Spectral Curves and Whitham Equations in Isomonodromic Problems of Schlesinger Type
Extracted source text ยท arXiv solv-int/9704004
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Spectral Curves and Whitham Equations in %%
%% Isomonodromic Problems of Schlesinger Type %%
%% By Kanehisa Takasaki, KUCP-0105 %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% global parameters %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%
\documentstyle[12pt]{article}
\renewcommand{\baselinestretch}{1.3}
\setlength{\oddsidemargin}{0cm}
\setlength{\evensidemargin}{0cm}
\setlength{\topmargin}{0cm}
\setlength{\textheight}{22.5cm}
\setlength{\textwidth}{16cm}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% modifying LaTeX counter %%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\catcode`\@=11
\def\theequation{\thesection.\arabic{equation}}
\@addtoreset{equation}{section}
\@addtoreset{footnote}{section}
%%% \@addtoreset{footnote}{subsection}
\catcode`\@=12
%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% my macros %%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newtheorem{theorem}{Theorem}
\newtheorem{proposition}{Proposition}
\newtheorem{corollary}{Corollary}
%
\newcommand{\rd}{\partial}
\newcommand{\dfrac}[2]{ \frac{\displaystyle #1}{\displaystyle #2} }
\newcommand{\binom}[2]{ {#1 \choose #2} }
\newcommand{\Res}{\mathop{\mbox{\rm Res}}}
\newcommand{\diag}{\mathop{\mbox{\rm diag}}}
\newcommand{\Tr}{\mathop{\mbox{\rm Tr}}}
\newcommand{\const}{\mbox{\rm const.}\;}
%
\newcommand{\gl}{\mbox{gl}}
\newcommand{\GL}{\mbox{GL}}
%
\newcommand{\bfC}{{\bf C}}
\newcommand{\bfZ}{{\bf Z}}
\newcommand{\calA}{{\cal A}}
\newcommand{\calB}{{\cal B}}
\newcommand{\calF}{{\cal F}}
\newcommand{\calO}{{\cal O}}
\newcommand{\calT}{{\cal T}}
\newcommand{\Yhat}{\hat{Y}}
\newcommand{\mutilde}{\tilde{\mu}}
\newcommand{\omegatilde}{\tilde{\omega}}
\newcommand{\Omegatilde}{\tilde{\Omega}}
\newcommand{\Ctilde}{\tilde{C}}
\newcommand{\Ftilde}{\tilde{F}}
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\title{Spectral Curves and Whitham Equations in
Isomonodromic Problems of Schlesinger Type}
%
\author{Kanehisa Takasaki
\thanks{Partly supported by the Grant-in-Aid for Scientific
Researches, Priority Area 231 ``Infinite Analysis'',
the Ministry of Education, Science and Culture, Japan}
\\
{\normalsize Department of Fundamental Sciences, Kyoto University}\\
{\normalsize Yoshida, Sakyo-ku, Kyoto 606, Japan}\\
{\normalsize E-mail: \tt takasaki@yukawa.kyoto-u.ac.jp}
}
%
\date{}
%
\maketitle
\begin{abstract}
It has been known since the beginning of this century that
isomonodromic problems --- typically the Painlev\'e
transcendents --- in a suitable asymptotic region look like
a kind of ``modulation'' of isospectral problem. This
connection between isomonodromic and isospectral problems
is reconsidered here in the light of recent studies related
to the Seiberg-Witten solutions of $N = 2$ supersymmetric
gauge theories. A general machinary is illustrated in a
typical isomonodromic problem, namely the Schlesinger equation,
which is reformulated to include a small parameter $\epsilon$.
In the small-$\epsilon$ limit, solutions of this isomonodromic
problem are expected to behave as a slowly modulated
finite-gap solution of an isospectral problem. The modulation
is caused by slow deformations of the spectral curve of the
finite-gap solution. A modulation equation of this slow
dynamics is derived by a heuristic method. An inverse
period map of Seiberg-Witten type turns out to give general
solutions of this modulation equation. This construction of
general solution also reveals the existence of deformations
of Seiberg-Witten type on the same moduli space of spectral
curves. A prepotential is also constructed in the same way as
the prepotential of the Seiberg-Witten theory.
\end{abstract}
\bigskip
\begin{flushleft}
KUCP-0105\\
solv-int/9704004
\end{flushleft}
\newpage
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Introduction}
The notion of ``isomonodromic deformations'' was first
discovered by R. Fuchs \cite{bib:RFuchs} in 1907 as
a new interpretation of the 6th Painlev\'e equation
(P$_{\rm VI}$), and developed in diverse directions
in the next decade. R. Fuchs considered a second order
scalar ODE
\[
\frac{d^2 y}{d\lambda^2} + p(\lambda)y = 0
\]
of Fuchsian type with four regular singularities,
one of which is an apparent singularity. Garnier
\cite{bib:Garnier1912}
generalized the work of R. Fuchs in two different
forms. One generalization is to consider more than
four regular singularities. This leasd to a
multi-variable generalization of P$_{\rm VI}$.
The other is to include irregular singularities.
The other five Painlev\'e equations
(P$_{\rm I}$ -- P$_{\rm V}$) can be derived
from this generalization. Schlesinger
\cite{bib:Schlesinger} obtained the so called
Schlesinger equation from isomonodromic deformations
of first order matrix ODE
\[
\frac{dY}{d\lambda} = M(\lambda) Y,
\]
where $M(\lambda)$ is an $r \times r$ matrix of the form
\[
M(\lambda) = \sum_{i=1}^N \dfrac{A_i}{\lambda - t_i}.
\]
Garnier noticed later \cite{bib:Garnier1926} that
the $2 \times 2$ Schlesinger equation is equivalent
to his isomonodromic deformations of 2nd order scalar
Fuchsian equation.
It is also in this decade that a link with
``isospectral deformations'' was uncovered.
This is again due to Garnier \cite{bib:Garnier1917}.
He proposed an autonomous analogue of the Schlesinger
equations, and pointed out that it can be integrated by
Abelian functions. Remarkably, Garnier substantially
arrived at the notion of isospectral deformations
therein. Let us briefly review Garnier's discovery.
Schlesinger's equation can be written
\[
\dfrac{\rd A_i}{\rd t_j}
= (1 - \delta_{ij}) \dfrac{[A_i,A_j]}{t_i - t_j}
- \delta_{ij} \sum_{k (\not= i)} \dfrac{[A_i,A_k]}{t_i - t_k}.
\]
This is a non-autonomous system, because the right hand side
contains the independent variables $t = \{t_i\}$ explicitly.
Garnier's proposal was to replace these $t_i$'s by constants,
$t_i \to c_i$. The outcome is an autonomous system of the form
\[
\dfrac{\rd A_i}{\rd t_j}
= (1 - \delta_{ij}) \dfrac{[A_i,A_j]}{c_i - c_j}
- \delta_{ij} \sum_{k (\not= i)} \dfrac{[A_i,A_k]}{c_i - c_k}.
\]
As Garnier noticed, this gives isospectral deformations of
$M(\lambda)$, namely, the characteristic polynomial
$\det\Bigl( M(\lambda) - \mu I\Bigr)$ is invariant under
the $t$ flows. The algebraic curve defined by the
characteristic equation
\[
\det\Bigl( M(\lambda) - \mu I \Bigr) = 0
\]
in the $(\lambda,\mu)$ plane (and its appropriate
compactification) is nowadays called the ``spectral curve.''
What Garnier did is to solve the above autonomous system
in terms of Abelian functions on this algebraic curve.
Isospectral problems of the same type were studied
by Moser \cite{bib:Moser} in the 70's. Moser proposed
a matrix nonlinear system as a unified framework for
a number of classically well known completely integrable
dynamicals systems, such as the Neumann and Rosochatius
systems, geodesic flows on an ellipsoid, etc. Moser's
idea and Munford's related work \cite{bib:Munford}
were reformulated by the Montreal group (Adams, Harnad,
Hurtubise and Previato) \cite{bib:AHHP,bib:AHH}
and Beauville \cite{bib:Beauville} to the isospectral
problem of a rational matrix of the form $M(\lambda)$ (or
$U + M(\lambda)$, where $U$ is a constant matrix).
Harnad and his collaborators later applied their method
to isomonodromic problems
\cite{bib:Harnad-dual,bib:HTW,bib:HR,bib:HW}.
Garnier's proposal, however, originally aimed at
a quite different issue. He considered the autonomous
system as a tool for studying asymptotic behavior of
solutions of the Schlesinger equation in a neighborhood
of singularities. A similar problem concerning the
Painlev\'e equations had been pursued by Boutroux
\cite{bib:Boutroux}. Boutroux obtained an asymptotic
expression of Painlev\'e transcendents as a ``modulated''
elliptic function. Here ``modulation'' means that
parameters of the elliptic function also depend
(but ``slowly'') on the independent variable.
In Garnier's program, Boutroux's elliptic curve
is replaced by a more general algebraic curve.
Flaschka and Newell \cite{bib:FN}
revisited Garnier's program in their study of
isomonodromic and isospectral problems, and noted
an important remark: They pointed out that a JWKB
approximation converts the monodromy problem into
a spectral problem.
We reconsider this issue in the light of researches
in the last ten years. Since the end of the 80's,
low dimensional string and topological field theories
have provided new subjects of isomonodromic problems.
A central subject is the so called ``string quations''
of two-dimensional quantum gravity \cite{bib:Moore-review}.
They are the first (or second) Painlev\'e equation and
its higher dimensional generalizations. These string
equations have been studied from several different
points of view. Among them, we are particularly interested
in the approach \cite{bib:Novikov,bib:Krichever-ETH,
bib:Fucito-etal,bib:Vereschagin-string}
from the ``Whitham averaging method'' \cite{bib:Whitham}
(also referred to as the ``nonlinear JWKB method''
\cite{bib:D-Maslov}). This method may be viewed as
a modernized version of Boutroux's analysis. We now
consider the Schlesinger equation in the same philosophy.
The Whitham averaging method, or the nonlinear JWKB method,
covers a wide area of modulational phenomena in nonlinear
waves. The most relevant for our problem is the case of
modulation of Abelian function solutions (which are usually
called ``finite-gap'' or ``quasi-periodic'' solutions
\cite{bib:DMN,bib:Krichever1977}) to a soliton equation.
An ``unmodulated'' finite-band solution generally takes
the form
\[
u_0 = u_0 \Bigl( \sum_i U_i t_i \mid \{ I_n \} \Bigr),
\]
where $u_0$ is an Abelian function, $U_i$'s are $g$-dimensional
constant vectors ($g$ is the genus of the spectral curve),
and $I_n$'s are other parameters of the solution.
The parameters $U_i$'s and $I_n$'s are eventually determined
by the spectral curve (as period integrals of meromorphic
differentials). If the problem in question contains a small
parameter $\epsilon$, one may consider a solution with the
following asymptotic form
\[
u \sim u_0 \Bigl( \sum_i U_i(T) t_i \mid \{I_n\} \Bigr)
\]
as $\epsilon \to \infty$. The parameter $U_i$ and $I_n$ now
depend on the ``slow variables''
\[
T = \{ T_i \}, \quad T_i = \epsilon t_i.
\]
This is the ``modulation'' of a finite-gap solution.
The Whitham averaging method is a method to determine this
slow dynamics in $T_i$ in the form of differential equations.
This kind of differential equations are generally called
``modulation equations'' (or ``Whitham equations'').
In the case of finite-gap solutions, the modulation
equation can be formulated as a dynamical system on
the moduli space of spectral curves.
These modulation equations of finite-gap solutions
are known to possess a number of remarkable properties
\cite{bib:DN-review}. It is Flaschka, Forest and McLaughlin
\cite{bib:FFM} who first pointed out that this type of
modulation equations have a universal structure. They
demonstrated, in the case of the KdV equation, that the
modulation equation boils down to the universal form
\[
\frac{\rd}{\rd T_i} d\Omega_j
= \frac{\rd}{\rd T_j} d\Omega_i,
\]
where $d\Omega_i$'s are meromorphic differentials on
the spectral curve. Krichever \cite{bib:Krichever-Whitham}
and Dubrovin \cite{bib:Dubrovin-Whitham} presented
an abstract reformulation of this type of equations
(``Whitham hierarchies''), and constructed many special
solutions (with applications to geometry and physics).
Our concern lies in special solutions of the universal
Whitham hierarchy that represent slow dynamics of a
spectral curve in isomonodromic problems. Remarkably,
it seems likely that this class of solutions of the
universal Whitham hierarchy are always characterized
by a differential equation of the form
\[
\frac{\rd}{\rd T_i}dS = d\Omega_i, \quad
dS = \mu d\lambda.
\]
This is indeed the case for the string equations
\cite{bib:Novikov,bib:Krichever-ETH,
bib:Fucito-etal,bib:Vereschagin-string}.
In this paper, we shall derive a modulation equation
of this form from the Schlesinger equation.
We use a very heuristic method to derive the modulation
equation. This heuristic method was developed in an
attempt \cite{bib:TN} at an isomonodromic interpretation
of integrable structures in supersymmetric gauge theories
\cite{bib:SW-integrable}.
The modulation equation turns out to possess almost the
same properties as the so called ``Seiberg-Witten solutions''
of $N=2$ supersymmetric gauge theories. In particular, we
introduce a period map of Seiberg-Witten map, and prove that
the inverse period map solves the modulation equation. This
also reveals the existence of another set of commuting flows
on the moduli space of spectral curves. We also show that
the notion of prepotential can be generalized to this case.
These results will be strong evidence for the validity of
the heuristic derivation.
