Integrable Systems in The Realm OfAlgebraic Geometry

Lecture Notes in Mathematics 1638
Editors:
J.-M. Morel, Cachan
F. Takens, Groningen
B. Teissier, Paris
Springer
Berlin
Heidelberg
New York
Barcelona
Hong Kong
London
Milan
Paris
Singapore
Tokyo
Pol Vanhaecke
Integrable Systems
in the realm
of Algebraic Geometry
Second Edition
Y,Vkl
"841
Springer
Author
Pol Vanhaecke
D6parternent de Math6matiques
UFR Sciences SP2MI
Universit6 de Poitiers
T616port 2
Boulevard Marie et Pierre Curie
BP 30179
86962 Futuroscope Chasseneuil Cedex, France
E-mail: Pol.Vanhaecke@mathlabo.univ-poitiers.fr
Cataloging-in-Publication Data applied for
Die Deutsche Bibliothek -

CIP-Einheitsaufnahme
Vanhaecke, Pol:
Integrable systems in the realm of algebraic
geometry / Pol Vanhaecke. 2. -
ed.. Berlin ; Heidelberg New York ; Barcelona

-
; Hong Kong ; London

Milan ; Paris ; Singapore Tokyo : Springer, 2001
(Lecture notes in mathematics ; 1638)
ISBN 3-540-42337-0
Mathematics Subject Classification (2000): 14K20, 14H70, 17B63, 37J35
ISSN 0075- 8434

ISBN 3-540-42337-0 Springer-Verlag Berlin Heidelberg New York
ISBN 3-540-61886-4 (Ist edition) Springer-Verlag Berlin Heidelberg New York
This work is subject to copyright. All rights are reserved, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, re-use
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other
way, and storage in data banks. Duplication of this publication or parts thereof is
permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from
Springer-Verlag. Violations are liable for prosecution under the German Copyright
Law.
Springer-Verlag Berlin Heidelberg New York

a member of BertelsmannSpringer Science+Business Media GmbH
http:Hwww.springer.de
0 Springer-Verlag Berlin Heidelberg 1996, 2001
Printed in Germany
The of
general descriptive names, registered names, trademarks, etc. in this
use
publication does not

imply, even in the absence of a specific statement, that such
names are exempt from the relevant protective laws and regulations and therefore free
for general use.
Typesetting: Camera-ready TEX output by the author

SPIN:10844943 41/3142/LK -
543210 Printed on acid-free paper
Preface to the second edition
The present edition of this book, five years after the first edition, has been spiced with
naturally in the point of view that had been adapted in the
several recent results which fit
original text and with some new examples and constructions that will help the reader to
appreciate better our approach to integrable systems.
On this occasion I wish to thank my collaborators from the last five years, to wit Christina
Birkenhake, Peter Bueken, Rui Fernandes, Masoto Kimura, Vadim Kuznetsov, Marco Pe-
droni, Michael Penkava, Luis Piovan and Claude Roger for a fruitful interaction and for their
warm friendship. Most of the results that have been added axe taken from, or are inspired
by, joint work with some of them; I acknowledge their permission to add these, sometimes
unpublished, results.
The colleagues at my newest working environment, the University of Poitiers (aance),
created for me a pleasant and stimulating working enviromnent. I wish to acknowledge the
support of all of them. Special thanks go to Marc van Leeuwen, Claude Quitt6 and Patrice
Tauvel for sharing their insights with me, which usually led to a real improvement of parts
of the text.
Last but not least, Yvette Kosmann-Schwambach, who was not acknowledged in the
first version of this book -
most probably because my gratitude to her was too big and too
obvious! -
is thanked here in all possible superlatives, for her constant support and for her
sincere friendship. Merci Yvette!
Acknowledgments
The help of many people indispensable for establishing and presenting the results
was
which are contained in this work.Not enough credit can be given to those who created
at home, at the Max-Planck-Institut in Bonn, at the University of Lille and finally at the
University of California at Davis a pleasant and stimulating atmosphere. Even some people
I don't know by name should be thanked here.
Special thanks are due to Mark Adler and Pierre van Moerbeke, whose fundamental work
on a.c.i. systems was the starting point for the research contained in this book. Stimulating
discussions with them have led to an improvement of many of the results and to a better
understanding of the subject. Also Michble Audin deserves a special plarce here for sharing
her insights with me through long discussions and letters. Extremely helpful for a thorough
understanding were several algebraic-geometric explanations by Laurent Gruson.
I wish to thank my collaborators Jos6 Bertin and Marco Pedroni for a fruitful interac-
tion. I have also benefited from discussions with my colleagues at Lille, in particular Jean
d'Almeida, Robert Gergondey, Johannes Huebschmann, Rapha6l Freitas, Armando Treibich,
Gijs Taymnan and Alberto Verjowski and at UC Davis, in particular Josef Mattes, Motohico,
Mulase, Michael Penkava, Albert Schwarz and Craig Tracy.
I also acknowledge my other friends scattered around the globe, to wit, Christina Birken-
hake, Robert Brouzet, Peter Bueken, Jan Denef, Paul Dhooghe, Jean Fastr6, Ljubomir
Gavrilov, Luc Haine, Horst Knbrrer, Franco Magri, Askold Perelomov, Luis Piovan, Elisa
Prato and Taka Shiota for their interest in my work and helpful related discussions.
For useful comments on the manuscript I am indebted to Mich6le Audin, an anonymous

referee and several students in my graduate course in UC Davis (Spring 1996).
Last but not least, special thanks to my wife Lieve for her constant assistance through
this adventure.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.
IT. Integrable Hamiltonian systems on affine Poisson varieties . . . . . .

17.
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17.
2. Affine Poisson varieties and their morphisms . . . . . . . . . . . . . . . .

19.
2.1. Affine Poisson varieties . . . . . . . . . . . . . . . . . . . . . . .
19.
2.2. Morphisms of affine Poisson varieties . . . . . . . . . . . . . . . . .

26.
2.3. Constructions of affine Poisson varieties . . . . . . . . . . . . . . . .
28.
2.4. Decompositions and invariants of affine Poisson varieties . . . . . . . . .
37.
3. Integrable Hamiltonian systems and their morphisms . . . . . . . . . . . .

47.
3.1. Integrable Hamiltonian systems on affine Poisson varieties . . . . . . . .

47.
3.2. Morphisms of integrable Hamiltonian systems . . . . . . . . . . . . .
54.
3.3. Constructions of integrable Hamiltonian systems . . . . . . . . . . . .

57.
3.4. Compatible and multi-Hamiltonian integrable systems . . . . . . . . . .
62.
4. Integrable Hamiltonian systems on other spaces . . . . . . . . . . . . . . .

65.
4.1. Poisson spaces . . . . . . . . . . . . . . . . . . . . . . . . . . .

65.
4.2. Integrable Hamiltonian systems on Poisson spaces . . . . . . . . . . . .
69.
111. Integrable Hamiltonian systems and symmetric products of curves .

71.
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
71.
2. The systems and their integrability . . . . . . . . . . . . . . . . . . . .

73.
2. 1. Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73.
2.2. The compatible Poisson structures -jwd 73.
2.3. PolynomiaJs in involution for I., -1d' 78.
2.4. The hyperelliptic case . . . . . . . . . . . . . . . . . . . . . . . .

83.
3. The geometry of the level manifolds . . . . . . . . . . . . . . . . . . . .

85.
3.1. The real and complex level sets . . . . . . . . . . . . . . . . . . . .

85.
3.2. The structure of the complex level manifolds . . . . . . . . . . . . . .

87.
3.3. The structure of the real level manifolds . . . . . . . . . . . . . . . .

89.
3.4. Compactification of the complex level manifolds . . . . . . . . . . . .

93.
3.5. The significance of the Poisson structures -j'Pd . .

95.
viii
IV. Interludium: the geometry of Abelian varieties . . . . . . . . . .
97.
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
97.
2. Divisors and line bundles . . . . . . . . . . . . . . . . . . . . . . . .

99.
2.1. Divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
99.
2.2. Line bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . .

100.
2.3. Sections of line bundles . . . . . . . . . . . . . . . . . . . . . . .

101.
2.4. The Riemann-Roch Theorem . . . . . . . . . . . . . . . . . . . . .

103.
2.5. Line bundles and embeddings in projective space . . . . . . . . . . . .

105.
2.6. Hyperelliptic curves . . . . . . . . . . . . . . . . . . . . . . . . .

106.
3. Abelian varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . .
108.
3.1. Complex tori and Abelian varieties . . . . . . . . . . . . . . . . . .

108.
3.2. Line bundles on Abelian varieties . . . . . . . . . . . . . . . . . . .

109.
3.3. Abelian surfaces . . . . . . . . . . . . . . . . . . . . . . . . . .

111.
4. Jacobi varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
114.
4.1. The algebraic Jacobian . . . . . . . . . . . . . . . . . . . . . . .

114.
4.2. The analytic/transcendental Jacobian . . . . . . . . . . . . . . . . .

114.
4.3. Abel's Theorem and Jacobi inversion . . . . . . . . . . . . . . . . .

119.
4.4. Jacobi and Kummer surfaces . . . . . . . . . . . . . . . . . . . . .

121.
5. Abelian surfaces of type (1,4) . . . . . . . . . . . . . . . . . . . . . . .

123.
5.1. The generic case . . . . . . . . . . . . . . . . . . . . . . . . . .

123.
5.2. The non-generic case . . . . . . . . . . . . . . . . . . . . . . . .

124.
V. Algebraic completely integrable Hamiltonian systems . . . . . . . .

127.
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
127.
2. A.c.i. systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
129.
3. Painlev6 analysis for a.c.i. systems . . . . . . . . . . . . . . . . . . . .

135.
4. The linearization of two-dimensional a.c.i. systems . . . . . . . . . . . . .

138.
5. Lax equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
140.
VI. The Mumford systems . . . . . . . . . . . . . . . . . . . . .

143.
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
143.
2. Genesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
145.
2.1. The algebra of pseudo-differential operators . . . . . . . . . . . . . .

145.
2.2. The matrix associated to two commuting operators . . . . . . . . . . .

146.
2.3. The inverse construction . . . . . . . . . . . . . . . . . . . . . . .

150.
2.4. The KP vector fields . . . . . . . . . . . . . . . . . . . . . . . .

152.
ix
3. Multi-Hamiltonian structure and symmetries . . . . . . . . . . 155.
. . . . . .
3.1. The loop algebra 91, . . . . . . . . . . . . . . . . . . . . .

155.
. . .
3.2. Pteducing the R-brackets and the vector field V . . . . . . . . . . . 157.

. .
4. The odd and the even Mumford systems . . . . . . . . . . . . . . .

161.
. . .
4.1. The (odd) Mumford system . . . . . . . . . . . . . . . . . . . . .

161.
4.2. The even Mumford system . . . . . . . . . . . . . . . . . . 163.
. . . .
4.3. Algebraic complete integrability and Laurent solutions . . . . . . . . . .

164.
5. The general case . . . . . . . . . . . . . . . . . . . . . . . . . . . .
168.
VII. Two-dimensional a.c.i. systems and applications . . . . . . . 175.

. .
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
175.
.
2. The genus two Mumford. systems . . . . . . . . . . . . . . . 177.

. . . . . .
2.1. The genus two odd Mumford, system . . . . . . . . . . . . . . .

177.
. .
2.2. The genus two even Mumford system . . . . . . . . . . . . . . . . .

179.
2.3. The Bechlivanidis-van Moerbeke system . . . . . . . . . . . . . . . .
181.
3. Application: generalized Kummer surfaces . . . . . . . . . . . . . . . .
185.
.
3.1. Genus two curves with an automorphism of order three . . . . . . . . .

185.
3.2. The 94 configuration on the Jacobian of r . . . . . . . . . . . . . . .
186.
3.3. A projective embedding of the generalised Kummer surface . . . . . . . .
190.
4. The Gaxnier potential . . . . . . . . . . . . . . . . . . . . . . . .
196.
. .
4.1. The Garnier potential and its integrability . . . . . . . . . . . . . . .

196.
4.2. Some moduli spaces of Abelian surfaces of type (1,4) . . . . . . . . . .
202.
4.3. The precise relation with the canonical Jacobian . . . . . . . . . . . .
206.
4.4. The relation with the canonical Jacobian made explicit . . . . . .
211.
. . .
4.5. The central Garnier potentials . . . . . . . . . . . . . . . . . . . .

216.
5. An integrable geodesic flow on SO(4) . . . . . . . . . . . . . . . . . .
220.
.
5.1. The geodesic flow on SO(4) for metric II . . . . . . . . . . . . . .

220.
. .
5.2. Linearizing variables . . . . . . . . . . . . . . . . . . . . .

222.
. . . .
5.3. The map M -+ M . . . . . . . . . . . . . . . . . . . . . . 226.

. . . .
6. The H6non-Heiles hierarchy . . . . . . . . . . . . . . . . . . . 230.

. . . .
6.1. The cubic H6non-Heiles potential . . . . . . . . . . . . . . . 230.

. . . .
6.2. The quartic H6non-Heiles potential . . . . . . . . . . . . . . 232.

. . . .
6.3. The H6non-Heiles hierarchy . . . . . . . . . . . . . . . . . 233.

. . . .
7. The Toda lattice . . . . . . . . . . . . . . . . . . . . . . 235.

. . . . . .
7.1. Different forms of the Toda lattice . . . . . . . . . . . . . . 235.

. . . .
7.2. A morphism, to the genus 2 even Mumford system . . . . . . . . 237.

. . . .
7.3. Toda and Abelian surfaces of type (1,3) . . . . . . . . . . . 240.

. . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.
Index . . . . . . . . . . . . . . . . . . . . . . . . . . 253.
. . . .
x
Chapter II
Integrable Hamiltonian systems

on affine Poisson varieties
1. Introduction
In this chapter give the basic definitions and properties of integrable Hamiltonian
we
systems on affine Poisson varieties and their

morphisms. In Section 2 we define the notion of
a Poisson bracket (or Poisson structure) on an affine algebraic variety. The Poisson bracket is
precisely what is needed to define Hamiltonian mechanics on a space, as is well-known from

the theory of symplectic and Poisson manifolds. We shortly describe the simplest Poisson
structures (i.e., constant, linear, affine and quadratic Poisson structures; also general Poisson
structures on C2 and C') and describe two natural decompositions of affine Poisson varieties,
one is given by the algebra of Casimirs, the other comes from the notion of rank of a Poisson
structure (at a point). We also describe several ways to build new affine Poisson varieties
from old ones.
Morphisms of affine Poisson variety are regular maps which preserve the Poisson bracket.
Isomorphisms preserve the rank at each point, leading to a polynomial invariant for affine
Poisson varieties. This invariant permits us on the one hand to distinguish
many different
affine Poisson varieties, on the other hand it allows us to display in a structured
way the basic
characteristics of the Poisson structure. It will be computed for many different examples and
a refinement of this invariant is also discussed.
In Section 3
we turn to integrable Hamiltonian systems. We motivate our definition
by
severalpropositions and (counter-) examples. The notions of super-integrability, compatibility
and integrable multi-Hamiltonian systems fit very well into the picture and most of our
propositions are easily adapted to the case that the integrable Hamiltonian systems under
discussion have one of these extra structures. The notion of momentum map leads to a
decomposition of the variety, as the one given by the Casimirs (however

it is much finer).
17
P. Vanhaecke: LNM 1638, pp. 17 - 70, 1996, 2001

© Springer-Verlag Berlin Heidelberg 1996, 2001
Chapter 11. Integrable Hamiltonian systems
We also define morphisms of integrable Hamiltonian systems; they are Poisson mor-
phisms which preserve thealgebra of functions in involution. It allows one to state precisely
the relation between different integrable Hamiltonian systems, for example between new sys-
tems and the old ones from which they were constructed. Our discussion is parallel to the
one of affine Poisson varieties (up to some modifications). Some really interesting examples
of integrable Hamiltonian systems will be given in later chapters.
The final section (Section 4) is devoted to a generalization of our definitions to the

caseof other spaces. We draw special attention to the case of real Poisson manifolds. The
main difference is that on the one hand the algebras we work with in the case of an affine
Poisson variety are in general not finitely generated so that many constructions do not apply
(e.g., polynomial invariant), on the other hand many local constructions (e.g., Darboux
the
coordinates, action-angle variables) which cannot be performed for affine Poisson varieties,
play a dominant role in the study of some other Poisson spaces, including Poisson manifolds.
Apart from Section 4 we will in this chapter always work over the field of complex
numbers.
18
2. Affine Poisson varieties and their morphisms
2.1. Affine Poisson varieties
Phase space will always be an affine vaxiety in the sense of [Har], i.e., an irreducible
closed subset of C' (closed for the Zariski topology). Such a variety M C CI is the zero
locus of prime ideal Im of C[xi.... Xn], and its ring (or C-algebra) of regular functions
a 1
is
denoted, resp. defined by

C[Xi'...' Xn]
O(M) =
IM
O(M) integral domain (it has no zero divisors) and it is finitely generated; M can be
is an
reconstructed, up to isomorphism, from O(M) as SpecmO(M), the set of closed points in

SpecO(M).
The extra structure which we use to describe Hamiltonian systems on M is given by a
Poisson bracket on its algebra of regular fanctions.
Definition 2.1 Let M be an affine variety. A Poisson bracket or Poisson structure on M

is a Lie algebra structure I-, j on O(M), which is a bi-derivation, i.e., for any f G O(M) the
C-linear map
Xf:O(M)-+O(M)
g -+Ig,fl
is a derivation (satisfies the Leibniz rule),
Xf (gh) =
(Xf g) h + gXf h (2.1)
for all g, h E O(M). The derivation Xf is called the Hamiltonian derivation associated to
the Hamiltonian f and we write Ham (M, f -, -1) for the (vector) space
Ixf =
I., f I I f E OMI
of Hamiltonian derivations. A function f E O(M) whose Hamiltonian vector field is zero,

0, is called Casimir function Casimir and we denote
Xf = a or a
Cas (M, 1., -1) =

If G O(M) I Xf =
01
for the(vector) space of Casimirs; it is the center of the Lie bracket I-, j hence it is a Lie
ideal of(O(M), I-, J). When no confusion can arise, either argument in Ham (M, I-, J) and
Cas (M, f J) is omitted.
-
Remarks 2.2
1. Xf being a derivation may be refrased in a geometric way by saying that it is a global
section of the tangent sheaf to M, i.e.,
TM, HO(M, Tm) (for the definition of the sheaf
Xf E
of differentials and the tangent sheaf to an algebraic variety see [Hax] Section 11.8). For this
reason we usually call the elements Xf of Ham (M, 1., -1) Hamiltonian vector fields.
Using the above mentionned correspondence between an affine variety and its algebra
2.
of regular functions we have that affine Poisson varieties correspond to finitely generated
Poisson algebras without zero divisors.
19
Chapter Il. Integrable Hamiltonian systems
3. Turning the above definition upside down one gets at the following, equivalent defini-
n
tion of a Poisson bracket. Let denote the vector space Hom(A 0 (M), 0 (M)) by C' (M)
us
and its subspace of skew-symmetric n-derivations by Der' (M). For every p, q > 0 a bilinear
map
F -1
,
: CP M X C, (M) -+ CP+"- I (M)
is defined for P E CP(M), Q E Cq(M) and for fi,..., fp+,-i E O(M) by

[Pj Q] (fl I... I fp+q-1)
o,ESq,p-i
+ 1: ...
i fa(p+q-1))
aESp,q_i
where Sp, denotes the set of (p,q) shuffles (permutations a of 11,...'p + qj such that
o-(1) < ...
< a(p) and a(p + 1) < < a(p + q); c(a) is the sign of a). It is easy to see
...
that if P E DerP (M) and Q E Derq (M) then [P, Q] E Derp+q-1 (M). Thus restricts to a
bracket
-Is : DerP (M) x Derq (M) -+ DerP+q- 1 (M),
2
called the Schouten bracket. For P E Der (M) we have that
[P, P]s(f, g, h) =
2(P(P(f, g), h) + P(P(g, h), f) + P(P(h, f), g)),
so that Poisson structures can skew-symmetric bi-derivations P such that
also be defined as
[P, PIS = 0. This point of view leads following interesting interpretations. If P G

to the
2
Der (M) defines a Poisson structure then the (graded) Jacobi identity for [-, -Is implies that
Der* (M) becomes a complex when the coboundary operator
8p : Derq (M) -+ Derq+1 (M)
is defined for Q E Derq (M) by 8p (Q) corresponding cohomology is called

=
[P, Q] s. The
Poisson cohomology. One observes that precisely the 0-cocycles and that the
the Casimirs are
Hamiltonian vector fields are the 1-coboundaries. For X E Deri (M), 8p X -,CxP, where =
LX is the Lie derivative of P with respect to X, hence the 1-cocycles are the vector fields
that preserve the Poisson structure P (such vector fields are called Poisson vector fields). A
similar interpretation of the 2-cocycles and the 2-coboundaxies will be given at the end of
this section.
The following properties follow at once from Definition 2.1.
Proposition 2.3 Let (M, 1-, -1) be an affine Poisson variety.

(3.) Ham(M) subalgebra of Der' (M) (with the commutator
is a Lie as Lie bracket);
Ham(M) is however in general not an O(M)-module, as opposed to Der(M);

(2) Cas(M) is a subalgebra of O(M);
(3) The adjoint map ad : O(M) -+ Ham(M) which is defined by f i-+ -Xf is a Lie
algebra homomorphism;
(4) For all f, g E O(M), ad(fg) f ad(g) + ad(f)g; equivalently, Xf,
=
f X, + gXf; =
(5) There is a short exact sequence of Lie algebras
ad
0 Cas(M) -,
O(M) __
-o-
Ham(M) 0
20
2. Afflne Poisson varieties and their morphisms
By (2.1) the Hamiltonian vector field Xf is completely determined by its action on
any system of generators gi, - - -

, g8 of 0 (M). It leads to a system of first order polynomial
differential equations on O(M), namely
i =
Xf gi =
Igi, f (i =
17 ... 14 (2.2)
where j is a convenient notation for Xf gi when a particular choice of f E O(M) has been
fixed. Similarly a Poisson structure I-, j is, in view of the Leibniz rule, completely described
in terms of the system of generators gj, . . .
98 by the Poisson matrix
,
9 =
Q9i19jD1<i,j<81
whose entries gij axe polynomials in gj, . . .
, g, - Morevoer, the Jacobi identity will be satisfied
for any three functions in O(M) as soon as it is satisfied for all triples of generators gi, 9j, gk
of O(M), with i < j < k.
Every vector field, in
particulax every Hamiltonian vector field, can be (non-uniquely) ex-
tended to vector field
the ambient affine space of M. Namely, suppose that M C C8, let
a on
P1,. .g8 be the standard coordinates of C8 and denote the corresponding generators of O(M)
.
,
by g.1, 98. The vector field determined by j

. . .
, Fj (gi, g,) then leads to a vector field =
. . .
,
on C8 determined by gi Fj (pl, p,). Conversely, a system 9i Fj (gi,

=
. . of first or-
.
=
, .
der polynomial differential equations on C8 defines a derivation on O(M) C[g.1, g.]Ilm =

- -
-,
precisely when j E Im for every f E Im C i.e., when the vector field is tan-
gent to M as a subvariety of C". Similarly, a Poisson bracket on O(M) defines a skew-
symmetric bi-derivation on C', but the latter is in general not a Poisson structure because
the Jacobi identity need not hold. Explicitly, let #jj -gji denote any extensions of gij =
to C[gj .... g,] (I < ij:5 k) anddefine askew-symmetric bi-derivationP of

C[gj,..-'9,] by
a a
P=I:gij -99i A
api
(2.3)
i<j
Then the Jar-obi identity for P is tantamount to the following system of partial differentia.1
equations.
a a
I 9jk + 9ji
2 g=gki + 9ki 1-9ij
I 'g
E Em, (1 < i < j < k < (2.4)
showing that the Poisson matrix does not define a Poisson bracket on C[gj, 08]
in general.
We will often define a Poisson bracket by specifying its Poisson matrix (in terms of a
system of generators); necessary and sufficient conditions for a matrix to define a Poisson
structure are given by the following proposition.
21
Proposition 2.4 Let M C[ 1,...' g,]/-E be an affine variety and let 91,
=
g, denote -..,
the cormsponding system of generators of O(M). Let g be a skew-symmetric .5 x s matrix,

with entries in O(M), and let p be any skew-symmetric matrix, whose entries are extensions
of the corresponding elements in g to Then g is the Poisson matrix (in terms
of gi, gj of a Poisson bracket on M if an only if
. . .
,
(i) the Jacobi identity (2.4) is satisfied for all triples (Pi, 9i 9k) (With i
i < i < k);
(2) for any G E Im C C[p,.... g.] I
one has
"
OG
E_ gjj E IM, U =
1,-, 8). (2.5)
Proof
We have seen(.t) is necessary; if G E Im then for any gj one has fgj, GI 0
above that =
(in O(M)), implies (2.5) upon using (2.3). Let us show that (i) and (2) are also
which
sufficient. Define a skew-symmetric bi-derivation I I' on C [pi, 9, ] by (2.3). It is not
-
,
-
- - -
,
necessarily a Poisson bracket since (i.) only guarantees that the Jacobi identity is satisfied
in 0 (M), i.e., its right hand side belongs to Im. Condition (2) implies that for all F1, F2 E
C[gj'...' #,] and for all G1, G2 Elm we have
IF, + G1, F2 + G2 11 =
IF,, F21' + G3
for some G3 E Im hence we have an induced bracket I -
,
-
I on 0 (M). It satisfies the Leibniz
rule because the original bracket does; because of condition (:L) it satisfies the Jacobi identity
for a system of generators hence for all regular functions. Hence it defines a Poisson bracket
on O(M). I
Remark 2.5 It is natural to ask if for every affine Poisson variety M one can find a system
of generators gl,...,g, of O(M) (realizing M as a subvariety of C') and extensions of their
Poisson brackets to C[gj,...' g,], such that the bi-derivation (2.3) is a Poisson bracket on C".
The answer to this question is at present unknown.
Definition 2.6 Let (M, 1-, -1) be an affine Poisson variety and let x G M. We define the
rank of the Poisson structure at x, denoted Rkxl., -1, as the even number defined by either
one of the following:
(i) The rank at x of any Poisson matrix representing 1., -1;

(ii) The number of linearly independent Hamiltonian derivations at x.
The rank of the Poisson structure is defined as maxjRkxj-, -11 x E MI and is denoted by
Rkj-, -1 and the co-rank dimM -
Rkj-, .1 is denoted by CoRkj-, -1. The Poisson structure

(or variety) is called regular if it has constant rank and trivial if its rank is zero.
equivalence of the two definitions is easily established; since (ii) is intrinsic we see
The
that (i) depend on the chosen system of generators of O(M). Notice that the
does not
rank is also defined at the singular points of M and that, in view of (i) it is maximal on
a Zariski open subset of M (see also Paragraph 2.4 below). Moreover, since at the general
point of M the number of independent Hamiltonian derivations is bounded by dimM we

have that Rkj-, -1 < dimM. We give some first examples of affine Poisson varieties.
22
Example 2.7 Any constant skew-symmetric n x n matrix is the matrix of a Poisson struc-
ture on Cn, in terms of its standard coordinates, as follows from (2.4). We refer to such a
Poisson structure as a constant Poisson structure. By the classification theorem for skew-
symmetric bilinear forms the Poisson matrix takes the standard form
0 1, 0
(0-1, 0
0
0)
0
after alinear change of coordinates. Here 2r is the rank of the Poisson matrix. Thisstructure
is often called the canonical or standard Poisson structure of rank 2r on Cn. Since in the
case of a constant bracket (2.5) degree deg G I we see that a necessary condition for
is of -
such a bracket to define a bracket on an affine subvariety M of Cn is that all elements of

inimal degree in Im are Casimirs.
Example 2.8 Let g be any finite-dimensional (complex) Lie algebra, with Lie bracket
By Proposition 2.4, the Lie bracket extends to the symmetric algebra Symg of 0, making
Sym. g into a Poisson algebra. Since Sym. g -5-- 0(g*) the dual space g* of g carries a natural
Poisson structure. For f g E 0 (g*) it is given at
,
E g* by
If, g I W RR), Qb 1 (2.6)
the hat denoting the natural pairing (,g*)*

-+ g. This Poisson structure is known as the
canonical Poisson structure or the Lie-Poisson structure of g*. The rank at the origin is
always zero, so it is never regular (unless g is commutative so that I -} is trivial). Notice -
that the resulting Poisson algebra is characterized by having independent generators with
respect to which the Poisson matrix is linear.
Choosing a non-degenerate bilinear form on g we have an isomorphism X : Sym g

O(q) defined by
x
(" ej)
sym
11(ei,
It allows us to transfer the Poisson structure to 9; it is easy to see that it is given, for
f,g E 0(g) at x E g by
V, 91W =
(X, [Vf W, Vg(XT'
where the gradient Vf (x) of f E (9(g) at x is defined by
d
Vy E 0 Tf W, Y) =
f (X + ty)-
wt-lt=o
Example 2.9 For quadratic Poisson structures (on C') a general theory is not known. A
3-dimensional family of quadratic structures on C' was given by Sklyanin (see [Skll]). Let
XO; X1 I X2 and X3 be linear coordinates on C4 and consider the following matrix:
0 bjX2X3 b2X1X3 b3XIX2
(-b3XIX2
-b.IX2X3
-b2XIX3
0
-a3XOX3
-a2XOX2
a3XOX3
0
-aixox,
a2 XOx-
aixox,
0
9.
23
Chapter II. Integrable Hamiltonian systems
By Proposition 2.4 four checks of the Jacobi identity sufficeto show that this is a Poisson
matrix and they are all satisfied if and only if
alb, -
a2b2 + a3b3 :::--
0;
notice that this is also

equivalent to the vanishing of the determinant of the Poisson matrix
(for all values of the This gives a 5-dimensional family of quadratic Poisson structures
xi).
on C4. Except for the trivial structure they are all of rank two and two Casimirs are given
2
by ajX21 a2X 2 + a3X2
-
3
and ajX20 b3X22 + b2X2.
-
3
The Poisson structures given by Sklyanin
correspond to
al =
-a2 =
a3,
bi + b2 + b3 == 0-
We will see in Example 2.52 that the 5-dimensional family given above is more general (up
to isomorphism) than the family given by Sklyanin.
Example 2.10 It is easy to describe afl Poisson structures on C2 since in this case the Ja,
cobi identity is satisfied for any skew-symmetric bracket. Let x and y be coordinate functions
then any polynomial V in two variables defines a Poisson bracket on C2 by jx, yj W(x, y) =
and conversely every Poisson bracket is of this form. It is regular if and only if W is constant,
otherwise the rank drops to zero along the algebraic curve W(x, y) 0. =
Example 2.11 For C3 the Jacobi identity is not trivially satisfied. Indeed the matrix
0 H
( -H
G
0
-F -G)
F
0
will be a Poisson matrix if and only if
(V X P) -0
- ==
0, (2.7)
where P =
(F, G, H). For X and W arbitrary polynomials j0 XVW is a solution to (2.7),
=
leading to laxge number of

a Poisson structures on C3; explicitly for such .9 the Poisson
matrix is given by
0 OW -
OW
(
OZ Oy
OW 0 OW
X 8Z OX I
E -OW 0
Oy OX
and W is seen to be a Casimir of this Poisson structure. The rank is two except at the zero
locus of X and at the points where the gradient of o vanishes. Notice that not all solutions
to (2.7) axe of this form, for example the Poisson matrix
0 0 X
(-X 0)
0 0
-Y
Y
is not of the form XVW (with X and W regular) -
24
In many cases, especially in the theory of integrable systems, the space comes naturally
equipped with several Poisson structures, moreover they are often compatible in the following
sense.
Definition 2.12 Let M be an affine vaxiety and let and two Poisson brackets
on M. Then we say that they axe compatible if for any a, b E C the linear combination
1','Ia,b = a 1-, -11 + b I- J2

i
is a Poisson bracket on M. Similaxly n Poisson brackets on M are said to be compatible if
any linear combination of them is a Poisson bracket on M.
implies that 1-, -11 and {* '12 are compatible if and only if [aj-, -11 +
Remark 2.2
bj- '12, al-, .11 bl. i'12]S 0, for all a, b E C, which is equivalent to [1-, -Ili J* *12]S 0-
+ =
7
=
This means that 1', *12 defines a 2-cocycle in the Poisson cohomology of 1-, .11. It is from
this point of view a natural question to ask whether the cohomology class defined by 1., .12
is trivial, i.e., whether there exist a vector field X on M such that Lxf-, -11 f' *12- Such =
1
a vector field is said to have the deformation property with respect to 1-, -11. Notice that
if P and LxP are both Poisson structures then they are automatically compatible Poisson
structures since
[P, Lx P] S =
_[p76PX] = _62P X = 0.
Explicitly, the condition [1-, -11, 1- *121S 1

= 0 takes the following form
Iff, g1l, h12 + f Ig, h1l, f12 + ff h, fIll 912 +

(2.8)
jjfj 912, h1l + 11g, h12) f}1 + j1h, fl21 911 =
0,
for all fig and h in O(M) (or, equivalently, for a system of generators). It is also cleax
from the Schouten bracket that n brackets are compatible if and only if they are pairwise
compatible.
Example 2.13 Constant Poisson structures are always compatible.
Example Examples 2.7 and 2.8 describe all possible constant and linear Poisson
2.14
structures Cn, i.e., Poisson structures whose Poisson matrix (with respect to some, hence
on
with respect to any system of linear coordinates for Cn) consists only of constant resp. linear
elements. One may wonder about their combination, that is, Poisson brackets on Cn whose
Poisson matrices (as above) have entries of degree at most one; let us call them affine Poisson
brackets on Cn. Obviously both the constant and linear paxt of such Poisson structures are
Poisson structures, hence a constant Poisson structure on Cn which is compatible with a
Lie-Poisson structure on C' defines (by taking their sum) an affine Poisson structure on Cn
and any affine Poisson structure on Cn is of this form. These affine structures are known
as modified canonical Poisson structures or modified Lie-Poisson structures. If we denote
the Poisson bracket determined by the linear part by I-, -I, and the one determined by the
constant part by I- ,
-
jo then we see that condition (2.8) reduces in terms of linear coordinates
X1, - -
-, Xn for Cn to
JjXi, Xjjl XkJO + JjXj, Xkjl, XiJO + JjXki Xill, XjJ0 _-:: 0- (2.9)
25
Using the hat-notation from Example 2.8, look at Cn the dual of Lie
we as a algebra 0 with
basis J i and Lie bracket determined by
I&i, &j 4--'-i I

Then the second bracket (the constant bracket) determines a linear map
C: 9 A 9 -+ C : (.- j, &j) F-+ Ixi, xj jo-
With this notation
jjXi7 Xjj1i XkJO ---

C(jXij XjJ17 -:4) Q-'Ni &j], 41);
hence (2.9) becomes
Q-'N,&j1i&'k) + C&ji&k1iM + Q&ki&d7: j):--:: 0-

This formula expresses that C is a 2-cocycle in the cohomology' of
g associated with the
trivial representation of 9 on C, giving another way to describe affine Poisson structures on
a vector space. As
an application of this point of view, recall that if
g is semisimple the first
and second cohomology groups are trivial; then C is a coboundary, C W', written out, =
C( tj,: j) C([., j,&j]) and we see that the affine Poisson bracket is nothing but the original
=
Lie-Poisson bracket with xi replaced by xi + C'(&j), i.e., both brackets are the same
up to
an affine change of variables.
2.2. Morphisms of affine Poisson varieties
We recall that a map 0 : M, -+ M2 of affine varieties is called a regular

map or a
morphism if O*O(M2) C O(MI), where 0*(f) f o 0 for functions f E O(M2); thus 0
=
induces and is uniquely defined by an algebra homomorphism 0* : O(M2) -+

O(MI). A
regular map which has a regular inverse is called a biregular map or an isomorphism.
Definition2.15 Let (M1j*,'}1) and (M2,1',*12) be two affine Poisson varieties, then a
map 0: M, -+ M2 is called a Poisson morphism or a morphism of affine Poisson varieties if
(:L) 0 is a morphism, O*O(M2) c O(MI);
(2) For all f,g E O(M2), O*Jf7912 JOV4*911- =
Both conditions are conveniently summarized by the commutativity of the following diagram:
1',*12
O(M2) X O(M2) O(M2)
0.
X"*1 1"*
O(MI) X O(Mi) O(MI)
T-
5
For an introduction to the cohomology of Lie algebras see e.g. Appendix 5 in [LM3].
26
The standard adjectives which axe used for morphisms of affine varietie's (e.g., injective,
dominant, finite, -) may also be used for Poisson morphisms. A biregular map is a Poisson
..
morphism if and only if its inverse is a Poisson morphism; we call such a map a (Poisson)
isomorphism. When M, is an affine subvariety of M2 and the inclusion map (MI, 1-, .11)
(W J* *12) is a Poisson morphism then M, is called an affine Poisson subvariety of M2.
1
image of an affine variety by a morphism. needs not be an affine variety; consider

The
forexample the image of the map (x, y) -+ (x, xy), defined on C2. If the image of a Poisson
morphism is an affine subvaxiety (i.e., a (Zariski) closed subset) of the target spar-O then
there is an induced Poisson structure on it, making it into an affine Poisson subvariety, as
shown in the following proposition.
Proposition2-16 If 0: M, -+ M2 is a morphism of affine Poisson varieties and the image

O(MI) is an affine subvariety of M2 then O(MI) has a unique structure of an affine Poisson
variety such that 0 can be factorized as the composition of a Su7jective and an injective
Poisson morphism, as in the following diagram:
M, M2
O(MI)
Proof
By assumption 0 can be factorized as 0 o , where M, -+ O(MI) is a surJective
mOrphism (of affine varieties) and s : O(MI) -+ M2 is an inclusion map. Since 0 is surJective,
0* : O(O(Mi)) -+ O(MI) is injective; since O(MI) -+ M2 is an inclusion map, t* -
O(M2) -+ O(O(Mi)) is surJective.

This leads to thefollowing definition of a Poisson bracket on O(MI). Let f g E.O(O(Ml)), ,
then by surJectivity of%* there exist f' and g' in O(M2) such that f z*f' and g =s*g. We =
define
If; 9} =:= Z* lfi 91 12; (2.10)
and verify that it is independent of the choices made for f'
and g; since 0* is injective it
suffices to show that *S* Ift 9'}2 depends only on f
5
and g. Using the fact that 0 is a Poisson
morphism we find
0*ZIf 9112,
::=
f *fi *911- (2.11)
It follows from (2.10) that f -, -1 satisfies the Jacobi identity and that Z is a Poisson morphism.
Now (2.11) can also be written as
g} *g} I,
so that is a Poisson morphism.
We now show how the rank of the Poisson structure at a point and at its image by a
Poisson morphism are related; it implies equality for an isomorphism.
6
Examples include closed, proper and finite morphisms, see [Har] Ch. II 4.
27
Proposition 2.17 Let (Mi, I- ji)

, , (i =
1, 2) be two affine Poisson varieties and let 0
M, -+ M2 a Poisson morphism. Then Rk.,j-, -11 : Rk,5( &, *12 for any x E MI.
Proof
Let gj, . . .
, g,
be a system of generators of 0 (M2). Then hi =
O*gj, (i 1, =
s) can
. . .
,
be completed into a system of generators hi, , h,+t

of 0 (MI). . . .
By definition of the rank
we have
Rkx I == Rk (I hi, hj
> Rk (1hi, hj I (x)) I<i,j<,,
=:
Puko(m)l'i *12-
The proposition implies that in general an affine subvariety of an afline Poisson variety
does not carry a Poisson structure which makes it into an affine Poisson subvariety. Necessary
and sufficient conditions for this to happen will be given in Proposition 2.18.
2.3. Constructions of affine Poisson varieties
Thereare four (known) basic constructions of PoiSson brackets on finite dimensional
spaces (here taken to be affine varieties). The first one is that of the canonical Lie-Poisson
structure (Example 2.8). In the most important examples, at least from the point of view of
integrable Hamiltonian systems, the relevant Lie bracket is an R-bracket, see Section V.5. The
second one consists of the canonical Poisson structure associated to a symplectic structure
(see Example 4.2 below), the prime example being here the one of cotangent bundles. Notice
that in the first case the Poisson structure is never regular while in the second case it is
always regular. Both these axe very classical, as opposed to the other two which will not be
discussed in detail here. The first of these two deals with Poisson structures on Lie groups,
in particular Lie-Poisson groups; an excellent account of this is given in Semenov's paper
in [BCKII. The last one consists of the construction of higher order brackets, starting from
a Lie bracket, also within the R-matrix approach (see [LP]).
Apart from these four basic constructions there are also several constructions which allow
one to build new Poisson brackets from given ones. We will discuss these here in the context
of affine varieties, in fact we will show how the standard constructions by which new affine
varieties are built from given ones, have their equivalents for afline Poisson varieties.
We think here of the following:

(1) an affine subvariety;
the restriction to
(2) the product of two affine Poisson varieties;
(3) the quotient and fixed point set of an affine Poisson variety under the action of a
finite reductive group;

or
(4) removing a divisor.

They are considered next in the above order: we start by giving a precise condition for a
Poisson structure to restrict to an affine subvariety (for animportant example, see Proposi-
tion 2.38 below).
Proposition 2.18 Let (M, I., -J) be an affine Poisson variety and suppose that N is an
affine subvariety of M. Then the following are equivalent.-
28
(i) There exists a Poisson structure on N with respect to which N is an affine Poisson
subvariety of M;
(ii) The ideal of N is a Poisson ideal of O(M);
(iii) The restriction of every Hamiltonian vector field Xf, f E O(M) to N is tangent
to N.
Proof
Theequivalence of (ii) and (iii) is immediate from the definition of a Poisson ideal: an
ideal of Poisson algebra is called a Pois8on ideal if it has the additional property of being
a
a Lie ideal; thus, -EN, the ideal of regular functions vanishing on N, is a Poisson ideal if and
only if I-EN, 0 (M) I C -TN which is in turn equivalent to
Xf g =
1g, f I vanishes at all points of N,
for every Hamiltonian vector field Xf, f E O(M) and every g E IN.
If IN is a Poisson ideal of 0 (M) then for any f, g E 0 (M) and n c- N
If + IN, g + -TN} (n) =

I f gj (n),
, (2.12)
and we can define 1',*IN at n E N by j%*fjZ*9jN(n) =

If,gl(n), where z : N -+ M is
the inclusion map; clearly the latter becomes a Poisson morphism. This shows that (ii)
implies (i). If z : N -+ M is a Poisson morphism then (2.12) holds, in particular JIN, gj
vanishes on N for any g E O(M). Since IN is a prime ideal it follows that JIN, gj C IN for
any g E O(M), so (i) implies (ii). I
Example 2.19 Suppose that is an ideal of a Lie algebra g and denote by (0) the ideal
of Sym g generated by 0. For h E 0 and for gi, g,,, F- g a direct application of the Leibniz
rule shows that every term in
1h, g, "
92' gm
belongs to (0), where denotes the canonical Poisson structure

on Sym g. This shows
that is a Poisson ideal of Symg. Under the canonical
isomorphism Symo 1--- 0(,g*) the
ideal corresponds to the ideal of functions vanishing on . Therefore, Proposition 2.18
implies that the subspace of g* which consists of the elements of Z* that vanish on 0 is an
affine Poisson subvariety of 9* with its Lie-Poisson structure.
Remark 2.20 There are, of course, other ways in which a subvariety may inherit in -
one way or another a Poisson structure from its ambient affine Poisson variety. Think for
-
example of a proper symplectic subvariety of a symplectic, manifold, which is never a Poisson

subva,riety, but still carries a natural Poisson structure "inherite&' by the symplectic 2-form.
on its ambient manifold. This will be discussed later in this paragraph.
29
Second, we consider products of affine Poisson varieties.
Proposition 2.21 Let (Mi, 1., -11) and (M2, 1" '12) be two affine Poisson varieties, then 7
the product M, x M2 has a natural structure of an affine Poisson variety such that the
projection maps iri : M, x M2 -+ Mi are Poisson morphisms. Moreover it has rank Rkf +
Rkf-,'121 the rank at (MI, M2) G M1 x M2 being given by Rk,,,, + Rk,,,,I-
Proof
The algebra of regular functions on the product Mi x M2 is given by
0 (MI X M2) =
7r* 0 (MI) 0 lr * 0 (M2) 1 (2.13)
hence the construction amounts to making the tensor product of two Poisson algebras into
a Poisson algebra. Formula (2.13) implies that O(M., x M2) is generated by the func-
tions ir,*L fl, irl fn, 7r2*gl,
. . . 7r2*g,,,, where fl,
, fn is an arbitrary system of generators
. ..
, . . .
,
of O(MI) and gl,...,g,,, for O(M2). In order for -7r, and ir2 to be Poisson morphisms it is
necessary and sufficient to define J1r1*fi,-7r1*fjJ = ?r1*Jfi,fjJ1 and J7r2*gi,7r;gjJ
2 ir2J9i,9jJ2 =
for all i and j. A natural choice for the remaining brackets firl*fi, 7r2*gjl is to make them
all zero: with this choice the Jacobi identity is surely satisfied. The Poisson matrix of 1., .1
with respect to the generators ?rif1,...'7r2*g,,, has a block form, hence Rk(,,,,,,,n,)I-, J =
Rk,n, I , I I + Rk,,,, I 1 '12 for any (tnl , M2) E M1 X M2 It is also easy to see that the algebra
- - -
-
of Casimirs of (Mi x M2, is given by ir,* Cas(MI) 0 7r;2 Cas(M2).
Definition 2.22 The Poisson bracket on M, x M2 given by Proposition 2.21 is called the
product bracket and is denoted by I-, -1m, xm,.
Example 2.23 Suppose that (G, 1-, -1) is an affine Poisson variety and that G is an alge-
braic group with multiplication X: Gx G -* G. Then (G, 1-, .1) is called an affine Lie-Poisson
group if X is a Poisson morphism, the Poisson bracket on G x G being the product bracket.
Example A related construction appears when having a family of affine Poisson

2.24
varieties whichdepend on a (or several) parameter(s). More precisely we assume that there is
given a dominant morphism 7r : P -+ N of affine varieties (N being the parameter space) and
a Poisson bracket I* Jn on each non-empty fiber 7r-'(n)
i
Define for f g E O(P) and p E P .
,
ff'gl(p) =
(p). (2.14)
If for any f,g c O(P) one has ff,gl E O(P) (roughly speaking, if the brackets J.,-In
va,ry regularly with n Ei N) then 1-, .1 defines a Poisson bracket on P and for any n E N the
bracket f -In makes every irreducible component
-
i
of 7r-1 (n) into an affine Poisson subvariety
of (P, I-, J). As a special case, let P M x N, = where M is an affine Poisson variety and
take ir as projection on the second factor. Clearly this leads to a Poisson structure on each
of the fibers of .7r which varies regularly with n E N. The resulting Poisson structure on P
coincides with the product bracket on M x N, where N is given the trivial (zero) bracket.
Third, we consider the Poisson structure on the fixed point set and on the quotient
space MIG finite group or a reductive algebraic group which is
where G is a equipped with
a Poisson structure (notice that it needs not be an affine Lie-Poisson group).
30
Before doing thiswe recall a few facts about group actions on affine varieties. All groups
considered here will be either finite or reductive; moreover, when we want to consider Poisson
structures on reductive groups we will assume that they are affine varieties, so as to stay in the
category of affine Poisson varieties. A (finite or algebraic) group G is said to act on an affine
variety M C C' if the action is the restriction to M of representation of G on Cn. When a
G is finite every representation on Cn is completely reducible, i.e., if the action of G leaves

invariant some subspace of Cn then it leaves invariant a complementary subspace. For infinite
groups the above property characterizes reductive groups (since we are working over C). We
recall also that there is an induced action of G on O(M), given for g E G, f E O(M) and
x E M by g*f (x) f (g-'x).
=
If G is finite then we say that the action of G on M is a Poisson action if for every g E G
the isomorphisin M -+ M defined by m i-+ gm is Poisson. If G is infinite, say G is an affine
algebraic group, it may itself carry a Poisson structure and the proper generalization of the
above notion of Poisson action is that the map G x M -4 M is a Poisson map, where G x M
is given the product Poisson structure.
Proposition 2.25 Let (M, I-, Jm) and affine Poisson variety and let G be an affine alge-
braic group acting on M.
(i) If there is a Poisson structure on G for which the action is a Poisson action, then
the algebra O(M)G of G-invariant functions is a Poisson subalgebra of O(M).
(2) If G is moreover reductive or finite, then O(M)G is finitely generated, hence cor-
responds to an affine Poisson variety MIG, leading to the following commutative

diagram of Poisson morphisms (7r2 is projection onto the second component).
x
GxM I M
1r2
1r (2.15)
M -,r MIG
Proof
Clearly f E 0 (M) is G-invariant if and only if the following diagram is commutative.
x
GxM -
M
7r2 f
M - C
f
Thus, if f, g E O(M)G and X is Poisson then
X*Jf19JM--`JX*f1X*gJ G xm --flr2Yi r2*91GXM:--:7r2*ffiglMl

and we see that the bracket of any two G-invariant functions is G-invariant. Therefore the
subalgebra O(M)G of O(M) is aJso a Lie subaJgebra of O(M), i.e., it is a Poisson subalgebra
of O(M).
31
Chapter Il. Integrable Hamiltonian systems
Assume now that G is reductive or finite. Then O(M)G is finitely generated (see
e.g.[Muml] or [Spr]) hence is the algebra of regulax functions on an affine variety, denoted
MIG and called the (categorical) quotient; MIG is naturally identified with the orbit space
of an open subset of M. The natural projection map M -+ MIG is regular and yields the
commutative diagram (2.15). Granted (1) this proves (2). M
We next consider
generalization of the above proposition. We consider again a group G
a
(assumed finite or reductive) acting

on M and leaving some subvariety N of M invariant.
We will show that N may inherit a Poisson structure from M, even if N is not a Poisson
subariety of M. Let us denote by O(M, N)G, the algebra of regular functions on M that
restrict to G-invariant functions on N, and by p : O(M,N)G -+ O(N)G the natural map
induced by the inclusion N M.
Definition 2.26 Let (M, -1) be an affine Poisson vaxiety, X : G x M -+ M a Poisson

action and N a subvaxiety of M which is G-stable. Then the triple (M, G, N) is called
Poisson-reducible if O(M, N)G is a Poisson subalgebra of O(M) and if there exists a Poisson
bracket on O(N)G such that
JP(FI)i P(F2)10(N)G PJFJ, F21 (2.16)
holds for all F1, F2 E O(M, N)G.
Formula (2.16) says that in order to compute the Poisson bracket of two G-invariant
functions on N one computes the Poisson bracket of any extensions to M and then restricts
the result to N. Notice also that (2.16) uniquely defines a bracket on O(N)G (if it exists)
since p issurJective. In the following proposition we give necessary and sufficient conditions
for (M, G, N) to be Poisson-reducible (for a proof see [PV1]).
Proposition 2.27 Let (M, 1., -1) be an affine Poisson variety, X: G x M -+ M a Poisson
action and N a subvariety of M which is G-stable. Then (M, G, N) is Poisson-reducible if
and only if
p f O(M, N)G7 I(N)} =
0; (2.17)
it is implicit in this condition that its left hand side makes sense.
In a slightly different vein a Poisson structure is also inherited by the fixed point variety
of group action. This fact was first shown in
a [FV] in the case of a Poisson involution and
is generahzed in Proposition 2.29 below. First we need a lemma about the ideal of a fixed
point variety.
Lemma 2.28 Let G be a finite or reductive group acting on an affine Poisson variety and
let N be its fixed point variety. The ideal IN is generated by functions fj for which there
exist gj E G and j (E C 11 such that gj* fj =
j fj
Proof
Since G is finite or reductive the representation space C' which contains M decomposes
as a direct of spaces VO,
sum ,
V,, such that VO is the fixed point set of the action and
. . .
such that the action of G on the other Vi is irreducible. Then vo n m N and IN is =
generated by ?ri* Vi*) where 7ri is the natural projection Cn -+ Vi and i =

