Текст
                    Integrability:
The Seiberg-Witten and Whitham Equations
Edited by
H.W. Braden
University of Edinburgh, Scotland
and
I.M. Krichever
L,D. Landau Institute of Theoretical Physics
Moscow, Russia
GORDON AND BREACH SCIENCE PUBLISHERS
Australia • Canada • France • Germany • India • Japan
Luxembourg • Malaysia • The Netherlands • Russia
Singapore • Switzerland


Copyright © 2000 OPA (Overseas Publishers Association) N.V. Published by license under the Gordon and Breach Science Publishers imprint. All rights reserved. No part of this book may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying and recording, or by any information storage or retrieval system, without permission in writing from the pubHsher. Printed in Singapore. Amsteldijk 166 1st Floor 1079 LH Amsterdam The Netherlands British Library Cataloguing in Publication Data ISBN: 90-5699-281-3
Contents Preface vii 1 Baker-Akhiezer Functions and Integrable Systems 1 I.M. Krichever 2 Algebraic Geometry, Integrable Systems, and Seiberg-Witten Theory 23 E. Markman 3 Seiberg-Witten Theory and Integrable Systems 43 E. D 'Hoker and D.H. Phong 4 Seiberg-Witten Curves and Integrable Systems 69 A. Marshakov 5 Integrability in Seiberg-Witten Theory 93 A. Morozov 6 WDVV Equations and Seiberg-Witten Theory 103 A. Mironov 1 On Geometry of a Special Class of Solutions to Generalised WDVV Equations 125 A.P. Veselov 8 Picard-Fuchs Equations, Hauptmoduls and Integrable Systems 137 J. Hamad 9 Painleve Type Equations and Hitchin Systems 153 MA. Olshanetsky 10 World-sheet Instantons and Virasoro Algebra 175 T Eguchi
vi CONTENTS 11 Dispersionless Integrable Systems and their Solutions 199 Y. Kodama 12 A^-Component Integrable Systems and Geometric Asymptotics 213 M.S. Alter 13 Systems of Hydrodynamic Type from Poisson Commuting Hamiltonians 229 A.P. Fordy 14 Integrability of Equations of Hydrodynamic Type from the End of the 19th to the End of the 20th Century 251 S.P. Tsarev Appendix: List of Participants 267 Index 275
Preface Integrable systems, like invariant theory, has had a somewhat chequered history. The nineteenth century saw many aspects of geometry and analysis, particularly the development of Abelian functions, drawn together in the study of integrable systems. The works of Jacobi, Abel, Riemann and Weierstrass enabled a number of important integrable problems of mechanics and physics to be solved, including the problem of geodesic motion on a tri-axial ellipsoid (Jacobi), the motion of a heavy body around a fixed point (Kowalevski), the motion of a rigid body in a liquid (Kirchoff, Clebsch and Steklov) and the motion of a material point on a sphere under the influence of a quadratic potential (Neumann). Geometric aspects of integrable systems were developed by Darboux, Levi-Civita and Stackel. The turn of the twentieth century however saw a change in attitude towards integrable systems. After the results of Poincare and de Bruns, integrable systems lacking any 'obvious' group symmetry were perceived as something exotic, and ideas about the structure of nontrivial integrable systems did not exert any real influence on the development of physics for much of that century. This situation changed drastically with the discovery of soliton theory. Nonlinear phenomena could now be treated, and the ever-growing interest in this theory is connected with the fact that it is applicable to equations which (as became clear in the mid-sixties) possess remarkable universality properties. They arise in the description of the most diverse phenomena in plasma physics, the theory of elementary particles, the theory of superconductivity and in nonlinear optics. Among the equations referred to are the Korteweg- de Vries equation, the nonlinear Schrodinger equation, the Sine-Gordon equation and many others. Perhaps surprisingly, finite-dimensional integrable systems again appear in this new setting, now (for example) describing the evolution of exact solutions of the partial differential equations. Thus many classical integrable systems, together with several families discovered at this time (such as the Toda and Calogero-Moser systems), can be viewed as either describing particular soliton solutions, or as finite-dimensional systems their own rights. There is a rich interplay between the Inverse Scattering techniques developed to study the integrable partial differential equations of soliton theory and the techniques of integrable systems. This ubiquity of integrable systems together with the beautiful structures that
viii PREFACE underly them has led to a renewed interest in the area. Geometry and algebraic geometry, functional equations and special functions, Lie algebras and groups all come together in their study. The most recent burst of interest is due to the unexpected connections of these systems to N=2 supersymmetric gauge theories. In 1994 Seiberg and Witten suggested a new way to deal with four-dimensional N=2 supersymmetric gauge theories, both for pure gauge theories and those with matter hypermultiplets. The consequences of this work are still being assimilated. The low energy effective actions of such supersymmetric theories are described by a function, the prepotential ^and depends on a finite dimensional moduli space. This moduli space characterises the vacuum of the theory. Seiberg and Witten introduced an ansatz identifying this moduli space with the moduli space of a certain family of algebraic curves. The prepotential F which encodes the effective quantum theory is derived from a (Seiberg-Witten) one-form on this moduli space. At this stage there was no a priori reason to expect the appearance of integrable systems. The first hint of a connection was the observation by Gorsky that the Seiberg-Witten curve corresponding to the pure A^ = 2 SUSY gauge theory with gauge group SU (N^) could be identified with the family of spectral curves of the A^^-periodic Toda lattice. Subsequently, integrable systems were associated to the spectral curves arising from A^ = 2 gauge models coupled to matter hypermultiplets in varying representations. (A fuller list of models and references will appear in the text.) The connection between the Seiberg-Witten ansatz and integrable systems was further strengthened by the identification of the Seiberg-Witten one-form with that one-form central to the Whitham equations. The Whitham equations in view here are those that arise when applying Whitham averaging to finite-gap solutions of soli ton bearing equations. Whitham averaging is essentially the WKB method applied to nonlinear equations. (The Whitham equations are the first order equations of an asymptotic expansion in the parameter describing the slow variation.) In the context of finite-gap solitons (which depend on a finite number of parameters or moduli) exact results are possible, and as these moduli slowly vary Whitham averaging describes the evolution of the solitons. The equations themselves are a quasi-linear, first order system of equations of hydrodynamic type. Remarkably, this one- form, central to both Whitham theory and Seiberg-Witten theory, is the generator of a canonical transformation to action-angle variables in the Hamiltonian setting of the soliton theory. Yet although many connections now exist between Seiberg-Witten theory and integrable systems, the correspondence remains poorly understood. The present volume arose from a meeting devoted to the further elucidation of these connections. The chapters of this book consist (in most part) of surveys given by plenary speakers at this meeting, and cover various areas impinging on the subject. Overviews are seldom easy, and we are very grateful to the contributing authors for holding so strictly to their remit. We hope that this book will provide an excellent introduction to the ideas and methods surrounding these exciting theories. An overview of the book is as follows. The first chapter sets the theme, introducing many of the main characters that will be developed further in the book. The Baker-Akhiezer functions central to finite-gap integration, the Whitham equations, topological and Seiberg- Witten theories are all introduced. Following this are three detailed chapters on various
PREFACE ix aspects of Seiberg-Witten theory. Markman overviews the special and algebraic geometry. D'Hoker and Phong survey the physical background to Seiberg-Witten theory and focus on connections with the Calogero-Moser family of integrable systems. Marshakov describes the construction of Seiberg-Witten curves associated with a wide range of integrable systems. Morozov's contribution then tackles the fundamental connection between Seiberg- Witten and integrability: why should integrability appear at all? Connections with matrix models, x-functions and the Kontsevich model amongst other things are touched upon. The next two chapters deal with the WDVV equations. Both the Seiberg-Witten prepotential and the Whitham equations give rise to solutions of the WDVV equations that characterise two-dimensional topological theories. These equations describe relations between various third derivatives of the prepotential. They define an associative algebra known as a Frobenius algebra; the geometrical data defines a Frobenius manifold. Mironov surveys the connections between these equations and Seiberg-Witten theory, while Veselov describes a special class of solutions related to deformations of Calogero-Moser systems. Both Hamad and Olshanetsky describe particular classes of equations associated with surfaces and integrable systems, the former that of Picard-Fuchs equations and the latter, Painleve equations and Hitchin systems. Eguchi's contribution returns to the topological field theoretic aspects touched upon in several of the earlier chapters, notably in connection with the WDVV equations. In particular, the Virasoro conjecture on quantum cohomology is reviewed. Topological theories are related to integrable systems of hydrodynamic type, and Kodama describes this connection together with those of dispersionless hierarchies and Frobenius manifolds. Alber's contribution deals with geometric asymptotics which further relates solutions of partial differential equations with integrable systems. The final two chapters focus on equations of hydrodynamic type. Fordy surveys the connection between such systems and the equations arising by requiring two (quadratic) Hamiltonian vector fields to Poisson commute. Tsarev concludes with an extensive overview of equations of hydrodynamic type, with particular emphasis on the connections with classical geometric problems such as the existence of orthogonal curvi-linear coordinates and many others. Each chapter can be read alone, yet together they convey something of the rich flavour of present day integrable systems. It remains our pleasure to thank various individuals and organisations. The meeting 'Integrability: The Seiberg-Witten and Whitham Equations' which took place in Edinburgh in 1998 and from which this book stems, was largely funded by EPSRC. We are grateful to them for this support. The meeting was organised under the auspices of the International Centre for Mathematical Sciences, Edinburgh, and we wish to thank Tracey Dart and her assistants for their organisational skills and hard work. We are also grateful to our co- organisers David Fairlie and Ian Strachan for their substantial efforts. Finally we thank all of those who attended the meeting. (The list of participants is given as an appendix.) Their informal discussions and the shorter research presentations given at this time were deemed very helpful by many, and we were delighted to see several research collaborations ensue from the meeting. H.W. Braden I.M. Krichever
1 Baker-Akhiezer Functions and Integrable Systems I.M. KRICHEVER Columbia University, 2990 Broadway, New York, NY 10027, USA and Landau Institute for Theoretical Physics, Kosygina str 2, 117940 Moscow, Russia; e-mail: krichev @ math. Columbia, edu. Key ideas of the algebra-geometric methods in the theory of solitons are presented. Unexpected Hnks between various theories in which the same objects emerge repeatedly, albeit under different names, like r-function in the Whitham theory, partition function in topological filed theories, and prepotential in Seiberg-Witten theory mainly are discussed. 1.1 Introduction The main goal of this chapter is to present key ideas which unify the algebro-geometric methods in the theory of soliton equations and recent developments in the theory of N = 2 super symmetric gauge models. Solitons arose originally in the study of shallow water waves. Since then, the notion of soliton equations has widened considerably. It embraces now a wide class of non-linear partial differential equations, which all share the characteristic feature of being expressible as a compatibility condition for an auxiliary pair of linear differential equations. A variety of methods have been developed over the years to construct exact solutions for these equations. Since the middle seventies algebraic geometry has become one of the most powerful tools among them. In the next section we outline basic elements of the, so-called, finite-gap theory which were originated in (Novikov [1994]; Dubrovin et al. [1976], Lax [1975], McKean et al. [1975]) in the framework of the Floquet spectral theory of periodic Schrodinger operators combined with a theory of completely integrable Hamiltonian systems. Analytical properties
2 L KRICHEVER of Bloch solutions of finite-gap Schrodinger operators with respect to an auxiliary spectral parameter, established in this remarkable series of papers, were a starting point in a definition of the Baker-Akhiezer functions which are the core of a general algebro-geometric construction of exact periodic and almost periodic solutions to soliton equations proposed in (Krichever [1976, 1977a, 1977b]). Section 3 is devoted to brief description of the Whitham method which is a generalization to the case of partial differential equation of the classical Bogolyubov-Krylov averaging method. It turns out that differential equations describing a slow modulation of integrals of finite-gap solutions of soliton equations, called Whitham equations, are deeply connected with a theory of deformations of topological quantum field models. This connection between Whitham theory and the, so-called, Witten-Dijgraaf-Verlinde-Verlinde (WDVV) equations is discussed in Section 4. In the last section we show that the Seiberg-Witten theory ofN = 2 supersymmetric gauge theories can be considered on one hand as a part of the Whitham theory and at the same time leads to a new general approach to Hamiltonian theory of soliton equations proposed in (Krichever ^r fl/. [1977, 1999]). Our discussion of unexpected links between various theories in which the same objects emerge repeatedly, albeit under different names, like r-function in the Whitham theory, partition function in topological field theories, and prepotential in Seiberg-Witten theory mainly follows Krichever et ah [1999]), where more details can be found. 1.2 Finite-gap Solutions of Integrable Systems The finite-gap or algebro-geometric integration method is uniformly applicable to all soliton equations. In the case of spatial one-dimensional evolution equations it is instructive enough to consider as a basic example equations that have Lax representation dtL = [A,L], (1) where the unknown functions {m, (x, y, 0}/^Jo ^ {^7 (-^^ y^ ^'^)T=o ^^ ^^^ coefficients of the ordinary differential operators n—2 m—2 /=0 7=0 A preliminary classification of equations of the form (1) is by the orders n, m of the operators L and A. In the case n = 2 the operator L is just the usual Schrodinger operator L = —d^-\-u(x,t), and for A = d^ — 3/2udx — 3/4ux equation (1) is equivalent to the KdV equation 4Ut - 6uUx + Uxxx = 0. From (1) it follows that certain spectral quantities of the operator L are integrals of motion. In the framework of the finite-gap theory these integrals are organized in the form of the so-called spectral curve. In all the cases, i.e. finite-dimensional integrable systems, spatial one- or two-dimensional evolution equations, the spectral curve is defined by a characteristic equation R(w, E) = det(M; - T(t, E)) = 0, (3)
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 3 where T(t, E) is a, finite-dimensional matrix depending on a spectral parameter E. In the case of finite-dimensional (or (0+1)) integrable systems, which have the Lax representation Lt(t, E) = [A(t, E), L(t, £")], where L and A are finite-dimensional matrices depending on the spectral parameter, the matrix T(t, E) defining the spectral curve is the Lax matrix L by itself, i.e. T(t, E) = L(t, E). In the infinite-dimensional case the spectral curve can be defined for special classes of solutions, only. For spatial one-dimensional systems these classes are singled out by the constraint that there exists an additional operator T which commutes with L and (dt — A). For example, if the coefficients of the operator L of the form (2) are periodic functions of the variable x with period T, then the operator L commutes with the shift operator f : y(x) h^ y(x + T). (4) Therefore, the finite-dimensional linear space C(E) of the solutions of the ordinary differential equation y(x) e C(E) : Ly = Ey (5) is invariant with respect to T. Restriction of the shift operator onto C defines a finite- dimensional linear operator T(E) = f\ciE)^ (6) A point Q of the spectral curve F is just a pair Q = (E,w) of complex numbers that satisfy (3). They parametrize Bloch eigenfunctions of the operator L, i.e. common eigenfunctions of L and the monodromy operator LV^(x, Q) = Eif(x, 2), if(x -\-T,Q) = wif(x, Q). (7) In a generic case the corresponding Riemann surface is a smooth surface of infinite genus. If its genus is finite then the corresponding operators are called finite-gap or algebro-geometric operators. It should be emphasized that in such a case the Riemann surface defined by the characteristic equation is a singular surface. After resolving the singularities we get a finite genus smooth Riemann surface, i.e. an algebraic curve. For example, let L = — 9^ + u(x) be the Schrodinger operator with a periodic potential u(x) = u(x -\-T). Then equation (3) has the form w^ - 2Q(E)w + 1=0, 2Q(E) = Tr T(E). (8) The roots 6, of the equation Q^(E) = 1 are points of the periodic or anti-periodic spectrum of the Schrodinger operator. Equation (8) can be rewritten in the form oo y^ = Y[(E-ei), y = w-Q(E). (9) i=i If all of the edges 6, are distinct then (9) defines smooth infinite genus hyperelliptic Riemann surface. The finite-gap operators correspond to the degenerate case when all but a finite number of eigenvalues of periodic or anti-periodic spectral problem for L are multiple. Let £"1 < • • • < £"2^+1 be the simple eigenvalues. Then a finite genus smooth algebraic curve of the Bloch functions is defined by 2g-\-l
4 I. KRICHEVER Note that Ei are the edges of the spectrum bands of L, being considered as an operator in the space of square integrable functions on the whole Une. The finite-gap theory was initiated by the work (Novikov [1974]), where the spectral theory of periodic operators was combined with an approach based on a use of the KdV hierarchy. The KdV equation (as well as any soliton equation) is compatible with an infinite hierarchy of commuting flows. They have the Lax representation diL = [Ai,L], di = ^. (11) where A, is an ordinary differential operator of order 2/ + 1. Consider stationary solutions of a linear combination of these flows, i.e. solutions of the ordinary differential equation g [L,A]=0, A = Y^CiAi, (12) i=\ As it was shown in (Novikov [1974]), equation (12) is a completely integrable Hamiltonian system. Therefore, its general solution is a quasi-periodic function of x. Periodic solutions are finite-gap potentials. Let A{E) be the restriction of the operator A commuting with L onto C{E). The matrix elements of A(£') are polynomial functions of the spectral parameter. Therefore, the characteristic equation det(A(£') — iti) = 0 defines an algebraic curve P. It turns out that this curve coincides with the spectral curve (10). Note that the operator equation (12) is a particular case of the more general problem of the classification of commuting ordinary differential operators L„ and L^ of orders n and m, respectively. As a purely algebraic problem it was considered and partly solved in the remarkable works of Burchnall and Chaundy (Burchall et al [1922, 1928]) in the 1920s. They proved that for any pair of such operators there exists a polynomial R(X, /jl) in two variables such that R(Ln, Lm) = 0. If the orders n and m of these operators are co-prime, (n,m) = 1, then for each point Q = (X, /jl) of the curve F defined in C^ by the equation R(X, /jl) = 0 there corresponds a unique (up to a constant factor) common eigenfunction V^(jc, Q) of Ln and Lm Ln-^ix, Q) = X\lf{x, Q)\ Lmif(x, Q) = fiir(x, Q). The logarithmic derivative V^^ V^~^ is a meromorphic function on P. In the general position (when r is smooth) it has g poles yi(jc),... , y^(jc) in the affine part of the curve, where g is the genus of P. The commuting operators L„ and Lm (in this case of co-prime orders) are uniquely defined by the polynomial R and by a set of g points yi(jco),... , y^(^o) on P. In such a form, the solution of the problem is one of pure classification: one set is equivalent to the other. Even the attempt to obtain exact formulae for the coefficients of commuting operators had not been made. Baker proposed making the program effective by pointing out that the eigenfunction -[jr has analytical properties that were introduced by Clebsch, Gordan and himself as a proper generalization of the notion of exponential
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 5 functions on Riemann surfaces. The Baker program was rejected by the authors of (Burchnall et al. [1922, 1928]) consciously (see the postscript of Baker's paper [1928]) and all these results were forgotten for a long time. This program was realized only in (Krichever [1976, 1977a]) (though at that time the author was not aware of the remarkable results of Burchnal, Chaundy and Baker) where the commuting pairs of ordinary differential operators were considered in connection with the problem of constructing solutions to the KP equation. Spatial two-dimensional integrable systems of the KP type have an analogue of the Lax representation of the form [dy-L,dt -A] = 0, (13) where, as before, L and A are ordinary differential operators of the form (2) but now with the coefficients depending on the variables x,y,tAn two dimensions in order to single out special classes of solutions for which a spectral curve can be defined one needs to impose two constraints. For example, that can be done if we assume that in addition to (13) there exist two ordinary differential operators of orders n and m such that [a^-L,L,] = 0, [a^-L,L^] = 0. (14) Such operators commute with each other, and commute with the operator (9? — A). The corresponding spectral curve is a spectral curve of commuting operators L„, L^. It does not depend on {x,y,t). (Classification of commuting operators of arbitrary orders was completed in Krichever [1978]). The common eigenfunction of commuting operators is a particular case of the general definition of the scalar multi-point multi-variable Baker-Akhiezer function. Let F be a non-singular algebraic curve of genus g with A^ punctures Pa and fixed local parameters k~^{Q)m neighbourhoods of these punctures. For any set of points yi,... , y^ in general position, there exists a unique (up to constant factor c{toi,i)) function ij/it, Q), t = (taj). Of = 1,... , A^ ; / = 1,... , such that: (i) the function i// (as a function of the variable Q e F) is meromorphic everywhere except for the points Pa and it has at most simple poles at the points yi,... , y^ (if all of them are distinct). (ii) in a neighbourhood of the point Pa the function i// has the form oo oo V^(r, Q) =txp{J2taA){J2^sAt)ka').ka =ka(Q). (15) i = l s=0 The Baker-Akhiezer function i// depends on the variables t = [tij,... , t^j} as on external parameters. From the uniqueness of the Baker-Akhiezer function it follows that for each pair (a, n) there exists a unique operator La,n of the form n-l 7 = 1
6 I. KRICHEVER where daj = 9/9^a,i, such that {da,i-La,n)^(t,Q)=0. (17) The idea of the proof of theorems of this type proposed in (Krichever [1976]) is universal. For any formal series of the form (15) there exists a unique operator La,n of the form (16) such that oo i = l The coefficients of Lc^^„ are differential polynomials with respect to ^^ c^. They can be found after substitution of the series (15) into (18). It turns out that if the series (15) is not formal but is an expansion of the Baker-Akhiezer function in the neighbourhood of Pa, then the congruence (18) becomes an equality. Indeed, let us consider the function iri={da,n-La,n)if(t,Q). (19) It has the same analytical properties as V^, except for one. The expansion of this function in a neighbourhood of Pa starts from 0(k~^). From the uniqueness of the Baker-Akhiezer function it follows that t/^i = 0 and the equality (17) is proved. A corollary is that the operators La,n satisfy the compatibility conditions The equations (20) are gauge invariant. For any function g(t) operators La,n = gLa,ng~^ + (^a,ng)g^~^^ have the same form (16) and satisfy the same operator equations (20). The gauge transformation corresponds to the gauge transformation of the Baker-Akhiezer function V^i(r, Q) = g(t)\l/(t, Q). In the one-point case the Baker-Akhiezer function has an exponential singularity at a single point Pi and depends on a single set of variables. Let us choose the normalization of the Baker-Akhiezer function with the help of the condition §o, i = 1, i-^- an expansion of i// in the neighbourhood of Pi is oo oo ir(tut2, ... , 2) = exp(^r,/:')(l +J^^,(t)k-'). (21) i=l s=l In this case the operator L„ has the form n-2 L, = af + ^t.f>a|. (22) If we denote ^i, ^2, t^ by jc, y, t, respectively, then from (20) it follows (for n = 2, m = 3) that u(x, y, r, ^4,...) satisfies the KP equation 3uyy = (4ut — 6uux + Uxxx)x- The exact formula for these solutions in terms of the Riemann theta-function is based on the exact formula for the Baker-Akhiezer function.
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 7 Let us fix the basis of cycles a,, Z?,, i = 1,... , ^ on F with the canonical matrix of intersections: at o aj = bi o bj = 0, at o bj = 8ij. The basis of normalized holomorphic differentials coj(Q), j = I,..., g is defined by conditions ^^ coj = 8ij. The b -periods of these differentials define the so-called Riemann matrix Bkj = fi^.^k- The basic vectors Ck of C^ and the vectors Bk, which are the columns of matrix B, generate a lattice BinC^. The ^-dimensional complex torus J(T) = C^/B, B = J2^kek+mkBk, rik.mkeZ, (23) is called the Jacobian variety of F. A vector with coordinates Ak(Q) = / cok defines the Abel map A : F —> /(F) which depends on the choice of the initial point qo. The Riemann matrix has a positive-definite imaginary part. The entire function of g variables 0(z) = 0(z\B) = J2 ^2^'^^'^>+^'^^^'^>, (24) meZs z = (zi,... , Zn). m = (mi,... , rUn), (z, m) = z\m\ + ... + z„m„, is called the Riemann theta-function. It has the following monodromy properties e{z + ck) = 0(z). 0(z + Bk) = ^-2^'^^-^'^^^^(^). (25) The function 0(A(Q) — Z) is a multi-valued function of Q. But according to (25), the zeros of this function are well-defined. For Z in a general position the equation 0(A(Q) - Z) = 0 (26) has g zeros yi,... , y^. The vector Z and the divisor of these zeros are connected by the relation g Zk = J^A(ys)+IC, (27) i=i where /C is the vector of Riemann constants. Let us introduce the normalized Abelian differentials dQ^ • of the second kind. The differential dQ^ ■ is holomorphic on F except for the puncture Pa. In the neighbourhood of Pa point it has the form J<, =J(4 + 0(1)). (28) "Normalized" means that it has zero a-periods, f^ dQ^ ■ = 0. Consider the function 0(A(Q)-\-J2ai^ccjUaj-Z) /-_, rQ „\ ir(t, Q) = ^^aj_i—J / \Ytai dOP^ . (29) ^''"^^ e{A{Q)-Z) \t7 4 '7 where the coordinates of the vector Uaj are equal to
8 I. KRICHEVER Equations (25-27) imply that t/^ is a single valued function on F and has all the analytical properties of the Baker-Akhiezer function. That proves the existence of the Baker-Akhiezer function. Let ^jr be any function with the same analytical properties. The ratio -ilr/1// is a meromorphic function with at most g poles. The Riemann-Roch theorem implies that such a function is equal to a constant. Hence, the uniqueness of the Baker-Akhiezer function (up to a constant factor) is also proved. The coefficients of the operators Laj which are defined by the equations (17) are differential polynomials in the coefficients of the expansions of the second factor in (29) near the punctures. Hence, they can be expressed as differential polynomials in terms of Riemann theta-functions. For example, the algebraic-geometrical solutions of the KP hierarchy have the form m(jc,>;, ^,^4,...) = 2d^lnO(xUi -\-yU2-\-tU3 H + Z)-\-const. (31) The common eigenfunction of commuting operators of co-prime orders is the particular case of a one-point Baker-Akhiezer function corresponding to ^i = jc, ^2 = 0, ^3 = 0, Therefore, the coefficients of such operators (in general position) are differential polynomials in terms of the Riemann theta-f unctions. This has an important corollary. The coefficients of commuting differential operators of co-prime orders are meromorphic functions of the variable x. Moreover, in general position they are quasi-periodic functions of X. The last statement presents evidence that the theory of commuting operators is connected with the spectral Floquet theory of periodic differential operators. These connections were missing in (Burchnall et al [1922, 1928], Baker [1928]). Let us introduce real normalized Abelian differentials dQ.a,i of the second kind. The differential dQ.oi,i is holomorphic on F except for the puncture Pa. In the neighbourhood of this point it has the same form as dOP^ ., i.e. dO^aj = d{¥^ + 0(1)). Real normalization means that for any cycle on F the period of the differential is pure imaginary, i.e. Kt{§^dQ.oc,i)=0. From (29) it follows that the algebro-geometric solutions corresponding to F are periodic functions of the variable taj with a period T if and only if the periods of the corresponding differential have the form Ini i dQaj = -zrnc, (32) T where nc SiTQ integers. The spectral theory of two-dimensional periodic operators was developed in (Krichever [1989]). It was proved that for the operator (dy — 9^ + m(jc, y)) with a real analytic periodic (in X and y) potential the spectral curve does exist. Points of this curve Q = (wi, W2) G F parametrize Bloch solutions of the equation (dy — 9^ + u(x, }?))V^(jc, y, ifi, W2) = 0, i.e. they parametrize pairs of complex numbers (wi,W2) such that there exists a solution to the equation with the following monodromy properties: '\lf(x-\-luy,wuW2) = wi'\lf(x,y,wuW2), i^ix, y-\-l2,wuW2) = W2ilf(x, y,wuW2). (33) In a general case the spectral curve has infinite genus. For the algebro-geometric potentials the spectral curve has finite genus and coincides with the spectral curve of a corresponding pair of the commuting differential operators.
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 9 The space of algebro-geometric data defining solutions of the full hierarchy of spatially two-dimensional KP type systems is infinite dimensional because it contains a choice of the local coordinates near the punctures. At the same time the space of algebro-geometric solutions of a single equation of the zero-curvature form (13) is finite-dimensional. If L and A are operators of orders n and m with scalar coefficients, then this space can be described as follows (see details in (Krichever et al [1997])). Let Mg{n, m) be the space (F, E, Q) of pairs of Abelian integrals on a smooth genus g algebraic curve F, where E and Q have poles of orders n and m, respectively, at a puncture Po • Then we define a local coordinate k~^ near the puncture by the equality k^ = E. This choice of the local coordinate corresponds to the identification of the variable y with a basic time variable y = tn. In the presence of a second Abelian integral Q, we can select a second time t, by writing the singular part Q-{-(k) of 2 ^s a polynomial in k and setting Q^(k) = aik-\ h flm^"", ti = ait, 1 <i <m. (34) This means that we consider the Baker-Akhiezer function -{//(x, y,t; k) with the essential singularity exp(kx -\-k^y -\- Q^(k)t), and construct the operators L and A by requiring that (dy - L)\lr = (dt — A)\lr = 0. The pair (L, A) provides then a solution of the zero-curvature equation. By rescaling t, we can assume that A is monic. The proper interpretation of the full geometric data (F, £", 2; yi, • • • y^) is as a point in the bundle J\fg(n, m) over Mg{n, m), whose fiber is the ^-th symmetric power S^{T) of the curve: Ml(n,m)^^ Mg(n,m) (35) The ^-th symmetric power can be identified with the Jacobian of F via the Abel map. More generally, we can construct the bundles Afg(n, m) with fiber S^{T) over the bases Mg(n,m). Thus the bundle J\f^^^(n, m) =Mg{n, m) is the analogue in our context of the universal curve. 1.3 Whitham Equations We have seen that soliton equations exhibit a unique wealth of exact solutions. Nevertheless, it is desirable to enlarge the class of solutions further, to encompass broader data than just rapidly decreasing or quasi-periodic functions. Typical situations arising in practice can involve Heaviside-like boundary conditions in the spatial variable jc, or slowly modulated waves which are not exact solutions, but can appear as such over a small scale in both space and time. The non-linear WKB method (or, as it is now also called, the Whitham method of averaging) is a generalization to the case of partial differential equations of the classical Bogolyubov-Krylov method of averaging. This method is applicable to nonlinear equations which have a moduli space of exact solutions of the form uo(Ux + Wt + Z\I). Here uoizi,... ,Zg\I) is a periodic function of the variables Zi\ U = (f/i,... , Ug), W = (Wi, ... , W^) are vectors which like u itself, depend on the parameters / = (/i,... , /yv).
10 I. KRICHEVER i.e. U = U(I), W = W(I). These exact solutions can be used as a leading term for the construction of asymptotic solutions m(jc, t) = uo(6~^S(X, T) + Z(X, r)|/(X, T)) + suiix, t) + s'^U2{x, t) + - • • , (36) where / depend on the slow variables X = sx.T = st and and £ is a small parameter. If the vector-valued function S{X, T) is defined by the equations dxS = U(I(X,T)) = U(X,T), dTS=W(I(X,T)) = W(X,T), (37) then the leading term of (36) satisfies the original equation up to order one in s. All the other terms of the asymptotic series are obtained from the non-homogeneous linear equations, whose homogeneous part is just the linearization of the original non-linear equation on the background of the exact solution mq- In general, the asymptotic series becomes unreliable on scales of the original variables x and t of order 6~^.ln order to have a reliable approximation, one needs to require a special dependence of the parameters /(X, T). Geometrically, we note that 6~^S(X, T) agrees to first order with Ux + Vt, and jc, ^ are the fast variables. Thus u{x,t) describes a motion which is to first order the ongmdX fast periodic motion on the Jacobian, combined with a slow drift on the moduli space of exact solutions. The equations which describe this drift are in general called Whitham equations, although there is no systematic scheme to obtain them. One approach for obtaining these equations in the case when the original equation is Hamiltonian is to consider the Whitham equations as also Hamiltonian, with the Hamiltonian function being defined by the average of the original one. In the case when the phase dimension g is greater than one, this approach does not provide a complete set of equations. If the original equation has a number of integrals one may try to get the complete set of equations by averaging all of them. This approach was used in (Flashka et al. [1980]) where Whitham equations wqyq postulated for the finite-gap solutions of the KdV equation. The Hamiltonian approach for the Whitham equations of (l+l)-dimensional systems was developed in (Dubrovin et al. [1983]) where the corresponding bibliography can also be found. In (Krichever [1988]) a general approach for the construction of Whitham equations for finite-gap solutions of soliton equations was proposed. It is instructive enough to present it in the case of zero-curvature equation (13) with scalar operators. Recall from the previous section that the coefficients m,(jc, y, t), Vj{x, y, t) of the finite-gap operators Lq and Aq satisfying (13) are of the form Ui = Ui^oiUx -\-Vt-\-Wt-\- Z\I), Vj = Vj^oiUx -\-Vt-\-Wt-\- Z\I), (38) where m,,o and Vj^o are differential polynomials in ^-functions and / is any coordinate system on the moduli space Mg(n, m). (A helpful example is provided by the solutions (31) of the KP equation, where / is the moduli of a Riemann surface, and U, V, W are the i5)t-periods of its normalized differentials J^i, ^^2, and ^^3.) We would like to construct operator solutions of (13) of the form L = Lo + £Li + • • • , A = Ao + £Ai + ... , (39)
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 11 where the coefficients of the leading terms have the form ui = ui^o{8-^s{x, y, T) + z(x, y, r)|/(x, y, r)), vj = vj,o{8-^s{x, y, T) + z(x, y, r)|/(x, y, r)) (40) If / is a system of coordinates on M.g(n,m), then we may introduce a system of coordinates (z, /) on Afg(n, m) by choosing a coordinate along the fiber P. The Abelian integrals p, E, Q arc multi-valued functions of (z, /), i.e. p = p(z, I), E = E(z, I), Q = (2(z, /). If we describe a drift on the moduli space of exact solutions by a map (X, y, D ^ / = /(X, y, r), then the Abelian integrals p, E, Q become functions of (z, X, y, r). The following was established in (Krichever [1988]): A necessary condition for the existence of the asymptotic solution (4) with leading term (5) and bounded terms L\ and A\ is that the equation \dT dY J dz \dT dx) dz \dY dX) dp (dE Vz is satisfied. The equation (41) is called the Whitham equation for (13). It can be viewed as a generalized dynamical system on Mg(n, m), i.e., a map (X, y, T) -^ Mg(n,m). Some of its important features are: • Even though the original two-dimensional system may depend on y, Whitham solutions which are y-independent are still useful. As we shall see later, this particular case has deep connections with topological field theories. If we choose the local coordinate z along the fiber as z = £", then the equation simplifies to dTP = dxQ. (42) • Naively, the Whitham equation seems to impose an infinite set of conditions, since it is required to hold at every point of the fiber P. However, the functions involved are all Abelian integrals, and their equality over the whole of P can actually be reduced to a finite set of conditions. • The equation (41) can be represented in a manifestly invariant form without explicit reference to any local coordinate system z. Given a map (X, y, T) -^ Mg{n, m), the pull-back of the bundle A/'^H'^, m) defines a bundle over a space with coordinates X, y, T. The total space M^ of this bundle is 4-dimensional. Let us introduce on it the one-form a = pdX + EdY + QdT, (43) Then (41) is equivalent to the condition that the wedge product of da with itself be zero (as a 4-form on J\f^) da Ada = 0. (44)
12 I. KRICHEVER • It is instructive to present the Whitham equation (41) in yet another form. Because (41) is invariant with respect to a change of local coordinate we may use p = p(z, I) by itself as a local coordinate. Then we may view E and Q as functions of p, X,Y and T, i.e. E = E(p, X, y, r), Q = Q(p, X, F, T). With this choice of local coordinate (41) takes the form dTE-dYQ + [E,Q} = 0, (45) where {•, •} stands for the usual Poisson bracket of two functions of the variables p and X, i.e. {/,^} = fpgx -gpfx^ • Above we had focused on constructing an asymptotic solution for a single equation. This corresponds to a choice of A, and thus of an Abelian differential Q, and the Whitham equation is an equation for maps from (X, Y, T) to Mg(n, m). As in the case of the KP and other hierarchies, we can also consider a whole hierarchy of Whitham equations. This means that the Abelian integral Q is replaced by the real normalized Abelian integral Q^i which has the following form Q^i =k' -\-0(k-^), r = E, in a neighbourhood of the puncture P. The whole hierarchy may be written in the form (44) where we set now i In (Krichever [1988]) a construction of exact solutions to the Whitham equations (41) was proposed. We present the most important special case of this construction, which is also of interest to topological field theories and supersymmetric gauge theories. It should be emphasized that for these applications, the definition of the hierarchy should be slightly changed. Namely, the Whitham equations describing modulated waves in soliton theory are equations for Abelian differentials with a real normalization. In what follows we shall consider the same equations, but where the real-normalized differentials are replaced by differentials with the complex normalization §^ dQ = 0. The two types of normalization coincide on the subspace corresponding to M-curves, which is essentially the space where all solutions are regular and where the averaging procedure is easily implemented. Thus, the two forms of the Whitham hierarchy can be considered as different extensions of the same hierarchy. The second one is an analytic theory, and we shall henceforth concentrate on it. In the rest of this chapter we shall restrict ourselves to the hierarchy of "algebraic geometric solutions" of Whitham equations, that is, solutions of the following stronger version of the equations (45) dT,E = {Qi,E]. (46) We note that the original Whitham equations can actually be interpreted as consistency conditions for the existence of an £" satisfying (46). Furthermore, the solutions of (46) can be viewed in a sense as 'T-independent" solutions of Whitham equations. They play the same role as Lax equations in the theory of (2+l)-dimensional soliton equations. As stressed earlier, F-independent solutions of the Whitham hierarchy can be considered even for two-dimensional systems where the };-dependence is non-trivial in general.
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 13 Equations (46) define a system of commuting flows on the moduli space of Abelian integrals. For the one puncture case this space is a union of the spaces Mg(n) of Abelian integrals with the pole of order n at the puncture. The complex dimension of Mg(n) is equal to dim A4g(n) = 4g -\-n — I. Let us describe a special system of coordinates for it. The first 2g coordinates are still the periods of dE, Ta,,e = <p dE, Tb,,e = <p dE. (47) The differential dE has 2g -\- n — 1 zeros (counting multiplicities). When all the zeroes are simple, we can supplement (47) by the 2^ + n — 1 critical values Es of the Abelian differential £", i.e. Es = E{qs). dE(qs) =0, 5 = 1, ... , 2^ + n - 1. (48) Let V^ be the open set in Mg(n) where the zero divisors of dE and dp, namely the sets {z\dE(z) = 0} and {z\dp(z) = 0}, do not intersect and where all zeros of dE are simple. As shown in Krichever et al [1997], the set (Ta^^e, Tb^^e, Es) define a local coordinate system on The Whitham equations (46) define a system of commuting meromorphic vector fields (flows) on A4g(n) which are holomorphic on V C M.g(n) and have the form a a do^i — Ta^,e=0, —Tb^^e=0, dT^Es = -^(qs)dxEs. (49) aTj aTj ^ dp An important consequence of (49) is that the space Mg(n) admits a natural foliation by the joint level sets of the functions 7a.,£, T^^. £. The leaves of the foliation are smooth (2g -\-n — 1)-dimensional submanifolds, and are invariant under the flows of the Whitham hierarchy (46). A special case of the construction of exact solutions to (46) in [Krichever [1988]) may now be described as follows: the moduli space Mg{n,m) provides the solutions of the first n + m-flows of (46) parametrized by 3^ constants, which are the set ^Ai,Q = (p dQ, Tb,,q = (p dQ, at = (h QdE. JAi JBi JAi Let us consider the joint level set of functions (47, 50). Then the functions (50) Ti = -Resp,{E-''''QdE) (51) define coordinates on its open set V^ where the zero divisors of dE and dQdiO not intersect. The projection Mgin, m) -^ Mgin) : (F, E, Q) h^ (F, E) (52) defines (F, £") as a function of the coordinates onMg{n,m). For each fixed set of parameters TAi,E^ TBi,E,TAi,Q,TAi,Q, «/, the map (7^))^i~^^ -^ Mg(n) satisfies the Whitham equations (46).
14 I. KRICHEVER For the proof of this statement it is enough to note that if we use E(z) as a local coordinate on F, then as we saw earlier, the equations (46) are equivalent to the equations dTiP(E, T) = dx^i(E, T). These are the compatibility conditions for the existence of a generating function for all the Abelian differentials dO^i. In fact, if we set dS = QdE, (53) then it turns out that dT,dS = dQi, dxdS = dQ, (54) (For the proof of (54), it is enough to check that the right and the left hand sides of it have the same analytical properties.) Consider now the second Abelian integral 2 as a function of the same parameters Tt, 1 < / < n -\-m. Then Q(p, T) satisfies the same equations as £", i.e. dT,Q = {ni,Q}. (55) Furthermore, {E.Q} = L (56) We note that (56) can be viewed as a Whitham version of the so-called string equation (or Virasoro constraints) in a non-perturbative theory of 2-d gravity (Douglas [1990], Witten [1991]). The solution of the Whitham hierarchy can be summarized in a single r-function defined as follows. The Icey underlying idea is that suitable submanifolds of Mg(n,m) can be parametrized by Whitham times Ta, to each of which is associated a "dual" time Tda^ and an Abelian differential J^a, which generates with the help of equation (46) the Ta-Aow. Recall that the coefficients of the pole of dS determine n-\-m Whitham times (51). Their dual variables are TDj = Resp{z-JdS), (57) and the associated Abelian differentials are the familiar dO^i of (28) (complex normalized). When ^ > 0, the moduli space Mg (n, m) has in addition 5^ more parameters. We consider only the foliations for which 3^ parameters 7a,,£, T'^.,^, and Ta^^q (defined by (47, 50) are fixed. Thus the case ^ > 0 leads to two more sets of g Whitham times, namely each ak and TBk,Q- Their dual variables are aok = -:^^ <p dS, T^j^ =—^ (p EdS. In I Jb, '"' 2ni Ja~ (58) (Because EdS has a jump on the cycle, one has to be careful in choosing a side of integration. The superscript A^ here means the left hand side of the cut with respect to the natural orientation.) The corresponding Abelian differentials are respectively the holomorphic differentials dcok and the differentials dQ^, defined to be holomorphic everywhere on r except along the Aj cycles, where they have discontinuities J^f + - dQ^~ = SjkdE. (59)
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 15 We denote the collection of all 2g -\- n -\-m times by Ta = (Tj.ak, T^ = Tb^^q). We can now define the r-function of the Whitham hierarchy by 1 1 ^ inr(r) = nr) = ^Y.^^^^^ + -r-'Y.''^^k E^^knBk). (60) where Ak H Bk is the point of intersection of the Ak and Bk cycles. Note that the definition of the r-function for a general case of the universal Whitham hierarchy (for which a corresponding moduli space is the space of curves with fixed pair of Abelian integrals with several poles) is given by the same formula. The only difference is that there are more times and more corresponding differentials (see (Krichever et al [1997, 1999]). As shown in Krichever [ 1994] the derivatives of T with respect to the 2^+n+m Whitham times Ta are given by 1 ^ ^T.J" =TDA + -;—y\ Sa,,AT^E{Ak n Bk). dl,A^ = ^ hiAk n Bk)8^E,k),A - i d^A , (61) ai,c^=E''-..r-^^^jii^j These formulae show that the r-functions encodes the whole hierarchy, because the coefficients of expansions of the differentials at the puncture, as well as their periods are given by derivatives of r. (Note that formulae (61) require some modifications in the multi-puncture case for differentials with nonzero residues (see D'Holcer et al. [1997]).) 1.4 Topological Landau-Ginzburg Models on Riemann Surfaces In general, a two-dimensional quantum field theory is specified by the correlation functions < 0(zi) • • • 0(z7v) >^ of its local physical observables (pi (z) on any surface F of genus g. Here 0/ (z) are operator-valued tensors on P. The operators act on a Hilbert space of states with a designated vacuum state |^ >. Topological field theories are theories where the correlation functions are actually independent of the insertion points Zi. Thus they depend only on the labels of the fields 0, and the genus ^ of P. This independence implies that for all practical purposes, the operator product (l)i{zi)(j)j{zj) can be replaced by the formal operator algebra k
16 I. KRICHEVER The associativity of operator compositions translates into the associativity of the operator algebra (62). Furthermore, the operator algebra is commutative. As shown in (Dijkgraaf et al. [1990,1991a, 1991b], the partition function J^(xi,... , jc„) for the marginal deformations of a topological field theory with n primary fields 0i,... , 0^ satisfies an overdetermined system of equations which are equivalent to the condition that the commutative algebra with generators (pk and the structure constants defined by the third derivatives of J^: CklmM = -J -1 , (63) cl>kcl>i = c^iix) (l>m\ c^i = cm rf"^'. mrl"^ = K^ (64) is an associative algebra, i.e. c\.{x)c[^{x) = c]^{x)c\^{x) (65) In addition, it is required that there exist constants r^ such that the constant metric r] in (64) is equal to mi =r'^ckim(x). (66) In terms of J^ the conditions (65) become a system of non-linear equations called the Witten-Dijkgraaf-Verlinde-Verlinde (WDVV) equations. In recent years these equations have become a key element of the theory of Gromov-Witten invariants and have been applied for solving various problems of enumerative geometry. In the original work (Dijkgraaf et al. [1991a,b]) a solution to WDVV equations for some topological Landau-Ginzburg theories was found. In Krichever [1992], it was noted that the calculation of Dijkgraaf [1991], are similar to the construction of solutions to the dispersionless Lax equations which are the zero genus case of the Whitham hierarchy. The results of Krichever [1992] were generalized for higher genus Whitham hierarchies for Lax equations in Dubrovin [1992]. The general case of the universal Whitham hierarchy was considered in Krichever [1994]. Let us consider the space Mg (n) = {F, E] of normalized Abelian integrals on genus g curves with a single pole of order n at the puncture (for simplicity, we consider only the one-puncture case). As before, we identify Mg(n) with Mg(n, 1) by the choice dQ = dQi. The relevant leaf within Mg(n, 1) is of dimension n — l-\-2g and is given by the constraints n Tn = 0, r„+i = AZ + 1 (f) dE =0, (f) dE = fixed, (f) dQ=0. (67) The leaf is parametrized by the {n — \) Whitham times rA,A = l,•••,« — 1, and by the periods Uj and T^ = Tbj,q defined by (50). The fields 0a of the theory can be identified with dQj/dQ. We take the 2g additional fields to be given by dcoj/dQ and dO^J/dQ, where the differentials dcoj and J^^ are the ones associated to aj and T^, as described earlier.
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 17 Let rjAB and cabc be defined by E dQAdQB \-^ „ dQAdQBdQc .^ox '''''''-^^' '^'^ -L^^^.^ dEdQ ' ^^^^ where ^^ are the zeroes of dE, and the indices A,B,C are running this time through the augmented set of n — 1 + 2g indices given by Ta = (Tt, aj, T^). Then r]ij = Sj-^j^n^ Vaj,iE,k) = ^j,k' AH Other pairings vanish. The algebra 0a 05 = ^a^^c can be identified with the algebra of functions at zeros qs of the differential dE which is obviously associative. From (61) it follows that ^AfiC-^^^) ~ CABC' We have tjab = ciab, also. Therefore, the r-function of the Whitham hierarchy J^(Ti, aj, Tf) restricted to the leaf (67) is a solution of the WDVV equations. Remarkably, the larger spaces Mg(n,m) can accommodate the gravitational descendants of the fields 0a. More precisely, consider for ^ = 0 the leaf of the space Mo(n, mn + 1) given by the following normalization ^^ Tin =0, / = !,..., m, r^m+i = nm + 1 The space of Whitham times is automatically increased to the correct number by taking all the coefficients of QdE. The additional m(n — I) fields may be identified with the first m gravitational descendants of the primary fields. Namely, the p-ih descendant <yp{(j)i) of the primary field 0/ is just dQpn-{-i/dQ. This statement is a direct corollary of the following result proved in Krichever [1994]. The correlation functions given by < (I)a(I>b(I>c >= ^abc^ ^^^^ cfp((j)i) = dQt-^pn/dQ satisfy the factorization properties for descendant fields < ap((j)i)(l)B(l)c > = < cfp-\{(l)i)(l)j > rjJ^ < (l)k(l>B(l>c >, where (l)i,i = 1, ... , n — 1 are primary fields, and(l)A are all fields (including descendants). Factorization properties for descendant fields were derived by Witten [1988a, 1988b, 1991, 1992]. 1.5 Seiberg-Witten Solutions of N=2 SUSY Gauge Theories Moduli spaces of geometric structures are appearing increasingly frequently as the key to the physics of certain supersymmetric gauge or string theories. One recurring feature is a moduli space of degenerate vacua in the physical theory. The physics of the theory is then encoded in a Kahler geometry on the space of vacua, or, in presence of powerful constraints such as N=2 supersymmetry, in an even more restrictive special geometry, where the Kahler potential is dictated by a single holomorphic function ^, called the prepotential. In Seiberg [1994a,b] Seiberg and Witten introduced the following fundamental ansatz that for A^ = 2 SUSY gauge theories: (i) the quantum moduli space should be parametrized a family of Riemann surfaces F (a), now known as the spectral curves of the theory;
18 I. KRICHEVER (ii) on each r(fl), there is a meromorphic one-form dX, such that its derivatives along the moduli space are meromorphic differentials; (iii) J^ is determined by the periods of dX ak = Q) dX, aD,k = z-r Q) dX, -— = aD,k- JAk 271 i Jb, dak (69) In Gorsky et ah [1995], it was noticed that the moduli space of curves for SU(N) theories can be identified with spectral curves of the A/^-periodic Toda lattice. It was also noted that the generating differential dX coincides with the generating differential dS (c.f. 53) of the Whitham hierarchy. A general approach for solution of the Seiberg-Witten ansatz was developed in Krichever et al [1997]. We present here Icey elements of this approach. Let now n = iria), m = {nia), a = 1,... , N, be multi-indices, and Mg(n, m) be the moduli space (F, £", Q) of pairs of Abelian differentials on F with poles of orders ria and rua at punctures Pa. The dimension of this space is equal to dimMgin, m) = 5g - 3-\-3N -\- Y^(na + m^). (70) The Whitham coordinates on this space can be introduced in a similar way to the one puncture case. The Abelian integral E defines a coordinate system Za near each Pa by E=Za''"-\-R^logZa. (for simplicity we assume that ria is strictly positive). Then the formulae Taj = --Resp^{z'aQdE), Ta,o = Resp^(QdE), (71) I define I]^^i(AXa -\- rUa) -\- N - I parameters (X!a Ta,o = 0). The remaining parameters needed to parametrize Mg (n,m) consist of the 2N—2 residues of dE and dQ R^ = RespJE, R^ = RespJQ, of = 2, • • • , A^, (72) and 5^ parameters which are the periods of dE, dQ and a-periods of dS = qdE given by (47, 50). In Krichever et al [1997], it was shown that the joint level sets of all parameters except ak = §j^ dS define a smooth foliation of the open seiV^ of Mg(n,m), which is independent of the choices made to define the coordinates themselves. This intrinsic foliation is central to the Seiberg-Witten theory and the Hamiltonian theory of soliton equations. We shall refer to it as the canonical foliation. Our goal is to construct now a symplectic form co on the complex 2^-dimensional space obtained by restricting the fibration J\fg(n, m) to a ^-dimensional leaf M of the canonical foliation of M.g{n,m).
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 19 Let us consider the Abelian integrals E and Q as multi-valued functions on the fibration. Despite their multivaluedness, their differentials along any leaf of the canonical fibration are well-defined. In fact, E and Q are well-defined in a small neighbourhood of the puncture Pi. The ambiguities in their values anywhere on each Riemann surface consist only of integer combinations of their residues or periods along closed cycles. Thus these ambiguities are constant along any leaf of the canonical foliation, and disappear upon differentiation. The differentials along the fibrations obtained this way will be denoted hy 8E and 8Q. Restricted to vectors tangent to the fiber, they reduce to the differentials dE and dQ. These arguments show that (Krichever et al. [1997]) the following two-form on the fibration M^in, m) restricted to a leaf M. of the canonical foliation of M.g{n,m) g g COM = MI] 2(K/M^(n)) = I]5e(n) A dEin) (73) defines a holomorphic symplectic form which is equal to g COM = /_] ^^i ^ ^^i' ^'^^^ i = l where (pk are canonical coordinates on the Jacobian of the curve. Note that the first set of formulae (61) implies that the restriction of the logarithm of the r-function of the Whitham hierarchy on a leaf of the canonical foliation satisfies relations (69) for the prepotential, and therefore, the function J^(T) given by (60) is a solution of the Seiberg-Witten ansatz. Although the results presented above suggest deep relations between N=2 gauge theories, soliton equations, their Whitham theory, and Landau-Ginzburg type models, such relations are still not fully understood at the present time. Nevertheless, the parallels between these fields allows us to apply to the study of the prepotential J^ of gauge theories the methods developed in the theory of solitons. In D'Hoker et al. [1997], with the help of these methods, the renormalization group equation for the prepotential J^ for SU(A^c) gauge theories with Nf < 2Nc hypermultiplets of masses my in the fundamental representation was derived. It was shown that this equation is powerful enough to generate explicit expressions for the contributions of instanton processes to any order. We conclude this chapter by a discussion of connections of the symplectic form (73) with the Hamiltonian theory of soliton equations. The Hamiltonian theory of finite-dimensional and spatial one-dimensional soliton equations is a rich subject which has been developed extensively over the years (see Faddeev et al. [1987], Diclcey [1991]). However, until recently much less was known about the 2D case. In Krichever et al. [1997, 1999], a new algebro-geometric approach to the Hamiltonian theory of soliton equations was developed. This approach is uniformly applicable for all integrable systems: finite-dimensional, spatial one- or two-dimensional evolution equations. Its universality is based on a universal symplectic form which can be defined on a space of operators in terms of operators and their eigenfunctions, only. For simplicity we consider here the Lax equations (1) for the operators (2) with scalar coefficients.
20 I. KRICHEVER Let V^ (jc, /:) be a formal solution of the form V.(x, k) = /^(l + ^§,(x)/:-0 (75) to the equation Li// = k^i/r, normalized by the condition ijfiO.k) = 1. The coefficients hi of the expansion oo dAnjlf = k-\-J2hsk~' (76) s=l are differential polynomials in the coefficients m, of L, i.e hi = hi(u). They are densities of integrals of motions Hs = J hi(u)dx of the Lax equation (1). Let us introduce the dual formal solution oo r=e-^'(\ + Y.^t{x)k-') (71) s=l of the formal adjoint equation V^*L = /:"t/^*, normalized by the condition / V^* t/^Jjc = 1. The main ingredients are the one-forms 8L and S^o- The one-form 5L is given by n-2 8L = J2^^iK^ i=0 and can be viewed as an operator-valued one-form on the space of operators L. Similarly, the coefficients of the series i// are explicit integro-differential polynomials in m,. Thus Si// can be viewed as a one-form on the space of operators with values in the space of formal series. Consider the following two-form on the space of operators L ( f(if''8LA8ilf)dxjdp, CO = Resoo ( / (V^*5L A 8ilf)dx j dp, (78) where p = k -\- ^^ Hsk~^. In Krichever et al. [1999] it was shown that on the subspaces of the operators L defined by the constrains {Hi = const, / = 1,... , n — 1} the form co (i) defines a symplectic structure, i.e, a closed non-degenerate two-form; (ii) the form co is actually independent of the normalization point {x = 0) for the formal Bloch solution i/rix, k); (iii) the flows (1) are Hamiltonian with respect to this form, with the Hamiltonians InHm+niu). Consider now the leaves A^^ of the canonical foliation on M.g{n, 1) corresponding to zero values of variables § dE = 0. These leaves correspond to spectral curves of one-dimensional finite-gap Lax operators, i.e. we have a geometric map of the Jacobian bundle A/^ over M^ to the space of operators g:Af^ ^ (L).
BAKER-AKHIEZER FUNCTIONS AND INTEGRABLE SYSTEMS 21 A connection of the Hamiltonian theory of soliton equations with the previous construction of algebro-symplectic structures associated with the Seiberg-Witten theory was established in Krichever [1997]. Namely, it was proved that (iv) the restriction of the symplectic form co given by (78) via the geometric map to the leaf M^ of the canonical foliation is holomorphic symplectic form equals g The results presented above are a particular case of more general settings. It turns out the the algebro-geometric symplectic structure (73) on the general leaves of the canonical foliation on Mg (n, 1) is a restriction of the basic symplectic structure for 2D soliton equations. The symplectic structure on leaves of the foliation for Mg (n, m) for m > 1 is the restriction of higher symplectic forms for soliton equations. It is necessary to emphasize that the same soliton equations are Hamiltonian flows with respect to all these structures but generated by different Hamiltonians. A variety of examples which show the universality of our approach can be found in Krichever et al. [1997, 1999]. Acknowledgement Research supported in part by National Science Foundation under the grant DMS-98-02577 and by the grant RFFI-98-01-01161. References Baker, H.F, Note on the foregoing paper "Commutative ordinary differential operators", Proc. Royal Soc. London 118, 584-593 (1928). Burchnall, J.L., and Chaundy, T.W., Commutative ordinary differential operators. I, Proc. London Math Soc. 21, 420-440 (1922). Burchnall, J.L., and Chaundy, TW., Commutative ordinary differential operators. II, Proc. Royal Soc. London 118, 557-583 (1928). D'Hoker, E., Krichever, I., and Phong, D.H., The renormalization group equation for N=2 supersymmetric gauge theories, Nucl. Phys. B494, 89-104, hep-th/9610156 (1997). Dickey, L.A., Soliton equations and Hamiltonian systems. Advanced Series in Mathematical Physics, Vol. 12 (1991) World Scientific, Singapore. Dijkgraaf, R., and Witten, E., Mean field theory, topological field theory, and multi-matrix models, Nucl. Phys. B 342, 486-522 (1990). Dijkgraaf, R., Verlinde, E., and Verlinde, H., Topological strings in ^ < 1, Nucl. Phys. B 352, 59-86 (1991). Dijkgraaf, R., Verlinde, E., and Verlinde, H., Notes on topological string theory and 2D quantum gravity, in "String theory and quantum gravity". Proceedings of the Trieste Spring School 1990, M. Green (ed), World-Scientific, 1991. Douglas, M., and Shenker, S., Strings in less than one dimension, Nucl Phys. B 335, 635-654 (1990). Dubrovin, B., and Novikov, S., The Hamiltonian formalism of one-dimensional systems of the hydrodynamic type, and the Bogoliubov-Whitham averaging method, Sov. Math. Doklady 27, 665-669 (1983).
22 I. KRICHEVER Dubrovin, B., Hamiltonian formalism of Whitham-type hierarchies and topological Landau-Ginzburg models, Comm. Math, Phys. 145, 195-207 (1992). Dubrovin, B., Matveev, V., Novikov, S., Non-linear equations of Korteweg-de Vries type, finite zone linear operators and Abelian varieties, Uspekhi Mat. Nauk 31(1), 55-136 (1976). Faddeev, L., and Takhtajan, L., "Hamiltonian methods in the theory of soliton". Springer-Verlag, 1987. Flaschka, H., Forrest, M., and McLaughlin, D., Multiphase averaging and the inverse spectral solution of the Korteweg-de Vries equation, Comm. PureAppl. Math. 33, 739-784 (1980). Gorsky, A., Krichever, I., Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B 355, 466-474 (1995). Krichever, I., and Phong, D.H., On the integrable geometry of soliton equations and N = 2 supersymmetric gauge theories, J. Differential Geometry 45, 349-389 (1997). Krichever, I., and Phong, D.H., Symplectic forms in the theory of solitons. Surveys in Differential Geometry: Integral Systems, vol. 4, 239-313. International Press, Boston (1999). Krichever, I., Averaging method for two-dimensional integrable equations, Funct. Anal. Appl. 22, 37-52 (1988). Krichever, I., Methods of algebraic geometry in the theory of non-linear equations, Russian Math Surveys 32, 185-213 (1977a). Krichever, I., Spectral theory of two-dimensional periodic operators and its applications, Uspekhi Mat. Nauk 44(2), 121-184 (1989). Krichever, I., The r-function of the universal Whitham hierarchy, matrix models, and topological field theories, Comm. PureAppl. Math. 47, 437^75 (1994). Krichever, I., The algebraic-geometric construction of Zakharov-Shabat equations and their solutions, Doklady Akad. Nauk USSR 227, 291-294 (1976). Krichever, I., The commutative rings of ordinary differential operators. Funk. anal, ipril. 12(3), 20-31 (1978). Krichever, I., The dispersionless Lax equations and topological minimal models, Comm. Math. Phys. 143(2), 415^29 (1992). Krichever, L, The integration of non-linear equations with the help of algebro-geometric methods. Funk. anal, ipril. 11(1), 15-31 (1977b). Lax, P., Periodic solutions of Korteweg'de Vries equation, Comm. Pure and Appl. Math. 28, 141-188 (1975). McKean, H., and van Moerbeke, P., The spectrum of Hill's equation. Invent. Math. 30, 217-274 (1975). Novikov, S., Periodic Problem for the Korteweg-de Vries equation, Funct. anal, i pril. 8(3), 54-66 (1974). Seiberg, N., and Witten, E., Electric-magnetic duality, Monopole Condensation, and Confinement in N = 2 Supersymmetric Yang-Mills Theory, Nucl. Phys. B 426, 1952, hep-th/9407087 (1994). Seiberg, N., and Witten, E., Monopoles, Duality and Chiral Symmetry Breaking in N = 2 Supersymmetric QCD, Nucl. Phys. B 431, 484-550, hep-th/9410167 (1994). Witten, E., Topological Quantum Field Theory, Comm. Math. Phys. Ill, 353-386 (1988). Witten, E., Topological sigma models, Comm. Math. Phys. 118, 411^49 (1988). Witten, E., Two-dimensional gravity and intersection theory on moduli space. Surveys in Differential Geometry 1, 281-332 (1991). Witten, E., Mirror manifolds and topological field theory, in "Essays on Mirror Manifolds", ed. by S.T. Yau, International Press (1992), Hong-Kong, 120-159.
Algebraic Geometry, Integrable Systems, and Seiberg-Witten Theory EYAL MARKMAN Department of Mathematics and Statistics, University of Massachusetts Amherst, MA, USA e-mail: emarkman @ math, umass.edu This survey chapter reviews the relationship between integrable systems and the moduli spaces of A^ = 2 supersymmetric Yang-Mills theories in 4 dimensions. Our starting point is the special Kahler structure on these moduli spaces. We then review the geometry of the generalized Hitchin systems. Donagi and Witten conjectured that a particular Hitchin system produces a special Kahler structure equal to the one coming from the moduli of the supersymmetric Yang-Mills theory when the gauge group is SU(n). We conclude with a closer look at this particular Hitchin system. 2.1 Introduction Physical arguments imply that the moduli space B of N = 2 supersymmetric Yang-Mills theories in 4 dimensions is a finite dimensional Kahler manifold endowed with an extra structure called a special Kahler structure. The mysterious connection between these Yang-Mills theories and integrable systems is formally explained by Theorem 2.1 (see Donagi and Witten [1996], Donagi [1997] and Freed [1999], Theorem 3.4) The following data are equivalent: L the data given by an integral special Kahler structure on a complex manifold B, and 2. the data given by an algebraically completely integrable hamiltonian system A^- B, Algebraic integrable systems are defined in section 2.2. The definition of a special Kahler structure and the derivation of Data 2 from Data 1 is sketched in Section 2.3. In Section 2.4 we will briefly explain how a special Kahler structure is defined on the base of every algebraic integrable system. In view of the above equivalence, it is natural to look for a 23
24 E. MARKMAN known integrable system A ^^ B, which special Kahler structure on B is the one coming from supersymmetric Yang-Mills theory. Such an identification enables one to compute the real-world observables of the N = 2 supersymmetric Yang-Mills theories ("the low energy effective Lagrangian") in terms of periods of a family of abelian varieties. Donagi and Witten conjecture that the integrable system for the SU(n) theory is a generalized Hitchin system [1996]. In Sections 2.5 we review the general construction of these systems and their Poisson structure. In Section 2.6 we describe the complete integrability of the symplectic leaves of the generalized Hitchin systems. Section 2.7 is devoted to the geometry of the particular Hitchin system considered by Donagi and Witten: the moduli space of KP elliptic solitons of order Az. The moduli space of KP elliptic solitons fits also in a family of integrable systems associated to root systems; the Calogero-Moser systems with elliptic parameter. The Calogero-Moser systems are conjectured to provide the identification of the special Kahler structure of the supersymmetric Yang-Mills theories for all simple groups. Lax pairs for these systems are described in Bordner et ah [1999] and Hoker and Phong [1998]. An interpretation of the Calogero-Moser systems as Hitchin systems is worked out in Hurtubise and Markman [1999]. Most of the material below is contained in the excellent survey of Donagi [1997]. One exception is a sufficient condition for the generic fiber of the Hitchin map in a given symplectic leaf to be smooth and compact (Proposition 2.10 and Corollaries 2.11 and 2.13). We include this more detailed discussion in response to questions asked at the workshop. 2.2 Algebraically Completely Integrable Systems An algebraically completely integrable Hamiltonian system is a complex algebraic variety M (the phase space) with a holomorphic symplectic structure a (a (2, 0)-form) and a lagrangian fibration tt : M ^^ B. Given a function / on M, the symplectic structure translates the 1-form df to a (Hamiltonian) vector field. The algebra of functions on the base B gives rise to a maximal sub-algebra of commuting vector fields on M. Classically, such systems were discovered in the study of differential equations. A classical example is the geodesic flow on ellipsoids. In that context, one starts with a hamiltonian (function) on M (e.g. the kinetic energy in the case of geodesic flow) and the coordinates of the lagrangian fibration n correspond to a complete set of conserved quantities. Compact smooth lagrangian fibers are tori by Liouville's Theorem. The integral curves of the Hamiltonian vector field are "straight lines" in the lagrangian fibers. When these tori are abelian varieties, one uses theta functions to obtain explicit formulas for the solution of the ordinary differential equation given by the Hamiltonian vector field. Recall, that an n-dimensional compact complex torus A is an abelian variety, if it admits an embedding as a subvariety of the complex projective space. Integration / //i(A,Z) ^ //^'^(A)* (1) embeds the first homology as a full lattice in the complex vector space dual to the space of global holomorphic one-forms. As a complex torus A is naturally the quotient
SEIBERG-WITTEN THEORY 25 //^'^(A)*///i(A, Z). An embedding of A into projective space is determined by a positive line-bundle whose first Chem class is an integral Kahler (1, l)-form co (the polarization). We can choose a symplectic basis yi,..., y^; y^+i,..., y2n for //i(A, Z) so that span[y\,...,]/«} and span{yn-\-i, -. - ^yin) are isotropic with respect to a> and co(yi, yn-\-j) = ^i • ^ij' The non-zero integers 5i,..., 5„ are called the divisors of the polarization and the polarization in principal if they are all equal to 1. As elements in //^'^(A)* we can express the last n elements y„+i,..., y2n in terms of yi,..., y„ n Yn+i = E^O?- (2) The two conditions (i) a> is a (1, l)-form and (ii) co is Kahler, translate to Riemann's two bilinear conditions: (i) then x n matrix r is symmetric and (ii) its imaginary part Im(z) is positive definite (Griffiths and Harris [1978]). Consequently, any polarized abelian variety (A, co) is obtained as the quotient of C" by the lattice generated by the columns of an n x 2n matrix (A5; r) where A5 is a diagonal matrix with integral entries 8i and r is a point in the Siegel upper half space Mn := {n X n symmetric matrices with a positive definite imaginary part}. Changing the choice of a symplectic basis yi,..., y^; y^+i,.. •, y2n introduces an action of an integral subgroup Fs of Sp(2n, R) on H^. If the polarization is principal, the modular group r is Sp(2n, Z). Two points in M^ correspond to the same polarized abelian variety if and only if they belong to the same orbit of Vs. 2.3 Integral Special Kahler Geometry Let G be a complex reductive group of rank r, T a maximal torus, g, t, their Lie algebras, and W the Weyl group. The moduli space ofN = 2 supersymmetric Yang-Mills theory in 4 dimensions is the vector space B := i/W.liis equipped with an integral special Kahler structure. Locally, this special Kahler structure consists of the following data: 1. Electric charges ai,... , a^ thought off as holomorphic coordinates on B. 2. A holomorphic function J-'(ai,... ,ar) on B (the prepotential) expressing the relations between the electric and magnetic charges. The goal, which was achieved in Serberg and Witten [1994] for SL(2, C), is to compute the functions at and the prepotential J^ in terms of the natural coordinates ut on the vector space t/ W. The coordinates ut are those obtained by a choice of generators for the algebra of W-invariant polynomials on t. The magnetic charges are given by af := f (3) and are thought off as dual coordinates on B. Clearly, the electric and magnetic charges determine the prepotential up to an additive constant. We get also a symmetric matrix r(a)ij := -^ (4) dai auj
26 E. MARKMAN whose imaginary part is required to be positive definite (a point in the Siegel upper half-space Mr). It follows that ds^ := Im{xdada) = Im(da^dd), K{a,a) := -Im ( ^S^a^di j , and tj are a Kahler metric, its Kahler potential, and its Kahler form. Analytic continuation of the local holomorphic functions {at, aj^] globally on B (away from a bad codimension 1 locus) forces multi-valuedness. The electric-magnetic "duahty" is the statement that a finite index subgroup of the integral symplectic group Sp(2r, Z) is a symmetry group for the data {at, «,^}. More precisely, any other local branch [bj, bj^] is related to {at, aP] by an affine integral symplectic monodromy transformation Above, R is an integral symplectic matrix, the constants m, are the masses and nt.nf are integral vectors. Consequently, the differential one-forms transform via R (db^\ ^ j,(da^\ \db ) \da ) In particular, in the absence of masses, the electric and magnetic charges can be interchanged (up to sign) by the involution R = \ j f\Y The fact that R is symplectic implies that the Kahler form co is global (single-valued). The integrality of R implies that [dai^da^] is a local symplectic basis of a global family A := Uh^B^b consisting of a rank 2r integral lattice A^ in each holomorphic cotangent space T^ B. Consequently, we get a family A:= {T'^B)/A of abelian varieties (away from a bad locus of codimension 1 in i5). Moreover, the natural symplectic form on r*i5 descends to A because, locally, the lattice A is generated by closed 1-forms. We see that an integral special Kahler structure gives rise to an integrable system A ^^ B whose lagrangian fibers are principally polarized abelian varieties. One can formalize the above discussion and define an integral special Kahler structure on a Kahler manifold M via the above local data and subject to the above integral symplectic gluing transformations. An equivalent coordinate-free definition was found by Freed [1999]. Definition 2.2 A special Kahler structure on a Kahler manifold M is a real flat torsion-free symplectic connection v on the tangent bundle of M with respect to which, the complex structure I satisfies dyl =0. The special Kahler structure is integral if there is a lattice A* C TM, flat with respect to v» whose dual A C T'^M is a complex lagrangian submanifold.
SEIBERG-WITTEN THEORY 27 In case G is SL(2, C), the moduli space i5 is a complex line, r in (4) is a point in the upper half plane M and A ^^ B isa family of complex elliptic curves. Seiberg and Witten used physical arguments to show that, in the absence of mass, the monodromy group is the level-2 congruence subgroup r(2) of SL(2, Z) generated by (-2 ?) ^'^ (o^ -l)- This is precisely the monodromy of the family of elliptic curves Eu given by the equation / = (jc + l)(jc - l)(jc - m), m g C \ {1, -1}. 2.4 The Special Kahler Structure of an Integrable System We sketch in this section the construction of a special Kahler structure on the base i9 of an algebraically integrable system n : A ^^ B with a section s : B ^^ A. We assume that all fibers are smooth polarized abelian varieties. In particular, the local system of integral first homologies R\^nJ^ corresponds to a representation 7i\{B, bo) ^^ F^ of the fundamental group of the base in the integral modular subgroup Fs of Sp(2n, R) (see Section 2.2). We may assume also that the section s is lagrangian (otherwise, replace the symplectic structure a by the difference a — {sonYa). Since tt is a lagrangian fibration, the symplectic structure induces an isomorphism between the tangent bundle and the conormal bundle of each fiber A^. Both are trivial bundles modelled after the vector spaces //^'^(A^,)* and T^B respectively. We get an isomorphism i : (H^^^y ^ T*B (6) of vector bundles over the base B. Composing l with a relative version of the integration homomorphism (1) we get an embedding of the local system of integral first homologies Ri^j^^Z eis a, holomorphic family A C r*i5 of full lattices in the cotangent bundle of the base. It follows that the phase space A is canonically isomorphic to the quotient T*B/A. The section s of A pulls back to A C T*B. Hence, A is lagrangian with respect to the puUback or of (7 to r*i5. In particular, the zero section of r*i5 is lagrangian, which implies that a is the standard symplectic structure on r*i5. It follows that local sections of A are closed holomorphic 1-forms. We can choose a symplectic basis of local sections of A Jzi,..., Jz„; dzi ,... ,dzn corresponding to a symplectic basis of local sections of Rx^-^JL. Being closed, we can lift the local basis to a set of holomorphic functions {zi. ^f }• The equations defining the period matrix (2) translate to the system of linear equations ^.? = E'.f •
28 E. MARKMAN The symmetry of Zij implies that the 1-form dT:^y:zr-^ is closed. Hence, locally on the base, we get a holomorphic function J^ satisfying oZi Setting ai := |- and aP := zf we get the local data of a special Kahler structure. Remark 2.3 1. (Donagi and Markman [1996a]) While the local sections dT and Yl,ij dlfa^^i ^ ^^J of r*i5 and Sym^{T''B) depend on the choice of a symplectic basis, the cubic tensor obtained locally as the third derivative of ^ is a global section c of Sym^(T'^B). Under the identification of r*i5 with the Hodge bundle H^'^ via (6) and the polarization, c is mapped to the differential of the classifying map B -^ M^/ F^, which is a global section of Hom(TB, Sym^H^i). 2. The special Kahler structure we get, locally on the base, is the one coming from the family A ^^ B of principally polarized abelian varieties, where A^ is obtained from the quotient of A^ by the subgroup of order YYi=i ^i spanzi-—, • • •. -r~} I spanzidzu • • •, dzn) 01 On of torsion points. In general, there does not exist a canonical choice of such a subgroup. Thus, the global analogue A, of the lattice span[dai, da^} defining the special Kahler structure, would be constructed on a finite cover B ^^ B parametrizing such choices. 2.4.1 Seiberg- Witten Differentials The special Kahler structure on the moduli space B of supersymmetric Yang-Mills theories involves the electric and magnetic charges rather than their differentials. In other words, it includes a global lift of the lattice of closed 1-forms A C r*i5 via a homomorphism J : A ^^ Ob into the sheaf of holomorphic functions. The lift is subject to the symplectic affine monodromy transformations (5). In the case when all the masses m, vanish (and the transformations are symplectic), such lifts I are in one-to-one correspondence with equivalence classes of holomorphic 1-forms XonA satisfying a = dX. (7) In particular, the symplectic form a on >l is exact. Two such 1-forms Xi, X2 give rise to the same left / if and only if Xi — X2 is the pulback of a 1-form on B. Indeed, given such a 1-form X and a local section y of A, we consider y as a section of the lattice of first homologies Ri,Tr^Z and define the value of I(y) at Z? g i5 to be the y-period of the restriction of X to Ah. Conversely, given such a lift I, the local vector field J^t ^i j^. patches to a global vector field V on B. Let /x^ be the corresponding function on 7*5 and a the tautological (action) 1-form on r*i5. Then the 1-form X := ot — dfjCy on r*i5 is invariant under translations by local sections of A and descends to a 1-form X on A satisfying (7). (See Lemma 2 in Donagi and Markman [1996a] for more details).
