ K/min.). At
constant temperature, both components of the ac-susceptibility relax
downward by about the same absolute amount. However, the relative
decay of the out-of-phase is much larger than the relative decay of
the in-phase component, and in the following we mainly focus on results
from the out-of-phase component of the susceptibility.
\section{Results}
\label{resu}
\subsection{Double memory}
The results of a double memory experiment are presented in Fig. 2.
The initial data is recorded on continuously cooling the sample
including a first halt at the temperature $T_{1}$=12 K (0.72$T_{g}$)
for $t_{w1}$=7 $hrs$ and a second halt at $T_{2}$=9 K (0.54$T_{g}$)
for $t_{w2}$=40 $hrs$. The cooling is then continued to $T$=5 K,
from where a new set of data is taken on increasing the temperature at
a constant heating rate whithout halts. A reference curve, measured
on continuous heating after cooling the sample without intermittent
halts, is included in the figure.
\begin{figure}
\vskip 2.1cm
\resizebox{0.48\textwidth}{!}{%
\includegraphics{fig2kjev.eps} }
\caption{ Out-of-phase susceptibility vs. temperature. While cooling
down (open diamonds), two intermittent halts are made; at $T_{1}$=12
$K$ during $t_{w1}$=7 $hrs$, and at $T_{2}$=9 K during $t_{w2}$=40
$hrs$. The system is then reheated at a constant heating rate (full
circles). The reference curve (solid line) is measured on heating the
sample after cooling it without intermittent halts. The inset shows
the in-phase susceptibility in the same measurement procedure, at
temperatures around $T_{1}$=12 K.}
\label{deux}
\end{figure}
A first important feature can be noted on the curve recorded on
cooling with intermittent halts. After aging 7 $hrs$ at 12 K, $\chi''$
has relaxed downward due to aging. But when cooling resumes, the
curve rises and merges with the reference curve, as if the aging at 12 K
was of no influence on the state of the system at lower temperatures.
This chaos-like effect (in reference to the notion of chaos in
temperature introduced in \cite{braymoore}) points out an important
difference from a simpler description of glassy systems, in which
there are equivalent equilibrium states at all low temperatures and
aging at any temperature implies that this equilibrium state is
further approached. Here, as pictured in more detail in other
experiments \cite{uppsalaSaclay}, only the last temperature interval
of the cooling procedure does contribute to the approach of the
equilibrium state at the final temperature.
Note that this notion of
`last temperature interval' depends on the observation time scale
of the measurement ($\chi''$ is only sensitive to dynamical processes
with a characteristic response time of order $1/\omega$ which in these
experiments corresponds to $\approx 4$ s). In magnetisation relaxation
experiments, the observation time corresponds to the time elapsed after
the field change, and effects of aging at a higher temperature can be
seen in the long-time part ($10^3-10^4$ s) of the relaxation curves
\cite{marcos,DjurCuMn} in a correspondingly enlarged `last temperature
interval' compared to that of our current ac-susceptibility
experiments.
The curve recorded on re-heating in Fig. 2 clearly displays the
memory effect: the dips at $T_{1}$ and $T_{2}$ are recovered. The
long wait time (40 $hrs$) at $T_{2}$= 9 K has no apparent influence on
the memory dip associated with $T_{1}$=12 K. This experiment
displays a double memory where no interference effects are present,
i.e. the two dips at $T_{1}$ and $T_{2}$ are, within our experimental
accuracy, fully recovered when reheating the sample. A similar result
has been obtained on a metallic Cu:Mn spin-glass sample
\cite{uppsalaSaclay,DjurCuMn}, confirming the universality of aging
dynamics in very different spin-glass realizations.
This experimental procedure has been recently reproduced in extensive
simulations of the 3d Edwards-Anderson model. Although weaker and
more spread out in temperature, similar effects of a restart of
aging upon cooling and of a memory effect upon heating have been
found \cite{Takayama}.
The memory phenomenon is also observable in the in-phase
susceptibility, $\chi$'. In the inset of Fig. 2, $\chi$' is plotted
in the region around $T_{1}$=12 $K$. A relaxation due to aging is
visible, and it is also clear that when cooling resumes after aging,
the $\chi$' curve rather rapidly merges with the reference curve
(chaos-like effect). Upon re-heating, the memory effect can be
distinguished, the memory curve clearly departs from the reference in
the 11-13K range. The relative weakness of the deviation compared to
the large $\chi$' value can be understood in the following way.
Aging in spin glasses mainly affects processes with relaxation times
of the order of the age of the system; processes with shorter
relaxation times are already equilibrated, and processes with longer
relaxation times are not active. $\chi$' measures the integrated
response of all short time processes up to the observation time
$1/\omega$, whereas $\chi$'' only probes processes with relaxation
times of order $1/\omega$. The relative influence of aging is thus
smaller in $\chi$' than in $\chi$''.
\subsection{Memory erasing by heating}
\begin{figure}
\vskip 2.0cm
\resizebox{0.49\textwidth}{!}{%
\includegraphics{fig3kjev.eps} }
\caption{ Memory erasing by overheating. The memory effect at 12K,
recorded during heating with the same procedure as in Fig.2, is shown
in full diamonds. Now, when reaching $T^*$ ($T^*$= 12.1, 12.3, 12.5,
12.7 and 12.9 K), re-heating stops and the sample is cooled back. The
$\chi''$ signal (alternatively open circles and crosses for the
various $T^*$ values) during cooling shows the progressive erasing of
the 12K memory for increasing $T^*$. Reference curves, measured
during a continuous cooling (dashed line) and heating (solid line),
are also shown. }
\label{trois}
\end{figure}
The memory of aging at $T_1$ remains imprinted in the system during
additional aging stages at sufficiently lower temperatures, and is
recovered when heating back to $T_1$. We have investigated what remains of this $T_1$-memory
after heating up to $T^*>T_1$ (keeping of course $T^*<T_g$). The
experiments were performed using only one intermittent stop at
$T_1=12$ K for $t_{w1}$= 3 $hrs$, and continuing the cooling to about 10
K. Then
$\chi''$ was recorded upon heating the sample to $T^*$ and
immediately re-cooling it to 10 K. The results are displayed in Fig. 3.
For higher and higher $T^*$, the memory dip at $T_1$ becomes weaker
and weaker, finally fading out at $T^*\sim$ 13 K. The additional
shallow dips observed in a limited temperature region just below $T^*$
and just above 10 K, the two temperatures where the temperature change
is reversed, are due to the finite heating/cooling rate and to the overlap within this temperature range between the state created
on heating (cooling) and the desireable state on re-cooling
(re-heating) the sample \cite{jonssonetal}.
\subsection{Memory interference}
The memory is also affected by aging at a lower temperature, provided
this temperature is close enough to $T_{1}$ or the time spent there is
long enough. In order to systematically investigate this interference
effect, we have performed double memory experiments in which we have
varied the parameters $T_{2}$ and $t_{w2}$ of the aging stage at the
lower temperature, but kept the initial aging temperature $T_{1}$= 12
K and wait time $t_{w1}$=3 $hrs$ fixed.
\begin{figure}
\vskip 2.0cm
\resizebox{0.48\textwidth}{!}{%
\includegraphics{fig4kjev.eps} }
\caption{ Effect on the memory at $T_{1}$=12 K ($t_{w1}$=3 $hrs$) of
aging at slightly lower temperatures $T_{2}$ (9.5 K and 10.5 K) during
$t_{w2}$=6 $hrs$. All data shown was taken on reheating after the
various histories. In addition to the reference curve (solid line), a
`single memory' curve (open circles) is presented; it shows the 12K
memory when no additional aging at $T_2$ is performed. For $T_2$=9.5 K
(crosses), there is almost no effect; for $T_2$=10.5 K, the 12K memory
is partly erased.
}
\label{quatre}
\end{figure}
Fig. 4 shows the results using a fixed value of $t_{w2}$=6 $hrs$ but
two different values of $T_2$ (9.5 and 10.5 K). Two reference heating
curves are added for comparison; one is recorded after a cooling
procedure where no halts are made, and the other after cooling with
only a single halt at $T_1$=12 K for 3 $hrs$ (`single memory'). This
latter curve is a reference for a pure memory effect at 12 K. The
obtained 12 K dip is about the same for the pure memory and the double
memory curve with $T_{2}$=9.5 K. However, in the experiment performed
with $T_{2}$=10.5 K , the memory of the dip achieved at $T_{1}$=12 K
has become more shallow. Thus, for a temperature difference $\Delta
T$=1.5 K, the memory of aging gets partly re-initialised, while for
$\Delta T$=2.5 K no re-initialisation is observed.
\begin{figure}
\vskip 2.0cm
\resizebox{0.48\textwidth}{!}{%
\includegraphics{fig5kjev.eps} }
\caption{ Effect on the memory at $T_{1}$=12 $K$ ($t_{w1}$=3 $hrs$) of
aging at $T_{2}$=10.5 K during $t_{w2}$=6 (crosses) and 12 $hrs$ (full
diamonds). Same conventions as in Fig.4.
}
\label{cinq}
\end{figure}
Fig. 5 shows the results using a fixed value of $T_{2}$=10.5 K but two
different values of the wait time at $T_2$ $t_{w2}$=6 $hrs$
and $t_{w2}$=12 $hrs$ . The reference curves are the
same as in Fig. 4. The longer the time spent at $T_{2}$, the larger
is the part of the memory dip at $T_{1}$ that has been erased.
\section{Discussion}
\label{discu}
\subsection{Memory and chaos effects in phase space pictures}
Some effects related to these memory and memory interference phenomena
have been explored in the past through various experimental
procedures, as well in ac as in dc (magnetisation relaxation following
a field change) measurements \cite{hierarki,Sitges,UppsDT}. The
hierarchical phase space picture \cite{hierarki} has been developed
as a guideline that accounts for the various results of the
experiments. Although this hierarchical picture deals with metastable
states as a function of temperature, it is obviously reminiscent of
the hierarchical organization of the pure states as a function of
their overlap in the Parisi solution of the mean field spin glass
\cite{mpv}. We want to recall that some more quantitative analyses of the
experiments \cite{sacorbach} have shown that the
barrier growth for decreasing temperatures should be associated with a
divergence of some barriers at any temperature below $T_g$. Thus,
what can be observed of the organization of the metastable states
might well be applicable to the pure states themselves. From a
different point of view, another link between the hierarchical picture
and mean field results has been proposed in a tree version of
Bouchaud's trap model \cite{jpbdean}.
The restart of aging when the temperature is again decreased after
having been halted at some value indicates that the
free-energy landscape has been strongly perturbed. The metastable states
have been reshuffled, but not in any random manner, because the
memory effect implies a return to the previously formed landscape
(with the initial population distribution) when the temperature is
raised back. The restart of aging then corresponds to the growth of
the barriers and to the birth of other ones, which subdivide the
previous valleys into new ones where the system again starts some
ab initio aging. This hierarchical ramification is easily reversed
to produce the memory effect when the temperature is increased back.
The full memory effect seen in the experiment of Fig. 2 requires a
large enough temperature separation $\Delta T=T_1-T_2$, as is shown
from the memory interference effects displayed in Figs. 4 and 5.
If one forgets the restart of aging at the lower temperature, the memory
effect can be given a simple explanation: the slowing down related to
thermal activation is freezing all further evolution of the system.
However, the memory effect takes place while important relaxations
occur at lower temperatures, and it is clear that there must be some
smaller limit of $\Delta T$ below which the $T_2$ evolution is of
influence on the $T_1$ memory. From other measurements
\cite{hierarki,Sitges}, it has been shown that in the limit of small
enough $\Delta T$'s (of order 0.1-0.5 K), the time spent at $T_2$
contributes essentially additively to the aging at $T_1$, as an
effective supplementary aging time. In that situation of small
$\Delta T$, the landscape at $T_2$ is not very different from that at
$T_1$ (large `overlap'); the same barriers are relevant to the aging processes,
although being crossed more slowly at $T_2$.
But this is not the case
in Figs. 4 and 5, where intermediate values of $\Delta T$ have been
chosen.
The memory interference effect
demonstrated in Figs. 4 and 5 is in agreement with earlier ac and dc
experiments which used negative temperature cycling procedures
\cite{hierarki,UppsDT} and intermediate magnitudes of $\Delta T$.
Such experiments were performed so that the sample first was aged at
$T_1$ a wait time $t_{w1}$, and thereafter cooled to $T_{2}$ and kept
there a substantial wait time $t_{w2}$, after which it was re-heated to
$T_1$, where the relaxation of the ac or dc signal was recorded. The
results of these experiments are that a partial re-initialisation of the
system has occurred, but that simultaneously a memory of the original aging at $T_{1}$ remains. In Figs. 4 and
5, the partial loss of the $T_1$ memory dip corresponds to such a
partial reinitialisation.
In this case of an intermediate value of $\Delta T\sim 1$ K, there are
indeed differences between the landscapes at both temperatures.
Still, they are hierarchically related, since a memory effect is
found. But the memory loss of Figs. 4 and 5 suggests that the
free-energies of the bottom of the valleys are different at $T_1$ and
$T_2$, meaning that the thermodynamic equilibrium phase is different
from one temperature to another. As discussed previously
\cite{uppsalaSaclay}, the restart of aging when the temperature is
lowered is suggestive of chaos between the equilibrium correlations at
different temperatures. The conclusion from the current results is
thus that the free-energies of the metastable states vary
chaotically with temperature, which reinforces the idea of a
`chaotic nature of the spin glass phase' \cite{braymoore}.
\subsection{Towards an understanding of memory and chaos effects in
real space}
While these phase space pictures allow a good description of many
aspects of the experimental results, a correct real space picture would
form the basis for a microscopic understanding of the
physics behind the phenomena. As aging proceeds, $\chi$'' decreases,
which means a decrease of the number of dynamical processes that have
a time scale of order $1/\omega$. Aging corresponds to an overall
shift of a maximum in the spectrum of relaxation times towards longer
times, as was understood from the early observations of aging in
magnetisation relaxation experiments \cite{lundgren83}. Thinking of
the dynamics in terms of groups of spins which are simultaneously
flipped, longer response times are naturally associated with larger
groups of spins. In such `droplet' \cite{FH} and `domain' \cite{Henk}
pictures, aging corresponds to the progressive increase of a typical
size of spin glass domains. Difficulties are encountered in this real
space description with `memory and chaos' effects
\cite{uppsalaSaclay}. On the one hand, the restart of aging
processes when the temperature is lowered indicates the growth of
domains of different types at different temperatures. On the other
hand, the memory of previous aging at a higher temperature can be
retrieved; thus, the low temperature growth of domains of a given type
does not irreversibly destroy the spin structures that have developed
at a higher temperature.
However, a heuristic interpretation of aging in spin glasses in terms
of droplet excitations and growth of spin glass equilibrium domains,
along the lines suggested in \cite{uppsalaSaclay} and developed in
\cite{jonssonetal}, is perhaps able to include the memory phenomena
discussed in this paper. The carrying idea of this phenomenology is
that at each temperature there exists an equilibrium spin glass
configuration that is two fold degenerate by spin reversal symmetry.
The simple picture of Fisher-Huse \cite{FH}, where only compact
domains are considered, is hard to reconcile with the memory effect
reported here \cite{uppsalaSaclay}. As suggested in various
theoretical work \cite{fractaldiv,fractclust,jpbdean,mfnoneq}, we
assume that the initial spin configuration results in an
interpenetrating network of fractal `up' and `down' domains of all
sizes separated by rough domain walls. We furthermore propose to
modify somewhat the original interpretation of the `overlap length'
$L_{\Delta T}$. The standard picture states that the equilibrium
configurations corresponding to two nearby temperatures $T$ and
$T\pm\Delta T$, are completely different as soon as one looks at a
scale larger than $L(\Delta T)$. However, in a non-equilibrium
situation, we believe that some fractal large scale (larger than
$L_{\Delta T}$) `skeletons', carrying robust correlations, can survive
to the change of temperature, and are responsible for the memory
effects. An assumption of this sort is, we think, needed to account
for the existence of domains of all sizes within the initial
condition. If the initial spin configuration was purely random as
compared to the equilibrium one, then the problem would be tantamount
to that of percolation far from the critical point, where only small
domains exist.
The allowed excitations in this system
are droplets of correlated `up' or `down' regions of spins of all
sizes. Within this model, the magnetisation of the sample in response
to a weak magnetic field is caused by polarisation of droplets, and the
out-of-phase component of the susceptibility directly reflects the
number of droplet excitations in the sample with a relaxation time
equal to $1/\omega$. The size of spin glass domains is in the following
denoted $R$ and the size of a droplet excitation $L$. As a function of time,
the size of the excited droplets grows as $L(T,t_a)$.
The effect of a droplet excitation on the spin configuration is
different depending on the size and position of the droplet and the
age of the spin glass system. A small droplet excitation $L\ll R$
most probably is just an excitation within an equilibrium spin glass
configuration, yielding no measurable change of the spin system. An
excitation of size $L\approx R$ may
(i) remove the circumventing domain wall separating an up domain from a
down domain, (ii) slightly displace an existing domain wall or (iii)
just occur within an equilibrium spin configuration.
After a few decades in time, the result of the numerous dispersed
droplet excitations of sizes $L\leq L(T,t_{a})$ is that most domains
of size $R\ll L(T,t_{a})$ are removed, whereas most larger domains remain
essentially unaffected, only having experienced numerous slight domain
wall displacements. In other words, in this picture, the structure of the
large domains is unaffected by the dynamics. If the temperature now is
changed to a
temperature where the overlap length, $L_{\Delta T}$, is smaller than the
typical droplet size $L(T_1,t_{a})$ active at the original temperature
$T_{1}$, the new
initial condition still leads to an interpenetrating network of up and down
domains of all sizes relevant to the new temperature (chaos implies that
the domain spin structure is different from the equilibrium configuration
at $T_{1}$). A similar
process as discussed above now creates a spin configuration of
equilibrium structure on small length scales, where only small domains
(up to a size $L(T_2,t_{a})$) are erased, leaving large ones essentially
unaffected. Returning to the original temperature, there will exist a new
domain
pattern on small length scales corresponding to droplets with short
relaxation time and there will additionally remain an essentially
unperturbed domain pattern on large length scales originating from the
initial wait time at this temperature. The small length scale domains
are rapidly washed out and a domain pattern equivalent
to the original one is rapidly recovered.
These dynamic features of the spin structure are reflected in the
susceptibility experiments discussed above. The out of phase
component gives a measure of the number of droplets that have a
relaxation time equal to the observation time of the experiment,
$1/\omega$, the decrease of the magnitude of the susceptibility with
time implies that the number of droplets of relaxation time $1/\omega$
decays toward an equilibrium value obtained when `all' domains of this
size are extinguished and all subsequent droplet excitations of this
size occur within spin glass ordered regions. The fact that a dip
occurs when the temperature is recovered mirrors that the long length
scale spin configuration is maintained during an aging period at
lower temperatures, where only droplet excitations on much smaller
length scales are active. The memory interference effects are within
this picture immediate consequences of that the droplet excitations at
the nearby temperature $T_{2}$ are allowed to reach the length scales
of the original domain growth at $T_{1}$ leading to an enhanced number
of droplet excitations of the size of this reconstructed domain
pattern. When heating above the temperature for the aging process,
Fig. 3, and outside the region of overlapping states, the
equilibration processes at the high temperature reach longer length
scales than at the aging temperature, and the memory of the
equilibration at $T_{1}$ is rapidly washed out. On the other hand, in
the experiments where the sample is cooled (Figs. 4 and 5) below
$T_{1}$, the processes require longer aging time to reach the length
scales of the aging at $T_{1}$ and the intereference effects become
larger with increased time and higher temperature.
\section{Conclusions}
\label{conc}
When cooling a spin glass to a low temperature in the spin glass
phase, a memory of the specific cooling sequence is imprinted in the
spin configuration and this memory can be recalled when the system is
continuously re-heated at a constant heating rate
\cite{uppsalaSaclay}. E.g. in a `double memory experiment' two
intermittent halts, one at $T_{1}$ a time $t_{w1}$ and another at
$T_{2}$ a time $t_{w2}$ are made while cooling the sample. Depending
on the parameters $T_{2}$ and $t_{w2}$ it is possible to partly
reinitialise (erase) or fully keep the memory of the halt at the higher
temperature $T_{1}$.
These memory and memory interference effects can on the one hand be
incorporated in hierarchical models for the configurational energies
at different temperatures including a chaotic nature of the spin glass
phase. On the other hand, a preliminary phenomenological
real space picture has been proposed to account for the observed
phenomena. In our mind, much remains to be done on the theoretical side to
put this `fractal' droplet picture on a firmer footing.
\section{Acknowledgments}
Financial support from the Swedish Natural Science Research Council
(NFR) is acknowledged.
We are grateful to L.F. Cugliandolo, T. Garel, S. Miyashita, M. Ocio and
H. Takayama for valuable discussions, and to L. Le Pape for his
technical support.
|
\section{Introduction}
Attractions among stable colloidal particles
lead to a diverse
phase behavior. Colloidal attractions, unlike the molecular attractions,
act usually over a relatively short (compared to the particle size) range.
It is by now well established that when the range of the colloidal
attraction is decreased the phase diagram undergoes a progression
from gas--liquid--solid to fluid--solid coexistence,
the latter with a subcooled
critical point which is metastable with respect to fluid--solid coexistence
\cite{Gas83,Gas86,Tej94,Hag94,Lek95}.
Numerous experimental studies show that
suspensions often form incompletely equilibrated solids
with the appearance of gels where one expects a fluid--solid
\cite{Pat89,Emm90,Poo93,Ile95,Ver97,Poo97b,Jan86,Che91,Gra93}
or gas--liquid \cite{Ver95} phase separation from equilibrium theory.
The systems studied include
mixtures of colloids and non-adsorbing polymer
\cite{Pat89,Emm90,Poo93,Ile95,Ver97,Poo97b} and
sterically stabilized colloids in marginal
solvents \cite{Jan86,Che91,Gra93,Ver95,Rue97,Rue98}.
In the former case, the attractions stem from
depletion of the polymer coils from the regions between closely
spaced particles \cite{Asa54,Vri76}, and in the latter they are caused
by surface grafted chain--chain interactions
\cite{Rou88}. The gel transition is observed
when the range of attraction
is short compared to the particle size.
In the colloid--polymer mixtures this is achieved by choosing a
small ratio of polymer to colloid size, whereas
the overlap length of the surface grafted chains sets essentially the
range of the attraction in the sterically stabilized particle
systems.
While the equilibrium phase behavior
of these systems is well understood \cite{Gas83,Gas86,Tej94,Hag94,Lek95},
the nature of the gel transition
remains to be clarified.
The gel state appears to be related to a ramified structure with
interconnected particle clusters \cite{Ver97,Poo97b}. Temporal
density fluctuations are very slow close to the gel transition
and the suspensions acquire a yield stress
and a finite low-frequency elastic shear modulus in the gel
\cite{Poo97b,Che91,Gra93,Rue98,Ver96}.
In the past the transition to
the gel state has most often been interpreted as either a
static percolation transition \cite{Che91,Ver95,Wou94},
where a sample-spanning cluster of
particles forms, or due to the fluid--solid phase transition \cite{Gra93}.
Comparison between integral equation predictions
for the percolation transition and experimental data, however, shows that
the gel transition is
confined to the region in the phase diagram
between the static percolation threshold and
the gas--liquid spinodal \cite{Gra93,Ver95}.
Colloidal gels have also been attributed to states inside a gas--liquid
binodal which is metastable with respect to fluid--solid coexistence.
Such metastable binodals have indeed been observed for suspensions of
globular proteins \cite{Mus97}, which also form
gels when the ionic strength is sufficiently high \cite{Mus97,Geo94,Ros95}.
In this work we present an alternative interpretation of
the dynamical arrest of the gel structure which causes colloidal systems
to become disordered solids.
We propose that colloidal gels are nonergodic systems that form
when a dynamic gel transition is traversed. We further suggest that
the gel transition is a low temperature
extension of the liquid--glass transition.
The gels, however, differ physically from colloidal hard sphere glasses
in that they generally display a larger elastic shear modulus
and that the particles are more strongly
localized in the gels; both of these effects are due to particle clustering
induced by a short--range attraction among particles.
We demonstrate that this is a possible explanation for
colloidal gel formation by applying
the idealized mode coupling theory (MCT)
of the form used to study the liquid--glass
transition \cite{Ben84,Got91,Got91b,Got92,Fuc95b}
to systems in which the attraction is restricted to short ranges.
This study is motivated by
the observations made by Verduin and Dhont \cite{Ver95},
who noted a structural arrest (nonergodicity)
in connection with the gel transition
similar to that observed for hard sphere colloidal
glasses \cite{Pus87,vMe91,vMe94,vMe95}. Also Poon et al. \cite{Poo97b}
have made such observations in the so-called transient gelation region
of the colloid--polymer phase diagram
for short polymers.
Krall and Weitz \cite{Kra98} have shown that, in the limit of strong
particle aggregation, suspensions become nonergodic even at low colloid
densities. To date, however, only a speculative
connection has been made between the gel
and liquid--glass transitions \cite{Ver95}.
We conduct a study of ergodicity breaking
in two model systems: Baxter's adhesive hard sphere (AHS) system
and the hard core attractive Yukawa (HCAY) system.
Both systems supply analytical solutions
for the static structure factor,
the former within Percus-Yevick (PY) theory \cite{Bax68b}, and the latter
within the mean spherical approximation (MSA) \cite{Wai73,Hoy77,Cum79}.
This study provides more information on the AHS phase diagram and, in
addition, serves to complement
a recent independent MCT study \cite{Fab98}
on the temperature dependence of the AHS glass transition.
Further, the HCAY system provides a likely candidate for the gel transition
in colloidal systems as an ergodicity breaking dynamic transition
of the same type as the liquid--glass transition,
suggesting that the experimentally observed gel transition
is a low temperature extension of the glass transition.
In what follows, the idealized
mode coupling theory of the liquid--glass
transition, suitable for colloidal suspensions,
is briefly summarized. Results for the temperature dependence
of the glass transition are then presented and compared to the AHS
phase diagram as predicted by density functional theory.
Results are also shown for the HCAY system,
which show that the MCT glass transition
extends to the critical and subcritical region at low temperatures.
The way in which the glass transition line traverses this part of the
phase diagram is shown to be a strong function of the
range of attraction.
\section{Mode coupling theory}
The mode coupling theory (MCT) of the liquid--glass transition
provides a dynamic description of the
transition \cite{Ben84,Got91,Got91b,Got92,Fuc95b,Leu84}.
For sufficiently strong interactions the
dynamical scattering functions do not decay to zero with time,
leaving instead finite residues -- the nonergodicity parameters,
also known as Edwards--Anderson parameters or glass form factors.
Generically, concurrent with this long-time diffusion
ceases and the zero-shear viscosity diverges, both due to a diverging
relaxation time. This structural relaxation time is in turn
related to the particles' inability of escaping their
nearest neighbor cages.
The glass transition within the framework of the idealized MCT is not
a conventional thermodynamic phase transition;
the constrained motion of the particles leads to a difference
between time and ensemble averages, i.e. an
ergodicity breaking transition.
Application of the MCT to the Mori--Zwanzig reduced equations of motion
for the density correlators, together with a $t\rightarrow \infty $ limit,
leads to the following set of closed
equations~\cite{Ben84,Got91,Got91b,Got92,Fuc95b,Sza91}
\begin{eqnarray}
\label{coh}
\frac{f_q}{1-f_q} &=& \frac{\rho }{2(2\pi )^3q^2}\int d{\bf k}\,
V({\bf q},{\bf k})^2\, S_qS_kS_{|{\bf q}-{\bf k}|}f_kf_{|{\bf q}-{\bf k}|}
\\
V({\bf q},{\bf k}) &=& {\hat {\bf q}}\cdot ({\bf q}-{\bf k})
\, c_{|{\bf q}-{\bf k}|}
+{\hat {\bf q}}\cdot {\bf k}\; c_k+q\rho c^{(3)}({\bf q},{\bf q}-{\bf k})
\nonumber
\\
\label{incoh}
\frac{f_q^s}{1-f_q^s} &=& \frac{\rho }{(2\pi )^3q^2}\int d{\bf k}\,
V^s({\bf q},{\bf k})^2 S_kf_kf^s_{|{\bf q}-{\bf k}|}
\\
V^s({\bf q},{\bf k}) &=& {\hat {\bf q}}\cdot {\bf k}\; c_k
\nonumber
\end{eqnarray}
where $\rho $ is the number density,
$S_q$ is the static structure factor, $c_q=(S_q-1)/\rho S_q$
is the Fourier--transformed Ornstein--Zernike direct correlation
function, and $c^{(3)}$ is the triplet direct correlation function.
In this study we use primarily the so-called convolution approximation
($c^{(3)}=0$) \cite{Got91,Got92,Jac62}
for the triplet direct correlation function.
Note that the coupling vertices $V({\bf q},{\bf k})$ and
$V^s({\bf q},{\bf k})$ in Eqs. \ref{coh} and \ref{incoh}
are free of singularities and vary smoothly with a set of external
control parameters, e.g., the particle density.
The coherent ($f_q$) and incoherent ($f_q^s$) nonergodicity parameters
are defined as the long-time limits of the intermediate scattering function
$F_q(t)$ and the self-intermediate
scattering function $F^s_q(t)$, according to
\begin{eqnarray}
f_q &=& \lim_{t\rightarrow \infty}\left(F_q(t)/S_q\right)
\label{def1}
\\
f_q^s &=&\lim_{t\rightarrow \infty}\left(F_{q}^s(t)\right).
\label{def2}
\end{eqnarray}
The nonergodicity parameters determine the properties of the glass.
The zero-frequency elastic shear modulus of the colloidal glass
(in units of $k_BT/\sigma ^3$, with
$k_BT$ the temperature and $\sigma $ the particle diameter)
is given by \cite{Got91,Nag98}
\begin{equation}
G = \frac{\sigma ^3}{60\pi ^2}\int_0^\infty dk
k^4 \left(\frac{d\ln{S_k}}{dk}f_k\right)^2.
\label{G}
\end{equation}
The incoherent nonergodicity parameter
$f_q^s$ is found to be well approximated by a Gaussian, the
half-width of which is proportional to the mean-square
displacement in the glass state \cite{Ben84,Got91}.
The localization length $r_s$
is defined as the root-mean-square displacement in the glass,
and is determined from $f_q^s = 1-q^2r_s^2$ for $q\to 0$ \cite{Got91}.
Eqs. \ref{coh} and \ref{incoh} are
solved self-consistently for the nonergodicity
parameters as functions of specified external control parameters: the
reduced temperature $\tau $ and
volume fraction $\phi=\pi \rho \sigma ^3/6$ for the AHS system,
and the reduced temperature K$^{-1}$, screening parameter b,
and volume fraction for the HCAY system (cf. below).
The solution proceeds by iteration,
first on $f_q$, and
subsequently on $f_q^s$.
Transition lines delineating ergodic and nonergodic states were
found by bracketing, and the monotonicity property of the iteration
\cite{Got95} was employed.
The integrations
were performed numerically using Simpson's rule on a uniformly
discretized wavevector grid: $q_i=i\Delta q$,
$i=0, \ldots , N$.
The parameters $\Delta q$ and $N$ used varied somewhat, but most results
were obtained using $0.15 < \Delta q\sigma < 0.30$ and $600 < N <1000$.
The iterative solution scheme
was complemented occasionally by an algorithm to speed up convergence by using
stored previous iterates \cite{Ng74}.
We have also directly integrated
the equations of motion \cite{Fuc91} (adjusted to obey Smoluchowski dynamics).
This yields the entire time-evolution of the density correlators
$F_q(t)$ and $F^s_q(t)$,
the long-time limits of which were found to be identical to
the solutions of Eqs. \ref{coh} and \ref{incoh}.
The result $f_q=f^s_q=0$ is always a solution to Eqs. \ref{coh}
and \ref{incoh},
implying that correlations among density fluctuations vanish for
long times. At low densities this is the only solution,
hence the system is in a fluid, possibly metastable fluid state.
Above a critical volume fraction $\phi _c$
also non-zero solutions appear, which correspond to
nonergodic glass states. The physical solutions to
Eqs. \ref{coh} and \ref{incoh}, corresponding to
the long-time limits defined by Eqs. \ref{def1} and \ref{def2},
were identified by choosing the largest solutions for
$f_q$ and $f_q^s$ \cite{Got91,Got95}.
\section{Adhesive hard sphere system}
The AHS pair potential consists
of an infinitely deep and narrow well
located at particle contact,
given explicitly by \cite{Bax68b}
\begin{equation}
u(r)/k_BT = \lim_{d\rightarrow \sigma ^+}\left\{ \begin{array}{ll}
\infty & 0 < r < \sigma \\
\ln{\left[\frac{12\tau(d-\sigma )}{d}\right]} \;\;\; & \sigma < r < d \\
0 & d < r
\end{array}
\label{ghs}
\right.
\end{equation}
where $\tau $ is a reduced temperature and $r$ is the interparticle
separation distance.
The $\tau \rightarrow \infty $ limit of the PY--AHS system corresponds to the
PY hard sphere system. Using this as the starting point,
the earlier MCT result for the hard sphere glass
transition volume fraction $\phi _c =0.516$ \cite{Ben84} was reproduced.
With more accurate hard sphere
static inputs one obtains $\phi _c=0.525$ \cite{Got91,Bar89,Ben86b}.
Experiments locate the
colloidal hard sphere glass
transition at $\phi _c\approx 0.58$ \cite{Pus87,vMe91,vMe94,vMe95},
showing that the idealized MCT prediction for
the hard sphere $\phi _c$ is too low;
this is presumably caused by the strong restriction of the modes
of structural relaxation imposed in the MCT.
The locus of critical glass transition points is shown
in Fig. \ref{PH} as a function of the particle volume fraction $\phi $ and
the reduced temperature $\tau $, with the shaded region
denoting nonergodic glass
states. Upon decreasing $\tau $ the glass
transition point moves along the line B1 in Fig. \ref{PH}
to higher density, contrary to the findings for
particles interacting via a molecular interaction potential of
Lennard--Jones form~\cite{Ben86b}.
We see that starting with a hard
sphere glass and introducing a short--range
attraction leads to an ergodicity restoring transition, provided
the density of the isochore is not too high.
The glass transition
for $\tau =2$ occurs at a volume fraction of $0.5527$, significantly
higher than the MCT hard sphere result.
The strength of attraction needed to shift the
glass transition to higher density
is weak; a second virial coefficient mapping
reveals that a reduced temperature $\tau =2$
corresponds roughly
to a $0.05\sigma $ wide square-well with a depth of $\sim $0.6 $k_BT$.
An examination of the static structure factors along the critical
glass transition boundary B1 in Fig. \ref{PH}
shows that they are not markedly different from those of hard sphere
suspensions; however, subtle changes in the structure lead to
significant changes in the critical glass transition density.
At high temperatures we observe
that the MCT predictions for the localization length $r_s$ follow roughly
a Lindemann criterion, given by the hard sphere result
$r_s \approx 0.074\sigma$, in agreement with
the results of the Lennard--Jones study~\cite{Ben86b}.
Below $\tau \sim 5$, however,
the Lindemann criterion is violated, the particles being now more
strongly localized in the glass state.
We note also that inclusion of improved triplet correlations in the manner of
Barrat et al. \cite{Bar89}, who used an approximation due to
Denton and Ashcroft for $c^{(3)}$ \cite{Den89b},
leads to a qualitatively similar
phase diagram with boundaries shifted slightly
to lower densities and temperatures relative to those shown in Fig. \ref{PH}.
The AHS phase behavior has been the subject of several studies, most
of them using density functional theory
(DFT) \cite{Cer85,Smi85,Zen90,Tej93,Mar93}. Selecting the most
recent one by Marr and Gast \cite{Mar93} for comparison, who used the
modified weighted--density approximation (MWDA) \cite{Den89a}, we find that
the B1 glass transition is confined to the metastable region between
the fluid--solid coexistence lines (see Fig. \ref{PH}).
One striking feature is that the
B1 glass transition line from MCT and the MWDA freezing transition line
track each other, the
quantity $\Delta \phi = \phi _c -\phi _f$, with $\phi _f$ the
volume fraction at freezing, being nearly temperature
independent.
This result has interesting consequences for the diffusion constants
at the freezing densities \cite{Fuc95b,Fuc95a}.
When sufficiently close to the
glass transition the long-time self diffusion coefficient
assumes its asymptotic behavior governed by the distance to
the glass transition singularity. The
normalized long-time self diffusion coefficient
has been found to exhibit universality along the fluid--solid
freezing transition \cite{Low93}.
The comparison made here shows that this condition
may be related to a universality of the proximity of the freezing
transition to the glass transition.
At least, it suggests a deeper
connection between MCT for the liquid--glass transition and DFT,
a topic that has been explored to some extent \cite{Kir87}.
In following the glass transition line from high to low temperatures
(the line denoted by B1 in Fig. \ref{PH}),
we find a line crossing similar to that studied within
schematic ($q$--independent) models \cite{Got91,Fuc91,Got88}.
This region in the phase diagram has been studied in detail recently by
Fabbian et al. \cite{Fab98}.
At the crossing between the B1 and B2 glass transition lines,
it is B2 that determines the behavior of the physical solution
as the nonergodicity parameters associated with
B2 are found to be always greater than those associated with B1.
Thus, along the B2 line bordering the fluid phase,
$f_q$ for each $q$ jumps discontinuously between 0 and finite values,
and between
smaller and larger finite values when the B2 line is traversed in the glass.
The appearance of the B2 line
is a result of an endpoint (cusp, A3)
singularity \cite{Got91,Fab98,Got88,Got89} in the AHS phase diagram,
where three solutions of Eq. \ref{coh} for $f_q$ coalesce.
This singularity appears as the termination point of the B2 transition line.
It is connected to another bifurcation point with triply degenerate
$f_q$ solutions located at lower temperature by the B3 transition line
shown in Fig \ref{PH}. Neither
the low temperature endpoint, the piece of B1 between it and B2,
nor the B3 transition line play a
role in determining the glass dynamics, but show
the connectivity among the bifurcation solutions of Eq. \ref{coh}.
The B2 line in Fig. \ref{PH} exhibits unusual properties.
Varying the numerical parameters $N$ and $\Delta q$, such that
the maximum wavevector $q_{max}=N\Delta q$ changes, shifts the location
of the B2 glass transition line
and the high temperature endpoint in the phase diagram.
Such a variation is not observed in connection with the B1 glass
transition line.
In addition, we were unable to identify a set of $N$ and $\Delta q$
such that the $f_q$ associated with the B2 glass transition
decays to zero within the prescribed wavevector range.
The results shown in Fig. \ref{PH} were obtained using $N=700$ and
$\Delta q\sigma =0.2$.
It is possible that the anomalous behavior of the B2 glass solutions
results from the atypical behavior
of the AHS $S_q$ in the large $q$ limit caused by the singular
nature of the AHS pair potential. The AHS $S_q$ decays slowly
for large $q$ as $S_q \approx 1+2\phi \lambda _{\rm{B}} \sin{(q\sigma )}/q\sigma $,
where $\lambda _{\rm{B}}$ is the solution of Baxter's quadratic equation \cite{Bax68b}.
We do not consider the behavior of the B2 glass solutions here further;
instead, we show in the next section that the HCAY system exhibits
glass transition lines that extend to low temperatures and densities.
Several properties of these low temperature glasses
are in qualitative agreement with experiments on colloidal gels.
\section{Hard core attractive Yukawa system}
\label{HCAYsec}
In this section we examine the effect on the glass transition
of introducing a finite range of attraction via the HCAY system.
The HCAY pair potential is given by
\begin{equation}
u(r)/k_BT = \left\{ \begin{array}{ll}
\infty & 0 < r < \sigma \\
-\frac{{\mbox{\small K}}}{r/\sigma }e^{-{\mbox{\scriptsize b}}(r/\sigma -1)} \;\;\; & \sigma < r
\end{array}
\label{hcay}
\right.
\end{equation}
where the dimensionless parameter K regulates the depth of the attractive
well and the reduced screening parameter b sets the range of the attraction.
Using the MSA static structure factor \cite{Wai73,Hoy77,Cum79} as input,
the MCT was solved for three different screening parameters: b = 7.5,
20, and 30. The progression of the glass transition can be traced from the
(PY) hard sphere limit, corresponding to K=0,
to lower temperatures in terms of
the reduced temperature K$^{-1}$.
In Fig. \ref{PH2} we show the gas--liquid spinodal curves,
studied in detail by Cummings, Smith, and Stell \cite{Cum79,Cum83}.
The critical temperature is sensitive to the range of the attraction,
decreasing with increasing b. The spinodal curves are shown as indicators
of where gas--liquid phase coexistence will occur, should there be a stable
liquid phase.
Included in the diagrams in Fig. \ref{PH2} are the corresponding loci of
glass transition points.
At high temperatures and small values of b (b=7.5) they are relatively
insensitive to the strength of the attraction, showing only a minor initial
movement toward higher densities.
Increasing the value of the screening parameter b (b=20 and 30),
which decreases the range of the attraction,
leads to a small initial increase of the glass transition density upon
lowering the temperature,
although this trend is not as pronounced as in the AHS system;
subsequently, at lower temperatures the glass transition is induced
at increasingly lower densities.
For b=7.5 and 20 the nonergodicity transition lines reach subcritical
temperatures and approach the liquid side of the spinodal.
For b=30 the nonergodicity transition line lies entirely within the fluid phase
above the two phase region, and extends to subcritical temperatures at
low densities.
The MCT used here does not account for
large concentration gradients and additional critical slowing
of relaxations. As there is no small expansion parameter
in MCT, it is difficult to ascertain when, upon approaching a
critical point, this form of MCT should
be replaced by a more complete theory,
including a more sophisticated handling of the critical dynamics
(see ref. \cite{Kaw76} and references therein).
At high temperatures wavevectors around the primary peak of the
structure factor contribute the most to the
mode coupling integrals in Eqs. \ref{coh}
and \ref{incoh}.
With decreasing temperature, on the one hand, the small wavevector
structure in $S_q$ leads to a stronger coupling on large length scales;
on the other hand, the attractive interactions become of increasing importance
on all length scales. The former effect, which can be expected to appear for
all ranges of attractions, leads to nonergodicity transitions
which trace the spinodal lines. These transitions will be discussed in the
appendix as here the present MCT is least reliable because it does not
include all relevant mode couplings and will not correctly describe
the dynamics near the critical points \cite{Kaw76}.
The latter effect, important for systems with short--range attractions,
can be studied by an asymptotic analysis of the
MCT equations and will be seen to dominate the low density glass transitions
for large values of b.
At low densities and in the limit of strong attractive interactions,
the Ornstein--Zernike direct correlation function becomes independent of
density. Specifying this to the MSA
of the HCAY system, this limit corresponds to $\phi \to 0$ and K $\to \infty$.
Considering the asymptotic limit
\begin{equation}
\phi \to 0 \; \quad\mbox{and }\;\;\; \rm{K}\to \infty \; , \quad\mbox{so that }\;
\Gamma = \frac{\rm{K}^2\phi }{\rm{b}} = \mbox{constant}\; ,
\label{asymp}
\end{equation}
the MCT equations simplify because $S_q\to 1$ follows.
The nonergodicity transitions then occur at $\Gamma =\Gamma _c(\rm{b})$,
leading to the asymptotic prediction $\rm{K}_c \propto 1/\sqrt{\phi _c}$.
For short--range attractions, in the limit of $\rm{b}\to \infty $, a further
simplification arises because the coupling constant $\Gamma $ approaches
a unique value at the transition, $\Gamma _c \to 3.02$ for $\rm{b}\to \infty $,
and the nonergodicity parameters now depend only on
the rescaled wavevector $\tilde{q}=q\sigma /\rm{b}$:
$f^c_q\to \tilde{f}^c(\tilde{q})$.
The asymptotic transition lines are shown in Fig. \ref{PH2} as
the chain curves.
We find excellent agreement with the MCT transition line for b=30 at low density,
demonstrating that the low density nonergodicity transitions are
not driven by the divergence of the small wavevector
limit of $S_q$. Moreover, the asymptotic model is seen to capture the
behavior of the full MCT transition lines -- where present --
qualitatively and even semi--quantitatively at higher densities.
We further point out that Eq. \ref{incoh} for the single particle
dynamics and, thus, the incoherent form factors $f_q^s$ are
not dominated by small wavevector variations in the static structure factor.
Instead, $f_q^s$ and the localization length $r_s$ are dominated by
the small distance or large wavevector behavior of the liquid structure.
In the asymptotic limit of Eq. \ref{asymp} this also holds for the collective
particle dynamics and $\tilde{f}^c(\tilde{q})=\tilde{f}^s(\tilde{q})$ is
obtained, where both functions show rather large non--Gaussian
corrections.
For the system with b=20, Fig. \ref{PH2} shows that the glass transition
nearly meets the critical point (see also Fig. \ref{PH3} in the appendix).
This aspect is in qualitative agreement with the behavior of the
sterically stabilized suspensions studied by Verduin and Dhont \cite{Ver95}.
They observed a gel transition that traversed the phase diagram
from high density and temperature to
the critical point. The transition, which they refer to as a
static percolation transition, was associated with a non-decaying
intermediate scattering function and non-fluctuating dynamic light scattering
speckle patterns. Thus, it has the expected properties of the
ergodic--nonergodic dynamic transition predicted by the idealized MCT.
Moreover, they were able to follow the transition into the unstable
region inside the spinodal curve, showing that complete phase separation
does not occur because of interference from the
gel transition. We cannot extend the calculation of the
MCT gel transition line into the unstable region because an
appropriate $S_q$ is not available and the theory assumes closeness
to equilibrium \cite{Got92}.
Restricting the range of the attraction sufficiently, as for b=30,
Fig. \ref{PH2} shows that the glass transition line passes above the
critical point and reaches subcritical temperatures on the gas side of
the metastable spinodal. For such systems we may speculate that the
glass transition renders the entire spinodal curve and the
liquid phase dynamically inaccessible.
This feature appears to agree with some measurements on
sterically stabilized suspensions \cite{Che91,Gra93,Rue98}, in
which only a liquid-gel transition was observed.
Recent measurements by Poon et al. \cite{Poo97b}
suggest that the colloid--polymer
mixtures with a small polymer/colloid size ratio ($\xi \approx 0.08$)
may belong to the class of HCAY phase diagrams with
b $ < 20$, where the nonergodicity transition line meets the spinodal on the
liquid side.
This interpretation includes a possible explanation
for the growth of the small-angle scattering peak for samples with
low colloid concentrations \cite{Poo97b}, and
that the denser colloid domains arrest in the transient gelation region.
To further clarify the physical mechanism of the gel transition
and the properties of the gel states, various aspects of the solutions
of the MCT will be discussed for the three
cases: b=7.5, 20, and 30.
In Fig. \ref{FK} we show the evolution of the coherent
nonergodicity parameter $f_q$ along the critical glass transition
boundary corresponding to b=30 in Fig. \ref{PH2}.
As seen, the width of $f_q$
increases with decreasing temperature (increasing K).
This behavior of $f_q$ with decreasing temperature is a result
of a corresponding increase in the range of $S_q$, which results
from particles being strongly correlated near contact, i.e. due to particle
clustering. For longer range attractions, like the b=7.5 system, the
width of $f_q$ and $f_q^s$ remain essentially unchanged
along the glass transition
boundary, which reflects the lower degree of
particle clustering in this system.
Note that $f_q$ becomes a density and
temperature independent function, $f_q \to \tilde{f}(q\sigma /\rm{b})$, in the
limit of strong attractions. This prediction is shown in Fig. \ref{FK}
as the bold line, and agrees almost quantitatively with the full MCT $f_q$ solutions
for large values of K.
The localization length $r_s$
in the glass decreases along the glass transition boundary when the
attraction is sufficiently short range.
This decrease in the localization length is shown in Fig. \ref{L},
and is caused by
the increased contribution from large wavevectors in the MCT integrals in
Eqs. \ref{coh} and \ref{incoh}. For longer range attractions,
like the b=7.5 case, the localization length stays close to the
value dictated by the Lindemann criterion and found at the glass transition
in the hard sphere system \cite{Ben84,Ben86b}. Thus, for short--range
attractions the particles are more strongly localized in the glass
than for systems with somewhat longer range attractions.
At low temperatures the localization length saturates at a limiting value
which is inversely proportional to b because of Eq. \ref{asymp},
which leads to the prediction $r_s \to 0.91\sigma /\rm{b}$ for b $\to \infty $.
In addition to an increased width at low temperatures,
the small wavevector behavior of $f_q$ changes
dramatically, such that at low temperatures the intermediate
scattering function for small $q$ practically does not decay with time at all
(see Fig. \ref{FK}).
This indicates that large scale assemblies of particles behave
essentially as static objects, where the single particles are
tightly bound to the particle clusters (see Fig. \ref{L}).
The asymptotic limit, Eq. \ref{asymp} and b $\to \infty $, which
results in $\tilde{f}(\tilde{q})-1 \propto \tilde{q}^2$, stresses that
this is caused by the short--range attraction. Such a rise in $f_q$ for small $q$
is observed also in the b=7.5 system, where it is
caused by a different mechanism, namely the increase in the isothermal
compressibility on approaching
the gas--liquid spinodal. There, this leads to coherent nonergodicity parameters which are
essentially hard sphere--like except for a large $q\to 0$ value.
In Fig. \ref{MOD} we show the zero-frequency elastic shear modulus as a function
of the reduced temperature along the glass transition lines in Fig. \ref{PH2}.
When the range of attraction is comparatively large (b=7.5)
the shear modulus remains
constant at the hard sphere value, even for suspensions near the
spinodal. This illustrates that the shear modulus,
like the localization length, is
determined by the large wavevector behavior of the liquid structure $S_q$, and
is unaffected by long-wavelength density fluctuations in our calculations.
For shorter range
attractions (b=20 and 30), the shear modulus
is dominated by particle clustering; it increases strongly with decreasing temperature
because of the stronger binding among particles, eventually showing a maximum
for suspensions close to the bend in the K$_c$ versus $\phi _c$ curves,
where $\phi _c$ begins to decrease strongly with decreasing temperature.
At lower density the shear modulus becomes linear in the density at the transition,
according to $G \propto \rm{K}^2_c\phi _c/\rm{b}$, as predicted by
the asymptotic solution in Eq. \ref{asymp},
and as observed in the dilute limit of the b=30 system.
These results show that low temperature nonergodic structures,
proposed to be colloidal gels here, are distinct from colloidal glasses in
that they generally display a larger static shear modulus and
strongly localized particles bound in clusters.
To more clearly connect this study of the low temperature
behavior of the glass transition to the experimental studies
of the gel transition, we have calculated the intermediate scattering function
upon approaching the glass transition at fixed volume fraction.
This mimics the Verduin and Dhont study \cite{Ver95}, in which they performed
low--$q$ dynamic light scattering experiments
on a series of suspensions at fixed volume fraction close to the gel transition.
As noted already, the HCAY system with b=20 exhibits a glass
transition line that nearly meets the critical point.
This qualitative aspect is shared
with the experimental phase diagram determined by
Verduin and Dhont.
We have selected four suspensions with b=20 and $\phi =0.4$
at different reduced temperatures (shown as open circles
in Fig. \ref{PH2}), such that the suspension with the
lowest temperature (K=12) is located in the glass.
The resulting normalized intermediate scattering functions corresponding to
these suspensions are displayed in Fig. \ref{DLS} for a fixed wavevector
$q\sigma $=0.2, the same wavevector as that used in the measurements
by Verduin and Dhont.
As the normalized intermediate scattering function is the quantity that one
measures in dynamic light scattering experiments,
we can compare Fig. \ref{DLS}
with the results of Verduin and Dhont (see their Fig. 11).
This comparison shows excellent
qualitative agreement between their dynamic light scattering data
and our calculated $F_q(t)/S_q$.
Away from the transition the decay of $F_q(t)/S_q$ is approximately
exponential for this wavevector.
The decay becomes slower when the temperature is decreased until
$K=12$, when $F_q(t)/S_q$ no longer decays to zero. Instead,
a long-time plateau with a value near unity is obtained,
which corresponds to the
nonergodicity parameter $f_q$ at $q\sigma =0.2$.
Note that additional incoherently scattered light due to particle
size polydispersity \cite{Ver95} may contribute appreciably and
cause $f_q$ to attain such a large value. Nevertheless, the dynamical
arrest of $F_q(t)/S_q$ agrees with our proposed ergodic--nonergodic
transition for the gel transition and is captured by the
idealized MCT.
As has been shown in the past,
the range of the colloidal
attraction dictates where the fluid--solid freezing transition
passes through the phase diagram and whether there is a stable
liquid phase \cite{Gas83,Gas86,Tej94,Hag94,Lek95}.
In the same manner, Fig. \ref{PH2} shows that
the range of the attraction determines
how the glass transition traverses the phase diagram relative
to the critical point.
We have not compared the
diagrams in Fig. \ref{PH2} with results for the fluid--solid and gas--liquid
transitions (see e.g., \cite{Hag94,Ren91,Has98}).
The MCT relies on the static structure factor input, which was
provided using the MSA.
For short--range
attractions the MSA produces relatively poor
structural and thermodynamic predictions.
Thus, a fair comparison should be made
with a theory, e.g., MWDA \cite{Den89a},
which can use the same input as that supplied to the MCT.
Alternatively, the MCT can be solved using a more accurate static
structure factor, such as that from HMSA theory \cite{Zer86}.
This would enable a determination of the location of the glass
transition relative to that of the fluid--solid freezing transition,
testing the conjecture made here that the glass transition line
tracks the freezing transition at higher density in the phase diagram.
\section{Discussion and Conclusions}
\label{Discsec}
The idealized MCT has been shown to provide a possible
explanation for important aspects of the colloidal gel transition.
In this scenario the arrest of the dynamics during
the gel transition is caused by a low temperature
liquid--glass transition.
The underlying phenomenon is a breaking of ergodicity,
caused by long-time structural arrest.
It is accompanied by the cessation of hydrodynamic diffusion and
the appearance of relatively large finite elastic moduli as
the particles are tightly localized in ramified clusters.
The AHS system was found to have endpoint singularities in the phase diagram.
The spectacular dynamics close to the MCT endpoint singularity
has been the focus of a recent study \cite{Fab98}.
However, the glass transitions that occur at low temperatures in the AHS
system are accompanied by numerical difficulties,
resulting from the singular nature of the AHS pair interaction potential.
Nevertheless, the AHS system provides insight into the
temperature dependence of the glass transition in systems with weak
short--range attractions.
Subtle changes in the structure caused by the attractions
lead to an initial increase
in the glass transition density with decreasing temperature.
The particles forming the glassy cage tend to stick together, thereby
creating openings in the collective cage around a central particle
which have to be filled by increasing the critical colloid density.
We suggest that the recrystallization of glass samples at high
densities upon addition of short polymers, reported in \cite{Poo93},
is explained by this shift of the glass transition density to higher
values. Moreover, the MCT glass transition line was observed to lie parallel
to the DFT fluid--solid freezing transition at high temperatures
in the AHS phase diagram.
Introduction of a finite range of attraction and replacement of
the PY theory with the MSA via the HCAY system,
yields glass transition lines that
extend to low temperatures in the phase diagram.
For HCAY systems with moderate--range attractions
the glass transition
line crosses the liquid binodal. When the range of the attraction is
further restricted the glass transition line passes above the
critical point, likely rendering
part of the (metastable) equilibrium phase diagram irrelevant.
Preliminary solutions of the dynamical MCT equations for
the b = 30 HCAY system indicate that nearby $A_l$ singularities
with $l > 2$ appreciably distort the time dependent structural correlators in
the intermediate time windows, even though no $A_3$ singularity
\cite{Got91,Got92,Fab98,Got89} could be found in the phase diagram.
The nonergodicity transitions of the HCAY system are influenced by
two mechanisms which are absent or not dominant in
the hard sphere and
Lennard--Jones \cite{Ben86b} systems.
In the latter two, the temperature dependence of the
critical density $\rho _c$ is either trivially
absent or arises from the soft repulsive part of the pair interaction
potential. Along the MCT transition line in the Lennard--Jones liquid
the temperature dependent packing fraction $\phi (\rho, T)$,
resulting from the effective excluded volume diameter
$\sigma = \sigma ^{\rm eff}(T)$ \cite{Wee71}, is roughly constant
$\phi (\rho,T) \approx 0.52$ and approximately equal to its hard sphere
value \cite{Ben86b}. As the soft repulsion of the Lennard--Jones system
leads to $\sigma ^{\rm eff} \propto T^{-1/12}$, the critical density
smoothly decreases with temperature \cite{Nau97}.
Note that this observation indicates that the nonergodicity transitions
resulting from the solution of Eq. \ref{coh} for the Lennard--Jones system
are dominated by the excluded volume effect, i.e. the
primary peak of the structure factor $S_q$, as is
also the case for the hard sphere system.
Because the idealized MCT has been developed from approximations
aimed at describing the connected physical mechanism, called cage--
or back--flow effect, the quality of the mode coupling approximation
is expected to be unaffected by temperature changes as long as these
excluded volume effects dominate in Eqs. \ref{coh} and \ref{incoh}.
The nonergodicity transition lines of the HCAY system on the other hand
are additionally affected by the low wavevector fluctuations in the
fluid structure factor, $S_q$ for $q\to 0$, and by the increase
in $S_q$ at large wavevectors arising from the short--range nature of the
attraction.
The first aspect, which also occurs for longer range attractions,
leads to nonergodicty transitions tracking the spinodal curve (see appendix).
The short--range nature of the attraction causes the stronger localization
of the particles, i.e. the shorter localization length $r_s$, upon
decreasing the temperature. Also the strong increase in the elastic shear
moduli along the nonergodicity transition line occurs only for sufficiently
short--range attractions as shown in Fig. \ref{MOD}. Again, the ability
of the MCT Eqs. \ref{coh} and \ref{incoh} to describe such local
interparticle correlations is not known. Note, however, that the
HCAY results are independent of the numerical parameters chosen.
Clearly, theories aimed at long wavelength phenomena at the gel
transition cannot incorporate these variations of the elastic modulus
as described by the MCT because it arises from local potential
energy considerations.
The asymptotic model, defined by Eq. \ref{asymp} (and b $\to \infty $),
which highlights the effects of strong short--range attractions,
captures all aspects of the low density MCT nonergodicity transitions
qualitatively and even semi--quantitatively. Furthermore, it clearly
demonstrates that the gel transitions are not driven by long--range
structural correlations. It can be expected that such an asymptotic
model can be found for other theories of liquid structure with strong
short--range attractive potentials, but the detailed predictions presented
here rest upon the use of the MSA for the HCAY system.
Based on this suggested interpretation of the MCT nonergodicity
transitions,
several features of the computed HCAY density--temperature
diagrams agree qualitatively with experimental
observations made on colloid--polymer mixtures and sterically stabilized
suspensions \cite{Pat89,Emm90,Poo93,Ile95,Ver97,Poo97b,Jan86,Che91,Gra93,Rue97,Rue98}.
First, the gel transitions appear to lie at lower temperatures
than, but otherwise track, the freezing line when present.
Second, for short--range attractions the gel transition can shift
to comparable or higher temperatures than those required for gas--liquid
phase separation. Third, the gel line does not show such a strong density
dependence as the static percolation transition.
We emphasize that this suggested interpretation of the colloidal gel
transitions is based on an extension of the idealized MCT of the glass
transition beyond the range its approximations were aimed at.
Our speculation, however, can be decisively tested by dynamic light
scattering experiments. Nonergodicity transitions within the MCT
exhibit universal dynamical properties
\cite{Got91,Got92,Got89,Got85}, which for example led to the identification
of the colloidal hard sphere glass transition by van Megen and
coworkers \cite{Pus87,vMe91,vMe94,vMe95}.
As more complicated
bifurcation scenarios, $A_l$ with $l>2$ \cite{Got91,Got92,Fab98,Got89},
can be expected, the
dynamics at the gel transitions should be very nonexponential and anomalous.
Moreover, the short--range attractions lead to couplings among more wavevector--modes,
as can be seen from the asymptotic model defined by Eq. \ref{asymp},
resulting in an MCT
exponent parameter (see ref. \cite{Got91,Got85} for a
definition and details on its calculation)
$\lambda = 0.89$ for b $\to \infty $, considerably
larger than values found for systems not characterized by short--range
attractions (see e.g., \cite{Fab98,Bar89,Nau97}).
The proposed connection between the MCT nonergodicity
transitions in the HCAY system and the non-equilibrium
transitions in colloidal suspensions is further supported
by the following observations.
For moderate--range attractions, like the b=7.5 diagram in
Fig. \ref{PH2}, the coexistence region can
be tentatively divided into three regions.
For somewhat lower temperatures than the critical temperature,
gas--liquid phase separation occurs, provided
a thermodynamically stable liquid phase exists.
For temperatures (just) below the triple point temperature,
gas--crystal phase separation takes place. Decreasing
K$^{-1}$ still more, gas--glass coexistence may be expected if --- as argued
from computer simulations \cite{tWo97} --- the way to crystallization
proceeds via the initial formation of a liquid droplet,
whose density lies above the nonergodicity transition line.
As the glass states for this system
are rather close to the ideal hard sphere glass state,
we expect signatures of this well studied transition to
be observed \cite{Got91b,Got92,Pus87,vMe91,vMe94,vMe95}.
We speculate that the vanishing of the homogeneously nucleated
crystallites in the colloid--polymer systems upon addition of sufficient
large molecular weight polymer, observed in \cite{Ile95},
signals the presence of a nonergodicity
transition as found in the colloidal hard sphere system
\cite{vMe95,vMe93,Har97}.
This suggestion can be tested by studying the
dynamic density fluctuations close to the transition
as has been demonstrated in the hard sphere system
\cite{Got91b,Got92,Pus87,vMe91,vMe94,vMe95}.
For suspensions with short--range attractions, like the
b=20-- or b=30--curve in Fig. \ref{PH2},
it seems possible that the long--range density fluctuations,
likely induced by the hidden critical point, become arrested
when the denser domains of the system cross
the MCT nonergodicity transition line.
Nonergodic gel states characterized by large small-wavevector
form factors $f_q$, rather short localization lengths,
and finite, rather large elastic moduli can be expected.
We suggest that these nonergodicity transitions cause
the gel transitions observed in the colloid--polymer mixtures
and sterically stabilized suspensions, and anticipate that
they may also play a role in other colloidal systems, such as
emulsions \cite{Bib92}, emulsion--polymer mixtures \cite{Mel98},
and suspensions of globular proteins \cite{Mus97,Geo94,Ros95},
in which short--range attractions also dominate.
We caution again that a proper extension of the MCT used here to
include a full description of the critical dynamics close
to critical points has yet to be formulated.
Experimental tests of the dynamics close to the gel transitions would
be required to test our suggestion.
\acknowledgments
J. B. acknowledges financial support from the National Science Foundation
(Grant No. INT-9600329)
and the kind hospitality of Professor R. Klein
(Universit\"at Konstanz).
The work was further supported by the Deutsche Forschungsgemeinschaft (DFG)
(Grant No. Fu309/2-1).
Useful discussions with G. N\"agele and N.~J. Wagner are
acknowledged. We also thank Fabbian et al. \cite{Fab98}
for making their work available to us prior to publication.
This work was conducted independently of theirs.
|
\section{Introduction}
\setcounter{equation}{0}
\label{secintro}
Cosmic strings \cite{vilsh,KibbleH} were introduced into cosmology by
Kibble \cite{Kibble1}, Zel'dovich \cite{Zel} and Vilenkin \cite{Vil1} as
(linear) topological defects, which may have been formed during phase
transitions in the early universe. Cosmic strings are considered as possible
sources for density perturbations and hence for structure formation in the
universe.
The simplest and most common field-theoretical model, which is used in order to
describe the generation of cosmic strings during a phase transition, is the
Abelian Higgs model. This model is known to have (magnetic) flux tube solutions
\cite{NO}, whose gravitational field is represented by an asymptotically conic
geometry. These are the local cosmic string solutions.
This conic space-time is a special case of the general static and
cylindrically-symmetric vacuum solution of Einstein equations
\cite{exact-sol}, the so-called Kasner solution:
\begin{eqnarray}
ds^{2} = (kr)^{2a}dt^{2} - (kr)^{2c}dz^{2} - dr^{2}
- \beta^{2} (kr)^{2(b-1)}r^{2}d{\phi}^2
\label{Kasner1}
\end{eqnarray}
where $k$ sets the length scale while $\beta$ represents the asymptotic
structure, as will
be discussed below. Notice also that $(a, b, c)$ must satisfy the Kasner
conditions:
\begin{equation}
a + b + c=a^2 + b^2 + c^2=1
\label{Kasner2}
\end{equation}
The standard conic cosmic string solution \cite{Vil1,Garfinkle1} is
characterized
by an
asymptotic behavior given by (\ref{Kasner1}) with $a=c=0,\; b=1$ which is
evidently locally flat. In this case, the parameter $\beta$ represents a
conic angular
deficit \cite{Marder2,Bonnor}, which is also related to the mass distribution
of the source.
A first approximation to the relation between the
angular deficit $\delta\phi = 2 \pi (1-\beta)$ and the ``inertial mass" (per
unit length) $\tilde{m}$, of a local string was found
to be \cite{Vil1,Gott,Hiscock,Linet1}:
\begin{equation}
\delta\phi = 8 \pi G\tilde{m}
\label{angdef}
\end{equation}
Further corrections to (\ref{angdef}) were calculated in the following papers
\cite{LagMatz,GarfinkleLag,LagGarfinkle,CNV,Ver}.
\vskip 6pt
In order to fully analyse the string-like solutions of the Abelian Higgs
system one
needs to solve the full system of coupled field equations for the
gravitational
field and matter fields (scalar + vector). But this is not necessary in
order to
obtain the asymptotic geometry. It is sufficient to note that for cylindrical
symmetry the flux tube has the property of ${\cal T}^0_0={\cal T}^{z}_{z}$,
where
${\cal T}^\mu_\nu$ is the energy-momentum tensor. This means that the
solution will
have a symmetry under boosts along the string axis, i.e., $a=c$. The Kasner
conditions (\ref{Kasner2}) then leave only two options; either the locally
flat
case:
\begin{equation}
a=c=0 \;\;\; , \;\;\; b=1
\label{SL}
\end{equation}
which we shall refer to as the cosmic string branch, or
\begin{equation}
a=c=2/3\;\;\; ,\;\;\;b=-1/3
\label{MW}
\end{equation}
which is the same behavior as that
of the Melvin solution \cite{Melvin}. We shall therefore refer to (\ref{MW}) as
the Melvin branch.
The Melvin branch does not provide
the same characteristics used to describe a ``standard" cosmic
string, and thus it has been
disregarded in previous investigations.
However, the Melvin-like solution seems completely well-behaved and one may
wonder whether it exists at all (asymptotically), or maybe it is just an
artifact of the too general reasoning above. Although several authors
\cite{CNV,Ver,Ortiz,FIU,Rayc} explicitly mention this possibility of a second
Melvin branch of solutions in the Abelian Higgs system, it has never been
properly discussed.
In this paper we show that indeed the Melvin branch actually exists
in the Abelian Higgs model. It turns out that each conic cosmic string
solution has a ``shadow" in the form of a corresponding solution in the Melvin
branch.
These two solutions, conic cosmic string and asymptotic Melvin, are found
only in
a part of the
two dimensional parameter space of the system, bounded by the curve of
maximal angular deficit of
$2\pi$.
There also seem to be some open questions in the literature
\cite{LagGarfinkle,Ortiz} about
the nature of the solutions beyond the border of maximal angular deficit of
$2\pi$. We will see that also in this region two types
of solutions coexist, both of them with a finite radial extension.\\
\section{General analysis of the Abelian Higgs system}
\setcounter{equation}{0}
\label{secgeneral}
The action of the gravitating Abelian Higgs model is:
\begin{equation}
S = \int d^4 x \sqrt{\mid g\mid } \left({1\over 2}D_{\mu}\Phi ^
{\ast}D^{\mu}\Phi
- {{\lambda }\over 4}(\Phi ^{\ast} \Phi - v^2)^2 -
{1\over 4}{F}_{\mu \nu}{F}^{\mu \nu} + \frac{1}{16\pi G}
{\cal R}\right)
\label{higgsaction}
\end{equation}
where ${\cal R}$ is the Ricci scalar, ${F}_{\mu \nu}$ the Abelian field
strength, $\Phi$ is a complex scalar field with vacuum expectation value $v$
and $D_\mu = \nabla _{\mu} - ieA_{\mu}$ is the usual gauge covariant
derivative. We use units in which $\hbar=c=1$.
Because of the cylindrical symmetry of the source, we will use a line element
of the form:
\begin{equation}
ds^{2} = N^{2}(r)dt^{2} - d{r}^{2} - L^{2}(r)d{\phi}^2 - K^{2}(r)dz^{2}
\label{lineelement}
\end{equation}
and the usual Nielsen-Olesen ansatz for the +1 flux unit:
\begin{eqnarray}
\Phi=vf(r)e^{i\phi} \hspace{10 mm} , \hspace{10 mm}
A_\mu dx^\mu = {1\over e}(1-P(r))d\phi
\label{NOansatz}
\end{eqnarray}
This gives rise to the following field equations for the Abelian Higgs flux
tube:
\begin{eqnarray}
\frac{(NKLf')'}{NKL} + \left(\lambda v^2 (1-f^2) -
\frac{P^2}{L^2}\right)f = 0 \\
\frac{L}{NK}\left(\frac{NK}{L} P'\right)' - e^2 v^2 f^2 P = 0
\label{fluxtube}
\end{eqnarray}
With the line element (\ref{lineelement}), the components of the Ricci tensor
are:
\begin{eqnarray}
{\cal R}^{0}_{0} = - \frac{(LKN')'}{NLK} & , &
{\cal R}^{r}_{r} = - \frac{N''}{N} - \frac{L''}{L} - \frac{K''}{K}
\nonumber \\
{\cal R}^{\phi}_{\phi} = - \frac{(NKL')'}{NLK} &,&
{\cal R}^{z}_{z} = - \frac{(NLK')'}{NLK}
\label{Ricci}
\end{eqnarray}
The source is described by the energy-momentum tensor with the
following components:
\begin{eqnarray}
{\cal T}^{0}_{0} &=& \rho = \varepsilon _s +\varepsilon _v +
\varepsilon _{w} + u \nonumber \\
{\cal T}^{r}_{r} &=& -p_r = -\varepsilon _s -\varepsilon _v +
\varepsilon _{w} + u \nonumber \\
{\cal T}^{\phi}_{\phi} &=& -p_{\phi} = \varepsilon _s -
\varepsilon _v -\varepsilon_{w} + u \nonumber \\
{\cal T}^{z}_{z} &=& -p_z = \rho
\label{Tmunu}
\end{eqnarray}
where:
\begin{eqnarray}
\varepsilon_s = {v^2 \over 2} f'^2 \hspace{5 mm},\hspace{5 mm}
\varepsilon_v = \frac{P'^2}{2e^2 L^2} \hspace{5 mm},\hspace{5 mm}
\varepsilon_{w} = \frac{v^2 P^2 f^2}{2L^2} \hspace{5 mm},\hspace{5 mm}
u = \frac{\lambda v^4}{4} (1-f^2)^2
\label{densities}
\end{eqnarray}
It turns out to be convenient to use Einstein equations in the form (${\cal
T}\equiv {\cal
T}^{\lambda}_{\lambda}$):
\begin{equation}
{\cal R}_{\mu\nu} = -8 \pi G({\cal T}_{\mu \nu} -
\frac{1}{2} g_{\mu \nu}{\cal T})
\label{EinsteinR}
\end{equation}
from which one obtains:
\begin{eqnarray}
\frac{(LKN')'}{NLK} &=& 4 \pi G(\rho +p_r+p_{\phi} +p_z)=
8 \pi G(\varepsilon_v - u)
\label{Einst0}
\end{eqnarray}
\begin{eqnarray}
\frac{(NKL')'}{NLK} &=& -4 \pi G(\rho -p_r+p_{\phi} -p_z)
= - 8 \pi G(\varepsilon_v + 2 \varepsilon_{w} + u)
\label{Einstphi}
\end{eqnarray}
\begin{eqnarray}
\frac{(LNK')'}{NLK} &=& -4 \pi G(\rho -p_r-p_{\phi} +p_z)=
8 \pi G(\varepsilon_v - u)
\label{Einst3}
\end{eqnarray}
and instead of the ``radial" part of (\ref{EinsteinR}), we take the following
combination:
\begin{eqnarray}
\frac{N'}{N} \frac{L'}{L} + \frac{L'}{L} \frac{K'}{K}+
\frac{K'}{K} \frac{N'}{N} = 8 \pi G p_r =
8 \pi G(\varepsilon _s +\varepsilon _v - \varepsilon _{w} - u)
\label{constraint}
\end{eqnarray}
which is not an independent equation but serves as a constraint. In vacuum, the
right-hand-sides of these equations vanish, and the first three of them are
trivially integrated. In this way we may get back Kasner's line element
(\ref{Kasner1}). This is therefore the asymptotic form of the metric tensor
around any (transversally) localized source and especially around an Abelian
Higgs flux tube.
Moreover, it is easy to get convinced that due to the symmetry under boosts
along the string axis, $K=N$.
The equations become more transparent if we express all lengths in terms of the
scalar characteristic length scale $1/\sqrt{\lambda v^2}$ (the ``correlation
length" in the superconductivity terminology). We therefore change to the
dimensionless length
coordinate $x=\sqrt{\lambda v^2}r$ and we introduce the metric component
$L(x)=\sqrt{\lambda
v^2}L(r)$. We also introduce the two parameters
$\alpha=e^2/\lambda$ and $\gamma=8\pi Gv^2$. In terms of these new
quantities, we
get a two parameter system of four coupled non-linear ordinary differential
equations (the prime now denotes $d/dx$):
\begin{eqnarray}
\frac{(N^2 Lf')'}{N^2 L} + \left(1-f^2 - \frac{P^2}{L^2}\right)f = 0
\label{systemNO1}\\
\frac{L}{N^2}\left(\frac{N^2 P'}{L}\right)' - \alpha f^2 P = 0
\label{systemNO2}
\end{eqnarray}
\begin{eqnarray}
\frac{(LNN')'}{N^2 L} &=&\gamma\left(\frac{P'^2}{2\alpha L^2} - \frac{1}{4}
(1-f^2)^2\right)
\label{systemE1}\\
\frac{(N^2 L')'}{N^2 L}
&=& - \gamma\left(\frac{P'^2}{2\alpha L^2} + \frac{P^2 f^2}{L^2} + \frac{1}{4}
(1-f^2)^2\right)
\label{systemE2}
\end{eqnarray}
We have also to keep in mind the existence of the constraint
(\ref{constraint}),
which gets the
following form:
\begin{eqnarray}
\frac{N'}{N} \left(2\frac{L'}{L} + \frac{N'}{N}\right) =
\gamma\left(\frac{f'^2}{2} + \frac{P'^2}{2\alpha L^2} -
\frac{P^2 f^2}{2L^2} - \frac{1}{4}(1-f^2)^2 \right)
\label{constraintmod}
\end{eqnarray}
In order to get string-like solutions, the scalar and gauge fields should
satisfy
the following boundary conditions:
\begin{eqnarray}
f(0)=0 &, & \lim_{x\rightarrow \infty} f(x) = 1 \nonumber \\
P(0)=1 &, & \lim_{x\rightarrow \infty} P(x) = 0
\label{boundarycond}
\end{eqnarray}
Moreover, regularity of the geometry on the symmetry axis $x=0$ will be
guaranteed by the
``initial conditions":
\begin{eqnarray}
L(0)=0 &, & L'(0) = 1 \nonumber \\
N(0) = 1 &, & N'(0) = 0
\label{initcond}
\end{eqnarray}
The purpose of the present paper is to map the two dimensional
$\alpha$-$\gamma$
parameter space, and thereby to classify all the string-like solutions of the
Abelian
Higgs system.
It is well-known that, even in flat space, the field equations can only be
solved numerically. However, much can be said about the nature of the solutions
even without explicitly solving the field equations
\cite{Garfinkle1,LagGarfinkle,CNV,FIU}.
The standard Abelian Higgs
string solution, usually considered in the literature, has a vanishing
gravitational mass. This
means that the spacetime around the string is locally flat except in the
core of the string, while
there is a non-trivial global effect namely a conical structure of the
space which is quantified by
an angular deficit. However, this does not at all saturate all the
possibilities of solutions of
(\ref{systemNO1})-(\ref{systemE2}). There are further types of solutions
with the same
boundary and ``initial" conditions, (\ref{boundarycond})-(\ref{initcond}),
which
are not asymptotically flat but have interesting physical interpretations.
In this paper, we
will show that a point in the $\alpha$-$\gamma$ plane always represents two
solutions, except at the curve representing angular deficit of $2 \pi$.
The various solutions are distinguished by, among other things,
their asymptotic geometries. \\
For analysing the solutions and obtaining the above-mentioned features and some
additional ones, we
introduce the Tolman mass (per unit length), $M$:
\begin{equation}
GM=2\pi G\int _{0}^\infty dr\; N^2 L(\rho +p_r+p_{\phi}
+p_z)=\frac{\gamma}{2} \int _{0}^\infty
dx\; N^2 L \left(\frac{P'^2}{2\alpha L^2} - \frac{1}{4} (1-f^2)^2\right)
\label{mass}
\end{equation}
Using the field equations, one can show that \cite{Ver}:
\begin{equation}
GM=\frac{1}{2} \lim_{x\rightarrow\infty} (LNN')
\label{tolmanmass}
\end{equation}
We also define the magnetic field $B$:
\begin{equation}
B = -\frac{1}{eL(r)}\frac{dP(r)}{dr}=-\frac{\gamma}{\alpha}\left(
\frac{e}{8\pi
G}\right)
\frac{P'(x)}{L(x)}
\label{magn}
\end{equation}
and the dimensionless parameter ${\cal B}=8\pi GB(0)/e$.
We find that the central value of the magnetic field (its value in the core
of the
string) can be expressed as \cite{Ver}:
\begin{equation}
{\cal B}= 1+2GM-\lim_{x\rightarrow\infty}(N^2 L')
\label{centralmag}
\end{equation}
The asymptotic form of the metric tensor is easily found by direct integration
of the two Einstein equations (\ref{systemE1})-(\ref{systemE2}) using the
boundary conditions and the definitions of $M$ and ${\cal B}$. It is of
the Kasner
form
\begin{eqnarray}
N(x)=K(x)\sim \kappa x^a \hspace{6 mm},\hspace{6 mm} L(x)\sim \beta x^b
\label{asymptKasner1}
\end{eqnarray}
with:
\begin{eqnarray}
a=\frac{2GM}{6GM+1-{\cal B}} \hspace{6 mm},\hspace{6 mm}
b=\frac{2GM+1-{\cal B}}{6GM+1-{\cal B}} \hspace{6 mm},\hspace{6 mm}
\kappa ^2\beta=6GM+1-{\cal B}
\label{asymptKasner2}
\end{eqnarray}
The constant $\kappa$ appears free in the asymptotic solution, but it is
uniquely fixed in the complete one by the boundary conditions on the metric.
Due to (\ref{constraintmod}) we get the following relation
which is equivalent to the quadratic Kasner condition in Eq. (\ref{Kasner2}):
\begin{equation}
GM(3GM+1-{\cal B})=0
\label{MB}
\end{equation}
This immidiately points to the possibility of two branches of solutions
corresponding
to the vanishing of either factor in Eq. (\ref{MB}). These two possibilities of
course represent the two branches already discussed in Section 1.\\
Consider first the cosmic string branch ($a=c=0,\; b=1$) in a little more
detail. In this case, we find a simple physical interpretation to the
parameters
in Eq. (\ref{asymptKasner1}).
the constant $\beta=L'(\infty)$ defines the deficit angle
$\delta\phi=2\pi(1-\beta)$,
while the constant $\kappa=N(\infty)$ is the red/blue shift of
time between infinity and
the string core. Moreover, relations (\ref{tolmanmass}), (\ref{centralmag})
give:
\begin{equation}
M=0
\end{equation}
\begin{equation}
{\cal B}=1-\kappa ^2 \left(
1-\frac{\delta\phi}{2\pi}\right)
\end{equation}
That is to say, the Tolman mass vanishes and the central magnetic field is
directly expressed in
terms of the red/blue shift $\kappa $ and the deficit angle $\delta\phi$. \\
Then consider the Melvin branch ($a=c=2/3,\; b=-1/3$). In this case there
is no simple physical
interpretation of the parameters $\beta$ and $\kappa$ in Eq.
(\ref{asymptKasner1}).
As for
the Tolman mass, we
notice that it is non-zero, but the
equations (\ref{tolmanmass}), (\ref{centralmag}) lead to a simple
relation to the central
magnetic field
\cite{Ver}:
\begin{equation}
{\cal B}=1+3GM
\end{equation}
For more discussion of these general relations (and some other ones), we
refer to Ref. \cite{Ver}.
In the remaining sections of this paper, we shall construct explicitly the
various types of
solutions to (\ref{systemNO1})-(\ref{systemE2}) by using a relaxation
procedure
to solve the four
coupled differential equations numerically. The constraint
(\ref{constraintmod})
has been used for estimating the numerical errors. \\
\section{Open solutions: Cosmic strings and Melvin branch}
\setcounter{equation}{0}
\label{secopen}
The existence of cosmic string solutions in the Abelian Higgs system is very
well established.
However, the existence of the solutions of the second type,
i.e. the Melvin branch type which is implied by (\ref{MW}) and (\ref{MB}), has
not been
properly
studied. We have found that for any conic cosmic string solution, an associated
solution exists in the Melvin branch. The main difference with respect to the
cosmic string branch is the fact that asymptotically the azimuthal circles have
vanishing circumference.
A related difference is the non-vanishing total mass (per unit
length) of the Melvin-like solutions. Figure \ref{figopn}
shows an example of the conic and the Melvin-like solution at the
point $(\alpha ,\gamma )=(2,1.8)$ in parameter space. Only the (square root of
the) metric components, $N$ and $L$, are plotted, as the scalar and vector
fields for the various solutions
deviate only very little from the standard and well examined cosmic
string configuration. The two branches are obviously quite different. The
cosmic string spacetime is asymptotically flat as $N$ is constant (actually
$N=1$ here as $\alpha = 2$ is the Bogomol'nyi limit \cite{Linet2,GibbCom})
and $L$
becomes
proportional to $x$ far from the core. In contrast hereto, the associated
solution in the Melvin branch has $N \propto x^{2/3}$ and $L \propto x^{- 1/3}$
in the asymptotic region. Therefore the circumference of circles lying in
planes
perpendicular to the core with $x=0$
as center will eventually decrease as one moves to larger $x$.
The curves shown in
Figure \ref{figopn} are representative for the cosmic string and Melvin
branches.
In the first
case, increasing
$\gamma$ at fixed $\alpha $ will
simply shift $N$ towards a higher constant value in the asymptotic region,
whereas $L$ will become less steep (meaning larger angular deficit). An
increase of $\alpha $ when $\gamma $ is held constant will have the opposite
effect, i.e. shifting $L$ to lower asymptotic values and decreasing the angular
deficit. This observation holds for all solutions (including the cylindrical
solutions and the closed solutions to be described in Section
\ref{secclosed}): only
the
$\gamma / \alpha$ ratio seems to matter for the asymptotic behavior. Still,
because of the region near the core, no obvious symmetry is observed that could
render the parameter space effectively one dimensional. In the case of the
Melvin branch, an increased $\gamma$ (again at fixed $\alpha$) tends to flatten
$N$, whereas the global maximum of $L$ takes a larger value at higher $x$. \\
Moving towards more massive strings in the parameter
space (larger $\gamma$ to $\alpha$ ratio) the angular deficit of the conic
solution will increase until it reaches a maximum of $2 \pi$.
The corresponding solution represents an asymptotically cylindrical manifold
with $N$ and $L$ both asymptotically constant.
One such solution exists for any $\alpha$ and the curve $\gamma _*(\alpha)$,
where $\delta \phi = 2 \pi$, plays a very special role in the
classification of
the various solutions. First of all, it obeys a power law \cite{LagGarfinkle}.
Secondly, as one
approaches a point on this curve in parameter space along the Melvin branch
the solution
converges to the same cylindrical solution as for the cosmic string; that
is, the conic solutions and the Melvin-like solutions coincide in a single
asymptotically cylindrical solution on the curve of maximal angular deficit.
Figure \ref{figmag} illustrates how the central magnetic field
${\cal B}$ varies
throughout the parameter space for both the cosmic string and the Melvin-like
universes. Notice how the two surfaces intersect along the curve of maximal
angular deficit, where the two branches coincide in the cylindrical solutions.
\\
\section{Closed solutions: Inverted cone and singular Kasner solutions}
\setcounter{equation}{0}
\label{secclosed}
The issue of ``supermassive cosmic strings" in the Abelian Higgs system was
discussed at early days \cite{LagGarfinkle,Ortiz}. Two types of closed
solutions were
found. One \cite{LagGarfinkle} in which $N(x)$ vanishes at a finite distance
from the axis ($x=x_{max}$) while $L(x)$ diverges at that point. In the other,
$N(x)$ stays
finite for any $x$ while $L(x)$
decreases linearly outside the core of the string, and vanishes at a finite
distance from the axis. The geometry is still conic but of an inverted one
where the apex of the cone is at the point where $L=0$ \cite{Gott,Ortiz}.
Using the same numerical methods as in the previous section, we have found
that
these two types of solutions are encountered just as the curve
$\delta\phi = 2 \pi$ is traversed. But the lack of asymptotic behavior for
closed solutions forces us to replace the boundary conditions
(\ref{boundarycond}) by the condition
\begin{eqnarray}
f(0)=0 &, & f(x_{max}) = 1 \nonumber \\
P(0)=1 &, & P(x_{max}) = 0
\label{bc2}
\end{eqnarray}
This ensures that the solutions are unit flux tubes, and makes the numerical
work simpler. For quite large radial extension this technicality is not of any
relevance as the variables $f$ and $P$ describing the scalar and vector fields,
respectively, converge quite fast to their values at ``infinity". \\
Figure \ref{figclo} shows an example of both
kinds of solutions at the point $(\alpha,\gamma)=(2,2.05)$ in parameter
space. This is
just above the curve of maximal angular deficit, which lies at $\gamma _*
(2)=2$ for fixed $\alpha =2$, and the solutions still have considerable sizes.
Strictly speaking, these two types of solutions
have no asymptotic behavior as they are both representing closed solutions that
pinch off at a finite radial extension ($x_{max}$). But for slightly
supermassive configurations we have
\begin{eqnarray}
N(x)=K(x)\sim \kappa (x_{max} -x)^a \hspace{6 mm},\hspace{6 mm}
L(x)\sim \beta (x_{max} -x)^b
\label{singKasner1}
\end{eqnarray}
The inverted cone behaves according to Eq. (\ref{SL}) and
the singular Kasner solution behaves according to Eq. (\ref{MW}).
The curves shown in Figure \ref{figclo} are representative for the
solutions above
the curve of maximal angular deficit. As $\gamma$ increases ($\alpha $ fixed)
both solutions become smaller in radial extent. For the inverted cone $N$ gets
shifted to a lower constant value and $L$ intersects zero at smaller $x$.
Likewise $x_{max}$ decreases for the singular Kasner solution
causing $L$ to diverge at similarly low $x$. We have found that for any given
$\alpha $ and $\gamma > \gamma _* (\alpha)$, the solution with geometry as an
inverted cone has the larger size. \\
As mentioned, the solutions shrink when $\gamma $ is increased ($\alpha $
fixed). At some point the spacetime described by the solution becomes smaller
than a few times the characteristic core thickness which scales as $\alpha
^{-1/2}$ \cite{NO}. At even higher values of $\gamma$ the choice of
boundary conditions
becomes more and more important to fundamental questions as flux quantization
and topological stability; the solutions exist mathematically but are probably
unphysical.
Figure \ref{figlim} shows three curves in parameter space which have all been
fitted to a power law, $\gamma = c_1 \alpha ^{c_2}$. The lowest one
is simply the curve of angular deficit $2 \pi$. This curve separates the two
open
solutions from the two closed ones, and represents itself a linear family of
asymptotically cylindrical solutions. The middle one represents the curve in
parameter space where the singular Kasner-like solutions have a radial extent
of $10 \, \alpha ^{-1/2}$, i.e. ten times the characteristic thickness of the
core. Similarly, the upper one represents the curve where the inverted cone
solutions pinches off at $x_{max} = 10 \, \alpha ^{-1/2}$. Starting on one of
these two curves and moving downwards (towards lower $\gamma$ for fixed
$\alpha$), the two closed solutions will become more and more similar and will
eventually coincide in the cylindrical solution. Moving
upward instead will yield rather small and increasingly unphysical
solutions. \\
\section{Conclusion}
\setcounter{equation}{0}
Cylindrically symmetric configurations of the gravitating Abelian
Higgs model have been
examined by application of relaxation techniques to Einstein equations
which simplify in this case to a system of coupled
ordinary differential equations. Everywhere in parameter space two
kinds of
solutions exist, except at the curve of angular deficit $2\pi $, where
only one
cylindrical solution exists. The two open solutions are the standard
asymptotically conic cosmic
string and a solution with a Melvin like asymptotic behavior. The two
closed solutions are the inverted
cone and a singular Kasner-like solution. Supermassive solutions pinch
off
very fast and are probably unphysical. \\
Here only the unit flux tube, $n=1$, has been considered, but
introduction of an
arbitrary $n \neq 0$ is expected to give similar families of
solutions, which
can be examined in an analogous way. \\
Finally, it is worthwhile noting that the singularities of static
field
configurations generally seem to be lifted when time-dependence is
reintroduced \cite{stringinfl}. Therefore the nature of the singularities
of the supermassive strings
may be understood in the framework of time dependent analysis.
|
\section{Introduction}
\vspace*{-0.5pt}
\noindent
The models containing interactions of electrons and phonons have been
the object of intense study in the recent years, mainly due to the
discovery of high $T_c$ superconductors ($HTS$) and colossal
magnetoresistance manganites ($CMR$). When trying to apply the
theoretical results to the real materials,
however, one is confronted with the
difficulty that a large degree of arbitrariness is
present, stemming from two origins. First,
different forms of interactions
have been introduced, essentially on an {\it ad hoc} basis. Second, the
value of the interaction parameters are estimated from the
experimental data, which, however, can be interpreted only after a
definite theoretical model, i.e. a set of electron-phonon interactions,
has been assumed to be relevant for the material under measurement.
Therefore, besides unavoidable
quantitative approximations in fitting the data,
also qualitative uncertainties are introduced {\it a priori} in the
analysis.
The purpose of the present work is to
provide, in a simple but not trivial model, both
rigorous prescriptions about the admissible electron-phonon
interaction terms, and the quantitative evaluation of the interaction
strength parameters. The model we consider is a dimer where the
electrons occupy a non-degenerate orbital described by Wannier
functions built from atomic orbitals of Gaussian shape. All the
one- and two-body electronic parameters in that model have been
analytically evaluated in Ref.1. The electron-phonon interactions
will be evaluated following the general definitions introduced in
Refs.2 and 3.\par The structure of the paper is as follows: in
Sec.2 we describe the model and the general procedure of
evaluation; Secs.3 and 4 are devoted to the analytical evaluation
of the coupling parameters in the non-adiabatic, and adiabatic
limits, respectively, also defining the correct form of the
corresponding terms in the Hamiltonian; Sec.5 discusses the
numerical evaluation of the coupling parameters; Sec.6 is devoted
to the conclusions. The Appendices contain details of the
analytical calculations.
\section{The Model}
\noindent In a general electron-phonon Hamiltonian
$H=H_{el}+H_{ph}+H_{el-ph} \label{ham1}$ the interacting term
$H_{el-ph}$ originates from the development of the
electronic part $H_{el}$ to first order in the phonon-induced displacement of
the ion positions. In our analysis we shall consider the terms due to
the
development of both the one-body and two-body contributions
to the electronic Hamiltonian, which, in standard
notation{\cite{acquarone1}, reads:
\[
H_{el}=\epsilon \sum_{\sigma }(n_{1\sigma }+n_{2\sigma})+\sum_{\sigma}
[t+X(n_{1-\sigma }+n_{2-\sigma})](c_{1\sigma }^{\dagger }c_{2\sigma }+H.c.)
+U(n_{1\uparrow}n_{1\downarrow}+n_{2\uparrow}n_{2\downarrow})
\]
\begin{equation}
+Vn_1n_2-2J_zS^z_1S^z_2-J_{xy}(S^+_1S^-_2+H.c.)
+P(c^{\dagger}_{1\uparrow}c^{\dagger}_{1\downarrow}
c^{}_{2\downarrow}c^{}_{2\uparrow}+H.c.).
\label{helbare}
\end{equation}
The electrons are assumed to occupy a non-degenerate orbital described
by Wannier functions built from atomic orbitals of Gaussias shape.
\subsection{Undisplaced Wannier functions}
\noindent The ions occupy the undeformed positions $\Rv_1^0=(-a/2,0,0)$ and $
\Rv_2^0=(a/2,0,0)
$. The unit vector ${\bfm e}_{12}=(\Rv_2^0-\Rv_1^0)/a=(1,0,0)$
points from ion 1 towards ion 2;
if not specified otherwise, $\bfm{e}=\bfm{e}_{12}$.
The position of the electron is $\bfm{r}=(x,y,z)$. We associate to
each site $i=1,2$ a
Gaussian atomic-like orbital $\phi_i(\bfm{r}-\bfm{R}_i)$. By defining
$N\equiv\left (2/ \pi\right)^{3/4}\Gamma^{3/2}$, so that
$<\phi_i|\phi_i>=1$, they read:
\[
\phi_1(\bfm{r}-\Rv_1^0)=
N\exp\left\{-\Gamma^2\left[\left(x+a/2\right)^2+y^2+z^2\right]\right\},
\]
\begin{equation}
\phi_2(\bfm{r}-\Rv_2^0)=
N\exp\left\{-\Gamma^2\left[\left(x-a/2\right)^2+y^2+z^2\right]\right\}.
\label{wan2}
\end{equation}
Their overlap $S_0 \equiv \langle \phi_1|\phi_2\rangle =
\exp(-\Gamma^2a^2/2)$ is non-vanishing. Then the Wannier functions
$\Psi_1, \Psi_2$ can be written as:
\[
\Psi_1(\bfm{r} -\Rv_1^0,\bfm{r}- \Rv_2^0)=A(S_0) \phi_1(\bfm{r}-\Rv_1^0) + B(S_0)
\phi_2(\bfm{r}-\Rv_2^0),
\]
\begin{equation}
\Psi_2(\bfm{r}-\Rv_1^0,\bfm{r}- \Rv_2^0)=B(S_0) \phi_1(\bfm{r}-\Rv_1^0) + A(S_0)
\phi_2(\bfm{r}-\Rv_2^0).
\label{wan1}
\end{equation}
The requirement
$<\Psi_i|\Psi_j>=\delta_{ij}$ yields the mixing coefficients $A,B$
as: \begin{equation} A(S_0)\equiv {1 \over 2} \left [ {{1}\over{ \sqrt{1+S_0}
}} +
{{1}\over{ \sqrt{1-S_0} }} \right],
\qquad
B(S_0)\equiv {1 \over 2} \left [ {{1}\over{ \sqrt{1+S_0} }} -
{{1}\over{ \sqrt{1-S_0} }} \right].
\label{wan4}
\end{equation}
In the limit $S_0\rightarrow 0$, one has $\Psi_i \rightarrow \phi_i$
$(i=1,2)$,
so that one can connect unambiguously each Wannier function to a well
defined site. \par
\subsection{Displaced Wannier functions}
\noindent We shall consider only deformations $\uv^{~}_1=u_1\bfm{e}_{12}$ and
$\uv^{~}_2=u_2\bfm{e}_{12}$ altering the length of the dimer,
so that only their $x$-components are non vanishing.
Their signs are defined, on both sites, with
respect to the same unit vector $\bfm{e}_{12}$, so as
to have a single reference system. Symmetry requirements impose
$\uv^{~}_1=-\uv^{~}_2$ so that the relative displacement $u\equiv
u_2-u_1=-2u_1=2u_2$. Even though we shall never introduce the
exact time dependency in the following calculation, it is
important to keep in mind that the displacements vary periodically
in time with the frequency $\Omega$ of the phonon. The
instantaneous positions of the ions are therefore defined as
\begin{equation}
\bfm{R}_1=\Rv_1^0+\uv^{~}_1=(-a/2+u_1, 0,0),\qquad \bfm{R}_2=\Rv_2^0+\uv^{~}_2=(a/2+u_2, 0,0).
\label{istpos}
\end{equation}
\noindent In the adiabatic limit, the orbitals
on each site are assumed to
adjust instantaneously to the displaced positions of the ions.
Then, if we define
\[
\phi_1[\bfm{r}-(\Rv_1^0+\bfm{u}_1)]=
N\exp\left\{-\Gamma^2\left[\left(x+a/2-u_1\right)^2+y^2+z^2\right]\right\},
\]
\begin{equation}
\phi_2[\bfm{r}-(\Rv_2^0+\bfm{u}_2)]=
N\exp\left\{-\Gamma^2\left[\left(x-a/2-u_2\right)^2+y^2+z^2\right]\right\},
\label{wan2d}
\end{equation}
the Wannier functions $\Psi_i$ ($i=1,2$)are
written as:
\[
\Psi_1(\Rv_1^0,\Rv_2^0,u_1,u_2)=A(u_1,u_2) \phi_1(\bfm{r}-\Rv_1^0-\bfm{u}_1) +
B(u_1,u_2) \phi_2(\bfm{r}-\Rv_2^0-\bfm{u}_2),
\]
\begin{equation}
\Psi_2(\Rv_1^0,\Rv_2^0,u_1,u_2)=B(u_1,u_2)
\phi_1(\bfm{r}-\Rv_1^0-\bfm{u}_1) + A(u_1,u_2)
\phi_2(\bfm{r}-\Rv_2^0-\bfm{u}_2). \label{wan1d}
\end{equation}
In principle we
have four coefficients $A(u_1,u_2),B(u_1,u_2)$ to determine. We
make the reasonable guess that, as for $u_i\rightarrow 0$ the $A$'s ( and
the $B$'s) have the same shape on both sites, then they keep the
same shape also for $u_i \ne 0$. We shall define $A_0, B_0,S_0$
as $\lim_{u_1,u_2\rightarrow 0}A(u_1,u_2)$ etc. As $\langle
\phi_1(u_1)|\phi_1(u_1)\rangle=\langle \phi_2(u_2)|\phi_2(u_2)\rangle=1$ still
holds, by defining
$S\equiv \exp\{-( {{\Gamma^2}/{2}})[a-(u_2-u_1)]^2\} $
the condition $\langle \Psi_i(u_1,u_2)|\Psi_j(u_1,u_2)\rangle=\delta_{ij}$
now yields:
\begin{equation}
A(S)\equiv {1 \over 2} \left [ {{1}\over{ \sqrt{1+S} }} +
{{1}\over{ \sqrt{1-S} }} \right],
\qquad
B(S)\equiv {1 \over 2} \left [ {{1}\over{ \sqrt{1+S} }} -
{{1}\over{ \sqrt{1-S} }} \right].
\label{AsBs}
\end{equation}
Namely, $A(S)$ and $B(S)$ depend on $S(u)$ as $A_0$ and $B_0$ depend
on $S_0$.\par
\subsection{The electronic parameters}
\noindent Given the Wannier functions, all the electronic
parameters can be evaluated: this was done in Ref.1.
To make clear our method of calculation, it is
convenient to explicitate
the one-body electronic interactions,
corresponding to the local energy $\epsilon$
and to the hopping amplitude $t$.
\begin{equation}
\epsilon_i=\int\Psi_i^{\ast}(\bfm{r},\bfm{R}_1,\bfm{R}_2)
\left[-{{\hbar^2}\over{2m}}\nabla^2+V_1(\bfm{r}-\bfm{R}_1)+V_2(\bfm{r}-\bfm{R}_2)\right]
\Psi_i^{~}(\bfm{r},\bfm{R}_1,\bfm{R}_2) d^3\bfm{r},
\label{epsilon0}
\end{equation}
\begin{equation}
t=\int \Psi_1^{\ast}(\bfm{r},\bfm{R}_1,\bfm{R}_2)
\left[-{{\hbar^2}\over{2m}}\nabla^2+V_1(\bfm{r}-\bfm{R}_1)+V_2(\bfm{r}-\bfm{R}_2)\right]
\Psi_2^{~}(\bfm{r},\bfm{R}_1,\bfm{R}_2) d^3\bfm{r}.
\label{hopping}
\end{equation}
In the
formulas above, the potentials originating from the ion cores at
the displaced positions $\bfm{R}_1$ and $\bfm{R}_2$ are:
\[
V_1\equiv V(\bfm{r}-\bfm{R}_1)=-e^2Z \left[\left(x+{a \over 2}-u_1\right)^2
+y^2+z^2\right]^{-1/2},
\]
\begin{equation}
V_2\equiv V(\bfm{r}-\bfm{R}_2)=-e^2Z\left[\left(x-{a \over 2}- u_2
\right)^2+y^2+z^2\right]^{-1/2}, \label{hol5}
\end{equation}
where $-e$ is
the electron charge, and $+Ze$ is the charge of the ion core. \par
The local energy $\epsilon$ (actually site-independent)
can be decomposed into three terms,
respectively corresponding to the contributions from the Laplacian
kinetic operator ($\epsilon_{\nabla}^{~}$) and from each one of
the ionic potentials ($\epsilon_{V_1}^{~},\epsilon_{V_2}^{~}$).
The modulation of the term where both charge distribution and
potential refer to the same ion gives rise to an Holstein-type
coupling. The other term, where the charge around one ions feels
the displaced potential of the other ion, we shall call
"crystal-field" coupling. In the literature\cite{barisic}
this latter term is usually assumed to be negligible
but we shall see that actually is of the same order as the other
ones, and can even become the dominant one.\par The modulation of
the hopping amplitude $t$ gives rise to the so-called
Su-Schrieffer-Heeger ($SSH$) interaction\cite{SSH}. Similarly to
$\epsilon$, it can be decomposed into kinetic ($t^{~}_{\nabla}$)
and potential ($t{~}_{V_1,V_2}$) contributions. \par
We need to distinguish between the adiabatic (labelled {\it ad})
and the non-adiabatic (labelled {\it na})
limit in evaluating the electron-phonon interactions, because the
integrals have different kernels in the two cases. Indeed, in the
adiabatic limit, when $\hbar\Omega<t$,
the displacements affect both the potentials and
the electronic Wannier functions, expressing the requisite that the
electronic charge distribution adjusts itself instantaneously at the
position of the ions. \par
We shall schematize the opposite situation $\hbar\Omega>t$,
where the electrons are
slower than the ions, as realized by the
electronic charge distribution staying centred around the undisplaced
ion position, while the potentials are centred on the displaced ions.
We shall call this the extreme anti-adiabatic limit. Not only the
strength of the interactions, but
also the form of the terms contributing to the Hamiltonian will turn out
to differ in the two cases.\par
The anti-adiabatic limit will be treated first, as the notation is easier
to establish in that case.
\vfill\eject
\section{Couplings in the non-adiabatic limit}
If the Wannier functions keep being centred around the
undisplaced ionic positions, neither $\epsilon^{~}_{\nabla}$ nor
$t^{~}_{\nabla}$
does change, therefore
no electron-phonon coupling originates from them.
The couplings derived from the two-body interactions are also identically
vanishing in this limit, because they involve the Wannier
functions and the inter-electronic Coulomb potential
which are both insensitive to the displacements of the ions.\par
The only non-vanishing
types of electron-phonon non-adiabatic couplings are those arising
from the variation of the potential contributions to $\epsilon$ and
$t$.
\subsection{The Holstein-type coupling $g_0$}
\noindent The Holstein-type coupling $g_0$ is usually introduced
as the site-independent amplitude of a term in the Hamiltonian
connecting the local charge with the local deformation:
\begin{equation}
g_0\sum_\sigma (n_{1\sigma} u_1+n_{2\sigma} u_2). \label{g0def1}
\end{equation}
\noindent This
interaction originates from the perturbation of the on-site atomic
energy $\epsilon_{0}$ due to the ion motion in the non-adiabatic
limit, namely:
\[
\langle\Psi_i(\Rv_1^0,\Rv_2^0)|V_i(\bfm{R}_i)|\Psi_i(\Rv_1^0,\Rv_2^0)\rangle-
\langle\Psi_i(\Rv_1^0,\Rv_2^0)|V_i(\bfm{R}_{i}^0)|\Psi_i(\Rv_1^0,\Rv_2^0)\rangle
\]
\begin{equation} \equiv g_0^{(i)}n_iu_i +{\cal{O}}(u_i^2),\qquad (i=1,2)
\label{g0def} \end{equation}
Our $g_0^{(i)}$
is therefore different from the interaction defined by
Holstein\cite{holstein}
because he considered the interaction as due to the
variation of the internuclear distance between the two components of
each dimer attached at the nodes of a frozen chain. We use the
"Holstein" label because this interaction has
the same form as the traditional Holstein term.
We also allow for a possible site-dependency of $g_0^{(i)}$, to show
later that this is not the case.\par
For the ion at $\bfm{R}_1$ one writes $g_0^{(1)} $ as:
\begin{equation}
g_0^{(1)} \equiv
\left(\lim_{u_1\rightarrow 0}{{\partial\epsilon^{(1)}_{V_1}}\over{\partial u_1}}\right)
=\left(\lim_{u_1\rightarrow 0} {{\partial } \over{\partial u_1}}\right)
\int_{-\infty}^{\infty}\Psi_1(x,y,z,\Rv_1^0,\Rv_2^0)^2V_1(u_1^{~})dxdydz.
\label{g01}
\end{equation}
In the integral
\[
\int_{-\infty}^{\infty}\Psi_1(x,y,z,\Rv_1^0,\Rv_2^0)^2V_1(u_1^{~})dxdydz =
\]
\begin{equation}
\int^{\infty}_{-\infty}
\left\{A\phi_1\left[\left(x+{{a}\over{2}}\right),y,z\right]+
B\phi_2\left[\left(x-{{a}\over{2}}\right),y,z\right]\right\}^2
V_1\left[\left(x+{{a}\over{2}}-u_1^{~}\right),y,z\right]dxdydz,
\label{g0def2}
\end{equation}
let us change variable form $x$ to $p=x+a/2-u_1^{~}$, to shift the
origin of $x$ coordinate onto the displaced ion, without changing
the shape of the Wannier functions. The reason is to avoid the
derivation with respect to $u_1^{~}$ of $V_1(u_1^{~})$ which is
discontinuous in the integration range.
The above expression changes into:
\begin{equation}
\int^{\infty}_{-\infty}
\left\{A\phi_1\left[\left(p+u_1^{~}\right),y,z\right]+
B\phi_2\left[\left(p-a+u_1^{~}\right),y,z\right]\right\}^2
V_1\left[(p,y,z)\right]dxdydz.
\label{g0u}
\end{equation}
In this form the kernel has a continuous derivative with respect to $u_1^{~}$
everywhere. Notice also that $A$ and $B$ still do not depend on $u_1^{~}$.
To derive with respect to $u_1^{~}$ Eq.\ref{g0u}, we use
$u_2^{~}=-u_1^{~}$ and notice that:
\begin{equation}
\lim_{u_1^{~}\rightarrow
0}{{\partial\phi_1^{~}(p+u_1^{~},y,z)}\over{\partial u_1^{~}}}
={{\partial\phi_1^{~}(x,y,z)}\over{\partial x}}, \qquad
\lim_{u_2^{~}\rightarrow 0}
{{\partial\phi_2^{~}(p-a-u_2^{~},y,z)}\over{\partial u_2^{~}}}
=-{{\partial\phi_2^{~}(x,y,z)}\over{\partial x}}, \label{derphi}
\end{equation}
yielding:
\[
\lim_{u_1^{~}\rightarrow 0}{{\partial}\over{u_1^{~}}}
\left\{A\phi_1\left[\left(p+u_1^{~}\right),y,z\right]+
B\phi_2\left[\left(p-a+u_1^{~}\right),y,z\right]\right\}^2
\]
\begin{equation}
=2\Psi_1^{~}\left[A{{\partial\phi_1^{~}}\over{\partial x}}
+B{{\partial\phi_2^{~}}\over{\partial x}}\right]
=2\Psi_1^{~}{{\partial\Psi_1^{~}}\over{\partial x}}.
\label{kernel1}
\end{equation}
Therefore we can write:
\begin{equation}
g_0^{(1)}=2\int^{\infty}_{-\infty}
\Psi_1^{~}{{\partial\Psi_1^{~}}\over{\partial x}} V_1^{~}dxdydz.
\label{hol4}
\end{equation}
By substituting $\Psi_1$ and its derivative
into Eq.{\ref{hol4}} and by noticing that
$\langle \phi_1|\left(x+{a / 2} \right)V_1|\phi_1\rangle =0$,
as the kernel is odd in $x+a/2$, we obtain:
\begin{equation} g_0^{(1)}=-4\Gamma^2\int^{\infty}_{-\infty} \left(B^2\phi_2^2
+2AB\phi_1\phi_2\right)xV_1 dxdydz +
2\Gamma^2aB^2\int^{\infty}_{-\infty}\phi_2^2 V_1dxdydz.
\label{hol6} \end{equation} The integrals in Eq.{\ref{hol6}} are listed in
Appendix 2. The final result is: \begin{equation}
g_0^{(1)}={{2\Gamma\sqrt{2/\pi} } \over {a}} \left\{B^2
\left[F_0(2a^2\Gamma^2) -S^4\right] +4ABS \left[
F_0\left({{a^2\Gamma^2}\over{2}}\right)-S \right] \right\}.
\label{hol27c} \end{equation}
\noindent That $g_0^{(i)}$ is
site-independent can be proved by considering that, for equal
charges and displacement amplitudes, the energies
$E^{(i)}=g_0^{(i)}n_i{\bf u}_i\bullet{\bf e}_{ij}$ $(i,j=1,2)$ on
both sites must coincide, i.e.
\begin{equation}
g_0^{(1)}{\bf u}_1\bullet{\bf
e}_{12}=g_0^{(2)}{\bf u}_2\bullet{\bf e}_{21} =-g_0^{(2)}{\bf
u}_2\bullet{\bf e}_{12} \label{g0equal}
\end{equation}
Now symmetry
requires $u_1=-u_2$ from which $g_0^{(2)}=g_0^{(1)}\equiv g_0$
follows.
\subsection{The crystal-field coupling}
\noindent This term expresses the change in the energy of
interaction between the charge on site $i$ and the potential
centred on the site $j$, that is:
\begin{equation}
\langle\Psi_i^0|V_j(\bfm{R}_j^0+u_j)|\Psi_i^0\rangle -
\langle\Psi^0_i|V_j(\bfm{R}_{j}^0)|\Psi^0_i\rangle\equiv g_{cf}^{(i)}n_iu_j
+\cal{O}(u_j^2)
\label{gcfnadef}
\end{equation}
In the literature, there is some confusion
on the form of this term, so we shall discuss it in some details,
following the line of reasoning used to prove that
$g_0^{(1)}=g_0^{(2)}$. Consider site $1$, with charge $n_{1}$.
Its energy, after a displacement ${\bf u}_{2}$, changes by an amount $%
E^{(1)}=g_{cf}^{(1)}n_{1}$ ${\bf u}_{2}\bullet {\bf e}_{12}$. \ This can
be considered as the quantity measured by an observer sitting on ion $1$ and
watching the ion $2$ \ moved by ${\bf u}_{2}$ .\ The equivalent measurement
done by an observer \ on ion $2$ watching the ion $1$ displaced by ${\bf u}%
_{1}$,\ yields $E^{(2)}=g_{cf}^{(2)}n_{2}{\bf u}_{1}\bullet {\bf
e}_{21}$.
Notice that each observer uses its own unit vector, pointing from the
observer's ion towards the other, displaced ion. Formally, $E^{(2)}$ can
be obtained by performing a site label permutation on $E^{(1)}.$ An
external observer would find it more convenient to refer both quantities to
the same reference system, \ ${\bf e}_{12}$ say, yielding$\
E^{(2)}=-g_{cf}^{(2)}n_{2}$ ${\bf u}_{1}\bullet {\bf e}_{12}.$ Assuming
equal charge density and displacement amplitude in the two observations,
one must have $E^{(1)}=E^{(2)}.$ \ It follows, dropping the charge
densities:
\begin{equation}
g_{cf}^{(1)}{\bf u}_{2}\bullet {\bf e}_{12}=-g_{cf}^{(2)}{\bf u%
}_{1}\bullet {\bf e}_{12}\Longrightarrow
g_{cf}^{(1)}u_{1}=-g_{cf}^{(2)}u_{2}\Longrightarrow
g_{cf}^{(1)}=g_{cf}^{(2)}=g_{cf}^{{}}, \label{cfdef1}
\end{equation}
because the only displacements allowed by symmetry are such that \ $%
u_{1}=-u_{2}$. Therefore for the dimer as a whole one writes this term as
\begin{equation}
g_{cf}^{{}}\sum_{\sigma}(n_{1\sigma}u_{2}+n_{2\sigma}u_{1}).
\label{cfdef2}
\end{equation}
Passing now to the explicit evaluation for site $1$ we have:
\begin{equation}
g_{cf}^{(1)}=\lim_{u_2^{~}\rightarrow 0} {{\partial } \over{
\partial u_2}}
\int^\infty_{-\infty}\Psi_1(\Rv_1^0,\Rv_2^0,x,y,z)^2V_2^{~}(u_2^{~})dxdydz.
\label{cfna1} \end{equation}
We change variables from $x$ to
$p=x/a/2-u_2^{~}$ and proceed in strict analogy to the evaluation
of $g_0^{(1)}$, arriving at:
\begin{equation}
g_{cf}^{(1)}=
2\int^\infty_{-\infty}\Psi_1{{\partial \Psi_1} \over{ \partial
x}}V_2 dxdydz. \label{cfna2}
\end{equation}
By explicitating $\Psi_1$ and
its derivative we obtain:
\[
g_{cf}^{(1)}=-4\Gamma^2\int^{\infty}_{-\infty}
\left(A^2\phi_1^2+B^2\phi_2^2 +2AB\phi_1\phi_2\right)xV_2 dxdydz
\]
\begin{equation} - 2\Gamma^2a\int^{\infty}_{-\infty}
\left(A^2\phi_1^2-B^2\phi_2^2\right)V_2dxdydz. \label{cfna3}
\end{equation}
>From the matrix elements evaluated in Appendix 2 we obtain:
\begin{equation}
g^{(1)}_{cf}=-2A\left({{\Gamma}\over{a}}\right)\sqrt{{{2}\over{\pi}}}
\left[AF_0(2a^2\Gamma^2)+4BSF_0(a^2\Gamma^2/2)-4BS^2-AS^4\right].
\label{g1cfnadexpl} \end{equation}
\vfill\eject
\subsection{The Su-Schrieffer-Heeger coupling}
\noindent This term is a non-diagonal inter-site coupling,
due to the modulation of
the hopping amplitude $t$.
As the Wannier functions keep being centred on the static lattice
positions, the difference in kinetic energy caused by the displacements
is due only to the contribution to $t$ from $V_1$ and $V_2$:
\[
\int \Psi_1(\bfm{r}-\Rv_1^0)[V_1(\bfm{r}-\Rv_1^0-\uv^{~}_1)+V_2(\bfm{r}-\Rv_2^0-\uv^{~}_2)]\Psi_2(\bfm{r}-\Rv_2^0)
d^3\bfm{r}
\]
\[
-\int \Psi_1(\bfm{r}-\Rv_1^0)[V_1(\bfm{r}-\Rv_1^0)+V_2(\bfm{r}-\Rv_2^0)]\Psi_2(\bfm{r}-\Rv_2^0)
d^3\bfm{r}
\]
\begin{equation} \equiv \gamma_{12}\sum_\sigma (c_{1\sigma}^\dagger
c^{~}_{2\sigma}+c_{2\sigma}^\dagger
c^{~}_{1\sigma})(u_2-u_1)+{\cal{O}}[(u_2-u_1)^2]. \label{g12na1}
\end{equation} Notice that, to preserve the invariance of the Hamiltonian
under site permutation, the $SSH$ coupling has to be odd under the
same operation: $\gamma_{12}=-\gamma_{21}$ (see e.g Refs. 2,3).\par
Let us distinguish
the two contributions as $t_{V_{1}}^{~}$ and $t_{V_{2}}^{~}$,
defined as:
\[
t_{V_{1}}^{~} =\int^{\infty}_{-\infty}
{{\Psi_1^{~}(u=0)\Psi_2^{~}(u=0)}\over{\sqrt{(x+a/2-u_1^{~})^2+y^2+z^2}}}
dxdydz,
\]
\begin{equation}
t_{V_{2}}^{~} =\int^{\infty}_{-\infty}
{{\Psi_1^{~}(u=0)\Psi_2^{~}(u=0)}\over{\sqrt{(x-a/2-u_2^{~})^2+y^2+z^2}}}
dxdydz.
\label{tv1v2}
\end{equation}
To evaluate $t_{V_1}^{~}$, let us define $p=x+a/2-u_1^{~}$ and use the
symmetry relation $u_1^{~}=-u/2$, so that we can derive the kernel:
\begin{equation}
{{\partial t_{V_1}^{~}}\over{\partial u}}=\int^{\infty}_{-\infty}
{{\partial [\Psi_1^{~}(p,u,y,z)\Psi_2^{~}(p,u,y,z)]}\over{\partial
u}}V_1^{~}(p,y,z)dpdydz.
\label{dtudu}
\end{equation}
Next we notice that
\begin{equation}
\lim_{u\rightarrow 0} {{\partial
[\Psi_1^{~}(p,u,y,z)\Psi_2^{~}(p,u,y,z)]}\over{\partial u}}
=-{{1}\over{2}}\lim_{u\rightarrow 0} {{\partial
[\Psi_1^{~}(p,u,y,z)\Psi_2^{~}(p,u,y,z)]}\over{\partial p}}
=-{{1}\over{2}}{{\partial (\Psi_1^{~}\Psi_2^{~})}\over{\partial
x}},
\label{psider}
\end{equation}
so that finally we can write:
\begin{equation}
\lim_{u\rightarrow 0}{{\partial t^{~}_{V_1}}\over{\partial u}}=
-{{1}\over{2}}\int^{\infty}_{-\infty} {{\partial
[\Psi_1^{~}\Psi_2^{~}]}\over{\partial x}}V_1^{~}dxdydz.
\label{dtv1du}
\end{equation}
By performing similar manipulation on
$t_{V_2}^{~}(u_2)$, but now with $u_2^{~}=u/2$, we arrive at:
\begin{equation}
\lim_{u\rightarrow 0}{{\partial t^{~}_{V_2}}\over{\partial u}}=
{{1}\over{2}}\int^{\infty}_{-\infty} {{\partial
[\Psi_1^{~}\Psi_2^{~}]}\over{\partial x}}V_2^{~}dxdydz.
\label{dtv2du}
\end{equation}
Summing the two contributions yields:
\begin{equation}
\gamma_{12}=-{1 \over
2}\int^\infty_{-\infty}dxdydz \left[{{\partial (\Psi_1\Psi_2)}
\over{ \partial x}} \right]V_1 +{1 \over
2}\int^\infty_{-\infty}dxdydz \left[{{\partial (\Psi_1\Psi_2)}
\over{ \partial x}}\right] V_2 \equiv -\cal{X}+\cal{Y}.
\label{g12na2}
\end{equation}
The first integral, $\cal{X}$, after
explicitating $\Psi_1$ and $\Psi_2$ becomes:
\begin{equation}
{\cal{X}}\equiv
{1 \over 2}\int^\infty_{-\infty}dxdydz {{\partial } \over {
\partial x}}\left[AB(\phi_1^2+\phi_2^2)+
(A^2+B^2)\phi_1\phi_2\right]V_1(x,y,z). \label{C1}
\end{equation}
The
derivatives:
\begin{equation}
{{\partial (\phi_1\phi_2)} \over { \partial
x}}=-4\Gamma^2x\phi_1\phi_2, \qquad {{\partial \phi_{1}^2} \over {
\partial x}}=-4\Gamma^2\left(x+{a \over 2}\right)\phi_{1}^2, \qquad
{{\partial \phi_{2}^2} \over { \partial x}}=-4\Gamma^2\left(x-{a
\over 2}\right)\phi_{2}^2. \label{C2} \end{equation}
when substituted into
\Eq{\ref{C1}} yield
\begin{equation}
{\cal{X}}= -2\Gamma^2
\left\{AB[\langle\phi_1|(x+{a \over 2})V_1|\phi_1\rangle + \langle\phi_2|(x-{a
\over 2})V_1|\phi_2\rangle] +(A^2+B^2)\langle
\phi_1|xV_1|\phi_2\rangle\right\}. \label{g12na3}
\end{equation}
Similarly we get
\[
{\cal{Y}} \equiv
{1 \over 2}\int^\infty_{-\infty}dxdydz
\left[{{\partial (\Psi_1\Psi_2)} \over{ \partial x}}\right]V_2,
\]
\begin{equation}
{\cal{Y}}=-2\Gamma^2\left\{ AB\left[\langle\phi_1|(x+{a \over
2})V_2|\phi_1\rangle +\langle\phi_2|(x-{a \over 2})V_2|\phi_2\rangle \right]
+(A^2+B^2)\langle\phi_1|xV_2|\phi_2\rangle\right\}. \label{C3}
\end{equation}
The
contributions from $\langle\phi_1|(x+a/2)V_1|\phi_1\rangle $ in
Eq.\ref{g12na3} and $\langle\phi_2|(x-a/2)V_2|\phi_2\rangle$ in
Eq.\ref{C3} vanish due to the
parity of the kernel.
Substituting the matrix elements from Appendix 2 we get:
\[
\gamma_{12}=4\sqrt{{{2}\over{\pi}}}\left({{\Gamma}\over{a}}\right)
\left\{{{AB}\over{2}}\left[S_0^4-(1-4a^2\Gamma^2)F_0(2a^2\Gamma^2)\right]
\right\}
\]
\begin{equation}
+4\sqrt{{{2}\over{\pi}}}\left({{\Gamma}\over{a}}\right)
\left\{(A^2+B^2)S_0\left[S_0-F_0(a^2\Gamma^2/2)\right]\right\}.
\label{ga12nadexpl}
\end{equation}
Under site permutation $a\rightarrow -a$ and
$A\rightarrow B$ so that $\gamma_{12}=-\gamma_{21na}$ as expected.
In conclusion, in the non-adiabatic limit the electron-phonon
Hamiltonian is given by:
\[
H^{na}_{el-phon}= g_0\sum_\sigma (n_{1\sigma} u_1+n_{2\sigma} u_2)+
g_{cf}^{{}}\sum_{\sigma}(n_{1\sigma}u_{2}+n_{2\sigma}u_{1})
\]
\begin{equation} +\gamma_{12}\sum_\sigma (c_{1\sigma}^\dagger
c^{~}_{2\sigma}+c_{2\sigma}^\dagger c^{~}_{1\sigma})(u_2-u_1).
\end{equation}
\section{Couplings in the adiabatic limit}
One can use the method introduced in the non-adiabatic limit to
evaluate the couplings in the adiabatic case. However, one readily
verifies that, in the explicit expression of the different electronic
interactions in the adiabatically displaced state, $u$
invariably enters in the combination $a+u$. Therefore the procedure of
first deriving the
integral kernels with respect to $u$, then taking the limit $u\rightarrow 0$, and
finally evaluating the integrals is equivalent to first
evaluating the interactions
for $u=0$, and then deriving them with respect to $a$. All the
parameters in Eq.\ref{helbare} were explicitly evaluated in Ref.1,
so we shall simply derive them with respect to the dimer length
$a$. \par
\subsection{The coupling term derived from $\epsilon$}
There is some confusion in the literature about the correct form
of the electron-phonon Hamiltonian obtained in this limit from the
variation of the local energy $\epsilon$, therefore we shall
devote some space to discussing this point. \par \noindent In the
adiabatic limit, $u_i\ne 0$ in both the charge distributions and
in the potentials. As the origin of the $x$-coordinate can be
placed onto one of the displaced ions, this interaction couples
the charge on site $i$ to the position of site $j$ through the
modification of both the kinetic and the potential contributions.
One has then to take
into account the {\it relative} displacement of the ions.
To obtain coupling terms from $\epsilon_{\nabla}$ and
$\epsilon^{(i)}_{V_{i}}$ it is essential that one uses the proper
Wannier functions. If one instead adopted the local orbitals, such
couplings would vanish. In the adiabatic limit, therefore, the
overall $\epsilon$-derived electron-phonon coupling term in the
Hamiltonian is:
\begin{equation}
g_{\epsilon}^{(1)}\sum_\sigma n_{1\sigma}(u_2-u_1)
+g_{\epsilon}^{(2)}\sum_\sigma n_{2\sigma}(u_1-u_2).
\label{gcfaddef}
\end{equation}
\noindent As $g_{\epsilon}^{(1)}=g_{\epsilon}^{(2)}$
we can drop the site indexes, and write the total adiabatic
contribution from local energy terms to the electron-phonon
Hamiltonian as:
\begin{equation}
H^{\epsilon}_{el-phon}=
g_{\epsilon}\sum_\sigma(n_{2\sigma}-n_{1\sigma})(u_1-u_2).
\label{cfad13}
\end{equation} An expression similar to Eq.\ref{cfad13} has been proposed in
Ref.7. To convince the skeptical reader, in Appendix 4 we show,
using
the contribution from $\epsilon_{V_{2}}^{(1)}$ as an
example, that, starting from the general definition of the coupling,
one verifies Eq.\ref{cfad13}.\par
\subsection{Explicit expressions for the adiabatic couplings.}
\subsubsection{Couplings originated from one-body electronic
interactions.}
There are two type of couplings: one ($g_{\epsilon}$),
derived from the local energy:
\begin{equation}
g^{~}_{\epsilon}\equiv g_{\nabla}^{(i)}+ g_{V_{i}}^{(i)}+
g_{V_{j}}^{(i)} \qquad (i,j=1,2),
\label{epgeps}
\end{equation}
and one ($\gamma_{12}$) from the hopping term
(Su-Schrieffer-Heeger coupling):
\begin{equation}
\gamma_{12}\equiv
\gamma_{12\nabla}+\gamma_{12V}.
\label{gam12def}
\end{equation}
Below we list each contribution to the electronic
interactions and the derived electron-phonon couplings. From the
local energy on site $1$
\[
\epsilon^{(1)} \equiv \epsilon_{\nabla}^{~} +\epsilon_{V_1}^{(1)}
+\epsilon_{V_2}^{(1)},
\]
\[
\epsilon^{~}_{\nabla} ={{\hbar^2}\over{2m}}\left[3\Gamma^2+
\Gamma^4\left({{a^2S^2}\over{1-S^2)}}\right)\right],
\]
\[
\epsilon_{V_{1}}^{(1)}=-Ze^2\left(2\Gamma\sqrt{ {{2}\over{\pi}}
}\right)
\left[A^2+B^2F_0(2a^2\Gamma^2)+2ABSF_0(a^2\Gamma^2/2)\right],
\]
\begin{equation}
\epsilon_{V_2}^{(1)}=-Ze^2\left(2\Gamma\sqrt{{{2}\over{\pi}}}\right)
\left[B^2+A^2F_0(2a^2\Gamma^2)+2ABSF_0(a^2\Gamma^2/2)\right],
\label{eps3}
\end{equation}
one obtains the three contributions to
\[
g_{\epsilon}^{(1)}\equiv
g_{\nabla}^{~}+g_{V_1}^{(1)}+g_{V_2}^{(1)},
\]
\[
g_{\nabla}=
-{{\hbar^2}\over{2m}}\left[{{a\Gamma^4S^2}\over{(1-S^2)^2}}\right]
\left[2\left(1-a^2\Gamma^2-S^2\right)\right],
\]
\[
g^{(1)}_{V_1}=
-Ze^2\left(2\Gamma\sqrt{{{2}\over{\pi}}}\right) \left[{{\partial
A^2}\over{\partial u}}+{{2ABS^2+B^2S^4}\over{a}}
+F_0(2a^2\Gamma^2)\left({{\partial B^2}\over{\partial u}}-
{{B^2}\over{a}}\right)\right],
\]
\[
-Ze^2\left(2\Gamma\sqrt{{{2}\over{\pi}}}\right)
(2S)\left[{{\partial AB}\over{\partial
u}}-{{AB}\over{a}}\left(1+a^2\Gamma^2)\right)\right],
\]
\[
g^{(1)}_{V_2}=-Ze^2\left(2\Gamma\sqrt{{{2}\over{\pi}}}\right)
\left[{{\partial B^2}\over{\partial u}}+{{2ABS^2+A^2S^4}\over{a}}
+F_0(2a^2\Gamma^2)\left({{\partial A^2}\over{\partial u}}-
{{A^2}\over{a}}\right)\right],
\]
\begin{equation}
-Ze^2\left(2\Gamma\sqrt{{{2}\over{\pi}}}\right)
(2S)\left[{{\partial AB}\over{\partial u}}
-{{AB}\over{a}}\left(1+a^2\Gamma^2)\right)\right].
\label{geps}
\end{equation}
The two potential contributions depend on the site for which
they are evaluated, but their sum, $g^{(1)}_{V_1}+g^{(1)}_{V_2}$,
does not.\par For the hopping term, we can lump together the
potential contributions which individually have no particular
physical meaning. Then from
\[
t\equiv t_{\nabla}^{~}+t_V^{~},
\]
\[
t_{\nabla}={{\hbar^2}\over{2m}}{{a^2S\Gamma^4}\over{(1-S^2)}},
\]
\begin{equation} t_{V}= -Ze^2\left(2\Gamma\sqrt{{{2}\over{\pi}}}\right)
\left\{2AB\left[1+F_0(2a^2\Gamma^2)\right]+2(A^2+B^2)
SF_0(a^2\Gamma^2/2)\right\},
\label{tnablaV}
\end{equation}
the
Su-Schrieffer-Heeger coupling $\gamma^{~}_{12}\equiv
\gamma^{~}_{12\nabla}+\gamma^{~}_{12V}$ is obtained as:
\[
\gamma_{12\nabla}=
{{\hbar^2}\over{2m}}\left[{{aS\Gamma^4}\over{(1-S^2)^2}}\right]
\left[2(1-S^2)-a^2\Gamma^2(1+S^2)\right],
\]
\[
\gamma_{12V}=
-Ze^2\left(4\Gamma\sqrt{{{2}\over{\pi}}}\right)\Bigg\{
\left[{{\partial AB}\over{\partial u}}
+{{A^2+B^2}\over{a}}S^2+{{AB}\over{a}}S^4\right] +
\left[{{\partial AB}\over{\partial u}}
-{{AB}\over{a}}\right]F_0(2a^2\Gamma^2),
\]
\begin{equation}
+S\left[{{\partial (A^2+B^2)}\over{\partial u}}
-{{A^2+B^2}\over{a}}(1+a^2\Gamma^2)\right]F_0(a^2\Gamma^2/2)\Bigg\}.
\label{SSHV}
\end{equation}
Notice that, as the partial derivatives are linear in $a$, then both
$\gamma_{12\nabla}^{~}$ and $\gamma^{~}_{12V}$ change
sign under site permutation, as expected from Refs.2,3.
\par
The electron-phonon Hamiltonian in the adiabatic limit has
therefore the form:
\begin{equation}
H^{ad}_{el-phon}=
g_{\epsilon}^{{}}\sum_{\sigma}(n_{1\sigma}-n_{2\sigma})(u_{2}-u_{1})
+\gamma_{12}\sum_\sigma (c_{1\sigma}^\dagger
c^{~}_{2\sigma}+c_{2\sigma}^\dagger c^{~}_{1\sigma})(u_2-u_1).
\end{equation}
\subsubsection{Couplings originated from two-body electronic
interactions.}
Also the two-body electronic interactions $U,V,J(=P),X$ of Eq. (1)
give origin to electron-phonon couplings which, to the best of our
knowledge, have never been considered in the literature up to now.
The electron-phonon coupling originating from the modulation of the
kinetic exchange in the $t-J$ Hamiltonian has been considered in
Ref.17. However, the kinetic exchange
$J_{kin}\equiv 2t^2/(U-V)$ is a composite
quantity different from the direct exchange $J_{xy}=J_z=P$ in Eq. (1).
Here we list the values the two-body electron-phonon interactions
have in our model. They were obtained
by deriving with respect to $a$ the two-body interactions
evaluated explicitly in Ref.1. Notice that, as
$U(a)=J(a)+e^2\Gamma/\sqrt{\pi}$, then $dU/da=dJ/da=dP/da$
\[
{{dX}\over{da}}=-e^2{{\Gamma}\over{\sqrt{\pi}}}
\left[\left(-a\Gamma S\right)
\frac{\left( 1+3S^{2}\right) }{(1-S^{2})^{3}}\right]
\left[1+2S^{2}+F_{0}\left( a^{2}\Gamma ^{2}\right)
-2(1+S^{2})F_{0}\left( \frac{a^{2}\Gamma ^{2}}{4}\right) \right]
\]
\[
-e^{2}\frac{\Gamma }{\sqrt{\pi }}\left[ \frac{S/a}{(1-S^{2})^{2}}\right]
\left\{ 4a^{2}\Gamma ^{2}S^{2}\left[ F_{0}\left( \frac{a^{2}\Gamma ^{2}}{4}%
\right) -1\right] +S^{2}-F_{0}\left( a^{2}\Gamma ^{2}\right) \right\}
\]
\begin{equation}
-e^{2}\frac{\Gamma }{\sqrt{\pi }}\left[
\frac{S/a}{(1-S^{2})^{2}}\right] \left\{ 2(1+S^{2}) \left[
F_{0}\left( \frac{a^{2}\Gamma ^{2}}{4}\right) -\sqrt{S}\right]
\right\}, \label{dXda}
\end{equation}
\[
\frac{dU}{da}=
e^{2}\frac{\Gamma }{\sqrt{\pi }}\left[ \frac{-4a\Gamma
^{2}S^{2}}{(1-S^{2})^{3}}\right] \left[ 2-S^{2}+2S^{4}+S^{2}F_{0}\left(
a^{2}\Gamma ^{2}\right) -4S^{2}F_{0}\left( \frac{a^{2}\Gamma ^{2}}{4}\right)
\right]
\]
\[
+e^{2}\frac{\Gamma }{\sqrt{\pi }}\left[ \frac{S^{2}/a}{(1-S^{2})^{2}}\right] %
\Bigg\{2a^{2}\Gamma ^{2}\left[ 1-4S^{2}-F_{0}\left( a^{2}\Gamma ^{2}\right)
+4F_{0}\left( \frac{a^{2}\Gamma ^{2}}{4}\right) \right]
\]
\begin{equation}
+S^{2}-F_{0}\left( a^{2}\Gamma ^{2}\right) +4\left[ F_{0}\left( \frac{%
a^{2}\Gamma ^{2}}{4}\right) -\sqrt{S}\right] \Bigg\},
\label{dUda}
\end{equation}
\[
\frac{dV}{da}=e^{2}\frac{\Gamma }{\sqrt{\pi }}\left[- \frac{4a\Gamma
^{2}S^{2}}{(1-S^{2})^{3}}\right] \left[ 3-S^2-8S^4-(7-5S^2)F_0(a^2\Gamma^2)
-4(1-3S^2)F_0\left({{a^2\Gamma^2}\over{4}}\right)\right]
\]
\[
+e^{2}\frac{\Gamma }{\sqrt{\pi }}\left[ \frac{1/a}{(1-S^{2})^{2}}\right] %
\Bigg\{2a^{2}\Gamma ^{2}S^{2}\left[ -1-4S^{2}+F_{0}\left( a^{2}\Gamma
^{2}\right) +4F_{0}\left( \frac{a^{2}\Gamma ^{2}}{4}\right) \right]
\]
\begin{equation}
+2S^2+{{3}\over{2}}S^4-\left(2-{{3}\over{2}}S^2\right)F_0(a^2\Gamma^2)
+2S^{2} \left[ F_{0}\left( \frac{a^{2}\Gamma
^{2}}{4}\right)-\sqrt{S} \right] \Bigg\}, \label{dVda}
\end{equation}
\[
\frac{dJ}{da}\equiv\frac{dP}{da}\equiv\frac{dU}{da}. \label{dJda}
\]
All the above interactions change sign under site permutation. The
terms they contribute to the electron-phonon Hamiltonian all have
the same form, namely: \begin{equation} H_{Y}^{ep}= \frac
{dY}{da}F(c^{\dagger}_{i\sigma},c^{~}_{j\sigma}) (u_j-u_j)\qquad
(i,j=1,2) \end{equation} \noindent where $Y=U,V,X,J$ and
$F(c^{\dagger}_{i\sigma},c^{~}_{j\sigma})$ $(i,j=1,2)$ is the
function of Fermi operators representing the two-body interaction
whose amplitude is $Y$
\section{Results}
\noindent By definition,
as $n_{i\sigma}$ is the number of electrons per site
(a dimensionless quantity), then the interaction parameters have
dimensions $[energy][length]^{-1}$. To get them in energy units
(eV) we shall measure the deformations $u_i$ in units of the
characteristic phonon length $L=\sqrt{\hbar/(2\Omega M)}$. We
choose the phonon frequency such that $\hbar\Omega=0.1$ eV, which
is appropriate to $HTS$ and $CMR$, and $M$ equal to the mass of
$^{16}O$.
\begin{figure}
\centerline{\psfig{figure=fig1.eps,height=6cm,width=6cm}}
\caption {Non-adiabatic coupling constants
$g_0,g_{cf},\gamma_{12}$ (in eV) versus the dimer length (in \AA),
evaluated assuming $\Gamma=1.0 \AA^{-1}$.} \label{fig1}
\end{figure}
Figs.1 and 2 show the behaviour versus the dimer length $a$ (in \AA)
of the non-adiabatic coupling parameters
$g_0,g_{cf},\gamma_{12}$ (in eV) evaluated by assuming two
representative values $\Gamma=1.0,2.0$\AA$^{-1}$ for the
shape-controlling parameter of the Wannier functions. The most
unexpected result concerns $g_{cf}$. While usually neglected in
the literature\cite{barisic} on metallic systems, this coupling
has been recognized as relevant to polar materials\cite{barisic2}.
We find indeed that, when $\Gamma=1.0$\AA$^{-1}$, $g_{cf}$ is
larger than $g_0$ for any $a$,
and it becomes the largest
parameter for $a>2.2$\AA. For small $a$, the $SSH$ coupling is the
largest.
The strength of all the couplings decreases with $a$,
less quickly for $g_{cf}$, which, for large $a$, is still
comparable to the values of the hopping amplitude $t(a)$. \par
\begin{figure}
\centerline{\psfig{figure=fig2.eps,height=6cm,width=8cm}}
\caption {Non-adiabatic coupling constants $g_{cfad},
\gamma_{12ad}$ (in eV) versus the dimer length (in \AA), for
$\Gamma=2.0$\AA$^{-1}$.} \label{fig2}
\end{figure}
When $\Gamma=2.0$\AA$^{-1}$(see Fig.2) $g_0$ is negligible for any
$a$, $\gamma_{12}$ is large for small $a$ but drops very rapidly to
negligible values as $a$ increases, while $g_{cf}$ keeps appreciable
values for all $a$ values. We can conclude that, in the non-adiabatic
limit, the more localized
are the orbitals, the more relevant is the role of $g_{cf}$ in
relation to the other admissible couplings.\par
\begin{figure}
\centerline{\psfig{figure=fig3.eps,height=6cm,width=8cm}}
\caption {Adiabatic coupling constants $g_{\epsilon},
\gamma_{12ad}$ (in eV) versus the dimer length (in \AA), for
$\Gamma=1.0$ \AA$^{-1}$.} \label{fig3}
\end{figure}
Fig.3 is the adiabatic counterpart of Fig.1. The Holstein type
interaction is absent, being identically vanishing as discussed above.
Here we find that $g_{\epsilon}$ is always larger than the $SSH$
interaction $\gamma_{12}$, and particularly for large $a$ there is
an order of magnitude difference between them. \par
\begin{figure}
\centerline{\psfig{figure=fig4.eps,height=6cm,width=8cm}}
\caption {Adiabatic coupling constants from two-body interactions
$dX/da, dU/da, dV/da$ (in eV) versus
the dimer length (in \AA), for $\Gamma=1.0$ \AA$^{-1}$.}
\label{fig4}
\end{figure}
Fig.4 shows the couplings derived from the two-body electronic
interactions for the same parameters as Fig.3. In general, their
values are smaller than those of $g_{\epsilon}$ and $\gamma_{12}$,
with the possible exception of $dV/da$. Indeed, that coupling
arises from a physical mechanism not very different from the one
originating $g^{(i)}_{V_j}$, i.e. the vibration of the charge on
site $j$ as felt by site $i$. Similarly to $g_{\epsilon}$ also
$dV/da$ decreases slowly with $a$, so that for large $a$ those two are
the only relevant couplings. It is worth stressing that, though
$U$ is larger than $J$, their derivatives coincide. Besides, the
derivative of the interaction coupling $X$ is a non-monotonic
function of the lattice constant. Indeed, $X$ exhibits a maximum
for $a\simeq$ 1.8 and then sharply decreases to a negligible
value.
\par
\begin{figure}
\centerline{\psfig{figure=fig5.eps,height=6cm,width=8cm}}
\caption {Adiabatic coupling constants $g_{\epsilon}, \gamma_{12},
dV/da$ (in eV) versus the dimer length (in \AA), for $\Gamma=2.0$
\AA$^{-1}$.} \label{fig5}
\end{figure}
When $\Gamma=2.0$\AA$^{-1}$, as shown in Fig.5, $g_{\epsilon}$ and
$\gamma_{12}$ have trends similar to the non-adiabatic case, with
larger values for small $a$. As $a$ increases, $\gamma_{12}$ quickly
reduces to negligible values, while $g_{\epsilon}$ decreases more slowly.
Of the two-body couplings, only $dV/da$ is non-negligible, and it has
a strength very close (in absolute value) to $g_{\epsilon}$.\par
\section{Conclusions}
\noindent We have presented the analytical evaluation of the electron-phonon
coupling parameters derived from both one- and two-body electronic interactions
in a model of a dimer with a non-degenerate orbital
built from atomic orbital of Gaussian shape. The approach we have
followed (proposed in general terms in Refs.2,3)
consists in inserting the site displacements $u_1,u_2$
in the kernels of the
integrals defining the one-body electronic parameters $\epsilon_0$ and
$t$, and then considering the first-order terms in the expansion of
$\epsilon_0(u_1,u_2)$ and $t(u_1,u_2)$. In the adiabatic case
this procedure is equivalent to the simpler
one\cite{labbe,barisic,kuzemsky}$^{-}$\cite{petru}
consisting in evaluating the
electronic parameters as functions of the lattice parameter
(corresponding to the dimer length $a$) and then deriving with
respect to $a$. In the anti-adiabatic case, however, the latter
procedure can not be applied, because $a$ and $u$ enter
independently the kernel of the integrals. From the physical point
of view, the "$a$-derivative" method describes how the electronic
parameters vary under a quasi-static (low frequency) change of the
equilibrium position of the ions, as could be realized e.g. under
pressure. In general the electron-phonon couplings are the
effect of the ion oscillations at frequency $\Omega$, which causes the
system to be alternatively compressed and elongated over a time
$\approx \Omega^{-1}$. It is this implicit dynamics of the
displacements which allows for distinguishing between the
adiabatic and non-adiabatic regimes, and forbids the use of the
"$a$-derivative" method in the latter case.\par A novel result is
the evaluation of the couplings originating from the two-body
electronic interactions. We have shown that at least one of them,
generated by the Coulomb repulsion between the charges on
different sites, is comparable, or even larger, than the couplings
derived from the one-body interactions.\par The quantitative
results for the coupling parameters, even if agreeing in order of
magnitude with some estimates from experimental data (see, e.g.
Ref.13) are obviously model-depending. However, their ratios
should be more close to the reality. In particular, the obtained
values of the various couplings, when compared to the values of
the electronic interactions obtained from the same Wannier
functions\cite{acquarone}
suggests that, for dimer lengths comparable to the lattice parameters
in $HTS$ and $CMR$, only $dU/da$ and $dX/da$ can be safely dropped,
while neglecting {\it any} of the other
electron-phonon interactions is a questionable approximation. \par
Finally, we have been able to determine the correct form of the
admissible electron-phonon coupling one-body terms in the
interacting Hamiltonian. In the literature one sometimes finds
proposals for such terms which are incompatible with our results.
This is the case, for instance, of the electron-phonon Hamiltonian
of Refs.14,15.
\nonumsection{Acknowledgements}
\noindent It is a pleasure to
thank J.R. Iglesias, M. A. Gusm\~{a}o,
A. Painelli, M. Cococcioni, A. Alexandrov, and particularly A.A. Aligia,
for critical discussions and comments. This work was supported by
I.N.F.M. and by MURST 1997 co-funded project "Magnetic Polarons in
Manganites" .
\par
\vfill\eject
\section{Appendix 1}
\noindent Here we list some results of frequent use in the
calculations. We shall use the zero-displacement derivatives of
$A$ and $B$: \begin{equation} {{\partial A} \over{ \partial u}}\Bigg|_{u \rightarrow
0}= {{a \Gamma^2 S_0}\over{4}} \left[ - (1+S_0)^{-3/2}+
(1-S_0)^{-3/2} \right], \label{wandadu}
\end{equation}
\begin{equation} {{\partial B}
\over{ \partial u}}\Bigg|_{u \rightarrow 0}= {{a \Gamma^2 S_0}\over{4}}
\left[ - (1+S_0)^{-3/2}- (1-S_0)^{-3/2}\right].
\label{wandbdu}
\end{equation}
\noindent In the following we shall use the substitution
$t=x+a/2-u_1$ in both $\Psi_1$ and $\Psi_2$, yielding:
\[
\Psi_1(x, u_1,u_2,y,z)\Rightarrow \Psi_1(t, u,y,z)=
A(u)Ne^{-\Gamma^2[t^2+y^2+z^2] }
+ B(u)Ne^{-\Gamma^2[(t-a+u)^2+y^2+z^2]},
\]
\begin{equation} \Psi_2(x,u_1,u_2,y,z)\Rightarrow \Psi_2(t, u,y,z)=
B(u)Ne^{-\Gamma^2[t^2+y^2+z^2]} +
A(u)Ne^{-\Gamma^2[(t-a+u)^2+y^2+z^2]}. \label{wanpsi} \end{equation} To
write down their derivatives in the zero-displacement limit, it is
convenient to revert to $x=t-a/2$ so that: \begin{equation} {{\partial
\Psi_1(x,y,z,u)} \over{\partial u}}\Bigg\vert_{u\rightarrow 0}
=\phi_1(x,y,z) {{\partial A} \over{\partial u}} + \phi_2(x,y,z)
\left[ {{\partial B} \over{\partial u}} -2\Gamma^2(x-{a\over
2})B\right], \label{dpsi1dub} \end{equation} \begin{equation} {{\partial
\Psi_2(x,y,z,u)} \over{\partial u}}\Bigg\vert_{u\rightarrow 0}
=\phi_1(x,y,z) {{\partial B} \over{\partial u}} + \phi_2(x,y,z)
\left[ {{\partial A} \over{\partial u}} -2\Gamma^2(x-{a\over
2})A\right]. \label{dpsi2dub} \end{equation} We shall also use the other
substitution $t\equiv x-a/2-u_2$, yielding:
\[
\Psi_1(x, u_1,u_2,,y,z)\Rightarrow
A(u)Ne^{-\Gamma^2[(t+a-u)^2+y^2+z^2] } +
B(u)Ne^{-\Gamma^2[t^2+y^2+z^2]},
\]
\begin{equation} \Psi_2(x,u_1,u_2,y,z)\Rightarrow
B(u)Ne^{-\Gamma^2[(t+a-u)^2+y^2+z^2]} +
A(u)Ne^{-\Gamma^2[t^2+y^2+z^2]}. \label{wanpsi2}
\end{equation}
In the limit
of vanishing deformations it follows, after reverting
to $x=t+a/2$:
\begin{equation} {{\partial \Psi_1(x,y,z,u)} \over{\partial
u}}\Bigg\vert_{u\rightarrow 0} = \phi_1(x,y,z) \left[ {{\partial A}
\over{\partial u}} +2\Gamma^2(x+{a\over 2})A\right] +\phi_2(x,y,z)
{{\partial B} \over{\partial u}}, \label{dpsi1dub2} \end{equation} \begin{equation}
{{\partial \Psi_2(x,y,z,u)} \over{\partial u}}\Bigg\vert_{u\rightarrow 0}
= \phi_1(x,y,z) \left[ {{\partial B} \over{\partial u}}
+2\Gamma^2(x+{a\over 2})B\right] +\phi_2(x,y,z) {{\partial A}
\over{\partial u}}. \label{dpsi2dub2} \end{equation}
\section{Appendix 2}
\noindent For convenience reasons we shall list here various
matrix elements, all evaluated in the limit of vanishing
deformations, which enter the calculations. Those of them which do
not follow straightforwardly from standard properties of Gaussian
integrals\cite{rizhik}
are evaluated in details in the Appendix 3. By defining:
\begin{equation} F_0(x)={{1} \over{ \sqrt{x} }}\int^{\sqrt{x}}_0e^{-t^2}dt,
\label{hol27} \end{equation} we have: \begin{equation} \langle
\phi_i|-\nabla^2/2|\phi_i\rangle={{3\Gamma^2} \over {2}}+
{{\Gamma^4a^2S_0^2} \over{2(1-S_0^2)}}, \qquad \langle
\phi_i|-\nabla^2/2|\phi_j\rangle=-{{\Gamma^4a^2S_0}
\over{2(1-S_0^2)}}, \qquad (i,j=1,2) \label{matel1} \end{equation} \begin{equation} \langle
\phi_i|V_i|\phi_i\rangle=2\Gamma\sqrt{2/\pi}, \qquad \langle
\phi_i|V_j|\phi_i\rangle=2\Gamma\sqrt{2/\pi}F_0(2a^2\Gamma^2),
\label{matel1b} \end{equation} \begin{equation} \langle
\phi_i|V_i|\phi_j\rangle=2\Gamma\sqrt{2/\pi}S_0F_0
\left({{a^2\Gamma^2}\over{2}}\right), \label{matel2} \end{equation} \begin{equation} \langle
\phi_1|xV_1|\phi_1\rangle=-a\Gamma\sqrt{2/\pi}, \qquad \langle
\phi_2|xV_2|\phi_2\rangle=a\Gamma\sqrt{2/\pi}, \label{matel3} \end{equation}
\[
\langle\phi_1|xV_2|\phi_1\rangle= {{1}\over{2a\Gamma}}\sqrt{{2\over \pi} }
\left[-S_0^4+ \left(1-2a^2\Gamma^2 \right)F_0(2a^2\Gamma^2)
\right],
\]
\begin{equation} \langle\phi_2|xV_1|\phi_2\rangle=-\langle\phi_1|xV_2|\phi_1\rangle,
\label{matel4} \end{equation}
\[
\langle\phi_1|xV_1|\phi_2\rangle= {{ S_0\sqrt{2/\pi} } \over{a\Gamma
}}\left[S_0 -F_0\left({{a^2\Gamma^2}\over{2}}\right)\right],
\]
\begin{equation} \langle\phi_1|xV_2|\phi_2\rangle=-\langle\phi_1|xV_1|\phi_2\rangle.
\label{matel5} \end{equation} Some other useful relations are: \begin{equation} \langle
\phi_1|V_1|\phi_1\rangle = \langle \phi_2|V_2|\phi_2\rangle, \qquad \langle
\phi_2|V_1|\phi_2\rangle = \langle \phi_1|V_2|\phi_1\rangle. \label{hol12}
\end{equation} We can directly evaluate \begin{equation} \langle\phi_1|xV_1|\phi_1\rangle = -a
\Gamma \sqrt{2/\pi}= - \langle\phi_2|xV_2|\phi_2\rangle. \label{hol13}
\end{equation} Indeed, due to the odd parity of the kernel,
$\langle\phi_1|(x+a/2)V_1|\phi_1\rangle =0$ from which \Eq{\ref{hol13}}
follows. We can also prove that \begin{equation} \langle\phi_1|xV_2|\phi_2\rangle = -
\langle\phi_2|xV_1|\phi_1\rangle, \qquad \langle\phi_1|xV_2|\phi_1\rangle = -
\langle\phi_2|xV_1|\phi_2\rangle. \label{hol14} \end{equation} Indeed, if $x$
changes sign, then $V_1\rightarrow V_2$ and $\phi_1\rightarrow\phi_2$. Therefore
the functions $\phi_1(V_1+V_2)\phi_2$ and $ \phi_1^2V_2+
\phi_2^2V_1$ are even in $x$, so that:
\[
\langle\phi_1|xV_2|\phi_2\rangle + \langle\phi_2|xV_1|\phi_1\rangle =
\langle\phi_1|x(V_1+V_2)|\phi_2\rangle= 0,
\]\begin{equation}
\langle\phi_1|xV_2|\phi_1\rangle + \langle\phi_2|xV_1|\phi_2\rangle =
\int^\infty_{-\infty} x(\phi_1^2V_2+ \phi_2^2V_1)dxdydz =0,
\label{hol15} \end{equation} because the integrands are odd functions of x.
\vfill\eject
\section{Appendix 3}
\noindent Let us now evaluate the unknown integrals in \Eq{\ref{hol6}}. To do
that, consider
\[
\langle\Psi_1|V_1|\Psi_1\rangle=
\int^\infty_{-\infty}\Psi_1(a/2,x,y,z)^2V_1(a/2,x,y,z)dxdydz,
\]
and its derivative with respect to $a/2$:
\[
{{\partial\langle\Psi_1|V_1|\Psi_1\rangle} \over{\partial(a/2)}}=
2\int^\infty_{-\infty}\Psi_1
\left[ {{\partial \Psi_1} \over{ \partial(a/2)}} \right]
V_1dxdydz
+\int^\infty_{-\infty}\Psi_1^2
{{\partial V_1} \over{ \partial(a/2)}} dxdydz
\]
\begin{equation} = 2\int^\infty_{-\infty}\Psi_1 \left[ {{\partial \Psi_1}
\over{ \partial(a/2)}} \right] V_1dxdydz - g_0^{(1)},
\label{hol17} \end{equation} where we use the \Eq{\ref{hol4}} and
$\lim_{u_1\rightarrow 0} {{\partial V_1} / {\partial u_1}}=
-{{\partial V_1} / {\partial x}}= -{{\partial V_1} / {\partial (a/2)}}$.
To evaluate the first integral we need also:
\begin{equation} {{\partial \Psi_1} \over{ \partial(a/2)}} ={{\partial A}
\over{ \partial(a/2)}}\phi_1 + {{\partial B} \over{
\partial(a/2)}} \phi_2 - 2
\Gamma^2\left[A\left(x+a/2\right)\phi_1-B(x-a/2)\phi_2\right].
\label{hol18} \end{equation} Inside the integral, it is convenient to use
$\partial/\partial x$,
instead of $\partial/\partial(a/2)$.
As \begin{equation} {{\partial \Psi_1} \over{ \partial x}}= -2 \Gamma^2
\left[A\left(x+a/2\right)\phi_1+B(x-a/2)\phi_2\right],
\label{hol19} \end{equation} then \begin{equation} {{\partial \Psi_1} \over{
\partial(a/2)}} ={{\partial A} \over{ \partial(a/2)}}\phi_1 +
{{\partial B} \over{ \partial(a/2)}} \phi_2 +{{\partial \Psi_1}
\over{ \partial x}} +4\Gamma^2B(x-a/2)\phi_2, \label{hol20} \end{equation}
so that \Eq{\ref{hol17}} becomes, recalling \Eq{\ref{hol4}}:
\[
{{d\langle\Psi_1|V_1|\Psi_1\rangle} \over{d(a/2)}}=
2{{\partial A} \over{ \partial(a/2)}}
\int^\infty_{-\infty}\Psi_1\phi_1V_1 dxdydz
+ 2{{\partial B} \over{ \partial(a/2)}}
\int^\infty_{-\infty}\Psi_1\phi_2V_1 dxdydz
\]
\begin{equation} +8\Gamma^2B\int^\infty_{-\infty}\Psi_1(x-a/2)\phi_2V_1dxdydz.
\label{hol21} \end{equation} Developing $\Psi_1=A\phi_1+B\phi_2$
and reordering, one arrives at:
\[
8\Gamma^2 \left(
AB\langle\phi_1|xV_1|\phi_2\rangle+B^2\langle\phi_2|xV_1|\phi_2\rangle \right)=
{{\partial \langle\Psi_1|V_1|\Psi_1\rangle} \over{\partial (a/2)}}
-{{\partial A^2} \over{ \partial(a/2)}}\langle\phi_1|V_1|\phi_1\rangle
\]
\begin{equation} +\left[4a\Gamma^2B^2- {{\partial B^2} \over{ \partial(a/2)}}
\right]\langle\phi_2|V_1|\phi_2\rangle + \left[4a\Gamma^2AB -2{{\partial
(AB)} \over{ \partial(a/2)}} \right]\langle\phi_1|V_1|\phi_2\rangle.
\label{hol22} \end{equation} This equation connects the unknown integrals on
the left hand side to known ones. We need another relation, which
is provided by similar manipulations on $\langle\Psi_2|V_1|\Psi_2\rangle$.
\[
{{ \partial\langle\Psi_2|V_1|\Psi_2\rangle} \over{ \partial(a/2)}}
={{ \partial} \over{ \partial(a/2)}}\left [\int^\infty_{-\infty}
\Psi_2(a/2,x,y,z)^2V_1(a/2,x,y,z)dxdydz \right]
\]
\begin{equation}
=2\int^\infty_{-\infty} \Psi_2 {{ \partial\Psi_2} \over{
\partial(a/2)}}V_1dxdydz
+\int^\infty_{-\infty} \Psi_2^2 {{ \partial V_1} \over{
\partial(a/2)}}dxdydz. \label{hol23} \end{equation} After substituting $
\partial V_1/ \partial (a/2)=
\partial V_1/ \partial x$ the second
integral is done by parts and one arrives at: \begin{equation} {{
\partial\langle\Psi_2|V_1|\Psi_2\rangle} \over{ \partial(a/2)}}
=2\int^\infty_{-\infty} \Psi_2 \left[ {{ \partial\Psi_2} \over{
\partial(a/2)}} - {{ \partial\Psi_2} \over{ \partial x}} \right]
V_1dxdydz. \label{hol24} \end{equation}
>From the definition of $\Psi_2$ one
has:
\[
{{\partial \Psi_2} \over{ \partial(a/2)}} ={{\partial B} \over{
\partial(a/2)}}\phi_1 + {{\partial A} \over{ \partial(a/2)}}
\phi_2 - 2
\Gamma^2\left[B\left(x+a/2\right)\phi_1-A(x-a/2)\phi_2\right],
\]
\begin{equation} {{\partial \Psi_2} \over{ \partial x}}= -2 \Gamma^2
\left[B\left(x+a/2\right)\phi_1+A(x-a/2)\phi_2\right],
\label{hol24a} \end{equation} so that \begin{equation} {{ \partial\Psi_2} \over{
\partial(a/2)}} - {{ \partial\Psi_2} \over{ \partial x}}= {{
\partial B} \over{ \partial(a/2)}} \phi_1 +{{ \partial A} \over{
\partial(a/2)}} \phi_2 +4\Gamma^2A(x-a/2)\phi_2. \label{hol25} \end{equation}
Substituting \Eq{\ref{hol25}} into \Eq{\ref{hol24}}, developing
$\Psi_2=B\phi_1+A\phi_2$ and reordering yields finally:
\[
8\Gamma^2 \left ( AB\langle\phi_1|xV_1|\phi_2\rangle+A^2\langle\phi_2|xV_1|\phi_2\rangle
\right)=
{{\partial\langle\Psi_2|V_1|\Psi_2\rangle} \over{\partial(a/2)}}
-{{\partial B^2} \over{ \partial(a/2)}}
\langle\phi_1|V_1|\phi_1\rangle
\]
\begin{equation} +\left[4a\Gamma^2A^2- {{\partial A^2} \over{ \partial(a/2)}}
\right]\langle\phi_2|V_1|\phi_2\rangle + \left[4a\Gamma^2AB -2{{\partial
(AB)} \over{ \partial(a/2)}} \right]\langle\phi_1|V_1|\phi_2\rangle,
\label{hol26} \end{equation} which is the second equation needed to evaluate
the integrals entering the \Eq{\ref{hol6}} for $g_0^{(1)}$. For
the explicit evaluation we need also:
\[
\langle\Psi_1|V_1|\Psi_1\rangle=
2\Gamma\sqrt{2/\pi}\left[A^2+B^2F_0(2\Gamma^2a^2)+2ABSF_0(\Gamma^2a^2/2)\right],
\]
\begin{equation} \langle\Psi_2|V_1|\Psi_2\rangle=
2\Gamma\sqrt{2/\pi}\left[B^2+A^2F_0(2\Gamma^2a^2)
+2ABSF_0(\Gamma^2a^2/2)\right]. \end{equation} \noindent It is convenient to
subtract \ the \Eq{\ref{hol22}} from \Eq{\ref{hol26}} yielding:
\[
8\Gamma ^{2}\left( A^{2}-B^{2}\right) <\phi _{2}|xV_{1}|\phi
_{2}>=\frac{d} { d\left( a/2\right) }\left[ <\Psi _{2}|V_{1}|\Psi
_{2}>-<\Psi _{1}|V_{1}|\Psi _{1}>\right]
\]
\begin{equation}
+\frac{d\left( A^{2}-B^{2}\right) }{d\left( a/2\right) }<\phi
_{1}|V_{1}|\phi _{1}>+\left[ 4\Gamma ^{2}a\left( A^{2}-B^{2}\right) -\frac{%
d\left( A^{2}-B^{2}\right) }{d\left( a/2\right) }\right] <\phi
_{2}|V_{1}|\phi _{2}>, \label{cadd1}
\end{equation}
\noindent
while their sum yields:
\[
16\Gamma ^{2}AB<\phi _{1}|xV_{1}|\phi _{2}>+8\Gamma ^{2}\left(
A^{2}+B^{2}\right) <\phi _{2}|xV_{1}|\phi _{2}>=
\]
\[
\frac{d}{d\left( a/2\right) }\left[ <\Psi _{2}|V_{1}|\Psi _{2}>+<\Psi
_{1}|V_{1}|\Psi _{1}>\right] -\frac{d\left( A^{2}+B^{2}\right) }{d\left(
a/2\right) }<\phi _{1}|V_{1}|\phi _{1}>+
\]
\[
+\left[ 4\Gamma ^{2}a\left( A^{2}+B^{2}\right) -\frac{d\left(
A^{2}+B^{2}\right) }{d\left( a/2\right) }\right] <\phi _{2}|V_{1}{}|\phi
_{2}>
\]
\begin{equation}
+\left[ 8\Gamma ^{2}aAB-4\frac{d\left( AB\right) }{d\left(
a/2\right) }\right] <\phi _{1}|V_{1}{}|\phi _{2}>. \label{cadd2}
\end{equation}
\noindent To proceed, one has
\[
<\Psi _{2}|V_{1}|\Psi _{2}>-<\Psi _{1}|V_{1}|\Psi _{1}>=-\Gamma \sqrt{\frac{2%
}{\pi }}\left( A^{2}-B^{2}\right) \left[ 1-F_{0}\left( 2\Gamma
^{2}a^{2}\right) \right],
\]
\[
<\Psi _{2}|V_{1}|\Psi _{2}>+<\Psi _{1}|V_{1}|\Psi _{1}>=
\]
\begin{equation}
\Gamma \sqrt{\frac{2}{\pi }}\left\{\left( A^{2}+B^{2}\right)
\left[ 1+F_{0}\left( 2\Gamma ^{2}a^{2}\right) \right]
+4ABSF_{0}\left( \Gamma ^{2}a^{2}/2\right) \right\}. \label{cadd3}
\end{equation}
The derivatives of interest are:
\begin{equation}
\frac{\partial S}{\partial \left( a/2\right) }=-2\Gamma ^{2}aS,
\end{equation}
\begin{equation}
\frac{\partial \left( A^{2}+B^{2}\right) }{\partial \left( a/2\right) }=-%
\frac{4\Gamma ^{2}aS^{2}}{\left( 1-S^{2}\right) ^{2}}, \qquad
\frac{\partial \left( A^{2}-B^{2}\right) }{\partial \left( a/2\right) }=-%
\frac{2\Gamma ^{2}aS^{2}}{\left( 1-S^{2}\right) ^{3/2}}, \qquad
\frac{\partial AB}{\partial \left( a/2\right) }=\frac{\Gamma
^{2}aS\left( 1+S^{2}\right) }{\left( 1-S^{2}\right) ^{2}},
\label{cadd4}
\end{equation}
\begin{equation}
\frac{\partial }{\partial \left( a/2\right) }\left[ F_{0}\left(
2\Gamma ^{2}a^{2}\right) \right] =\frac{2}{a}\left[
S^{4}-F_{0}\left( 2\Gamma ^{2}a^{2}\right) \right], \qquad
\frac{\partial }{\partial \left( a/2\right) }\left[ F_{0}\left(
{{\Gamma^{2}a^{2}}\over{2}}\right) \right] =
\frac{2}{a}\left[ S-F_{0}\left( {{\Gamma^{2}a^{2}}\over{2}}\right)
\right]. \label{cadd5}
\end{equation}
Therefore we obtain:
\[
\frac{\partial }{\partial \left( a/2\right) }\left[<\Psi _{2}|V_{1}|\Psi
_{2}>-<\Psi _{1}|V_{1}|\Psi _{1}>\right]=
\]
\begin{equation}
\left[ \frac{2\Gamma \sqrt{2/\pi }}{a\left( 1-S^{2}\right) ^{3/2}}%
\right] \left\{ \Gamma ^{2}a^{2}S^{2}\left[ 1-F_{0}\left( 2\Gamma
^{2}a^{2}\right) \right] +\left( 1-S^{2}\right) \left[
S^{4}-F_{0}\left( 2\Gamma ^{2}a^{2}\right) \right] \right\},
\label{cadd6}
\end{equation}
and
\[
\frac{\partial }{\partial \left( a/2\right) }[<\Psi _{2}|V_{1}|\Psi
_{2}>+<\Psi _{1}|V_{1}|\Psi _{1}>]=
\]
\[
\left[ \frac{2\Gamma \sqrt{2/\pi }}{a\left( 1-S^{2}\right) ^{2}}%
\right] \{-2\Gamma ^{2}a^{2}S^{2}-\left( 2-S\right) S^{3}\left(
1-S^{2}\right) -F_{0}\left( 2\Gamma ^{2}a^{2}\right) \left[ 1-S^{2}\left(
1-2\Gamma ^{2}a^{2}\right) \right]
\]
\begin{equation}
+F_{0}\left( \frac{\Gamma ^{2}a^{2}}{2}\right) \left[ 4\Gamma
^{2}a^{2}S^{2}+2S^{2}\left( 1-S^{2}\right) \right] \}.
\label{cadd7}
\end{equation}
By substituting the equations above deduced and the other matrix
elements into the \Eq{\ref{cadd1}} and \Eq{\ref{cadd2}} we obtain
the unknown matrix elements as:
\begin{equation} \langle\phi_2|\left(x-{a \over 2}\right)V_1|\phi_2\rangle=
{{\sqrt{2/\pi} }
\over{2a\Gamma}}\left[S^4-F_0(2a^2\Gamma^2)\right], \label{hol27a}
\end{equation} \begin{equation} \langle\phi_1|xV_1|\phi_2\rangle= {{ S\sqrt{2/\pi} }
\over{a\Gamma }}\left[S
-F_0\left({{a^2\Gamma^2}\over{2}}\right)\right]. \label{hol27b}
\end{equation}
\section{Appendix 4}
Let us evaluate the crystal field coupling for site $1$ in the
adiabatic limit, in order to verify Eq.\ref{gcfaddef}.
For the charge on site 1 the variation of the local energy defines
$g_{cfad}^{(1)}$ as:
\[
\int^\infty_{-\infty}\Psi_1(\Rv_1^0-\uv^{~}_1,\Rv_2^0-\uv^{~}_2,\bfm{r})^2V_2(\Rv_2^0-\uv^{~}_2,\bfm{r})d^3\bfm{r}
-\int^\infty_{-\infty}\Psi_1(\Rv_1^0,\Rv_2^0,\bfm{r})^2V_2(\Rv_2^0,\bfm{r})d^3\bfm{r}
\]
\begin{equation} \equiv g_{cfad}^{(1)}n_1(u_2-u_1)+\cal{O} [(u_2-u_1)^2].
\label{cfad1} \end{equation} Substituting $t=x-a/2-u_2$ and developing
$\Psi_1(u,t,y,z)^2$ to first order in $u$ yields:
\begin{equation} g_{cfad}^{(1)}=2 \int^\infty_{-\infty} \Psi_1(0,t,y,z)
{{\partial \Psi_1} \over{\partial(u)}}\bigg\vert_{u=0} V_2(t,y,z)
\bullet \bfm{e}_{12}. \label{cfad4} \end{equation} By substituting
$t=x-a/2-u_2$ we obtain:
\[
{{\partial \Psi_1} \over{\partial u}}\bigg\vert_{u=0}=
{{\partial A} \over{\partial u}} \bigg\vert_{u=0}
Ne^{-\Gamma^2[(t+a)^2+y^2+z^2]}
+{{\partial B} \over{\partial u}} \bigg\vert_{u=0}
Ne^{-\Gamma^2[t^2+y^2+z^2]}
\]
\begin{equation} -2\Gamma^2A(t+a)Ne^{-\Gamma^2[(t+a)^2+y^2+z^2]}.
\label{cfad3} \end{equation} Substituting $\Psi_1$ and \Eq{\ref{cfad3}} into
\Eq{\ref{cfad4}} yields:
\[
g_{cfad}^{(1)}=
{{\partial (A^2)} \over{\partial u}}\bigg\vert_{u=0}\langle\phi_1|V_2|\phi_1\rangle +
{{\partial (B^2)} \over{\partial u}}\bigg\vert_{u=0}\langle\phi_2|V_2|\phi_2\rangle +
2 {{\partial (AB)} \over{\partial u}}\bigg\vert_{u=0}\langle\phi_1|V_2|\phi_2\rangle
\]
\begin{equation} +4\Gamma^2\left[ A_0^2 \langle\phi_1|(x+a/2)V_2|\phi_1\rangle +
2A_0B_0\langle\phi_1|xV_2|\phi_2\rangle \right]. \label{cfad10} \end{equation} The
corresponding coupling for the other site is obtained by the site
permutation $1 \rightleftharpoons 2$, \begin{equation}
H_{cfad}^{(2)}=g_{cfad}^{(2)}\sum_\sigma n_{2\sigma}(u_1-u_2),
\label{cfad10b} \end{equation} that is, with $t=x+a/2-u_1$:
\begin{equation} g_{cfad}^{(2)} =2\int^\infty_{-\infty} \Psi_2(0,t,y,z)
{{\partial \Psi_2} \over{\partial(u_1-u_2)}}\bigg\vert_{0}
V_1(t,y,z). \label{cfad11} \end{equation} Performing the corresponding
calculations we obtain:
\[
g_{cfad}^{(2)} = g_{cfad}^{(1)} -4\Gamma^2\Biggl\{
AB \langle\phi_1|x(V_1+V_2)|\phi_2\rangle +A^2
(\left[ \langle\phi_1|xV_2|\phi_1\rangle + \langle\phi_2|xV_1|\phi_2\rangle) \right]
\]
\begin{equation} +B^2 (\left[ \langle\phi_2|xV_2|\phi_2\rangle +
\langle\phi_1|xV_1|\phi_1\rangle) \right] \Biggr \}. \label{cfad12} \end{equation}
The difference between $g_{cfad}^{(2)}$ and $ g_{cfad}^{(1)}$
vanishes because of the results reported in the Appendix 2, \Eq{
\ref{hol15}}.
\nonumsection{References}
\noindent
|
\section{Introduction}
\label{Intro}
Some twenty years ago, Heiles (\markcite{79}1979, \markcite{84}1984,
\markcite{90}1990) discovered large, shell--like structures in the
neutral hydrogen (\ion{H}{1}) distribution of our Galaxy by inspection
of maps made in the 21--cm line. Since this discovery, a wealth of
observations obtained with powerful Synthesis Radio Telescopes such as
the Very Large Array (VLA) or the Westerbork Synthesis Radio Telescope
(WSRT) revealed similar structures in our nearest neighbors. Prominent
examples are M~31 (Brinks \markcite{BRI81}1981, Brinks \& Bajaja
\markcite{BRI86}1986), M~33 (Deul \& den Hartog \markcite{DEU90}1990),
Holmberg~II (Puche \emph{et al.\ } \markcite{PUC92}1992, hereafter referred to as
PWBR92), the galaxies M~101 and NGC~6946 (Kamphuis
\markcite{KAM93}1993) and IC~10 (Wilcots \& Miller
\markcite{WIL89}1998). The latest, impressive additions to this list
are the SMC (Staveley--Smith \emph{et al.\ } \markcite{STA97}1997) and LMC (Kim
\emph{et al.\ } \markcite{KIM98}1998). All these observations indicate that the
interstellar medium (ISM) of medium to late--type galaxies is
dominated by features, which are variously described as shells, rings,
holes, loops, bubbles or cavities.
Obviously, one of the basic questions is which physical process is
able to produce these observed structures. The general picture which
has emerged is that they are the result of combined stellar winds and
supernova explosions produced by young stellar associations. For
review articles on that topic see Tenorio--Tagle \& Bodenheimer
\markcite{TEN88}(1988), and van der Hulst \markcite{HUL96}(1996, and
references therein). Based on a simple model of holes created by O and
B stars, Oey \& Clarke \markcite{OEY97}(1997) successfully predict the
observed number distribution of SMC holes, lending support to this
hypothesis. It should be noted, however, that this general picture is
not without its flaws. Searches for the remnant stellar populations of
the OB associations thought to be responsible for some of the holes in
Ho~II have not led to the expected result (Rhode \emph{et al.\ }
\markcite{RHO97}1999). And in the case of the largest observed shells,
these seem to have energy requirements surpassing the output of
stellar winds and supernovae. To explain those structures an
alternative mechanism was proposed, the infall of gas clouds (see
Tenorio--Tagle \emph{et al.\ } \markcite{TEN87}(1987) for a numerical simulation
and van der Hulst \& Sancisi \markcite{HUL88}(1988) for observational
evidence in the case of one of the largest holes in M~101). And
recently, even Gamma Ray Bursters (GRBs) have been suggested as a
possible explanation for the largest \ion{H}{1} holes (Efremov,
Elmegreen \& Hodge \markcite{EFR98}1998, Loeb \& Perna
\markcite{LOE98}1998).
The idea that supernovae deposit large amounts of energy into the ISM
is not new. Cox \& Smith \markcite{COX74}(1974) already discussed the
idea that the energy released by supernovae sets up tunnels of hot gas
in the ISM. This picture was a substantial modification of the Field,
Goldsmith \& Habing \markcite{FIE69}(1969) model which consists of a
two phase medium. The ideas proposed by Cox \& Smith were incorporated
within the model elaborated by McKee \& Ostriker \markcite{MCK77}(1977)
who suggested that the ISM has three distinct phases, a cool
(T$\,\approx10^2\,$~K), a warm (T$\,\approx10^4\,$~K), and a hot phase
(T$\,\approx10^6\,$~K), all in pressure equilibrium. Various
refinements have been proposed but this picture basically still stands.
Knowing how much kinetic energy is transferred in expanding shells and
filaments is of direct importance for assessing the energy balance in
the ISM. In addition to an analytical approach, various authors are
pushing forward numerical simulations (e.g., Palou\v{s}, Franco \&
Tenorio--Tagle \markcite{PAL90}1990, Silich \emph{et al.\ } \markcite{SIL96}1996,
Palou\v{s}, Tenorio--Tagle \& Franco \markcite{PAL94}1994), in order
to model the evolution of the shape of the \ion{H}{1} shells in
various environments, the fragmentation of matter on the rim of the
shells and to investigate induced star formation. This is important
not only for tracking down the mechanisms which produce the
\ion{H}{1}\ shells and filaments, but also for determining what
fraction of the ISM breaks out into the halo and might eventually rain
back onto the disk.
Our aim is to investigate the properties of the \ion{H}{1}\ shells and
bubbles in the ISM of galaxies. Intimately linked to this are the
questions when fragmentation on the rim of the shells starts and what
drives star formation. Unfortunately, the determination of the
characteristics of \ion{H}{1} shells in our Galaxy is not
straightforward due to our unfavorable location within the disk.
Looking at the nearest spiral galaxies (e.g., M~31, M~33, M~101,
NGC~6946) does not improve matters all that much since the structures
in their ISM are strongly modified by spiral density waves and
differential rotation, both effects dramatically influencing the shape
and formation of holes and shells. Therefore, we decided to direct our
attention to nearby dwarf galaxies instead. Dwarfs are slow rotators,
generally show solid body rotation, and lack density waves. This
implies that once features like shells have formed, they won't be
deformed by galactic shear and therefore tend to be long
lived. Moreover, the overall gravitational potential of a dwarf is
much smaller than in a normal spiral. The same amount of energy input
of a star forming region therefore has a more pronounced impact on the
overall appearance of the ISM, as shown by PWBR92. Since dwarf
galaxies also tend to have a much thicker disk compared to massive
spirals (e.g., Staveley--Smith, Davies \& Kinman
\markcite{STA92}1992), the \ion{H}{1}\ volume density of dwarf
galaxies is lower, again facilitating the creation of large holes. And
because of their smaller size, the probability to suffer impacts from
infalling clouds is lower compared to that for larger
galaxies. Despite all these arguments in favor, only very few such
\ion{H}{1} studies of dwarf galaxies have been published. Beside our
two neighbors, the SMC and LMC a detailed analysis was performed in
one case only, that of the dwarf galaxy Holmberg~II (PWBR92), member
of the M~81 group.
In this paper, we concentrate on IC~2574, a gas--rich dwarf galaxy in
the same group. Table~\ref{ic_info} summarizes some general
information on this object. IC~2574 (= UGC~5666 = DDO~81, also
sometimes referred to as Coddington's Nebula) has been studied by
several authors in the past. Observations with single dish telescopes
in the 21--cm line of neutral hydrogen (\ion{H}{1}) were published by
Rots \markcite{ROT80}(1980) and Huchtmeier \& Richter
\markcite{HUC88}(1988) and a first study using an interferometer was
made by Seielstad \& Wright \markcite{SEI73}(1973). A recent
\ion{H}{1} study, concentrating on the dynamics and contribution to
the total mass by dark matter was performed by Martimbeau, Carignan \&
Roy \markcite{MAR94}(1994, hereafter abbreviated MCR94), who observed
the galaxy with the Westerbork Synthesis Radio Telescope. Recent
broadband optical studies were published by Tikhonov \emph{et al.\ }
\markcite{TIK91}(1991). Studies of \ion{H}{2} regions in IC~2574 have
been presented by Miller \& Hodge (\markcite{MIL94}1994), Kennicutt
\markcite{KEN88}(1988) and Arsenault \& Roy
\markcite{ARS86}(1986). Spectroscopy of some prominent \ion{H}{2}
regions was performed by Miller \& Hodge (\markcite{MIL96}1996) and
H$\alpha$ velocity fields were recently obtained by Tomita \emph{et al.\ }\
(\markcite{TOM98}1998). In addition, four Wolf--Rayet stars were
discovered by Drissen, Roy \& Moffat (\markcite{DRI93}1993).
\placetable{ic_info}
We present \ion{H}{1}\ observations made with the
NRAO\footnote{The National Radio Astronomy Observatory (NRAO) is
operated by Associated Universities, Inc., under cooperative agreement
with the National Science Foundation.} Very Large Array (VLA) in its
B--, C-- \& D--array configurations supplemented by optical
observations obtained at the Calar Alto Observatory.
The VLA data were originally obtained by Puche, Brinks
and Westpfahl with the intention of following up on their work on Ho
II. So far, only a low resolution integrated \ion{H}{1} map was
published by Puche \& Westpfahl \markcite{PUC94}(1994). This paper
presents the VLA data in a complete form and compares them with
optical observations. The original data were recalibrated by us to
produce far superior maps using the latest calibration and imaging
techniques within {\sc aips}. The data reduction and results are discussed
in Secs.~\ref{obser}~\&~\ref{histu}. In Sec.~\ref{hihol}, a catalog
of \ion{H}{1} holes and shells is presented. Sec.~\ref{ha-hi}
discusses the relation between \ion{H}{1} and H$\alpha$ features. A
general discussion is given in Sec.~\ref{discu} and a summary of the
results is presented in Sec.~\ref{summa}.
\section{Observations}
\label{obser}
\subsection{Radio Observations}
IC~2574 was observed with the VLA in B--, C-- and
D--configuration. Part of the D--array observations were affected by
solar interference, which was removed by deleting {\it uv--}spacings
shorter than typically 0.4 k$\lambda$ and editing out bad
time--ranges. In total, 17 hours were spent on source which were
divided into 4 hours for D--array, 2.5 hours in C--array and 10.5
hours in B--array (see Table~\ref{VLAsetup} for a detailed description
of the VLA observations). The longer observing time in B--array only
partly compensates for the lower surface brightness sensitivity in
this extended configuration. The absolute flux calibration was
determined by observing 1328+307 (3C286) for approximately 20 minutes
during each observing run, assuming a flux density of 14.73 Jy
according to the Baars \emph{et al.\ } \markcite{BAA77}(1977) scale. This
calibrator was also used to derive the complex bandpass
corrections. The nearby calibrators 1031+567 and 0945+664 were used as
secondary amplitude and phase calibrators and their fluxes were
determined to be 1.80 Jy and 2.22 Jy, respectively. Since the systemic
velocity of IC~2574 ($\approx 53$ $\,\mbox{km}\,\mbox{s}^{-1}$) is close in velocity to
\ion{H}{1} emission from our Galaxy, each calibrator observation was
split in two parts to form a pair of observations, one part having the
velocity shifted by +300 $\,\mbox{km}\,\mbox{s}^{-1}$\ and the other by -300 $\,\mbox{km}\,\mbox{s}^{-1}$\ to avoid
contamination by Galactic emission. In the course of the calibration
these observations were then averaged to give interpolated amplitude
and phase corrections. IC~2574 was observed using a 1.56 MHz bandwidth
centered at a heliocentric velocity of 38 $\,\mbox{km}\,\mbox{s}^{-1}$. This band was divided
into 128 channels resulting in a velocity resolution of 2.58 $\,\mbox{km}\,\mbox{s}^{-1}$\
after online Hanning smoothing.
The data for each array were edited and calibrated separately with the
{\sc aips} package\footnote{The Astronomical Image Processing System
({\sc aips}) has been developed by the NRAO.}. The {\it uv--}data were
inspected and bad data points due to either interference or cross talk
between antennae were removed, after which the data were
calibrated. We Fourier transformed our B--, C-- and D--array
observations separately to assess their quality; subsequently we
combined all data to form a single dataset which was used for
mapping. All the results presented in this paper are based on this
combined dataset.
Channels with velocities lower then $-40$ $\,\mbox{km}\,\mbox{s}^{-1}$\ and higher than $+136$
$\,\mbox{km}\,\mbox{s}^{-1}$\ were found to be line free; these channels determined the
continuum emission which was subtracted in the {\it uv--}plane. The same
channels were used to produce a radio continuum map of IC~2574.
After that, two sets of datacubes (1024 $\times$ 1024 pixels $\times$
68 channels each) were produced using the task {\sc imagr} in {\sc
aips}, each of them {\sc clean}ed to a level of two times the rms
noise (H\"ogbom \markcite{HOG74}1974, Clark \markcite{CLA80}1980). The
first data set was made with natural weighting, leading to a
resolution of $12.4'' \times 11.9''$ and emphasizing large scale
structures. A second cube was produced using the {\sc robust}
weighting scheme (Briggs \markcite{BRI95}1995). This scheme achieves
the high sensitivity of natural weighting combined with a well behaved
synthesized beam, at a resolution close to that of uniform
weighting. To obtain the optimum {\sc robust} value for our
observation, the {\sc robust} parameter space was searched by
producing maps in the sensitive regime of $-2<${\sc robust}$<2$ and
measuring for each map the rms noise and beamsize. Eventually, we
chose a value of {\sc robust}=0.25, resulting in a beamsize of $6.4''
\times 5.9''$ and an rms noise of 0.7 mJy beam$^{-1}$. A comparison
with the beamsize of the high resolution uniform map ($4.7''\times
4.3''$) and the rms noise of the low--noise, natural map ($\approx 10
\%$ increase in noise) shows the strength of this mapping
algorithm. Unless otherwise mentioned, the results presented in the
following are based on this {\sc robust} cube.
Although IC~2574 is situated at a galactic latitude of 43.6$^{\circ}$,
some of the channels were clearly confused by Galactic emission (in
particular between $+10$ and $-10$ $\,\mbox{km}\,\mbox{s}^{-1}$). Since galactic emission
shows its presence on predominantly large scales, we were able to
remove the most prominent emission features by blanking the {\it uv--}data
out to 0.1 k$\lambda$ in these channels.
To separate real emission from the noise, the following procedure was
applied: the natural weighted data cube was convolved to a circular
beam with a FWHM of $45''$. The smoothed map was then tested at the
$2\sigma$ level; if a pixel fell below this level, the counterpart in
the cube was blanked. After that, the remaining peaks were
inspected. Emission that was present in 3 consecutive channels was
considered to be real while all other remaining spikes were considered
to be noise and blanked. The final result was named the master
cube. This master cube can be used as a mask to blank the natural and
{\sc robust} data cubes. This method ensures that the same regions are
included when inspecting cubes at different resolutions and with
different signal--to--noise ratios.
\placetable{VLAsetup}
\subsection{Optical Observations}
\label{optical}
IC~2574 was observed with the 2.2--m telescope of the Calar Alto
Observatory\footnote{The Calar Alto Observatory is operated by the
Max--Planck--Institute for Astronomy, Heidelberg, jointly with the
Spanish Commission for Astronomy.} on 1999 January 25 and 26 in
Johnson R--band (20 min) as well as in H$\alpha$ (45 min). A $2048
\times 2048$ pixel CCD was used leading to $0.5''$ per pixel. The
seeing was around $1.3''$. A focal reducer (CAFOS) allowed for a field
of view of radius $8'$ to be imaged. The usual calibration steps were
followed using the {\sc iraf} package\footnote{{\sc iraf} is
distributed by National Optical Astronomy Observatories, which is
operated by the Association of Universities for Research in Astronomy
(AURA), Inc., under contract with the National Science Foundation.}.
Each exposure was corrected for zero offset (bias) and was flatfielded
using skyflats. After that, frames were inspected for bad pixels and
shifted to get them aligned. Lastly, the exposures were combined to
eliminate cosmic ray hits and to create a final image. As it turned
out, conditions were not photometric during the observation, hence no
calibration using standard stars was performed. In order to get an
approximate calibration for IC~2574, values were taken from the
literature (\markcite{MAR94}MCR94). The deep R--band image was used
for the continuum subtraction of the H$\alpha$ map. The continuum
level was determined by simply scaling the continuum image relative to
the line plus continuum image such that the foreground stars
disappeared when the two images were subtracted. For the calibration
we relied on the H$\alpha$ fluxes published by Miller \& Hodge
\markcite{MIL94}(1994, hereafter referred to as MH94). In their
paper, these authors identify 289 \ion{H}{2} regions in IC~2574 (for a
thorough discussion on the H$\alpha$ emission and its calibration see
the Appendix in \markcite{MIL94}MH94). Figs.~\ref{ic_r}
and~\ref{ic_ha} show our R--band and the H$\alpha$ image,
respectively.
Coordinates for the maps were obtained using the task {\sc koords}
within the {\sc karma} software package (for a description of {\sc
karma} see Sec.~\ref{hicat}). We used as input an image from the
Digital Sky Survey. Using 14 stars found in our CCD image, it was
possible to determine the coordinate system to an accuracy of better
than one arcsecond, more than adequate for our needs.
\placefigure{ic_r}
\placefigure{ic_ha}
\section{HI Studies}
\label{histu}
\subsection{Global \ion{H}{1}\ Properties}
The channel maps obtained by the robust weighted cube are presented in
Fig.~\ref{ic_cm_1}. As can be seen, most traces of contamination by
Galactic emission were successfully removed. To save space, only every
other channel over the frequency range where line emission is present
is shown. The heliocentric radial velocities (in $\,\mbox{km}\,\mbox{s}^{-1}$) are plotted in
the upper right--hand corner of each plane. The beamsize ($6.4'' \times
5.9''$) is indicated in the top left--hand panel (note that the
beamsize is so small that it cannot be properly reproduced in our
maps). Many holes and shell--like structures are visible in the channel
maps; these features will be the main topic of this paper.