1, . . .
,
s. Let i be
32
fixed (between I and s) and take any non-zero element 0 E Vi*. Since the action of G on Vi
is irreducible there exists g E G such that g*0 -A 0. Since G is reductive (or finite), G is
generated by its semi-simple elements (see [Hum] p. 162). Therefore, let g be a semi-simple
element for which g*0 =54 0. Then we have a linear basis 01, ...' 0, of Vi*, where g*oj =
joj
for j =
1,...,v with j = 0 for j =
1,...,u and j :A 0 for j = u1,...,v. Since g*0 7 0
+
it follows that u < v and we have at least one function V) for which g*'O O, with 7 1.
=
Consider now the subspace Wj* of Vj* which is the span of all functions fj E Vj* for which
there exists gj E G and j : :4 1 such that gj*fj jfj- We have already established that Wj*
=
contains a non-zero element. Therefore it suffice to verify that Wj* is invariant to conclude
that Wj* =
Vi*. Let f rjn-, cj fj (=- Wi*, with fj as above, and let g E G. We need to show
=
that Ejn-, cjg*fj E Wi*. This follows at once from
(g-,gjg)*g*fj =
g*gj*fj =
6g*h,
since j :7 1.
Proposition 2.29 Suppose that G is a finite or reductive group acting on an affine Poisson
variety (M, 1-, -1). We assume that for every g E G the isomorphism -ID, : M -+ M which
corresponds to the action of g is a Poisson map. Let N be the subvariety of M consisting of
the fixed points of d) and denote the inclusion map N -+ M by i. Then N carries a (unique)
Poisson Structure j",jN such that
Z*IFIIF2} =
jS*F1jZ*F2jN
for all F1, F2 E 0 (M) that are G-invariant.
Proof
For fl, f2 E O(N) we choose F1, F2 E O(M) such that f, t*Fl and f2 %*P2. Since = =
G is finite or reductive we may assume that F, and F2 are G-invariant. We define
Iflif2IN =
Z*fFliF21 (2.18)
and show that this definition is independent of the choice of F, and F2. To do this it is
sufficient to show that if F, and F2 are G-invariant, with z*Fl 0, then z* IF,, F21 0. We
= =
will actually show that if F2 is G-invariant and F1 E IN then z*IFI, F21 = 0. Let us denote
by fl, . system of generators of IN as given by the previous lemma. If F is G-invariant

. .
, ft a
then the fact that (D, is a Poisson map implies
z* jfj, Fj = z*-I)* jfj,

.9
Fj = z* 1-(D*fj,
9
- 9*F} =
jz* jfj, FI,
showing our claim, since j =A 1. Note also that the bracket of any two G-invariant functions
is G-invariant. In view of this and because (2.18) is independent of the choice of F, and F2
we have for any fl, f2, f3 E O(N) that
jjf17f2jN)f3 IN= Z*jjF1iF2jiP3}i
leading at once to the Jacobi identity for Similarly the fact that j* JN
I
is an anti-
symmetric biderivation follows.
33
We next give a few examples of Proposition 2.25 in the case of finite group actions 0, Cn.
For examples which involve larger groups see Section VI.3.
Example 2.30 Consider the following automorphism of C2,
?,1(P17P2) (-PliP2)i
which corresponds to a diagonal action Of Z2 on C2 which has a line of fixed points. Let
us compute the algebras of invariants for the induced action and derive from it the Poisson
structures on C2 which descend to the quotient. If we denote the standard coordinates on C2
by x, and X2 then the invariant functions are the polynomials whose terms are even in x1,
hence the quotient again C2 and the projection map is given by (PI P2) -+ (ql, q2)
C2/Z1 is i
=
(P2,
I P2).
The map zi is a Poisson morphism, if and only if I-XI X21 Z*JX1i
1 X21- If we denote 7
=
JX1, X21 F(X1 X2) then this condition means that -F(XI, X2) F(-XI, X2) i.e., F is odd
=
I
=
in xjL and it follows that F can be factorized as x, times an invariant polynomial. Then the
Poisson structure on the quotient is given by JY1 Y2 1 0 2y, G (yl, Y2) where G is defined by i
=
F(XI,X2) ==
xG(X2,
I X2).
Notice that these Poisson structures on C2 and C2/t, have a line
where the rank is zero: if non-trivial they are never regular.
Example 2.31 The only other possibility (up to isomorphism) for Z2 to act non-trivially
on C2 corresponds to the following automorphism of C2:
t2 (PI P2) 7
=
(-PI -P2) i
In the notations of the previous example, a polynomial function is now invariant if and only
only of terms which axe of even total degree in x, and X2. Therefore the algebra
if it consists
of invariants is generated by y, x 2, Y2 XIX27 Y3 =X22, with the single relation y22
=
y1y3 = =
1
and the quotient space is a quadratic cone. The projection map is then given by (plIP2) 1-+
(ql, q2, 93) 1 PIP27 p2
(p2' 2) and 1.2 is a Poisson morphism. if and only if the polynomial F
=
which defines the bracket, JX1 X2} F(xi, yl) is even, F(X1, X2)
i
=
F(-XI, -X2) In this =
-
case there exists a polynomial G(yl, Y21 Y3) for which G(X2, 2)
1 XIX2iX 2 F(X1, X2)- In terms =
of the generators Yli Y2 and y3 the Poisson structure on the quotient is then described by the
following Poisson matrix-
0 YJ 2Y2
2G(y, 7 Y2 i Y3) -Y1 0 Y3
-2Y2 -Y3 0
Even if the original Poisson structure is regular (e.g., if F 1, in which case (C2 /221 1*;'10) =
is an affine Poisson subvariety of,01(2)*, with its standard Lie-Poisson structure) the quotient
Poisson structure (if non-trivial) is never regular: it always has rank zero at the vertex of the
cone.
Example 2.32 The two preceding examples are easily generalized, giving all possible ef-
fective actions of a cyclic group on C2 Namely let .
Z3 (P1 i P2) =
(TI 6P2)i 7
ep = 6q = 1.
Here p and q are assumed to be the smallest integers satisfying ep 6q 1. The map Z3 = =
corresponds to an effective action of the cyclic group of order 1 1.c.m.(p, q) and by the =
34
remarks made above every such action is of this form. Suppose first that p and q are not
coprime, let d denote their g.c.d. and p p'd, q q'd. Then y, 4 and Y3 Xq2 are invariant
= = = =
'
functions and, since EP' and dq are primitive d-th roots of unity, we may suppose that x, and
X2 are chosen such that Y2 =4 1x"

2
is also invariant. Then the quotient is a cone in C3 given
by y1y3 yd.
2
The bracket JX1, X21
=
F(xl, X2) defines a Poisson bracket for which 23 is a
=
Poisson action if and only if flexi, 6X2) == e6flxi, X2)- If 6: :k 1 then F(xi, X2) is X1X2 times
an invariant polynomial and we may define G(yl, Y2i Y3) by XlX2G(xp, 1,

Xp'Xq'
1 q)
2,X2 =F(x,,X2)-
It is easy to check that the Poisson structure on the quotient is then described by
0 PY1Y2 pqy.lyf3
G(y,IY2iY3)
( -PYIY2
-pqyly3
0
-qY2Y3
qY2Y3
0
If on the other hand 6 = 1 then p =

q and v =
X1X2- Then F(xl, X2) is invariant and we
may define G(yi, Y27 Y3) by G(4, XIX2i4) =

F(xi, X2). The Poisson structure is in this case
described by
0 Y1 PYP2-1
pG(y17Y21Y3)
( -PYP2
Y1
I
0
-Y3
Y3
0
Finally, if p and q arecoprime then 4 and Xq2 generate the algebra of invariants, hence this
algebra is a polynomial algebra. As above F is divisible by XIX2 and we define G(u, v) by
G(4, xq)
2
=
F(xi, X2) and find that the Poisson structure on the quotient is determined by
JYI Y210 =pqy1Y2G(y1,y2).
Example 2.33 The cyclic permutation
Z4(PliP27N) =
(P27NiPl)
gives an action Of Z3 on C3 which will also appear later on (see Section VII.7). It is not in
diagonal form; in order to diagonalize the induced action on C [XI i X2 i X31 let e be a primitive
cubic root of unity and define
U1 =
X1 + X2 + X31
U2 =
X1 + 'EX2 + 6 2X31 (2.19)
U3 =
X1 + 6 2X2 + C:X3-
Then the action of I E Z3 on C[Ul, U2, U31 is given by T*,ul =

U1, T*JU2 =
15U2 and VJU3
2 3 3
6 U3. The algebra of invariants is now generated by ul, v =
U21 W =
U3 and t =
U2U3 with
the single relation t3 showing that the quotient C3 /Z4 is a cylinder over a cubic surface.
=
vw,
In terms of the new projection map is given by (q q2, q3)
coordinates the (ql, q 3 q3, q2q3). , ,
Assign to ui a weight i 1 and let X and-
W be polynomials in ul, U2 and U3, all of whose

terms have the same weight modulo 3. According to Example 2.11 these lead to a Poisson
structure on C3; the above action Of Z3 will be a Poisson action if and only if all terms in the
product XW have weight 0 modulo 3; equivalently X and W must both be weight homogeneous
and the sum of their weights must be a multiple of 3. The resulting Poisson matrix for the
quotient is easily written down.
35
Example 2.34 As a final example let us consider the natural action of the symmetric
group Sd on Cd. It is well-known that the algebra of invariant functions for this action is
freely generated by the elementary symmetric functions, in particular the invariant :ftmctions
constitute a polynomial algebra and the quotient is just Cd, the projection map being
(pi I ... 1Pd) -+ (PI +P2 + - - - +Pd, ...

1P1P2 * '
*Pd).
By a transformation similar to (2.19) the action can be partly diagonalized. The Poisson
brackets which descend to the quotient are the ones for which the bracket of any two sym-
metric polynomials is a symmetric polynomial.
The fourth and final construction is that of removing a divisor from an affine Poisson
variety. We show that the resulting space still has the structure of an affine Poisson variety.
Proposition 2.35 Let (M, I-, -1m) be an affine Poisson variety and let f E O(M) be a
regular function which is not constant. Then there exists an affine Poisson variety (N&,*JN)
and a Poisson morphism N -+ M which is dominant, having the complement (in M) of the
zero locus of f as image.
Proof
Consider a new variable t and define an affine variety N by its ring O(N) of regular
functions as follows:
O(M)[t]
O(N) =
idl(f t 1) -
Let us denote the canonical projection by 7r : N -+ M. Clearly ir is dominant and its image
is the complement of the zero divisor of f. If 7r is to be a Poisson morphism, then we axe
forced to define 19i , gj IN Jgi, gj I for a system of generators gi I

=
I A
Of O(M) (we made ...
a notational identification between -7r*gi and gi). In view of Proposition 2.4 the only way to
extend this to a Poisson structure on N is upon using I f t -
1, gi 0 (for i ==
1, . . .
, k).
Thus one needs to add the brackets
19i, tIN =
_t21g,J f IN- (2.20)
By the same proposition it now suffices to check the Jacobi identity on the system of gen-
erators gi, . . .
,gk, t; since we know it is valid for gi gk the following easily established
identity suffices:
119ii gjJ7 tIN + Jjgj t1i giIN + JJt7 gili gjIN = 0
(one uses JJ9ii9j}itJN = -t2 ffgi gjJ7 fIN7 which is an immediate consequence of (2.20)).
This gives the desired Poisson bracket on N. Note that if f G Cas(M) then
Cas(M)[t]
Cas(N) =
idl(f t 1) -
Remark 2.36 Another way to state the above result is that the algebra of rational func-
tions on an affine Poisson variety with poles only at some fixed divisor (which need not
be irreducible) is also a (finitely generated) Poisson algebra. Clearly, the field of rational
functions of an affine Poisson variety can in a similar way be turned into a Poisson algebra.
36
2.4. Decompositions and invariants of affine Poisson varieties
There are three natural decompositions of an affine Poisson variety, two of which con-
stitute of algebraic varieties. The first one discussed here is a level decomposition by the
Casimirs, the second is a decomposition according to rank and the last one the non- -
algebraic one is the decomposition into symplectic leaves. Due to its non-algebraic nature,
-
the latter will only indirectly (via the other decompositions) be used in this book and is
discussed in Section 4.
a The Casimir decomposition

The decomposition of an affine Poisson variety which is naturally associated to its algebra
of Casimirs applies (and will be applied) equally for other subalgebras of O(M), so let us
introduce it for an arbitrary subaJgebra A (containing 1) of O(M). To each m E M we may
associate an algebra homomorphism X,,, : A -+ C by f -+ X,,, (f f (m), allowing us to
associate to m E M the ideal
If -
x,,(f) I f E Al
of A, which is a point in Spec A, the spectrum of A; we will denote this natural map M
Spec A by irA. Another way to see how this map comes out is as follows. The inclusion map
z : A c M allows one to associate to a prime ideal I of O(M) the prime ideal I n A of A,
hence leading to a morphism
?,* : Spec 0 (M) -+ Spec A
of affine schemes. The space Spec O(M) contains M as the set of its closed points this -
set may also be seen as the maximal spectrum, the space of all maximal ideals, of O(M) -
and our just the restriction of %* to M. We prefer to work with M rather than
map irA is
with Spec O(M) since that is the space we originally started from; however we like to keep
Spec A, even when A is finitely generated so that its underlying variety is an affine variety,
because we will also be interested in the fibers of z* over points which are not closed. Notice
that the irreducible components of each fiber of 7rA are affine varieties and a complete set
of equations (from which we may choose a finite generating set) for the fiber which contains
m E M is given by
Vf E A: f (x) xn(f). =
(2.21)
Often our statements will be about the general fibers of irA: when saying that some
property holds for a general fiber we mean that it holds for the fiber over a general point, i.e.,
for all closed points which do not belong to a certain divisor. We denote the Krull dimension
of A by dim A; if A is finitely generated then it is a basic result that dim A dim Spec A. =
The following proposition relates the dimension of A to the dimension of the fibers of irA
(see [Sha] Ch. 116).
Proposition2.37 All (non- empty) fibers of irA : M -+Spec A have co-dimension at most
dim A and the general fiber has co-dimension precisely dim A.
Let us apply this to the case where A is the algebra of Casimirs of an affine Poisson
variety (M, 1-, -1), which we denoted by Cas(M). As an application of Proposition 2.18, the
following proposition shows that the fibers of M -+ Spec Cas(M) inherit a Poisson structure,
thereby giving each irreducible component the structure of an affine Poisson variety.
37
Proposition 2.38 Every (non-empty) fiber.F of -7rc,,(M) inherits a Poisson bracket

from 1-, -1 and all Hamiltonian vector fields Xf, f E O(M) are tangent to these fibers.
Moreover RkJ +F < RkJ -
I with equality for a general fiber .97.
,
-
,
-
Proof
It suffices to prove the property for a (non-empty) fiber over a closed point. Let F be
any such fiber and m E F. In view of (2.21) the ideal of F is generated by
If -
x,,,(f) I f E Cas(M)l
so it is a Poisson ideal of O(M) and Proposition 2.18 applies. In order to determine the rank
of the restricted Poisson bracket we use item (ii) of Definition 2.6. It shows that the rank
of both structures is the same at every point. This leads at once to the inequality; in order
to obtain the equality for a general fiber, just recall that the rank of is maximal on a
Zariski open subset of M. I
Referring to Example 2.24 we see that the Poisson structure on M can be reconstructed
from the Poisson structure on the fibers of M -+
Spec Cas M, where Spec Cas M plays the role
of the space of parameters. Therefore we will call Spec Cas M the parameter space and the
map M -+ Spec Cas M the parameter map of the affine Poisson variety (M, 1-, .1). The fibers
of the parameter map will also be called level sets of the Casimirs because picking a -fiber
(resp. over a closed point) corresponds to fixing some (resp. all) Casimirs. The decomposition
of M into the'fibers over closed points is called the Casimir decomposition.
Remarks 2.39
1. A Poisson morphism does not necessarily induce a map of the corresponding parameter
spaces (for a counterexample, see Example 3.14 below).
2. Since Cas(M) is a Poisson subalgebra, of O(M) (actually a Lie ideal), 17rCas(M) is a
Poisson morphism, Spec Cas (M) having the trivial Poisson structure.
3. For special fibers F of the parameter map the rank of the Poisson structure may be
strictly smaller, see Example 2.54 below.
Proposition 2.40 Let (M, 1-, .1) be any affine Poisson variety. Then
dim Cas(M) : CoRkJ-, J.
Proof
Let F be a general fiber of the parameter map and let I-, -JY be the induced Poisson
structure. Then dimM -
dim.F dimCas(M) and RkJ-, .1jr.
= =
RkJ-, .1. Since dim,r (resp.
RkJ-, jp) equals the number of independent derivations (resp. Hamiltonian derivations)
of 0(,F) at a general point of 17 we find that
dim Cas(M) = dimM -

dim.F < dimM -
RkJ-, .1 =
CoRkJ-, -1.
38
studying integrable Hamiltonian systems we will exclusively be interested in affine

When
Poisson varieties for which the above inequality is an equality, because only in that case, the
varieties or manifolds which are traced out by the flows of the integrable vector fields, can be
affine (sub-) varieties (of the phase space). It motivates the following definition.
Definition 2.41 Let (M, be an affine Poisson variety. We say that its algebra of
Casimirs is maximal if
dim Cas(M) =
CoRkj-, J,
i.e., if the general level set of the Casimirs has dimension equal to the rank of the Poisson
structure.
Following the above comment, we will want to know if maximality of the algebra of
Casimirs is preserved by the different constructions we made (restriction, product, quotient
and taking away a divisor). We show that maximality of the algebra of Casimirs is indeed
preserved by restriction of the Poisson structure to a general level of (a subalgebra of) the
algebra of Casimirs. Similar propositions for the other constructions which we have discussed
are easier to obtain and axe not made explicit here.
Proposition 2.42 Let (M, 1-, -1) be an affine Poisson variety whose algebra of Casimirs is
maximal. Then for any subalgebra A of Cas(M) the induced Poisson structure on the general
fiber of M -+ Spec A is also maximal.
Proof
Let A be any subalgebra of Cas(M) and let Y denote a general fiber of the natural map
M -+ Spec A so that dim -T = dim M -
dim A. Obviously, if f E Cas(M) then fly E Cas(.F)
and
dimCasY > dimCas(M)ly =
dimCas(M) -
dimA.
Since Cas(M) is maximal and Jr is general it follows that
dim.F -
dim Cas(.F) < dim M -
dim Cas(M) =
Rkj-, j =
Rkj-, jy,
showing that for a general fiber F the algebra of Casimirs of I -
,
-
IT is maximal.
Example 2.43 The algebra of Casimirs needs not be maximal. The simplest counterex-
ample is given by the following Poisson matrix, defining a Lie-Poisson structure on C3,
0 0 ax
( 0
-ax
0
y
-Y
0
)1
coming from a solvable Lie algebra. F(x, y, z) will be a Casimir if and only if
8F i9F 9F
i9z =ax5x -y-=0-
C9Y
These equations easily solved giving F cxy" +d where c and d are integration constants.
are =
If 0
a N then F is nota regular function, if a 0 Q then the level sets of F are not even
algebraic varieties. It should however be remarked that for any a p1q E Q we can restrict =
the Poisson structure to a general level, which is an affine algebraic surface given by xqyP C, =
where C E C.
39
Example 2.44 Taking up Example 2.11 again we see that every polynomial F(x,'Y, z) al>-
pears as a Casimir for some C3. Namely, consider the Poisson
Poisson structure on structure
on C3 defined by the following Poisson matrix (which corresponds to X 1 and 0 = =
F):
OF OF
0
OZ OY
-OF OF
0
OZ Ox
OF OF
0
;9_Y _5X_
Then F is a Casimir of this Poisson structure. All the fibers of the parameter map are two-
dimensional and if F is non-constant then they all have rank two, the singularities of these
fibers being precisely the points where the rank drops to zero.
Example 2.45 We also give an example to show that the fibers (of the parameter map)
lying over closed points may have higher dimension. Consider on C4 the following Poisson
matrix:
0 0 0 X
0 0 0 -Y
0 0 0 -Z
-X y Z 0
It is easy to verify that the algebra of Casimirs is given by Cas(C4) C[xy,xz]. Thus =
the parameter space Spec CaS (C4) is isomorphic to C2 and the fibers of the parameter map
are given by xy a, xz
= b for a, b E C. For (a, b) 54 (0, 0) the fiber is two-dimensional,
=
however xy xz = 0 has x
= 0 as a component, hence is three-dimensional. Notice that
=
on this special fiber the algebra of Casimirs of the induced Poisson structure is not maximal
anymore.
We prove one more proposition about the algebra of Casimirs which may be useful for
its computation; the type of axgument used in the proof will be used several times in the next
section.
Proposition 2.46 The algebra of Casimirs of an affine Poisson variety (M, is inte-
grally closed in O(M).
Proof
Suppose F E O(M) and Ein-0 ajF' =
0, with ai E Cas(M), an = 1 and n minimal. We
need to show that F E Cas(M). Taking the bracket with any g G O(M) we find
0 =
Ig, ajF'j (
'
E
i=O
=
E(i
i=1
-
I)aiF'-') Ig, F1,
for all g G O(M). Since n was supposed minimal, it follows that Ig, Fj = 0 for all g EE O(M),
i.e., F E Cas(M).
40
9 The rank decomposition

Another decomposition relates to the rank of the affine Poisson variety at each point. Given
(M, define for 0 < i < 1
2 Rkf -, -1
Mi =
lp E M I Rkj-, -1 < 2ij.
Then the components of each Mi are affine varieties: let g be the Poisson matrix of I. , j
with respect to arbitrary system of generators, then Mi
an can be described as a so-called
determinantal variety (see [ACGH] Ch. 2)
Mi =
lp E M I all determinants of order 2i + I of g vanish at pl,
7
a description which also gives the equations defining these algebraic sets Obviously Mi C .
Mi+,; also for r Rkj-, -1/2

= one has M Mr and the Poisson structure
= is regular if and
only if Mr-1 0; finally Mo M if and only if the bracket is trivial. We call the (singular)
= =
stratification by the Mi the mnk decomposition of M.
The algebraic setsMi may be reducible and their components may be of varying dimen-
sion. Therefore we define for 0 < i < I
2 Rkj-, .1
and j E N
Mij E Mi I 3D irred. comp. of Mi, p E D and dim D =

j
Thus Mij is the (finite) union of the j-dimensional irreducible components of Mi. This leads
at once to the following polynomial invariant for an affine Poisson variety.
Definition 2.47 Let (M, 1-, -1) be an affine Poisson variety. Its invariant polynomial
p(M) =
p (M, -1) E Z[R, S] is defined by
p(M) =
E pij RSj, pij irred. comp. of Mij.
The polynomial can also be represented by a matrix (of minimal size), called the invariant
matrix of the Poisson structure by taking as (i, j)-th entry the integer pij (labeling of matrix
entries starts here from zero).
Proposition 2.48 Let (MI, 1-, .11) and (M2, 1*) '12) be two affine Poisson varieties. If
there exists a biregular map 0: MI -+ M2 which is Poisson, then p(MI) =
p(M2).
Proof
We noticed that the inverse of biregular Poisson morphism is also Poisson. From
a
Proposition 2.17 it follows that for all points x, the rank at x equals the rank at O(x).
Thus the restriction of 0 to any of the subvarieties M, or Mr, is a biregular map onto the
subvarieties N, and Nr,. It follows that p(M) =
p(N). 1
7
Strictly speaking the Mi are defined here as affine schemes, i.e., the ideal generated by
these determinants is in general not reduced. The invariant, defined below, leads to another
invariant when passing to the radical of this ideal, but all properties listed below also hold
for this (coarser) invariant.
41
Proposition 2.49 Let (M, 1-, -1) be an affine Poisson variety of rank 2r and dimension d
and let p =
p(R, S) its invariant.
(i.) p(M) R'Sd(I + O(R-', S-')),

=
(2) If M is non-singular the coefficients pij of p E pijRSj, satisfy joij 0 for 2i > j.
= =
(3) For any r < d/2, there exists an affine Poisson variety whose invariant is RrSd.
Proof
Since M is irreducible and Md M we have Prd =
1, all other Mij have by definition
=
lower rank and being given as the intersection of hypersurfaces in M they also have lower
dimension. This shows (iL).
As for (2) we need to
rely on the symplectic foliation, described in Section 4 below; an
algebraic proof which would allow to remove the assumption about M being non-singular
is still missing (in view of Proposition 2.18 it would suffice to show that the irreducible
components of the Mi are affine Poisson subvarieties of M). Through every point of M
passes a leaf which inherits a symplectic structure from the Poisson structure, so on the one
hand all Hamiltonian vector fields at this point (which span a subspace of dimension equal
to the rank 2r of the Poisson structure at this point) are tangent to such a leaf, on the other
hand such a leaf is entirely contained in the subset M2,; thus every irreducible component
of M2, has dimension at least 2r showing (2).
For (3) take the canonical Poisson structure of rank 2r on C2d (Example 2.7). 0
Before we give a refinement of the invariant, let us consider some first examples.
Example 2.50 An affine Poisson variety is regular if and only if its invariant polynomial
is a monomial, i.e., is of the form R'S', where 2r is the rank and s the dimension of the
variety. In particular the invariant polynomial of the trivial structure on an afline Poisson
variety of dimension s is S'.
Example 2.51 For the Poisson structures on C2, which axe defined by a single polynomial
W(x, y) jx, yj, with W:A 0 we have p RS2 + kS, where k is the number of components of
= =
the plane curve defined by W(x, y) 0. Its invariant matrix is thus given by
=
0 k
( 0 0 0).
1
It follows in particular that the polynomial invariant is not a complete invariant: all non-
constant irreducible polynomials W(x, y) lead to a Poisson structure on C2 with invariant
p = RS2 + S.
Example 2.52 The Sklyanin brackets and their generalizations (see Example 2.9) lead for
the various values of the parameters to a lot of different invariant polynomials, giving an easy
proof that many of these Poisson structures are different. We give the different polynomials
-
which are easily computed -
in the following table (the integers i, j, k are taken different
and range from 1 to 3; a dash means that the values of the parameters are incompatible with
the relation alb, a2b2 + a3b3-
0)- :--
42
2. Alfine Poisson varieties and their morphisms
P all a = 0 ai =
aj = 0 ak = 0 all a : 0
all b = 0 s4 RS4 + 2S3 RS4 + S3 + S2 RS4 + S' + S
bi =
bj = 0 RS4 + 2S3 RS4 + 3S2
bi =
bk = 0 RS4 + 2S3 RS4 + S3 + S2
bi = 0 RS4 + S3 +,52 RS4 +,52 + 2S
bk = 0 RS4 + S3 + S2 RS4 + S3 + S RS4 + 2S2 RS4 + S2 + 2S
all b =k 0 RS4 +3S2 RS4 + s2 +2S RS4 + 4S
Table I
A more precise description of an affine Poisson variety can be given by combining the
above polynomial invariant with Proposition 2.38. We know from that proposition that
there corresponds to each point of the affine variety Spec Cas(M) a fiber whose irreducible
components are affine Poisson vaxieties. Then we may define a polynomial invariant p,(M)
for each c E Spec Cas(M) by
P,-(M) =
P (-7r-I (M) (C)
Cas
under the assumption that the fiber over c is irreducible; if not then the right hand side in
just replaced by the sum over all irreducible components. Thus we label
this definition is
each point of Spec Cas(M) by the invariant polynomial of the corresponding fiber over it and
obtain in this way a more sensitive invariant for affine Poisson varieties. In the examples
which follow we will only consider the fibers over closed points c.
Example 2.53 The simplest non-trivial example is given by the Lie-Poisson structure on
the dual of a three-dimensional semi-simple Lie algebra (see Example 2.8). A basis Ix, y, zJ
of this space can be chosen such that the corresponding Poisson matrix takes the form
0 -Z Y
( Z
-Y -X
0 X
0
. (2.22)
The algebra of Casimirs is clearly given by C[X2 y2 Z2] hence Spec Cas(M) can be ,
2
identified with C by evaluation on the element X2 Y Z 2; we denote the corresponding
coordinate by u. Since (2.22) has only rank zero at the origin, which lies in the fiber over
zero, we conclude that P = RS3 + 1 and
RS2 if U(c) :;:A 0,

Pc
RS2 +1 if U(C) 0.=
43
It may also be depictured as follows.
0
X U
RS2+1 RS2
Example 2.54 For the Heisenberg algebra the Lie-Poisson structure can be written as
jxj yj jx, zj
=
0, ly, zj x. As above one finds that the algebra of Casimirs is given
= =
by C[x], and again its spectrum can be identified with C (with coordinate u) by evaluation
on the Casimir x. The Poisson structure has now rank zero on the plane x = 0 which is an
entire level of the Casimirs (showing that equaJity in Proposition 2.38 needs not hold for all
level sets). It follows that p = RS3 + S2 and
PC
f RS2 if U (c) 0,
S2 if U(c) 0.
This case is depictured as follows.
0
X U
S2 RS2
Example 2.55 An interesting example is found by taking the Lie-Poisson structure on

gf(2)*. Consider the following basis
1 0 0 0 0
X =
( 0) 0 ,
Y=
(0 1), 0
Z=
0 0), 0
T=
(0 0),
1
for g and let x, . . .

,
t be the generators of 0 (Z), X, T. The corresponding Poisson
matrix is given by
0 Y -Z 0
-Y 0 X -
t Y
Z t -
X 0 -Z
0 -Y Z 0
and we have Cas(q*) C[x+t, xt-yz]. It follows that Spec Cas(g*) is in this case isomorphic
=
to C2 ; we pick the isomorphism. such that the standard coordinates u and v on C2 correspond
to x + t and xt -
yz (in that order). Since the rank of the Poisson structure is two except for
the points on the line y = z =
0, x =
t, we find that in this case p RS4 + S and =
RS2 if U2(e) 4v(c),

pe
RS2 +I if U2 (C) 4V (C).
44
Example 2.56 The following example will come up later when studying the Toda lattice
(Section VII.7). In terms of coordinates t6l for C' we consider the Lie-Poisson
structure determined by the Poisson matrix
0 -t2 t3
0 tT
(-T ) 0
with T
-t1
tj
t2
0 -t3
0
(2.23)
We will show later (in Paragraph VII.7.1) that CaS(C6) C[t1t2t3j t4 + t5 + t6], so that
=
Spec Cas (C6) can be identified withC2, with coordinates u and v, corresponding to t1t2t3
and t4 + t5 + t6 (in that order). By computing a few determinants one sees that,the rank is zero
on the three-plane tj =
t2 t3
=
0, two on the three four-planes ti
=
tj 0 (1 < i < j ! 3) = =
and four elsewhere. From it one easily obtains the following invariant polynomials:
p=R2,56 + 3R84 + S3,

R2S4 if U(c) 0,
PC ==
f3R 2S4 + 3RS3 + S2 if U(C) 0.
It is represented by the following diagram.
;3+S2
Proposition 2.57 Let (M, I., .1m) and (N, I* JN) be two affine Poisson varieties and let
their product M x N be equipped with the product bracket. Then
p(M x N) =
p(M)p(N).
In particular, if the invariant polynomial of an affine Poisson variety is irreducible then this
Poisson variety is not a product (with the product bracket).
Proof
We use as above Mi, Nj and (M x N)i as notation for the determinantal varieties as-
sociated to M, N and M x N respectively. The coefficients of the invariant polynomials
p(M), p(N) and p(M x N) are written as pi'., and pi2j
By Proposition 2.21, we have
pi'j.
(M x N)i U Mk x N1.
k+l=i
45
Using the fact that the irreducible components of a product are precisely the products of
irreducible components, we find
pixj #j-dim. irred. comp. of (M x N)j
E #j-dim. irred. comp. of Mk x N,

k+l=i
E 1: (#m-dim. irred. comp. of Mk) (#n-dim. irred. comp. of NI)

k+l=i m+n=j
1: 1: PklrnPin -
k+l=i m+n=j
This shows that p(M x N) =

p(M)p(N).
Remark 2.58 It would be interesting to determine the invariant(s) of the Lie-Poisson

structure of an arbitrary semi-simple Lie algebra and to relate it to the theory of (co-)
adjoint orbits.
46
3. Integrable Hamiltonian systems and their morphisms
In the study of semi-simple Lie algebras the notion of a Cartan subaJgebra plays a
dominant role. The corresponding object for affine Poisson spaces is an integrable algebra:
a maximal commutative (in this context called involutive) subaJgebra. An affine Poisson
variety with a fixed choice of integrable algebra is what we call an integrable Hamiltonian
system. The study of integrable Hamiltonian systems can be seen as a chapter in Poisson
geometry; for example we will see that all propositions which we proved for affine Poisson
varieties have their equivalents for integrable Hamiltonian systems.
Our definition is an adaption of the classical definition of an integrable system on a
symplectic manifold (see e.g., [AMI]) to the case of an affine Poisson variety. Notice that
we do not ask that the rank of the Poisson variety be maximal
(or constant). Another
difference is that the classical definition demands for having the right number of independent
functions in involution, while we ask for having a complete algebra (of the right dimension) of
functions in involution, completeness meaning here that this algebra contains every function
which is in involution with all the elements of this algebra. On the one hand this adaption
is very natural, it is even inevitable if one wants to discuss morphisms and isomorphisms of
integrable Hamiltonian systems. On the other hand it is not easy to verify completeness of an
involutive algebra, e.g., the (polynomial) algebra generated by a maximal number of functions
in involution needs not be complete. Accordingly we will also prove some propositions in this
section which will be useful for describing and determining explicitly the integrable
algebra
in the case of concrete examples.
3.1. Integrable Hamiltonian systems on affine Poisson varieties
Definition 3.1 Let (M, -1) be an affine Poisson variety. A subalgebra A of O(M) is
called involutive if JA, Al 0; we say that it is complete if moreover for any f E O(M)
one has f f, Al = 0 -#> f c- A. The triple (M, A) is called a (complete) involutive
Hamiltonian system -
Lemma 3.2 Let (M, A) be an involutive Hamiltonian system.

(i.) If A is complete then A is integrally closed in O(M);
(2) The integral closure of A in O(M) is also involutive and is finitely generated when
A is finitely generated.
Proof
The proof of (i.) goes in
exactly the same way as the proof of Proposition 2.46, replacing
Cas(M) by A and g O(M) by g Ei A. It is well-known that if A is finitely generated then
E
its integral closure in O(M) (defined as the set of all elements 0 of O(M) for which there
exists a monic polynomial with coefficients in A, which has 0 as a root) is also a finitely
generated algebra (see e.g., [AD] Ch. 5). To check that it is involutive, we first check that
every element of the integral closure of A is in involution with all elements of A. Thus, let 0
be an element of O(M) for which there exists a polynomial
p(X) = Xn + a1Xn-1 + - - -
+ an
for which P(0) = 0 and with all ai belonging to A; we that the polynomial isassume
of minimal degree. For any f E A the equality f P(o), f Jimplies as in the proof of= 0
Proposition 2.46 that 10, f I 0, upon using the minimality of P. Using this, it can now be
=
checked by a similar argument that any two functions in the integral closure are in involution.1
47
Every involutive algebra is contained in an involutive algebra which is complete, but the
latter is in general not unique. This is contained in the following lemma.
Lemma 3.3 Let (M, 1-, .1, A) be an involutive Hamiltonian system and denote by A the
integral closureof the field of fractions of A.
(3.) The subalgebra An o(m) of O(M) is also involutive;
(2) If A is complete then A n O(M) A; =
(3) A is contained in an involutive subalgebra B of O(M) which is complete; it is unique

if dim B dim A.
=
Proof
Recall (e. g., from [AD] Ch. 5) that A n o (m) can be identified as the set of elements 0 of
O(M) for which there exists a polynomial (which is not necessarily monic) with coefficients
in A, which has 0 as a root. if 0 E A n O(M) and
P(X) =
aoXn + aXn-I + - - -
+ an
is polynomial of minimal degree (with coefficients ai in A) for which P(O)

a 0, then =
implies 10, Al 0, using the minimality of P (again in the proof of

JP(O), Al 0 =
upon = as
Proposition 2.46). In turn this implies that if 0' is another element of An 0 (M) the equality
JP(O), O'l 0 leads to 10, O'l 0. Thus A n O(M) is involutive, showing (i.); from it (2)
= =
follows at once.
complete we pass to AO
If A is involutive but not A n O(M); if the latter is complete =
it is theunique involutive subalgebra of O(M) which contains A and is complete. If not, we

add ail element f E O(M) \ AO for which If, AO I 0 and repeat the above construction to =
obtain A,. Since dim A, dim AO + 1 we are done after a finite number of steps; because of
==
the choice of f the algebra which is obtained is not unique in general (interesting examples
of this are given below). 0
In this text only be interested in involutive algebras of the maximal possible

we will
dimension, given by proposition. We know from Lemma 3.3 that such an algebra A
the next
has a unique completion, which we will denote by Compl(A) (or by Complf fl, A I if A
is generated by If,, A 1) . . .
,
-
Proposition 3.4 Let (M, A) be an involutive Hamiltonian system. Then
1
dim A ::' , dim M -
Rkj-, .1. (3.1)

2
Proof
Consider a general fiber.F of the map M -+ SpecA which is induced by the inclusion
map A C O(M). By Proposition 2.37,
dim.F = dim M -
dim A. (3.2)
dim.F also equals the number of independent derivations of O(Y) at a general point of F and
involutivity of A implies that such derivations can be constructed using functions from A.
48
To see the latter, recall that the ideal of F is generated by the functions f -
X'-"(f) where
m E 97 is arbitrary but fixed and f ranges over A. For any g E A we have
Xg(f -
X-M) =
If, gj =
0,
hence X. is tangent to the locus defined by the ideal of F, i.e., to Y and we can construct
derivations of O(Y) using elements of A. Next we show that the elements of A lead to
dim A -
dim Cas(M) independent derivations, giving a lower bound for diM.F. Consider a
nested sequence of subalgebras
Cas =
Ao C Ai C A2 C ...
c A, =
O(M),
where dim Aj+j dim A, + 1, in particular r

=
Rkj 1. If ni denotes the number of
= - -
independent vector fields on M coming from A, (i.e., having independent vectors at a general
point) then obviously ni < ni+l :5 ni + 1, no 0 and n, r. It follows that ni i for all i.
= = =
It gives the following lower bound
dim.F > dim A -

dim Cas (M). (3.3)
Combining (2.40), (3.2) and (3.3) we find
I
dimA < (dim M + dim Cas (M)) < dim M -
Rkj-, .1. (3.4)

2
We finally get to the definition of an integrable Hamiltonian system (on an affine Poisson
variety).
Definition3.5 If (Mj-,-j) is an affine Poisson variety whose algebra of Casimirs is max-
imal and A is a complete involutive subalgebra of O(M) then A is called integrable if
dimA = dimM -
1Rkj-,
2
(3.5)
The triple (M, A) is then called an integrable Hamiltonian system and each non-zero
vector field in
Ham(A) =
JXf I f E Al
is called integrable vector field. The dimension of A is called the dimension or the degrees
an
of freedom of the integrable Hamiltonian system. M is called its phase space and Spec A its
base space.
If A, and A2 axe two different subalgebras of 0 (M) which make 0 (M) into an integrable
Hamiltonian system then every non-zero vector field in the intersection Ham(AiL) n Ham(,42)
is called a super-integrable vector field.
49
Remarks 3.6
1. What we call an integrable vector field is in the literature often called an integrable
system; the distinction we make is motivated by the fact that the datum of one integrable
vector field Xf (or its corresponding Hamiltonian f) does not suffice in general to determine
A (see Examples 3.10 and 3.11 below).
2. In view of (3.4) the condition that the algebra of Casimirs is maximal follows
from (3.5); it was added in the hypotheses to stress that it is a condition on the Poisson
structure -
in our approach affine Poisson varieties whose algebra of Casimirs is not maxi-
maJ do not admit integrable Hamiltonian systems.
3. Completeness of the integrable algebra A implies that Cas(M) c A and A can be
seen as an intermediate involutive object between Cas(M) and O(M); for example, it follows
from (3.4) and (3-5) that
dim A I(dim M + dim Cas(M)),

2
which supports this assertion.
The commutative triangle of inclusions
OM
I
\
A -
Cas(M)
induces, as explained in Paragraph 2.4, the following commutative triangle of dominant (Pois-
son) morphisms.
M
"ro-(M
0",(M,
-7rA
Spec A -
7r
-
Spec Cas(M)
Thus the parameter map irc , m, which maps the phase space to the parameter space, can be
factorized via the map 7rA : M -+ Spec A from the phase space to the base space; we call the
latter map the momentum map. The irreducible components of the fibers of the momentum
map axe affine varieties which will play a dominant role in this text. We call them
the level
sets of the integrable Hamiltonian system or the level sets of A for short.
We now come technicallity alluded to at the beginning of this section. We know

to the
from Lemma 3.3 abstractly how to complete an involutive algebra A (say of the maximal
possible dimension), but it does not lead to an explicit description of the completion when
studying concrete examples. The following proposition gives sufficient and checkable condi-
tions for such an algebra A to be complete; it will be used several times when we get to the
examples.
50
Proposition 3.7 Let (M, I., J) be an affine Poisson variety and let A be an involutive
subalgebra of O(M) of dimension
dimA = dimM -
IRkj-, J.
2
Then A is complete, hence integrable, if the fibers of 7rA : M -+ Spec(A) have the following
two properties
(i.) the general fiber is irreducible;
(2) the fibers over all closed points have the same dimension.
Proof
Let us suppose that A is not
complete, i.e., f 0 A and If, Al 0 for some f E O(M). We =
denote by A! the
algebra generated by f and the elements of A, which has by Proposition 3.4
the same dimension as A. By Lemma 3.3 f belongs to the integral closure of the quotient
field of A. Thus f Ei O(M) is a root of a polynomial Q(t) E A[t]. Consider the following
commutative diagram which is induced by the inclusion A C W.
IrAl
Spec A! Spec A
If Q(t) has degree at least two then z is a ramified covering map of degree at least two, hence
the fiber of -7rA over a general point P has at least two components, which axe the fibers
of -7rA, over the antecedents %--l (P). This is in conflict with assumption (L), hence Q(t) is of
degree one, Q (t) Since f E 0 (M) \ A neither p, nor P2 are constant. Therefore
::::::
P1 t + P2 -
there is a closed Spec A which corresponds to an algebra homomorphism onto C

point P in
which sends both p, and P2 to 0. This closed point is the image under z of a point which
is not closed, namely the corresponding algebra homomorphism. can take any value on f.
Then the fibers of 7rA, over these points have dimension one less than the dimension of the
fiber 7r. '(P)
which is dim M dim A by assumption (2). Since A! has the same dimension
-
as all fibers of 7rA, have dimension at least dimM

A, -
dimA, a contradiction. It follows
that A is complete. I
We have seen in Proposition 2.38 that all Hamiltonian vector fields Xf, f E O(M) are
tangent to all fibers of the parameter map. Similarly we show now that all integrable vector
fieldsXf, f E A are twigent to all fibers of the momentum map; in addition they have the
special property to pairwise commute.
Proposition 3.8 Let (M, 1., -1, A) be an integrable Hamiltonian system. Then all Hamilto-
nian vector fields in Ham(A) are tangent to all fibers of the momentum map 7rA : M -+ Spec A
and they all commute; the irreducible components of these fibers are affine varieties and the
dimension of the general fiber is 12 RkI., -1, which coincides with the number of independent
vector fields in Ham(A).
51
If (M, A,) is another integrable Hamiltonian system, then super-integrable vector

fields in Ham(A) n Ham(Al) are tangent to the (strictly smaller) intersection of the fibers of
the corresponding maps irA and irA,
Proof
Let g E Ham(A). f E Ham(A) we have X,f
Then for any if, gJ 0, hence X, = =
is tangent Clearly these fibers are affine varieties and commutativity of

to all fibers of 7rA.
the vector fields in Ham(A) follows from item (3) in Proposition 2.3. The dimension of a
1 in view of Proposition 2.37. Our claim about
general fiber is dimM dimA 2 RkJ-, .1
- =
super-integrable vector fields follows at once from the first paxt of the proposition. 0
We now get to some first examples of integrable Hamiltonian systems, in particulax we
will give two examples of a super-integrable vector field.
Example3.9 If (M, I., -J) is anaffine Poisson variety of rank two whose algebra of Casimirs
is maximal, then any function F which does not belong to Cas(M) leads to an integrable
Hamiltonian system. Namely A ComplICas(M), F1 is obviously involutive and dim A
= =
dim Cas (M) + I dim M 1, hence A is integrable; clearly its level sets are just algebraic
= -
curves.
This well-known fact is often expressed by saying that in one degree of freedom all
Hamiltonian systems are integrable (although the condition that the algebra of Casimirs
should be maximal is never stated explicitly; when assuming implicitly that M has dimension
two this condition is of course automatically satisfied).
Example 3.10 Another trivial class of integrable Hamiltonian systems is defined on Cn,
with a regular Poisson bracket, by considering linear functions; the example shows that the
integrable algebra is not always determined by just one of its (non-trivial) elements. For
simplicity let us take the case n 4 with a constant Poisson structure of rank 4. As we know
=
from Example 2.7 lineax coordinates q, p, q2 P2 on C4 may be picked such that Jqi, pj I
I I i 6ij =
and Jq1, q2} Jp1 P21

= 0- Take F
7 aql + bq2 + CP1 + dP2 with e.g. a =A 0 and look for a
= =
linear function G alql + b'q2 + 41 + dIP2 which is in involution with F. Replacing G by

=
G -
Fa'/a if necessary we may assume that a' == 0 and we find
G =
b1q2 + (db' -
bd)pl + dP2
as the most general solution (up to adding multiples of F). Here Y, d' EE C are arbitrary, so
that we essentially a one-paxameter family of possibilities for G (paxametrized by d1b'),
have
all leading to an integrable subalgebra A of O(C4) The Poisson bracket of two of these .
possibilities for G, is given by
Jb'q2 + (dY -
bd)pl + 421 Vq2 + (db" -
bdll)Pl + dIP21 = Yd' -

db"
which is non-zero if they are different (i.e., non-proportional), in agreement with Proposi-
tion 3.4. The general fiber of A is in this just a plane and all integrable vector fields
case
Ham(A) are constant when restricted to each plane. Clearly, all these vector fields axe super-
integrable and their flow evolves on a (straight) line.
52
The above examples are the most trivial classes of examples of integrable Hamiltonian
systems -
apart from the really trivial class where affine Poisson spaces of rank zero are
considered. To increase complexity one may consider Poisson structures which are of higher
rank and not constant (in particular they are never regular), also polynomials of higher degree
may be considered and ambient affine
variety of higher dimension. It turns out that in these
an
cases it is a non-trivial matter to find

integrable Hamiltonian systems. There are of course
some trivial ways to obtain new systems from old ones, one
may for example take the product
of two integrable Hamiltonian systems or rewrite a simple system in a complicated
way by
changing variables (see Section 3.3), but these results are in reality often only interesting in
the other sense, namely for reducing large or complicated integrable Hamiltonian
systems
to smaller or simpler ones. A general scheme for either constructing
integrable Hamiltonian
systems or for deciding whether a given Hamiltonian vector field is integrable is not known.
We will come back to this in Chapters III and VI.
Example 3.11 Let E be a compact oriented topological surface of genus g > 1 with
fundamental group 7r, (E) and let G be a reductive algebraic group. Then Hom (7r, (E), G) is
an affine variety on which G acts by conjugation, more precisely if
p : 7r, (F,) --+ G and g E G
then g-p is the homomorphism ir,(E) -+ G defined by
g-P M =
g(P(O)g
for C E ?r,(E). It turns out (see [Gol]) that the quotient
M =
Hom(7r, (F,), G) IG
(which is an affine
variety since G is reductive) has a natural Poisson structure which can
very explicitly be described for the classical groups. For simplicity let us consider the case
G = SL (n) in the standard representation. For a curve C G 7ri (E) the function
fc : M -+ G : p j-+ T ace(p(C))
is a well-defined regular function on M and it can be shown that these functions generate
O(M). It was shown by Goldman (see [Gol]) that on such functions a Poisson bracket of
maximal rank is given by
I fc, fc, I (p; C, C) fc, C', -
fc fc, (3-6)
n
PEC#C1
The sum runs over the intersection points of C and C' (one may suppose that the curves
intersect transversally) and e(p; C, C') is a sign which is determined by the way the (oriented)
curves C and C' intersect at p, upon using the orientation of E. Finally, CpCp' is the curve
on E, based at p which is obtained by first following C and then following C'.
A large involutive
algebra for this bracket is obtained
as follows. E can be decomposed (in
several ways) into so-called

trinions; a trinion, also called a pair of pants, is just a three-holed
sphere and such a decomposition will consist of 2g 2 trinions (in the case of genus two there
-
exist precisely two such decompostions) Each trinion being bounded by three curves (which
are identified two by two) one gets 3g 3 curves on E and what is important here is that they
-
are non-intersecting. Calling these curves C, Gg-3 we find from Goldman's formula (3.6)
...
I I
that the functions fo are in involution; thus one obtains an involutive algebra
......
53
A =
Compllfc...... fc,,, -j and its dimension is computed to be 3g -
3. Since the rank of
the Poisson bracket is maximal, A will be integrable if and only if
1 1
3g -
3 = dimM -
2
Rkj-, -1 =
2
dimM,
i.e., for dim M 6g

= 6. Since iri(E) has a system of 2g generators, which are bound
-
by one relation, dim Hom(iriL (E), G) has dimension (2g 1) dim G, hence M has dimension -
(2g 2) dim G and A is integrable if and only if

-
6g -
6 =
(2g -
2) dim G,
i.e., for dimG = 3. Since we restricted ourselves to G =

SL(n) we find that A is only
integrable for G SL(2); it is clear from the above pictures that the
= Hamiltonian vector
fields corresponding to all functions fc, are actually super-integrable.
3.2. Morphisms of integrable Hamiltonian systems
In parallel with our discussion of morphisms of affine Poisson varieties we now turn to
morphisms of integrable Hamiltonian systems.
Definition3.12 Let and (M2&,'j2,A2) be two integrable Hamiltonian

-+ (M2ij*i'j2iA2) is a morphism 0: M,
systems, then a morphism 0: (Mj,j-,-jj,Aj)
M2 with the following properties
(j-) 0 is a Poisson morphism;
(2) 0* CaS(M2) C Ca$(MI);
(3) O*A2 CAI -
Schematically, regularity of the map and (2) and (3) can be represented as follows:
Cas(M2) - A2 O(M2)
0. 0* 0* (3.7)
Cas(MI) -
--------
Al - O(Mi)
A morphism 0: (M17j*7"j1iA1) -+ (W J'i *12, A2) which is biregular has an inverse which
is automatically a morphism: we call such a map an isomorphism (it forces all inclusion maps
in the diagram to be bijective).
Ikom the very definition it is clear that the composition of two morphisms is a mor-
phism (hence we have a category). It is also immediate that for any biregular map 0 :
M, -+ M2 and for any integrable Hamiltonian system (MI, 1- 7 -11, A,) there exists a unique
Poisson bracket 1. -12 on M2 and a unique integrable algebra A2 C O(M2) such that
0: (Ml J* i'll IAI)
7
-+ (M2 I i
*
i j 2 A2)
1
is an isomorphism; explicitly A2 A, and
Ifi 912 (0-1)* 10*f7 0*911 Vig G O(m2)-
54
Conditions (i.) and (2) axe conditions at the level of the Poisson structures, rather than
on the level of the integrable algebras. Condition (2) resp. (3) implies that 0 induces a mor-
phism of the corresponding paxameter spaces resp. base spaces, as is shown in the following
proposition.
Proposition 3.13 Let 0: (MI, 1-, -11, A,) -+ (M2i I' *121 A2) 1
be a morphism of integrable
Hamiltonian systems. Then 0 induces a morphism
0: Spec Cas(MI) -+ Spec Cas(M2)
which makes the following diagram commutative,
M, M2
7rc-(Ml)I I7rC-(M2)
Spec Cas(MI) Spec Cas(M2)
as well as a morphism
: Spec A, Spec A2
which makes the following diagram commutative.
M, M2
IrAjI I-A2
Spec A, Spec A2
If 0* Cas(M2) =
Cas(MI) (resp. O*A2 =
Aj then (resp. ) is injective.
Proof
The first assertions are immediate from diagram (3.7) by taking spectra; also surjectivity
of 0* implies injectivity at the level of the corresponding spectra. 0
Saiddifferently, condition (3) in Definition 3.12 implies that each level set of A, is
mapped a level set Of A2 and if O*A2
into A, then different level sets of A, are mapped
into different level sets of A2; condition (2) can be given a similax interpretation. We further
illustrate the meaning and relations between the three conditions in Definition 3.12 in the
following examples and propositions.
55
Example 3.14 Let us show that in Definition 3.12 neither (2) nor (3) follow from (1).
Consider C4 (with coordinates q1, q2 P1 P2) with the canonical Poisson structure Jqi,
7 i pj I =
8ij, Jqi, qj I fpi, pj I

=
0, and C-3 (with coordinates q1, q21 PI) with Jq1, p, I
= I and q2 =
as Casimir. We look at this C3 as the qlq2PI-plane in C4 and denote by 0 the projection
map along P2. Then 0 is a Poisson morphism, however O*q2 is not a Casimir of C4 showing ,
that (3.) does not imply (2). Notice that in this case 0 does not induce a map 0 as in
Proposition 3.13. Taking two different functions on C2 (i.e., the algebras generated by them)
shows that (i.) does not imply W-
There is however a of morphisms for which condition (2) in Definition 3.12

large class
follows from (i.), namely that of
universally closed morphisms; these include the proper
morphisms and, in particular, the finite morphisms (see [Har] pp. 95-105). We prove this in
the following proposition, however we restrict ourselves to the case of finite morphisms, since
we will only use the result in this case (the proof however generalizes verbatim to the case of
universally closed morphisms).
Proposition 3.15 Let (MI, -11) and (M2, J*)'12) be two affine Poisson varieties and
suppose that 0 : M, -4 M2 is a finite morphism (for example a (possibly ramified) covering
map). If 0 is a Poisson morphism then 0* Cas(M2) C Cas(MI); if 0 is moreover dominant
then Cas(Mi) is the integral closure of 0* Cas(M2) in O(Mi).
Proof
Let us show that if 0 is finite then for any f E Cas(M2), O*f is in involution with all
elements of O(MI). The main property which is used about finite (or universally closed)
morphisms is that if 0: M, -+ M2 is such a morphism then O(MI) is integral over O*O(M2).
Thus any element g E O(MI) is a root of a monic polynomial P (of minimal degree) with
coefficients in O*O(M2). As in the proof of Proposition 2.46 we find
0 =
10*f, PWI =
P,(g)lo*f gI ,
where P' denotes the derivative of the polynomial P. By minimality of P we find JO*f gJ ,
= 0
as desired. We have shown that 0* Cas(M2) C Cas(MI).
Next take an element g E Cas(MI) and call P its polynomial as above, with coef-
we
O*O(M2). We show that P has actually its coefficients in 0* Cas(M2), thereby

ficients in
proving that Cas(MI) is the integral closure of 0* Cas(M2). To do this, let O*f E O*O(M2)
be arbitrary, then
0 =
10*f P(g)11
,
=
JO*f, g' + O*alg'-' +... + 0 *a.11
10*f, O*alllg'-' + + 10*f, O*a.11
O*Jf,a1J2gn-1 +'*'+ O*Ifi anJ2-
Since this polynomial has its coefficients in O*O(M2) and since P was supposed of minimal
degree, we find that 0* If, ai I 0 for all i. Since 0 is dominant it follows that If, ai
= 0
for all f E O(M2), so that ai E Cas(M2) for i 1, n. =
- -
.'
0
56
It can be seen in a similar way that if 0 : (MI, {-, -11, A,) -+ (M21 A2) is a
morphism of integrable Hamiltonian systems which is finite and dominant then A, is the
integral closure Of O*A2 in O(Mi) (for a proof, use completeness of A,). It leads to the
following corollary.
Corollary 3.16 Let (MI, I j 1, A,) -+ (M2 1 21 A2) be a morphism which is finite and
whose image is an affine subvariety of M2. Then 0 is the composition of an injective and a
su7jective morphism.
Proof
We know from Proposition 2.16 that, as a Poisson morphism, 0 can be decomposed via
(O(Mi), 1-, -1) say 0 s o . Define

=
A =
If E O(O(Mi)) I *f E Ai}
For f,g E A we have *Jf,gj J *f, *g}
=
0; by injectivity of * we see that A is
=
involutive. If If, A} = 0 thenJ *f, A,} 0 since A, is the integral closure of O*A in O(MI).
=
Then *f E A, by completeness of A, and A is also complete. Finally the dimension count

for O(MI) is the same as the one for M, since 0 is finite. It follows that (O(MI), 1-, -1, A) is
an integrable Hamiltonian system. Clearly % and axe morphisms of integrable Hamiltonian

3
systems.
Example If a Poisson morphism 0 : (MI, l'I'll) -+ (W 1'7 *12) is finite but not
3.17
dominant then Cas(M.1) may be larger than the integral closure of 0* Cas(M2) in O(MI).
Take for example for (M2, J* '}2) the Lie-Poisson structure for the Heisenberg algebra (Ex-
ample 2.54), for M, the plane x 0 with the trivial Poisson structure and for 0 the inclusion
=
map. Then Cas(M2) C[xl hence 0* Cas(M2) C, while Cas(MI) O(MI).