SEIBERG-WITTEN THEORY 29 In the general case, when the masses m, do not vanish, a lift I: A ^- Ob satisfying (5) is provided by a meromorphic 1-form X satisfying (7) and having poles along A^ divisors with residue m, along the i-ih divisor. 2.5 Yang-Mills Theory and Algebraic Geometry The special Kahler structure on the r-dimensional moduli space ofN = 2 supersymmetric Yang-Mills theories in 4 dimensions (with one hypermultiplet of mass in the adjoint representation) is known to depend on the (complexified) gauge group G of rank r, one mass parameter m, and an ^L (2, Z) orbit of a point Xd in the upper half plane. We are thus looking for a known family of 2r dimensional integrable systems depending on these parameters. A natural first guess would be to consider integrable systems arising from classical Yang-Mills theory. We describe in this section a family of integrable systems, the generalizedHitchin systems, depending canonically on gauge theoretic data ('E,G,C,DARp]peD) where S is a Riemann surface, G a complex reductive group, c the topological type of a principal G-bundle (the first Chem class in the case of GL{n)), D an effective divisor on S, and Rp is a marking of each point /? g D by a non-zero coadjoint orbit Rp in g* (or, possibly, in the dual of the positive half of the loop algebra) (see Hitchin [1987a], Bottacin [1995] and Markman [1994]). The phase space of this integrable system is the moduli space M := M(i;, G, c, D, [Rp}p^D) of Higgs pairs defined below. If G is 5 L(r +1) and we choose S to be the elliptic curve determined by r^/, then the dimension of M (if non empty) is equal to the dimension of YipeD ^p- ^^^ ^^^y choices (D, YipeD Rp^ resulting in a 2r dimensional system are: (i) D empty, in which case we get the Hitchin system, (ii) D consists of the zero point po of the elliptic curve and Rp^ is the minimal non-zero nilpotent orbit, in which case we get another birational model of the Hitchin system, or (iii) D consists of the zero point po and Rp^ = mR, where R is the minimal non-zero semi-simple coadjoint orbit of /-I 0 ... 0\ 0 -1 ' (8) -1 0 \ 0 ... 0 rj Donagi and Witten performed a sequence of (quite involved) tests verifying that the geometry of this family of integrable systems agrees with the physical predictions [1996]. We will come back to this system in Section 2.7. The dimension of non-zero coadjoint orbits of other simple groups is larger than 2r. Hence, non-trivial choices of data (D, YipeD ^/?) ^^^^ ^^ integrable systems of too large a dimension. The Calogero-Moser systems are conjectured to be the correct choice for more general groups (see Bordner et al [1999], D'Hoker and Phong [1998], and Hurtubise and Markman [1999]).
30 E. MARKMAN Definition 2.4 A (holomorphic) Higgs pair (P,(p) over S consists of a principal G bundle P and a holomorphic section cp of the twist ad(P) (8) Kj: of its adjoint Lie algebra bundle by the canonical line bundle. The pair (P,(p) is a meromorphic Higgs pair if instead, (p is a meromorphic section of ad(P) (g) Kj:. The Higgs pair is stable (semi-stable) if every ad (P)-invariant subbundle p C ad(P) of parabolic subalgebras has negative (resp. non-positive) degree. When G is GL{n), then P is the frame bundle of a vector bundle E and cp G EndC^") (g) K^, is a 1-form with values in the endomorphism bundle of E. The moduli space M(i;, G, c) parametrizes isomorphism classes of (stable) holomorphic Higgs pairs of topological type c (where c is a class in 7ri(G)). If the principal bundle P is itself stable, then standard deformation theory identifies the space H^{Ti, ad{P)) with the tangent space to the moduli space of principal bundles. Using Serre's duality we get that the space //^(S, ad(P) (8) ^e ), of Higgs fields with a fixed bundle P, is the cotangent space to moduli. Hence, the cotangent bundle of the moduli space of principal bundles is an open set of M(i;, G, c). M(i;, G, c) is the phase space of the Hitchin system. Remark 2.5 The stability condition is significant for two reasons: It enables one to construct the moduli space as an algebraic (quasi-projective) variety (see Faltings [1993], Hitchin [1987a], Nitsure [1991] and Simpson [1994, 1995]). There is also a real-analytic isomorphism between M(i;, G, c) and the moduli space of irreducible representations of the fundamental group of S (or of a central extension of 7ri(i;)). In that context, stability of the Higgs pair corresponds to irreducibility of the representation. There is also a relationship between meromorphic Higgs pairs (possibly with additional parabolic data) and representations of the fundamental group of the complement of the polar divisor (Hitchin [1987a] and Simpson [1990]). Denote by M(i;, G, c, D) the moduli space of (stable) meromorphic Higgs pairs whose polar divisor is contained in a fixed effective divisor D. We will denote it also by M(D) for short. M(D) is smooth if the group is GL(n) or SL(n), but it may have quotient singularities in general (in which case the moduli stack is smooth) (see Faltings [1993]). Some insight into the geometry of M(D) is gained by its realization "as" the quotient of a symplectic variety by a finite dimensional Hamiltonian group action. We will obtain a rough description of a Zariski open subset of M(D) as a bundle of Hitchin systems over the dual g^ of a Lie algebra (or a reduction of g^ if S is rational or elliptic). In particular, when the Hitchin system (of holomorphic Higgs pairs) is a single point, the moduli space of meromorphic Higgs pairs admits a group theoretic description. This happens when S is rational (Example 2.6) or elliptic (Example 2.8). Example 2.6 (Alder and van Moerbeke [1980], Adams et al. [1990], Beauville [1990] and Reiman and Semenov-Tian-Shansky [1994]) Let us describe this open subset of M(D) when S is rational, c is the trivial topological type and D is a collection of distinct points D = pi-\-p2-\ \-pd^ d > 3. An open subset of M(D) consists of Higgs pairs with a trivial G-bundle P. Hence, it consists of conjugacy classes of sections cp of //^(P^, g* (8)c K(D)) by the group G (considered as the group of global automorphisms of P). By the residue
SEIBERG-WITTEN THEORY 31 theorem, H^(F^, g* (8)c K(D)) embeds as the kernel of (01,...0j) h> ^0,- 1 = 1 where /x is the moment map for the diagonal conjugation action. Hence, an open subset of M(D) is the reduction of the dual §^ of the Lie algebra of the group Gd of maps from D to G. In particular, M(D) has a Poisson structure. When D contains a point with multiplicity, the above statements holds if Gd is the group of jets of maps. 2.5.7 The Poisson Structure The moduli space of holomorphic Higgs pairs is a partial compactification of the cotangent bundle of the moduli space of principal bundles. Hence, it has an algebraic symplectic structure (even a hyperkahler structure). The moduli space M(D) of meromorphic Higgs bundles has a natural Poisson structure. A canonical global formula for this Poisson structure is obtained via cohomological techniques once the tangent and cotangent spaces to moduli at a Higgs pair (P, ^) are identified with the first hyper-cohomology of the complexes r(p,^)M(D) = H\ad{P) ^ ad{P) 0 K^{D)] T^^P^^^M(D) = H\ad{P){-D) ^ ad{P) 0 K^] (9) The contraction with the Poisson structure T^p .M(D) -^ T(p^(p)M(D) is induced by the natural homomorphism of complexes (Bottacin [1995], Markman [1994] and Faltings [1993]). A geometric description of the Poisson structure on a Zariski open subset of M(D) was given in Markman [1994]. Recall that when a group G is acting symplectically on a symplectic variety X with a moment map /x : X ^- g*, then a "nice" quotient X/G has a natural Poisson structure. Given a coadjoint orbit R C Q*, the symplectic reduction fji~^(R)//G (when connected) is a symplectic leaf of X/G. For example, the standard Poisson structure on the dual of a Lie algebra g* can be recovered via the above construction if we consider g* as the quotient of the cotangent bundle r*G via the lifted action of G. Let V{Ti,G,c, D) be the moduli space of framed principal bundles. We denote it by V{D) for short. A point in V{D) is a pair (P, vj) consisting off a principal bundle P and a lift T] \ D ^^ P of the divisor D to an embedding in the total space of P. When G is GL(az), P is the frame bundle of a vector bundle £", and D consists of distinct points, then r] corresponds to a choice of a frame in each fiber of E over D. Standard deformation theory identifies the tangent bundle of V{D) at (P, vi) with //^(S, ad(P)(-D)). Serre's duality identifies the cotangent space with H^(i:,ad(P) (g) K(D)). Hence, the cotangent bundle T^'ViD) paramQinzQs framed Higgs pairs. A point in T*V(D) is a triple (P,(p,r]) where (P, ^) is a Higgs pair and ?; is a framing of P along D.
32 E. MARKMAN Note that the data of a framing t] is equivalent to a trivialization of the restriction of the bundle P to D. The group Go of maps from D to G acts naturally on V(D) by changing the framing data rj. Hence, it act symplectically on its cotangent bundle. The quotient space "is" the moduli space of Higgs pairs (over open sets where the various stability notions agree). The action of Go factors through its quotient Go •= Gd/Z(G) by the diagonal embedding of the center of G (because the center Z(G) embeds in the automorphism group of P). The moment map (P^cp,ri)^((pi^r (10) conjugates the restriction (p\^ via the trivialization rj into q (S>c (K(D))\^. The latter space is naturally identified with g^ via the trace-residue pairing. We get that a coadjoint orbit R = YipeD ^p ^^ 5d determines a symplectic leaf M(R) in the moduli space M(D) of Higgs pairs. Coadjoint orbits of g^ embed via the inclusion g^ ^-> g^ as those coadjoint orbits of 0^ which central summand satisfies the constraint imposed by the global residue theorem. Example 2.7 In the setup of Example 2.6, the quotient Gd/G is the open subset in the moduli space of (stable) framed bundles V(F^, G, c, D) consisting of isomorphism classes of pairs (P, rj) with P trivial. The action of Go on Gd/G lifts to the cotangent bundle T'^iGo/G). The moment map is Gd -equivariant. In particular, it is equivariant with respect to the stabilizing group G/Z(G). As the action of Go on Gd/G is transitive, we get that the quotient map T\Gd/G) -^ M(D) factors through the moment map as the composition T*{GdIG) ^ qI ^ Ql/G. This is precisely the birational embedding of M{D) in 02)/G described in Example 2.6. Example 2.8 (The elliptic Gaudin system (Reiman and Semenov-Tian-Shansky [1994]) and (Markman [1994]) Section 9.2). A second example of a rigid bundle is obtained when the curve S is elliptic, the group G is GL{n) but we fix the determinant line bundle of the vector bundle det(£') = L and require its first Chem class c to be relatively prime to the rank n. Atiyah proved that there exists a unique such stable vector bundle E (Atiyah [1957]). The moduli space V(L, D) of framed bundles contains, as a Zariski open subset, a copy of SL(n)D' The moduli space M(L, D) of traceless Higgs pairs (F, cp) with fixed determinant line bundle L contains the vector space //^(S, End(£')o <S>K(D))2is the Zariski open subset in which F is stable (and hence isomorphic to E). An argument analogous to the one in Example 7 provides an isomorphism between this open subset of M(L, D) and the quotient (slnTo ^f T'^[SL(n)D] by SL(n)D' This isomorphism is compatible with the poisson structures because the quotient map is the moment map. In particular, we get that coadjoint orbits of the Lie algebra (5/^)^ are completely integrable systems!
SEIBERG-WITTEN THEORY 33 2.6 Complete Integrability Hitchin made the remarkable observation that the moduli space of holomorphic Higgs pairs is a completely integrable system [1987]. The lagrangian fibration is given by the characteristic polynomial map (also known as the Hitchin map). The analogous construction induces a lagrangian fibration on the symplectic leaves M(R) of the moduli space of meromorphic Higgs pairs. Let ai,... a^ be generators of the algebra of invariant polynomials on g, dt the degree of at and set Char(D) := e^^i/Z^S,/^(D)®^0- Since a, is invariant, it can be evaluated at the section cp of ad(P) (g) K(D) to give a global section of //^(S, K{D)®^'). We get the characteristic polynomial map char : M{D) -^ Char(D). (11) It is easy to check that the dimension of the generic fiber of char is equal to half the dimension of the maximal dimensional symplectic leaves M(R) corresponding to regular (of maximal dimension) coadjoint orbits R of g^ (See Markman [1994], Proposition 7.17). Moreover, the algebra of polynomials on Char(D) pulls back to a Poisson involutive algebra on M(D). The generic fiber of char is a smooth compact abelian variety contained in a unique symplectic leaf M(R) of maximal dimension (here we need to assume that the derived subgroup of G is simply connected to assure that the fiber is connected). This abelian variety is the Jacobian of a branched (spectral) cover of S if the group is GL(n) (see Beauville et al [1989]). In that case, we choose the invariant polynomials at on g/„ to be the coefficients of the characteristic polynomial. Given a section Z? := (Z?i,..., Z?„) of Char(D), we get a morphism of complex surfaces pb : K{D) -^ K(D)®^ which sends a section y of K(D) to the section y"" + biy""-^ + • • • + Z?« of KiD)®"". The spectral curve Cb is the inverse image in K(D) of the zero section of K{D)®^. If b is the characteristic polynomial of a Higgs pair (E,(p : E ^^ E (S) K(D)), then the spectral curve parametrizes eigenvalues of cp. The puUback of E to Cb contains a canonical eigenline-sub-bundle. This defines a map from the fiber of char over b to the group PIcq of line bundles on Cb. The inverse PIcq -^ char~^ (b) of the above map is obtained as follows (up to a shift by the ramification line-bundle): The pushforward tt^L of a line-bundle L on Cb to S is a rank n vector bundle £" on S. The fiber of E over a point jc G S is the direct sum of the fibers of L at the points of Cb over x (if a point ic is a ramification point of tt : C^? ^- S, then the space of jets of sections of L at ic of appropriate height is taken as a direct summand). L determines also a Higgs field (p \ E ^^ E^ K(D). Observe first that the pullback 7t''K{D) of the line bundle K{D) to its total space has a canonical section y. Away from the fibers over D, the surface is the cotangent bundle of S and y is the action 1-form on the surface. It extends to the whole surface as a meromorphic 1-form with poles along 7r~^(D). We get a homomorphism <^yb \ L ^> L ^ n'^K(D) where yb is the restriction of y to Cb- The Higgs field cp is the pushforward of (Siyb- The meromorphic 1-form y provides the Seiberg-Witten differential needed to derive a special Kahler structure from the Hitchin systems (see Section 2.4.1).
34 E. MARKMAN For a general reductive group G, the generic fiber of char is described in terms of a curve in the twisted Cartan bundle t (8)c K(D). One observes that the vector bundle ^^■^■^KiD)'^^' is the quotient of the total space of t (8)c K(D) by the natural action of the Weyl group A point b in Char(D) is a section of the quotient bundle. The inverse image Cb := q~^(b) is a curve embedded in t (S>c K{D). We call Cb the cameral cover. It is a Galois W-cover of S. Denote by A the weight lattice Hom(T, C^). The data of a principal T-bundle C on Cb is equivalent to the data given by the homomorphism sending a weight of T to the associated principal C^ bundle, i.e., to a line-bundle on Cb. Hence, the moduli space of principal 7-bundles on Cb is Horn (A, Pic^ ), whose connected component is the cartesian product of r copies of the Jacobian /^ . The Weyl group acts on A and J^ and the generalized Prym is the connected component of the subgroup of W-equi variant homomorphisms Prynib := Homw(A, J'^)^. The fiber of char over a characteristic polynomial b with a smooth cameral cover Cb is isogenous to Prynib (and is isomorphic to Prynib if the derived subgroup [G, G] is simply connected) (see Donagi [1995], Faltings [1993] and Scognamillo [1998]). Example 2.9 Let us compare the two descriptions of the fiber of the Hitchin map when the group G is GL(az). The Weyl group W is the symmetric group. Cb is an n! sheeted branched cover of S, while the n sheeted spectral cover Cb is the quotient of Cb by the stabilizer Synn-i of the highest weight of the standard representation of GL(n), If ^i,..., ^„ is the standard basis of A* and W^. is the stabilizer of ei, then , w Prymb:=(Jc,^zA^r={ H ^cJ = (^f'><'••>< ^f") We ~ The invariant subvariety /~ ' is isomorphic to the Jacobian of the quotient Cb/We- which is Cb ' Cb. Thus, the right hand side is the diagonal embedding of the Jacobian /c^ of the spectral cover in the r-th cartesian product of the Jacobian of the cameral cover. Hence, Prynib is isomorphic to /q . It is also easy to describe the map from the fiber of char in M(D), over a generic characteristic polynomial b, to a coset of Prynib in the moduli space of principal T-bundles on Cb. Fix a Borel subgroup B C G and a maximal torus T C B. One observes that the cameral cover Cb parametrizes pairs (x, fi) of a point jc in S and a Borel subalgebra fi in ad(P)x containing the centralizer of cpx. Over Cb, the puUback of P admits a canonical reduction to a principal B bundle. Using the natural homomorphism B ^' T v/q associate to the i5-bundle a principal T-bundle on C^?. The main effort in Donagi [1995] and Scognamillo [1998] is devoted to the proof that this map is invertible (under suitable conditions on G and the cameral cover).
SEIBERG-WITTEN THEORY 35 2.6.1 Smooth Compact Hitchin Fibers in a Symplectic LeafM(R) Consider now the fibers of the restriction of the Hitchin map (11) to a symplectic leaf charR : M(R) —> Char(R) C Char(D) (12) for a general coadjoint orbit /? in g^. Understanding the geometry of these fibers could be quite complicated. The difficulty arises from singularities which a general R could impose upon the spectral (and cameral) curves (see Hurtubise and Markman [1998], Section 6.4). When the cameral cover Cb is singular over points in the divisor D C S, the fiber of (11) over b may intersect several symplectic leaves M(R) in M(D). Two simplifying technical assumptions are: (i) R is regular, or (ii) R is closed in g^. The regularity assumption fits well with the results of (Donagi [1995]) and helps to describe the generic fiber of (12) as a generalized Prym of the cameral cover, even if the latter is singular over points in D. Condition (ii) holds, for example, if each Rp is a semi-simple coadjoint orbit of 0*. The closedness of R helps to prove that the generic fiber of chavR is an abelian variety (Corollary 2.11). A further relaxation of the closedness condition leads to the same conclusion (Corollary 2.13). Proposition 2.10 Let (P, cp) be a Higgspair with poles along D and polar tail in R. If the characteristic polynomial b := char(P, cp) determines a reduced and irreducible spectral (or cameral) cover, then M(R) is non-empty. If, in addition, R is a closed orbit in g^, then the generic fiber of (12) is a (smooth compact) lagrangian abelian variety. Sketch of proof: M(R) is non-empty because the pair (P, cp) is necessarily stable. The moduli space M(D) of stable Higgs pairs is contained in its partial compactification M(Dy^\ the moduli space of semi-stable Higgs pairs. The morphism char extends to a proper morphism from M(Dy^ to Char(D). The condition on the spectral curve implies that every Higgs pair with characteristic polynomial b is stable and its automorphism group is minimal (equal to the center Z(G)). Hence, the compact fiber in M(Dy^ is contained in the smooth locus of M(D). Next we show that if R is closed, M(R) is a closed subset of M(D). We conclude that the fiber of (12) over b is compact. Consider the moduli space M(D) of triples (P, ^, rj) consisting of a stable Higgs pair (P,(p) and a framing rj along D. M(D) is similar to the cotangent bundle in (10) except that we require the Higgs pair (rather than the framed bundle) to be stable. The morphism /x given by (10) extends to M(D). The morphism 7T : M(D) -^ M(D) is surjective and an open map in the complex topology (it is even a principal Go bundle over the locus of Higgs pairs with automorphism group Z(G)). Thus, M(R) is closed if and only if 7T~^(M(R)) is closed. The latter is closed because we have the equality 7T-\M(R)) = jn'^R). Since M(R) is a finite union of symplectic leaves, its intersection with the smooth locus of M(D) is smooth. Hence, the generic fiber of (12) is smooth. It remains to prove that the fiber of (12) over b is lagrangian. It would follow that a compact smooth fiber is an abelian variety by Liouville's Theorem. We already know that the polynomial algebra on Char(D) pulls back to a Poisson involutive algebra on M(D) (and in particular on M(R)). Hence, the fiber is involutive. Showing that the fiber is isotropic amounts to a formal cohomological calculation
36 E. MARKMAN using the identification (9) of the Poisson structure. One proceeds as in the proof of Theorem II.5 in Faltings [1993]. It is easier to carry out the cohomological calculation on the level of stacks. This enables us to drop the stability assumption and to use functorial operations, such as pull-back to a finite cover of S, reduction to a Borel bundle, induction to a torus bundle, pushforward via a faithful representation of G, and decomposition into rank 1-bundles. These operations reduces the calculation to the case of Higgs line bundles in which the Higgs field ^ is a meromorphic 1-formon S andM(/?) isthecoset [^ + //^(i;, K)] xPic^ of the cotangent bundle of Pic^ in //^(S, K(D)) x Pic^. □ As a corollary of the Proposition we get: Corollary 2.11 If the genus of S is > 2, then M(R) is non-empty for every orbit R of 0^. If /? is closed then the generic fiber of (12) is a (smooth compact) lagrangian abelian variety. Note: A partial analogue for the genus 1 case is provided in Remark 2.15. Whether M(R) is empty or not is a delicate question in the genus 0 case (see Simpson [1991]). Proof of Corollary 2.11 Let P be a stable principal bundle on S. Then H^(ad(P)) and H^(ad(P) (8) K) are the center 3 of g and its dual 3*. As the evaluation homomorphism from H^(ad(P) 0 K(D)) to ker[ad(P) (g) K(D)i^ -^ H\ad(P) (g) K)] is surjective, the moment map homomorphism from H^(ad(P) (8) K(D)) to g^ is also surjective. Choose cp in H^(ad(P) (g) K(D)) mapping to R. The whole coset cp + H^(ad(P) (g) K) is contained in M(R). Moreover, if P is generic (more precisely, very stable) the characteristic polynomial map from the vector space of holomorphic Higgs fields H^(ad(P)(S}K) -^ ^'i^^H^iK®"^^) is surjective (see Kouvidakis and Pantev [1995] Lemma 1.4; their proof for GL(n) works for a general reductive G as the cited theorem of Laumon holds in general). Since the generic cameral cover in t (8) ^ of the Hitchin system is smooth and irreducible (Faltings [1993]), the generic cameral cover in t (g) K(D) of Higgs fields in the coset cp + H^(ad(P) (g) K) is reduced and irreducible. The Corollary follows from Proposition 2.10. D Remark 2.12 Note that if R is closed but not regular, then the spectral and cameral covers of every Higgs pair in M(R) are necessarily singular. In that case, when the group is GL(az), the generic fiber of (12) in Proposition 2.10 is the Jacobian of the normalization of the singular spectral curve. Proposition 2.10 can be extended to more general coadjoint orbits R of g^ which are not closed but which admit a "lift" to a closed orbit once we consider higher jets. In this case, the symplectic leaf M{R) is not closed but the morphism (12) factors through a coarser foliation of M(R) whose generic leaf is closed in M(D). The following simple example illustrates this situation. Let D = Ip consist of a single point p and R be the coadjoint orbit
SEIBERG-WITTEN THEORY 37 in s/(2)* of the regular nilpotent orbit of the zero order germ (p\^ of the Higgs field \cz -az ) z Clearly R is not a closed orbit. If we consider however iht first order germ (p\^^ and c 7^ 0, then we get a closed orbit of SL(2)2p. On the other hand, the orbit R^ in formula (6.9) in Hurtubise and Markman [1998], Section 6.4 does not admit a lift to a closed orbit. Moreover, none of the fibers of M{R') -^ Char{R') is compact. Corollary 2.13 \fG\sGL{n)oxSL{n), the genus of S is > 2, and D consists of distinct points, then the generic fiber of (12) is a lagrangian abelian variety for every orbit R of g^ whose generic lift ^ is a closed orbit of G2d- This abelian variety is the Jacobian or prym of a smooth compact irreducible curve. Sketch of Proof (GL(n) case) We need to prove that the germ (pi^^ of the generic Higgs field (p : E ^^ E (S) K(D) in M(R) belongs to a closed orbit R of G2d- The argument used in the proof of Proposition 2.10 would then imply that the generic fiber of (12) is an abelian variety. The orbit R is closed if and only if each factor Rp, p e D, is closed. Thus, it suffices to prove that for a generic stable vector bundle £", the evaluation homomorphism H^(End(E) (S> K) ^^ [End(£') (g) K]\^ is surjective. For a stable £", End(£') is semi-stable and the evaluation homomorphism fails to surject if and only if End(£') has a line subbundle of the form O^iiq — p), q 7^ P- The corresponding section / of End(£')(/7) must be an isomorphism E ^^ E(p — q). Hence, Oj^iq — /?) is a line-bundle of order dividing n and £" is a fixed point of the automorphism £" i-> £" (g) O^ (q — p). If the genus of S is > 2, the generic stable vector bundle is not fixed by any of the n^^ — I such automorphisms. It is known, that if the spectral curve Cb is reduced and irreducible, then the fiber char~^(b) in M(D) is the compactification of its Jacobian via the moduli space of rank 1 torsion free sheaves of a fixed Euler characteristic (Simpson [1994, 1995]). The Jacobian J^ acts on this compactification. The fiber of (12) over bis a, union of orbits of J^ . Every such orbit is a bundle over the Jacobian of the normalization of Cb with rational fibers. Being an abelian variety, the generic fiber can not contain any orbit other than the Jacobian of the normalization of C^,. D Coadjoint orbits as germs of maps: The results of Section 2.6.1 admit the following algebro-geometric interpretation when the group is GL(n). Simpson introduced a dictionary under which (1) meromorphic Higgs pairs with poles along D correspond to (2) sheaves on the total space S of K(D) (Simpson [1994, 1995]). The infinitesimal version of this correspondence associates to a coadjoint orbit R in 0/(n)^ an O^-module Fr of length n ' deg(D). The generic fiber of (12) is the Jacobian of a smooth curve, if and only if this module Fr is the germ of a morphism y : C ^- C C 5 of degree 1 from a smooth irreducible curve C onto a curve C in the linear system of n sheeted spectral covers in S. By a germ of a morphism we mean that Fr is isomorphic to v^[Oc/Oc(—7T~^D)].
38 E. MARKMAN There are infinitesimal as well as global obstructions for an O^-module Fr to be the germ of such a map. The results of Section 2.6.1 give criteria for Fr to be such a germ. We also prove that if Fr is the germ of such a map, then the generic such C is smooth away from the fibers over D, Moreover, the singularity over D is "as mild as it could be" in the following sense: the difference between the arithmetic and geometric genus of C is equal to half the difference between the maximal dimension of coadjoint orbits of Ql(nYj^ and the dimension of R. 2.7 KP Elliptic Solitons Donagi and Witten conjectured that the generalized Hitchin system M{11, SL(n), po.mR) provides the special Kahler structure on the moduli space of N = 2 supersymmetric Yang-Mills with adjoint matter when S is an elliptic curve, po its origin, and the coadjoint orbit R in s/* is given by (8). We describe in this section the geometry of this system. A Higgs pair (E, cp) in M(R) consists of a rank n vector bundle E with a trivial determinant line bundle and a homomorphism cp \ E ^^ E ^ Kj: (po)- The generic (semi-stable) vector bundle E is represented by the direct sum ofn line bundles Oj: (pi—po) 0 • • • 0 Os (Pn—po) with ^ Pi in the linear system \n - po\. We get a rational morphism from M(R) to P"~^ The differential of the i-ih invariant polynomial a, on s/* vanishes along R to order / — 2. Thus, the characteristic polynomials (Z?2, - • • ,bn) of Higgs pairs in M{R) are sections of //^(S, 0^^2^s iPo)®^) in a fixed coset Br of the n — 1 dimensional subspace //^(S, 0^^2^?')- If '^ > 2, the generic spectral curve Cb in Ky:(po) is singular. It is convenient to identify the fiber of Kj: (po) over po with C via the residue map. Cb has one smooth point po passing through the point m(n — l) over po and n — 1 branches through the point —m meeting pairwise transversally. Separating these n — l branches of a generic Cb we get a smooth curve Cb of genus n (see Remark 2.15 for a proof). The spectral curves described above are the celebrated tangential covers (Krichever [1980], Treibich and Verdier [1989, 1990]) (see also Donagi and Markman [1996b] Section 6.3.5). They have the property that the Abel-Jacobi image of Cb in /q (sending a point p to p — po)is tangent at po to the image of /^ in /q . In fact, they are characterized by this property. If the Abel-Jacobi image of a curve C is tangent at po to an elliptic curve S, then C admits a morphism to S. Moreover, the morphism factors through a curve C in the total space of Ky: (po) intersecting the fiber over po transversally at po and at one additional point with multiplicity n — l (where n is the degree of the morphism). Krichever constructed solutions of the KP hierarchy from algebro-geometric data. Krichever's construction identifies M(R) as a moduli space of finite dimensional solutions in which the orbit of the first KP equation is S. Again we have a group theoretic description of a Zariski open subset of M(R): Lemma 2.14 The moduli space M(R) is naturally birational to the coadjoint orbit R. The orbit R is isomorphic to the unique non-trivial affine bundle over P'^"! with a principal r*P"-i structure. Proof There is a one-to-one correspondence between line bundles on the (normalized)
SEIBERG-WITTEN THEORY 39 spectral curves Cb and Higgs fields in M(i;, GL(n), c, R), with arbitrary first Chem class c (see Remark 2.12). Under this correspondence L <—> (£", cp), the vector bundle E is the push-forward of L. In particular, the Euler characteristic h^(L) — h^(L) of the line bundle is equal to that of the vector bundle E. Riemann-Roch on S and Cb implies that a line bundle of degree n — \-\- c corresponds to a vector bundle of degree c. The condition that E has trivial determinant implies that L varies in an abelian subvariety Prym^~^ of J^~^ of codimension 1. M(R) is naturally the relative prym of degree r of the family of tangential covers (the union Prym^~^ ''=^beBR Prym^~^ of their Pry ms). The point ^o with residue m(Az—l) over po is a marked smooth point on each spectral curve Cb- Hence, we have a natural birational isomorphism between Prym^~^ and Prym^~^ (isomorphism along the locus with reduced and irreducible spectral curves). The latter is the moduli space M{Oy. (—po), R) of Higgs pairs with a fixed determinant line-bundle O^ (—po)- This is precisely a symplectic leaf of the elliptic Gaudin system of Example 2.8 and we saw that it contains the coadjoint orbit /? as a Zariski open subset. There is a unique non-trivial extension 0^-r*P"~^^-y^- Opn-i ^- 0 of the trivial line bundle by r*P"-i. If P^-^ = P(//), then a concrete model of V is obtained by choosing the fiber of V over a line € in // to be the subspace of End(H) of traceless endomorphisms leaving £ invariant and whose image in End(H/£) is a multiple of the identity. The section — 1 of Opn-i determines an affine subbundle of V which maps isomorphically onto the orbit RinEnd(H). D Let us describe one of the tests performed in Donagi and Witten [1996] to check the compatibility of the geometry of this integrable system charR : M(R) -^ Br with the physical predictions. Let A C i^^? be the discriminant divisor parametrizing singular tangential covers. The most singular locus of A parametrizes the most singular covers, those with geometric genus 1. It is a zero dimensional locus and t'Hooft predicted that there is a one-to-one correspondence between points in this locus and index n subgroups of Z/n X Z/n. There is indeed a one-to-one correspondence between degrees genus 1 covers y : F ^- S and index n subgroups of the group of line bundles of order n on S. We send such a cover to the kernel of y* : Pic^ -^ PiCp. It remains to show that any such y admits a factorization y : r ^- f C K(po) (unique up to deck transformations of y) as an n-sheeted spectral cover f intersecting the fiber over po with residues (—1,...,—1,az — 1). Pulling back the tautological meromorphic 1-form y on K(po), we would get a 1-form (p := v'^y onF with simple poles along n points y~^(po) with residue n — 1 along one point po and residue — 1 along the others—1 points. We may choose po arbitrarily, as any two choices are related by a deck transformation of the covering. Conversely, any such 1-form (p would give rise to a factorization v : F ^^ K(po) whose image is the spectral curve of the Higgs field obtained by pushing forward (8)0 : L ^^ L(S}K(y~^ po) where L is any line bundle on F (see Section 2.6). The existence of a unique such 1-form 0 follows from the residue theorem. Remark 2.15 Note that the above argument provides a construction of a reduced and irreducible n-sheeted spectral cover with Higgs fields in M{R') for every non-zero orbit
40 E. MARKMAN R' = YipeD ^p of 5/(az)^ where D consists of distinct points and each R^ is semi-simple. In particular, Proposition 2.10 applies and the generic fiber of (12) is an abelian variety which is the prym of the normalization of a spectral curve. One needs to place the residues of (p at points in y~^(D) in an arrangement, which is not invariant with respect to any non-trivial subgroup of Gal(T/Ti). This is always possible, though it may impose some restrictions on the choice of GaliT/H) (a cyclic group will do). Such an arrangement assures that y : r ^- f is of degree 1. Acknowledgements I would like to thank the organizers for the invitation to participate in the conference and Ron Donagi for pleasant and helpful conversations. This work was partially supported by NSF grant number DMS-9802532. References Adams, M.R., J. Hamad, J., and Hurtubise, J., Isospectral Hamiltonian flows in finite and infinite dimensions, 11. Integration of flows. Commun. Math. Phys. 134, 555-585 (1990). Adler, M., and van Moerbeke, R, Completely integrable systems, Euclidean Lie algebras, and curves. Advances in Math. 38, 267-379 (1980). Atiyah, M., Vector bundles over an elliptic curve. Proc. Lond. Math. Soc. 7, 414-452 (1957). Beauville, A., Jacobiennes des courbes spectrales et systemes hamiltoniens completement integrables. Acta Math. 164, 211-235 (1990). Beauville, A., Narasimhan, M.S., and Ramanan, S., Spectral curves and the generalized theta divisor. J. ReineAngew. Math. 398, 169-179 (1989). Bordner, A.J., Sasaki, R., and Takasaki, K., Calogero-Moser models II: Symmetries and Foldings, Progr. Theoret. Phys., 101(3), 487-518 (1999). Bottacin, R, Symplectic geometry on moduli spaces of stable pairs. Ann. Sci. Ecole Norm. Sup. 4, 28(4), 391^33 (1995). D'Hoker, E., and Phong, D.H., Calogero-Moser Lax pairs with spectral parameter for general Lie algebras, Nucl. Phys. B 530(3), 537-610, hep-th/9804124 (1998). Donagi, R., and Markman, E., Cubics, integrable systems, and Calabi-Yau threefolds, Israel Math. Conference Proceedings 9 (alg-geom eprint no. 9408004) (1996a). Donagi, R., and Markman, E., Spectral covers, algebraically completely integrable Hamiltonian systems, and moduli of bundles. Springer Lecture Notes in Math. 1620, 1-119 (1996b). Donagi, R., and Witten, E., Supersymmetric Yang-Mills theory and integrable systems, Nucl. Phys B 460, 299 (1996). HepTh, 9510101. Donagi, R., Principal bundles on elliptic fibrations. Asian J. Math. 1, 214-223 (alg-geom eprint no. 9702002) (1997). Donagi, R., Seiberg-Witten integrable systems. Algebraic geometry-Santa Cruz 1995, 3^3, Proc. Sympos. Pure Math., 62, Part 2, Amer. Math. Soc, Providence, RI (1997). (alg-geom eprint no. 9705008). Donagi, R., Spectral covers, in: Current topics in complex algebraic geometry. Math. Sci. Res. Inst. Publ. 28, 65-86 (Berkeley, CA 1992/93) (alg-geom eprint no. 9505009) (1995). Faltings, G., Stable G-bundles and Projective Connections, Jour. Alg. Geo. 2, 507-568 (1993). Freed, D., Special Kahler Manifolds, Comm. Math. Phys., 203(1), 31-52 (1999). Griffiths, P.A., and Harris, J., Principles of algebraic geometry. Pure and Applied Mathematics. Wiley-Interscience 1978.
SEIBERG-WITTEN THEORY 41 Hitchin, N.J., Stable bundles and integrable systems. Duke Math. J. 54(1), 91-114 (1987). Hitchin, N.J., The self-duality equations on a Riemann surface. Proc. Lond. Math. Soc. 55, 59-126 (1987a). Hurtubise, J., and Markman, E., Rank 2 Integrable Systems of Prym Varieties. Adv. Theor. Math. Phys. 2, 633-695 (1998). Hurtubise, J., and Markman, E., Calogero-Moser systems and Hitchin systems. Math. AG/9912161 (1999). Hurtubise, J., Integrable systems and Algebraic Surfaces, Duke Math. J. 83(1), 19-50 (1996), and Erratum:, Duke Math. J. 84(3), 815 (1996). I.M. Krichever, Elliptic solutions of the K-P equation and integrable systems of particles. Funct. Anal 14,45-54(1980). Kouvidakis, A., and Pantev, T., Automorphisms of the moduli spaces of stable bundles on a curve. Math. Ann. 302(2), 225-268 (1995). Markman, E., Spectral curves and integrable systems, Compositio Math. 93, 255-290, (1994). Nitsure, N., ModuU space of semistable pairs on a curve, Proc. London Math. Soc. 3(62), 275-300 (1991). Reiman, A.G., and Semenov-Tian-Shansky, M.A., Integrable Systems II, chap. 2, in Dynamical Systems VII, Encyclopaedia of Mathematical Sciences, vol 16., V.I. Arnold and S.P Novikov, eds.. Springer-Verlag, Berlin, 1994. Scognamillo, R., An elementary approach to the abelianization of the Hitchin system for arbitrary reductive groups. Compositio Math. 110(1), 17-37 (1998). Seiberg, N., and Witten, E., Monopoles, duaUty and chiral symmetry breaking in N = 2 supersymmetric QCD, Nucl. Phys. B 431, 484-550 (1994). Simpson, C, Harmonic bundles on non-compact curves, J. ofAMS, 3(3), (1990). Simpson, C, Higgs bundles and local systems. Inst. Hautes Etudes Sci. Publ. Math. IS, 5-95 (1992). Simpson, C, Moduli of representations of the fundamental group of a smooth projective variety. I, II, Inst. Hautes Etudes Sci. Publ. Math. 79, 47-129 (1994), and No. 80 (1994), 5-79 (1995). Simpson, C, Products of matrices. Differential geometry, global analysis, and topology (Halifax, NS, 1990), 157-185, CMS Conf. Proc, 12, Amer. Math. Soc, Providence, RI, 1991. Treibich, A., and Verdier, J.-L., Solitons elliptiques. The Grothendieck Festschrift, III, 437^80, Birkhauser Boston, 1990. Treibich, A., and Verdier, J.-L., Varietes de Kritchever des solitons elliptiques de KP. Proceedings of the Indo-French Conference on Geometry (Bombay, 1989), 187-232, 1993.
3 Seiberg-Witten Theory and Integrable Systems ERIC D'HOKERi and D.H. PHONG^ ^Department of Physics, University of California^ Los Angeles, CA 90095, USA ^Department of Mathematics, Columbia University, New York, NY 10027, USA We summarize recent results on the resolution of two intimately related problems, one physical, the other mathematical. The first deals with the resolution of the non-perturbative low energy dynamics of certain J\f = 2 supersymmetric Yang-Mills theories. We concentrate on the theories with one massive hypermultiplet in the adjoint representation of an arbitrary gauge algebra Q. The second deals with the construction of Lax pairs with spectral parameter for certain classical mechanics "Calogero-Moser" integrable systems associated with an arbitrary Lie algebra Q. We review the solution to both of these problems as well as their interrelation. 3.1 Introduction Some of the most important physical problems of contemporary theoretical physics concern the behaviour of gauge theories and string theory at strong coupling. For gauge theories, these include the problems of confinement of colour, of dynamical chiral symmetry breaking, of the strong coupling behaviour of chiral gauge theories, and of the dynamical breaking of supersymmetry. In each of these areas, major advances have been achieved over the past few years, and a useful resolution of some of these difficult problems appears to be within sight. For string theory, these include the problems of dynamical compactification of the 10-dimensional theory to string vacua with 4 dimensions and of supersymmetry breaking at low energies. Already, it has become clear that, at strong coupling, the string spectrum is radically altered and effectively derives from the unique 11-dimensional M-theory. This rapid progress was driven in large part by the Seiberg-Witten solution of J\f = 2 supersymmetric Yang-Mills theory for SU(2) gauge group (see Seiberg and Witten [1994]) and by the discovery of D-branes in string theory. Some of the key ingredients underlying these developments are: 43
44 E. D'HOKER and D.H. PHONG (1) Restriction to solving for the low energy behaviour of the non-perturbative dynamics, summarized by the low energy effective action of the theory. (2) High degrees of supersymmetry. This has the effect of imposing certain holomorphicity constraints on parts of the low energy effective action, and thus of restricting its form considerably. For gauge theories in 4-dimensions, we distinguish the following degrees of supersymmetry. • J\f = I supersymmetry supports chiral fermions and is the starting point for the Minimal Supersymmetric Standard Model, the simplest extension of the Standard Model to include supersymmetric partners. 9 J\f = 2 supersymmetry only supports non-chiral fermions and is thus less realistic as a particle physics model, but appears better "solvable". This is where the Seiberg-Witten solution was constructed. • J\f = 4 is the maximal amount of supersymmetry, and a special case of Af = 2 supersymmetry with only non-chiral fermions and vanishing renormalization group )S-function. Dynamically, the latter theory is the simplest amongst 4-dimensional gauge theories, and offers the best hopes for admitting an exact solution. (3) Electric-magnetic and Montonen-Olive duality. The free^Maxwell equations are invariant under electric-magnetic duality when E ^^ B and B -^ — £". In the presence of matter, duality will require the presence of both electric charge e and magnetic monopole charge g whose magnitude is related by Dirac quantization e • g ^ h. Thus, weak electric coupling is related to large magnetic coupling. Conversely, problems of large electric coupling (such as confinement of the colour electric charge of quarks) are mapped by duality into problems of weak magnetic charge. It was conjectured by Montonen and Olive that the A/* = 4 supersymmetric Yang-Mills theory for any gauge algebra Q is mapped under the interchange of electric and magnetic charges, i.e. under e <^ l/e into the theory with dual gauge algebra Q^. When combined with the shift-invariance of the instanton angle 0 this symmetry is augmented to the duality group SL(2, Z), or a subgroup thereof. (4) Finally, the Maldacena equivalence between Type IIB superstring theory on AdSs x S^ and 4-dimensional J\f = 4 superconformal Yang-Mills theory is conjectured to hold at strong coupling. The AdS/SCFT correspondence thus establishes a link between certain non-perturbative phenomena in string theory and in gauge theory. Of central interest to many of these exciting developments is the 4-dimensional supersymmetric Yang-Mills theory with maximal supersymmetry, J\f = 4, and with arbitrary gauge algebra Q. Here, we shall consider a generalization of this theory, in which a mass term is added for part of ihcj\f = 4 gauge multiplet, softly breaking ihej\f = 4 symmetry to A/^ = 2. As an A/^ = 2 supersymmetric theory, the theory has a ^-gauge multiplet, and a hypermultiplet in the adjoint representation of Q with mass m. This generalized theory enjoys many of the same properties as the A^ = 4 theory: it has the same field contents; it is ultra-violet finite; it has vanishing renormalization group y^-function, and it is expected to have Montonen-Olive duality symmetry. For vanishing hypermultiplet mass m = 0, the A/* = 4 theory is recovered. For m -^ oo, it is possible to choose dependences of the gauge coupling and of the gauge scalar expectation values so that the limiting theory is one of many interesting M — 2 supersymmetric Yang-Mills
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 45 theories. Amongst these possibihties for Q = SU(N) for example, are the theories with any number of hypermultiplets in the fundamental representation of SU(N), or with product gauge algebras SU(Ni) x SU(N2) x • • • x SU(Np), and hypermultiplets in fundamental and bi-fundamental representations of these product algebras. Remarkably, the Seiberg-Witten theory for A/^ = 2 supersymmetric Yang-Mills theory for arbitrary gauge algebra Q appears to be intimately related with the existence of certain classical mechanics integrable systems. This relation was first suspected on the basis of the similarity between the Seiberg-Witten curves and the spectral curves of certain integrable models (see Gorsky et aL [1995]). Then, arguments were developed that Seiberg-Witten theory naturally produces integrable structures (see Donagi and Witten [1996]). But a connection derived from first principles between Seiberg-Witten theory and integrable models still seems to be lacking. For the Af = 2 supersymmetric Yang-Mills theory with massive hypermultiplet, the relevant integrable system appears to be the elliptic Calogero-Moser system. For SU(N) gauge group, Donagi and Witten [1996] proposed that the spectral curves of the SU(N) Hitchin system should play the role of the Seiberg-Witten curves. Krichever (in unpublished work), Gorsky and Nekrasov [1996], and Martinec [1996] recognized that the SU(N) Hitchin system spectral curves are identical to those of the SU(N) elliptic Calogero-Moser integrable system. That the SU(N) elliptic Calogero-Moser curves (and associated Seiberg-Witten differential) do indeed provide the Seiberg-Witten solution for the J\f = 2 theory with one massive hypermultiplet was fully established by D'Hoker and Phong [1998], where it was shown that: (1) the resulting effective prepotential J^ (and thus the low energy effective action) reproduces correctly the logarithmic singularities predicted by perturbation theory; (2) J^ satisfies a renormalization group type equation which determines explicitly and efficiently instanton contributions to any order; (3) the prepotential in the limit of large hypermultiplet mass m (as well as large gauge scalar expectation value and small gauge coupling) correctly reproduces the prepotentials for Af = 2 super Yang Mills theory with any number of hypermultiplets in the fundamental representation of the gauge group. The fundamental problem in Seiberg-Witten theory is to determine the Seiberg-Witten curves and differentials, corresponding to an A/^ = 2 supersymmetric gauge theory with arbitrary gauge algebra Q, and a massive hypermultiplet in an arbitrary representation R of G, subject to the constraint of asymptotic freedom or conformal invariance. With the correspondence between Seiberg-Witten curves and the spectral curves of classical mechanics integrable systems (see Donagi and Witten [1996]), this problem is equivalent to determining a general integrable system, associated with the Lie algebra Q and the representation R. The Af = 2 theory for arbitrary gauge algebra Q and with one massive hypermultiplet in the adjoint representation was one such outstanding case when Q ^ SU(N). Actually, as discussed previously, upon taking suitable limits, this theory contains a very large number of models with smaller hypermultiplet representations R, and in this sense has a universal aspect. It appeared difficult to generalize directly the Donagi-Witten construction of Hitchin systems to arbitrary Q, and it was thus natural to seek this generalization directly amongst the
46 E. D'HOKER and D.H. PHONG elliptic Calogero-Moser integrable systems. It has been known now for a long time, thanks to the work of Olshanetsky and Perelomov [1976, 1981], that Calogero-Moser systems can be defined for any simple Lie algebra. Olshanetsky and Perelomov also showed that the Calogero-Moser systems for classical Lie algebras were integrable, although the existence of a spectral curve (or Lax pair with spectral parameter) as well as the case of exceptional Lie algebras remained open. Thus several immediate questions are: • Does the elliptic Calogero-Moser system for general Lie algebra Q admit a Lax pair with spectral parameter? • Does it correspond to the J\f = 2 supersymmetric gauge theory with gauge algebra Q and a hypermultiplet in the adjoint representation? • Can this correspondence be verified in the limiting cases when the mass m tends to 0 with the theory acquiring Af = 4 supersymmetry and when m ^- oo, with the hypermultiplet decoupling in part to smaller representations of ^? The purpose of this chapter is to review the solution to these questions, which were obtained in D'Hoker and Phong [1998a,b,c]. In summary, the answers can be stated succinctly as follows. • The elliptic Calogero-Moser systems defined by an arbitrary simple Lie algebra G do admit Lax pairs with spectral parameters. • The correspondence between elliptic Q Calogero-Moser systems and Af = 2 supersymmetric Q gauge theories with matter in the adjoint representation holds directly when the Lie algebra Q is simply-laced. When Q is not simply-laced, the correspondence is with new integrable models, the twisted elliptic Calogero-Moser systems introduced in D'Hoker and Phong [1998a,b]. • The new twisted elliptic Calogero-Moser systems also admit a Lax pair with spectral parameter (see D'Hoker and Phong [1998a]). • In the scaling limit m = Mq ^ -^ oo, M fixed, the twisted (respectively untwisted) elliptic Q Calogero-Moser systems tend to the Toda system for {Q^^^Y (respectively Q^^"^) for 8 = i^T (respectively 8 = ~). Here hg and h^ are the Coxeter and the dual Coxeter numbers of Q (see D'Hoker and Phong [1998b]). The remainder of this chapter is organized as follows. In section 3.2, we briefly review supersymmetric gauge theories, the set-up and basic constructions of Seiberg-Witten theory. In section 3.3, we discuss the elliptic Calogero-Moser systems introduced by Olshanetsky and Perelomov long ago, and present the new twisted elliptic Calogero-Moser systems introduced in D'Hoker and Phong [1998a,b]. In section 3.4, we show how these systems tend to the Toda systems in certain limits and discuss their integrability properties and Lax pairs with spectral parameter in section 3.5. Finally, in sections 3.6 and 3.7, we discuss the Seiberg-Witten solution for the A/^ = 2 supersymmetric Yang-Mills theories and a massive hypermultiplet in the adjoint representation of the gauge algebra for Q = SU(N) and for arbitrary Q respectively.
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 47 3.2 Seiberg-Witten Theory Supersymmetric Yang-Mills theories are ordinary field theories of scalar, spin 1/2 fermions and gauge fields, with field contents fitting into representations of the supersymmetry algebra and with certain special relations between the gauge, Yukawa and Higgs self-couplings. For each of A/* = 1, 2, 4, there is a gauge multiplet (g) in the adjoint representation of the gauge algebra Q and foYj\f= 1, 2 there are matter multiplets (m) in an arbitrary representation Rofg. 3.2.1 Supersymmetry Multiplets (1) For M = 1, we have (g) the gauge multiplet (A^, X) containing a gauge field A^ and a Majorana fermion X; (m) the chiral multiplet ((p^i/r) containing a complex scalar (p and a chiral fermion i//. (2) For A/" = 2, we have (g) the gauge multiplet (A^, X±, 0) containing a gauge field A^, a Dirac fermion X± and a complex scalar 0, which we shall often refer to as the gauge scalar. Under an A/^ = 1 supersymmetry subalgebra, the A^ = 2 gauge multiplet is the direct sum of the A/* = 1 gauge multiplet and an A/* = 1 chiral multiplet in the adjoint representation of ^. (m) the hypermultiplet (V^±, H±) contains a Dirac fermion \l/± and two complex scalars H±. Under an A/* = 1 subalgebra, this multiplet is the sum of one left and one right M = \ chiral multiplets. (3) For A/^ = 4, we have (g) the gauge multiplet (A^, Xa, (l>i), containing a gauge field A^, four Majorana spinors X^, a = 1, • • • , 4 and six real scalars 0/, / = 1, • • • , 6. Under an A/^ = 2 subalgebra, the multiplet is the sum of an A/^ = 2 gauge multiplet and an A/^ = 2 hypermultiplet in the adjoint representation of the gauge algebra Q. (m) there is no matter multiplet for J\f = 4. 3.2.2 Supersymmetric Lagrangians For the study of Seiberg-Witten theory, we shall need both the A/^ = 2 supersymmetric microscopic (renormalizable) Lagrangian as well as A/* = 2 supersymmetric effective Lagrangians. Both types may be viewed as general Lagrangians involving the multiplets given above, but with the restriction that only terms are retained with at most two derivatives on any term involving boson fields, and one derivative on any term involving fermion fields. This is the usual approximation made when dealing with effective low energy theories, and also happens to be one of the criteria for renormalizability. These effective Lagrangians are always polynomial in the gauge and fermion fields, but depend upon the various scalar fields through possibly general functions. Supersymmetry imposes certain holomorphicity conditions on some of these functions, a property fundamental in the Seiberg-Witten analysis. Henceforth, we restrict to considering only such Lagrangians.
48 E. D'HOKER and D.H. PHONG For M = I supersymmetric theories with gauge multiplet (A^, X^), a = 1, • • • , dim Q and chiral multiplets {(p\'\lf^),i = 1, • • • , A^/, the key parts of the most general Lagrangian are given by the kinetic terms type and potential terms for the fields (all other terms such as Yukawa couplings are omitted, as we shall not need their form) £ = - ,,.;[D,^'dV + ifh^D^r] - 1 ,7^^ ^ o(p d(pJ /i\ - \Tabi<p)[{^F^^.F'^^' - ^F^^.F^"' + X"a^D^x'] + c.c. + ■■■ Here g-j = 3idjK((p, cp) is the Kahler metric oti the scalar fields, D^ are suitable covariaivt derivatives with respect to the gauge field (and the Kahler connection for D^ on fermions), and F^y is the field strength of A^. The superpotential W((p) and the gauge coupling field Tab((p) are constrained by AA = 1 supersymmetry to be complex analytic functions of (p. For M = I supersymmetric theories, it is very convenient to derive the above results from an A/^ = 1 superfield formulation, in which the complex analyticity of W and Xab emerges from the fact that these functions arise in F-terms, while the Kahler potential comes from a D-term. In F-terms, only superfields of one chirality enter; since a chiral fermion is in the same multiplet as the complex scalar field cp, but not ip, all ^-dependence emerging from F-terms is inherently complex analytic. With a generalization io J\f = 2 and A/* = 4 in mind, where no convenient off-shell superfield formulation is available, we prefer here to use component language throughout. For M = 2 supersymmetric theories, the gauge multiplet consists of an A^ = 1 gauge multiplet and an A/^ = 1 chiral multiplet in the adjoint representation of the gauge algebra. Thus, part of the components of the chiral field (^', V^') are in the adjoint representation, and we shall denote that part by (0^, t/^^), with the index a running through the adjoint representation. (The remaining components make up hypermultiplets.) Since the adjoint representation is real, there is no distinction between a and a. We shall concentrate on that part of the Lagrangian (1) which involves only the A/* = 2 vector multiplet fields. Enforcing M = 2 supersymmetry on the A/* = 1 Lagrangian (1) for the vector multiplet is not so easy. However, it is straightforward to enforce some necessary conditions. The M = 2 supersymmetry algebra is invariant under an SU(2)r group which rotates the two independent supercharges into one another, and thus rotates the two spinors in the J\f = 2 gauge multiplet into one another as well. In the A/* = 1 language used in (1), these two spinors are V^^ and X^.Af = 2 supersymmetry requires invariance under SU(2)r, and thus invariance of the Lagrangian under this symmetry. Invariance of the kinetic terms for X and t//" in (1) immediately yields a relation between the Kahler metric and the gauge coupling function Since Tab((p) is a complex analytic function of 0, the partial derivative of (2) with respect to (p^ is complex analytic, and thus
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 49 for some complex analytic function Tabc of (j). The most general solution to (3) is very easily obtained by integrating up twice, and may be expressed in terms of a single complex analytic function T, called the superpotential. In terms of T, the quantities r and K are given by This restricted form of the A/^ = 2 effective action is closely related with special geometry. Imposing SU(2)r -symmetry on the Yukawa couplings requires that the superpotential for the gauge scalars be similarly restricted. The resulting expressions are rather complicated, and we shall give below only the special cases needed for our analysis. Analogous conditions are required upon inclusion of hypermultiplets, but we shall not give those here. Once these necessary conditions arising from SU(2)r invariance have been imposed, it may in fact be shown that the Lagrangian obtained in this way is indeed J\f = 2 supersymmetric (see de Wit and van Proeyen [1984]). 3.2.3 The Set- Up for Seiberg- Witten Theory The starting point for Seiberg-Witten theory is an A/* = 2 supersymmetric Yang-Mills theory with gauge algebra Q and hypermultiplets in a representation RofQ with masses my. The microscopic Lagrangian is completely fixed by J\f = 2 supersymmetry in terms of the gauge coupling g and the instanton angle 0, and is given by £ = ^^Mv^"" + sl^^Mv^"" + D^4>D''<I> + m, <A]2 + •.. (5) where we have neglected hypermultiplet and fermion terms. The low energy effective theory corresponding to this model can be analyzed by studying first the structure of the vacuum. Af = 2 supersymmetric vacuum states can occur whenever the vacuum energy is exactly zero. Since the energy is always positive in a supersymmetric theory, we are guaranteed that any zero energy solution is a vacuum. This is the case here for vanishing gauge fields and constant gauge scalar fields cp for which the potential energy term also vanishes. The potential energy vanishes if and only if [0, 0] = 0, a condition equivalent to the vacuum expectation value of 0 being a linear combination of the Cartan generators of the gauge algebra Q, n < (j) >= 2_]ajhj, n = rank Q. (6) 7 = 1 Here, the complex parameters aj are usually referred to as the quantum moduli, or also as the quantum order parameters of the A/* = 2 vacua.
50 E. D'HOKER and D.H. PHONG For generic values of the parameters aj, the ^-gauge symmetry will be broken down to f/(l)"/Weyl(^), and the low energy theory is that ofn different Coulomb fields, up to global identifications by Weyl(^). Since A/^ = 2 supersymmetry is unbroken in any of these vacua, the low energy effective Lagrangian will have to be invariant under Af = 2 supersymmetry. But, we have already given a description of all such effective actions before, in terms of a complex analytic superpotential ^(0). In the case of n different U(l) gauge fields, this effective Lagrangian is particularly simple, and we have >Ceffective = -Im(r,y) F^,F^^^' + -Re(r,7)F;,F^^^' + d^^^d^cl^Dj + fermions (7) Here, the dual gauge scalar 0d and the gauge coupling function r,y are both given in terms of the prepotential T (pBi = T// = (8) ^^ d(j)j '^ d(j)id(j)j ^ ^ The form of the effective Lagrangian (7) is the same for any of the values of the complex moduli of J\f = 2 vacua, with the understanding that the fields 0y take on the expectation value < (pj >= aj. Since the prepotential J-'icp) is a function of the fields 0 only, but not of derivatives of 0, the prepotential will be completely determined by its values on the vacuum expectation values of the field, namely by its values on the quantum order parameters aj. 3.2.4 The Seiberg-Witten Solution The object of Seiberg-Witten theory is the determination of the prepotential T{aj), from which the entire low energy effective action will be known. This is achieved by exploiting the physical conditions satisfied by T (see Seiberg and Witten [1994]), (1) !F{aj) is complex analytic in aj in view of J\f = 2 supersymmetry, as shown in b) above. (2) The matrix Im r,y = Im9,9y^ is positive definite, since by (7), it coincides with the metric on the kinetic terms for the gauge fields Aj. (3) The large aj behaviour is known from perturbative quantum field theory calculations and asymptotic freedom, and is given by J^(a) ^ (at — aj)^ ln(ai — aj)^. More precisely, for gauge algebra Q and hypermultiplets in the representation R of Q, T{a) is of the form __[ ^ (o,.a)in—^^- Y^ (X.a+m)ln -^ ]. 871/
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 51 Here A is a dynamically generated scale introduced by renormalization, /i^ is the quadratic Casimir of Q (equal to the dual Coxeter number), I(R) is the Dynkin index of the representation R, and 7Z(Q) and W(R) denote respectively the roots of Q and the weights of the representation R. The terms on the right hand side of (9) represent respectively the classical prepotential, the one-loop perturbative corrections (higher loops do not contribute in view of non-renormalization theorems), and the instanton corrections jr{d) _ jr^^{2h^-i{R))d ^^ ^^X orders d. In general, it is prohibitively difficult to determine the coefficients Td from field theory methods. For conformally invariant theories, the expansion (9) is replaced by a similar one where the dynamical scale A is replaced by a modular invariant q = ^^^'^ (see e.g. (49) and (54-56) below). As a result of the requirements points 1 and 2 above, it follows immediately that !F cannot be a single-valued function of the aj. For if it were, S r/y would be both harmonic and bounded from below, which would imply that it must be independent of aj. But, from point 3, we know that xtj is not constant at large aj. And indeed, from point 3 again, it is clear that neither T nor zij are single valued functions of the aj. As is clear from the large aj behaviour r/y (a) ~ In (a, — fly), one of the ways in which Xij (a) is multiple valued is by shifts of any of the matrix elements by an integer. This ambiguity does not affect the physics of the low energy effective action (7), because the constant shifts in Re(r/y) are like the shifts of the instanton angle 0 by In times an integer and not observable. A more complicated multiple-valuedness consists in taking r -^ — r ~ ^ and corresponds to electric-magnetic duality, as shown by Seiberg and Witten [1994]. The combination of these two types of transformations produces the full duality group SL (2n, Z) of monodromies of r. A natural setting in which the above monodromy problem may be solved is provided by families of Riemann surfaces, called the Seiberg-Witten curves, denoted by P. Indeed, letting the quantum moduli ay correspond to moduli of the Riemann surfaces, there is automatically a complex analytic period matrix, whose imaginary part is positive definite, and whose monodromy group corresponds to the modular group of the surface. For g = SU{2) gauge group and no hypermultiplets for example, the Seiberg-Witten curve is a of genus 1, and may be represented as a double sheeted cover of the complex plane, r(M) = {(x, y)\ y^ = {x — A)(x + A){x — u)}. Here m is an auxiliary parameter, which will be related to the quantum modulus a, and A is the renormalization scale. We shall choose the branch cut between the points jc = ibA. The quantum modulus and prepotential are then given by 1 / dx dT(a) 1 / dx a(u) = —^ (p (x - u)— aoiu) = —- = —^ (p (x - u)— (10) Ini J A y da Ini Jb y where the A-cycle may be chosen around the branch cut between ibA and the 5-cycle between the branch points +A and u.Asu -^ ±A, the elliptic curve produces a singularity which physically is interpreted as caused by the vanishing of the mass of a magnetic monopole or dyon. Starting from the Seiberg-Witten solution for gauge group Q = SU(2), one may abstract the general set-up of the Seiberg-Witten solution, expected for arbitrary gauge algebra Q with rank n and general hypermultiplet representation. The ingredients are
52 E. D'HOKER and D.H. PHONG (1) The Seiberg-Witten curve is a family of Riemann surfaces r(Mi, • • • , m„) dependent on n auxiliary complex parameters Uj, which are related to the quantum moduli aj. The Seiberg-Witten curve will also depend upon the gauge coupling g and ^-angle and on the hypermultiplet masses ruk. (2) The Seiberg-Witten meromorphic differential 1-form dX on F, whose residues are linear in the hypermultiplet masses nik. Since the hypermultiplet masses receive no quantum corrections as aj varies, the derivatives d(dX)/daj are holomorphic 1-forms. (3) The quantum moduli and the prepotential are given by 1 / dJ=' I r aj = -—7 (b dX, aoj = t— = t—: (p dX. (11) 27TI /a, 9fly 27TI Jb- Shortly after the initial work of Seiberg and Witten, the curves and differentials for general SU(N), with and without hypermultiplets in the fundamental representation were proposed, as well as generalizations to the gauge groups.^'OCA^) and Sp(N) (see Lerche [1996,1997], Krichever and Phong [1997], Donagi [1997] and Marshakov [1997] for reviews). Use was made of the /?-charge assignments of the fields, the singularity structure of the degenerations of the Seiberg-Witten curve, and much educated guess work. 3.3 Twisted and Untwisted Calogero-Moser Systems 3,3,1 The SU(N) Elliptic Calogero-Moser System The original elliptic Calogero-Moser system is the system defined by the Hamiltonian 1 1 H(x,p) =-Y^pf --m^Y^pixi-Xj) (12) Here m is a mass parameter, and p (z) is the Weierstrass p-function, defined on a torus C/(2coiZ + 2co2Z). As usual, we denote by r = C02/CO1 the moduli of the torus, and set q = ^^TTir jYic well-known trigonometric and rational limits with respective potentials --m^y^ ^ ^ ^ and - :::^^ Y^ : ^ 2 ^4sh2(^) 2 ^.(Xi-Xj)^ arise in the limits coi = —in, C02 ^^ 00 and co\,cl>2 -^ 00. All these systems have been shown to be completely integrable in the sense of Liouville, i.e. they all admit a complete set of integrals of motion which are in involution (see Calogero [1975], Moser [1975] and Braden [1998]). For a recent review of some applications of these models see Polychronakos [1999].
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 53 Our considerations require however a notion of integrability which is in some sense more stringent, namely the existence of a Lax pair L(z), M(z) with spectral parameter z. Such a Lax pair was obtained by Krichever [1980] in 1980. He showed that the Hamiltonian system (12) is equivalent to the Lax equation L(z) = [L(z), M(z)], with L(z) and M(z) given by the following N x N matrices Lijiz) = PiSij - m(\ - 8ij)<^(xi - Xj,z) Mijiz) = mSij Y^p(xi - Xk) - m{\ - 8ij)^\xi - Xj, z). (13) The function 0(x, z) is defined by C|>(;C,z) = ^^^.-^(^>, (14) a(z)cr(x) where (7 (z), ^(z) are the usual Weierstrass a and ^ functions on the torus C/(2co[Z-\-2co2Z). The function 0(jc, z) satisfies the key functional equation <t>(x,zW(y,z) - <^(y,zW(x,z) = (p(x) - p(y))<^(x + y,z). (15) It is well-known that functional equations of this form are required for the Hamilton equations of motion to be equivalent to the Lax equation L(z) = [L(z), M(z)] with a Lax pair of the form (13). Often, solutions had been obtained under additional parity assumptions in X (and >^), which prevent the existence of a spectral parameter. The solution 0(jc, z) with spectral parameter z is obtained by dropping such parity assumptions for general z. It is a relatively recent result of Braden and Buchstaber [1997] that, conversely, general functional equations of the form (15) essentially determine 0(jc, z). 3.3.2 Calogero-Moser Systems defined by Lie Algebras As Olshanetsky and Perelomov [1976,1981] realized very early on, the Hamiltonian system (12) is only one example of a whole series of Hamiltonian systems associated with each simple Lie algebra. More precisely, given any simple Lie algebra Q, Olshanetsky and Perelomov [1976, 1981] introduced the system with Hamiltonian 1 ' 1 i=l aeniG) where r is the rank of Q, 71{G) denotes the set of roots of Q, and the m |c^| are mass parameters. To preserve the invariance of the Hamiltonian (16) under the Weyl group, the parameters m\a\ depend only on the orbit |Qf | of the root a, and not on the root a itself. In the case of Atv-i = ^^/(A^),it is common practice to use A^ pairs of dynamical variables(jc/, /?,),since the roots of Ayv-i lie conveniently on a hyperplane in C^. The dynamics of the system are unaffected if we shift all Xi by a constant, and the number of degrees of freedom is effectively N — I = r. Now the roots of SU(N) are given by a = et — ej,l < /, j < A, / ^ j. Thus we recognize the original elliptic Calogero-Moser system as the special case of (16) corresponding to A^-i. As in the original case, the elliptic systems (16) admit rational and trigonometric limits. Olshanetsky and Perelomov succeeded in constructing a Lax pair for all these systems in the case of classical Lie algebras, albeit without spectral parameter [1976, 1981].
54 E. D'HOKER and D.H. PHONG 3.3.3 Twisted Calogero-Moser Systems defined by Lie Algebras It turns out that the Hamiltonian systems (16) are not the only natural extensions of the basic elliptic Calogero-Moser system. A subtlety arises for simple Lie algebras Q which are not simply-laced, i.e., algebras which admit roots of uneven length. This is the case for the algebras i5„, C„, G2, and F4 in Cartan's classification. For these algebras, the following twisted elliptic Calogero-Moser systems were introduced by D'Hoker and Phong [1998a,b] ^twisted ^1 ^^2 _1 ^ ml^^^^.,{a.x). (17) i=\ aen{G) Here the function v(a) depends only on the length of the root a. If ^ is simply-laced, we set y(Qf) = 1 identically. Otherwise, for Q non simply-laced, we set y(Qf) = 1 when a is a long root, v(a) =2 when a is a short root and Q is one of the algebras i5„, C„, or F4, and v(a) = 3 when a is a short root and ^ = G2. The twisted Weierstrass function Pv(z) is defined by y-l Pv(z) = y2p(z-\-2cOa-), (18) where coa is any of the half-periods a>i, a>2, or coi + a>2- Thus the twisted and untwisted Calogero-Moser systems coincide for Q simply laced. The original motivation for twisted Calogero-Moser systems was based on their scaling limits (which will be discussed in the next section) (see D'Hoker and Phong [1998a,b]). Another motivation based on the symmetries of Dynkin diagrams was proposed subsequently by Bordner, Sasaki, and Takasaki [1998]. 3.4 Scaling Limits of Calogero-Moser Systems 3.4.1 Results of Inozemtsev for An-i For the standard elliptic Calogero-Moser systems corresponding to A^-i, Inozemtsev [1989a,b] has shown in the 1980s that in the scaling limit i_ m = Mq 2A^, ^ ^- 0 (19) Xi = Xi - 2a>2 —, 1 < / < A^ (20) where M is kept fixed, the elliptic A^-i Calogero-Moser Hamiltonian tends to the following Hamiltonian ^ TV N-l The roots et — et^i, I < i < N — I, and e^ — ei can be recognized as the simple roots of the affine algebra aJ^_j. (For basic facts on affine algebras, we refer to Goddard and Olive [1986]). Thus (21) can be recognized as the Hamiltonian of the Toda system defined
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 55 3.4.2 Scaling Limits based on the Coxeter Number The key feature of the above scaling limit is the collapse of the sum over the entire root lattice of A TV-1 in the Calogero-Moser Hamiltonian to the sum over only simple roots in the Toda Hamiltonian for the Kac-Moody algebra A)^_j. Our task is to extend this mechanism to general Lie algebras. For this, we consider the following generalization of the preceding scaling limit m = Mq-2\ (22) X = X - 2co28p'', (23) Here x = (xt), X = (X,) and p^ are r-dimensional vectors. The vector x is the dynamical variable of the Calogero-Moser system. The parameters 8 and p^ depend on the algebra Q and are yet to be chosen. As for M and X, they have the same interpretation as earlier, namely as respectively the mass parameter and the dynamical variables of the limiting system. Setting coi = —in, the contribution of each root a to the Calogero-Moser potential can be expressed as m^p(ax) = -M^ ^ — -—. (24) 2 ^—^ ch(Qf • X — 2nco2) — 1 n=-oo ^ ^ It suffices to consider positive roots a. We shall also assume that 0 < 5 a • p^ < 1. The contributions of the n = 0 and n = —\ summands in (24) are proportional to ^2ft>2(5-5a p"^) and ^2ft>2(5-i+5a p"^) respectively. Thus the existence of a finite scaling limit requires that 5 <5Qf •p'' < 1-5. (25) Let Of,, 1 < f < r be a basis of simple roots for ^. If we want all simple roots a, to survive in the limit, we must require that Gti p"" = 1, 1 <^' <r. (26) This condition characterizes the vector p^ as the level vector. Next, the second condition in (18) can be rewritten as 5{1 + maxa (ot • p^)} < 1. But hg = 1 + max (Of -p^) (27) a is precisely the Coxeter number of Q, and we must have 8 < -^. Thus when 8 < ^, the contributions of all the roots except for the simple roots of Q tend to 0. On the other hand, when 8 = 7;^, the highest root ofo realizing the maximum over a in (27) survives. Since ng —Qfo is the additional simple root for the affine Lie algebra Q^^\ we arrive in this way at the following theorem, which was proved in D'Hoker and Phong [1998b].
56 E. D'HOKER and D.H. PHONG Theorem 3.1 Under the limit (22-23), with 8 = -^, and p^ given by the level vector, the Hamiltonian of the elliptic Calogero-Moser system for the simple Lie algebra Q tends to the Hamiltonian of the Toda system for the affine Lie algebra Q^^\ 3.4.3 Scaling Limit based on the Dual Coxeter Number If the Seiberg-Witten spectral curve of the A/* = 2 supersymmetric gauge theory with a hypermultiplet in the adjoint representation is to be realized as the spectral curve for a Calogero-Moser system, the parameter m in the Calogero-Moser system should correspond to the mass of the hypermultiplet. In the gauge theory, the dependence of the coupling constant on the mass m is given by i v/ m ~^ X = —/iXln —y ^=> m = Mq ''q (28) ATT M^ where /i^ is the quadratic Casimir of the Lie algebra Q. This shows that the correct physical limit, expressing the decoupling of the hypermultiplet as it becomes infinitely massive, is given by (22), but with 5 = tV . To establish a closer parallel with our preceding discussion, we recall that the quadratic Casimir h^ coincides with the dual Coxeter number of Q, defined by /j^ z^l+maxCa^'-p), (29) where a^ = ^ is the coroot associated to a, and P = ^ l^a>o ^ ^^ ^^^ well-known Weyl vector. For simply laced Lie algebras Q (ADE algebras), we have hg = h^, and the preceding scaling limits apply. However, for non simply-laced algebras (Bn, Cn, G2, F4), we have hg > ^g, and our earlier considerations show that the untwisted elliptic Calogero-Moser Hamiltonians do not tend to a finite limit under (28), ^ ^- 0, M is kept fixed. This is why the twisted Hamiltonian systems (17) have to be introduced. The twisting produces precisely to an improvement in the asymptotic behaviour of the potential which allows a finite, non-trivial limit. More precisely, we can write 2 ^^ 2 m'pAx) = - Y. T-; 1 TT- (^^> 2 ^—^ chv(x — 2nco2) — 1 Setting X = X — 2co28^p, we obtain the following asymptotics = v^mA 2 2 21 -^ 2a>2(ra-p-«-)-«-.X^^-2<«2(l-^-«--p-r)+a-.X^ ifaislong; " ^ ' ' e-2'»2(^"« p-^ )-a^-x if a is short. (31) This leads to the following theorem (see D'Hoker and Phong [1998b]).