Obtaining the \ion{H}{1} flux of each channel and hence the global
\ion{H}{1}\ profile of IC~2574 is not straightforward. As first
pointed out by J\"ors\"ater \& van Moorsel \markcite{JOR95}(1995),
fully cleaned maps do not exist and any cleaned map consists of the
sum of two maps: one containing the restored clean components and the
other the residual map. In the former, the unit is Jansky per clean
beam area, and in the latter Jansky per dirty beam area. Usually,
fluxes are determined on a cleaned map, assuming that the clean beam
is the correct one for the entire map. In reality, the flux is
calculated correctly only for the cleaned fraction of the map; for the
residual map the flux is overestimated when the dirty beam area is
bigger than the clean beam area (which is usually the case). This
becomes a real problem for extended objects, especially in the case of
combined array data, like for our VLA data set. For a full discussion
on this topic see J\"ors\"ater \& van Moorsel \markcite{JOR95}(1995).
Following their prescription, the real flux of a channel map is given
by:
$$
G=\frac{D \times C}{D-R}
$$
where $C$ is the cleaned flux, and $R$ and $D$ are the 'erroneous'
residual flux and the 'erroneous' dirty flux over the same area of a
channel map, respectively. We followed this procedure for every single
channel to obtain the global \ion{H}{1}\ profile as presented in
Fig.~\ref{ic_glob}. Without this correction, we would have
overestimated the \ion{H}{1} flux by a factor of about two!
Figure~\ref{ic_glob} shows the flux-- and primary--beam corrected
fluxes for every channel of IC~2574 (open squares) plotted together
with the single dish profile published by Rots \markcite{ROT80}(1980)
and the WSRT data by MCR94\markcite{MAR94} (the arrows indicate the
velocity range for which the data points were interpolated due to
contamination by galactic emission). From our data, we derive a total
integrated \ion{H}{1}\ flux of 373 Jy $\,\mbox{km}\,\mbox{s}^{-1}$\ which should be compared
with the value of 399 Jy $\,\mbox{km}\,\mbox{s}^{-1}$\ obtained by Rots
\markcite{ROT80}(1980), who used the Greenbank 300 ft telescope to map
IC~2574. This is excellent agreement, given the uncertainties of
removing Galactic emission and systematic errors in the absolute
calibration. This also means that we hardly miss any \ion{H}{1} flux
due to missing short spacings. Other values obtained by
interferometric observations are: $440\pm 50$ Jy $\,\mbox{km}\,\mbox{s}^{-1}$\ found by
Seielstad \& Wright \markcite{SEI73}(1973), using the Owens Valley
Radio Interferometer and 286 Jy $\,\mbox{km}\,\mbox{s}^{-1}$\ obtained by MCR94 from their
Westerbork observations. As can be seen from Fig.~\ref{ic_glob}, the
new VLA observations recover substantially more flux than the WSRT
data. Adopting a distance of 3.2 Mpc, our flux translates to a total
neutral hydrogen mass of $M_{HI}=(9.0\pm 0.1)\times 10^8\,
\mbox{M}_{\odot}$ for IC~2574.
\placefigure{ic_cm_1}
\notetoeditor{please note that Figure 3 -- the channel maps --
actually consists of 3 Figures (a, b and c) -- the filenames are
walter.fig3an.ps, walter.fig3bn.ps and walter.fig3cn.ps)}
\placefigure{ic_glob}
\subsection{Global Characteristics}
\label{global}
Fig.~\ref{ic_mom0} shows the \ion{H}{1} surface density map of IC~2574
obtained by adding all channels containing \ion{H}{1} emission of the
robust cube, after blanking them with the master cube. Note that the
beamsize is so small that it can hardly be properly reproduced. Note
also that, although many hole--like structures and cavities are
visible, specific features turn out to be less prominent in the total
surface brightness map than in the channel maps. This is related to
the fact that the galaxy is highly inclined, leading to overlapping of
shell--like structures along the line--of--sight in the integrated
map. For a more complete discussion on that topic, see
Sec.~\ref{hicat}.
Fig.~\ref{ic_mom1} shows the velocity field of IC~2574 which was
calculated in the usual manner by taking the first moment. Since the
high resolution velocity field is dominated by small--scale motions
within the ISM, the natural weighted cube was convolved to $30''$ (as
indicated by the beamsize) to put more emphasis on the global
characteristics of the velocity field. The underlying greyscale is a
linear representation of the \ion{H}{1} surface brightness map at full
resolution. Although the overall velocity field shows the typical
behavior for a dwarf galaxy (solid body rotation in the center,
flattening of the rotation curve towards the very edge of the galaxy),
the general rotation is obviously disturbed by noncircular motions
(e.g., note the prominent band of lower velocities to the north east
of the center of the galaxy, parallel to its major axis). These
deviations are linked to holes in the neutral interstellar medium.
Finally, Fig.~\ref{ic_mom2} shows the second moment map, or velocity
dispersion of IC~2574. Although some regions near shells reach values
as high as 15 $\,\mbox{km}\,\mbox{s}^{-1}$, the overall velocity dispersion in the unperturbed
parts is much lower. Using \ion{H}{1} profiles, we derive the velocity
dispersion in the disk by looking at quiescent regions. Several lines
of sight were averaged to give a one sigma velocity dispersion of
$7\pm 1$$\,\mbox{km}\,\mbox{s}^{-1}$\ (similar to what is found in Ho~II). This shows that our
velocity resolution of 2.5 $\,\mbox{km}\,\mbox{s}^{-1}$\ (FWHM) was quite sufficient to
resolve the lines of neutral hydrogen.
\placefigure{ic_mom0}
\placefigure{ic_mom1}
\placefigure{ic_mom2}
In this paper, we focus on the small--scale structure of the ISM in
IC~2574. For our analysis, however, we need to adopt values for the
orientation of the object which are usually derived on the basis of a
kinematical analysis (rotation curve fitting). A full analysis of the
dynamics of IC~2574 based on our observations, however, is beyond the
scope of this paper. Luckily, an adequate analysis has already been
published (see the paper by \markcite{MAR94}MCR94). In the course of
their study they derive a systemic velocity of $58\pm 7$ $\,\mbox{km}\,\mbox{s}^{-1}$, adopt
an inclination of $75^{\circ} \pm 7^{\circ}$ and a position angle of
$52^{\circ}$. Depending on their dynamical model, they find that
$70-90\%$ of the total mass of the galaxy is in the form of dark
matter. Following the same procedure as performed for the dwarf galaxy
II~Zw~33 (Walter \emph{et al.\ } \markcite{WAL97}1997) we carried through a first
analysis of our data confirming the results by MCR94 regarding the
rotation curve and the global orientation of IC~2574. For ease of
comparison we therefore adopted their values throughout this paper.
\subsection{The scale height of the \ion{H}{1} disk}
\label{scale}
The scale height $h$ of the \ion{H}{1} layer of IC~2574 plays an
important role in the determination of the properties of the holes,
e.g., when translating the observed \ion{H}{1} surface density into an
\ion{H}{1} volume density. Here $h$ is defined as the $1\sigma$ value
for a Gaussian distribution of the \ion{H}{1} layer. Recent studies
have shown that in general dwarf galaxies have thicker \ion{H}{1}
disks than more massive spiral galaxies. Their aspect ratio tends to
be of the order of 10:1 rather than 100:1, e.g., Ho~II: $h=625$ pc
(PWBR92) or NGC~5023: $h=460$ pc (Bottema \emph{et al.\ } \markcite{BOT86}1986;
see also Staveley--Smith \emph{et al.\ }\ \markcite{STA92}1992). In the case of
IC~2574 there are also indications that the \ion{H}{1} is distributed
in a thicker disk. MCR94 already noticed that, looking at the apparent
axial ratio of the integrated \ion{H}{1} map and assuming a real
inclination of $75^{\circ}$, the \ion{H}{1} must be distributed in a
somewhat thicker disk. Moreover, the fact that we observe, as we will
argue below, spherically expanding shells with diameters of up to almost
1 kpc implies that the extent of the \ion{H}{1} layer must be of the
same order of magnitude. Hence, the scale height of the \ion{H}{1} gas
should be of order 350--400 pc.
An approximate value for the scale height can be obtained fairly
simply, as shown by PWBR92. We follow their approach here. The scale
height is proportional to the velocity dispersion of the gas and
inversely proportional to the square root of the mass density in the
disk (Kellmann \markcite{KEL72}1972; van der Kruit
\markcite{KRU81}1981). For a density distribution that can be
described as $\rho(z,R)=\rho(0,R)\,\mbox{sech}^2(z/z_0)$, $z_0$ is
given by
$$
z_0(R)=\frac{\sigma_{gas}}{\sqrt{2\pi\,G\,\rho(0,R)}}
$$ where $R$
is the galactocentric distance, $\sigma_{gas}$ the velocity dispersion
of the gas in quiescent regions of the galaxy (Sec.~\ref{global}), and
$\rho(R)$ the total volume density (gas, stars and dark matter) of the
galaxy in the plane. However, since $\mbox{sech}^2(z/z_0)\approx
\mbox{exp}(-z^2/z_0^2)$ (to within 5$\%$ in the region of interest) we
can also assume a Gaussian distribution in $z$. The $1\sigma$ scale
height $h$ is then defined as $h = z_0 / \sqrt{2}$. To calculate the
average density of IC~2574 in the plane we assume that the total mass
of the galaxy can be estimated from the last measured point
($R_{max}$) of the observed rotation curve (MCR94) and that this is
distributed over a spherical volume of radius $R_{max}$. MCR94 derive
a rotation curve with $v(R_{max})=66$ $\,\mbox{km}\,\mbox{s}^{-1}$\ at $R_{max}=8.0$ kpc. This
corresponds to a total dynamical mass of $8.3 \times 10^9$
M$_{\odot}$. Distributing this mass over a spherical volume with
radius $R_{max}$ gives an average value for $\rho$ of 0.0039
M$_{\odot}$ pc$^{-3}$. Since this is only the mean density of the
galaxy and the density near the plane is certainly higher, we take
twice this derived value (see PWBR92 for a justification) to get a
first approximation to the real volume density $\rho(R)$ near the
center. Using this approach, we find $\rho(R)=0.008$ M$_{\odot}$
pc$^{-3}$.
Substituting the observed velocity dispersion of 7 $\,\mbox{km}\,\mbox{s}^{-1}$\ (see
Sec.~\ref{global}) in the equation given above we derive a scale
height of $h=345$ pc for IC~2574. Note that this value, given all the
uncertainties in the derivation, is in good agreement with the value
derived based on the sizes of the largest holes ($h=350-400$ pc). We
therefore adopted a scale height of $h=350$ pc for the \ion{H}{1}
layer of IC~2574.
We can now calculate the volume density of the \ion{H}{1} from the
projected and corrected to face--on \ion{H}{1} surface density. The
surface density profile was obtained by elliptically averaging the
flux-- and primary--beam corrected total \ion{H}{1} map (assuming an
inclination of 75$^{\circ}$). Again, this is correct to first order
only as in the case of a thick \ion{H}{1} distribution the
deprojection of the observed projected \ion{H}{1} distribution is not
trivial. We then calculated the volume density $n_{HI}(R)$ of
the \ion{H}{1} in the plane of the disk using
$$
N_{HI}(R)=\int_{-\infty}^{+\infty}n_{HI}(R) \, \exp{(\frac{-z^2}{2
h^2})}\,\mbox{dz} = \sqrt{2\pi} \, h \, n_{HI}(R)
$$
\section{\ion{H}{1} Holes in IC~2574}
\label{hihol}
\subsection{The \ion{H}{1}--hole catalogue}
\label{hicat}
We conducted a search for \ion{H}{1} structures in IC~2574 using the
tasks {\sc kview, kpvslice} and {\sc kshell} that are built into the
recently developed {\sc karma}\footnote{The {\sc karma} visualization
software package was developed by Richard Gooch of the Australia
Telescope National Facility (ATNF)} visualization software
package. {\sc kview} offers many different ways to look at a movie of
a datacube whereas {\sc kpvslice} produces real--time position
velocity (pV) cuts through a cube in any orientation which
tremendously facilitates the search for and identification of
hole--like structures. Finally, {\sc kshell} was used to make
radius--velocity plots of some of the more prominent shells and holes.
First, the {\sc robust} cube was searched for the smaller holes in
IC~2574. In order to be sensitive to more extended structures and
therefore bigger holes, we smoothed our {\sc robust} cube to 11$''$,
15$''$ and 30$''$, respectively, and inspected each of them. For each
\ion{H}{1} hole in IC~2574 we determined its position, size and
expansion velocity. Both authors independetly produced a list of holes
to reduce personal bias, based on some simple criteria regarding the
determination of the center, size, expansion velocity and systemic
velocity of the holes. After that, the two lists were compared and
merged. There was an overlap of about 75$\%$ in terms of well defined
holes which lends some credibility to the identification process. We
reexamined the remaining structures and, erring on the side of
caution, ended up with a catalogue of 48 holes. Although this list
cannot be free from personal bias, the method closely
follows that employed in the case of M~31 and Ho~II which should make
a comparison between those objects and IC~2574 valid.
It is interesting, at this point, to reflect on the reality of the
features. Do they really constitute large--scale, coherent structures
or are we tricked by nature into seeing coherence in what is basically
a turbulent, perhaps fractal medium? Mac~Low \emph{et al.\ }
(\markcite{MAC98}1998) recently produced sets of model data cubes,
similar to those obtained via radio (\ion{H}{1}) or optical
(H$\alpha$) observations (i.e., position, position and velocity). We
analysed, using the same techniques as explained above, a cube of
uniform, isotropic, isothermal, supersonic, super--Alfv\'enic,
decaying turbulence. The simulations assume that the gas is optically
thin and emits with a strength directly proportional to its
density. The model corresponds to a weakly magnetized ISM. Single cuts
through the data cube show features remarkably similar to the ones
seen in the \ion{H}{1} data. However, when focussing on a potential
feature and investigating its morphology in position--position or
position--velocity space, no coherent structure emerges. This lends us
to believe that shells due to random (or at least turbulent) processes
make up a small fraction of the holes we classified. Instead, it is
more likely that the \ion{H}{1} holes and shells are due to mechanisms
which act on large scales, such as stellar winds and (multiple)
supernovae.
Following Brinks \& Bajaja \markcite{BRI86}(1986), we distinguished
between three different types of holes based on their appearance in
the pV--diagrams. A Type 1 hole corresponds to a total blowout (i.e.,
we don't observe the receding nor the approaching side of the
shell). In that case, it is impossible to determine the expansion
velocity and hence the kinematics of the hole. Figure~\ref{pvfinal}c shows a
pV--diagram of a typical Type~1 hole (our hole number 30) using {\sc
kpvslice}.
A Type~2 hole corresponds to a hole which is offset with respect to
the plane of the galaxy. In that case, we witness a deformation in
the pV--diagram towards the plane of the galaxy and a blowout to the
opposite side (see Silich \emph{et al.\ } \markcite{SIL96}1996 for numerical
simulations of such a case --- their case C). The amount of
deformation yields the expansion velocity of the shell towards the
midplane. The part of the shell that points away from the plane is
usually not visible due to lack of sensitivity. Figure~\ref{pvfinal}d
shows a pV--diagram of a typical Type~2 hole (our hole number 33).
Type~3 is the 'classic hole' and corresponds to a more or less
complete expanding shell. The signature in pV--space is that of an
elliptical hole. The expansion velocity is equal to half the
difference between the velocity of the approaching and receding
sides. The systemic velocity of the hole is defined by its
center. Figure~\ref{pvfinal}a shows a pV--diagram of a typical example of
a Type~3 hole (our hole number 21). Figure~\ref{pvfinal}b shows the same
hole, but analyzed with the task {\sc kshell}. In this figure, the
horizontal axis corresponds to radius $r$ (distance from the center of
the hole) and the intensity corresponds to the average \ion{H}{1}
density within an annulus with radius $r$ around the center of the
hole. The vertical axis is again velocity. Since we are looking at a
radius--velocity diagram, we will call such a representation an
rV--diagram in the following. For an ideal shell, a half ellipse
should be visible in which the diameter in the $v$--direction is twice
the expansion velocity and its dimension in the $r$--direction the
radius of the shell. The advantage of looking at a near circular
\ion{H}{1} shell with {\sc kshell} lies in a substantial increase in
signal--to--noise ratio as one averages over many pixels whereas in a
pV--diagram one merely cuts through an \ion{H}{1} structure.
Note that in the case of a Type~1 hole, the absence of detectable
emission corresponding to the front and back of a shell causes a hole
in rV--space. For obvious reasons, a Type~2 hole does not show a
prominent signature in an rV--diagram either.
\placefigure{pvfinal}
As was already mentioned earlier, the holes were found to be much more
prominent in the channel maps than in the integrated \ion{H}{1}
map. This is due to the high inclination of IC~2574 which leads
to several \ion{H}{1} shells at different locations within the disk
and possibly different velocities overlapping along the
line--of--sight, as is the case in M~31. The total \ion{H}{1}
surface brightness map was therefore of little use for the
identification of a hole. Note that in the case of Ho~II, {\em only}
the integrated map was used because of the much lower inclination of
the galaxy. This implies that the approach to search for holes needs
to be adapted to the orientation of the galaxy.
Figure~\ref{ic_holes} shows a plot of the distribution of the 48
\ion{H}{1} holes detected with confidence in IC~2574. The greyscale
map is a linear representation of the total \ion{H}{1} surface
brightness. Some of the holes are clearly visible in the total
\ion{H}{1} map but the majority of holes doesn't show up at all and
can only be detected in the channel maps or pV--diagrams.
\placefigure{ic_holes}
The final list of the detected \ion{H}{1} holes is given in
Table~\ref{holecat}. The properties listed are as follows:
\begin{description}
\item[Column 1:] Number of the hole (increasing with increasing right
ascension)
\item[Column 2, 3:] Position of the center of the hole in right
ascension and declination, respectively (B1950.0 coordinates). Typical
errors are 3$''$.
\item[Column 4:] Heliocentric velocity $v_{\rm hel}$ of the hole defined
by the channel in which the hole is largest or most prominent. The
accuracy of $v_{\rm hel}$ is about 2.5 $\,\mbox{km}\,\mbox{s}^{-1}$\ (the velocity resolution of
our observations).
\item[Column 5:] Diameter $d$ of the hole in kpc. Note that most of
the holes are circular. In the case of elliptical holes, the diameter
along the major axis, along the minor axis and the position angle
($2 \times b_{\rm maj}, 2 \times b_{\rm min}, PA$) is given
(but see Sec.~\ref{single}). The accuracy of
$d$ is around 95 pc (the FWHM of our synthesized beam).
\item[Column 6:] The expansion velocity $v_{\rm exp}$ of the hole in
$\,\mbox{km}\,\mbox{s}^{-1}$. Note that in the case of a Type 1 hole, no expansion velocity
can be measured. Typical errors are about 2 $\,\mbox{km}\,\mbox{s}^{-1}$.
\item[Column 7:] The estimated \ion{H}{1} volume density $n_{\rm HI}$ (in
M$_{\odot}\,$pc$^{-3}$) of the region where the hole is situated before
the hole was created. $n_{\rm HI}$ was calculated based on the
\ion{H}{1} surface density, as described in Sec.~\ref{scale}.
Given the uncertainties in the deprojection and the assumptions which
went into the derivation of the mass density within the disk, we
estimate a relative error in the value of $n_{\rm HI}$ to be of order
30$\%$.
\item[Column 8:] The type of the hole (1--3), as defined in
Sec.~\ref{hicat}.
\end{description}
Table~\ref{propcat} summarizes the physical properties derived from
Table~\ref{holecat}. The properties listed are as follows:
\begin{description}
\item[Column 1:] Number of the hole
\item[Column 2:] Diameter $d$ of the hole. For elliptical holes, the
diameter was defined using the geometric mean
$$
d=\sqrt{2 \, b_{\rm maj} \times 2 \, b_{\rm min}}
$$
\item[Column 3:] Expansion velocity $v_{\rm exp}$ of the hole in
$\,\mbox{km}\,\mbox{s}^{-1}$. Same as column 6 in Table~\ref{holecat}.
\item[Column 4:] Kinematic age of the hole in units of $10^6$
years. This age determination assumes that over its entire lifetime
the hole was expanding with $v_{\rm exp}$. Hence:
$$
t=\frac{(d/2)}{v_{\rm exp}}
$$
The accuracy of the kinetic age is
typically 20$\%$. Because of the assumption of a constant expansion
velocity, the kinematic age is an upper limit.
\item[Column 5:] Indicative \ion{H}{1} mass $M_{\rm HI}$ in units of
$10^4$ M$_{\odot}$. It is the mass which filled the observed hole
under the following assumptions: i) the hole is spherical, ii) the gas
was distributed uniformly in the galaxy, and iii) the hole is
completely empty now. Note that if one can assume that only a minor
fraction of the surrounding \ion{H}{1} is ionized, this mass corresponds
also to the total \ion{H}{1} mass on the rim of the hole. $M_{\rm HI}$ was
calculated using
$$
M_{\rm HI}=\frac{4}{3}\,\pi \, (d/2)^3 \, n_{\rm HI}(R)
$$ where
$n(R)$ is the volume density of the gas at the galactocentric distance
$R$ (column 7 in Table~\ref{holecat}). To correct this value for the
amount of primordial helium, a factor of 1.33 has to be applied. The
relative error is, as $d$ enters to the third power, of order
30$\%$. Note that in the case of the largest holes, this value is more
likely to be an upper limit since $n_{\rm HI}$ decreases with
increasing distance $z$ from the plane.
\item[Column 6:] The initial total energy $E_{\rm E}$ that was
released by the supernova explosions which created the hole (in units
of $10^{50}$ ergs). Note that the kinetic energy of the expanding
shell is about 1$\%$ to $10\%$ of $E_{\rm E}$ (depending on the
ambient density, see Chevalier \markcite{CHE74}1974). To determine
$E_{\rm E}$, we used the equation by Chevalier \markcite{CHE74}(1974)
who derived a fit to numerical results of an expanding shell created
by a single explosion (see also McCray \& Snow \markcite{MAS79}1979).
Expressing $E_{\rm E}$ in terms of observable quantities $d$, $v_{\rm
exp}$ and $n_0$, he obtained
$$
E_{\rm E}=5.3\times 10^{43}\, n_0^{1.12}\, (d/2)^{3.12}\, v_{\rm exp}^{1.4}\,
\mbox{ergs}
$$
where $v_{\rm exp}$ is the expansion velocity in $\,\mbox{km}\,\mbox{s}^{-1}$, $n_0$ the
density of the ambient medium in particles per cubic centimeter, and
$d$ is the diameter of the shell in parsecs. In our calculations we
replaced $n_0$ by $n_{\rm HI}$ as was done, e.g., for M~31 and
Ho~II. This is not strictly correct as we are ignoring contributions
by He and heavier elements. To correct $E_{\rm E}$ for this
contribution, all values should be multiplied by a factor of $\approx
1.5$. In addition, the derived \ion{H}{1} densities are average values
for a given galactocentric distance which introduces another source of
uncertainty. Note that the numerical coefficient in Chevaliers
equation depends on the radiative energy loss rate and thus on the
heavy element abundance in the expanding shell. Not only is this
abundance different for various galaxies, but it decreases with
galactic radius, both for our own and external galaxies and by more
than an order of magnitude (e.g., Heiles \markcite{HEI79}1979). The
last phase in the expansion of a shell is reached when, after the
shell has continued to collect more of the ambient medium it will slow
down to velocities comparable to the random motions of the ISM (around
7 $\,\mbox{km}\,\mbox{s}^{-1}$) and hence stops expanding. Most probably the largest
structures we observe are in this latter phase (see
Sec.~\ref{large}). From this brief discussion it should be clear that
the derived energies $E_{\rm E}$ are only order of magnitude
estimates.
\placetable{holecat}
\placetable{propcat}
\end{description}
\subsection{Notes on single holes}
\label{single}
The following holes deserve a more detailed description:
\begin{description}
\item[Hole $\#$35:] This hole causes a major disturbance in the
pV--diagrams of IC~2574. It is one of the few elliptical holes, with
dimensions of $1000 \times 500$ pc. Although there is some
circumstantial evidence that it is a superposition of two spherical
holes, it was treated as one hole. As we will show in
Sec.~\ref{ha-hi}, the most prominent \ion{H}{2} region in IC~2574 is
located on the rim of this supergiant shell. The shell is filled with
soft X--ray emission as observed with the ROSAT telescope. A detailed
multiwavelength analysis (optical, radio, X--ray) of this particular
supergiant shell has been presented elsewhere (Walter \emph{et al.\ }
\markcite{WAL98}1998)
\item[Hole $\#$22:] Catalogued elliptical in shape, this hole is
actually a superposition of several holes, leading to the overall
impression of an elliptical feature; we treated this as a single
object since subdividing it in, say, three holes would have been
arbitrary.
\item[Holes $\#$ 13, 23, 26:] As can be seen from the channel maps
(Fig.~\ref{ic_cm_1}), some holes are not clear--cut cases in terms of
the definition of a hole given in Sec.~\ref{hicat}. They are possibly
chance superposition of old shells and fragments leading to the
impression of a hole in the integrated \ion{H}{1} map (see also
Sec.~\ref{large}). We therefore decided not to include these holes in
the statistical analysis.
\end{description}
\subsection{Notes on large--scale structures}
\label{large}
As can be seen from Table~\ref{holecat}, only few shells larger then
$\approx 1000$~pc have been catalogued in IC~2574. It has to be
mentioned, however, that there appear coherent features larger then
1~kpc in the channel maps. Several of these can be identified in the
channel maps (Fig.~\ref{ic_cm_1}), e.g., between 45 and 70 $\,\mbox{km}\,\mbox{s}^{-1}$. Note
that these structures are related to the major disturbance in the
velocity field due north of the center of the galaxy. Obviously, it is
very difficult to catalogue these features as they lack the clear
coherence, both in position as well as in velocity, of \ion{H}{1}
shells. Moreover, because of their large size, their dynamics is not
dominated by some leftover expansion but rather by large scale
galactic rotation modified by the effects of nearby, smaller
holes. The reason why we wish to mention these large scale features is
because along some of the apparent arcs we find \ion{H}{2} regions.
\section{The \ion{H}{1}--H$\alpha$ connection in IC~2574}
\label{ha-hi}
\subsection{Global correlations}
Until now, we have focussed on the structure and dynamics of the
neutral ISM of IC~2574. In the following, we will compare these
results with our H$\alpha$ observations which reveal the sites of
recent or ongoing SF in IC~2574. As noted earlier (Sec.~\ref{Intro}),
the expanding \ion{H}{1} structures are widely thought to result from
recent star formation. They sweep up gas causing the volume density on
the rim of the shell to increase until gravitational instabilities
lead to molecular cloud formation and hence secondary SF (e.g.,
Palou\v s \emph{et al.\ } \markcite{PAL90}1990, Elmegreen
\markcite{ELM94}1994). Therefore, a comparison of the position of
\ion{H}{1} holes with \ion{H}{2} regions should show some correlation,
the H$\alpha$ emission being predominantly found along the rims of
larger holes, as evidence for propagating SF. In the case of the
smaller, young holes one would expect some to be still filled with
diffuse H$\alpha$ emission coming from the ionized wind--blown
cavities around the still present, young O and B stars.
Figure~\ref{haho} shows an overlay of the H$\alpha$ map with the
\ion{H}{1} holes we detected in our survey (Sec.~\ref{hicat}). The
width of the rims of the \ion{H}{1} holes is of course in reality
larger then indicated in this plot. The subregions A--C will be
discussed in more detail in the following section. As can be seen from
this overlay, there is a prominent correlation of ongoing SF with the
rims of the large \ion{H}{1} shells which is in general agreement with
the picture painted above. There is a strong tendency for \ion{H}{2}
regions to be situated on the rims of the \ion{H}{1} shells (for some
very nice examples see holes 8 and 35) and only in a very few cases
diffuse H$\alpha$ emission does not seem to be related to an
\ion{H}{1} structure. Note that in one case only (hole 32), diffuse
H$\alpha$ emission (partly) fills an \ion{H}{1} hole.
Fig.~\ref{hahic} presents an overlay of the H$\alpha$ map with one
contour of the \ion{H}{1} surface brightness map (flux and primary
beam corrected) plotted at $1.7\times 10^{21}$ atoms cm$^{-2}$. As can
be seen from this overlay, H$\alpha$ emission is globally correlated
with a high \ion{H}{1} surface brightness (for an exception in the
very north see below). Kennicutt \markcite{KEN89}(1989) used the disk
instability relation of Toomre \markcite{TOO64}(1964) to show that,
for spirals, there is a threshold density for star formation. For gas
densities above the threshold, the star formation rate (SFR) is
consistent with a Schmidt law. Near the threshold, the correlation
breaks down, and one sees bursts of SF. Much below the threshold the
SFR is extremely low. Skillman \markcite{SKI87}(1987) has
observationally determined this threshold for dwarf galaxies to be on
the order of $1\times 10^{21}$ atoms cm$^{-2}$ (at a resolution of
500~pc) which is in general agreement with our results (see also
below). For a detailed discussion on the models that describe the
relationship between gas, stars and star formation in irregular
galaxies, the reader is referred to the paper by Hunter, Elmegreen \&
Baker \markcite{HUN98}(1998).
\placefigure{haho}
\placefigure{hahic}
\subsection{Analysis of individual \ion{H}{2} regions}
\label{sha}
Using our high resolution \ion{H}{1} data of IC~2574 we are now able
to perform a much more detailed analysis and comparison of the
properties of the \ion{H}{2} regions with the \ion{H}{1} features that
are related to them. When making overlays of the H$\alpha$ map with
the high resolution cube we found that in virtually all cases, single
\ion{H}{2} regions are correlated with distinct \ion{H}{1} structures
like clumps or clouds. We defined 40 regions where H$\alpha$ emission
and \ion{H}{1} clouds seem to be related. These regions are outlined
by the squares in Figs.~\ref{har1}--\ref{har3} which are meant to
indicate which \ion{H}{2} regions were lumped together. They are
numbered 1$'$ to 40$'$ to avoid confusion with the numbers of the
\ion{H}{1} holes. Note, that sometimes two or more H$\alpha$ features
belong to the same \ion{H}{1} structure (like a clump) and were thus
considered as one region (e.g., in region 40$'$), whereas in other
cases individual \ion{H}{2} regions were treated as separate entities
(e.g., regions 11$'$ and 12$'$). The H$\alpha$ fluxes of the regions
tell us something about the current SF rate and the average number
density of the ionized gas whereas the \ion{H}{1} data yield
approximations for the number density of neutral hydrogen, the
velocity dispersion of the gas in that region and the thresholds for
SF. Figs.~\ref{har1}--\ref{har3} are enlargements of the regions A--C
marked in Fig.~\ref{haho} and display the regions defined by us. The
position and size of the \ion{H}{1} holes are plotted as well for
comparison (please refer to Fig.~\ref{haho} for cross referencing of
the holes). The fluxes of the \ion{H}{2} regions were taken from the
observations of \markcite{MIL94}MH94. Table~\ref{cross} summarizes
the cross identifications between their and our regions. In this table
we list the coordinates of the centers of our regions as well.
Table~\ref{Ha-regions} summarizes the results of our analysis of the
\ion{H}{2} regions. The first column refers to the number of the
\ion{H}{2} region as defined in Figs.~\ref{har1}--\ref{har3}. From the
fluxes published by \markcite{MIL94}MH94, we deduced the total flux
$F(H\alpha$) of our groups by simply summing their fluxes in the
respective regions (column~2, Table~\ref{Ha-regions}). From this
number, the H$\alpha$ luminosity can be calculated using
$L(H\alpha)=4\pi\,D^2 F(H\alpha),$ which, in the case of IC~2574
($D$=3.2~Mpc), translates to $L(H\alpha)=3.1\times 10^{17} F({\rm
H\alpha})$ where $F(H\alpha)$ is given in ergs~cm$^{-2}$~s$^{-1}$ and
$L(H\alpha)$ in units of solar luminosities $L_{\odot}$ ($1
L_{\odot}=3.85\times 10^{33}$ ergs~s$^{-1}$). The number of Lyman
continuum photons per second is given by $N_{\rm Lyc}=2.82\times
10^{45} L({\rm H\alpha}).$ Table~\ref{sprop} (see Devereux \emph{et al.\ }
\markcite{DEV97}1997 and references therein) summarizes the expected
properties of high mass stars; column~1 gives the spectral type,
column~2 the stellar mass, column~3 the bolometric stellar luminosity,
column~4 the number of Lyman continuum photons per second, column~5
the H$\alpha$ luminosity L$(H\alpha)$, and column~6 the ratio of
bolometric to H$\alpha$ luminosity. These equations and values can be
used to estimate, for example, the equivalent number of O5 stars, for
each region (see below).
Column 3 in Table~\ref{Ha-regions} presents the volume density of the
ionized free electrons over the area where we find H$\alpha$ emission
(as defined by MH94). This was arrived at by first calculating
the emission measure (EM) of each of their regions according to:
$$
EM=2.41\times 10^3 \, T^{0.92}\, F(H\alpha)\, {\rm cm}^{-6}\, {\rm pc}
$$
(Peimbert \emph{et al.\ } \markcite{PEI75}1975), where $F(H\alpha)$ is
the H$\alpha$ flux in units of ergs~cm$^{-2}$~s$^{-1}$~sterad$^{-1}$
and $T$ is assumed to be the canonical value of $10^4$~K.
From that, the mean electron density for every
single region published by MH94 was derived using
$$
EM=n_{\rm e}^2\, d\, {\rm cm}^{-6}\, {\rm pc}
$$
where $n_{\rm e}$ is the electron density and $d$ the path length in
pc through the \ion{H}{2} region (assuming a spherical distribution of
the H$\alpha$ emission). Since the derived values for $n_{\rm e}$ were
very similar for a single region, we calculated the value for the
electron density for each of our regions by taking an
intensity--weighted average of the contributing MH94 regions. The
average electron densities range from 0.5 to 2 cm$^{-3}$. This is,
most likely, a lower limit for the actual electron density due to
internal absorption within IC~2574. Note that in the above, effects
due to the filling factor of the gas are ignored. The real densities
in the \ion{H}{2} regions are likely much higher.
In order to derive properties of the \ion{H}{1} features that are
physically related to the \ion{H}{2} regions, we defined areas $A$ around
each of the \ion{H}{2} regions with an effective radius given by $r_{\rm
eff}=\sqrt{A/\pi}$, where $r_{\rm eff}$ is given in pc (column
4). Although this is a somewhat subjective process, we found that slight
variations in the size of the aperture didn't change the derived
properties by more than $20-30\%$. Column 5 represents the total
\ion{H}{1} mass found within the aperture using the flux and primary beam
corrected \ion{H}{1} data. For every single region, we measured the
velocity dispersion of the \ion{H}{1} by taking spectra through the
\ion{H}{1} cube. Column 6 summarizes the measured velocity widths
defined as the full width at half maximum (FWHM) from which the real
velocity dispersion $\sigma_v$ (assuming a Gaussian velocity
distribution) can be calculated using
$
\Delta v=2\sqrt{2\ln 2}\,\sigma_{\rm v}=2.355\, \sigma_{\rm v}
$
Interestingly, these values tend to be high where two or more shells
overlap (e.g., the \ion{H}{2} complexes 16, 31, 32, 33). These high
dispersions are probably caused by blending along the line of sight of
these shells.
Column 7 presents the average density of the neutral hydrogen which
was calculated using
$$
\rho_{\rm HI}=M_{\rm HI}/(\frac{4}{3}\, \pi\, r_{\rm eff}^3)
$$
with $M_{\rm HI}$ and r$_{\rm eff}$ the values from columns 5 and 4,
respectively. Values for $\rho_{\rm HI}$ range from about 1 cm$^{-3}$
to values as high as 6 cm$^{-3}$. Note that these are only averages
and that locally the \ion{H}{1} density might be more
enhanced. Based on the analysis by MCR94 and some crude mass modeling
(Sec.~\ref{scale}) we derive an average density in the disk of
0.008 $\rm M_\odot\,pc^{-3}=0.3\, atom\,cm^{-3}$. The regions in which
star formation is currently active thus have an overdensity
of an order of magnitude (see Table~\ref{Ha-regions}).
The average surface density of the \ion{H}{1} regions is shown in
column 8. Note that virtually all values are larger than $1\times
10^{21}$~cm$^{-2}$ which is in excellent agreement with the
observational value given by Skillman \markcite{SKI87}(1987). In
other words, SF is taking place if the projected \ion{H}{1} surface
density reaches values of about $10^{21}$ cm$^{-2}$ not only for a
resolution of 500 pc, as observed by Skillman but also for almost an
order of magnitude improvement ($\approx 95\, \rm pc$)! The highest
surface density found is of order $2.5\times 10^{21}$~cm$^{-2}$.
Finally Column 9 summarizes some remarks regarding the regions. Most
of the \ion{H}{2} regions are related to the rim of an \ion{H}{1}
shell. It should be noted at this point that in the course of checking
for correlations between H$\alpha$ and \ion{H}{1} emission, we
detected 5 new tiny holes at the positions were there is H$\alpha$
emission. However, as these holes were not detected in our
inspection of the \ion{H}{1} cubes we didn't include them in the final
catalogue which is thought to be complete for holes with radii larger
then about 100 pc only (as discussed in Secs.~\ref{hicat}
and~\ref{stat}).
We catalogued four of the \ion{H}{1} features as ``clumps'' (regions
1$'$, 2$'$, 6$'$ \& 22$'$). By 'clump' we mean a well--defined
\ion{H}{1} region in 2 spatial dimensions as well as in velocity
space. These clumps look like small, filled circles in
position--velocity space. For those cases we asked ourselves whether
these clumps are gravitationally confined or not. To answer that
question, we calculated the virial masses of these objects using
$$
M_{\rm vir}=250\, \Delta v^2 \, r_{\rm eff} \, {\rm M}_\odot
$$
(Rohlfs \& Wilson \markcite{ROH96}1996), where $\Delta v$ is the full
velocity width
at half maximum (FWHM) in km~s$^{-1}$ and $r_{\rm eff}$ the radius of
the clump in pc. The virial masses of the four regions are: $M_{\rm
vir}(1$'$) = 13\times 10^6 \,{\rm M}_\odot$, $M_{\rm vir}(2$'$) =
38\times 10^6 \, {\rm M}_\odot$, $M_{\rm vir}(6$'$) = 21\times 10^6 \,
{\rm M}_\odot$ and $M_{\rm vir}(22$'$) = 7\times 10^6 \, {\rm
M}_\odot$. As we see, the virial masses are all about one order of
magnitude higher than the \ion{H}{1} masses. So, the clumps are either
not gravitationally bound or they are held together by as yet
undetected material. One possibility is that they contain an order of
magnitude more mass in the form of molecules. This will be hard to
confirm, however, as dwarf galaxies have been notoriously difficult to
detect in CO, the most abundant tracer for molecular gas (e.g., Sage \emph{et al.\ }
\markcite{SAG92}1992). Dark Matter could play a role, but the mass
density, including DM, near the above mentioned SF regions is of order
0.008~$M_\odot\,{\rm pc^{-3}}$ (see Sec.~\ref{scale}), insufficient by
a wide margin.
If the observed velocity dispersion is not due to gravitation, the
question remains, why {\em do} we see clumps? We can think of two more
mechanisms, magnetic confinement or pressure. The former is unlikely
to play a large role. Dwarf galaxies do not possess large scale
magnetic fields (the exceptions being NGC~4449 and IC~10, e.g., Chyzy \emph{et al.\ }
\markcite{CHY98}1998), any non--thermal radio continuum radiation being confined to the
strongest regions of active star formation (Deeg \emph{et al.\ }
\markcite{DEE93}1993).
We did not find any correlation of H$\alpha$ flux with the surrounding
\ion{H}{1} density. However, one wouldn't really expect any strong
correlation of this kind. Not only is the H$\alpha$ emissivity of a
star forming region strongly dependent on the age of the region (see,
e.g., Gerritsen \& Icke 1997), but it is also dangerous to try to
deduce actual physical properties of the SF regions from mean values,
like average electron or \ion{H}{1} densities, while ignoring
spatial variations in their volume filling factors.
In some regions within IC~2574 (e.g., region 40'), we find high
\ion{H}{1} surface densities but only little H$\alpha$ emission
indicating that possibly a major star formation burst is about to
commence there and that therefore most of the gas is still
neutral. With the help of the equations given above we can calculate
how many stars are approximately needed to create the observed
H$\alpha$ flux. In the case of region 40$'$ we find a flux of
$1.5\times 10^{-14}\, \rm ergs\, cm^{-2}\, s^{-1}$ and deduce a
luminosity of $4.7\times 10^3\, {\rm L}_\odot$ which is the equivalent
of a few O7 stars only (see Table~\ref{sprop}). Note, however, that
there exist also regions where the situation is reversed and most of
the \ion{H}{1} is being ionized by the Lyman continuum photons of the
most massive stars of the powerful \ion{H}{2} regions. Region 34' is a
prominent example (see also Fig.~\ref{hahic}). From the H$\alpha$ flux
of $F(H\alpha)\approx 1\times 10^{-12}\, \rm ergs\, cm^{-2}\, s^{-1}$
we derive an H$\alpha$ luminosity of $L(H\alpha)=3.1\times10^5\, {\rm
L}_\odot$ which corresponds to an equivalent of 22 O5 stars, in other
words a major SF region.
\placetable{cross}
\placetable{Ha-regions}
\placetable{sprop}
\placefigure{har1}
\placefigure{har2}
\placefigure{har3}
\section{Discussion}
\label{discu}
\subsection{What created the \ion{H}{1} holes?}
\label{what}
The ISM of IC~2574 is dominated by expanding \ion{H}{1} holes and
shells. Similar structures have been found in other gas--rich dwarf
galaxies (e.g., Holmberg~II (PWBR92), IC~10 (Wilcots \& Miller
\markcite{WIL98}1998) and in more massive spiral galaxies like our
Galaxy (Heiles \markcite{HEI79}1979), M~31 (Brinks \& Bajaja
\markcite{BRI88}1988) and M~33 (Deul \& den Hartog
\markcite{DEU90}1990). One of the major questions concerning these
structures is what physical process or processes are responsible for
their origin.
Heiles (\markcite{HEI79}1979), who was the first to recognize
expanding \ion{H}{1} shells in the Galaxy, concluded that 'many of the
\ion{H}{1} structures [...] were probably produced by large amounts
of energy injected directly to the interstellar medium'. One obvious
mechanism is the after effects of massive star formation: strong
stellar winds and supernova explosions (see the review by
Tenorio--Tagle and Bodenheimer 1988). Detailed numerical modeling by
various authors (e.g., Palou\v{s}, Franco \& Tenorio--Tagle
\markcite{PAL90}1990, Silich \emph{et al.\ } \markcite{SIL96}1996, Palou\v{s},
Tenorio--Tagle \& Franco \markcite{PAL94}1994) has strengthened
confidence in this explanation, as has the work by Oey \& Clarke
\markcite{OEY97}(1997) who in a (semi--)analytical way managed to
predict the \ion{H}{1} shell size distribution in the SMC based on the
observed H$\alpha$ luminosity function.
Having said that, there are potentially problems with this
interpretation, as mentioned by Radice, Salzer \& Westpfahl
(\markcite{RAD95}1995) and Rhode, Salzer \& Westpfahl
(\markcite{RHO99}1999). Looking for remnant stellar clusters at the
centers of several large holes in Holmberg~II they failed to find a
population of A and F main sequence stars. Moreover, it is difficult
to imagine that the largest \ion{H}{1} shells which are located beyond
the optical galaxy should have been created by star formation (an
argument which holds equally well for some of the holes in
IC~2574). This is corroborated by ultraviolet imaging of Holmberg~II
by Stewart (\markcite{STE99}1999) who finds no FUV emission from the
outer regions of Holmberg~II.
It should be noted, though, that the nondetection of population A and
F stars does not necessarily mean that none of the \ion{H}{1} are
created by SF. In dwarf galaxies, in the absence of shear, \ion{H}{1}
shells can reach large dimensions and grow so old that no obvious
bright cluster remnant will remain visible in the center (Efremov,
Elmegreen \& Hodge \markcite{EFR98}). Some of the observed supershells
are so old that the brightest members of stellar cluster which created
the hole may have evolved off the main sequence and dispersed (assuming
a similar IMF as in the Galaxy). This may make a detection of old
clusters difficult. Such a case (LMC~4/Constellation~III) is described
by Efremov \& Elmegreen (\markcite{EFE98}1998).
Another potential problem with the conventional picture for the origin
of the holes is that for the largest ones the energy requirements seem
to surpass the amount that can possibly be released into the ambient
ISM via stellar winds and supernova explosions. The giant shell in
M~101 (van der Hulst \& Sancisi \markcite{HUL88}1988) is just one
example of such a structure. Tenorio--Tagle (\markcite{TEN87}1987)
proposed as an alternative the infall of high velocity clouds which
could deposit an amount of energy of 10$^{52}$ to 10$^{54}$~ergs per
collision (see the review by Tenorio--Tagle \& Bodenheimer
\markcite{TEN88}1988 for further references). Even though this
mechanism seems a viable explanation for at least some holes, it is
unlikely that it plays a dominant role. Certainly it can't be as
important in dwarf galaxies because of their much smaller collisional
cross section.
Recently, an alternative explanation for the creation of supergiant
\ion{H}{1} shells in galaxies has been proposed, namely GRBs or Gamma
Ray Bursters (Efremov, Elmegreen \& Hodge \markcite{EFR98}1998, Loeb
\& Perna \markcite{LOE98}1998). Even though the basic physics of GRB
are not yet known the picture to emerge is that they are highly
energetic events (of order $10^{54}$~ergs) that release huge amounts
of energy in a very short period of time (order of a few seconds).
Although we recognize that perhaps not all holes can be explained by
stellar winds and supernova explosions, we feel that there is
sufficient evidence for the conventional interpretation to warrant an
analysis in those terms, which is what we pretend in the rest of this
paper.
\subsection{Statistical properties of the holes: a comparison
with other galaxies}
\label{stat}
For the first time we are now in a position to address an important
question regarding the creation and evolution of hole--like structures
in the interstellar medium: how are their properties related to their
environment, i.e., the type of host galaxy. In what follows we will
compare the observed and derived \ion{H}{1} hole properties of IC~2574
with those found in M~31 (Brinks \& Bajaja \markcite{BRI86}1986), an
example of a massive spiral galaxy similar to our own, M~33 (Deul \&
den Hartog \markcite{DEU90}1990), a less massive spiral and Ho~II
(PWBR92\markcite{PUC92}), another dwarf galaxy in the same group of
galaxies, yet four times less massive than IC~2574. In other words,
the sequence M~31 -- M~33 -- IC~2574 -- Ho~II spans a large range of
different Hubble types from massive spirals to low--mass dwarfs.
Despite the fact that a similar study, and at considerably higher
sensitivity, is now available for the SMC (Staveley--Smith \emph{et al.\ }
\markcite{STA97}1997), we decided not to include their results. The
main reason being that the linear scales (or spatial frequencies)
sampled by the ATNF hardly overlap with those observed in the galaxies
listed above. The linear resolution, at 28 pc, is almost four times
higher than, for example, the VLA maps on IC~2574. At the other end of
the spectrum, because of the lack of short spacing information,
structures larger than a few hundred parsec will have been
missed. Moreover, the SMC is a very disturbed system, being torn apart
by tidal forces due to interactions with the LMC and the Galaxy.
Limiting the comparison to the four galaxies listed above has some
further advantages. The linear resolutions are very similar, as are
the velocity resolutions with which they have been observed (see
Tab.~\ref{diff} for a summary). In addition, all four galaxies were
examined in more or less the same fashion, one of the authors (EB)
having taken part in the analysis of three of the four objects. We
are aware that the results for the four galaxies suffer partially from
low statistics and incompleteness due to personal bias and
observational constraints (such as the beamsize). However, we feel
that these effects, to first order, affect a comparison in a similar
way and that it is valid to try to find global trends as a function
of Hubble type.
For the sake of brevity, rather than first show the statistical
properties of the holes in IC~2574 separately, we immediately go to a
comparison with the other three objects. In
Figs.~\ref{massc}--\ref{velc} we compare the distributions of
swept--up matter, the energies, diameters, expansion velocities and
ages of the holes in the 4 galaxies. In each graph we plot in the form
of histograms the relative number of holes, in percent, in order to
make a direct comparison possible. In each histogram, the data for the
individual objects are binned in the same way. To improve the
presentation, however, the respective bins were slightly shifted when
plotting the results.
Fig.~\ref{massc} shows an overlay of the indicative masses of the
\ion{H}{1} holes, binned logarithmically. The values range from
typically $10^4$ to $5 \times 10^6$ M$_\odot$. Even though a
two--sample Kolmogorov--Smirnov test rejects the hypothesis of
pairwise equality of the distributions, a casual inspection shows that
the indicative masses span roughly the same range. The mean masses
(in 10$^5$M$_{\odot}$) for the dwarf galaxies (Ho~II: 10.5, IC~2574:
8.4) are slightly higher as compared to the spirals (M~33: 4.1, M~31:
3.2). This is somewhat surprising if one realizes that the volume
density in the plane of a spiral galaxy is typically higher by a
factor of three as compared to that found in a dwarf
galaxy. \ion{H}{1} column densities have a much smaller variation from
object to object, though (see Tab.~\ref{diff}) which probably
explains why the swept up matter works out to be more or less the
same. As a warning, it should be born in mind that, in order to
determine the \ion{H}{1} densities we smeared out any fine scale
structure, potentially adding somewhat to the errors.
Fig.~\ref{energyc} is similar to Fig.~\ref{massc}, showing instead the
distribution of the logarithm of the energies needed to create the
observed \ion{H}{1} holes (under the assumption of them being created
by supernova explosions). The derived energies range from $10^{50}$ to
$10^{53}\, \rm ergs$. We find the important result that the
distributions are identical, at least to first order (mean energies
(10$^{52}$\,ergs): Ho~II: 1.4, IC~2574: 1.1, M~33: 1.6, M~31: 1.5). A
two--sample Kolmogorov--Smirnov test reveals that there is no
significant pairwise difference at the 95\% level between the four
energy distributions. From this we conclude that regardless of Hubble
type of the parent galaxy, the star clusters (or whatever other
process created the \ion{H}{1} holes) deposit more or less the same
amount of energy into the ISM.
Fig.~\ref{diamc} shows an overlay of the size distribution of the
holes found in the four galaxies. In this and the following plots the
bins are on a linear scale. Note that there is a clear difference with
Hubble type! The size distribution for holes in M~31 and M~33 cuts off
sharply near 600 pc. In contrast, holes in IC~2574 and Ho~II reach
sizes of 1200 to 1500 pc, respectively. The lack of holes with sizes
smaller than $\sim 100$~pc is due to our resolution limit. The fact
that the holes get larger for ``later'' Hubble types (i.e., galaxies
with lower mass) can be understood as follows. Smaller galaxies have
lower masses and their mass surface density in the disk is lower as
well. The one--dimensional velocity dispersion varies little, if any,
among galaxies. We find in the four objects under scrutiny, taking
into account an uncertainty of $\pm$ 1$\,\mbox{km}\,\mbox{s}^{-1}$, a range of 6--9
$\,\mbox{km}\,\mbox{s}^{-1}$. This, combined with the lower mass surface density, leads to
dwarf galaxies having much thicker disks with aspect ratios of 1:10
rather than 1:100. Hence, for the same amount of energy deposited, as
argued above, an \ion{H}{1} shell can grow much larger, because of a
lower gravitational potential and a lower ambient density. Moreover,
in dwarf galaxies shells are prevented from breaking out of the disk
because it is thicker (see Sec.~\ref{scale}).
The age distribution of the holes (Fig.~\ref{agec}) also shows a
sequence with Hubble type. The mean values for the ages of the holes
are: $10 \times 10^6$ years (M~31); $12\times 10^6$ years (M~33); $23
\times 10^6$ years (IC~2574) and $62\times 10^6$ years (Ho~II). In
the spirals no holes are found which are older than 30 Myr whereas the
distribution in dwarfs is flatter and spans an age range of up to 120
Myr. This is not unexpected since in more massive galaxies shear and
mixing due to the passage of density waves tend to destroy holes. In
dwarfs, unless a shell is hit by a neighboring explosion, and
subsequently expanding structure, a feature can potentially persist
for a long time, until the rim disperses and mixes with the
surrounding ISM.
Finally, the distribution of the expansion velocities is presented in
Fig.~\ref{velc}. The predominantly low values for the dwarfs
(especially Ho~II) reflect, again, the fact that we see many large
holes in the dwarfs which have almost stalled. A word of caution
should be added as the velocity resolution of the IC~2574 and Ho~II
observations was a factor of about 3 better than those of M~31. Very
young (i.e., smaller than $\sim 100$ pc) and fast expanding holes are
missed, of course, but this holds true for all galaxies in the sample.
Figure~\ref{v_d_comp} displays an alternative way of looking at the
properties of the \ion{H}{1} holes on a galaxy--wide scale. It shows
the expansion velocities of the shells in the four nearby galaxies
plotted as a function of their diameter (crosses). Based on the
hydrodynamical models for expanding shells (Chevalier
\markcite{CHE74}1974), we plotted lines of constant energy for the
expanding structures in Fig.~\ref{v_d_comp} (assuming a constant
density of the ambient medium of 0.15 cm$^{-2}$). The plotted curves
represent energies at a level of $10^{49}$, $10^{50}$, $10^{51}$ and
$10^{52}$ ergs, respectively (from left to right). The evolution of
an individual hole of a given energy is along an equi--energy line
from the top of the plot downwards. In addition, lines of equal ages
are shown in the figure, ranging from $10^7$ years (steepest slope)
via $5 \times 10^7$ years to $10^8$ years.
There is a marked difference between the two spiral galaxies and the
dwarfs. The \ion{H}{1} shells in M~31 and M~33 are smaller and the
range of expansion velocities is larger. In fact, when looking at the
objects in order of decreasing mass (M~31 -- M~33 -- IC~2574 -- Ho~II)
there seems to be a gradually shifting pattern. Whereas in the larger
spirals there is an absence of older holes (older than a few $\times
10^7$ years, holes in IC~2574 range between $5 \times 10^7$ and $10^8$
years and holes in Ho~II cluster near $10^8$ years
This suggests an alternative explanation for why we don't see younger
holes in IC~2574 and Ho~II. The SF histories of these dwarf galaxies
could be quite different from the spirals. In the latter, the global
SF rate is, to first order, constant with time. Therefore, there is
always a population of young holes. In contrast, in gas--rich dwarf
irregulars, SF occurs in bursts which transforms them into an
\ion{H}{2} or blue compact dwarf (BCD) galaxy. It might be that
IC~2574 and Ho~II have undergone a recent burst, perhaps triggered by
a close passage with another member of the M~81 group, and are now in
their post--starburst phase and that this is why we don't see many
young holes. Our data alone don't allow us to choose between the two
possibilities.
\placetable{diff}
\placefigure{massc}
\placefigure{energyc}
\placefigure{diamc}
\placefigure{agec}
\placefigure{velc}
\placefigure{v_d_comp}
Oey \& Clarke \markcite{OEY97}(1997), using the standard, adiabatic
shell evolution predict the size distribution for the populations of
OB superbubbles. In their paper they compare their predictions with
observations of the SMC and the three same galaxies as we compared our
IC~2574 data with, namely M31, M33, and Ho~II. Without going into
detail, their prediction fits the observations for the SMC
surprisingly well. This despite the fact that they ignore the
selection effect (missing short spacings) which renders the SMC maps
insensitive to large structures which should dominate the appearance
of dwarf galaxies. Also, they emphasize stalled shells, i.e., objects
which have reached their largest diameter. This is in contradiction
with the fact that the SMC shells, especially the smaller ones, are
all found to be rapidly expanding. Their approach works less well for
the other objects. They ascribe this to the data being incomplete, not
only when considering the smallest detectable shells, but overall, due
to the reduced sensitivity as compared to the SMC data. In that
respect, our IC~2574 are not better than the maps on M31, M33 or
Ho~II. Hence, a thorough test of Oey \& Clarke's predictions will have
to await higher resolution and higher sensitivity data on the nearest
(dwarf) galaxies.
\subsection{The intrinsic shape of the shells}
\label{shape}
One of the results from our analysis is that over 90\% of the holes are
well approximated by a circular shape (the few exceptions are
discussed in Sec.~\ref{single}). This is a major difference with
other galaxies studied thus far in similar detail (i.e., Ho~II, M~31,
M~33, M~101, NGC~6946) which tend to have much more elliptical holes.
One attempt to understand this is in terms of projection effects
(different inclinations for the galaxies observed thus far). If the
shells are intrinsically triaxial, their projected shape on the sky
will invariably be an ellipse with an ellipticity which varies as a
function of the intrinsic axial ratios, the inclination of the galaxy,
{\em and} the position angle of the major axis of the hole within the
plane of the galaxy. This should lead to a wide range of
ellipticities. The fact that we don't see this in IC~2574 can only
mean that the shells in this galaxy are {\em intrinsically}
spherical.
This can be understood by considering the shapes of the
rotation curves of the galaxies. IC~2574 is the only object which
exhibits a rising rotation curve throughout, as is typical for dwarf
galaxies. As the rotation curve does not turn over, leading to a flat
part dominated by differential rotation, there is no shear to distort
and ultimately destroy any large scale structures such as \ion{H}{1}
shells. For a more complete discussion on this topic, see Silich \emph{et al.\ }
\markcite{SIL96}(1996) and Palou\v{s} \emph{et al.\ } \markcite{PAL90}(1990). Indeed,
PWBR92\markcite{PUC92} estimate that the amount of shear over 1 or 2 kpc
in radius, once the flat part of the rotation curve is reached in
Ho~II, is sufficient to stretch the holes to an axial ratio of 0.5 over
a timescale of $10^8$ years.
IC~2574, however, is the only object studied to date which
shows solid body rotation almost throughout the disk
(MCR94\markcite{MAR94}) which ensures that holes stay circular until
long after their creation.
There is one fly in the ointment. Why should the shells in IC~2574 be
intrinsically spherical? The expansion rate perpendicular to the disk
should be different from that in the disk. In the direction
perpendicular to the disk, the volume density rapidly decreases,
favoring faster expansion in that direction. But on the other hand,
the gas which moves upwards starts to feel the gravitational pull from
the mass surface density below it, slowing it down. It thus seems as
if nature conspires to create, at least in this particular object,
holes which are near spherical in shape.
Based on the absence of elliptical holes, an interesting conclusion
can be drawn concerning magnetic fields. Usually, the influence of
magnetic fields is neglected in numerical studies of expanding bubbles
in the ISM. However, a few analytical approaches have been put
forward in recent years to investigate the effects of the evolution of
shell--like structures in the presence of magnetic fields (e.g.,
Tomisaka \markcite{TOM91}1991). The results of these investigations are
that if global magnetic fields are present, this would inhibit the
expansion of a supernova driven bubble in directions perpendicular to
the field lines. The expansion along the field lines would not be
affected, thus leading to elliptical, elongated structures. The fact
that we observe only spherical holes in IC~2574 shows that either
magnetic fields play a minor role or that the magnetic fields in
IC~2574 are really weak.
\section{Summary}
\label{summa}
High resolution \ion{H}{1} and H$\alpha$ imaging using the VLA and the
Calar Alto observatory of the dwarf galaxy IC~2574 have been presented
in this paper. The main results and conclusions are:
1. IC~2574 is a gas--rich dwarf galaxy which shows a stunning amount
of detail in the form of \ion{H}{1} shells and holes in its
interstellar medium. These features are similar to those found in the
Galaxy, the Magellanic Clouds, M~31, M~33, M~101 and NGC~6946, IC~10
and Ho~II. From the global \ion{H}{1} profile we derive a total
\ion{H}{1} mass of $9.0\times 10^8\,\mbox{M}_{\odot}$.
2. Using two independent approaches, the thickness of the \ion{H}{1}
layer of IC~2574 has been derived to be $\approx 350$ pc. This implies
that IC~2574, as is the case for Ho~II, is a 'thick' dwarf galaxy
(compared to more massive spirals) which can be understood in terms of
its lower gravitational potential.
3. Large scale structures emptied of \ion{H}{1} (believed to be
expanding \ion{H}{1} shells) leave roughly circular footprints,
\ion{H}{1} holes. They are prominent in the high resolution VLA
channel maps of IC~2574. In total, 48 \ion{H}{1} holes have been
catalogued which are distributed across the entire galaxy. The
diameters of the holes range between 100 pc and 1200 pc. The lower
limit of the sizes is due to the limit of our spatial resolution. Most
of the holes are found to be expanding with typical expansion
velocities ranging between 8 and 12 $\,\mbox{km}\,\mbox{s}^{-1}$. The indicative ages of the
\ion{H}{1} holes range between 10 and $60\times 10^6$ years. Typical
energies that are required to create the observed features are of the
order of 10--100$\times 10^{50}$~ergs assuming that they were created
by the combined effects of stellar winds and supernova explosions. The
\ion{H}{1} masses that were present at the position of the holes
before creation (and that might now be present on the rim of the
shell) are of the order of 10--100$\times 10^4$~M$_{\odot}$.
4. In addition to the well defined holes, some large scale ($> 1000$
pc), coherent features are visible in the channel maps. They might be
the remainder of an older shell population.
5. The radial expansion of the holes, the indicative ages and the
energy requirements for their formation suggest that they have been
created by the combined effects of stellar winds and multiple
supernova explosions. Despite their large dimensions, the energy
requirements are not extreme and infall of material such as clouds is
not needed to explain the observed features. However, alternative
mechanisms cannot be ruled out.
6. The shells can grow to these large dimensions because of several
conditions which are fulfilled in dwarf galaxies. The volume density
in the plane is low, which facilitates expansion. In the
direction perpendicular to the disk, the gravitational pull is smaller
than in a massive spiral. Also, because of the thick \ion{H}{1}
layer, shells are easily contained and unlikely to blow--out. Lastly,
solid body rotation and a lack of spiral density waves prevent holes
from being rapidly destroyed.
7. A comparison with other galaxies studied in similar detail so far
(M~31, M~33 and Ho~II) shows that the size distribution of \ion{H}{1}
holes found in a galaxy is related to its Hubble type in the following
way. The size of the largest \ion{H}{1} shells is inversely
proportional to the global gravitational potential (and hence mass
surface density). The shells in dwarf galaxies show lower expansion
velocities, most likely due to the fact that the shells are preserved
much longer due to the absence of shear (differential rotation). The
energies needed to create these structures, though, are found to be
roughly the same for all types of galaxies. If we follow the
conventional view that star forming regions are responsible for the
structures found in \ion{H}{1}, the energy output of a typical
star forming region does not seem to be related to the host galaxy.
8. The fact that virtually all holes in IC~2574 are circular in shape
is a major difference with the other galaxies studied in comparable
detail and can be explained, again, by the absence of shear and spiral
density waves in IC~2574. This finding also indicates that magnetic
fields play only a minor role in shaping expanding holes or that the
magnetic fields in IC~2574 are weak.
9. Current star formation, as traced by H$\alpha$ emission, is
predominantly found along the rims of the larger \ion{H}{1}--holes,
indicating propagating star formation. A detailed comparison of the
H$\alpha$ and the \ion{H}{1} properties yields that, on a global
scale, star formation is only present if the \ion{H}{1} surface
density reaches values larger then $10^{21}$~cm$^{-2}$ (on a linear
scale of $\approx 95$~pc).
In summary, our results on IC~2574 confirm the findings by PWBR92 and
show that Ho~II is not a special case. Our analysis, and the
comparison with other galaxies studied in similar detail suggest that
there is a trend of hole characteristics with Hubble type (or
mass). If the conclusion that the energy output of a SF region is
independent of the host galaxy holds then it follows that there is a
lower mass limit for the formation of dwarf galaxies. In other words,
if the mass of a galaxy falls below a certain threshold, the first
burst of star formation will tear the whole galaxy apart (see also,
e.g., Puche \& Westpfahl \markcite{PUC94}1994). Mac~Low \& Ferrara
\markcite{MAF98}(1998) predict that this lower mass limit will be near
$10^6$~M$_{\odot}$, which is well below the mass of the objects dealt
with in this paper.
\acknowledgments
The authors are indebted to Mordecai--Marc Mac~Low for providing them
with a model data cube of a turbulent ISM in advance of publication and
to Johan van Horebeek for valuable advise regarding some of the
statistical techniques employed. FW acknowledges the 'Deutsche
Forschungsgemeinschaft (DFG)' for the award of a stipendium in the
framework of the Graduiertenkolleg on "The Magellanic Clouds and other
Dwarf Galaxies" and would like to thank Uli Klein for continuous
support and Neb Duric for fruitful discussions. FW appreciates the
help of the staff of the Array Operation Center of the NRAO during his
stay in Socorro, New Mexico and also wants to thank the staff of the
Calar Alto observatory in Spain for excellent working conditions. FW
is grateful to Jan Palou\v s for illuminating discussions regarding
the numerical simulations of holes and shells during his stays at the
Academy of Science in Prague, Czech Republic. EB acknowledges a grant
awarded by CONACyT (grant number 0460P--E) and is grateful for the
support by the Graduiertenkolleg without which this collaborative
effort would not have been possible. Last but not least we would like
to thank Klaas de Boer for carefully reading the manuscript and an
anonymous referee for useful suggestions. This research has made use
of the NASA/IPAC Extragalactic Database (NED) which is operated by the
Jet Propulsion Laboratory, Caltech, under contract with the National
Aeronautics and Space Administration (NASA), NASA's Astrophysical Data
System Abstract Service (ADS), and NASA's SkyView.
|
\section{Introduction}
Now, it has become a standard point of view that a fundamental theory (well
defined for the extremely high energy end) underlies the present QFTs that
are in fact low energy (LE) effective theories for the phenomena in LE
ranges \cite{wein}. But as far as the author knows, we are still lacking a
formulation that can yield finite results in a natural way that fully makes
use of the standard point of view. A new strategy is proposed in Ref. \cite
{YYY} that indicates the power of the standard point of view if one uses it
appropriately.
To focus on the UV problem, we will assume from now on that there were no
unphysical IR singularity in the LE models in our discussions or we have
already had an IR regular formulation for the LE QFTs. (We will discuss in
our future works about the IR structure's contribution to the whole
formulation--it should be included to arrive at a totally satisfying
formulation, especially for QCD-like theories where the IR singularity is
rather severe and affects the theories' predictions \cite{Bigi}).
It is convenient to employ a generating functional formalism \cite{Shirkov}
or a path integral formalism to assemble the Green functions for the LE
sectors of the underlying theory. It is natural to expect that the
well-defined path integral for an effective sector take the following form,
\FL
\begin{equation}
Z_{\{\sigma\}}(\{J^{i}\})= \int D{\mu}(\phi^{i}_{\{\sigma\}}) \exp
\{iS(\phi^{i}_{\{\sigma\}};\{J_{i}\}; \{\sigma\})\}
\end{equation}
where $\{\sigma\}$ are the fundamental parameters (some fundamental
constants, probably including the gravitation constant) from the underlying
theory and $\{J^{i}\}$ are the external sources for LE sectors. The
'elementary fields' $(\phi^{i}_{\{\sigma\}})$ in the LE sectors are here
appended by the underlying parameters to indicate that they are in fact
effective ones. It is easy to see that for different LE physics, the LE
limit operation may act upon sets of underlying parameters that differ in
part.
Within the path integral formalism, we can easily see that in the LE limit
(denoted by ${\bf L}_{\{\sigma \}}$)
\FL
\begin{eqnarray}
&&Z^0(\{J^i\};\overline{\cdots })\equiv {\bf L}_{\{\sigma \}}\int D{\mu }%
(\phi _{\{\sigma \}}^i)\exp \{iS(\phi _{\{\sigma \}}^i;\{J_i\};\{\sigma
\})\}
\\
&\neq &\int D{\mu }(\phi ^i)\exp \{iS(\phi ^i;\{J_i\})\},
\end{eqnarray}
where the symbols not appended by the underlying parameters refer to the
constants and field parameters given by the present QFTs (i.e., the LE limit
has been applied on these 'effective' fields or objects). Thus, generally
speaking, we can not let the operation of the LE limit cross the summation
over the paths or the intermediate states. In other words, the LE limit
operation and the summation over intermediate states {\it do not commute},
\FL
\begin{equation}
\Delta \equiv [{\bf L}_{\{\sigma \}},\sum_{\{paths\}}]\neq 0.
\end{equation}
If one ignores such a subtle fact, the resulting formulations would be ill
defined and needs regularizations that introduce unjustified artificial
substitutes of the true underlying structures, which will in turn lead to
divergences may appear and subtraction is needed. This is our opinion about
the appearance of the unphysical infinities in some conventional QFT
formulations.
In the path integral formalism it is immediate to see that the spectra given
by the conventional Hamiltonian models would differ from true ones given by
the underlying theory, especially in the UV regions. The deviation is
signaled by the ill-definedness or UV divergence in the conventional QFTs.
More severe UV divergence implies more severe deviation.
What we are trying to present in the following is that if one starts merely
with the existence of the underlying theory and $\{\sigma\}$ (without
knowing the details), there is a simple strategy to calculate the amplitudes
wanted without introducing any {\it ad hoc } regularization or cutoff that
leads to UV divergence. But, due to the lacking of the true underlying
structures, there {\it must remain} in our approach certain ambiguities
signaling the missing of the underlying structures, which are to be fixed in
principle by phenomenology and experiments.
This paper is organized in the following way: We will first exemplify our
strategy in Feynman graph language for the one-loop case in section II. The
treatment of the multi-loop cases is given in section III. Then we discuss
some nonperturbative examples in section IV to further illustrate our
proposal. Finally we summarize our presentation in Section V. We should note
in advance that here we do not claim a satisfactory, systematic and final
presentation of a new approach, instead we just put forward an alternative
strategy for dealing with ill-definedness.
\section{How Can UV Finite Results be Derived}
From our discussion in section I, we see that the Hamiltonians (and hence
the propagators and vertices) in their present forms are the LE limits of
the ones characterized by $\{\sigma \}$. To interest most readers we will
exemplify our strategy in the perturbative Feynman graph language (we remind
that our strategy is definitely not confined to the perturbative case and we
will exhibit the use of the strategy in some nonperturbative cases in
section IV). Thus, according to our postulate, the conventional ill-defined
(or divergent) Feynman amplitudes (FAs) are results of illegitimate
operation order with the LE limit operation and the internal momenta
integration. In formula, if the integrand $f(\{Q_i\},\{p_j\},\{m_k\})$ (of
an ill-defined FA) corresponds to the integrand $\bar{f}(\{Q_i\},\{p_j\},%
\{m_k\};\{{\sigma }_l\})$ given by the underlying theory with $%
\{Q_i\},\{p_j\},\{m_k\}$ and $\{{\sigma }_l\}$ being respectively loop
momenta, external momenta, masses and the fundamental parameters in the
underlying theory, then
\FL
\begin{eqnarray}
&&\Gamma ^0(\{p_j\},\{m_k\};\{\bar{c}\})={\bf L}_{\{\sigma \}}\overline{%
\Gamma }(\{p_j\},\{m_k\};\{\sigma _l\})={\bf L}_{\{\sigma \}}\int
\prod_id^nQ_i\bar{f}(\{Q_i\},\{p_j\},\{m_k\};\{\sigma _l\}) \\
&\neq &\int \prod_id^nQ_i{\bf L}_{\{\sigma \}}\bar{f}(\{Q_i\},\{p_j\},\{m_k%
\};\{\sigma _l\})=\int \prod_id^nQ_if(\{Q_i\},\{p_j\},\{m_k\}),
\end{eqnarray}
where $\{\bar{c}\}$ denotes the definite constants arising from the LE limit
operation. $\Gamma ^0$ and $\overline{\Gamma }$ are well-defined (finite)
but $\int \prod_id^nQ_if(\{Q_i\},\{p_j\},\{m_k\})$ is ill defined. That
means, the commutator
\FL
\begin{equation}
\delta _{\{\sigma \}}=\left[ {\bf L}_{\{\sigma \}},\int \prod_id^nQ_i\right]
\end{equation}
only vanishes identically for convergent (i.e., well-defined) FAs, otherwise
we encounter divergence or ill-definedness in FAs. The deviation of the
effective formalism is not detected by the convergent FAs, or these
amplitudes are well defined in the present QFTs. This is an extremely
important fact for our purpose in the following.
As the underlying theory or the amplitudes $\bar{f}(...;\{\sigma_{l}\})$ are
unavailable by now, we have to find a way to approach the truth--the $%
\Gamma^{0}(\{p_{j}\},\{m_{k}\})$'s--from our present partial knowledge about
the LE sectors of the underlying quantum theory --the present form of the
Hamiltonians or Lagrangians as LE sectors of the underlying theory.
Here is our strategy for extracting finite results out of effective
formulations of QFT: (1) First we try to perform certain {\it legitimate}
operations (say, $\Omega $) on the objects {\it so that} the LE limit
operation commutes with the summation on the resulted objects if they do not
commute on the original objects. (2) Then we can safely perform the
intermediate states summation (e.g., loop integration) on the resulted
objects. (3) At last, we perform the inverse operations (say, $\Omega
^{-1}$%
, which should also be legitimate) to go back and the final expressions
should be (UV) finite by construction but probably ambiguous at the meantime
due to our lack of knowledge about the underlying structures. The
ambiguities should be fixed from the phenomenological and experimental
inputs as mentioned above. The operations ($\Omega $'s) and their inverse
are of the main concern in our strategy and they would often be operations
with respect to the parameters external to the intermediate states since the
objects of interests are expressed in terms of these parameters after all.
The more general framework effecting the natural strategy proposed here will
be the subject of our future investigations. Here, we only show how the
strategy works in a simple way for the Feynman graph approach, i.e., we
exemplify our simple strategy in the Feynman graph language. First we show
that the following important relation holds for 1-loop case ill-defined FAs
(c.f. Eq.(5) for 1-loop case)
\FL
\begin{equation}
\int d^{n}Q \left ({\partial}_{p_{j}} \right )^{\omega}
f(Q,\{p_{j}\},\{m_{k}\})= \left ( {\partial}_{p_{j}} \right )^{\omega}
\Gamma^{0} (\{p_{j}\},\{m_{k}\}),
\end{equation}
with $\omega-1$ being the usual superficial divergence degree of $\int
d^{n}Q f (Q,\{p_{j}\},\{m_{k}\})$ so that the left hand side of Eq.(8)
exists (finite), $\left ({\partial}_{p_{j}} \right
)^{\omega} $ denoting
the differentiation's with respect to the external parameters $\{p_{j}\}$'s
of the amplitude and $\Gamma^{0}(...)$ is the LE limit of the amplitude
calculated in the underlying theory (i.e., the internal momentum integration
is performed first). It is easy to see that the operation $\left
({\partial}%
_{p_{j}} \right )^{\omega}$ leads to convergent graphs or objects.
The proof is very simple, since
\FL
\begin{eqnarray}
&&\int d^{n}Q \left ({\partial}_{p_{j}} \right )^{\omega} f
(Q,\{p_{j}\},\{m_{k}\})= \int d^{n}Q \left ({\partial}_{p_{j}}
\right)^{\omega} {\bf L}_{\{\sigma\}} \bar{f} (Q,\{p_{j}\},\{m_{k}\};\{%
\sigma_{l}\}) \nonumber \\
&&= \int d^{n}Q {\bf L}_{\{\sigma\}} \left ({\partial}_{p_{j}} \right
)^{\omega}\bar{f} (Q,\{p_{j}\},\{m_{k}\};\{\sigma_{l}\}) = {\bf L}%
_{\{\sigma\}} \int d^{n}Q \left ({\partial}_{p_{j}} \right )^{\omega}\bar{f}
(Q,\{p_{j}\},\{m_{k}\};\{\sigma_{l}\}) \nonumber \\
&&={\bf L}_{\{\sigma\}} \left ({\partial}_{p_{j}} \right)^{\omega}
\overline{%
\Gamma} (\{p_{j}\},\{m_{k}\};\{\sigma_{l}\}) = \left ( {\partial}_{p_{j}}
\right )^{\omega} \Gamma^{0} (\{p_{j}\},\{m_{k}\}).
\end{eqnarray}
The second and the fifth steps follow from the commutativity of the two
operations $\left ({\partial}_{p_{j}} \right )^{\omega}$ and ${\bf L}%
_{\{\sigma\}}$ as they act on different arguments, the third step is due to
the existence of $\int d^{n}Q \left({\partial}_{p_{j}} \right )^{\omega}
f(Q,...)$ and the fourth is justified from the existence of $\int d^{n}Q
\bar{f} (Q,...;\{\sigma_{l}\}) ( = \overline {\Gamma} (...;\{\sigma_{l}\}))$
by postulate.
It is clear that here the differentiation with respect to the external
momenta and its 'inverse'--indefinite integration with respect to the same
momenta play the role of the certain operations stated above.
The right hand side of Eq.(8) can be found now as the left end exists as a
nonpolynomial (nonlocal) function of external momenta and masses, i.e.,
denoting it as $\Gamma _{(\omega )}^0$,
\FL
\begin{equation}
\left( {\partial }_{p_j}\right) ^\omega \Gamma ^0(\{p_j\},\{m_k\})=\Gamma
_{(\omega )}^0(\{p_j\},\{m_k\}).
\end{equation}
To find $\Gamma ^0(\{p_j\},\{m_k\})$, we integrate both sides of Eq.(10)
with respect to the external momenta ''$\omega $'' times indefinitely to
arrive at the following expressions
\FL
\begin{eqnarray}
\left( \int_{{p}}\right) ^\omega \left[ ({\partial }_{{p}})^\omega \Gamma
^0(\{p_j\},\{m_k\})\right] =\Gamma ^0(\{p_j\},\{m_k\})+N^\omega
(\{p_j\},\{c_\omega \}) =\Gamma _{npl}(\{p_j\},\{m_k\})+N^\omega
(\{p_j\},\{C_\omega \})
\end{eqnarray}
with $\{c_\omega \}$ and $\{C_\omega \}$ being arbitrary constant
coefficients of an $\omega -1$ order polynomial in external momenta $%
N^\omega $ and $\Gamma _{npl}(\{p_j\},\{m_k\})$ being a definite
nonpolynomial function of momenta and masses. Evidently $\Gamma
^0(\{p_j\},\{m_k\})$ is not uniquely determined within conventional QFTs at
this stage. That the true expression
\FL
\begin{equation}
\Gamma ^0(\{p_j\},\{m_k\})=\Gamma _{npl}(\{p_j\},\{m_k\})+N^\omega
(\{p_j\},\{\bar{c}_\omega \}),\ \ \ \bar{c}_\omega =C_\omega -c_\omega
\end{equation}
contains a definite polynomial part (unknown yet) implies that it should
have come from the LE limit operation on $\overline{\Gamma }%
(\{p_j\},\{m_k\};\{\sigma _l\})$ (see Eq.(5)) as the usual convolution
integration can not yield a polynomial part--an indication of the
incompleteness of the formalism of the QFTs.
We can take the above procedures as efforts for rectifying the ill-defined
FAs and ''represent'' the FAs with the expressions like the right hand side
of Eq.(11), i.e.,
\FL
\begin{equation}
\int d^nQf(Q,\{p_j\},\{m_k\})>=<\Gamma _{npl}(\{p_j\},\{m_k\})+N^\omega
(\{p_j\},\{C_\omega \})
\end{equation}
with ''$>=<$'' indicating that left hand side is rectified as the right hand
side. That the ambiguities reside only in the local part means that the QFTs
are quite effective in the LE limit.
To find the $\{\bar{c}_{\omega}\}$'s in Eq.(12) we need inputs from the
physical properties of the system (such as symmetries, invariances and
unitarity) and a complete set of data from experiments \cite{CK,LL} (if we
can derive them from the underlying theory all these requirements would be
automatically fulfilled). In other words, all the ambiguities should be
fixed according to this principle. Similar approach had been adopted by
Llewellyn Smith to fix ambiguities on Lagrangian level by imposing high
energy symmetry, etc. on relevant quantities \cite{LL}. It is also the
physical reasoning followed by the conventional renormalization programs.
As we have seen, the $\bar{c}_\omega $'s arise in fact from the low energy
limit operation on the objects calculated in the underlying theory, they
should be uniquely defined up to possible equivalence. Different or
inequivalent choices of these constants simply correspond to different LE
theories (amount to being defined by different underlying theories). Since
different regularizations and/or renormalization conditions might correspond
to inequivalent choices of the constants, (the problem might be more severe
in nonperturbative cases \cite{QMDR}), they might lead to different LE
theories that even could not faithfully describe relevant low energy
physics. Although the underlying structures do not appear explicitly in the
LE formulations, they are not totally decoupled from the effective ones and
they 'stipulate' the effective sectors indirectly through the constants $\{%
\bar{c}\}$. In other words, these constants missing from the effective
Hamiltonian models are just what we need to define the quantum theory
completely in addition to the canonical parameters (masses and couplings).
Their missing in the Hamiltonian or Lagrangian level does not mean they do
not exist. As we will see in section IV, the former studies on the
self-adjoint extension \cite{QM} of some quantum mechanical Hamiltonians
just support our point of view here.
\section{Multi-loop Case}
Since the UV divergence would appear if one first take the limit before
doing loop momenta integrations, our strategy here is just to move the limit
operator ${\bf L}_{\{\sigma \}}$ across the loop integration operations in
such a way that no potential divergence is left over.
For any multi-loop graph $\Gamma $ (we will use the same symbol to denote
the graph and the associated FA if it is not confusing), we should again
start with the amplitude derived from the underlying theory, i.e., $%
\overline{\Gamma }(\ldots ;\{\sigma \})$ with the same graph structure. All
the internal lines and vertices are again understood to be given by the
underlying theory, characterized by the presence of the parameters $\{\sigma
\}$. Then the LE limit of $\overline{\Gamma }(\ldots ;\{\sigma \})$ produces
the definite constants $\{c^0\}$ that are unknown to us yet
\FL
\begin{equation}
\Gamma ^0(\ldots ;\{c^0\})={\bf L}_{\{\sigma \}}\overline{\Gamma }(\ldots
;\{\sigma \})={\bf L}_{\{\sigma \}}\int \prod_ld^nl{\bar{f}}_\Gamma
(\{l\},\ldots ;\{\sigma \})
\end{equation}
where $\bar{f}_\Gamma (\{l\},\ldots ;\{\sigma \})$ denotes the integrand
obtained from the underlying theory corresponding to the graph $\Gamma $ and
the dots refer to the LE parameters like external momenta, mass parameters
and coupling constants. Other symbols are self-evident. We will use in the
following $\omega _\gamma -1$ to denote the overall divergence index \cite
{Shirkov} for any graph $\gamma $ and $\{l\}$ to represent the internal
momenta and all the partial differentiation operators and their 'inverse'
($%
\partial _{\omega _\gamma }^{-1}$) act upon the momenta only 'external' to
the very internal integration of the graph under consideration. In certain
cases, masses are appropriate or efficient external parameters to work with.
If the graph is totally convergent, then $\Gamma ^0$ contains no UV
ambiguity and the limit operation can cross all the internal integrations to
act upon the integrand to yield the product of the propagators and vertices
given by the present QFTs. But once there is any potential UV
ill-definedness associated with any internal integration, one must proceed
in the following way (suppose that a graph $\Gamma $ contains at least an
overall divergence)
\FL
\begin{eqnarray}
&&\Gamma ^0(\ldots ;\{c_i^0\})={\bf L}_{\{\sigma \}}\int \prod d^nl\bar{f}%
_\Gamma (\{l\},\ldots ;\{\sigma \}) \nonumber \\
&\Rightarrow &\partial _{\omega _\Gamma }^{-1}{\bf L}_{\{\sigma \}}\int
\prod d^nl\partial ^{\omega _\Gamma }\bar{f}_\Gamma (\{l\},\ldots ;\{\sigma
\})=\sum_{\{\gamma \}=\partial ^{\omega _\Gamma }\Gamma }\partial _{\omega
_\Gamma }^{-1}{\bf L}_{\{\sigma \}}\int \prod d^nl\bar{f}_\gamma
(\{l\},\ldots ;\{\sigma \}).
\end{eqnarray}
The differentiation with respect to the external parameters will give rise
to a sum of graphs $\{\gamma \}$ (without overall divergence) from the
original graph $\Gamma $. (Note that any overlapping divergence will be
killed by the differentiation operation, only non-overlapping divergences
remain, i.e., the overlapping divergences are disentangled \cite{CK}). If
there is no more ill-definedness (in any subgraph), one can move the limit
operator across all the internal integrations to act directly upon the
integrands $\bar{f}_\gamma (\{l\},\ldots ;\{\sigma \})$ just like in the
totally convergent graph case. Then one can carry out all the loop
integrations without any trouble for each graph $\gamma $ and then sum them
up and finally apply the 'inverse' operator with respect to the parameters
(usually momenta) external to the graph $\Gamma $ (and each $\gamma $).
Now suppose there are still some ill-defined subgraphs in each $\gamma $. In
this case, each graph in the set ${\partial }^{\omega _\Gamma }\Gamma $ can
be expressed as a 'product' of disconnected divergent (at least
superficially divergent) subgraphs (each subgraph itself may contain
overlapping divergences). The LE limit operator crossed all the other parts
and stopped before the divergent subgraphs. In formula, for each graph $%
\gamma $,it is
\FL
\begin{eqnarray}
&&\partial _{\omega _\Gamma }^{-1}{\bf L}_{\{\sigma \}}\int \prod
d^nl\bar{f}%
_\gamma (\{l\},\ldots ;\{\sigma \})=\partial _{\omega _\Gamma }^{-1}\left\{
\int \prod d^n\overline{l^{\prime }}g_{\gamma /{[\gamma ^{\prime }]}}(\{%
\overline{l^{\prime }}\},\ldots ){\bf L}_{\{\sigma \}}\prod_{\gamma
_j^{\prime }}\int \prod_{i\epsilon \gamma
_j^{\prime }}d^nl_i^{\prime }{\bar{%
f}}_{\gamma :\gamma _j^{\prime }}(\{l^{\prime }\}_{\gamma _j^{\prime
}},\ldots ;\{\sigma \})\right\} \nonumber \\
&=&\partial _{\omega _\Gamma }^{-1}\left\{ \int \prod d^n\overline{l^{\prime
}}g_{\gamma /{[\gamma ^{\prime }]}}(\{\overline{l^{\prime }}\},\ldots
)\prod_{\gamma _j^{\prime }}\left[ {\bf L}_{\{\sigma
\}}{\overline{\Gamma }}%
_{\gamma :\gamma _j^{\prime }}(\ldots ;\{\sigma \})\right] \right\} , \\
&&(\bigcup_{\gamma _j^{\prime }}\{l^{\prime }\}_{\gamma _j^{\prime
}})\bigcup \{\overline{l^{\prime }}\}=\{l\},\ \ \ \ \ [\gamma ^{\prime
}]\bigcup \gamma /{[\gamma ^{\prime }]}=\gamma ,\ \ \ [\gamma ^{\prime
}]=\prod_j\gamma _j^{\prime },\gamma _j^{\prime }\bigcap \gamma _k^{\prime
}=0,\ \ j\neq k,
\end{eqnarray}
where all the dots in the expressions refer to the parameters 'external' to
the loop integrations for the subgraphs (i.e., to $\gamma _j^{\prime }$%
)--they contain the external parameters for the original graph $\Gamma $
(also for all the graphs in $\{\gamma \}$) {\it and} the internal momenta in
the set $\{\overline{l^{\prime }}\}$. $\overline{\Gamma }_{\gamma :\gamma
_j^{\prime }}$ refers to amplitude derived from the underlying theory that
corresponds to each subgraph $\gamma _j^{\prime }$ contained in $\gamma $.
{\it Since some loop momenta are 'external' to certain subgraphs, one can
not first carry out these loop integrations before the ill-defined subgraphs
are treated. } This is in sheer contrast to the totally convergent graphs
where the loop integration order does not matter.
As the ill-defined subgraphs in $[\gamma ^{\prime }]$ are disconnected with
each other, we now treat each of them separately as a new 'total' graph just
like what we have done with the total graph $\Gamma $ starting from Eq.(14).
Then we go through the procedures from Eq.(14) to Eq.(17) till we encounter
new disconnected and ill-defined subgraphs that are in turn to be treated as
before. Finally, we will go to the smallest subgraphs that are completely
convergent. Now we can finally remove the LE limit operator to get the
integrands totally expressed with propagators and vertices given by the
effective theories and we can begin to perform all the loop integrations
with insertions of various pairs of $\partial _{\omega _\gamma }^{-1}$ and
$%
\partial ^{\omega _\gamma }$ and there will appear a natural order of
integration due to our procedures above: the smallest subgaphs are
integrated first, then the ''larger'' subgraphs with their subgraphs done
already, and then the still ''larger'' subgraphs and so on, till all
integrations are done. [It is worthwhile to note that at each level of the
subgraphs, the loop integrations are guaranteed to be convergent due to
Weinberg's theorem \cite{weinth}].
The resulting expression will be a definite nonlocal functions plus nonlocal
ambiguities (due to subgraph ill-definedness) and local ambiguities if $%
\Gamma$ is suffering from overall divergence,
\FL
\begin{eqnarray}
&&\Gamma^0(\ldots; \{c^0\}) \Rightarrow \Gamma(\ldots;\{C\})
=\Gamma^{npl}_{0}(\ldots)+ \Gamma^{npl}_{1}(\ldots;\{C^{\prime}\})+
N^{\omega_{\Gamma}}(\ldots;\{\bar{C^{\prime}}\}), \\
&&\{C^{\prime}\} \bigcup \{\bar{C^{\prime}}\}=\{C\}.
\end{eqnarray}
Here again we used $N^{\omega_{\Gamma}}$ to denote the polynomial containing
the ambiguities ($\{\bar{C^{\prime}}\}$) appearing due to the overall
divergence. Others are nonlocal functions. Different from the single loop
case, there are nonlocal ambiguities in this multiloop graph suffering from
subgraph divergences (as evident from our treatment) in addition to the
nonlocal definite part and the local ambiguous part. The result we obtained
(%
$\Gamma (\ldots;\{C\})$) is not what we are really after ($%
\Gamma^0(\ldots;\{c^0\})$), but that is the best we can do with the present
QFT.
Now some remarks are in order:
{\bf A}. It is evident that overlapping divergences are just automatically
resolved in our proposal, because the differentiation operators just remove
the overlapping ill-definedness by 'inserting' internal lines and vertices
to reduce the overall divergence. Thus one need not worry about them any
more. This is the utility derived from the differentiation with respect to
external parameters (momenta, masses or other massive parameters that might
appear in the LE propagators) \cite{CK}.
{\bf B}. The amplitudes given by the underlying theory should be invariant
under any linear transformations of the internal integration variables. In
our treatment of the ill-defined graphs, since every loop integration
actually performed is convergent, these transformations do not alter the
results of the loop integrations. Due to the 'inverse' operations, these
linear transformations would at most change the polynomial part. But that
does not matter at all. This observation implies that one should not worry
about the variable shifting and routing of the external momenta that belong
to the transformations just described if he has noted the ambiguous part.
{\bf B1}. An immediate corollary to this observation is that, the chiral
anomaly, which is conventionally interpreted as due to the variable shifting
in relevant linearly divergent amplitude, must have been due to other
definite properties. Otherwise, if it were totally due to the local
ambiguities, one can well remove them away by choosing appropriate
definitions of the constants (or appropriate renormalization conditions).
Our direct calculation shows that \cite{JF1}, one kind of definite rational
terms (independent of masses) originated the chiral anomaly. Since they are
nonlocal and unambiguous, one can not attribute them simply as UV effects.
The trace anomaly is also shown to be originated by such kind of rational
terms \cite{JF1}. To our best knowledge, this nontrivial structure
(independent of the UV ambiguities) has never been noted before in the old
renormalization framework.
{\bf B2}. Another utility derived from the observation is that, one can
choose the routes of flows of the external momenta to be as simple as
possible to make the treatments of an ill-defined multi-loop amplitude as
easy as possible. For the single loop cases, sometimes one may only focus on
the parts of the amplitude that are really divergent. This may yield fewer
ambiguities.
{\bf C}. With the above deduction, one can easily see that the results of
any regularization and/or renormalization scheme can be readily reproduced
by corresponding choices of the constants. That is, our proposal can lead to
a universal formulation for all the regularization and renormalization
schemes at least in the perturbative framework. If one wish to pursue
calculation efficiency in the first place, one can choose a regularization
scheme that saves labor and then replace all the divergent expressions with
an ambiguous polynomial in external parameters to arrive at the same
expressions that can be obtained in the procedures described above.
Conventionally one has to check whether the symmetry properties of a scheme
would affect the main body of a theory and/or works very hard in order to
setup a scheme as consistently as possible. Our proposal spares the labor of
checking every corner of a scheme and just make use of its efficiency.
For the ambiguities, as we have discussed in the section II, we may first
impose some novel symmetries and invariances on the amplitudes to reduce the
ambiguities to certain degree, then one has to resort to the experimental
physics data, a strategy effectively employed in the conventional
renormalization approaches. The W-T identities have served as constraints in
the conventional renormalization schemes and saved a lot of labor of
calculation. They are just certain kinds of graphical relations among the
Feynman graphs. So, in the Feynman graph language, we can derive constraints
upon the ambiguities from the graph structures \cite{YYY}.
Moreover, it is worthwhile to note that our strategy works in any kind of
models as long as they are physically effective and consistent. That is, it
applies efficiently to the unrenormalizable theories as well as the
renormalizable ones in the conventional terminology. It would be interesting
to integrate our approach with the BV anti-field formalism of QFT \cite{BV}
that has been used to deal with the unrenormalizable theories quite recently
\cite{gom}. In contrast, we do not need any counterterm in our strategy for
whatever kind of unrenormalizable models simply because we do not encounter
any divergence, the only trouble is to fix the ambiguities..
As we have pointed out, the Feynman Amplitudes or the 1PI functions are
generally parametrized by more than one constants (I will refer to them as
'agent constants') in addition to the phenomenological ones. If the changes
in the radiative constants could be completely compensated by that in the
phenomenological ones (which is only possible for rather special kind of
models), then we might implement a redefinition invariance of the
phenomenological constants or parameters for the FAs like in the RG case.
Recently we have rederived the Callan-Symanzik like equations governing the
scale transformation behaviors of QFTs without divergence and bare
parameters. The influences of the underlying structures effected through the
agent constants are clearly shown and these agent constants complete the
harmony of the scale transformation of the effective QFTs in the sense
discussed in Ref. \cite{Scale}. Moreover, the new equations improved the
conventional ones in quite some aspects.
\section{Nonperturbative Examples}
Iit is obvious that our proposal works in principle for any model, whether
it is a QFT or not. The key observation that the UV ill-definedness is
caused by illegitimate order of 'operations' is valid in both perturbative
and the nonperturbative contexts. That is to say, our simple and natural
strategy should also be applicable to nonperturbative case.
Recently, the cutoff and Dimensional regularizations are compared in
nonperturbative context \cite{QMDR} in the hot topic of applying the idea of
effective field theory method (EFT \cite{EFFT}) to LE nuclear physics
following Weinberg's suggestion \cite{WeinEFT}. The framework is a
non-relativistic quantum mechanics with Delta-potentials, which is ill
defined in the short-distance. According to our discussions above, such
ill-definedness means that the effective LE models must have failed in the
higher energy end. Then it is illegitimate to simply work with the
propagators and vertices (or Green functions and potentials) given by such
models. Great care must also be taken with respect to the regularization
effects. Thus the inequivalence between the cutoff scheme and Dimensional
regularization exhibited in Ref. \cite{QMDR} well evidenced the correctness
of our arguments given above.
Now let us try to treat the problem within our proposal. Generally, the
Lippmann-Schwinger equation for $T$-Matrix in the simple two-body problems
reads (we follow the notation conventions of Ref. \cite{QMDR})
\FL
\begin{equation}
T(p^{\prime},p; E)=V(p^{\prime},p)+ \int \displaystyle\frac{d^d k}{(2\pi)^d}
V(p^{\prime},k ) \displaystyle\frac{1}{E^{+}-k^2/(2\mu)} T(k,p; E),
\end{equation}
where $E^{+}$ is $E + i \epsilon$, with $E$ non-negative, and $\mu$ denotes
the reduced mass in the two-body problem. In our point of view, this
equation is not well-defined and should be written as the LE limit of that
derived from the well-defined fundamental underlying theory which is
unavailable to us by now. [The underlying parameters will be always denoted
as $\{\sigma\}$. For different problems or different LE ranges, the contents
may differ.] So in our language, Eq.(20) should be rewritten as
\FL
\begin{eqnarray}
&&T(p^{\prime},p; E ; \{\sigma\}) = V(p^{\prime},p ; \{\sigma\}) + \int %
\displaystyle\frac{d^d k}{(2\pi)^2} V(p^{\prime},k ; \{\sigma\})
G(E^{+}-k^2/(2\mu); \{\sigma\}) T(k,p;E; \{\sigma\}), \\
&&V(p^{\prime},p) \equiv {\bf L}_{\{\sigma\}} V(p^{\prime},p ; \{\sigma\}),
\displaystyle\frac{1}{E^{+}-k^2/(2\mu)}\equiv {\bf L}_{\{\sigma\}}
G(E^{+}-k^2/(2\mu); \{\sigma\}).
\end{eqnarray}
Eq.(21) is now well-defined in the underlying theory. Thus Eq.(20) is
correct only when there is no UV infinities (there is no IR problem in the
following discussions for the Delta-potential problem) so that the LE limit
operator can cross the internal momentum integration (summation over
intermediate states). Otherwise we have to find a legitimate way to let the
LE limit operator cross the internal momenta integration.
In the case of Delta-potential, $V(p^{\prime },p)=C$, but the $V(\ldots
;\{\sigma \})$ is generally a nonlocal potential before the LE limit is
taken. To be rigorous, we write formally
\FL
\begin{eqnarray}
T(p^{\prime },p;E;\{c^0\}) &=&C+{\bf L}_{\{\sigma \}}\left\{ \int %
\displaystyle\frac{d^dk}{(2\pi )^d}V(p^{\prime },k;\{\sigma \})
G(E^{+}-k^2/(2\mu );\{\sigma \})T(p^{\prime },p;E;\{\sigma \})\right\} ,
\end{eqnarray}
here it is not generally legitimate to move the $V(\ldots ;\{\sigma \})$ out
of the integration to be directly subject to the LE limit operator--which is
exactly what was done in the conventional calculations (with only the
propagator regularized)--and it is definitely illegitimate to apply the LE
limit operator to all the other objects before the integration is done.
Thus, in principle, even when the LE potential is local (of course $V(\ldots
;\{\sigma \})$ is nonlocal), it might be dangerous to simply reduce Eq.(21)
to an algebraic one. Only when the ill-definedness is mainly caused by $%
1/(E^{+}-k^2/(2\mu ))$ (i.e., it differs greatly from $G(\ldots ;\{\sigma
\}) $ in the UV region where $V(\ldots )$ differs less from $V(\ldots
;\{\sigma \})$), could we pull out the true potential to subject it directly
to the action of the LE limit operator. In other words, to put Eq.(21) (a
correct formulation for Eq.(20)) or Eq. (23) into an algebraic one requires
quite nontrivial properties of the potential and the propagator. To focus on
the main point, we temporarily assume this condition is satisfied, then we
have the well-defined form of the algebraic equation for the $T-$matrix
(which is now parametrized by the new constants $\{c^0\}$ from the LE limit
in addition to $E$),
\FL
\begin{equation}
\displaystyle\frac 1{T^{on}(E;\{c^0\})}=\displaystyle\frac 1C-I(E;\{c^0\}),
\end{equation}
with
\FL
\begin{equation}
I(E;\{c^0\})={\bf L}{\{\sigma \}}\int \displaystyle\frac{d^dk}{(2\pi )^d}%
G(E^{+}-k^2/(2\mu );\{\sigma \}).
\end{equation}
Now we can employ the technique described in sections II and III to
calculate the integrals, i.e., first differentiate $G(E^{+}-\ldots
;\ldots )$
with respect to $E^{+}$ (which is the 'external' parameter in the integral)
for appropriate times, secondly perform the LE limit legitimately and carry
out the integral thus obtained, finally do the 'inverse' operation with
respect to $E$ and we find the followings (note that here one
differentiation with respect to $E$ reduces the divergence degree by two)
\FL
\begin{eqnarray}
&&I_{d;odd}(E;\{c^0\})\Rightarrow I_{d;o}(E;\{c^{\prime }\})
=-i\displaystyle%
\frac{2\mu {\pi }^{d/2+1}}{\Gamma (d/2)(2\pi )^d}(2\mu
E)^{d/2-1}+N^{[(d-1)/2]}(2\mu E;\{c^{\prime }\}); \\
&&I_{d;even}(E;\{c^0\})\Rightarrow I_{d;e}(E;\{c^{\prime }\})
=\displaystyle%
\frac{2\mu \pi ^{d/2}}{\Gamma (d/2)(2\pi )^d}(2\mu E)^{d/2-1}\ln (2\mu
E/c_0^{\prime })+N^{[d/2]}(2\mu E;\{c^{\prime }\})
\end{eqnarray}
with $\{c^{\prime }\}$ being arbitrary constants--the ambiguities. These
expressions can again be viewed as universal parametrization and compared
with that given in cutoff regularization and dimensional regularization
schemes (C.f. Ref \cite{QMDR}) with the latter ones as special cases.
In terms of the ambiguous (but {\it finite }) integrals given by Eq.(26,27),
the $T-$matrix is now parametrized by $\{c^{\prime}\}$ in addition to $E$
like
\FL
\begin{equation}
\displaystyle\frac{1}{T^{on}(E;\{c^{\prime}\})} = \displaystyle\frac{1}{C} -
I_{d;\ldots}(E;\{c^{\prime}\}).
\end{equation}
Again we need to fix the constants $\{c^{\prime}\}$ rather than to
renormalize the interaction constant $C$.
It is easy to see that following the normalization condition of Ref. \cite
{QMDR}, we can reproduce the result derived by Weinberg \cite{WeinEFT} in
two or three dimensional space-time. However, there seems to be no necessary
constraints on the phenomenological constant $C$ as it is physical rather
than ' bare' in our proposal. Thus, this LE framework \'{a} la Weinberg
works equally well for both the attractive interactions and the repulsive
ones in our approach , contrary to the conclusions that EFT framework failed
in the repulsive cases where the LE models are believed to be trivial \cite
{Beg}. The nontriviality of the Delta-potential dynamics has also been
investigated by Jackiw \cite{Jackiw}.
This problem can also be examined from another angle. Jackiw had already
pointed out \cite{Jackiw} that the Hamiltonians for such models are not
automatically Hermitian and need self-adjoint extension. This has already
been dealt with by mathematicians in the operator theory and has also been
extensively discussed in a number of approaches (please refer to \cite
{Albeverio} for a comprehensive list of the references). The key point is,
in such cases, the self-adjointness of the Hamiltonian is never an automatic
property. That is, in contrast to the normal case, the contact potential
problem \'{a} la Schr\"{o}dinger equation is UV ill-defined. The resolution
of the problem gives rise to a family of self-adjoint extensions of the
original Hamiltonian operator parametrized by an additional constant, which
upon different choices leads to different or inequivalent (LE) physics \cite
{QM}. {\it This additional 'family' parameter is just the constant that will
surely be predicted from the LE limit operation in our proposal}, as has
been emphasized in section II and III. As a matter of fact, there is an
approach that is quite the same as ours in spirit, the one based on
resolvent formalism \cite{Grossmann} where an important object is defined
through an equation in which it appeared in a form differentiated with
respect to the 'external' parameter--the resolvent variable (energy). Thus
this important object is only defined up to an additional parameter (the
family parameter in operator theory approach) which is to be determined by
other input, just like in our proposal.
I would like to mention a recent calculation \cite{GEP} of Higgs masses in
nonperturbative context employing our proposal. The results thus obtained
are neat and clear, in contrast to that performed within the old
renormalization framework (see the references in \cite{GEP}). Especially,
the physical pictures are different from that using the old renormalization,
which is easy to see from the discussions above.
One could expect that great ease can be found in employing our proposal or
its equivalents (in any form known or unknown) in his/her studies in the
nonperturbative contexts and the outcome would be quite different and
significant. Moreover, within our approach, those phenomenologically
oriented models which are unrenormalizable in the usual renormalization
schemes could become quite tamed. One can test it with the NJL model and
chiral perturbation theory \cite{CPT} and with gravity \cite{Dono}. We will
further apply our strategy to more concrete calculations and develop more
rigorous framework of this strategy in the future.
\section{Discussions and summary}
For a complete representation of the world, we should expect that the
underlying theory is also well defined in the IR sector. (Our discussions in
the introduction about the spectra are partial as we deliberately omitted
the IR issues to focus on the UV structures.) It is conceivable that the
phenomenological models give wrong information of the IR end spectra
signaled by the unphysical IR infinities.
The underlying theory, 'postulated' here, if exists, should contain all the
nontrivial UV and IR structural information lost in the effective theories.
Then an interesting scenario dawns upon us: for each effective model
dominating certain energy range (say, theory $T_{mid}$), there should exist
two other effective models (or sectors) that are most adjacent to this model
from the IR end and UV end respectively (say, $T_{IR}$ and $T_{UV}$). Then
it is imaginable that the phenomenological parameters in $T_{IR}$ and/or $%
T_{UV}$ would at least quite nontrivially improve the IR and/or UV behaviors
of the theory $T_{mid}$. While on the other hand, the $T_{mid}$ contains
what $T_{IR}$ (or $T_{UV}$) needs to improve its UV (or IR) behaviors. Put
it another way, the active and 'elementary' modes or fields in $T_{IR}$ will
break up in $T_{mid}$ and give way to the new 'elementary' modes active in
$%
T_{mid}$. Similarly, the 'elementary' modes in $T_{mid}$ will go
'hibernating' as the energy goes down while 'new' elementary modes 'emerge'
to dominate spectra in $T_{IR}$. The relation between the elementary modes
in $T_{mid}$ and $T_{UV}$ is in principle just like that between those in $%
T_{IR}$ and $T_{mid}$. Of course, there may be modes active in several
successive effective models, some may even be active and stable through all
energy levels---the 'fossil' modes or fields we mentioned in the
introduction. Evidently, the information about those 'elementary' modes in
$%
T_{IR}$ and $T_{UV}$ missing from $T_{mid}$ (i.e., missing from the
effective spectrum given by $T_{mid}$) can contribute to improve the IR and
UV behavior of the latter.
Of course the author does not have a solid idea for the answer right now.
Thus we only indicate some interesting but plausible picture about the
resolution of the unphysical IR problem or the reformulation of the IR end
of a theory suffered from IR problems like QCD.
From our proposal, the 'elementary'commutator for a field and its conjugate,
if calculated (or formulated) from the underlying theory, must have been at
least a nonlocal function(al) parametrized by the underlying parameters of
the underlying theory and must have been closely related with the
gravitational interaction and perhaps new fundamental ones, rather than a
highly abstract Dirac delta function containing least information. In a
sense, the incompleteness of the present QFTs or their ill-definedness is
inherent in the present quantization procedure whose most elementary
technical building block is the Dirac delta function (called as distribution
by mathematicians) that is {\it extremely singular and oversimplified in the
UV ends.} That the distribution theory works necessarily with test function
space or appropriate measure, if viewed from physical angle, is equivalent
to that we need more 'fundamental structures' in order for some singular
functions to make sense, i.e., a necessity of introducing underlying theory
or its artificial substitute--regularization.
In a short summary, we discussed the strategy recently proposed by the
author and the important consequences following from it. The proposal could
overcome many typical difficulties and shortcomings associated with old
regularization and renormalization framework and can be applied in principle
to any UV ill-defined quantum theories.
\section*{Acknowledgement}
The author is very grateful to Professor F. V. Tkachov for his encouragement
and enlightening remarks.
|
\section{Introduction}
The Cosmic Microwave Background (CMB) provides a unique probe of the
early universe (see White, Scott, \& Silk 1994 for a review). If CMB
fluctuations are consistent with inflationary models, future
ground-based and satellite experiments will yield accurate
measurements of most cosmological parameters (see Zaldarriaga,
Spergel, \& Seljak 1997; Bond, Efstathiou, \& Tegmark 1997 and
reference therein). These measurements rely on the detection of
primordial anisotropies produced at the surface of last
scattering. However, various secondary effects produce fluctuations at
lower redshifts. The study of these secondary fluctuations (or
extragalactic foregrounds) is important in order to isolate primordial
fluctuations. In addition, secondary fluctuations are interesting in
their own right since they provide a wealth of information on the
local universe.
In this contribution, I present an overview of the different
extragalactic foregrounds of the CMB. The foregrounds produced by
discrete sources, the thermal Sunyaev-Zel'dovich (SZ) effect, the
Ostriker-Vishniac (OV) effect, the Integrated Sachs-Wolfe (ISW)
effect, gravitational lensing, and other effects, are briefly
described. I show their relative importance on the multipole-frequency
plane, and pay particular attention to their impact on the future CMB
missions \hbox{MAP} (Bennett et al.\ 1995) and Planck
Surveyor(Bersanelli et al.\ 1996). A more detailed account of each
extragalactic foreground can be found in the other contributions to
this volume. In this article, I have focused on the latest literature,
and have not aimed for bibliographical completeness. This overview is
based on a more detailed study of extragalactic foregrounds in the
context of the \hbox{MAP} mission (Refregier et al. 1998).
\section{Comparison of Extragalactic Foregrounds}
\label{foregrounds}
To assess the relative importance of the extragalactic foregrounds, I
decompose the temperature fluctuations of the CMB into the usual
spherical harmonic basis,
$\frac{\delta{T}}{T_{0}}(\theta)=\sum_{\ell,m} a_{lm} Y_{lm}(\theta)$,
and form the averaged multipole moments $C_{l}\equiv \langle
|a_{lm}|^{2} \rangle$. Following Tegmark \& Efstathiou (1996), I
consider the quantity $ \Delta T_{\ell} \equiv \left[ \ell(2 \ell+1)
C_{\ell}/4 \pi \right]^{\frac{1}{2}} T_{0}$, which gives the {\it rms}
temperature fluctuations per $\ln \ell$ interval centered at
$\ell$. Another useful quantity that they considered is the value of
$\ell=\ell_{eq}$ for which foreground fluctuations are equal to the
CMB fluctuations, i.e.\ for which $C_{\ell}^{\rm foreground} \simeq
C_{\ell}^{\rm CMB}$. Note that, since the foregrounds do not
necessarily have a thermal spectrum, $\Delta T_{\ell}$ and $\ell_{eq}$
generally depend on frequency.
The comparison is summarized in table~\ref{tab:foregrounds} and in
figure~\ref{fig:lnu}. Table~\ref{tab:foregrounds} shows $\Delta
T_{\ell}$ and $\ell_{eq}$ for each of the major extragalactic
foregrounds at $\nu=94$ GHz and $\ell=450$, which corresponds to a
FWHM angular scale of about $\theta \sim .3$ deg. These values were
chosen to be relevant to the \hbox{MAP} W-band ($\nu \simeq 94$ GHz
and $\theta_{\rm beam} \simeq 0\fdg21$. I also indicate whether each
foreground component has a thermal spectrum.
Figure~\ref{fig:lnu} summarizes the importance of each of the
extragalactic foregrounds in the multipole-frequency plane. It should
be compared to the analogous plot for galactic foregrounds (and
discrete sources) shown in Tegmark \& Efstathiou (1996; see also
Tegmark 1997 for an updated version). These figures show regions on
the $\ell$-$\nu$ plane in which the foreground fluctuations exceed the
CMB fluctuations, i.e.\ in which $C_{\ell}^{\rm foreground} >
C_{\ell}^{\rm CMB}$. As a reference for $C_{\ell}^{\rm CMB}$, a COBE
normalized CDM model with $\Omega_{b}=0.05$ and $h=0.5$ was used. Also
shown in figure~\ref{fig:lnu} is the region in which \hbox{MAP} and
Planck Surveyor are sensitive, i.e.\ in which $\Delta C_{\ell}^{\rm
noise} < C_{\ell}^{\rm CMB}$, where $\Delta C_{\ell}^{\rm noise}$ is
the {\it rms} uncertainty for the instrument. Note that this figure is
only intended to illustrate the domains of importance of the different
foregrounds qualitatively.
In the following, I briefly describe each extragalactic foreground
and comment on its respective entries in table~\ref{tab:foregrounds}
and figure~\ref{fig:lnu}.
\begin{table}
\caption{Summary of Extragalactic Foregrounds for $\nu=94$
GHz and $\ell=450$.}
\label{tab:foregrounds}
\begin{center}\small
\begin{tabular}{rrrrrr}
Source & $\Delta T_{\ell}$ ($\mu$K)\tablenotemark{a} &
$\ell_{\rm eq}$\tablenotemark{b} & Thermal\tablenotemark{c} &
Note & Ref.\tablenotemark{d} \\
\tableline
CMB\tablenotemark{e} & 50 & & yes & & 1\\
Discrete\tablenotemark{f} & 5 & 1800 & no & $S<1.0$ Jy & 2 \\
& 2 & 3100 & no & $S<0.1$ Jy & 2 \\
SZ\tablenotemark{g} & 10 & 1900 & no & C &3 \\
& 7 & 2300 & no & NC &3 \\
OV\tablenotemark{h} & 2 & 2900 & yes & $z_{r}=50$ & 4 \\
& 1 & 3100 & yes & $z_{r}=10$ & 4 \\
ISW & 1 & 5000 & yes & $\Omega h=0.25$ & 5 \\
& 0.9 & 5800 & yes & $\Omega h=0.50$ & 5 \\
Lensing & 5 & 2400 & yes & & 6
\end{tabular}
\end{center}
\tablenotetext{a}{$\Delta T_{\ell} \equiv [\ell(2\ell +1)C_{\ell}/4\pi]^{1/2}$
centered at $\ell=450$ and $\nu=94$ GHz.}
\tablenotetext{b}{Value of $\ell$ for which $\Delta T_{\ell}=\Delta
T_{\ell,{\rm CMB}}$}
\tablenotetext{c}{Thermal (yes) or nonthermal (no) spectral dependence}
\tablenotetext{d}{1: Seljak \& Zaldarriaga 1996; 2: Toffolatti
et al.\ 1998; 3: Persi et al.\ 1995; 4: Hu \& White 1996;
5: Seljak 1996a; 6: Zaldarriaga \& Seljak 1998a}
\tablenotetext{e}{Primordial CMB fluctuations for a CDM model
with $\Omega_{m}=1$, $\Omega_b=0.05$, and $h=0.5$}
\tablenotetext{f}{Discrete sources with 94 GHz removal threshold of
0.1, 1 Jy, respectively}
\tablenotetext{g}{SZ effect with (C) and without (NC) cluster cores
respectively.}
\tablenotetext{h}{OV effect with two different reionization redshifts
$z_{r}$}
\end{table}
\begin{figure}
\centerline{\epsfig{file=fig1.ps,width=12cm}}
\caption{Summary of the importance of extragalactic foregrounds of the
CMB. Each filled area on the multipole-frequency plane corresponds to
regions where the foreground fluctuations exceed those of the CMB. The
sparse and dense dotted regions correspond to discrete sources (with
$S<1$ Jy) and the Sunyaev-Zel'dovich effect (with cluster cores),
respectively. The horizontal, descending and ascending hashed regions
correspond to the Ostriker-Vishniac effect (with $z_{r}=50$),
gravitational lensing, and the integrated Sachs-Wolfe effect (with
$\Omega h = 0.25$), respectively. The areas marked \hbox{MAP} (thick
line) and Planck (thin line) show the regions of sensitivity for each
of the future missions, i.e.\ regions where the CMB fluctuations
exceed the noise for each of the instruments (see text).}
\label{fig:lnu}
\end{figure}
\subsection{Discrete Sources}
Discrete sources produce positive, point-like, non-thermal
fluctuations. While not much is known about discrete source counts
around $\nu \sim 100$ GHz, several models have been constructed by
interpolating between radio and IR observations
(Toffolatti et al.\ 1998; Gawiser \& Smoot 1997; Gawiser et al. 1998;
Sokasian et al. 1998). Here, I adopt the model of
Toffolatti et al.\ and consider the two flux limits $S <1$ and 0.1 Jy
for the source removal in table~\ref{tab:foregrounds}. The sparsely
dotted region figure~\ref{fig:lnu} shows the discrete source region
for $S <1$ Jy. In the context of CMB experiments, the Poisson shot
noise dominates over clustering for discrete sources (see Toffolatti
et al.\ ). As a result, the discrete source power spectrum,
$C_{\ell}^{\rm discrete}$, is essentially independent of
$\ell$.
\subsection{Thermal Sunyaev-Zel'dovich Effect}
\label{foregrounds_sz}
The hot gas in clusters and superclusters of galaxies affect the
spectrum of the CMB through inverse Compton scattering. This effect,
known as the Sunyaev-Zel'dovich effect (for reviews see Sunyaev \&
Zel'dovich 1980; Rephaeli 1995), results from both the thermal and
bulk motion of the gas. We first consider the thermal SZ effect, which
typically has a larger amplitude and has a non-thermal spectrum (see
the \S\ref{OV} below for a discussion of the kinetic SZ effect). The
CMB fluctuations produced by the thermal SZ effect have been studied
using the Press-Schechter formalism (see Bartlett 1997 for a review),
and on large scales using numerical simulations (Cen \& Ostriker 1992;
Scaramella, Cen, \& Ostriker 1993) and semi-analytical methods (Persi
et al.\ 1995). Here, I consider the SZ power spectrum, $C_{\ell}^{\rm
SZ}$, calculated by Persi et al.\ (see their figure 5). In
table~\ref{tab:foregrounds}, I consider their calculation both with
and without bright cluster removal. In figure~\ref{fig:lnu}, only the
spectrum without cluster removal is shown.
\subsection{Ostriker-Vishniac Effect}
\label{OV}
In addition to the thermal SZ effect described above, the hot
intergalactic medium can produce thermal CMB fluctuations as a result
of its bulk motion. While this effect essentially vanishes to first
order, the second order term in perturbation theory, the
Ostriker-Vishniac effect (Ostriker \& Vishniac 1986; Vishniac 1987),
can be significant on small angular scales. The power spectrum of the
OV effect depends on the ionization history of the universe, and has
been calculated by Hu \& White (1996), and Jaffe \& Kamionkowski
(1998; see also Persi et al.\ 1995). We use the results of Hu \& White
(see their figure 5) who assumed that the universe was fully reionized
beyond a redshift $z_{r}$. In table~\ref{tab:foregrounds}, I consider
the two values $z_{r}=10$ and 50, while in figure~\ref{fig:lnu}, I
only plot the region corresponding to $z_{r}=50$. For consistency, the
standard CDM power spectrum is still used as a reference, even though
the primordial power spectrum would be damped in the event of early
reionization. (Using the damped primordial spectrum makes, at any
rate, only small corrections to both table~\ref{tab:foregrounds} and
figure~\ref{fig:lnu}.)
\subsection{Integrated Sachs-Wolfe Effect}
The Integrated Sachs-Wolfe Effect (ISW) describes thermal CMB
fluctuations produced by time variations of the gravitational
potential along the photon path (Sachs \& Wolfe
1967). Linear density perturbations produce non-zero ISW fluctuations
in a $\Omega_m \neq 1$ universe only. Non-linear perturbations produce
fluctuations for any geometry, an effect often called the Rees-Sciama
effect (Rees \& Sciama 1968). Tuluie
\& Laguna (1995) have shown that anisotropies due to intrinsic changes
in the gravitational potentials of the inhomogeneities and
anisotropies generated by the bulk motion of the structures
across the sky generate CMB anisotropies in the range of
$10^{-7} \la \frac{\Delta T}{T} \la 10^{-6}$ on scales of about
$1^{\circ}$ (see also Tuluie et al. 1996). The power
spectrum of the ISW effect in a CDM universe was computed by
Seljak (1996a; see also references therein). In
table~\ref{tab:foregrounds}, I consider values of the density
parameter, namely $\Omega h=0.25$ and $0.5$. In figure~\ref{fig:lnu},
only the $\Omega h=0.25$ case is shown. As above, the standard CDM
($\Omega =1$, $h=0.5$) spectrum is still used as a reference.
\subsection{Gravitational Lensing}
Gravitational lensing is produced by spatial perturbations in the
gravitational potential along the line of sight (see Schneider,
Ehlers, \& Falco 1992; Narayan \& Bartelmann 1996). This effect does
not directly generate CMB fluctuations, but modifies existing
background fluctuations. The effect of lensing on the CMB power
spectrum was calculated by Seljak (1996b) and Metcalf \& Silk
(1997). Recently, Zaldarriaga \& Seljak (1998a) included the lensing
effect in their CMB spectrum code (CMBFAST; Seljak \& Zaldarriaga
1996). This code was used to compute the absolute lensing correction
$|\Delta C_{\ell}^{\rm lens}|$ to the standard CDM spectrum, including
nonlinear evolution. The results are shown in
table~\ref{tab:foregrounds} and figure~\ref{fig:lnu}.
\subsection{Other Extragalactic Foregrounds}
In addition to the effects discussed above, other extragalactic
foregrounds can cause secondary anisotropies. For instance, patchy
reionization produced by the first generation of stars or quasars can
cause second order CMB fluctuations through the doppler effect
(Aghanim et al. 1996a,b; Gruzinov \& Hu 1998; Knox, Scoccimaro,
\& Dodelson 1998; Peebles \& Juzkiewicz 1998).
Calculations of the spectrum of this effect are highly uncertain, but
show that the resulting CMB fluctuations could be of the order of 1
$\mu$K on 10 arcminute scales, for extreme patchiness. More likely
patchiness parameters make the effect negligible on these scales, but
potentially important on arcminute scales. Another potential
extragalactic foreground is that produced by the kinetic SZ effect
from Ly$_{\alpha}$ absorption systems, as was recently proposed by
Loeb (1996). The resulting CMB fluctuations are of the
order of a few $\mu$K on arcminute scales, and about one order of
magnitude lower on 10 arcminute scales. Because of the uncertainties
in the models for these two foregrounds and because they are small on
10 arcminute scales, they are not included in
table~\ref{tab:foregrounds} and figure~\ref{fig:lnu}.
\section{Discussion and Conclusion}
An inspection of table~\ref{tab:foregrounds} shows that, at 94 GHz and
$\ell=450$, the power spectra of the largest extragalactic foregrounds
considered are a factor of 5 below the primordial CDM spectrum. As can
be seen in figure~\ref{fig:lnu}, the dominant foregrounds for
\hbox{MAP} and Planck Surveyor are discrete sources, the thermal SZ
effect and gravitational lensing. Note that, for Planck surveyor,
these three effects produce fluctuations which are close to the
sensitivity of the instrument. The spectra of the OV and ISW effects
will produce fluctuations of the order of $1 \mu$K , and are thus less
important for a measurement of the power spectrum. The effect of
gravitational lensing is now incorporated in CMB codes such as
CMBFAST, and can thus be taken into account in the estimation of
cosmological parameters. The other two dominant extragalactic
contributions, discrete sources and the thermal SZ effect, must also
be accounted for, but are more difficult to model. Note that, on large
angular scales, extragalactic foregrounds produce relatively small
fluctuations, and are thus not detectable in the COBE maps (Boughn \&
Jahoda 1993; Bennett et al.\ 1993; Banday et al.\ 1996; Kneissl et
al.\ 1997)
While I have concentrated above on the power spectrum, secondary
anisotropies are also a source of non-gaussianity in CMB
maps. Discrete sources and the SZ effect from clusters of galaxies
mainly produce Poisson fluctuations and are thus clearly
non-gaussian. The other extragalactic foregrounds (SZ, OV, ISW, and
lensing) are also non-gaussian and trace large-scale structures in the
local universe. As a consequence of the latter fact, extragalactic
foregrounds can be probed by cross-correlating CMB maps with galaxy
catalogs, which act as tracers of the large scale structure. Such
technique can be used to detect the ISW effect (Boughn et al. 1998,
and reference therein), gravitational lensing (Suginohara et al. 1998)
and the SZ effect by superclusters (Refregier et
al. 1998). Gravitational lensing is particularly interesting since it
produces a specific non-gaussian signature (Bernardeau 1998). This
signature can be used to reconstruct the gravitational potential
projected along the line of sight (Zaldarriaga \& Seljak 1998b).
Further non-gaussian signatures result from the fact that the
different extragalactic foregrounds are spatially correlated. For
instance, a detection of the cross-correlation signal between
gravitational lensing and the ISW and SZ effects would allow us to
determine the fraction of the ionized gas and the time evolution of
gravitational potential (Goldberg \& spergel, 1998; Seljak \&
Zaldarriaga 1998). A detection of secondary anisotropies would help
break the degeneracy between cosmological parameters measured from
primary anisotropies alone.
\acknowledgments I thank David Spergel and Thomas Herbig for active
collaboration and discussions on this project. This work was
supported by the \hbox{MAP} MIDEX program.
|
\section{Prologue}
This paper deals with a topic of current interest in Theoretical Physics
and is presented to the forum of historians and philosophers of science.
It assumes familiarity at the popular level with developments
in High Energy Physics and with basic Quantum Physics. In
rephrasing technical statements, an attempt is made to remain
close to the truth, albeit selectively. Math is used sparingly
but some formulae are displayed in the hope that they will give
the reader an opening into more detailed literature. Sections
2 and 3 deal with the paradigm of symmetry as it has come to be
understood in this century.
Supersymmetry is an elegant symmetry principle, but seems
to not be operating in nature in its simplest version.
Section 4 deals with Supersymmetry; the idea, its appeal
and its failings. Section 5 presents two
alternatives for the metaphysical status of Supersymmetry
in case it is indeed discovered.
\section{The intangible microworld}
The macroscopic world directly impinges on the senses
and demands systematizing principles. Presenting itself
in many different contexts, it also provides ample clues for
arriving at such principles.
Majority of the phenomena of common experience are correctly
described by Newtonian principles, complemented by laws
governing electromagnetism, hydrodynamics and so on.
The microscopic world is nevertheless present. It is intangible
except for a few tangible and powerful clues. The shape and
solidity of the world relies on Fermi statistics. Several
phenomenological constants contain Planck's constant
or Avogadro's number. The scientist has had to progressively
become a detective relying on skimpy evidence to pursue the trail
of an elusive and magnificent if unseeable reality.
Consider an example of clues leading to detection.
Valence was wrested from Chemical phenomena, after much confusion
and controversy. Mendeleev's periodic table, at first based on
valence, systematized the elements and predicted new ones. The
raison d'etre of the table remained a mystery until the electronic
structure of the atom could be understood. This in turn needed
the Pauli exclusion principle for its explanation; in turn bringing us
to the very heart of microscopic phenomena, the fundamental
indistinguishability of quanta. The message of Quantum Mechanics
is that only the possible quantum states of a collection of quanta
that are distinct, not the quanta themselves.
\subsection{The metaphysics of insight}
An important paradigm for theoretical progress in this
century has been symmetry principles. This is in contrast
to the development of Electromagnetism which occurred over
about two centuries, during which theory and experiment
progressed step by step, aiding each other. The exploration
of the microworld beginning with radioactivity did not
enjoy such a luxury of wealth of data, nor of easily
constructible and repeatable experiments.
The developments starting during 1880's and culminating in 1930's
therefore relied on deep insights guided by
certain metaphysical assumptions.
Here and in the following, by metaphysical we shall mean
principles external to the discipline of physics itself,
but nevertheless conceived and used by professionals.
It is the implicit use of such principles that is the subject
of this paper. Also, being external to the discipline itself,
they appropriately form the subject matter of Philosophy of Science.
There are two simple but deep principles used universally
in science. These are, (1) universal applicability of the concepts,
(2) consistency of the epistemy. By the latter we mean
the expectation that existing technical frameworks or
formalisms will apply also to a relatively new domain of phenomena.
An example of above principles at work is provided by
the discovery of Bose Statistics. Planck's explanation
of Black Body radiation relied on assumption of absorption
and emission of radiation energy in quanta. Einstein made
this fact into a new concept, that of a photon, and used
it to explain photoelectric effect. This put the photon
on a more general footing. The next step, which took many
years in coming, was taken by Bose who assumed that the
thermodynamics of photons must be deducible from a
counting of states just as in classical Statistical
Mechanics. This is what we mean by principle (2). There was
one revolutionary new input required however. The counting of
states is based on strict indistinguishability among the photons.
We cite these as examples of insight working in conjunction
with above guiding principles. In the Quantum domain
however, both proved unreliable. Neither the concepts could
be universally applied, nor the epistemy. Mathematically precise
entities and rules were in some sense the only infallible
guide. Which concepts would remain robust and what exactly
these rules were took a long time for its understanding.
Barring possible new phenomena such as
Hawking radiation, Quantum Mechanics as we know today is
consistent and complete but does not cease to evoke
disbelief even in eminent practitioners.
There was however another metaphysical principle which was
emerging as means of guessing ahead. Very loosely it may
be called the principle that the equations must be elegant
and must incorporate a certain symmetry.
It was based on this principle that
Einstein's theory of Gravitation and Dirac's theory for the
relativistic electron were accepted by Physics community with
awe and excitement even before they could be completely
established. In its highly evolved form today, it has come
to be further formalized as a demand for
the existence of precise, mathematically implementable
symmetry principles.
The origins of this metaphysical
principle go back to the nineteenth century, when Maxwell
achieved an elegant unification of the laws of electromagnetism.
He organized several laws and rules of thumb then known
so that the Electric and Magnetic forces appeared on par with each
other displaying an uncanny similarity between the two.
The laws were also stated in the form of mathematical equations
that permitted easy geometrical visualization and were yet
so far reaching in their import and applicability that an
eminent colleague is reported to have exclaimed,
quoting Goethe, ``was it a god who wrote those lines?"
With hindsight we know that in fact the symmetry they
displayed went much farther than a nineteenth century esthete
could have discerned, for they were the first equations
to be known which were covariant under Special
Relativistic transformations.
What we are trying to identify as a principle is
actually rather broad and perhaps contains more logically
distinct positions than one. We shall focus here on a much
more restrictive and precise aspect of this principle.
Specifically we refer to the use of symmetry principles that tend
to restrict theories and introduce economy of phenomenological
parameters in it. We shall refer to it broadly as Gauge Symmetry.
It started with Einstein trying to formulate
relativistically consistent laws of Gravitation, and
later also helped to shape the laws of strong and weak interactions.
The gauge symmetry underlying General Relativity on the one
hand and Gauge Field Theories of strong and weak interactions
on the other hand have several technical differences. But
as has been emphasized by Weinberg\cite{weingrco}, they
have an essential similarity to permit being viewed as
manifestations of the same basic principle.
This is the topic of the next section.
\section{Gauge Symmetry}\label{sec:ggesym}
The most common example of a mathematically implementable
symmetry principle in Physics is the idea of
rotations.~\footnote{This section
is an abridged version of the author's contribution to
the Seminar on Philosophy of Science, IIT Bombay, 1993.}
We do not expect outcomes of experiments to depend on the
directional orientation of the apparatus. Considered as a
set of operations, the rotations form a mathematical
structure called Group.
One representation of this group is in terms of
matrices, acting on vectors such as the position vector or
the electric field vector.
The Special Relativity principle is a similar principle, in
fact a generalized rotation involving time, such that the
ordinary rotations form a subgroup of this bigger group.
But this notion of symmetry was completely revolutionized
by Einstein in his subsequent work, viz., General
Relativity. This theory is actually a theory of Gravity,
generalized from its Newtonian version and made consistent
with Special Relativistic rotations. The prescription of
General Relativity can be summarized in two parts : (1) The
space-time should be treated as curved, like the surface of
a ball. Thus gravitational influences are described by a
set of space-time dependent functions that specify the
distance and angle measurement prescriptions. In
a curved space these replace the Pythagorian distance law
from point to point.
(2) In a curved space one does not choose a rigid,
Cartesian system of coordinates, but
any convenient curvilinear coordinates. So the laws of
physics must be such as to remain invariant under arbitrary
choices of curvilinear coordinates. This translates to
invariance of the laws under rotations that can be different at
different points. This two part law was called the
Principle of General Covariance.
This theory was a great speculative triumph. At the time of its
invention, there was no evidence for it. No one had
suspected that the perihelion precession of Mercury had
contributions from Relativistic effects,
requiring fundamental reformulation of Newtonian Gravity.
No other experimental evidence
existed that demanded such a generalization.
In the 1930's the notion of rotational symmetry was extended by
Heisenberg in a very profound way. It is known that the
strong nuclear force does not depend separately on the
physical state of the proton or that of the neutron. In
Quantum Mechanics, the physical state of a system is
described by a complex wavefunction denoted $\psi$. Heisenberg's
proposal was that instead of using $\psi_p$ (for proton)
and $\psi_n$ (for neutrons), if we used
\begin{eqnarray}
\tilde{\psi_p} &= c_1 \psi_p + c_2 \psi_n\cr
&\cr
{\rm and}\qquad
\tilde{\psi_n} &= c_3 \psi_n + c_4 \psi_p
\end{eqnarray}
the physics would remain unchanged. Here the $c_1, c_2$, $c_3, c_4$
are complex numbers satisfying some constraints. The relations
above can be thought of as a complex rotation, an abstract
generalization from the case of real vectors.
This rotation, called an isospin rotation is a
symmetry (although approximate) of the strong nuclear force.
Several decades later, Yang and Mills proposed Gauge Field
Theories. These were a generalization of isospin symmetry
in much the same way as General Relativity generalized Special
Relativity. Both prescribed the form of the interaction,
although the requirement was stated as a geometrical law.
To summarize, precise mathematical principles were used
as a {\it strategy} to guess at a theory with insufficient
experimental evidence. The theory could well have proved
wrong. This too has happened many times, as for instance
with the original Kaluza-Klein theory. But the success of
the cases in which this approach has worked is spectacular.
\subsection{Broken symmetry} The curious fact about
symmetries is that sometimes they may not be manifest in
the data. This can happen due to two different reasons.
One reason is that the symmetry may be only approximate.
That is, only by ignoring some of the data or by modifying
their values does one see the symmetry principle at work.
For this to be true, the contaminating effects should be
small in a quantitative sense. But this case of
non-manifest symmetry is not as interesting as the next
one.
It has been found that in some systems, the governing
equations possess a certain symmetry. However, the
complexity of the interactions drives the system to
solutions that do not reflect the symmetry. This case
is called Spontaneous Breakdown of symmetry. In case of
weak nuclear interactions, one seeks a theory that obeys
gauge invariance, somewhat similar to the two-part
principle of General Covariance. However, gauge invariance
implies masslessness of the mediating particles, whereas
the mediators of the weak force are known to be massive.
The resolution of this paradox lay in realizing that there
could be additional particles, known as Higgs whose complex
dynamics leads to the ground state of the system not
explicitly displaying the gauge symmetry. Under these
circumstances, the interaction of the gauge particles with
the Higgs particles makes the former massive.
In the second type of broken symmetry, the symmetry is all
the time present, being made invisible by the particular
state in which the system is available to us.
In this case, guessing the governing equations is difficult
but symmetry can be used as a guiding principle.
\section{Supersymmetry}
This brief history prepares us for a description of the
new proposal of Supersymmetry. The origins of the search
for this rather bizarre symmetry to be described lie in
two unrelated motivations. One was a direct one, asking
whether photon and neutrino, the only two particles
known to be massless in 1960's had anything more in common.
Specifically whether they were two manifestations of the
same particle ``species" masquerading as two.
Secondly, it was also
a search for {\it the most general} type of symmetries
allowed by interactions that respect the basic rotational
and Special Relativistic symmetries of space time.
There was also an enigma in the distinction between
the gauge symmetry of Gravity which involved space-time
itself and the gauge symmetry of Nuclear forces, which
seemed to operate in an abstract space of wave functions.
The General Covariance of Gravity came to be called an
external gauge symmetry and the Gauge symmetry of the
nuclear forces internal symmetries. The possible kinds of
internal symmetries were soon classified in terms of
the mathematical theory of Lie Groups. It is a rich variety
of possible symmetries. The question was,
what were the most general kinds of {\it external} symmetry
and whether there could be any mixing between external and
internal.
The above question was supposedly answered with some degree
of finality by a so called "No-go" theorem which appeared
in the late sixties. It said that all the possible symmetries
one could possibly have, consistent with Quantum Mechanics
were the ones already known, viz., a variety of internal
symmetries like isospin on the one hand and the
already known external symmetries, those of the Special
Theory of Relativity (subsuming the old known symmetry of
rotations) on the other hand. There was nothing new to
be added to the category of external.
There was a loop hole however. In order to understand it,
let us look at a mathematical statement of internal
consistency of several symmetry operations is formulated.
It can be checked by some amount of careful experimentation
that a small amount of $x$-axis rotation followed by a
small amount of $y$-axis rotation, is not the same as
the $y$-rotation first followed by same
$x$-rotation.
This can actually be checked by holding up a pen. The
results of the two operations differ by a small $z$-axis
rotation! This was first put in the form of equations by Hamilton
in mid-nineteenth century. In modern notation one says
\begin{equation}
L_x L_y - L_y L_x = L_z
\ee
Here $L_x$ stands for the operation of small $x$-axis
rotation. The product $L_xL_y$ has to be read right to left
for its factors. The left hand side is called the
commutator of the two operators.
The case when the commutator of two operators vanishes, is
when the two operations are really independent of each other.
For example small linear motion in $x$ direction is
completely independent of small linear motion in $y$
direction. So they can be taken up in any order, giving the
same result. This fact is expressed by the equation
\begin{equation}
P_x P_y - P_y P_x = 0
\ee
In the Quantum Theory, formulated in terms of space-time
dependent fields, there is a different kind of
``commutator". It was known since the 1930's that to obtain
a consistent Quantum theory of particles of spin $1/2$, one
must require an {\it anti-commutation} relation. The independence
of quantum field operator at far away points $x$ and $y$
has to be expressed by
\begin{equation}
\psi(x)\psi(y) + \psi(y)\psi(x) = 0\qquad\qquad {\rm(Fermionic)}
\ee
What is unusual about this relation is the plus sign where
minus should be; and this indeed expresses independence.
The correctness of this rule is amply borne out by the
Pauli Exclusion Principle whereby two spin $1/2$ particles
can never occupy the same state.
The other kind of particles, those with integer spin and
called Bosons, obey a more familiar algebra, where
independence is expressed by
\begin{equation}
\phi(x)\phi(y) - \phi(y)\phi(x) = 0 \qquad\qquad {\rm(Bosonic)}
\ee
These algebraic operations were however considered special
to the Quantum Fields representing real particles, not to
be confused with operators representing symmetry
operations. The breakthrough against the No-go theorem lay
in realizing that perhaps one could allow ``fermionic"
algebra even between symmetry operations. Thus consider
a linear displacement along two directions which are
independent, and require
\begin{equation}
\theta_1\theta_2 + \theta_2 \theta_1 = 0
\ee
Here $1$ and $2$ are some ``directions" whose meaning
is yet to be clarified, $\theta$ the corresponding
operators. In what way this can be visualized and in
what sense this is an independence are questions not
easy answer. For the moment we take symbols and their
algebra as guides and check for internal consistency of
various operations. Miraculously, it turned out that one
could indeed expand the algebra of the Special Relativistic
generators in this way, provided all the new generators
were fermionic rather than bosonic. This was a possibility
not considered by the authors of the No-go theorem.
\subsection{Superspace}
Here we elaborate a little on the technical idea of
Supersymmetry. Supersymmetry was first formulated as a set of
operations on Quantum Fields. An interpretation closer to that for
usual Special Relativistic symmetries was later formulated, pioneered
among others by Abdus Salam. To every of the four dimensions $(t, x, y, z)$
there corresponds a superspace dimension, and these are labeled
as $(\theta^1, {\bar \theta}^1, \theta^2, {\bar
\theta}^2)$. They are supposed to obey anticommuting algebra.
The mathematics of classical (non-Quantum) variables of this
kind was known to the mathematicians as Grassmann algebra.
Just as rotations led to a mixing of the axes, a
supersymmetric translation leads to mixing of ordinary and
superspace axes. To give an example, if $\theta^1$ is
shifted to $\theta^1 + \alpha$ then the $x$ coordinate shifts
as
\begin{equation}
x \longrightarrow x - i\alpha {\bar \theta}^2
\ee
This does not mean anything to us according to usual
intuition. But this is how things proceed in gleaning
secrets of the microworld. From abstract operations on
fields, we proceeded to further compatibility with usual
space time picture; the metaphysical principle (2) of section 3.
Perhaps future knowledge of new
phenomena will help us visualize these operations better.
\subsection{Predictions and extensions}
There are two main results that follow from assuming that
there is supersymmetry in nature. The first is that for every
fermion of given mass there is a boson of identical mass
and vice versa. This means that corresponding to the observed
photon, there must exist a spin half particle which has been
named photino. Similarly corresponding to the electron,
there must exist a spin zero particle which has been named
selectron (abbreviating `scalar electron'). This nomenclature
pattern is followed for all the hypothetical supersymmetric
partners of known particles.
The problem is, we don't have a single known pair of species
which may be considered superpartners of each other. It is worth
recalling that the original motivation for searching for
supersymmetry was to identify the almost massless neutrino
as the superpartner of the photon. This however cannot be
true because other quantum numbers as required by the symmetry
principle do not match.
The second and very powerful implication of supersymmetry is that
it subsumes the usual gauge symmetry principle and predicts
{\it all} the possible forms of interactions between the particles.
This is a very desirable and attractive. This was the main benefit
of pursuing symmetry principles. They should help us to guess the
form of the interaction. Since we do not see any superpartners
yet, the confirmation or otherwise of this prediction is in the
future.
There are many other attractive features of supersymmetry from
the theoretical point of view but are more technical.
And the simplest predictions seem to be unviable.
Does this mean supersymmetry is of no use? The experience
of searching for gauge symmetry for the weak interactions tells us
that we should keep the possibility open that this symmetry too
is not realized in nature in its simple uncomplicated version,
but perhaps exists in a broken form.
The problem of broken supersymmetry is technically more involved
than breaking of the known gauge symmetries. In some sense this is
because the principle is really very strong. It is difficult
to understand how the symmetry breaks. Developing an understanding
of that is itself a theoretical challenge.
Supersymmetry and its possible manifestations constitutes a
subject of extensive scientific investigation
at present. There are several hypothetical models with mechanisms
for breaking supersymmetry and several giant accelerators being constructed
to check these models. In addition, as true with all Elementary
Particle Physics, these models ought also to have left their
imprints in the early Universe. Efforts are also therefore on to
validate or invalidate some of these models based on cosmological
observations being carried out today.
\section{Philosophical positions}
The esthetic appeal of the symmetry paradigm lies in the
elegance of the mathematical structure.
It seems to generalize the ordinary notion of the freedom to
choose the frame of reference. (See sec. \ref{sec:ggesym}).
It also permits mixing or ``rotating" of distinct particle
species into each other, thus entailing economy in the number
of particle types or species. On the utilitarian side, the unification
achieved requires fewer coupling constants, since several
are dictated to be identical and others have to be simple
multiples of a basic value. This success however has not been
unqualified. In the Standard Model of elementary particles
for example, although the advertised benefits are
present, and the coupling constants are fewer, there do remain
a large number of unknown parameters. These arise primarily in
the form of unknown masses of fermionic and scalar species.
Supersymmetry has merited so much attention due to its elegance.
But it does require introducing a large number of new species
or types of particles. Since many of these are fermionic and
scalar, their masses again require a large number of unknown
parameters. The whole picture is further complicated by the need
to have the symmetry broken. The mechanisms that explain this
breakdown have to rely on more unknown physics, thus invoking
whole new unknown sectors of the theory. Some of the new particles
are supposed to be unobservable by themselves
and their influences on the observable world are only through
the fact that supersymmetry appears broken. There is supposed
be very little additional observable evidence about them even
in principle.
How do we view the situation from outside Physics? I submit that
there are two possibilities. One is that Supersymmetry is indeed
a deep new principle. The other is that it is an expedient
necessary to tide us over till further experiments provide more
clues. The third uninteresting possibility of course remains,
viz., it shares the fate of several other profound speculations
about nature, beautiful but irrelevant.
\subsection{A principle ...}
There are several technical reasons advanced to support
why it must indeed be a fundamental principle of nature. For example
it is meant to rationalize some of the mystery surrounding the
Electroweak symmetry breaking. We can not enter into this
discussion. In the spirit of staying close to fundamental facts,
we may yet advance a reason for the same position along the following
lines.
There is no known classical analogue to fermions. For
several decades in early Quantum Mechanics they were treated
with awe and mystery. Their wavefunction can distinguish between
$360^{\circ}$ and $720^{\circ}$ rotations. Pauli referred to this
as ``non-classical two-valuedness". However, later developments
have required and guided parallel treatments for bosons and fermions.
In the path integral approach, fermions could be
elegantly included by inventing rules for integration over
fermionic variables. But more importantly, a set of simple consistent
rules are suggested by the formalism itself. This is the first
time that one treats {\it classical} fermionic variables with
impunity and gets the required answers.
This begins to suggest
that the bias towards bosons as more natural is perhaps purely
classical. As Dirac\cite{dirac} emphasizes, the early Quantum Mechanics
was developed only for those systems that have a classical analogue.
It was not possible to ``quantize" other systems. But such may
nevertheless exist. Over the decades
that have elapsed, the only other kind of system that seems to
require quantization is the fermionic one. (Modulo phenomena we
still have no reasonable explanation for). Thus the microworld
dictates that we expand our classical notions to include the Grassmann
numbers as well. The notion of Superspace brings further parity
between bosonic and fermionic dimensions. In fact if the fermionic
dimensions did not exist, bosons would still get away with a special
status. What more elegant framework could we have for understanding
these new dimensions than to require that they are partners in a
very rigorous sense to the bosonic dimensions of common experience.
\subsection{... or an expedient?}
But if the symmetry is so fundamental, why do we not see it in its
pure form? Should the fact that the ideal principle has already been
disproved make us abandon the search for supersymmetry?
This brings us to our second, an inelegant but more pragmatic view.
A usual argument for assuming supersymmetry is as follows.
The newer, higher energy accelerators have to be built with special
purposes in mind, i.e. to search specific energy ranges, and to
detect particles with expected decay products.
We need guiding principles to channel our searches and supersymmetry
seems to be only elegant guiding principle aside from gauge
symmetry. But we shall go a little bit beyond this position.
Suppose that supersymmetry is indeed discovered. It may manifest
itself in an ugly and highly disproportionate form. Would it be still
worth discovering? The answer is an affirmative. The caveat is
that it may not be due to the pristine principles we advanced.
But supersymmetry may yet act as an ``organizing principle".
By this we mean a rule such as the valence of elements. This
concept allows us to understand the possible compounds the element
can form. Empirical study then also reveals for the same element
several possible valence states. But once discovered in one context,
that valence state can be sought for in other contexts as well.
Valence is not any kind of fundamental principle. But it does
put strong restrictions on the kind of molecules that can exist.
An ugly manifestation of supersymmetry would be worth having for
the same reason. It may still place a strong restriction on
the kind of fundamental particles that can exist and the qualitative
nature of their interactions. Valence of course is vindicated by
centuries of further developments which make it a direct descendant
of indeed a deep and beautiful principle.
Perhaps the same is true of supersymmetry.
\section{Conclusion}
We argued that the need to understand the microworld has demanded
greater ingenuity from the theoretical physicist in the twentieth
century. The response to this challenge has been in the form of
educated metaphysical principles evolved by the pioneers of the
subject. A prime one of these in retrospect, has been the principle
of symmetry with specific mathematical connotations.
Its great bounty has been the ability to guess at a whole theory
starting from very scarce evidence.
In particular, gauge symmetry has shaped our understanding of
the fundamental forces of nature, beginning with General Theory
of Relativity. Supersymmetry is another such principle, proposed
in advance of any empirical data but with many compulsive reasons
for its correctness. However, even if discovered, the symmetry
would have to be found in a badly broken form. This seems to detract
from its original appeal. We have presented a very general reason
why we may expect the symmetry to be fundamental. Yet there is an
alternative possibility that it may be an organizing principle
similar to valence. In either case it is worth stepping up the efforts
to verify its presence or otherwise in nature.
|
\section{FORMALISM}
To proceed quantitatively, we need to employ a model which is compatible
with the deep inelastic scattering data on a free nucleon and then we
impose the restrictions due to nuclear effects for the bound nucleons.The
picture that comes in mind is the so-called {\it{Valon}} model of
R.C.Hwa \cite{9}. In
this model the nucleon is considered as a system of bound constituent
quarks, which themselves have structure. The bound state problem is a
nonperturbative effect which is taken into account in the distribution of
constituent quarks. The structure of constituent quark is produced
perturbatively and is free of bound state problem. In such a picture the
structure function of a free nucleon is the convolution of constituent quark
distribution in a nucleon and the structure function of the constituent
quark itself:
\begin{equation}
F_2^N(x,Q^2)=2e^{2}_{U}\int dy G_{\frac{U}{N}}(y) {\it{f}}^{c}(x,Q^2) +e^{2}_{D}
\int dy G_{\frac{D}{N}}(y) {\it{f}}^{c}(x,Q^2)
\end{equation}
where $G_{\frac {U}{N}}(y)$ and $G_{\frac {D}{N}}(y)$ are the distribution
of $U$ and $D$-type constituents in a
nucleon and ${\it{f_2^C}}(z, Q^2)$ is the structure function of the
constituent calculated in QCD. At sufficiently high $Q^2$, ${\it{f_2^C}}(z,Q^2)$
can be described accurately in the Leading-Order results in QCD. They are also
calculated in the Next-to-Leading Order[8]; but here we will restrict ourselves
to the leading order, because, we are mainly interested in the application of the
model and it describes the nuclear structure functions sufficiently accurately.
The moments of ${\it{f_2^C}}(z,Q^2)$ can be expressed in terms of the evolution
parameter:
\begin{equation}
s=\it{ln}\frac{\it{ln}\frac{Q^{2}}{\Lambda^{2}}}{\it{ln}\frac{Q_{0}^{2}}
{\Lambda^{2}}}
\end{equation}
where $Q_{0} =0.233 {\frac{GeV}{c}}^2$ and $\Lambda=0.345 {\frac{GeV}{c}}^2$ are scale parameters determined from the data.
The moments of the singlet and non-singlet constituent quark structure function
in the leading order solution of the renormalization group equation are given
as[9]:
\begin{equation}
\begin{array}{c}
M^{NS}(n,Q^2)=\exp (-d_{NS}\ s), \\
M^S(n,Q^2)=\frac 12(1+\rho )\exp (-d_{+}\ s)+\frac 12(1-\rho )\exp (-d_{-}\
s),
\end{array}
\end{equation}
The anomalous dimensions, $d'^{s}$, and other associated parameters are :
\begin{equation}
\begin{array}{c}
\rho =(d_{NS}-d_{gg})/\Delta , \\
\Delta =d_{+}-d_{-}=[(d_{NS}-d_{gg})^2+4d_{gQ}d_{Qg}]^{1/2}, \\
d_{NS}=\frac 1{3\pi b}[1-\frac 2{n(n+1)}+4\sum_{2}^{n}\frac 1j],\\
d_{gQ}=\frac{-2}{3\pi b}\frac{2+n+n^2}{n(n^2-1)},\\
d_{Qg}=\frac{-f}{2\pi b}\frac{2+n+n^2}{n(n+1)(n+2)}, \\
d_{gg}=\frac{-3}{\pi b}[-\frac 1{12}+\frac 1{n(n-1)}+\frac 1{(n+1)(n+2)}-\frac
f{18}-\sum _{2}^{n}\frac 1j], \\
d_{\pm }=\frac
12[d_{NS}+d_{gg}\pm \Delta ], \\
b=(33-2f)/12\pi ,
\end{array}
\end{equation}
Since these moments, $M(n,s)$, are known, using inverse Mellin transformation
technique, we can obtain the distributions of various components in the
constituent quark; namely, valence, sea quarks, and gluon distributions,
$P^{v(sea, g)}$. They are as follows:
\begin{equation}
\begin{array}{c}
P^{v(s)g}(z,Q^2)=\frac 1{2\pi i}\int_cdnz^{-n+1}M_{v(s)g}(n,s), \\
M_v(n,s)=M^{NS}(n,s), \\
M_s(n,s)=(2f)^{-1}(M^S(n,s)-M^{NS}(n,s)), \\
M_g(n,s)=M_{gQ}(n,s),
\end{array}
\end{equation}
and $M_{gQ}(n,s)$ is the quark-to-gluon evolution function given by:
\begin{equation}
M_{gQ}(n,S)=\Delta ^{-1}d_{gQ}[\exp (-d_{+}s)-\exp (-d_{-}s)],
\end{equation}
To account for the $SU(2)$ asymmetry of the nucleon sea which is evident
form the violation of the Gottfried sum rule, we follow the same procedure
as in\cite{7} and replace the distribution of d-quark in the sea by:
\begin{equation}
P_{d/C}(z,Q^2)=\frac 1{(1-x)}P_{u/C}(z,Q^2),
\end{equation}
The calculation of the moments given in eq.(4) is simple. Instead of
exhibiting the moment distribution, we present the results in parametric
form. That is, for every $s$ we fit the moments by a form give below:
\begin{equation}
\begin{array}{c}
P_{C}^v(z,Q^2)=a_v(Q^2)\ z^{b_v(Q^2)}\ (1-z)^{c_v(Q^2)}, \\
P_{C}^{s}(z,Q^2)=\sum_{1}^{2}a_{s_i}(Q^2) (1-z)^{b_{s_i}(Q^2)}, \\
P_{C}^{g}(z,Q^2)=\sum_{1}^{2}a_{g_i}(Q^2) (1-z)^{b_{g_i}(Q^2)}
\end{array}
\end{equation}
for the valence, sea quarks, and gluon respectively. The dependence of the
coefficients $a_j$, $b_j$ and $c_j$ on $Q^{2}$, or rather on $s$ is given in the
appendix. To complete the calculation of $F_{2}^{N}$, we also need to specify
constituent quark distributions, $G_{\frac{U}{N}}$ and $G_{\frac{D}{N}}$.
The simplest approach to evaluate these distributions is to write down the inclusive
momentum distribution of the constituent quark in a free nucleon. We will limit
ourselves to proton here. these inclusive distributions can be written as:
\begin{equation}
G_{UUD}(y_1,y_2,y_3)=\alpha \ y_1^a\ y_2^a\ y_3^b\ \delta (y_1+y_2+y_3-1)
\end{equation}
where $a=0.65$ and $b=0.35$ are two free parameters and $y_1, y_2$ refer
to $U$-type and $y_3$ refers to the $D$-type constituents. By double
integration
over the unspecified variable we get the exclusive distributions:
\begin{equation}
\begin{array}{c}
G_{\frac UP}(y)=[B(a+1,a+b+2)]^{-1}\ y^a\ (1-y)^{a+b+1} \\
G_{\frac DP}(y)=[B(b+1,2a+2)]^{-1}y^b\ (1-y)^{2a+1}
\end{array}
\end{equation}
where $B(x,y)$ is the Euler Beta function. The above distributions satisfy
also the following normalization conditions:
\begin{equation}
\int_0^1dy\ G_{\frac UP}(y)=\int_0^1dy\ G_{\frac DP}(y)=1
\end{equation}
In figure 1, we present the result for $F_2^P(x,Q^2)$ for several values
of $x$ and $Q^2$, its agreement with the experimental data in a wide range
of kinematics is rather good.\\
Having determined the nucleon structure function, next we extend the same
procedure to the nuclear medium. As we mentioned earlier, in the nuclear
medium the constituent quark distribution has to be modified,
but its structure function, or rather parton distribution in a constituent
will not be affected by the nuclear environment. The modification to
$G_{\frac CP}(y)$ is achieved by introduction of a distortion factor,
$\delta$. Hence, we write the inclusive momentum distribution as follows :
\begin{equation}
G_{UUD}^{\prime }(y)=\alpha (y_1y_2)^{a+\delta}y_3^{b+\delta}\delta
(y_1+y_2+y_3-1)
\end{equation}
which leads to :
\begin{equation}
\begin{array}{c}
G_{\frac UP}^{\prime }(y)=\alpha \ y^{a+\delta}\ (1-y)^{a+b+2\delta+1} \\
G_{\frac DP}^{\prime }(y)=\alpha ^{\prime }\ y^{b+\delta}\
(1-y)^{2a+2\delta+1}
\end{array}
\end{equation}
with :
\begin{equation}
\begin{array}{c}
\alpha =[Beta(a+\delta+1,a+b+2\delta+2)]^{-1} \\
\alpha ^{\prime }=[Beta(b+\delta+1,2a+2\delta+2)]^{-1}
\end{array}
\end{equation}
we note that $\delta$ is a function of atomic number, $A$. In Ref.\cite{8} a
probabilistic argument is used to find $\delta$. Here we do generalize
the same result as follows :
\begin{equation}
\delta=\frac{0.0104-0.014 \ln A+ 0.014(\ln A)^2}{%
1-0.83 \ln A+0.295(\ln A)^2-0.024(\ln A)^3}
\end{equation}
So far we have determined the constituent quark distribution in a bound
nucleon. We need to find its distribution in a nucleus. For that matter,
however, one needs also to know the nucleon distribution in the nucleus.
we can use the Fermi distribution and arrive at the following result :
\begin{equation}
\phi _N^A(z)=\int_{k_F}^{\left| k\right| }d^3k\ \rho (k)\ \delta (z-1-\frac{%
k^{+}}{M_n})
\end{equation}
where $k^{+}=k^0+k^z$ and $k^0=M_{N}$ are the light cone coordinates
and $\rho (k)=\frac 3{4\pi k_F^3}\theta (k_F-\left| k\right| )$ is the nucleon
density in the nucleus. So,
straightforward integration would yield the nucleon distribution in a
nucleus. However, this also means that the bound nucleon is nearly on shell
and therefore, the nuclear binding effect is neglected. To be more
realistic, and include this effect, we replace $z-1$ by $z-\eta _A$ where $%
\eta _A=1-\frac{B_A}{M_N^{*}}$ and $M_N^{*}=M_N-B_A$ is the effective mass
of the nucleon. With these fine tunings we arrive at the following function
for the nucleon distribution in a nucleus :
\begin{equation}
\begin{array}{c}
\phi _N^A(z)=\frac 34
\frac{M_N^3}{k_F^3}(\frac{k_F^2}{M_N^2}-(z-\eta _A)^2)\ ;\eta _A-\frac{k_F^A
}{M_N}\langle \ z\ \langle \eta _A+\frac{k_F^A}{M_N} \\
\phi _N^A(z)=0\ ;otherwise
\end{array}
\end{equation}
The values of the Fermi momentum, $k_F^A$, and the binding energy, $B_A$,
are given by the requirements of nuclear physics. Going one step further, we
carry the Fermi motion to the constituent quark level. This is done
easily by renormalization of the distribution function give in Eq.(14) this
leads to the following constituent quark distribution per proton in a nucleus :
\begin{equation}
G_{\frac{U}{P}}^{A}(y)\approx (\frac 1\eta )G_{\frac UP}^{\prime }(\frac y\eta
)+\frac {1}{10}\lambda ^2(
\frac{d^2}{dz^2})(\frac 1z)G_{\frac UP}^{\prime }(\frac yz)\mid _{z=\eta }
\end{equation}
\begin{equation}
G_{\frac DP}^A(y)\approx (\frac 1\eta )G_{\frac DP}^{\prime }(\frac y\eta
)+\frac 1{10}\lambda ^2(\frac{d^2}{dz^2})(\frac 1z)G_{\frac DP}^{\prime
}(\frac yz)\mid _{z=\eta }
\end{equation}
where $\lambda =\frac{k_F}{M_N^{*}}$ .
Having collected all the ingredients, now we are
in a position to evaluate the nuclear structure function ratios,
$R=\frac{F_2^A(x,Q^2)}{F_2^D(x,Q^2)}$. Notice that all of our discussions about the constituent quark distribution are pertinent to proton. Finally,
taking the neutron excess in the nucleus into consideration, we get :
\begin{equation}
F_{2}^{A}(x, Q^2)=(\frac{1}{A})\{F_{2A}(x, Q^2)-\frac{1}{2}(N-Z)[F_2^n(x,Q^2)-F_2^p(x,Q^2)]\}
\end{equation}
where $F_2^p(x,Q^2)$ and $F_2^n(x,Q^2)$ are the free proton and free neutron
structure functions, respectively, and $F_{2A}(x,Q^2)$ is obtained using
Eq.(1). The results of the model is presented in figure $2$ for a variety of
nuclei ranging from $A=4$ to $A=204$. As it is apparent from Fig.(2), we
see that the model calculation is reasonable only down to $x\geq 10^{-1}$.
For smaller values of $x$, $x\leq 10^{-1}$ an extra ingredient is
required. An obvious and natural place to look for the dynamics of small $x$
would be the shadowing effects, where the growth of parton densities are
suppressed due to recombination and annihilation of the partons. There are
some attempts in the literature to calculate the shadowing corrections,
pioneered by Qiu\cite{12}. However we found that such a
parameterization is inadequate in describing the data. Following Ref.[12] we
modify the shadowing correction of Qiu in two ways: (1) we take the
general idea of vector meson dominance which describes low $Q^2$ virtual
photon interaction, to model the approximate scaling of shadowing by pomeron
exchange. It turns out that the experimental data on $F_{2p}(x,Q^2)$ up to $%
x<0.05$ can be well described by a sub-asymptotic pomeron with the effective
anomalous dimension going down logarithmically with $x$\cite{13}.(2) The magnitude of the shadowing
effect in such models is sensitive to the value of vector meson-nucleon
cross section. This sensitivity is explored by G. Show \cite{14} and it is
described by a quantity :
\begin{equation}
\frac{A_{eff}}A=\frac{\sigma _{\gamma \ A}}{A\sigma _{\gamma \ N}}
\end{equation}
This ration is particularly manifested for heavy nuclei and reflects the
ratio of nuclear radius $R$ to the mean free path, $\lambda$, of the
vector meson inside the nucleus; which also depends on the nuclear density .
That is, the shadowing effect at small $x$ values increases with increasing
nuclear radius and density. Utilizing these modifications to the original
work of Qiu [12], finally we present the following modification of parton
distributions in a nucleus :
\begin{equation}
\begin{array}{c}
P_{C/A}^s(z,Q^2)=R_s(x,Q^2,A_{eff})P_{C/N}^s(z,Q^2), \\
P_{C/A}^v(z,Q^2)=P_{C/N}^v(z,Q^2),
\end{array}
\end{equation}
where superscript $s(v)$ stands for sea (valence) components and
$R_s(x,Q^2,A_{eff})$ is the shadowing factor affecting parton distributions
of the constituent quark. These distributions in the nuclear medium are
compared to the corresponding distribution in a free nucleon. The
parameterized form of this factor is :
\begin{equation}
R_{s}(x, Q^2, A_{eff})=R_{s}(Q^{2},A_{eff})(0.94+0.076A_{eff}^{0.5})x^{0.5}ln(x))
\end{equation}
And $A_{eff}$ is given by:
\begin{equation}
A_{eff}=-10.97-0.704A+.0042A^2-0.00019A^{2.5}+8.63A^{0.5}
\end{equation}
In equation (24) $R_s(Q^2,A_{eff})$ is evaluated by the $n^{th}$-moment
equations for the parton distribution in the sea of a constituent quark\cite{15}:
\begin{equation}
\begin{array}{c}
<F_{c/N}^s(Q^2)>_n=<F_{c/N}^s(Q_c^2)>_nK_{qq}^n(Q^2)+<F_{c/N}^g(Q_c^2)>_nK_{qg}^n(Q^2)+ \\
<F_{c/N}^v(Q_c^2)>_nK_{NS}^n(Q^2)
\end{array}
\end{equation}
where $<F(Q^2)>_n=\int_0^1z^{n-2}P(z,Q^2)dz$. Obviously, nuclear shadowing
do not affect the distribution of valence quarks in the constituent quarks.
If the momentum loss owing to the recombination process of the shadowed sea
quarks is negligible, the momenta carried by each parton component would be
conserved separately and approximately,
\begin{equation}
\begin{array}{c}
<F_{c/N}^s(Q^2)>_2\cong <F_{c/A}^s(Q^2)>_2 \\
<F_{c/N}^g(Q^2)>_2\cong <F_{c/A}^g(Q^2)>_2
\end{array}
\end{equation}
and the following inequalities will be satisfied for $Q^2> Q_{0}^{2}$[15] :
\begin{equation}
<F_{c/N(A)}^{s(g)}(Q^2)>_2\gg <F_{c/N(A)}^{s(g)}(Q^2)>_3
\end{equation}
As a result, following Re.\cite{12} we acquire:
\begin{equation}
R_sea(Q_0^2,A_{eff})=1-k_s(Q_0^2)(A_{eff}^{1/3}-1)=\lim_{x\rightarrow 0}
\frac{F_A^s(x,Q^2)}{F_N^s(x,Q_0^2)}\approx \frac{<F_{C/N}^s(Q_0^2)>_3}{%
<F_{C/A}^s(Q_0^2)>_3}
\end{equation}
\begin{equation}
R_g(Q_0^2,A_{eff})=1-k_g(Q_0^2)(A_{eff}^{1/3}-1)=\lim_{x\rightarrow 0}
\frac{F_A^g(x,Q_0^2)}{F_N^g(x,Q_0^2)}\approx \frac{<F_{C/N}^g(Q_0^2)>_3}{%
<F_{C/A}^g(Q_0^2)>_3}
\end{equation}
where $K_{s(q)}$ is the shadowing strength and and depends on the starting scale
$Q_{0}^{2}$ for the evolution to set in.
It is easy to verify that the primitive shadowing for gluon in a constituent
quark is weaker than that for the sea partons[12] at such scale.The form of
shadowing correction is in terms of evolution kernels is given as:
\begin{equation}
\begin{array}{c}
R_s(Q^2,A_{eff})=
R_s(Q_0^2,A_{eff})\frac{%
<F_{c/N}^s(Q_0^2)>_3K_{qq}^3(Q^2)+<F_{C/N}^g(Q_0^2)>_3K_{qg}^3(Q^2)+<F_{C/N}^v(Q_0^2)>_3K_{NS}^3(Q^2)
}{<F_{C/N}^s(Q_0^2)>_3K_{qq}^3(Q^2)+\lambda
<F_{C/N}^g(Q_0^2)>_3K_{qg}^3(Q^2)+R_s(Q_0^2)<F_{C/N}^v(Q_0^2)>_3K_{NS}^3(Q^2)%
}
\end{array}
\end{equation}
where $K_{ab}^{n}$ are the evolution kernels. Defining $s=\it{ln}L$, these
kernel functions are[15]:
\begin{equation}
\begin{array}{c}
K_{qq}^n(Q^2)=\alpha _nL^{-a_n^{-}}+(1-\alpha _n)L^{-a_n^{+}} \\
K_{qg}^n=\beta _n(L^{-a_n^{-}}-L^{-a_n^{+}}) \\
K_{NS}^n(Q^2)=\left[ \alpha _nL^{-a_n^{-}}+(1-\alpha
_n)L^{-a_n^{+}}-L^{-a_{NS}^n}\right] \\
a_{NS}=
\frac{\gamma _{qq}}{\beta _{\circ }}\ ,a_{\pm }^n=\frac{\gamma ^{\pm n}}{%
\beta _{\circ }} \\ \alpha _n=
\frac{\gamma _{qq}-\gamma ^{+}}{\gamma ^{-}-\gamma ^{+}} \\ \beta _n=\frac{%
\gamma _{qq}}{\gamma ^{-}-\gamma ^{+}}
\end{array}
\end{equation}
\section {CONCLUSION}
We have calculated the nucleus structure function ratio in the context of
constituent quark picture, utilizing the essence of the so called valon model.
As we can see from the figures; the model describes the data rather well for
a wide range of $A$. It appears that the $Q^2$ dependence of the ratio of
structure functions is week. The shadowing effect is quite large for small
value of $x$, $(x\langle 0.01)$ and dies away very rapidly as $x$ increases.
Also, the shadowing correction becomes more pronounced with increasing
atomic number $A$. We further note that in this model, which is based on
the constituent quark structure, we have not included explicitly the
anti-shadowing effects notwithstanding that there are some debates on the
subject particularly pertinent to the middle range of $x$. we simply did
not need it to describe the data. Our anticipation is that such an effect
would not be very large .
\section{APPENDIX}
In this appendix we provide numerical values for the coefficients given in
equation ( ). These are the results of our $Q^2$ dependence parameterization
of parton distributions in a constituent quark of proton. They are calculated
in the leading order.
\begin{equation}
\begin{array}{c}
a_{s1}=-0.119+0.0296\exp (s/0.66) \\
b_{s1}=5.510+8.50\exp (s/0.95) \\
a_{s2}=-0.080+0.066\exp (s/3.12) \\
b_{s2}=-1.194+2.303\exp (s/2.76) \\
a_{g1}=-2.490+1.027\exp (s/0.69) \\
b_{g1}=5.920+6.440\exp (s/0.80) \\
a_{g2}=.174+.345s^{0.09} \\
b_{g2}=-1.407+1.763\exp [s/2.42]; \\
a_v=-0.300+1.254s-0.368s^2+0.034s^3 \\
b_v=0.196+1.286\exp [-s/3.14] \\
c_v=-0.844+.451s^{.17}
\end{array}
\end{equation}
\newpage
\begin{figure}
\label{fig1}
\caption{ $F^{p}_{2}$ as a function of $x$ at several $Q^{2}$.}
\end{figure}
\begin{figure}
\label{fig2}
\caption{ The ratio $R=\frac{F_2^{A_1}(x,Q^2)}{F_2^{A_2}(x,Q^2)}$ for
a variety of nuclei at different $Q^2$. Dashed-dotted line represents
the results without the shadowing effects. Others include the shadowing
corrections.}
\end{figure}
\newpage
|
\section{Introduction}
\object{AFGL 2290} (OH~39.7+1.5, IRAS~18560+0638, V1366 Aql) belongs
to the group of type II OH/IR stars, which can be defined as infrared
point sources with a maximum of the spectral energy distribution (SED)
around $6 - 10\,{\rm\mu m}$, with the $9.7\,{\rm\mu m}$ silicate
band in absorption, and with OH maser emission in the 1612 MHz line
(Habing \cite{Hab96}). Most of these objects show a long-period
variability in the infrared and the OH maser emission (Engels
\cite{Eng82}; Herman \& Habing \cite{HerHab85}), although also a small
fraction either varies irregularly with small amplitude or does not
vary at all. OH/IR stars are surrounded by massive circumstellar
envelopes composed of gas and small solid particles (dust, grains).
These circumstellar dust shells (CDS) are produced by the ejection of
matter at large rates ($\dot{M}>10^{-7}$ ${\rm M_{\odot}\,yr^{-1}}$)
and low velocities ($\sim 15\,{\rm kms^{-1}}$), and in some cases they
totally obscur the underlying star. Based on the luminosities ($\sim
10^{4}\,\rm L_{\odot}$), the periods (500d to 3000d) and bolometric
amplitudes ($\sim 1$~mag), the kinematical properties and galactic
distribution, the majority of OH/IR stars are highly evolved low-- and
intermediate--mass stars populating the asymptotic giant branch (AGB)
(Habing \cite{Hab96}). They extend the sequence of optical Mira
variables to longer periods, larger optical depths and higher mass
loss rates (Engels et al.\ \cite{EKSS83}; Habing \cite{Hab90}; Lepine
et al.\ \cite{LOE95}).
The improvements of the observational techniques, especially at
infrared wavelengths, and the elaboration of increasingly
sophisticated theoretical models have provided a wealth of new
information on the structure, the dynamics, and the evolution of the
atmospheres and circumstellar shells of AGB stars, although many
details still remain to be clarified (see the review by Habing
\cite{Hab96}). A general picture has become widely accepted in which
both the large amplitude pulsations and the acceleration by radiation
pressure on dust contribute to the mass loss phenomenon for AGB stars.
From observations, correlations are found between the period and the
infrared excess (indicating the mass loss rate)
(DeGioia--Eastwood et al.\ \cite{DHGG81}; Jura \cite{Jur86}), and
between the period and the terminal outflow velocity (Heske
\cite{Hes90}). On the theoretical side, hydrodynamical models showed
that due to the passage of shocks generated by the stellar pulsation
the atmosphere is highly extended, thus enabling dust formation and
the subsequent acceleration of the matter (Wood \cite{Wood79}; Bowen
\cite{Bow88}). The inclusion of a detailed treatment of dust formation
revealed a complex interaction between pulsation and dust formation,
which results e.g.\ in a layered dust distribution and affects the
derived optical appearance (Fleischer et al.\ \cite{fgs92},
\cite{fgs95}; Winters et al.\ \cite{wfgs94}, \cite{wfgs95}).
Until now most interpretations of observations as well as most
theoretical models are based on the assumption of a spherically
symmetric dust shell, often motivated by the circularity of the OH
maser maps. However, observations show that some objects have
substantial deviations from spherical symmetry (e.g.\ Dyck et al.\
\cite{DZLB84}; Kastner \& Weintraub \cite{KaWe94}; Weigelt et al.\
\cite{WBBFOW98}). This suggests that the asymmetries observed in many
post--AGB objects and planetary nebulae (cf.\ Iben \cite{Ib95}) may
already start to develop during the preceding AGB phase, which
provides new challenges for the modeling of the mechanisms and processes
determining the structure of the dust shells around AGB stars.
High spatial resolution observations can yield direct information on
important properties of the dust shells around AGB stars, such as the
dimensions and geometry of the shell. Therefore, such observations
contribute additional strong constraints for the modeling of these
circumstellar environments, which supplement the information from the
spectral energy distribution.
Measurements of the visibility at near-IR wavelengths, for example,
can be used to determine the radius of the onset of dust formation as
well as to constrain the dominant grain size (Groenewegen
\cite{Groe97}). To gain information on details of the spatial
structure, in particular on asymmetries and inhomogeneties of the CDS,
the interferometric imaging with large single--dish telescopes is
especially well suited because one observation provides all spatial
frequencies up to the diffraction limit of the telescope and for all
position angles simultaneously, allowing the reconstruction of true
images of the object.
We have chosen \object{AFGL 2290} for our study because it represents
a typical obscured OH/IR star with a high mass loss rate, whose
location is not too far away from us. The distance to
\object{AFGL 2290} can be determined directly with the phase lag
method (cf.\ Jewell et al.\ \cite{JEWS79}), which gives $D =
0.98\,{\rm kpc}$ (van Langevelde et al.\ \cite{LHS90}). For the
bolometric flux at earth a value of $f_{\rm b} \sim
2.4\,10^{-10}\,{\rm W m^{-2}}$ is derived by van der Veen \& Rugers
(\cite{VeRu89}) from infrared photometry between $1\,{\rm\mu m}$ and
$12\,{\rm\mu m}$ and the IRAS fluxes. At $0.98\,{\rm kpc}$ the
luminosity is $L=7200\,{\rm L_{\odot}}$, which is within the typical
range for an oxygen--rich AGB star. The long period of $P=1424\,{\rm
d}$ determined from the variation of the OH maser (Herman \& Habing
\cite{HerHab85}) and the high mass loss rate suggest, that the star is
in a late phase of its AGB evolution.
So far, Chapman \& Wolstencroft (\cite{ChWo87}) reported the only high
angular--resolution infrared observations of \object{AFGL 2290}. From
1--dimensional slit--scan speckle interferometry with the UKIRT 3.8~m
telescope at $3.8\,{\rm\mu m}$ and $4.8\,{\rm\mu m}$ they derive
1--dimensional visibilities and determine Gaussian FWHM diameters.
Radiative transfer models for the \object{AFGL 2290} dust shell have
been presented by Rowan--Robinson (\cite{RowRob82}), Bedijn
(\cite{Bed87}), Suh (\cite{Suh91}) and recently by Bressan et al.\
(\cite{BGS98}). These models yield dust shell properties within the
typical range of OH/IR stars, e.g.\ a dust mass loss rate of about
$4\,10^{-7}\,{\rm M_{\odot} yr^{-1}}$ (Bedijn \cite{Bed87}; Bressan et
al.\ \cite{BGS98}), or an optical depth at $9.7\,{\rm\mu m}$ of about
10 (Bedijn \cite{Bed87}; Suh \cite{Suh91}). However, none of these
studies includes constraints from high spatial resolution infrared
measurements.
In Sect.~\ref{observations} we present the results of our speckle
masking observations of \object{AFGL 2290}. The approach for the
radiative transfer modeling is described in
Sect.~\ref{radtrf-approach} comprising a short description of the
code, the selection of the photometric data and a discussion of input
parameters for the models. In Sect.~\ref{radtrf-modeling} we present
the results of the radiative transfer modeling starting with the
discussion of a model, which yields a good fit of the observed SED at
all wavelengths but does not reproduce the observed $2.11\,{\rm\mu m}$
visibility. In search of an improved model the changes of the
resulting SED and visibility under variations of the input parameters
are investigated in the following sections. We finish the paper with a
summary of the results and our conclusions in Sect.~\ref{summary}.
\section{Speckle masking observations}
\label{observations}
The \object{AFGL 2290} speckle data presented here were obtained with
the Russian 6~m telescope at the Special Astrophysical Observatory
(SAO) on June 14 and 16, 1998. We recorded a total number of 1200
speckle interferograms of \object{AFGL 2290} (600 on June 14 and 600
on June 16) and 2400 speckle interferograms of the unresolved
reference star HIP~93260 (1200 on each of the two nights) with our
256$\times$256 pixel NICMOS~3 camera through an interference filter
with center wavelength 2.11\,$\mu$m and a bandwidth of 0.192\,$\mu$m.
The exposure time per frame was 100~ms, the pixel size was 30.61~mas
and the field of view $7\farcs8\times7\farcs8$. The $2.11\,{\rm\mu m}$
seeing was about $\sim 1\farcs2$.
A diffraction-limited image of \object{AFGL 2290}
was reconstructed from the speckle data by the speckle masking
bispectrum method (Weigelt \cite{Weig77}; Lohmann et al.\
\cite{LohmWeWi83}; Weigelt \cite{Weig91}). The process includes the
calculation of the average power spectrum and of the average
bi--spectrum and the subtraction of the detector noise terms from
those. The modulus of the object Fourier transform was determined with
the speckle interferometry method (Labeyrie \cite{Lab70}). The Fourier
phase was derived from the bias--compensated average bispectrum.
\begin{figure}\unitlength 1cm
\resizebox{\hsize}{!}{\includegraphics{H1209F1.eps}}
\caption{
Two--dimensional $2.11\,{\rm\mu m}$ visibility function of
\object{AFGL 2290} derived from the speckle interferograms.
The contour levels are plotted from 20\% to 80\% of the peak
value in steps of 10\%.}
\label{Powerspectrum}
\end{figure}
\begin{figure}\unitlength 1cm
\resizebox{\hsize}{!}{\includegraphics{H1209F2.eps}}
\caption{
Azimuthally averaged $2.11\,{\rm\mu m}$ visibility function of
\object{AFGL 2290} and errorbars.}
\label{Visibility}
\end{figure}
Figures \ref{Powerspectrum} and \ref{Visibility} show the visibility
function of \object{AFGL 2290} at $2.11\,{\rm\mu m}$. The azimuthally
averaged visibility decreases steadily to values below $\sim 0.40$ of
the peak visibility at the diffraction cut--off frequency ($13.5\,{\rm
arcsec^{-1}}$). Thus, the circumstellar dust shell is almost totally
resolved, and
the contribution of the unresolved stellar component to the
monochromatic flux at $2.11\,{\rm\mu m}$ must be less than $\sim
40\,\%$, suggesting a rather high optical depth at this wavelength.
In order to derive diameters for the dust shell, the object visibility
function was fitted with an elliptical Gaussian model visibility
function within a range of $1.5\,{\rm arcsec^{-1}}$ up to $7.5\,{\rm
arcsec^{-1}}$.
We obtain a Gaussian fit diameter of 43~mas$\times$51~mas
for \object{AFGL 2290} corresponding to 42~AU$\times$50~AU for an
adopted distance of 0.98~kpc or $5.7\,r_{*}\times6.8\,r_{*}$ for an
adopted distance of 0.98~kpc and an adopted stellar radius of
$r_{*}=7.5\,{\rm mas}$ (cf.\ Sect.~\ref{best-fit-sed}),
respectively.
\begin{figure}\unitlength 1cm
\resizebox{\hsize}{!}{\includegraphics{H1209F3.eps}}
\caption{Diffraction--limited $2.11\,{\rm\mu m}$ speckle masking image
of the \object{AFGL 2290}. North is at the top and east to the left.
The contours level intervals are 0.25~mag. The lowest contour level
is 3.25~mag fainter than the peak intensity.}
\label{object-intensity}
\end{figure}
\begin{figure}\unitlength 1cm
\resizebox{\hsize}{!}{\includegraphics{H1209F4.eps}}
\caption{Azimuthally averaged image of \object{AFGL 2290} (solid line)
and of the unresolved reference star HIP~93260 (dashed line).}
\label{Intensitycuts}
\end{figure}
Figure \ref{object-intensity} shows the reconstructed $2.11\,{\rm\mu
m}$ speckle masking image of \object{AFGL 2290}. The resolution is
75~mas. Figure \ref{Intensitycuts} shows the azimuthally averaged
images derived from the reconstructed 2--dimensional images of
\object{AFGL 2290} and the reference star HIP~93260. In the
2-dimensional \object{AFGL 2290} image a deviation from
spherical symmetry can be recognized. The intensity contours are
elongated in the south--eastern direction along an axis with a position
angle of $130^{\circ}$.
\section{The radiative transfer modeling approach}
\label{radtrf-approach}
\subsection{The radiative transfer code}
\label{radtrf-code}
The radiative transfer calculations are performed with the code DUSTY
developed by Ivezi{\'c} et al.\ (\cite{INE97}), which is publicly
available.
The program solves the radiative transfer problem for a spherically
symmetric dust distribution around a central source of radiation and
takes full advantage of the scaling properties inherent in the
formulation of the problem. The formulation of the radiative transfer
problem, the model assumptions and the scaling properties are
described in detail by Ivezi{\'c} \& Elitzur (\cite{IE97}). Therefore,
we give only a brief discussion here.
The problem under consideration is a spherically symmetric dust
envelope with a dust free inner cavity surrounding a central source
of radiation. This geometry is not restricted to the dust shell of a
single star. It can as well describe a dust envelope around a group
of stars (e.g\ a binary) or even around a galactic nucleus. The radial
dependence of the dust density between the inner and outer boundary
can be chosen arbitrarily.
To arrive at a scale invariant formulation two assumptions are
introduced: i) the grains are in radiative equilibrium with the
radiation field, and ii) the location of the inner boundary $r_1$ of
the dust envelope is controlled by a fixed temperature $T_1$ of the
grains at $r_1$. Due to radiative equilibrium this temperature is
determined by the energy flux at $r_1$, which in turn is controlled
by the energy flux from the central source via the radiative transfer
through the dusty envelope. Then prescribing the dust temperature at
$r_1$ is equivalent to specifying the bolometric flux at the inner
boundary, and the only relevant property of the input radiation is its
spectral shape (Ivezi{\'c} \& Elitzur \cite{IE97}). Similarly, if the
overall optical depth of the dust envelope at some reference
wavelength is prescribed, only dimensionless, normalized distributions
describing the spatial variation of the dust density and the
wavelength dependence of the grain optical properties enter into the
problem.
This formulation of the radiative transfer problem for a dusty
envelope is well suited for model fits of IR observations, because it
minimizes the number of independent model parameters. The input
consists of:
\begin{itemize}
\item the spectral shape of the central source of radiation, i.e.\ the
variation of the normalized monochromatic flux with wavelength,
\item the absorption and scattering efficiencies of the grains,
\item the normalized density distribution of the dust,
\item the radius of the outer boundary in units of the inner boundary,
\item the dust temperature at the inner boundary,
\item the overall optical depth at a reference wavelength.
\end{itemize}
For a given set of parameters, {\small DUSTY} iteratively determines
the radiation field and the dust temperature distribution by solving
an integral equation for the energy density, which is derived from a
formal integration of the radiative transfer equation. For a
prescribed radial grid the numerical integrations of radial functions
are transformed into multiplications with a matrix of weight factors
determined purely by the geometry. Then, the energy density at every
point is determined by matrix inversion, which avoids iterations over
the energy density itself and allows a direct solution of the pure
scattering problem. Typically fewer than 30 grid points are needed to
achieve a relative error of flux conservation of less than 1\%. The
number of points used in angular integrations is 2--3 times the number
of radial grid points, and the build--in wavelength grid has 98 points
in the range from $0.01\,{\rm\mu m}$ to 3.6 cm (see Appendix C in
Ivezi{\'c} \& Elitzur \cite{IE97}).
The distributed version of the code provides a variety of quantities
of interest including the monochromatic fluxes and the spatial
intensity distribution at wavelengths selected by the user, but not
the corresponding visibilities.
Since we want to employ the visibilities obtained from our high
spatial resolution measurements as constraints for the radiative
transfer models, we have supplemented the code with routines for the
calculation of synthetic visibility functions.
\subsection{Selection of photometric data}
\label{photometry}
An important ingredient for the radiative transfer modeling of
circumstellar dust shells around evolved stars is the spectral energy
distribution (SED). Due to the variability of Miras and OH/IR stars,
the SED of such objects ideally has to be determined from coeval
observations covering all wavelengths of interest.
Unfortunately, no such coeval photometric data set for the wavelength
region from $\lambda \approx 1\,{\rm\mu m}$ to $\lambda \ge
20\,{\rm\mu m}$ is available in the literature for \object{AFGL
2290}. Thus, we have to define a `composite' SED, which is derived
from observations made by different authors at different epochs, but
at about the same photometric phase (Griffin \cite{Grif93}).
\begin{table*}
\caption{Infrared photometry of \object{AFGL 2290} ordered by the date
of observation}
\label{IR-obstab}
\begin{tabular}{|r|c|c|c|r|r|r|r|r|r|r|r|r|r|}\hline
No. & Julian Date & ${\rm Phase^{*}}$ & Ref.\ &
\multicolumn{10}{c|}{Wavelengths}\\ \hline
& 244 0000+ & $P=1424\,{\rm d}$ & &
\multicolumn{10}{c|}{[$\,{\rm\mu m}$]}\\ \hline
1 & 1045 & 0.320 $(-3)$ & 1&
& & & 4.2& & & & 11.0& & 19.8 \\
2 & 2295 & 0.198 $(-2)$ & 1&
& & & & & & & 11.0& & 19.8 \\
3 & 2725 & 0.500 $(-2)$ & 2&
& 2.2& 3.6& 5.0&8.4, 8.8& & 10.4 10.6& 11.6& 12.6& \\
4 & 3726 & 0.203 $(-1)$ & 3&
& 2.3& 3.6& 4.9& 8.7& & 10.0& 11.4& 12.6& 19.5 \\
5 & 3972 & 0.376 $(-1)$ & 4&
1.25, 1.65& 2.2& 3.7& 4.8& & & & & & \\
6 & 4082 & 0.453 $(-1)$ & 3&
& 2.3& 3.6& 4.9& 8.7& & 10.0& 11.4& 12.6& 19.5 \\
7 & 4348 & 0.640 $(-1)$ & 3&
& 2.3& 3.6& 4.9& 8.7& & 10.0& 11.4& 12.6& 19.5 \\
8 & 4533 & 0.770 $(-1)$ & 4&
1.25, 1.65& 2.2& 3.7& 4.8& 8.2& 9.6& 10.2& & 12.2&19.6\\
9 & 5146 & 0.200 $(+0)$ & 5&
& & 3.8& 4.8& 8.7& 9.7& 10.5& 11.5& 12.5& 20 \\
10 & 7816 & 0.075 $(+2)$ & 6&
1.63& 2.23& 3.79& & & & & & & \\
11 & 7832 & 0.087 $(+2)$ & 7&
1.26, 1.68& 2.28& 3.80& & & & & & & \\
12 & 8041 & 0.233 $(+2)$ & 8&
1.24, 1.63& 2.19& 3.79& 4.64& & & & & & \\
\hline
\end{tabular}\\
References: 1) Price \& Murdock \cite{PM76},
2) Lebofsky et al.\ \cite{LKRL76},
3) Gehrz et al.\ \cite{GKMHG85},
4) Engels \cite{Eng82},
5) Herman et al.\ \cite{HISH84},
\mbox{6) Noguchi et al.\ \cite{NQWW93}},
7) Xiong et al.\ \cite{XChG94},
8) Nyman et al.\ \cite{NHB93}.\\
${}^{*}$ Numbers in parantheses give the cycle with respect to epoch
JD 244~4860.8.
\end{table*}
From the infrared photometry of \object{AFGL 2290} available in the
literature, we selected those publications which specify the date of
observation and present the fluxes in tabulated form, either in
physical units (e.g.\ Jy) or in magnitudes with given conversion
factors (at least as a reference).
Table \ref{IR-obstab} lists the references, the date and phase of
observation and the wavelengths. The phases were determined from the
period $P = 1424$d and the epoch of maximum, JD = 244 4860.8, which
has been derived from the monitoring of the OH maser emission by
Herman \& Habing (\cite{HerHab85}). Engels et al.\ (\cite{EKSS83})
determined periods of OH/IR stars from infrared observations and
found that the periods and phases are in agreement for objects in
common with the sample Herman \& Habing (\cite{HerHab85}).