= = =
Example3.18 Even if a Poisson morphism 0: (Mill* 7,11) -+ (W J*,*12) is finite and

dominant then Cas(MI) may be different from 0* Cas(M2). Take for example on C3 the
Poisson structure from Example 3.14 and consider the finite covering map 0 : C3 -+ C' given
by O(ql, pl, q2) (qj, pl, q22). Obviously this is a Poisson morphism; however the Casimir q2
=
is not of the form O*F for any function F E O(C3). Notice that C -+ C is in this
2
case not injective, being given by (q2) =
q2 .
A similar remark applies to condition (3) in
Definition 3.12.
3.3. Constructions of integrable Hamiltonian systems
In Section 2.3 we gave several constructions to build new affine Poisson varieties from old
ones. Using these we now give the corresponding constructions for integrable Hamiltonian
systems on them. We first show that an integrable Hamiltonian system restricts to a general
fiber of the parameter map.
Proposition 3.19 Let (M, A) is an integrable Hamiltonian system and T an irre-

ducible component of a general level of the Casimirs. Then (,F, f -, JI.F, is an integrable AI.F)
Hamiltonian System and the inclusion map is a morphism. The property also holds for the
general levels of any subalgebra of the Casimirs.
57
Proof
Let B be anysubalgebra of Cas(M) and let Y be an irreducible component of a general
fiber of M Spec B. We know already from Proposition 2.38 that Y has an induced Poisson
-+
structure and from Proposition 2.42 that the algebra of Casimirs of this structure is maximal.
If we restrict A to Y then we get again an involutive algebra Ap which is complete since A
is complete and Y is general. Thus it suffices to compute the dimension of A,77,
I
dimY -
dim A dimM -
dim B -
(dim A -
dim B) =
RkJ-, .1 RkJ-, -I.F.
2 2
This shows that Ay is an integrable algebra. Clearly the inclusion map is a morphism. 2
Definition3.20 Any integrable Hamiltonian system obtained from (M, A) by Propo-

sition 3.19 is called a trivial subsystem.
One may think of a trivial subsystem as being obtained by fixing the values of some of
the Casimirs.
Example 3.21 In the examples one has however to be careful when picking a particular
fiber.F (i.e., in the choice of values assigned to (some of) the Casimirs). Namely one has to
check that F is general enough in the sense that both the dimension and rank of Y coincide
with those of a general fiber. The dimension of a special fiber F may be higher and/or its
rank may be lower; then
dim.F -
Rkf-, .1y > dhnA dimAly,
so (F, AI.F) is not an integrable Hamiltonian system. Reconsider e.g. Example 2.54:
none of the Hamiltonian systems
integrable on C' for this Poisson structure will lead to an
integrable Hamiltonian system on the fiber x =

0, since the induced Poisson structure on
that fiber is trivial, while Al, 54 0(.F)
r
Proposition 3.22 For i E 11, 21 let (Mi, I., Ji, A,) be an integrable Hamiltonian system
and let -7ri denote the natural projection map M, x M2 -+ Mi Then
(MI X M27 f'i -1m, xm2,-7r,*Al 0 *7r2*A2) (3.8)
is anintegrable Hamiltonian system and the projection maps 7ri are morphisms. Each level
set of the integrable Hamiltonian system is a product of a level set of (MI, f -, -11, A,) and a
level Set Of (W J* i'}27 A2)-
Proof
The Poisson-part of this proposition was already given in Proposition 2.21. As for
involutivity,
firi Ai (2) 7r2* A2 7r,*1 Al

,
(9 7r;2 A2 I mi . m, -"::::
7ri*
1 J& A111 + 1r*2JA2,
2 A212 0-
58
We count dimensions:
dim -7r,*Al 0 7r2*A2 = dim A, + dim A2

1 1
= dim Mi -
RkJ-, -11 + dim M2 -
Itkf* 1'12
2 2
=
dim(Mi X M2) Rk Q -, JM1 xM2)
2
Since ?r,*Al (8) ?r2*A2 is complete and involutive with respect to the product bracket, this
computation shows that 7r,*Ai (8),7r2*A2 is integrable. Since for earch of the projection maps iri
one has -7ri*Ai C 7r1*A1 0 7rM2, these projection maps are morphisms. The fibers of the
momentum map are given by the fibers of M, x M2 -+ Spec(7r,*Al 0 lr2*A2), that is, of the
product map M, x M2 -+ Spec A, x Spec A2 hence all fibers are products of level sets of A,
and A2. I
It is easy to show in addition that Ham(-7r,* A, (9 7r2* A2) contains a super-integrable vector
field if Ham(Ai) (or Ham(A2)) does.
Definition3.23 We call (3-8) the product of (M1,J-,J1,A1) and (M2,J',*}2,A2)-
A construction which is related to (but different from) the product construction and
which will be used several times in the next chapters, is obtained when dealing with integrable
Hamiltonian systems which depend on parameters. By this we mean that we have an affine
Poisson variety (M, I , J) and for all possible values c of a set of parameters we have an
-
integrable algebra A, on it. This set of parameters is assumed here to be the points on an
affine variety N and we assume that A, (i.e., its elements) depends regularly on c. Then we
can build a big affine Poisson variety which contains all the integrable Hamiltonian systems
(M, 1., .1, A,) as trivial subsystems. This is given by the following proposition.8
Proposition 3.24 If N is an for each c r= N an integrable Hamilto-

affine variety and
nian system (M, I., Jm, A,.), depending regularly
is given on an affine Poisson variety
on c
(M, 1-, .1) then M x N has a structure of an affine Poisson variety (M x N, I-, J) and
O(M x N) contains an integrable subalgebra A such that each (M, I-, Jm, A,) is isomorphic
to a trivial subsystem of (M x N, 1-, -1, A) via the inclusion maps
0,: M -+ M x N: m i-+ (m,c).
Proof
For N one takes the trivial structure so that Cas(N) =
O(N) which makes M x N into
a Poisson manifold. The algebra of Casimirs on this product is maximal since the one on M
is maximal and Cas(MN) Cas(M) (9 O(N). The fact that A, depends regularly on c
x =
means that there exists a subalgebra A of O(M x N) which restricts to A, on the fiber over c
of the projection p, : M x N -+ N. Clearly its dimension is given by dim A dim A, +dim N =
8
generalizes to the situation considered in Example 2.24, namely when
The proposition
ir : P -+ N is a morphism, for each n E N, I-, -In is a Poisson bracket on the
dominant
fiber -7r(-) (n) and An is an involutive subaJgebra of 0 (-7r(- 1) (n)) which is integrable for
general n; both I-, Jn and An axe supposed to depend regularly on n G N. Proposition 3.24
corresponds to the special case P = M x N considered at the end of Example 2.24.
59
that dim A 1
so =
dim(M x N) -
2 Rkf
since A is complete and involutive it is integrable.
Since O(N) is a subalgebra of Cas(M x N) the fiber over p is a level set of the Casimirs and
the restriction of the Poisson structure corresponds to the one on M via the morP hism.
which is an isomorphism when restricted to such a fiber.
The next construction we discuss is that of taking a quotient. This is of interest, because
many of the classical integrable Hamiltoniau systems possess discrete or continuous symmetry
groups. The algebraic setup which we use here has the virtue to allow to pass easily to the
quotient (one does not need to worry about the action being free, picking regular values and
so on).
Proposition 3.25 Let G be a finite or reductive group and consider a Poisson action
X: G x M, where (M, 1-, -1) is an affine Poisson variety. If A is an involutive algebra
M -+
such that for each g (=- G the biregular map X, : M -+ M defined by

X, (m) X(g, m) leaves =
A invariant, i.e., X*A C A, then (MIG, j.'.10, AG) is an involutive Hamiltonian system
9
and the quotient map -7r is a morphism. Here 1., -10 is the quotient bracket on MIG given by
Proposition 2.25. If G is finite then (MIG, f.,.}O, AG) is integrable.
Proof
Involutivity of AG is immediate from Proposition 2.25. Suppose now that G is finite.
Then completeness of A implies completeness of A n O(M)G. As for dimensions, since G is
a finite group we have
dimAn O(M)G = dim.A
= dimM -
1Rkf
2
-, -1
I
= dim M/G -
Rkf -, -jo,
2
where we used in the first equality that dim O(M)' dim O(M) and A c O(M). Similarly
=
one shows that the algebra of Casimirs is maximal, being given by Cas(M) n O(M)G. Thus
A n O(M)G is integrable; obviously -7r*(A n O(M)G) C A, hence the quotient map is a
morphism. 0
We will encounter a lot of examples later. Here are some first observations.
Example 3.26 A special case occurs when A C O(M)G (which implies Cas(MIG) c
O(M)G). Namely, in this case each level set of (M, f -, J, A) is stable for the action of G and
the level sets of (MIG, I -
, j o A)
5
are precisely the quotients of the level sets of (M, f - -
1, A).
,
A similar result applies for the level sets of the Casimirs in case Cas(MIG) C O(M)G.
Example 3.27 The quotient construction leads to a lot of new integrable Hamiltonian
systems which look interesting. One may e.g. start with an integrable Hamiltonian system
(M, I-, -}, A) and consider its (M
M, I-, -Imxm, A (9 A). The group Z2 acts on
square x
M x M by interchanging the factors in the

product. Obviously this is a Poisson action
and the action leaves A (& A invariant, thereby leading to a quotient. The level sets which
correspond to the diagonal are symmetric products of the original level sets.
60
Notice that the group G in Proposition 3.25 can be seen as a subgroup of the automor-
phism. group of M. For future use we introduce also the slightly more general notion of a
quasi-automorphism.
Definition3.28 (A I-, J,A) bean integrable Hamiltonian system. An automorphism

Let
is an -+ (M, I-, -}, A). More generally, if 1., -11 and J* *12 are two
isomorphism (M, I-, -}, A) 1
Poisson brackets on M then an isomorphism (M, -.11, A) -+ (A {-, '12, A) is called a

quasi-automorphism.
The final construction is to remove a divisor from phase space.
Proposition 3.29 Let (M, 1-, -1, A) be an integrable Hamiltonian system and let f E O(M)
be a function which is not constant. Then there exists an integrable Hamiltonian system
(N, f"i'lN, AN) and a morphism (N, J* 7'IN7 AN) -+ (M, 1-, -1, A) which is dominant, having
the complement (in M) of the zero locus of f as image.
Proof
proof (the Poisson part) was given in Proposition 2.35 and we
Most of the use the
notation of that proposition. We start with the case f E A. If we define AN :--
7r*A[t]
then AN is involutive since 7r is a Poisson morpbism and it has the right dimension in order
to be integrable. We need to verify completeness. Let Ein-0 fiti EE O(N) then j-
n n
i=O
fit', AN
IN 0 :> Effi, 7r*A[t]lNti
i=O
n
0
Effii lr*AlNfn-i 0
i=O
n
1:1& Alfn-i 0
i=O
E ffn-i, A 0
i=O
n
E fjn-i E A
i=O
n
1: ffn-itn G AN
i=O
n
1: fit' CE AN-
i=O
Since AN is involutive the last line also implies the first line, so we have established the
desired equivalence.
If f an explicit description Of AN is still available if (M, I
A then J,A) satisfies -
the conditions of Proposition 3.7. In that case the fibers of N Spec 7r*A also satisfy the
conditions of Proposition 3.7 hence -7r*A is complete and AN 7r*A. In general one has
AN Compl(-7r*A) and a more explicit description is not available.
=
61
3.4. Compatible and multi-Hamiltonian integrable systems
We now introduce a few concepts which relate to compatible integrable Hamiltonian

systems.
Definit ion 3.30 Let i =-=

1, n be n (linearly independent) compatible Poisson
brackets on an affine variety M. If (M, I-, ji, A) is
integrable Hamiltonian system for each
an
i 1, n then these systems axe called compatible integrable Hamiltonian

=
. . .
, systems. Any
non-zero vector field Y on M which is integrable (in particular
Hamiltonian) with respect to
all Poisson structures i.e., for which there exist fl, f,, E A such that . . .
,
Y =
f., fill =
...
=
1', Aln,
is called a multi-Hamiltonian (bi-Hamiltonian if n 2) vector field, since it is Hamiltonian in

=
many different ways; any of the integrable Hamiltonian systems (M, I-, ji, A) is then called
an integrable multi-Hamiltonian system (bi-Hamiltonian when n =
2).
Remark 3.31 We do not demand in the definition of an integrable multi-Hamiltonian

system that all the integrable vector fields be multi-Hamiltonian. Although this condition is
satisfied in Examples 3.33 and 3.34 it is far too restrictive in general.
All propositions and basic constructions given above are easily adapted to the case of
compatible or multi-Hamiltonian structures, but this will not be made explicit here. Just
one example: an action of a reductive group which is a Poisson action with respect to both
Poisson structures of two compatible integrable Hamiltonian systems yields on the quotient
two compatible integrable Hamiltonian systems. Here are some properties which are specific
to compatible integrable Hamiltonian systems.
Proposition 3.32
(1) Compatible integrable Hamiltonian systems have the same level sets;
(2) The Poisson brackets of compatible integrable Hamiltonian systems have the same
rank, which also equals the rank of a general linear combination of these Poisson
structures
(3) If (M, I., -1j, A) are compatible integrable Hamiltonian system then for ageneral
linear combination I-, +x of the Poisson structures the system (M, A) is an
integrable Hamiltonian system.
Proof
The proof of (l.) is obvious since the level sets by A only. Since Rkf ji
are determined
2 dimM-2 dimA we find that the rank ofall structures
equal. To determine the rankI., ji is
of a linear combination of these structures one looks at the corresponding Poisson matrix (with
respect to a system of generators of O(M)) which is given by the same linear combination of
the Poisson matrices of the structures I-, ji. Now a general linear combination of invertible
matrices is invertible, which applied to a non-singular minor of size Rkj-, ji leads to (2).
For a linear combination I-, .1,\ of (maximal) rank Rkj-, jj one has that JA, A},\ 0 and =
dimA dimM 1L
= -
2 Rkj-, ji, hence (M, I-, +\, A) is an integrable Hamiltonian system,

showing W-
62
We will encounter in this text many (non-trivial) examples of compatible integrable

Ha,miltonian systems and of integrable multi-Hamiltonian systems. Here are two simple
examples of integrable bi-Hamiltonian systems.
Example 3.33 Consider the Poisson structures 1-, -11 and 1' J21
on C4 (with coordinates
qj, q2, p, and P2) defined by the Poisson matrices
0 0 1 0 0 0 0 1
0 0 0 1 0 0 1 0
and
-1 0 0 0 0 -1 0 0
0 -1 0 0 -1 0 0 0
For A c O(C4) take those functions which are independent of q, and q2. Then both Poisson
structures are compatible and since their integrable vector fields are of the form
a C9
I f
9ql
+ g
9q2
1 f,g E A
they are all bi-Hamiltonian.
Example 3.34 Recall from Example 2.11 that the matrix

OF -OF
0
Oz OY
U( OF
OF
5
;9__V
-OF
0
TX_
OF
Ox
0
defines for any u and F in O(C') a Poisson structure on C3 F is assumed non-constant

here in order to obtain a non-triviaJ Poisson structure. Let us denote this Poisson structure
by J* juF.
1
If G is any other non-constant element of O(C3) then I-, Ju,F + l'i"ju,G __"
j','ju,F+G hence1' 1 *}u,F and J* , ju,G are compatible and, assuming that F and G are in-
dependent, A ComplIF, G} defines an integrable Hamiltonian system on (C3, J* ju,F)-
=
,
However, by interchanging the roles of F and G. we find that A also defines an integrable
Hamiltonian system on (C3, J* ju,G) hence leading to a pair of compatible integrable Hamil-
, I
tonian systems. Since moreover the Hamiltonian vector fields with respect to both Poisson
structures are given by
fuoVF x VG 10 c Al
we conclude that A defines an integrable bi-Hamiltonian system on C3.
Closely related to the concept of an integrable multi-Hamiltonian system is that of a

multi-Hamiltonian hierarchy. Let us define this in the case of a bi-Hamiltonian hierarchy and
explain its use. Let 1-, -11 and J",'}2 be two compatible Poisson brackets on M. Then a
sequence of functions jfj I i E ZI is called a bi-Hamiltonian hierarchy if
I-, fiJ2 --::

I' fi+111i
i (i E Z).
The following property is essentially due to Lenaxd and Magri.
63
Proposition 3.35 All functions fi of a bi-Hamiltonian hierarchy jfj I i E Z} are in

involution with respect to both Poisson brackets (hence with respect to any linear combination).
If one of these functions is a Casimir (for either of the structures) then all these fi are also
in involution with the elements of any other bi-Hamiltonian hierarchy.
Proof
If jfj I i E ZI forms a hierarchy, then for any i < j E Z
JA fj}l Ifii fj-1}2

U41, fj-l}l
1h fib
I
so Ifi, fj}l = 0 by skew-symmetry. They are also in involution with respect to the sec-
ond bracket since Jfi)fjj2 =
Jfjjj+jjj. In the same way, if jgj I j E Z} is another
bi-Hamiltonian hierarchy and fk is a Casimir, say of 1., .11 then for any i, j E Z
Ifi; 9jj1 =
jfkj gi+j-k}l = 0.
The above proposition leads to many interesting integrable Hamiltonian systems; said
differently it can be used to give an elegant proof of the involutivity of many integrable
Hamiltonian systems.
64
4. Integrable Hamiltonian systems on other spaces
In this section
we wish to consider briefly integrable Hamiltonian systems on spaces other
than affine algebraic vaxieties. One possible generalization is to consider spaces which are not
necessarily algebraic, but have a differential structure (real or complex analytic), at least on a
dense open subset. Examples include smooth manifolds, analytic varieties and orbifolds. Note
however that extra generality comes also from the fact that one can often choose which algebra
of functions on these spaces to consider, for example one may consider an affine vaxiety with
its algebra of rational functions; however these algebras should be reasonably big in order to
lead to integrable Hamiltonian system, as is cleax from the example of a projective algebraic
vaxiety with its regular functions (which axe only the constant functions). Another possible
generalization, closely related to the problem raised by the latter example is to consider
(reasonable) ringed spaces or schemes. We will only consider the first generalization here.
4.1. Poisson spaces
At first we define a general class of spaces, which includes both affine algebraic vaxieties
and smooth manifolds, onwhich it is possible to define the notion of an integrable Hamiltonian
system.
Definition 4.1 Let M be a topological space which has at least on a dense open subset
a smooth (or holomorphic) structure. Also let R be an algebra of functions on M which
is big enough to distinguish (smooth) points in M, and whose elements axe smooth (resp.
holomorphic) on a dense open subset of M. A Poisson bracket on (M, R) is as in the case of
affine varieties a Lie bracket 1-, -1 R x R -+ R : (f , g) i-+ If , gJ, which satisfies the Leibniz
-
rule in each of its arguments. We call (M, R, I-, J) (or (M, 1-, -1) for short) a Poisson space;
in the special case that M is a manifold and R CI(M) (resp. = R =
Cw(M)) (M, is
called a Poisson manifold (resp. analytic Poisson manifold).
The algebra of Hamiltonian vector fields and the algebra of Casimir8 axe defined as in
case of affine Poisson varieties. The Hamiltonian vector fields axe of course only (real or
holomorphic) vector fields
on the non-singular paxt of the space. On this non-singular part
a representing the Poisson bracket can be defined and also there is a
Poisson tensor notion
of rank at a non-singular point. Notice that all this was in the case of affine Poisson spaces
even defined at the singular points.
Exarnple 4.2 example which originated the theory of Poisson brackets and Pois-
The
son symplectic manifolds. A symplectic manifold (M,w) is a mani-
manifolds is that of
fold equipped with a closed two-form w (a symplectic two-form) which is non-degenerate
(as a bilinear form on each tangent space). A vector field XF is associated to any func-
tion f E C'(M) by
w(Xf, df
and a skew-symmetric bracket is defined on smooth functions by
If, g1 =
W(Xf' Xg)-
Notice that this new definition of Xf is consistent with the definition Xf f I which we
gave in the case of affine Poisson varieties.
65
Clearly is a derivation in each of its arguments and the Jacobi identity for this
bracket is equivalent to the fact that w, is closed. Thus a symplectic manifold is a Poisson
manifold in a natural way. Such a Poisson manifold is regular and its dimension equals its
rank (in particular it is even). Conversely every regular Poisson manifold of maximal rank is a
symplectic manifold in a natural way. In turn, the main examples of symplectic manifolds are
provided by the cotangent bundle to any manifold and by Khhler manifolds. The literature
on symplectic manifolds is immense. See e.g. [AL], [AM1j and [LM3].
A fundamental property of symplectic manifolds is that they admit locally so-called

canonical coordinates (the Daxboux Theorem). The following theorem provides the proper
generalization of this property to Poisson manifolds. This theorem is due to A. Weinstein;
fora proof we refer to [CW].
Theorem 4.3 Let (M, 1-, -1) be a Poisson manifold and letp E M be arbitrary. There exists
a coordinate neighborhood V of p with coordinates (qj, q,, pi.... Pr, Y1 y.) centered
7 ....
at p, such that
8
I-, JV
aqi
A
api
+ -
2
E Okl(Y)y-
Yk
A
yj
k'1=1
where the functions Oki are smooth functions which vanish at p.
The rank of is 2r but is not necessarily constant on a neighborhood of p. When the

1., -1
rank is constant neighborhood of p the neighborhood V can be chosen such that, on V,
on a
the functions Oki vanish, yielding the following canonical brackets for the above coordinates:
lqi)qjl ==
fPiiPjl =
fqi7Ykl =
fPi,Ykl =
fYk7YI1 =
0, f%jpjj =
6ij7 (4.1)
where I < i, j < r and I < k, I < s. In this form Weinstein s Theorem is usually referred
to as the Darboux Theorem and the above local coordinates are called Darboux coordinates
or canonical coordinates. The Darboux Theorem may be refrased by saying that the rank
of the Poisson manifold at a point where it is locally constant is the only local invariant of
a Poisson manifold. A stronger version of the Darboux. Theorem says that a collection of
independent functions (around the point) which satisfy canonical commutation relations can
be extended to a complete set of canonical coordinates. In this stronger form the Darboux
Theorem is false for affine Poisson variety, consider for example on C' the Poisson bracket
Ix, yj x at a point not on the Y-axis and let the incomplete collection consist just of jyj.
=
The only way to complete it with f such that If, yj 1, is to take f ln(x) which is not a
= =
regular function on any Zariski open subset of C'. Canonical coordinates (which are regular
on a Zariski open subset) exist however for this bracket, for example one has 11, -yxj
X
(clearly canonical coordinates which are regular on C' do not exist). It is unlikely that a set
of independent regular (on an open subset) functions, satisfying commutation relations as in
the Darboux Theorem, can be found for any affine Poisson variety, but a counterexample (if
any) is missing.
Although there is a notion of rank at each point of a Poisson manifold, it is not true as
in the case of affine Poisson varieties that the rank is constant on an open dense subset of
the Poisson manifold which may result in some nasty behavior of the algebra of Casimirs.
Consider the following example.
66
Example 4.4 We first construct a bump Poisson structure on the plane R1. Let W be a
non-zero function on R2 whose support Supp(W) is compact and connected. Clearly Ix, yJ =
W(x, y) defines a Poisson bracket on R2 and there is an open subset where the rank is two but
also an open subset where the rank is zero. Moreover its algebra of Casimirs is non-trivial
since it contains all functions whose support is disjoint from Supp(w). Thus Supp(w) is a
level set as as every point in M \ Supp(W). The former level set is never a manifold (in
well
the best case might be a manifold with boundary, but it is in general singular as well). Of
it
course all this is typical for the smooth case; when analytic brackets axe considered then the
rank is constant on an open dense subset, the fibers of a (real or complex) analytic map will
be analytic varieties and so on.
We have discussed in 2.4 two decompositions of affine Poisson varieties, the

Paragraph
Casimir decomposition and the rank decomposition. From what we said it is clear that the
rank decomposition does not have its counterparts in a smooth setting. There is however
in the case of Poisson manifolds another decomposition (singular foliation) the symplectic
decomposition or symplectic foliation which is very useful. Its name stems from the fact that
the Poisson structure restricts to a regular structure of maximal rank on each leaf, hence
the Poisson structure permits to define a symplectic structure on each leaf. On an affine
Poisson variety the leaves of the symplectic foliation need not be algebraic (as e.g. in the
Example 2.43) and they (i.e., equations for them) are difficult to determine explicitly in
general (for example it is a well-known result that in the Lie-Poisson case (see Example 2.8)
the symplectic leaves coincide with the co-adjoint orbits, i.e., the orbits of the corresponding
group G acting on 9* via the co-adjoint action; even in low dimensions these orbits may be
very hard to compute).
The easiest way to obtain the symplectic foliation is by using Weinstein's Theorem.
Indeed, a subvariety of M around p is obtained by taking y, =
...
=
y, = 0 and along this
subvariety I -, J restricts to a symplectic structure and this (local) subvariety is the only one
containing p on which f-, -1 restricts to a Poisson bracket of maximal rank. Hence we may
globalize this construction to find a unique symplectic leaf passing through each point. Notice
that these leaves are immersed submanifolds and not closed submanifolds in general; each
leaf may even be dense in M, as is shown in the following example (the example also shows
that, even in the case of Poisson manifolds, the algebra of Casimirs needs not be maximal).
Example4.5 Take on R3 an orthogonal basis el, e2, e3 with e3 (1,a,,3) where 1,a and,6
==
are linearly independent over Q. The bivector el A e2 determines by parallel translation a
Poisson structure on R3 which descends to a Poisson structure 1., -1 on the torus WIZ3.
All symplectic leaves two-dimensional, but they are dense, hence
are none of them can be a
level set of the Casimirs, such level sets being always closed.
As a final remark about the symplectic foliation, we wish to point out that Weinstein's
proof is easily seen to be valid also in the holomorphic case, yielding a holomorphic symplectic
foliation on any holomorphic Poisson manifold. For affine Poisson varieties this leads to a
holomorphic symplectic foliation on its smooth part (which is a complex manifold).
In the following definition we generalize Definition 2.15 to the case of general Poisson
spaces.
67
Definition 4.6 Let (Ml, R1, J'7 *11) and(M2, R21 J* '12) be two Poisson spaces, then a
map 0: M, -+ M2 is called a a Poisson morphism if

(1) 0*7Z2 C R1,
(2) 0*1figJ2 10*f,0*911i
'-- for all f,g E R2-
A Poisson morphism which has an inverse is called a Poisson isomorphism.
In terms of integral curves the relevance of Poisson morphisms for (integrable) Hamilto-
nian systems (as defined below) is formulated by the following proposition.
Proposition 4.7 Let (MI, 1-, -11) and (M2, J* *12) be two Poisson manifolds and suppose1
that 0: M, -+ M2 is a Poisson morphism. Then the integral curves of a Hamiltonian vector

field XH, H E C'(M2) which intersect O(MI) are entirely contained in O(MI) and are the
projections under 0 of the integral Curves Of XO-H.
Proof
If -y is an integral curve of O*H then 0 oy is an integral curve of H. Indeed, let gi be
any local coordinates, then
(gi o 0 oy)* =
Igi o 0, H o 01 oy =
Igi, HI o 0 oy.
If P E O(Ml) c M2, let Q E M, be lying over P, then the above computation shows that
the integral curve of H o 0 through Q projects (via 0) onto the (unique) integral curve of H
passing through P in particular this integral curve cannot leave O(MI).
-
We wish to point out that a similar proposition, stating that all integral curves Of XH
are projections of integral curves Of XO*Hi is given in [Wei2] (Lemma 1.2 p. 528), but this
cannot be true: it would imply surJectivity of the map 0 (at least onto the non-singular part).
Even when dealing with integrable Hamiltonian systems on symplectic manifolds one
should by the above
proposition consider Poisson morphisms rather than symplectic maps.
It is seen from the following simple example that the two concepts do not agree in general
and that the above proposition needs not hold for symplectic maps.
Example 4.8 Take M2 R4 (with coordinates X1 Y1 X2 Y2) and M, C R4 the plane

=
5 i i
given by X2 Y2==-0- On both M, and M2 we put the standard symplectic structure:

==
wl dxl A dyl and W2

= dxl A dy, + dX2 A dy2. Then there are obvious projection and
=
inclusion maps
2 2
7r: R4 -+ R and z: R _+ W,
and it is easy to check that -7r is Poisson but not symplectic and z is symplectic but not
Poisson.
Example 4.9 Let us show by a simple modification of the previous example that Propo-
sition 4.7 needs not be true for symplectic maps. Instead of the obvious inclusion map we
consider now the symplectic map
0: R2 -+ W : (xi, yi) -+ (xi, y, , x, , 0) -
The function X2 on W has all integral curves parallel to the Y2 axis, hence none of them is
included in the image of 0.
68
Thepolynomial invariant which we associated to affine Poisson varieties does not gener-
ahze to general Poisson spaces since the rank decomposition may not lead to (a finite number
of) reasonable spaces, so it may not be clear how to count "components". For analytic brack-
ets our construction goes however over verbatim. A lot of attention has been given over the
last few years to global invariants for symplectic manifolds, a good introduction and more
references are given in [AL].
4.2. Integrable Hamiltonian systems on Poisson spaces
As for integrable Hamiltonian systems on general Poisson spaces we would like to copy
Definition 3.5, but a few modifications axe needed.
At wish the rank of the Poisson space to be constant on some open dense subset,
first, we
otherwise may run into complications such as in Example 4.4 in which at some open subset
we
the level sets of the integrable Hamiltonian system are given by the levels of -the Casimirs
and in some other open subset they axe given by the level sets of the integrable algebra. In
such the Poisson space can be split in two, so it is a mild assumption that the rank is
case
constant on an open dense subset; this constant is then called the rank of the Poisson space.
Second, the notions of spectrum and dimension for an algebra A C R need to be modified.
our algebras A have no spectrum nor a dimension; the dimension is naturally replaced by the
number of independent functions (we say that a collection of functions is independent if their
differentials axe independent at every point of some open dense subset). As for the spectrum,
which we needed in order to define the momentum map, we could take Hom(A, R) (resp.
Hom(A, C)) or the real spectrum (in the case of manifolds) but this may be a very complicated
(and ugly) object; in particular we will not have a smooth or holomorphic projection map
M -+ Hom(A, R); however for any system of generators fl,... fn as above, we will have a ,
smooth (resp. holomorphic) map M -- Rn (resp. M _+ Cn).
Third, it is not clear at all how to show for general Poisson spaces that some algebra is
complete (in the sense of Definition 3.1). Recall that we insisted in having completeness in
order not to call two systems non-isomorphic while their algebras have the same completion.
A solution to this is not to insist on completeness in the definition of an integrable Hamiltonian
system but to call two systems isomorphic when some involutive extension of their integrable
algebras coincide.
These remarks lead to the following definition.
Definition 4.10 Let be a Poisson space which is of constant rank on an

open dense subset of M and whose algebra of Casimirs is
maximal, i.e., it contains dimM -
CoRkJ-, -1 independent functions. An involutive subalgebra A of Z is called integrable if it

contains dimM 1
2 RkJ-, -1
-
independent functions. The quadruple (M, 7Z, I-, J, A) is then

called an integrable Hamiltonian system and each non-zero vector field in
Ham(A) =
JXf I f E A}
is called an integrable vector field.
69
Example 4.11 In its original form the three body Toda lattice is given on RI with the
standard symplectic structure E dqi A dpi by the algebra generated by the following two
smooth functions:
H= IE Pk+Ee
2
2
k=1 k=1
PIP2P3 -
EPkeqk+l
k=1
Since the translations
(qi, q2i q37P1 iP2i P3) -+ (ql + a, q2 + a, q3 + a,pi, P21 P3)
define a Poisson action, the quotient of RI by these translations, which is R5 inherits a

Poisson structure. It leads on every hyperplane pi + P2 + P3 c (c E R any fixed constant) =
to a symplectic Since the group action leaves the functions H and I invariant
structure.
they descend to this quotient and since they are in involution they are also in involution on
the quotient. Clearly they axe also independent, hence the algebra generated by H and I is
integrable.
In view of the exponentials this is not what we called an integrable Hamiltonian system
on an affine Poisson space; it is however closely related to one, see Section VII.7.
Example 4.12 A second example is given by the elliptic Calogero-Moser system, studied
in detail (especially from the point of view of algebraic geometry) by Treibich and Verdier
(see (TV]). The setup is the same as for the Toda lattice above but the exponentials axe
replaced by the Weierstrass p function. In the simplest case of three "particles" the involutive
algebra is generated by the following two meromorphic functions (P is the Weierstrass function
associated to a fixed elliptic curve)
3 3
1 2
H =
2.EPk -
2 1: p(qk+l -
qk-1),
k=1 k=1
3 3
3
K =
3- 1: Pk
-
2 E (Pk+l -
Pk-I)P(qk+l -
qk-1)
k=1 k=1
As in the Toda case there axe many different versions of the Calogero system (rational,
trigoniometric, relativistic, ...) and as in that case they are all closely related to integrable
Hamiltonian systems on affine Poisson varieties.
Finally, here is the definition of a morphism. of integrable Hamiltonian system on Poisson

spaces. Notice that in property (3) below we do not ask that O*A2 C Al, in accordance with
the third remark, just before Definition 4.10.
Definition4.13 Let (M1,R1,J*,'J1,A1) and (M2,7Z2&,*J2,A2) be integrable Hamilto-

nian systems, then a map 0 : M, -+ M2 is a morphism if it has the following properties.
(1.) 0 is a Poisson morphism,
(2) 0* CaS(M2) C CaS(MI);
(3) O*A2 C A3, where A3 C Ri is an involutive algebra which contains A,.
70
Chapter III
Integrable Hamiltonian systems
and symmetric products of curves
1. Introduction
chapter is devoted to the construction and a geometric study of a big family of

This
integrable Hamiltonian systems. The phase space is C2d equipped with an infinite dimen-
,
sional vector space of Poisson structures: for each non-zero W E C[x,y) we construct (in
Paragraph 2.2) a Poisson bracket I-, J1 d
which makes (C2d, d
into an affine poiSSon
vaxiety. Each of these brackets has maximal rank 2d (in paxticular the algebra of Casimirs
is trivial) and they are all compatible. An explicit formula for all these brackets is given;
they grow in complexity (i.e., degree) with W so that only the first members are (modified)
Lie-Poisson structures.
What is surprising is that all these structures (for fixed d) have many
integrable algebras
in common; moreover a system of generators of these algebras
given by a very compact
are
and simple formula. Namely there is one integrable algebra corresponding to each polynomial
F(x, y) in two vaxiables (it is assumed here that the polynomial depends on y). The magical
formula is given by
H(,X) =
F(.\, v(A)) mod u(A);
in this formula u(A) is a monic polynomial ofdegree d and v(A) is a polynomial of degree
less than d and the 2d coefficients of these two polynomials are the coordinates on CU. The
integrable algebra is obtained from this formula by taking
AF,d =
C[HOi ... 1Hd-11i
where Hi is the coefficient of Xi in H(.X). It leads to many integrable Hamiltonian systems and
for fixed F(x, y) they are all compatible; the integrable vector fields which correspond to them
a,re however different so that these do not give integrable multi-Hamiltonia-n systems. Their
71

Chapter III. Integrable Hamiltonian systems and symmetric products of curves
integrability is shown in
Paragraph 2.3. We will look at the special case for which F(x, y)
Y2 _
f(X) in Paragraph 2.4; in this case we are able to write down Lax equations for the
vector fields.
A closer study of the fibers

of the momentum map reveals the meaning of the poly-
nomial F(x, y). We describe the fiber FFd over (Ho,...' Hd-,) (0, ...' 0) and obtain a
=
description of the other fibers by a slight change in F. If the algebraic curve IPF (in C2)
defined by F(x, y) 0 is non-singular then TFd is non-singular and we show that in this case
=
the fiber -FFd is isomorphic to an affine part of the d-fold symmetric product of the algebraic
curve ]Pp (we also give an explicit description of the divisor which is missing). This shows
that basically all our systems (for different F) axe different and that the d-fold symmetric
products of any curve (smoothly embedded in C2) appears as a level set of some integrable
Hamiltonian system. We deduce from the description of the general fibers of the momentum
map a description of their real parts. For d 2 (when surfaces are obtained as level sets) the
=
description is easily visualized and shows at once that a large family of topological types is
present. The level sets are described in Paragraph 3.2 and their real parts in Paragraph 3.3.
The effect of changing the Poisson structure
(keeping F(x, y) and d fixed) manifests itself
only at the level of the integrable vector fields
(the Poisson structure is not seen from the
fibers of the momentwn map since these depend on F(x, y) and d only). These vector fields
axe all tangent to the same fibers and span the tangent space at each (non-singular) point,
hence these vector fields must be related; they are in the present example even related in a
very simple
way, however these vector fields are different for all choices of V so that changing
V also leads to different (i.e., non-isomorphic) systems. The effect of varying the Poisson
structure is given in Paragraph 3.5.
Later in the text we will refer on several occasions to the systems described in this
chapter. For a futher generalization of these systems, in which F(x, y) is replaced by a family
of algebraic curves, we refer to [Van5]. For a more abstract, but less explicit, construction of
these systems, where C2 is replaced by any Poisson surface, see [Bot].
72
2. The systems and their integrability
In this section we show how there is associated to every polynomial F(x, y) an algebra
of functions which is integrable with respect to a family of compatible Poisson structures
on C2d, which is parametrized by the set of all polynomials W(x, y) in two variables.
2.1. Notation
C2d is viewed throughout this chapter as the spare of pairs of polynomials (u(A), v(A)),
with u(A) monic of degree d and v(,\) of degree less than d, via
U(A) = Ad + Ud- 1Ad-1 + ...

+ UI'\ + UO'
(2.1)
V(A) =
Vd-I Ad-I +... + VIA + VO,
so the coefficients ui and vi serve as coordinates on C2d. Some formulas below are simplified
by denoting Ud = 1-
For any rational function r(,\), we denote by [r(A)]+ its polynomial part and we let
r(A) [r(,\)]+. If f (,\) is any polynomial and g(,\) is a monic polynomial, then
= -
f (,\) mod g (A) denotes the polynomial of degree less

than deg g (A), defined by
f (A) mod g (A) =

g (A) [ fg ((,\)1111
sof (A) f (A) mod g(,\) + h(,\)g(,\) for a unique polynomial h(,\) and f (A) mod u(A)
=
,
is easily
computed as the rest obtained by the Euclidean division algorithm.
2.2. The compatible Poisson structures
Any polynomial w(x, y) specifies a Poisson bracket on C2 by ly, xj W(x, y), hence also =
on the cartesian product (C2)d C2 X X C2 (by taking the product bracket). Explicitly
= ...
fyi, Xj I =
6ij W(Xj' yi), 1xi, Xj I =
lyi, Yj I =
0, (2.2)
where (xi, yi) are the coordinates on the i-th factor, coming from the chosen coordinates
on C2 Let A denote the closed subset of
.
(C2)d defined by
A : --
I ((XI Y1)) (X2 Y2)
i 7 i ...
I (Xdi Yd)) I xi =
xj for some i 0 jJ,
and consider the map S : (C2)d \ A _4 C2d' given by
A -
Xj
((X1iYI)i (X21Y2)1 ...
I (Xd, Yd)) -+ (U(A) V(,X)) i (A_ Xi), Yi rl Xi -
Xj
(2.3)
73
Chapter 111. Integrable Hamiltonian systems and symmetric products of curves
This map can be interpreted as a morphism of affine Poisson varieties

upon using Proposi-
tion 11.2.35. This is done as follows. Define
MI I (XO (XI Y0
I I I ...
I (Xdi Yd)) X0 H(X, Xj)2 _
=
11 C C X (C2)d,
i<j
and
M2 =
J(t, u(A), v(A)) I t disc(u(A)) =
11 C C X CU.
Then we have a commutative diagram
M, M2
P1 P2
(C2)d \A S
C2d
with S a morphism between the affine varieties M, and M2. By Proposition 11.2.35 the
Poisson bracket j.'.1 on (C2)d leads to a Poisson bracket on MI, also denoted by
upon using the relation x0 11i<j (X, _
Xj) 2 =
1; namely one adds the brackets
jx0' Xjj =
0, and 1X0, A -X0
2
I ]I(X,
i<j
_
Xj)2, A
the latter being computed from
IX0, Yk} 11(xi

i<j
-
xj)
2
+ JI(X, Xj)2, y1c
i<j
_
I XO = 0.
d
The natural action of the permutation group Sd on (C2) lifts to a free action of Sd on M, and
d
since it is a Poisson action on (C2) it is also a Poisson action on (MI, I- ,
-
1) NowS is a d!
. : I
(unramified) covering morphism variety onto the affine M2 and S is invariant for the action
of Sd on MI, hence M2 may be identified with the quotient MIlSd. By Proposition 11.2.25
M2 has a unique Poisson structure such that S is a Poisson morphism. It will be denoted
by I-, -J d'
We would like to transport this Poisson structure on M2 to C2d by the morphism P2
M2 _+ CU. Of course this is in general impossible, however in the present case it turns
out that there does exist a (unique) Poisson structure on C2d, also denoted by -1'*,
d
such
that P2 is a Poisson morphism. It is given in the following proposition.
Proposition 2.1 There exists a (unique) Poisson structure I-, J'd on C2d such that P2
M2 _+ C2d is a Poisson morphism. In terms of the coordinates U,' V, for O(CM) the Poisson
bracket is given by
IUN I Uj Id IV(A),Vjld' =
01
IU('\)'Vj}d' =
1Uj1V(A)1d' =
W(AIVN) I U(11)
Aj+1
mod u (A), 0 < j :5 d - 1.
(2.4)
74
Except for the zero bracket 1., -10,

d
all Poisson brackets I-, J'd are of rank 2d and they are all
compatible.
As a special and most important case, ify and x are canonical variables, i.e., V(x, y) 1, =
then the Poisson structure I-, -I'd, also denoted by 1"'Id; is regular; the nonzero part of the
Poisson bracket (2.4) reduces in this case to
1U(A)'Vj1d=1Uj1V(A)Id= [u`1)
),j+l
(2.5)
and its matrix of Poisson brackets with respect to the coordinate functions ui and vj, takes
the form
0 0 1
(0 ...
)
0 0 ... I Ud-1
0
P=
(-U U) 0
where U
0 1 U3 U2
I Ud-1 U2 Ul
In terms of I') Jdy the Poisson structure I-, -I'Pd is given by
1U(A)7 Adp W(A7 VN) 1U(A)7 f1d mod u(A),

0 (2.6)
MA) fl d 7 (P(Al VN) IVN f1d mod u(A), I
where f is any element of 0 (CM).
Proof
We compute explicitly on M2 the Poisson brackets of uo.... i Ud-li VOi Vd-1 (without ...
worrying about their brackets with t) and observe that they are independent of t; by Propo-
sition 11.2.4 this leads to a Poisson bracket f. .IVd on C2d which makes p2 into a Poisson
morp1hism (the unicity of this bracket is immediate). Clearly Ju(A), u(p) J d' 0. If I < j :5 d,
then
JUd-jMA)jd (-I
ji1<i2<**'<?j Xil Xi2 xij
1=1
yl 11 X1
kol
A -
-
Xk
Xk Id
A Xk
1: EjXi1Xi2-XiJ7y1 1W 11 X1 d
Xk
i1<i2<-<ij 1=1 k961
i
A -
Xk
Xil Xi2 ist xij W(xit IYO
X't -
Xk
i1<i2<-<ij t=1 kOit
A -
Xk
(-W-1 XilXi2
X1 -
Xk
10fi1<i2<***<ij-1j k961
d j-1
A Xk -1
(-W-1 Axt, YO I)j X7nUd -j+m+l
X1 Xk
k961 M=0
d j-1
-
Xk
E E Xm1Ud-j+M+1(P(X1 7 YO
X1 -
Xk'
1=1 M=0 kol
75
Chapter M. Integrable Hamiltonian systems and symmetric products of curves
Substituting A= x, in the right hand side one sees that IUd-hV(,XWd'O is the
(unique) polyno-
I
mial in A of degree less than d which takes at A =
xj the value Ej rn =0 Xl"'Ud-j+m+l W (Xii V(X1))7
for I =
1,...,d. As the xj axe the zeros of u(A) and since yj =
v(xl) the same is true for
J-1
E M=O \M Ud-j+m+1W(A, v (,\)) mod u(,\), and we find
j-1
jUd-j V(A) Id
,
=
1: 1\ M
Ud-j+m+lW(/\i V(,\)) mod u(A)
,n=O
U(,\)
=
W(A,V('M I\d-j+l mod u(,\),
which proves the second equality in (2.4). For the first equality in (2.4), notice that
d W
IU001 V(A)ld
I HP\ -
X.) YZ
Id H X1
j961
/Z -
-
Xj
Xj
d
(A x .) Ut -
xj)
=EW(X1,Y0jj
1=1 jol
is symmetric in and p, which leads at once to 1UNIVild' =

lUi,V( A d*
In order to show that Ivi, vj I'd ==

0, let us simplify the formulas by chosing (p 1; for
general W the result then follows from (2.6). By construction
IV (/\), V (A) I d =
E
i,k=l
A
( Xi
\ -
-
xj
Xj )I Yi, 11 Xk
196k
-
X,
X1
-
(i ++ k, p ++ (2.7)
where (i ++ k, p denotes a term similar to the first one, obtained by exchanging i with
k as well as p with As for the first term, its terms corresponding to k i axe given by
=
Yi
( -Xj) ( O-x,)r-
Xi -
Xj Xi -
X1 Xi -
X1,
which is symmetric in X and p so that these cancel when substracting the symmetric term
in (2.7). The
remaining terms, which correspond to i =A k can be rewritten as
k96i
Yk
( "
Xi
-
-
xi
Xj ) (Ini 96k,
/,
Xk -XI
-
X,
) P
(Xi
-
-
Xk
Xk)21
a polynomial which evaluates to 0 for all (X, p) =

(x,, xt) with =A t. It follows that both
s
2
terms in (2.7) are of degree less than d which agree on the d couples (X, p) (x,, xt) and
=
we may conclude that IV W 7

V (11) Id = 0-
Compatibility pf the brackets derives from the formula
W+1P
I -'J'd =I
P
I- Td
I
+ .
I
.
Id
which is an easy consequence of (2.4).
76
Notice that it is also seen from formula (2.4) that -j'Pd really is a map from O(CM) X
O(CM) __ O(C2d): it suffices to use that if a polynomial (in several variables) is reduced by
using a monic polynomial then the result is also a polynomial (in all these variables).
For W = 1 one obtains (2.5), because the degree of IU 1+ is less than d for any
j =
0, . . .
,
d -
1, which also leads at once to the matrix representation of d

-
since the
determinant of this matrix equals 1, it is regular of rank 2d. Note also that if d > I then
1* "Id is not compatible with the standard Poisson structure on CU.
To see where the rank of the Poisson structure I-, J1 d

fails to be maximal, we need to
investigate the determinant of the matrix of Poisson brackets f ui, vj 1'.d Using elementary
properties of determinants one finds that for any values -ol, Xdi ...
i
det (jui, v(xj+,)Idl) O<i,j:5d-1 = det RUi1VjPd0) O<i,j<d-I 11 (Xk -
X0 (2.8)
k<1
Choosing x, Xd to be the roots of u(A), we get from (2.4)
det RN V (Xj+l) I 0)0<i,j<d-l=det

I d W(xj+,,v(xj+,)) Aj+1 )' +P1=0j+') O<i,j<d-1
d
= det
Ai+1 11 W(X-' V(X.))
+ -\=xd + O<i,j < d- I
d
det (N' V(Xj+l)}d) 0<,,j<d-, rl V(Xml V(X-))

M=1
d
M
( _1)[d/2] JI(Xk -
XI) H W(XM' V(XM))'

k<1 M=1
where in (i) we used (2.8) for W 1. It follows that (even if u(A) has multiple roots)
d
[d/2]
det RUil VAd 11 W(XM' V(X.)),
M=1
d
on all of C2d, hence the Poisson structure is of lower rank on the lo cus rl.!_-, W(xj, v(xj)) 0,
which for given W and d is easy written as the equation of an algebraic hypersurface in CU.
Finally, (2.6) follows immediately from the Leibniz property of Poisson brackets. 9
If W depends only on x and has degree at most d, then 1., -11d is a modified Lie-Poisson
structure (see Example 11.2.14). Explicitly, for W x1, 0 < n < = d the Poisson matrix P,,
with respect to the above coordinates, taken in the order Ud-1, ...
iUO,Vd-1, ...
7 vo I is given
by
0 0 Un 0
0 0 0 -Un'
Pn
Un 0 0 0
0 Un/ 0 0
77
where
0 0 1
(0 ...
) Un-1 Un-2 * -
Ul UO
0 0 Ud-1 Un-2 Un-3 -
UO 0
Un -'-- : and Un1
0 1 Un+3 Un+2 Ul UO ...