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 57 ]_ Theorem 3.2 Under the limit x = X -\- 2co2T7-p, m = Mq ^^Q , with p the Weyl vector and q ^^ 0, the Hamiltonian of the twisted elliptic Calogero-Moser system for the simple Lie algebra Q tends to the Hamiltonian of the Toda system for the affine Lie algebra (Q^^^Y. So far we have discussed only the scaling limits of the Hamiltonians. However, similar arguments show that the Lax pairs constructed below also have finite, non-trivial scaling limits whenever this is the case for the Hamiltonians. The spectral parameter z should scale as ^^ = Z^ 2, with Z fixed. The parameter Z can be identified with the loop group parameter for the resulting affine Toda system. 3.5 Lax Pairs for Calogero-Moser Systems 3.5.1 The General Ansatz Let the rank of ^ be n, and d be its dimension. Let A be a representation of Q of dimension A^, of weights Xi, I < I < N. Let uj e C^ be the weights of the fundamental representation of GL(N, C). Project orthogonally the m/'s onto the X/'s as sui = Xi -\-vi, Xi A.VJ. (32) It is easily verified that 5"^ is the second Dynkin index. Then ocij =Xi -Xj (33) is a weight of A (g) A* associated to the root uj —uj of GL(N, C). The Lax pairs for both untwisted and twisted Calogero-Moser systems will be of the form L = P-\-X, M = D + X, (34) where the matrices P,X,D, and Y are given by ^ = Y. ^iJ^iJ^'^JJ^ ^)Eij, Y = J2 Cij<^u(aij, z)Eij (35) and by P = ph, D=d'(h®h)-\-A. (36) Here /i is in a Cartan subalgebra Hq for Q,h is in the Cartan-Killing orthogonal complement of Hq inside a Cartan subalgebra H for GL(N, C), and A is in the centraHzer of Hq in GL(N, C). The functions <^ij(x, z) and the coefficients Cjj are yet to be determined. We begin by stating the necessary and sufficient conditions for the pair L(z), M(z) of (34) to be a Lax pair for the (twisted or untwisted) Calogero-Moser systems. For this, it is convenient to introduce the following notation Pu = ^uiocij • X, zWjii-ocij -x.z)- ^ij(-aij • X, zWjiiajj • x, z). (37)
58 E. D'HOKER and D.H. PHONG Then the Lax equation L(z) = [L(z), M(z)] implies the Calogero-Moser system if and only if the following three identities are satisfied ^ CljCjip'jjOCij =S^ Yl ^]a\Pv{ot){OL ' X) (38) /// aeniG) J2 CijCjip'iji^i -vj) = 0 (39) Y CikCkj{^ik^'kj-^'iK^Kj)=sCij^ijd'{vi-vj)+ Y^ A ijCkj^kj Ki^I,J Ki^IJ - Y. CiK^iK^KJ (40) The following theorem was established in D'Hoker and Phong [1998a]: Theorem 3.3 A representation A, functions <^ij, and coefficients Cjj with a spectral parameter z satisfying (38^0) can be found for all twisted and untwisted elliptic Calogero- Moser systems associated with a simple Lie algebra Q, except possibly in the case of twisted G2' In the case of Es, we have to assume the existence of a ±1 cocycle. 3.5.2 Lax Pairs for Untwisted Calogero-Moser Systems We now describe some important features of the Lax pairs we obtain in this manner. • In the case of the untwisted Calogero-Moser systems, we can choose ^ij{x,z) = 0(x, z), pij(x) = p(x) for all g. • A = 0 for all ^, except for E^. • For An, the Lax pair (13-14) corresponds to the choice of the fundamental representation for A. A different Lax pair can be found by taking A to be the antisymmetric representation. • For the BCn system, the Lax pair is obtained by imbedding Bn in GL(N, C) with N = 2n -\- I. When z = coa (half-period), the Lax pair obtained this way reduces to the Lax pair obtained by Olshanetsky and Perelomov [1976, 1981]. • For the B^ and D„ systems, additional Lax pairs with spectral parameter can be found by taking A to be the spinor representation. • For G2, a first Lax pair with spectral parameter can be obtained by the above construction with A chosen to be the 7 of G2. A second Lax pair with spectral parameter can be obtained by restricting the 8 of Bt, to the 7 0 1 of G2. • For F4, a Lax pair can be obtained by taking A to be the 26 0 1 of F4, viewed as the restriction of the 27 of Ee to its F4 subalgebra. • For £"6, A is the 27 representation. • For F7, A is the 56 representation.
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 59 • For £"8, a Lax pair with spectral parameter can be constructed with A given by the 248 representation, if coefficients c/j = ±1 exist with the following cocycle conditions c(A,, X — 8)c(X — 8, fji) = c(X, /x + 5)c(/x + 5, /x) when 8 • X = —8 ' fji = I, X - fjt = 0 c(X, fji)c(X — 8, fji) = c(X, X — 8) when 8 ' X = X • fji = I, 8 - fjt = 0 c{X, fji)c(X, X — fji) = — c(X — fji, —fji) whenX • /x = 1. (41) The matrix A in the Lax pair is then the 8 x 8 matrix given by Aab = Yl '^{c(fia.S)c(8, fib) + Cifia, Pa " ^)c{pa " 5, Pb))p{^ ' x) - ^ —{c{fia.^)c{8,fib)+c{fia.Pa-^)c{Pa-^.Pb))p{^'X) Aaa= Yl ^2p(8 ' X) -\-2m2p(Pa ' X), (42) where y^a, 1 < fl < 8, is a maximal set of 8 mutually orthogonal roots. • Explicit expressions for the constants Cjj and the functions d(x), and thus for the Lax pair are particularly simple when the representation A consists of only a single Weyl orbit of weights. This is the case when A is either (1) the defining representation of A„, C„ or D„; (2) any rank p totally anti-symmetric representation of A„; (3) an irreducible fundamental spinor representation of Bn or D„; (4) the27of £6;the56of £7. The, the weights X and /x of A provide unique labels instead of / and /, and the values of C/7 = Cxi^ are given by a simple formula Cxi^ = \a\ when a = X — fji is a root 0 otherwise The expression for the vector d may be summarized by sd'Ux= ^ mi8\p(8'x) ),-8=l; 8^=2 (For Cfi, the last equation has an additional term, as given in D'Hoker and Phong [1998a].) In each case, the number of independent couplings m\a\ equals the number of different root lengths.
60 E. D'HOKER and D.H. PHONG 3.5.3 Lax Pairs for Twisted Calogero-Moser Systems Recall that the twisted and untwisted Calogero-Moser systems differ only for non-simply laced Lie algebras, namely i5„, C„, G2 and F4. These are the only algebras we discuss in this paragraph. The construction (38-^0) gives then Lax pairs for all of them, with the possible exception of twisted G2. Unlike the case of untwisted Lie algebras however, the functions O/7 have to be chosen with care, and differ for each algebra. More specifically, • For Bn, the Lax pair is of dimension N = 2n, admits two independent couplings m 1 and m2, and f 0(jc,z), ifI-J^O,±n 0/j(jc,z)= . (43) 1 02(^Jc,z), ifI-J = ±n Here a new function O2 (x, z) is defined by ^2(-x, z) = —^ ., (44) 2 "^(couz) • For Cn, the Lax pair is of dimension N = 2n-\-2, admits one independent coupling m2, and 0/j(x,z) = 02(x+a>/7,z), where cou are given by coij 0, if / 7^ / = 1,2, ••• ,2az + 1; 0)2. if 1 < / < 2az, J = 2n + 2; (45) -0)2, if 1 < / < 2az, I = 2n + 2. • For F4, the Lax pair is of dimension N = 24, two independent couplings mi and m2, f 0(x,z), if X/x = 0; ^Xf^(x,z)= \ ^i(x,z), ifX/x=^; (46) [ <^2(^x,z), ifX'fji = -1. where the function Oi (jc, z) is defined by Oi(jc, z) = 0(jc, z) - ^^'^^^>+^^^0(jc + a>i, z) (47) Here it is more convenient to label the entries of the Lax pair directly by the weights X = Xi and fjt = Xj instead of / and /. • For G2, candidate Lax pairs can be defined in the 6 and 8 representations of G2, but it is still unknown whether elliptic functions <^ij(x, z) exist which satisfy the required identities. We note that recently Lax pairs of root type have been considered (see Bordner etal. [1998, 1998] and Bordner and Sasaki [1998]) which correspond, in the above Ansatz (34-36), to A equal to the adjoint representation of Q and the coefficients Cjj vanishing for / or / associated with zero weights. This choice yields another Lax pair for the case of Fg.
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 61 3.6 Calogero-Moser and Seiberg-Witten Theory for SU(N) The correspondence between Seiberg-Witten theory for J\f = 2 super-Yang-Mills theory with one hypermultiplet in the adjoint representation of the gauge algebra, and the elliptic Calogero-Moser systems was first established in D'Hoker and Phong [1998d], for the gauge algebra Q = SU(N). We describe it here in some detail. All that we shall need here of the elliptic Calogero-Moser system is its Lax operator L(z), whose N X N matrix elements are given by Lijiz) = PiSij - m(l - 8ij)<t>(xi - Xj, z) (48) Notice that the Hamiltonian is simply given in terms of L by //(jc, /?) = ^trL(z)^ + Cp (z) withC = -^m^A^CA^ - 1). 3.6.1 Correspondence of Data The correspondence between the data of the elliptic Calogero-Moser system and those of the Seiberg-Witten theory is as follows. (1) The parameter m in (48) is the hypermultiplet mass; (2) The gauge coupling g and the ^-angle are related to the modulus of the torus £ =C/(2a>iZ + 2a>2Z)by 0)2 0 Ani ^^^^ T = — = — + -^ ; (49) (0\ ATT g^ (3) The Seiberg-Witten curve F is the spectral curve of the elliptic Calogero-Moser model, defined by r = {(k, z) G C X E, dQi{kI - L(z)) = 0} (50) and the Seiberg-Witten 1-form is dX = k dz. T is invariant under the Weyl group of SU(N). (4) Using the Lax equation L = [L, M], it is clear that the spectral curve is independent of time, and can be dependent only upon the constants of motion of the Calogero-Moser system, of which there are only A^. These integrals of motion may be viewed as parametrized by the quantum moduli of the Seiberg-Witten system. (5) Finally, dX = kdz is meromorphic, with a simple pole on each of the A^ sheets above the point z = 0on the base torus. The residue at each of these poles is proportional to m, as required by the general set-up of Seiberg-Witten theory, explained in section 3.2. 3.6.2 Four Fundamental Theorems While the above mappings of the Seiberg-Witten data onto the Calogero-Moser data is certainly natural, there is no direct proof of it, and it is important to check that the results inferred from it agree with known facts from quantum field theory. To establish this, as well as a series of further predictions from the correspondence, we give four theorems (the proofs may be found in D'Hoker and Phong [1998d] for the first three theorems, and in D'Hoker and Phong [1998] for the last one).
62 E. D'HOKER and D.H. PHONG Theorem 3.4 The spectral curve equation det(/:/ — L{z)) = 0 is equivalent to ^1 f ^a - m^)\z\H(k) = 0 (51) where H(k) is a monic polynomial in k of degree N, whose zeros (or equivalently whose coefficients) correspond to the moduli of the gauge theory. If H(k) = Y[i=i(k — h), then 1 ^^0 27ii Jj^. ' 2 lim i) kdz = ki m. ^ '^TTi J A, Here, i^i is the Jacobi ?^-function, which admits a simple series expansion in powers of the instanton factor q = e^^^^, so that the curve equation may also be rewritten as a series expansion Y^(-Yq-2^^^-^^e^'H(k - Az . m) = 0 (52) neZ where we have set coi = —in without loss of generality. The series expansion (52) is superconvergent and sparse in the sense that it receives contributions only at integers that growlike Az^. Theorem 3.5 The prepotential of the Seiberg-Witten theory obeys a renormalization group-type equation that simply relates T to the Calogero-Moser Hamiltonian, expressed in terms of the quantum order parameters aj ^^ 2ni J A = H{x,p) = UxL{zf + Cp{z) (53) Furthermore, in an expansion in powers of the instanton factor q = e^^^^, the quantum order parameters aj may be computed by residue methods in terms of the zeros of H{k). The proof of (53) requires Riemann surface deformation theory (see D'Hoker and Phong [1998d]). The fact that the quantum order parameters may be evaluated by residue methods arises from the fact that Ay-cycles may be chosen on the spectral curve T in such a way that they will shrink to zero as ^ ^- 0. As a result, contour integrals around full-fledged branch cuts Aj reduce to contour integrals around poles at single points, which may be calculated by residue methods only. These methods were originally developed in D'Hoker et ah [1997a,b]). Knowing the quantum order parameters in terms of the zeros kj of H{k) =0 is a relation that may be inverted and used in (53) to obtain a differential relation for all order instanton corrections. It is now only necessary to evaluate explicitly the r-independent contribution to T, which in field theory arises from perturbation theory. This may be done easily by retaining only the n = 0 and n = \ terms in the expansion of the curve (52), so that z = lnH(k) —In H(k — m). The results of the calculations to two instanton order may be summarized in the following theorem (see D'Hoker and Phong [1998d]).
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 63 Theorem 3.6. Theprepotential, to 2 instanton order is given by T = ^^P^^^ -\-!F^^^ -\-T^^. The perturbative contribution is given by 2 ^-^ ' Sni ^ (a,- —aj) In (a; —Oj) — (at —Oj —m)^ ln(a; —aj —m)^ ' '-J while all instanton corrections are expressed in terms of a single function (54) as follows q I ^-^ ^ {at - ajY {at - aj - my jr(2) _ _^ Tv^ cv.Ai^2c./.A I A V^ Si{ai)Sj{aj) Si(ai)Sj(aj) Sni i^j (56) The perturbative corrections to the prepotential of (54) indeed precisely agree with the predictions of asymptotic freedom. The formulas (56) for the instanton corrections ^^^^ and are new, as they have not yet been computed by direct field theory methods. Perturbative expansions of the prepotential in powers of m have also been obtained in Minahan et ah [1997]. The moduli /:/, 1 < / < N,of the gauge theory are evidently integrals of motion of the system. To identify these integrals of motion, denote by S be any subset of {1, • • , A^}, and let 5* = {1, • • • , A^} \ 5, p(S) = p(xi — Xj) when S = {/, j). Let also ps denote the subset of momenta pt with i e S. We have (see D'Hoker and Phong [1998] Theorem 3.7 For any K,0 < K < NJetcrK(ki, - - - ,kN) = cfRik) be the K-th symmetric polynomial of {k\, • • • , k^), defined by H(u) = X1a:=o(~)^^^(^)"^~^- ^^^'^ [K/2] I CTKik) = CTKip) + ^ m^^ J2 ""K-lliPiul^S^y) Y[^P(Si) + ^] (57) 1=1 \SinSj\=28ij ' i = l ^1 3.6.3 Partial Decoupling of the Hypermultiplet and Product Gauge Groups The spectral curves of certain gauge theories can be easily derived from the Calogero-Moser curves by a partial decoupling of the hypermultiplet. Indeed, • the masses of the gauge multiplet and hypermultiplet are \ai — aj \ and \ai — aj + m\. In suitable limits, some of these masses become 00, and states with infinite mass decouple. The remaining gauge group is a subgroup of SU(N).
64 E. D'HOKER and D.H. PHONG • When the effective coupling of a gauge subgroup is 0, the dynamics freeze and the gauge states become non-interacting. Non-trivial decoupling limits arise when r ^- oo and m ^- oo. When all at are finite, we obtain the pure Yang-Mills theory. When some hypermultiplets masses remain finite, the U(l) factors freeze, the gauge group SU(N) is broken down to SU(Ni) x • • • x SU(Np), and the remaining hypermultiplets are in e.g. fundamental or bifundamental representations. For example, let A^ = 2Ni be even, and set ki = vi-\- Xi, kNi-\-j = V2-\- yj, 1 < /, j < Nu ^ith J^fli ^i = Jlfli yj = ^' (The term i; = fi — i;2 is associated to the U(l) factor of the gauge group). In the limit m ^- oo, ^ ^- 0, with xt, yj, fi = v — m and A = mq^ kept fixed, the theory reduces to a SU(Ni) x SU(Ni) gauge theory, with a hypermultiplet in the bifundamental (M, A^i) 0 (A^i, A^i), and spectral curve A(x) - t(-)^'B(x) - 2^'A^'(- - t^) = 0, (58) where A(x) = Ylf^liix - jc/), B(x) = Y[f=i(^ + M - yj)^ t = e^. This agrees with the curve found by Witten [1997] using M Theory, and by Katz et ah [1998] using geometric engineering. The prepotential of the S U (Ni )xSU (Ni) theory can be also read off the Calogero-Moser prepotential. It is convenient to introduce x\ \ I = 1,2, by jc^. = jc/, x^ ^ = >;/, 1 < / < A^i.Set Al=Y[(x-xP), B\x)= Y\{^Ji±{x-xf')), 5/(x) = ^^, where the ib sign in B^ (jc) is the same as the sign of J — I. Then the the first two orders of instanton corrections to the prepotential for the SU(Ni) x SU(Ni) theory are given by (1) _(-2A)^ y. Y^^i. (IK ^^^ 7=1,2 /G/ ^SU(N,.SU(N,- 8^, ^2^^Z.^(^.- ) 3^(/)2 +Z. (^(/)_ (/))2 • (59) We note that an alternative derivation of (51) was recently presented in Vaninslcy [1998]. 3.7 Calogero-Moser and Seiberg-Witten Theory for General Q We consider now the A/* = 2 supersymmetric gauge theory for a general simple gauge algebra Q and a hypermultiplet of mass m in the adjoint representation. Then (see D'Hoker and Phong [1998]):
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 65 the Seiberg-Witten curve of the theory is given by the spectral curve F = {(k, z) G C X S; dci(kl — L(z)) = 0} of the twisted elliptic Calogero-Moser system associated to the Lie algebra Q. The Seiberg-Witten differential dX is given by dX = kdz. The function R{k, z) = det(/:/ — L(z)) is polynomial in k and meromorphic in z. The spectral curve F is invariant under the Weyl group of ^. It depends on n complex moduli, which can be thought of as independent integrals of motion of the Calogero-Moser system. The differential dX = kdz is meromorphic on F, with simple poles. The position and residues of the poles are independent of the moduli. The residues are linear in the hypermultiplet mass m. (Unlike the case of SU(N), their exact values are difficult to determine for general G-) In the m ^- 0 limit, the Calogero-Moser system reduces to a free system, the spectral curve F is just the producer of several unglued copies of the base torus S, indexed by the constant eigenvalues of L(z) = p - h. Let kt, I < i < n,bcn independent eigenvalues, and Ai, Bt be the A and B cycles lifted to the corresponding sheets. For each /, we readily obtain cii = T— (p dX = -— (b dz = -r-^ki, ciDi = :z— <p dX= -— (p dz= :r-^rki. 2ni Jsi 2711 Jb 27TI Thus the prepotential J^ is given by ^ = | Yl^=i ^f- This is the classical prepotential and hence the correct answer, since in the m ^- 0 limit, the theory acquires an A/^ = 4 super symmetry, and receives no quantum corrections. The m ^- 00 limit is the crucial consistency check, which motivated the introduction of the fw/5'^^ J Calogero-Moser systems in the first place (see D'Hoker and Phong [1998a,b]). In view of Theorem 3.2 and subsequent comments, in the limit m ^- oo, ^ ^- 0, with !_ X = X-\-2co2 A-p,m = Mq ^^^ with X and M kept fixed, the Hamiltonian and spectral curve for the twisted elliptic Calogero-Moser system with Lie algebra Q reduce to the Hamiltonian and spectral curve for the Toda system for the affine Lie algebra (^^^^)^. This is the correct answer. Indeed, in this limit, the gauge theory with adjoint hypermultiplet reduces to the pure Yang-Mills theory, and the Seiberg-Witten spectral curves for pure Yang-Mills with gauge algebra Q have been shown by Martinec and Warner [1996] to be the spectral curves of the Toda system for (^^^^)^. The effective prepotential can be evaluated explicitly in the case of ^ = D„ forn < 5. Its logarithmic singularity does reproduce the logarithmic singularities expected from field theory considerations. As in the known correspondences between Seiberg-Witten theory and integrable models (D'Hoker and Phong [1998d] and D'Hoker et al [1997a]), we expect the following equation ^-f^Hf''^{x,p\ (60) ax to hold. Note that the left hand side can be interpreted in the gauge theory as a renormalization group equation.
66 E. D'HOKER and D.H. PHONG • For simply laced Q, the curves R(k,z) = 0 are modular invariant. Physically, the gauge theories for these Lie algebras are self-dual. For non simply-laced Q, the modular group is broken to the congruence subgroup To(2) for Q = Bn.Cn, F4, and to To(3) for G2. The Hamiltonians of the twisted Calogero-Moser systems for non-simply laced Q are also transformed under Landen transformations into the Hamiltonians of the twisted Calogero-Moser system for the dual algebra Q^. It would be interesting to determine whether such transformations exist for the spectral curves or the corresponding gauge theories themselves. Spectral curves for certain gauge theories with classical gauge algebras and matter in the adjoint representation have also been proposed in Uranga [1998] and Yokano [1998], based on branes and M-theory. Relations with integrable systems were discussed in Gorsky [1997], Gorsky et al [1995] and Cherkis and Kapustin [1997]. Acknowledgements We are happy to acknowledge invitations to lecture on the material presented in this paper by Harry Braden and Igor Krichever at the workshop "Integrability, the Seiberg-Witten and Whitham Equations" at Edinburgh, Sepember 1998, by Ryu Sasaki and Takeo Inami at the "Workshop on Gauge Theory and Integrable Models" at Kyoto, January 1999, and by Tohru Eguchi and Norisuke Sakai at "Supersymmetry and Unified Theory of Elementary Particles" at Kyoto, February 1999. We wish to thank these organizers for their warm hospitality and generous support. This research was supported in part by the National Science Foundation under grants PHY-95-31023 and DMS-98-00783. References Bordner, A., and Sasaki, R., "Calogero-Moser systems III: Elliptic potentials and twisting", hep- th/9812232. Prog. Theor. Phys. 101, 799-829 (1999b). Bordner, A., Corrigan, E., and Sasaki, R., "Calogero-Moser systems: a new formulation", hep- th/9805106. Prog. Theor Phys. 100, 1107-1129 (1998). Bordner, A., Sasaki, R., and Takasaki, K., "Calogero-Moser systems II: symmetries and foldings", hep-th/9809068. Prog. Theor Phys. 101, 487-518 (1999a). Braden, H.W., and Buchstaber, V.M., "The general analytic solution of a functional equation of addition type", Siam J. Math. Anal. 28, 903-923 (1997). Braden, H.W., "A conjectured R-matrix", J. Phys. A 31 (1998) 1733-1741. Calogero, E, "Exactly solvable one-dimensional many-body problems", Lett. Nuovo Cim. 13,411^16 (1975). Carroll, R., "Prepotentials, and Riemann surfaces", hep-th/9802130. Cherkis, A., and Kapustin, A., "Singular monopoles and supersymmetric gauge theories in three dimensions", hep-th/9711145. Nucl. Phys. B525, 215-234 (1998). D'Hoker, E., and Phong, D.H., "Calogero-Moser Lax pairs with spectral parameter for general Lie algebras", Nucl. Phys. B 530, 537-610, hep-th/9804124 (1998a). D'Hoker, E., and Phong, D.H., "Calogero-Moser systems in SU(N) Seiberg-Witten theory", Nucl. Phys. B 513, 405^44, hep-th/9709053 (1998d).
SEIBERG-WITTEN THEORY AND INTEGRABLE SYSTEMS 67 D'Hoker, E., and Phong, D.H., "Calogero-Moser, and Toda systems for twisted, and untwisted affine Lie algebras", Nucl Phys. B 530, 611-640, hep-th/9804125 (1998b). D'Hoker, E., and Phong, D.H., "Order parameters, free fermions, and conservation laws for Calogero-Moser systems", hep-th/9808156, to appear in Asian J. Math. (1998d). D'Hoker, E., and Phong, D.H., "Spectral curves for super Yang-Mills with adjoint hypermultiplet for general Lie algebras", Nucl Phys. B 534, 697-719, hep-th/9804126 (1998c). D'Hoker, E., Krichever, LM., and Phong, D.H., "The effective prepotential for A/" = 2 supersymmetric SU{Nc) gauge theories", Nucl. Phys. B489, 179, hep-th/9609041 (1997b). D'Hoker, E., Krichever, LM., and Phong, D.H., "The renormalization group equation for AT = 2 supersymmetric gauge theories", Nucl. Phys. B 494, 89-104, hep-th/9610156 (1997a). de Wit, B., and Van Proeyen, A., Nucl. Phys. B245, 89 (1984). Donagi, R., and Witten, E., "Supersymmetric Yang-Mills, and integrable systems", Nucl. Phys. B 460, 288-334, hep-th/9510101 (1996). Donagi, R., "Seiberg-Witten integrable models", in: Surv. Differ. Geom. IV, International Press, M.A. (1998). alg-geom/9705010. Goddard, P., and Olive, D., "Kac-Moody, and Virasoro algebras in relation to quantum physics", InternationalJ. Mod. Phys. A, I, 303^14 (1986). Gorsky, A., and Nekrasov, N., "Elliptic Calogero-Moser System from two-dimensional current algebra", hep-th/9401021. Gorsky, A., Gukov, S., Mironov, A., "SUSY field theories, integrable systems, and their stringy brane origin", Nucl. Phys. B1518, 689-713 (1998), hep-th/9710239. Gorsky, A., Krichever, LM., Marshakov, A., Mironov, A., and Morozov, A., "Integrability and Seiberg-Witten exact solution", Phys. Lett. B355, 466, hep-th/9505035 (1995). Gorsky, A., "Branes, and IntegrabiUty in the AT = 2 SUSY YM theory". Int. J. Mod. Phys. A12,1243, hep-th/9612238 (1997). Inozemtsev, L, "Lax representation with spectral parameter on a torus for integrable particle systems", Lett. Math. Phys. 17, 11-17 (1989a). Inozemtsev, L, "The finite Toda lettices", Comm. Math. Phys. 121, 628-638 (1989b). Katz, S., Mayr, P., and Vafa, C, "Mirror symmetry, and exact solutions of 4D A/" = 2 gauge theories". Adv. Theor. Math. Phys. 1, 53, hep-th/9706110 (1998). Krichever, LM., and Phong, D.H., "Symplectic forms in the theory of solitons", hep-th/9708170. Surveys in Differential Geometry, Vol. IV, 239-313, International Press, M.A. (1998). Krichever, I.M., "Elliptic solutions of the Kadomtsev-Petviashvili equation, and integrable systems of particles", Funct. Anal. Appl 14, 282-290 (1980). Lerche, W, hep-th/9611190, Nucl. Phys. Proc. Suppl. B55, 83 (1997). Lerche, W., "Introduction to Seiberg-Witten theory, and its stringy origins". Proceedings of the Spring School, and Workshop on String Theory, ICTP, Trieste (1996). Marshakov, A., "On integrable systems, and supersymmetric gauge theories", Theor Math. Phys. 112, 791-826, hep-th/9702083 (1997). Martinec, E., and Warner, N., "Integrable systems, and supersymmetric gauge theories", Nucl. Phys. B 459, 97-112, hep-th/9509161 (1996). Martinec, E., "Integrable structures in supersymmetric gauge, and string theory", hep-th/9510204. Minahan, L, Nemeschansky, D., and Warner, N., "Instanton expansions for mass deformed J\f = 4 super Yang-Mills theory", hep-th/9710146. Nucl. Phys. B528, 109-132 (1998). Moser, L, "Three integrable Hamiltonian systems connected with isospectral deformations", A(iv«nc^5 Math. 16, 197 (1975). Nekrasov, N., "Holomorphic bundles, and many-body systems", Comm. Math. Phys. 180,587 (1996). Olshanetsky, M.A., and Perelomov, A.M., "Completely integrable Hamiltonian systems connected with semisimple Lie algebras", Inventiones Math. 37, 93-108 (1976); "Classical integrable finite-dimensional systems related to Lie algebras", Phys. Rep. 71 C, 313^00 (1981). Polychronakos, A.P., "GeneraUzed Statistics in one dimension", hep-th/9902157. Les Houches Lectures 1998. Seiberg, N., and Witten, E., "Electro-magnetic duahty, monopole condensation, and confinement in M = 2 supersymmetric Yang-Mills theory", Nucl. Phys. B 426, 19-53, hep-th/9407087 (1994).
68 E. D'HOKER and D.H. PHONG Uranga, A.M., "Towards mass deformed M = A SO(N), and Sp(K) gauge theories from brane configurations", NucL Phys. B 526, 241-277, hep-th/9803054 (1998). Vaninsky, K., "On explicit parametrization of spectral curves for Moser-Calogero particles, and its applications", Intemat. Math. Res. Notices, pp. 509-529 (1999). Witten, E., "Solutions of four-dimensional field theories via M-Theory", Nucl. Phys. B 500, 3, hep-th/9703166(1997). Yokono, T., "Orientifold four plane in brane configurations, and Af = 4 USpilN), and SO (IN) theory", Nucl. Phys. B 532, 210-226, hep-th/9803123 (1998).
4 Seiberg-Witten Curves and Integrable Systems A. MARSHAKOV Theory Department, Lebedev Physics Institute, Moscow 117924, and ITEP, Moscow 117259, Russia E-mail address: mars@lpi.ru, andrei@heron.itep.ru This chapter gives an introduction into the subject of Seiberg-Witten curves and their relation to integrable systems. We discuss some motivations and origins of this relation and consider the explicit construction of various families of Seiberg-Witten curves in terms of corresponding integrable models. 4.1 Introduction Some time has already passed since the observation of Gorsky et al. [1995] that the effective theories for A/* = 2 vector supermultiplets (Seiberg and Witten [1994a,b] and Klemm et al. [1995]) can be reformulated in terms of integrable systems. This connection, though still not clearly understood, has become a beautiful example of the appearance of hidden integrable structures in (multi-dimensional) quantum gauge theories. Accordingly it has become quite a popular topic at many different conferences. In this chapter I was asked to review basic facts known of this relationship, to explain briefly the constituents of this correspondence, and to present a list of problems which deserve further investigation. The formulation of the Seiberg-Witten (SW) solution (Seiberg and Witten [1994a,b]) itself is very simple: the (Coulomb branch) low-energy effective action for the 4D A/* = 2 SUSY Yang-Mills vector multiplets can be described in terms of an auxiliary Riemann surface (complex curve) S equipped with meromorphic 1-differential dS. The complex structures arise here because supersymmetry requires the metric on the moduli space of massless complex scalars from M = 2 vector supermultiplets to be of "special Kahler form". The the Kahler potential ^(a, a) = Im ^- a, |^ is expressed through a holomorphic function T = ^(a) , the prepotential. This setup possesses several peculiar properties: 69
70 A. MARSHAKOV • The number of "live" moduli (of the complex structure) of S is strongly restricted (roughly "3 times" less than for a generic Riemann surface). The genus of S for SU(N) gauge theories is exactly equal^ to the rank of gauge group — i.e. to the number of independent moduli. • The variation of generating 1 -form dS over these moduli gives holomorphic differentials. • The periods of the generating 1-form =/. dS A &j) — <t dS give the set of "dual" masses — the W-bosons and the monopoles — while the period matrix TijiH) provides the set of couplings in the low-energy effective theory. The prepotential is a function of half of the variables (1), say T = .f (a), and Tij = 8fl^ _ d^j^ ^^^ daj dai daj This data means that the effective SW theory is formulated in terms of a classical finite-gap integrable system (see, for example, (Dubrovin et al [1985]) and references therein) and their Whitham deformations (Gorsky etal [1995]). The corresponding integrable models are well-known members of the KP/Toda family — for example, pure gauge theory corresponds to a periodic Toda chain (Gorsky et al [1985] and Martinec and Warner [1996]), the theories with broken J\f = 4 SUSY by an adjoint mass can be formulated in terms of the elliptic Calogero-Moser models (Donagi and Witten [1996]), the theories with extra compact dimensions (or Kaluza-Klein modes) give rise to appearance of relativistic integrable systems (Nekrasov [1998], Braden etal. [1998, 1999]). The aim of this chapter is rather modest, to give some arguments in favour of the appearance of complex curves S in the context of SUSY gauge theories and to present in a clear way how the form of these curves can be explicitly found in by means of Lax representations of well-known finite-dimensional integrable systems. I should stress that the curves S (and corresponding integrable systems) are auxiliary from a physical point of view since the quantities (1), (2) describing the effective theories depend only on the moduli of the SW curve, that is only on half of the variables of the classical integrable system. This dependence is governed by integrable systems which are, in a sense, derivative from those we are going to consider below — the hierarchies of Whitham and (generalized) associativity equations. Both these subjects are, however, beyond the scope of these notes and are described elsewhere in the volume. ^ For generic gauge groups one should speak instead of genus — the dimension of the Jacobian of a spectral curve — about the dimension of the Prym variety. In practice this means that for other than A a^-type gauge theories one should consider spectral curves with involution and only invariants under the involution possess physical meaning. We consider in detail only the A^ theories, the generalization to the other gauge groups is straightforward: for example, instead of periodic Toda chains (Toda [1981]), corresponding to A^ theories (Gorsky et al. [1995]), one has to consider the "generalized" Toda chains (Martinec and Warner [1996]), first introduced for different Lie-algebraic series (B, C, D, E, F and G) in (Bogoyavlensky [1981]).
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 71 4.2 Motivations and Origins Despite the absence of any consistent or complete explanation at the moment of why the SW curves are identical to the curves of integrable systems, let us start from some physical motivations. We consider, first, the perturbative limit of Af = 2 SUSY gauge theories and show that (degenerate) spectral curves and corresponding integrable systems already appear at this level. The situation is much more complicated for the non-perturbative picture and an explanation of the origin of the full SW curve exists only in the framework of non-perturbative string theory or M-theory. In the second part of this section we shall briefly discuss how the Lax representations of the SW spectral curves arise in this context. 4.2,1 Perturbative Spectral Curves Amazingly enough the relation between SW theories and integrable systems can already be discussed at the perturbative level, where J\f = 2 SUSY effective actions are completely defined by the 1-loop contributions (see Seiberg and Witten [1994a] and references therein). The scalar field 4> = ||0/^ || of the A/^ = 2 vector supermultiplet acquires nonzero VEV 4> = diag(0i,..., 07v) and the masses of "particles" — W-bosons and their superpartners are proportional to 0/y = 0/ — 0y due to the Higgs term [A^, 4>],^ = A^ (0, — 0^) in the SUSY Yang-Mills action. These masses can be written altogether in terms of the generating polynomial w = Pn(X) = det(X -<!>) = Yl(^ - (t>i) (3) (4> — the adjoint complex scalar, Tr4> = 0) via the residue formula rriij ^ (p Xdlogw = (p XdlogPN(X), (4) which for a particular "oo-like" contour Cij around the roots X = (j)i and X = (j)j gives rise to the Higgs masses. The contour integral (4) is defined on a complex plane (the X-plane with the A^ roots of the polynomial (3) points), a degenerate Riemann surface. The masses of the monopoles are naively infinite in this limit, since the corresponding contours (dual to Cij) start and end in the points where dS has pole singularities. This means that the monopole masses, proportional to the squared inverse coupling, are renormalized in perturbation theory and defined naively up to the masses of particle states times some divergent constants. The effective action (the prepotential) T, or the set of effective charges Tij (2), are defined in A/* = 2 perturbation theory by 1-loop diagrams which give rise to the logarithmic contribution (si2nr\ rj. v-^ 1 (mass)2 (0. _0.)2 (5 T).. = Tij ^ }^ log —j^ = log p . (5) masses Here A = Aqcd and the last equality is written only for pure gauge theories, since the only masses we have in this are the Higgs masses (4). That is all one has in the perturbative weak-coupling limit of the SW construction, when the instanton contributions to the prepotential being proportional to the powers of A^^ (or q^^ = ^^ttitTV ^^ uV-finite theories with bare coupling r) are (exponentially) suppressed, consequently one keeps
72 A. MARSHAKOV only those terms proportional to r or log A. We shall list several more examples below and demonstrate that these degenerate rational spectral curves are related to the family of trigonometric Ruijsenaars-Schneider and Calogero-Moser-Sutherland systems and the open Toda chain or Toda molecule. We start with the original case of SU(2) pure gauge theory. Eq. (3) turns into w = X^ —u, (6) with u = ^Tr4>^. In the parameterization of (Seibeg and Witten [1994a]) X = w = e^ = }? — M, Y — wX, and the same equation can be written as Y^ = X^(X-\-u). (7) The masses (4) are now defined by the contour integrals of X^dX XdX XdX , dX dS = Xdlog w = 2- = + -—- = VX + M —. (8) X^ — u X — y/u X + ^u X One notices that Eqs. (6)-(8) may be interpreted as the integration of the open SL(2) (the Liouville) Toda chain with the co-ordinate X = w = e^, momentum p = X and Hamiltonian (energy) u. The integration of the generating differential dS = pdq over the trajectories of the particles gives rise, in fact, to the monopole masses in the SW theory. This is actually a general rule, the perturbative J\f = 2 theories of the "SW family" give rise to the "open" or trigonometric families of integrable systems such as the open Toda chain, the trigonometric Calogero-Moser or Ruijsenaars-Schneider systems. This may easily be established at the level of the spectrum (4) and the effective couplings (5). For the A^-particle Toda chain case the corresponding (rational) curves are given by (3) and (4), while for the the trigonometric Calogero-Moser-Sutherland model we have dS = X—. (9) For the trigonometric Ruijsenaars-Schneider system we have Plf^\^) dw yo= .4 . . ds = \ogx—. (10) It is easy to see that the (perturbative) spectra are given by general formula (see Braden etal [1999]) nn e + nn ^^ /? R (11) n eZ. and contain, in addition to the Higgs part 0,;, the Kaluza-Klein (KK) modes ^ and the KK modes for fields with "shifted" (by e) boundary conditions. The e parameter here can be treated as a Wilson loop of a gauge field along the compact dimension — a different (or dual) kind of moduli in the theory and in a subclass of models plays the role of the mass of the adjoint matter multiplet.
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 73 4,2,2' Nonperturbative SW Curves, M-Theory and the d-Equation As an example, let us again consider the case of pure SU(2) J\f = 2 gauge theory (Seiberg and Witten [1994a]). The gauge group has rank 1 and from the "integrable" point of view the situation is trivial, since the corresponding integrable model is "one-dimensional" (that is, the phase space is two dimensional) and this case can be always solved explicitly. The full ("blown-up") spectral curve has the form A^ 7 7 w-\- — = 2A^ coshz = X^ -u = P2(X) (12) (cf. with (3)) and coincides with the equation relating the Hamiltonian (energy) u with the co-ordinate z = iq (or z = q) and momentum X = p of a particle moving in the SL(2) Toda-chain potential. This is the well-known physical pendulum (rather than the previously considered "Liouville wall"^.) There are two other (hyper)elliptic parameterizations of the SU{2) SW curve: the first one was proposed in Seiberg and Witten [1994a] Y^ = (X2-4A^)(X + m) X = P2{X) = X^-u Y = Xy with ^5 = (X + /i)^, Jr = ^, and another one used in Seiberg and Witten [1996] y2 = yo^ -\-hw^ + A^w w = A e^ y = Xw endowed with dS = y^ = ^^'+^^~+^^^^ and dt = ^. Since any 1-dimensional system with conserved energy is integrable, (12) gives Jr = ^ = ^=2^, (15) p X y where y^ = (X^ - uf - 4A^ = 4A^ sinh^ z. (16) Then rdx ^ rdx is the Abel map for (13) or (12). The (normalized) action integrals in the two different ("sine and sinh-Gordon") phases are the periods (a,aD)=(p dS = d) pdq = d) dqJu-\-A^ cos q, (18) J{A,B) J{A,B) J{A,B) 2> ^Note, that for the SW theory one has to consider both phases — the sine-Gordon and the sinh-Gordon — of an integrable system together, that is why sometimes people speak in this respect about a complex integrable system.
74 A. MARSHAKOV In the "sine phase" if m >> A^ the interaction is inessential and, the main effect for the integral (18) comes from a ^ Vm I dq ^ \fu x const (19) J A In the other "phase", corresponding to imaginary values of q (q -^ iq + n) the action integral (18) turns into a^ = (p pdq = (p dqJu — A^ cosh^ (20) which in the first approximation is equal to a^ ^ I dqJu — A^ cosh q ^ ^/u I dq ^ ^/u x q^ = ^/u log—r:. (21) Jb ^ Jb A^ Here we have denoted by q^ the "turning point" A^e^* = m. It is easy to see that expressions (19) and (21) give the perturbative values of the W-boson m^ = u and monopole m\f = (^) masses in the SU(2) SW theory. The prepotential for m >> A^ is thus This is an obvious perturbative (1-loop) result of J\f = 2 SUSY Yang-Mills theory. The main hypothesis of Seiberg and Witten [1994a], is that formula (18) (and the prepotential or the set of coupling constants they define) are valid beyond perturbation theory. In this case the right hand side of (22) would contain an infinite instanton series (in ^ for the SU(2) gauge group) but this can be encoded into a relatively simple modification of the SW curves. In the construction inspired by M-theory, moduli arise either as the positions of the D-branes, 0i ^ ^, or as the monodromies of the gauge fields e = § AMdx^ along the compactified directions (and are related to the positions of branes by T-duality), i.e. they are given by the set of data (Am, 4>^'^). The perturbative spectral curves (6), (9), (10) correspond in this picture to the /7-branes (/7-dimensional hypersurfaces with the (p + 1)-dimensional world volume) with D(p — l)-brane sources (see Witten [1997]). In the particular case when the configuration can be described by two real (or one complex component) of both kind of fields (A, 4>) one can define the full (smooth) spectral curves S^ as a cover of some bare spectral curve So (usually a torus, which can be naturally chosen when one has at least two compactified dimensions) by generalization of the equation (3) det(X - 4)(z)) = 0. (23) Here 4> = 4>(z) is now function (in fact a 1-differential) on the bare curve So and obeys (see Hitchin [1987] and Gorsky and Nekrasov [1994]) aO + [A, O] = ^ /^"^5^2)(p _ p^^ (24)
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 75 i.e. it is holomorphic in the complex structure determined by A. The invariants of A can be thought of as co-ordinates (one commuting set of variables) while the invariants of O as hamiltonians (another commuting set of variables) of an integrable system. The M-theory point of view implies that the VEV 4> becomes a function on some base spectral curve So (usually a cylinder or torus) appearing as a part of the (compactified) brane world volume and satisfies a 9-equation. Such holomorphic (or, better, meromorphic) objects were introduced long ago in Dubrovin et al [1985] as Lax operators for finite-dimensional integrable systems, holomorphically depending on some spectral parameter. The (first-order) equation (24) arises from the BPS-like condition (Diaconescu [1996] and Marshakov et al. [1998]) of the type Qi/r = FmDm^ + ^mn^mn = 0 which determines the form of the Lax operator O and, thus, the shape of the curve (23). On torus with p marked points zi,..., z^ we obtain (/, 7 = 1,..., A^) 94>o- + {qi - qj)<i>ij = Y, 4f^(z - Za). (25) a=l The solution has the form {qij = qi — qj)^ '^ijiz) = Sij I Pi + J2 4"'9 log^*(2 - 2«|-^)) + (26) ^ "^ V '' o.{z-zoc)e*e^qij) The exponential (nonholomorphic) part can be removed by a gauge transformation <^ij{z)^iU-'<^U)ij{z) (27) with Uij = e^'^Sij. The additional conditions to the matrices J^j E^.f-O (28) imply that the sum of all the residues of the function O,, vanish. Setting Tr/^"^ = ma (29) with m-a = const, the m^ are parameters ("masses") of the theory. The spectral curve equation becomes TV V(X; z) ^ det (X - <l>(z)) = X^ + Vx^'V^a) = 0 (30) NxN *•—' '''ks, usual, we denote by ^*(z|t) = ^n (z|t) the (only on a torus) odd theta-function ^*(0|t) =0.
76 A. MARSHAKOV where fk{z) are some functions (in general with k poles) on elliptic curve. If, however, the J {a) ^^ further restricted by rank/^"^</, / < A^, (31) the functions fk{z) will have poles at zi,..., z^ of order not greater than /. The generating differential, as usual, should be dS = Xdz (32) and its residues at the marked points (z^, )^^^\za)) (different i correspond to the choice of different sheets of the covering surface) are related to the mass parameters (29) by TV rua = YCSz^Xdz = y^jcs^-\ ^^X^^\z)dz = res^^Tr4>Jz. (33) We shall see that general form of the curve (30) coincides with the general curves arising in the SW theory (in many important cases the torus should degenerate into a cylinder). For example, the /? = 1, / = 1 case gives rise to the elliptic Calogero-Moser model (see eq. (56) below). Thus, in the M-theory picture, it becomes clear that the full non-perturbative SW curves are smooth analogues of their degenerate perturbative cousins. This blowing up corresponds to the "massive" deformation of the previous family of integrable systems, or, more strictly, the non-perturbative SW curves correspond to the family of periodic Toda chains (Toda [1981]) and Calogero-Moser (Calogero [1969], Krichever [1980], Ruijsenaars and Schneider [1986] and Ruijsenaars [1987]) integrable models. 4.3 The Zoo of Curves and Integrable Systems Now let us turn to the question of the zoo of SW integrable systems, that is some classification of relations between the SUSY YM theories and corresponding integrable models. We shall consider the cases of broken A/* = 4 SUSY theories, or J\f = 2 SUSY Yang-Mills with extra adjoint matter multiplets, theories with soft Kaluza-Klein (KK) modes or with an extra 5th compact dimension and, as a separate question, theories with fundamental matter (all for a SU(N) gauge group). The last case has been less investigated yet and there still exist some open problems, mostly related with the "conformal" case Nf = IN. The first two classes can be formulated in a uniform manner since there exists a "unifying" integrable system, the elliptic Ruijsenaars-Schneider model (Ruijsenaars and Schneider [1986] and Ruijsenaars [1987]) (with bare coupling r, "relativistic" parameter R—the radius of compact dimension and an extra parameter e — see (11)), and this gives rise to all known models of these two classes in its various degenerations (Braden et ah [1999]): • If /? ^- 0 (with finite e) the mass spectrum (11) reduces to a single point M = 0, i.e. all masses arise only due to the Higgs effect. This is the standard four dimensional M = 2 SUSY YM model associated with the periodic Toda chain. In this situation J\f = 2 SUSY in four dimensions is insufficient to ensure UV-finiteness, thus bare coupling diverges r -^ /00, but the dimensional transmutation substitutes the dimensionless r by the new dimensionful (andfinite) parameter A^ = e^^^^(e/R)^.
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 77 If R ^^ 0 and e ^ mR fox finite m, then UV finiteness is preserved. The mass spectrum reduces to the two points M = 0 and M = m. This is the four dimensional YM model with A/* = 4 SUSY softly broken to J\f = 2. The associated finite-dimensional integrable system (Donagi and Witten [1996]) is the elliptic Calogero-Moser model (Calogero [1969] and Krichever [1980]). The previous case is then obtained by Inosemtsev's [1989] double scaling limit when m ^- 0, r ^- /oo and A^ = m^e^^^^ is fixed. If /? 7^ 0 but 6 ^- /oo the mass spectrum reduces to a single Kaluza-Klein tower, M = Ttn/R, n e Z. This compactification of the five dimensional model has N = I SUSY and is not UV-finite. Here z ^^ ioo and e -^ ioo, such that Inz — Ne remains finite. The corresponding integrable system (Nekrasov [1998]) is the relativistic Toda chain (Ruijsenaars [1990]). Finally, when R ^ 0 and e and r are both finite one distinguished case still remains: 6 = 7r/2.'^Here only periodic and antiperiodic boundary conditions occur in the compact dimension. This is the case analyzed in Braden et al [1998, 1999] and interested reader can find all details there. 4.3,1 Toda chain This is the most well-known example of SU(N) pure gauge theory (see Klemn etal [1995]). The spectral curve is u) (34) Pn(X) = det(X -^) = X^ -J2^k^^ and the generating 1-form dS = X^ satisfies <^moduli"*^ — <^moduli"*^ lu;=const ^ v'^moduli'^) W (35) J2 ^^^Uk dw ^c-A X^dX ; = > 8uk = holomorphic where dPN dx The spectral curve (34) corresponds to the periodic A^-particle Toda chain (Date and Tanaka [1976] and Krichever [1978]). Let us now recall how this can be derived via the language of integrable systems. We will use two different forms of the Lax representation with spectral parameter for the periodic Toda chain. "^The case e = 0 of fully unbroken five dimensional N = 2 supersymmetry is of course also distinguished, but trivial: there is no evolution of effective couplings (renormalisation group flows) and the integrable system is just that of A^ non-interacting (free) particles.
78 A. MARSHAKOV The Toda chain system (Toda [1981], see also Ueno and Takashi [1984]) is a system of particles with nearest neighbour, pairwise exponential interactions. The equations of motion are^ ^-^=Pi ^ = e^'-^-^' - e^'-^'-\ (37) dt ^ dt ^ where one assumes (for the periodic problem with the "period" A^) that qi^^ = qt and Pi_^j^ = pi. It is an integrable system, with A^ Poisson-commuting Hamiltonians, hi = ^ Pi = P, /i2 = ^{^pj + e^'~^'-^) = E, etc. Starting naively from an infinite-dimensional system of particles (37), the periodic problem can be formulated in terms of (the eigenvalues and the eigenfunctions of) two commuting operators: the Lax operator C (or the auxiliary linear problem for (37)^) and a second operator T. In the case of periodic boundary conditions T is chosen to be a monodromy or shift operator in a discrete variable (the number of a particle): Tqn = qn^-N Tpn = Pn-{-N T^n = ^n-{-N - (40) The existence of common spectrum of these two operators^ C^ = X^ T^ = w^ [£, r] = 0 (41) means that there is a relation between them V(C, T) = 0. This can be expressed in terms of a spectral curve (see Krichever [1977]) S: V(X, w) = 0. Here there is an important difference with the perturbative case considered above, where only one (Lax) operator was really defined and there was no analogue of the second T-operator. The generating function for the integrals of motion, the Toda chain Hamiltonians, can be written in terms of both the C and T operators and Toda chains possess two different (though, of course, equivalent) formulations of this kind. ^For simplicity in this section we consider a periodic Toda chain with coupling constant equal to unity. It is easy to restore it in all the equations; then it becomes clear that it should be identified with A = Aqcd - the scale parameter of pure M = 2 SUSY Yang-Mills theory. ^Equation (39) is a second-order difference equation and it has two independent solutions which we shall often denote below as ^"^ and ^~. In more general framework of the Toda lattice hierarchy these two solutions correspond to the two possible choices of sign in the time-dependent form of the Lax equation (39) ^Let us point out that we consider a periodic problem for the Toda chain when only the BA function can acquire a nontrivial factor under the action of the shift operator while the coordinates and momenta themselves are periodic. The ^M«5/periodicity of coordinates and momenta — when they acquire a nonzero shift — corresponds to a change of the coupling constant in the Toda chain Hamiltonians.
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 79 In this first description the Lax operator (39) when re-written in the basis of the T-operator eigenfunctions and becomes ih^ N x N matrix, simple a -ao + ive ( e^ P\ {qi-qx) 0 1 ^^ ^2 \Le\^q\-qN) q 0 (qs-qi) P3 0 0 0 (42) PN I It explicitly depends on the spectral parameter w, the eigenvalue of the shift operator (40), and is defined on a cylinder. The matrix (42) is almost tri-diagonal, the only extra nonzero elements appear in the off-diagonal comers are due to the periodic conditions (40) which in this naive way reduce the infinite-dimensional constant matrix (39) to a finite-dimensional one, but depending on spectral parameter w. The eigenvalues of the Lax operator (42) are defined from the spectral equation P(X, w) = det (C^^iw) -X)=0. NxN Substituting the explicit expression (42) into (43), one gets: V(X, w) 1 w-\- Pn(X) w 0 (43) (44) i.e. eq. (34), where Pn(^) is a polynomial of degree A^, with the mutually Poisson- commuting coefficients: p^(X) =X^ -^ hiX^-^ + ^(/i2 - hl)X^-^ + ... (45) (hk = J2iLi pf + " ')y parameterizing (a subspace) in the moduli space of the complex structures of the hyperelliptic curves S^^ of genus N — I = rank SU(N). An alternative description of the same system arises when one (before imposing the periodic conditions!) solves explicitly the auxiliary linear problem (39) which is a second-order difference equation. To solve it one rewrites (39) as ^l,i^, = (X - pi)^i - e^ ■^/-i (46) or, since the space of solutions is 2-dimensional^, it can be rewritten as ^z+i = L/(X)^/ where ^/ is a set of two-vectors and L/ — a chain of 2 x 2 Lax matrices. After a simple gauge" transformation Li -^ f//+iL,f/. ^ where Ut = diag(^2^s e 2^'-') (and replacing be written in the f Pi -^ —Pi) these matrices can be written in the form (Faddeev and Takhtadjan [1986]) 'X + Pi ""' l,...,N (47) ^The initial condition for the recursion relation (46) consists of two arbitrary functions, say, ^j and ^2, which are, of course, linear combinations of ^+ and ^~ from (38).
80 A. MARSHAKOV The matrices (47) obey quadratic r-matrix Poisson bracket relations: the canonical Poisson brackets of the co-ordinates and momenta of the Toda chain particles [qi, Pj} — ^ij can be rewritten equivalently in the "commutator" form {Li(X)^LjiX^)] = 8ij [r(X - V), Li(X) 0 Ly(X')] . (48) Here the (/-independent!) numerical rational r-matrix r(X) = ^ Xla=i ^a^cr^ satisfies the classical Yang-Baxter equation. As a consequence, the transfer matrix N>i>l satisfies the same Poisson-bracket relation {TN(X)nN(^')} = [r(X - XO, Tn(X) 0 Tn(X')] (50) and the integrals of motion of the Toda chain are generated by another representation of the spectral equation det (Tn(X) -w) = w^ - wTyTn(X) + det Tn(X) 2x2 = w^ -wTyTn(X)-\-1 =0. (51) Here we have that det2x2 L(X) = 1 leading to det2x2 T]si(X) = 1. Thus V(X, w) = w-\- TyTn(X) = w-\- Pn(X) = 0, (52) w w which coincides with (44). The polynomial PNiX) = TyTn(X) in (52) is of degree A^, its coefficients are the integrals of motion since {TrTM(X\ TrlMiX')} = Tr {Tn(X)nNi^')} = Tr [r(X - V), Tn(X) 0 Tn(X')] = 0. (53) Finally, let us note that the Toda chain N x N Lax operator (42) can be brought by gauge transformation to another familiar form y simple a J \ simple a 0 |;^5(^3-^2) ^3 0 \le-2^q^-qN) 0 0 PN / Uij = v'Sij w = v^. (54)
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 81 Formally this corresponds to change of gradation of the Toda chain Lax operator. The form (54) is especially natural and relates the Toda chain Lax operator with the sl(N) Kac-Moody algebra. In the form (54) the periodic Toda chain can be thought of as a special "double-scaling" limit of the SL(N) Hitchin system on a torus with a marked point — the Calogero-Moser model (see below). It is clear that the Lax operator (54) satisfies the 9-equation (24) on a cylinder with trivial gauge connection (see Marshakov [1999]). simple a I ' . (55) e-"°*£„„+ Y. «"*£-„ j 5(Poo) simple OL J and this can be easily solved giving rise to (54). 432 Broken N=^4 SUSY and the Elliptic Calogero-Moser Model Now let us turn to the observation that ih^ N x N matrix Lax operator (42) can be thought of as a "degenerate" case of the Lax operator for the A^-particle Calogero-Moser system (see Krichever [1980]). C'^^(z) = h H + J^Fiq a\z)EA = Pi F(qi-q2\z) ... F(qi-qN\z)\ (56) F(q2-qi\z) P2 ... F(q2-qN\z) .F(qN-qi\z) F(qN-q2\z) ... Pn The matrix elements F(q\z) = ^aiq)aiz)^^^^^^ ^^ expressed in terms of the Weierstrass sigma-functions so that the Lax operator C{z) is defined on an elliptic curve E{x) y^ = {x - e\){x - e2)(x - ^3), 1 , (57) x = p(z) y = :^p(z), or a complex torus with modulus r and one marked point z = 0,jc = 00, y = 00. The Lax operator (56) corresponds to a completely integrable system with Hamiltonians h\ = Yli Pi = P^ ^2 = Zl, p} + ^^ Hi<j P(^i ~ ^7)' ^t^- The coupling constant m in the 4 dimensional interpretation plays the role of the mass of adjoint matter J\f = 2 hypermultiplet breaking J\f = 4 SUSY down ioAf = 2 (see Donagi and Witten [1996]). From the Lax representation (56) it follows that the spectral curve S ^^ for the A/^-particle Calogero-Moser system det (C^^(z) -X)=0 (58) NxN
82 A. MARSHAKOV covers N times the elliptic curve (57) with canonical holomorphic 1-differential dz = ^. (59) The SW BPS masses a and sld are now the periods of the generating 1-differential dS^^ = 2Xdz = X— (60) y along the non-contractable contours on S^^^. In order to recover the Toda-chain system, one has to take the double-scaling limit (see Inozemtsev [1989]), when m and —it both go to infinity and qt -qj ^ 2 [^^ ~ -^^ ^^^^ "^ ^^' ~ ^J^^ ^^^^ such that the dimensionless coupling r gets substituted by a dimensionful parameter A"^ ~ m^e^^'^. The idea is to separate the pairwise interacting particles far away from each other and to adjust the coupling constant simultaneously in such a way, that only the interaction of neighbouring particles survives (and turns out to be exponential). In this limit, the elliptic curve degenerates into a cylinder with co-ordinate w = e^e^^'^ so that ^5^^ -> dS^"^ = k—. (62) w The Lax operator of the Calogero system turns into that of the A^-periodic Toda chain (42): C^^iz)dz^C^'^iw)— (63) w and the spectral curve acquires the form (43). In the simplest example ofN = 2 the spectral curve S^^ has genus 2. Indeed, in this particular case, Eq. (58) turns into V(X;x,y) = X^ -\-u-m^x =0. (64) This equation says that with any value of jc one associates two points of S^^ X = ±iy/u — mP-x (65) i.e. it describes S^^ as a double covering of an elliptic curve ramified at the points x = -^ and JC = 00. In fact, x = -^ corresponds to a pair of points on E(t) distinguished by the sign of y. This would be true for jc = oo as well, but x = oo is one of the branch points, thus, two cuts between x = -^ and jc = oo on every sheet of E(z) touching at the common end at JC = 00 become effectively a single cut between (-^, +) and (-%,—). Therefore, we can consider the spectral curve S^^ as two tori E(t) glued along one cut, i.e. I1^^2 ^^^ genus 2. It turns out to be a hyperelliptic curve (for N = 2 only!) after substituting in (64) JC from the second equation to the first one. ^Let us point out that the curve (58) has genus g = N (while in general the genus of the curve defined by N X N matrix grows as A^^). However, the integrable system is still 2(A^ — 1)-dimensional since the sum of the periods of (60) vanishes due to specific properties of Xdz and there are only A^ — 1 = g — I independent integrals of motion.
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 83 Two holomorphic 1-differentials on S^^ (^ = A^ = 2) can be chosen to be dx XdX dz dx dX dv^ =dz = — = , dv- = — = —- = —, (66) 2y y A Lyk y so that dx dx and dS = 2Xdz = X— = —y/u -m^jc, (67) y y ddS ^ dx --- = —- =dv-. (68) au 2yX The fact that only one of two holomorphic 1-differentials (66) appears on the right hand side of (68) is related to their different parity with respect to the Z2 (8) Z2 symmetry of S ^^: y -^ —y,X -^ —Xanddv± -^ ibJi;±. Since J5 has a definite (positive) parity, its integrals along two of the four elementary non-contractable cycles on S^^ automatically vanish leaving only two non-vanishing quantities a and ao, as necessary for the 4 dimensional interpretation. Moreover, the two remaining nonzero periods can be defined in terms of the "reduced" curve of genus ^ = 1 3 y2 = -(yXf = {u- m^x) Y\(x - ea), (69) a=l equipped with J5 = (u — m^x) ^^^. Since for this curve jc = 00 is no longer a ramification point, dS has simple poles when x = 00 (on both sheets of 5]^^^^^^) with the residues ±m. The opposite limit of the Calogero-Moser system with vanishing coupling constant m^ ^- 0 corresponds to the J\f = 4 SUSY Yang-Mills theory with identically vanishing yS-function. The corresponding integrable system is a collection of free particles and the generating differential dS = y/u • dz is just a holomorphic differential on E(t). 4.3.3 Relativistic Toda Chain and the Ruijsenaars-Schneider Model So far we have considered theories with only Higgs and adjoint matter contributions to the spectrum (11). If, however, one also adds soft KK modes (Nekrasov [1998]), the resulting integrable systems would correspond to the Ruijsenaars-Schneider family (Ruijsenaars and Schneider [1986], Ruijsenaars [1987, 1990]), which is often described as a relativistic integrable model. The 1-loop contributions to the effective charge (5) are now of the form Tij - ^ log (aij + -^ j ^ log ]^ [R^aij + m) - log sinh {Rsaij) (71) ^^This curve can be obtained by simple integration of the equations of motion since we again here deal with the only degree of freedom in our integrable system. The concerning energy u = p^ +m^p{q) gives J Jh -m^piq) J J{x y/h -m^piq) J y/{x - ei)ix - e2)(x - e^){u - m^x) i.e. exactly the Abel map on the reduced curve (69) (70)
84 A. MARSHAKOV i.e. in going from 4 dimensions up to 5 dimensions one should make a trigonometric substitution a -^ sinhfl/?5, at least, in the formulas for the perturbative prepotential. A similar change of variables corresponds to relativization of integrable systems, which implies a sort of "Lie group generalization" of the "ordinary" integrable systems related rather to Lie algebras. The relativization of an integrable system replaces the momenta of particles by their exponentials and it results in period matrices or effective charges of the form (71). For example, in the case of SU(2) pure gauge theory it gives instead of Hamiltonian of the Toda chain (12) cosh z = cosh p — h. (72) The net result in the case of relativistic Toda chain is that the spectral curve is a minor modification of (52), + - = (A5X)-^/2p(X), u; + - = (A5X)-^^/^P(X), (73) w which can be again rewritten as a hyperelliptic curve in terms of the new variable Y^ = P^{X)-AaI^X^ . (74) Here X = fu? = e^^, where § can be chosen as a spectral parameter of the relativistic Toda chain and A 5 is its coupling constant. The most general picture in these terms corresponds to the "unifying" elliptic Ruijsenaars- Schneider model (Nekrasov [1998] and Braden et al [1999]). The Lax operator for the elliptic Ruijsenaars model is (Ruijsenaars [1987]) 'J a{qij+e)a{z) e^' = eP' Y[VP(^)-P(m)^ (75) and in the trigonometric limit it turns into ^TR _ ^p, sinh(^,^ + z) sinh(6) "'^ sinh(^/^ + e) sinh(z) ^^' = eP^ sinh^6 k^i T sinh2(^,^) n> (76) Introducing v, = e^*, f = e^z ^ud q = e^^ one finds that '■' qvi — Vj 1 — ^ f^oo "' 'q — Ht) p, _ n, T-T ^/qvi-vj^/vi-qvj ki-i (77)
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 85 Only the leading term in (77) is usually taken as an expression for the Lax operator of trigonometric Ruijsenaars system. The elliptic Ruijsenaars spectral curve acquires the form (Ruijsenaars [1987] and Braden et al. [1999]) det(X-£^a)) = ^(-X)^- where^^ k=0 E(-^) N-k k=0 ^^^^1^^ = ^-^ ' a^(z)a^(^-i)(.) ■ (Do(z\e) = 1 and Di(z\e) = 1.) The generating differential is dS = log Xdz. In the simplest case of 1 degree of freedom (A^ = 2) Eq. (78) reads Dk(z\e)Hk=0 (78) (79) (80) X^-uX + D2(z\e) = X^-uX- ^(^ .Z?"^,^^!^^ = X^-uX-\-p(z) - p(e) =0 (81) cr'^(z)(y'^(e) and one gets a spectral curve identical to (64). The same sort of arguments as for the elliptic Calogero-Moser model shows that its full genus is ^ = 2 while the SL(2) integrable system lives effectively on a reduced spectral curve of genus one. In the trigonometric limit (76), (77) the spectral curve (78) turns into (10). The interested reader can find the details concerning various limits of the Ruijsenaars-Schneider model in Braden et al [1999]. 4.3.4 Fundamental Matter Let us finally turn to the case of SUSY QCD, those SW theories with matter multiplets in the lowest dimensional representations of the gauge group. According to Seiberg and Witten [1994b] and Hanany and Oz [1995] the spectral curves for J\f = 2 SQCD with gauge group SU(N) and any number of matter multiplets Nf < 2N have the form / = p2(X)-4A2^-^/P;v,(A) /=1 Pm(X) ^ pI^\x) + Rn-i(X) ^ Y\(X - Xi(<t>, m)), P^^(X)^Y[(X-ma). (82) a=l ''in Braden et al. [1999] we have used for technical reasons a slightly different form of the Ruijsenaars spectral curve, which can be obtained from Eq, (78) substituting X -^ Ac(z) (or D^(z|e) -^ D^(z|e)c*^(z)) with some (independent of Hamiltonians and e) function c{z). Such substitution does not change the "symplectic form" ^ Adz and utilising this function some particular formulas can be brought to more simple form.
86 A. MARSHAKOV Here P^^ (X) and Rn-i(X) are "moduli independent" polynomials of X (i.e. they depend only on "external" masses of matter hypermultiplets (see Hanany and Oz [1995]) and the spectral curves are still described by hyperelliptic equation. In contrast to a pure gauge theory one immediately runs into a problem of parameter counting: there are A^ — 1 moduli ^Tr4>^ together with Nf masses rua give N -\- Nf — I parameters and this is more than the number of hyperelliptic moduli 2g — 2 = 2N — 4 of the equation (82) for Nf > N — 3. In other words, the moduli space of Eq. (82) is too small to be parameterized by all the parameters of the theory. That is why there are some complications in retranslating these models into the language of integrable systems^^. However, the form (82) immediately implies (see Marshakov [1996]) that for A/^ = 2 SQCD there should exist a 2 x 2 representation of the kind (51) with monodromy matrix r/v(X), whose invariants have the form Tr Tm(X) = Pn(^), dtiTN(X) = PM,(X), A natural proposal then is (see Gorsky [1996]) to use a generalization of the 2x2 Toda chain Lax representation, i.e. to deform the Lax matrix (47) preserving the Poisson brackets (48) and (50). The monodromy matrix T(X) is again constructed by multiplication of L,(X)'s (satisfying (48)) at different sites giving rise to the spectral curve equation dQi(TN(X)-w) = 0 (84) where from T-matrix one has to require (83). A wide class of such systems is given by the family of spin chains. To get the most general form of the transfer matrix one should consider the inhomogeneous spin chain ^^ Tn(X)= n ^i(^-^i) (85) N>i>l If the Lax matrices L, (X) were of size p x p, equation (84) would have the form of p-ih degree polynomial in w. When /? = 2 it is exactly the general case of (82)^"^ w + ^^^ = TxTn(X), (86) ^^In particular, there are different ideas of interpretation of the equation (82) (at least in the case of small Nf < N) in the context of integrable models. Some of these can be found in Marshakov [1996], Ahn and Nam [1996] and Krichever and Phong [1997]. We shall consider below, following (see Gorsky et al [1996a] and Gorsky et al. [1996]) only one such option, which seems at present to be the most attractive, since it leads to a family of integrable systems which admits, for examples in the case of adjoint matter, inclusion of a KK sector, and natural relativistic generalizations. ^^For the Toda chain the inhomogeneity parameters /, are not independent variables — they can be reabsorbed into the definition of momenta. ^"^We introduce a new variable W = . ^ in order to make analytic representation of the spectral curve more symmetric. In the picture of M-theory these redefinitions are related with different brane pictures of the theories where presence of fundamental matter multiplets are caused by extra semi-infinite branes.
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 87 or W-\- — = , ^' ^ = /' . (87) w vdetryvw y?;;^ As in the Toda chain case, TyTn (A.) has degree A^ but now det Tn (A.) is not longer unity but some general polynomial of degree Nf < 2N. The generating 1-form, according to general rules, becomes dS = X^, (88) Vr The right hand side of the equations (86), (87) contain the dynamical variables of the spin system only in special combinations of the Hamiltonians (which are all in involution or mutually Poisson-commuting) and the inhomogeneities which commute with all dynamical variables. The 2x2 Lax matrix for the simplest 5"/(2) rational XXX chain is 3 L(X) = X-l-\-J2Sa ■cr'' (89) a=l and the Poisson brackets on the space of the dynamical variables Sa,a = 1, 2, 3 are implied by quadratic r-matrix relations (48) with the same rational r-matrix as for the Toda chain (see Sklyanin [1989]). The r-matrix relations (48) for the Lax matrix (89) are equivalent to the well-known 5"/(2) commutation (Poisson bracket) relations {Sa,Sb} = ieabcSc; (90) the vector [Sa) plays the role of angular momentum ("classical spin") variables giving the name "spin-chains" to the whole class of systems. The algebra (90) has an obvious Casimir operator (an invariant Poisson commuting with all generators Sa), 3 K^=S^ = J2SaSa. (91) Thus a=l dciL(X) = X^ - K^, 2x2 dtiTN(X)= n dti Li(X-li)= n {(X-lif-Kf) 2x2 ^ ^ 2x2 1 1 V ' ^ /QO) 1</<7V l<i<N ^^ ^ = n i^ + rn^K^ + mr). l<i<N We have assumed that the values of spin K can be different at different nodes of the chain, and set^^ mf = -liTKi. (93) ^^Eq. (93) implies that the Hmit of vanishing masses (all mf = 0) is associated with the homogeneous chain (all /, = 0) and vanishing spins at each site (all Ki = 0).
88 A. MARSHAKOV The determinant of the monodromy matrix (92) depends on the dynamical variables only through the Casimirs Kt of the Poisson algebra, and the trace Pn(^) = \^^2x2Tn(^) generates the Hamiltonians or integrals of motion. In the case of 5"/(2) XXX spin chains, the spectral equation acquires precisely the form (86) or (87) where the number of matter multiplets, Nf < IN (which determines the degree of polynomials in (83)) depends on the particular "degeneracy" of the full chain. In this picture the rational si (2) XXX spin chain literally corresponds to Nf < 2N J\f = 2 SUSY QCD. Things are not so simple in the "conformal" Nf = IN case when an additional dimensionless parameter appears. A naive toric generalization of the XXX-picture leads to the XyZ chain with Hamiltonian structure given by the elliptic Sklyanin algebra (see Sklyanin [1989]). This model was proposed in Gorsky et al. [1996] as a candidate for the integrable system behind the Nf = 2N theory, and, as it was discussed later (see, for example, Marshakov and Mironov [1998] and references therein), it should rather be interpreted as a 4 dimensional theory with two extra compact dimensions (or a compactified 6 dimensional theory).^^ The condition Nf = 2N which can be easily broken in 4 and 5 dimensions situations is very strict in 6 dimensions and one may think of the corresponding theory as of blowing up all possible compactified dimensions in the Nf = 2N 4 dimensional theory with vanishing )^-function. In general, one finds, that the spectral curves for SUSY QCD with Nf fundamental multiplets (and prepotentials) in the 4, 5 and 6 dimensional cases are described by similar formulae (see Marshakov and Mironov [1998]). The spectral curves in all cases can be written in the form w + -—^ = 2P^^H§), (95) w or 1 2P(^>(§) w W + —= -7===, W= . (96) The generating differentials are dS = ^dlogW (97) and the perturbative part of the prepotential is always of the form ^^Indeed, in the 6 dimensional theory with two extra compactified dimensions of radii R^ and R^ one should naively expect m,n ~ log ]~[ (^«5au + m + «^^ ~ loge (RiOij Uj-) (94) i.e. coming from 4 dimensions or 5 dimensions to 6 dimensions one should replace the rational (trigonometric) expressions by the elliptic functions, at least, in the formulas for the perturbative prepotential, the (imaginary part of) modular parameter being identified with the ratio of the compactification radii R5/Re-
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 89 TTie functions /^^^ introduced here are defined via: 2(4)(^) ^ Yl(^ _ ^^)^ 0(5)(^) - ]^sinh(§ - mj a a TV TV p(4) ^ Y[{^ _ ^.), p(5) ^ Yl sinhd - a,) (99) >(6) T-nr6'*(^ -a,) (100) 0*(§-^,) (in P*^*(§), there is also some exponential of § unless iV/ = 2N) /W(x) = x2logx m,n ^ ^ ^ "^ n ^ "^ = ( - |x^| - -Li3,^ (^~^'^') + quadratic terms J (101) so that /^^^'' = logJc /^^^''(jc) = logsinhjc f^^^\x) = logO^(x), (102) Note that in the 6 dimensional case, we have always Nf = IN. The variables §, above are inhomogeneities of the integrable system, and, in 5 and 6 dimensions there is the restriction that ^at =YlHi = \Yl^oi^ which implies that gauge moduli should rather be associated with at shifted by the constant 577 XI ^«- 4.4 Conclusion In this chapter I have tried to formulate the main issues in the correspondence between integrable systems and the Seiberg-Witten curves governing exact solutions to A/* = 2 SUSY gauge theories. During the last 4 years there has been much progress in this area. However, there are still many questions which are not yet understood and which deserve further investigation. Let me point out at least some of them: • The first, and the most essential question is still open: how to derive the SW/integrable systems correspondence from first principles. The significance of the integrable system's dynamics in terms of SUSY gauge theory remains unclear.