It can be seen from the entries in Table \ref{IR-obstab} that
the wavelength range from $\lambda=1.2 \,{\rm\mu m}$ to
$\lambda=20\, {\rm\mu m}$ is only fully covered by observations
around phase 0.2 (see entries 2, 4, 9, 12).The respective fluxes are
shown in Fig.\ \ref{IR-obsfig}.
\begin{figure}\unitlength 1cm
\includegraphics{H1209F5.eps}
\caption{
IR--fluxes of \object{AFGL 2290} observed about phase 0.2 and at
phase 0.77. The data are taken from Price \& Murdock \cite{PM76}
($+$), Herman et al.\ \cite{HISH84} ($\diamond $), Gehrz et al.\
\cite{GKMHG85} ($\triangle $), and Nyman et al.\ \cite{NHB93}
($\sq $).
Also shown are the colour corrected IRAS fluxes adopted from van
der Veen et al.\ \cite{VOHM95} (filled sqares), and the IRAS low
resolution spectra (dashed line). The IRAS data are multiplied by a
factor of 1.95 in order to match the photometric data at $12\,{\rm
\mu m}$. The insert shows mm measurements by Walmsley et al.\
\cite{WCKSFO91} ($\bullet$) and van der Veen et al.\ \cite{VOHM95}
($\bigtriangledown$, $3\,\sigma$ upper limits).}
\label{IR-obsfig}
\end{figure}
The measurements of Herman et al.\ (\cite{HISH84}) and Nyman et al.\
(\cite{NHB93}) match each other quite well at $\lambda = 3.8$ and
$\lambda = 4.8\,{\rm\mu m}$, although the observations are separated
by two periods. The fluxes of Price \& Murdock (\cite{PM76}) and Gehrz
et al.\ (\cite{GKMHG85}) agree with the Herman et al.\ and Nyman et
al.\ data within the errors given by the authors.
To represent the SED of \object{AFGL 2290} we adopt the data of Herman
et al.\ (\cite{HISH84}), Gehrz et al.\ (\cite{GKMHG85}), and Nyman et
al.\ (\cite{NHB93}). The scatter between the different data sets gives
a rough estimate of the uncertainty of the `composite' SED at phase
0.2 of $\approx 0.25$. We do not correct for interstellar extinction
because the corrections are less than, or of the same order as, the
uncertainty estimated above. For \object{AFGL 2290} Herman et al.\
(\cite{HISH84}) give a value of $A_{V} = 1.6$ for an adopted distance
of 1.19 kpc which reduces to $A_{V} = 1.3$ at a distance of 0.98
kpc. With the wavelength dependence of the interstellar extinction in
the infrared from Becklin et al.\ (\cite{BNMW78}) one obtains a
correction factor of 1.35 at $\lambda = 1.25\,{\rm\mu m}$, 1.03 at
$\lambda = 4.8\,{\rm\mu m}$, and 1.15 at $\lambda = 9.5\,{\rm\mu
m}$.
\object{AFGL 2290} was observed by IRAS (IRAS Point Source Catalog
\cite{IRAS85a}). We adopt the colour corrected broadband fluxes given
by van der Veen et al.\ (\cite{VOHM95}) and the IRAS low resolution
spectra from the IRAS Catalog of Low Resolution Spectra
(\cite{IRAS87}). The latter are corrected according to Cohen et al.\
(\cite{CWW92}). Since the broadband fluxes and spectra are averages of
several measurements taken at different phases (IRAS Explanatory
Supplement \cite{IRAS85b}), the flux levels, e.g.\ at $12\,{\rm\mu
m}$, are lower than the fluxes from ground based observations around
phase 0.2. Therefore, we multiply the IRAS data with a factor of 1.95
to join them with the ground based data.
Finally, observations at mm wavelengths were reported by Walmsley et
al.\ (\cite{WCKSFO91}) who measured a flux of 0.025 Jy at $\lambda =
1.25\,{\rm mm}$ at phase 0.18 (JD 2447960), and by van der Veen et
al.\ (\cite{VOHM95}) who derived $3\sigma$ upper limits of 0.13 Jy and
0.14 Jy at 0.76 mm and 1.1 mm respectively for phase 0.25 (JD
2448069), which are consistent with the 1.25 mm flux.
\subsection{Selection of input parameters}
\label{input-parameters}
We represent the central star by a blackbody with an effective
temperature $T_{\rm eff}$. In contrast to the visible M type Mira
variables with $T_{\rm eff} \la 3500\,{\rm K}$, the effective
temperature of OH/IR stars with optically thick dust shells cannot be
directly determined. However, if OH/IR stars can be considered as an
extension of the Mira sequence to longer periods and larger optical
depths, one might extrapolate the period--$T_{\rm eff}$ relation for
Mira variables derived by Alvarez \& Mennessier (\cite{AlMe97}) to
$P > 650\,{\rm d}$, which yields $T_{\rm eff} < 2500\,{\rm K}$ in
agreement with the values expected for the tip of the AGB.
The dust density distribution is obtained from the velocity law, which
results from an approximate analytic solution for a stationary dust
driven wind with constant mass loss rate (e.g.\ Schutte \& Tielens
\cite{SchTi89}). If the gas pressure force is neglected and the flux
averaged absorption coefficient is assumed to be constant with the radius
$r$ in the wind, the velocity distribution is given by
\begin{equation}\label{veleq}
v(r) = v_{\infty} \sqrt{ 1 - \frac{r_{1}}{r}
\left( 1 -
\left(\frac{v_{1}}{v_{\infty}}\right)^{2}\right) }
\end{equation}
where $v_{1}$ denotes the velocity at the inner boundary $r_{1}$, and
$v_{\infty}$ is the velocity at infinity. The relevant free parameter
is the ratio of these velocities $\delta = v_{1}/v_{\infty}$, because
only the normalized density distribution enters into the calculation.
We adopt this velocity law, because it accounts for the changing
density gradient due to the acceleration of the matter by radiation
pressure on dust in the innermost parts of the dust shell. Compared to
a dust shell with a $1/r^{2}$ density distribution and equal optical
depth the dust density at $r_{1}$ is higher by a factor of $0.5
(1+\delta)/\delta$ and the mass loss rate is lower by a factor of $0.5
(1+\delta)$ (cf.\ Le Sidaner \& Le Bertre \cite{LSLB93}).
According to the theory of dust driven winds the velocity at the inner
boundary, where efficient grain condensation takes place and the
acceleration of the matter by radiation pressure on dust starts, is
close to local sound velocity $c_{s}$ (see Gail \cite{Gail90}). This
is supported by observations of the velocity separation of the SiO
maser emission in OH/IR stars (Jewell et al.\ \cite{JBWW84}), which
presumably originates from the dust forming region. With $c_{s} \la
2\,{\rm kms^{-1}}$ for temperatures of about $1000\,{\rm K}$ and with
the measured outflow velocity of \object{AFGL 2290} of $v_{\infty} =
16\,{\rm kms^{-1}}$ (Herman \& Habing \cite{HerHab85}) one obtains
$\delta \approx 0.12$, which we adopt as the standard value for
$\delta$.
As described in the previous section, the location of the inner
boundary $r_{1}$ of the dust shell is determined by the choice of the
dust temperature $T_{1}$ at $r_{1}$. For the outer boundary $r_{\rm
out}$ we adopt a default value of $r_{\rm out} = 10^{3}\, r_{1}$. As
shown in the following section a larger outer boundary affects only
the far infrared fluxes for $\lambda = 100\,{\rm\mu m}$ without
altering the other properties of the model.
We consider spherical grains of equal size described by the grain
radius $a_{\rm gr}$. This is certainly a simplification because based
on theoretical and observational arguments, one expects the presence of
a grain size distribution $n(a_{\rm gr})$. Therefore, a size
distribution similiar to the one observed in the ISM ($n(a_{\rm gr})
\propto a_{\rm gr}^{-3.5}$) is often assumed for radiative transfer
models of circumstellar dust shells (e.g.\ Justtanont \& Tielens
\cite{JusTi92}, Griffin \cite{Grif93}). Consistent models for {\em
stationary} dust driven winds, which include a detailed treatment of
(carbon) grain formation and growth, in fact yield a broad size
distribution which can well be approximated by a power law (Dominik et
al.\ \cite{DGS89}). However, in circumstellar shells around {\em
pulsating} AGB stars the conditions determining the condensation of
grains change periodically. The time available for the growth of the
particles is restricted by the periodic variations of the temperature
and density. This results in a narrower size distribution
(Gauger et al.\ \cite{GGS90}, Winters et al.\ \cite{WFBS97}) which
might roughly be approximated by a single dominant grain size.
For the dust optical properties we adopt the complex refractive index
given by Ossenkopf et al.\ (\cite{OHM92}) for `warm, oxygen--deficient'
silicates. The authors consider observational determinations
of opacities of circumstellar silicates as well as laboratory data and
discuss quantitatively the effects of inclusions on the complex
refractive index, especially at shorter wavelengths ($\lambda <
8\,{\rm\mu m}$). These constants yield a good match of the overall
spectral shape of the observed SED of \object{AFGL 2290}, especially
of the $9.7\,{\rm\mu m}$ silicate feature.
However, we will also discuss the effects on the radiative transfer
models resulting from different optical constants in the Appendix.
With the tabulated values of the complex refractive index the
extinction and scattering efficiencies are calculated from Mie theory
for spherical particles assuming isotropic scattering.\\
Once a satisfactory fit of the spectral shape is achieved with
suitably chosen values for the remaining input parameters $T_{\rm
eff}$, $T_{1}$, $a_{\rm gr}$, and $\tau_{0.55}$ (the optical depth at
the reference wavelength $0.55\,{\rm\mu m}$) the match of the
normalized synthetic SED with the observed SED determines the
bolometric flux at earth $f_{\rm b}$.
Combined with the effective temperature one obtains the angular
stellar diameter $\theta_{*}$ and thereby the spatial scale of the
system. With an assumed distance $D$ to the object, the luminosity
$L_{*}$, the radius of the inner boundary in cm, and the dust mass
loss rate $\dot{M}_{\rm d}$ can be calculated. The latter quantity is
given by:
\begin{equation}\label{mdoteq}
\dot{M}_{\rm d} = 2 {\rm\pi} r_{1} v_{\infty} (1+ \delta)
\rho_{\rm gr}
\frac{\tau_{\rm d}(\lambda)}
{\tilde{Q}_{\rm ext}(\lambda)}
\end{equation}
where $\rho_{\rm gr}$ denotes the specific density of the grain
material, $\tau_{\rm d}(\lambda)$ is the dust optical depth at
wavelength $\lambda$, and $\tilde{Q}_{\rm ext} = \kappa_{\rm ext}
/ V_{\rm gr}$ is the extinction cross section $\kappa_{\rm ext}$ per
unit volume $V_{\rm gr}$ of the grains. For single sized grains
$\tilde{Q}_{\rm ext}$ is proportional to the extinction efficiency
divided by the grain radius $Q_{\rm ext} / a_{\rm gr}$, and in the
Rayleigh limit $2 \pi a_{\rm gr} \ll \lambda$ it is independent of the
grain radius $a_{\rm gr}$. For the specific density of silicate grains
we adopt $\rho_{\rm gr} = 3 {\rm gcm^{-3}}$ as a typical value. For
the distance to \object{AFGL 2290} we use $D = 0.98\,{\rm kpc}$ (van
Langevelde et al.\ \cite{LHS90}).
\section{Radiative transfer modeling of \object{AFGL 2290}}
\label{radtrf-modeling}
\subsection{A radiative transfer model for the SED of \object{AFGL 2290}}
\label{best-fit-sed}
Starting with the parameters of previous radiative transfer models for
\object{AFGL 2290} presented by Rowan--Robinson (\cite{RowRob82}),
Bedijn (\cite{Bed87}), and Suh (\cite{Suh91}), we achieved a
satisfactory match of the observed SED with the set of parameters
given in Table \ref{tab-14} after a few trials. Henceforth we will
refer to this parameter set as model A, and we will first discuss its
properties before we use it as a reference for an investigation of the
sensitivity of the results on the parameters.
\begin{table}
\caption{Parameters and resulting properties of model A.}
\label{tab-14}
\begin{tabular}{ccccc}\hline
$T_{\rm eff}$ [K] & $ T_{1}$ [K] & $a_{\rm gr}$ [${\rm\mu m}$] &
$r_{\rm out}/r_{1}$& $\tau_{0.55}$\\
2000 & 800 & 0.10 & $10^{4}$ & 100\\ \hline
& & & & \\
$\dot{M}_{\rm d}$ [${\rm M_{\odot}yr^{-1}}$]& $r_{1}$ [$R_{*}$] &
$f_{\rm b}$ [${\rm W m^{-2}}$] & $\theta_{*}$ [mas]& $\tau_{10}$\\
2.7 $10^{-7}$ & 7.80 & 3.0 $10^{-10}$ & 7.50 & 7.49 \\ \hline
\end{tabular}
\end{table}
The SED of model A is shown in Fig.\ \ref{sed-14}. Figure \ref{lrs-14}
displays an enlargement of the $5-25\,{\rm\mu m}$ region with the
$9.7\,{\rm\mu m}$ silicate feature. From the shortest wavelength at
$\lambda = 1.25\,{\rm\mu m}$ up to $\lambda = 1.25\,{\rm mm}$ model A
provides a good fit to the observations. The location, shape and
strength of the silicate feature around $10\,{\rm\mu m}$ is well
reproduced with the adopted optical data from Ossenkopf et al.\
(\cite{OHM92}). Only in the $18\,{\rm\mu m}$ region is there a
noticeable deviation because the model shows a weak, broad emission
which is absent in the IRAS LRS spectrum.
In addition to the input parameters, Table \ref{tab-14} also
lists the derived properties of model A.
Our value for the bolometric luminosity at earth of $f_{\rm b} = 3\,
10^{-10}\,{\rm W m^{-2}}$ is consistent with the values of $2.4\,
10^{-10}\,{\rm W m^{-2}}$ determined by van der Veen \& Rugers
(\cite{VeRu89}). With $T_{\rm eff}=2000\,{\rm K}$ the angular stellar
diameter is $\theta_{*}=7.5\,{\rm mas}$ and we obtain a stellar radius
$R_{*}=790\,{\rm R_{\odot}}$ and a luminosity $L_{*}=9000\,{\rm
L_{\odot}}$ adopting a distance $D=0.98\,{\rm kpc}$.
The bolometric flux should be quite accurate considering the quality
of the fit. However, the errors in the distance determination (e.g.\
$\sigma_{D}=0.34$, van Langevelde et al.\ \cite{LHS90}) and the
uncertainty in the determination of $T_{\rm eff}$ from the radiative
transfer modeling (see Sect.\ \ref{Teff-effects}) are rather large,
resulting in correspondingly large uncertainties for $L_{*}$, $R_{*}$
and $\dot{M}_{\rm d}$.
The derived dust mass loss rate of $2.7\, 10^{-7}\,{\rm
M_{\odot} yr^{-1}}$ is close to the values derived from radiative
transfer models by other authors. Bressan et al.\ (\cite{BGS98})
obtain $4.5\, 10^{-7}\,{\rm M_{\odot} yr^{-1}}$, and the model of
Bedijn (\cite{Bed87}) yields\footnote{
We have calculated $\dot{M}_{\rm d}$ from the values for $r_{1}$
and $\tau_{10}$ given by Bedijn (\cite{Bed87}) adopting
$\tilde{Q}_{\rm ext}(10\,{\rm\mu m})=5.6\,10^{3}\,{\rm cm^{-1}}$.}
$4\, 10^{-7}\,{\rm M_{\odot}yr^{-1}}$.
From a relation between the strength of the $10\,{\rm\mu m}$ feature
and the color temperature Schutte \& Tielens (\cite{SchTi89}) obtained
$\dot{M}_{\rm d} = 2.4\, 10^{-7}\,{\rm M_{\odot} yr^{-1}}$ for
\object{AFGL 2290}, and Heske et al.\ (\cite{HFOVH90}) estimated
$\dot{M}_{\rm d} = 1.2\,10^{-7}\,{\rm M_{\odot} yr^{-1}}$ from the
$60\,{\rm\mu m}$ IRAS flux.
\begin{figure}\unitlength 1cm
\includegraphics{H1209F6.eps}
\caption{
Spectral energy distribution for model A (solid line) and adopted
photometry of \object{AFGL 2290} at about phase 0.2 (see Fig.\
\ref{IR-obsfig} for the references and corresponding symbols). The
insert shows the SED in the mm range. The dotted line displays the
SED for a model with the parameters of model A and an outer
boundary of $r_{\rm out}=10^{3}\,r_{1}$.}
\label{sed-14}
\end{figure}
\begin{figure}\unitlength 1cm
\includegraphics{H1209F7.eps}
\caption{
$5-25\,{\rm\mu m}$ spectrum of model A (solid line) and the
corrected IRAS LRS spectra (dashed lines). The IRAS data are scaled
by a factor of 1.95 to match the ground based photometry (see Fig.\
\ref{IR-obsfig} for the references and corresponding symbols).}
\label{lrs-14}
\end{figure}
All together the parameters and derived properties of model A lie in
the typical range of values obtained from radiative transfer models
for OH/IR stars showing the silicate feature in absorption. In the
calculations of Lorenz--Martins \& de Ara{\'u}jo (\cite{LMA97})
$T_{\rm eff}$ ranges from $1800\,{\rm K}$ to $2400\,{\rm K}$, $T_{1}$
from $650\,{\rm K}$ to $1200\,{\rm K}$, and $\tau_{9.7}$ from 7 to 17.
Justtanont \& Tielens (\cite{JusTi92}) derive dust mass loss rates
between $2.6\, 10^{-7}\,{\rm M_{\odot}yr^{-1}}$ and
$2.2\,10^{-6}\,{\rm M_{\odot}yr^{-1}}$, and optical depths
$\tau_{9.7}$ between 4.85 and 19.6 for their sample of OH/IR stars.
In addition to model A with $r_{out}/r_{1} = 10^{4}$, Fig.\
\ref{sed-14} also displays the SED for a model with the same
parameters but with an outer boundary, which is ten times smaller,
i.e.\ $r_{out}/r_{1} = 10^{3}$ (dotted line).
The spectra are virtually indistinguishable up to $\lambda \ga 60
{\rm\mu m}$. The additional cold dust due to the larger outer
boundary of model A increases the far infrared fluxes (see insert of
Fig.\ \ref{sed-14}), but does not affect the SED at shorter
wavelengths (cf.\ Justtanont \& Tielens \cite{JusTi92}). Since the
values of the derived properties ($f_{\rm b}$, $\dot{M}_{\rm d}$, ...)
are determined essentially by the shape of the SED below $60\,{\rm\mu
m}$, they do not change for $r_{out}/r_{1} \ga 10^{3}$.
\begin{figure}\unitlength 1cm
\includegraphics{H1209F8.eps}
\caption{
Synthetic $2.11\,{\rm\mu m}$ visibility for model A (solid line)
and the azimuthally averaged $2.11\,{\rm\mu m}$ visibility obtained
from the speckle observation (squares).}
\label{vis-14}
\end{figure}
Figure \ref{vis-14} shows a comparison of the $2.11\,{\rm\mu m}$
visibility calculated for model A with the azimuthally averaged
$2.11\,{\rm\mu m}$ visibility $V_{\rm obs}$ derived from our speckle
observations. Although model A yields a good fit to the observed SED,
it fails to reproduce the observed visibility. The slope of the model
visibility is much shallower than the slope of the observed one, and
the model visibility levels off at $q \ga 20\,{\rm arcsec^{-1}}$ with
$V_{2.11}=0.6$. From the observed visibility we obtained an upper limit
for the contribution of the star to the monochromatic flux at
$2.11\,{\rm\mu m}$ of $\approx 40\,\%$, whereas model A yields
$62\,\%$, indicating that the model optical depth of $\tau_{2.11}=3$
at $2.11\,{\rm\mu m}$ is too low. Furthermore the steeper slope of
$V_{\rm obs}$ suggests that a levelling--off should occur at a smaller
spatial frequency compared to the model visibility, which means that
the observed intensity distribution would be rather more extended
than the intensity profile of model A.
To summarize, a dust shell model, which yields a visibility in
agreement with the observations, requires a larger optical depth at
$2.11\,{\rm\mu m}$ and a more extended intensity distribution
compared to model A, but it has to produce the same spectral energy
distribution. It turns out that these requirements can be partially
fulfilled with suitably modified parameter values.
\subsection{Effects of parameter variations on the calculated SED and
the model visibility}
\label{parameter-study}
In order to find a model which matches both the observed SED and the
visibility, as well as exploring the sensitivity of the SED and
the $2.11\,{\rm\mu m}$ visibility of the models on the input
parameters, we have calculated various sequences of models. In each
sequence one parameter was varied within a certain range, while the
other parameters were kept fixed at the values of model A, except for
the optical depth at the reference wavelength $\tau_{0.55}$. This
quantity was adjusted for each model in order to obtain a match of the
SED with the observation, especially with the observed strength of the
$9.7\,{\rm\mu m}$ silicate feature. It turns out that models, for
which the SED fit the feature equally well have (almost) equal dust
mass loss rates, i.e.\ with our procedure we force the models
in a sequence to have equal mass loss rates instead of equal
$\tau_{0.55}$.
\subsubsection{Effects of different dust temperatures at the inner
boundary}
\label{T1-effects}
The effects of different dust temperatures at the inner boundary on
the calculated SED and the $2.11\,{\rm\mu m}$ visibility are displayed
in Fig.\ \ref{sed-Ow92-T1}.
The dust temperature is varied between $T_{1}=600\,{\rm K}$ and
$T_{1}=1200\,{\rm K}$, and the derived properties of the corresponding
models are given in Table \ref{tab-Ow92-T1}.
\begin{table}
\caption{Resulting properties for models with the parameters of model
A, but with different dust temperatures at the inner
boundary.}
\label{tab-Ow92-T1}
\begin{center}
\begin{tabular}{crcrrc}\hline
$T_{1}$ & $\tau_{0.55}$ &
$\dot{M}_{\rm d}$ & $r_{1}$ &
$\tau_{10}$ & $f_{\rm b}$ \\
{}[K] & &
$[{\rm M_{\odot}yr^{-1}}$]& [$R_{*}$] &
& [${\rm W m^{-2}}$] \\ \hline
600 & 65 & 2.86 $10^{-7}$ & 12.89 & 4.87 & $2.60\,10^{-10}$ \\
800 & 100 & 2.66 $10^{-7}$ & 7.80 & 7.49 & $3.00\,10^{-10}$ \\
1000 & 150 & 2.70 $10^{-7}$ & 5.27 & 11.23 & $3.12\,10^{-10}$ \\
1200 & 200 & 2.66 $10^{-7}$ & 3.91 & 14.97 & $3.25\,10^{-10}$ \\
\hline
\end{tabular}
\end{center}
\end{table}
For the values of $T_{1}$ presented here, the SED is changed at
wavelengths below $\lambda \approx 9\,{\rm\mu m}$. Compared
to model A with $T_{1}=800\,{\rm K}$ a smaller value of $T_{1}$
results in a higher flux below $\lambda=3\,{\rm\mu m}$ and a
deficiency of flux at wavelengths between $3\,{\rm\mu m}$ and
$9\,{\rm\mu m}$. Values of $T_{1}>800\,{\rm K}$ produce a deficiency
of flux below $\lambda=3\,{\rm\mu m}$ and an excess of flux between
$3\,{\rm\mu m}$ and $9\,{\rm\mu m}$. As a consequence, the
bolometric fluxes are different for these models.
The changes of the SED for a variation of $T_{1}$ with fixed other
parameters, including $\tau_{0.55}$, have been studied already by
Ivezi{\'c} \& Elitzur (\cite{IE97}). For high $\tau_{0.55} > 100$
decreasing $T_{1}$ lowers the flux at shorter wavelengths, increases
the strength of the silicate feature and raises the flux at larger
wavelengths, similiar to the changes produced by increasing
$\tau_{0.55}$. Since we have adjusted $\tau_{0.55}$ for each value of
$T_{1}$ to reproduce the observed strength of the $9.7\,{\rm\mu m}$
silicate feature, the model SED are only changed at shorter
wavelengths in a modified way. This behaviour can be understood from
the competition of the effects caused by lowering $T_{1}$ and
simultaneously decreasing $\tau_{0.55}$. Up to $\lambda \la 3\,{\rm
\mu m}$ the increase of the monochromatic flux induced by the smaller
optical depth more than compensates the decrease of the flux due to a
lower $T_{1}$, but in the region $3\,{\rm\mu m} \la \lambda \la
8\,{\rm \mu m}$ the effect of lowering $T_{1}$ dominates and the
monochromatic flux decreases with $T_{1}$.
\begin{figure}\unitlength 1cm
\includegraphics{H1209F9.eps}
\caption{
SED (a) and visibilities (b) for models with the parameters of
model A, but with different dust temperatures at the inner
boundary:
$T_{1}=600\,{\rm K}$ (long dashed line),
$800\,{\rm K}$ (solid line),
$1000\,{\rm K}$ (short dashed line),
$1200\,{\rm K}$ (dotted line).
The corresponding optical depths and derived model properties are
given in Table \ref{tab-Ow92-T1}. The spectra have been scaled with
different $f_{\rm b}$ to match the observations at $\lambda >
8\,{\rm\mu m}$. The symbols denote the observations (see
Figs.\ \ref{Visibility} and \ref{IR-obsfig}).}
\label{sed-Ow92-T1}
\end{figure}
The variation of $T_{1}$ mainly affects the slope of the $2.11\,{\rm
\mu m}$ visibility $V_{2.11}$ via the different $\tau_{0.55}$. The
optical depth increases with $T_{1}$ resulting in a broader intensity
profile and a smaller stellar contribution to the monochromatic
flux. This leads to a steeper decline of $V_{2.11}$ (see Ivezi{\'c}
\& Elitzur \cite{IE96}). The curvature of the visibility only noticeably
changes at higher values of the spatial frequency. This change of
the slope of the visibility indicates the onset of a levelling off of
$V_{2.11}$, especially for the model with $T_{1}=600\,{\rm K}$. Since
its inner boundary is located at $12.9 R_{*}$ or 97~mas, the
visibility should approach a constant value at $q \approx 13\,{\rm
arcsec^{-1}}$. An increase of $T_{1}$ above 1000~K yields only a
marginally steeper slope of the visibility, although $\tau_{0.55}$
increases and the stellar contribution to the $2.11\,{\rm\mu m}$ flux
correspondingly decreases. However, the inner boundary is shifted
simultaneously to smaller radii, resulting in similiar slopes of
$V_{2.11}$ below $q \la 13\,{\rm arcsec^{-1}}$.
To summarize, the changes of the SED due to a variation of $T_{1}$ can
in principle be compensated by a corresponding variation of
$\tau_{0.55}$, but the presence of the silicate feature constrains
the choice of the optical depth to values which reproduce the strength
of the observed feature, i.e. to combinations of $T_{1}$ and
$\tau_{0.55}$ which yield a fixed value for $\dot{M}_{\rm d} \propto
r_{1} \tau_{10}$. Increasing $T_{1}$ yields a steeper slope of the
$2.11\,{\rm\mu m}$ visibility. Above $T_{1}=1000\,{\rm K}$, however,
these changes are small and will not result in a model which matches
the observed visibility.
\subsubsection{Effects of different effective temperatures on the SED
and the visibility}
\label{Teff-effects}
Figure \ref{sed-Ow92-Teff} displays the effects of different effective
temperatures on the calculated SED and the $2.11\,{\rm\mu m}$
visibility. The effective temperature is varied between $T_{\rm
eff}=1600\,{\rm K}$ and $T_{\rm eff}=2400\,{\rm K}$. The derived
properties of the corresponding models are given in
Table \ref{tab-Ow92-Teff}.
\begin{figure}\unitlength 1cm
\includegraphics{H1209F10.eps}
\caption{
SED (a) and visibilities (b) for models with the parameters of
model A, but with different effective temperatures:
$T_{\rm eff} = 1600\,{\rm K}$ (dashed line),
$2000\,{\rm K}$ (solid line),
$2400\,{\rm K}$ (dotted line).
The corresponding optical depths and derived properties are
given in Table \ref{tab-Ow92-Teff}. The spectra have been scaled
with different $f_{\rm b}$ to match the observations at $\lambda >
8\,{\rm\mu m}$. The symbols denote the observations (see
Figs.\ \ref{Visibility} and \ref{IR-obsfig}).}
\label{sed-Ow92-Teff}
\end{figure}
\begin{table}
\caption{Resulting properties for models with the parameters of model
A, but with different effective temperatures.}
\label{tab-Ow92-Teff}
\begin{center}
\begin{tabular}{crcrrcc}\hline
$T_{\rm eff}$ & $\tau_{0.55}$ &
$\dot{M}_{\rm d}$ & $r_{1}$ &
$\tau_{10}$ & $\theta_{*}$ &
$f_{\rm b}$ \\
{}[K] & &
[${\rm M_{\odot}yr^{-1}}$] & [$R_{*}$] &
& [mas] &
[${\rm Wm^{-2}}$] \\ \hline
1600 & 110 & 2.61 $10^{-7}$ & 4.45 & 8.23 & 11.7 &
$3.1\,10^{-10}$ \\
2000 & 100 & 2.66 $10^{-7}$ & 7.80 & 7.49 & 7.5 &
$3.0\,10^{-10}$ \\
2400 & 92 & 2.73 $10^{-7}$ & 12.60 & 6.86 & 5.2 &
$2.9\,10^{-10}$ \\
\hline
\end{tabular}
\end{center}
\end{table}
The effects of the variation of $T_{\rm eff}$ on the model SED are
qualitatively similiar to the effects induced by a variation of
$T_{1}$. Lowering $T_{\rm eff}$ from \mbox{2400 K} to \mbox{1600 K}
decreases the monochromatic flux below $\lambda \la 2.5\,{\rm\mu m}$
and increases the flux at longer wavelengths up to $\lambda = 8\,{\rm
\mu m}$. The main cause for these changes is the shift of the
wavelength $\lambda_{m}$, where the stellar (blackbody) spectrum
reaches its maximum, from $\lambda_{m}=1.5\,{\rm\mu m}$ to
$\lambda_{m}=2.3\,{\rm\mu m}$. The monochromatic flux of the `hotter'
star is larger below a certain wavelength, which is, in addition,
affected by the slightly different optical depths. However, for the
large optical depths considered here, the effects of different
$T_{\rm eff}$ on the SED are small as already shown by e.g.\
Rowan--Robinson (\cite{RowRob80}). It is interesting to note that the
changes of the SED due to a variation of $T_{1}$ can be compensated
for by a corresponding variation of $T_{\rm eff}$, at least within a
certain range of values.
The variation of $T_{\rm eff}$ has only a negligible effect on the
$2.11\,{\rm\mu m}$ visibility for the temperature range considered
here. Since the changes of the optical depths are small, both the
angular diameter of the inner boundary and the stellar contribution to
the $2.11\,{\rm\mu m}$ flux only moderately increase with
$T_{\rm eff}$. Correspondingly, the visibility approaches a slightly
higher constant value at a slightly smaller spatial frequency
resulting in the minor differences for $V_{2.11}$ at spatial
frequencies $q<13.5\,{\rm arcsec^{-1}}$. Thus, changing the
$T_{\rm eff}$ within a reasonable range cannot produce a model, which
matches the observed visibility.
\subsubsection{Effects of different grain radii}
\label{a0-effects}
The effects of different grain radii on the calculated SED and the
$2.11\,{\rm\mu m}$ visibility are displayed in Fig.\ \ref{sed-Ow92-a0}.
The grain radius is varied between $a_{\rm gr}=0.04\,{\rm\mu m}$ and
$a_{\rm gr}=0.12\,{\rm\mu m}$. The derived properties of the
corresponding models are given in Table \ref{tab-Ow92-a0}. Figure
\ref{sed-Ow92-a0-qext} shows the extinction coefficient per unit
volume of the grains $\kappa_{\rm ext}/V_{\rm gr}$ obtained from the
optical data for `warm' silicates from Ossenkopf et al.\
(\cite{OHM92}).
The choice of the grain radius only affects the short wavelength tail
of the SED below $\lambda \la 3\,{\rm\mu m}$, because at these
wavelengths $\kappa_{\rm ext} / V_{\rm gr}$ still depends on $a_{\rm
gr}$, but it becomes independent of $a_{\rm gr}$ at longer
wavelengths (see Fig.\ \ref{sed-Ow92-a0-qext}). This behaviour
is caused by two factors: the contribution of scattering to
extinction and the dependence of the absorption efficiency on the
grain size. The scattering efficiency per unit volume of the grains,
which is $\propto a_{\rm gr}^3$, steeply declines with increasing
wavelength and can be neglected above a certain wavelength depending
on $a_{\rm gr}$. The absorption efficiency depends on the grain size
only at short wavelengths and becomes independent of $a_{\rm gr}$ once
the grains are sufficiently small compared to the wavelength.
Therefore, the grain radius is constrained by the observed fluxes at
the shortest wavelengths $\lambda \la 2\,{\rm\mu m}$. In our case the
photometry at $\lambda=1.65\,{\rm\mu m}$ excludes grain radii $a_{\rm
gr} \ga 0.12\,{\rm\mu m}$ and $a_{\rm gr} \la 0.06\,{\rm\mu m}$
and the photometry at $\lambda=1.25\,{\rm\mu m}$ restricts the grain
radii to values close to $a_{\rm gr} = 0.1\,{\rm\mu m}$. However, the
values of the absorption and scattering efficiency depend on the
adopted optical data. A different data set can result in vastly
different grain radii (see Appendix A).
\begin{figure}\unitlength 1cm
\includegraphics{H1209F11.eps}
\caption{
SED (a) and visibilities (b) for models with the parameters of
model A, but with different grain radii ranging from
$a_{\rm gr}=0.04\,{\rm\mu m}$ to $a_{\rm gr}=0.16\,{\rm\mu m}$.
The corresponding optical depths and derived model properties are
given in Table \ref{tab-Ow92-a0}. The symbols denote the
observations (see Figs.\ \ref{Visibility} and \ref{IR-obsfig}).}
\label{sed-Ow92-a0}
\end{figure}
\begin{table}
\caption{Resulting properties for models with the parameters of model
A, but with different grain radii.}
\label{tab-Ow92-a0}
\begin{center}
\begin{tabular}{cccccc}\hline
$a_{\rm gr}$ & $\tau_{0.55}$ & $\dot{M}_{\rm d}$ &
$r_{1}$ & $\tau_{10}$ & $\tau_{2.2}$ \\
$[ \mu $m] & &[${\rm M_{\odot}\,yr^{-1}}$]&
[$R_{*}$] & & \\ \hline
0.04 & 20.0 & 2.60 $10^{-7}$ & 6.93 & 8.22 & 2.86 \\
0.06 & 37.5 & 2.69 $10^{-7}$ & 7.15 & 8.24 & 2.96 \\
0.08 & 65.6 & 2.66 $10^{-7}$ & 7.41 & 7.86 & 2.98 \\
0.10 & 100 & 2.66 $10^{-7}$ & 7.80 & 7.49 & 3.07 \\
0.12 & 140 & 2.63 $10^{-7}$ & 8.22 & 7.03 & 3.20 \\
0.14 & 170 & 2.60 $10^{-7}$ & 8.70 & 6.59 & 3.40 \\
0.16 & 150 & 2.65 $10^{-7}$ & 9.24 & 6.33 & 3.76 \\ \hline
\end{tabular}\\
$f_{\rm b}=3\,10^{-10}\,{\rm W m^{-2}}$, $\theta_{*} = 7.5$ mas.
\end{center}
\end{table}
The variation of the grain radius has two effects on the $2.11\,{\rm
\mu m}$ visibility. First, the slope of $V_{2.11}$ steepens with
increasing values of $a_{\rm gr}$, because models with larger grain
radius require a higher optical depth at $2.11\,{\rm\mu m}$ in order
to match the observed SED for $\lambda>2\,{\rm\mu m}$. With
increasing optical depth the intensity distribution becomes broader
and the stellar contribution to the monochromatic flux at $2.11\,{\rm
\mu m}$ decreases. Correspondingly, the decline of visibility with
spatial frequency becomes steeper (see Ivezi{\'c} \& Elitzur
\cite{IE96}). The second effect is the change of the curvature of
$V_{2.11}$, which is noticable at low spatial frequencies.
The curvature changes its sign at about $a_{\rm gr}<0.1\,{\rm\mu m}$.
This behaviour reflects the changes of the
spatial intensity distribution. At large radial offsets $b$ from the
star the intensity decreases approximately as $I(b) \propto b^{-3}$,
because the optical depth along the line of sight at b becomes small
(see Jura \& Jacoby \cite{JuJa76}). For smaller offsets $b$, however,
the decline of the intensity steepens and the slope of the visibility
changes accordingly, i.e.\ the curvature of V indicates the changing
slope of the spatial intensity distribution. Thus, the visibility
constrains the grain radii not only via its slope, but also via its
curvature.
The observed visibility $V_{\rm obs}$ declines in almost a straight
line to values below $\sim 0.40$ at $q=13.5{\rm arcsec^{-1}}$ with
only a slight curvature. $V_{\rm obs}$ is fairly well matched
by the model with $a_{\rm gr}=0.16\,{\rm\mu m}$, although the
curvature of the model visibility is a little too strong. Since the
visibilities for models with $a_{\rm gr}=0.15\,{\rm\mu m}$ and
$a_{\rm gr}=0.17\,{\rm\mu m}$ already fall outside the error bars of
the observation, at least for certain spatial frequencies, the grain
radius is determined by $V_{\rm obs}$ with an uncertainty
$<0.01\,{\rm\mu m}$ (cf.\ Groenewegen \cite{Groe97}).
However, the SED for the model with $a_{\rm gr}=0.16\,{\rm\mu m}$,
shows a deficit of flux below $\lambda = 3\,{\rm\mu m}$.
This deficit cannot be removed by a change of $T_{1}$ or
$T_{\rm eff}$. Although lowering $T_{1}$ increases the flux at
$\lambda \la 3\,{\rm\mu m}$ it also decreases the flux at longer
wavelengths. Furthermore, the inner boundary of the dust shell is
moved outwards which deforms the resulting visibility in a way that
destroys the fit. Increasing $T_{\rm eff}$ yields a similiar
behaviour.
\begin{figure}\unitlength 1cm
\includegraphics{H1209F12.eps}
\caption{
Extinction coefficient per unit volume of the grains
$\kappa_{\rm ext}/V_{\rm gr}$ for different grain radii ranging
from $a_{\rm gr}=0.04 \,{\rm\mu m}$ to
$a_{\rm gr}=0.16 \,{\rm\mu m}$ calculated with the optical data
from Ossenkopf et al.\ (\cite{OHM92}) for `warm' silicates.}
\label{sed-Ow92-a0-qext}
\end{figure}
The deficit of the short wavelength model flux could have several
causes, but the clear evidence for a non-spherical dust distribution
around \object{AFGL 2290} from our speckle masking observations favors
the explanation that the deficiency of flux in the model is due to the
assumption of a spherically symmetric circumstellar dust shell.
A more general assumption would be that the CDS has an axisymmetric,
`disk--like' structure. Theoretical investigations show that the
variation of the effective optical depth with the inclination of a
disk--like dust distribution affects the shape of the SED up to far
infrared wavelengths, as well as the monochromatic intensity
distributions and the corresponding visibilities (e.g.\ Efstathiou \&
Rowan--Robinson \cite{EfRo90}; Collison \& Fix \cite{CoFi91};
Lopez et al.\ \cite{LML95}; Men'shchikov \& Henning \cite{MeHe97}). If
the disk--like dust distribution is viewed at an intermediate
inclination one expects more flux at visual wavelengths than in the
case of a spherically symmetric dust distribution, due to scattered
light escaping from the optically thinner polar region located either
above or below the equatorial plane. In other words, we expect a
deficiency of the model flux at short wavelengths if we model the SED
of an aspherical dust distribution assuming a spherically symmetric
dust distribution.
\section{Summary and conclusions}
\label{summary}
We have presented the first diffraction--limited $2.11\,{\rm\mu m}$
speckle masking observations of the circumstellar dust shell around
the highly obscured type II OH/IR star \object{AFGL 2290}. The
resolution achieved with the SAO 6 m telescope is 75~mas, which is
sufficient to partially resolve the circumstellar dust shell at this
wavelength. From a 2--dimensional Gaussian fit of the visibility
function the diameter was determined to be 43~mas$\times$51~mas, which
corresponds to a diameter of 42~AU$\times$50~AU for a distance of
0.98~kpc. The reconstructed image shows deviations from a spherical
structure with an elongation at position angle $130^{\circ}$.
Our new high resolution spatial measurements provide additional strong
constraints for radiative transfer models for the dust shell of
\object{AFGL 2290}, supplementing the information provided by the
spectral energy distribution (SED). In order to investigate the
structure and the properties of the circumstellar dust shell we have
performed radiative transfer calculations assuming a spherically
symmetric dust distribution. The spectral energy distribution at phase
$\sim 0.2$ can be well fitted at all wavelengths by a model with an
effective temperature of 2000~K, a dust temperature at the inner
boundary of 800~K, an optical depth at $0.55\,{\rm\mu m}$ of 100, and
a radius for the single--sized grains of $0.1\,{\rm\mu m}$, using the
optical constants for `warm' silicates from Ossenkopf et al.\
(\cite{OHM92}). From this fit we derived e.g.\ a bolometric flux at
earth of $3\, 10^{-10}\,{\rm W m^{-2}}$, a radius of the inner
boundary of the dust shell of $r_{1} = 7.8\, R_{*}$, and a dust mass
loss rate of $2.7\,10^{-7}\,{\rm M_{\odot}yr^{-1}}$, in agreement with
the results of previous radiative transfer models for \object{AFGL
2290}. However, this model {\em does not} reproduce the observed
$2.11\,{\rm\mu m}$ visibility function.
We have, therefore, investigated the changes of the calculated SED and
the model visibility with the input parameters in search of an improved
model.
We found that the grain size is the key parameter in achieving a
fit of the observed visibility, while retaining at least a partial
match of the SED. Both the slope and the curvature of the visibility
react sensitively to the assumed grain radii.
With the assumption of single--sized grains we obtain an uncertainty
of less then $\pm 0.01\,{\rm\mu m}$ for $a_{\rm gr}$.
Another result was that the dust mass loss rate is well constrained
by the shape of the SED at longer wavelengths and, especially, by the
shape of the silicate absorption feature. For given optical constants
the value of the dust mass loss rate, as derived from the match of the
feature, is not very sensitive to changes of the input parameters.
The uncertainty of $\dot{M}_{\rm d}$ is $\sim 3\,10^{-8}\,{\rm
M_{\odot}yr^{-1}}$. The effective temperature and the dust
temperature at the inner boundary, however, are less well
constrained. We roughly estimate a range of $\pm 300\,{\rm K}$ for
$T_{\rm eff}$ and $\pm 100\,{\rm K}$ for $T_{1}$.
The shape of the observed visibility and the
strength of the silicate feature constrain the possible grain radii
and optical depths of the model. The observed visibility
can be reproduced by a model with a larger grain size of
$0.16\,{\rm\mu m}$ and a higher $\tau_{V}=150$, preserving
the match of the SED at longer wavelengths. Nevertheless, the model
shows a deficiency of flux at short wavelengths, which can be
explained if the dust distribution is not sperically symmetric. If the
CDS of \object{AFGL 2290} has in fact a disk--like structure, the
radial optical depths vary between the equatorial and polar
direction. Due to the scattered radiation escaping from the optically
thinner polar regions one expects more flux at shorter wavelengths
than from a spherically symmetric system with equal optical depth
towards the star.
\begin{acknowledgements}
This research made use of the SIMBAD database, operated by CDS in
Strasbourg.
\end{acknowledgements}
|
\section{Introduction}
In this symposium, there are several talks devoted to the recent
developments of the large $N$ limit of gauge theories and the
matrix models for string unification. We will discuss
a specific matrix model in zero dimension based on $USp(2k)$
introduced in ref.~\cite{IT1,IT2}. ( We will refer to zero
dimensional matrix model in general as reduced matrix model.)
We will present the criteria, the logic and the construction leading to the
model as well as theoretical implications from our present understanding.
Gauge fields and strings - the two notions occupying our mind- have had
interesting relationship: which of the two notions is more fundamental has
shifted from one to the other over the decades.
Without making our talk historical, let us begin with mentioning that
our current practise is to construct string theory from noncommuting
matrix degrees of freedom which originated from gauge fields.
The major goal is
to overcome the difficulties of the first
quantized superstring theory which have prevented us from
predicting physical quantities: this will include
the one associated with the existence of
the infinitely degenerate perturbative vacua.
Let us first recall the five consistent perturbative superstrings in ten
dimensions constructed by the end of 1984.
\bigskip
{\tiny
\renewcommand{\arraystretch}{2}
\begin{tabular}[t]{|c|c|} \hline
10 dim N =2 (32 supercharges) & 10 dim N=1 (16 supercharges)\\ \hline
type IIA \hspace{1cm} type IIB &
type I \hspace{0.5cm} $SO(32)$ het \hspace{0.5cm} $E_8 \times E_8$ het
\\ \hline
\end{tabular}
}
\bigskip
\noindent
Type $IIB$ superstrings are related to type $I$ superstrings by the
twist operation and the addition of the open-string sectors.
The rest are related by the Wilson lines and the T duality in nine
dimensions and by the S duality.
The reduced model of type $IIB$ superstrings has been proposed before
\cite{IKKT}. We will focus on the reduced model
which descends from the first quantized nonorientable
type $I$ superstrings
\cite{GSano}: they are related to heterotic strings
by S duality.
The $USp(2k)$ matrix model thus has a phenomenological
perspective accessible to us by the presence of gauge bosons,
matter fermions and other properties.
The reduced model in general lays its basis on the
correspondence with
the covariant Green-Schwarz superstrings
in the Schild gauge \cite{Schild}.
In this sense, the applicability of the reduced model is by no means
limited to low energy phenomena although its equivalence
with the first quantized critical superstrings has so far been
eatablished at the level of classical equations of motion on the
two-dimensional worldsheet.
One dimensional matrix model \cite{BFSS}
of $M$ theory \cite{M} has, on the other hand,
obtained successes on the agreement of the spectrum and other properties
with the low energy eleven-dimensional
supergravity theory.
It has been demonstrated in \cite{IT1,IT2} that the $USp(2k)$ matrix model
is uniquely selected by the three requirements:
\bigskip
{\bf Requirements :}
1)having eight dynamical and eight
kinematical supercharges.
2)obtained by an appropriate projection
from the $IIB$ matrix model and an addition of the degrees of freedom
corresponding to open strings.
3)nonorientable.
\bigskip
In the next section, we summarize the criteria and the logic leading to
the model \cite{IT2}.
We will begin with presenting the
closed string sector of the model. This will include
introduction of the $USp$ projector and its commutativity with
$8+8$ supersymmetry. We also discuss
the reduced model- Green-Schwarz correspondence \cite{IKKT}.
After the discussion \cite{Itsu}
on the open string sector, the loop variables and the Chan-Paton symmetry,
we will present the action of the model in its final
form.
The reduced matrix model is a constructive approach to superstring theory.
It is at the same time dynamical theory of spacetime points, which
we briefly discuss in the last subsection.
In quantum mechanics, space is a dynamical variable while time is
considered to be a parameter. In (relativistic) quantum field theory,
both space and time are parameters. In reduced model, both space and time
are dynamical variables appearing as eigenvalue distributions.
This is an ideal setup for pursuing
quantum gravity which regards spacetime as a derived concept.
In the subsequent sections, we
discuss three subjects which are relevant to the properties
of the $USp(2k)$ matrix model.
Leaving aside the issue of lifting the degenerate perturbative
vacua and the true scaling limit of the model, the matrix model
permits us to consider a series of such vacua
with $D$ objects through its $T$ dualized worldvolume representation.
We will discuss \cite{IT2} in section three a particular series of
perturbative vacua
associated with the $USp(2k)$ matrix model and the consistency of the model
with the literature, examining properties of $4d,5d,6d$ worldvolume
field theories.
In section four, we study the model by $T$-dualizing in the time direction,
namely, by the $T$-dualized quantum mechanics.
The main purpose here is to reveal the existence of branes as quantized
degrees of freedom which the degrees of freedom in the fundamental
representation are responsible for. This is done by examining
properties of the fermionic integrations via the (non)-abelian
Berry phase \cite{IM,CIK}.
We find formation of extended objects such as
Dirac monopoles and its nonabelian generalization.
The complete consideration of the model can be established by
the Schwinger-Dyson/loop equations \cite{Itsu}, which are considered to be
the second quantized formulation of the reduced matrix model.
We present the derivation in the final section.
These loop equations
exhibit a complete set of the joining and splitting interactions
required for the nonorientable $Type I$ superstrings. The study
at the linearized level provides us with the Virasoro conditions and
the mixed Dirichlet/Neumann boundary conditions acting on the
closed and open loop variables.
We will make remarks not mentioned in the original papers along
the way the discussion goes. Unless it is necessary,
we will suppress the $USp$ indices in most of our discussion.
More extensive references are found in the original papers
\cite{IT1,IT2,Itsu,IM,CIK}.
\section{Criteria and logic leading to the model}
\subsection{projection from $IIB$ matrix model}
We begin with the closed string sector. According
to the criteria mentioned in the introduction, this sector should be
obtained from the action of the $IIB$ matrix model via the appropriate
projection which we will determine in the next subsection.
\subsubsection{$IIB$ matrix model}
The action of the $IIB$ matrix model is
\begin{eqnarray}
S_{IIB}(\underline{v}_{M}, \underline{\Psi} ) =
\frac{1}{g^{2}} Tr \left( \frac{1}{4} \left[\underline{v}_{M},
\underline{v}_{N} \right]
\left[ \underline{v}^{M}, \underline{v}^{N} \right] - \frac{1}{2}
\bar{\underline{\Psi}} \Gamma^{M} \left[ \underline{v}_{M},
\underline{\Psi} \right] \right) \;\;.
\end{eqnarray}
The symbols with underlines lie in the adjoint representation
of $U(2k)$ and
\begin{eqnarray}
\label{eq:phiv}
\Psi = \left(
\lambda , 0, \psi_{\Phi_{1}}, 0 ,\psi_{\Phi_{2}}, 0 , \psi_{\Phi_{3}}, 0 ,
0, \bar{\lambda}, 0,
\bar{\psi}_{\Phi_{1}}, 0,
\bar{\psi}_{\Phi_{2}}, 0, \bar{\psi}_{\Phi_{3}} \right)^t \;\;.
\end{eqnarray}
is a thirty two component Majorana-Weyl spinor
satisfying
\begin{eqnarray}
C \bar{\Psi}^t = \Psi \;
, \; \Gamma_{11} \Psi = \Psi \;\;\;.
\label{eq:majorana-weyl con}
\end{eqnarray}
Later we will also use
\begin{eqnarray}
\Phi_{i} = \frac{1}{\sqrt{2}} \left( v_{3+i} +i v_{{6+i}}
\right) \;\;. \;
\end{eqnarray}
The ten dimensional gamma matrices have been denoted by
$\Gamma^{M}$.
\subsubsection{covariant Green-Schwarz superstrings in the Schild gauge}
In the large $k$ limit, one can check that this action
$S_{IIB}(\underline{v}_{M}, \underline{\Psi} )$ goes to that of
the covariant Green-Schwarz superstrings in the Schild gauge.
Let us sketch how this is seen. In the large $k$ limit, the group $SU(\infty)$
goes to the group of area preserving diffeomorphisms $(APD)$ on, for
example, torus. ( We ignore the issue of the worldsheet geometry to consider
and other subtleties here.) The generators are represented by
two index objects $L_{\vec{m}}$, $\vec{m} = (m_{1}, m_{2})^{t}$
with $m_{1}$, $m_{2}$ being integers. The algebra reads
\begin{eqnarray}
\left[ L_{\vec{m}}, L_{\vec{n}} \right] = \left( m_{1}n_{2}-
m_{2}n_{1} \right) L_{\vec{m}+\vec{n}} \;\;.
\end{eqnarray}
The $\underline{v}_{M}$ are expanded in $L_{\vec{m}}$ and can be written as
\begin{eqnarray}
\underline{v}_{M} &=& \int \frac{d^{2}\sigma}{(2\pi)^{2}} \tilde{\underline{v}}_{M}
\left( \vec{\sigma} \right) \tilde{L} \left( \vec{\sigma} \right)
\nonumber \\
&\equiv& \tilde{L}_{ \tilde{\underline{v}}_{M}} \;\;,
\end{eqnarray}
where
\begin{eqnarray}
\tilde{L} \left( \vec{\sigma} \right) \equiv \sum_{\vec{m}}
e^{-i\vec{m} \vec{\sigma}} L_{\vec{m}} \;\;.
\end{eqnarray}
Using
\begin{eqnarray}
\left[ \tilde{L}_{f}, \tilde{L}_{g} \right] =
- \tilde{L}_{ \{ f,g \}_{P.B} } \;\;,
\end{eqnarray}
one can readily derive
\begin{eqnarray}
\lim_{k \rightarrow \infty} S_{IIB} &=& S_{Schild} \\
S_{Schild} &\equiv& \frac{1}{2\pi \alpha^{\prime}}
\int \frac{d^{2}\sigma}{(2\pi)^{2}} \left( \frac{1}{2}
\tilde{\Sigma}_{MN} \tilde{\Sigma}^{MN} + 2i \tilde{\bar{\psi}} \Gamma^{M}
\left\{ \tilde{\underline{v}}_{M}, \tilde{\psi} \right\}_{P.B} \right) \;\;.
\end{eqnarray}
Here
\begin{eqnarray}
\tilde{\Sigma}_{MN} \equiv \left\{ \tilde{\underline{v}}_{M}, \tilde{\underline{v}}_{M}
\right\}_{P.B.} \;\;.
\end{eqnarray}
On the other hand, starting from the covariant Green-Schwarz action,
we can fix the local $\kappa$ symmetry via the condition \cite{IKKT}
\begin{eqnarray}
\theta^{1} = \theta^{2} \equiv \psi \;\;,
\end{eqnarray}
where $\theta^{1,2}$ are Majorana-Weyl spinor fields on the worldsheet
having the sama chirality.
We obtain
\begin{eqnarray}
S_{GS}^{(fix)} &=& - \frac{1}{2\pi \alpha^{\prime}} \int
\frac{d^{2}\sigma}{(2\pi)^{2}} \left( \sqrt{-\frac{1}{2} \Sigma^{2}}
\right. \nonumber \\
&& \left. +2i \bar{\psi}\Gamma^{M}\left\{X_{M}, \psi \right\}_{P.B.}
\right) \;\;
\end{eqnarray}
in the Nambu-Goto form.
Here $ \Sigma_{MN} \equiv \left\{ X_{M}, X_{N}
\right\}_{P.B.} $.
Equation of motion obtained from $ S_{GS}^{(fix)}$ reduces to
that from $S_{Schild}$ provided $\partial_{a} \left( \Sigma^{2}
\right) = 0 $, which can be shown by using again equation of motion
obtained from $S_{Schild}$.
\subsubsection{USp/So projector}
In order to make the closed string sector nonorientable, we need
a projection of $u(2k)$ Lie algebra valued matrices,
which corresponds to the twist operation $\Omega$ on the worldsheet.
Natural structure to consider is an embedding of
$usp$ and $so$ Lie algebras into the $u(2k)$ Lie algebra.
In both cases, it is expedient to introduce the following projector:
\begin{eqnarray}
\label{projector}
\hat{\rho}_{\mp} \bullet \equiv \frac{1}{2} \left( \bullet \mp F^{-1}
\bullet^{t} F \right) \;\;
\end{eqnarray}
Using this projector, one can uniquely decompose $u(2k)$ Lie algebra
valued matrices into the adjoint(= symmetric) and the antisymmetric
representations of the $usp$ Lie algebra or into the adjoint(= antisymmetic)
and symmetric representations of the $so$ Lie algebra. This is
schematically drawn as
\bigskip
\begin{tabular}{lcc}
$F = \left(
\begin{array}{cc}
0 & I \\
-I & 0
\end{array}
\right)$ &
$U(2k)$ \mbox{adj} &
$\begin{array}{l}
\nearrow \, USp \ \mbox{adj(= sym)} \\
\searrow \, USp \ \mbox{asym}
\end{array}$ \\
$F = \left(
\begin{array}{cc}
0 & I \\
I & 0
\end{array}
\right)$ &
$U(2k)$ \mbox{adj} &
$\begin{array}{l}
\nearrow \, SO \ \mbox{adj(= asym)} \\
\searrow \, SO \ \mbox{sym}
\end{array}$
\end{tabular}
\bigskip
We have found out in references \cite{IT2,Itsu} that the analysis
based on the planar diagrams, the consistency with the worldvolume
field theory and the Chan-Paton factor of the open loop variable
all lead to the choice of the $USp$ case. In this talk we will only
include this third discussion in subsection $2.3$.
Let $L_{\vec{m}}$ be a generator belonging to the antisymmetric
representation of $USp$. In this case,
$L_{\vec{m}}^{t} =L_{\vec{m}^{t}}$ with
$\vec{m}^{t} \equiv \left( -m_{1},m_{2} \right)$. We find
\begin{eqnarray}
\label{twistop}
F^{-1} \underline{v}_{M} F = \underline{v}_{M}^{t} =
\int \frac{d^{2}\sigma}{(2\pi)^{2}} \tilde{\underline{v}}_{M}
\left(-\sigma_{1},\sigma_{2} \right) \tilde{L} \left(
\vec{\sigma} \right) \;\;.
\end{eqnarray}
The matrix $F$ is in fact the matrix counterpart of
the twist operation $\Omega$. If
$L_{\vec{m}}$ be a generator belonging to the symmetric(=adjoint)
representation of $USp$, we obtain an extra minus sign
to eq. (\ref{twistop}), telling us an orientifold operation.
\subsubsection{reduced model of closed nonorientable superstrings}
To summarize, the closed string sector of the reduced matrix model
descending from the type $I$ superstrings must take the following form:
\begin{eqnarray}
S_{close} \equiv S_{0} ( v_{m} , \Phi_{I},
\lambda, \psi_{\Phi_{I}},
\bar{\Phi}_{I}, \bar{\lambda}, \bar{\psi}_{\Phi_{I} })
= S_{IIB}( \hat{\rho}_{b\mp}\underline{v}_{M}, \hat{\rho}_{f\mp}\underline{\Psi} ) \;\;.
\label{eq:equiv}
\end{eqnarray}
Here $\hat{\rho}_{b\mp}$ is a diagonal matrix acting on the vector
indices while $\hat{\rho}_{f\mp}$ is a diagonal matrix
acting on the spinor indices. Each entry of these two matrices is
either $\hat{\rho}_{-}$ or $\hat{\rho}_{+}$.
How these are chosen while preserving $8+8$ supersymmetry is the subject of
the next subsection.
\subsection{$USp$ projector and supersymmetry}
In order to make $8+8$ supersymmetry and the $USp$ projector compatible,
the model must implement a set of conditions under which the projectors
$\hat{\rho}_{b \mp}$, $\hat{\rho}_{f \mp}$,
and dynamical $\delta^{(1)}$ as well as
kinematical $\delta^{(2)}$ supersymmetry commute.
This constraint of commutativity turns out to be very stringent
and essentially leads to the unique possibility. We will repeat the
discussion of \cite{IT2}.
Start with
\begin{eqnarray}
\delta^{(1)} \underline{v}_{M} &=& i \bar{\epsilon} \Gamma_{M} \underline{\Psi}
\label{eq:d11} \\
\delta^{(1)} \underline{\Psi} &=& \frac{i}{2} \left[ \underline{v}_{M}, \underline{v}_{N} \right]
\Gamma^{MN} \epsilon \label{eq:d12} \\
\delta^{(2)}\underline{v}_{M} &=& 0 \label{eq:d21} \\
\delta^{(2)} \underline{\Psi} &=& \xi \;\;\;. \label{eq:d22}
\end{eqnarray}
Let us write generically
\begin{eqnarray}
\label{eq:prv}
v_{M} &\equiv& \delta_{M}^{~N} \hat{\rho}_{b \mp}^{(N)}
\underline{v}_{N} \;\;\; \nonumber \\
\Psi_{A} &\equiv& \delta_{AB} \hat{\rho}_{f \mp}^{(B)} \underline{\Psi}_B \;\;.
\end{eqnarray}
The condition
$\left[\hat{\rho}_{b \mp}, \delta^{(1)} \right] \underline{v}_{M} =0$ together
with eq.~(\ref{eq:d11}) gives
\begin{eqnarray}
\label{eq:1}
\sum_{A=1}^{32} \left( \bar{\epsilon} \Gamma_{M}\right)_{A} \left(
\hat{\rho}_{f \mp}^{(A)}- \hat{\rho}_{b \mp}^{(M)} \right) \underline{\Psi}_{A}
= 0 \;\;,
\end{eqnarray}
with index $M$ not summed.
The condition
\begin{eqnarray}
\label{eq:con2}
\left.
\left[\hat{\rho}_{f \mp}, \delta^{(1)} \right] \underline{\Psi}
\right|_{\underline{v}_{M}
\rightarrow \hat{\rho}_{b \mp}\underline{v}_{M} } =0
\end{eqnarray}
together with eq.~(\ref{eq:d12}) provides
\begin{eqnarray}
\label{eq:2}
\left( 1 - \hat{\rho}_{f \mp}^{(A)} \right) \left[ \hat{\rho}_{b \mp}^{(M)}
\underline{v}_{M}, \hat{\rho}_{b \mp}^{(N)} \underline{v}_{N} \right]
\left( \Gamma^{MN} \epsilon \right)_{A} =0 \;\;\;.
\end{eqnarray}
The restriction at eq.~(\ref{eq:con2}) comes from the fact that
eq.~(\ref{eq:d12}) is true only on shell.
Eq. ~(\ref{eq:d21}) does not give us anything new while
$\left[\hat{\rho}_{f \mp}, \delta^{(2)} \right] \underline{\Psi} =0$ with
eq.~(\ref{eq:d22}) gives
\begin{eqnarray}
\xi_{A} 1 = \xi_{A} \hat{\rho}_{f \mp}^{(A)} 1 \;\;\;,
\label{eq: condition kin susy}
\end{eqnarray}
with index $A$ not summed.
In order to proceed further, we rewrite eq.~(\ref{eq:prv}) explicitly as
\begin{eqnarray}
\hat{\rho}_{b \mp}^{(M)} &\equiv& \Theta (M\in {\cal M}_{-}) \hat{\rho}_{-}
+ \Theta (M\in {\cal M}_{+}) \hat{\rho}_{+} \nonumber \\
\hat{\rho}_{f \mp}^{(A)} &\equiv& \Theta (A\in {\cal A}_{-}) \hat{\rho}_{-}
+ \Theta (A\in {\cal A}_{+}) \hat{\rho}_{+} \;\;\; ,
\label{eq: explicit proj}
\end{eqnarray}
where
\begin{equation}
{\cal M}_{-} \cup {\cal M}_{+}
= \{ \{ \; 0,1,2,3,4,5,6,7,8, 9 \; \} \} \;\; ,
\;\;
{\cal M}_{-} \cap {\cal M}_{+} = \phi \;\;,
\label{eq: M condition}
\end{equation}
\begin{equation}
{\cal A}_{-} \cup {\cal A}_{+}
= \{ \{ \; 1,2,5,6,9,10,13,14,19,20,23,24,27,28,31,32 \; \} \} \;\; ,
\;\;
{\cal A}_{-}\cap {\cal A}_{+}=\phi \; .
\label{eq: A condition}
\end{equation}
We find that eq.~(\ref{eq:1}) gives
\begin{eqnarray}
\label{eq:set1}
\left( \bar{\epsilon} \Gamma_{M_{-}} \right)_{A_{+}} = \left( \bar{\epsilon}
\Gamma_{M_{+}} \right)_{A_{-}} =0 \;\;,
\end{eqnarray}
while eq.~(\ref{eq:2}) gives
\begin{eqnarray}
\label{eq:set2}
\left( \Gamma^{M_{-}N_{+}} \epsilon \right)_{A_{-}} &=& 0 \nonumber \\
\left( \Gamma^{M_{-}N_{-}} \epsilon \right)_{A_{+}} &=&
\left( \Gamma^{M_{+}N_{+}} \epsilon \right)_{A_{+}} =0 \;\;.
\end{eqnarray}
Equation (\ref{eq: condition kin susy}) gives
\begin{eqnarray}
\label{eq:am}
\xi_{A_{-}} = 0 \;\;.
\end{eqnarray}
As we consider the case of eight kinematical
supersymmetries, the number of elements of the sets
denoted by $\mbox{\boldmath $\sharp$} ({\cal A}_{\pm})$ must be
\begin{equation}
\mbox{\boldmath $\sharp$} ({\cal A_{-}}) = 8 \;\; \mbox{ and } \;\;
\mbox{\boldmath $\sharp$}({\cal A_{+}}) = 8
\label{eq:number of A+A-} \;.
\end{equation}
Eqs.~(\ref{eq:set1}) and (\ref{eq:set2}) are regarded as the ones which
determine the anticommuting parameter $\epsilon$, and
the sets ${\cal A}_{+}$, ${\cal A}_{-}$, ${\cal M}_{+}$ and
${\cal M}_{-}$. In addition they must satisfy the conditions
(\ref{eq: M condition}), (\ref{eq: A condition}) and
(\ref{eq:number of A+A-}).
We search for solutions by first trying out as an input an appropriate
thirty-two component anticommuting parameter $\epsilon$
satisfying Majorana-Weyl condition.
Given $\epsilon$, we see if we can determine ${\cal A}_{+}$, ${\cal A}_{-}$,
${\cal M}_{+}$ and ${\cal M}_{-}$ successfully.
We have tried out many cases.
The case leading to our model is
\begin{equation}
\epsilon = (\epsilon_0, 0, \epsilon_1, 0, 0,0,0,0, 0, \bar{\epsilon}_0, 0 , \bar{\epsilon}_1, 0,0,0,0 )^t \;\; .
\end{equation}
Note that $\epsilon_0$, $\epsilon_1$, $\bar{\epsilon}_0$ and $\bar{\epsilon}_1$ are two-component
anticommuting parameters.
We have found out that
\begin{eqnarray}
\hat{\rho}_{b\mp}
&=& diag
(\hat{\rho}_{-},\hat{\rho}_{-},\hat{\rho}_{-},
\hat{\rho}_{-},\hat{\rho}_{-},\hat{\rho}_{+},
\hat{\rho}_{+},\hat{\rho}_{-}, \hat{\rho}_{+},\hat{\rho}_{+} )
\nonumber \\
\hat{\rho}_{f\mp} &=&
\hat{\rho}_{-} 1_{(4)} \otimes
\left( \begin{array}{cccc}
1_{(2)}& & & \\
& 0 & & \\
& & 1_{(2)} & \\
& & &0
\end{array}
\right)
+
\hat{\rho}_{+} 1_{(4)} \otimes
\left( \begin{array}{cccc}
0& & & \\
& 1_{(2)} & & \\
& & 0 & \\
& & & 1_{(2)}
\end{array}
\right)
\;\; ,
\end{eqnarray}
is a solution to the above set of equations. We adopt this choice
as the projectors of our model.
We have found only one solution other than this one, which is given
by
\begin{equation}
\epsilon = (\epsilon_0, 0, \epsilon_1, 0, 0,0,0,0, 0, 0,0,0,0, \bar{\epsilon}_2, 0 , \bar{\epsilon}_3 )^t \;\; .
\label{eq:80susy para}
\end{equation}
The consistent sets
\begin{eqnarray}
{\cal A}_{-} &=& \{ \{ \; 1,2,5,6,27,28,31,32 \; \} \} \;\; ,
\nonumber\\
{\cal A}_{+} &=& \{ \{ \; 9,10,13,14,19,20,23,24 \; \} \} \;\; ,
\label{eq:A-A+08}
\end{eqnarray}
\begin{eqnarray}
{\cal M}_{-} &=& \{\{ \; 4,7 \;\}\} \;\; ,
\nonumber\\
{\cal M}_{+} &=& \{\{ \; 0,1,2,3,5,6,8,9 \;\}\} \;\;.
\label{eq:M-M+08}
\end{eqnarray}
have been obtained.
The projectors (\ref{eq: explicit proj}) are
\begin{eqnarray}
\hat{\rho}_{b\mp}
&=& diag
(\hat{\rho}_{+},\hat{\rho}_{+},\hat{\rho}_{+},
\hat{\rho}_{+},\hat{\rho}_{-},\hat{\rho}_{+},
\hat{\rho}_{+},\hat{\rho}_{-}, \hat{\rho}_{+},\hat{\rho}_{+} )
\nonumber \\
\hat{\rho}_{f\mp} &=&
\hat{\rho}_{-} 1_{(4)} \otimes
\left( \begin{array}{cccc}
1_{(2)}& & & \\
& 0 & & \\
& & 0 & \\
& & & 1_{(2)}
\end{array}
\right)
+
\hat{\rho}_{+} 1_{(4)} \otimes
\left( \begin{array}{cccc}
0& & & \\
& 1_{(2)} & & \\
& & 1_{(2)} & \\
& & & 0
\end{array}
\right)
\;\; .
\label{eq:2-8 case projector}
\end{eqnarray}
This is the case considered in ref. \cite{Mtwist,Ba}
in the context of $M$ theory compactification to
the lightcone heterotic strings.
The spinorial parameters $\epsilon_0$, $\epsilon_1$, $\bar{\epsilon}_2$ and $\bar{\epsilon}_3$ in
eq. (\ref{eq:80susy para}) are all real, however, and the closed string sector
obtained from this choice is not regarded as a projection from
the $IIB$ matrix model.
\subsection{closed and open loops and Chan-Paton symmetry}
Loop variables play a decisive role in the second
quantized formulation of the theory; they eventually act as string
fields. This will be discussed in
section $5$. We include the part of
the discussion here to appreciate the role played by the degrees of
freedom in the (anti-)fundamental representation and
the attendant Chan-Paton symmetry.
\subsubsection{adding fundamentals and flavor symmetry}
It is well known that nonorientable closed strings by themselves
are not consistent. We need to add degrees of freedom corresponding to
open strings, keeping $8+8$ susy. Let us, therefore, consider the
following ``hypermultiplet'' in the fundamental/antifundamental
representation of $usp(2k)$:
\begin{eqnarray}
\left( Q_{i}, \tilde{Q}^{i}, \psi_{Q}{}_{i}, \psi_{\tilde{Q}}^{i}
\right) \;\;.
\end{eqnarray}
The number of this multiplet is denoted as $n_{f}$.
To make this ``flavour''symmetry (= local gauge symmetry of strings) manifest,
let us introduce complex $2n_{f}$ dimensional vectors
\begin{eqnarray}
{\bf Q} \equiv \left\{ \begin{array}{ll}
Q_{(f)}\;\;, & f=1 \sim n_{f} \\
F^{-1} \tilde{Q}_{(f- n_{f})}\;\;, & f=n_{f} +1 \sim 2n_{f} \;,
\end{array}
\right. \;\;\;
{\bf Q}^{*} \equiv \left\{ \begin{array}{ll}
Q_{(f)}^{*} \;\;, & f=1 \sim n_{f} \\
\tilde{Q}_{(f- n_{f})}^{\ast} F \;\;, & f=n_{f} +1 \sim 2n_{f} \;.
\end{array}
\right.
\end{eqnarray}
Similarly,
\begin{eqnarray}
\mbox{\boldmath$\psi_{Q}$} \equiv \left\{ \begin{array}{ll}
\psi_{ Q_{(f)}} \;\;, & f=1 \sim n_{f} \\
F^{-1} \psi_{\tilde{Q}_{(f- n_{f})} } \;\;, & f=n_{f} +1 \sim 2n_{f} \;,
\end{array}
\right.
\mbox{\boldmath$\psi_{Q}$}^{\ast} \equiv \left\{ \begin{array}{ll}
\overline{\psi}_{Q_{(f)} } \;\;, & f=1 \sim n_{f} \\
\overline{\psi}_{\tilde{Q}_{(f- n_{f})} } F \;\;, & f=n_{f} +1
\sim 2n_{f} \;.
\end{array}
\right.
\end{eqnarray}
We denote the $f$-th components of these vectors by ${\bf Q}_{(f)}$
etc.
\subsubsection{Choice of variables}
Let us first introduce a discretized path-ordered exponential
which represents a configuration of a string in momentum superspace:
\begin{eqnarray}
U[p^{M}_{.},\eta_{.};n_1,n_0] \equiv
P \exp (-i \sum_{n=n_0}^{n_1} (p^{M}_{n} v_{M}+\bar{\eta}_{n} \Psi))
= \stackrel{\leftarrow}{\prod_{n=n_0}^{n_1}}
\exp (-i p^{M}_{n} v_{M}-i \bar{\eta}_{n} \Psi)\;\;, \nonumber
\end{eqnarray}
where $p^{M}_{n}$ and $\eta_n$ are respectively the sources
or the momentum distributions for $v_{M}$ and those for $\Psi$.
The closed loop is then defined by
\begin{equation}
\Phi[p^{M}_{.},\eta_{.};n_1,n_0] \equiv Tr U[p^{M}_{.},\eta_{.};n_1,n_0]\;\;.
\end{equation}
To consider an open loop, let us introduce
$\Xi = \left( \xi, \xi^{*} \right)$ as bosonic sources for $\mbox{\boldmath$Q$}_{(f)}$
and $\mbox{\boldmath$Q$}_{(f)}^{\ast}$, and $\Theta = \left(\theta, \bar{\theta} \right)$
as Grassmannian ones for $\mbox{\boldmath$\psi_{Q}$}_{(f)}$ and $\mbox{\boldmath$\psi_{Q}$}_{(f)}^{\ast}$:
$ \left( \Xi \mbox{\boldmath$\Omega$}_{(f)} \right) = \xi \mbox{\boldmath$Q$}_{(f)}
+F^{-1} \xi^{\ast}\mbox{\boldmath$Q$}_{(f)}^{\ast}$,
$\left( \Theta \mbox{\boldmath$\Upsilon$}_{(f)} \right) = \theta \mbox{\boldmath$\psi_{Q}$}_{(f)}
+F^{-1} \bar{\theta} \mbox{\boldmath$\psi_{Q}$}^{\ast}_{(f)}.$
We write these collectively as
\begin{equation}
\left( \Lambda \mbox{\boldmath$\Pi$}_{(f)} \right) \equiv \left( \Xi \mbox{\boldmath$\Omega$}_{(f)}
\right)
+
\left( \Theta \mbox{\boldmath$\Upsilon$}_{(f)} \right)\;\;.
\end{equation}
The open loop is defined by
\begin{equation}
\Psi_{f'f}[k^{m}_{.},\zeta;l_1,l_0;\Lambda',\Lambda]
\equiv \left( \Lambda' \mbox{\boldmath$\Pi$}_{(f')} \right)
F U[k^{m}_{.},\zeta_{.};l_1,l_0]
\left( \Lambda \mbox{\boldmath$\Pi$}_{(f)} \right) \;\;,
\end{equation}
where $f$ and $f'$ are the Chan-Paton indices.
The open and closed loops generate all of the observables in the theory
under question.
We now turn to the question of the nonorientability of the
closed and the open loops.
Using $v^{t}_{M}=\mp F v_{M} F^{-1}, \Psi^{t}= \mp F \Psi F^{-1},$
and $F^{t}= -F$, we readily obtain
\begin{eqnarray}
\Phi[p^{M}_{.},\eta_{.};n_1,n_0]
=Tr(\stackrel{\rightarrow}{\prod_{n=n_0}^{n_1}}
\exp (-i p^{M}_{n} v^{t}_{M}-i \bar{\eta}_{n} \Psi^{t}))
=\Phi[\mp p^{M}_{.},\mp \eta_{.};n_0,n_1]\;\;,
\label{csnonori}
\end{eqnarray}
and
\begin{equation}
\Psi_{f'f}[k^{M}_{.},\zeta_{.};l_1,l_0;\Lambda',\Lambda]
=-\Psi_{ff'}[\mp k^{M}_{.},\mp \zeta_{.};l_0,l_1;\Lambda,\Lambda']\;\;.
\label{osnonori}
\end{equation}
These equations relate a string configuration to the one with its
orientation and the Chan-Paton factor reversed and drawn pictorially
as
\begin{figure}[t]
\epsfysize=3cm
\centerline{\epsfbox{fig1.eps}}
\label{fig1}
\end{figure}
\begin{figure}[t]
\epsfysize=3cm
\centerline{\epsfbox{fig2.eps}}
\label{fig2}
\end{figure}
The minus signs in front of
$p^{m}_{.}$, $\eta_{.}$, $k^{m}_{.}$ and $\zeta_{.}$ in eq.
(\ref{csnonori}) and eq. (\ref{osnonori}) reflect the orientifold
structure of the $USp(2k)$ matrix model.
The overall minus sign in the last line of (\ref{osnonori})
is of interst: it comes from $F^{t}=-F$ of the $usp$ Lie algebra
and implies the $SO(2 n_f)$ gauge group. This is the cleanest one
of the three rationales for the choice of the $usp$ Lie algebra:
we present this as a table.
\bigskip
\begin{tabular}{ccccc}
& & & original(worldsheet) & \\
& C-P factor & & Lie algebra & \\
\hline
$-$ $\colon$ & $so(2n_{f})$ & $\Leftrightarrow$ & $usp(2k)$ &
$\Leftarrow$ our choice \\
$+$ $\colon$ & $usp(2n_{f})$ & $\Leftrightarrow$ & $so(2k)$ &
\end{tabular}
\bigskip
In order to inherit the infrared stability of perturbative vacua
\cite{GS,IMo} of superstrings, the model must be based on
the $usp$ as opposed to the $so$ Lie algebra and $n_f =16$.
This latter property also follows from the anomaly cancellation
of the $6D$ worldvolume gauge theory \cite{IT2}, which we will
discuss in subsection $3.2$.
\subsection{the model}
\subsubsection{the action of the $USp$ matrix model}
We finally come to the action of our $USp(2k)$ reduced matrix model.
It is obtained from the dimensional reduction of
${\cal N}=2, d=4$ $USp(2k)$ supersymmetric gauge theory with one
hypermultiplet in the antisymmetric representation and $n_{f}$ hypermultiplets
in the fundamental representation. This makes manifest the presence of the
eight dynamical supercharges.
In the ${\cal N}=1$ superfield notation with
spacetime dependence all dropped, we have a vector superfield $V$ and a
chiral superfield $\Phi \equiv \Phi_{1}$ which are $usp$ Lie algebra valued
\begin{eqnarray}
\label{eq:adj}
V^{t} = - FV F^{-1} \;\;, \;\; \Phi^{t} = - F \Phi F^{-1} \;\;,
\;\; V^{\dagger} = V \;\;, \;\; {\rm with} \;\; F =
\left(
\begin{array}{cc}
0 & I \\
-I & 0
\end{array} \right)\;\;,
\end{eqnarray}
and the two chiral superfields $\Phi_{I} ,\; I =2,3$ in the antisymmetric
representation which obey
\begin{eqnarray}
\label{eq:asym}
\Phi_{I}^{t}= F \Phi_{I} F^{-1} \;\;\; {\rm for\; I=2,3}\;\;.
\end{eqnarray}
We can write
$V = \hat{\rho}_{-} \underline{V},\;\; \Phi_{1} = \hat{\rho}_{-}
\underline{\Phi_{1}} ,\;\;
\Phi_{I} = \hat{\rho}_{+} \underline{\Phi_{I}} , \;\; I = 2,3.$
The total action is written as
\begin{eqnarray}
\label{eq:vvaym}
S &=&
\frac{1}{4 g^{2}} \; Tr \left(
\int d^2 \theta W^{\alpha} W_{\alpha} + h.c. +
4 \int d^2 \theta d^2 \bar{\theta} \Phi^{\dagger}_{I} e^{2V} \Phi_{I} e^{-2V} \right) \\
&+& \frac{1}{g^{2}} \sum_{f=1}^{n_f}
\int d^2 \theta d^2 \bar{\theta}
\left( Q_{(f)}^{*} \left( e^{2V} \right) Q_{(f)}
+ \tilde{Q}_{(f)} \left( e^{-2V} \right) \tilde{Q}_{(f)}^{*} \right)
+ \frac{1}{g^{2}} \left( \int d^2\theta W ( \theta) + h.c. \;\right)\;,
\nonumber
\end{eqnarray}
where the superpotential is
\begin{eqnarray}
W ( \theta) = \sqrt{2} Tr \left(
\Phi_{1} \left[ \Phi_{2}, \Phi_{3} \right] \right)
+ \sum_{f=1}^{n_{f}} \left( m_{(f)} \tilde{Q}_{(f)} Q_{(f)}
+ \sqrt{2} \tilde{Q}_{(f)} \Phi_{1}Q_{(f)} \right) \;\;\;.
\end{eqnarray}
It is of interest to to express $S$ as
\begin{equation}
S= S_{close} + \Delta S \;\;.
\end{equation}
Fot this, we need to
render $S$ to its component form.
Solving the equation for the $D$ term, we obtain
\begin{eqnarray}
D = \left[ \Phi^{\dagger}_{I}, \Phi_{I}\right] - \hat{\rho}_{-}
\sum_{f=1}^{n_{f}} \left( Q_{(f)} Q_{(f)}^{*} -
\tilde{Q}_{(f)}^{*} \tilde{Q}_{(f)} \right) \;\;,
\end{eqnarray}
where we have placed the $USp$ vectors $Q_{(f)},\; \tilde{Q}_{(f)}$ and their
complex conjugates in the form of dyad. The $F$ terms are such that
\begin{eqnarray}
- \delta W = \sum_{I=1,2,3} tr F^{\dagger}_{\Phi_{I}} \delta \Phi_{I}
+ F^{*}_{Q_{(f)}} \delta Q_{(f)}
+ F^{*}_ {\tilde{Q}_{(f)}} \delta \tilde{Q}_{(f)} \;\;\;.
\end{eqnarray}
Explicitly
\begin{eqnarray}
\label{eq:F}
F^{\dagger}_{\Phi_{1}} &=& - \sqrt{2} \left[ \Phi_{2}, \Phi_{3}\right]
- \sqrt{2} \hat{\rho}_{-} \left( \sum_{f=1}^{n_{f}} Q_{(f)}
\tilde{Q}_{(f)} \right),\;
F^{\dagger}_{\Phi_{2}} = - \sqrt{2} \left[ \Phi_{3}, \Phi_{1}\right],
\;\;
F^{\dagger}_{\Phi_{3}} = - \sqrt{2} \left[ \Phi_{1}, \Phi_{2}\right] \;,
\nonumber \\
F^{*}_{Q_{(f)}} &=& - \left( m_{(f)} \tilde{Q}_{(f)} + \sqrt{2}
\tilde{Q}_{(f)} \Phi_{1} \right) \;,\;\;
F^{*}_{\tilde{Q}_{(f)}} = - \left( m_{(f)} Q_{(f)} + \sqrt{2} \Phi_{1}
Q_{(f)} \right) \;\;.
\end{eqnarray}
As for the Yukawa couplings, they can be read off from
\begin{eqnarray}
\delta^{2} W \equiv \sum_{A,B} \frac{\partial^{2}W}{\partial A \partial B}
\delta A \delta B \;\;,
\end{eqnarray}
where the summmation indices $A,\;B$ are over all chiral superfields
$\Phi_{I}\; I=1,2,3,$ and $Q_{(f)}, \tilde{Q}_{(f)}\;, \;
f= 1, \cdots n_{f}$.
Using the complex $2n_{f}$ dimensional vectors and spinors introduced before,
we find, after some algebras,
\begin{eqnarray}
\Delta S &=& \Delta S_{b} + \Delta S_{f} =
\left( S_{g-s} + {\cal V}_{scalar} + S_{mass} \right)
+ \left( S_{g-f} + S_{Yukawa} \right) \;\;, \\
S_{g-s} &=& -\frac{1}{g^{2}} tr \left( \sum_{\nu = 0,1,2,3,4,7}
v_{\nu} v^{\nu} + \sum_{I=2,3} \left[ \Phi_{I}, \Phi^{\dagger I} \right]
\right) {\bf Q} \cdot {\bf Q}^{*} \nonumber \\
&+& \frac{1}{g^{2}} tr \left[ \Phi_{2}, \Phi_{3} \right] F^{-1}
{\bf Q}^{*} \cdot \Sigma {\bf Q}^{*}
- \frac{1}{g^{2}} tr \left[ \Phi_{2}^{\dagger}, \Phi_{3}^{\dagger}
\right]
{\bf Q} \cdot \Sigma F {\bf Q} \;\;, \\
S_{mass} &=&
- \frac{1}{g^{2}} tr \left( {\bf Q} \cdot M^{2} {\bf Q}^{*} \right)
-\frac{2}{g^{2}} tr \left(v_{4} {\bf Q} \cdot M {\bf Q}^{*} \right)\;\;, \\
{\cal V}_{scalar} &=& - \frac{1}{2g^{2}} tr {\bf Q} \cdot
\Sigma {\bf Q}
{\bf Q}^{*} \cdot \Sigma {\bf Q}^{*}
- \frac{1}{8g^{2}} tr \left[
{\bf Q} \cdot {\bf Q}^{*} - F^{-1} {\bf Q}^{*} \cdot {\bf Q} F \right]^{2}
\;, \\
S_{g-f} &=& \frac{1}{g^{2}} \left\{ \mbox{\boldmath$\psi_{Q}$}^{\ast}
\overline{\sigma}^{m} v_{m} \cdot \mbox{\boldmath$\psi_{Q}$}
+ i \sqrt{2} {\bf Q}^{*} \lambda \cdot
\mbox{\boldmath$\psi_{Q}$} - i \sqrt{2} \mbox{\boldmath$\psi_{Q}$}^{\ast}
\overline{\lambda} \cdot {\bf Q} \right\} \;, \\
S_{Yukawa} &=& - \frac{1}{g^2}\left\{ \sum_{(c_{1}, c_{2})=
(Q, \tilde{Q}), (Q, \Phi_{1}), (\Phi_{1},\tilde{Q})}
\frac{\partial^{2} W_{matter}}
{\partial C_{1} \partial C_{2}} \psi_{C_{2}} \psi_{C_{1}} + h.c. \right\}
\nonumber \\
&=& \frac{1}{g^2}\left( \frac{1}{2} \mbox{\boldmath$\psi_{Q}$} \cdot \Sigma F
\left( \sqrt{2} \Phi_{1} + M \right) \mbox{\boldmath$\psi_{Q}$}
+ \sqrt{2} {\bf Q} \cdot \Sigma F \psi_{\Phi_{1}} \mbox{\boldmath$\psi_{Q}$}
+ h.c. \right) \;\;.
\end{eqnarray}
Here
\begin{eqnarray}
\Sigma &\equiv& \left(
\begin{array}{cc}
0 & I \\
I & 0
\end{array} \right)\;\;, \\
M &\equiv& {\rm diag} \left( m_{(1)}, \cdots, m_{(n_f)}
- m_{(1)}, \cdots, -m_{(n_f)} \right) \;\;, \\
W_{{\rm matter}} &=& \sum_{f=1}^{n_{f}} \left( m_{(f)}
\tilde{Q}_{(f)} Q_{(f)}+ \sqrt{2} \tilde{Q}_{(f)} \Phi Q_{(f)} \right)\;\;,
\end{eqnarray}
and $\cdot$ implies the standard inner product with respect to the
$2n_{f}$ flavour indices.
\subsection{Notion of spacetime points}
We have seen the validity and the rationales of the
$USp(2k)$ reduced matrix model
as a constructive model descending from type $I$ superstrings.
As we will see, the model is also dynamical theory of
spacetime points..
As is mentioned at the introduction, the dynamical degrees of freedom
of spacetime points $X_{M}^{(i)}$ are embedded in the matrices.
We write
\begin{eqnarray}
v_{M} = X_{M}+ \tilde{v}_{M} \;\;\;,
\end{eqnarray}
where $\tilde{v}_{M}$ are off-diagonal matrices.
One can imagine integrating out the bosonic off-diagonals
and the fermions. This will give dynamics
to the spacetime points. The object we study is the effective action for
the spacetime points \cite{IM,CIK,AIKKT,HNT}:
\begin{equation}
\ln Z\left[ X_{M}^{(i)} \right] \;\;.
\end{equation}
It is sometimes instructive to study the case in which the spacetime points
are assumed to interact weakly and are widely separated.
In this case we can approximate the $USp$ matrices by
individual $SU(2)$ blocks. We will find below that this approximation
has a direct connection to string theory results obtained
from the worldvolume gauge theories in various dimensions.
\section{Matrix $T$ dual and representation as worldvolume gauge theories}
Before going into the subject of this section, let us note that
the classical solution of the model which gives vanishing action
is broken by $Z_{2}$ for six of the adjoint directions.
In fact, the commutator of $\tilde{\delta} \equiv \delta^{(1)}
+ \delta^{(2)}$ with itself closes into translation only for
four of the antisymmetric matrices as is clear from eq. (\ref{eq:am}).
This in fact means the presence of
the $O3$ fixed surfaces.
The discussions in later sections convince ourselves of
the presence of $D3$ branes as well in the original representation.
These appear in such a way
that the total $RR$ flux of the system cancels \cite{Pol}.
\subsection{general remarks and the series of degenerate vacua associated
with the model}
What we would like to put forward with the reduced matrix model is
a constructive approach to unified string theory. We presume that
the lifting of perturbative vacua requires genuinely nonperturbative
mechanism and this can only be accomplished by finding the proper scaling limit
of the model, which is a difficult task at this moment.
As a separate theme of research,
we try to establish the consistency of the $USp$ matrix model with
string perturbation theory in the presence of D-branes
(semiclassical object) via the worldvolume representation.
In particular, one can think of toroidal compactifications in various
dimensions via the recipe of \cite{WT}.
While some nonperturbative effects can be
seen as exact results on worldvolume gauge theories, this representation
leaves aside the original goal of lifting perturbative string vacua.
In fact, one naively sets $k\rightarrow
\infty $ and has no hope of capturing the true vacuum.
The procedure of matrix toroidal compactification is well known
and will not be reviewed here. Instead, we just tabulate the relevant
properties and the correspondence with the ``classical'' counterparts.
We have seen in the last section that $F$ is a matrix counterpart of
the twist operation $\Omega$.
The matrix $T$ dual is nothing but a Fourier transform $\hat{T}$.
We can consider the combined transformation of these.
We have observed in \cite{IT2}
that the sign flip occurs provided the worldvolume
gauge fields are odd under parity.
These are summarized as a table.
\noindent
\begin{tabular}{|c|c|c|}
\hline
& "classical" & matrices \\
\hline
twist & $\Omega$ & $F$ \\
\hline
O3 fixed & $X_{\nu}(\bar{z}, z) = - \Omega X_{\nu}(z, \bar{z})
\Omega^{-1}$ &
$v_{\nu}^{t} = - Fv_{\nu}F^{-1}$ \\
surface & $X_{I}(\bar{z}, z) = + \Omega X_{I}(z, \bar{z}) \Omega^{-1}$ &
$v_{I}^{t} = + Fv_{I}F^{-1}$ \\
\hline
$T$ transf & $\hat{T}[X_{M}] \equiv X_{M \ \mbox{R}}(z) - X_{M
\mbox{L}}(\bar{z})$ &
$\hat{T}((v_{M})_{\vec{a}, \vec{b}}) = <x|\hat{v}_{M}|x>$ \\
& sign flip & sign flip OK \\
& under $\Omega T$ & if $\tilde{v}_{\nu}(\vec{x}) = -
\tilde{v}_{\nu}(\vec{x})$ \\
\hline
\end{tabular}
\noindent
From matrix toroidal compactification,
we find a series of degenerate perturbative vacua associated with
the $USp(2k)$ matrix model.
In the remainder of this section, we will see the consistency
of the worldvolume representation of the $USp$ matrix model in various
dimensions with some literature. When some of the adjoint
directions get compactified and become small, it is preferrable to T-dualize
the system into these diretions.
One can imagine this as a figure:
\begin{figure}[t]
\centerline{\epsfbox{fig3.eps}}
\label{fig3}
\end{figure}
\subsection{6d,5d,4d worldvolume representations}
We now look at the $T$ dualized form of the $USp$ matrix model
for the cases of the $6d,5d,4d$ worldvolume theories, which
respectively represent type$I$ theory
in ten dimensions, its orientifold compactification
on $S^{1}/Z^{2}$ ($9d$ string theory) and that on $T^{2}/Z^{2}$
($8d$ string theory). We will demand that all of the
degenerate perturbative vacua discussed above be infrared stable.
We will find out
$n_{f}=16$ and in each case we distribute these evenly around
the fixed surfaces.
\subsubsection{6d}
For consistency, we have added the fields lying in
the fundamental representation.
They are responsible for creating an open string sector.
In string perturbation theory, the infrared stability is seen
through the (global) cancellation of dilaton tadpoles between disk and
$RP^{2}$ diagrams \cite{GS}, \cite{IMo}, leading to the $SO(32)$ Chan-Paton
factor. This survives toroidal compactifications with/without discrete
projection \cite{AIKT}.
We have found out that in the case of all six adjoint directions
compactified, the infrared stability gets translated into
the consistency of the six dimensional worldvolume $USp(2k)$ gauge theory
with matter in the antisymmetric and fundamental representations.
In fact, by acting
\begin{eqnarray}
\Gamma_{(6)} \equiv \Gamma^{0}\Gamma^{1}\Gamma^{2}\Gamma^{3}
\Gamma^{4} \Gamma^{7}
\end{eqnarray}
on $\Psi$, we see that the adjoint fermions $\lambda$ and $\psi_{1}$ have
chirality plus while $\psi_{2,3}$ have chirality minus. The fermions in the
fundamental representation have chirality minus.
The standard technology
to compute nonabelian anomalies is provided by family's index theorem
and the descent equations.
We find that the condition for the anomaly cancellation:
\begin{eqnarray}
& & tr_{adj}F^{4} - tr_{asym} F^{4} - n_{f} tr F^{4} \;\; \nonumber \\
& & = (2k+8) trF^{4} + 3 \left( trF^{2}\right)^{2}
- \left( \left(2k-8\right) trF^{4} + 3 \left( trF^{2}\right)^{2}
\right)
- n_{f} trF^{4} \nonumber \\
& & = \left( 16-n_{f} \right) tr F^{4} =0 \;\;\;,
\end{eqnarray}
where we have indicated the traces in the respective representations.
The case $n_{f}=16$ is selected by the consistency of the theory.
In the case discussed in eq. (\ref{eq:2-8 case projector}),
we conclude from similar calculation that
the anomaly cancellation of the worldvolume two-dimensional
gauge theory selects sixteen complex fermions.
This result is reasonable from the ``classical'' consideration
of the $RR$ flux. We know that the flux is from $O3$ to $n_{f}$ of
$D3$ and their mirror and is in the adjoint directions.
When all of the six adjoint directions are compactified,
the flux cannot escape to infinity and the total flux had better vanish.
In this way, we also get $n_{f}=16$.
Although we will not discuss here, a version of anomaly inflow argument is
operating which relates the conservation of the $RR$ flux to
the nonabelian anomaly.
\subsubsection{space-dependent (axion-) dilaton background field}
The simplest quantity to be computed in the worldvolume representation of
matrix models in general is the effective running coupling constant
$g^{2}_{eff}(u)$ obtained from the low-energy effective action.
Here $\vec{u}$ is a $vev$ of the scalars which labels quantum moduli space.
The $g^{2}_{eff}(\vec{u})$ represents the marginal scalar
deformation of the original action to a type of nonlinear $\sigma$ model.
The background field appearing through this procedure is a massless
(axion-)dilaton field. The running coupling constant is
identified as the space-dependent (axion-) dilaton background field.
One strategy to compute $g^{2}_{eff}(u)$ is
$i)$ to begin with computing the part of the effective action associated with
the fermionic integrations, $ii)$ to invoke supersymmetry to find
out $S_{eff}$, and $iii)$ to take the hessian of $S_{eff}$ to obtain
$g^{2}_{eff}(u)$.
\subsubsection{5d}
Let us first consider the widely separated case, namely,
$SU(2)$ case with eight flavours per fixed surface.
$X_{7}=u$ and $X_{5,6,8,9}$ decouple.
Omitting the derivation, we write the result \cite{Seiberg}
\begin{eqnarray}
\frac{1}{g^{2}_{eff} (u)} = \frac{1}{g^{2}} + 16X_{7}
- \sum_{f=1}^{8} \mid X_{7}- m_{f} \mid -
\sum_{f=1}^{8} \mid X_{7}+ m_{f} \mid \;\;.
\end{eqnarray}
(See also \cite{PW}.)
The $USp(2k)$ matrix model provides a natural generalization
to this. The answer this time depends on the $k$ spacetime points
$X_{7}^{(i)}$ and on those of the antisymmetric directions
$X_{5,6,8,9}^{(i)}$. We have completed the first step of the
computation \cite{AI} and the answer appears to reflect that the
spacetime is curved in the $X_{5,6,8,9}^{(i)}$ directions.
\subsubsection{4d}
It has been shown that the model is able to describe
\cite{IT1,IT2} the $F$ theory compactification on an elliptic fibered
$K3$ \cite{V,Sen}. We will not repeat the discussion here.
In the widely separated case
($SU(2)$ with four flavours and
$u= X_{4}+ iX_{7}$ ), the $4d$ worldvolume theory obtaind
is exactly Sen's scaling limit to $F$ on $K3$ \cite{Sen}.
The space-dependent
axion/dilaton background field is controlled by the Seiberg-Witten curve.
The model again provides an interesting generalization to
the $USp(2k)$ case and much remains to be worked out.
\section{Formation of extended objects from (non)-abelian Berry phase}
We now show that the model contains degrees of freedom
corresponding to D-objects. We study the effective action for the
spacetime points.
In particular, we study the effects of fermionic integrations
which contain the information of the $RR$ sector to deduce the
coupling of D-objects to spacetime. This will be the effect which
survives the cancellation of the bosonic
integrations against the fermionic ones. We will suppress here the
nature of time in the reduced model as dynamical variable in order to
form a loop in the parameter space ($=$ spacetime points.)
We will study the coupling in the representation of the model as
$T$ dualized quantum mechanics via the Berry phase.
\subsection{path-dependent effective action}
We will make the effective action dependent on
the paths $\{ \Gamma_{A}^{(R)} \}$ in the parameter space
labelled by the five sets of the adjoint spacetime points
$X_{\nu} = diag(X_{\nu}^{(1)}, \cdots X_{\nu}^{(k)},$
$ -X_{\nu}^{(1)}, \cdots -X_{\nu}^{(k)}.)$ $\nu =1,2,3,4,7$.
We choose $v_{0}=0$ gauge.
\begin{eqnarray}
\label{eq:adprocess}
Z \left[ X_{\nu}; m_{(f)}, \{\Gamma_{A}^{(R)} \},
\{ \sigma_{i A}^{(R)} \}, \{ \sigma_{f A}^{(R)} \}
\right] = \int \left[ D \tilde{v}_{M} \right] \prod_{f=1}^{n_{f}}
\left[ D Q_{(f)}\right] \left[ D Q_{(f)}^{*}\right]
\left[D {\tilde{Q}}_{(f)} \right] \nonumber \\
\left[ D \tilde{Q}_{(f)}^{*} \right] e^{iS_{B}}
\lim_{T \rightarrow \infty} \langle \Psi ; \{ \sigma_{i A}^{(R)} \}
\mid P \exp
\left[ -i \int_{0}^{T} dt H_{fermion}(t) \right]
\mid \Psi ; \{ \sigma_{f A}^{(R)} \} \rangle \;.
\end{eqnarray}
Here $v_{M} = X_{M}+ \tilde{v}_{M}$ and $S^{B}$ is the pat of $S$
which does not contain any fermion. For simplicity, we have set the
dependence of the remaining antisymmetric spacetime points
$X_{I}= diag(X_{I}^{(1)},
\cdots X_{I}^{(k)},$$ X_{I}^{(1)}, \cdots X_{I}^{(k)})$ $\;I= 5,6,8,9$
to zero.
The operator $H_{fermion}$ is the sum of the respective
Hamiltonians $H_{{\rm fund}},H_{{\rm adj}}$ and $H_{{\rm asym}}$ obtained
from the fermionic part of
$ S_{{\rm fund}}, S_{{\rm adj}}$ and $S_{{\rm asym}}$ after $T$ duality.
Their $t$ dependence comes from that of $X_{\nu}$ which acts as external
parameters on the Hilbert space of fermions.
We consider a set of degenerate
eigenstates. The degeneracy of the initial state and that of the final one
are respectively specified by a set of labels $\{ \sigma_{i A}^{(R)} \}$
and $\{ \sigma_{f A}^{(R)} \}$, where the indices $A$ and $(R)$ specify
the species of fermions.
\subsection{reduction to the first quantized problem}
Let $R = {\rm fund, adj, antisym}$. We denote by $e_{(R)}^{(A)}$
the standard eigenbases belonging to the roots of $sp(2k)$ and the weights
of the fundamental representation and those of the antisymmetric
representation respectively.
Let us expand the two component fermions as
\begin{eqnarray}
\label{eq:expand}
\psi^{(R)} = \sum_{A}^{N_{(R)}} b_{A}^{(R)}
e_{(R)}^{(A)} /\sqrt{2} \;,\;\;
\bar{\psi}^{(R)} = \sum_{A}^{N_{(R)}} \bar{b}_{A}^{(R)}
e_{(R)}^{(A) \dagger}/ \sqrt{2} \;\;\;,
\end{eqnarray}
where $N_{({\rm adj})}= 2k^{2} +k$,
$N_{({\rm antisym})}= 2k^{2} -k$ and $N_{({\rm fund})}= 2k$.
We find that all of the three Hamiltonians
$H_{{\rm fund}},H_{{\rm adj}}$ and $H_{{\rm asym}}$
are expressible in terms of a generic one
\begin{eqnarray}
\label{eq:abelian}
g^{2} H_{0} \left(X_{\ell}, \Phi, \Phi^{*}; (R), A \right) =
-\bar{b}_{A\dot{\alpha}}^{(R)} {\sigma}^{m\dot{\alpha}\alpha}X_{m}b_{A\alpha}^{(R)}
-d^{(R)\alpha}_{A} \sigma_{\alpha\dot{\alpha}}^{m}X_{m} \bar{d}_{A}^{(R) \dot{\alpha}}
+ \sqrt{2} \Phi b^{(R) \alpha}_{A} d_{A\alpha}^{(R)} \nonumber \\
+\sqrt{2} \Phi^{*}\bar{b}_{A\dot{\alpha}}^{(R)}
\bar{d}^{\dot{\alpha} (R) }_{A} \;\;
\end{eqnarray}
provided we replace the five parameters
\begin{equation}
X_{\ell},\;\;\; \Phi = \frac{X_4 + iX_7}{\sqrt{2}}, \;\;\;
\Phi^{*} = \frac{X_4 - iX_7}{\sqrt{2}} \;\;\; \ell =1,2,3
\end{equation}
by the appropriate ones. (See argument of ${\cal A}$ in
eq.~(\ref{eq:answer1}) below.)
The Berry connection appears in one or three particle
state of $H_{0}$ with respect to
the Clifford vacuum $\ket{\Omega}$;
$b^{\alpha}\ket{\Omega} = \bar{d}_{\dot{\alpha}}\ket{\Omega} = 0$.
We suppress the labels $A$ and $(R)$ seen in
eqs. (\ref{eq:expand}),(\ref{eq:abelian}) for a while.
Let us write $\mid \Psi \rangle = \left( h_{\alpha} d^{\alpha}
+ \bar{h}^{\dot{\alpha}} \bar{b}_{\dot{\alpha}} \right) \mid \Omega \rangle $ and
$\psi \equiv \left( h_{\alpha} , \bar{h}^{\dot{\alpha}} \right)^{t}$.
The transition amplitude reads
\begin{eqnarray}
\label{eq:formula}
\lim_{T \rightarrow \infty} \langle \Psi \mid P \exp \left[ -i
\int_{0}^{T}
dt H_{0}(t) \right] \mid \Psi \rangle
= \psi^{\dagger}
P \exp \left[ -i \int_{0}^{\infty} E(t) dt
+ i \int_{\Gamma} d\gamma (X_{m}, \Phi, \Phi^{*}) \right]
\psi\;.
\end{eqnarray}
Here $\Gamma$ is a path in the parameter space. The
connection one-form is
\begin{eqnarray}
id\gamma (t)= - \psi^{\dagger} (t) d \psi (t)
\equiv -i {\cal A} \;\;\;,
\end{eqnarray}
which is in general matrix-valued.
Let us consider for definiteness a set of two degenerate adiabatic
eigenstates with positive energy, which is specified by an index
$\sigma =1,4$. Using the completeness
$ {\displaystyle \sum_{\sigma =1,4} } \psi_{\sigma}
\psi_{\sigma}^{\dagger} = {\bf 1}_{(2)}$,
we find
\begin{equation}
\label{eq:factor}
\sum_{\sigma = 1,4} \lim_{T \rightarrow \infty} \langle \Psi_{\sigma}
\mid P \exp \left[ -i \int_{0}^{T}
dt H_{0}(t) \right] \mid \Psi_{\sigma} \rangle =
tr P \exp \left[ -i \int_{0}^{\infty} E(t) dt
- i \int_{\Gamma} {\cal A}(X_{\nu}) \right]
\end{equation}
from eq. (\ref{eq:formula}).
Here the trace is taken with respect to the two-dimensional subspace.
\subsection{computation of the Berry phase and the BPST instanton}
Now the problem is to obtain the nonabelian $(su(2)$ Lie algebra valued)
Berry connection associated with the fist quantized hamiltonian
\begin{equation}
{\cal H} = \frac{R}{g^2} \sum_{\nu=1,2,3,4,7} N^{\nu} \Gamma_{\nu}\;, \;
R \equiv \sqrt{(X^1)^2 +(X^2)^2 +(X^3)^2 +(X^4)^2 +(X^7)^2}\;,
\; N^{\nu} \equiv \frac{X^{\nu}}{R} \;,
\end{equation}
where $\Gamma_{\nu}$ are the five dimensional gamma matrices obeying
the Clifford algebra and the explicit representation can be read off
from eq.~(\ref{eq:abelian}).
The projection operators are
\begin{equation}
P_{\pm} = \frac{1}{2} ({\bf 1}_4 \pm N^{\nu} \Gamma_{\nu}) \;\;, \;\;
{P_{\pm}}^2 = P_{\pm} \;\;,\;\;
P_+^{\dag} = P_+\;\;,
\end{equation}
which satisfy
\begin{equation}
{\cal H} P_{\pm} = \pm \frac{R}{g^2} P_{\pm}\;\;.
\end{equation}
Denoting by ${\bf e}_i$ $(i=1,2,3,4)$ the unit vector in the $i$-th
direction, we write a set of normalized eigenvectors belonging to
the plus eigenvalue as
\begin{equation}
\psi_i = \frac{1}{{\cal N}_i} P_{+}{\bf e}_i \;\;.
\end{equation}
Here, $i=1,4$ refer to the sections around the north pole $X^{3}=R$
while $i=2,3$ to the ones around the south pole $X^{3}=-R$.
The ${\cal N}_i$ are the normalization factors:
\begin{equation}
{\cal N} \equiv {\cal N}_1 = {\cal N}_4 = \sqrt{\frac{1+N^3}{2}}\;\;,\;\;
{\cal N}^{\prime} \equiv{\cal N}_2 = {\cal N}_3 = \sqrt{\frac{1-N^3}{2}}\;\;.
\end{equation}
We focus our attention on the sections near the north pole. The Berry
connection is
\begin{equation}
i{\cal A} = \left( \begin{array}{l}
\psi_1, \\
\psi_4,
\end{array} \right)
d ( \psi_1, \: \psi_4 )
\end{equation}
We will here present the final answer only.
We parametrize $S^{3}$ of unit radius by the coordinates
\begin{equation}
Y^{\nu} \equiv \frac{1}{\sqrt{R^2 - (X^3)^2}} X^{\nu}\:\:,\:\:\:
(\nu = 1,2,4,7)
\end{equation}
\begin{equation}
(Y^1)^2 + (Y^2)^2 + (Y^4)^2 + (Y^7)^2 = 1\;\;.
\end{equation}
Also let ${\bf Y} \equiv ( Y^4, Y^7, Y^1 )^{t}$.
The $Y$ coordinates parametrize the $SU(2)$ group element as well:
\begin{equation}
T \equiv Y^2 {\bf 1}_2 + i {\bf Y} \cdot \mbox{\boldmath$\sigma$}\;\;,
\end{equation}
from which we can make the pure gauge configuration
\begin{equation}
\begin{array}{rcl}
dT T^{-1}
&=& (dY^2 {\bf 1}_2 + i d{\bf Y} \cdot \mbox{\boldmath$\sigma$})
(Y^2 {\bf 1}_2 - i {\bf Y} \cdot \mbox{\boldmath$\sigma$})\\
&=& i (d{\bf Y} \times {\bf Y} + Y^2 d{\bf Y} -{\bf Y} dY^2) \cdot
\mbox{\boldmath$\sigma$}.
\end{array}
\end{equation}
The final answer we have found out is
\begin{equation}
\label{eq:BPST}
{\cal A} \left( X^{\nu} \right) = p(R, X^{3}) dT T^{-1}\;\;.
\end{equation}
The prefactor $p(R, X^{3})$ is of interest and can be written as
\begin{equation}
p(R, X^{3}) = \frac{\tau^2}{\tau^2 + \lambda^2} \;\; , \;\;\;
\tau = \sqrt{R^2 - (X^3)^2}\;\;,\;\;\;
\lambda = R + X^3.
\end{equation}
The nonabelian connection ${\cal A}$ is in fact the BPST instanton
configuration. The size of the instanton
$\lambda$
is not a ${\it bonafide}$ parameter of the model but is chosen to be
the fifth
coordinate in the five dimensional Euclidean space. For fixed $\lambda$,
the four dimensional subspace embedded into the ${\bf R}^{5}$ is a
paraboloid wrapping the singularity. An observer on this recognizes
the pointlike singularity as the BPST instanton. As $\lambda$ goes to zero,
this paraboloid gets degenerated into an $SU(2)$ counterpart of the Dirac
string connecting the origin and the infinity.
Note also that the prefactor is written in terms of the angle measured
from the north pole as
\begin{equation}
p(\theta) = \frac{1}{2}(1- \cos \theta )\;\;,\;\;
N^3 \equiv R \cos \theta \;\;.
\end{equation}
\subsection{coupling to spacetime points}
Returning to the expression (\ref{eq:adprocess}) and
taking a sum over the labels $\sigma_{f A}^{(R)} = \sigma_{i A}^{(R)}$,
we find that the second line is expressible as the product of the factors
\begin{eqnarray}
\label{eq:answer1}
&& Tr P \exp \left( -i \sum_{f=1}^{n_{f}} \sum_{A=1}^{2k}
{\bf 1} \otimes \cdots \int_{\Gamma_{A,f}^{(fund)} } {\cal A}
\left[ {\bf w}^{A}\cdot {\bf X} _{\ell},\;
\frac{ m_{(f)}}{\sqrt{2}} +{\bf w}^{A} \cdot {\bf\Phi},\;
\frac{ m_{(f)}}{\sqrt{2}} + {\bf w}^{A} \cdot {\bf \Phi }^{\dagger}
\right] \cdots \otimes {\bf 1} \right)
\nonumber \\
&& Tr P exp \left( -i \sum_{A=1}^{2k^{2}}
{\bf 1} \otimes \cdots \int_{\Gamma_{A}^{(adj)} }
{\cal A} \left[ {\bf R}^{A} \cdot {\bf X}_{\ell},\;
i {\bf R}^{A} \cdot {\bf \Phi},\;
i {\bf R}^{A} \cdot {\bf \Phi}^{\dagger} \right] \cdots
\otimes {\bf 1} \right) \nonumber \\
&& Tr P exp \left(
-i \sum_{A=1}^{2k^2 - 2k} {\bf 1} \otimes \cdots
\int_{\Gamma_{A}^{(asym)} }
{\cal A} \left[ {\bf w}_{{\rm asym}}^{A} \cdot {\bf X}_{\ell},\;
{\bf w}_{{\rm asym}}^{A} \cdot {\bf \Phi},\;
{\bf w}_{{\rm asym}}^{A} \cdot {\bf \Phi}^{\dagger} \right] \cdots
\otimes {\bf 1} \right) \;.
\end{eqnarray}
We have included the energy dependence seen in eq. (\ref{eq:factor})
in $S_{B}$ as this is perturbatively cancelled by the contribution from
bosonic integration. The symbols seen in the arguments are
\begin{eqnarray}
\{ \{ {\bf w}^{A} \mid 1 \leq A \leq 2k \}\}
&=&
\{ \{ \pm {\bf e}^{(i)}
\; , 1 \leq i \leq k \}\} \;\; \nonumber \\
\{ \{ {\bf R}^{A} \mid 1 \leq A \leq 2k^{2} \}\}
&=&
\{ \{ \pm 2 {\bf e}^{(i)}, {\bf e}^{(i)}-{\bf e}^{(j)} ,
\pm \left({\bf e}^{(i)} +{\bf e}^{(j)} \right)
\; 1 \leq i,j, \leq k \}\} \;\; \nonumber \\
\{ \{ {\bf w}_{ {\rm asym} }^{A} \mid 1 \leq A \leq 2k^{2} -2k
\}\}
&=&
\{ \{ \pm \left( {\bf e}^{(i)} +{\bf e}^{(j)} \right) ,
{\bf e}^{(i)} - {\bf e}^{(j)}, \; 1 \leq i,j, \leq k \}\} \;\;.
\end{eqnarray}
The second and the third lines
are respectively the nonzero roots and the weights in the
antisymmetric representation
of $usp(2k)$. We have denoted by ${\bf e}^{(i)} \;(1 \leq i \leq k)$
the orthonormal basis vectors of $k$-dimensional Euclidean space and
\begin{eqnarray}
{\bf X}_{\ell} = \sum_{i=1}^{k} {\bf e}^{(i)} X_{\ell}^{(i)},\;
{\bf \Phi} = \sum_{i=1}^{k}{\bf e}^{(i)}
\frac{X_{4}^{(i)} + iX_{7}^{(i)}}{\sqrt{2}}, \;
{\bf \Phi}^{\dagger}= \sum_{i=1}^{k}{\bf e}^{(i)}
\frac{X_{4}^{(i)} - iX_{7}^{(i)}}{\sqrt{2}}
\;\;\;.
\end{eqnarray}
Let us exploit the symmetry of the roots and the weights
under ${\bf e}^{(i)} \leftrightarrow -{\bf e}^{(i)}$.
In general, neither the second line nor the third one collapse to unity.
Observe that, due to this symmetry, we can pair
the two-dimensional vector space associated with $ {\bf R}^{A}$
(or ${\bf w}_{ {\rm asym} }^{A}$) and that with $-{\bf R}^{A}$
(or $-{\bf w}_{ {\rm asym} }^{A}$). Let us symmetrize the tensor product
of these two two-dimensional vector spaces. On this, the nonabelian
Berry phase is reduced to the pure gauge configuration
\begin{equation}
{\cal A}(X^{\nu})_{ \{i }^{ \;\;\{j } \delta_{k \} }^{\;\; \ell \} }
+ {\cal A}(-X^{\nu})_{ \{i }^{ \;\;\{j } \delta_{k \} }^{\;\; \ell \} }
= \left( dT T^{-1} \right)_{ \{i }^{ \;\;\{j }
\delta_{k \} }^{\;\; \ell \} } \;\;\;,
\end{equation}
and this can be gauged away.
As for the first line of eq. (\ref{eq:answer1}), the mass terms prevent
this from happening.
After all these operations, eq. (\ref{eq:answer1}) becomes
\begin{eqnarray}
\label{eq:answer2}
Tr P \exp \left( -i \sum_{f=1}^{n_{f}} \sum_{A=1}^{2k}
{\bf 1} \otimes \cdots \int_{\Gamma_{A,f}^{(fund)} } {\cal A}
\left[ {\bf w}^{A}\cdot {\bf X} _{\ell},\;
\frac{m_{(f)}}{\sqrt{2}} +{\bf w}^{A} \cdot {\bf\Phi},\;
\frac{m_{(f)}}{\sqrt{2}} + {\bf w}^{A} \cdot {\bf \Phi }^{\dagger}
\right] \cdots \otimes {\bf 1} \right) \; \nonumber \\
\;\;
\end{eqnarray}
Let us note that before the symmetrization, the nonabelian Berry phase
is present in the $IIB$ case.
\subsection{abelian approximation}
In order to understand better the formation of the BPST instanton
obtained from the nonabelian Berry phase, we will study
this problem ignoring the degeneracy {\it i.e.} in the abelian
approximation. The section then takes the tensor product form of two
two-component wave functions
\begin{eqnarray}
\psi_{i} \equiv \psi_{A, a} \equiv
\left(
\begin{array}{c}
h_{\alpha} \\
\bar{h}^{\dot{\alpha}}
\end{array}
\right) \;\;.
\end{eqnarray}
Separation of variables is done by the following five
dimensional spherical coordinates
\begin{eqnarray}
X_2 &=& r \sin \phi_1 \sin \theta_1 \cos \theta_2 \;\;\;, \nonumber
\\
X_1 &=& r \cos \phi_1 \sin \theta_1 \cos \theta_2 \;\;\;, \nonumber
\\
X_3 &=& r \;\;\;\;\;\;\;\;\; \cos \theta_1 \cos \theta_2
\;\;\;, \nonumber \\
X_4 &=& r \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;
\sin \theta_2 \cos \phi_2 \;\;\;, \;\; \qquad 0 \leq \phi_2 \leq 2
\pi \;\;, \nonumber \\
X_7 &=& r \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;
\sin \theta_2 \sin \phi_2 \;\;\;, \;\; \qquad 0 \leq \theta_2 \leq
\pi \;\;.
\end{eqnarray}
The local form of the section
around $\left( \theta_{2},\theta_{1} \right)$ denoted by $(N,N)$
is
\begin{eqnarray}
\label{eq:efn}
\psi_{A, a}^{(N,N)} =
\left(
\begin{array}{c}
\sin \frac{\theta_2}{2} \, e^{-i \phi_2} \\
\cos \frac{\theta_2}{2} \\
\end{array}
\right)_{A}
\otimes
\left(
\begin{array}{c}
\cos \frac{\theta_1}{2} \\
\sin \frac{\theta_1}{2} \, e^{i \phi_1} \\
\end{array}
\right)_{a} \;\;\;.
\end{eqnarray}
The connection one-form is
\begin{eqnarray}
\label{eq:BC}
{\cal{A}}^{(N,N)} =
- \frac{i}{2} ( 1 - \cos \theta_2 ) d \phi_2
+ \frac{i}{2} ( 1 - \cos \theta_1 ) d \phi_1 \;\;.
\end{eqnarray}
We see that the BPST instanton is made of a monopole
anti-monopole pair and the flux of the monopole and
that of the antimonopole spread in the different
directions of spacetime.
\subsection{brane interpretation}
This time, cancellation noted before occurs without symmetrization.
The cancellation occurs as well to the part from the fundamental
representation which does not involve $X_{4}^{(i)}$ or $X_{7}^{(i)}$.
We find that the exponent of eq.~(\ref{eq:answer2}) is
written as
\begin{eqnarray}
&& -i \sum_{f=1}^{n_{f}} \sum_{i=1}^{k}
\gamma_{\Gamma}^{({\rm Berry})}\left[ X_{3}^{\prime (i)},
m_{f}+ X_{4}^{(i)}, X_{7}^{(i)} \right]
\nonumber \\
&& -i \sum_{f=1}^{n_{f}} \sum_{i=1}^{k}
\gamma_{\Gamma}^{({\rm Berry})}\left[ X_{3}^{\prime (i)},
m_{f}- X_{4}^{(i)}, X_{7}^{(i)} \right] \;\;.
\end{eqnarray}
Here
\begin{eqnarray}
\gamma_{\Gamma}^{({\rm Berry})} \left[ X_{3}^{\prime}, X_{4}, X_{7}
\right]
&=& \int {\cal A}^{({\rm Berry})} \\
{\cal{A}}^{(N)({\rm Berry})} &=&
- \frac{i}{2} ( 1 - \cos \theta_2 ) d \phi_2 \;, \;\;
{\cal{A}}^{(S)({\rm Berry})} =
+\frac{i}{2} ( 1 + \cos \theta_2 ) d \phi_2 . \label{eq:config}
\end{eqnarray}
and $X_{3}^{\prime (i) 2} = X_{1}^{(i)2} + X_{2}^{(i)2}+ X_{3}^{(i)2}$.
It is satisfying to see a pair of magnetic monopoles sitting at
$X_{4}^{(i)} = \pm m_{(f)}$ from the orientifold surface
for $i= 1 \sim k$. These monopoles live in the
parameter space,
which is the spacetime points of the matrix model.
Coming back to eq.~(\ref{eq:adprocess}), we conclude that the Berry
phase generates an interaction
\begin{eqnarray}
\label{eq:induced}
Z \left[ x_{\ell}, x_{I}=0 ; \cdots\right] =
\int \left[ D \tilde{v}_{M} \right] \prod_{f=1}^{n_{f}}
\left[ D Q_{(f)}\right] \left[ D Q_{(f)}^{*}\right]
\left[D {\tilde{Q}}_{(f)} \right]
\left[ D \tilde{Q}_{(f)}^{*} \right] \exp [ iS_{B} +
i \gamma_{\Gamma}^{({\rm total})} ] . \nonumber
\end{eqnarray}
Let us give this configuration we have obtained a brane
interpretation first from the six dimensional and
subsequently from the ten dimensional
point of view. It should be noted that the two coordinates
which the connection ${\cal{A}}^{({\rm Berry})}$ does not depend
on are the angular cooordinates $\theta_{1}, \phi_{1}$, so that
$X_{1}, X_{2}$ are not quite separable from the rest of the
coordinates $X_{3}, X_{4}, X_{7}$ in
eq.~(\ref{eq:config}). Only in the asymptotic region
$\mid X_{3}^{\prime} \mid >> \mid X_{3} \mid$, there exists
an area of size $\pi \mid X_{3}^{\prime} \mid^{2}$ transverse to
the three dimensional space where the Berry phase is obtained.
In this region, the magnetic flux obtained from the $b(=1)$-form
connection embedded in $d(=6)$-dimensional spacetime looks
approximately as is discussed in \cite{Nepomechie}: the flux
no longer looks coming from a poinlike object but from
a $d-b-3(=2)$ dimensionally extended object.
The magnetic monopole obeying the Dirac quantization behaves
approximately like a $D2$ brane extending to the
(1,2) directions, which are perpendicular to the orientifold surface.
In fact, the presence of this object and its quantized magnetic flux
have been detected by quantum mechanics of a point particle (electric
$D0$ brane) obtained from the $n=1$ and $n=3$ particle states of
the fermionic sector in the fundamental representation.
The induced interaction is a minimal one. We conclude that the $D0$
represented by the first and the third excited states of the quantum
mechanical problem given above is under the magnetic field created
by $D2$. To include the four remaining coordinates
$(X_5, X_6, X_8, X_9)$ of the antisymmetric directions, we appeal
to the translational invariance which is preserved in these
directions. The simplest possibility is that they appear in the
coupling through the derivatives
\begin{eqnarray}
\int {\cal A}^{({\rm Berry})} =
\int\prod_{I=5,6,8,9} dX^{I}dX^{\nu} A_{\nu 5689} \;\;.
\end{eqnarray}
With this assumption,
the $D0$ brane is actually a $D4$ bane extended in $(5,6,8,9)$
directions while the $D2$ still occupies $(1,2)$: the
quantization condition is preserved in ten dimensions as well.
\section{Schwinger-Dyson equations}
We will here repeat the basic part of the discussion in \cite{Itsu}.
\subsection{derivation of S-D/loop equations}
Let us derive S-D/loop equations, employing the open and the
closed loop variables
introduced in section $2.3$.
We first introduce abbreviated notation:
\begin{eqnarray}
\Phi[(i)] &\equiv& \Phi[p^{(i)}_{.},\eta^{(i)}_{.};n^{(i)}_{1},n^{(i)}_{0}]\;,
\;\Psi[(i)] \equiv
\Psi_{f^{(i)'}f^{(i)}}[k^{(i)}_{.},\zeta_{.}^{(i)};l^{(i)}_{1},l^{(i)}_{0};
\Lambda^{(i)'},\Lambda^{(i)}]\;, \nonumber\\
\int d\mu \cdots &\equiv& \int [dv][d\Psi][d\mbox{\boldmath$Q$}][d\mbox{\boldmath$Q$}^{\ast}]
[d\mbox{\boldmath$\psi_{Q}$}][d\mbox{\boldmath$\psi_{Q}$}^{\ast}] \cdots \;.
\nonumber
\end{eqnarray}
We begin with the following set of equations consisting of $N$ closed loops
and $L$ open loops:
\begin{eqnarray}
&&0=\int d\mu
\frac{\partial}{\partial X^{r}}
\left\{ Tr(U[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)}_{2},n^{(1)}_{1}+1] T^r
U[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)}_{1},n^{(1)}_{0}]) \right. \nonumber\\
&& \hspace*{4.5cm}
\left. \Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)]
\, e^{-S} \right\} \;\;,
\label{loopeq1}
\\
&& 0=\int d\mu \frac{\partial}{\partial X^{r}}
\left\{ \left( \Lambda^{(1)'} \mbox{\boldmath$\Pi$}_{(f^{(1)'})} \right)
F U[k^{(1)}_{.},\zeta^{(1)}_{.};l^{(1)}_{2},l^{(1)}_{1}+1]T^r
U[k^{(1)}_{.},\zeta^{(1)}_{.};l^{(1)}_{1},l^{(1)}_{0}]
\left( \Lambda^{(1)} \mbox{\boldmath$\Pi$}_{(f^{(1)})} \right) \right. \nonumber\\
&&\hspace*{4.5cm}
\left. \Phi[(1)] \cdots \Phi[(N)]
\Psi[(2)] \cdots \Psi[(L)]
\, e^{-S} \right\} \;\;,
\label{loopeq2}
\\
&& 0=\int d\mu
\frac{\partial}{\partial \mbox{\boldmath$Z$}_{(f) i}}
\left\{ (U[k^{(1)}_{.},\zeta^{(1)}_{.};l^{(1)}_{1},l^{(1)}_{0}]
\left( \Lambda^{(1)} \mbox{\boldmath$\Pi$}_{(f^{(1)})} \right) )_{i} \right. \nonumber\\
&& \hspace*{5cm} \left. \Phi[(1)] \cdots \Phi[(N)] \,
\Psi[(2)] \cdots \Psi[(L)]
\, e^{-S} \right\} \;\;,
\label{loopeq3}
\end{eqnarray}
where $X^r$ denotes $v_{M}^{r}$ or $\Psi^r_{\alpha}$ while
$\mbox{\boldmath$Z$}_{(f)i}$ denotes
$\mbox{\boldmath$Q$}_{(f)i}$ or $\mbox{\boldmath$\psi_{Q}$}_{(f) i \alpha}$.
We will exhibit eqs.
(\ref{loopeq1}) $\sim$ (\ref{loopeq3}) in the form of loop equations.
We will repeatedly use
\begin{equation}
\sum_{r=1}^{2 k^2 \pm k} (T^r)_{i}^{\; j} (T^r)_{k}^{\; l}
=\frac{1}{2} (\delta_{i}^{\; l} \delta^{j}_{\; k}
\mp F^{-1}_{ik} F^{lj}) \;\;,
\end{equation}
which is nothing but the expression for the projector (\ref{projector}).
In these equations below,
\begin{eqnarray}
P^{(i)}_{n}=\left\{ \begin{array}{l}
p^{(i)M}_{n} \; \mbox{if} \; X^r=v^{r}_{M} \;, \\
-\bar{\eta}^{(i)}_{n} \; \mbox{if} \; X^r=\Psi^r \;,
\end{array} \right. \;\;\;
K^{(i)}_{n}=\left\{ \begin{array}{l}
k^{(i)M}_{n} \; \mbox{if} \; X^r=v^{r}_{M} \;, \\
-\bar{\zeta}^{(i)}_{n} \; \mbox{if} \; X^r=\Psi^r \;,
\end{array} \right.
\end{eqnarray}
and $\Lambda^{(i)}$ not multiplied by $\Pi$ represents either
$\Xi^{(i)}$ or $\Theta^{(i)}$.
The symbol $\hat{b}$ denotes an omission of
the $b$-th closed or open loop.
\begin{eqnarray}
\bullet
(\ref{loopeq1}) \Rightarrow \; 0 =
(1)\; \mbox{\underline{kinetic term (Fig. \ref{closedkin}),
\ref{closed-open})}}
\;+\; (2)\; \mbox{\underline{splitting and twisting
(Fig. \ref{closed})}} \\
+ (3)\; \mbox{\underline{joining with a closed string
(Fig. \ref{closedclosed})}}
\;+\; (4)\; \mbox{\underline{joining with an open string
(Fig. \ref{openclosed})}} \;\;. \nonumber
\end{eqnarray}
Here
\begin{eqnarray}
&(1)& =
\frac{1}{g^2} \left\langle \left( \delta_{X} \Phi[(1); X^r] \right)
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;,
\label{eq:loop1kin} \\
&(2)& =
\left\langle
\left( -\frac{i}{2}\sum_{n=n^{(1)}_{0}}^{n^{(1)}_{1}} P^{(1)}_{n} \right)
\left\{ \Phi[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)}_{1},n+1]
\Phi[p^{(1)}_{.},\eta^{(1)}_{.};n,n^{(1)}+1] \right.\right.\nonumber\\
&\pm & \left. Tr(U[p^{(1)}_{.},\eta^{(1)}_{.};n,n^{(1)}_1+1]
U[\mp p^{(1)}_{.},\mp \eta^{(1)}_{.};n+1,n^{(1)}_1]) \right\}
\left.
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle
\nonumber\\
&+& \left\langle
\left( -\frac{i}{2}\sum_{n=n^{(1)}_{1}+1}^{n^{(1)}_{2}} P^{(1)}_{n} \right)
\left\{ \Phi[p^{(1)}_{.},\eta^{(1)}_{.};n,n^{(1)}_{1}+1]
\Phi[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)},n+1] \right.\right.\nonumber\\
&\pm& \!\!\!\!\left. Tr(U[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)}_{1},n+1]
U[\mp p^{(1)}_{.},\mp \eta^{(1)}_{.};n^{(1)}_{1}+1,n]) \right\}
\left.
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle ,
\nonumber \\
&& \;\; \\
&(3)& =
\left\langle
\left( -\frac{i}{2}\sum_{b=2}^{N} \sum_{n=n^{(b)}_{0}}^{n^{(b)}_{1}}
P^{(b)}_{n} \right)
\left\{ Tr(U[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)}_{1},
n^{(1)}_{1}+1] U[p^{(b)}_{.},\eta^{(b)}_{.};n,n+1]) \right.\right.\nonumber\\
&& \mp \left. Tr(U[\mp p^{(1)}_{.},\mp
\eta^{(1)}_{.};n^{(1)}_{1}+1,n^{(1)}_{1}]
U[p^{(b)}_{.},\eta^{(b)}_{.};n,n+1]) \right\} \nonumber\\
&&\left. \Phi[(2)] \cdots \hat{b} \cdots \Phi[(N)]
\Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;, \\
&(4)& =
\left\langle
\left( -\frac{i}{2}\sum_{b=1}^{L} \sum_{l=l^{(b)}_{0}}^{l^{(b)}_{1}}
K^{(b)}_{l} \right) \right. \nonumber \\
&& \left\{ \left( \Lambda^{(b)'} \mbox{\boldmath$\Pi$}_{(f^{(b)'})} \right) F
U[k^{(b)}_{.},\zeta^{(b)}_{.};l^{(b)}_{1},l+1]
U[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)}_{1},n^{(1)}_{1}+1]
U[k^{(b)}_{.}, \zeta^{(b)}_{.};l,l^{(b)}_{0}]
\left( \Lambda^{(b)} \mbox{\boldmath$\Pi$}_{(f^{(b)})} \right) \right. \nonumber\\
&\mp& \!\!\!\!\left. \left( \Lambda^{(b)'} \mbox{\boldmath$\Pi$}_{(f^{(b)'})} \right) F
U[k^{(b)}_{.},\zeta^{(b)}_{.};l^{(b)}_{1},l+1]
U[\mp p^{(1)}_{.},\mp \eta^{(1)}_{.};n^{(1)}_{1}+1,n^{(1)}_{1}]
U[k^{(b)}_{.},\zeta^{(b)}_{.};l,l^{(b)}_{0}]
\left( \Lambda^{(b)} \mbox{\boldmath$\Pi$}_{(f^{(b)})} \right) \right\} \nonumber\\
&& \hspace*{4cm}\left.
\Phi[(2)] \cdots \Phi[(N)]
\Psi[(1)] \cdots \hat{b} \cdots \Psi[(L)] \right\rangle \;\;\;.
\label{loopeq1complete}
\end{eqnarray}
The term $\delta_{X} \Phi[(1); X^r]$
comes from the variation of the action and
contains terms representing closed-open transition.
For its explicit form, see the original paper \cite{Itsu}.
We present the pictures associated with the terms
$(1) \sim (4)$ as figures.
\begin{figure}
\centerline{\epsfbox{closedkin.eps}}
\caption{\small infinitesimal deformation of a closed string}
\label{closedkin}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{closed-open.eps}}
\caption{\small closed-open transition}
\label{closed-open}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{closed.eps}}
\caption{\small splitting and twisting of a closed string}
\label{closed}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{closedclosed.eps}}
\caption{\small joining of two closed strings}
\label{closedclosed}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{openclosed.eps}}
\caption{\small joining of a closed string and an open string}
\label{openclosed}
\end{figure}
\begin{eqnarray}
\bullet
(\ref{loopeq2}) &\Rightarrow& \;\;0 =
(1)\; \mbox{\underline{kinetic term (Fig. \ref{openkin1},\ref{open2})}}
+ (2)\; \mbox{\underline{splitting and twisting (Fig.\ref{open1} )}} \\
&+& (3)\;\mbox{\underline{joining with a closed string
(Fig. \ref{openclosed})}}
+(4)\;\mbox{\underline{joining with an open string (Fig. \ref{openopen1})}}
\;\;.
\nonumber
\end{eqnarray}
We will present here only the figures associated with
$(1) \sim (4)$ of eq. (\ref{loopeq2}).
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{openkin1.eps}}
\caption{\small infinitesimal deformation of an open string: case one}
\label{openkin1}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{open2.eps}}
\caption{\small splitting of an open string}
\label{open2}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{open1.eps}}
\caption{\small splitting and twisting of an open string}
\label{open1}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{openopen1.eps}}
\caption{\small joining of two open strings: case one}
\label{openopen1}
\end{figure}
\begin{eqnarray}
\bullet
(\ref{loopeq3}) &\Rightarrow& \;\;0 =
(1)\; \mbox{\underline{kinetic term (Fig. \ref{openkin2})}}
+ (2)\; \mbox{\underline{open-closed transition
(Fig. \ref{open-closed})}} \nonumber \\
&+& (3)\;
\mbox{\underline{joining with an open string (Fig. \ref{openopen2})}} \;\;.
\end{eqnarray}
Again, we will present here only the figures associated with
$(1) \sim (3)$ of eq. (\ref{loopeq3}).
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{openkin2.eps}}
\caption{\small infinitesimal deformation of an open string: case two}
\label{openkin2}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{open-closed.eps}}
\caption{\small open-closed transition}
\label{open-closed}
\end{figure}
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{openopen2.eps}}
\caption{\small joining of two open strings: case two}
\label{openopen2}
\end{figure}
We have checked that all three of the loop equations
are expressed by the closed and open loops
$\Phi$ and $\Psi$ and their derivatives with respect to the sources
introduced.
For example, the expression
$\mbox{\boldmath$\psi_{Q}$}^{\ast} \cdot \bar{\sigma}^m
U[p^{(1)}_{.},\eta^{(1)}_{.};n^{(1)}_{1},n^{(1)}_{1}+1] \mbox{\boldmath$\psi_{Q}$}$
contained in $\left( \delta_{X} \Phi[(1); X^r] \right)$
in eq. (\ref{eq:loop1kin}) is represented as
\begin{equation}
\sum_{f=1}^{2n_f}\bar{\sigma}^{\dot{\alpha}\alpha m}
\frac{\partial}{\partial\bar{\theta}^{'\dot{\alpha}}}
\frac{\partial}{\partial\theta^{\alpha}} \Psi_{f f}[p^{(1)}_{.},
\eta^{(1)};n^{(1)}_{1},n^{(1)}_{1}+1;\Lambda',\Lambda]\;\;.
\end{equation}
In this sense, the set of loop equations we have derived is closed.
It is noteworthy that all of the terms in the above loop equations
are either an infinitesimal deformation of a loop or a consequence from
the two elementary local processes of loops which are illustrated
in Fig. \ref{elementaryprocess}.
It is interesting to discuss the system of loop equations we have derived
in the light of string field theory. In addition to the lightcone
superstring field theory constructed earlier in \cite{GSSFT}, there is
now gauge invariant string field theory for closed-open bosonic system
\cite{KT}. We find that the types of the interaction terms of our equations
are in complete agreement with the interaction vertices seen in
\cite{GSSFT} and
the second paper of \cite{KT}. In particular, Figures $2\sim5, 7\sim9,
11\sim12$ for the interactions of our equations are in accordance with
$U, V_{\infty}, V_{3}^{c}, U_{\Omega}, V_{3}^{0}, V_{\alpha}, V_{4}^{0}$
of \cite{KT}.
While BRS invariance determines the coefficients of the interaction
vertices in \cite{KT}, the (bare) coefficients are already determined
in our case from the first quantized action. This may give us insight
into properties of the model which are not revealed.
\begin{figure}
\vspace*{1cm}
\centerline{\epsfbox{elementaryprocess.eps}}
\caption{\small two kinds of elementary local processes}
\label{elementaryprocess}
\end{figure}
\subsection{Linearized loop equations and a free string}
Let us consider the all three loop equations
eqs. (\ref{loopeq1}), ({\ref{loopeq2}) and (\ref{loopeq3}) in the linearized
approximation, namely, ignoring the joining and splitting of the loops.
Let us first introduce a variable conjugate to $p_{A n}$ or $k_{A n}$
and that to $\eta_{n}$ or $\zeta_{n}$ by
\begin{eqnarray}
\hat{X}^{A}_{n} &=& i \frac{\delta}{\delta p_{A n}} \; \mbox{or} \;
\; i \frac{\delta}{\delta k_{A n}} \;\;, \\
\hat{\Psi}_{n} &=& i \frac{\delta}{\delta \eta_{n}} \; \mbox{or} \;
\; i \frac{\delta}{\delta \zeta_{n}}\;\;.
\end{eqnarray}
By acting $\hat{X}^{A}_{n}$ and $\hat{\Psi}_{n}$ on a loop, we obtain
respectively an operator insertion
of $v^{A}$ and that of $\Psi$ at point $n$ on the loop.
Now consider eq. (\ref{loopeq1}) and eq. (\ref{loopeq2})
for the case $X^r=v_{M}^{r}$, multiplying them by $p_{nM}^{(1)}$ and
$k_{nM}^{(1)}$ respectively. Consistency requires that,
for these terms, we must take into account
the term from the interactions which represents splitting of
a loop with infinitesimal length.
This in fact occurs when
the splitting point $n$ coincides with the point $n_{1}^{(1)}$
at which $T^r$ is inserted.
We obtain
\begin{eqnarray}
0&=&\frac{1}{g^2} p_{nM}^{(1)}
\left\langle \left( \delta_{X} \Phi[(1); v_{M}^{r}] \right)
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)]
\right\rangle \nonumber\\
&&-\frac{i}{2}2k p_{n}^{(1)2}
\left\langle \Phi[(1)]
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;,\\
0&=&\frac{1}{g^2} k_{nM}^{(1)}
\left\langle \left( \delta_{X} \Psi[(1); v_{M}^{r}] \right)
\Phi[(1)] \cdots \Phi[(N)] \Psi[(2)] \cdots \Psi[(L)] \right\rangle \nonumber\\
&&-\frac{i}{2}2k k_{n}^{(1)2}
\left\langle \Phi[(1)]
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;.
\end{eqnarray}
These equations lead to the half of the Virasoro conditions \cite{FKKT}:
\begin{eqnarray}
0&=&(p_{n}^{(1)2}+\hat{X}_{n}^{(1)'2}+\mbox{(fermionic part)}) \nonumber\\
&&\left\langle \Phi[(1)]
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;, \\
0&=&(k_{n}^{(1)2}+\hat{X}_{n}^{(1)'2}+\mbox{(fermionic part)}) \nonumber\\
&&\left\langle \Phi[(1)]
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;,
\end{eqnarray}
where $\prime$ implies taking a difference between two adjacent
points $n$ and $n+1$.
The reparametrization invariance of the Wilson loops leads to
the remaining half of the Virasoro conditions:
\begin{eqnarray}
0&=&(p_{n}^{(1)M}\hat{X}_{nM}^{(1)'}+\mbox{(fermionic part)}) \nonumber\\
&&\left\langle \Phi[(1)]
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;, \\
0&=&(k_{n}^{(1)M}\hat{X}_{nM}^{(1)'}+\mbox{(fermionic part)}) \nonumber\\
&&\left\langle \Phi[(1)]
\Phi[(2)] \cdots \Phi[(N)] \Psi[(1)] \cdots \Psi[(L)] \right\rangle \;\;.
\end{eqnarray}
Next, let us consider eq. (\ref{loopeq3}), ignoring joining and splitting
of the loops. Again consistency appears to require that
we drop the cubic terms consisting of $\mbox{\boldmath$Q$}$ and $\mbox{\boldmath$Q$}^{\ast}$
in $\delta_{\mbox{\boldmath$Z$}} \Psi[(1);\mbox{\boldmath$Z$}]$ .
To write explicitly, the following expression must vanish
\begin{eqnarray}
&& \left\{ \mbox{\boldmath$Q$}^{\ast}_{(f)}(v_{\nu}v^{\nu}+[\Phi_I,\Phi^I])
+(\mbox{\boldmath$Q$} \Sigma)_{(f)}F[\Phi_{2}^{\dagger},\Phi_{3}^{\dagger}]
+(\mbox{\boldmath$Q$}^{\ast}M^2)_{(f)}+2 (\mbox{\boldmath$Q$}^{\ast}M)_{(f)}v_4
- i \sqrt{2} \mbox{\boldmath$\psi_{Q}$}^{\ast}_{(f)}\bar{\lambda} \right. \nonumber\\
&& \left. -\sqrt{2}(\mbox{\boldmath$\psi_{Q}$} \Sigma)_{(f)} F \psi_{\Phi_1} \right\}
U[k^{(1)}_{.},\zeta^{(1)}_{.};l^{(1)}_{1},l^{(1)}_{0}]
(\Lambda^{(1)} \mbox{\boldmath$\Pi$}_{f^{(1)}}) \approx 0
\label{linearizedloopeq1} \;\;\;, \\
&& \left\{ \mbox{\boldmath$\psi_{Q}$}^{\ast}_{(f)} \bar{\sigma}^{m} v_{m}
+i \sqrt{2} \mbox{\boldmath$Q$}_{(f)}^{\ast} \lambda
+ \left( \mbox{\boldmath$\psi_{Q}$} \Sigma F (\sqrt{2}\Phi_1 + M) \right)_{(f)} \right.
\nonumber \\
&&\left. +\sqrt{2} (\mbox{\boldmath$Q$} \Sigma F \psi_{\Phi_1})_{(f)} \right\}
U[k^{(1)}_{.},\zeta^{(1)}_{.};l^{(1)}_{1},l^{(1)}_{0}]
(\Lambda^{(1)} \mbox{\boldmath$\Pi$}_{f^{(1)}}) \approx 0 \;\;\;,
\label{linearizedloopeq2}
\end{eqnarray}
when inserted in
\begin{equation}
\left\langle \Phi[(1)]
\Phi[(2)] \cdots \Phi[(N)] \Psi[(2)] \cdots \Psi[(L)] \right\rangle \;\;.
\end{equation}
As we stated before, the lefthand sides of eqs.
(\ref{linearizedloopeq1}) and (\ref{linearizedloopeq2}) are expressible
as an open loop with some functions of $\hat{X}^{A}_{l^{(1)}_{1}}$ and
$\hat{\Psi}_{l^{(1)}_{1}}$ acting on the loop. Let us see by inspection how
eqs. (\ref{linearizedloopeq1}) and (\ref{linearizedloopeq2}) are satisfied by
the source functions alone.
Consider the following configuration of $\hat{X}_{n}$ and $\hat{\Psi}_{n}\;,$
$n= l^{(1)}_{1}$:
\begin{eqnarray}
&& \hat{X}^{\mu} \approx 0 \;\; \mbox{for}\;\; \mu = 0,1,2,3,7 \;\;,
\;\;\;\hat{X}^{4}=\pm m_f \;\;\;. \label{boundary1} \\
&& \hat{X}^{'I} \approx 0 \;\; \mbox{for}\;\; I = 5,6,8,9
\label{boundary2} \\
&& \hat{\Gamma}_{3} \hat{\Psi} \approx -\hat{\Psi} \;\;\;,
\label{boundary3}
\end{eqnarray}
where $\hat{\Gamma}_{3} \equiv \Gamma_5 \Gamma_6 \Gamma_8 \Gamma_9$.
Again these equations should be understood in the sense of
an insertion at the end point of the open loop.
Eqs. (\ref{boundary1}), (\ref{boundary2}), and (\ref{boundary3}) tell us
that the open loop $\Psi[(1)]$ obeys the Dirichlet boundary
conditions for $0,1,2,3,4,7$ directions and the Neumann boundary
conditions for $5,6,8,9$ directions.
We find that the configuration given by eqs. (\ref{boundary1}),
(\ref{boundary2}) and (\ref{boundary3}) solves the linearized loop equations
(\ref{linearizedloopeq1}), (\ref{linearizedloopeq2}).
This configuration clearly tells us the existence of $n_f$ $D3$ branes
and their mirrors each of which is at a distance $\pm m_f$ away from the
orientifold surface in the fourth direction.
This conclusion consolidates both the semiclassical picture
in section three and the picture emerging from the
fermionic integrations in section four.
\section*{Acknowledgements}
We thank the organizers of the Nishinomiya symposium and
the YITP workshop. This work is supported in part
by the Grant-in-Aid for Scientific Research (10640268,10740121)
and by the Grant-in-Aid for Scientific Research Fund (97319)
from the Ministry of Education, Science and Culture, Japan.
|
\section{Introduction}
The Advection Dominated Accretion Flows (ADAFs) have been reviewed
by a number of authors, most recently by Narayan, Mahadevan, \& Quataert
(\cite{NMQ}), Kato, Fukue, \& Mineshige (\cite{kato}), and Lasota
(\cite{jpl98}). The ADAFs represent a class of optically thin solutions of
the accretion flow, which radiate so inefficiently, that almost all the
heat dissipated inside the fluid is subsequently advected toward the black
hole horizon. Since the cooling of matter is negligible, the equations of
fluid dynamics are independent of the equations describing the emission,
absorption and scattering of radiation. This allows one to address these two
topics separately. In this paper we are concerned mostly with the radiation
processes inside the flow.
The fully relativistic dynamics of ADAFs has been described and some
solutions have been obtained in Lasota (\cite{jpl94}),
Abramowicz et al. (\cite{ACGL}, hereafter ACGL),
Abramowicz, Lanza \& Percival (\cite{ALP}),
Peitz \& Appl (\cite{peitz}),
Jaroszy\'{n}ski \& Kurpiewski (\cite{paperI}, hereafter Paper I),
Gammie \& Popham (\cite{gammie98}),
and Popham \& Gammie (\cite{popham98}).
The most extensive survey of the parameter
space is probably presented by Popham \& Gammie (\cite{popham98}), where
the dependence of the flow on the black hole spin $a$, the viscosity
parameter $\alpha$, the gas adiabatic index and on the advection parameter
$f$ (where $f=1$ means fully advective flow without any cooling) is
investigated. This work shows rather strong dependence of the flow
characteristics on parameters mentioned, especially on the black hole spin.
The dependence on advection parameter is also substantial, but for a narrow
range of $f$, which can represent flows with negligible cooling
($0.9 \le f \le 1$) one can use the $f=1$ solutions. Similarly,
the solutions depend strongly on the viscosity parameter $\alpha$, but for
the limited range of this parameter ($\alpha \ge 0.01$), which is more
physically relevant (Balbus, Gammie \& Hawley \cite{balbus95}), the
differences between the solutions are not dramatic, in particular the
topology of the isobars is the same.
In our calculations we use the solutions of the equations describing the
flow dynamics presented in Paper I.
All our models use $f=1$ and $\alpha=0.1$. Both values are
representative of the physically relevant ADAFs. The black hole spin, we
consider, is limited to the three values ($a=0$, $0.5$, and $0.9$).
The main purpose of this paper is a self-consistent
treatment of photon Comp\-to\-ni\-za\-tion in a two temperature plasma of
an ADAF solution. To do so we need a 3D distribution of matter density,
velocity and temperature, while the standard solutions give only the
vertically averaged quantities as measured at the equator. At this point it
is important to choose the relevant method of vertical averaging and we
follow Abramowicz et al. (\cite{ALP}), Quataert \& Narayan (\cite{QN}) and
Paper I in choosing averaging on spheres (and not on cylinders). In this
approach the matter distribution in space is limited by the centrifugal
forces barrier and does not resemble the infinite isothermal atmosphere,
obtained in the alternative approach.
We calculate the spectrum of photons leaving the flow. The photons
originate in brems\-strahlung and thermal synchrotron processes, which can be
described locally. The scattering (if any) can take place at any point of
the photon trajectory inside matter distribution and is nonlocal. We
simulate this process using Monte Carlo approach. Our simulation of photon
Comp\-to\-ni\-za\-tion is quite standard, but the flow, where the processes
take place is rather complicated. Since the optical depth for scattering is
low, a photon can travel to a distant part of the fluid before undergoing
any interaction. The relative motion of the fluid elements, where
consecutive interactions of photons with matter take place, can be
substantial. Also, the photons traveling in the densest parts of the flow,
near the horizon, can be deflected by the gravitational field. This and
other relativistic effects in photon motion are included in our study.
The main aim of our investigation is the self-consistent treatment of
Comp\-to\-ni\-za\-tion. Most non-analytical thermal Comp\-to\-ni\-za\-tion
models use iterative methods of solving the kinetic equations
(e.g. Poutanen \& Svensson \cite{poutanen}). This method has been adopted
by Narayan, Barret \& McClintock (\cite{NBC}) to calculate the spectrum of
soft X-ray transient source V404 Cyg in which, they believe, ADAF exists.
In our calculations we use Monte Carlo method which has been described
in most detail by Pozdnyakov, Sobol \& Sunyaev (\cite{PSS})
and G\'orecki \& Wilczewski (\cite{gorecki}). This method allows us to
follow only one photon at a time and we are not able to take into account
the creation of $e^+e^-$ pairs. However as Bj{\"o}rnsson et
al. (\cite{bjorn96}) and Kusunose \& Mineshige (\cite{kusunose96}) showed,
the role of $e^+e^-$ pairs in ADAFs is not significant.
In the next Section we briefly characterize the model of ADAF. The
method of Monte Carlo Comp\-to\-ni\-za\-tion is presented in Sec.~3. In
Sec.~4 we present the results of calculations, showing the ADAFs spectra.
The discussion and conclusions follow in the last Section.
\section{The 3D description of the fluid}
\subsection{The accretion flow}
We investigate the stationary flow of matter and the propagation of light
in the gravitational field of a rotating Kerr black hole
using the Boyer-Lindquist coordinates $t$, $\phi$, $r$, $\theta$ and the
metric components $g_{ab}(r,\theta)$ as given by Bardeen (\cite{Bard}).
We follow the $(-,+,+,+)$ signature convention.
We use geometrical units, so the speed of light
$c \equiv 1$ and the mass of the hole $M \equiv 1$.
The Kerr parameter $a$ ($0 \le a <1$) gives
the black hole angular momentum in geometrical units. We use the
Einstein summation convention, where needed, and a semicolon for the
covariant derivative. The normalization of four velocity with the chosen
metric signature reads $u^a u_a=-1$.
The system of equations we use follows in general ACGL. Since we
are neglecting the cooling processes at this stage, and treat the
accretion flow in the disk approximation, two components of velocity
and the speed of sound $c_{\mathrm{S}}$ given as functions of radius, fully
describe the dynamics. We use the angular velocity $\Omega$ and the
physical radial velocity $V$ as measured by locally nonrotating observers
as main kinematic variables (compare ACGL).
After the {\it vertical averaging} the velocity perpendicular to the
equatorial plane is neglected ($u^\theta \equiv 0$) and other
components of the four velocity are given as:
\begin{equation}
u^t={1 \over \sqrt{1-V^2}
\sqrt{-g_{tt}-2 \Omega g_{t\phi}-\Omega^2 g_{\phi\phi}}}
\end{equation}
\begin{equation}
u^\phi=\Omega u^t
\end{equation}
\begin{equation}
u^r={1 \over \sqrt{g_{rr}}}{V \over \sqrt{1-V^2}}~~.
\end{equation}
The above velocity components have to be known only at the equatorial plane
and its close vicinity if one needs to obtain a system of equations
describing an ADAF in the slim disk approximation (Abramowicz et
al. \cite{maa88}). To describe the matter distribution in space (also far
from the equatorial plane) one has to introduce further specifications of
the velocity and angular momentum distribution, since only the averaged
values enter the equations. We assume that the poloidal velocity component
($u^\theta$) is absent and that the radial velocity component $V$ depends
on the radius only:
\begin{equation}
V=V(r) ~~~~~~u^\theta \equiv 0
\end{equation}
(The BL velocity component $u^r$ depends on $\theta$ through $g_{rr}$ -
compare Eq.3)
The choice of angular momentum distribution is less obvious. Close to the
horizon, where the velocities are highest and possible kinematic effects
most important, the specific angular momentum $\ell \equiv -u_\phi/u_t$ is
approximately constant (Paper I, ACGL, Peitz \& Appl, \cite{peitz}).
We assume $\ell$ to be exactly constant on spheres:
\begin{equation}
\ell = \ell(r)
\end{equation}
Since the metric components do depend on $\theta$ the angular velocity is
not constant on spheres:
\begin{equation}
\Omega
={g^{\phi t}u_t+g^{\phi\phi}u_\phi \over g^{tt}u_t+g^{t\phi}u_\phi}
={g^{\phi t}-\ell g^{\phi\phi} \over g^{tt}-\ell g^{t\phi}}
\end{equation}
The other kinematic quantities are given as:
\begin{equation}
u_t={1 \over \sqrt{1-V^2}
\sqrt{-g^{tt}+2 \ell g^{t\phi}-\ell^2 g^{\phi\phi}}}
\end{equation}
\begin{equation}
u_\phi=-\ell u_t
\end{equation}
We assume that the accreting plasma contains small scale,
isotropically tangled magnetic field resulting from magnetohydrodynamical
instability (Balbus et al. \cite{balbus95}). Hence we write the total
pressure as
\begin{equation}
p=p_\mathrm{g}+p_\mathrm{m}
~~~~p_\mathrm{g}= \beta p
~~~~p_\mathrm{m} \equiv {B^2 \over 8\pi} = (1-\beta) p
\end{equation}
where $p_\mathrm{g}$ is the gas pressure, $p_\mathrm{m}$ - the
magnetic pressure, $B$ - the magnetic field, and $\beta$ is a constant
parameter. The pressure, the rest mass density $\rho_0$ and the sound
velocity $c_\mathrm{S}$ are related:
\begin{equation}
p=\rho_0 c_\mathrm{S}^2
\end{equation}
In our calculations we use a two temperature plasma with a small amount
of magnetic field to represent the matter properties. Thus the ion
pressure dominates, the gas is non-relativistic and the specific enthalpy
$\mu$ is given
as:
\begin{equation}
\mu \equiv {\epsilon+p \over \rho_0}=1+{5 \over 2}c_\mathrm{S}^2
\end{equation}
where $\epsilon$ is the total (rest mass plus thermal) energy density.
In the ADAF set of equations only the vertically averaged sound speed
is used. It is in spirit of our approximations to postulate:
\begin{equation}
c_\mathrm{S}=c_\mathrm{S}(r)
\end{equation}
which means that the gas is isothermal on spheres.
The electron temperature $T_\mathrm{e}$
is much lower than the ion temperature, so
its influence on the equation of state and the structure of the flow can be
neglected.
Our detailed (but approximate) description of the fluid kinematics makes it
possible to find the density dependence on the angular coordinate $\theta$.
The set of equations
describing ADAF contains viscosity terms, which are included in the energy
equation, but neglected in the mechanical equilibrium equations (ACGL,
Peitz \& Appl \cite{peitz}, Paper I). Thus it is sufficient to use the
ideal fluid energy momentum tensor:
\begin{equation}
T^a_b=(\epsilon + p)u^a u_b + p \delta^a_b
\end{equation}
The $\theta$ conservation equation, $T^a_{\theta;a}=0$ reads:
\begin{equation}
u^a u_{\theta;a} = {-p_{,\theta} \over \epsilon + p}
\end{equation}
After some algebra we obtain:
$$
{1 \over 1-V^2}(\ln u_t)_{,\theta}
-{1 \over 2}{V^2 \over 1-V^2}(ln(r^2+a^2\cos^2 \theta))_{,\theta}~~~~~~
$$
\begin{equation}
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
= -{c_\mathrm{S}^2 \over \mu}(\ln \rho_0)_{,\theta}
\end{equation}
Since $V$ does not depend on $\theta$, the LHS of the equation is a full
gradient of a quantity which can be called a potential $\psi$. This implies
the solution for the density:
\begin{equation}
\rho_0(\theta)=\rho_{0,\mathrm{eq}}\exp\left(-{\mu(\psi(\theta)-\psi_\mathrm{eq})
\over c_\mathrm{S}^2}\right)
\end{equation}
where the $\rho_{0,\mathrm{eq}}$, $\psi_\mathrm{eq}$ denote the values measured
at the equatorial plane.
The constant sound speed on the spheres may suggest that the exponential
atmosphere never ends. For the rotating configuration there is, however, an
infinite potential barrier close to the rotation axis, where $u_t
\rightarrow \infty$ and $\rho_0 \rightarrow 0$. Thus the vicinity of the
rotation axis is empty and the density falls steeply down near this
region. In this respect the ADAF solutions are similar to the so called
thick accretion disks (Abramowicz, Jaroszy{\'n}ski, \& Sikora \cite{AJS78}).
Both have empty funnels around their rotation axes.
The mass flow through the $r=const$ surface can be calculated as:
\begin{equation}
{\dot M}=-\int_0^\pi \mathrm{d}\theta \mathrm{d}\phi
\sqrt{-g} \rho_0(\theta) u^r
\end{equation}
\begin{equation}
\equiv 2\pi \sqrt{V^2 \over 1-V^2} \int_0^\pi \mathrm{d}\theta
\sqrt{(g_{t\phi}^2-g_{tt}g_{\phi\phi})g_{\theta\theta}}\rho_0(\theta)
\end{equation}
Combining the last equation and the formula for the $\theta$ dependence
of the density we obtain the equatorial value of the density.
\subsection{Two temperature plasma}
While modeling the dynamics of ADAF we neglect the heat transfer and all
the radiation processes, assuming that only a small part of the total energy
generated by viscous processes can be affected by them. Now we are going to
model the radiation processes.
We assume that the whole energy dissipated is transferred to
the ions. The ions heat the electrons via Coulomb collisions and the
electrons lose their energy by the synchrotron, bremsstrahlung and
inverse Compton cooling processes. In the ADAFs thermalization time-scale
greatly exceeds the dynamical time-scale and the plasma remains two temperature.
We find the electron and ion temperatures, $T_\mathrm{e}(r,\theta)$
and $T_\mathrm{i}(r,\theta)$, self-consistently using the equation of state
\begin{equation}
p_\mathrm{g}=\beta p=\frac{\rho
kT_\mathrm{i}}{\mu_\mathrm{i}m_\mathrm{H}}+
\frac{\rho kT_\mathrm{e}}{\mu_\mathrm{e}m_\mathrm{H}}
\label{state}
\end{equation}
where $\mu_\mathrm{i}=1.29$ and $\mu_\mathrm{e}=1.18$ are effective molecular
weights of the ions and electrons, and the condition of thermal equilibrium
applied locally
\begin{equation}
q^\mathrm{+}=q^\mathrm{-}_\mathrm{br}+q^\mathrm{-}_\mathrm{br,C}
+q^\mathrm{-}_\mathrm{S}+q^\mathrm{-}_\mathrm{S,C}
\label{equilib}
\end{equation}
where $q^\mathrm{+}$ is the rate of Coulomb heating of electrons by
ions (e.g. Mahadevan \cite{mahadevan}), $q^\mathrm{-}_\mathrm{S}$ and
$q^\mathrm{-}_\mathrm{br}$ are the synchrotron and brems\-strahlung cooling
rates, and $q^\mathrm{-}_\mathrm{S,C}$ and $q^\mathrm{-}_\mathrm{br,C}$
are the Compton cooling rate of synchrotron and brems\-strah\-lung photons,
respectively.
The calculation of the synchrotron cooling rate is somewhat complicated
because the optical depth to absorption (Self Synchrotron Absorption) for
majority of the synchrotron photons is high.
In our approach we assume that a photon can either escape
from the medium carrying out its energy and taking part in cooling
process, or be absorbed very close to the point of its original
emission and not contributing to the cooling. In reality all photons
carry energy and some are absorbed far from the emission point
transferring energy to distant portions of the fluid. Our assumption
neglects the heat transport between different volume elements, but makes
calculations doable.
To find the probability of a photon escape from given location we follow
$N \sim 10^2$ rays with directions randomly chosen at the frame
comoving with the fluid. The optical depth along a ray measured at the
frequency $\nu$ is
\begin{equation}
\tau_\nu=\int_0^{l_\infty}
\frac{\epsilon_\mathrm{S}(\nu)}{4\pi B_{T_\mathrm{e}}(\nu)}\mathrm{d}l
\end{equation}
where $\epsilon_\mathrm{S}$ is the synchrotron emissivity,
$B_{T_\mathrm{e}}$ is the
Planck function corresponding to the electron temperature
$T_\mathrm{e}$, and the integration is over the proper distance,
$l_\infty$ corresponding to the point on the fluid boundary.
Since the integrand in the above formula is the
function of the electron temperature we must assume the approximate
distribution of the electron temperature in the ADAF. As a first
approximation we use our results from Paper I based on the approach of
Narayan \& Yi \cite{NYi}.
Averaging one gets the probability of escape:
\begin{equation}
\mathrm{e}^{-\tau_{\nu}} =
{1 \over N}~\sum_{i=1}^N~~\mathrm{e}^{-\tau_\nu^{(i)}}
\label{tau}
\end{equation}
Since we assume that cooling is provided only by the escaping photons,
we have:
\begin{equation}
q^\mathrm{-}_\mathrm{S}=\int_0^\infty~~\epsilon_\mathrm{S}(\nu)~
\mathrm{e}^{-\tau_\nu}~\mathrm{d}\nu
\end{equation}
for the synchrotron cooling.
We adopt the expressions for synchrotron emissivity
from Pacholczyk (\cite{pacholczyk}) and Mahadevan, Narayan \& Yi
(\cite{MNY}).
For the brems\-strahlung cooling the solution is straightforward.
The absorption of low frequency brems\-strahlung photons has no
practical
meaning, so we adopt their frequency integrated emission as cooling rate
\begin{equation}
q^\mathrm{-}_\mathrm{br}=\int_0^\infty~\epsilon_\mathrm{br}(\nu)~\mathrm{d}\nu
\end{equation}
We take the expression for brems\-strahlung cooling rate from Stepney
\& Guilbert (\cite{stepney}).
Finally we take the cooling by Comptonization of both synchrotron and
brems\-strahlung photons using the formulae of Esin et al. (\cite{esin}).
The mean optical depth to Compton scattering is calculated in the same
manner as above.
Solving the equations \ref{state} and \ref{equilib} we get the electron
temperature and the ion temperature.
\subsection{Spectra neglecting Comptonization}
As a by-product of the calculations of the previous subsections one can
obtain the spectrum of the model, which would be valid if the
Comp\-to\-ni\-za\-tion were unimportant. For sufficiently low frequencies
a photon is rather absorbed than scattered, so both approaches, neglecting
and including Comp\-to\-ni\-za\-tion, should give similar results in this
regime. This gives a chance of a self check of the simulations.
Calculation of many rays sent from a given fluid element can also be used to
find the contribution of this element to the total luminosity of the
configuration as seen by a distant observer. If the frequency
$\nu_\mathrm{em}$ and the direction of a photon in the fluid frame is known,
its frequency in the Boyer-Lindquist coordinate frame $\nu_\mathrm{obs}$
can be calculated and this is the frequency that would be measured by
a distant observer, unless the photon goes under the horizon. For photons
going to infinity the redshift factor can be defined:
\begin{equation}
1+z={\nu_\mathrm{em} \over \nu_\mathrm{obs}}
\end{equation}
Distant observers measure photon energies divided by the factor $(1+z)$.
Also the time interval between the detection of two signals is the interval
between their sending multiplied by $(1+z)$.
Thus the contribution to the total luminosity from the fluid element of the
volume $\Delta V$, measured by distant
observers in their frequency interval $\Delta\nu_\mathrm{obs}$ is given as:
\begin{equation}
L_(\nu_\mathrm{obs})\Delta\nu_\mathrm{obs}=
{1 \over N}~\sum_{i=1}^N~~
{\epsilon(\nu_\mathrm{em})\Delta\nu_\mathrm{em}
~\mathrm{e}^{-\tau_{\nu_\mathrm{em}}^{(i)}}
\over (1+z_i)^2}~~\Delta V
\end{equation}
where for the i-th ray:
\begin{equation}
\nu_\mathrm{em}=\nu_\mathrm{obs}(1+z_i)~~~\Delta\nu_\mathrm{em}=
\Delta\nu_\mathrm{obs}(1+z_i)
\end{equation}
The summation is limited to rays which reach infinity. Integration over
the volume of the fluid gives the total luminosity of the model.
The emission from the configuration is not isotropic, so the observers
at different position angles measure a different flux of radiation.
The luminosity calculated above is in fact an average of luminosities
assigned to the disk by observers uniformly distributed on a sphere
around the object. One can also find the average luminosity that would
be measured by observers from a limited solid angle $\Delta\Omega$.
To do so it is sufficient to neglect in the summation all the rays
which do not enter the region of interest and multiply the result
by the correction factor $4\pi/\Delta\Omega$.
\subsection{Spatial distribution of the photon emissivity}
In the Monte Carlo simulations we follow individual photons as they travel
through the fluid undergoing consecutive scatterings. We have to know what
is the distribution of the points of emission of the photons. We divide the
flow into several spherical layers. The radius in the middle of the
layer numbered $j$ is $r_j$. Each layer is then subdivided
into annuli of limited range in the polar angle $\theta$ between fluid
boundaries
$\theta_\mathrm{min}(r_j)$ and $\pi-\theta_\mathrm{min}(r_j)$.
The angular coordinate at the middle of the annulus numbered $jk$
is $\theta_{jk}$.
One can assume that all fluid parameters are almost uniform inside each
annulus, and take their values at $(r_j,\theta_{jk})$ as representative.
Using similar arguments as in the previous subsection we calculate the
rate of photon emission from the region numbered $jk$
of the volume $\Delta V_{jk}$:
\begin{equation}
{\dot {\cal N}}_{jk}=
{\Delta V_{jk} \over N}~\sum_{i=1}^N~~
\int_0^\infty{\epsilon(\nu_\mathrm{em})
~\mathrm{e}^{-\tau_{\nu_\mathrm{em}}^{(i)}}~\mathrm{d}\nu_\mathrm{em}
\over h\nu_\mathrm{em}(1+z_i)}
\label{ndotjk}
\end{equation}
We adopt the expressions for synchrotron cooling
from Pacholczyk (\cite{pacholczyk})
and Mahadevan, Narayan \& Yi (\cite{MNY}),
and for brems\-strahlung cooling from Svensson (\cite{svensson}) [and
references therein].
The cooling rates are functions of the electron temperature,
the number density of ions and electrons and the magnetic field density
(synchrotron radiation) and hence they are functions of $r$ and $\theta$.
We take $\epsilon=\epsilon(r_j,\theta_{jk})$.
The expression under the integral is regular at low frequencies despite
the presence of $\nu_\mathrm{em}$ in the denominator because
$\tau_\mathrm{em}^{(i)} \rightarrow \infty$ when $\nu_\mathrm{em}
\rightarrow 0$. The redshift factor in the denominator takes care of the
difference between clock rates in the fluid frame and at infinity.
\section{The Comptonization}
We follow the method of Comp\-to\-ni\-za\-tion described by G\'{o}recki \&
Wilczewski (\cite{gorecki}).
\subsection{Basic concepts}
The differential cross section for Compton scattering is given by the
following formula (Akhiezer \& Berestetski \cite{akhiezer}):
\begin{equation}
{\mathrm{d}\sigma \over \mathrm{d}\Omega^\prime}={r^2_0 \over
2\gamma^2}X(1-\vec{v}\vec{\Omega}/c)^{-2}
{\left( {h\nu^\prime \over h\nu} \right)}^2
\end{equation}
where $h\nu$, $\vec{\Omega}$, $h\nu^\prime$ and $\vec{\Omega}^\prime$
are respectively
the energy and the direction of the photon before and after the scattering,
$\vec{v}$ is the velocity of the electron, $\gamma$ is the Lorentz factor and
$r_0$ is the classical electron radius. The symbol $X$ denotes the
invariant part of the cross section :
\begin{equation}
X={x \over x^\prime}+{x \over x}^\prime+4\left({1 \over x}-{1 \over
x^\prime}\right)+4{\left({1 \over x}-{1 \over x^\prime}\right)}^2
\end{equation}
where
\begin{equation}
{x \over 2}={h\nu \over mc^2}\gamma(1-\vec{v}\vec{\Omega}/c) ~,~
{x \over 2}^\prime={h\nu^\prime \over
mc^2}\gamma(1-\vec{v}\vec{\Omega}^\prime/c)
\end{equation}
are the energies of the incoming photon and of the scattered photon,
respectively, expressed in units of $mc^2$ in the reference frame of the
electron. The energies $h\nu$ and $h\nu^\prime$ are related by the
Compton formula
\begin{equation}
h\nu^\prime={h\nu(1-\vec{v}\vec{\Omega}/c) \over
1-\vec{v}\vec{\Omega}^\prime/c+{h\nu \over
\gamma mc^2}(1-\vec{\Omega} \vec{\Omega}^\prime)}
\label{compton}
\end{equation}
We use the total Compton cross section which is given by (Berestetski,
Lifshitz \& Pitaevski \cite{berestetski})
\begin{eqnarray}
\lefteqn{\sigma (x) =} \nonumber \\
& = 2\pi r^2_0{1 \over x} \left[\left(1-{4 \over x}-{8 \over
x^2}\right)ln(1+x)+{1 \over 2}+{8 \over x}-{1 \over 2(1+x)^2}\right]
\end{eqnarray}
The basic concept of this method is to follow the photon trajectory from
the moment of emission until the photon leaves the flow. The probability
that a photon leaves the flow without scattering is
\begin{equation}
P_i=exp\left\{-\int_{\vec{r}_i}^{\vec{r}_\infty}
n_e\langle\sigma\rangle \mathrm{d}l\right\}
\end{equation}
where the integral is taken along the photon trajectory from the point of
the last ($i$-th) scattering to the boundary of the flow $\vec{r}_\infty$,
$n_e=\int n_e(\vec{v})\mathrm{d}^3v$ is the electron density,
$n_e(\vec{v})$ is the electron velocity distribution, and
\begin{equation}
\langle \sigma \rangle = {1 \over n_e}\int
n_e(\vec{v})(1-\vec{v}\vec{\Omega}/c)
\sigma(x)\mathrm{d}^3v
\end{equation}
is a mean cross section averaged over the electron velocity distribution.
$n_e(\vec{v})$.
The probability $P_i$ enables us to find the statistical weights of
the number of photons leaving the flow without scattering (and thus
contributing to the emerging spectrum) and the photons which remain in
the flow and undergo the next ($[i+1]$-th) scattering. These weights are
given by $w_i P_i$ and $w_{i+1}=w_i (1-P_i)$, respectively, where
$i=0$, $1$, $2$, $3$, $...$ is the index denoting the succeeding
scatterings. We
assume $w_0=1$. We follow the trajectory of the photon until $w$ becomes
less than a certain minimal value $w_\mathrm{min}$. Since ADAFs are optically
thin (in our model the Thomson optical depth is about 0.1 in
equatorial directions) the mean number of scatterings is 4 - 5 for
$w_\mathrm{min}=10^{-7}$.
\subsection{Generating the random variables}
The random variables are generated from the probability distributions
using Monte Carlo methods: the inversion of the cumulative distribution
function or von Neumann's rejection technique. Multi-dimensional
distributions are modeled using the conditional probability
distributions.
(i) At first we generate the position vector at which the
photon is initially emitted.
According to our approximation this position is uniformly distributed
in space within each of the annuli $jk$. The probability that a
photon is emitted from a region of the given number $jk$ is :
\begin{equation}
f_{jk}=\frac{{\dot {\cal N}}_{jk}}
{\sum_{j^\prime,k^\prime} {\dot {\cal N}}_{{j^\prime}{k^\prime}} }
\label{efjotka}
\end{equation}
where ${\dot {\cal N}}_{jk}$ are given in Eq.~\ref{ndotjk}. In
Eq.~\ref{ndotjk} we take into account the optical depth due to the
absorption, so the number of the low frequency photons we use in the
simulation
is the expected number of the photons which have a chance of escape. The
absorption does not have to be considered on the further photon
trajectory. (After a scattering with relativistic electrons a photon gains
so much energy that the possibility of its absorption can be
neglected. The probability of absorption on the original trajectory is
included in Eq.~\ref{ndotjk}.)
(ii) As the position of input photon is determined we can generate
the initial energy of the photon $h\nu_0$ using a probability distribution
specified by the photon spectrum of synchrotron or brems\-strahlung emission
\begin{equation}
f_{jk}(\nu)=\frac{n_{\nu}(r_j,\theta_{jk})}
{\int_0^{\infty}n_{\nu}(r_j,\theta_{jk})\mathrm{d}\nu}
\end{equation}
where $n_{\nu}(r_j,\theta_{jk}) [Hz^{-1}cm^{-3} s^{-1}]$ is the photon spectrum
approximated by formulae of Pacholczyk (\cite{pacholczyk})
and Mahadevan, Narayan \& Yi (\cite{MNY}) for synchrotron emission
or Svensson (\cite{svensson}) for brems\-strahlung
emission. The photon spectrum is determined from the energy spectrum
by dividing the last one by $h\nu$.
(iii) We assume that the emission of input photons is isotropic. Hence
the direction of the photon in a comoving Cartesian coordinate frame
$\vec{\Omega}$= $(\sin\Theta\cos\Phi$, $\sin\Theta\sin\Phi$, $\cos\Theta)$
is generated
from uniform distributions in the ranges $\cos\Theta\in[-1;1]$ and
$\Phi\in[0;2\pi]$.
In this way we determine the set of parameters
\{$\vec{r}_0$, $h\nu_0$, $\vec{\Omega}_0$, $w_0=1$\}
describing the initial point of
the trajectory of the photons beam. We calculate the next points of the
trajectory, i.e.
\{$\vec{r}_{i+1}$, $h\nu_{i+1}$, $\vec{\Omega}_{i+1}$, $w_{i+1}$\}
($i=0$, $1$, $2$, $3$, $...$) until $w=w_{min}$. The way of computing
the weights $w_i$ is described in Sec.~3.1. Below we present the way of
computing the position, energy and direction of a photon after following
scatterings.
(iv) The position $\vec{r}_{i+1}$ is found on the photon trajectory
at the proper distance $l$ from the starting point
$\vec{r}_i$, from the probability distribution:
\begin{equation}
f(l)=
\frac{\mathrm{e}^{-\tau(l)}\frac{\mathrm{d}\tau(l)}{\mathrm{d}l}}
{\int_0^{l_\infty}e^{-\tau(l)}\frac{\mathrm{d}\tau(l)}
{\mathrm{d}l}\mathrm{d}l}
\end{equation}
where
\begin{equation}
\tau(l)=\int_0^{l} n_\mathrm{e}{\langle \sigma \rangle}\mathrm{d}l
\label{tauscat}
\end{equation}
(v) The two remaining parameters $h\nu_{i+1}$ and $\vec{\Omega}_{i+1}$
of the $(i+1)-th$ point of the photon trajectory are obtained by simulating
the scattering of the photon of energy $h\nu_i$ and direction
$\vec{\Omega}_i$ by an electron with velocity $\vec{v}$. To describe the
probability distribution of this scattering we use the differential
cross section (1) :
\begin{eqnarray}
f_{(h\nu_i,\vec{\Omega}_i)}(\vec{v},\vec{\Omega}_{i+1}) & = &
\nonumber \\
& = &
\frac{n(\vec{v})(1-\vec{v}\vec{\Omega}_i/c)\frac{\mathrm{d}\sigma}{\mathrm{d}\Omega_{i+1}}}
{\int\!\!\int
n(\vec{v})(1-\vec{v}\vec{\Omega}_i/c)
\frac{\mathrm{d}\sigma}{\mathrm{d}\Omega^{\prime}}
\mathrm{d}^3\Omega^{\prime}\mathrm{d}^3\vec{v}}
\end{eqnarray}
We model the multi-dimensional probability distribution (13) as a
product of the probability distribution of $\vec{v}$ and the
conditional probability distribution of $\vec{\Omega}_{i+1}$
\begin{equation}
f_{(h\nu_i,\vec{\Omega}_i)}(\vec{v},\vec{\Omega}_{i+1})=
f_1(\vec{v})f_2(\vec{\Omega}_{i+1}|\vec{v})
\end{equation}
where
\begin{equation}
f_1(\vec{v})=\frac{n(\vec{v})}{n_e}(1-\vec{v}\vec{\Omega}_i/c)
\frac{\sigma(x)}{\langle \sigma \rangle}
\label{velocity}
\end{equation}
and
\begin{equation}
f_2(\vec{\Omega}_{i+1}|\vec{v})=\frac{1}{\sigma(x)}
\frac{\mathrm{d}\sigma}{\mathrm{d}\Omega_{i+1}}
\label{direction}
\end{equation}
We generate the velocity $\vec{v}$ from Eq.~\ref{velocity}
and then the direction $\vec{\Omega}_{i+1}$ from Eq.~\ref{direction}.
Having $\vec{\Omega}_{i+1}$ , the energy
of the scattered photon $h\nu_{i+1}$ can be obtained from the Compton
formula (\ref{compton}). The detailed description of the method of
modeling the
probability distributions $f_1(\vec{v})$ and
$f_2(\vec{\Omega}_{i+1}|\vec{v})$ can be found in G\'orecki \&
Wilczewski (\cite{gorecki}).
\section{Results}
We have performed trial calculations of the ADAFs spectra employing the
method described. We use the models of ADAFs from Paper I. For the black hole
mass, accretion rate and parameter $\beta$ we use the parameters of
Lasota et al. (\cite{jpl96}),
$M_\mathrm{BH}=3.6 \times 10^7 M_{\sun}$, ${\dot m}=0.016$, $\beta=0.95$,
which they apply
in the modeling of \object{NGC 4258}. We do not, however, attach
cold, thin
disks at large radii to our ADAF solutions.
In Fig.~\ref{spectrum} we show the input spectra of
synchrotron and brems\-strahlung photons as well as the resulting
Comptonized spectrum of the disk around $a=0.9$ black hole.
All the spectra in this and other diagrams are shown as $\lg(\nu)$ versus
$\lg(\nu F_\nu)$ plots.
For the synchrotron input the results
are based on calculations including $10^6$ input photons and
following more than $5\times 10^6$ branches of photon trajectories. The
Comp\-to\-ni\-za\-tion plays a less important role in the case of brems\-strahlung
radiation, so we use $\sim 10$ times fewer photon trajectories to
obtain the spectra in this case. Since the Comp\-to\-ni\-za\-tion preserves the
photon number, we normalize the spectra using Eq.~\ref{ndotjk} with
either synchrotron or brems\-strahlung emissivity under the integral to
obtain the relative numbers of seed photons of each kind.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{fig1.eps}}
\caption{
The spectra of the input synchrotron and brems\-strahlung photons
and the resulting spectrum after Comp\-to\-ni\-za\-tion for $a=0.9$.
The resulting spectrum is shown as a solid line. Synchrotron
input (left) and brems\-strahlung (right) use dotted lines.
}
\label{spectrum}
\end{figure}
The Comptonized spectra of synchrotron radiation are smooth enough to
allow a power law fit with power index $\Gamma$ (i.e. $L_\nu \sim
\nu^{-\Gamma}$). The fit is not valid at the vicinity of the first peak,
which is due to the seed photons. In Fig.~\ref{fits} we show synchrotron
spectra with fits. As can be seen in the plots the slopes of the
spectra depend on the model. Since the black hole mass and the accretion
rate are the same for all three cases, the differences must be attributed
to the black hole angular momentum and its influence on the flow structure.
The power law indices estimated from fits
are $\Gamma=0.89$, $0.85$, and $0.81$ for $a=0.$, $0.5$, and
$0.9$ respectively.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{fig2.eps}}
\caption{
The power law fits to the Comptonized synchrotron emission for
models with black hole angular momentum
$a=0$ (dashed lines), $a=0.5$ (dotted) and $a=0.9$ (solid).
The corresponding straight lines represent the fits.
}
\label{fits}
\end{figure}
The three spectra resulting from combined effects of synchrotron and
brems\-strahlung emission with Comp\-to\-ni\-za\-tion are shown in
Fig.~\ref{widma}.
\begin{figure}
\resizebox{\hsize}{!}{\includegraphics{fig3.eps}}
\caption{
The resulting spectra of ADAFs for the three cases including the
brems\-strahlung component. The conventions follow Fig.~\ref{fits}
}
\label{widma}
\end{figure}
We have also checked the dependence of the observed total luminosity of our
models on the observer's position. For the brems\-strahlung photons the
dependence is absent. The synchrotron radiation observed from the
equatorial plane is stronger by 10 to 20\% as compared to the measurement
from the axis of rotation.
Following the photons we are able to find the fraction which goes under the
horizon. For the synchrotron photons the numbers are: $0.07$, $0.05$, and
$0.04$ for $a=0$, $0.5$ and $0.9$ respectively. The brems\-strahlung photons
are emitted at relatively larger distances from the horizon and less than
1\% of them are lost in all cases.
The fraction of photons emitted by the fluid and going under the horizon
is a decreasing function of the black hole angular momentum according to
our simulations.
We have checked this result of our simulations making an independent
calculation. We have compared isotropic sources of radiation at the
same distance from the black hole, comoving with the matter of the three
models we use. Again the fraction of rays going under the horizon is the
smallest for the case of the most rapidly rotating hole. The effect must
be attributed to the differences in matter kinematics between the three
models.
\section{Discussion and conclusions}
We use the ADAF models from Paper I, but we apply some changes to the
treatment of the 3D structure of the flow. The models are still based on
vertically averaged, stationary equations for the radial structure of the
flow, which uses the vertical scale height and quantities measured at
the equator as flow variables. Such treatment can be inadequate near
the flow boundary, especially close to the rotation axis. We assume the
specific angular momentum of matter, its radial velocity, and the sound
speed to be constant on spheres. Under such assumptions it is
possible to introduce an effective potential on each sphere, which
acts as infinite centrifugal potential barrier near the rotation axis.
This infinite barrier causes the sharp drop to zero of matter density
there despite the fact that the gas is approximately isothermal on spheres;
it effectively removes matter with artificially high angular velocity.
Our models are based on the assumption that all the heat dissipated in the
flow goes into the ions. We neglect the direct viscous heating of electrons
and the fact that their entropy changes as they fall toward the black hole,
which represents the advection of heat. Both mechanisms influence the
energy balance equation and, as shown by Nakamura et
al. (\cite{nakamura97}), Narayan et al. (\cite{narayan98}) and Quataert \&
Narayan (\cite{QN}), have an impact on the electron temperature and the
resulting spectrum of the model. While advection of heat by electrons is a
well defined process, the viscous heating must be introduced using another
free parameter. The combined effects of viscosity and advection on
electrons would influence all our models in a similar way, not greatly
changing the differences between them.
In our calculations we use three models of the matter flow onto the black
hole, with the same accretion rate and the same black hole mass, but with
different black hole angular momentum. The spin of the black hole has
strong influence on the density and temperature of the matter near the
horizon, which are both increasing functions of the rotation rate.
Similar behavior of the gas parameters can also be seen in much broader
investigation of ADAFs parameter space by Popham \& Gammie
(\cite{popham98}). We are not able to present a full discussion of the
ADAF structure - spectrum dependence, but we can point out some trends.
Our calculations show a strong dependence of the ADAFs spectra on the
flow structure resulting from the differences in the black hole angular
momentum. The synchrotron seed photons are produced mainly in the
central parts of the flow, which are the densest and the hottest. The
total energy emitted as synchrotron photons increases with the black
hole angular momentum. Also the influence of Comp\-to\-ni\-za\-tion is
increased the same way. In the case of $a=0.9$ model, the Comptonized
synchrotron radiation dominates all the way to the highest frequencies,
making the usual brems\-strahlung peak invisible. For other cases
considered ($a=0.5$ or $0$) this is not true and the brems\-strahlung
components dominate at highest frequencies.
The standard theory of Comp\-to\-ni\-za\-tion (Rybicki \& Lightman
\cite{rybicki})
enables one to estimate the spectral index of low energy radiation,
which undergoes multiple scatterings with thermal relativistic
electrons of given temperature and optical depth. In our case both
parameters can be defined as averages over the configuration. We have
tried several simple prescriptions for calculating the averages, but we
have not obtained a quantitative agreement between our results and the
estimates, the calculated spectra having spectral indexes by $\approx
0.1$ higher (i.e. being steeper). The discrepancies must be attributed
to the complexity of the flow and effects such as the relative motion of
the starting point of a photon and the place of its interaction with the
electrons.
We are not attempting to model \object{NGC 4258}, but using the
parameters from the model of Lasota et al. (\cite{jpl96}) we get the
right luminosity in the X-rays for the $a=0.9$ model. The slope of the
calculated spectrum at this frequency ($10^{18}~\mathrm{Hz}$) is within the
observational bounds. The spectrum of \object{NGC 4258} has been also
modeled by Gammie, Narayan, \& Blandford (\cite{gnb}).
Although they use slightly different values of the ADAF parameters,
their results are very similar to those of Lasota et al. (\cite{jpl96}).
\begin{acknowledgements}
This work was supported in part by the Polish State Committee
for Scientific Research grant 2-P03D-012-12
\end{acknowledgements}
|
\subsection*{1. Introduction}
{\bf 1.} Self-interactions of the Higgs field in the scalar sector
induce the breaking of the electroweak symmetry $SU(2)_L\times U(1)_Y$
down to the electromagnetic symmetry $U(1)_{EM}$ of the Standard Model
(SM). Gauge bosons and fermions acquire masses by interactions with
the non-zero Higgs field $v=1/(\sqrt{2} G_F)^{1/2}$ in the ground
state of the scalar potential. It is therefore an important
experimental task to reconstruct the elements of the Higgs potential
which gives rise to the spontaneous breaking of the electroweak
symmetry. The shape of the potential is determined by the mass $M_H$
of the physical Higgs boson field, and its trilinear and quadrilinear
couplings. The trilinear coupling \cite{hhhlc},
\begin{eqnarray}
\lambda_{HHH} = 3M_H^2/M_Z^2
\end{eqnarray}
in units of $\lambda_0 = M_Z^2/v$, can be measured directly in the
production of Higgs-boson pairs at high energy colliders. In proton
collisions at the LHC, Higgs pairs can be produced through double
Higgs-strahlung off $W$ and $Z$ bosons \cite{hrad}, $WW$ and $ZZ$
fusion \cite{fusion}, and gluon-gluon fusion \cite{glover}; in generic
notation:
\begin{eqnarray}
\begin{array}{l l l c l c l}
\mbox{double Higgs-strahlung}& \hspace{-0.3cm} : & q\bar{q} &
\hspace{-0.3cm} \to &\hspace{-0.1cm} W^*/Z^* &
\hspace{-0.3cm} \to &\hspace{-0.1cm} W/Z + HH
\nonumber \\[0.1cm]
WW/ZZ\ \mbox{double-Higgs fusion}& \hspace{-0.3cm} : & qq &
\hspace{-0.3cm} \to & \hspace{-0.1cm} qq + WW/ZZ &
\hspace{-0.3cm} \to & \hspace{-0.1cm} HH
\nonumber \\[0.1cm]
\mbox{gluon} \; \mbox{fusion} & \hspace{-0.3cm} : & gg &
\hspace{-0.3cm} \to & \hspace{-0.1cm} HH & & \nonumber
\end{array}
\end{eqnarray}
Characteristic diagrams of the three processes are shown in
Fig.\ref{smdiag}. With values typically near 10~fb, high integrated
luminosities are needed to generate a sufficiently large ensemble of
signal events and to cope with the large number of background events.
