Dispersionless Hierarchies, Hamilton-Jacobi Theory and Twistor Correspondences
Extracted source text ยท arXiv solv-int/9705013
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%% Title: Dispersionless Hierarchies, Hamilton-Jacobi Theory
%%%% and Twistor Correspondences
%%%% Authors: Partha Guha and Kanehisa Takasaki
%%%% Comments: 20 pages, latex, no figures
%%%% Report-no: RIMS-1124
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\documentstyle[12pt]{article}
\renewcommand{\baselinestretch}{1.4}
\setlength{\oddsidemargin}{0cm}
\setlength{\evensidemargin}{0cm}
\setlength{\topmargin}{0cm}
\setlength{\textheight}{22.5cm}
\setlength{\textwidth}{16cm}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newtheorem{theorem}{Theorem}
\newenvironment{acknowledge}%
{\bigskip\noindent{\bf Acknowledgements}\bigskip}%
{\bigskip}
%
\newcommand{\beqn}{\begin{equation}}
\newcommand{\eeqn}{\end{equation}}
\newcommand{\beqnarray}{\begin{eqnarray}}
\newcommand{\eeqnarray}{\end{eqnarray}}
%
\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{\cB}{{\cal B}}
\newcommand{\cBbar}{{\bar{\cB}}}
\newcommand{\bC}{{\bf C}}
\newcommand{\Edot}{{\dot{E}}}
\newcommand{\cF}{{\cal F}}
\newcommand{\cL}{{\cal L}}
\newcommand{\cLbar}{{\bar{\cL}}}
\newcommand{\cM}{{\cal M}}
\newcommand{\cMbar}{{\bar{\cM}}}
\newcommand{\cO}{{\cal O}}
\newcommand{\cP}{{\cal P}}
\newcommand{\Pbar}{{\bar{P}}}
\newcommand{\cQ}{{\cal Q}}
\newcommand{\qbar}{{\bar{q}}}
\newcommand{\Sbar}{{\bar{S}}}
\newcommand{\cT}{{\cal T}}
\newcommand{\tbar}{{\bar{t}}}
\newcommand{\ubar}{{\bar{u}}}
\newcommand{\vbar}{{\bar{v}}}
\newcommand{\lambdabar}{{\bar{\lambda}}}
\newcommand{\mubar}{{\bar{\mu}}}
\newcommand{\xibar}{{\bar{\xi}}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\date{ }
\title{Dispersionless Hierarchies, Hamilton-Jacobi Theory
and Twistor Correspondences}
\author{
Partha Guha$^1$
\footnote{On leave from S.N. Bose National Center
for Basic Science, Salt-Lake, Culcutta-91, India}
and Kanehisa Takasaki$^2$\\
\\
$^1${\normalsize
Research Institute for Mathematical Sciences, Kyoto University}\\
{\normalsize Sakyo-ku, Kyoto 606, Japan}\\
$^2${\normalsize
Department of Fundamental Sciences, Kyoto University}\\
{\normalsize Sakyo-ku, Kyoto 606, Japan}
}
\maketitle
\bigskip
\begin{center}
{\it Dedicated to Sir Roger Penrose on his 65th birthday}
\end{center}
\bigskip
\begin{abstract}
\noindent The dispersionless KP and Toda hierarchies possess
an underlying twistorial structure. A twistorial approach is
partly implemented by the method of Riemann-Hilbert problem.
This is however still short of clarifying geometric ingredients
of twistor theory, such as twistor lines and twistor surfaces.
A more geometric approach can be developed in a Hamilton-Jacobi
formalism of Gibbons and Kodama. \\
{\bf AMS Subject Classifiation} (1991): 35Q20, 58F07,70H99
\end{abstract}
\newpage
\section{Introduction}
The dispersionless KP and Toda hierarchies are the
most typical dispersionless integrable hierarchies.
The ordinary KP and Toda hierarchies have a Lax
representation of the form
\beqn
\dfrac{\rd L}{\rd t_n} = [B_n, L], \quad
\mbox{etc} \ldots
\eeqn
where $L,B_n$, etc. are pseudo-differential or difference
operators. Their dispersionless limit is a kind of
``quasi-classical'' limit in which $L,B_n$, etc. are
phase space functions and the commutator $[\ ,\ ]$ is
replaced by a Poisson bracket $\{\ ,\ \}$. This causes
a drastic change, especially in the Toda hierarchy --
the one-dimensional lattice in the Toda hierarchy turns
into a continuous line. The dispersionless hierarchies
nevertheless inherit many aspects of integrability from
the original KP and Toda hierarchies, such as the notion
of $\tau$ functions, Hirota equations, infinite-dimensional
symmetries, etc. (For an overview, we refer to the review
\cite{takasaki-takebe-95}.)
In some aspects, however, the dispersionless hierarchies
require an entirely new approach. One of them is the problem
of constructing special solutions. A standard recipe is to
consider a suitable ``reduction'' of the original infinite
hierarchy into a system with a finite number of unknown
functions. Usually, such a reduced system of the dispersionless
hierarchies is a hydrodynamic system, and solvable by a
generalized ``hodographic'' method as developed by Tsarev
\cite{tsarev-85}. Many special solutions of the dispersionless
KP and Toda hierarchies have been indeed constructed by
Gibbons and Kodama \cite{kodama-88,gibbons-kodama-89,kodama-90}
by a generalized hodographic method.
Proof of integrability of the dispersionless hierarchies
themselves, too, has to be established by a new method.
Such an approach is provided by the method of Riemann-Hilbert
problem \cite{takasaki-takebe-91,takasaki-takebe-92a}. This
Riemann-Hilbert problem is discovered as an analogue of the
Riemann-Hilbert problem for the four-dimensional self-dual
Einstein equation \cite{boyer-plebanski-85} and an associated
infinite hierarchy \cite{takasaki-90}. In the case of the
self-dual Einstein equation, the Riemann-Hilbert problem is
an analytic expression of Penrose's curved twistor theory
\cite{penrose-76}. Thus the method of Riemann-Hilbert method
may be called a twistorial approach to the dispersionless
hierarchies. (For more details on this analogy, we refer to
the review \cite{takasaki-92}.)
Although thus nicely exhibiting a link with twistor theory,
however, the previous approach by the Riemann-Hilbert problem
\cite{takasaki-takebe-91,takasaki-takebe-92a}
is still lacking a geometric language. Let us recall that
a central idea of twistor theory is to relate a space-time
manifold $M$ with a complex manifold $\cT$ (twistor space) by
the ``twistor correspondence'', i.e., a manifold $\cF$ with a
double fibration $\pi_1: \cF \to M$ and $\pi_2: \cF \to \cT$.