This paper is organized as follows. Sections 2 and 3 are of
preliminary nature. In Section 2, we review basic properties
of the Schlesinger equation. In Section 3, we consider the
geometric structure of spectral curves along the lines of
approach by the Montreal group and Beauville. Section 4, 5,
and 6 are focussed on the derivation of the modulation equation.
We begin with a reformulation of the Schlesinger equation
in Section 4. The reformulated Schlesinger equation has a
small parameter $\epsilon$. Garnier's autonomous system
emerges in the limit of $\epsilon \to 0$. Our modulation
equation is derived in Section 5, along with comments on other
possible approaches. The structure of meromorphic differentials
$d\Omega_i$, which are also basic constituents of our
modulation equation, is specified in Section 6. Section 7
is devoted to solving the modulation equation by the inverse
period map. Section 8 deals with the notion of prepotential.
We conclude this paper in Section 9.
\section{Schlesinger Equation}
In this section, we review basic properties of the Schlesinger
equation. For details and related topics, we refer to a series
of papers by the Kyoto school (Jimbo, Miwa, M\^ori, Sato and Ueno)
\cite{bib:JMMS,bib:JMU,bib:JM}.
\subsection{Coadjoint Orbit and Hamiltonian Structure}
Let $\gl(r,\bfC)^N$ denote a direct sum of $N$ copies of
$\gl(r,\bfC)$. This is the space of $N$-tuples
$(A_1,\cdots,A_N)$ of $r \times r$ matrices. $\GL(r,\bfC)$
acts on this space by the diagonal coadjoint action:
$A_i \mapsto g A_i g^{-1}$. The Schlesinger equation
can be written
\begin{equation}
\dfrac{\rd A_i}{\rd t_j}
= \left[ A_i,
(1 - \delta_{ij}) \dfrac{A_j}{t_i - t_j}
- \delta_{ij} \sum_{k(\not= i)}
\dfrac{A_k}{t_i - t_k} \right],
\end{equation}
each coadjoint orbit $\calO_i$ is left invariant under the
$t$-flows. Thus the Schlesinger equation is actually a
collection of non-autonomous dynamical systems on a
direct product $\calO_1 \times \cdots \times \calO_N$
of coadjoint orbits in $\gl(r,\bfC)$. Usually, only semi-simple
orbits are considered; such an orbit is labeled by the
eigenvalues $\theta_{i\alpha}$ ($\alpha = 1, \cdots, r$)
of $A_i$. In other words, these eigenvalues (and, in general,
the Jordan canonical form of $A_i$'s) are invariants of the
Schlesinger equation.
Actually, there are some extra invariants. They are the
matrix elements of
\begin{equation}
A_\infty = - \sum_{i=1}^N A_i,
\end{equation}
which are invariant under the Schlesinger equation:
\begin{equation}
\dfrac{\rd A_\infty}{\rd t_i} = 0.
\end{equation}
This matrix, too, is usually assumed to be semi-simple,
and it is customary to diagonalize this matrix in
advance by a constant ``gauge transformation''
$A_i \mapsto C A_i C^{-1}$. Thus only the eigenvalues
$\theta_{\infty \alpha}$ ($\alpha = 1,\cdots,r$) of
$A_\infty$ are nontrivial invariants.
Geometrically, this gauge-fixing may be interpreted as
the Marsden-Weinstein construction of a ``reduced phase
space.'' The ``unreduced phase space'' is a coadjoint orbit
$\calO_1 \times \cdots \calO_N \times \calO_\infty$ in
the vector space $\gl(r,\bfC)^{N+1}$ of $(N+1)$-tuples
$(A_1,\cdots,A_N,A_\infty)$. In order to reproduce the
Schlesinger equation, one has to impose the linear constraint
\begin{equation}
\sum_{i=1}^N A_i + A_\infty = 0,
\end{equation}
and ``gauge away'' redundant degrees of freedom by the
action of $\GL(r,\bfC)$. The left hand side of the linear
constraint is essentially a moment map of this diagonal
$\GL(r,\bfC)$ action.
Analytically, as we see show below, the coadjoint orbit
invariants $\theta_{i\alpha}$ ($i = 1,\cdots,N,\infty$,
$\alpha = 1,\cdots,r$) give local monodromy exponents of
Schlesinger's monodromy problem.
The coadjoint orbit structure leads to a Hamiltonian
formalism of the Schlesinger equation. Let us introduce
a Poisson structure on the vector space $\gl(r,\bfC)^N$
by defining the Poisson bracket of matrix elements of
$A_i = (A_{i,\alpha\beta})$ as:
\begin{equation}
\{ A_{i,\alpha\beta}, A_{j,\rho\sigma} \}
= \delta_{ij} \Bigl(
- \delta_{\beta\rho} A_{i,\alpha\sigma}
+ \delta_{\sigma\alpha} A_{i,\rho\beta} \Bigr).
\end{equation}
In each component of the direct sum, this is just the
ordinary Kostant-Kirillov Poisson bracket. The Schlesinger
equation can be written in the Hamiltonian form
\begin{equation}
\dfrac{\rd A_j}{\rd t_i} = \{ A_j, H_i \},
\end{equation}
where the Hamiltonians are given by
\begin{equation}
H_i = \Res_{\lambda=T_i} \frac{1}{2} \Tr M(\lambda)^2
= \sum_{j(\not= i)} \Tr\left(
\dfrac{A_i A_j}{t_i - t_j} \right),
\end{equation}
and involutive,
\begin{equation}
\{ H_i, H_j \} = 0.
\end{equation}
\subsection{Isomonodromic Deformations}
The Schlesinger equation gives isomonodromic deformations of
the first order ODE
\begin{equation}
\frac{dY}{d\lambda} = M(\lambda)Y
\end{equation}
with the rational coefficient matrix
\begin{equation}
M(\lambda) = \sum_{i=1}^N \dfrac{A_i}{\lambda - t_i}.
\end{equation}
Note that $A_\infty$ is the residue of $M(\lambda)$ at
$\lambda = \infty$.
Usually, this type of isomonodromic problems are considered
under the following
\medskip
\noindent {\bf Assumption}
\begin{itemize}
\item The residue matrices $A_i$ ($i = 1,\cdots,N,\infty$)
are diagonalizable.
\item The eigenvalues $\theta_{i\alpha}$ ($\alpha = 1,\cdots,r$)
of each residue matrix $A_i$ have no integer difference,
i.e., $\theta_{i\alpha} - \theta_{i\beta} \not\in \bfZ$ if
$\alpha \not= \beta$.
\end{itemize}
We assume them throughout this paper. These assumptions
ensure that local solutions at the singular points
$\lambda = t_1,\cdots,t_N,\infty$ develop no logarithmic
term (see below).
The isomonodromic deformations are generated by the deformation
equations
\begin{equation}
\frac{\rd Y}{\rd t_i} = - \dfrac{A_i}{\lambda - t_i} Y.
\end{equation}
These deformation equations and the above first order ODE
comprise an ``auxiliary linear problem'' of the Schlesinger
equation. Its Frobenius integrability conditions can be written
in the ``zero-curvature form''
\begin{equation}
\left[ \dfrac{\rd}{\rd t_j} + \dfrac{A_i}{\lambda - t_i}, \
M(\lambda) - \dfrac{\rd}{\rd \lambda} \right] = 0,
\quad
\left[ \dfrac{\rd}{\rd t_i} + \dfrac{A_i}{\lambda - t_i}, \
\dfrac{\rd}{\rd t_j} + \dfrac{A_j}{\lambda - t_j} \right] = 0,
\end{equation}
and one can easily check that these zero-curvature equations
are equivalent to the Schlesinger equation.
\subsection{Local Solutions at Singular Points and Tau Function}
Since $\lambda = t_1,\cdots,t_N$ and $\lambda=\infty$ are
regular singularities of the above first order ODE, one can
construct a local solution of the following form at each of
these singular points:
\begin{itemize}
\item Local solution at $\lambda = t_i$:
\begin{equation}
Y_i = \Yhat_i \cdot (\lambda - t_i)^{\Theta_i}, \quad
\Yhat_i = \sum_{n=0}^\infty Y_{in} (\lambda - t_i)^n.
\end{equation}
\item Local solution at $\lambda = \infty$:
\begin{equation}
Y_\infty = \Yhat_\infty \cdot \lambda^{-\Theta_\infty}, \quad
\Yhat_\infty = \sum_{n=0}^\infty Y_{\infty n} \lambda^{-n}.
\end{equation}
\end{itemize}
Here $Y_{in}$ are $r \times r$ matrices, the leading coefficients
$Y_{i0}$ and $Y_{\infty 0}$ are invertible, and $\Theta_i$ and
$\Theta_\infty$ are diagonal matrices of local monodromy exponents.
Inserting these expressions into the first order ODE gives the
relations
\begin{equation}
A_i= Y_{i0} \Theta_i Y_{i0}^{-1}, \quad
i = 1,\cdots,N,\infty.
\end{equation}
In particular, local monodromy exponents coincide with the
eigenvalues of $A_i$:
\begin{equation}
\Theta_i = \diag( \theta_{i1}, \cdots, \theta_{ir} ).
\end{equation}
It is not hard to check that these expressions of $A_i$ and
$\Theta_i$ are also consistent with the other equations of
the auxiliary linear problem.
The $\tau$ function of the Schlesinger equation is defined
in two equivalent ways. One way is to define $\log\tau$ as a
potential of the Hamiltonians $H_i$:
\begin{equation}
d\log\tau = \sum_{i=1}^N H_i dt_i.
\end{equation}
Another equivalent definition, which is more suited for
generalization, is based on the equations
\begin{equation}
\frac{\rd \log \tau}{\rd t_i} = \Tr \Theta_i Y_{i0}^{-1}Y_{i1}.
\end{equation}
The equivalence can be verified as follows:
\begin{eqnarray*}
H_i &=& \sum_{j(\not= i)} \Tr \dfrac{A_i A_j}{t_i - t_j} \\
&=& \Res_{\lambda=t_i}
\Tr \dfrac{A_i}{\lambda - t_i} M(\lambda) \\
&=& \Res_{\lambda=t_i}
\Tr \dfrac{Y_{i0} \Theta_i Y_{i0}^{-1}}{\lambda - t_i}
\frac{\rd Y_i}{\rd \lambda} Y_i^{-1} \\
&=& \Res_{\lambda=t_i}
\Tr \dfrac{\Theta_i}{\lambda - t_i}
Y_{i0}^{-1} \frac{\rd Y_i}{\rd \lambda}
Y_i^{-1} Y_{i0} \\
&=& \Tr \Theta_i Y_{i0}^{-1} Y_{i1}.
\end{eqnarray*}
The closedness of the 1-form $\sum H_i dt_i$, or equivalently
the integrability condition
\begin{equation}
\dfrac{\rd H_i}{\rd t_j} = \dfrac{\rd H_j}{\rd t_i},
\end{equation}
is ensured by the Schlesinger equation itself.
\section{Spectral Curve}
By ``spectral curve,'' we mean the plane algebraic
curve defined on the $(\lambda,\mu)$ plane by
\begin{equation}
\det\Bigl( M(\lambda) - \mu I \Bigr) = 0
\end{equation}
and its suitable compactification. We first discuss
its roles in isomonodromic and isospectral problems,
then consider its geometric properties.
\subsection{Spectral Curve in Isomonodromic Problem}
Isomonodromic deformations such as the Schlesinger
equations are non-isospectral, namely, the spectral
curve varies in deformations.
Let us present an interesting formula (essentially
due to Vereschagin \cite{bib:Vereschagin-painleve})
which show {\it qualitatively} that the characteristic
polynomial of $M(\lambda)$ varies under isomonodromic
deformations. First, by the the well known identities
of linear algebra, we have
\begin{eqnarray*}
\frac{\rd}{\rd t_i} \log \det \Bigl( M(\lambda) - \mu I \Bigr)
&=& \frac{\rd}{\rd t_i} \Tr\log
\Bigl( M(\lambda) - \mu I \Bigr) \\
&=& \Tr \frac{\rd M(\lambda)}{\rd t_i}
\Bigl( M(\lambda) - \mu I \Bigr)^{-1}.
\end{eqnarray*}
By the first equation of the zero-curvature representation,
$\rd M(\lambda)/\rd t_i$ can be rewritten
\[
\frac{\rd M(\lambda)}{\rd t_i}
= \dfrac{A_i}{(\lambda - t_i)^2}
- \left[ \dfrac{A_i}{\lambda - t_i}, M(\lambda) - \mu I \right].
\]
The second term on the right hand side has no contribution,
because of the identity
\[
\Tr \left[ \dfrac{A_i}{\lambda - t_i}, M(\lambda) - \mu I \right]
\Bigl( M(\lambda) - \mu \Bigr)^{-1} = 0.
\]
Thus, eventually, we obtain the formula
\begin{equation}
\frac{\rd}{\rd t_i} \log\det \Bigl( M(\lambda) - \mu I \Bigr)
= \Tr \dfrac{A_i}{(\lambda - t_i)^2}
\Bigl( M(\lambda) - \mu I \Bigr)^{-1}.
\end{equation}
This clearly shows that the $t$-dependence of the characteristic
polynomial of $M(\lambda)$ is driven by the ``anomalous'' term
$A_i/(\lambda - t_i)^2$.