\vspace{\baselineskip}
\noindent {\bf 2.} The Minimal Supersymmetric extension of the Standard
Model (MSSM) incorporates a quintet of Higgs bosons: $h$, $H$, $A$,
$H^{\pm}$; the particles $h$, $H$ are neutral and CP-even while $A$ is
neutral and CP-odd. The mass of the light CP-even Higgs boson $h$ is
limited to less than about 130~GeV. The masses of the other Higgs
bosons are typically of the order of the electroweak symmetry breaking
scale $v$, yet they may extend up to values of order 1~TeV. The MSSM
Higgs system is described inherently by two parameters which are
generally chosen as the mass $M_A$ of the pseudoscalar Higgs boson and
the mixing parameter $\tan\beta$, ratio of the vacuum expectation
values of the two neutral Higgs fields. Radiative corrections
introduce the mass and mixing parameters of the heavy $t/ \tilde{t}$
and $b/\tilde{b}$ chiral multiplets into the system. \s
In CP-invariant theories, six types of trilinear couplings are
realized among the neutral Higgs fields \@ifnextchar [{\@tempswatrue\@citexr}{\@tempswafalse\@citexr[]}{hhhlc,djouadi}:
\begin{eqnarray}
\begin{array}{l l l l}
hhh, & Hhh, & HHh, & HHH \nonumber \\
hAA, & HAA & & \nonumber
\end{array}
\end{eqnarray}
\begin{fmffile}{fd}
\begin{figure}[ht]
\begin{flushleft}
\underline{double Higgs-strahlung: $q\bar q\to ZHH/WHH$}\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{10mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmflabel{$q$}{i1} \fmflabel{$\bar q$}{i3}
\fmf{boson,lab=$W/Z$,lab.s=left,tens=3/2}{v1,v2}
\fmf{boson}{v2,o3} \fmflabel{$W/Z$}{o3}
\fmf{phantom}{v2,o1}
\fmffreeze
\fmf{dashes,lab=$H$,lab.s=right}{v2,v3} \fmf{dashes}{v3,o1}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$H$}{o2} \fmflabel{$H$}{o1}
\fmfdot{v3}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmf{boson,tens=3/2}{v1,v2}
\fmf{dashes}{v2,o1} \fmflabel{$H$}{o1}
\fmf{phantom}{v2,o3}
\fmffreeze
\fmf{boson}{v2,v3,o3} \fmflabel{$W/Z$}{o3}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$H$}{o2} \fmflabel{$H$}{o1}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmf{boson,tens=3/2}{v1,v2}
\fmf{dashes}{v2,o1} \fmflabel{$H$}{o1}
\fmf{dashes}{v2,o2} \fmflabel{$H$}{o2}
\fmf{boson}{v2,o3} \fmflabel{$W/Z$}{o3}
\end{fmfgraph*}
\end{fmfshrink}
}
\\[2\baselineskip]
\underline{$WW/ZZ$ double-Higgs fusion: $qq\to qqHH$}\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{10mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(24,20)
\fmfstraight
\fmfleftn{i}{8} \fmfrightn{o}{8}
\fmf{fermion,tens=3/2}{i2,v1} \fmf{phantom}{v1,o2}
\fmflabel{$q$}{i2}
\fmf{fermion,tens=3/2}{i7,v2} \fmf{phantom}{v2,o7}
\fmflabel{$q$}{i7}
\fmffreeze
\fmf{fermion}{v1,o1}
\fmf{fermion}{v2,o8}
\fmf{boson}{v1,v3}
\fmf{boson}{v3,v2}
\fmf{dashes,lab=$H$}{v3,v4}
\fmf{dashes}{v4,o3} \fmf{dashes}{v4,o6}
\fmflabel{$H$}{o3} \fmflabel{$H$}{o6}
\fmffreeze
\fmf{phantom,lab=$W/Z$,lab.s=left}{v1,x1} \fmf{phantom}{x1,v3}
\fmf{phantom,lab=$W/Z$,lab.s=left}{x2,v2} \fmf{phantom}{v3,x2}
\fmfdot{v4}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,20)
\fmfstraight
\fmfleftn{i}{8} \fmfrightn{o}{8}
\fmf{fermion,tens=3/2}{i2,v1} \fmf{phantom}{v1,o2}
\fmf{fermion,tens=3/2}{i7,v2} \fmf{phantom}{v2,o7}
\fmffreeze
\fmf{fermion}{v1,o1}
\fmf{fermion}{v2,o8}
\fmf{boson}{v1,v3}
\fmf{boson}{v4,v2}
\fmf{boson,lab=$W/Z$,lab.s=left}{v3,v4}
\fmf{dashes}{v3,o3} \fmf{dashes}{v4,o6}
\fmflabel{$H$}{o3} \fmflabel{$H$}{o6}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,20)
\fmfstraight
\fmfleftn{i}{8} \fmfrightn{o}{8}
\fmf{fermion,tens=3/2}{i2,v1} \fmf{phantom}{v1,o2}
\fmf{fermion,tens=3/2}{i7,v2} \fmf{phantom}{v2,o7}
\fmffreeze
\fmf{fermion}{v1,o1}
\fmf{fermion}{v2,o8}
\fmf{boson}{v1,v3}
\fmf{boson}{v3,v2}
\fmf{dashes}{v3,o3} \fmf{dashes}{v3,o6}
\fmflabel{$H$}{o3} \fmflabel{$H$}{o6}
\fmffreeze
\end{fmfgraph*}
\end{fmfshrink}
}
\\[2\baselineskip]
\underline{$gg$ double-Higgs fusion: $gg\to HH$}\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{10mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(30,12)
\fmfstraight
\fmfleftn{i}{2} \fmfrightn{o}{2}
\fmflabel{$g$}{i1} \fmflabel{$g$}{i2}
\fmf{gluon,tens=2/3}{i1,v1} \fmf{phantom}{v1,v2,v3,o1}
\fmf{gluon,tens=2/3}{w1,i2} \fmf{phantom}{w1,w2,w3,o2}
\fmffreeze
\fmf{fermion}{w1,x2,v1}
\fmf{dashes, lab=$H$}{x2,x3}
\fmf{dashes}{o1,x3,o2}
\fmffreeze
\fmf{fermion,label=$t$,label.s=left}{v1,w1}
\fmflabel{$H$}{o1} \fmflabel{$H$}{o2}
\fmfdot{x3}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(30,12)
\fmfstraight
\fmfleftn{i}{2} \fmfrightn{o}{2}
\fmf{gluon}{i1,v1} \fmf{phantom}{v1,v3} \fmf{dashes}{v3,o1}
\fmf{gluon}{w1,i2} \fmf{phantom}{w1,w3} \fmf{dashes}{w3,o2}
\fmffreeze
\fmf{fermion}{v1,w1,w3,v3,v1}
\fmflabel{$H$}{o1} \fmflabel{$H$}{o2}
\end{fmfgraph*}
\end{fmfshrink}
}
\end{flushleft}
\caption{\textit{%
Processes contributing to Higgs-pair production in the Standard Model
at the LHC: double Higgs-strahlung, $WW/ZZ$ fusion, and $gg$ fusion
(generic diagrams).
}}
\label{smdiag}
\end{figure}
They can be expressed in terms of the two mixing angles $\beta$ and
$\alpha$, mixing angle in the CP-even Higgs sector. The couplings
involving two light Higgs bosons, for example, are given by the
trigonometric functions
\begin{eqnarray}
\lambda_{hhh} &=& 3 \cos2\alpha \sin (\beta+\alpha)
+ 3 \frac{\epsilon}{M_Z^2} \frac{\cos \alpha}{\sin\beta} \cos^2\alpha
\\
\lambda_{Hhh} &=& 2\sin2 \alpha \sin (\beta+\alpha) -\cos 2\alpha
\cos(\beta
+ \alpha) + 3 \frac{\epsilon}{M_Z^2} \frac{\sin \alpha}{\sin\beta}
\cos^2\alpha \nonumber
\label{coup}
\end{eqnarray}
The couplings are defined in units of $\lambda_0$. They are
renormalized indirectly by the re\-nor\-ma\-li\-za\-tion of the mixing
angle $\alpha$ and directly by additive terms proportional to the
radiative correction parameter $\epsilon$ which, to leading order, is
given by $\epsilon = 3 G_F M_t^4/(\sqrt{2}\pi^2\sin^2\beta)\cdot
\ln(M^2_{\tilde{t}}/M_t^2)$ \cite{okada}. In the subsequent numerical
analysis the complete one-loop and the leading two-loop corrections to
the Higgs masses and couplings \cite{carena} are included . The size
of the couplings has been exemplified for a set of parameters in
Ref.\cite{hhhlc}. \s
The trilinear MSSM Higgs couplings are involved in a large number of
multi-Higgs processes at the LHC \@ifnextchar [{\@tempswatrue\@citexr}{\@tempswafalse\@citexr[]}{conf,plehn}:
\begin{eqnarray}
\begin{array}{l@{:\quad}l@{\;\to\;}l l l l}
\mbox{double Higgs-strahlung} & $$q\bar{q}$$ & W/Z + H_i H_j & \mathrm{and}
& W/Z + AA & [H_{i,j}=h,H] \nonumber\\[0.2cm]
\mbox{triple Higgs production} & $$q\bar{q}$$ & AH_i H_j & \mathrm{and}
& AAA & \nonumber \\[0.2cm]
WW/ZZ\ \mbox{double-Higgs fusion} & qq & qq+ H_i H_j & \mathrm{and}
& qq+ AA & \nonumber \\[0.2cm]
gg\ \mbox{fusion} & gg & H_i H_j,\quad H_iA & \mathrm{and}
& AA & \nonumber
\end{array}
\end{eqnarray}
In this analysis we will restrict ourselves to a specific class of
final states while a comprehensive description of all processes will
be deferred to a subsequent report \cite{muehl}; we will study final
states involving two light Higgs bosons $h$:
\begin{eqnarray}
pp &\to& gg \to hh \\
pp &\to& Z/W + hh \qquad {\rm and} \quad A+hh \nonumber
\end{eqnarray}
They are generated either in the continuum or, for moderate values of
$\tan\beta$, in cascade decays, too \cite{djoukal}:
\begin{eqnarray}
H \to hh, \quad A \to Zh \quad {\rm and} \quad H^{\pm} \to W^{\pm}
h
\end{eqnarray}
A set of typical diagrams is shown in Fig.\ref{fig:mssm}. We will also
present selected results on cascade decays involving heavy Higgs
bosons $H$ in the final state; they can be generated in the chains
\begin{eqnarray}
q\bar{q} &\to& Z^* \to AH \to ZHh \\
q\bar{q} &\to& W^* \to H^{\pm}H \to WHh \nonumber
\end{eqnarray}
These chains give rise to large rates, yet they do not involve
trilinear Higgs couplings but only gauge couplings. The corresponding
diagrams are shown in Fig.\ref{nocoup}.\s
\begin{figure}
\begin{flushleft}
\underline{double Higgs-strahlung: $q\bar q\to Zhh/Whh$}\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{5mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmflabel{$q$}{i1} \fmflabel{$\bar q$}{i3}
\fmf{boson,tens=3/2,label=$W/Z$, label.s=left}{v1,v2}
\fmf{boson}{v2,o3} \fmflabel{$W/Z$}{o3}
\fmf{phantom}{v2,o1}
\fmffreeze
\fmf{dashes,lab=$h,,H$,lab.s=right}{v2,v3} \fmf{dashes}{v3,o1}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$h$}{o2} \fmflabel{$h$}{o1}
\fmfdot{v3}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmf{boson,tens=3/2}{v1,v2}
\fmf{dashes}{v2,o1}
\fmf{phantom}{v2,o3}
\fmffreeze
\fmf{dashes,lab=$H^\pm/A$,lab.s=left}{v2,v3}
\fmf{boson}{v3,o3} \fmflabel{$W/Z$}{o3}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$h$}{o2} \fmflabel{$h$}{o1}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmf{boson,tens=3/2}{v1,v2}
\fmf{dashes}{v2,o1}
\fmf{phantom}{v2,o3}
\fmffreeze
\fmf{boson}{v2,v3,o3} \fmflabel{$W/Z$}{o3}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$h$}{o2} \fmflabel{$h$}{o1}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmf{boson,tens=3/2}{v1,v2}
\fmf{dashes}{v2,o1} \fmflabel{$h$}{o1}
\fmf{dashes}{v2,o2} \fmflabel{$h$}{o2}
\fmf{boson}{v2,o3} \fmflabel{$W/Z$}{o3}
\end{fmfgraph*}
\end{fmfshrink}
}
\\[2\baselineskip]
\underline{triple Higgs production: $q\bar q\to Ahh$}
\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{5mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmflabel{$q$}{i1} \fmflabel{$\bar q$}{i3}
\fmf{boson,tens=3/2,label=$Z$, label.s=left}{v1,v2}
\fmf{dashes}{v2,o3} \fmflabel{$A$}{o3}
\fmf{phantom}{v2,o1}
\fmffreeze
\fmf{dashes,lab=$H,,h$,lab.s=right}{v2,v3} \fmf{dashes}{v3,o1}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$h$}{o2} \fmflabel{$h$}{o1}
\fmfdot{v3}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmf{boson,tens=3/2}{v1,v2}
\fmf{dashes}{v2,o1} \fmflabel{$h$}{o1}
\fmf{phantom}{v2,o3}
\fmffreeze
\fmf{dashes,lab=$A$,lab.s=left}{v2,v3}
\fmf{dashes}{v3,o3}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$h$}{o2} \fmflabel{$A$}{o3}
\fmfdot{v3}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmf{boson,tens=3/2}{v1,v2}
\fmf{dashes}{v2,o1}
\fmf{phantom}{v2,o3}
\fmffreeze
\fmf{boson}{v2,v3}
\fmf{dashes}{v3,o3} \fmflabel{$A$}{o3}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$h$}{o2} \fmflabel{$h$}{o1}
\end{fmfgraph*}
\end{fmfshrink}
}
\\[2\baselineskip]
\underline{$WW/ZZ$ double-Higgs fusion: $qq\to qqhh$}
\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{5mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(24,20)
\fmfstraight
\fmfleftn{i}{8} \fmfrightn{o}{8}
\fmf{fermion,tens=3/2}{i2,v1} \fmf{phantom}{v1,o2}
\fmf{fermion,tens=3/2}{i7,v2} \fmf{phantom}{v2,o7}
\fmflabel{$q$}{i2} \fmflabel{$q$}{i7}
\fmffreeze
\fmf{fermion}{v1,o1}
\fmf{fermion}{v2,o8}
\fmf{boson}{v1,v3}
\fmf{boson}{v3,v2}
\fmf{dashes,lab=$H,,h$}{v3,v4}
\fmf{dashes}{v4,o3} \fmf{dashes}{v4,o6}
\fmflabel{$h$}{o3} \fmflabel{$h$}{o6}
\fmffreeze
\fmf{phantom,lab=$W/Z$,lab.s=left}{v1,x1} \fmf{phantom}{x1,v3}
\fmf{phantom,lab=$W/Z$,lab.s=left}{x2,v2} \fmf{phantom}{v3,x2}
\fmfdot{v4}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,20)
\fmfstraight
\fmfleftn{i}{8} \fmfrightn{o}{8}
\fmf{fermion,tens=3/2}{i2,v1} \fmf{phantom}{v1,o2}
\fmf{fermion,tens=3/2}{i7,v2} \fmf{phantom}{v2,o7}
\fmffreeze
\fmf{fermion}{v1,o1}
\fmf{fermion}{v2,o8}
\fmf{boson}{v1,v3}
\fmf{boson}{v4,v2}
\fmf{boson,lab=$W/Z$,lab.s=left}{v3,v4}
\fmf{dashes}{v3,o3} \fmf{dashes}{v4,o6}
\fmflabel{$h$}{o3} \fmflabel{$h$}{o6}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,20)
\fmfstraight
\fmfleftn{i}{8} \fmfrightn{o}{8}
\fmf{fermion,tens=3/2}{i2,v1} \fmf{phantom}{v1,o2}
\fmf{fermion,tens=3/2}{i7,v2} \fmf{phantom}{v2,o7}
\fmffreeze
\fmf{fermion}{v1,o1}
\fmf{fermion}{v2,o8}
\fmf{boson}{v1,v3}
\fmf{boson}{v4,v2}
\fmf{dashes,lab=$H^\pm/A$,lab.s=left}{v3,v4}
\fmf{dashes}{v3,o3} \fmf{dashes}{v4,o6}
\fmflabel{$h$}{o3} \fmflabel{$h$}{o6}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(24,20)
\fmfstraight
\fmfleftn{i}{8} \fmfrightn{o}{8}
\fmf{fermion,tens=3/2}{i2,v1} \fmf{phantom}{v1,o2}
\fmf{fermion,tens=3/2}{i7,v2} \fmf{phantom}{v2,o7}
\fmffreeze
\fmf{fermion}{v1,o1}
\fmf{fermion}{v2,o8}
\fmf{boson}{v1,v3}
\fmf{boson}{v3,v2}
\fmf{dashes}{v3,o3} \fmf{dashes}{v3,o6}
\fmflabel{$h$}{o3} \fmflabel{$h$}{o6}
\end{fmfgraph*}
\end{fmfshrink}
}
\\[2\baselineskip]
\underline{$gg$ double-Higgs fusion: $gg\to hh$}\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{5mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(30,12)
\fmfstraight
\fmfleftn{i}{2} \fmfrightn{o}{2}
\fmflabel{$g$}{i1} \fmflabel{$g$}{i2}
\fmf{gluon,tens=2/3}{i1,v1} \fmf{phantom}{v1,v2,v3,o1}
\fmf{gluon,tens=2/3}{w1,i2} \fmf{phantom}{w1,w2,w3,o2}
\fmffreeze
\fmf{fermion}{w1,x2,v1}
\fmf{dashes, lab=$h,,H$}{x2,x3}
\fmf{dashes}{o1,x3,o2}
\fmffreeze
\fmf{fermion,label=$t,,b$,label.s=left}{v1,w1}
\fmflabel{$h$}{o1} \fmflabel{$h$}{o2}
\fmfdot{x3}
\end{fmfgraph*}
\hspace{15mm}
\begin{fmfgraph*}(30,12)
\fmfstraight
\fmfleftn{i}{2} \fmfrightn{o}{2}
\fmf{gluon}{i1,v1} \fmf{phantom}{v1,v3} \fmf{dashes}{v3,o1}
\fmf{gluon}{w1,i2} \fmf{phantom}{w1,w3} \fmf{dashes}{w3,o2}
\fmffreeze
\fmf{fermion}{v1,w1,w3,v3,v1}
\fmflabel{$h$}{o1} \fmflabel{$h$}{o2}
\end{fmfgraph*}
\end{fmfshrink}
}
\end{flushleft}
\caption{\textit{%
Processes contributing to double and triple Higgs production involving
trilinear couplings in the MSSM.
}}
\label{fig:mssm}
\end{figure}
\begin{figure}
\begin{flushleft}
\underline{cascade decay: $q\bar q\to AH/HH^\pm \to ZHh/WHh$}\\[1.5\baselineskip]
{\footnotesize
\unitlength1mm
\hspace{5mm}
\begin{fmfshrink}{0.7}
\begin{fmfgraph*}(24,12)
\fmfstraight
\fmfleftn{i}{3} \fmfrightn{o}{3}
\fmf{fermion}{i1,v1,i3}
\fmflabel{$q$}{i1} \fmflabel{$\bar q$}{i3}
\fmf{boson,tens=3/2,lab=$W/Z$,lab.s=right}{v1,v2}
\fmf{dashes}{v2,o1}
\fmf{phantom}{v2,o3}
\fmffreeze
\fmf{dashes,lab=$H^\pm/A$,lab.s=left}{v2,v3}
\fmf{boson}{v3,o3} \fmflabel{$W/Z$}{o3}
\fmffreeze
\fmf{dashes}{v3,o2}
\fmflabel{$h$}{o2} \fmflabel{$H$}{o1}
\end{fmfgraph*}
\end{fmfshrink}
}
\end{flushleft}
\caption{\textit{%
Processes which contribute to double light plus heavy Higgs production
in the MSSM but which do not involve trilinear couplings.
}}
\label{nocoup}
\end{figure}
The cross sections for continuum production are generally small and it
will be difficult to discriminate the signal from the background, after
the decay of the Higgs particles into pairs of $b$ quarks, for
instance. Cascade decays, on the other hand, have been proposed to
search for these particles at the LHC \cite{richter}. \s
The present paper has got a limited goal. We have built up the general
theoretical formalism for multiple Higgs production at the LHC in the
Standard Model and the MSSM, and we discuss a few examples in detail,
see also Ref.\cite{muehl}. Just setting the base for these processes
we do not intend to consider background reactions in a systematic way;
such simulations can only be carried out by taking proper account of
detector properties, what is beyond the scope of this paper.
\subsection*{2. Higgs Pairs in the Standard Model}
The cross sections for double Higgs-strahlung off $W/Z$ bosons and for
vector-boson fusion can be evaluated, {\it mutatis mutandis}, at the
quark level for the LHC in the same way as for $e^+e^-$ collisions,
cf.~Ref.\cite{hhhlc}; just the couplings have to be adjusted
properly. The proton cross sections are derived by folding the parton
cross sections $\hat{\sigma}(qq'\to HH;\hat{s})$ of the quark
subprocesses with the appropriate luminosities $d{\cal L}^{qq'}/d\tau$:
\begin{eqnarray}
\sigma (pp\to HH) = \int_{4M_H^2/s}^1 d\tau
\frac{d{\cal L}^{qq'}}{d\tau}
\hat{\sigma} (qq' \to HH; \hat{s} = \tau s)
\end{eqnarray}
where
\begin{eqnarray}
\frac{d{\cal L}^{qq'}}{d\tau} = \int_\tau^1 \frac{dx}{x}
q(x;Q^2) q'(\tau/x;Q^2)
\end{eqnarray}
with $q$ and $q'$ denoting the parton densities in the proton
\cite{pdflib}, taken at a typical scale $Q\sim M_H$.\s
\begin{figure}
\begin{center}
\epsfig{figure=sm.eps,width=13cm}
\caption{The cross sections for gluon fusion, $WW/ZZ$ fusion and
double Higgs-strahlung $WHH$, $ZHH$ in the SM. The vertical arrows
correspond to a variation of the trilinear Higgs coupling from $1/2$
to $3/2$ of the SM value.}
\label{fig:SM}
\end{center}
\end{figure}
The large number of gluons in high-energy proton beams provides an
additional mechanism for the production of Higgs pairs: gluon fusion
$gg \to HH$ \cite{glover}. The proton cross section is derived by
folding the parton cross section $\hat{\sigma}(gg\to HH)$ with the
gluon luminosity. The coupling between gluons and SM Higgs bosons is
mediated by heavy top-quark loops. As expected from single Higgs
production \cite{mspira}, QCD radiative corrections are particularly
important for this channel. They have been determined in the
low-energy limit of small Higgs masses $M_H^2 \ll 4M_t^2$, leading to
a $K$ factor $K\approx 1.9$ \cite{dawson}. A $K$ factor of similar
size is generally expected for Higgs masses beyond the top-quark
threshold.\s
The cross sections are shown in Fig.\ref{fig:SM} for the intermediate
Higgs mass range discussed above. Gluon fusion dominates over the
other mechanisms. The $WW/ZZ$ fusion mechanisms are the next important
channels. In addition to the four $b$ jets, the four $W W^{(*)} W
W^{(*)}$ bosons or the mixed $bbWW^{(*)}$ pairs generated in the
decays of the two Higgs bosons, the light-quark jets associated with
the equivalent $W/Z$ bosons in the fragmentation, $q\to W/Z +q$ can be
exploited to tag fusion events; these jets are emitted at average
transverse momenta $p_{T} \sim 1/2 M_{W/Z}$. $WW$ fusion dominates
over $ZZ$ fusion at a ratio $WW$:$ZZ\approx 2.3$. The cross sections for
double Higgs-strahlung are relatively small. This follows from the
scaling behavior of the cross sections which drop $\sim 1/\hat{s}$.
The cross sections for Higgs-strahlung off $W$ and $Z$ bosons are
combined in Fig.\ref{fig:SM}; their relative size is close to
$W/Z\approx 1.6$. The vertical arrows indicate the sensitivity of the
cross sections to the size of the trilinear Higgs coupling; they
correspond to a modification of the trilinear SM coupling
$\lambda_{HHH}$ by the {\it ad hoc} rescaling coefficient
$\kappa=1/2\to3/2$.
\subsection*{3. Higgs Pairs in Supersymmetric Theories}
With appropriate modifications, the pattern of MSSM Higgs
pair-production at the LHC is similar to the characteristics in
$e^+e^-$ collisions \cite{hhhlc,djouadi}. An important exception
however is the additional gluon-fusion channel \cite{plehn}. \s
\begin{figure}
\begin{center}
\epsfig{figure=tan3.eps,width=13cm}
\end{center}
Figure 5a:{\it Total cross sections for MSSM hh production via double
Higgs-strahlung $Whh$ and $Zhh$, $WW/ZZ$ fusion and gluon fusion at the
LHC for $\tan\beta=3$, including mixing effects ($A=1$~TeV,
$\mu=-1$~TeV).}
\end{figure}
\noindent
{\bf 1.} We will focus on the production of pairs of light Higgs
bosons: $pp \to hh$. For moderate values of $\tan\beta$, the $hh$
production channels follow the pattern of the Standard Model, with
gluon fusion being dominant, Fig.5a. However, within the cascade-decay
regions of the heavy Higgs bosons $H$, $H^{\pm}$ the cross sections
rise dramatically. These domains are marked in the figures explicitly
by arrows. Large contributions to the cross sections are generated by
heavy Higgs formation $gg/VV \to H \to hh$ in the fusion channels, and
$H^{\pm} \to W^{\pm} h$ decay in Higgs-strahlung $W^{\pm*} \to H^{\pm}
h\to W^{\pm} hh$. As expected \cite{plehn}, the gluon-fusion $hh$
cross section becomes very large in the $H$ decay region, giving rise
to a sample of about a million $hh$ events. This process therefore
provides an important channel for searching for MSSM Higgs bosons at
the LHC \cite{richter}. The sensitivity of the cross sections with
regard to a variation of $\lambda_{hhh}$ by the rescaling factor
$\kappa = 1/2$ to $3/2$ is close to ten per-cent in the continuum
while the sensitivity of $H$ cascade decays to a variation of
$\lambda_{Hhh}$ is indicated by arrows. \s
\begin{figure}
\begin{center}
\epsfig{figure=tan50.eps,width=13cm}
\end{center}
Figure 5b:{\it Total cross sections for MSSM hh production via double
Higgs-strahlung $Whh$, $Zhh$, $WW/ZZ$ fusion and gluon fusion at the
LHC for $\tan\beta=50$, including mixing effects ($A=1$~TeV,
$\mu=1$~TeV).}
\end{figure}
\setcounter{figure}{5}
For large $\tan\beta$, a huge ensemble of $hh$ continuum events is
generated by gluon fusion, Fig.5b. The enhancement is due to the
large $hbb$ Yukawa coupling, $\sim m_b \tan\beta$, in the $b$-quark
loops connecting the gluons with the Higgs bosons. Since the box
diagrams are enhanced quadratically compared to the triangle diagrams,
the sensitivity to the trilinear coupling is small.\footnote{The
multi-$b$ final states $pp \to hh \to (b\bar{b})(b\bar{b})$ with two
resonance structures and large transverse momenta provide an
outstanding signature which may be exploited to search for $h$
Higgs bosons in the range of large $\tan \beta$ (and moderate $M_A$)
not covered hitherto at LHC in standard channels.} The continuum
cross sections of the $VV$ fusion and Higgs-strahlung channels are
suppressed with respect to the Standard Model until the decoupling
limit is reached. For large $\tan\beta$ cascade decays do not play a
role in $hh$ pair production; the kinematical decay thresholds are
reached only for masses for which the decoupling limit is being
approached. \vspace{\baselineskip}
\begin{figure}
\begin{center}
\epsfig{figure=res3.eps,width=13cm}
\caption{\it Total cross sections for MSSM $Hh$ production in the
processes $WHh$ and $ZHh$ for $\tan\beta=3$, including mixing effects
($A=1$~TeV, $\mu=-1$~TeV).}
\label{heavy}
\end{center}
\end{figure}
\noindent
{\bf 2.} Two processes involve Higgs-pair final states including a
light plus a heavy Higgs boson. However for cross sections in excess
of 1~fb, Fig.\ref{heavy}, the final states are generated in cascade
decays by gauge interactions:
\begin{eqnarray}
\begin{array}{l l l l l}
pp & \hspace{-0.1cm} \to & AH &
\hspace{-0.3cm} \to &\hspace{-0.1cm} ZHh \\
\\[-0.8cm]
& \hspace{-0.1cm} \scriptstyle{Z}& & & \\[0.1cm]
pp & \hspace{-0.1cm} \to & H^{\pm} H &
\hspace{-0.3cm} \to & \hspace{-0.1cm} W^{\pm} Hh
\\ \\[-0.8cm]
& \hspace{-0.1cm} \scriptstyle{W}& & &
\end{array}
\end{eqnarray}
These processes are therefore not suitable for measuring trilinear
Higgs couplings.
\subsection*{4. Conclusions}
In the present paper we have analyzed the production of neutral
Higgs-bosons pairs in various channels at the LHC which can eventually
be used to measure fundamental trilinear Higgs self-couplings. In a
first step we have compared the production cross sections in the
Standard Model, assuming a Higgs-boson mass in the intermediate range.
Moreover, we have calculated the cross sections for pairs of light
Higgs bosons in the Minimal Supersymmetric extension of the Standard
Model. Earlier results haven been combined with new calculations in
these analyses.\s
The continuum cross sections are generally small in the SM and MSSM,
yet not for gluon fusion. The trilinear SM Higgs coupling and the
trilinear coupling $\lambda_{hhh}$ in the MSSM may thus be accessible
experimentally, provided the backgrounds can be rejected sufficiently
well. If Higgs cascade decays $H\to hh$ occur in the MSSM, they can be
exploited to measure one of the couplings between light and heavy
CP-even neutral Higgs bosons, $\lambda_{Hhh}$. \vspace{\baselineskip}
\subsubsection*{Acknowledgements}
We are grateful to M.~Spira for discussions and for providing us with a
source code for gluon fusion of Higgs pairs.
\end{fmffile}
\newpage
\baselineskip15pt
|
\section{Introduction}
Two branches in general relativity have met with varying
degrees of success over the years. One is the use of strings
to model, or reflect, the structure of matter. This topic,
while still relatively young, has provided substantial hope
into a successful program of quantum gravity,\footnotemark\ and
generates great interest at the classical level as well.\footnotemark\
The other branch reaches further back and modifies Einstein's
original theory by taking the affine connection to be
asymmetric. The antisymmetric part of the affine
connection --- the torsion, has a twisted history,
but gained renewed popularity in the 1970s when it was
used in developing a local Poincar\`e gauge theory
of gravity.\footnotemark\ The interest in torsion gained a wider audience
when it was shown that it could could act as the antisymmetric
field that is required in string theory,\footnotemark\
and gained further
notice when it appeared in supergravity theories.\footnotemark\
The two once disparate branches have become intertwined,
and the purpose here is to tap into the synergism
generated by this growth.
One of the underlying bonds that seems to unite string theory
and general relativity with torsion is the assumption
that the torsion may be derived from an antisymmetric
potential $\psi_{\mu\nu}$ according to
\begin{equation}\label{tordef}
S_{\mu\nu\sigma}=\psi_{[\mu\nu,\sigma]}
.\end{equation}
When this assumption was used to develop a theory of gravitation
with torsion, the physical interpretation that fell upon torsion
was that it was created from intrinsic spin.\footnotemark\ The interpretation
resulted from an analysis of the conservation laws of the
equations of motion. Some of the salient features that were
discussed and shown in \cite{sg} are the following. It was shown
that it was necessary to introduce an intrinsic vector $\xi^\mu$
to represent the source for torsion. This may be viewed as a generalization
of other intrinsic quantities used in describing sources. For
example, to describe a point particle we may consider its mass
$m$ as an intrinsic quantity, or its charge $q$ as an
intrinsic quantity. The source tensors are built up from these
quantities. In order to couple a source to an antisymmetric
field, as is the case for torsion, it is necessary
to go beyond scalar intrinsic quantities and adopt the intrinsic
vector approach. This gave rise to several aspects of the theory.
One is that intrinsic spin is derived from this intrinsic
vector. Thus, intrinsic spin arises, not from motion or rotation,
but from structure and persists in the limit of a static configuration.
Another result was that the particle
had to have structure, and labeling each point on the structure
by a vector $\xi^\mu_n$, it was shown that the conservation laws
imply $\sum_n\xi^\mu_n=0$. The conservation laws also
predicted a definite spin interaction term, that in principle,
could be observed. This interaction term was confirmed when
the Dirac Lagrangian was used in place of the intrinsic
vector material action. In the low energy limit, the Dirac equation
yielded that same interaction as that predicted by the intrinsic
vector approach,\footnotemark\ and this was used to place an upper
bound on the coupling constant.\footnotemark\
In addition, it was shown that a scalar
arises naturally, \footnotemark\ and it was also shown that this scalar
field might be interpreted as the scalar field of string theory.\footnotemark\
Thus, the theory of gravitation with torsion with (\ref{tordef})
adopts many characteristics of string theory --- from the mirrored
Lagrangian of the low energy string theory limit, to a source
which must be at least one dimensional. It is natural then to
consider replacing the intrinsic vector material Lagrangian by
that of a string. In this case, many plausible results follow.
First, the intrinsic vector will be seen to correspond to the
tangent vector of the string. Also the condition given above that
the intrinsic vectors sum to zero, is replaced by the much more
natural result $\oint d\zeta=0$, which is true for closed strings.
Moreover, there is a natural coupling between the torsion
potential and the area swept out by the string. It will be shown
that many of the results of the intrinsic vector approach will be
reproduced by use of the string source, and in fact, in many ways,
do so in a much more natural fashion. It will also be shown that
strings have intrinsic spin, which follows from their structure
(not motion), and the equations of motion will follow from the
conservation laws. In what follows only closed strings will be
considered, the discussion of open strings will be considered
separately. Finally, I would like to give one more preview of
things to come. It is customary to obtain the equation of motion
of a string by adopting the geodesic postulate, i.e., setting the
variations of the action with respect to the coordinate equal to
zero, or in a more rigorous approach, using invariance of the
material action.\footnotemark\ However, in general relativity the equations
of motion of an object follow from the field equations, and
variations with respect to the coordinate will not be considered.
Thus, certain conditions or properties of the string well known
from string theory are not necessarily appropriate to this case.
\section{Field equations and the conservation laws}
\subsection{Field equations}
The material in this section is general in the sense that
no explicit form of the energy momentum tensor is given.
Details on this material may be found in \cite{sg}.
The field equations are given by the action principle
\begin{equation}
\delta (I_g +I_m)=0
\end{equation}
where
\begin{equation}\label{R}
I_g=\int\sqrt{-g}{R\over2k}d^4x
,\end{equation}
$I_m$ is the material action, and $ k=8\pi$.
The curvature scalar $R$ is that of $U_4$ spacetime,
and therefore contains torsion. The
unknown quantities are the 10 components of the symmetric metric
tensor and the 6 components of the antisymmetric torsion
potential. The equations are obtained by considering independent
variations of these potentials. (One may note that this action
principle yields second order differential equations in the torsion
potential, whereas
an action given by (\ref{R}), but with variations taken with respect
to the torsion itself, yields non-propagating torsion.)
However, an equivalent procedure
is, defining $\phi_{\mu\nu}= g_{\mu\nu}+\psi_{\mu\nu}$, to consider
variations with respect to $\phi_{\mu\nu}$. Now, taking variations
with respect to $\phi_{\mu\nu}$ gives a set of 16 equations. The
symmetric part is equivalent to those obtained by considering
variations with respect to the metric tensor, and will be called
the gravitational field equations. The antisymmetric part is
equivalent to the equations obtained by performing variations with
respect to the torsion potential and will be called the torsional
field equations. In this case, a non-symmetric energy momentum
tensor was defined (see \cite{sg}) according to
\begin{equation}\label{emt}
\delta I_m={1\over2}\int d^4x\sqrt{-g}T^{\mu\nu}\delta \phi_{\mu\nu}
.\end{equation}
The symmetric part of the energy momentum tensor in
(\ref{emt}) is the source in the gravitational field equations,
and the antisymmetric part is the source in the torsional field
equations. With these remarks, the field equations are given by
\begin{equation}\label{fe}
G^{\mu \nu} - 3S^{\mu \nu \sigma}_{\ \ \ \ ; \sigma}
-2S^{\mu}_{\ \alpha \beta}S^{\nu \alpha \beta} = kT^{\mu \nu}
\end{equation}
where
\begin{equation}
G^{\mu\nu}=R^{\mu\nu}-{1\over2}g^{\mu\nu}R
\end{equation}
and $R^{\mu\nu}$ is the (asymmetric) Ricci tensor in $U_4$ spacetime.
In line with the above remarks, the torsional field equations
are given by
\begin{equation}\label{tfe}
S^{\mu \nu \sigma}_{\ \ \ \ ;\sigma} =-kj^{\mu \nu}
\end{equation}
where $j^{\mu\nu}\equiv (1/2)T^{[\mu\nu]}$.
In the above, and below, it is convenient to use two
different kinds of covariant differentiation. First, the
fundamental definition of a covariant derivative is given
by, for any vector $A^\mu$,
\begin{equation}
\nabla_\sigma A^\mu=A^\mu_{\ ,\sigma}+ \Gamma_{\sigma \nu}^{\ \ \ \mu}A^\nu
\end{equation}
which contains the full (asymmetric) affine connection
$\Gamma_{\sigma \nu}^{\ \ \ \mu}$.
However, sometimes the antisymmetric part drops out, and it is useful
to define the ``Christofell'' covariant derivative by
\begin{equation}
A^\mu_{\ ;\sigma}=A^\mu_{\ ,\sigma}+ \{_{\sigma \nu}^{\; \mu}\}A^\nu
\end{equation}
where $\{_{\sigma \nu}^{\; \mu}\}$ is the (symmetric)
Christofell symbol. The relation between the affine connection
and the Christoffel symbol is obtained by the requirement
$\nabla_\sigma g_{\mu\nu}=0$, which yields
\begin{equation}
\Gamma_{\mu \nu}^{\ \ \ \sigma}=\{_{\mu \nu}^{\; \sigma}\}
+S_{\mu\nu}^{\ \ \ \sigma}+S^\sigma_{\ \mu\nu}+S^\sigma_{\ \nu\mu}
\end{equation}
which becomes, with (\ref{tordef}),
\begin{equation}
\Gamma_{\mu \nu}^{\ \ \ \sigma}=\{_{\mu \nu}^{\; \sigma}\}
+S_{\mu\nu}^{\ \ \ \sigma}
.\end{equation}
\subsection{Conservation laws}
Ultimately, the correctness of any theory can only
be ascertained by experiment. In a theory
of gravitation this means that one must derive
the equations of motion, which predict acceleration,
interaction energy or something that is measurable.
Not only do the equations of motion provide the
means to evaluate the theory, they are also
instrumental in developing the physical interpretation of the theory.
A powerful aspect of general relativity is that the
equations of motion follow from the field equations,
and therefore already establish the machinery for the analysis
of the predictions of the theory. In fact, this comes about in
two ways, each of which yields the same result. One is that
the Bianchi identities must be obeyed, which are,
in $U_4$ spacetime,
\begin{equation}\label{bi}
\nabla_{\nu}G^{\mu \nu} = 2S^{\mu \alpha \beta}R_{\beta \alpha}
-S_{\alpha \beta \gamma}R^{\mu \gamma \beta \alpha}
.\end{equation}
To use the Bianchi identities, one operates with
$\nabla_\nu$ on (\ref{fe}) and uses (\ref{bi}). This establishes
a differential relation on the source. The other way to proceed
is to capitalize on the requirement that the material
action is a scalar, and that under the manifold mapping
$x^\mu\rightarrow x^\mu+\epsilon^\mu$,
\begin{equation}
\int d^4x\sqrt{-g}T^{\mu\nu}{\cal L}_\epsilon\phi_{\mu\nu}=0
\end{equation}
where ${\cal L}_\epsilon$ is the Lie derivative. Either approach
yields
\begin{equation}\label{conlaw}
T^{\mu \nu}_{\ \ \ ; \nu} = {3 \over 2}T^{\alpha \beta}S^{\mu}_{\ \alpha
\beta}
.\end{equation}
This result shows that when the torsion vanishes, we obtain the
conventional result that the covariant
derivative of the energy momentum vanishes.
The equations of motion then follow from this result. With torsion,
(\ref{conlaw}) shows that there will be additional forces due to
torsion, and that geodesic motion should not be expected.
\section{Equations of motion}
\subsection{The general case}
The equations of motion can be found using the method of
Papapetrou.\footnotemark\ At first, the method will be kept general,
meaning that the actual source will not be specified. After that,
the formulation will be used to find the equation of motion
of a small string in an external field. Some general
comments about this method are: It is not generally covariant.
Volume integrals over the test object are considered at
constant $x_0$. Also, under consideration is the motion of a
small test object in the presence of a large object. The gravitational
field of the test object is essentially ignored, except in the way the
inertial mass is defined. These notions will be clarified as we go.
The starting point is the identity
\begin{equation}\label{id}
\tilde T^{\mu\nu}_{\ \ ,\nu}= \sqrt{-g}T^{\mu\nu}_{\ \ \ \ ;\nu}
-\{_{\alpha \beta}^{\; \mu}\} \tilde T^{\alpha\beta}
\end{equation}
where the tilde implies density according to
$\tilde T^{\mu\nu}=\sqrt{-g}T^{\mu\nu}$.
Now, consider a small volume $d^3x$ that completely enclosed the test
object, and integrate this equation over that volume. In that region, the
energy momentum tensor of the large body is zero, so from here on the
energy momentum tensor is that of the small body---the string.
In other words, we suppose that
\begin{equation}
I_m=I_b+I_s
\end{equation}
where $I_b$ is the material action corresponding to
the energy momentum tensor of the big body, and $I_s$
that of the small body so that
\begin{equation}\label{emts}
\delta I_s={1\over2}\int d^4x\sqrt{-g}T^{\mu\nu}_s\delta \phi_{\mu\nu}
,\end{equation}
and from here on the subscript $s$ is dropped.
The metric tensor that appears in these formulas is total
gravitational field of both objects, however, assuming that the large
object has much more mass than the small object, the metric tensor
that appears in (\ref{id}) is approximated by that of the large body.
Discarding surface terms, and using the conservation law
(\ref{conlaw}), (\ref{id}) becomes
\begin{equation}
{d\over dx^0}\int \tilde T^{\mu0}=
\frac3 2\int\tilde T^{\alpha\beta}S^\mu_{\ \alpha\beta}
-\int\tilde T^{\alpha\beta}\{_{\alpha\beta}^{\; \mu}\}
\end{equation}
where the volume element has been, and will be, suppressed.
The following definitions will be useful,
with $\tau^{\alpha\beta}\equiv\tilde T^{(\alpha\beta)}$:
\begin{equation}
M^{\mu \nu} = v^{0}\int \tilde T^{\mu \nu}
,\end{equation}
\begin{equation}
M^{\alpha \mu \nu} = -v^{0}\int \delta x^{\alpha}\tau^{\mu \nu}
\end{equation}
\begin{equation}\label{m}
m^{\alpha \mu \nu} = v^{0}\int \delta x^{\alpha}\tilde j^{\mu \nu}
\end{equation}
\begin{equation}\label{Jdef}
J^{\mu \nu} =\int (\delta x^{\mu}\tilde T^{\nu o} -\delta x^{\nu}
\tilde T^{\mu 0})
.\end{equation}
We consider $x^\alpha$ as the coordinate
from the origin to a point on the
small body. Then $y^\alpha$ is defined according to
$x^\alpha=y^\alpha+\delta x^\alpha$, where $\delta x^\alpha<<y^\alpha$.
To proceed, the Cristoffel symbols and the torsion
tensor are expanded in a Taylor series about the
point $y^\mu$. These quantities may be then taken outside
the integrals, and using the above equations one may show
that
\begin{equation}\label{trans0}
{d \over d\tau}\biggl({M^{\mu 0}\over v^{0}}\biggr) +
\{_{\alpha\beta}^{\ \mu}\}M^{\alpha\beta}=
\{_{\alpha\beta}^{\ \mu}\},_{\eta}
M^{\eta \alpha \beta}
+\frac3 2 M^{\alpha\beta}S^\mu_{\ \alpha\beta}
+3S^{\mu}_{\ \alpha \beta},
_{\eta}m^{\eta \alpha \beta}
.\end{equation}
In the appendix it is shown that we may take
\begin{equation}\label{int0}
\int\tilde j^{\mu\nu}=0
.\end{equation}
With this
\begin{equation}\label{pdef}
p^\mu\equiv{M^{\mu0}\over v^0}=\int\tau^{\mu0}dV
,\end{equation}
and (\ref{trans0}) becomes
\begin{equation}\label{trans1}
{dp^\mu \over d\tau}+
\{_{\alpha\beta}^{\ \mu}\}M^{\alpha\beta}=
\{_{\alpha\beta}^{\ \mu}\},_{\eta}
M^{\eta \alpha \beta} +3S^{\mu}_{\ \alpha \beta}, _{\eta}m^{\eta
\alpha \beta} .\end{equation}
If the torsion vanishes, this equation reduces
to the equation derived by Papapetrou many years ago. The first
term on the right side represents the force on a particle with
structure due the gradient of the field. This gives rise to the
well know result (still no torsion) that only point particles with
no structure follow along geodesics. Any structure to a particle
will give rise to $M^{\alpha \mu \nu}$, which by (\ref{trans1})
gives rise to non-geodesic motion. Thus we anticipate
that strings, even in Riemannian space,
will not follow geodesic motion. The second term
on the right side of (\ref{trans1}) is the force on the particle
due to the torsion. It was shown that when the torsion was
constructed from the intrinsic vector, this force was due to the
interaction of the intrinsic spin of the particle and the external
torsion field.
The Papapetrou method also gives rise to the equation for angular
momentum by starting with the identity
\begin{equation}
(x^\alpha\tilde T^{\beta\gamma})_{,\gamma}=\tilde T^{\beta\alpha}
+x^\alpha\tilde T^{\beta\gamma}_{\ \ ,\gamma}
,\end{equation}
which gives
\begin{equation}
\int\tilde T^{\beta\alpha}={d\over dx^0} \int(y^\alpha+\delta x^\alpha)
\tilde T^{\be0}- \int(y^\alpha+\delta x^\alpha) \tilde T^{\beta\gamma}_{\ \
,\gamma}
.\end{equation}
Using the same kinds of manipulations as above this
gives
\begin{equation}\label{djdt} {d \over d\tau}J^{\alpha
\beta}+{dy^\alpha\over d\tau}{M^{\beta 0} \over v^{0}} -{dy^\beta\over
d\tau}{M^{\alpha 0} \over v^{0}}= 6S^{[\beta}_{\ \mu \nu}m^{\alpha
]\mu \nu} +2\{_{\mu\nu}^{[ \beta}\}M^{\alpha ]\mu \nu}
.\end{equation}
In the limit that the gravitational and torsional field go to zero
the right side of this equation becomes zero.
In addition, in this limit, (\ref{trans1}),
with (\ref{pdef}), show that $M^{\al0}/v^0$ is the constant
momentum. In this case (\ref{djdt}) yields, upon integration,
\begin{equation}\label{jconst}
J^{\alpha\beta}+y^\alpha p^\beta-y^\beta p^\alpha=\mbox {constant}
.\end{equation}
This shows that,
since the second two terms on the left represent the
orbital angular momentum, $J^{\alpha\beta}$ must represent the total
rotational angular momentum plus intrinsic spin.
With the torsion set equal to zero, these results are identical
to Papapetrou's results. The main differences we are about to
encounter below are that torsion is not zero, and the energy
momentum tensor is not symmetric, as explained above.
In fact, one may see already that the non-symmetric part of the energy
momentum tensor enters into $J^{\alpha\beta}$ through its definition,
and therefore we see in general, from this result, that
the antisymmetric part of the energy momentum tensor
represents intrinsic spin.
\subsection{Enter the string}
Now we would like to consider that the energy momentum tensor
that describes the particle discussed above is that of a string, so
that, for a simple Nambu-Goto string we assume
\begin{equation}\label{stac}
I_s=\mu\int\sqrt{-\gamma}d^2\zeta+{\mu\eta\over2}
\int\sqrt{-\gamma}\psi_{\mu\nu}d\sigma^{\mu\nu}
\end{equation}
where
\begin{equation}
d\sigma^{\mu\nu}=\epsilon^{ab}x,_a^\mu x,_b^\nu d^2\zeta
\end{equation}
and $\sqrt{-\gamma}\epsilon^{10}=1$, etc.
The first term is the usual Nambu-Goto action, and is the
conventional way in which the string is introduced
into gravity. In
order to couple something to the torsion potential $\psi_{\mu\nu}$,
an antisymmetric source term must be invoked. A very natural
choice arises with strings, and this is to couple the torsion
potential to the worldsheet area element, as is done above.
This coupling was first used by Kalb and Ramond.\footnotemark\
The string coordinates are labeled by $\zeta^a$ where $a$
and $b$ range from 0 to 1. Due to the coordinate invariance
of the string action we may choose $\zeta^0=x^0$, and call
$\zeta^1=\zeta$. In curved space, we are not allowed therefore
to choose the conformal gauge, and we won't.
Using this and the definition (\ref{emts})
one obtains
\begin{equation}\label{emtst}
T^{\sigma\nu}={\mu\over\sqrt{-g}}\int
d^2\zeta\sqrt{-\gamma}\delta(x-x(\zeta))x^\sigma_{,a}x^\nu_{,b}
(\gamma^{ab}+\eta\epsilon^{ab})
\end{equation}
where
\begin{equation}\label{gab}
\gamma_{ab}=x^\mu_{,a}x^\nu_{,b}g_{\mu\nu}
.\end{equation}
It is worth emphasizing some differences that arise between these
equations and those that appear in string theory or in the study
of cosmic strings. First, of course, this is a completely
classical presentation. However, we do not impose that variations
of the string action with respect to the coordinate vanish. Thus,
the common `geodesic' condition that is often imposed on the string
coordinates, which in curved space this disallows a static rigid string
(from the Nambu-Goto action), is not enforced.
The equation of motion of the string is derived below
from the Bianchi identity. Also, since we chose $x^0=\zeta^0$,
we are not able to arbitrarily choose a gauge for the string
metric $\gamma^{ab}$. In this case, we are considering the equation
of motion of the string in an external field, so $\gamma^{ab}$ is
determined from this external field according to (\ref{gab}). In
fact, from (\ref{gab}) we see that $\gamma_{00}=(v^0)^{-2}$. We also
see that $\gamma_{01}=g_{01}\partial x^1/\partial\zeta$. Assuming that $g_{01}<<
g_{00}, \ g_{11}$, we may ignore the off diagonal components of
the two dimensional metric. Finally, we see that
$\gamma_{11}=g_{mn}(dx^m/d\zeta)(dx^n/d\zeta)$ where $m,n=1-3$. To
calculate this we assume that $g_{mn}$ in the last equation
can be replaced by $\eta_{mn}$. To justify this, we should
examine the equation of motion (\ref{trans1}) where this will
be used. If we limit the discussion to weak fields so
that $g_{\mu\nu}=\eta_{\mu\nu}+h_{\mu\nu}$ where
$h_{\mu\nu}<<\eta_{\mu\nu}$, then we may be content to carry
only terms linear in $h_{\mu\nu}$ in the equation of motion.
Now, since $T^{\mu\nu}$ (through the quantities $M^{\mu\nu}$
etc.,), which contains $\gamma_{ab}$, is multiplied
by $g_{\mu\nu}$ and its derivatives (and products), to this
order it is sufficient to replace
$g_{\mu\nu}$ by $\eta_{\mu\nu}$ in (\ref{emtst}). With this,
in the appendix it is shown that
$\gamma_{11} =-1$, so that
the energy momentum tensor of the string becomes
\begin{equation}\label{emtmink}
\tilde T^{\alpha\beta}={\mu\over v^0}
\int d^2\zeta\delta(x-x(\zeta))[(v^\alpha v^\beta-x'^\alpha x'^\beta)
+\eta(v^\beta x'^\alpha-v^\alpha x'^\beta)]
\end{equation}
where $v^\alpha=dy^\alpha/d\tau$ is the four velocity of the center of mass
and $x'^\alpha=dx^\alpha/d\zeta$.
With this, we can put the equation of motion (\ref{trans1})
can be put into a more recognizable, or useful, form.
For the present we will content ourselves to obtain
the equation of motion in the lowest order. This means that the terms
of the right hand side of (\ref{trans1}) will be neglected
for now. A more detailed examination of these terms will be
reserved for future work.
With (\ref{emtmink}) one obtains
\begin{equation}
M^{\alpha\beta}=\mu\int d\zeta(v^\alpha v^\beta-x'^\alpha x'^\beta)
.\end{equation}
With this, (\ref{trans1}) becomes
\begin{equation}\label{eqm2}
{d\over d\tau}(mv^\sigma)+\{^{\ \sigma}_{\alpha\beta}\}M^{\alpha\beta}=0
.\end{equation}
where
\begin{equation}\label{mass}
m\equiv {M^{00}\over (v^0)^2}=\mu\int d\zeta
.\end{equation}
Now, (\ref{eqm2}) becomes
\begin{equation}\label{eqm3}
{dm\over d\tau}v^\sigma +m {dv^\sigma\over d\tau}
+\{^{\ \sigma}_{\alpha\beta}\}\left( mv^\alpha v^\beta-
\mu \int d\zeta x'^\alpha x'^\beta \right)=0
.\end{equation}
Also defining
\begin{equation}
{Dv^\sigma\over d\tau}={dv^\sigma\over d\tau} +
\{_{\mu \nu}^{\; \sigma}\}v^\mu v^\nu
,\end{equation}
we can use the identity $v_\sigma Dv^\sigma/d\tau=0$, so that (\ref{eqm3})
gives
\begin{equation}
{dm\over d\tau}=\mu v_\sigma\{^{\ \sigma}_{\alpha\beta}\}
\int d\zeta x'^\alpha x'^\beta
\end{equation}
which puts us in the momentarily awkward position
that, even neglecting the terms on the right
side of (\ref{trans1}), we do not have geodesic
motion and the inertial mass is not conserved.
However, consider
\begin{eqnarray}
\int_0^L d\zeta x'^\alpha x'^\beta=
\int dx^\alpha x'^\beta
=-\int x^\alpha d(x'^\beta)=\\ \nonumber
-\int(y^\alpha+\delta x^\alpha) d(x'^\beta)\approx-y^\alpha\int
d(x'^\beta)=0
\end{eqnarray}
where integration by parts was used and it is assumed that
{\em the string is a closed loop} so that the end
point terms contribute nothing,
and the last step follows for a closed loop that has
no kinks. This shows, finally, the inertial mass is conserved
and that (\ref{eqm3}) reduces to
\begin{equation}
{dv^\sigma\over d\tau}
+\{^{\ \sigma}_{\alpha\beta}\}v^\alpha v^\beta=0
.\end{equation}
This result is received as good news for several reasons. First,
of course, the inertial mass is conserved. Second, in this lowest
order approximation, the string moves along a geodesic.
We also see that the structure of the string will cause
its actual motion to deviate from the geodesic, and these
effects can be calculated from (\ref{trans1}). These results
also show that a sensible equation of motion results from the
Bianchi identity (or from the conservation laws) as it should.
Thus, there is no need to make the additional ``geodesic''
postulate, that variations of the matter action
with respect to the coordinates vanish. In fact, they probably
do not.
To understand better the physical significance
of this energy momentum tensor, and in particular to
see that this implies that strings have intrinsic spin,
we start from the spin vector, defined by
\begin{equation}\label{spin0}
S_\gamma={v^\sigma\over2} J^{\alpha\beta}\epsilon_{\sigma\\alpha\beta\gamma}
,\end{equation}
which for low velocities becomes
\begin{equation}\nonumber
S_\gamma\rightarrow\frac1 2J^{\alpha\beta}\epsilon_{0\alpha\beta\gamma}
,\end{equation} and make the same
approximations for $T^{\alpha\beta}$ as we did above. When
(\ref{emtmink}) is used in (\ref{Jdef}), which is used in
(\ref{spin0}), two kinds of terms arise, those that depend on
velocity (3-velocity) and those that do not. The velocity
dependent terms represent a conventional $\bm{r}\times\bm{p}$
angular momentum of string, about its center of mass, due to its
oscillations or rotations. If we restrict our attention to the rigid
loop, these terms vanish. Even if the string is not rigid, but
is small, i.e., elementary
particle size, this term is negligible. Thus, we restrict
our attention to the terms that do not depend on
velocity, and obtain, for the circular loop,
\begin{equation}\label{spinloop}
S_\gamma=2\int d\zeta\epsilon_{0\alpha\beta\gamma}\delta x^\alpha{dx^\beta\over d\zeta}
,\end{equation}
or
\begin{equation}\label{spinloop3}
{\bm S}=\mu\eta\int\bm{r}\times\bm{r}'d\zeta\ \ \Rightarrow
S=2\mu\eta\times\mbox{Area}
\end{equation} where ${\bm r'}=d{\bm r}/d\zeta$, and
${\bm r}$ is a vector from the center the center of mass to a
point on the string.
This result shows that strings, due to the fact that they have structure,
give rise to intrinsic spin when they are coupled to the torsion
potential.
(One may note that the conventional method of discussing
a spinning string in string theory is described by the introduction
the Grassmannian $\psi$, leading to the Raymond model
or the Neveu-Schwarz.\footnotemark\
Here, we will have intrinsic spin from the Nambu-Goto action alone,
due only to its structure.)
Now we may turn our attention to the torsion
field they generate.
\section{Torsion from a string}
The adoption of the string action (\ref{stac}) has two significant
consequences. One, as shown above, ties intrinsic spin to the string.
The other consequence, and in fact the original motivation in
adopting (\ref{stac}), is that it acts a source of torsion.
In this section, we shall consider the Minkowski
limit, and solve the torsional field equations in this limit.
Of course, since torsion enters in the gravitational field
equations, strictly speaking, torsion cannot exist without
a concomitant gravitational field. However, as has been shown, when
torsion arises from intrinsic spin the torsion field is very small
and Minkowski spacetime is a very good approximation.
It is helpful to use the dual to torsion
\begin{equation}\label{bmu}
b_{\mu} = \epsilon_{\mu \alpha \beta \gamma}S^{\alpha \beta \gamma}
\end{equation}
and $\bm{b}=(b_n)$.
With this in mind (\ref{tfe}) becomes
\begin{equation} \Box\psi^{\mu\nu}=-3k j^{\mu\nu}
.\end{equation}
Now we consider the static case, so that this reduces to
\begin{equation}
\label{lap} \nabla^2\bm{A}=\bm{N}
\end{equation} where
\begin{equation}
A_n\equiv2\psi_{0n},\ \ \ \ \ \ \bm{A}=(A_n)
\end{equation} and
\begin{equation}
N^n\equiv6kj^{0n},\ \ \ \ \ \ \bm{N}=(N^n)
\end{equation} and a `Lorentz'
gauge is chosen so that \begin{equation} \psi^{\sigma\mu}_{\ \ ,\sigma}=0
,\end{equation}
which is allowed due to the gauge invariance
$\psi_{\mu\nu}\rightarrow\psi_{\mu\nu}+\xi_{[\mu,\nu]}$.
Ordinary Green's function techniques may now be used to give the
solution to ({\ref{lap}) as
\begin{equation}
\bm{A}(x)={1\over4\pi}\int
G(\bm{x}-\bm{y})\bm{N}(\bm{y})d^3y
\end{equation} where
\begin{equation}
G={1\over|\bm{x}-\bm{y}|}=\frac1 x+{\bm{x}\cdot\bm{y}\over
x^3}+...
\end{equation}
\noindent To lowest order (using only the $1/x$ term in the
expansion) gives
\begin{equation} A_n={3k\over2\pi x}\int j^{0n}d^3y
\end{equation}
where
\begin{equation} j^{0n}={\eta\mu\over2}\int
d^2\zeta\delta^4(x-x(\zeta))x_{,a}^0x_{,b}^n\epsilon^{ab}
\end{equation} so
\begin{equation} \int
{2\over\eta\mu}j^{0n} d^3y=\int_0^Ldx^n=0
\end{equation}
for closed strings. Now, using the
next term in the expansion we have
\begin{equation} A_n={\bm {x}\over4\pi
x^3}\cdot\int\bm{y}N^nd^3y
.\end{equation}
\noindent In these formulas $\bm{y}$ is
the body (string) centered coordinate, and represents the spatial
part of $\delta x^\mu$ used above. Using (\ref{spinloop3}) the
solution be comes \begin{equation}
\bm{A}={3k\over8\pi}{\bm{S}\times\bm{x}\over x^3}
.\end{equation}
With this, the torsional field equations yield,
letting ${\bm r}$ replace ${\bm x}$,
and considering the case that the spin points in the $z$ direction,
\begin{equation}\label{torsol} {\bm
b} = {3k\over8\pi}{S\over r^{3}} \left(2 \cos(\theta) \bm {\hat r} +
\sin(\theta)\bm{\hat \theta} \right)
.\end{equation}}
The physical significance
of this result is that the intrinsic spin, which results from
the structure of the string, gives rise to a dipole field---the torsion
field.
\section{Conclusions}
A string not only acts as the source of a gravitational field,
it also becomes the source of torsion. In fact, when the string
worldsheet area is coupled to the torsion potential, the physical
property of the string that gives to torsion is intrinsic spin.
The work presented here focused on closed strings only. The
intrinsic spin of the string does not arise from motions of the
string, but is due to the structure and spatial extent of the string.
The Bianchi identity may be used to find the equation of motion
of the string, and we found that the motion is geodesic
only in lowest order.
\section{Appendix}
Above, the volume integral of the torsion source
tensor was taken to vanish. This was done in (\ref{int0}).
This can be shown to be true for two situations. One corresponds
to the case that the string is a static rigid circle.
It is noted that this assumption cannot hold for large cosmic
strings under the influence of their own gravity.
This result comes from the equation of motion of string, which is obtained
by taking variations with respect to $x^\mu$. However, this `geodesic'
postulate is not adopted here, the equation of motion of the string
is obtained from the Bianchi identities, so this restriction does
not hold. Allowing oscillations of the string nevertheless,
we may then show that the time average of the integral will vanish.
Assuming the period is very small, this is just as good as the rigid
string assumption.
First, without any assumptions, one may see that,
putting back the 3-volume element
\begin{equation}
\int\tilde j^{\si0}d^3x=0
.\end{equation}
From the definition $j^{\mu\nu}\equiv (1/2)T^{[\mu\nu]}$ and
(\ref{emtst}) we have
\begin{equation}
j^{\sigma\nu}={\mu\eta\over2\sqrt{-g}}\int
d^2\zeta\sqrt{-\gamma}\delta(x-x(\zeta))x^\sigma_{,a}x^\nu_{,b}\epsilon^{ab}
.\end{equation}
Integrating this over a volume one has,
\begin{equation}
\int\tilde j^{\si0}d^3x={\eta\mu\over2}\int_0^Ld\zeta{dx^\sigma\over d\zeta}=0
\end{equation}
where again this holds for closed strings.
Now it is shown that $\int\tilde j^{mn}d^3x=0$ for a rigid loop, or
on average. For a rigid loop this becomes
\begin{equation}
\int\tilde j^{mn}d^3x=
-{\eta\mu\over2}\int(y^m+\delta x^m)d\left({dx^n\over dx^0}\right)
\end{equation}
after integration by parts. The first term vanishes since $y^m$
goes to the center of mass and if $d(\delta x^n)/dx^0=0$,
as it would for a rigid string, then the second term vanishes too,
so that we have shown
\begin{equation}
\int \tilde j^{\mu\nu}=0
.\end{equation}
Alternatively, we may allow the string to undergo periodic
oscillations in which case it is to be shown that
\begin{equation}
<\int j^{\mu\nu}d^3x>\equiv\frac1 T\int dx^0\int j^{\mu\nu}d^3x=0
.\end{equation}
In this case we have
\begin{equation}
<\int j^{\mu\nu}d^3x>=
{\eta\mu\over T}\int dx^0d\zeta\left({dx^\nu\over dx^0}{dx^\mu\over d\zeta}
-{dx^\mu\over dx^0}{dx^\nu\over d\zeta}\right)=0
.\end{equation}
Finally, one may see that the average over one period
of the symmetric part of the energy momentum tensor does not
vanish, and essentially gives back the energy momentum tensor,
as we would expect.
Now we show the same kind of thing for the spatial part of the
worldsheet metric. First, if we assume that the string is a rigid
circular loop and, defining the $\phi$ as an angular measure as
$\delta x^\mu$ traverses the loop, we get $\gamma_{11}
=-(\cos^2\phi+\sin^2\phi)=-1$. On the other hand, if again we assume
that there are periodic oscillations it is shown that $<\gamma_{11}>$
is approximately equal to -1. To see this, note that
\begin{equation}
<\gamma^{11}>=\frac1 T\int dx^0{dx^m\over d\zeta}{dx^n\over d\zeta}g_{mn}
.\end{equation}
Now use $x^m=y^m+\delta x^m$ and assume further that $\delta x^m$
may be written as $\delta x^m=a^m+\epsilon^m(t)$ where $a^m$ is
time independent and $\epsilon^m$ is periodic in $T$. Then,
\begin{equation}
<\gamma_{11}>\approx \frac1 T\int dx^0{d\over d\zeta}(a^m+\epsilon^m)
{d\over d\zeta}(a^n+\epsilon^n)\eta_{mn}
.\end{equation}
Now, retaining terms to order $\epsilon^m$ and using the periodicity
of $\epsilon^m$ we have
\begin{equation}
<\gamma_{11}>={d a^m\over d\zeta}{d a^n\over d\zeta}\eta_{mn}
+{2\over T}{d a^m\over d\zeta}{d \over d\zeta}\int_0^Tdx^0\epsilon^n\eta_{mn}
.\end{equation}
The second terms is zero due to the periodicity and the first
terms replicates the result for the rigid loop, so we have
\begin{equation}
<\gamma^{11}>=-1
.\end{equation}
|
\section{Introduction}
The measurement of the different observables related to the so-called
four-fermion processes constitutes an important part of LEP2 physics.
The most important group of observables is related to $W$-pair production
and decay. It is not only the total cross section we are interested in,
but also the wealth of different spin and angular asymmetries. This leads
to rather sophisticated observables depending on multidimensional distributions
that are also often strongly affected by experimental cut-offs.
There are {\it nine} distinct $W$ decay channels and, owing to small statistics
of $W$-pair production, one is often bound to combine data of rather different
experimental signatures to obtain statistically meaningful results.
The situation of the forthcoming $ZZ$ physics or so-called single-$W$ physics
is rather similar. In all cases, theoretical predictions, to be comparable
with the data, have to be provided in the form of
four-fermion Monte Carlo simulations.
Monte Carlo integration over four-body
phase space%
\footnote{
In reality the case is even more complicated, because of
additional bremsstrahlung photons. In this paper, however,
we restrict ourselves
to pure Born-level phase space and {\it do not} discuss any
issues related to photonic corrections. Let us note, only, that in the general
case the Born-level phase space can form well-defined building-blocks of the
whole algorithm.
}
is a non-trivial problem. Why is it so?
First of all, there is over a hundred different four-fermion
final states allowed within the Standard Model. They differ by the flavours
of final-state quarks and/or leptons. Each of these states, even at the Born
level, is described by many, often over a hundred, Feynman graphs.
For distinct four-fermion final states one finds thus diametrically different
structures of singularities
depending on the set of Feynman graphs describing the particular
process. The singularities, or more precisely sharp peaks, have to be
carefully tackled by the generators to assure their efficiency. They can
affect the distribution in one or more variables of the particular
parametrization of the phase space.
As an example we can think of the $t$-channel singularities $1/t$,
$s$-channel ones such as resonances, etc.
These singularities occur in various parts of the phase space. Also, if they
are present in more than one variable
they can affect one another, either destructively or by
increasing the degree of singularity of the second one. Finally one particular
four-fermion process can have many distinct singular sub-processes, which need
to be treated individually.
Of course, there are also some similarities and symmetries amongst
processes as well as individual Feynman graphs. If properly used, they
can somewhat reduce the size of the problem, but unfortunately its complexity
remains intact. At the same time, if one departs from the Standard
Model and introduces any new particles or interaction types, the
size and complexity of the problem grows.
At the top of the above list of difficulties, there is also a strong cut-off
dependence of the cross-section. This, in the case
of electrons close to the beam directions,
lead either to instability of the Monte Carlo integration
or makes the generation ineffective since
many events are generated outside the detector acceptance.
In the present paper we show a working solution to the above phase-space
integration problem. It is based on the
{\it multichannel} approach that is
used in the {\tt KORALW} Monte Carlo code \cite{koralw:1995a,koralw:1998}.
Our aim was to enable the generation of all four-fermion
$e^+e^-\to f_1f_2\bar f_3\bar f_4$ final states
of the Standard Model over the full phase space or its sub-space,
within reasonable CPU time.
At the time of the LEP2 Workshop \cite{yr:lep2},
several dedicated four-fermion Monte Carlo codes were presented.
As far as the integration method is
concerned, some general strategies can be identified amongst these
codes:
the adaptive
algorithms based on {\tt VEGAS}-type routines
\cite{alpha,wphact,wwf,wwgenpv}, the {\tt BASES-SPRING}
algorithm \cite{grc4f,comphep,wwf}, the importance-sampling technique
\cite{wwgenpv,lpww02,wopper}
and the {\it multichannel} importance-sampling algorithms.
In the latter group, one finds two approaches: of the {\tt EXCALIBUR} code
\cite{berends:1994,excalibur:1995}, followed by {\tt ERATO}
\cite{erato:1997}, and of the {\tt KORALW} \cite{koralw:1995a,koralw:1998},
the subject of this paper.
The major advantage of the {\it multichannel} approach is, in our opinion,
that it gives almost complete control over the presampling process.
The branches correspond to physical configurations in a transparent
way. At the same time, since one can implement many branches,
the structure of the generator can be very rich.
No complicated dividing of the multidimensional phase space is
necessary and no {\it potentially uncontrolled} automated optimization
in the core of
the algorithm needs to be performed (although some {\em auxiliary}
optimization at the top levels can be useful).
The {\it multichannel} approach to few-body phase-space integration
has been used already in the past. For instance,
the {\tt TAUOLA} \cite{tauola:1990,tauola:1992,tauola:1993}
package for $\tau$ decays works in such a manner up to five-body decays. The
{\tt FERMISV} \cite{kleiss:1993}
code generates the NC-type four-fermion processes in that way also.
As it happens with the Monte Carlo algorithms, it may be difficult
to point to the exact source of the {\it multichannel} concept. Traces
of it can be found in a number of
other papers as well. Let us mention
just a few of these: \cite{fowl,mustraal,jadach:1985,topki,zodec}.
It is instructive to note that
there are two ``layers'' in the structure of {\it multichannel}
algorithms. The
core of the algorithm is determined by the internal structure of the
{\it channels} that contain the process-specific information on the
actual generation
algorithm. However, on the top of separate channels there is another,
external layer,
responsible for the ``branching'' process itself.
Recently \cite{jadach:MC-www} this second step has
been explained in detail.
The main advantage of the algorithm presented here is, in our opinion, the
following. Out of various, already known
as well as new, ideas, we show how to create in a systematic way,
with simple building blocks, a
universal and extendable presampler for four-body final states.
Despite its simplicity, this presampler is shown to work effectively
in an actual Monte Carlo code {\tt KORALW}. The important advantage of
the algorithm proposed here is its
modularity and simplicity: all the branches are built out of two
rather simple elementary pieces.
In this paper we present a complete description of the multichannel
four-fermion Monte Carlo algorithm of the {\tt KORALW}
code.
We start with the detailed description of the ``external layer'' --
that is the branching process -- which is quite universal.
Using notation of ref.\ \cite{jadach:MC-www}, we show
how it works in the existing algorithm of {\tt KORALW}. Starting from
section\ 3, we discuss the ``inner layer'' -- the actual
structure of individual channels dedicated to generating the four-fermion
phase space.
We end the paper with a short summary.
\section{``Multichannel'' Master Formula}
\label{MASTER}
As a basis of the generation of the four-fermion phase space we take
a multibranch type of Monte Carlo algorithm.
In this section, let us summarize the details of the ``external layer''
of four-fermion phase space.
Our goal is to calculate the following integral
\begin{eqnarray}
\label{SIGMA}
\sigma &=& \int d\Phi(P,\underline q) \vert M(P,\underline q) \vert^2
\\
d\Phi(P,\underline q)
&=&\frac{1}{2P^2} (2\pi)^4\delta^4\left(P-\sum_{k=1}^r q_k\right)
\frac{1}{2^{r-1}}
\prod_{k=1}^r \frac{1}{(2\pi)^3}\frac{d^3\vec{q_k}}{2E_k},
\;\; r=4,
\label{PHASE_SP}
\end{eqnarray}
where $P$ denotes the sum of beam momenta, $q_k$ denote the four-momenta of
final particles, and $\underline q$ denotes collectively all $q_k$
four-momenta.
Note that eq.\ \reff{PHASE_SP}
includes all the normalization
factors, including the flux factor $1/2P^2$.
Our strategy in solving eq.\ \reff{SIGMA} is to simplify $\vert M\vert^2$
until it reaches the analytically calculable level. Then, we generate
this simple distribution and reweight it to an exact $\vert M\vert^2$.
Let us then replace $\vert M\vert^2$ with the crude distribution
$\tilde\psi_{CR}$, in which we introduce the
structure of branches
\begin{equation}
d\Phi(P,\underline q) \tilde \psi_{CR}
= d\Phi(P,\underline q) \sum_i^{N_{BR}} \tilde p_i
\frac{\tilde \psi_{CR}^i(P,\underline q)}
{{\cal J}^i(P,\underline q)}.
\label{TREE}
\end{equation}
The summation goes over branches, with $N_{BR}$ denoting the total number of
branches.
In this equation we introduced two set of functions: $\tilde \psi_{CR}^i$
and ${\cal J}^i$. The first
represents the actual (crude) distributions to be generated over the phase space.
It is in these functions that all the physical information is
located. Their specific form is the main subject of this paper and will
be analysed in detail in the next sections.
Here we confine ourselves to the formal structure of
the arrangement of branches.
${\cal J}^i$ is composed of
the Jacobians of the change
of variables from four-momenta to angular and invariant-mass variables
(as defined in formulae below), which we anticipate in
each channel $i$ and of which we want to get rid (and reintroduce later by
appropriate reweighting). They are isolated in eq.\ \reff{TREE} for
simplicity reasons.
Finally, $\tilde p_i$ are some positive coefficients to be specified later.
We change variables $(P,\underline q)$ into
angles and masses
$(\cos\underline \theta^i,\underline \phi^i,\underline s^i)$
defined for each branch in different Lorentz frames. By
$\cos\underline \theta^i,\underline \phi^i,\underline s^i$
we denote collectively all the angles and masses of a given branch $i$.
The crude distribution becomes
\begin{equation}
\sum_i^{N_{BR}}
d\Phi^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)
\tilde p_i
\tilde \psi_{CR}^i(\cos\underline \theta^i,\underline \phi^i,
\underline s^i),
\label{TREE-BRANCH}
\end{equation}
where we introduced the explicit form of the Jacobian
${\cal J}^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)$
in new variables
\begin{eqnarray}
{{\cal J}^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)}
&=& \frac{d\Phi(P,\underline q)}
{d\Phi^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)}
= \prod_{j=1}^3 \lambda_j^i,
\label{TREE-JACOBIAN}
\\
\lambda_j^i=\lambda(1,s_{2_j}^i/s_{1_j}^i,s_{3_j}^i/s_{1_j}^i) &=&
\frac{1}{s_{1_j}^i}
\sqrt{ \left( s_{1_j}^i -s_{2_j}^i -s_{3_j}^i \right)^2
-4 s_{2_j}^i s_{3_j}^i},
\label{TREE-LAMBDA}
\\
d\Phi^i (\cos\underline \theta^i,\underline \phi^i,\underline s^i)
&=& \frac{1}{4P^2} \frac{1}{(4\pi)^{8}}
ds_1^ids_2^i \prod_{j=1}^{3} d\cos\theta_j^i d\phi_j^i.
\end{eqnarray}
The mass-like variables $s_1^i$ and $s_2^i$ present in the integration element
denote squares of the invariant masses of systems consisting of two or three
outgoing fermions. They are present in the definition of the $\lambda$-factors
of the $\cal J$ as well, but in this case $s^i_{j_k}$
can denote squares of masses of the outgoing fermions as well.
We have found that
it is convenient to enlarge integration domains of the branches.
This can be done for each branch in a different way. We do it by
introducing $\Theta_i$ functions
\begin{equation}
\label{THETA}
\tilde \psi_{CR}^i(\cos\underline \theta^i,\underline \phi^i,
\underline s^i) =
\psi_{CR}^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)
\Theta_i(\Omega^i)
\end{equation}
and extending the integration areas
$\Omega^i \to \Omega^i_{CR}$.
The $\Theta_i$ enforce the exact phase-space limits. Now, by setting
$\Theta_i\to 1$, we arrive at the final form of the crude distribution
\begin{equation}
\sum_i^{N_{BR}}
d\Phi^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)
\vert_{\Omega^i_{CR}}
\tilde p_i \psi_{CR}^i(\cos\underline \theta^i,\underline \phi^i,
\underline s^i).
\label{TREE-CRUDE}
\end{equation}
We assume that eq.\ \reff{TREE-CRUDE}
is integrable analytically, with the result
\begin{equation}
\int\limits_{\Omega^i_{CR}} \tilde p_i
d\Phi^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)
\psi_{CR}^i(\cos\underline \theta^i,\underline \phi^i,
\underline s^i)
= \tilde p_i{\cal N}_{CR}^i.
\label{TREE-NORM}
\end{equation}
The random choice of the branch to be used in the generation of a particular
point in the phase space is performed with the help of probabilities $P_i$:
\begin{equation}
\label{PI}
P_i=\frac{\tilde p_i{\cal N}_{CR}^i}
{\sum\limits_k^{N_{BR}} \tilde p_k {\cal N}_{CR}^k}.
\end{equation}
Later, following ref.\ \cite{jadach:MC-www}, we generate points
according to distributions $\psi_{CR}^i$ of eq.\ \reff{TREE-CRUDE}.
The last step to be done is to reintroduce all the simplifications in
the form of the appropriate weights to complete the prescription on how to
calculate the integral \reff{SIGMA}.
For each channel $i$, the transition from $\psi_{CR}^i$ of eq.\
\reff{TREE-CRUDE} to $\tilde \psi_{CR}^i$ of eq.\ \reff{TREE-BRANCH}
can be achieved with the help of a simple step-like weight:
\begin{equation}
w^a(i)=\frac{\tilde\psi_{CR}^i}{\psi_{CR}^i}=\Theta_i.
\label{WA}
\end{equation}
The whole distribution of eq.\ \reff{TREE-BRANCH} is then generated from
$\psi_{CR}^i$ with the help of the same weight $w^a(i)$, the argument $i$
being the number of the chosen branch for a particular event. The
distribution generated this way
\begin{equation}
\sum\limits_i^{N_{BR}} P_i \frac{\psi_{CR}^i}
{\tilde p_i {\cal N}_{CR}^i} w^a(i)
= \frac{\sum\limits_i^{N_{BR}}\tilde p_i\tilde\psi_{CR}^i}
{\sum\limits_i^{N_{BR}}\tilde p_i {\cal N}_{CR}^i}
\end{equation}
is, up to a normalization factor, equal to that of eq.\
\reff{TREE-BRANCH}.
The normalization (integral) of eq.\ \reff{TREE-BRANCH} can
be calculated as
\begin{eqnarray}
\tilde{\cal N}_{CR}
&=& \sum\limits_i^{N_{BR}}\tilde p_i \tilde {\cal N}^i_{CR}
= \sum\limits_i^{N_{BR}}\tilde p_i {\cal N}^i_{CR}
\left<w^a(i)\right>_i
\\
&=& \left(\sum\limits_k^{N_{BR}}\tilde p_k {\cal N}^k_{CR} \right)
\left(\sum\limits_i^{N_{BR}} P_i \left<w^a(i)\right>_i \right)
= \; \Bigl<_{P_i}\left<w^a(i)\right>_i \Bigr>
\sum\limits_k^{N_{BR}}\tilde p_k {\cal N}^k_{CR},
\end{eqnarray}
where we have made use of the fact that $P_i$ defines a ratio of
the number of events generated
in channel $i$ to the total number of events. With $\left<x \right>_i$ we
denote the
average of $x$ within the $i$-th channel and with
$\left<_{P_i} \; ... \; \right>$ we denote consecutive
average with respect to the branches:
\begin{equation}
\Bigl<_{P_i} \left<w(i,\dots)\right>_i \Bigr>=
\sum\limits_i^{N_{BR}} \frac{N_i}{N} \left<w(i,\dots)\right>_i
\; ; \;
\lim_{N \to \infty }
\Bigl<_{P_i} \left<w(i,\dots)\right>_i \Bigr>=
\sum\limits_i^{N_{BR}} P_i \left<w(i,\dots)\right>_i.
\label{szesnascie}
\end{equation}
In practice, the external summation over channels is realized
with the help of the Monte Carlo generation
of the channel number with the probabilities $P_i$
and, for every event, only that individual weight of the particular
channel is calculated and used.
Furthermore, we do not need to
calculate $\Bigl<_{P_i} \left<w(i,\dots)\right>_i \Bigr>$
as an explicit sum of averages $\left<w(i,\dots)\right>_i$, calculated
and stored
for each channel separately, as in eq.\ \reff{szesnascie}.
Instead we calculate only {\it one} average over the whole sample
of generated events of all channels:
\begin{equation}
\Bigl<_{P_i} \left<w(i,\dots)\right>_i \Bigr>
= \sum\limits_i^{N_{BR}} \frac{N_i}{N} \frac{1}{N_i}
\sum\limits_{events~in~i}^{N_{i}} w(i,\dots)
= \frac{1}{N} \sum\limits_{all~events}^{N} w(i,\dots)
\equiv \left< w(i,\dots)\right>.
\end{equation}
Here the argument $i$ denotes that for the event generated
with channel $i$ we calculate a weight $w(i,\dots)$.
\footnote{
Let us also note that in general, in our paper,
we will omit the symbol $\lim_{N\to \infty}$.}
Including the fundamental matrix-element weight
\begin{equation}
w^b(P,\underline q)
=\frac{\vert M(P,\underline q)\vert^2}{\tilde\psi_{CR}(P,\underline q)}
\end{equation}
and repeating the above steps
we end up with the following formula for the integral \reff{SIGMA}
\begin{eqnarray}
\label{SIGMA2}
\int d\Phi(P,\underline q) \vert M(P,\underline q) \vert^2
&=&
\left< \tilde w(k,P,\underline q) \right>
\sum\limits_i^{N_{BR}}\tilde p_i {\cal N}^i_{CR},
\\
\tilde w(k,P,\underline q) &=& w^a(k) w^b(P,\underline q)
=
\frac{\vert M(P,\underline q)\vert^2}{\tilde\psi_{CR}(P,\underline q)}
\Theta_k(\Omega_k).
\end{eqnarray}
Let us concentrate for a moment on the coefficients $\tilde p_i$.
If we rewrite them in
a more specific form as $\tilde p_i =p_i/{\cal N}^i_{CR}$, with the
condition $\sum p_i =1$, we find from eq.\ \reff{PI} that $P_i = p_i$, so
that the
$p_i$ have a nice interpretation as the actual branching probabilities.
The master equation \reff{SIGMA2} then evolves into
\begin{eqnarray}
\label{SIGMA3}
\int d\Phi(P,\underline q) \vert M(P,\underline q) \vert^2
&=&
\left<w(k,P,\underline q)\right>,
\\
w(k,P,\underline q) &=&
\vert M(P,\underline q)\vert^2
\left[ \sum\limits_i^{N_{BR}}
\frac{p_i \tilde \psi_{CR}^i(P,\underline q)}
{{\cal N}^i_{CR}{\cal J}^i(P,\underline q)}
\right]^{-1}
\Theta_k(\Omega_k).
\end{eqnarray}
Much as in ref.\ \cite{jadach:MC-www}, we will now discuss
a possible variation of the above algorithm. Turning back to
eq.\ \reff{TREE},
we modify it by removing the Jacobians ${\cal J}_i$:
\begin{equation}
d\Phi(P,\underline q) \tilde \psi_{CR}
= d\Phi(P,\underline q) \sum_i^{N_{BR}} \tilde p_i
\tilde \psi_{CR}^i(P,\underline q)
\label{TREE-BIS}
\end{equation}
and instead introduce them in eq.\ \reff{TREE-BRANCH}
\begin{equation}
\sum_i^{N_{BR}}
d\Phi^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)
\tilde p_i
{\cal J}^i(\cos\underline \theta^i,\underline \phi^i,\underline s^i)
\tilde \psi_{CR}^i(\cos\underline \theta^i,\underline \phi^i,
\underline s^i).
\label{TREE-BRANCH-BIS}
\end{equation}
On the course to eq.\ \reff{TREE-CRUDE} we simplify
${\cal J}^i_{CR} \to 1$ and, as a result, weights $w^a(i)$ get modified:
\begin{equation}
w^a(i)=\frac{{\cal J}^i\tilde\psi_{CR}^i}{\psi_{CR}^i}
={\cal J}^i\Theta_i.
\label{WA-BIS}
\end{equation}
Accordingly, eq.\ \reff{SIGMA3} becomes
\begin{eqnarray}
\label{SIGMA3-BIS}
\int d\Phi(P,\underline q) \vert M(P,\underline q) \vert^2 &=&
\left<w(k,P,\underline q)\right>,
\\
w(k,P,\underline q) &=&
\vert M(P,\underline q)\vert^2
\left[ \sum\limits_i^{N_{BR}}
\frac{p_i \tilde \psi_{CR}^i(P,\underline q)}
{{\cal N}^i_{CR}}
\right]^{-1}
{\cal J}^k(P,\underline q)\Theta_k(\Omega_k).
\end{eqnarray}
Note that here the Jacobians $\cal J$ are {\em not} summed over branches but
calculated for the branch of a given event only. As the actual form of
the Jacobians is rather simple, we should not expect a significant
difference in the performance of the two algorithms.
Finally, let us remark that {\em both} of our algorithms are in fact of
the ``second'' type in the classification of
ref. \cite{jadach:MC-www}.
In the first one (eq.\ \reff{SIGMA3}), we attribute zero weight to some
events already inside branches in the form of trivial step
functions $\Theta_i$. In the second algorithm,
weights $\Theta_i {\cal J}^i$ are continuous functions between zero and one.
In the language of measure theory the difference of the two algorithms
(defined respectively
by formulae \reff{SIGMA3}
and \reff{SIGMA3-BIS} ) lies in the choice of the basic measure on the
phase-space
manifold. In the second, more natural case, the Lorentz-invariant phase-space
element takes this role.
\section{The Branches}
\label{BRANCHES}
In the previous section we
gave the complete general description of the
``external layer'' of the multichannel algorithm responsible for the
alignment of separate branches. In this section we will present the
construction of the actual branches in the case of four-fermion final
states that we have developed for the {\tt KORALW} code \cite{koralw:1998}.
Let us begin with remark.
Each branch $\psi^i_{CR}$ of eqs.\
\reff{TREE-BRANCH}, \reff{TREE-CRUDE}, \reff{TREE-BRANCH-BIS}
is intended to describe a certain type of
singularities that one encounters in the Feynman graphs,
but there is no unique definition of
the $\psi^i_{CR}$ functions. They are dummy functions that cancel out in
the final result. On the one hand, their role is to mimic as close as
possible the
complicated structure of singularities of the true matrix element, but
on the other hand they should not be too complicated. First of
all they {\it must}
be analytically integrable over the generation area in order to be able
to normalize the generator.
Secondly, because of the complexity of the problem, it is, in our
opinion, very useful for them to
have a modular structure. This allows us
to write the program and its documentation in a compact and
transparent way, leaving room for easier modifications in the
future.
Finally, these simplifications obviously cannot go too far
as the $\psi^i_{CR}$ functions have to be quite universal and flexible
to accommodate a variety of matrix-element singularities.
How do we fulfil in practice all these, partly contradictory, requirements?
How do we construct the actual $\psi^i_{CR}$ functions?
The crucial assumption that we make
is the partial
factorization of $\psi^i_{CR}$ with respect to the angular and mass
variables (we skip the branch index $i$ for the next two sub-sections):
\begin{equation}
\psi_{CR}(\cos\underline \theta,\underline \phi,\underline s)
=f(\underline s)g(\cos\underline \theta,\underline \phi,\underline s).
\label{FACTORISATION}
\end{equation}
We request that
$g(\cos\underline \theta,\underline \phi,\underline s)$, upon integrating
over angular variables, becomes independent of masses $\underline s$
\begin{equation}
\int d\cos\underline \theta d\underline \phi
g(\cos\underline \theta,\underline \phi,\underline s)
= \hbox{const}.
\end{equation}
Finally, we assume that masses $\underline s$ will always be
generated first, before the angles $\underline \theta,\underline
\phi$.
With all these simplifications, one may worry that we are too
restrictive. For example, it may happen that, for certain configurations
of singularities, it would be natural to reverse the order of generation --
angles first and then masses, or even, further, that one should use some
combination of these variables as more ``physical''.
This approach would, however, betray to some extent the modularity of the
program: it is always up to the author of an algorithm to decide where
to draw the line.
Our priority was set on simplicity and modularity of the algorithm.
We believe we achieved those to a large extent.
The solution presented here may look
trivial and simplistic, but the point is
that the efficiency of the generation
is sufficient for our purposes, and we do not need to pay
the price of formulas several pages long, which may be necessary
e.g. in the
algorithm of refs.\
\cite{kleiss:1993,excalibur:1995}.
Now we can proceed to the details
of the mass and angular distributions $f$ and
$g$.
\subsection{Mass Distributions}
As explained above we intend to
generate $s_1, s_2$ variables first, and then the angles.
Generically there are two possible ways of aligning the two mass
variables: a ``chain''-like and a ``fork''-like, see Fig.\ 1.
\noindent
\unitlength=1mm
\begin{picture}(140,70)(0,0)
\put(10,40){\line(1,0){60}}
\put(25,40){\line(1,-1){10}}
\put(40,40){\line(1,-1){10}}
\put(55,40){\line(1,-1){10}}
\put(10,42){$s_{max}$}
\put(30,42){$s_{1}$}
\put(45,42){$s_{2}$}
\put(65,42){$m_{4}$}
\put(65,26){$m_{3}$}
\put(50,26){$m_{2}$}
\put(35,26){$m_{1}$}
\put(10,5){Fig.\ 1a: ``Chain''}
\put(85,40){\line(1,0){15}}
\put(100,40){\line(1,1){25}}
\put(100,40){\line(1,-1){25}}
\put(115,55){\line(1,-1){10}}
\put(115,25){\line(1,1){10}}
\put(85,42){$s_{max}$}
\put(100,47){$s_{1}$}
\put(100,32){$s_{2}$}
\put(127,13){$m_{4}$}
\put(127,35){$m_{3}$}
\put(127,43){$m_{2}$}
\put(127,65){$m_{1}$}
\put(100,5){Fig.\ 1b: ``Fork''}
\end{picture}
For each of these configurations we need to give the exact phase-space
limits $\Theta(\Omega)$ corresponding to eq.\ \reff{THETA}.
Then we define an extension
$\Omega \to \Omega_{CR}$
of these limits that enable the analytic integration of the
distributions to simple results, allowing
numerically fast normalization to unity.
For the two cases defined above we continue independently:
\noindent
\unitlength=1mm
\begin{picture}(150,80)(0,0)
\label{STRIPES}
\put(0,0){
\begin{picture}(75,80)(-10,-10)
\put(10,10){\framebox(40,40)}
\put(10,10){\line(1,1){40}}
\put(35,0) {\large $s_{1}$}
\put(-5,35){\large $s_{2}$}
\put(47,6){$s_{max}$}
\put(1,50){$s_{max}$}
\put(6,6){$0$}
\put(-5,-10){Fig.\ 2a: ``Chain''-type phase space}
\end{picture}
\put(80,0){
\begin{picture}(75,80)(-10,-10)
\put(10,10){\framebox(40,40)}
\qbezier(50,10)(10,10)(10,50)
\put(10,20){\line(1,0){1}}
\put(20,10){\line(0,1){1}}
\put(20,20){\line(1,0){30}}
\put(20,20){\line(0,1){30}}
\put(35,0) {\large $s_{1}$}
\put(-5,35){\large $s_{2}$}
\put(47,6){$s_{max}$}
\put(1,50){$s_{max}$}
\put(17,6){$\frac{s_{max}}{4}$}
\put(1,20){$\frac{s_{max}}{4}$}
\put(6,6){$0$}
\put(-5,-10){Fig.\ 2b: ``Fork''-type phase space}
\end{picture}
\end{picture}
\begin{itemize}
\item
The ``chain''.\\
The exact phase-space limits $\Theta(\Omega^C)$ of eq.\ \reff{THETA}
in this case are given by:
\begin{equation}
\sqrt{s_{max}} \geq \sqrt{s_1}+m_1, \;\;
\sqrt{s_1} \geq \sqrt{s_2}+m_2,\;\;
\sqrt{s_2} \geq m_3+m_4.
\end{equation}
By dropping masses, the above domain $\Omega^C$ extends
to the simpler one, $\Omega^C_{CR}$ of\ Fig
2a:
\begin{equation}
s_{max} \geq {s_1} \geq {s_2} \geq 0.
\end{equation}
Next, we assume that $s_1$ and $s_2$ distributions are identical and that
the function $f(s_1,s_2)$ factorizes
\begin{equation}
f(s_1,s_2)= \frac{1}{{\cal F}(s_{max})}f_1(s_1) f_1(s_2).
\end{equation}
This is a somewhat restrictive assumption, but as there will be practically
no restrictions on the form of a single distribution $f_1(s_1)$,
we can always use a sufficiently general form of it.
It costs some efficiency, but this is
compensated by the (almost) perfect representation of the phase-space limits
already at a crude level and by the speed of the code.
With our assumptions, the integral
${\cal F}$ can be easily calculated:
\begin{eqnarray}
{\cal F}(s_{max}) &=& \int_{\Omega^C_{CR}} f(s_1,s_2)ds_1 ds_2 =
\int_0^{s_{max}} ds_1 f_1(s_1)
\int_0^{s_1} ds_2 f_1(s_2)
\\
&=& \frac{1}{2}\left[ F_1(s_{max})- F_1(0)\right]^2,
\\
F_1(x) &=& \int^x f_1(s)ds
\end{eqnarray}
and the distribution can be normalized, assuming one knows the $F_1$
function.
\item
The ``fork''.\\
Also in this configuration we assume factorization of the function
$f(s_1,s_2)$ into the one-dimensional functions $f_1(s_1)$ and $f_2(s_2)$:
\begin{equation}
f(s_1,s_2)= \frac{1}{{\cal F}(s_{max})}f_1(s_1) f_2(s_2).
\end{equation}
The phase-space limits
($\Theta(\Omega^F)$ of eq.\ \reff{THETA}) are the following:
\begin{equation}
\sqrt{s_1} +\sqrt{s_2} \leq \sqrt{s_{max}},\;\;
\sqrt{s_1} \geq m_1 +m_2,\;\;
\sqrt{s_2} \geq m_3 +m_4.
\end{equation}
This area
can be extended to the polygonal $\Omega^F_{CR}$ domain, or explicitly
to the surface defined by the difference between the two squares, as shown in
Fig.\ 2b.
The integration is easy:
\begin{eqnarray}
{\cal F}(s_{max}) &=&
\int_{\Omega^F_{CR}} f_1(s_1)f_2(s_2)ds_1 ds_2
\\
&=&
\left[\int_0^{s_{max}} -\int_{s_{max}/4}^{s_{max}}\right]
ds_1 ds_2 f_1(s_1) f_2(s_2)
\\
&=& \left[ F_1(s_{max})- F_1(0)\right]
\left[ F_2(s_{max})- F_2(0)\right]
\\
&&- \left[ F_1(s_{max})- F_1(\frac{s_{max}}{4})\right]
\left[ F_2(s_{max})-
F_2(\frac{s_{max}}{4})\right],
\;\;\;
\end{eqnarray}
\begin{equation}
F_j(x) = \int^x f_j(s)ds, \;\; j=1,2.
\end{equation}
\end{itemize}
Finally we make a choice of a suitable form of $f_j$ functions.
We use a multi-branch form of it, each branch $\alpha$ being
associated with a basic mass singularity type:
\begin{eqnarray}
f_j(s) &=&
\sum_\alpha^{N_{MAS}} a_{j\alpha}
\frac{ f_{j\alpha}(s)}{\int_0^{s_{max}} f_{j\alpha}(x) dx},\;\;
\sum_\alpha^{N_{MAS}} a_{j\alpha} =1,
\\
f_{j\alpha}(s) &=& \left\{ \begin {array}{l}
1,\\ \\
{(s+m_R^2)^{-n}},\\ \\
\left(s+m_R^2\right)^{-1}
{\log^n\left((s_{max}+m^2_R)/(s+m^2_R)\right)},\\ \\
{\left((s-M^2_k)^2 +M^2_k\Gamma^2_k\right)^{-1}},\\ \\
{\left(s_{max}+m_R^2-s\right)^{-1}}.
\end{array}
\right.
\label{FMASS}
\end{eqnarray}
The $N_{MAS}$ denotes the total number of ``mass branches''.
The parameter $m_R^2$ takes a role of a regulator, preventing distributions from
blowing up to infinity at the phase-space limit.
It should be stressed that $m_R^2$ does not play the role of a cut-off,
and that the complete phase space is covered by the generation.
The $M_k$ and $ \Gamma_k$ are parameters of the $k$-th resonance.
All the functions presented above are easily integrable into the elementary
functions.
\subsection{Angular Distributions}
In the next step we generate the angular variables: $\cos\theta_j$ and
$\phi_j$ (the index $j=1,2,3$ denotes the number of angle
within the same branch).
The role of the angular variables is to describe the class of $t$- and
$u$-channel-type
singularities of the Feynman graphs. There is a number of different
transfers that can be built, based on two beams and
four external final-state four-vectors.
In principle each of them can develop a singularity in some of the
Feynman graphs for some configurations
of final-state fermions.
The general angular function
$g(\cos\underline \theta,\underline \phi,\underline s)$ of
eq.\ \reff{FACTORISATION} is further
factorized with respect to the angles of each step of the
generation. Unlike the mass distributions case, the factorization is
not complete -- each subsequent angle uses previously generated angles
and masses (in the form of four-momenta). We write it symbolically
as:
\begin{equation}
g(\cos\underline \theta,\underline \phi,\underline s)=
g_1(\cos\theta_1,\phi_1;\underline s)
g_2(\cos\theta_2,\phi_2;\theta_1,\phi_1,\underline s)
g_3(\cos\theta_3,\phi_3;\theta_1,\phi_1,\theta_2,\phi_2,\underline s).
\end{equation}
Much as with the mass distributions, we use branching structure to
build individual $g_j$ functions (we always take
flat distributions in $\phi_j$ variables and thus their generation
decouples):
\begin{eqnarray}
g_j(\cos\theta_j,\phi_j;...,\underline s) &=&
\sum_{\alpha}^{N_{ANG}}
b_{j\alpha}
\frac{ g_{j\alpha}(\cos\theta_j,\underline s) }{
\int_{-1}^{1} d\cos\theta \int_{0}^{2\pi} d\phi
g_{j\alpha}(\cos\theta,\underline s)},\;\;
\sum_{\alpha}^{N_{ANG}} b_{j\alpha} =1,
\;\;
\end{eqnarray}
with $N_{ANG}$ denoting the number of ``angular branches''.
The basic angular distribution functions $g_{j\alpha}$ are:
\begin{eqnarray}
g_{j\alpha}(\cos\theta_j,\underline s) &=& \left\{ \begin {array}{l}
1,
\\ \\
{\left(-t_j+m_A^2\right)^{-n}},
\\ \\
\left(-t_j+m_A^2\right)^{-1}
{\log^n\left((-t_{j,max}+m_A^2)/(-t_j+m_A^2)\right)},
\\ \\
{\left(-u_j+m_A^2\right)^{-n}},
\\ \\
\left(-u_j+m_A^2\right)^{-1}
{\log^n\left((-u_{j,max}+m_A^2)/(-u_j+m_A^2)\right)},
\end{array}
\right.
\label{GANGLE}
\end{eqnarray}
where $t_j$ and $u_j$ are functions of $\cos\theta_j$ and
$\underline s$-variables. The
$\phi_j$ variables are independent and generated
with a flat distribution.
The $m_A^2$ is, as before, a small mass introduced to regularize the
distributions at singularity points.
\subsection{Event Construction}
Now, we can continue step by step with the generation and construction
of the explicit form of the event with the help
of the variables generated as explained in the previous sections.
Let us first define the relation between angles and $t_j$ and $u_j$
transfers.
For any two-to-two sub-process of our generation
(see Fig. 1), the transfers $t_j$ and $u_j$ are
defined as
\begin{equation}
t_j=(P_{1}-q_{1_j})^2,\;\;\;
u_j=(P_{1}-q_{2_j})^2,
\label{TUTRANSFERS}
\end{equation}
where $P_{1}$, $P_{2}$ denote four-momena of the first
and second beams, and $q_{1_j}$, $q_{2_j}$ denote the four-momenta of the
first and second outgoing objects. These outgoing objects
can either be the final-state particles or groups of
particles (``intermediate states''). Later on, we will use the masses of the
objects to label them. Thus $[m_1^2]$, $[s_1]$ will denote respectively the
first final-state fermion and the system consisting of second, third and fourth
particles (case of fig. 1a) or the
system consisting of first and second particles
(case of fig. 1b). In every case, the invariant mass
of the object is known, either from the input parameters or from the mass
generation in the previous step of the algorithm. In this
way one has an access to all possible transfers $t_j$ and
$u_j$.
Angles corresponding to the invariants are easiest to generate in the
local CMS frames of outgoing four-vectors $(q_{1_j},q_{2_j})$.
In such a frame the $t_j(\cos\theta_j,\underline s)$ and
$u_j(\cos\theta_j,\underline s)$ can
be expressed in a standard way
(we drop the index $j$ here):
\begin{eqnarray}
t &=& (P_1-q_1)^2 = q_1^2+P_1^2 -2q_1^0P_1^0
+2\vert\vec q_1\vert\vert\vec P_1\vert\cos\theta,
\\
u &=& (P_1-q_2)^2 = q_1^2+P_2^2 -2q_1^0P_2^0
-2\vert\vec q_1\vert\vert\vec P_1\vert\cos\theta,
\end{eqnarray}
with
\begin{eqnarray}
q_1^0 &=& \frac{1}{2\sqrt{s_{12}}} \left(s_{12}+q_1^2-q_2^2\right),
\\
P_1^0 &=& \frac{1}{2\sqrt{s_{12}}} \left(s_{12}+P_1^2-P_2^2\right),
\\
\vert\vec q_1\vert^2 &=& \frac{1}{4}s_{12}
\lambda\left(1,q_1^2/s_{12},q_2^2/s_{12}\right),
\\
\vert\vec P_1\vert^2 &=& \frac{1}{4}s_{12}
\lambda\left(1,P_1^2/s_{12},P_2^2/s_{12}\right),
\\
s_{12} &=& (P_1+P_2)^2.
\end{eqnarray}
Finally we need to define a series of \ frames (i.e. four-momenta
of group of particles) used
to construct subsequent angles. This can be, as a matter of fact,
decoded from figs.~1a and 1b.
For the configuration of fig.\ 1a we start by generating the angle
$\angle ([m_1^2],P_{1})$
in the rest frame of the two beams
$P_{1}$ and $P_{2}$.
Next, we subtract the generated final four-momentum
$[m_1^2]$ from the second beam $P_{2}$ and repeat the previous step for the
angle $\angle ([m_2^2],P_{1})$ in the rest frame of the pair $([m_2^2],[s_2])$.
Finally, in the third step, after subtracting again
$[m_2^2]$ from $P_{2}-[m_1^2]$, we build the angle
$\angle \left({[m_3^2]},{P_{1}}\right)$ in the rest frame of the pair
$\left({[m_3^2]},{[m_4^2]}\right)$.
In the case of fig.\ 1b, the first and second angles,
$\angle \left({[s_1]},{P_{1}}\right)$ in the rest frame of the pair
$ \left({[s_1]},{[s_2]}\right)$ and
$\angle \left({[m_1^2]},{P_{1}}\right)$ in the rest frame of the pair
$ \left({[m_1^2]},{[m_2^2]}\right)$,
are generated as
before. In the definition of the third angle,
$\angle \left({[m_3^2]},{P_{2}}\right)$ in the rest frame of the pair
$\angle \left({[m_3^2]},{[m_4^2]}\right)$,
the direction of the
second beam $P_{2}$ is chosen as the $z$-axis
instead of the first one.
\subsection{Normalization}
The last step is to check out the normalization
${\cal N}^i_{CR}$ of a given
$\psi^i_{CR}$ used in the definition of the branch $i$ of eq.\ \reff{TREE-NORM}.
We need to calculate the integral
$\int d\Phi^i \psi_{CR}^i$:
\begin{eqnarray}
\label{NORM}
{\cal N}^i_{CR} &=&
\int\limits_{\Omega^i_{CR}}
d\Phi^i(\cos\underline \theta^i,\underline \phi^i,\underline s)
\psi_{CR}^i (\cos\underline \theta^i,\underline \phi^i,\underline s)
\\
&& =
\int ds_1 ds_2 f(s_1,s_2)
\int \prod_{j=1}^3 d\cos\theta_j d\phi_j
g_j(\cos\theta_1,\phi_1;\dots,\theta_j,\phi_j,\underline s)
\nonumber
= 1.
\end{eqnarray}
The normalization to unity follows from the specific form of the $f$ and $g$
functions of our choice.
The property that normalization factors ${\cal N}^i_{CR}=1$, although
not necessary for the correctness of the algorithm, simplifies the
practical use.
Let us elaborate more on this point. Thanks to the normalization,
and if in formulae \reff{FMASS} and \reff{GANGLE} only single branches
are present, the product of functions $f$
and $g$ forms the Jacobian of the transformation from the
$\Omega^{i}_{CR}$ domain to the eight-dimensional unit cube.
The inverse transformation induces the measure on $\Omega^{i}_{CR}$, which is
the (crude) probability distribution. In the multichannel case,
when coefficients $a_{j_\alpha}$ and $b_{j_\alpha}$ can
all be non-zero, the measure (probability distribution) on $\Omega^{i}_{CR}$
is defined as the weighted%
\footnote{
More precisely, weighted with appropriate products
of coefficients/probabilities $a_{j_\alpha}$ and $b_{j_\alpha}$.
}
sum of measures defined by individual transformations.
\subsection{Optimization}
Finally, we have to discuss the choice of the coefficients
$p_i$, $a_{j\alpha}$ and $b_{j\alpha}$. It is these parameters that
assure flexibility of the algorithm.
The coefficients $p_i$, $a_{j\alpha}$ and $b_{j\alpha}$ are free
parameters of the generator. Their choice depends on the final state
chosen as well as on the external cuts used in the calculation of
cross-section.
An optimal choice of these parameters is a delicate and important part of
the algorithm. We did it ``manually'', by looking at the
physical structures they approximate. Our strategy was to eliminate
the most overweighted events by iterative adjustments of these
coefficients and, if needed, adding more branches to the master sum of eq.\
\reff{TREE-BRANCH}. One can easily imagine other, more sophisticated,
optimization procedures, for example by assuming functional
dependence of these coefficients on the type of the final fermions or by
allowing for functional dependence on external cut-offs. This latter
optimization may be especially interesting. In the case of
strongly peaked cross-sections, strong cut-offs can downgrade the efficiency
of the algorithms by causing excessive rejection, unless compensated by
retuning of these internal parameters.
Yet another option would be a numerical
optimization of some sort. One can mention in this context
the approach of Ref.\ \cite{kleiss:1994}, for example.
To summarize, such a large number of free parameters makes it more
complicated to find the optimal configuration of the coefficients, but
on the other hand it gives a direct access to virtually {\em
every} singular configuration and gives us a possibility for its direct
modelling.
We also have to remember that
$p_i$, $a_{j\alpha}$ and $b_{j\alpha}$ are dummy
parameters. It means that, by varying them, one gets a
strong tool for testing the correctness of the integration.
\section{The Algorithm}
At this point we are ready to put together all the pieces and write down
the complete formula for the overall weight and normalization for our
algorithm. To be able to calculate $\sigma$ of eq.\ \reff{SIGMA},
the following steps have to be completed:
\begin{enumerate}
\item
Generate the branch number $i$ according to probabilities $p_k$.
\item
Generate point
$(\cos\underline \theta^i,\underline \phi^i,\underline s)$
according to the $\psi^i_{CR}$
distribution of the chosen $i$-th branch.
\item
Construct the weight $w^a(i)$.
\item
If $w^a(i)\neq 0$ construct the phase-space point $(\underline q)$ out of
$(\cos\underline \theta^i,\underline \phi^i,\underline s)$.
\item
Calculate the total weight $w(i,P,\underline q)$
of eq.\ \reff{SIGMA3} (or \reff{SIGMA3-BIS}) in terms of
four-vectors.
\item
Calculate the cross-section as an average of the weight
$w(i,P,\underline q)$
\begin{equation}
\sigma = \left<w(i,P,\underline q)\right>.
\end{equation}
\end{enumerate}
A few comments are in order here.
While constructing a set of all branches,
one has to include all the necessary permutations
of external momenta of beams and outgoing fermions.
The generic two branches
--- ``chain'' and ``fork'', need to be applied to all configurations of
external momenta to describe all the singularities properly. All together,
this leads to over fifty branches in the case of four-fermion final states%
\footnote{
If one wanted to be even more precise, and counted separately
each ``basic'' branch of eqs.\ \reff{FMASS} and \reff{GANGLE}, the total
number of branches would easily exceed $10^6$!
}.
The algorithm as described in this note is the one with variable weights.
By means of the standard rejection technique
it can provide unweighted events as well.
\section{Summary}
In this paper we gave a complete description of the Monte Carlo
algorithm for the generation of the four-fermion phase space, as needed
for LEP2 applications. It is based
on the ``multichannel'' approach, as explained elsewhere e.g. in
\cite{jadach:MC-www}.
Strong emphasis is given to the construction of separate
branches with the help of rather simple modular building blocks.
This approach allowed for transparent writing of the code and testing,
as well as for easy extensions and modifications in the future.
The major disadvantage of the present version, or rather the important
place for future improvements,
is optimization of internal coefficients. In particular, in the
case of big changes of external cut-offs or centre-of-mass energy,
the coefficients may need retuning. At present this must be done
by hand.
The algorithm has been successfully implemented in the {\tt KORALW}
\cite{koralw:1998} Monte Carlo program. Generation can cover the full
phase space, or any sub-region defined by cut-offs\footnote{
In {\tt KORALW} some of the cut-offs can be optionally implemented
before lengthy calculation of the matrix element.
This is essential, e.g. in the generation
of $e^+e^- f \bar f$ final states with tagged electrons when the rejection
rate is very high.
}. This is the case
of all four-fermion final states at LEP2 centre-of-mass energies.
Numerically stable results, with a statistical precision of few per mille,
can be obtained from the runs easily attainable in a few hours of CPU time
of any modern PC.
\vskip 0.3 cm
\centerline{ \bf \large Acknowledgements}
\vskip 0.3 cm
Authors are indebted to S. Jadach, W. P\l{}aczek, A. Vallasi, M. Witek for
stimulating discussions and cooperation at different steps of
the algorithm development. One of the authors (MS) acknowledges
support of the CERN ALEPH group.
|
\section*{Acknowledgments}
This work was supported by the U. S. Department of Energy
under grant DE-FG05-85ER40226.
|
\section{\@startsection {section}{1}{\z@}%
{-3.5ex \@plus -1ex \@minus -.2ex}%
{2.3ex \@plus.2ex}%
{\normalfont\large\bfseries}}
\renewcommand\subsection{\@startsection{subsection}{2}{\z@}%
{-3.25ex\@plus -1ex \@minus -.2ex}%
{1.5ex \@plus .2ex}%
{\normalfont\normalsize\bfseries}}
\renewcommand\subsubsection{\@startsection{subsubsection}{3}{\z@}%
{3.25ex plus 1ex minus .2ex}{-.5em}%
{\normalfont\normalsize\bfseries}}
\makeatother
\setlength{\oddsidemargin}{-9.4 true mm}
\setlength{\evensidemargin}{-9.4 true mm}
\setlength{\textwidth}{178 true mm}
\setlength{\textheight}{238 true mm}
\setlength{\topmargin}{-17 true mm}
\begin{document}
\title{The triangulated Hopf category $n_+SL(2)$}
\author{Volodymyr Lyubashenko}
\address{
Institute of Mathematics\\
National Academy of Sciences of Ukraine\\
3, Teresh\-chen\-kiv\-ska st.\\
Kyiv-4, 01601\\
Ukraine
}
\email{<EMAIL>}
\date{math.QA/9904108; Edited March 28, 2001; Printed \today}
\begin{abstract}
We discuss an example of a triangulated Hopf category related to SL(2).
It is an equivariant derived category equipped with multiplication
and comultiplication functors and structure isomorphisms.
We prove some coherence equations for structure isomorphisms.
In particular, the Hopf category is monoidal.
\end{abstract}
\thanks{Research was supported in part by National Science Foundation
grant 530666.}
\maketitle
\newtheorem*{acknowledgements}{Acknowledgements}
\newtheorem*{notations}{Notations}
\newlength{\texthigh}
\setlength{\texthigh}{\textheight}
\addtolength{\texthigh}{-0.12pt}
\newlength{\textwids}
\setlength{\textwids}{\textwidth}
\addtolength{\textwids}{-0.12pt}
\providecommand{\text{\cit\Zh\/}}{\mathsf{X\mkern-8.2mu I\mkern3mu}}
\allowdisplaybreaks[1]
\section{Introduction}
Crane and Frenkel proposed a notion of a Hopf category
\cite{CraneFrenk:4-dim}. It was motivated by Lusztig's approach to
quantum groups -- his theory of canonical bases. In particular, Lusztig
obtains braided deformations $U_q\mathfrak n_+$ of universal enveloping
algebras $U\mathfrak n_+$ for some nilpotent Lie algebras
$\mathfrak n_+$ together with canonical bases of these braided Hopf
algebras \cite{Lus:canonical,Lus:quiver,Lus:book}. Elements of the
canonical basis are identified with isomorphism classes of simple
perverse sheaves -- certain objects of equivariant derived categories.
They are contained in the subcategory of semisimple complexes. One of
the proposals of Crane and Frenkel is to study this category rather
than its Grothendieck ring $U_q\mathfrak n_+$. Conjectural properties
of this category were collected into a system of axioms of a Hopf
category, equipped with functors of multiplication and
comultiplication, isomorphisms of associativity, coassociativity and
coherence which satisfy four equations \cite{CraneFrenk:4-dim}. A
mathematical framework and some examples were provided by
Neuchl~\cite{Neuchl-rtHc}.
Crane and Frenkel \cite{CraneFrenk:4-dim} gave an example of a Hopf
category resembling the semisimple category encountered in Lusztig's
theory corresponding to one-dimen\-sio\-nal Lie algebra $\mathfrak n_+$
-- nilpotent subalgebra of $\mathfrak{sl}(2)$. We want to discuss an
example of a related notion -- triangulated Hopf category -- the whole
equivariant derived category equipped with multiplication and
comultiplication functors and structure isomorphisms. In particular,
it is a monoidal category. New feature of coherence is that additive
relations of \cite{CraneFrenk:4-dim} are replaced with distinguished
triangles. This new structure does not induce a Hopf category
structure of Crane and Frenkel on the subcategory of semi-simple
complexes. The missing component is a consistent choice of splitting of
splittable triangles. Verification of some of the consistency
equations is still an open question.
To give more details let us first recall some braided Hopf algebra
$H$. As an algebra $H$ is the algebra of polynomials of one variable
over $R=\ZZ[q,q^{-1}]$. More precisely, $H\subset\QQ(q)[x]$ is an
$R$\n-submodule spanned by the elements
\[ y_n = \frac{x^n}{(n)_{q^{-2}}!}, \qquad n\ge0, \]
\[(n)_{q^{-2}}! = \prod_{m=1}^n (m)_{q^{-2}}, \qquad (m)_{q^{-2}} =
\frac{1-q^{-2m}}{1-q^{-2}}. \]
The basis $(y_n)_{n\ge0}$ is called a canonical basis.
Multiplication table in this basis is
\[ y_k\cdot y_l = \binom{k+l}k_{q^{-2}} y_{k+l},\qquad
\binom{k+l}k_{q^{-2}} = \frac{(k+l)_{q^{-2}}!}{(k)_{q^{-2}}!
(l)_{q^{-2}}!} \in R. \]
Comultiplication by definition is
\[ y_n= \sum_{k+l=n} y_k\tens y_l. \]
These operations make $H$ into a $\ZZ$\n-graded $R$\n-algebra and
coalgebra. We equip the category ${\mathcal C}$ of $\ZZ$\n-graded
$R$\n-modules with the braiding
\[ c: M\tens_R N \to N\tens_R M, \qquad c = \sum_{k,l} c_{k,l}, \]
\[
\begin{tanglec}
\object{k}\Step\object{l} \\
\xx \\
\object{l}\Step\object{k}
\end{tanglec}
= c_{k,l}: M_k\tens N_l \to N_l\tens M_k, \qquad c_{k,l}=q^{-2kl}, \]
where $M=\sum_{k\in\ZZ}M_k$, $N=\sum_{l\in\ZZ}N_l$ are graded
$R$\n-modules. Then $H$ is a Hopf algebra in a braided category ${\mathcal C}$ as
defined in \cite{Ma:bg}. Using graphical notations
\[ m=\sum_{k,l}m_{k,l}, \qquad
\vstretch 60
\begin{tanglec}
\object{k}\Step\object{l} \\
\@d{}{0,0,}} \def\dd{\@d{0,0,}{}\dd \\
\s \\
\object{k+l}
\end{tanglec}
=m_{k,l}: H_k\tens_R H_l \to H_{k+l}, \]
\[ \Delta=\sum_{k,l}\Delta_{k,l}, \qquad
\vstretch 60
\begin{tanglec}
\object{k+l} \\
\n \\
\dd\@d{}{0,0,}} \def\dd{\@d{0,0,}{} \\
\object{k}\Step\object{l}
\end{tanglec}
=\Delta_{k,l}: H_{k+l} \to H_k\tens_R H_l, \]
we can write the coherence axiom as an equation
\begin{equation}\label{HnHmHpHq}
\vstretch 120
\begin{tanglec}
\nodeld{n}\Step\noderd{m}\\
\nw1\nodeld{n+m}\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\ne1\\
\n\\
\nodelu{p}\sw1\se1\noderu{q}
\end{tanglec}
\quad\; = \quad
\sum_{\substack{i+j=n \\ k+l=m \\ i+k=p \\ j+l=q}} \quad
\vstretch 72
\begin{tanglec}
\nodeld{n}\@ifnextchar[{\@step}{\@step[1]}[3]\noderd{m}\\
\nodeld{i}\n\noderd{j}\@ifnextchar[{\@step}{\@step[1]}[3]\nodeld{k}\n\noderd{l}\\
\sw1\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\se1\\
\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\Step\hbx(1,2){\EMl@ne(0,20,3,14)\EMl@ne(0,0,10,20)\EMl@ne(10,0,7,6)}\Step\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\\
\nw1\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\ne1\\
\nodelu{p}\s\@ifnextchar[{\@step}{\@step[1]}[3]\s\noderu{q}
\end{tanglec}
\quad: H_n\tens_R H_m \to H_p\tens_R H_q
\end{equation}
for all $n,m,p,q\in\ZZ_{\ge0}$ such that $n+m=p+q$.
This algebra was obtained by Lusztig\cite{Lus:book} from the
following setup:
${\mathcal H}_{n_1,\dots,n_k}$ are $\CC$\n-linear categories, depending
symmetrically on parameters $n_1$,\dots,$n_k\in\ZZ_{\ge0}$,
$k\in\ZZ_{\ge0}$;
$m_{k,l}:{\mathcal H}_{k,l}\to{\mathcal H}_{k+l}$, $\Delta_{k,l}:{\mathcal H}_{k+l}\to{\mathcal H}_{k,l}$
are $\CC$\n-linear functors of
multiplication and comultiplication;
$c_{k,l}:{\mathcal H}_{k,l}\to{\mathcal H}_{l,k}$ are braiding functors;
there are associativity isomorphisms
\begin{diagram}
{\mathcal H}_{k,l,n} & \rTTo^{1\dlgn m_{l,n}} & {\mathcal H}_{k,l+n} \\
\dTTo<{m_{k,l}\dlgn1} & \simeq & \dTTo>{m_{k,l+n}} \\
{\mathcal H}_{k+l,n} & \rTTo^{m_{k+l,n}} & {\mathcal H}_{k+l+n}
\end{diagram}
where the meaning of $\dlgn$ will be specified further;
there are similar coassociativity isomorphisms.
The category ${\mathcal H}_{n_1,\dots,n_k}$ is
$D^{b,c}_{GL(n_1)\times\dots\times GL(n_k)}(pt)$ -- the bounded
constructible equivariant derived category of a point. It has a
distinguished object $Y_{n_1,\dots,n_k}$ -- the constant sheaf,
which is the complex
\[ \dots \to 0 \to 0 \to \CC \to 0 \to 0 \to \dots \]
concentrated in degree 0. It turns out that the collection
$(Y_{n_1,\dots,n_k})$ is closed under multiplication and
comultiplication (up to coefficients which are graded vector spaces):
\[ m_{k,l}(Y_{k,l}) \simeq H^{\scriptscriptstyle\bullet}(\Gr_k^{k+l}(\CC),\CC)\tens_\CC
Y_{k+l}. \]
The coefficient vector space here is de Rham
cohomology of the Grassmannian $\Gr_k^{k+l}(\CC)$ -- manifold of
$k$\n-dimensional subspaces of a $k+l$\n-dimensional space.
Cohomology is concentrated in even degrees and the Betti numbers
\[ \beta_i = \dim_\CC H^{2i}(\Gr_k^{k+l}(\CC),\CC) \]
are coefficients of the expansion of a $q$\n-binomial coefficient
in powers of $q^{-2}$:
\[ \binom{k+l}k_{q^{-2}} = \sum_{i\ge0} \beta_i q^{-2i}. \]
Replacing the degree with the power of $q$ we get the multiplication
table for the canonical basis $(y_k)$. Comultiplication law is
recovered from
\[ \Delta_{k,l}(Y_{k+l}) \simeq Y_{k,l}. \]
The braiding functor
\[ c_{k,l} = \bigl( D_{GL(k)\times GL(l)}(pt) \simeq D_{GL(l)\times
GL(k)}(pt) \rTTo^{[-2kl]} D_{GL(l)\times GL(k)}(pt) \bigr) \]
is essentially the degree shift by $-2kl$. It translates into
multiplication by $q^{-2kl}$ for the braiding in algebra setting.
In the present paper we shall discuss coherence at the category
level. If one replaces linear mappings in equation~\eqref{HnHmHpHq}
with functors and $\sum$ with $\oplus$
the equation fails: the left and the right hand side functors
${\mathcal H}_{n,m}\to{\mathcal H}_{p,q}$ are, in general, not isomorphic. (Restricted
to $Y_{n,m}$ they give, however, isomorphic results.) One of the results
of the present paper is the following. Value of the left hand
side functor on an object $X$ of ${\mathcal H}_{n,m}$ is a repeated extension
of values on $X$ of summands in the right hand side in the sense
of distinguished triangles. Precise analogy is as follows: a sheaf
$S$ on a topological space $W$ is an
extension of its quotient-sheaf $S_F$ supported on closed subset $F$ by
subsheaf $S_U$ supported on its open complement $U=W-F$.
Technically, this is achieved by introducing new operations-functors
with two inputs and two outputs
\[
\vstretch 80
\begin{tanglec}
\nodeld{n}\Step\noderd{m}\\
\nw1\noder{{\mathcal O}}\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\ne1\\
\nodelu{p}\sw1\se1\noderu{q}
\end{tanglec}
\qquad : D^{b,c}_{GL(n)\times GL(m)}(pt) \to D^{b,c}_{GL(p)\times GL(q)}(pt),
\]
which depend on a parameter ${\mathcal O}$ -- a $P_{p,q}\times P_{n,m}$\n-invariant
subset of $GL(n+m)$, where $p+q=n+m$, the parabolic subgroup
$P_{n,m}\subset GL(n+m)$ consists of matrices preserving
$\CC^n\subset\CC^{n+m}$. Minimal such subsets ${\mathcal O}$ are double cosets --
points of the double coset space $P_{p,q}\backslash GL(n+m)/P_{n,m}$.
This is a finite set, it is in bijection with the set of quadruples
$(i,j,k,l)$, $i,j,k,l\in\ZZ_{\ge0}$, which satisfy the equations
$i+j=n$, $k+l=m$, $i+k=p$, $j+l=q$. Hence, we may index the
$P_{p,q}\times P_{n,m}$\n-orbits with these quadruples, say,
${\mathcal O}_{ijkl}$.
First, we prove that the left and the right hand sides of
\eqref{HnHmHpHq} are isomorphic to the above mentioned operations:
\[
\phantom0\quad
\vstretch 100
\begin{tanglec}
\nodeld{n}\Step\noderd{m}\\
\nw1\nodeld{n+m}\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\ne1\\
\n\\
\nodelu{p}\sw1\se1\noderu{q}
\end{tanglec}
\quad\;\simeq\;\quad
\vstretch 150
\begin{tanglec}
\nodeld{n}\Step\noderd{m}\\
\nw1\noder{GL(n+m)}\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\ne1\\
\nodelu{p}\sw1\se1\noderu{q}
\end{tanglec}
\qquad\qquad;\qquad
\vstretch 60
\begin{tanglec}
\nodeld{n}\@ifnextchar[{\@step}{\@step[1]}[3]\noderd{m}\\
\nodeld{i\;}\n\noderd{\;j}\@ifnextchar[{\@step}{\@step[1]}[3]\nodeld{k\;}\n\noderd{\;l}\\
\sw1\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\se1\\
\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\Step\hbx(1,2){\EMl@ne(0,20,3,14)\EMl@ne(0,0,10,20)\EMl@ne(10,0,7,6)}\Step\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\\
\nw1\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\ne1\\
\nodelu{p}\s\@ifnextchar[{\@step}{\@step[1]}[3]\s\noderu{q}
\end{tanglec}
\quad\;\simeq\;\quad
\vstretch 150
\begin{tanglec}
\nodeld{n}\Step\noderd{m}\\
\nw1\noder{{\mathcal O}_{ijkl}}\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\ne1\\
\nodelu{p}\sw1\se1\noderu{q}
\end{tanglec}
\qquad.
\]
Since $GL(n+m)=\sqcup_{i,j,k,l}{\mathcal O}_{ijkl}$, the former functor above
is a repeated extension of the latter functors via distinguished
triangles.
This shows usefulness of operations with many inputs and outputs for
our purposes. They are also used to prove an equation for associativity
isomorphisms which makes $\cupD^{b,c}_{GL(k)}(pt)$ into a monoidal
category. Similar equation for coassociativity isomorphisms is proven
as well.
\begin{acknowledgements}
I am grateful to L.~Crane and D.~Yetter for fruitful discussions which
stimulated this study. Commutative diagrams in this paper are drawn
with the help of a package {\tt diagrams} of Paul Taylor.
\end{acknowledgements}
\numberwithin{equation}{subsection}
\section{Preliminaries}
A definition of equivariant derived categories
is given by Bernstein and Lunts~\cite{BernsL:Equivariant}.
First we explain basic terms.
With a topological space $X$ is associated the category $Sh(X)$
of sheaves of topological spaces.
Its derived category is denoted $D(X)$.
The subcategory consisting of bounded complexes of sheaves
is denoted $D^b(X)$.
If $X$ is a complex algebraic variety,
we call a sheaf \emph{constructible}
if it is constructible with respect to some stratification
by algebraic submanifolds
and stalks are finite-dimensional vector spaces.
A complex is cohomologically constructible
if its cohomology sheaves are constructible.
Subcategory of bounded constructible complexes
is denoted $D^{b,c}(X)$.
\subsection{Equivariant derived categories}
Assume that a complex linear algebraic group $G$
acts algebraically on a complex algebraic variety $X$.
In this setting Bernstein and Lunts~\cite{BernsL:Equivariant}
define bounded constructible equivariant derived category
$D^{b,c}_G(X)$, as a fiber category.
A $G$\n-variety $P$ is called free if $G$ acts freely on $P$
and the quotient map $q:P\to G\backslash P=\overline{P}$
is a locally trivial fibration with the fiber $G$.
A $G$\n-resolution of a $G$\n-variety $X$ is a $G$\n-map $P\to X$,
where the $G$\n-variety $P$ is free.
Let $j:{\mathcal J}\to\Resto(X,G)$, $P\mapsto (jP:JP\to X)$ be a functor
to the category of $G$\n-resolutions.
Let ${\mathcal T}$ denote the category of complex algebraic varieties.
Let us denote $\Phi:\Resto(X,G)\to{\mathcal T}$,
$(R\to X)\mapsto\overline{R}=G\backslash R$ quotient functor.
Consider the composite functor
\[ \Psi:{\mathcal J} \rTTo^j \Resto(X,G) \rTTo^\Phi {\mathcal T}, \qquad
P\mapsto (jP:JP\to X) \mapsto \overline{JP}, \]
and define the fiber-category $D^{b,c}(\Psi)$ as follows.
\begin{definition}[Bernstein and Lunts~\cite{BernsL:Equivariant}]
An object of $D^{b,c}(\Psi)$ is a function
$M:\Resto(X,G)\ni P\mapsto M(P)\in D^{b,c}(\overline{JP})$
equipped with isomorphisms
$\alpha_\nu:(\overline{J\nu})^*(M(R))\to M(P)$ given for any
$\nu:P\to R\in\Mor{\mathcal J}$, such that for any pair of composable morphisms
$P \rTTo^\nu R \rTTo^\mu S$ we have
\[ \alpha_{\mu\nu} = \Bigl[ \overline{J(\mu\nu)}^*M(S) \simeq
\overline{J\nu}^*\overline{J\mu}^*M(S)
\rTTo^{\overline{J\nu}^*\alpha_\mu} \overline{J\nu}^*M(R)
\rTTo^{\alpha_\nu} M(P)\Bigr]. \]
A morphism $\phi:M\to N$ is a collection $\phi(P):M(P)\to N(P)$,
$P\in\Ob{\mathcal J}$, compatible with $\alpha_\nu$
for any $\nu:P\to R\in\Mor{\mathcal J}$:
\begin{diagram}
(\overline{J\nu})^*(M(R)) & \rTTo^{\alpha_\nu^M} & M(P) \\
\dTTo<{(\overline{J\nu})^*\phi(R)} & = & \dTTo>{\phi(P)} \\
(\overline{J\nu})^*(N(R)) & \rTTo^{\alpha_\nu^N} & N(P)
\end{diagram}
Define \emph{equivariant derived category} as $D^{b,c}_G(X)=D^{b,c}(\Phi)$
in the case of identity functor $j=\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}:{\mathcal J}\rId\Resto(X,G)$.
\end{definition}
We shall also use the notation $\edc{G}X=D^{b,c}_G(X)$
for equivariant derived category.
Notice that, if $X$ is $G$\n-free, then $\edc{G}X$ is equivalent to
$D^{b,c}(G\bs X)$.
Without freeness assumption the former and the latter categories
are not equivalent, in general.
Bernstein and Lunts compute the equivariant derived category
in the case when $X$ is a point.
\begin{theorem}[Bernstein and Lunts \cite{BernsL:Equivariant}
Theorem~12.7.2]
Assume that $G$ is a connected linear algebraic group.
The triangulated category $D^{b,c}_G(pt)$
is equivalent to the derived category of the category
of finitely generated differential graded $A$\n-modules,
where the graded algebra $A=H^{\scriptscriptstyle\bullet}(BG,\CC)$
is equipped with zero differential.
\end{theorem}
For $G=GL(n,\CC)$ the algebra $A$ is the algebra of symmetric
polynomials of $n$ variables
\[ A_n=\CC[x_1,\dots,x_n]^{{\mathfrak S}_n} \simeq \CC[e_1,\dots,e_n], \]
where $\deg x_i=2$ and $e_j$ are elementary symmetric functions. For
$G=GL(n_1,\CC)\times\dots\times GL(n_k,\CC)$ we have
$A=A_{n_1}\tens_\CC\dots\tens_\CC A_{n_k}$.
\subsection{Equivariant derived functors}
\subsubsection{The inverse image functor.}
Suppose that $\phi:G\to H$ is a group homomorphism,
and $f:X\to Y$ is a $\phi$\n-equivariant map.
We want to define a functor $f^*=(f,\phi)^*:D^{b,c}_H(Y)\toD^{b,c}_G(X)$.
First, denote ${\mathcal J}=\Resto(f,\phi)$ the category, whose objects
are $\phi$\n-maps $\underline{f}:P\to R$ of resolutions, that is,
\begin{diagram}
P & \rTTo^{\underline{f}} & R \\
\dTTo && \dTTo \\
X & \rTTo^f & Y
\end{diagram}
commutes. Morphisms $\nu:\underline{f}\to\underline{f'}$
are pairs of morphisms of resolutions $\nu_1:P\to P'$, $\nu_2:R\to R'$
such that
\begin{diagram}
P & \rTTo^{\underline{f}} & R \\
\dTTo<{\nu_1} && \dTTo>{\nu_2} \\
P' & \rTTo^{\underline{f'}} & R'
\end{diagram}
commutes. Use the functor $j:\Resto(f;\phi)\to\Resto(X,G)$,
$j(\underline{f}:P\to R)=P$ and $\Psi=j\Phi$.
Bernstein and Lunts \cite{BernsL:Equivariant} have shown that the
restriction functor $D^{b,c}_g(X)\toD^{b,c}(\Psi)$ is an equivalence.
Similarly to Bernstein and Lunts \cite{BernsL:Equivariant} we
define the first version of the inverse image functor:
\[ f^*: D^{b,c}_H(Y) \to D^{b,c}(\Psi:\Resto(f;\phi)\to{\mathcal T}), \]
\begin{multline*}
f^*(M:R\mapsto M(R)\inD^{b,c}(\overline{R})) \\
= \bigl[ f^*M: (P \rTTo^{\underline{f}} R) \mapsto
\overline{\underline{f}}^*(M(R)) \inD^{b,c}(\overline{P}) \bigr].
\end{multline*}
Next thing is to choose for all $(f,\phi)$ an equivalence
\[ F_{f,\phi}:D^{b,c}(\Psi:\Resto(f;\phi)\to{\mathcal T}) \to D^{b,c}_G(X) \]
quasi-inverse to the canonical restriction functor
\begin{align*}
\Can_f: D^{b,c}_G(X) &\longrightarrow D^{b,c}(\Psi:\Resto(f;\phi)\to{\mathcal T}), \\
[P\mapsto M(P)] &\longmapsto [(f:P\to R)\mapsto M(P)].
\end{align*}
The chosen isomorphisms of the composition
with the identity functor are denoted
\begin{multline*}
\eta_f:\bigl[ D^{b,c}(\Resto(f;\phi)\to{\mathcal T}) \rTTo^{F_{f,\phi}} D^{b,c}_G(X)
\rTTo^{\Can_f} D^{b,c}(\Resto(f;\phi)\to{\mathcal T}) \bigr] \\
\to \Id,
\end{multline*}
\begin{equation*}
\epsilon_f: \Id \to \bigl[ D^{b,c}_G(X) \rTTo^{\Can_f}
D^{b,c}(\Resto(f;\phi)\to{\mathcal T}) \rTTo^{F_{f,\phi}}
D^{b,c}_G(X) \bigr].
\end{equation*}
Define the \emph{inverse image functor} of $(f,\phi)$ as
\[ f^*=(f,\phi)^*: D^{b,c}_H(Y) \rTTo^{f^*} D^{b,c}(\Psi:\Resto(f;\phi)\to{\mathcal T})
\rTTo^{F_{f,\phi}} D^{b,c}_G(X). \]
For composable maps
\[ X \rTTo^f Y \rTTo^g Z \rTTo^h W \quad\text{ over }\quad
G \rTTo^\phi H \rTTo^\psi K \rTTo^\chi L \]
we define categories $\Resto(f,g;\phi,\psi)$ and
$\Resto(f,g,h;\phi,\psi,\chi)$, whose objects are pairs
$(\underline{f},\underline{g})$ (resp. triples
$(\underline{f},\underline{g},\underline{h})$)
of morphisms of resolutions over $(f,g)$ (resp. $(f,g,h)$)
\begin{diagram}
P & \rTTo^{\underline{f}} & R & \rTTo^{\underline{g}} & S &
\rTTo^{\underline{h}} & Q \\
\dTTo && \dTTo && \dTTo && \dTTo \\
X & \rTTo^f & Y & \rTTo^g & Z & \rTTo^h & W
\end{diagram}
Morphisms are triples $P\to P'$, $R\to R'$, $S\to S'$ (resp. quadruples
\dots, $Q\to Q'$) of morphisms of resolutions compatible with
$(\underline{f},\underline{g},\underline{h})$ and
$(\underline{f'},\underline{g'},\underline{h'})$.
There is an isomorphism $\iota:f^*g^*\simeq (gf)^*$ determined uniquely
by the equation
\begin{multline}
\begin{diagram}[inline,nobalance]
&& D^{b,c}_H(Y) &&&& D^{b,c}(\Resto(f,g;\phi,\psi)\to{\mathcal T}) \\
& \ruTTo^{g^*} & \dTwoar<{\iota_{g,f}} & \rdTTo^{f^*} &&& \uTTo>\wr \\
D^{b,c}_K(Z) && \rTTo_{(gf)^*} && D^{b,c}_G(X) & \rTTo_\sim &
D^{b,c}(\Resto(f;\phi)\to{\mathcal T})
\end{diagram}
= \\
\begin{diagram}[inline,width=2em,height=1em,objectstyle=\scriptstyle]
D^{b,c}_H(Y) && \rTTo^{f^*} && D^{b,c}(\Resto(f;\phi)\to{\mathcal T}) &&
\rTTo^{F_{f,\phi}} && D^{b,c}_G(X) \\
& \rdTwoar(2,4)_{\eta_g} \rdTTo(4,4) &&&& \rdId(4,4) &&
\ldTwoar_{\eta_f} & \\
\uTTo<{F_{g,\psi}} &&&& = && \phantom0 && \dTTo \\
&&&& &&&& \\
D^{b,c}(\Resto(g;\psi)\to{\mathcal T}) && \rId && D^{b,c}(\Resto(g;\psi)\to{\mathcal T}) &&&&
D^{b,c}(\Resto(f;\phi)\to{\mathcal T}) \\
& \rdTwoar(4,4)^{i_{g,f}} &&&& \rdTTo(4,4)^{f^*} &&& \\
\uTTo<{g^*} &&&& &&&& \dTTo \\
&&&& &&&& \\
D^{b,c}_K(Z) &&&& \hspace*{-4.5mm} D^{b,c}(\Resto(gf;\psi\phi)\to{\mathcal T})
&& \rTTo_\sim && D^{b,c}(\Resto(f,g;\phi,\psi)\to{\mathcal T}) \\
&&& \ruId(4,4) &&&&& \\
\dTTo<{(gf)^*} && \phantom0 && \uTTo>\wr && = && \uTTo>\wr \\
&&& \rdTwoar^{\quad\eta_{gf}^{-1}} &&&&& \\
D^{b,c}(\Resto(gf;\psi\phi)\to{\mathcal T}) && \rTTo_{F_{gf,\psi\phi}} && D^{b,c}_G(X)
&& \rTTo_\sim && D^{b,c}(\Resto(f;\phi)\to{\mathcal T})
\end{diagram}
\label{eq-dia-iota}
\end{multline}
\subsubsection{The direct image functor.}
Let $f:X\to Y$ be a $\phi$\n-map, $\phi:G\to H$.
Assume that $X$ is $K$\n-free, where $K=\Ker\phi\subset G$.
For our purposes it suffices to define $f_*:D^{b,c}_G(X)\toD^{b,c}_H(Y)$
as a right adjoint functor to $f^*:D^{b,c}_H(Y)\toD^{b,c}_G(X)$.
Furthermore,
we shall use it mainly in the quotient equivalence situation:
$H=G/K$, $\phi:G\to G/K$ is the canonical projection,
$X$ is $K$\n-free, $Y=K\bs X$,
$f=\pi:X\to K\bs X$ is the canonical projection.
\subsubsection{Quotient equivalence.}
\begin{subtheorem}[Bernstein and Lunts~\cite{BernsL:Equivariant}]
Let $K$ be a normal subgroup of $G$,
let $X$ be a $G$\n-space which is free as a $K$\n-space.
Then the quotient map $\pi:X\to K\backslash X$ gives an equivalence
\[ \pi^*: D^{b,c}_{G/K}(K\backslash X) \to D^{b,c}_G(X) \]
with a quasi-inverse $\pi_*$.
\end{subtheorem}
In this situation we shall make a concrete choice of a right adjoint
(and quasi-inverse) functor to $\pi^*$
\[ \pi_*: D^{b,c}_G(X) \to D^{b,c}_{G/K}(K\bs X), \quad N \mapsto \pi_*N, \]
\begin{multline*}
(\pi_*N)(R\to K\bs X) = N(R\times_{K\bs X}X\to X) \\
\in D^{b,c}(G\bs(R\times_{K\bs X}X)) \simeq D^{b,c}((G/K)\bs R),
\end{multline*}
the equivalence is due to the isomorphism
$G\bs(R\times_{K\bs X}X) \simeq (G/K)\bs R$.
\subsubsection{Induction equivalence.}
\begin{subtheorem}[Bernstein and Lunts~\cite{BernsL:Equivariant}]
Let $H$ be a subgroup of $G$, let $X$ be an $H$\n-space.
Then the induction map $i:X\to G\times_HX$, $x\mapsto(1,x)$
gives an equivalence
\[ i^*: D_G(G\times_HX) \to D_H(X) \]
with a quasi-inverse $i_*$.
\end{subtheorem}
\subsubsection{The direct image with proper supports.}
The following definition belongs to Bernstein and Lunts~
\cite{BernsL:Equivariant}.
\begin{definition}
Let $f:X\to Y$ be a map of $G$\n-varieties.
For any resolution $\pi:P\to Y$ there is a pull-back resolution
$\underline{\pi}$ of $X$.
\begin{diagram}
P\times_YX & \rTTo^{\underline{f}} & P \\
\dTTo<{\underline{\pi}} && \dTTo>{\pi} \\
X & \rTTo^f & Y
\end{diagram}
The functor $f_!:D^{b,c}_G(X)\toD^{b,c}_G(Y)$ maps an object
$M:(R\to X)\mapsto M(R\to X)\inD^{b,c}(\overline{R})$ to the object
$f_!M:(P\to Y)\mapsto
\overline{\underline{f}}_!(M(P\times_YX\to X)) \inD^{b,c}(\overline{P})$
equipped with an isomorphism
\begin{multline*}
\alpha_\nu^{f_!M}: \overline{\nu}^*\overline{f_R}_!M(R\times_YX\to X)
\rTTo^\beta_\sim
\overline{f_P}_!\overline{\underline{\nu}}^*M(R\times_YX\to X) \to \\
\rTTo^{\overline{f_P}_!\alpha_\nu^M} \overline{f_P}_!M(P\times_YX\to X)
\end{multline*}
for $\nu:P\to R$. Here $\beta$ is a base change isomorphism
obtained from the top square of the following prism.
\begin{diagram}[height=1.5em,width=2.7em,nobalance]
&&&& \overline{R\times_YX} &&& \rTTo^{\overline{f_R}} &&&
\overline{R} \\
&&& \ruTTo(4,2)^{\overline{\underline{\nu}}} & \uTTo &&&&&
\ruTTo(4,2)^{\overline{\nu}} & \\
\overline{P\times_YX} &&& \rTTo^{\overline{f_P}} & \HonV &&
\overline{P} &&&& \uTTo \\
&&&& \vLine && \uTTo &&&& \\
\uTTo &&&& R\times_YX & \hLine & \VonH & \rTTo^{f_R} &&& R \\
&&& \ruTTo(4,2)^{\underline{\nu}} \ldLine(1,2) &&&&&& \ruTTo(4,2)^{\nu}
\ldTTo(2,4) & \\
P\times_YX &&& \HonV & \rTTo^{f_P} && P &&&& \\
& \rdTTo & \ldTTo(1,2) &&&&& \rdTTo &&& \\
&& X &&& \rTTo &&& Y &&
\end{diagram}
Required property of $\alpha$ follows from
\lemref{lem-distr-base-chge} (see Appendix).
\end{definition}
\subsubsection{The equivariant base change isomorphism.}
\label{sec-equi-bas-chg}
Let the pull-back square
\begin{diagram}
W\SEpbk & \rTTo^e & X \\
\dTTo<h && \dTTo>f \\
Z & \rTTo^g & Y
\end{diagram}
consist of equivariant maps:
$h$ is a $G$\n-map, $f$ is an $H$\n-map and $e$, $g$ are $\phi$\n-maps,
where $\phi:G\to H$ is a group homomorphism.
There is a commutative cube
\begin{diagram}[height=1.8em]
P\times_ZW && \rTTo^{\underline{e}} && R\times_YX && \\
& \rdTTo_{\underline{h}} &&& \vLine & \rdTTo^{\underline{f}} & \\
\dTTo && P & \rTTo^{\underline{g}} & \HonV && R \\
&& \dTTo && \dTTo && \\
W & \hLine & \VonH & \rTTo^e & X && \dTTo \\
& \rdTTo_h &&&& \rdTTo^f & \\
&& Z && \rTTo^g && Y
\end{diagram}
where the left and the right walls and the bottom are pull-back squares.
It follows that the top is also a pull-back square.
We define a version of direct image functor with proper support:
\[ h_!: D^{b,c}(\Resto e\to{\mathcal T}) \to D^{b,c}(\Resto g\to{\mathcal T}),
\quad N\mapsto h_!N, \]
\[ (h_!N)(\underline{g}) = \overline{\underline{h}}_!(N(\underline{e}))
\inD^{b,c}(\underline{P}), \]
where $N(\underline{e})\inD^{b,c}(\overline{P\times_ZW})$.
Notice that the functor $h_!$ depends on $f$ as well,
which is not reflected in notation.
The base change isomorphism
\begin{diagram}[width=6em]
D^{b,c}(\Resto e\to{\mathcal T}) & \lTTo^{e^*} & D^{b,c}_H(X) \\
\dTTo<{h_!} & \luTwoar^\beta & \dTTo>{f_!} \\
D^{b,c}(\Resto g\to{\mathcal T}) & \lTTo^{g^*} & D^{b,c}_H(Y)
\end{diagram}
comes from the standard one for quotient spaces. The collection
\[ (g^*f_!M)(\underline{g}) =
\overline{\underline{g}}^*[(f_!M)(R\to Y)] =
\overline{\underline{g}}^*\overline{\underline{f}}_![M(R\times_YX\to X)]
\inD^{b,c}(\overline{P}) \]
is isomorphic to the collection
\[ (h_!e^*M)(\underline{g}) =
\overline{\underline{h}}_![(e^*M)(\underline{e})] =
\overline{\underline{h}}_!\overline{\underline{e}}^*[M(R\times_YX\to X)]
\inD^{b,c}(\overline{P}) \]
via base change isomorphism
$\beta:\overline{\underline{g}}^*\overline{\underline{f}}_! \to
\overline{\underline{h}}_!\overline{\underline{e}}^*$.
Finally, we define a full form of base change isomorphism
as the following diagram suggests:
\begin{diagram}[height=2.400000em,width=6em]
D^{b,c}_G(W) & \pile{\lTTo^{F_{f,\phi}}\\ \Downarrow\eta\\ \rTTo_{\Can_e}}
& D^{b,c}(\Resto e\to{\mathcal T}) & \lTTo^{e^*} & D^{b,c}_H(X) \\
\dTTo<{h_!} & = & \dTTo<{h_!} & \luTwoar^\beta & \dTTo>{f_!} \\
D^{b,c}_G(Z) &
\pile{\rTTo^{\Can_g}\\ \Downarrow\epsilon \\ \lTTo_{F_{g,\phi}}} &
D^{b,c}(\Resto g\to{\mathcal T}) & \lTTo_{g^*} & D^{b,c}_H(Y)
\end{diagram}
Namely, a full base change isomorphism is:
\begin{multline*}
\beta: g^*_{\text{full}}f_! = F_{g,\phi}g^*f_! \rTTo^{F\beta}
F_{g,\phi}h_!e^* \rTTo^{Fh_!\eta_f^{-1}}
F_{g,\phi}h_!\Can_eF_{f,\phi}e^* \\
= F_{g,\phi}\Can_gh_!F_{f,\phi}e^* \rTTo^{\epsilon_g^{-1}}
h_!F_{f,\phi}e^* = h_!e^*_{\text{full}}.
\end{multline*}
\section{The Hopf category $n_+SL(2)$}
\subsection{Setup and notations}
We partially follow Lusztig \cite{Lus:quiver} and
\cite{Lus:book}, Chapter~9 in notations.
Let $V$ be a vector space and
\[ G = G_V = GL(V) .\]
Let us make the product of $D_G(pt)$ over varying $\dim V$
into a sort of a graded Hopf category.
Assume we are given a decomposition
\[ {\mathcal V} :\qquad V^1\oplus V^2\oplus\dots\oplus V^k = V \]
into vector subspaces. Associate with it a filtration of $V$
\[ 0=V^{(0)}\subset V^{(1)}\subset \dots\subset V^{(n)}=V ,\qquad
V^{(m)} = V^1 \oplus\dots\oplus V^m .\]
$P_{\mathcal V}$ is the corresponding parabolic group
\[ P_{\mathcal V} = \{ g\in G_V \mid \forall m\ g(V^{(m)}) \subset V^{(m)} \}
= \{ g\in G_V \mid \forall m\ g(V^m) \subset V^{(m)} \} \]
and $U_{\mathcal V}$ is its unipotent radical. The group
\[ L_{\mathcal V} = \{ g\in G_V \mid \forall m\ g(V^m) \subset V^m \} =
\prod_{m=1}^k G_{V^m} \]
is a Levi subgroup of $P_{\mathcal V}$.
Notice that $P_{\mathcal V}$, $U_{\mathcal V}$ need only a filtration to be defined
unlike $L_{\mathcal V}$, which requires a direct sum decomposition.
\subsection{Suggestions for a monoidal 2-category}
To provide a final framework for Hopf categories
one would be interested to have a symmetric monoidal 2\n-category
of equivariant derived categories.
Tensor product of categories would be similar
to that of abelian \kd-linear categories introduced
by Deligne~\cite{Del:cat}. In particular,
$D^{b,c}_G(X)\dlgnD^{b,c}_H(Y)\simeqD^{b,c}_{G\times H}(X\times Y)$
is desirable. However, this wish does not look realistic.
To achieve it, possibly, one has to replace equivariant derived
categories with some other kind of categories, so that inverse
image functors and direct image functors make sense,
and the usual relations still hold.
Let us consider a question of tensor product of functors.
Let $f:X\to Y$, $g:Z\to W$ be maps of algebraic varieties.
Denote by $f^*$, $g^*$, $f_!$, $g_!$
the corresponding equivariant derived functors.
It is explained in \cite{Lyu-sqHopf} that one choice for $f^*\dlgn g^*$
is as good as another as long as they are isomorphic.
If in the isomorphism class of $f^*\dlgn g^*$ there is a functor
$(f\times g)^*$, we can modify the definition of $\dlgn$ and set
$f^*\dlgn g^*=(f\times g)^*$. If $f_!\dlgn g_!$ is isomorphic to
$(f\times g)_!$, we can set $f_!\dlgn g_!=(f\times g)_!$.
The isomorphism
\[ (\Id\dlgn g^*)\circ(f_!\dlgn\Id) \simeq
(f_!\dlgn\Id)\circ(\Id\dlgn g^*) \]
can be chosen as the base change isomorphism
$\beta:(Y\times g)^*(f\times W)_!\to(f\times Z)_!(X\times g)^*$,
constructed for the pull-back square
\begin{diagram}
X\times Z\SEpbk & \rTTo^{X\times g} & X\times W \\
\dTTo<{f\times Z} && \dTTo>{f\times W} \\
Y\times Z & \rTTo^{Y\times g} & Y\times W
\end{diagram}
We stress again that $\dlgn$ is far from being constructed.
Nevertheless, we prove some statements,
which can be interpreted as axioms of a Hopf category.
\subsection{Braiding}
Pretending that the categories
$D_{G_{a_1}\times\dots\times G_{a_k}}(pt)$ form
a monoidal 2\n-category, where $a_i$ are some vector spaces,
we define a braiding in it via functor
\begin{equation*}
c: D_{\prod G_{a_i}\times\prod G_{b_j}}(pt) \rTTo^{[-2d]}
D_{\prod G_{a_i}\times\prod G_{b_j}}(pt)
\rTTo^{\sigma^*} D_{\prod G_{b_j}\times\prod G_{a_i}}(pt) ,
\end{equation*}
where $d=(\sum_i\dim_\CC a_i)(\sum_j\dim_\CC b_j)$ and $\sigma$
is a permutation isomorphism of groups. The shift functor
$K\mapsto K[-2d]$ (the $R$\n-matrix) is related to other functors
that we are using via the following lemma.
\begin{lemma}
Let $h:E\to B$ be an affine linear $C^\infty$\n-bundle.
Then there is a (canonical in $B$) isomorphism of functors
\[ \bigl( D^{b,c}(B) \rTTo^{h^*} D^{b,c}(E) \rTTo^{h_!} D^{b,c}(B) \bigr)
\simeq T^{-2\dim_\CC h}. \]
\end{lemma}
\begin{proof}
Using Propositions 10.1 and 10.8(2) from \cite{Borel-Int_Cohom-V}
we find for any $K\inD^{b,c}(B)$
\[ h_!h^*K \simeq h_!h^*(\CC\tens K) \simeq h_!(h^*\CC\tens h^*K)
\simeq (h_!h^*\CC)\tens K. \]
Thus we have to prove that $h_!h^*\CC_B\simeq h_!\CC_E$ is isomorphic
to $\CC[-2d]$ for $d=\dim_\CC h$.
Choose a flat connection $\nabla$ on the bundle $h$. By definition
this is a $C^\infty(B)$-linear homomorphism of Lie algebras of
vector fields
\[ \nabla: \Vect(B) \to \Vect(E), \qquad \xi \mapsto \nabla_\xi, \]
such that for each point $e\in E$ we have
$Th({\nabla_\xi}_e)=\xi_{h(e)}$, where $Th:T_eE\to T_{h(e)}B$ is the
tangent map to $h$. The fields $\nabla_\xi$ (the horizontal vector
fields) form a $2b$\n-dimensional (over $\RR$) distribution in $E$,
where $b=\dim_\CC B$. This distribution is involutive, therefore,
by Frobenius theorem, locally there exist coordinates $(z_i,u_j)$,
$i=\overline{1,2b}$, $j=\overline{1,2d}$ in $E$ in which leaves of
the obtained foliation are described by equations $u_1=\const$, \dots,
$u_{2d}=\const$. Moreover, $z_i$ can be chosen as $z_i=x_i\circ h$,
where $x_i$ are local coordinates on the base $B$.
The sheaf of differential graded algebras of \emph{fibrewise forms}
$\Omega^{\scriptscriptstyle\bullet}_\nabla$ is defined as a subsheaf of $C^\infty$
differential forms on $E$
\[ \Omega^{\scriptscriptstyle\bullet}_\nabla(U) =
\{ \omega\in\Omega^{\scriptscriptstyle\bullet}_E(U) \mid \forall \xi\in\Vect(B) \quad
i_{\nabla_\xi}\omega=0 \text{ and } i_{\nabla_\xi}d\omega=0 \}. \]
The second condition may be replaced with $L_{\nabla_\xi}\omega=0$,
since Lie derivative can be computed as
$L_{\nabla_\xi}\omega=di_{\nabla_\xi}\omega+i_{\nabla_\xi}d\omega$.
Therefore, in local coordinates forms $\omega\in\Omega^{\scriptscriptstyle\bullet}_\nabla(U)$
are written as $f(u)du^\beta$. Absence of $dz_i$ is implied by the
first condition, and the Lie derivative condition implies independence
of the coefficients on $z_i$ coordinates. We conclude that the complex
$\Omega^{\scriptscriptstyle\bullet}_\nabla$ is a $c$\n-soft on fibers of $h$ resolution of
the constant sheaf $\CC_E$. Hence, it can be used to compute
the complex $h_!\CC_E$.
Let $V$ be an open subset of $B$. Then $h_!\CC_E(V)$ is a complex of
vector spaces $\Omega^{\scriptscriptstyle\bullet}_{\nabla,c}(h^{-1}(V))$, where $c$ indicates
such forms $\omega$ that $h^{-1}(K)\cap\supp\omega$ is compact for each
compact subset $K$ of $B$. This complex is exact everywhere except the
maximal degree $2d$. It has a map into the algebra of functions on $V$
\[ \alpha(V): \Omega^{2d}_{\nabla,c}(h^{-1}(V)) \to C^\infty(V), \quad
\omega \longmapsto (v\mapsto \int_{h^{-1}(v)}\omega), \]
given by fibrewise integration. Local presentation of $\omega$ implies
that the function $\alpha(V)(\omega)$ is locally constant, hence,
is in $\CC_B(V)$. The obtained chain map $\alpha:h_!\CC_E\to\CC_B[-2d]$
is a quasi-isomorphism.
\end{proof}
Another (complex analytic) construction of $\alpha$ will be published
elsewhere. It follows the lines of \cite{Lyu-l-adic-Mat-Studii}, where
the case of quasicompact schemes over $\overline{\FF_p}$ and
$\ell$\n-adic sheaves is considered. In the analytic setting one can
also prove that for any pull-back square
\begin{diagram}
F \SEpbk & \rTTo^j & A \\
\dTTo<g && \dTTo>f \\
E & \rTTo^h & B
\end{diagram}
where $h$ is an affine linear bundle (and so is $j$), we have an equation
\[
\begin{diagram}[inline,width=2.5em]
&& D^{b,c}(E) && \\
& \ruTTo^{h^*} & \dTTo>{g^*} & \rdTTo^{h_!} & \\
D^{b,c}(B) && D^{b,c}(F) && D^{b,c}(B) \\
\dTTo<{f^*} & \ruTTo^{j^*} & \dTwoar>{\alpha_A}
& \rdTTo^{j_!} & \dTTo>{f^*} \\
D^{b,c}(A) && \rTTo_{[-2d]} && D^{b,c}(A)
\end{diagram}
\quad = \quad
\begin{diagram}[inline,width=2.5em]
&& D^{b,c}(E) && \\
& \ruTTo^{h^*} & \dTwoar>{\alpha_B} & \rdTTo^{h_!} & \\
D^{b,c}(B) && \rTTo_{[-2d]} && D^{b,c}(B) \\
\dTTo<{f^*} && = && \dTTo>{f^*} \\
D^{b,c}(A) && \rTTo_{[-2d]} && D^{b,c}(A)
\end{diagram}
\]
This is the precise meaning of canonicity of isomorphism $\alpha$.
\begin{corollary}\label{cor-aff-bun-equi}
Let $h:E\to B$ be an affine linear $G$\n-bundle (an affine linear bundle
equipped with a group homomorphism $G\to\Aut_{\text{af.lin.bun.}}(h)$).
Then there is an isomorphism of functors
\[ \bigl( D^{b,c}_G(B) \rTTo^{h^*} D^{b,c}_G(E) \rTTo^{h_!} D^{b,c}_G(B) \bigr)
\simeq T^{-2\dim_\CC h}. \]
\end{corollary}
\begin{proof}
The system of isomorphisms $\alpha$ in $D^{b,c}(\overline{P})$ for
$G$\n-resolutions $P\to X$ is coherent due to canonicity of $\alpha$.
\end{proof}
\subsection{Operations}
Let two decompositions of $V$ into a direct sum be given:
\begin{alignat}2
{\mathcal V} &:&\qquad V^1\oplus V^2\oplus\dots\oplus V^k &= V
\label{eq-decomp-V-to-Vi2} \\
{\mathcal W} &:&\qquad W^1\oplus W^2\oplus\dots\oplus W^l &= V.
\label{eq-decomp-V-to-Wj2}
\end{alignat}
Let ${\mathcal O}\subset G$ be a left $P_{\mathcal W}$\n-invariant and right
$P_{\mathcal V}$\n-invariant subset. We associate with it an operation
\[ \text{\cit\Zh\/}_{\mathcal W}^{{\mathcal O};{\mathcal V}} = \quad
\begin{tangle}
\vstretch 120
\object{V^1}\Step\object{V^2}\@ifnextchar[{\@step}{\@step[1]}[6]\object{V^k} \\
\nw4\nw0\nw2\nodel{{\mathcal O}}\nw0\n\ne0\ne2\ne0\ne4 \\
\sw4\sw0\sw2\sw0\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se0\se2\se0\se4 \\
\object{W^1}\Step\object{W^2}\@ifnextchar[{\@step}{\@step[1]}[6]\object{W^k}
\end{tangle}
\quad = \quad
\vstretch 68
\begin{tangle}
\object{V^1}\Step\object{V^2}\@ifnextchar[{\@step}{\@step[1]}[6]\object{V^k} \\
\nw4\nw0\nw2\nodel{{\mathcal O}}\nw0\n\ne0\ne2\ne0\ne4 \\
\@ifnextchar[{\@step}{\@step[1]}[4]\idash \\
\sw4\sw0\sw2\sw0\s\se0\se2\se0\se4 \\
\object{W^1}\Step\object{W^2}\@ifnextchar[{\@step}{\@step[1]}[6]\object{W^k}
\end{tangle}
\quad = \pitchfork_{\mathcal W}^{\mathcal V} \circ \Psi_{\mathcal W}^{\mathcal V}.
\]
The components of it are defined below.
\begin{notations}
In graphical notations a straight line $\Big|X$ labeled by a vector
space $X$ denotes category $D_{G_X}(pt)$. A dashed line
$
\begin{tangle}
\idash
\end{tangle}
\:{\mathcal X}$ labeled by a filtration ${\mathcal X}$ of a vector space $X$ denotes
category $D_{P_{\mathcal X}}(pt)$. If a horizontal line crosses a diagram
intersecting transversally with several (solid or dashed) open intervals
labeled by spaces $Y_1$, $Y_2$,\dots and filtrations ${\mathcal X}_1$, ${\mathcal X}_2$
of spaces $X_1$, $X_2$,\dots , this line represents the category
\( D_{G_{Y_1}\times G_{Y_2}\times\dots\times
P_{{\mathcal X}_1}\times P_{{\mathcal X}_2}\times\dotsb}(pt) \).
The order of factors repeats the order of intersection points on the
horizontal line.
A vertex (or a horizontal row of vertices) represents a functor from
the category, corresponding to horizontal line just above the vertex,
to the category represented by a line just below the vertex.
\end{notations}
\subsubsection{The multiplication.}\label{sec-Multiplication}
Multiplication operation is
\begin{multline*}
\qquad
\vstretch 60
\begin{tangle}
\nodelu{V^1}\Step\nodelu{V^2}\@ifnextchar[{\@step}{\@step[1]}[6]\noderu{V^k} \\
\nw4\nw0\nw2\nodel{{\mathcal O}}\nw0\n\ne0\ne2\ne0\ne4 \\
\@ifnextchar[{\@step}{\@step[1]}[4]\idash\noder{{\mathcal W}} \\
\end{tangle}
\quad = \Psi_{\mathcal W}^{{\mathcal O};{\mathcal V}} = \Bigl( D_{L_{\mathcal V}}(pt) \rTTo^{\phi^*}
D_{P_{\mathcal W}\times L_{\mathcal V}}({\mathcal O}/{U_{\mathcal V}}) \\
\rTTo^{\pi_*} D_{P_{\mathcal W}}({\mathcal O}/{P_{\mathcal V}}) \rTTo^{\alpha_!}
D_{P_{\mathcal W}}(pt) \Bigr),
\end{multline*}
Here the scheme of multiplication is similar to that in \cite{Lus:book}:
\begin{equation}
pt \lTTo^\phi {\mathcal O}/{U_{\mathcal V}} \rTTo^\pi {\mathcal O}/{P_{\mathcal V}} \rTTo^\alpha pt,
\label{eq-p1p2p32}
\end{equation}
where $\pi$ is a canonical projection.
The action of $L_{\mathcal V}=P_{\mathcal V}/U_{\mathcal V}$ in ${\mathcal O}/{U_{\mathcal V}}$ is
induced from the action $p.o=op^{-1}$, $p\in P_{\mathcal V}$.
In particular, for $l=1$ we have $P_{\mathcal W}=G_V$, ${\mathcal O}=G_V$ and
the multiplication operation is
\begin{multline*}
\qquad
\vstretch 60
\begin{tangle}
\nodelu{V^1}\Step\nodelu{V^2}\@ifnextchar[{\@step}{\@step[1]}[6]\noderu{V^k} \\
\nw4\nw0\nw2\nodel{{\mathcal O}}\nw0\n\ne0\ne2\ne0\ne4 \\
\@ifnextchar[{\@step}{\@step[1]}[4]\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\end{tangle}
\quad = \Psi_1^{{\mathcal V}} = \Bigl( D_{L_{\mathcal V}}(pt) \rTTo^{\phi^*}
D_{G_V\times L_{\mathcal V}}(G_V/{U_{\mathcal V}}) \\
\rTTo^{\pi_*} D_{G_V}(G_V/{P_{\mathcal V}}) \rTTo^{\alpha_!} D_{G_V}(pt) \Bigr),
\end{multline*}
Here the scheme of multiplication is precisely that of Lusztig~
\cite{Lus:book}:
\begin{equation*}
pt \lTTo^\phi G_V/{U_{\mathcal V}} \rTTo^\pi G_V/{P_{\mathcal V}} \rTTo^\alpha pt.
\end{equation*}
The particular case $k=1$, ${\mathcal O}=G_V$ is also important. We have then
$L_{\mathcal V}=P_{\mathcal V}=G_V$, $U_{\mathcal V}=1$, and
\[ \Psi^1_{\mathcal W} = \bigl( D_G(pt) \rTTo^{\pr_2^*}
D_{P_{\mathcal W}\times G}(G) \rTTo^{d_{1*}} D_{P_{\mathcal W}}(pt) \bigr) ,\]
where $d_1:G\to pt$.
\begin{proposition}\label{pro-psi1=res}
The functor $\Psi^1_{\mathcal W}$ is isomorphic to the restriction functor
$\Resto_{P_{\mathcal W},G}:D_G(pt)\to D_{P_{\mathcal W}}(pt)$.
\end{proposition}
\begin{proof}
A map $d_1$ is the map $G\rEpi pt$.
Hence, $d_{1*}: D_{P_{\mathcal W}\times G}(G) \to D_{P_{\mathcal W}}(pt)$
is an equivalence with a quasi-inverse
$d_1^*: D_{P_{\mathcal W}}(pt) \to D_{P_{\mathcal W}\times G}(G)$.
The map $s_0:pt\to G$, $s_0(pt)=1$, is a $\Delta$\n-map
with respect to the homomorphism of groups
$\Delta:P_{\mathcal W}\to P_{\mathcal W}\times G$, $\Delta(p)=(p,p)$.
Furthermore, $s_0$ is an induction map
$pt \rTTo^i (P\times G)/P \rTTo^\rho_\sim G$, where $i(pt)=(1,1)$,
$\rho(p,g)=pg^{-1}$ and $\rho$ is a $P\times G$\n-map.
Therefore, $s_0^*: D_{P_{\mathcal W}\times G}(G) \to D_{P_{\mathcal W}}(pt)$
is an induction equivalence. Since
\( \bigl( pt \rTTo^{s_0} G \rTTo^{d_1} pt \bigr) = \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\),
we have
\[ \bigl( D_{P_{\mathcal W}}(pt) \rTTo^{d_1^*} D_{P_{\mathcal W}\times G}(G)
\rTTo^{s_0^*} D_{P_{\mathcal W}}(pt) \bigr) \simeq \Id .\]
Therefore, $s_0^* \simeq d_{1*}$. Hence,
\[ \Psi^1_{\mathcal W} \simeq \bigl( D_G(pt) \rTTo^{\pr_2^*}
D_{P_{\mathcal W}\times G}(G) \rTTo^{s_0^*} D_{P_{\mathcal W}}(pt) \bigr)
\simeq \Resto_{P_{\mathcal W},G},\]
for $\bigl(pt \rTTo^{s_0} G \rTTo pt\bigr) = \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}$ and
$P_{\mathcal W} \rTTo^\Delta P_{\mathcal W}\times G \rTTo^{\pr_2} G$ is an inclusion.
\end{proof}
\subsection{Associativity}
Assume that besides decomposition \eqref{eq-decomp-V-to-Vi2} of $V$ into
direct sum of $V^i$ we have also decompositions
\[ {\mathcal V}^i:\qquad V^i = \bigoplus_{m=1}^{m^i} V_m^i .\]
We can produce out of these the decomposition
\[ {\mathcal U}:\qquad V = \oplus_{i,m} V_m^i ,\]
where the lexicographic order of summands is used
$V_1^1$, \dots,$V_{m^1}^1,V_1^2$, \dots,$V_{m^2}^2$,
\dots,$V_1^l$, \dots,$V_{m^l}^l$.
An associativity isomorphism is obtained from the isomorphism
\[
\vstretch 80
\begin{tangle}
\nw1\nodelu{{\mathcal V}^1}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\nodelu{{\mathcal V}^2}\n\ne1\@ifnextchar[{\@step}{\@step[1]}%
\nw1\n\noderu{{\mathcal V}^k}\ne1 \\
\@ifnextchar[{\@step}{\@step[1]}\nw3\@ifnextchar[{\@step}{\@step[1]}\nodel{{\mathcal O}}\@ifnextchar[{\@step}{\@step[1]}\n\@ifnextchar[{\@step}{\@step[1]}\noder{{\mathcal V}}\@ifnextchar[{\@step}{\@step[1]}\ne3 \\
\@ifnextchar[{\@step}{\@step[1]}[4]\idash
\end{tangle}
\quad \rTTo^\sim \quad
\vstretch 120
\begin{tangle}
\nw4\nw3\nw2\nw1\n\ne1\noderu{\mathcal U}\ne2\ne3\ne4 \\
\@ifnextchar[{\@step}{\@step[1]}[4]\idash
\end{tangle}
\quad = \Psi^{\mathcal U}.
\]
This isomorphism can be read from the diagram in \figref{fig-ass}.
\begin{figure}
\begin{diagram}[height=2.400000em]
\eDc{\prod L_{{\mathcal V}^i}}{pt} && \rId && \eDc{L_{\mathcal U}}{pt} \\
\dTTo<{\prod\phi^{i*}} &&&& \dTTo>{\phi_{\mathcal U}^*} \\
\eDc{\prod{G_{V^i}\times L_{{\mathcal V}^i}}}{\prod_iG_{V^i}/{U_{{\mathcal V}^i}}}
& \rTTo^{\pr_{Y2}^*} &
\eDc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}
{{\mathcal O}/{U_{\mathcal V}}\times\prod_iG_{V^i}/{U_{{\mathcal V}^i}}}
& \rTTo^{\pi'_*}_\sim & \eDc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/{U_{\mathcal U}}} \\
\dTTo<{\prod\pi_*^i}>\wr && \dTTo<{\pi''_*}>\wr &&
\dTTo<\wr>{\pi_{{\mathcal U}*}} \\
\eDc{\prod G_{V^i}}{\prod_iG_{V^i}/{P_{{\mathcal V}^i}}} &
\rTTo^{\pr_{X2}^*} &
\eDc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}\times\prod_iG_{V^i}/{P_{{\mathcal V}^i}}}
& \rTTo^{\pi'''_*}_\sim & \eDc{P_{\mathcal W}}{{\mathcal O}/{P_{\mathcal U}}} \\
\dTTo<{\prod\alpha^i_!} && \dTTo<{\pr_{X1!}} && \dTTo>{\beta_!} \\
\eDc{L_{\mathcal V}}{pt} & \rTTo^{\phi_{\mathcal V}^*} & \eDc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}}
& \rTTo^{\pi_{{\mathcal V}*}}_\sim & \eDc{P_{\mathcal W}}{{\mathcal O}/{P_{\mathcal V}}} \\
&&&& \dTTo>{\alpha_{{\mathcal V}!}} \\
&&&& \eDc{P_{\mathcal W}}{pt}
\end{diagram}
\caption{Associativity isomorphism}
\label{fig-ass}
\end{figure}
The action of $P_{\mathcal W}\times L_{\mathcal V}$ in
$X={\mathcal O}/{U_{\mathcal V}}\times\prod_iG_{V^i}/{P_{{\mathcal V}^i}}$ is the following
\[ (p,l).[o\times\prod g_i] =
pol^{-1} \times\prod l_ig_i, \quad l=\prod l_i .\]
The map $\pi'''$ is the quotient map
\[ \pi''': X \to L_{\mathcal V}\backslash X \simeq {\mathcal O}/{P_{\mathcal U}} .\]
The action of $P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}$ in
$Y={\mathcal O}/{U_{\mathcal V}}\times\prod_iG_{V^i}/{U_{{\mathcal V}^i}}$ is the following
\[ (p,l,\bar l).[o \times\prod g_i] =
pol^{-1} \times\prod l_ig_i\bar l_i^{-1}, \quad
l=\prod l_i,\;\bar l=\prod\bar l_i .\]
The maps $\pi'$, $\pi''$ are the quotient maps
\[ \pi': Y \to L_{\mathcal V}\backslash Y \simeq {\mathcal O}/{U_{\mathcal U}} ,\quad
\pi'': Y \to L_{\mathcal U}\backslash Y \simeq X .\]
The canonical projection $\beta$ comes from the inclusion
$P_{\mathcal U}\subset P_{\mathcal V}$,
\[ \beta: {\mathcal O}/{P_{\mathcal U}} \to {\mathcal O}/{P_{\mathcal V}} .\]
\subsection{The associativity equation}
Assume that we have the following decompositions
of finite-dimensional $\CC$\n-vector spaces:
\begin{alignat*}4
{\mathcal W} &:&\quad V &= \oplus_j W^j, &\qquad
{\mathcal V} &:&\quad V &= \oplus_i V^i, \\
{\mathcal V}^i &:&\quad V^i &= \oplus_m V^i_m, &\qquad
{\mathcal V}^i_m &:&\quad V^i_m &= \oplus_p V^i_{m,p}.
\end{alignat*}
These decompositions imply also the following decompositions:
\begin{equation*}
{\mathcal U} :\; V = \oplus_{i,m} V^i_m, \quad\;
{\mathcal Y}^i :\; V^i = \oplus_{m,p} V^i_{m,p}, \quad\;
{\mathcal X} :\; V = \oplus_{i,m,p} V^i_{m,p}.
\end{equation*}
Let ${\mathcal O}\subset G_V$ be a left $P_{\mathcal W}$\n-invariant
and right $P_{\mathcal V}$\n-invariant subset.
Associativity isomorphisms give two isomorphisms
between the composite operation and the single operation
as in diagram
\begin{diagram}[height=4.5em,LaTeXeqno]
\vstretch 75
\begin{tangle}
\nw1\nodel{{\mathcal V}^1_1\;}\n\@ifnextchar[{\@step}{\@step[1]}\nw1\n\@ifnextchar[{\@step}{\@step[1]}\nw1\n\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1 \\
\@ifnextchar[{\@step}{\@step[1]}\nw1\nodel{{\mathcal V}^1\;}\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\ne1\Step\nw1\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\ne1 \\
\Step\nw2\@ifnextchar[{\@step}{\@step[1]}\nodelu{{\mathcal V}\;}\nodeld{{\mathcal W}\;}\noder{\;{\mathcal O}}\@ifnextchar[{\@step}{\@step[1]}\ne2 \\
\Step\sw2\sw1\s\se1\se2
\end{tangle}
& \rTTo^{\assoc} &
\begin{tangle}
\nw2\nw1\nodel{{\mathcal Y}^1\;}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw2\nw1\n\ne1\ne2 \\
\Step\nw2\@ifnextchar[{\@step}{\@step[1]}\nodelu{{\mathcal V}\;}\nodeld{{\mathcal W}\;}\noder{\;{\mathcal O}}\@ifnextchar[{\@step}{\@step[1]}\ne2 \\
\Step\sw2\sw1\s\se1\se2
\end{tangle}
\\
\dTTo<{\assoc} & = & \dTTo>{\assoc} & . \\
\begin{tangle}
\nw1\nodel{{\mathcal V}^1_1\;}\n\@ifnextchar[{\@step}{\@step[1]}\nw1\n\@ifnextchar[{\@step}{\@step[1]}\nw1\n\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1 \\
\@ifnextchar[{\@step}{\@step[1]}\nw3\nw0\nw1\nodelu{{\mathcal U}\;\;\;}\nodeld{{\mathcal W}\;}\noder{\;{\mathcal O}}
\ne1\ne0\ne3\\
\Step\sw2\sw1\s\se1\se2
\end{tangle}
& \rTTo^{\assoc} &
\vstretch 150
\begin{tangle}
\nw4\nw3\nw2\nw1\nodelu{{\mathcal X}\;\;\;}\nodeld{{\mathcal W}\;}\noder{\;{\mathcal O}}
\n\ne1\ne2\ne3\ne4\\
\Step\sw2\sw1\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se1\se2
\end{tangle}
\label{dia-asso-square}
\end{diagram}
\begin{proposition}[Associativity equation]
Diagram \eqref{dia-asso-square} is commutative.
\end{proposition}
\begin{proof}
The left-lower associativity composition is given on diagram in
\figref{fig-ll-assoc-start}.
\begin{figure}
\begin{diagram}[width=3.5em,height=2.400000em,nobalance]
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi_{\mathcal X}^*} &
\edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} & \rTTo^{\pi_{{\mathcal X}*}}_\sim &
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} & \rTTo^{\alpha_{{\mathcal X}!}} & \edc{P_{\mathcal W}}{pt} \\
\dTTo<{\prod\phi_m^{i*}} && \uTTo<\sim>{\pi'_*} &
\ruTTo(2,4)^\sim_{\pi''_*} && \ruTTo(2,4)_{\alpha_{{\mathcal U}!}} & \\
\edc{\prod G_{V^i_m}L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} &
\edc{P_{\mathcal W}\times L_{\mathcal U}\times L_{\mathcal X}}
{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
&& \dTTo>{\beta_{{\mathcal X}!}} && \uTTo>{\alpha_{{\mathcal V}!}} \\
\dTTo<{\prod\pi^I_{m*}}>\wr && \dTTo<\wr>{\pi''_*} &&&& \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pr_2^*} &
\edc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &&
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal U}} & \rTTo^{\beta_!} & \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal V}} \\
&& \dTTo>{\pr_{1!}} & \ruTTo^{\pi_{{\mathcal U}*}}_\sim &&& \\
&& \edc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal U}} && \uTTo<\wr>{\pi'''_*} &&
\uTTo<\wr>{\pi_{{\mathcal V}*}} \\
\dTTo<{\prod\alpha^i_{m!}} & \ruTTo(2,4)^{\phi^*_{\mathcal U}} &
\uTTo<\wr>{\pi'_*} &&&& \\
&& \edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/U_{{\mathcal V}^i}} &
\rTTo^{\pi''_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal V}^i}} &
\rTTo^{\pr_{1!}} & \edc{P_{\mathcal W} L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}} \\
&& \uTTo>{\pr_2^*} && \uTTo>{\pr_2^*} && \uTTo>{\phi_{\mathcal V}^*} \\
\edc{\prod G_{V^i_m}}{pt} & \rTTo^{\!\!\!\!\!\!\prod\phi^{i*}\!\!\!\!\!\!} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}} &
\rTTo^{\!\!\!\!\!\!\prod\pi_*^i\!\!\!\!\!\!}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal V}^i}} &
\rTTo^{\!\!\!\!\!\!\prod\alpha_{i!}\!\!\!\!\!\!} & \edc{L_{\mathcal V}}{pt}
\end{diagram}
\caption{First composition of associativity isomorphisms}
\label{fig-ll-assoc-start}
\end{figure}
The right-upper associativity composition is given on diagram in
\figref{fig-ru-assoc-start}. We have to check that isomorphism in this
figure equals to the one in \figref{fig-ll-assoc-start}.
\begin{figure}
\begin{diagram}[width=2.6em,height=2.400000em,inline]
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi_{\mathcal X}^*} &
\edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} & \rTTo^{\pi_{{\mathcal X}*}}_\sim &
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} & \rTTo^{\alpha_{{\mathcal X}!}} & \edc{P_{\mathcal W}}{pt} \\
& \rdTTo(2,4)_{\prod\phi^*_{{\mathcal Y}^i}} & \uTTo<\wr>{\pi'_*} &&&
\rdTTo^{\beta_!} & \uTTo>{\alpha_{{\mathcal V}!}} \\
&& \edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal V}^i}} && \uTTo<\wr>{\pi'''_*} && \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal V}} \\
\dTTo<{\prod\phi_m^{i*}} && \uTTo>{\pr_2^*} & \rdTTo_\sim^{\pi''_*} &&&
\uTTo<\wr>{\pi_{{\mathcal V}*}} \\
&& \edc{\prod G_{V^i}\times L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal Y}^i}} &&
\hspace*{-3mm} \edc{P_{\mathcal W}\times L_{\mathcal V}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} &
\rTTo^{\pr_{1!}} & \edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}} \\
&& \uTTo<\wr>{\pi'_*} & \rdTTo(2,4)_\sim^{\prod\pi_{{\mathcal Y}^i*}} &&& \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} & \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}\times
L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &&
\uTTo>{\pr_2^*} && \\
\dTTo<{\prod\pi^i_{m*}}>\wr && \dTTo<\wr>{\pi''_*} &&&&
\uTTo>{\phi_{\mathcal V}^*} \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pr_2^*} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod
G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pi'''_*}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}} && \\
\dTTo<{\prod\alpha^i_{m!}} && \dTTo>{\pr_{1!}} &&
\dTTo>{\prod\beta_{i!}} & \rdTTo^{\prod\alpha_{{\mathcal Y}^i!}} & \\
\edc{\prod G_{V^i_m}}{pt} & \rTTo^{\!\!\!\!\!\!\prod\phi^{i*}\!\!\!\!\!\!} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}} &
\rTTo^{\!\!\!\!\!\!\prod\pi_*^i\!\!\!\!\!\!}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal V}^i}} &
\rTTo^{\!\!\!\!\!\!\prod\alpha_{i!}\!\!\!\!\!\!} & \edc{L_{\mathcal V}}{pt}
\end{diagram}
\caption{Second composition of associativity isomorphisms}
\label{fig-ru-assoc-start}
\end{figure}
The subdiagram of diagram in \figref{fig-ru-assoc-start}
between the third and the forth columns is transformed via
\propref{prop-distr-base-chge} to a 3\n-column subdiagram.
The right part of it coincides with the right subdiagram in
\figref{fig-ll-assoc-start} between the third and the fourth columns
and can be canceled.
The left subdiagram of \figref{fig-ll-assoc-start} between the first
and the third columns is also transformed via
\propref{prop-distr-base-chge}.
We come to equation between the isomorphisms in Figures
\ref{fig-ll-assoc-1} and \ref{fig-ru-assoc-1}.
\begin{figure}
\rotatebox{90}{%
\hspace*{1mm}%
\begin{diagram}[width=4.0em,height=2.600000em,inline]
&&&&&& \\
&&&&&& \\
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi_{\mathcal X}^*} &
\edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} && \rTTo^{\pi_{{\mathcal X}*}}_\sim &&
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
\dTTo<{\prod\phi_m^{i*}} && \uTTo<\sim>{\pi'_*} &&&
\ruTTo_\sim^{\pi'''_*} \ruTTo(2,4)_{\pi_*}^\sim & \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} &
\edc{P_{\mathcal W}\times L_{\mathcal U}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal U}\times\prod
G_{V^i_m}/U_{{\mathcal V}^i_m}} & \rTTo^{\pi''_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &&
\dTTo>{\beta_{{\mathcal X}!}} \\
&& \dTTo<\wr>{\pi''_*} & \ruId & \uTTo<\wr>{\pi'_*} && \\
\dTTo<{\prod\pi^i_{m*}}>\wr &&
\edc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo_\sim^{\pi^{\prime*}} &
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &&
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal V}} \\
& \ruTTo^{\pr_2^*} && \ruTTo(4,2)_{\pr_3^*} & \dTTo>{\pr_{12!}} &
\ruTTo^{\pi_*}_\sim & \uTTo<\wr>{\pi'''_*} \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &&&&
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal V}^i}} & \rTTo^{\pi''_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal V}^i}} \\
\dTTo<{\prod\alpha^i_{m!}} &&& \ruTTo(4,2)^{\phi^*} & \uTTo>{\pr_2^*}
&& \uTTo>{\pr_2^*} \\
\edc{\prod G_{V^i_m}}{pt} && \rTTo^{\prod\phi^{i*}} &&
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}} &
\rTTo^{\!\!\!\!\!\!\prod\pi_*^i\!\!\!\!\!\!}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal V}^i}}
\end{diagram}
\hspace*{1mm}
}
\caption{First isomorphism}
\label{fig-ll-assoc-1}
\end{figure}
\begin{figure}
\rotatebox{90}{%
\hspace*{1mm}%
\begin{diagram}[width=4.0em,height=2.4500000em,inline]
&&&&&& \\
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi_{\mathcal X}^*} &
\edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} & \rTTo^{\pi_{{\mathcal X}*}}_\sim &
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} & \rTTo^{\beta_{{\mathcal X}!}} & \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal U}}
\\
& \rdTTo(2,4)_{\prod\phi^*_{{\mathcal Y}^i}} & \uTTo<\wr>{\pi'_*} &&
\uTTo<\wr>{\pi'''_*} && \uTTo<\wr>{\pi'''_*} \\
&& \edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal Y}^i}} & \rTTo^{\pi''_*} &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} &
\rTTo^{\beta_!} &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal V}^i}} \\
\dTTo<{\prod\phi_m^{i*}} && \uTTo>{\pr_2^*} &&&& \\
&& \edc{\prod G_{V^i}\times L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal Y}^i}} &&&& \\
&& \uTTo<\wr>{\pi'_*} & \rdTTo(2,4)_\sim^{\prod\pi_{{\mathcal Y}^i*}} &
\uTTo>{\pr_2^*} && \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} & \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}\times
L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &&
&& \uTTo>{\pr_2^*} \\
\dTTo<{\prod\pi^i_{m*}}>\wr && \dTTo<\wr>{\pi''_*} &&&& \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pr_2^*} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod
G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pi'''_*}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}} && \\
\dTTo<{\prod\alpha^i_{m!}} && \dTTo>{\pr_{1!}} &&&
\rdTTo>{\prod\beta_{i!}} & \\
\edc{\prod G_{V^i_m}}{pt} & \rTTo^{\!\!\!\!\!\!\prod\phi^{i*}\!\!\!\!\!\!} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}} &&
\rTTo^{\prod\pi_*^i}_\sim &&
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal V}^i}}
\end{diagram}
\hspace*{1mm}
}
\caption{Second isomorphism}
\label{fig-ru-assoc-1}
\end{figure}
\figref{fig-ll-assoc-1} can be transformed further using
\propref{prop-distr-base-chge} to the form of \figref{fig-ll-assoc-2}.