By this correspondence, a space-time point $x \in M$
determines (and is determined by) a rational curve (twistor
line) $\pi_2(\pi_1^{-1}(x))$ in the twistor space, and a
twistor point $\xi \in \cT$ a submanifold (twistor surface)
$\pi_1(\pi_2^{-1}(\xi))$ of the space-time. These notions
have been left obscure in the case of the dispersionless
hierarchies.
The goal of this paper is to establish a dictionary between
the twistor geometry and the method of Riemann-Hilbert problem
for both the dispersionless KP and Toda hierarchy. It turns
out that the Hamilton-Jacobi approach of Gibbons and Kodama
\cite{kodama-gibbons-90,gibbons-kodama-94}, another proof of
integrability of the dispersionless KP hierarchy, provides
such a geometric interpretation of the method of Riemann-Hilbert
problem. In Section 2, we review the previous results on the
dispersionless KP hierarchy, and point out the problem.
In Section 3, we reformulate the Hamilton-Jacobi approach
of Gibbons and Kodama in our language, and present the
twistorial interpretation. Section 4 is devoted to a
similar treatment of the dispersionless Toda hierarchy.
This part may be read as an independent result on an
extension of the approach of Gibbons and Kodama.
Section 5 is added for comments on a relation to Frobenius
structures.
\section{Dispersionless KP hierarchy and Twistor Theory}
\subsection{Lax formalism}
Let us recall the Lax formalism of the dispersionless KP hierarchy
\cite{krichever-92,takasaki-takebe-92a}.
Let $(p,x)$ be canonical coordinates in a two dimensional
phase space with Poisson bracket
\beqn
\{ F, G \} = \frac{\rd F}{\rd p} \frac{\rd G}{\rd x}
- \frac{\rd F}{\rd x} \frac{\rd G}{\rd p}.
\eeqn
Let $t = (t_2,t_3,\ldots)$ be a set of ``time variables''.
The dispersionless KP hierarchy has the Lax representation
\beqn
\dfrac{\rd \cL}{\rd t_n} = \{ \cB_n, \cL\},
\eeqn
where $\cL$ is a Laurent series of $p$ of the form
\beqn
\cL = p + \sum_{n=1}^\infty u_{n+1}(x,t) p^{-n}.
\eeqn
$\cB_n$ are the polynomial (in $p$) part of the $n$-th
power of $\cL$,
\beqn
\cB_n = \Bigl( \cL^n \Bigr)_{\ge 0},
\eeqn
and obey the Zakharov-Shabat equations
\beqn
\dfrac{\rd \cB_m}{\rd t_n} - \dfrac{\rd \cB_n}{\rd t_m}
+ \{ \cB_m, \cB_n \} = 0.
\eeqn
One can extend these Lax equations by adding another
Laurent series $\cM$ of the form
\beqn
\cM = \sum_{n=2}^\infty n t_n \cL^{n-1} + x
+ \sum_{n=1}^\infty v_n(x,t) \cL^{-n-1}
\eeqn
that satisfies the Lax equations
\beqn
\dfrac{\rd \cM}{\rd t_n} = \{ \cB_n, \cM\}
\eeqn
and the canonical Poisson bracket relation
\beqn
\{ \cL, \cM \} = 1.
\eeqn
The dispersionless KP hierarchy is thus a ``quasi-classical''
limit of the ordinary KP hierarchy with commutators replaced by
Poisson brackets. $\cL$ corresponds to the Lax operator of the
ordinary Lax formalism of the KP hierarchy. $\cM$ is the
dispersionless version of the Orlov-Schulman operator $M$
\cite{orlov-schulman-86}.
\subsection{Twistorial structure}
The extended Lax formalism has another expression
\cite{takasaki-takebe-92a} which resembles twistor theory.
Let $\omega$ be the 2-form
\beqn
\omega = dp \wedge dx + \sum_{n=2}^\infty d\cB_n \wedge dt_n.
\eeqn
In terms of this 2-form, the Zakharov-Shabat equations and
the Lax equations can be rewritten in a very compact form as
\beqn
\omega \wedge \omega = 0,
\eeqn
and
\beqn
d\cL \wedge d\cM = \omega,
\eeqn
respectively.
These equations show that $\omega$ is a degenerate symplectic
form (on the infinite dimensional $(p,x,t)$ space) and that
$\cL$ and $\cM$ are its Darboux variables. The 2-form $\omega$
is an analogue of Gindikin's ``bundle of 2-form'', which lies
in the heart of his reformulation of curved twistor theory
\cite{gindikin-82}. This suggests that the dispersionless
KP hierarchy will give a kind of curved twistor theory.
This analogy is further deepend by the following characterization
of $\cL$ and $\cM$ in terms of a Riemann-Hilbert problem
\cite{takasaki-takebe-92a}:
\begin{theorem}
The functions $\cL = \cL(p,x,t)$ and $\cM = \cM(p,x,t)$
obeys the functional equations (Riemann-Hilbert problem)
\beqnarray
\cL(p,x,t) &=& f\Bigl( \cLbar(p,x,t), \cMbar(p,x,t) \Bigr),
\nonumber \\
\cM(p,x,t) &=& g\Bigl( \cLbar(p,x,t), \cMbar(p,x,t) \Bigr),
\eeqnarray
where $\cLbar = \cLbar(p,x,t)$ and $\cMbar = \cMbar(p,x,t)$ are
a (unique) solution of the equations
\beqnarray
&&
\dfrac{\rd \cLbar}{\rd t_n} = \{ \cB_n, \cLbar \}, \quad
\dfrac{\rd \cMbar}{\rd t_n} = \{ \cB_n, \cMbar \},
\nonumber \\
&&
\{ \cLbar, \cMbar \} = 1,
\eeqnarray
under the initial condition
\beqn
\cLbar(p,x,0) = p, \quad \cMbar(p,x,0) = x,
\eeqn
and $f = f(p,x)$ and $g = g(p,x)$ are given by
\beqn
f(p,x) = \cL(p,x,0), \quad g(p,x) = \cM(p,x,0).
\eeqn
\end{theorem}
{\bf Proof}: Note that $f$ and $g$ give a canonical pair
\beqn
\{ f, g \} = 1.