0 0
I Ud-1 * -
Un+2 Un+1 J UO 0 ...
0 0/
In paxticular, if 0 < n < d then the bracket is

product bracket (of a regular Poisson structure
a
of rank 2(d -
n) on C2(d-n) and
non-regular Poisson structure of rank 2n on C2n) For
a .
n = d one finds a Lie-Poisson bracket which is given by
Ud-1 Ud-2 Ul UO
Ud-2 Ud-3 UO 0
Pd (0Ud' _Ud) 0 ,
where Ud
I
0 0
Ul UO ...
UO 0 ...
0 0/
For V =
"=O E!
cixi the corresponding Poisson matrix is given by E! %= OciPi. Notice that
I
these axe the only W (x, y) for which I -
,
-
Id is a modified Lie-Poisson structure and that
,p(X, y) =
rXd' (c E C) is the only one which gives a Lie-Poisson structure.
For p(x, y) Xn, = 0 < n < d it is easy to compute the invariant polynomial (defined in
Definition 11.2.47) of the Poisson structure, which we will denote by Pd,n. Namely, since Un
is non-singular it suffices to look at the matrix U' whose determinant is zero if and only if
.
uO =0. By induction Un' has co-rank at least k if and only if uO ul = = ... =

Uk-1 = 0-
In conclusion the rank is at most 2d on all of C14, it is of rank at most 2d -
2 on the
hyperplane uO =
0, ...'
it is of rank at most 2d -
2n on the (2d -
n)-dimensional space
UO =U1 =
un-i
... = = 0 and
the latter space it is regular. Thus we have established the
on
following formula for the polynomial associated to the Poisson structure I-, (0 :5 n :5 d): -
1dX'
Pd,n = R dS2d + Rd-ISM-l +---+Rd-nS2d-n
= Rd-nS2d-n (I + RS + R2S2 + - - -
+ Rn Sn)
Notice that the fact that pd,,, is reducible reflects the fact that the Poisson structure is a
product. The polynomial has the simple (d + 1) X (2d + 1) matrix representation
0 0
0 In+,
where I is the identity matrix of size n + 1.
2.3. Polynomials in involution for I-, -j"d
We will now show how an arbitrary polynomial F(x, y) leads to a natural set of d
polynomials on C2d which have the remarkable property to be in involution for all the Poisson
structures I -
,
- 1.
1 d These polynomials generate a d-dimensional algebra (under the assuinption
that F (x, y) is not independent of y), hence they define an integrable Hamiltonian system
on C2d for any structure .1'.
d
Since all the brackets are compatible this means that we
78
have for each F(x, y) (which is not independent of y) a large class of compatible integrable
Hamiltonian systems. They are however not multi-Hamiltonian as we will see in Section 3.5
below (see however also Paragraph VI.3 and [Van5]).
Let F(x, y) E C[x, y] \ C[x] and let us view Cd as the space of polynomials (say in A) of
d
degree less than d. Then there is a natural map flPd from (C2) \'6' to Cd, which assigns to
a d-tuple ((xl yl) ,(Xdi Yd)) the unique polynomial in C[A] of degree less than d, which
I ...
I
takes for A xi the value F(xi,yi) (for i

=
1,...,d). We thereby arrive at the following
=
commutative diagram
M, M2
PI P2
(C2)d \A C2d (2.9)

S
\f1ii-111
fIll, d
cd
in which the existence of the dotted arrow is guaranteed by the following lemma.
Lemma 2.2 There exists a (unique) morphism HFd C2d _+ Cd such that the triangle
in (2.9) is commutative. Hpd is explicitly given by
HF,d(U(A), V(,\)) =
F(A, v(A)) mod u(A). (2.10)
Proof
P*lfIF,d : MI _+Cd is a morphism which is invariant for the action of Sd hence it can be
factorized via the quotient MIISd which we identified with M2. This means that we have a
morphism p3: M2 _4 Cd. It associates to (t, u(A), v(A)) E M2 the unique polynomial (in A)
of degree at most d -
I whose value for A =
xi, xi any root of u(A), is given by F(xi, yi). A
compact formula for P3 can be given:
P3 (t, u(A), v (A)) =

F(A, v (A)) mod u(A).
To check this formula, note that the right hand side is clearly a polynomial (in A) of degree
at most d I and for any xi which is aroot of u(A) it evaluates to F(xi, v(xi))
-
=
F(xi, yi).
Since the map P3 does not depend on t it can be factorized in turn via P2) i.e., P3 =
P*2HFd
and HFd is explicitly given by (2.10). 0
The d components of the map HFd define d regular functions (polynomials) on C2d'
which will be simply denoted by Hd-,,..., HO (omitting the dependence on F and d in the
notation), i.e., HFd(U(A), V(,X)) = Hd-,Ad-1 + Hd-2,\d-2 + + HO.
The main result of this section is the following.
79
Chapter Ill. Integrable Hamiltonian systems and symmetric products of curves
Proposition2.3 For any polynomial F(x,y) E C[X,y]\C[X], let AFd C[Ho, .,Hd-,.], =
-
where HO,...' Hd_j are the coefficients (in A) of H(A) F(A, v(A)) mod u(A). The triple=
(C2d, I., J,
d
AFd) defines for any non -zero W(x, y) E C[x, y] an integrable Hamiltonian sys-
tem and these systems are all compatible.
Before proving this proposition we prove a key lemma and write down explicit equations
for the Hamiltonian vector fields XH , =
I-, Hilld
Lemma2A Letp(A), q(A) and r(A) be polynomials, with degq(A) ! degr(A) and let i E N.
(1) r(A) [A-'q(A)] + mod q(A) =

r(A) [A-'q(,\)] + -
q(A) [A-'r(A)]
deg q deg q
(2.11)
(2) p p(A) [A-lq(A)] + mod q(A) A'-'p(p) [tt'q(p)] + mod q(M).
Proof
For the proof of (i) note that if deg r(A) :! deg q(A) then the right hand side of the
identity
r(A) [A-'q(A)] + -
q(A) [A-r(A)] + =
-r(A) [A-'q(A)] _
+ q(A) [A-'r(A)]
is a polynomial of degree less than deg q(A), hence also the left hand side. To show (2) we
may assume that deg p(A) < deg q(A) because the equality depends only on p(A) mod q(A).
Then
deg q deg q
[M-lq(/.A)]+ mod q(p) Al-I (p(p) [lj,-lq(p)] +

-
q(p) [IL-lp(p)] +)
deg q
(p(A) [A-1q(A)] +
-
q(A) [A-1p(A)] +
deg q
[A-lq(A)]+ mod q(A).
In (i) applied part (i) of this lemma; the exchange property

we in (ii) is proven at once by
expanding the polynomials or by induction on deg q(X). M
Proposition2.5 The coefficients Hi ofF(A, v(A)) mod u(A) determined polynomial vector
fields XH 'i on C2d' which are explicitly given by
aF U(A)
XH U(A) =
W(AIV(X))
ay
GX' v(A)) I'Xi+1 modu(A),
(2.12)
[F(A,v(A))
XW
Hiv(A) =
W(A' VNI ,
U(A) 1+ Ai+1
mod u(A).
80
Moreover, the following remarkable identities hold for all 0 < i, j :5 d -

1:
jui, HjjId =
JU31 H-l'd
-
"
and Ivi, Hjj'Pd =
jvj, HiJ'*d (2.13)
Proof
Writing XH, as a shorthand for XHI,, we first compute XH,u(A) ju(A),Hjj4, which
we obtain as the coefficient of /.jd-i in JU(,X)iH-Fd(U(P),V(/4))Id*
d-1
(9HF d
'
JU(A), HFd(U(/Z)) V(/'))Id =

1: IUN I Vi Id
avi
'
(U(/')'V(/4))
j=O
d-I
=
E [U(,\)]
Aj+1
19HFd(U(/.,),V(,,))
j=0 + avj
d-I d-j-1
aF
=
E E Ud-kA d-j-k-I (/,, v(p))pj mod u(p)
k=O
i9y
j=O
d-I d-I
49F
=
Z E uj+,Al-l ay (/.z, v(IL))pj mod u(p)
1=1 j=O
d
OF
A,-,
ay
Ca, V(/Z)) IU(P)I
A +
mod u(p)
d
aF
ay
(A, v(A)) IU(I11
),I
+
modu(A)
where we used the exchange property (2.11) in the last step. Since Hi is the coefficient of Ai in
H(A) this leads to equation (2.12) for XHu(A) in case p(x, y) = 1. In a similar way XHv(A)
is found, the computation of -OujLHFd(U(A), V(/I)) is however more involved: let 0 < j < d -
1
then
a F(p, v(p))
a
(F (p, v (p)) mod u (p))
'9Uj (u(t) U(m)
F(p, v(p))
(U(/Z) I U01) I +)
/d pi F(p,v(p))]
-U(/')
RM-) [F(p,v(p))]+_
U(P) U(P) U(A) I +)
y
U(A) [F(g, v(p)) +]
modu(p).
U(O +
81
In (i) used that if R R(p) and P P(p) rational functions, with

we = = axe
[R]+ =
0, then
R [Pj+ -
[RP]+ = R [P]+ -
[R [P]+] + =
[R [P]+]
Granted this, we obtain as above
d-1
jV(A),HF,d(U(A)i d
=
1=0
tit [U(,X) ] [F(A,U(X)v(A)) ]
X1+1
+ +
modu(X),
which leads at once to the expression (2.12) for XHj v (X) in case W(x, y) = 1. Having obtained
the formulas (2.12) for XHu(,\) and XHv(X), the formulas for XH",u(,\) and XHI%v(X), are
obtained at once upon using (2.6).

Finally, the exchange property (2.11) implies that X and M are everywhere interchangeable
in the above computations so we get JU(,X), HFd(U(A)i V(ltffld o JU(/1)7 HFd(U(X)i V(,\))Ilpd =
I
which is tantamount to the identity juj,Hjj' juj,Hjj'. The second formula in (2.13)
=
d d
follows in the same way.
Proof of Proposition 2.3

We first prove that jHi, HPd(U(X) i 0 for 0 < i < d VNWd
1, which shows that = -
AFd
is involutive. To make the proof more transparent, we use the following abbreviations:
aF F(,\, v(,\)) W(X, V (,X))

FV =
19Y
(,\,V(,\)), F(U) =
U(A)
and Uj =
U(A) 1U(11)I
Ai+1
+
I
so that (2.12) is rewritten as X"

Hj u(A)
=
u(A) [UiF,] and XH v(A) =
u(A) [Uj [F(u)] +]
Then
jHF,d(U(A)iv(A)),Hjj O=X.'Fj,
d U(A) I F(A,(A)v(A))
U
-
-)
lp
-=XHjUG ) [F(u)]_+u(A) I XH' 0,F(A,v(A))
U(A)
F(u)XH Ou(A)
U(A) I -
=
U(A) [[UiF,]- [F(u)] _
+ FV [Uj [F(u)] +1 -
-
F(u) [UiFv]_]
_
W
u(A) [- [UiF,] [F(u)] +
+ F, Uj [F(u) ] + I
=
u(A) [[UiFv]+ [F(u)] +]
_
=: 0.
In (i) we used the fact that polynomial, i.e., [F,,]_

F,, is a 0. =
We now show that the d coefficients of Hpd(U(1\), V(I\)) F(A, v(A)) mod u(A) are inde-
pendent, showing that dim AFd d dim C2d 1Rk I
=
IClearly
d
== the last d coefficients -
,
-
f1d-1,...' AO of F(A, v(A)) are independent because vi appears only in f1d-1,...' f1i (it does
appear since F(x,y) 0 C[x]). Reducing F(A,v(A)) modulo u(A) amounts to substracting
from ki polynomials of lower degree in the variables vj, so it cannot make these functions
dependent and the independence of JHO,...' Hd-11 follows.
Finally A.Fd is also complete. Namely we will show (in Proposition 3.3) that the general
fiber of M --+ Spec(AFd) is irreducible and (in Lemma 3.4) that all fibers have the same
dimension (d). Thus AF,d satisfies all conditions of Proposition 11.3.7 which implies its
completeness.
82
Amplification 2.6 Suppose that F(x, V) and F(x, y) differ only by a polynomial c(x)
which is independent of y and is of degree less than d in x, say
d-1
F(x, y) =
F(x, y) + E cix',
i=O
then
d-1
F(A, v(.X)) mod u(A) =

F'(,X, v(A)) mod u(A) + E ci) ,
i=O
hence the polynomials in involution which they determine are up to constants the same and
we find APd =
API,d, that is both systems are isomorphic. We might reformulate our result
by saying that -
for W(x, y) fixed we have associated an integrable Hamiltonian system
-
to a family
d-1
Fc (x, Y) = F (x, y) + cix' I ci E C

i=O
Suppose now bigger family M is given, i.e., F,.(x, y) depends on one or several extra
that a
parameters which we suppose to parametrize an affine algebraic variety. One observes that
the Hamiltonians Hl,...,Hd depend polynomially on the coefficients of F, hence also on ,,
the parameters c, so by Proposition 11.3.24 there is an integrable Hamiltonian system on

M X C2d with O(M) as its algebra of Casimirs and with projection M x C2d -+ M Such
that the fiber of this morphism over any closed point c E M is precisely our original system
on C2d corresponding to the polynomial F,(x, y) where the parameters have been given the
fixed value c. We will often prefer to work on these bigger systems, see Chapters VI and VII.
For a further generalization, in which arbitrary families of algebraic curves are considered,
see [Van5].
2.4. The hyperelliptic case
We now turn to a case which will be important later: the case that F(x, y) y2 f(X) = _
for some polynomial f (x). We call it the hyperelliptic case because F(x, y) 0 now defines a =
hyperelliptic curve (see Paragraph IV.2.6). In the following proposition we give Lax equations
for the hyperelliptic case (Lax equations will be explained in more detail in Section V.5; in
this section a Lax equation is no more than a neat way to write down the differential equations
describing an integrable vector field).
Proposition 2.7 If F(x, y) has the hyperelliptic form F(x, y) y2 f (X) for Some poly_ = _
nomial f (x) then the differential equations describing the vector fields XHI, are written in the
Lax form (with spectral parameter X)
X '
Hi A(,\) =
[A(,\), [Bi (A)]+] , (2.14)
where
V(A) U(A) O(A'v(A))
A(A) =
w (X) -
V (X
Bi(A) -
U(,\) [u(11) I
Ai+'
+
A(A)
83
and
rF.(/X, v
W(A) --
-
F-R-A)l +'
The spectral curve det(A(A) M Id) 0, which -
= is preserved by the flow of the vector fields
XHI',, given by t12 f (A) HPd(U(A), V(A))
is _ =
-
Proof
If we define the polynomial w(X) as stated above, then equations (2.12) axe rewritten as
V(A)
XH ',u(A) =
2W(A,v(A))v(A) -
2u(,\) W(A, v(A))

U(A) [U(A)I+l
Ai+1
W(A)
XH'jV(A)=-WGX1V(A))W(A) [U1111
),i+l
+ U(A)
1W (A, v W)
U(A) [U(),)]+]
Xi+1
+
upon using
(9F
(A, v (A)) = 2v (A).
5y-
Rom (2.15) we can compute X Hj
P w(A); observe how in this calculation the explicit dependence
on F disappears completely!
F (A, (A))
v
9
XH
i
WOO =
-X3
'P
j 1 I U(A) +
V(A) v(A)) XH Ou(A )

= -2
l (A) u
v(A) 1 [F(A,U(A) UN I
XV
41
+
+
+
= 2
[v(A) [O(A) O()k'v(A))
U(A) 1 1 w(A) U(A) Aj+j
+
-
I +
WN
= 2w (A)
I W (A, v (A))
VN
U(A) [ (I` 1
U
Ai+l
+
1 +
2v(.\) WO,, v(A))
U(A) [U(),) 1
Ai+1
+
1 +
-
This leads at once to the above Lax equations. The associated spectral curve is computed as
follows:
v2 (,\) f (A)
I
_
2 2
det(A(A) -
p Id) =
/4
_
_V (A) + U( A )
U(A) +
I 2(A)U(A)f(A)
V _
= IA2 _
U(,\)
= /42
_
_
f (A) -
HFd(U(A), V(A))-
For example, if we restrict ourselves to d = 1 (i.e., one degree of freedom), then u(A)
A + uO, v(A) =
vo and
HPl (UO' VO) =

(V2 f (A)) mod u (A)
=
(V20 -
f (A)) mod(A + uo)

= V20 _
f(_UO)'
and is the standard bracket on C', so we find that for W hyperelliptic case in
= 1 the
one degree of freedom corresponds exactly to the case of polynomial potentials on the line.
84
3. The geometry of the level manifolds
In this section we determine the nature of the (general) level sets of (C2d, J.'.JW AFd) ,
These level sets are the fibers of the momentum mapSpecAPd where, as dwe have
C2d _+
seen, SpecAPd is isomorphic to Cd. In particulax they axe independent of the polynomial
V(x, y) which dictates the Poisson structure (the impact of W (x, y) presents itself only at the
level of the integrable vector fields and is discussed in Section 3.5). It was generally believed
that the general level set of an integrable Hamiltonian system with polynomial invariants is
an (affine part of) a complex torus (Abelian variety) or an extensions of a complex torus
by C*n (see Chapters IV and V). It will turn out that the level sets encountered in these
examples are of a different nature. We will also look at the real parts of the smooth fibers:
whereas the real parts of Abelian varieties axe quite special (see [Sil]), it will turn out that
we find here a very rich class of topological types which appear as real parts of the fibers of
the momentum map.
3.1. The real and complex level sets
Since Spec(APd) is isomorphic to Cd and since the functions Ho,..., Hd_1 are indepen-
dent, the fibers over closed points are given by the level sets of HO, Hd- I or equivalently . . .
,
by the level sets of Hpd(U('\), V(,\))- Since HPd(U(,\), V(A)) is defined as F(A, v(A)) mod u(,\),
the fiber over an arbitrary polynomial c(,\) of degree smaller than d is the same as the fiber
over 0 for HFI,di Where lr(X7 Y)F(x, y) c(x). Therefore it suffices to describe the fiber
= -
lying over 0 for all polynomials F(x, y). We denote this fiber by "T'Fd; thus, by definition,
TP,d is given by
F(A, v (A))
TF,d (U(A), V (.X)) C C2d
u (,\) I -
0
I -
(3.1)
The real level sets axe defined as follows: we denote the fixed point set of the complex
conjugation map r : C2d _+ C2d : z -+ 2 as Fix(,r) and we define
TF d =
Fix(7-) n FPd- (3.2)
(TF,d, T) is real
algebraic variety (see [Sil]), whose real Part is TP d; in fact, if F(x, y) is a
a
real polynomial Y7d are nothing but the level sets of the corresponding
then the level sets
real integrable Hamiltonian system ( btained by replacing in all definitions C by R, see
Paragraph 11.4-2). We determine the non-singular real and complex fibers in the following
two propositions.
Proposition 3.1 If the algebraic curve rp C C2 defined by F(x, y) = 0 is non-singular,

then the fiber ): d C C2d is also non-singular.
Proof
Y Fd will be smooth if and only if H.Fd is submersive at each point Of )7 dl i.e., if and
only if
Rk
(9Ud-1"'*' 8UO'avd-1'***' 5vO)O<i<d-1 d, along -FF d,
85
From the proof of Proposition 2.3 and the definition of -FRd,

F,
the i-th and (d + i)-th columns
of this matrix are respectively given by
F(A, v (A)) OF
Ai modu(A) and Aj (,\,v(,X))modu(,\).
UN VY
It is therefore sufficient to show that if r, is smooth then the dimension of the linear space
OF
R,(A)F(X'v('\)) +R2(A) (A, v (M) mod u (A), deg R (,X) < d, (3.3)
U(A) Oy
equals d. Let A, be the distinct roots of u(,\), Ai having multiplicity si. We claim
that
F(Ai, v(,\i)) c9F
= 0 and (,\i, v 0 (3.4)
U(,\i) 19Y
cannot hold simultaneously if IPF is smooth. For otherwise (xi, yi) =
(Ai, v(Ai)) would be
'9F
a singular point of rp: if (3.4) holds then clearly 6-Y (xi, yi) 0, = but also F(xi, yi)
'9-F
yi) 0 because in this F(x, yi) has double
67X (xi,
= case a zero at x =
xi.
The dimension of investigated by using the fact that for any polynomial p(,\),
(3.3) is now
the value of AX) just p('Xi), and the values of the first si
mod u(,\) at \i is 1 derivatives -
of p(,\) mod u(A) at Xi are given by the values of the corresponding derivatives of p(,X) at \i
(si is the multiplicity of Ai in p(A)). Let us suppose that the different roots of u(A) are
ordered such that A, ......\t are also zeros of !M- (A, v, (A)), while At+,, A, are not. As a . . .
,
OY
first restriction, let Rl(,X) (resp. R2(,\)) be such that its first si 1 derivatives vanish at Xi -
for t + I < i < (resp. I < i < t). As a further restriction it is (by the first restriction and
r
as (3.4) cannot happen) now easy to see that R, (,\) (resp. R2 (,X)) can be determined such
that the polynomial given by (3.3) and the first si I derivatives of (3.3) take any given
-
values at Xi for 1 < i < t (resp. t + I < i < r). These d conditions are independent, hence
the dimension of (3.3) equals d and )7& is smooth. I
Proposition 3.2 The curve r, C C2 is non-singular if and only if the fiber TFd C C2d
is non-singular.
Proof
If rF has a singular point P, =
(xi, VI), choose for i =
2,. .
.'
d -
1 an extra point
Pi (xi, yi) on IPF and define (u(A), v(A)) S((xi, yi),
=
7 (Xd , Yd))
= E .97,vd. All polynomials
...
given by (3.3) vanish for X xi, hence they span a linear space of dimension less than d. Thus
=
HF,d is not submersive at (u(A), v(,\)) and AFd is singular at this point. This shows the if part
of the proposition; the only if part is proven verbatim as in the real case (Proposition 3.1). 0
It will be seen that a clear understanding of the structure of the complex level sets 97Fd
(for IPF smooth), leads also to a precise description of their real parts )7 d*
86
3.2. The structure of the complex level manifolds
d
We will show that -FFd is an affine part of the d-fold symmetric product Sym rp of
17F C C2. Recall (e.g. from [Gun]) that SyMd rF is defined as the orbit space of the natural
action of the permutation group Sd on the cartesian product rdF rF X x l7p (d factors),
i.e.,
S ym,dr.F = rdF ISd
Symd rF inherits its structure as an algebraic variety from the algebraic structure of r,
d
Moreover if IFF is smooth then the same holds true for Sym IPF: namely each point P
(PMJ,..., p.,n,) C
I r
Symd
I, (with all Pi different; mi is the multiplicity of Pi in P) has a
neighborhood which is neighborhood of ((Pm1),...' (Pm-)) in Sym7l IPF x

isomorphic to a
...
x sym- rF, and a point (Pim') on the diagonal of symm, rF admits local coordinates
given by the mi elementary symmetric functions of the mi coordinate functions on 17'pl.
Proposition 3.3If the algebraic curve rF in C', defined by F(x, y) 0 is non-singular, =
d
then -FFd is biholomorphic to the (Zariski) open subset of Sym ]pF, obtained by removing
from it the divisor
x(Pi) x(Pj) with Pi =A Pj,
(
or
f
=
DF,d= (P]L,...'Pd)J3iJ:1<i<j! d,
Pi =
Pj is a ramification point of x
In particular FPd is irreducible.
Proof
0 Construction of the map OFd : -T-7Fd -+ SYMdrF \ DFd
Given a point (u(A),v(A)) E TFd, a point in SymdrF is associated to it as follows: for
F(A,v(X))
every root Ai of u(A) one has F(Ai, v(,Xi)) =
0, because I U(A) I -
=
0, so each root \i of
u(A) determinespoint (Ai, v(Ai)) on r, Thus there corresponds to (u(A), v(A)) E FFd an
a
d
unordered set of d points (P.1, Pd) E Sym r.F, where Pi is defined by (x (Pi), y (Pffl
(Ai, v(Ai)). Clearly, if x(Pi) x(Pj) then Pi =
Pj; therefore, to show that (P,,..., Pd)
=
stays away from DFd we only need to prove that Pi Pj cannot occur for i :/= j if Pi is a =
ramification point for x, i.e., y (Pi)

if is a multiple root of F (x (Pi), y) (as a polynomial in y) -
As Pi Pj (i : :4 j) implies that u(A) has a multiple root x(Pi), in such a case F(x, y(Pi))
=
F(,\,v(,\))
would have a multiple root x =
x(Pj), again because [ U(A) I -
= 0. If moreover Pi is a
"
ramification point of x then also
ey
(x (Pi), y (Pi)) = 0 and it follows that (x (Pi), y (Pi)) is a
singular point of 17F, a contradiction.
0 DFd is a divisor on SyMd IPF

This means that DFd is given locally holomorphic function. If
as the zero locus of'a
d} be decomposed as S, U U Sn, such
(PI Pg) E VFd let the set of indices 11,
I... I
. . .
,
...
that all points Pi, with i running through one of the subsets Sj, have the same x-coordinate,
which is disjoint from the x-coordinates of the points which correspond to the other subsets.
For each Pi (i d) let xi denote the lifting of x to a small neighborhood of (P1,...' Pd)
=
(corresponding to the factor Pi). Then a local defining equation Of VFd is given by
n
H J1 (Xj -
Xk) = 0-
i=1 J,kESi
j<k
87
0 OFd is a biholomorphism
We first construct the inverse Of OFd, which is closely related to the map S, as given by (2.3).
Let (Pi Pd) E SYmd ]PP \ DPd- Clearly u(A) is taken as
d
U(A) =
MA -
x(pi)). (3.5)
i=1
If all x(Pi) axe different then v(A) is uniquely determined the

as polynomial of degree d -
I
whose value at A =
x(Pi) is y(Pj), i.e., v(A) is given by
d
A -
X(Pk)
V(, 1: Y(PI) 11 X(pl) -
X(Pk)
(3.6)
1=1 kol
and is holomorphic there. If two values coincide, x(P2), then P,

say x(PI)
P2 is not
= =
a ramification point (since we stay away from DFd), hence the equation F(x,y) 0 can =
be solved uniquely as y =
f (x) in a neighborhood of P, P2. For Pl' and P2' in this
=
neighborhood, substitute
d
f(X(Pl)) =
f(x( P1 )) + (X(Pj) x (pl)) (X( P1 )) + 0 (X( P11) X(PI ))2, (i =
1, 2)
dx
for yj and Y2 in (3.6), to obtain that v(A) has no poles as Pl', P2' -+ P3., hence extends to
a holomorphic function on the larger subset where at most two
points coincide. Since the
complement of this larger subset in SymdrF \DFd is of codimension at least two, v(A) extends
to aholomorphic function on SyMd ]pF \ DFd- It also follows that this holomorphic function
is the inverse Of OFd on all Of SYM'1 ]PP \ I)Fd: if the point Pi has multiplicity 3j, then the
first si I derivatives of v(A) at x(Pi) coincide with those of f (A) at x(Pi), hence
-
F(A, y(Pi))
has a zero of order si at A x(Pi). Finally, the inverse of a holomorphic bijection between
=
complex manifolds is always holomorphic (see [GH] Ch. 0.2), hence OFd is a biholomorphism.
Since the symmetric product of any non-singular curve is irreducible, the same holds true
for -FPd- 0
Having a biholomorphism between affine varieties does not imply that this is an iso-
morphism. However this is so in the present case, as can be proven by using Newton's
interpolation Theorem. This was done by Mumford for a special case (where the curve is
hyperelliptic of genus d); since his (quite long) proof applies verbatim to the general case it
is not repeated here (see [Mum5] pp. 3.23 3.25). -
Lemma 3.4 Even if -FPd is singular its dimension equals d.
Proof
We may still associate to each point in Fd a divisor consisting of d points on the
singular curve IPF defined by F(x, y) = 0. The d-fold symmetric product of such a curve is
singular but is still of dimension d. Therefore it suffices to show that the constructed
map is
finite on a Zaxiski open subset. In fact, if we stay away from the singular points in both TFd
and in Sym drp then the proof of Proposition 3.3 applies to show that this map is bijective
on these subsets. Thus all fibers (over closed points) have dimension d. I
Proposition 3.3 and Lemma 3.4 finish the proof of Proposition 2.3 upon applying Propo-
sition 11.3.7.
88
3.3. The structure of the real level manifolds
SinceY Fd is given as TFd n Fix(-r), it Consists of those polynomials (u(X), v(,X)) E J: F'd
R"
whose coefficients are all real. We figure out what this means for the corresponding point
in SymdrF-
Proposition 3.5 Under the biholomorphism OFd, the real level manifolds TFd correspond
to the set of all unordered d-tuples of points (Pi, Pd) on I7p, consisting only of real points
Pi E R2nr .F and complex conjugated pairs Pi P each ramification point (of x) occurring
at most once, and x(Pi) x(Pj) only if Pi=
Pj Moreover its manifold structure derives
=
-
from the structure of the d-fold symmetric product of rp.
Proof
u(A) only if its roots consist only of real roots and roots which occur
is real if and
in complex conjugate pairs. Obviously, if v (A) is real, v (A) and its derivatives take complex
conjugate values when evaluated at complex conjugate points (and real values at real points).
Also, if a polynomial of degree smaller than d is specified in d points that are real or occur
in complex conjugated pairs then that polynomial is real (it is sufficient to prove this in case
in which all points are distinct, which is easily done by using (3.6)). Since v(xi) yi, this =
means that the real polynomials (u(A), v(,\)) on TFd correspond to those points (PI, Pd) ...
5
in Sym drF consisting of real points Pi (x(Pi), y(Pi)) E R2 and complex conjugated pairs
=
Pj =
Wpj)' Y(pj)) =
(xX(Pk), i(A))= Pk, but not belonging to DFd, i.e., the multiplicity
of each ramification point (of x) is at most one, and x(Pi) =
x(Pj) only if Pi =
Pp I
Proposition 3.5 can be used to obtain a precise description of the topology of the real
2
level manifolds -FF d, as we show now for d = 2 (for d 1, -TF d is just IPFn R the real part
=
,
2
of 17F). For a fixed F such that rp is smooth, let the connected components of rpn R (if
any) be denoted by r,, , r,
and define for 1 < i, j, k < s and i <
. . .
roo =
(P, P) I P E 1Pp, x(P) 0 RJ,
I7ij =
(PI P2) E ri x rj I X(PI) X(P2)
i
= =* P1 =
P21i
]Pkk -
(pl, P2) .
rk x
S2
rk
I x(PI) =
x(P2) => (Pi =
P2 and is not a ramification point) I -
Then the union of ]POO with all the sets l7ij and ]Pkk is easy identified with 0F,2(AF,2), the
surface to be described. One observes that the only paths in it which are not contained in R2,
in in fact roo connects exactly the surfaces rlk. Therefore, if i then rij is not
are roo,
connected to any other IF .... .rkk, nor to ]POO.
Therefore we first concentrate I7ij, say on r12- If the intervals x(rl)
on such a subset
and X(I72) are disjoint, r, r2, 1712
then 1712is either homeomorphic to a torus, a
= x SO
cylinder or a disc, depending on whether the components r, and 172 are closed or open. If
x(ri) and x(r2) have a point P in common, then one finds again these surfaces, but with a
number of punctures (holes), equal to
#IQ E ri I X(Q) =
X(P)J.
89
if x(ri) and x(r2) have an interval in common, r12 may even disconnect in different pieces.
The structure of these pieces is easily determined from a picture of the real part of the curve.
Namely, on a square representing rl X 172 the divisor I (PI, P2) E ri X r, I x (PI)
, x (P2)1 is =
drawn by counting points on the vertical lines x constant, the only care
= one needs to take
is that if 171 (or r2) is closed, then an origin should be marked on it, and if one passes this
origin, one needs to pass over the corresponding edge of the rectangle. The following table
shows some examples (all possibilities for which 1P, and r2 are closed, and for which x is 2: 1
when restricted to 171 and r2)-
r, and r2 Divisor Component 1712 Picture
CD CD 1-1 torus
torus minus point
C (torus minus disc) + disc C ) 0

CD
C) two cylinders
CD FV1
I A I cylinder + disc
00
C) VN two discs
00
Table 2
In the same way ]Pkk is investigated by drawing the divisor
P, P2
I (PI, P2) E ri x ri I x(PI) =
x(P2) and
( P, P2 is
or
a ramification point of x
on a rectangle representi r,' x ri. Either triangle cut off from the rectangle by its main
M,
diagonal then represents S2 and I7jj is the complement of the divisor in the triangle.
For example, consider a as in Figure La below. Then Figure Lb shows

component ri
a torus with a circle (the
on it
anti-diagonal of the rectangle), which is the divisor D to be
removed. The resulting piece r1l is drawn in Figure Lc and is redrawn in a simpler way
in Figure I.d. For every I7j such a piece is found and will be glued to ]POO precisely along
90
the part of its boundary which comes from the diagonal in the rectangle (the solid lines in
Figure Ld).
D
b --- ---
F, A22
A A
a
_D a
A
1 a I 2
D
b a D
(a) (b) (c) (d)
Figure 1
In order to explain how roo is described, we recall the classical picture of a (smooth, complete)
algebraic curve P. An equation F(x, y) 0 of such a curve defines a m : I ramified covering
=
map to P' by (x, y) -+ x, when m is the degree

of F(x, y) in y. This may be visualized by
drawing concentric spheres (called sheets), on which there are marked some non-intersecting
intervals (called cuts, every cut is equally present on all sheets). The topology is such that
if you are walking on a sheet i and pass a cut j (from a fixed side) then you move to a sheet
pj (i), each pj being a permutation

of 11, ml. It is clear that the datum of cuts and their
. . .
,
corresponding permutations determines the topology of the curve completely. Since each cut
connects two ramification points (of x), these cuts may, for a real curve, be taken on the real
axis and orthogonal to it.
roo is now given as follows. Consider the described picture for the smooth completion
PF Of 17F. Clearly the conjugation map interchanges the upper and lower hemispheres and
is fixed on the equator(s) IP E PF I x(P) E R U ool. It follows that the open upper (lower)
hemispheres give precisely ]POO. A convenient way to represent them is by drawing a disc for
each upper hemisphere and labeling the different parts of the boundary which correspond to
the horizontal and vertical cuts. A moment's thought reveals that the different sheets are to
be connected along those lines which correspond to the vertical cuts, while the pieces I7kk
are to be connected to the corresponding horizontal cuts.
This gives a topological model of
rOO U U8k=l rkk as a disc with holes. The following example may highlight the different steps.
Example 3.6 hyperelliptic curve F(x, y)

We consider a yl f (x) 0, where f has five = - =
real zeros and one pair of complex conjugate zeros. The curve has genus
three and its graph
and related representation as a cover of P1 are given by
X2 X3
CD `4 X7
x
"6
X7
x,
Figure 2
from the graph. For roo get

where the imaginary ramification points (of x) are not seen we
two upper hemispheres
91
H21
DV DVI
H, H2
H,3 H23
V, V,
V2' V2'
Figure 3
which become one disc after gluing the vertical cut V11, V21- We also get two subsets 1713. and
r22 given as in Figure Ld by
--------
--------
H,' H,' H,1 H11

--------
--------
Figure 4
and one disconnected piece 1733 (since 00 IE 13)-

3 3
H H
Figure 5
Now glue Figures 3, 4 and 5 according to their labels Hj
to find a disc with two holes. Since
the other components Of AP,2 are direct products of the real components, we find that
AF,2 one torus + two cylinders + one disc with two holes.
It is shown in the same way that, if F(x, y) is of the form
n m
F(x, y) =
y2 + ]I (X _
Ce,) 11 (X2 + 3
i=1 j=j
with aj, #j E R (all ai being different, as well as all pi2)' then
AP,2 ((n-l)/2)
2
tori + V
2 cylinders + one disc with g -
1 holes if n is odd,
AF,2 (n)
2
tori + one disc with g holes if n is even,
where g is the genus [n+] 2

+ m -
I of the curve F(x, y) = 0.
92
3.4. Compactification of the complex level manifolds
We now discuss the (smooth) compactification of the manifolds TPd. There is one
obvious and natural compactification, namely the compact manifold Synid PF, a defined in
d
similar way as Sym rp; as above rF denotes the smooth compactification of rp. However
,FP,d has the disadvantage that none of the vector fields XH, extends holomorphically to it
-
a compactification such that at least one of these vector fields extends

in a holomorphic way
to it, will simply be called good. The interest in good compactifications is that it allows one
to integrate the corresponding vector fields in terms of theta functions, or degenerations of
theta functions. The purpose of this paragraph is to show that even for very simple choices
of F(x, y), a good compactification Of -FFd does not exist. We believe that this is true for
almost all choices of F(x, y). A special class of examples for which a good compactification
does exist is considered in Chapter V.
At firstcompute, for fixed F(x, y) how the vector fields XH. behave on the compact
we
d
manifold Symd PF, which relates to TFd (which we identified with Sym IFP\DPd) as follows:
d
Sym PP = TFd U DFd U EFd-
d
Here E)Fd is the closure of E)Fd in Sym rF andEpd is a divisor whose irreducible components
SF,d(00i) correspond to the points ooi in PF \ rF, namely
9F,d(00k) "
f(00ki& ... A) I Pk E Vp for 2 < k < dj
Each vector field XH, being a polynomial vector field on CId' it is holomorphic on TPd- We
determine its behavior along the irreducible components of f)Fd and 9Fd, which may be done
by computing the order of vanishing Of XH, at a generic point of each component, which in
turn is done by using local coordinates at such a point.
Proposition 3.7 Every vector field XHj has a simple pole along all irreducible components
of tpd. It has a zero of order pk along Gd(00k) (i.e., a pole of order -pk if pk < 0), where
pk -
vk + d + I if Vk < 0,
Pk = (3.7)
Pk -
Vk + I if Vk > 0;
at 00k, and vk is the order (resp. the

pk is the order (of vanishing) of Pj'v-
i9v (x, y)
of x at 00k
order Of X -
X(000 if x finite
is at 000-
Proof
We first write down the vector field XH, at a generic point (u(A), v (A)) E F.Fd; the gener-
icity condition taken here is that for 0Fd(u(A),vGX)) ((XIIYI)I. I (XdYd)) all xi arediffer-
:--: --
ent and none points (xi, yi) is a ramification point

of the Varying the point (u (A), v (M
of x.
in a small neighborhood, each xi gives a local coordinate on a neighborhood Uj C rF Of
(xi, yi) as well as a local coordinate on a neighborhood U C 37Fd Of ((XI Y0 (Xd Yd)) I I
...
I I
Since on the one hand the derivative of u(A) =

rld=l(,\
k
_
Xk) at A =
xj is XH,u(xj)
(xj -
xj) XH, xj, while on the other hand, direct substitution in (2.12) gives
d
o9F
XHAXj) =
i(Xj'Yj) F, UkXjk-i
k=i+l
93
we find that
OF
X.U,xj jj(Xj Xj)-j_
19Y
(3.8)
10i
where o-j(.,Bj) is the i-th symmetric function in x,.... ) Xd, evaluated at xj = 0.
The right hand side of (3.8) has at a generic point Of DF,d a simple pole, hence each
vector field XH, has a simple pole on (every component of) DPd. The behavior of XH, along
ISF,d is slightly more complicated since it depends on F(x, y), and may even behave differently
on each component 9Pd(00k). For a generic point in a neighborhood of a point of 9Fd(ook),
let us introduce coordinates xi as above. If we denote by Mk and vk the integers introduced

in the statement of the proposition, then clearly x, is given in a neighborhood of 00k in terms
of a local parameter tj at 00k as x, t'Ik (vk < 0), or as x,
=
cl + t'
Ik (vk > 0), depending =
on whether x is infinite in a neighborhood of ook or has a finite value cl E C at 00k; also

"
(xl (t) yj (t)) t"k (f, + 0 (t)) with
= 0. We define for 2 < j :! , d local parameters tj
&Y I
(centered at Pj, which may be assumed to be generic) by xj x(Pj) + tj. Direct substitution =
in (3.8) yields
XHj tl t'j' (Ci + 0 (4)),

=
XHjtj =
Cj + 0(tl)5 (j =
2,..., d);
where Pk is defined in (3.7). We conclude that XH, has a zero (resp. pole) of order jPkj along
46F,d(00k) if pl, ! 0 (resp. Pk < 0)- 1
Thus we Sym drF is not a good compactification, since 0 vector fields

have shown that
XH, have at least pole along T)Pd. This divisor can be contracted in some cases, as we
a
will show in Chapter V. The following example shows that a good compactification does not
exist in general.
Example3.8 Let F(x,y) yl+f (x), where thedegree off is at least three, andlet d= 2.
=
To show that J7F 2 has no good compactification we use some results about algebraic surfaces
which can be found in [Har] Ch. V. Suppose that Y is a good compactification of YC F,2
then Y and Sym2 PF are birational; for surfaces this means that there exists a finite series
of monoidal transformations (also known as blow-up's) which transforms Y into SYM2 P J'.
2
Then there exist (Zariski) open subsets U C Y and V C SYM PF to which all these monoidal
transformations restrict as isomorphisms and the vector fields on U and V correspond exactly
under this isomorphism. In particular DP,2 is entirely contained in the complement of V and
must be contracted by one of the monoidal transformations, so at least we know that the
genus Of DF,2 must be 0 (only P"s can be contracted).
We may however compute the genus Of DP,2 directly. Recall that it consists of the
points (PI, P2) with x(P.1) = x(P2) for which P, =A P2 or P, =
P2 is a ramification point, so
its smoothness is easy checked. However the map x expresses DF,2 as a 3 : I cover of P1,
ramified at the n deg f points
=
(xi, yi) for which f (xi) 0 (and at infinity if n is not divisible
=
by 3). So it has the same ramification divisor as PF, hence genus(f)F,2) =

genus(rF) > 0, a
contradiction.
Bxample 3.9 In the one-dimensional 1) the level manifolds are punctured Rie-
case (d =
mann surfaces and have unique compactification. If the genus of such a Riemann surface
a
exceeds one, then it supports no holomorphic vector fields, so for d I good compactifications =
Of -TFd rarely exist.
94
As anapplication of (3.8), let us show how the vector fields can be integrated on these
real level sets. Summing up (3.8) over all j (and for any W) we find that for any fixed integers
i,r < d,
d d d
XW
HjXj -i+k-I
E X
3 ap UkX;
j=1 W(xj,yj)-E'-'(Xj,yj) j=1 k=d-j+l
xj
-
XI
Therefore the d functions
d-I
X,r XH O Xj
Xr =
E W(xj'YjA-'(xj'Yj)' i
r =
0,..., d -
1, (3.9)
Oy
have linear dynamics in time and lead to the integration of the vector fields XH, along the
(connected components of the) real manifolds 7:F d*
3.5. The significance of the Poisson structures d
We have constructed for any positive integer d and for any F (x, y) E C [x, y] \ C [x] a
family of compatible integrable Hamiltonian systems (C2d, 1.,.}W
d i AFd)
which is indexed by
p(x, y) E C[x, y] \ 10}. For a fixed F(x, y) all vector fields in Ham(f-, -I'd AFd) ,
are tangent
to the d-dimensional fibers of the momentum map C2d
fixed W these
-+ Spec AFj. Since for a
vector fields generate at a general point the tangent space, the vector fields obtained for one
choice of W can be written down in terms of another choice of W. This relation between the
vector fields of the different (compatible) integrable Hamiltonian systems is given explicitly
by the following proposition.
Proposition 3.10 For fixed F(x, y) and d let Hi (i 0,..., d 1) denote the polynomials = -
on C2d, defined in (2.10), and let Xj' and Xf denote their Hamiltonian vector fields with
respect to 1', Jd and I-, j1d. Then the transfer matrix TI'P, which is defined by
(X p Hd-11
...
I X. F'J") (X1=
Hd-11-"
X,HO ) 7-1 0'
is given by
(-Ud-1 -Ud-2 -Ud-3 -UO

1 0 0 ...
0
where M 0 1 0 0 (3.10)
w(M, v(M)), ...
0 ...
0 1 0
The general transfer matrices 7- 2 are immediately computed from (3.10) upon using the
cocycle identities
-4 and -"P
7W 2 7W21 X'
7WV2 7W2371
95
Proof
It suffices to express 711 in the neighborhood of a generic point (u(A), v(,\)) in terms of
any local coordinates. At such a generic point all roots xi of u(A) axe different and serve as
coordinates; we denote v(xi) = yj as before. Then by (3.8),
XH o = WOXH with A P =
diag(V(xi, yj), W(Xd, Yd));
XH',o denotes the matrix with entries (XH")ij =

XH;,xi and XH X1.
H
The above formula
implies that V, is given by
TjW =
(XH)-l,&WXH
W
= VAWV-1
=
VW(AX, V(A'))V-l
W (VA'V-', v(VA'V-'))
WA V(M)),
where M == VA'V-1 is easily checked to have the form announced in (3.10). Step (i)
requires some extra work (one uses (3.8)); also we have introduced the notation V for the
Vandermonde matrix
d-1 d-i
(Xd-I
I- X2 ...
Xd
d 2 d-2 d-2
X1 X2 ...
Xd
V=
Remark 3.11 In the special case where d2 and W(x, y) x the transfer matrix has the
simple form
-
U1 -UO
1 0
and we obtain what Caboz et al. call a (p, s) bi-Hamiltonian structure (see [CGR]). Our def-
inition of compatible integrable Hamiltonian systems and Proposition 3.10 largely generalize
and clarify this concept.
96
Chapter IV
Interludium: the geometry
of Abelian varieties
1. Introduction
For the convenience of the reader who wishes to go on reading the rest of the text we
include here a chapter about Abelian varieties. There is nothing new in this chapter, our
intention was give a compact and coherent presentation of the theory of Abelian varieties in
to
a applications to integrable Hamiltonian systems. Our exposition is paxtly
form suitable for
algebraic partly analytic, we think that both approaches highlight different aspects of the
theory of Abelian varieties, see for example the theorems of Abel, Jacobi and Pdemann in
Paragraph 4.3. Moreover, for applications to the theory of integrable systems a reasonable
amount of understanding of both aspects and their interplay is needed. The main references
for the theory of Abelian varieties are [LB], [Kem], [Mum2] and [Mum3]. However the
relevant chapters in [ACGH], [GH] and [Mum4] are also highly recommended to learn this
subject.
We start by recalling the basic definitions of divisors and line bundles (on a complex
manifold) and recall how they are related. The sections of an ample line bundle are used to
construct embeddings in projective space and the dimension of this space is computed from
the Riemann-Roch Theorem. As an illustration we show that every compact Riemann surface
can be embedded in projective space, hence it is an algebraic curve. It is therefore
some
common not to distinguish between compact Riemann surfaces and algebraic curves, the
latter of course being assumed complete, non-singular, irreducible, reduced; we will also use
these terms interchangeably, choosing the term compact Riemann surface or algebraic curve
according to whether the analytic or the algebraic structure is more relevant.
In Section 3 we give the Pdemann conditions which tell which complex tori are Abelian,
i.e., can be embedded in projective space. The sections of an embedding line bundle are
explicitly described by theta functions and their number is easily computed. A lot is simplified
97

Chapter IV. The geometry of Abelian varieties
here because every effective divisor on an (irreducible) Abelian variety defines an ample line
bundle and the third power of an ample line bundle on an Abelian variety always provides an
embedding. In particular in the case of Abelian surfaces everything can be computed very
explicitly (except the embedding itself but that is where we win use the theory of integrable
systems for!), an ample divisor is then nothing but an embedded curve and its genus relates
to the number of sections of the line bundle defined by this divisor.
We also treat in detail the Jacobian of an algebraic curve since it is the most important
exaxnple of an Abelian variety. There axe basically two (very different) ways to define it,
there is algebraic definition (using divisors on the curve) and an analytic/transcendental
an
definition(using integration of differential forms over cycles). The fact that these correspond
to the same object is a deep theorem, due to Abel and Jacobi. We could not resist to
reproduce a proof of it here. We close Section 3 by discussing the Kummer surface of a

Jacobi surface and its 16r, configuration, which will be used in Section VIIA.
We use the following notation and terminology. On a complex manifold M we will use,
aside from 0, its sheaf of holomorphic functions', and M, its sheaf of meromorphic functions,
also the constant sheaves Z, Q, R and C and the sheaves 0*, M*, QP defined for each open
set U by
0* (U) = the multiplicative group of non-zero holomorphic functions on U,