We have to prove that it equals to isomorphism in
\figref{fig-ru-assoc-1}.
\begin{figure}
\rotatebox{90}{%
\hspace*{1mm}%
\begin{diagram}[width=4.0em,height=2.600000em,inline]
&&&&&& \\
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi_{\mathcal X}^*} &
\edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} &&&& \\
\dTTo<{\prod\phi_m^{i*}} && \uTTo<\sim>{\pi'_*}
& \rdTTo(4,2)^{\pi_{{\mathcal X}*}}_\sim &&& \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
& \rTTo^{\pr_2^*} &
\edc{P_{\mathcal W}\times L_{\mathcal U}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal U}\times\prod
G_{V^i_m}/U_{{\mathcal V}^i_m}} & \rTTo^{\pi''_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}}
& \rTTo_\sim^{\pi'''_*} & \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
&& \dTTo<\wr>{\pi''_*} & \ruId & \uTTo<\wr>{\pi'_*}
& \ruTTo^{\pi_*}_\sim \ruTTo(2,4)_{\pi_*}^\sim & \\
\dTTo<{\prod\pi^i_{m*}}>\wr &&
\edc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo_\sim^{\pi^{\prime*}} &
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}}
&& \dTTo>{\beta_{{\mathcal X}!}} \\
& \ruTTo^{\pr_2^*} && \ruTTo(4,2)_{\pr_3^*} \ruTTo(2,4)^{\pr_{23}^*} &
\dTTo>{\pi_*} && \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &&&&
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} &&
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal U}} \\
& \rdTTo>{\pr_2^*} &&& \uTTo>{\pr_2^*} & \rdTTo(2,4)^{\beta_!} & \\
\dTTo<{\prod\alpha^i_{m!}} && \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo^{\pi'''_*}_\sim & \edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}} &&
\uTTo>{\pi'''_*}<\sim \\
&& \dTTo>{\pr_{1!}} && \dTTo>{\beta_!} && \\
\edc{\prod G_{V^i_m}}{pt} & \rTTo^{\prod\phi^{i*}} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}} &
\rTTo^{\!\!\!\!\!\!\prod\pi_*^i\!\!\!\!\!\!}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal V}^i}}
& \rTTo^{\pr_2^*\hspace*{1em}} &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal V}^i}}
\end{diagram}
\hspace*{1mm}
}
\caption{Modified first isomorphism}
\label{fig-ll-assoc-2}
\end{figure}
Two lower squares and two rightmost squares in diagrams in
\figref{fig-ll-assoc-2} and \figref{fig-ru-assoc-1} cancel and we have
to prove the following equation.
\begin{figure}
\begin{diagram}[width=4.0em,height=2.400000em]
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi_{\mathcal X}^*} &
\edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} & \rTTo^{\pi_{{\mathcal X}*}}_\sim &
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
\dTTo<{\prod\phi_m^{i*}} && \uTTo<\wr>{\pi'_*} &
\ruTTo(2,4)^{\pi''_*}_\sim \ruTTo(2,6)_{\pi_*}^\sim & \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} & \edc{P_{\mathcal W}\times L_{\mathcal U}\times
L_{\mathcal X}}{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} && \\
&& \dTTo<\wr>{\pi''_*} && \uTTo<\wr>{\pi'''_*} \\
\dTTo<{\prod\pi^i_{m*}}>\wr &&
\edc{P_{\mathcal W}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal U}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}}
&& \\
& \ruTTo^{\pr_2^*} & \uTTo<\wr>{\pi'_*} && \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pr_3^*} &
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo^{\pi_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} \\
& \rdTTo^{\pr_2^*} & \uTTo>{\pr_{23}^*} && \uTTo>{\pr_2^*} \\
&& \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo^{\pi'''_*}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}}
\end{diagram}
\caption{Part of the first isomorphism}
\label{fig-ll-assoc-3}
\end{figure}
We have to show that the isomorphism in \figref{fig-ll-assoc-3} is
equal to the isomorphism in \figref{fig-ru-assoc-3}.
\begin{figure}
\begin{diagram}[width=4.0em,height=2.400000em]
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi_{\mathcal X}^*} &
\edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} & \rTTo^{\pi_{{\mathcal X}*}}_\sim &
\edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
& \rdTTo(2,6)_{\phi^*} \rdTTo(2,4)_{\prod\phi^*_{{\mathcal Y}^i}} &
\uTTo<\wr>{\pi'_*} && \uTTo<\wr>{\pi'''_*} \\
&& \edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal Y}^i}} & \rTTo^{\pi''_*} &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} \\
\dTTo<{\prod\phi_m^{i*}} && \uTTo>{\pr_2^*} && \\
&& \edc{\prod G_{V^i}\times L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal Y}^i}} && \\
&& \uTTo<\wr>{\pi'_*} & \rdTTo(2,4)_\sim^{\prod\pi_{{\mathcal Y}^i*}} &
\uTTo>{\pr_2^*} \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} & \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}\times
L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &&
\\
\dTTo<{\prod\pi^i_{m*}}>\wr && \dTTo<\wr>{\pi''_*} &
\rdTTo_{\pi_*}^\sim & \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pr_2^*} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod
G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pi'''_*}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}}
\end{diagram}
\caption{Part of the second isomorphism}
\label{fig-ru-assoc-3}
\end{figure}
Using \propref{pro-iota-cocycle} we reduce both isomorphisms to the
expressions in \figref{fig-ll-assoc-end} and \figref{fig-ru-assoc-end}.
\begin{figure}
\begin{diagram}[width=4.0em,height=2.400000em]
&& \edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} && \\
& \ruTTo^{\phi_{\mathcal X}^*} & \uTTo<\wr>{\pi_*} & \rdTTo^{\pi_{{\mathcal X}*}}_\sim &
\\
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi^*} &
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}\times L_{\mathcal X}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
& \rTTo^{\pi_*}_\sim & \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
\dTTo<{\prod\phi_m^{i*}} & \ruTTo^{\pr_3^*} & \dTTo<\wr>{\pi_*} &
\ruTTo^{\pi_*}_\sim & \uTTo<\wr>{\pi'''_*} \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
&& \edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}{{\mathcal O}/U_{\mathcal V}\times\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo^{\pi_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} \\
\dTTo<{\prod\pi^i_{m*}}>\wr & \ruTTo^{\pr_3^*} & \uTTo>{\pr_{23}^*} &&
\uTTo>{\pr_2^*} \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pr_2^*} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo^{\pi'''_*}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}}
\end{diagram}
\caption{First simplified isomorphism}
\label{fig-ll-assoc-end}
\end{figure}
\begin{figure}
\begin{diagram}[width=4.0em,height=2.400000em]
&& \edc{P_{\mathcal W}\times L_{\mathcal X}}{{\mathcal O}/U_{\mathcal X}} && \\
& \ruTTo^{\phi_{\mathcal X}^*} & \uTTo<\wr>{\pi_*} & \rdTTo^{\pi_{{\mathcal X}*}}_\sim &
\\
\edc{\prod G_{V^i_{mp}}}{pt} & \rTTo^{\phi^*} &
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}\times L_{\mathcal X}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
& \rTTo^{\pi_*}_\sim & \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
\dTTo<{\prod\phi_m^{i*}} & \rdTTo^{\phi^*} & \uTTo>{\pr^*_{23}} &
\rdTTo^{\pi_*}_\sim & \uTTo<\wr>{\pi'''_*} \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
& \rTTo^{\pr_2^*} & \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}\times
L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &&
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} \\
\dTTo<{\prod\pi^i_{m*}}>\wr && \dTTo<\wr>{\pi''_*} &
\rdTTo^{\pi_*}_\sim & \uTTo>{\pr_2^*} \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} & \rTTo^{\pr_2^*} &
\edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &
\rTTo^{\pi'''_*}_\sim &
\edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}}
\end{diagram}
\caption{Second simplified isomorphism}
\label{fig-ru-assoc-end}
\end{figure}
The two upper triangles in \figref{fig-ll-assoc-end} and
\figref{fig-ru-assoc-end} coincide. Due to \propref{pro-iota-cocycle}
the lower parts are equal to isomorphisms in \figref{fig-assoc-final}.
\begin{figure}
\rotatebox{90}{%
\hspace*{2mm}%
\begin{diagram}[width=4.0em,height=2.600000em,inline]
&&&&&& \\
\edc{\prod G_{V^i_{mp}}}{pt} && \rTTo^{\phi^*} &&
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}\times L_{\mathcal X}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
&& \rTTo^{\pi_*}_\sim && \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
\dTTo<{\prod\phi_m^{i*}} &&& \ruTTo(4,2)^{\pr_3^*} \ruTTo^{\pr_{23}^*} &&
\rdTTo<\wr>{\pi_*} && \ruTTo^{\pi'_*}_\sim & \uTTo<\wr>{\pi'''_*} \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} & \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}\times
L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
\hspace*{-2cm} &&&& \hspace*{-2cm} \edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}}
& \rTTo^{\pi_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} \\
\dTTo<{\prod\pi^i_{m*}}>\wr &&& \rdTTo^{\pi''_*} && \ruTTo>{\pr_{23}^*}
&&& \uTTo>{\pr_2^*} \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} && \rTTo^{\pr_2^*}
&& \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &&
\rTTo^{\pi'''_*}_\sim && \edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}}
\hspace*{1mm} \\
\hspace*{1mm} \\
\edc{\prod G_{V^i_{mp}}}{pt} && \rTTo^{\phi^*} &&
\edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}\times L_{\mathcal X}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
&& \rTTo^{\pi_*}_\sim && \edc{P_{\mathcal W}}{{\mathcal O}/P_{\mathcal X}} \\
\dTTo<{\prod\phi_m^{i*}} & \rdTTo^{\pr_3^*} && \ruTTo^{\pr_{23}^*} &&
\rdTTo<\wr>{\pi_*} \rdTTo(4,2)^{\pi_*}_\sim &&& \uTTo<\wr>{\pi'''_*} \\
\edc{\prod G_{V^i_m}\times L_{{\mathcal V}^i_m}}{\prod G_{V^i_m}/U_{{\mathcal V}^i_m}} &
\rTTo^{\pr_2^*} & \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}\times
L_{{\mathcal Y}^i}}{\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/U_{{\mathcal V}^i_m}}
\hspace*{-2cm} &&&& \hspace*{-2cm} \edc{P_{\mathcal W}\times L_{\mathcal V}\times L_{\mathcal U}}
{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}}
& \rTTo^{\pi_*}_\sim &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/U_{\mathcal V}\times\prod G_{V^i}/P_{{\mathcal Y}^i}} \\
\dTTo<{\prod\pi^i_{m*}}>\wr &&& \rdTTo^{\pi''_*} && \ruTTo>{\pr_{23}^*}
&&& \uTTo>{\pr_2^*} \\
\edc{\prod G_{V^i_m}}{\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} && \rTTo^{\pr_2^*}
&& \edc{\prod G_{V^i}\times L_{{\mathcal V}^i}}{\prod
G_{V^i}/U_{{\mathcal V}^i}\times\prod G_{V^i_m}/P_{{\mathcal V}^i_m}} &&
\rTTo^{\pi'''_*}_\sim && \edc{\prod G_{V^i}}{\prod G_{V^i}/P_{{\mathcal Y}^i}}
\end{diagram}
\hspace*{5mm}
}
\caption{Final isomorphisms}
\label{fig-assoc-final}
\end{figure}
Finally, the last two isomorphisms are equal to each other due again to
\propref{pro-iota-cocycle}.
\end{proof}
\subsection{Comultiplication}
The comultiplication functor is
\begin{equation*}
\qquad
\vstretch 60
\begin{tangle}
\@ifnextchar[{\@step}{\@step[1]}[4]\idash\noderu{{\mathcal W}} \\
\nodeld{W^1}\sw4\sw0\nodeld{W^2}\sw2\sw0\s\se0\se2\se0\se4\noderd{W^l}
\end{tangle}
\quad = \pitchfork_{\mathcal W}^{\mathcal V} =
\Bigl( \edc{P_{\mathcal W}}{pt} \rTTo^{\Resto_{L_{\mathcal W},P_{\mathcal W}}} \edc{L_{\mathcal W}}{pt} \Bigr)
= i_{L_{\mathcal W},P_{\mathcal W}}^*.
\end{equation*}
Together with multiplication definition of \secref{sec-Multiplication}
it gives general operation (in the set-up of
\eqref{eq-decomp-V-to-Vi2}--\eqref{eq-decomp-V-to-Wj2})
\begin{multline*}
\text{\cit\Zh\/}_{\mathcal W}^{{\mathcal O};{\mathcal V}} = \qquad
\vstretch 80
\begin{tangle}
\nodelu{V^1}\Step\nodelu{V^2}\@ifnextchar[{\@step}{\@step[1]}[6]\noderu{V^k} \\
\nw4\nw0\nw2\nodel{{\mathcal O}}\nw0\n\ne0\ne2\ne0\ne4 \\
\nodeld{W^1}\sw4\sw0\nodeld{W^2}\sw2\sw0\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se0\se2\se0\se4\noderd{W^l}
\end{tangle}
\qquad = \Bigl( D_{L_{\mathcal V}}(pt) \rTTo^{\phi^*}
D_{P_{\mathcal W}\times L_{\mathcal V}}({\mathcal O}/{U_{\mathcal V}}) \\[2mm]
\rTTo^{\pi_*} D_{P_{\mathcal W}}({\mathcal O}/{P_{\mathcal V}}) \rTTo^{\alpha_!} D_{P_{\mathcal W}}(pt)
\rTTo^{\Resto_{L_{\mathcal W},P_{\mathcal W}}} D_{L_{\mathcal W}}(pt) \Bigr)
\end{multline*}
\[
\simeq \bigl( D_{L_{\mathcal V}}(pt) \rTTo^{\phi^*}
D_{L_{\mathcal W}\times L_{\mathcal V}}({\mathcal O}/{U_{\mathcal V}}) \rTTo^{\pi_*} D_{L_{\mathcal W}}({\mathcal O}/{P_{\mathcal V}})
\rTTo^{\alpha_!} D_{L_{\mathcal W}}(pt) \bigr).
\]
In the particular case $k=1$ we have ${\mathcal O}=L_{\mathcal V}=P_{\mathcal V}=G_V$, $U_{\mathcal V}=1$,
and the comultiplication operation is isomorphic to
\begin{multline*}
\vstretch 80
\begin{tanglec}
\n \\
\noded{W^1}\sw4\sw0\noded{W^2}\sw2\sw0\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se0\se2\se0\se4\noded{W^l}
\end{tanglec}
\simeq \bigl( D_{G_V}(pt) \rTTo^{\Resto_{P_{\mathcal W},G_V}} D_{P_{\mathcal W}}(pt)
\rTTo^{\Resto_{L_{\mathcal W},P_{\mathcal W}}} D_{L_{\mathcal W}}(pt) \bigr) \\
\simeq \bigl( D_{G_V}(pt) \rTTo^{\Resto_{L_{\mathcal W},G_V}} D_{L_{\mathcal W}}(pt) \bigr)
\end{multline*}
by \propref{pro-psi1=res}.
\subsection{The coassociativity isomorphism}
Assume that besides decomposition \eqref{eq-decomp-V-to-Wj2} of $V$
into direct sum of $W^j$ we have also decompositions
\[ {\mathcal W}^j:\qquad W^j = \oplus_{m=1}^{m^j} W_m^j .\]
We can produce out of these the decomposition
\[ {\mathcal U}:\qquad V = \oplus_{j,m} W_m^j ,\]
where the lexicographic order of summands is used
$W_1^1$, \dots,$W_{m^1}^1,W_1^2$, \dots,$W_{m^2}^2$,
\dots,$W_1^l$, \dots,$W_{m^l}^l$.
Coassociativity isomorphism
$coass:\pitchfork_{\mathcal U}^{\mathcal W}\circ\text{\cit\Zh\/}_{\mathcal W}^{\mathcal V}\to\text{\cit\Zh\/}_{\mathcal U}^{\mathcal V}$ between
functors
\[
\quad
\hstretch 90
\vstretch 80
\begin{tanglec}
\nw2\nw1\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\ne1\noderu{{\mathcal V}}\ne2 \\
\nodeld{{\mathcal W}^1\;}\sw3\Step\nodeld{{\mathcal W}^2\;}\s\Step\se3\noderd{{\mathcal W}^l} \\
\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1
\end{tanglec}
\;= \Bigl( \edc{L_{\mathcal V}}{pt} \rTTo^{\Psi^{\mathcal V}_{\mathcal W}} \edc{P_{\mathcal W}}{pt}
\rTTo^{i^*_{L_{\mathcal W},P_{\mathcal W}}} \edc{L_{\mathcal W}}{pt} \rTTo^{i^*_{L_{\mathcal U},L_{\mathcal W}}}
\edc{L_{\mathcal U}}{pt} \Bigr)
\]
\[
\text{and}\quad
\vstretch 100
\begin{tanglec}
\nw2\nw1\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\ne1\noderd{{\mathcal U}}\ne2 \\
\sw4\sw3\sw2\sw1\s\se1\se2\se3\se4
\end{tanglec}
\quad = \Bigl( D^{b,c}_{L_{\mathcal V}}(pt) \rTTo^{\Psi^{\mathcal V}_{\mathcal U}} D^{b,c}_{P_{\mathcal U}}(pt)
\rTTo^{i^*_{L_{\mathcal U},P_{\mathcal U}}} D^{b,c}_{L_{\mathcal U}}(pt) \Bigr)
\]
is given by the diagram
\begin{diagram}[height=2.7em,nobalance]
\edc{L_{\mathcal V}}{pt} & \rTTo^{\phi^*} & \edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}} &
\rTTo^{\pi_*} & \edc{P_{\mathcal W}}{{\mathcal O}/{P_{\mathcal V}}} & \rTTo^{\alpha_!} &
\edc{P_{\mathcal W}}{pt} & \rTTo^{i^*_{L_{\mathcal W},P_{\mathcal W}}} & \edc{L_{\mathcal W}}{pt} \\
\dId && \dTTo~{i^*_{P_{\mathcal U}\times L_{\mathcal V},P_{\mathcal W}\times L_{\mathcal V}}} &&
\dTTo~{i^*_{P_{\mathcal U},P_{\mathcal W}}} && \dTTo~{i^*_{P_{\mathcal U},P_{\mathcal W}}} &
\rdTTo^{i^*_{L_{\mathcal U},P_{\mathcal W}}} & \dTTo~{i^*_{L_{\mathcal U},L_{\mathcal W}}} \\
\edc{L_{\mathcal V}}{pt} & \rTTo^{\phi^*} & \edc{P_{\mathcal U}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}} &
\rTTo^{\pi_*} & \edc{P_{\mathcal U}}{{\mathcal O}/{P_{\mathcal V}}} & \rTTo^{\alpha_!} &
\edc{P_{\mathcal U}}{pt} & \rTTo^{i^*_{L_{\mathcal U},P_{\mathcal U}}} & \edc{L_{\mathcal U}}{pt}
\end{diagram}
\subsection{The coassociativity equation}
Assume that we have the following decompositions
of finite-dimensional $\CC$\n-vector spaces:
\begin{alignat*}4
{\mathcal V} &:&\quad V &= \oplus_i V^i, &\qquad
{\mathcal W} &:&\quad V &= \oplus_j W^j, \\
{\mathcal W}^j &:&\quad W^j &= \oplus_m W^j_m, &\qquad
{\mathcal W}^j_m &:&\quad W^j_m &= \oplus_p W^j_{m,p}.
\end{alignat*}
These decompositions imply also the following decompositions:
\begin{equation*}
{\mathcal U} :\; V = \oplus_{j,m} W^j_m, \quad\;
{\mathcal Y}^j :\; W^j = \oplus_{m,p} W^j_{m,p}, \quad\;
{\mathcal X} :\; V = \oplus_{j,m,p} W^j_{m,p}.
\end{equation*}
Let ${\mathcal O}\subset G_V$ be a left $P_{\mathcal W}$\n-invariant
and right $P_{\mathcal V}$\n-invariant subset.
Coassociativity isomorphisms give two isomorphisms
between the composite operation and the single operation
as in the diagram
\begin{diagram}[height=4.5em,LaTeXeqno]
\vstretch 75
\begin{tanglec}
\nw2\nodelu{{\mathcal V}}\nodeld{{\mathcal W}}\nw1\n\ne1\noder{{\mathcal O}}\ne2 \\
\nodeld{{\mathcal W}^1}\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}}\sw2\Step\se2\hbx(0,0){\c@@rds(0,0,\FillCircDiam,0){\circle*{\h@z}}} \\
\nodeld{{\mathcal W}^1_1}\sw1\se1\Step\sw1\se1 \\
\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\s\se1\@ifnextchar[{\@step}{\@step[1]}\s\se1\@ifnextchar[{\@step}{\@step[1]}\s\se1
\end{tanglec}
& \rTTo^{coass} &
\begin{tanglec}
\nw2\nodelu{{\mathcal V}}\nodeld{{\mathcal W}}\nw1\n\ne1\noder{{\mathcal O}}\ne2 \\
\nodeld{{\mathcal Y}^1}\sw2\Step\se2 \\
\sw2\sw1\s\se1\se2\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\se2
\end{tanglec}
\\
\dTTo<{coass} & = & \dTTo>{coass} & . \\
\begin{tanglec}
\nw2\nodelu{{\mathcal V}}\nodeld{{\mathcal U}}\nw1\n\ne1\noder{{\mathcal O}}\ne2 \\
\nodeld{{\mathcal W}^1_1}\sw3\sw0\sw1\se1\se0\se3 \\
\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\s\se1\@ifnextchar[{\@step}{\@step[1]}\s\se1\@ifnextchar[{\@step}{\@step[1]}\s\se1
\end{tanglec}
& \rTTo^{coass} &
\vstretch 150
\begin{tanglec}
\nw2\nodelu{{\mathcal V}}\nodeld{{\mathcal X}}\nw1\n\ne1\noder{{\mathcal O}}\ne2 \\
\sw4\sw3\sw2\sw1\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se1\se2\se3\se4
\end{tanglec}
\label{dia-coass-square}
\end{diagram}
\begin{proposition}[The coassociativity equation]
Diagram \eqref{dia-coass-square} is commutative.
\end{proposition}
\begin{proof}
The graphically expressed equation means equality
of the following two isomorphisms
\begin{diagram}[nobalance,width=2.5em]
\edc{L_{\mathcal V}}{pt} & \rTTo^{\phi^*} & \edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}} &
\rTTo^{\pi_*} & \edc{P_{\mathcal W}}{{\mathcal O}/{P_{\mathcal V}}} & \rTTo^{\alpha_!} &
\edc{P_{\mathcal W}}{pt} & \rTTo^{i^*_{L_{\mathcal W},P_{\mathcal W}}} & \edc{L_{\mathcal W}}{pt} \\
\dId && \dTTo>{i^*_{P_{\mathcal U}\times L_{\mathcal V},P_{\mathcal W}\times L_{\mathcal V}}} &&
\dTTo>{i^*_{P_{\mathcal U},P_{\mathcal W}}} && \dTTo>{i^*_{P_{\mathcal U},P_{\mathcal W}}} &
\rdTTo^{i^*_{L_{\mathcal U},P_{\mathcal W}}} & \dTTo>{i^*_{L_{\mathcal U},L_{\mathcal W}}} \\
\edc{L_{\mathcal V}}{pt} & \rTTo^{\phi^*} & \edc{P_{\mathcal U}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}} &
\rTTo^{\pi_*} & \edc{P_{\mathcal U}}{{\mathcal O}/{P_{\mathcal V}}} & \rTTo^{\alpha_!} &
\edc{P_{\mathcal U}}{pt} & \rTTo^{i^*_{L_{\mathcal U},P_{\mathcal U}}} & \edc{L_{\mathcal U}}{pt} \\
\dId && \dTTo>{i^*_{P_{\mathcal X}\times L_{\mathcal V},P_{\mathcal U}\times L_{\mathcal V}}} &&
\dTTo>{i^*_{P_{\mathcal X},P_{\mathcal U}}} && \dTTo>{i^*_{P_{\mathcal X},P_{\mathcal U}}} &
\rdTTo^{i^*_{L_{\mathcal X},P_{\mathcal U}}} & \dTTo>{i^*_{L_{\mathcal X},L_{\mathcal U}}} \\
\edc{L_{\mathcal V}}{pt} & \rTTo^{\phi^*} & \edc{P_{\mathcal X}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}} &
\rTTo^{\pi_*} & \edc{P_{\mathcal X}}{{\mathcal O}/{P_{\mathcal V}}} & \rTTo^{\alpha_!} &
\edc{P_{\mathcal X}}{pt} & \rTTo^{i^*_{L_{\mathcal X},P_{\mathcal X}}} & \edc{L_{\mathcal X}}{pt}
\end{diagram}
\begin{diagram}[nobalance,width=3.0em,tight]
\edc{L_{\mathcal V}}{pt} & \rTTo^{\phi^*\hspace*{-1mm}} &
\edc{P_{\mathcal W}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}} & \rTTo^{\hspace*{-1mm}\pi_*} &
\edc{P_{\mathcal W}}{{\mathcal O}/{P_{\mathcal V}}} & \rTTo^{\alpha_!} &
\edc{P_{\mathcal W}}{pt} & \rTTo^{i^*_{L_{\mathcal W},P_{\mathcal W}}} & \edc{L_{\mathcal W}}{pt} \\
&&& &&& &&& \rdTTo>{i^*_{L_{\mathcal U},L_{\mathcal W}}} & \\
\dId &&
\dTTo~{\hspace*{-1mm}i^*_{P_{\mathcal X}\times L_{\mathcal V},P_{\mathcal W}\times L_{\mathcal V}}\hspace*{-1mm}}
&& \dTTo~{i^*_{P_{\mathcal X},P_{\mathcal W}}} && \dTTo~{i^*_{P_{\mathcal X},P_{\mathcal W}}} &&
\dTTo~{i^*_{L_{\mathcal X},L_{\mathcal W}}} && \edc{L_{\mathcal U}}{pt} \\
&&& &&& &&& \ldTTo>{i^*_{L_{\mathcal X},L_{\mathcal U}}} & \\
\edc{L_{\mathcal V}}{pt} & \rTTo^{\phi^*\hspace*{-1mm}} &
\edc{P_{\mathcal X}\times L_{\mathcal V}}{{\mathcal O}/{U_{\mathcal V}}} & \rTTo^{\hspace*{-1mm}\pi_*} &
\edc{P_{\mathcal X}}{{\mathcal O}/{P_{\mathcal V}}} & \rTTo^{\alpha_!} &
\edc{P_{\mathcal X}}{pt} & \rTTo^{i^*_{L_{\mathcal X},P_{\mathcal X}}} & \edc{L_{\mathcal X}}{pt}
\end{diagram}
Four triangular prisms, which follow from \propref{pro-iota-cocycle}
and \propref{prop-distr-base-chge},
imply that the above isomorphisms are equal.
\end{proof}
\subsection{The coherence isomorphism}
Assume given a vector space decomposition
\[ V = \oplus_{m=1}^k \oplus_{r=1}^l V_r^m .\]
We shall denote it ${\mathcal Y}$ if the summands are ordered as follows
\[ V_1^1,V_2^1,\dots,V^1_l,V^2_1,V^2_2,\dots,V^2_l,\dots,
V_1^k,V_2^k,\dots,V_l^k .\]
The same decomposition with order of summands
\[ V_1^1,V_1^2,\dots,V_1^k,V_2^1,V_2^2,\dots,V_2^k,\dots,
V_l^1,V_l^2,\dots,V_l^k \]
will be denoted ${\mathcal X}$. We also have decompositions
\begin{alignat*}4
{\mathcal V}^m &:&\quad V^m &= \oplus_{r=1}^l V^m_r &\qquad
{\mathcal V} &:&\quad V &= \oplus_{m=1}^k V^m \\
{\mathcal W}_r &:&\quad W_r &= \oplus_{m=1}^k V_r^m &\qquad
{\mathcal W} &:&\quad V &= \oplus_{r=1}^l W_r
\end{alignat*}
which produce filtrations $V^m_{(r)}$ of $V^m$, $V^{(m)}$ of $V$,
$V_r^{(m)}$ of $W_r$ and $W_{(r)}$ of $V$. We associate with these
filtrations parabolic groups $P_{{\mathcal V}^m}\subset G_{V^m}$,
$P_{\mathcal V}\subset G_V$, $P_{{\mathcal W}_r}\subset G_{W_r}$ and $P_{\mathcal W}\subset G_{\mathcal W}$.
\begin{notations}
Let $A$, $B$ be vector spaces. We denote the braiding functor as
\begin{equation*}
\begin{tanglec}
\xx
\end{tanglec}
= \bigl( D_{G_{A}\times G_{B}}(pt) \xra\tau D_{G_{A}\times G_{B}}(pt)
\xra{\sigma^*} D_{G_{B}\times G_{A}}(pt) \bigr),
\end{equation*}
where $\sigma$ is the permutation isomorphism of groups and modules and
the functor $\tau$ is the shift \( \tau(L) = L[-2\dim A\dim B] \).
\end{notations}
We claim that for any collection of indices and
for any collection of bi-invariant locally closed subsets
\(({\mathcal O}'_1,\dots,{\mathcal O}'_k,{\mathcal O}''_1,\dots,{\mathcal O}''_l)\),
which may occur in the following diagram,
there exists a bi-invariant locally closed subset ${\mathcal O}$
and a coherence isomorphism
\[
\vstretch 78
\begin{tanglec}
\nw1\nodel{{\mathcal O}'_1\;}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\nodel{{\mathcal O}'_k\;}\n\ne1 \\
\sw1\s\se1\nodeu{\;l}\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\nodeu{\;l}\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\nodeu{\;l} \\
\ffbox{9}{\sigma_{k,l}} \\
\nw1\nodel{{\mathcal O}''_1\;}\n\noderu{k}\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\noderu{k}\ne1\@ifnextchar[{\@step}{\@step[1]}%
\nw1\nodel{{\mathcal O}''_l\;}\n\noderu{k}\ne1 \\
\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1
\end{tanglec}
\quad\equiv\quad
\vstretch 60
\begin{tangle}
\nodel{{\mathcal Y}} \\
\nodel{{\mathcal V}}\nw1\noder{{\mathcal O}'_1}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\noder{{\mathcal O}'_k}\n\ne1 \\
\nodel{{\mathcal U}}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1 \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \nodel{{\mathcal X}}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\nodel{{\mathcal W}}\nw1\noder{{\mathcal O}''_1}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\noder{{\mathcal O}''_l}\n\ne1 \\
\nodel{{\mathcal Z}}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1
\end{tangle}
\quad \rTTo^{\Coher} \;
\vstretch 195
\begin{tangle}
\nw4\nw3\nw2\nodel{{\mathcal O}}\nw1\n\ne1\ne2\ne3\ne4 \\
\sw4\sw3\sw2\sw1\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se1\se2\se3\se4
\end{tangle}
,\]
explicitly described below.
Here $\sigma_{k,l}=(s_{k,l})_+\sptilde$ is the braid, corresponding
to the permutation $s_{k,l}$ of the set $\{1,2,\dots,kl\}$,
\begin{equation}
s_{k,l}(1+t+nl)=1+n+tk \text{ \ for \ } 0\le t<l, 0\le n<k,
\label{eq-s-permute}
\end{equation}
under the standard splitting $S_{kl}\to B_{kl}$, which maps
the elementary transpositions to the generators of the braid group.
The subset ${\mathcal O}$ is computed as follows.
Bi-invariance of initial parametrising subsets means that
\begin{align*}
G_{V^m} \supset {\mathcal O}'_m &\in P_{V_-^m}\sets P_{{\mathcal Y}^m}, \\
G_{W_r} \supset {\mathcal O}''_r &\in P_{{\mathcal Z}_r}\sets P_{V_r^-},
\end{align*}
where ${\mathcal Y}^m$ is the incoming filtration of $V^m$, and ${\mathcal Z}_r$ is the
outgoing filtration of $W_r$. The subsets
\begin{align*}
P_{{\mathcal V}}\supset \overline{{\mathcal O}} &\overset{\text{def}}=
U_{\mathcal V} \cdot \prod_m{\mathcal O}'_m = \prod_m{\mathcal O}'_m \cdot U_{\mathcal V}
\in P_{V_-^=}\sets P_{{\mathcal Y}}, \\
P_{{\mathcal V}}\supset \underline{{\mathcal O}} &\overset{\text{def}}=
U_{\mathcal W} \cdot \prod_r{\mathcal O}''_r = \prod_r{\mathcal O}''_r \cdot U_{\mathcal W}
\in P_{{\mathcal Z}}\sets P_{V_=^-}
\end{align*}
are locally closed. The subset
\begin{gather*}
G_V\supset {\mathcal O} \overset{\text{def}}= \und{\mathcal O}\cdot\overline{\mathcal O}
\in P_{{\mathcal Z}}\sets P_{{\mathcal Y}}, \quad
{\mathcal O} = \und{\mathcal O} P_{\mathcal U}\times_{P_{\mathcal U}}\overline{\mathcal O} =
\und{\mathcal O}\times_{P_{\mathcal W}\cap P_{\mathcal V}}\overline{\mathcal O},
\end{gather*}
is also locally closed. Indeed, $\und{\mathcal O}\times\overline{\mathcal O}$ is locally
closed in $P_{\mathcal W}\times P_{\mathcal V}$, hence,
$\und{\mathcal O}\times_{P_{\mathcal W}\cap P_{\mathcal V}}\overline{\mathcal O}$ is locally closed in
$P_{\mathcal W}\times_{P_{\mathcal W}\cap P_{\mathcal V}}P_{\mathcal V}\simeq P_{\mathcal W}\cdot P_{\mathcal V}$.
The subspace $C=P_{\mathcal W}\cdot P_{\mathcal V}$ of $G_V$ is locally closed, since it
is an orbit of $P_{\mathcal W}\times P_{\mathcal V}$ in $G_V$, and the number of such
orbits is finite. Indeed, $C$ is embedded in
its closure as
\[ C = \overline{C} - \bigcup\{\overline{{\mathcal O}_a} \mid
{\mathcal O}_a\subset\overline{C} \text{ is an orbit of } P_{\mathcal W}\times P_{\mathcal V},
\; {\mathcal O}_a\neq C \} \]
due to dimension considerations.
Both associativity isomorphism and coassociativity isomorphism
are particular cases of the general coherence isomorphism.
In the particular case, when the upper row of operations consists of
comultiplications (operations with single input), and the lower row of
operations consists of multiplications (operations with single output),
we have necessarily ${\mathcal O}'_m=G_{V^m}$ and ${\mathcal O}''_r=G_{W_r}$. In such
cases we omit the parametrising set, since it is unique. Graphical
notation here is the following:
\[
\vstretch 100
\hstretch 200
\begin{tanglec}
\nodeld{V^1}\@ifnextchar[{\@step}{\@step[1]}[3]\nodeld{V^2}\@ifnextchar[{\@step}{\@step[1]}[3]\noderd{V^k} \\
\nodeld{V_1^1}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\noderd{V_l^1}\@ifnextchar[{\@step}{\@step[1]}[3]\nodeld{V_1^2}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}%
\noderd{V_l^2}\@ifnextchar[{\@step}{\@step[1]}[3]\nodeld{V_1^k}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\noderd{V_l^k} \\
\sw1\nodeu{V_2^1\ \ }\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1%
\nodeu{V^2_2\ \ }\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\nodeu{V_2^k\ \ }\s\se1 \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\noded{V_1^2\ \ }\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}%
\noded{V_2^2\ \ }\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\noded{V_l^2\ \ }\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\nw1\nodelu{V_1^1}\n\noderu{V_1^k}\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\nodelu{V_2^1}\n%
\noderu{V_2^k}\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\nodelu{V_l^1}\n\noderu{V_l^k}\ne1 \\
\nodelu{W_1}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}[3]\nodelu{W_2}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}[3]\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\noderu{W_l}
\end{tanglec}
\quad \rTTo^\Coher \qquad
\vstretch 267
\hstretch 120
\begin{tangle}
\nodelu{V^1}\Step\nodelu{V^2}\@ifnextchar[{\@step}{\@step[1]}[6]\noderu{V^k} \\
\nw4\nw0\nw2\nodel{{\mathcal O}}\nw0\n\ne0\ne2\ne0\ne4 \\
\nodeld{W_1}\sw4\sw0\nodeld{W_2}\sw2\sw0\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se0\se2\se0\se4\noderd{W_l}
\end{tangle}
\quad. \]
By general formulas we find here $\overline{{\mathcal O}}=P_{\mathcal V}$,
$\underline{{\mathcal O}}=P_{\mathcal W}$, hence, ${\mathcal O}=P_{\mathcal V} P_{\mathcal W}$.
The general coherence isomorphism is built as the composition
\begin{multline*}
\Coher:\quad
\hstretch 120
\vstretch 68
\begin{tangle}
\nodel{{\mathcal Y}} \\
\nodel{{\mathcal V}}\nw1\noder{{\mathcal O}'_1}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\nodel{{\mathcal O}'_k}\n\ne1 \\
\nodel{{\mathcal U}}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1 \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \nodel{{\mathcal X}}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\nodel{{\mathcal W}}\nw1\noder{{\mathcal O}''_1}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\nodel{{\mathcal O}''_l}\n\ne1 \\
\nodel{{\mathcal Z}}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1
\end{tangle}
\quad=\quad
\vstretch 52
\begin{tangle}
\nw1\nodel{{\mathcal O}'_1}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\noder{{\mathcal O}'_k}\n\ne1 \\
\@ifnextchar[{\@step}{\@step[1]}\idash\noderu{{\mathcal V}^1}\@ifnextchar[{\@step}{\@step[1]}[3]\idash\@ifnextchar[{\@step}{\@step[1]}[3]\idash\nodelu{{\mathcal V}^k} \\
\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1 \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\def\hm@dehalf{.5}\def\hm@de{1 \hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\xx\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\@ifnextchar[{\@step}{\@step[1]}\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}} \\
\nw1\nodel{{\mathcal O}''_1\;}\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\noder{\,{\mathcal O}''_l}\n\ne1 \\
\@ifnextchar[{\@step}{\@step[1]}\idash\noderu{{\mathcal Z}_1}\@ifnextchar[{\@step}{\@step[1]}[3]\idash\@ifnextchar[{\@step}{\@step[1]}[3]\idash\nodelu{{\mathcal Z}_l} \\
\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1
\end{tangle}
\quad \rTTo^{\coher} \\[5mm]
\hstretch 120
\vstretch 88
\begin{tangle}
\nw1\nodel{{\mathcal O}'_1}\n\noded{{\mathcal V}^1}\ne1\@ifnextchar[{\@step}{\@step[1]}\nw1\n\ne1%
\@ifnextchar[{\@step}{\@step[1]}\nw1\noder{{\mathcal O}'_k}\n\noded{{\mathcal V}^k}\ne1 \\
\@ifnextchar[{\@step}{\@step[1]}\nwd3\@ifnextchar[{\@step}{\@step[1]}\nodel{\und{\mathcal O} P_{\mathcal U}}\@ifnextchar[{\@step}{\@step[1]}\nd\Step\ned3 \\
\@ifnextchar[{\@step}{\@step[1]}[4]\idash\noderu{{\mathcal Z}} \\
\@ifnextchar[{\@step}{\@step[1]}\nodeu{{\mathcal Z}_1}\swd3\Step\sd\Step\sed3\nodeu{{\mathcal Z}_l} \\
\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1\@ifnextchar[{\@step}{\@step[1]}\sw1\s\se1
\end{tangle}
\rTTo^{\assoc}_{\coass}
\hstretch 100
\vstretch 147
\begin{tangle}
\nw4\nw3\nw2\nodel{{\mathcal O}}\nw1\n\ne1\ne2\ne3\ne4 \\
\@ifnextchar[{\@step}{\@step[1]}[4]\idash\noderu{{\mathcal Z}} \\
\sw4\sw3\sw2\sw1\s\se1\se2\se3\se4
\end{tangle}
=
\vstretch 221
\begin{tangle}
\nw4\nw3\nw2\nodel{{\mathcal O}}\nw1\n\ne1\ne2\ne3\ne4 \\
\sw4\sw3\sw2\sw1\hbx(0,\hm@de){\c@@rds(0,0,0,\hm@detens){\line(0,1){\v@t}}}\se1\se2\se3\se4
\end{tangle}
.
\end{multline*}
The three components of the coherence isomorphism are defined next.
\begin{diagram}[height=2.6em,nobalance]
\Dec{pt}{\prod G_{Y_s^m}} & \rTTo^{\phi^*} &
\Dec{\prod {\mathcal O}'_m}{\prod P_{{\mathcal V}^m}\times P_{{\mathcal Y}^m}}
& \rTTo^{\pi_*} & \Dec{\prod {\mathcal O}'_m/{P_{{\mathcal Y}^m}}}{\prod P_{{\mathcal V}^m}}
& \rTTo^{\alpha_!} & \Dec{pt}{\prod P_{{\mathcal V}^m}} \\
\dTTo<{\phi^*} && \dTTo>{\phi^*} && \dTTo>{\phi^*} && \dTTo>{\phi^*} \\
\Dec{{\mathcal O}}{P_{\mathcal Z}\times P_{\mathcal Y}} & \rTTo^{\pi^*} &
\Dec{\und{\mathcal O} P_{\mathcal U}\times\overline{{\mathcal O}}}{P_{\mathcal Z}\times P_{\mathcal U}\times P_{\mathcal Y}}
& \rTTo^{\pi_*} &
\Dec{\und{\mathcal O} P_{\mathcal U}\times\overline{{\mathcal O}}/{P_{\mathcal Y}}}{P_{\mathcal Z}\times P_{\mathcal U}}
& \rTTo^{1\times\alpha_!} & \Dec{\und{\mathcal O} P_{\mathcal U}}{P_{\mathcal Z}\times P_{\mathcal U}} \\
& \rdId<{\Id} & \dTTo>{\pi_*} && \dTTo>{\pi_*} && \dTTo>{\pi_*} \\
&& \Dec{{\mathcal O}}{P_{\mathcal Z}\times P_{\mathcal Y}} & \rTTo^{\pi_*}
& \Dec{{\mathcal O}/{P_{\mathcal Y}}}{P_{\mathcal Z}} & \rTTo^{\beta_!}
& \Dec{\und{\mathcal O} P_{\mathcal U}/{P_{\mathcal U}}}{P_{\mathcal Z}} \\
&& \assoc &&& \rdTTo_{\alpha_!} & \dTTo>{\alpha_!} \\
&&&&&& \Dec{pt}{P_{\mathcal Z}}
\end{diagram}
\[ \coass: \Bigl(\Dec{pt}{P_{\mathcal Z}} \rTTo^{\iota^*}
\Dec{pt}{\prod_rP_{{\mathcal Z}_r}} \rTTo^{\iota^*}
\Dec{pt}{\prod_{n,r}G_{Z^n_r}}\Bigr)
\rTTo^\sim \Bigl(\Dec{pt}{P_{\mathcal Z}} \rTTo^{\iota^*}
\Dec{pt}{\prod_{n,r}G_{Z^n_r}}\Bigr). \]
\begin{figure}
\resizebox{\textwids}{!}
\begin{diagram}[height=2.5em,width=2.5em,inline,nobalance]
\dec{pt}{\prod_mP_{{\mathcal V}^m}} &&& \rTTo^{\iota^*}
&&& \dec{pt}{\prod_{m,r}G_{V^m_r}} \\
& \rdId^{\Id} & (1) &&& \ldTTo(4,2)^{p^*} \ldTTo>{p^*}
& \dTTo>{\phi^*} \\
\dTTo<{\phi^*} && \dec{pt}{\prod_mP_{{\mathcal V}^m}}
&& \dec{\und{\mathcal O}/{U_{\mathcal W}\cap P_{\mathcal V}}}{\prod_rP_{{\mathcal Z}_r}\times P_{{\mathcal W}_r}}
&& \dec{\prod_r{\mathcal O}_r''}{\prod_rP_{{\mathcal Z}_r}\times P_{{\mathcal W}_r}} \\
&& \dTTo^{p^*} & \ldTTo^{q^*} & \dId<{\ttt(2)\hspace*{0.7em}}>{\Id}
& \ldTTo^{\rho^*} & \dTTo<{\ttt(3)\hspace*{0.5em}}>{T^{-2A}} \\
\dec{\und{\mathcal O} P_{\mathcal U}}{P_{\mathcal Z}\times P_{\mathcal U}} & \rTTo^{i^*} &
\dec{\und{\mathcal O}}{(\prod P_{{\mathcal Z}_r})\times(P_{\mathcal W}\cap P_{\mathcal V})}
& \rTTo^{q_*} &
\dec{\und{\mathcal O}/{U_{\mathcal W}\cap P_{\mathcal V}}}{\prod_rP_{{\mathcal Z}_r}\times P_{{\mathcal W}_r}}
& \rTTo^{\rho_!} &
\dec{\prod_r{\mathcal O}_r''}{\prod_rP_{{\mathcal Z}_r}\times P_{{\mathcal W}_r}} \\
\dTTo<{\pi_*} & (4) & \dTTo<{q_*} & \ldTTo>{q_*} & (5)
&& \dTTo>{\pi_*} \\
\dec{\und{\mathcal O} P_{\mathcal U}/{P_{\mathcal U}}}{P_{\mathcal Z}} & \rTTo^{\iota^*}
& \dec{\und{\mathcal O}/{P_{\mathcal W}\cap P_{\mathcal V}}}{\prod P_{{\mathcal Z}_r}} && \rTTo^{\rho_!}
&& \dec{\prod_r{\mathcal O}_r''/{P_{{\mathcal W}_r}}}{\prod_rP_{{\mathcal Z}_r}} \\
\dTTo<{\alpha_!} && (6) & \rdTTo(4,2)^{\alpha_!} &&& \dTTo>{\alpha_!} \\
\dec{pt}{P_{\mathcal Z}} &&& \rTTo^{\iota^*} &&& \dec{pt}{\prod_rP_{{\mathcal Z}_r}}
\end{diagram}%
}
\caption{Coherence isomorphism $\coher$}
\label{dia-fig-coher-1-6}
\end{figure}
Coherence isomorphism in \figref{dia-fig-coher-1-6} is composed of
several isomorphisms. Isomorphism (1) exists by \propref{pro-pi-p-i}.
Isomorphism (2) follows from the fact that $q_*$ is an equivalence.
Isomorphism $T^{-2A}\to\rho_!\rho^*$ marked by (3) is obtained from
a sequence of affine linear $\prod P_{{\mathcal W}_r}$-bundles
\begin{multline*}
U_{\mathcal W}/(U_{\mathcal W}\cap P_{\mathcal V}) \rEpi \dots \rEpi
U_{\mathcal W}/[U_{\mathcal W}^{(k+1)}(U_{\mathcal W}\cap P_{\mathcal V})] \\
\rEpi U_{\mathcal W}/[U_{\mathcal W}^{(k)}(U_{\mathcal W}\cap P_{\mathcal V})] \rEpi \dots \rEpi pt,
\end{multline*}
where $U=U^{(1)}\supset U^{(2)}\supset\dots\supset U^{(m)}=1$ is
the lower central series of $U$. The group $\prod P_{{\mathcal W}_r}$
acts by conjugation. Pull-back of this sequence along
$\prod{\mathcal O}''_r\times F_{{\mathcal W}_r}|\prod P_{{\mathcal Z}_r}\times P_{{\mathcal W}_r}\to
pt|\prod P_{{\mathcal W}_r}$
gives a sequence of affine linear
$\prod P_{{\mathcal Z}_r}\times P_{{\mathcal W}_r}$-bundles
\begin{multline*}
\prod{\mathcal O}''_r\times U_{\mathcal W}/(U_{\mathcal W}\cap P_{\mathcal V}) \rEpi \dots \rEpi
\prod{\mathcal O}''_r\times U_{\mathcal W}/[U_{\mathcal W}^{(k+1)}(U_{\mathcal W}\cap P_{\mathcal V})] \\
\rEpi \prod{\mathcal O}''_r\times U_{\mathcal W}/[U_{\mathcal W}^{(k)}(U_{\mathcal W}\cap P_{\mathcal V})]
\rEpi \dots \rEpi \prod{\mathcal O}''_r.
\end{multline*}
Multiplication map $\prod{\mathcal O}''_r\times U_{\mathcal W}\to\underline{{\mathcal O}}$ is an
isomorphism of $\prod P_{{\mathcal Z}_r}\times P_{{\mathcal W}_r}$-spaces, hence, we may
replace the first space with the other one. The composition of the above
maps equals $\rho:{\mathcal O}/(U_{\mathcal W}\cap P_{\mathcal V})\to\prod_r{\mathcal O}''_r$. The type (3)
isomorphism for $\rho$ is a composition of isomorphisms, constructed for
each affine bundle of the sequence in \corref{cor-aff-bun-equi}.
Isomorphism (4) is obtained from a commutative square of equivariant
maps. Isomorphism (6) is a base change isomorphism. Isomorphism (5) is
obtained from a base change isomorphism by inverting $q^*$ and $\pi^*$
and replacing them with their quasi-inverses $q_*$ and $\pi_*$.
We use the facts
\begin{gather*}
U_{\mathcal W} = 1 + \oplus_{m,n;r>s} \Hom(V_r^m,V_s^n) ,\\
U_{\mathcal W}\cap P_{\mathcal V} = 1 + \oplus_{m\ge n;r>s} \Hom(V_r^m,V_s^n) ,\\
U_{\mathcal W}/(U_{\mathcal W}\cap P_{\mathcal V}) \simeq \oplus_{m<n;r>s} \Hom(V_r^m,V_s^n),
\end{gather*}
The dimension of fiber bundle $\rho$
\begin{equation*}
A \overset{\text{def}}= \dim_\CC U_{\mathcal W}/(U_{\mathcal W}\cap P_{\mathcal V})
= \sum_{m<n;r>s} \dim V_r^m \cdot \dim V_s^n
\end{equation*}
multiplied by $-2$ equals to the total shift, obtained from all
braidings in the source functor of coherence isomorphism $\Coher$.
\subsection{Distinguished triangles}
Consider left $P_{\mathcal W}$\n-invariant and right $P_{\mathcal V}$\n-invariant subsets
$F\subset Y\subset G_V$, such that $F$ is closed in $Y$.
Denote $S=Y-F$ and consider inclusions
\begin{gather*}
F/{U_{\mathcal V}} \rTTo^{i_U} Y/{U_{\mathcal V}} \lTTo^{j_U} S/{U_{\mathcal V}} ,\\
F/{P_{\mathcal V}} \rTTo^{i_P} Y/{P_{\mathcal V}} \lTTo^{j_P} S/{P_{\mathcal V}}.
\end{gather*}
Denote $\phi^X$, $\pi^X$, $\alpha^X$ maps~\eqref{eq-p1p2p32} for
${\mathcal O}=X\in\{Y,F,S\}$. Let $K\in D_{L_{\mathcal V}}(pt)$,
$K_U=\phi^{Y*}K\in D_{P_{\mathcal W}\times L_{\mathcal V}}(Y/{U_{\mathcal V}})$,
$K_P=\pi_{*}^YK_U\in D_{P_{\mathcal W}}(Y/{P_{\mathcal V}})$
(see the middle row in the following diagram).
\begin{diagram}
D_{L_{\mathcal V}}(pt) & \rTTo^{\phi^{S*}} &
D_{P_{\mathcal W}\times L_{\mathcal V}}(S /{U_{\mathcal V}}) &
\rTTo^{\pi_{*}^S}_\sim & D_{P_{\mathcal W}}(S /{P_{\mathcal V}}) &
\rTTo^{\alpha_{!}^S} & D_{P_{\mathcal W}}(pt) \\
\dEq && \uTTo<{j_U^*} && \uTTo<{j_P^*} \dTTo>{j_{P!}} && \dEq \\
D_{L_{\mathcal V}}(pt) & \rTTo^{\phi^{Y*}} &
D_{P_{\mathcal W}\times L_{\mathcal V}}(Y /{U_{\mathcal V}}) &
\rTTo^{\pi_{*}^Y}_\sim & D_{P_{\mathcal W}}(Y /{P_{\mathcal V}}) &
\rTTo^{\alpha_{!}^Y} & D_{P_{\mathcal W}}(pt) \\
\dEq && \dTTo<{i_U^*} && \dTTo<{i_P^*} \uTTo>{i_{P!}} && \dEq \\
D_{L_{\mathcal V}}(pt) & \rTTo^{\phi^{F*}} &
D_{P_{\mathcal W}\times L_{\mathcal V}}(F /{U_{\mathcal V}}) & \rTTo^{\pi_{*}^F}_\sim &
D_{P_{\mathcal W}}(F /{P_{\mathcal V}}) & \rTTo^{\alpha_{!}^F} & D_{P_{\mathcal W}}(pt)
\end{diagram}
Applying $\alpha_{!}^Y$ to a standard triangle for $K_P$ we get a
distinguished triangle
\[ \alpha_{!}^Yj_{P!}j_P^*K_P \to \alpha_{!}^YK_P
\to \alpha_{!}^Yi_{P!}i_P^*K_P \to \]
The above diagram shows that it is isomorphic to certain following
sequences:
\[ \alpha_{!}^Sp_{2*}^Sj_U^*K_U \to \alpha_{!}^Y\pi_{*}^YK_U \to
\alpha_{!}^F\pi_{*}^Fi_U^*K_U \to \]
\[ \alpha_{!}^S\pi_{*}^S\phi^{S*}K \to \alpha_{!}^Y\pi_{*}^Y\phi^{Y*}K \to
\alpha_{!}^F\pi_{*}^F\phi^{F*}K \to \]
This is the triangle
\[ \Psi_{\mathcal W}^{S;{\mathcal V}}K \to \Psi_{\mathcal W}^{Y;{\mathcal V}}K \to
\Psi_{\mathcal W}^{F;{\mathcal V}}K \to \]
from which the structure triangle
\begin{equation}
\text{\cit\Zh\/}_{\mathcal W}^{S;{\mathcal V}}K \rTTo^{a^{SY}} \text{\cit\Zh\/}_{\mathcal W}^{Y;{\mathcal V}}K \rTTo^{b^{YF}}
\text{\cit\Zh\/}_{\mathcal W}^{F;{\mathcal V}}K \rTTo^{c^{FS}}
\label{eq-trian-abc}
\end{equation}
is obtained by application of the exact functor $\pitchfork_{\mathcal W}$.
This triangle replaces non-existing isomorphism
\(\text{\cit\Zh\/}_{\mathcal W}^{Y;{\mathcal V}}K \simeq \text{\cit\Zh\/}_{\mathcal W}^{S;{\mathcal V}}K \oplus
\text{\cit\Zh\/}_{\mathcal W}^{F;{\mathcal V}}K \)
(this would be too much to ask for).
\begin{proposition}
For any pair of closed embeddings $F\subset Z\subset Y$ set
$S=Y-F$, $Q=Z-F$, $R=Y-Z$. The following diagram made with
distinguished triangles \eqref{eq-trian-abc} is an octahedron.
\begin{equation*}
\begin{diagram}[inline,width=2.8em]
\text{\cit\Zh\/}^{R} && \rTTo^{a^{RY}} && \text{\cit\Zh\/}^{Y} \\
& \luTTo^1_{c^{ZR}} & d & \ldTTo_{b^{WZ}} & \\
\uTTo<{c^{QR}}>1 & = & \text{\cit\Zh\/}^{Z} & = & \dTTo>{b^{YF}} \\
& \ruTTo^{a^{QZ}} & d & \rdTTo^{b^{ZF}} & \\
\text{\cit\Zh\/}^{Q} && \lTTo_1^{c^{FQ}} && \text{\cit\Zh\/}^{F}
\end{diagram}
\quad
\begin{diagram}[inline,width=2.8em]
\text{\cit\Zh\/}^{R} && \rTTo^{a^{RY}} && \text{\cit\Zh\/}^{Y} \\
& \rdTTo_{a^{RS}} & = & \ruTTo_{a^{SY}} & \\
\uTTo<{c^{QR}}>1 & d & \text{\cit\Zh\/}^{S} & d & \dTTo>{b^{YF}} \\
& \ldTTo^{b^{SQ}} & = & \luTTo_1^{c^{FS}} & \\
\text{\cit\Zh\/}^{Q} && \lTTo_1^{c^{FQ}} && \text{\cit\Zh\/}^{F}
\end{diagram}
\label{dia-zhe-octa}
\end{equation*}
That is, the four triangles marked ``='' commute,
as well as two squares
\[
\begin{diagram}[inline]
\text{\cit\Zh\/}^Z & \rTTo^{b^{ZF}} & \text{\cit\Zh\/}^F \\
\dTTo<{c^{ZR}} && \dTTo>{c^{FS}} \\
T\text{\cit\Zh\/}^R & \rTTo^{Ta^{RS}} & T\text{\cit\Zh\/}^S
\end{diagram}
\qquad
\begin{diagram}[inline]
\text{\cit\Zh\/}^S & \rTTo^{a^{SY}} & \text{\cit\Zh\/}^Y \\
\dTTo<{b^{SQ}} && \dTTo>{b^{YZ}} \\
\text{\cit\Zh\/}^Q & \rTTo^{a^{QZ}} & \text{\cit\Zh\/}^Z
\end{diagram}
\quad.
\]
\end{proposition}
The proof follows from \corref{cor-octa-equiv}.
|
\section{Introduction}
The usual way to apply more computing power to the dynamo problem is
to integrate more and more complex models. That's not the only way, of
course, because even simple models have multiple free parameters---starting
with the dynamo number itself---whose role in controlling the dynamics
is rarely investigated in any detail. The fundamental concern is that
if the behaviour of a system depends sensitively on the values of
unknown parameters, then it becomes far more difficult to say anything
definite about its interpretation as a solar model. I have spent some
time mapping the behaviour of some simple 1D dynamos as a function of
dynamo number and assembling the bifurcation diagrams that show the results.
I do not want to make any strong claims about the particular physical
relevance of these models, but will rather use them as an illustrative
and cautionary example.
This material is presented at greater length in Roald (1998a, b).
\section{Models}
Here we have a standard 1D mean-field $\alpha\omega$ dynamo, with a
dynamical quenching of the shear (Moffatt 1978; Jennings \& Weiss
1991; Roald \& Thomas 1997; Roald 1998a, b),
\begin{eqnarray}
{\partial A\over\partial t} &=& \mathcal D \cos x\: B
+ {\partial^2 A\over \partial x^2},
\label{s1A}\\
{\partial B\over\partial t} &=&
(\sin x +\omega){\partial A\over\partial x}
+ {\partial^2 B\over \partial x^2},
\label{s1B}\\
{\partial \omega\over\partial t} &=&
- \left({\partial A\over\partial x}\right)B
+ \nu{\partial^2 \omega\over\partial x^2},
\end{eqnarray}
where $A$ and $B$ are the toroidal components of the vector potential
and magnetic field, respectively; $x$ is latitude in a quasi-Cartesian
approximation; $\omega=\partial u^\ast/\partial r^\ast$ is the radial
shear, and the third equation controls it by requiring that the
nonlinear terms in the system (i.e., $\omega\partial A/\partial x$ and
$-B\partial A/\partial x$) only exchange energy; and $\nu$ is the
turbulent magnetic Prandtl number. (The geometry is Cartesian except
that the $\alpha$ and $\omega$ effects have been assigned cosine and
sine dependence.) Boundary conditions are
\begin{equation}
A = B = \omega = 0\mbox{ at } x = 0,\pi.
\end{equation}
I compare this system to an essentially similar two-layer, though
still 1D, version of the model that provides a simple representation
of an interface dynamo. The idea is that the $\alpha$ effect
functions in one layer, just inside the base of the convection zone
(\emph{CZ}), while the $\omega$ effect is concentrated in the shear
layer, which is assumed to be outside and beneath the CZ. The system
therefore consists of two partial copies of the basic 1D
$\alpha\omega$ dynamo equations (\ref{s1A}--\ref{s1B}), one in the CZ
with an $\alpha$ effect and one in the radiative zone with an $\omega$
effect. We can connect the two with an analogue of Newton's law of
cooling, such that the flux between layers is simply proportional to the
difference between them.\footnote
{This, of course, is a fair approximation only if the dynamo period is much
longer than the diffusion time between layers. This condition is at best
marginally satisfied, and at worst, quite violated.}
This brings in two additional dimensionless free parameters: the
ratio of the layers' effective diffusivities,
\begin{equation}
\kappa \equiv \nu_\mathrm{rad}/\nu_\mathrm{conv},
\end{equation}
and the ratio of the shell radius to the separation between layers,
which will enter in the form
\begin{equation}
\lambda \equiv (R_\mathrm{shell}/d)^2.
\end{equation}
The system of equations resulting from the above construction is:
\begin{eqnarray}
{\partial a\over\partial t}
&=& \mathcal D \cos x\: b + {\partial^2 a\over \partial x^2}
+\kappa\lambda (A-a),
\label{s2a}\\
{\partial b\over\partial t}
&=& \hspace{2em}{\partial^2 b\over \partial x^2}
+ \kappa\lambda (B-b)
\label{s2b}\\[1ex]
{\partial A\over\partial t}
&=& \hspace{2em}\kappa{\partial^2 A\over \partial x^2}
+ \kappa\lambda (a-A),
\label{s2A}\\
{\partial B\over\partial t}
&=& (\sin x +\omega){\partial A\over\partial x}
+ \kappa {\partial^2 B\over \partial x^2},
+ \kappa\lambda (b-B),
\label{s2B}\\[1ex]
{\partial \omega\over\partial t}
&=& - \left({\partial A\over\partial x}\right)B
+ \nu{\partial^2 \omega\over\partial x^2},
\end{eqnarray}
where $a$ and $b$ describe the $\alpha$ layer, and $A$ and $B$ describe
the $\omega$ layer.
These partial-differential equations were solved by making a spectral
expansion in latitude, then using a standard continuation-method
ordinary-differential-equation solver (\textsc{auto}97).
\section{Results}
\label{Sbd}
One set of results from these models is summarised here on a pair of
bifurcation diagrams (Figures \ref{F1} and \ref{F2}), for the same
magnetic Prandtl number and a comparable range of dynamo numbers
\begin{figure
\epsfxsize=\textwidth
\epsfbox{cbr_e24-.valla.grey.ps}
\vspace{-3.25in}
\caption{Bifurcation diagram for the one-layer system,
$-2000<\mathcal D <0$, $\nu=0.5$, truncation level $N=24$. See
further discussion in \S\ref{Sbd}.}
\label{F1}
\end{figure
\begin{figure
\epsfxsize=\textwidth
\epsfbox{cbr_v16a.valla.grey.ps}
\caption{Bifurcation diagram for the two-layer system, $-2000<\mathcal D <0$,
$\nu=0.5$, $\kappa=10^{-2}$, $\lambda=600$, truncation level $N=16$.
See further discussion in \S\ref{Sbd}.}
\label{F2}
\end{figure
Each curve in these diagrams represents a physically distinct solution
of the equations; different solutions are characterised by the
value of their poloidal magnetic field (in the $\alpha$ layer, in the case
of the two-layer model), averaged over latitude and time. Each branch
is labelled with its symmetry, as defined in Roald \& Thomas (1997).
Bifurcation points are marked.
The isolated points in the lower panels of Figure \ref{F2} (those
plotted with filled circles, boxes, and triangles) show stable
behaviour in parts of parameter space (specifically, for $\mathcal D
\la -200$) where I was unable to get \textsc{auto}97 to lock onto a
solution. These points were evaluated by time integration and at
lower truncation ($N=8$), and so I have marked them with different
symbols than the \textsc{auto}-computed branches. Furthermore, these
solutions were classified qualitatively from the appearance of their
Poincar\'e sections, and it should be understood that this technique
is more art than algorithm, particularly for distinguishing between
quasiperiodic and chaotic solutions. Because transients took a very
long time to decay, I could only evaluate a few of these points for
large negative $\mathcal D$. And lastly, note that their correct time
average of their poloidal field strength is not determined, so they
are all plotted at a single, arbitrary value.
\section{Discussion}
The graphs for the one-layer (Figure \ref{F1}) and two-layer (Figure \ref{F2})
models have many common features, as one would hope for related models.
\begin{itemize}
\item The first bifurcation from the trivial solution---the horizontal
axis on the diagrams---is a quadrupole equilibrium, which loops
back after a fold.
\item There is another (unstable) quadrupole equilbrium, branching from
the trivial solution near $\mathcal D = -700$.
\item A periodic dipole solution branches from the trivial solution near
$\mathcal D = -100$.
\item A ``mixed quadrupole'' (\emph{mq}) solution branches from the first
equilibrium solution shortly before the fold.
\end{itemize}
Similar structure was also found by Jennings \& Weiss (1991) in another
1D model with a different $\omega$-quenching prescription. The
mixed quadrupole branch here produces a Sun-like dynamo mode
(Figure \ref{Fmq}).
\begin{figure
\epsfxsize=3in
\epsfbox{cbr_ctr-e24-mq.ps}
\caption{Contour plots of the variation in time of the toroidal field,
one-layer model,
for the stable \emph{mq} periodic branch at $\mathcal D =
-1011$, $\nu = 0.5$, truncation order $N=24$.}
\label{Fmq}
\end{figure
On the other hand, we must also note two important differences:
\begin{itemize}
\item the two-layer model (Figure \ref{F2}) does not show an unstable
periodic quadrupole solution bifurcating from the trivial solution
anywhere in the range examined,
\end{itemize}
and critically,
\begin{itemize}
\item the stability of the steady quadrupole solution is destroyed in a
subcritical Hopf bifurcation at $\mathcal D = -276$, before the
supercritical bifurcation to the \emph{mq} solution at $-486$.
\end{itemize}
The consequence of this last is that in the two-layer model, the solar-like
\emph{mq} mode is no longer stable, even though it exists in much the
same form as in the one-layer version.
What do we have instead? The steady quadrupole mode loses stability
after bifurcation with a quadrupole periodic mode. This mode tracks
back toward smaller $\mathcal D$, and somewhere around $-60$ appears to
be destroyed in a homoclinic connection with zero. At that point, a
spray of stable unsteady solutions seems to be created, starting with
a chaotic quadrupole. These do not appear to connect conventionally
with any other solutions, and so I couldn't get the continuation-method
solver to lock on. Falling back to simple time-integration, we can
determine the stable behaviour of the system from Poincar\'e sections.
Figure \ref{Fpcar}
\begin{figure
\vspace{-1.2in}
\epsfxsize=4in\epsfbox{cbr_v8a.pcar.bw.ps}
\vspace{-0.7in}
\caption{Poincar\'e sections for the two-layer system, $\nu = 0.5$,
$\kappa=10^{-2}$, $\lambda=600$, truncation $N=8$, taken on the
section plane
$B_3 = 0.005$. Each section is labelled with its dynamo
number. Those for $\mathcal D = -300$ through $-250$ are single
points, while $\mathcal D = -240$ and $-230$ are period-6 (with the
six points of $-240$ too close together to be distinguishable
at this scale). $\mathcal D = -220$ and $-210$ are apparently
chaotic. All of these solutions have quadrupole symmetry.}
\label{Fpcar}
\end{figure
shows solutions for ten different dynamo numbers at the lower end of
the range that shows time-dependent behaviour. We start with two
chaotic solutions at $\mathcal D = -220$ and $-210$, and progress
through a series of periodic solutions. This is clearly the crudest
first pass at a serious mathematical investigation of this
system---for example, there are theorems that require some kind of
very complicated dynamics to be going on between the period-6 orbit at
$\mathcal D = -240$ and the period-1 orbit at $-250$---but pursuing it
any farther seems unprofitable from a physical point of view.
These solutions have been marked on the bifurcation diagram, Figure
\ref{F2}. Figure \ref{Fqpo} shows a typical simulated butterfly
diagram from the quasiperiodic range.
\begin{figure
\epsfxsize=3in
\epsfbox{cbr_v8-a.ctr.qpo-B.ps}
\caption{Contour plots of the variation in time of the shear-layer
toroidal field, two-layer model,
for the stable quadrupole quasiperiodic branch at $\mathcal D =
-1200$, $\nu = 0.5$, $\kappa=10^{-2}$, $\lambda=600$,
truncation order $N=8$.}
\label{Fqpo}
\end{figure
None of the computed modes of the two-layer model shows a Sun-like
butterfly.
\section{Conclusions}
So, despite a recogniseably similar bifurcation structure, the stable
behaviour of the two-layer model is entirely different from that of the
one-layer version. This is not in itself a problem, because the two
models do represent different physics. The concern, however, is that the
only point to studying these simplified, or over-simplified, models is in
the hope that something universal and robust can be identified from them
and applied to our understanding of more sophisticated and computationally
expensive models. Instead, however, we find almost frightening fragility.
Let me be careful in stating the conclusion here: I have shown here that
one pair of admittedly unrealistic models are surprisingly sensitive to
the addition of a simple bit of physics. This does not \emph{directly
predict} anything about the behaviour of more complex models. What it
does force, however, is the indirect question: can we be confident that
similar fragility does not occur in other models? Because if it does, the
models are effectively useless. Having worked out quite a few bifurcation
diagrams, my personal impression is we would be very lucky to find
significant details in common between 1D models like these and more
realistic dynamo models.
|
\section{Evidence for the disappearance of muon neutrinos}
The data of Super--Kamiokande (SK) \cite{SK,SK1} and other detectors
have given strong evidence that $\nu_\mu$'s and $\overline{\nu}_\mu$'s
`disappear'.
This evidence comes from the observation of three
experimental effects:
(i) the detection of an up--down asymmetry
for the $\mu$--like events,
(ii) the detection of a small $\mu/e$ ratio,
(iii) the detection of a distortion of the zenith angle
distribution and a suppression of the $\nu$ induced upward
going muon flux.
The three effects are listed in order of `robustness'
with respect to systematic uncertainties.
The statistical significance of the effects in SK,
especially for (i) and (ii), is very strong.
In the following we will discuss how theoretical
uncertainties in the predictions
cannot `reabsorb' the observed effects,
but do play a role in the {\em interpretation} of the data.
\section{Systematic uncertainties in the predictions}
In the prediction of the event rates for
an atmospheric $\nu$ experiment
one needs to: (i) consider an initial flux of cosmic ray particles,
(ii) model the hadronic
showers produced by these particles in the Earth atmosphere,
(iii) describe the $\nu$ cross sections,
(iv) describe the detector response
to $\nu$ interactions (and possible background sources).
Here we will consider only the first three `theoretical' elements of
the calculation, and will argue that there are significant uncertainties,
that influence the absolute normalization, the
shape of the energy spectrum, the angular distribution,
and the $\mu/e$ ratio of the events.
The primary cosmic ray (c.r.) flux has been a major source
of uncertainty (see \cite{Gaisser} for a detailed discussion)
because of the discrepant\footnote{It is highly unlikely that the
obseved differences are the result of time variations.}
results obtained by two groups (Webber 79 and LEAP 87)
differing by $\sim 50\%$ (see fig.~1).
Recently, new measurements of the c.r. proton flux
have given results consistent with the lower
normalization.
If the lower normalization is accepted as correct,
the uncertainty in the primary c.r. flux can be
reduced, however the description of
the primary c.r. flux (see fig.~1) used in
two calculations of
the atmopheric $\nu$ fluxes: Honda et al. (HKKM) \cite{HKKM} and
Bartol \cite{Bartol} used in predictions for SK
and other detectors are then too high (by $\sim 30\%$ and
$\sim 10$\%).
\begin{figure} [bt]
\centerline{\psfig{figure=proton.ps,height=5cm}}
\caption {Measurements of the c.r. proton flux (points)
(see \protect\cite{Gaisser} for the references), and the representations
used as input in the Bartol \protect\cite{Bartol} and
HKKM \protect\cite{HKKM} calculations (lines).}
\end{figure}
A second source important source of uncertainty is our lack of
knowledge on the properties of particle production
in $p$--nucleus and nucleus--nucleus interactions.
The calculations of Bartol \cite{Bartol} and HKKM \cite{HKKM} use different
descriptions for the multiplicity and
energy spectrum of the pions produced
in $p$--Nitrogen (Oxygen) interactions.
For the same primary c.r. flux, this would result
in a $\sim 20\%$ higher $\nu$ event rate
for the Bartol calculation.
Some controversy exists about what description
of hadronic interactions is in better agreement with the existing data.
An experimental program,
studying in detail the structure
of particle production in the relevant energy range
(a broad region centered at $E_0 \sim 20$~GeV),
would result in an improvement in the predictions.
The similar normalization of the two \cite{Bartol,HKKM} calculations
is the result of a cancellation between
a higher (lower) flux for the c.r. flux,
and a lower (higher) $\nu$ yield per primary particle
for the HKKM (Bartol) calculation.
This cancellation, to a large extent, is not
casual, but is the consequence of fitting
the (same) data on $\mu^\pm$ fluxes at ground level.
This underlies the importance of these measurements.
It is very desirable to repeat them with greater accuracy.
High altitude meaurements
with balloons \cite{Gaisser} also offer a great potential.
A third source of `theoretical' uncertainty, of comparable importance to
the other two, is related
to the description of $\sigma_\nu$.
At high $E_\nu$, when most of the phase space
for $\nu$ interactions is in the deep--inelastic region,
$\sigma_\nu$ is reliably calculable in terms of well determined
parton distribution functions (PDF's). However for $E_\nu \sim 1$~GeV
the description of $\sigma_\nu$ is
theoretically more difficult. Quasi--elastic scattering is
the most important mode, but also events with the production
of one or more pions (where the additional particles are
undetected or are reabsorbed in the target nucleus) are important
contributions to the signal. The production
of $\Delta$'s and other resonances is
important, nuclear effects have to be included.
A relatively small modification in the description of
a fraction of $\sigma_\nu$ in the SK montecarlo:
the choice of a new set of PDF's
(GRV94LO replacing the CCFR parametrization) has resulted in an increase
of the predicted number of partially contained events by
approximately 7\%
(compare the MC predictions in \cite{SK} and \cite{SK1}).
It appears very difficult to calculate accurately
$\sigma_\nu$ from first principles in the
relevant energy region. The existing data do not
determine the absolute value of the cross section
and the energy spectrum of the final state lepton better that
$\sim 15\%$. Additional data
could help in improving the situation. The
K2K $\nu$ beam with a spectrum not too different
from the atmospheric one offers interesting possibilities.
\section {Robust properties of the predictions and observed effects}
Two properties of the $\nu$ fluxes
are to a large extent independent from the details of the
calculation and provide `self calibration' methods:
(i) the fluxes are approximately up/down symmetric:
$\phi_{\nu_\alpha} (E_\nu, \theta) \simeq
\phi_{\nu_\alpha} (E_\nu, \pi - \theta)$,
(ii) the $\nu_\mu$ and $\nu_e$ fluxes
are strictly related to each other because they are
produced in the chain decay of the same
charged mesons (as in $\pi^+ \to \nu_\mu \, \mu^+$ followed
by $\mu^+ \to \overline{\nu}_\mu \nu_e e^+$). Writing
$\phi_{\nu_\mu} (E,\theta) = r (E_\nu, \theta) \times \phi_{\nu_e} (E,\theta)$
the factor $r (E_\nu, \theta)$ varies slowly with
energy and angle and is quite insensitive to the details of
the calculation.
These two properties are at the basis of the robustness of the
evidence for oscillations.
The up--down symmetry follows as a simple
and {\em purely geometrical} consequence from two assumptions:
the primary c.r. flux is isotropic,
the Earth is spherically symmetric.
The c.r. flux at 1 A.U. of distance form the sun
is isotropic to a precision better than $10^{-3}$,
as can be measured observing in a fixed direction and
looking for time variations while the Earth rotates.
The isotropy is spoiled by the geomagnetic field that
bends the particle trajectories and forbids the lowest
rigidity ones from reaching the Earth's surface introducing directional
(east--west) and location (latitude) effects.
\begin{figure} [bt]
\begin {minipage} [h] {15cm}
\psfig{figure=asym.ps,height=4.4cm}
\psfig{figure=r.ps,height=4.4cm}
\end{minipage}
\caption {The left panel shows
the up/down asymmetry (no--oscillation hypothesis)
of cc--interacting $\nu_\mu$'s
and $\overline{\nu}_\mu$'s in the Bartol \protect\cite{Bartol}
and HKKM \protect \cite{HKKM} calculations for the Kamioka site.
The right panel shows the $\mu/e$ ratio for the two models.}
\end{figure}
These effects vanish at large momentum (see fig.~2).
The measurement of an (oscillation independent)
east--west effect for atmospheric $\nu$'s
in agreement with predictions, by SK \cite{SK-east-west}
is an important test, that validates the calculations.
Note that at Kamioka (near the magnetic equator)
geomagnetic effect have the opposite effect to $\nu$--oscillations,
and produce an up--going $\nu$ flux
{\em larger} than the down--going one
(both magnetic poles are {below}
the detector).
At the Soudan mine (near the magnetic pole)
the opposite is true.
Note also that the predicted asymmetry at low energy
(see fig.~2) has some model dependence with the Bartol calculation
predicting a higher no--oscillation asymmetry.
This is important for the
detection of a zenith angle modulation in the Soudan detector
and is also relevant for the interpretation of the SK sub--GeV events.
The right panel of fig.~2 shows how different
calculations of the atmospheric $\nu$ flux predict very similar
$\mu/e$ ratios.
This however refers to a {\em fixed} value of $E_\nu$.
In fig.~3
\begin{figure} [hbt]
\centerline{\psfig{figure=ener.ps,height=5.0cm}}
\caption {
Approximate energy distributions
of interacting neutrinos for different
classes of events in SK. Note the difference
in the shape of the response for $\mu$--like and $e$--like events }
\end{figure}
we show an estimate of the energy distributions of the neutrinos
that produce the SK events, note how the distributions for
$\mu$ and $e$ like events differ.
For the multi--GeV samples,
a harder $\nu$ spectrum, or a faster raise with energy of
$\sigma_\nu$, result in a larger (smaller)
increase in the predicted rate of $\mu$--like ($e$--like) events,
therefore in a smaller double ratio $R$ (because of a
larger denominator) and finally to a larger $\Delta m^2$
(for the same mixing) in the $\nu_\mu \leftrightarrow \nu_\tau$
interpretation to explain the larger suppression.
At this conference, SK has presented
a new estimate
for the (90\% C.L.) allowed region in the
($\sin^2 2 \theta$, $\Delta m^2$) plane for the
$\nu_\mu \leftrightarrow \nu_\tau$ hypothesis,
considering a larger exposure and
a slightly modified MC calculation.
The new \cite{SK1} allowed region is smaller that the
previously published one \cite{SK} not including the
interval $|\Delta m^2| \simeq 0.5$--$1.0 \times 10^{-3}$~eV$^2$,
a result very encouraging for the LBL programs.
The use of a new set of PDF's in the description of $\sigma_\nu$
has the qualitative effect to enhance the contribution of
high energy events, and for the argument outlined above
is an important contribution to the exclusion of the low
$\Delta m^2$ interval.
\section{Outlook}
The detection of oscillations in atmospheric $\nu$
experiments is a result of great importance.
The detailed study of this phenomenon
and the precise measurement of the parameters
(masses and mixing) involved is a great opportunity
and a difficult challenge.
SK has a remarkable potential to
obtain more convincing evidence and more precise measurements.
New data on primary c.r. fluxes, hadron--nucleus interactions,
$\nu$--nucleus interactions and $\mu^\pm$ fluxes could help
the interpretation of present and future data.
Long baseline $\nu$ beams, have also the
potential to confirm the results
and study the phenomenon.
This could happen very soon with the K2K project,
the existence of two (similar to each other) LBL projects in the
US and in Europe is seen by some as a beneficial case
of scientific competition, and by others as a dangerous waste of
resources.
\noindent {\bf Acknowledgments:} Special thanks to prof. Y.~Suzuki
for kind explanations.
\section*{References}
|
\section{Introduction}
Low-mass giant stars with enhanced lithium are rare.
Indeed, Li is destroyed in all but the outermost 1--2\% of a
main-sequence star such as the Sun, and further reduced as the convection
zone deepens during giant branch evolution (Pilachowski \mbox{{\it et al.}}\ 1993).
Yet Li-enhanced giants do indeed exist.
Wallerstein \& Sneden (1982) discovered that the somewhat metal-rich giant
HD~112127 has a Li abundance log~$\epsilon$(Li)~=
+3.0~$\pm$~0.2.\footnote
{[A/B]~$\equiv$~log$_{\rm 10}$(N$_{\rm A}$/N$_{\rm B}$)$_{\rm star}$~--
log$_{\rm 10}$(N$_{\rm A}$/N$_{\rm B}$)$_{\odot}$, and
log~$\epsilon$(A)~$\equiv$~log$_{\rm 10}$(N$_{\rm A}$/N$_{\rm H}$)~+~12.0.}
Brown \mbox{{\it et al.}}\ (1989) noted that this is a factor of 30 higher than expected;
their survey of 644 field giants revealed only one more of similarly high
Li abundance, plus eight with moderate Li enhancements.
These stars are of lower mass and luminosity, and are probably less evolved,
than the ``classical'' Li-rich giants of the Galaxy and the Small Magellanic
Cloud (Plez \mbox{{\it et al.}}\ 1993).
The possibly distinct group of Li-rich stars associated with dust
shells is discussed by de~la~Reza \mbox{{\it et al.}}\ (1996, 1997).
This paper reports the serendipitous discovery of another extremely
Li-rich low-mass giant: IV-101, an otherwise normal member of the
globular cluster M3 (NGC 5272).
The light-element abundances of IV-101 indicate a moderate
degree of mixing and its color favors an evolutionary status on
the first ascent of the giant branch prior to the He core flash.
We suggest that it represents a fleeting phase of Li enrichment
immediately following a significant mixing event, and that such episodes of
enrichment are not uncommon among metal-poor and other low-mass giants ---
they are merely short-lived.
\section{Observations and Data Reduction}
The spectroscopic data discussed here are part of an abundance survey
of 47 M3 giant stars conducted in 1998 March with the Keck~I High
Resolution Echelle Spectrograph (HIRES --- Vogt \mbox{{\it et al.}}\ 1994).
The target stars are largely situated near the center of M3,
selected from the $VI$ photometry and coordinates
obtained with the post-repair Hubble Space Telescope by Marconi \mbox{{\it et al.}}\
(1998) and communicated by Ferraro \& Dorman (1998).
Also included were giants from the outer parts of M3, with photometry from
Johnson \& Sandage (1956), to provide a nearly complete sample of
cluster members within 1.5~mag of the red giant branch (RGB) tip.
With a slit width of $0\farcs86$, the Keck~I HIRES
achieved a two-pixel spectral resolution of R~=~45,000
on the Tektronics 2048$\times$2048 detector.
The S/N ratio was typically 100 per pixel for V~$<$~14.
Spectral coverage extended from \wave{4470} to \wave{6720} with gaps at the
ends of each echelle order.
While the strong \nion{Li}{i} resonance doublet at \wave{6707} was included,
the weaker \nion{Li}{i} \wave{6104} feature fell in a gap and was not recorded.
We also gathered exposures of quartz and Th-Ar lamps for spectrum
calibrations, and of rapidly rotating B-type stars for telluric line removal.
Spectrum reduction procedures were identical to those described in
earlier papers of this series (\mbox{{\it e.g.}}, Sneden \mbox{{\it et al.}}\ 1997; Kraft \mbox{{\it et al.}}\ 1998).
The equivalent width (EWs) measures are available by request to us and
will be published in the future as part of our larger M3 study.
During inspection of the data we spotted an extremely strong
\wave{6707} \nion{Li}{i} feature in IV-101.
Here we report abundance analyses for IV-101 and three very
similar M3 giant stars.
Basic data are gathered in Table~1.
All four stars are radial-velocity members of the cluster
(Gunn \& Griffin 1979; Pryor \mbox{{\it et al.}}\ 1988).
These and our own radial velocity measurements given in Table~1 show no
velocity variations at the 0.5~\mbox{km~s$^{\rm -1}$}\ level over periods of decades.
Our spectra in the vicinity of the \nion{Li}{i} features are
displayed in Figure~\ref{liobs}.
The great strength of the Li feature in IV-101 is obvious
(EW~$\simeq$~520~m\AA), and the \nion{Al}{i} doublet
($\lambda\lambda$6696, 6699~\AA) is much stronger in IV-101 than in
I-21 and IV-77.
The \nion{Al}{i} line strengths of vZ~746 are comparable to those of IV-101,
yet vZ~746 has no detectable \nion{Li}{i} line.
\begin{figure}[p]
\plotfiddle{kraft_fig1.ps}{120pt}{90}{40}{40}{160pt}{-75pt}
\vskip 0.65truein
\caption{Spectra of the \nion{Li}{i} transition region in the
four program stars.
The \nion{Li}{i} EW is $\simeq520$~m\AA\ in IV-101, but the line is
undetected in the other three stars.}
\label{liobs}
\end{figure}
\section{Abundance Analysis}
Abundances were derived in the same manner as in
our previous papers (\mbox{{\it e.g.}}, Sneden \mbox{{\it et al.}}\ 1997; Kraft \mbox{{\it et al.}}\ 1998).
We first considered the photometric information of the stars.
The positions of IV-101, I-21, and IV-77 (Johnson \& Sandage 1956)
in the M3 CMD of Buonanno \mbox{{\it et al.}}\ (1994) are shown in Figure~\ref{buonanno};
these stars lie hard against the right-hand edge of the
RGB--AGB sequence in this figure.
The CMD and an assumed distance modulus of $(m-M)_0$~=~15.02
(Djorgovski 1993) yield $M_{\rm bol}$~=~--2.8 for IV-101.
The central object vZ~746 (von Zeipel 1908) was not observed by
Buonanno \mbox{{\it et al.}}.
We have estimated its $B-V$ value from its observed $V-I$ and the
$B-V$ vs $V-I$ relation for other M3 giants (\mbox{{\it e.g.}}, Marconi \mbox{{\it et al.}}\ 1998),
but not plotted it in Figure~\ref{buonanno}.
vZ~746 is $\sim$0.2~mag brighter than this sequence in the Marconi \mbox{{\it et al.}}\
$V$ vs.\ $V-I$ diagram.
Probably IV-101, I-21, and IV-77 are on the first-ascent RGB,
but a double-shell AGB phase cannot be entirely ruled out,
especially for vZ~746.
None of the four stars was seen to vary in 90 B-band plates obtained
over a two-year period by Corwin \& Carney (1999).
\begin{figure}[p]
\plotone{kraft_fig2.ps}
\caption{The $V$, $B-V$ diagram for the most luminous stars in M3 from the
observations of Buonanno \mbox{{\it et al.}}\ (1994). Three of the program stars are
indicated in the figure, while the fourth star, vZ~746, was not observed by
Buonanno \mbox{{\it et al.}}}
\label{buonanno}
\end{figure}
Trial values of \mbox{T$_{\rm eff}$}\ and log~g were estimated for each star based on
the earlier calibration (Sneden \mbox{{\it et al.}}\ 1992; Kraft \mbox{{\it et al.}}\ 1995) of these
quantities as functions of \mbox{$M_{\rm V}^{\rm 0}$}\ and \mbox{$(B-V)^{\rm 0}$}\ for giants having
metallicities [Fe/H]~$\sim$~--1.5.
Final model parameters were determined by iteration, to satisfy
standard requirements that abundances deduced from iron lines show no
dependence on line excitation, ionization, or EW.
These parameters are listed in Table~\ref{abunds}.
All four stars are virtually identical in \mbox{T$_{\rm eff}$}, log~g,
v$_{\rm t}$, and [Fe/H].
The mean spectroscopic gravity of the four stars, $<$log~g$>$~$\sim$~+0.8,
agrees well with the mean evolutionary gravity derived from knowledge of the
M3 distance modulus, an assumed mass of 0.85~M$_{\odot}$, and our values of \mbox{T$_{\rm eff}$}.
For most elements, the abundances given in Table~\ref{abunds} were derived
from the EWs, while for Li, N, O, Na, and Ba we used spectral syntheses.
The \nion{Li}{i} line has doublet structure, and since the line components
have unequal transition probabilities, an asymmetrical profile is computed.
Yet IV-101's observed line is remarkably symmetrical, possibly indicating
a small velocity gradient desaturating this very strong line.
Thus our estimate of log~$\epsilon$(Li)~$\simeq$~+3.0 could be an
overestimate.
We found no evidence for $^{\rm 6}$Li, but the anomalous \nion{Li}{i}
profile precludes a definitive statement on this.
For the other three giants, the estimated upper limit
log~$\epsilon$(Li)~$\lesssim$~--0.5 is compatible with the low values of
halo field giants (Pilachowski \mbox{{\it et al.}}\ 1993).
For three stars, C abundances were taken, with minor extrapolation, from the
study of the \wave{4300} CH band in M3 and M13 giants by Smith \mbox{{\it et al.}}\ (1996).
For vZ~746, we assumed that
[C/Fe]$_{\rm vZ~746}$~$\simeq$~[C/Fe]$_{\rm IV-101}$, since these two
stars have comparable abundances of the other light ``proton-capture''
elements O, Na, Mg and Al.
Abundances of N were then estimated from synthesis of several very
weak CN lines surrounding the [\nion{O}{i}] lines; these agree well with
[N/Fe] values based on the calibration of the CN bands using the index
S(3839) by Smith \mbox{{\it et al.}}\ (1996).
Our derived [N/Fe] in IV-101 is 0.4--0.5~dex larger than those
of the other three stars, but this agrees with Suntzeff's (1981)
claim that IV-101 has a relatively large N abundances.
Abundances of the remaining elements are comparable in all four stars, and
very similar to their abundances in other cluster and halo field giants.
The abundances of Eu and Ba are also ``normal'' for the metallicity of
M3 (Armosky \mbox{{\it et al.}}\ 1994; Shetrone 1996a,b).
The [Ba/Eu]-ratios range from about $-0.2$ to $-0.4$, and the four stars
do not differ significantly from one another in this ratio.
Thus there is no evidence for any s-process abundance enhancements.
Overall, the general abundance patterns simply suggest that proton-capture
nucleosynthesis has gone further in IV-101 and vZ~746 than in I-21
and IV-77: C, O and possibly Mg are depleted, while N, Na, and Al
are enhanced in IV-101 and vZ~746 compared to I-21 and IV-77.
These four stars display Na vs.\ O and Al vs.\ Mg anticorrelations,
symptoms of deep mixing among M3 and M13 giants ({\it cf.},
Kraft \mbox{{\it et al.}}\ 1997; Shetrone 1996a,b).
Except for the overabundance of Li, IV-101 has essentially the same
abundances as do well-mixed giants in M3 and M13.
There are more of these stars in M13 than in M3, but they are not
rare in either cluster.
\section{Lithium Production and Depletion in Low-Mass Stars}
It is improbable that M3 IV-101 has retained its primordial Li abundance
while still occupying a normal place in the M3 color-magnitude array.
The Li abundance of M3 IV-101 is higher than its probable initial value
of log~$\epsilon$(Li)~$\simeq$~+2.2, that of halo turnoff stars
(\mbox{{\it e.g.}}, Spite \& Spite 1982; King \mbox{{\it et al.}}\ 1996).
Thus accretion of a substellar body that has not depleted its Li
(e.g.\ Brown \mbox{{\it et al.}}\ 1989) cannot provide enough additional Li.
Accretion of Li from a circumstellar disk also appears unlikely.
The presence of substantial cool material would also be seen in the
\nion{Na}{i} D~lines, whose line formation conditions are similar to those
of the \nion{Li}{i} 6707~\AA\ doublet.
But the D~lines in IV-101 are indistinguishable from those of two of
the other three giants.
Instead, fresh Li must have been produced and convected into the
envelope of IV-101.
This probably happened in a nucleosynthetic ``event'' that included the
``beryllium transport process'' first proposed by Cameron \& Fowler (1971).
This fusion sequence is
$^{\rm 3}$He($\alpha$,$\gamma$)$^{\rm 7}$Be(e,$\nu$)$^{\rm 7}$Li, with
the requirement that the $^{\rm 7}$Be created in the nuclear processing
zone be transported to cooler envelope regions so that the eventual
$^{\rm 7}$Li product will not be destroyed by proton captures.
Cameron \& Fowler originally wanted this mechanism to operate in
He-shell fusion zones, in order to simultaneously create the
Li and s-process overabundances that are observed
(\mbox{{\it e.g.}}, Plez \mbox{{\it et al.}}\ 1993) in some S and C stars.
But the greatly altered proton-capture element abundances, unaccompanied
by a barium abundance enhancement, probably rules out a He-fusion
zone origin for the large Li abundance in IV-101.
Li synthesis in a hydrogen shell fusion zone seems more likely.
Deep envelope mixing in low mass stars is explored by
Sackmann \& Boothroyd (1999), who call their mechanism
``cool bottom processing'', and focus on its application to the
Li-rich giants of de~la~Reza \mbox{{\it et al.}}\ (1996, 1997).
They studied both single ``nucleosynthetic event'' and ``steady state''
models of various mass/metallicity combinations.
They argue that the $^{\rm 7}$Li production process, which
depletes $^{\rm 3}$He, should be even more efficient in globular cluster
giants than in the giants discussed by de la Reza \mbox{{\it et al.}}, since the lower
metallicity cluster stars have hotter CNO H-burning shells.
Our abundance results cannot shed light on the physics of the
nucleosynthesis event (a hydrogen shell-source instability?
see von~Rudloff \mbox{{\it et al.}}\ 1988).
But we can argue that the Li seen in IV-101 was most probably
synthesized recently, because the Li so produced is likely to be
destroyed on relatively short time scales.
In classical models, an M3 star with M$_{\rm bol}$~=~--2.8 has
T~=~1.1$\times$10$^{\rm 6}$~K at the base of its convective envelope
(Sweigart \& Gross 1978).
Li is destroyed by proton capture at temperatures roughly twice this
value, so even modest overshoot of moving fluid cells below the
convection zone would bring about Li depletion.
Such mixing probably occurs quasi-continuously for all stars on the RGB,
since virtually all bright globular-cluster
giants show some evidence for enhanced proton-capture products.
To date, Li has been detected in only two other stars out of $\sim$100
globular cluster giants examined at high spectral resolution.
Thus the depletion time scale for Li in globular-cluster giants
must be $\lesssim$4$\times$10$^{\rm 4}$~yr, 2\% of the
evolutionary time scale to the RGB tip of 2$\times$10$^{\rm 6}$~yr
(Rood 1972; Kraft \mbox{{\it et al.}}\ 1993, Fig.~8).
The two other known Li-rich globular cluster giants differ
somewhat from the abundance pattern of IV-101.
The M5 W~Vir variable V42 (Carney \mbox{{\it et al.}}\ 1998) shares the
low O, high Na, and normal Ba/Eu of IV-101, but neither its Mg nor
Al are significantly altered, and the Li enhancement is
far less: log~$\epsilon$(Li)~$\simeq$~+1.8.
Li in this AGB star surely must have been produced on the AGB itself,
again possibly due to H-shell burning, but shortly before the
star ``looped'' into the W~Vir domain.
Too much time would have elapsed since the first RGB ascent for Li
synthesized then to have survived in V42.
In NGC~362, Smith \mbox{{\it et al.}}\ (1999) discovered a strong
\nion{Li}{i} line in the giant-branch tip variable V2.
Their derived abundance is log~$\epsilon$(Li)~$\simeq$~+1.2; other
light elements show no obvious abundance anomalies.
The Sackmann \& Boothroyd (1999) ``cool bottom processing'' may have
been at work in these two stars, but their cases are not as easy to
make as in M3 IV-101.
de~la~Reza \mbox{{\it et al.}}\ (1996, 1997) argue that their Li-rich low-mass,
high metallicity field giants have recently ejected Li-rich dusty shells.
IRAS surveys (\mbox{{\it e.g.}}, Gregorio-Hetem \mbox{{\it et al.}}\ 1993;
de la Reza \mbox{{\it et al.}}\ 1996) detect shells in a large fraction of these stars.
We looked for evidence for a dusty shell surrounding IV-101 by comparing
IRAS 100\/\micron\ fluxes centered on each of the four stars listed
in Table~\ref{abunds}.
Using the DIRBE-calibrated IRAS maps of Schlegel \mbox{{\it et al.}}\ (1998),
we found that the 100\/$\mu$m fluxes at the positions of IV-101, I-21,
vZ~746, and IV-77 agree to within 15\%.
Unfortunately, 100\/\micron\ flux might not be detectable from a
shell in which the metallicity is reduced forty-fold.
We also examined the radial velocity of the deep absorption cores of the
Na~D lines which, in cool giants, often show evidence of outflow in the
form of a cold expanding shell (Peterson 1981).
Relative to neighboring weak \nion{Fe}{i} and \nion{Ti}{i} lines, the core
displacements are --2.3, --2.4, +2.1, and --2.5~\mbox{km~s$^{\rm -1}$}, respectively,
for these four stars, again suggesting that IV-101 is in no way unusual.
We cannot definitely say whether the proposed mixing event is a rare
one due to individual circumstances, or whether such events commonly occur
during the RGB and AGB evolution of low-mass giants.
A mixing event that produces a Li enhancement may truly be
an anomaly in globular-cluster stellar evolution.
The similarity of the CNO and NaMgAl abundance ratios in IV-101 (with
high Li) and vZ~746 (with undetectable Li) might suggest this.
Perhaps circulation currents move faster to transport $^{\rm 7}$Be in
IV-101 than in vZ~746, indicative of a general spread of internal angular
momenta among cluster giants ({\it cf.} Pinsonneault 1997).
Since Peterson (1983,1985) showed that BHB stars in M13 have larger
rotational velocities than do both M3 and M5 BHB stars, giants in the
globular cluster M13 might also be expected to show Li enhancements.
However, our inspection of the spectra of Kraft \mbox{{\it et al.}}\ (1997) revealed
no prominent \nion{Li}{i} lines in their 18 giants.
Mixing events may be relatively common features of the normal stellar
evolution of globular-cluster and other low-mass giants, but might be
rarely observed because of the rapid destruction of newly minted Li.
In this scenario, IV-101 has recently experienced such an event;
a similar event may also have occurred in vZ~746 but sufficiently far in the
past ($>$10$^{\rm 4}$~yr) that the Li has by now been completely destroyed.
Such events probably occur at various luminosities along the giant branch.
This is consistent with the appearance all along the giant branch
of changes in CNO and NaMgAl abundance ratios.
Should Li enrichment be a common occurrence, then roughly 1--2\%
of all such giants should show some Li enhancement.
Should the He flash be involved, then the red horizontal branch
stars should show enhancements more frequently.
\medskip
\bigskip
\bigskip
We thank Franz Ferraro and Ben Dorman for providing new M3
HST photometry and astrometry, and appreciate the helpful comments
on this work by Inese Ivans and Bruce Carney.
This research was supported by NSF grants AST-9217970 to RPK and JPF,
AST-9618502 to RCP, and AST-9618364 to CS.
This paper is dedicated to the memory of our beloved colleague Ed Langer,
who died after a brief illness on February 16, 1999.
Ed brought a unique theoretical perspective to our globular cluster
abundance studies.
His career truly embodied the academic ideals of inspiration in both teaching
and research.
He made friends wherever he traveled, and was an inspiration to students.
We will miss him greatly.
\onecolumn
\begin{deluxetable}{lrrrr}
\tablenum{1}
\tablewidth{4.7in}
\tablecaption{M3 Data}\label{abunds}
\tablecolumns{5}
\tablehead{
\colhead{Quantity} &
\colhead{vZ 265} &
\colhead{vZ 746} &
\colhead{vZ 334} &
\colhead{vZ 1392} \nl
\colhead{} &
\colhead{IV--101} &
\colhead{\nodata} &
\colhead{IV--77} &
\colhead{I--21}
}
\startdata
\cutinhead{Basic Data and Model Atmosphere Parameters}
V & 13.18& 12.96& 13.11& 13.11\nl
B--V & 1.35& [1.32]& 1.32& 1.35\nl
T$_{\rm eff}$ & 4200& 4200& 4250& 4200\nl
log g & 0.85& 0.85& 0.75& 0.75\nl
v$_{\rm t}$ & 1.7& 1.7& 1.6& 1.6\nl
[Fe/H] & --1.50& --1.48& --1.52& --1.47\nl
\cutinhead{Absolute log $\epsilon$ Lithium Abundances}
Li (Li I) & 3.0: & $\leq$--0.5& $\leq$--0.5& $\leq$--0.5\nl
\cutinhead{Relative [X/Fe] Abundances}
C (CH) & --0.55& --0.50& --0.25& --0.20\nl
N (CN) & +1.0& +0.6& +0.5& +0.5\nl
O ([O I]) & --0.03& --0.15& +0.25& +0.30\nl
Na (Na I) & +0.18& +0.28& --0.28& --0.18\nl
Mg (Mg I) & +0.20& +0.16& +0.36& +0.30\nl
Al (Al I) & +0.91& +0.91& --0.01& --0.09\nl
Si (Si I) & +0.27& +0.16& +0.08& +0.19\nl
Ca (Ca I) & +0.24& +0.27& +0.29& +0.27\nl
Sc (Sc II) & --0.32& --0.16& --0.36& --0.38\nl
Ti (Ti I) & +0.41& +0.26& +0.13& +0.27\nl
V (V I) & +0.03& +0.08& --0.02& +0.01\nl
Fe (Fe I) & --0.02& +0.01& +0.01& +0.03\nl
Fe (Fe II) & +0.02& +0.00& --0.01& --0.02\nl
Ni (Ni I) & --0.05& --0.07& --0.06& --0.18\nl
Ba (Ba II) & +0.22& +0.24& +0.21& +0.16\nl
Eu (Eu II) & +0.63& +0.48& +0.39& +0.44\nl
\enddata
\end{deluxetable}
\twocolumn
|
\section{Introduction}
The subject of this report covers a wide range of items and problems
in the theoretical description of nuclear systems, the common link
among them being the strong interaction intervening between
the constituents of composite systems and the many--body complexity
of the latter.
Nuclear and nucleonic structure has been experimentally investigated
using probes of quite different nature, whose interaction with the
nuclear system involves forces of different nature, from the
electroweak one to the most complicated strong interactions between
relativistic heavy ions.
We will discuss here at length the information which can be
obtained through unpolarized and polarized electron--nucleus
scattering, briefly touching also neutrino scattering.
Strongly interacting probes, like mesons ($\pi, K$) and nucleons will
be considered in connection with the specific spin--isospin channels
they can excite in a nuclear system.
Finally we will briefly illustrate the problems connected with
subnucleonic degrees of freedom, in particular the transition from
nuclear matter to quark matter: the so--called deconfinement
phase transition, perhaps the only one occurring in the early stages of
the Universe which one can hope to reproduce in the laboratory.
Different theoretical schemes are employed to test the performances
and limitations of various models for describing nuclear dynamics and
spectra. Indeed in the nuclear medium many degrees of freedom come
into play and require different techniques to be enlightened.
For example, sticking to the case of electromagnetic probes, one has
to consider not only the one--body nucleonic current, but also the
two--body ones, which involve pairs of nucleons correlated via
the exchange of charged mesons. In the domain of deep inelastic
scattering the high momentum photons exchanged between the electron
and the nucleus shed light on the quark structure of the nucleon,
namely on the complications associated with the
subnucleonic degrees of freedom and colour confinement.
Just mentioning all the above items and themes can give the
impression of many different, disconnected fields of research: on
the contrary, theoretical and experimental nuclear physics can
be viewed as a ``magic circle'' which involves, with various and
strong interrelations, all different subfields. One can start,
e.g., from the single nucleon to proceed toward the structure of
nuclei and hypernuclei, down to the ideal ``nuclear matter'' system,
then to neutron stars and astrophysical systems, in which realistic
models envisage the existence of a quark--gluon plasma phase. And
then back, through the confinement mechanism, to nucleons and
nuclear systems.
\section{Weakly interacting probes: electrons and neutrinos}
The main advantage of electron--nucleus scattering stems from the
fairly good knowledge of the basic electromagnetic interaction vertex;
moreover the virtual exchanged photon can penetrate well inside the
nuclear system before being absorbed, thus involving into the
process any nuclear constituent in the whole volume of the nucleus.
This probe allows to investigate, depending upon the four--momentum
transfer, both {\it nuclear} and {\it nucleonic} structure. In the
second case, which concerns deep inelastic scattering experiments,
the partonic structure and perturbative QCD corrections can be tested,
together with various quark models for the confined hadronic states.
In the former case, which refers to relatively smaller momentum transfers,
nuclear structure studies can be performed by comparing the theoretical
evaluation of the inelastic electron scattering cross sections with the
measured data, in a wide range of energy and momentum transfers as well
as in different experimental setups, appropriate for inclusive $(e,e')$,
one--exclusive $(e,e'p)$, etc. measurements.
In all situations we test the theoretical description of complex nuclei
in connection with both the many--body scheme employed and the role played
by the NN interaction in shaping cross sections and amplitudes
for the excitation of specific nuclear states.
Taking advantage from the fact that electrons interact with the whole
nuclear volume, one expects that surface effects are of minor importance:
thus it is convenient to perform calculations in the framework of nuclear
matter.
Both perturbative and variational techniques have been employed to
test different model descriptions of:
\begin{enumerate}
\item
the two (and three)--body nucleon--nucleon interaction
\item
the nuclear electromagnetic currents and form factors
(including the single nucleon current as well as the two--body,
meson exchange currents)
\item
the weak nuclear and nucleonic currents (e.g. to investigate
the strange form factors of the nucleon)
\item
relativistic effects both in the kinematics and in the currents.
\end{enumerate}
All the above items concur in determining the so--called
{\bf nuclear response functions}, which can be directly confronted with
the ones extracted from the experimental data.
Different types of response functions can be obtained from the cross
sections for inelastic electron scattering; we shall list below a few
items which have been extensively explored.
\begin{description}
\item[$\diamondsuit$]
{\sl Inclusive scattering of unpolarized electrons off nuclei}
In Born approximation the differential cross section (with respect to
the energy, $\epsilon'$, and the scattering angles,
$\Omega\equiv(\theta,\varphi)$, of the final electron) is given by:
\begin{equation}
\frac{d^2\sigma}{d\Omega d\epsilon'} =
\sigma_M \left\{ v_L R_L(q,\omega)
+ v_T R_T(q,\omega)\right \}
\label{cross}
\end{equation}
where $\sigma_M$ is the Mott cross section, $q_{\mu}^2=\omega^2-q^2$
the squared four--momentum transfer, $v_L=(q_{\mu}^2/q^2)^2$,
$v_T=\left[\tan^2\left(\theta/2\right) - q_{\mu}^2/2q^2\right]$ and
$R_{L,T}$ the longitudinal and transverse (separated) electromagnetic
nuclear response functions. Relevant physical issues are connected with
these response functions, in particular the so--called {\it Coulomb
sum rule} (from the longitudinal response integrated over the energy)
and the scaling properties, both in non--relativistic and in relativistic
regimes.
\item[$\diamondsuit$]
{\sl Inclusive scattering of polarized electrons (off unpolarized
targets)}
From the difference between the ${\vec e}$--nucleus cross sections,
with polarization of the electron parallel and antiparallel to the
incident beam it is possible to measure the {\it asymmetry}:
\begin{eqnarray}
{\cal A} &=&
\frac{d^2\sigma^+ - d^2\sigma^-}{d^2\sigma^+ + d^2\sigma^-}
\nonumber\\
&\equiv&
{\cal A}_o\frac{ v_L R_{AV}^L(q,\omega) +
v_T R_{AV}^T(q,\omega) +v_{T'} R_{VA}^{T'}(q,\omega)}
{v_L R_L(q,\omega) + v_T R_T(q,\omega)}
\label{asymm}
\end{eqnarray}
which contains, in the numerator, the parity violating (PV) nuclear
response functions, stemming from the interference between the nuclear
electromagnetic and weak (vector and axial) neutral currents.
In the above
\begin{equation}
{\cal A}_o =\frac{G_F Q^2}{2\pi\alpha\sqrt{2}}
\approx 3.1\times 10^{-4}\tau \qquad\quad
\left(\tau=\frac{Q^2}{4M^2}\right)
\end{equation}
where $G_F$ is the Fermi constant, $\alpha$ the electromagnetic coupling
constant, $Q^2=-q_{\mu}^2 >0$ and $v_{T'}=\sqrt{\tan^2\left(\theta/2\right)
- (q_{\mu}^2/q^2)\tan(\theta/2)}$. It is worth mentioning that the PV
response functions can be sensitive to novel aspects of the nuclear
dynamics, different from the ones explored by the electromagnetic vertex.
In particular they give access to the isoscalar NN force and to the
neutrons' distribution in nuclei.
\item[$\diamondsuit$]
{\sl Exclusive versus inclusive electron scattering}
One--exclusive processes, like $(e,e',N)$, in which one nucleon is
observed in coincidence with the outgoing electron, are more sensitive
tests of the nuclear models. Indeed the probability of a
nucleon to be emitted strongly reflects, for example, the momentum
distribution inside the nucleus; striking differences are obviously
found by employing models like a realistic shell model, the so--called
Hybrid model or even the Fermi Gas model (the latter being, manifestly,
not appropriate to deal with the details of nuclear structure).
A variety of nuclear response functions can be separated by an
adequate selection of the kinematics; in particular some interference
between the longitudinal and transverse components of the electromagnetic
current contribute to the one--exclusive cross sections, at variance
with the inclusive ones. Thus one can test different components and
matrix elements of the nuclear electromagnetic current, including
delicate questions like the off--shellness of the nucleonic current
in the nucleus, the contribution of meson exchange currents, the
limitations of non--relativistic approaches.\cite{Amaro98,Cenni97}
Finally $(e,e',N)$ is fairly sensitive to the strong interaction of the
ejected nucleon with the residual nucleus, the so--called final state
interaction (FSI), which can be responsible for a sizeable distortion
of the outgoing nucleon wave. Recently a semiclassical approach has
been proposed\cite{Chanfray} to
describe this specific aspect: it will be shortly
mentioned here, since it provides a natural bridge between nuclear
matter and finite nuclei. The main research activity on the electromagnetic
interaction in light nuclei is reported elsewhere\cite{Ciofi}.
\end{description}
Coming back to the main subject of this report, we shall now concentrate
on the different theoretical methods which have been employed for an
accurate evaluation of the electromagnetic response functions in
nuclear matter. They can be roughly grouped into three categories:
\begin{enumerate}
\item
The variational method, which is based on the Fermi Hypernetted
Chain (FHNC) scheme
\item
The functional method, which employs the path--integral approach
for developing a consistent bosonic loop expansion
\item
The traditional perturbation theory, which encompasses the most
frequently utilized models, from the Fermi gas (non--relativistic
and relativistic), to the Hartree--Fock (HF) model combined together
with the important correlations provided by the Random Phase
Approximation (RPA).
\end{enumerate}
In concluding this section we briefly mention the case of neutral
current (NC)
neutrino-- and antineutrino--nucleus scattering: these processes
offer the possibility to extract information on the strange form factors
of the nucleon. For this purpose two quantities have been considered,
the $\nu-{\bar\nu}$ asymmetry\cite{Bai}:
\begin{equation}
{\cal A}_p = {\displaystyle
\frac{\left(\sigma\right)_{\nu p\rightarrow \nu p}^{NC} -
\left(\sigma\right)_{{\bar\nu}p\rightarrow {\bar\nu}p}^{NC} }
{\left(\sigma\right)_{\nu n\rightarrow \mu^- p}^{CC} -
\left(\sigma\right)_{{\bar\nu} p\rightarrow \mu^+ n}^{CC} }
}\,
\label{nuasym}
\end{equation}
and the ratio of the cross sections for inelastic
NC scattering of $\nu({\bar\nu})$ on nuclei, with the emission of a
proton or, respectively, of a neutron:
\begin{equation}
{\cal R}^{\nu(\bar\nu)}_{p/n}=
\frac{\displaystyle\left( \sigma \right)^{NC}_{\nu({\bar\nu}),p} }
{\displaystyle \left(\sigma\right)^{NC}_{\nu({\bar\nu}),n} }
\, .
\label{ratio}
\end{equation}
Calculations of both (\ref{nuasym}) and (\ref{ratio}) have been
performed, in an energy range from 200 MeV to 1 GeV, within two
relativistic independent particle models (Fermi gas and shell model);
the final state interactions of the ejected nucleon have been taken
into account through relativistic Optical model
potentials.\cite{Sam1,Sam2,Sam3}.
While the values of the cross sections significantly depend on the
nuclear model (especially in the lower energy range),
the NC/CC neutrino--antineutrino asymmetry and the ratio p/n
show a rather mild dependence on the model and allow one to disentangle
different values of the strangeness parameters entering into the weak NC
axial and vector strange form factors of the nucleon.
We remind that the $Q^2=0$ limit of the axial strange form
factor, $g_A^s$, is closely related to the problem of the proton spin,
a long lasting puzzle raised by the measurements, in the deep inelastic
polarized lepton scattering, of the polarized structure function $g_1$
of the proton. Moreover the information on the strange form factors
of the nucleon extracted fron neutrino scattering is complementary
to the similar information, which can be obtained in the above quoted
PV polarized electron scattering experiments.
\section{ Electromagnetic response functions}
\subsection{Variational method}
The inclusive {\it transverse response function} of symmetric
nuclear matter at saturation density, $R_T(q,\omega)$, has been studied
by Fabrocini\cite{Fab1} within the Correlated Basis Function (CBF)
perturbation theory. The main goal of this work is to ascertain how
$R_T$ is affected by the NN correlations, with a special emphasis
on the MEC contributions.
CBF calculations are based upon a set of correlated wave functions
\begin{equation}
|n>={\cal S}\left[\prod_{i,j} f(i,j)\right]|n>_{FG}
\label{CBFstate}
\end{equation}
obtained by applying a symmetrized product of two--body correlation
operators, $f(i,j)$, to the FG states $|n>_{FG}$. The following
effective structure is adopted for $f$:
\begin{eqnarray}
&&f(i,j)=\sum_{q=1,6} f^{(q)}(r_{ij}) O_{ij}^{(q)},
\label{fij}\\
O_{ij}^{(q=1,6)}&&= (1,{\sigma}_i\cdot{\sigma}_j, S_{ij})
\otimes(1,{\tau}_i\cdot{\tau}_j)
\label{operij}
\end{eqnarray}
$S_{ij}$ being the tensor operator; $f(i,j)$ depends upon a set of
variational parameters, which are fixed by minimizing the expectation
value of a realistic, non--relativistic Hamiltonian on the correlated
ground state. The g.s. energy is calculated via the FHNC cluster
summation technique, using the correlation corresponding to the
Argonne $V_{14}$~+~Urbana~VII three--nucleon interaction model of
ref.\cite{Adel21}.
The Jastrow (scalar) component of (\ref{operij}) accounts for the
short range NN repulsion, while, among the remaining correlators, the
most relevant are the spin--isospin and the tensor--isospin ones,
which mostly stem from the one--pion exchange (OPE) long range part
of the potential.
Configurations up to correlated one particle--one hole (1p--1h)
intermediate states
are considered; the spreading due to the decay of particle (hole)
states into 2p--1h (2h--1p) states is taken into account via a
realistic optical potential model.
The transverse response is given by
\begin{equation}
R_T(q,\omega) =\frac{1}{A}\sum_n \left\vert<0|{\vec j}({\vec q})|n>
\right\vert^2\delta(\omega-\omega_n)
\label{RTadel}
\end{equation}
where the sum goes over the intermediate excited states $|n>$, with
excitation energy $\omega_n$; ${\vec j}({\vec q})$ is the
electromagnetic current operator
\begin{equation}
{\vec j}({\vec q}) = {\vec j}^{(1)}({\vec q}) + {\vec j}^{(2)}({\vec q})
\label{curr}
\end{equation}
sum of the one--body, magnetic current [$ {\vec j}^{(1)}({\vec q}$] and
the two--body exchange currents [${\vec j}^{(2)}({\vec q})$]. For the
latter, the Schiavilla--Pandharipande--Riska model\cite{SPR} is adopted,
which satisfies, by construction, the continuity equation with realistic
$V_{14}$ Argonne and Urbana potentials. Currents due to intermediate
$\Delta$--isobar excitations are also included. The sharp energy
boundaries of the 1p--1h FG response are smoothed out by an appropriate
folding with a width $W(\omega)$:
\begin{equation}
R_T(q,\omega) =\frac{1}{\pi}\int\,d\omega'\, R_T^{1p-1h}(q,\omega')
\frac{W(\omega')}{(\omega-\omega')^2+[W(\omega')]^2},
\label{RTwidt}
\end{equation}
where $W(\omega)={\mathrm Im} W_o(\omega)/M^*$, $W_o$ being
the optical potential and $M^*$ the nucleon effective mass.
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=17.5cm \epsfbox{w3.eps2}
}
\caption[ ]{ Transverse response at $q=300$ (a), 380 (b), and 570 (c)
MeV/c for $^{40}$Ca and nuclear matter. See text. }
\label{fig1}
\end{figure}
Fig.~1 shows the comparison of the nuclear matter response evaluated in
ref.\cite{Fab1} and the experimental data on $^{40}$Ca; both the impulse
approximation (IA) and the full calculation (MEC+IA) are shown. The data
are taken from from ref.\cite{Adel4} ($\times$), ref.\cite{Adel6} (circles)
and ref.\cite{Adel9} (black circles).
The global contribution of the two--body currents turns out to be
positive and provides an enhancement of the one--body response,
ranging from $\sim 20\%$ for the lower momenta (300~MeV/c) to $\sim 10\%$
for the higher ones (about 500 MeV/c). The tensor--isospin component
of the correlation is crucial to obtain these results.
Several variational calculations have been performed in the past for the
{\it longitudinal response function} as well, both in nuclear matter and
light nuclei. In a recent work Amaro {\it et al.}\cite{Fab2} have
tested a simple approximation to deal with the short range correlations
(SRC) affecting the 1p--1h
excited states, which build up the nuclear response functions.
The basic idea of the model (previously tested for the ground state
properties) consists in truncating the CBF expansion in
such a way to retain only those terms containing a single Jastrow--type
correlation line,
\begin{equation}
h(r)=f^2(r)-1\, .
\label{Jastrow}
\end{equation}
The expansion in powers of $h(r)$ has the property of conserving
the proper normalization of the correlated many--body wave function.
The nuclear charge is conserved as well, at variance with other
truncation schemes. The correlation employed is the scalar component
of a complicated state dependent correlation, fixed to minimize the
nuclear binding energy in a FHNC calculation with the Urbana $V_{14}$
NN potential.
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=11cm \epsfbox{w4.eps2}
}
\caption[ ]{Diagrams considered in the model of ref.\cite{Fab2}. The dotted
lines represent the correlation function. The oriented lines represent
particle and hole wave functions. The black circle indicates an integration
point, while the black squares indicates the integration point, where
the charge operator is acting. }
\label{fig2}
\end{figure}
Fig.~2 shows the diagrams retained in \cite{Fab2}. The
comparison between the calculation of the longitudinal response
function in the full FHNC and in the proposed approximation shows
that the difference between the corresponding results does not exceed a
few parts $\times 10^{-4}$, in a fairly large interval of momentum
transfers. This proves the reliability of the truncated scheme also
for dealing with 1p--1h excited states; it needs to be further tested,
however, for the 2p--2h responses.
\subsection{Functional (path integral) method}
The path integral approach offers a consistent theoretical foundation
of the nuclear response functions. Indeed it employs the approximation
schemes of QFT, namely the perturbative approach and the loop
expansion, to provide a fully consistent criterium of selecting
classes of diagrams, still preserving the fundamental symmetries and
invariances of the choosen model Lagrangian.
Starting from a Lagrangian which describes interacting nucleons,
pions and eventually $\rho$ mesons, all coupled to an external
electromagnetic field, one derives the nuclear
polarization propagator and response functions through
derivatives of the generating functional
\begin{eqnarray}
Z[j_\mu^{ext}] &&= \frac{1}{{\cal N}}\int {\cal D}\left[ {\bar\psi},
\psi,\Phi,A_\mu\right] \exp\left\{i\int dx[j_\mu^{ext}A_\mu]\right\}
\nonumber\\
&&\times\exp\left\{i\int dx\left({\cal L}+{\cal L}_{em}+J_\mu A^\mu
+B_{\mu\nu} A^\mu A^\nu\right)\right\}
\label{funZ}
\end{eqnarray}
where $\psi,{\bar\psi}$ are the nucleon fields, $\Phi$ and $A_{\mu}$,
respectively, the pion and electromagnetic fields and
$\j_{\mu}^{ext}$ the external source of the latter. In the above
\begin{equation}
{\cal L}= {\bar\psi}(i\gamma^\mu\partial_\mu-M)\psi +
{\scriptstyle\frac{1}{2}}(\partial_\mu \Phi)^2 -\frac{m_\pi^2}{2}\Phi^2
-ig{\bar\psi}\gamma_5\tau\psi\cdot\Phi
\label{funLag}
\end{equation}
is the model Lagrangian (without the $\rho$ meson) and
\begin{eqnarray}
J_\mu &=& e{\bar\psi}{\scriptstyle\frac{1}{2}}(1+\tau_3)\gamma_\mu\psi +
e[\Phi\times\partial_\mu\Phi]_3
\label{funcurrent}\\
B_{\mu\nu} &=& e^2g_{\mu\nu}\Phi^+\Phi^-
\label{fun2pion}
\end{eqnarray}
the required couplings of nucleons and pions to the e.m. field.
The results obtained, within this formalism, for the longitudinal and
transverse responses in the inclusive electron scattering are shown in
the work by Cenni {\it et al.}\cite{Cenni1}
It is worth mentioning that an important connection between the
variational technique and the path integral approach has been recently
explored in ref.\cite{Cenni2}. These authors suggest a new method, based on
the variation, within a path integral framework, of a trial Hamiltonian:
it turns out that a particular choice of the latter
corresponds exactly to the use of a Jastrow correlated Ansatz for the
wave function in the FHNC approach. Thus the new formalism generalizes
the FHNC and CBF techniques, allowing for their extension to
relativistic and field theoretical problems.
\subsection{Perturbation theory}
Several approaches based on perturbative schemes have been applied to
the evaluation of nuclear response functions both for inclusive
(parity conserving and parity violating) and exclusive processes. Most
calculations are based on non--relativistic Hamiltonians and currents,
but attempts to include relativistic effects, especially in the
electromagnetic currents and vertices, have been done.
Many--body perturbative techniques have been developed to
treat quasielastic electron scattering in nuclear matter.
The Green's function method is particularly suited to express
inelastic cross sections through the imaginary part of the polarization
(particle--hole) propagator; nuclear matter results compare fairly
well with the experimental data for energy transfers in the region of
the quasielastic peak and above, providing the $\Delta_{33}$ resonance
and mesonic degrees of freedom are incorporated in the
nuclear matter polarization propagators.
The role played by the NN interaction in reshaping the nuclear
matter response functions with respect to the pure FG one has been widely
investigated, by employing both phenomenological effective interactions
and G--matrix parameterizations based on the one--boson exchange
potential. Starting from the naive Fermi gas description, one can easily
obtain the Hartree--Fock (HF) or Brueckner--Hartree-Fock (BHF)
polarization propagators; the next step is then to include RPA
correlations, both in the ``minimal'' version of the so--called
{\sl ring} diagrams and in the framework of the fully antisymmetrized
RPA.
Numerical calculations of the HF+RPA response functions can require
large computing time. Indeed, although in nuclear matter the
ring approximation for the polarization propagator
can be analytically evaluated, as soon as
one dresses the particle and hole propagators with HF self--energies,
even the simple HF p--h propagator, $\Pi_{HF}(q,\omega)$, requires
numerical integrations when the HF self--energy is derived from any
realistic NN interaction.
Moreover the simple algebraic equation for the ring polarization
propagator turns into an infinite series of complex integrals when
the exchange matrix elements of the p--h interaction are taken into
account in the fully antisymmetrized RPA, which is diagrammatically
shown in Fig.~3. An approximate, but reliable, treatment of
these contributions is thus compulsory.
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=5cm \epsfbox{a2.eps}
}
\caption[ ]{Diagrammatic representation of the perturbative expansion
for the polarization propagator in RPA. The dashed line represents the
particle--hole interaction $V$. }
\label{fig3}
\end{figure}
This has been achieved with the method of the {\sl continued fraction}
(CF) expansion\cite{Arturo}, which will be briefly sketched in the
following. The CF--like expansion for the polarization propagator can
be written as
\begin{equation}
\Pi^{RPA}={\displaystyle\frac{\Pi^{(0)}}
{\displaystyle 1-A-\frac{\displaystyle B}{\displaystyle 1-C-
\frac{D}{1-\dots}}}}
\label{Picf}
\end{equation}
where $\Pi^{(0)}$ is the free (or HF) particle--hole propagator and
all symbols appearing in (\ref{Picf}) are functions of both the energy
and momentum transfers. The CF approach at n--th order reproduces
exactly the perturbative series expressing the antisymmetrized RPA
at the same order while it approximates higher orders.
Setting, for example,
\begin{equation}
\Pi^{RPA} =\sum_{n=0}^{\infty}\, \Pi^{(n)}
\label{pisum}
\end{equation}
the first order in CF provides the following approximation for
$\Pi^{(n)}$:
\begin{equation}
\Pi^{(n)}\simeq \Pi^{(0)}\left[\frac{\Pi^{(1)}}{\Pi^{(0)}}
\right]^n,
\end{equation}
where $\Pi^{(1)}\equiv \Pi^{(0)}4V\Pi^{(0)} +\Pi^{(1)ex}$ is the sum
of the direct and exchange first order terms. With this approximation
the summation in (\ref{pisum}) is trivial, yielding
\begin{equation}
\Pi^{RPA}_{CF1} =
\frac{\Pi^{(0)}}{1- \Pi^{(1)}/\Pi^{(0)}}
=\frac{\Pi^{(0)}}{1- 4V\Pi^{(0)}- \Pi^{(1)ex}/\Pi^{(0)}}\, .
\label{Picf1}
\end{equation}
The next step extends the approximation to the exact second order
term; the corresponding RPA propagator reads:
\begin{eqnarray}
\Pi^{RPA}_{CF2} &=&
\frac{\Pi^{(0)}} {1- \Pi^{(1)}/\Pi^{(0)} -
\{ \Pi^{(2)}/\Pi^{(0)} - [\Pi^{(1)}/\Pi^{(0)}]^2 \} }
\label{Picf2}\\
&=&\frac{\Pi^{(0)}}{1- 4V\Pi^{(0)}- \Pi^{(1)ex}/\Pi^{(0)}
- \{ \Pi^{(2)ex}/\Pi^{(0)} - [\Pi^{(1)ex}/\Pi^{(0)}]^2 \} }
\, ,
\nonumber
\end{eqnarray}
where the approximation provided by the first order CF expansion
has been subtracted off.
Numerical calculations for the longitudinal and transverse responses,
using a G--matrix based on the Bonn potential, show that the CF
expansion rapidly converges, giving a few percent correction between
the first and the second order CF results. On the contrary,
the differences between ring and RPA (in the CF1 or CF2 approximation)
responses appear to be sizeable, both in the longitudinal and in the
transverse channel, with, perhaps,
the exception of the spin--isovector channel, as it is illustrated in
Fig.~3.
We also remind here that a very useful (and accurate) approximation,
consisting in a ``bi--parabolic'' parameterization of the HF self--energy
of the nucleon\cite{Bai96}, allows to express analytically $\Pi_{HF}$, thus
considerably reducing the computational time.
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=11cm \epsfbox{a1.eps}
}
\caption[ ]{Nuclear matter responses for $k_F=195$~MeV/c at $q=500$~MeV/c:
the dotted line corresponds to the free response, the dashed one to the
ring approximation and the solid line to the RPA--CF1 response. The
kinematics is relativistic and the particle--hole interaction is derived
from a realistic G--matrix. }
\label{fig4}
\end{figure}
Within the perturbative approach it is also possible to deal with
{\it relativistic effects}, which might become important for energy/momentum
transfers larger than 0.5~GeV or so. Besides the corrections stemming from
the relativistic kinematics and vertices, which have been completely
taken into account in the framework of the Fermi gas model,
several investigations, based on a fully
relativistic Lagrangian with nucleonic and pionic degrees of freedom,
both coupled to the external electromagnetic field, have been performed.
These approaches
usually include both the single--nucleon e.m. current and the two--body
MEC and account for relativistic effects within an expansion in terms
of suitably ``small'' variables, like the energy/momentum transfers in
units of twice the nucleon mass ($\kappa=q/2M, \lambda=\omega/2M$).
A new relativistic expansion for the matrix elements of one-- and
two--body e.m. currents has been recently proposed by Amaro
{\it et al.}\cite{Amaro98} in terms of the parameters $\eta_i\equiv p_i/M$
(characteristically of the order of 1/4),
$\{p_i\}$ being the initial--state nucleon momentum inside the nucleus:
clearly this formalism will be especially suited in the regime of GeV
energies, where, on the contrary, $\kappa$ and $\lambda$ are no longer small
with respect to 1.
\subsubsection{y--scaling and Coulomb sum rule}
\par
The concepts of y--scaling and Coulomb sum rule have been throughly
discussed in the past since they offer important tests for the
accuracy and reliability of the nuclear model description
underlying the calculation of the electromagnetic response functions.
Both of them require a careful investigation in order to
disentangle, from the full expression of the longitudinal and
transverse response functions, those dividing factors such that the
reduced responses have scaling properties or (for what concerns the
longitudinal response) satisfy the Coulomb sum rule. This task is
not trivial since, as shown by the Relativistic Fermi Gas (RFG)
model, the e.m. form factors to be divided by
are not simply factorizable, due to the structure of the relativistic
one--nucleon current.
In the work by Barbaro {\it et al.}\cite{Bai98} the dividing factors
$G_L,\,G_T$ are constructed in the RFG for the longitudinal and
transverse response functions in such a way that the reduced responses,
so obtained, scale; another dividing factor, $H_L$, yields a (different)
reduced response which fulfills the Coulomb sum rule (CSR). The relevant
point here is that $G_L,\,G_T$ and $H_L$ are all found to be only
weakly model--dependent, thus providing essentially universal
dividing factors.
It becomes thus possible to test whether specific models scale and
satisfy the CSR. This investigation has been carried out for
the Hybrid model\cite{Cenni97} and the Quantum Hadrodynamic
model\cite{Walecka} (QHM), each one with different
types of interaction effects which go beyond the strict RFG.
It is found that while the Hybrid model scales and obeys the CSR,
the QHM does neither. The Coulomb and higher order sum rules have
also been investigated by Amore {\it et al.}\cite{Amore} in the RFG,
by employing two different methods, namely by exploiting the scaling
properties of the longitudinal response function and by enforcing
the completeness of the states in the space--like domain via the
Foldy--Wouthuysen transformation.
\subsection{Exclusive versus inclusive electron scattering scattering}
The connections between exclusive and inclusive electron--nucleus
scattering within the framework of the plane--wave impulse
approximation (PWIA) have been investigated in a work by Cenni {\it
et al.}\cite{Cenni97}. These authors test the interplay between the (model
independent) kinematical constraint and the (model
dependent) features of the spectral
function in providing the exclusive (and inclusive) nuclear responses.
The RFG and the Hybrid model are employed to provide a link between
finite and infinite Fermi systems and to assess the impact of the
confinement of the struck nucleons on the inclusive charge response.
One of the important outcomes of this work is that, with an energy--shift
and rescaled Fermi momentum, an effective RFG response can be obtained
whose first three energy--weighted moments agree quite well with the
analogous quantities evaluated within the (confined) Hybrid model.
A more recent work\cite{Chanfray} evaluates semi--classically
the exclusive $(e,e',p)$ cross sections, yet partially preserving
the simplicity of nuclear matter calculation, but going beyond the
PWIA. Indeed this approach allows to include the distortion of the
outgoing nucleon wave (the so--called final state interaction, FSI)
at least within the mean field approximation.
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=5.5cm \epsfbox{fig5.ps}
}
\caption[ ]{ }
\label{fig5}
\end{figure}
Sticking to the IA, the cross section for the $(e,e',p)$ scattering
process, which is schematically illustrated in Fig.~5, reads:
\begin{equation}
\frac{d^4\sigma}{d\epsilon' d\Omega_e dE_{\scriptstyle N} d\Omega_{\scriptstyle N}}
= E_{\scriptstyle N} p_{\scriptstyle N} \left(\frac{d\sigma}{d\Omega_e}\right)_{e{\scriptstyle N}}
S({\vec p}_m,{\cal E}).
\label{exclcr}
\end{equation}
In the above:
\begin{equation}
\left(\frac{d\sigma}{d\Omega_e}\right)_{e{\scriptstyle N}}=\sigma_{Mott}
\frac{1}{\cos^2(\theta/2)}\frac{m_e^2}{\epsilon\epsilon'}
\eta^{\mu\nu}\frac{M^2}{E({\vec p})E_{\scriptstyle N}}
W_{\mu\nu}({\vec p},{\vec p}+{\vec q})
\label{sncr}
\end{equation}
is the single nucleon cross section, while
\begin{equation}
S({\vec q},q_0)=\frac{1} {(2\pi)^3} \frac{M} {E({\vec q})}
\frac{<A|{\hat a}^{\dagger}_q
\delta\left( q_0 - {\hat{\cal H}}-\mu\right){\hat a}_q|A>}{<A|A>}
\label{spectral}
\end{equation}
is the nuclear spectral function, which in (\ref{exclcr}) depends upon
the missing momentum ${\vec p}_m={\vec q}-{\vec p}_{\scriptstyle N}$ and the missing
energy ${\cal E}=E_{\scriptstyle N}-\omega$ (see Fig.~5 for the other symbols).
Different nuclear models generally provide different spectral functions,
as one can see, for example, by comparing the ones obtained in the RFG,
\begin{equation}
S^{RFG}({\vec p},{\cal E})=\frac{\Omega} {(2\pi)^3}
\theta(k_F-p)\delta\left\{{\cal E}-\sqrt{k_F^2+M^2}-
\sqrt{p^2+M^2}\right\}
\label{sfRFG}
\end{equation}
and in the Hybrid model (harmonic oscillator bound states combined
with a continuum of unbound states):
\begin{equation}
S^{HM}({\vec p},{\cal E})=\sum_{N=0}^{N_{max}}
\delta\left\{ {\cal E}-(N_{max}-N)\omega_0\right\} n_{N}(p),
\label{sfHM}
\end{equation}
where
\begin{equation}
n_{N}(p)=\sum_{n,\ell=N-2n}\frac{2\ell+1}{4\pi}
|\varphi_{n\ell}(p)|^2
\end{equation}
is the momentum distribution for the N--th shell. The spectral
function (\ref{sfHM}) is illustrated in Fig.~6, where a dashed line
is drawn in correspondence with the support for the spectral function
of the RFG, eq.~(\ref{sfRFG}), which is infinite along that line and
zero elsewhere.
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=9cm \epsfbox{w2.eps2}
}
\caption[ ]{The spectral function in the Hybrid Model. }
\label{fig6}
\end{figure}
Within the semiclassical framework, the Wigner transform of the
spectral function reads:
\begin{equation}
\left[ S({\cal E})\right]_W(R,{\vec p}) =
\theta\left[\epsilon_F -\frac{p^2}{2M^*}-V(R)\right]
\delta\left\{ {\cal E}-\left[\epsilon_F-\frac{p^2}{2M^*}-V(R)
\right]\right\}
\label{sfsemicl}
\end{equation}
where $V(R)$ is a suitable shell model potential (e.g. the harmonic
oscillator or the Woods--Saxon potentials) and $\epsilon_F\equiv
k_F^2(R)/2M^* +V(R)$ the local Fermi energy.
It is shown in ref.\cite{Chanfray} that in the semiclassical approach
it is possible to account for the distortion of the ejected nucleon
(illustrated by the black dot in Fig.5) in a relatively simple way:
an appropriate distortion operator is introduced, for which
two heuristic assumptions are made, corresponding to the eikonal
and to the uniform approximations, respectively.
The corresponding exclusive cross sections turn out to
be quite different, hence stressing the relevance of the FSI in the
analysis of the exclusive processes.
\section{Strange probes and strange matter}
\subsection{$K^+$--nucleus scattering}
The quasielastic $K^+$--nucleus scattering has been thoroughly
investigated by De Pace {\it et al.}\cite{Art1,Art2,Art3}.
This subject shares several motivations with the items in the
previous sections. The first one relies on the relatively small
(as compared with strong interaction processes)
$K^+ N$ cross sections: as a consequence the $K^+$ projectiles
can penetrate deeply inside the nucleus, similarly to electrons
and photons, thus allowing to investigate collective effects in
the nuclear response. The second reason of interest is the
possibility of exploring scalar--isoscalar excited states, which
are the dominant channel for the $K^+$--nucleon coupling. This
channel can be partly explored in the longitudinal electron scattering
response functions, but an isospin separation in addition to the
customary longitudinal--transverse one has never been, till now,
carried out in $(e,e',X)$ experiments.\footnote{Such a separation
should become possible with PV electron scattering experiments and
it will be of extreme interest to compare the latter with the $K^+$
quasielastic scattering.}
Finally the observed excess in elastic $K^+$--nucleus scattering
cross sections seems to require, to be explained, some enhancement
induced by the nuclear medium itself.
A simple phenomenological approach, which expresses the above
mentioned cross sections by means of an effective number of
participant ($N_{eff}$),
\begin{equation}
\frac{d^2\sigma}{d\Omega d\omega} = N_{eff}
\frac{d\sigma_{K^+N} }{d\Omega} R(q,\omega),
\label{Neff}
\end{equation}
seems to require values of $N_{eff}$ too large with respect to
what expected in the Glauber theory; moreover (\ref{Neff}) fails
in describing the observed collective phenomena at low momentum
transfer.
The model proposed in \cite{Art1} allows to evaluate the nuclear
response $R(q,\omega)$ within a continuum RPA, using an effective
G--matrix particle--hole interaction and a Woods Saxon mean field,
complemented by an appropriate spreading width of the ph states.
The kinematics is relativistic. An improved Glauber theory is used
for the reaction mechanism, including one-- and two--step
contributions.
\begin{figure}
\centerline{
\epsfxsize=12cm
\epsfysize=8cm \epsfbox{a3.eps}
}
\caption[ ]{The $K^+ - ^{40}$Ca cross sections at $q=290$~MeV/c
(left) and $q=480$~MeV/c (right); data are taken from ref.\cite{Korman}.
For the explanation of the theoretical curves see text. }
\label{fig7}
\end{figure}
In Fig.~7 we show an example of the results obtained in this
framework: the experimental data for $K^+-^{40}$Ca scattering are
compared with the 1 step (free)+ 2 step calculation (dotted line),
the 1 step (using the $N_{eff}$ formula)+ 2 step (dot-dashed line)
and the 1 step (RPA)+ 2 step (dashed line). While the free response
is clearly inadequate to reproduce the data, the other two calculations
fail in reproducing the excess cross section appearing in the low
energy region at small momentum transfers. However if one empirically
reduces by about 50\% the effective ph interaction in the scalar
isoscalar channel, then the 1 step (RPA)+ 2 step turns out to be in
very good agreement with the experimental data, in the whole range
of momentum transfers and nuclei ($^{12}$C and $^{40}$Ca) explored.
This is a new, challenging information on the ph force acting
in this channel.
\subsection{Hypernuclei (structure and decay)}
In the previous paragraph we have considered the interaction of a
strange particle, the $K^+({\bar s}u)$, with the nuclear medium.
Should a $K^-(s{\bar u})$ hit a nucleus, most of the times
it gets absorbed by a nucleon, leading to the formation of a
hypernucleus, namely a nucleus containing one (or more) hyperon $(sqq)$.
The lightest and more easily produced hyperon is the $\Lambda$ particle
and a great interest is focussed on the investigation of the ground
state properties and decay mechanisms of $\Lambda$--hypernuclei.
Several experimental data are already
available on light and medium hypernuclei,
but a reacher set of measurements are expected from the FINUDA experiment,
which is just starting operating in the Frascati National Laboratory of INFN.
The {\bf weak decay width} of $\Lambda$--hypernuclei can be evaluated with
the Green's function (polarization propagator) method, which is
conveniently implemented in nuclear matter and extended to finite
systems through the so--called Local Density Approximation (LDA).
The width of the hypernucleus is derived from the $\Lambda$ self--energy
\begin{equation}
\Gamma_{\Lambda}(k,\rho)= -2{\mathrm Im}\Sigma_{\Lambda}(k,\rho)
\label{GammaL1}
\end{equation}
after integrating over the $\Lambda$ momentum distribution.
Three different mechanisms contribute to the weak decay of a $\Lambda$
in the nucleus:
\begin{itemize}
\item the {\it mesonic decay},
\[ \Lambda\longrightarrow \pi N \qquad (\Gamma_{M}), \]
which also occurs in free space and is relevant only in light hypernuclei,
since the Pauli blocking prevents the low--momentum outgoing nucleon to
be emitted in a medium--heavy nucleus;
\item the {\it two--body non--mesonic (NM) decay}
\[ \Lambda N \longrightarrow NN \qquad (\Gamma_1), \]
\item the {\it three--body non--mesonic decay}
\[ \Lambda NN \longrightarrow NNN \qquad (\Gamma_2). \]
\end{itemize}
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=8cm \epsfbox{hyper1.eps}
}
\caption[ ]{Lowest order terms for the $\Lambda$ self--energy in
nuclear matter. }
\label{fig8}
\end{figure}
A new evaluation of the $\Lambda$ self--energy has been carried
out\cite{Garba}, which includes the 1p-1h and 2p-2h polarization
propagators (related to $\Gamma_1$ and $\Gamma_2$, respectively)
in a RPA scheme. Fig.~8 illustrates a few examples
of the diagrams whose imaginary part is contributing to the
hypernuclear decay width.
The particle--hole interaction embodies the
exchange of $\pi$ and $\rho$ mesons, together with short range
repulsive correlations described by the two Landau--Migdal
parameters, $g'$ and $g'_{\Lambda}$ (the latter being used when
one vertex of the exchanged pion coincides with the weak $\Lambda\pi N$
vertex). The nuclear matter calculation is adapted to finite nuclei
using the LDA, which amounts to replace the constant Fermi momentum
with a local one, expressed by $k_F(r)=[3\pi^2\rho(r)/2]^{1/3}$,
$\rho(r)$ being the nuclear density. The width associated to the decay
of a $\Lambda$ with momentum $k$ is then
\begin{equation}
\Gamma_{\Lambda}({\vec k})=\int d{\vec r} |\psi_{\Lambda}({\vec r})|^2
\Gamma_{\Lambda}[{\vec k},\rho(r)]
\label{GammaL2}
\end{equation}
where $\psi_{\Lambda}({\vec r})$ is the $\Lambda$ wave function in the
nucleus. The results one obtains turn out to be
very sensitive to the latter, the best choice corresponding to
a Woods Saxon potential well which reproduces the measured $s$ and
$p$ energy levels of the $\Lambda$ in the nucleus.
Finally the total width of the hypernucleus is obtained by
integrating (\ref{GammaL2}) over the momentum distribution of the
$\Lambda$ itself:
\begin{equation}
\Gamma_{\Lambda}=\int d{\vec k}\, |A_{\Lambda}({\vec k})|^2\,
\Gamma_{\Lambda}({\vec k}).
\label{GammaL3}
\end{equation}
The results obtained within this framework are illustrated in Fig.~9,
where the decay width of several hypernuclei (from the ``light'' $^{12}$C
to the heaviest $^{209}$Bi and $^{238}$U) is shown, in units of the free
$\Lambda$ decay width, $\Gamma_{\mathrm free}$, as a function of the mass
number. The separate contributions, $\Gamma_M$, $\Gamma_1$ and $\Gamma_2$,
are also displayed.
\begin{figure}
\centerline{
\epsfxsize=10cm
\epsfysize=10cm \epsfbox{hyper2.eps}
}
\caption[ ]{ $\Lambda$ decay widths in finite nuclei as a function of
the mass number $A$; the labels of the various curves correspond to:
the mesonic width ($\Gamma_M$), the one-- ($\Gamma_1$) and two--body
($\Gamma_2$) induced decay widths, the total non--mesonic width
($\Gamma_{NM}=\Gamma_1+\Gamma_2$) and the total ($\Gamma_T$), sum
of all independent contributions. }
\label{fig9}
\end{figure}
The theoretical results are in good agreement with the data over the
whole hypernuclear mass range explored. The saturation property of the
$\Lambda N\to NN$ interaction in nuclei clearly appears in the
flattening of the decay width as the mass number increases. In spite of
these satisfactory results, there remains to be explained the large
experimental value of the ratio $\Gamma_n/\Gamma_p$, which does not seem
to be in agreement with the present theoretical schemes.
Concerning the {\bf ground state properties of hypernuclei} we should
mention here a research project by Co {\it et al}\cite{Co99}, who intend
to investigate the structure of hypernuclei within the FHNC scheme.
Starting from a suitable basis of single particle wave functions
(Woods--Saxon and/or HF wave functions with finite range interaction)
a Jastrow correlated many--body state will be obtained by developing
a FHNC formalism for a system of A nucleons plus one impurity (the
hyperon). Binding energy and first excited states can also be calculated.
\section{Asymmetric nuclear matter}
\subsection{Equation of state of asymmetric matter}
Nuclear matter is usually considered as an isospin symmetric system,
$Z=N=A/2$, where the Coulomb interaction between protons is switched off.
In reality heavier nuclei are more and more isospin asymmetric, the
larger number of neutrons being there to balance the increasing Coulomb
repulsion. An interesting, ideal extension of these systems is the
asymmetric nuclear matter, where the neutron excess is measured by the
asymmetry parameter $I=(N-Z)/A$.
Several investigations have been performed on the ground state properties
of (strongly) asymmetric nuclear matter, with large asymmetry parameter,
up to values as large as 0.8. In the extreme limit $I\to 1$ one obtains
neutron matter, which is obviously unbound as far as nuclear forces
are concerned, but deserves very interesting applications in the
description of neutron stars, which are bound by gravitational force,
but can be considered as a unique realization of this ideal system.
The equation of state of asymmetric nuclear matter contains a symmetry
term which depends upon the asymmetry parameter and the density of the
system; the equilibrium conditions are crucially determined by the
NN interaction: various effective forces can produce quite different
behaviours with $\rho$ in the potential energy contribution
to the symmetry energy.\cite{Ditoro1}
A study of non--equilibrium properties of asymmetric nuclear matter
has been carried out by Di Toro {\it et al.}\cite{Ditoro2} starting from
two Vlasov equations, for proton and neutron liquids, coupled through
the mean field. The unstable growth of density perturbations is
influenced by the initial asymmetry: in neutron rich nuclear matter
the formation of larger fragments, in which isospin
symmetry is restored, is favoured.
At low (subsaturation) densities the system develops collective
dynamical instabilities of isovector nature, the so--called
{\it spinodal decomposition}: it corresponds to the growth of
density pertubations leading to a liquid--gas phase separation in
which symmetric nuclei coexist with a neutron gas. The instability
region (for example in the $(T,\rho)$ plane, turns out to be reduced
in strongly asymmetric systems.
This phenomenon can be tested in heavy ion reactions with radioactive
beams and could provide important information on large systems of
astrophysical interest.
\subsection{Neutron stars: structure and equation of state}
A large amount of information on neutron stars should be available in
the next few years from the new generation of X-- and $\gamma$--ray
satellites. In view of this, neutron stars are the subject of many
theoretical studies, aimed to predict their structure on the basis of
the properties of dense matter. The equation of state (EOS) of neutron
stars covers a wide density range, from $\sim 10$~g/cm$^3$ in the
surface to several times nuclear matter saturation density
($\rho\equiv\rho_0\sim
2.8\times 10^{14}$g/cm$^3$) in the center of the star.
The core (interior part) of a neutron star is believed to consist of
asymmetric nuclear matter with a consistent fraction of leptons. At
ultra--high densities, matter might undergo a transition giving rise
to other exotic hadronic components, like hyperons, a $K^-$ condensate
or a deconfined phase of quark matter. This occurrence, in turn, could
critically influence the evolution of neutron stars and could
eventually lead to the formation of black holes.\cite{Bomb97}
Baldo {\it et al.} have derived the static properties of non--rotating
neutron stars in a conventional framework, using the microscopic EOS
for asymmetric nuclear matter, derived from the Brueckner--Bethe--Goldstone
many--body theory.\cite{Baldo97} They use the Argonne $V_{14}$ and the
Paris two--body interactions, implemented by the Urbana model for the
three--body force. The latter are incorporated into the Brueckner scheme
by reducing them to affective two--body, density dependent forces. The
parameters are adjusted to reproduce the empirical nuclear matter
saturation point.
The calculated EOS allows to compute masses and radii as a
function of the central density $n_0$. Assuming that a neutron star
is a spherically symmetric distribution of mass in hydrostatic
equilibrium, and neglecting the effects of rotations and magnetic field,
the equilibrium configurations are obtained by solving the Tolman--
Oppenheimer--Volkoff (TOV) equations for the total pressure and the
enclosed mass:
\begin{eqnarray}
\frac{dP(r)}{dr} &=& -\frac{Gm(r)\rho(r)}{r^2}
\frac{\displaystyle{\left(1+\frac{P(r)}{c^2\rho(r)}\right)
\left(1+\frac{4\pi r^3 P(r)}{c^2 m(r)}\right)}}
{\displaystyle{\left(1-\frac{2G m(r)}{rc^2}\right)}}
\label{dpress}\\
\frac{dm(r)}{dr} &=& 4\pi r^2 \rho(r),
\label{dm}
\end{eqnarray}
where $G$ is the gravitational constant. The maximum mass configuration
obtained in ref.\cite{Baldo97} range from 1.8 to 2.13 solar masses,
while the corresponding radii vary from 8 to 10.6~Km. These values are
consistent with the present observations.
A challenging suggestion is found in a work of Bombaci\cite{Bomb98},
concerning the semiempirical mass--radius relation extracted for the
X--ray burst source 4U 1820--30: the latter cannot be reproduced by
neutron star models based on conventional EOS. It is thus suggested
that the onset of a phase transition to a $K^-$ condensate could explain,
with a suitable choice of the parameters, the observed mass--radius
relation.
\subsection{Neutron stars: the effects of superfluidity}
As already mentioned in the above subsections,
it is believed that the inner crust of a neutron star, at a density of
about $3\times 10^{11}$~g/cm$^3$ (corresponding to about
$10^{-3}\rho_0$), is a superfluid and
inhomogeneous system, consisting of a lattice of nuclei
immersed in a sea of neutrons and an approximately uniform
sea of electrons. Such a configuration persists up to roughly
$\rho_0/2$. At low temperatures, this system is superfluid with a
positive Fermi energy $\epsilon_F$: the estimate of the pairing gap
in this inhomogeneous medium has been carried out by Barranco
{\it et al.}\cite{Barranco1}. They solve the Hartree--Fock--Bogoliubov
(HFB) equations in a Wigner--Seitz cell, with a Woods--Saxon
potential; the gap is calculated self--consistently without altering
the single particle levels. Use is made of the Argonne two--body
interaction.
The HFB method is suitable to describe the halo--nuclei and/or
neutron--rich nuclei. Surface collective modes (phonons) mediate
the induced interaction producing Cooper pairs and a bosonic
condensate.
In a second work\cite{Barranco2} the quantum
calculation is compared to the LDA. Here the two--body interaction
is assumed to be a Gogny force and is included in the HFB
equations via the pairing field.
It is found that the LDA leads to a spatial variation of the
gap near the surface of a nucleus, which is stronger than the one
obtained in the HFB calculation. This is caused by the neglect
of the proximity effects and the delocalized character of the
single--particle wave functions close to the fermi energy.
From the energy of the system,
\begin{equation}
<E> =\sum_q n_q E_q,
\end{equation}
where $n_q=(1+e^{E_q/T})^{-1}$ is the occupation number (at the
temperature $T$) for the quasi--particle state $q$, one can obtain
the {\it specific heat} of the system:
\begin{equation}
C_v=\frac{1}{V}\frac{\partial <E>}{\partial T}\, .
\label{specH}
\end{equation}
This quantity turns out to be very sensitive to the pairing gap,
and hence to the presence of the inhomogeneity induced by the
presence of the nuclei in the crust.
The overestimate of the gap, as obtained in LDA, leads to a
specific heat of the system which is too large at low temperatures,
as compared with the quantal result. Incidentally, a reliable
estimate of the pairing gap is also important for the determination
of the cooling time of the neutron star.
In concluding this argument, it is worth mentioning that pairing
correlations have been studied within a relativistic mean field
approach based on a field theory of nucleons coupled to neutral
($\sigma$ and $\omega$) and charged ($\rho$) mesons. The HF and
pairing fields are calculated in a self--consistent way\cite{Matera}.
The energy gap is the result of strong cancellations between the
scalar and vector components of the pairing field. It is found
that the pair amplitude vanishes beyond a certain value of momentum
of the paired nucleons; the estimated gap is in agreement with
non--relativistic calculations of this quantity.
\subsection{Neutron stars: massive quark matter}
While in the previous subsection the attention was focussed on the
surface region of a neutron star, where the low density advocates
nucleonic degrees of freedom, its high central density, up to
$5\div 10~\rho_0$, seems to imply the existence of a core of massive
quark matter. The latter can be described in a variety of different
models, among which we shall consider here the so--called Color
Dielectric Model (CDM), which has been successfully employed in
connection with the single nucleon properties (structure functions,
e.m. form factors) as well as for the nucleon--nucleon interaction.
The CDM entails confinement of the quarks in ordinary hadronic
matter but also allows to describe, in mean field approximation,
a phase transition to deconfined quark matter\cite{Drago1, Drago2,
Drago3}. The model is defined by the following Lagrangian:
\begin{eqnarray}
{\cal L} &&= i{\bar\psi}\gamma^{\mu}\partial_{\mu}\psi
+\sum_{f=u,d}\frac{g_f}{f_{\pi}\chi}{\bar\psi}_f\left(\sigma +
i\gamma_5\tau\cdot\pi\right)\psi_f
\nonumber\\
&& + \frac{g_s}{\chi}{\bar\psi}_s\psi_s
+\frac{1}{2}\left(\partial_{\mu}\chi\right)^2
-\frac{1}{2}{\cal M}^2\chi^2
\label{CDMlag}\\
&& +\frac{1}{2}\left(\partial_{\mu}\sigma\right)^2
+\frac{1}{2}\left(\partial_{\mu}\pi\right)^2 -U(\sigma,\pi)
\nonumber
\end{eqnarray}
where $\chi$ is the color dielectric field and $U(\sigma,\pi)$ the
``mexican hat'' potential of the chiral sigma--model. This Lagrangian
describes a system of interacting $u$, $d$ and $s$ quarks, pions,
sigmas and a scalar--isoscalar chiral singlet field $\chi$. The latter
is related to the fluctuations of the gluon condensate around its
vacuum expectation value. The coupling constants are given by
$g_{u,d}=g(f_\pi\pm \xi_3)$ and $g_s=(2f_K-f_\pi)$, where $f_\pi=93$~MeV
and $f_K=113$~MeV are the pion and kaon decay constants, respectively,
while $\xi_3=f_{K^{\pm}}-f_{K^0}=-0.75$~MeV. Hence the model contains
only two free parameters, $g$ and ${\cal M}$, which are fixed to the
values $g=0.023$~GeV, ${\cal M}=1.7$~GeV.
Confinement is obtained via the effective quark masses, which diverge
outside the nucleon.
The CDM is employed to describe the deconfined quark matter phase in
the neutron star, while the relativistic field theoretical model of
Walecka is used for the hadronic phase. Applying Gibbs criteria to this
composite system, Drago {\it et al.}\cite{Drago2} find that the pure
hadronic phase ends at $0.11$~fm$^{-3}$ while the mixed (quark and hadronic)
phase extends up to $0.31$~fm$^{-3}$.
According to this calculation, neutron stars with masses in the range
$1.3 < M/M_{\odot} <1.54$ and radii of about 9~Km are found. A neutron star
with total mass of $1.4~M_{\odot}$ will consist of a crust of pure
hadronic matter, a $\sim 1$~Km thick region of mixed phase and a core of
pure quark matter.
In a recent work\cite{Drago3} the EOS of quark matter based on the CDM
has been discussed, taking into account the effects associated with
finite temperatures. Both the evolution of neutron stars and the onset
of supernovae explosion crucially depend upon the EOS of matter at very
high densities and temperatures. At finite $T$ the EOS considered here
shows a decrease of the pressure and of the adiabatic index, leading to
a deconfinement transition for densities slightly larger than the one
corresponding to nuclear matter saturation. The presence of a mixed phase
region softens the EOS and could lead to a direct supernova explosion.
At larger densities the EOS is stiff enough to support a neutron star
compatible with observations.
\section{Quark matter}
Quantum Chromodynamics (QCD), the non--abelian theory of coloured quarks
and gluons, is currently accepted as the theory of strong interactions,
and its predictions have been tested in a variety of
elementary particle reactions at large momentum transfers.
Its behaviour in the high energy
(or short distance) limit approaches the one of a non--interacting
free field theory ({\it asymptotic freedom}), while at low energies (or
large length scales) QCD is a non--perturbative field theory,
quarks and gluons being permanently confined inside hadrons.
At very high nuclear densities and/or temperatures it is believed that
hadronic matter undergoes a phase transition to a deconfined phase of
quarks and gluons, the so--called Quark Gluon Plasma (QGP). In
this new phase hadrons dissolve, strong interactions become very weak
and an ideal colour--conducting plasma of quarks and gluons is formed.
In the QGP the long--range colour force is Debye--screened due to collective
effects and the quarks can only interact via a short--range effective
potential.
During the last decade the attention of nuclear and particle physics
has been attracted by the possibility of producing this new state of
matter in the laboratory, by means of ultra--relativistic heavy
ion collisions. A large number of experiments have been carried
out at the AGS (Brookhaven) and at the SPS (CERN). Both laboratories
have planned considerable ``upgrading'' of the existing facilities,
the so--called RHIC (Relativistic Heavy Ion
Collider), which is going into operation in 1999 and the LHC (Large
Hadron Collider), whose program will be partly dedicated to high energy
particle physics and partly to heavy ion collisions.
The most recent measurements refer to S+U reactions at 200~GeV/A and
Pb+Pb at 158~GeV/A, reaching an (estimated) energy density of
$2\div 5$~GeV/fm$^3$.
Many aspects of the transition from hadrons to deconfined quark matter,
among which the order of the phase transition and the nature of
experimentally observable signatures, are still under debate.
Recently the NA50 experiment at CERN\cite{NA50} has shown an anomalous
suppression of the J/$\Psi$ survival probability measured
in Pb on Pb collisions\cite{Satz0}, which might be ascribed to the
formation of QGP at some stage of the collision.
Other interesting signals could be revealed by the dilepton spectra,
which are sensitive to the masses of vector mesons: the latter
can be significantly altered by the presence of the QGP, an argument
which is still widely discussed.
\subsection{Quark deconfinement}
At high densities and temperatures, the
interaction between quarks and gluons dresses their propagation,
so that gluons develop an effective mass, which in
turn produces screening of the long--range colour--electric forces.
For example, the free gluon propagator
$D_0(\omega,k)\sim (\omega^2-k^2)^{-1}$ is modified by summing an infinite
chain of one--loop insertions. One obtains\cite{Kapust2} the following
longitudinal gluon propagator
\begin{equation}
D_L(\omega,k)=\frac{1}{k^2\varepsilon_L(\omega,k)}
\label{gluonl}
\end{equation}
$D_L$ being a scalar function of the variables
$\omega=k^0$ and $k=|\vec k|$ and $\varepsilon_{L}$ the so--called
colour dielectric function. From explicit calculations, one can
show that the static ($\omega=0$) longitudinal colour fields are
screened, being
\begin{equation}
D_L(0,k)= \frac{1}{k^2\varepsilon_L(0,k)}= \frac{1}{k^2+g^2 T^2}
\equiv \frac{1}{k^2+\lambda^{-2}_D}
\label{static}
\end{equation}
which defines the Debye length $\lambda_D=(gT)^{-1}$.
The above mentioned features of the gluon propagator entail several
important consequences, among which we focus here on the fact that
the potential between two static colour charges (e.g. two heavy
quarks) is {\it screened} in the quark--gluon plasma phase. Indeed the
Fourier transform of $D_L(0,k)$, eq.(\ref{static}), yields the potential
\begin{equation}
V_{Q{\bar Q}}(r)\simeq \frac{1}{r}e^{-r/\lambda_D}
\label{Vscreen}
\end{equation}
with screening length $\lambda_D\simeq 0.4$~fm at $T=250$~MeV.
This screening of the long range colour forces is believed to be
at the origin of quark deconfinement in the high temperature phase.
It also leads to the disappearence of the bound states of a charmed
quark pairs $(c{\bar c})$ in the QGP\cite{Matsui}.
\begin{figure}
\centerline{
\epsfxsize=9cm
\epsfysize=9.5cm \epsfbox{pot.ps}
}
\caption[ ]{ The quark--quark potential, eq.(\ref{potsatz}) at $T=250^{\circ}$
(dashed line) and $T=0$ (continuous line). }
\label{fig10}
\end{figure}
Fig.~10 illustrates the deconfinement effect of a thermodynamical enviroment
of interacting light quarks and gluons on the interquark potential employed
in ref.\cite{KMS} to investigate the deconfinement conditions for heavy
quarkonia.:
\begin{equation}
V(r,T)=\frac{\sigma}{\mu(T)}\left(1-e^{\mu(T)r}\right)
-\frac{\alpha}{r}e^{-\mu(T)r}
\label{potsatz}
\end{equation}
where $\mu(T)=1/\lambda_D(T)$. The potential (\ref{potsatz}) reduces
to the Cornell potential in the zero temperature limit (we remind that
$\lambda_D\to\infty$ for $T\to 0$); in addition to the one gluon exchange
contribution ($1/r$) this potential also contains a phenomenological
linear binding term, as suggested by lattice QCD calculations for the
interaction between static (heavy) colour charges.
It would be important to find some ``QCD inspired'' model, suitable
to describe both the confined (hadronic) and the deconfined (plasma)
phases of a system of {\it many} quarks. This is obviously not trivial
to achieve, due to the non--perturbative nature of the quark--quark
interaction in the confined phase and also to the unavoidable
approximations required to describe a many--body system, even when
the interactions are known.
In a recent work Alberico {\it et al.}\cite{Piotr} have developed
a three--dimensional model for quark matter with a {\it density
dependent} quark--quark (confining) potential, which allows
to describe a sort of deconfinement transition as the system evolves
from a low density assembly of bound structures to a high density
free Fermi gas of quarks. Different confining potentials are considered,
some of which successfully utilized in hadron spectroscopy, like the
Cornell one.
We find that a proper treatment of the many--body correlations
induced by the medium is essential to disentangle the different nature
of the two (hadronic and deconfined) phases of the system. For this
purpose the ground state energy per particle and the pair
correlation function are investigated. The latter can be obtained
from the expectation value of the two--body density operator:
\begin{equation}
g(r) =\frac{N(N-1)}{\rho^2}
<\Psi|\rho_2(|{\vec r}_1-{\vec r}_2|)|\Psi>,
\label{defgr}
\end{equation}
where $|\Psi>$ is the exact (normalized) ground state of the system.
For a system of strongly correlated pairs $g(r)$ will show up a
well localized peak at small values of $r$, while the Fermi gas
pair correlation function has a fairly constant behaviour, with
the exception of small $r$ values, where the Pauli principle prevents
particles to be close to each other.
As an example, Fig.~11 shows the pair correlation function derived from
the Bethe--Goldstone wave functions of the screened Cornell potential:
\begin{equation}
V_{Cornell}(r,\rho) = \left(-\frac{a}{r} + br +K\right)e^{-c\rho r}
\label{Vcornel}
\end{equation}
where $a,b,c$ and $K$ are constants, while $\rho$ is the density of
the system. The free Fermi gas two--body correlation function is
recovered for $\rho\simeq 0.6$~fm$^{-3}$, in agreement with the value
of the ``transition density'' which can be obtained from the
corresponding equation of state.
\begin{figure}
\centerline{
\epsfxsize=9.5cm
\epsfysize=9cm \epsfbox{gcor2.ps}
}
\caption[ ]{The pair correlation function $g(r)$ obtained with the
potential (\ref{Vcornel}) is displayed as a function of the
relative interquark distance at various densities:
$\rho=0.05$~fm$^{-3}$ (thick solid line), $\rho=0.1$~fm$^{-3}$ (dashed
line), $\rho=0.2$~fm$^{-3}$ (dot--dashed line),
$\rho=0.35$~fm$^{-3}$ (dashed line) and $\rho=0.6$~fm$^{-3}$
(dotted line). For the last density value the Fermi gas correlation
function is also shown (thin solid line). }
\label{fig11}
\end{figure}
\subsection{Finite size effects in the QGP}
Before concluding this Section it is worth mentioning that one of the
major difficulties in the analysis of the relativistic heavy ion
collisions stems from the numerous uncertainties in the thermostatistical
evolution of the system. Among the various steps, the hadronization
process is crucial in determining the final states measured in the
detectors.
Brink {\it et al.}\cite{Brink} have studied the hadronization of a
plasma, in order to establish whether the final particles are
evaporated from droplets of hadronic matter inside the plasma or
rather from the whole volume of the plasma itself. The two situations
can be distinguished on the basis of the entropy density of the system.
Let us assume that at equilibrium (close to the critical temperature
$T\sim T_c$) the volume of the plasma ($V_{tot}$) is filled by $N$ droplets
of QGP (each occupying a volume $V_{glob}$) and a gas of relativistic
pions. For a fixed total volume and energy density, the calculated
entropy density of the systems turns out to be larger when the plasma
fills a big, unique glob rather than several globs of smaller radius.
This is partially illustrated in Fig.~12. Finite size corrections are
taken into account in the density of levels which enters into the
calculation of the entropy. Fluctuations with respect to the number of
globs are small, hence there is no tendency for a large glob to break
spontaneously into smaller ones.
\begin{figure}
\centerline{
\epsfxsize=11cm
\epsfysize=8cm \epsfbox{w1.eps2}
}
\caption[ ]{ Entropy density for a gas of quarks and gluons as a
function of the glob radius $R$. The temperature is fixed to be $T=200$~MeV. }
\label{fig12}
\end{figure}
This argument concludes the tour which was foreseen in the introduction,
showing how the various many--body techniques can be applied to strongly
interacting systems, either of nucleons or of quarks, in an attempt
to provide a unified microscopic description of the intriguing, complex
structure of nuclear systems.
\section*{Acknowledgments}
I would like to thank Dr. A. de Pace and Prof. A. Molinari for
enlightening discussions and to acknowledge their valuable help.
|
\section{A model of $v_{12}(r)$}
In this presentation we report on the the possibility of using the
``mean tendency of well-separated galaxies to approach each other''
\cite{peebles80} to measure the cosmological
density parameter, $\Omega$. The statistic we consider\cite{ferreira98} is the mean relative
pairwise velocity of galaxies, ${v_{{\scriptscriptstyle 12}}}$. It was introduced
in the context of the BBGKY theory \cite{davis77},
describing the dynamical evolution of a collection of particles
interacting through gravity. In this discrete picture,
${{\vec v}_{{\scriptscriptstyle 12}}}$ is defined as the mean value of the peculiar
velocity difference of a particle pair at separation ${\vec r}$.
In the fluid limit, its analogue is the
pair-density weighted relative velocity
\cite{fisher94,rj98a},
\begin{eqnarray}
{{\vec v}_{{\scriptscriptstyle 12}}}(r) \; = \; \langle \, {{\vec v}_{{\scriptscriptstyle 1}}} - {{\vec v}_{{\scriptscriptstyle 2}}} \, \rangle_{\scriptscriptstyle {\rho}} \; = \;
{{ \langle({{\vec v}_{{\scriptscriptstyle 1}}} -
{{\vec v}_{{\scriptscriptstyle 2}}} ) (1 + {{\delta}_{{\scriptscriptstyle 1}}})(1 +{{\delta}_{{\scriptscriptstyle 2}}} ) \rangle}
\over {1 \; + \; \xi(r)}} \; ,
\label{def}
\end{eqnarray}
where $\, {{\vec v}_{{\scriptscriptstyle A}}} \,$ and $\, {{\delta}_{{\scriptscriptstyle A}}} = {{\rho}_{{\scriptscriptstyle A}}} / \langle \rho \rangle - 1 \,$
are the peculiar velocity and fractional density contrast
of matter at a point ${{\rv}_{{\scriptscriptstyle A}}}$, $r=|{{\vec r}_{{\scriptscriptstyle 1}}} - {{\vec r}_{{\scriptscriptstyle 2}}}|$, and $\xi(r) =
\langle{{\delta}_{{\scriptscriptstyle 1}}}{{\delta}_{{\scriptscriptstyle 2}}}\rangle$ is the two-point correlation function.
The pair-weighted average, $\langle\cdots\rangle_{\scriptscriptstyle {\rho}}$,
differs from simple spatial averaging, $\langle\cdots\rangle$,
by the weighting factor
${{\rho}_{{\scriptscriptstyle 1}}}{{\rho}_{{\scriptscriptstyle 2}}} \,\langle {{\rho}_{{\scriptscriptstyle 1}}}{{\rho}_{{\scriptscriptstyle 2}}}\rangle^{-1}$,
proportional to the number-density of particle pairs.
In a recent letter \cite{rj98}, one of us has shown that an excellent approximation to ${v_{{\scriptscriptstyle 12}}}$ is given by
\begin{eqnarray}
{v_{{\scriptscriptstyle 12}}} (r) \; &=& \; - \, {\textstyle{2\over 3}} \, H r f \,
\bar{\hspace{-0.08cm}\bar{\xi}} (r)[1 + \alpha \; \bar{\hspace{-0.08cm}\bar{\xi}} (r)] \;,
\label{2nd-order}\\
\bar{\xi}(r) \; &=& \; (3/r^3)\, \int_0^r \, \xi(x) \, x^2 \, dx \;
\equiv \; \bar{\hspace{-0.08cm}\bar{\xi}}(r) \, [\, 1 + \xi(r) \, ]
\label{xb}
\end{eqnarray}
Here
$\alpha$ is a parameter, which depends on the logarithmic slope
of $\xi(r)$, while
$f = d \, \ln D/d \ln a$, with
$D(a)$ being the standard linear growing mode
solution and $a$ -- the cosmological
expansion factor.
Finally, $H = 100\;h\;\mbox{km}\;\mbox{s}^{-1}\;\mbox{Mpc}$ is
the present value of the Hubble constant.
This approximate solution of the pair conservation equation,
accurately
reproduces results of high resolution N-body simulations
in the entire dynamical range \cite{rj98}.
If we restrict ourselves to $\, r = 10 h^{-1}\, \mbox{Mpc} \,$,
one can use the APM catalogue of galaxies \cite{efst96} for an
estimate of galaxy correlation function, $\xi=(r/r-0)^{-\gamma}$ ;
the slope at the separation considered
is $\gamma=1.75\pm0.1$ (the errors we quote are conservative).
One obtains then
\begin{equation}
{v_{{\scriptscriptstyle 12}}}(10 h^{-1}\;\mbox{Mpc}) \; =
\; - \, 605 \, {{\sigma_{{\scriptscriptstyle 8}}}}^2 \Omega^{0.6}
\,(1 + 0.43{\sigma_{{\scriptscriptstyle 8}}}^2)\,/(1+ 0.38{{\sigma_{{\scriptscriptstyle 8}}}}^2)^2 \, {\rm km/s}
\; .
\label{10mpc}
\end{equation}
The above relation shows that at $r = 10 h^{-1}\;\mbox{Mpc}$, ${v_{{\scriptscriptstyle 12}}}$ is
almost entirely determined by the values of two parameters: ${\sigma_{{\scriptscriptstyle 8}}}$ and
$\Omega$. It is only weakly dependent on $\gamma$.
The uncertainties in the
observed $\gamma$ lead to an error in Eq.~\ref{10mpc} of less than
$10\%$ for ${\sigma_{{\scriptscriptstyle 8}}}\le 1$.
\section{The estimator}
Since we observe only the line-of-sight component of the
peculiar velocity, ${s_{{\scriptscriptstyle A}}} = {{\rv}_{{\scriptscriptstyle A}}}\cdot {{\vec v}_{{\scriptscriptstyle A}}}/r \equiv
{{\hat r}_{{\scriptscriptstyle A}}}\cdot {{\vec v}_{{\scriptscriptstyle A}}}$, rather than the
full three-dimensional velocity ${{\vec v}_{{\scriptscriptstyle A}}}$, it is not possible
to compute ${v_{{\scriptscriptstyle 12}}}$ directly. Instead, we propose to use the
mean difference between radial velocities of a pair of
galaxies,
$\langle \, {s_{{\scriptscriptstyle 1}}} - {s_{{\scriptscriptstyle 2}}} \, \rangle_{\scriptscriptstyle {\rho}} \; = \;
{v_{{\scriptscriptstyle 12}}} \, {\hat r}\cdot({{\hat r}_{{\scriptscriptstyle 1}}} + {{\hat r}_{{\scriptscriptstyle 2}}})/2$,
where ${\vec r} = {{\vec r}_{{\scriptscriptstyle 1}}} - {{\vec r}_{{\scriptscriptstyle 2}}}$.
To estimate ${v_{{\scriptscriptstyle 12}}}$, we use the simplest least squares techniques,
which minimizes the quantity
$\chi^2(r) \; = \; \sum_{{_{{\scriptscriptstyle A,B}}}} \, \left[ ({s_{{\scriptscriptstyle A}}} - {s_{{\scriptscriptstyle B}}}) - {p_{{\scriptscriptstyle AB}}}
\,{{\tilde v}_{{\scriptscriptstyle 12}}}(r)/2 \, \right]^2 \; \;$,
where ${p_{{\scriptscriptstyle AB}}} \equiv {\hat r} \cdot ({{\hat r}_{{\scriptscriptstyle A}}} + {{\hat r}_{{\scriptscriptstyle B}}})$ and
the sum is over all pairs at fixed separation $r = |{{\rv}_{{\scriptscriptstyle A}}} - {{\rv}_{{\scriptscriptstyle B}}}|$.
The condition $\partial \chi^2 / \,\partial{{\tilde v}_{{\scriptscriptstyle 12}}} = 0$ implies
\begin{eqnarray}
{{\tilde v}_{{\scriptscriptstyle 12}}} (r) \; = \; {
{2 \sum \, ({s_{{\scriptscriptstyle A}}} - {s_{{\scriptscriptstyle B}}})\, {p_{{\scriptscriptstyle AB}}} }\over
{\sum {p_{{\scriptscriptstyle AB}}}^2}} \;\; .
\label{estimator}
\end{eqnarray}
The above expression is a sum over positive quantities
and so is stable. This estimator is appropriate to be
applied to a point process which will sample an underlying continuous
distribution. The sampling is quantified in terms of the selection
function, $\phi({\vec r})$. The continuum limit of Eq.~\ref{estimator}
is then
\begin{eqnarray}
{{\tilde v}_{{\scriptscriptstyle 12}}} (r)=\frac{2 \int d {m_{{\scriptscriptstyle 1}}} \,
d {m_{{\scriptscriptstyle 2}}} \, {\Phi_{{\scriptscriptstyle 12}}} \,
({s_{{\scriptscriptstyle 1}}} - {s_{{\scriptscriptstyle 2}}}){p_{{\scriptscriptstyle 12}}}}
{\int d {m_{{\scriptscriptstyle 1}}} \,
d {m_{{\scriptscriptstyle 2}}} \, {\Phi_{{\scriptscriptstyle 12}}} \,
{p_{{\scriptscriptstyle 12}}}^2} \; ,
\label{estcont}
\end{eqnarray}
with $d {m_{{\scriptscriptstyle A}}} = {{\rho}_{{\scriptscriptstyle A}}}~d^3 {{\rv}_{{\scriptscriptstyle A}}}$, and
a two-point selection function given by
${\Phi_{{\scriptscriptstyle 12}}} \; = \; \delta_D(|{{{\vec r}_{{\scriptscriptstyle 1}}} - {{\vec r}_{{\scriptscriptstyle 2}}}}| - r)
\, \phi({{\vec r}_{{\scriptscriptstyle 1}}}) \, \phi({{\vec r}_{{\scriptscriptstyle 2}}}) \;$,
where $\delta_D$ is the Dirac delta function.
For ease of notation we shall denote the denominator
in Eq.~\ref{estcont} by $W(r)$. If
we take the ensemble average of Eq.~\ref{estcont} we
find that $\langle {{\tilde v}_{{\scriptscriptstyle 12}}}(r) \rangle={v_{{\scriptscriptstyle 12}}} (r)$. Note that,
unlike the estimators for the velocity correlation tensor proposed in
\cite{gorski89}, the ensemble average of the estimator is ${v_{{\scriptscriptstyle 12}}} (r)$
independent of
the selection function. For an isotropic selection function this
estimator is insensitive to systematic effects such as a bulk flow,
large scale shear and small scale random velocities
(as one might expect from virialized objects).
\section{Biases and errors}
How unbiased is this estimator? We have applied our
statistic to mock catalogues extracted from N-body simulations of a
dust-filled universe with $\Omega=1$ and $P(k) \propto 1/k$.
\begin{figure}[t]
\hspace{0.7in}
\psfig{figure=Fig2.ps,width=3.3in}
\vskip -0.7in
\caption{${{\tilde v}_{{\scriptscriptstyle 12}}}(r)$ (points) and its variance (dashed lines)
evaluated from mock catalogues: A) random
observers with a deep selection function compared to ${v_{{\scriptscriptstyle 12}}} (r)$
evaluated from Eq.~\ref{2nd-order} (solid line); b)
A fixed observer with a deep selection function; c) a fixed observer
with a shallow selection function. The variance is estimated
from the scatter over 20 (a) or 9 (b,c) mock catalogues.
\label{fig2}}
\end{figure}
In Figure~\ref{fig2}(a) we plot ${{\tilde v}_{{\scriptscriptstyle 12}}}(r)$ with one standard
deviation calculated with 20 mock catalogues extracted with a
deep selection function. Each catalogue has a
different observation position within the simulation volume and so an
average over this set should resemble a true ensemble average. The mean
is consistent with what one would expect from a
direct calculation with Eq.~\ref{2nd-order} (which is plotted in
Figure~\ref{fig2}(a) as a solid line). We have also performed this
analysis without collapsing the cores; the results changed by very
little.
We repeat this
calculation for a set of 9 catalogues all constructed from the same
observation point using a deep (Figure~\ref{fig2}b) or shallow
(Figure~\ref{fig2}c) selection
function to randomly sample a fraction of galaxies within the
simulation box.
The variance
in ${{\tilde v}_{{\scriptscriptstyle 12}}}(r)$ is now solely due to finite sampling (``shot noise'');
for catalogues with 2000 to 3000 galaxies we expect the
variance to be $\sqrt{2}$--$\sqrt{3}$ times larger.
We find that a shallow selection function
changes the functional form, or slope, of the mean,
making it a more sharply decreasing
function of $r$ than the ensemble average. It is therefore crucial when
analyzing a catalogue to restrict oneself to scales much smaller than
the effective cutoff scale of the selection function.
Errors in distance measurements will naturally affect our
results; the best estimators
use empirical correlations between intrinsic properties of
the galaxies and luminosities and
lead to log-normal errors in the estimated distance of around
$20\%$. These errors will naturally lead to biases in cosmological
estimators involving distance measurements and peculiar velocities and
are generically called Malmquist bias.
We shall model our errors assuming a Tully--Fisher law which
resembles that inferred from the Mark III catalogue. To correct for
Malmquist bias we use the prescription put forward
in \cite{landy92}.
In Figure~\ref{fig3}(a), we plot the results for the uncorrected simulations;
Malmquist errors systematically lower the values of ${{\tilde v}_{{\scriptscriptstyle 12}}}$
on small scales while enhancing its amplitude on large scales
(where the effect should be more dominant).
However in Figure~\ref{fig3}(b) we show
that with the correction for general Malmquist errors to the distance
estimator, it is possible to overcome this discrepancy. The 1-$\sigma$
errors now encompass the true ${{\tilde v}_{{\scriptscriptstyle 12}}}$ over a wide range of
scales.
\begin{figure}[t]
\hspace{0.7in}{
\psfig{figure=Fig3.ps,width=3.3in}}
\vskip -0.7in
\caption{${v_{{\scriptscriptstyle 12}}} (r)$ and its variance evaluated from 100 mock catalogues
with errors (described in the text) and the full selection function. The
solid points are the ${{\tilde v}_{{\scriptscriptstyle 12}}}$ of the error-free simulation seen from the
same observation point,
the solid line is the mean and dashed lines are the 1 $\sigma$. a)
uncorrected distances; b) distances corrected for Malmquist bias
\label{fig3}}
\end{figure}
\section{Conclusions}
In this contribution we report on a recent proposal to estimate the
mean pairwise streaming velocities of galaxies
directly from peculiar velocity samples.
We identified three possible sources of systematic errors in
estimates of ${v_{{\scriptscriptstyle 12}}}$ made directly from radial peculiar velocities
of galaxies. We also found ways of
reducing these errors; these techniques were successfully tested
with mock catalogues. The potential sources of errors and their
proposed solutions can be summarized as follows.
(1) On the theoretical front, assuming a linear theory
model of ${v_{{\scriptscriptstyle 12}}}(r)$ at $r \approx
10h^{-1}\;\mbox{Mpc}$ can introduce a considerable systematic error in
the resulting estimate of ${\sigma_{{\scriptscriptstyle 8}}}^2\Omega^{0.6}$.
For example, if ${\sigma_{{\scriptscriptstyle 8}}} = 1$ using the linear prediction for
${v_{{\scriptscriptstyle 12}}}$ at $r = 10 h^{-1}\;\mbox{Mpc}$ would introduce a 25\%
systematic error (see eq.~[\ref{10mpc}]).
We solve this problem by using a nonlinear expression
for ${v_{{\scriptscriptstyle 12}}}$.\cite{rj98}
(2) On the observational front,
a shallow selection function induces a
large covariance between ${{\tilde v}_{{\scriptscriptstyle 12}}}$ on different scales. This
must be taken into consideration by measuring ${{\tilde v}_{{\scriptscriptstyle 12}}}(r)$ only
on sufficiently small scales. A rule of thumb is that for
estimating ${{\tilde v}_{{\scriptscriptstyle 12}}}$ at $10h^{-1}\;\mbox{Mpc}$, the selection function
should be reasonably homogeneous out to at least $30h^{-1}\;\mbox{Mpc}$.
(3) Finally, care must be taken with generalized Malmquist bias due to
log-normal distance errors; these induce a systematic error in ${{\tilde v}_{{\scriptscriptstyle 12}}}$.
We have shown that, under certain assumptions about selection and
measurement errors, the method of Landy $\&$ Szalay\cite{landy92} for corrected
distance estimates allows one to recover the true ${{\tilde v}_{{\scriptscriptstyle 12}}}$. Naturally,
this particular correction must be addressed on a case-by-case basis,
given that different data sets will have different selection criteria
and correlations between galaxy position and measurement errors.
In a future publication we shall analyze the Mark III \cite{willick97}
and the SFI \cite{dacosta96} catalogues of
galaxies with this in mind.
\section*{Acknowledgments}
We thank the organizers for an enjoyable meeting.
We thank Jonathan Baker, Stephane
Courteau, Luis da Costa and Jim Peebles,
for useful comments and suggestions.
This work was supported by
NSF, NASA,
the Poland-US M. Sk{\l}odowska-Curie Fund,
the Tomalla Foundation and JNICT.
\section*{References}
|
\section{Introduction}
Detection of faint emission in the high excitation [O\,{\sc iv}] 25.9$\mu$m
infrared (IR) line has been reported by Lutz et al.\ \cite*{LUTZetal98}
in a number
of well-known starburst galaxies from observations with the
Short Wavelength Spectrometer on board {\sl ISO}.
IR line ratios such as [O\,{\sc iv}] 25.9$\mu$m / [Ne\,{\sc ii}] 12.8$\mu$m and
[O\,{\sc iv}] 25.9$\mu$m / [Ne\,{\sc iii}] 15.6$\mu$m were used
previously \cite{LUTZetal96,GENZELetal98} to establish the dominant
source (starburst or AGN) of ionizing radiations in ultraluminous infrared
galaxies.
Actually, rather than normalizing just to [Ne\,{\sc ii}] or [Ne\,{\sc iii}],
Lutz et al.\ \cite*{LUTZetal98} use [O\,{\sc iv}]/([Ne\,{\sc ii}] + 0.44[Ne\,{\sc iii}])
as a critical line ratio to reveal different excitation conditions in starburst
galaxies.
Results from pure photoionization models or from pure shock models are
used by these authors in order to explain the observed line ratios.
However, these models were not successful in reproducing the
IR line ratios of several starburst galaxies. In fact, models with a
power-law radiation producing enough [O\,{\sc iv}] line give too high values
for the [Ne\,{\sc iii}]/[Ne\,{\sc ii}] line ratio.
Those with two-temperature (39000 K + 80000 K) black body radiation
fail in explaining the low-excitation starbursts.
Finally, pure shock models could provide [O\,{\sc iv}] line intensities similar to
observations but not enough [Ne\,{\sc ii}] and [Ne\,{\sc iii}] which should come from the
H{\sc ii}\ region.
These results indicate that a composite model,
which accounts for the coupled effects of shocks and
photoionization, must be used.
Indeed, in a disturbed region such as that containing
starbursts, it is expected that shocked gas is embedded in
stellar radiation.
This kind of composite model has successfully explained the AGN
spectral features (Viegas \& Contini 1997 and references
therein) as well as those of single galaxies where
the AGN and starburst activities are of comparable importance
(e.g. the Circinus galaxy, Contini et al.\ 1998).
These models were provided by the SUMA code
(see Viegas \& Contini 1994 and references therein).
In the present study we suggest that shocks result from supernova
explosions in the nuclear
starburst region of the galaxy, which provide enough kinetic energy
to create high velocity superwinds.
These in turn will accelerate gas clouds on large scales.
The physical conditions in the ionized interstellar clouds are determined
by shocks coupled to photoionization by massive hot stars (O and Wolf-Rayet).
This paper, rather than being an answer to
Lutz et al.\ \cite*{LUTZetal98}, is aimed at carrying on
their investigation. The distinction between AGN and starburst activity
in galaxies being already established \cite{LUTZetal96,GENZELetal98},
our aim is to find out which models (provided by the SUMA code),
better fit the observed trend between IR line ratios in starburst galaxies.
We will not discuss in detail the
single galaxy structures, but will use the fit of calculated
to observed line ratios to define the critical parameters
(and their ranges of values) which prevail in the emitting gas.
In other words, we will treat the
emission-line regions of the observed galaxies
as schematic systems of gaseous clouds moving radially outward from
star-forming regions ionized by hot stars and shocks, avoiding a deep
investigation on the internal structure of single objects.
\begin{table}
\caption[]{Parameters for single-cloud models}
\begin{flushleft}
\begin{tabular}{lrrrr}
\hline
\hline
Galaxy & $V_{\rm s}$ & $n_{\rm 0}$ & $U$ & $D$ \\
& ($\rm km\, s^{-1}$) & ($\rm cm^{-3}$) & & (pc) \\
\hline
II Zw 40 & 50 & 100 &0.02 &0.10 \\
NGC 5253 & 100 & 100 &0.01 &0.01 \\
& & & & \\
NGC 3690 B/C & 300 & 100 &0.01 &0.03 \\
NGC 3256 & 400 & 500 &0.01 &1.70 \\
M 82 & 400 & 500 &0.01 &2.70 \\
M 83 & 500 & 500 &0.01 &0.70 \\
NGC 253 &600 & 800 &0.01 &0.23 \\
NGC 4945 & 600 & 800 &0.01 &0.20 \\
\hline
\hline
\end{tabular}
\end{flushleft}
\label{table1}
\end{table}
\section{The results by a composite model}
\label{model}
As commented above, Lutz et al.\ \cite*{LUTZetal98}
concluded that shocks are the
most plausible ionizing mechanism for explaining the [O\,{\sc iv}] line, while
the [Ne\,{\sc ii}] and [Ne\,{\sc iii}] come from the photoionized region. However,
shock results should be obtained by consistent models
concomitantly with photoionization from hot stars.
The SUMA code consistently accounts for both the shock effect accompanying
cloud expansion and photoionization by hot stars, in a plane-parallel
symmetry. The geometrical thickness of the clouds range from 0.01 to
2.70 pc (see Table~\ref{table1}), which is small compared with the
radius of the starburst regions in galaxies.
In our model, we assume that the gas clouds move radially outward from
the starburst region, i.e. the hot star population is facing the inner
edges of the clouds while a shock develops on the opposite
side.
For all models an average $stellar \, temperature$,
$T_{\rm *}$ = $5 \times 10^4$ K, for the hot star population,
and a preshock magnetic field, $B_{\rm 0}$ = $10^{-4}$ gauss, are adopted,
as well as cosmic abundances.
Lower abundances should lead to a lower cooling rate of the gas
downstream which could be reajusted by adopting higher preshock densities.