\eeqn Therefore $\cL$ and $f(\cLbar,\cMbar)$ turn out to satisfy the same
Lax equations. Furthermore they have an identical initial value at $t = 0$.
Thus they coincide. Similarly, we find that $\cM = g(\cLbar,\cMbar)$.
$\Box$
\bigskip
Let us specify why the above functional equations are interpreted
as Riemann-Hilbert problem. In a suitable analytical setting,
$\cLbar$ and $\cMbar$ are holomorphic functions of $p$ in a
neighborhood $\bar{D}$ of $p = 0$ (because $\cB_n$ are polynomials
of $p$), whereas $\cL$ and $\Xi = \cM - \sum n t_n \cL^{n-1}$
are holomorphic functions in a neighborhood $D$ of $p = \infty$.
Suppose that $D$ and $\bar{D}$ cover of the whole Riemann sphere
(e.g., this is the case if $t$ is small). A precise meaning of
the above functional equations is that these four functions
$(\cL,\Xi,\cLbar,\cMbar)$ satisfy the equation
\beqn
\cL = f(\cLbar,\cMbar), \quad
\Xi = g(\cLbar,\cMbar)
- \sum_{n=2}^\infty n t_n f(\cLbar,\cMbar)^{n-1}
\eeqn
over the intersection $D \cap \bar{D}$. This is a Riemann-Hilbert
problem of the type that typically arise in curved twistor
theory \cite{penrose-76,boyer-plebanski-85}.
If the flows are restricted to the $N$-dimensional subspace
$(x,t) = (x,t_1,\ldots,t_N,0,\ldots)$, the analogy with
curved twistor theory becomes more rigorous. One can indeed
``twist'' $(f,g)$ into a map connecting $(\cL^N, \cM \cL^{1-N}/N)$
and $(\cLbar,\cMbar)$, use it as a transition function for
constructing a mini-twistor space \cite{hitchin-82,jones-tod-85}
by gluing two coordinate patches. The four functions then become
``twistor functions''.
In the full space of time evolutions (i.e., if $N = \infty$),
the mini-twistor space ceases to exist. The Riemann-Hilbert
problem itself is meaningful, but geometric ingredients
of twistor theory, such as ``twistor correspondence'',
``twistor lines'' and ``twistor surfaces'' remain to be
clarified. This is the problem that we are addressing.
Our claim is that the Hamilton-Jacobi approach of Gibbons
and Kodama \cite{kodama-gibbons-90,gibbons-kodama-94} may be used
to resolve this problem.
\section{Hamilton-Jacobi Theory and Twistor Geometry}
\subsection{Multi-time Hamiltonian system}
In this and next subsections, we reformulate the Hamilton-Jacobi
approach of Gibbons and Kodama in our language. See also Carroll's
paper \cite{carroll-94} which presents an exposition of the
Hamilton-Jacobi approach in a form more faithful to the formulation
of Gibbons and Kodama.
Let $\lambda$ and $\mu$ be parameters, and define two functions
$p = p(\lambda,\mu,t)$ and $x = x(\lambda,\mu,t)$ implicitly
by the equations
\beqn
\cL(p,x,t) = \lambda, \quad
\cM(p,x,t) = \mu.
\eeqn
\begin{theorem}
$p = p(\lambda,\mu,t)$ and $x = x(\lambda,\mu,t)$ satisfy
the multi-time Hamiltonian system
\beqn
\dfrac{dp}{dt_n} = \dfrac{\rd \cB_n}{\rd x}, \quad
\dfrac{dx}{dt_n} = - \dfrac{\rd \cB_n}{\rd p}
\eeqn
with time-dependent Hamiltonians $\cB_n = \cB_n(p,x,t)$.
\end{theorem}
{\bf Proof}: For avoiding complicated notations, let us write
$p(\lambda,\mu,t)$ and $x(\lambda,\mu,t)$ as
$p(t)$ and $x(t)$. By the definition, we have the identities
\beqn
\cL\Bigl( p(t), x(t), t \Bigr) = \lambda, \quad
\cM\Bigl( p(t), x(t), t) \Bigr) = \mu. \nonumber
\eeqn
Differentiating these identities in $t_n$ gives
\beqnarray
\dfrac{\rd \cL}{\rd p} \dfrac{dp}{dt_n} +
\dfrac{\rd \cL}{\rd x} \dfrac{dx}{dt_n} +
\dfrac{\rd \cL}{\rd t_n} = 0,
\nonumber \\
\dfrac{\rd \cM}{\rd p} \dfrac{dp}{dt_n} +
\dfrac{\rd \cM}{\rd x} \dfrac{dx}{dt_n} +
\dfrac{\rd \cM}{\rd t_n} = 0.
\eeqnarray
The Lax equations for $\cL$ and $\cM$ can be used here to
evaluate the last terms in these equations. This gives
\beqnarray
\left( \begin{array}{rr}
\dfrac{\rd \cL}{\rd p} & \dfrac{\rd \cL}{\rd x} \\
\dfrac{\rd \cM}{\rd p} & \dfrac{\rd \cM}{\rd x}
\end{array} \right)
\left( \begin{array}{r}
\dfrac{\rd dp}{\rd t_n} \\
\dfrac{\rd dx}{\rd t_n}
\end{array} \right)
=
\left( \begin{array}{r}
- \dfrac{\rd \cL}{\rd t_n} \\
- \dfrac{\rd \cM}{\rd t_n}
\end{array} \right) \nonumber \\
=
\left( \begin{array}{rr}
\dfrac{\rd \cL}{\rd p} & \dfrac{\rd \cL}{\rd x} \\
\dfrac{\rd \cM}{\rd p} & \dfrac{\rd \cM}{\rd x}
\end{array} \right)
\left( \begin{array}{r}
\dfrac{\rd \cB_n}{\rd x} \\
- \dfrac{\rd \cB_n}{\rd p}
\end{array} \right).
\eeqnarray
The common $2 \times 2$ array is invertible (because of the
canonical Poisson commutation relation of $\cL$ and $\cM$)
and can be removed. The resulting equations are what we have
sought for.
$\Box$
\bigskip
The multi-time Hamiltonian system of Gibbons and Kodama can be
thus reproduced from our extended Lax formalism.
This result can be restated in terms of canonical transformations.