\subsection{Spectral Curve in Isospectral Problem}
We now turn to Garnier's autonomous analogue of the
Schlesinger equation. This equation has the following
zero-curvature representation:
\begin{equation}
\left[ \dfrac{\rd}{\rd t_i} + \dfrac{A_i}{\lambda - c_i}, \
M(\lambda) \right] = 0,
\quad
\left[ \dfrac{\rd}{\rd t_i} + \dfrac{A_i}{\lambda - c_i}, \
\dfrac{\rd}{\rd t_j} + \dfrac{A_j}{\lambda - c_j} \right] = 0.
\end{equation}
Repeating the same calculations as above, we now find that
\begin{equation}
\frac{\rd}{\rd t_i} \det\Bigl( M(\lambda) - \mu I \Bigr) = 0,
\end{equation}
because there is no ``anomalous'' term like $A_i/(\lambda - t_i)^2$.
Thus one can confirm that Garnier's autonomous system is
indeed an isospectral problem. An auxiliary linear problem
is given by
\begin{equation}
\mu \psi = M(\lambda) \psi, \quad
\frac{\rd \psi}{\rd t_i} = - \dfrac{A_i}{\lambda - c_i} \psi.
\end{equation}
Here $\psi$ is understood to be a column vector. The first
equation means that $\psi$ is an eigenvector of $M(\lambda)$
with eigenvalue $\mu$. The second set of equations generate
isospectral deformations.
It is nowadays well known that this type of isospectral
problems can be mapped to linear flows on the Jacobian
variety of the spectral curve \cite{bib:DMN,bib:Krichever1977}.
Reproducing the original nonlinear problem is identical
to Jacobi's inversion problem, and indeed solvable by
theta functions (or Baker-Akhiezer functions). The solution
thus obtained is written in terms of period integrals and
Abelian functions.
Although isomonodromic problems (in a generic case) cannot
be solved in that way, the notion of spectral curve still
plays a role in the study of Hamiltonian structures.
Indeed, Harnad and Wisse \cite{bib:Harnad-dual,bib:HW}
presented a construction of special Darboux coordinates
(``spectral Darboux coordinates'') using the language
of spectral curves. The case of $r = 2$ is particularly
interesting, because this is the case where the 6th
Painlev\'e equation (P$_{\rm VI}$) and Garnier's
multi-variable version of P$_{\rm VI}$ (called the
``Garnier system'' by Okamoto \cite{bib:Okamoto}) emerge.
The spectral Darboux coordinates in this case coincide
with Okamoto's Darboux coordinates for the ``Garnier system,''
which Okamoto discovered without using the notion of spectral
curve.
\subsection{Geometry of Spectral Curves}
The structure of spectral curves of the above type
has been elucidated in detail by the Montreal group
\cite{bib:AHHP,bib:AHH} and Beauville \cite{bib:Beauville}.
We present basic part of their results, which will be used
in the subsequent sections. In the following, the poles of
$M(\lambda)$ are written $c_i$ rather than $T_i$.
\subsubsection{Spectral Curve on Plane.}
Let $C_0$ be the spectral curve on the $(\lambda,\mu)$ plane:
\begin{equation}
F(\lambda,\mu) = \det\Bigl( M(\lambda) - \mu I \Bigr) = 0.
\end{equation}
This becomes a ramified covering of the punctured Riemann sphere,
\begin{equation}
\pi: C_0 \to \bfC P^1 \setminus \{ c_1,\cdots,c_N,\infty \},
\quad \pi(\lambda,\mu) = \lambda.
\end{equation}
For a generic value of $\lambda$, the inverse image
$\pi^{-1}(\lambda)$ consists of $r$ points
$(\lambda,\mu_\alpha)$ ($\alpha = 1,\cdots,r$). The
$\mu$-coordinates $\mu_\alpha$ of these points are
eigenvalues of $M(\lambda)$. In a neighborhood of
$\lambda = c_i$, $\mu_\alpha$'s behave as
\begin{equation}
\mu_\alpha = \dfrac{\theta_{i\alpha}}{\lambda - c_i}
+ \mbox{non-singular},
\end{equation}
where $\theta_{i\alpha}$ are the eigenvalues of $A_i$.
Similarly, in a neighborhood of $\lambda = \infty$,
\begin{equation}
\mu_\alpha = - \theta_{\infty\alpha} \lambda^{-1}
+ O(\lambda^{-2}),
\end{equation}
where $\theta_{\infty\alpha}$ are the eigenvalues of
$A_\infty = - \sum_{i=1}^N A_i$.
(Of course, the numbering of $\mu_\alpha$'s is meaningful
only locally.)
\subsubsection{Compactification of Spectral Curve}
We now compactify $C_0$ by adding several points over
the punctures of the Riemann sphere. In a neighborhood
of $\lambda = c_i$, let us consider the following
$\mutilde$ in place of $\mu$:
\begin{equation}
\mutilde = f(\lambda) \mu, \quad
f(\lambda) = \prod_{i=1}^N (\lambda - c_i).
\end{equation}
In terms of the new coordinates $(\lambda,\mutilde)$,
the equation of the spectral curve becomes
\begin{equation}
\Ftilde(\lambda,\mutilde)
= \det \Bigl( f(\lambda)M(\lambda) - \mutilde I \Bigr) = 0.
\end{equation}
The inverse image $\pi^{-1}(\lambda)$ consists of $r$ points
$(\lambda,\mutilde_\alpha)$ ($r = 1,\cdots,r)$ such that
\begin{equation}
\mutilde_\alpha =
f'(c_i) \theta_{i\alpha} + O(\lambda- c_i)
\end{equation}
as $\lambda \to c_i$. Since the eigenvalues
$\theta_{i\alpha}$ ($\alpha = 1,\cdots,r$)
are pairwise distinct (recall the assumptions in Section 2),
we add to $C_0$ extra $r$ points $(\lambda,\mutilde)
= \bigl(c_i,f'(c_i)\theta_{i\alpha} \bigr)$ to fill the holes
above $\lambda = c_i$.
Similarly, in a neighborhood of $\lambda = \infty$, we use
\begin{equation}
\mutilde = \lambda \mu
\end{equation}
in place of $\mu$. $\pi^{-1}(\lambda)$ now consists of the
$r$ points $(\lambda,\mutilde_\alpha)$ ($r = 1,\dots,r)$
such that
\begin{equation}
\mutilde_\alpha = - \theta_{\infty\alpha}
+ O(\lambda^{-1})
\end{equation}
as $\lambda \to \infty$. The eigenvalues
$\theta_{\infty\alpha}$ ($r = 1,\cdots,r$), too, are
pairwise distinct. Therefore we add to $C_0$ the $r$ points
$(\lambda,\mutilde) = (\infty,-\theta_{\infty\alpha})$
to fill the holes above $\lambda = \infty$.
Thus, by adding altogether $rN + r$ points to $C_0$, we obtain
a compactification $C$ of $C_0$. The covering map $\pi$
uniquely extends to $C$, and gives a ramified covering
$\pi: C \to \bfC P^1$ of the Riemann sphere.
\subsubsection{Genus of Compactified Spectral Curve}
If there are multiple eigenvalues of $A_i$, the compactified
spectral curve $C$ has singular points (nodes) over
$\lambda = c_i$. In that case, one has to take a
desingularization $\rho: \Ctilde \to C$
for further consideration. Fortunately, this does not occur
in our case because of the assumptions introduced in Section 2.
One can show, by a standard method, that $C$ has the genus
\begin{equation}
g = \frac{1}{2}(r-1)(rN - r - 2).
\end{equation}
\subsubsection{Structure of Characteristic Polynomial}
In order to examine the structure of the characteristic
polynomial of $M(\lambda)$ (or, rather, $f(\lambda)M(\lambda)$),
we take the rational matrix
\begin{equation}
M^0(\lambda) = \sum_{i=1}^N \dfrac{A^0_i}{\lambda - c_i},
\quad
A^0_i = \diag( \theta_{i1}, \cdots, \theta_{ir} ),
\end{equation}
as a reference point on the coadjoint orbit that $M(\lambda)$
belongs to. Now compare the characteristic polynomials
$\Ftilde(\lambda,\mutilde)$ and $\Ftilde^0(\lambda,\mutilde)$
of $f(\lambda)M(\lambda)$ and $f(\lambda)M^0(\lambda)$.
Note, first, that
$\Ftilde(\lambda,\mutilde) - \Ftilde^0(\lambda,\mu)$
vanishes at $\lambda = c_i$. This will be obvious from
the following expression of these polynomials as
$\lambda \to c_i$:
\begin{eqnarray*}
\Ftilde(\lambda,\mutilde) =
\det\Bigl( f'(c_i)A_i - \mutilde I \Bigr) + O(\lambda - c_i), \\
\Ftilde^0(\lambda,\mutilde) =
\det\Bigl( f'(c_i)A^0_i - \mutilde I \Bigr) + O(\lambda - c_i).
\end{eqnarray*}
(The two determinants on the right hand side are equal,
because $A_i$ and $A^0_i$ are on the same coadjoint orbit.)
Therefore $\Ftilde(\lambda,\mutilde) - \Ftilde^0(\lambda,\mu)$
is divisible by $f(\lambda)$. Furthermore, these
characteristic polynomials have the following expansion
in powers of $\mutilde$:
\begin{eqnarray*}
\Ftilde(\lambda,\mutilde) =
(-\mutilde)^r + \Tr M(\lambda) (-\mutilde)^{r-1} + \cdots, \\
\Ftilde^0(\lambda,\mutilde) =
(-\mutilde)^r + \Tr M^0(\lambda) (-\mutilde)^{r-1} + \cdots.
\end{eqnarray*}
Since the first two terms on the right hand side are equal,
respectively, the difference of the two polynomials contains
no terms proportional to $\mu^r$ and $\mu^{r-1}$. From these
observations, one can conclude that the difference of the
characteristic functions can be written
\[
\Ftilde(\lambda,\mutilde) - \Ftilde(\lambda,\mutilde) =
f(\lambda) \sum_{\ell=2}^r p_\ell(\lambda) \mutilde^{r-\ell},
\]
where $p_\ell(\lambda)$'s are polynomial functions of $\lambda$.
A simple power counting argument (assigning weight $1$ to
$\lambda$ and weight $N-1$ to $\mutilde$) shows that the degree
of $p_\ell(\lambda)$ does not exceed the positive integer
\begin{equation}
\delta_\ell = (N-1)\ell - N.
\end{equation}
Therefore $p_\ell(\lambda)$ can be written
\[
p_\ell(\lambda) =
\sum_{m=0}^{\delta_\ell} h_{m\ell} \lambda^m.
\]
The leading coefficient of this polynomial is a
function of $\theta_{i\alpha}$'s only, because it
can be writen
\begin{equation}
h_{\delta_\ell,\ell}
= (-1)^r \Bigl( \sigma_\ell(A_\infty)
-\sigma_\ell(A^0_\infty) \Bigr),
\end{equation}
where $\sigma_\ell(A_\infty)$ and $\sigma_\ell(A^0_\infty)$
are the $\ell$-th elementary symmetric function of
$A_\infty$ and $A^0_\infty = - \sum_{i=1}^r A^0_i$.
Thus, eventually,
we arrive at the following expression of
$\Ftilde(\lambda,\mutilde)$:
\begin{equation}
\Ftilde(\lambda,\mutilde) =
\Ftilde^0(\lambda,\mutilde)
+ \sum_{\ell=2}^r \sum_{m=0}^{\delta_\ell}
h_{m\ell} \lambda^m \mutilde^{r-\ell}.
\end{equation}
\subsubsection{Parameters of Spectral Curve}
Apart from the leading coefficients, the coefficients
$h_{m\ell}$ ($\ell = 2,\cdots,r$, $m = 0,\cdots,\delta_\ell - 1$)
of $p_\ell(\lambda)$'s give arbitrary parameters (``moduli'')
of the spectral curve. Their total number coincides with
the genus of the spectral curve:
\begin{equation}
\sum_{\ell=2}^r \delta_\ell
= \sum_{\ell=2}^r \Bigl( (N-1)\ell - N \Bigr)
= g.
\end{equation}
In the isospectral problem, these parameters $h_{m\ell}$
are constants of motion (Hamiltonians of commuting
isospectral flows). They should not be confused with the
Hamiltonians $H_i$ of the isomonodromic deformations.
Actually, the characteristic polynomial has yet another
set of parameters --- the position $c_i$ of poles of
$M(\lambda)$. They play the role of deformation
variables in isomonodromic problem.
In summary, the spectral curve has three distinct sets of
parameters:
\begin{itemize}
\item Position of poles $c_i$ ($i=1,\cdots,N$).
\item Coadjoint orbit invariants $\theta_{i\alpha}$
($i=1,\cdots,N,\infty$).
\item Isospectral invariants $h_{m\ell}$
($\ell=2,\cdots,r$, $m=0,\ldots,\delta_\ell-1$).
\end{itemize}
\section{Schlesinger Equation with Small Parameter}
\subsection{Reformulation Including Small Parameter}
The first step towards the derivation of our modulation
equation is to reformulate the Schlesinger equation by the
following substitution rule:
\begin{equation}
\frac{\rd}{\rd t_i} \to \epsilon \frac{\rd}{\rd T_i},
\quad
\frac{\rd}{\rd \lambda} \to \epsilon \frac{\rd}{\rd \lambda},
\quad
\dfrac{A_i}{\lambda - t_i} \to \dfrac{A_i}{\lambda - T_i}.
\end{equation}
Note, in particular, that the deformation variables are
renamed as $t_i \to T_i$. $T_i$'s will play the role of
``slow variables.'' The reformulated Schlesinger equation
reads:
\begin{equation}
\epsilon \dfrac{\rd A_i}{\rd T_j}
= (1 - \delta_{ij}) \dfrac{[A_i,A_j]}{T_i - T_j}
- \delta_{ij} \sum_{k (\not= i)} \dfrac{[A_i,A_k]}{T_i - T_k}.