*
M (U) = the multiplicative group of non-zero meromorphic functions on U,
9F(U) = the vector spare of holomorphic p-forms on U.
We add M as a subscript in the notation when it is not clear from the context that we are
talking about a sheaf on M.
9
In this chapter 0 will only be used in this sense, so no confusion can arise with the
notation O(M) for the ring of regulax functions on an affine variety, which is used in other
chapters.
98
2. Divisors and line bundles
we discuss divisors, line bundles and the way in which they are related. We
In this section
also explain how the holomorphic sections of a (very ample) line bundle lead to embeddings
in projective space. We fix a compact complex manifold M of dimension n, the dimensions
n = 1 and n 2 being the most important for us.
=
2.1. Divisors
A subset V of M which is locally given by the zero locus of a holomorphic: function

is called analytic hypersurface. V is called reducible if it is the union of at least two
an
hypersurfaces; otherwise it is called irreducible. A divisor on M is a finite (formal) sum
D =
1: aiVi (ai (E Z)
of irreducible analytic hypersurfaces of M. Two divisors are added in the obvious way and
the resulting group, which is called the group of divisors on M, is denoted by Div(M). If all
ai are positive the divisor is called effective. For curves (n 1), the integer E ai is called ==
the degree of D and is noted by deg D.
If V is an irreducible hypersurface then to any meromorphic function f there can be

assigned an integer, "the order of vanishing of f along V" as follows. By definition, V has a
local defining function for some neighborhood around any point P of V; if g is such a local
defining function then this integer is the largest integer n for which gn divides f in the ring
of germs of holomorphic functions at P. This is well-defined because it turns out that this
integer is independent of the point P and the local defining function g. We denote this integer
by ordV f and say that f has a zero (pole) of order a if ordV f a > 0 (ordV f -a < 0). = =
Then a divisor (f) can be assigned to every meromorphic function f by
(f) =
1: ordv f -
V,
V
the (finite) sum running over all irreducible analytic hypersurfaces V for which ordV f :74- 0; it
is not hard to show that deg(f 0 for any meromorphic function f on a compact Riem
surface.
holomorpbic top-form w E Qn (M), ordv w is defined as follows. If w

For a f,,,dzi A =
A dzn coordinate neighborhood U,,, then we define ordv w

on some ordv f,, for any =
irreducible analytic hypersurfarce V intersecting U,, (again this is well-defined because the
transition functions lie in 0*). Then the divisor of w is the effective divisor
E ordv w -
V,
V
the (finite) sum running over all irreducible analytic hypersurfaces V for which ordv W 7 0.
Div(M) admits a sheaf-theoretic interpretation, namely there is a natural isomorphisin
Div(M) --- HO (M, M* /0*).
99
To see this let f =

ffj be a global section of M*10*, then ordv f,, =
ordv fa because
f.. E ow., n up)

b
and the divisor associated to this section is naturally taken to be
D =
1: ordv f,, -
V,
where the sum is taken over all irreducible analytic hypersurfaces V for which v n u,, = A o.
Conversely, given a divisor, a global section of M*/O* is defined over small opens by taking
the product of the local d,efining functions of all irreducible analytic hypersurfaces appearing
in the divisor (with the coefficients as exponents). Also Div(M) and HO(M,M*10*) are
easily seen to have the same group structure.
2.2. Line bundles
By a line bundle f- on M we will always mean a holomorphic vector bundle of rank

one over M. For any two overlapping open sets U,, and Ujq the transition Anctions gq
I
U, n Up -+ C* of L axe defined in terms of the trivializations 0,, : -7r- (U,,) -+ U,, x C by
9.,&-) =
(0. - E C*'
(L., is the fiber over z E u,, n U,a) and take values in C*; one has that g,, 'q E 0 *(u,, n q 3
and these transition functions satisfy the so-called cocycle identities
909fla
ga'3gj3'Yg-Ya
hence the transition functions represent a Oech 1-cochain which is actually a 1-cocycle because
-1
(61g.'a 1) g6 9 PE 9-ya =
g6eg-!qfg-Y6 =
1,
(6 is the coboundary operator). It is a standard result that to any set of functions 1gq I satis-
fying cocycle identities there corresponds a line bundle with these functions as transition
the
functions. Thus, the tensor product of two line bundles is again a line bundle and the set of
all line bundles on M up to isomorphism becomes a commutative group, the Picard group of
M, Pic(M). The inverse of a line bundleC is nothing but its dual and will be denoted byC*.
If other transition functions jh,61 are given for L, coming from triviahzations X,,, then
there are functions f,, E 0*(U,,), defined by X,, such that
f.
hag =
ag,
fo
i.e., is a Nch coboundary. The upshot is that Pic(M) =

HI(M, 0*) and both sets
have the same group structure.
100
To every divisor there corresponds a line bundle in the following way. Let V E -Div(M)
be locally defined by functions f E M* (U,,), then the corresponding line bundle [V] is
defined by the transition functions
gap
461
and the definition is clearly independent of the choice of the local defining functions. The
map
Div(M) -+ Pic(M)
is a homomorphism of groups, which is surjective if M is algebraic. Its kernel is given by
Ker[-] =
IV E Div(M) 13f E M(M) for which (f) =
D},
because, on the one hand, the transition functions for [(f)] are given by
f 1U.
90 =
Alb =
f lu",
and on hand, if the line bundle [D] is trivial, then the local defining functions fc, give
the other
rise to f for which (f )
a function D, by defining f I u. fc, on U, which is well-defined
= =
when the local defining functions axe chosen in such a way that the transition functions
g,,,6 1. If two divisors V and D' are defined to be linearly equivalent when V -1 V, if
[D] [D], then we see that if M is (in addition) an algebraic variety, then
Div(M) -
Pic(M)
-1
in a natural way. The map [-] corresponds under the identifications Pic(M) !-2-' HI (M, 0*) and
Div(M) -5-- H'(M,.A4*10*) to the connecting homomorphism. P in the long exact sequence
- - -
-+ HO(M, M*) -+ HO(M, M*/O*) f+ H'(M, 0*) -+ - - -
coming from the short exact sequence 0 -+ 0* -+ M* -+ M*10* -+ 0.
2.3. Sections of line bundles
Let L be a line bundle over M with projection 7r : C -+ M, vector bundle charts
0,, : -7r-'(U,,) -+ U,, x C and transition functions g,q. A (holomorphic) section of L is a
(holomorphic) map s : M -+ L for which -7r o s 1m. The composition

=
0" 0 s I U. : U" -+ 7r-, (U") -+ U" X C
defines a (holomorphic) map s,, : U,, -+ C, for each a. Two of these, say s,, and s,8, with U, n
U,3 76 0, axe related by s,, g,,9 s,3. Conversely, a set of holomorphic maps Is,, I satisfying s,,
= =
g,,,,6s,6 on each non-emptyu, nU,9 determines a global holomorphic section of the line bundle
L. The sheaf of holomorphic sections of a line bundle L will be denoted by O(L) and the sheaf
O(L) (9 QP will be denoted by QP(L), the sheaf of holomorphic differentials with values in L.
We also define the sheaf of meromorphic sections of L as O(L) M; then the meromorphic
sections are given by meromorphic functions IsJ satisfying s,, g,6so on each non-empty
101
u,, n U,6. Given two meromorphic sections s, t of L, the local defining functions Is,, I and It,, I
satisfy s,,It,, =
s,61t,9. Hence we can define a meromorphic function f (the "quotient" of s
and t) by f,, =
s,,It,,. For example, for D E Div(M), defined locally by f,, E M(U,,), we have
taken [D] to be the line bundle with transition functions gq =
fJil, so that f,, =
g,,,6f,6.
Therefore every divisor D determines a line bundle [D] and any set of local defining functions
of D define a meromorphic section sf of [D]. It is customary to write O(D) for the sheaf
0([D])-
Just as we did for meromorphic functions, we can associate a divisor to
every mero-
morphic section of a line bundle by first defining, for any irreducible hypersurface V c
M, ordV s ordV s,, for any a for which u,, n v =A 0, and then setting
=
(s) =
E ordv s -
V
V
where the sum runs over all irreducible hypersurfaces V C M. As before, ordV is well-defined
because the transition functions gq s,,Isg are in 0*(U,, n U,3). Taking up the previous
=
example again, the divisor of sf can be read off from the local defining functions f,,, giving
(sf) = D. More generally, if s is any
meromorphic section of [D], then (s) -1 D because
the "quotient" of sf and defines
meromorphic function whose divisor is exactly E) (s).
s a -
It follows that [(s)] [D] and that there is for any divisor V -1 D, a section s of [D] for
=
which (s) V. Notice that line bundles which come from divisors are exactly those which
=
have (non-zero) meromorphic sections. Since a section s is holomorphic if and only if (s) is
effective, it follows that a line bundle is the line bundle of an effective divisor if and only if
it has a non-trivial global holomorphic section.
Actually, instead of working with the space HO (M, 0 (D)) of holomorphic sections of a
line bundle [D], one can work with the space L(D) of meromorphic functions on M defined
by
L(D) =
If E M(M) I (f) +D > 0}-
To see this, fix wiy meromorphic section so of [V] for which (so) D. Then every holomorphic =
section s E Ho (M, 0 (D)) corresponds to a meromorphic function f, by taking the "quotient",

f, =
s1so, and
(h) +D =
(8) -
(80) + D =
(s) > 0,
so that Ei L(V) and the proof of the converse is similar. We have established a non-
canonical isomorphism
L(D) ++ HI (M, O(D)). (2.1)
Denoting by ID1 the set of all effective divisors linearly equivalent to D, we have the additional
correspondence
ID1 5--- P (L (D)) P (Ho (M, 0 (D)))
for compact manifolds M, because V E ID1 implies the existence of a function f E L(D)
such that V D + (f) and f is uniquely determined up to a constant. Therefore,
=
IVI is
a projective space; any subspaces of it axe classically called linear systems and ID1 itself is
called a complete linear system. The common intersection of the divisors in a linear
system
is called the base locus of the system.
102
2.4. The Riemann-Roch Theorem
The calculation of the dimension of HO (M, O(V)) is done by using the Riemann-Roch
Theorem. We will give this theorem here only for (algebraic) curves, postponing the analogous
theorem for surfaces to a later paragraph. To give three equivalent versions of this theorem,
we need some terminology. The holomorphic Euler characteristic X(L) of a line bundle L
over a compact complex manifold M is the integer
X(L) =
E(-1)-"dimH-(M, O(L)),
P>O
which is a topological invariant of L and M.

paxticular the holomorphic Euler characteristic
In
of the trivial line bundle over M is denoted
by X(Om) and is a topological invariant of M.
Also, every compact complex manifold M has another distinguished line bundle, its canonical
line bundle KM, defined as the line bundle corresponding to the divisor of any holomorphic
top-form on M; this line bundle is well-defined because the "quotient" of two holomorphic
top-forms defines a meromorphic function on M which establishes the linear equivalence of
their divisors. The divisor of any holomorphic top-form is called a canonical divisor and will
be denoted in the same way as its corresponding line bundle (although this divisor is only
uniquely determined up to linear equivalence). Notice that the holomorphic sections of KM
n
correspond to the holomorphic top-forms on M, 0 (Km) c ' QM (isomorphism of sheaves). -
Then the Riemann-Roch Theorem for curves can be stated in the following equivalent
forms.
Theorem2.1 LetI7 be a compact Riemann surface ofgenus g, V be a divisor on rand Kr

a canonical divisor on 1P. Then the following equivalent formulas hold:
(i) XQD1) X(Or) +deg V,

=
(ii) dim HO (r, o (D)) dim HO (r, n, (-v)) + deg D g

= -
+
(iii) dim HO (r, O(D)) dim HO (r, O(K. D)) + deg D
= - -
g +
Proof
Instead of proving the Riemann-Roch Theorem, we prefer to show the equivalence of the
three formulas and to use this theorem to compute the degree of the canonical bundle on a
curve.
First, the equivalence of (ii) and (iii) is obvious from the isomorphism O(Kr)
To show the equivalence between (i) and (ii) we need another fundamental theorem, the
Kodaira-Serre duality Theorem, which we will formulate in a more general form, since we will
also use it for studying sections of line bundles on surfaces.
Theorem 2.2 Let M be a compact complex manifold of dimension n, then the vector spaces
Hq(M, W(L)) and Hn-q(M, gn-p(,C*))* are isomorphic.
For the curve r, n equals 1 in this theorem. Taking p = 0 and a line bundle L on r, we find
H q(r, o(,c)) =
Hq(r, flo(L)) -5-- Hl-q(]p, 01 (L*))*, (2.2)
so that Hq(r, o(,c)) = 0 for q > 1. Therefore, the holomorphic Euler chaxacteristic of C, X(,C),
equals
X(,C) = dim HO (r, o(r_)) -
dim H1 (r, o (,q).
103
On the other hand, for q = 1 (2.2) gives H1 (r, o(,c)) - --'HO (r, n, (,c*))* so that
X(L) =
dimHo(]P, O(L)) -
dimHO(r, n, (,c*)).
Taking for C the line bundle [D] and the trivial line bundle over 17, this gives respectively
X(['D]) = dim Ho (r, o (D)) -

dim H'(r, Q1 (-v)),
X(00 = dim Ho (r, Or) -

dim Ho (r, Q1) = I -
dim Ho (1P, Q').
The equivalence of (i) and (ii) is established if we can show that dim Ho (r, Q') =
g. To show
it, we consider the short exact sequence of sheaves
0 -+ C -+ 0 -4 Q, -+ 0,
which gives a long exact sequence in cohomology, namely
0 HO (r, C) Ho (r, o) Ho (r, n')

H' (r, q H1 (r, o) H' (r, P3.)
-+ H2 (]p, C) _+ 0,
because H2 (r, 0) = 0 from KodairaSerre duality. Also we know that
0 0 1 I
H (r, c) 5--- H (r, 0) 5--- H (r, n ) 5--- H2(r, q c,
c
and
H' (1P, 0) c--' Ho (17, Q').
Therefore, counting dimensions in the above exact sequence, we find
dim Ho (r, r2l) dim H' (r, q =

g,
proving the claim.
We proceed to calculate deg Kr, the degree of the canonical bundle on a curve r. Taking
D = Kr in the Riemann-Roch Theorem,
dim Ho (r, O(Kr)) dim Ho (17, 0) + deg Kr -
g + I
2 -
g + degKr.
Using Kodaira-Serre duality again,
deg Kr = dim HO(r, O(Kr)) +g -

2
=
dimHO(r, Q'(Kr Kr)) -
+g -
2
=
dimHO(r, f2l) + g 2 -
=
2g -
2,
which shows that the degree of the canonical bundle equals 2(g -
1).
104
2.5. Line bundles and embeddings in projective space
One useful application of linear systems is to construct embeddings of compact complex

manifolds in projective space, thereby realizing them as algebraic varieties. Let M be a
compact complex manifold and IVI a complete linear system (one proceeds in the same way
for general linear systems). If the base locus of JDJ is empty then, for every p, E M, the set
of sections vanishing at p forms a hyperplane Hp in P(HO(M, O(D))), so we get a map
t
[,D1 : M -+ P (Ho (M, 0 (D))) *
p F-+ H,.
If z[-D] is an embedding then [D] is called a very ample line bundle. If f- is a line bundle
and Ck is very ample for some k > 0 then f- is called ample. A necesSaXy condition for
%['D]
to be an embedding is expressed in the following theorem (Kodaira's Theorem).
Theorem 2.3 Let M be a compact complex manifold andC a positive line bundle. Then
there exists an integer ko such that for k > ko the map
M _+ pN
is a well-defined embedding of M in pN P(H0(M,0(&)))*. In the language of divisors,

if [D] is positive, then for some k E N, the functions with a pole of order at most k along D
will provide an embedding of M into projective space. I
It is plain from the theorem that the positivity of L is crucial. A line bundle L is called
positive if there exists a metric on L with curvature form 0 such that is a positive
(1, I)-form (see [GH] Ch. 1.2). Actually, the condition that the line bundle is positive turns
out to be a topological condition and Kodaira's Theorem can be reformulated as follows.
Theorem 2.4 A compact complex manifold M is an algebraic variety if and only if it has
a closed, positive (1, 1) -form w whose cohomology class [W] is rational, i.e., [W] E H2(M, Q).
Such a (1, 1) -form is called a Hodge form.
As anapplication of Kodaira's embedding Theorem, we show that every compact Rie-

mann surface isan algebraic curve, thereby justifying our terminology. Let g be any Hermitian
metric on the compact Riemann surface IP and w the associated (1, I)-forrn Q g. Multiplying
w by a constant if necessary, we may suppose that fr W 1, i.e., [W] is an integral cohomol-
=
ogy class. It follows that r can be embedded in projective space by using the meromorphic
functions on IP with poles at one point only. By Chow's Theorem, the embedded curve P
is given by the zero locus of a set of homogeneous polynomials. Since the variety of chords
of V has dimension three, we can project P to a hyperplane in this projective space and we
can repeat this process, until we finally obtain an embedding of r in p3. In general r can-
not be embedded in p2 ; however, since the set of tangents to the embedded curve has only
dimension two, we can project the curve birationally to a curve f whose only singularities,
are isolated (ordinary) double points, which means that the map IP -+ f is I : 1, except in a
finite number of points. Conversely, it can be shown that every irreducible algebraic curve P
has a normalization, i.e., there exists a compact Riemann surface r (which is unique up to
biholomorphism) and a map v : IP which is 1: 1 except at isolated points.
105
2.6. Hyperelliptic curves
Most compact Riemann surface of genus g ! 2 can be embedded in projective space in

a canonical way, namely by using the canonical bundle of the Riemann surface. Suppose r
has genus g ! 2 and let Jwi, .,wgl be a basis of W(r). In local coordinates, we can write
- -
wi =
fi(z)dz, and the canonical mapping ZK is given by
IP Pg_1
P (f, (P) (P)).
The complete linear system has no base points by the Riemann-Roch Theorem, hence SK is
an embedding when it is injective and immersive. The compact Rlemann surfaces for which
the above map is not an embedding axe called hyperelliptic (compact Riemann surfaces of
genus I being called elliptic). Hyperelliptic curves will be the most interesting for us because
of the following theorem.
Theorem 2.5 Let r be a compact Riemann surface of genus g > 2.
(1) hyperelliptic if and only if it has a non-constant meromorphic function f for

r is
(f) + P + Q : 0, for some points P, Q E IP;

which
(2) r is hyperelliptic if and only if IP is the normalization of an algebraic curve, given by
an affine equation of the form y2 f(X), where f is a monic polynomial of degree
=
2g + 1 or 2g + 2 without multiple roots;

(3) If g 2 then r is hyperelliptic.
=
Proof
The map W is I : 1 and immersive if and only if for any two points P, Q G 1P, there is
a holomorphic differential w for which w(P) :A w(Q) and there is an w' vanishing once at P
exactly. These conditions are equivalent to dim HO (r, 0 (K P Q)) < dim HO (r, 0 (K P)) - - -
for any P, Q E 1P. This reduces to dim HO (r, o (P + Q)) < 2 upon using dim HO (r, 0 (K -
P)) =
g -
1 and
dimHO(r, O(K -
P -
Q)) =
g -
3 + dimHO(r, O(P + Q)).
Since the constants obviously belong to L(P + Q) we conclude that a compact Riemann
surface is hyperelliptic if and only if it admits a meromorphic function with only two poles.
This proves (i).
As for (2), Consider the map from IF to P' constructed in (:L). It is a 2 : 1 map and
by an elementary count (the Riemann-Hurwitz formula) the number of branch points of this
map (i.e., the number of points where the map is I : 1) equals 2g -

2. Denoting these points
by xi, ...
I X2,+2
let
r, =
J(x,y) E C2 1 Y2 =
]J(X _
X,)J,
where the all i for which xi :A oo. The projection map x : r, -+ C is seen
sum runs over
to extend to holomorphic map r -+ P1, when TP is obtained from r, by adding one or two
a
points "at infinity". Finally r is shown to be a compact Riemann surface isomorphic to 1P.
To show (3), take IP of genus 2 and substitute V = K in the Pdemann-Roch Theorem.
Since deg K =
2g -
2 =
2, we find
dimHO(r, O(K)) = 1 + 2 -
2 +1 = 2.
We conclude the proof by applying part (i) of the theorem to K = P + Q.
106
In more general terms, a point P on a compact Riemann surface of genus g is called a

Weierstrass point if there exists a function which has a pole of order at most 9 at P and
which is holomorphic elsewhere. It follows that a hyperelliptic Riemann surface of genus g
has 2g + 2 Weierstrass points which are the points for which there is a function with a double
pole in one point only. Notice that these 2g + 2 points are intrinsically defined.
FinaJly, it is easy to check that on a compact hyperelliptic Riemann surface 1P, whose
corresponding algebraic curve is given by an equation y2 f (X), the g differentials
xi-ldx
=
Wi
are holomorphic. Since they are independent, they form a basis of the vector space of all
holomorphic differentials on the Riemann surface. A hyperelliptic Riemann surface has a
(holomorphic) involution (which is unique), the hyperelliptic involution, which is given by
(x, y) -+ (x, -y) when an equation for the corresponding algebraic curve is written as y2 =
f (x). Hyperelliptic Riemann surfaces and their Jacobians will be dealt with in detail in
Chapter VI. They will also appear frequently in Chapter VIL
107
3. Abelian varieties
3.1. Complex tori and Abelian varieties
If A is any lattice (i.e., a discrete subgroup of maximal rank) in a complex vector spare V
of dimension g, then the quotient
T9 =
V/A
is a compact complex manifold, calleda complex torus. It is a commutative complex Lie
group, but in general it is not an algebraic variety; a complex torus which is at the swne
i.e., which can be embedded in projective space, is called
time a projective algebraic variety,
an Abelian variety. The conditions on A for the derived torus T9 to be an algebraic variety,
which are computed from Theorem 2.4, are expressed by the famous Riemann conditions:
Theorem 3.1 T-9 =

V/A is an Abelian variety if and only if there exists an integral basis
I-X17 X2g}
...
7 of A and complex basis Jel,
a eg I of V such that the matrix of A with respect
to fel'..., eg} is given by
A =
(A,5 Z),
with A6 diag(61,..., 6,) a diagonal matrix whose diagonal elements are positive integers
=
satisfying dil&+, and Z a symmetric matrix whose imaginary part, Q (Z), is positive definite.
In terms of coordinates xi, X2g dual to this basis of A, the Hodge form w is given by
...
I
w 6idxi A dxi+,.
The Hodge form w and its cohomology class [w] carry extra information about particular
embeddings of V/A in projective space. Namely, up to an integral multiple, w is obtained
from an embedding
z: V/A -+ pN
as t* 91%(V/A) i where Q is the associated (1, I)-form associated to the standard KRhler structure
of pN. Different embeddings may lead to classes[w] which are different, even up to a multiple;
the cohomology class [w] of w on an Abelian variety is called a polarization and the pair
(V/A, [w]) is called a polarized Abelian variety. When the embedding is done by using the
sections of a (very ample) line bundle L, as explained in Paragraph 2.5, then the polarization
is precisely the Chern class of C and any polarization is the Chern class of an embedding line
bundle (Theorem 3.3 below) -
importance of considering polarized Abelian varieties, rather than Abelian varieties,

The
stems also from the fact that their moduli spaces are simpler. They break up in components,
given by the polarization type as follows. The integers 6i in w E 6i dxi A dxi+g are invariants =
of the cohomology class of w and are called the elementary divisors of the polarization and
(V/A, [w]) is said'O to have polarization type or type 6g); if the elementary divisors
of w are all one then w is said to define a principal polarization and (V/A, [w]) (or just V/A) is
called a principally polarized Abelian variety. Thus the moduli space of all polarized Abelian
10 abbreviation "Abelian of type instead

We will often use the common variety
of "polarized Abelian variety of type 6g)"
108
varieties breaks up in a natural way in components indexed by the polarization type and are
thus studied separately for each polarization type.
A holomorphic map between Abelian varieties is a group homomorphism followed by a
translation. If such a group homomorphism is surjective with finite kernel then it is called
an isogeny. Surjectivity is almost automatic in view of the following theorem, known as the
Poincar6 reducibility Theorem.
Theorem 3.2 If Tq is an Abelian

variety which contains a non-trivial Abelian subvariety
A then there exists 8ubvariety B of T9 such that A n B is a finite subgroup and
an Abelian
such that there exists a suijective homomorphism A ED B -+ T9 whose kernel equals A n B:
up to an isogeny such an Abelian variety is a product of Abelian varieties.
Abelian varieties which contain a non-triviaJ subtorus are called reducible Abelian va-
rieties and it follows e.g. from an easy dimension count that a general Abelian variety is
irreducible. Interesting isogenies, used below, are obtained by the following: any polarized
Abelian variety is isogenous to a principally polarized Abelian variety (but not in a unique
way). Another interesting isogeny is an isogeny between a polarized Abelian variety and its
dual which will be described in the next paragraph.
3.2. Line bundles on Abelian varieties
The Riemann conditions give us necessary and sufficient conditions for a complex torus
to be Abelian variety and Kodaira's Theorem says that the embedding can be done using
an
the sections of a positive line bundle. The positive line bundles on an Abelian variety can
be described very explicitly and a basis of the space of holomorphic sections can be written
down. To show this, let T9 V/A be a complex torus andC a line bundle on T-9. Then 7r*,C
=
is trivial (ir : V -4 T9) because V is contractible. Hence, there exists a global trivialization
0: lr*L -+ V X C.
Then (ir*L), =
(-7r*L),+,x and since 0., maps (ir*L)-. to C we get
C (7r*L)-, =
(ir*L),+,\ -+ C,
giving a linear automorphism C C, i.e., multiplication by a non-zero number e,&). The

functions 1eA E 0*(V)I.\EA are called the multipliers of L and satisfy
e,\, (z + A) e,\ (z) =

e,\ (z + A') e,\, (z) =
e,\+,\, (--) -
Conversely, multipliers which satisfy these relations define a unique line bundle with these
multipliers.
Line bundles can be constructed using multipliers je,\(z)j of a simple character. From
the exponential sequence
e
0 -4 Z -+ 0 !!T 0* -+ 0
we get
H1 (7'9, 0) -+ H1 ('T9, 0*) 54 H2 (T9' Z) -+ H2(Tg, 0)
109
where cl (L) is called the Chem class of the line bundle L. If W is a positive integral form of
type (1, 1) then
6,,dx,,, A dx,,+,,,,
with respect to the basis JxI.... I X2gJ dual to some basis JAI.... .X2.1 of A. Setting e,\
,\,, 16,, and letting zi, z, be linear coordinates dual to eI, , e, ,
we get the following . . .
theorem.
Theorem 3.3 The line bundle C -+ 7-9 with multipliers
e,\. =
1,
-21riz ,
= e
e,\,,,.
(for a = 1 .... g) has Chem class cl(,C) ==

[w].
Up to a translation in T9 every line bundle is uniquely determined by its Chern class

and this Chern class is c, (L) =
'j [ ' 7-r E)]
so it is a positive integral form of type (1, 1).
The fact that the line bundle is given by simple multipliers allows us to construct ex-
plicitly its holomorphic sections; they can be seen as functions on C9 which are periodic in g
directions and "quasi-periodic" in g other directions. The number of independent holomor-
phic sections is given by
dim HO (T9, 0 (,C)) 6,,, (3.1)
where (61 6,) axe the elementary divisors of the polarization cl(f-). For a line bundle
.....
defining a principal polarization, for example, there is only one section which, as a quasi-
periodic function on C9, is given by Riemann's theta function
ew'(',Zl)e21ri(l,z) (A =
(I Z)). (3.2)
IEZn
Its divisor of zeros, E), is determined uniquely by L, hence up to a translation by [c, (,C)] and
is called the Riemann theta divisor.
The group of all line bundles of degree 0 on a polarized Abelian variety 7-9 is a complex
torus, called its dual and denoted by t9. If T9 corresponds to a period matrix (A8, Z) then
a "dual" basis can be picked such that the matrix defining the lattice defining'fg is given by
K A6 1, bn A6 IZA6 1) - (3.3)
representation it is easy to check the Riemann conditions, which show that the dual
In this
an Abelian variety. For L a fixed positive line bundle on 7-9 one defines an isogeny
is indeed
between T9 and its dual by v -+,C-I (& Tv*,C. The degree of this isogeny is 11 6il.
If T9 is irreducible then the line bundle of any effective divisor is ample; moreover the
third power of any ample line bundle on an Abelian variety is very ample, hence gives an
embedding in projective space (these two properties axe particular for Abelian varieties, for
general algebraic varieties both are false; the last property is due to Lefschetz). For example,
if L defines a principal polarization on ail irreducible Abelian variety T9, then C3 induces
a polarization of type (3,3,...,3), and hence every irreducible principally polarized Abelian
variety can be embedded in PHO(T-9, 0(,C3)) which is by (3.1) isomorphic to p311-I.
110
3-3. Abelian surfaces
Since we are mainly interested in two dimensional a.c.i. systems, the Abelian varieties
which we will encounter often Abelian surfaces. In what follows we give some useful
are
techniques to study these surfaces.
As we have seen in the case of curves, varieties

are often studied by examining the (mero-
morphic) functions them,

specifically by examining the divisors of these functions;
on more
we will call an effective divisor C on an algebraic surface S an

(embedded) curve on S. The
curve C is said to be smooth if it is a submanifold of S
(taken with multiplicity 1) and
irreducible if it is not the union of two effective divisors.
A fundamental result here is that the canonical bundle of a curve on a surface and the
canonical bundle of the surface itself are intimately related, is expressed in the
as following
adjunction formula.
Theorem 3.4 If S is an algebraic surface and C a smooth curve on S, then the canonical
bundles KS and KC of S and C are related by
KC (KS 0 [C]) I C.
Since we have shown that deg(Kc) 2g -

2 if C has genus g, we can calculate the genus
of C by
I
g =
deg (Ks 0 [C]) I c + 1.
2
Now on S there is a natural non-degenerate intersection pairing
-
: H2(Sj Z) x H2(S, Z) -+ Z7
which counts the (signed) number of intersection points of arbitrary transversely meeting
2-cycles representing the homology classes. The pairing also gives a natural definition of the
intersection of divisors by taking the intersection of their fundamental classes in
H2(S, Z).
Under the natural isomorphism H2 (S, R) -+ H 2(S, R) -- HD2R which derives from
it, each
divisor D corresponds to a two-form qD E HD2R(S), its Poincar6 dual, and the intersection of
cycles can be shown to be Poincar6 dual to the wedge product of forms, i.e., if Q denotes the
top-form corresponding to the natural orientation, then " A ?7v, (V D')Q. This suggests = -
to define the intersection C -,C' for two line bundles L and V as the
cup product of their first
Chem classes, thought of as an element of Z,
C1 (L) U C2 (,Cl) =
(,C V) Q
Using the fact that the first Chem class of [D] is Poincar6 dual to D (which can be shown by
computing the curvature form of a metric connection on [D]) we see that
([E)] [D']) Q
* ==
cl ([D]) U cl ([D']) -
77v A 77D, =
(D D') Q
*
Hence, [D] [D']- = D -

D', and the two definitions correspond under the basic correspondence
between line bundles and divisors. It follows that we can also define the intersection of a line
bundle f- and a divisor D by putting
'C -
D = L -
[D] =
cl(,C)(D).
ill
Using the intersection pairing we can calculate the genus of the smooth curve C on S
more explicitly as
I Ks C+ C C - -
g =
deg(Ks 0 [C])Ic + I (Ks + C) - C+ I = +
2 2 2
For arbitrary curves on S we define the virtual genus -7r(C) by this formula. Hence, 7r(C) is
the genus of any smooth curve homologous to C. Let o : r -+ C be a normalization of C
and let JP1, , P, I
-
denote the
- - singularities of C, i.e. the points where C is not smooth. If
we by ki the multiplicity of C in Pi (i.e., the number of sheets in the projection, in a
denote
small coordinate disc in S around Pi, of C onto a generic disc), then
Ks-C+C-C ki(ki -
1)
9 Or) < +1- (3.4)
2 2
and the equality holds if and only if all Pi axe ordinary singularities (i.e., the ki tangents
at Pi are different).
We are now ready to give and prove the Riemann-Roch Theorem for line bundles on a
surface.
Theorem 3.5 LetC be a line bundle on a (smooth) surface S with canonical bundle KS.
Then
L - L -
L -
KS
X(L) =
X(OS) +
2
Proof
We give the proof for line bundles of the form L =
[D], where D is a smooth, irreducible
curve on S. Staxting from the short exact sequence
0 -+ OS -+ OS(,C) -+ OD(,C) -+ 01
and expressing the fact that the alternating sum of the dimensions of the vector spaces
appearing in the associated long exact sequence is zero, we obtain
X(,C) =
X(OS) + X(OV (,C)).
is the Euler characteristic of C restricted to the D, and it can be

Now X(OD (,C)) curve
calculated using the Rlemann-Roch Theorem for curves, giving
C -
L -,C -
KS
x(O-D(L)) =
X(Ov) + deg(,CI-D) = I -
g +,C -,C =
which leads to the proposed formula for X(,C).
112
It remains to compute the holomorphic Euler characteristic of the trivial bundle over S.
This result is given by Noether's formula:
Theorem 3.6 Let S be a complex surface and KS its canonical bundle. Then
1
X(OS) =
1-2 (KS Ks
-
+ X(S)),
where X(S) denotes the Euler-Poincar6 characteristic of S.
In the case of an Abelian surface 7-2 containing an axbitrary curve C, the adjunction
formula and the Riemann-Roch formula become extremely simple:
Theorem 3.7 Let T2 be an Abelian surface and C a curve on T. Then
C-C
7r(C) =
+ 1 =
XQCD + L
If [C] is positive and induces a polarization of type (91, 2) on S, then
C-C
-7r(C) = + 1 =
dimL(C) + I =
6162 + 1- (3.5)
Moreover, for a general Abelian surface the intersection of two line bundles [C] and [D] is
deduced from these formulas by replacing C and D by linear equivalent multiples of one divisor.
Proof
Let (ZI, Z2) be the coordinates on 7' coming from C'. Then the two-form
w =
dzj A dz2
has no zeros. Hence its canonical bundle vanishes, K 0. Because r has a natural cell-
=
decomposition with (4)

i
cells of dimension i, (i =
0,. .4), its Euler-Poincax6 characteristic
.
,
equals I 4 + 6 4
- -
+ I 0, leading to the first
=
string of formulas.
We now show that X ([D])
dim L (D) for positive line bundles, the equality dim L (D)
=
6162 being given by (3. 1). Clearly, it suffices to show that dim Hi (T2, 0 (D)) 0 for i > 1. For =
this purpose, we use the Kodaira vanishing Theorem, which states that HP (MI Qq (f-)) 0 for =
any positive line bundle L over a compact complex n-dimensional manifold M, if p + q > n.
For Abelian surfaces, Q2(,C) O(K)(L) O(C), because K 0 and the Kodaira vanishing
= = =
Theorem reduces for Abelian surfaces to
HP (7, 0 (L)) = 0 for p > 0.
The last claim follows from the fact that the N6rou-Sevieri group of a general Abelian surface
is isomorphic to Z.
113
4. Jacobi varieties
There are two very different ways to define the Jacobian of

a non-singular curve, the
equivalence of the two definitions

being given by two fundamental theorems, Abel's Theorem
and the Jacobi inversion Theorem. We prefer to give both definitions here because of the
importance of both of them in application to integrable Hamiltonian systems. We start by
giving the algebraic definition, then we give the analytic/transcendental definition and finally
prove their equivalence. It is also shown that the Jacobian of a curve of genus g is a principally
polarized Abelian variety of dimension g.
4.1. The algebraic Jacobian
We fix a curve (compact Pdemann surface) 17 of genus g. For the algebraic definition,
recall that constructed the group Pic(]P) of all line bundles on r, and showed that this
we
group is isomorphic to Div(r)/ -1, the group of all divisors modulo linear equivalence. Since
deg(f) 0 for any meromorphic function f , it follows that D -1 D' implies deg E) deg V,
= =
hence deg induces a homomorphism
Div(r) Z.
deg_ : -+
-1
We define the Jacobian of 1P, Jac(r), to be Kerdeg_. Said differently, Jac(]P) =

Divo(r)/
the group of divisors of divisors of degree zero modulo linear equivalence. The map deg_
allows us to define the degree of a line bundle C [D] by deg L = deg_ D.
=
Defining Pic'(r)
as the group of (isomorphism classes of) line bundles of degree i on r we have, with the
above definition of the Jacobian, that Jac(r) is canonically isomorphic to the group PicO(r).
Of course, PicO(r) is isomorphic to Pid'(r) for any other integer i but the isomorphism is
not canonical. For reasons that will become cleax later some authors prefer to define Jac(r)
as Picg-'(r).
4.2. The analytic/transcendental Jacobian
t
For the analytic definition, choose any basis 10 (wi, w,) of the space of holomorphic
=
differentials on IP and let A denote the discrete subgroup of C9 consisting of all vectors in C9
of the form
'Y
c , with -y running through HI(r, z). Since 1w,.... 1Wg' j'...' gj generate
HD'R (r) H1,0 E) H0,1 (the first the Rham group of IP and its splitting in the holomorphic and
=
anti-holomorphic paxt), A is actually a lattice (called the period lattice of Jac(I7)). This shows
that Cg/A is a complex torus, the analytic Jacoblan of r. More intrinsically, the analytic
Jacobian of r is defined by
HO(r, Q1)*
Jac(I') =
'
H, (r, Z)
where H, (r, Z) is viewed as a subgroup of HO (1P, Q')* via the natural injective homomor-
phism T which maps -y E HI (1P, Z) to the linear map
,P(-y) : Ho (r, Q1) c
W
fy
114
4. Jacobi varieties
Before
showing how the algebraic and the analytic Jacobian are related we want to show
that the analytic Jacobian is a (principally polarized) Abelian variety. Choosing a basis of
H, (r, Z), for example a symplectic basis JAI A,, Bl,...,B,}, i.e., a basis for which
......
Ai Aj -
Bi Bj
= 0 and Ai Bj
- =
Jjj, the lattice A is then conveniently represented as the
- =
column space (over Z) of the matrix
fAj WI ...
fAg W1 fB, WI fBg W,
fAj W9 fA,, W9 fB, W9 fBg W9

The first g x g block is called the matrix
of A-periods and the last g x g block the matrix of
B-periods. For
single 1-form its i-th period (I < i < 2g) is its integral over the cycle Ai
a
if i < g, otherwise it is its integral over the cycle

Bi-,. The following theorem states that
the (analytic) Jacobian of IP is a principally polarized Abelian variety.
Theorem 4.1 Let JAI,_, Ag BI,

..' Bg}
be a symplectic basis of HI (r, Z). For any
I -
basis of H'(r, QI) the of A-periods is non-singular and hence a basis of the latter
matrix
space can be chosen such that the matrix ofA-periods is the identity matrix. In this basis the
matrix of B-periods is symmetric and its imaginary part is positive definite. Thus
Jac(r) is
a principally polarized Abelian variety.
The main ingredient in the proof (given below) is the following proposition (known as
the reciprocity law for differentials of the first and third kind).
Proposition 4.2 Let IP be a compact Riemann surface of genus g, equipped with a sym-
plectic basis JAI,_, Ag, BI,-, B,} of HI(r, Z). Let w be a holomorphic 1-form and let 77
be a meromorphic 1-form whose poles are simple; we call these poles S,....
8n. Also so E r 7
denotes an arbirary fixed point. If IIi (resp. M) denotes the i-th period of W (resp.
ofy) then
9 n
81
1101g+k -
llg+k Wk 27ri Res,, q W. (4.1)

k=I 3=1
Proof
Representatives for the Ai and Bi may be chosen such that none of them pass throug
thepoles of 77 and by cutting r along the cycles the surface may be represented by a polygon IF
(with 4g sides) as in the following figure.
115
Figure 6
The boundary pieces which correspond to the cycles Ai and Bi will simply be called
-(i resp. -yi+, and accordingly the cycles are sometimes denoted by the uniform notation
Ji, i =
1,. 2g, (thus with this notation yj is just the boundary piece corresponding to the
. .
,
cycle Ji). Let
7r: C: 8 W,
80
then 7ris holomorphic and -7r?7 is a meromorphic 1-form with poles in the points si. Since
these poles are simple we have by the residue theorem
f o f;
?r,q=2-7riERes,,iri7=27ril:Res,,,q
i i
I. W,
which gives the right hand side of (4.1). Also
29 2g
-7r?7 r Fn =
E (7r(p+) _
7r(P 77(p).
r
Here P+ and P- are the two points corresponding to P which lie on -yj resp. yj-'; we used
that 77(P+) =
?I(P ). To finish, look at the abovefigure and compute
P+
ir(P+) -
7r(P-) =
fp, -
W
IBI W =
-Hj+g
if P G yj, I < j < g. Similarly for P E -tj+,, g + I < j < 2g one finds
P+
7r(p+) -
7r(P-) =
f 'P
W, =
fA W =
11j.
From this we find the left hand side of (4.1):
fat '7r?J E
j=I
-IIj+gq(P) +
+9
Ilin(p) E IIJI'9+k
k=1
IIg+kIIk- (4.2)
116
4. Jacobi varieties
Proof of Theorem 4.1

For a holomorphic 1-form. 77 the right hand side in (4.1) vanishes and the
gives simple
relation
9
IIkIII,+k -
IIg+kI-Ik --` 0- (4.3)

k=1
Denoting the complex conjugate of 77 by one has
A, wA = A; dir A =
A d(7r -
q) =
fe 7r -
'
which, upon using (4.2) to
A; wAq=EIlkfg+k -Hg+k k=1

k -
(4.4)
For a holomorphic Morm w =

f (z)dz
w A Co =
-2ilf (z) 12 dx A dy,
which leads for w =A 0 in combination with (4.4) to
0 > Q
ffr wA =
2!aEIIkfIg+k-
k=1
(4.5)
A first conclusion is that if all A-periods of a holomorphic 1-form w are zero then w = 0.
This implies that the matrix of A-periods of the basis JW1,
wg I is non-singular: otherwise. . .
,
some non-trivial combination of the elements of this basis would have all its A-periods equal
to zero.
By changing our basis of the space of holomorphic differential$ (if necessary) we may
therefore assume that the wi are normalized in the sense that
iAj Wi =
6ij.
The matrix of periods now takes the form (I Z), where Z is the matrix of B-periods. For
the B-periods of two I-forms wi and wj one finds from (4.3) that IIg'+i lIg+j, i.e., =
fBi
and Z is a symmetric matrix.
FinaJly!a,Z is positive definite. Let 0 =4 w ckwk with Ck E R. Then
Ilk CjWj =
Ck
k
j=1
k k
Hk+g 4k E i=1
cjwj =
E CjZjki
j=1
117
which leads by substitution in (4.5) to
1: Ck (O Zkj) Ej < 0-
k=1
This shows that the period matrix has the desired form and we conclude from Theorem 3.1
that Jac(r) is a principally polarized Abelian variety whose dimension is the genus of 1P.
Thus the Jacobian of a curve of genus g is a principally polarized Abelian variety whose
dimension g. The following converse also holds: every irreducible principally polarized
equals
Abelian variety of dimension 2 or 3 is the Jar-obian of a curve of genus 2 or 3. In higher
dimensions this is no longer true (as can be checked by an easy dimension count), and there
is the famous Schotky problem which asks for a characterization of those matrices Z for which
(Idg Z) is a (see [Mull], [Mum3] and [Shi]).

Jacobi variety
The Riemann theta function, defined in (3.2), can be used to construct (all) meromorphic
functions on Jac(17). Namely from the definition one checks the following quasi-periodicity
of this entire function on C9: if m E Z9 then
V(Z + M) ='O(Z),
(4.6)
V(z + Zm) e`ri(,rn,Zm)e -27ri(m,z)V(_,).
=
From this it is clear that for all 1 < i, j < g the meromorphic function
a2
u (Z) =
-log 19 (Z)
k9Zj'9Zj
isperiodic, u(z+m) u(z+Zm) u(z), hence it descends to a meromorphic function on the

= =
quotient with a double pole on the theta divisor. Another way to construct such functions,
n
which will also be used later, is by using theta functions with characteristics. For a, b E Q
(called characteristics) one defines
,i(1+a,Z(1+a)) e21ri(l+a,z+b)
,,
[ab ] (Z) =
1: (4.7)
IEZ9
Formulas (4.6) now become
[a]b (.v + M) = e2ri(a,m)V [a]b (4.8)

[a]b (z + Zm) = e-"(7n zm)e -27ri(m,z),-27ri(b,m),9 [a]b
From these formulas it is easy to see that if aj, bi and a'j, b'i are characteristics (for i =
1, . . .
, n)
such that E(aj a'j) (E Z9 and
-
S
E(bi bil) (E Z9 then
-
n V [ai ] (_,)
bi
H (4.9)
'=I V
[ail
b,
(Z)
is a meromorphic function on Jac(r).
118
4. Jacobi varieties
4.3. Abel's Theorem and Jacobi inversion
We now show that the algebraic and the analytic/transcendental Jacobians correspond
to the same object, namely we show that the Abel-Jacobi map
9 9 Pi
Ab : Divo (r) -+ Cg/A :
E Pi -
Q + EI c (mod A),
i=1 i=1 Qi
with c and A as defined above, is surJective with kernel consisting of the principal divisors.
We start with the latter which is the content of Abel's Theorem.
Theorem 4.3 For any divisor D of degree 0, Ab(V) = 0 if and only if D is the divisor of
a meromorphic function.
Proof
If V then d log f is a well-defined meromorphic 1-form with simple poles on the
support of V rn=,(pk
'k Qk) and residue I (resp. 1) at PI, (resp. at Qk). Denote by Wpk Q,,
= - -
the unique meromorphic 1-form with the same poles and residues as d log f but with its A-
periods zero. Then their difference is a holomorphic I-form
n 9
dlog f -
E WPkQk E CjWjl
k=1 j=I
C =
(cl, . . .
, cg) is just the vector of A-periods of d log f. Another application of Proposi-
tion 4.2, now with w =
wj and 17 =WPkQk leads to
fBj wpk Q, = 2-7ri

fPk
Qkh
Wi-
This allows us to compute explicitly the integrals appearing in the Abel map:
n P n
k=1
fQ kk Wj =
1
Iri
Ei
k=1
13
WpkQk
,1,J
i7ri BFj
dlogf- CI
fB WI
nj -
EmIZjI,
1=1
in which the integers mj and nj are up to a factor 27ri the A- resp. B-periods of d log f .
This shows one direction of Abel's Theorem. For the other direction, if Ab(D) = 0 one
is tempted to define a function on r by
f (P) =
exp
(27ri 77) PO
119
where 71 is a 1-form with residue in the points Pk and ;'

27ri 27ri
in Qk. Indeed, if f is well- j:7r-'j
defined then (f) D and we are done. The problem is thus to find an
71 with all its periods
=
integral. Any such 77 can however be written as

9
,q =
770 + E c1W1
1=1
where 970 has its A-periods equal to zero. We must take ck integral in order to have integral
A-periods for 77. As for the B-periods,
n P- 9
fB I
17
h=1
fQ k
Wj + cl Z1j'
where computed the B-periods of qo from Proposition 4.2 (with w wj and 71

we
71o). Since
= =
Ab(D) = 0 the first term has the form nj + (Zm)j where ni, mi E Z (and m (m, .... M9))- =
Thus it suffices to choose cj =

-mj to obtain a Morm 77 which has all its periods integral. n
To show surjectivity, the Jacobi inversion Theorem, it is better to define other maps
which axe quite similar to the Abel-Jacobi map. To do this we fix any PO E 17 and we define
for any d E N
d
P!
Abd : SYM
dr-+C9/A:(Pj,...'Pd) -+Ef. j=1
PT,
c (modA). (4.10)
Theorem 4.4 The map Abg is surjective.
Proof
symg r and Cg/A are both compact connected complex manifolds and Abg is a holo-
morphic map. Hence the image of Ab. is a compact subvariety of Cg/A. Its differential in a
general point (PI, P,) is given by
. . .
,
Wi(PI) ...
wg(pl)
W, (Pg) ...
Wg Wg)
where we have written wj locally as

wj fj(z)dz and wj(P) is a (bad) notation for fj(P).
=
By Riemann-Roch there exists no holomorphic differential with its zeros on a general set
fpj'...' Pgj of points, hence this matrix is invertible. This shows that the image of Abg has
dimension g, hence Ab. is surjective.
The link between the algebraic and the analytic/transcendental Jacobian is pursued
further in the following theorem, attributed to Riemann.
Theorem 4.5 There is a constant,& E C9 (called Riemann's constant) such that
g-1
V(Z) == 0 4=* 3PI, -
..' P, E r : Z = Ab E(Pi -
P) A (mod A). (4.11)
120
4. Jacobi varieties
The important condition in the right-hand side is that the sum runs over g -
I points
I
only. Because of this theorem mwiy people prefer to define the Jacobian by Pic-q- (r) (rather
than Pico (17)) since with the former definition the theta divisor is canonical determined (while
otherwise it is defined only up to a translation which depends on the choice of a point on r).
4.4. Jacobi and Kummer surfaces
Everything discussed above is easily specialized to Jacobi surfaces, i.e., Jacobians of

genus two curves. They are especially important when working with Abelian surfaces since,
as we explained, every Abelian surface is isogenous to a principally polarized Abelian surface
and these are (under the asswnption of irreducibility) Jacobians of genus two curves (recall
also that every Pdemann surface of genus two is hyperelliptic, hence can it can easily be
described explicitly). The theta divisor of a Jacobi surface is by Riemann's Theorem nothing
but a copy of the curve, embedded in its Jacobian; conversely if in some Abelian surface a
(non-singular) curve of genus two is found, then it defines by (3.7) a principal polarization
on the surface which necessitates it to be the Jacobian of this curve. Since the third power
of any ample line bundle on an Abelian variety is very ample, we may take the third power
of the line bundle corresponding to the theta divisor to construct an embedding of a Jacobi
surface in P8.
Associated to an Abelian variety T9 is a singulax variety, its Kummer variety, which is de-
fined the quotient surface K
as T9 / (- 1), where I is the involution given by (ZI,
= -
, zg)
1-+ . . .
(-zl,..., -z,) in linear coordinates coming from the universal covering space C9 of T9.
Clearly the Kummer variety has 229 singular points which correspond to the two-torsion
points (also called half-periods) of T9. We will consider in this text only the case g 2, =
the case of Kummer surfaces. Let A be an Abelian surface of type (611 62) Its Kummer -
surface K Al(-I) has sixteen singular points ej,...' e16. The desingularization of K can
=
be described as follows. Let p : A -* A be the blow-up of A at all its half periods and denote
the corresponding exceptional divisors by Ej. (-I)A extends to an involution (-l)A On A
and the quotient ff Al(-l)A is a K-3 surface (see [Beall, Proposition VIII.11). ff is the
=
desingularisation (minimal resolution) of K and we have the following commutative diagram.
P
A i A
IrI I
k - K
Associated to A there are surfaces which are desingulariza-

also several intermediate Kummer
tions of K at some but not all
singular points. A similar construction can be performed when
the Abelian surface admits an automorphism, different from the (-1)-involution, leading to
a generalized Kummer surface. See Paragraph VII.3 for an example.
2
In the case of the Kummer surface of surface, taking C
a Jacobi [6], all sections Of r- =
axe even and the singular surface can be embedded in P1 by using these sections. Being two-
dimensional the image is given by a single equation. To compute the degree of this equation,
which is the degree of the hypersurface, we use the fact that this degree is given by fK Q,
where Q is associated (1, I)-form of the standard Mililer structure on P1. Clearly this is*
twice the volume of K, which itself is half the volume of the Jacobi surface (embedded with
121
the polarization of type (2, 2)). For w 2dxj A dX2 + 2dx3 A dX4 we get f W2
=
8, hence the =
Jacobi surface has degree 8, its volume is 4, the volume of K is 2 and the degree of K is 4.
Explicit equations for this quartic polynomial, in terms of an equation for the underlying
algebraic curve, will be given in Section VIIA.
Another classical result about the Kummer surface of a Jacobi surface is that it has
a 166 configuration. Namely for IP a curve of genus two, the (-I)-involution t on Jac(]P) has,
apart from 16 fixed points aJso 16 invariant theta curves (i.e., translates of the theta divisor);
each of these 16 curves contains 6 of the fixed points and through any of the 16 fixed
points
pass 6 of the invariant theta curves. The configuration can be described completely as follows
(see [Hud]). Let W1,...' W6 be the Weierstrass points on r, then the points
Wij =
fW", 0' (mod A)
are half-periods ofJac(]P) since 2Wj -1 2Wj. There are sixteen half-periods in total since the
equalities Wij =
Wji and Wii Wjj hold for all i, j
=
1, 6; apart from these they are
=
. . .
,
all different. The sixteen invariant theta curves rij in Jac(IF) are the translates
wij + rkk of
the single curve ]PIi = *
IP66, which can e.g. be taken as jAb(P WI) I P E rj. aoin
* * = -
this it is clear that IPII, hence aJI ]Pij, pass through six points Wk, and also that each point
belongs to six lines ]Pij. Since the whole configuration is invariant under z it goes down to the
Kummer surface in P3 and gives there a 166 configuration of points and planes, classicaJly
called Kummer's configuration. The sixteen points are nodes (singular points) and the sixteen
planes the lines belong to are tropes (singular planes) of the Kummer surface.
The 166 configuration is best visuahzed by the incidence diagram, which consists of a
pair of square diagrams, such as the following.
W11 W12 W23 W13 ri, r.2 P23 IP13