The Hamiltonian system is derived as equations of motion of
a point $(p,x)$ keeping $\cL$ and $\cM$ constant. In other words,
$\cL$ and $\cM$ are invariants of the Hamiltonian flows:
\beqnarray
\cL\Bigl( p(t),x(t),t \Bigr) &=& \cL\Bigl( p(0),x(0),0 \Bigr),
\nonumber \\
\cM\Bigl( p(t),x(t),t \Bigr) &=& \cM\Bigl( p(0),x(0),0 \Bigr).
\eeqnarray
Meanwhile, by the canonical Poisson commutation relation between
$\cL$ and $\cM$, the two-dimensional map $(p,x) \mapsto
(\lambda,\mu) = \Bigl(\cL(p,x,t),\cM(p,x,t)\Bigr)$ is symplectic.
The above invariance property of $\cL$ and $\cM$ then implies that
this symplectic map is a canonical transformation converting the
multi-time Hamiltonian system to the Hamiltonian system
\beqn
\dfrac{d\lambda}{dt_n} = 0, \quad
\dfrac{d\mu}{dt_n} = 0
\eeqn
with {\it zero Hamiltonians}. This Hamiltonian system is further
transformed into the Hamiltonian system
\beqn
\dfrac{d\lambda}{dt_n} = 0, \quad
\dfrac{d\xi}{dt_n} = - n \lambda^{n-1}
\eeqn
by the simple transformation
\beqn
\xi = \mu - \sum_{n=2}^\infty n t_n \lambda^{n-1}.
\eeqn
(Recall that the last transformation is just a disguise of the
transformation $\Xi = \cM - \sum n t_n \cL^{n-1}$.) The last
canonical variables $(\lambda,\xi)$ are exactly those of Gibbons
and Kodama. As they pointed out, the Hamiltonian system in these
variables resembles the time evolution of {\it scattering data}
in the conventional inverse scattering problem, which are also
{\it action-angle variables}.
\subsection{Generating function}
Gibbons and Kodama formulated the above canonical transformation
$(p,x) \mapsto (\lambda,\xi)$ in terms of a generating function.
In our formulation, it is more convenient to consider the
canonical transformation $(p,x) \mapsto (\lambda,\mu)$. Let
$S(\lambda,x,t)$ be a generating function for the latter.
The relations among the canonical variables $(p,x)$,
$(\lambda,\mu)$ and the Hamiltonian $\cB_n$ are now written
\beqnarray
&& \dfrac{\rd S(\lambda,x,t)}{\rd \lambda} = \mu, \quad
\dfrac{\rd S(\lambda,x,t)}{\rd x} = p, \nonumber \\
&& \dfrac{\rd S(\lambda,x,t)}{\rd t_n} = \cB_n.
\eeqnarray
These relations can be cast into a more compact form:
\beqn
dS = \mu d\lambda + pdx + \sum_{n=2}^\infty \cB_n dt_n.
\eeqn
This shows that the generating function is nothing but the
$S$-function that was discovered in an entirely different context
\cite{krichever-92,takasaki-takebe-92a}.
The generating function of Gibbons and Kodama, say
$S_{GK}$, is related to ours as
$S_{GK} = S - \sum \lambda^n t_n$.
The generating function $S(\lambda,x,t)$ is also related to the
linear problem
\beqnarray
&& \lambda \Psi = L\Bigl(\hbar\frac{\rd}{\rd x}\Bigr)\Psi, \quad
\hbar \frac{\rd \Psi}{\rd \lambda}
= M\Bigl(\hbar\frac{\rd}{\rd x}\Bigr)\Psi, \nonumber \\
&& \hbar \dfrac{\rd \Psi}{\rd t_n}
= B_n\Bigl(\hbar\frac{\rd}{\rd x}\Bigr)\Psi
\eeqnarray
of the KP hierarchy \cite{takasaki-takebe-92b,takasaki-takebe-95}.
The WKB approximation
\beqn
\Psi \sim \exp \hbar^{-1}S(\lambda,x,t).
\eeqn
to this {\it multi-time Schr\"odinger system} gives a system of
Hamilton-Jacobi (or ``eikonal'') equations of the form
\beqnarray
&&
\lambda = \cL\Bigl(\frac{\rd S(\lambda,x,t)}{\rd x}\Bigr),
\quad
\dfrac{\rd S(\lambda,x,t)}{\rd \lambda}
= \cM\Bigl(\frac{\rd S(\lambda,x,t)}{\rd x}\Bigr) \nonumber \\
&&
\dfrac{\rd S(\lambda,x,t)}{\rd t_n}
= \cB_n\Bigl(\frac{\rd S(\lambda,x,t)}{\rd x}\Bigr).
\eeqnarray
These Hamilton-Jacobi equations turn into the above defining
relations of $S(\lambda,x,t)$ by the change of coordinates
$(\lambda,x,t) \to (p,x,t)$ with
\beqn
p = \frac{\rd S(\lambda,x,t)}{\rd x}.
\eeqn
Thus, the generating function $S(\lambda,x,t)$ links
a {\it particle picture} (multi-time Hamiltonian system) and
a {\it wave picture} (multi-time Schr\"odinger system).
\subsection{Twistor geometry}
Let us now compare these results with curved twistor
theory. As the Riemann-Hilbert problem suggests,
$\lambda$ and $\mu$ may be thought of as coordinates on a
(virtual) twistor space, and $\cL$ and $\cM$ as defining a
twistor correspondence between space-time points and twistor
points. According to the ordinary curved twistor theory
of 4D space-times \cite{penrose-76}, the level surfaces
of $\cL$ and $\cM$ should be twistor surfaces. Actually,
the present setting rather resembles the curved twistor theory
of 3D space-times\cite{hitchin-82,jones-tod-85}. From that
point of view, it seems more suitable to interpret $(x,t)$
as space-time coordinates and $p$ as the fiber coordinate
of a projectivized ``spinor bundle''. Accordingly, $p(t)$
is a ``covariantly constant spinor field'' evaluated at
the space-time point $(x,t) = (x(t),t)$ of a twistor
surface.
Summarizing, we have the following dictionary between
the dispersionless KP hierarchy and the twistor geometry
(in particular, the space-time side):
\begin{itemize}
\item $(\lambda,\mu)$ $\longleftrightarrow$ mini-twistor space
\item $(x,t)$ $\longleftrightarrow$ space-time
\item $p$ $\longleftrightarrow$ fiber of projectivized spinor bundle
\item $(x(t),t)$ $\longleftrightarrow$ twistor surface
\item $p(t)$ $\longleftrightarrow$ covariantly constant spinor
field on this twistor surface
\end{itemize}
Of course the notion of mini-twistor space is only virtual, but
the others have a definite meaning.