The other input parameters for SUMA: the shock velocity, $V_{\rm s}$,
the preshock density, $n_{\rm 0}$, the ionization parameter\footnote{The
ionization parameter is the ratio of the photon density of the ionizing
radiations reaching the nebula to the gas density in the inner part of
the nebula.}, $U$, and the
geometrical thickness of the clouds, $D$, are chosen to vary
within the ranges consistent with observations.
\begin{table*}
\caption{Calculated line fluxes (in $\rm erg\, cm^{-2}\, s^{-1}$) with multi-cloud models}
\begin{flushleft}
\tiny{
\begin{tabular}{|l|rrrrr|rrrr|rrrr|}
\hline
&&& &&& &&& &&&& \\
{\normalsize Line} & \multicolumn{5}{c|}{\normalsize M 82} & \multicolumn{4}{c|}{\normalsize NGC 3256} &
\multicolumn{4}{c|}{\normalsize NGC 253} \\
&&& &&& &&& &&&& \\
&&& &&& &&& &&&& \\
($\mu$m) & A1 & A2 & A3 & A4 & Av & B1 & B2 & B3 & Av & C1 & C2 & C3 & Av \\
&&& &&& &&& &&&& \\
\hline
[C\,{\sc ii}]\ 158&0.05 & 0.026 & 0.007 & 0.420& 2.56 & 0.03 & 0.012 & 0.007 & 1.6 & 0.03 & 0.07 & $6\times 10^{-4}$&7.06 \\
&0.01 & 16.5 & 27.0 & 56.5& - & 0.04 & 75 & 24.5 &- & 0.02 & 99.1 & 0.85 & - \\
[O\,{\sc iii}]\ 88&9.6 & 0.002 & 0.008 & 0.001 & 0.92 & 6.2 & 0.002 & 0.008 & 0.8 & 2.10 & $3\times 10^{-5}$ &0.0076& 0.78 \\
&6.30 & 4.2 & 89.0 & 0.39 & - & 13.9 & 23.8 & 62.4 & - & 11.3& 0.3 &88.5 &- \\
[O\,{\sc i}]\ 63 & 0.004&0.067 & 0.007 & 0.348 & 2.96 & 0.002 & 0.025 & 0.007 & 2.9 & 0.78 & 0.075 & $5\times 10^{-4}$ & 7.59 \\
& 0.0008 &37.0 & 23.0 & 40.0 & - & 0.002 &86.5 & 13.5 & - & 0.5 & 98.8 & 0.66 & - \\
[O\,{\sc iii}]\ 52 & 89.6 & 0.004 & 0.008 & 0.003 & 1.46 & 56.6 & 0.004 & 0.008 & 2.0 & 20 & $6.7\times 10^{-5}$ & 0.0078&1.73 \\
&36.7 & 4.5 & 57.8 & 0.9 & - & 54.6 & 20.4 & 26.5 & - & 53.7& 0.4 &45.7 &- \\
[Si\,{\sc ii}]\ 35 & 25.6 & 0.166 & 0.027 & 2.170 & 13.20 & 15.9 & 0.073 & 0.027 & 9.4& 3.70 & 0.44 &0.01 & 45.2 \\
& 1.2 & 20.8 & 20.5 & 57.6 & - & 3.2 & 77.4 & 19.4 & - & 0.4& 97.4 &2.2 & - \\
[S\,{\sc iii}]\ 33 & 12.0 & 0.055 & 0.001 & 0.570 & 3.10 & 7.5 & 0.024 & 0.001 & 2.6 & 3.30 & 0.02 & $2\times 10^{-4}$ &2.17 \\
& 2.3 & 29.3 & 4.2 & 64 & - & 5.5 & 91.6 & 3.0 & - & 7.1 & 91.9 & 0.92 & - \\
[S\,{\sc iii}]\ 18 & 96.0 & 0.031 & $6.6\times 10^{-4}$ & 0.114 & 1.54 & 60.6 & 0.013 & $6.6\times 10^{-4}$ & 2.5 & 30 & 0.005& $1.2\times 10^{-5}$ &1.85 \\
& 1.9 & 32.9 & 4.2 & 25.8 & - & 46.2 & 52.2 & 0.02 & - & 75 & 24.5 & 0.64 & - \\
&&& &&& &&& &&&& \\
[O\,{\sc iv}]\ 25.9 & 7 & $1.2\times 10^{-4}$ & $6.6\times 10^{-5}$ & $2.3\times 10^{-4}$ & 0.05 & 7 &$1.2\times 10^{-4}$ &
$6.6\times 10^{-5}$ & 0.14 & 5.6 & $8.8\times 10^{-5}$ & $6.3\times 10^{-5}$&0.3 \\
& 82 & 4 & 13 & 1 & -& 89 &8 & 3 & - & 94& 3 &2 & - \\
[Ne\,{\sc iii}]\ 15.5&123 & 0.014 & 0.008 & 0.01 & 1.80 & 80 & 0.0075 & 0.008 & 2.77 & 46 & $8\times 10^{-4}$& 0.0063&2.8 \\
&40& 13 & 45 & 2 & - & 55 & 27 & 18 & - & 75 & 3 & 22 & - \\
[Ne\,{\sc ii}]\ 12.8 & 960 & 0.025 & 0.004& 0.15 & 7.06 & 606 & 0.01 & 0.004 & 12.6 & 490 & 0.0055 & 0.0016 & 23.3 \\
& 82 & 6 & 5 & 7& - & 90 & 8 & 2 & - & 97 & 2 & 0.7 & - \\
&&& &&& &&& &&&& \\
$W$ (\%) & 0.005& 13.8&83.3 &2.895 &-&0.01 & 60.4 & 39.59&- & 0.007 &49.493 & 50.5&- \\
\hline
$V_{\rm s}$\ ($\rm km\, s^{-1}$)& 400 & 200 & 100 & 200 &-& 400 & 200 & 100 &-& 600 & 100 & 100 &- \\
$n_{\rm 0}$\ ($\rm cm^{-3}$)& 500 & 50 & 40 & 100 &-& 500 & 50 & 40 & -& 800 & 200 & 40 & - \\
$U$ & 0.01 & 0.01 & 0.01 & 0.01 &- & 0.01 & 0.01 & 0.01 &-& 0.01 & 0.001 & 0.01&- \\
$D$ (pc) & 2.67 & 3.33 & 3.33 & 20&-& 1.67 & 1.67 & 3.33 &-& 0.23 & 10 & 0.5& - \\
\hline
\noalign{\smallskip}
\noalign{\small For each line, we report as a second entry the percentage
contribution of each model in the weighted sum flux.}
\end{tabular}
}
\end{flushleft}
\label{table2}
\end{table*}
\subsection{Single-cloud models}
\label{scm}
First we perform calculations by single-cloud models,
which are compared with the observations of each galaxy in Fig.~\ref{fig1}.
The input parameters in the models
are given in Table~\ref{table1}.
Filled squares represent the data from the
Lutz et al.\ \cite*{LUTZetal98}
sample and open squares the results of model calculations.
Upper limits are not included. As shown in
Lutz et al.\ (1998, Fig. 2) they are located in
the left region of Fig.~\ref{fig1}.
The fit is very good for almost all galaxies.
A perfect fit for individual data is senseless because many different
conditions coexist in single objects.
Notice indeed that NGC 4945 is a composite object with
starburst/Seyfert 2 nuclear activity. This could explain its higher
intensity of [O\,{\sc iv}] 25.9$\mu$m relative to model predictions for
starburst only. A possible hidden AGN has also been reported in the
merger galaxy NGC 3256 \cite{KOTILAINENetal96}.
\begin{figure}[t]
\resizebox{\hsize}{!}{\includegraphics{h1255_f1.eps}}
\caption{
[O\,{\sc iv}] 25.9$\mu$m / ([Ne\,{\sc ii}] 12.8$\mu$m + 0.44 [Ne\,{\sc iii}] 15.5$\mu$m)
versus the [Ne\,{\sc iii}] 15.5$\mu$m / [Ne\,{\sc ii}] 12.8$\mu$m line ratio.
The observational data from Lutz et al.\ (1998)
are indicated by filled squares. Open symbols correspond to results
for single-cloud (squares, input parameters listed in
Table~\ref{table1}) and multiple-cloud (triangles, input
parameters listed in Table~\ref{table2}) models.
}
\label{fig1}
\end{figure}
The parameters listed in Table~\ref{table1} indicate that the
shock velocity and the preshock density are the critical parameters
to reproduce the observed
trend of the IR line ratios in Fig.~\ref{fig1}.
The velocity growth is generally followed by an increase of the preshock
density.
An ionization parameter of 0.01 was adopted for nearly all the models
except for II Zw 40 for which $U = 0.02$.
For this galaxy the effect of radiation dominates the shock effect
(see Sect.~\ref{obs}). The geometrical thickness of the clouds, $D$,
ranges between 0.01 pc for NGC 5253 and $<$ 3 pc for M 82.
Lutz et al.\ \cite*{LUTZetal98} already suggested that shocks
are the most plausible source of [O\,{\sc iv}] emission observed in starburst
galaxies. Our results
show that shocks are also necessary to reproduce the low observed values
of [Ne\,{\sc iii}]/[Ne\,{\sc ii}] (see Sect.~\ref{physic} for details).
Moreover, Fig.~\ref{fig1} shows that the objects with [Ne\,{\sc iii}]/ [Ne\,{\sc ii}]
$<$ 1
are well fitted by models characterized by high velocity and density clouds.
However, previous investigations on the physical conditions in the ISM of
NGC 253 \cite{CARRALetal94},
NGC 3256 \cite{CARRALetal94,RIGOPOULOUetal96},
and M 82 \cite{HOUCKetal84,LORDetal96}
found relatively low electronic densities on the basis of the
[S\,{\sc iii}] 33$\mu$m/18$\mu$m and [O\,{\sc iii}] 88$\mu$m/52$\mu$m line ratios,
and low velocities derived from the FWHM of the IR line profiles.
\begin{table}
\caption[]{Observed vs. calculated IR line fluxes$^{\rm a}$}
\begin{flushleft}
{\footnotesize
\begin{tabular}{lrrrrrr}
\hline
\hline
Line & \multicolumn{2}{c}{NGC 253} & \multicolumn{2}{c}{NGC 3256} & \multicolumn{2}{c}{M 82} \\
($\mu$m) & Obs$^{\rm b}$ & Cal & Obs$^{\rm b}$ & Cal & Obs$^{\rm b}$ & Cal \\
\hline
[O\,{\sc iii}] 88 & 0.10$\pm$0.02 & 0.1 & 0.4$\pm$0.3 & 0.5 & 0.6$\pm$0.2 & 0.4 \\
[O\,{\sc i}] 63 & 0.9$\pm$0.1 & 1.0 & 1.2$\pm$0.4 & 1.7 & 1.1$\pm$0.4 & 1.2 \\
[O\,{\sc iii}] 52 & 0.20$\pm$0.04 & 0.2 & $<$1.3 & 1.2 & 0.7$\pm$0.2 & 0.5 \\
[Si\,{\sc ii}] 35 & 1.1$\pm$0.2 & 6.2 & 1.9$\pm$0.5 & 5.6 & 1.0$\pm$0.4 & 5.4 \\
[S\,{\sc iii}] 33 & $<$1.3 & 0.3 & 1.2$\pm$0.3 & 1.6 & 1.1$\pm$0.3 & 1.2 \\
[S\,{\sc iii}] 18 & \ldots & 0.2 & 0.9$\pm$0.2 & 1.5 & 0.7$\pm$0.3 & 0.6 \\
\hline
\hline
\noalign{\smallskip}
\noalign{\small $^{\rm a}$ Fluxes are relative to [C\, {\sc ii}] 158$\mu$m.}
\noalign{\small $^{\rm b}$ Observed line fluxes from Carral et al.\
\cite*{CARRALetal94} for NGC 253 and NGC 3256,
Rigopoulou et al.\ \cite*{RIGOPOULOUetal96} for NGC 3256,
and Lord et al.\ \cite*{LORDetal96} for M82.}
\end{tabular}
}
\end{flushleft}
\label{table3}
\end{table}
\subsection{Multi-cloud models}
Many different conditions coexist in the ISM of single galaxies.
Therefore, we performed multi-cloud model calculations to reproduce all the
observed IR line ratios available in the literature for the starburst galaxies
NGC 253, NGC 3256, and M 82.
In these models, we assume that both {\it low} and {\it high}
density-velocity (d-v) clouds are present in single objects.
In Table~\ref{table2}, the calculated IR line fluxes are given for
single-cloud models and for the weighted sums of the multi-cloud models.
Notice that the line fluxes are calculated at the
emitting nebula while the observations are given at Earth.
The input parameters of single-cloud models are also listed.
Line fluxes are proportional to the square of the gas emitting density, and
compression downstream increases with the shock velocity. Thus,
the line fluxes computed from high d-v models (i.e. A1, B1, and C1) are higher
than those calculated from low d-v models (A2, A3, A4, B2, B3, C2, and C3)
by some orders of magnitude, except for lines [O\,{\sc i}]\ and [C\,{\sc ii}].
The gas in high d-v models is ionized to very high temperatures
(Fig.~\ref{fig2}b), and so low-level lines like [O\,{\sc i}]\ and [C\,{\sc ii}]\ are weak.
Recall that the first ionization potential of C is lower than
that of H (and Ne). Consequently, [C\,{\sc ii}]\ lines, as well as [O\,{\sc i}]\ lines,
are emitted by relatively cold gas and can be weak in models
which show strong [Ne\,{\sc ii}]\ lines.
In order to fit the observations, a balance must be achieved between high
and low d-v models in the multi-cloud spectra, and the final result is a
weighted sum (A$_{\rm v}$ in Table 2) with the weights chosen adequately.
For each model, the weight
$W$ represents the relative number of clouds (in \%) characterized by that
model. The weights of low d-v models must be by a factor of $10^4 - 10^5$
higher than for high d-v models.
In other words, the number of high d-v clouds must be very low compared with
the number of low d-v clouds.
In Table~\ref{table2}, for each emission line, the second entry
corresponds to the percentage contribution of the single-cloud model to
the line flux, taking into account the corresponding weight.
The high d-v clouds
contribute up to 90\% to the [O\,{\sc iv}]\ and [Ne\,{\sc ii}]\ emission lines,
whereas the [Ne\,{\sc iii}]\ line is produced at a same level both by low and high
d-v clouds for M 82 and NGC 3256.
The others IR lines are mainly emitted by the low d-v clouds except
the [O\,{\sc iii}]\ 52 line.
The results of multi-cloud model calculations are compared to
the observed IR line ratios in Table~\ref{table3} and plotted in
Fig.~\ref{fig1} (open triangles).
The fit is good for all line ratios (especially for the density critical
ones, see sect.~\ref{scm}), except for the [Si\,{\sc ii}] line which is
overpredicted by a factor of about 5.
This might indicate that silicium is depleted and locked in
dust grains. Adopting a relative abundance Si/H lower by a factor of $\sim$ 5
than cosmic will not affect the predictions for the other line ratios
listed in Table~\ref{table3}.
Most of the IR lines reported in Table~\ref{table2}
(e.g. [O\,{\sc i}], [C\,{\sc ii}], etc)
are mainly produced by the low d-v clouds.
Thus, in the weighted sum, the IR lines ratios [S\,{\sc iii}] 33$\mu$m/18$\mu$m
and [O\,{\sc iii}] 88$\mu$m/52$\mu$m
are determined by low d-v models, while the [O\,{\sc iv}]/([Ne\,{\sc ii}]+0.44[Ne\,{\sc iii}])
and [Ne\,{\sc iii}]/[Ne\,{\sc ii}] line ratios are mainly determined by the high d-v cloud
component.
Once fitted the line ratios, it is now possible to examine the
FWHM of the line profiles, because the weights adopted in Table~\ref{table2}
for high d-v models are very low and could lead to hardly observable broad
components in the line profiles.
The calculated line profiles for M 82 and NGC 253 are
shown in Fig~\ref{lineprof}. The effect of galaxy rotation is included
in the calculation. Rotation velocities are about 76 $\rm km\, s^{-1}$\ and 189 $\rm km\, s^{-1}$\ for
M 82 and NGC 253 respectively.
Two significant lines are chosen for each galaxy: [O\,{\sc iv}]\ and [Ne\,{\sc iii}]\ for M 82,
and [O\,{\sc iv}]\ and [O\,{\sc iii}]\ 88 for NGC 253. The [O\,{\sc iii}]\ 88 line is chosen because it
represents the case of a low contribution of the high d-v model (C1)
relatively to a strong contribution of the low d-v model (C3).
Single-cloud model profiles are also plotted in the figures using their
percentage contribution reported in Table~\ref{table2}.
The broad component from the high d-v clouds dominates the weighted averaged
profile of [O\,{\sc iv}], while in the [O\,{\sc iii}]\ 88 and [Ne\,{\sc iii}]\ profiles
the broad component is faint and hardly observable, due to the low
contribution from high d-v clouds. For M 82, the predicted FWHM of the
[O\,{\sc iv}]\ profile is $\sim$ 300 $\rm km\, s^{-1}$, in good agreement with the observed
value (FWHM $\sim$ 360 $\rm km\, s^{-1}$) reported by Lutz et al.\ \cite*{LUTZetal98}.
\begin{figure}[t]
\vspace{-1cm}
\resizebox{\hsize}{!}{\includegraphics{h1255_f2.eps}}
\vspace{-2.5cm}
\caption{Predicted line profiles for NGC 253 (top) and M 82 (bottom) using
multi-cloud models. The final profile (solid line) is the weighted sum
of different cloud components: high density-velocity clouds (dotted line)
and low density-velocity clouds (long-dashed line: $V_{\rm s}$\ = 200 $\rm km\, s^{-1}$;
short-dashed line: $V_{\rm s}$\ = 100 $\rm km\, s^{-1}$)}
\label{lineprof}
\end{figure}
\begin{figure*}[]
a)
\psfig{figure=h1255_f3a.eps,height=8.8cm}
b)
\psfig{figure=h1255_f3b.eps,height=8.8cm}
\caption{
The physical conditions across low-density (a, NGC 5253) and high-density
(b, NGC 253) clouds are shown in three panels, each one with a vertical
line indicating the center of the emitting clouds.
The electron temperature (dotted line) and electron density (solid line)
appear in the top panel.
The fractional abundances of $\rm O^{+3}$ (long-dashed line),
$\rm Ne^{+2}$ (solid line), and $\rm Ne^{+1}$ (short-dashed line) are
displayed in the central panel.
The ionization rates for $\rm Ne^{+2}$ (in logarithmic units) due to
electron collisions C (long-dashed line), to the stellar radiation P
(solid line) and to the diffuse radiation D (short-dashed line) are
shown in the bottom panel.
The shock front is on the left side and the edge of the cloud
photoionized by the stellar radiation is on the right of each panel.
The horizontal axis scale is logarithmic and symmetric in order to show
an equal view
of both sides of the emitting cloud
}
\label{fig2}
\end{figure*}
\section{The physical conditions across emitting clouds}
\label{physic}
To better understand the role of the input parameters in determining
the physical conditions of the emitting gas, the distribution of some
important physical quantities across the cloud is given in
Figures~\ref{fig2}a and \ref{fig2}b respectively for two significant
models listed in Table~\ref{table1}: NGC 5253, which represents the
low density-velocity cloud component, and the high density-velocity
component of NGC 253. Each figure shows three
panels where the shock front is on the left and the photoionized edge
is on the right. The top panel displays the electron temperature, $T_{\rm e}$,
and density, $n_{\rm e}$. The fractional abundances of the ions $\rm O^{+3}$/O,
$\rm Ne^{+2}$/Ne, and $\rm Ne^{+1}$/Ne are shown in the central panel.
The ionization rates for $\rm Ne^{+2}$ due to the primary radiation
(P), to the diffuse radiation (D), and to electronic collisions (C)
are given in the bottom panel.
The [O\,{\sc iv}] line is always
emitted from the shock dominated region where the temperatures are high,
with no contribution from the photoionized zone.
In particular, Fig.~\ref{fig2}a shows that the critical temperatures
are about $10^5$ K.
For NGC 5253 the shock effect becomes negligible at a
distance of about $10^{15}$ cm from the shock front, and most of the
physical conditions in the cloud are then dominated by the effect of
stellar radiation which leads to [Ne\,{\sc iii}]/[Ne\,{\sc ii}] $>$ 1.
On the other hand, due to strong compression in the NGC 253 gas clouds
(Fig.~\ref{fig2}b), the temperature rapidly drops to $T_{\rm e}$\ $< 10^4$ K.
This can be explained by the high density ($\sim 10^5$ $\rm cm^{-3}$) of the gas
in the innermost region of the cloud which strengthens the cooling rate.
The role of the primary radiation is to keep the gas at a temperature
slightly below $10^4$ K corresponding to a relatively high $\rm Ne^{+1}$/Ne
compared to $\rm Ne^{+0}$/Ne and $\rm Ne^{+2}$/Ne.
Therefore, for this galaxy, [Ne\,{\sc iii}]/[Ne\,{\sc ii}] is $<$ 1.
Owing to the high optical thickness of the gas, the stellar radiation
flux reaching the inner edge of the cloud rapidly decreases,
the size of the $\rm Ne^{+2}$ photoionized zone is reduced, and the diffuse
radiation effect prevails accross the cloud.
For NGC 253, temperatures as high as $7 \times 10^6$ K deduced
from the ROSAT PSPC data (Heckman 1996 and references
therein) are in agreement with $5.4 \times 10^6$ K obtained by the
model in the postshock region (Fig.~\ref{fig2}b).
Precursor soft X-rays radiation from the hot gas could affect the
ionization conditions in the upstream gas clouds. However, due to the
adiabatic jump and compression, the densities $n$ upstream are lower
than in the downstream gas by at least a factor of 4.
Considering that line intensities depend on $n^2$, the contribution of
the line emission from upstream gas to the final spectrum will be small.
\section{Comparison with the observed starburst properties}
\label{obs}
One might interpret the observed trend in Fig.~\ref{fig1} as a
continuous sequence between two extreme cases:
starburst regions where shocks in high-velocity clouds are
{\em necessary} to reproduce the observed critical IR line ratios
(bottom left part of the diagram) and the ``radiation-dominated'' ones
(upper right part of the diagram), where the shock contribution is
relatively less important, because formed only in low-velocity clouds.
Let us now check wether the critical parameters derived from the models
along this sequence are in agreement with the observed properties
of the starburst galaxies.
\begin{table}
\caption[]{Galaxian properties}
\begin{flushleft}
\begin{tabular}{llrrr}
\hline
\hline
Galaxy & Morph.$^{\rm a}$ & $M_{\rm abs}^{\rm a}$ & $d^{\rm b}$ & $\log L_{\rm X}^{\rm b}$ \\
& & &(Mpc)& ($\rm erg\, s^{-1}$) \\
\hline
II Zw 40 & BCDG & $-$14.5 & 9.2 & ... \\
NGC 5253 & Irr & $-$16.2 & 4.1 & 38.6 \\
& & \\
NGC 3690 B/C & Merger & $-$21.1 & 42.4 & 41.0 \\
NGC 3256 & Merger & $-$22.2 & 33.2 & 41.7 \\
M 82 & Sd & $-$19.0 & 4.3 & 39.7 \\
M 83 & SABc & $-$21.2 & 5.1 & 40.2 \\
NGC 253 & SABc & $-$20.9 & 3.3 & 39.9 \\
NGC 4945 & SBcd & $-$20.1 & 5.1 & 39.9 \\
\hline
\hline
\noalign{\smallskip}
\noalign{\small $^{\rm a}$ From the LEDA database}
\noalign{\small $^{\rm b}$ The
distance ($d$) is derived from the Galactic Standard of Rest (GSR)
velocity, using a Hubble constant $H_0 = 75$ km s$^{-1}$ Mpc$^{-1}$,
except for the distance of NGC 5253 taken from Saha et al.\ (1995).
The X-ray luminosity is derived from the flux reported in the
{\em ROSAT} catalog (Brinkmann et al.\ 1994) or in
the {\em EINSTEIN} catalog (Fabbiano et al.\ 1992).}
\end{tabular}
\end{flushleft}
\label{table4}
\end{table}
Galaxies in the first group (the last six galaxies in Table~\ref{table4})
are massive
and chemically evolved spiral galaxies which exhibit a single
energetic and compact starburst in their nucleus, and are commonly
named Starburst Nucleus Galaxies (SBNGs; Balzano 1983;
Coziol et al.\ 1994).
Evidence for powerful large-scale starburst-driven superwinds in the nuclear
region of SBNGs
(like M 82, NGC 253, NGC 4945 and NGC 3690 in the present sample)
has been reported by Heckman et al.\ \cite*{HAM90}.
Moreover, the strong X-ray emission
($L_{\rm X} \sim 10^{40}-10^{42}$, see Table~\ref{table4})
in SBNGs provides strong evidence
for the formation of hot superbubbles due to the combined action
of stellar winds from massive stars and supernova
explosions in the central starburst cluster.
Expanding velocities as high as $\sim 500 - 600$ $\rm km\, s^{-1}$\ have been
estimated for the ionized gas in
M 82 \cite{SB-H98},
NGC 4945 \cite{MOORWOODetal96},
and NGC 253 \cite{H96}.
In the central region of SBNGs,
part of the ISM is overpressured relative to the ISM in the Milky Way
\cite{HAM90,CARRALetal94},
due to compression by superwinds.
The nuclear starbursts usually contain
many compact radio sources associated with bright and small
supernova remnants
(Muxlow et al.\ 1994;
Huang et al.\ 1994 for M 82;
Ulvestad \& Antonucci 1997 for NGC 253;
Zhao et al.\ 1997 for NGC 3690).
Such a population of dense and compact objects is generally
attributed to the high pressure in the central region of SBNGs which allows
the formation of dense ionized clouds (see e.g. Carilli 1996).
The galaxies in the second group (the first two galaxies
in Table~\ref{table4})
are mostly metal-poor dwarf irregular galaxies with multiple star-forming
regions distributed randomly, and are known as H{\sc ii}\ galaxies \cite{F80}.
The arche\-types of this class of galaxies are II Zw 40 and NGC 5253. Their
starbursts are very young ($\leq$ 4 Myr), as proved by the
numerous Wolf-Rayet stars still present in II Zw 40 \cite{KS81,VC92}
and NGC 5253 \cite{SCHAERERetal97,SCK99}, and only contain a few supernova
remnants (Vanzi et al.\ 1996 for II Zw 40;
Beck et al.\ 1996 for NGC 5253). Moreover,
there is no evidence for powerful starburst-driven superwinds
in these galaxies (Martin 1998 for II Zw 40;
Marlowe et al.\ 1995 for NGC 5253) which are faint
or even undetected in X-rays
(see Table~\ref{table4}).
Thus, the distinction between H{\sc ii}\ galaxies and SBNGs can be understood
in terms of temporal evolution of the starbursts.
In the former galaxies (II Zw 40 and NGC 5253), the star-forming event
is still young enough to contain a high proportion of massive O and
Wolf-Rayet stars.
These populations of hot ($T_{\rm eff} \sim 60 000$ K) stars produce
large amounts of ionizing radiation, but did not have the time to
evolve off the main sequence to produce supernova explosions.
In SBNGs, like M 82 or NGC 253, the starburst is much older
($\geq$ 20 Myr, e.g. Engelbracht et al.\ 1998),
and the most massive stars had the time to evolve
through supernovae.
As shown by the models of Leitherer \& Heckman \cite*{LH95},
the ratio between the mechanical energy and the ionizing Lyman continuum,
both injected in the ISM by the massive star population, rapidly increases
by a factor of $\sim$ 100 between $\sim$ 5 and 20 Myr.
This explain the difference between the ``radiation-dominated'' H{\sc ii}\
galaxies and the ``shock-dominated'' SBNGs.
The presence of high density-velocity clouds in SBNGs, which host a
relatively ``old'' starburst, require large quantities of mechanical
energy to power large-scale superwinds. These energetic events disrupt
molecular clouds through shock waves and drive them out of the nucleus
at high velocities.
\section{Conclusions}
\label{conclu}
The observed relation between [O\,{\sc iv}]/([Ne\,{\sc ii}]+ 0.44[Ne\,{\sc iii}])
and [Ne\,{\sc iii}]/[Ne\,{\sc ii}], and single object IR line ratios are well
reproduced by models accounting consistently for both shock and
photoionization effects occurring in starburst regions.
The critical parameters which reproduce the observed trend of the IR
line ratios are the preshock density and the shock velocity which are
consistently correlated.
Most of the shocks are produced in low density-velocity ($n_{\rm 0}$ $\sim$ 100 $\rm cm^{-3}$\
and $V_{\rm s}$\ $\sim 50 - 100$ $\rm km\, s^{-1}$) clouds which represent the bulk of
the ionized gas in starburst galaxies.
However, even if they are by many orders less numerous, high-velocity
($\sim 400 - 600$ $\rm km\, s^{-1}$) shocks in dense ($\sim 500 - 800$ $\rm cm^{-3}$) clouds are
necessary to reproduce the critical IR line ratios observed in the
low-excitation SBNGs.
These model predictions are in good agreement with the observed ISM
properties in starburst galaxies. SBNGs are the only ones to exhibit
powerful starburst-driven superwinds and highly pressured ISM.
On the contrary, the high-excitation H{\sc ii}\ galaxies do not show any
clear signature of large scale outflows of gas.
This difference between H{\sc ii}\ galaxies and SBNGs can
be interpreted in terms of temporal evolution of their starbursts.
In the former, the starburst is too young to release large amounts
of mechanical energy through supernova explosions whereas in the
latter, characterized by older starbursts, the kinetic energy
compete with the ionizing radiation in the heating of the ISM.
\begin{acknowledgements}
M.C. is grateful to the Instituto Astron\^omico e Geof\' \i sico, USP
for warm hospitality.
We thank Emmanuel Davoust for a careful reading of the
manuscript and the anonymous referee for helpful comments and suggestions.
This research has made use of the Lyon-Meudon Extragalactic Database
(LEDA) supplied by the LEDA team at the CRAL - Observatoire de Lyon (France).
This work was partially supported by FAPESP and PRONEX/\-Finep,
Brazil.
\end{acknowledgements}
|
\section{Introduction.}
The study of quantum field theories in the perturbative regime can sometimes
become a very challenging and arduous issue to deal with, especially when one
needs to go on calculating multi-loop Feynman integrals. There are several
techniques that have been developed and refined in the course of time, but
usually the calculations are rather involved and analytic results are difficult
to obtain. Depending on the approach one uses to tackle the problem, the
obstacles to surmount can be more or less demanding. For example, if one uses
the standard Feynman parametrization, the integral over momentum space is
straightforward --- in the sense that nowadays one can easily fint them out even
tabulated in textbooks on quantum field theories --- but the resulting
parametric integrals that are left over become quite complicated and cumbersome
to tackle, especially with increasing number of internal momenta in the loops.
Of course, if one is clever enough, he/she may be able to overcome some of
these hurdles and even perform really hard calculations
\cite{kramer,gonsalves}.
There are, of course, other techniques which have been considered in the
literature, e.g., integration by parts and Gegenbauer polynomial method in
configuration space \cite{russo}; the Mellin-Barnes representation for massive
propagators \cite{boos} to cite just a few of them \cite{davyd}, each one of
those with its own strengths and weaknesses.
On the other hand, there is this novel technique known as negative dimensional
integration method (NDIM{}), first devised by Halliday and Ricotta
\cite{halliday}. It has as its starting point the principle of \an{} continuation{}, it has
the remarkable property of being equivalent to Grassmannian integration in
(positive) $D$-dimensional space \cite{halliday2}, it requires only the
integration of Gaussian-like integrals and solving of linear algebraic
equations and is suitable to deal with massless \cite{lab} or massive
\cite{box} diagrams in on- and off-shell regimes \cite{preprint} and even to
carry out light-cone gauge integrals, with added troublesome gauge-dependent
poles \cite{probing}.
Our aim in this work is to demonstrate the simplicity of NDIM calculating some
scalar three-point integrals, with five and six massless propagators, at
two-loop level and some two-point three-loop integrals also in the massless
case. The outline for our work is as follows: in section II we present a
detailed calculation of a two-loop integral in the NDIM{} approach; in section
III we write down the results for the remaining two- and three-loop scalar
integrals, while in section IV we give our concluding remarks about this work.
\section{Two-loop three-point vertex with six massless propagators.}
This computation is performed following just a few simple steps
\cite{lab}.
Fist of all, let us consider the Gaussian-like integral,
\begin{eqnarray}
\label{gauss}
F &=& \int\int d^D\!r\; d^D\!q \;\;\exp\left[-\alpha q^2 -\beta r^2
-\gamma (q-r)^2 -\delta (p+q)^2 -\omega(q-k)^2-\theta (q-r-k)^2\right]\nonumber\\
&=& \left(\frac{\pi^2}{\lambda}\right)^{D/2} \exp{\left[ -
\frac{1}{\lambda}\left(\alpha\beta\delta+ \alpha\gamma\delta+
\beta\gamma\delta+\alpha\theta\delta\right)p^2\right]} ,
\end{eqnarray}
where the arguments in the exponential function of the integrand correspond to
propagators in the diagram of Fig.1, with $k^2=t^2=0$ satisfying the on-shell
condition. For the sake of compactness, we define
\[
\lambda=\alpha\beta+\alpha\gamma+\alpha\theta+\beta\gamma+\beta\delta+
\gamma\delta+\theta\delta+\beta\omega+\gamma\omega+\theta\omega+\beta\theta.
\]
The parameters $(\alpha,\beta,\gamma,\delta,\omega,\theta)$ are quite general
and arbitrary except that their real parts must be positive to make sure we
have well-defined objects over the whole integration region.
Expanding the exponential in the second line of (\ref{gauss}) in Taylor series
and using the multinomial expansion for the argument of the exponential, we
obtain,
\begin{eqnarray}
F &=& \pi^D\sum_{\{x,y,z=0\}}^\infty \frac{(-p^2)^{\Sigma x} (-\Sigma
x-D/2)! \alpha^a \beta^b \gamma^c \delta^d \omega^e \theta^f}{x_1!\cdot\cdot
\cdot x_4!y_1!\cdot\cdot\cdot y_9!z_1!z_2!} \delta_{-\Sigma x-D/2,\Sigma
y+\Sigma z},
\end{eqnarray}
where we have a fifteen-fold sum, with $\Sigma x=x_1+x_2+x_3+x_4$ and we
define,
\begin{eqnarray}
a&=&x_1+x_2+x_4+y_1+y_2+y_3\:,\nonumber\\
b&=&x_1+x_3+y_1+y_4+y_5+y_8+z_2\:,\nonumber\\
c&=&x_2+x_3+y_2+y_4+y_6+y_9\:,\nonumber\\
d&=&x_1+x_2+x_3+x_4+y_5+y_6+y_7\:,\nonumber\\
e&=&y_8+y_9+z_1\:,\nonumber\\
f&=&x_4+y_3+y_7+z_1+z_2\:.
\end{eqnarray}
On the other hand, taking the Taylor expansion of the exponential that is under
the integration sign in (\ref{gauss}), we have,
\begin{equation}
\label{proj} F = \sum_{i,j,l,m,n,s=0}^\infty \frac{(-1)^{i+j+l+m+n+s}}{i!j!l!m!n!s!}
\alpha^i\beta^j\gamma^l\delta^m\omega^n\theta^s {\cal S}_{NDIM}^1, \end{equation}{}
where ${\cal {S}}_{NDIM}^1$ is the relevant integral in negative $D$, defined
by
\begin{equation} \label{Indim} {\cal S}_{NDIM}^1 = \int d^D\!q\;d^D\!r
(q^2)^i(r^2)^j(q-r)^{2l}(q+p)^{2m} (q-k)^{2n}(q-r-k)^{2s} .
\end{equation}
Now comparing both expressions for the original integral $F$ we are led to the
conclusion that, in order to have equality between these two expressions, the
factor ${\cal S}_{NDIM}^1$ must be given by the multiple series,
\begin{equation}
\label{geral}
{\cal S}_{NDIM}^1 = P_1 \sum_{x,y,z=0}^\infty
\frac{1}{x_1!\cdot\cdot\cdot x_4!y_1!\cdot\cdot\cdot y_9!z_1!z_2!}
\; \delta_{a,i}\;\delta_{b,j}\;\delta_{c,l}\;\delta_{d,m}\;\delta_{e,n}
\;\delta_{f,s} ,
\end{equation}
with
\[ P_1 = (-\pi)^D(p^2)^\sigma\Gamma(1+i)\Gamma(1+j)\Gamma(1+l)\Gamma(1+m)\Gamma(1+n)\Gamma(1+s)
\Gamma(1-\sigma-\frac{1}{2} D),
\]
and for convenience we use the definition $\sigma=i+j+l+m+n+s+D$. The system we
must solve has, therefore, fifteen ``unknowns'' (the indices $x,y,z$) and only
seven equations --- six of which comes from comparing the exponents of
$(\alpha,\beta,\cdots)$ and one from the multinomial expansion, namely, $\Sigma
y+\Sigma z=-\Sigma x-D/2$. So, such a system can only be solved in terms of
some of these ``unknowns'', and we conclude, from the combinatorics, that there
are $C^{15}_7=6435$ systems $(7\times 7)$ of linear algebraic equations, i.e.,
there are $6435$ different ways of choosing these ``unknowns''. Of this grand
total possible solutions, $4113$ yield trivial solutions and for this reason
present no interest at all. Yet, the remaining $2322$ solutions are quite a big
number! Fortunately, not all the remnant non-vanishing solutions need to be
considered in order to solve the Feynman integral. In fact, only linearly
independent series will be needed. In our massless case and where there is only
one non-vanishing external momentum, the argument of such series is unit. Since
NDIM provides, in general, hypergeometric series (see, for example ref.
\cite{probing} where we have double and triple series of such), we obtain in
this case $_pF_q(\cdots|1)$, and from the theory of generalized hypergeometric
functions \cite{bailey,luke}, a linear combination of linearly independent
series of this type has up to $p$ series.
Consider then, the solution where the indices $y_2, y_3, y_4, y_5, y_7, y_8, y_9,
z_2$ are left undetermined. Solving this system we get,
\begin{eqnarray}
x_1&=&-l+m+n-s+y_2+y_3+y_4-y_5-y_8+z_2\:,\nonumber\\
x_2&=&i+l+n+D/2-y_2-y_9+y_5+y_7+z_2\:,\nonumber\\
x_3&=&j+l+s+D/2-y_4-z_2\:,\nonumber\\
x_4&=&-n+s-y_3-y_7+y_8+y_9-z_2\:,\nonumber\\
y_1&=&-m-n-D/2-y_2-y_3-y_4-z_2\:,\nonumber\\
y_6&=&-\sigma+m-y_5-y_7\:,\nonumber\\
z_1&=&n-y_8-y_9\:,
\end{eqnarray}
which substituted in (\ref{geral}) yields an eight-fold summation. Seven of
these can be rewritten in terms of gamma functions. The procedure is as
follows: Firstly, the factorials must be converted into gamma functions and
these into Pochhammer symbols, defined by,
\begin{equation}
(n|k) \equiv (n)_k = \frac{\Gamma(n+k)}{\Gamma(n)} .
\end{equation}
Secondly, use the properties,
\begin{equation}
\label{prop}
(n|j+k) = (n+j|k)(n|j),\qquad\qquad
(n|-k) = \frac{(-1)^k}{(1-n|k)},
\end{equation}
which follow from the definition above. Then, it is possible to rewrite some
series in terms of Gaussian hypergeometric function{}s, which can, in general, be summed when its
argument is unit \cite{bateman}.
\begin{equation}
\label{somaf21}
_2F_1(a,b;c|1) = \frac{\Gamma(c)\Gamma(c-a-b)}{\Gamma(c-a)\Gamma(c-b)}.
\end{equation}
Exception for this occurs when the parameters are such that $c-a-b=0$.
Therefore, if we rearrange these series conveniently, we can sum them, and
applying this procedure for the series defined by indices $y_4, y_3, y_2, y_8,
y_9, y_7, y_5$, they all can be summed out. The last series, defined by index
$z_2$, however cannot be summed in this fashion since it is a $_3F_2$ function,
which is summable only when it is of Saalsch\"utzian type \cite{bailey}, and
this is not our present case.
Many of the gamma functions that appear in this process simplify among
themselves and in the end, what remains are fourteen gamma functions
distributed in a fraction, i.e., seven in the numerator and seven in the
denominator multiplying the $z_2$ series. This is the result in the {\it
negative} dimension region and positive exponents $(i,j,l,m,n,s)$, see
(\ref{proj}). Now we must be able to bring this result back, that is,
analytically continue it, to our real physical world, that of positive $D$. This
is carried out very easily: group the gamma functions in convenient Pochhammer
symbols and use the second property (\ref{prop}).
Considering now the solution where $x_1, x_2, x_3, y_1, y_5, y_6, y_9, z_2$ are
left undetermined, we get another $_3F_2(a,b,c;e,f|z)$ function. Let us recall
that such functions have three singular points, namely $(0,1,\infty)$. Here, we
are interested in solutions in the vicinity of $z=1$. No$\!\!\!$/rlund \cite{norlund}
proved that when the difference between $e$ and $f$ is an integer, some of the
series concide or are without meaning. This is precisely our case. This means
that our final result for the Feynman integral shall exhibit only two linearly
independent generalized hypergeometric function{}s of the type $_3F_2$, instead of the usual three.
We get, as a result, then
\begin{equation}
{\cal S}_{NDIM}^1 = \pi^D (p^2)^\sigma \{{\bf A}\:_3F_2(a_1,b_1,c_1;e_1,f_1|1)+{\bf
B}\:_3F_2(a'_1,b'_1,c'_1;e'_1,f'_1|1)\}
\end{equation}
where
\begin{eqnarray}
{\bf A}&\equiv & (-m|\sigma)(\sigma+D/2|-2\sigma-D/2) (-j|j+l+s+D/2)(-l|j+l+s+D/2)
\nonumber\\
&\times &(-i-j-l-s-D/2|i)(j+l+s+D|-j-l-s-D/2+m+n),
\end{eqnarray}
and
\begin{eqnarray}
{\bf B}&\equiv & (-i|\sigma)(-m|\sigma)(\sigma+D/2|-2\sigma-D/2)\nonumber\\
&\times &(-j|-l-s-D/2)(j+l+s+D|-j-D/2) (-l-s|j+l+s+D/2),
\end{eqnarray}
where the set of parameters are defined as
$(a_1,\,b_1,\,c_1;\,e_1,\,f_1)=(-s,\,m+n+D/2,\,-j-l-s-D/2;\,-i-j-l-s-D/2,\,1-j-s-D/2)$,
and
$(a'_1,\,b'_1,\,c'_1;\,e'_1,\,f'_1)=(-\sigma,\,-l,\,-j-l-s-D/2;\,-l-s,\,1+i-\sigma)$.
Note that $(e_1-f_1)$ is an integer. Observe also that, in the special case
where all the exponents are equal to minus one, the hypergeometric function
$_3F_2$ reduces to $_2F_1$, which can be summed again, using (\ref{somaf21}).
So, for this special case of negative unity exponents, the final result is
expressed in terms of gamma functions only.
Let us point out here that our final result agrees with that given by
Davydychev and Osland \cite{davyd}, i.e., meaning that equation (27) in ref.
\cite{kramer} disagrees with ours. The difference is in the $9\zeta(4)$ factor
in \cite{kramer} which should correctly be read $9\zeta(2)$. This gives the
correct result for $i=j=l=m=n=s=-1$ and $D=4-2\epsilon$.
\section{Results for some two and three-loop massless scalar integrals.}
The scalar integrals for the remaining diagrams (Fig.2-7) are, in negative $D$,
given by (number superscripts 2-7 label the corresponding diagrams):
\begin{equation} {\cal S}_{NDIM}^2 = \int d^D\!q\;d^D\!r \; (r^2)^i (q^2)^j
(p+r)^{2l}(q+k)^{2m} (p+r-q)^{2n}, \end{equation}
\begin{equation} {\cal S}_{NDIM}^3 = \int d^D\!q\;d^D\!r \; (r^2)^i (q^2)^j
(k-q-r)^{2l}(k-r)^{2m} (r+p)^{2n}, \end{equation}
\begin{equation} {\cal S}_{NDIM}^4 = \int d^D\!q\;d^D\!r\;d^D\!k \; (r^2)^i (k^2)^j
(q^2)^l (p-q-r)^{2m} (r-k)^{2n}, \end{equation}
\begin{equation} {\cal S}_{NDIM}^5 = \int d^D\!q\;d^D\!r\;d^D\!k \; (q^2)^i (k^2)^j
(r^2)^l (p-r)^{2m} (r-q)^{2n} (p-r-k)^{2s}, \end{equation}
\begin{equation} {\cal S}_{NDIM}^6 = \int d^D\!q\;d^D\!r\;d^D\!k \; (r^2)^i (k^2)^j
(q^2)^l (r+k)^{2m}(p-k-r)^{2n} (r+k-q)^{2s}, \end{equation}
and
\begin{equation} {\cal S}_{NDIM}^7 = \int d^D\!k\;d^D\!q\;d^D\!r\;
(q^2)^i(k^2)^j (p-k-r)^{2l} (k-q)^{2m} (p-k)^{2n} (r^2)^s .\end{equation}
All these can be evaluated in a similar way following the steps of the previous
section. For this reason we quote here only the final results, after
continuation to positive dimension and negative exponents of propagators:
\begin{equation}
{\cal S}_{NDIM}^2 = \pi^D (k^2)^\rho \{{\bf C}\:_3F_2(a_2,b_2,c_2;e_2,f_2|1)+{\bf
D}\:_3F_2(a'_2,b'_2,c'_2;e'_2,f'_2|1)\} ,
\end{equation}
where
\begin{eqnarray}
{\bf C}&\equiv & (-i|\rho)(\rho+D/2|-2\rho-D/2)(i+D/2-\rho|\rho)(\rho-i-l|-j-D/2) \nonumber\\
&\times &(-j|-m-n-D/2)(-n|j+m+n+D/2)\;
\end{eqnarray}
and
\begin{eqnarray}
{\bf D}&\equiv & (-i|\rho)(-l|\rho)(\rho+D/2|-2\rho-D/2)(j+m+n+D|-j-m-D/2)\nonumber\\
&\times &(-j-m|-n-D/2)(-n|j+m+n+D/2)\; ,
\end{eqnarray}
with the set of parameters defined by
$(a_2,\,b_2,\,c_2;\,e_2,\,f_2)=(-m,\,i+D/2,\,-j-m-n-D/2;\,i+D/2-\rho,\,1-m-n-D/2)$
and $(a'_2,\,b'_2,\,c'_2;\,e'_2,\,f'_2)=(-j,\,-\rho,\,-j-m-n-D/2;1+l-\rho,\,-j-m)$
and $\rho=\sigma-s$. Note that here also we have a linear combination of two $_3F_2$
hypergeometric function{}s, since $(e_2-f_2)$ is an integer and No$\!\!\!/$rlund's theorem applies.
\begin{eqnarray}
{\cal S}_{NDIM}^3 &=& \pi^D (k^2)^\rho
(-l|\rho)(\rho+D/2|-2\rho-D/2)(i+l+D/2-\rho|\rho)(-j|j+n+D/2) \nonumber\\
&&\times (-n|j+n+D/2)(j+n+D|-2j-2n-3D/2),
\end{eqnarray}
\begin{eqnarray} {\cal S}_{NDIM}^4 &=& \pi^{3D/2}(p^2)^{\rho'} (-j|j+n+D/2)(-n|j+n+D/2)
(-l|l+m+D/2)\nonumber\\
&&\times (-m|l+m+D/2) (\rho'+D/2|-2\rho'-D/2)(j+n+D|i)\nonumber\\
&&\times (-j-n-D/2|-i),
\end{eqnarray}
where $\rho'=\rho+D/2$.
\begin{eqnarray}
{\cal S}_{NDIM}^5 &=& \pi^{3D/2}(p^2)^{\sigma'} (-i|i+n+D/2)(-n|i+n+D/2)
(-j|j+s+D/2)\nonumber\\
&& \times (-s|j+s+D/2)(-i-l-n-D/2|l)(-j-m-s-D/2|m)\nonumber\\
&&\times (\sigma'+D/2|-2\sigma'-D/2)(i+n+D|l)(j+s+D|m),
\end{eqnarray}
where $\sigma'=\sigma+D/2$,
\begin{eqnarray}
{\cal S}_{NDIM}^6 &=& \pi^{3D/2}(p^2)^{\sigma'} (-i|i+j+D/2)(-j|i+j+D/2)
(-l|l+s+D/2)(-s|l+s+D/2) \nonumber\\
&& \times (\sigma'+D/2|-2\sigma'-D/2)(-i-j-l-m-s-D|l+m+s+D/2)
\nonumber\\
&&\times (i+j+D|l+m+s+D/2)(-n|-l-s-D/2+n)\nonumber\\
&&\times (l+s+D|-l-s-D/2+n).
\end{eqnarray}
Note that this result contains a special case (for $m=0$) calculated
numerically in \cite{easther}. Our result, on the other hand, was analytically
obtained and is more general, since the exponents of propagators are arbitrary
(cf. results in the second reference in \cite{lab}).
Finally,
\begin{eqnarray}
{\cal S}_{NDIM}^7 &=& \pi^{3D/2} (p^2)^{\sigma'}
(-i|i+m+D/2)(-m|i+m+D/2)(-l|l+s+D/2)\\
&&\times
(\sigma'+D/2|-2\sigma'-D/2)(-i-j-m-D/2|j)(l+s+D|-2l-2s-3D/2)
\nonumber\\
&&\times(-l-n-s-D/2|\sigma')(i+m+D|-i+l-m+n+s)(-s|l+s+D/2)\nonumber .
\end{eqnarray}
\section{Conclusion.}
Although NDIM{} is not a regularization method, it shares all the concepts of
dimensional regularization since it is based on the principle of analytic
continuation in the dimension of space-time. In this work we showed that with
an adequate technique such as NDIM{} it is possible to calculate some two- and
three-loop Feynman integrals with relative easiness and without much elaborate
machinery, just the performance of some Gaussian-type integrals plus resolution
of systems of linear algebraic equations. Of course, in view of the great
number of systems of linear algebraic equations that are involved, we need the
help of computer facilities, but the task is simple, which a PC can handle
properly.
The results we obtained can be used to study on-shell form factors in massless
QCD at the two-loop level. Moreover, to illustrate how NDIM{} works beyond
two-loops, we also performed some calculations of scalar three-loop, two-point
integrals with arbitrary exponents for propagators and dimension, in Euclidean
space.
\acknowledgments{A.G.M.S. gratefully acknowledges FAPESP (Funda\c c\~ao de Amparo \`a Pesquisa do
Estado de S\~ao Paulo, Brasil) for financial support.}
\vspace{1cm}
|
\section*{Acknowledgements}
|
\section{\label{intro}Introduction}
Context sensitive rewrite rules have been widely used in several areas
of natural language processing. Johnson \cite{john:form72} has
shown that such rewrite rules are equivalent to finite state
transducers in the special case
that they are not allowed to rewrite their own output. An algorithm
for compilation into transducers was provided by
\cite{kapl:regu94}. Improvements and extensions to this algorithm
have been provided by \cite{kart:95}, \cite{kart:repl97},
\cite{karttunen:96} and \cite{mohri-sproat:96}. In this paper,
the algorithm will be extended to provide a limited form of
backreferencing. Backreferencing has been implicit in previous
research, such as in the ``batch rules'' of \cite{kapl:regu94},
bracketing transducers for finite-state parsing \cite{karttunen:96},
and the ``LocalExtension'' operation of
\cite{roche_schabes95}. The explicit use of
backreferencing leads to more elegant and general solutions.
Backreferencing is widely used in editors, scripting languages and
other tools employing regular expressions \cite{frie:mast97}. For
example, Emacs uses the special brackets \verb+\(+ and \verb+\)+ to
capture strings along with the notation \verb+\+$n$ to recall the
$n$th such string. The expression {\bf \verb+\+(a*\verb+\+)b\verb+\+1}
matches strings of the form $a^nba^n$. Unrestricted use of
backreferencing thus can introduce non-regular languages. For NLP
finite state calculi \cite{kart:regu96,noord:fsa97} this is
unacceptable. The form of backreferences introduced in this paper
will therefore be restricted.
The central case of an allowable backreference is:
\begin{equation} \label{base}
x\Rightarrow T(x)/\lambda\mbox{\_\_}\rho
\end{equation}
\noindent This says that each string $x$ preceded by $\lambda$ and
followed by $\rho$ is replaced by $T(x)$, where $\lambda$ and $\rho$
are arbitrary regular expressions, and $T$ is a
transducer.\footnote{The syntax at this point is merely
suggestive. As an example, suppose that $T_{acr}$ transduces
phrases into acronyms. Then
\[x\Rightarrow T_{acr}(x)/\langle
\mbox{abbr}\rangle\mbox{\_\_}\langle /\mbox{abbr}\rangle\]
would transduce {\tt <abbr>non-deterministic finite automaton</abbr>}
into {\tt <abbr>NDFA</abbr>}.
To compare this with a backreference in Perl, suppose that
$T_{acr}$ is a subroutine that converts phrases into acronyms and
that $R_{acr}$ is a regular expression matching phrases that can
be converted into acronyms. Then (ignoring the left context) one can
write something like:
s/($R_{acr}$)(?=$\langle$/ABBR$\rangle$)/$T_{acr}$(\$1)/ge;.
The backreference variable, \$1, will be set to whatever string
$R_{acr}$ matches.} This contrasts sharply with the rewriting
rules that follow the tradition of Kaplan \& Kay:
\begin{equation} \label{noback}
\phi\Rightarrow \psi/\lambda\mbox{\_\_}\rho
\end{equation}
\noindent In this case, any string from the language $\phi$ is replaced
by any string independently chosen from the language $\psi$.
We also allow multiple (non-permuting)
backreferences of the form:
\begin{equation}\label{multi}
x_1 x_2\ldots x_n \Rightarrow T_1(x_1) T_2(x_2) \ldots
T_n(x_n)/\lambda\mbox{\_\_}\rho
\end{equation}
\noindent Since transducers are closed under concatenation,
handling multiple backreferences reduces to the problem
of handling a single backreference:
\begin{equation}
x\Rightarrow (T_1 \cdot T_2 \cdot \ldots \cdot T_n)(x)/\lambda\mbox{\_\_}\rho
\end{equation}
\noindent A problem arises if we want capturing to follow the
POSIX standard requiring a longest-capture strategy. Friedl
\cite{frie:mast97} (p.\/ 117), for example, discusses matching the
regular expression (to$\mid$top)(o$\mid$polo)?(gical$\mid$o?logical)
against the word: {\bf topological}. The desired result is that (once
an overall match is established) the first set of parentheses should
capture the longest string possible ({\bf top}); the second set should
then match the longest string possible from what's left ({\bf o}), and
so on. Such a left-most longest match concatenation operation is
described in \sref{topo}.
In the following section, we initially concentrate on the simple case
in (\ref{base}) and show how (\ref{base}) may be compiled assuming
left-to-right processing along with the overall longest match strategy
described by \cite{karttunen:96}.
The major components of the algorithm are not new, but straightforward
modifications of components presented in \cite{karttunen:96} and
\cite{mohri-sproat:96}. We improve upon existing approaches
because we solve a problem concerning the use of special marker
symbols (\sref{alpha}). A further contribution is that all steps are
implemented in a freely available system, the FSA Utilities of
\cite{noord:fsa97} (\sref{fsu}).
\section{The Algorithm \label{algorithm}}
\subsection{Preliminary Considerations}
Before presenting the algorithm proper, we will deal with a couple of
meta issues. First, we introduce our version of the finite state
calculus in \sref{fsu}. The treatment of special marker symbols
is discussed in \sref{alpha}. Then in \sref{util}, we
discuss various utilities that will be essential for the algorithm.
\subsubsection{\label{fsu}FSA Utilities}
\begin{table}
\begin{tabular}{cl}
\tt [] & empty string \\
\tt [E1,\dots En] & concatenation of \tt E1 \dots En \\
\tt \verb+{}+ & empty language \\
\tt \verb+{+E1,\dots En\verb+}+ & union of \tt E1,\dots En\\
\tt E* & Kleene closure\\
\tt E\verb+^+ & optionality\\
\tt \~{}E & complement\\
\tt E1-E2 & difference\\
\tt \verb+$+ E & containment\\
\tt E1 \verb+&+ E2 & intersection\\
\tt ? & any symbol\\
\tt A:B & pair\\
\tt E1 x E2 & cross-product\\
\tt A o B & composition\\
\tt domain(E) & domain of a transduction\\
\tt range(E) & range of a transduction\\
\tt identity(E) & identity transduction\\
\tt inverse(E) & inverse transduction\\
\end{tabular}
\centering
\caption{\label{notation}Regular expression operators.}
\end{table}
The algorithm is implemented in the FSA Utilities \cite{noord:fsa97}.
We use the notation provided by the toolbox throughout this paper.
Table~\ref{notation} lists the relevant regular expression operators.
FSA Utilities offers the possibility to define new regular expression
operators. For example, consider the definition of the nullary
operator {\tt vowel} as the union of the five vowels:
\begin{verbatim}
macro(vowel,{a,e,i,o,u}).
\end{verbatim}
In such macro definitions, Prolog variables can be used in order to
define new n-ary regular expression operators in terms of existing
operators. For instance, the {\em lenient\_composition} operator
\cite{karttunen:98} is defined by:
\begin{verbatim}
macro(priority_union(Q,R),
{Q, ~domain(Q) o R}).
macro(lenient_composition(R,C),
priority_union(R o C,R)).
\end{verbatim}
Here, \code{priority\_union} of two regular
expressions \code{Q} and \code{R} is defined as the union of \code{Q}
and the composition of the complement of the domain of \code{Q} with
\code{R}. Lenient composition of \code{R} and \code{C}
is defined as the priority union of the composition of \code{R} and
\code{C} (on the one hand) and \code{R} (on the other hand).
Some operators, however, require something more than simple macro
expansion for their definition. For example, suppose a user wanted
to match $n$ occurrences of some pattern. The FSA Utilities already
has the '*' and '+' quantifiers, but any other operators like this
need to be user defined. For this purpose, the FSA Utilities
supplies simple Prolog hooks allowing this general quantifier to be
defined as:
\begin{verbatim}
macro(match_n(N,X),Regex) :-
match_n(N,X,Regex).
match_n(0,_X,[]).
match_n(N,X,[X|Rest]) :-
N > 0,
N1 is N-1,
match_n(N1,X,Rest).
\end{verbatim}
\noindent For example: \verb+match_n(3,a)+ is equivalent to the
ordinary finite state calculus expression \verb+[a,a,a]+.
Finally, regular expression operators can be defined in terms of
operations on the underlying automaton. In such cases, Prolog hooks
for manipulating states and transitions may be used.
This functionality has been used in \cite{noord-gerd99} to provide
an implementation of the algorithm in \cite{mohri-sproat:96}.
\subsubsection{\label{alpha}Treatment of Markers}
Previous algorithms for compiling rewrite rules into transducers have
followed \cite{kapl:regu94} by
introducing special marker symbols ({\em markers})
into strings in order to mark off
candidate regions for replacement. The assumption is that these
markers are outside the resulting transducer's alphabets. But
previous algorithms have not ensured that the assumption holds.
This problem was recognized by \cite{karttunen:96}, whose algorithm
starts with a filter transducer which filters out any string
containing a marker. This is problematic for two reasons. First, when
applied to a string that does happen to contain a marker, the
algorithm will simply fail. Second, it leads to logical problems in
the interpretation of complementation. Since the complement of a
regular expression R is defined as $\Sigma-R$, one needs to know
whether the marker symbols are in $\Sigma$ or not. This has not been
clearly addressed in previous literature.
We have taken a different approach by providing a contextual way of
distinguishing markers from non-markers. Every symbol used in the
algorithm is replaced by a pair of symbols, where the second member of
the pair is either a 0 or a 1 depending on whether the first member is
a marker or not.\footnote{This approach is
similar to the idea of laying down tracks as in the compilation of
monadic second-order logic into automata
\nocite{klar:mona97}Klarlund (1997, p. 5). In
fact, this technique could possibly be used for a more efficient
implementation of our algorithm: instead of adding transitions over
0 and 1, one could represent the alphabet as bit sequences and then
add a final 0 bit for any ordinary symbol and a final 1 bit for a
marker symbol. }
As the first step in the algorithm, 0's are inserted
after every symbol in the input string to indicate that initially
every symbol is a non-marker. This is defined as:
\begin{verbatim}
macro(non_markers,[?,[]:0]*).
\end{verbatim}
Similarly, the following macro can be used to insert a 0 after every
symbol in an arbitrary expression E.
\begin{verbatim}
macro(non_markers(E),
range(E o non_markers)).
\end{verbatim}
\noindent
Since \code{E} is a recognizer, it is first coerced to
\code{identity(E)}. This form of implicit conversion is standard in
the finite state calculus.
Note that 0 and 1 are perfectly ordinary alphabet
symbols, which may also be used within a replacement. For example, the
sequence [1,0] represents a non-marker use of the symbol 1.
\subsubsection{\label{util}Utilities}
Before describing the algorithm, it will be helpful to have at our
disposal a few general tools, most of which were described already in
\cite{kapl:regu94}. These tools, however, have been modified so
that they work with our approach of distinguishing markers from
ordinary symbols. So to begin with, we provide macros to describe the
alphabet and the alphabet extended with marker symbols:
\begin{verbatim}
macro(sig,[?,0]).
macro(xsig,[?,{0,1}]).
\end{verbatim}
The macro \code{xsig} is useful for defining a specialized version of
complementation and containment:
\begin{verbatim}
macro(not(X),xsig* - X).
macro($$(X),[xsig*,X,xsig*]).
\end{verbatim}
The algorithm uses four kinds of brackets, so it will be convenient to
define macros for each of these brackets, and for a few disjunctions.
\begin{verbatim}
macro(lb1,['<1',1]).
macro(lb2,['<2',1]).
macro(rb2,['2>',1]).
macro(rb1,['1>',1]).
macro(lb,{lb1,lb2}).
macro(rb,{rb1,rb2}).
macro(b1,{lb1,rb1}).
macro(b2,{lb2,rb2}).
macro(brack,{lb,rb}).
\end{verbatim}
As in Kaplan \& Kay, we define an Intro(S) operator that
produces a transducer that freely introduces instances of
S into an input string. We extend this idea to create
a family of Intro operators. It is often the case that we want to
freely introduce marker symbols into a string at any position {\em
except} the beginning or the end.
\begin{verbatim}
macro(intro(S),{xsig-S,[] x S}*).
macro(xintro(S),{[],[xsig-S,intro(S)]}).
macro(introx(S),{[],[intro(S),xsig-S]}).
macro(xintrox(S),{[],[xsig-S],
[xsig-S,intro(S),xsig-S]}).
\end{verbatim}
This family of Intro operators is useful for defining a family
of Ignore operators:
\begin{verbatim}
macro( ign( E1,S),range(E1 o intro( S))).
macro(xign( E1,S),range(E1 o xintro( S))).
macro( ignx(E1,S),range(E1 o introx(S))).
macro(xignx(E1,S),range(E1 o xintrox(S))).
\end{verbatim}
In order to create filter transducers to ensure that markers are
placed in the correct positions, Kaplan \& Kay introduce the operator
\code{P-iff-S(L1,L2)}. A string is described by this expression iff
each prefix in \code{L1} is followed by a suffix in \code{L2} and each
suffix in \code{L2} is preceded by a prefix in \code{L1}. In our
approach, this is defined as:
\begin{verbatim}
macro(if_p_then_s(L1,L2),
not([L1,not(L2)])).
macro(if_s_then_p(L1,L2),
not([not(L1),L2])).
macro(p_iff_s(L1,L2),
if_p_then_s(L1,L2)
&
if_s_then_p(L1,L2)).
\end{verbatim}
To make the use of \code{p\_iff\_s} more convenient, we introduce a new
operator \code{l\_iff\_r(L,R)}, which describes
strings where every string position is preceded by a string in
\code{L} just in case it is followed by a string in \code{R}:
\begin{verbatim}
macro(l_iff_r(L,R),
p_iff_s([xsig*,L],[R,xsig*])).
\end{verbatim}
Finally, we introduce a new operator \code{if(Condition,Then,Else)} for
conditionals. This operator is extremely useful, but in order for it
to work within the finite state calculus, one needs a convention as to
what counts as a boolean true or false for the condition
argument. It is possible to define {\tt true} as the universal
language and {\tt false} as the empty language:
\begin{verbatim}
macro(true,? *). macro(false,{}).
\end{verbatim}
With these definitions, we can use the complement operator as
negation, the intersection operator as conjunction and the union
operator as disjunction. Arbitrary expressions may be coerced to
booleans using the following macro:
\begin{verbatim}
macro(coerce_to_boolean(E),
range(E o (true x true))).
\end{verbatim}
\noindent Here, \code{E} should describe a recognizer.
\code{E} is composed with the universal transducer, which transduces from
anything (\code{?*}) to anything (\code{?*}).
Now with this background, we can define
the conditional:
\begin{verbatim}
macro(if(Cond,Then,Else),
{ coerce_to_boolean(Cond) o Then,
~coerce_to_boolean(Cond) o Else
}).
\end{verbatim}
\subsection{Implementation}
A rule of the form $x \rightarrow T(x)/\lambda\mbox{\_\_}\rho$
will be written as \code{replace(T,Lambda,Rho)}. Rules of the more general
form $x_1 \ldots x_n \Rightarrow T_1(x_1) \ldots
T_n(x_n)/\lambda\mbox{\_\_}\rho$ will be discussed in \sref{topo}. The
algorithm consists of nine steps composed as in figure~\ref{replace}.
\begin{figure*}
\begin{verbatim}
macro(replace(T,Left,Right),
non_markers
o
r(Right)
o
f(domain(T))
o
left_to_right(domain(T))
o
longest_match(domain(T))
o
aux_replace(T)
o
l1(Left)
o
l2(Left)
o
inverse(non_markers)).
\end{verbatim}
\caption{\label{replace}Definition of \code{replace} operator.}
\end{figure*}
The names of these steps are mostly derived from \cite{kart:95} and
\cite{mohri-sproat:96} even though the transductions involved are
not exactly the same. In particular, the steps derived from Mohri \&
Sproat (\code{r}, \code{f}, \code{l1} and \code{l2}) will all be
defined in terms of the finite state calculus as opposed to Mohri \&
Sproat's approach of using low-level manipulation of states and
transitions.\footnote{The alternative implementation is provided in \cite{noord-gerd99}.}
The first step, \code{non\_markers}, was already defined above.
For the second step, we first consider a simple special case. If the
empty string is in the language described by \code{Right},
then \code{r(Right)} should insert an
\code{rb2} in every string position. The definition of
\code{r(Right)} is both simpler and more efficient if this is treated
as a special case. To insert a bracket in every possible string
position, we use:
\begin{verbatim}
[[[] x rb2,sig]*,[] x rb2]
\end{verbatim}
If the empty string is not in \code{Right}, then we must use
\code{intro(rb2)} to introduce the marker \code{rb2}, followed
by \code{l\_iff\_r} to ensure that such markers are immediately
followed by a string in \code{Right}, or more precisely a string
in \code{Right} where additional instances of \code{rb2} are freely
inserted in any position other than the beginning. This expression is
written as:
\begin{verbatim}
intro(rb2)
o
l_iff_r(rb2,xign(non_markers(Right),rb2))
\end{verbatim}
Putting these two pieces together with the conditional yields:
\begin{verbatim}
macro(r(R),
if([] & R,
[[[] x rb2,sig]*,[] x rb2],
intro(rb2)
o
l_iff_r(rb2,xign(non_markers(R),rb2)))).
\end{verbatim}
The third step, \code{f(domain(T))} is implemented as:
\begin{verbatim}
macro(f(Phi), intro(lb2)
o
l_iff_r(lb2,[xignx(non_markers(Phi),b2),
lb2^,rb2])).
\end{verbatim}
The \code{lb2} is first introduced and then, using
\code{l\_iff\_r}, it is constrained to occur immediately before every
instance of (ignoring complexities) \code{Phi} followed by an
\code{rb2}. \code{Phi} needs to be marked as normal text
using \code{non\_markers} and then \code{xign\_x} is used to allow
freely inserted \code{lb2} and \code{rb2} anywhere except at
the beginning and end. The following \code{lb2}\verb+^+ allows an
optional \code{lb2}, which occurs when the empty string is in
\code{Phi}.
The fourth step is a guessing component which (ignoring complexities)
looks for sequences of the form \code{lb2 Phi rb2} and
converts some of these into \code{lb1 Phi rb1}, where the
\code{b1} marking indicates that the sequence is a candidate for
replacement. The complication is that \code{Phi}, as always, must be
converted to \code{non\_markers(Phi)} and instances of \code{b2}
need to be ignored. Furthermore, between pairs of \code{lb1}
and \code{rb1}, instances of \code{lb2} are deleted. These
\code{lb2} markers have done their job and are no longer needed.
Putting this all together, the definition is:
\begin{verbatim}
macro(left_to_right(Phi),
[[xsig*,
[lb2 x lb1,
(ign(non_markers(Phi),b2)
o
inverse(intro(lb2))
),
rb2 x rb1]
]*, xsig*]).
\end{verbatim}
The fifth step filters out non-longest matches produced in the
previous step. For example (and simplifying a bit), if \code{Phi} is
\code{ab*}, then a string of the form $\ldots$ rb1 a b lb1 b
$\ldots$ should be ruled out since there is an instance of \code{Phi}
(ignoring brackets except at the end) where there is an internal
\code{lb1}. This is implemented as:\footnote{The line with
\code{\$\$(rb1)} can be optimized a bit: Since we know that an
\code{rb1} must be preceded by \code{Phi}, we can write:
\code{[ign\_(non\_markers(Phi),brack),rb1,xsig*])}. This
may lead to a more constrained (hence smaller) transducer. }
\begin{verbatim}
macro(longest_match(Phi),
not($$([lb1,
(ignx(non_markers(Phi),brack)
&
$$(rb1)
),
rb
]))
o
inverse(intro(rb2))).
\end{verbatim}
The sixth step performs the transduction described by \code{T}. This
step is straightforwardly implemented, where the main difficulty is
getting \code{T} to apply to our specially marked string:
\begin{verbatim}
macro(aux_replace(T),
{{sig,lb2},
[lb1,
inverse(non_markers)
o T o
non_markers,
rb1 x []
]
}*).
\end{verbatim}
The seventh step ensures that \code{lb1} is preceded by a string
in \code{Left}:
\begin{verbatim}
macro(l1(L),
ign(if_s_then_p(
ignx([xsig*,non_markers(L)],lb1),
[lb1,xsig*]),
lb2)
o
inverse(intro(lb1))).
\end{verbatim}
The eighth step ensures that \code{lb2} is not preceded by a
string in \code{Left}. This is implemented similarly to the
previous step:
\begin{verbatim}
macro(l2(L),
if_s_then_p(
ignx(not([xsig*,non_markers(L)]),lb2),
[lb2,xsig*])
o
inverse(intro(lb2))).
\end{verbatim}
Finally the ninth step, \code{inverse(non\_markers)}, removes the
\code{0}'s so that the final result in not marked up in any special
way.
\section{\label{topo}Longest Match Capturing}
As discussed in \sref{intro} the POSIX standard requires that multiple
captures follow a longest match strategy. For multiple captures as in
(\ref{multi}), one establishes first a longest match for
$domain(T_1)\cdot \ldots\cdot domain(T_n)$. Then we
ensure that each of $domain(T_i)$ in turn is
required to match as long as possible, with each one having priority
over its rightward neighbors. To implement this, we define a macro
\code{lm\_concat(Ts)} and use it as:
\begin{verbatim}
replace(lm_concat(Ts),Left,Right)
\end{verbatim}
\begin{figure*}
\small
\begin{verbatim}
macro(lm_concat(Ts),mark_boundaries(Domains) o ConcatTs):-
domains(Ts,Domains), concatT(Ts,ConcatTs).
domains([],[]).
domains([F|R0],[domain(F)|R]):- domains(R0,R).
concatT([],[]).
concatT([T|Ts], [inverse(non_markers) o T,lb1 x []|Rest]):- concatT(Ts,Rest).
macro(mark_boundaries(L),Exp):-
boundaries(L,Exp0),
greed(L,Exp0,Exp).
boundaries([],[]).
boundaries([F|R0],[F o non_markers, [] x lb1 |R]):- boundaries(R0,R).
greed(L,Composed0,Composed) :-
aux_greed(L,[],Filters), compose_list(Filters,Composed0,Composed).
aux_greed([H|T],Front,Filters):- aux_greed(T,H,Front,Filters,_CurrentFilter).
aux_greed([],F,_,[],[ign(non_markers(F),lb1)]).
aux_greed([H|R0],F,Front,[~L1|R],[ign(non_markers(F),lb1)|R1]) :-
append(Front,[ignx_1(non_markers(F),lb1)|R1],L1),
append(Front,[non_markers(F),lb1],NewFront),
aux_greed(R0,H,NewFront,R,R1).
macro(ignx_1(E1,E2), range(E1 o [[? *,[] x E2]+,? +])).
compose_list([],SoFar,SoFar).
compose_list([F|R],SoFar,Composed):- compose_list(R,(SoFar o F),Composed).
\end{verbatim}
\caption{\label{lmconcat}Definition of \code{lm\_concat} operator.}
\end{figure*}
Ensuring the longest overall match is delegated to the \code{replace}
macro, so \code{lm\_concat(Ts)} needs only ensure that each individual
transducer within \code{Ts} gets its proper left-to-right longest
matching priority. This problem is mostly solved by the same
techniques used to ensure the longest match within the \code{replace}
macro. The only complication here is that \code{Ts} can be of
unbounded length. So it is not possible to have a single expression in
the finite state calculus that applies to all possible lenghts. This
means that we need something a little more powerful than mere macro
expansion to construct the proper finite state calculus expression.
The FSA Utilities provides a Prolog hook for this purpose. The
resulting definition of \code{lm\_concat} is given in
figure~\ref{lmconcat}.
Suppose (as in \cite{frie:mast97}), we want to match the following
list of recognizers against the string \code{topological} and insert a
marker in each boundary position. This reduces to applying:
\begin{verbatim}
lm_concat([
[{[t,o],[t,o,p]},[]: '#'],
[{o,[p,o,l,o]},[]: '#'],
{[g,i,c,a,l],[o^,l,o,g,i,c,a,l]}
])
\end{verbatim}
This expression transduces the string {\tt topological} only to the
string {\tt top\#o\#logical}.\footnote{An anonymous reviewer suggested that
\code{lm\_concat} could be implemented in the framework of
\cite{karttunen:96} as:
\[ \mbox{[t o $\mid$t o p $\mid$ o $\mid$ p o l o ] $\rightarrow$ ... \#}; \]
Indeed the resulting transducer from this
expression would transduce {\tt topological} into {\tt top\#o\#logical}.
But unfortunately this transducer would also transduce
{\tt polotopogical} into {\tt polo\#top\#o\#gical}, since the notion of
left-right ordering is lost in this expression.}
\section{\label{conclude} Conclusions}
The algorithm presented here has extended previous algorithms for
rewrite rules by adding a limited version of backreferencing. This
allows the output of rewriting to be dependent on the form of the
strings which are rewritten. This new feature brings techniques
used in Perl-like languages into the finite state calculus. Such an
integration is needed in practical applications where simple
text processing needs to be combined with more
sophisticated computational linguistics techniques.
One particularly interesting example where backreferences are
essential is cascaded deterministic (longest match) finite state
parsing as described for example in Abney \cite{abne:part96} and
various papers in \cite{roche_schabes97}. Clearly, the standard
rewrite rules do not apply in this domain. If {\it NP} is an NP recognizer,
it would not do to say {\it NP} $\Rightarrow [${\it NP}$]/\lambda \mbox{\_\_}\rho$. Nothing would force the string matched by the
{\it NP} to the left of the arrow to be the same as the string
matched by the {\it NP} to the right of the arrow.
One advantage of using our algorithm for finite state parsing is that
the left and right contexts may be used to bring in top-down
filtering.\footnote{The bracketing operator of \cite{karttunen:96},
on the other hand, does not provide for left and right contexts.} An
often cited advantage of finite state parsing is robustness. A
constituent is found bottom up in an early level in the cascade even
if that constituent does not ultimately contribute to an S in a later
level of the cascade. While this is undoubtedly an advantage for
certain applications, our approach would allow the introduction of
some top-down filtering while maintaining the robustness of a
bottom-up approach. A second advantage for robust finite state parsing
is that bracketing could also include the notion of ``repair'' as in
\cite{abney:90}. One might, for example, want to say something
like: $\mbox{\it xy} \Rightarrow [_{NP}\, \mbox{\it RepairDet(x)
RepairN(y)}\,]/\lambda \mbox{\_\_}\rho$\footnote{The syntax here has
been simplified. The rule should be understood as:
replace(lm\_concat([[]:'[np', repair\_det, repair\_n,
[]:']'],lambda, rho).} so that an {\it NP} could be parsed as a
slightly malformed {\it Det} followed by a slightly malformed {\it
N}. RepairDet and RepairN, in this example, could be doing a variety
of things such as: contextualized spelling correction, reordering of
function words, replacement of phrases by acronyms, or any other
operation implemented as a transducer.
Finally, we should mention the problem of complexity. A critical
reader might see the nine steps in our algorithm and conclude that the
algorithm is overly complex. This would be a false conclusion. To
begin with, the problem itself is complex. It is easy to create
examples where the resulting transducer created by any algorithm would
become unmanageably large. But there exist strategies for keeping the
transducers smaller. For example, it is not necessary for all nine
steps to be composed. They can also be cascaded. In that case it will
be possible to implement different steps by different
strategies, e.g.\/ by deterministic or non-deterministic transducers or
bimachines \cite{roch:lang97}. The range of possibilities
leaves plenty of room for future research.
\bibliographystyle{plain}
|
\section{Introduction}
The quark model predicts that the
charge conjugation (C) and parity (P) of a meson
with spin $S$ and orbital angular momentum $L$ is
\begin{equation}
P = (-1)^{L+1} \; \; \;
C = (-1)^{L+S}
\label{eq:JCP}
\end{equation}
States with quantum numbers not produced
by eq.~\ref{eq:JCP},
such as
\begin{equation}
J_{exotic}^{PC} = 1^{-+}, 0^{+-}, 2^{+-}, 0^{--}
\end{equation}
are known as exotics~\cite{Burnett:1990aw}.
Exotic states are allowed by QCD.
There are a number of different possibilities for the structure of an
exotic state. An exotic signal could be: a hybrid meson, which is
quark and anti-quark and excited glue, or bound state of two quarks and
two anti-quarks ($\overline{Q}\overline{Q}QQ$). The two most popular
guesses for the structure of the ($\overline{Q}\overline{Q}QQ$) state
are either a molecule of two mesons or diquark anti-diquark bound
state. In this paper I review the latest lattice results for the
masses of exotic hybrid mesons, concentrating on the $1^{-+}$ state,
obtained from lattice QCD.
\section{Lattice simulations of exotic mesons}
Many numerical predictions of QCD can be
determined from the path integral
\begin{equation}
c(t) \sim \int dU \int d\psi \int d\overline{\psi}
\;
\sum_{\underline{x}} O(\underline{0},0)
O(\underline{x},t)^{\dagger}
e^{-S_F - S_G }
\label{eq:reallyQCD}
\end{equation}
where $S_F$ is the fermion action (some lattice
discretization of the Dirac action) and $S_G$
is the pure gauge action.
The path integral in eq.~\ref{eq:reallyQCD} is put on
the computer using a clever finite
difference formalism~\cite{Montvay:1994cy}, due
to Wilson, that maintains gauge invariance.
The physical picture for eq.~\ref{eq:reallyQCD} is that a
hadron is created at time 0, from where it propagates to the time t,
where it is destroyed.
The fermion integration can be done exactly in
eq.~\ref{eq:reallyQCD} to produce the fermion determinant.
Simulations that include the effect of the determinant are very
expensive computationally, so typically it is
not included in the simulation (the quenched approximation).
However there has been some recent work that includes the
effect of the determinant~\cite{Lacock:1998A} on the light
exotic spectrum.
The standard interpolating operator for the pion, which can be used
in eq.~\ref{eq:reallyQCD}, is
\begin{equation}
O_{\pi} (\underline{x} , t ) = \overline{\psi}(\underline{x},t)
\gamma_5
\psi (\underline{x},t)
\end{equation}
which has the correct $J^{PC} = 0^{-+}$ quantum numbers.
One possible interpolating operator~\cite{Bernard:1997ib}
for an exotic $1^{-+}$ particle is
\begin{equation}
O_{1^{-+}} (\underline{x} , t ) = \overline{\psi}(\underline{x},t)
\gamma_j F_{ij} (\underline{x},t)
\psi (\underline{x},t)
\label{eq:interponemp}
\end{equation}
where $F$ is the QCD field strength tensor.
It is essential to use operators that have
some spatial separation between the quarks in the meson
to get a good signal. Recently the MILC collaboration has
attempted to measure the ``wave function'' of the $1^{-+}$ hybrid meson
in coulomb gauge~\cite{McNeile:1998cp}. Unfortunately the operator
used did not have the correct charge conjugation quantum number, so
the published wave function~\cite{McNeile:1998cp} is incorrect.
In this formalism a gauge invariant interpolating operator, for any
possible exotic hybrid particle or four particle state can be
constructed. The dynamics then determines whether the resulting state
has a narrow decay width, which can be detected experimentally. In the
large $N_c$ (number of colours)
limit~\cite{Burnett:1990aw,Cohen:1998jb} both exotic hybrid mesons and
non-exotic mesons have widths vanishingly small compared to their
masses.
The data from the simulation is extracted using a
fit model~\cite{Montvay:1994cy}:
\begin{equation}
c(t) = a_0 exp( -m_0 t ) + a_1 exp( -m_1 t ) + \cdots
\label{eq:fitmodel}
\end{equation}
where $m_0$ ($m_1$) is the ground (first excited) state
mass and the dots represent higher excitations.
Although in principle excited state masses can be extracted from a
multiple exponential fit, in practice this is numerically non-trivial
because of the noise in the data from the simulation.
Simulations that involve the calculation of the properties
of exotic hybrid mesons are harder than those that concentrate
on the non-exotic hadrons, because the signal to noise ratio is
worse for exotic mesons than for $\overline{Q}Q$ mesons.
In an individual lattice simulation there are errors from the finite
size of the lattice spacing and the finite lattice volume. State of the art
lattice simulations in the quenched theory, run at a number of
different lattice spacings and physical volumes and extrapolate the
results to the continuum and infinite volume. For the exotic mesons,
this is done for heavy quarks (see section~\ref{se:heavyRES}), but as
yet, the continuum extrapolation has not been done for light exotic mesons.
How successful is lattice QCD in practice? One way to tell is to
compare the predictions for the masses from lattice QCD for well known
particles (proton, $\rho$, etc.) with experiment. The most accurate
quenched calculation to date has recently been completed by the
CP-PACS collaboration~\cite{Kanaya:1998sd}. From the masses of 11
light hadrons, they conclude that the quenched approximation disagrees
with experiment by at most 11\%. The comparison of results from
simulations that include dynamical fermions with experiment is less
clear, because of their high computational cost (see
Kenway~\cite{Kenway:1998ew} for a review of the latest results).
The results from lattice QCD also provide insight into the underlying
dynamics of light hadrons. Lattice QCD simulations can test the
various assumptions made in models of the QCD dynamics. For example
there are a number of models of exotic states based on the idea of a
bound diquark anti-diquark
pair~\cite{Uehara:1996hc}. A
critical assumption in diquark models is that two quarks actually do
cluster to form a diquark. This assumption has recently been tested in a
lattice gauge theory simulation by the Bielefeld
group~\cite{Hess:1998sd}, where they found no deeply bound diquark
state in Landau gauge.
\section{Results for light exotic mesons}
In the last year the MILC collaboration have
repeated~\cite{McNeile:1998cp} their initial
simulations~\cite{Bernard:1997ib} using the clover fermion action.
The clover action is ``closer'' to the continuum than
the Wilson fermion action, because it has the leading order lattice
spacing terms removed. There are also new results for the hybrid
masses from the SESAM collaboration (reported by Lacock and Schilling)
~\cite{Lacock:1998A}, that include some effects from dynamical sea
fermions.
The results for the mass of the $1^{-+}$ exotic state are summarised in
table~\ref{onempRES}. All the results from the various simulations
are essentially consistent.
with the mass of the $1^{-+}$ state around $2$ GeV.
In table~\ref{describeLIGHT} we show the
physical parameters for each of the simulations. The interpolating
operators used to create the exotic meson states in the MILC
calculations~\cite{Bernard:1997ib} are different to those used in the
UKQCD~\cite{Lacock:1997ny} and SESAM simulations~\cite{Lacock:1998A}.
The observation that the results for the mass of the $1^{-+}$
hybrid meson are consistent for very
different simulations gives us confidence in the final result.
Although I would prefer to see simulation results at lighter
quark masses. Simulations at the point where the ratio of the
pseudoscalar mass to vector mass ($M_{PS}/M_V \sim 0.5 $) are
possible with current algorithms and computers~\cite{Kanaya:1998sd}.
\begin{table}
\centering
\caption{ \it Masses of the light $1^{-+}$ hybrid from
lattice gauge theory.
}
\vskip 0.1 in
\begin{tabular}{|c|c|c|} \hline
Simulation & Group & mass GeV \\
\hline
\hline
A & UKQCD~\cite{Lacock:1997ny} &
$ 2.0(2)$ \\
B & MILC~\cite{Bernard:1997ib,McNeile:1998cp} & $ 2.0(1) \pm sys $ \\
C & MILC~\cite{Bernard:1997ib,McNeile:1998cp} & $ 2.1(1) \pm sys $ \\
D & Lacock and Schilling~\cite{Lacock:1998A} & $ 1.9(2) $ \\
\hline
\end{tabular}
\label{onempRES}
\end{table}
\begin{table}
\centering
\caption{\it Parameters of light exotic meson simulations,
}
\vskip 0.1 in
\begin{tabular}{|c|c|c|c|c|c|} \hline
Simulation & Action & fermions & length
(fm) & $a^{-1}$ GeV & $M_{PS} / M_V$ \\
\hline
\hline
A & clover & quenched & $1.6$ & $2.0$ &
$0.76$ \\
B & Wilson & quenched & $2.3$ &
$2.8$ & $0.96,0.93,0.88,0.77$ \\
C & clover & quenched & $2.3$ &
$2.8$ & $0.94,0.90,0.72$ \\
D & Wilson & dynamical & $1.4$ &
$2.3$ & $0.83,0.81,0.76,0.69$ \\
\hline
\end{tabular}
\label{describeLIGHT}
\end{table}
\section{Results for heavy exotic mesons} \label{se:heavyRES}
There has been a lot of work on calculating the spectroscopy of $\overline{c}c$
and $\overline{b}b$ mesons from lattice gauge theory
(see Davies~\cite{Davies:1997hv} for a review).
The main technical complication in heavy quark simulations is that the lattice
spacing of current simulations is not smaller than the heavy quark
mass. So various effective field theory Lagrangian approximations
to QCD are simulated.
The NRQCD (nonrelativistic QCD) Lagrangian is one such effective
field theory approximation to QCD, with the
expansion parameter equal to the velocity squared.
NRQCD has been particularly successful
in simulating the Upsilon spectrum~\cite{Davies:1997hv},
but is less well converged for charmomium, (particularly for
spin splittings), because the charm quarks move with higher
velocity~\cite{Davies:1997hv}.
The NRQCD Langrangian
correct up to $O(Mv^2)$ is
\begin{equation}
{\cal L}^{NRQCD} = \overline{\psi}( -\frac{\bigtriangleup^2}{2 M }
-\frac{c_0 \bigtriangleup^4}{8 M^2}
-\frac{c_1 \sigma B } { 2 M } ) \psi
\label{NRQCDL}
\end{equation}
where $c_0$ and $c_1$ are coefficients obtained by a
perturbative matching procedure to QCD.
In table~\ref{hybridRESHEAVY} the results of all the recent
NRQCD simulations of the $\overline{b}bg$ hybrids in the quenched
approximation are compiled. In the
hybrid meson simulations no spin terms are included in the Lagrangian
($c_1 = 0 $ in eq.~\ref{NRQCDL}), so the $1^{-+}$, $0^{+-}$, and $2^{+-}$
states are degenerate. Both the results from the CP-PACS
collaboration~\cite{Manke:1998qc} and from Juge, Kuti and
Morningstar~\cite{Juge:1999ie} were shown to
be independent of the lattice spacing and lattice volume.
For example, the CP-PACS
collaboration~\cite{Manke:1998qc} found that the masses of the hybrids
were independent of the box size above
$1.2 \; fm$.
The ``asymmetric'' comment in table~\ref{hybridRESHEAVY} refers to the
technique of treating space and time asymmetrically. A smaller lattice
spacing was used in the time direction than in the space direction,
which allowed the signal to be seen for further, for a given spatial
volume. This technique has helped to reduce the errors. The
practicalities of this idea were demonstrated by Morningstar and
Peardon~\cite{Morningstar:1997ff} for the glueball spectrum.
\begin{table}
\centering
\caption{ \it Mass splitting between the \protect{$\overline{b}b$}
$1^{-+}$ hybrid and the \protect{$\overline{b}b$} 1S state.
}
\vskip 0.1 in
\begin{tabular}{|c|c|c|} \hline
Group & comments & mass GeV \\
\hline
\hline
UKQCD~\cite{Manke:1998yg} & $O(Mv^4)$ errors & $1.68(10)$ \\
CP-PACS~\cite{Manke:1998qc} & Asymmetric, $O(Mv^4)$ errors & $1.542(8)$ \\
Juge et al.~\cite{Juge:1999ie} & Asymmetric, $O(Mv^4)$ errors & $ 1.49(2)(5) $ \\
\hline
\end{tabular}
\label{hybridRESHEAVY}
\end{table}
In table~\ref{hybridRESCHARM} the results of the mass splitting of the
$1^{-+}$ states and the $1S$ state are shown in the charmonium system.
The MILC collaboration used
the standard Wilson and clover actions
to simulate the charm quark in their simulations
of heavy exotic mesons, as previous work
has shown this to be reliable~\cite{Davies:1997hv}.
\begin{table}
\centering
\caption{ \it Mass splitting between the \protect{$\overline{c}c$}
$1^{-+}$ hybrid and the \protect{$\overline{c}c$} 1S state.
}
\vskip 0.1 in
\begin{tabular}{|c|c|c|} \hline
Group & comments & mass MeV \\
\hline
\hline
MILC~\cite{Bernard:1997ib,McNeile:1998cp} & Wilson action & $1340^{+60}_{-150}+sys$ \\
MILC~\cite{McNeile:1998cp} & clover action & $1220^{+110}_{-190}+sys$ \\
CP-PACS~\cite{Manke:1998qc} & NRQCD $O(Mv^4)$ errors, Asymmetric & $1323(13)$ \\
\hline
\end{tabular}
\label{hybridRESCHARM}
\end{table}
The first results for heavy exotic~\cite{Perantonis:1990dy} hybrids
were done in the adiabatic surfaces approach, where the effect of the
excited glue is subsumed in a potential measured on the lattice
(see~\cite{Kuti:1998rh,Michael:1998sm} for a review). Juge, Kuti and
Morningstar~\cite{Juge:1999ie} have completed a systematic study of
these potentials. The NRQCD approach is a more accurate approximation to
QCD than the adiabatic potential technique; however Juge, Kuti
and Morningstar~\cite{Juge:1999ie} found that the adiabatic potential
approach reproduced the level splittings from NRQCD up to 10 \%. A
preliminary result for the calculation of the adiabatic surfaces with
the effects of dynamical fermions included has been reported by
Bali~\cite{Kuti:1998rh} and collaborators.
No dramatic differences between the quenched theory
result were observed.
\section{Conclusions}
All the lattice simulations agree that the light $1^{-+}$ state has a
mass of $2.0(2) GeV$. The first simulation that included the effects of
dynamical fermions has not changed the result.
The experimental results for light exotics are reviewed by
S.U. Chung~\cite{Chung:1999}, so I just briefly compare the
lattice results to experiment.
There is an experimental signal for a $1^{-+}$ state at $1.4$ GeV with
a decay into $\eta\pi$~\cite{Chung:1999} from E852, Crystal barrel,
VES, and KEK. It is surprising that this state is only seen in the
$\eta\pi$ channel, as this decay is theoretically suppressed relative
to other decays~\cite{Page:1997rj}. The E852 collaboration have also
reported~\cite{Adams:1998ff} a signal for $1^{-+}$ state decaying into
$\rho\pi$ with a mass of around $1.6$ GeV. The decay width is in
reasonable agreement with theoretical calculations~\cite{Page:1997xs}.
Clearly the agreement between the possible experimental signals for
the $1^{-+}$ states and the lattice results is very poor.
The errors on the lattice results for $1^{-+}$
states are large relative to the errors on $\overline{Q}Q$ states. To
quantify the disagreement between experiment and lattice results the
systematic errors on the lattice simulation results should be reduced.
In particular the masses of the quarks used in the lattice simulation
should be reduced. It is possible that the states seen experimentally
are really $\overline{Q}\overline{Q}QQ$ states, in which case the
operators used in the lattice simulations would not couple strongly to
them.
Although the adiabatic lattice potential approach is not expected to
be a good description of the physics of light hybrids, we note that
the results of Juge, Kuti and Morningstar~\cite{Juge:1999ie}, show
that the splitting between the ground and first excited state is about
$200$ MeV, in broad agreement with the experimental results of
the E852 collaboration. Although no insight is gained about the different
decay widths.
To definitely identify an exotic hybrid meson requires both the
calculation of the mass as well as the decay widths.
There has been
very little work on hadronic decays on the lattice.
The most obvious hadronic process to study using lattice gauge theory is
the $\rho \rightarrow \pi\pi$ decay, however there have
only been a
few attempts to calculate the $g_{\rho \pi \pi}$
coupling~\cite{Gottlieb:1984rh} The GF11 lattice group
has recently computed the decay widths for the decay of the $0^{++}$
glueball to two pseudoscalars~\cite{Sexton:1995kd}.
The MILC collaboration~\cite{Bernard:1997ib} have started
to investigate the
mixing between the operator in eq.~\ref{eq:interponemp} and
the operator ($\pi \otimes a_1$) eq.~\ref{eq:QQQQ}.
\begin{equation}
\overline{\psi}^a \gamma_5 \psi^{a}
\overline{\psi}^b \gamma_5 \gamma_i \psi^{b}
\label{eq:QQQQ}
\end{equation}
which has the quantum numbers $1^{-+}$. This type of correlator would
naively be expected to yield the decay width of the $1^{-+}$ state to
$\rho$, and $a_1$. Unfortunately the analysis of Maiani and
Testa~\cite{Maiani:1990ca} shows that the matrix element required in
the computation of the decay width is hidden beneath an unphysical
term that increases exponentially with time. The origin of the
unphysical term comes from the requirement that both final mesons
should be onshell and is deeply related to theory being defined in
Euclidean space (required for us to have a well defined theory to
simulate). Some information may be extracted for onshell
processes~\cite{Maiani:1990ca,Sexton:1995kd}, using the methods
proposed by Michael~\cite{Michael:1989mf}.
\section{Acknowledgements}
I thank Chris Michael and Doug Toussaint for
discussions about exotic mesons.
|
\section*{Acknowledgements} The author would like to express his thanks to Jim Stasheff and Alice Rogers for useful communications on a earlier version of this preprint.
\section{Witten's postulates.}
In this section we shall describe to the reader the formulation of the model and the postulated algebra as originally given by Witten \cite{TSM}. Although the arguments are of the type common in the supersymmetry literature they are heuristic in nature, and from the point of view taken in this paper, mask the true origin of the ground from which the structures Witten defined actually arise. The aim of this paper is to elucidate this ground.
As explained in the foreword the basic idea is to consider embeddings of a Riemann surface $\Sigma$ into a symplectic manifold $M$, viz.
\begin{equation}
\phi:\Sigma \longrightarrow M \isto \phi:x^\alpha\longrightarrow u^i(x^\alpha).
\end{equation}
Witten then defined a multiplet by introducing the following field structures:
\begin{itemize}
\item $u^i(x^\alpha)$: the local embedding coordinates.
\item $\chi^i(x^\alpha)$: an anticommuting section of $\phi^*(TM)$.
\item $\varrho_\beta^i$: an anticommuting field ($\beta$=1,2 a tangent index to $\Sigma$, and $i=1,\dots,m$ runs over a basis of $\phi^*(TM)$), obeying the ``self-duality'' constraint:
\begin{equation}
\varrho_\beta^i\epsilon^\beta_\alpha =\varrho_\alpha^j \text{J}_j^i \label{duality1}\tag{D1}
\end{equation}
\item $H^{\alpha i}$: a commuting field ($\beta$=1,2 a tangent index to $\Sigma$, and $i=1,\dots,m$ runs over a basis of $\phi^*(TM)$), obeying the ``self-duality'' constraint:
\begin{equation}
H^{\beta i}\epsilon_{\beta}^{ \alpha} =H^{\alpha j}\text{J}_j^i \label{duality2}\tag{D2}
\end{equation}
\end{itemize}
Clearly the ``self-duality'' constraints where inspired by the pseudoholomorphic embedding condition (\ref{phe}).
Then Witten ``postulated'' the following ``fermionic'' transformation laws:
\begin{equation}
\delta u^i(x^\alpha) = \varepsilon \chi^i(x^\alpha) \label{s2} \tag{P1}
\end{equation}
\begin{equation}
\delta \chi^i(x^\alpha)=0 \label{s1}\tag{P2}
\end{equation}
The two transformations (\ref{s1}) and (\ref{s2}) are the usual nilpotent supersymmetry.
\begin{equation}
\delta \varrho_\alpha^i =\varepsilon (H_{\alpha}^i + \tfrac{1}{2}\epsilon_{\alpha \beta}(D_kJ^i_j)\chi^k\varrho^{\beta j}) - \varepsilon \Gamma^i_{jk}\chi^j\varrho^{k}_\alpha\label{s3}\tag{P3}
\end{equation}
The rather complicated structure of the transformation (\ref{s3}) is explained by Witten as follows: The term proportional to $D_kJ^i_j$ is needed for ``consistency'' with the duality relation (\ref{duality1}). The non-covariant term is needed for covariance under reparametrisation under the $u^i$. The transformation (\ref{s3}) is then regarded as defining $ H_{\alpha}^i$ so that any terms which might be added to the right hand side of (\ref{s3}) could be absorbed in the redefinition of $H$.
Finally Witten postulates the ``unpromising-looking formula'':
\begin{equation}
\begin{split}
\delta H^{\alpha i}&=-\tfrac{\varepsilon}{4}\chi^k\chi^l(\text{R}_{kl}{}^i{}_t +\text{R}_{klrs}J^{ri}J^s_t )\varrho^{\alpha t} \\
&+\varepsilon\epsilon_{\alpha \beta}(D_kJ^i_j)\chi^kH^{\beta j} \\
&-\tfrac{\varepsilon}{4}(\chi^kD_kJ^i_s)(\chi^lD_lJ^s_t)\varrho^{\alpha t} \\
&-\varepsilon \Gamma^i_{jk}\chi^j H^{ \alpha k}\label{s4}
\end{split}\tag{P4}
\end{equation}
After a rather lengthy calculation, one finds that $\delta^2 H^{\alpha i}=0$.
Witten then goes on to construct an action as a $\delta$-variation of some functional of the fields. After the construction of this action Witten then derives the Noether current corresponding to the $\delta$-variation symmetry, viz.
\begin{equation}
{\mathcal J}^\alpha =g_{ij}H^{\alpha i}\chi^j + J^{is}\varrho_s^\alpha D_kJ_{ij}\chi^k\chi^j \label{cnc}
\end{equation}
So summarising, in \cite{TSM} Witten:
\begin{itemize}
\item Introduces the field multiplet and certain constraints inspired by the psuodohomorphic embedding condition.
\item Postulates a Grassmann-odd Symmetry of this multiplet.
\end{itemize}
In this paper we shall elucidate the ground from which the definitions and postulates of Witten arise. We identify the Noether current (\ref{cnc}), an ``a posteriori derived object'' with the ``canonical'' Grassmann-odd momentum observable whose adjoint action (in the sense of the Leibniz bracket), generates the action induced by the prolongation to the graded covariant phase space of the original morphism of the configuration bundle. We shall see that all of the above structures are natural geometric structures on the graded covariant phase space of the authors covariant Hamiltonian BRST formalism as expounded in \cite{jdg2}. In \cite{TSM} Witten also formally introduces a generator Q which is said to generate the above transformations (\ref{s2})-(\ref{s4}) by the adjoint action of bracket $\{\text{Q},\cdot\}$. No explicit details of this formally introduced object are ever given. The notion that these transformations could be the transformations generated by a Hamiltonian BRST charge was taken up in \cite{Igarashi}. But here the authors do not derive the above transformations but simply restate the postulates in the BFV language. In this paper we give substance to these formal statements.
\section{Ambient diffeomorphism deformations of first-order embedded submanifolds}
In the multisymplectic formulation of field theories the space of configurations of classical field theories is defined to be a bundle over some parameter space the sections of which are the fields of interest. In the case of embeddings the ``fields'' are the sections of the bundle:
\[
\xymatrix{ M \times N \ar[d]^\pi \\ N \ar@/_15pt/[u]_{\phi}
}
\]
and as such are the graphs of the embeddings of N into M. Note that not all the fields of physics are sections of vector bundles, for here we have a bundle who's fibres are differentiable manifolds. The local coordinates on M are $(u^i)_{i=1,\cdots,m}$ and those on N be $(y^\mu)_{\mu=1,\cdots,n}$.
The covariant kinematical canonical phase space corresponding to ``first order embeddings''\footnote{``first order embeddings''$:=$ ``embeddings specified up to first order contact and no higher''.}, is the vertical covariant canonical phase space bundle (e.g. as described in \cite{jdg2}). The total space of this bundle is the affine dual of the first jet bundle, $\djb$, (see \cite{Crampin,jdg2}), the base space is the to be embedded submanifold N. The local coordinates on the covariant canonical phase space bundle, adapted to the bundle $\pi$ are $(y^\mu,u^i,p_\mu^i)$. Note that because the configuration bundle is trivial we may choose the trivial connection $\mathfrak{A}$=0 in this case the covariant canonical phase space bundle $\djb$ is endowed with the following multisymplectic (n+1)-form \cite{kan1}:
\begin{equation}
\Omega^V= du^i \wedge dp_i^\mu \wedge d^{n-1} y_\mu
\end{equation}
where $d^n y$ is the volume form on N and $d^{n-1} y_\mu:=\tfrac{\partial}{\partial y^\mu}\hook d^n y$. We then have:
{\defn
The {\it vertical covariant canonical phase space of first order embeddings} is the quadruple $(\djb, \tau^o , N,{\Omega^V})$.
}
In the multisymplectic formalism symmetries of field theories arise as morphisms of the configuration bundle which prolong to mutisymplectomorphisms of the covariant canonical phase space. We are interested here in the possible symmetries of an embedded submanifold inherited from the ambient space. It will be this symmetry which will underly the postulated transformations (\ref{s2})-(\ref{s4}) of Witten.
We wish to consider the covariant multimomentum mapping arising from the vertical bundle morphism of the configuration bundle:
\[
\xymatrix{M\times N \ar[rr]^{\text{Diff(M)}\times\text{Identity}} \ar[dr]^\pi& & \ar[ld]_\pi M\times N\\
& N
}
\]
given by the action of Diff(M) on the fibres $\pi^{-1}(u)\cong M$.
The Lie algebra corresponding to the infinite dimensional group of diffeomorphisms of M is denoted D(M) and is isomorphic to the space of all smooth vector fields on M, the corresponding diffeomorphisms being generated by the 1-parameter flows generated by the elements of D(M).
We wish to construct covariant Noether currents which will encode the invariance of our embeddings under this ambient diffeomorphism symmetry. The key idea will be to make use of the fact that we may re-express the notion of diffeomorphism invariance in terms of vertical automorphisms of the frame bundle.
Consider the automorphisms, Aut$\mathcal{F}$(M), of the frame bundle ($\mathcal{F}$(M)$,\pi^\mathcal{F}$,M). $\phi\in$Aut$\mathcal{F}$(M) iff $\phi$ is a diffeomorphism:
\begin{equation}
\phi:\mathcal{F}\text{(M)}\longrightarrow\mathcal{F}\text{(M)} \text{ s.t. }
\phi(\mathfrak{f}\cdot g)=\phi(\mathfrak{f})\cdot g
\end{equation}
for each $\mathfrak{f}\in$$\mathcal{F}$(M) and for all $g\in$G, where G is the structure group of $\mathcal{F}$(M). Such a $\phi$ satisfies $\pi^\mathcal{F}\circ\phi(\mathfrak{f}\cdot g)=\pi^\mathcal{F}\circ\phi(\mathfrak{f})$ and hence determines a diffeomorphism $\bar{\phi}$ of M defined by
\begin{equation}
\bar{\phi}(\pi^\mathcal{F}{\mathfrak f}):=\pi^\mathcal{F}\circ\phi({\mathfrak f}).
\end{equation}
One therefore has a homomorphism
\(j:\text{Aut} {\mathcal{F}}(M)\longrightarrow \text{Diff(M)}\). The kernel of this homomorphism, Gau$\mathcal{F}$(M), is called the {\it gauge group of the second kind}, the structure group G being known as the {\it gauge group of the first kind}. One then clearly has the following short exact sequence of groups:
\begin{equation}
\xymatrix{0 \ar[r] & \text{Gau}{\mathcal F}\text{(M)}\ar[r]^i &\text{Aut} {\mathcal{F}}\text{(M)}\ar[r]^j & \text{Diff(M)}\ar[r] & 0
} \label{sesg}
\end{equation}
Correspondingly one has (see \cite{Guill}) the following short exact sequence of Lie algebras:
\begin{equation}
\xymatrix{0 \ar[r] & \text{gau}{\mathcal F}\text{(M)}\ar[r]^i &\text{aut} {\mathcal{F}}\text{(M)}\ar[r]^j & \text{D(M)}\ar[r] & 0
} \label{sesl}
\end{equation}
where:
\begin{itemize}
\item $\text{gau}{\mathcal F}$ is the Lie algebra of G-invariant vertical vector fields,
\item $\text{aut} {\mathcal{F}}\text{(M)} $ is the Lie algebra of G-invariant vector fields,
\item $\text{D(M)} $ the algebra of smooth vector fields on M.
\end{itemize}
Let $\mathfrak{s}$ be any section of $\mathcal{F}$(M) representing a Cartan moving frame, that is:
\begin{equation}
\mathfrak{s}:M \longrightarrow \text{L}\mathcal{F}\text{(M)} \text{ :: }
\mathfrak{s}_A:u^i \longrightarrow {\Hat{\theta}_A}:=E_A^i \frac{\partial}{\partial u^i}
\end{equation}
where the $E_A^i$ are a (not orthonormal) vielbien basis (see Appendix I), and $A=1,\cdots,m$ is an anholonomic index.
Observe that a choice of moving frame (the section $\mathfrak{s}_A$) thus corresponds to a choice of equivalence class in the quotient D(M)$\cong\tfrac{\text{aut} {\mathcal{F}}\text{(M)}}{\text{gau}{\mathcal F}\text{(M)}}$.
\begin{figure}[h]
\[
\input{diag5.pstex_t}\]\caption{Diff(M) as global gauge transformations of the second kind.}\label{CT}
\end{figure}
The salient point is that the invariance under Diff(M) can be re-expressed in the form of a gauge invariance under a subgroup of vertical automorphisms of the linear frame bundle achieved by mapping all infinitesimal diffeomorphisms in M into infinitesimal gauge transformations of the second kind that take place along the fibres of $\mathcal{F}\text{(M)}$, (see \cite{Prugo} for a more detailed discussion).
For any $\psi\in$Diff(M) we can define a corresponding Cartan transformation, $ \text{D}(\psi)$, by:
\begin{equation}
\text{D}(\psi):\mathfrak{s}=({\Hat{\theta}_A})\mapsto\mathfrak{s'}:=
\psi_*(\mathfrak{s}\circ\psi^{-1})
\end{equation}
to another frame $\mathfrak{s'}$ with the same domain as that of $\mathfrak{s}$. Covering $\mathcal{F}\text{(M)}$ with a bundle atlas of local trivialisations associated with local sections of $\mathcal{F}\text{(M)}$, we can thus associate to each $\psi\in$Diff(M) a vertical automorphism $ \text{D}(\psi)$ \cite{Prugo}. This gives a isomorphism between Diff(M) and a subgroup G$_A$(Diff(M)) of the vertical automorphisms of $\mathcal{F}\text{(M)}$. Thus for each $ \text{D}(\psi)\in$G$_A$(Diff(M)) the action of the diffeomorphism $\psi\in$Diff(M) has been transformed into a gauge transformation of the second kind mapping the section $\mathfrak{s}$ into another section $\mathfrak{s'}$, see Figure \ref{CT}.
Given a choice of moving frame, ${\Hat{\theta}_A}$, any other can be obtained from it by virtue of a gauge transformation of the second kind. In this way one can generate from any arbitrary equivalence class in
\begin{equation} \text{D(M)}\cong\frac{\text{aut} {\mathcal{F}}\text{(M)}}{\text{gau}{\mathcal F}\text{(M)}}\end{equation} all others by the action of G$_A$(Diff(M)).
It is thus sufficient to show that our embedded submanifolds are invariant under the 1-parameter group of diffeomorphisms generated by an arbitrary choice of moving frame, or equivalence class in D(M)$\cong\tfrac{\text{aut} {\mathcal{F}}\text{(M)}}{\text{gau}{\mathcal F}\text{(M)}}$.
Let M be a Riemannian manifold with metric g, of signature (p,q). Then we may reduce the structure group of the frame bundle to {\bf SO}(p,q), and thus obtain the orthonormal frame bundle L$\mathcal{F}$(M). We shall henceforth work with an orthonormal frame.
We are free to choose the Levi-Civita connection on the the Riemannian manifold M.
A Cartan moving frame satisfies the following algebra:
\begin{equation}
[\Hat{\theta}_A,\Hat{\theta}_B]=C(u)^C_{AB} \Hat{\theta}_C
\end{equation}
where [ , ] is the Lie bracket on vector fields and $C(u)^C_{AB}$ are known as the Cartan structure functions defined by:
\begin{equation}\text{C}^A_{BC}:=\Gamma^A_{BC}-\Gamma^A_{CB}.\label{CSF}\end{equation}
The corresponding lifted momentum observable \cite{GIMMSY,jdg2} is:
\begin{equation}
\mathfrak{J}_A=\mathfrak{i}_{\mathcal{M}\pi,J^1\pi^\star}^* \circ \tau^* (\Hat{\theta}_A\hook z)=E_A^ip_i^\mu dy_\mu=:p_A^\mu dy_\mu \label{CovNC}
\end{equation}
We choose new coordinates on $J^1\pi^\star$, viz. $(y^\mu,u^i,p_A^\mu)$.
We shall also have occasion to use $\theta^A=E^A_idu^i$. In terms of these new coordinates the vertical multisymplectic potential takes the form:
\begin{equation}
\Theta^V=p_A^\mu{\Hat{\theta}}^A\wedge dy_\mu
\end{equation}
Then the vertical multisymplectic form may be written as:
\begin{equation}
\Omega^V={\Hat{\theta}}^A \wedge dp_A^\mu\wedge dy_\mu +\tfrac{1}{2}p_A^\alpha \text{C}^A_{BC}{\Hat{\theta}}^B \wedge{\Hat{\theta}}^C\wedge dy_\mu
\end{equation}
The Hamiltonian vector field $\mathfrak{X}[\mathfrak{J}_A]$, solving the multisymplectic structural equation,
\begin{equation}
\mathfrak{X}[\mathfrak{J}_A]\hook\Omega^V=d^V\mathfrak{J}_A
\end{equation}
is given by:
\begin{equation}
\mathfrak{X}[\mathfrak{J}_D]={\Hat{\theta}}_D +p_E^\nu \text{C}^E_{DG}\frac{\partial}{\partial p_G^\nu}
\end{equation} The vertical exterior derivative is d$^V:=d u^i \frac{\partial}{\partial u^i} +d p_i^\mu\frac{\partial}{\partial p_i^\mu}$.
By a short computation we then have:\newline
\begin{boxedminipage}[t]{12.75cm}
{\thm \label{T1}The Algebra generated by the covariant Noether currents $\mathfrak{J}_A$ is non-Abelian and is given by:}
\[
\{\mathfrak{J}_B ,\mathfrak{J}_C\}=\text{C}^A_{BC}\mathfrak{J}_A
\]where $\{\mathfrak{J}_B ,\mathfrak{J}_C\}=\mathfrak{X}[\mathfrak{J}_B]\hook d^V \mathfrak{J}_C$.
\end{boxedminipage}
{\rmk We argued earlier in this section that the ambient diffeomorphism invariance of first order embedded submanifolds could be encoded into the covariant Noether currents (\ref{CovNC}), since they give a representation of the Lie algebra of the 1-parameter diffeomorphisms generated by an arbitrary choice of the Cartan moving frame, or equivalently to an equivalence class in D(M)$\cong\tfrac{\text{aut} {\mathcal{F}}\text{(M)}}{\text{gau}{\mathcal F}\text{(M)}}$. Theorem \ref{T1} expresses the fact that the covariant Noether currents corresponding to the ambient diffeomorphism symmetry \[Diff(M):M \times N\ \longrightarrow M \times N \quad :\quad : \quad \Phi:(u(y),y)\longrightarrow (\Phi u(y),y)\] acting by bundle automorphisms over the identity on $\Sigma$ satisfy a non-Abelian algebra.
}
{\rmk \label{R2} Note that the Cartan structure functions are defined as the antisymmetrised connections in the real analytic anholonomic basis. We may therefore locally choose a synchronous frame in which the connections and therefore the Cartan structure functions vanish. In this way we see that the deformation algebra of Theorem \ref{T1} may be locally Abelianised. We shall see below that the particular form of the structure functions needed to obtain the algebra of Witten (\ref{s2})- (\ref{s4}) also depends on a special choice of local basis, viz. a Darboux basis.}
{\section{The covariant phase space of pseudoholomorphic maps}}
We now turn our attention to a particular example of first order embeddings, viz. the pseudoholomophic embeddings of Riemann surfaces $(\Sigma,\epsilon)$ with complex structure $\epsilon$ into an almost K\"{a}hler manifold with metric $g$, compatible almost complex structure J, and symplectic structure $\omega$, viz. $(M,\omega,g,J)$. Following the discussion in the last section we then have:
{\defn The {\it configuration bundle } $\pi^{\Sigma}$ for the embeddings of Riemann surfaces into a symplectic manifold is the triple \(( M \times \Sigma,\pi^{\Sigma},\Sigma)\).
}
The kinematical phase space is a particular subbundle of the vertical covariant canonical phase space of embeddings. In order to explicitly construct this subbundle we shall require an explicit coordinisation.
In order to obtain coordinates with the correct dimensionality we now introduce
the J-pseudoholomorphic anholonomic basis vielbeins on M, viz. (E$^a_i$,E$^{a*}_i$) and also vielbeins taking us from the analytic basis indices $\alpha$ on $\Sigma$ to the complex analytic holonomic basis indices $(\kappa,\kappa*)$, viz. $(E_\alpha^\kappa,E_\alpha^{\kappa*})$. See {\scshape Appendix I} for more details.
In terms of these veilbeins one then has the identities:
\begin{alignat}{2} \label{multi-momentum identities}
p^\alpha_i&=\text{E}^{a}_ip_{a}^{\kappa}\text{E}_{\kappa}^\alpha &+
\text{E}^{a*}_ip_{a*}^{\kappa}\text{E}_{\kappa}^\alpha &+
\text{E}^{a}_ip_{a}^{\kappa*}\text{E}_{\kappa*}^\alpha +
\text{E}^{a*}_ip_{a*}^{\kappa*}\text{E}_{\kappa*}^\alpha \\
p_{a}^{\kappa} &= \text{E}_{a}^ip_i^\alpha \text{E}_\alpha^{\kappa} &\quad
p_{a*}^{\kappa} &= \text{E}_{a*}^ip_i^\alpha \text{E}_\alpha^{\kappa} \\
p_{a}^{\kappa*} &= \text{E}_{a}^ip_i^\alpha \text{E}_\alpha^{\kappa*} &\quad
p_{a*}^{\kappa*} &= \text{E}_{a*}^ip_i^\alpha \text{E}_\alpha^{\kappa*}
\end{alignat}
The vielbeins have be so constructed that they are infact eigenvectors of the respective complex structures (see \cite{Hsuing}). That is,
\begin{alignat}{2}
J^i_jE^j_a & =iE^i_a &\quad J^i_jE^j_{a*} & =-iE^i_{a*} \\
J_i^jE_j^a & =iE_i^a &\quad J_i^jE_j^{a*} & =-iE_i^{a*} \\
\varepsilon^\alpha_\beta E^\beta_{\kappa}&=iE^\alpha_{\kappa} &\quad
\varepsilon^\alpha_\beta E^\beta_{\kappa*}&=-iE^\alpha_{\kappa*} \\
\varepsilon_\alpha^\beta E_\beta^{\kappa}&=iE_\alpha^{\kappa} &\quad
\varepsilon_\alpha^\beta E_\beta^{\kappa*}&=-iE_\alpha^{\kappa*}
\end{alignat}
We then have the following:\newline
\begin{boxedminipage}[t]{12.75cm}
{\prop\label{Embb}
If $p_i^\alpha$ satisfies the pseudoholomorphicity condition $J^i_jp_i^\alpha = \varepsilon^\alpha_\beta p_j^\beta$ then:
\begin{gather}
p_{a}^{\kappa*}=0 \text{ and } p_{a*}^{\kappa}=0
\end{gather}
}
\end{boxedminipage}
\begin{proof}
If $p_i^\alpha$ satisfies the pseudoholomorphicity condition then:
\begin{equation}
\begin{split}
p_{a*}^{\kappa} &= \text{E}_{a*}^ip_i^\alpha \text{E}_\alpha^{\kappa} \\
&= \text{E}_{a*}^i[-J_i^j \varepsilon^\alpha_\beta p_j^\beta ] \text{E}_\alpha^{\kappa} \\
&= -(i)(-i)\text{E}_{a*}^ip_i^\alpha \text{E}_\alpha^{\kappa} \\
\text{then } p_{a*}^{\kappa}&=-p_{a*}^{\kappa} \\
\Longrightarrow p_{a*}^{\kappa}&= 0
\end{split}
\end{equation}
That $p_{a}^{\kappa*}=0$ follows by complex conjugation.
\end{proof}
{\rmk One may easily verify that no constraints result from pseudoholomorphicity on $p_{a*}^{\kappa*}$ nor $p_{a}^{\kappa}$.}
Proposition \ref{Embb} enables us to define the subbundle ${\text{J}^1 \pi^*_+{}^\Sigma}$ by the vanishing of half of the components of the multimomenta.
{\defn The subjective submersion $\mathbb{P_+}$ is the fibrewise map:
\begin{equation}
\mathbb{P_+}:{\text{J}^1 \pi^*{}^\Sigma} \longrightarrow {\text{J}^1 \pi^*_+{}^\Sigma} \hspace{5pt} : \hspace{5pt}: \hspace{5pt}
\mathbb{P_+}:(p_i^\alpha)=(p_a^\kappa, p_{a*}^\kappa ,p_a^{\kappa *} ,p_{a*}^{\kappa *} ) \longrightarrow (p_a^\kappa,0,0,p_{a*}^{\kappa *})
\end{equation}
}
The local adapted coordinates on $\pdjb$ may therefore be written $(x^\alpha,u^i,p_a^\kappa,p_{a*}^{\kappa *})$.
Let $\mathfrak{i}_{J^1\pi^\star|_+, J^1\pi^\star}:J^1\pi^\star|_+\longrightarrow J^1\pi^\star$ denote the natural embedding of $J^1\pi^\star|_+$ into $J^1\pi^\star$.
{\defn
The { vertical canonical covariant phase space of pseudoholomorphic maps} is $(\pdjb,{\Omega^V_+})$ where
$\Omega^V_+:= {\mathfrak{i}_{J^1\pi^\star|_+, J^1\pi^\star}}^*[\Omega^{V}].$ }
\section{The induced ambient diffeomorphism deformations of pseudoholomorphic mappings}
In order to effect the analysis of the ambient diffeomorphism symmetry of psudoholomorphic embeddings
we now introduce new vielbeins taking us from the real analytic anholonomic basis with indices A to the J-pseudoholomorphic anholonomic basis (a,a*), viz $(E_A^a,E_A^{a*})$, see {\scshape Appendix I}.
We may then induce the symmetry generated by the Noether currents and the multisymplectic potential from $J^1\pi^\star$ to $J^1\pi^\star|_+$ via the pull-back by $\mathfrak{i}_{J^1\pi^\star|_+, J^1\pi^\star}$. The induced objects are then:
\begin{equation}
\begin{split}
\mathfrak{i}_{J^1\pi^\star|_+, J^1\pi^\star}^*(\mathfrak{J}_A)&=
\mathfrak{i}_{J^1\pi^\star|_+, J^1\pi^\star}^*
(E_A^{a}p_{a}^{\kappa} E_{\kappa}^\alpha dx_\alpha + E_A^{a*}p_{a*}^{\kappa*} E_{\kappa*}^\alpha dx_\alpha\\ &+ E_A^{a*}p_{a*}^{\kappa} E_{\kappa}^\alpha dx_\alpha + E_A^{a}p_{a}^{\kappa*} E_{\kappa*}^\alpha dx_\alpha) \\
&=E_A^{a}p_{a}^{\kappa} E_{\kappa}^\alpha dx_\alpha + E_A^{a*}p_{a*}^{\kappa*} E_{\kappa*}^\alpha dx_\alpha
\end{split}
\end{equation}
By a similar calculation one also has:
\begin{equation}
\mathfrak{i}_{J^1\pi^\star|_+, J^1\pi^\star}^*
{\Hat{\theta}}^V = E_A^{a}p_{a}^{\kappa}{\Hat{\theta}}^A \wedge dx_{\kappa} + E_A^{a*}p_{a*}^{\kappa*}{\Hat{\theta}}^A \wedge dx_{\kappa*}
\end{equation}
Let us denote the pull-back of $\mathfrak{J}_A$ by $\mathfrak{J}_A|_+$ and the pull-back of $\Theta$ by $\Theta|_+$.
The multisymplectic form corresponding to $\Theta|_+$ is then:
\begin{equation}
\begin{split}
\Omega^V|_+ &= E_A^{a}{\Hat{\theta}}^A \wedge dp_{a}^{\kappa} \wedge dx_{\kappa}
+ E_A^{a*}{\Hat{\theta}}^A \wedge dp_{a*}^{\kappa*} \wedge dx_{\kappa*} \\
&- p_{a}^{\kappa}(\omega^{a}_{b}E^{b}_A + \omega^{a}_{b*}E^{b*}_A){\Hat{\theta}}^A \wedge dx_{\kappa} \\
&- p_{a*}^{\kappa*}(\omega^{a*}_{b}E^{b}_A + \omega^{a*}_{b*}E^{b*}_A){\Hat{\theta}}^A \wedge dx_{\kappa*}
\end{split}
\end{equation}
Where $\omega^a_b=\Gamma_{Cb}^a{\Hat{\theta}}^C$ are the corresponding connection 1-forms.
$\mathfrak{J}_A|_+$ is a Hamiltonian form w.r.t. $\Omega^V|_+$ with Hamiltonian vector field given by:
\begin{equation}
\begin{split}
\mathfrak{X}[\mathfrak{J}_D|_+]= {\Hat{\theta}}_D &+ p_{a}^{\kappa}(\Gamma^{a}_{(Cb)}E^{b}_D + \Gamma^{a}_{(Cb*)}E^{b*}_D)E^C_{d}\frac{\partial }{\partial p_{d}^{\kappa}} \\
&+ p_{a*}^{\kappa*}(\Gamma^{a*}_{(Cb)}E^{b}_D + \Gamma^{a*}_{(Cb*)}E^{b*}_D)E^C_{d*}\frac{\partial }{\partial p_{d*}^{\kappa*}} \\
&+ 2p_{a}^{\kappa}(\Gamma^{a}_{[Cb]}E^{b}_D + \Gamma^{a}_{[Cb*]}E^{b*}_D)E^C_{d}\frac{\partial }{\partial p_{d}^{\kappa}} \\
&+ 2p_{a*}^{\kappa*}(\Gamma^{a*}_{[Cb]}E^{b}_D + \Gamma^{a*}_{[Cb*]}E^{b*}_D)E^C_{d*}\frac{\partial }{\partial p_{d*}^{\kappa*}}
\end{split}
\end{equation}
By a rather lengthy, but straightforward calculation one then has the following result:\newline
\begin{boxedminipage}[t]{12.75cm}
{\thm \label{T2}The Algebra generated by the covariant Noether currents $\mathfrak{J}_A|_+$ is non-Abelian and is given by:}
\[
\{ \mathfrak{J}_B|_+ ,\mathfrak{J}_C|_+\}=\text{C}^A_{BC}\mathfrak{J}_A|_+
\]
\end{boxedminipage}
{\rmk Theorem \ref{T2} shows us that the non-Abelian deformation diffeomorphism symmetry of embeddings is inherited by the subclass of pseudoholomorphic embeddings. The origin of the remarks of Witten describing his construction of the algebra (\ref{s2})-(\ref{s4}) that the algebra must be amended in order to be consistent with the ``duality constraints'' (\ref{duality1}) and (\ref{duality2}) would thus seem to be at variance with the point of view taken here.
However, if one works locally then the algebra of deformations can be Abelianised by choosing an synchronous frame (see Remark {\ref{R2}}). It was shown in an earlier version of this preprint that one can still obtain Witten's algebra from the local Abelian algebra if one considers the algebra induced by an extrinsic embedding of $\pdjb$ into $\djb$. The approach originally taken by Witten thus seems to be this extrinsic point of view. The intrinsic point of view taken in this article has the advantage of: (i) clearly elucidating the origin of the symmetry in Diff(M), (ii) does not make use of an overdetermined coordinate system. }
\section{The graded canonical kinematical phase space of pseudoholomorphic embeddings}
In this section we shall briefly describe the construction of the graded canonical kinematical phase space of pseudoholomorphic embeddings as follows from the general covariant Hamiltonian BRST formalism developed in \cite{jdg2}. In order to contain this section within a reasonable length we have assumed that the reader will have some familiarity this reference.
Consider the natural projection $\iota: M \times \Sigma\longrightarrow M$. We may use this natural projection to pull-back the linear frame bundle L$\mathcal{F}$(M) onto $M \times \Sigma$. One may then construct the graded bundles of the covariant Hamiltonian BRST formalism according to Figure \ref{F3} (reading from right to left):
\begin{figure}[h]
\[
\xymatrix{J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_{+} \ar[rr]^{\tau^0_{1+}} \ar[rrrd]_>>>>>>>{\tau^0_+} & &\Pi[\iota^*[\text{L}\mathcal{F}]] \ar[dr]^{\pi^{\Pi\text{L}\mathcal{F}\Sigma}} & & \ar[ll]_\Pi \iota^*[\text{L}\mathcal{F}] \ar[dl]^{\pi^{\text{L}\mathcal{F}\Sigma}} \ar[dd]^{\iota^*({\pi^{\text{L}\mathcal{F}}})} \ar[r]& \text{L}\mathcal{F}\ar[dd]^{\pi^{\text{L}\mathcal{F}}} \\
& \ar[ul]_{\mathbb{P}_{++}} \ar[dl]^{\mathbb{P}_{--}} J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star \ar[ur]_<<<<{\tau^0_1} & & \Sigma & & \\
J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_{-} & & \ar[ul]^\varrho \mathcal{S}\pi^{\Pi\text{L}\mathcal{F}} \ar[ur]^\kappa \ar[uu]^\mu \ar[rr]& &M\times \Sigma \ar[ul]^\pi \ar[r]^\iota& M
}\]\caption{The construction of the graded bundles.}\label{F3}
\end{figure}
The triple $(J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_{+},\tau^0_{+},\Sigma)$ is the sought after {\it graded vertical covariant canonical phase space of pseudoholomorphic embeddings}. The construction is outlined as follows (see \cite{jdg2} for the general construction):
\begin{itemize}
\item First construct the pull-back of the frame bundle as a bundle over $M\times\Sigma$ by the natural projection.
\item Then construct the {\it Legendre bundle} $(\iota^*[\text{L}\mathcal{F}],\pi^{\text{L}\mathcal{F}\Sigma},\Sigma)$ where $\pi^{\text{L}\mathcal{F}\Sigma}:=\pi \circ\iota^*(\pi^{\text{L}\mathcal{F}})$.
\item The {\it graded configuration bundle} is then given by the inversion functor (see \cite{voronov}) to be $(\iota^*[\text{L}\mathcal{F}],\pi^{\Pi\text{L}\mathcal{F}\Sigma},\Sigma)$. The adapted local coordinates are $(x^\alpha,u^i,\eta^A)$. The $\eta^A$ carries an $\mathfrak{so}$(m) index and is Grassmann-odd.
\item One then constructs the {\it graded multiphase space } $\mathcal{S}\pi^{\Pi\text{L}\mathcal{F}}$ over the graded configuration bundle according to the usual method. It is endowed with the tautologous canonical form:
\begin{equation*}
{\Theta}^{\text{E}}=p\wedge d^2x + p_i^\alpha du^i \wedge dx_\alpha + \mathcal{P}_A^\alpha d \eta^A \wedge dx_\alpha
\end{equation*}
\item The {\it graded covariant canonical phase space bundle of pseudoholomorphic embeddings} is then defined by the short exact sequence to be $J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star$. It has local adapted coordinates $(x^\alpha,u^i,\eta^A,p_i^\alpha,\mathcal{P}_A^\alpha)$.
\end{itemize}
Then just as in equation (\ref{multi-momentum identities}) we also have:
\begin{alignat}{2} \label{graded multi-momentum identities}
\mathcal{P}^\alpha_A&=\text{E}^{a}_A\mathcal{P}_{a}^{\kappa}\text{E}_{\kappa}^\alpha &+
\text{E}^{a*}_A\mathcal{P}_{a*}^{\kappa}\text{E}_{\kappa}^\alpha &+
\text{E}^{a}_A\mathcal{P}_{a}^{\kappa*}\text{E}_{\kappa*}^\alpha +
\text{E}^{a*}_A\mathcal{P}_{a*}^{\kappa*}\text{E}_{\kappa*}^\alpha \\
\mathcal{P}_{a}^{\kappa} &= \text{E}_{a}^A\mathcal{P}_A^\alpha \text{E}_\alpha^{\kappa} &\quad
\mathcal{P}_{a*}^{\kappa} &= \text{E}_{a*}^A\mathcal{P}_i^\alpha \text{E}_\alpha^{\kappa} \\
\mathcal{P}_{a}^{\kappa*} &= \text{E}_{a}^A\mathcal{P}_A^\alpha \text{E}_\alpha^{\kappa*} &\quad
\mathcal{P}_{a*}^{\kappa*} &= \text{E}_{a*}^A\mathcal{P}_A^\alpha \text{E}_\alpha^{\kappa*}
\end{alignat}
in terms of the ``combined vielbeins'', see {\scshape Appendix I}. Moreover,\newline
\begin{boxedminipage}[t]{12.75cm}
{\prop
If $p_i^\alpha$ and $\mathcal{P}^\alpha_A$ satisfy the pseudoholomorphicity conditions $J^i_jp_i^\alpha = \varepsilon^\alpha_\beta p_j^\beta$ and $J^A_B\mathcal{P}_A^\alpha = \varepsilon^\alpha_\beta \mathcal{P}_B^\beta$ then:
\begin{gather}
p_{a}^{\kappa*}=0 \text{ ; } p_{a*}^{\kappa}=0\text{ ; }\mathcal{P}_{a}^{\kappa*}=0 \text{ and } \mathcal{P}_{a*}^{\kappa}=0
\end{gather}
}
\end{boxedminipage}
{\rmk As before, one may easily verify that no constraints result from pseudoholomorphicity on $p_{a*}^{\kappa*}$, $p_{a}^{\kappa}$, $\mathcal{P}_{a*}^{\kappa*}$ nor on $\mathcal{P}_{a}^{\kappa}$.}
{\defn The subjective submersion $\mathbb{P_{++}}$ is the fibrewise map:
\begin{equation}
\begin{split}
\mathbb{P}_{++}&:J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star \longrightarrow J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_+ \hspace{5pt} : \hspace{5pt}: \hspace{5pt} \\
\mathbb{P}_{++}&:(p_i^\alpha,\mathcal{P}^\alpha_A)=(p_a^\kappa, p_{a*}^\kappa ,p_a^{\kappa *} ,p_{a*}^{\kappa *},\mathcal{P}_a^\kappa, \mathcal{P}_{a*}^\kappa ,\mathcal{P}_a^{\kappa *} ,\mathcal{P}_{a*}^{\kappa *} ) \longrightarrow (p_a^\kappa,0,0,p_{a*}^{\kappa *},\mathcal{P}_a^\kappa,0,0,\mathcal{P}_{a*}^{\kappa *})
\end{split}
\end{equation}
}
The local adapted coordinates on $J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_+$ may therefore be written in the following form: $(x^\alpha,u^i,\eta^A, p_a^\kappa,p_{a*}^{\kappa *},\mathcal{P}_a^\kappa,\mathcal{P}_{a*}^{\kappa *})$.
{\defn The graded extended covariant kinematical phase space of embeddings is the quadruple $(J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_{+},\tau^0_{+},\Sigma,\Omega^V|_+)$ where $\Omega^{\text{EV}}|_+:=(\mathbb{P}_{++} \circ \varrho )^*[-d\Theta^{\text{EV}}]$.
}
In terms of the local adapted coordinates on $J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_+$ one has the following local expression for the induced vertical Cartan form:
\begin{equation}
\begin{split}
\Omega^{\text{EV}}|_+ &= E_A^{a}{\Hat{\theta}}^A \wedge dp_{a}^{\kappa} \wedge dx_{\kappa}
+ E_A^{a*}{\Hat{\theta}}^A \wedge dp_{a*}^{\kappa*} \wedge dx_{\kappa*} \\
&-E_A^{a}d\eta^A \wedge d\mathcal{P}_{a}^{\kappa} \wedge dx_{\kappa}
-E_A^{a*}d\eta^A \wedge d\mathcal{P}_{a*}^{\kappa*} \wedge dx_{\kappa*} \\
&- p_{a}^{\kappa}(\omega^{a}_{b}E^{b}_A + \omega^{a}_{b*}E^{b*}_A) \wedge {\Hat{\theta}}^A \wedge dx_{\kappa} \\
&- p_{a*}^{\kappa*}(\omega^{a*}_{b}E^{b}_A + \omega^{a*}_{b*}E^{b*}_A) \wedge{\Hat{\theta}}^A \wedge dx_{\kappa*}\\
&+ \mathcal{P}_{a}^{\kappa}(\omega^{a}_{b}E^{b}_A + \omega^{a}_{b*}E^{b*}_A)\wedge d\eta^A \wedge dx_{\kappa} \\
&+ \mathcal{P}_{a*}^{\kappa*}(\omega^{a*}_{b}E^{b}_A + \omega^{a*}_{b*}E^{b*}_A)\wedge d\eta^A \wedge dx_{\kappa*}
\end{split}
\end{equation}
\section{The BRST algebra of the ambient diffeomorphism deformations of pseudoholomorphic embeddings}
Recall that in Theorem \ref{T1} we proved that the algebra of the covariant Noether currents corresponding to the ambient diffeomorphism deformations of embeddings satisfied a non-Abelian algebra whose structure functions are the structure functions of Cartan. That is:
\[
\{ \mathfrak{J}_B ,\mathfrak{J}_C\}=\text{C}^A_{BC}\mathfrak{J}_A
\]
According to the general formalism developed in \cite{jdg2}, the covariant BRST Noether current is the Grassmann-odd 1-form:
\begin{equation}
\Upsilon=\eta^A\mathfrak{J}_A-\tfrac{1}{2}\text{C}^A_{BC}\eta^B\eta^C\mathcal{P}_A^\alpha dx_\alpha
\end{equation}
We shall not pause to consider the BRST algebra generated by $\Upsilon$ but shall pass on to the case of pseudoholomorphic embeddings. The algebra is given, as in Theorem \ref{T2}, by:
\[
\{ \mathfrak{J}_B|_+ ,\mathfrak{J}_C|_+\}=\text{C}^A_{BC}\mathfrak{J}_A|_+
\]
The the covariant BRST Noether current is the Grassmann-odd 1-form corresponding to this algebra is the pull back $\Upsilon|_+=(\mathbb{P}_{++}\circ\varrho)^*[\Upsilon]$ which is given by:
\begin{equation}
\Upsilon|_+=\eta^A\mathfrak{J}_A|_+ -\tfrac{1}{2}\text{C}^A_{BC}\eta^B\eta^C\mathcal{P}_A^\alpha|_+ dx_\alpha
\end{equation}
where $\mathcal{P}_A^\alpha|_+=\text{E}^{a}_A\mathcal{P}_{a}^{\kappa}\text{E}_{\kappa}^\alpha +
\text{E}^{a*}_A\mathcal{P}_{a*}^{\kappa*}\text{E}_{\kappa*}^\alpha$.
In order to calculate the BRST algebra we must first calculate the Hamiltonian vector fields corresponding to the observables, which are to be built from the ``field content'' of the theory. The ``fields'' in this case are the local fibre coordinates on $(J^1{\pi^{\Pi\text{L}\mathcal{F}}}^\star|_{+},\tau^0_{+},\Sigma,\Omega^V|_+)$.That is, the field content is: \[(u^i,\eta^A, p_a^\kappa,p_{a*}^{\kappa *},\mathcal{P}_a^\kappa,\mathcal{P}_{a*}^{\kappa *}).\]
We construct from this field multiplet the following Hamiltonian forms:
\begin{alignat}{1}
u^i; &\quad \mathfrak{J}_A|_+; \\
\eta^A; &\quad \mathcal{P}_A|_+.
\end{alignat}
The corresponding Hamiltonian form-valued (multi-)vector fields are:
\begin{equation}
\mathfrak{X}[u^i]=\text{E}^i_{a} \frac{\partial}{\partial z^{\kappa *}} \wedge \frac{\partial}{\partial p_{a}^{\kappa}} + \text{E}^i_{a*} \frac{\partial}{\partial z^{\kappa }} \wedge \frac{\partial}{\partial p_{a*}^{\kappa *}}
\end{equation}
\begin{equation}
\begin{split}
\mathfrak{X}[\mathfrak{J}_D|_+]= {\Hat{\theta}}_D &+ p_{a}^{\kappa}(\Gamma^{a}_{(Cb)}E^{b}_D + \Gamma^{a}_{(Cb*)}E^{b*}_D)E^C_{d}\frac{\partial }{\partial p_{d}^{\kappa}} \\
&+ p_{a*}^{\kappa*}(\Gamma^{a*}_{(Cb)}E^{b}_D + \Gamma^{a*}_{(Cb*)}E^{b*}_D)E^C_{d*}\frac{\partial }{\partial p_{d*}^{\kappa*}} \\
&+ 2p_{a}^{\kappa}(\Gamma^{a}_{[Cb]}E^{b}_D + \Gamma^{a}_{[Cb*]}E^{b*}_D)E^C_{d}\frac{\partial }{\partial p_{d}^{\kappa}} \\
&+ 2p_{a*}^{\kappa*}(\Gamma^{a*}_{[Cb]}E^{b}_D + \Gamma^{a*}_{[Cb*]}E^{b*}_D)E^C_{d*}\frac{\partial }{\partial p_{d*}^{\kappa*}} \\
&- \mathcal{P}_{a}^{\kappa}(\Gamma^{a}_{Db}E^{b}_A + \Gamma^{a}_{Db*}E^{b*}_A)d\eta^A \text{E}^E_{e}\frac{\partial}{\partial p_{e}^{\kappa}}\wedge {\Hat{\theta}}_E \\
&- \mathcal{P}_{a*}^{\kappa*}(\Gamma^{a*}_{Db}E^{b}_A + \Gamma^{a*}_{Db*}E^{b*}_A) d\eta^A \text{E}^E_{e*}\frac{\partial}{\partial p_{e*}^{\kappa*}}\wedge {\Hat{\theta}}_E
\end{split}
\end{equation}
\begin{equation}
\mathfrak{X}[\eta^A]= - \biggl(\text{E}^A_{a} \frac{\partial}{\partial z^{\kappa *}} \wedge \frac{\partial}{\partial \mathcal{P}_{a}^{\kappa}} + \text{E}^i_{a*} \frac{\partial}{\partial z^{\kappa }} \wedge \frac{\partial}{\partial \mathcal{P}_{a*}^{\kappa *}} \biggr)
\end{equation}
\begin{equation}
\begin{split}
\mathfrak{X}[\mathcal{P}_A|_+]=-\frac{\partial}{\partial \eta^A}
\end{split}
\end{equation}
The BRST algebra is then obtained by taking the interior product of the above
form-valued (multi-)vector fields with the exterior derivative of $\Upsilon|_+$. The latter object may be expressed as:
\begin{equation}
\begin{split}
d\Upsilon|_+&= d\eta^A \wedge \mathfrak{J}_A|_+ +\eta^A d \mathfrak{J}_A|_+
-\text{C}^A_{BC}(d\eta^B)\eta^C\mathcal{P}_A^\alpha|_+ dx_\alpha \\
&-\tfrac{1}{2}\text{C}^A_{BC}\eta^B\eta^C d\mathcal{P}_A^\alpha|_+ dx_\alpha
-\tfrac{1}{2}[\Hat{\theta}_D\text{C}^A_{BC}\Hat{\theta}^D]\eta^B\eta^C\mathcal{P}_A^\alpha|_+ dx_\alpha
\end{split}
\end{equation}
By a straightforward calculation we then have:\newline
\begin{boxedminipage}[t]{12.75cm}
{\thm[Ambient diffeo BRST Algebra: the pseudoholomorphic case] \label{T3} The BRST algebra corresponding to the diffeomorphism deformation algebra of pseudoholomorphic embeddings is given by:}
\begin{align}
\delta_{\Upsilon|_+}[u^i]&:= \mathfrak{X}[u^i]\hook d \Upsilon=\text{E}^i_A\eta^A=:\chi^i \\
\delta_{\Upsilon|_+}[\chi^i]&:= \mathfrak{X}[\chi^i]\hook d \Upsilon|_+ =0 \\
\delta_{\Upsilon|_+}[\mathcal{P}_A|_+]&:= \mathfrak{X}[\mathcal{P}_A|_+]\hook d \Upsilon|_+ = - [\mathfrak{J}_A|_+ +
\text{C}^B_{AC}\eta^C\mathcal{P}_B^\alpha|_+ dx_\alpha ] \\
\delta_{\Upsilon|_+}[\mathfrak{J}_A|_+]&:= \mathfrak{X}[\mathfrak{J}_A|_+]\hook d \Upsilon|_+= \eta^C \text{C}^D_{AC}\mathfrak{J}_D|_+
- \tfrac{1}{2}[\text{D}_A\text{C}^D_{BC}]\eta^B\eta^C\mathcal{P}_D^\alpha|_+ dx_\alpha \label{LT} \end{align}
\end{boxedminipage}
The covariant derivative appears in the last term of the transformation (\ref{LT}) by virtue of identity (\ref{id12}) in Appendix II.
\section{Relation to Witten's BRST algebra}
We can already see the beginnings of an emergence of Witten's algebra in Theorem \ref{T3}. What is missing is the fact that the structure functions that are implicit in the construction of Witten involve the covariant derivatives of the almost complex structures. This form of the structure functions, as we shall show below, are in fact the Cartan structure functions expressed in a Darboux basis on M. They are therefore an artifact of the local geometry of the symplectic manifold M. Once we establish this fact the algebra of Witten then follows.\newline
\begin{boxedminipage}[t]{12.75cm}
{\prop \label{P1} The Levi-Civita connection on a symplectic manifold, expressed in terms of a real anholonomic Darboux basis is given by the following expression:}
\begin{equation}
\Gamma^A_{BC}=\tfrac{1}{2}\text{J}^A_D\text{D}_B\text{J}^D_C\label{E3}
\end{equation}
\end{boxedminipage}
\begin{proof}
We perform all computations in a real anholonomic Darboux basis.
Consider the compatibility equation between the almost complex structure J, symplectic form $\omega$ and metric $g$, viz. \(\omega_{AB}=J^C_Ag_{CB} \). Taking the covariant derivative of both sides and by use of the compatibility equation we find that:
\begin{equation}
\begin{split}
\text{D}_E\omega_{AB}&=(\text{D}_EJ^C_A)g_{CB} \\
&=-(\text{D}_EJ^C_A)J^F_CJ_F^Gg_{GB}\\
&=-J^F_C(\text{D}_EJ^C_A)\omega_{FB}\label{E1}
\end{split}
\end{equation}
Computing the covariant derivative in local Darboux basis on M:
\begin{equation}
\text{D}_E\omega_{AB}= -2\Gamma^F_{AE}\omega_{FB}\label{E2}
\end{equation}
On comparison of equations (\ref{E1}) and (\ref{E2}) one obtains (\ref{E3}).
\end{proof}
We then obtain the structure functions of Witten by antisymmetrising:\newline
\begin{boxedminipage}[t]{12.75cm}
{\cor The Cartan structure functions in a real anholonomic Darboux basis on M are identical to Witten's structure functions, viz. }
\begin{equation}
\text{C}^A_{BC}:=\Gamma^A_{[BC]}=\tfrac{1}{2}\text{J}^A_D\text{D}_{[B}\text{J}^D_{C]}\label{E4}
\end{equation}
\end{boxedminipage}
Together with Theorem \ref{T3}, this Corollary then yields Witten's BRST algebra:\newline
\begin{boxedminipage}[t]{12.75cm}
{\thm[Witten's Algebra] \label{T5}
The BRST algebra corresponding to the diffeomorphism deformation algebra of pseudoholomorphic embeddings in a real anholonomic Darboux basis is given by:}
\begin{align}
\delta_{\Upsilon|_+}[u^i]&:= \mathfrak{X}[u^i]\hook d \Upsilon=\text{E}^i_A\eta^A=:\chi^i \\
\delta_{\Upsilon|_+}[\chi^i]&:= \mathfrak{X}[\chi^i]\hook d \Upsilon|_+ =0 \\
\delta_{\Upsilon|_+}[\mathcal{P}_A|_+]&:= \mathfrak{X}[\mathcal{P}_A|_+]\hook d \Upsilon|_+ = - [\mathfrak{J}_A|_+ +
\tfrac{1}{2}\text{J}^B_D\text{D}_{[A}\text{J}^D_{C]}\eta^C\mathcal{P}_B^\alpha|_+ dx_\alpha ] \\
\delta_{\Upsilon|_+}[\mathfrak{J}_A|_+]&:= \mathfrak{X}[\mathfrak{J}_A|_+]\hook d \Upsilon|_+= \eta^C \tfrac{1}{2}\text{J}^D_F\text{D}_{[A}\text{J}^F_{C]}\mathfrak{J}_D|_+ \\
&- \tfrac{1}{4} [(\text{D}_{A} J^D_F ) (\text{D}_{B}\text{J}_{C}^F)
+ J^D_F\text{R}_{AB}{}^{\text{ } F}{}_{\text{}K}\text{J}_{C}^K
+ J^D_F\text{R}_{CB}J^F_{A}]\eta^B\eta^C\mathcal{P}_D^\alpha|_+ dx_\alpha \tag*{} \end{align}
\end{boxedminipage}
\begin{proof}
The first three equations are obtained from those of Theorem \ref{T3} by direct substitution from equation (\ref{E4}). The final equation follows by substitution and by identity (\ref{id13}) from Appendix II.\end{proof}
{\rmk $\chi^i$ is to be identified with the anticommuting parameter introduced by Witten. We now see that the commuting field H$^{i\alpha}$ is the multimomenta p$_i^\alpha$ with one index raised, and the anticommuting field $\varrho^\alpha_i$ is the ghost multimomenta $\mathcal{P}^\alpha_i$. }
{ \rmk The extra {\it ``...non-covariant looking terms...''} which Witten \cite{TSM} added to his transformations do not occur here because we know the \(p_i^\alpha|_{+}\) and the \(\mathcal{P}_A^\alpha|_{+}\) are geometrically local fibre coordinates on an affine bundle. They are therefore expected to transform affinely under reparametrisations of the \(u^i\). We have therefore achieved the aim of this section and given the first derivation of this BRST algebra in any Hamiltonian formulation of BRST.}
\newpage
\section*{Appendix I: Vielbein Identities}
We shall have use of three sets of vielbeins:
\begin{itemize}
\item Real analytic anholonomic basis vielbeins on M.
\item J-pseudoholomorphic anholonomic basis vielbeins on M.
\item Complex analytic holonomic basis vielbeins on $\Sigma$.
\end{itemize}
We collect together here the identities satisfied by these vielbeins.
We denote the indices of the anholonomic orthonormal vielbein basis (for more details see \cite{Nakahara}) on M by upper case Latin letters: $A,B,\cdots=1,\cdots,m$ and the coordinate basis by lower case middle Latin letters: $i,j=1,\cdots,m$. They satisfy the following identities:
{\hspace{5pt}}\newline
\begin{boxedminipage}[t]{12.75cm}
{ \scshape Real analytic anholonomic basis vielbeins on M:}
\begin{alignat}{2}
\text{E}_A^i\text{E}_i^B &= \delta^B_A &\qquad\text{E}_A^i\text{E}_j^A &= \delta^i_j \tag{RA1}\\
\Hat{\theta}^A &= \text{E}_j^A du^j &\qquad \Hat{\theta}_A &=\text{E}_A^i\frac{\partial}{\partial u^i}\tag{RA2} \\
g(\Hat{\theta}_A,\Hat{\theta}_B):&=\delta_{AB} &\qquad g_{ij} &=\text{E}_A^i\text{E}_B^j\delta_{AB} \tag{RA3}
\end{alignat}
\end{boxedminipage}
{\hspace{5pt}}\newline
The following characteristic of almost complex manifolds may be found in in \cite{Hsuing}:
{\thm A necessary and sufficient condition for a differentiable 2k-dimensional (2k=m) manifold $M^{\text{2k}}$ to admit an almost complex structure is that on $M^{\text{2k}}$ there exist two distributions $\Pi_{k}$ and $\Bar{\Pi}_{k}$ of complex dimension k, which are conjugate to each other, have no common direction and span a linear space of dimension 2n.}
So by this theorem let \(( \text{E}^i_a )_{a=1,\cdots,k}\) span $\Pi_{k}$ and \( (\text{E} ^i_{a*} )_{a*=1,\cdots,k}\) span $\Bar{\Pi}_{k}$. Since the 2n vectors \( ( \text{E}^i_a, \text{E}^i_{a*} )_{a,a*=1,\cdots,k}\) are also linearly independent, the matrix \( ( \text{E}^i_a, \text{E}^j_{a*} )\) has an inverse which is denoted by \( ( \text{E}_i^a, \text{E}_j^{a*} )\).
Then the following hold:
{\hspace{5pt}}\newline
\begin{boxedminipage}[t]{12.75cm}
{ \scshape J-pseudoholomorphic anholonomic basis vielbeins on M:}
\begin{alignat}{2}
\text{E}_{a}^i\text{E}_i^{b} &= \delta^{a}_{b} &\qquad \text{E}_{a*}^i\text{E}_i^{b*} &= \delta^{a*}_{b*}\tag{JA1}\\
\text{E}_{a}^i\text{E}_i^{b*} &= 0 &\qquad \text{E}_{a*}^i\text{E}_i^{b} &= 0\tag{JA2}\\
\text{E}_{a}^i\text{E}_j^{a} + \text{E}_{a*}^i\text{E}_j^{a*} &=\delta^i_j
\tag{JA3}\\
\Hat{\theta}^a &= \text{E}_j^a du^j &\qquad \Hat{\theta}^{a*} &= \text{E}_j^{a*} du^j \tag{JA4}\\
\Hat{\theta}_{a} &=\text{E}_{a}^i\frac{\partial}{\partial u^i} &\qquad
\Hat{\theta}_{a*} &=\text{E}_{a*}^i\frac{\partial}{\partial u^i}\tag{JA5}
\end{alignat}
\end{boxedminipage}
For more details see \cite{Hsuing,Goldberg}.
\newpage
Since the almost complex structure on $\Sigma$ is actually integrable we have the following holonomic vielbeins corresponding to the complex coordinate basis.
{\hspace{5pt}}\newline
\begin{boxedminipage}[t]{12.75cm}
{ \scshape Complex analytic holonomic basis vielbeins on $\Sigma$:}
\begin{alignat}{2}
\text{E}_{\kappa}^\alpha\text{E}_\alpha^{\lambda} &= \delta_{\kappa}^{\lambda} &\qquad \text{E}_{\kappa*}^\alpha\text{E}_\alpha^{\lambda*} &= \delta^{\lambda*}_{\kappa*}\tag{CH1}\\
\text{E}_{\kappa}^\alpha\text{E}_\alpha^{\lambda*} &= 0 &\qquad \text{E}_{\kappa*}^\alpha\text{E}_\alpha^{\lambda} &= 0\tag{CH2}\\
\text{E}_{\kappa}^\alpha\text{E}_\beta^{\kappa} + \text{E}_{\kappa*}^\alpha\text{E}_\beta^{\kappa*} &=\delta^\alpha_\beta \tag{CH3}\\
\Hat{\theta}^\kappa &= \text{E}_\alpha^\kappa dx^\alpha=dz^\kappa &\qquad \Hat{\theta}^{\kappa*} &= \text{E}_\alpha^{\kappa*} du^\alpha=dz^{\kappa*}\tag{CH4} \\
\Hat{\theta}_{\kappa} &=\text{E}_{\kappa}^\alpha\frac{\partial}{\partial x^\alpha}= \frac{\partial}{\partial z^\kappa} &\qquad
\Hat{\theta}_{\kappa*} &=\text{E}_{\kappa*}^\alpha\frac{\partial}{\partial x^\alpha}=\frac{\partial}{\partial z^{\kappa*} }\tag{CH5}
\end{alignat}
\vspace{5pt}
\end{boxedminipage}
\vspace{10pt}
We shall also need to make use of the following ``combination vielbeins'' on M,
defined by $\text{E}_{a}^A:=\text{E}_{a}^i\text{E}_{i}^A$ and satisfying the following identities:
{\hspace{5pt}}\newline
\begin{boxedminipage}[t]{12.75cm}
{ \scshape ``combination'' anholonomic basis vielbeins on M:}
\begin{alignat}{2}
\text{E}_{a}^A\text{E}_A^{b} &= \delta^{a}_{b} &\qquad \text{E}_{a*}^A \text{E}_A^{b*} &= \delta^{a*}_{b*}\tag{CBA1}\\
\text{E}_{a}^A\text{E}_A^{b*} &= 0 &\qquad \text{E}_{a*}^A \text{E}_A^{b} &=0\tag{CBA2} \\
\text{E}_{a}^A\text{E}_B^{a} + \text{E}_{a*}^A\text{E}_B^{a*} &=\delta^A_B \tag{CBA3}
\end{alignat}
\end{boxedminipage}
\newpage
\section*{Appendix II:On almost K\"ahler manifolds}
We shall give a brief review and some useful geometric identities satisfied by of almost K\"ahler structures. Almost K\"ahler manifolds are one class of a richly structured interrelated hierarchy of manifolds. Only the K\"ahler class has drawn much attention from the theoretical physics community, via the pre-eminence of Calabi-Yau manifolds in string theory\footnote{Although the almost K\"ahler class, in the form of symplectic manifolds, is the axiomatic foundation of classical mechanics and plays a indispensable role in geometric quantisation.}. This hierarchy\footnote{For more details see \cite{Hsuing}. } may be depicted graphically as in Figure \ref{K}:
\begin{figure}[h]
\[
\xymatrix{ & \mathcal{AK} \ar@0[dr]^{{\subset}}\\
\mathcal{K} \ar@0[ur]^\subset \ar@0[dr]_\subset& &\mathcal{QK} \hspace{10pt} \subset & \mathcal{ASK}\hspace{10pt} \subset & \mathcal{AH} \\
& \mathcal{NK} \ar@0[ur]_\subset
}
\]\caption{The hierarchy of almost K\"ahler manifolds.}\label{K}
\end{figure}
\index{almost Hermitian manifold}\index{almost semi- K\"ahler manifolds manifold}\index{quasi-K\"ahler manifold}\index{nearly K\"ahler manifold}\index{almost K\"ahler manifold}\index{K\"ahler manifold} where \(\mathcal{K}\) denotes K\"ahler, \(\mathcal{AK}\) denotes almost K\"ahler, \(\mathcal{NK}\) denotes nearly K\"ahler, \(\mathcal{QK}\) denotes quasi-K\"ahler, \(\mathcal{ASK}\) denotes almost semi-K\"ahler and \(\mathcal{AH}\) denotes almost Hermitian.
All almost complex manifolds admit a Hermitian metric, and they are thus all {\it almost Hermitian}. The next most general class are the {\it almost semi- K\"ahler manifolds}. They are defined to be those almost Hermitian manifolds whose almost complex structure obeys the identity:
\begin{equation}
\text{J}_i:=-\text{D}_k\text{J}_i^k=0\label{id1} \tag{A.1}
\end{equation}
One then obtains the {\it quasi-K\"ahler} manifolds by imposing the identity:
\begin{equation}
\text{D}_k\text{J}_i^j +\text{J}_r^k\text{J}_s^i\text{D}_r\text{J}_s^j=0\label{id2}\tag{A.2}
\end{equation}
At this point the classification bifurcates into two classes. The { \it almost K\"ahler} class was described in the introduction. The {\it nearly K\"ahler} class is obtained from the quasi-K\"ahler class by imposing the identity \(\text{D}_{[k}\text{J}_{i]}^l=0\). Finally both the almost K\"ahler and nearly K\"ahler classes coalesce to the ``integrable'' {\it K\"ahler} case upon certain extra conditions upon their respective almost complex structures.
Having described the wider context in which almost K\"ahler manifolds fit we now detail some identities which are satisfied by the corresponding almost complex structure. Knowledge of the wider context will prove to be useful in proving the various identities which follow.
As a consequence of the fact that the almost complex structure is an anti-idempotent endomorphism of the tangent bundle, viz \( \text{J}^m_n \text{J}^n_l=-\delta^m_l \), one finds that:
\begin{equation}
\text{J}^m_n \text{D}_k\text{J}^n_l =- \text{J}^n_l \text{D}_k \text{J}^m_n
\text{ or equivalently, } \text{J}^m_n \partial_k\text{J}^n_l =- \text{J}^n_l \partial_k \text{J}^m_n \label{id3}\tag{A.3}
\end{equation}
Now consider the tensor \( \text{C}_{mij}:=\tfrac{1}{2}\text{J}_{mn} \text{D}_{[i} \text{J}_{j]}^n \) which plays the role of structure functions for the non-Abelian deformation algebra.
\begin{equation}
\begin{split}
\text{C}_{mij}&:=\tfrac{1}{2}\text{J}_{mn} \text{D}_{[i} \text{J}_{j]}^n \\
&=- \tfrac{1}{2}(\text{D}_{[i|}\text{J}_{mn}) \text{J}_{|j]}^n \\
&=- \tfrac{1}{2} (\text{D}_{[i|} \text{J}_{m}^t) \text{J}_{|j]t} \\
&=- \text{C}_{jim}
\end{split}\tag{A.4}
\end{equation}
The tensor \( \text{C}_{mij} \) is therefore antisymmetric in all three indices.
By virtue of the quasi-K\"ahler identity (\ref{id2}) we have:
\begin{equation}
\begin{split}
\text{C}_{mkl}&=-\tfrac{1}{2}\text{J}^n_m\text{J}^r_{[k|} \text{J}^s_n \text{D}_r\text{J}_{s|l]} \\
&=\tfrac{1}{2}\text{J}^r_{[k|}\text{D}_r\text{J}_{m|l]} \\
&=-\tfrac{1}{2}\text{J}^n_{m}\text{D}_n\text{J}_{[kl]} \\
\Longleftrightarrow 0&=\text{J}_{m}^n \text{D}_{([k|} \text{J}_{l]|n)}
\end{split}\label{id5}\tag{A.5}
\end{equation}
We now demonstrate that the covariant derivative of the structure functions satisfies an antisymmetry. More precisely:
\begin{equation}
\text{D}_q\text{C}_{mkl}=:\tfrac{1}{2}\text{D}_q\{
J^n_m\text{D}_{[k|}\text{J}_{n|l]}
\}=\tfrac{1}{2}\text{D}_{[q|}\{ J^n_m\text{D}_{[|k]|}\text{J}_{n|l]}
\} \label{id6}\tag{A.6}
\end{equation}
By virtue of the Leibniz rule the covariant derivative of the structure functions splits into two parts:
\begin{equation}
\text{D}_q\text{C}_{mkl}= \tfrac{1}{2}(\text{D}_q J^n_m ) (\text{D}_{[k|}\text{J}_{n|l]})
+ \tfrac{1}{2}J^n_m(\text{D}_q\text{D}_{[k|}\text{J}_{n|l]} )\label{id7}\tag{A.7}
\end{equation}
We shall establish the antisymmetry of each of these terms and thus of the whole. Consider the first term:
\begin{equation}
\begin{split}
(\text{D}_q J^n_m ) (\text{D}_{[k|}\text{J}_{n|l]})&=
-(\text{D}_q J^n_{[l} ) (\text{D}_{k]}\text{J}_{nm})\\
&=(\text{D}_q J_{[l|n} ) (\text{D}_{|k]}\text{J}_{m}^n) \\
&=-(\text{D}_q J_{n[l} ) (\text{D}_{k]}\text{J}_{m}^n) \\
&= - (\text{D}_{[k|}\text{J}_{m}^n)(\text{D}_q J_{n|l]} )\label{id8}
\end{split}\tag{A.8}
\end{equation}
where we have used the complete antisymmetry of the structure functions and have carefully raised and lowered using the covariantly constant metric on M. Now we consider the second term, the antisymmetry of which will follow by demonstrating that the following is true:
\begin{equation}
J^n_m\text{D}_{(q}\text{D}_{[|k)|}\text{J}_{n|l]}=0\label{id9}\tag{A.9}
\end{equation}
To prove this consider the following:
\begin{equation}
\begin{split}
J^n_m\text{D}_{(q}\text{D}_{[k)|}\text{J}_{n|l]}
&=- J^n_m\text{D}_{(q}\text{D}_{[l|}\text{J}_{n||k])} \\
&=- J^n_m\text{D}_{([k|}\text{D}_{l]}\text{J}_{n|q)} \\
&= J^n_m\text{D}_{([l|}\text{D}_{k]}\text{J}_{n|q)} \\
&=- J^n_m\text{D}_{[l|}\text{D}_{n}\text{J}_{|(k]q)} \\
&= 0 \label{id10}
\end{split}\tag{A.10}
\end{equation}
where the second from last equality follows from (\ref{id5}). We therefore conclude that \( \text{D}_q\text{C}_{mkl}\) is antisymmetric in q and k as required, i.e. \( \text{D}_q\text{C}_{mkl}\) is completely antisymmetric. This also holds in the real anholonomic basis so that \( \text{D}_A\text{C}_{BCD}\) is completely antisymmetric. It follows from the complete antisymmetry of \( \text{D}_A\text{C}_{BCD}\) that:
\begin{equation}
\text{D}_A\text{C}_{BCD}=\Hat{\theta}_A\text{C}_{BCD}\tag{A.11}\label{id11}
\end{equation}
and since the metric in the anholonomic basis is given by \(g_{AB}=\delta_{AB}\)one finds that:
\(
\Hat{\theta}_A\text{C}_{BCD}=(\Hat{\theta}_A\text{C}_{CD}^E)\delta_{EB}
\).
From which it follows:
\begin{equation}
\Hat{\theta}_A\text{C}_{CD}^E=\text{D}_A\text{C}_{CD}^E\tag{A.12}\label{id12}
\end{equation}
By virtue of the Ricci identity, see \cite{Hsuing}, one has \(\text{J}^m_n[\text{D}_q,\text{D}_k]J^n_l=\text{J}^m_n \text{R}_{qk}{}^{\text{ } n}{}_{\text{}r}\text{J}^r_l+ \text{J}^m_n\text{R}_{lk}J^n_q\). We then find (in the anholonomic basis):
\begin{equation}
\text{D}_A\text{C}_{BC}^D=(\text{D}_{A} J^D_F ) (\text{D}_{B}\text{J}_{C}^F)
+ J^D_F\text{R}_{AB}{}^{\text{ } F}{}_{\text{}K}\text{J}_{C}^K
+ J^D_F\text{R}_{CB}J^F_{A}\tag{A.13}\label{id13}
\end{equation}
|
\section{Introduction}
\setcounter{equation}{0}
Anti de Sitter spaces (AdS) were left aside initially in General Relativity in detriment of
de Sitter (dS) spaces due to the (dubious) interpretation of the negative
cosmological constant that supports them.
A posteriori were relived in the context of Supergravity as possible vacuum
solutions in compactification mechanisms \cite{peter} and the study of black hole
thermodynamics on them \cite{hp}.
Emphasis was certainly put in $(1,3)$ signatures and euclidean
versions in the quantum gravity context.
However during the past year a huge amount of work rounding AdS spaces in different
dimensions was made in relation with Maldacena's conjectures relating on-shell quantum
gravity theories on $AdS_{p+1}$ with off-shell conformal field theories (CFT)
that live on its $p$ dimensional boundary \cite{malda1}.
The reason behind this relation can be heuristically found in the fact \cite{malda1},
\cite{w1} that any theory of gravity with a vacuum value of the metric tensor given by
$d+1$-dimensional $AdS$ space, having in mind the existence of a time-like boundary, should
depend on boundary conditions defined on the $d$ dimensional boundary manifold and should
realize the isometry group of this space that becomes automatically the conformal group
acting on the boundary.
So it seems natural that the degrees of freedom living on the boundary should realize
this symmetry.
Of course things are not so easy, the growing number of people working on the field
extended in many directions a representation of this fact, neither obvious \cite{ps}.
It is the aim of this paper to clarify several aspects of this relation from
the purely mathematical point of view, particularly the group actions on this
spaces and their relation with the corresponding ones on the boundaries, and to
study geodesics on them, all in the general case of $(d_-,d_+)$ arbitrary signature
(since now on $AdS_{d_-,d_+}$), generalizing in some cases results known in cases
of earlier physical interest \cite{he}.
\section{Review of $AdS_{d_-,d_+}$ spaces}
\setcounter{equation}{0}
Several definitions of $AdS_{d_-,d_+}$ spaces are possible (in Section 5 we will introduce
another different to that given here useful for our purposes there) and many
parametrizations of them were introduced in the literature, each one useful in particular
contexts.
For sake of completeness and later use we believe pertinent to give a brief summary of them.
\subsection{The defining coordinates}
Let us consider cartesian coordinates $\;\{ w^M , M = -d_-, \ldots,0,\ldots,d_+ \}\;$ in
$\Re^{1+ d_- + d_+}$ with $\;d_\pm \geq 0,$
and endows it with the minkowskian metric
\begin{equation}}\newcommand{\ee}{\end{equation}
\eta_{d_- +1,d_+} = \eta_{MN}\; dw^M\; dw^N\;\; , \;
\eta \equiv \matriz{-1_{d_- + 1}}{0}{0}{1_{d_+}}\label{gmin}
\ee
This is a maximally symmetric space (MSS) with the pseudo-Poincar\`e isometry
group $ISO(d_- + 1,d_+)$.
The $AdS_{d_-,d_+}$ space is introduced as the submanifold defined by the constraint
\begin{equation}}\newcommand{\ee}{\end{equation}
\eta_{MN}\; w^M\; w^N = -r_{0}{}^2 \;\;, \;\; r_{0}{}>0\label{cons}
\ee
where $r_{0}{}$ is the radius (curvature scale) of the space, with the
metric $G$ induced by \eqn{gmin}.
Clearly the constraint respects only $SO(d_- + 1, d_+)$ invariance and then it
becomes the isometry group of $AdS_{d_-,d_+}$; reduction of the dimensionality combines with
the reduction of symmetry to make it also a MSS.
As such its curvature tensor verifies
\footnote{We adopt the conventions of \cite{wald}; also the convention of rising and lowering
indices with the corresponding (from the context) $\eta$-matrix is implicit all along the paper.
}
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
{\cal R}_{abcd} &=& r_{0}{}^{-2}\; ( G_{ad}\; G_{bc} - G_{ac}\; G_{bd} )\cr
{\cal R}_{ab} &=& \frac{2}{{d_{-}} +{d_{+}} -2}\; \lambda\; G_{ab}\cr
{\cal R} &=& -\;r_{0}{}^{-2}\; (d_- + d_+ -1 ) \; (d_- + d_+ )
\eea
where $\; \lambda = -\; (2r_{0}{}^2 )^{-1}\; (d_- + d_+ -1 )\;(d_- + d_+ -2 )\;$ is the
cosmological constant.
In this defining coordinates we can solve for one of the coordinates, i.e.
\begin{equation}}\newcommand{\ee}{\end{equation}
w^{-d_-} = \pm (w^\mu w_\mu + r_{0}{}^2 )^{\frac{1}{2}} \label{w-}
\ee
and parametrize (each sheet of) $AdS_{d_-,d_+}$ by
\begin{equation}}\newcommand{\ee}{\end{equation}
\{ w^\mu , \;\;\mu = -d_- +1, \ldots,0,\ldots,d_+\} = \left(\begin{array}{c}
{\vec w}_{-} \\ {\vec w}_{+} \end{array}\right)\;\in\Re^{d_-}\times\Re^{d_+}\label{dg}
\ee
The metric takes the form
\begin{equation}}\newcommand{\ee}{\end{equation}
G_{d_-,d_+} = \left(\eta_{\mu\nu} - \frac{w_\mu\; w_\nu}{w^\rho\;w_\rho +
r_{0}{}^2}\right)\; dw^\mu\; dw^\nu\label{gdc}
\ee
In the euclidean case $d_-=0$ the topology corresponds to a two-sheeted hyperboloid with
basis in $w^{0} = \pmr_{0}{} $; we will consider the connected component given by the $+$
sign in $\eqn{w-}$, the sheet based on $w^0 = +r_{0}{}$ with naive topology $ \Re^{d_+}$.
If instead $d_+ = 0$ the topology is compact corresponding to a $S^{d_-}$ of radius
$r_{0}{}$ and ``wrong" sign metric.
In the general case $d_\pm >0$ the topology is $S^{d_-}\times \Re^{d_+}$
corresponding to a hyperboloid, the (non) compact directions being
(space) time like.
It is important to note that the solution of \eqn{w-} imposes the
condition
\begin{equation}}\newcommand{\ee}{\end{equation}
w^\mu \; w_\mu + r_{0}{}^2 > 0 \label{conboundc}
\ee
\subsection{The plane coordinates}
They are defined by the following one-to-one map
\begin{equation}}\newcommand{\ee}{\end{equation}
z^\mu = \frac{w^\mu}{1+ (1+\frac{w^\rho w_\rho}{r_{0}{}^2} )^\frac{1}{2}}\;
\longleftrightarrow\; w^\mu = \frac{2\; z^\mu}{1 - \frac{z^\rho z_\rho}{r_{0}{}^2}} \label{plc}
\ee
In terms of them the metric takes the simple diagonal form where the signature is
manifest,
\begin{equation}}\newcommand{\ee}{\end{equation}
G_{d_-,d_+} = \frac{4}{(1 - \frac{z^\rho z_\rho}{r_{0}{}^2} )^2} \;
\eta_{\mu\nu}\; dz^\mu\; dz^\nu
\ee
The definition \eqn{conboundc} togheter with \eqn{plc} is then equivalent to
\begin{equation}}\newcommand{\ee}{\end{equation}
-r_{0}{}^2 < z^\mu \; z_\mu < r_{0}{}^2 \label{conbounpc}
\ee
Let us remember that unless ${d_{-}} =0$ these coordinates cover only
one-half of $AdS_{d_-,d_+}$; in fact from the relations
\begin{equation}}\newcommand{\ee}{\end{equation}
z^\rho\; z_\rho = \frac{w^\rho\; w_\rho}{\left( 1 +
(1 + \frac{w^\rho\; w_\rho}{r_{0}{}^2})^{\frac{1}{2}}\right)^2 }\;\; \longleftrightarrow \;\;
w^\rho\; w_\rho = \frac{4\;z^\rho\; z_\rho}{\left( 1 - \frac{z^\rho\; z_\rho}{r_{0}{}^2} \right)^2}
\ee
we can have
\begin{equation}}\newcommand{\ee}{\end{equation}
w^{-{d_{-}}} = \pm\;r_{0}{}\;\frac{r_{0}{}^2 + z^\rho\; z_\rho}{r_{0}{}^2 -z^\rho\; z_\rho }
\ee
corresponding to the $\pm$ choice in \eqn{w-}.
\subsection{The global coordinates}
Given the constraint \eqn{cons} it is natural to introduce in our context hyperbolic
coordinates covering the sector of $\; w^M w_M < 0\;$ of ${\cal M}_{d_- +1,d_+}$ space as follows,
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\left(\begin{array}{l} w^{-{d_{-}}}\\ {\vec w}_{-} \end{array}\right) &=& r\;\cosh r_L\; {\check n}_{-} \cr
{\vec w}_{+} &=& r\;\sinh r_L\; {\check n}_{+} \label{rc}
\eea
where $r>0 , r_L \geq 0,$ and ${{\check n}_\pm}^t{{\check n}_\pm} = 1$,
so ${\check n}_{-} , {\check n}_{+}$ define points in $S^{d_{-}}, S^{{d_{+}} -1}$ respectively (the case ${d_{+}} = 0$ is
handled by putting $r_L\equiv 0)$).
Obviously in this coordinates the constraint \eqn{cons} is
equivalent to fix $r = r_{0}{} $; the metric is written as
\begin{equation}}\newcommand{\ee}{\end{equation}
G_{d_-,d_+} = r_{0}{}^2\; \left( - \cosh^2 r_L\; d^2 \Omega_{{d_{-}}} + d^2 r_L +
\sinh^2 r_L\; d^2 \Omega_{{d_{+}} -1} \right)\label{grc}
\ee
where $d^2 \Omega_{p}$ stands for the standard measure of (radius $1$) $S^p$.
The signature as well as the topology is manifest in this form, with ${\check n}_{-}$ and
$(r_L,{\check n}_{+} )$ parametrizing $S^{{d_{-}}}$ and $\Re^{{d_{+}}}$ respectively.
The $``+"$ ($``-"$) sign in \eqn{w-} corresponds to consider the semi-sphere of $S^{d_{-}}$
containing the north (south) pole.
It is common in general relativity contexts \cite{hp} to introduce the radial variable
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
r_H &=& \sinh r_L \;\;\; , \;\; r_H\geq 0\cr
G_{d_-,d_+} &=& r_{0}{}^2\; \left( - (1 + r_H{}^2 ) \; d^2 \Omega_{{d_{-}}} + \frac{d^2 r_H}{1 + r_H{}^2 }
+ r_H{}^2\; d^2 \Omega_{{d_{+}} -1}\right)\label{ghc}\label{hc}
\eea
where the near horizon geometry $r_H \sim 0$ is manifestly minkowskian.
Another variant is to introduce
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
r_H &=& \tan \rho \;\; ,\;\; 0 \leq \rho <\frac{\pi}{2} \cr
G_{d_-,d_+} &=& r_{0}{}^2\; \sec^2\rho\; \left( - d^2 \Omega_{{d_{-}}} + d^2\rho + \sin^2\rho\; d^2
\Omega_{{d_{+}} -1}\right)\label{grhoc}\label{rhodef}
\eea
which shows that $AdS_{d_-,d_+}$ is conformal to the geometry product of a time-like $S^{d_{-}}$ and a
``trumpet", the well known (for $(1,3)$ signature) Einstein static universe (for exactness,
one half of it \cite{he}).
\subsection{The Poincar\`e coordinates}
In ${\cal M}_{d_- +1,d_+}$ are introduced in terms of $(w^M )$ the following coordinates ($d_+ >0$)
\begin{equation}}\newcommand{\ee}{\end{equation}
\left\{
\begin{array}{l} x^\alpha = r_{0}{}\; (w^{-{d_{-}}} + w^{d_{+}} )^{-1}\; w^\alpha \\
x^+ = r_{0}{}^2 \;(w^{-{d_{-}}} + w^{d_{+}} )^{-1} \\
\;V = r_{0}{}^{-2} \;(w^{-{d_{-}}} - w^{d_{+}} )\end{array}\right.
\;\;\longleftrightarrow\;\;
\left\{ \begin{array}{l} w^{-{d_{-}}} = \frac{r_{0}{}^2}{2}\; ( (x^+)^{-1} + V)\\
\;\;\; w^\alpha = r_{0}{}\; (x^+)^{-1}\; x^\alpha\\
\;\; w^{{d_{+}}} = \frac{r_{0}{}^2}{2}\; ( (x^+)^{-1} - V)\end{array}\right.\label{pc}
\ee
where $\;\alpha = -{d_{-}} +1, \ldots, 0, \ldots, {d_{+}} -1\;$. The constraint \eqn{cons} looks
\begin{equation}}\newcommand{\ee}{\end{equation}
-r_{0}{}^{2}\; (x^+)^{-1}\; V + (x^+)^{-2}\; \eta_{\alpha\beta}\; x^\alpha\; x^\beta + 1 = 0
\label{conspc}
\ee
and solving for $V$ the metric reads in Poincar\`e parametrization ($ x^+ \equiv x^{d_{+}}$)
\begin{equation}}\newcommand{\ee}{\end{equation}
G_{d_-,d_+} = \frac{r_{0}{}^2}{(x^+)^2} \; \eta_{\mu\nu}\; dx^\mu\;dx^\nu
\label{gwc}
\ee
These coordinates were considered in \cite{w1} to do computations in the boundary theory;
the reason of this fact should be clear in Section 3.
A little modification of this system of coordinates, namely
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
U &=& (x^+ )^{-1}\cr
G_{d_-,d_+} &=& r_{0}{}^2 \; \left(\frac{d^2 U}{U^{2}} + U^2\; \eta_{\alpha\beta}\;
dx^\alpha\;dx^\beta\right)\label{gpc}
\eea
was introduced in \cite{malda1} where it appears in a natural way coming from the usual
parametrization of $p$-brane geometries in the decoupling limit.
\subsection{The boundary of $AdS_{d_-,d_+}$ spaces}
For ${d_{+}} >0$ we saw that $AdS_{d_-,d_+}$ is a non compact space.
It is possible however to adscribe to it a boundary through Penrose's concept of conformal
infinity.
The rough idea (see \cite{he} for exact statements) consists in trying to map the space-time
at hand to a (possible finite) region in which null curves are at $\pi/4$, and in such a way
that both geometries are Weyl-related.
Then causality concepts will coincide in both spaces and we should be able to identify
the conformal boundaries where any curve should born and die.
In our case we have the job made by considering the form \eqn{grhoc} of the metric; from there
it follows that the boundary corresponds to the hypersurface defined by $\rho=\pi/2$,
topologically homeomorphic to $\;S^{d_{-}}\times S^{{d_{+}} -1}$.
\footnote{
When ${d_{+}} = 1$ the boundary consists of two disconnected $S^{d_{-}}$ ( $S^0 \sim Z_2$).
}
It is time-like in character; when ${d_{-}} =1$ this fact is responsable among other things
for the lack of global hyperbolicity (if ${d_{-}} >1$ clearly the concept of a Cauchy surface
is not obvious).
Note also that this definition is the same as to consider the $r_L=\infty$ limiting surface, the
same way in spirit in which the boundary was introduced in \cite{w1}.
In plane coordinates it corresponds to the hypersurface
\begin{equation}}\newcommand{\ee}{\end{equation}
z^\rho \; z_\rho = r_0^2
\ee
In Poincar\`e coordinates it is sitted at $x^+ = 0$ ($U= \infty$); however some points
``at infinity" should be added to the space parametrized by $(x^\alpha)$ to get the
conformal compactification of ${\cal M}_{{d_{-}},{d_{+}}-1}$, i.e. $\;S^{d_{-}}\times S^{{d_{+}} -1}$.
Of course this fact a bit obscure in this coordinatization follows from the relations
\eqn{pc} and becomes evident from the action of $SO({d_{-}} +1, {d_{+}})$ on the boundary given
in the next section.
\section{The $SO({d_{-}} +1,{d_{+}} )$ action}
\setcounter{equation}{0}
The action of the isometry group on $AdS_{d_-,d_+}$ is that induced by the
linear action on ${\cal M}_{d_- +1,d_+}$.
Let $\Lambda_{{d_{-}} +1,{d_{+}}} \in SO({d_{-}}+1,{d_{+}} )$; it admits a coset decomposition under its
maximal compact subgroup $SO({d_{-}} +1)\times SO({d_{+}})$ in the form \cite{gil}
\begin{eqnarray}
\Lambda_{{d_{-}} +1,{d_{+}} }(S,P, Q) &=& K_{{d_{-}} +1,{d_{+}}}(S)\; H(P,Q) \cr
K_{{d_{-}} +1,{d_{+}}}(S) &=& \matriz{(1 + S S^{t})^{\frac{1}{2}}}{S}{S^{t}}{(1 + S^{t}S)^{\frac{1}{2}}}
= \exp\matriz{0}{N}{N^t}{0} \cr
S &\equiv& \frac{ \sinh (N\; N^t)^\frac{1}{2}}{(N\; N^t)^\frac{1}{2}}\;
N\;\; \in \Re^{({d_{-}} +1 )\times{d_{+}}}\cr
H(P,Q ) &=& \matriz{P}{0}{0}{Q} \;\; , \; P\in SO({d_{-}} +1)\; ,\; Q\in SO({d_{+}}) \label{parortpq}
\end{eqnarray}
The linear action on $(w^M)\equiv (w^{-{d_{-}}}, {\vec w}_{-} ,{\vec w}_{+} ) \in {\cal M}_{d_- +1,d_+}$ is then given by
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\left( \begin{array}{l} {}^\Lambda w^{-{d_{-}}} \\ {}^\Lambda{\vec w}_{-} \end{array}\right)
&=&(1 + S S^{t})^{\frac{1}{2}}\; P\;\left( \begin{array}{l} w^{-{d_{-}}} \\ {\vec w}_{-} \end{array}\right)
+ S\; Q\; {\vec w}_{+}\cr {}^\Lambda{\vec w}_{+} &=& S^t\; P\; \left( \begin{array}{l} w^{-{d_{-}}} \\ {\vec w}_{-}
\end{array}\right) + (1 + S^{t}S)^{\frac{1}{2}}\;
Q\;{\vec w}_{+}\label{lintran}
\eea
By restricting ourselves to the submanifold \eqn{cons} and using
\eqn{w-} we get the non linear realization searched for.
This is easier acomplished in the global coordinates $( r_L , {\check n}_{-} , {\check n}_{+} )$ of
\eqn{rc}; in terms of them the transformations are given by
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\sinh {}^\Lambda r_L &=& |\cosh r_L \; S^t\; P\; {\check n}_{-} + \sinh r_L \;(1 + S^{t} S)^{\frac{1}{2}}\;
Q \; {\check n}_{+} | \cr
{}^\Lambda{\check n}_{-} &=&
\frac{\cosh r_L (1 + S S^{t})^{\frac{1}{2}}\; P\;{\check n}_{-} + \sinh r_L \; S\; Q\; {\check n}_{+}}
{| \cosh r_L (1 + S S^{t})^{\frac{1}{2}}\; P\;{\check n}_{-} + \sinh r_L \; S\; Q\; {\check n}_{+}|}\cr
{}^\Lambda{\check n}_{+} &=&
\frac{\cosh r_L \; S^t\; P\; {\check n}_{-} +\sinh r_L (1 + S^{t}S)^{\frac{1}{2}}\; Q\;{\check n}_{+} }
{|\cosh r_L \; S^t\; P\; {\check n}_{-} +\sinh r_L (1 + S^{t}S)^{\frac{1}{2}}\; Q\;{\check n}_{+} |}
\label{nltgc}
\eea
From here it is clear that the boundary results invariant under the $SO({d_{-}} +1,{d_{+}})$
transformations; in fact $r_L = \infty$ is mapped to ${}^\Lambda r_L = \infty$ while the
variables in the boundary manifold realize the isometry in the way
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
{}^\Lambda{\check n}_{-} & \stackrel{r_L=\infty}{=}&
\frac{(1 + S S^{t})^{\frac{1}{2}}\; P\;{\check n}_{-} + S\; Q\; {\check n}_{+}}
{| (1 + S S^{t})^{\frac{1}{2}}\; P\;{\check n}_{-} + S\; Q\; {\check n}_{+}|}\cr
{}^\Lambda{\check n}_{+} &\stackrel{r_L=\infty}{=}&
\frac{ S^t\; P\; {\check n}_{-} +(1 + S^{t}S)^{\frac{1}{2}}\; Q\;{\check n}_{+} }
{| S^t\; P\; {\check n}_{-} +(1 + S^{t}S)^{\frac{1}{2}}\; Q\;{\check n}_{+} |}
\eea
These expressions provide a global, non linear realization of generalized conformal
transformations.
This realization is made not on ${\cal M}_{{d_{-}},{d_{+}}-1}$ but on the conformal compactification of
it, i.e. $S^{d_{-}}\times S^{{d_{+}} -1} $.
For example in the euclidean case ${d_{-}} = 0$ there is no ${\check n}_{-}$ and the boundary
$S^{{d_{+}} -1}$ is parametrized by ${\check n}_{+}\equiv\vec n$; with $S^t\equiv {\vec S}$ we have
\begin{equation}}\newcommand{\ee}{\end{equation}
{}^\Lambda\vec n =
\frac{ Q\; \vec n + \left( (1 + |{\vec S}|^2)^{\frac{1}{2}} - 1\right)\; {\check S}^t\; Q\;\vec n\;
{\check S} }{\left( 1 + ({\vec S}^t\; Q\;\vec n)^2 \right)^\frac{1}{2}}
\ee
The (linear) transformations ${\vec S}= {\vec 0}$ with $Q\in SO({d_{+}})$ are
just the isometries of $S^{{d_{+}} -1}$ and those with $Q=1$ corresponds
to scale and special conformal transformations parametrized by
${\vec S}\in\Re^{d_{+}}$; both generate the conformal group $SO(1,{d_{+}} )$.
But although compact and clear in meaning they are not enough useful for most current applications.
For this we move to get the realization in Poincar\`e coordinates.
From \eqn{pc} it should be clear that the parametrization \eqn{parortpq} is not a convenient one.
Instead we will proceed in two steps.
First we decompose $SO({d_{-}} +1 ,{d_{+}} )$ under $SO({d_{-}} , {d_{+}} )$
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\Lambda_{{d_{-}} +1,{d_{+}}}( \vec {S} , Q_{{d_{-}}{d_{+}}}) &=&
K_{{d_{-}} + 1,{d_{+}}}(\vec {S} )\; H(1,Q_{{d_{-}},{d_{+}}}) \cr
K_{{d_{-}} + 1,{d_{+}}}(\vec {S} ) &=& \matriz{C}{ {\vec {S}}^t \; \eta_{{d_{-}}{d_{+}}} }{\vec {S}}{1_{{d_{-}} ,{d_{+}}} +
\frac{{\vec {S}} \; {\vec {S}}^t \;\eta_{{d_{-}},{d_{+}}}}{1+C} } \label{dec1}
\eea
where $\; Q_{{d_{-}},{d_{+}}}\in SO({d_{-}} ,{d_{+}} )\;,\; {\vec S} \in \Re^{{d_{-}}+{d_{+}}}\;$, and
$\; C^2 = 1 + \vec {S}^t \eta_{{d_{-}},{d_{+}}}\vec {S} \geq 0\;$ define the coset manifold.
In what follows we will consider separately the $-{d_{-}}$ and ${d_{+}}$ directions,
$w^{\pm d_\pm } \equiv w^\pm$, and vectors, $\eta$ matrices, etc., will be
(${d_{-}} + {d_{+}} -1$)-dimensional; scalar products of vectors will be wrt
$\eta_{{d_{-}},{d_{+}} -1}\equiv\eta$, i.e.
${\vec v}_1 \cdot {\vec v}_2 \equiv {\vec v}_1{}^t\eta {\vec v}_2$.
In this spirit we introduce
\begin{equation}}\newcommand{\ee}{\end{equation}
\vec {S} \equiv \left( \begin{array}{c} {{\vec s}_1} \\ \sigma \end{array}\right)\;\;\; ,\;\;
C^2 = 1 + {{\vec s}_1}^2 + \sigma^2 \label{S}
\ee
With this in mind the second step is to decompose $Q_{{d_{-}},{d_{+}}}$ under $SO({d_{-}} , {d_{+}} -1)$
in a similar fashion
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
Q_{{d_{-}} ,{d_{+}}}( {{\vec s}_2} , Q) &=&
K_{{d_{-}} ,{d_{+}}}({{\vec s}_2} )\; H(Q, 1) \cr
K_{{d_{-}} ,{d_{+}}}({{\vec s}_2} ) &=& \matriz{1 - \frac{{{{\vec s}_2}} \; {{{\vec s}_2}}^t\;\eta}{1+c_2} }
{{{\vec s}_2}}{-{{{\vec s}_2}}^{t} \;\eta}{c_2} \label{dec2}
\eea
where $\; c_2{}^2 = 1 - {{\vec s}_2}^2 \geq 0\;$.
By plugging \eqn{S},\eqn{dec2} in \eqn{dec1} we get another parametrization of $SO({d_{-}}+1,{d_{+}})$
different from \eqn{parortpq} in terms of a $SO({d_{-}},{d_{+}} -1)$ matrix $Q$, two
real $({d_{-}} +{d_{+}} -1)$-dimensional vectors ${{\vec s}_1} ,{{\vec s}_2}$ and a real number $\sigma$.
As in \eqn{lintran} is straightforward to get the linear transformations of the
$(w^- , {\vec w} , w^+)$-coordinates.