90 A. MARSHAKOV • It does appear clear that the SW construction itself is much more transparent from the point of view of non-perturbative string theory or M-theory. However, there are still many open questions in this direction. For example, what is the M-theory picture of the Ruijsenaars-Schneider model and double-elliptic systems, etc. It is clear that the Ruijsenaars Lax operator (75) satisfies an equation similar to the 9-equation (24), but there is no clear interpretation of this equation in terms of D-brane constructions. • I have already mentioned some general problems with fundamental matter. Of particular interest is the conformal Nf = IN case, which should be clearly related, on one hand, with double-elliptic systems and, on the other hand, with the K3 and elliptic Calabi-Yau compactifications of string and M-theory. • In this chapter only the SU(N) gauge theories were considered in detail. I did not touch at all on the many problems related to the generalizations to the other gauge groups and representations. From the point of view of integrable systems this is a question, in part, about different representations of the Lax pairs for integrable systems and overlaps, for example, with the problem of different Lax pairs for the Calogero-Moser systems found recently in Bordner et al [1998] and D'Hoker and Phong [1998]. Acknowledgements I am grateful to H.W. Braden, I. Krichever, A. Mironov and A. Morozov for illuminating discussions and to the organisers of the conference for the nice time in Edinburgh. The work was partially supported by the RFBR grant 98-01-00344 and the INTAS grant 96-482. References A. Hanany, and Oz, Y, Nucl Phys. B452, 283, hep-th/ 9505075 (1995). Ahn, C, and Nam, S., Phys. Lett B387, 304, hep-th/ 9603028 (1996). Bogoyavlensky, O., Toppling Solitons. Nonlinear Integrable Equations, Moscow, Nauka, 1981. Bordner, A., Corrigan, E., and Sasaki, R., Prog. Theor. Phys. 100, 1107 (1998), hep-th/9805106. Braden, H., Marshakov, A., Mironov, A., and Morozov, A., Nucl. Phys. B558, 371-390 (1999b), hep-th/9902205. Braden, H., Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B448, 195-202 (1999a), hepth/9812078. Calogero, R, J. Math. Phys. 10, 2191, 2197 (1969). D'Hoker, E., and D. Phong, Nucl. Phys. B530, 537 (1998), hep-th/9804124. Date, E., and Tanaka, S., Prog. Theor. Phys. 55, 457 (1976). Diaconescu, D.-E., Nucl. Phys. B503, 220 (1997), hep-th/ 9608163. Donagi, R., and Witten, E., Nucl. Phys. B460, 299, hep-th/9510101 (1996). Dubrovin, B., Krichever, I., and Novikov, S., Integrable systems — /, Sovremennye problemy matematiki (VINITI), Dynamical systems - 4, 179 (1985). Faddeev, L., and Takhtadjan, L., Hamiltonian Approach to the Theory of Solitons, Moscow, Nauka, 1986. Gorsky, A., and Nekrasov, N., hep-th/ 9401021 (1994). Gorsky, A., Krichever, I., Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B355, 466, hep-th/9505035 (1995).
SEIBERG-WITTEN CURVES AND INTEGRABLE SYSTEMS 91 Gorsky, A., Marshakov, A., Mironov, A., and Morozov, A., in Problems in Modem Theoretical Physics, Dubna 1996, 44, hep-th/ 9604078. Gorsky, A., Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B380, 75, hep-th/ 9603140 (1996). Hitchin, N., Duke. Math. Journ. 54, 91 (1987). Inozemtsev, v., Comm. Math. Phys. 121, 629 (1989). Klemm, A., Lerche, W., Theisen, S., and Yankielowicz, S., Phys. Lett. 344B, 169; (1995) hep-th/ 9411048; hep-th/9412158. Krichever, L, and Phong, D., J. Diff. Geom. 45 349, hep-th/ 9604199 (1997). Krichever, L, Func. Anal. & Appl. 14, 282 (1980). Krichever, I., Func. Anal. & Apps. 11, 15 (1977). Krichever, I., Soviet. Math. Surveys 33, N4 215 (1978). Marshakov, A., and Mironov, A., Nucl. Phys. B518 59, hep-th/ 9711156 (1998). Marshakov, A., in New Developments in Quantum Field Theory, Plenum Press 1999, NATO ASI Series B: Physics 366 (Eds. PH. Damgaard, and J. Jurkiewicz), 279; hep-th/ 9709001. Marshakov, A., in Proceedings of 10th International Conference "Problems of Quantum Field Theory", Dubna 1996, hep-th/ 9607159. Marshakov, A., MartelUni, M., and Morozov, A., Phys. Lett. B418 294; hep-th/ 9706050 (1998). Marshakov, A., Mod. Phys. Lett. All, 1169; hep-th/ 9602005 (1996). Martinec, E., and Warner, N., Nucl. Phys. 459, 97-112, hep-th/9509161 (1996). Nekrasov, N., Nucl. Phys. B531, 323, hepth/9609219 (1998). Ruijsenaars, S., and Schneider, U.,Ann. Phys. (NY.) 170, 370 (1986). Ruijsenaars, S., Comm. Math. Phys. 110 191 (1987). Ruijsenaars, S., Comm. Math. Phys. 133, 217 (1990). Seiberg, N., and Witten, E., hep-th/9607163 (1996). Seiberg, N., and Witten, E., Nucl. Phys. B426,19, hep-th/9407087 (1994a). Seiberg, N., and Witten, E., Nucl. Phys. B431 484, hep-th/9408099 (1994b). Sklyanin, E., J. Sov Math. 47 2473 (1989). Toda, M., Theory of Nonlinear Lattices, Springer-Verlag, 1981. Ueno, K., and Takasaki, K., Adv. Studies in Pure Math. 4, 1 (1984). Witten, E., Nucl. Phys. B500, 3^2 (1997), hep-th/ 9703166.
5 Integrability in Seiberg-Witten Theory A. MOROZOV 777259, ITEP, Moscow, Russia A review is given of the theory of effective actions and the hypothetical origins of integrabihty in Seiberg-Witten theory. This conference has emphasised the remarkable links between the recently discovered Seiberg-Witten ansatz (Seiberg et al [1994]) for the low-energy effective action of N = 2 supersymmetric Yang-Mills theories in 4 and 5 space-time dimensions and the old theory of classical (0+l)-dimensional integrable systems, based on the notions of Lax representations and spectral Riemann surfaces. The basic question about this whole subject is why integrability properties should arise at all in the study of high-dimensional models, and it remains obscure and not even addressed in most publications. Still, in Gorsky et al [1995] integrability has been deliberately searched for — and found — as the hidden structure behind the Seiberg-Witten ansatz. It is the goal of this brief presentation to collect the implicit arguments in favour of the a priori existence of a hidden integrable structure. These arguments are non-convincing and somewhat non-constructive, yet they deserve being kept in mind in the study of Seiberg-Witten theory and its predecessors (like matrix models) and possible generalizations. 5.1 Why Integrability? The question about the origins of integrability in Seiberg-Witten theory is in fact a part of broader question: what is the role of integrability in physics? Different people would give 93
94 A. MOROZOV different answers to this question, and most of them would choose between the following options: • Integrable models provide funny examples. • These exactly solvable examples can be (and actually are) used as the starting points for perturbative expansions describing approximately the really relevant physical models. • Whatever is solvable is integrable, i.e. if anybody succeeded in providing an exact solution to some physical model, there should be a (hidden) integrable structure in this model, responsible for the very possibility to solve it. This a posteriori integrability principle, while certainly true, is not very interesting: we would prefer to know about the existence of integrable structures in advance. • Integrability theory is rather a branch of mathematics (group theory), and can be of little physical significance. What we are going to advocate in the present notes is a less obvious hypothesis'. integrability is the general property of effective actions. However, in order to understand this hypothesis it is important to have an adequately adjusted (and broad enough) notion of integrability. 5.2. What is Integrability? (Morozov [1992,1994]) Effective actions are functional integrals like ZU|0o}^ I Dcl>e'^'\^K (1) These quantities depend on two types of variables: on the shape of the "bare action" 5(0) and on the boundary conditions 0o for the integration variables (fields) 0. In "physical language" these are (respectively) dependencies on the "coupling constants" t and on the choice of "vacuum" 0o. Obviously, being a result of integration, an (exponentiated) effective action Z exhibits peculiar features, unfamiliar from the studies of other classes of objects in quantum field theory and mathematical physics. Identification of such features should be the key to characterization of the new class of special functions, which will be used to represent the answers to any questions in non-perturbative quantum field (string) theory. Among those already discovered, the most important are (Morozov [1994]): (i) covariance of Z under variation of the coupling constants (the shape of 5(0)), reflecting invariance of an integral under a change of integration variables (fields) 0; and (ii) certain relations between the dependencies of Z on the shape 5(0) and on the choice of the vacuum (boundary conditions) 0o, reflecting dependence of vacuum states on the shape of the action. The first feature turns out to imply that effective actions exhibit a (hidden) integrable structure and hypothetically effective actions are (always?) the (generalized) r-functions of integrable hierarchies. So far, we have a restricted set of examples where this hypothesis is explicitly formulated and checked:
INTEGRABILITY IN SEIBERG-WITTEN THEORY 95 • Ordinary matrix models. Here one can straightforwardly study the r-dependence of matrix integrals and find the links to the simplest KP/Toda-like r-functions. • Character formulas and Chem-Simons theory. Here the dependence on boundary conditions can be investigated and again the links with KP/Toda r-functions are easily established. This result can be less expected than the previous one, yet still it is true. • Generalized Kontsevich Model (GKM). The GKM theory allows us to study the interplay between the t- and 00-dependencies. • Seiberg-Witten conjecture and Donaldson N=4 SUSY 4d Yang-Mills model (the topological 4d Yang-Mills theory). In this context the quasiclassical (Whitham) r-functions enter the game. All these examples are distinguished by the property that the relevant r-functions belong to the well studied KP/Toda family (i.e. are associated with the simplest Kac-Moody algebra U{\)]^=\). New examples are likely to be discovered after further advances in the theory of geometric quantization, when the notion of r -functions for other groups becomes more familiar. 5.3 Ordinary Matrix Model (Ambjom et ah [1990], Gerasimov et al [1991], Mironov etal [1990], Itoyama etal [1991], Morozov [1992]) In this case the role of the functional integral is played by an ordinary integral over N x N Hermitean matrices 0: Z{t] = fd""'^ exp (f^tkTvcpn (2) Invariance of the integral under arbitrary change of integration variable 0 ^- /(0) leads to covariance of Z[t} under the change of its arguments: the coupling constants tk. Namely (Mironov et al [1990], Ambjom et al [1990], Itoyama et al [1991]), if 50 = 60"+^ then the action S((p) = Tr J^k ^^^^ changes by 8S = Tr6 J^k ^^^0^^"^ ^^^ — if 6 is a number, not a matrix - this variation can be reinterpreted as the result of the change of r-variables: 8tk-{-n = ^ktk. This implies a relation like Z[t] ^ Z{t -\- 8t]. Such a relation would imply that Z{t} is actually independent of r. In fact, one should also take into account the change of the measure dcp under the variation of 0, and an exact statement, expressing the covariance of ZU), is: LnZ = 0 forn > -1, This covariance still implies that Z{r} is essentially independent of its arguments tk, but a more precise statement is that a dependence exists, but is of a very peculiar form: Z[t} = ZKp{t\Go}. (4) i.e. Z[t} is a KP r-function at some special point Go of the universal Grassmannian (a particular group element of L^(l)^^i). This statement can be checked by explicit evaluation of the matrix integral (Gerasimov etaL[\99\\) in the formalism of orthogonal polynomials.
96 A. MOROZOV Thus, the dependence of an effective action on the coupling constants can be represented in terms of the t-functions. 5.4 Generalized t-Functions and Dependence on the Boundary Conditions The generalized t-function (Gerasimov et al. [1995]) is a generating function for all the matrix elements of a given group element G in a given representation R\ ZR{t\G) = {0\e''^'G\0)R (5) Here t^ are parameters of the generating function and J^ are generators of the Lie algebra under consideration. As already mentioned, the conventional KP/Toda-like t functions are associated with the free fermion realization of the simplest affine algebra U{\)]^=i\ J(z) = V^V^(z), S = f^-^di//. Here C is a Riemann surface (complex curve), which can be used instead of the group element G = expXlm,n ^mn'^m'^n in the definition of t for affine (1-loop) algebras. Also, in this case it is convenient to parametrize the set of f-variables by a single function A: TKP(A\C) = {cxpU AJhc (6) The standard form of the KP tau-fiinction is a particular case of this one, associated with The same KP t-function can alternatively be represented in terms of a topological f/ir^^-dimensional field theory: abelian Chem-Simons model (Morozov [1998b]) (this is a simple example of the "AdS/CFT-correspondence" (Maldacena et al [1998], Gubser et al [1998], Witten [1998])). Here TKP(A\C)= f VAoxpScs, (7) where (8) Scs == I Ad A -i-cb AA Jm JdM and M is the "filled Riemann surface C" — the 3 dimensional manifold with boundary C. This identity illustrates the important hypothesis: the boundary conditions dependence of the functional integral (that of the Chem-Simons theory in this particular example) can also be (like that of the coupling constants) represented in terms of t-functions. This may not be too surprising since the dependence on coupling constants and boundary conditions are in fact deeply interrelated.
INTEGRABILITY IN SEffiERG-WITTEN THEORY 97 5.5 Generalized Kontsevich Model (Kharchev et ah [1992,1993], Morozov [1994]) This interrelation can be illustrated with the example of the Generalized Kontsevich Model (GKM), defined as the matrix integral over n x n Hermitean matrices Z, Zgkm[L\Vp^x} = C j d""" ^^tr(-y^+i(X)+XL)^ ^9^ Here L is a n x n Hermitean matrix and Vp+i (jc) is a polynomial of degree /?+1, its derivative Wp{x) = y' J (jc) is a polynomial of degree p. In GKM theory one introduces two sets of "time-variables" to parametrize the dependence on L and on the shape of V^+i (jc): Tk^^TrA-^, f;t^^TrA-^ L = Wp(A) = AP, (10) and (Krichever [1992]) tk = ^^^^res^W;-^/^(/x)J/x. (11) In (9) C is a prefactor used to cancel the quasiclassical contribution to the integral around the saddle point X = A, C = exp (try^+i(A) - trAy;+i(A)) dci^^^ddVp^iiA). (12) The central result of the GKM theory is that (Kharchev et al [1993]) ZGKM{L\Vp+i} = e-''^^^^\'^\p{fk+tk) (13) where Zp is a t-function of the "/7-reduced KP hierarchy", and the shape of Zp as a function of time-variables (i.e. the associated group element of U{\)) depends only on degree p and not on the shape of the polynomial Wp{x). The exponential factor in (13) contains J'pifkVk) = -Y^Aij{t){fi+ti){fj+tj), ^ ij (14) Aij =reSoo W'^P(X)dwl^^(X) which is a "quasiclassical (Whitham) r-function". At first glance this GKM example does not seem very close to (1): neither of the two types of variables in (9) — the shape of V^+i and the matrix L — is obviously interpreted as a boundary condition or vacuum. Instead the whole expression (9) looks very much like a (matrix) Fourier transform, and the interrelation (13) between the L- and Vp+i-dependencies, which puts them essentially on equal footing, is not too surprising: usually the dependence of the Fourier transform (Z) on its argument (L) contains entire information about the shape of transformed function (e~^p+^^^^). However, the GKM example is not irrelevant to our considerations. As shown in Gorsky et al. [1998], the GKM partition function (9) can be a natural member (the oversimplified case, associated with the genus-zero spectral curve) of the family of the Seiberg-Witten prepotentials, which are clearly quantities of the type (1).
98 A. MOROZOV 5.6 The Low Energy Effective Actions Low energy effective actions describe the effective dynamics of light modes, arising after all the heavy modes have been integrated away. In the case of the Seiberg-Witten theory the light modes are abelian supermultiplets (their lowest components parametrize the valleys in the potential, i.e. the moduli space of vacua) and the Chem-Simons degree of freedom (not a field!) K = f d^xTr(AdA + f A^), peculiar to Yang-Mills theories in four dimensions with the property that the (non-perturbative) effective potential is always a periodic function of ^. It remains an unresolved problem to derive the low-energy effective action for the fluctuations along the valleys in interaction with the dynamics in ^-direction — which should be just a (0+l)-dimensional problem — directly from the non-perturbative N = 2 SUSY Yang-Mills theory. Still one can attempt to guess what the transition to the low-energy effective actions should mean from the point of view of the T-functions. Normally, in higher-dimensional field theories the functional integrals depend on an additional parameter: the normalization point /x, which is the IR cutoff in the integration over fluctuations with different momenta. Effective actions with a given /x describes the effective dynamics of excitations with wavelengths exceeding /x~^ The low-energy effective action arises in the limit /x -> 0, when only finite number of excitations (the zero-modes of the massless fields) remain relevant. Of course, different theories can possess the same low-energy action (e.g. in all the theories without massless particles there are no degrees of freedom, surviving in the low-energy approximation, and the low-energy effective action is just zero for all of them): these actions are pertinent for universality classes rather than for particular field theory models. As usual in the study of universality classes it is instructive to look for the simplest representative of the given class. The matrix integrals, which are (0+0)-dimensional models are such simple representatives of the class which also contains the 2d topological sigma-models interacting with 2d gravity. Similarly, the Seiberg-Witten conjecture can be interpreted so that N = 2 SUSY Yang-Mills models belong to universality class, of which the simplest representatives are the (0+l)-dimensional integrable systems. The natural hypothesis (suggested in Gorsky et al. [1995]) about interpretation of the low-energy effective actions in terms of the t-functions states that: • The low-energy effective actions are quasiclassical (Whitham) t-functions. • The proper coordinates (the flat structure) on the moduli space of vacua are provided by the adiabatic mvariauts {ac = ^^ pdq). • TYve lime-variables, associated with the low-energy correlators (i.e. renormalized coupling constants) are Whitham times (the deformations of symplectic structure). Seiberg-Witten theory provides us with two kinds of pieces of evidence in support of this hypothesis: • Dynamics of branes seems to imply the shapes of the spectral curves and their period matrices (Witten [1997], Marshakov et al [1998] and Kapustin et al [1998], Kapustin [1998]).
INTEGRABILITY IN SEIBERG-WITTEN THEORY 99 • The correlation functions in Donaldson theory seem to be indeed described in terms of the relevant Whitham fc^w-functions (Losev et al. [1998] and Gorsky et al [1998]). 5.7 Dynamics of 5-Branes (Gorsky [1997], Witten [1997], Marshakov et al [1998], Kapustin et al [1998], Kapustin [1998]). The fundamental object associated with the J = 11 supergravity — and thus, presumably, with the entire M-theory — is a membrane, i.e. a 2-brane. Its dual — which should be equally fundamental — is a 5-brane with a 6-dimensional world volume. In the first-quantized approach the dynamics of a 5-brane is described in terms of some 6d field theory, which — as a first choice - can be either a non-abelian super-Yang-Mills model or that of abelian self-dual 2-forms. If the topology of the world volume is /^"^ x C one gets a family of Ad Yang-Mills models, parametrized by the choice of vacua, labelled by Riemann surfaces C. The possible choice of C is restricted by the equations of motion of the 6d theory. According to Marshakov et al [1998], these equations for 6d super-Yang-Mills model, provide C in the form of the spectral curve, C: det(L($)-A) =0 (15) with the flat coordinates on the moduli space of such curves defined as c^c = fc^- ^^ ^^^ further considers the theory of abelian 2-forms on such curves, one immediately gets (Witten [1997]) the Seiberg-Witten prepotential (the N = 2 SUSY substitute of the low-energy effective action) J^(aA), of which the second derivative is the period matrix of the spectral curve C: as = dJ^/daA, T = das/da a- Eq. (15) is in fact the equation of motion for scalar fields O/y in the 6d super-Yang-Mills theories: D^<t> = fermionic v.e.v. (16) For the topology R^xT with 2 dimensional torus T (the bare spectral curve) and the 5-brane wrapped around the torus, this equation becomes (in the gauge A =0 and Aij = diag(^/)) (d + A)(aO) = source. (17) If there is no source at the right hand side, O/y = const — this is the case of unbroken N =4 supersymmetry in four dimensions. Breakdown of supersymmetry (down to A^ = 2 as result, say, of non-trivial boundary conditions imposed along T on some fields) somehow produces a non-vanishing source on the right hand side If, for example, the source is a 5-function, m(l - 8ij)8^^H^ - ^o)Ahcn ( ^(§-§0 + ^(^/-^7))\ Lij = d^ij ^ d^ USij + m(l - 6ij) ^' 0(^-^0) I ' ^^^^ This is the Lax operator of elliptic Calogero model. In the double-scaling (Inosemtsev's) limit m -^ oo, z -^ ioo, A^^ = w?-^e^^^'^ the spectral equation det (l{z) - i) = 0 (19)
100 A. MOROZOV turns into familiar equation for the Toda-chain system: z + - = W{X), i = X— (20) z z where W(X) is a polynomial of degree A^ and z ~ e^^^^. Thus we see, that the brane dynamics forces the brane, wrapped around the bare spectral curve (with some SUSY breaking boundary conditions), to split its A^ sheets in such a way as to form thQ full spectral curve. 5.8 Correlation Functions for A^ = 2 SUSY YM in 4d The main observables in Donaldson theory are made from the scalar fields 0: the lowest components of the supermultiplets. With Ok = tr 0^ we are interested in the quantities Ck,...k, = (Ok,...OkJ. (21) The ordinary anomalies in Yang-Mills theory, like ^ ^ ' (22) imply that A^Ck„„k„=PC2k,...k,, (23) in particular A^(0k)=P(020k) (24) dA In Losev et al [1998] the correlation functions {020k) were explicitly evaluated by methods of Donaldson theory as functions of the moduli. In Gorsky et al. [1998] it was shown that these answers are indeed consistent with an interpretation in terms of Whitham T-functions, which implies that (Uk, ...OkJ = — —— (25) dTk, ...dTk„ The prepotential (Whitham t-function) J^ and the Whitham times T are introduced by the following construction (Krichever [1992], Dubrovin [1992], Itoyama etal. [1997], Nakatsu et al. [1996], and Gorsky et al. [1998]). The differential dS = X which is the eigenvalue of the Lax matrix 1-form L/y = 30/^ has the property that ddS holomorphic differential on C. (26) Omoduli
INTEGRABILITY IN SEIBERG-WITTEN THEORY 101 If one allows C to have punctures, then "holomorphic" means that the first-order poles are allowed at punctures. If punctures collide, then higher order poles are allowed. The differential dS can be now deformed: dS = X + Yl TkdQk (27) where dQk has a pole of degree ^ +1 at a given point. One can now include the cycles going around the punctures into the set of A-cycles and those going between the punctures into the set of 5-cycles. The enlarged period matrix Tkl can be used to define the prepotential: TkL = TlK = ^rr r,rr (28) The prepotential itself is given by the sum over the enlarged set of A and B cycles: Presumably it should satisfy the generalized WDVV equations (Marshakov et al [1996, 1997], Morozov [1998a], Mironov et al [1998]). In the case of the Toda chain one can explicitly evaluate the period matrix (Gorsky et al. [1988]): dTmoTn 27X1 \ mn da^ aaJ ^ / .^r.^ mn This result is in agreement with the calculation of (Losev et al. [1998]). On the other hand, it is obviously a direct generalization of Eq. (14) in the case of GKM. Acknowledgements It is a pleasure to thank H.W. Braden and other organizers of the meeting. My work is partly supported by the Russian President's grant 96-15-96939 and RFBR grant 98-02-16575. References Ambjom, J., Jurkiewicz, and Makeenko, Yu., Phys. Lett. B251, 517 (1990). Dubrovin, B., Nucl Phys. B379, 627 (1992). Gerasimov, A., Khoroshkin, S., Lebedev, D., Mironov, A., and Morozov, A., Int. J. Mod. Phys. AlO, 2589, hepth-9405011 (1995).
102 A. MOROZOV Gerasimov, A., Marshakov, A., Mironov, A., Morozov, A., and Orlov, A., Nucl Phys. B357, 565 (1991). Gorsky, A., Krichever, I., Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B355, 466, hepth-9505035 (1995). Gorsky, A., Marshakov, A., Mironov, A., and Morozov, A., Nucl. Phys. 527B, 690, hepth-9802007 (1998). Gorsky, A., Phys. Lett. B410, 22, hepth-9612238 (1997). Gubser, S., Klebanov L, and Polyakov, A., Phys. Lett. 428B, 105, hepth-9802109 (1998). Itoyama H., and Matsuo, Y., Phys. Lett. B255, 202 (1991). Itoyama H., and Morozov, A., Nucl. Phys. B491, 529, hepth-9512161 (1997). Kapustin, A., and Sethi, S., Adv. Theor. Math. Phys. 2, 571 (1998). Kapustin, A., Nucl Phys. B534, 531 (1998). Kharchev, S., Marshakov, A., Mironov, A., and Morozov, A., Mod. Phys. Lett. A8, 1047; hepth- 9208046 (1993). Kharchev, S., Marshakov, A., Mironov, A., Zabrodin, A., et al., Phys. Lett. 275B, 311, hepth-9111037 (1992); Nucl. Phys. B380, 181, hepth-9201013 (1992). Krichever, I., Comm. Math. Phys. 143, 415, hepth-9205110 (1992). Losev, A., Nekrasov, N., and Shatashvili, S., Nucl. Phys. B534, 549, hepth-9711108 (1998). Maldacena, J., Int. J. Theor. Phys. 38, 1113 (1999), hepth-9711200. Marshakov, A., Martellini, M., and Morozov, A., Phys. Lett. B418, 294, hepth-9706050 (1998). Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett B389, 43, hepth-9607109 (1996); Mod. Phys. Lett. Ml, 113, hepth-9701014; hepth-9701123 (1997). Mironov, A., and Morozov, A., Phys. Lett. B252, 47 (1990). Mironov, A., and Morozov, A., Phys. Lett. B424, 48, hepth-9712177 (1998). Morozov, A., hepth-9810031 (1998b). Morozov, A., Phys. Lett. B427, 93, hepth-9711194 (1998). Morozov, A., Russian Physics Uspekhi 35, 671 (1992) (162 #8 (1992) 84 in Russian edition). Morozov, A., Russian Physics Uspekhi 37,1 (1994) (164 (1994) 3 in Russian edition), hepth-9303139; hepth-9502091. Nakatsu T., and Takasaki, K., Mod. Phys. Lett. 11, 417, hepth-9509162 (1996). Seiberg, N., and Witten, E., Nucl. Phys. B426, 19, hepth-9407087 (1994); Nucl. Phys. B431, 484, hepth-9408099 (1994). Witten, E., Adv. Theor. Math. Phys. 2, 253 (1998), hepth-9802150. Witten, E., Nucl. Phys. B500, 3, hepth-9703166 (1997).
6 WDVV Equations and Seiberg-Witten Theory A. MIRONOV Theory Department, Lebedev Physics Institute, Moscow 117924, and ITEP, Moscow 117259, Russia E-mail: mironov@lpi.ru & mironov@itep.ru We present a review of the results on the associativity algebras and WDVV equations associated with the Seiberg-Witten solutions of N = 2 SUSY gauge theories. It is mostly based on the integrable treatment of these solutions. We consider various examples of the Seiberg-Witten solutions and corresponding integrable systems and discuss when the WDVV equations hold. We also discuss a covariance of the general WDVV equations. 6.1 What is WDVV More than two years ago N. Seiberg and E. Witten [1994] proposed a new way to deal with the low-energy effective actions ofN = 2 four-dimensional supersymmetric gauge theories, both pure gauge theories (i.e. containing only vector supermultiplet) and those with matter hypermultiplets. Among other things, they have shown that the low-energy effective actions (the end-points of the renormalization group flows) fit into universality classes depending on the vacuum of the theory. If the moduli space of these vacua is a finite-dimensional variety, the effective actions can be essentially described in terms of a system with^n/f^-dimensional phase space (# of degrees of freedom is equal to the rank of the gauge group), although the original theory lives in a many-dimensional space-time. These effective theories turn out to be integrable. Integrable structures behind the Seiberg-Witten (SW) approach have been found in Gorsky et al. [1995] and later examined in detail for different theories in Martinec etal. [1996], Gorsky etal. [1996a], [1996b], [1998], [1999], Nakatsu etal. [1996], Donagi etai [1996], Nekrasov [1998], Braden^f^/. [1999a],Marshakov^f^/. 1998],Itoyama^f^/. [1996]. 103
(Fi)jk = ^.. ^.. ^.. ^ ij,k = l,...,n (2) 104 A. MIRONOV The second important property of the SW framework which merits the adjective "topological" has been more recently revealed in the series of papers (Marshakov et al. [1996], [1997a], [1997b], Morozov [1998], Mirozov et al. [1998], Braden et al [1999b]) and has much to do with the associative algebras. Namely, it turns out that the prepotential of SW theory satisfies a set of Witten-Dijkgraaf-Verlinde-Verlinde (WDVV) equations. These equations have been originally presented in Witten et al. [1991], Dijkgraff et al. [1991] (in a different form, see below) FiFr'Fk = FkFr'Fi (1) where F/ 's are matrices with the matrix elements that are the third derivatives of the unique function F of many variables ai 's (prepotential in the SW theory) parameterizing a moduli space: d^F dai daj dak' Although generally there are a lot of solutions to the matrix equations (1), it is an extremely non-trivial task to express all the matrix elements through the only function F. In fact, until recently only two different classes of non-trivial solutions to the WDVV equations were known, both being intimately related to the two-dimensional topological theories of type A (quantum cohomologies (Manin [1996], Kontsevich et al. [1994])) and of type B (A^ = 2 SUSY Landau-Ginzburg (LG) theories that were investigated in a variety of papers, see, for example, Krichever [1994], Dubrovin [1992], and references therein). Thus, the existence of a new class of solutions connected with the four-dimensional theories looks quite striking. It is worth noting that both two-dimensional topological theories and SW theories reveal integrability structures related to the WDVV equations. Namely, the function F plays the role of the (quasiclassical) t-function of some Whitham type hierarchy (Krichever [1992, 1994], Gorsky^f^/. [1995]). In this brief review, we will describe the results of papers (Marshakov et al. [1996, 1997a, 1997b], Morozov [1998], Mironov and Morozov [1998] and Braden etal [1999b]) that deal with the structure and origin of the WDVV equations in the SW theories and, to some extent, with their general properties ^ To give some insight of the general structure of the WDVV equations, let us consider the simplest non-trivial examples ofn = 3 WDVV equations in topological theories. The first example is the A^ = 2 SUSY LG theory with the superpotential W\X) = X^ — q (Krichever [1994]). In this case, the prepotential reads as F = -aial + -aja3 + -^2^3 (3) and the matrices Ft (the third derivatives of the prepotential) are /O 0 1\ /O 1 0\ /I 0 0\ Fi = ( 0 1 0 ) , F2 = I 1 0 0 ) , F3 = I 0 0 ^ ) . (4) \1 0 0/ \0 0 q/ \0 q 0/ One can easily check that these matrices do really satisfy the WDVV equations (1). ^We tried to make this review self-consistent. Some related points can be found in other talks presented at the Workshop, in particular, delivered by A. Marshakov and A. Morozov.
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 105 The second example is the quantum cohomologies of CP^. In this case, the prepotential is given by the formula (Manin [1996]) 1 1 '^ AT ^3A:-1 where the coefficients A^;^ (describing the rational Gromov-Witten classes) counts the number of the rational curves in CP^ and are to be calculated. Since the matrices F have the form /O 0 1\ /O 1 0 \ /I 0 0 \ Fi = ( 0 1 0 ) , F2 = ( 1 F222 ^^223 I , F3 = ( 0 F223 F233 I (6) \1 0 0/ \0 F223 /^233/ \0 F233 /^333/ the WDVV equations are equivalent to the identity ^333 — ^223 ~ ^m^ll>l> 0) which, in turn, results into the recurrent relation defining the coefficients A^;^: A^;^ Y- a^b(3b - l)b(2a - b) ^^ = > NnNh. (8) (3k-4)\ V, (3a - l)\(3b - l)\ '^ a-\-b=k The crucial feature of the presented examples is that, in both cases, there exists a constant matrix F\. Following Krichever [1994], one can consider it as a flat metric on the moduli space. In fact, in its original version, the WDVV equations have been written in a slightly different form, that is, as the associativity condition of some algebra. We will discuss this later, and now just remark that, having a distinguished (constant) metric r] = F\, one can naturally rewrite (1) as the equations CiCj=CjCi (9) for the matrices C/ = r]~^Fi, i.e. (C/);[ = rjj^Fnk. Formula (9) is equivalent to (1) with j = I. Moreover, this particular relation is already sufficient (Marshakov et al. [1997a, 1997b] to reproduce the whole set of the WDVV equations (1). Indeed, since Ft = FiQ, we obtain FiFr'Fk = F,(CiCj;'Cj) (10) which is obviously symmetric under the permutation / ^^ j. Let us also note that, although the WDVV equations can be fulfilled only for some specific choices of the coordinates at on the moduli space, they still admit any linear transformation. This defines the flat structures on the moduli space, and we often call at "flat coordinates". In fact, the existence of the flat metric is not necessary for (1) to be true, as we explain below. Moreover, the SW theories give an example of just this case where there is no distinguished constant matrix. This matrix can be found in topological theories because of existence their field theory interpretation where the unity operator is always present.
106 A. MIRONOV 6.2 Perturbative SW Prepotentials Before going into the discussion of the WDVV equations for the complete SW prepotentials, let us note that the leading perturbative part of them should by itself satisfy equations (1) (since the classical quadratic piece does not contribute to the third derivatives). In each case this can be checked by straightforward calculation. On the other hand, if the WDVV equations are fulfilled for the perturbative prepotential, it is a necessary condition for them to hold for the complete prepotential. The perturbative prepotential can be obtained from the one-loop field theory calculations. To this end, let us note that there are two origins of masses in A^ = 2 SUSY YM (SYM) models: first, they can be generated by vacuum values of the scalar 0 from the gauge supermultiplet. For a supermultiplet in representation R of the gauge group G this contribution to the prepotential is given by the analog of the Coleman-Weinberg formula (from now on, we omit the classical part of the prepotential from all expressions): F/,=±^Tr/,02log0, (11) and the sign is "+" for vector supermultiplets (normally they are in the adjoint representation) and "—" for matter hypermultiplets. Second, there are bare masses rriR which should be added to 0 in (11). As a result, the general expression for the perturbative prepotential is F = ^ ^TrA(0 + MnUflogicI) + MnU) vector mplets - - ^Tr/?(0 + m/?//?)^log(0 + m/?//?) + /(m) hyper mplets where the term f(m) depending only on masses is not fixed by the (perturbative) field theory but can be read off from the non-perturbative description, and Ir denotes the unit matrix in the representation R. As a concrete example, let us consider the SU(n) gauge group. Here the perturbative prepotential for the pure gauge theory acquires the form ^v^"" ""jJl (^^ ~ ^jf ^^^ (^^ ~^j) (13) This formula establishes that when vacuum expectation values of the scalar fields in the gauge supermultiplet are non-vanishing (perturbatively ar are eigenvalues of the vacuum expectation matrix (0)), the fields in the gauge multiplet acquire masses nirr' = ar — ar' (the pair of indices {r,r') label a field in the adjoint representation of G). In the 5'C/(n) case, the eigenvalues are subject to the condition ^ • ai — 0. The analogous formula for the adjoint matter contribution to the prepotential is ^pert ^ _ V^ ^^. _ ^^ _j_ ^^ i^g ^^. _ ^^ _j_ ^^ (14)
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY while the contribution of the fundamental matter reads as 107 (15) Similar formulae can be obtained for the other groups. The eigenvalues of (0) in the first fundamental representation of the classical series of the Lie groups are Bn{S0{2n + \)) Cn (Spin)) Dn {S0{2n)) {a\, ..., Qn, 0, —a\,..., —an}\ (16) while the eigenvalues in the adjoint representation have the form Bn Cn Dn {^aj\ ±aj ± ak), j <k <n\ [^2aj\ ±aj :^ak}, j < k < n; {±aj ± ak), j < k <n. (17) Analogous formulae can be written for the exceptional groups too. The prepotential in the pure gauge theory can be read off from the formula (17) and has the form Bn Cn Dn i,i i Fo = - ^ ((^/ - ajf log {ai - aj) + {at + ajf log {at + aj)j + 2 ^^f log^,; Fo = - ^ [{ai - aj) log [at - aj) + [at + aj) log [at + aj)j i,i (18) The perturbative prepotentials are discussed in detail in Marshakov et al. [1997a]. In that paper is also contained the proof of the WDVV equations for these prepotentials. Here we just list some results of Marshakov et al. [1997a] adding some new statements recently obtained. i) The WDVV equations always hold for the pure gauge theories F^^^ = Fy^^ (including the exceptional groups)^. In fact, in Marshakov et al. [1997a] it has been proved that, if one starts with the general prepotential of the form F = -Y2[0(- {at - ajf log {at - aj)-\r a+ {at + ajf log [at + ^7)) + ^ XI ^^ ^^g at i,i i (19) ^The rank of the group should be bigger than 2 for the WDVV equations not to be empty, thus for example in the pure gauge G2-model they are satisfied trivially. It have been checked in Marshakov et al. [1997a, 1998] (using the MAPLE package) that the WDVV equations hold for the perturbative prepotentials of the pure gauge F4, Ee and Ej models. Note that the corresponding non-perturbative SW curves are not obviously hyperelliptic.
108 A. MIRONOV the WDVV hold iff a+ = a_ or a-^ =0,r] being arbitrary. A new result due to A. Veselov [1999] shows that this form can be further generalized by adding boundary terms. For this latter case, however, we do not know the non-perturbative extension. ii) If one considers the gauge supermultiplets interacting with the n/ matter hypermulti- plets in the first fundamental representation with masses m^ ^pert ^ ^pert ^ ^^pert ^ ^y^(^) (20) (where r and K are some undetermined coefficients), the WDVV equations do not hold unless K = r^/4, the masses are regarded as moduli (i.e. the equations (1) contain the derivatives with respect to masses) and for the SU(n) gauge group and "'^ (22) + ^^ 2^w„logw„ for other classical groups, 5^ = 2 for the orthogonal groups and s = —2 for the symplectic ones. Note that at value r = —2 the prepotential (20) can be considered as that in the pure gauge theory with the gauge group of the higher rank = rankG + n/. At the same time, at the value r = 2 with like at's lying in an irrep of G, the masses m^'s can be regarded as lying in an irrep of some G. If G = A„, C„, D„ then G = An, Dn.Cn- (This is nothing but the notorious (gauge group <—> flavor group) duality, see, e.g. Argyres et al [1996].) These correspondences "explain" the form of the mass term in the prepotential /(m). iii) The set of the perturbative prepotentials satisfying the WDVV equations can be further extended. Namely, one can consider higher dimensional SUSY gauge theories (Nekrasov [1998], Gorsky et al [1996b], Braden et al [1999a] and Marshakov et al [1998]), in particular, 5d theories compactified onto the circle of radius R, so that in four-dimensions it can be seen as a gauge theory of infinitely many vector supermultiplets with masses Mk = Tzk/R. Then, the perturbative prepotentials in the pure SU(n) gauge theory of such type reads as ^"" - I E {\4 + ^i3 (e-'-'O) -jZ-f (23) ij ^ ^ i>j>k where aij = at — aj and Li3(jc) is the standard tri-logarithm function. The first sum in this expression tends to the usual logarithmic prepotential F^^^ as /? ^- 0, while the second one vanishes. It warrants mentioning that the cubic terms do not come from any field theory calculation, but correspond to the Chem-Simons term Tr (A A F A F) in the field theory Lagrangian (Seiberg et al [1996]. It is similar to the C/^-terms of the perturbative
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 109 prepotential F^^^ of the heterotic string (Harvey etal. [1996]). The presence of these terms turns to be absolutely crucial for the WDVV equations to hold. Further details on this case, and on the prepotential with fundamental hypermultiplets included can be found in Marshakov ^f ^/. [1998]. One can also consider other classical groups. Then, the perturbative prepotentials acquire the form (18) with all jc^ log jc substituted by ^x^ + ^Lis (e~^^^^). One can easily check, along the line of Marshakov et al. [1998] that these prepotentials satisfy the WDVV equations. iv) If in the four dimensional theory adjoint matter hypermultiplets are present, i.e. ^pert ^ ^pert ^ ^pert ^ y^(^) ^^e WDVV equations never hold. At the same time, the WDVV equations are fulfilled for the theory with matter hypermultiplets in the symmetric/antisymmetric square of the fundamental representation^ if and only if the masses of these hypermultiplets are equal to zero. v) Our last example (Braden et al. [1999a,b]) has the most unclear status, at the moment. It corresponds to the pure gauge 5d theory with higher, N = 2 SUSY in five dimensions. Starting with such a five dimensional model one may obtain four dimensional N = 2 SUSY models (with fields only in the adjoint representation of the gauge group) by imposing non-trivial boundary conditions on half of the fields: (l>(x5-\-R)=e^^'(l>(x5). (25) If 6 = 0 one obtains N = 4 SUSY in four dimensions, but when e ^ 2iin this is explicitly broken to N = 2. The low-energy mass spectrum of the four dimensional theory this time contains two towers of Kaluza-Klein modes: Tin € -\- Tin M = — andM = --^^ , N e Z. (26) R R The prepotential for the group SU(n) (i = I.. .n) should be i,j n=~oo I - {Raij -\-7Tn- ef log {Raij -\-7Tn-€)\ = - Y^' fiatj) i ij with f(a) = Lis (^-''■'^") - Lis (^-2^(^-+0^ (28) ^These hypermultiplets contribute to the prepotential Fs=- - 2J(^' + ^j + '")^ log(^' + ^j + '^) Fas = -- 2_^(^/ + ^j + ffi)^ log(^/ + ^j + f^) {Raij + Tin) log {Raij + nn) (27)
no A. MIRONOV The prepotential (27) satisfies the WDVV equations if and only if e = tz (Braden et al. [1997b]). Moreover, it gives the most general solution for a general class of perturbative prepotentials Fpen assuming the functional form F = J2f(aa), (29) where the sum is over the root system O of a Lie algebra. The mystery about this prepotential is that, on the one hand, it never satisfies the WDVV equations unless e = tt and we do not know if the corresponding complete non-perturbative prepotential satisfies the WDVV even for 6 = tt. On the other hand, in the limit R -^ 0 and e ^ mR for finite m, (when the mass spectrum (26) reduces to the two points M = 0 and M = m), the theory is the four dimensional YM model with N = 4 SUSY softly broken to N = 2, i.e. includes the adjoint matter hypermultiplet. This corresponds to item iv) when the WDVV always do fail. For all these reasons, this exceptional case deserves further investigation. From the above consideration of the WDVV equations for the perturbative prepotentials, one can learn the following lessons: • masses'are to be regarded as moduli • as an empirical rule, one may say that the WDVV equations are satisfied by perturbative prepotentials which depend only on the pairwise sums of the type (at -^bj), where moduli at and bj are either periods or masses'^. This is the case for the models that contain either massive matter hypermultiplets in the first fundamental representation (or its dual), or massless matter in the square product of those. Troubles arise in all other situations because of the terms with ai±bj ±Ck± (The converse statement is false - there are some exceptions when the WDVV equations hold in spite of the presence of such terms - e.g., for the exceptional groups.) Note that for the non-UV-finite theories with perturbative prepotential satisfying the WDVV equations, one can add one more parameter to the set of moduli - the parameter A that enters all the logarithmic terms as jc^ log jc/A. Then, some properly defined WDVV equations still remain correct despite the matrices F.~ no longer existing (one just needs to consider instead of them the matrices of the proper minors) (Bertoldi et al. [1998]) — see footnote 9 in section 6.6. 6.3 Associativity Conditions In the context of the two-dimensional LG topological theories, the WDVV equations arose as associativity condition of some polynomial algebra. We will prove below that the equations in the S W theories have the same origin. Here we briefly recall the main ingredients of this approach in the standard case of the LG theories. In this case, one deals with the chiral ring formed by a set of polynomials {0/ (X)} and two co-prime (i.e. without common zeroes) fixed polynomials Q(X) and P(X). The polynomials "^This general rule can be easily interpreted in D-brane terms, since the interaction of branes is caused by strings between them. The pairwise structure (a, ± bj) exactly reflects this fact, a, and bj should be identified with the ends of string.
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 111 O form the associative algebra with the structure constants C^- given with respect to the product defined by modulo P': <t>i<t>j = C^j<t>kQ' + (*)P' -^ C^j^kQ' (30) the associativity condition being (cD,c|>.)cD;^ = cD. (cD-cD;^), (31) i.e. CiCj=CjCi. (Q)i=cik (32) Now, in order to get from these conditions the WDVV equations, one needs to properly choose the flat moduli (Krichever [1994]): —res (P'/" J(2), n = ord(P) (33) at = i(n — i) Then, there exists the prepotential whose third derivatives are given by the residue formula On the other hand, from the associativity condition (32) and the residue formula (34), one obtains that Fijk = (Qyj FQUk, i.e. Ci=FiF-}. (35) Substituting this formula for C/ into (32), one finally obtains the WDVV equations in the form FiG-^Fj = FjG-^Fi, ^ ^ (36) G^Fq. The choice Q' = O/ gives the standard equations (1). In two-dimensional topological theories, there is always the unity operator corresponding to Q' = 1, and this leads to the constant metric Fq' . Thus, from this short study of the WDVV equations in the LG theories, we can get three main ingredients necessary for these equations to hold. These are: • associative algebra; • flat moduli (coordinates); and • residue formula. We will show that in the S W theories only the first ingredient requires a non-trivial check, while the other two are automatically present because of appropriate integrable structures. 6.4 SW Theories and Integrable Systems Now we turn to the WDVV equations emerging within the context of the SW construction (Seiberg and Witten [1994]) and show how they are related to integrable system underlying the corresponding SW theory. The most important result of (Seiberg and Witten [1994]),
112 A. MIRONOV from this point of view, is that the moduli space of vacua and low energy effective action in SYM theories are completely given by the following input data: • Riemann surface C; • moduli space M. (of the curves C; and) • meromorphic 1-form dS on C. As pointed out in Gorsky et al. [1995, 1996] and Marshakov et al. [1997a], this input can naturally be described in the framework of some underlying integrable system. Let us consider a concrete example — thQ SU(n) pure gauge SYM theory that is associated with the periodic Toda chain with n sites. This integrable system is encoded by the Lax operator / pi ^^'-^2 m;^^«-^i\ L(w) = e^^-^2 p^ (37) X^-e^^-^^ ... pn J The Riemann surface C of the SW data is nothing but the spectral curve of the integrable system, which is given by the equation dct(L(w)-X) =0. (38) Taking into account (37), one can get from this formula the equation 1 " 1 = 1 i where the ramification points A/ are Hamiltonians (integrals of motion) parameterizing the moduli space M of the spectral curves. The substitution Y = w — \/w transforms the curve (39) into the standard hyperelliptic form Y^ — P^ — 4, the genus of the curve being n — \. The same integrable system, i.e. the periodic Toda chain may be alternatively rewritten in terms of 2 x 2 Lax matrices Li each associated with the site of the chain: The Lax operator Li can be considered as an "infinitesimal" transfer matrix that shifts from the /-th to the / + 1-th site of the chain f:i(X)^i{\) = ^i^x{X) (41) where ^/ (A) is the two-component Baker-Akhiezer function. One also needs to consider proper boundary conditions. In the SU(n) case, they are periodic. The periodic boundary conditions are easily formulated in terms of the Baker-Akhiezer function and read as ^i^n(^) = w^i(X) (42)
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 113 where w; is a free parameter (diagonal matrix). The Toda chain with these boundary conditions can be naturally associated with the Dynkin diagram of the group aJ^^j. One can also introduce the transfer matrix shifting / to i -\-n T(X)=Cn(X)...Ci(X), (43) Now the periodic boundary conditions are encapsulated in the spectral curve equation dct(T(X) - w;. 1) = 0, (44) or w^ - TtT(X)w + det r(A) = 0. (45) This curve coincides with (39), since det T(X) = I for the Toda Lax operators (40). The last important ingredient of the construction is the meromorphic 1-form dS = X^ = X^. From the point of view of the Toda chain, it is just the action ''pdq'' along the non-contractible contours on the Hamiltonian tori. Its defining property is that the derivatives of dS with respect to the moduli (ramification points) are holomorphic differentials on the spectral curve. After this concrete example, we are ready to describe how the SW data emerge within a more general integrable framework and then discuss more on the concrete examples of the SW construction. As before, we start with the theories without matter hypermultiplets. First, we introduce a bare spectral curve E (that is the torus j^ = jc^ + g2X^ + g3 for the UV finite SYM theories) with the associated holomorphic 1-form dco = dx/y. This bare spectral curve degenerates into the double-punctured sphere (annulus) for asymptotically free theories: x -^ w-\- \/w, y -^ w — \/w, dco = dw/w. On this bare curve we are further given either a matrix-valued Lax operator L(x,y) (if one considers an extension of the (37) Lax representation), or another matrix Lax operator Ci (x,y) associated with an extension of the representation (40) and defining the transfer matrix T(x,y). The corresponding dressed spectral curve is defined either from the formula det(L — A) = 0, or from det(r — w)=0. This spectral curve is a ramified covering of E given by the equation V(X;x,y)=0. (46) In the case of the gauge group G = SU(n), the function P is a polynomial of degree n in X. Thus, the moduli space M of the spectral curve is given just by coefficients of V. The generating 1-form dS = Xdco is meromorphic on C ("=" denotes the equality modulo total derivatives). The prepotential and other "physical" quantities are defined in terms of the cohomology class of dS: ai = (p dS, (47) dS, Ai oBj =8ij.
114 A. MIRONOV The first identity defines here the appropriate flat moduli while the second one defines the prepotential. The defining property of the generating differential dS is that its derivatives with respect to moduli give holomorphic 1-differentials. In particular, ddS dai = dcoi (48) and, therefore, the second derivative of the prepotential with respect to ai 's is the period matrix of the curve C: dai daj (49) The latter formula allows one to identify the prepotential with the logarithm of the t-function of the Whitham hierarchy (Krichever ^f ^/. [1992, 1994]): F = logr. So far we reckoned without matter hypermultiplets. In order to include them, one just needs to consider the surface C with punctures. Then, the masses are proportional to residues of dS at the punctures, and the moduli space has to be extended to include these mass moduli. All other formulas remain in essence the same (see Marshakov et al. [1997a] for more details). The known correspondences between SYM theories and integrable systems associated with the SW construction are collected in the table^. TABLE 6.1. SUSY gauge theories ^^=^ integrable systems correspondence. SUSY physical theory 4d 5d 6d Comments Pure gauge SYM theory, gauge group G Toda chain for the dual affine G^ Relativistic Toda chain There are two Lax representations SYM theory with fund. matter XXX spin chain XXZ spin chain XYZ spin chain Spherical bare curve SYM theory with adj. matter Calogero-Moser system Ruijsenaars- Schneider model^ Elliptic bare curve ^In Table 6.1 we considered only the classical groups. ^This case is better associated with the specific boundary conditions imposed on the fields in the fifth dimension (see Braden et al. [1999a,b] and section 6.2) than with the adjoint matter added Nekrasov [1998].
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 115 The table reflects the several possible generalizations of the periodic Toda chain. First of all, one can extend the representation in terms of the "large" Lax matrices (37). It naturally leads to the system whose potential is a doubly-periodic function of the coordinates. This system is the Calogero-Moser system. It is associated with an elliptic bare spectral curve and generating 1-form dS = Xd^, § being the coordinate on the bare torus. On the physical side, this system corresponds to including the adjoint matter hypermultiplets (Donagi et al. [1996]). The second possible extension is to generalize the 2 x 2 Lax representation of the Toda chain (40). This leads to the XXX spin chain with a cylindrical bare spectral curve and the same generating 1-form dS = Xdw/w. Physically, this system describes the inclusion of fundamental matter hypermultiplets (Gorsky et al. [1996, 1998]). Now, each of these systems can be further generalized to higher dimensional (5d and 6d) SYM gauge theories (Gorsky et al. [1996b], Braden et al. [1999a], Marshakov et al. [1998]), with the target space (the fifth and sixth dimension) being accordingly a cylinder or a torus. Within the integrable framework, this means putting the momenta of the system onto the cylinder (Hamiltonians periodic in momenta) and the torus (Hamiltonians doubly-periodic in momenta) respectively. In the generating 1-form one needs simply to substitute respectively X -^ log A in 5 dimensions and A ^- f in 6 dimensions, f being the coordinate on the target space torus. At the same time, the bare spectral curve associated with coordinate dependence is unchanged in this extension. In a word, adding adjoint matter makes the coordinate dependence (and the bare spectral curve) elliptic, while going to higher dimensions provides trigonometric and elliptic momentum dependence (and the corresponding target space). Let us discuss a little the dressed spectral curves. In the adjoint matter case, the spectral curve is non-hyperelliptic, since the bare curve is elliptic. Therefore, it can be described as some covering of the hyperelliptic curve. We will not go into further details here, just referring to Marshakov et al. [1997a] and Braden et al. [1999a], since the WDVV equations do not hold, at least, in the standard form (1), with the elliptic bare curve (see below). Instead, we will describe in more explicit terms the dressed spectral curves for the 4 dimensional theories without adjoint matter and with classical gauge group. Let us note that in all these cases the curves are hyperelliptic, since all of them follow from some 2x2 Lax representation and are, therefore, the spectral curves of the form (45). More concretely, for the SYM gauge theory with the gauge group G, one should consider the integrable system given by this 2x2 Lax representation on the Dynkin diagram for the corresponding dual affine algebra G^ (Gorsky and Mironov [1999]). The spectral curves can be described by the general formula Q(X) V(X, w) = 2P(X) -w- ^^ (50) w Here P(X) is the characteristic polynomial of the algebra G itself for all G 7^ C„, i.e. P(X) = det(G - XI) = Y\(^ - ^i) (51)
116 A.MIRONOV where the determinant is taken in the first fundamental representation and the A/'s are the eigenvalues of the algebraic element G. For the pure gauge theories of the classical groups (Martinec etal. [1996]), Q(X) = X^' and'^ An-i : P(X) = Y[(X-Xi), s=0; i=\ n '=' (52) « 2 Cn: P(X) = Y\(X^-Xf)--^, s = -2; Dni P(X) = Y[(^^-^h ^ = 2 /=1 For the exceptional groups, the curves arising from the characteristic polynomials of the dual affine algebras do not acquire this hyperelliptic form, although the WDVV equations still seem to be fulfilled. In order to include np massive hypermultiplets in the first fundamental representation one can simply change X^^ for Q(X) = X^^ n"=i(^ ~ ^i) if G = A„ and for Q{X) = X^' n"li(^^ - ^?) ifG = BnXn,Dn (Gorsl^ and Mironov [1999], Argyres and Sharpere [1996], Hanany [1996]). Note that the 5 dimensional theories can be also described by the same curves but with different 1-forms dS (Marshakov et al [1998]). 6.5 WDVV Equations in SW Theories As we already already discussed, in order to derive the WDVV equations along the line used in the context of the LG theories, we need three crucial ingredients: flat moduli, a residue formula and an associative algebra. The first two of these are however always contained in the SW construction provided the underlying integrable system is known. Indeed, one can derive (see Marshakov et al. [1997a]) the following residue formula Fijk = res —-—^- , (53) da)=o dcodX where the proper flat moduli ai 's are given by formula (47). Thus, the only point to be checked is the existence of the associative algebra. The residue formula (53) hints that this algebra is to be the algebra Q^ of the holomorphic differentials dcoi. In the following discussion we restrict ourselves to the case of pure gauge theory, the general case being treated in complete analogy. ^In the symplectic case, the curve can be easily recast in the form with polynomial F(A) and 5=0.
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 117 Let us consider the algebra Q^ and fix three differentials dQ, dco, dX e Q^. The product in this algebra is given by the expansion dcoidcoj = C^jdcokdQ + (^)dco + (*)JA (54) that should be factorized over the ideal spanned by the differentials dco and dX. This product belongs to the space of quadratic holomorphic differentials: Q^ -Q^ eQ^ = Q^ ■ (dQ 0 Jo; 0 dX) (55) Since the dimension of the space of quadratic holomorphic differentials is equal to 3g — 3, the left hand side of (54) with arbitrary dcot's is a vector space of dimension 3g — 3. At the same time, on the right hand side of (54) there are g arbitrary coefficients Cf in the first term (since there are exactly this many holomorphic 1-differentials that span the arbitrary holomorphic 1-differential Cf-dcok), g — 1 arbitrary holomorphic differentials in the second term (one differential should be subtracted to avoid the double counting) and g — 2 holomorphic 1-differentials in the third one. Thus, in total, we get that the right hand side of (54) is spanned also by the basis of dimension g -\- (g — I) -\- (g — 2) = 3g — 3. This means that the algebra exists in the general case of the SW construction. However, this algebra is generally not associative. This is because, unlike the LG case, when it was the algebra of polynomials and, therefore, the product of the two belonged to the same space (of polynomials), the product in the algebra of holomorphic 1-differentials no longer belongs to the same space but to the space of quadratic holomorphic differentials. Indeed, to check associativity, one needs to consider the triple product of ^^: Q^ 'Q^ 'Q^ eQ^ = Q^' (dQf e Q^' dco e Q^■ dX (56) Now let us repeat our calculation: the dimension of the left hand side of this expression is 5g — 5, that is the dimension of the space of holomorphic 3-differentials. The dimension of the first space in expansion of the right hand side, is g, the second one is 3g — 4 and the third one is 2g — 4. Since g + (3g — 4) + (2g — 4) = 6g — 8 is greater than 5g — 5 (unless g S 3), formula (56) does not define the unique expansion of the triple product of Q^ and, therefore, the associativity fails. The situation can be improved if one considers the curves with additional involutions. As an example, let us consider the family of hyperelliptic curves: y'^ = Pol2gj^2{X). In this case, there is the involution, a : y -^ —y and Q} is spanned by the a-odd holomorphic 1-differentials ^' ^^, / = 1,..., g. Let us also note that both dQ and dco are a-odd, while dX is a-even. This means that dX can only be meromorphic on a surface without punctures (which is, indeed, the case in the absence of mass hypermultiplets). Thus, dX drops from formula (54) which thus acquires the form Qi^ = Q}_'dQ®Q}_' dco, (57) where we expanded the space of holomorphic 2-differentials into the parts with definite a-parity: Q^ — ^^ 0 ^?^, which are manifestly given by the differentials ^' ^^^ , / = 1,..., 2g — 1 and -—y^, i = 1,..., g — 2 respectively. Now it is easy to understand that the dimensions of the left hand side and right hand side of (57) coincide and are equal to2g-l.
118 A. MIRONOV Analogously, in this case, one can check the associativity. It is given by the expansion qI=qI- (dQf eQl'dco (58) where both the left hand side and right hand side have the same dimension: 3g — 2 = g + (2g — 2). Thus, the algebra of holomorphic 1-differentials on hyperelliptic curve really is associative. This completes the proof of the WDVV equations in this case. Now let us briefly look at the cases when there exists an associative algebra arising from the spectral curves discussed in the previous section. First of all, it exists in the theories with gauge group A„, both in the pure gauge 4 dimensional and 5 dimensional theories and in the theories with fundamental matter, since, in accordance with the previous section, the corresponding spectral curves are hyperelliptic of genus n. The theories with the gauge groups SO(n) or Sp(n) are also described by the hyperelliptic curves. The curves, however, are of higher genus 2n — I. This would naively destroy all the reasoning of this section. The arguments, however, can be restored by noting that the corresponding curves (see (52)) have yet another involution, p : X -^ —X. This allows one to expand further the space of holomorphic differentials into the pieces with definite p-parity: ^L = ^L_ 0 ^L_^ etc. so that the proper algebra is generated by the differentials from ^L_. One can easily check that it leads again to an associative algebra. Similar considerations are even more tricky for the exceptional groups, when the corresponding curves appear non-hyperelliptic. However, additional symmetries should allow one to obtain associative algebras in these cases too. There are more cases when an associative algebra exists. First of all, there are 5 dimensional theories, with and without fundamental matter (Marshakov et al. [1997a]). One can also consider the SYM theories with gauge groups being the product of several factors, with matter in the bi-fundamental representation (Witten [1997]). These theories are described by SL(p) spin chains (Gorsky et al. [1998]) and the existence of an associative algebra in this case has been checked in Isidro [1999]. The situation is completely different in the adjoint matter case. In four dimensions, the theory is described by the Calogero-Moser integrable system. Since, in this case, the curve is non-hyperelliptic and lacks enough additional symmetries, one needs to include into the considerations both the differentials do) and dX for algebra to exist. However, under these circumstances, the algebra is no longer associative, as was demonstrated above. This can also be verified by direct calculation for the first values of n (see Marshakov et al. [1997a]). This also explains the lack of the perturbative WDVV equations in this case (see section 6.2). 6.6 Covariance of the WDVV Equations Having discussed the role of the (generalized) WDVV equations in SYM gauge theories of the Seiberg-Witten type, let us briefly describe the general structure of the equations themselves. We look at them now just as at some over-defined set of non-linear equations
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 119 for a function (prepotential) of r variables^ (times), F(f'), / = 1,... , r, which can be written in the form (36) FiG-^Fj = FjG-^Fi, k=\ Fi being r xr matrices (Fi)jk = Fjjk = d/dt^t'' ^^^ ^^^ "metric" matrix G is an arbitrary Hnear combination of F^'s, with coefficients r]^(t) that can be time-dependent.^ The WDVV equations imply consistency of the following system of differential equations (Morozov [1998]): fFjjk^-Fjji^^r(t)=0, ^iJ^k (60) Contracting with the vector r]^(t), one can also rewrite it as -JL.=Cj^Dr, yij (61) where Ck=G-^Fk, G = n'Fi, D^n'di (62) ^We deliberately chose different notations for these variables, t instead of a in the gauge theories, in order to point out the more general status of the discussion. ^We already discussed in section 6.2 that one can add to the set of times (moduli) in the WDVV equations the parameter A (Bertoldi and Matone [1996]). In this case, the prepotential that depends on one extra variable r^ = A can be naturally considered as a homogeneous function of degree 2: see Krichever et al. [ 1992] for the general theory. As explained in Bertoldi and Matone [ 1996], the WDVV equations (59) for F(t') can be also rewritten in terms of T(t^): TiQ-'Tj=TjQ-'Ti, V/,7 = 0, 1,... ,r;{r?^(0} where this time Ti are (r + 1) x (r + 1) matrices of the third derivatives of ^ and Q = Y^^''TK. g-i=(detg)g- /t=0 Note that the homogeneity of T implies that r^-derivatives are expressed through those with respect to f, e.g. t ^,oij = —^jjkt , t ^,ooi = ^jkit t , t JFooo = —^,kimt t t'" etc. Thus, all the "metrics" Q are degenerate, but Q~^ are non-degenerate. One can easily reformulate the entire present section in terms of ^.Then, e.g., the Baker-Akhiezer vector-function \l/(t) should be just substituted by the manifestly homogeneous (of degree 0) function \l/(t' /t^). The extra variable t^ should not be mixed with the distinguished "zero-time" associated with the constant metric in the 2d topological theories which generically does not exist (when it does, see comment 6.2.3 below, we identify it with t'').
120 A. MIRONOV (note that the matrices Q and the differential D depend on choice of {r)\t)}, i.e. on choice of the metric G) and (59) can be rewritten as [Q,C,]=0, V/,; (63) As we already discussed, the set of the WDVV equations (1) is invariant under linear change of the time variables with the prepotential unchanged (Marshakov et al. [1997a]). According to the second paper of Krichever et al. [1994] and especially to Losev [1997], there can also exist non-linear transformations which preserve the WDVV structure, but they generically change the prepotential. In Mironov et al. [1998], it is shown that such transformations are naturally induced by solutions of the linear system (60): (64) t' -^ F(t) -^ ins intact: 7 . . — ''•^ dt^dtJ p = r(t). F(t), dPdtJ ^^J (65) Now let us make some comments. 1. As explained in Morozov [1998], the linear system (60) has infinitely many solutions. The "original" time-variables are among them: x//^ (t) = t^. 2. Condition (65) guarantees that the transformation (64) changes the linear system (60) only by a (matrix) multiplicative factor, i.e. the set of solutions {i/^(t)} is invariant of (64). Among other things this implies that successively applying (64) one does not produce new sets of time-variables. 3. We already discussed that, in the case of 2d topological models (Manin et al [1996], Krichever et al [1994] and Losev [1997]), there is a distinguished time-variable, say, t^, such that all Fjjk are independent of t^: — Fijk=0 V/,y,^ = l,...,r (66) (equivalently, -^Frjk = 0 V/, j, k). Then, one can make the Fourier transform of (60) with respect to t^ and substitute it by the system ±jfl=zCi,fl ^ij (67) where f^it^,... ,t''~^) — Jf^{t'^,... , t''^^, fje^''df. In this case, the set of transformations (64) can be substituted by a family, labeled by a single variable z: f -^ P= xkUt) (68)
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 121 In the limit z -^ 0 and for the particular choice of the metric, G = Fr, one obtains the particular transformation a7^ = C'jj^hl", hi" = const, (69) discovered in Losev [1997]. (Since C/ = 3/C, one can also write Ci = C[h^, 4. Parameterizations like (69) can be used in the generic situation (64) as well (i.e. without distinguished f'^-variable and for the whole family (64)), the only change is that h^ is no longer a constant, but a solution to {dj-DCj)[h^=0 (70) (h^ = Di//^ is always a solution, provided i//^ satisfies (61)). Note also that, although we have described a set of non-trivial non-linear transformations which preserve the structure of the WDVV equations (59), the consideration above does not prove that all such transformations are of the form (64), (65). Still, (64) is already unexpectedly large, because (59) is an over-defined system and it could seem to be very restrictive, if to have any solutions at all. 6.7 Concluding Remarks To conclude this short review, let us emphasize that a lot of problems have to be solved before we get any real understanding of what the WDVV structure means. We already mentioned the problem of lack of the WDVV equations for the Calogero-Moser system. The way to resolve this problem might be to construct higher associativity conditions as has been done by E.Getzler in the elliptic case (Caporaso et al. [1996]) (that is to say, for the elliptic Gromov-Witten classes). The other kind of problem is that the WDVV equations for the type A topological theories themselves still await an explanation in terms of associative algebras. All these problems are to be resolved in order to establish to what extent there is a really deep reason for the WDVV equations to emerge in topological and Seiberg-Witten theories. Note that the latter two are connected through the Whitham hierarchies (see, e.g., Moore et al. [1997], the second paper in Krichever et al [1992] and the review by A. Morozov at the Workshop) and there are also tight connections of the Whitham hierarchies with the WDVV equations (Krichever [1992]). The other problem is more on the structure of the WDVV equations themselves: at present we do not understand what the class of solutions to the equations is, how wide it is, only having some particular examples in hands. Perhaps even more obscure are the origins and implications of the covariance of the WDVV equations. Finally we still have no clear understanding of the connections between the WDVV and integrable structures. Rather what we have is a set of random observations.
122 A. MIRONOV All these problems call for better understanding before the associative algebras and WDVV equations can be put on any solid footing. Acknowledgement I am grateful to H.W. Braden, A. Gorsky, A. Marshakov and A. Morozov for useful discussions and to H.W. Braden for strongly encouraging me to write this review. I also acknowledge the hospitality of University of Edinburgh and the Royal Society for support under a joint project. The research is partly supported by the RFBR grant 98-01-00328, INTAS grant 96-482 and the program for support of the scientific schools 96-15-96798. References Ahn, C, and Nam, S., Phys, Lett. B387, 304, hep-th/9603028 (1996). Argyres, R, and Shapere, A., Nucl. Phys. B461, 437, hep-th/9509175 (1996). Bertoldi, G., and Matone, M., Phys. Rev. D57, 6483-6485; hep-th/9712109 (1998). Braden, H.W., Marshakov, A., Mironov, A., and Morozov, A., hep-th/9812078; Phys. Lett. B448,195 (1999). Braden, H.W., Marshakov, A., Mironov, A., and Morozov, A., hep-th/9902205; Nucl. Phys. B558,371 (1999). Brandhuber, A., and Landsteiner, K., Phys. Lett. B358, 73, hep-th/9507008 (1995). Caporaso, L., Harris, J., Invent. Math. 131, 345-392 (1998) alg-geom/9608025. Carroll, R., hep-th/9712110. Danielsson, U., and Sundborg, B., Phys. Lett. B358, 273, hep-th/9504102 (1995). Dijkgraaf, R., Verlinde, E., Verlinde, H., Nucl. Phys. B352, 59 (1991). Donagi, R., Surv. Differ. Geom. IV, Int. Press, Boston, MA, 1998, alg-geom/9705010. Donagi, R., and Witten, E., Nucl. Phys. B460, 299, hep-th/9510101 (1996). Dubrovin, B., Nucl. Phys. B379, 627; hep-th/9407018 (1992). Eguchi, T., and Yang, S.K., Mod. Phys. Lett. All, 131—138, hep-th/9510183 (1996). For reviews see: Itoyama H., and Morozov, A., "Frontiers in quantum field theory", 301-324, World Sci., NJ, 1996, hep-th/9601168. Getzler, E., J. Amer. Math. Soc. 10, 973-998 (1997) alg-geom/9612004; for the latest review, and further references, see: Getzler, E., Contemp. Math. 241, Amer. Math. Soc, 1999, math/9812026. Gomez, C, and Hernandez, R., hep-th/9711102. Gorsky, A., and Marshakov, A., Phys. Lett. B375, 127, hep-th/9510224 (1996). Gorsky, A., and Mironov, A., hep-th/9902030; to appear in Nuclear Physics B550, 513 (1999). Gorsky, A., Gukov, S., and Mironov, A., Nucl. Phys. B517, 409, hep-th/9707120 (1998). Gorsky, A., Gukov, S., and Mironov, A., Nucl. Phys. B518, 689, hep-th/9710239 (1998). Gorsky, A., Krichever, I., Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B355, 466, hep-th/9505035 (1995). Gorsky, A., Marshakov, A., Mironov, A., and Morozov, A., Nucl. Phys. B550, 513, hep-th/9604078 (1999). Gorsky, A., Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B380, 75, hep-th/9603140 (1996). Gorsky, A., Marshakov, A., Mironov, A., Morozov, A., Nucl. Phys. B527, 690-716, hep-th/9802007 (1998). Hanany, A., Nucl. Phys. 466, 85 (1996), hep-th/9509176. Harvey, J., and Moore, G., Nucl. Phys. B463, 315-368; hep-th/9510182 (1996).
WDVV EQUATIONS AND SEIBERG-WITTEN THEORY 123 Intriligator, K., Morrison, D.R., and Seiberg, N., Nucl. Phys. B497, 56 (1997); hep-th/9702198. Isidro, J.M., NucL Phys. B539, 379^02, hep-th/9805051 (1999). Ito, K., and Yang, S.-K., Phys. Lett. B433, 56-62 (1998). Itoyama H., and Morozov, A., Nucl. Phys. B477, 855, hep-th/9511125; hep-th/9512161 (1996). Ketov, S., hep-th/9710085. Klemm, A., hep-th/9705131. Kontsevich, M., Manin, Yu., Comm. Math. Phys. 164, 525 (1994). Krichever, I., and Phong, D., Surv. Differ. Geom. TV, Int. Press, Boston, MA, 1998, hep-th/9708170. Krichever, I., Comm. Math. Phys. 143, 415, hep-th/9205110 (1992). Krichever, I., Comm. Pure Appl. Math. 47, 437 (1994). Losev, A., Nekrasov, N., and Shatashvili, S., Nucl. Phys. B534, 549-611, hep-th/9711108; hep- th/9801061 (1998). Losev, A., JETP Lett. 65, 374 (1997). Manin, Yu., Frobenius manifolds, quantum cohomology, and moduli spaces, Amer Math Soc. Colloquium Publications 47, Amer Math. Soc. (1999). Marshakov, A., and Mironov, A., NucL Phys. B518, 59-91, hep-th/9711156 (1998). Marshakov, A., Mironov, A., and Morozov, A., hep-th/9701123. Marshakov, A„ Mironov, A., and Morozov, A., Mod. Phys. Lett. All, 773, hep-th/9701014 (1997). Marshakov, A., Mironov, A., and Morozov, A., Phys. Lett. B389, 43, hep-th/9607109 (1996). Marshakov, A., Int. J. Mod. Phys. All, 1607, hep-th/9610242 (1997); Theor & Math. Phys. Ill, 791, hep-th/9702083 (1997). Martinec, E., and Warner, N., hep-th/9511052; Marshakov, A., Mod. Phys. Lett. All, 1169, hep- th/9602005 (1996). Martinec, E., and Warner, N., Nucl Phys. B459, 97-112, hep-th/9509161 (1996). Martinec, E., Phys. Lett. B367, 91, hep-th/9510204 (1996). Mironov, A., and Morozov, A., Phys. Lett. B424,48-52, hep-th/9712177 (1998). Mironov, A., Nucl. Phys. B. Suppl. 61A, 177-185 (1998), hep-th/9704205. Mironov, A., hep-th/9801149. Moore, G., and Marino, M., Nucl. Phys. Prod. Suppl. 68, 336-347, hep-th/9712062 (1998). Moore, G., and Witten, ^.,Adv. Theor Math. Phys. 1, 298-387 (1997), hep-th/9709193. Morozov, A., Phys. Lett. B427, 93-96, hep-th/9711194 (1998). Nakatsu T., and Takasaki, K., Mod. Phys. Lett. 11, 417, hep-th/9509162 (1996). Nekrasov, N., Nucl. Phys. B531, 323, hep-th/9609219 (1998). Seiberg, N., and Morrison, D., Nucl. Phys. B483, 229; hep-th/9609070 (1997). Seiberg, N., and Witten, E., Nucl. Phys. B426, 19, hep-th/9407087 (1994); Nucl Phys., B431, 484, hep-th/9408099 (1994). Seiberg, N., Phys. Lett. B388, 753; hep-th/9608111 (1996). Takasaki, K., Asian J. Math. 1, 1049-1078 (1998), solv-int/9704004; Lett. Math. Phys. 43, 123-135 (1998), solv-int/9705016. Veselov, A., Phys. Lett. A261, 297-302 (1999), hep-th/9902142. Witten, E., Nucl Phys. B500, 3, hep-th/9703166 (1997). Witten, E., Surv. Diff. Geom. 1, 243 (1991).