Having obtained this twistorial interpretation of the dispersionless
KP hierarchy, we now turn to the dispersionless Toda hierarchy.
\section{Dispersionless Toda hierarchy}
\subsection{Lax formalism}
The Lax formalism of the dispersionless Toda hierarchy
\cite{takasaki-takebe-91}, too, is based on a two-dimensional
phase space with coordinates $(P,q)$. $q$ is a continuum limit
of the lattice coordinate, whereas $P$ corresponds to the unit-shift
operator. The Poisson bracket is given by
\beqn
\{ F, G \} =
P \frac{\rd F}{\rd P} \frac{\rd G}{\rd q}
- \frac{\rd F}{\rd q} P \frac{\rd G}{\rd P},
\eeqn
The corresponding 2-form is $d\log P \wedge dq$. In other words,
$(\log P, q)$ gives a canonical pair.
The dispersionless Toda hierarchy consists of commuting flows
of two Lax functions $\cL$ and $\cLbar$ with time variables
$( t = t_1, t_2, \cdots )$ and
$ ( \tbar = \tbar_1, \tbar_2, \cdots )$.
The Lax equations are given by
\beqnarray
\frac{\rd \cL}{\rd t_n} = \{ B_n , \cL \},
&\quad&
\frac{\rd \cL}{\rd \tbar_n} = \{ \cBbar_n , \cL \},
\nonumber \\
\frac{\rd \cLbar}{\rd t_n} = \{ B_n , \cLbar \},
&\quad&
\frac{\rd \cLbar}{\rd \tbar_n} = \{ \cBbar_n , \cLbar \},
\eeqnarray
where Lax functions $\cL$ and $\cLbar$ are two formal Laurent series
of a variable $P$ of the form
\beqn
\cL = P + \sum_{n=0}^{\infty} u_{n+1} (q,t,\tbar)P^{-n}, \quad
\cLbar = \sum_{n=1}^{\infty} {\bar u}_n q,(t,\tbar) P^n,
\eeqn
Furthermore, $\cB_n$ and $\cBbar_n$ are given by
\beqn
\cB_n = (\cL^n )_{\geq 0} \quad
\cBbar_n = ( \cLbar^{-n})_{\leq -1},
\eeqn
where $(\quad)_{\geq 0}$ and $(\quad)_{\leq -1}$ denote the projection
of Laurent series of $P$ into positive and negative powers respectively.
They obey the Zakharov-Shabat equations
\beqnarray
&&
\dfrac{\rd \cB_m}{\rd t_n} - \dfrac{\rd \cB_n}{\rd t_m}
+ \{ \cB_m , \cB_n \} = 0, \nonumber \\
&&
\dfrac{\rd \cBbar_m}{\rd \tbar_n} - \dfrac{\rd \cBbar_n}{\rd \tbar_m}
+ \{ \cBbar_m,\cBbar_n \} = 0, \nonumber \\
&&
\dfrac{\rd \cB_m}{\rd \tbar_n} - \dfrac{\rd \cBbar_n}{\rd t_m}
+ \{ \cB_m , \cBbar_n \} = 0.
\eeqnarray
This Lax formalism of the dispersionless Toda hierarchy can
be extended by adding two counterparts $\cM$ and $\cMbar$ of
the $\cM$ of the dispersionless KP hierarchy. $\cM$ and
$\cMbar$ are Laurent series of the form
\beqnarray
\cM &=& \sum_{n = 1}^{\infty} nt_n \cL^n + q
+ \sum_{n =1}^{\infty} v_n (q,t,\tbar)\cL^{-n},
\nonumber \\
\cMbar &=& -\sum_{n = 1}^{\infty} n\tbar_n \cLbar^{-n} + q
+ \sum_{n = 1}^{\infty} \vbar_n (q,t,\tbar)\cLbar^n
\eeqnarray
and satisfy the Lax equations
\beqnarray
\frac{\rd \cM}{\rd t_n} = \{ \cB_n , \cM \},
&\quad&
\frac{\rd \cM}{\rd \tbar_n} = \{ \cBbar_n , \cM \},
\nonumber \\
\frac{\rd \cMbar}{\rd t_n} = \{ \cB_n , \cMbar \},
&\quad&
\frac{\rd \cMbar}{\rd \tbar_n} = \{ \cBbar_n , \cMbar \}
\eeqnarray
and the canonical Poisson commutation relation
\beqn
\{ \cL , \cM \} = \cL, \quad
\{ \cLbar, \cMbar \} = \cLbar.
\eeqn
\subsection{Twistorial structure}
The above extended Lax formalism, too, can be reformulated
\`a la Gindikin \cite{gindikin-82}. Let $\omega$ be the 2-form
\beqn
\omega = d\log P \wedge dq
+ \sum_{n=1}^{\infty} d\cB_n \wedge dt_n
+ \sum_{n=1}^{\infty} d\cBbar_n \wedge d\tbar_n.
\eeqn
The Zakharov-Shabat equations then imply that
\beqn
\omega \wedge \omega = 0,
\eeqn
and the Lax equations and the canonical Poisson commutation
relations can be rewritten
\beqn
d\log\cL \wedge d\cM = \omega
= d\log\cLbar \wedge d\cMbar.
\eeqn
Thus $\omega$ is a denegerate symplectic form, and $(\cL,\cM)$
and $(\cLbar,\cMbar)$ are two different pairs of Darboux coordinates.
It is now straightforward to derive a Riemann-Hilbert problem
\cite{takasaki-takebe-91}. The two pairs of Darboux coordinates
are connected by functional relations of the form
\beqnarray
\cL(P,q,t,\tbar) &=&
f( \cLbar(P,q,t,\tbar), \cMbar(P,q,t,\tbar), t,\tbar),
\nonumber \\
\cM(P,q,t,\tbar) &=&
g( \cLbar(P,q,t,\tbar), \cMbar(P,q,t,\tbar), t,\tbar ).
\eeqnarray
The transition functions $f = f(P,q)$ and $g = g(P,q)$ are
required to satisfy the the (twisted) canonical Poisson
commutation relations
\beqn
\{ f(P,q), g(P,q) \} = f(P,q).
\eeqn
This gives the Riemann-Hilbert problem.