\end{equation}
An auxiliary linear problem is given by
\begin{equation}
\epsilon \frac{\rd Y}{\rd \lambda} = M(\lambda) Y, \quad
\epsilon \frac{\rd Y}{\rd T_i} = - \dfrac{A_i}{\lambda - T_i} Y.
\end{equation}
The above $\epsilon$-dependent Schlesinger equation can be indeed
reproduced from the the Frobenius integrability conditions
\begin{eqnarray}
\left[ \epsilon \dfrac{\rd}{\rd T_j} + \dfrac{A_i}{\lambda - T_i}, \
M(\lambda) - \epsilon \dfrac{\rd}{\rd \lambda} \right] = 0,
\nonumber \\
\left[ \epsilon \dfrac{\rd}{\rd T_i} + \dfrac{A_i}{\lambda -T_i}, \
\epsilon \dfrac{\rd}{\rd T_j} + \dfrac{A_j}{\lambda - T_j}
\right] = 0.
\end{eqnarray}
A few comments on the above reformulation will be in order:
(i) Reformulating the Schlesinger equation as above is
inspired by the work of Vereschagin
\cite{bib:Vereschagin-painleve}, who considered all the six
Painlev\'e equations (P$_{\rm I}$--P$_{\rm VI}$) in such an
$\epsilon$-dependent form. This is different from the way
Boutroux \cite{bib:Boutroux} and Garnier \cite{bib:Garnier1917}
derived an Abelian function approximation,
but, as Vereschagin stresses, they are asymptotically related.
(ii) The string equations of two-dimensional quantum gravity,
which are P$_{\rm I}$, P$_{\rm II}$ and their higher order
generalization, contain such a small parameter from the
beginning. The small parameter i interpreted as
``string coupling constant'' \cite{bib:Moore-review}.
\subsection{Isospectral Deformations as First Approximation}
One can now see, at least intuitively, that an isospectral
problem emerges from the above $\epsilon$-dependent
Schlesinger equation. Let us introduce the ``fast variables''
\begin{equation}
t_i = \epsilon^{-1} T_i,
\end{equation}
and rewrite the above equation as
\begin{equation}
\dfrac{\rd A_i}{\rd t_j}
= (1 - \delta_{ij}) \dfrac{[A_i,A_j]}{T_i - T_j}
- \delta_{ij} \sum_{k (\not= i)} \dfrac{[A_i,A_k]}{T_i - T_k}.
\end{equation}
Suppose we now observe this system {\it in the scale of the
fast variables\/} $t_i$. In this scale, $T_i$'s may be
treated as being {\it approximately constant}, because
a finite displacement in $t_i$'s corresponds to a small
(i.e., $O(\epsilon)$) displacement in $T_i$'s. If
$T_i$'s were true constants, the above equation would
be exactly Garnier's autonomous system. Thus, observed
in the scale of the fast variables, our $\epsilon$-dependent
Schlesinger equation looks {\it approximately} like an
isospectral problem.
Of course, this is no more than an approximation.
In fact, $T_i$'s are not constant, but vary slowly in
the order of $O(\epsilon)$. Accordingly, the spectral
curve, too, deforms slowly because the defining
equation of the spectral curve contains $T_i$'s
explicitly. A precise description of the isospectral
approximation has to take into account such slow
deformations of the spectral curve.
It is, however, not only $T_i$'s that vary; the isospectral
invariants $h_{m\ell}$ also change values slowly. They are
both responsible for deformations of the spectral curve.
In order to see how this occurs, recall the derivative
formula of the characteristic polynomial under isomonodromic
deformations (Section 3.1). In the present setting,
this formula takes the form
\begin{equation}
\frac{\rd}{\rd T_i} \log\det \Bigl( M(\lambda) - \mu I \Bigr)
= \Tr \dfrac{A_i}{(\lambda - T_i)^2}
\Bigl( M(\lambda) - \mu I \Bigr)^{-1}.
\end{equation}
The right hand side plays the role of a ``driving force''
for slow deformations of the characteristic polynomial.
More precisely, in order to extract a true driving force
of slow deformations, one has to take an average of the
right hand side over the quasi-periodic motion in the
fast variables $t_i$. This is a central idea of
Vereschagin's averaging method
\cite{bib:Vereschagin-string,bib:Vereschagin-painleve}.
Meanwhile, the coadjoint orbit invariants $\theta_{i\alpha}$
remain constant, because they are also invariant of
isomonodromic deformations.
\subsection{Concept of Multiscale Analysis}
Although frequently lacking mathematical rigor,
``multiscale analysis'' is widely accepted in applied
mathematics as a powerful tool for dealing with this
kind of problems \cite{bib:multiscale}.
The Whitham averaging method, too, has been developed
in the framework of multiscale analysis. The nonlinear
JWKB method of Dobrokhotov and Maslov \cite{bib:D-Maslov}
is an attempt at a rigorous reformulation of multiscale
analysis. Let us show how our problem may be formulated
in the language of multiscale analysis.
A key of multiscale analysis is to treat the fast
variables $t = \{ t_i \}$ and the slow variables
$T = \{ T_i \}$ as {\it independent variables\/}.
$A_i$'s are thus assumed to be a function of $t$ and
$T$ (and the small parameter $\epsilon$),
\begin{equation}
A_i = A_i(t,T,\epsilon).
\end{equation}
The relation $t_i = \epsilon^{-1} T_i$ is imposed only
in the differential equation in question. (For a more
precise description, one may introduce a series of
``slower'' variables $t^{(2)}$, $t^{(3)}$, $\cdots$,
related with $t_k$'s as
\begin{equation}
t^{(k)}_i = \epsilon^k t_i.
\end{equation}
For our purpose, only the first two scales are sufficient.)
In this multiscale ansatz, derivative terms in the equation
are given by a sum of $t$-derivatives and $T$-derivatives.
In terms of differential operators, this amounts to substituting
\begin{equation}
\epsilon \frac{\rd}{\rd T_i} \to
\frac{\rd}{\rd t_i} + \epsilon \frac{\rd}{\rd T_i}.
\end{equation}
We now assume an asymptotic expansion of the form
\begin{equation}
A_i(t,T,\epsilon) =
A^{(0)}_i(t,T) + A^{(1)}_i(t,T) \epsilon + \cdots,
\end{equation}
and plug all these stuff into the Schlesinger equation.
From each order of $\epsilon$, we obtain a differential
equation for the coefficients of the above expansion.
The lowest order equation is give by
\begin{equation}
\dfrac{\rd A^{(0)}_i}{\rd t_j}
= (1 - \delta_{ij}) \dfrac{[A^{(0)}_i,A^{(0)}_j]}{T_i - T_j}
- \delta_{ij} \sum_{k (\not= i)}
\dfrac{[A^{(0)}_i,A^{(0)}_k]}{T_i - T_k}.
\end{equation}
If we consider $T_i = c_i$, this is nothing but Garnier's
autonomous system. Note that this derivation of Garnier's
isospectral problem is more understandable than the
intuitive derivation in the last subsection. In the derivation
of the last subsection, $t$ and $T$ were not independent
and constrained by the relation $t_i = \epsilon^{-1}T_i$,
thereby we had to say that $T$ is ``approximately constant'';
in the setting of multiscale analysis, $t$ and $T$ are
{\it independent\/}. This shows a conceptual advantage of
multiscale analysis.
The lowest order equation, however, carries no information
on the $T$-dependence of $A^{(0)}_i$'s, which are to be
determined by the next order equation. The next order
equation is given by
\begin{eqnarray}
\dfrac{\rd A^{(1)}_i}{\rd t_j}
&=& (1 - \delta_{ij})
\dfrac{[A^{(0)}_i,A^{(1)}_j] + [A^{(1)}_i,A^{(0)}_j]}{T_i - T_j}
\nonumber \\
& & - \delta_{ij} \sum_{k (\not= i)}
\dfrac{[A^{(0)}_i,A^{(1)}_k] + [A^{(1)}_i,A^{(0)}_k]}{T_i - T_k}
\nonumber \\
& & + (\mbox{terms containing $A^{(0)}$'s and
their $T$-derivatives only}).
\end{eqnarray}
This equation contains $T$-derivatives of $A^{(0)}_i$ as well
as $t$-derivatives of $A^{(1)}_i$. A standard prescription of
multiscale analysis is to eliminate the latter by ``averaging
over the $t$ space.'' This procedure usually takes the form
of an ``integrability condition'' for the above equation
to have a solution $A^{(1)}_i$ under suitable a boundary
condition (e.g., requiring the same quasi-periodicity as
$A^{(0)}_i$'s possess).
One thus obtains a differential equation (in $T$) for
$t$-averaged functionals of $A^{(0)}_i$'s. This is one
of various possible expressions (``modulation equations'')
of modulational dynamics. As Whitham pointed out
\cite{bib:Whitham}, such a modulation equation frequently
appears in the form of ``averaged conservation laws.''
Having this fact in mind, Flaschka, Forest and McLaughlin
\cite{bib:FFM} considered averaged conservation lows of
the KdV equation, and derived their compact expression of
this problem.
\subsection{Our approach to Modulation Equation}
There are many technical difficulties in deriving a
modulation equation from the Schlesinger equation along
the line presented above. A main obstacle is the fact
that the spectral curve is no longer hyperelliptic for
$r > 2$. Most attempts in the literature, including the
work of Flaschka, Forest and McLaughlin, have been limited
to hyperelliptic spectral curves. This considerably reduces
the obstacles (though a complete treatment of the problem
is still by no means an easy task). The spectral curve
of the Schlesinger equation is hyperelliptic only if
$r = 2$. In the general case, the averaging method is
inevitably confronted with delicate problems of geometry
of spectral curves. (In this respect, Krichever's
averaging method \cite{bib:Krichever-averaging}
for general spectral curves is quite remarkable.)
Since our main concern is the structure of the modulation
equation rather than the averaging method itself, we now bypass
these delicate issues by a very heuristic argument.
\section{Modulation Equation}
\subsection{Multiscale Analysis of Auxiliary Linear Problem}
The first step of our heuristic argument is to apply the
concept of multiscale analysis to the auxiliary linear
problem of the Schlesinger equation. Besides the
multiscale expression of $A_i$'s, we now assume the
following ansatz to $Y$ (which is now understood to be
{\it vector-valued}):
\begin{equation}
Y = \Bigl( \phi^{(0)}(t,T,\lambda)
+ \phi^{(1)}(t,T,\lambda) \epsilon + \cdots \Bigr)
\exp \epsilon^{-1} S(T,\lambda).
\end{equation}
Here $\phi^{(k)}(t,T,\lambda)$ are vector-valued function
and $S(T,\lambda)$ a scalar-valued function. This is a
kind of JWKB ansatz (inspired by the work of Flaschka and
Newell \cite{bib:FN} and Novikov \cite{bib:Novikov}).
Note that the ``phase function'' $S$ is independent of $t$.
We can now write the auxiliary linear problem in the
following multiscale form:
\begin{eqnarray}
\epsilon \frac{\rd Y}{\rd \lambda}
= M(\lambda)Y, \\
\left( \frac{\rd}{\rd t_i} +
\epsilon \frac{\rd}{\rd T_i} \right) Y
= - \dfrac{A_i}{\lambda - T_i} Y.
\end{eqnarray}
The leading order equation should reproduce the auxiliary
linear problem of the isospectral problem. Let us confirm
that this is indeed the case. The leading order equation
is given by
\begin{eqnarray}
\frac{\rd S(\lambda)}{\rd \lambda} \phi^{(0)}(\lambda)
= M^{(0)}(\lambda) \phi^{(0)}(\lambda), \\
\frac{\rd \phi^{(0)}(\lambda)}{\rd t_i}
+ \dfrac{\rd S(\lambda)}{\rd T_i} \phi^{(0)}(\lambda)
= - \dfrac{A^{(0)}_i}{\lambda - T_i} \phi^{(0)},
\end{eqnarray}
where
\begin{equation}
M^{(0)}(\lambda)
= \sum_{i=1}^N \dfrac{A^{(0)}_i}{\lambda - T_i}.
\end{equation}
We now define
\begin{equation}
\mu = \dfrac{\rd S(\lambda)}{\rd \lambda}, \quad
\psi = \phi^{(0)}(\lambda)
\exp \sum_{i=1}^N t_i
\frac{\rd S(\lambda)}{\rd T_i}.
\end{equation}
In terms of these quantities, the leading order equation
of multiscale expansion can be rewritten
\begin{equation}
\mu \psi = M^{(0)}(\lambda) \psi, \quad
\frac{\rd \psi}{\rd t_i}
= - \dfrac{A^{(0)}_i}{\lambda - T_i} \psi.
\end{equation}
This is exactly the auxiliary linear problem of the
isospectral problem that we derived in Section 4!
\subsection{Matching with Baker-Akhiezer Function}
The next step is the most crucial part of our heuristics.
We now compare the above $\psi$ with the Baker-Akhiezer
function in the ordinary algebro-geometric approach
\cite{bib:DMN,bib:Krichever1977} to finite-band solutions
of soliton equations.
It is well known that the solution of the auxiliary
linear problem for a finite-band solution can be
constructed as a (scalar- or vector-valued)
Baker-Akhiezer function. In the present setting,
such a Baker-Akhiezer function can be written
\begin{equation}
\psi = \phi \exp \sum_{i=1}^N t_i \Omega_i.