7 On Geometry of a Special Class of Solutions to Generalized WDVV Equations A.R VESELOV Department of Mathematical Sciences, Loughborough University, Loughborough, Leicestershire, LE 11 3TU, UK Landau Institute for Theoretical Physics, Kosygina 2, Moscow, 117940, Russia E-mail: A.RVeselov@lboro.ac.uk A special class of solutions to the generalized WDVV equations related to a finite set of covectors is considered. We describe the geometric conditions (v-conditions) on such a set which are necessary and sufficient for the corresponding function to satisfy the generalized WDVV equations. These conditions are satisfied for all Coxeter systems but there are also other examples discovered in the theory of the generalized Calogero-Moser systems. As a result some new solutions for the generalized WDVV equations are found. 7.1 Introduction The WDVV (Witten-Dijgraaf-Verlinde-Verlinde) equations have been introduced first in topological field theory as some associativity conditions (see Witten [1991] and Dijkgraaf et al. [1991]). Dubrovin has found an elegant geometric axiomatization of the these equations introducing a notion of Frobenius manifold (see Dubrovin [1992]). What we will discuss in this paper are their generalized versions that appeared in the Seiberg-Witten theory (see Seiberg [1994] and Marshakov [1996]). The generalized WDVV equations are the following overdetermined system of nonlinear partial differential equations: FiF-^Fj=FjF^^Fi, /,i,^ = l,...,n, (1) where Fm is the n x n matrix constructed from the third partial derivatives of the unknown 125
126 A.R VESELOV function F = F(jc\ ... , jc"): _ d^ F In this form these equations have been presented by A. Marshakov, A. Mironov and A. Morozov, who showed that the Seiberg-Witten prepotential in A^ = 2 four-dimensional supersymmetric gauge theories satisfies this system [1996]. We will consider the following special class of the solutions to (1): F^ = ^(a,Jc)2log(a,Jc)^ (3) where 21 be a finite set of noncoUinear vectors a in R". It is known to be a solution of the generalized WDVV equations in case when 21 is a root system (see Marshakov et al. [1997b] for the classical root systems and Martini [1999] for the general case). In the paper by Veselov [1999] it was observed that the same is true for any Coxeter configuration where the general geometric conditions on 21 have been found which guarantee that (3) satisfies the generalized WDVV equations (the so-called v-conditions). We describe these conditions in the first section below. It turned out (see Veselov [1999]) that these conditions are satisfied not only for the root systems but also for their deformations discovered by O. Chalykh, M. Feigin and the author in the theory of the generalized Calogero-Moser systems (Veselov et al. [1996], Chalykh et al. [1998] and Chalykh et al. [1999]). The corresponding families of the solutions to WDVV equations have the form n 1 " F ^ ^(x,- -;c;)2logte -xy)2 + - J^xflogxf (4) KJ 1 = 1 with an arbitrary real value of the parameter m and n F=kJ2 [(^i + ^jf log (xi + Xjf + (xi - Xjflog (xi - Xjf] + i<j n + Yl i^^i "^ ^n+l)^l0g (Xi + Xn-\-\f + (Xi - X„+i)^log (Xi - Xn-\-\f] + (5) n + 4m ^ xf logxf + 4/x^_^i log x^_^i, where the real parameters k,mj satisfy the only relation ^2/ + l) :=2m + l. (6)
GENERALIZED WDV V EQUATIONS 127 When m = 1 the formula (4) gives the well-known solution to WDVV equations, corresponding to the leading perturbative approximation to the exact Seiberg-Witten prepotential for the gauge group SU(n + 1) (see Marshakov et al. [1996]). For the general m it corresponds to the deformation A„ (m) of the root system An related to the Lie algebra su(n + 1) (see Veselov et al. [1996] and below). The corresponding solution (4) was first found in Marshakov et al. [1997b] (see the formula (3.15) and the calculations after that) although the fact that it is related to the configurations with non-Coxeter geometry seems only to have been realized in Veselov [1999]. The second family of the solutions to WDVV equations (5) was found in Veselov [1999]. When k = m = I = I they correspond to the root system C„+i; in the general case, to its deformation C„+i (m, /) (see Chalykh et al. [1999] and below). In this paper, which is an extended version of Veselov [1999], we present also a new family of solutions which is related to a configuration discovered in the theory of integrable Schrodinger operators by Yu. Berest and M. Yakimov [1997]. It has a form F(xu ...,x^, ji, ...J/), k I F = X^fe - ^jf log i^i - xjf + J2 ^^^^yp - y^^^ ^^s ^yp - y^^^-^ (7) k I ^ ^ + Yl Yl^^^^' ~ yp^^ ^^^ ^^^'" ~ yp^^' for any integer k and / and arbitrary parameter /x. The fact that all the families of the configurations discovered so far in the theory of multidimensional integrable Schrodinger operators satisfy the v-conditions seems to be remarkable and calls for better understanding. 7.2 v-Systems and a Particular Class of Solutions to WDVV Equations It is known (see Marshakov etal [1996,1997]) that WDVV equations (1), (2) are equivalent to the equations FiG-^Fj=FjG-^Fi, /,; = l,...,n, (8) n where G = ^ r]^Fk i^ any particular invertible linear combination of Ft with the k=\ coefficients, which may depend on jc. Introducing the matrices F/ = G~^Fi one can rewrite (8) as the commutativity relations [f,,F,]=0, /,; = l,...,n. (9) We will consider the following particular class of the solutions to these equations. Let y be a real linear vector space of dimension n, V* be its dual space consisting of the linear functions on V (covectors), 21 be a finite set of noncoUinear covectors a G V*.
128 A.R VESELOV Consider the following function on V: F^ = ^(a,jc)2log(a,Jc)^ (10) where (a, x) = a(x) is the value of covector a e V* on a vector x e V. For any basis ^1,... ,en we have the corresponding coordinates jc\ ... , jc" in V and the matrices Ft defined according to (2). In a more invariant form for any vector a e V one can define the matrix Fa = J2'''^i' i=\ By a straightforward calculation one can check that Fa is the matrix of the following bilinear form on V ^ (a, jc) where a 0 P(u, v) = a(u)P(v) for any u,v e V and a, fi e V*. Another simple check shows that G^ which is defined as as F^, i.e. is actually the matrix of the bilinear form G^ = ^a0a, (11) which does not depend on jc. We will assume that the covectors a g 21 generate V*, in this case the form G^ is non-degenerate. This means that the natural linear mapping (p^ : V -^ V* defined by the formula ((^2i(w), v) = G^{u, v), u,v e V is invertible. We will denote (p^^(a), a G V* as a^. By definition Y^a"" (S>a = Id as an operator in V* or equivalently (a,i;) = ^(a,r)(y^,^)- (12) )6g21 for any a e V*,v e V. Now according to (9) the WDVV equations (1,2) for the function (10) can be rewritten as [e^f]-0 (13)
GENERALIZED WDV V EQUATIONS 129 for any a,b e V, where the operators F^ are defined as A simple calculation shows that (15) can be rewritten as 1^ } \ra \ aA/3 = 0, (15) ^^/rr^.9f (a,x)(y6,x) where and a A P =a(Si P - P(SiOi Ba,fi(a, b)=aA P(a, b) = a(a)P(b) - a(b)P(a). Thus the WDVV equations for the function (10) are equivalent to the conditions (15) to be satisfied for any x,a,b e V. Notice that WDVV equations (1,2) and, therefore, the conditions (15) are obviously satisfied for any two-dimensional configuration 21. This fact and the structure of the relation (15) motivate the following notion of the v-systems (Veselov [1999]). Recall first that for a pair of bilinear forms F and G on the vector space V one can define an eigenvector e as the kernel of the bilinear form F — AG for a proper X: (F -XG)(v,x) =0 for any v e V. When G is non-degenerate e is the eigenvector of the corresponding operator F = G-^F: F(e) = G-^F(e) =Xe. Now let 21 be as above any finite set of non-coUinear covectors a e V^, G = G^ be the corresponding bilinear form (11), which is assumed to be non-degenerate, a^ are defined by (12). Define now for any two-dimensional plane IT c V* a form Definition. We will say that 21 satisfies the v-conditions if for any plane U g V^ the vectors a^, a G O U 21 are the eigenvectors of the pair of the forms G^ and G^. In this case we will call ^ a V-system. The V-conditions can be written explicitly as J2 P{cc^)^^^Xa\ (17) for any a g O fl 21 and some A, which may depend on IT and a.
130 A.RVESELOV If the plane 11 contains no more that one vector from 21 then this condition is obviously satisfied, so the v-conditions should be checked only for a finite number of planes 11. If the plane 11 contains only two covectors a and ^ from 21 then the condition (17) means that a^ and y^^ are orthogonal with respect to the form G^: ^(a^)=G^(a^,y6^)=0. If the plane 11 contains more that two covectors from 21 this condition means that G^ and G^ restricted to the plane 11^ c V are proportional: G^ln.-Mn)G^|nv (18) Theorem 7.1. A function (10) satisfies the WDW equations (1) if and only if the configuration 21 is a V-system Proof We have shown above that the function (10) satisfies the generalized WDW equation if and only if the relations (15) are satisfied. Rewriting these relations as ^ G'^(a\nBa,^(a,b) 2^ 7^-T oi A Pka,x)=o = 0 for any a g 21 it is easy to see that they are equivalent to ^ G^(a^,r)^.,^(^,^) ,., for any a G 21 and any two-dimensional plane 11 containing a (cf. Chalykh et al. [1999]). The last conditions are equivalent to J2 G^(a^,r)5c.,M^,^)=0 (19) for any a g 21 and 11 such that a g 11. We would like to show that these relations are actually equivalent to the v-conditions. If n contains only two covectors a and P from 21 then it is obvious since in this case both of these relations are simply saying that G^(a^, fi^) =0. Assume now that 11 contains more than two covectors. We should show that in this case G^ is proportional Gpj after the restriction to 11^. First of all the relations (19) are obviously satisfied if we replace G^ by Gp|: J2 Gl(a'',p'')Ba,p(a,b)=0 (20)
GENERALIZED WDVV EQUATIONS 131 for any a G 21 and any plane 11 containing oc. This follows for example from the fact that in two dimensions any function satisfies the generalized WDVV equations, but can be easily checked in a straightforward way as well. In particular this immediately implies that the V-conditions are sufficient for (10) to satisfy the generalized WDVV equations. To prove that they are also necessary for this let us suppose that this is not the case, i.e. G^ is not proportional GpJ on 11^. Then we can find such a constant c that the restriction of the form G^ - cG"^ onto n^ has a rank 1: G^ - cG^llnv = 6)/ 0 Kinv, 6 = +1 or - 1. for some y g V*. Without loss of generality we can assume that (y, a.^) > 0 for all a G 21. Let a^ be "the very right" vector from 21^ in the half-plane (y, i;) > 0, f G n^ such that ^ao,)S(^' b) = ocq A P(a, b) has the same sign for all y6 G 21. Now put in the relations (19) and (20) a = ao and subtract from the first relation the second one multiplied by c: J2 (G'^-cGlXa^, P'')BaoM^, b) = £(>/, c^o^) ^ (y, ,6^)5,„^(^, b)=0 ;6/ao,)SGnn2l )6/ao,)8Gnn2l Since due to the choice of y and ocq all the summands have the same sign this is possible only if all of them are zero, i.e. (y, fi^) = 0 for all y6 G n PI 21 different from ao- But this contradicts to the assumption that we have more than two noncoUinear covectors from 21 belonging to n. This completes the proof of the Theorem 7.1. 7.3 Examples of v-Systems and New Solutions to Generalized WDVV Equations Let V be now Euclidean vector space with a scalar product (, ), and G be any irreducible finite group generated by orthogonal reflections with respect to some hyperplanes (Coxeter groups, see Bourbaki [1981]). Let 7^ be a set of normal vectors to the reflection hyperplanes of G. We will not fix the length of the normals but assume that IZ is invariant under the natural action of G and contains exactly two normal vectors for any such hyperplane. Let us choose from each such pair of vectors one of them and form the system 7^+: 7^ = 7^+u(-7^+). Usually 7^+ is chosen simply by taking from IZ vectors which are positive with respect to some linear form on V. We will call a system 7^+ a Coxeter system and the vectors from 7^+ as roots. Theorem 7.2. Any Coxeter system 7^+ is a y-system. The proof is very simple. First of all the form (11) in this case is proportional to the euclidean structure on V because it is invariant under G and G is irreducible. By the same reason this is true for the form G n (16) if the plane 11 contains more than two roots from 7^+. When n contains only two roots they must be orthogonal and therefore satisfy v-conditions.
132 A.RVESELOV Corollary. For any Coxeter system 7^+ the function F = J2 (oi,xf\og(a,xf (21) aen+ satisfies WDVV equations (1), (2). When the Coxeter system is a root system of some semisimple Lie algebra this result has been proven in Marshakov et al. [1997b] and Martini [1999]. Notice that even when G is a Weyl group our formula (21) in general gives more solutions since we have not fixed the length of the roots. It is remarkable that the v-conditions are also satisfied for the following deformations of the root systems discovered in the theory of the generalized Calogero-Moser systems in Veselov etal [1996], Chalykh etal. [1998, 1999] and Berest [1997]. To show this let us first make the following remark. One can consider the class of functions related to a formally more general situation when the covectors a have also some prescribed multiplicities /jta /r(2i,M) ^ J2 f^a(a, xf log (a, xf. (22) But it is easy to see that this actually will give no new solutions because F^^'^^ = F^-\- quadratic terms, where 21 consists of covectors y/JI^a. The following configurations An(m) and C„+i(m, /) have been introduced in Veselov et al. [1996], Chalykh et al [1998, 1999]. They consist of the following vectors in R"+i: \ _ f ^/ ~ ^7' 1 < « < y < w, with multiplicity m, " ~ \ei — y/men-\-\, / = 1, ... , n with multiplicity 1, and Iei ±ej, 1 < « < 7 < w, with multiplicity k, let, / = 1, ... , n with multiplicity m, et ± ^fken+i, / = 1,... , n with multiplicity 1, 2\/^^„+i with multiplicity /, whereJb = ^. When all the multiplicities are integer the corresponding generalization of Calogero- Moser system is algebraically integrable, but usual integrability holds for any value of multiplicities (see Veselov ^f^/. [1996], Chalykh ^f^/. [1998, 1999]). Notice that when m = 1 the first configuration coincides with the classical root system of type An and when ^ = m = / = 1 the second configuration is the root system of type C„+i. So these families can be considered as the special deformations of these roots systems. One can easily check that the corresponding sets {^{et - ej), I <i < j <n, et - v^^„+i, / = 1,... ,n
GENERALIZED WDVV EQUATIONS 133 and Cn-\-\(mJ) I y/kei ± y/kej, I < i < j < n 2y/mei, / = 1, ... , n ei ± V^^„+i, / = 1, ... ,n [ 2^/lden-\-\, with k = ^j^ satisfy the v-conditions. To write down the corresponding solution in a more simple form it is suitable to make the following linear transformation: An(m) = et — ej, I < i < j < n, 1 et, / = 1,... , n and C„+i(m,/) f Vk(ei ± ej), I < i < j < n, 2y/mei, / = 1, ... , n et ±^„+i, / = 1, ... ,n where again k = ^^• Now the corresponding functions F have the form (4), (5) written in the Introduction. The third family of solutions (7) is related to the configuration which I would denote as Ak^Ai(fM) Ak^Aiifi) Ci -ej. I <i < j <k. Kfp - fq). I < P <q <l. y f^et - fp, i = 1,... ,k, p = l,, where et and fp are the basic vectors in R^ and R^ correspondingly. This configuration has been discovered by Yu. Berest and M. Yakimov [1997] who were looking for a special "isomonodromial deformation" of the Calogero-Moser problem related to a direct sum of the root systems of types Ak and A/. When the parameter /x = 1 it coincides with the standard Ak-{-i-\-i root system. Again the fact that this family of configurations satisfies the v-conditions can be checked in a straightforward way. As a corollary we have the following Theorem 7.3. The functions F given by the formulas (4), (5), (7) satisfy the generalized WDW equations.
134 A.P.VESELOV It is easy to see that these solutions are really different, i.e. they are not equivalent under a linear change of variables, which is a symmetry of the generalized WDVV equations. Indeed the bilinear form G^ is determined by the configuration 21 in an invariant way and thus induces an invariant Euclidean structure on V and V*. In particular, all the angles between the covectors of the configuration are invariant under any linear transformation applied to the configuration. A simple calculation shows that these angles depend on the parameters of the families and the only case when these configurations have the same geometry is when m = 1 in the first family and /x = ±1 in the third one. In this case we have simply the root system of type A^^. 7.4 Concluding Remarks At the moment there are no satisfactory explanations why the deformed root systems arisen in the theory of the generalized Calogero-Moser problems turned out to be v-systems. It may be that it is a common geometrical property of all the so-called locus configurations (see Chalykh et al. [1999]). In this connection I would like to mention that v-systems can be naturally defined in a complex vector space. Their classification seems to be very interesting and important problem. Another very interesting problem is the investigation of the corresponding almost Frobenius structures related to these systems (see Dubrovin [1992, 1993]). Dubrovin discovered some very interesting duality in the Coxeter case with the Frobenius structures on the spaces of orbits of Coxeter groups (Dubrovin [1993, 1999]). The natural question is whether this can be generalized for the non-Coxeter v-systems and what are the corresponding dual Frobenius structures. Acknowledgements I am grateful to M. Feigin for the useful comments on the preliminary version of this paper, to Yu. Berest for the suggestion to look at the configurations he discovered with M. Yakimov and to B. Dubrovin for the inspiring discussions at MSRI, Berkeley in February 1999. I am grateful also to A. Marshakov who attracted my attention to a very important paper by Marshakov et al. [1997b] where in particular one of the families of our solutions (4) has been first discovered. References Berest, Yu., and Yakimov, M., Private communication (October 1997). Bourbaki, N., Groupes et algebres de Lie. Chap. VI, Masson, 1981. Chalykh, O.A., Feigin, M.V., and Veselov, A.P., Multidimensional Baker-Akhiezer Functions and Huygens' Principle. Commun. Math. Physics, 206, 533-566 (1999). Chalykh, O.A., Feigin, M.V., and Veselov, A.P, New integrable generalizations of Calogero-Moser quantum problem. J. Math. Phys. 39(2), 695-703 (1998).
GENERALIZED WDVV EQUATIONS 135 Dijkgraaf, R., Verlinde, E., and Verlinde, H., Notes on topological string theory and 2D quantum gravity. Nucl Phys. B 352, 59 (1991). Dubrovin, B., Geometry of 2D topological field theories. In: Springer Lecture Notes in Math. 1620, 120-348 (1996). Dubrovin, B., Private communication (February 1999). Dubrovin, B., Differential Geometry of the space of orbits of a Coxeter group, hep-th/9303152 (1993). Marshakov, A., Mironov, A., and Morozov, A., More evidence for the WDVV equations in N=2 SUSY Yang-Mills theories, hep-th/9701123 (1997). Marshakov, A., Mironov, A., and Morozov, A., WDVV equations from algebra of forms. Mod. Phys. Lett. All, 773, hep-th/9701014 (1997a). Marshakov, A., Mironov, A., and Morozov, A., WDVV-like equations in A^ == 2 SUSY Yang-Mills theory. Phys. Lett. B 389, 43-52, hep-th/9607109 (1996). Martini, R., and Gragert, RK.H., Solutions of WDVV equations in Seiberg-Witten theory from root systems. J. of Nonlinear Math. Physics 6(1), 1-4 (1999). Seiberg, N., and Witten, E., Electro-magnetic duality, monopole condensation, and confinement in A^ = 2 supersymmetric Yang-Mills theory. Nucl. Phys. B 426, 19-53 (1994). Veselov, A.R, Deformations of the root systems and new solutions to generalized WDVV equations. Phys. Lett. A. 261, 297-302 (1999). Veselov, A.R, Feigin, M.V., and Chalykh, O.A., New integrable deformations of quantum Calogero- Moser problem. Usp. Mat. Nauk 51(3), 185-186 (1996). Witten, E., Two-dimensional gravity and intersection theory on moduli space. Surv. Diff. Geom. 1, 243-210(1991).
8 Picard-Fuchs Equations, Hauptmoduls and Integrable Systems J. HARNAD Department of Mathematics and Statistics, Concordia University, 7141 Sherbrooke W., Montreal, Que., Canada H4B 1R6, Centre de recherches mathematiques, Universite de Montreal, C P. 6128, succ. centre ville, Montreal, Que., Canada H3C 3J7, E-mail: hamad@ crm. umontreal. ca The Schwarzian equations satisfied by certain Hauptmoduls (i.e., uniformizing functions for Riemann surfaces of genus zero) are derived from the Picard-Fuchs equations for families of elliptic curves and associated surfaces. The inhomogeneous Picard-Fuchs equations associated to elliptic integrals with varying endpoints are derived and used to determine solutions of equations that are algebraically related to a class of Painleve VI equations. 8.1 Differential Equations for Modular Functions There are a number of differential systems whose general solutions may be expressed in terms of modular functions. An example is the Darboux-Halphen system (see Halpen [1881a,b,c]), w[ = W\(W2 + W^) — W2W3 w'2 = W2{Wi -\r W3) - W\W3 (1) w;3 = W3(wi + W2) - W1W2, which was already thoroughly studied in the last century, but recently has recurred in several applications in mathematical physics (see Gibbons and Pope [1979], Atiyah and Hitchen [1988], Chakravarty et al. [1990], Tod [1994], Hitchin [1995] and Dubrovin [1996]). Its 137
138 J. HARNAD symmetrization under the symmetric group in three variables ^3 gives the Chazy equation [1910] W'" = VNW" - ?>W'^, (2) whose solutions are related to those of (1) by W = 2{wi+W2 + w^). (3) The physical contexts in which these equations have appeared include: 1. the dynamics of magnetic monopoles pairs (Atiyah and Hitchin [1988]). 2. homogeneous, 50(3) invariant solutions of self-dual Einstein equations (Gibbons and Pope [1979], Tod [1994] and Kitchen [1995]). 3. solutions of the WDVV equations in topological field theory (Dubrovin [1996]). The general solution to (1) was determined in 1881 by Halphen [1881a,b,c] and Brioschi [1881] in terms of the elliptic modular function X{z) = k^T), (4) where ^(t) is the elliptic modulus, viewed as a function of the ratio t of the elliptic periods. A particular solution to (1) is given by 1 d X' \ d ^ X' w\\=-—-m—, w;2 •=--;—In 2dz X' 2dz {X-\y \ d ^ X' ^^ wx := In . 2dz X{X-\) The general solution is obtained by composing this with a general Mobius transformation (-) Similarly, a particular solution to (2) is given by T:x-^ ?7T^ = ^^''^' I " ^ I ^ "^^(2, C). (6) ^^=^-'-^i^"7^(7!^IF' ^'^ where / denotes Klein's /-function _ 4(^2 - A + 1)3 27X2(A, - 1)2 ' (8) and the general solution is again obtained by composing / with a Mobius transformation (6).
PICARD-FUCHS EQUATIONS, HAUPTMODULS AND INTEGRABLE SYSTEMS 139 The key element in deriving these solutions is to first note that the period integrals, when viewed as functions of A, are hypergeometric functions /■ Jo f Jo dt IT (9) , dt in iK2 = / , = ^Fa,{; 1; 1 - A), and hence satisfy the hypergeometric equation of Legendre type d^y dy 1 Since X(t) is the inverse of the function t(X) given by the ratio of the periods, it follows that it satisfies the Schwarzian equation A A -[- 1 ,2 where is the Schwarzian derivative (see Gerretson and Sansone [1969]). This implies that (5) defines a solution of (1) and, by the SL(2, C) invariance of the system, composition with the Mobius transformations (6) gives the general solution. Similarly, the fact that (7) determines the general solution to (2) follows from the Schwarzian equation satisfied by / which is obtained from (11) by composing with (8). This latter Schwarzian equation equation is analogously related to the hypergeometric equation J(l-jA^C--lj)^-^y = 0. (14) Thus / is the inverse of the function obtained by taking the ratio of two linearly independent solutions of (14). The hypergeometric equations (10) and (14) may both be viewed as examples of Picard-Fuchs equations for elliptic pencils; that is, as Fuchsian differential equations determining the variation of elliptic integrals over an affine parametric family of elliptic curves. In each of these cases, the relevant inverse function is a modular function. This follows from the fact that the projectivized monodromy groups for the associated hypergeometric equations (10) and (14) are both commensurable with the modular group r := PSL(2, Z). For the case (10), this gives the principal congruence subgroup r(2), which is the automorphism group of the modular functions A(t), while for (14), it is the full modular group, the automorphism group of /(r).
140 J. HARNAD There is another sense in which Picard-Fuchs equations may be related to nonlinear equations of interest in mathematical physics; namely, the class of isomonodromic deformation equations, such as the family of Painleve equations Py/(a, P, y, 8) X(X-i)(X-t) / pt^ y(t-i) 8t(t-i)\ (15) It was shown in the work of Hitchin [1995], Dubrovin [1996] and Mazzocco [1998] that solutions to particular cases of (15) with special values of the parameters (a, fi, y, 8) could be given in terms of solutions to (1). For example, for the case Pyi{oc = 2, fi = 0,y = 0, 5 = 5). we have the one-parameter family of Chazy solutions given by (W2W3 - WiW2 - WiWsf ^ ^ —;; ~r ^—' (i^>^ 4w\W2U}3{'W\ — W3) where the independent parameter is taken as w\ — W3 t:=— -=X, (17) W2 — W3 and (w;i, W2, W3) is a general solution to (1). (Only one of the three SL(2, C) parameters introduced by (6) is effective.) More generally, a sequence of Chazy-type solutions for parameter values a = \{2^i — 1)^, M + 5 G Z\l are derived in Mazzocco [1998] by application of discrete symmetry transformations. Another class of solutions to (15) for the case (a = 0, y6 = 0, y = 0, 5 = ^) was already known to Picard [1889], who expressed them in terms of elliptic integrals with variable end-points. This allows us to relate this case to an inhomogeneous Picard-Fuchs equation. Namely, consider the 1-parameter family of elliptic curves / ^ 4;^(;^ _ l)(;^ _ X) , (18) and corresponding period integrals: /oo y ' If the elliptic integral with varying (A-dependent) endpoints .{X{k),Y{k)) j^ y ^1 = / —, K2 = Joo y Joo g (A.-depe Joo (19) (20) is required to satisfy the same hypergeometric equation as do the period integrals Ki, K2, i.e., if it is set equal to a linear combination K = AKi + BK2, (21)
PICARD-FUCHS EQUATIONS, HAUPTMODULS AND INTEGRABLE SYSTEMS 141 it follows that X = X(X) satisfies Pvi(oi = 0, fi = 0,y = 0, 5 = ^), providing a 2-parameter family of solutions. More generally, Fuchs [1907] showed that Pyi (a, P, K, 5) is equivalent to the inhomogeneous Picard-Fuchs equation d^K dK 1 Ml-X)^+ (1-2.)---/. X{X Y ( , px' , yQ--1) , /, i\Mi-x)\ (A more recent perspective on such equations and their algebro-geometric meaning may be found in Manin [1998]). In the following sections a number of further examples of modular functions having similar properties will be considered. These all provide solutions to certain associated systems of nonlinear differential equations whose origins may be traced to Picard-Fuchs equations for families of elliptic curves. In each case, an associated inhomogeneous Picard-Fuchs equation may also be determined, and shown equivalent to an equation that is algebraically related to the Picard case of Pyi. 8.2 Generalized Halphen Equations 8.2.1 Triangular Cases Halphen [1881a,b,c] also considered generalizations of the system (1) related to the general hypergeometric equation d^y dy f(^-f)-7^-^(c-(a-^b-^ l)f)^ - aby = 0. (23) Assuming the inverse function of the ratio of two linearly independent solutions of (23) to exist (which by no means is always the case in a global sense, in view of the infinite-valued multiplicity of the solutions of (23) due to monodromy), it also satisfies a Schwarzian equation of the form {/,r} + 2/?(/)/2 = 0, (24) where with the parameters (A, /x, y) defined by A := 1 — c, iJi\=c — a — b, v:—b — a. (26) The generalized Halphen-Brioschi variables W\ \— In —, W2 := In 2dT /' " 2dz (/-I)' 1 d f W3:= In -^ (27) 2dT /(/-I)
142 J. HARNAD then satisfy the general Halphen system: W[ = Wi(W2 + W3) - W2W3 + X W2 = W2(Wi + W3) -WiW3-^X W; = W3(Wi + W2) -WiW2-\- X, (28) + 0} - /x^ - v^W3W\ + (/x^ -1} - y^)W2W3. A sufficient condition for having a well-defined inverse function /(t) is that the projectivized monodromy group of the hypergeometric equation (23) be a Fuchsian group of the first kind. This essentially means that it acts properly discontinuously and there is a tesselation of the image space by fundamental domains with a finite number of vertices. In the present case, the domains are necessarily triangular (with circular arcs as sides), since the vertices must map to the three regular singular points (0, 1, 00). (The more general case, with n singular points is considered in the next subsection.) The projectivized image of the monodromy representation is just the automorphism group of the function. The parameters (A, /x, y) are the fractions of tt giving the angles at the vertices of the fundamental domain. In general, the automorphism group 0/ of a function / (t) (under Mobius transformations (6)) is said to be commensurable with the modular group F if it is a subgroup of PGL(2, Q) whose intersection with F is of finite index in both F and 0/. Such a function and its automorphism group will be referred to as modular. By a HauptmoduU we understand a uniformizing function for a genus zero Riemann surface; that is, the quotient H/0/ of the upper half r-plane by the automorphism group defines a genus zero Riemann surface. The fact that such a function is the sole generator of the field of meromorphic functions on the Riemann surface implies that / must satisfy a Schwarzian equation of the type (24) for some rational function R{f). A special class of such Hauptmoduls, referred to as replicable functions (due to their replication properties under the action of generalized Hecke operators (see Conway and Norton [1979])), have the additional property of containing a finite index subgroup of the type -l(:2) Fo(A^) := { ,\ eSL{2,Z), c = OmodN\, (29) (Such functions arise in connection with "modular moonshine", either as character generators for the Monster sporadic group, or in relation to these through the generalized Hecke averaging procedure (see Conway and Norton [1979] and Ford et al. [1994])). Each such function, of which there are only a finite number, is analytic in the upper half-plane, has a finite number of vertices in its fundamental domains, a cusp at 00 and admits, up to an affine transformation, an expansion as a normalized McKay-Thompson ^-series 1 "^ F(q) = af(T) + P = - +Yanq\ q := ^''^^ (30) n=\ convergent in the upper half-plane. (For the cases considered here, we also have an G Z.)
PICARD-FUCHS EQUATIONS, HAUPTMODULS AND INTEGRABLE SYSTEMS 143 TABLE 8.1. Triangular Replicable Functions. Name (a,b,c) (k,(ji,v) Po /(i") r Vl2'12'3/ V3'2' / Vl -1/ SAi^^i^^i^l Vl2' 12' 3/ \3' 2'7 VI -1/ al^) ni,o) r-') V8' 8'4/ V4 2' / V2 -2/ /I 1 5\ /I 1 \ /O -IX (,^^(r)+27,^^(3r))^ \6'6'6/ V6'2' / \3 -3/ 108r;i2(T)r;i2(3T) 'ro(2) V4'4'2j V2' ' / \2 -l) ^ (<(r)+z^:(r)) ^ro(3) V3'3'3/ V3' / \3 -l) 2l\r){3x)f ^ro(4) V2 2 ; V4 -,) =l^Um) 2a (- i ?^ ("i i 0^ ("^ "^^ x/3/(^--'/3^3^(2t)-i>,^(2t))' 4a (- - l\ (- ^ 0] ("^ ~^\ i {^li2T) + i^li2T)y V4'4'47 V4'4' / VS -8/ 82>4(2t)2>32(2t)2>2(2t) /I 1 5\ /^^ ^ \ { ^ ~'^ \ V3i{r]H2T)-\-3V3ir]H6T)f ^^ V3'3'6/ V6'6' / V12 -12/ 36r]^(2T)r]^{6T) The table above (which is taken from Hamad and Mackay [1998]) contains a complete list, up to equivalence under affine transformations in the t variable, of the replicable functions of triangular type (i.e., those for which the fundamental domain has three vertices). These coincide with the arithmetic triangular functions of noncompact type (see Takeuchi [1977]). Through formulae (27), they provide solutions to the general Halphen systems (28). In this table, the first column gives the labelling according to the notation of Conway and Norton [1979] and Ford et al. [1994], consistent with the finite group atlas. The second and third columns give the hypergeometric parameters (a,b,c) and (A, /x, y). The fourth column contains the generator of the automorphism group which stabilizes a vertex mapping to 0. The corresponding generator stabilizing the cusp at 00 is Poo = (q j), (31)
144 J. HARNAD and the third generator p\, stabiHzing a vertex mapping to 1 is determined by the relation PooP\Po=l' (32) The last column in Table 8.1 gives explicit expressions for the Hauptmoduls in terms of null theta functions ?^2(^), ?^3(^), ?^4(^) or the Dedekind eta-function rjir). In section 8.3, it will be shown how the corresponding hypergeometric equations, which imply the Schwarzian equation (24), and hence provide solutions to the generalized Halphen equations for the triangular cases, may be derived as Picard-Fuchs equations for families of elliptic curves. But first we consider some further generalizations of the Halphen system (1) corresponding to second order Fuchsian equations with more than three regular singular points. Such generalized systems were introduced by Ohyama [1997]. 8.2.2 Generalized Halphen Systems with n Singular Points Consider second order Fuchsian equations of the form d2_y df -^ + RU)y = 0, (33) where /?(/) is a rational fiinction of the form A^(/) (£>(/))2' ^(^> = 7^' o(/) = !!(/-«'•). (34) / = 1 and A^(/) is a polynomial of degree < 2n — 2. (Any second order Fuchsian equation is projectively equivalent to one of this form; i.e., it may be transformed to this form, with no first derivative term, by multiplication of the solutions by a suitably chosen function.) Let t(/) again denote the ratio r(/) :- - (35) of two linearly independent solutions of (33), and suppose again that the inverse function / = /(t) is well-defined. It then satisfies the Schwarzian differential equation {/,r} + 2/?(/)/2^0, (36) and conversely, at least locally, all solutions of the Fuchsian equation (33) are expressible as: (A + ^r(/)) y = \ • (37) {TV The image of the monodromy representation M : 7ri(P- {a\, ...an, M::^i(P-{ai,...a„,oo}) -^ GL(2,C) a b\ (38)
PICARD-FUCHS EQUATIONS, HAUPTMODULS AND INTEGRABLE SYSTEMS 145 defined up to global conjugation by K:(Ji,J2)I/o = (ji,J2)I/oM^, (39) determines a subgroup 0/ C GL(2, C) that acts on t by Mobius transformations (6), leaving /(t) invariant. Introducing the new variables (see Ohyama [1996] and Hamad and Mckay [1998]) u:=Xo = -'—. Vi:=-(Xo-Xi) = --!—, (40) 2 f 2 2 f -at these satisfy the set of quadratic constraints (at - aj)viVj + (aj - ak)vjVk + (ak - ai)vkVi = 0, (41) and the differential equations: V- = -2vf + 2uvi, / = 1,... n (42) n u' = U^ — 2_. ^ij'^i'^j^ (43) where the quadratic form Yl^ j=\ ^ij'^i'^j appearing (43) is determined by expressing of the rational function /?(/) in the form 1 " 4 .^j (f - ai)(f - aj) (There is a nonuniqueness in such expressions for /?(/), but this just corresponds to the freedom of adding any linear combination of the vanishing quadratic forms (41) to the right hand side of Eq. (43) In this case, if the automorphism group of / is again Fuchsian, the angles {a/7r}/=i „ at the finite vertices of the fundamental polygon are related to the diagonal part of the quadratic form by rii = l-af. (45) Table 8.2 below, which is a shortened version of one appearing in Hamad and Mckay [1998], again contains a list of Hauptmoduls that are replicable functions. But in these cases, their fundamental domains have four vertices, and hence there are four generalized Halphen variables (w, v\, V2, fs) appearing in Eqs. (42), (43). The first column again identifies the functions and their groups according to the notation of Conway and Norton [1979] and Ford etal [1994], the second gives the values of /(t) at the finite vertices (i.e., the location of the poles of R(f)), and the third lists the generators of the automorphism group of / corresponding to these vertices (i.e., the projectivized monodromy group generators). The fourth column gives the quadratic form appearing in (43), and serves to define the rational function R(f) through (44). The last colunm lists explicit formulae for /(t) in terms of the Dedekind ?;-function. This table contains all the geometrical and group theoretical data characterizing the Hauptmoduls listed. A similar set of data may be determined for all the replicable functions appearing in Conway and Norton [1979] and Ford et al. [1994], and from this the corresponding Schwarzian equations and generalized Halphen equations may be deduced. (It should be noted that the number of vertices for the corresponding fundamental domains never exceeds 26, and that there are algebraic relations interlinking all these various cases.)
146 J. HARNAD TABLE 8.2. Four Vertex Replicable Functions. Name P\ {ax ,^2,^3) p2 2_^ ^'J^'^J /(^) - 1 P3 iJ=^ 6C (-3,0,1) 6D (ro(8)) -I+V2- 6E 1 (ro(6)) ' 8' (-1,0,1) 95 {o),^,l) (ro(9)) co-e"^ (I a {-: c ii (1 (,', (« {-: ii c (-1 ii ii {-', ID :0 -n :n :D -:) :?) :?) -,°) :D :;) -:) ::) ::) .:) Iv^ + lvl + vj -\V2VJ -V,V3 Iv^ + lvj + vj -mMVky,y^ v\ + v\ + v\ v\^v\^ v\ -V\V2 — (1 — (i>)V\V^ — (1 — (JL>)V2V'i, 1 r;6(T)r;6(3T) 4 ?;6(2t)?;6(6t) 1 r;4(T)r;4(2T) 4 ^\^x)^\(,x) 1 ^'(T)r7(3T) 8 r;(2T)r;5(6T) 1 r]\x)^^{^x) 4 r;2(2T)r;4(8T) 1 r]\x) 3 ?;3(9t:) 8.3 Picard-Fuchs Equations on Elliptic Families In this section, the parametric families of elliptic curves whose associated Picard-Fuchs equations underlie the Schwarzian equations governing these Hauptmoduls will be given for three of the examples appearing in the tables above. Only those cases are treated which are actually subgroups of the full modular group F, but an indication will be given at the
PICARD-FUCHS EQUATIONS, HAUPTMODULS AND INTEGRABLE SYSTEMS 147 end of this section how the other cases may be similarly derived. (A more complete version of these results will appear elsewhere Hamad [1999].) 8.3.1 Arithmetic Triangular Subgroups of F Four of the cases appearing in Table 8.1 involve automorphism groups that are contained in the full modular group; these are: lA -> T, 25 -- ro(2), 3B -> ro(3) and 4C -> To (4) ~ r(2). For each of these, it is possible to find a 1-parameter family of elliptic curves for which the associated Picard-Fuchs equation gives the required hypergeometric equation. We illustrate this below for the two cases: lA ~ F and 2B ~ ro(2). The first of these leads to the hypergeometric equation (14) associated with the /-function. (The case 4C ~ To (4) ~ r(2) gives the corresponding Legendre hypergeometric equation (10) associated with A, and hence the original Halphen system.) 1. Hauptmodul I A. The associated family of elliptic curves is given by the elliptic pencil j2 = 4(^ _ i);^3 _2ax-a (a = J). (46) Denote by K and L the elliptic integrals of the first and second kind. Joo y Joo (47) where we allow the endpoints to depend upon the parameter a. Differentiating these with respect to a, taking the endpoint contributions into account, we deduce the inhomogeneous Gauss-Manin system ^, (5^ - 2)K _ ^ ^ a + {2 + a)X + 2{\-a)X^ r 12^(^ - 1) 6a 6a(a - 1)Y Y K + {Ua + 4)L _ -^ - ^X + (4 + 2a)X^ XX' ^ 24a(a - 1) 12^(^ - l)Y "^ ~1^* (Usually, the term "Gauss-Manin system" refers to the equations satisfied by the corresponding differential cohomology classes, but here we consider integrals along a suitable path, with varying endpoints.) If instead of taking variable endpoints, we integrate around a cycle, the right hand sides of Eqs. (48), (49) vanishes, and we have the more usual homogeneous Gauss-Manin system. Eliminating the L integral from this system gives the inhomogeneous Picard-Fuchs equation „ (2a - 1) , (36a^ - Ala - A) 1 . „ . ,9 . , . (50)
148 J. HARNAD where ^.(X):='^<-'>f-^'- ,51, + (60^ - Sla^)X^ + (16 - 140^ + 124^2)X^^ (53) Y^ := 4(a - 1)X^ - 3aX - a. (54) Taking the path defining ^ to be a cycle, the right hand side of (50) disappears and we obtain the homogeneous Picard-Fuchs equation, which is projectively equivalent to the hypergeometric equation (14) under the identification a = J. (More precisely, the function a^(a — 1)4^, taken over a generating pair of cycles gives a basis of solutions of (14).) Setting the integral K with variable endpoints equal to a linear combination K = AKi + BK2, (55) where Ki and K2 denote the values of the integral taken over a basis of cycles, amounts to choosing X(a) as an elliptic function, defined as the inverse of the elliptic integral K, with its argument taken as AK\ + BK2. This implies that the right hand side of of (50) disappears, giving an equation for X(a) of the same type as Picard's case of Pyi- (In fact, the two are algebraically related through through (8).) 2. Hauptmodul 2B. The associated family of elliptic curves in this case is given by the elliptic pencil y^ = 4x^ - 3(1 + 3a)x + 9^-1. (56) (To be precise, the Hauptmodul f2B normalized as in Table 8.1 corresponds to the inverse ^ of the parameter appearing in (56).) Denoting again by K and L the elliptic integrals of the first and second kind, respectively, defined as in (47), and differentiating with respect to the parameter, we obtain the corresponding inhomogeneous Gauss-Manin system: , (3a -\)K-2L 1 + 3^ + (1 - 3a)X - 2X^ X' T^' I _i 1 2 _L. rs7^ 12^(^ - 1) 6a(a - l)Y Y ^ ^ , (l + 3^)^ + (2-6^)L _ 1 - 9^ + (1 + 3^)X - (2 - 6^)X2 XX' "^ 24^(^ - 1) ~ 12^(^ - 1)F "^ ~Y~ .^g. and the resulting inhomogeneous Picard-Fuchs equation of type 2B ^" + ^7^^' + 77-7^^ = ^ i^" - -42(X)X'2 - AiiX)X' - Ao(X)), a{a — 1) l6a{a — 1) Y ^ ' (59)
PICARD-FUCHS EQUATIONS, HAUPTMODULS AND INTEGRABLE SYSTEMS 149 where 1 2 4- 4X -^^^^ •= 8a(a-l)(l-9a + 4X+4X2)- ^^^^ Again, if the integrals are taken over cycles, the right hand side of (59) vanishes, and the independent variable transformation a -^ ^ gives an equation that is projectively equivalent to the hy pergeometric equation satisfied by F{\, \, \\ ^). Choosing K once again to equal a linear combination of the period integrals as in (55); i.e., expressing X{a) again as an elliptic function of this argument, this defines a 2-parameter family of solutions to the equation obtained by equating the right hand side of (59) to zero. (This again is algebraically related to the Picard type solutions of Py/.) A further class of examples is provided by the 4-vertex cases listed in Table 8.2 that correspond to subgroups of P. Up to projective equivalence, these are 6E ~ To(6), ^E' - ro(8)' - ri(4) n r(2), and 9B' - ro(9)' - r(3). (The primes ' in this notation just denote composition with transformations of the type t -^ t/2 or t ^- t/3, which do not affect the resulting Schwarzian equations.) These are all cases of elliptic pencils corresponding to Beauville's elliptic surfaces (see Beauville [1982]). The only case of Beauville's surfaces that does not appear in Table 8.2 is the one corresponding to Fi (5), which does not give a replicable function since this contains no Tq{N) subgroup. Nevertheless, it can be can similarly dealt with. As an illustration, here are the details in the case of the Hauptmodul with automorphism group To (8). The elliptic pencil is defined by x^ + 2xy + a{x^ - j^) + jc = 0. (63) In this case, the elliptic integrals of first and second kinds are _ r^-') dx r^^''> xdx Joo x-ay Joo X -ay' (64) The inhomogeneous Gauss-Manin system is aK + L _ (l+a^)X + 2aX2 X' ^ a^-\ a{a^-\){X-aY)^ X-aY ^ ' ^ aK-^L 2aX + (l+a^)X^ XX' a{a^-V)~ a{cfi-l){X-aY) X-aY' and the inhomogeneous Picard-Fuchs equation is (3a2 - 1) , 1 _ 1 a(a2 - 1) ^ "*" (a2-l)^"" X-aF ^" + f,!',!!^' + 712^^ = ^F^ i^" - ■42(X)X'2 - ^,(X)X' - AoiX)), (67)
150 J. HARNAD where A .v^ a + 2(l+a2)X + 3flX^ ^'^^^ '-^ 2(aX3 + (l+a2)X2+aX) ^^^^ _ 2a3-(l-4a2-a4)x + 2a3x2 •^'^^•~ a(a2-l)(aX2 + (l+a2)X + a) ^ ^ X(X^ — 1) 2a{aX^ + (1 + ^ )a + a) The associated Fuchsian equation for this case is the corresponding homogeneous Picard- Fuchs equation, and the result of choosing the argument of the elHptic function defining X(ci) as in (55) again defines a Picard type solution of an equation algebraically related to We conclude with the following remarks. 1. The nonlinear monodromy (Dubrovin [1996] and Mazzocco [1998]) of the Picard-type solutions of these Py/-like equations coincides in each case with the monodromy of the associated homogeneous Picard-Fuchs equation, since it just involves a linear transformation of the coefficients in the linear combination (55), giving the argument of the elliptic function that defines the solution X{a). 2. The Fuchsian and associated Schwarzian equations governing the other cases of Hauptmoduls, which do not involve subgroups of F, may also be derived from Picard- Fuchs equations, since they are obtained by taking extensions of the automorphism groups by Atkin-Lehner involutions (Conway and Norton [1979] and Ford etal [1994]). Instead of considering families of elliptic curves, however, one must consider families of surfaces obtained essentially by taking the topological products of the pairs curves involved. (Further details will be provided in Hamad [1999].) 3. The various Py/-like equations that arise are all related by the same algebraic transformations that connect the corresponding homogeneous Picard-Fuchs (and Schwarzian) equations (see Hamad and Mckay [1998]). From the viewpoint of generalizations and applications of integrable dynamical systems involving such modular functions, it seems natural to try to extend these considerations to families of higher genus curves, and also to determine whether analogs of the Chazy solutions exist, which might provide new cases of Frobenius manifolds. Acknowledgements I would like to thank J. McKay, J. Hurtubise and C. Doran for helpful discussions relating to this material. The results quoted in Section 8.2, in particular the contents of Tables 8.1 and 8.2, are taken from Hamad and Mckay [1998], where the associated dynamical systems and the algebraic relations between the cases are developed in greater detail. This research was supported in part by the Natural Sciences and Engineering Research Council of Canada and the Fonds FCAR du Quebec.
PICARD-FUCHS EQUATIONS, HAUPTMODULS AND INTEGRABLE SYSTEMS 151 References Atiyah, M.F., and Hitchin, NJ., The Geometry and Dynamics of Magnetic Monopoles, Princeton University Press, Princeton (1988). Beauville, A., "Les families stables de courbes elliptiques sur P' admettant quatre fibres singulieres, C R. Acad. Sci. Paris 294, 657-660 (1982). Brioschi, M., "Sur un systeme d'equations differentielles", C.R. Acad. Sci. Paris 92, 1389-1393 (1881). Chakravarty, S., Ablowitz, M.J., and Clarkson, PA., "Reductions of Self-Dual Yang-Mills Fields and Classical Systems", Phys. Rev. Let. 65, 1085-1087 (1990). Chazy, J,, "Sur les equations differentielles dont I'integrale generale possede une coupure essentielle mobile", C. R. Acad. Sc. Paris, 150, 456-^58 (1910). Conway, J., and Norton, S.P, "Monstrous moonshine". Bull. Land. Math. Soc. 11, 308-339 (1979). Dubrovin, B.A., "Geometry of 2D topological field theories". Lecture Notes in Math. 1620, Springer-Verlag, Berlin, Heidelberg, New York (1996). Ford, D., McKay, J., and Norton, S., "More on replicable functions" Comm. in Algebra 22,5175-5193 (1994). Fuchs, R., "Uber lineare homogene Differentialgleichungen zweiter Ordnung mit im endlich gelegene wesentlich singularen Stellen." Math. Ann. 63, 301-321 (1907). Gerretson, J., and Sansone, G., Lectures on the Theory of Functions of a Complex Variable. II. Geometric Theory. Walters-Noordhoff, Groningen (1969). Gibbons, G.W,, and Pope, S.N., "The Positive Action Conjecture and Asymptotically Euclidean Metrics in Quantum Gravity", Commun. Math. Phys. 66, 267-290 (1979). Halphen, G.-H., "Sur des fonctions qui proviennent de I'equation de Gauss", C.R. Acad. Sci. Paris 92, 856-858 (1881a); "Sur un systeme d'equations differentielles", ibid. 92, 1101-1103 (1881b); "Sur certains systemes d'equations differentielles", ibid. 92, 1404—1406 (1881c). Hamad, J., and McKay, J., "Modular Solutions to Equations of Generalized Halphen Type", to appear Proc. Roy. Soc, preprint CRM-2536 (1998), solv-int/98054006 Hamad, J., "Picard-Fuchs equations for elliptic families and Schwarzian equations for Hauptmoduls", (CRM preprint (1999), in preparation). Hitchin, N., "Twistor Spaces, Einstein metrics and isomondromic deformations", J. Diff. Geom. 42, 30-112(1995). Manin, Y., "Sixth Painleve equations, universal elliptic curve and mirror of P^", Bonn preprint (1998), alg-geom/9605010. Mazzocco, M., "Picard and Chazy Solutions to the Painleve VI Equation", preprint SISSA 89/98 (1998). Ohyama, Y, "Systems of nonlinear differential equations related to second order linear equations", Osaka J. Math. 33, 927-949 (1996);"Differential equations for modular forms with level three", Osaka Univ. preprint (1997). Picard, E., Memoire sur la Theorie des Functions Algebriques de deux Varables, Journal de Liouville 5, 135-319(1889). Takeuchi, K., "Arithmetic triangle groups", J. Math. Soc. Japan 29, 91-106 (1977). Tod, K.P, "Self-dual Einstein metrics from the Painleve VI equation", Phys. Lett. A190, 3-4 (1994).
9 Painleve Type Equations and Hitchin Systems M.A. OLSHANETSKY Max-Plank-Institut fur Matematiky Bonn, Institut of Theoretical and Experimental Physics, Moscow, Russia E-mail: olshanet@ mpim-bonn. mpg. de; olshanet@ heron, itep. ru In this survey we present the interpretation of isomondromy preserving equations on Riemann surfaces with marked points as reduced Hamiltonian systems. The upstairs space is the space of smooth connections of GL(N) bundles with simple poles at the marked points. We discuss relations of these equations with the Whitham quantization of the Hitchin systems and with the classical limit of the Knizhnik-Zamolodchikov-Bemard equations. The main example is the one-parameter family of Painleve VI equation and its multicomponent generalization. 9.1 Introduction The famous Painleve VI equation depends on four free parameters (PVIa,p,y,s) and has the form d^X dt^ 1/1 1 1 \ fdxy n 1 1 \ dx 2\x^ x-\ ^ x-t) ydTJ ~ V ^ r-i "^ x-t) ~dt^ , X{X-\){X-t) ( ^.t t-\ t(t-\)\ (1) It was discovered by B. Gambier [ 1910] in 1906. He accomplished the Painleve classification program of the second order differential equations whose solutions have no movable critical points. This equation and its degenerations PV — PI have a lot of applications in classical and quantum integrable systems (see, for example NATO [1990]), topological field theories (Dubrovin [1992]), general relativity (Hitchin [1995] and Korotkin et al. [1998]), and in the Seiberg-Witten theory (D'Hoker et al. [1998]). In this paper we discuss two important and interrelated aspects of PVI: 153
154 M.A. OLSHANETSKY • PVI and isomonodromic deformations of linear differential equations; • The Hamiltonian structure of PVI. The derivation of the PVI equation as the monodromy preserving condition was given by Fuchs [1907], while the Hamiltonian structure of PVI was introduced by Malmquist [1992-3]. We incorporate the one-parameter family PVI^ _yL )^ ^_yL =PVIy in a wide class of 4 ' 4 ' 4 ' 2 4 nonlinear equations. They preserve monodromies of systems of linear equations on Riemann curves with marked points when the complex structures of curves are changed. These systems come from the flatness condition of vector bundles over the curves. We restrict ourself by considerations of smooth connections with simple poles only, and therefore don't include the Stokes phenomena. In such general form the isomondromy preserving equations were considered in Iwasaki [1992]. Our investigation of these systems is inspired by methods developed in classical and quantum integrable systems. In general all the systems can be derived using three different constructions: 1. The symplectic reduction procedure from free infinite dimensional theory. This approach is very similar to the derivation of the Hitchin integrable systems (Hitchin [1987]); 2. The Whitham quantization of the Hitchin systems (Krichever [1994]); 3. Classical limit of the Knizhnik-Zamolodchikov-Bemard (KZB) equations (Knizhnik etal [1984] and Bernard [1988]). We discuss these constructions separately and then demonstrate their application on the multicomponent generalization of PVIy. The presentation is based for the most part on a previously published paper (Levin et al. [1999]). First, in Section 9.2 we consider the elliptic form of PVI. In this form the relations of PVI with the Hitchin systems and KZB equations become transparent. Then we discuss these three approaches to the isomonodromic deformations. In Section 9.6 the multicomponent generalization of PVI is described. Finally, we discuss some open problems related to PVI and its generalizations. 9.2 Elliptic Form of PVI 9.2.1 Elliptization Procedure Soon after discovering of PVI (1) by Gambier, Painleve presented it in terms of the Weierstrass elliptic functions (Painleve [1906]). This paper was almost forgotten for ninety years and the elliptic form was rediscovered recently in Manin [1998] and Babich et al [1997]. We follow the derivation presented in Manin [1998], where the Hamiltonian form and symmetries of PVI in terms of elliptic functions are treated. Consider the family of elliptic curves Er = C/(Z + Zt) (2) where r e H = {Imr > 0}. Let p(m|t) be the Weierstrass function .(M|r) = ^ + J2 {-TT ^-Z ^ - 7 T ^2) ' (^ = ")• u^ ^-^ \{u-\-mu)\-\-nuyi) (mcoi + nc02) / m (3)
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 155 In the most part of the paper we put a;i = landp(M|T) = p(m|1, a;2). p(m|t) uniformize the elHptic curve Pu(u\t) = 4(p(m|t) - ei(T))(p(u\T) - e2(r))(p(u\T) - ^sC^^)), (4) ei = P (j\r) , (To,... , Ts) = (0, 1, r, 1 + r). We consider two kind of transformations. The first one is the lattice action u -^ u -\-m -\-nr, r -^ z. (5) It leaves p(u\t) and Pu(u\t) invariant. The second is the modular transformation by PSL2(Z) / u ar-\-b\ ,^o / u az-\-b\ ^ P( —-I —- =(cT-\-drp(u\T), Pu —-I -— ={cZ+dyPu{u\T), \cx-\-d cT-\-dJ \cT-\-d cT-\-dJ (6) Now consider another family of elliptic curves Et -^ B, Y^ = X(X — l)(X — t) parameterized by B = {t e P^ \ (0, 1, oo)}. There exists the morphism [Er] -^ {Et} defined as (u,t) -^ [X = , Y = , t = . (7) Theorem 9.1 In terms of(u, r) Pyia,p,y,8 takes the form TT(u\ U \U \ (ao. r), U(u\ ."3) T) = -(«: 1 (271 i .-/S, 3 \2 Z^">^ y,\-s). The proof of the equivalence of (1) and (8) is based on the Picard-Fuchs equation on elliptic curves. The Picard-Fuchs operator acting on the holomorphic differential co = (dE/Bx)/y yields the exact differential \dE/B (x-t)^' T^^ Picard-Fuchs equation just means that periods of dE/sx/y are annihilated by Lt. Using the Picard-Fuchs operator Fuchs proved that PVI (1) is equivalent to the following equation f^ ( t t-\ \ t(t-\)\ (9)
156 M.A. OLSHANETSKY The equivalence of (1) and (9) follows from the following equality ., .r f^ , , ^t{t-\)Y (fX \(\ 1 1 \{dX\^ (7-7^-x^)f' (^^-(-^)(- 0). The proof is straightforward. Thus, PVI can be written in the form of the so-called /x-equation (Manin [1998]) Joo muLt j CO = S(a,fi,y,8)(X), (10) Joo where the right hand side is a special section of the bundle Ef. It can be fixed by the symmetries of the equation. Under the morphism (7) the holomorphic differential dE/HZ on Er is transformed into dE/Bx/y, and jp into Lf. More exactly, the left hand side of (10) takes the form dz. (ei - e2){e\ - ^3)(^2 - ^i)^ ^^^ Taking into account that / Jo Y = -{e2-ex)-^'^Pu{u,T) we come finally to (8). 9.2.2 Hamiltonian Structure The Hamiltonian form of (8) is defined by the standard symplectic form co^^'>=8v8u, (11) and the Hamiltonian H = --U(u\T). (12) Consider the bundle V over the moduli space M. = if/PSL2(Z) with the symplectic fibers parameterized by the local coordinates (v, u). It plays the role of the extended phase space for the non-autonomous Hamiltonian system (11), (12). The equation of motion (8) can be derived from the action J^ onV 8J^ = v8u-H8t. (13) The symmetries of the non-autonomous Hamiltonian systems are determined by the invariance of the two-form co on V CO = 0)^^^ - 8H8t = 8v8u - 8H8t. (14)
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 157 It follows from (3) and (6) that the symmetry group is the semi-direct product of Z + Zt and the group r(2) c PSL2(Z). We consider a simplified version of this action in Section 9.7 in detail. 9.2.3 Calogero-Inozemtsev Equation and PVI Let us introduce the new parameter k and instead of (14) consider 1 CO = coo 8H8t. (15) K It can be achieved by the rescaling the dynamical variables (f, u) and periods a)i,(02 by V -^ K~2^ u -^ K^, coi -^ K2, C02 -^ K2. Then, (8) takes the form ^^t4 = -^uU(u\t). (16) Put T = To-\- Kt^ and consider the system in the limit k -^ 0. We come to the equation Jdi^ ^-duU{u\To) (17) corresponding to the autonomous Hamiltonian system with the time-independent potential U{u\zo). It is just the rank one elliptic Calogero-Inozemtsev equation (CIa,^,y,8) (Calogero [1976] and Inozemtsev [1989]). The potential U(u\to) was considered first by Darboux [1882]. It arises also in the soliton theory (Treibich [1990]). Thus, we have in this limit PVIa,fi,y,s"-^CIa,^,y,S. (18) There is the inverse procedure (Whitham quantization) that allows one to construct approximations of non-autonomous systems starting from integrable autonomous systems. It will be discussed in Section 9.4. Inozemtsev considered degenerations of C/(m |to) playing with the coupling constants, the periods, and u. In this way he obtained trigonometric, rational and exponential interactions. Presumably, they describe the degenerations of PVI to PV-PI in terms of degenerations of elliptic functions. Here is one of his potentials: 1 1 ao ^ f-ai ^ f-a2expM -\-a3Qxp2u. sinh^ u sinh^ 2m In what follows we consider only the subfamily PVIy corresponding to aj = v^.
158 M.A. OLSHANETSKY 9.3 Isomonodromic Deformations Here we describe the monodromy preserving equations as reduced Hamiltonian systems. The original phase space is infinite-dimensional and almost all degrees of freedom are killed by the symplectic reduction. Our approach differs from Hamad [1996], where PI-PVI equations are treated as a result of symplectic reduction of a finite-dimensional space. 9.3.1 Hamiltonian Approach Let Tig be a Riemann curve of genus g. Consider the space FBun^^G of flat vector bundle Vg, where G = SL(N, C) with smooth connection A. The flatness means that its curvature vanishes FA=dA-^-[AA]=0. (19) Let us fix the complex structure on S^. Then for ^ = (A, A) we have locally a consistent system of matrix differential equations (a + A)vI/=0, (a + A)vI/ =0. We modify this system in the following way. First, introduce formally a parameter k eR (the level) and consider the operator k d instead of d in the first equation. Let /x be a Beltrami differential on S^ (/x G ^ ^~ ^'^^(S^)). It means that in local coordinates/x = /x(z, z)^<^dz. It allows us to deform the complex structure on S^ such that the new complex coordinates are 3^(z, z) w = z- e(z, z), w = z, i^(z, z) l-d€(z,z)' The holomorphic operator dw = 3 + /x3 annihilates the one-form dw likewise d annihilates dZ' We do not touch the anti-holomorphic operator Kd.ln the new coordinates (19) takes the form F^ = (a + a/x)A - KdA + [A, A] = 0. (20) Thus, we come to the system (Aca + A)vl/= 0, (21) (a + /xa + A)vI/=0. (22) Represent the Beltrami differential as /x = J2a=i ^^ M^' where /x p ... , /x|^ is the basis in the tangent space to the moduli space Mg of complex structures on S^, (/ = dim Mg = 3g — 3, for g > 0). In other words, t = (t\,... , f/) are coordinates of a tangent vector to Aig. To fix a fundamental solution of (21), (22), impose the following normalization for some reference point (zo, ^o) ^ Tg ^(zo,zo) = /.
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 159 Let y be a homotopically nontrivial cycle in S^ such that (zo, ^o) ^ Y and y is the corresponding monodromy transformation y(y) = ^(zo, zo)\y = Pexp * A. The set of matrices {y(y)} generates a representation of the fundamental group tti (E^, zo) in SL(A^, C). Independence of the monodromy y on the deformations of the complex structure means that the linear equations day = 0, (a = h...J)(da=dO (23) are consistent with (21), (22). Proposition 9.1 Equations (23) are consistent with (21), (22) iff daA=0, (« = l,... ,/), (24) daA = -Atxl (a = l,...J). (25) K The proof is straightforward. Proposition 9.2 Equations of motion (24), (25) are Hamiltonian. Endow the space FBun^^c with the symplectic form L (0^^^ = / <SA, SA >, (<, >= tr), (26) and the set of Hamiltonians < A,A>/if, (a = l,...,Z). (27) ""IL Then (24), (25) are Hamiltonian equations with respect to co^^^ and Ha. Consider the bundle V over the moduli space Mg with FBun^^c as the fibers. The triple (A, A, t) can be considered as the local coordinates of the total space of the bundle. It is useful to consider V as the extended phase space (Arnold [1978]). There is a closed two-form on V r..-rJ^)--\^SHa8ta. (28) K a Though CO is degenerate on V it produces the equations of motion (24), (25), since the form co^^^ is non-degenerate along the fibers. The gauge transformations in the deformed complex structure take the form A ^ f-'Kdf + f-'Af, A -^ f-\d + ^l^)f + f-'Af. (29)
160 M.A. OLSHANETSKY The form co is invariant under these transformations, though its constituents co^^"^ and Ha separately are not invariant. Introduce a new pair of connection components A = {A, A'), where A' = A — ^f^A. In terms of (A, AO the form co (28) takes the canonical form = f <8A,8A' > . 0)= <8A,8A' > . (30) 9.3.2 Symplectic Reduction Gauge fixing along with the flatness condition (20) is nothing but symplectic reduction from the space of smooth connections Sm^^c in the bundle Vg to the reduced space FBun^^G = FBun^^c/G = Sm^^of/G' The double slashes means that we impose the moment constraints (20) and fix the gauge. FB wn E, G is the moduli space of flat connections of the bundle Vg • In terms of the symplectic reduction procedure the flatness condition is called the moment constraint equation. Let us fix the gauge in a such way that the A component of A becomes anti-holomorphic dL =0, (L = f-\d + ^l^)f + f-'Af). (31) We can do this because the antiholomorphity of /~^(3 + /x3)/ + f~^Af amounts to the classical equations of motion for the Wess-Zumino-Witten functional Swzwif^ A) for the gauge field / in the external field A. Denote the gauge transformed field A as L L = f-'Kdf + f-'Af. Then (20) takes the form (a + a/x)L + [L, L] = 0. (32) Thus, the moduli space of flat connections FB un^^c^^ characterized by the set of solutions of the linear differential equation (32) along with the condition (31). The moduli space FBun^^G is a finite-dimensional space dim F^n^^G = 2(N^ - l)(g - 1), g > 1. After gauge fixing we come to the bundle V over Mg with FBun^^G as the fibers. The system of linear differential equations (21), (22) and (23) after gauge fixing takes the form (Aca + L)vI/=0, (33) (a + /xa + L)vI/=0, (34) (Kds-\-Ms)^=0, (35)
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 161 where we replaced ^ by f~^^ and Ms = —Kdsff~^. The gauge transformations do not spoil the consistency of the system. The consistency of (33) and (34) is provided by (32) and (31). In fact, the consistency (35) with (33) and (34) leads to the Lax form of the equations of isomonodromic deformations dsL - KdM + [M, L] = 0, (36) KdsL - /x^L = (a + fMd)Ms - [Ms, L]. (37) They play the role of (24), (25) correspondingly. The last equation allows us to find Ms in terms of dynamical variables L, L. The symplectic form co on the reduced phase space V is = f <8L,8L>—y^SHsSts, 0)=^ <8LJL>—> SHsSts, (38) Hs = l f <L,L>fMfK (39) Introduce the local coordinates (v, u) in FBun^c- L = L(v, u, t), L = L(v, u, t), V = (fi, . . . , i;(A^2_i)(^_i)), U = (mi, . . . , M(A^2_i)(^_i)). Assume for simplicity that this parameterization leads to the canonical form on FBun^^c .(0) -i < 8L(\, u, t), 8L(\, u, t) >= ((5v, (5u), (40) 5]. where the pairing in the right hand side is induced by the trace. The form on the extended phase space is o) = (5v, 8u) - - Y^Ks(\, u, t)8ts, (41) s and the variations of the Hamiltonians Ks in the new variables take the form 8Ks = I [< L,8L> /xf ^ + k(< 8L, dsL > - < dsL, 8L >)]. (42) Now, due to (32), the Hamiltonians depend explicitly on times. Consider the one-form (the integral invariant of Poincare-Cartan) = 8-^CO = (v, (5u) - - V ^,(v, u, t)8ts 1^- ^—'
162 M.A. OLSHANETSKY There exists a 3g — 3 = dim Al^-dimensional space of vector fields Vs that annihilate 0 Vs=Kds-\-{Hs,'l (s = h,..J), (43) It can be checked that V^ satisfy the following conditions KdsHr - KdrHs + {/f„ /f,L(0) = 0. (44) Thereby, they define a flat connection in the bundle V. These conditions are called the Whitham hierarchy (WH). The equations for any function /(v, u, t) on V take the form !ffi^=«^^%^ + (H„/, (45, dts dts They are called the hierarchy of isomonodromic deformations (HID). Both hierarchies can be derived from variations of the prepotential T. It is defined as the integral over the classical trajectories in the extended phase space V /»U,t J^(u, t) = J^(uo, to) + / Csdts. (46) •>'Uo,to where Cs{dsVL, u, t) = (v, O^u) — ^^(v, u, t), {dsVL = ^) is the Lagrangian. T satisfies the set of the Hamilton-Jacobi equations ^a,^+/f,(^,u,t)=0. (47) The logarithm of T is called the tau-function of HID. 9.3,3 Singular Curves Singular curves are important for applications since they produce nontrivial systems for curves of low genus (g = 0, 1). In these cases explicit calculations of hamiltonians are available. Consider a curve I)^,„ of genus g with n marked points (jci,... , jc„). The number of times is equal to dimension of the moduli space Al^,^. We extend the space of connections FBun^^G = {A^ ^} by adding the coadjoint orbits of G = GL(N, C) at the marked points (Oi,...,a), ot = {pb = gp^g-'}, where p^ fixes the conjugacy class of Ob- We allow the A component of the connection to have simple poles at the marked points, while the Beltrami differentials vanish there. The Hamiltonian formalism is provided by the modified symplectic form / < 8A, 8A > -\r27ii ^ < 8(pbg^^), Sgb w*"* = / < <5A, <5A > +2ni ^ < SipbSh'), Sgb > ■ (48) b=l
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 163 Finally, we come to the same linear system (33)-(35), but due to the singularities of A the following relation between L and L holds (d + a/x)L + [L, L] = 0, L\,^,, - 27zi-^^ (49) Z- Xb As before, the linear equations are equivalent to the equations of motion of HID coming from the symplectic form 1 ^ CO = co^^\y, u, p) - - J2^^s(y, u, p, t)8ts, (I = dim(Mg,n), (50) '^ 5 = 1 where p = (pi,... , Pn),coo is determined by the reduction from (48) J^"^ = < 8L(\, u, p), (5L(v, u, p) > +27r/ ^ < 8(pbg^^), Sgt >, (51) •^^^^ b=i and Ks (42). 9.4 Hitchin Systems and Their Whitham Deformations 9.4.1 Hitchin Systems Consider the moduli space 1Zg^M of stable holomorphic SL(N, C) vector bundles V over Tig. It is a smooth variety of dimension dimng,M =g = (N^- l)(g - 1). (52) Let T^lZg^M be the cotangent bundle to Tlg^M with the standard symplectic form on it. Hitchin [1987] defined a completely integrable system on T*lZg m- The space T*7lg^M can be obtained by symplectic reduction from the space T*7Z^ ^ = (O, A), where A is a smooth connection of the stable bundle corresponding to 3 + A and O is the Higgsfield O G OPiTg, EndV 0 K) (K is the canonical bundle of Tg). There is the well defined symplectic form on this space -L a;(0) ^ j < 50, (5A > . (53) This form is invariant with respect to the gauge group Q = C^MapiTig, SL(N, C) action ^ -^ /"'^/, A -^ /-la/ + f-'Af. (54) In particular, 1Zg^M = ^^,a^/^- Let p^j^ = ps,k^z~^ ^ ^^ ^^ ^^^ (~^ + ^' 1)-differentials (Ps^k ^ H^(Tig, r^-i)), and s enumerates the basis in H^(Tig, r^~^), (ps,2 = f^s)- E>ue to the Riemann-Roch theorem dim//i(S„r^-i) = (2^-l)(g-l).
164 M.A. OLSHANETSKY These differentials allow us to define the gauge invariant Hamiltonians Hs,k = \ I <<^^> Ps,k, (k = h...,N, s = h...,(2k- l)(g - 1)). (55) The Hamiltonian equations take the form da<^=0, (da = ^,a = (s,k)), (56) Ota daA = <l>''-^Ps,k (57) The gauge action produces the moment map /x : T*1Z^ -^ Lie*(SL(N, C)). It follows from (53),(54) that /x = 30 + [A, O]. The reduced phase space is the cotangent bundle we started with The Hitchin hierarchy (HH) is the set of Hamiltonian equations with Hs^k (55) on the reduced phase space T^lZg^M- Hitchin observed that the number of integrals Hs,k J2(2k-l)(g-l) = (N^-l)(g-l) k=2 which coincides with dimension g of the coordinate space TZg^M (52). Since they are independent and Poisson-commute, HH is the set of completely integrable Hamiltonian systems. Let us fix the gauge of the field A A = fdf-' + fLf-\ Then L = f-'<t>f. is a solution of the moment constraint equation dL + [L, L] = 0. (58) The space of solutions of this equation is isomorphic to//^(S^, End y0^) -the cotangent space to the moduli space Tlg^M- The gauge transformation / defines the element Ma in Lie(SL(A^, C) Ma = daff~^. Proposition 9.3 The system of linear equations (A + L)F=0, (59) (a -\-J2x^-hs,kPs,k + L)Y = 0, (60) s,k (da-^Ma)Y=0 (61) is consistent and defines the equations of motion for HH.
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 165 Proof The consistency of (59) and (60) follows from (58). In terms of L the equations of motion (56), (57) take the form daL + [Ma, L] = 0 (the Lax equation), (62) dal - dMa + [Ma, L] = L^-'Ps,k, (a = (s, k)). (63) The Lax equation provides the consistency of (60) and (61), while the second equation of motion (63) plays the same role for the pair (60) and (61). This equation allows us to determine Ma from L and L. The bundles over singular curves can be incorporated in this approach as well (Nekrasov [1996]). The Higgs field has simple poles at the marked points and (58) is replaced by n dL + [L, L] = 2jTiJ2^^(^b)Pb^ (64) b=i The form a;^^> on T^Ug^N o)^^^ = / <8LJL> -\-27Ti Y < 8(ptg^^), Sgb > (65) J^^ b=x Note that the space T*lZg^M and the space of flat bundles FBun^^c have the same dimensions. Moreover, it follows from the comparison the moment constraints (49), (64) and the symplectic forms (51), (65) that they are isomorphic as symplectic manifolds. 9.4.2 Spectral Description Due to the Liouville theorem the phase flows of HH are restricted to the Abelian varieties, corresponding to a level set of the Hamiltonians Hs,k = Cs,k' The phase flow takes a simple form in terms of action-angle coordinates. They are defined in a such way that the angle type coordinates are angular coordinates on the Abelian variety, and the Hamiltonians depend on the action coordinates only. To describe them consider the characteristic polynomial of the matrix L P(A, z) = det(A + L) = A^ + ... + bjX^-J + ... + Z?a^, (66) bj = 22 ^i^j^ (Mirij — principle minors of order j, b^ = det L). The spectral curve C c T* S^ is defined as the zero set of P C = {P(X,z)=0}. It is a well defined object, because the coefficients bj are gauge invariant.
166 M.A. OLSHANETSKY Since L G H^i^g, End V 0 K), the coefficients bj g H^i^g, KJ) and we obtain the map p : T*ng,N -^ B = ejL2^^^^' ^^')- (67) The space B can be considered as the moduH space of the family of spectral curves parameterized by the Hamiltonians Hs^k- The fibers of p are Lagrangian subvarieties of T^lZg^N' The spectral curve C is the A^-fold covering of the basis curve Tig^M TC \ C -^ ^g,N' There is a line bundle C with an eigenspace of L(z) corresponding to the eigenvalue X as a fiber over a generic point (X, z) CckcT(X-^L) C7r*(y). It defines a point of the Jacobian Jac(C), the Liouvillean variety of dimension g = dimB. Conversely, if z g S^ is not a branch point one can reconstruct V for a given line bundle onC as Let (Oj, j = I.., ,g be the canonical holomorphic one-differentials on C such that for the cycles ai,... , a^; y^i,... ,Pg, ocfOCj = fit • fij = 0, a/ • fij = 8ij, /^. coj = (5/y. Then the symplectic form co^^^ (53) can be written in the form f <8L,8L >=T] f ^h^^j- Here ^j are the diagonal elements of sLs ^ where s diagonalizes L, sLs ^ diag(Ai,... , Aa^). Then we obtain «(«> = f sxs^. Because A is a holomorphic one-form on C, it can be decomposed as A = X!f=i ^j^j- Thereby a;^^> 7 = 1 The action variables can be identify with f Xj, (j = l...,g), (68) To define the angle variables, put locally ^ = d log x//. If (/7^) is a divisor of i/^ then / ^7^? = y] / ^7 log V^ = %•
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 167 Thus (pj are linear coordinates on Jac(C) and g 9.4.3 Scaling Limit Consider HID in the limit k -^ 0. The value ac = 0 is called critical. We prove that on the critical level HID coincide with part of HH relating to the quadratic Hamiltonians (55). Note first, that in this limit the A-connection is transformed in the Higgs field O: A —> O, and therefore FBun^^G —> T*1Zg^M' But the form co on the extended phase V appears to be singular (see (28), (38)). To get around this we rescale the times t = T + Act^, (69) where t^ are the fast (Hitchin) times and T are the slow times. Assume that only fast times are dynamical. This means that 5/x(t):=^^/xf5ff, (/Xf =an,). s After this rescaling the forms (28), (38) become regular. The rescaling procedure means that we blow up a vicinity of the fixed point /x^ ^ in Mg^n and the whole dynamic is developed in this vicinity. This fixed point is defined by the complex coordinates wo = z-Y^ Ts€s(z, z), u'o = z- (70) s Now compare the Baker-Akhiezer function of HID vj/ (33), (34),(35) with the Baker- Akhiezer function of HH Y (59), (60), (61). Using the WKB approximation, assume that c(0) vI/ = Oexp( +5^^^), (71) K where O is a group valued function and S^^\ S^^^ are diagonal matrices. Let us substitute (71) in the linear system (33), (34), (35). If dwo dt^ there are no terms of order ac ~ ^ It follows from the definition of the fixed point in the moduli of complex structures (70) that 5(0) ^ s^O)^j^^ ^^^^Ti\z-Y, Ts^siz, z)). (72)
168 M.A. OLSHANETSKY We take also S^^^ = dS^^^ J2s ts"^s(z, z). In the quasi-classical limit we put dS^^^ = X. (73) In the zero order approximation we come to the linear system of HH (33), (34), (35), defining by the Hamiltonians Hk,s^ k = 2. The Baker-Akhiezer function Y takes the form \^ fH _d_ o(0) Y = ^e^^ ^^ 9^^ ^ . (74) Our goal is the inverse problem. We need to reconstruct the dependence on the slow times T starting from solutions of HH. Since T is a vector in the tangent space to the moduli of curves Aig^n, it defines a deformation of the spectral curve in the space B (see (55), (67)). Solutions Y of the linear systems (59), (60), (61) take the form Y = <^e^^ ^' ^\ v/here Qs are diagonal matrices. Their entries are primitive functions of meromorphic differentials with singularities matching the corresponding poles of L. Then according with (74) we can assume that dS = dQs- dTs These equation define the approximation to the phase of ^ in the linear problems (33), (34), (35) of HID along with do — coj. daj ^ The differential dS plays the role of the Seiberg-Witten differential. The important point is that only part of the spectral moduli, that connected with Hk^s^ k = 2, is deformed. As a result there is no matching between the action parameters of the spectral curve aj, j = 1,... , g (68) and deformed Hamiltonians. A detailed analysis of this situation in the rational case is undertaken in Takasaki [1997]. Another object of the Whitham quantization is the prepotential J^ (46). It depends on the action variables aj . This dependence is compatible with the Hamilton-Jacobi equation (47) with slow times Ts as the independent variables. These equations are discussed in Takasaki [1997] and Itoyama et al [1997]. 9.5 Classical Limit of the Knizhnik-Zamolodchikov-Bemard Equations The Knizhnik-Zamolodchikov-Bemard equations (KZB) are the system of differential equations having the form of the non-stationary Schrodinger equations with the times coming from Mg^n (see, for example, Ivanov [1996]). They arise in the geometric quantization of the moduli of flat bundles FBun^G (Axelrod et al. [1991] and Hitchin [1990]). Let V = Vi x • • 0 y„ be the tensor product of finite-dimensional irreducible representations associated with the marked points. The Hilbert space of the quantum system is a space of sections of the bundle f y,/c««««' (S^,n) over FBun^G depending on non-negative number /c^"^"^ with the V-fibers. It is the space of conformal blocks of the WZW theory on S^,„.