One can give a twistorial interpretation of this
Riemann-Hilbert problem in exactly the same way as in the
case of the dispersionless KP hierarchy. Note however that
that, unlike the case of the dispersionless KP hierarchy,
$(\cL,\cM)$ and $(\cLbar,\cMbar)$ are both dynamical variables.
It should be also added that a twistorial formulation of
the dispersionless Toda equation (the lowest member of the
hierarchy) is presented by Ward \cite{ward-90} in the
conventional mini-twistor language \cite{hitchin-82,jones-tod-85}.
\subsection{Multi-time Hamiltonian system}
Because of the presense of two canonical variable pairs
$(\cL,\cM)$ and $(\cLbar,\cMbar)$, the Hamilton-Jacobi formalism
of the dispersionless Toda hierarchy is more complicated than
the case of the dispersionless KP hierarchy.
Let $(\lambda,\mu)$ and $(\lambdabar,\mubar)$ be two such pairs
of parameters. One can define two functions
$P = P(\lambda,\mu,t,\tbar)$ and $q = q(\lambda,\mu,t,\tbar)$
implicitly by
\beqn
\cL(P,q,t,\tbar) = \lambda, \quad
\cM(P,q,t,\tbar) = \mu,
\eeqn
and similarly, $P = \Pbar(\lambdabar,\mu,t,\tbar)$ and
$q = \qbar(\lambdabar,\mubar,t,\tbar)$ by
\beqn
\cLbar(P,q,t,\tbar) = \lambdabar, \quad
\cMbar(P,q,t,\tbar) = \mubar.
\eeqn
We now have the following result:
\begin{theorem}
Both $\Bigl(P(\lambda,\mu,t,\tbar), q(\lambda,\mu,t,\tbar)\Bigr)$
and $\Bigl(\Pbar(\lambdabar,\mu,t,\tbar),
\qbar(\lambdabar,\mubar,t,\tbar)\Bigr)$
satisfy the same multi-time Hamiltonian system
\beqnarray
\frac{dP}{dt_n} = P \frac{\rd \cB_n}{\rd q},
&\quad&
\frac{dq}{dt_n} = - P \frac{\rd \cB_n}{\rd P},
\nonumber \\
\frac{dP}{d\tbar_n} = P \frac{\rd \cBbar_n}{\rd q},
&\quad&
\frac{dq}{d\tbar_n} = - P \frac{\rd \cBbar_n}{\rd P}.
\eeqnarray
\end{theorem}
{\bf Proof}: The proof is almost the same as the case of the
dispersionless KP hierarchy. The only difference is the Poisson
bracket. (It will be more convenient to do calculations in
the canonical pair $p = \log P$ and $q$ rather than in $P$
and $q$.)
$\Box$
\bigskip
Since this Hamiltonian system lives on a two-dimensional
phase space, its trajectories should form just a two-dimensional
family. Accordingly, the two sets of parameters $(\lambda,\mu)$
and $(\lambdabar,\mubar)$ should be functionally related.
Let us write the functional relations as
\beqn
\lambda = f(\lambdabar,\mubar), \quad
\mu = g(\lambdabar,\mubar).
\eeqn
This implies that the four functions $(\cL,\cM,\cLbar,\cMbar)$
satisfy the functional relations
\beqnarray
&&
\cL(P,q,t,\tbar) =
f\Bigl(\cLbar(P,q,t,\tbar), \cMbar(P,q,t,\tbar)\Bigr),
\nonumber \\
&&
\cM(P,q,t,\tbar) =
g\Bigl(\cLbar(P,q,t,\tbar), \cMbar(P,q,t,\tbar)\Bigr).
\eeqnarray
Thus the Riemann-Hilbert problem of the previous subsection
can be reproduced from the multi-time Hamiltonian system.
Twistorial interpretation of these results is quite parallel
to the case of the dispersionless KP hierarchy.
\subsection{Generating functions}
The two different parametrizations of trajectories of the multi-time
Hamiltonian system lead to two different canonical transformations
and multi-time Hamiltonian systems with zero Hamiltonians. The first
canonical transformation $(P,q) \mapsto (\lambda,\mu)$ is defined by
a generating function $S(\lambda,q,t,\tbar)$ as
\beqnarray
&&
\frac{\rd S}{\rd \lambda} = \mu, \quad
\frac{\rd S}{\rd q} = \log P,
\nonumber \\
&&
\dfrac{\rd S}{\rd t_n} = \cB_n, \quad
\dfrac{\rd S}{\rd \tbar_n} = \cBbar_n.
\eeqnarray
The transformed Hamiltonian system is given by
\beqn
\dfrac{d\lambda}{dt_n} = 0, \quad
\dfrac{d\mu}{dt_n} = 0.
\eeqn
Similarly, the second canonical transformation
$(P,q) \mapsto (\lambdabar,\mubar)$ is defined by a generating
function $\Sbar = \Sbar(\lambdabar,q,t,\tbar)$ as
\beqnarray
&&
\frac{\rd \Sbar}{\rd \lambdabar} = \mubar, \quad
\frac{\rd \Sbar}{\rd q} = \log P,
\nonumber \\
&&
\dfrac{\rd \Sbar}{\rd t_n} = \cB_n, \quad
\dfrac{\rd \Sbar}{\rd \tbar_n} = \cBbar_n.
\eeqnarray
The transformed Hamiltonian system is given by
\beqn
\dfrac{d\lambdabar}{dt_n} = 0, \quad
\dfrac{d\mubar}{dt_n} = 0.
\eeqn
These Hamiltonian system can be further mapped to a system of
Gibbons and Kodama type by a simple change of variables
$(\lambda,\mu) \mapsto (\lambda,\xi)$ and
$(\lambdabar,\mubar) \mapsto (\lambdabar,\xibar)$ as:
\beqn
\xi = \mu - \sum_{n=1}^\infty n t_n \lambda^n, \quad
\xibar = \mubar + \sum_{n=1}^\infty n \tbar_n \lambdabar^{-n}.
\eeqn
The defining equations of the canonical transformations can be
rewritten
\beqnarray
dS &=& \mu d\lambda + q
+ \sum_{n=1}^\infty \cB_n dt_n
+ \sum_{n=1}^\infty \cBbar_n d\tbar_n ,
\nonumber \\
d\Sbar &=& \mubar d\lambdabar + q
+ \sum_{n=1}^\infty \cB_n dt_n
+ \sum_{n=1}^\infty \cBbar_n d\tbar_n .