\end{equation}
Here $\phi$ is a vector-valued function on the spectral
curve, also depending on $t$; each entry is a combination
of theta functions. We do not specify its detailed
structure, because it is irrelevant in the following
consideration. Meanwhile, $\Omega_i$ is the primitive
function of a meromorphic differential $d\Omega_i$
on the spectral curve,
\begin{equation}
\Omega_i = \int^{(\lambda,\mu)} d\Omega_i.
\end{equation}
Note that these quantities also depend on $T$ through
the $T$-dependence of the spectral curve. Another
important remark is that such an expression of the
Baker-Akhiezer function is available only after
selecting a ``symplectic homology basis'' of the spectral
curve, i.e., cycles $A_I,B_I$ ($I = 1,\cdots,g$) with
intersection numbers $A_I \cdot A_J = B_I \cdot B_J = 0$
and $A_I \cdot B_J = \delta_{IJ}$. (This issue will be
discussed in detail in Section 6.)
We now {\it assume} that, for a suitable symplectic
homology basis, this Baker-Akhiezer function coincides
with the $\psi$ derived from the leading order equation
of multiscale expansion. More precisely, we require
their ``amplitude part'' and ``phase part,'' respectively,
to coincide:
\begin{equation}
\phi^{(0)} = \phi, \quad
\frac{\rd S}{\rd T_i} = \Omega_i.
\end{equation}
Actually, the first relation may be relaxed as
\begin{equation}
\phi^{(0)} = \phi h(T,\lambda),
\end{equation}
where $h(T,\lambda)$ is a scalar function independent
of $t$; this takes into account the obvious symmetry
\begin{equation}
\phi^{(0)} \to \phi^{(0)} h(T,\lambda)
\end{equation}
of the lowest order equation of multiscale expansion.
Such a factor $h(T,\lambda)$ might be necessary to
proceed to the next order approximation of multiscale
analysis. We shall not go into this issue here.
\medskip
{\it Remark\/}.
We take this opportunity to correct an error in the
previous publication \cite{bib:TN}. Substantially
the same JWKB ansatz also assumed therein, but with
an extra factor of the form $(\rd^2 S/\rd \lambda^2)^{1/2}$
--- see Eqs. (13) and (21) therein. This factor
itself is nonsense. Fortunately, this does not
affect the leading order of multiscale analysis.
As mentioned above, however, something like this
factor will be necessary in the next order. The
arbitrary function $h$ stands for such a compensating
factor.
\subsection{Modulation Equation}
Thus, under several assumptions, we have been able to
derive a series of relations that link the isomonodromic
and isospectral problems. In particular, the following
equations are obtained:
\begin{equation}
\frac{\rd S}{\rd \lambda} = \mu, \quad
\frac{\rd S}{\rd T_i} = \Omega_i.
\end{equation}
We propose these equations as the modulation equation
that governs slow dynamics of the spectral curve
\begin{equation}
\det\Bigl( M^{(0)}(\lambda) - \mu I \Bigr) = 0.
\end{equation}
Remember that one has to select a symplectic homology
basis $A_I,B_I$ ($I = 1,\cdots,g$) in order to derive
these equations. Actually, selecting a proper homology
basis is a non-trivial problem. We discuss this issue
in more detail in Section 6, along with a precise
characterization of the meromorphic differentials $d\Omega_i$.
\subsection{Relation to Generic Whitham Equation}
As a piece of evidence that our modulation equation is
a reasonable one, we now show that solutions of this
equation give a special subfamily of generic Whitham
deformations.
The first equation of the modulation equation can be rewritten
\begin{equation}
dS = \frac{\rd S}{\rd \lambda} d\lambda = \mu d\lambda.
\end{equation}
In other words, $S$, like $\Omega_i$'s, is the primitive
function of a meromorphic differential $dS$ on the
spectral curve:
\begin{equation}
S = \int^{(\lambda,\mu)} \mu d\lambda.
\end{equation}
The second part of the modulation equation implies the
equations
\begin{equation}
\frac{\rd}{\rd T_i} dS = d\Omega_i.
\end{equation}
Accordingly, the generic Whitham equation
\begin{equation}
\frac{\rd}{\rd T_i} d\Omega_j = \frac{\rd}{\rd T_j} d\Omega_j
\end{equation}
follows immediately. (Here, as usual, $\lambda$ is understood
to be {\it constant} under the $T$-derivation. Geometrically,
this should be treated as a connection. See the paper of
Krichever and Phong \cite{bib:Krichever-Phong}.)
Thus, our modulation equation turns out to yield a special
subfamily of generic Whitham deformations. We shall see
in Section 8 that an inverse period map yields generals
solution of this equation.
\subsection{Possible Approach from Averaging Method}
Our heuristic argument has to be justified in a more
rigorous form, or at least cross-checked by some other
method. A possible check will be to derive the above
equations by an averaging argument.
Krichever's averaging method \cite{bib:Krichever-averaging}
(see also the papers by Fucito et al. \cite{bib:Fucito-etal}
Carroll and Chang \cite{bib:Carroll-Chang}) seems to
be particularly promising. This method employs the
so called ``dual Baker-Akhiezer function'' $\psi^*$, i.e.,
a solution of the dual auxiliary linear problem
\begin{equation}
\mu \psi^* = \psi^* M^{(0)}(\lambda), \quad
\frac{\rd \psi^*}{\rd t_i}
= \psi^* \dfrac{A^{(0)}_i}{\lambda - T_i},
\end{equation}
along with $\psi$. One can derive, for instance, the
following formula evaluating $T$-dependence of the
eigenvalue $\mu$ of $M^{(0)}$:
\begin{equation}
\frac{\rd \mu}{\rd T_i}
= \dfrac{\Bigl< \psi^* A^{(0)}(\lambda - T_i)^{-2} \psi \Bigr>}
{<\psi^* \psi>}.
\end{equation}
Here $<\cdots>$ means the average over the fast variables $t$.
This formula can be readily translated to the language of $dS$
(though we have been unable to calculate the average and to
identify the resulting equation with the above equation for
$dS$). Note that the term $A^{(0)}(\lambda - T_i)^{-2}$ in
the above formula is the same as the ``driving force'' term
that we encountered in the derivative formula of the characteristic
polynomial of $M(\lambda)$.
\section{Structure of Meromorphic Differentials}
In this and subsequent sections, we omit the superfix
``$^{(0)}$'' and write $A^{(0)}_i$, $M^{(0)}(\lambda)$,
etc. as $A_i$, $M(\lambda)$, $\cdots$. This is just
for simplifying notations.
\subsection{General Remarks}
The vector-valued Baker-Akhiezer function $\psi$ is single
valued on the spectral curve, meromorphic outside the points
of $\pi^{-1}(\{T_1,\cdots,T_N\})$, and has essential
singularities of a particular exponential form at these
exceptional points. The entries of $\phi$ have a common
pole divisor $D$, part of which may overlap with the
essential singularities. (In the following, we consider
a generic case where this overlapping does not occur.)
The essential singularities at points in $\pi^{-1}(T_i)$
are generated by poles of the meromorphic differential
$d\Omega_i$.
The meromorphic differentials $d\Omega_i$ have to be
selected for the essential singularities of $\psi$ to
match the auxiliary linear problem. This yields conditions
on the poles of the meromorphic differentials. Our task
in the following is to specify those conditions.
In order to consider this problem, it is convenient to build
a matrix solution $\Psi$ of the auxiliary linear problem
from the vector-valued Baker-Akhiezer function $\psi$. Let
$P_\alpha(\lambda)$ ($\alpha = 1,\cdots,r$), denote the points
of $\pi^{-1}(\lambda)$. (Of course, also here, such numbering
is meaningful only locally.) $\Psi$ is given by
\begin{equation}
\Psi = \Bigl( \psi(P_1(\lambda)),\cdots,\psi(P_r(\lambda)) \Bigr).
\end{equation}
This matrix has the following factorized form:
\begin{equation}
\Psi = \Phi \exp \diag\Bigl(
\sum t_i \Omega_i( P_1(\lambda) ), \cdots,
\sum t_i \Omega_i( P_r(\lambda) ) \Bigr).
\end{equation}
$\Phi = \Phi(\lambda)$ is an $r \times r$ matrix originating
in the ``amplitude part'' $\phi$, and invertible at a generic
point.
\subsection{Conditions on Meromorphic Differentials}
\subsubsection{Poles of $d\Omega_i$}
All necessary information on poles of $d\Omega_i$ can be
derived from the following relation, which is obtained by
inserting the above expression of $\Psi$ into the auxiliary
linear problem:
\begin{equation}
\frac{\rd \Phi}{\rd t_i} \Phi^{-1}
+ \Phi \diag\Bigl(
\Omega_i( P_1(\lambda) ), \cdots,
\Omega_i( P_r(\lambda) ) \Bigr) \Phi^{-1}
= - \dfrac{A_i}{\lambda - T_i}.
\end{equation}
We now compare the singular part of both hand side at
$\lambda = T_i$. Since the first term on the left hand side
is non-singular, it turns out that $\Omega_i$ has a pole of
first order at each point of $\pi^{-1}(T_i)$, and the residue
is equal to $(-1)$ times an eigenvalues of $A_i$. Since the
eigenvalues of $A_i$ are $\theta_{i\alpha}$, this implies that
(after suitably renumbering the eigenvalues) $A_i$ can be written
\begin{equation}
A_i = \Phi(T_i) \diag\Bigl( \theta_{i1},\cdots,\theta_{ir} \Bigr)
\Phi(T_i)^{-1},
\end{equation}
and that
\begin{equation}
\Omega_i = - \dfrac{\theta_{i\alpha}}{\lambda - T_i}
+ \mbox{non-singular}
\end{equation}
in a neighborhood of $P_\alpha(T_i)$.
In particular, the meromorphic differential $d\Omega_i$
behaves as
\begin{equation}
d\Omega_i
= \dfrac{\theta_{i\alpha}}{(\lambda - T_i)^2} d\lambda
+ \mbox{non-singular}
\end{equation}
in the same neighborhood of $P_\alpha(T_i)$. Therefore,
if $\theta_{i\alpha} \not= 0$, $P_\alpha(T_i)$ is a pole
of $d\Omega_i$ of second order. These are all poles that
$d\Omega_i$ is required to have.
\subsubsection{Period Integrals of $d\Omega_i$}
The above conditions on poles determine the meromorphic
differential $d\Omega_i$ up to a difference of holomorphic
differential. We now select a symplectic homology basis
$A_I,B_I$ ($I = 1,\dots,g$), and put the standard
normalization condition
\begin{equation}
\oint_{A_I} d\Omega_i = 0, \quad I = 1,\dots,g.
\end{equation}
$d\Omega_i$ is thus uniquely determined.
\subsection{Back to Modulation Equation}
This is the end of the precise description of our modulation
equation. Let us finally reconfirm the roles of the three
sets of parameters $\theta_{i\alpha}$, $T_i$ and $h_{m\ell}$
in the defining equation of the spectral curve:
\begin{itemize}
\item The coadjoint orbit invariants $\theta_{i\alpha}$ are
constant parameters.
\item The positions of poles $T_i$ are ``time variables.''
\item The isospectral invariants $h_{m\ell}$ are ``dynamical
variables.''
\end{itemize}
\subsection{Remarks on Symplectic Homology Basis}
The choice of symplectic homology basis is a non-trivial
problem. The work of Flaschka, Forest and McLaughlin
\cite{bib:FFM} provides a typical example for considering
this problem. In their work, the meromorphic differentials
$d\Omega_i$ are normalized along the so called ``$\mu$-cycles.''
These cycles are homologous to trajectories of the
``auxiliary spectrum,'' which are nothing but the degree
$g$ divisor $\sum_{i=1}^g (\lambda_i,\zeta_i)$ of zeros of
a Baker-Akhiezer function on the KdV hyperelliptic spectral
curve
\begin{equation}
\zeta^2 = \prod_{i=1}^{2g+1} (\lambda - e_i).
\end{equation}
The Whitham averaging over the quasi-periodic motion on
the Jacobian variety eventually boils down to period
integrals along these trajectories. As this typical
example shows clearly, the normalization condition
of $d\Omega_i$ is by no means arbitrary, but determined
by the geometric structure of motion of a divisor on
the spectral curve. (In fact, as Ercolani, Forest and
McLaughlin noted \cite{bib:EFM}, this issue is already
considerably delicate in the case of the sine-Gordon
equation.)
Meanwhile, our formulation of the modulation equation
itself works for any choice of the symplectic homology
basis. Furthermore, the results of the subsequent
sections also hold irrespective of the choice of the
symplectic homology basis. Presumably, it is only a
subset of solutions of our modulation equation that
actually correspond to the true modulational description
of isomonodromic problems. This issue, too, forms
part of the hard analytical problems that we do not
pursue in this paper.
\section{Solutions of Modulation Equation}
\subsection{Period Integrals \`a la Seiberg-Witten}
In the following, let $h = \{ h_I \mid \ I=1,\cdots,g \}$
denote the $g$-tuple of isospectral invariants $h_{m\ell}$
($\ell = 2,\cdots,r$, $m = 0,\cdots,\delta_\ell - 1$)
(see Section 3). With the other parameters $\theta_{i\alpha}$
and $T_i$ being fixed, the spectral curve forms a $g$-dimensional
deformation family parametrized by these isospectral invariants.
This is the same situation as the Seiberg-Witten solution and
its various generalizations \cite{bib:SW-integrable}. One can
indeed derive the following analogous results.
\begin{itemize}
\item
A set of (local) coordinates on the $g$-dimensional moduli space
of spectral curves are given by the period integrals
\begin{equation}
a_I = \oint_{A_I} dS, \quad I = 1,\cdots,g.