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 169 The Hitchin systems are the classical limit of the KZB equations on the critical level (Nekrasov [1996] and Ivanov [1996]). The classical limit means that one replaces operators by their symbols and generators of finite-dimensional representations in the vertex operator acting in the spaces Vj by the corresponding elements of coadjoint orbits. To pass to the classical limit in the KZB equations (K^^'''''ds-^Hs)F = 0, (75) we replace the conformal block by its quasi-classical expression F=exp|, (76) where h = (/c^"««^)-i. Consider the classical limit /c^"^"^ -^ oo and assume that values of the Casimirs Q, (/ = !,..., rankG, a = 1,... ,n) corresponding to the irreducible representations defining the vertex operators also go to infinity. Let all values lim ^^^ be finite. This allows us to fix the coadjoint orbits at the marked points. In the classical limit (75) is transformed to the Hamilton-Jacobi equation for the action ^ = log t (47) of HID. The integral representations of conformal blocks are known for WZW theories over rational and elliptic curves (Schechtman et al [1991], Falceto et al. [1996], Etingof et al. [1994] and Felder et al. [1996]). Then (76) allows us to extract the prepotential JT of HID. The KZB operators (75) play the role of flat connections in the bundle p^"^"^ over the moduli of curves Mg,n with the fibers 5y,/c««««'(^^,n) (Hitchin [1995] and Felder [1995]) [K^^^^^ds + Hs, /t^"^"'a, + Hr] = 0. These equations are the quantum counterpart of the Whitham hierarchy (44). 9.6 Multicomponent Generalization of PVIy Consider FBun^^c over the family of elliptic curves with a one marked point Ali,i. The space M\,\ is one-dimensional, because the position of one point on a torus is irrelevant. Thus, we have only one time t and Mi,i ~ Er (2). In this case the Beltrami differential takes the form T - TO M = ^• T - To Consider the most degenerate orbit O = (gp^g~^) of SL(N, C) sitting at the marked point z =0 with p^ = v[(h^_^f 0(1,...,!)- Id], (11) N N For stable bundles the gauge transforms allow us to put the A component in diagonal form 27zi L = ^u, u = diag(Mi,... , um) eH — Cartan algebra. (78) T — To
170 M.A. OLSHANETSKY It means that / Ldwdw = u. (79) Let L = d log 0. Then the integral j Ldwdw = I logcpdw. JEr JPo defines the Abel map E^ in the product of A^ Jacobians. The remaining gauge transforms do not change the gauge fixing. These transformations are generated by the Weyl subgroup W of G and elements f(w,w) G Map(r^^, Cartan(G)). The orbit variables can be gauged away by these transforms and we are left with/7(^> (77). The solution L of the moment constraint d^L + [L, L] = 27zi8^(0)p^^^ takes the form L = P + X, P = 2jTi(— AC-), (80) 1 -/x p u = diag(Mi,... , un), V = diag(i;i,... ,vn) w — w Xjk = x(uj - Uk) = (t - ro)vexp27r/{ —{uj - Uk)}(t>{oc{uj - Uk). w), T — To e(u)e(z) ^ The operator M defining the phase flow according with the Lax equation (36) can be extracted from (37) M = —D -\-Y, D = diag(Ji,... , d^), dj = ^^^(^j ~ "/)' ^(^) = -p(w) + (81) Yjk = y(uj - Uk), y(u, w, w) = . :—dux(u, w, w). 2niK{z — ro) The functions x,y,z satisfy the functional equations x{u, z, z)y{v, z, z) - x{v, z, z)y{u, z, z) = {s{v) - s(u))x(u + v, z, z). (82) This equation is derived from the Lax equation (36). The symplectic form co is reduced to (o = ((5v, Su) SHSr, K where 2 ^ (8\,8\) y- j<k
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 171 They define the Hamiltonian flow $ = (^E«^«(«-«^l^)- (83) k<j For N = 2 one can put u\ = —U2 = u. Then the potential y2 y2 3 ^p(2w|r) = ——2X^p(w + ^|r) produces PVIy (see (8). The remaining gauge symmetries implies that co is invariant under the Weyl transformations W of (\, u) and the lattice actions (compare with (5) V -^ 5V, V + Acn, u -^ 5U, u — m + rn, (s e W, n e Z^). It is also invariant under the PSL2(Z) action on t nr 4- h V -^ V(CT -\- d) — KCU, U -^ U(CT + J) This invariance follows from the invariance of the upstairs system under the diffeomor- phisms of J^g^n (see Levin et al. [1999] for details). On the critical level ac -^ 0, r — tq = KtwQ obtain the elliptic Calogero N-body system. This system is a particular example of the Hitchin systems (Nekrasov [1996]). Note, that the functions jc, y and s defining the Lax matrices satisfy the same functional equation (82) as in the Calogero-Hitchin limit k =0 (Calogero [1976]). 9.7 Conclusion Here we propose a few open problems in the context of the topics discussed above. • The obvious problem is a description of PVI with four arbitrary constants as a reduced Hamiltonian system. The first step in this direction is the Lax form of PVIa,^,y,8' The Lax form is unknown even on the critical level, i.e. for the Calogero-Inozemtsev system. It will be interesting to generalize this approach to the A^-body Calogero-Inozemtsev system and the A^-component PVI with four coupling constants. • The degenerations of PVI to PV-PI in terms of elliptic functions. • There exists a generalization of the Calogero systems related to any simple group. In addition to degrees of freedom coming from the moduli of bundles (the coordinates of particles), these systems certainly contain degrees of freedom related to the coadjoint orbits. Recently new Lax equations based on arbitrary root systems without the orbit coordinates were proposed (Phong [1998] and Bordner et al. [1998]). This construction is purely algebraic and does not use the symplectic reduction. How can these systems can be incorporated in the Hitchin approach, or, more generally, in the isomonodromic deformation construction?
172 M.A. OLSHANETSKY • Gonsider the A^ = 2 elliptic Calogero system. The solution u(t), corresponding to the fixed value /i2 of the Hamiltonian 2 2 H = — -\- —^p(2u\to) = /i2, 2 47r^ is implicitly described by the elliptic integral of the first kind 1 f^^dx , r ^ where y = 4(x — e\ (tq)) (jc — ^2 (^0)) (^ — ^3 (^0)) • As was mentioned at the end of Section 9.6 this can serve for calculating solutions to PVIy. This procedure may be accomplished by the Krichever averaging method (Krichever [1998]). It will be interesting to compare this approximation with explicit solutions presented recently in Deift et al. [1998] and Kitaev et al. [1998] for some particular value of the coupling constant v. As suggested in Section 9.5 another way of approximation comes from the classical limit of conformal blocks for SL2(C) theory on elliptic curves with one marked point (Etingof et al. [1994] and Felder et al. [1995]). Which method gives the better approximation? We considered deformations with respect to the moduli of complex structures of curves. They describe only part of the moduli of the spectral curves C. The remaining moduli of C come from p^j^, k > 2. They correspond to the so-called W-geometry of the basic curve ^g,n' This geometry is poorly understood. On the other side, there are no examples of isomonodromic deformation equations with respect to these moduli spaces, as well as the corresponding higher order KZB equations. Any progress in understanding of one of these subjects will shed light on another. Acknowledgements The work is supported in part by grants RFFI-96-02-18046INTAS 96-518 and 96-15-96455 for support of scientific schools. I am grateful to the Max-Planck-Institut fiir Mathemamatik in Bonn for the hospitality, where this chapter was prepared. References Arnold, V.I., Mathematical methods of classical mechanics. Springer-Verlag, NY (1978). Axelrod, S., Delia Pietra, S., and Witten, E., Geometric quantization of the Chem-Simons gauge theory, Joum. Diffi Geom. 33, 787-902 (1991). Babich, A., and Bordag, L., The elliptic form of the sixth Painleve equation. Preprint NTZ 25/1997. Bernard, D., On the Wess-Zumino-Witten models on the torus, Nucl. Phys. B303, 77 (1988); On the Wess-Zumino-Witten models on the Riemann surfaces, Nucl. Phys. 309, 145 (1988). Bordner, A.J., Corrigan, E., and Sasaki, R., Calogero-Moser Models: A New Formulation, hep- th/9805106. Prog. Theor Phys. 100, 1107-1129 (1998). Calogero, F, Exactly solvable one-dimensional many-body problems, Lett. Nuovo Cim. 13,411^17 (1976),
PAINLEVE TYPE EQUATIONS AND HITCHIN SYSTEMS 173 D'Hoker, E., and Phong, D.H., Calogero-Moser Systems in SU(N) Seiberg-Witten Theory, Nucl Phys. B513, 405-444 (1998). Darboux, M.G., Sur une equation lineare, Comptes Rendues de VAcademie de Science de Paris XCIV(N25), 1645-1648 (1882). Deift, P., Its, A., Kapaev, A,, and Zhou, X., On the algebro-geometric integration of the Schlesinger equations, and on elliptic solutions of the Painleve VI equation. Preprint lUPUI 98-2 (1998). Dubrovin, B., Integrable systems in topological field theories, Nucl. Phys. B 379, 627-689 (1992). Etingof, P., and Kirillov, A., Representations of affine Lie algebras, parabolic differential equations, and Lame functions, Duke Math. J. 74, 585-614 (1994). Falceto, F, and Gawedzky, K., Elliptic Wess-Zumino-Witten model from elliptic Chem-simons theory, hep-th/9502161, Lett. Math. Phys. 38, 155 (1996). Felder, G., and Varchenko, A., Integral representation of solutions of the elliptic Knizhnik- Zamolodchikov-Bemard equations, hep-th/9502165. Int. Math. Res. Notices 221-233 (1995). Felder, G., andWeiczerkovski, C, Conformal blocks on elliptic curves, and Knizhnik-Zamolodchikov- Bemard equations, hep-th/9411004, Comm. Math. Phys. 176, 113 (1996). Felder, G., The KZB equations on Riemann surfaces, hep-th/9609153, to appear in the Proceedings of the 1995 les Houches Summer School. Fuchs, R., Uber lineare homogene Differentialgleichungen zweiter ordnung mit im endlich gelegne wesentlich singularen Stellen, Math. Annalen, 63, 301-323 (1907). Gambler, B., Sur les equations differentielles du second ordre et du premier degre dont Tintegral generale a ses points critiques fixes. Acta Math. Ann 33, 1-55 (1910). Hamad, J., Dual Isomonodromic Deformations, and Moment Maps to Loop Algebras, Commun. Math. Phys. 166, 337-366 (1994); Hamad, J., and M.-A. Wisse, Loop Algebra Moment Maps, and Hamiltonian Models for the Painleve Transcendants, Fields Inst. Commun. 7, 155-169 (1996). Hitchin, N., Flat connections, and geometric quantization, Comm. Math. Phys. 131, 347-380 (1990). Hitchin, N., Stable bundles, and Integrable Systems, Duke Math. Joum. 54, 91-114 (1987). Hitchin, N., Twistors spaces, Einstein metrics, and isomonodromic deformations, Joum. Diff. Geom. 3, 52-134 (1995). Inozemtsev, V., Lax Representation with spectral parameter on a toms for integrable particle systems, Utt. Math. Phys. 17, 11-17 (1989). ItoyamaH., andMorozov, A., Prepotential, and the Seiberg-Witten Theory, A^mc/. Phys. B491,529-573 (1997). Ivanov, D., KZB eqs. as a quantization of nonstationary Hitchin systems, hep-th/9610207 (1996). Iwasaki, K., Fuchsian moduli on a Riemann surface — its Poisson stmcture, and Poincare-Lefschetz duality. Pacific J. Math. 155, 319-340 (1992). Kitaev, A., and Korotkin, D., On solutions of the Schlesinger Equations in Terms of S-Functions, math-ph/9810007. Intemat. Math. Res. Notices 17, 877-905 (1998). Knizhnik V., and Zamolodchikov, A., Current algebra, and Wess-Zumino model in two dimensions, Nucl. Phys. B247, 83 (1984). Korotkin, D., and Matveev, V., On solutions of the Schlesinger system, and Ernst equation in terms oftheta-functions. Preprint AEI-087, 1998. Krichever, I., Method of averaging for two-dimensional "integrable" equations, Func. Anal. Appl. 22, 200-213 (1988). Krichever, I., The tau-function of the universal Whitham hierarchy, matrix models, and topological field theories, Comm. on Pure, and Appl. Math. XLVII, 437-475 (1994). Levin, A., and Olshanetsky, M., Isomonodromic deformations, and Hitchin systems, Amer. Math. Soc. Transl. 2, 191 (1999). Malmquist, J., Sur les euations differentielles du second odre dont Tintegrale generale a ses points critiques fixes, ArA:. Mat. Astr Fys. 17, 1-89 (1922/23), Manin, Yu.L, Sixth Painleve equation, universal elliptic curve, and mirror of P^, alg-geomi/9605010, Amer. Math. Soc. Transl. 2(186), 131-151 (1998). Nekrasov, N., Holomorphic bundles, and many-body systems, Comm. Math. Phys. 180, 587-604; hep-th/9503157(1996).
174 M.A. OLSHANETSKY Painleve, Sur les equations differentielles du second odre a points critics fixes, CRAS, 143, 111 1-1117 (1906). Phong, D.H., Talk given at this conference; D'Hoker, E., and Phong, D.H., Calogero-Moser Lax Pairs with Spectral Parameter for General Lie Algebras, Nucl. Phys. B530, 537-610 (1998). Schechtman V., and Varchenko, A., Arrangements of hyperplanes, and Lie algebra homology, /nv. Math. 106, 139-194 (1991). Takasaki, K., Spectral Curves, and Whitham Equations in the Isomonodromic Problems of Schlesinger Type, A5/«/i/. Math. 2, 1049-1078 (1998), solv-int/9704004 (1997). Treibich, A., and Verdier, J.-L., Revetements tangentiels et sommes de 4 nombres triangulaires, C.R. Ac. Sci. Paris, sen Math., 311, 51-54 (1990). Workshop on Painleve Transcedents, Their asymptotics, and Physical Applications, NATO ASI Sen B: Physics Vol. 278 (1990) Sainte Adele, Quebec, ed. by D. Levi, and P. Wintemitz.
10 World-sheet Instantons and Virasoro Algebra TOHRU EGUCHI Department of Physicsy Faculty of Science, University of Tokyo, Tokyo 113, Japan In this chapter I would Hke to review our recent results on the structure of the topological string theory and in particular the so-called Virasoro conjecture on quantum cohomology theory: a certain infinite set of differential operators forming a Virasoro algebra annihilate the partition function of the string theory and determine the number of world-sheet instantons at arbitrary genus. I start from an elementary introduction to a topological a model coupled to topological gravity and then discuss the case of the CP^ model, which describes holomorphic maps from the Riemann surface onto CP^ A present the Virasoro operators which annihilate the partition function of the CP^ model and show that they reproduce the known results on the number of holomorphic curves in C P^. I then present the operators in the case of an arbitrary Kahler manifold M which form a Virasoro algebra with a central charge c = x i^) (X is the Euler number). It is pointed out that these Virasoro operators possess a free-field realization with a free bosonic (fermionic) field corresponding'to each even (odd) homology cycle of the target manifold M. Thus the "quantum geometry" of a Kahler manifold M is described by means of a 2-dimensional conformal field theory with a central charge x i^)- 10.1 Introduction The approach of topological field theory may play a fundamental role in our attempts at understanding the geometrical principles behind string theory. String theory exposes its geometrical structures in a most transparent manner in its topological formulation and we may gain geometrical insights from the study of the topological version of string theories. When a string is compactified on a Kahler manifold M, the theory is described by an A^ = 2 supersymmetric non-linear a-model. By twisting N = 2 theory we obtain a topological field theory, the topological a-model (A-model). It is well-known that the amplitude of the topological a-model is given by the sum over holomorphic maps (world-sheet instantons) from the Riemann surface onto the target manifold M (see Witten [1998]). a-model correlation functions are interpreted as the intersection numbers among various homology cycles on the manifold M (Gromov-Witten 175
176 T. EGUCHI invariants) and the world-sheet instantons describe the quantum deformations of classical geometry (quantum cohomology). Evaluating the instanton amplitudes amounts to counting the number of holomorphic curves of M. When the target space Af is a Calabi-Yau manifold, the first Chem class vanishes ci (Af) = 0 and the dimensions of the moduli space of instantons do not depend on their degrees. In the case of a Calabi-Yau 3-fold the (virtual) dimension of the moduli space vanishes and the instantons are all isolated. It is well-known that the mirror symmetry has been used to count the number of genus=0 instantons (see Yau [1992]). The prediction of the number of rational curves of arbitrary degrees in the quintic Calabi-Yau manifold was a spectacular success of mirror symmetry (see Candelas etal. [1991]). The A-model partition function at genus ^ = 1 may also be determined by using the holomorphic anomaly (see Bershadsky ^f^/. [1993, 1994]). It is well-known that the one-loop beta function of the supersymmetric a-model is proportional to the (negative of) 1st Chem class of Af. Thus the Calabi-Yau condition c\ (Af) = 0 corresponds to the scale invariance of the theory. On the other hand, in the case of manifolds with ci (Af) > 0 (Fano varieties), the theory is asymptotically free and develops a mass gap due to dimensional transmutation. Thus the theory resembles QCD. Examples of Fano varieties are given by complex projective spaces CP^, Grassmannians, hypersurfaces of lower-degree in projective spaces etc.. (It seems that there is an interesting subtlety. In the case of a degree-^ hypersurface in projective space CP^ there is an anomaly at the "weakest" Fano variety k = N. In this case the theory does not seem to have a mass gap although the beta function is still negative. This anomalous behaviour is detected by a shift in the relation of quantum cohomology (see Givental [1996] and Jinzenji [1996].) The situation of world-sheet instantons changes very much in the case of Fano varieties. Holomorphic curves now have moduli and are able to move inside the target manifold. The dimension of the moduli space increases with the degree of the curve. In order to count the number of curves one imposes an extra condition that the curve pass through a certain number of fixed points. This condition corresponds to an insertion of extra operators into the or-model Correlation functions. Recently an algorithm based on the associativity of the quantum cohomology ring was developed and used to determine recursively the number of rational curves in various Fano varieties (see Kontsevich and Manin [1994], Itzykson [1994] and Jinzenji and Sun [1996]). The associativity approach is equivalent to the use of the WDVV equation in 2-dimensional topological field theories (see Dubrovin [1996]). The machinery is powerful, however, it is essentially limited to the case of lower genera (for the case of genus-1 Gromov-Witten invariants see Getzler [1996]). Thus far our knowledge of Gromov-Witten invariants is limited to the case of low genus curves. Instead of studying the theory perturbatively, genus by genus, however, we would like to develop an approach which enables us to study the theory as a whole non-perturbatively. In our previous communications (see Eguchi et al. [1997b, 1998] and Eguchi and Xiong [1998]) we have proposed a new approach to quantum cohomology theory: we consider the case of the topological cr-model coupled to the 2-dimensional gravity (gravitational quantum cohomology) so that we introduce Riemann surfaces of arbitrary genus and incorporate
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 177 gravitational descendants. We control the structure of the gravitational quantum cohomology at an arbitrary genus by an infinite set of differential operators which form a Virasoro algebra. We have shown that the Virasoro conditions (Virasoro operators annihilating the partition function of the theory) together with topological recursion relations reproduce exactly the known Gromov-Witten invariants of various Fano manifolds at lower genus (see Kontsevich and Manin [1994], Vainsencher [1993], Itzykson [1994], Jinzenji and Sun [1996], Caporaso and Harris [1996] and Getzler [1996]). Our "Virasoro conjecture" has now been proved rigorously in the case of ^ = 0 and 1 (see Liu and Tian [1998], Getzler [1998] and Dubrovin and Zhang [1998]). In the following we first discuss some basic materials of topological a-model coupled to 2-dimensional gravity. We then propose Virasoro generators in the case of CP^ manifolds and show that the Virasoro conditions correctly reproduce the results on instanton numbers of projective spaces. Next we present the general form of the Virasoro operators for an arbitrary Kahler manifold. We then construct a free field description of Virasoro operators and quantum cohomology. We shall show that to each even (odd) homology class of a Kahler manifold we have a free bosonic (fermionic) field and Virasoro operators are given by a simple bilinear form of these fields. It turns out that the central charge of the Virasoro algebra equals the Euler characteristic of the manifold M. Finally we discuss topological recursion relations which are auxiliary equations converting correlation functions of gravitational descendants to those of primaries. 10.2 Basics 10.2.1 Lagrangian It is well-known that when the target manifold of a supersymmetric non-linear a model is Kahler, a new supersymmetry generator can be constructed from the complex structure of the manifold and the theory acquires an extended N = 2 supersymmetry. When a theory has an A^ = 2 supersymmetry it is possible to twist the theory by redefining its spinor quantum numbers and convert it into a topological field theory. After twisting, the spinor fields are turned into ghost fields, i.e. anti-commuting fields with integer spins. The topological a model is obtained from twisting the A^ = 2 supersymmetric non-linear a model. The Lagrangian of the topological theory is obtained from the standard Lagrangian of non-linear a model with spinor fields being replaced by ghosts p, x and the Dirac operator being replaced by a suitable covariant derivative operator = it I d^z{Q,V)-itxd, V = \8ij{/4^-z<l>' + Pi9z<^), d = i j gij{d,<l>'d-,<l>J - di4>'d,(t>^). (1)
178 T. EGUCHI Here d is the degree of the map 0 : E -^ Af. 2 is the BRST operator given by the scalar component of the twisted N = 2 supercharge operator. BRST transformations of the fields are given by 8(1)' =/6x', S(t>' =i€x\ 8x'=0, 8x'=0, 8pi = -ed,<p^-iex^r)^p^, (2) The BRST variation 8 is nilpotent, 8^ = 0, up to equations of motion. Thus the Lagrangian of the topological a model is given by a BRST commutator (modulo a term proportional to the degree of the maps) Lagrangian = [Q, V], (3) When the Lagrangian is a BRST commutator, the WKB approximation of the path-integral becomes exact and there are no higher-order corrections beyond the one-loop Gaussian fluctuations. In the case of the topological a model, the saddle point of the action is found by saturating the inequality gi]{dz(t>'dict>^ + d-,ct>'d,ct>^) + gij{ - d,ct>'dict>^ + 3^0^'3^00 = 2g,jd-,ct>'d,ct>^ > 0. (4) It is given by the holomorphic maps ^z<P' = 3z0^ = 0. (5) Eq. (4) is analogous to the inequality in non-Abelian gauge theories TrF^ + TrFF* = -Tr(F + F*f > 0 (6) which is saturated by the instantons, i.e. the anti-self-dual gauge fields. Due to this analogy the holomorphic maps (5) are called as world-sheet instantons. Thus the path-integral of topological a model reduces to a sum over world-sheet instantons. 10.2.2 Instantons For the sake of simplicity let us now consider instantons from a genus 0 Riemann surface, i.e. CP\ onto the target space which is also CP^ If we use local coordinates z and w for the world-sheet and target space CP\ respectively, a degree-J instanton is described as w =■ J J—-. . (7) The number of parameters of this map is given by 2(d-\-l)-l=2d-\-l (8)
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 179 where one parameter was subtracted since only the ratio matters in (7). When the system is coupled to gravity, we have to divide by the automorphism group SL(2, C) of the sphere and hence the number of parameters reduces to 2d - 2. (9) We can write the above result more formally as dimMo,o(CP^; d)=2d-2 (10) where M.g,s (A^; d) denotes the moduli space of degree-J instantons from genus g Riemann surfaces with s punctures onto a manifold M. A general formula is known for the dimensionality of the (virtual) moduli space of holomorphic maps dimA^^,,(M; d) = ci(M)d + (1 - ^)(dimM - 3) + ^ (11) where ci(Af) is the first Chem class of the manifold M. Note that ci(CP^) = 2 and we recover (10). 10.2.3 Observables Physical observables in the topological a-model come from the de Rham cohomology classes of the target manifold. In fact, given a closed differential p-form coa = (Ji^ciii i (<l>)d<f)^^ A" 'd(l>^p € /f^(Af; R) on Af we can construct a BRST invariant operator a ' Woe = 0. (12) using the BRST transformation law (2). In the case of C P^ there are two de Rham classes 1 (identity) and co (Kahler class) and the corresponding physical observables O are denoted as P and (2. In the case of CP^ there are three de Rham classes, \,a),co^ and the corresponding physical operators are denoted as P, Q, R, respectively. These operators are fields on the world-sheet and depend on z, z, however, coordinates are usually suppressed since correlation functions are position independent in topological field theories. 10.2.4 Ghost-Number Conservation The ghost number of an observable Oa is given by 1/2 of the degree of its corresponding differential form ghost number (Oa) = ^^ = - degree of form coa- (13) Thus qp =0,qQ = I in the CP^ model and qp = 0, ^g = 1, ^/? = 2 in the CP^ model, etc. We introduce parameters [t^} in order to describe the coupling of the fields [C7^]. The derivative of the free energy with respect to t" gives the expectation value of the operator ^ = (O.). (14)
180 T.EGUCHI When gravity is coupled to the topological a model, new BRST observables, gravitational descendants cfniOa) (n = 0, 1,2, •••) appear for each primary field Oa- The 0-th descendants croiOa) are identified as the primary fields Oa themselves. Descendants or„ (Oa) carry a ghost number ghost number (cfniOa)) =n-\-qa. (15) The coupling parameters of the descendants OniOa) are denoted as t^ {t^ = t^). When only the primary couplings t^ are non-vanishing, the theory is said to be in the small phase space. On the other hand, when descendant couplings are turned on, the theory is in the large phase space. Now the ghost number conservation law reads as s s ([\cfn,{Oa,))g,s 7^ 0 4=> Y.{ni^qcc,) = dinLM^,,(M; d) i=\ i=\ = cx{M)d + (1 - ^)(dimM - 3) + ^. (16) The above formula implies that the correlation function (n/=i ^n, {Oai))g,s is non-zero and receives degree-^ instanton contributions if one can solve the equation ^\=\{ni 4- qat) — c\(M)d 4- (1 — ^)(dimAf — 3) + 5- for a non-negative integer d given the values of {w/}, {c^i}» ^, •$■ and ci(Af). We note the crucial difference between the case of Calabi-Yau manifolds with c\ (M) = 0 and the cases of Fano varieties ci (Af) > 0. In the former case the degree d does not enter in the right-hand-side of (16) and non-vanishing correlation functions receive contributions from instantons of all degrees. On the other hand, in the latter case, contributions come from instantons of a definite degree. Instantons carry a weight proportional to c\ (M) which is the coefficient of the 1-loop beta function. 10,2.5 Metric A metric in the space of primary fields {C>a} in the matter theory is given by the two-point functions on the sphere, riafi = {OaOp)s=o. (17) When gravity is coupled, the sphere with two-punctures becomes unstable and we should instead consider a three-point function rja^ = {POaOp)g=o. (18) Ghost number conservation (16) reads (only the "classical" piece d = 0 contributes to the metric) qcti + qctj = dimM (19) and thus the metric corresponds to the classical intersection pairing /^ Oa aO^ = rja^. In the case of the CP^ model the only non-vanishing element of the metric is given by {PPQ) = 1. Thus the classical piece ofthe genus 0 free energy has the form, Fq^ = ^tj,tQ. In the case of CP^ , {PQQ) = ^, (PPR) = 1 while others vanish. Here the classical free energy is given by Fq^ = ^tptQ-\- jtptR.
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 181 10.2.6 Instant on Contributions Let us next evaluate the instanton contributions to the genus 0 free energy in the small phase space. If we consider the case of C P ^ and correlation functions of the type {P - " P Q - - - Q) wheren P'sandm 2's appear, the ghost number conservation law reads m = 2J—24-n4-m. Hence n has to vanish for J > 0 and correlation functions with P insertion receive no contributions from instantons. On the other hand correlation functions of the type (Q - " Q) with any number of Q's receive a unique degree-1 instanton contribution and we obtain {Q- " Q) = exp(fg). Thus we find the genus 0 free energy of the CP^ model to be Fo = F^'-hF^'' = ^tltQ-he'^. (20) In the case of CP^ the situation becomes more complex. If we consider correlation functions of the type {P • " P Q " - QR - " R) inihQ small phase space with n P's,m Q's and i /?'s, ghost number conservation gives m-\-2l = 3d — l-\-n-\-m-\-i. When n = 0, there exist non-vanishing correlation functions for all d with I = 3d — l. Thus the genus 0 free energy has an expansion Fo = F^^ + Fo^" = ^tptl + ^tltR + f(tQ, tR), .3d-l /(^e,^/e) = E<^5^.'"^. (21) A^^ is the number of genus-0 degree-J instantons which pass through 3d — I fixed points on C P^. It can be computed recursively in the following manner (see Kontsevich and Manin [1994]). Let us introduce a basic set of two-point functions at genus 0 u = (PP)o, V = {PQ)o, w = {PR)o. (22) •)o denotes the genus-0 correlation function. In the small phase space they are give i3d-l)\ by u = tR, V = tQ, w = tp. Then we have {QQ)o = tp + E (^^ryy^'^ ^^"^^^ w 4- d^f(u, v)/dv^ where ,j3d-\ /(«'^)-E<'(^^''"- (23) Similarly we find (QR) = du^vf, (RR) = 9h/ in the small phase space. These are the so-called constitutive relations of the CP^ model. By making use of topological recursion relations it is possible to show that these relations actually hold in the large phase space (see Dijkgraff and Witten [1990]). Thus we obtain formulas {QQ)o = w + U, {QR)o = fuv, {RR)o = fuu. (24) Now consider the operator-product-expansion (OPE) relations QQ = {QQPhR + {QQQhQ + {QQR)oP = R + fvvvQ + fuw, QR = (QRP)oR + (QRQ)oQ + {QRRhP = fuwQ + fuuv, RR = {RRP)oR + {RRQ)qQ + {RRR)oP = fuuvQ + fuuu. (25)
182 T.EGUCHI By imposing the associativity of OPE, {QfR = Q(QR), Q(RR) = R(QR), we find the basic relation Juvv ^^ Juuu "T Jvvvjuuv' \^^) This leads to a recursion formula for the expansion coefficients A^^ <'=E<'<'^4'(^:2)-<3':0^ k+i=d (27) 10.2.7 Gravitational Quantum Cohomology In the following we are going to analyze topological a-models coupled to 2-dimensional gravity. Thus we study the theory in the presence of gravitational descendants (in the large phase space) and at arbitrary genus of Riemann surface. Such a system may be called gravitational quantum cohomology (see Eguchi et al. [1997a]). Some authors, however, are reluctant to introduce gravitational descendants and prefer to work within the small phase space. The introduction of infinitely many extra variables t^ may appear an unnecessary complications. We would like to stress, however, that these additional variables actually simplify the problem: in some sense they linearize the system so that we can write down a universal formula for differential operators which are bilinear in these variables and annihilate the partition function of the theory. We can then study the system by using the representation theory of Virasoro algebra. These constructions are universal and equally apply to arbitrary Kahler manifolds. If we wish, we may eliminate the descendant variables by making use of topological recursion relations. (Topological recursion relations (TRR's) were available only at genus 0 and 1 (see Witten [1990]). Recently, however, TRR has been obtained at genus 2 (see Getzler [1998] and Belorousski and Pandharipande [1998]) and conjectured at all genus (see Eguchi and Xiong [1998]).) This is exactly what we do when we compare predictions of Virasoro conditions with the known numbers of holomorphic curves. Contrary to our construction, the conventional quantum cohomology theory appears to be a difficult non-linear problem where each manifold has to be studied individually. 10.3 CP^ Model 10.3.1 Virasoro Operators Virasoro conditions take the form LnZ = 0, n = -l,0,1,2, ••• (28) where Z is the partition function related to the free energy of the theory as Z=exp(^x2«-2F^). (29) g=0
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 183 X is the genus expansion parameter. Operators L„ form the Virasoro algebra [L„, L^] = (n -m)Ln+m' The cohomology classes of the complex projective space M = CP^ are given by {1, (W, <w^, • • • , o)^} where co is the Kahler class. The corresponding fields are denoted as {Oa, a = 0,1, 2, • • • , A^} and their coupling parameters are given by [t^]. Gravitational descendants of Oa are denoted as OniOa), n = 1, 2, • • • and their parameters by [t^], Virasoro operators for CP^ are then defined by A^ oo 1 ^ a=Om=l a=0 A^ oo A^—1 oo ^tm^m-l,a-\-l a=Om=0 a=0 m=0 1 ^"^ 1 oo A^ N—a ^n = E E E (^ + D^C^-''^'", n)f«9„+„-^,„+,- n > 1 m=Oa=0;=0 (32) .2 N N-an-j-l + y E E E (^ + l)''0«)(m, n)dZdn-m-J-l,a+j a=0;=0 m=0 + :4 E (N + ir+h"t„+„+u a=0 where C^j\m, n) = (be, + m)(bcc + m + 1) • • • (^a + m + n) X m<^i<€2< and Dy>(m,n) = b'^ib'' + 1).. • (b''-hm)ba(ba + 1) • • • (Z.^ +n - m - 1) -\<ii<i2<-<ii<n-m-\ ^i=\ ^ ' ^/ }; iMi-^^)- (34) -m-l<€i<€2< Here dn,a = 9/9f" and Z?^ is related to the degree q^ of the cohomology class coa as ba^qa- \(N - 1), coa € /f2^«(M). (35)
184 T. EGUCHI The indices are raised and lowered as usual by the metric ^"^=5a+^,^. (36) Thus q^ =N - qcc, ba-\-b"" = 1. Note that the factor A^ + 1 which appears in the right-hand-side of Eqs. (31), (32) is the magnitude of the first Chem class of CP^, ci (CP^) = (N-\- l)co. Thus the terms jt = 0 represent the broken scale invariance of the a-model with the target manifold M which is a Fano variety. If one considers a fictitious manifold with a vanishing first Chem class with a fractional dimension N/N + 2, then it turns out that the above Virasoro operators reduce to the well-known expressions in the theory of KP hierarchy and the two-dimensional gravity coupled to minimal models (see Dijkgraaf ^f al. [1991]). Thus Eqs. (30-32) may be regarded as the generalization of Virasoro conditions of 2-dimensional gravity to the case of a well-defined target space. We also note that when A^ = 1 the above expressions reduce to those already obtained in the case of CP^ model (see Eguchi and Yang [1994]). The Virasoro algebra [Ln,Lm] = (n-m)Ln-\-m, n,m>-l, (37) may be verified by making use of identities as i J2 Ct'^k, n)C'^Jl_j{k + n-i+j, m) = {ba+k + n)C^\k, n+m) + C^'^Hk, n + m), (38) i J2 Dt'\K n)C^Jli_j(n -k-i-hj-hm) j=0 = (ba+n-k- l)D^J\k, n + m) + D^-^\k, n + m). (39) Our original derivation of the Virasoro conditions (28) uses the algebraic recursion relations for genus-0 correlation functions derived in Eguchi et al. [1997a]. See Eguchi et al [1997b] and Getzler [1996] for details. 10.3.2 Number of Rational Curves We have made an extensive check of the Virasoro conditions Eq. (28). For the sake of illustration we consider the case of CP^ and the equation LiZ = 0. We denote the three primary fields corresponding to the cohomology classes l,a),a)^ as P, Q, R and denote their couplings ast^,t^,t^. The genus-0 free energy is given by (21). In the small phase space with ^" = 0 for all a and n > I except for fj^ = — 1, the LiZ = 0 equation reads as - l(cr2(P))o + ^t^(cTi(R))o + lt^(cri(Q))o - ^t""{ai(P))o - 6{ai(Q))o 4 4 4 4 + 6tQ(R)o - 9{R)o - l{P)o(R)o + ^(2)0(2)0 + ^(^'')' = 0. (40)
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 185 We take the second derivative of (40) in f*2 and set f ^ =t^ =0. We obtain - \{02{P)QQh + ^t^{oi{R)QQ)o + \{cr\{Q)Q)o - 6{ai(Q)QQ)o (41) + U{RQ)o - 9{RQQ)o - ^{PQQ)o{R)o - l{PQh{RQ)o - l{Ph{QQR)o + \{QQh{QQh + \{QQQh{Qh = 0. In order to eliminate descendant fields in (41) we use the topological recursion relation (TRR) at genus-0 (see Witten [1990]), {an{Occ)XY)o = {cfn-x{0„)Op)o{OPXY)o (42) (X, Y are arbitrary fields) and also the "flow equations" (ori(P)P)0 = ««; + -V^ (ori(P)e)o = VW - f^ + ufuv + u/vt, (43) {ai(P)R)o = -u;2 + «/„„ + vfuv - fu, {cri{Q)Q)o = vfuv + ^(w + Uf. Flow equations are derived using TRR. Then Eq. (41) is rewritten as 3t''{RR}o + 6(QR}o - 9{QQR)o + {QhiQQQh + ^t^{QRUQQQh -^{QQ)o{QQQ)o + i{QQ)of = 0. (44) By using the expansion of the free energy (21) we can convert (44) into a relation for the instanton numbers. After some algebra we recover the result of Kontsevich and Manin (27) nP = (3^-4)! J]^^^-i^^^^|—^^2^[3H+£-2^]. (45) 10,3.3 Number of Elliptic Curves We may also consider the genus-1 instanton numbers and check against the recent results of Getzler [1996] and Caporaso and Harris [1996]. The genus-1 free energy of CP^ has an expansion —fQ ffR\^d We again consider the equation L\Z = 0 for simplicity. This time we use the TRR at genus-1 (see Witten [1990]) {crn(Oa))i = ^{crn-i(Oa)O^Oho + {crn-i(0^)0^)o{Ohi (47)
186 T.EGUCHI and the flow equation Eq. (43). After some algebra we find (2)0(01 + 3f^(e/?>o(e>i + It^'iQQRh - liQQQh 8 4 - 6{QQ)o(Q)i + ^(22)0 - 9(/?>i = 0 (48) where (• • •)! denotes genus-1 correlators. Using the free-energy expansion (46) we find 1 A^^^^ A^^^^ I <^ = ^Pj^did - Did - 2) + ^ (3J - l)!-_^_|^-(3^2 - 2k), (49) The above equation has a form somewhat different from the one of Getzler [1996], yet nonetheless predicts identical instanton numbers (see Pandharipande [1997]). We can also check that the genus-1 instanton numbers of CP^ are correctly reproduced by the Virasoro conditions. 10.3.4 Virasoro Operators for the Negative Branch It is interesting to see if we can construct the negative branch L_„, n > 2 of Virasoro operators and compute the central charge of the algebra. It turns out that in the case of CP^ with A^ =even it is possible to construct {L_„}. They are given by oo L-^Yl E^^ + iyAiJ\m,n)C+„+jdm.a+j m=0 a,j + y E E (^ + ^yB^J\m, n)t^_^^j_,tm,a-,j, n>l. (50) a,j m=0 The coefficients are defined by A<J\m,n) = {-iy ^- iba+m + j + l)(ba + m + j+2)' ^ £,..., (n^:^^;^^)- "■> ■■■(ba+m+j + n-1) j^^_^^ and Bi^\m,n) = (-1^77^ .,,,„ ^ , r^ -ttttt \ : {b" - j){b"+ \-j)---{b"+m-\-J i„ + j)ibce + \+j)---{bcc+n-m-2 + J\<i,<i,h<ej<n-2 \U ba+j-m + ii) (52)
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 187 We can check the entire algebra and find that the central charge is given by c = A^4-l. (53) This suggests that there exists a realization of the algebra by means of A^ + 1 free scalar fields. In the case of A^ =odd some factors in the denominators of A^ \m, n), B^ (m, n) vanish and the above expressions become singular. We do not understand the origin of this disparity between even and odd values of A^. 10.4 General Kahler Manifolds We now turn to discussion of the case of general Kahler manifolds M. We first note that the L_i operator of (30) is universal and does not depend on the data of the manifold M. The L_ 1Z = 0 gives the well-known string equation. The Lq operator of (31), on the other hand, is a formula of Hori [1995] adapted for the case of CP^. In the case of general manifolds Hori's equation reads as m=0 ^^m m=l K-l ^^ 1 /3-dimM f \ + ^( ^ xiM)- J^ci(M)cdimM-i(M)\ (54) where x(^) is the Euler characteristic of M and C denotes the matrix of the first Chem class of M defined by (CI a^= I Ci(M) ACOcc A COp. (55) JM (54) may be derived using the intersection theory on the moduli space of Riemann surfaces. The form of Virasoro operators in CP^ now suggests a natural generalization for a general Fano manifold. We consider OO Ln = Y.llY. <^a-''*('"' ")(^ ')JC^m+n-j,P « > 1 (56) m=0 a,p j + y E E E ^«'('»' «)(^ ^')/a^9n-m-;-M + ^2 T,(c "+')Jt%, a,p j m=0 ap OO L-n = Y.mi ^«''('"' «)(^ ')JC^n+j^m,p, « > 1 (57) m=0 a,p j + y E E E ««■'■'('"' «)(^ ^')/C.+;-l'm,^. ci,p j m=0 (dimM-1) 1 bc( = qa , Qa — - degree of differential form co^ e H (M)
188 T.EGUCHI where C ^ is the 7-th power of the matrix C and a, fi run over the cohomology classes of M. Equations (56, 57) reduce to (32, 50) in the case of CP^. 10.4.1 Equation for Betti Numbers The operators (56), (57) (together with Lq of (54)) satisfy the Virasoro algebra if and only if the following condition is obeyed -^T.^^- = Ya\ 2 X(M)-y^ci(M)AcdiinM-iW)- (58) (This follows from [Li, L_i] = 2Lo). Equation (58) is an interesting relation depending only on the geometrical data of M. It is easy to check that the formula holds in various Kahler manifolds, i.e. projective spaces, rational surfaces, Grassmannians etc. In fact there is a theorem (see Libgober and Wood [1990]) that (58) holds if and only if the manifold has only analytic classes (i.e. classes of the type H^'P in the Hodge decomposition) as in the case of projective spaces. In the general case when "off-diagonal" classes if ^'^, pt = q arQ present, one has instead > E(»-^-^)(^-^)'.''(->)-'=hm^L-'-L'"^'-')- <''> (n is the dimension of M). Comparison of (59) with (58) suggests that one should define the factor ba in the Virasoro operators asba = p — (n — l)/2 for a cohomology class coa € HP'^(M)(p ^ q). One may as well define ba = q — {n — \)/2 since h^'^ is symmetric in /?, q and the left hand side of the above equation remains invariant under /?, q interchange. Let us call these values of ba as the holomorphic and anti-holomorphic dimensions. Based on the above observation S. Katz [1997] has proposed to define holomorphic (anti-holomorphic) Virasoro operators where the factors ba are given by the holomorphic (anti-holomorphic) dimensions. The sign factor (—1)^+^ in (59) is taken care by the Fermi statistics in the case of cohomology classes with p + ^=odd which are described by anti-commuting variables. Predictions of the holomorphic (anti-holomorphic) Virasoro conditions have not been studied yet in detail. In Eguchi et al. [1998] we made a sample calculation in the case of a cubic hypersurface in CP^ and reproduced known instanton numbers up to genus 0 and 1. 10.5 Free Field Representation Let us next come to a free-field realization of the Virasoro operators. We may assemble variables [t^} into field operators {O"} and rewrite Virasoro operators as their bilinear form. It turns out that we have a free bosonic field for each even cohomology class W^^^ and a free fermionic (ghost) field for each odd cohomology class H^^^. Together they form a conformal field theory (CFT) of central charge c — x^M).
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 189 10.5.1 Bosonic Fields We first consider cohomology classes coa € if ^'^ with /?+^=even so that the corresponding parameters [t^} are ordinary commuting numbers. By assembling these into Fourier series we obtain a basic set of bosonic fields 0", _ V^ I (g> +n) Q^ j^_iyci a _ r.01 01 _ f i\n^a », ^ n (f^CW = -Z^ p.^c^x ^n^ ' ^n=%^ ^_n-l = (-1) ^n' " ^^- (^O) Here z is an arbitrary complex variable (not to confused with the local coordinate of the Riemann surface in section 10.2). It is easy to see that the scalar fields {0"} have a canonical OPE. In terms of the currents, j^(z) = az0"(z) it is given by ja(z)j^(w) ^ " (61) (z — wY As usual the stress tensor is defined as the generating function of the Virasoro operators For convenience we introduce a generalization of the currents j"(z) by an additional parameter e •a. . . Y^ r(Z7^+n + l-6) ^ ^„_^«_i+. (63) We then find that the stress tensor can be represented as IbosonKZ) N a N a,i6=0 m,£ ^ riba-m-l + e) (=0 + ^(E^"^«)z-^ (65)
190 T. EGUCHI Derivatives in e of the ratio of gamma functions in (65) reproduce the coefficient functions A, B,C, D of Virasoro operators. Since the stress tensor involves derivatives in 6, the basic currents j"(z) themselves do not have a good conformal property. Instead we consider new operators defined by «(-ir+ir(Z7« + 6) 6=0 a = ^(expCa,) ^=0 ^ r(bp-m-l-\-€) <. (66) 6=0 After some algebra we find the following OPE Ja(z)J^(w) is regular for a > fi. These currents /« are primary fields with respect to the stress tensor Tboson(z)r(w) ^ --^r(w) + —L-^du^r(w), (68) (z - wY (z - w) The stress tensor itself is expressed as a bilinear form in terms of /« Ttosoniz) = \J2'' G"^J-(z)Mz) : + J(^ b,b-)z-\ G"^ = rj-^ - C"^. (69) Thus we have a standard Sugawara form for the stress tensor (if necessary, we may further rotate the currents /« and bring G"^ to the unit matrix). We should note, however, the operators Ja(z) contain powers of logarithms of z and are not single valued. Logarithms come from the derivative in e hitting the exponent of z in (66). We may define currents /« which are single valued as (these operators were denoted as /« in Eguchi ^f ^/. [1998].) "(-ir+^nb^+e) p=o ^ T{bp-m-\-\-e) aP,z-'"-^'-\ (70) €=0 Then the currents Jaiz) have a modified OPE and conformal property as Uz)P{w)^(-^—\ ( ^ +^"^"~^^^ ly p>a, (71) \l-CJ„ \(z-wy 2{z-w) wj Tiz)riw) « -J—^/«(„;) + —1— id^J"(w) + J2 (t^) " b/-^ (z-wY (z-w)l ^\\-CJp '^ w (72)
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 191 The stress tensor itself, however, is expressed in exactly the same manner as /« 1 2 Tbosoniz) = IJ2'' G"^Jc.(z)Mz) : A(J2b^b'')z-\ (73) Thus our stress tensor Thoson is somewhat different from that of standard CFT: primary fields are multiple-valued while single-valued fields are not exactly primary. This is due to the mixing of the variables {t^} in Virasoro operators caused by the non-vanishing of the 1st Chem class of M. When ci(M) vanishes as in the case of K3 surface or Calabi-Yau manifolds, mixing disappears and the basic fields 0" themselves become primary fields. 10.5.1 Ghost Fields Now we consider odd cohomology classes coa € if ^'^ with p + ^=odd. In this case the parameters coupled to co^ are anti-commuting numbers 0^. Odd classes always occur in pairs as coa € if ^'^ and co^ € H^'P. We thus obtain a pair of anti-commuting variables {Oa,n}, {On) for ^ach odd class. By collecting these variables together we construct ghost fields „ror(^"+« + i) +4^ r(M 30n,/ ' ^'^^ Here ba denotes the holomorphic dimension pa — (dim M — l)/2 of (W^. OPE's are given by bAz)cHw) = S^-^(^r ^ 8t (—^ + -) . (76) z — w w \{z — w) w J The stress tensor of the be system is defined as T,host{z) = Y,:d,b^{z)c''{z)'.-]^{Y,b^b^)z-^. (11) a a The total stress-tensor is given by the sum of (69) and (77). As is well-known, a free boson contributes 1 to the central charge of the Virasoro algebra. On the other hand, a pair of be ghosts (of spin 0 and 1) contributes —2 to the central charge. Hence the odd-dimensional cohomologies contribute — (number of odd classes) in total. Thus the Virasoro central charge equals c — number of even classes — number of odd classes = x (M) (78) for a general manifold M where x(M) is its Euler characteristic. 10.6 Topological Recursion Relations In order to test the predictions Virasoro conditions against the data on the number of holomorphic curves we need TRR's which convert descendant correlation functions into
192 T. EGUCHI those of primaries as we have seen in section 10.3. (It seems unlikely that the descendant correlation functions are evaluated directly by geometrical methods.) At the time of our proposal of Virasoro conditions TRR's were available only at genus 0 and 1. Recently, however, genus-2 TRR was obtained by Getzler [1998] and further discussed in Belorousski and Pandharipande [1998]. Its derivation is based on the analysis on the linear dependence among homology cycles in the moduli space of stable maps. In Eguchi and Xiong [1998] we proposed a somewhat different form of TRR at higher genera starting from a simple assumption on the dependence of higher-genus free energies on genus=0 primary correlation functions. Our recursion relation at genus g involves gravitational descendants of degree n,n — I,-- ,n — 3g -\- 1. In the case of genus 2 descendants of degree n,n — I,-- ,n —5 appear and thus our relation is weaker than that of Getzler's which involves descendants of degree n,n — l,n —2. Nevertheless, these TRR's can be used together with the Virasoro conditions in order to completely determine the number of holomorphic curves of arbitrary degree and genus. In the case of CP^ at genus g = 2, 3 we explicitly verify that our TRR and Virasoro conditions reproduce the known results of Vainsencher [1993], Itzykson [1994] and Caporaso and Harris [1996]. 10.6,1 TRR at Higher Genus Our basic assumption is that the genus-g free energy of topological string theory is a function depending only on the primary multi-point functions of genus g = 0. Let us consider the case of a theory with primary fields {Oa}(o( — 0, 1 • • • , A^) coupled to the perturbation parameters {f"}. Gravitational descendants and their couplings are denoted as {<yn(Pa)\ and {f"} (n = 0,1, 2, • • •), respectively. We define genus=0 correlation functions as Maia2..«; = (^^ai^a2 * * * ^a^ )o where P denotes the puncture operator C>o. Then our assumption is that the genus-g free energy Fg is a function only of the genus=0 correlation functions Uoi,,Ua,a2. • • , Ua.a^-oi^.-x Fg(t) = Fg{Ua,{t), Ua^ajO)^ '" ^ Wda^-ag.-i (0), ^ > 1- (79) Here t stands for all the couplings [t^] of the large phase space. Note that the dependence on the parameters t of the free energy Fg occurs only through the functions Ua^aj-.a (J — l,2,...,3g-l). Equation (79) is known to hold in the 2-dimensional pure gravity theory. For instance, the genus 1,2 and 3 free energies are given by Dijkgraaf and Witten [1900] and Itzykson and Zuber [1992] Fi = -l0g., ^^ = ^^^4-J^^^^-[Y^2^ (80) 5u''^ 59u''\^^^ 83w''^w(3)2 ^^^(3)3 83^//3^(4) X213u'u^^^u^^^ f^ — 1 1 1 648m'^ 3024m'^ 7168m'^ 64512m'^ 15120m'^ 322560m'^ __103w^ 35?>U''\^'^ 531.(3)^(5) ^^.^(6) ^(7) 483840m'4 322560m'^ 161280m'^ 46080m'^ 82944m'3
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 193 where M = {PP)o and ' denotes the f^-derivative. Eq. (79) is also known to hold in some cases of the 2-dimensional gravity coupled to minimal matter at lower genera (see Eguchi et al. [1995]). It is also valid in the case of CP^ model (Eguchi and Yang [1994]). Eq. (79) means that higher genus amplitudes are expressed in terms of the genus=0 data and suggests a possible reinterpretation of the world-sheet topological theory as a field theory on the target space (see Itzykson and Zuber [1992] and Eguchi et al, [1995]). We now assume that (79) is a universal feature of 2-dimensional topological field theories coupled to gravity. It is then easy to derive our TRR's. Let us first consider for simplicity the case of genus=l. Genus-1 free energy depends on {ua} and {m^^}. We then have dF, du^ dF, du^, dF, dF, d{an(Oa)O.P)o dF, At n = 0 Eq. (82) becomes (a>i = {OccO^P)o^ + {OccO^O,P)o^. (83) We use the genus=0 TRR {an{Oa)AB)o = {Gn-xOyUO^ AB)o (84) to rewrite (82) as dFx {crn(Oa))l = (crn-l(Oa)0^)o(O^O^P)o du^ + ((crn-i(Oa)0^)o{0^0^0,P)o + (crn-i(Oa)0^0^)o(0^0,P)o) ^^' dujj^v (85) Putting n = 1 in (85) gives dFi du^ + ({0^0^)o{0^0^0,P)o + {O^O^Op)o{0^0,P)o)^ = {OceOpUOhi + {OaO^Ofi)o{Of'0,P)o^ (86) where (83) has been used. By making use of (83) and (86). Eq. (85) is then reexpressed as (or„(0„)>i = (a„-i(a)O^)0(O^>i + {cr„-2(0„)0y)o({<ri(0y))i - {OWpUOl'h). (87) Eq. (87) is our TRR at genus=l. It appears somewhat different from the standard TRR (see Witten [1910] {On{Ooc))x = j^{(rn-i(0„)OfiOf)o + {<r„-i(Oc)Op)o{Of)i. (88)
194 T. EGUCHI However, when one uses the structure of the genus=l free energy Fi = — log dti(uap) + /i (ua) (89) one may easily check (87) and (88) are equivalent. By repeating the same procedure as above we can derive the TRR at genus=2 + {crn^i(Oa)0^)oA^^ + {an(Oa)Op)oA^^ (90) where A^o ^ {0^2 (91) A^,^(ai(0^))2-(O^Oy)oAl (92) A^ ^ {a2(0^))2 - {cri(0^)Oy)oAl - {O^Oy)o • A\ (93) A^ ^ (or3(C^^)>2 - {a2(0^)Oy)oAl - {ai(0^)Oy)o • A\ - {O^Oyh • A^ (94) A^ = (or4(C^^)>2 - {cr3(0^)Oy)oAl - {a2(0^)0y)o • A[ - {ai(0^)Oy)o • A^ -(C^^C^^>o-A3^ (95) Thus all the descendants {(JniOa), n > 5} may be eliminated in favour of {cri(Oa), i — 1, 2, 3, 4} at genus g = 2. Similarly TRR's at arbitrary genus (g > 1) are given by 3^-2 (0r„+3^_l(a)>^ = Y. (^n+3^-2-;(aC^^))0A^^ (96) ;=o aI ^ {Oh,. (97) A^. = {a^iPh), - Y.{a^(PhOh^A]_^_^, (98) . k=^ Thus the descendant degrees are reduced to n < 3g — 2 at genus g. These TRR are not quite as efficient as the standard TRR at genus 0 and 1 which reduce the descendant degrees all the way to zero. However, they are powerful enough to determine instanton numbers of arbitrary genera when combined together with the Virasoro conditions. We present the result of the calculation of instanton numbers in CP^ up to genus 3 in the Appendix. 10.7 Discussion Recently, by assuming the Virasoro conjecture for constant maps Getzler and Pandharipande [1998] obtained strong results on the intersection theory of the Hodge bundle on Mg^n which in part reproduce previously conjectured results (see Faber [1997]). Dubrovin and Zhang, on the other hand, proposed a generalization of the Virasoro conjecture for general 2-dimensional topological field theories (Dubrovin and Zhang [1998]). It seems that there is now strong empirical evidence for the Virasoro conjecture of quantum cohomology although its complete mathematical proof may not become available for some time.
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 195 Finally I would like to draw attention to a striking analogy between our result and the formula for the Betti numbers of the Hilbert schemes of points which featured in recent studies on string duality. The well-known formula of Gottshe [1990] for the generating function for the Poincare polynomial of the Hilbert scheme of points reads as Here M is a smooth projective surface and bii — 0, • • • 4 are Betti numbers of M. mS^^ denotes the Hilbert scheme parameterizing n points in M and Pt{M^^^) is its Poincare polynomial, Pt(M^^^) = X!/ t^hi(M\^^). If one puts ^ = 1, one obtains the sum over the number of cohomology classes of M^^^ E^-£M"")= "-■,"_;ytll- 000) We recognize that the system is described by Z?o + ^2 + ^4 = beven free bosons and b\ + Z?3 = bodd free fermions. On the other hand, if we set f = —1, we obtain WTien t is set to —1, fermions are converted into ghosts and we have a partition function of a CFT with a central charge=x(M). Thus we find an identical CFT as ours with beven bosons and bodd ghosts describing the cohomology of M^^^ (formula (101) also holds for higher-dimensional manifolds if one uses the orbifold Euler number (see Dijkgraaf et al. [1997]). Therefore an isomorphic CFT describes both (i) holomorphic curves in M and (ii) cohomology of the symmetric products of M. This is an extremely curious coincidence worth further study: it may possibly be explained by some kind of duality relating the moduli space of curves in M with that of vector bundles on M. Acknowledgement I would like to thank K. Hori, C.S. Xiong and M. Jinzenji for their collaborations. This work is supported in part by the Grant-in-Aid for Scientific Research on Priority Area 707 "Supersymmetry and Unified Theory of Elementary Particles", Japan Ministry of Education. 10.8 Appendix Instanton Numbers ofCP^ N^ denotes the numbers of degree d and genus g curves
196 T. EGUCHI genus 0 A^. (0) = 1, A^f ^ = 1, A^f ^ = 12, A^f ^ = 620, A^f ^ = 87304, A^f ^ = 26312976, A^^^^^ = 14616808192, A^g^^^ = 13525751027392, A^^^^^ = 19385778269260800 genus 1 A^{^^ = 0, A^2^^^ = 0, A^3^^^ = 1, A^^^^ = 225, A^5^^^ = 87192, A^^^^ = 57435240, .(1) ,(1) .(1) A^)^^ = 60478511040, N^'^ = 96212546526096, A^9''' = 220716443548094400 genus 2 A^P> = 0, A^f ^ = 0, A^f ^ = 0, A^f ^ = 27, A^f ^ = 36855, N^^^ = 58444767, A^f ^ = 122824720116, N^^^ = 346860150644700, N^^^ = 1301798459308709880 genus 3 A^P> = 0, A^f ^ = 0, A^f ^ = 0, A^f ^ = 1, A^f ^ = 7915 A^f ^ = 34435125, A^f ^ = 153796445095, N^^^ = 800457740515775, A^9^^^ = 5039930694167991360 N^^^ = 38747510483053595091600 References Belorousski, P., and Pandharipande, R., A Descendent Relation in Genus 2, math/9803072 (1998). Bershadsky, M., Cecotti,'s., Ooguri, H., and Vafa, C, Comm. Math, Phys, 165, 311 (1994). Bershadsky, M.,Cecotti, S., Ooguri, H., and Vafa, C, Nucl Phys, B405, 279 (1993). Borisov, L.A., On Betti Numbers, and Chem Classes of Varieties with Trivial Odd Cohomology Groups, alg-geom/9703023 (1997). Candelas, P, de la Ossa, X.C, Green, PS., and Parkes, L., Nucl, Phys, B359, 21 (1991). Caporaso, L., and Harris, J., Counting Plane Curves of Any Genus, alg-geom/9608025. Invent. Math. 131, 345-392 (1998). Collino, A., and Jinzenji, M., On the Structure of Small Quantum Cohomolgy Rings for Projective Hypersurface, Hep-th/9611053 (1996); Commun, Math. Phys, 206, 157-183 (1999). Di Francesco, P., and Itzykson, C, Quantum Intersection Rings, in Dijkgraaf, R., Faber, C, and G. van der Geer eds.. The ModuU Space of Curves, Birkhauser, Boston-Basel-Berlin (1995). Dijkgraaf, R., and Witten, E., Nucl Phys. B342, 486 (1990). Dijkgraaf, R., Moore, G., Verlinde, H., and Verlinde, E., Commun, Math. Phys. 185, 197 (1997). Dijkgraaf, R., Verlinde, E., and Verlinde, H., Nucl. Phys, B348, 435 (1991). Dubrovin, B., and Zhang, Y., Frobenius Manifolds, and Virasoro Constraints, math/9808048 (1998); Selecta Math. 5, 423-466 (1999). Dubrovin, B., Geometry of 2D Topological Field Theories, in: Integrable Systems, and Quantum Groups, Montecalini, Terme, 1993. M. Francaviglia, S. Greco eds.. Springer Lecture Notes in Math. (1996).
WORLD-SHEET INSTANTONS AND VIRASORO ALGEBRA 197 Eguchi, T., and Xiong, C.S., Adv. Theor Math. Phys. 2, 219 (1998). Eguchi, T., Hori, K., and Yang, S.-K., Int. J. Mod. Phys. AlO, 4203 (1995). Eguchi, T., and Yang, S.-K., Mod. Phys. Lett. A9, 2893 (1994). Eguchi, T., Hori, K., and Xiong, C.S., Int. J. Mod. Phys. All, 1743 (1997a). Eguchi, T., Hori, K., and Xiong, C.S., Phys. Lett. B402, 71 (1997b). Eguchi, T., Jinzenji, M., and Xiong, C.S., Nucl. Phys. B510 [PM], 608 (1998). Eguchi, T., Yamada, Y, and Yang, S.-K., Rev, Math. Phys. 7, 279 (1995). Essays on Mirror Manifolds, ed. S.-T. Yau, International Press (1992). Faber, C, A Conjectured Description of the Tautological Ring of the Moduli Space of Curves, math/9711218. Aspects Math. E33, Vieweg, Braunschwerg (1997). Fukuma, M., Kawai, H., and Nakayama, R., Int. J. Mod. Phys. A6, 1385 (1991). Getzler, E., and Pandharipande, R., Virasoro constraints, and the Chem classes of the Hodge bundle, math/9805114. Nucl Phys. B530, 701-714 (1998a). Getzler, E., Intersection Theory on M\,4, and Elliptic Gromov-Witten Invariants, alg-geom/9612004. J. Amer Math. Soc. 10, 973-998 (1997). Getzler, E., Topological recursion relations in genus 2, math/9801003 (1998). In "Inetgrable Systems, and Algebraic Geometry", World Scientific (NJ) (1998b). Givental, A.B., A Mirror Theorem for Toric Complete Intersections, alg-geom/9701016 (1997); Prog. Math. 160, Birkauser, Boston MA (1998). Givental, A.B., Equivariant Gromov-Witten Invariants, alg-geom/9603021. Intemat. Math. Res, Notices 13, 613-663 (1996). Givental, A.B., Stationary Phase Integrals, Quantum Toda Lattices, Flag Manifolds, and the Mirror Conjecture, alg-geom/9612001 (1996). Gottsche, L., Math. Ann. 286, 193 (1990). Hori, K., Nucl Phys. B439, 395 (1995). Itzykson, C, and Zuber, J.-B., Int. J. Mod. Phys. AT, 5661 (1992). Itzykson, C, Int. J. Mod. Phys. B8, 3703 (1994). Jinzenji, M., and Sun, Yi., Int. J. Mod. Phys. All, 171 (1996). Jinzenji, M., On Quantum Cohomology Rings for Hypersurfaces in CP^~\ hep-th/9511206. J. Math. Phys. 38, 6613-6638 (1995). Katz, S., unpublished, March 1997. Kontsevich, M., and Manin, Yu., Comm. Math. Phys. 164, 525 (1994). Libgober, A.S., and Wood, J.W, J. Diff. Geom. 32, 139 (1990). Liu, X., andTian, G., Virasoro Constraints For Quantum Cohomology, math/9806028. J. Differential Geom. 50, 537-590 (1998). Pandharipande, R., A geometric construction of Getzler's relation, alg-geom/9705016. Math. Ann. 313,715-729(1999). Vainsencher, I., Enumeration of n-fold Tangent Hyperplanes to a Surface, alg-geom/9312012. J. Algebraic. Geom. 4, 503-526 (1995). Witten, E., Commun. Math. Phys. 118, 411 (1988). Witten, E., Nucl Phys. B340, 281 (1990).
11 Dispersionless Integrable Systems and their Solutions YUJI KODAMA Department of Mathematics, Ohio State University, 231 West 18th Avenue, Columbus, OH 43210, USA E-mail: kodama @ math, ohio-state, edu This chapter concerns the general properties of dispersionless integrable systems in their connections to topological field theory. We consider the dispersionless KP hierachy as a prototype example of the system. A solution to the hierarchy determines a Frobenius manifold in a parameterization with the flows of the hierarchy. We also discuss an example of a degenrate Frobenius submanifold. 11.1 Introduction In this chapter, we summarize a basic method to find expHcit solutions of dispersionless integrable systems, and discuss their fundamental connections to topological field theory (TFT). We will pay particular attention to the dispersionless KP (dKP) hierarchy with certain finite dimensional reductions such as the A^-component vector nonlinear Schrodinger (NLS) equation. Most of the basic results presented here can be found in the papers (Kodama and Gibbons [1990] and Aoyama and Kodama [1996]). We also discuss an embedding of a degenerate Frobenius manifold recently discussed by Strachan [1999], and suggest a possible stratification of the Frobenius manifold based on the zeros of the t-function (or the singularity of the free energy). This chapter is organized as follows: In Section 11.2, we begin with a summary of the general properties of integrable systems of hydrodynamic type (Dubrovin and Novikov [1989]) which are a special class of 1st order quasi-linear system of pdes. The integrability is then defined as the existence of infinitely many symmetries which are expressed as commuting flows of the system. Then the solution 199
200 Y. KODAMA method based on the decomposition of the higher commuting flows is presented, and the basic properties connecting this to TFT are summarized. The symmetry parameters in the commuting flows then give the coupling constants in the TFT, and finding a particular solution of the dKP hierarchy determines a Frobenius manifold which characterizes the theory. In Section 11.3, we give a brief summary of the KP theory placing particular emphasis on the T-function (Date et al. [1983]), and then formulate the dKP hierarchy as a quasi-classical limit of the KP hierarchy. The main feature of the dKP theory is that the spectral problem of the Lax operator of the KP hierarchy provides a genus zero algebraic curve giving an invariant level set of the dKP hierarchy. The dKP hierarchy is an integrable system of hydrodynamic type, and as a consequence of the quasi-linearity one can treat all the symmetry parameters on an equal basis. This unifies two of the main dispersionless hierarchy, the dKP and dispersionless (or continuous limit of) Toda hierarchy (Kodama [1990] and Aoyama and Kodama [1996]). In Section 11.4, the main concern here is to discuss an example of a degenerate Frobenius manifold recently discussed by Strachan [1999]. The degeneracy in this example can be obtained by a singular reduction of a nondegenerate higher dimensional Frobenius manifold. The nondegenerate example discussed here is given by the dispersionless N-component NLS equation (Zakharov [1980] and Gibbons [1981]). We also expect a nested structure of the Frobenius manifold similar to the case of the Birkhoff strata of the Grassmannian for the KP hierarchy (Pressley and Segal [1986] and Adler and van Moerbeke [1994]). 11.2 Integrable Systems of Hydrodynamic Type Let y be a vector valued function of (jc, t) on C^ which satisfies the following equation of hydrodynamic type, ^^A^ (1) : dt dx where A € Mnxn(C) is a function of V only. The symmetry of the system (1) plays an essential role in the method of solution, and it is defined by dv dv Definition 11.1 A quasi-linear system — = B— with B e Mnxn(C) is a symmetry ds dx of the system (1), if it commutes with (1), that is, — = —• (2) dtds dsdt Here the variable s represents the symmetry parameter Since A does not depend on the coordinates x explicitly, we have jc-translational invariance, i.e. with B = I,ihc N X N identity matrix, ^ = ^ (3) 3fO dx ^ '
DISPERSIONLESS INTEGRABLE SYSTEMS 201 In this chapter, we consider a system having an infinite number of symmetries, which we call an integrable system of hydrodynamic type. We write the system in the form of symmetry, dv dv — =An — , forn =0,1,2,..., (4) dt^ dx in which the first N coefficient matrices {A„}^~ are assumed to form a basis of all commuting matrices with Ai = A (we also assign Aq = I,i^^NxN identity matrix). Note here that the commutativity of the matrices A„ is necessary but not sufficient for the flow with An being a symmetry. Then by means of the Cay ley-Hamilton theorem, the commuting matrices A„ can be also expressed as the polynomials of the reference matrix A in (1), i.e. N-l A„:=0„(A) = ^4"\A)^ (5) k=0 where aj^ are scalar functions of V. Thus we have a commutative ring of polynomials, where Ca(0 is the characteristic polynomial, Ca(0 = det(f / — A). The commutative matrix algebra having the basis {A„}^~^ forms a Frobenius algebra with a nondegenerate inner product given by the trace of the matrices, (A,A,>:=tr(A,A,). (7) In the generic case of diagonalizable matrices, one can write N where f„ are the eigenvalues of At. With the polynomial expression of A„ in (5), we have the residue formula, ' 27TiT^=^ Ca(0 c,in=o\ Ca(0 J In terms of the TFT, each matrix A„ in the basis gives a primary field representing the matter fields, and the algebra of the fields N-l AnAm = Y^C^nmAkAo, (10) k=0 gives a fusion rule with the structure constants c^^. Taking the inner product, we have N-l Cnml := {AnAmAi) = ^ C^^ (Ai^A/Ao). (11) k=0
202 Y. KODAMA which ^ive the 3-point functions of the TFT. In particular the 3-point function ckio defines a metric rjki providing a matrix of lowering the indices (the inverse of the matrix ('nki)~^ •= (^^0 then provides the matrix of raising the indices), i.e. rjki := (AkAiAo). (12) These variables Cnmi, Vki are the functions of the coupling constants tn through the variable y, that is, we have a deformation of the Frobenius algebra by an integrable system (4). In particular, if the metric r]ki is constant, the deformation defines a Frobenius manifold introduced by Dubrovin [1996]. An explicit solution V can be obtained as follows: Through the higher order symmetries, we have: Theorem 11.1 Kodama and Gibbons [1989] A solution of the systems (4) can be obtained in the following implicit form (hodograph type), t-^Y^CkUl^ (13) k=0 where Ck are arbitrary constants, and fx^ are the coefficients in the expansion of the matrix Af^^k if^ the primaries {A„}q ~ , N-l n=0 Remark 11.1 The solution formula (13) is identified as the string equation of the corresponding TFT, and the expansion (14) gives the so-called topological recursion relation (see Dijkgraaf and Witten [1990] and Aoyama and Kodama [1996]). We will give a more precise connection of these relations in the case of Landau-Ginzburg model associated with the dispersionless KP hierarchy in the next section. Remark 11.2 A formula equivalent to (14) was obtained by Tsarev [1985] in somewhat different way which is called the generalized hodograph method. His formula is x-\-^j^t = Vk, forA: =!,••• ,N, (15) where ^k and Vk are respectively the eigenvalues of the matrices A of the flow (1) and B of its symmetry (both flows are assumed to be diagonalizable hamiltonian systems of hydrodynamic type). 11.3 The Dispersionless KP Hierarchy Here we present the basic structure of the dKP theory as a quasi-classical limit of the KP theory. Let us begin with a brief summary of the KP theory.
DISPERSIONLESS INTEGRABLE SYSTEMS 203 113.1. The KP Theory Let L be a formal pseudo-differential operator given by L = 8 + ^f/^8-(^+i\ (16) /=o where U^ = U^ (X, Ti,T2,'"), the symbol d implies the derivative with respect to X, and d~^d = dd~^ = I. The operation with d^ is given by the generalized Leibnitz rule, ''^■=§0 ^^^rd'-J' , i gZ. (17) 5" = —^[L"+^]+ . (19) 7=0 The KP hierarchy is then defined by the so-called Lax formula, IL = IB\L]:= B'^L - LB\ n=0,h'' (18) where the differential operator 5" is given by the differential part of L^~^^/(n H-1), denoted by 1 The commutativity of those flows is just a direct consequence of the definition of B^ in (19) (Date etal. [1983]). The Lax equation (18) with (16) gives a system with an infinite number of the variables {U^}q^. There are several finite dimensional reductions of the equation (18) to 1 -h 1 soliton equations. For example, the N-KdV equation can be derived by the condition (Gelfand-Dickey reduction [1975]), [L^+i]+ = L^+i = 8^+1 + vid""-' -^'--^VN, (20) (when N = 1 this corresponds to the KdV equation while N = 2 yields the Boussinesq equation). The reduced L given by N L = D-\-Y^ VPk(^ - Uk)~^^, with D = id (21) k=i provides the A^-component NLS equation for the vector valued function q = (qi,-- - ,qN)^ with qk = y/^exp(i f^ UkdX), where q^ = (qi,- - , qN) (see for example Kodama [1999]).