\eeqnarray
Thus $S$ and $\Sbar$ are, in fact, the same as those which
are already known \cite{takasaki-takebe-91} and connected
with the linear problem of the Toda lattice hierarchy
\cite{takasaki-takebe-93,takasaki-takebe-95}.
\section{Concluding Remarks}
The dispersionless KP and Toda hierarchies are also related to
another geometric structure -- Frobenius manifold.
Let us recall the notion of Frobenius algebra and Frobenius
manifold. For more details, we refer to lecture notes of
Dubrovin \cite{dubrovin-lecture} and Hitchin \cite{hitchin-lecture}.
A Frobenius algebra is a finite-dimensional commutative and
associative algebra $V$ with an identity element $e$ and a
linear form $\theta \in V^*$ for which
$g(a,b) = \theta(ab)$, $a,b \in V$, is a nondegenerate
inner product on $V$. This determines a symmetric form
$g \in S^2 V^*$. Furthermore, the multiplicative structure
$V \otimes V \to V$ determines an element of
$V \otimes V \otimes V^*$. Identifying $V^* \simeq V$ by the
nondegenerate inner product, one obtains an element
$c \in V^* \otimes V^* \otimes V^*$. By commutativity
$c$ becomes totally symmetric, i.e., $c \in S^3 V^*$.
The three data $(\theta,g,c)$ conversely determines a
Frobenius algebra.
Let $M$ be an $n$-manifold with a smoothly varying structure
of Frobenius algebra in the tangent space at each point.
This amounts to giving the following threee data:
\begin{itemize}
\item $\theta \in C^\infty(T^*M)$
\item $g \in C^\infty(S^2 T^*M)$
\item $c \in C^\infty(S^3 T^*M)$
\end{itemize}
Let $e$ denote the identity element of the point-wise defined
Frobenius algebra, and $\nabla$ the covariant derivative of
the Levi-Civita connection determined by the metric $g$.
$e$ is a vector field on $M$ and called the Euler vector field.
A manifold $M$ with these data is called a Frobenius manifold
if the following conditions are satisfied:
\begin{enumerate}
\item $(M,g)$ is flat,
\item the Euler vector field $e$ is covariantly constant,
\item $\nabla c$ is symmetric.
\end{enumerate}
The flatness of the metric implies the existence of special
local coordinates (``flat coordinates'') $t^i$ for which the
coefficents of metric are constant:
\beqn
g = \sum \eta_{ij} dt^i dt^j, \quad
\eta_{ij} = \mbox{constant}.
\eeqn
The third condition above, meanwhile, implies the existence of
another set of distinguished coordinates (``canonical
coordinates''). They are local orthogonal coordinates $u^i$
for which the metric is written in the following special
diagonal form with a potential $\phi$:
\beqn
g = \sum \mu_i (du^i)^2, \quad
\mu_i = \dfrac{\rd \phi}{\rd u^i}.
\eeqn
(A metric of this form is called, in general, an Egorov metric.)
A particularly important class of Frobenius manifolds are
those with homogeneity. This means that all structure functions
are homogeneous with respect to the Euler vector field $e$.
Dubrovin \cite{dubrovin-92} pointed out that the moduli spaces
of marked Riemann surfaces of arbitrary genus and with an
arbitrary number of marked points (``Hurwitz spaces'') have
a Frobenius structure (with homogeneity).
Two special classes of those Frobenius manifolds emerge as
solutions of the dispersionless KP and Toda hierarchies.
(Furthermore, these are examples where the Riemann-Hilbert
problem can be solved rather explicitly.)
The first example is related to the dispersionless KP hierarchy.
Let $M$ be the space of complex polynomials
\beqn
E(p) = p^{n+1} + a_1 p^{n-1} + \cdots + a_n.
\eeqn
This is an affine space and its tangent vectors at the point
is represented by a polynomial $\Edot(p)$ of the form
\beqn
\Edot(p) = \Edot_1 p^{n-1} + \cdots + \Edot_n.
\eeqn
We define a metric on $M$ by
\beqn
g(\Edot_1,\Edot_2)
= \res_{p = \infty} \frac{\Edot_1 \Edot_2}{E'}dp,
= - \sum_i \res_{p = \alpha_i} \frac{\Edot_1 \Edot_2}{E'}dp,
\eeqn
where $\alpha_i$'s are the roots of $E'(p) = dE(p)/dp$, and
the summation is over all those roots. $M$ becomes a Frobenius
manifold with canonical coordinates $u^i = E(\alpha_i)$.
In the context of the dispersionless KP hierarchy, $E(p)$
is related to $\cL$ as:
\beqn
E(p) = \cL^{n+1}.
\eeqn
This gives a ``reduction'' of the dispersionless KP hierarchy
\cite{krichever-92}, which has only a finite number of unknown
functions ($a_1,\ldots,a_{n-1}$ and $a_n$ in the above notation). Solving
this relation for $p$ yields a Laurent series of
the form
\beqn
p = \cL + b_1 \cL^{-1} + b_2 \cL^{-2} + \cdots.
\eeqn
The first $n$ coefficients $b_1,\ldots,b_n$ then give flat
coordinates.
Similarly, the family of trigonometric polynomials of the form
\beqn
E(p) = e^{np} + a_1 e^{(n-1)p} + \cdots + a_n + a_{n+1}e^{-p}
\eeqn
gives another example of Frobenius manifold. If we identify
$P = e^p$, this corresponds to a reduction of the dispersionless
Toda hierarchy defined by the relations \cite{takasaki-93}
\beqn
\cL^n = E(p) = \cLbar^{-1}.
\eeqn
A very intriguing open problem is to interpret the ``unreduced''
dispersionless KP and Toda hierarchies as a kind of infinite
dimensional Frobenius manifolds. This will require a drastic
modification of the notion of Frobenius manifold. This situation
is reminiscent of our twistorial interpretation of these hierarchies
-- the dispersionless hierarchies have no twistor space in the
ordinary sense, but still retain a virtual counterpart of twistor
space and twistor lines (encoded in the Riemann-Hilbert problem)
as well as the notion of twistor functions and twistor lines
(related to a multi-time Hamiltonian system).
The same problem may be raised to Krichever's ``universal
Whitham hierarchy'' \cite{krichever-94}, which includes all
Hurwitz spaces of arbitrary genus and number of marked points
as special
solutions.
\begin{acknowledge}
\noindent The authors are grateful to Professor Yuji Kodama
for fruitful discussions. One of the author (P. G.) would
like to thank Professor Masaki Kashiwara for invitation to RIMS.