W45 W36 W16 W26 r4, 1 3, P,6 r2,
W46 W35 W15 W25 IP46 IP3. r,, r25
W56 W34 W14 W24 r.6 r34 r14 r24
Namely, the points incident with a line at position (m, n) in the second square diagram are
those six points in the m-th row and n-th column, but not in both, of the first square diagram.
Dua,lly, the same applies for the lines incident with a point. The 242 incidence diagrams
obtained by permuting the rows or columns of both square diagrams in an incidence diagram
(in the same way) are defined to be the same as the original incidence diagram (we will see in
Section VIIA that there are 20 incidence diagrams which are different in this sense). These
incidence diagrams will be used in Section VIIA to describe Abelian surfaces of type (1,4).
122
5. Abelian surfaces of type (IL 14)
5. Abelian surfaces of type (1,4)
Closely related to Jacobi surfaces are Abelian surfaces of type (1, 4). Their geometry
is discussed here in detail because they will appear in the two examples discussed in Chap-
ter VII. The results in this section axe taken from [BLS]. As in that paper we will without
further mention always restrict ourselves to those Abelian surfaces of type (1, 4) which are not
isomorphic to a product of elliptic curves as polarized Abelian surfaces. Let PC be a line bun-
dle of type (1, 4) on an Abelian surface 7-2. It follows from (3.7) that dim HO (7-2, 0 (,C)) 4 =
and C induces a rational map Oc : 7-2 _+ p3.
5.1. The generic case
In the generic case, the image of this map 0 Oc(T2) = C p3 is an octic and O'C is
birational on its image. Let K(L) be the kernel of the isogeny
1'C :r -+ r
a -+ taiC 0,C_1
between r and its dual 'P (defined as the set of all line bundles on T2 of degree 0; t,, is the
translation by a Ei T), then K(L) is a group of translations, isomorphic to Z/4Z ED Z/4Z.
Picking any such isomorphism, let a and -r be generators of the subgroups corresponding
to this decomposition. Then homogeneous coordinates (VO : yj : Y2 : Y3) for P3 can be
picked, such that o-, -r and the (-I)-involution % on 7' (defined as %(Z1, Z2) (-Z11 -Z2) for =
(zl, z2) E C2/A) act as follows (see [LB]):
0'(YO Y1 Y2 Y3) =
(Y2 Y3 YO
*
-YI),
T(YO Y1 Y2 Y3) =
(Y1 YO iY3 : iY2)i (5-1)
t(YO Y1 Y2 Y3) =
(YO Y1 Y2: -Y3)i
(strictly speaking it may be necessary to replace r by 3-r; it is easily checked that these
coordinates exist only for (a, -r) and (3a, 3-r) or for (a, 3,r) and (3a, -r)). [BLS] show that the
octic 0 is given in these coordinates by
2 2 2 2 2
,\2OYOYIY2Y3 + '\2I ( YO4)Y41
Y4Y34)
2 2 + Y14Y4)
+ A2(YO4Y4
2 + 3 +A3 (Y04Y343 + Y4IY24) +
0 1 + Y2Y2)(Y2Y2
2A,1\2 (Y2Y2 3 1 32 Y2Y2)
0 2 + 2A,1\3 (Y2Y2
_
0 3 Y2Y2)
1 2 (Y2Y2
0 1 Y2Y2)
2 3 +
_ _
(5.2)
2 2 2
21\2X3(YlY2+YOY32)( Y12Y23 + Y2Y22)
0 0, =
for some (,\o : \, : 1\2 : X3) E p3 \ S where S is some divisor of p3 which we will determine
later (Paragraph VII.4.4). Notice that for any ej 1, the coordinates (IEOYO : 'ElYl : IE2Y2 :
=
1EO611E2Y3) will also satisfy (5.1) and these are the only coordinates with this property. It is also
seen that, if (a,,r) is replaced by (3a, 3,r), then the coordinates (yo : yj : Y2 : Y3) are replaced
by (YO : Y1 : Y2 : -Y3)- Since the equation of 0 depends only on yj2 these choices do not affect
the equation (5.2), so there is associated to a decomposition K(,C) K, (D K2 (where K, =
and K2 are cyclic of order 4) an equation for 0. [BLS] also show that the polarized Abelian
surface as well as the decomposition of IC(L) can be recovered from (5.2) and that every octic
123
of the type (5.2) (with (AO : 1\1 : /\2 : )k3) S) is the image Oc (T2) of some (1, 4)-polarized
Abelian surface (7-2, L) -
If we denote by 140(1,4) the moduli space of (isomorphism classes of) (1, 4)-polarized
Abelian surfaces for which OC is birational, equipped with a decomposition of K(L) as above,
then it follows that
'40 _
p3\S
(1,4)
-",::
ToZ -AO (5.3)
Moreover, if we denote by K the subgroup of K(,C) of two-torsion elements,
K =
10, 2o-, 2T, 2T + 2o-J,
then 71/K is a principal polarized Abelian surface, which is the Jacobian of a curve of genus
two; call T2/K the canonical Jacobian associated to 7-2. Recall that for a two-dimensional
we
Jacobiau J its Kummer surface is the image of 0[28] C p3' where 0 is the theta divisor of J.
Then it is seen from (5.1) that an equation for the Kummer surface of 7/K is given by
the quartic Q in p3' obtained by replacing yi2 by zi in the equation (5.2) for 0 and there
is a natural projection fi : 0 -+ Q. In fact, choosing the origin of 7' such that L becomes
symmetric, L is the pull-back of a line bundle Ar on T11K of type (1,I) via the canonical
projection
p : T2 -* 7-2 1K,
and OAr2 induces the Kummer mapping; [BLS] prove that the following diagram commutes
O.C
-2
7 0
P1 If (5.4)
Q
O,vn
5.2. The non-generic case
If Oc is not birational, then it is 2 : I and Oc (T2) is a quaxtic in p3' given by one of the
equations
A Y6 Y21 +
Y2Y2)
2 3 + A2 (Y2Y2
1 3 Y2Y2)
0 2
_
=
0,
A 1 (Y2Y2
2 3 Y2Y2
0 1)_
+ A3 (Y2Y2
1 2 Y2Y32)
0
_
=
07
2 2
/N2(Y1Y3 + y2y2)
0 2 + \3(y2y22
1 + Y02y32) =
07
depending on the choice of the decomposition; in this
case the Abelian surface as well as
the decomposition of K(L) can only partly be recovered from these equations and r1K is
124
5. Abelian surfaces of type (1,4)
a product of elliptic curves (in particular T' is isogeneous to a product of elliptic curves).
Squaring each of these equations we find equation (5.2) respectively with
A20 2 ('X22 + A2)

3
1\2,X3 0, A22 - 'X23'A 0,
0
A20 -2 (,\21 + A2)

3
'X1A3 : 07 \21 _ 'X237 0,
=
(5.5)
X2 = 0
2
1\ 0 2 (,\ 21 \2)
f
=
2
\IX2 7 07 A21 + '\22A 0.
X3 = 0
Summarizing, case (the generic case), OC(V) is an octic, 7-2/K is a Jacobian

in the first
and 7-2 as well asdecomposition of K(L) can be reconstructed from the octic; in the other
the
case OC (7-2) is a quartic, 71 IK is a product of elliptic curves
and T2 cannot be reconstructed
from the quaxtic. The rational map OC provides us with a natural surjective map
,00 AO(I 4) -+
((p3 \ S) U (three rational curves in S missing eight points)) /(Ao 0),
denotes the moduli space of (isomorphism classes of) (1, 4)-polarized Abelian
where AO(,,4)
surfaces with decomposition of K(L) (as above). The map 00 extends the bi-
together a
'4 0(1,4) of AO and maps the (two-dimensional)

jection (5.3) defined on the dense subset
(1,4)
complement of .40(1,4) to the three rational curves, which axe thought of as lying inside the
boundaxy of 00 (A(,,4)),
0
i.e., in S; the generic point of S however does not correspond
of Abelian
to Abelian surfaces, but to surfaces which can be interpreted as degenerations
surfaces (see P3LS]).
125
Chapter V
Algebraic completely integrable

Hamiltonian systems
1. Introduction
In many integrable Hamiltonian systems of interest the general level set of the momentum
map axe isomorphic to an affine paxt of an Abelian vaxiety and the flow of the integrable
vector fields is linearized by this isomorphism. These two properties lead to the definition
of an algebraic completely integrable Hamiltonian system (a.c.i. system). We will discuss
three quite different definitions of an a.c.i. system, which have been proposed by different
authors, and we extract from it a definition which is consistent with our approach to integrable
Hamiltonian systems. All constructions of integrable Hamiltonian systems easily speciahze
to the case of a.c.i. systems, except in the case of the quotient, which requires a real proof.
The definitions and these properties will be considered in Section 2.
Painlev6 analysis isimportant tool for studying an (irreducible) a.c.i. system, since
an
it allows us to determine the nature of the general level sets of its momentum map and
to construct an explicit embedding of the completed general level set (which is isomorphic
to an Abelian variety) in projective space. Although the nature of these Abelian varieties
can often also be deduced from a Lax representation of the a.c.i. system (see e.g. [Aud3%
Painlev6 analysis is at present the only available method to construct an embedding of these
Abelian varieties in projective space. Since several such embeddings will be constructed and
used, we will explain what Painlev6 analysis is about in Section 3. We wish to point out that
Painlev6 analysis can, by a theorem of Adler and van Moerbeke (see [AM7]), in principle
be used to prove that a certain integrable Hamiltonian system is a.c.i., but applying this
theorem requires even for the simplest systems a considerable ainount of work. In Section 4
we recall from [Van2] our algorithm which allows one to lineaxize explicitly two-dimensional
a.c.i. systems starting from the differential equations for one of the integrable vector fields.
It is used in this text to construct morphisms of two-dimensional a.c.i. systems.
127

Chapter V. A.c.l. systems
In Section 5 we will explain briefly how Lax equations appeax and what is their relevance
for a.c.i. systems, in paxticular when dealing with Lax equations with a spectral parameter.
Lax equations have by now been obtained for all integrable Hamiltonian systems which were
classically known, however some of these have been obtained after a lot of effort, reflecting the
fact that it is not clear how to write down Lax equations for a given integrable Hamiltonian
system, even if the underlying geometry has been completely revealed. For the general
integrable Hamiltonian systems introduced by us in Chapter III for example it is not clear
how to write Lax equations for the integrable vector fields (apaxt from the very special case
treated in Paragraph 111.2.4). Lax equations constitute a complete chapter in the theory
of integrable systems, however we will not treat them as being basic in their study since
very often they only appear at the end. Of course, once obtained, Lax equations beautifully
exhibit many aspects of the integrable Hamiltonian system in a unifying way; moreover, a
careful analysis of the underlying algebra leads in many cases to generalizations and to a
quantification of the integrable system.
128
2. Ax.i. systems
2. Ax.i. systems
We startby discussing three definitions of algebraic complete integrability. Often in the

literature this term is
being used meaning that the general fiber of the integrable Hamiltonian
system under consideration is an Abelian variety, an affine part of an Abelian variety or just
a dense subset of an Abelian variety, the flow of the integrable vector fields being in any case
linear on it. Indeed that is the main idea and will be present in the different definitions of
a.c.i. systems given below.
Let us start with the Adler-van Moerbeke definition (see [AM7]). They consider a
polynomial integrable Hamiltonian system (i.e., the Poisson bracket is polynomial as well
as the functions in involution) on IV' (fixing a basis of the integrable algebra, but that is
not relevant here) which they complexify in the natural way. They suppose the following
conditions about the general level sets of its momentum map and the flows on them to be
verified:
1) The general fiber of the momentum map is (isomorphic to) an affine part of an
Abelian variety;
2) The flow of the integrable vector fields on the general fiber of the momentum map
is linear;
3) on C' restrict to the general level sets as regular functions
The coordinate functions
which extend to meromorphic functions on the corresponding Abelian varieties;
4) The divisor to be adjoined to the general level set is minimal in the sense that for
each irreducible component of this divisor there is at least one of these meromorphic
functions which has a pole on it.
Next we give Mumford's definition (see [Mum5J pp. 3.51 3.54). He starts from a real -
symplectic manifold (M, w) of dimension 2n and a smooth function H on it. His requirement
for algebraic complete integrability is the existence of a smooth real algebraic variety (MC, T)
such that:
1) M is a component of the real part of (MC, -r);

2) MC is endowed with a (co-)symplectic structure (D which restricts to w along M;
3) There exists on MC a proper map which is submersive onto a Zariski open sub-
set of Cn and whose components are in involution, H being dependent on these
component functions (restricted to M).
The conditions are sufficient to imply that the fibers are Abelian varieties or extensions of
these by C*1 and that the flow of the integrable vector fields is linear.
Third, let us recall Kn6rrer's definition (see [Kn6]). He considers an (irreducible) complex
manifold M of dimension 2n with a holomorphic symplectic structure w, a holomorphic
function H on M and n holomorphic functions in involution (and in involution with H)
which define a submersive map onto an open dense subset B of Cn,
F: F-I(B) c M -+ B c Cn.
For algebraic complete integrability he demands the existence of
1) a bundle 7r : A -+ B of Abelian varieties;

2) a divisor E) C
M;
3) an isomorphism 0 : F-1 (B) -+ A \ D;
4) a vector field Y on A \D which restricts to a linear vector field on the fibers of ir;
129
Chapter V. A.c.i. systems
so that the diagram

F -1(B) A\D
P
I Z11
B
is commutative and such that the vector field XH is 95-related to Y.

Let us comment on these definitions and deduce from it the definition which is appropri-
ate in the case of integrable HandItonian systems on affine Poisson spaces. We don't discuss
here the fact that all three choosea specific basis of the integrable algebra since we discussed
this aspect already in Chapter II (and it is not really relevant here). The last two definitions
are situated on symplectic manifolds, while the first is on Poisson manifolds, a difference
which is easy to overcome. The first two definitions start from a real system and complexify
it, the last one is completely situated in the complex. Also here, there is nothing to be worried
about; from the point of view of physics and when one wants to study the real topology of
integrable Hamiltonian systems, one may want to impose a reality condition, but it is clear
that the condition is not essential in the definition. So we think it should not be encoded in
the definition but put as a restriction when the applications which one has in mind ask so.
We now come to the more essential points in the definitions which all relate to the nature
of the general level sets of the momentum map.
The Adler-van Moerbeke definition can in this respect be reformulated saying that
the fiber of the momentum map (on
general Cn), with its algebraic structure which comes
from the ambient space C', is isomorphic to an affine part of an Abelian variety; the added
condition that the missing divisor is minimal relates to the possibility to perform the com-
pactification of this general fiber into an Abelian variety by using the Laurent solutions to
the differential equations describing one of the vector fields (see Section 3 below). We think
that one should not insist on putting this condition in the general definition.
D. Mumford's definition does not demand about the nature of the

;ber of the momentum map but his assumptions anything
imply what it is- it is an
general
Abelian variety
or an extensions of an Abelian variety by C*11. The key is that he asks properness of the
morphism (which plays the role of the momentum map). As such, this is very strict: in
the examples we know of one rarely finds complete Abelian varieties (but C*In are found in
several examples). After completion however, as is discussed below, a proper map and hence
an a.c.i. system in this sense" is found. An advantage of his definition is that he allows fibers
which are more general than Abelian varieties.

H. Kn6rrer's definition asks that the level manifolds be
isomorphic to affine parts of
lbelian varieties, but family. Whereas the Adler-van Moerbeke definition asks that
even as a
every generic level manifold can be compactified into an Abelian variety, his definition asks
for a so-called partial compactification of the momentum map (at least over a Zariski open
subset); this would then lead to a compactification of each general fiber by restriction. With
these remarks in mind we will now define two notions of a.c.i. systems, one notion being
stronger (but harder to verify) than the other.
Strictly speaking the symplectic or Poisson structure may not extend to the completion.
130
2. A.c.i. systems
Definition 2.1 Let (M, A) be an integrable Hamiltonian system on an affine Poisson

variety and denote its momentum map by -7rA : M -+ Spec(A). Then (M, 1-, -1, A) is called
an algebraic completely integrable Hamiltonian system or an a.c.i. system if there exists a
Zariski open subset B C Spec(A) and a bundle of Abelian groups 7r A -+ B such that for
each b E B there exists a divisor Vb C 7r- I (b) and an isomorphism Ob 7r. ' (b) -+ 7r-'(b) \ Z)b i
such that the restriction of each vector field in Ham(A) to ir. '(b) is Ob-related to a linear"
vector field on 7r-'(b).
If there exists instead a bundle of Abelian groups 7r : A -+ B, where B C Spec A is a
Zaxiski open subset, a divisor D on A and an isomorphism 0 such that
0
,ir. ' (B) - -------
P-
A\V
VA
I/
B
ir
is a commutative diagram and such that each vector field in Ham(A) is 0-related to a vector
field on\ D which restricts to a linear vector
A field on each level set of ir, then we say that
(M, 1-, .1, A) is a completable a.c.i. system.
Here are some remarks about this definition. Clearly a completable a.c.i. system is an
a.c.i. system: defines Ob and 1)J-7r-1(b)- Whether every a.c.i. system is

one =
01-7rAl(b)
completable is a question which has been studied in more general terms in algebraic geometry.
13
In simple terms the way they put this question is demanding for a partial compactification of
a morphism whose fibers over closed points are (isomorphic to) affine parts of algebraic groups;
this partial compactification should be such that the compactified fibers are isomorphic to
the corresponding algebraic groups. Their solution is to construct the points to be added as
a SUM14 of two points in the affine part; more precisely they construct the fibred product
of -7r. ' (B) with itself and define an equivalence relation by which two pairs of points are
equivalent if they belong to the same fiber and they have the same sum (the sum to be taken
15
in the completed fiber). The question is now if this leads to an algebraic quotient. In general
the answer is no, it works only up to a base extension. The problem of completing an a.c.i.
system comes from the monodromy of the base space; in a real setting the monodromy is an
obstruction for the global existence of action-action variables, as was shown by Duistermaat
(see [Dui]). As Michble Audin pointed out to me, these ideas are very close.
It turns out that many a.c.i. systems are actually completable (more precisely: a coun-
terexample is not known). One way to verify this is to compute an embedding of the general
fiber of the momentum map in some projective space PI; this can be done explicitly by using
"
Bylinear vector field on an Abelian group G we mean a G-invariant vector field.
a
13
partial compactification only the fibers are compactified, not the base space.
In a
14
In our case the Abelian varieties will appear without origin, i.e., as homogeneous spaces
and one cannot assume the existence of a section over B which picks an origin on each of
these Abelian varieties. Then the construction is modified by picking the fibred cube instead
of the fibred square, using a similar equivalence relation.
It is given that we can complete each level separately.
131
the Laurent solutions to the differential equations describing one of the integrable vector fields
will explain in Section 3. In all examples we have seen the embedding of Fb
as we
=,7r. '(b)
(b B,
E Zariski open subset of Spec A) in projective space is given by functions which are
a
regular on M and independent of the chosen fiber, i.e., there exist fi,..., fN E
O(M) such
that the map 0: M _+ pN given by
P -+ (1: h(p): fN(P))
has thefollowing property: for any b G B the restriction of 0 to -Fb is an embedding and the
image is an Abelian group (isomorphic to ir-'(b)). In
closure of its this case we obviously
have an embedding -7rAl (B) _+ pN x B given by

P -+ ((1 -, h (p) : * * *
: fN (P)) , IrA (P))
and the closure of (7r. ' (13)) is the desired paxtial compactification. The same construction
applies when the functions fl,..., fN depend regularly on B, i.e., if fi G O(B x M) for i
N.
aom now on we will almost exclusively (i.e., except in Section VII.4.5) deal with irre-
ducible a.c.i. systems, defined as follows.
Definition 2.2 An a.c.i. system is called irreducible if the general fiber of the momentum
map is an affine part of an irreducible Abelian variety.
Example 2.3 If one of the vector fields of

an a.c.i. system is super-integrable then the a.c.i.
system is not irreducible.

Namely suppose that its general level sets are Abelian varieties,
then the flow of the super-integrable vector field, being lineax on the Abelian variety on the
one hand and being contained in a subvariety of lower dimension on the other hand, must
evolve on an Abelian subvariety, hence the general fiber of the momentum map would not
be an irreducible Abelian variety. Whether an a.c.i. system which is not irreducible admits
a super-integrable vector field is not known.
Example 2.4 product of two a.c.i. systems is an a.c.i. system which is never irreducible.
The
In this case every integrable vector field of the original systems leads to a super-integrable
vector field on the product.
On each fiber ir. '(b) of an irreducible a.c.i. system the divisor A induces a polariza-
tiOn C-1 ([Dbl) since any effective divisor on an irreducible Abelian variety is ample. Thus one
may think of the general fiber of the momentum map of an irreducible a.c.i. system as being
a polarized Abelian variety. If the a.c.i. system is moreover completable then the polarization
type of this general level set is constant since it is discrete; probably the assumption that the
system is completable is superfluous here but we don't have a proof of this. In any case, if
the polarization type of the general level sets of an irreducible a.c.i. system is constant then
we call it the (polarization) type of the a-c-i. system.
If a divisor is removed from a (completable) a.c.i. system as in Proposition 11.2.35 then

the resulting integrable Hamiltonian system is also a (completable) a.c.i. system. There are
two very distinct possibilities.
-
If the function f whose zero divisor is removed belongs to A then some level sets
are left out, the others remain intact. Of course the ones which are left out are not
general, so we still have an a.c.i. system;
132
2. A.c.i. systems
-
If not belong to A its zero divisor cuts off from every level set a divisor (a
f does
dramatic change), of course the general level set is still an affine part of an Abelian
variety (or Abelian group in general).
For a.c.i. system which depends on parameters as in Proposition 11.3.24 we easily see
an
that the big integrable Hamiltonian system which is given by the latter proposition is also
a.c.i.; clearly it contains as its level sets aJ1 level sets of all the a.c.i. systems obtained by
freezing the parameters.
Some care has to be taken when restricting a (completable) a.c.i. system to a level set of
the Casimirs. By Proposition 11.3.19 one gets on a general level
integrable Hamiltonian
set an
system; in order for it to be a (completable) a.c.i. system one has to verify in addition that
the general level set of this restriction is contained in the collection of general fibers (which
are known to be affine parts of Abelian groups) of the original system. For special level sets
of the Casimirs we may not even get an integrable Hamiltonian system.
Less obvious is the useful property that the quotient construction also leads to a.c.i.
systems as given in the following proposition 16.
Proposition 2.5 Let (M, 1-, -1) be a Poisson variety with a Poisson action by a finite
group G. If (M, 1-, -1, A) is a (completable) a.c.i. system with an affine part of an Abelian
variety as its general level set and for each g E G one has g A C A then (MIG, J.'.10' AG) -
is a (completable) a.c.i. system (with an affine part of an Abelian variety as its general level
set) and the quotient map 7r is a morphism (I., jo is the quotient bracket on MIG).
Proof
Proposition 11.3.25 it suffices to identify the general fiber of the momentum
In view of
map being isomorphic to an affine part of an Abelian variety. Clearly the action descends
as
to an action on Spec A, (denoted in the same way) namely for each g Ei G one has the
following commutative diagram.
M M
WA
I I WA
Spec A -
91
Spec A
Since (M, A) is a-c-i. there is a Zaxiski open subset B C Spec A and a bundle -7r : T -+ B
whose fibers are Abelian varieties which compactify the fibers of 7rA over B. Since G is
finite, B may be assumed stable for the action of G; also by passing to a smaller group if
necessary we may assume that the action is effective.
Let b E B be a general point. Since each g- maps level sets (of 7rA) to level sets and
since all g- invertible, each g- restricts to an isomorphism 7r '(c) -+ ir' '(g b) which after
are -
composing with Ob and 09-b leads to an isomorphism ir-'(b) \Vb -+ *7r-' (9 b) \Vg-b of the affine ,
parts of the corresponding Abelian vaxieties; since g- is a Poisson map and Ob and 09-b linearize
16
The proposition is generated by a quasi-automorphism. (of filinite
not valid if the action is
order). If one considers for

example the Z2 action generated by a (-1)-involution then the
general level set of the quotient is an affine part of the Kummer variety of the Abelian variety;
obviously also the vector fields do not descend to the quotient.
133
the Hamiltonian vector fields, each g. extends to an isomorphism g. : ir-1(b) -+ -7r-'(g b) -
hence extends to T. The upshot is that the action of G on M induces an action on T (such
that 0 is equivariant). As we have seen inParagraph IV.3.1 this implies that each map
g.
-
T -+ T restricts to each fiber as a group homomorphism followed by a translation.
At first suppose that for any g E G the level set of ir over any closed point is mapped
to another level set of ir then we easily pass to the quotient, giving a bundle TIG -+ BIG
whose fibers are isomorphic to the original fibers of 7r : T -+ B, hence are Abelian varieties
and done. If for any g E G the level set of 7r over a general point is mapped to another
we axe
level set of ir then we may replace B and T by Zariski open subsets which put us in the
previous situation leading again to an a.c.i. system. In this case, if the original a.c.i. system
is completable then obviously the quotient is also completable.
The situation is very different when the general fiber is stable for the action of G, because
in this case we get non-trivial quotients of these level sets. In this case all points of SpecA
axe fixed (for the induced action), hence all fibers are stable and 4 C O(M)G. Since for
any g E G the morphism g. is Poisson, all vector fields Xf, f E A are preserved by the
action; by linearity of these vector fields the ar-tion of ear-h g E G on the level sets of ir is by
translation over an integral part of a period. The quotient of an Abelian variety by a Enite
group of translations is again an Abelian variety and we may form the quotient of T by the
action of G, obtaining a new bundle TIG of Abelian varieties over B. Thus in this case the
quotient is also a.c.i., but the Abelian varieties which appeax in this quotient system are not
isomorphic but merely isogeneous to the Abelian varieties which appear in the original a.c.i.
system (in particular the quotient system will most often have a different polarization type).
Again completability of the a.c.i. system (if present) is not lost. Notice also that in this last
case the group G acts as a group of translations, hence G is commutative if its action is
effective.
We are left with the case in which some elements of G map the level sets over every (or
the closed point to another level set, while some other elements fix all these level
generic)
sets. In this case one may take the quotient in two steps, since the subgroup of G which
corresponds to the latter elements is a normal subgroup of G. 0
The following converse of the above proposition is not true. Let (M, be a Pois-
son variety with a Poisson action by a finite group G and suppose that (M, -1,A) is an
integrable Hamiltonian system such that for each g E G one has g -A C A. If the quotient
(MIG, 1.'.10' AG) (which we know to be an integrable Hamiltonian system) happens to be
an a.c.i. system, then it does not follow that (M, 1-, -1, A) is itself a.c.i. See Paragraph VII.6.2
for a counterexample.
Notice that in the first case treated in the proof of the proposition the map g- : B -+ B
is a covering the moduli space of Abelian varieties which appear as level sets in the
map over
a.c.i. system. The quotient BIG is an intermediate object between B and this moduli space.
134
3. Painlev,6 analysis for a.c.i. systems
3. PalnleW analysis for ax.i. systems
The differential equations describing an integrable vector field of an irreducible a.c.i.

system possess families of Laurent solutions (see [AM7]). In slightly different terms this was
already known to Kowalevski; her original idea was taken up and extended by Adler and van
Moerbeke to give necessary and sufficient conditions for algebraic complete integrability in
terms of Laurent solutions (see [AM7]). We restrict ourselves here to part of their result17.
Proposition 3.1 Let (M, 1-, -1, A) be an irreducible a.c.i. system. Then for any integrable
vector field XH, H E .4 the space of Laurent solutions has dimension dim M -
1.
Proof
Pick a general fiber F of the momentum map -7r : M -+ Spec.4. By assumption F

is isomorphic to T \ D where T is an Abelian variety and D is a divisor on T; we denote
this isomorphism by 0. Let Z be any point on 7- and let us choose a system of generators
zi, . zm of 0 (M); upon restriction to F they lead to a system of generators of 0 (.F), which
. .
,
we still denote by zi. The functions zi o 0-' provide a system of generators of 0 (T\ V) and in
terms of these the differential equations for the vector fields 0,,XH (H E A) axe linear, hence
if Z E 7- \ D then one finds a solution for zi o 0', hence also for zi, which is holomorphic
in t. It is just the description of the integral curve of the vector field XH starting from the
point 0-'(Z) E M. Of course the dimension of the space of such solutions is dim M, since
we have precisely one solution for every point of M.
Next, suppose that Z E D is such that one or several of the functions zi o 0-' have a
pole at Z; notice that the functions zi o 0-' which are regular on T \ D uniquely extend to
meromorphic functions which have their poles on at least one irreducible component of V.
The divisor of e.g. z, o 0-' can be written (uniquely) as
k I
niVi -
miEX (mi, ni E N \ 10}),
where all Di and Di' axe different and irreducible. If z, o 0-' is not holomorphic around Z
then Z belongs to one or more Di, but it may belong as well to some of the Di. In any case, if
we pick for each irreducible component of D a local defining function around Z, say fi for Di
and gi for Di' (if Z does not belong to some divisor then the local defining function may be
taken as the constant function 1), then z, o 0-1 is written around Z as
fnl
1 fn2
2
...
fknk
Z' 0 f M1 M2 I
91 92 ...
gMJ
1
with f holomorphic around Z and f (Z) 9 0.
We may take linear coordinates x, ti X21 7 Xn

for the torus, and think of the local...
defining functions as being expressed in terms of these. The t-axis cannot be contained in
any of the divisors Di or Di' since otherwise the general fiber would contain a subtorus, i.e.,
17
talking in this section about the solutions to some differential equations (in
Since we are
other words theintegral curves of a vector field) we will talk here about holomorphic and
meromorphic functions rather than regular and rational maps.
135
be reducible, contrary to our assumptions. It follows that all these functions can (again up
to nonvanishing holomorphic function) be written as a (Weierstrass) polynomial in t (by
a
the Weierstrass Preparation Theorem) and we see that the zero or pole z, has in Z (as a
function of t) depends on the components of the divisor of Z, to which Z belongs but also on
the singularity these divisors have in Z (since then the first few terms in the series vanish)
and on the contact the vector field XH has with these divisors (for the same reason).
Proceeding in this way for all functions zi we find a Laurent solution to the differential
equations, which staxts from Z. Since Z is an arbitrary point of a divisor of the general fiber,
the space of Z from which there staxts a Laurent solution is of dimension dim M -
1.
Notice that this divisor is contained in but is not

necessarily equal to the divisor which
needs to be adjoined to the level in order to complete it into an Abelian variety. If there
is for every component of D at least one of the functions zi o 0-' which has a pole on this
component, then a Laurent solution stafts from an arbitraxy point of D and we have an exact
bijection of the space of Laurent solutions and the points to be added to the general level
sets in order to complete them into Abelian varieties. This is the case in all examples that
we will consider. 0
The Laurent series organize themselves naturally in families as follows- for every zi, fix
an intersection of some divisors (contained in the divisor of poles of (zi)), fix an order of
singularity and an order of tangency of the vector field. On this set all zi are written as
Laurent series depending on a number of free paxameters, equal to the dimension of this set
(corresponding to the staxting point of the series which can be chosen in it) and in a dense
subset the order of pole each expansion experiences is fixed. The pole may however become
less severe in an analytic subset, obtained from the intersection with one of the divisors on
which zi has a zero; in such a case the leading coefficient of the Laurent series must be
(dependent on) a free parameter, so that it can in paxticular take the value 0. Thus, there is
always at least one family of Laurent solutions which depends on dim M 1 free parameters -
(called a principal balance). If there is for every component of D at least one of the functions
zi o 0-1 which has a pole on this component, then there are as many principal balances
as irreducible components in D. There axe also always families of Laurent solutions which
depend on fewer free parameters, known as lower balances.

The geometric study of the Laurent solutions of one of the integrable vector fields of an
integrable Hamiltonian system is called the Painlev6 analysis of this system.
Remark 3.2 The different sets which correspond to the different balances do not give a
stratification of the Abelian variety in general; indeed, if, for example, z, and Z2 both have
a pole on some smooth divisor and the intersection of these divisors is singulax, then the
singular locus of this intersection will in general not show up as a sepaxate family of Laurent
solutions, leading to a singulax stratum. An example where the Laurent solutions do lead to
a (family of) stratification(s) of all hyperelliptic Jacobians will be given in Paragraph VI.4.2.
Finding all Laurent solutions in a direct way is in general a difficult task. One encounters
the following problems.
1) It is not clear by looking at the differential equations with which exponents to start
in order to find a (all) solution(s);
2) For a given choice of exponents one needs to solve a nonlinear system of alge-
braic equations (called the indicial equations) for the leading term, which may be
136
3. Painlev6 analysis for a.c.1. systems
verydifficult, especially when the number of variables is indefinite (see e.g., Para-
graph VI.4.2);
3) The presence of free parameters (giving information about the dimension of the cor-
responding subset) can in favourable cases be detected by computing the eigenvalues
of a matrix, depending on these
leading terms, but this is again very difficult when
the number of variables, hence the size of the
matrix, is indefinite;
4) One also has to show the convergence of all Laurent solutions;
5) It is not obvious to figure out how the different sets the different families of Laurent
solutions correspond to are related (see [AM7]).
The main use that we will make of Painlev6 analysis in the remaining chapters is not to
detect a.c.i. systems (see e.g. [Hai2]) or algebraic complete integrability (see [AM7]),
to prove
but to construct explicit projective embeddings of the Abelian varieties whose affine parts
appear as the fibers of the momentum map.
Proposition3.3 Let T bean Abelian variety and letV some divisor such thatAr= T\V
is an affine variety; consider also a linear vector field Y on T. If for every irreducible
component of D the space of Laurent solutions, which corresponds to a general point of it,
is explicitly known (in the form of the first few terms), then an explicit embedding of T in
projective space can be computed concretely from it.
Proof
We know that any irreducible component Di of D is ample since T is irreducible; more-
over 3Di is very ample and we may construct an embedding of the Abelian variety by us-
ing 3Di. For any irreducible component of D these Laurent solutions express a system of
generators zi, . . .
, z,, of 0 (M) (restricted to Ar) as Laurent series in t, hence can be used
to express any element of O(M) (restricted to JV) in terms of t. What is important here is
that the pole which the Laurent solution of a function zi on JV has in t coincides with the
pole the extension of f to T has on the divisor which corresponds to the Laurent solution;
this follows as above by writing zi (or zi o 0-1) as a quotient of holomorphic functions which
in turn are expressed in terms of Weierstrass polynomials, but now this is done at a general
point of the divisor. This allows to look for elements of O(M) which lead to independent
meromorphic functions on T which have a certain pole at D' but axe holomorphic on the
other divisors in D. This is a finite proces: we can first look for functions of degree one, then
of degree two and so on and by Formula (IV-3.1) we know when we have found a complete
set of (independent) functions and these provide the embedding. 0
Amplification 3.4 One can also construct an embedding by using some combination of
the different components in the divisor, taking these components only with multiplicity one
or two. It is easy to figure out which set of multiplicities suffices when the N6ron-Severi
group of the Abelian variety is trivial (i.e., is equal to Z). Notice that since this is the
case for a generic Abelian variety this is a mild assumption. Then every component Vi is
algebraically equivalent to a multiple of some fixed divisor and it suffices that, under this
algebraic equivalence, the divisor which one picks to construct the embedding is equivalent
to (at least) three times this divisor. One fixes such a choice of embedding divisor D' and
determines as before the Laurent solutions which correspond to the irreducible components
which belong to this divisor as well as the Laurent solutions which correspond to the remaining
components in D; from it a concrete embedding can be constructed.
137
Chapter V. Ax.i. systems
4. The linearization of two-dimensional a.c.i. systems
Let (M, 1-, -1, A) be a two-dimensional a.c.i. system and let us denote for any c E Spec A
the fiber of the momentum map over c by F,. If c is general then T, completes into an
Abelian surface by adding one or several (possibly singular) curves. The following algorithm,
proposed in [Van2] leads to an explicit linearization (i.e., integration) of any integrable vector
field H, H E .4. (steps (1) and (2) are due to Adler and van Moerbeke, see [AM9]).
(1) Compute the first few terms of the Laurent solutions to the differential equations,
and use these to construct an embedding of the general fiber.F, in projective space
(Proposition 3.3).
(2) Deduce from the embedding the structure of the divisors V, to be adjoined to T,
in order to complete F,, into an Abelian surface 7,,. At this point the type of
polarization induced by each irreducible component of D, can also be determined
(see Section IV.3.3).
(3) a) If one of the components of D, is a smooth curve r, of genus two, compute the
image of the rational map
2
0[2r,. _+ p3
which is a singular surface in P3, the Kummer surface IC, of Jac(r,).

b) Otherwise, if one of the components of D, is a d : I unramified cover Cr. of a smooth
curve IP, of genus two, p : C,, -+ ]U, the map p extends to a map jac(r.
In this case, let 8r, denote the (non-complete) linear system p*12]P,l C 12C.1 which
corresponds to the complete linear system 12r,J and compute now the Kummer
surface Kc of Jac(]P,) as the image of
oe'c : V _+ p3.
c) Otherwise, change the divisor at infinity so as to arrive in case a) or b). This can
always be done for any irreducible Abelian surface.
(4) Choose a Weierstrass point W on the curve IP, and coordinates (zo : z, : Z2 : Z3)
for p3 such that 0[2r,,] (W) =
(0 : 0 : 0 : 1) (3) a) and Oe.JW)
in case =
(0: 0: 0: 1)
in case (3) b). Then this point will be a singular point (node) for IC, and IC, has an
equation
2
P2(Z0iZ15Z2)Z3 + P3(Z07ZI7Z2)Z3 + P&O,ZI Z2) =
0)
where the pi are polynomials degree of i. After a projective transformation which
fixes(0 : 0 : 0 : 1) we may assume that
2
P 2 ( Zo 'ZI, Z2 ) =
ZI -
4zoz2.
(5) Finally, let x, and X2 be the roots of the quadratic equation _,OX2 + _,IX + Z2 =
0,
whose discriminant is P2 (ZO ZI i Z2) , with the zi expressed in terms of the original
variables qI, , q4.
.
Then the differential equations describing the vector field XH
. .
are rewritten by direct computation in the classical Weierstrass form
dxl dX2
- + -
=
aldt,
A/f(XI) Vff (X2)
xldxl -X2dX2
+ a2dt,
VY XIJ *V7f(X2)
138
4. The linearization of two-dimensional a.c.i. systems
where a, and a2 depend on c (i.e., on the torus) only. From it, the symmetric
functions X1 + X2 (--'z -zl/zo) and X1X2 (:-- Z2/ZO) and hence also all functions in
O(M) can be written in terms of the Riemann theta function associated to the
curve y2 =
f(X).
We will show below (in Section VII.5) on a non-trivial example that this algorithm is very
effective and easy to apply. Other worked-out examples can be found in [Van2]. We wish to
make the remark that it is also shown in [Van2] how a Lax representation and action-angle
variables for the system derive from the above linearization.
139
Chapter V. Ax.i. systems
5. Lax equations
An interesting way to construct integrable Hamiltonian systems is by means of Lax

equations. In many cases they turn out to be even a.c.i. We give a sketch of how this
works in the case of finite dimensional integrable Hamiltonian systems. The literature on
Lax equations is immense, original references are [Lax], [Flal], [Fla2], [AM2], [AM3], [RSI]
and [RS2]; the approach we describe here is due to Semenov-Tian-Shansky (see [Sem]). We
also wish to note here Garnier's paper [Gar], since it is the first paper (as early as 1919) in
which Lax equations (with spectral parameter!) were written down.
Let g be a Lie algebra with Lie bracket [-, -] and R an endomorphism of Z. The new
bracket
[Xi YJR 1QRX, Y] + [X, RYI)

2
satisfies the Jacobi identity (hence defines another Lie bracket on g) if and only if
[BR(X, Y), Z] + [Bp,(Y, Z), X] + [BR(Z, X), Y1 =

0, (5.1)
with
BR (X, Y) =
[RX, RY] -
R ([RX, YJ + [X, RY]) .
If so, then R is said to define a structure of a double Lie algebra on g. A paxticular class of
solutions to (5. 1) is found by looking for solutions to BR (X7 Y) =
0, i.e.,
[RX, RYJ = R QRX, Y] + [X, RY]) .
the so-called classical Yang-Baxter equation; more general solutions are obtained via solutions
to the modified classical Yang-Baxter equation
BR(X, Y) =
-c[X, Y], (5.2)
where c is a constant. If for

example 9 is a direct sum (as a vector space) of two Lie
subalgebras, ig ig+ =
+ g- then the
corresponding projection operators P+ : 0 -+ 0+ and
P_ -+,g- lead to a solution of this modified Yang-Baxter equation (for c 1), by taking =
R =
P+ -
P_, (5.3)
as is easy to verify. Notice that in this case the formula for the R-bracket takes the following
simple form
[X1 YIR =
1P+X1 P+y] -
1P_X' P_Y]. (5.4)
Having two Lie structures on g we also have two Lie-Poisson structures on g*, denoted
by 1., -1 and I-, -JR. Then the relevance of double Lie algebras for integrable Hamiltonian
systems comes from the following proposition:
Proposition 5.1 Cas({-, -1) is involutive for the R-bracket
Proof
Recall from Example 11.2.8 the explicit formula
If, gl(O df-( ) Wg co

,
140
5. Lax equations
for the Lie-Poisson bracket (f , g E 0 (9*), E g*). From it, it follows that
f E Cas 0 (g*) = - V E 9*, Vx E 9 ( 1&R(), x] ) = 0-
Therefore, if f , g E Cas 0 (g*) then
If, 91R( ) =
6 ( [RdfT6), dg- 6)1) + 6 ( [jfT6), RcFg 6)1 ) =
0,
showing that they are in involution with respect to the R-bracket.
Thus the algebra generated by Cas(j-, J) and Cas(j-, JR) is a good candidate of being
integrable on in order for Cas(I., J) to be big enough to imply integrability one
(ig*, I-, JR);
often has to restrict the phase space to a Poisson subvariety. For many choices of g and R
one finds indeed interesting integrable Hamiltonian systems in this way; for example if Z is
a simple Lie algebra then the root space decomposition of g leads to a natural choice for R
similar to (5.3) and one finds a large family of integrable Hamiltonian systems, the generalized
Toda lattices.
If g admits a nondegenerate, invariant bilinear form then the phase space Z* can be
identified with 9 and the differential equations which describe the integrable vector fields
Xf =
I-, f}R on g (for f E Cas(j-, -1)) have the following peculiar form
1 -
Xf L =
[L, Mf] where Mj 2R(df (L)). (5.5)
Such an equation is called a Lax equation and an integrable Hamiltonian system on a Poisson
subspace of the dual of a double Lie algebra will be said to be in Lax form or to be of Lax
type (to be distinguished from Lax representations, defined below). The determination of the
integral curves of (5.5) can be reduced to the Riemann problem (see [RS1]).
One may however also consider the loop algebra jg[X, A-'] of a Lie algebra g, which
inherits a Lie structure from g and has a natural decomposition
=
9[/\] ED A-I
with corresponding R-bracket given by (5-3). It a nondegenerate, invariant bilin-

also has
ear form if g has one. In this equations (5.5) retain their form but L
case the above Lax
and Mf are now dependent on A; this A is called a spectral parameter and the Lax equations
are also said to be Lax equations with a spectral parameter. Of course the loop algebra is
infinite dimensional but in concrete examples a finite-dimensional Poisson subspace is usually
considered. An example will be discussed in detail in Section VI.3.
Integrable Hamiltonian systems of Lax type for which the Lie algebra is a loop algebra
often turn out to be a.c.1. Still assuming that g has a nondegenerate, invariant bilinear
form, we may identify Z* with g and the spectral invariants of 0 are Casimirs of -1 when
viewed as functions on g*. Applied to the case of a loop algebra, L(,\) E and the
characteristic polynomial det(L(,X) zId) defines the affine part of an algebraic curve rL(,\)
-
in C x C*, namely the curve
rL(A) : det(L(A) -
zId) = 0
141
which is called the spectral curve. Thus there is associated to each matrix L(A) an algebraic
curve and L(A) moves under the flow of Xf in such a way that the curve which is associated
to it is constant; what moves under this flow is
a line bundle of degree 0 on this spectral
curve,
which is constructed from the eigenvectors of L(A) (for the precise construction, see [Gri]) and
thus the flow of L(,\) can be seen as a flow on Pico (rL(A)) i.e., on the Jacobian of
, rL(,\) (more
precisely on the Jacobian of the Riemann surface obtained by compactifying L(,) over 0
and oo). Linearity of this flow is not guaxanteed, although for many examples of interest
this vector field is linear and the above eigenvector map is said to linearize the integrable
Hamiltonian system; the Lax equations constructed in Paragraph 111.2.4 for example do not
have this property except in the special case considered in Paragraph VI.4.2. This can for
example be checked by the necessary and sufficient conditions for the eigenvector map of a
vector field (which is defined by a Lax equation) to linearize, which is given in [Gri].
The name Lax equation is also often used in the following sense (but it is not explained
in this way). Let (M, 1-, .11, A,) be an integrable
Hamiltonian system and (N, 1-, *12, A2)
be another integrable Hamiltonian system which is assumed to be of Lax type (as defined
above). Then a finitemorphism 0 : (M, 1-, -11, A,) -+ (N, J* ij2i A2) is called a Lax rep-
resentation of the integrable Hamiltonian system (M, 1., -11, A,.). Notice that even if the
integrable Hamiltonian system on N is a.c.i. the one on M needs not be, moreover the Lax
equations (5.5) lead to similar equations for (some of) the integrable vector fields on M,
namely we may compose the Lax equation (5.5) with 0 to obtain
Xf L 0 o =
[L o 0, Mj o 0], (5.6)
which represents the integrable vector field X4,.f on M. Notice that not all integrable vector
fields Xg with g E A, axe of this form since O*A2 = 6 A, (in general). Also the vector
fields (5.6) are not vector fields on an affine subspace of the dual of a Lie algebra. Special
caxe has to be taken when determining information about the level sets of A,, which axe
covers of (Zariski open subsets of) the level sets of A2; even when the degree is one, these
need not be isomorphic.
142
Chapter VI. The Mumford systems
In Paragraph 4.3 we will exhibit some interesting features of the even and the odd
Mumford systems. Namely we will show how Painlev6 analysis leads to a stratification of the
Jacobians which appear as the fibers of the momentum map; notice that the stratification
which is obtained in the even case is very different from the one which is induced in the
odd case. These stratifications axe also obtained from the natural stratification of the Sato
Grassmannian via (an extension of) the Krichever map, but this will not be discussed here
(see [VanQ-
(Section 5) we show how to construct for any smooth curve in C2
In the final section
an integrable Hamiltonian system which has a level set (of the momentwn map) which is
isomorphic, to an affine part of the Jacobian of this curve. Moreover, the restriction of the
integrable vector fields to this level are linearized by this isomorphism. For a large and
most important class of curves a.c.i. system. The construction which we
this leads to an
give here is a modification of the Chapter III, in particular the construction

one given in
is also completely explicit. A generalization, which is not a.c.i. was given in Chapter III
and is further generalized in [Van5] to arbitrary families of curves. For a generalization of
the construction in Section 2 to matrix differential operators, see [KV]. A variant of the
(even) Mumford system in which the polynomials have parities (odd/even) has recently be
constructed in [FV].
144
2. Genesis
2. Genesis
2.1. The algebra of pseudo-differential operators
We start by reviewing the basic definitions and properties of (formal) pseudo-differential

operators. Let D denote the non-commutative algebra C[[x]][i9] of differential operators, the
multiplication being given by juxtaposition and applying the commutation rule [19, a(x)] =
a'(x); here a(x) is any formal power series, a(X) E C[[x]], and a'(x) its (formal) derivative.
D has as a distinguished maximal commutative subalgebra the algebra D' C [a] of constant =
coefficient differential operators. D is contained in the larger algebra T C[[x]](((9-1)) of =
pseudo-differential operators, the (associative) rule of multiplication being formally derived

from the commutation rule [,9, a(x)] a'(x), i.e., =
00
,9-'a(x) =
i=O
An element Q C- T, Q EiL
z_=-.
aii9', is said to have order q if aq =A 0 and is said to be monic
(of order q) when aq
1; if, in addition, ag-1 0 then Q is called normalized (of order q).
=
The subalgebra C((a-')) of T consisting of constant coefficient pseudo-differential operators

is denoted by T1. The following properties of differential operators are easily established.
Proposition 2.1
(1) Every monic pseudo- differential operator Q of order q has a unique inverse Q`
in T. In particular the monic pseudo- differential operators of order zero form a
group, called the Volterra group and denoted by Volt. Its subgroup of constant coef-
ficient operators is denoted by Volt'.
(2) Every normalized differential operator Q of order q > 0 has a unique monic q-th
root Q11q in T. This root Q11q is normalized and has order 1.
(3) Every normalized differential operator Q of order q > 0 is conjugated to a q, i. e .,
Q = T` 9qT for some T E Volt which is unique up to left multiplication by an

element of Volt'.
Finally we recall the definition of the Sato Grassmannian Gr. Let us denote by 6 Dirac's
delta function, thought of as a zeroth order differential operator. It has the fundamental
property that for any Q E T there exists a unique Q' E T' such that QJ Q'6. The left =
coset
V:= C((,9-1))d = 'F-6
is a left T-module in a natural way: for P E T and for Q E C((,9-1)) c T we define

P -
(Q6) =
(PQ)6. A distinguished subspace of V is defined by
H =
C[a]6 = D6.
The multiplication allows us to associate to each element in XF a subspace of C((,9-'))6 in a
natural way, namely given Q E IF define WQ C C ((a-')) 6 by WQ Q H. We call the set of = -
all WT which correspond to elements T in the Volterra group the Sato Grassmannian,
Gr:= JWT I T E VOltJ
145
Chapter V1. The Mumford systems
If T E Volt the linear space WT has a basis, similar to the standard basis of H, namely it
has a basis whose elements have the form
T - a' =
(0" + 0(&-1))6, (i > 0). (2.1)
It follows that the map Volt -+ Gr given by P i-+ Wp is injective hence bijective. The
following proposition gives a useful characterization of differential operators in terms of this
map (see Nul3]).
Proposition 2.2 Let Q E T. Then Q E V if and only if WQ C H.
Since no confusion is possible we remove 6 from the elements of WQ, i.e., we identify V
with C((a-1)).
2.2. The matrix associated to two commuting operators
In this section we assume that P and Q are monic differential operators which commute,
[P, Q] = 0. We will assume that one of them, say Q, is normalized. Finally we assume that
the orders p and q of P and Q axe positive and coprime. In this paragraph we show how to
associate a unique element M E ]g I, [A] to (P, Q); it will be shown in the next section how to
reconstruct (P, Q) from M.
In a associate to the pair (P, Q) a pair (P, W) of elements P E Te and
first step we
W E Gr unique up to an equivalence specified below). Since Q is normalized

(this pair is
there exists by Proposition (2.1) an element T E Volt such that Q T-1,917T. Choosing such =
an element T we define W WT T H G Gr.