204 Y. KODAMA The hierarchy (18) is also given by the compatibility conditions of the following linear equationsforthewavefunction^(ro, Ti, • • OwithJo = X and the iso-spectrality condition — -0- dTn = 5"^. (23) (24) The wave function ^ with prescribed asymptotic condition is called the Baker-Akhiezer function and is given by Date et al. [1983], ^(T,X) = T(T) where exp ^(J, X) is the free space solution with oo . ^(T,X) = T—-X-^'Tn, Here t(T) is the tau-function of the KP hierarchy, and r(T - [X-i]) :=r(To-^,Ti-^,^>\ = exp k=0 x^+i dTk r(7b,ri,---). (25) (26) (27) 11.3.2. The Dispersionless KP Theory The dKP hierarchy is defined as the "quasi-classical" limit of the KP theory (see for example Kodama [1988]). Let /i be a small parameter, and introduce the variables. tn := hTn, for n = 0,1, • • , u' =u'(to,tu'"):=U'(To,Tu"'), for / = 0,1, (28) (29) which lead to the replacement -^ = h-^. Then writing the wave function ^ in the WKB form, (30) ^(To,Tu'" ;X)= exp -5(^0,^1,... ; X) [, where the function S is called the action, and it plays a fundamental role in the dKP theory in the framework of the Hamilton-Jacobi theory (Kodama and Gibbons [1990] and Guha Takasaki [1998]). The expression of the action S can be found from the Baker-Akhiezer
DISPERSIONLESS INTEGRABLE SYSTEMS 205 function (25) in the dispersionless limit, ^(T, X) = — exp[^(r, X)] = —r exp -^(t, X) t(T) t(0 L^ J ^exp^-ga,X)+log .^^^ J = exp r^^a, X) + (e-^T.^ii; _ l^ iogf(01 = exp \^^(t, X) - ^ -l^i-(/^iogf(0) + Oih^)] (31) from which we can write oo . oo . s(t, x) = y -^x-^'tn - y TTZT—^t), (32) The function T{t) is the free energy (see for example Takasaki and Takebe [1995] and Carroll and Kodama [1995]), T{t) = lim h^logf(0 = lim h^logr (-] := logZdKP, (33) where the raw-function t^kp plays the role of the partition function of the TFT. The free energy can be also directly derived from the dKP hierarchy. The following argument for the derivation of the free energy is particularly important and necessary for the case of rational reduction of the dKP hierarchy (Aoyama and Kodama [1996]): With (25), the quasi-classical limit leads to, for / € Z, ^-^^-!L=K-^^-^ -^ p\ as h -^ 0. (34) where p is the quasi-momentum defined by dS P = ^' (35) ax Using (34), Eqs. (23) and (24) become X = p-\- — -\--r-\-''' , (36) p p^ dp 82" — = , (Hamilton-Jacobi equation). (37) dtn dx where the Hamiltonian 2" given by lim [5„^/^] is the polynomial part of X""^V('^ + 1) in p. Thus Q^ can be expressed by oo Q- = -—[X-^^]^ = -—X-^' - T G'\ (38)
206 Y. KODAMA where the coefficient G^" can be calculated by the residue form, G^" = - res [X^ff^dX] = —!— res \x^^^—dp\ (39) k=oo k -\- 1 P=oo |_ dp ] which also shows the symmetric property G^" = G"^. Equation (36) for the spectral parameter of the Lax operator gives an algebraic constraint of a level set of the dKP hierarchy. Equation (37) can be expressed in terms of the conservation laws as the compatibility conditions, (40) - X. This implies the integrability of G"'" as expn in terms of the free energy in (32), Note in particular Q^ = p and to = x. This implies the integrability of G"'" as expressed G^^ = ——J^^ (41) dtn dtm which defines the constitutive equations G"'" as 2-point functions (Dijkgraaf and Witten [1990] and Aoyama and Kodama [1996]). In the present formulation of the dKP hierarchy, X is considered to be a constant given by the spectral parameter of L. However with Eq. (36) one can treat /? as a constant instead of X (recall that Eq. (36) gives an algebraic constraint, i.e. F(X, p, x) = 0). This formulation then leads directly to the quasi-classical limit of the Lax equation (18), ^.{2«,,^.^^_^^. (42) dtn dp dx dx dp These formulations can be put into the differential three-form, oo dQ:=J2dQ'' A dtn AdX = 0, (43) ; n=o from which we have also the general form of the Lax equation, {G^M" = {G^M'"•.= f^^-f^^. (44) dp dtm atm dp This equation (44) indicates the democracy of the coupling constants tn, the symmetry parameters, and there is no particular significance of the variable x, unlike the original KP theory. This observation also connects the relation between the continuous limit (or dispersionless) of Toda hierarchy and the dKP hierarchy (Kodama [1990] and Aoyama and Kodama [1996]). In the case of the dispersionless Toda hierarchy similar to the example discussed in the next section, we have a log-function for some of the Hamiltonian 2"; e.g. Q~^ = logip — s) then introducing a new momentum P = p — s,ihc Lax equation (44) with m = — 1 is expressed in the pair (P, r_i) as dx ~dt, ^^p(dQ;^^_dg^dX\ V dP dt-i dt-i dPj'
DISPERSIONLESS INTEGRABLE SYSTEMS 207 which shows the Poisson bracket for the dispersionless Toda hierarchy as commonly used (Takasaki and Takebe [1991]). 11.4 Example of Degenerate Frobenius Manifold Let us now consider the example of the dispersionless limit of the N- component vector NLS equation (32) (Zakharov [1980] and Gibbons [1981]). We then construct a degenerate Frobenius manifold as a singular reduction of the manifold associated with this example. We start by presenting the general structure of the dispersionless vector NLS equation. 11.4.1. The Dispersionless Vector NLS Equation The spectral parameter of the equation is given by the dispersionless limit of (21), N The generating functions 2" of the symmetry flow are then given by Krichever [1994] and Aoyama and Kodama [1996], QiO,n) _ _!_[;^n+l] for n = 0, 1, 2, • • • , n -\-1 = log(p-Uk), forA: = l,. • ,iV (47) [/x^-^]_, forA: = l," ,A/^, n = 2,3, ••• . Q{k-n) n-\ Here /Xjt are the local coordinates near p = Uk having the asymptotic forms, Pk /Xjt = + 0(1), near p = Uk, (48) p-Uk which also satisfy the global relations /xi = • • • = /x^^ = ^, and [/x^~^]_ represents the part of /x^~^ containing the negative powers of {p — Uk). The Lax equations with 2" in (47) are given by (42), i.e. ^={Q\^\ (49) ata and they are shown to be all commute (Aoyama and Kodama [1996]). In the variable V = (pi, Ml, • • • , pN, unV, the flows (49) gives an integrable system of hydrodynamic dV ^dV type, — = A — with x = ^(o,0)- The matrix A" can be expressed by dtct dx A" = -^:=cj>", (50) dp
208 Y. KODAMA with p in the rational function 0"(/?) substituted for the matrix A^^'^\ Since A" is a matrix with the size IN x IN, we have a 2A/'-dimensional basis such as B = .".. j.».-.j;^,.j.».-j;:;]. (5i) Because of (50), this basis can be also expressed by the rational functions 0" defined in (50), the primary fields, and the space spanned by the basis forms a rational ring given by A where the ideal of the ring is given by X' = |^ = l-E-^=0. (53) Equation (53) gives k=\ which is the expansion of the function 0^^'"^^ in terms of the primaries in the basis B. The rational functions 0" which are not in the basis B are called the gravitational descendants of the primary fields, generating the higher symmetry flows of the dispersionless vector NLS equation, and we denote them as ctm{(I>''), for M > 1, a € a. (55) where A is the set of primary indices. These are given by the rational functions defined in (50) with (47) except those of the primaries 0^^'"^^ which are expressed as 8 /rx^ 1 \u^ aM(0^^'-'^) := — ( I ]t7i(log^ - cm) I - ^(log/^it - cm) (56) where cm = T^ - (Eguchi and Yang [1994] and Aoyama and Kodama [1996]). Despite the appearance of the log-terms in (53), trM(0(^'~^^) is a rational function in C[/7, {p — u\)~^,' " , {p — un)~^] (see Aoyama and Kodama [1996] for the detailed calculations having the log terms). Then with the ideal (53) they can be decomposed into the primary fields with the coefficients determined completely by the constitutive equations, the 2-point functions. This is precisely the topological recursion relation on the TFT (Dijkgraaf and Witten [1990]) (see Eq. (14)), and one can state:
DISPERSIONLESS INTEGRABLE SYSTEMS 209 Lemma 11.1 Aoyama and Kodama [ 1996] For each primary 0", the topological recursion relation can be written as cTMir) = ^(trM-i(0")0^>0^ (mod X'). (57) forM>\ andaoicj)'') = 0". Note here that the 2-point function (crM-i((t>^)(t>^) can be expressed as the 3-point function, that is, using the relation J^y ^ay^^^ = ^a, (crM(r)(l>^(t>^^^^^) = (aM-i(0")0^>. (58) The 3-point functions {(p^(p^(j)'^) is then expressed by the residue formula (Dubrovin [1996] and Aoyama and Kodama [1996]), (0 0^0>^) = res [ -— dp . (59) In terms of those n-point functions, the dispersionless vector NLS equation is just the integrability relations, 7^(0^0^) = 7^(0"0n = (0"0^0^> (60) ota Ot^ which implies that the 3-point functions are indeed the derivatives of the free energy J^ i^r<t>^V) = ^~-^T. (61) oiqi^ at^ aty To find an explicit form of !F, we first note Theorem 11.1, which we can restate in terms of the 2-point functions in Lemma 11.1, Proposition 11.1 Aoyama and Kodama [1996] A solution of the system (49) can be obtained in the hodograph form, ^"=X^(^m-i(0^)0">Cm,^, (62) M>1 /36A where Cm,^ cire arbitrary constants. In particular, if we take Cm,^ = ^M,i8p,^o with an arbitrary choice of the primary index Po, the solution leads to a flat metric rj"^ = (0"0^0(^'^^) = constant, which is called the flat solution and determines a Frobenius manifold of the corresponding TFT. Note here that one can construct 2N different flat solutions in this model (Aoyama and Kodama [1996]).
210 Y. KODAMA 11.4.2 Degenerate Frobenius Manifold We now give an explicit solution of the system having a degenerate metric as a singular reduction of the dispersionless 2-component NLS equation. To find the flat solution, we first give the 2-point functions (0"0^) as the 4x4 symmetric matrix, /P1+P2 Wi Ml log Pi ((0"0^» = Pi \W2 Ml - P2 Pi Ml - Pi - P2 M2 — Ml PlP2 M2 \ log(M2 - Wl) Pi M2—Ml (W2—Mi)2 M2—Ml Pi l0g(M2 -Ml) log p2 M2 — Ml (63) / where we have ordered the indices as [(0, 0), (1, —1), (1, —2), (2, —1)]. In finding (63), we have used the residue formulae similar to (39). The dispersionless 2-component NLS equation is explicitly expressed with those 2-point functions as (60). In particular, one should note from the form log P2 that the equation has a particular solution P2 = 0 which reduces the equation into the single dispersionless NLS equation. This observation is a key to a singular reduction of the Frobenius manifold. Using Proposition 11.1, a flat solution is obtained by setting Cm,^ = 5m,i5^,(2,-i), that is, t- = (0(2'-iV">, (64) from which the corresponding (nondegenerate) flat metric is given by (n"h^ 0 1 0 1\ 10 10 0 10 0 .10 0 0/ The free energy can be computed by integrating (63), and we obtain jr ^ ,0,1,2 _ 1,1(^2)2 + 1 (,2)2 ^i„g,2 _ 3^ _ ^2^, (65) (66) where *' for j" = 0, • • • , 3 are defined by ,0 _ ,(0,0)^ ,1 _ jd.-l)^ ,2 ._ ,(1.-2)^ ,3 _ ,(2,-1) (67) Now we consider the singular limit t^ -> —oo which corresponds to the case p2 -> 0. In this limit, the ideal X' = 0 leads to a further constraint on the primaries, X' P2=0 ^ 0(0.0) _ ^(1,-2)^0 (68)
DISPERSIONLESS INTEGRABLE SYSTEMS 211 from which we have, for any primaries 0" and 0^, (0«0^V) = ± ((0^0<«-«)} - {0^0<'--2>}) = 0. (69) Then from (63) with p2 = 0, one finds a nontrivial Casimir C, = U2-\- —^— =t^ - t^. (70) U2 — U\ Note also C = X(p = m2, P2 = 0) (Strachan [1999]). Then the free energy constrained on the level surface C =constant and p2 =0 turns out to be a system with a degenerate metric for {t^.t^j^) having the form, /O 1 0\ (tJ"^) =10 1. (71) \0 1 0/ In the free energy J^ in (66) on the constraint C =constant, the r^-variable appears as a linear term except the e^ (= P2) term. Thus the terms containing t^ in ^ do not influence the theory of the reduced system. Then the corresponding free energy can be written as i(.¥(.og,^-2)-, ^= V(rV + ^(^')Mlog^' - ^ ) - ^V\ (72) which has been obtained by Strachan [1999] as an example of a degenerate Frobenius manifold. Namely his degenerate example can be considered as a singular reduction of the higher dimensional nondegenerate system. It is also interesting to note that this singular reduction can be viewed as a constraint on the zero set of r-function TdKP = exp J^ in (33). Namely the degenerate system lives on the constraint given by t^j^p (^^, • • • , ^^) = 0, and it gives a possible stratification of the Frobenius manifold. This may relate to a recent study on the Birkhoff strata of the Grassmannian based on the zeros of the KP r-function (Adler and van Moerbeke [1994]). This problem is currently under investigation. Acknowledgement I would like to thank S. Aoyama for a general discussion on the topological field theory and I. A. B. Strachan for explaining to me his example of the degenerate Frobenius manifold. My work is partially supported by NSF grant DMS9403597. References Adler, M., and van Moerbeke, P., Birkoff Strata, Backlund Transformations, and Limits of Isospectral Operators, A^v. Math. 108, 140-204 (1994).
212 Y. KODAMA Aoyama, S., and Kodama, Y., Topological Landau-Ginzburg theory with a rational potential, and the dispersionless KP hierarchy Comm. Math. Phys. 182, 185-219 (1996). Carroll, R., and Kodama, Y, Solutions of the dispersionless Hirota equations. J. Phys. A: Math. Gen. 28, 6373-6387 (1995). Date, E., Jimbo, M., Kashiwara, M., and Miwa, T., Transformation Groups for Soliton Equations in Nonlinear Integrable Systems — Classical, and Quantum Theory, Eds. Jimbo, M., and Miwa, T., Proc. RIMSSymp. (World Scientific, Singapore, 1983) 39-119. Dijkgraaf, R., and Witten, E., Mean field theory, topological field theory and multi-matrix models. Nucl. Phys. B342, 486-522 (1990). Dubrovin, B., and Novikov, S.R, Hydrodynamics of weakly deformed soliton lattices. Differential geometry, and Hamiltonian theory, Russian Math. Surveys 44(6), 35-124 (1989). Dubrovin, B., Geometry of 2D topological field theories, in "Integrable Systems, and Quantum Groups" Montecalini 1993, Eds. M. Francaviglia, and S. Greco, Springer lecture notes in mathematics 1620, 120-348 (1996). Eguchi, T., and Yang, S., The topological CP^ model, and the large A^ matrix integral. Mod. Phys. Lett. A9, 2893-2902 (1994). Gel'fand, I.M., and Dikii, L.A., Asymptotic behavior of the resolvent of the Sturm-Liouville equation, and the Algebra of the Korteweg-de Vries equation, Russ. Math. Surv. 30, 67-100 (1975). Gibbons, J., CoUisionless Boltzmann equations, and integrable moment equations, Physica D 3D, 503-513(1981). Guha, P., and Takasaki, K., Dispersionless hierarchies, Hamilton-Jacobi theory, and twistor correspondences, J. Geom. Phys. 25, 326-340 (1998). Kodama, Y, A solution method for the dispersionless KP equation. Prog. Theor Phys. Suppl. 94, 184-194(1988). Kodama, Y, and Gibbons, J., A method for solving the dispersionless KP hierarchy, and its exact solutions n. Phys. Lett. 135, 167-170 (1989). Kodama, Y, and Gibbons, J., Integrability of the dispersionless KP hierarchy, in Proc. of Workshop "Nonlinear, and Turbulent Processes in Physics" (NonUnear World) Kiev 1989, (World Scientific, 1990) 160-180. Kodama, Y, Solutions of the dispersionless Toda equation, Phys. Lett. 147A, 477-482 (1990). Kodama, Y, The Whitham equations for optical communications: mathematical theory of NRZ, Sept. 1997, SIAMJ. ofAppl Math., 59, 2162-2192 (1999). Krichever, I., The raw-function of the universal Whitham hierarchy, matrix models, and topological field theories. Comm. PureAppl. Math. 47, 437-475 (1994). Pressley, A., and Segal, G., Loop Groups (Oxford University Press, Oxford, 1986). Strachan, I.A.B., Degenerate Frobenius manifolds, and the bi-Hamiltonian structure of rational Lax equations, J. Math. Phys. 40, 5058-5079 (1999). Takasaki, K., and Takebe, T., Integrable hierarchies, and dispersionless limit. Rev. Math. Phys., 7, 743-808 (1995). Takasaki, K., and Takebe, T, SDiff(2) Toda equation — hierarchy, tau function, and symmetries, Lett. Math. Phys. 23, 205-214 (1991). Tsarev, S.R, On Poisson brackets, and one dimensional hamiltonian systems of hydrodynamic type. Sov. Math. Dokl 31, 488-491 (1985). Zakharov, V.E., Benney equation, and quasi-classical approximation in the inverse transform model, Funct. Anal. Appl 14, 15-24 (1980).
12 A/^-Component Integrable Systems and Geometric Asymptotics MARK S. ALBER Department of Mathematics, University of Notre Dame, Notre Dame, IN 46556, USA E-mail: Mark. S.Alberl@ nd. edu A method of complex geometric asymptotics for the Hamihonian systems on Riemann surfaces is reviewed. Then a connection between integrable billiards and weak piecewise solutions of nonlinear evolution equations including shallow water equation, Dym type equations and a class of N-component systems is described. 12.1 Introduction The method of geometric asymptotics was introduced to investigate semiclassical asymptotic solutions of the wave and Schrodinger equations in the presence of caustics (that is, focal surfaces of the corresponding geodesic flow). For example, this was done in Keller and Rubinow [1960] to explain the whispering gallery phenomenon of acoustics. This method was developed into one of the areas of research in geometric analysis by Leray, Hormander, Guillemin and Sternberg, Kostant, Weinstein, Arnold, Maslov, Duistermaat, Souriau and many others. (For details see Guillemin and Sternberg [1977].) Maslov also developed complex WKB method (see Maslov [1994] and Mishchenko et al. [1990]). There have been many important developments in which the methods of complex and algebraic geometry have been used to investigate the eigenfunctions of Hill's operator in the context of integrable equations. For example, Bloch eigenfunctions of Hill's operators, which are meromorphic on the associated spectral curves play an important role in the inverse scattering transform method for nonlinear soliton equations. For details see amongst others, Ablowitz and Segur [1981] and Dubrovin, Krichever and Novikov [1985]. 213
214 M.S.ALBER Algebraic-geometric methods also yield construction of action-angle variables for the Hamiltonian systems on Riemann surfaces. In particular, action-angle variables were used in Krichever and Phong [1997] to introduce the notion of the symplectic form on Jacobian fibrations in terms of the Kadomtsev-Petviashvili (KP) Lax operator and the associated KP Baker-Akhiezer wave-function and its conjugate. Action-angle variables play a central role in the techniques for establishing a link between mechanical systems and their quantum mechanical counterparts (see Arnold [1978], Veselov and Novikov [1982] and Lazutkin [1993]). One such technique uses the WKB method to analyze asymptotic limits of eigenfunctions and thereby establish the corresponding semiclassical solutions (modes). This technique can be used to establish certain semiclassical limits of the spatial flows related to integrable nonlinear evolution equations. In this way a connection is established between special classes of solutions of nonlinear equations and mechanical systems, to form a semiclassical theory for these nonlinear equations. For example the Jacobi problem of geodesies on n-dimensional quadrics and billiards in domains bounded by quadrics is related to the Dym equation and the C. Neumann problem for the motion of particles on an n-dimensional sphere in the field of a quadratic potential has the KdV flow associated with it. Recently the Camassa-Holm shallow water equation was linked to the geodesic motion in the presence of a potential on an n-dimensional ellipsoid (see Alber et al. [1999a,b]). For the C. Neumann problem, the spectral parameter appears linearly in the potential of the corresponding Schrodinger equation: V = u — X. In contrast, Antonowicz and Fordy [1987, 1988, 1989] investigated potentials with poles in the spectral parameter for what they refer to as energy dependent Schrodinger operators connected to certain systems of evolution equations. Specifically, they obtained multi-Hamiltonian structures for A/'-component integrable systems. The presence of a pole in the potential was shown in Alber et al. [1994, 1995] to be essential for the existence of weak billiard solutions of the nonlinear equations including the Dym equation and equations in its hierarchy. These weak billiard solutions can be obtained for large classes of the A/'-component systems. In this chapter we giye a review of a method of complex geometric asymptotics for the Hamiltonian systems on Riemann surfaces and illustrate it by constructing semiclassical modes for the families of geodesies in domains outside of n-dimensional quadrics in the context of problems of diffraction and for the n-dimensional complex spherical pendulum in the context of semiclassical monodromy. Then we describe a connection between integrable billiards and weak billiard solutions of nonlinear equations, including shallow water and Dym type equations. Notice that weak billiard solutions can be obtained for the whole class of N-component systems. 12.2 N-component Systems The hierarchy of an N-component systems of nonlinear equations is obtained as the compatibility condition for the eigenfunction of the linear system of equations LV^ = 0, (1) xlrt=Axlr, (2)
A/^-COMPONENT INTEGRABLE SYSTEMS AND GEOMETRIC ASYMPTOTICS 215 The time flow is produced by the Unear differential operator where B(x,t, E) is a specified rational function. The operator L is assumed to be of the energy dependent Schrodinger type, J2 L = - with the potential in the form L = —^-hV(x,t,E), (4) V(x,t,E) = ^=^^^^^^-—— (5) where rj are constants and the Vj (jc, 0 are functions of the variable jc, the parameter t and the spectral parameter E is complex. In particular, the potential is chosen as E and U(x, t) V(E) = -^+4 (7) E to obtain the Dym type (HD) equation U^^t + 2U^U^^ + UU^^^ - 2kU^ = 0 (8) and the (SW) equation, derived from the Euler equations of hydrodynamics in Camassa and Holm [1993], Ut + 3UU^ = U^^t + 2U^U^^ + UU^^^ - 2kU^ (9) respectively. Stationary flow of the system determined by (6) coincides with the geodesic flow on quadrics. Mechanical systems associated with the potential (7) were studied in Braden [1982]. One chooses V{E) = kE^ -h m(jc, t)E -h v(x, t) , (10) to recover the cKdV system ut = v' - ^uu' -\-Kiu' , (11) vt = \u''' - vu' - \uv' -\-Kiv\ (12) or V(E) = m(jc, t)-\-KE-\- -^ , (13)
216 M.S.ALBER to recover the cDym system Ut = \ujcxx - \UU^ +V^+ KlUx (14) Vt = -UjcV - \uvj, + KxVj, , (15) The compatibility condition of (l)-(2) leads to the Lax operator equation L, = [L,A]. (16) Using the definition for the Schrodinger operator and the differential operator A in (3) and (1) in this Lax equation yields dV _ \d^B dB dV dt 2 dx^ dx dx which is a generating equation for the coefficients of the differential operator A. By taking B to be the rational function, m n B(x, t,E)=J2 ^'"-^(^' ^) ^^ = ^"' n^^ - ^^(^' ^» ' (1^> k=-r k=l substituting it into the generating equation (17) and equating like powers of £", a recurrence chain of equations for the coefficients bj is obtained. Assuming that Vt = 0 in (17) and integrating gives the stationary generating equation which has the form -B^^B -h \{B^f + 2B^V = C(E) , (19) where the choice of B(E) ensures that C (E) is a rational function with constant coefficients. These coefficients are the first integrals and parameters of the N-component system of equations. The time evolution of B is obtained using a dynamic generating equation (17). At each instant, that is for each value of the parameter ^, 5 is a solution of the the stationary generating equation (19). bifferentiating (19) with respect to t and requiring consistency of Vt with (17) yields, B.^B'„B^-BiB., i.e, |(^) = ^ (|) , (20) where Bi = J2j=z0^j^^~^ i^ ^ solution of the dynamical generating equation (17). The order n is calculated using the number of roots of the spectral polynomial. These may be endpoints of gaps or isolated poles corresponding to solitons. Thus, n is the dimension of the solution space and / labels the equation in the hierarchy. To show how the generating equations yield the quasiperiodic n-phase solutions of evolution equations, one introduces the roots /xy, of B. Substituting X = /xy, y = 1,..., n one by one into (19) one obtains a system of ordinary differential equations for the spatial flow of the /x-variables: /^) = ™—^ ^J^(t^j) ' 7 = 1,...,«. (21)
A/^-COMPONENT INTEGRABLE SYSTEMS AND GEOMETRIC ASYMPTOTICS 217 The motion in t is produced by substituting X = fXj with B this time in (20) so that, Solutions of this system of equations for /Xy (jc, 0 can be related to the Vj(x, t) to obtain solutions of the original hierarchies of evolution equations generated by (17). Basic facts about these systems are therefore of great interest. The systems (21) and (22) have a Hamiltonian structure with "'g n;;.a.,-.,) ■ ' = ' "■ '^'' where D(fXj) = 1 and D(fXj) = Bi (/Xy) in the stationary and dynamical cases, respectively. We think of C^" as being the cotangent bundle of C", with configuration variables /xi,... , /x„ and with canonically conjugate momenta Pi,... , P„. The two Hamiltonians (23) on C^" both have the form ^=^^'"'>/ + V(/xi,.../x„), (24) where gJJ is a Riemannian metric on C". The two Hamiltonians are distinguished by different choices of the diagonal metric. They have the same set of first integrals, which are of the form P? = C(/x,), y = !,...,«, where C is a rational function of fXj. Thus, we get two commuting flows on the symmetric product of n copies of the Riemann surface R defined by P' = C(/x). These Riemann surfaces can be regarded as complex Lagrangian submanifolds of C^". We call this the ^i-representation of the problem. Recall that a Hamiltonian system is linearized when written in action-angle variables on the complex Jacobian. For details about integrating the /x — representation by reducing to the Jacobi inversion problem and using Riemann ^-functions in case of the KdV equation see Dubrovin [1981] and Ercolani and McKean [1990] and for details about the generating equations see J. Alber [1981] and Alber ^r a/. [1997]. 12.3 Complex Semiclassical Modes An important part of geometric asymptotics is the establishment of a link between the Schrodinger equation (using the Laplace-Beltrami operator in its kinetic part plus a potential part, in the usual way) and certain integrable nonlinear Hamiltonian systems. One does this by considering a class of solutions of the Schrodinger equation of the form U = ^Ait(/xi,... ,/x„)exp(/w;5it(/xi,... ,/x„)), (25)
218 M.S.ALBER where the /x variables evolve according to the phase flow of an associated Hamiltonian system —- = [W, HI W = (All,... , /x„, Pi,... , Pn). (26) ax Here {,} are the standard canonical Poisson brackets and H is a. Hamiltonian function of the form kinetic plus potential energy corresponding to the quantum Hamiltonian; this Hamiltonian determines a flow on the phase space that we denote by g, : M^" -^ M^\ (27) which is a 1-parameter group of diffeomorphisms of the phase space, complex 2n- dimensional manifold M^". The function Ak is the so-called amplitude, which contains all the information about caustics (that is, the set of focal or conjugate points of the extremal, or geodesic, field) The function Sk is called the phase function and one can show that it is the generating function of the Lagrangian submanifold of the phase space obtained by transporting an initial Lagrangian submanifold by the Hamiltonian flow. Here w; is a parameter, and in WKB theory, one normally takes w = 1/h where h is Planck's constant. Semiclassical solutions (modes) are constructed in the form of functions of several complex variables on the moduli of Jacobian varieties of compact multisheeted Riemann surfaces. This method enables one to use, in the neighborhood of a caustic, a circuit in the complex plane. By gluing together different pieces of the solution in this fashion, one can obtain global geometric asymptotics. Quantum conditions are defined as conditions of finiteness on the number of sheets of the Riemann surface. This procedure, together with the transport theorem for integrable problems on Riemannian manifolds, facilitates the construction of geometric asymptotics for a whole class of quasiperiodic solutions of integrable systems on hyperelliptic Jacobi varieties. For every spatial (stationary) Hamiltonian (24) there is a corresponding stationary Schrodinger equation which has the form W^WjU -h w^(E - V)U = 0. (28) Here w; is a parameter as before, and V^ and Vy are covariant and contravariant derivatives defined by the metric tensor g^^: g'' J=ly/Ul=l\8ll\^^^^ n1/=1 We consider geometric asymptotics to be solutions of equation (28) of the form (25) defined on the covering of the Jacobi variety in the phase space of the integrable problem. Substituting (25) into (28) and equating coefficients for different orders of w results in the following system V^ (AlWj Sk) = 0 (transport equation), (30) VJ'Sk'^jSk -V = -E (eikonal equation). (31)
A/^-COMPONENT INTEGRABLE SYSTEMS AND GEOMETRIC ASYMPTOTICS 219 We can interpret the eikonal equation as the Hamilton-Jacobi equation of the corresponding classical problem. Solutions can be constructed using symmetry properties of the Riemann metric, which in turn determines the quantum equation. Then modes of the form (25) are constructed which link the Schrodinger operators on Riemannian manifolds with integrable systems corresponding to the class of metrics mentioned above. 12.4 Complex Diffracted Modes In this section we describe geometric asymptotics for billiards outside of a quadric in connection with the problem of diffraction by an n-dimensional ellipsoid. The main idea of the collapsing construction can be described as follows. One first considers the geodesic flow on a quadric in M""^^. Associated with this flow is some underlying complex geometry, first integrals of the motion, and a complex Hamiltonian. We fix the value of the first integrals and let a„+i, the shortest semiaxis (in the case of an ellipsoid and the semiaxis with the smallest absolute value in the case of a hyperboloid), tend to zero. This yields corresponding first integrals and Hamiltonians for the geodesic flow in a domain in W bounded by a quadric. In the hyperbolic case, the trajectories may be regarded as complex billiards. We will use complexification of the problem to resolve the singularities at caustics and to extend the semiclassical mode into the shadow domain. (For details see Alber [1991] and Alber and Marsden [1994].) We apply the above construction to the geodesic flow on hyperboloids. The first integrals and Hamiltonian for geodesies in the domain outside the (n — 1)-dimensional ellipsoid, after applying the collapsing construction, have the form Pj = ± and n d2 T n2n 2n-l LoY[(f^j-^k), j = l,...,n (32) The quantities fXj = t^j(x) are functions of the variable x and the diagonal metric tensor has the expression g" = f^—7^ T (34) and the potential energy is given by
220 M.S. ALBER Now let (/Xy, 7 = 1,... , n — 1) vary along cycles on the Riemann surface over the cuts with end points at m^ and let /x„ vary over an infinite cut from m2n-i to —oo. We call domains on the real axis other than cuts of the Riemann surface shadow domains. For example, after collapsing the hyperboloid by means of the limiting process (a^ = W3) -> 0, and making the choice of parameters and first integrals given by m4 < m3 < 0 < m2 < mi, one obtains the interval ]m4, 0[ as one of the shadow domains. Applying the method of geometric asymptotics, as described above, one obtains a diffracted mode that has the following form: [n+l n -1-1/4 n n^^y-'^/) (36) X exp I /— — i — Do — wD\ -\- iwD2 1. (37) The mode (36) can be constructed independently in each domain. Here Do is a vector of Maslov indices, Di is given by Di = (-lY^Tn(mn,f^n)-\-knTn(mn,0); Tn(a,b) = f PndfXrt, Ja and D2 is the real part of the phase function S. We keep track of the number of reflections on the boundary of the quadric and number of tangent points with caustics by introducing indices k and Maslov indices r. 12.5 Semiclassical Monodromy Cushman and Duistermaat [1988] considered the semiclassical spherical pendulum and by studying Bohr-Sommerfeld quantum conditions numerically, detected monodromy in the semiclassical spectrum of the Schrodinger operator. They suggested that semiclassical solutions could be used for the asymptotic description of the eigenfunctions of the Schrodinger operator obtained as a result of quantization of the spherical pendulum. They also pointed out that the main difficulty is related to the construction of action-angle variables at the singular points. Guillemin and Uribe [1989] suggested that one should relate monodromy to Maslov effects. In this section complex modes are constructed for the n-dimensional spherical pendulum to demonstrate complex monodromy.
A/^-COMPONENT INTEGRABLE SYSTEMS AND GEOMETRIC ASYMPTOTICS 221 The Hamiltonian of the n-dimensional spherical pendulum in Cartesian coordinates Qj and their conjugate momenta Pj has the form:- (38) Here the acceleration due to gravity is taken to be unity. We also constrain the length of Q to be one. The same Hamiltonian in the n-dimensional spherical coordinates can be expressed in the following "nested" form 1 " / " 1 \ H= :^y^Po\ n T I -h/?COSl9„. 2^'pr H,i/^i (sin^,)V (39) The change of coordinates Zj = (cos Oj)^, Pl(l-zj)zj = Pi. j = h Zn =COS0n, Pl(l-zl) = Pl, results in the Hamiltonian n-l ,n-h (40) ^7 = 1 ^k=j-^l ^^^ ^" The nested structure of the Hamiltonian (39) shows that one has the following first integrals for the n-dimensional pendulum: r P^ ^l Pi + ^l (sin6i2)^ PL + «2 r'n- (sin0„_i)2 -Pi 2 - /q2 (42) K''^+si?) ^'""■=* 2V ^" (sin i9„ Here ^j are constants along solutions of the corresponding Hamiltonian system. Let Kj(z) = fif(l-z)-fif_,, 7=2,...,n-l, and Kn(z) = 2(p^^-z)(l-z^)-P^_,.
222 M.S. ALBER In what follows we extend our system into the complex domain by considering fij to be complex numbers and let the variables Zj be defined on the associated Riemann surfaces: Ri : Wf = Pi z\{\ -zxY ^ Z2(l-Z2)^ P . W2 Kn-l(Zn-l) (43) Rn : W^ = Zn-l(l -Zn-l)^' KniZn) (^-zl) 2\2- We call the Hamiltonian system with Hamiltonian (41) and first integrals (42) on the Riemann surfaces (43) a complex n-dimensional spherical pendulum. To make things concrete, we shall apply the general construction of geometric asymptotics to the case of the 2-dimensional spherical pendulum. In this case the action function S can be represented in terms of angle variables {a\, 0^2) as follows: S = -^ia\ - ^la2 -if ^^ zidzi ;? VMfe) (44) The last two terms correspond to the holomorphic and meromorphic parts of the action function. The holomorphic part is proportional to the angle variable of the classical problem. The amplitude A can be found after calculating D and /. We find that 1 ^ = \/gng22 = and det/-' = dUi dzj V(l -^1)^1 2)^2 "VM(z2)zi(l-zi)* This results in the following form of the function U: U= Yl AoVl^2(M(z2)r'^^cxp\iwJ2SkAzj)\ (45) where and *.(^.)=_(:/^ Zl)Z] dz\ +k\Tx Sk2(Z2) -L 1/ T, 2T2 ^^^2 + k2T2 + ^r-, (46) (47)
A^-COMPONENT INTEGRABLE SYSTEMS AND GEOMETRIC ASYMPTOTICS 223 where r2 is the Maslov index and (48) 4 VMfe) "^^ 4 (1 - dz2 Quantum conditions of Bohr-Sommerfeld-Keller type can be imposed as conditions on the number of sheets of the covering space of the corresponding Riemann surface for each coordinate Zj: Iwk\Tx =27tNu TT ^ (49) -r2 + wk2T2 = 2nN2. Here Ni, N2 are integer quantum numbers. The quantum conditions (49) include a monodromy part after transport along a closed loop in the space of parameters (fii and P2)' This semiclassical monodromy consists of a classical part as well as a contribution from complex monodromy and the Maslov phase. (For details about monodromy see Alber et al [1997] and Bates and Zou [1993] and about Maslov's phases see Littlejohn [1988].) Quantum numbers Nj are considered in the form of continuous functions of the parameters. It is justified by Berry's idea of fast switching quantum conditions passing rational winding numbers through the dense sets of ratios. (See Berry [1985] and Littlejohn [1988].) Only integer shifts of Nj are considered after transporting the system along a closed curve in the space of parameters as allowed. It means that the changing of parameters could lead to a changing of quantum state. 12.6 Quasiperiodic Solutions of (HD) and (SW) Equations In this section we describe Jacobi problems of inversion associated with the equations (8) and (9). Let us substitute (18) into (19) and (20) and set £" = /xi,..., /x„ subsequently. This results in the following two systems of equations /^f = ^— = Sign(/x/, m2/-i, m2M st)——p- (50) and f^i = -T7 = Sign(/x,,m2/-i,m2M^0—pz^r-^ 7' / = !,...,«, D = /xi H h /Xn, in X and t, where R(fM) = —Lq/x Y[r=i (/^ ~ ^r) and R(fM) = /x Yir^ (/^ ~ ^r) in the case of equations (8) and (9) respectively. By rearranging and summing equations (50) and (51) and integrating on particular branches of hyperelliptic Riemann surface F, one obtains the following nonstandard Jacobi inversion problem n k-\ . f^ A: =!,...,«-2,
224 M.S. ALBER which contains (n — I) holomorphic differentials and one meromorphic differential on F. Thus, the number of holomorphic differentials is less than genus of the Riemann surface, which implies that the corresponding inversion problem cannot be solved in terms of meromorphic functions of jc and t. Stationary case: t = to. To transform (52) to a standard inversion problem, let us introduce a new space variable 5'i defined as follows r = / /^i ^0 'fXn/Lodsi (53) Then, using the well-known Jacobi identities /^f n]M^i-f^j) 0 k = 0,,,.,n-2, 1 k = n-l, (54) D k = n, we rearrange the equations of (50), obtaining the following system k-i EfM. dfMi \ dsi k = 1, . , ly^RQlO ~ i 0 k = 2,...,n-l. (55) /=i The map can be inverted, resulting in expressions for algebraic symmetric functions of /x-variables in terms of theta-functions of n arguments depending linearly on 5-1 and to. Then, by using the trace formula, one can obtain a quasi-periodic stationary solution (profile) U(si, to) of equations (HD) or (SW). In addition, by substituting the theta-functional expression for the product fxi - - - fXn into (53), we obtain a quadrature. Taking it, one finds x as a meromorphic fiinction of s\ depending on to as on a parameter. Dynamical case: To study time-dependent solutions, we introduce a new variable ^i defined by ^o(mi H + /^«) This results in the following canonical Abel-Jacobi mapping '*-! dix m- ^, 2^/ROl) \ si-\-ti-\-(pi k= 1, (t>k k = 2,...,n-h (57) t -\-(t>n k = n. where 0i,..., 0„ are constant phases. To describe the relation between ti,t and 5-1, we take the expressions for the symmetric functions /xi, • • • , /x„, /xi H h /x„ and substitute them into (56). As a result, in contrast to the quadrature (53) relating x and 5-1, now we get a differential equation of the form dt -— = F(t,tusi), at\
A^-COMPONENT INTEGRABLE SYSTEMS AND GEOMETRIC ASYMPTOTICS 225 where F is a transcendental function of ^ ^i involving 5'i as a parameter. It can be shown that the equation has a transcendental integral. It follows that, in contrast to the s\-flow considered above, the flow generated by (51) (t-flow) gives rise to a nonlinear flow on the Jacobi variety Jac(T). From the point of view of algebraic geometry, this constitutes the main difference between, say, the KdV equation and equations (HD) or (SW). For detailed formulae in terms of ^-functions see Alber et al. [1999a,b, 2000]. 12.7 Billiard Dynamical Systems and Billiard Solutions of PDE's In this section a link is described between certain integrable billiards and weak billiard solutions of nonlinear integrable equations. (For details see Alber et al. [1999a,b, 2000].) We start by considering a family of confocal quadrics in W~^^ = (jci,..., jc„+i) Q(s) = \ —^i— -h ... -h ^"""^^ = 1 [, s eR, 0 < a„+i < ai < • • • < a„. yai-s an-\-i-s j These define elliptic coordinates /xi,..., /x„+i in IR"+^ in a standard way: ^/ = (a,-c) P/r'^"^'^,. y = l,...,« + l. (58) Uk=i,k^j(^j-^k) It is well-lmown that the problem of geodesies on the ellipsoid Q = Q(0) is completely integrable. Moreover, as shown by many authors (see e.g. Ranch-Wojciechowski [1995]), there exists an infinite hierarchy of integrable generalizations of the problem describing a motion on Q in the force field of certain polynomial potentials Vp(xi,,.,, jc„+i), /? € N. The simplest integrable potential is a Hooke potential or the potential of an elastic string joining the center of the ellipsoid Q to the point mass on it: Vi = 2^^"^^ "^ "^ -^n+i)' ^ = ^^^st. In this case, under the substitution (58) the total energy (Hamiltonian) takes the Stackel form: ''^sX:—^(i^^—(^j ^2^.^- a>(/x) = (/x - ai) • • • (/x - a„+i). Geodesic motion and motion in the field of a Hooke potential on a quadric are closely related to the quasi-periodic stationary solutions of equations (HD) and (SW), respectively. Namely, we have n n n U(si ,to) = J2 f^J = J2 ^/<*i. z) + X^ «.•. (59) j=l y=l ,=1
226 M.S. ALBER Now suppose that a„+i -> 0. In the limit, Q passes into the interior of the {N — 1) -dimensional ellipsoid Q = {x\lax -h • • '^xllan = 1} € M", M" = (jci, JC2 ..., Jc„). The motion on Q transforms to billiard motion inside the ellipsoid Q.The motion on Q under the Hooke force passes to the motion inside Q under the action of the Hooke force with the potential V = a{xl -\- • • • -\- x^)/2 and again with elastic reflections along Q, Thus, we have "a generalized ellipsoidal billiard with potential**. Under the limit a„+1 -> 0, the Jacobi inversion problem (55) can be written in the following integral form = (pi = const, I = I,... ,n — I, I 1 JUc IfM^/RQl) k=l -^^0 R(f^) = -(/^ - ai) • • (/x - a„)[(/x - ci) • • • (/x - Cn-i) - or/x"], (60) which contain n — 1 holomorphic differentials on the Riemann surface of genus g =n — 1 and one differential of the third kind having a pair of simple poles Q_, Q+ on C with /x(Q±) = 0. Billiard solutions of PDE's: In the above limit the system (52) is formally reduced to the Abel-Jacobi mapping. The profiles of billiard weak solutions of the (SW) and (HD) equations are associated with the solutions of the Jacobi inversion problem by using the trace formula at time ^ = ^o- For or = 0 we get ; U(x,to) = Y.^tMj = Y^^at -x" -^^^ , (61) while for a ^ 0 we get lu .^ ^ ^ g"^[A](zo + q/2) + e-'e[A]izo - q/2) where zo = (zio, • • • , Zgo)^ (63) is a vector of constant phases and U € C^ is a vector of ^-periods of the normalized differential of the second kind on C with a double pole at the infinite point and ^'s are the standard theta-functions. (For details see Alber et al. [1999a,b], [2000].)
iV-COMPONENT INTEGRABLE SYSTEMS AND GEOMETRIC ASYMPTOTICS 227 Solitary peakon solutions: Collapsing pairwise roots of the spectral polynomial R(E) yields solitary type billiard solutions. We demonstrate this approach by deforming spectral polynomial of the genus 2 quasi-periodic solution. This results in the following inversion problem dui , dLi2 dX h-T^ 7 + h , 7 = «2 = a2Y (64) J/xi J/X2 dX h—; -+h—; r = «i = axY (65) /Xl(/Xl - a2) /X2(/^2 - (12) /X1/X2 We consider three different cases: {l\ = 1, /2 = 1), {l\ = 1, /2 = —1) and {l\ = —1, h = — 1). In each case we integrate and invert the integrals to calculate symmetric polynomials of /x's. After substituting these expressions in the trace formula for the solution, this results in three different parts of the profile of the solution defined on different subintervals on the real line. The union of these subintervals covers the whole line. On the last step these three parts are glued together to obtain a wave profile with two peaks. The dynamics of the two peakon solutions is described by using the same approach. (For details see Alber and Miller [2000].) Acknowledgement Research partially supported by NSF grant DMS 9626672 and NATO grant CRG 950897. References Ablowitz, M.J., and Segur, H., Solitons, and the Inverse Scattering Transform, Philadelphia: SIAM (1981). Alber, M.S., Camassa, R., Fedorov, Y., Holm, D.D., and Marsden, I.E., On Billiard Solutions of Nonlinear PDE's, Phys. Lett A 264, 171-178 (1999a). Alber, M.S., Camassa, R., Fedorov, Y., Holm, D.D., and Marsden, I.E., The geometry of new classes of weak billiard solutions of nonlinear pde's (subm.) (1999b). Alber, M.S., Camassa, R., Holm, D.D., and Marsden, I.E., On Umbilic Geodesies, and Soliton Solutions of Nonlinear PDE's, Proc. Roy. Soc, London Ser A 450, 677-692 (1995). Alber, M.S., Camassa, R., Holm, D.D., and Marsden, I.E., The geometry of peaked solitons, and billiard solutions of a class of integrable pde's, Lett. Math. Phys. 32, 137-151 (1994). Alber, M.S., and Miller, C, On Peakon Solutions of the Shallow Water Equations, Appl. Math. Lett. (to appear) (2000). Alber, M.S., and Fedorov, Yu.N., Geometric Solutions for Nonlinear Evolution Equations and Flows on the NonUnear Subvarieties of Jacobians, 7. Phys. A: Math. Gen. (to appear) (2000). Alber, M.S., Hyperbolic Geometric Asympiotics, Asymptotic Anal. 5, 161-172 (1991). Alber, M.S., Luther, G.G., and Marsden, I.E., Energy Dependent Schrodinger Operators, and Complex Hamiltonian Systems on Riemann Surfaces, Nonlinearity 10, 223-242 (1997). Alber, M.S., and Marsden, I.E., Complex Geometric Asymtotics for Nonlinear Systems on Complex Varieties, TopoL Methods Nonlinear Anal. 4, 237-251 (1994). Alber, S.J., On stationary problems for equations of Korteweg-de Vries type. Comm. Pure Appl. Math. 2, 259-272 (1981).
228 M.S. ALBER Antonowicz, M., and Fordy, A.P., Coupled Harry Dym equations with multi-Hamiltonian structures, J. Phys. A 21, L269-L275 (1988). Antonowicz, M., and Fordy, A.P., Coupled KdV equations with multi-Hamiltonian structures, Physica Z) 28, 345-357 (1987). Antonowicz, M., and Fordy, A.P., Factorisation of energy dependent Schrodinger operators: Miura maps, and modified systems, Comm. Math. Phys. 124, 465-486 (1989). Arnold, V.I., Mathematical Methods of Classical Mechanics, Springer-Verlag: New York, Heidelberg, Berlin (1978). Bates, L., and Zou, M., Degeneration of Hamiltonian monodromy cycles, Nonlinearity 6, 313-335 (1993). Berry, M.V., Classical adiabatic angles, and quantal adiabatic phase, J. Phys. A: Math. Gen. 18, 15 (1985). Braden, H.W., A completely integrable mechanical system, Lett. Math. Phys. 6, 449-452 (1982). Camassa, R., and Holm, D.D., An integrable shallow water equation with peaked solitons. Phys. Rev. Lett. 71, 1661-1664 (1993). Cushman, R., and Duistermaat, J.J., The Quantum Spherical Pendulum, Bull. Amer. Math. Soc. (new series) 19, 475-479 (1988). Dubrovin, B.A., Krichever, I.M., andNovikov, S.P., Integrable systems. I. (Russian) Current problems in mathematics. Fundamental directions, 4, 179-284, 291 (1985); Itogi Nauki i Tekhniki, Akad. Nauk SSSR, Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow (1985). Dubrovin, B.A., Theta-functions, and nonlinear equations. With an appendix by I.M. Krichever. Russ. Math. Surveys 36, n{\9U). Ercolani, N., and McKean, H., Geometry of KdV(4): Abel sums, Jacobi variety, and theta function in the scattering case. Invent. Math 99, 483-544 (1990). Guillemin, V., and Sternberg, S., Geometric Asymptotics, Math. Surveys 14, Amer. Math. Soc, Providence, Rhode Island (1977). Guillemin, V., and Uribe, A., Monodromy in the Quantum Spherical Pendulum, Commun. Math. Phys. 122,563-574(1989). Keller, J.B., and Rubinow, S.I., Asymptotic solution of eigenvalue problems, Ann. Phys. 9, 24 (1960). Krichever, I.M., and Phong, D.H., On the integrable geometry of soliton equations and N = 2 supersymmetric gauge theories, J. Differential Geom. 45(2), 349-389 (1997). Lazutkin, V.F., KAM theory, and semiclassical approximations to eigenfunctions. With an addendum by A.I. Shnirel'man. Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics, and Related Areas (3), 24. Springer-Verlag, Berlin (1993). Littlejohn, R.G., Cyclic:Bvolution in Quantum Mechanics, and the Phases of Bohr-Sommerfeld, and Maslov, Phys. Rev Lett. 61, 2159 (1988). Maslov, V.P., The complex WKB method for nonlinear equations. I. Linear theory. Progress in Physics, 16. Birkhuser Verlag, Basel (1994). Mishchenko, A.S., Shatalov, V.E., and Stemin, B.Yu., Lagrangian manifolds, and the Maslov operator. Springer Series in Soviet Mathematics. Springer-Verlag, Berlin (1994). Rauch-Wojciechowski, S., Mechanical systems related to the Schrodinger spectral problem. Chaos, Solitons, and Fractals 5, 12, 2235-2259 (1995). Veselov, A.P, and Novikov, S.P, Poisson brackets that are compatible with the algebraic geometry, and the dynamics of the Korteweg-de Vries equation on the set of finite-gap potentials. Sov Math. Doklady 26, 357 (1982).
13 Systems of Hydrodynamic Type from Poisson Commuting Hamiltonians A.R FORDY Department of Applied Mathematics and Centre for Nonlinear Studies^ University of Leeds, Leeds LS2 9JT, UK E-mail: allan @ amsta. leeds. ac. uk In this chapter we discuss systems of hydrodynamic type (PDEs) which arise through the commutativity of two functions on a finite dimensional phase space (each of which is associated with a system of ODEs (the corresponding Hamiltonian flow)). These functions should be quadratic in the momenta, but may contain first degree terms, in which case the resulting hydrodynamic system is of "non-homogeneous" type. Not all systems of hydrodynamic type can be obtained in this way, so one of our tasks is to identify and (if possible) classify those that can. Another question concerns the integrability of the resulting hydrodynamic type systems. The answer is simple only for the 'homogeneous' systems. A number of interesting open questions remain for the non-homogeneous case. We give an alternative derivation of the non-homogeneous system, which may help to answer some of these questions in the future. This paper essentially constitutes a review of the papers (Ferapontov et al [1997a,b, 1999]). 13.1 Introduction Many wave phenomena in gas dynamics and hydrodynamics (not to mention traffic flow, chromatography and many other applications) are modelled by first order, quasi-linear hyperbolic systems. There is a rich body of knowledge on how to handle such equations, masterfully summarized by Whitham [1974], in which there is much emphasis on the close connection between dispersive and hyperbolic wave equations, which is the origin of the "Whitham Equations" of the title of this meeting. This connection was exploited in the context of equations of Korteweg-de Vries type in Flaschka et al [1980] and Dubrovin et al [1983]. In Dubrovin et al [1984], Dubrovin and Novikov developed the theory of Poisson brackets of hydrodynamic type, the coefficients 229
230 A.R FORDY of which take particular geometric significance as a result of the Jacobi identities. For both these aspects see the review (Dubrovin et al. [1989]). This theory was further developed in Tsarev [1985], in which it is shown that a "semi-Hamiltonian", diagonal system (see below) possesses an infinite number of first integrals of "hydrodynamic type", which are in involution with respect to the hydrodynamic Poisson bracket. It was further shown that these generate an infinite number of commuting flows of hydrodynamic type and that these can be solved by the generalized hodograph method (see the review Tsarev [1991] and also in this volume (Tsarev [1999])). There is a large literature on such systems, more recently reviewed in Ferapontov [1994]. In the present chapter we do not use this Hamiltonian property of hydrodynamic type systems. The Poisson bracket mentioned in the title is the canonical one acting on a finite dimensional phase space, with coordinates q\ pt. Furthermore, the inverse metric coefficients ^" which occur in this chapter are not those of the Dubrovin-Novikov bracket, but those which describe the kinetic energy of our finite dimensional Hamiltonians. Any two Poisson commuting functions H, F on this phase space generate commuting Hamiltonian flows, which we suppose to be respectively parameterised by x and t. The commutativity means that we may consider a 2—dimensional surface in phase space, parameterised by x and t. On this surface we have: i ^H , dF opi dpi from which (in principle) we can eliminate pt to obtain a system of first order PDEs for q'{x,t): qi = K'[q\'",q\ql"',q:i (1) This system is not usually quasi-linear, but is when the functions H and F are quadratic in momenta, which is the case considered in this chapter. There is no guarantee that this system of PDEs is in any sense integrable, but any functions q^(x,t), obtained by solving the two systems of ODEs, will giVe a solution of (1). When H generates a completely integrable Hamiltonian system (always the case in 2 degrees of freedom), then integrating this finite dimensional system gives a 2n—parameter family of solutions of our PDEs. Remarkably, there is a class of such PDEs whose general solution can be obtained by these "finite dimensional" methods (see section 13.3). By eliminating some of the variables in the system (1) it is possible to generate higher order (in jc—derivatives) PDEs which are sometimes integrable (but sometimes not!). For instance, choosing the pair of functions: H=^(pi-^pl)-^h(q\q\ F = q^pxP2-q^pl + f{q\q^), then the system (1) is of hydrodynamic type: 1 2 2 2 2 1 o 1 2 (It =^ ^x^ ^t =^ ^x-^^ ^x^
POISSON COMMUTING HAMILTONIANS 231 whose diagonal form is the linearly degenerate system (16). In this calculation, h and / play no role (but are respectively conserved density and flux for the hydrodynamic system). To eliminate one of the q^ we need the particular form of h. The functions h for which H and F commute belong to the Stackel class for parabolic coordinates. The Henon-Heiles potential h = {q^)^ -\-\q^ {q^)^ leads to the KdV equation for ^ ^ whilst the quartic potential h = 16(^1)4 -h 12(^1)2(^2)2 _^ (^2)4 generates: ^' 48 dx iA(^,«(,v), which is not integrable (a//the integrable cases of such equations are listed in canonical form in Mikhailov et al. [1991]). We currently have no way of knowing in advance whether or not the resulting system will be integrable, although some progress has been made recently (Blaszak etal. [1999]). The rest of this chapter is concerned with systems of hydrodynamic type which arise in this way. In section 13.2 we derive the general equations which result from commuting two quadratic (diagonal) Hamiltonians. In section 13.3 we consider the homogeneous case and show how the separation of variables for ihQ finite dimensional Hamiltonian system leads to the general solution for the corresponding homogeneous system of hydrodynamic type. The simplest non-homogeneous examples, associated with Killing vectors of the given metric, are presented in section 13.4. The general case (in 2 dimensions) is considered in section 13.5. When the homogeneous version is just linear, the corresponding non-homogeneous terms are derivatives of Weierstrass elliptic functions. Otherwise, we obtain a finite sequence of specific functions (see Figure 13.1), but have no proof that these are the only solutions. In section 13.6 we present a duality between pairs of our metrics, enabling new solutions to be constructed from old. In section 13.7 we present an alternative way of constructing (some of) our solutions, involving the extension of Noether constants to non-homogeneous form. Some open problems are identified in the conclusions (section 13.8). 13.2 Commuting Quadratic Hamiltonians In Ferapontov et al. [1997b] we considered Poisson commuting, quadratic Hamiltonians with electromagnetic term. We mainly considered diagonalised Hamiltonians H and F of the form: 1 " H=-Y,g'\pi-Aif-hh, (2) ^ n n F = :^J2s"^'(Pi - ^')^ + J2'^'iPi - Ai) + f. (3) ^ ,=1 ,=1 Such diagonalization is always possible in two degrees of freedom, provided the metric gtj has Euclidean signature, but generally is a restriction. In this diagonal case the commutativity
232 A.P. FORDY conditions can be written in the form: diV^ = 0 for any i = 1,- ,n, (4) djln(g'') =-^^ for any / # y, (5) 11/ — lit yJ — v^ n (6) 1 9<<o' - -gii Y^ (p'^dkg" for any i = 1, ■ • ■ , n, ^ k=i (v''-v')(diAk-dkAi)=gudk(p'+gkkdi(p'' for any i jl: k, (7) n dif-v^dih-\-J2^^(^kAi-diAk)=0 forany / = !,...,«, (8) it=i ^/8it/i = 0. (9) k=i In this case the system (1) takes the guise of a non-homogeneous system of hydrodynamic type in Riemann invariant form: ql = v'qi-\-(p', i = l,",n. (10) Condition (4) means that our system (10) is linearly degenerate. Cross differentiation of (5) gives: 9*(^) = 'J (^) '^'^y '• ^> ^M'•• (11) This means, that the "homogeneous" part ql = v^q^^ of system (10) is semi-Hamiltonian, thus possessing an infinite number of first integrals of hydrodynamic type, together with the corresponding commuting flows (Tsarev [1985, 1991]). In fact, the homogeneous equation can be solved exactly either by a reciprocal transformation (Ferapontov [1991]) or by separating the variables of the associated finite dimensional Hamiltonian system (Ferapontov et al. [1997a]). The latter approach is briefly described in section 13.3. With nontrivial At and (p^, we provide a natural way to add nontrivial right hand sides to homogeneous, linearly degenerate systems, but generally destroying integrability. For instance, the semi-Hamiltonian property no longer guarantees the existence of an infinite number of first integrals. In Ferapontov et al. [1997b] we presented the simplest solutions of equations (4)-(9), for which Af = 0. In this case (p^ are the components of a Killing vector of the diagonal metric ga. This includes some interesting examples, such as the Gibbons-Tsarev equation (Gibbons et al. [1996]). These results were extended in Ferapontov et al. [1999], where we presented examples for which both At and (p^ are nontrivial. For the case of 2 degrees of freedom it is possible to completely solve equations (4)-(8), subject only to the constraint (9). For higher degrees of freedom the situation is more complicated, but some progress can be made (Aujla [1999] and Aujla et al. [1999]).
POISSON COMMUTING HAMILTONIANS 233 Thequestion of the integrability of our non-homogeneous systems is an open problem. In Ferapontov et al. [ 1997b] we present both integrable and non-mtegrable examples. However, the non-homogeneous hydrodynamic systems obtained by our construction are very special, possessing a higher order conservation law dtC = dxJ^, where: ^ n n with the density C a quadratic expression in the first derivatives q]^. However, they seem to possess only finitely many symmetries in general. 13.3 The Homogeneous Case We briefly describe homogeneous, linearly degenerate, semi-Hamiltonian systems: ql=v'{q)qi,, / = l,...,n, (13) with the eigenvalues v^ satisfying the equations (4) and (11). Ferapontov [1991] has given the general solution of this system. For any particular solution we can solve (5) for ^", the compatibility of which is guaranteed by (11). The resulting functions v^ and ^" are, in fact, solutions regardless of whether or not we are in the homogeneous or non-homogeneous case, so can be used later. Since the q^ derivative of ^" is not given, this metric coefficient is only determined up to an arbitrary function /^(^^) of a single variable. Therefore, the general solution of (5) (for given v^) contains n arbitrary functions, each of one variable. In the case (p^ = 0, equations (7) imply diAk — dkAt = 0, so that Ak is a gradient (Ak = dkA) and we can set Ak = 0 by performing the canonical transformation Pk 1-^ pk -\- dkA. Thus in the homogeneous case, Hamiltonians H and F contain no first degree (in pt) terms. The remaining equation (8) reduces to: dif = v'dih for any / = 1, ••-,«, (14) which means that h(q) is a conserved density of the system (13) with the corresponding flux/: dt dx The integrability conditions dt dj f = 9/ di f give a system of^n(n — l) second order, linear equations for h, which take the form: didj((v' - vJ)h) = 0, for / # y, (15)
234 A.P. FORDY after using the linear degeneracy condition (4). Taken with the semi-Hamiltonian property, this leads to h being parameterised by n arbitrary functions of 1 variable (Tsarev [1991]). Therefore, any semi-Hamiltonian system in Riemann invariants possesses infinitely many integrals of hydrodynamic type. Thus, in the homogeneous case, our Hamiltonians H and F depend on 2n arbitrary functions of 1 variable. This gives the possibility of integrating all linearly degenerate semi- Hamiltonian systems through the Hamilton-Jacobi approach, applied to the corresponding Hamiltonian flows for H and F (Ferapontov et al. [1997a]). In fact, we have too many arbitrary functions, so can set n of them to zero and still obtain the general solution of the hydrodynamic system. The resulting formulae coincide with those obtained previously in Ferapontov [1991] within a different approach. Example 13.3.1 (2-Component System) There is essentially only one possibility for the 2 component linearly degenerate system for which neither of v^ is constant: Then 1 2 1 2 12 q^ — q"^ q"^ — q^ t\q')-f^{q^) . f\qW- h = 7 9 , / = r- (16) (17) where /', \{r' are arbitrary functions. The corresponding Hamiltonians H and F assume the form: ^ {f\q')pi - f\q^)P2 + ^\q') - ir\q^)) , {f\qWp\ - fHqWpl + f\qW - i^HqW) ■ 11 F = q' q' -q^ 1 -q^ This is easily extended to n-components. Example 13.3.2 (n-Component System). This extension takes the form: qi=v\q)qi, v'=J2^'-q\ (18) with corresponding Hamiltonians H and F: „ ^^f(q')pf + r(q') , ^ A if'(q')pf + r(q') H=y — ; , and F= > v — ; , ^ Uk^iiq^ - q^) ^ ^k^M' - q^) which is of Stackel type.
POISSON COMMUTING HAMILTONIANS 235 To obtain the general solution of (18) we can set \/r\q^) = 0. Since the metric is of Stackel type, the Hamilton-Jacobi equation for the generating function S(q,a): A/'(^')(^)' 1 V^ X-. , - > — r—— = const. (19) is separable, meaning that S(q, a) = Si(q^, a) -\- • - • -\- Sn(q^, a). It is an easy matter to show that the numerators in (19) must be the same function of their respective variables and that this must be polynomial: where we introduce the notation r(^) = ai -\- ^2? H h «n?"~^- Hence In the new canonical variables at, b^ = dS/dat the Hamiltonians H and F become an an-i which generate trivial linear (in jc and f) flows. With fo" = |+const., b"~^ — —j+const. and b' = const, otherwise, S(q, a) generates the canonical transformation: which constitutes an implicit solution for (18), containing n arbitrary functions of a single variable. When n = 2 we recover the solution of (16). By selecting some particular form for P{q^) and allowing the functions V^^{q^) to be arbitrary, we obtain a different, but equivalent, solution of the hydrodynamic equation (Ferapontov et al. [1997a]). These solutions were obtained in a different way (using reciprocal transformations) in Ferapontov [1991]. 13.4 The Killing Vector Case After the homogeneous case the next simplest is that for which dtAk — dkAt = 0, but (p^ ^ 0. In this case (6) and (7) are respectively the diagonal and off-diagonal components of Killing's equations for the metric gu (best seen in the general non-diagonal case, written in terms of the covariant derivative (Ferapontov et al. [1997b])), so that (p^ are just the components of a Killing vector (infinitesimal symmetry) of this metric. Equation (8) again corresponds to a local conservation law and the integrability conditions again lead to an overdetermined system of linear wave equations for the function h. However, in this case, h must satisfy the constraint (9), so only very special potential energy functions (conserved densities for our non-homogeneous hydrodynamic systems) are allowed.
236 A.P. FORDY We have no classification of those of our metrics which possess Killing vectors, but can write down several classes of flat and constant curvature metrics, each of which therefore possesses the maximal number of symmetries for the corresponding dimension. In 2 dimensions (with v^ as in (16)) the metric is given by (17), and this is of constant curvature R = —1«3 when the functions take the same cubic form: f{q') = ao-^aiq'-\-a2(q'f-\-a3(q'f. This includes the flat case when a^ = 0. Here we just give some examples. Example 13.4.1 [f = 1] The flat metric M 1 ^22 1 S 1 9 » 6 9 1 has 3 Killing vectors, with: (cp\(p^) = (-^^,^^), ((p\(p^) = (q^-^3q\3q^-hq^), and ((p\(p^) = (1, 1) q^ -q^ q^-q^ This gives rise to the 3—parameter family of non-homogeneous systems: ql = qVx + -T^ + ^(^^ + 3^^) + ^' q^ -q^ qf = q^ql + ^^ + ^(3^^ + q^) + c, H H where a, b and c are constants. This includes the Gibbons-Tsarev system, corresponding to ^ = c = 0. Example 13.4.2 {f = {q'f) The metric: 11 iq'? 22 (^')' ^ ~ -1 _9 ' ^ ~ 1 9 » 6 9 1 has constant curvature (= — i). The 3 Killing vectors have the form: , 1 2, , 1 2, , 1 2, (iq'rr -q\q'Y\ , 1 2, /v±i! ii+v\ giving rise to the 3—parameter family of systems: ql = q^ql + ««' + ^'"1 ' ''2 + ^ (q^)^q^ , 3q^+q^ q^-q^ ' 3^2 ' qf^qq,+aq +b-^^—^+c ^^, , where a, ^ and c are constants.
POISSON COMMUTING HAMILTONIANS 237 These examples can be extended to A/'—components. For instance, the Gibbons-Tsarev equation is extended to: Here H and F are of the form N 2 N i 2 , f. _ V Pf p_x- v'pf + pi trn*#,(9'-?')' ttWk^M' -<!")' 13.5 Systems with Nontrivial At in 2 Degrees of Freedom In 2 degrees of freedom v^ is either constant or an arbitrary (non-constant) function of ^^~^. In Ferapontov et al. [1997b] we gave the general solution of (4)-(9) in the case when v^ = d were constants. In this case: with q} = c'ql + d2rlr\q\ qf = c^ql + 91^^^^), 8iVl = a{f^f + M^ + Yx, dlf^ = -ccif^f + M^ + Yi. where a, p^i and yi are constants. Thus V^^ are Weierstrass elliptic functions. All the other quantities can be written in terms of V^^. For instance: Al = n 5-' ^2 = ^5 r. The details can be found in Ferapontov et al, [1997b]. We now consider the system of equations (4)-(9) for the case when neither of v^ is constant. In this case we can re-define q^ so that: ,1_.2 2_ 1 11 _ /H^^) 22 _ /'(^') for arbitrary functions /^ (^^). Equation (6) for / = 1 takes the form: 2(q' - q^)di<p' = Uq' - q^)^ - l) cp'-^ <p\ with the i = 2 equation obtained by the interchange 1 <^ 2. In Ferapontov et al. [1999] we derived the formulae: 9'=y[F7^^2^. H>^ = {P~P^\^. (22)
238 A.P. FORDY where the function \/r(q^, q^) satisfies: -2(q^ - q^)did2if -\- diif - d2if = 0, (23) for all functions /^ /^. I repeat, for emphasis, that this equation is independent of the particular metric. This equation is discussed in the appendix, where a number of useful solutions are listed. With these forms for (p^, equation (7) for vector potential Ai takes the form: d2Ai - diA2 = di I Jj^dix/r I - 82 JjYd2^ I • (24) If the right hand side of this is zero, then we may choose a gauge (performing a canonical transformation) in which At = 0, returning us to the Killing vector case. Otherwise, up to the usual gauge freedom, (7) has solution: IP P JI^2ir, A2 = -Jj, ^i = -i/Tr^2V^, A2 = - j^dixlr. (25) We are left with equation (8) for di f: dif = q^dih - dix/r (Jpdl (/pdlA " /ph LfpdlA) , (26) and similarly for 82/ by the interchange 1 <^ 2. The integrability conditions give an equation for/i: did2((q'-q^)h) = ^182 {f{d2ff - f\diff) + ^82 {dixlf {2 f dlxjf + d2fd2ir)) - ^di {d2f(2pdlf + dipdiif)) , (27) which is a nonhomogeneous form of the wave equation (36) (see the appendix), with a = —I. The wave operator on the left of (27) will be called L_i in what follows. We can explicitly integrate this equation to obtain the general solution: ^ = 2( l^_ 2) (^(^') + ^(^') + / (2/'92V + hfhir) dix/rdq' - j (2/l8iV + ai/^aiV^) ^lirdq^ + fOl^rf - fHdix/rA , (28) where a (^ ^) and ^(^^) are arbitrary functions, representing the kernel of the wave operator. This solution must satisfy the constraint (9), now taking the form: d2\lrdih-\-dix/rd2h=0. (29)
POISSON COMMUTING HAMILTONIANS 239 Handling this constraint in general seems to be difficult. For a given metric, what are the consistent choices of solution \/r of (23)? Alternatively, given the function \/r, what are the consistent choices of functions /^ ? These are just two different formulations of the same problem, but the second has the advantage that the unknown functions are each of just one variable. For given ^/z the integrals / 91 V^9i V^J^^ etc., can be explicitly evaluated, since the coefficients /^ etc., are independent of the integration variable. In this situation it is possible to separate variables and fully determine /^. Once we have determined h, xjr and /^, it is straightforward to integrate equations (26) to find /. Particular choices of ^ give, for (^^,^^), Killing vectors for the flat and constant curvature metrics. In the following table (Figure 13.1) these are denoted by a •. Consistent (but "non-Killing") solutions are denoted by o. A x denotes a verified "no solution" case. The functions Pi, etc., are defined in the appendix. Qi Q\ Go tnh Ig Pi P2 P^ Pa f(q) X X • • o o • • • o • • • o o • • o o X • o o X o o X X o I q q' q' q' FIGURE 13.1. The pattern of solutions 13.5.1 Examples We now briefly illustrate the above procedure with a few examples. We only list those for which curl A 7^ 0. Examples corresponding to Killing vectors can be found in Ferapontov et al [1997b]. We only need give the single component (p^, since all our solutions have a symmetry 1 <^ 2. The first examples are listed against specific metrics as in Figure 13.1, ordered from left to right. Then we give an alternative form of the calculation, in which we specify the function V^ and determine the viable metrics. In the next section we show that there is a "duality relation" between certain metrics, which enables us to map corresponding solutions onto each other. 13.5.1.1 Metric with f^q) = 1 With this metric we have the following solutions for which A is not a gradient field:
240 A.P. FORDY 1. V^ = tnh-^ y' = . ^/, ■ , Ai=-<p\ The function h satisfies the usual homogeneous equation L-\h = 0, giving (after the constraint): , C2 . -1 C2{q^+q^) ^ = ^1 TT' f = TT^ n • q^q"^ Sq^q"^ q^q^^ 2. V^ = P3 -^ ^l=3((^l)2 +2^1^2^5(^2)2)^ Ai=-(p\ The function h satisfies the usual homogeneous equation L_i/i = 0, so can choose /i = 0, leading to: / = -6(15(^1)4 - m.q'fq' - 6(q'f(q^f - I2q\q^f + 15(^2)^). 3. xlf = P4 =^ (p^=4(5(q^)^-\-9(q^f(q^)-\-l5q\q^f-\-35(q^)^). At =-(p\ The function h satisfies L-ih = 12288(^1 - q^)P3, giving: h = 1024(^1 - ^2)2(^1 + q^XKq^f + 2q^q^ + Kq^f), f = -256((q' - q'fOS^q')^ + 60(^1)^^2 ^ 66(^1^2)2 ^ 60^1(^^)3 ^ 35(^2)4)^ 13.5.1.2 Metric with f{q) = q^ With this metric we have the following solutions for which A is not a gradient field: 1. V^ = Pl ^ ^1=^2 ^^1^2^ Al=-f^. ^2 = -fi, and h just satisfies (36) with a = — 1. The constraint forces /i to be a constant, so we take: /, = 0 and / =-1(^1-^2)2 and h satisfies L-\h = 96 (^i — q^) Pi. We take /i to be the particular integral: h = \6{q'+q'){q'-q^f, f = -18 {(q'f + (q^) + 8^1^2 ((^1)2 ^ (^2)2>| ^ 20(^1^^)2 13.5.1.3 The Solution when \/r is Specified Previously we calculated the solutions x// which are consistent with a given metric function /^(q^). In our diagram, this corresponds to moving in a horizontal line. We now consider which metric functions /^ (q^) are consistent with a given function \/r. In our diagram, this corresponds to moving in a vertical line.