\end{equation}
Here (and in the following), $ A_I, B_I$ ($I = 1,\cdots,g$)
are the same symplectic homology basis as used in the definition
of the modulation equation.
\item
Another set of (local) coordinates are given by the dual period
integrals
\begin{equation}
b_I = \oint_{B_I} dS, \quad I = 1,\cdots,g.
\end{equation}
In the Seiberg-Witten theory, they are denoted by $a^D_I$.
\item
There is a (locally defined) function $\calF = \calF(a)$
(``prepotential'') of $a = \{ a_I \}$ such that
\begin{equation}
\frac{\rd \calF}{\rd a_I} = b_I.
\end{equation}
\item
The second derivatives of $\calF$ coincide with matrix elements
of the period matrix,
\begin{equation}
\frac{\rd^2 \calF}{\rd a_I \rd a_J} = \calT_{IJ}
= \oint_{B_I} d\omega_J,
\end{equation}
where $d\omega_I$ ($I = 1,\cdots,g$) are a basis of holomorphic
differentials uniquely determined by the normalization condition
\begin{equation}
\oint_{A_I} d\omega_J = \delta_{IJ}.
\end{equation}
\end{itemize}
In a sense, the rest of this paper is devoted to verifying
these results in an $T$-dependent form.
The goal of this section is to show that the inverse period
map $a \mapsto h$ solves our deformation equation. We first
establish the invertibility of the period map $h \mapsto a$.
The inverse map $a \mapsto h$ then turns out to satisfy a
deformation equation of Seiberg-Witten type (with respect
to $a_i$'s), as well as our modulation equation (with respect
to $T_i$'s). Most results and proofs are a rather straight
forward generalization of those of the Seiberg-Witten
solution \cite{bib:SW-integrable}.
The notion of prepotential will be discussed in detail in
Section 8.
\subsection{Invertibility of Period Map}
This subsection is organized as follows. Firstly, we construct
a basis $d\omegatilde_I$ ($I = 1,\cdots,g$) of holomorphic
differentials as derivatives of $dS$ with respect to $h_I$'s.
Secondly, we examine the linear relations between this basis
and the normalized basis $d\omega_I$ ($I = 1,\cdots,g$). As a
corollary, (local) invertibility of the period map $h \to a$
follows.
\subsubsection{Basis of Holomorphic Differentials}
We here consider $\mu$ to be a (multivalued) function
$\mu = \mu(T,h,\lambda)$ of $T$, $h$ and $\lambda$, and
differentiate $dS(T,h,\lambda) = \mu(T,h,\lambda) d\lambda$
by $h_I$. This gives the differential
\begin{equation}
d\omegatilde_I = \frac{\rd}{\rd h_I} dS
= \frac{\rd \mu(T,h,\lambda)}{\rd h_I} d\lambda.
\end{equation}
We now show that this gives a basis of holomorphic differentials
on the spectral curve. The following reasoning is borrowed
from the work of Adams, Harnad and Hurtubise \cite{bib:AHH}.
This definition of $d\omegatilde_I$ can be cast into a more
tractable form as follows. First rewrite it in terms of
$\mutilde = f(\lambda) \mu$ (see Section 3):
\[
d\omegatilde_I
= \frac{1}{f(\lambda)}
\frac{\rd \mutilde(T,h,\lambda)}{\rd h_I} d\lambda.
\]
Differentiating the equation
\[
\Ftilde(\lambda,\mutilde)
= \det \Bigl( f(\lambda)M(\lambda) - \mu I \Bigr) = 0
\]
of the spectral curve in the $(\lambda,\mutilde)$ coordinates
gives
\[
\frac{\rd \mutilde}{\rd h_I}
= - \dfrac{\rd \Ftilde(\lambda,\mutilde)/\rd h_I}
{\rd \Ftilde(\lambda,\mutilde)/\rd \mutilde}.
\]
Now recall that $h_I$ is just an abbreviation of one of
$h_{m\ell}$'s in the following formula of Section 3:
\[
\Ftilde(\lambda,\mutilde) =
\Ftilde^0(\lambda,\mutilde)
+ f(\lambda) \sum_{\ell=2}^r \sum_{m=0}^{\delta_\ell}
h_{m\ell} \lambda^m \mutilde^{r-\ell}.
\]
Therefore, for $h_I = h_{m\ell}$,
\[
\frac{\rd \Ftilde(\lambda,\mutilde)}{\rd h_{m\ell}}
= f(\lambda) \lambda^\ell \mutilde^{r-m}.
\]
Thus, eventually, we obtain the following expression
of $d\omegatilde_I = d\omegatilde_{m\ell}$:
\begin{equation}
d\omegatilde_{m\ell}
= - \dfrac{\lambda^\ell \mutilde^{r-m}}
{\rd \Ftilde(\lambda,\mutilde)/\rd \mu} d\lambda.
\end{equation}
The last expression may be viewed as the Poincar\'e residue
of a meromorphic 2-form on the $(\lambda,\mutilde)$ plane
with pole divisor along the spectral curve. Since
$2 \le \ell \le r$ and $0 \le m \le \delta_\ell - 1$,
a standard argument leads to the following result:
\begin{proposition}
The differentials $d\omegatilde_I$ ($I = 1,\cdots,g$), form
a basis of holomorphic differentials on the spectral curve.
\end{proposition}
\subsubsection{Linear Relations between Two Bases}
Since $d\omegatilde_I$ and $d\omega_I$ both give a basis of
holomorphic differentials, they should be linked by an
invertible linear transformation:
\begin{equation}
d\omegatilde_I = \sum_{J=1}^g \calA_{IJ} d\omega_J.
\end{equation}
The matrix $\calA = (\calA_{IJ})$ of the coefficients is
invertible.
In fact, the matrix $\calA$ is given by the period integrals
\begin{equation}
\calA_{IJ} = \oint_{A_J} d\omegatilde_I,
\end{equation}
as one can readily see by integrating the above linear relation
of $d\omegatilde_I$ and $d\omega_J$ along $A_J$. Similarly,
integrating along $B_J$ yields the matrix relation
\begin{equation}
\calB = \calA \calT,
\end{equation}
where the matrix elements of $\calB = (\calB_{IJ})$ are given by
\begin{equation}
\calB_{IJ} = \oint_{B_J} d\omegatilde_I.
\end{equation}
In particular, the period matrix $\calT$ can be written
\begin{equation}
\calT = \calA^{-1} \calB.
\end{equation}
\subsubsection{Invertibility of Period Map}
We now show the following result, which shows that the matrix
is in fact the Jacobian matrix of the period map $h \mapsto a$.
\begin{proposition}
\begin{equation}
\calA_{IJ} = \dfrac{\rd a_J}{\rd h_I}.
\end{equation}
\end{proposition}
{\it Proof\/}. $a_J$ is defined by the period integral
\[
a_J = \oint_{A_J} dS.
\]
Its $h_I$-derivative can be calculates as follows:
\[
\dfrac{\rd a_J}{\rd h_I}
= \oint_{A_J} \dfrac{\rd}{\rd h_I} dS
= \oint_{A_J} d\omegatilde_I
= \calA_{IJ}.
\]
Q.E.D.
Since $\calA$ is an invertible matrix, we have:
\begin{corollary} The period map $h \mapsto a$ is (locally)
invertible.
\end{corollary}
\subsection{Solving Deformation Equations by Inverse Period Map}
We now take into account the variables $T = \{T_i\}$,
and consider the $I$-th coordinate of the inverse map
$a \mapsto h$ as a function of $T$ and $a$:
\begin{equation}
h_I = h_I(T,a).
\end{equation}
(In order to obtain these functions, one has to solve
the defining equation of the $a_I$'s for the parameters
$h_I$. This is by no means an easy task.) Accordingly,
$dS = \mu d\lambda$ becomes a meromorphic differential
depending on the parameters $(T,a)$:
\begin{equation}
dS = dS(T,a).
\end{equation}
We can now prove that $dS$ satisfies the modulation equation
with respect to $T$:
\begin{proposition}
$dS$ satisfies the modulation equation
\[
\frac{\rd}{\rd T_i} dS = d\Omega_i.
\]
\end{proposition}
{\it Proof\/}. Let us tentatively define
\[
d\Omegatilde_i = \frac{\rd}{\rd T_i} dS
= \frac{\rd \mu}{\rd T_i} d\lambda.
\]
We show that this differential satisfies all conditions
that characterize $d\Omega_i$. By the uniqueness, then,
$d\Omegatilde_i$ coincides with $d\Omega_i$, and the
proof is completed.
(i) {\it Locations of poles of $d\Omegatilde_i$\/}.
Differentiating the equation of the spectral curve
\[
F(\lambda,\mu) = \det \Bigl( M(\lambda) - \mu I \Bigr) = 0
\]
gives the relation
\[
\frac{\rd \mu}{\rd T_i}
= - \dfrac{\rd F(\lambda,\mu)/\rd T_i}
{\rd F(\lambda,\mu)/\rd \mu}.
\]
Therefore, apart from the poles of $\mu$, the derivative
$\rd \mu/\rd T_i$ can have poles at the zeros of
$\rd F/\rd \mu$ (i.e., at ramified points of $\pi$).
Let $e$ be the ramification index at a ramified point.
$\rd \mu/\rd T_i$ has a pole of order at most $e-1$ there.
This pole, however, is canceled by zeros of $d\lambda$
because $d\lambda$ has a zero of order $e - 1$ at the
same point. Meanwhile, in a neighborhood of each point of
$\pi^{-1}(\infty)$,
\[
\frac{\rd \mu}{\rd T_i}
= \left( \theta_{\infty\alpha} \lambda^{-2}
+ O(\lambda^{-2}) \right) d\lambda,
\]
so that the derivative $\rd \mu/\rd T_i$ cancels the
second order poles of $d\lambda$ at these points.
Thus, poles of $d\Omegatilde_i$ are limited to points
in $\pi^{-1}(\{T_1,\cdots,T_N\})$.
(ii) {\it Singular behavior of $d\Omegatilde_i$ at poles\/}.
We examine the singular behavior of $d\Omegatilde_i$ at each point
of $\pi^{-1}(T_j) = \{P_{j1},\cdots,P_{jr}\}$. Recall that,
in a neighborhood of $P_{i\alpha}$, $\mu$ behaves as
\[
\mu = - \dfrac{\theta_{i\alpha}}{\lambda - T_j}
+ \mbox{non-singular}.
\]
Therefore
\[
\frac{\rd \mu}{\rd T_i} = \left\{ \begin{array}{ll}
\mbox{non-singular} & (j \not= i) \\
- \dfrac{\theta_{i\alpha}}{(\lambda - T_i)^2}
+ \mbox{non-singular} & (j = i)
\end{array} \right.
\]
Consequently, $d\Omegatilde_i$ has poles only at the points over
$\lambda = T_i$, and exhibits there the same singular
behavior as $d\Omega_i$ does.
(iii) {\it Period Integrals of $d\Omegatilde_i$\/}. Since
the deformations leave $a_I$'s invariant, we have
\[
\oint_{A_I} \frac{\rd}{\rd T_i} dS
= \frac{\rd}{\rd T_i} \oint_{A_I} dS
= \frac{\rd a_I}{\rd T_i} = 0.
\]
Thus $d\Omegatilde_i$ satisfies the normalization condition of periods, too.
Q.E.D.
Similarly, $dS$ turns out to satisfy a deformation equation of
Seiberg-Witten type with respect to $a_I$'s:
\begin{proposition}
$dS$ satisfies the deformation equation
\[
\frac{\rd}{\rd a_I} dS = d\omega_I, \quad I = 1,\cdots,g,
\]
of Seiberg-Witten type.
\end{proposition}
{\it Proof\/}. This is just a consequence of the chain rule
of differentiation:
\[
\frac{\rd}{\rd a_I} dS
= \sum_J \frac{\rd h_J}{\rd a_I} \frac{\rd}{\rd h_J} dS
= \sum_J (\calA^{-1})_{IJ} d\omegatilde_J
= d\omega_I.
\]
Q.E.D.
Thus, we have been able to show that the inverse period map
$a \mapsto h$ solves the coupled deformation equations
\begin{equation}
\frac{\rd}{\rd T_i} dS = d\Omega_i, \quad
\frac{\rd}{\rd a_I} dS = d\omega_I.
\end{equation}
(Reconfirm, once again, that the differentiation is
understood to leave $\lambda$ constant, i.e.,
$\rd \lambda/\rd T_i = \rd \lambda/\rd a_I = 0$.)
\subsection{Remarks on Structure of $dS$}
It seems remarkable that although our meromorphic
differential $dS$ has simple poles with non-zero residues,
their derivatives in deformation variables are holomorphic
differentials or meromorphic differentials of second kind
(i.e., meromorphic differentials with higher order poles
and no residue). This is reminiscent of the structure of
the Seiberg-Witten meromorphic differential in
supersymmetric gauge theories coupled with matter fields
\cite{bib:SW-integrable}.
Because of this fact, $dS$ cannot be written as a linear
combination of $d\omega_I$ and $d\Omega_i$; we need extra
meromorphic differentials of the third kind, say $d\Pi$,
with the same singularity structure as $dS$ but with
vanishing $A_I$-cycles. $dS$ can be written
\begin{equation}
dS = \sum_{I=1}^g a_I d\omega_I + d\Pi.
\end{equation}
Of course $d\Pi$ can be further decomposed into a sum of
more elementary meromorphic differentials, but can never
be a linear combination of $d\Omega_i$'s.