= If we let P= TPT-1 then 15 E T is monic
- =
of order p and [&, P] 0. We claim that the latter implies that P E Tc. To show this it is
=
sufficient to show that if a(x) E C[[xl] is such that [aq a(x)]

,
= 0 for some q > 0 then a(x) is
constant. Since
0 =
[Igq a(x)]
,
= aq a(x) -
a(x) aq =
qa'(x)a7-' + Opq-2),
we find indeed that a(x) = 0.

the choice of T: if T is replaced by T'T where
The pair (P, W) cleaxly depends on
Te E Volt' then (P, W) is replaced by
T' -
(15, W) = (TP(T')-1, T' W) - T' -
W),
since constant coefficient differential operators commute. The above equation defines an
action of Volt' on T' x Gr; we say that two elements in T1 x Gr are equivalent when they
correspond under this action. We now give a characterizing property of the pair (P, W) and
we use it to show that the pair (P, Q) can be reconstructed from it.
Proposition 2.3 The element W E Gr is stable under the action of & and
W,
(2.2)
P-WCW.
146
2. Genesis
Conversely, suppose that (P, W) E T- x Gr satisfies (2.2), P being monic. The pair (P, W) is
associated to a unique pair (P, Q) of commuting differential operators, such that P is monic
and Q is normalized.
Proof
The verification of (2.2) is easy, for example
aq -
W = T -
(Q H)-
C T -
H =
W,
in which the inclusion Q - H C H holds because Q E V. Let us show how to reconstruct

(P, Q) from (P, W). Since W E Gr it is of the form W = T -
H for a unique T E Volt. If we
define
Q = T-1 aqT)
P = T-'PT,
then Q is normalized of order q while P is monic of order p; also [P, Q] 0. The crucial =
property is that (2.2) implies that P and Q are differential operators. We have that WQ c H
and Wp C H because e.g. for P one computes
Wp = P -
H = T-1 -
(P W)
-
c T-1 W - = H.
Using this, the fact that P and Q are differential operators follows from Lemma (2.2). Clearly,
if we replace (P, W) by an equivalent pair then the same pair (P, Q) is obtained. 0
Notice that in the above proposition the orders of P and Q need not be coprime.
The next step is to associate to the pair (-P, W) (up to equivalence) a matrix M E 0 [q [A]
from which (j5, W) can be reconstructed. Since P E Te and P -
W c W it follows that P
is endomorphism of W, hence can be represented by a semi-infinite matrix (with entries
an
in C) by choosing any basis of W. This matrix becomes a periodic matrix (with period q)
when a periodic basis (EO, El, ) is chosen, i.e., a basis such that for any basis element Ej
. . .
,
the element &E, is also a basis element. Note that the existence of a periodic basis follows
from the inclusion &W C W. We choose q vectors in W of the form
Ej = 04 + 0(ai-1), (0 < i < q -
1),
and extend them to a basis of W by introducing, for arbitrary i

> 0, the vectors Ei+q &Ei- =
Periodic matrices can be rewritten as square matrices at the price of allowing entries which
are polynomials in 'X = aq . Explicitly, if we define the matrix'8 M (mij) Ei [A] by =
q-1
-
P -
Ej =
E MijEj, (0 < i < q -
1), (2.3)
j=0
The rows and columns of the matrix M and of the matrices eij, introduced below, are
labelled from 0 to q -
1.
147
then the elements of M have the following degree constraints:
:5 [p1q] [p1q] [p1q] -

I
< [p1q] [p1q]

(2.4)
[p1q] + I < [p1q]
< [p1q] + 1 [p1q] + I [p1q]
since
P -
Ei Ev+i + lower order in 19
P+i
-[ Q
'E(p+i) mod q
+ lower order in a.
implicit in (2.4) that when the degree is exact (no inequality signs) then the corresponding
It is
polynomial is monic. We denote the affine space of matrices of the form (2.4) by MV,q Let .
N - c GL(q) denote the subgroup of lower triangular matrices, which acts on Mp'q by
conjugation.
Proposition 2.4 The above procedure associates to the pair (P, Q) a well-defined element
of MMIN -.
Proof
Two bases of the form (2.1) axe related by an element of Nq- hence their matrices axe
N --conjugated. When replaced by any other representative T' (fl, W) then

(P, W) is -
precisely the same matrix M is obtained when using the basis of T' W which is obtained by -
multiplying all elements of the basis of W by T'. Notice that such a basis is aJways of the
form (2.1). 1
As it turns out the quotient space Mp,qlN - is in a natural way isomorphic to an affine
subspace f4P,q of MPq. To show this we need to introduce some notation which is motivated
by (2.4). For 0 < i, j :! q 1 let eij denote the

-
(q x q)-matrix with a I at position (i, j)
q-
and zeros elsewhere and decompose g Iq as 9 Iq 0 -
'7=11 q &Y where &Y is the subspace of 9 Iq

,
generated by the matrices eij for which j i - Wien Z E

&Y we also write deg Z
The projection 0 Iq -+ &, will be denoted by IIy. Let d =
p mod q, 0 < d < q, and let S and R
be the elements Of 91q, given by
S=
( It7-Od 0)' 0
R=
(0 1d).
0 0
We have that deg S = -d and that deg R =

q -
d. With this notation Mp,q consists of all
matrices F[Plql+l M.Ai

_i=O %
for which
M[plql+l E S+
-y<-d
M[,I,] E R +ED jgy.

,y<q-d
148
2. Genesis
The affine subspace M- of M- consists of those matrices in MP," for which
0 Ad
ML./"]+' =
S, ML./ql =
(* *),
where the stars denote arbitraxy matrices of the appropriate size. We also introduce an
intermediate space Mp"', which is also an affine subspace of Mp,q. The subspace MM
RS Rs
consists of all matrices in Mp,q for which
M[plq]+l =
S, MLIl (E R +ED 9,Y.
,y<q-d
Let us denote the Lie algebra of N - by n and let gs denote the isotropy algebra of S,
OS =
fX E nq I [X, S] =
01,
which is the Lie algebra of Gs exp9s, the stabilizer of S. We have that the (adjoint)
=
action of Gs on Mp,q leaves MPIq invariant. Notice that 9 s is given explicitly as the algebra
RS
of strictly lower triangular matrices M for which Mi+d,j+d Mij for all 0 < j < i < q = -
Proposition 2.5 Mpq A- is isomorphic Mp,q

RS IGs which,
in turn, is isomorphic to the
q(p + q -
2d) -dimensional affine space Mp,q.
Proof
We show that every element in Mp,q is unique element of 'A' p,q.
N --conjugated to a
The proof then follows from the fact that the corresponding f4p,q
map Mp,q -+ is regular.
Let M E[p/qj+l
=
i=O
M.Ai C Mp,q.
%
We first show that M[plq]+l is N --conjugated to S. Take
any element E g- 1 and let g =
exp E N - .
Then
Ad, MLI,]+, =
expadC M[Plq]+, =
M(Plqj+l + [ , M[Plq]+,]
which after projection on 10-d-I becomes
H-d-1 (Ad, M[plq]+I) =

II-d-1 (M[pl,]+I) + [ , S1-
Since the linear map ads 9-1 _+ 9-d-I is surJective we can pick g (i.e.
:
6) such that
II-d-I (AdgML,,1q]+I) = 0. Repeating this procedure with 6 E
0-,y where y
2,3.... and =
using the surJectivity of ads : 19-y _+ 9-d-,y we find that ML"1qI+I is N --conjugated to S.
Since, by definition, Gs c N - is the stabilizer of S and since the map Mp,q _+ Mp,q can be
RS
picked regulax it follows that MP,qlNq- is isomorphic to MP,qlGs.
RS
0 Id
We next show that ML,,Iq]
Notice that if M E MP' q then M[plq E R +
is Gs-conjugated to a unique element of the form
(* ) *
-
RS I G-j<q-d gy. Fix any 6 E 11.... 7q 11 a

consider the space of matrices of the form
* ...
* *1 0 ...
0 1 0)
0 ...
0 1
(2.5)
149
where the diagonal with the stars diagonal with the I,, axe precisely a
and the
distance 6 apart. The number s by s
is minjd, q 61. It is easy to see
given in terms of 6 = -
that the adjoint action of G6 exp(o-6 no,) on 0 Iq leaves the space (2.5) invariant. Also, at
=
level q-d-6 (the diagonal where the *i in (2.5) are) the adjoint action induces for any element
g exp
= E G6 the affine map (translation) l9q-d-6 -- 19q-d-6 : Z -+ Z + [R, ]. In turn this
map induces on C' the translation by II[R, ] where Il is

the linear map O[q -+ C' which
the matrix (2.5) to (*,, *,). We denote this map by X , thus X&) z + II[R, ] for =
maps -
..'
z G C8. We wish to show that there exists for any z C: C' a unique E 9-, n gs such that
XC (z) 0. Equivalently, that for any z E C' the affine map X , : g-, n gs -+ Ca :
= -+ Xc (z)
is a bijection. Taking the corresponding linear map this means that we need to show that
X : 9-, n 9. -+ C8 : -+ II[R, 6]
is an isomorphism. Since both spaces have dimension s =

minjd, q -
61 it suffices to show
that X is injective. Let be in the kernel of X. If 6 : q -
d then s =
q -
6 >
d 6 and
-
X( ) =
(01 01 ... 10,61, ... 16d-6) -
(611 ... 168)

so that X is injective. When 6 < q d the proof is more delicate and depends on the fact that
-
p and q are relatively prime (indeed if p and q have a common divisor then X is not injective
for some values of 6). Then
X( ) =
(01 ...
1 01 C17 C2, ...
7 Cd-6) -
(Cq-d-5+1 7... I Cq-6)i
where appears at position 8 +

j 1, the length of these vectors being 3 d. If is in the =
kernel of x then q-d-6+1 q-d== 0 and

...
G Cq-d+k
= for k = d 6. But = = -
remember that G Ck+d because C E gs. This means that the indices of C may be thought
=
of as lying in Zd. Now the fact that p and q are relatively prime implies that d and q are
coprime, hence also q d and q. The fact that G
-
Cq-d+k for k 1, d 6 then implies

= =
. . .
,
-
that all .y are equal, hence they are all equal to 0. Thus X is injective in all cases and we can
make all *j in (2.5) equal to zero by using a unique element of G6; doing this consecutively
for 6 1, 2,.. -, q
= I leads to the desired result.
-
We call M E Mp,q the matrix of (P, Q) or of W).
2.3. The inverse construction
We will show that every element M E Mp,q (with p and q coprime) is the matrix
now
of pair (P, Q)
a of commuting differential operators such that P is monic of order p and Q is
normalized of order q. By Proposition 2.3 it suffices to show that M is the matrix of a pair
(P, W) E Tc x Gr satisfying aq. W C W, W C W and P monic. Equivalently we need to
show that there exist a monic element P T' of order p and q vectors E0, . . .
, Eq- I in W
such that ord Ej =

i, and such that
q-1
Ej 1: mijEj. (2.6)
j=0
150
2. Genesis
In order to do this we expand and Ej in terms of 0,
P qP
+P20W-2
Ej
=
=
0
a+ g,
+ plo
iF-1
,,gi-I + g2iai-2 +...'
(2.7)
and we define for r E Z an element MH E of, by

MH where mij (2.8)
r
Lemma 2.6
(3.) M(r) 0 for r < 0;
1 if p+i-j=Omodq,
(2) M 27)
O if P+i-j:AOmodq;
(3) M(O) is the matrix of a cyclic permutation or of 10, q -
11.
Proof
If r < 0 then
P+i-i
p + i -
j -
r> q
I q
ord mij
and so mi(jr), which is the coefficient of ap+i-i-r, is

7,3
zero. The same inequality holds for r 0
when q does not divide p + i -
j so that for such values of i and j we also have m 9) = 0. If q
divides p + i -
j then mij
is monic of order p + i j hence 1. It implies that M(O) is -
m 9) =
a permutation matrix. The fact that this permutation is cyclic follows from the fact that p
and q are coprime; indeed, this permutation corresponds to the translation over p in Z.. I
If we plug (2.7) and (2.8) into (2.6) then we find that for any -Y 0 and for 0 < i < q -
I
y q- I
Eprg') Y-r 1: 1: MlVg j-r, (2.9)

r=O r=O j=O
which can also be written as
0 0
Pr! -r !4-r
E
r=O
Prg,qy
Prgi-r
-
I
r
LM(r)
r=O
9 -r
9j-r
q-1
(2.10)
for any -y > 0. We show that (2.10) can be solved recursively for pj and g, . Since po gi0 = 1
for i 0,
=
q 1,
. . .
,
-
and since M(O) is a permutation matrix, the equation (2. 10) is satisfied
identically for y 0. = Let us assume that we have constructed Pr and gr' for all r < -y and
i E 10, q -
1}. Then equation (2. 10) can be written as
Ay 90
2
(AyRy
(M(O) -
1q)
gq
g'q-1
Y
+ (known stuff)
151
where "(known stuff)" involves only the previous gi and p,. Summing up the q rows of this
equation we find p, because the sum of the rows of any permutation matrix equals the sum
'
of the rows of the identity matrix. We take = I and we use (2.10) to solve for I q-1
9 7 9 ...
194
To see how this is done, rewrite (2.10) as follows,
0
9
I
9,
(MO) -
1,) (known stuff).
g,q-
Y
Recall that M(I) is a permutation matrix which corresponds to a permutation a of the set
10, 1, . . .
, q -
11. Since o- is a cyclic permutation of order q we can solve for the g'Y in the
following order: solve first for i or(O), then for i 0.2 (0) and so on. Notice that the last
= =
equation is precisely the equation defining p, showing that the solution exists and is unique
once the vector g.0, has been chosen. Clearly the freedom in choice for
gOY corresponds to the
left action of Volt'. Notice that it is only in the very last step that we used that p and q are
coprime.
Summarizing we have shown the following proposition.
Proposition 2.7 Let ME Mp,q, where p and q are coprime. Then there exists a pair
(P, W) E T' x Gr such that P W C W and aq_W C W and such that M is the matrix
-
of P.
The pair (P, W) is unique up to the left action of Volt'. The correspondence which associates
to thepair (P, Q) its matrix is a bijection between mp,q and the space of all pairs (P, Q) of
commuting differential operators with P monic of order p and Q normalized of order q.
2.4. The KP vector fields
In this section we will realize the KP vector fields, which are a naturaJ collection of
commuting vector fields on the Sato Grassmannian Gr, as a collection of commuting vector
fields on the affine spare 1CIp, q. In order to write down the KP vector fields on the Grass-
mannian, let us first show that the tangent space at a point W of the finite-dimensional
Grassmannian G =
G(k, n) of k planes in Cn is naturaJly given by Hom(W, Cn/W). To see
this, we consider G as the homogeneous space GLn1 Stab(W), where GLn GL(n, C) and =
Stab(W) C GLn is the stabilizer of W, with Lie aJgebra
stab(W) =
10 E g1n I O(W) c Wj.
Then TWG =
TW(GLnl Stab(W)) =
gIn/stab (W) =
Hom(W, Cn/W); for the last equality
one associates to a representative 0 of an element of gIn/stab(W) the composite map
0
W --, Cn ,
Cn ,
V/W
In the of the Sato Grassmannian (which is infinite-dimensional) we define the tangent

case
space at point W E Gr to be given by Hom(W, TI1W), where we consider W in the

a
last equation as a subspace of V. In this language the i-th KP vector field is given by
Vi : W -4 TI1W : w i-+ a'w mod W.
152
2. Genesis
We first transfer these vector fields to the space of pairs (P, Q) of commuting scalar
differential operators with P monic of order p and Q normalized of order q. To do this we
use the bijection Volt -+ Gr : T i-+ WT, which gives the following commuting vector
fields
on Volt
dT
T(T-la'T)-,
dti
(see [1\4ul3]). If we have a constant coefficient pseudo-differential operator & and we define
U U(t) T-I&T then
= =
dU
[T-lft (T-'O'T)-] =
[U, Q"q] =
[Q'Il,
+ U].
Tti
Applying this for & given by lgq and by 16 we find the following Lax representation for the
KP vector fields on the above space of commuting operators (P, Q),
Q i dP i/q
t-i
=
[Q+/q7 Q17 Tti
=
[Q+ p]. ,
We proceed to write these vector fields down on MP'q. More precisely, we will write down the
vector fields that correspond to the constant coefficient scalar differential operators [,5i/),k] ;
each KP vector field is then a linear combination of these vector fields. We fix i, k and denote
the derivative in the direction of the vector field corresponding to
[P'/Ak] +
by a dot. We
choose a periodic basis of W and denote the column vector containing its first q elements
by e. By the above interpretation of the tangent space at W to the Grassmannian, we can
write
e =
[P'/Ak] +
6 -
A(aq) e, (2.11)
where A is the polynomial matrix (in A =

aq) such that the order of the i-th component
of the right hand side of (2.11) is smaller than i. Also equation (2.3), which is the defining
of the matrix M E Mp,q with respect to the chosen periodic basis, can be rewritten
equation
as
Pe = Me. (2.12)
If we differentiate (2.12) then we find P6 = Me+ Me, which is easily rewritten as
k,5 =
(MA -
PA) 6 =
[M, A] 6,
(one that elements of D'- commute among themselves and with matrices which are
uses
independent of x, such as M and A). Rom the last equality we can conclude that k [M, A], =
because k, M and A are polynomials in A & (rather than in 8). We claim that A can be
=
taken as (MiIAI)+. To prove this we must show that the i-th component of
(jii/Ak) +
e -
(Mi/Ak 415
has order smaller than i (for i =

1, . . .
, q). Since P commutes with M we have that
(pi/,\k) e =
(Mi/Ak) 15
153
and it suffices to show that the order of the i-th component of
(.Pi/Ak)- 6_ (Mi/,Xk)_,5
is smaller than i. Since the i-th component of 6 has order i this is clearly the case for the
first term; in the second term every component has negative order because M
depends on
A =,9q only. Thus we have shown that the KP vector fields lead to the following
commuting
vector fields on Mp,q:
dM
dtij= IM, [Mi/A!"]+]- (2.13)
Given M E MP,q have shown in Section 2.2 that there exists

we a unique g G
N - such that
Y =
gMg-1 E kPA. Differentiating Y =
gMg-' we find for any i, j the following commuting
vector fields on kP,9:
dY
=
ly, lyi/Aj+lj+ -
w1l. (2.14)
dtij
In the next section we will show that these vector fields are Hamiltonian with respect to a
family of compatible Poisson structures on .1 jp,q.
154
3. Multi-Hamiltonian structure and symmetries
In the previous section we have constructed a family of commuting vector fields on the
affine space Mp,q C q[q [A]. In this section we will show that these vector fields are multi-
Hamiltonian. We will do this by using the loop algebra i(q Of 91q-
3.1. The loop algebra i-t,,

The loop algebra j_1q of qIq is defined by
91q == =
131JAI ED A-I
Elements of the loop algebra will be denoted by capital letters; for an element X =
X(A) =
E xi Ai iE j_Iq we write X =
X+ + X_ according to the above (vector space) decomposition.
We define an inner product on 91q by (x, y) = T ace xy, which is non-degenerate and ad-
invariant, (x, [y, z]) ([x, y], z). According to Example 11.2.8,
=
0(91q) carries a natural Poisson
bracket, which we will denote by I., J. The inner product leads to an ad-invariant, non-
degenerate inner product (-, -) on i1q via
(X (,X), Y (A)) =
Res,\=o (X (,X), Y (A)) ,
where the right hand side is a shorthand for Ei+j=-, (xi, yj) =
EiEz Trace xjy_j_j.
We introduce an algebra 0(j_Q of functions on j-1. for which we can define a gradient
and a Poisson bracket as in Example 11.2.8, but which is large enough to contain functions of
the type X(A) -+ Res H(X'(X)) (for H E 0(g[,)), which will be important later. We define
on an algebra of functions by
0(b<n) = C IV] qf*
where C' denotes, for C E q(q and for .5 E Z, the linear map FxjAj i-+ C(x.). On it, we
consider the following algebra of functions:
0(q[q)= F:j 1-+CjVnEZ:Fjb E O&Jj- q <n -
Thus, elements of O(ZI.) restrict to polynomials on all subspaces B<n- As in Example 11.2.8
the gradient VF(X) of a function F E 0(j_[,) at X E it, is defined by

d
(VF (X), Y) =
dt lt=o
F(X + tY) VY E i1q. (3.1)
Proposition 3.1 For any X E 91, and F E 0(gl,), VF(X) is well-defined by (3.1) and
belongs to 4 - Moreover, for any F, G E 0 61q) the Poisson bracket IF, Gj, defined by
IF, Gj(X) =
(X, [VF(X), VG(X)])
belongs to 0(i-io, making 0(j[q) into a Poisson algebra.
155
For I c Z let R, denote the endomorphism of of, defined by
R: glq -+ Oil : X + X+ -
X-,
RI: DIq of,: X -+ R(A'X).
Since any linear combination of the endomorphisms RI satisfies the modified classical Yang-
Baxter equation (V.5.2) the brackets
IF, G11 (X) =

(X, [RIVF(X), VG(X)] + [VF(X), RIVG(X)]), (3.2)
2
form, for I E Z, a family of compatible Poisson brackets. We call them R-brackets to distin-
guish them from the above-defined canonical Lie-Poisson bracket on 0( i-iq) -

Consider
i+1
the Ad-invariant functions Ki : x t-+ T ace '
(i =
0'..., q -
1) on jg[q and define, for

i+1 , any
j E Z, the function
(X (A))
Hjj:j(q-+C:X=X(A) -+Res Ki),j+l (3.3)
Clearly Hij E O(j(q) and VHij(X) = XiA-i-1. Therefore each Hij is a Casimir for 1-, -1
and Proposition V.5.1 implies that all functions Hij are in involution with respect to all
R-brackets 1-, -11, which can also be deduced immediately from (3.2). As we have seen in
Section V.5 the Hamiltonian vector fields that correspond to such functions can be written
in Lax form; taking for example the 0-th R-bracket the function
Hij leads to the vector field
Xjj =
I , Hij 10, which takes the Lax form
- -
dX 1
=
2
, RVHij (X)] .
(3.4)
dtij
Two alternative ways to write this are
dX
=
[X, (VHij (X)) j = -
[X, (VHij (X)) ]. (3.5)

dtij _
Notice that the equations (3.5) and (2.13) are formally the same but axe defined on different
spaces. The vector field (3.5) is in fact Hamiltonian with respect to all brackets I-, J, since
I -, Hij 10 =
I -, Hij+j 11. (3.6)
Therefore, Xjj can be written in Lax form with respect to all endomorphisms Rj,
dX .1 [X, R,
=
2
VHi,j+l (X)] .
dtij
We now show that on it-, all the R-brackets I-, JI, the Hamiltonians Hij and the vector fields
Xjj axe linked by a vector field V, which has the deformation property with respect to all
these R-brackets. V is defined the infinitesimal generator of the action of C
as on 91, given
by "shift in A",
8, E XiXi) Xi (A + S)i;
156
here we use for negative powers of A the formal expansion
+ s)-1 =
i>O
which is actually convergent for small s, in particular it is the right definition if one wants
to consider the fundamental vector field V of this action: the latter is easily computed as
Vx(,\) =
'9
'6-1N X(A), i.e., V ' (i + 1)&' for any E gl*.
=
q
The properties of V are given by the
following proposition, whose proof is an easy consequence of the definitions.
Proposition 3.2 Let i, 1 E Z -
(i.) The Lie derivative of the 1-th R- bracket is (up to a factor -1) the (I -
1) -th R-bracket,
,CVIF, G11 -
1,CVF, G11 -
IF, f-VGjj =
-11F, G}1-1, (3.7)
for any F, G E 0 (it,);

(2) CVHjj =
(j + I)Hi,j+,;
(3) CVXij =
[V, Xii] U + 1)Xi,j+i
=
The conclusion is that we have for every 1, m E Z and for every Hi a bi-Hamiltonian
A typical fragment of it is (for
hierarchy with respect to the R-brackets 1-, -11 and I-, jm-
m =
0) depicted as follows (we omit the coefficients).
'C" 'C
Hij Hjj+j -
Hjj+21
0 0 0
1
Xij Xi,j+l Xi,j+21

'ClV 'ClV
3.2. Reducing the R-brackets and the vector field V
We will now apply Proposition 11.2.27 to obtain a family of Poisson brackets and a vector
field
relating them on Mp,q Before doing this we truncate g(q to a finite-dimensional Poisson
.
subspace with respect to some of the brackets.
Proposition 3.3 Let M'jq denote the affine subspace of it, defined by
mp"
S
=
I[Plql+l
E
i=O
MiA' 1E Ofq I M[p1q] +1 =
S
-[plqj+l
Mp" is a Poisson subvariety of 91q with respect to the Poisson structure E1=0 C11- I JI,
S
where thecomplex numbers ci are arbitrary. Moreover V is tangent to Mp,q
S ,
hence it has the
to each of these restricted Poisson structures on M11".
deformation property with respect S
157
Proof
By Proposition 11.2.18, M jq is a Poisson subvariety of gl, if and only if its ideal I is a
Poisson ideal of 0 (i-Q. For 0 < i, j < q let 6ij =

(- eij)
,
E 0 [,* and consider for 8 E Z the
linear function QIj on i-tq; it is convenient to write QIj as follows
(- 1)
-
illj ==
, eij A' -
The ideal I is generated by the functions jfj where s f 10, [p1q] + I I and by the functions
. .
jljl -
jj(S), where s =
[p1q] + 1. Since V j8j eijA-1-1= we find for any I E Z the following
formula for the RZ-bracket.
Cj', di I I = 68
1 (jkA81+t+1 -Ji4aj (3.8)
where eft = I if s, t < I and eft = -I if s, t : 1; otherwise eft = 0. Substituting any of the
above generators of I in (3.8), together with an arbitrary ktj, one finds that the resulting
linear function contains only terms where w 0 10,..., fp/q] + 11, showing that -E is a
Poisson ideal. The fact that V is tangent to MP'q follows in a similar way from the explicit
formula V jfj (s + = for V. 1) j'j+' N
The restriction of the above Poisson structures to M"'

S
and the restriction of V to Mp1q
S
will be denoted in the same way as the corresponding structures on if,-

We are now precisely in the case of Proposition 11.2.27: M"'
S
is an affine Poisson variety
(w.r.t. where 0 < I < jp1q] + 1) on which the stabilizer GS of S acts; MP1q
RS
is a
subvariety of Mp,q
S
which is Gs-stable.
Proposition 3.4 The triple (Mp,q

S GS, M,,q
RS ),
is Poisson-reducible with respect to the Pois-
son bracket E v/ql+'

1=0 &, -11, where the complex numbers ci are arbitrary.
Proof
We first show that the action of Gs on Mp,q
S
is Poisson (where Gs is given the trivial
Poisson structure). Actually (diagonal) the action of

GL(q) on gfq is Poisson with respect
to 1-, -11 for any I r= Z. To see this, take any 1 E Z and g E GL q). It is sufficient to show
*
that (Ad,) I fl, f2 11 I (Ad,) fl, (Ad,) f2 11 for any fi, f2 E 0 (jTq) that are linear. For fi
=
* *
linear one has that (Ad,)*fi is also lineax hence Vfj(X) is independent of X G gl, and we
can omit the argument X. Since
d
Wt- jt=O f, (Ad, (X
+ tY)) =
f, (Ad, Y) =
(Vfi, Ad, Y)
we find that (V(Ad,)*f,,Y) =

(Ad,--i Vfi,Y) giving V(Ad,)*fj. =
Ad,-i Vfi. Then
f (Ad,)*fl, (Ad,)*f2j, (X) =

(X, [Adg-i Vfi, Adg-, Vf2].,,,)
=
(Ad., X, [Vfi, Vf2]R,)
=
ffi, f2 11 (Ad, X)
=
(Adg)*jfj, f2j,(X).
158
Since Ap,q is Gs-invariant we are only left with the verification of condition (11-2.17). The
ideal _T of MP,q
RS
in Mp" is generated by those elements of the form
S &" S3
-
ij(R) for which

i < d q. For 1 0, [p1q] these elements are Casimirs of if X E Mp'q
-
=
Indeed, S
andO<m,n<q- I andO< s< [plqj then
'&[Plq],
'ij Cnn 1, (X)
.8
= '[Plq]'a
I (81111d/ ql+8+1-1(X)
in P
-1
(X))
which is zero: for s zA 1 this is seen at once, while for s = I one computes
j6?J1q]i 6minjI(X) 6in6m[Pj/q]+1(X)

= -
6mjp
[Vlq]+l
in (X)
=
(din emj 6,njein, S)
-
=
(femn eij] S)
i
=
([eij, S], emn)
and one finds again zero because j i + deg S j i d < -q. We have shown (11.2.17)
-
= - -
when I :A [p1q] + 1. Suppose now that I [p1q] + 1. Take F E O(Mp,q, Mp,q)G

= s
and notice
S RS
that this implies that F satisfies the infinitesimal condition
([VF(X), X], V) =
0, VX C f4p,q, VV E OS. (3.9)
Verifying (11.2.17) amounts to showing that 101q] Fl,,I,,+,(X) =OforanyX Gf4P,q and
wI
for ij suchthatj-i <d-q. But
1 ?3 /"I, Fj [plq]+l (X)

%3
1
=
X, [eij, VF(X)] [A-[Vlql-leij,R (,X[p/ql+lVF(X))]
2
+
1
=
2 (X, [eij, VF(X)] [A-W'71-'eij,AWq1+'VF(X) (,\[Plql+'VF(X)

+ -
2
=
(X, [eij, VF(X)]) (X, (AWq1+'VF(X)) j )
-
Both terms vanish: the first one in view of (3.9) and the second one because it can be
rewritten as
(S, [eij, (AIP/ql+l VF(X)) ([S, eijj, (A[Plql+-'VF(X)) _1
In view of Proposition 2.5 we have shown that Mp,q carries a ([plq] + 2)-dimensional
family of compatible Poisson structures. We will now show that the restriction of V to X4p,q
has the deformation property with respect to these Poisson structures.
Proposition 3.5 Let 1-, -1 =

E Plql+'
1=0 cjj-, J, be any R-bracket on Mpj" and let I., JV
,CVj-, .1. Denote by W the restriction of V to M-14. Then Cwf- j'-j' I-, J'V where I-, -I' =
,
(resp. V
is the reduced brucket of I-, j (resp. I., JV) on MP,q. In particular W has
the deformation property with respect to
159
Proof
First notice that the triple (M111,
S GS, M,,q
RS )
is Poisson-reducible because
[p/q]+l
I-, -IV Cal-,
an immediate consequence of (3.7). Let fl, f2 0 (Mp,q)Gs

RS
4'p,q and let us denote the
restriction of V to MpWr7S by W (notice that its further restricition to 'Qp,q is also denoted by
W). We need to show that
Iflif211V =LWIflif2li-fcWfl)f2lt-ffl7,CWf2li-
To see that this formula makes sense one it suffices to prove that (Mp,q)Gs
RS
C (.A4p,
RS q)Gs
to show this the fact that the actions of LW and GS commute, a consequence of
inclusion, use
the fact that the flow of DW is given by 0, : X (A) --+ X (A + s), while GS acts by simultaneous
conjugation. If we let F1, F2 C 0 (MP,q)
S
such that f, =
p(FI) and f2 =
p(F2) then
Iflif211V =pfF1,F2JV
=
pCV IF,, F21 -
pJCVF1, F21 -
pJF1, LvF21
=
pCV IF,, F21 -
JpCVF1, f2l -
Ifl, pCvF21
so pCv(F) =,Cwp(F) for any F E 0 (M,,q,

it suffices to show that
S
M_,,q
RS ).
Let us denote
by (Mp'q)'s
ir* the inclusion map 0
RS
__+ 0 (Mpq) (which may be thought of as coming ftom.
RS
"
the quotient map Mp"
RS
-+ M' RS IGs)
and by Mp" _+ Mp,q the inclusion map. Then
RS S
z*(F) ir*p(F) for any F E 0 (Mp,q,
=
S
MpqRS ).
Therefore
7r*LWp(F) =,CW7r*p(F) =
Lwz*(F) =
z*,CVF _Q 7r*pCVF
and the result follows from injectivity of 7r*. In (i) we used the fact that
GS
1CV 0 (Mp,q,
S
Mp,
RSD COO4p,q
S 7
Mp,q)Gs
RS
(Mp,q)Gs o,q Gs.

an easy consequence of f
_, RS C WRSJ '
Z.,CVO (Mp,q,
S Mp,q)Gs
RS
=
LW%*O (Mp,q,
S MV,q)Gs
RS ='CJ'V7r*O (Mpq)Gs
S
Gs
=
7r*LWO (MPi") c 0 (Mp,q,
S Mp,q)Gs.
RS
There is a final, innocent, reduction that can be done on these systems. Namely, since
VHoj (X) =
Idq A_j_' the coefficients of Trace(X) axe Casimirs and we can restrict phase
space to traceless matrices. The multi-Hamiltonian structure, the commuting vector fields
and the vector field W are just the restrictions of the original ones to this smaller phase
space.
160
4. The odd and the even Mumford systems
4.1. The (odd) Mumford system
In the preceeding section we have described a family of commuting Hamiltonian vector

fields affine subspace of it,, which, as we said, can be taken as an affine subspace Of ;Iq
on an -
For q =2 and p 2g + 1 we call the corresponding system the (genus g) Mumford system,
=
because it was Mumford who first wrote down (in [Mum5]) explicitly these commuting vector
fields and showed their algebraic complete integrability (the Hamiltonian structure is absent
in [Mum5l). For reasons that will become clear in the next paragraph we will often refer to
the Mumford system as the odd Mumford system.
Since we are considering special values for p and q we will simplify the notation, used
earlier in this chapter; the new notation will also take into account the fact that we only
consider traceless matrices from now on. We will denote the phase spare of the genus g
Mumford system, which consists of the traceless matrices in X4, 2g+1,2 by Mg. EXplicitlyj Mg
consist of all matrices
v(A) U (A) )
A(A) -
( W(A) -V(A) J '
where u and w axe monic, deg u =

g =
deg w -
1 and deg v < g. We have that S =
(0 0)
1 0
0 1 1
and R =
0 0 ) so that the group GS consists of all matrices of the form

a 0)
I
with
Liealgebra gs g-,. The space of all

= traceless matrices in M2g+I
RS
'g
resp. M2g+l,g
S
will be
denoted by M9RS resp. M9S. A general element of either one of these spaces will be denoted
by
A(A) -
03 (A) ii (A)
for which t7v is monic of degree g + 1, while both ii and 0 have degree at most g, with the extra
condition that fi is monic when A(,\) E Mg R s.
Since GS acts by conjugation the quotient
map Mgs -+ Mg is explicitly given by
Ui =
iii,
Vi =
Oi -
fiiOgI i 07 11 ...
191 (4.1)
wi =
iv-i + 20iO, -
fii F, 2,
g
where ui, fii.... denotes the coefficient of Ai in u (A), ii (A), Using (4. 1) it is easy to compute
the reduced brackets on Mg, as given by (11.2.27). To do this, notice first that the Poisson
brackets (for 0 < I < g + 1) on MgS are given by
ij
ffiw jll ='51 i4+j+1_11
10i'l7vj}1
Jt7vi, fij I I =
2e'jJ Oi+j+l -1
t
(recall that e-I 1 if s, t < I and el = -
1; otherwise el t
I if s, t > 0) and all other brackets
=
(between linear functions) are zero. For example if I =A n then Ivi, wj 11 is found from
10i -
fiiO, iv-j + NjO, -
fijOg211 fii+j+1-IOg2 + 20i+j+,-,Og) + fiidj',

161
giving Ivi, wjjt =

ej'3'wj+j+j-j +%M. In this way the reduced brackets are found to be
given, for I 0, 1,
=
g, by
lui, Vj 11 =
61 Ui+j+1-11 lui, Uj 11 =
0,
Ivi, WjIl =e ijwi+i+l-l + UX Ivi, VjIl =
0, (4.2)
jwj, uj 11 =
2dj'j Vj+j+j-j' jwj, wj 1, =
26ivj -
261jvi,
while the bracket I-, -lg+l is quadratic and is given by
lui, Vj 19+1 =
?g,+, ui+j-g -
UiUj' lui, Uj Ig+1 =

0,
N, Wj Ig+1 =
ei"+l wi+j+-g -
UiWj' IN, Vi Jg+1 =

01 (4.3)
jwj, ujl,+, = 2e'j 2ujvi, jwj, wj Ig+1 =
2viwj -
2vjwi.
9+ lvi+j-g
-
For linear combinations of the Poisson structures i =

0... g a compact formula
can be written. Namely, let W =
Ej9-0 cjA' be a polynomial of degree at most g and let'9
I., JV = -
Eq 0 cif-, -Ii. Then it follows from (4.2) that I-, -1 0 can be written in terms of
generating functions as follows:
IUM, UMII =
IVM' V(Y)I1 0, =
U(X)W(Y) U(Y) O(X) -
IUMMOI, =
X-Y
-2v(X)W(y) -
V(Y)V(X)
ju(x),w(y)j'P =
7 (4.4)
X-Y .
W(X)W(Y) -
W(Y)W(X)
IV(X), WMY =
U(X)W(Y),
X-Y
jw(x), w(y)j o = 2 (v(x)W(y) -
v(y)V(x)) .
The Hamiltonians in involution are the coefficients of TraceA(A)2/2, i.e., the coefficients
of H(A) U(,\)W(,\) + V2(,\). Rom (4.4) we can compute the Hamiltonian vector fields
=
Xy =
I-, H(y)J1, where y E C as well as the vector fields Xi I-, Hil', where Hi is the =
i-th coefficient of H(A) (notice that the function Hi, used here, coincides to the function Hji,
defined by (3.3)). For example
2u(A)v(y) U(Y)V(A)
X1,(U(A)) =
IU(A), U(Y)W(y) + V2(Y)j1 =
A-V
leading to
A(Y) 0
d-A(A) =
A(A), - -
0)]
WKY A -
Y U (Y ) 0
Writing Hi =
Resv=o H(y)/yi+l we find
d 0
Tti
A(A) =
[A(A), (A(,\)/,\'+')+ ( 0)]. -
Ui 0
(4.5)
19
The minus sign has been put in to make these brackets coincide (when q =
2)
with the brackets defined in Chapter III.
162
4.2. The even Mumford system
In the previous sectionhave constructed the (odd) Mumford system. Alternatively it

we
can also be obtained from thehyperelliptic case considered in Paragraph 111-2.4. We show how
this can be done be constructing an integrable Hamiltonian system that is
very similar to the
Mumford system. It is an open problem to obtain this variant of the Mumford system (which
was first constructed in [Van2]) from the algebra of pseudo-differential
operators and/or from
a reduction on the loop algebra 4-
In the language of Paragraph 111.2.4 we take F of the form F(x, y) y2 f(X), where = _
deg f =
2g + 2, we take W 1 and choose d
=
g, the genus of the hyperelliptic curve y2
=
f (X).
Then the Lax equation (111.2.14) takes the form
dti
A(A) =
[A(,\), [Bi(,\)]+] ,
where
V(A) U(A) A(A)

A(,\) =
( W(A) -V(A)
Bi(,\) =
u(A) [u(A)]Ai+1
+
and w(,X)=_ [ F(l\ (A)v(,\))]
u
+
(4.6)
The highest flow A(,X) =

Xa.,-,A(A) is the most important for us and is simply given by
V(,X) U(,X)
A(A) =
[A(,\), B(A)], A(,X) =
(W(A) -V(A ) ) , B(A) =

(bo)
('X 1), 0
(4.7)
where one computes b(A) from
V2(,\)
b(A) =
M11)]
('X)
U
+
and W(A) I U(,\) ]+' (4.8)
We want these systems to be extended to a larger phase space in the sense of Amplifica,
tion 111.2.6. As deformation family M we take
M= fR (X' Y) =
y2 _
(X2g+2 + a2g+1 X2g+l + - - -
+ aix + ao)j,
which is an affine variety isomorphic to C2g+l .

Notice that M contains an (odd) equation for
every hyperelliptic curve of genus g. The g + 2 coefficients a ..... a2,+l are Casimirs for this
larger system and the above Lax pair, supplemented with the equations coming from these
Casi:rnirs, describes the integrable Hamiltonian system on the larger phase space Cg+2 X C2g
C3g+2 (which is equipped with coordinates uo, Ug-1 Vo vg-,, a2g+1 -
agl.
- -
, 7 .... , ...
,
There is however better way to look at this larger system. Notice that in the way in
a
which we have presented it, the g + I coefficients of
W(A) = A9+2 + W g+,Ag+" + WgAg + - - -

+ WJA + Wo
are computed from (4.8); of course they will be polynomials in the ui, vi and ai. One sees
however that by inverting these relations the ai can also be written as polynomials, but now
in terms of ui, vi and wi. The upshot is that we can look at this system as defined on C3g+2
163
with coordinates juo,...' ug-li'vol ...

7Vq-1iWO)Wli ...
, w,+, I and the Lax operator A(A) is
now an arbitrary element
V(A) U(A)
( W(A) -V(A)
where
deg u(A) =
g, u(A) monic,
deg v(A) < g -
1,
degw(A)=g+2, w(A)monic.
A family of Poisson brackets, similar to one for the odd Mumford system, is given, for any
polynomial W of degree at most g by
IUW' UM11 =
NX), VM11 =
0,
U(X)W(Y) -
U(Y)V(X)
1U(X)'V(Y)l1 =
X-Y
V (T) W(Y) -
V (Y) W (X)
ju (x), w (y) 1W = -2
X -
Y
W(X)V(Y) -
WM O(X)
1V(X)'W(Y)l1 =
a(x + y)u(x)W(y),
X -
jw(x), w(y)J P =
2a(x + y) (v(x)V(y) -
v(y)W(x)).
where a(u) =u + w,+, u,-I. Rom these formulas one obtains, as in the case of the odd
-
Mumford system, easily Lax equations for the integrable vector fields X, I-, H(y)ll, as well =
as the vector fields Xi I-, Hill, where Hi is the coefficient of Xi in H(,\) u(,\)W(,\)+V2 (X).
= =
It is worthwile to point out that when the odd Mumford system is constructed in this way
then one arrives precisely at the Poisson structures 1.,Jw, given in (4.4). This means that
we have two different constructions for the multi-Hamiltonian structure of the odd Mumford
system.
4.3. Algebraic complete integrability and Laurent solutions
Having shown integrability of the odd and the even Mumford systems in Chapter III we
now comment on its algebraic complete integrability. It was shown by Mumford in [Mum5]
that the general level set of the odd Mumford system is an affine part of a Jacobian (of the
corresponding hyperelliptic curve) obtained by removing the theta divisor and that the flow
of the integrable vector fields is linear on it, hence leading to algebraic complete integrability.
In Section 5 we will modify our construction given in Chapter III in order to obtain similar
systems which axe a.c.i. Since this new construction coincides with the original one in the
hyperelliptic case it will give a proof that the odd and even Mumford systems are a-c-i- For
the odd case explicit solutions in terms of theta functions are given as follows (see [Mum5] for
details). Let D(t) Ab(D(t)) be an integral curve starting at D D(O) then D(t) At+D,
== = =
where A is a fixed vector which depends on the chosen vector field (going with t) and the
corresponding polynomial u(A) is computed from its values at the Weierstrass points ak of
the curve using
2
0 [J + 19k] (At + D)
U(ak) =
Ck
( 0 [9] (At + D) ) (4.9)
the vector 77k and 8 are characteristics which are described explicitly in [Mum5]. In view
of (IV.4.9) these define indeed meromorphic functions on the Jacobian.
164
For both cases the smooth level sets of the momentum map appear as the first level of
a stratification which we describe next (for more details and proofs see [VanQ.
Let r be a smooth(complete, irreducible) complex curve (say, a compact Riemann

surface) of genus g which ishyperelliptic. The hyperelliptic involution o- extends linearly
to Div(]P) giving an involution D -+ D'. We introduce a decomposition of Jac(IF) with
respect to an arbitrary fixed point P on the (hyperelliptic) curve r. Let E, denote the set
-T, =
1(m, n) E N x N 10 < m + n < gJ
which we order by (m, n) :5 (m', n) if and only if m < m' and n < n. Then for (m, n) E Ig
we define a subset Divmn (17, P) of Div(r) by
9-m-n
Divmn (17, P) Pi +mP+nP' -gP I Pi E r\JPPJ and i:A:j =* Pi =,I=Pj'
the term gP is introduced here in order to make every element in Divmn (r, P) of degree 0.
If we introduce20
g g-n
Divo (r, P) =
U U Divmn (r, P)
n=O m=O
then the Abel map Ab : Div'(r) -+ jac(r) : D -+ Ab(D) restricts to a bijection Ab

-+ Jac (17). It is shown in [Van4] that the sets
Divo (17, P)
Jm,n (r, P)t"T Ab(Divmn (r, P))
(or
Jm (r, P)t'e-f Ab (Divmo (r, P))
in P
case PI) define a stratification of Jac(r), meaning that they are disjoint differen-
=
tiable manifolds, whose boundary is a finite union of lower-dimensional sets i,,(r, P) (resp.
J,,(I`,P)). In the case P = 4'PO* the stratification is completely described by the following
proposition.
Proposition 4.1 If P =A P' then Jac(r) is stratified by the (g -

m -
n)-dimensional
submanifolds Jm,n Wi P), whose closure is given by the (finite) union
-7m,n (r, P) U ik,1 (r, P) (4.10)

(k,l) !(m,n)
Each stratum Jm,n (17, P) has two boundary components which are translates of each other by
P., P
,F =
Ab(Pa -
P) fp W11 ...
Ifp W9) (mod A).
20
Divo (r, P) is not to be confused with Divo (r), the group of divisor of degree zero on r,
of which it is a subset.
165
More generally, of dimension g i are translates of each other by n15 for some
all i + 1 strata -
n 11,... i}.
E ,
The closures
of the (g I)-dimensional strata ji, (r, P) and Jo,j (r, P) are
-
translates of the theta divisor and are tangent along their intersection jij (r, P).
Thus the different strata fit together as dictated by the partial order < on 1g: if we
represent the different spaces jm,n (171 P) by .7m,n i put those of equal dimension on the same
horizontal line and depict inclusions by axrows, then the stratification is schematically rep-
resented by the following.
JO, 0
40 jo, 1
-70,2
j, 0 J019
We also give the corresponding proposition for P PO".
Proposition 4.2 If P P' then Jac(r) is stratified by the (g

= -
m)-dimensional subman-
ifolds J,,, (r, P), whose closure is given by the (finite) union
im (r, P) U jk (r, P)
k>m
and each stratum j, (17, P) has just one boundary component.
In this case the stratification is simply depicted as
Jg -+ Jg-1 -+ Jg-2 -4 J1 A
jo =
Jac(r), j, is a translate of the theta divisor and ig is the origin in Jac(r).
The relevance of these stratifications for the evenand odd Mumford systems resides on
the fact that each stratum corresponds to precisely one family of Laurent solutions for one
special vector field of these systems (the most basic one). These Laurent solutions are given
by the following two propositions (for the relation to the Sato Grassmannian, see [Van4]).
Proposition4.3 For the odd Mumford system there are g+1 families of Laurent solutions.
The m-th family corresponds to the stratum J,(]P, P) and the functions ul,..., ug have the
following Laurent expansions starting at points of the stratum J,,, (17, P)
.(2i-l)!!(m+i)! 1
U
g-i
=
(- 1)
"
--
2"il t2i
+O(t-2i+l) (i M),
(M i)! -
(4.11)
Ug-i =
O(t -2i+l)
166
Proof
Equations (4.7) are written out in the case of the odd Mumford system (corresponding
to P =
PO') as
fi(x) =
2v(x),
,b(x) =
-w(x) + (x2u,-I)u(x),
-
ib(x) =
-2(x -
2u.-I)v(x),
or just as a third order equation,
; i (x) = 4 (fii- I -
2ug- I fii -
fig- I ui) (i =
0, - - -
, g -
1; u -
I =
0). (4.12)
The ansatZ21
00
Ui
t2i
uijti
j=O
leads for the leading coefficients ai u9-i,O to the recursion relation
2i + 1 [i(i + 1)
ai+1
j-+-1
+ [ 2
+ a,
1 ai. (4.13)
To solve this recursion relation, notice that if ai = 0 then ai+1 =

0; since ai 0 for at least
one i < g + 1, we find that
a, m(m + 1) (4.14)
2
for some m E 10,..., gJ which leads by induction immediately to the formula
(2i -
1)!! (m + i)!
aj (i M),
2ii! (M -
i)!
and a,,,+, a. 0, hence also to (4.11). The series for vi and wi follow immediately
=
from it by differentiation, in particular they do not give rise to separate families of Laurent
solutions. M
As a corollary of the proposition we see that the odd Mumford system induces a strati-
fication on Jac (17) which coincides with the stratification by the subsets Jm (17, P).
In the case of the even Mumford system we have the following result (for a proof
see [Van4)).
Proposition 4.4 For the even Mumford system there are (g+l)(g+2) 2 families of Laurent
solutions, one for each element of the set -Eq. The (m, n) -th family corresponds to the stratum
Jm,n(r, P) and the functions uo,..., ug-1 have the following Laurent expansion in t:
m-n
ug-1
t (4.15)
ug-i 0(0), (i =
2, .
..' g).
In particular, the even Mumford system induces a stratification on Jac(r) which coincides
with the stratification by the subsets J,,,,(I7, P).
It can be shown that this gives all Laurent solutions, see [Van4].
167
5. The general case
We will now modify our construction of Chapter III to construct for a large class of
curves an a.c.i. system whose general fiber of the momentum map is the affine part of the
Jacobian of a deformation of this curve. This new construction will coincide with the previous
one in one case (called the hyperelliptic case), considered in Section 4.
We start with a smoothplane curve rp C C', defined by an equation F(x, y) 0. Let g

denote the genus of the smooth completion Pp of ]Pp, which we assume to be non-zero. Each
holomorphic differential w on Vp can be written as
R(x,y)
!2y- UX)
(X, Y)
'9Y
for some polynomial R(x, y), hence the choice of a basis of the space of holomorphic differ-
entials leads to g polynomials Ro (x, y), , R,- iL (x, y). Having fixed such a basis
- - - we define
for any c (co,...' c,-,) c C9 a polynomial F,. with corresponding curve rp, by
=
g-1
F,(x, y) =
F(x, y) + E ckRk (Xi Y) - (5.1)
k=O
The following will in one of the statements below be assumed on the curve l7p:
Assumption For a general point c E C9, a basis of the space of holomorphic differentials
on r,, is given by
Rk(XiY)
dx (k =
0,..., g -
1).
I
0 V (X, Y)
This assumption, which is easily checked for any concrete curve at hand, is obviously valid
forhyperelliptic, trigonal, say n-gonal curves. For curves with a bad singularity at infinity
however, the condition may fail; an example of such a curve was kindly communicated to us
by H. Kn6rrer and will be given below (Example 5.2).
by constructing the affine Poisson vaxiety on which our a.c.i. system will live;
We staxt
closely related to the one in Paragraph 111.2.2 to which we will refer several
the construction is
times. We suppose that rp and the polynomials have been fixed (leaving aside the assumption
at the moment). For simplicity we take the standard Poisson structure jyj, xj I Jjj on C9 =
(which corresponds to W I in Paragraph 111.2.2); also we take d equal to the genus g of the
=
curve r, Let f h E O(Cg) be defined by