\end{acknowledge}
\begin{thebibliography}{99}
\bibitem{boyer-plebanski-85} C.P. Boyer and J.f. Plebanski 1985,
An infinite hierarchy of conservation laws and nonlinear
superposition principles for self-dual Einstein spaces,
{\it J. Math. Phys. 26}, 229-234.
\bibitem{carroll-94} R. Carroll 1994,
Remarks on dispersionless KP, KdV, and 2D gravity,
{\it J. Nonlinear Sci. 4}, 519-544.
\bibitem{dubrovin-92} B. Dubrovin 1992,
Hamiltonian formalism of Whitham-type hierarchies and
topological Landau-Ginzburg models,
{\it Comm. Math. Phys. 145}, 195-207.
\bibitem{dubrovin-lecture} B. Dubrovin 1994,
Geometry of 2D Topological Field Theories,
{\it hep-th/9407018}.
\bibitem{gibbons-kodama-89} J. Gibbons and Y. Kodama 1989,
A method for solving the dispersionless KP hierarchy
and its exact solutions II,
{\it Phys. Lett. 135A}, 167-170.
\bibitem{gibbons-kodama-94} J. Gibbons and Y. Kodama 1994,
Solving dispersionless Lax equqations,
{\it Singular Limit of Dispersive Waves}, Lyon 1991,
NATO/ASI Series B:Physics, vol. 320, Plenum Press,
61-66.
\bibitem{gindikin-82} S.G. Gindikin 1982,
Integral geometry and twistors,
in {\it Twistor Geometry and Non-linear Systems},
H.D. Doebner and T. Weber (eds.),
Lecture Notes in Mathematics vol. 970, Springer-Verlag,
2-42.
\bibitem{hitchin-82} N.J. Hitchin 1982,
Complex manifolds and Einstein's equations,
in {\it Twistor Geometry and Non-linear Systems},
H.D. Doebner and T. Weber (eds.),
Lecture Notes in Mathematics vol. 970, Springer-Verlag,
73-99.
\bibitem{hitchin-lecture} N. Hitchin, 1996,
Lecures on Frobenius manifold, Notes by D. Calderbank.
\bibitem{jones-tod-85} P.E. Jones and K.P. Tod 1985,
Minitwistor spaces and Einstein Weyl spaces,
{\it Class. Quant. Grav. 2}, 565-577.
\bibitem{kodama-88} Y. Kodama 1988,
A method for solving the dispersionless KP hierarchy
and its exact solutions,
{\it Phys. Lett. 129A}, 223-226.
\bibitem{kodama-90} Y. Kodama 1990,
Solutions of the dispersionless Toda equation,
{\it Phys. Lett. 147A}, 477-482.
\bibitem{kodama-gibbons-90} Y. Kodama and J. Gibbons 1990,
Integrability of the dispersionless KP hierarchy,
in {\it Nonlinear and Turbulent Processes in Physics},
Kiev 1989, World Scientific,
166-180.
\bibitem{krichever-92} I.M. Krichever 1992,
Topological minimal models and dispersionless Lax equations,
{\it Commun. Math. Phys. 143}, 415-429.
\bibitem{krichever-94} I.M. Krichever 1994,
The $\tau$ function of the universal Whitham hierarchy,
matrix models and topological field theories,
{\it Comm. Pure. Appl. Maths. 47}, 437-476.
\bibitem{lebedev-manin-79} D. Lebedev and Y. Manin 1979,
Conservation laws and Lax representations on Benny's
long wave equations,
{\it Phys. Lett. 74A}, 154-156.
\bibitem{orlov-schulman-86} A. Orlov and E. Schulman 1986,
Additional symmetries for integrable and conformal algebra
representation,
{\it Lett. Math. Phys. 12}, 171-179.
\bibitem{penrose-76} R.P. Penrose 1976,
Non-linear gravitons and curved twistor spaces,
{\it Gen. Rel. Grav. 7}, 31-52.
\bibitem{takasaki-90} K. Takasaki 1990,
Symmetries of hyper-K\"ahler (or Poisson gauge field) hierarchy,
{\it J. Math. Phys. 31}, 1877-1888.
\bibitem{takasaki-92} K. Takasaki 1992,
Area-preserving diffeomorphisms and nonlinear integrable systems,
in {\it Topological and geometrical methods in field theory},
J. Mickelsson and O. Pekonen (eds.), World Scientific,
383-397.
\bibitem{takasaki-93} K. Takasaki 1993,
Integrable hierarchy underlying topological
Landau-Ginzburg models of D-type,
{\it Lett. Math. Phys. 29}, 111-121.
\bibitem{takasaki-takebe-91} K. Takasaki and T. Takebe 1991,
SDiff(2) Toda equation-hierarchy, tau function and symmetries,
{\it Lett. Math. Phys. 23}, 205-214.
\bibitem{takasaki-takebe-92a} K. Takasaki and T. Takebe 1992a,
SDiff(2) KP hierarchy,
{\it Int. J. Mod. Phys. A7, Suppl. 1B}, 889-922.
\bibitem{takasaki-takebe-92b} K. Takasaki and T. Takebe 1992b,
Quasi-classical limit of KP hierarchy,
W-symmetries and free fermions,
in Proceedings of Lobachevsky Semester of
Euler International Institute, St. Petersburg 1992;
{\it hep-th/9207081}.
\bibitem{takasaki-takebe-93} K. Takasaki, and T. Takebe 1993,
Quasi-classical limit of Toda hierarchy
and W-infinity symmetries,
{\it Lett. Math. Phys. 28}, 165-176.
\bibitem{takasaki-takebe-95} K. Takasaki and T. Takebe 1995,
Integrable hierarchies and dispersionless limit,
{\it Reviews in Mathematical Physics 7}, 743-808.
\bibitem{tsarev-85} S.P. Tsarev 1985,
On the Poisson brackets and one-dimensional Hamitonian systems
of hydrodynamic type,
{\it Soviet Math. Dokl. 31}, 488-491.
\bibitem{ward-90} R.S. Ward 1990, Einstein-Weyl spaces
and SU($\infty$) Toda fields,
{\it Class. Quant. Grav. 7}, L95-L98.
\bibitem{zakharov-81} V.E. Zakharov 1981,
On the Benny's equations,
{\it Physica 3D}, 193-202.
\end{thebibliography}
\end{document}