\section{Prepotential}
\subsection{Definition of Prepotential}
The notion of prepotential has been formulated in a quite
general framework by Krichever \cite{bib:Krichever-Whitham}
and Dubrovin \cite{bib:Dubrovin-Whitham}.
Following their formalism we can now define a prepotential
$\calF = \calF(T,a)$ by the following differential equation.
\begin{proposition}
The following differential equations is integrable in the
sense of Frobenius:
\begin{equation}
\frac{\rd \calF}{\rd a_I} = \oint_{B_I} dS, \quad
\frac{\rd \calF}{\rd T_i}
= \sum_{\alpha=1}^r \Res_{P_{i\alpha}}
\dfrac{\theta_{i\alpha}}{\lambda - T_i} dS.
\end{equation}
Here $P_{i\alpha}$, $\alpha = 1,\cdots,r$, denote the points
of $\pi^{-1}(T_i)$ such that
\begin{equation}
\mu = \dfrac{\theta_{i\alpha}}{\lambda - T_i}
+ \mbox{non-singular}
\end{equation}
in a neighborhood of $P_{i\alpha}$.
\end{proposition}
\subsection{Equivalent Definition}
Before proving the above result, we note here that the
defining equations of $\calF$ can be rewritten in the
following more compact form:
\begin{equation}
\frac{\rd \calF}{\rd a_I} = b_I, \quad
\frac{\rd \calF}{\rd T_i} = H_i.
\end{equation}
The equivalence of the first equation is obvious from the
definition of $b_I$. Let us verify the equivalence of the
second equation in some detail. We first rewrite the
definition of $H_i$ into a contour integral
\[
H_i = \frac{1}{2\pi\sqrt{-1}}
\oint_{|\lambda - T_i| = \delta}
\frac{1}{2} \Tr M(\lambda)^2 d\lambda
\]
along a sufficiently small circle (of radius $\delta$)
around $\lambda = T_i$. Let $\mu_\alpha$,
$\alpha = 1,\cdots,r$, denote the eigenvalues of
$M(\lambda)$; they correspond to the $r$ sheets of
the spectral curve in a neighborhood of $\pi^{-1}(T_i)$.
Expressing the trace of $M(\lambda)^2$ in terms of
these eigenvalues, we can rewrite the above integral
formula as:
\begin{eqnarray*}
H_i
&=& \frac{1}{2\pi\sqrt{-1}}\oint_{|\lambda - T_i| = \delta}
\sum_{\alpha=1}^r \frac{\mu_\alpha^2}{2} d\lambda \\
&=& \sum_{\alpha=1}^r
\frac{1}{2\pi\sqrt{-1}}\oint_{C_{i\alpha}}
\frac{\mu^2}{2} d\lambda.
\end{eqnarray*}
Now recall that $\mu$ has a Laurent expansion of the form
\[
\mu = \dfrac{\theta_{i\alpha}}{\lambda - T_i}
+ c_{i\alpha,0} + c_{i\alpha,1}(\lambda - T_i)
+ \cdots
\]
in a neighborhood of $P_{i\alpha}$. From this fact,
we can easily derive the relation
\[
\oint_{C_{i\alpha}} \frac{\mu^2}{2} d\lambda
= \oint_{C_{i\alpha}}
\dfrac{\theta_{i\alpha}}{\lambda - T_i} dS.
\]
This relation and the above expression of $H_i$ give
\[
H_i = \sum_{\alpha=1}^r \Res_{P_{i\alpha}}
\dfrac{\theta_{i\alpha}}{\lambda - T_i} dS.
\]
This leads to the second expression of the defining
equation of the prepotential.
\subsection{Proof of Proposition}
The idea of proof is the same as the standard one based on
Riemann's bilinear relation. Integrability conditions to be
checked are the following:
\begin{equation}
\dfrac{\rd H_j}{\rd T_i} = \dfrac{\rd H_i}{\rd T_j}, \quad
\dfrac{\rd H_j}{\rd a_I} = \dfrac{\rd b_I}{\rd T_j}, \quad
\dfrac{\rd b_J}{\rd a_I} = \dfrac{\rd b_I}{\rd a_J}.
\end{equation}
Here $b_I$, $b_J$, $H_i$ and $H_j$ are used for simplifying
notations; in the verification below, we have to return to the
integral expressions of these quantities.
These three relations can be verified in much the same way.
We show the derivation of the second relation in detail.
First, differentiating the integral formula
\[
H_j = \sum_\alpha \frac{1}{2\pi\sqrt{-1}} \oint_{C_{j\alpha}}
\dfrac{\theta_{j\alpha}}{\lambda - T_j} dS
\]
by $a_I$ gives
\[
\dfrac{\rd H_j}{\rd a_I}
= \sum_\alpha \frac{1}{2\pi\sqrt{-1}} \oint_{C_{j\alpha}}
\dfrac{\theta_{j\alpha}}{\lambda - T_j} d\omega_I.
\]
Similarly, from the integral formula of $b_I$,
\[
\dfrac{\rd b_I}{\rd T_j}
= \oint_{B_I} d\Omega_j.
\]
By Riemann's bilinear relation, $d\Omega_j$ and $d\omega_I$
satisfies the relation
\[
(*) \quad
\frac{1}{2\pi\sqrt{-1}} \oint_{\rd\Delta} \Omega_j d\omega_I
= \sum_J \left(
\oint_{A_K} d\Omega_j \oint_{B_K} d\omega_I -
\oint_{B_K} d\Omega_j \oint_{A_K} d\omega_I \right).
\]
Here $\Delta$ is a simply connected surface with boundary
($4g$-gon) obtained in a standard way by cutting the Riemann
surface of the spectral curve along $2g$ paths. The boundary
$\rd\Delta$ is oriented in the direction encircling interior
points anti-clockwise. We now evaluate both hand sides.
As for the left hand side of $(*)$, $\Omega_j$ has poles at
the points $P_{j1},\cdots,P_{jr}$ in $\Delta$ whereas
$d\omega_I$ has of course no pole. Therefore the contour
integral along $\rd\Delta$ splits into a sum of contour
integrals along $C_{j1},\cdots,C_{jr}$. Furthermore, the
singular behavior of $\Omega_j$ in a neighborhood of
$P_{j\alpha}$ is such that
\[
\Omega_j = - \theta_{j\alpha}/(\lambda - T_j)
+ \mbox{non-singular}.
\]
Therefore
\[
\mbox{LHS of $(*)$}
= - \sum_\alpha \frac{1}{2\pi\sqrt{-1}} \oint
\dfrac{\theta_{j\alpha}}{\lambda - T_j} d\omega_I.
\]
Now consider the the right hand side of $(*)$. The integrals
of $d\Omega_j$ along $A_K$'s
all vanish by the normalization conditions of $d\Omega_j$.
Its integrals along $B_K$'s do not vanish in general, but
now the integral of $d\omega_I$ along $A_K$'s vanish
except for $K = I$; in the case of $K = I$, the latter
integral is equal to $1$ by the normalization conditions
of $d\omega_I$. Thus
\[
\mbox{RHS of $(*)$} = - \oint_{B_I} d\Omega_j.
\]
These relations imply the second integrability condition.
In the verification of the first integrability condition,
it is convenient to express $d\Omega_i$ as a sum of
meromorphic differentials $d\Omega_{i\alpha}$ with just
one pole at $P_{i\alpha}$ and normalized by the same
condition of vanishing $A$-periods. The integrability
condition can then be reduced to Riemann's bilinear
relation for $d\Omega_{i\alpha}$ and $d\Omega_{j\beta}$.
The third integrability condition is the easiest to
verify. This case, however, is the most interesting
in the context of the Seiberg-Witten theory.
Differentiating $b_J$ now by $a_I$ gives
\[
\dfrac{\rd b_J}{\rd a_I} = \oint_{B_J} d\omega_I
= \calT_{IJ}.
\]
This is equal to $\rd b_I/\rd a_J$, because the period
matrix $\calT$ is symmetric. (One can also directly
deduce this conclusion by applying Riemann's bilinear
relation to $d\omega_I$ and $d\omega_J$. This is
indeed a usual proof of the relation
$\calT_{IJ} = \calT_{JI}$ !)
This completes the proof of the proposition.
As a corollary of the proposition and the final part of
the proof, we obtain the following result:
\begin{corollary}
The second $a$-derivatives of $\calF = \calF(T,a)$
coincide with the matrix elements of the period matrix:
\begin{equation}
\dfrac{\rd^2 \calF}{\rd a_I \rd a_J} = \calT_{IJ}.
\end{equation}
\end{corollary}
Thus our treatment of the prepotential incorporates
all essential aspects of the prepotential of
Seiberg-Witten type.
\subsection{Final Remarks}
(i) The definition of prepotential shows a link with the
notion of $\tau$ function. Recall that $H_i$ is equal to
a logarithmic derivative of the $\tau$ function of the
Schlesinger equation. In the $\epsilon$-dependent
formulation,
\begin{equation}
H_i = \epsilon \frac{\rd \log \tau}{\rd T_i}.
\end{equation}
Note, however, that $\log\tau$ and $\calF$ are by no means
in a simple proportional relation, because some averaging
operation intervenes.
(ii) Unlike the prepotentials in topological conformal
field theories and the Seiberg-Witten theory,
our prepotential $\calF$ seems to possess no manifest
homogeneity of degree two. This issue seems to be
linked with the unusual structure of $dS$ that
we pointed out in the end of the last section.
Presumably our prepotential (and the modulation
equation) will have hidden homogeneity, which emerges
after introducing more deformation variables.
\section{Conclusion}
The body of this paper consists of two part. The
first part is concerned with the derivation of the
modulation equation. This equation is expected to
describe slow dynamics of the spectral curve in
isospectral approximation, in the sense of Garnier,
of the ($\epsilon$-dependent) Schlesinger equation.
Although our method for deriving this modulation
equation is heuristic, we believe that this gives
a correct answer (under a suitable choice of the
symplectic homology basis). The second part of
this paper is devoted to a complete description of
solutions of the modulation equation, as well as the
notion of prepotential. We have been able to obtain
the following fundamental results:
\begin{itemize}
\item General solutions are obtained by an inverse period
map. The period map is given by period integrals of
Seiberg-Witten type.
\item The modulation equation can coexist with a deformation
equation of Seiberg-Witten type. They altogether form a
commuting set of flows on the $g$-dimensional moduli
space of spectral curves.
\item A prepotential can be defined on this extended commuting
flows. This construction is parallel to already known
examples of prepotentials.
\end{itemize}
These results, too, strongly support the validity of the
modulation equation.
A number of problems are left unanswered. The most crucial
is of course the issue of validity of our derivation of
modulation equation. We have presented a few fragmental
ideas in this direction. Krichever's averaging method
seems to be the most powerful and universal approach to
this problem; we, however, have no idea how to calculate
the averaged quantities to obtain period integrals.
Presumably, Vereschagin's method will be a hint to this
issue. Also, we would like to stress that generalizing
the averaging calculation of Flaschka, Forest and
McLaughlin to non-hyperelliptic spectral curves is still
a very important problem; the spectral Darboux coordinates
of the Montreal group should be used in place of the
classical auxiliary spectrum.
The methods presented in this paper will be generalized
to other types of isomonodromic problems. An immediate, but
also interesting generalization is the case with irregular
singular points. For instance, if an irregular singular
point of Poincar\'e rank one is added to $\lambda = \infty$,
the resulting isomonodromic problem becomes the so called
JMMS equation \cite{bib:JMMS}. This equations exhibits a
remarkable ``duality'' \cite{bib:Harnad-dual}, which will
be inherited to the modulation equation. This issue will
be discussed in a separate paper \cite{bib:takasaki-dual}.
An even more interesting direction is a generalization
to isomonodromic problems on an elliptic (and higher genus)
Riemann surface. Examples of such isomonodromic problems
have been constructed by Okamoto's group in a geometric
framework \cite{bib:nonzero-genus}. Recently, Levin and
Olshanetsky presented another framework based on the
method of Hamiltonian reduction, and pointed out a link
with the Whitham equation \cite{bib:levin-olshanetsky}.
\begin{thebibliography}{99}
\bibitem{bib:RFuchs}
R. Fuchs,
\"Uber lineare homogene Differentialgleichungen
zweiter ordnung mit drei in Endlich gelegene wesentlich
singul\"aren Stellen,
Math. Ann. {\bf 63} (1907), 301-321.
\bibitem{bib:Garnier1912}
R. Garnier,
Sur des \'equations diff\'erentielles du troisi\`eme ordre
dont l'int\'egral g\'en\'erale et uniforme et sur une class
d'\'equations nouvelles d\'odre sup\'erieur dont l'int\'egrale
g\'en\'erale a ses points critiques fixes,
Ann. Sci. \'Ecole Norm. Sup. (3) {\bf 29} (1912), 1-126.
\bibitem{bib:Schlesinger}
L. Schlesinger,
\"Uber eine Klasse von Differentialsystemen beliebiger
Ordnung mit festen kritischen Punkten,
J. f\"ur Math. {\bf 141} (1912), 96-145.
\bibitem{bib:Garnier1926}
R. Garnier,
Solution du probl\`eme de Riemann pour les syst\`emes
diff\'erentiels li\'eaires du second ordre,
Ann. Sci. \'Ecole Norm. Sup. (3) {\bf 43} (1926), 177-307.
\bibitem{bib:Garnier1917}
R. Garnier,
Etude de l'int\'egrale g\'en\'erale de l'\'equation VI
de M. Painlev\'e dans le voisinage de ses singularit\'e
transcendantes,
Ann. Sci. \'Ecole Norm. Sup. (3) {\bf 34} (1917), 239-353.