POISSON COMMimNG HAMILTONIANS 241 We find L_i/i = ^OiV^ - a|/^), giving: The constraint (29) leads to: q^d^f' - q'dlf = 2(dia + 82^) + q'd^.f' - q'dlf. giving: f 1 /2 = a(^ V + )S2(^¥ + m^ + 52, a=ao--()Sl+)S2)(^')'-«(^')^ ^ = -«o + ^(Pi-hP2KqY-^a(q^)\ 1 . 1 . t .. 1 1 I ^2n , ^_/^l ^2x2 /^ = ^(n + y2) + -(A +)S2)(^^ +r) + ^ctiq-q'T' 13.6 A Duality Between Metrics The notation used in the definition of H and F suggests that the Hamiltonian vector field of H describes the motion of a particle on a manifold with metric gtj, electromagnetic potential A and scalar potential h. However, since F is also quadratic in momenta, we can, instead, regard the quadratic part of F as defining a metric. A simple change of coordinates will transform F into "Stackel form" and H into a form analogous to the current F. We define the canonical coordinates: G^=i, Pi = -(q'fpi, / = 1,2, q' and define dual functions H^ and F^ by: 1 1 ^ H"^ = F(-, -Q^P) = - J2s''(Pi - Aif + h 1 1 2 2 F"" = H(-, -Q^P) = - Y,rv\Pi - Aif ^Y.^\Pi - Ai) + /, where f)^ = 2^ ^nd: (GMVUtf) ^^^ = - )i -n2
242 A.P. FORDY 22' 22^ f^-h with similar formulae after making the change 1 <^ 2. It can be seen that this transformation gives a duality between metric, with /^ = {q^Y and /^ = (2^)^"- For instance, there is such a duality between a flat {n = 2) and constant curvature {n = 3) metric. The functions \lr and ^, corresponding to cp^ and ^^ are: x/r Qo tnh /^ Pi P2 ^ Pi /^ tnh Qo Qx Thus, for the metric with f^(q) = q^we can immediately construct five solutions, two of which have curl A 7^ 0: 1. ^ = lg =^ <P^^-^i^, M--j^, with: 2. ir = Pi ^ <p'=<p2 = (q'q^)y\ Ai = - (|^)^^^ , Notice that the transformation does not respect the Killing property. Indeed, the duality between n = 0 and n = 5 cannot respect the Killing property, since when n = 5 there are no Killing vectors. Remark 13.6.1 The formula for 0^ is easily constructed by using (10) to calculate expressions for Q]:
POISSON COMMUTING HAMILTONIANS 243 leading to: which corresponds to reversing the roles of x and t. The nonhomogeneous term is just (p^, given above. Example 13.6.2 (The duality between n = 6 and n = —1) Here we just give one example. For the metric with f^(q^) = (q^)^ we have: q^-q'' (q')Hq'-q^)' and h = ci-^C2(q'-hq^)-h\(q'-hq^)\ f = C2q'q'-\{q'+q'f{{q'f-q'q" ^{q'f). Under the above transformation, we reach the solution with yjr = — 2tnh for the metric with/(^^) = (^^)-i: q^(q^-q^) (l-\-2c2q^q^) ^ C2(q^-\-q^) q^-\-q^ h = ^.,1 9x9 > f = -c\- 2(^1^2)2 ' ^ qlql (^1^2)2' 13.7 An Alternative Way to Isolate our Solutions We have solved (4-9) in terms of a function \lr{q^,q^), which satisfies (23). We were left with the constraint (9), which contains the metric functions f^(q^). This is complicated constraint on \/r, p (q^) and upon the two functions a(q^), b{q^). However, it turns out that (for some metrics) these particular choices of xjr can be isolated by postulating the existence of a first degree (in pi) constant of the motion. We have already seen that since the flat and constant curvature metrics have an algebra of Killing vectors, the second degree (in pi) part of H has a corresponding algebra of Noether constants of the form ki = ^l p\ -\- ^fpi- Our non-homogeneous solutions for H sometimes possess a non-homogeneous form of these Noether constants (with ^/ and ^f unchanged): ki=^lpX+Hfp2+Hl Whilst ^? can be gauged to zero, it is fixed by our choice of gauge in the definition (25). This accounts for our rows of 5 solutions (both • and o), but not for the 'odd' case '^ = tnh for f^(q^) = 1. This case corresponds to a conformal Killing vector, satisfying: [k, H] = aH, a a constant. (30)
244 A.P. FORDY Our choices of At (equivalently \/r) correspond to the absence of pt dependence in [k, H] (or in [k, H] — aH). However, with \/r fixed by this simple calculation, the remaining equations in [F, H] = 0 arc straightforward to solve (when compatible), after which we can check the precise form of [k, H}. This can vanish, but in some cases k satisfies: [k, [k, [k, H}}} = 0. Not all of these calculations have yet been carried out. I present below only the case of the simplest metric with f^(q^) = 1. 13.7.1 The Metric with f(q') = l In example 13.4.1 we listed the 3 Killing vectors. Consider the 3 non-homogeneous extensions of the corresponding Noether constants: ki = ^\^-^^f, k2 = px+P2+Hh h = {q'+^q')pi+0q'+q^)P2+Hl We consider H with (as yet) undetermined At and h. Since the "homogeneous" part of kt is Noether for the "kinetic" part of H, the quadratic parts of {k, H] vanish. Requiring the degree 1 parts of [k, H] to vanish gives us a formula for the partial derivatives dj^f, whose integrability conditions lead to a restriction on the gauge invariant quantity Bn = diA2 -92^1. For ki, the explicit formulae are: ai^f = (q^ - q^KdiAi - diAi) -\-Ai - A2, 92?? = (q^ - q^)(dlA2 - d2A2) + A2 - Ai, whose integrability condition can be written as: M^'5i2 = r^^Oi - 92) - , 1 ^ 2m] ^12 = 0' (31) Iq^-q^ (^1-^2)2 J which leads to B12 = (q^ - q^)y\q^ + ^^), Y an arbitrary function. (32) Introducing the derivative y^ rather than y is convenient at the next step. The operator M^^ has a simple formula in terms of the function \/r^^ = Log (q^ — q^) , used to build the homogeneous part of A:i (?/ = 82^^^^, ^l = 9iV^^0- M^' = d2xlr^'di-\-dixlr^'d2-\-2did2xlr^'. (33) For each k we have formula (33) and equation (31) in terms of the corresponding V^^. Having fixed the form of Bn (up to a function of 1 variable), (24) gives us an equation for \/r, which, in the present case, is: Ol - d^Oi^ = Bn, (34)
POISSON COMMUTING HAMILTONIANS 245 whose general solution contains arbitrary functions ofq^-\-q^ and q^ —q^. These are fixed by requiring V^ to be a solution of (23). When B\2 has the form (32), equation (34) can be solved explicitly: f = -\iq'-q'fY+ci{q' + q')+ b{q'- q"). (35) Equation (23) then gives: ' - - ■■ ^' ^a" + -y. 8 q q where the primes denote derivatives with respect to appropriate arguments. The only non-separated part of this equation is that containing y^\ giving us that y'' = 2c2, a constant. The remaining equation separates, so can be integrated: Y=cx{q^+q^)+C2{q^+q^f, a = a,{q'+q^)+a2{q'+q^f - |(^^ +^V - |(^' +^V, b = b, logiq' - q') + |(^1 - q'f - '^(q' - q')\ The parts of \/r corresponding to ai, a2 and bi give rise to the 3 Killing vector cases (denoted by • in Figure 13.1). The coefficients of ci and C2 respectively give P3 and P4. The o corresponding to "tnh" is not obtained in this way. For each \/r arising in this way, the remaining part of the calculation of [F, H] = 0 proceeds as before. We give the two cases Ci ^ 0. The case ci 7^ 0: V^ = P3 Here we have: o h = 0, f = -^(qi - q2f(5(q'f + eq'q"" + 5(^^)2) With the compatibility conditions satisfied, we can integrate the equations for ^f to obtain ^f = —\c\{q^ -h q^). Indeed, since this Bn satisfies M^'Bu = 0 for / = 1, 2, 3, we can extend all 3 of the Noether constants: The resulting kt satisfy: [k\,k2} = -2c\, {kx^k^} = 4ki, {fe, k3} = -4k2, with Casimir function: 1 ci H = -k\k2 + —k^.
246 A.P. FORDY In terms of ki,F = ^ikiks — k^). The Hamiltonian H with the 2 first integrals ki, k2 (for instance) is a super-integrable system, which can therefore be solved (non-parametrically) by fixing the values of these constants. The case C2 ^Oixjr = P4 In this case: Ai = -cp' = ^(^(^'f + 9(^')V + l5q\qY + 35(^^)3)^ Bn = C2((q'f-(qY), h = j^(q' - q^fCKq'f + 9(q'fq'' + 9q\qY + l(q^f), j/o Only the compatibility conditions for ki are satisfied, and ^f = — j^C2(3^ ^ -\-q^)(q ^ +3^^). The function ki is not a constant of the motion, but satisfies: [kuH} = --^P3 and hence [ku [ku {ku H}}} = 0, where P3 is defined in the appendix. 13.7.1.1 The Conformal Killing Vector Case We now seek a function w = w^pi-\-w^p2-\-w^, satisfying (30). We can no longer appeal to our previous knowledge of Killing vectors, but the calculation follows the same pattern. The quadratic (in pi) part of (30) gives: -2(q^-q^)diw^-\-w^-w^ =a(q^-q^), 2(q^ - q^)diw^ -\-w^ -w^ = a(q^ - q^), diw^-d2W^=0. Adding the first two of these leads to w;^ = 82^^^, w^ = 9iV^^, where x//^ satisfies a non-homogeneous form of (23), together with the linear wave equation: -2(^1 - q^)did2r + ^ir - ^iV = ct{q^ - q^). of - di)r = 0. Choosing the solution x//^ = q^q^ (having a = —3) leads to an equation of the form (31) with operator (33) defined in terms of V^^, with solution: We solve (23) and (34) in terms of a function of the single variable q^/q^. We obtain the function "tnh" of the appendix, giving our 'extra' solution of Figure 13.1. With this choice we find that w;^ = 0.
POISSON COMMUTING HAMILTONIANS 247 13.8 Conclusions In this chapter we have seen a connection between systems of hydrodynamic type (both homogeneous and non-homogeneous) and commuting quadratic integrals with terms linear in momenta. We have presented various classes of such systems. The simplest are associated with Killing vectors of the corresponding metric. Both these and solutions with nontrivial "electro-magnetic" term At are associated with solutions of the (linear) Euler-Poisson-Darboux equation (23) and fit nicely into the pattern depicted in Figure 13.1. Most of these results were presented in the 2 degrees of freedom (2 dimensional hydrodynamic system) context, but can be extended to higher dimensions. For higher dimensional flat and constant curvature metrics we can immediately write down the Killing vector solutions, such as the extension of the Gibbons-Tsarev equation given after example 13.4.2. In 2 dimensions there is a unique (up to coordinate change) semi-Hamiltonian, linearly degenerate system (equation (16)). In higher dimensions Ferapontov has given a general formula for such v^, involving n(n — I) arbitrary functions of a single variable (n — 1 functions of q^ for each /) (Ferapontov [1994]), the simplest being v^ = Xlj^i q^ — q\ which arose in the extension of the Gibbons-Tsarev equation. For such examples it is possible to find a system of equations analogous to (23) and a pattern of solutions analogous (but much larger than) Figure 13.1 (see Aujla [1999] and Aujla et al. [2000] for the 3 dimensional case), but once again no way of classifying all solutions. In higher dimensions, not all of our finite dimensional Hamiltonian systems are completely integrable, since our construction deals with first integrals of H and not involutivity. We are left with a number of open problems: 1. can we classify all non-homogeneities (p^ which arise in our scheme?; 2. can the "non-homogeneous Noether constant" approach of section 7 be applied in all cases?; 3. can we find a criterion for the integrability of our hydrodynamic systems?; and 4. for the integrable cases can we find a unified way of solving them or must we tackle each equation individually? We have discussed the first of these above. Regarding the second problem, the "duality" of section 13.6 may help, since the Poisson algebra of functions is invariant under canonical transformations. One interpretation of these "non-homogeneous Noether constants" is as usual Noether constants for a metric in an extended space. We may consider the linear terms in our functions H and F as derived from quadratic terms piP2, in a 3 degrees of freedom system with ignorable coordinate q^. This places the expression —g^^Ai in the (/, 3) slot of the 3 dimensional metric. The third and fourth questions are undoubtedly the most pressing, but also the most difficult. It is certainly true that there exist non-integrable cases, but all of our systems have special properties and some are integrable (sometimes in a trivial way). Gibbons and Tsarev have recently (Gibbons et al. [1999]) given a construction of the solutions of their equation (see example (13.4.1)). This is much more complicated than either the generalised hodograph method of Tsarev (see his article in this volume) or the inverse spectral transform.
248 A.P. FORDY Acknowledgements Most of the results presented in this chapter were obtained in collaboration with E.V. Ferapontov. References Aujla, K.S., and Fordy, A.P., Three component non-homogeneous systems of hydrodynamic type and commuting Hamiltonians. Preprint, 2000. Aujla, K.S., Hamiltonian systems and related equations of hydrodynamic type. PhD thesis. University of Leeds, 1999. Dubrovin, B.A., and Novikov, S.P, Hydrodynamics of weakly deformed soliton lattices, differential geometry and Hamiltonian theory. Russ. Math. Surveys 44, 35-124 (1989). Dubrovin, B.A., and Novikov, S.P., On Poisson brackets of hydrodynamic type. Sov. Math. Dokl 30, 651-4(1984). Dubrovin, B.A., and Novikov, S.P, Hamiltonian formalism of one-dimensional systems of hydrodynamic type and bogolyubov-whitham averaging method. Sov. Math. Dokl 27, 665-9 (1983). Ferapontov, E.V., and Fordy, A.P, Commuting quadratic Hamiltonians with velocity dependent potentials. Rep. Math. Phys., 44, 71-80 (1999). Ferapontov, E.V., and Fordy, A.P, Nonhomogeneous systems of hydrodynamic type, related to quadratic Hamiltonians with electromagnetic term. Physica D 108, 350-64 (1997). Ferapontov, E.V., and Fordy, A.P, Separable Hamiltonians and integrable systems of hydrodynamic type. J. Geom. andPhys. 21, 169-82 (1997). Ferapontov, E.V., Hydrodynamic-type systems. In N.H. Ibragimov, editor, CRC Handbook of Lie Group Analysis of Differential Equations, volume 1, chapter 14, pages 303-31. CRC Press, USA, 1994. Ferapontov, E.V., Integration of weakly nonlinear hydrodynamic systems in Riemann invariants. Phys. i>to. A 158, 112-8(1991). Flaschka, H., Forest, M.G., and McLoughlin, D.W., Multiphase averaging and the inverse spectral solution of the Korteweg-de Vries equation. Comm. PureAppl. Math. 33, 739-84 (1980). Gibbons, J., and Tsarev, S.P, Conformal maps and reductions of the Benney equations. Phys. Lett. A 258, 263-71 (1999). Gibbons, J., and Tsarev, S.P, Reductions of the Benney equations. Phys. Lett. A 211, 19-24 (1996). Blaszak, M., and Fordy, A.P, Solvable nonUnear evolution equations from integrable mechanical systems. In preparation, 1999. Mikhailov, A.V., Shabat, A.B., and Sokolov, V.V., The symmetry approach to classification of integrable systems. In Zakharov, V.E., editor. What is Integrability, pages 115-84. Springer- Verlag, Berlin, 1991. Tsarev, S.P, On Poisson brackets and one-dimensional Hamiltonian systems of hydrodynamic type. Sov. Math. Dokl. 31, 488-91, (1985). Tsarev, S.P, The geometry of Hamiltonian systems of hydrodynamic type, the generaUsed hodograph transform. USSRIzv. 37, 397-419 (1991). Tsarev, S.P, Integrability of equations of hydrodynamic type, from the end of the 19th to the end of the 20th century. In Braden, H.W., editor, Integrability, the Seiberg-Witten and Whitham Equations. Gordon and Breach/Harwood, New York, 1999. Whitham, G.B., Linear and Nonlinear Waves. Wiley, New York, 1974.
POISSON COMMUTING HAMILTONIANS 249 13.9 Appendix: The Euler-Poisson-Darboux Equation The Euler-Poisson-Darboux equation: Laf{x, y) = a{x - y)\lrxy -\-i^x - i^y =0, (36) occurs frequently in mathematical physics and is satisfied by circularly symmetric solutions of the standard linear wave equation in two spatial dimensions. It is important in several contexts related to this chapter. It arose in the calculation of non-homogeneities, with a = —2 (see equation (23)), and then as the governing equation for the potential function h (equation (27)), in which case the general solution can be written down: ,, , a(x)-\-b(y) h(x,y) = . x-y In the case a = 2 it is the equation satisfied by the flat coordinates for our flat metrics. It has a sequence of polynomial solutions, generated by the solution: T/r^ = (l-^jc)l/^(l-^3;)l/^ for arbitrary b. There exist "raising" and "lowering" operators: ra = -a(x^d^-\-y^dy)-\-x-\-y, l^ =-(d^c-\-dy), which shift us up and down this sequence. The Kernel of r^ is: for arbitrary function of one variable A (^). Choosing A (^) = ^ ~ ^ /^, we obtain the particular member: ' Go = (^j)'/'. Operating on Qo with la generates a sequence of non-polynomial solutions of (36), labelled Gn. For a = —2 some of these are: Po = 1, Pi=x-\-y, P2 = 3x^ -h 2xy -h 3/, P3 = 5jc^ -h 3x^y -h 3jc/ -h 5y^, P4 = 35jc^ -h 20x^y -\- ISjc^/ -h 20xy^ -\- 35/, eo = ^,en = (^,lg = Log(.-,).tnh = tanh-(yf). These functions satisfy the recursion relations: r(P„)(xP„+i, l{P„)(xP„.i, riQ„)(xQ„-i, l(Q„) oc Q„+u with: '•(20) = 0, /(tnh)aeo, /(lg) = 0.
14 Integrability of Equations of Hydrodynamic Type From the End of the 19th to the End of the 20th Century SERGUEI P. TSAREV Krasnoyarsk State Pedagogical University^ Lebedevoi 89, 660049 Krasnoyarsk, Russia tsarev @ edk. krasnoyarsk. su A remarkable parallelism between the theory of integrable systems of first-order quasilinear PDEs and some old results in the differential geometry of orthogonal curvilinear coordinates, conjugate nets, Laplace equations and their Bianchi-Backlund transformations is exposed. The applications of these results range from models in topological field theories to the celebrated Bieberbach Conjecture in the theory of conformal univalent mappings. 14.1 Introduction The modem theory of integrable systems of hydrodynamic type — i.e. quasilinear systems of first-order PDEs '"/\ /v\(u) '" vj^iu)^ /"^ .v-,(u) ... <(m) (1) where u^ = u^(x,t),i = l,-- - ,n are the dependent variables (unknown functions), x and t are one-dimensional space and time variables — started with the paper by Dubrovin and Novikov [1983] where a general hamiltonian formalism for such systems was proposed. Physical examples of (1) with a "natural" Hamiltonian structure are numerous, see e.g. (Bao et al. [1985], Dzyaloshinskii and Volovick [1980], Holm and Kupershmidt [1983, 1988], Holm [1985, 1986a, 1986b], Katz and Lebedev [1985] and Novikov [1982]) for physical examples of systems with Hamiltonian structures originating from infinite-dimensional Lie algebras naturally related to the respective systems; such Hamiltonian structures are a 251
252 S.P. TSAREV particular case (when the coefficients g^J (u) are linear in u^, and b^j^ are constant — see (9), (10) in the next section) of the more general structure described by Dubrovin and Novikov. Their generalization allows us to cast into Hamiltonian form another vast class of systems (1), derived from integrable nonlinear equations via the so called "averaging" procedure, which gives a system (usually of type (1)) describing slowly varying ("modulated") quasiperiodic solutions of the respective initial nonlinear equation — see Witham [1974], Avilov etal [1987], Chierchia [1987], Dubrovin and Novikov [1989], Lax [1957], Lax and Levermore [1979, 1983]). It is the Hamiltonian structure of Dubrovin and Novikov which made possible integration of such physically interesting systems of hydrodynamic type as the Whitham equations and many other systems originating from different models of gas or liquid motion. Some particular cases of this Hamiltonian structure for averaged equations were studied in Hayes [1973]. The simplest system (1) known to be integrable (via the classical hodograph transformation) since the end of the 19th century is the Euler system describing 1-dimensional perfect barothropic gas motions: Pt + (pu)x = 0 Ut -h UUx -h Px/P =0, p = p(p), where p(x,t) is the gas density, u(x,t) is the gas velocity. The classical hodograph transformation (see for example Rozhdestvenski and Yanenko [1983]) interchanges the dependent (p, u) and independent (jc, t) variables producing a linear 2x2 first-order system for the new unknown functions x(p,u), t(p,u); this system should be further integrated (for special barothropic laws p{p)) using different methods specific for linear systems of hyperbolic PDEs. Hereafter we will accept a "weak" definition of integrability: if the problem of solution of a nonlinear system is transformed to a linear problem, we will consider the initial nonlinear system "integrated". Fortunately for many integrable systems of hydrodynamic type the respective linear problem may be solved, this will be described below for the Whitham and Benney equation as well as ideal chromatography/electrophoresis equations. The classical hodograph method is limited to systems (1) with n = 2. Some physical examples of (1) such as the aforementioned Whitham equations (the averaged 1-phase KdV equation) (see Whitham [1974]): u^ = v^ (u)u^ v^ (u) - ^'+^'+^' - 2(^^-^^)/s:(5) Uj —V\\U)U^, V\\U) — 3 ?,{K{s)-E{s))' (here s^ — (u^ — u^)/{u^ — m'); E{s), K(s) are the complete elliptic integrals), Zakharov reduction of the Benney system (see Benney [1973] and Zakharov [1980]): hi+(q'h%=0, i^l,...,N, qi+q'qi+S^=0, S = h^ + ...+h^. or ideal chromatography equations (see Rozhdestvenski and Yanento [1983]): c-4 + (a'(m)+«'■), =0, i = l,...,N, a' = atkiu'/V, V = 1 + J2"=i ksu\
INTEGRABILITY OF EQUATIONS OF HYDRODYNAMIC TYPE 253 (where a/, kt and c are constants, u^ and a^ are concentrations of the non-absorbed and the absorbed i-ih component respectively) as well as equivalent to (5) isotachophoresis equations (see Babskii et ai [1988]) have n > 2. The integrability of these equations was established in Tsarev [1985, 1991] where a generalized hodograph method suitable for reducing the problem of solution of the above systems to a linear problem was given. Independently a partial result was obtained in Serre [1983]. This method may be applied to a wide range of "averaged" integrable equations which possess alongside with Dubrovin-Novikov Hamiltonian structure the remarkable property of diagonalizability: after a suitable change of the dependent variables u^ in (1) one obtains a system mJ = Vi(u)u^^, / = !,...,«, (6) (no summation over repeated indices hereafter!) with diagonal matrix Vi(u)8j. These new variables u^ are called Riemann invariants. For n = 2 every hyperbolic system (1) may be diagonalized; for n > 2 this is not true in general, and a special criterion (see Haantjes [1955]) shall be applied to check diagonalizability of (1). On the other hand the methods of Tsarev [1985, 1991] lead to an unexpected link between the theory of Hamiltonian diagonalizable hydrodynamic type systems and a classical object of local differential geometry intensively studied at the end of 19th and beginning of 20th century — orthogonal curvilinear coordinates in flat Euclidean space R^. Practically all "physical" entities (conservation laws, symmetries etc.) have their counterparts in the theory of orthogonal curvilinear coordinates; amazingly enough the basic formulae relating conservation laws, symmetries and the Hamiltonian structure of integrable systems of hydrodynamic type may be found in Darboux [1910]! Further investigation discovered a deep relation between other types of integrable nonlinear systems studied in the modem theory of integrable physical equations and classical problems studied in Darboux [1887-1896, 1910], Bianchi [1955], Guichard [1905, 1935, 1936] and other papers at the turn of this century. Differential-geometric methods of integration for (6) were later extended to some classes of non-homogeneous systems as well as some (2+l)-dimensional homogeneous systems 'w\(u) ■■■ w^„(u) (7) (8) In fact one may say that the integrability of systems (7), (8) describing different local geometric objects was estabUshed by Darboux, Bianchi, Tzitz6ica et al. some 80 years ago.
254 S.P. TSAREV This review is devoted to (an inevitably brief) exposition of this remarkable link as well as some recently discovered deeper relations between the classical differential geometry that flourished at the end of the 19th century and the modem theory of integrable nonlinear PDEs appearing in different applications to mathematical physics. Applications of this theory to topological quantum field theory given for the first time in Dubrovin [1991] and Krichever [1994] are described in more detail in other reviews. The bibliography given in the end of the review is certainly very incomplete; a better review of different branches of current research undertaken in the theory of integrable systems of hydrodynamic type even in the past decade would require a separate volume; tracing all links of this theory to the results in differential geometry would enlarge it by an order of magnitude. The author apologies for all inevitable omissions and expects that a reader interested in a particular topic will be able to find further papers of the authors given. 14.2 Diagonal Systems of Hydrodynamic Type and Orthogonal Curvilinear Coordinate Systems in R^ Let us recall briefly the main results of Dubrovin and Novikov [1983], Tsarev [1985,1991]. A (generally nondiagonal) system (1): u\ = Yl]=i ^j (^Wx is caUed Hamiltonian if there exist a Hamiltonian — a functional of the following form: H = f h(u)dx — and a Hamiltonian operator, A'J =g'\u)^-hl/j^\u)4 (9) which define a skew-symmetric Poisson bracket on functionals r 81 ^ 8J J dU^(x) duJ{x) This bracket should satisfy the Jacobi identity and generate the system u\(x) = {u'(x), H] = J^Aij-^ = J2^g'Jdkdjh-^l/j/djh)4 = ^4(m)w^, (10) j "^^ ^^^ j,k k where ds = d/du^. B.A. Dubrovin and S.P. Novikov [1983] proved that necessary and sufficient conditions for Atj to be a Hamiltonian operator in the case of a non-degenerate matrix g^J are: (a) g^J = gj^, i.e. the inverse matrix g~^ defines a Riemannian metric. (b) b^j^ = ~S^^^ik ^^^ ^^ standard Christoffel symbols F^^ generated by gij: 'j ^ 2 \ dui duJ du^ J ' (c) the metric gij has identically vanishing curvature tensor: RW^ = 0. In this case we have Vj(u) = V^Vjh = g^^VsVjh with the covariant derivatives defined by^/y.
INTEGRABILITY OF EQUATIONS OF HYDRODYNAMIC TYPE 255 Lemma 14.1 (Tsarev [1985]). In order that a matrix vUu) be a matrix of a Hamiltonian system (1) with a nondegenerate metric in A^^ it is necessary and sufficient that there exists a nondegenerate zero curvature metric gij (u) such that (a) gikv^ = gjkv^ and (b) Vyi;[ = Vjtuj, where V is the covariant differentiation generated by the metric gij. For a diagonal matrix v^. (u) = Vj (M)3y this implies that gij (u) is also diagonal and Ji^!!^ = rli = ldi\ngkk, di = d/du^ (11) Vi -Vk 2 (we recall that we do not imply the summation on repeated indices!). From (11) we deduce dj-^^ = di-^^^, i^jj^k. (12) Vi - Vk Vj - Vk From a differential geometric point of view, giving a zero curvature nondegenerate diagonal metric is equivalent to giving an orthogonal curvilinear coordinate system on a flat (possibly pseudo-Euclidean) space (see Darboux [1910]). Locally these coordinate systems are determined by n(n — l)/2 functions of two variables (see Bianchi [1955]). A striking fact can be discovered: formula (11) was found in Darboux [1910] (p. 353)! This formula is crucial for the integrability property of diagonal Hamiltonian systems (6): if we interpret it as an overdetermined (compatible in view of the zero curvature property of g) system on n unknown functions Vj(u) (ga given) we can generate from every solution Wj(u) a symmetry (commuting flow) u\ = Wi{u)u\, i = 1,..., n, of (6) and a solution of (6) (the generalized hodograph method): Theorem 14.1 (T«arev [1985, 1991]). Any smooth solution u^ (jc, t) of a semihamiltonian system (6) in a neighbourhood of a generic point (jcq, ^o) (where m^ are non-zero for all i) can be found as the solution of the system Wk(u) = Vk(u)t-\-X, / = l,...,n, (13) where Wk(u) are solutions of the linear system diWk = r^,(M) • (^i - Wk), i # k, rli = diVk/(vi - Vk) = l/2di \ngkk. One can prove (see Tsarev [1991]) the completeness property for this class of symmetries and solutions parameterized by n functions of 1 variable, the generic Cauchy data for our diagonal system (6). The corresponding geometric notion used in the theory of orthogonal curvilinear coordinate systems corresponding to (11) is the so called Combescure transformation (see Darboux [1910]).
256 S.P. TSAREV Definition. ^ Two orthogonal curvilinear coordinate systems x^ = jc^(m^ ... , m") and jc^ = jc^(m^ ... , m") m the same flat (pseudo)Euclidean space R^ = {(x^ ... , x")} are said to be related with a Combescure transformation (or simply parallel) if and only if their tangent frames ei = dx/du^ andet = dx/du^ are parallel in points corresponding to the same values of curvilinear coordinates u\ Let us take the quantities Hi (u) = |2f | = y^, Hi (u) = \ei \ (Lame coefficients). Lemma 14.2 The quantities Wi(u) = Hi(u)/Hi(u) satisfy (11) with V^- = diHk/Hk, the connection coefficients for the metric gu = Hf. Conversely^ for any solution Wi of (11) gu = (wiHi)^ will give an orthogonal curvilinear coordinate system related to the coordinate system with the metric gu = Hf by a Combescure transformation. The theory of Combescure transformations coincides with the theory of integrable diagonal systems of hydrodynamic type. Physical examples of such systems (Whitham equations, Benney equations) have Hamiltonian structures (9), (10) with diagonal metrics gu possessing the so called Egorov property: digkk = \gii' As we have demonstrated earlier (Tsarev [1991]) this is a consequence of Galilei invariance of the original systems. See also Dubrovin [1990] for the algebro-geometric background of this property for averaged integrable systems. Using this property and homogeneity of coefficients one can find explicit formulas for solutions of (11) for the systems in question (see Tsarev [1994], Kudashev and Sharapov [1991a,b] and Tian [1994]). The class of Egorov orthogonal curvilinear coordinate systems is interesting in itself and merits our special attention. 14.3 Egorov Coordinate Systems; Nonhomogeneous and (2+l)-dimensional Integrable Systems of Hydrodynamic Type Introducing Pik{u) =^diHk/Hi, i 7^ k, Puiu) = 0 (rotation coefficients of a given orthogonal curvilinear coordinate system with gu = Hf) one can easily check the following: (a) the vanishing of the curvature tensor is equivalent to dj^ik = fiijfijk, i+i+ K (15) s^i,k (b) the Egorov property digkk = hga reduces to Pik = Pki^ (17) In the Egorov case condition (16) is equivalent to r^S/it =0,T = di-\-.. .-|-9„. Consequently the problem of classification of Egorov coordinate systems is reduced to description of all off-diagonal symmetric matrices (fiik) satisfying (15) and Tfiik = 0. B.A. Dubrovin [1990] had observed that this problem coincides with the purely imaginary reduction of the well-known integrable system describing resonant A/'-wave interactions.
INTEGRABILITY OF EQUATIONS OF HYDRODYNAMIC TYPE 257 Namely, restriction of ^ik on any (jc, t) plane u^ = a^x -\-b^t gives (compare, for example, Nowikow etal. [1984]) [A, r,] - [B, rj = [[A, n, [5, r]], A = diag(a\ ... , a"), 5 = diag(b^,... , ^"), T = (Ptk) with additional reduction Im r = 0, r^ = r. For the case A/^ = 3 this reduces to b}-\-cibl =Kb^b^, bj-\-C2bl =Kb^b^, (18) b^-\-C3bl =Kb^b^. This is a system of type (7), integrable by the 1ST method (see Novikov et al. [1984]). Now we can compare the progress achieved in the modem integrability theory for (18) and the results obtained more than 90 years ago in the theory of Egorov coordinate systems initiated by G. Darboux in 1866 and continued by D.Th. Egorov in 1901 in his thesis ([1970]). It was Darboux [1910] who proposed to call this special type of coordinate systems Egorov type systems. From the point of view of integrability properties remarkable progress was achieved by L. Bianchi in 1915 ([1955]). He found a Backlund transformation for this problem and established the permutability property as well as the superposition formula for it. We also note that the pioneering results on Backlund transformations and their permutability in the well-known theory of constant curvature surfaces in R^ are due to Bianchi as well. For an exposition of this theory see Tsarev [1992] and Ganzha and Tsarev [1996]. One can enjoy reading (Case and Chiu [1977] and Kaup [1981]) where these formulas were rediscovered in the context of the 3-wave system. So the basic integrability results for (18) were established long ago by Darboux, Egorov and Bianchi certainly with the exception of the 1ST transformation. An unexpected result (hidden in Darboux [1910]) consists of the existence of a homogeneous system (1) of three equations related to (18) by a nonlocal transformation. Geometrically this is trivial: given an orthogonal curvilinear coordinate system in R^ we have in each its point P(jco, jo, ^o) the orthogonal 3-frame of tangent planes Z = Pk(x - xo) -h qk(y - yo) + zo, k = 1, 2, 3. (19) Let us parameterize it by 3 functions A (jc, y, z), B(x,y, z),C(x,y, z), with the coefficients Pk, qk of the tangent planes being the three solutions of pq+Ap + Bq= 0, p^-q^+ 2(Cp -h Hq) = 0, 2(BC - AH) + 1=0, different from the trivial solution p = q = 0 (see Bouligand [1953]). Then the Frobenius compatibility conditions for these three families of distributions (19) gives a system of three homogeneous first-order equations of type (8): 2(AQ - CAz) = 2Cy -\-By-Ajc 2(BHz - HBz) = 2H^ + By - A^ (20) AH, - HA, + BC, - CB, = Ay - B, where H = (2BC - 1)/2A .
258 S.R TSAREV For this, system one can reformulate the Backlund-like transformation given in Bianchi [1955] in terms of Pik(u). A number of different transformations producing (with quadratures) solutions of (20) parameterized by arbitrary many functions of one variable may be found in Darboux [1910]. Thus (20) is integrable. If we search for solutions of (20) which do not depend on z then a remarkable nondiagonalizable integrable system (1) with three equations appears. Since one can easily prove the equivalence of z-independence in (20) and the Egorov property (17) we have found a homogeneous system related to (18) by a nonlocal change of variables. In the Euler (p, x//, 0 parameterization of orthogonal 3-frames it reads ' V^A / — cos^ (p — sin ^ cos (p/ sin 0 ^^1 = 1 — sin ^ sin ^ cos ^ — sin^^ 0 ) ( ^;c | (21) ,(Pt / \ —cos^(l-hcos^^) — sin^cos^cos^/sin^ 1> This nonlocal change does not affect the existence of higher order conserved densities. Recently Ferapontov [1993] has proved the uniqueness result for such 3x3 homogeneous systems possessing higher-order conserved densities: they may be transformed to (21) by reciprocal and point transformations. Also another nonlocal transition from (18) to (21) was given there. The matrix of (21) has constant eigenvalues — 1, 0, -h 1 but its eigenvector fields (properly normalized) form a so(3) Lie algebra, consequently (21) is a non-diagonalizable (1) integrable system. The complete system (21) certainly may be called a (2+l)-dimensional generalization of the (l+l)-dimensional 3-wave system (18). Orthogonal curvilinear coordinate systems in R^ also provide only a (2-\-1 )-dimensional generalization of the (l+l)-dimensional N-wave system since they are parameterized hyn(n — l)/2 functions of two variables (L. Bianchi). 14.4 Semihamiltonian Diagonal Systems and Coordinate Systems with Conjugate Lines The class of integrable diagonal systems (6) is wider than the class of Hamiltonian systems of this type. Namely, the property (12) which is a weaker consequence of the hamiltonian property is sufficient (Tsarev [1991]). Let us call a diagonal system semihamiltonian if n = 2 or if n > 2 and vt (u) satisfy (12). As a physical example of a semihamiltonian (but non-hamiltonian for n > 3) system one can mention the ideal Langmuir chromatography and electrophoresis systems (Tsarev [1991]). To every semihamiltonian system we can relate a diagonal metric ga (u) via 8/ In gkkf^ = ^i^kli'^i — Vk)- This metric is not flat in general though some coefficients of the curvature tensorvanishasaconsequenceof(12).Namelyintroducing/fi = y/gii ,fiikM = dtHk/Ht, we can find out that (12) is equivalent to the set (15) of equations on Pik. Solutions of (15) may be parameterized by n(n — 1) functions of 2 variables. This system coincides with the compatibility conditions for a linear system ^ifk = Pikfi. ii^k.
INTEGRABILITY OF EQUATIONS OF HYDRODYNAMIC TYPE 259 Restricting (15) on 3-dimensional planes u^ = a^ x-\-b^ y-\-c^ z iTi R^ we obtain (for general nonvanishing constants a\b\d)2i (2+1 )-dimensional nonhomogeneous system on n (n — 1) quantities ^ik{x, j, z). As we have seen earlier the theory of Hamiltonian diagonal systems (6) is closely related to the theory of orthogonal curvilinear coordinate systems in R^. The geometric background for the theory of semihamiltonian systems is given by the theory of coordinate systems with conjugate coordinate lines (see Darboux [1910], 1887-1896], t. 4, ch. 12). A general (non-orthogonal) coordinate system x{u\,U2,m)in R^ is called a system with conjugate coordinate lines (or simply a conjugate coordinate system) if on every coordinate surface St^ = [uq = const} at every point P(jco, jo, ^o) the lines of intersection of this surface with two other coordinate surfaces belonging to other one-parameter families of coordinate surfaces and containing P(jco, jo, ^o) are conjugate on St^ (with respect to its second fundamental form); this is equivalent to the equations (22) below. Every orthogonal curvilinear coordinate system is conjugate due to Dupin's theorem mentioned above. The theory of conjugate coordinate systems was developed by Darboux and others and borrowed many results from the classical theory of conjugate coordinate nets on surfaces inR^ (known as "nets" or "reseaux", see Eisenhart [1962], Guichard [1905, 1935, 1936] and Tzitzecia [1924]. A number of Backlund-like transformations for these coordinate systems was given with permutability properties. Every conjugate coordinate system x^ = Jc^(M^ ... , m") in /?" = {(jc^ ... , jc") is characterized by the conditions of conjugacy of coordinate lines: didkx = rli(u)dkx -h rij^(u)dix, i # k. (22) This system of equations coincides with the system describing hydrodynamic type conserved quantities of a semihamiltonian system (seeTsarev [1991]). The quantities F^- in (22) satisfy the compatibility conditions which are equivalent to the semihamiltonian property (12). Introducing Hi(u) as solutions of diHk(u) = r^-Hk(u) and fitk = diHk/Hi, i ^ k, we obtain a set of fiik satisfying (15). The converse is also true: given a solution fitk of (15) one can find (a number of) semihamiltonian systems related to it. Any semihamiltonian system also may be related to a Combescure transformation of conjugate coordinate systems (see Darboux [1910]). This geometric interpretation provides another example of an integrable system (8). Namely, given a conjugate coordinate system in R^ one can take the field of its (non- orthogonal) tangent 3-frames (?i, ?2, ^s) and parameterize it by 6 independent functions e[(x,y,z),i = 1, 2, A: = 1, 2, 3 (the coefficients el may be set to 1 due to normalization). Then the Frobenius compatibility conditions give 3 homogeneous first-order PDE's on e[(x,y,z)J = 1, 2, A: = 1, 2, 3. Another 3 equations are given by the conjugacy condition det((?/ • V)2jt, 2/, ek) = OJ < k. This system of 6 equations is a homogeneous system (8) in question. Its z-independent solutions satisfy a system of type (1) enjoying properties analogous to those of (21): it has constant eigenvalues —1, 0, -hi (all doubly degenerate) and 6 linearly independent fields of eigenvectors forming (if properly normalized) a nontrivial Lie algebra. This remarkable system was studied in many recent publications (see papers
260 S.R TSAREV by E.V. Ferapontov, E.V. Zakharov, M. Manas, A. Doliwa). A generalization of the classical Laplace transformations of hyperbolic equations Zxy = «(jc, y)zx + b{x, y)zy -h c(jc, y) (see Darboux [1887-1896], t. II for the detailed exposition of Laplace transformations — which have nothing to do with the Laplace transforml) to the case of systems (22) was given in Darboux [1887-1896], t. IV; see also Ferapontov [1987] and references therein. 14.5 Further Topics In the past 15 years a lot of progress has been achieved in the study of Hamiltonian structures of more general type than that of Dubrovin and Novikov. A nonlocal formalism was developed in Ferapontov [1991], Mokhov and Ferapontov [1991], Mokhov [1992a,b]; this formalism is capable of encompassing virtually all semi-hamiltonian diagonal systems as well as many others. All known physical examples of integrable systems of hydrodynamic type (6) (including Whitham and Benney systems) have multi-hamiltonian structure: see Tsarev [1991] and subsequent publications by M.V. Pavlov, E.V. Ferapontov and O.I. Mokhov. Weakly nonlinear semihamiltonian systems (i.e. systems (6) with dtVi = 0 — no summation on / — such systems are also called "linearly degenerate") were studied in Ferapontov [1990] and Serre [1989]. The theory of such systems is connected to the theory of n-webs on Euclidean plane, Dupin eyelids and Stackel metrics (E.V. Ferapontov). Among the results are: quasiperiodic behaviour of their solutions (Serre [1989], complete description of such systems and complete sets of their hydrodynamic symmetries (Ferapontov [1990]). E.V. Ferapontov communicated to the author the following fact: any n-phase (n-zone) quasiperiodic (or a n-soliton) solution of the KdV equation can be represented by a solutionof a weakly nonlinear semihamiltonian system/?J = (Jlk^i R^)Rx,i = I, .. ,n. These results may be compared with Curro and Fusco's results [1987] in the soliton-like interactions of Riemann simple waves for some 2x2 systems. Temple type systems (Temple [1983]) were studied in Agafanov and Ferapontov [1996] and recent publications by D. Serre. As proved in Tsarev [1991], the Whitham and Benney equations possess in addition to conservation laws and symmetries of hydrodynamic type studied here other higher order symmetries and recursion operators. Many physical 2x2 systems in fact have these properties, see Sheftel [1986, 1993], Teshukov [1989], Pavlov and Tsarev [1991] and Serre [1983]. IST-like methods were developed in Geogdzhaev [1987], Kodama and Gibbons [1989, 1991] for some diagonal Hamiltonian systems of physical importance. Certainly this approach should be related to our geometric methods. An important and rapidly growing field is the theory of discretizaton of continuous geometric objects studied in this review; see e.g. Bobenko and Hertrich-Jeromin [1997] and papers by A. Doliwa, M. Manas, P. Santini (some of them are obtainable from solv-int@xxx.lanl.gov). The important relation of Backlund transformations and the discretized systems deserves detailed study. Investigation of finite-parametric reductions of the initial (infinite) Benney moment
INTEGRABILITY OF EQUATIONS OF HYDRODYNAMIC TYPE 261 equations A^l + A^+i + nA^'-^Al =0, n = 0, 1,... , oo, (23) has resulted in a classification theory of such reductions (the mentioned above Zakharov reduction (4) is a particular example of such reduction of (23) to a finite system (1)) bearing a close resemblance to the Loewner method which was used by de Branges in his final proof of the celebrated Bieberbach Conjecture in the theory of univalent conformal mappings (see an exposition of the history of this Conjecture in Fomenko and Kuz'mina [1986], and Gibbons and Tsarev [1986, 1989] for the details). Acknowledgements The author enjoys the occasion to express his gratitude to the organizers of the Workshop "Integrability: the Seiberg-Witten and Whitham equations" (Edinburgh, September 14th- 19th, 1998) for their hospitality; my participation in the Workshop was supported by a grant from EPSRC. This work was partially supported by INTAS grant 96-0770 and Russian Presidential grant 96-15-96834. References Agafonov, S.L, and Ferapontov, E.V., Systems of conservation laws from the point of view of projective theory of congruences, hvestia RAN (Math.) 60, 1-30 (1996). Avilov, V.V., Krichever, I.M., Novikov, S.P., Evolution of the Whitham zone in Korteweg-de Vries theory, Soviet Phys. Doklady 32 (1987). Babskii, V.G., Zhukov, M. Yu., and Yudovich, V.I., Mathematical theory of electrophoresis. Kiev, 1983; Plenum Press, N.Y., 1988. Bao, D., Marsden, J., Walton, R., The Hamiltonian structure of general relativistic perfect fluid, Comm. Math. Phys. 99, 319-345 (1985). Benney, D.J., Some properties of long nonlinear waves. Stud. Appl. Math. 52, 45-50 (1973). Bianchi, L., Opere 3: Sisteme tripli ortogonali, Ed., Cremonese, Roma (1955). Bikbaev, R.F., and Novokshenov, V.Yu., Self-similar solutions of the Whitham equations and the Korteweg-de Vries equation with finite-gap boundary condition, Proc. 3 Intern. Workshop "Nonlinear and turbulent processes in physics", Kiev, v.l, p.32 (1987). Bobenko, A., Hertrich-Jeromin, U., On discrete triply orthogonal systems and Clifford algebras. Preprint SFB-288, Berlin, 1997. Bouligand, G., Une forme donnee a la recherche des systemes triples orthogonaux, C.R. Acad. Sci. Paris 236, 2462-2463 (1953). Case, K.M., and Chiu, S.C, Backlund transformation for the resonant three-wave process, Phys. Fluids 20, 76S (1977). Chierchia, L., Ercolani, N., and D.W McLauhglin, On the weak limit of rapidly oscillating waves, Duke Math. J. 55, 759-764. (1987). Curro, C, and Fusco, D., On a class of quasilinear hyperbolic reducible systems allowing for special wave interactions, Zeitschr Angew. Math. andPhysik 38, 580 (1987). Darboux, G., Legons sur la theorie generale des surfaces et les applications geometriques du calcul infinitesimal t. 1-4, Paris (1887-1896). Darboux, G., Legons sur les systemes orthogonaux et les coordonnees curvilignes, Paris (1910).
262 S.R TSAREV Demoulin, A., Recherches sur les systemes triples orthogonaux. Memoirs Soc. Roy. Sciences Liege 11, 98 (1921). Dubrovin, B.A., and Novikov, S.R, Hydrodynamical formalism of one-dimensional systems of hydrodynamic type and the Bogolyubov-Whitham averaging method, Soviet Math. Doklady 27, 665-669 (1983). Dubrovin, B.A., and Novikov, S.R, Hydrodynamics of weakly deformed soliton lattices. Differential geometry and Hamiltonian theory, Russ. Math. Surveys 44, 29-98 (1989). Dubrovin, B. A., Differential geometry of moduU spaces and its applications to soliton equations and to topological conformal field theory. Preprint n.l 17, Scuola Normale Superiore, Risa, November, p.31 (1991). Dubrovin, B.A., Geometry and integrability of topological-antitopological fusion. Commun. Math. Phys. 152, 539-564 (1993). Dubrovin, B.A., Geometry of 2D topological field theories. In: Springer Lecture Notes in Math. 1620, 120-348 (1996). Dubrovin, B.A., Hamiltonian formalism of Whitham-type hierarchies and topological Landau — Ginsburg models. Commun. Math. Phys. 145, 195-207 (1992). Dubrovin, B.A., Integrable systems in topological field theory, Nucl. Physics B 379,627-689 (1992). Dubrovin, B.A., On the differential geometry of strongly integrable systems of hydrodynamic type. Funk. Anal 24, 280-285 (1990). Dubrovin, B.A., The differential geometry of moduli space and its applications to soUton equations and to the topological conformal field theory, Surv. Differ Geom. TV, 213-238, Int. Press, Boston MA, (1998). Dubrovin, B.A., Weakly perturbed sohton lattices. Nucl. Phys. B (Proc. Suppl.) 18A, 23 (1990). Dzyaloshinskii, I.E., and Volovick, G.E., Roisson brackets in condensed matter physics, Ann. of Phys.125, 67-97 (19S0). Egorov, D.Th., Collected papers on differential geometry, Nauka, Moscow (1970) (in Russian). Eisenhart, L.R, Transformations of surfaces, Princeton (1923), 2nd ed., Chelsea (1962). Ercolani, N., et al., Hamiltonian structure for the modulation equations of a sine-Gordon wavetrain, Duke Math. J. 55, 949-983 (1987) Ferapontov, E.V., and Fordy, A.R, Separable Hamiltonians and integrable systems of hydrodynamic type, J. of Geometry and Physics 21, 169-182 (1997). Ferapontov, E.V., and Mokhov, O.I., The equations of associativity as hydrodynamic type systems: Hamiltonian representation and the integrability. In Proc. NEEDS-95, Lecce, Italy. Ferapontov, E.V., Differential geometry of nonlocal Hamiltonian operators of hydrodynamic type, Func. Anal. itsAppl'.lS (1991). Ferapontov, E.V., Dupin hypersurfaces and integrable Hamiltonian systems of hydrodynamic type, which do not possess Riemann invariants. Diff. Geometry and its Appl. 5, 121-152 (1995). Ferapontov, E.V., Integrability of weakly nonlinear hydrodynamic systems in Riemann invariants, Phys. Letters A 158, 112-118 (1991). Ferapontov, E.V., Integration of the weakly nonlinear semihamiltonian systems of hydrodynamic type with the web theory methods. Mat. Sbomik 181, 1220 (1990). Ferapontov, E.V., Isoparametric hypersurfaces in spheres, integrable nondiagonalizable systems of hydrodynamic type, and A^-wave systems. Diff. Geometry and Appl, 1995. Ferapontov, E.V., Laplace transformations of hydrodynamic-type systems in Riemann invariants: Periodic sequences, J. Phys A; Math. Gen. 30, 6861-6878 (1997). Ferapontov, E.V., On the integrability of 3 x 3 semihamiltonian hydrodynamic type systems which do not possess Riemann invariants, Physica D 63, 50-70 (1993). Ferapontov, E.V., On the matrix Hopf equation and integrable Hamiltonian systems of hydrodynamic type, which do not possess Riemann invariants. Physics Letters A 179, 391-397 (1993). Raschka, H., Forest, M.G., and McLaughlin, D.W., Multiphase averaging and the inverse spectral solution of the Korteweg-de Vries equation, Comm. Pure Appl Math. 33, 739 (1980). Fomenko, O.M., and Kuz'mina, G.V., The last 100 days of the Bieberbach conjecture. Math. Intelligencer 8( 1), 40-47 (1986).
INTEGRABILITY OF EQUATIONS OF HYDRODYNAMIC TYPE 263 Ganzha, E.L, and Tsarev, S.P., An algebraic formula for superposition and the completeness of the Backlund transformations of (2+l)-dimensional integrable systems, Russian Math. Surveys 51 (1996) No 6, 1200-1202. also solv-int@xyz,lanl.gov, 1996, No 9606003. Geogdzhaev, V.V., Solutions of Benney equations by the method of the inverse scattering problem, Theor Math. Phys. 73, 255 (1987). Gibbons, J., and Tsarev, S.P., Conformal maps and reductions of the Benney equations. Physics Letters A, 1999. Gibbons, J., and Tsarev, S.P., Reductions of the Benney equations. Physics Letters A 211, 19-24 (1986). Grinevich, RG., Nonisospectral symmetries of the KdV equation and the corresponding symmetries of the Whitham equations, NATO Adv. Sci. Inst. Ser B. Phys. 320, Plenum NY (1984). Guichard, C, Sur les systemes triplement indetermines et les systemes triples orthogonaux. Collection Scientia, no. 25, Paris, 95 p. (1905). Guichard, C, Theorie des reseaux. Memorial Sci. Math. 74, Paris (1935). Guichard, C, Theorie generale des reseaux, appUcations, Memorial Sci. Math. 77, Paris (1936). Gumral, H., and Nutku, Y, Hamiltonian structure of equations of hydrodynamic type, J. Math. Phys. 31, 2606 (1990). Haantjes, J., On X„_i-forming sets of eigenvectors, Proc. Konink. Nederl. Akad. Wet. 58(2); Indagationes Mathematicae, 17(2), 158-162 (1955). Hayes, W.D., Group velocity and nonlinear wave propagation, Proc. Roy. Soc. (London) Ser. A 332, 199-221 (1973). Holm, D.D., and Kupershmidt, B. A., Hamiltonian formulation of ferromagnetic hydrodynamics, Phys. Letters A 129, 93-100 (1988). Holm, D.D., and Kupershmidt, B.A., Poison brackets and Clebsch representation for magnetohydro- dynamics, multifluid plasmas, and electricity, Physica Z), 6, 347-363 (1983). Holm, D.D., Hamiltonian formulation for Alfven wave turbulence equations, Phys. Letters A 108, 445-447 (1985). Holm, D.D., Hamiltonian formulation of a charged fluid, including electro- and magnetodynamics, Phys. Letters A 114, 137-141 (1986b). Holm, D.D., Hamiltonian formulation of baroclinic quasigeostrophic fluid equations, Phys. Fluids 29, 7-8 (1986a). Katz, E.L, and Lebedev, V., Nonlinear dynamics of smectic liquid chrystals with orientational ordering in the layer, Soviet Phys. JETP 61 (1985). Kaup, D.J., The lump solutions and Backlund transformation for the 3-dimensional 3-wave resonant interaction, J. Math. Phys. 22(6), 1176-1181 (1981). Kodama, Y, and Gibbons, J., Integrability of the dispersionless KP hierarchy, in: Proc. Intern. Workshop "Nonlinear and turbulent processes in physics", Kiev, 1989. Kodama, Y, and Gibbons, J., Solving dispersionless Lax equations, Proc. NATO ARW "Singular Limits and Dispersive Waves", Lyon, France, 8-12 July, 1991. Krichever, I., The t-function of the universal Whitham hierarchy, matrix models and topological field theories, Commun. Pure and Applied Math. 47, 437-475 (1994). Krichever, I.M., Averaging method for two-dimensional "integrable" equations. Functional Anal. Appl. 22 (1988). Kudashev, V.R., and Sharapov, S.E., Hydrodynamic symmetries for the Whitham equations for the nonlinear Schrodinger equation (NSE), Phys. Lett. A, 154, 445 (1991b). Kudashev, V.R., and Sharapov, S.E., The inheritance of KdV symmetries under Whitham averaging and hydrodynamic symmetries for the Whitham equations, Theor Math. Phys. 87, 40 (1991a). Lax PD, Hyperbolic systems of conservation laws II. Comm. in Pure and Appl. Math. 10, 537-566 (1957). Lax, P.D., and Levermore, CD., The Small Dispersion Limit for the Korteweg-de Vries Equation I, II, III, Comm. Pure Appl. Math. 36, 253-290, 571-593, 809-830 (1983). Lax, P.D., and Levermore, CD., The Zero Dispersion Limit for the Korteweg-de Vries Equation, Proc. Nat. Acad. Sci. USA 76(8), 3602-3606 (1979).
264 S.R TSAREV Lax, P.D., Hyperbolic systems of conservation laws 11. Comm. in Pure andAppl. Math. 10, 537-566 (1957). Mokhov, O.I., and Ferapontov, E.V., On the nonlocal Hamiltonian operators of hydrodynamic type connected with constant-curvature metrics, Russian Math. Surveys, 45 (1991). Mokhov, O.I., Differential equations of associativity in 2D topological field theories and geometry of nondiagonalizable systems of hydrodynamic type. In: Abstracts of Intemat. Conference on Integrable Systems "Nonlinearity and Integrability: from Mathematics to Physics", February 21-24, 1995, Montpellier, France (1995). Mokhov, O.I., Hamiltonian systems of hydrodynamic type and constant curvature metrics. Phys. Letters A, 166, 215-216 (1992). Mokhov, O.I., Symplectic and Poisson geometry on loop spaces of manifolds and nonlinear equations. Trans. Amer Math. Soc. ser. 2. v.170. 121-151 (1995). Mokhov, O.I., Symplectic and Poisson geometry on loop spaces of manifolds and nonlinear equations. Advances in Soviet Mathematics. S.P Novikov seminar. Ed. S.P Novikov. 22, AMS, Providence, USA (1995). Novikov, S.P, Hamiltonian formalism and a muti-valued analogue of Morse theory. Russian Math. Surveys 37 (1982); Theory ofSolitons, eds. S.R Novikov et al.. Plenum Press, N.Y., 1984. Pavlov, M.V., andTsarev, S.R, On the conservation laws for Benney equations, Russian Math. Surveys, 45 (1991). Rozhdestvenski, B.L., and Yanenko, N.N., Systems of quasilinear equations and their appUcations to gas dynamics, AMS, 1983. Schief, W.K., Rogers, C.S., and Tsarev, S.R, On a 2+1-Dimensional Darboux System — Integrable and Geometric Connections, Chaos, Solitons & Fractals 5(12), 2357-2366 (1995). Serre, D., Int'egrabiUt'e d'une classe de systemes de lois de conservation. Forum Math 4, 607-623 (1992). Serre, D., Invariants du premier ordre des systemes hyperboliques semi-Uneares. J. ofDiff. Eq. 49, 270-280 (1983). Serre, D., Systemes d'EDO invariants sous Taction de systemes hyperboliques d'EDP, Ann. Inst. FoMr/er39, 953 (1989). Sheftel, M.B., Higher integrals and symmetries of semi-Hamiltonian systems, Dijf. Equations 29(10), 1782-1795 (1993). Sheftel, M.B., Integration of Hamiltonian systems of hydrodynamic type with two dependent variables with the aid of the Lie-Backlund group, Func. Anal Appl. 20, 227-235 (1986). Sheftel, M.B., On the infinite-dimensional noncommutative Lie-Backlund algebra associated with the equations of one-dimensional gas dynamics, Theor Math. Phys. 56, 878-891 (1983). Sheftel, M.B., On the integration of the Hamiltonian systems of hydrodynamic type with two dependent variables with the help of a Lie-Backlund groups. Funk. Anal. 20 (1986). Sheftel, M.B., On the Lie-Backlund groups admissible by the equations of one-dimensional gas dynamics, Vestnik of Leningrad State Univ. No. 7, 37-41 (1982) (in Russian). Temple, B., Systems of conservations laws with invariant submanifolds. Transactions of the American Mathematical Society 280(2), 781-795 (1983). Teshukov, V.M., HyperboUc systems admitting a nontrivial Lie-Backlund group. Preprint LEAN, Leningrad, 1989. Tian, F.R., On the InitialValue Problem of the Whitham Averaged System, Proc. NATO ARW "Singular Limits and Dispersive Waves", Lyon, France, 8-12 July, 1991. N.M. Ercolani, (ed.) et al.. New York, NY: Plenum, NATO ASI Ser., Ser. B, Phys. (1994). Tsarev, S.R, Classical differential geometry and integrabiUty of systems of hydrodynamic type, in: Proc. NATO ARW "Applications of analytic and geometric methods to nonlinear differential equations", Exeter, UK, NATO ASI Series C 413, (ed. RA.Clarkson), Kluwer Publ, 241-249, 14-19 July, 1992. Tsarev, S.R, On Poisson brackets and one-dimensional Hamiltonian systems of hydrodynamic type, Dokl. Akad. NaukSSSR 282, 534-537 (1985); {Sov. Math. Dokl. 31, 488-491 (1985)).
INTEGRABILITY OF EQUATIONS OF HYDRODYNAMIC TYPE 265 Tsarev, S.P., On the integrability of the averaged KdV and Benney equations, Proc. NATO ARW "Singular Limits and Dispersive Waves", Lyon, France, 8-12 July, 1991. Ercolani, N.M. (ed.) et al.. New York, NY: Plenum, NATO ASI Ser., Ser. B, Phys. 320, 143-155 (1994). Tsarev, S.P, The geometry of Hamiltonian systems of hydrodynamic type. The generalised hodograph method. Math, USSR Izv. 37, 397-419 (1991). Tzitzeica, G., Sur une nouvelle classe de surfaces, C.R, Acad. Sci. Paris 150, 955-956, 1227-1229 (1910). Tzitzeica, G., Geometrie differentielle projective des reseaux, Paris-Bucarest, 1924. Verosky, J., Higher-order symmetries of the compressible one-dimensional isentropic fluid equations, J, Math. Phys. 25, 884 (1984). Whitham, G.B., Linear and nonlinear waves, N.Y, John Wiley (1974). Zakharov, V.E., Benney equations and the quasiclassical approximation in the inverse spectral problem method. Funk. Anal 14 (1980). Zakharov, V.E., Description of the n-Orthogonal Curvilinear Coordinate Systems and Hamiltonian Integrable Systems of Hydrodynamic lype. Part 1. Integration of the Lame Equations. Duke Math. J. 94, 103-139 (1998).
List of Participants ALBER, Mark University of Notre Dame, Mathematics Department, Room 370, CCMB, Notre Dame, Indiana 46556-5683, USA Email: mark.s.alber. 1 @nd.edu ATHORNE, Chris Glasgow University, Department of Mathematics, 15 University Gardens, Glasgow, G12 8QW, UK Email: c.athome@maths.gla.ac.uk BEGGS, Edwin J. University of Wales Swansea, Department of Mathematics, Singleton Park, Swansea, SA2 8PP, UK Email: e.j.beggs@swansea.ac.uk BELOKOLOS, Eugene Institute of Magnetism, Vemadsky Str. 36, Kiev-142, 252142, Kiev, Ukraine Email: bel@im.imag.kiev.ua BIELAWSKI, Roger The University of Edinburgh, Dept. of Mathematics and Statistics, The James ClerkMaxwell Building, King's Buildings, Mayfield Road, Edinburgh EH9 3JZ, UK Fax: 0131 650 6553 BOALCH, Phillip Oxford University, Mathematical Institute, 24-29 St Giles, Oxford, OXl 3LB, UK Email: boalch@maths.ox.ac.uk 267
268 LIST OF PARTICIPANTS BRADEN, Harry University of Edinburgh, Department of Mathematics, James Clerk Maxwell Building, Mayfield Road, Edinburgh, EH9 3JZ, UK Email: hwb@maths.ed.ac.uk BUCHSTABER, Victor M. Steklov Mathematical Institute, Dept. of Mathematics and Mechanics, Moscow State University, Moscow 119899, Russia Email: buchstbe@nw.math.msu BYATT-SMITH, John The University of Edinburgh, Department of Mathematics, James Clerk Maxwell Building, King's Buildings, Mayfield Road, Edinburgh EH9 3JZ, UK Email: byatt@maths.ed.ac.uk CARROLL, Robert Univ. of Illinois at Urbana-Champaign, Mathematics Department, 1409 W Green Street, Urbana,IL 61801, USA Email: rcarroll@math.uiuc.edu CHALYKH, Oleg Loughborough University, Loughborough, Leicestershire LEll 3TU, UK and MSU, Russia Email: 0.Chalykh@lboro.ac.uk CORRIGAN, Edward E University of Durham, Dept. of Mathematical Sciences, Science Laboratories, South Road, Durham DHl 3LE, UK Email: edward.corrigan@durham.ac.uk DUBROVIN, Boris SISSA, Via Beirut, 2-4,1-34013, Trieste, Italy Email: dubrovin@sissa.it DUNAJSKI, Maciej Oxford University, Mathematics Institute, 24-29 St Giles, Oxford OXl 3LB, UK Email: dunajski@maths.ox.ac.uk EDELSTEIN, Jose D. University of Santiago de Compostela, Department of Particle Physics, Faculty of Physics, E-15706, Santiago de Compostela, Spain Email: edels@fpaxpl.usc.es
LIST OF PARTICIPANTS 269 EGUCHI, Tohru The University of Tokyo, Department of Physics, Faculty of Science, Bunkyo-ku, Tokyo, Japan 113 Email: eguchi@ siren.phys.s.u-tokyo.ac.jp FAIRLIE, David B. University of Durham, Dept. of Mathematical Sciences, Science Laboratories, South Road, Durham DHl 3LE, UK Email: david.fairlie@durham.ac.uk FELDER, Giovanni ETH Zentrum, Dept. of Mathematics, CH-8092 Zurich, Switzerland Email: felder@math.ethz.ch FLUME, Rainald Universitat Bonn, Mathematics Institute, Rheinische Friedrich-Wilhelms, D-53115 Bonn, Germany Email: unp06d@ibm.rhz.uni.bonn.de FORDY, Alan R University of Leeds, Dept. of Applied Mathematics, Leeds, LS2 9JT, UK Email: allan@amsta.leeds.ac.uk FREEMAN, Mike King's College London, Department of Mathematics, Strand , London, WC2R 2LS, UK Email: mfreeman@ mth.kcl.ac.uk GANZHA, Elena Krasnoyask State Pedagogical University, Maths. Dept., Lebedevoi, 89, Krasnoyarsk 660089, Russia Email: ganzha@edk.kraznoyarsk.su GIBBONS, John Imperial College, Dept. of Mathematics, Huxley Building, 180 Queens Gate, SW7 2BZ, UK Email: j.gibbons@ic.ac.uk GILSON, Claire University of Glasgow, Department of Mathematics, Glasgow, G12 8QW, UK Email: claire@maths.gla.ac.uk GRANT, James D.E. University of Hull, Department of Mathematics, Cottingham Road, Hull, HU6 7RX, UK Email: j.d.grant@maths.hull.ac.uk
270 LIST OF PARTICIPANTS HARNAD.John Concordia University, Department of Mathematics, 7141 Sherbrooke St. W., Montreal, QC, Canada H4B 1R6 Email: hamad@crm.umontreal.ca HASSNER, Martin IBM Research Division, Almaden Research Division, Almaden Research Centre, 650 Harry Road, San Jose, CA 95120-6099, USA Email: hassner@almaden.ibm.com HAWKSLEY, Ruth The University of Edinburgh, Dept. of Mathematics, James Clerk Maxwell Building, King's Buildings, Mayfield Road, EH9 3JZ, UK Email: ruthh@maths.ed.ac.uk HITCHIN, Nigel Oxford University, Department of Mathematics, 24-29 St Giles, Oxford, OXl 3LB, UK Email: hitchin@maths.ox.ac.uk HOUGHTON, Conor J. University of Cambridge, Dept. of Applied Maths & Th. Physics, Silver Street, Cambridge, CBS 9EW, UK Email: c.j.houghton@damtp.cam.ac.uk JOHNSON, Peter Max-Plank Institut fiir Gravitationsphysik, Albert-Einstein-Institut, Schlaatzwegi 1„ D- 14473 Potsdam, Germany Email: johnson@aei-potsdam.mpg.de JOHNSTON, Des Heriot-Watt University, Dept. of Mathematics, Riccarton Campus, Edinburgh, EH14 4AS, UK Email: des@ma.hw.ac.uk KODAMA,Yuji Ohio State University, Department of Mathematics, Ohio State University, 231 West, 18th Avenue, Columbus, OH 43210, USA Email: kodama@math.ohio-state.edu ROSTOV, Nikolay Bulgarian Academy of Science, Institute of Electronics, Boul. Trakia 72, BG-Sofia, 1784 Bulgaria Email: n_a_kostov@yahoo.com
LIST OF PARTICIPANTS 271 KMCHEVER, Igor Columbia University, Department of Mathematics, 2990 Broadway, mcode 4406, New York, NY 10027, USA Email: igor@landau.ac.ru LEYKIN, Dmitry V. Institute of Magnetism NASU, Vemadsky Str. 36, Kiev-142, 252142, Kiev, Ukraine Email: dile@d24.imp.kiev.ua MANTON, Nick S. University of Cambridge, Department of Applied Mathematics, Silver Street, Cambridge, CBS 9EW, UK Email: n.s.manton@damtp.cam.ac.uk MARKMAN, Eyal University of Massachusetts, Department of Maths and & Statistics, LGRT, Box 34515, Amherst, MA 01003-4515, USA Email: markman@math.umass.edu MARSHAKOV, Andrei FIAN, Leninsky pr., 53, 117924, Moscow, Russia Email: mars@td.lpi.ac.ru MASON, Lionel St Peter's College, Oxford, 0X1 2DL, UK Email: lmason@maths.ox.ac.uk MCKAY, John Concordia University, Dept. of Maths and Computer Sci., Sir George Williams Campus, 1455 De Maisonneuve Blvd. West, LB 903 15, Montreal, QC Canada H3G IM* Email: mckay@cs.concordia.ca MIKHAILOV, Sasha University of Leeds, Dept. of Applied Mathematical Studies, Leeds, LS2 9JT, UK Email: sashamik@lpm.univ-montp2.fr MIRAMONTES, J. Luis University of Santiago, Dept. De Fisica De Particulas, Facultad De Fisca, Santiago De Compostela, E-15706, Spain Email: niiramont@fpaxpl.usc.es MIRONOV, Andrei ITEP, B. Cheremushinskaya, 25-117259, Moscow, Russia Email: niironov@heron.itep.ru
272 LIST OF PARTICIPANTS MOKHOV, Oleg I. Centre for Nonlinear Studies, Landau Institute for Theoretical Physics, Kosygina 2, Moscow, GSP-1, 117940, Russia Email: mokhov@mi.ras.ru MOROZOV, Alexei Inst, of Theoretical and Experi. Physics, B. Cheremushinskaya 25,117259, Moscow, Russia Email: morozov@vitep5.itep.ru NIJHOFF, Frank W. University of Leeds, Department of Applied Mathematical Studies, Leeds, LS2 9JT, UK Email: frank@amsta.leeds.ac.uk NIMMO, John C. University of Glasgow, Department of Mathematics, 15 University Gardens, Glasgow, G12 8QW, UK Email: j.nimmo@maths.gla.ac.uk OLIVE, David University of Wales Swansea, Department of Mathematics, Singleton Park, Swansea, SA2 8PP, UK Email: d.I.olive@swansea.ac.uk OLSHANETSKY, Misha ITEP, B. Cheremushinskaya 25, 117259, Moscow, Russia Email: olshanet@heron.itep.ru PHONG, Duong H. Columbia University, Department of Mathematics, 2990 Broadway, 4406, New York, NY 10027, USA Email: phong@math.columbia.edu RAMANAN, S. TATA Institute, School of Mathematics, Tata Institute of Fundamental Research, Colaba, Mumbai 400 005, India Email: ramanan@math.res.in ROBBERS, Jonathan BRIMS, Hewlett-Packard Laboratories, Filton Road, Stoke Gifford, BS12 6QZ, UK Email: j.robbins@bristol.ac.uk
LIST OF PARTICIPANTS 273 SANCHEZ-GUILLEN, Joaquin University of Santiago, Dept. De Fisica De Particulas, Facultad De Fisca, Santiago De Compostela, E-15706, Spain Email: joaquin@gaes.usc.es SASAKI, Ryu Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto, 606-8502 Japan Email: ryu@yukawa.kyoto-u.ac.jp SCHROERS, Bemd University of Amsterdam, ITEP, Valckenier Straat 65, 1018 XE Amsterdam, The Netherlands Email: schroers @phys.uva.nl SINGER, Michael The University of Edinburgh, Dept. of Mathematics and Statistics, The James ClerkMaxwell Building, King's Buildings, Mayfield Road, Edinburgh EH9 3JZ, UK Email: michael@maths.ed.ac.uk STRACHAN,IanA.B. University of Hull, Department of Applied Mathematics, Cottingham Road , Hull, HU6 7RX,UK Email: i.a.strachan@maths.hull.ac.uk SUTCLIFFE, Paul M. University of Kent, Institute of Mathematics, Canterbury, Kent, CT2 7NF, UK Email: p.m.sutcliffe@ukc.ac.uk TAKASAKI, Kanehisa Kyoto University, Department of Fundamental Sciences, Sakyo, Kyoto, 606-8501 Japan Email: takasaki@yukawa.kyoto-u.ac.jp TSAREV, Serguei Krasnoyarsk State Pedagogical University, Maths. Dept., Lebedevoi, 89, Krasnoyarsk, 660089 Russia Email: tsarev@edk.krasnoyarsk.su VAN DE LEUR, Johan University of Twente, Faculty of Mathematical Sciences, PO Box 217,7500 AE Enschede, The Netherlands Email: vdleur@math.utwente.nl VANINSKY, Kirill Kansas State University, Department of Mathematics, Manhatten, Kansas, 66506 USA Email: vanisky@math.ksu.edu
274 LIST OF PARTICIPANTS VARELA, Victor Edinburgh University, Dept. of Mathematics and Statistics, James Clerk Maxwell Building, King's Buildings, Mayfield Road, Edinburgh EH9 3JZ, UK Email: varela@maths.ed.ac.uk VESELOV, Sasha Loughborough University, Department of Mathematical Sciences, Loughborough, Leicestershire, LE11 3TU, UK Email: a.p.veselov@lboro.ac.uk
Index algebraically completely integrable 23, 24, 27, associativity 105, 110, 117, 121, 176, 182 B Baker-Akhiezer function (BA) 5, 6, 8, 78, 112, 167, 204, 214 billiards 219, 225 billiard solutions 214, 226 BPS 75, 82 ; Burchnall and Chaundy 4 Calogero-Moser 24, 29, 45, 46, 52, 54, 56, 57, 65, 70, 72, 76, 81, 99, 114, 118, 126, 177 Camassa-Holm (SW) 214, 215 cameral cover 34, 35 Chem-Simons 95, 98 Combescure 257, 258 D diagonalizable 233, 255, 257, 258, 261 dispersionless 16, 206 dispersionless KP 199, 202, 204 E Egorov property 258-260 Fano varieties 176, 180, 187 Frobenius manifold 125, 150, 199, 201, 202, 207, 209-211 T function 14, 15, 17, 19, 95, 96, 114, 211 G Gaudin system 32, 39 geometric asymptotics 214, 217 Gibbons-Tsarev 238, 239, 249 Gromov-Witten invariants 16, 105, 175, 176, 177 H Hamilton-Jacobi 204, 205, 219, 236, 237 Hauptmodul 142, 144, 150 Harry Dym equation (HD) 214, 215, 223 Hitchin map 35, system 24, 29, 30, 33, 38, 45, 81, 154, 163, 169, 171 hodograph 202, 209, 254, 257 holomorphicity 44, 47 hydrodynamic 199, 200, 207, 232, 234, 253, 255, 261 275
276 INDEX isomonodromic deformations 158, 162, 172 Ruijsenaars-Schneider 72, 76, 83, 84, 114 Jacobi 214, 224 Jacobian 9, 19, 20, 33, 34, 36, 166, 218, 224 K Killing vector 234, 237, 241, 245, 248 Knizhnik-Zamolodchikov-Bemard equations (KZB) 168 linearly degenerate 233-235, 249 M Maslov indices 220, 223 matrix models 95 moduli 51, 61, 169 moduli space 11, 13, 17, 30-33, 37, 38, 98, 104, 112, 156, 159, 160, 163, 192 monodromy 220, 223 N Neumann 214 nondiagonalizable 260 Painleve VI equations 140, 153 partition function 16, 176, 182 peakon 227 Picard-Fuchs equation 139, 140, 144, 146, 149, 155 prepotential (^) 19, 25, 45, 50-52, 62-65, 69, 71, 74, 88, 99, 101, 106, 108, 109, 127, 168 Prym 34, 37, 39, 70 Seiberg-Witten curves 45, 51, 52, 61, 65 differential 28, 33, 52, 61, 65, 168 semi-Hamiltonian 234, 235, 249, 257, 260-262 special geometry 17, 49 special Kahler 23, 25-29, 33, 38, 69 spectral cover 33, 38, 45, 56 spectral curve 2, 8, 18, 36, 45, 56, 62, 66, 74, 75, 77, 82, 100, 113, 165 spin chains 86, 88, 114, 115, 118 string equation 14, 202 supersymmetry 47, 48-50, 177 Shallow Water equation (SW) 223 tau-function (t) 14, 15, 94, 96-98, 162, 204, 211 Toda systems 46, 54-56, 70, 72, 73, 76, 77, 83, 99, 101, 112, 114, 206 topological field theory (TFT) 12, 15, 16, 96, 104, 105, 138, 175, 193, 194, 199, 200, 201, 205, 208 topological recursion relations (TRR) 177, 181, 182, 191, 192, 202, 208, 209 topological Sigma model 98, 175, 178 two-dimensional gravity (2d gravity) 14, 98, 184 V Virasoro condition 177, 182, 184, 186, 188, 192, 194 conjecture 177, 194 operators 182, 186, 188, 190, 191 v-conditions 129 quantum cohomologies 104, 105, 176, 182, 194 reciprocal transformation 234, 237 Riemann invariants 255 W WDW equations 16, 17, 101, 104, 107- 109, 116, 119, 121, 125, 130, 132, 133, 138 Whitham equations (hierarchy) 9-13, 114, 162, 169, 231, 254, 258, 262 times 14-17, 98, 100