With them and \eqn{pc}, \eqn{conspc} we finally obtain
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\frac{2 x^+}{ ^\Lambda x^+}\; \frac{^\Lambda{\vec x}}{r_{0}{}} &=&
( 1 + \frac{{{\vec s}_1}\cdot{{\vec s}_2} + \sigma c_2}{C + 1})\;{{\vec s}_1} + {{\vec s}_2} +
\left(( 1 - \frac{{{\vec s}_1}\cdot{{\vec s}_2} + \sigma c_2}{C + 1})\;{{\vec s}_1} -
{{\vec s}_2} \right)\; \frac{{\vec x}^2 + {x^+}^2}{r_{0}{}^2} \cr
&+& 2\; \left[1 + \left( \frac{{{\vec s}_1} {{\vec s}_1}^t}{1+ C} - \frac{{{\vec s}_2} {{\vec s}_2}^t}{1+
c_2} - (\frac{{{\vec s}_1}\cdot{{\vec s}_2}}{c_2 + 1} + \sigma)\;\frac{{{\vec s}_1} {{\vec s}_2}^t}{1+ C}
\right)\; \eta\right]\; Q\; \frac{{\vec x}}{r_{0}{}}\cr
\frac{2 x^+}{ ^\Lambda x^+} &=&
C + \sigma + c_2 + ( {{\vec s}_1}\cdot{{\vec s}_2} + \sigma c_2 )\; (1 +\frac{\sigma}{C+1})\cr
&+& 2\;\left[ (1 + \frac{\sigma}{C+1})\; {{\vec s}_1} -
\left( (1 + \frac{\sigma}{C+1})\; (\frac{{{\vec s}_1}\cdot{{\vec s}_2}}{c_2 + 1} +
\sigma)+ 1 \right)\; {{\vec s}_2}\right]\cdot Q\frac{{\vec x}}{r_{0}{}}\cr
&+& \left(C + \sigma - c_2 - ( {{\vec s}_1}\cdot{{\vec s}_2} + \sigma c_2 )\; (1 +
\frac{\sigma}{C+1})\right) \; \frac{{\vec x}^2 + {x^+}^2}{r_{0}{}^2}\label{nltpc}
\eea
These are the expressions we liked to find.
They realize non linearly the group of isometries $SO({d_{-}} + 1,{d_{+}})$ on the whole
$AdS_{d_-,d_+}$ space, and act on the invariant boundary $x^+ = 0$ (modulo
compactifications) in the usual way.
More explicitly, it is not difficult to show that the standard
conformal transformations are given by
\begin{itemize}
\item\underline{Pseudo-Poincar\`e transformations.}
\begin{equation}}\newcommand{\ee}{\end{equation}
^{(P,\vec a )}{\vec x} = P\; {\vec x} + \vec a \;\;\longrightarrow\;
\left\{ \begin{array}{l}
Q = P\cr
{{\vec s}_1} = r_{0}{}^{-1}\; \vec a \cr
{{\vec s}_2} = (1 +\frac{{\vec a}^2}{4r_{0}{}^2} )^{-1}\; \frac{\vec a}{r_{0}{}}\cr
\sigma = -\frac{{\vec a}^2}{2r_{0}{}^2} \end{array}\right.\label{pointr}
\ee
\item\underline{Dilations}
\begin{equation}}\newcommand{\ee}{\end{equation}
^\lambda{\vec x} = \lambda^{-1}\;{\vec x}
\;\;\longrightarrow\;
\left\{ \begin{array}{l}
Q = 1\cr
{{\vec s}_1} = \vec 0\cr
{{\vec s}_2} = \vec 0\cr
\sigma = \frac{1}{2}\; ( \lambda - \frac{1}{\lambda} )
\end{array}\right.\label{diltr}
\ee
\item\underline{Special conformal transformations}
\begin{equation}}\newcommand{\ee}{\end{equation}
^{\vec b}{\vec x} = \frac{{\vec x} + {\vec x}^2\;{\vec b}}{1 + 2\;{\vec b}\cdot{\vec x} + {\vec b}^2\;{\vec x}^2}
\;\;\longrightarrow\;
\left\{ \begin{array}{l}
Q = 1\cr
{{\vec s}_1} = r_{0}{}\; \vec b\cr
{{\vec s}_2} = -(1 + \frac{r_{0}{}^2}{4}\; {\vec b}^2)^{-1} \;r_{0}{}\;\vec b\cr
\sigma = \frac{{\vec b}^2}{2r_{0}{}^2}\end{array}\right.\label{spcontr}
\ee
\end{itemize}
\section{The $AdS_{d_-,d_+}$ and conformal algebras}
\setcounter{equation}{0}
Here we consider the representation of the corresponding algebras and their meaning in
each context.
The $SO({d_{-}} +1, {d_{+}})$ Lie algebra is ($X_{MN} = - X_{NM}$)
\begin{equation}}\newcommand{\ee}{\end{equation}
[X_{MN}, X_{PQ}] = \eta_{NP}\; X_{MQ} + \eta_{MQ}\; X_{NP} - (M\leftrightarrow N)
\label{stalg}
\ee
An arbitrary element $\delta\Lambda$ in the vector representation obeys:
$\;\delta\Lambda^t\; \eta + \eta\; \delta\Lambda =0\; $, and then
$\;\delta\Lambda = \frac{1}{2} \delta\Lambda^{MN} V_{MN}\;$,
where the generators in this representation are
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
V_{MN} &\equiv& V(X_{MN}) = E_{MN} - E_{NM}\cr
(E_M{}^N)^P{}_Q &\equiv& \delta_M^P\; \delta_Q^N
\eea
The infinitesimal generators acting on functions in ${\cal M}_{d_- +1,d_+}$ are
introduced in the standard way in order to obey \eqn{stalg}
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\delta_{\delta\Lambda} w^M &\equiv& - \delta\Lambda^{PQ}\; {\hat X}_{PQ} w^M\cr
{\hat X}_{MN} &=& w_M \;\partial_N - w_N\; \partial_M \label{infgen}
\eea
Then the realization of the generators on $AdS_{d_-,d_+}$ in $(w^\mu)$ coordinates are gotten by taking
into account the constraint \eqn{cons}
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
{\hat X}_{\mu\nu} &=& w_\mu \; \partial_\nu - w_\nu \; \partial_\mu\cr
{\hat p}_\mu &=& \pm \; (w^\rho w_\rho + r_{0}{}^2)^\frac{1}{2}\; \partial_\mu = {\hat X}_{\mu -} |_{AdS}
\eea
and obeys the $AdS_{d_-,d_+}$ algebra
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
[{\hat X}_{\mu\nu}, {\hat X}_{\rho\sigma}] &=& \eta_{\nu\rho}\; {\hat X}_{\mu\sigma} +
\eta_{\mu\sigma}\; {\hat X}_{\nu\rho} - (\mu\leftrightarrow \nu)\cr
[ {\hat p}_\rho ,{\hat X}_{\mu\nu}] &=& \eta_{\mu\rho}\; {\hat p}_\nu -
\eta_{\nu\rho}\; {\hat p}_\mu \cr
[{\hat p}_\mu , {\hat p}_\nu ] &=& {\hat X}_{\mu\nu }\label{infgenads}
\eea
Now let us see the realization in Poincar\`e coordinates.
From \eqn{nltpc} the infinitesimal transformations are
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\delta_{\delta\Lambda} {\vec x} &=& \delta P\; {\vec x} + \delta\vec a
- \delta\lambda\; {\vec x} + \delta\vec b\; ( {\vec x}^2 + (x^+)^2 ) - 2\;\delta\vec b\cdot {\vec x}\;\; {\vec x}\cr
\delta_{\delta\Lambda} x^+ &=& - (\delta\lambda + 2\; \delta\vec b\cdot {\vec x}\ )\; x^+
\eea
where in view of \eqn{pointr}, \eqn{diltr}, \eqn{spcontr} we have introduced
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\delta P &=& \delta Q\cr
\delta\vec a &=& \frac{r_{0}{}}{2}\; ( \delta{{\vec s}_1} + \delta{{\vec s}_2} )\cr
\delta\lambda &=& \delta\sigma\cr
\delta\vec b &=& \frac{1}{2r_{0}{}}\; ( \delta{{\vec s}_1} - \delta{{\vec s}_2} )
\eea
and according to the definition in the first line of \eqn{infgen}
we get the infinitesimal generators associated to $\;(\delta P ,
\delta\vec a , \delta\lambda, \delta\vec b )\;$ respectively as
($\;\partial_\mu\equiv \frac{\partial\;\;}{\partial x^\mu}$)
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
{\hat J}_{\alpha\beta}&=&x_\alpha \;\partial_\beta - x_\beta\; \partial_\alpha = {\hat X}_{\alpha\beta}\cr
{\hat P}_\alpha &=& -\partial_\alpha = \frac{1}{r_{0}{}}\;( {\hat p}_\alpha + {\hat X}_{\alpha +})\cr
{\hat D} &=& x^\beta\;\partial_\beta + x^+\;\partial_+ = {\hat p}_+\cr
{\hat K}_\alpha &=& \left( -({\vec x}^2 + (x^+)^2 )\;\delta_\alpha^\beta +
2\; x_\alpha\; x^\beta\right) \; \partial_\beta + 2\; x_\alpha\;
x^+\;\partial_+ = r_{0}{}\;( {\hat p}_\alpha - {\hat X}_{\alpha+})\label{infgencon}
\eea
These operators obeys the $SO({d_{-}} + 1,{d_{+}})$ algebra in the form
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
[J_{\alpha\beta}, J_{\gamma\delta}] &=& \eta_{\beta\gamma}\; J_{\alpha\delta} +
\eta_{\alpha\delta}\; J_{\beta\gamma} -(\alpha\leftrightarrow\beta)\cr
[{\hat P}_\gamma , {\hat J}_{\alpha\beta}] &=& \eta_{\alpha\gamma}\;{\hat P}_\beta -
\eta_{\beta\gamma}\;{\hat P}_\alpha\cr
[{\hat K}_\gamma , {\hat J}_{\alpha\beta}] &=& \eta_{\alpha\gamma}\;{\hat K}_\beta -
\eta_{\beta\gamma}\;{\hat K}_\alpha\cr
[{\hat P}_\alpha ,{\hat D} ] &=& {\hat P}_\alpha \cr
[{\hat K}_\alpha ,{\hat D} ] &=& -{\hat K}_\alpha \cr
[{\hat K}_\alpha , {\hat P}_\beta] &=& 2\; (\eta_{\alpha\beta}\; {\hat D} +{\hat J}_{\alpha\beta})\cr
0 &=& [{\hat P}_\alpha ,{\hat P}_\beta ] = [{\hat K}_\alpha , {\hat K}_\beta] = [{\hat D} , {\hat J}_{\alpha\beta}]
\eea
which is the standard form of the conformal algebra.
In conclusion, the standard $SO({d_{-}} +1,{d_{+}})$ linear realization leads to the usual
representation of their infinitesimal generators \eqn{infgen} in the space of functions
on ${\cal M}_{d_- +1,d_+}$; the non-linear realization instead yields the representation \eqn{infgenads}
of them in the space of functions on $AdS_{d_-,d_+}$, which in turns take the form \eqn{infgencon}
in Poincar\`e coordinates where the conformal algebra point of view is clear, in particular
at the boundary $x^+ = 0$.
\section{Coset representation and geodesics}
\setcounter{equation}{0}
$AdS_{d_-,d_+}$ spaces can be defined as pseudo-Riemannian homogeneous symmetric spaces, i.e.
a coset space $G/H$; more precisely: $\; AdS_{d_-,d_+} \sim SO({d_{-}} +1 ,{d_{+}})/SO({d_{-}},{d_{+}})\;$.
This can be seen directly from the coset decomposition introduced in
\eqn{dec1}; let us rewrite the coset element in the form
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
K(\vec {S}) &=& \exp \matriz{0}{\vec u^t \eta}{\vec u}{0} =
\matriz{C}{\vec {S}^t \; \eta}{\vec {S}}{1 + \frac{\vec {S} \; \vec {S}^t\;\eta}{1+C} }\cr
\vec {S} &=& \frac{\sinh \sqrt{\vec u^2}}{\sqrt{\vec u^2}}\; \vec u\cr
-1 &=& - C^2 + \vec {S}^2\label{adscos}
\eea
From here and \eqn{cons} we can identify our defining coordinates as
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
w^- &\equiv& r_{0}{}\; C\cr
w^\mu &\equiv& r_{0}{}\; S^\mu \label{coorads}
\eea
Furthermore it is possible to show \cite{gil} that the metric induced on it
by the invariant metric on $SO({d_{-}} +1,{d_{+}})$ is given by \eqn{gdc}.
There exists an elegant method in this context to get the geodesics wrt this metric \cite{gil}.
It consists simply in lifting up to the coset element the geodesics through the null element
in the Lie algebra wrt the Cartan-Killing metric on it.
In this way we obtain a geodesic through the identity in the coset space that we identify
with the point $(w^{-} = r_{0}{} , \vec w =\vec 0)$ in $AdS_{d_-,d_+}$.
\footnote{
More exactly, this is so if ${d_{-}}=0$ according to the choice of connected component
discussed in Section 2.1.
If ${d_{-}} >0$ we have two sets of geodesics if they are not time-like, one covering the region
$w^- >0$ corresponding to take $w^- = +{r_{0}{}}$ as the origin, and one covering the region
$w^- < 0$ if starting through $w^- = -r_{0}{}$.
If they are time-like instead, there is just one family that rolls up the time direction,
see below.
}
In order to get a geodesic through an arbitrary point we have to act by left action with an
element of the coset.
Let us show how all this works here.
First of all let us consider geodesics through the origin of $AdS_{d_-,d_+}$.
They are given by \eqn{adscos} with $\vec u$ replaced by
\begin{equation}}\newcommand{\ee}{\end{equation}
\vec{\bar u}(\tau) = \vec u\; \tau\label{geoalg}
\ee
where the ``velocity" $\vec u$ parametrizes the family.
This is so because \eqn{geoalg} are the geodesics through the null element in the Lie
algebra.
Therefore according to \eqn{adscos}, \eqn{coorads} the geodesics in $AdS_{d_-,d_+}$ through the origin
will be given in $(w^\mu)$-coordinates by
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
{\bar w}^\mu (\tau) = r_{0}{}\; \left\{ \begin{array}{l}
\sinh \omega\tau \; {\check u}^\mu \;\; ,\; {\check u}^2 = +1 \;\;,\; space-like\\
\;\;\sin \omega\tau \; {\check u}^\mu \;\; ,\; {\check u}^2 = -1 \;\;,\; time-like\\
\;\tau \;u^\mu\;\;\;\;\;\;\;\;\;\; , \;\vec u^2 = 0 \;\;\;\;\;,\; null\end{array}\right.
\eea
where $\; \omega = \sqrt{|\vec u^2|}\; , {\check u}=\vec u /\omega$.
For sake of interpretation is useful to have them in global coordinates $(r_L ,{\check n}_{+} ,{\check n}_{-} )$.
If $\vec u = (\vec u_- , \vec u_+)$ and we reintroduce the vector
\begin{equation}}\newcommand{\ee}{\end{equation}
{\vec w}_{+} \equiv r_{0}{}\; \sinh r_L \; {\check n}_{+} \;\; \in \Re^{d_{+}}
\ee
parametrizing the space-like sections of $AdS_{d_-,d_+}$, then
\bigskip
\noindent\underline{Space-like geodesics:}
with $\; {\check u}_+{}^2 = 1 + {\check u}_-{}^2 \ge 1$ we have
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\vec{\bar w}_+ (\tau) &=& r_{0}{}\; {\check u}_+ \; \sinh \omega\tau\cr
\check{\bar n}_- (\tau) &=& (1 + {\check u}_+{}^2 \sinh^2\omega\tau )^{-\frac{1}{2}}\;
\left( \begin{array}{l} \cosh\omega\tau\\ \sinh\omega\tau\;{\check u}_- \end{array}
\right)\label{geosp}
\eea
\noindent\underline{Time-like geodesics:}
with $\; {\check u}_-{}^2 = 1 + {\check u}_+{}^2 \ge 1$ we have
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\vec{\bar w}_+ (\tau) &=& r_{0}{}\; {\check u}_+ \; \sin \omega\tau\cr
\check{\bar n}_- (\tau) &=& (1 + {\check u}_+{}^2 \sin^2\omega\tau )^{-\frac{1}{2}}\;
\left( \begin{array}{l} \cos\omega\tau\\ \sin\omega\tau\;{\check u}_- \end{array} \right)
\label{geoti}
\eea
\noindent\underline{Null geodesics:}
with $\; {\vec u}_+{}^2 = {\vec u}_-{}^2$ we have
\begin{eqnarray}} \newcommand{\eea}{\end{eqnarray}
\vec{\bar w}_+ (\tau) &=& r_{0}{}\; {\vec u}_+ \;\tau\cr
\check{\bar n}_- (\tau) &=& (1 + {\vec u}_+{}^2 \tau^2 )^{-\frac{1}{2}}\;
\left( \begin{array}{c} 1\\ \tau\;{\vec u}_- \end{array} \right)\label{geonu}
\eea
From these explicit expressions it is easy for example to compute for a non-null
geodesic the distance between the origin and a point at $\tau$; it is simply:
$\; s(\tau) = \omega\;r_{0}{}\;\tau$.
Null geodesics reach the time-like boundary asymptotically in a conserved spatial
direction $\vec u_+/\sqrt{\vec u_+{}^2}$.
Instead time-like geodesics are bounded and periodic with period $T = \frac{2\pi}{\omega}$;
they never reach the boundary.
\footnote{
For ${d_{-}} =1 $, $\;{\check n}_+ = \left(\begin{array}{c} \cos t\\
\sin t\end{array}\right)\;$ with $\; t\sim t+ 2\pi$ and then they wrap the time-like $S^1$
and re-converge to the starting point.
If, as usual made not to have closed time-like loops, $S^1$ is replaced by $\Re$, the
geodesics converge to an image point and diverge from it to re-converge to another one and so on.
}
We complete the analysis given the expression of the geodesics through an arbitrary
point.
From the stated at the beginning of this section we need to get the map
\begin{equation}}\newcommand{\ee}{\end{equation}
\vec {S} \longrightarrow {}^{\vec {S}_0}\vec {S} \label{map}
\ee
defined by the left action of an element in the coset in the form
\begin{equation}}\newcommand{\ee}{\end{equation}
K(\vec {S}_0 )\; K(\vec {S} ) \equiv K( {}^{\vec {S}_0}\vec {S} )\; H(1,Q) \;\;\; , \; Q\in SO({d_{-}},{d_{+}})
\ee
Then having obtained the map ${}^{\vec {S}_0}\vec {S}$ a geodesic through $\vec {S}_0$ is just
\begin{equation}}\newcommand{\ee}{\end{equation}
\vec{\bar S} (\tau; \vec {S}_0, \vec u ) = {}^{\vec {S}_0}\vec{\bar S}(\tau)
\ee
where $\vec{\bar S}(\tau)$ is a geodesic passing through the origin.
After a little bit of algebra we get the map in the simple form
\begin{equation}}\newcommand{\ee}{\end{equation}
\left( \begin{array}{l} {\bar C}(\tau; \vec {S}_0 ,\vec u) \\ \vec{\bar S}(\tau; \vec {S}_0 ,\vec u)\end{array}
\right) = K(\vec {S}_0)\;\left( \begin{array}{l}{\bar C}(\tau)\\ \vec{\bar S}(\tau)\end{array}\right)
\ee
From here, the identifications \eqn{coorads} and the results \eqn{geosp}, \eqn{geoti},
\eqn{geonu} we have the explicit whole set of geodesics in $AdS_{d_-,d_+}$ spaces.
\section{Conclusions}
\setcounter{equation}{0}
First of all it is worth to say that we have concentrated along the paper on $AdS_{d_-,d_+}$ spaces,
but it should be clear that $dS_{d_-,d_+}$ ones are straigthforwardly studied in our metric
signature convention starting from \eqn{cons} with a $``+"$ sign on its r.h.s. and working
on $\;{\cal M}_{{d_{-}},{d_{+}}+1}\;$.
In particular the coset representation for them is given by $SO({d_{-}},{d_{+}} +1)/SO({d_{-}},{d_{+}})$.
We obtained explicit global non linear realizations of the isometry groups, equations
\eqn{nltgc}, \eqn{nltpc}, by using various coset decompositions where the parameter space of
the transformations is manifest at difference what happens by considering infinitesimal
transformations.
From these realizations issues like the conformal compactification of the boundary as well as
the relation between standard generators of orthogonal, AdS and conformal algebras are
evident.
Finally it is none to say that geodesics in spaces of physical interest are well-known,
\footnote{
See for example reference \cite{he}, chapter 5.
}
in particular $AdS_{1,3}$.
We believed useful and instructive however to extend these results to arbitrary $AdS_{d_-,d_+}$ spaces
giving the explicit expressions obtained through the coset space method without any reference
to the metric.
We think our results can be of usefulness among other things in the non trivial problem of
fixing boundary conditions in the search of solutions of gravity-matter classical field theories
in the presence of a non zero cosmological constant \cite{ls}.
\section*{Acknowledments} We thank to Fidel Schaposnik for discussions and references.
|
\section{Introduction}
The sine-Gordon model is an exactly solvable model which has an
enormous number of applications in
condensed matter physics and statistical mechanics. It has been
studied for years with many remarkable results being
obtained. However, most of the effort has been concentrated on studying
this
model as a quantum field theory. In this paper we discuss the
sine-Gordon model together with a less famous sinh-Gordon model as
models of classical statistical mechanics analyzing the behavior of
their specific heat in the various range of parameters.
Let us consider the
{\it classical} sine- and sinh-Gordon models whose partition functions are
given by
\begin{eqnarray}
Z &=& \int {\cal{D}}\varphi e^{- S[\varphi]}\quad , \nonumber \\
S_{\rm sin} &\equiv& \frac{1}{T}E_{\rm sin}[\varphi]=\int d^2 x\left[
\frac{\rho_{\rm s}}{2T}\left| \nabla \varphi -
{\bf Q}\, \right|^2+ \frac{m}{T}\left(1-\cos\varphi \right)\right] \quad , \\
S_{\rm sinh} &\equiv& \frac{1}{T}E_{\rm sinh}[\varphi]=\int d^2
x\left[\frac{\rho_{\rm s}}{2T}\left| \nabla \varphi
\right|^2+ \frac{m}{T}\left(\cosh\varphi - 1\right)\right] \quad .
\end{eqnarray}
The first model describes, for example, the commensurate-incommensurate
transition \cite{Talapov}. Most recently it has been applied to
double-layered Quantum Hall systems \cite{MacDonald}.
The incommensurate phase appears when $|{\bf Q}|$
exceeds some critical value and is characterized by nonzero average
value $\left<{\bf Q}\nabla\varphi\right>$.
Redefining the field variable $(\rho_{\rm s}/T)^{1/2}\varphi = \phi$ we reduce
the above Euclidean action to the canonical sine-Gordon form
(see, for example, \cite{Zamolodchikov}):
\begin{equation}
S_{\rm sin} = \int d^2x\left[\frac{1}{2}\left| \nabla \phi -
{\bf h}\beta/2\pi\, \right|^2+ \mu\left(1 - \cos\beta\phi\right)\right] \quad ,
\label{sg}
\end{equation}
with $\beta^2 = T/\rho_{\rm s}$ and $\mu = m/T$, $|{\bf h}| = 2\pi
|{\bf Q}|/\beta^2$.
In a similar fashion the sinh-Gordon action
becomes
\begin{equation}
S_{\rm sinh} = \int d^2x\left[\frac{1}{2}\left| \nabla \phi\, \right|^2+
\mu\left(\cosh\beta\phi - 1\right)\right] \label{sh} \quad .
\end{equation}
For this model there are no kinks and creation of nonzero field
gradient would require imaginary field $h$. We do not consider such
possibility.
One can consider the quantum field theory in (1+1)-dimensions with
$Z=\int \protect{\cal{D}} \phi e^{-S[\phi]} $ where $S$ is given by (\ref{sg})
or (\ref{sh}). The exact solution for both models is known and
we can take advantage of the fact that the free energy of
a $D$-dimensional classical model is related to the ground state
energy $E_{\rm 0}$ of the
corresponding quantum field theory living in a space of
$(D-1)$-dimensions. More specifically, for $D = 2$
the partition function of the classical theory defined on
a rectangle $L_x\times L_y$ with periodic boundary conditions in the
$x$-direction at temperature $T$ is equal to the partition function of the
quantum field theory at temperature $L_x^{-1}$ with the coupling constant
$\beta^2 = T$. The limit $L_x \rightarrow \infty$
corresponds to the limit of zero temperature in the quantum field theory when
its free energy is equal to the ground state energy
$E_{\rm 0} = L_y{\cal E_{\rm 0}}$.
Thus we get the following relation between the free energy per unit area of the 2-dimensional classical model ${\cal F}$
and the ground state energy per unit length of the (1+1)-dimensional field
theory:
\begin{equation}
{\cal F}(T) = T{\cal E}_{\rm 0}[\beta(T)] \quad .
\end{equation}
The ground state energy ${\cal E}_{\rm 0}$ of the quantum sine-Gordon model
as a function of parameters $\beta(T)$ and $\mu$
is known exactly (\cite{Zamolodchikov}, \cite{Vega}) and the corresponding
expression for the sinh-Gordon model can be extracted from
\cite{Fateev}.
\section{Sine-Gordon Model at $Q$ = 0}
We start our discussion of the sine-Gordon
model with the case $Q = 0$ which is already quite
nontrivial. In the Bethe ansatz approach the ground state energy of
the quantum sine-Gordon model is calculated by regularizing the model
by putting it on a lattice. The lattice constant $a$ and the inverse
coupling constant $\theta$ of the
regularized model are related to the mass of physical particles.
According to \cite{Vega} the ground state energy (at $Q\,=0$) is given by:
\begin{equation}
{\cal E}_{\rm 0}=\frac{1}{a^2}\int_{-\infty}^{+\infty} \frac{\sin 4\theta t}{t}
\frac{\sinh(\pi \tau t)}{\cosh[\pi(1 - \tau) t] \ \sinh (\pi t)} dt \quad ,
\label{grounden}
\end{equation}
where $\tau =T/(8\pi \rho_{\rm s}) \equiv T/T_{\rm c}$.
The parameters $\theta$ and $a$ are related to the kink's
mass:
\begin{equation}
m_{\rm s}=\frac{4}{a} e^{- \theta/(1 - \tau)}
\quad ,
\end{equation}
To exclude $\theta$ from Eq. (\ref{grounden}) we use
the T-dependence of the kink's mass
given in \cite{Zamolodchikov}:
\begin{equation}
m_{\rm s}=\frac{1}{a} \frac{2}{\sqrt{\pi}}
\frac{\Gamma{\left(\frac{1}{2}\frac{T}{T_{\rm c}-T}\right)}}{\Gamma{
\left(\frac{1}{2}\frac{T_{\rm c}}{T_{\rm c}-T}\right)}}
\left\{\frac{\pi m}{T_{\rm c}}\frac{\Gamma{\left(1-\frac{T}{T_{\rm c}}\right)}}
{\Gamma{\left(1 + \frac{T}{T_{\rm c}}\right)}}\right\}^{
\frac{1}{2\left(1-\frac{T}{T_{\rm c}}\right)}} \quad .
\end{equation}
Note that in the limiting case $\tau \ll 1$ we have
\begin{equation}
m_{\rm s}=\frac{4}{a}\tau^{-1}(m/\pi T_{\rm c})^{1/2} \label{limit}
\quad .
\end{equation}
We should stress that the above expressions make sense
only for $m_{\rm s}a\ll 1$ when the continuous
approach works. Therefore to
calculate the ground state energy in the continuous
approximation in Eq. (\ref{grounden})
one has to keep
only the pole closest to real axis.
Near the point $\tau = 1/2$ two poles at
$t=i$ and $t=i/2(1 - \tau)$ compete and one has to take
into account both of them.
Taking this into account we obtain from (\ref{grounden})
the general expression for the free energy:
\[
F = F_{\rm 1} + F_{\rm 2} \quad ,
\]
\begin{eqnarray}
\label{F1}
F_{\rm 1} & = & T\ \frac{m_{\rm s}^2}{4} \cot\left[\frac{\pi}{2 (1 - \tau)}
\right]
\\ [3mm]
& = & \frac{T}{\pi a^2}\cot\left(\frac{\pi}{2(1 - T/T_{\rm c})}\right)
\frac{\Gamma^2{\left(\frac{1}{2}\frac{T}{T_{\rm c}-T}\right)}}{\Gamma^2{
\left(\frac{1}{2}\frac{T_{\rm c}}{T_{\rm c}-T}\right)}}
\left\{\frac{\pi m}{T_{\rm c}}\frac{\Gamma{\left(1-\frac{T}{T_{\rm c}}\right)}}
{\Gamma{\left(1 +\frac{T}{T_{\rm c}}\right)}}\right\}^{
\frac{1}{\left(1-\frac{T}{T_{\rm c}}\right)}} \quad,
\nonumber \\ [3mm]
F_{\rm 2} & = &
-\ T\ \frac{2}{a^{2}}\ \left(\frac{m_{\rm s}a}{4}\right)^{4(1 - \tau)}
\ \tan\left[\pi(1 - \tau)\right] \quad .
\end{eqnarray}
We emphasize that the necessity to keep both terms in the expression for the
free energy exists only close to the free fermion point. At $\tau < 1/2$ the
free energy remains finite in the continuous limit
($a \rightarrow 0, m_{\rm s} =$const), while at $\tau > 1/2$ it diverges.
The latter fact is in agreement with the perturbation theory in $m$:
\[
\int d^2x\left<\cos\beta\phi(x)\cos\beta\phi(0)\right> \sim
\int d^2x (x/a)^{-4(1 - \tau)}
\]
diverges at small distances.
\begin{figure}
\unitlength=1mm
\begin{picture}(165,156)
\put(-3,4){\line(1,0){162}}
\put(-3,54){\line(1,0){162}}
\put(-3,104){\line(1,0){162}}
\put(-3,4){\line(0,1){152}}
\put(159,4){\line(0,1){152}}
\put(-3,156){\line(1,0){162}}
\put(79,4){\line(0,1){152}}
\put(0,4){\epsfig{file=math2.000001.eps,height=48mm}}
\put(0,106){\epsfig{file=math2.01.eps,height=48mm}}
\put(81,106){\epsfig{file=math2.001.eps,height=48mm}}
\put(0,54){\epsfig{file=math0.0001.eps,height=48mm}}
\put(81,54){\epsfig{file=math0.00001.eps,height=48mm}}
\put(81,4){\epsfig{file=math0.0000001.eps,height=48mm}}
\put(40,146){\makebox(0,0)[cc]{(a)}}
\put(120,146){\makebox(0,0)[cc]{(b)}}
\put(40,95){\makebox(0,0)[cc]{(c)}}
\put(120,95){\makebox(0,0)[cc]{(d)}}
\put(40,47){\makebox(0,0)[cc]{(e)}}
\put(120,47){\makebox(0,0)[cc]{(f)}}
\put(43,140){\makebox(0,0)[tc]{$m=0.01$}}
\put(123,140){\makebox(0,0)[tc]{$m=0.001$}}
\put(47,89){\makebox(0,0)[tc]{$m=0.0001$}}
\put(127,89){\makebox(0,0)[tc]{$m=0.000\,01$}}
\put(47,41){\makebox(0,0)[tc]{$m=0.000\,001$}}
\put(127,41){\makebox(0,0)[tc]{$m=0.000\,0001$}}
\end{picture}
\caption{Plots of the specific heat as a function of $T/Tc$, for
different values of the parameter $m$, are shown.}
\end{figure}
Since we always keep $a$ finite, we plot the specific heat at finite values of
$(\pi m/T_{\rm c})$, and
it is convenient to separate the interval of $\tau$ into
three regions. First, for $T/T_{\rm c}\in (0.1,0.35)$, we take the
contribution of only the nearest pole $i/2(1 - \tau)$.
Second, the region $T/T_{\rm c}\in(0.35,0.7)$ where we
take the contribution of both poles. Third, for
$T/T_{\rm c}\in(0.7,0.9)$ we take only the contribution of
the pole $t=i$.
We combine these results to find the specific heat $C_{\rm v}(T)=
-T\partial^2 F/\partial^2 T$ as a function of
\begin{figure}
\unitlength=1mm
\begin{picture}(162,56)
\put(-3,4){\line(1,0){162}}
\put(-3,56){\line(1,0){162}}
\put(-3,4){\line(0,1){52}}
\put(78,4){\line(0,1){52}}
\put(159,4){\line(0,1){52}}
\put(0,6){\epsfig{file=ms0.0001.eps,height=48mm}}
\put(81,6){\epsfig{file=ms0.0000001.eps,height=48mm}}
\put(47,46){\makebox(0,0)[cc]{(a)}}
\put(127,46){\makebox(0,0)[cc]{(b)}}
\put(47,40){\makebox(0,0)[tc]{$m=0.0001$}}
\put(127,40){\makebox(0,0)[tc]{$m=0.000 0001$}}
\end{picture}
\caption{Plots of $(a m_{\rm s})$ as function of $(T/Tc)$, for different
values of the parameter $m$, are shown. The condition of continuous
approach works on the region $(a m_{\rm s})\ll 1$.}
\end{figure}
$T/T_{\rm c}$ and $m/T_{\rm c}$,
in $T/T_{\rm c}\in (0,1)$.
On Fig. 1 we present plots of the temperature of the specific heat
for various values of $m$. Fig. 2 gives temperature dependence
of the kink's mass. From this picture one one can estimate the region where
the condition $m_{\rm s}a\ll 1$ is fulfilled.
At $1 - \tau \ll 1$ and at $\tau \ll 1$ expressions (\ref{F1}) for the
specific heat simplify. In the first case we have
\begin{equation}
a^2 F = - \frac{2\pi^3 m^2}{T_{\rm c}}\left\{\frac{1}{1 - \tau} +
2\ln\left[\frac{e^2}{\pi(1 - \tau)}\right] + O\left((1 - \tau)\ln^2(1 -
\tau)\right)\right\} \quad ,
\end{equation}
\begin{equation}
C_{\rm v} = \frac{4\pi^3}{a^{2}}\frac{m^2T_{\rm c}}{(T_{\rm c} - T)^3} + ...
\quad .
\end{equation}
This singularity is associated with the Kosterlitz-Thouless transition
at $T = T_{\rm c}$.
At $\tau \ll 1$ we have
\begin{eqnarray}
a^2 F &=& - 4\pi^2 m\left(\frac{\pi m}{T_{\rm c}}\right)^{1/(1 - \tau)}
\quad , \\
C_{\rm v} &=& 4\pi\left(\frac{\pi m}{T_{\rm c}}\right)^2
\ln\left(\frac{T_{\rm c}}{\pi m}\right)\left[2 +
\ln\left(\frac{T_{\rm c}}{\pi m} \right)\right] \tau
\exp\left\{- \tau\ln\left(\frac{T_{\rm c}}{\pi m} \right)\right\} \quad .
\end{eqnarray}
The latter expression explains the existence of the maximum in the
specific heat: at $\ln(T_{\rm c}/\pi m) \gg 1$ the maximum occurs at
$\tau^* = [\ln(T_{\rm c}/\pi m)]^{-1}$.
\section{Sine-Gordon Model in the Incommensurate Phase}
At
\begin{equation}
Q > Q_{\rm c} = 4\tau m_{\rm s}(\tau) \quad \label{condit}
\label{criteq}
\end{equation}
the sine-Gordon model is in the
incommensurate phase characterized by a condensate of
kinks $\left<{\bf Q}\nabla\phi \right> \neq 0$. The ground state
energy of the corresponding quantum field theory acquires an additional
contribution originating from the condensate. The corresponding change
in the free energy of the classical model is
\begin{eqnarray}
\delta F = \frac{\rho_{\rm c} Q^2}{2} + \frac{T m_{\rm s}}{2\pi}\int_{-B}^B
d\theta\cosh\theta \epsilon(\theta) \quad .
\end{eqnarray}
The nonpositive function $\epsilon(\theta)$ is defined inside the
interval $-B< \theta < B$ and satisfies the integral equation (see,
for example \cite{Zamolodchikov})
\begin{eqnarray}
\epsilon(\theta) + \int_{-B}^B d\theta' K(\theta -
\theta')\epsilon(\theta') = m_{\rm s}\cosh\theta - \frac{Q}{4\tau} \quad ,
\label{integ}
\end{eqnarray}
where the Fourier image of the kernel is
\[
K(\omega) = \frac{\sinh\frac{\pi(1 - 2\tau)\omega}{2(1 -
\tau)}}{2\cosh(\pi\omega/2)\sinh\frac{\pi\tau\omega}{2(1 - \tau)}} \quad .
\]
The kernel $K(\theta)$ encodes the information about the soliton-soliton
scattering.
The limit $B$ is determined by the condition $\epsilon(\pm B) = 0$. A
possibility of $\epsilon$ being negative appears when the right hand
side becomes not positively defined which corresponds to condition
(\ref{condit}).
We have solved the integral equation for the function $\epsilon(\theta)$
numerically and the plots of its dependence on $\tau$ and $Q$ are shown in
Fig. \ref{epstheta}.
The dependence of
the limit $B$ on the temperature and $Q$ is shown on
Fig. \ref{boundaryB}.
\begin{figure}
\unitlength=1mm
\begin{picture}(160,55)
\put(-3,-3){\epsfig{file=critline2.eps,height=55mm}}
\put(92,0){\makebox(68,55)[b]{\begin{minipage}[b]{68mm}
\protect\caption{\sloppy
The critical line as function of $(T/T_{\rm c})$ for a
fixed value of the parameter $m$ is shown. The lattice constant $a$ is
taken to be unity.}
\end{minipage}}}
\end{picture}
\end{figure}
It is curious that the critical field has a finite limit at $\tau
\rightarrow 0$ (this limit was first considered in \cite{fowler}). According to Eq. (\ref{limit}), we have
\begin{equation}
aQ_{\rm c}(0) = 16\left(\frac{m}{\pi T_{\rm c}}\right)^{1/2} \quad .
\end{equation}
In the vicinity of the critical line in the incommensurate phase
one can expand solution of
Eq. (\ref{integ}) in series in $B$ and get for the additional free energy:
\begin{eqnarray}
\frac{1}{T_{\rm c}}\delta F & = & \frac{Q^2}{2}
\\ [3mm]
& & - \frac{\tau m_{\rm s}^2}{6\pi}
\left(\frac{Q}{Q_{\rm c}}- 1\right)^{3/2}\left[1 -
K(0)\left(\frac{Q}{Q_{\rm c}}- 1\right)^{1/2} + \left(0.1 +
7 \frac{K^2(0)}{6}\right)\left(\frac{Q}{Q_{\rm c}}- 1\right) + ...\right]\ .
\nonumber
\end{eqnarray}
At small $\tau$, $K(0) \approx \frac{1}{\pi^2\tau}\ln(1/\tau)$ and the
expansion is valid for
\begin{equation}
\pi^2\tau/\ln(1/\tau) \gg (Q/Q_{\rm c} - 1) \label{cond} \quad .
\end{equation}
Plots of the additional specific heat are shown on Fig. \ref{Cvalltogether}a),
whereas plots of the total specific heat are shown in
Fig. \ref{Cvalltogether}b) for some values
of the field $Q$.
The $Q/\tau \rightarrow \infty$ analytic structure of the total free
energy $F(Q)$ is given by Zamolodchikov \cite{Zamolodchikov}
\begin{equation}
\label{eq22}
F(Q)-F(0) = \frac{\rho_{\rm c} Q^2}{2} - T\ \frac{m_{\rm s}^2}{4}
\cot\left[\frac{\pi}{2 (1 - \tau)} \right]
- \left(\frac{Q}{4 \tau}\right)^2 \frac{k(Q/4\tau)}{\pi} \quad .
\end{equation}
The factor $k(Q/4\tau)$ is given as a power series
\begin{figure}
\unitlength=1mm
\begin{picture}(160,103)
\put(-2,53){\epsfig{file=intequ1.eps,height=50mm}}
\put(78,53){\epsfig{file=intequ3.eps,height=50mm}}
\put(-2,3){\epsfig{file=intequ2.eps,height=50mm}}
\put(78,3){\epsfig{file=IQbaa.MET,height=50mm}}
\put(6,99){\makebox(0,0)[cc]{(a)}}
\put(150,99){\makebox(0,0)[cc]{(b)}}
\put(6,49){\makebox(0,0)[cc]{(c)}}
\put(150,49){\makebox(0,0)[cc]{(d)}}
\end{picture}
\caption{(a) Plots of $(-\epsilon(\theta))$ for different fixed
$\tau$. On figure (b) the parameter $Q = 0.090$, which corresponds to
crossing
the critical line at $\tau\approx 0$.}
\label{epstheta}
\end{figure}
\begin{figure}
\unitlength=1mm
\begin{picture}(160,53)
\put(-2,3){\epsfig{file=Btau1.eps,height=50mm}}
\put(78,3){\epsfig{file=Btau2.eps,height=50mm}}
\end{picture}
\caption{Plots of temperature dependence of the parameter $B$ in
Eq. (\ref{integ}).
The parameter $Q$ is taken to have values $0.060$ and $0.120$,
respectively.}
\label{boundaryB}
\end{figure}
\begin{equation}
k(Q/4\tau)=\sum_{n=0}^{\infty}K_{\rm n}y^n \quad ,
\end{equation}
with
\begin{equation}
y = \left[\frac{2 m_{\rm s}\sqrt{\pi}}{Q}\frac{\Gamma\left(\frac{1}{2(1-\tau)}
\right)}{\Gamma\left(\frac{\tau}{2(1-\tau)}\right)}\right]^{4(1-\tau)}
\quad .
\end{equation}
\begin{figure}
\unitlength=1mm
\begin{picture}(160,53)
\put(77,3.1){\line(1,0){82}}
\put(159,3.1){\line(0,1){50}}
\put(77,53.1){\line(1,0){82}}
\put(-3,3){\epsfig{file=IQLAS.MET,height=50mm}}
\put(78,3){\epsfig{file=Cvallnew.eps,height=50mm}}
\put(3,49){\makebox(0,0)[cc]{(a)}}
\put(154,49){\makebox(0,0)[cc]{(b)}}
\end{picture}
\caption{(a) Three dimensional plot of the contribution to the
specific heat coming from the soliton
condensate
as function of
$\tau$ and $Q$. The presence of the condensate
leads to decrease in the specific heat. (b) The total specific heat of
the
system for different values of the parameter $Q$.
Each of the lines 1-5 merges with the line 0 ($Q=0$) at the
commensurate-incommensurate transition temperature.}
\label{Cvalltogether}
\end{figure}
The first two coefficients are given by
\begin{eqnarray}
K_{\rm 0} = \tau \quad , \qquad
K_{\rm 1} = 2 \frac{\Gamma(\tau)}{\Gamma(-\tau)}
\frac{\Gamma(5/2-\tau)}{\Gamma(1/2+\tau)}
\frac{\tau}{(2\tau-1)(3-2\tau)} \quad .
\end{eqnarray}
Putting them together in Eq. (\ref{eq22}) we get
\begin{eqnarray}
F(Q) -F(0) & = & \frac{\rho_{\rm c} Q^2}{2}- T\ \frac{m_{\rm s}^2}{4}
\cot\left[\frac{\pi}{2 (1 - \tau)} \right]
-T\left(\frac{Q}{4 \tau}\right)^2 \frac{\tau}{\pi}
\\ [3mm]
& & -T\ \frac{2}{\pi}\left(\frac{Q}{4 \tau}\right)^2
\frac{\Gamma(\tau)}{\Gamma(-\tau)}\frac{\Gamma(5/2-\tau)}{\Gamma(1/2+\tau)}
\frac{\tau}{(2\tau-1)(3-2\tau)}\left[\frac{2 m_{\rm
s}\sqrt{\pi}}{Q}\frac{\Gamma\left(\frac{1}{2(1-\tau)}
\right)}{\Gamma\left(\frac{\tau}{2(1-\tau)}\right)}\right]^{4(1-\tau)} \ .
\nonumber
\end{eqnarray}
The second term on the right hand side of the above equation gives the
free energy of the system in absence of the field ${\bf Q}$ (with
opposite sign) and cancels with $F(0)$ of the left hand side.
On the other hand the first and the third terms on the right
cancel each other (keeping in mind that $T_{\rm c}=8 \pi \rho_{\rm c}$).
The total free energy of the system in the presence of the field ${\bf
Q}$ is given, to this order of approximation, by:
\begin{eqnarray}
F(Q) & = & -T\ \frac{2}{\pi}\left(\frac{Q}{4 \tau}\right)^2
\frac{\Gamma(\tau)}{\Gamma(-\tau)}\frac{\Gamma(5/2-\tau)}{\Gamma(1/2+\tau)}
\frac{\tau}{(2\tau-1)(3-2\tau)}\left[\frac{2 m_{\rm s}
\sqrt{\pi}}{Q}\frac{\Gamma\left(\frac{1}{2(1-\tau)}
\right)}{\Gamma\left(\frac{\tau}{2(1-\tau)}\right)}\right]^{4(1-\tau)}
\nonumber \\ [3mm]
& = &
-Q^{-2+4\tau}\ \frac{T_{\rm c}}{8 \pi}
\frac{\Gamma(\tau)}{\Gamma(-\tau)}\frac{\Gamma(5/2-\tau)}{\Gamma(1/2+\tau)}
\frac{1}{(2\tau-1)(3-2\tau)}\left[2 m_{\rm s} \sqrt{\pi} \
\frac{\Gamma\left(\frac{1}{2(1-\tau)}
\right)}{\Gamma\left(\frac{\tau}{2(1-\tau)}\right)}\right]^{4(1-\tau)}
\ .
\end{eqnarray}
The expansion is valid only for small values of $\tau < 1/2$, where the
free energy of the system in absence of the field $Q$ is given by
$F_{\rm 1}$ of Eq. (\ref{F1}).
\section{Sinh-Gordon Model; Thermodynamic Instability}
Now we consider the sinh-Gordon model. Here the exact
solution was suggested by Fateev \cite{Fateev} who has taken the
sine-Gordon
two-body S-matrix for the first breathers and changed the sign of the
coupling constant $\beta^2$ in it. Comparing Eqs.(\ref{sg}, \ref{sh}) we
see that to get the sinh-Gordon action out of the sine-Gordon one we have to
change $\beta \rightarrow i\beta$ and $m \rightarrow - m$.
Doing this substitution in the expression for
the free energy (see also
\cite{Fateev}), we get
\begin{eqnarray}
F = - mI(T/T_{\rm c})(1 + T/T_{\rm c})\left[\frac{T_{\rm c}
\Gamma(1 - T/T_{\rm c})}{\pi
m\Gamma(1 + T/T_{\rm c})}\right]^{\frac{T}{T + T_{\rm c}}} \label{Fsinh}
\quad ,
\end{eqnarray}
\begin{eqnarray}
I(x) = \exp\left\{2\int_0^{\infty} \frac{dt}{t}\left[- \frac{\cosh^2(xt)\sinh(xt)}{\sinh
t\cosh((1 + x)t)} + xe^{-2t}\right]\right\} \label{I}
\quad ,
\end{eqnarray}
where $T_{\rm c} = 8\pi\rho_{\rm s}$. This expression is valid for
$T < T_{\rm c}$.
The integral (\ref{I}) can be calculated to give the expression
\begin{equation}
I(x)=\frac{\Gamma\left(\frac{1}{2}+\frac{x}{2(1+x)}\right)}
{\Gamma\left(\frac{1}{2}-\frac{x}{2(1+x)}\right)}
\frac{\Gamma(x)}{\Gamma(-x)}
\frac{\Gamma\left(-\frac{x}{2(1+x)}\right)}
{\Gamma\left(\frac{x}{2(1+x)}\right)}
\quad ,
\end{equation}
with $0<x<1$. $I(x)$ is monotonically decreasing function taking values in
the interval $(0,1)$.
In the light of the following discussion it will be instructive also
to have the expression for the mass (the inverse correlation length)
of the sinh-Gordon theory. To get this expression one has to change
$\beta$ to $i\beta$ for the sine-Gordon mass which corresponds to
reversal of the sign of $T_{\rm c}$ in Eq.(8):
\begin{eqnarray}
m_{\rm s} a = \frac{4\sqrt\pi}{\Gamma\left(\frac{T_{\rm c}}{2(T + T_{\rm c})}
\right)\Gamma\left(1 +
\frac{T}{2(T + T_{\rm c})}\right)}\left[\frac{\pi m}{T_{\rm c}}
\frac{\Gamma\left(1 + T/T_{\rm c}\right)}{\Gamma\left(1 - T/T_{\rm c}\right)}
\right]^{\frac{T_{\rm c}}{2(T + T_{\rm c})}} \label{masssinh}
\quad .
\end{eqnarray}
(This expression coincides with the mass of the sine-Gordon {\it breather}
after the substitution $T \rightarrow -T$).
We would like to attract the reader's attention to the fact that $m(T)$
never diverges at $T < T_{\rm c}$ and actually goes to zero at $T
\rightarrow T_{\rm c}$.
For $\tau = (1 - T/T_{\rm c}) \ll 1$ we obtain from
(\ref{I}) $I(\tau) \sim \tau$ and substituting this into
Eq. (\ref{Fsinh}) we get at $\ln(T_{\rm c}/\pi m) \gg 1$
\begin{equation}
F \sim - (1 - T/T_{\rm c})^{1/2}e^{- (1 - T/T_{\rm c})\ln(T_{\rm c}/\pi m)}
\quad .
\end{equation}
It is easy to see that the specific heat becomes negative at
\begin{equation}
(1 - T/T_{\rm c}) < (1 + \sqrt 2) [2\ln(T_{\rm c}/\pi m)]^{-1} \label{inst}
\quad ,
\end{equation}
such that and at temperatures sufficiently close
to $T_{\rm c}$ the model is thermodynamically unstable.
This thermodynamic instability is not completely unexpected.
It occurs in the strong coupling regime when the coupling
constant $T/T_{\rm c}$ approaches its critical value 1. Let the reader recall
that at this value of the coupling constant
the ultraviolet limit of the sinh-Gordon model - the Liouville model becomes
unstable \cite{zamol2} (its central charge approaches the value of 25).
Another indication of the instability comes from
the temperature dependence of the inverse correlation length given by
Eq. (\ref{masssinh}): at $T \rightarrow T_{\rm c}$ the mass becomes zero which
one does not expect to happen to the sinh-Gordon model which can be thought as
the Gaussian theory perturbed by a strongly relevant operator. One
can check that at the
instability point (\ref{inst}) the mass of the sinh-Gordon particle
is still much larger than its value at $ T = 0$.
\section{Conclusions}
The summary of our results on the classical sine-Gordon model is well
represented by Fig. 6. The specific heat has a peak well below the
Kosterlitz-Thouless transition and the temperature dependence becomes
even more complicated in the incommensurate phase. We suppose that
all these features are detectable experimentally in the relevant
systems like the one described in \cite{MacDonald}.
\newpage
{\bf Acknowledgments}
E. P. would like to thank Prof. H. D. Dahmen for the hospitality he offered
during the stay at the University of Siegen, Germany, where part of this work
was done. He also would like to thank Dave Allen, Joseph Betouras,
Pedro Ferreira and especially Tilo Stroh for many helpful discussions.
A. M. T. is grateful to H. Saleur, F. Smirnov V. Fateev
and above all to Yu Lu
for stimulating discussions and valuable remarks.
|
\section{Introduction}
\label{sec:intro}
Elliptical galaxies in the local universe are known to populate only a
two-dimensional manifold, known as the Fundamental Plane (hereafter
FP; Djorgovski \& Davis 1987; Dressler et al.~1987), in the
three-dimensional space defined by effective radius $R_{\tx{e}}$, mean
surface brightness within the effective radius SB$_{\tx{e}}$
(SB$_{\tx{e}}=-2.5\log\langle I \rangle_{\tx{e}}+$const; I is the
surface brightness in linear flux), and central velocity dispersion~$\sigma$. The FP is described by:
\begin{equation}
\label{eq:FP} \log R_{\tx{e}} = \alpha \log~\sigma + \beta~\tx{SB}_{\tx{e}} + \gamma,
\end{equation}
$R_{\tx{e}}$ is in kpc (while $r_{\tx{e}}$ is in arcsecs), $\sigma$ in
km~$\tx{s}^{-1}$, SB$_{\tx{e}}$ in $\tx{mag arcsec}^{-2}$ and $H_0$ is hereafter
assumed to be 50 $h_{50}$ km~$\tx{s}^{-1}$ $\tx{Mpc}^{-1}$. Typical values of
the coefficients are for example $\alpha=1.25$, $\beta=0.32$ and
$\gamma=-8.895$ in Johnson B (Bender et al.\ 1998; hereafter B98). It is still
unclear whether the thickness of the FP is intrinsic, i.e., due to
scatter in the properties of elliptical galaxies, or is due to
observational errors.
The existence of the empirical scaling described by the FP has strong
implications in terms of galactic evolution and formation theories.
For example, the FP suggests that the distributions of dark and
luminous matter are related, since $\sigma$ depends on the total
gravitational potential, whereas SB$_{\tx{e}}$ and $R_{\tx{e}}$ trace
only the stellar mass. The FP can be explained in terms of homology,
of the virial theorem, and of the existence of a well-defined relation
between luminosity and mass, $ L \propto M^{\eta}$ (Faber et al.,
1987; see also van Albada, Bertin \& Stiavelli~1995). This
interpretation requires a relation between the derived values of
$\alpha$ and $\beta$, namely $\alpha-10\beta+2=0$
($\eta=0.2~\alpha/\beta$).
Measuring the FP parameters for ellipticals at intermediate redshift $
z \approx 0.3$--$0.8 $ has only recently become feasible (Pahre,
Djorgovski \& de Carvalho 1995; van Dokkum \& Franx 1996; Bender et
al.~1996; Kelson et al. 1997; van Dokkum et al. 1998). This opens up
two important questions which bear on the age, formation history, and
internal properties of elliptical galaxies: {\it i)} how far in the
past does an FP-like relation apply? and {\it ii)} do its parameters
evolve significantly with time? The existence of a tight FP-like
relation at substantial look-back times would suggest the existence of
a universal relation between mass and formation time for ellipticals,
since, e.g., population differences would be amplified as the redshift
increases. Existing measurements of the properties of ellipticals at
intermediate redshifts have concentrated so far on relatively rich
clusters of galaxies, where bright ellipticals are easier to find
\cite{PDdC95,DF96,B96,ZB97,KDFIF,DFKI98} and observations can be
carried out efficiently with multi-object spectrographs. When
comparing, e.g., the FP of intermediate redshift and zero redshift
clusters one must take into account the evolution with redshift of the
population of galaxies in clusters, and environmental effects.
Although the evolution of clusters of galaxies is still somewhat
controversial \cite{Post96}, the galaxy population in rich clusters is
likely to become less uniform (in terms, e.g., of age and metallicity)
with time, due to accretion of isolated galaxies and small groups.
Moreover, the question of whether there are systematic differences
between cluster and field ellipticals (the latter including galaxies
in loose groups or poor clusters) is still open even at low redshift
\cite{dCD92}. A study of the galaxy properties as a function of
look-back time provides a sensitive probe of the possible evolutionary
differences between cluster and field ellipticals. For these reasons,
we have begun a study of the empirical scaling laws of field
ellipticals at intermediate redshift.
We have started out by defining a sample of ellipticals which is not
biased in favour of rich clusters. Details on the selection process
are given in Section~\ref{sec:sample}. Sections~\ref{sec:foto} and
\ref{sec:spec} describe the data reduction and the error analysis.
Measurements at non-zero redshift have to be carried out with
photometric filters and effective apertures that differ from those
used to derive the local scaling laws; Section~\ref{sec:corr}
describes the conversion of our measurements to the standard
quantities. The main results are reported in Section~\ref{sec:res},
where our measurements are compared to local samples and intermediate
redshift cluster samples. A summary is given in Section~\ref{sec:conc}.
\section{Sample selection}
\label{sec:sample}
With this study we aim at addressing two questions: {\it i)} does the FP of field ellipticals evolve with redshift? and {\it ii)} do, at any
given redshift, the field ellipticals have FP parameters different from those
found in rich cluster environments? For this we need to select a
primary sample of galaxies biased against rich cluster membership
at intermediate redshift, where the project is feasible and the look-back time
($\approx 4$--$5$ Gyrs) is high enough that evolution can be noticeable.
The targets used in this study have been chosen among a sample of
random ellipticals found in the WFPC2 parallel images collected by
the Medium Deep Survey \cite{MDS}. The target pool has
been selected originally from available HST images in the
appropriate RA, $\delta $ range according to the following criteria:
\begin {enumerate}
\item Apparent magnitude $ I < 20.5 $
\item Morphology clearly defined as elliptical, with an apparent
effective radius $ r_{\tx{e}} > 0\farcs 5 $.
\item $ V - I $ colour in agreement with the empiric relation for
ellipticals in the redshift range $ 0.2 \leq z \leq 0.5 $ ($ 1.1 < V-I
< 1.7 $)
\end {enumerate}
Magnitudes and colours above are as defined by the Medium Deep Survey
group \cite{MDS}; $ V $ and $ I $ are the magnitudes in the WFPC2
filters F606W and F814W, computed in the WFPC2 Flight system (Holtzman
et al.~1995), and the effective radius $ r_{\tx{e}} $ is the value found from
the two-dimensional image fitting carried out by the MDS. All
photometric parameters have been rederived here with two different
techniques (see Section~\ref{sec:foto}) to ensure a
self-consistent treatment.
Individual targets were then selected from this pool of approximately 25
candidates on the basis of convenience and availability during each
observing night; in the spirit of this exploratory study, we did not
attempt further to achieve uniformity or completeness within our pool of
candidates.
Since we did not explicitly exclude members of poor clusters (no rich
clusters were observed in these random fields), the ellipticals in our
sample should be representative of a random, magnitude-selected sample
of non-rich-cluster objects. In fact, four of the six objects for
which we have obtained FP parameters happen to be members of a group
(or a poor cluster), which was randomly observed in the available HST
images. Future observations will enable us to achieve a better level
of completeness and thus to better characterise our sample.
Our colour selection criterion will bias against actively star-forming
ellipticals, which are known to occur, albeit infrequently, in
complete samples \cite{CFRS}. We plan to relax the blue cutoff in
future observations.
\subsection{Target list}
The characteristics of the objects we observed are summarised in
Table~\ref{tab:obj}. The final sample used in the discussion in
Section~\ref{sec:res} is composed of the galaxies successfully fitted
with the two different photometric techniques, isophotal profile fit and
two-dimensional fit, and with measured kinematics.
The photometry of {\bf C, L, M, N} was of low quality due to
companion galaxies, stars in the field of view and the faintness of the
objects. The 2D fit was performed only on the galaxies with a high
quality isophotal profile. We were able to determine high quality
photometric parameters for {\bf G} by subtracting a model of its
companion {\bf H} from the image before the 2D fit.
\begin{table}
\caption{Galaxy identification. (1) Adopted names, (2) HST Medium
Deep Survey field name, (3) WFPC2 chip, (4) ID within the field, (5)
and (6) pixel positions of the center of the galaxy, (7) Sizes of the
F606W and F814W images used in the 2D fits and displayed in Figures 2
and 3. The image used for {\bf I} in F814W was $5\farcs 4 \times 4
\farcs 8$. Dashes in Column 7 indicate that the 2D fit was not
performed.}
\label{tab:obj}
\begin{tabular}{c c c c c c c}
galaxy & field & chip & n & x & y & size \\
(1) & (2) & (3) & (4) & (5) & (6) & (7) \\
\hline
A & ut800 & 4 & 1 & 324 & 444 & $18''\times 18''$ \\
B & ut800 & 4 & 2 & 265 & 741 & $8''\times 8''$ \\
C & ur610 & 3 & 1 & 678 & 691 & -\\
D & ur610 & 3 & 2 & 723 & 642 & $9\farcs 5\times 6\farcs 4$\\
E & u5405 & 2 & 1 & 335 & 324 & $8''\times 7''$\\
F & u5405 & 2 & 2 & 332 & 221 & $5\farcs 4 \times 6 \farcs 4$\\
G & u5405 & 2 & 3 & 192 & 288 & $10'' \times 10''$\\
H & u5405 & 2 & 4 & 194 & 255 & -\\
I & u5405 & 2 & 5 & 72 & 63 & $5\farcs 2 \times 4 \farcs 3$ \\
L & u5405 & 3 & 6 & 254 & 711 & -\\
M & u5405 & 3 & 7 & 266 & 715 & -\\
N & u5405 & 3 & 8 & 747 & 732 & -\\
O & ust00 & 3 & 1 & 226 & 523 & $5'' \times 5''$\\
P & ust00 & 3 & 2 & 336 & 404 & $5'' \times 5''$\\
Q & ust00 & 3 & 3 & 443 & 298 & $5 \farcs 2 \times 5 \farcs 4$\\
R & urz00 & 2 & 1 & 719 & 242 & -\\
\hline
\end{tabular}
\end{table}
\section{Photometry}
\label{sec:foto}
The images are taken from the Medium Deep Survey of the
Hubble Space Telescope \cite{MDS}. For each field there are images
from WFPC2 through filters F606W and F814W. Table~\ref{tab:phlog}
lists total exposure times and number of exposures.
\begin{table}
\caption{Photometric data: number of exposures (nexp) and total exposure times (texp).}
\label{tab:phlog}
\begin{tabular}{|l|l|r|r|}
field & filter & nexp & texp (s)\\
\hline
u5405 & F606W & 1 & 800 \\
& F814W & 1 & 800 \\
ur610 & F606W & 2 & 1500 \\
& F814W & 2 & 1600 \\
urz00 & F606W & 3 & 5400 \\
& F814W & 4 & 8400 \\
ust00 & F606W & 10 & 16500 \\
& F814W & 11 & 23100 \\
ut800 & F606W & 2 & 1430 \\
& F814W & 3 & 4430 \\
\hline
\end{tabular}
\end{table}
\subsection{Reduction}
The basic reduction was done at the Space Telescope Science Institute
using the standard pipeline with the best reference files available. As
bias and dark files we used the superbias and the superdark from the
Hubble Deep Field \cite{HDF} for their high signal-to-noise ratio. The
cosmic ray removal for the fields with multiple images was done using
the {\sc iraf/stsdas} task {\sc crrej}, with a cutoff at 5 standard deviations
from the mean. Since
only one exposure per filter was available for u5405, the original image
of this field was compared with a smoothed version (using the {\sc
midas} command {\sc filter/cosmic}). In all cases we checked by eye
that the pixels removed in the area of the targets were only those
affected by cosmic rays.
For u5405, chip 2 of WFPC2 is affected by a bad column, very close to
the center of galaxies {\bf E} and {\bf F} (see Figure~\ref{fig:2D}).
We interpolated through it in the outer parts of the galaxies, but we
chose not to modify the 4-5 pixel diameter central zone because it was
too steep for any interpolation to be meaningful (see also
Figure~\ref{fig:profs}). In order to test how much this bad column
affected the determination of the photometric parameters, we fitted
the isophotal profile of {\bf E} excluding the innermost part of the
profile. We tried excluding a 2 pixel and a 4 pixel diameter circle;
the effective radius and the effective surface brightness change
significantly (up to 20\% in $r_{\tx{e}}$), towards the value given by
the 2D fit (smaller $r_{\tx{e}}$ and brighter surface
brightness). These new values are still consistent with the average
value within the internal errors (that are the largest of the sample,
see Subsection~\ref{ssec:fotoerr}). Furthermore it should be noticed
that the derived values for $r_{\tx{e}}$ and SB$_{\tx{e}}$ are highly
correlated and therefore the variation of the combination that enters
the FP, $\log r_{\tx{e}} - \beta \tx{SB}_{\tx{e}}$, is much smaller
than the variation on the single coefficients (see
Subsection~\ref{ssec:fotoerr} and Table~\ref{tab:fotcorr}).
A constant sky contribution was subtracted from each image. The sky
contribution used was the mean of the averages of areas with no evidence
for light sources. This procedure does not take into account smooth and
diffuse background sources, which is one of the main sources of uncertainty
when using curve of growth technique (see Section~\ref{ssec:fotoerr}).
\subsection{Isophotes and two-dimensional fits}
The photometric parameters needed for the FP are the effective radius
$r_{\tx{e}}$, and the effective surface brightness SB$_{\tx{e}}$,
measured as the best $r^{1/4}\,$ light profiles parameters.
In order to obtain the best results and robust error estimates, the
photometric parameters were derived using two independent techniques:
fits to the isophotal luminosity profiles and two-dimensional fits to
the images. Both procedures require a PSF to convolve the models
with. We used synthetic PSFs calculated with Tiny Tim 4.0 \cite{TT4}
with a 15 mas jitter. The quality of the synthetic PSF was checked by
comparing the isophotal profiles of two stars and the profiles of two
PSFs created in the same spot. No significant difference was
noticed. As a double-check we fitted an isophotal profile using both a
synthetic PSF profile and a real star profile: the differences were
less than 1\% in $r_{\tx{e}}$ and less than 0.02 mag in
SB$_{\tx{e}}$. The photometric parameters of each galaxy were derived
independently in the two bandpasses F814W and F606W.
\subsubsection{Isophotes}
\label{sssec:iso}
Isophotes were fitted to the images with the center of the ellipse,
the semiaxes $a$ and $b$, and the position angle as free parameters;
variations of ellipticity ($e$) and position angle ($pa$) with the
``circularised'' radius $\sqrt{ab}$ were allowed. Isophotes were
derived using a version of the {\sc midas} command {\sc fit/ell3}
modified to better deal with steep gradients in the luminosity profile
in the innermost pixels (M{\o}ller, Stiavelli \& Zeilinger, 1995).
The profiles obtained were fitted with an exponential law, an $r^{1/4}\,$ law,
and a linear combination of the two. In all cases, except for {\bf B},
we found that the objects were best fitted by the $r^{1/4}$ law. A
slightly better fit of the light profile of galaxy {\bf B} was obtained
by adding a small exponential component to the $r^{1/4}\,$ law, suggesting that the
bump on the residual might be due to a fainter disk component (see
Figures \ref{fig:profs} and \ref{fig:2Da}). The fit was done using
the least squares fitting software used by Carollo et al.\
\shortcite{MCetal}. For each galaxy we used a specific PSF calculated at its
position. In Figure~\ref{fig:profs} we show the
best-fit $r^{1/4}\,$ laws convolved with the PSFs, superimposed on the data
points.
\begin{figure*}
\mbox{\epsfysize=18cm \epsfbox{profiles3.eps}}
\caption{Isophotal profiles. Crosses are the data points, solid lines
represent the best-fit $r^{1/4}\,$ profile, convolved with the PSF. The upper
data refer to F814W, the lower to F606W. The error bars (which are
smaller than the crosses for the innermost part of the profile) take
into account the sky-subtraction error. The bump on the {\bf B}
profile is possibly the signature of a disk component (see text,
Section \ref{sssec:iso}, and the two dimensional residual in Figure
\ref{fig:2Da}). The central surface brightness decline in {\bf E}
is likely to be caused by the bad column right at the center of the
galaxy.}
\label{fig:profs}
\end{figure*}
\subsubsection{Two-dimensional fits}
This technique consists in fitting a two-dimensional model of an
elliptical galaxy, as seen on the CCD chip, to the WFPC2 image. The
model galaxy has an $r^{1/4}\,$ luminosity profile, with fixed position angle
and ellipticity (we used the value derived from the isophotal fit near
the effective radius). Each fit has four free parameters: the
effective surface brightness, the effective radius, and the position
of the centre. We have two sets of results, depending on how the fit
was performed (see (iv) below). The software we developed works as
follows:
\begin{enumerate}
\item The code generates a 2D projected image of the galaxy using the
$r^{1/4}\,$ law on a subsampled grid. The pixel size is chosen to be
approximately ten times smaller than the galaxy effective radius in
order to make discretisation problems negligible.
\item The resulting ``ideal'' model is rebinned to the WFPC2 chip scale.
\item The rebinned model is convolved with the Tiny Tim \cite{TT4} PSF.
The Tiny Tim (non resampled) PSF includes the effects of diffraction
(i.e. of the telescope and camera optics) and the Pixel
Response Function (i.e. the spread caused by electron diffusion in the
WFPC2 CCDs).
\item This model is fitted to the data either by least $\chi^2$ or by
least squares, using a simplex algorithm to find the minimum ({\sc
amoeba}, Press et al.~1992).
\end{enumerate}
For each galaxy we ran the code twice, minimising both the $\chi^2$
and the unweighted residuals (least squares). In the case of {\bf G}
the fit was not straightforward because of the companion {\bf H}.
Therefore we subtracted from the data, before performing the fit, a
model for {\bf H}, obtained from an $r^{1/4}\,$ profile with $r_{\tx{e}}$,
SB$_{\tx{e}}$, $pa$, and $e$ taken from the isophotal profile fit.
The result was stable with respect to variations of the parameters of
{\bf H}. In Figure~\ref{fig:2D} and \ref{fig:2Da} the original WFPC2
image and the fit residual are shown for all the galaxies fitted with
the 2D fits.
\begin{figure*}
\begin{center}
A
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwA6.ps}}
\mbox{\epsfysize=3cm \epsfbox{A6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwA8.ps}}
\mbox{\epsfysize=3cm \epsfbox{A8r.ps}}
}
D
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwD6.ps}}
\mbox{\epsfysize=3cm \epsfbox{D6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwD8.ps}}
\mbox{\epsfysize=3cm \epsfbox{D8r.ps}}}
E
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwE6.ps}}
\mbox{\epsfysize=3cm \epsfbox{E6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwE8.ps}}
\mbox{\epsfysize=3cm \epsfbox{E8r.ps}}}
F
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwF6.ps}}
\mbox{\epsfysize=3cm \epsfbox{F6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwF8.ps}}
\mbox{\epsfysize=3cm \epsfbox{F8r.ps}}}
GH
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwGH6.ps}}
\mbox{\epsfysize=3cm \epsfbox{GH6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwGH8.ps}}
\mbox{\epsfysize=3cm \epsfbox{GH8r.ps}}}
I
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwI6.ps}}
\mbox{\epsfysize=3cm \epsfbox{I6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwI8.ps}}
\mbox{\epsfysize=3cm \epsfbox{I8r.ps}}}
\end{center}
\caption{WFPC2 images and 2D fit residuals for the galaxies with
measured kinematics: {\bf A},{\bf D}, {\bf E}, {\bf F}, {\bf G}, {\bf
I}. From the left to the right we display the F606W image, the F606W
residual, the F814W image and the F814W residual. The images (and
their sizes) are the ones used for the 2D fits, see
Table~1. Galaxies {\bf A}, {\bf D} show a residual in the
innermost pixels which is in agreement with the deviation from the $r^{1/4}\,$
law in the luminosity profile. The bad column in the centre of {\bf
E} is clearly evident in the residual. {\bf F} shows a peculiar pot
handle structure just to the upper left of the bad column.}
\label{fig:2D}
\end{figure*}
\begin{figure*}
\begin{center}
B
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwB6.ps}}
\mbox{\epsfysize=3cm \epsfbox{B6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwB8.ps}}
\mbox{\epsfysize=3cm \epsfbox{B8r.ps}}}
O
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwO6.ps}}
\mbox{\epsfysize=3cm \epsfbox{O6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwO8.ps}}
\mbox{\epsfysize=3cm \epsfbox{O8r.ps}}}
P
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwP6.ps}}
\mbox{\epsfysize=3cm \epsfbox{P6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwP8.ps}}
\mbox{\epsfysize=3cm \epsfbox{P8r.ps}}}
Q
\mbox{
\mbox{\epsfysize=3cm \epsfbox{bwQ6.ps}}
\mbox{\epsfysize=3cm \epsfbox{Q6r.ps}}
\mbox{\epsfysize=3cm \epsfbox{bwQ8.ps}}
\mbox{\epsfysize=3cm \epsfbox{Q8r.ps}}}
\end{center}
\caption{WFPC2 images and 2D fit residuals for the galaxies without measured
kinematics: {\bf B},{\bf O},{\bf P},{\bf Q}. In the residual of {\bf B} a
spiral pattern is recognizable.}
\label{fig:2Da}
\end{figure*}
\subsubsection{Results}
All galaxies are well fitted by the $r^{1/4}\,$ law (de Vaucouleurs 1948)
with the exception of galaxy {\bf B} which shows face-on spiral-like
residuals and might be an early-type spiral.
\begin{figure}
\mbox{\epsfysize=8cm \epsfbox{check2Dprof.eps}}
\caption{Comparison of
photometric parameters from the isophotal profile fit (x-axis) with
those from one of the 2D fits (the least $\chi^2$, y-axis). Subscripts 6 and
8 indicate filters F606W and F814W.}
\label{fig:cf2D}
\end{figure}
Standard magnitudes are obtained from the data following the Holtzman
et al.~\shortcite{WFPC2} calibration. In Figure~\ref{fig:cf2D} we
plot the photometric parameters obtained from one of the 2D fits (the
least $\chi^2$) versus those given by the isophotal profile fit. The
results are in very good mutual agreement, even though the procedures
differ in some important details, e.g., the isophotal profile fit is
in magnitudes and allows for $pa$ and $e$ variation, while the 2D fits
are in counts and have fixed $pa$ and $e$. As final value of the
parameters we use the average of the three results; the photometric
parameters are shown in Tables~\ref{tab:fotores} and
\ref{tab:fotores2}. Integrated luminosities are those of the
best-fitting $r^{1/4}$ law.
\begin{table}
\caption{Photometric parameters through filter F606W. Errors (see
Section 3.4) do not include sky-subtraction contributions, which we
estimate to be 0.5 per cent on $r_{\tx{e}}$, 0.05 mag
$\tx{arcsec}^{-2}$ on SB$_{\tx{e}}$ and 0.02 mag on m (see text for
details).}
\label{tab:fotores}
\begin{tabular}{cllllll}
galaxy & $r_{\tx{e}}$ & $\delta r_{\tx{e}}$ & SB$_{\tx{e}}$ & $\delta \tx{SB}_{\tx{e}}$ & m & $\delta \tx{m}$ \\
\hline
A & 1.65 & 0.13 & 20.64 & 0.13 & 17.57 & 0.12 \\
B & 0.588 & 0.038 & 19.73 & 0.11 & 18.90 & 0.05 \\
D & 0.793 & 0.042 & 20.99 & 0.09 & 19.5 & 0.03 \\
E & 0.794 & 0.071 & 21.60 & 0.13 & 20.12 & 0.06 \\
F & 1.145 & 0.017 & 22.56 & $<0.01$ & 20.27 & 0.03 \\
G & 1.589 & 0.064 & 21.81 & 0.07 & 18.81 & 0.01 \\
I & 0.636 & 0.046 & 20.92 & 0.13 & 19.92 & 0.04 \\
O & 0.795 & 0.011 & 23.15 & 0.05 & 21.65 & 0.02 \\
P & 0.526 & 0.029 & 22.20 & 0.11 & 21.66 & 0.01 \\
Q & 0.612 & 0.018 & 23.20 & 0.06 & 22.27 & 0.02 \\
\hline
\end{tabular}
\end{table}
\begin{table}
\caption{Photometric parameters through filter F814W (same as Table 2).}
\label{tab:fotores2}
\begin{tabular}{cllllll}
galaxy & $r_{\tx{e}}$ & $\delta r_{\tx{e}}$ & SB$_{\tx{e}}$ & $\delta \tx{SB}_{\tx{e}}$ & m & $\delta \tx{m}$\\
\hline
A & 1.99 & 0.15 & 19.92 & 0.19 & 16.44 & 0.13 \\
B & 0.548 & 0.014 & 18.58 & 0.05 & 17.89 & 0.04 \\
D & 0.645 & 0.058 & 19.29 & 0.16 & 18.27 & 0.04 \\
E & 0.872 & 0.080 & 20.31 & 0.12 & 18.64 & 0.07 \\
F & 1.29 & 0.12 & 21.33 & 0.11 & 18.81 & 0.10 \\
G & 1.419 & 0.075 & 20.29 & 0.09 & 17.53 & 0.02 \\
I & 0.606 & 0.051 & 19.47 & 0.15 & 18.58 & 0.03 \\
O & 0.654 & 0.007 & 21.02 & 0.05 & 19.94 & 0.02 \\
P & 0.502 & 0.018 & 20.48 & 0.08 & 19.98 & $<0.01$ \\
Q & 0.547 & 0.053 & 21.14 & 0.17 & 20.48 & 0.03 \\
\hline
\end{tabular}
\end{table}
\subsection{A cross-check with the growth curve technique}
The growth curve technique is a standard procedure for nearby samples.
Two problems of this technique are: {\it i)} the statistical
dependence between different aperture magnitudes, e.g. in the presence
of light coming from a nearby galaxy or of diffuse background sources,
the systematic error affects the flux within all the apertures; {\it
ii)} the sensitivity to sky-subtraction errors. We measured the
photometric parameters by fitting the growth curve of an $r^{1/4}$
profile to the aperture magnitudes and the results were in reasonably
good agreement with the other two methods.
\subsection{Error analysis}
\label{ssec:fotoerr}
We define as internal error of the parameters the standard deviation
of the results of the two 2D fits and the isophotal profile fit. Since
photon shot noise is negligible in our context, the main source of
error not included in the internal error is the uncertainty due to sky
subtraction (0.5 per cent on $r_{\tx{e}}$, 0.02 mag on m, 0.05 mag
$\tx{arcsec}^{-2}$ on SB$_{\tx{e}}$). We estimated this uncertainty by
measuring how the photometric parameters changed with the sky level,
and we added it quadratically to the internal error, so as to obtain
the total error; more precisely we used the variation of the
photometric parameters when the sky is shifted by one standard
deviation, calculated as the scatter of the sky levels measured in
different areas (a typical value is 2\%). The photometric parameters
$r_{\tx{e}}$ and $ \langle I \rangle_{\tx{ e}} $ are correlated, as
can be seen from the errors calculated directly and reported in
Table~\ref{tab:fotcorr}. The particular combinations of effective
radius and surface brightness are chosen because they enter the
definition of the $\kappa$ space (Bender, Burstein \& Faber 1992).
Also the error on the colour ($m_{\tx{\small F606W}}-m_{\tx{\small
F814W}}$) is different from the sum of errors on the single filter
magnitudes (see Table~\ref{tab:fotcorr}) since the errors in the two
filters are correlated: it generally happens that when one method
(e.g. 2D $\chi^2$) gives a lower magnitude than another (e.g. profile
fits), the same thing is repeated in both filters.
\begin{table}
\caption{Photometric parameters II. Errors are shown as examples of
the correlation of the photometric uncertainties. The quantity
$colour$ is defined as $m_{\tx{\small F606W}}-m_{\tx{\small F814W}}$.}
\begin{tabular}{ccccc}
galaxy & colour & $\delta$colour & $\delta \langle I \rangle_{\tx{e}}^2
r_{\tx{e}}^{-1}/ \langle I \rangle_{\tx{e}}^2 r_{\tx{e}}^{-1}$ & $\delta \langle I \rangle_{\tx{e}} r_{\tx{e}}/ \langle I \rangle_{\tx{e}} r_{\tx{e}}$ \\
\hline
A & 1.127 & 0.012 & 0.23 & 0.089 \\
B & 1.015 & 0.012 & 0.24 & 0.047 \\
D & 1.226 & 0.018 & 0.19 & 0.027 \\
E & 1.486 & 0.013 & 0.29 & 0.034 \\
F & 1.463 & 0.068 & 0.025 & 0.011 \\
G & 1.273 & 0.007 & 0.15 & 0.028 \\
I & 1.333 & 0.005 & 0.28 & 0.044 \\
O & 1.705 & 0.006 & 0.091 & 0.028 \\
P & 1.678 & 0.013 & 0.26 & 0.045 \\
Q & 1.789 & 0.051 & 0.15 & 0.031 \\
\hline
\end{tabular}
\label{tab:fotcorr}
\end{table}
\section{Spectroscopy}
\label{sec:spec}
The spectroscopic observing run took place in the period April 18-22
1996 at the 3.6-m ESO telescope. Long slit ($ 1 \farcs 5 $ slit width)
spectroscopy was obtained with the EFOSC spectrograph, with the Orange
150 grism in the spectral range between approximately 5200 and 7000
\AA\, (512$\times$512 CCD, with pixel size 30 $\mu m$ equivalent to
about $ 0 \farcs 57 $). The slit length of about $ 180 \arcsec $
projects to only about 320 pixels on the CCD. For each target we
obtained multiple exposures. The estimated seeing, measured with R-band
acquisition images, was $ 0 \farcs 8 $ FWHM. Table~\ref{tab:splog}
gives the basic observation log.
\begin{table}
\caption{Spectroscopic data: number of observations (nexp), total exposures
times (texp), and average signal-to-noise ratio per pixel (S/N).}
\label{tab:splog}
\begin{tabular}{|l|r|r|r|}
galaxy & nexp & texp (s) & S/N \\
\hline
A & 4 & 7200 & 65 \\
B & 4 & 7200 & 40 \\
D & 4 & 12000 & 22 \\
E & 5 & 13200 & 17 \\
F & 5 & 13200 & 12 \\
G & 3 & 10800 & 35 \\
I & 3 & 10800 & 22 \\
LM & 2 & 4800 & - \\
N & 2 & 4800 & - \\
O & 10 & 36000 & 7 \\
P & 10 & 36000 & 6 \\
Q & 10 & 36000 & - \\
\hline
\end{tabular}
\end{table}
The slit positions were carefully chosen to include as many objects
(two or three at a time) as possible. Every night we obtained an
exposure with the He-Ar lamp at zenith; before and after every galaxy
spectrum we took a He lamp spectrum as a check of the dispersion
relation shifts. We observed two spectrophotometric standard white
dwarfs in order to perform relative-flux calibration of the spectra.
The identification of our targets as normal, non star-forming,
ellipticals was confirmed by the absence of emission lines in their
spectra.
\subsection{Reduction}
A superbias file was obtained by averaging about a hundred bias
images and was subtracted from the data. Dome flatfield exposures
were used to produce a flatfield along the wavelength direction (CCD
columns). The slit function was obtained from sky frames and was
found to be independent of wavelength. Dark current subtraction was
done by producing a dark frame from the average of 8 darks 1 hour long. Since this mean dark did not contain structures, we only
subtracted the mean value.
Throughout the reduction and analysis we have avoided (non-linear)
rebinnings. To this end, we did the reduction before wavelength
calibration. The extraction of the spectra was a crucial point of the
reduction process, given the faintness of the signals. Averaging too
many rows would have meant too much noise, using only the central row
would have meant losing important signal. As we could not simply
extract a fixed pixel width (along the spatial direction) because of
optical distorsions, we used a code developed by M{\o}ller \&
Kj{\ae}rgaard (1992) to obtain a one-dimensional spectrum, extracted
with optimal signal-to-noise weighting. The code also removes cosmic
rays, identified as pixels five standard deviations away from the
expected signal. All spectra were carefully inspected by eye before
and after extraction to check the proper cosmic ray removal.
Given the faint signal of the spectra, we took special care during sky
subtraction. For each galaxy we extracted two sets of columns, to the
left and to the right of the galaxy. The columns were selected to be
as close as possible to the galaxy to limit optical distorsion
problems. Columns of one set were combined neglecting the highest and
lowest pixels to remove cosmic rays. The result was eye-inspected and
then the two spectra obtained were averaged to obtain a high
signal-to-noise sky spectrum. This procedure allowed us to check the
optical distorsion by comparing the emission lines between the left
and the right spectra. The sky spectra obtained had broader lines
than the single column ones, because we combined spectra
wavelength-shifted by optical distorsion. Sky contribution was
subtracted during spectrum extraction by deconvolving the artificial
additional width. The resulting one-dimensional spectra were combined
to obtain the final galactic spectrum, after verifying that the
relative shifts of the positions along the y-axis of the emission
lines were negligible. The reduced spectra are shown in Figure
\ref{fig:spectra}. The average signal-to-noise ratio per pixel of the
spectra is listed in Table \ref{tab:splog}. The average
signal-to-noise ratio ranges from 65 per pixel for the brightest
galaxy ({\bf A}) to 12 per pixel for the faintest galaxy for which we
obtained a velocity dispersion ({\bf F}); the estimated error is
accordingly larger for {\bf F} than for {\bf A}. For comparison,
Davies et al. (1987) in their low redshift study, with a total
signal-to-noise ratio equivalent to 5$\cdot10^5$ photons, estimate the
random error to be about 10 \%. The analysis in Section \ref{sec:res}
is done both with the entire sample and excluding the two galaxies
{\bf E} and {\bf F} with the lowest signal-to-noise spectra among
those with measured velocity dispersion.
\subsection{Wavelength calibration and measurement of the instrumental
resolution}
\label{sec:wl}
We found the dispersion relation for each He-Ar lamp using the {\sc
midas gui/long} software. The different calibrations were all within
the errors. We used short He lamp images to check that shifts between
single exposures were negligible. The He-Ar lamps were also used to
measure the instrumental resolution: the line shapes were fitted with
Gaussian functions to obtain line widths as a function of CCD pixel
number, and, via dispersion relation, of wavelength. The FWHM is independent of $\lambda$ within the measurement scatter and it was determined to be:
\begin{equation} \Delta
\lambda = (8.39\pm 0.26) \tx{\AA},
\label{res}
\end{equation}
Translated into velocities it corresponds to a kinematic resolution
ranging from 206 to 155 km~$\tx{s}^{-1}$\, from the blue to the red. The 3 \%
error is due to the scatter in the measurement (see also Section
4.4). The instrumental resolution measurement is very important for the
kinematic fit, as we discuss in Section 4.3. Therefore we carefully
examined all lines, to avoid ghost lines which were found and which
would have affected the measurement. We checked the resolution by
measuring the width of the sky lines and the results were consistent
with the ones found with the He-Ar lamp (8.11$\pm$0.39 \AA).
For comparison, the instrumental resolution of the Lick/IDS
spectrograph was between 8-10 \AA\, at shorter wavelengths (see
e.g. Dalle Ore et al., 1991). The velocity dispersions we find are
larger than half the instrumental resolution, which is commonly
considered as a limit (see e.g. Dressler 1979; Kormendy 1982; Bender,
1990). It has been noticed that when the velocity dispersion becomes
of the order of the kinematic resolution and the signal-to-noise is
low a systematic error may be introduced (see e.g. J\o rgensen, Franx
\& Kj\ae rgaard, 1995). In order to study this effect we broadened
some stellar templates to 150,~200,~250,~300 km~$\tx{s}^{-1}$ and we added noise, thus
producing artificial galaxies with S/N per pixel ranging between 10
and 60. The artificial galaxies were then fitted with the same stellar
templates to measure how well the original velocity dispersions were
recovered. The largest velocity dispersions were recovered within few
percents (2--4~\%) down to S/N=10, while the smaller one (150 km~$\tx{s}^{-1}$)
resulted systematically underestimated by $\sim10$\% at S/N 10. No
systematic trends were noticed above S/N=15. This bias is negligible
to our measurement, as galaxies {\bf E} and {\bf I}, the only with
measured velocity dispersion below 200 km~$\tx{s}^{-1}$ (see Table 7), have high
enough S/N (see Table \ref{tab:splog}), while {\bf F}, the only with
S/N below 15 has $\sigma\approx 200$ km~$\tx{s}^{-1}$.
\begin{figure*}
\mbox{\epsfysize=18cm \epsfbox{spectra.eps}}
\caption{Wavelength (in \AA) calibrated spectra of the galaxies with
measured velocity dispersion. The shaded spectral regions were
affected by strong sky-emission lines. These regions have not been
used in the kinematic fit. The standard deviation per pixel is also
shown (lower curves). The arrows identify some of the main absorption
features, such as Ca H (3968.5 \AA) and K (3933.7 \AA), the G band
(4304.4 \AA), Mg$_b$ (5175.4 \AA) and Na D (5892.5 \AA).}
\label{fig:spectra}
\end{figure*}
\subsection{Kinematics}
\label{ssec:km}
In order to derive a robust measurement of the kinematics of our
sample of galaxies we used two different and independent codes: {\it
i)} the {\sc Gauss-Hermite Fourier Fitting Software} \cite{gh1,gh2},
hereafter GHFF, and {\it ii)} the {\sc Fourier Quotient} \cite{fq},
hereafter FQ, in the version modified by Rose (Dressler, 1979) and
used e.g. by Stiavelli, M{\o}ller \& Zeilinger (1993). The results of
the two methods are mutually consistent, within the estimated
errors. However, the GHFF, allows a reliable estimate of the errors
(see, e.g., Rix \& White, 1992, and our discussion below) while the
version of the FQ we used does not; moreover methods like the GHFF are
known to be less sensitive to template mismatches than the FQ (see
e.g. Bender 1990, and our discussion below). Furthermore the GHFF
provides more insight in the fitting procedure, because the continuum
subtraction, the tapering, and the filtering of both observed and
template spectrum can be checked step by step. Finally it provides a
quality parameter (the $\chi^2$) and the residual spectrum. Therefore
we shall consider the values obtained with the FQ only as a check of
the values obtained with the GHFF. We report the complete results
obtained with both codes (Tables 7 and 8 for the GHFF, Table 9 for the
FQ). The discussion of the FP properties in Sections \ref{sec:res}
and \ref{sec:conc} is based on the GHFF method, but the FQ method
leads to the same conclusions.
\subsubsection{Template spectrum}
We used star spectra from the Jacoby et al.~\shortcite{J84} library as
starting bricks for the template construction. We simulated the
measured experimental resolution, taking into account the
non-negligible width of the original Jacoby spectra (4.5 \AA\,
FWHM), by convolving the redshifted template spectra
with a gaussian of width $\sigma_{t}^2=8.39^2-(4.5(1+$z$))^2$, where
$z$ is the galaxy redshift\footnote{The redshift was first determined
by recognizing the most evident absorption lines and then by iteratively refining the
measurement.}. The choice of the correct spectral
class for the template is critical in order to avoid systematic
errors. Therefore, we decided to fit the galactic spectra with each
of the spectral types from F0 to K7 to be sure to include all the most
relevant spectral types. By setting some fit quality criteria (see the
following section) we selected a set of ``good'' spectral types. The
range of variation of the central velocity dispersion within the
``good'' spectral types provides a check of the robustness of the
result, and at the same time an estimate of the systematic error
possibly induced by residual template mismatches. We avoided the
alternate approach of using synthetic galaxy spectra because of its
heavy model dependence on the underlying stellar population.
\subsubsection{The {\sc Gauss-Hermite Fourier Fitting Software}(GHFF) fit}
The GHFF, developed by R.~P. van der Marel \& M.~Franx (Franx,
Illingworth \& Heckman, 1989; van der Marel \& Franx, 1993) models the
broadening function $B$ as the first terms of a Gauss-Hermite series,
thus allowing one to derive a better estimate of the velocity
dispersion in case of non-Gaussian broadening. The advantages and
caveats of this approach are discussed in detail in van der Marel \&
Franx (1993). In order to achieve a meaningful estimate of the first
non-Gaussian terms however, very high signal-to-noise ratio is
required \cite{gh1,fcq} and generally the studies on the Fundamental
Plane with this software fit only the Gaussian component (see e.g. J\o
rgensen, Franx \& Kj\ae rgaard, 1995). Therefore we used the code in
Gaussian-fitting mode.
The GHFF code provides the redshift ($z_{\tx{ghff}}$\,), the line strength ratio ($\gamma_{\tx{ghff}}$\,) and the velocity dispersion ($\sigma_{\tx{ghff}}$\,). The fit gives also an error value
for the parameters that we will call $\delta v_{\tx{ghff}}$, $\delta
$$\gamma_{\tx{ghff}}$\,, $\delta \sigma_{\tx{ghff}}$.
Generally, the templates within the spectral types G0-K2 provided the
best fit, as could be judged by looking at the $\chi^2$, the residuals
and the broadened template, but no template was clearly a better
approximation than the rest. In order to better understand the
systematic trends, we created some ``artificial galaxies'' by
broadening the templates and adding noise to reproduce a
signal-to-noise ratio of about 12 per pixel. The GHFF with the
templates in the range G0-K2 was able to recover the correct velocity
dispersion, while generally the F stars gave lower sigmas and the K3-7
higher ones. The scatter in the values obtained with different noise
patterns was comparable with the formal error of the fit when the same
star was used both as ``galaxy'' and template, but was higher even in
case of small mismatching. For example a ``galaxy'' produced and
fitted with a G5 spectrum gave on average 96~\% the correct value,
with 15~\% average formal error and a measured scatter of 17~\%; using
a G2 star as template we obtained equally 96~\% of the correct value,
with 22~\% scatter and 17~\% average formal error; with a G6 the same
figures were 104~\%, 21~\% and 19~\%. Therefore we decided to keep
only the G0-K2 interval, where no systematic trend was noticed.
We defined the scatter of the velocity dispersion of the ``good''
templates to be the estimate of the systematic error due to residual
mismatches. In order to obtain results comparable to the FQ ones, we
selected the good fits with criteria applicable also to the FQ (that
in the version we have does not provide the $\chi^2$). We used the
information provided by the code as an indicator of the quality of the
fit. Specifically, we used the line strength ratio $\gamma_{\tx{ghff}}$\, as an
estimator of the mean ratio of the abundances of the absorbers (i.e.,
roughly speaking, the metallicity) and $\delta$$\sigma_{\tx{ghff}}$\, as a measure of
the mismatch in the shape (primarily due to differences in the
spectral type). A fit is considered acceptable if it satisfies the
three conditions: {\it i)} 0.5 $<$ $\gamma_{\tx{ghff}}$\, $<$ 1.5, {\it ii)}
$\sigma_{\tx{gh}}>0$ and {\it iii)} $\delta$$\sigma_{\tx{ghff}}$\,$<$$\sigma_{\tx{ghff}}$\,. As final
value we adopt the average of all acceptable results weighted on the
fit formal errors.
We checked what templates were rejected by
this ``$\chi^2$ blind'' algorithm, and the criteria were very
effective in selecting the best $\chi^2$ and the best matching
templates. The GHFF compared to the FQ produced generally a larger
number of fits within the quality criteria and a lower scatter as a
function of template, thus confirming to be less sensitive to template
mismatches. As a further check we also considered a simple average of
the five best $\chi^2$ and the changes were much smaller than the
error bars. The average values are listed in Table
\ref{tab:cineresGH}. The fit formal errors listed refer to the
template which gave the least $\chi^2$.
We have omitted galaxy {\bf B} in our kinematic discussion since its
kinematic fit does not satisfy our quality criteria and indicates a
very small velocity dispersion, below our resolution limit and
consistent with the morphological appearance as a face-on early spiral
(see Figure~\ref{fig:2Da}).
\begin{table}
\caption{Kinematic results from the GHFF fit I. For each galaxy we list
the average central velocity dispersion ($\sigma_{\tx{ghff}}$\,) together with the
random error ($\delta$ $\sigma_{\tx{ghff}}$\,) and the systematic error ($\Delta$
$\sigma_{\tx{ghff}}$\,). The random error is the fit formal uncertainty for the
best-fitting template, while $\Delta$ $\sigma_{\tx{ghff}}$\, is the scatter of the
values obtained with the ``good'' templates (see text).}
\label{tab:cineresGH}
\begin{tabular}{lccc}
gal & $\sigma_{\tx{ghff}}$ & $\delta\sigma_{\tx{ghff}}$ &
$\Delta$$\sigma_{\tx{ghff}}$\, \\
\hline
A & 232 & 11 & 28 \\
D & 271 & 24 & 16 \\
E & 175 & 28 & 28 \\
F & 195 & 41 & 74 \\
G & 226 & 15 & 25 \\
I & 166 & 21 & 29 \\
\hline
\end{tabular}
\end{table}
\begin{table}
\caption{Kinematic results from the GHFF fit II. Same as Table
\ref{tab:cineresGH} for the redshifts ($z_{\tx{ghff}}$\,) and line strength ratios
($\gamma_{\tx{ghff}}$\,).}
\label{tab:cineres2GH}
\begin{tabular}{lcccccc}
gal & $z_{\tx{ghff}}$\, & $\delta$ $z_{\tx{ghff}}$\, & $\Delta$ $z_{\tx{ghff}}$\, & $\gamma_{\tx{ghff}}$\, & $\delta$ $\gamma_{\tx{ghff}}$\, & $\Delta$ $\gamma_{\tx{ghff}}$\,\\
\hline
A & 0.1462 & 3$\cdot 10^{-5}$& 3$\cdot 10^{-4}$ & 0.90 & 0.02 & 0.14\\
D & 0.3844 & 5$\cdot 10^{-5}$& 1$\cdot 10^{-4}$ & 0.80 & 0.03 & 0.04\\
E & 0.2936 & 6$\cdot 10^{-5}$& 1$\cdot 10^{-4}$ & 0.91 & 0.04 & 0.14\\
F & 0.2933 & 1$\cdot 10^{-4}$& 1$\cdot 10^{-4}$ & 0.95 & 0.10 & 0.19\\
G & 0.2942 & 4$\cdot 10^{-5}$& 1$\cdot 10^{-4}$ & 0.93 & 0.03 & 0.16\\
I & 0.2924 & 4$\cdot 10^{-5}$& 1$\cdot 10^{-4}$ & 0.84 & 0.04 & 0.16\\
\hline
\end{tabular}
\end{table}
\subsubsection{{\sc Fourier Quotient}(FQ) fit}
\label{sssec:fq}
The basic idea of the FQ is the hypothesis that one can
describe a galaxy spectrum G as the convolution $\tx{G}=\tx{B}\circ
\tx{S}$ of an average star template spectrum S with a broadening
function B. The broadening function is purely kinematic, while the
template spectrum includes the broadening due to the instrumental
resolution. The template spectrum has to be a good approximation of
the galaxy spectrum especially in its absorption features because FQ
works on the spectra after subtraction of the continuum. Selecting
the best template is critical to the measurement since, as shown,
e.g., by Bender \shortcite{fqc}, FQ is very sensitive to the spectral
type of the star used as a template. To perform the kinematic fit with
the FQ we needed spectra free of their continuum component. The
continuum was subtracted out by interpolating between regions with no
spectral features.
The version of the FQ code available to us requires a
template and a galaxy spectrum linearly rebinned in wavelength and
then it rebins them in $\log \lambda$. We modified it to avoid the
extra rebinning. The new version requires the one-dimensional galaxy
spectrum as a function of the CCD pixel number and the dispersion
relation given as polynomial coefficients; the rebinning is done
directly in $\log{\lambda}$. The spectral regions affected by the
strongest sky emission lines were masked out.
Free parameters in the fit are the redshift
($z_{\tx{fq}}$\,), the width ($\sigma_{\tx{fq}}$\,), and a normalising
factor ($\gamma_{\tx{fq}}$\,, the line strength). The fit gives also a formal error value
for the parameters that we will call $\delta v_{\tx{fq}}$, $\delta
$$\gamma_{\tx{fq}}$\,, $\delta \sigma_{\tx{fq}}$.
A detailed study of the results shows no correlations between fit
parameters and errors. The only general trend is that best results
are obtained for templates of spectral type G, and they become worse
towards the edges of the range we considered, F0 and K7 (for the
complete results see Treu 1997), consistently with the findings of the GHFF. Generally, the F star templates gave
systematically lower velocity dispersion, while the K3-7 templates
gave systematically higher values (see also the discussion in the GHFF
section). In Table
\ref{tab:cineres} we list the average velocity dispersion, line
strength ratio and redshift, together with the scatter in the results between different templates as for the GHFF.
Consistently with what is
found with the GHFF, the velocity dispersion derived for {\bf B} was
below our resolution limit.
\begin{table}
\caption{Kinematic results from the FQ fit. For each galaxy we list
the average central velocity dispersion, redshift and line strength ratio ($\sigma_{\tx{fq}}$\,,$z_{\tx{fq}}$\,, $\gamma_{\tx{fq}}$\,) together with the
systematic error ($\Delta$ $\sigma_{\tx{fq}}$\,, $\Delta$ $z_{\tx{fq}}$\,, $\Delta$ $\gamma_{\tx{fq}}$\,), that is the scatter of the values
obtained with the ``good'' templates (see text).}
\label{tab:cineres}
\begin{tabular}{lcccccc}
gal & $\sigma_{\tx{fq}}$ & $\Delta$$\sigma_{\tx{fq}}$\, & $z_{\tx{fq}}$ & $\Delta$$z_{\tx{fq}}$\, & $\gamma_{\tx{fq}}$\, & $\Delta$$\gamma_{\tx{fq}}$\,\\
\hline
A & 208 & 39 & 0.1462 & 1$\cdot 10^{-4}$ & 0.92 & 0.16\\
D & 252 & 13 & 0.3844 & 1$\cdot 10^{-4}$ & 0.84 & 0.14\\
E & 211 & 10 & 0.2935 & 2$\cdot 10^{-4}$ & 0.67 & 0.10\\
F & 233 & 74 & 0.2934 & 2$\cdot 10^{-4}$ & 0.61 & 0.01\\
G & 216 & 55 & 0.2924 & 1$\cdot 10^{-4}$ & 0.56 & 0.03\\
I & 214 & 39 & 0.2918 & 2$\cdot 10^{-4}$ & 0.65 & 0.10\\
\hline
\end{tabular}
\end{table}
\subsection{Error analysis}
\label{sec:cinerr}
The errors on $\sigma$, $\gamma$, $z$ are a combination of a
systematic component due to template errors and mismatch, and a random
component due to the limited signal. For the systematic error we adopt
the standard deviation $\Delta \sigma$ of the sample of the acceptable
$\sigma$ values (see Tables~\ref{tab:cineresGH} and
\ref{tab:cineres}). For the random error, we adopt the formal error
given by the GHFF fit with the best-fitting template, although it is
probably an overestimate because it includes possible residual
template mismatches. The random and systematic errors for the line
strength ratio ($\gamma$) and the redshift ($z$), defined in the same
way, are listed in Tables \ref{tab:cineres} (FQ) and
\ref{tab:cineres2GH} (GHFF).
The uncertainty in the measurement of the instrumental resolution
introduces a systematic error in the measurement of the velocity
dispersion that is in our case negligible. For example, considering
{\bf A}, the galaxy with the highest signal-to-noise spectrum, the
random error is estimated by us to be 5 \%, the systematics due to
template mismatches 12 \%, and therefore a further 3 \%, to be added
in quadrature, does not change the total error significantly. As a
double check we repeated the fits with templates with different
resolution ($\delta \sigma_t$ equivalent to 0.26 \AA) and the results
changed as expected:
\begin{equation}
\label{eq:resch}
\delta \sigma = \delta \sigma_t \left(\frac{\sigma_t}{\sigma}\right).
\end{equation}
For galaxy {\bf I} we obtained the larger difference (7 km~$\tx{s}^{-1}$), still
negligible with respect to the other sources of error.
The spectrum of galaxy {\bf D} includes the lines Ca H and K. While
widely used to perform such measurements (e.g. Dressler 1979), these
lines has been reported to induce a slight overestimate of the
velocity dispersion (Kormendy, 1982, estimates it to be between 5 and
20 \%; see also Kormendy and Illingworth, 1982). They suggest the
problem may be caused by the intrinsic width of the lines or by the
steepness of the continuum in that spectral region. However, Dressler
(1984), suggest that they may be best suited for measurement of large
velocity dispersion for faint objects, such as {\bf D}. For these
reasons we performed the kinematic fit also exluding the region of Ca
H and K and we found a velocity dispersion 10 \% higher in this case,
a variation of the order of the estimated errors. In the analysis presented in this paper we shall
consider the results from the entire spectrum, for its higher
signal-to-noise ratio. Nevertheless, when a larger sample of these
objects will be available we plan to better address this issue by
studying systematic variations induced by Ca H and K on the sample.
Similarly the spectrum of {\bf A} includes the region of NaD, which
may be affected by interstellar absorption even in ellpitical
galaxies (e.g. Dressler, 1984). We repeated the analysis exluding the
NaD region and found that this resulted in a 3\% increase in the velocity dispersion, again within
the estimated errors.
The error on $z$ is caused by uncertainties in the wavelength
calibration (the dominant source), in the fit results and in the
absolute motion of the template stars, for the Jacoby et
al.~\shortcite{J84} spectra were not corrected for peculiar motions
(estimated to be of order 0.0001 by the authors). The total error
adds up to about 0.0005. In the cases with insufficient signal, the
redshift has been measured by identifying the main spectral features,
with a total estimated error of $\approx0.005$. These redshifts were
then been checked via superimposition of redshifted templates. The
listed error corresponds to $\sim 10$ pixel in the studied spectral
range.
\subsection{Relative flux calibration}
The instrumental throughput is not constant as a function of
wavelength. In order to obtain a good relative flux calibration for
our spectra, we observed two spectrophotometric standard white dwarfs
through our instrumental setup for a total of five exposures. By
comparison with the spectra taken from the {\sc midas} libraries, we
obtained the response function separately for each exposure, using
{\sc midas gui/long} and the values of galactic extinction from
Burstein \& Heiles \shortcite{maps}. The results were all within two
percent and we adopted their average function to correct the
spectra for the instrumental sensitivity.
\section{Photometric and kinematic corrections}
\label{sec:corr}
The spectroscopic and photometric results described so far require a number of
corrections. Photometry should be corrected for galactic extinction
(Subsection~\ref{ssec:gaex}), while WFPC2 magnitudes should be
transformed to standard restframe magnitudes (Subsection~\ref{ssec:K})
and $\sigma_{\tx{ghff}}$ should be corrected into a standard central
velocity dispersion $\sigma$ (Subsection~\ref{ssec:scorr}).
\subsection{Galactic extinction}
\label{ssec:gaex}
For galactic extinction, we used E(B-V) values from the Burstein \&
Heiles \shortcite{maps} maps, and the relations $A_{\tx{\small F606W}}=2.907\,
\tx{E(B-V)}$ and $A_{\tx{\small F814W}}=1.902\, \tx{E(B-V)}$, calculated by
Romaniello \shortcite{Rom}. The extinction coefficients obtained for
the various fields are given in Table~\ref{tab:gaex}.
\begin{table}
\caption{Galactic extinction coefficients.}
\label{tab:gaex}
\begin{tabular}{lccc}
field & E(B-V) & $A_{\tx{F606W}}$ & $A_{\tx{F814W}}$\\
\hline
u5405 & 0.18-0.21 & 0.57 & 0.37 \\
ur610 & 0.03-0.06 & 0.13 & 0.09 \\
urz00 & 0.06-0.09 & 0.22 & 0.14 \\
ust00 & 0 & 0 & 0 \\
ut800 & 0.03-0.06 & 0.13 & 0.09 \\
\hline
\end{tabular}
\end{table}
\subsection{An improved way to calculate K-correction}
\label{ssec:K}
The photometric parameters measured through WFPC2 filters have different
physical meaning depending on the galaxy redshift. The common technique to
obtain standard magnitudes is to introduce the so-called K-correction,
\begin{equation}
\textrm{K}\equiv-2.5\log\left[\frac{\int F_{\lambda}
(\lambda)S_o(\lambda)d\lambda}{\int F_{\lambda} (\lambda)S_o
(\lambda(1+z))d\lambda}\right],
\end{equation}
where $F_{\lambda}$ is the spectral flux density of the galaxy and
$S_{o}$ is the response function of the instrumental setup used during the
observations. To calculate the K-correction at intermediate redshift,
knowledge of a large spectral range is required. In order to reduce the
required spectral range and to minimise the correction, it is convenient
to calculate the restframe absolute magnitudes ($M_{\tx{w}}$) in different
filters from the observational ones ($m_{\tx{o}}$):
\begin{equation}
M_{\tx{w}}\equiv \Delta m_{\tx{wo}}(z;F_{\lambda}) + m_{\tx{o}} - DM(z;\Omega,\Omega_{\Lambda}),
\end{equation}
where
\begin{eqnarray}
\Delta m_{\tx{wo}}\equiv&-2.5\log\Big[\frac{\int F_{\lambda}(\lambda)S_{\tx{w}}(\lambda)d\lambda \nonumber}{\int F_{\lambda}^{st}(\lambda)S_{\tx{w}}(\lambda)d\lambda}\Big]\nonumber\\
&+2.5\log\Big[\frac{\int F_{\lambda}(\lambda)S_{\tx{o}}(\lambda (1+z))d\lambda}{\int F_{\lambda}^{st}(\lambda)S_{\tx{o}}(\lambda)d \lambda} \Big],
\end{eqnarray}
$DM(z;\Omega,\Omega_{\Lambda})$ is the bolometric distance modulus,
and $F_{\lambda}^{st}$ is the spectral flux density of a standard
star. For each redshift, we calculated corrections from F606W to
Johnson B ($\Delta m_{\tx{\small BF606W}}$) and from F814W to Johnson
V ($\Delta m_{\tx{\small VF814W}}$).
To perform this calculation we chose the following, purely empirical,
approach. For each galaxy we selected from the synthetic library of
Bruzual \& Charlot (1993; GISSEL96 version) the spectrum which best
approximates the galactic spectrum under consideration in the observed
spectral range. Assuming that the selected spectrum provides a good
description of the spectral energy distribution of our galaxy also
outside the observed wavelength range, we used it to compute $\Delta
m_{\tx{wo}}$. This method also provided us with an error estimate based on
the range $\Delta m_{\tx{wo}}$ obtained from all the spectra not
significantly different from the observed one; the typical error on
the K-correction is $\approx 0.05$ mag. Since we were particularly
interested in the continuum shape, which is the main contribution to
large-passband fluxes, we needed flux calibrated spectra. The
so-called age-metallicity degeneracy \cite{Worthey} with respect to
the continuum shape is not a significant concern, because we were only choosing
the best phenomenological approximation. In Table~\ref{tab:Kcorr} we
list the computed $\Delta m_{\tx{wo}}$ values. As a cross-check we
compared the WFPC2 colours of our galaxies with the WFPC2 colours
``measured'' on the redshifted synthetic spectra. The colours from
the best approximation spectrum were all within 0.05 magnitudes from
the observed one.
\begin{table}
\caption{Photometric corrections.}
\label{tab:Kcorr}
\begin{tabular}{lccc}
galaxy& $z$ & $\Delta m$ & $\Delta m$\\
& & B F606W & V F814W \\
\hline
A & 0.146 & 1.11 & 1.12 \\
B & 0.146 & 1.01 & 1.04 \\
D & 0.384 & 0.44 & 0.86 \\
E & 0.294 & 0.68 & 1.00 \\
F & 0.293 & 0.68 & 0.99 \\
G & 0.294 & 0.63 & 0.92 \\
I & 0.292 & 0.65 & 0.94 \\
O & 0.647 & -0.28 & 0.65 \\
P & 0.550 & -0.05 & 0.74 \\
\hline
\end{tabular}
\medskip
For the galaxies {\bf B}, {\bf O}, {\bf P} with insufficent
signal-to-noise ratio, the redshift was found by identifying the
strongest spectral features. The redshift of {\bf P} is somewhat less
certain because the signal-to-noise ratio is particularly low.
\end{table}
\subsection{Reduction to a standard central velocity dispersion}
\label{ssec:scorr}
In Section~\ref{sssec:fq} $\sigma_{\tx{ghff}}$\, was defined as the broadening of the
projected spectrum integrated over the solid angle subtended by the
CCD. The velocity dispersion in elliptical galaxies varies with
distance from the center. Usually, the velocity dispersion is defined
as the kinematic broadening of a spectrum integrated over a reference
solid angle. The choice of a standard reference solid angle makes it
possible to compare values obtained in different studies. Previously
used reference solid angles are, e. g., a fixed angular size at a fixed
distance or a fraction of the effective radius (J{\o}rgensen, Franx \&
Kj{\ae}rgaard 1996; hereafter JFK96). The latter choice is to be
preferred because it is related to an intrinsic length scale of the
galaxy and does not depend on cosmological parameters. Following
JFK96, we correct all measurements to a circular aperture of $r_{\tx{e}}/4$.
To calculate the correction we need to model the large-scale kinematic
profile of elliptical galaxies. A power law:
\begin{equation}
\sigma(r)\propto \left(\frac{r}{r_e}\right)^d,
\end{equation}
is generally a good description of the kinematic
profile (see, e.g., the radially extended profiles in Carollo \&
Danziger 1994a, 1994b, Bertin et al. 1994 with $-0.1< d<0$).
We assume that
\begin{equation}
\sigma^2({\mathcal A})\simeq\int_{\mathcal{A}} 2 \pi r dr \sigma^2(r) DV(r),
\end{equation}
where $r$ is the projected angle variable, $\mathcal{A}$ is the
projection on the focal plane of the solid angle subtended by the
instrument and $DV(r)$ is the $r^{1/4}$ law, appropriately normalised.
Under these assumptions
\begin{equation}
\sigma^2=\sigma^2_{\tx{ghff}} \frac{\int_0^{1/4} dx x
\sigma^2(x r_{\tx{e}})DV(x r_{\tx{e}})}{\int_{\mathcal{A}} dr r \sigma^2(r)DV(r)}
\equiv \sigma^2_{\tx{ghff}} {\mathcal B}^2(d).
\end{equation}
The correcting factors were computed by taking into account the number
of pixels used during each spectrum extraction and are summarised in
Table~\ref{tab:scorr}. Given the lack of information about the
kinematic profiles, the correcting factors is taken to be
${\mathcal B}\equiv[{\mathcal B}(-0.1)+{\mathcal B}(0)]/2$, with an
error estimate $\delta {\mathcal B}\equiv [{\mathcal
B}(-0.1)-{\mathcal B}(0)]/2\sqrt{3}$ ($\approx$5\%).
\begin{table}
\caption{Kinematic correcting factors $\mathcal{B}$ with standard
deviations $\delta \mathcal{B}$ (see text) and final central velocity
dispersion $\sigma$ in km~$\tx{s}^{-1}$.}
\label{tab:scorr}
\begin{tabular}{ccccc}
galaxy & $\sigma_{\tx{ghff}}$\, & $\mathcal{B}$ & $\delta \mathcal{B}$ & $\sigma$ \\
\hline
A & 232 & 1.07 & 0.04 & 248 \\
D & 271 & 1.14 & 0.08 & 309 \\
E & 175 & 1.10 & 0.06 & 192 \\
F & 195 & 1.08 & 0.05 & 211 \\
G & 226 & 1.09 & 0.05 & 246 \\
I & 166 & 1.11 & 0.06 & 184 \\
\hline
\end{tabular}
\end{table}
\subsection{Effective radii}
The radii $R_{\tx{eV}}$
and $R_{\tx{eB}}$ are calculated for the central rest wavelength of each
filter by linear interpolation/extrapolation in wavelength between the
measured radii in the WFPC2 filters F606W and F814W at their observed
central wavelengths. We take the central wavelengths of B, V,
F606W and F814W to be 4400, 5500, 5935, and 7921~\AA, respectively,
resulting in the following explicit formulae for the angular sizes:
\begin{eqnarray}
r_{\tx{e}{\sc B}}=&\big\{r_{\tx{e}{\sc F606W}}[7921-(1+z)4400]\nonumber\\
&+r_{\tx{e}{\sc F814W}}[(1+z)4400-5935]\big\}/1986 \\
r_{\tx{e}{\sc V}}=&\big\{r_{\tx{e}{\sc F606W}}[7921-(1+z)5500]\nonumber\\
&+r_{\tx{e}{\sc F814W}}[(1+z)5500-5935]\big\}/1986 \, .
\end{eqnarray}
The typical correction is of order of 3\%.
\section{Results}
\label{sec:res}
We now consider the astrophysical implications of our measurements in
terms of the changes in scaling laws as a function of redshift. We
emphasize that our sample, although small (only six galaxies with a
reliable measurement of the velocity dispersion), is unique in that it
focuses on {\it field} galaxies. Therefore, we will first consider
our sample by itself (Subsection~\ref{ssec:ourres}), and then proceed
to compare with {\it cluster} galaxy measurements from the literature
(Subsection~\ref{ssec:compres}).
\subsection {Results from our sample: Scaling laws for field galaxies at
intermediate redshifts}
\label {ssec:ourres}
The input data for the galaxies in our sample are presented in
Table~\ref{tab:scorr} for the central velocity dispersion and in
Table~\ref{tab:fres} for the photometry. Absolute magnitudes and metric
sizes in Table~\ref{tab:fres} are computed using $H_0$=50 $h_{50}$ km~$\tx{s}^{-1}$
$\tx{Mpc}^{-1}$, $\Omega=1$, $\Omega_{\Lambda}=0$.
\begin{table}
\caption{Final corrected photometric values. Effective radii are
expressed in $h_{50}^{-1}$kpc. B and V refer to the corresponding Johnson
passbands. Quantities are computed using $\Omega=1$, $\Omega_{\Lambda}=0$.}
\begin{tabular}{lcclccl}
galaxy& $R_{\tx{eB}}$ & SB$_{\tx{eB}}$ & $M_{\tx{B}}$ & $R_{\tx{eV}}$ & SB$_{\tx{eV}}$ & $M_{\tx{V}}$ \\
\hline
A & 5.02 & 21.03 & -21.23 & 5.73 & 20.37 & -22.30 \\
B & 2.02 & 20.02 & -20.00 & 1.94 & 18.95 & -20.94 \\
D & 4.93 & 19.89 & -22.16 & 4.22 & 18.65 & -22.94 \\
E & 4.30 & 20.60 & -21.13 & 4.60 & 19.83 & -22.09 \\
F & 6.18 & 21.57 & -20.97 & 6.76 & 20.85 & -21.92 \\
G & 9.02 & 20.75 & -22.5 & 8.39 & 19.72 & -23.29 \\
I & 3.52 & 19.89 & -21.35 & 3.40 & 18.93 & -22.20 \\
O & 5.61 & 20.70 & -21.83 & 4.58 & 19.50 & -22.61 \\
P & 3.58 & 20.29 & -21.21 & 3.43 & 19.32 & -22.09 \\
\hline
\end{tabular}
\label{tab:fres}
\end{table}
We consider two scaling laws: the FP (Eq.~\ref{eq:FP}) and the
Kormendy relation \cite{K77}
\begin{equation}
SB_{\tx{e}} =a_1\log R_e+a_2.
\label{eq:HK}
\end{equation}
The latter is useful for our purposes because it does not involve
internal kinematics, and thus it can be followed to higher redshifts
($ z = 0.647 $ for our sample), although it has a larger intrinsic scatter
than the FP and its coefficients are much more sensitive to sample
selection biases (see e.g., Capaccioli, Caon \& D'Onofrio 1992).
Because of the small number of points in our sample, we do not try to
rederive the parameters $ \alpha $ and $ \beta $ for the FP
independently, but we compare our sample of field ellipticals with the
local relations, adopting their slopes. In Figures~\ref{fig:allvscoma}
and \ref{fig:Ben} we compare our galaxies with a sample of Coma
ellipticals in Johnson V (Lucey at el. 1991, hereafter L91) and with a
recent determination of the FP in Johnson B, $\alpha=1.25, \beta=0.32,
\gamma=-8.895$, (B98). The mean offset $\Delta \gamma $ (defined
as the mean of $\log R_e + \log h_{50} - \alpha \log \sigma-\beta {\tx{SB}}_{\tx{e}}
-\gamma$ over the data points) is 0.10 with respect to the L91 sample
(using the slopes given in L91, i.e. $\alpha=1.23$ and $\beta=0.328$)
and 0.07 in Johnson B. We also compared our data {\it i)} to the 7
Samurai sample \cite{7S} obtaining $\Delta \gamma $=0.08 and 0.07,
respectively in Johnson B and V and {\it ii)} to the FP in Gunn $r$
(JFK96), finding $\Delta \gamma \approx 0.03$ with respect to the Coma
zero point, even if the colour corrections increase in this case the
uncertainty.
The rms scatter of our points is 0.16 in $\log R_e$ giving a standard
deviation of the mean value of about 0.07 in $\Delta
\gamma$. Considering other smaller sources of error such as the
peculiar velocity of Coma and the colour transformations we can
estimate the uncertainty on $\gamma$ to be $\approx 0.1$. The offset
found can be interpreted as a result of the evolution of the mean
$M/L$ ratio, $\Delta \log M/L\approx 0.1$. In other words, our field
galaxies are more luminous than a local elliptical with the same
effective radius and $\sigma$ (dimension and mass), consistently with
the expected evolution of stellar populations, and in agreement with
the results found by Bender et al. (1996), Ziegler \& Bender (1997),
Kelson et al.\ (1997), Schade, Barrientos \& L\'opez-Cruz (1997) and van
Dokkum et al.\ (1998), for the cluster environment.
If we exclude from the analysis the two galaxies {\bf E} and {\bf F}
with lower signal-to-noise spectra, the mean offset
becomes 0.15 with respect to the L91 sample, 0.11 to B98, 0.13 and
0.12 to the 7 Samurai B and V and 0.12 to JFK96. The scatter is
practically unchanged (0.16). The change in the offset is not
negligible, but is not statistically significant. More data are needed
to extend this preliminary study to larger sample in order to overcome
small number fluctuations and to measure the offset and scatter with
smaller errors. Moreover when a larger sample will be available, it
will be possible to measure the offset in smaller redshift bins, thus
providing homogeneous subsamples.
Clusters provide a homogeneous environment for the formation of
ellipticals. In contrast, there is no {\it a priori} reason to believe
that randomly selected field galaxies have formed at the same epoch,
and any differences in the formation redshift will be amplified with
look-back time. Therefore, one might have expected to find a larger
scatter for our field relation (if any relation was to be found at
all) than in the cluster FP. The scatter we find for our data
points (0.16 in $\log R_e$, using the local slopes) is in fact larger than
the scatter found by, e.g., L91 (0.075). Part of this scatter can be
explained because we are considering a small sample of galaxies at
different redshifts (and therefore different $\gamma$; from a
geometric point of view, the plane is thicker because is the
superimposition of parallel planes). It is not clear however whether
there is a residual scatter, and a larger sample is required to
investigate the intrinsic variations of the formation and evolutionary
history for galaxies not in rich clusters.
\subsection {Comparison with other results: The evolution of the
Fundamental Plane}
\label{ssec:compres}
Obviously, a larger sample of field galaxies is necessary to assess
whether they truly obey an FP-like relation and whether, and how much,
their properties vary with redshift. However, a significant body of
data is now available on ellipticals at intermediate redshifts,
mostly from clusters, and it is instructive to look at our sample
in the context of the cluster data as well.
We consider here a total of 25 ellipticals from the papers of van
Dokkum and Franx (1996, 4 objects), Kelson et al.~(1997, 15 objects),
and from our sample (6 objects).
The first consideration is that these galaxies, much like
ours, fit reasonably well with the zero-redshift relations (see
Figure~\ref{fig:allvscoma}). The zero point determined from all these
galaxies using the L91 slopes is $ -8.60 $
(Figure~\ref{fig:allvscoma}), very similar to the value obtained with
our six field galaxies alone.
\begin{figure}
\mbox{\epsfysize=8cm \epsfbox{Lucey.eps}}
\caption {The intermediate-redshift (filled) points plotted against
the L91 (Lucey et al.\ 1991; empty pentagons) Coma cluster
sample. The filled points are from this paper (large squares), from
van Dokkum and Franx 1996 (triangles), and from Kelson et
al.~1997 (small squares and pentagons). All surface
brightnesses are in Johnson V. The line corresponds to the FP relation
of L91. The rms scatter in $ \log R_e $ is 0.14. The mean offset of
the field ellipticals is 0.10 $ \log R_e $, of the cluster sample is 0.17 }
\label{fig:allvscoma}
\end{figure}
\begin{figure}
\mbox{\epsfysize=8cm \epsfbox{Bender.eps}}
\caption {The field intermediate-redshift galaxies parameters in
Johnson B are shown as filled squares. The solid line is the FP
relation given by Bender et al. 1998. The mean offset is 0.07 in $\log
\tx{R}_{\tx{e}}$. The rms scatter in $\log \tx{R}_{\tx{e}}$ is 0.15.}
\label {fig:Ben}
\end{figure}
\begin{figure}
\mbox{\epsfysize=8cm \epsfbox{UDCH.eps}}
\caption {The intermediate-redshift points plotted against their own
best-fitting FP relation. The points are from this paper (large
squares), from van Dokkum and Franx 1996 (open triangles), and from Kelson
et al.~1997 (open squares and pentagons). The rms scatter in $ \log
R_e $ is 0.10.}
\label {fig:allvshigh}
\end{figure}
With this sample, it is also possible to determine the FP slopes $
\alpha $ and $ \beta $ independently of the low-redshift measurements.
The results depend on the method used for the minimisation ($\alpha$
especially, while $\beta$ is better determined). Using the technique
described in JFK96, which equally weights the three FP parameters, we
obtain $ \alpha = 0.91 $ and $ \beta = 0.307 $
(Figure~\ref{fig:allvshigh}). We estimate the errors to be
$\sigma_{\alpha}\approx 0.15$ and $\sigma_{\beta}\approx0.02$. This
choice of the slopes reduces the scatter in $ \log~R_e $ from 0.13
(with respect to the local relation) to about 0.10. The relation
$\alpha-10\beta+2=0$ (see Section \ref{sec:intro}) still holds within
the errors. If the reasoning is carried out with the assumption of the
single parameter $\eta$ ($L\propto M^{\eta}$), this would lead to
$\Delta \eta \approx -0.2$. If confirmed with better statistics, this
variation might imply a differential passive evolution as a function
of mass, i.e. a mass-dependent formation redshift.
In Figure~\ref{fig:totkor} we show the Kormendy relation between
SB$_{\tx{e}}$ and $R_{\tx{e}}$ for nine galaxies in our field sample (up to z=0.647)
plus the 25 other intermediate redshift cluster ellipticals. The
intermediate redshift data are compared in the figure with the L91
sample of Coma ellipticals. An average offset of $-0.52$ magnitudes is
found between the intermediate redshift cluster and field sample and the local
Coma sample. The intermediate redshift galaxies are therefore on
average brighter (at the same $R_{\tx{e}}$), consistently with what is found using
the FP. In order to determine the slope $a_1$, a sample with well-controlled
selection biases is required. The
intermediate redshift sample was not chosen to this aim and therefore
we do not attempt to derive the value of $a_1$.
More data and carefully selected subsamples are needed to assess if
the field Kormendy relation at intermediate redshift differs
significantly from the cluster one.
\begin{figure}
\mbox{\epsfysize=8cm \epsfbox{korComa.eps}}
\caption{Surface brightness (SB$_{\tx{eV}}$) vs. effective radius
(R$_{\tx{e}}$) for the same sample as in Figure~\ref{fig:allvscoma}
(filled symbols); the large squares are our data, the small squares
and the pentagons are the galaxies from CL1358+62 and MS2053-04
(Kelson et al.\ 1997), and the triangles are the four ellipticals of
CL0024 (van Dokkum \& Franx 1996). The data of a sample of Coma
ellipticals (Lucey et al. 1991) are shown as empty pentagons for
comparison, and their best-fit Kormendy relation is overplotted. The
mean average offset is $\approx-0.5$ magnitudes. Three galaxies are
also shown ({\bf B}, {\bf O}, {\bf P}) for which we have measured
photometry and redshift but no central velocity dispersion. }
\label{fig:totkor}
\end{figure}
\section{Summary}
\label{sec:conc}
With the instrumental setup and the procedure described Sections 3, 4
and 5, we have measured the Fundamental Plane parameters of
intermediate redshift field elliptical galaxies. As an important part
of the data-reduction procedure we have carried out accurate
K-corrections and spectroscopic aperture corrections. Given the
importance of the template spectrum choice for the kinematic
fit, we have decided to make use of a wide set of spectra, checking
how the results vary with the spectral type, as described in Section
4, obtaining at the same time an estimate of the systematic error
produced by the choice of the template spectrum.
While the number of galaxies studied here is too small to draw firm
conclusions on the properties of field ellipticals at intermediate
redshift, the preliminary evidence suggests that:
\begin{enumerate}
\item Our six field ellipticals at redshift $z\sim 0.3$,
are in agreement with the local FP relation, with a variation of
the zero point ($\Delta \gamma \approx 0.1$), and a scatter of 0.16 in
$\log R_{\tx{e}}$. This means that the stellar populations of our sample of field
galaxies are brighter than the local ellipticals, with the same size and mass.
This data fit into a scenario in which our galaxies, at a look-back time
of $\approx 4$--$5$ Gyrs, are evolving passively. The small sample and the
sample selection criteria do not allow any more general conclusion.
\item The FP obtained from our data and the cluster ellipticals at
intermediate redshift \cite{DF96,KDFIF} is well defined with a scatter
of 0.13 in $\log R_e$ with respect to the local relation. The full
intermediate redshift sample is large enough to perform the fit the
FP coefficients independently. Using the fitting technique used by JFK96,
we find $\alpha=0.91$ and $\beta=0.307$, and the scatter is reduced
to 0.10. By interpreting the Fundamental Plane as a result of
homology, the virial theorem and the existence of a relation $L\propto
M^{\eta}$ \cite{3M}, different slopes at different redshift imply a
variation of $\eta$ with time, i.e., a mass-dependent evolution. The
good agreement between the field and cluster galaxies suggests that
there are no significant differences in the history of the two
environments. More data are needed to perform separate fit to the two
subsamples.
\item The Kormendy relation between SB$_{\tx{e}}$ and $R_{\tx{e}}$ for
all the intermediate redshift data shows an offset of $\approx-0.5$
magnitudes in surface brightness with respect to the Coma ellipticals
(L91), in agreement with what is found using the FP. Even though
caution is needed, due to possible sample selection biases, the
variation of the Kormendy relation suggests that the evolution
found via the Fundamental Plane, extends to galaxies at higher
redshift. Carefully selected samples of field
and cluster intermediate redshift are needed to study the
the Kormendy relation in greater detail and to
investigate the possible differences between the two environments.
\end{enumerate}
Further
discussion on the evolution of elliptical galaxies will be given in a
follow-up paper (Paper II; Stiavelli et al.\ 1999) with the help of
information on the line strengths.
\section{Acknowledgments}
Tommaso Treu's work at Space Telescope Science Institute (STScI) was
financially supported by the STScI Summer and Graduate Student
Programs, the Scuola Normale Superiore (Pisa), the Italian Space
Agency (ASI), and by STScI DDRF grant 82216. We thank an anonymous
referee for several valuable comments which significantly improved the
presentation of the kinematic measurement. The use of Gauss-Hermite
Fourier Fitting Software developed by R.~P. van der Marel and M. Franx
is gratefully acknowledged.
|
\section{Introduction}
Descriptions of fluid motion are conventionally based on the principles
of conservation of mass and linear momentum. One might hope that all
such descriptions would accordingly exhibit key energetic properties
(nonlinear, exponential--type dissipation in the absence of forcing and
long--term stability under conditions of time dependent loading)
consistant with the principle of energy conservation. These properties
are intrinsic to real flows and the conventional, Eulerian
Navier--Stokes equations.
A completely general reference description of an incompressible,
Newtonian fluid, which reconciles the differences between the
so--called arbitrary Lagrangian Eulerian (A.L.E.) formulation of {\sc
Hughes, Liu} and {\sc Zimmerman} \cite{h:1} (deformation gradients
absent) and that of {\sc Soulaimani, Fortin, Dhatt} and {\sc Ouellet}
\cite{f:1} (deformation gradients present, but use is problematic), is
derived for the purposes of this investigation. The implications of the
resulting description are investigated in the context of energy
conservation in a similar, but broader, approach to that taken by
others (eg. {\sc Simo} and {\sc Armero} \cite{s:1}) for the
conventional, Eulerian Navier--Stokes equations.
The work subsequently focusses on establishing a class of time
discretisations which inherit these self--same energetic properties
irrespective of the time increment employed. The findings of this
analysis have profound consequences for the use of certain classes of
difference schemes in the context of deforming references. It is
significant that many algorithms presently in use do not automatically
inherit the fundamental qualitative features of the dynamics. An
``updated'' approach as a means of avoiding ever burgeoning deformation
gradients which arise from the accumulated step--wise deformation of
meshes and a still further simplified implementation are further topics
explored.
The main conclusions of this work are based on a new inequality and a
number of lemmas. These lemmas are mainly concerned with the new
convective term. The new inequality is used in place of where the
Poincar\'{e}--Friedrichs inequality might otherwise have limited the
analysis. This analysis is extended in that non--zero boundaries,
so--called free boundaries and time--dependent loads are considered.
\section{A Completely General Reference} \label{130}
The implementation of most numerical time integration schemes would be
problematic were a conventional Eulerian\footnotemark[1] description of
fluid motion to be used in instances involving deforming domains. The
reason is that most numerical time integration schemes require
successive function evaluation at fixed spatial locations. On the other
hand meshes rapidly snarl when purely Lagrangian\footnotemark[2]
descriptions are used. \footnotetext[1]{{\sc Eulerian} or {\sc spatial}
descriptions are in terms of fields defined over the current
configuration.} \footnotetext[2]{{\sc Lagrangian} or {\sc material}
descriptions are made in terms of fields defined over a reference
(material) configuration.}
Eulerian and Lagrangian references are just two, specific examples of
an unlimited number of configurations over which to define fields used
to describe the dynamics of deforming continua. They are both special
cases of a more general reference description, a description in which
the referential configuration is deformed at will. A deforming finite
element mesh would be a good example of just such a deforming reference
in practice. The transformation to the completely general reference
involves coordinates where used as spatial variables only and the
resultant description is therefore inertial in the same way as
Lagrangian descriptions are.
\subsection{Notation}
Consider a material body which occupies a domain $\Omega$ at time $t$.
The material domain, $\Omega_0$, is that corresponding to time $t =
t_0$ (the reference time, $t_0$, is conventionally, but not always,
zero). A third configuration, ${\tilde \Omega}$, which is chosen
arbitrarily is also defined for the purposes of this work. The three
domains are related in the sense that points in one domain may be
obtained as one--to--one invertible maps from points in another.
\begin{figure}[h]
\begin{center}
\mbox{\epsfbox{domains.eps}}
\end{center}
\caption{Schematic Diagram of Domains and Mappings Used in a
Completely General Reference Description} \label{91}
\end{figure}
For any general function $f({\mbox{\boldmath{$x$}}},t)$, a function, $\tilde
f(\tilde{\mbox{\boldmath{$x$}}},t) \ \equiv \ f({\mbox{\boldmath{$\lambda$}}}^{\ast}(\tilde{\mbox{\boldmath{$x$}}},t),t)$, can
be defined in terms of the domains and one--to--one, invertible
mappings illustrated in Figure \ref{91}. Similarly, $f_0({\mbox{\boldmath{$x$}}}_0 ,t)
\ \equiv \ f({\mbox{\boldmath{$\lambda$}}}({\mbox{\boldmath{$x$}}}_0 ,t),t)$ can be defined. This notation
can be generalised for the component--wise definition of higher order
tensors. The key to understanding much of this work lies possibly in
adopting a component-wise defined notation.
In contrast to the function notation just established, the definition
of the operators ${\tilde \nabla}$ and $\widetilde {\mbox{div}}$ is not
based on $\nabla$ and $\mbox{div}$. They are instead the referential
counterparts, that is
\[
{\tilde \nabla} = \frac{\partial}{\partial {\tilde {\mbox{\boldmath{$x$}}}}}
\hspace{10mm} \mbox{and} \hspace{10mm} {\widetilde {\mbox{div}}} =
\frac{\partial}{\partial {\tilde x}_1} + \frac{\partial}{\partial
{\tilde x}_2} + \frac{\partial}{\partial {\tilde x}_3}.
\]
The notation ${\mbox{\boldmath{$A$}}}:{\mbox{\boldmath{$B$}}}$ is used to denote the matrix inner product
$A_{ij}B_{ij}$ throughout this work, $\left< \ \cdot \ , \ \cdot
\ \right>_{L^2( \ \cdot \ )}$ denotes the $L^2$ inner product and
$\left|\left| \ \cdot \ \right|\right|_{L^2( \ \cdot \ )}$ the $L^2$
norm.
\subsection{Some General Results for Functions Defined on the Three
Domains}
Three important results are necessary for the derivation of the
completely general reference description and these are presented
below.
\subsubsection*{The Material Derivative in Terms of a Completely
General Reference}
The material derivative of any vector field ${\tilde {\mbox{\boldmath{$v$}}}}$ in terms
of a completely general, reference is
\begin{eqnarray} \label{25}
\frac{\partial \tilde{\mbox{\boldmath{$v$}}}}{\partial t} + {{\tilde \nabla} \tilde
{\mbox{\boldmath{$v$}}}} \left[ {\tilde {\mbox{\boldmath{$F$}}}}^{-1} (\tilde{\mbox{\boldmath{$v$}}} - \tilde{\mbox{\boldmath{$v$}}}^{ref})
\right].
\end{eqnarray}
where $\tilde{\mbox{\boldmath{$v$}}}^{ref}$ is the velocity of the reference deformation,
and $\tilde {\mbox{\boldmath{$F$}}}$ is the deformation gradient given by
\[
{\tilde {\mbox{\boldmath{$F$}}}}({\tilde {\mbox{\boldmath{$x$}}}}) = \frac{\partial {\mbox{\boldmath{$\lambda$}}}^*}{\partial
{\tilde {\mbox{\boldmath{$x$}}}}}.
\]
This result (taken from {\sc Hughes, Liu} and {\sc Zimmerman}
\cite{h:1}) is obtained by recalling that the material derivative
(total derivative) is the derivative with respect to time in the
material configuration. Thus
\begin{eqnarray} \label{26}
\frac{D\tilde{v}_i}{Dt} & = & \frac{\partial}{\partial t} \{ {\tilde
v}_i ( {\tilde {{\mbox{\boldmath{$\lambda$}}}}} ({\mbox{\boldmath{$x$}}} _0,t) ,t) \} \nonumber \\
& = & \frac{\partial \tilde{v}_i}{\partial t} + \frac{\partial
\tilde{v}_i}{\partial \tilde{x}_j}\frac{\partial
\tilde{\lambda}_j}{\partial t} \ .
\end{eqnarray}
A more practical expression is needed for $\displaystyle \frac
{\partial\tilde{\lambda_j}} {\partial t}$ (the velocity as perceived in
the distorting reference). This can be obtained by considering
\[
\lambda_k({\mbox{\boldmath{$x$}}} _0,t) = \lambda^{\ast}_k(\tilde {{\mbox{\boldmath{$\lambda$}}}} ({\mbox{\boldmath{$x$}}}
_0,t),t) \hspace{10mm} \mbox{(see Figure \ref{91} on page
\pageref{91})}
\]
so that
\[
{\left.\frac{ \partial \lambda_k }{\partial t} \right|} _{ {\mbox{\boldmath{$x$}}} _0
\ fixed} = {\left.\frac{ \partial \lambda^{\ast}_k }{\partial t}
\right|}_{ {\tilde{\mbox{\boldmath{$x$}}} } \ fixed} + \frac{ \partial \lambda^{\ast}_k
}{\partial \tilde{x}_j}\frac{\partial \tilde{\lambda}_j}{\partial t}
\]
or
\[
\frac{\partial \tilde{\lambda}_j}{\partial t} = \frac{\partial
\tilde{x}_j}{\partial x_k} \left( {\left. \frac{ \partial \lambda_k
}{\partial t} \right|}_{ {\mbox{\boldmath{$x$}}} _0 \ fixed} - {\left. \frac{ \partial
\lambda^{\ast}_k }{\partial t} \right|}_{ {\tilde {\mbox{\boldmath{$x$}}} } \ fixed}
\right).
\]
Substituting this expression into equation (\ref{26}), the desired,
suitably practicable result is obtained.
\subsubsection*{An Element of Area in Terms of a Distorting Reference}
The second important result can be recalled from general continuum
mechanics. Consider an element of area, size $dA$, with an outward unit
normal ${\mbox{\boldmath{$n$}}}$. Then
\begin{eqnarray} \label{27}
{\mbox{\boldmath{$n$}}} dA = {\tilde{\mbox{\boldmath{$F$}}}}^{-t}\tilde{\mbox{\boldmath{$N$}}} {\tilde J}
d\tilde{A}
\end{eqnarray}
where $d{\tilde A}$ and $\tilde{\mbox{\boldmath{$N$}}}$ denote the respective analogous
size and outward unit normal of this element of area in the referential
configuration (capital ``n'' so as to remain consistent with the
notation, since ${\tilde N}_i \neq n_i$ in this case) and ${\tilde J}
= \det{\tilde {\mbox{\boldmath{$F$}}}}$. This result is demonstrated in most popular
textbooks on continuum mechanics (eg. {\sc Lai}, {\sc Rubin} and {\sc
Krempl} \cite{lrk:1}).
\subsubsection*{The Kinematic Result ${\dot {\cal J}}_0 = {\cal J}_0
\mathop{\rm div}{\mbox{\boldmath{$v$}}}$} \label{5}
The material derivative of the Jacobian ${\cal J}_0$ is given by the
relation
\[
{\dot {\cal J}}_0 = {\cal J}_0 \mathop{\rm div}{\mbox{\boldmath{$v$}}}
\]
where ${\cal J}_0$ is defined as follows,
\[
{\cal J}_0 \equiv \det \left\{ \frac{\partial {\mbox{\boldmath{$\lambda$}}}}{\partial
{\mbox{\boldmath{$x$}}}_0} \right\}.
\]
This result is demonstrated in most popular textbooks on continuum
mechanics (eg. {\sc Lai}, {\sc Rubin} and {\sc Krempl} \cite{lrk:1}).
\subsection{Derivation of the Completely General Equation}
One way in which to derive a completely general reference
description of an incompressible, Newtonian fluid is to start with
the balance laws in global (integral) form, and to make the necessary
substitutions in these integrals. The desired numerical
implementation (similar to the conventional Navier--Stokes one which
has been thoroughly investigated and found to be stable) is then
obtained.
\subsubsection*{Conservation of Mass}
Let $\Omega(t)$ be an arbitrary sub--volume of material. The principle
of conservation of mass states that
\begin{eqnarray} \label{92}
\frac{d}{dt} \int_{\Omega(t)} \rho {d \Omega} &=& 0 \hspace{10mm}
\mbox{\it (rate of change of mass with time }= \ \mbox{\it 0)}
\nonumber \\
& & \nonumber \\
\frac{d}{dt} \int_{\Omega_0} {{\rho}_0} {\cal J}_0 {d \Omega_0} &=& 0
\hspace{10mm} \mbox{\it (reformulating in terms of the material}
\nonumber \\
& & \hspace{13mm} \mbox{\it configuration, } \Omega_0 \mbox{\it .)}
\nonumber \\
\int_{\Omega_0} \frac{\partial}{\partial t} \left\{ {{\rho}_0} {\cal
J}_0 \right\} {d \Omega_0} &=& 0 \hspace{10mm} \mbox{\it (since limits
are not time dependent in} \nonumber \\
& & \hspace{13mm} \mbox{\it the material configuration.)} \\
\int_{\Omega_0} \left( \rho_0 {\dot {\cal J}}_0 \ + \ {\dot \rho}_0
{\cal J}_0 \right) d \Omega_0 &=& 0 \hspace{10mm} \mbox{\it (by the
chain rule)} \nonumber \\
& & \nonumber \\
\int_{\Omega(t)} \left( \dot{\rho} \ + \ {\rho \mathop{\rm div}{\mbox{\boldmath{$v$}}}
} \right) {d \Omega} &=& 0 \hspace{10mm} \mbox{\it (using the
kinematic result } \dot{{\cal J}_0} = {\cal J}_0 \mbox{div}\, {\mbox{\boldmath{$v$}}}
\mbox{\it )} \nonumber \\
& & \nonumber \\
\int_{{\tilde \Omega}(t)} \left( {\dot {\rho}} \ + \ {\rho}
\frac{\partial \tilde{v}_i}{\partial \tilde{x}_j}\frac{\partial
\tilde{x}_j}{\partial x_i} \right) {\tilde J} {d
{\tilde \Omega}} &=& 0 \hspace{10mm} \mbox{\it (reformulating in terms
of the distorting} \nonumber \\
& & \hspace{13mm} \mbox{\it referential configuration, }{\tilde \Omega}
(t) \mbox{\it .)} \nonumber \\
\Rightarrow \left( {\dot {\rho}} \ + \ {\rho} {{{\tilde \nabla} \tilde
{\mbox{\boldmath{$v$}}} } : {\tilde {\mbox{\boldmath{$F$}}} }^{-t}} \right) {\tilde J} &=&
0 \hspace{10mm} \mbox{\it (integrand must be zero since the volume}
\nonumber \\
& & \hspace{13mm} \mbox{\it was arbitrary.)} \nonumber
\end{eqnarray}
Thus, for a material of constant, non--zero density,
\[
{\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}} : {\tilde {\mbox{\boldmath{$F$}}}}^{-t} = 0 \hspace{7mm}
\mbox{since} \hspace{7mm} {\tilde J} \neq 0
\hspace{12mm} \mbox{\it (mappings are one-to-one and invertible).}
\]
Notice also that equation (\ref{92}) implies
\begin{eqnarray} \label{93}
\frac{\partial}{\partial t} \left\{ {{\rho}_0} {\cal J}_0 \right\} &=&
0
\end{eqnarray}
since the volume was arbitrary and the integrand must therefore be
zero.
\subsubsection*{Conservation of Linear Momentum (and Mass)}
The principle of conservation of linear momentum for an arbitrary
volume of material $\Omega(t)$ with boundary $\Gamma(t)$ states that
\begin{eqnarray} \label{28}
\frac{d}{dt} \int_{\Omega(t)} \rho {\mbox{\boldmath{$v$}}} {d \Omega} = \int_{\Omega(t)}
\rho {\mbox{\boldmath{$b$}}} {d \Omega} \ + \ \int_{\Gamma(t)} {\mbox{\boldmath{$\sigma$}}} {\mbox{\boldmath{$n$}}} dA
\end{eqnarray}
where ${\rho}$ is density, ${\mbox{\boldmath{$b$}}}$ is the body force per unit mass,
${\mbox{\boldmath{$\sigma$}}}$ is the stress, {\mbox{\boldmath{$n$}}} the outward unit normal to the
boundary and {\mbox{\boldmath{$v$}}} is the velocity. The term on the lefthand side can
be rewritten as follows:
\begin{eqnarray*}
{\frac{d}{dt}} \int_{\Omega(t)} \rho {\mbox{\boldmath{$v$}}} {d \Omega} &=&
\frac{d}{dt} \int_{\Omega_0} {{\rho}_0} {{\mbox{\boldmath{$v$}}} _0} {\cal J}_0 {d
\Omega_0} \hspace{10mm} \mbox{\it (Reformulating in terms of the
material} \\
& & \hspace{42mm} \mbox{\it configuration, } \Omega_0\mbox{\it .)} \\
&=& \int_{\Omega_0} \frac{\partial}{\partial t} \left\{ {{\rho}_0}
{{\mbox{\boldmath{$v$}}} _0} {\cal J}_0 \right\} {d \Omega_0} \hspace{5mm} \mbox{\it (Since
limits are not time dependent in} \\
& & \hspace{41mm} \mbox{\it the material configuration.)} \\
&=& \int_{\Omega_0} \left( \frac{\partial {\mbox{\boldmath{$v$}}}_0}{\partial t}
{{\rho}_0} {\cal J}_0 \ + \ {{\mbox{\boldmath{$v$}}}_0}\frac{\partial}{\partial t} \left\{
{{\rho}_0} {\cal J}_0 \right\} \right) {d \Omega_0} \\
& & \\
& & \\
&=& \int_{\Omega(t)} \rho \dot{\mbox{\boldmath{$v$}}} {d \Omega} \hspace{22mm} \mbox{\it
(The second term above is zero as a} \\
& & \hspace{41mm} \mbox{\it consequence of equation (\ref{93}).)} \\
&=& \int_{{\tilde \Omega}(t)} {\rho} {\dot{\tilde {\mbox{\boldmath{$v$}}} }} {\tilde J}
{d {\tilde \Omega}} \hspace{20mm} \mbox{\it (Reformulating in terms of
the dist--} \\
& & \hspace{42mm} \mbox{\it orting referential configuration, } \tilde
\Omega \mbox{\it .)} \\
&=& \int_{{\tilde \Omega}(t)} {\rho} \left( \frac{\partial
\tilde{\mbox{\boldmath{$v$}}}}{\partial t} + { {{\tilde \nabla} \tilde {\mbox{\boldmath{$v$}}} } } \left[
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} (\tilde{\mbox{\boldmath{$v$}}} - \tilde{\mbox{\boldmath{$v$}}} ^{ref}) \right]
\right) {\tilde J} {d {\tilde \Omega}} \hspace{5mm}
\mbox{\it (Using result} \\
&& \hspace{79mm} \mbox{\it (\ref{25}) on page \pageref{25})}
\end{eqnarray*}
where $\dot{\mbox{\boldmath{$v$}}}$ denotes the material derivative of ${\mbox{\boldmath{$v$}}} $. The
surface integral becomes
\begin{eqnarray*}
\begin{array}{rcll}
\displaystyle \int_{\Gamma(t)} {\mbox{\boldmath{$\sigma$}}} {\mbox{\boldmath{$n$}}} dA &=& \displaystyle
\int_{\tilde {\Gamma}(t)} {\tilde {\mbox{\boldmath{$\sigma$}}}} \tilde{\mbox{\boldmath{$F$}}} ^{-t} {\tilde
{\mbox{\boldmath{$N$}}}} {\tilde J} d{\tilde A} & \mbox{\it (Reformulating in terms of a
distorting} \\
& & & \mbox{\it \ reference using result (\ref{27}) on page
\pageref{27}.)} \\
&=& \displaystyle \int_{{\tilde \Omega}(t)} \mathop{\widetilde {\rm
div}}\, \{ {\tilde {\mbox{\boldmath{$\sigma$}}}} {\tilde{\mbox{\boldmath{$F$}}} ^{-t}} {\tilde J} \} {d
{\tilde {\Omega}} } & \mbox{\it (By the divergence theorem).}
\end{array}
\end{eqnarray*}
Finally, the term involving body force becomes
\begin{eqnarray*}
\begin{array}{rcll}
\displaystyle \int_{\Omega(t)} \rho {\mbox{\boldmath{$b$}}} {d \Omega} &=& \displaystyle
\int_{{\tilde \Omega}(t)} {\rho} {\tilde {\mbox{\boldmath{$b$}}} } {\tilde J} {d {\tilde
\Omega}} \hspace{18mm} & \mbox{\it (Reformulating in terms of a
distorting} \\
& & & \mbox{\it \ reference.).}
\end{array}
\end{eqnarray*}
Substituting these expressions into (\ref{28}), remembering that the
volume used in the argument was arbitrary and that the entire integrand
must therefore be zero, the conservation principles of linear momentum
and mass may be written in primitive form as
\begin{eqnarray} \label{29}
{\rho} \left( \frac{\partial {\tilde {\mbox{\boldmath{$v$}}} }}{\partial t} + {{{\tilde
\nabla} \tilde {\mbox{\boldmath{$v$}}} } {\tilde {\mbox{\boldmath{$F$}}} }^{-1} } ({\tilde {\mbox{\boldmath{$v$}}} } -
{\tilde {\mbox{\boldmath{$v$}}} }^{ref}) \right) {\tilde J} &=& {\rho} {\tilde {\mbox{\boldmath{$b$}}} }
{\tilde J} + \mathop{\widetilde {\rm div}}{\tilde {\mbox{\boldmath{$P$}}}}
\end{eqnarray}
and
\begin{eqnarray} \label{30}
{{\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}} : {\tilde {\mbox{\boldmath{$F$}}} }^{-t} } \ = \ 0
\end{eqnarray}
where $\tilde{\mbox{\boldmath{$P$}}}$ is the Piola--Kirchoff stress tensor of the first
kind, $\tilde{\mbox{\boldmath{$P$}}} \ = \ \tilde{\mbox{\boldmath{$\sigma$}}} {\tilde {\mbox{\boldmath{$F$}}}}^{-t} {\tilde
J}$. In terms of the constitutive relation, $\mbox{\boldmath{$\sigma$}} = - p {\mbox{\boldmath{$I$}}} + 2
\mu {\mbox{\boldmath{$D$}}}$, for a Newtonian fluid,
\[
{\tilde {\mbox{\boldmath{$P$}}}} = \left( - p{\mbox{\boldmath{$I$}}} + \mu \left[ {\tilde \nabla}{\tilde
{\mbox{\boldmath{$v$}}}}{\tilde {\mbox{\boldmath{$F$}}}}^{-1} + \left({\tilde \nabla}{\tilde {\mbox{\boldmath{$v$}}}}{\tilde
{\mbox{\boldmath{$F$}}}}^{-1}\right)^t \right] \right){\tilde {\mbox{\boldmath{$F$}}}}^{-t} {\tilde J}
\hspace{8mm} \mbox{since} \hspace{8mm} {\tilde {\mbox{\boldmath{$D$}}}} = \frac{1}{2}
\left( {\widetilde {\nabla {\mbox{\boldmath{$v$}}}}} + \left( {\widetilde{\nabla {\mbox{\boldmath{$v$}}}}}
\right)^t \right).
\]
The derivation of a variational formulation is along similar lines as
that for the Navier-Stokes equations (the purely Eulerian
description). For a fluid of constant density, the variational
formulation
\renewcommand{\thefootnote}{\fnsymbol{footnote}}
\begin{eqnarray} \label{33}
{\rho} \int_{ \tilde \Omega } {\tilde {\mbox{\boldmath{$w$}}} } \cdot
\frac{\partial {\tilde {\mbox{\boldmath{$v$}}} }}{\partial t} {\tilde J}
{d{\tilde \Omega}} \ + \ {\rho} \int_{ \tilde \Omega } {\tilde
{\mbox{\boldmath{$w$}}}} \cdot {{\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}} }} \left[{\tilde {\mbox{\boldmath{$F$}}}
}^{-1} ({\tilde {\mbox{\boldmath{$v$}}}} - {\tilde {\mbox{\boldmath{$v$}}} }^{ref}) \right] {{\tilde J}}{d {\tilde \Omega}} \ = \nonumber \hspace{40mm} & & \\
{\rho} \int_{ \tilde \Omega } {\tilde {\mbox{\boldmath{$w$}}} } \cdot {\tilde
{\mbox{\boldmath{$b$}}}} {\tilde J} {d {\tilde \Omega}} \ + \ \int_{
\tilde \Omega } {\tilde p} {{{\tilde \nabla} \tilde {\mbox{\boldmath{$w$}}} } : {\tilde
{\mbox{\boldmath{$F$}}} }^{-t}} {\tilde J} {d {\tilde \Omega}} \ - \ 2
{\mu} \int_{ \tilde \Omega } {\tilde {\mbox{\boldmath{$D$}}} }(\tilde {\mbox{\boldmath{$w$}}} ) :
{\tilde {\bf D}}(\tilde {\mbox{\boldmath{$v$}}}) {\tilde J} {d {\tilde
\Omega}} \nonumber & & \\
+ {\rho} \int_{ \tilde \Gamma } {\tilde {\mbox{\boldmath{$w$}}}} {\tilde {\mbox{\boldmath{$P$}}}} {\tilde
{\mbox{\boldmath{$N$}}}} d {\tilde \Gamma} \hspace{60mm} & &
\end{eqnarray}
\begin{eqnarray} \label{34}
\int_{\tilde \Omega} {\tilde q} { {{\tilde \nabla} \tilde {\mbox{\boldmath{$v$}}} } :
{\tilde {\mbox{\boldmath{$F$}}} }^{-t} } {d {\tilde \Omega}} &=& 0
\end{eqnarray}
is obtained, where $\tilde q$ and ${\tilde {\mbox{\boldmath{$w$}}}}$ are respectively the
arbitrary pressure and velocity of the variational formulation. \renewcommand{\thefootnote}{\arabic{footnote}}
\subsection{Reconciling the Different Schools of Thought} \label{1001}
The equations (\ref{29}) and (\ref{30}) are the completely general
reference description of an incompressible, Newtonian fluid. They
reduce to the so--called A.L.E. equations of {\sc Hughes, Liu} and {\sc
Zimmerman} \cite{h:1} for an instant in which spatial and referential
configurations coincide. These simplified equations should, however,
not be implemented where the implementation requires evaluation about
more than one point within each time step (see Section \ref{44} for a
further, in--depth explanation). Under such circumstances the equations
of {\sc Hughes et al.} are an arbitrary Lagrangian Eulerian (A.L.E.)
description in the very true sense (this is not surprising considering
the equations have their origins in the arbitrarily, either Lagrangian
or Eulerian programmes of {\sc Hirt}, {\sc Amsden} and {\sc Cook}
\cite{h:5}). This fact is further borne out in observing that key
energetic properties, consistant with the principle of energy
conservation, are not automatically inherited by the equations of {\sc
Hughes et. al.} in the context of more general references.
The momentum equations of {\sc Soulaimani, Fortin, Dhatt} and {\sc
Ouellet} \cite{f:1} are flawed as a result of the mistaken belief that
${\tilde {\mbox{\boldmath{$\sigma$}}}} {\tilde {\mbox{\boldmath{$F$}}}}^{-1} {\tilde J}$ is the
Piola--Kirchoff stress tensor of the first kind (pg. 268 of {\sc
Soulaimani et al.}). Yet another problem is illustrated by rewriting
the conventional incompressibility condition using the chain
rule. The new incompressibility condition which arises is most
certainly
\[
\frac{\partial \tilde{v}_i}{\partial \tilde{x}_j}\frac{\partial
\tilde{x}_j}{\partial x_i} = 0 \hspace{10mm} \mbox{and not}
\hspace{10mm} \frac{\partial \tilde{v}_i}{\partial
\tilde{x}_j}\frac{\partial \tilde{x}_i}{\partial x_j} = 0.
\]
Further errors arising (eg. $\hat J$ omitted in the first term on the
right hand side of the momentum equation, equation (10) on pg. 268 of
{\sc Soulaimani et al.}) make the use of these equations problematic.
\section{The Energetic Implications of a Deforming Reference}
\label{69}
The effect of quantities parameterising reference deformation on key
energetic properties -- nonlinear, exponential--type dissipation in the
absence of forcing and long--term stability under conditions of time
dependent loading -- is investigated in this section. These properties,
\[
K({\mbox{\boldmath{$v$}}}) \le K({\mbox{\boldmath{$v$}}} \mid_{t_0}) \ e^{-2 {\nu} C t} \hspace{10mm}
\mbox{and} \hspace{10mm} \lim_{t \rightarrow \infty} \sup K({\mbox{\boldmath{$v$}}}) \le
\frac{M^2}{2 \nu^2 C^2}
\]
respectively (where $K = \frac{1}{2} {\rho} \left|\left| {\mbox{\boldmath{$v$}}}
\right|\right|_{L^2(\Omega)}^2$ is the total kinetic energy), are
intrinsic to real flows and the conventional, Eulerian Navier--Stokes
equations (see {\sc Temam} \cite{Temam:1}, \cite{Temam:2}, {\sc
Constantin} and {\sc Foias} \cite{Constantin:1} and {\sc Simo} and {\sc
Armero} \cite{s:1} in this regard). The effect of ${\tilde
{\mbox{\boldmath{$v$}}}}^{ref}$ on the afore mentioned aspects of conservation of the
quantity
\[
\frac{1}{2} {\rho} \left|\left| {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega})}^2
\]
is essentially what is being investigated, with a view to establishing
a set of conditions under which the discrete approximation can
reasonably be expected to inherit these self--same energetic
properties.
One might anticipate key energetic properties to be manifest only in
instances involving a fixed contributing mass of material, whether its
boundaries be dynamic, or not. An analysis of this nature only makes
sense in the context of a constant volume of fluid which, for
simplicity, will have material limits.
Inequalities of the Poincar\'e-Friedrichs type are a key feature of any
stability analysis of this nature. Gradient containing $L^2$ terms need
to be re--expressed in terms of energy. In the case of a ``no slip''
(${\mbox{\boldmath{$v$}}} = 0$) condition on the entire boundary the situation is
straightforward, in that it is possible to use the standard
Poincare-Friedrichs inequality: there exists a constant $C_1 > 0$ such
that
\[
\|\mbox{\boldmath{$v$}}\|_{L^2} \leq C_1 \|\nabla \mbox{\boldmath{$v$}}\|_{L^2}\ \ \mbox{for all}\ \mbox{\boldmath{$v$}} \in
[H^1_0(\Omega)]^n.
\]
The use of the classical Poincar\'{e}--Friedrichs inequality is
otherwise identified as a major limitation, even in the conventional
Navier--Stokes related analyses. The Poincar\'{e}--Friedrichs
inequality is only applicable in very limited instances where the
value for the entire boundary is stipulated to be identically zero.
For boundary conditions of a more general nature, such as those
encountered in this study, in which parts of the boundary may be
either a free surface or subject to traction conditions, a more
suitable inequality is required (notice that subtracting a boundary
velocity and analising the resulting equation is not feasible as the
equations are nonlinear). The Poincar\'{e}--Friedrichs inequality
does, furthermore, not hold on subdomains of the domain in question
and the constant is not optimal.
Further investigation ({\sc communication} \cite{comreddy:1}) reveals a
similar result, the so-called Poincar\'{e}--Morrey inequality, holds
providing the function attains a value of zero somewhere on the
boundary. The Poincar\'e-Morrey inequality states that a constant $C_2
> 0$ exists such that
\[
\|\mbox{\boldmath{$v$}}\|_{L^2} \leq C_2 \|\nabla \mbox{\boldmath{$v$}}\|_{L^2}\ \ \mbox{for all}\ \mbox{\boldmath{$v$}} \in
[H^1_0(\Omega)]^n.
\]
The proof of the Poincar\'e-Morrey inequality is, however, similar to
that of one of Korn's inequalities (see, for example, {\sc Kikuchi}
and {\sc Oden} \cite{kikuchi:1}). In particluar, it is
non--constructive, by contradiction and the constant cannot therefore
be determined as part of the proof. Viewed in this light the
forthcoming inequality amounts to a specification of the hypothetical
constant in the Poincar\'{e}--Morrey inequality for domains of a
particular geometry. The particular types of geometry considered are
those that arise in problems involving the motion of rigid bodies
such as pebbles on the sea bed; thus a free surface is present, and
the domain may be multiply connected.
\begin{inequality}[A New ``Poincar\'{e}'' Inequality] \label{94}
Suppose ${{\mbox{\boldmath{$v$}}}}$ is continuous and differentiable to first order and
that ${{\mbox{\boldmath{$v$}}}}$ attains a maximum absolute value, $c$, on an included,
finite neighbourhood of minimum radius $R_{\mbox{\it \scriptsize min}}$
about a point ${{\mbox{\boldmath{$x$}}}}^{\mbox{\scriptsize origin}}$ (as depicted in
Figure \ref{194}).
\begin{figure}[H]
\begin{center} \leavevmode
\mbox{\epsfbox{rs.eps}}
\end{center}
\caption{A Finite Neighbourhood of Minimum Radius $R_{\mbox{\it
\scriptsize min}}$ About a Point ${{\mbox{\boldmath{$x$}}}}^{\mbox{\scriptsize
origin}}$.} \label{194}
\end{figure}
If ${\Omega}$ is a bounded, star--shaped (about a point
${{\mbox{\boldmath{$x$}}}}^{\mbox{\scriptsize origin}}$)\footnotemark[1] domain in $R^3$,
then \footnotetext[1]{by which is meant that every point in the domain
can be reached by a straight line from ${{\mbox{\boldmath{$x$}}}}^{\mbox{\scriptsize
origin}}$ that does not pass outside of $\Omega$}
\[
\left|\left| {\mbox{\boldmath{$v$}}} \right|\right|_{L^2({ \Omega})} \le { \left[
\frac{(R_{\mbox{\it \scriptsize max}} - R_{\mbox{\it \scriptsize min}})
(R_{\mbox{\it \scriptsize max}}^3 - R_{\mbox{\it \scriptsize min}}^3)}
{3 R_{\mbox{\it \scriptsize max}} R_{\mbox{\it \scriptsize min}}}
\right]^{\frac{1}{2}}} \left|\left| { \nabla} { {\mbox{\boldmath{$v$}}}}
\right|\right|_{L^2({ \Omega})} + \left|\left| c
\right|\right|_{L^2({\Omega})}
\]
where $R_{\mbox{\it \scriptsize max}}$ is the distance to the farthest
point in ${ \Omega}$ from ${{\mbox{\boldmath{$x$}}}}^{\mbox{\scriptsize origin}}$.
\end{inequality}
{\sc Proof:} Consider the change to spherical coordinates
\[
{\breve v}_i(r, \theta, \phi) = { v}_i(r \mathop{\rm sin}\theta
\mathop{\rm cos}\phi - { x}_1^{\mbox{\scriptsize origin}}, r
\mathop{\rm sin}\theta \mathop{\rm sin}\phi - {
x}_2^{\mbox{\scriptsize origin}}, r \mathop{\rm cos}\theta -
{ x}_3^{\mbox{\scriptsize origin}})
\]
centred on ${{\mbox{\boldmath{$x$}}}}^{\mbox{\scriptsize origin}}$. Suppose the radial
limits of the domain and neighbourhood are denoted $R_b (\theta, \phi)$
and $R_a (\theta, \phi)$ respectively. By the fundamental theorem of
integral calculus
\begin{eqnarray*}
\left( {\breve v}_i(r, \theta, \phi) - {\breve v}_i\mid_{R_a (\theta,
\phi)} \right)^2 &=& \left( \int_{R_a (\theta, \phi)}^r \frac{\partial
{\breve v}_i}{\partial r} (\xi, \theta, \phi) d \xi \right)^2 \\
&& \\
&=& \left( \int_{R_a (\theta, \phi)}^{r} \frac{1}{\xi} \xi
\frac{\partial {\breve v}_i}{\partial r} (\xi, \theta, \phi) d \xi
\right)^2 \\
&& \\
&\le& \int_{R_a (\theta, \phi)}^r \frac{1}{\xi^2} d \xi \int_{R_a
(\theta, \phi)}^r \left( \frac{\partial {\breve v}_i}{\partial r}(\xi,
\theta, \phi) \right)^2 \xi^2 d \xi \\
&& \hspace{10mm} \mbox{\it (by Schwarz inequality)} \\
&\le& \int_{R_{\mbox{\it \scriptsize min}}}^{R_{\mbox{\it \scriptsize
max}}} \frac{1}{\xi^2} d \xi \int_{R_a (\theta, \phi)}^{R_b (\theta,
\phi)} \left( \frac{\partial {\breve v}_i}{\partial {r}}(\xi,
\theta, \phi) \right)^2 \xi^2 d \xi \hspace{10mm} \mbox{\it (for } r
\in {\breve \Omega} \mbox{\it)} \\
&& \\
&=& \frac{(R_{\mbox{\it \scriptsize max}} - R_{\mbox{\it \scriptsize
min}})} {R_{\mbox{\it \scriptsize max}} R_{\mbox{\it \scriptsize
min}}} \int_{R_a (\theta, \phi)}^{R_b (\theta, \phi)} \left(
\frac{\partial {\breve v}_i}{\partial r}(\xi, \theta, \phi)
\right)^2 \xi^2 d \xi \\
&& \\
&=& \frac{(R_{\mbox{\it \scriptsize max}} - R_{\mbox{\it \scriptsize
min}})} {R_{\mbox{\it \scriptsize max}} R_{\mbox{\it \scriptsize
min}}} {\breve V}_i (\theta, \phi)
\end{eqnarray*}
\[
\mbox{where} \hspace{10mm} {\breve V}_i (\theta, \phi) = \int_{R_a
(\theta, \phi)}^{R_b (\theta, \phi)} \left( \frac{\partial {\breve
v}_i}{\partial r}(\xi, \theta, \phi) \right)^2 \xi^2 d \xi.
\]
Integrating this result over that part of ${\breve \Omega}$ outside the
neighbourhood (angular extent being $\Theta_a(\phi) \le \theta \le
\Theta_b(\phi)$ and $\Phi_a \le \phi \le \Phi_b$)
\begin{eqnarray*}
&& \hspace{-7mm} \int_{\Phi_a}^{\Phi_b}
\int_{\Theta_a(\phi)}^{\Theta_b(\phi)} \int_{R_a (\theta, \phi)}^{R_b
(\theta, \phi)} \left( {\breve v}_i(r, \theta, \phi) - {\breve
v}_i\mid_{R_a (\theta, \phi)} \right)^2 r^2 \mathop{\rm sin}\theta d r
d \theta d \phi \\
&& \hspace{13mm} \le \frac{(R_{\mbox{\it \scriptsize max}} -
R_{\mbox{\it \scriptsize min}})} {R_{\mbox{\it \scriptsize max}}
R_{\mbox{\it \scriptsize min}}} \int_{\Phi_a}^{\Phi_b}
\int_{\Theta_a(\phi)}^{\Theta_b(\phi)} \int_{R_a (\theta, \phi)}^{R_b
(\theta, \phi)} {\breve V}_i (\theta, \phi) r^2 \mathop{\rm sin}\theta
d r d \theta d \phi \\
&& \\
&& \hspace{13mm} \le \frac{(R_{\mbox{\it \scriptsize max}} -
R_{\mbox{\it \scriptsize min}})} {R_{\mbox{\it \scriptsize max}}
R_{\mbox{\it \scriptsize min}}} \int_{\Phi_a}^{\Phi_b}
\int_{\Theta_a(\phi)}^{\Theta_b(\phi)} {\breve V}_i (\theta, \phi)
\left( \int_{R_{\mbox{\it \scriptsize min}}}^{R_{\mbox{\it \scriptsize
max}}} r^2 d r \right) \mathop{\rm sin}\theta d \theta d \phi \frac{}{}
\\
&& \\
&& \hspace{13mm} \le \frac{(R_{\mbox{\it \scriptsize max}} -
R_{\mbox{\it \scriptsize min}}) (R_{\mbox{\it \scriptsize max}}^3 -
R_{\mbox{\it \scriptsize min}}^3)} {3 R_{\mbox{\it \scriptsize max}}
R_{\mbox{\it \scriptsize min}}} \int_{\Phi_a}^{\Phi_b}
\int_{\Theta_a(\phi)}^{\Theta_b(\phi)} \int_{R_a (\theta, \phi)}^{R_b
(\theta, \phi)} \left( \frac{\partial {\breve v}_i}{\partial r}
\right)^2 {r}^2 \mathop{\rm sin}\theta d r d \theta d \phi \\
&& \\
&& \hspace{13mm} \le \frac{(R_{\mbox{\it \scriptsize max}} -
R_{\mbox{\it \scriptsize min}}) (R_{\mbox{\it \scriptsize max}}^3 -
R_{\mbox{\it \scriptsize min}}^3)} {3 R_{\mbox{\it \scriptsize max}}
R_{\mbox{\it \scriptsize min}}} \int_{\Phi_a}^{\Phi_b}
\int_{\Theta_a(\phi)}^{\Theta_b(\phi)} \int_{R_a (\theta, \phi)}^{R_b
(\theta, \phi)} \left[ \left( \frac{\partial {\breve v}_i}{\partial r}
\right)^2 + \frac{1}{r^2} \left( \frac{\partial {\breve v}_i}{\partial
\theta} \right)^2 \right. \\ && \hspace{83mm} \left. + \frac{1}{r^2
\sin^2 \theta} \left( \frac{\partial {\breve v}_i}{\partial
\phi} \right)^2 \right] r^2 \mathop{\rm sin}\theta d r d \theta d \phi
\\
&& \\
&& \hspace{13mm} = \frac{(R_{\mbox{\it \scriptsize max}} - R_{\mbox{\it
\scriptsize min}}) (R_{\mbox{\it \scriptsize max}}^3 - R_{\mbox{\it
\scriptsize min}}^3)} {3 R_{\mbox{\it \scriptsize max}} R_{\mbox{\it
\scriptsize min}}} \int_{\Phi_a}^{\Phi_b}
\int_{\Theta_a(\phi)}^{\Theta_b(\phi)} \int_{R_a (\theta, \phi)}^{R_b
(\theta, \phi)} \left( \nabla {\breve v}_i \right) \cdot \left( \nabla
{\breve v}_i \right) r^2 \mathop{\rm sin}\theta dr d \theta d \phi.
\end{eqnarray*}
Changing back to the original rectangular coordinates and defining
${{\mbox{\boldmath{$v$}}}}\mid_{\mbox{\it \scriptsize bndry}}$ to be a radially constant
function throughout $\Omega$ which takes the values of ${\breve {\mbox{\boldmath{$v$}}}}
\mid_{R_a (\theta, \phi)}$ for $r = R_a(\theta, \phi)$,
\begin{eqnarray*}
\int_{\Omega_*} \left( { v}_i({ {\mbox{\boldmath{$x$}}}}) - { v}_i\mid_{\mbox{\it
\scriptsize bndry}} \right)^2 d { \Omega} &\le& \frac{(R_{\mbox{\it
\scriptsize max}} - R_{\mbox{\it \scriptsize min}}) (R_{\mbox{\it
\scriptsize max}}^3 - R_{\mbox{\it \scriptsize min}}^3)} {3
R_{\mbox{\it \scriptsize max}} R_{\mbox{\it \scriptsize min}}}
\int_{\Omega_*} \left( \nabla { v}_i({ {\mbox{\boldmath{$x$}}}}) \right) \cdot \left(
\nabla { v}_i({ {\mbox{\boldmath{$x$}}}}) \right) d {\Omega}
\end{eqnarray*}
where $\Omega_*$ is $\Omega$ excluding the neighbourhood. Summing over
$i$,
\begin{eqnarray*}
\int_{\Omega_*} \left( { {\mbox{\boldmath{$v$}}}} - {\mbox{\boldmath{$v$}}}\mid_{\mbox{\it \scriptsize
bndry}} \right) \cdot \left( { {\mbox{\boldmath{$v$}}}} - { {\mbox{\boldmath{$v$}}}}\mid_{\mbox{\it
\scriptsize bndry}} \right) d { \Omega} &\le& \frac{(R_{\mbox{\it
\scriptsize max}} - R_{\mbox{\it \scriptsize min}}) (R_{\mbox{\it
\scriptsize max}}^3 - R_{\mbox{\it \scriptsize min}}^3)} {3
R_{\mbox{\it \scriptsize max}} R_{\mbox{\it \scriptsize min}}}
\int_{\Omega_*} \left( \nabla { {\mbox{\boldmath{$v$}}}} \right) : \left( \nabla { {\mbox{\boldmath{$v$}}}}
\right) d {\Omega}.
\end{eqnarray*}
Making use of either the Cauchy--Schwarz or triangle inequality,
\begin{eqnarray*}
\left( \left|\left| {\mbox{\boldmath{$v$}}} \right|\right|_{L^2({\Omega_*})} - \left|\left|
{\mbox{\boldmath{$v$}}}\mid_{\mbox{\it \scriptsize bndry}} \right|\right|_{L^2({\Omega_*})}
\right)^2 &\le& \frac{(R_{\mbox{\it \scriptsize max}} - R_{\mbox{\it
\scriptsize min}}) (R_{\mbox{\it \scriptsize max}}^3 - R_{\mbox{\it
\scriptsize min}}^3)} {3 R_{\mbox{\it \scriptsize max}} R_{\mbox{\it
\scriptsize min}}} \left|\left| \nabla {\mbox{\boldmath{$v$}}} \right|\right|_{L^2({\Omega_*})}^2,
\end{eqnarray*}
and remembering that $\sup \mid {\breve {\mbox{\boldmath{$v$}}}} \mid_{R_a (\theta,
\phi)} \mid \le c$,
\begin{eqnarray*}
\left|\left| {\mbox{\boldmath{$v$}}} \right|\right|_{L^2({\Omega_*})} &\le& \left[
\frac{(R_{\mbox{\it \scriptsize max}} - R_{\mbox{\it \scriptsize min}})
(R_{\mbox{\it \scriptsize max}}^3 - R_{\mbox{\it \scriptsize min}}^3)}
{3 R_{\mbox{\it \scriptsize max}} R_{\mbox{\it \scriptsize min}}}
\right]^{\frac{1}{2}} \left|\left| { \nabla}{ {\mbox{\boldmath{$v$}}}}
\right|\right|_{L^2({\Omega_*})} + \left|\left| c \right|\right|_{L^2({\Omega_*})}.
\end{eqnarray*}
Consider the terms $\left|\left| {\mbox{\boldmath{$v$}}} \right|\right|_{L^2}$ and
$\left|\left| c \right|\right|_{L^2}$. Comparing these terms under
circumstances of $\sup \mid {\mbox{\boldmath{$v$}}} \mid \le c$ leads to the conclusion
that the inequality holds over the neighbourhood and that the
inequality is therefore unaffected when the domain of integration is
extended to include the neighbourhood. Of course, the radial extension
of ${\mbox{\boldmath{$v$}}}\mid_{\mbox{\it \scriptsize bndry}}$ can be used in place of
$c$ in instances where inclusion of the neighbourhood is not required.
This inequality is similar to the Poincar\'{e}--Friedrichs inequality
when $c=0$, but is extended to a geometrical subclass of domains which
have free and partly non-zero boundaries. It has a further advantage in
that the constant is an order of magnitude more optimal when used under
the ``no slip'' Poincar\'{e}--Friedrichs condition (under such
conditions the domain can always be deconstructed into a number of
subdomains in which $R_{\mbox{\it \scriptsize min}} =
\frac{1}{3}R_{\mbox{\it \scriptsize max}}$). The
Poincar\'{e}--Friedrichs inequality is a special case of the above
inequality. The necessary lemma (below) follows naturally from the
above inequality.
\begin{lemma}[Deviatoric Stress Term Energy] \label{58}
The kinetic energy satisfies the bound $ \ \displaystyle \frac{C}{\rho}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}) \ \le \ \left|\left| {\tilde {\mbox{\boldmath{$D$}}}}({\tilde
{\mbox{\boldmath{$v$}}}}) {\tilde J}^{\frac{1}{2}} \right|\right|_{L^2({\tilde
\Omega})}^2$, where $C$ is related to the constant in Inequality
\ref{94}, $C > 0$.
\end{lemma}
{\sc Proof:} If, in particular, ${ {\mbox{\boldmath{$v$}}}}\mid_{\mbox{\it \scriptsize
bndry}} = 0$ in Inequality \ref{94},
\begin{eqnarray*}
\left|\left| { {\mbox{\boldmath{$v$}}}} \right|\right|_{L^2({ \Omega})} &\le&
\frac{\left|\left| { \nabla} { {\mbox{\boldmath{$v$}}}} \right|\right|_{L^2({
\Omega})}}{\sqrt {C}} \\
C\frac{1}{2}\left|\left| { {\mbox{\boldmath{$v$}}}} \right|\right|_{L^2({ \Omega})}^2
&\le& \left|\left| { {\mbox{\boldmath{$D$}}}}({ {\mbox{\boldmath{$v$}}}}) \right|\right|_{L^2({ \Omega})}^2
\end{eqnarray*}
(The relationship between ${\mbox{\boldmath{$D$}}}$ and ${\nabla} {\mbox{\boldmath{$v$}}}$ arises in the
context of the original equations involving $\mathop{\rm
div}{\mbox{\boldmath{$\sigma$}}}$. It is because \begin{eqnarray*}
\begin{array}{ccll} D_{ij,j} &=& \frac{1}{2} \left( v_{i,jj} + v_{j,ij}
\right) & \\
&=& \frac{1}{2} \left( v_{i,jj} + v_{j,ji} \right)
\hspace{10mm} & \mbox{\em (changing the order of differentiation)} \\
&=& \frac{1}{2} v_{i,jj} & \mbox{\em (${\mathop {\rm div}}{\mbox{\boldmath{$v$}}} = 0$ by
incompressibility}), \end{array} \end{eqnarray*}
assuming, of course, that ${\mbox{\boldmath{$v$}}}$ is continuous and differentiable to
first order.) Rewriting in terms of ${\tilde \Omega}$
\begin{eqnarray*}
\frac{C}{\rho} {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}) &\le& \left|\left| {\tilde
{\mbox{\boldmath{$D$}}}}({\tilde {\mbox{\boldmath{$v$}}}}) {\tilde J}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega})}^2.
\end{eqnarray*}
The following lemma is vital to the deforming reference analysis in
particular. It will form the basis to the next lemma and another (on
page \pageref{354}) concerned with the time discrete analysis.
\begin{lemma} [Basic to Lemmas \ref{355} and \ref{354}] \label{54}
The relation
\[
\left<{\tilde{\mbox{\boldmath{$u$}}}}, ({\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} {\tilde {\mbox{\boldmath{$w$}}}} {\tilde J} \right>_{L^2(\tilde
\Omega)} = - \left< {\tilde {\mbox{\boldmath{$v$}}}}, ({\tilde \nabla}{\tilde{\mbox{\boldmath{$u$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} {\tilde {\mbox{\boldmath{$w$}}}} {\tilde J} \right>_{L^2(\tilde
\Omega)} - \left< {\tilde{\mbox{\boldmath{$u$}}}} \left( {\tilde \nabla}{\tilde
{\mbox{\boldmath{$w$}}}}:{\tilde {\mbox{\boldmath{$F$}}}}^{-t} \right), {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J}
\right>_{L^2(\tilde \Omega)}
\]
is valid for
\[
{\tilde {\mbox{\boldmath{$w$}}}} \in W = \left\{ {\tilde{\mbox{\boldmath{$w$}}}} \ : \ {\tilde {\mbox{\boldmath{$w$}}}} =
{\bf 0} \mbox{ or } {\tilde {\mbox{\boldmath{$F$}}}}^{-t}{\tilde {\mbox{\boldmath{$N$}}}} \cdot {\tilde
{\mbox{\boldmath{$w$}}}} = 0 \mbox{ on } {\tilde \Gamma} \right\}.
\]
\end{lemma}
{\sc Proof:} Consider ${\tilde{\mbox{\boldmath{$u$}}}} \cdot ({\tilde \nabla}
{\tilde{\mbox{\boldmath{$v$}}}}) {\tilde {\mbox{\boldmath{$F$}}}}^{-1} {\tilde {\mbox{\boldmath{$w$}}}} {\tilde J}$:
\begin{eqnarray*}
{\tilde u}_i {\tilde v}_{i,j} {\tilde F}^{-1}_{jk}{\tilde w}_k {\tilde
J} &=& - {\tilde u}_{i,j} {\tilde v}_i {\tilde F}^{-1}_{jk} {\tilde
w}_k {\tilde J} - {\tilde u}_i {\tilde v}_i ({\tilde F}^{-1}_{jk}
{\tilde w}_k {\tilde J})_{,j} + ({\tilde u}_i {\tilde v}_i {\tilde
F}^{-1}_{jk} {\tilde w}_k {\tilde J})_{,j}
\end{eqnarray*}
by the product rule. In the terms arising from $({\tilde F}^{-1}_{jk}
{\tilde w}_k {\tilde J})_{,j}$, both ${\tilde F}^{-1}_{jk,j}$ and ${\tilde J}_{,j} {\tilde F}^{-1}_{jk}$ vanish under the condtions specified (in section \ref{1001}) for equations of {\sc Hughes, Liu} and {\sc
Zimmerman} \cite{h:1} to be a completely general reference description. Thus
\begin{eqnarray*}
{\tilde u}_i {\tilde v}_{i,j} {\tilde F}^{-1}_{jk} {\tilde w}_k {\tilde
J} &=& - {\tilde u}_{i,j} {\tilde v}_i {\tilde F}^{-1} _{jk} {\tilde
w}_k {\tilde J} - {\tilde u}_i {\tilde v}_i {\tilde F}^{-1}_{jk}
{\tilde w}_{k,j} {\tilde J} + ({\tilde u}_i {\tilde v}_i {\tilde
F}^{-1} _{jk} {\tilde w}_k {\tilde J})_{,j}.
\end{eqnarray*}
Integrating over the domain ${\tilde {\Omega}}$ and applying the
divergence theorem,
\begin{eqnarray} \label{309}
\left<{\tilde{\mbox{\boldmath{$u$}}}}, ({\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}}) {\tilde
{\mbox{\boldmath{$F$}}}}^{-1} {\tilde {\mbox{\boldmath{$w$}}}} {\tilde J} \right>_{L^2(\tilde
\Omega)} &=& - \left< {\tilde {\mbox{\boldmath{$v$}}}}, ({\tilde \nabla}{\tilde{\mbox{\boldmath{$u$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} {\tilde {\mbox{\boldmath{$w$}}}} {\tilde J}
\right>_{L^2(\tilde \Omega)} \nonumber \\
&& - \left< {\tilde{\mbox{\boldmath{$u$}}}} \left( {\tilde \nabla}{\tilde {\mbox{\boldmath{$w$}}}}:{\tilde
{\mbox{\boldmath{$F$}}}}^{-t} \right), {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J} \right>_{L^2(\tilde
\Omega)} \nonumber \\
&& + \left< {\tilde{\mbox{\boldmath{$u$}}}}, {\tilde {\mbox{\boldmath{$v$}}}} \left( {\tilde {\mbox{\boldmath{$F$}}}}^{-t}
{\tilde {\mbox{\boldmath{$N$}}}} \cdot {\tilde {\mbox{\boldmath{$w$}}}} \right) {\tilde J}
\right>_{L^2(\tilde \Gamma)}
\end{eqnarray}
The condition of this lemma dictates the manner in which the reference
must deform to ensure that the equation will inherit the desired
energetic properties. This lemma is crucial to the deforming reference
analysis. The lemma immediately below will facilitate the elimination
of the convective energy rate in the forthcoming analysis.
\begin{lemma} [Convective Energy Rate] \label{355} The relation
\begin{eqnarray*}
- {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, ({\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} \left( {\tilde {\mbox{\boldmath{$v$}}}} - {\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde J}
\right>_{L^2({\tilde \Omega})} &=& - \frac{1}{2} \rho \left< {\tilde
{\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$v$}}}} \frac{\partial {\tilde J}}{\partial t}
\right>_{L^2({\tilde \Omega})}
\end{eqnarray*}
is valid in instances where a purely Lagrangian description is used to
track free boundaries and/or boundaries are of a fixed impermeable
type.
\end{lemma}
{\sc Proof:} In instances where a purely Lagrangian description is used
to track free boundaries, ${\tilde {\mbox{\boldmath{$v$}}}} - {\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}}$ vanishes, as does ${\tilde
{\mbox{\boldmath{$F$}}}}^{-t}{\tilde {\mbox{\boldmath{$N$}}}}\cdot{\tilde {\mbox{\boldmath{$v$}}}}$ at fixed impermeable
boundaries. The condition at the boundary for Lemma \ref{54} is
therefore satisfied. Thus the term
\begin{eqnarray*} \label{310}
- {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, ({\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} \left( {\tilde {\mbox{\boldmath{$v$}}}} - {\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde J}
\right>_{L^2({\tilde \Omega})} &=& {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}},
({\tilde \nabla}{\tilde{\mbox{\boldmath{$v$}}}}) {\tilde {\mbox{\boldmath{$F$}}}}^{-1} \left( {\tilde
{\mbox{\boldmath{$v$}}}} - {\tilde {\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde
J} \right>_{L^2(\tilde \Omega)} \nonumber \\
&& + {\rho} \left< {\tilde{\mbox{\boldmath{$v$}}}} \left( {\tilde \nabla}\left( {\tilde
{\mbox{\boldmath{$v$}}}} - {\tilde {\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} \right):{\tilde
{\mbox{\boldmath{$F$}}}}^{-t} \right), {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J} \right>_{L^2(\tilde
\Omega)} \nonumber \\
&& \nonumber \\
&=& {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, ({\tilde \nabla}{\tilde{\mbox{\boldmath{$v$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} \left( {\tilde {\mbox{\boldmath{$v$}}}} - {\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde J}
\right>_{L^2(\tilde \Omega)} \nonumber \\
&& - {\rho} \left< {\tilde{\mbox{\boldmath{$v$}}}} \left( {\tilde \nabla}{\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}}:{\tilde {\mbox{\boldmath{$F$}}}}^{-t} \right),
{\tilde {\mbox{\boldmath{$v$}}}} {\tilde J} \right>_{L^2(\tilde \Omega)} \hspace{5mm}
\mbox{\em (by incomp--} \nonumber \\
&& \hspace{62mm} \mbox{\em ressibility)} \nonumber \\
&=& - \frac{1}{2} {\rho} \left< {\tilde{\mbox{\boldmath{$v$}}}} \left( {\tilde
\nabla}{\tilde {\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}}:{\tilde
{\mbox{\boldmath{$F$}}}}^{-t} \right), {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J} \right>_{L^2(\tilde
\Omega)} \nonumber \\
&& \nonumber \\
&=& - \frac{1}{2} \rho \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$v$}}}}
\frac{\partial {\tilde J}}{\partial t} \right>_{L^2({\tilde \Omega})}
\nonumber
\end{eqnarray*}
since $\displaystyle \frac{\partial {\tilde J}}{\partial t} = {\tilde
J} div {\mbox{\boldmath{$v$}}}^{\scriptsize ref}$ (which is ${\tilde J}{\tilde
\nabla}{\tilde {\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}}:{\tilde
{\mbox{\boldmath{$F$}}}}^{-t}$) in the same vein as ${\dot {\cal J}}_0 = {\cal J}_0
\mathop{\rm div}{\mbox{\boldmath{$v$}}}$ (the kinematic result on used earlier).
This lemma concludes the preliminaries required for the deforming
reference energy analysis.