\bibitem{bib:Moser}
J. Moser,
Geometry of quadrics and spectral theory,
in: {\it Chern Symposium, Berkeley, 1979\/}, pp. 147-188
(Springer-Verlag, Berlin, Heidelbeg, New York, 1980).
\bibitem{bib:Munford}
D. Munford,
Tata lectures on theta, vol. II
(Birkh\"auser, Boston, 1983).
\bibitem{bib:AHHP}
M.R. Adams, J. Harnad and E. Previato,
Isospectral Hamiltonian flows in finite and
infinite dimensions, I,
Commun. Math. Phys. {\bf 117} (1988), 451-500.
\newline
M.R. Adams, J. Harnad and J. Hurtubise,
{\it ditto} II,
Commun. Math. Phys. {\bf 134} (1990), 555-585.
\bibitem{bib:AHH}
M.R. Adams, J. Harnad and J. Hurtubise,
Darboux coordinates and Liouville-Arnold integration
in loop algebras,
Commun. Math. Phys. {\bf 155} (1993), 385-413.
\bibitem{bib:Beauville}
A. Beauville,
Jacobiennes des courbes spectrales et syst\`emes
hamiltoniens compl\`etement int\'egrables,
Acta Math. {\bf 164} (1990), 211-235.
\bibitem{bib:Harnad-dual}
J. Harnad,
Dual isomonodromic deformations and moment maps
to loop algebras,
Commun. Math. Phys. {\bf 166} (1994), 337-365.
\bibitem{bib:HTW}
J. Harnad, C.A. Tracy and H. Widom,
Hamiltonian structure of equations appearing in
Random matrices,
in: {\it Low Dimensional Topology and Quantum Field Theory\/}
(ed. H. Osborn), pp. 231-245 (Plenum, New York, 1993).
\bibitem{bib:HR}
J. Harnad and M. Routhier,
R-matrix construction of electromagnetic models for
Painlev\'e transcendents,
J. Math. Phys. {\bf 36} (1995), 4863-4881.
\bibitem{bib:HW}
J. Harnad and M.-A. Wisse,
Loop algebra moment maps and Hamiltonian models
for the Painlev\'e transcendents,
in: {\it Mechanics Day workshop Proceedings\/}
(eds. P.S. Krishnaprasad et al.),
AMS-Fields Inst. Commun. {\bf 7} (1996), 155-169.
\bibitem{bib:Boutroux}
P. Boutroux,
Recherches sur les transcendantes des M. Painlev\'e
et l'\'etude asymptotique des \'equations diff\'erentielles
du second ordre,
Ann. Sci. Ecole Norm. Sup. (3) {\bf 30} (1913
\bibitem{bib:FN}
H. Flaschka and A. Newell,
Monodromy- and spectrum-preserving deformations, I,
Commun. Math. Phys. {\bf 76} (1980), 65-116.
\newline
H. Flaschka and A. Newell,
Multiphase similarity solutions of integrable
evolution equations,
Physica {\bf 3D} (1981), 203-221.
\bibitem{bib:Moore-review}
G. Moore,
Matrix models of 2D gravity and isomonodromic deformations,
Prog. Theor. Phys. Suppl. {\bf 102} (1990), 255-285.
\bibitem{bib:Novikov}
S.P. Novikov,
Quantization of finite-gap potentials potentials and
nonlinear quasiclassical approximation in nonperturbative
string theory,
Funct. Anal. Appl. {\bf 24} (1990), 296-306.
\bibitem{bib:Krichever-ETH}
I.M. Krichever,
On Heisenberg relations for the ordinary linear
differential operators,
ETH Z\"urich preprint, June 1990.
\bibitem{bib:Fucito-etal}
F. Fucito, A. Gamba, M. Martellini and O. Ragnisco,
Nonlinear WKB analysis of the string equation,
Int. J. Mod. Phys. {\bf B6} (1992), 2123-2148.
%%% hep-th/9112074.
\bibitem{bib:Vereschagin-string}
V.L. Vereschagin,
Asymptotics of solutions of the discrete string equation,
Physica {\bf D95} (1996), 268-282.
\bibitem{bib:Whitham}
G.B. Whitham,
Linear and Nonlinear Waves
(Wiley-Interscience, New York, 1974).
\bibitem{bib:D-Maslov}
S.Yu. Dobrokhotov and V.P. Maslov,
Finite-zone, almost-periodic solutions in WKB approximations,
J. Soviet Math. {\bf 16}, No. 6 (1981), 1433-1487.
\bibitem{bib:DMN}
B.A. Dubrovin, V.B. Matveev and S.P. Novikov,
Nonlinear equations of Korteweg-de Vries type,
finite band operators and Abelian varieties,
Russian Math. Surveys {\bf 31}, No. 1 (1976), 59-146.
\bibitem{bib:Krichever1977}
I.M. Krichever,
Methods of algebraic geometry in the theory
of nonlinear equations,
Russian Math. Surveys {\bf 32}, No. 6 (1977), 185-214;
Integration of nonlinear equations by the
method of algebraic geometry,
Funct. Anal. Appl. {\bf 11} (1977), 12-26.
\bibitem{bib:DN-review}
B.A. Dubrovin and S.P. Novikov,
Hydrodynamics of weakly deformed soliton lattices:
Differential geometry and Hamiltonian theory,
Russian Math. Surveys {\bf 44}, No. 6 (1989), 35-124.
\bibitem{bib:FFM}
H. Flaschka, M.G. Forest and D.W. McLaughlin,
Multiphase averaging and the inverse spectral solution
of the Korteweg-de Vries equation,
Comm. Pure Appl. Math. {\bf 33} (1980) 739-784.
\bibitem{bib:Krichever-Whitham}
I.M. Krichever, The dispersionless Lax equations and
topological minimal models,
Commun. Math. Phys. {\bf 143} (1992), 415-429;
The $\tau$-function of the universal Whitham hierarchy,
matrix models and topological field theories,
Comm. Pure. Appl. Math. {\bf 47} (1994), 437-475.
\bibitem{bib:Dubrovin-Whitham}
B.A. Dubrovin,
Hamiltonian formalism of Whitham hierarchies and
topological Landau-Ginzburg models,
Commun. Math. Phys. {\bf 145} (1992), 195-207;
Geometry of 2D topological field theories,
Lecture Notes in Math. {\bf 1620}, pp. 120-348
(Springer-Verlag, Berlin, Heidelbeg, New York, 1996).
%%% hep-th/9407018.
\bibitem{bib:TN}
K. Takasaki and T. Nakatsu,
Isomonodromic deformations and supersymmetric
Yang-Mills theories,
Int. J. Mod. Phys. A 11 (38) (1996), 5505-5518.
%%% KUCP-0092, hep-th/9603069.
\bibitem{bib:SW-integrable}
A. Gorsky, I. Krichever, A. Marshakov, A. Mironov and A. Morozov,
Integrability and exact Seiberg-Witten solution,
Phys. Lett. {\bf 335B} (1995), 466-474. %%% hep-th/9505035.
\newline
E. Martinec and N. Warner,
Integrable systems and supersymmetric gauge theories,
Nucl. Phys. {\bf B459} (1996), 97-112. %%% hep-th/9509161.
\newline
T. Nakatsu and K. Takasaki,
Whitham-Toda hierarchy and $N=2$ supersymmetric Yang-Mills theory,
Mod. Phys. Lett. {\bf A11} (1996), 157-168. %%% hep-th/9509162.
\newline
R. Donagi and E. Witten,
Supersymmetric Yang-Mills theory and integrable systems,
Nucl. Phys. {\bf B460} (1996), 299-334. %%% hep-th/9510101.
\newline
T. Eguchi and S.-K. Yang,
Prepotentials of N = 2 supersymmetric gauge theories and
soliton equations,
Mod. Phys. Lett. {\bf A11} (1996), 131-138. %%% hep-th/9510183.
\newline
E. Martinec,
Integrable structures in supersymmetric gauge and string theory,
Phys. Lett. {\bf B367} (1996), 91-96. %%% hep-th/9510204.
\newline
A. Gorsky and A. Marshakov,
Towards effective topological theories on spectral curves,
hep-th/9510224.
\newline
H. Itoyama and A. Morozov,
Integrability and Seiberg-Witten theory. Curves and periods,
Nucl. Phys. {\bf B477} (1996), 855-877. %%% hep-th/9511126;
Prepotential and the Seiberg-Witten theory,
Nucl. Phys. {\bf B491} (1997), 529-573. %% hep-th/9512161.
\newline
A. Marshakov,
Exact solutions to quantum field theories and integrable
equation,
Mod. Phys. Lett. {\bf A11} (1996), 1169-1184. %%% hepth-/9602005.
\newline
C. Ahn and S. Nam,
Integrable structure in supersymmetric gauge theories
with massive hypermultiplets,
Phys. Lett. {\bf B387} (1996), 304-309. %%% hep-th/9603028.
\newline
A. Marshakov,
On integrable systems and supersymmetric gauge theories,
hep-th/9702083.
\bibitem{bib:JMMS}
M. Jimbo, T. Miwa, Y. M\^ori and M. Sato,
Density matrix of an impenetrable Bose gas and
the fifth Painlev\'e transcendents,
Physica {\bf 1D} (1980), 80-158.
\bibitem{bib:JMU}
M. Jimbo, T. Miwa and K. Ueno,
Monodromy preserving deformation of linear ordinary
differential equations with rational coefficients, I,
Physica {\bf 2D} (1981), 306-352.
\bibitem{bib:JM}
M. Jimbo and T. Miwa,
Monodromy preserving deformation of linear ordinary
differential equations with rational coefficients, II,
Physica {\bf 2D} (1981), 407-448;
{\it ditto} III, {\it ibid} {\bf 4D} (1981), 26-46.
\bibitem{bib:Vereschagin-painleve}
V.L. Vereschagin,
Nonlinear quasiclassics and Painlev\'e equations,
hep-th/9605092.
\bibitem{bib:Okamoto}
K. Okamoto,
Isomonodromic deformations and Painlev\'e equations,
and the Garnier system,
J. Fac. Sci. Univ. Tokyo, Sect. IA, Math,
{\bf 33} (1986), 575-612.
\bibitem{bib:multiscale}
A. Newell and J. Whitehead,
Finite bandwidth, finite amplitude convection,
J. Fluid Mech. {\bf 38} (1969), 279-303.
\newline
J.D. Gibbon and M.J. McGuiness,
Amplitude equations at the critical points of
unstable dispersive physical systems,
Proc. R. Soc. London {\bf A377} (1981), 185-219.
\newline
A. Newell,
The dynamics of patterns: A survey,
in: {\it Propagation in Systems far from Equilibrium\/}
(ed. J.E. Wesfeld), Les Houches 1987
(Springer-Verlag, Berlin, Heidelbeg, New York, 1988).
\bibitem{bib:Krichever-averaging}
I.M. Krichever,
Method of averaging for two-dimensional ``integrable'' equations,
Funct. Anal. Appl. {\bf 22} (1988), 200-213;
Spectral theory of two-dimensional periodic operators
and its applications,
Russian Math. Surveys {\bf 44}, No. 2 (1989), 145-225.
\bibitem{bib:Krichever-Phong}
I.M. Krichever and D.H. Phong,
On the integrable geometry of soliton equations and
N = 2 supersymmetric gauge theories,
J. Diff. Geom. {\bf 45} (1997), 349-389.
%%% hep-th/9604119
\bibitem{bib:Carroll-Chang}
R. Carroll and J.-H. Chang,
The Whitham equations revisited,
solv-int/9612010.
\bibitem{bib:EFM}
M.G. Forest and D.W. McLaughlin,
Modulations of sinh-Gordon and sine-Gordon wavetrains,
Stud. Appl. Math. {\bf 68} (1983), 11-59.
\newline
N.M. Ercolani, M.G. Forest and D.W. McLaughlin,
Modulational stability of two-phase sine-Gordon wavetrains,
Stud. Appl. Math. {\bf 71} (1984), 91-101.
\bibitem{bib:takasaki-dual}
K. Takasaki,
Dual isomonodromic problems and Whitham equations,
solv-int/9704004.
\bibitem{bib:nonzero-genus}
K. Okamoto,
On Fuchs's problem on a torus,
I. Funkcial. Ekvac. {\bf 14} (1971), 137-152;
{\it ditto} II,
J. Fac. Sci. Univ. Tokyo, Sect. IA, {\bf 24} (1977), 357-371;
D\'eformation d'une \'equation differ\'entielle lin\'eaire
avec une singularit\'e irr\'eguli\`ere sur un tore,
{\it ibid\/}, {\bf 26} (1979), 501-518.
\newline
K. Iwasaki,
Moduli and deformation for Fuchsian projective connections
on a Riemann surface,
J. Fac. Sci. Univ. Tokyo, Sect. IA, {\bf 38} (1991), 431-531;
Fuchsian moduli on a Riemann surface --- its Poisson
structure and Poincar\'e-Lefschetz duality,
Pacific J. Math. {\bf 155} (1992), 319-340.
\newline
S. Kawai,
Deformations of complex structures on a torus and
monodromy preserving deformations,
preprint.
\bibitem{bib:levin-olshanetsky}
A.M. Levin and M.A. Olshanetsky,
Classical limit of the Knizhnik-Zamolodchikov-Bernard
equations as hierarchy of isomonodromic deformations:
Free field apoproach, hep-th/9709207.
\end{thebibliography}
\end{document}