\subsection{Exponential Dissipation in the Absence of Forcing}
The issue of whether nonlinear, exponential--type dissipation in the
absence of forcing is a property intrinsic to the deforming reference
description is resolved as follows.
\begin{theorem}[Exponential Dissipation in the Absence of Forcing]
\label{60}
A sufficient condition for the completely general reference
description to inherit nonlinear, exponential type energy dissipation
\[
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}) \le {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}} \mid_{t_0})
\ e^{-2 {\nu} C t}
\]
(where ${\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}) \equiv \displaystyle \frac{1}{2}
{\rho} \left|\left| {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega})}^2$)
in the absence of forcing (an intrinsic feature of real flows and the
conventional, Eulerian Navier--Stokes equations) is that the reference
moves in a purely Lagrangian fashion at free boundaries.
\end{theorem}
{\sc Proof:} The first step towards formulating an expression involving
the kinetic energy is to substitute ${\tilde {\mbox{\boldmath{$v$}}}}$ for ${\tilde
{\mbox{\boldmath{$w$}}}}$ in the variational momentum equation (\ref{33}) on page
\pageref{33}. Then
\begin{eqnarray} \label{74}
{\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, \frac{\partial {\tilde {\mbox{\boldmath{$v$}}}}}{\partial
t} {\tilde J} \right>_{L^2(\tilde \Omega)} &=& \left< {\tilde p}
{\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}} , {\tilde {\mbox{\boldmath{$F$}}}}^{-t} {\tilde J}
\right>_{L^2({\tilde \Omega})} - 2 {\mu} \left< {\tilde {\mbox{\boldmath{$D$}}}}({\tilde
{\mbox{\boldmath{$v$}}}}), {\tilde {\mbox{\boldmath{$D$}}}}({\tilde {\mbox{\boldmath{$v$}}}}) {\tilde J} \right>_{L^2(\tilde
\Omega)} \nonumber \\
&& - {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, ({\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} \left( {\tilde {\mbox{\boldmath{$v$}}}} - {\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde J}
\right>_{L^2({\tilde \Omega})} \nonumber \\
&& + {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$b$}}}} {\tilde J}
\right>_{L^2({\tilde \Omega})} + \left< {\tilde{\mbox{\boldmath{$v$}}}}, {\tilde{\mbox{\boldmath{$P$}}}}
{\tilde {\mbox{\boldmath{$N$}}} } \right>_{L^2({\tilde \Gamma})}.
\end{eqnarray}
The term containing the pressure, that is
\[
\left< {\tilde p} {\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}} : {\tilde {\mbox{\boldmath{$F$}}}}^{-t}
{\tilde J} \right>_{L^2({\tilde \Omega})},
\]
vanishes as a result of incompressibility (equation (\ref{30})). The order of integration and differentiation are
interchangeable (limits are time--independent in the reference which
tracks the free boundary perfectly -- a description which becomes fully
Lagrangian at boundaries was stipulated). Equation (\ref{74}) can be
rewritten
\begin{eqnarray*}
\frac{1}{2}{\rho} \left( \frac{d}{dt} \left|\left| {\tilde {\mbox{\boldmath{$v$}}}}
{\tilde J}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega})}^2 -
\left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$v$}}}} \frac{\partial {\tilde
J}}{\partial t} \right>_{L^2({\tilde \Omega})} \right) &=& -2 {\mu}
\left|\left| {\tilde {\mbox{\boldmath{$D$}}}} {\tilde J}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega})}^2 \\
&& - {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, ({\tilde \nabla} {\tilde {\mbox{\boldmath{$v$}}}})
{\tilde {\mbox{\boldmath{$F$}}}}^{-1} \left( {\tilde {\mbox{\boldmath{$v$}}}} - {\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde J}
\right>_{L^2({\tilde \Omega})} \\
&& + {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$b$}}}} {\tilde J}
\right>_{L^2(\tilde \Omega)} + \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde
{\mbox{\boldmath{$P$}}}}{\tilde {\mbox{\boldmath{$N$}}}} \right>_{L^2(\tilde \Gamma)}
\end{eqnarray*}
as a result. The conditions of Lemma \ref{355} are also satisfied for a
description which becomes fully Lagrangian at free boundaries and an
expression
\begin{eqnarray*}
\frac{d {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}})}{dt} &=& -2 {\mu} \left|\left|
{\tilde {\mbox{\boldmath{$D$}}}} {\tilde J}^{\frac{1}{2}} \right|\right|_{L^2({\tilde
\Omega})}^2 + {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$b$}}}} {\tilde J}
\right>_{L^2(\tilde \Omega)} + \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde
{\mbox{\boldmath{$P$}}}}{\tilde {\mbox{\boldmath{$N$}}}} \right>_{L^2(\tilde \Gamma)}
\end{eqnarray*}
is therefore obtained, where ${\tilde K} = \displaystyle \frac{1}{2}
{\rho} \left|\left| {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega})}^2$ is the total kinetic energy.
Using Lemma \ref{58}
\begin{eqnarray} \label{59}
\frac{d {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}})}{dt} &\le& -2 {\nu} C {\tilde
K}({\tilde {\mbox{\boldmath{$v$}}}}) + {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$b$}}}}
{\tilde J} \right>_{L^2(\tilde \Omega)} + \left< {\tilde {\mbox{\boldmath{$v$}}}},
{\tilde {\mbox{\boldmath{$P$}}}}{\tilde {\mbox{\boldmath{$N$}}}} \right>_{L^2(\tilde \Gamma)}.
\end{eqnarray}
Equation (\ref{59}) has a solution of the form
\[
{\tilde K} \le {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}} \mid_{t_0}) \ e^{-2 {\nu} C t}
\]
in the absence of forcing ($\mbox{\it ``no forcing''} \Rightarrow
{\tilde {\mbox{\boldmath{$b$}}}} = {\tilde {\mbox{\boldmath{$P$}}}}{\tilde {\mbox{\boldmath{$N$}}}} = {\bf 0}$), providing a
purely Lagrangian description is used at free boundaries.
A nonlinear, exponential--type energy dissipation in the absence of
forcing is therefore an intrinsic property of the completely general
reference description. This contractive flow property is also an
intrinsic property of the conventional Navier--Stokes equations.
\subsection{Long--Term Stability under Conditions of Time--Dependent
Loading}
The formulation of suitable load and free surface bounds is necessary
before the issue of long-term stability ($L^2$--stability) under
conditions of time--dependent loading can be resolved. The following
lemma facilitates the formulation of load and free surface bounds.
\begin{lemma}[Force, Free Surface Bounds] \label{49} The inequality
\begin{eqnarray*}
{\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$b$}}}} {\tilde J}
\right>_{L^2({\tilde \Omega})} + \left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$P$}}}}
{\tilde {\mbox{\boldmath{$N$}}}} \right>_{L^2(\tilde \Gamma)} &\le& \displaystyle
\frac{\nu C}{2} \left( {\rho} \left|\left| {\tilde {\mbox{\boldmath{$v$}}}} {\tilde
J}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega})}^2 + \left|\left|
{\tilde {\mbox{\boldmath{$v$}}}} \right|\right|_{L^2({\tilde \Gamma})}^2 \right) \\
&& + \frac{1}{2 \nu C} \left( {\rho} \left|\left| {\tilde {\mbox{\boldmath{$b$}}}}
{\tilde J}^{\frac{1}{2}} \right|\right|_{L^2(\tilde \Omega)}^2 +
\left|\left| {\tilde {\mbox{\boldmath{$P$}}}} {\tilde {\mbox{\boldmath{$N$}}}} \right|\right|_{L^2(\tilde
\Gamma)}^2 \right)
\end{eqnarray*}
holds where $\nu C$ is a constant, $\nu C > 0$.
\end{lemma}
{\sc Proof:} In terms of the Cauchy--Schwarz inequality,
\begin{eqnarray*}
\left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$b$}}}} {\tilde J} \right>_{L^2({\tilde
\Omega})} &\le& \left|\left| {\tilde {\mbox{\boldmath{$v$}}}} {\tilde J}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega})} \ \left|\left| {\tilde {\mbox{\boldmath{$b$}}}}
{\tilde J}^{\frac{1}{2}} \right|\right|_{L^2(\tilde \Omega)} \\
&\le& \frac{\nu C}{2} \left|\left| {\tilde {\mbox{\boldmath{$v$}}}} {\tilde
J}^{\frac{1}{2}} \right|\right|_{L^2(\tilde \Omega)}^2 + \frac{1}{2 \nu
C} \left|\left| {\tilde {\mbox{\boldmath{$b$}}}} {\tilde J}^{\frac{1}{2}}
\right|\right|^2_{L^2(\tilde \Omega)} \hspace{5mm} \mbox{for }
\hspace{5mm} \nu C > 0
\end{eqnarray*}
by Young's inequality. Similarly,
\begin{eqnarray*}
\left< {\tilde {\mbox{\boldmath{$v$}}}}, {\tilde {\mbox{\boldmath{$P$}}}} {\tilde {\mbox{\boldmath{$N$}}}}
\right>_{L^2(\tilde \Gamma)} &\le& \frac{\nu C}{2} \left|\left| {\tilde
{\mbox{\boldmath{$v$}}}} \right|\right|_{L^2({\tilde \Gamma})}^2 + \frac{1}{2 \nu C}
\left|\left| {\tilde {\mbox{\boldmath{$P$}}}} {\tilde {\mbox{\boldmath{$N$}}}} \right|\right|^2_{L^2(\tilde
\Gamma)} \hspace{5mm} \mbox{for } \hspace{5mm} \nu C > 0.
\end{eqnarray*}
This done, the mathematical machinery necessary to the long--term
stability analysis is in place.
\begin{theorem}[Long--Term Stability] \label{61}
A sufficient condition for the completely general reference
description to inherit the property of long--term stability
\[
\lim_{t \rightarrow \infty} \sup {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}) \le
\frac{M^2}{2 \nu^2 C^2}
\]
under conditions of time--dependent loading (an intrinsic feature of
real flows and the Navier--Stokes equations), where this
time--dependent loading and the speed of the free surface is bounded in
such a way that
\[
{\rho} \left|\left| {\tilde {\mbox{\boldmath{$b$}}}} {\tilde J}^{\frac{1}{2}} \right|\right|_{L^2(\tilde \Omega)}^2 + \left|\left|
{\tilde {\mbox{\boldmath{$P$}}}}{\tilde {\mbox{\boldmath{$N$}}}} \right|\right|_{L^2({\tilde \Gamma})}^2 +
\nu^2 C^2 \left|\left| {\tilde {\mbox{\boldmath{$v$}}}} \right|\right|_{L^2({\tilde
\Gamma})}^2 \le M^2,
\]
is that the description becomes purely Lagrangian at free boundaries.
\end{theorem}
{\sc Proof:}
Using Lemma \ref{49} in equation (\ref{59}), then applying the above
bound,
\begin{eqnarray*}
\frac{d {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}})}{dt} + \nu C {\tilde K}({\tilde
{\mbox{\boldmath{$v$}}}}) &\le& \frac{M^2}{2 \nu C}.
\end{eqnarray*}
Using the Gronwall lemma (see {\sc Hirsch} and {\sc Smale}
\cite{smale:1}) leads to the differential inequality
\begin{eqnarray*}
\frac{d {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}})}{dt} &\le& \frac{M^2}{2 \nu C} e^{-
\nu C t},
\end{eqnarray*}
which, when solved, yields
\begin{eqnarray*}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}) &\le& e^{- \nu C t}{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}
\mid_{t=t_0}) + \left( 1 - e^{- \nu C t} \right) \frac{M^2}{2 \nu^2
C^2}.
\end{eqnarray*}
This in turn implies
\begin{eqnarray*}
\lim_{t \rightarrow \infty} \sup {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}) &\le&
\frac{M^2}{2 \nu^2 C^2}.
\end{eqnarray*}
The preceding analyses lead to natural notions of nonlinear dissipation
in the absence of forcing and long--term stability under conditions of
time--dependent loading for the analytic problem.
\section{The Energetic Implications of the Time Discretisation}
\label{6}
This section is concerned with establishing a class of time
discretisations which inherit the self--same energetic properties
(nonlinear dissipation in the absence of forcing and long--term
stability under conditions of time dependent loading) as the analytic
problem, irrespective of the time increment employed. In this section a
generalised, Euler difference time--stepping scheme for the completely
general reference equation is formulated and the energetic
implications investigated in a similar vein as the analytic equations
in the previous section.
This stability analysis is inspired by the approach of others to
schemes for the conventional Navier--Stokes equations. The desirability
of the attributes identified as key energetic properties is recognised
and they have been used as a benchmark in the analysis of various of
the conventional, Eulerian Navier--Stokes schemes by a host of authors.
Related work on the conventional, Eulerian Navier--Stokes equations can
be found in a variety of references, for example {\sc Temam}
\cite{Temam:2} and {\sc Simo} and {\sc Armero} \cite{s:1}.
The analyses presented here are extended, not only in the sense that
they deal with the completely general reference equation, but also in
that non--zero boundaries, so--called free boundaries and
time--dependent loads are able to be taken into account (the former two
as a consequence of the new inequality on page \pageref{94}). The
findings of this work have profound consequences for the implementation
of the deforming reference equations. It is significant that many
algorithms used for long--term simulation do not automatically inherit
the fundamental qualitative features of the dynamics.
\subsubsection*{A Generalised Time--Stepping Scheme}
An expression for a generalised Euler difference time--stepping scheme
can be formulated by introducing an ``intermediate'' velocity
\begin{eqnarray} \label{52}
{\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha} \equiv \alpha \ {\tilde {\mbox{\boldmath{$v$}}}}\mid_{t +
\Delta t} + (1 - \alpha) \ {\tilde {\mbox{\boldmath{$v$}}}}\mid_{t} \hspace{5mm}
\mbox{for} \hspace{5mm} \alpha \in [0, 1]
\end{eqnarray}
to the variational momentum equation (equation (\ref{33}) on page
\pageref{33}) where ${\tilde {\mbox{\boldmath{$v$}}}}\mid_t$ and ${\tilde {\mbox{\boldmath{$v$}}}}\mid_{t +
\Delta t}$ are the solutions at times $t$ and $t + \Delta t$
respectively, $\Delta t$ being the time step. It is in this way that a
generalised time--discrete approximation of the momentum equation
\begin{eqnarray} \label{70}
&& \frac{{\rho}}{\Delta t} \left< {\tilde {\mbox{\boldmath{$w$}}}}, ({\tilde {\mbox{\boldmath{$v$}}}}_{n +
1} - {\tilde {\mbox{\boldmath{$v$}}}}_{n }) {\tilde J}_{n + \alpha}
\right>_{L^2({\tilde \Omega}_{n + \alpha})} = \nonumber \\
&& \hspace{20mm} \left< {\tilde p} {\tilde \nabla} {\tilde {\mbox{\boldmath{$w$}}}},
{\tilde {\mbox{\boldmath{$F$}}}}_{n + \alpha}^{-t} {\tilde J}_{n + \alpha}
\right>_{L^2({\tilde \Omega}_{n + \alpha})} - 2 \mu \left< {\tilde
{\mbox{\boldmath{$D$}}}}({\tilde {\mbox{\boldmath{$w$}}}}), {\tilde {\mbox{\boldmath{$D$}}}}({\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha})
{\tilde J}_{n + \alpha} \right>_{L^2({\tilde \Omega}_{n +
\alpha})} \nonumber \\
&& \hspace{20mm} - {\rho} \left<{\tilde {\mbox{\boldmath{$w$}}}}, ({\tilde \nabla}
{\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}) {\tilde {\mbox{\boldmath{$F$}}}}_{n + \alpha}^{-1} \left(
{\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha} - {\tilde {\mbox{\boldmath{$v$}}}}_{n +
\alpha}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde J}_{n + \alpha}
\right>_{L^2({\tilde \Omega}_{n + \alpha})} \nonumber \\
&& \hspace{20mm} + {\rho} \left<{\tilde {\mbox{\boldmath{$w$}}}}, {\tilde {\mbox{\boldmath{$b$}}}}_{n +
\alpha} {\tilde J}_{n + \alpha} \right>_{L^2({\tilde \Omega}_{n +
\alpha})} + \left< {\tilde{\mbox{\boldmath{$w$}}}} , {\tilde{\mbox{\boldmath{$P$}}}}_{n + \alpha} {\tilde
{\mbox{\boldmath{$N$}}}_{n + \alpha}} \right>_{L^2({\tilde \Gamma}_{n + \alpha})}
\end{eqnarray}
is derived, where $\left< \ . \ \right>_{L^2({\tilde \Omega}_{n +
\alpha})}$ denotes the $L^2$ inner product over the deforming domain at
time $t + \alpha \Delta t$. ${\tilde \Gamma}_{n + \alpha}$, ${\tilde
{\mbox{\boldmath{$F$}}}}_{n + \alpha}$, ${\tilde J}_{n + \alpha}$, ${\tilde {\mbox{\boldmath{$D$}}}}_{n +
\alpha}$, ${\tilde {\mbox{\boldmath{$P$}}}}_{n + \alpha}$, and ${\tilde {\mbox{\boldmath{$b$}}}}_{n +
\alpha}$ are likewise defined to be the relevant quantities evaluated
at time $t + \alpha \Delta t$.
It will presently become clear that it makes sense to perform the
analyses for the time--discrete equation in the context of divergence
free rates of reference deformation only. This is since relevant
energy terms are not readilly recovered from the time-discrete
equations for deforming references in general. This investigation is
accordingly restricted to a subclass of reference deformations in
which ``reference volume'' is conserved. This is for reasons of
expedience alone and the subclass of deformations is thought to be
representative.
\begin{assumption} \label{assumption} The assumptions ${\tilde J}_{n} =
{\tilde J}_{n + \alpha}$ and ${\tilde J}_{n + 1} = {\tilde J}_{n +
\alpha}$ are made so that
\[
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_n) = \frac{1}{2} \rho \left|\left| {\tilde
{\mbox{\boldmath{$v$}}}}_n {\tilde J}_n^{\frac{1}{2}} \right|\right|_{L^2({\tilde
\Omega}_n)}^2 = \frac{1}{2} \rho \left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n}
{\tilde J}_{n + \alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde
\Omega}_{n + \alpha})}^2
\]
and
\[
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n+1}) = \frac{1}{2} \rho \left|\left|
{\tilde {\mbox{\boldmath{$v$}}}}_{n+1} {\tilde J}_{n+1}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega}_{n+1})}^2 = \frac{1}{2} \rho
\left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n+1} {\tilde J}_{n +
\alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})}^2
\]
(by equation (\ref{52}) and since the volume of material over which
integration is being performed is constant).
\end{assumption}
{\sc Remark:} Notice that $\displaystyle \frac{{\tilde J}_{n + 1} -
{\tilde J}_{n}}{\Delta t} = {\tilde J} \mathop{\rm
div}{\mbox{\boldmath{$v$}}}^{\mbox{\scriptsize {\it ref}}}_{n + \alpha}$, the discrete
form of $\displaystyle \displaystyle \frac{\partial {\tilde
J}}{\partial t} = {\tilde J} \mathop{\rm div}{\mbox{\boldmath{$v$}}}^{\mbox{\scriptsize
{\it ref}}}$, can consequently be rewritten as
\[
\mathop{\rm
div}{\mbox{\boldmath{$v$}}}^{\mbox{\scriptsize {\it ref}}}_{n + \alpha} = 0
\]
under the conditions of the above assumption. It is for the practical
expedience afforded by Assumption \ref{assumption} alone that this
analysis is limited to instances in which $\mathop{\rm
div}{\mbox{\boldmath{$v$}}}^{\mbox{\scriptsize {\it ref}}}_{n + \alpha} = 0$.
The following lemma will establish that the rate of energy change
associated with the convective term vanishes as a result of the
assumption.
\begin{lemma} [Discrete Convective Energy Rate] \label{354}
The discrete convective term
\[
- {\rho} \left<{\tilde {\mbox{\boldmath{$w$}}}}, ({\tilde \nabla}
{\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}) {\tilde {\mbox{\boldmath{$F$}}}}_{n + \alpha}^{-1} \left(
{\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha} - {\tilde {\mbox{\boldmath{$v$}}}}_{n +
\alpha}^{\mbox{\scriptsize {\em ref}}} \right) {\tilde J}_{n + \alpha}
\right>_{L^2({\tilde \Omega}_{n + \alpha})}
\]
vanishes under circumstances of $\mathop{\rm
div}{\mbox{\boldmath{$v$}}}^{\mbox{\scriptsize {\it ref}}}_{n + \alpha} = 0$ and a purely
Lagrangian description is used at free boundaries (alternatively
boundaries are of the fixed, impermeable type).
\end{lemma}
{\sc Proof:} The operator $\left< \ \cdot \ , ({\tilde \nabla} \ \cdot
\ ) {\tilde {\mbox{\boldmath{$F$}}}}^{-1} {\tilde {\mbox{\boldmath{$w$}}}} {\tilde J} \right>_{L^2({\tilde
\Omega})}$ is skew--symmetric for
\[
{\tilde {\mbox{\boldmath{$w$}}}} \in W = \left\{ {\tilde{\mbox{\boldmath{$w$}}}} \ : \ ({\tilde
\nabla} {\tilde{\mbox{\boldmath{$w$}}}}) : {\tilde {\mbox{\boldmath{$F$}}}}^{-t} = 0 \ \mbox{ on } {\tilde
\Omega}; \ {\tilde {\mbox{\boldmath{$w$}}}} = {\bf 0} \mbox{ or } {\tilde
{\mbox{\boldmath{$F$}}}}^{-t}{\tilde {\mbox{\boldmath{$N$}}}} \cdot {\tilde {\mbox{\boldmath{$w$}}}} = 0 \mbox{ on } {\tilde
\Gamma} \right\}
\]
by equation (\ref{309}) on page \pageref{309}. In instances where a
purely Lagrangian description is used to track free boundaries ${\tilde
{\mbox{\boldmath{$v$}}}}_{n + \alpha} - {\tilde {\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}}_{n
+ \alpha}$ vanishes. At fixed, impermeable boundaries ${\tilde
{\mbox{\boldmath{$F$}}}}_{n + \alpha}^{-t}{\tilde {\mbox{\boldmath{$N$}}}}_{n + \alpha} \cdot {\tilde
{\mbox{\boldmath{$v$}}}}_{n + \alpha}$ vanishes. The condition at the boundary is
therefore satisfied, under all of the afore--mentioned circumstances.
Apply the stipulated condition ${\tilde \nabla}
{\tilde{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em ref}}} : {\tilde {\mbox{\boldmath{$F$}}}}^{-t} = 0$
and set ${\tilde {\mbox{\boldmath{$w$}}}} = {\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha} - {\tilde
{\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\it ref}}}_{n + \alpha}$ etc.
{\sc Remark:}\label{assumpremark} Recall that in the investigation of
the analytic problem, a term arising from the manipulation of the
acceleration containing term (the term containing the rate of change
of the Jacobian) cancelled with the convective energy. It is
therefore not surprising that assumptions pertaining to the
acceleration containing term (in particular to the rate of change of
the Jacobian) in the discrete problem will, once made, also be
necessary for the corresponding discrete convective energy term to
vanish (reffering to the $\mathop{\rm div}{\mbox{\boldmath{$v$}}}^{\mbox{\scriptsize
{\it ref}}} = 0$ condition of Lemma \ref{354}). This is a good
prognosis for the energetic behaviour of the discrete problem in
circumstances of reference deformations excluded by Assumption
\ref{assumption}.
This concludes the preliminaries required for the analysis of the
time--discrete equation.
\subsection{Nonlinear Dissipation in the Absence of Forcing}
The following analysis establishes a class of time--stepping schemes
which exhibit nonlinear dissipation in the absence of forcing
regardless of the time increment employed.
\begin{theorem}[Nonlinear Dissipation in the Absence of Forcing]
\label{55} Suppose that the description is pure Lagrangian at any free
boundaries and that the deformation rate of the reference is divergence
free. A sufficient condition for the kinetic energy associated with the
generalised class of time--stepping schemes to decay nonlinearly
\begin{eqnarray*}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) - {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n })
&\le& - {\Delta t} \ 2 \mu \left|\left| {\tilde {\mbox{\boldmath{$D$}}}}({\tilde
{\mbox{\boldmath{$v$}}}}_{n + \alpha}) {\tilde J}_{n + \alpha}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega}_{n + \alpha})}^2
\end{eqnarray*}
in the absence of forcing and irrespective of the time increment
employed, is that the scheme is as, or more, implicit than central
difference. That is
\[
\alpha \ge \frac{1}{2}.
\]
\end{theorem}
{\sc Proof:} Expressing the intermediate velocities ${\tilde {\mbox{\boldmath{$v$}}}}_{n
+ \frac{1}{2}}$ and ${\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}$ in terms of equation
(\ref{52}) and subtracting, the result
\begin{eqnarray} \label{48}
{\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha} = \left( \alpha - \frac{1}{2} \right)
\left( {\tilde {\mbox{\boldmath{$v$}}}}_{n + 1} - {\tilde {\mbox{\boldmath{$v$}}}}_{n } \right) +
{\tilde {\mbox{\boldmath{$v$}}}}_{n + \frac{1}{2}}
\end{eqnarray}
is obtained. The first step towards formulating an expression
involving the kinetic energy of the generalised time stepping--scheme
(\ref{70}) is to replace the arbitrary vector, ${\mbox{\boldmath{$w$}}}$, with ${\tilde
{\mbox{\boldmath{$v$}}}}_{n + \alpha}$. By further substituting (\ref{48}) into
(\ref{70}) and eliminating the pressure containing term in a similar
manner to that in Theorem \ref{60}, an expression involving the
difference in kinetic energy over the duration of a single time step
is obtained.
The vector ${\tilde {\mbox{\boldmath{$v$}}}} - {\tilde {\mbox{\boldmath{$v$}}}}^{\mbox{\scriptsize {\em
ref}}}$ vanishes in instances where a purely Lagrangian description is
used to track free boundaries. The quantity ${\tilde {\mbox{\boldmath{$F$}}}}_{n +
\alpha}^{-t}{\tilde {\mbox{\boldmath{$N$}}}}_{n + \alpha} \cdot {\tilde {\mbox{\boldmath{$v$}}}}_{n +
\alpha}$ vanishes where boundary conditions are of a fixed impermeable
type. The condition at the boundary for Lemma \ref{354} is therefore
satisfied. Incompressibility and a restriction on reference
deformations to those for which ${\mathop {\rm div}}{\mbox{\boldmath{$v$}}}^{\scriptsize
ref}_{n+\alpha}$ is zero ensure that the remaining Lemma \ref{354}
condition is satisfied.
The equation
\begin{eqnarray} \label{50}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) - {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n })
&=& - {\rho} \left( \alpha - \frac{1}{2} \right) \left|\left| \left(
{\tilde {\mbox{\boldmath{$v$}}}}_{n + 1} - {\tilde {\mbox{\boldmath{$v$}}}}_{n} \right) {\tilde J}_{n +
\alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})}^2 \nonumber \\
&& - {\Delta t} \ 2 \mu \left|\left| {\tilde {\mbox{\boldmath{$D$}}}}({\tilde {\mbox{\boldmath{$v$}}}}_{n +
\alpha}) {\tilde J}_{n + \alpha}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega}_{n + \alpha})}^2 \nonumber +
{\Delta t} {\rho} \left< {\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}, {\tilde {\mbox{\boldmath{$b$}}}}_{n
+ \alpha} {\tilde J}_{n + \alpha} \right>_{L^2({\tilde \Omega}_{n +
\alpha})} \nonumber \\
&& + {\Delta t} \left< {\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}, {\tilde{\mbox{\boldmath{$P$}}}}_{n +
\alpha} {\tilde{\mbox{\boldmath{$N$}}}}_{n + \alpha} \right>_{L^2({\tilde \Gamma}_{n +
\alpha})},
\end{eqnarray}
is then obtained. Since it is assumed that there is no forcing,
\begin{eqnarray*}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) - {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n })
&\le& - {\rho} \left( \alpha - \frac{1}{2} \right) \left|\left| \left(
{\tilde {\mbox{\boldmath{$v$}}}}_{n + 1} - {\tilde {\mbox{\boldmath{$v$}}}}_{n} \right) {\tilde J}_{n +
\alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})}^2 \\
&& - {\Delta t} \ 2 \mu \left|\left| {\tilde {\mbox{\boldmath{$D$}}}}({\tilde {\mbox{\boldmath{$v$}}}}_{n +
\alpha}) {\tilde J}_{n + \alpha}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega}_{n + \alpha})}^2.
\end{eqnarray*}
Thus the kinetic energy inherent to the algorithmic
flow decreases nonlinearly in the absence of forcing, irrespective of
the time increment employed and for arbitrary initial conditions
provided that
\begin{eqnarray*}
\alpha \ge \frac{1}{2} \hspace{10mm} \mbox{and} \hspace{10mm}
\mathop{\rm div} {\mbox{\boldmath{$v$}}}_{n + \alpha}^{\mbox{\scriptsize {\em ref}}} =
0.
\end{eqnarray*}
The former requirement translates directly into one specifying the use
of schemes as, or more, implicit than central difference. Only for
descriptions which become fully Lagrangian at free boundaries can it be
guaranteed that energy will not be artificially introduced by way of
the reference.
{\sc Remark:} Notice (by Lemma \ref{58}) that for $\alpha =
\frac{1}{2}$ an identical rate of energy decay
\begin{eqnarray*}
\frac{{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) - {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n
})}{{\Delta t}} &\le& - 2 \nu C {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha})
\end{eqnarray*}
is obtained for the discrete approximation as was obtained for the
equations.
\subsection{Long--Term Stability under Conditions of Time--Dependent
Loading}
This second part of the time--discrete analysis establishes a class of
time stepping schemes which exhibit long--term stability under
conditions of time dependent loading irrespective of the time increment
employed. The following lemma is necessary to the analysis and is
concerned with devising a bound for the energy at an intermediate point
in terms of energy values at either end of the time step.
\begin{lemma}[Intermediate Point Energy] \label{57} The following bound
applies
\[
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}) \ge \alpha \left( \alpha - c +
\alpha c \right) {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) + (1-\alpha)
\left( 1-\alpha - \displaystyle \frac{\alpha}{c} \right) {\tilde
K}({\tilde {\mbox{\boldmath{$v$}}}}_{n })
\]
where $c$ is some constant, $c > 0$.
\end{lemma}
{\sc Proof:} By Young's inequality
\begin{eqnarray} \label{51}
\left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n+1} {\tilde J}_{n +
\alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})} \ \left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n} {\tilde J}_{n +
\alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})} &\le& \left(\frac{c}{2}\right) \left|\left| {\tilde
{\mbox{\boldmath{$v$}}}}_{n+1} {\tilde J}_{n + \alpha}^{\frac{1}{2}}
\right|\right|_{L^2({\tilde \Omega}_{n + \alpha})}^2 \nonumber \\
&& + \left( \frac{1}{2c} \right) \left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n}
{\tilde J}_{n + \alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde
\Omega}_{n + \alpha})}^2
\end{eqnarray}
for $c > 0$. Writing ${\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n+\alpha})$
explicitly in terms of the ``intermediate'' velocity definition,
(\ref{52}), leads to
\begin{eqnarray*}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}) &=& \alpha^2 {\tilde K}({\tilde
{\mbox{\boldmath{$v$}}}}_{n + 1}) + (1 - \alpha)^2 {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n }) + 2
\alpha(1-\alpha) \left< {\tilde {\mbox{\boldmath{$v$}}}}_{n + 1},{\tilde {\mbox{\boldmath{$v$}}}}_{n}
{\tilde J}_{n + \alpha} \right>_{L^2({\tilde \Omega}_{n + \alpha})} \\
&\ge& \alpha^2 {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) + (1 - \alpha)^2
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n }) \\
&& - 2 \alpha(1-\alpha)\left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n + 1} {\tilde
J}_{n + \alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})} \ \left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n} {\tilde J}_{n +
\alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})} \\
&\ge& \alpha \left[ \alpha - (1 - \alpha) c \right] {\tilde K}({\tilde
{\mbox{\boldmath{$v$}}}}_{n + 1}) + (1-\alpha) \left[ (1-\alpha) - \frac{\alpha}{c}
\right] {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n })
\end{eqnarray*}
using equation (\ref{51}). The optimal choice of the constant $c$ is
established farther on.
The following theorem establishes a class of time--stepping schemes
which exhibit long--term stability under conditions of time--dependent
loading regardless of the time increment employed.
\begin{theorem}[Long--Term Stability] \label{155}
Suppose that the description is pure Lagrangian at any free boundaries
and that the rate at which the reference is deformed is divergence
free. A sufficient condition for the algorithmic flow to exhibit
long--term stability under conditions of time--dependent loading
(intrinsic to real flows and the Navier--Stokes equations), assuming
this time--dependent loading and the speed of the free surface is
bounded in such a way that
\begin{eqnarray*}
{\rho} \left|\left| {\tilde {\mbox{\boldmath{$b$}}}}_{n + \alpha} {\tilde J}_{n +
\alpha}^{\frac{1}{2}} \right|\right|_{L^2({\tilde \Omega}_{n +
\alpha})}^2 + \left|\left| {\tilde {\mbox{\boldmath{$P$}}}}_{n + \alpha}{\tilde {\mbox{\boldmath{$N$}}}}_{n
+ \alpha} \right|\right|_{L^2({\tilde \Gamma}_{n + \alpha})}^2 + \nu^2
C^2 \left|\left| {\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha}
\right|\right|_{L^2({\tilde \Gamma}_{n + \alpha})}^2 &\le& M^2,
\end{eqnarray*}
is
\[
\alpha > \frac{1}{2}.
\]
\end{theorem}
{\sc Proof:} Substituting Lemma \ref{49} (page \pageref{49}) and Lemma
\ref{58} (page \pageref{58}) into equation (\ref{50}), applying the
above bound and choosing $\alpha \ge \frac{1}{2}$ one obtains
\begin{eqnarray*}
\frac{{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) - {\tilde K}({\tilde
{\mbox{\boldmath{$v$}}}}_{n })}{\Delta t} + \nu C {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + \alpha})
&\le& \frac{M^2}{2 \nu C}.
\end{eqnarray*}
From this point on the argument used is identical to that of {\sc
Simo} and {\sc Armero} \cite{s:1} for the conventional, Eulerian
Navier--Stokes equations. Substitution of Lemma \ref{57} leads to a
recurrence relation,
\begin{eqnarray*}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) &\le& \frac{1 - \nu C (1 - \alpha)(1
- \alpha - \frac{\alpha}{c})\Delta t}{1 + \nu C\alpha (\alpha - c +
\alpha c)\Delta t}{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n }) + \frac{M^2 \Delta
t}{2 \nu C \left[ 1 + \nu C\alpha (\alpha - 1 + \alpha c)\Delta t
\right]}.
\end{eqnarray*}
Using this recurrence relation to take cognisance of the energy over
all time steps,
\begin{eqnarray} \label{200}
{\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1}) &\le& \left[ \frac{1 - \nu C(1 -
\alpha)(1 - \alpha - \frac{\alpha}{c})\Delta t}{1 + \nu C\alpha (\alpha
- c + \alpha c)\Delta t} \right]^{n} {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_0)
\nonumber \\
&& + \frac{M^2 \Delta t}{2\nu C \left[ 1 + \nu C\alpha (\alpha - c +
\alpha c)\Delta t \right]} \sum_{k=0}^{n-1} \left[ \frac{(1 - \nu C(1 -
\alpha)(1 - \alpha - \frac{\alpha}{c})\Delta t)}{1 + \nu C\alpha
(\alpha - c + \alpha c)\Delta t} \right]^{k} \nonumber \\
&& \hspace{133mm}
\end{eqnarray}
is obtained. An infinite geometric series which converges so that
\begin{eqnarray*}
\lim_{n \rightarrow \infty} \sup {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1})
&\le& \frac{M^2 \Delta t}{2 \nu C \left[ 1 + \nu C \alpha (\alpha - c +
\alpha c)\Delta t \right]} \left[ 1 \frac{}{}^{}_{} \right. \\
&& \hspace{30mm} \left. - \ \frac{(1 - \nu C(1 - \alpha)(1 - \alpha -
\frac{\alpha}{c})\Delta t)}{1 + \nu C \alpha (\alpha - c + \alpha
c) \Delta t} \right]^{-1} \\
&& \\
&=& \frac{M^2}{2\nu C \left[ \nu C\alpha (\alpha - c + \alpha c) + \nu
C(1 - \alpha)(1 - \alpha - \frac{\alpha}{c}) \right]}
\end{eqnarray*}
results, providing the absolute ratio of the series is less than
unity. That is
\[
\left| \frac{1 - \nu C(1 - \alpha)(1 - \alpha - \frac{\alpha}{c})\Delta
t}{1 + \nu C\alpha (\alpha - c + \alpha c)\Delta t} \right| < 1.
\]
Therefore either
\[
- 1 - \nu C\alpha(\alpha - c + \alpha c)\Delta t < 1 - \nu C(1 -
\alpha) \left (1 - \alpha - \frac{\alpha}{c} \right) \Delta t
\]
or
\begin{eqnarray} \label{202}
1 - \nu C(1 - \alpha) \left (1 - \alpha - \frac{\alpha}{c} \right)
\Delta t < 1 + \nu C\alpha(\alpha - c + \alpha c)\Delta t
\end{eqnarray}
in order for the bound to exist. Notice, furthermore, that for this
desired convergence to be unconditional (regardless of the time
increment employed) requires
\begin{eqnarray} \label{201}
\alpha - c + \alpha c \ge 0.
\end{eqnarray}
The denominator in the series ratio might otherwise vanish for some
value of $\Delta t$.
For $\alpha \in \left[\frac{1}{2}, 1\right]$ equation (\ref{202}) and
equation (\ref{201}) together imply
\begin{eqnarray*}
\frac{(1 - \alpha)}{\alpha} < c \le \frac{\alpha}{(1 -
\alpha)}
\end{eqnarray*}
which in its turn implies
\begin{eqnarray*}
\frac{(1 - \alpha)}{\alpha} < \frac{\alpha}{(1 - \alpha)}.
\end{eqnarray*}
The choice of the parameter $\alpha > \frac{1}{2}$ therefore leads to
an infinite geometric series which forms the desired upper bound. The
minimum value of this bound occurs for $c$ chosen according to
\begin{eqnarray*}
\inf_{\frac{(1-\alpha)}{\alpha} < c \le \frac{\alpha}{(1-\alpha)}}
\frac{1}{\nu C\alpha(\alpha - c + \alpha c) + \nu C(1 - \alpha)(1 -
\alpha - \frac{\alpha}{c})} &=& \frac{1}{\nu C (2\alpha - 1)^2}.
\end{eqnarray*}
The value of this upper bound, which occurs for the choice of the
parameter $\alpha > \frac{1}{2}$, is then
\begin{eqnarray*}
\lim_{n \rightarrow \infty} \sup {\tilde K}({\tilde {\mbox{\boldmath{$v$}}}}_{n + 1})
&\le& \frac{M^2}{2 \nu^2 C^2 (2\alpha - 1)^2}.
\end{eqnarray*}
In this way one arrives at a class of algorithms which are
unconditionally (irrespective of the time increment employed) stable.
{\sc Remark:} Notice that for $\alpha = 1$ one obtains an identical
energy bound for the discrete approximation as was obtained for the
equations.
\section{An ``Updated'' Approach and a Simplified Implementation}
\label{44}
Ever burgeoning deformation gradients accumulate for a straight forward
implementation of the equations. Using an ``updated'' approach is one
way of coping with this otherwise rather daunting prospect. An
``updated'' approach is the result of a little, well--worthwhile
lateral thinking. An ``updated'' approach amounts to choosing a new
referential configuration during each time step.
In the case of time stepping schemes based about a single instant (eg.
the generalised class of Euler difference schemes investigated in
Section \ref{6}) a considerably simplified implementation can further
be achieved by a particularly appropriate choice of configurations.
Making the choice of a referential configuration which coincides with
the spatial configuration at the instant about which the time stepping
scheme is based allows the deformation gradient to be omitted
altogether (the deformation gradient is identity under such
circumstances). For such implementations (those which require
evaluation about a single point only) no error arises from the use of
the equations cited in {\sc Hughes, Liu} and {\sc Zimmerman}
\cite{h:1},
\begin{eqnarray}
{\rho} \left( \frac{\partial { {\mbox{\boldmath{$v$}}} }}{\partial t} + {{{
\nabla} {\mbox{\boldmath{$v$}}} } } ({ {\mbox{\boldmath{$v$}}} } - { {\mbox{\boldmath{$v$}}} }^{ref})
\right) &=& {\rho} { {\mbox{\boldmath{$b$}}} } + \mathop{ {\rm
div}}{ {\mbox{\boldmath{$\sigma$}}}} \label{307} \\
\mathop{ {\rm div}}{ {\mbox{\boldmath{$v$}}}} &=& 0. \label{308}
\end{eqnarray}
These equations are not valid for any, arbitrary choice of reference
or if the implementation requires the equation to be evaluated at
more than one point within each time step (eg. a Runge--Kutta or
finite--element--in--time scheme). It is important to remember that
in a discrete context the reference configuration is fixed for the
duration of the entire time increment. Although the referential
configuration is hypothetical and can be chosen arbitrarily for each
time step, once chosen it is static for the duration of the entire
time step. Once the coincidence of configurations is ordained at a
given instant, ${\tilde {\mbox{\boldmath{$F$}}}}$ is defined by the deformation, both
before and after, and must be consistant.
There would seem to be no reason why one would wish to define the
deformation about a configuration other than that at the instant about
which the implementation is based (assuming the implementation used is
indeed based about a single point eg. a finite difference) thereby
involving deformation gradients. Resolving the resulting difficulties
associated with the deformation gradients by means of a perturbation
seems unnecessarily complicated in the light of the above reasoning.
\section{Conclusions}
The correct equations, which describe the motion of an
incompressible, Newtonian fluid and which are valid for a completely
general range of reference deformations, are equations (\ref{29}) and
(\ref{30}). For implementations requiring the equations to be
evaluated about a single instant within each time step only (eg.
finite differences), the deformation gradients may be assumed
identity i.e. the equations of {\sc Hughes, Liu} and {\sc Zimmerman}
\cite{h:1} (equations (\ref{307}) and (\ref{308})) will suffice.
In this work it is shown (as was hoped) that nonlinear,
exponential--type dissipation in the absence of forcing and long--term
stability under conditions of time dependent loading are properties
automatically inherited by deforming reference descriptions. The single
provisor is that such descriptions become fully Lagrangian at any
moving boundaries. These properties are intrinsic to real flows and the
conventional, Eulerian Navier--Stokes equations.
Relevant energy terms are not readily recovered from the
time--discrete equations for deforming references in general. Only for
divergence free rates of reference deformation which become fully
Lagrangian at free boundaries could it consequently be guaranteed that
energy would not be artificially introduced to the algorithmic flow by
way of the reference. The divergence free assumption was made for
reasons of expedience alone and the limitations of the time--discrete
analysis are consequently not expected to detract from the use of the
method in any way. This is especially so when it is considered that, a
term arising from the manipulation of the acceleration containing term
(the term containing the rate of change of the Jacobian) cancelled with
the convective energy in the investigation of the analytic problem and
that assumptions pertaining to the acceleration containing term (in
particular to the rate of change of the Jacobian) in the discrete
problem were, once made, also necessary for the corresponding discrete
convective energy term to vanish (reffering to the $\mathop{\rm
div}{\mbox{\boldmath{$v$}}}^{\mbox{\scriptsize {\it ref}}} = 0$ condition of Lemma
\ref{354}). If one were to be overly cautious on this basis one would
be faced with the additional challenge of enforcing a fully Lagrangian
description for nodes situated on any free boundaries, while deforming
elements would be required to deform at a rate which is divergence
free. Such a totally divergence free description may, however, not be
possible. An alternative strategy would be to use a fully Lagrangian
description. Both the purely Lagrangian and purely Eulerian fluid
descriptions have divergence free rates of distortion.
There are inherent problems with using certain classes of
time--stepping schemes and the use of finite difference schemes more
implicit than central difference is consequently advocated. Such
differences exhibit the key energetic properties (nonlinear,
exponential--type dissipation in the absence of forcing and long--term
stability under conditions of time dependent loading) irrespective of
the time increment employed. A backward difference is the obvious
choice. Calculations at time $t + \alpha \Delta t$ would require an
intermediate mesh and associated quantities for instances in which
$\alpha \ne 1$ (since $\alpha > \frac{1}{2}$).
The author recommends a strategy in which a predominantly Eulerian
description is used, where possible, for the bulk of the problem (from
an efficiency point of view) and the completely general reference
description for the remainder is appropriate. Purely Eulerian
descriptions have the advantage of a ``one off'' finite element
construction and involve none of the hazards of a badly distorted
reference.
\section{Acknowledgements}
Grzegorz Lubczonok and Ronald Becker are thanked for their respective opinions on the inequality on page \pageref{94}, as is Daya Reddy. The use of Kevin Colville/George Ellis' printer is also gratefully acknowledged.
|
\section{Introduction}
The standard model has been verified to remarkable accuracy\cite{Altarelli}
down to scales of $10^{-15}$ cm, corresponding to energies up to $\simeq
100$ GeV. With the discovery of the top quark with a mass $m_t=175.6(5.5)$
GeV, all the required fermions in the standard model are now in place.
The Higgs particle, which represents a vital missing element in the
standard model, is yet to be discovered. When it is found, we could just
declare that particle physics is closed. However, there are conceptual
difficulties with the standard model, which point to new physics beyond it.
A successful unified theory of gravity and the standard model should at
least accomplish the following:
\begin{enumerate}
\item
Resolve the gauge hierarchy problem. The gauge hierarchy (or 't Hooft
naturalness) problem besets the Higgs sector. The standard model
cannot naturally explain the relative smallness of the weak scale of mass,
set by the Higgs mechanism at $M_{\rm WS}\sim 250$ GeV.
\item
Reduce the number of unknown parameters.
\item
Explain the origin of the three fermion generations in the standard model.
\item
Provide a mathematically consistent quantum gravity theory which leads to
finite scattering amplitudes to all orders in perturbation theory.
\item
Guarantee that the proton remains stable in accordance with the current
experimental lower bound on its decay lifetime $\tau_p$.
\end{enumerate}
A leading candidate for a unified theory of the standard model and gravity
has been superstring theory\cite{Schwarz}. String theory
bypasses the problem of ultraviolet divergences of gravity by replacing
the fundamental point-like object by a string, an extended one-dimensional
object. String compactifications lead to a myriad of possible vacuum states
as models of low energy particle physics, but recent developments in
duality and p-branes\cite{Duff} have led to new possibilities for
unification models. D-branes are associated with gauge fields living in
their world volume\cite{Witten,Dienes,Shiu}. The standard model gauge group
would correspond to gauge fields living in the world volume of 3-branes. If
the gauge group comes from open strings starting and ending on a set of
p-branes, then the string scale $M_s$ can be lowered much below the Planck
mass scale, $M_{\rm Planck}\sim 10^{19}$ GeV, by using the formula (for
3-branes): $M^4_s=\alpha_{\rm GUT}M_c^3 M_{\rm Planck}/\sqrt{2}$, where
$M_c$ is the compactification scale. If $M_s\sim 1$ TeV, then predictions
could be made that might be checked by the new generation of accelerators.
Dienes et al.\cite{Dienes2} have shown that a Kaluza-Klein orbifold
reduction can lead to a gauge coupling unification at a much lower energy
scale than the usual GUT scale with power law behaviour instead of the
familar logarithmic scaling behaviour. A more radical idea has been to
reduce the Planck scale of gravity to the TeV energy region\cite{Arkani},
thus resolving the hierarchy problem.
In previous work\cite{Moff}, the author developed a model based on the
gauge group ${\cal G}=SO(3,1)\times SU_c(3)\times SU(2)\times U(1)$ in four
dimensions. No attempt was made to unify the standard model with
gravity. The unknown parameters such as coupling constants and fermion
masses were to be determined by a relativistic, Schr\"odinger-type
eigenvalue equation, using the perturbatively finite and unitary formalism
of finite quantum field theory (FQFT)[10-18]. The stability of the proton
was guaranteed, for quarks and leptons were not combined in the same
irreducible fermion representation.
If, indeed, baryon number is conserved, then the small but non-zero matter
content of the universe is simply a matter of the initial conditions, and
its value cannot be explained within the standard domain of physics, which
is not a satisfactory state of affairs. Moreover, it is difficult to find
non-trivial solutions to the relativistic eigenvalue equation for the
mass spectrum and the coupling constants. In view of this, it is tempting
to pursue further the possibility of discovering a unified field theory of
gravity and the gauge fields of the standard model.
In the following, we shall pursue such a possible theory based on a
higher-dimensional unified field theory. In the early
eighties there was a revival of attempts to build a unified theory using
the idea of a Kaluza-Klein pure gravity theory in
D-dimensions\cite{Kaluza}. These ideas were abandoned when it was
discovered that assuming a Riemannian geometry in D-dimensions and an
internal compact space, the models failed to predict flavor chiral fermions
in the four-dimensional theory\cite{Wetterich}. Beginning with a spinor
coupled to gravity in D dimensions, one always ends in the four-dimensional
theory with vector-like fermion representations of the gauge group.
Applying a theorem due to Atiyah and Hirzebruch\cite{Hirzebruch}, it is
found that the number of chiral fermions derived from dimensional reduction
of a Weyl spinor coupled to Riemannian geometry with $D-4$ dimensions
describing a compact, orientable manifold without boundary is zero. This
situation is also found to exist for $N=8$ supergravity theories.
Moreover, except for certain special choices of parameters, the fermions
have masses of order the Planck mass. An explanation of the observed small
fermion masses requires chiral fermions, where the left-handed and
right-handed Weyl or Weyl-Majorana spinors belong to inequivalent
representations of the low energy group. Another reason for abandoning the
Kaluza-Klein approach to unification is that the gravity theory is not
renormalizable. The usual ultraviolet divergences that plague quantum
gravity are present, as in the standard point particle four-dimensional
quantum gravity theory.
We shall base our unification theory on a D-dimensional Einstein-Yang-
Mills-Higgs field theory, with supplementary gauge fields coupled to
gravity, which has chiral fermion representations corresponding
to massless Dirac modes, and is free of anomalies. We consider an
$SO(18)\supset SO(10)\times SO(8)$ model in twelve dimensions in which the
eight-dimensional internal space is described by spinor connection gauge
fields, associated with an $SO(8)$, which are topologically non-trivial. By
dimensional reduction the $SO(18)$ leads to the four-dimensional grand
unified theory (GUT) $SO(10)$ and the $SO(8)\supset Sp(4)\times SU(2)$ with
three families of chiral quarks and leptons.
The FQFT gauge formalism is applied to the D-dimensional theory to
guarantee a self-consistent quantum gravity theory coupled to the
Yang-Mills, Higgs and spinor fields. The formalism is free of tachyons and
unphysical ghosts and satisfies unitarity to all orders of perturbation
theory. It could incorporate supersymmetry if required, in the form of a
supergravity theory, but we shall not do so here, in order to aim for as
minimal a scheme as possible. The gauge hierarchy problem is resolved
because the finite scalar Higgs self-energy loop graphs are damped
exponentially at high energies above the physical Higgs scale $\Lambda_H$
set by the FQFT formalism and by choosing $\Lambda_H\sim 1$ TeV. The
compactification scale $M_c$ can be as low as $M_c\simeq 1-10$ TeV, while
the quantum gravity scale $\Lambda_G\sim 1-10$ TeV. This would predict that
at these energies future high-energy experiments could detect Kaluza-Klein
excitation modes at energies of several TeV.
If we choose $\Lambda_G\sim 1-10$ TeV, then quantum gravity loop
corrections are perturbatively weak all the way to the Planck energy. This
would obviate the need to find a non-perturbative quantum gravity
formalism.
\section{\bf Kaluza-Klein Theory and The Ground State}
We shall begin with the action:
\begin{equation}
W=W_{\rm grav}+W_{YM}+W_{\rm H}+W_{\rm Dirac},
\end{equation}
where
\begin{equation}
W_{\rm grav}=-\frac{1}{\kappa^2}\int d^Dz\sqrt{-g}(R+\lambda),
\end{equation}
\begin{equation}
W_{\rm YM}=-\frac{1}{4}\int d^Dz\sqrt{-g}\,{\rm tr}(F^2),
\end{equation}
\begin{equation}
W_{\rm H}=-\frac{1}{2}\int d^Dz\sqrt{-g}[D_M\phi^aD^M\phi^a+V(\phi^2)],
\end{equation}
\begin{equation}
W_{\rm Dirac}=\frac{1}{2}\int
d^Dz\sqrt{-g}\bar{\psi}\Gamma^Ae_A^M[\partial_M\psi-\omega_M\psi -{\cal
D}(A_M)\psi]+h.c. \end{equation}
Here, we use the notation: $z^M=(x^\mu;
\mu=0,1,2,3, y^m; m=1,2,...,D-4), M=0,...,D, g={\rm det}(g_{MN})$. The
Riemann tensor is defined such that
${{R_{LM}}^K}_N=\partial_L{\Gamma_{MN}}^K-\partial_M{\Gamma_{LN}}^K+
{\Gamma_{LC}}^K{\Gamma_{MN}}^C-{\Gamma_{MC}}^K{\Gamma_{LN}}^C$.
Moreover, h.c. denotes the Hermitian conjugate, $\bar\psi=\psi^{\dagger}
\Gamma^0$, and $e^M_A$ is a vielbein, related to the metric by
\begin{equation}
g_{MN}=\eta_{AB}e_M^Ae_N^B,
\end{equation}
where $\eta_{AB}$ is the D-dimensional Minkowski metric tensor associated
with the flat tangent space with indices A,B,C... Moreover, $F^2=
F_{MN}F^{MN}$, $R$ denotes the scalar curvature, $\lambda$ is
the cosmological constant and
\begin{equation}
F_{aMN}=\partial_NA_{aM}-\partial_MA_{aN}-ef_{abc}A_{bM}A_{cN},
\end{equation}
where $A_{aM}$ are the gauge fields of the Yang-Mills group with
generators $f_{abc}$ and $e$ is the coupling constant. $\kappa^2=16\pi
\bar G $, where $\bar G$ is related to Newton's constant $G$ by $\bar
G=GV$ and $V$ is the volume of the internal
space. The dimensions of $\bar G$ are $(\rm length)^{\delta}$($\delta=D-2$). Moreover,
$D_M$ is the covariant derivative operator:
\begin{equation}
D_M\phi^a=\partial_M\phi^a+ef^{abc}A_M^b\phi^c.
\end{equation}
The Higgs potential $V(\phi^2)$ is of the form leading to spontaneous
symmetry breaking
\begin{equation}
V(\phi^2)=\frac{1}{4}g(\phi^a\phi^a-K^2)^2+V_0,
\end{equation}
where $V_0$ is an adjustable constant and the coupling constant $g > 0$.
The spinor field is minimally coupled to the gauge potential $A_M$, and
${\cal D}$ is a matrix representation of the gauge group $G$ defined in
D-dimensions. The spin connection $\omega_M$ is
\begin{equation}
\omega_M=\frac{1}{2}\omega_{MAB}\Sigma^{AB},
\end{equation}
where $\Sigma^{AB}=\frac{1}{4}[\Gamma^A,\Gamma^B]$ is the spinor matrix
associated with the Lorentz algebra $SO(D-1,1)$. The components
$\omega_{MAB}$ satisfy
\begin{equation}
\partial_Me^K_A+{\Gamma_{MN}}^Ke^N_A-{\omega_{MA}}^Be_B^K=0,
\end{equation}
where ${\Gamma_{MN}}^K$ is the Christoffel symbol.
The field equations for the gravity-Yang-Mills-Higgs-Dirac sector are
\begin{equation}
R_{MN}-\frac{1}{2}g_{MN}R=-\frac{1}{2}\kappa^2(T_{MN}-\lambda g_{MN}),
\end{equation}
\begin{equation}
g^{LM}\nabla_LF_{MN}=g^{LM}\biggl(\partial_LF_{MN}-{\Gamma_{LM}}^KF_{KN}
-{\Gamma_{LN}}^KF_{MK}
$$
$$
+[A_L,F_{MN}]\biggr)=0,
\end{equation}
\begin{equation}
\frac{1}{\sqrt{-g}}D_M[\sqrt{-g}g^{MN}D_N\phi^a]
=\biggl(\frac{\partial V}{\partial\phi^2}\biggr)\phi^a,
\end{equation}
\begin{equation}
\Gamma^Ae^M_A[\partial_M-\omega_M-{\cal D}(A_M)]\psi=0.
\end{equation}
The Yang-Mills-Higgs contribution to the energy-momentum tensor is
\begin{equation}
T_{MN}^{\rm YMH}={\rm tr}(F_{MK}F^K_N)+D_M\phi^aD_N\phi^a
-\frac{1}{2}g_{MN}\biggl[\frac{1}{2}{\rm tr}(F^2)
$$
$$
+D_P\phi^aD^P\phi^a+V(\phi^2)\biggr].
\end{equation}
We must now choose an ansatz for the ground state of the four-dimensional
world. A general theory would start by assuming the ground state to be
$M^4\times B$, where $M^4$ is four-dimensional Minkowski space and $B$ is
a compact internal space. A simple ansatz for the compact space $B$ is
to assume a symmetric solution with the structure, $M^4\times S/H$, where
$S/H$ is a coset space of dimension $D-4$. For the metric we take
\begin{equation}
g_{MN}dz^Mdz^N=g_{\mu\nu}(x)dx^\mu dx^\nu+g_{mn}(y)dy^m
dy^n,
\end{equation}
\begin{equation}
A^a_Mdz^M=A^a_\mu dy^\mu.
\end{equation}
One possible symmetric choice for the ground state four-dimensional
spacetime is that $g_{\mu\nu}$ is a de Sitter solution of the
four-dimensional Einstein equation: \begin{equation}
R_{\mu\nu}=-\frac{1}{2}\kappa^2\lambda g_{\mu\nu}.
\end{equation}
Since we have additional Yang-Mills and Higgs scalar fields in our
higher-dimensional theory, it is possible for us to obtain classical
solutions to the field equations in spacetime as the product of flat
four-dimensional Minkowski spacetime and an internal compact space. This
"spontaneous compactification" can be achieved by going beyond a pure
Kaluza-Klein theory, which does not allow a flat four-spacetime unless the
curvature of the internal space is also zero\cite{Cremmer,Luciani}.
If the metric $g_{MN}$ for the space $M^4\times B$ is by construction a
solution of the Killing equation for $M^4\times B$, then one can carry out
the integrations over the $y$ coordinates in the action, and the dynamical
variables in the theory are functions of $x$ only. The dimensional
reduction reduces some of the gauge fields $A_M$ to $A_\mu(x)$, while
certain linear combinations of the gauge fields $A_m(x,y)$ become
geometrical scalar fields in four dimensions. However, the latter scalar
fields do not in general lead to a spontaneous symmetry breaking
Higgs mechanism, so that our additional Higgs field action $W_H$
is required to perform this task.
\section{\bf Reduction to a Flavor-Chiral Theory}
The problem of obtaining the correct quark and lepton quantum numbers is
more subtle and difficult than one might suspect in both Kaluza-Klein
theories and in string theories. One of the most striking features of
particle physics is the knowledge that the quantum numbers of fermions
{\it are not vector-like}, i.e., that left-handed fermions
transform under $SU(3)\times SU(2)\times U(1)$ differently from the way
right-handed fermions transform. Left-handed quarks are $SU(2)$ doublets
but right-handed quarks are $SU(2)$ singlets. Fermions of given helicity
form a complex representation of $SU(3)\times SU(2)\times U(1)$, so the
fermion representations are not self-conjugate. This fact plays an
important role in unified theories, because it means that the bare masses
of the quarks and leptons are ruled out by gauge invariance. The fermions
can acquire mass only through spontaneous symmetry breaking. Thus, arises
the problem of explaining the relative lightness of observed fermions. What
generates the smallness of the $SU(2)\times U(1)$ breaking scale?
Since the quantum numbers of the fermions are not vector-like, the spectrum
of light fermions depends only on universality class features of an
$SU(3)\times SU(2)\times U(1)$ invariant theory, i.e. the lightness of the
fermions cannot be modified by any $SU(3)\times SU(2)\times U(1)$
invariant perturbations. We shall consider for the present that the light
fermions are massless, ignoring the $SU(2)\times U(1)$ breaking.
Another striking feature of the fermion spectrum is that the anomalous
triangle graphs cancel, an important ingredient in a successful gauge
invariant unified theory. In addition, each family of fermions consists of
five irreducible representations of $SU(3)\times SU(2)\times U(1)$ and
their exists a redundancy of three families. The $SU(5)$ and $SO(10)$ grand
unified models successfully describe the fermion structure in one family
in terms of the representation
$\underline{\overline{5}}_L+\underline{10}_L$ in $SU(5)$, and
$\underline{16}_L$ in $SO(10)$\cite{Glashow,Fritzsch}. As far as the
replication of families is concerned, it seems natural and elegant to
describe this replication in terms of spinor representations of $SO(N)$ for
$N\geq 18$.
Witten\cite{Witten2} proved by using topological arguments that, in any
number of dimensions, the Dirac operator in a pure Kaluza-Klein
theory cannot admit a chiral spectrum. Wetterich\cite{Wetterich} showed
that the spectrum of fermions could be chiral only if the dimensionality of
the space is 2 mod 8. Only if additional Yang-Mills gauge fields are
included in a higher-dimensional theory can the fermion spectrum become
chiral under dimensional reduction. However, this is only true if the
compactification involves a topologically non-trivial configuration of
these gauge fields. The addition of extra gauge symmetries is indeed the
mechanism whereby superstring theories lead to a spectrum of chiral
fermions.
In higher-dimensional theories with a ground state $M^4\times B$, a
massless spectrum of particles is generated as zero modes of wave operators
on the internal compact space $B$. A massless fermion particle in $D=4+n$
dimensions obeys
\begin{equation}
\label{Dirac}
\Gamma^MD_M\psi=0,
\end{equation}
where $\Gamma^M$ are the gamma matrices. The quantum numbers remain
unchanged in the presence of non-minimal couplings, so we ignore them in
the present discussion. We can separate (\ref{Dirac}) into
\[
D^{(4)}\psi+D^{(n)}\psi=0,
\]
where $D^{(4)}=\Gamma^\mu D_\mu$ and $D^{(n)}=\Gamma^mD_m$, and we see that
$D^{(n)}$ is a mass operator, whose eigenvalues are the experimentally
determined fermion masses. The zero eigenvalues are the massless fermions.
In order to see that a pure Riemannian Kaluza-Klein theory
has difficulties, we employ an argument due to
Lichnerowicz\cite{Lichnerowicz}. If we square the internal Dirac operator,
we get $(iD^{(n)})^2=-D_mD^m+\frac{1}{4}R$,
and because $-D^mD_m$ is a non-negative operator, then if $R>0$
everywhere, it follows that the Dirac operator has no zero eigenvalues,
i.e. there are no massless fermions. By using the Atiyah-Hirzebruch
theorem\cite{Hirzebruch}, which states that the character-valued index of
the Dirac operator vanishes on any manifold with a continuous symmetry
group (in any even number of dimensions), Witten proved in general that a
compact Riemannian manifold (such as the internal space of a pure
Kaluza-Klein theory) does not possess zero mass chiral fermions.
We shall now assume that we have additional Yang-Mills gauge fields. If the
gauge quantum numbers of the fermions are vector-like, they will remain
vector-like after compactification, unless special Majorana and
Weyl-Majorana conditions are imposed together with gauge symmetry
conditions\cite{Slansky}. A serious problem of cancellation of
anomalies will occur, unless careful attention is payed to the gauge
couplings and the compactification. Non-vector-like couplings of gauge
fields will lead to anomalies, unless the condition is
satisfied\cite{Witten2}:
\begin{equation}
\label{anomaly}
{\rm tr}(M_L^a)^r={\rm tr}(N_R^a)^r,
\end{equation}
where $M_L^a$ and $N_R^a$ are matrices that couple the gauge fields $A_m^a$
to left-handed and right-handed spin $1/2$ fermions, respectively.
Moreover, $r=n+1, n-1, n-3,n-5,...$ in $2n$ dimensions with $n+1$ external
gluons, $n-1$ external gluons and two gravitons, $n-3$ external gluons and
four gravitons, etc. which all correspond to anomalous graphs.
Witten\cite{Witten2} found solutions
to Eq.(\ref{anomaly}) in $2n$ dimensions, which we shall use in the
following. Let us consider in $2n$ dimensions, a theory with the orthogonal
gauge group $SO(2n+6)$. We assert that the positive chirality spinors of
the Lorentz group $SO(2n-1,1)$ transform as positive chirality spinors of
the gauge group $SO(2n+6)$, while the negative chirality $SO(2n-1,1)$
spinors transform as negative chirality spinors of $SO(2n+6)$. For any $n
> 0$ this leads to an anomaly-free theory, and for $n=2$, it is the
familiar $SO(10)$ grand unified model in four dimensions\cite{Fritzsch}.
We must guarantee that, if we begin with a non-vector-like theory in $2n$
dimensions, then we retain this property under dimensional reduction. The
condition
\[
\Gamma^{(4+n)}=\Gamma^{(4)}\cdot\Gamma^{(n)}
\]
where $\Gamma^{(4+n)}=\Gamma_1...\Gamma_{4+n},
\Gamma^{4}=\Gamma_1...\Gamma_4$, and
$\Gamma^{(n)}=\Gamma_5...\Gamma_{4+n}$, shows that the $4+n$-dimensional
chirality operator $\Gamma^{(4+n)}$ differs from $\Gamma^{(4)}$ by a factor
$\Gamma^{(n)}$ that can equal plus 1 or minus 1, so that we can lose the
non-vector-like property under compactification. As shown by
Randjibar-Daemi, Salam and Strathdee\cite{Salam}, this can be avoided by
attributing a non-trivial topological structure to the internal gauge
group ${\cal K}$, associated with the internal Kaluza-Klein space. In
particular, this can be achieved by inserting a generalized Dirac monopole
inside the Kaluza-Klein space.
Let us consider an $SO(18)$ theory in twelve dimensions. We identify the
internal space spinor connections $\omega^a_j$ with gauge fields $B^a_j$,
which have a non-vanishing vacuum expectation value, $\langle
B^a_j\rangle_0\not= 0$. The $B^a_j$ are $SO(8)$ gauge fields on the
eight-dimensional Riemannian manifold of the Kaluza-Klein sector. The
$SO(8)$ is embedded in $SO(18)$ , such that we obtain the symmetry breaking
$SO(18)\rightarrow SO(8)\times SO(10)$, which leads to the breaking in four
dimensions: $SO(8)\times SO(10)\rightarrow SO(10)$. Depending on the type
of manifold assumed for the compact dimensions, we get different
numbers of families which equal the Euler characteristic of the space $B$.
The number of families in ten dimensions is always even. For example, for
the manifolds $S^2\times S^4$ and $CP^3$ the number of families of zero
mass fermions is four, while for $S^2\times S^2\times S^2$ the number of
fermion families is eight. For models based on eight and twelve dimensions,
the number of fermion families is odd.
Since the observed number of chiral families in four dimensions is
three, we shall restrict our attention to the orthogonal group $SO(18)$ in
twelve dimensions with an odd number of families. We can decompose the
$\underline{256}$-dimensional representation of $SO(18)$ as
\[
(\underline{8}_{sp},\underline{\overline{16}})+(\underline{8}'_{sp},\underline{16})
\]
of $SO(8)\times SO(10)$, where $\underline{8}_{sp}$ and
$\underline{8}'_{sp}$ are the two real inequivalent spinors of $SO(8)$. Now
consider $SO(8)\supset Sp(4)\times SU(2)$, where the vectorial octet
$\underline{8}_V$ of $SO(8)$ yields $(\underline{4},\underline{2})$ of
$Sp(4)\times SU(2)$ and $\underline{8}'_{sp}$ of $SO(8)$ equivalently,
whereas $\underline{8}_{sp}$ of $SO(8)$ yields
\[
(\underline{1},\underline{3})+(\underline{5},\underline{1})
\]
of $Sp(4)\times SU(2)$. This then gives for the $\underline{256}$
representation of $SO(18)$:
\[
(\underline{1},\underline{3},\underline{16})+(\underline{5},\underline{1},\underline{16})
+(\underline{4},\underline{2},\underline{\overline{16}})
\]
of $Sp(4)\times SU(2)\times SO(10)$. By identifiying $Sp(4)$ as a
supplementary factor of the exactly conserved colour group $SU^c(3)\times
Sp^c(4)'$, and $SU(2)$ as a gauged family subgroup of $SO(18)$, then only
the fundamental left-handed fermions without primed colour are three
families of the sixteen-dimensional representation of $SO(10)$, which
agrees with observations\cite{Gell-Mann,Zee}.
\section{\bf Finite Quantum Field Theory Formalism in D Dimensions}
It is well-known that higher-dimensional field theories are
non-renormalizable, for their structure has enhanced divergences. The
non-renormalizability arises from the presence of infinite towers of
non-chiral Kaluza-Klein states which circulate in all Feynman graphs. Even
if we choose to describe the space as consisting of D-dimensional flat
spacetime, the field theory would still be non-renormalizable, because we
need to integrate over D dimensions of uncompactified loop momenta.
Therefore, the well-known non-renormalizability of quantum gravity is
further exacerbated in higher-dimensional theories. However, by applying
our finite quantum field theory (FQFT) formalism, based on a nonlocal interaction
Lagrangian which is perturbatively finite, unitary and gauge invariant
[10-18], we can obtain a finite quantum field theory in higher dimensions
and, in contrast to string theory, achieve a {\it genuine quantum field
theory}, which allows vertex operators to be taken off the mass shell.
The finiteness draws from the fact that factors of $\exp[{\cal
K}(p^2)/2\Lambda^2]$ are attached to propagators which suppress any
ultraviolet divergences in Euclidean momentum space, where $\Lambda$ is an
energy scale factor. An important feature of FQFT is {\it that only the
quantum loop graphs have non-local properties}; the classical tree graph
theory retains full causal and local behaviour.
An important development in FQFT was the discovery that gauge invariance and
unitarity can be restored by adding series of higher interactions. The
resulting theory possesses a nonlinear, field representation dependent
gauge invariance which agrees with the original local symmetry on shell but
is larger off shell. Quantization is performed in the functional
formalism using an analytic and convergent measure factor which retains
invariance under the new symmetry. An explicit calculation was made of the
measure factor in QED\protect\cite{Moffat2}, and it was obtained to lowest
order in Yang-Mills theory\protect\cite{Kleppe2}. Kleppe and
Woodard\protect\cite{Woodard2} obtained an ansatz based on the derived
dimensionally regulated result when $\Lambda\rightarrow\infty$, which was
conjectured to lead to a general functional measure factor in FQFT gauge
theories.
A convenient formalism which makes the FQFT construction transparent is
based on shadow fields\protect\cite{Kleppe2,Woodard2}. We shall consider
the D-dimensional spacetime to be approximately flat Minkowski spacetime,
which is a valid approximation for circles in the internal space $B$ having
large fixed radii $R$.
Let us denote by $f_i$ a generic local field and write the standard local
action as
\begin{equation}
W[f]=W_F[f]+W_I[f],
\end{equation}
where $W_F$ and $W_I$ denote the free part and the interaction part
of the action, respectively, and
\begin{equation}
W_F=\frac{1}{2}\int d^Dzf_i{\cal K}_{ij}f_j.
\end{equation}
In a gauge theory $W$ would be the Becchi, Rouet, Stora, Tyutin (BRST)
gauge-fixed action including ghost fields in the invariant action required
to fix the gauge\cite{Becchi}. The kinetic operator ${\cal K}$ is fixed by
defining a Lorentz-invariant distribution operator: \begin{equation} {\cal
E}\equiv \exp\biggl(\frac{{\cal K}}{2\Lambda^2}\biggr) \end{equation} and
the shadow operator: \begin{equation}
{\cal O}^{-1}=\frac{{\cal K}}{{\cal E}^2-1}.
\end{equation}
Every local field $f_i$ has an auxiliary counterpart field $h_i$, and they
are used to form a new action:
\begin{equation}
W[f,h]\equiv W_F[\hat f]-P[h]+W_I[f+h],
\end{equation}
where
\[
\hat f={\cal E}^{-1}f,\quad P[h]=\frac{1}{2}\int d^Dzh_i{\cal
O}^{-1}_{ij} h_j.
\]
By iterating the equation
\begin{equation}
h_i={\cal O}_{ij}\frac{\delta W_I[f+h]}{\delta h_j}
\end{equation}
the shadow fields can be determined as functions, and the regulated
action is derived from
\begin{equation}
\hat W[f]=W[f,h(f)].
\end{equation}
We recover the original local action when we take the limit
$\Lambda\rightarrow\infty$ and $\hat f\rightarrow f, h(f)\rightarrow 0$.
Quantization is performed using the definition
\begin{equation}
\langle 0\vert T^*(O[f])\vert 0\rangle_{\cal E}=\int[Df]\mu[f]({\rm gauge\,
fixing})O[\hat f]\exp(i\hat W[f]).
\end{equation}
On the left-hand side we have the regulated vacuum expectation value of the
$T^*$-ordered product of an arbitrary operator $O[f]$ formed from the local
fields $f_i$. The subscript ${\cal E}$ signifies that a regulating Lorentz
distribution has been used. Moreover, $\mu[f]$ is a measure factor and
there is a gauge fixing factor, both of which are needed to maintain
perturbative unitarity in gauge theories.
The new Feynman rules for FQFT are obtained as follows: The vertices remain
unchanged but every leg of a diagram is connected either to a regularized
propagator,
\begin{equation}
\label{regpropagator}
\frac{i{\cal E}^2}{{\cal K}+i\epsilon}
=-i\int^{\infty}_1\frac{d\tau}{\Lambda^2}\exp\biggl(\tau
\frac{{\cal K}}{\Lambda^2}\biggr),
\end{equation}
or to a shadow propagator,
\begin{equation}
-i{\cal O}=\frac{i(1-{\cal E}^2)}{{\cal K}}=-i\int^1_0\frac{d\tau}
{\Lambda^2}
\exp\biggl(\tau\frac{{\cal K}}{\Lambda^2}\biggr).
\end{equation}
The formalism is set up in Minkowski spacetime and loop integrals are
formally defined in Euclidean space by performing a Wick rotation. This
facilitates the analytic continuation; the whole formalism could from the
outset be developed in Euclidean space.
In FQFT renormalization is carried out as in any other field theory.
The bare parameters are calculated from the renormalized ones and
$\Lambda$, such that the limit $\Lambda\rightarrow\infty$ is finite for
all noncoincident Green's functions, and the bare parameters are those of
the local theory. The regularizing interactions {\it are determined by the
local operators.}
The regulating Lorentz distribution function ${\cal E}$ must be chosen to
perform an explicit calculation in perturbation theory. We do not know the
unique choice of ${\cal E}$\cite{Feynman}. It maybe that there exists an
equivalence mapping between all the possible distribution functions ${\cal
E}$. However, once a choice for the function is made, then the
theory and the perturbative calculations are uniquely fixed. A standard
choice in early FQFT papers is\protect\cite{Moffat,Moffat2}:
\begin{equation}
\label{reg}
{\cal E}_m=\exp\biggl(\frac{\partial^2-m^2}{2\Lambda^2}\biggr).
\end{equation}
An explicit construction for QED was given using the Cutkosky rules as
applied to FQFT whose propagators have poles only where ${\cal K}=0$ and
whose vertices are entire functions of ${\cal K}$. The regulated action
$\hat W[f]$ satisfies these requirements which guarantees unitarity on the
physical space of states. The local action is gauge fixed and then a
regularization is performed on the BRST theory.
The infinitesimal transformation
\begin{equation}
\delta f_i=T_i(f)
\end{equation}
generates a symmetry of $W$, and the infinitesimal transformation
\begin{equation}
\hat\delta f_i={\cal E}^2_{ij}T_j(f+h[f])
\end{equation}
generates a symmetry of the regulated action ${\hat W}$. It follows
that FQFT regularization preserves all continuous symmetries including
supersymmetry. The quantum theory will preserve symmetries
provided a suitable measure factor can be found such that
\begin{equation}
\hat\delta([Df]\mu[f])=0.
\end{equation}
Moreover, the interaction vertices of the measure factor must be
entire functions of the operator ${\cal K}$ and they must not destroy the
FQFT finiteness.
In FQFT tree order, Green's functions remain local except for
external lines which are unity on shell. It follows immediately that since
on-shell tree amplitudes are unchanged by the regularization, $\hat W$
preserves all symmetries of $W$ on shell. Also all loops contain at least
one regularizing propagator and therefore are ultraviolet finite. Shadow
fields are eliminated at the classical level, for functionally integrating
over them would produce divergences from shadow loops. Since shadow field
propagators do not contain any poles there is no need to quantize the
shadow fields. Feynman rules for ${\hat W}[f,h]$ are as simple as those for
local field theory.
\section{\bf Finite Quantum Yang-Mills and Kaluza-Klein Gravity Theory}
Let us now consider the finite quantization of the D-dimensional Yang-Mills
sector in D-dimensional Minkowski flat space. The gauge field strength
$F_{aMN}$ is invariant under the familiar transformations:
\[
\delta A_{aM}=-\partial_M\theta_a+ ef_{abc}A_{bM}\theta_c.
\]
To regularize the Yang-Mills sector, we identify the kinetic operator
\[
{\cal K}_{ab}^{MN}=\delta_{ab}(\partial^2\eta^{MN}-\partial^M\partial^N).
\]
The regularized action is given by\cite{Kleppe2}
\begin{equation}
\hat{W}_{YM}[A]=\frac{1}{2}\int d^Dx\biggl\{\hat{A}_{aM}
{\cal K}^{MN}_{ab}\hat{A}_{bN}-B_{aM}[A]({\cal O}^{MN}_{ab})^{-1}
B_{bN}[A]\biggr\}
$$
$$
+W^I_{YM}[A+B[A]],
\end{equation}
where $B_{aM}$ is the Yang-Mills shadow field, which satisfies the
expansion:
\[
B^M_a[A]={\cal O}^{MN}_{ab}\frac{\delta W^I_{YM}[A+B]}{\delta B^N_b}
\]
\[
={\cal O}^{MN}_{ab}ef_{bcd}[A_{Nc}\partial_KA^K_d+A_{cK}\partial_NA^K_d
-2A_{cK}\partial^KA^N_d]+O(e^2A^3).
\]
The regularized gauge symmetry transformation is
\[
\hat{\delta}_{\theta}A^M_a=({\cal E}^{MN}_{ab})^2
\biggl\{-\partial^M\theta_a+ef_{bcd}(A_{cN}+B_{cN}[A])\theta_d\biggr\}
\]
\[
-\partial^M\theta_a+({\cal E}^{MN}_{ab})^2ef_{bcd}(A_{cN}
+B_{cN}[A])\theta_d.
\]
The extended gauge transformation is neither linear nor local.
We functionally quantize the Yang-Mills sector using
\begin{equation}
\langle 0\vert T^*(O[A])\vert 0\rangle_{\cal E}=\int[DA]
\mu[A]({\rm gauge\, fixing})
O[{\hat A}]\exp(i\hat W_{\rm YM}[A]).
\end{equation}
To fix the gauge we use Becchi-Rouet-Stora-Tyutin (BRST)\cite{Becchi}
invariance. The ghost structure of the BRST action comes from
exponentiating the Faddeev-Popov determinant. Since the FQFT algebra fails
to close off-shell, we need to introduce higher ghost terms into both the
action and the BRST transformation. In Feynman gauge, the local BRST
Lagrangian is
\[
{\cal L}_{YM\,BRST}=-\frac{1}{2}\partial_MA_{aN}\partial^MA^N_a
-\partial^M\bar{\eta_a}\partial_M\eta_a+ef_{abc}\partial^M\bar{\eta}_a
A_{bM}\eta_c
\]
\[
+ef_{abc}\partial_MA_{aN}A^M_bA^N_c-\frac{1}{4}e^2f_{abc}f_{cde}
A_{aM}A_{bN}A^M_dA^N_c.
\]
It is invariant under the global symmetry transformation:
\[
\delta A_{aM}=(\partial_M\eta_a-ef_{abc}A_{bM}\eta_c)\delta\zeta,
\]
\[
\delta\eta_a=-\frac{1}{2}ef_{abc}\eta_b\eta_c\delta\zeta,
\]
\[
\delta\bar{\eta}_a=-\partial_MA^M_a\delta\zeta,
\]
where $\zeta$ is a constant anticommuting c-number.
The gluon and ghost kinetic operators are
\[
{\cal K}^{MN}_{ab}=\delta_{ab}\eta^{MN}\partial^2,
\]
\[
{\cal K}_{ab}=\delta_{ab}\partial^2.
\]
The regularizing operators associated with the ghosts are
\[
\bar{\cal E}=\exp\biggl(\frac{\partial^2}{2\Lambda^2}\biggr),
\]
\[
\bar{\cal O}=\frac{\bar{\cal E}^2-1}{\partial^2}.
\]
The regularized BRST action is
\[
\hat{W}_{YM}[A,\bar{\eta},\eta]=\int d^Dz\biggl\{-\frac{1}{2}
\partial_N\hat{A}_{aM}\partial^N\hat{A}_a^M-\frac{1}{2}B_{aM}
\bar{\cal O}^{-1}B^M_a
\]
\[
-\partial^M\hat{\bar{\eta}}_a\partial_M\hat{\eta}_a
-\bar{\chi}_a\bar{\cal O}^{-1}\chi_a\biggr\}+W^I_{\rm YM}[A+B,\bar{\eta}
+\bar{\chi},\eta+\chi],
\]
where $\chi$ is the ghost shadow field.
The regularizing, nonlocal BRST symmetry transformation is
\[
\hat\delta A_{aM}=\bar{\cal E}^2\biggl\{(\partial_M\eta_a
+\partial_M\chi_a)-ef_{abc}(A_{bM}+B_{bM})(\eta_c+\chi_c)\biggr\}\delta\zeta,
\]
\[
\hat\delta\eta_a=-\frac{1}{2}ef_{abc}\bar{\cal E}^2(\eta_b+\chi_b)
(\eta_c+\chi_c)\delta\zeta,
\]
\[
\hat{\delta}\bar{\eta}_a=-\bar{\cal E}^2(\partial_MA^M_a
+\partial_MB^M_a)\delta\zeta.
\]
The full functional, gauge fixed quantization is now given by
\begin{equation}
\langle 0\vert T^*(O[A,\bar{\eta},\eta])\vert 0\rangle_{\cal E}=\int[DA]
[D\bar{\eta}][D\eta]
\mu[A,\bar{\eta},\eta]O[\hat{A},\hat{\bar{\eta}},\hat{\eta}]
$$
$$
\exp(i\hat{W}_{\rm YM}[A,\bar{\eta},\eta]).
\end{equation}
Kleppe and Woodard\cite{Kleppe2} have obtained the invariant measure
factor for the regularized Yang-Mills sector to first order in the
coupling constant $e$:
\begin{equation}
\ln(\mu[A,\bar{\eta},\eta])=-\frac{1}{2}e^2f_{acd}f_{bcd}\int d^Dz
A_{aM}{\cal M}A^M_b+O(e^3),
\end{equation}
where
\[
{\cal M}=\frac{1}{2^D\pi^{D/2}}\int^1_0d\tau
\frac{\Lambda^{D-2}}{(\tau+1)^{D/2}}\exp\biggl(\frac{\tau}{\tau+1}
\frac{\partial^2}{\Lambda^2}\biggr)
\]
\[
\biggl\{\frac{2}{\tau+1}+1-D+2(D-1)\frac{\tau}{\tau+1}\biggr\}.
\]
The existence of a suitable invariant measure factor implies that the
necessary Slavnov-Taylor identities also exist.
We shall now formulate in more detail the Kaluza-Klein gravitational sector
as a FQFT. This problem has been considered previously in the context of
four-dimensional GR\protect\cite{Moffat,Moffat2,Moffat4}. We shall treat
the theory as effectively being in D flat Minkowski dimensions, $D=4+d$.
A spacetime consisting of four flat Minkowski dimensions and $d$
circles of fixed radii $R=1/\mu_0$ cannot, in general, be equivalent to a
flat D-dimensional spacetime. However, as the energy scale $\mu$ increases,
the effective length scale decreases, so that the fixed radius $R$
appears to become large, and the D-dimensional flat spacetime becomes a
good approximation. In fact, FQFT can be formulated as a perturbative
theory by expanding around any fixed, classical metric background, but
for the sake of simplicity, we shall only consider in the
following expansions about flat spacetime. As in ref.(\cite{Moffat4}), we
will regularize the GR equations using the shadow field formalism.
We expand the local interpolating field ${\bf g}^{MN}=\sqrt{-g}
g^{MN}\quad (g={\rm Det}(g_{MN}))$ about Minkowski spacetime
\begin{equation}
{\bf g}^{MN}=\eta^{MN}+\kappa\gamma^{MN}+O(\kappa^2).
\end{equation}
We separate the free and interacting
parts of the action
\begin{equation}
W_{\rm grav}(g)=W^F_{\rm grav}(g)+W^I_{\rm grav}(g).
\end{equation}
The finite regularized gravitational action in FQFT is given by
\begin{equation}
\hat{W}_{\rm grav}(g,s)=W^F_{\rm grav}({\hat g})
-P_{\rm grav}(s)+W^I_{\rm grav}(g+s),
\end{equation}
where
\begin{equation}
{\hat g}={\cal E}^{-1}g,\quad P_{\rm grav}(s)
=\frac{1}{2}\int d^Dz{\cal F}(\sqrt{s},s_i{\cal O}_{ij}^{-1}s_j),
\end{equation}
$s$ denotes the graviton shadow field, and ${\cal F}$ denotes the detailed
expansion of the metric tensor formed from the shadow field.
The quantum gravity perturbation theory is locally $SO(D-1,1)$ invariant
(generalized, nonlinear field representation dependent transformations),
unitary and finite to all orders in a way similar to the non-Abelian gauge
theories formulated using FQFT. At the tree graph level all unphysical
polarization states are decoupled and nonlocal effects will only occur in
graviton and graviton-matter loop graphs. Because the gravitational tree
graphs are purely local there is a well-defined classical GR limit. The
finite quantum gravity theory is well-defined in D real spacetime
dimensions.
The graviton regularized propagator in a fixed de Donder gauge
is given in D-dimensional Minkowski space by\cite{Donder,Moff}
\[
D^{\rm grav}_{MNKL}
=(\eta_{MK}\eta_{NL}+\eta_{ML}\eta_{NK}
-\eta_{MN}\eta_{KL})
\]
\[
\biggl(\frac{-i}{(2\pi)^D}\biggr)\int d^Dk\frac{{\cal
E}^2(k^2)}{k^2-i\epsilon} \exp[ik\cdot(z-z')],
\]
while the shadow propagator is
\[
D^{\rm shad}_{MNKL}
=(\eta_{MK}\eta_{NL}+\eta_{ML}\eta_{NK}
-\eta_{MN}\eta_{KL})
\]
\[
\biggl(\frac{-i}{(2\pi)^D}\biggr)\int d^Dk
\frac{[1-{\cal E}^2(k^2)]}{k^2-i\epsilon} \exp[ik\cdot(z-z')].
\]
In momentum space we have
\[
\frac{-i{\cal E}^2(k^2)}{k^2-i\epsilon}=-i\int^{\infty}_1\frac{d\tau}
{\Lambda^2_G}
\exp\biggl(-\tau\frac{k^2}{\Lambda^2_G}\biggr),
\]
and
\[
\frac{i({\cal E}^2(k^2)-1)}{k^2-i\epsilon}=-i\int_0^1\frac{d\tau}
{\Lambda^2_G}
\exp\biggl(-\tau\frac{k^2}{\Lambda^2_G}\biggr),
\]
where $\Lambda_G$ is the gravitational scale parameter.
We quantize by means of the path integral operation
\begin{equation}
\langle 0\vert T^*(O[g])\vert 0\rangle_{\cal E}=\int[Dg]
\mu[g]({\rm gauge\, fixing})
O[\hat g]\exp(i\hat W_{\rm grav}[g]).
\end{equation}
The quantization is carried out
in the functional formalism by finding a measure factor
$\mu[g]$ to make $[Dg]$ invariant under the
classical symmetry.
To ensure a correct gauge fixing scheme, we write $W_{\rm grav}[g]$
in the BRST invariant form with ghost fields; the ghost structure
arises from exponentiating the Faddeev-Popov determinant. The algebra of
gauge symmetries is not expected to close off-shell, so one needs to
introduce higher ghost terms (beyond the normal ones) into both the
action and the BRST transformation. The BRST action will be regularized
directly to ensure that all the corrections to the measure factor are
included.
\section{\bf Quantum Nonlocal Behavior in FQFT}
In FQFT, it can be
argued that the extended objects that replace point particles (the latter
are obtained in the limit $\Lambda\rightarrow\infty$) cannot be probed
because of a Heisenberg uncertainty type of argument. The FQFT nonlocality
{\it only occurs at the quantum loop level}, so there is no noncausal
classical behavior. In FQFT the strength of a signal propagated over an
invariant interval $l^2$ outside the light cone would be suppressed by a
factor $\exp(-l^2\Lambda^2)$.
Nonlocal field theories can possess non-perturbative instabilities. These
instabilities arise because of extra canonical degrees of freedom associated
with higher time derivatives. If a Lagrangian contains up to $N$ time
derivatives, then the associated Hamiltonian is linear in $N-1$ of the
corresponding canonical variables and extra canonical degrees of freedom
will be generated by the higher time derivatives. The nonlocal theory can
be viewed as the limit $N\rightarrow\infty$ of an Nth derivative
Lagrangian. Unless the dependence on the extra solutions is arbitrarily
choppy in the limit, then the higher derivative limit will produce
instabilities\protect\cite{Eliezer}. The condition for the smoothness of
the extra solutions is that no invertible field redefinition exists which
maps the nonlocal field equations into the local ones. String theory does
satisfy this smoothness condition as can be seen by inspection of the
S-matrix tree graphs. In FQFT the tree amplitudes agree with those of the
local theory, so the smoothness condition is not obeyed.
It was proved by Kleppe and Woodard\protect\cite{Kleppe2} that the solutions
of the nonlocal field equations in FQFT are in one-to-one correspondence
with those of the original local theory. The relation for a generic field
$v_i$ is \begin{equation}
v_i^{\rm nonlocal}={\cal E}^2_{ij}v^{\rm local}_j.
\end{equation}
Also the actions satisfy
\begin{equation}
W[v]={\hat W}[{\cal E}^2v].
\end{equation}
Thus, there are no extra classical solutions.
The solutions of the regularized nonlocal Euler-Lagrange equations are
in one-to-one correspondence with those of the local action. It follows
{\it that the regularized nonlocal FQFT is free of higher derivative
solutions, so FQFT can be a stable theory.}
Since only the quantum loop graphs in the nonlocal FQFT differ from
the local field theory, then FQFT can be viewed as a non-canonical
quantization of fields which obey the local equations of motion. Provided
the functional quantization in FQFT is successful, then the theory does
maintain perturbative unitarity.
\section{\bf A Resolution of The Higgs Hierarchy Problem and Quantum
Gravity}
It is time to discuss the Higgs sector hierarchy problem\cite{Susskind}.
The gauge hierarchy problem is related to the spin $0^+$ scalar field
nature of the Higgs particle in the standard model with quadratic mass
divergence and no protective extra symmetry at $m=0$. In standard point
particle, local field theory the fermion masses are logarithmically
divergent and there exists a chiral symmetry restoration at $m=0$.
Writing $m_H^2=m_{0H}^2+\delta m_H^2$, where
$m_{0H}$ is the bare Higgs mass and $\delta m_H$ is the Higgs
self-energy renormalization constant, we get for the one loop Feynman
graph in $D=4$ spacetime:
\begin{equation}
\delta m_H^2\sim \frac{g}{32\pi^2}\Lambda_C^2,
\end{equation}
where $\Lambda_C$ is a cutoff parameter. If we want to understand
the nature of the Higgs mass we must require that
\begin{equation}
\delta m_H^2 \leq O(m_H^2),
\end{equation}
i.e. the quadratic divergence should be cut off at the mass scale of
the order of the physical Higgs mass. Since $m_H\simeq \sqrt{2g}v$,
where $v=<\phi>_0$ is the vacuum expectation value of the scalar field
$\phi$ and $v=246$ GeV from the electroweak theory, then in order to keep
perturbation theory valid, we must demand that $10\,{\rm GeV}
\leq m_H \leq 350\,{\rm GeV}$ and we need \begin{equation}
\label{Higgscut}
\Lambda_C =\Lambda_{\rm Higgs}\leq 1\, {\rm TeV},
\end{equation}
where the lower bound on $m_H$ comes from the avoidance of washing out
the spontaneous symmetry breaking of the vacuum.
Nothing in the standard model can tell us why (\ref{Higgscut}) should be true,
so we must go beyond the standard model to solve the problem. $\Lambda_C$
is an arbitrary parameter in point particle field theory with no physical
interpretation. Since all particles interact through gravity, then
ultimately we should expect to include gravity in the standard model, so we
expect that $\Lambda_{\rm Planck}\sim 10^{19}$ GeV should be the natural
cutoff. Then we have using (\ref{Higgscut}) and $g\sim 1$:
\[
\frac{\delta m_H^2(\Lambda_{\rm Higgs})}{\delta m_H^2
(\Lambda_{\rm Planck})} \approx \frac{\Lambda^2_{\rm Higgs}}
{\Lambda^2_{\rm Planck}}\approx 10^{-34}, \]
which represents an intolerable fine-tuning of parameters. This `naturalness' or
hierarchy problem is one of the most serious defects of the standard model.
There have been two strategies proposed as ways out of the hierarchy
problem. The Higgs is taken to be composite at a scale $\Lambda_C\simeq 1$ TeV,
thereby providing a natural cutoff in the quadratically divergent Higgs
loops. One such scenario is the `technicolor' model, but it cannot be
reconciled with the accurate standard model data, nor with the smallness of
fermion masses and the flavor-changing neutral current interactions. The other
strategy is to postulate supersymmetry, so that the opposite signs of the
boson and fermion lines cancel by means of the non-renormalization theorem.
However, supersymmetry is badly broken at lower energies, so we require
that
\[
\delta m_H^2\sim \frac{g}{32\pi^2}\vert\Lambda^2_{C\,{\rm bosons}}
-\Lambda^2_{C\,{\rm fermions}}\vert\leq 1\,{\rm TeV}^2,
\]
or, in effect
\[
\vert m_B-m_F\vert \leq 1\, {\rm TeV}.
\]
This physical requirement leads to the prediction that the supersymmetric
partners of known particles should have a threshold $\leq1$ TeV.
A third possible strategy is to introduce a FQFT formalism, and realize a field
theory mechanism which will introduce a natural physical scale in the
theory $\leq 1$ TeV, which will protect the Higgs mass from becoming large
and unstable.
Let us consider the regularized scalar field FQFT Lagrangian in
D-dimensional Minkowski space
\begin{equation}
{\hat{\cal L}}_S=\frac{1}{2}\hat\phi(\partial^2-m^2)\hat\phi
-\frac{1}{2}\rho{\cal O}^{-1}\rho+\frac{1}{2}Z^{-1}\delta m^2(\phi+\rho)^2
-\frac{1}{24}g_0(\phi+\rho)^4,
\end{equation}
where $\phi=Z^{1/2}\phi_R$ is the bare field, $\phi_R$ is the renormalized
field, $\hat\phi={\cal E}^{-1}\phi$, $\rho$ is
the shadow field, $m_0$ is the bare mass, $Z$ is the field strength
renormalization constant, $\delta m^2$ is the mass renormalization constant
and $m$ is the physical mass. The regularizing operator is given by
\begin{equation}
{\cal E}_m=\exp\biggl(\frac{\partial^2-m^2}{2\Lambda_H^2}\biggr),
\end{equation}
while the shadow kinetic operator is
\begin{equation}
{\cal O}^{-1}=\frac{\partial^2-m^2}{{\cal E}_m^2-1}.
\end{equation}
Here, $\Lambda_H$ is the Higgs scalar field energy scale in FQFT.
The full propagator is
\begin{equation}
-i\Delta_R(p^2)=\frac{-i{\cal E}_m^2}{p^2+m^2-i\epsilon}
=-i\int_1^{\infty}\frac{d\tau}{\Lambda_H^2}\exp\biggl[-\tau
\biggl(\frac{p^2+m^2}{\Lambda_H^2}\biggr)\biggr],
\end{equation}
whereas the shadow propagator is
\begin{equation}
i\Delta_{\rm shadow}
=i\frac{{\cal E}_m^2-1}{p^2+m^2}=-i\int^1_0\frac{d\tau}{\Lambda_H^2}
\exp\biggl[-\tau\biggl(\frac{p^2+m^2}{\Lambda_H^2}\biggr)\biggr].
\end{equation}
Let us define the self-energy $\Sigma(p)$ as a Taylor series expansion
around the mass shell
$p^2=-m^2$:
\begin{equation}
\Sigma(p^2)=\Sigma(-m^2)+(p^2+m^2)\frac{\partial\Sigma}{\partial p^2}(-m^2)
+{\tilde \Sigma}(p^2),
\end{equation}
where ${\tilde\Sigma}(p^2)$ is the usual finite part in the point
particle limit $\Lambda_H\rightarrow\infty$.
We have
\begin{equation}
{\tilde\Sigma}(-m^2)=0,
\end{equation}
and
\begin{equation}
\frac{\partial{\tilde\Sigma}(p^2)}{\partial p^2}(p^2=-m^2)=0.
\end{equation}
The full propagator is related to the self-energy $\Sigma(p^2)$ by
\begin{equation}
-i\Delta_R(p^2)=\frac{-i{\cal E}_m^2[1+{\cal O}\Sigma(p^2)]}{p^2+m^2
+\Sigma(p^2)}
=\frac{-iZ}{p^2+m^2+\Sigma_R(p^2)}.
\end{equation}
Here $\Sigma_R(p^2)$ is the renormalized self-energy which can be written
as
\begin{equation}
\Sigma_R(p^2)=(p^2+m^2)\biggl[\frac{Z}{{\cal E}_m^2(1+{\cal O}\Sigma)}
-1\biggr]+\frac{Z\Sigma}{{\cal E}_m^2(1+{\cal O}\Sigma)}.
\end{equation}
The 1PI two-point function is given by
\begin{equation}
-i\Gamma_R^{(2)}(p^2)=i[\Delta_R(p^2)]^{-1}=\frac{i[p^2+m^2+\Sigma(p^2)]}
{{\cal E}_m^2[1+{\cal O}\Sigma(p^2)]}.
\end{equation}
Since ${\cal E}_m\rightarrow 1$ and ${\cal O}\rightarrow 0$ as $\Lambda_H
\rightarrow\infty$,
then in this limit
\begin{equation}
-i\Gamma_R^{(2)}(p^2)=i[p^2+m^2+\Sigma(p^2)],
\end{equation}
which is the standard point particle result.
The mass renormalization is determined by the propagator pole at $p^2=-m^2$
and we have
\begin{equation}
\Sigma_R(-m^2)=0.
\end{equation}
Also, we have the condition
\begin{equation}
\frac{\partial\Sigma_R(p^2)}{\partial p^2}(p^2=-m^2)=0.
\end{equation}
The renormalized coupling constant is defined by the four-point function
$\Gamma^{(4)}_R(p_1,p_2,p_3,p_4)$ at the point $p_i=0$:
\begin{equation}
\Gamma_R^{(4)}(0,0,0,0)=g.
\end{equation}
The bare coupling constant $g_0$ is determined by
\begin{equation}
Z^2g_0=g+\delta g(g,m^2,\Lambda_H^2).
\end{equation}
Moreover,
\[
Z=1+\delta Z(g,m^2,\Lambda_H^2),
\]
\[
Zm_0^2=Zm^2-\delta m^2(g,m^2,\Lambda^2_H).
\]
A calculation of the scalar field mass renormalization gives\cite{Woodard2}:
\begin{equation}
\delta m^2=\frac{g}{2^{D+1}\pi^{D/2}}m^{D-2}
\Gamma\biggl(1-\frac{D}{2},\frac{m^2}{\Lambda_H^2}\biggr)+O(g^2),
\end{equation}
where $\Gamma(n,z)$ is the incomplete gamma function:
\begin{equation}
\Gamma(n,z)=\int_z^{\infty}\frac{dt}{t}t^n\exp(-t)=(n-1)\Gamma(n-1,z)
+z^{n-1}\exp(-z).
\end{equation}
We have
\begin{equation}
\Gamma(-1,z)=-E_i(z)+\frac{1}{z}\exp(-z),
\end{equation}
where $E_i(z)$ is the exponential integral
\[
E_i(z)\equiv \int^\infty_zdt\frac{\exp(-t)}{t}.
\]
For small $z$ we obtain the expansion
\begin{equation}
E_i(z)=-\ln(z)-\gamma+z-\frac{z^2}{2\cdot 2!}+\frac{z^3}{3\cdot 3!}-...,
\end{equation}
where $\gamma$ is Euler's constant. For large positive values of $z$, we
have the asymptotic expansion
\begin{equation}
E_i(z)\sim\exp(-z)\biggl[\frac{1}{z}-\frac{1}{z^2}+\frac{2!}{z^3}-...\biggr].
\end{equation}
Thus, for small $m/\Lambda_H$ we obtain in $D=4$ spacetime:
\begin{equation}
\label{lambdast}
\delta m^2=\frac{g}{32\pi^2}
\biggl[\Lambda_H^2-m^2\ln\biggl(\frac{\Lambda_H^2}{m^2}\biggr)
-m^2(1-\gamma)+O\biggl(\frac{m^2}{\Lambda_H^2}\biggr)\biggr]+O(g^2),
\end{equation}
which is the standard quadratically divergent self-energy, obtained from
a cutoff procedure or a dimensional regularization scheme.
We have for $z\rightarrow\infty$:
\begin{equation}
\label{Gamma}
\Gamma(a,z)\sim
z^{a-1}\exp(-z)\biggl[1+\frac{a-1}{z}
+O\biggl(\frac{1}{z^2}\biggr)\biggr]
\end{equation}
so that for $m\gg\Lambda_H$, we get in D-dimensional space
\begin{equation}
\delta
m^2\sim\frac{g}{2^{D+1}\pi^{D/2}}\biggl(\frac{\Lambda^D_H}{m^2}\biggr)
\exp\biggl(-\frac{m^2}{\Lambda_H^2}\biggr). \end{equation}
Thus, the Higgs self-energy one loop graph falls off exponentially fast for
$m\gg \Lambda_H$.
The lowest order contributions to the graviton self-energy in FQFT will
include the standard graviton loops, the shadow field graviton loops, the
ghost field loop contributions with their shadow field counterparts, and
the measure loop contributions. The calculated measures for regularized
QED, first order Yang-Mills theory and all orders in $\phi^4$ and $\phi^6$
theories lead to self-energy contributions that are controlled by an
incomplete $\Gamma$ function. For the regularized perturbative gravity
theory the first order loop amplitude is
\begin{equation}
\label{Aequation}
A_i=\Gamma\biggl(2-D/2,\frac{p^2}{\Lambda_G^2}\biggr)(p^2)^{D-2}
F_i(D/2).\quad (i=1,...,5)
\end{equation}
The dimensional regularization result is obtained by the replacement
\[
\Gamma\biggl(2-D/2,\frac{p^2}{\Lambda_G^2}\biggr)\rightarrow
\Gamma(2-D/2),
\]
yielding the result\protect\cite{Capper,Veltman}
\begin{equation}
A_i\sim \Gamma(2-D/2)(p^2)^{D-2}F_i(D/2)\sim \frac{1}{\epsilon}
(p^2)^{D-2}F_i(D/2),
\end{equation}
where $\epsilon=2-D/2$ and $\Gamma(n)$ is the gamma function.
Whereas the dimensional
regularization result is singular in the limit $\epsilon\rightarrow
0$, the FQFT result is {\it finite in this limit} for a
fixed value of the parameter $\Lambda_G$, resulting in a finite
graviton self-energy amplitude $A_i$.
For D-dimensional spacetime, using (\ref{Gamma}) and (\ref{Aequation}),
we obtain in the Euclidean momentum limit
\begin{equation}
A_i\sim \biggl(\frac{p^2}{\Lambda_G^2}\biggr)^{1-D/2}
\exp\biggl(-\frac{p^2}{\Lambda_G^2}\biggr)(p^2)^{D-2}F_i(D/2)\quad
(i=1,...,5).
\end{equation}
Thus, in the infinite Euclidean momentum limit the quantum graviton
self-energy contribution damps out and quantum graviton corrections
become negligible. It is often argued in the literature on quantum
gravity that the gravitational quantum corrections scale as
$\alpha_G =GE^2$, so that for sufficiently large values of the energy $E$,
namely, of order the Planck energy, the gravitational quantum
fluctuations become large. We see that in FQFT quantum gravity this may not
be the case, because for $\Lambda_G << M_{\rm Planck}\sim 10^{19}$ GeV, the
finite quantum loop corrections become negligible in the high energy limit.
Of course, the contributions of the tree graph exchanges of virtual
gravitons can be large in the high energy limit, corresponding to strong
classical gravitational fields. It follows that for high enough energies,
a classical curved spacetime would be a good approximation.
In contrast to recent models of branes and strings in which the
compactification scale is lowered to the TeV range\cite{Witten,Arkani}, we
retain the classical GR gravitation picture and its Newtonian limit. It is
perhaps a radical notion to entertain that quantum gravity becomes weaker
as the energy scale increases towards the Planck scale $\sim 10^{19}$ Gev,
but there is, of course, no known experimental reason why this should not
be the case in nature. This would have important implications about the
nature of singularities in gravitational collapse and the big bang
scenario, for we could not appeal to quantum gravity to alleviate the
singularity problem, but hope instead that a classical modification of GR
occurs for very small distances $\leq 10^{-33}$ cm.
If we choose $\Lambda_H=\Lambda_G\geq$ 1 -5 TeV, then due to the damping of
the gravitational loop graphs and the scalar loop graphs in the Euclidean
limit $p^2\gg\Lambda^2$, the Higgs sector is protected from large
unstable radiative corrections and FQFT provides a solution to the Higgs
hierarchy problem, without invoking low-energy supersymmetry or
technicolor. The universal fixed FQFT scale $\Lambda_H$ corresponds to the
fundamental length $\ell_H\leq 10^{-17}$ cm. For a Higgs mass much larger
than 1 TeV, the Higgs sector becomes non-perturbative and we must be
concerned about violations of unitarity\cite{Veltman2}.
\section{Kaluza-Klein Excitations Associated with Higher Dimensions}
In our D-dimensional gauge theory, we must concern ourselves with the
restrictions imposed on the scale size, $\Lambda$, and the compactification
size, $R$, imposed by Kaluza-Klein excitations associated with the extra
dimensions. Recently, Nath and Yamaguchi\cite{Nath}, and
others\cite{Graesser}, have studied the constraints on the compactification
scale imposed by the Fermi constant, the fine structure constant, and the
$W$ and $Z$ masses. Because of the importance of this issue for our theory,
we shall discuss these results here in some detail. Dienes et al. have
caluculated the effects of infinite Kaluza-Klein towers on the "running" of
the gauge coupling constants\cite{Dienes2}. Let us now adapt their results
to our FQFT formalism.
As before, we shall treat our four-dimensional space as an approximately
flat Minkowski spacetime, and evaluate the vacuum polarization diagram,
including the effects of the Kaluza-Klein excitations on the loop graphs.
Consider first the case of a single Dirac fermion. We can generalize this
result to the case of a realistic chiral fermion model as a final step. In
FQFT, the vacuum polarization tensor is the sum of three parts,
$\Pi_1^{\mu\nu}(p), \Pi_2^{\mu\nu}(p)$, and $\Pi_3^{\mu\nu}(p)$
corresponding to the standard loop graph, a tadpole graph and a
contribution from the invariant measure factor. We have
\begin{equation}
\Pi_1^{\mu\nu}=\Pi_1^T(p^2)\biggl(\eta^{\mu\nu}-\frac{p^\mu
p^\nu}{p^2}\biggr)+\Pi^L_1(p^2)\frac{p^\mu p^\nu}{p^2},
\end{equation}
where $\Pi^T$ and $\Pi^L$ denote the transverse and longitudinal parts,
respectively. The total transverse part is given by\cite{Moffat2}:
\begin{equation}
\label{vacuumstate}
\frac{\Pi^T(p^2)}{p^2}=-\frac{e^{(4)2}}{\pi^2}\sum^\infty_{n_i=-\infty}
\exp\biggl(-\frac{p^2}{\Lambda^2}\biggr)\int^{1/2}_0dx
x(1-x)E_i\biggl[x(1-x)\frac{p^2}{\Lambda^2}+\frac{m_n^2}{\Lambda^2}\biggr],
\end{equation}
where $e^{(4)}$ is the four-dimensional coupling constant and
\[
\sum^\infty_{n_i=-\infty}\equiv
\sum^\infty_{n_1=-\infty}\sum^\infty_{n_2=-\infty}...\sum^\infty_{n_d=-\infty},
\]
describes a summation over all Kaluza-Klein excitations with masses $m_n$,
and $m_0$ is the energy of the ground state. Moreover, we see that
$\Pi(p^2)$ vanishes exponentially fast in the Euclidean momentum region as
$p^2\rightarrow\infty$.
The sum over the Kaluza-Klein states can be performed by using the Jacobi
$\theta_3$ function\cite{Dienes2}:
\[
\theta_3(\tau)\equiv \sum^\infty_{n_i=-\infty}\exp(\pi i\tau n^2),
\]
where $\tau$ is a complex number. This function obeys the property
\[
\theta_3(-1/\tau)=\sqrt{-i\tau}\theta_3(\tau),
\]
where the branch of the square root with non-negative real part is chosen.
We obtain
\[
\frac{\Pi(p^2)}{p^2}=-\frac{e^{(4)2}}{\pi^2}\exp\biggl(-\frac{p^2}{\Lambda^2}\biggr)
\int^{1/2}_0dxx(1-x)E_i\biggl[x(1-x)\frac{p^2}{\Lambda^2}\biggr]
\biggl[\theta_3\biggl(\frac{it}{\pi R^2}\biggr)\biggr]^d.
\]
We then get
\begin{equation}
\biggl[\frac{\Pi(p^2)}{p^2}\biggr]_{p^2=0}=-\frac{e^{(4)2}}{12\pi^2}
\int^{r^2/\mu_0^2}_{r^2/\Lambda^2}\frac{dt}{t}\biggl[\theta_3\biggl(\frac{it}{\pi
R^2} \biggr)\biggr]^d,
\end{equation}
where
\[
r^2=\frac{\pi}{(X_d)^{2/d}}
\]
and
$$
X_d=\frac{\pi^{d/2}}{\Gamma(1+d/2)}
$$
relates our FQFT scale $\Lambda$ to the underlying physical mass scales. It
is to be noted that in contrast to the standard cut-off technique, the FQFT
calculation of the vacuum polarization is fully gauge invariant and unitary
and $\Lambda$ represents a {\it physical} scale in the theory.
For our full chiral gauge theory assuming that the Kaluza-Klein excitations
arise from the gauge bosons and Higgs fields only, we obtain
\begin{equation}
\biggl[\frac{\Pi(p^2)}{p^2}\biggr]_{p^2=0}
=\frac{e_i^{(4)2}(b_i-\tilde{b}_i)}{8\pi^2}\ln\biggl(\frac{\Lambda}{\mu_0}\biggr)
$$
$$
+\frac{e_i^{(4)2}\tilde{b}_i}{16\pi^2}
\int^{r^2/\mu_0^2}_{r^2/\Lambda^2}\frac{dt}{t}
\biggl[\theta_3\biggl(\frac{it}{\pi R^2}\biggr)\biggr]^d,
\end{equation}
where the $b_i$ are the one-loop beta-functions for the zero modes, while
the $\tilde{b}_i$ denote the beta-functions associated with the
Kaluza-Klein excitations.
We can now obtain the scaling behaviour of our gauge coupling constants
\begin{equation}
\alpha_i^{-1}(\Lambda)=\alpha^{-1}_i(\mu_0)-\biggl(\frac{b_i-\tilde{b}_i}{2\pi}\biggr)
\ln\biggl(\frac{\Lambda}{\mu_0}\biggr)
$$
$$
-\frac{\tilde{b}_i}{4\pi}\int^{r^2/\mu_0^2}_{r^2/\Lambda^2}
\frac{dt}{t}\biggl[\theta_3\biggl(\frac{it}{\pi R^2}\biggr)\biggr]^d,
\end{equation}
where $\alpha_i\equiv e_i^{(4)2}/4\pi$. This gives the
running of the gauge coupling constants as obtained by Dienes et
al.\cite{Dienes2}. The important result emerges that the Kaluza-Klein
exitations convert the standard logarithmic scaling of the gauge coupling
constants to a power law running behaviour.
By imposing matching conditions that the uncorrected value of the effective
four-dimensional coupling constant $\alpha_i$ must agree with the value of
the four-dimensional coupling $\alpha_i(\mu_0)$ at the scale $\mu_0$, we
get
\begin{equation}
\alpha_i^{-1}(\Lambda)=\alpha^{-1}_i(M_Z)-\frac{b_i}{2\pi}
\ln\biggl(\frac{\Lambda}{M_Z}\biggr)
+\frac{\tilde{b}_i}{2\pi}\ln\biggl(\frac{\Lambda}{\mu_0}\biggr)
-\frac{\tilde{b}_iX_d}
{2\pi d}\biggl[\biggl(\frac{\Lambda}{\mu_0}\biggr)^d-1\biggr],
\end{equation}
where $M_Z$ is the mass of the Z-boson. This result is valid for all
$\Lambda\geq \mu_0$. The Kaluza-Klein excitations cause {\it an
acceleration of the unification of the gauge coupling constants}
$\alpha_i$; this leads to the possibility of having gauge coupling
unification at energies well below the usual GUT scale of $10^{16-19}$ GeV.
Let us consider the unregulated
interaction Lagrangian, describing the coupling of fermions to zero modes
and to the Kaluza-Klein modes:
\begin{equation} {\cal L}_{\rm int}
=e^{(4)}_iJ^\mu(A_{\mu i}+\sqrt{2}\sum^{\infty}_{n=1}A^{(n)}_{\mu i}),
\end{equation}
where $A_{\mu i}$ are the zero modes and $A^{(n)}_{\mu i}$ are the
Kaluza-Klein modes, respectively. Integrating out the $W$ boson and its
Kaluza-Klein excitations gives for the effective standard model Fermi
constant\cite{Nath,Graesser}
\begin{equation} G^{\rm eff}_F=G^{\rm
SM}_FK_d\biggl(\frac{M_W^2}{M_R^2}\biggr),
\end{equation} where $d=D-4$ and
$K_d$ is \[
K_d(s)=\int^\infty_0 dt\exp(-t)\biggl[\theta_3\biggl(\frac{it}{s\pi}\biggr)
\biggr]^d.
\]
This integral diverges for $d > 1$, but a convergent result is obtained
in regularized FQFT, corresponding to a truncation of the Kaluza-Klein states when the
masses exceed the FQFT scale $\Lambda$, associated with the regulated
Lagrangian.
Nath and Yamaguchi obtain the ratio of the Kaluza-Klein contribution to the
Fermi constant to the standard model value of the Fermi constant for
$d\geq 3$:
\begin{equation}
\frac{\Delta G_F^{KK}}{G_F^{\rm SM}}\simeq \biggl(\frac{d}{d-2}\biggr)
\frac{\pi^{d/2}}{\Gamma(1+d/2)}
\biggl(\frac{\Lambda}{M_R}\biggr)^{d-2}\biggl(\frac{M_W}{M_R}\biggr)^2.
\end{equation}
The gauge coupling evolution constrains $M_R$ and $\Lambda$ for TeV scale
unification. Let us adopt the evolution equation\cite{Dienes2} used
by Nath and Yamaguchi:
\begin{equation}
\alpha_i(M_Z)=\frac{1}{\alpha_U}+\frac{b_i}{2\pi}\ln\biggl(
\frac{M_R}{M_Z}\biggr)-\frac{b_i^{KK}}{2\pi}\ln\biggl(\frac{\Lambda}{M_R}
\biggr)
+\Delta_i,
\end{equation}
where $\alpha_U$ is the effective GUT coupling constant, $b_i=(-3,1,33/5)$
for $SU(3)_c\times SU(2)\times U(1)$, $b^{KK}_i=(-6,-3,3/5)$ are the
$b_i$ minus the fermion sector contribution which has no Kaluza-Klein
excitations, and $\Delta_i$ are the Kaluza-Klein corrections. Given $d$ and
$M_R$ the unification of $\alpha_1$ and $\alpha_2$ fixes $\Lambda/M_R$. For
the cases of $d$ equal to 2, 3 and 4 extra dimensions, the Nath-Yamaguchi
analysis produces the lower limits on $M_R$ of 3.5 TeV, 5.7 TeV and 7.8
TeV. Thus, the observation of Kaluza-Klein excitations may be possible at
the Large Hadron Collider (LHC).
Nath and Yamaguchi\cite{Nath} have also considered the constraints arising
from an analysis of $g_\mu-2$ for extra $d$ dimensions. These constraints
on $M_R$ at the $2\sigma$ level are $M_R > 1.6$ TeV for $d=1$, $M_R > 3.5$
TeV for $d=2$, $M_R > 5.7$ TeV for $d=3$ and $M_R > 7.8$ TeV for $d=4$.
\section{Proton Decay Lifetime and Unified Theory Phenomenology}
The problem of proton decay must be considered in the context of the
unified models. When both colour triplet quarks and colour singlet leptons are
assigned to the same irreducible representation of a symmetry group, then
there exist vector bosons (leptoquarks) that transform leptons into quarks.
Unless there exists a conserved quantum number $A$ for which the proton is
the lowest mass state with $A=1$, then the proton can decay. We must
guarantee within our unified field theory that the proton is stabilized
sufficiently to not disagree with the experimental bounds on its lifetime,
$\tau_p\geq 10^{32}$ yrs. For example, as is well-known, the conventional
$SU(5)$ model of Georgi and Glashow\cite{Glashow} has been eliminated by
the experimental bound on the decay rate. The problem becomes much more
severe when we contemplate a compactification scale $M_c$ of order 1-10
TeV.
In our theory, there exist several energy scales to consider. There is the
Yang-Mills scale $\Lambda_{\rm YM}$, associated with the Yang-Mills
Lagrangian, ${\cal L}_{\rm YM}$, the Higgs scalar field scale $\Lambda_H$,
and the gravitational scale $\Lambda_G$. In addition, there is the
aforementioned compactification scale $M_c$. Let us first consider the
possibility that the Yang-Mills and the compactification scales are large,
$\Lambda_{\rm YM}\sim M_c\geq 10^{16}$ GeV.
For the $SO(10)$ model, we choose the left-right symmetric
$SU(2)$ Pati-Salam breaking pattern \cite{Pati}:
\[
SO(10)\rightarrow SU(4)\times SU(2)_L\times SU(2)_R\rightarrow SU(3)\times
SU(2)_L\times SU(2)_R\times U'(1)_{B-L}
\]
\[
\rightarrow SU(3)\times SU(2)\times
U(1)\rightarrow SU(3)\times U(1)_{\rm em}.
\]
Here, the sequence of characteristic Higgs potentials are described by the
representations $\underline{54}, \underline{45}, \underline{16}$ and
$\underline{10}$. At one loop level, the renormalization group equations
are for $\mu < M_c \sim\Lambda_{\rm YM}$\cite{Dienes2,Ross}
\begin{equation}
\sin^2\theta_W(M_W)=\frac{3}{8}-\frac{11}{3}\frac{\alpha(M_W)}{\pi}
\biggl[\frac{5}{8}\ln\biggl(\frac{M_U}{M_W}\biggr)-\frac{3}{8}
\ln\biggl(\frac{M_U}{M_S}\biggr)\biggr],
\end{equation}
and
\begin{equation}
\frac{\alpha_{\rm em}(M_W)}{\alpha_3(M_W)}=\frac{3}{8}-\frac{11}{8}
\frac{\alpha(M_W)}{\pi}\biggl[3\ln\biggl(\frac{M_U}{M_W}\biggr)
-\ln\biggl(\frac{M_U}{M_S}\biggr)\biggr],
\end{equation}
where $M_U$ denotes the mass of the exchanged boson that breaks $SO(10)$,
while $M_W$ is the weak scale mass. We have assumed that $M_C\sim M_U$,
where $M_C$ is the mass of the exchanged boson that breaks $SU(4)\times
SU(2)_L\times SU(2)_R$ to $SU(3)\times SU(2)_L\times SU(2)_R\times
U'(1)_{B-L}$. Moreover, $M_S$ denotes the mass associated with the breaking
of $SU(3)\times SU(2)_L\times SU(2)_R\times U'(1)_{B-L}$ to $SU(3)\times
SU(2)\times U(1)$. The term proportional to $\ln\biggl(M_U/M_W\biggr)$
corresponds to the low energy group $SU(3)\times SU(2)\times U(1)$. We can
choose $M_S$ to keep the prediction for $\alpha(M_W)/\alpha_3(M_W)$ fixed
at the $SU(5)$ value, while varying $M_U$ from the usual $SU(5)$ value for
$M_X$. This increases $\sin^2\theta_W(M_W)$ by approximately $0.005
\ln\biggl(M_U/M_X\biggr)$. We can still obtain reasonable values for
$\sin^2\theta_W$ for a large variation of $M_U$. Consequently, the proton
decay lifetime, which satisfies
\begin{equation}
\tau_p\propto 10^{30}\,{\rm yrs}\biggl(\frac{M_U}{5\times 10^{14}\,{\rm
GeV}}\biggr)^4
\end{equation}
can be made to satisfy the experimental bounds.
This illustrates the freedom in building an $SO(10)$ model, because of the
uncertainties in the values of $M_U$ and $\tau_p$, once we go beyond the
simplest, minimal symmetry breaking scheme. However, these models do
possess the possibility of having very light scales of intermediate
unification, which may be associated with baryon and lepton number
violating processes in other than proton channel decays.
We must now turn our attention to the Higgs hierarchy problem. With the
foregoing scenario, we can choose for the Higgs scale, $\Lambda_H\sim 1$
TeV, which guarantees in FQFT (see Sect.7) that the {\it radiative loop
contributions to the scalar Higgs self-energy are controlled}, leaving the
tree graphs untouched at higher energies, which can mediate Higgs
spontaneous symmetry breaking. We have the possibility of reducing the
graviton quantum loop contributions to the TeV scale by choosing
$\Lambda_G\sim 1$ Tev, or choosing a much higher value for $\Lambda_G$,
e.g. the Planck scale $\sim 10^{19}$ GeV.
This scenario does solve the Higgs hierarchy problem, while simultaneously
ensuring a sufficiently stable proton, but the extreme difference between
the Yang-Mills scale $\Lambda_{\rm YM}$ and the Higgs scalar field scale
$\Lambda_H$ does raise questions about "naturalness". However, we stress
that these energy scales are associated with {\it physical} scales in FQFT,
and not with arbitrary cut-offs.
Let us consider next a scenario in which $\Lambda_{\rm YM}\sim M_c\sim
\Lambda_H\sim \Lambda_G\sim$ 1-10 TeV. This produces, at first sight, an
attractive physical picture in which we can contemplate observing new
physics with the next generation of accelerators. The classical GR theory
is left unchanged, because only the graviton quantum loop graphs are
reduced to a scale of 1-10 TeV, leaving classical GR intact to all
energies. The gauge coupling constant will satisfy a power law behaviour
when $\Lambda_{\rm YM}\sim M_c > \mu$, so gauge coupling unification is
accelerated, as explained in Sect 8. There is no Higgs hierarchy problem in
four dimensions, and for our $SO(18)$ unification scheme {\it we obtain the
correct prediction of three chiral families and the standard model in four
dimensions}. However, we are faced with the {\it b${\hat e}$te noire} of
potentially fast proton decay.
There are two possible scenarios that we can adopt to circumvent the
problem of baryon and lepton number violation. The first is to introduce a
global $U(1)$ symmetry. The conservation of $A$ does not carry with it a
long-range force, so even though $A$ is an additive quantum number like
charge $Q$, it cannot be a generator of a local $U(1)$. However, if a
global $U(1)$ and a local $U(1)$ are both broken in such a way that a
linear combination of the generators is conserved, then the vector boson
acquires a mass and the unbroken linear combination yields an exact
conservation law\cite{Gell-Mann2}. Let $X$ be the generator of the local
$U(1)$ and $Z$ the generator of the global $U(1)$. We choose $Z=0$ for the
fermions and assign all fermions to a single irreducible representation (as
is the case for our $SO(10)$ in four dimensions). The Higgs fields have
non-zero values of $Z$, and if $Z$ and $X$ are broken and their sum is
conserved, then some of the scalar fields will acquire non-zero $A$,
resulting in a heavy boson and $A=X$ in the fermion sector. Thus, our
$SO(10)$ theory has a stable proton. It is possible to break the $Z$ and
$X$ so that no ``weird" fermions exist, but in general there will exist
"weird" fermions at higher energies.
Of course, the introduction of a global $U(1)$ symmetry into the theory is
considered by some to be an unpalatable way of solving the proton decay
problem. It goes against the spirit of local gauge symmetries. It means
that if we rotate a proton in our living room, then exactly the same
rotation is required to be performed in the Andromeda galaxy. There is also
the danger that gravitational interactions will induce fast proton decay in
higher-dimensional operators. Moreover, there is the more abstract issue
that black holes violate all non-gauged symmetries.
Another possible scenario to evade the proton decay issue has recently been
proposed by Arkani-Hamed and Schmaltz\cite{Schmaltz}. They ``stick" the
standard model fermions at different points on domain walls in the
extra $d$ dimensions. The couplings between them are suppressed due to the
exponentially small overlaps of their wave functions. We can adopt this
mechanism in our higher-dimensional field theory. The model can be simply
visualized in one extra dimension, in which the gauge fields and the Higgs
fields are allowed to propagate inside the wall, while the fermions are
constrained to different points in the wall. The fermion wave functions are
described by narrow Gaussian functions. The long-distance four-dimensional
theory can have exponentially small Yukawa couplings, generated by the
small overlap between left- and right-handed fermion wave functions. This
can lead to an exponentially suppressed proton decay rate, if the quarks
and leptons are localized to separate ends of the wall. This avoids the
issue of inventing symmetries in the theory which protect the proton from a
fast decay rate. Since this mechanism relies only on the dynamics of the
wall geometry and the placements of the fermions on the walls in the extra
dimensions, one may develop an uneasy feeling that the mechanism is
somewhat contrived, but at present there appears to be no obvious critical
reason why this could not be an acceptable way to resolve the proton decay
problem.
\section{\bf Conclusions}
A higher-dimensional unified field theory based on a
Kaluza-Klein-Yang-Mills-Higgs action and a ground state $M^4\times B$ is
developed. The gauge group has a topologically non-trivial
sub-group associated with the compact internal space, i.e. a
Dirac monopole is inserted into the latter group to guarantee that the
compactification to four dimensions retains the chiral non-vector-like
property of the fermions. We choose the minimal, anomaly free model
$SO(18)$ in twelve dimensions with the breaking to $SO(8)\times SO(10)$,
leading to a four-dimensional group $SO(10)$ GUT model, which contains the
standard model $SU(3)_c\times SU(2)\times U(1)$ and predicts {\it three
families of quarks and leptons}. The chiral nature of the fermion
representations is guaranteed in the dimensional reduction process by
making $SO(8)$ topologically non-trivial with the reduction
$SO(8)\rightarrow Sp(4)\times SU(2)$. The fermions have a non-vanishing
chirality number, whereby the left-handed quarks and leptons have the
correct physical interpretation.
The problem of the stability of Kaluza-Klein theories can be solved in
our model by means of the supplementary Yang-Mills fields or the magnetic
monopole that exists in the internal space, which can generate the
repulsive forces necessary to balance the gravitational forces.
The important gauge hierarchy problem, associated with the Higgs sector, is
solved by the exponential damping of the Higgs self-energy in the Euclidean
$p^2$ domain for $p^2 > \Lambda_H^2$, and for a $\Lambda_H$ scale in the
TeV range.
The unified field theory will have three coupling constants,
namely, the gravitational constant $\bar G$, the gauge coupling
constant $e$, and the scalar Higgs coupling constant g, which have
to be rescaled in four dimensions. Dimensional reduction to four dimensions
can explain charge quantization in terms of the compactification scale $R$.
In string theory, the coupling constants are determined by the string
dilaton scalar field, so in principle there are no arbitrary coupling
constants. However, this presupposes that one knows the {\it complete}
solution to the dynamics of the string equations. It is difficult to see
how the determination of the coupling constants in string theory can be
implemented in practice, particularly, since it would appear that only
non-perturbative solutions can be obtained in M-theory.
The critical issue of the finiteness of quantum gravity perturbation theory
in D dimensions is solved by applying the FQFT formalism. The nonlocal
quantum loop interactions reflect the non-point-like nature of the
field theory, although we do not specify the nature of the extended object
that describes a particle. Thus, as with string theories, the point-like
nature of particles is ``fuzzy" in FQFT for energies greater than the scale
$\Lambda$. One of the features of superstrings is that they provide a
mathematically consistent theory of quantum gravity, which is ultraviolet
finite and unitary. FQFT focuses on the basic mechanism behind string
theory's finite ultraviolet behavior by invoking a suppression of bad
vertex behavior at high energies, without compromising perturbative
unitarity and gauge invariance. FQFT provides a mathematically consistent
theory of quantum gravity at the perturbative level. If we choose
$\Lambda_G\sim$ 1-10 TeV, then quantum radiative corrections to the
classical tree graph gravity theory are perturbatively negligible to all
energies greater than $\Lambda_G$ including the Planck energy. If, on the
other hand, we choose $\Lambda_G\sim M_{\rm Planck}$, then we are forced to
seek a non-perturbative FQFT quantum gravity formalism at the Planck scale.
Our solution of the finiteness of quantum gravity is based on a {\it gauge
field theory}, which allows us to go off the mass shell when calculating
vertex operators. It is generally accepted that there is no self-consistent
string field theory, because such a theory would correspond to a $\phi^3$
theory with a Hamiltonian unbounded from below, causing the field theory to
be unstable. Our FQFT can be a stable theory as was shown by Kleppe and
Woodard\cite{Kleppe2}. However, our higher-dimensional field theory may be
linked to a final, realized stable version of M-theory, for the finiteness
of FQFT owes its existence to an ultraviolet suppression mechanism
akin to that of string theory, and a field theory version of M-theory could
possess a structure similar to our FQFT. Superstring theory and its
offspring appear to lead to fundamental changes in our understanding of
space and time\cite{Hooft}. In particular, our everyday notions of
causality may be altered at high energies and small distances. Our
introduction of nonlocal interaction Lagrangians in FQFT may be the
modification of local, point particle field theory at the quantum level
that is needed to achieve a mathematically consistent theory of quantum
gravity.
Supersymmetry is required if we wish to unify the particle spins of bosons
and fermions. This would be a mathematically beautiful achievement. We
could incorporate supersymmetry in our higher-dimensional FQFT, and indeed
in contrast to, for example, dimensional regularization, FQFT respects
continuous supersymmetry gauge transformations to all orders of
perturbation theory. However, supersymmetry partners have not, as yet, been
observed and as we have demonstrated, FQFT can resolve the Higgs hierarchy
problem without supersymmetry, thereby removing the primary reason for
promoting supersymmetry at the phenomenological level.
Because we are able to lower the compactification scale $M_c$ to the TeV
energy range, we anticipate that Kaluza-Klein excitations will be
observable at these energies by the LHC. In order to distinguish these
Kaluza-Klein signatures from the signatures of supersymmetry partners or
other exotic physics, we must analyze further the decay properties of the
Kaluza-Klein modes, so as to select possible unique features associated
with the excitations.
We have adopted the idea of reducing quantum gravity to lower energies by
choosing the quantum gravity scale $\Lambda_G$ in the TeV range, or even at
much lower energies. This is in accord with the recent interesting idea
that the gravitational scale could be in the TeV range\cite{Arkani}.
However, in contrast to the work in ref. (8), we only lower the energy
scale of the quantum gravity loops through the choice of the scale
$\Lambda_G$, without affecting the classical tree graphs which sum to give
local and causal, classical Newtonian and GR theories.
We have not considered the
implications of our theory for cosmology and black holes, nor have we
concerned ourselves with the important problem of the cosmological
constant. These issues will be addressed in future work.
\vskip 0.2 true in
{\bf Acknowledgments} \vskip 0.2 true in This work was supported by the
Natural Sciences and Engineering Research Council of Canada.
\vskip 0.5 true in
|
\section{Introduction}
\label{sec:Introduction}
In our previous papers \cite{1991MNRAS.249..498M,MSFR,MSCFG} we have
investigated the evolution of the counts and colours of faint field
galaxies, culminating in extremely deep optical imaging of a
representative region of sky we call the ``Herschel Deep Field''
\cite{MSF}. In this paper we present new near-infrared observations of
this field, and combine this data with our existing ultra-deep optical
imagery. Near-infrared observations have many advantages which are
oft-quoted in the literature, among them, the insensitivity of the $k-$
correction to type or star-formation history at intermediate redshifts
is perhaps the most relevant in a cosmological setting.
Previously we have considered our observations in the context of pure
luminosity evolution (PLE) models. Recently, additional support for
these models has come from fully complete, blue-selected, redshift
surveys \cite{CSH2} which have revealed the presence of an extended
tail in the redshift distribution, partially resolving the paradox that
non-evolving galaxy formation models could be used to fit the results
of the early faint redshift surveys \cite{1995MNRAS.275..169G,BEG}. Such
models, as we have shown, are able to reproduce all the observable
quantities of the faint field galaxy population (counts, colours,
redshift distributions) at least for low $\Omega_0$ Universes and
within current observational uncertainties. High $\Omega_0$ Universes
can be accommodated by the model if we add an extra population of low
luminosity galaxies with constant star-formation rates which boost the
counts at faint ($B>25^m$) magnitude levels \cite{MSCFG,1997ApJ...488..606C}.
In the PLE model, galaxies form monolithically at high redshift.
Changes in luminosity and colour after the formation event are modelled
with stellar population synthesis models. $K-$ selected galaxy samples,
at least until $K~\sim~20$, are dominated by early-type galaxies;
therefore, changes in optical-infrared galaxy colours are particularly
sensitive to evolution of this galaxy class. By investigating
near-infrared number counts and colour distributions we can potentially
address questions concerning the formation and evolution of elliptical
galaxies.
\begin{table}
\begin{center}
\caption{Photometric Limits of the Herschel Deep Field.}
\begin{tabular}{lccccc}
Filter& U & B & R & I & K \\
Limit (3$\sigma$) & 26.8 & 27.9/28.2 &26.3& 25.6&20.0/22.75\\
Area (arcmin$^2$) & 46.4 & 46.8/2.8 & 48.5 &52.2 & 47.2/1.78\\
\end{tabular}
\end{center}
\end{table}
\label{tab:intro.limits}
The two surveys presented here are complementary. The ``wide'' survey
covers a large, $\sim50$ arcmin$^2$ area to a shallow,
$K(3\sigma)=20^m$, depth, whilst the ``deep'' survey covers a much
smaller area of $\sim 1.8$ arcmin$^2$ to a much fainter
$K(3\sigma)=22.75^m$ limiting magnitude. In combination, both surveys
provide a $K-$ limited sample extending from $K\sim16^m$ to $K=22^m.75$.
This paper is organised as follows: Section~\ref{sec:ObsDat} describes
the data reduction techniques used in both the wide and the deep
surveys; in Section~\ref{sec:Results} number counts and colour
distributions for our both data sets are presented; in
Section~\ref{sec:Interpr-Modell} we discuss these results in terms of
our modelling procedure; and finally, in Section~\ref{sec:IntDisc} we
summarise the main conclusions we draw from our work.
\section{Observations and Data Reductions}
\label{sec:ObsDat}
\subsection{Reducing the ``Wide'' Survey}
\label{sec:RedWid}
The observations were made over four nights in July 1995 at the 3.8m
United Kingdom Infrared Telescope (UKIRT) using the IRCAM3
near-infrared detector with a standard $K-$ ($2.2~\mu$m) filter.
Conditions were photometric on two of the four nights. The mean seeing
was $\sim 1''$ FWHM\@. The centre of the region we imaged, RA $0$h
$19$m $59$s Dec $+0^o$ $4'$ $0''$, corresponds to the centre of our
deep optical fields which are described in \citeN{MSF}. For this region
we have imaging data complete to the limiting magnitudes listed in
Table~\ref{tab:intro.limits}. The optical observations were made at the
4.2m William Herschel and the 2.5m Isaac Newton telescopes (WHT and INT
hereafter) and cover approximately $\sim 50$ arcmin$^2$.
The IRCAM3 detector, a $256 \times 256$ element InSb chip, has a pixel
scale of 0.286$''$ and consequently each image covers only $\sim 1.5$ \
arcmin$^2$. For this reason we observed a mosaic of $6\times 6$ frames
which completely overlapped the previously-observed optical fields. The
entire mosaic was observed in each pass for two minutes at each
position, and offset slightly for each observation from the previous
pointing. In total, the entire survey area was covered fifteen times,
giving a total integration time at each sub-raster of 30 minutes. The
short exposures for each image is a consequence of the high sky
brightness in the near-infrared and is fact the longest exposure time
possible with IRCAM3 in the $K-$ before sky background causes the
detector to saturate. Our data reduction technique for both infrared
surveys follows the methods outlined in \citeN{1995ApJS...98..441G}.
A two-pass procedure is adopted: for the first stage in the data
reductions a dark frame is subtracted from each night's images,
calculated from the median of the dark frames taken that particular
night. Next, the images are sorted in order of time of exposure. Then,
a {\it running sky\/} is constructed from the median of the six images
nearest in time to the frame being processed. This is necessary as the
sky background varies on exceedingly short time-scales (i.e., under 30
minutes). This running sky is then subtracted from the current image
and the procedure repeated for all the images in the mosaic. Next, for
each night a ``superflat'' is constructed by medianing together all the
observations for that night. Each image is then divided by the
superflat.
Once all the images have been dark- and sky-subtracted and flat-fielded
they are grouped into 36 separate pointings and each stack averaged
together using the IRAF \textsc{imcombine} task. Next, masks were
constructed from each of the stacked images by flagging all pixels more
than $3\sigma$ from the sky background. These masks were then used in
re-calculating the running sky frames. This was necessary as bright
sources present in each of the individual exposures caused raised
counts in the original running sky frames. The whole procedure outlined
above was then repeated, using these new frames.
Finally, the $36$ stacked sub-rasters were combined into a single image
by matching each sub-raster with objects on the extremely deep $B-$
band frame and using this to calculate the correct position of each
sub-raster relative to the others. The final image, after trimming,
covers an area of $47.2$ arcmin$^2$, overlapping our WHT deep optical
fields.
We calibrated this data by repeatedly ($\sim 30$ times) observing the
faint UKIRT Standards \cite{CH92} each night at a range of airmasses,
although fortuitously the mean airmass of the standards and of the
program field images was approximately the same, so accurate
determination of the extinction coefficient was not necessary. The rms
scatter between our standards measurements was $\pm 0.03$ magnitudes.
\subsection{Reducing the ``Deep'' Survey}
\label{sec:RedDep}
Observations for the ``Deep'' survey were made over six nights in 1994
September at UKIRT using IRCAM3 with a $K-$ band filter. Conditions
were photometric during the first two nights, and the remainder of the
data were scaled to the photometry of bright objects in the field made
on those nights. The seeing ranged from 0.8$\arcsec$ to 1.4$\arcsec$.
The total observing time on the field was just over 100 ksec (27.9
hours), and was made up of 838 individual exposures of 120 seconds
each. Each individual exposure was made up of 12 co-adds of 10 seconds
each. The telescope was dithered using the offset guide camera in a
diamond pattern of 13 positions, each position separated by
5.25$\arcsec$ in both Right Ascension and declination. Data reduction
proceeded generally in the manner described in Section
\ref{sec:RedWid}. Dark frames were made several times each night, and
were subtracted from the images. A running sky was constructed, using
the median of 12 images (six before and six after). The running sky was
subtracted (after applying the two-pass object masking system described
above), and a ``superflat'' was constructed for each night and divided
into the data. The input dither pattern was recovered and adjusted by a
centroid of the brightest object in the field. The data taken each
night were shifted and co-added. The photometry of each night's data
were scaled to that of the photometric first two nights (which were
themselves consistent). The final image covers an area of 1.78
arcmin$^2$, and has a limiting depth of 3$\sigma$ = 22.75. The deep
field is centred at 00 19 59.6 +00 02 56 (1950.0) and covers 81$''$
square (140x140) 0.58'' pixels.
\begin{table*}
\begin{center}
\caption{Detection and measurement parameters used in both surveys.}
\begin{tabular}{lccccc}
Frame&Limiting isophote&Detection limit&Min. r&Kron multiplying &Correction to\\
&(mag/arcsec$^2$)&(K mag)&($''$)& factor&total (mag)\\
Deep K&24.50&23.50&1.25&1.40&0.32\\
Wide K&21.50&22.00&1.35&1.50&0.25\\
\end{tabular}
\end{center}
\label{tab:K.info}
\end{table*}
\subsection{Object Detection And Photometry}
\label{sec:ObjRed}
For both the deep and wide images we use the same prescription for
object detection and photometry as in our optical studies
\cite{MSFR,MSF}. Note that to improve the reliability of image
detection both images were binned $2\times2$ (the $0.286''$ pixel size
of IRCAM3, combined with median seeing of $\sim 1''$ FWHM, means that
our raw images are oversampled). We fit the sky background
approximately with a 3rd order polynomial and then subtract this from
the frame. Deep isophotal image detection is then undertaken (to a
magnitude about $1^m$ deeper than the $3\sigma$ limit for the frame --
see Table ~\ref{tab:K.info}) and the objects removed from the frame and
replaced by a local sky value. \footnote{Our $3\sigma$ limit is
defined as when the variance of the sum of the pixel values within our
detection aperture reaches this value.} The resulting
image is heavily smoothed (with several passes with box smoothing up to
$10$ pixels on a side) and then subtracted from the original. The
result is an extremely flat image on which the isophotal detection is
then rerun. At this stage, in order to reduce false detections, any
images within a few pixels of one another are recombined into one. A
Kron-type magnitude \cite{1980ApJS...43..305K} is then calculated for
each image using a local sky value. This is essentially an aperture
magnitude calculated to $mR_k$, where $R_k$ is the Kron radius and $m$
is the multiplying factor given in Table~\ref{tab:K.info}. It is
necessary to set a minimum aperture equivalent to that for an
unresolved object, and apply a correction to total magnitude. These are
also listed in Table~\ref{tab:K.info}. A more detailed discussion of
this technique is given in \citeN{1991MNRAS.249..498M}. Our
optical-infrared colours are measured inside fixed $1.5''$ radius
apertures. Note that to form the ``Deep'' $(B-K)$ colours, the $K-$
band magnitudes are matched with the $2'$ diameter, stacked, ultra-deep
$B-$ band image described in \citeN{MSF} (and see
Table~\ref{tab:intro.limits}). All other colours are formed from
matches with the $7'\times7'$ Herschel Deep Field optical images.
Star-galaxy separation is taken from \citeN{MSF}.
\subsection{Confusion corrections}
\label{sec:Confus}
The deeper the ground-based number-counts are extended the more
important the corrections for confusion becomes. This is of particular
relevance for the deep K-band data. To make an estimate of the
completeness of our counts on this frame we have added numerous
artificial stars of various known magnitudes to the real data frame and
then reanalysed the resulting images using the normal data reduction
procedure. We then compare the measured magnitudes with those which we
input. Table~\ref{tab:K.sim} gives the mean magnitudes, scatter, and
detection rate for these stars. Note that an image is considered
undetected if it is merged with another image and the combined
brightness is a factor two or more greater than the true magnitude, or
if it is not found within $\pm2$ pixels of its true position. As
expected the detection rates in the real data drop as the magnitude
falls. This is almost entirely due to objects being merged with other,
brighter objects - isolated objects are recovered with almost $100\%$
efficiency. Note, however, that the mean recovered magnitudes are close
to the true values. Although most objects at the limit of our data are
galaxies, they are in general close to being unresolved, and so we
correct our counts by the detection rates implied by our artificial
stars.
Due to the much wider mean separation between objects, the shallower
$K-$ band data do not suffer from confusion, and no correction is
made.
\begin{table*}
\begin{center}
\caption{Results of adding artificial stars to the deep K data.}
\begin{tabular}{ccc}
True magnitude&Measured magnitude&Detection rate (\%)\\
\\
20.75&$20.70\pm0.17$&90\\
21.25&$21.22\pm0.21$&88\\
21.75&$21.76\pm0.26$&85\\
22.25&$22.36\pm0.36$&76\\
22.75&$22.83\pm0.37$&51\\
\end{tabular}
\end{center}
\label{tab:K.sim}
\end{table*}
\begin{figure*}
\centering
\centerline{
\epsfysize = 18.5cm
\epsfbox{fig1.eps}}
\vspace{-1.0in}
\caption{Differential galaxy number counts in half-magnitude intervals as a
function of $K-$ limiting magnitude as measured from our UKIRT wide
survey (filled circles) and from our UKIRT deep survey (filled
squares). Also plotted is a compilation of published $K-$ band number
counts and the predictions of the two non-evolving models described
in the text for $q_0 = 0.05$ (dotted line) and $q_0=0.5$ (solid
line). Error bars are calculated from the number of galaxies in each
magnitude bin using Poisson $\sqrt n$ counting statistics. The inset
shows the number counts from our surveys alone (this time with open
symbols) compared to the non-evolving predictions.}
\label{fig:countscols.counts}
\end{figure*}
\nocite{DSP}
\nocite{1986MNRAS.223...11M}
\nocite{1995ApJS...96..117M}
\nocite{GSCF}
\nocite{1997ApJ...476...12H}
\section{Results}
\label{sec:Results}
The final ``wide'' catalogue contains 298 galaxies to a
$K(3\sigma)=20.0^m$ over an area of 47.2~arcmin$^2$. The deep catalogue
contains 86 galaxies to a $K(3\sigma) = 22.75^m$ over an area of
1.8~arcmin$^2$.
\subsection{$K-$ band Galaxy Number Counts}
\label{sec:countscols.Kband}
Number counts derived from both our catalogues are presented in
Figure~\ref{fig:countscols.counts}, which shows our data (with error
bars calculated using Poisson counting statistics) and a compilation of
published number counts. At bright magnitudes ($16^m < K^m < 20^m$) we
agree well with the counts from the literature, whereas at fainter
magnitudes ($20^m < K< 22^m$) our data favours the higher counts found
by \citeN{1998ApJ...505...50B} as opposed to the lower counts measured
by \citeN{DSP}. \citeANP{1998ApJ...505...50B} claim that the aperture
corrections employed by \citeN{DSP} resulted in an overestimation of
the depths of the latter survey by $\sim 0.5^m$; our results support
this interpretation.
It is worth noting that in our last bin, as in
\citeANP{1998ApJ...505...50B}, our counts are only approximately $50\%$
complete and therefore subject to large corrections. Furthermore, given
the small areas of surveys at this magnitude level (where areas covered
are typically $\sim 1$ arcmin$^2$) it is also possible that cosmic
variance could account for the large observed field-to-field variations
between the different groups.
Considering the slope of our number counts, we derive a value of
$d(\log N)/dm~=~0.37~\pm~0.03$ ($1\sigma$ errors) over the magnitude
range covered by both wide and deep catalogues (namely, $K~\sim
16-23$). This value is in good agreement with $d(\log
N)/dm~=~0.36~\pm~0.02$ found by \citeANP{1998ApJ...505...50B}. We do
not find the count slope for the deeper survey to be significantly
different for the shallower one (however, as pointed out by \citeN{GCW}
the compiled $K-$ band number counts show a slope change at $K\sim17$).
In Figure~\ref{fig:countscols.counts} we also plot the predictions of
two non-evolving galaxy count models: one with deceleration parameter
$q_0=0.5$ (solid line) and one with $q_0=0.05$ (dotted line). Both
models give a good fit the observations from $K\sim14$ to $K\sim22$.
These non-evolving model were computed from the $k-$ corrections of
\citeN{BC} and the $z=0$ luminosity function parameters given in
Table~\ref{tab:pars}. We compute $M^*_K$ by combining the tabulated
$M^*_B$ with the rest-frame $(B-K)$ colours. In our models we adopt the
same normalisation as in our previous papers. As we and others have
noted before \cite{1991MNRAS.249..498M,sh89}, normalising galaxy counts
at brighter ($B\sim16^m$) magnitudes requires that substantial amounts
of $B-$ band evolution must take place at relatively low ($z<0.1$)
redshift. Conversely, by normalising at $18^m<B<20^m$ we allow a
better fit at intermediate magnitudes and reduce the amount of
evolution required to fit the counts at fainter magnitudes, but
under-predict counts at brighter magnitudes. Support for this high
normalisation has recently come from the 2-micron all-sky survey (2MASS;
\shortciteNP{1992robt.proc..203K}) which finds a significant underdensity in
the galaxy counts at $10^m<K<13^m$ and is discussed further in a
forthcoming paper (Shanks et al 1999).
Could substantial galaxy luminosity evolution be occurring at these low
redshifts? Bright, $K<20^m$ samples are dominated by early-type
galaxies Morphologically-segregated number counts from the Medium Deep
Survey \cite{1994ApJ...437...67G} and the Hubble Deep Field \cite{WBD}
indicate that counts for elliptical galaxies at intermediate magnitude
levels ($18^m<I_{816W}<22^m$) follow the predictions of a non-evolving
model
\cite{1998ApJ...496L..93D,1995ApJ...453...48D,1995MNRAS.275L..19G}.
This would argue against an evolutionary explanation for the steep
slope observed in the range $10^m<K<14^m$. We are then faced with
finding an alternate explanation for the number counts under-density;
one suggestion comes for the findings of \citeN{MFS} and
\citeN{1997A&A...317...43B} who both show how scale errors (i.e.,
non-linearities) in the photographic APM photometry of
\citeN{1990MNRAS.247P...1M} could result in anomalously low galaxy
counts at brighter magnitudes in $B-$ selected surveys; however, the
$K-$ band surveys are carried out with electronic detectors, and such
effects are likely to be less significant (although the
\citeN{1986MNRAS.223...11M} survey was based on objects selected from
$B-$ band photographic plates). Another suggestion is that our galaxy
may reside in a locally under-dense region of the Universe \cite{sh89};
however, to produce the observed discrepancy in the number counts would
require that this void would be $\sim 150 h^{-1}$~Mpc, and have an
under-density of $\sim 30 \%$; fluctuations on this scale are difficult
to understand in terms of current cosmological models and also measured
large-scale bulk flows. Some evidence for a local void has come from
studies of peculiar velocities of type Ia supernovae
\cite{1998ApJ...503..483Z} although the size of the void they detect
($\sim70 h^{-1}$~Mpc) is not large enough to explain the low number
counts. For time being, resolution of this issue awaits a better
determination of the bright end of the $B-$ and $K-$ number counts by
much larger surveys.
\subsection{Optical-Infrared Colours}
\label{sec:Optic-Infr-Colo}
\begin{figure*}
\centering
\centerline{\epsfxsize = 12.0 cm
\epsfbox{fig2.eps}}
\caption{$K-$ magnitude against $(B-K)$ colour for galaxies in the
wide survey (open triangles) and the deep survey (open circles). Also
plotted are the median $(B-K)$ colours in one-magnitude bins for our
survey (filled squares), with $1\sigma$ error bars calculated by a
bootstrap resampling technique. Triangles show median colours from
the compilation of Gardner et al. 1993. The dotted line represents
the region of incompleteness; galaxies below this line have $B<28^m$.
Drop-outs are plotted at the upper limit of their colours as
right-pointing arrows. For comparison, predictions from two
non-evolving models are plotted; one for $q_0 = 0.05$ (dotted line)
and one for $q_0=0.5$ (solid line).}
\label{fig:medBK}
\end{figure*}
Figure~\ref{fig:medBK} shows the $(B-K)$ {\it vs} $K$ colour-magnitude
diagram for both samples for galaxies in both the ``wide'' (open
triangles) and the ``deep'' survey (open circles). Additionally the
median $(B-K)$ colour computed in one-magnitude bins is shown (filled
squares) and compared to the values given in \citeN{GCW} (filled
triangles). Error bars on the median colours were calculated using a
bootstrap resampling technique.
Incompleteness is represented by the dashed line; objects to the right
of this line have $B>28^m$. Several objects have been identified in
both surveys which are undetected in $B-$ but are detected in $K-$ and
these are plotted in the figure as right-pointing arrows and are
discussed in Section~\ref{sec:ExtRed}. Considering the median $(B-K)$
colour we see that it increases steadily until around $K\sim 18^m$
where it reaches a maximum value of $\sim 6$. After this, it sharply
becomes bluer and this blueward trend continues to the faintest limits
we have measured. Our median colours agree well with those in
\citeN{GCW}, even for brighter bins where the number of galaxies in our
sample are much smaller. We have also found our $K<18$ $(B-K)$
distributions to be in good agreement with the brighter survey of
\citeN{1997AJ....114..887S}. Finally, it is worth noting the extremely
large spread in colour in this diagram -- at $K\sim 20$ $(B-K)$ ranges
from $2$ to $9$.
Additionally, the predictions of the two non-evolving models shown in
Figure~\ref{fig:countscols.counts} are plotted. In general, the
$q_0=0.5$ model gives a slightly better fit to the observed colours, at
least until $K~\sim~20^m$; at fainter magnitudes, galaxies are
significantly bluer than both model predictions.
Another interesting feature in Figure~\ref{fig:medBK} is the apparent
deficit of galaxies with magnitudes in the range $K>21^m$
and colours from $(B-K) > 5$. The different areal coverages and
depths of the two surveys doubtless exacerbates this effect but its
presence is consistent with a population of galaxies which turns
rapidly blueward faintwards of $K~\sim~20$. We discuss the origin of
this feature in Section~\ref{sec:Medi-Colo-Counts}.
\begin{figure*}
\centering
\centerline{\epsfxsize = 12.0 cm
\epsfbox{fig3.eps}}
\caption{$(B-R)$ colour against $(R-K)$ colour for galaxies in the wide
survey (open circles) and the deep survey (filled circles). Fainter
galaxies are plotted with smaller symbols. Also shown are
evolutionary tracks for the five classes of galaxy in our model, as
computed from the population synthesis code of Bruzual \& Charlot
(93). The solid line represents type E/S0, the dotted line Sab and
the dashed, long dashed and dash-dotted lines types Sbc, Scd and Sdm
respectively. The dot-dashed line shows the path of of an elliptical
galaxy which has an IMF slope, x, of 3. For the elliptical tracks,
the labelled stars indicate the redshift. }
\label{fig:Col:Col}
\end{figure*}
In Figure~\ref{fig:Col:Col} we plot $(B-R)$ {\it vs.} $(R-K)$ colour
for all the galaxies in the wide survey (open circles) and the deep
survey (filled circles). Fainter galaxies are plotted with smaller
symbols. The ``deep'', $K<22.75^m$ sample is much bluer in both $(B-R)$
and $(R-K)$ than the ``wide'' than the brighter, $K<20^m$, sample,
which reaches $(R-K)\sim6$. We also plot colour-colour tracks
representing the different galaxy classes in our evolutionary models.
We defer discussion of these tracks until the following section.
\section{Interpretation and Modelling}
\label{sec:Interpr-Modell}
\subsection{Outline of the Models}
\label{sec:Outline-Models}
To interpret our results we investigate variants of a pure luminosity
evolution (PLE) model in which star-formation increases exponentially
with look-back time. Earlier versions of these models are discussed in
our previous papers \cite{1991MNRAS.249..498M,MSFR,MSCFG}. In this
paper we consider cosmologies in which $\Lambda=0$ and take
$H_0=50$~kms$^{-1}$Mpc$^{-1}$, although changing the value of $H_0$
does not affect any of the conclusions of this paper. Two values of the
deceleration parameter, $q_0$, $q_0 = 0.05$ and $q_0 = 0.5$, are
adopted, corresponding to open and flat cosmologies respectively.
(Zero-curvature cosmologies with $\Omega=0.3$ and $\lambda_0=0.7$ are
almost identical to our $q_0=0.05$ model because co-moving distance and
look-back time as a function of redshift are almost identical to that
found in the zero-$Lambda$ models, at least to $z\sim2$
\cite{1984ApJ...284..439P}). The input parameters to our models (given
in Table~\ref{tab:pars}) consist of observed {\it local\/} galaxy
parameters (namely, rest-frame colours and luminosity functions) for
each of the five morphological types (E/S0, Sab, Sbc, Scd and Sdm) we
consider in our models. These morphological types are divided into
elliptical (E/S0) and spiral (the remainder) and these two classes are
each given a separate star-formation history. We could, in principle,
sub-divide the spirals into different morphological types each with
different star formation histories but for simplicity we do not;
$(k+e)$ corrections for the different types are fairly similar to each
other in these models in any case . Instead, taking a Sbc model as
representative of all types we produce the other types by normalising
the Sbc track to the observed rest-frame $(B-K)$ colours from Table
\ref{tab:pars}. As we have already explained in
Section~\ref{sec:countscols.Kband} our $\phi^{*}$'s are chosen to match
the galaxy counts at $B\sim18-20$ and we seek to explain the low number
counts at bright magnitudes from a combination of photometric errors
and anomalous galaxy clustering, rather than substantial evolution at
low redshift. Our models also include the effects of the
Lyman-$\alpha$ forest, and, for spiral types, dust extinction
corresponding to the Large Magellanic Cloud as described in
\citeN{Pei}.
PLE models have difficulty in correctly reproducing the observed $K-$
band redshift distribution. Even essentially passive evolution
over-predicts the numbers of galaxies seen at $1<z<2$, and this fact
has been used by several authors (most recently
\citeN{1998MNRAS.297L..23K}) to argue for the hierarchical,
merger-driven galaxy formation models in order to reduce the mean
redshift of $K-$ selected galaxy redshift distributions. However, in
\citeN{MSCFG} we showed how PLE models {\it could\/} produce a $K-$
band redshift distribution compatible with the observations of
\citeN{CSH2} by assuming a very steep slope ($x=3$) for the stellar
initial mass function (IMF) in E/S0 types. This steep, low-mass-star
dominated IMF reduces the amount of passive evolution observed in $K-$
and allows us to reproduce the observed redshift distribution, which is
close to the predictions of the non-evolving model. It should be noted
however, that if the incompleteness in the \citeN{CSH2} n(z) were
entirely due to $z>1$ galaxies then the deficit relative to the PLE
model (with standard IMF) is only $\sim 3\sigma$ from $\sqrt N$
statistics alone and is likely to be less significant when galaxy
clustering is taken into account. Conversely, the $B-$ band redshift
distribution from the same survey does have a large fraction of high
redshift ($z>1$), high luminosity ($L>L^*$) galaxies. However, this is
not a problem for our models. Our steep luminosity function slope for
spiral types combined with a small amount of internal dust extinction
(corresponding to $A_B=0.3^m$ at $z=0$) allows us to reproduce this
result.
Although the resolution of the dataset discussed in this paper is too
low to extract useful size information for the faint galaxy population,
we mention in passing that observed faint galaxy sizes are compatible
with these models, models, as we have demonstrated in a recent paper
\cite{1998yugf.conf..102S}. We have used a modified version of Howard
Yee's PPP code to produce a simulated HDF frame which contains galaxies
with parameters based on the output of our PLE models. For this image,
spiral galaxies were generated assuming that luminosity evolution only
affects disk surface brightness and not size, combined with Freeman's
\citeyear{1970ApJ...160..811F}, law to relate disk size to absolute
magnitude at z=0. For bulges, we adopted the the diameter-magnitude
relation of \citeN{1990ApJ...361....1S}. By applying our photometry
package to this image we account for selection effects in a realistic
manner. We find that the galaxies in the simulation have sizes not
dissimilar from those measured in the HDF, in part because many are
intrinsically small galaxies seen well down the LF.
\begin{figure}
\begin{center}
\centerline{\epsfxsize = 12cm
\epsfbox{fig4.eps}}
\caption{Predicted redshift distributions for both ``Scalo'' ($\tau
= 1.0$~Gyr) and ``x=3'' ($\tau=2.5$~Gyr) models (dotted lines and
solid lines) compared with the results of the $K-$ (lower panel)
and $B-$ (upper panel) redshift distributions taken from Table 1
of Cowie et al. 1996. Incompleteness is represented by the open
boxes. Also plotted are the predicted redshift distributions for
elliptical types (long dashed and short dashed lines) for the
``Scalo'' and ``x=3'' models respectively.}
\label{fig:redshift.dndz}
\end{center}
\end{figure}
\begin{figure}
\begin{center}
\centerline{\epsfxsize = 9cm
\epsfbox{fig5.eps}}
\caption{The median redshift, $z_{med}$ of elliptical galaxies in our model
populations for the magnitude slices used in
Figure~\ref{fig:BKhists} and Figure~\ref{fig:IKhists}.These
galaxies have star-formation histories.}
\label{fig:medzed}
\end{center}
\end{figure}
In Figure~\ref{fig:redshift.dndz} the effect which changing the IMF
slope for early types has on predicted redshift distributions is
illustrated. The dotted line shows a model computed using $\tau =
1.0$~Gyr and a Scalo \cite{1986FCPh...11....1S} IMF, whereas the solid
line shows the $x=3$ model which has $\tau=2.5$~Gyr. Also plotted are
the redshift distributions taken from Table 1 of \citeN{CSH2}. Objects
which were marked as unidentified in each band in this table are shown
in their respective incompleteness boxes. We have normalised both model
predictions to the total number of {\it observed\/} objects (i.e.,
unidentified objects are included). We take $q_0 = 0.05$ and $z_f=6.4$
in all our models. The star-formation e-folding time $\tau$, is
$9.0$~Gyr for all the spiral types. The $x=3$ model, as it dominated
by dwarf stars produces {\it much smaller\/} amounts of passive
evolution in the $K-$ band and as a consequence the median redshift of
galaxies in this model is much lower than in the Scalo model, which
produces an extended tail in the redshift distribution. Notice the
effect of changing the IMF slope in the $B-$ band is negligible.
This point is further illustrated in Figure~\ref{fig:medzed} where we
show the median redshift of elliptical galaxies with a range of
star-formation histories for several $K-$ selected magnitude slices.
This panel shows that, in all all magnitude ranges, elliptical galaxies
in the ``Scalo'' model have a higher median redshift than in the
``x=3'' model, irrespective of $\tau$ (demonstrating that the
differences in \ref{fig:redshift.dndz} are due to changing the IMF and
not $\tau$). This point is discussed further in
Section~\ref{sec:Medi-Colo-Counts}.
\begin{table*}
\begin{center}
\caption {Luminosity function parameters and rest-frame colours used in all models}
\begin{tabular}{cccccc}
Parameter / Type & E/S0 & Sab & Sbc & Scd & Sdm\\\\
$\phi^{*}$ (Mpc$^{-3}\times 10^{-4}$) & $10.2$ & $5.09$ & $6.82$ & $3.0$ & $1.5$\\
$\alpha$ & $-0.7$ & $-0.7$ & $-1.1$ & $-1.1$ & $-1.5$\\
$M^*_B$ & $-21.0$ & $-21.0$ & $-21.32$ & $-21.44$ & $-21.45$\\
$(B-K)$ & $3.93$ & $3.78$ & $3.51$ & $2.90$ & $2.26$\\
$(I-K)$ & 1.57 & 1.63 & 1.7 & 1.47 & 1.0\\
\label{tab:pars}
\end{tabular}
\end{center}
\end{table*}
\begin{figure}
\begin{center}
\centerline{\epsfxsize = 12cm
\epsfbox{fig6.eps}}
\caption{{\bf a:} (lower panel) $(B-K)$ colour against redshift, $z$,
for an elliptical galaxy with an $x=3$ IMF and $\tau=2.5$~Gyr
(solid line), and for the Scalo IMF and with three values of
$\tau$; $\tau=0.5$~Gyr (dot-dashed line), $\tau=1.0$~Gyr (dotted
line) and $\tau=2.5$~Gyr (long dashed line). Also shown is a
non-evolving model (short dashed line). {\bf b:}(upper panel) As
for the lower panel, but with $(I-K)$ colour. }
\label{fig:colzedmed}
\end{center}
\end{figure}
\begin{figure}
\begin{center}
\centerline{\epsfxsize = 12cm
\epsfysize = 12cm
\epsfbox{fig7.eps}}
\caption{Galaxy number counts for two evolutionary
models -- the ``scalo'' model, (solid and dotted lines), and the
``x=3'' model, (short dashed and long-dashed), for low and high
values of $q_0$ respectively. $K-$ band counts are shown in the
upper panel; $B-$ band ones in the lower one.}
\label{fig:slope.removed}
\end{center}
\end{figure}
In Figure~\ref{fig:colzedmed} we illustrate the effect which varying
the IMF slope and $\tau$ has on the E/S0 colour-redshift relation. The
solid, dotted and long-dashed lines shows Scalo IMF tracks with $\tau =
0.5$~Gyr, $\tau=1.0$~Gyr and $\tau = 2.5$~Gyr respectively, whereas the
long dashed lines shows an $x=3$ model with $\tau = 2.5$~Gyr. As is
apparent from the plot, at $z\sim1$ $(B-K)$ colour depends {\it very
sensitively\/} on the assumed value of $\tau$ and the slope of the
IMF\@. For a given IMF, longer $\tau$'s cause the peak in the
$(B-K)$-$z$ relation to shift to progressively higher redshifts, a
point we will return to in Section~\ref{sec:Medi-Colo-Counts}. Finally,
the $x=3$ IMF model track, as it is dominated by low-mass stars, is in
general redder (for a given value of $\tau$) than the Scalo IMF track.
We note that the single-burst models of \citeN{1994ApJS...95..107W}
have shown that the amount of passive evolution in the rest frame $I-$
and $R-$ bands is sensitive to metallicity variations; consequently at
$z=1-2$ the predicted $K-$ magnitude from our essentially single-burst model
is also expected to be approximately independent of metallicity.
Therefore reasonable changes in metallicity seem unlikely to change the
rate of passive evolution at $K-$ which, in the context of these models,
only appears to depend on the slope of the IMF."
We now return to Figure~\ref{fig:Col:Col} where we have over-plotted on
our $(B-R)-(R-K)$ diagram evolutionary tracks for each of the galaxy
types in our model. The dot-dashed and solid lines shows the path of an
elliptical, whereas the remaining tracks show the four spiral types in
the models. On the elliptical tracks, the labelled stars indicate the
redshift at selected intervals. Generally, there is good agreement
between the model tracks and the observed colours. The low redshift
($0.5 < z < 0.0$) elliptical track is populated by the brightest
galaxies in our sample. Generally, we do not find a large fraction of
objects have colours which cannot be reproduced by the tracks (in
contrast with the findings of \citeN{MDS97} who claimed there was a
significant population of objects whose colours could not be reproduced
by the models). There are {\it some\/} objects which appear to be
redder than the tracks, but these do not constitute a significant
fraction of the observed galaxy population. The broad range of $(B-K)$
colours of objects with $(R-K) > 4$ (which according to the models
should all be E/S0 galaxies) is a real effect which we interpret as
evidence of the diverse range of star-formation and metallicities
histories present in the elliptical population.
\subsection{Counts and Colour Distributions}
\label{sec:Medi-Colo-Counts}
In Figure~\ref{fig:slope.removed} we present a compilation of all
published galaxy number counts in $B-$ and $K-$ bandpasses, as well as
the predictions from our two evolutionary models.
What is immediately apparent is that both the models discussed provide
a good fit to the number counts over a wide magnitude range for the low
$q_0$ case. Furthermore, in both $B-$ and $K-$ bandpasses at faint
magnitudes the differences between the Scalo and ``x=3'' models is {\it
much smaller\/} than the differences between the high- and low-$q_0$
models. It is also apparent, as has been noted by many others
\cite{MSCFG,1997ApJ...488..606C,BR} that, unless another population
(such as bursting dwarfs) is added to the models then the $q_0=0.5$
model under-predicts the observed $B-$ band galaxy counts. In the $K-$
band this model is marginally excluded at the faintest magnitudes.
Number counts are a coarse test of any model; the extremely deep
optical photometry covering the Herschel Field is one of main virtues
of this work. So, to test our models more stringently we now
investigate our observed optical-infrared colour distributions. We
confine our discussion to the $q_0=0.05$ case as these models provide
a better fit to the counts, although the colour distribution for the
$q_0=0.5$ models will be very similar.
\begin{figure*}
\centering \centerline{\epsfxsize = 16.0 cm
\epsfbox{fig8.eps}}
\caption{$K-$ selected $(B-K)$ colour distributions (solid histogram)
in six magnitude slices for both the wide (brightest four slices) and
the deep survey (faintest two slices). Also shown are predictions
from four models: `'x=3'' (solid line), ``Scalo'' (dotted and
dot-dashed lines) and ``non-evolving'' (long dashed line). The two
``Scalo'' models have differing star-formation histories; the dotted
and heavy-dashed line has $\tau=1.0$~Gyr whilst the heavy dot-dashed
line has $\tau=0.5$~Gyr. The $x=3$ models has $\tau=2.5$~Gyr. The
model predictions have been normalised to the {\it total\/} number of
galaxies in each slice. In the fainter ($K>18^m$) bins this includes
the $B-$ band non-detections discussed in Section~\ref{sec:ExtRed}
and the number of such objects in each magnitude slice is represented
by the incompleteness boxes. The position of the downward-pointing
arrow represents the colour of the {\it reddest\/} object at the
centre of each magnitude slice which could be detected; objects in
the incompleteness boxes must lie rightwards of this arrow. For all
models in this plot we adopt $q_0=0.05$. }
\label{fig:BKhists}
\end{figure*}
In Figure~\ref{fig:BKhists} we compare in detail our model predictions
with our observations for $K-$ selected $(B-K)$ colour distributions
and compare these with the data (shown as the solid histograms) in six
one-magnitude slices from $K=16^m$ to $K=22^m$. The ``Scalo'' model is
shown as a dotted and a dot-dashed line and the ``x=3'' model is
represented by a solid line. All models have been convolved with a
conservative $0.3^m$ colour measurement error (overestimating errors in
the brighter magnitude slices). For the non-evolving model, as for the
evolving models, $k-$ corrections are calculated from the models of
\citeN{BC}.
\begin{figure*}
\centering
\centerline{\epsfxsize = 16.0 cm
\epsfbox{fig9.eps}}
\caption{$K-$ selected $(I-K)$ colour distributions for
both the wide and deep surveys (solid histogram). All symbols are as
in Figure~\ref{fig:BKhists}. }
\label{fig:IKhists}
\end{figure*}
Incompleteness in the histograms was determined by counting how many
galaxies had $K-$ magnitudes but were not detected in $B-$ in each
magnitude slice (and consequently have undetermined colours). The
colour of the {\it reddest\/} object which could be detected is
represented by the downwards-pointing arrow, which is plotted at the
colour corresponding to the $K-$ magnitude at the centre of each
magnitude slice (these objects have $B>28^m$). The number of
non-detections is shown by the size of the box. Models have been
normalised to the total number of galaxies in each slice; in the latter
panels this includes $B-$ band non-detections.
The non-evolving model predictions (dashed line) rapidly diverge from
the data. Faintwards of the $K=17^m-18^m$ magnitude slice a bump
appears in the model colour distributions, corresponding to the
unevolved, red, $(B-K)\sim8$ elliptical population. At approximately
the same magnitude limit and at bluer colours (this time at $(B-K) \sim
4$) a second peak becomes apparent. This corresponds to the model
spiral population, and at fainter magnitudes this peak becomes the more
prominent of the two. This is a consequence of our adopted luminosity
function parameters and morphological mix.
Qualitatively, the evolutionary models follow the pattern of the
non-evolving model, though the location and amplitude of the E/S0 peaks
depend sensitively on the choice of IMF and $\tau$. On the blueward
side of the distributions, the spiral model colours appear to be a good
fit to the observations. Referring to Figure \ref{fig:colzedmed} it is
apparent that beyond $z\sim0.3$, the $(B-K)$ colour for the
$\tau=0.5$~Gyr E/S0 model is redder than the $\tau=1.0$~Gyr model;
furthermore, as the median redshift of both populations is almost the
same (Figure~\ref{fig:medzed}), changes in the colour distributions are
almost completely a product of differences in the $(B-K)$-$z$ relation.
In general, \citeN{BC}-type models for elliptical galaxies with short
$\tau$'s predict a $(B-K)-z$ relation in which galaxies are at their
reddest at $z\sim 1$, and become bluer thereafter until $z\sim3$ at
which point they turn redder once more as a result of Lyman-$\alpha$
forest reddening. The position of this turn-over depends on the
assumed $\tau$, the IMF and the redshift of formation $z_f$.
If we now we consider the ``x=3'', $\tau=2.5$~Gyr model, we
see that its colour distributions are in general {\it bluer\/} than
those predicted from the ``Scalo'' $\tau = 0.5$ or $\tau = 1.0$~Gyr
models. This is partly a consequence of the longer $\tau$ adopted in this
model, but also of the fact that (as we have seen in
Figure~\ref{fig:medzed}) galaxies lie at a lower redshift than the
``Scalo'' model and consequently are sampling a bluer region of the
$(B-K)$-$z$ relation. So, despite the $x=3$ model being low-mass star
dominated, the larger value of $\tau$ allows us to produce colours
which are blue enough to match the observations.
Returning to Figure \ref{fig:BKhists} we can now understand the origin
of the differences between the $(B-K)$ colour distributions predicted
by the models. Considering first the brighter magnitude slices in
Figure~\ref{fig:BKhists}, at $K=17.0^m-18.0^m$ we see that the
``Scalo'' $\tau=1.0$~Gyr model predicts a peak at $(B-K) \sim 7$ which
appears in the ``x=3'' model at $(B-K)\sim6$. At this magnitude,
elliptical galaxies in the ``Scalo'' model have $z_{med}\sim0.6$,
whereas for the ``x=3'' model $z_{med}\sim~0.4$ (from
Figure~\ref{fig:medzed}); as a consequence ``Scalo'' model colours are
redder than ``x=3'' ones. Furthermore, the essentially
passively-evolving Scalo $\tau=0.5$~Gyr model contains insufficient
recent star-formation to shift the peak at $(B-K)\sim9$ blueward; as we
have seen from our considerations of the non-evolving model, this peak
is from unevolved ellipticals. At fainter ($K>20^m$) magnitudes, both
the $x=3$ model and the Scalo $\tau =1.0$~Gyr model turn sharply
blueward; the Scalo $\tau=0.5$~Gyr model colour distribution, however,
has a extended tail which reaches $(B-K)\sim 9-10$.
It is clear from Figure \ref{fig:BKhists} that either a Scalo IMF with
a $\tau = 1.0$~Gyr or an $x=3$ IMF with $\tau=2.5$~Gyr model
reproduces the main features of the data histograms, including the
broadening of the colour distribution at the $K<18^m$ slice, and its
subsequent narrowing and blueward trend in the fainter slices.
Furthermore, as we will see in the following Section, these models are
consistent with the redwards limits of our data. Of these two models,
the $x=3$, $\tau=2.5$~Gyr model as commented in
Section~\ref{sec:Outline-Models}, gives a much better fit to the
observed $K-$ band redshift distributions. As is apparent from
Figures~\ref{fig:medzed} and~\ref{fig:colzedmed} distinguishing between
the effects of changes in IMF slope or the star-formation history on
the colour-redshift relation is difficult. Making this choice requires
additional information such as the redshift distribution of $K-$
selected galaxies.
Finally, these models also allow us to understand the deficit of
galaxies in the colour-magnitude relation at $(B-K)>5$ and $K>21^m$
which we commented upon in Section~\ref{sec:Optic-Infr-Colo}. Firstly,
as a consequence of our choice of luminosity function parameters
elliptical counts turn over at $K\sim20$ and spiral types dominate
faintwards of this. Secondly, from Figure \ref{fig:medzed} we see that
at $K\sim21$ E/S0's are at $z\sim1$; from these redshifts to $z>3$
Figure \ref{fig:colzedmed} indicates that elliptical colours turn
sharply blueward. In combination, these factors produce the rapid
blueward trend observed at $K>21^m$ and the absence of redder objects,
which happen in all our models regardless of our choice of IMF or
$\tau$.
\subsection{$(I-K)$ distributions and extremely red galaxies}
\label{sec:ExtRed}
In the previous section we discussed the broad characteristics of the
$K-$ selected $(B-K)$ colour distribution and showed how these features
can be understood in terms of our models; in this section we turn our
attention to very red objects which are at the outskirts of the colour
distributions. The red colours of these objects imply they are E/S0
galaxies at moderate ($z>1$) redshift and consequently it is
interesting to see if their observed number density is consistent with
current galaxy formation scenarios. As previous authors
\cite{AB,MDS97,1997Natur.390..377Z,1994ApJ...434..114C} who focussed
their attentions upon these galaxies discussed them in terms of $K-$
selected $(I-K)$ distributions, here we introduce our observations in
these bandpasses. Following these workers we define an ``extremely red
galaxy'' as an object with $(I-K)>4$; from Figure~\ref{fig:colzedmed}
we see that at $(I-K)\sim4$, $z\sim1$. In this section we will make
comparisons with models generated using the same set of parameters as
the $K-$ selected $(B-K)$ distributions discussed previously. In
interpreting $(I-K)$ colours one faces some significant differences to
$(B-K)$ observations; as is apparent from Figure \ref{fig:colzedmed},
$(I-K)$ colour is relatively insensitive to the amount of
star-formation; at $z\sim1$ the difference in $(I-K)$ between
elliptical galaxies with a Scalo IMF and $\tau=0.5,1,2.5$~Gyr is $0.5$;
by comparison, at this redshift $(B-K)$ ranges from $\sim3$ to $\sim5$.
This difference is reflected in the tightness of the $K-$ selected
$(I-K)$ distributions plotted in Figure \ref{fig:IKhists}. This
insensitivity makes $(I-K)$ distributions relatively poor probes of the
elliptical star-formation history. However, it is precisely this
behaviour which makes $(I-K)$ colours useful in detecting
high-redshift, evolved ellipticals; Figure~\ref{fig:colzedmed} shows
that until $z\sim2$ $(I-K)$ colour is approximately proportional to
$z$, and is relatively insensitive to variations in $\tau$ or the IMF.
Turning to work from the literature, \citeN{1994ApJ...434..114C}, in a
$5.9$~arcmin$^2$ survey, detected $13$ objects with $(I-K)>4$ to
$K<20.9^m$, and concluded that these galaxies do not dominate the faint
$K>20^m$ population. They also concluded that less that $10\%$ of
present day ellipticals could have formed in single-burst events. In
contrast, \citeN{MDS97} found 8 galaxies with $(I-K)>4$ and $K<22^m$
over a small, $2$~arcmin$^2$ area. They also isolated a population of
objects with blue optical ($(V-I)<2.5$) and red near-infrared colours
and argued that these colours could not be reproduced with
\citeN{BC}-type models. Although we do not have $V-$ band photometry
the top middle and top left of our Figure~\ref{fig:Col:Col} roughly
corresponds to the regions highlighted in the colour-colour plot shown
in Figure 9 of \citeANP{MDS97}. These ``red outlier'' objects, as
\citeANP{MDS97} describe them, viewed in the context of the models
discussed here are most likely elliptical galaxies at $1<z<2$. Finally,
both \citeN{1997Natur.390..377Z} and \citeN{AB} investigated extremely
red objects in the Hubble Deep Field (HDF).
\citeN{1997Natur.390..377Z}, using publicly-available Kitt Peak
near-infrared imaging data covering the $5$~arcmin$^2$ of the HDF found
that to a $50\%$ completeness limit of $K<22^m$ there were $\sim 2$
objects with $(V_{606}-K) >7$. \citeN{AB}, in a much larger $\sim
60$~arcmin$^2$ infrared survey comprising both the HDF and the HDF
flanking fields found 12 galaxies with $(I-K)>4$ and $19<K<20$.
\begin{table*}
\caption{ Numbers of objects with $(I-K)>4~$per~arcmin$^{2}$ with $\pm
1\sigma$ errors. The survey area in arcmin$^{2}$ is shown in
parentheses. In the deep survey ($20<K<22$) there are 7 objects
which are undetected in $I$ and therefore only lower limits can be
placed on their colours. They are included in their respective bins.
The number quoted from Moustakas et al.~(1997) covers the magnitude
range $20<K<22$.}
\begin{center}
\begin{tabular}{lrlrlrl}
Author&\multicolumn{2}{c}{$19<K<20$}&\multicolumn{2}{c}{$20<K<21$}&
\multicolumn{2}{c}{$21<K<22$}\vspace{1mm}\\
Model ($x=3$)&\multicolumn{2}{c}{$0.3$}&\multicolumn{2}{c}{$1.4$}&
\multicolumn{2}{c}{$1.8$}\vspace{1mm}\\
Model (scalo, $\tau=1.0$~Gyr)&\multicolumn{2}{c}{$0.6$}&
\multicolumn{2}{c}{$1.4$}&\multicolumn{2}{c}{$1.2$}\vspace{1mm}\\
Model (scalo, $\tau=0.5$~Gyr)&\multicolumn{2}{c}{$1.1$}&
\multicolumn{2}{c}{$3.1$}&\multicolumn{2}{c}{$2.8$}\vspace{1mm}\\
This work&$0.5^{+0.1}_{-0.1}$&(47)&$3.9^{+2.1}_{-1.4}$&(1.8)&
$1.1^{+1.5}_{-0.7}$&(1.8)\vspace{1mm}\\
\citeN{AB}&$0.2^{+0.1}_{-0.1}$&(62)&$1.2^{+0.5}_{-0.4}$&(7.8)&
\multicolumn{2}{c}{$-$}\vspace{1mm}\\
\citeN{1994ApJ...434..114C}&$0.3^{+0.4}_{-0.3}$&(7.1)&$1.5^{+0.7}_{-0.5}$&(5.9)&
$0.3^{+0.5}_{-0.2}$&(5.9)\vspace{1mm}\\
\citeN{MDS97}&\multicolumn{2}{c}{$-$}&&\multicolumn{1}{r}
{$4.0^{+2.0}_{-1.4}$}&\multicolumn{1}{l}{(2)}&\\
\end{tabular}
\label{tab:ext-col-obj}
\end{center}
\end{table*}
Table \ref{tab:ext-col-obj} presents a compilation of observations from
these papers as well as our current work. We also show the predictions
of the $x=3$ and Scalo $\tau=0.5$ model (our model which most closely
approximates a single burst), normalised to the total numbers of
objects each magnitude slice. Generally, in the brighter bin
$(19<K<20)$ the only substantial comparison that we can make is with
\citeANP{AB}. We see a factor $\sim 3$ more objects than they do,
although this represents only a $2\sigma$ discrepancy. In the fainter
bin $20<K<22$ our numbers of red objects agree well with
\citeANP{MDS97}. Although our numbers are higher than
\citeN{1994ApJ...434..114C} in this range there is still no significant
discrepancy due principally to our small area.
\citeANP{AB} compared their observations to the predictions of a
passively-evolving $\tau=0.1$~Gyr model elliptical population, and
concluded that the numbers of $(I-K)>4$ objects observed disagreed with
the predictions of this model. Our $\tau=0.5$~Gyr distribution (solid
dot-dashed line) in Figure~\ref{fig:IKhists} is quite similar to this
model, and in the $19<K<20$ magnitude slice we can see from
Table~\ref{tab:ext-col-obj} that it does indeed produce a large number
of objects with $(I-K)>4$ which are not seen in our observations. We
therefore conclude, as \citeANP{AB} did, that the low numbers of
objects with $(I-K)>4$ observed at $19<K<20$ disfavour non-evolving and
passively evolving models (the constraints on the models which can be
obtained from the distributions at fainter magnitudes are less
significant). However, we note that the $x=3$, $\tau=2.5$~Gyr and
$\tau=1.0$~Gyr Scalo models do correctly reproduce the observed colour
distributions. This is because both these models contain enough
on-going star-formation to move galaxy colours sufficiently blueward to
match the observations.
Lastly, could a merging model, like those adopted by
\citeN{1998MNRAS.297L..23K}, produce these results? It is not possible
to rule this model out from the redshift data and it may even be said
that the difficulty that models with standard Scalo IMF's have in
fitting the $K<19$ redshift distribution is evidence in favour of the
merging scenario. We therefore consider whether the $K-$ selected
$(B-K)$ and $(I-K)$ colour distributions presented here can
discriminate between a PLE $(x=3)$ model and a merging model. The $K-$
selected $(B-K)$ distributions in Figure~\ref{fig:BKhists} can be
understood in terms of either scenario; the red galaxies that are
missing at $17<K<22$ may either be due to them becoming blue at high
redshift due to continuing star-formation or passive evolution -- or
because they are fainter than expected due to de-merging (although our
lack of red galaxies even at the faintest limits implies either several
magnitudes of fading or that the pre-merger components are blue).
However, as we have already noted the $(I-K)$-$z$ relation for E/S0
galaxies is insensitive to changes in $\tau$ or the IMF. Furthermore,
in the range $18<K<20$ some galaxies with the $I-K>4$ colour expected
of $z=1$ elliptical galaxies are detected (Figure \ref{fig:IKhists}).
Indeed the numbers of these galaxies we see is within a factor of $\sim
2$ of what is predicted on the basis of the Scalo, $\tau=0.5$~Gyr model
and in even better agreement with the $x=3$, $\tau=2.5$~Gyr models or
even a Scalo $\tau=1.0$~Gyr model. If the merging model is the
explanation of the large deficiency of red galaxies in $(B-K)$ compared
to passively evolving models, then it might be expected that a similar
deficiency should be seen in $(I-K)$. Since the deficiency in $(I-K)$
is less, then this might be taken to be an argument {\it against}
merging and {\it for} the $x=3$ PLE model.
\section{Conclusions and Summary}
\label{sec:IntDisc}
In this paper we have presented the results of two near-infrared surveys to
$K\sim20$ and $K\sim23$ which cover our ultra-deep ($B\sim28$) optical
fields. We draw the following conclusions from this work:
\begin{description}
\item Our K number counts are consistent with the predictions of
non-evolving models with $0\leq q_0 \leq 0.5$.
\item As previously noted by \citeN{CSH2},\citeN{MSCFG}, and
\citeN{1998MNRAS.297L..23K} the $18<K<19$ $n(z)$ of \citeN{CSH2} is
also well fitted by non-evolving models. However, passively evolving
models with a Salpeter/Scalo IMF predict too many galaxies with
$z>1$. Dynamical merging is one possible solution to reduce the
numbers of these galaxies but we have shown that a dwarf-dominated
IMF for early-types could offer an alternative explanation.
\item Our $K-$ selected $(B-K)$ colour distributions display a strong
bluewards trend for galaxies fainter than $K\sim20$, confirming
results previously observed in the shallower surveys of \citeN{GCW}.
\item At brighter magnitudes ($K<20$~mag) our $K-$ selected $(B-K)$
distributions indicate a deficiency of red, early-type galaxies at
$z\sim1$ compared to the predictions of passively evolving models.
This implies either a PLE model where star-formation continues at a
low level after an initial burst or dynamical merging.
\item At fainter magnitudes ($20<K<22$) the continuing bluewards
trend observed in $(B-K)$ can be explained purely in terms of
passively evolving PLE models with no need to invoke any additional
mechanisms.
\item Our observed numbers of $(I-K)>4$ galaxies at $K\sim20$ are lower
than the predictions of passively evolving models or PLE models with
a low level of continuing star-formation, suggesting that at least
part of the larger deficiency observed in $(B-K)$ at $K\sim20$ may be
due to star-formation rather than dynamical merging.
\item In the range $19<K<20$, where our statistical uncertainties are
lowest, we detect $0.5\pm 0.1$ red galaxies arcmin$^{-2}$ with
$(I-K)>4$. We see a factor $\sim 3$ more objects than \citeANP{AB}
do although this represents only a $2\sigma$ discrepancy. The PLE
models discussed here suggest these galaxies will have redshifts
$1<z<2$. The numbers of these galaxies are consistent with
the predictions of PLE models with small amounts of ongoing
star-formation.
\end{description}
\section{Future Work}
\label{sec:Future-Work}
We have recently completed a $H-$ band survey of the Herschel field
using the $1024\times1024$ Rockwell HgCdTe array at the Calar Alto 3.5M
telescope. This survey is complete to $H<22.6^m$ and covers the entire
area of the Herschel Deep Field. Analysis of this dataset will be
presented in future papers.
\section{Acknowledgements}
\label{sec:Acknowledgements}
H.J.McCracken wishes to acknowledge financial support from PPARC and
the hospitality and generosity of Dr. Mariano Moles at IMAFF, Madrid
where an earlier version of this paper was written. Support for this
trip was also provided the University of Durham Rolling Grant for
Extragalactic Astronomy. This work was supported in part by a British
Council grant within the British/Spanish Joint Research Programme
(Acciones Integradas). NM acknowledges partial PPARC
support.
|
Subsets and Splits