,
f =
JI(X, Xj)2, _
i<j
2
h = det ( R. (Xj , Yj ))I<i,j<.
and let A' denote the union of their zero divisors, A' =
(f ) U (h) =
(f h). Notice that (h)
depends on F only (and not on the choice of Rk) so we will denote it by AF. Thus A'
168
5. The general case
A U AF. Proposition 11.2.35 leads to an affine Poisson variety (MI, and a Poisson
morphism M, -+ 09 \ A'. Explicitly MI is given by
MI (XO' X0, (XI, YO'... (X q, Y9

, X0 II(X, _
X j)2 = x
0
det' (R, (xj, yj)) = 1
i<j
In view of the squares in the definition of f and h the symmetric group S, defines a Poisson
action on M, leading to a quotient (M2 I i
*
i
*
1 2) 1
M2 =
I (t, t, u (A), v (A)) I t disc u (A) = t'R (u, v) =
1}.
Here we computed22 det2 (R, (xj, yj))

have in terms of the ui and vi and called the resulting
polynomial R(u, v). Finally we define
MF21 =
I (t', u (A), v (A)) I t'R(u, v) = I
which is the affine variety we were aiming at. This variety is determined up to isomorphism
by F (it does not depend on the choice of basis Ri so we use a subscript F instead of R). As
in Paragraph 111.2.2 we get a commutative diagram,
MI ------------
0' M2
PI1 IP2
(C2) \ A/ ,
S
M.F29
in which S is (a restriction of) the map (111.2.3) and all maps turn out to be Poisson maps:
as in Paragraph 111.2.2 the Poisson structure J* *12 on M2 descends to a
1
Poisson structure
on MF29, also denoted by J* '12- Apart
1
from the verification that the brackets of the ui and
the vi do not depend on t, in the present case one also has to verify that the same is true for
the brackets of these with t'. However, by Proposition 11.2.35 the latter are given by
jui't'll =
_t/21U, R(u, v)},,
IV,, till =
_t/2fV, R(u, v)jl,
so they are independent of t. For hyperelliptic curves f = h hence A' = A = A.F and the
present construction reduces to the construction which was given in Paragraph 111.2.2.
We now adapt our construction given in Section 111.2.3. The natural map fIFd is now
replaced by the map23

ftF : (C2) 2g Cg
22
effectively compute R(u, v) for a concrete example one first replaces Vi by
In order to
v(xi) symmetric polynomial in xl,..., x.. This polynomial is easily rewritten in

to obtain a
terms of the elementary symmetric functions ul,. Ug. ..'
23
We do not add the dimension as a subscript here, because it is implicit in F, namely
d =
g is the genus Of rF-
169
whose components A,-, are defined by requiring that the polynomial
g-1
j>1i(O11y1)7 ...
, (xg yg)) R., (x, y)
, (5.2)
i=O
has for (x, y) (xj, Vj) the value F(xj, yj), where j 1,
=
g. Solving for the f1i involves
=
-
only the determinant of the matrix with elements Ri (xj, yj), hence we arrive at a regular
morphism Hp : M -4 C9 which makes the following diagram commutative:
M, M2
PiI JP2
(C2)9 \ (A U A/) M29
F
S
H1,
C9
Although the components of Hp depend on the choice of R, the algebra AF generated by

these is clearly independent of it. It leads to an integrable Hamiltonian system (which is an
a.c.i. system if the assumption is satified) by the following proposition (for a different proof
of integrability see [Van5]).
Proposition 5.1 Let F(x,y) E C[x,y] be such that F(x,y) = 0 defines a non-singular
Curve rF C C2 of positive genus g. Then

(M.29, 1.
F '12, AF) defines an integrable Hamil-
tonian system whose level set over 0 is isomorphic to an affine part of the Jacobian of 17,v
and the flow of the restriction to this level of all integrable vector fields is linearized by this
isomorphism.
When the above assumption about I7p is satisfied then this integrable Hamiltonian system
is a.c.i. system whose general level set over c E C9 is isomorphic to
an an affine part of the
Jacobian of the deformation rF,, defined by (5.1) of the curve rF-
Proof
Since S is a Poisson map it suffices to show that the components of ftF are in involution.
Clearly IF(xi, yi), F(xj Yj)jh 7
= 0 for all i, j. From. the definition of the components Ak of
fl.p we have
IfloRo (xi, yi) +... +A,-,R,-,(xi,yi),floRo(xj,yj)+---+A,-iLR,-I(xj,yj)},=O. (5-3)

Since the second component depends only on (xj, yj) (although each individual Hi depends on
all (xi, y,)) it is in involution with all coefficients of R1, (xi, yi) and (5.3) can be rewritten
as
g-1
E IfIk, Al JRk (Xi, yi)RI (xj, yj) = 0.

k,1=0
170
5. The general case
Since the polynomials Rk are independent it follows that jAk, Al I 0 for all k and 1. =
Thus the algebra Ap generated by H1,...' H, is involutive. It has dimension g by the same
argument as used in the proof of Proposition 111.2.3, hence Ap C[HI, H,]. As for =
...'
completeness of A, notice that choosing a closed point in the spectrum Of AF Consists in

fixing the values of the Hi, hence the level sets (in MF29) axe given by
g-1
F(A, v(A)) -
ciRi (A, v(A)) = 0 mod u(A),

i=O
where the ci are these fixed values. Denoting the polynomial on the left hand side by F,
we see that they correspond to the level sets FF. described by Proposition 111.3.3 and
,, .
Lemma 111.3.4. More precisely they axe obtained from these by removing a divisor. In
particular the general level set is still irreducible and every level set has dimension g. This
completes the proof that A.F is an integrable algebra, hence that (M2g, 1.
F 12 1 AF) defines
an integrable Hamiltonian system.
In order to say more about the general level sets we restrict ourselves from now on to
those curves IPF for which the assumption (announced at the beginning of this paragraph) is
valid. If not then our argument is still valid for the fiber over 0 (i.e., the one which corresponds
to Ho = ... =
H,_1 =
0).
The main observation to be made here is that the restriction Ap, of the divisor 16kF Of
h to ageneral fiber c of the momentum map (which is an (affine part of) a g-fold symmetric
product of the curve l7pj is precisely the divisor which is blown down by the Abel-Jacobi
map in order to construct the Jacobi variety of IPF,, from the g-fold symmetric product of rF,,.
To see this, notice that the points of AF, correspond exactly to those divisors of degree g on
IPF, for which the matrix R,(xj, yj) becomes singular, meaning that the space of holomorphic
differentials with zeros at these points has positive dimension, hence (by Riemann-Roch) the
dimension of the linear system defined by this divisor has positive dimension and this whole
linear system is collapsed to a point by the Abel-Jacobi map. Thus, removing the zero divisor
of (h) from C2g, which led precisely to our phase space MF29, has the effect of removing from
each fiber the divisor which is blown down by the Abel-Jacobi map.
Recall however that we had already removed a divisor (with several irreducible compo-
nents) symmetric product which included at least one component which maps to a
from the
translate of the theta divisor (in the notation of Paragraph 111.3.4 the latter components are
the -6Fd(OOk))- Clearly each of the divisors in A. is linear equivalent to an effective divisor
which contains ook, hence is collapsed to a point in the image of SFd(ook) (under the Abel-
Jacobi map); since this is true for all k the image of A. is contained in the intersection of
the translates of the theta divisor which were all missing already in our original affine part
2
of the syrmnetric product. It follows that our level set in MF'9 is now isomorphic to an affine
part of the Jacobian of I7p.. If the assumption on the curve r F is not satisfied then we still
have that the fiber over 0 is isomorphic to all affine part of the Jacobian of rp (this property
may still hold for some other fibers).
We have checked now (under the assumption) that the general fiber is an affine part of an
Abelian variety; we may take base N the space of all curves ]FF,, which axe non-singular
as a
and construct 7- as the universal Jacobian over it (replacing every point by the corresponding
Jacobian), giving us the necessary ingredients for the proof that we have an a.c.i. system.
171
Only linearity of the flow on the Jacobians remains to be checked. By construction we have
ata general point
g-1
F(xj, yj) 1: AiRi (xj, yj),

i=O
where f1i are the components of Ap. Taking the bracket with xj, and recalling that
jyj, xj Ig =
Jij,we have
g-1 g-1
c9F "
57- (xj, yj) Jjk

Y
E Ri (xj, yj) X.#, xk + E Ai (Xj, Yj)Jjk-
i=O i=O
Restricted to the invariant manifolds Aj =

ci we have
9-1
i=O
Rt(Xj) Yj)Xftj Xk
-5Y (XjYj)Jjki
g-1
Ri (xj, yj)
2F Xf-ii Xk -Jjki
i=0 ev (Xj,
,
Yj)
which is easily rewritten as
g
Rj (xi, yi) X-
OF Hk
X-6'k-
ey
(X,, Y&
Since rp satisfies the assumption, we have on the left exactly a basis of the space of holo-
morphic differentials rF. so we find that the vector fields X.#. (hence also the vector
on
fields XHJ linearize under the Abel-Jacobi map, i.e., their flow is Unear on the Jacobian of
1P.F,. This shows that we have an a.c.i. system. Without the assumption we still have linearity
of the flow on the Jacobian of IP.F (the invariant manifold over 0).
Let us also show that these a.c.i. systems satisfy the Adler-van Moerbeke condition
for algebraic complete integrability (which we did not impose in our definition of an a.c.i.
system). That is, we show that along each of the components of the divisor which is missing
from Jac(:PF,j at least one of the functions ui, vi, t' has a pole. Notice that ui and vi (when
restricted to rF..) are given as symmetric functions on rF,, hence giving indeed meromorphic
functions on Jac(:rF. In the notation of 111-3.4 the missing components are the images
under the Abel-Jacobi map of the divisors DF, and gpg. As for the former, the functions vi
obviously all have a pole along it, for the latter, on a component 9F,,(OOk) where X(000 is
infinite, all ui have a pole, and if Y(000 is infinite, then at least vg-1 has a pole along it. m
Example 5.2 Here is Kn6rrer's example which shows that the above assumption is not
valid for all curves. Notice that the assumption implies that the genus of the curves 17F. is
constant and equal to the genus Of IPF (for generic c); in the present example the genus of
the general curves will be higher than the genus of ]Pp providing the counterexample. Take
F(x,,U) = x
8
Y2+ Y 10
172
5. The general case
and R, (x,y) X6y. It is a curve of degree 10 whose closure X8y2 +y1O ZIO 0 has (I : 0 : 0)
= _ =
as only singulax point. It is checked that this point is a cusp whose tangent has a contact
its
of order 10 to the curve. It follows that the genus of the smooth completion of the curve
(10-1)(10-2) 5 31. If however one adds a multiple of R, to the curve, which gives
equals 2
- =
RjX, Y) =
X8Y2 + y1O + CX6Y
then for general c E C it still has singularity at infinity but it is now a cusp whose
only one
tangent has a contact of order 6 to the curve, giving 33 as its genus. Thus for general c the
genus of rp,, is bigger than the genus of 1F.F; if other holomorphic differentials are taken into
account then it can of course only get worse. Thus I7p does not satisfy the assumption.
173
Chapter VII
Two-dimensional a.c.i. systems
and applications
1. Introduction
The last chapter is devoted to the study of several two-dimensional integrable Hainilto-
nian system which are of interest, especially from the point of view of algebraic geometry.
We specialize the odd and even Mumford systems to the case of genus 2 in Section 2.
We give explicit equations for these systems, which will be useful when studying the other
examples, which will all be shown to be related to the genus 2 Mumford systems. An appli-
cation of the (genus two) even Mumford system was worked out in collaboration with JoS6
Bertin (see [BV2]) and is given in Section 3. We consider an arbitrary curve of genus two
which has anautomorphism of order three. This automorphism can be extended to its Ja,
cobian and leads to a singular quotient, similax to the classical Kummer surface. We exhibit
a 94 configuration on this quotient and show that it is a complete intersection of a quadric
and cubic hypersurface in p4. By using the even Mumford system we explicitly compute
equations for the quadric and cubic hypersurface, thereby giving a projective realization of
these generalized Kummer surfaces.
We study in Section 4 an integrable quartic potential on the plane, which was discovered
by Garnier as very special case of a laxge family of integrable Hamiltonian systems which
a
he derived from the Schlesinger equations (see [Gar]). In Paragraph 4.1 we will show in two
different ways that the general fiber of its momentum map is an affine paxt of an Abelian
surface of type (1, 4). One method uses some of their specific geometry, as given in the
beautiful paper [BLS], the other one is based on a morphism to the odd Mumford system.
It follows that the Gaxnier potential is an a.c.i. system of type (1, 4) Moreover the morphism
to the odd Mumford system leads to a Lax representation of the Gamier potential. In
Paragraph 4.5 we consider the limiting case in which the potential is a central potential.
175

Chapter VIL Two-dimensional a.c.1. systems
Then the general fiber of the momentum map is not aai affine part of an Abelian variety but
is a C* bundle over an elliptic curve.
a (1, 4)-polarization there is associated a rational map

To the line bundle which defines
to p3(we assume here that the Abelian surface is not a product of elliptic curves). The image
of this map was shown in [BLS] to be an octic and its coefficients parametrize a 24: 1 cover of
the moduli space A(1,4) of (1, 4)-polaxized Abelian surfaces. The octic is an unramified cover
of the Kummer surface of a Jacobian which is canonically associated to the (1, 4)-polarized
Abelian surface. One way to characterize this Jacobian is that it is the unique Jacobian J
for which the map 2j (multiplication by 2) can be factorized via the Abelian surface. We
will give in Paragraph 4.3 some other characterizations of this Jacobian. It should also be
remarked that the morphism to the odd Mumford system precisely maps the general level
set, which is affine part of an Abelian surface of type (1,4) to its canonical Jacobian, in
an
particular we can use this morphism to explicitly describe the natural map between A(1,4) and
the moduli space of Riemann surfaces of genus two. The result is surprisingly simple; looked
at in another way it shows how to write down explicitly for a curve (given by its equation)
the equation of the Kummer surface of its Jacobian in the classical Rosenhain form.
We also get an explicit description of the moduli space A(1,4) It is obtained as a quotient
-
of the parameter space which is given by the coefficients of the octic. The group which is
acting on it has order 24 and is isomorphic to Z/4Z x 63; the action was suggested by an
automorphism of the system. The final result is that A(1,4) is birational to a cone M3 in
weighted projective space p(1,2,2,3,4) for which we give explicit equations.
Another a.c.i. system that is naturally related to Abelian surfaces of type (1, 4) is the
in
geodesic flow on SO(4) corresponding to some special metric, as was first pointed out
[BV4]. In this case, studied in Section 5, the general fiber of the momentum map is an affine
part of a Jacobian, but there is a group of translations acting on this fiber, so
hyperelliptic.
to this fiber Abelian surface of type (1, 4). The resulting
that there is naturally associated an
map from the base space to

the cone M3 in p(1,2,2,3,4) will be explicitly computed. See [BV4]
for its relation to the moduli problem for a.c.i. systems.
We look in Section 6 at a hierarchy of integrable potentials V,, on the plane. Their

is
easily checked but revealing the geometry of the general level sets of the mo-
integrability
mentum map requires more work. For the first non-trivial member V3 we define a morphism
to the odd Mumford system and deduce from it that V3 is algebraic completely integrable
and we identify the general fiber of the momentum map as an Abelian surface of polarization
Painlev6 analysis by Adler and van
type (1,2) (this result was previously established by using
Moerbeke, see [AM9]). For V4 there is a similar morphism to the even Mumford system and
we show that its general fiber is a 2 : 1 cover
of a hyperelliptic Jacobian, which is ramified
along two touching translates of the theta divisor; this result could also be obtained by using
a morphism to the Ber-hlivanidis-van

Moerbeke system, which is discussed in Paragraph 2.3.
to show that the level sets 2 -
1 unrarnified covers of a certain
For V,, we use a morphism axe
n+L
two-dimensional stratum in the Jacobian of hyperelliptic curve of genus [ 2 1. We also
a
exhibit Lax equations (with a spectral parameter) for the hierarchy.
In the last section we reconsider the periodic three body Toda lattice. We use again
The order three
a morphism to give a new proof of its algebraic complete integrability.
to an a.c.i.
automorphism which comes from a cyclic permutation of the three particles leads
concrete realization of this quotient system.
system of type (1,3) and we give a
176
2. The genus two Mumford systems
2.1. The genus two odd Mumford system
For future use we now give explicit formulas for the genus 2 Mumford systems. Before
doing this, let us restrict the odd and even Mumford systems to a Casimir which is common
to all the Poisson structures I., JW, where deg W :5 2, without restricting the class of
(hy-
perelliptic) curves that appear as spectral curves for these systems. Namely, we consider for
the odd (resp. even) Mumford system the hyperplane given by 0 (resp. H2,+I
H2, 0). = =
Since this fixing a level of a Casimir all formulas are easily transcribed: for
corresponds to
the odd (resp. even) case, simply substitute
w. -+ ug-I (resp. wg+l -+ ug-1) in all formulas.
We start with the odd case. Then the Lax operator is given by
VIA + Vo A2 + UI A + Uo
A(A) =
( A3 _
UI'X2 + WI'X + Wo -VIA -
Vo ) (2.1)
and the Lax equations for the Hamiltonian vector fields X, and X2 are given by XH,A(A)
[A (A), Bi (A)], where
0 1
BI (A) =
(A-2u, 0
Bo (A) =
('\2 u
v1
1A+WI -Uo \+Ul)
-VI
If we denote derivation in the direction of the first vector field by a dot and in the direction
of the zeroth vector field by a prime, these are written out as
,&(A) =
2v(A), u'(A) =
2v(,\) (A + ul) -
2v,u(A),
b(A) =
(A -
2u,)u(A) -
w(A), V/ (A) = U (,\) ('\2 _
U1 'X + W, -
Uo) -
(A + UIMA),
zb(A) =
-2v(A)(A -
2u,), w'(A) =
2w(A)vI -
2v(A) (,\2 _
UIA + W1 _
Uo).
In order to write down explicitly the compatible Poisson structures it is useful to have these
even further written out, namely in the following form:
fil 2vI, ul 2vo,

fio 2vo, uO 2(uIvo -
uovj),
2
I =
uo -
2u _
WI, v' =
-2u,uo -
wo,
o =
-2uuo -
wo, vo =
_UIwo + Uo (WI -
UO),
tbi =
4uIvI -
2vo, wl =
2(uIvo + uovI),
tbo =
4uvo, wl0 =
2(uovo + vIwo -
vow,).
Recall also that the corresponding integrable algebra A is a polynomial algebra generated by
the coefficients Of V2 (A) + U(A)W(,X), i.e., by
2
H3 =
uo + w, -
ul,
2
H2 =
-uluo + uIwI + wo + v, ,
(2.2)
HI =
ulwo + uow, + 2vIvo,
H0 =
UoWo + V201
177
Chapter VIL Two-dimensional ax.i. systems
(recall that we 0). We now explicit for this system the formulas for the three
have put H4 =
compatible I., -J'P, where W

Poisson structures 1, A"\2 Their Poisson matrices are easily
=
.
computed from (VI-4-4) in terms of the system of generators ul, uo, vi, vo, wi, wo (in that
order). Notice that since S4 is not a Casimir of I-, -IA' the latter Poisson structure does not
restrict to the level S4 = 0. The Poisson structure -I' is given by
0 0 0 1 0 0
0 0 1 U1 0 -2vi
0 -1 0 0 1 -2u,
0
(2.3)
-1 -U, 0 -Ul W1 -
UO
0 0 -1 U1 0 2v,
0 2vi 2% uO -
w, -2vj 0
It is easy to compute that H2 and H3 are Casimirs and that H, and HO give the first and
zeroth vector fields. Notice also that this structure is affine (i.e., it is a modified Lie-Poisson
structure, see Example 11.2.14). The Poisson matrix of the (affine) structure is given
by
0 0 1 0 0 0
0 0 0 -UO 0 2vo
-1 0 0 0 -2ui 0
(2.4)
0 UO 0 0 -UO -WO
0 0 2u, UO 0 -2vo
0 -2vo 0 WO 2vo 0
In this case HO and H3 are the Casimirs and H, and H2 give the two vector fields. Finally
%2
the Poisson matrix of is given by
0 0 -U, -UO 2vi 2vo

0 0 -UO 0 2vo 0
U1 UO 0 0 -WI -WO
0
(2.5)
UO 0 0 0 -WO
-2v, -2vo W, WO 0 0
\ -2vo 0 WO 0 0 0 )
In this case HO and H, are Casimirs and H2 and H3 give the two vector fields. Notice
that this structure is even linear. The Hamiltonian vector fields which correspond to the
polynomials HO, . . .
, H3 using these structures are summarized in the following table.
I., Hjw H3 H2 Hi HO
0 0 X, X0
0 X, X0 0
1. j'\2 XI X0 0 0
Table 3
178
2.2. The genus two even Mumford system
We nowgive the corresponding formulas for the genus 2 even Mumford system. The
Poisson structures I., .1w, which we computed from Proposition 111.2.1. These structures are
neither linear nor affine. The Lax operator is given by
VIA A2
A(A) =
( A4 _
UIA3 + W2A2
+ VO
+ WIA + Wo
+ UIA + UO
-VIA VO - )
and the Lax equations for the Hamiltonian vector fields X, and Xo are given by XHA(A)
[A(A), Bi (A)], where
0 1
Bi (A) =
2
A2 -2uA+2u,-UO+W2 0 ) ,
and
V1 A + U,
Bo(A) A3 -
UIA2 + (W2 -
uo)A + 2u,uo + w, -VI
They are written out as vector fields on C7 as
fil =
2vI, U1I =
2vo,
ito =
2vo, uO =
2(uIvo -
uovI),
i), = 2u3+
I UIW2 -
3uuo -
wi, V,I = 2u 12UO -

U2+ UOW2 -
WOi
0
,bo = 2u2UO
1 -U 0 2+ UOW2 -
WOi V
I
0 =uow, + u, (2U20 _
WO),
?b2 =
2(2uIvI -
vo), W,2 =
2(uIvo uovI),
+
?bI =
2(-2u2v,
I + 2uIvo -
VIW2 + VIUO)i W,I =

-2(2uuov, + VO(W2 -
UO)),
2
?bo =
-2vo(2u 1
_
UO + W2)5 W,0 =
2(vlwo -
vow, -
2u,uovo).
The corresponding integrable algebra A is now the polynomial algebra generated by
2
H4 =
W2 -
U1 + UO,
H3 =
WI -
U1UO + UIW27
2
H2 =
UOW2 + UIWI + WO + V17 (2.6)
H, =
ulwo + uow, + 2vIvo,
2
Ho =
uowo + vo,
We also write down the three compatible Poisson structures for the genus 2 even
Mumford system (for W 1, A, A2); their Poisson matrices are now written in terms of the
=
system of generators U1, U07 VIi VOi W21 Wli WO (in that order). Here is the Poisson matrix
of 1.,jI
0 0 0 1 0 0 0
0 0 1 U1 0 0 -2v,
0 -1 0 0 2
1 -2u, W2 -
uo + 2u,
-
1 -Ul. 0 0 -UI W2 -
UO w, + 2u,uo (2.7)
0 0 -1 U1 0 0 2v,
0 0 2u, UO -W2 0 0 -4uiv,
0 2vI uo -
2U21 -W2 -2uuo -
w, -2v, 4uIvj 0
179
Chapter VIL Two-dimensional a.c.i. systems
It has H2, H3 and H4 as Casimirs and H, and HO give the first and zeroth vector fields. The
Poisson matrix of is given by
0 0 1 0 0 0 0
0 0 0 -UO 0 0 2vo
-1 0 0 0 -2u, W2 -
UO + 2U21 0
0 UO 0 0 -UO 2ujuO -WO - (2.8)
0 0 2ul UO 0 0 -2vo
0 0 UO -
W2 -
2u 2I -2uuo 0 0 4u,vo
\ 0 -2vo 0 WO 2vo -4u,vo 0
This one has HO, H3 and H4 as Casimirs and H, and H2 give the two vector fields. A third
structure is given by
0 0 -Ul -UO 0 2v, 2vo
0 0 -UO 0 0 2vo 0
U1 UO 0 0 2U2I _
UO -WI -WO
UO 0 0 0 2uuo -WO 0 (2.9)

0 0 uO - 2U21 -2u,uo 0 -2vo + 4u,vl 4uvo
-2v, -2vo W1 WO 2vo 4u,vl
-
0 0
-2vo 0 WO 0 -4u,vo 0 0
In this case HO, H, and H4 axe Casimirs and H2 and H3 give the two vector fields. Finally, a
fourth structe, with H2, H3 and H4 as Casimirs, H, and H2 givng the two commuting vector
fields, is given by the following (in order to make this anti-symmetric fit on this page we only
print its upper triangular part)
2
0 0 U, -
UO UOU1 2v, 2(vo -
ulvi) -2u,vo
0 UOU1 U20 2vo -2uovl -2uovo
0 0UIW2 -
W1 U1W1 -
WO UIWO
0UOW2 -
WO UOWj UOWO (2.10)

0 2vjW2 2vOW2
0 2(vow, -
vlwo)
0
As above we summarize the Hamiltonian vector fields which correspond to HO, H4 upon
using these a table.

structures in
I., Hjjw H4 H3 H2 H, HO
I ,11 0 0 0 X, X0
1.'.Y\ 0 0 X, X0 0
\2 0 0
j.'j 0 X, X0
1-'-11'3 X, X0 0 0 0
Table 4
It can be checked by direct computation that these brackets also verify (VI.3.7), but
g1q) is
a conceptual proof (given for the odd Mumford systems through the loop algebra
missing.
180
2.3. The Bechlivanidis-van Moerbeke system
In this paragraph we wish to consider an integrable Hamiltonian system on C' which was
constructed by Bechlivanidis and van Moerbeke (see [BNq) in order to understand the geom-
etry of the Goryachev-Chaplygin top. This system is sometimes called the s even- dimensional
system, but since the dimension of an integrable Hamiltonian system was defined as the di-
mension of the general fiber of the momentum map, which is two in *this
case, we prefer not to
call it this way; we prefer to call it the Bechlivanidis-van Moerbeke system. It was constructed
as a pair of commuting vector fields together with five independent functions whose deriva-
tives in the direction of these vector fields was zero; this was done without making reference
to any Poisson structure for which the equations of motion were Hamiltonian. Several such
Poisson structures (which are compatible) were constructed for it by us in [Vanl]. Not only
will these be reproduced here but we will show that the Bechlivanidis-van Moerbeke system is
isomorphic to the genus 2 even Mumford system, thereby proving its algebraic complete inte-
grability (a proof of this has been given nowhere) and "explaining" where the bi-Hamiltonian
structure of this system comes from. There is for this system a third (compatible) structure
which is found in anatural way; it is closely related to the Poisson structure of the Toda
lattice discussed in the last section of this chapter; of course this structure can also be trans-
ported to the even Mumford system, thereby leading to another compatible structure for that
system.
We pick coordinates sl, 87 on C7 and consider the following algebra of functions,
A =
C[Sli S2, S3i S4, 651,
where
2
S, =
si -
432 -
8S41
S2 =
5132 + 4s6,
S3 =,92
4
_ S25 + 531
S4 S486 + 3537 + 52831

2
S5 -86 + 82-SIS3-
7
Also consider the Poisson matrix (18il 8jD,<ij<81 given by
0 0 -16s5 0 -8 0 16S2
0 0 0 0 0 0 4
16s5 0 0 2s5 2S4 -4S2S5 4S2S4
0 0 -2S5 0 -1 0 -2S2 (2.12)
8 0 -2S4 1 0 -2S2 0
0 2
0 4S2S5 0 2S2 0 -4S2 _
S,
-16S2 -4 -4S2S4 2S2 0 4S22 + S, 0
Although we do not suggest to the reader to do this (because it is long), it can be checked
by
direct computation that this matrix defines indeed a Poisson structure on C7; we will check
it in another way. What is however checked at once is that S1, S2 and S3 are Casimirs for
181
this structure and that S4 and S.5 are in involution; they lead to the vector fields (up to a
constant)
bi =
-8S7, si =
8(sIS5 + 2S2S7)j
b2 =
4s5, s'2 =
4s7,
h =
2(S4S7 + 506)1 S3 =
4(S2S586 + 82S4S7 -
2s3s5)7
h =
-4S2S5 -
S77 S14 =
s1s5 -
2SO7,
b5 =
S6 -
48234, S15 =
S134 + 232S6 -
4S3,
2
b6 =
-,5ls5 + 2S2S77 S16 =
-SIS7 -
2slS2S5 -
452S77
2
b7 =
S154 + 2S2S6 -
483. 817 =
882S3 + U01284 -
4s 2 S6 -
8186-
Before writing down other Poisson structures, let us prove that the Bechlivanidis-van Moer-
beke system is isomorphic to the even Mumford system by giving explicitly the isomorphism;
this map was constructed in [Van2] as an illustration of the algorithm (recalled in Section VA)
which was developed there. The isomorphism is given by the map
0: C7 _+ C7 : (Sl S7) '-+ (U (A) ,

V (A) 2
W (A))
where
U(,X) = A2 + 2S2A + Sli
v(A) =
8s5A -
8S7,
W(A) = A4 -
2821\3 + (SI -
16S4 -
4S2),\2
2
-
4(-92(Sl -
8S4 -
2S2)
2 + 4s6)A
+ 4(s, '922 + 85256 -
1653)-
Of course this map is regular, but it is even biregular: S1, 82, s5 and S7 are of course regular
in terms of the coefficients of u(A) and v(A); given this S4 is regular in terms of these and
the coefficient of 'X2 in w(A); the same follows for S6 and S3 upon using respectively the
coefficients in A and XO in w(X). Next it is easy to check that
uo) 2(s, 4s 2-
2
0*(W2 -
U1 + =
2 884), -
0* (WI -
UIUO + UlW2) -4(.5182 + 4s6),=
0*(UIWI + UOW2 + Wo + V2) =

-64(324 _ '925 +S3)+(SI -4 S2
2-8S4 )2,
1
0*(ulwo + uow, + 2vivo) =

-128(54,56 + S587 + 52S3) -
4(s, -
4 S22 -
84)(SIS2 + 4S6)7
2
0* (UOWO + V2)
0
=64 (8 7 _
'926 _
$183) + 4(s182 + 486 )2.

(2.13)
Thus 0* A! A, where A! denotes here the integrable aJgebra of the (genus 2) even Mumford
=
system. Since 0 is a biregular map the fact that A! is integrable for some Poisson structure,
O*A is integrable for the corresponding Poisson structure; moreover since A! is a.c.i. the same
holds true for A. The Poisson structure which corresponds to (2.9) is given (up to a constant)
by (2.12) hence (2.12) defines a Poisson structure and 0 is a Poisson map with respect to
the Poisson structures (2.9) and (2.12). Since 0 is biregular we can also transport the other
182
Poisson structures. If we transport the Poisson structure given by (2.8) then we get (up to a
constant) the following Poisson matrix:
0 0 -1657 0 0 0 -8s,
0 0 0 0 -4 0 0
1637 0 0 2S7 -2,36 -432-97 8S3 -
48256
0 0 -2S7 0 4S2 0 -SI
0 4 286 -432 0 -Sl 0
0 0 48237 0 31 0 2s,82
8s, 0 4S286 8S3
-
81 0 -2s,S2 0
In this case S1, S2 and S5 are Casimirs and S3 and S4 generate (up to a constant) the vector
fields X, and X2; this can be checked directly or by using (2.13). One might think now that
the structure (2.7) will give a Poisson bracket for which S1, S4 and S5 are Casimirs but S2
and S3 generate the two vector fields, but this is not the case. To see this, let us rewrite (2.13)
as follows:
O*H4 =
2S1,
O*H3 =
-4S2,
O*H2 =
S12 -
64S3,
0*H1 = -128S4 -
4SIS2,
0*H0 = 6485 + 4S22.
Since 0 is a isomorphism it maps Casimirs to Casimirs. Thus, if H2, H3 and H4
Poisson
are Casimirs then S.1, S2 and S3 are Casimirs for the corresponding structure. Similarly, if
H0, H3 and H4 are Casimirs then S1, S2 and S5 are Casimirs for the corresponding structure.
However, if H0, H, and H4 are Casimirs then S1, 32S4 + S, S2 and 16S5 + S21 are Casimirs, so
that (up to a multiple) S2 and S3 generate X, but S4 and S5 give respectively SiX2 and S2X2-
It follows even that there is no linear combination of these three Poisson structures which
has S1, S4 and S,5 as Casimirs, while S2 and S3 give the vector fields X, and X2. However
one finds by trial and error easily that the following structure does this job:
0 0 -4(s586 + S487) 287 -2S6 2s,85 + 482s7 452S6 -

2s, 84
0 0 0 -S5 -54 37 36
4(s586 + 54,57) 0 0 0 0 -28385 2S3,94

-2s7 85 0 0 0 0 -S3
286 94 0 0 0 -S3 0
-4S237 2s,85
-
-S7 2S385 0 -53 0 -2-92,53

2sS4 48286
-
-S6 -2S384 -3 0 2S283 0

(2.14)
This one important because it will appeax later when studying the Toda lattice. It is
is
compatible with the previous ones in view of Theorem 14 in [Van2] which says that if in a
two-dimensional integrable Hamiltonian system all integrable vector fields axe Hamiltonian
with respect to two different Poisson brackets, then these brackets are compatible. Of course
it leads also to a fifth Hamiltonian structure for the genus 2 even Mumford system. For a
generalization of this special structure to all even Mumford systems, see [FVJ.
Using the isomorphism everything can be transported from the genus 2 even Mumford
system to the Bechlivanidis-van Moerbeke system. Since we have Lax equations for the even
Mumford system we also have them for this system. Similaxly we obtain all possible Laurent
183
Chapter VIL Two-dimensional ax.i. systems
solutions from those of the even Mumford system, and since all coefficients of the are
map
real we can use our knowledge about the topology of the real level sets of the even Mumford
system to describe the topology of these level sets for the BechlivaiAdis-van Moerbeke system.
Finally, in the next section we will use the even Mumford system to study a certain problem in
algebraic geometry; clearly we could use the isomorphic Bechlivanidis-van Moerbeke system
instead.
184
3. Application: generalized Kummer surfaces
We now give an application of integrable Hamiltonian systems to algebraic geometry,

which was worked out in collaboration with Jos6 Bertin (see [BV2]). For other applications
which use the same technique, see [PV2].
3.1. Genus two curves with an automorphism of order three
We consider a curve equipped with an automorphism of order three,

r of genus two,
denoted by By7% quotient I'l-r has genus zero and r has
the Riemann-Hurwitz formula the
four fixed points. Since r has genus two it is also hyperelliptic; as before the hyperelliptic
involution will be denoted by a. We have the following diagram
IF -ITL11- P1
7r,I
1
P
-r necessarily points to Weierstrass points, hence the commutator IT, a]

maps Weierstrass
fixes all these points and we see that ra o--r since the only automorphisms which fix all
=
Weierstrass points are o- and identity. It follows on the one hand that -r induces on P'
a fractional linear transformation -7 of order three, and on the other
hand that the four
fixed points of -r consist of two o--orbits. We may therefore suppose that - is given by
exp(!'-), by choosing coordinate x on P' such that these two orbits
-T(x) =
ex, c =
3
a
correspond to x 0 and x
= oo. The images of the Weierstrass points form two orbits of
=
three points under f, which correspond to the roots of the equation X3 A3 and X3 A-3, =
possibly after a rescaling of x. Obviously A :;:4 0; since both orbits are different, A3 A-3,
i.e., A6 =? 1. This shows that IP has an equation
Y2 =
(X
3
_
X3) (X 3 _
A-3)
3
(3.1)
X6 + 2nx + 1,
with r.:;4 1.
of genus
every equation of the form (3.1) with K: 1, defines a smooth curve
Clearly,
two with automorphism. (x, y) -+ (ex, y) of order three; also, if K in (3. 1) is replaced by -K
an
then an isomorphic curve is obtained. Conversely, let there be given two isomorphic curves
r and 17' with respective automorphisms -T and r' of order three. We may suppose that the
isomorphism. 0 : IP -+ r' respects the automorphism, i.e., Or r'o. We claim that if r, =
and 172 are written as above as
Y2 = X6 + 2nix3 + 1,
then nj2
2
K2. To see this, notice that 0
=
i bviously pmmutes with a hence there is an induced
linear transformation 0 which satisfies O(ex) =
eo(x), for all x E P'. Thus O(x) =
px and
O(x, y) =
(px, y), iving pl = 1. It follows that X2 A 1
"
= can be taken as modular parameter.
185
The
automorphism group of r contains a subgroup which is isomorphic to S3 x Z/2Z, as
is immediately from (3.1); it actually coincides with this group, unless K 0 (in which
seen =
case the group of automorphisms jumps to D6 x Z/2Z).

Namely, there is, apaxt from the
hyperelliptic involution a, an action of S3 by means of which the Weierstrass points belonging
to oner-orbit can be at random permuted. For future use we choose an element
p of order two
in thissymmetry group S3 corresponding to a transposition in S3, say P(X, V) (X-1, yX-3) =
and notice that it commutes with a but not with r. Its fixed points axe the two points in
7r-1111, hence IP/p is an elliptic curve.
We will find it convenient to denote the fixed points of -r, which mapped by
are 7r, to 0
(resp. oo) by ol and 02 (resp. oo, and 002). Then o" 027 000 002 and we may suppose = =
p(oi) oo, giving also /L(02)

=
002- In the same way we denote the Weierstrass points
=
corresponding to the x3 Akorbit by Ai, -r(Ai)

=
Ai+1 (indices are taken modulo 3) and the
=
3
ones corresponding to the x A-3 -orbit by i, p(Ai)
=
i. Then the action of S3 x Z/2Z =
on these points is contained in the following table.
order 01 02 001 002 Ai i
3 01 02 001 002 Ai+1 i_l

0' 2 02 01 002 001 Ai i
A 2 001 002 01 02 Xi Ai
Table 5
3.2. The 94 configuration on the Jacobian of 1P
Let J(r) denote the Jacobian of r and for a divisor D of degree 0, let Ab(D) denote the
corresponding point in J(r) as before. Recall that for any fixed Q1, Q2 E r, every element
w r. J(r) can be written as w Ab(Pi + P2 Q1=
Q2); moreover this representation is
- -
unique if and only if P, =7 P21, all P + PO' and Q + Q11 (P, Q E 1P) being linearly equivalent,
P + P" -1 Q + Q1. In the present case of curves (3. 1) which have an automorphism T of order
three, the cover -7r, associated to -r provides in addition (using the notations of the previous
section for the fixed points of -r) the following linear equivalences
3o, -1 302 -1 3ool -1 3002- (3.2)
The automorphism -r extends in a natural way to an automorphism on J(I`), also denoted

by -r. It is given and well-defined for w =
Ab(PI + P2 Q,
-
Q2) as follows: -r(w)

-
Ab(-r(PI) + -r(P2) -,r(Ql) -
7(Q2))-
Proposition3.1 The automorphism -r has nine fixed points and nine invariant theta curves
on J(r).
Proof
The principal polarisation on J(]P) is invariant under Aut(r), hence the isomorphism
J(r) -+ i(r) from J(r) to its dual j(]P) is Aut(r)-invariant and the second statement
follows from the first one.
186
We count the number of fixed points of r in two different ways. At first we use the
holomorphic Lefsehetz fixed point formula
E (- I)-" Trace f HP,o (m)

det (I B,,)'
(3.3)
P f (P.) P.
for holomorphic map f : M -+ M, where B,, is the linear part of f at the fixed point p,,.
a
We apply it for f -r and M =

i(r); in this case HP,O(j(r)) may be identified with the
=
p-th skew-symmetric power of the cotangent bundle at any point of J(r). For the left-hand
side in (3.3), the basis of HPO (j(r)) may thus be taken in a point Ab(PI + P2 Q, Q2) as - -
jQIi n2l JWI (PI) +WI(P2)i W2(PI) + W2(p2)}, where wi x'-'dxly and Q1 A Q2 generates
= =
H2,0 (j(r)). Since -r* Qj e'Qi, (i 1, 2), the left hand side in (3.3) gives
= =
1: (- 1)" Trace -r* I H-0 (.T(r))

P=O
= 1 -
Trace
( 0' 02)
As for the right hand side, obviously all B,, are equal, in fact
0
B,, =
(5 )0 IE2
(3.4)
when local coordinates dual to Q, and Q2 axe picked around the point P,,. Therefore
det (I -
B,,) =
(1 _
6)(1 _
62) =
3,
and the number of fixed points of -r is indeed nine.
A second way to determine the number of fixed points of -r is by writing down an explicit
list: if we write every point W E J(r) as w Ab(PI + P2 2ool) then rw w if and only if
= - =
7*(Pl) +r(P2) -1 P, + P2, i.e., P, =

P20' or P, and P2 are both fixed points for -r. Using (3.2)
we arrive at the following list
10, 01 -
021 02 -
011 001 -
M21 002 -
0011 01 -
001) 02 -
0021 01 -
0021 02 -
0011- (3.5)
The nine invariant curves are then given by the nine translates over these points of the image
of IP in i(r) by the map x -+ Ab(x -
ool). Since this curve obviously contains exactly the

four fixed points
10, 002 -
0011 01 -
0011 02 -
0011i
each of the nine invariant curves will contain exactly four fixed points. Dually, every fixed
point belongs to four invariant curves since the origin 0 belongs to the four curves
Ix -+ Ab(x -
ooi), x -+ Ab(x -
oi), i =
1, 21.
Notice that the fixed points form a group F (isomorphic to Z/3Z ED Z/3Z) which is a
subgroup of J3 (1P), the three-torsion subgroup of J(]P). On J3 (1P) there is a non-degenerated
alternating form (-, -) induced by the Riemann form corresponding to the principal polarisa-
tion. The subgroup F C J3(r) has the following property.
187
Proposition 3.2 The group F of fixed points of -r on J(17) is a totally isotropic subgroup
of J3 (IF) with respect to the Riemann form
Proof
-r is a symplectic automorphism of j3(r) (Z/3Z)4, which satisfies 1 + T +,r 2 0; also =
dim ker(T 1) 2. It follows that F consists exactly of the elements of the form
-
=
-r(x) x -
where x E J3(r). Finally, if y E F, then obviously (y,,r(x) x) 0. -
0 =
Apart from the Riemann form, which coincides on J3 (1P)

pairing e3 (see [LB]) with Weil's
a function can be defined on F with values in the group of cubic roots of unity. It is
analogous
to Mumford's quadratic form (theta chaxacteristic) on the two-torsion subgroup J2
(17) of J(r)
and can be defined in complete generality. It measures the obstruction for a line bundle L
to descend to the quotient J(]P)Ir. One can define it as follows. Choose a lineaxisation of L
with respect to the cyclic group Z/3Z generated by r, i.e., an isomorphism 0 : C4,r*(L)
with 0(0) Idco. When x is a fixed point of T, then 0 induces an isomorphism, of C., which
=
is multiplication by a root of unity e(x), and e : x i-+ e(x) is the desired function. It depends
on the choice of C itself and not only on the polaxisation. If E) is the
(theta) divisor which
corresponds to L, i.e., C [E)J, then the corresponding e ee may be computed as follows.
= =
Let f 0 be a local defining function for 19 in x. Since the divisor E) is non-singular, the
=
leading part h of f is linear and we have r*(h) e(x)h. Since the singular points are of
=
type A2, as is seen from (3.4), there exist local coordinates ju, vJ at x such that r* (u) eu =
and T*(v) e 2v. Therefore we have either h u and e(x) h and e2 Also
= = =
e, or = v e(x)=
.
if x 0 19 then e(x) 1. It follows that ee is explicitly given

=
for all x EE F by ee(x) =
-r,,ITe,
or equivalently
ee(x)v =,r*v for all v G TxE). (3.6)
The automorphisms p and a act on F as well as on the set of invariant theta curves.
It is desirable to have a "totally symmetric" theta curve, i.e., invariant by -r, o- and p. The
ain observation of this paragraph, from which the 94-configuration is a consequence, is the
following.
Proposition There is a unique totally symmetric theta curve among the nine invariant
3.3
theta curves. The
function ee associated to this curve E) is a quadratic form on F; it is given
in a suitable basis of F and upon identification of the group of cubic roots of I with F3 by
2
ee(r, s) = r _ S2 (mod 3).
Proof
The existence of the curve is clear: since the polarisation is invariant by the group
Aut(r), may find
we an invariant line bundle which gives this polarisation, hence also an
invariant divisor. It isunique since if there are two Aut(]P)-invariant curves, then their (two)
intersection points must be invariant under Aut(r) which is impossible (see Table 5). It is
easy to identify 0: it is given by the image of P -+ Ab(P + oojL 2002). To see this, notice -
that this image can be written as
P -4 Ab(P + S, + S:OL -
3S2),
independent of the choice of Si, S2 E 1011 02 0017 002}. From this representation
1
it is also
clear that 19 contains the four points Ab(Sf SI)7 S, E 1011 021 0011 0021- -
188
Let us determine ee in terms of the basis 1 1, 21 where (I = Ab(002 ool) and (2 -
2
Ab(02 -
00- Sincer mra and a
= 1 it follows using the chain rule that if r(x)
= x and =
v E 7:x., E) then
ee(xo")v = -rv =
o-.,ro-,,v =
ee(x)o,.o-.v =
ee(x)v,
hence ee(x") =
ee(x). In the same way it follows fromr =
Mr-'11 that ee(p(x)) =
ee(x)-l.
Therefore, if we identify the group of cubic roots of unity with F3 by ee ((I) = I then ee is
given by
ee(rC, + s 2) = r2 _ 82 (mod 3). (3.7)
1
The 94 configuration is now described as follows. if w and w' are two fixed points, then
W E E) + W, - = ee (W -
W') :A 0.
It follows that every invariant theta curve passes through four fixed points and that every
fixed point belongs to four invariant theta curves. Moreover we have seen that the function ee
determines the direction of the tangent to 19 in the fixed points of r. Therefore, if w, w' E F
then 0 + w and E) + w' are tangent in a common point x E F if and only if
ee+,, (x) =
ee+w, (x).
Since ee+,,, (x) =

ee (x -
w), this condition is rewritten as
ee(x -
w) =
ee(x -
w')
which -is satisfied for w' 2x w (only). We conclude that the four invariant curves running
= -
through one fixed point come in two pairs: since any two theta curves always intersect in two
points (which may coincide), the curves of one pair are tangent in their unique intersection
point and the curves of opposite pairs intersect in two different points (see Figure 7, which
also contains the dual picture, equally present in the 94 configuration).
dual
Figure 7
189
,
Chapter V11. Two-dimensional a.c.i. systems
There is also
a neat way to display the incidence relations of the
points and the curves of
the 94 configuration by an incidence diagrain analogous to the one for the 166 configuration
on the Kummer surface, as we explained in Paragraph IVAA Let us define
W,,, r ,. + 8 2 =
and r,,, 0 + W, where (I

=
Ab(002 ool) and (2 =
Ab(02 01) as in the proof of
-
= -
Proposition 3.3; also E) is the totally symmetric theta curve given by Proposition 3.3. It
follows from (3.7) that one can use for the 94 Configuration the following incidence
diagrams.
WOO W01 W02 roo ro, 1'02
WIO W11 W12 rio ril r12
W20 W21 W22 r2o r2, IF22
3.3. A projective embedding of the generalised Kummer surface
In this section will compute explicit equations for the quotient s

we
i(r)/,r as an =
algebraic surface in p4. Since T has nine fixed points, S has nine singular points and we have
seen that they are of type A2. The minimal resolution of these singularities of S leads to a
K-3 surface (a generalized Kummer surface), which we will denote by X (see [Beal]). Let
ir : J(17) -+ S be the quotient and denote by E) the unique divisor given by Proposition 3.3.
Proposition 3.4 Let M be the divisor on S for which ir*(M) [30]. Then M = is very
ample, leading to an embedding of S as the complete intersection of a quadric and a cubic
threefold in p4.
Proof
Using the quadratic form ee we see that 'C03 =
[36] descends to a line bundle M on S,
i.e., lr* (M) = L03 Let

. us denote by N the line bundle on X which is the pull-back of M by
the canonical map from X to S. Then usingC -,C = 2 we find
18 ='C(&3.'C(&3 =
(deg 7r)M -
M = 3M -
M,
so that M M =
6, which is also the self-intersection of N. Therefore, find the
-
we by
Riemann-Roch Theorem (for K-3 surfaces),
N-N
X(N) =
X(Ox) + = 2 + 3 = 5.
2
It follows moreover from Serre duality and Kodaira vanishing that dim Hi(X, O(N)) = 0 for
i > 0, so that dimHO(X, O(N)) =
X(N) = 5.
The
morphism ON corresponding to N can be factorized via the blow-up p : X -+ S and
is shown to provide an injective morphisin 0 : S _+ p4. More precisely, it can be seen by
analyzing theta curves on J that ON is one to one away from the exceptional curves. If we
consider now the surJective map
Sym HO (X, N) -+ EDt>oHO (X, M t),

whose kernel leads to the defining equations for the image of S in p4' we see by a dimension
count as above that the kernel contains a quadratic as well as an (independent) cubic form.
Since the degree of N equals six, we see that the image is the complete intersection of a
quadric and a cubic hypersurface in p4.
190
Consider now the genus 2 even Mumford system; our system of generators (2.6) of
itsintegrable algebra have the property of being weight homogeneous when the variables
uo'...'wo are assigned the weights given by the weight table below. We call a Poisson
bracket weight homogeneous of degree i when the bracket of any two weight homogeneous
elements is weight homogeneous and
I_XkXk7 AIXII jXk , X1 I
for allweight homogeneous elements xk and x, (of weight k resp. 1), where s denotes the
weight Of fXk, x1j. Then the Poisson brackets f-, J1, I-, -J-X and f.'.j,\2 have weights 2, 3
and 4 respectively. Everything is contained in the following table.
weight 1 2 3 4 5 6
Ul UO VO W2
V1 W1
WO
H4 H3 H2 H, HO
2
1.1-1 1-74 1.1-1
It follows from the table that the genus 2 even Mumford system has an automorphism defined
by
(U1iUO V15VOiW1jW2) '-+ (Mlif 2UO, f 2Vi, VO, E2WO i W1 i EW2) (3.8)
upon taking the second structure as Poisson structure. Having a (finite) automorphism allows
us a quotient system (which is also a.c.i. in view of Proposition V.2.5).
to construct Here
however we are only interested in the level sets which are invariant for this automorphism.
Again by the weight table these are the subsets of C7 given by
2
W2 -U
I +UO =
0,
W1 -
UIUO + UIW2
2
UOW2 + UIWI + WO + V1 =
0, (3.9)
ulwo + u0w, + 2v,vo =
0,
2
UOWO + VO
for n and p arbitrary. This is an affine part of the Jacobian of the curve
Y2 = X6 + 2NX3 + /'t'
whenever this curve is smooth, i.e., p =A 0. These axe presicely the curves of genus two
with an automorphism of order 3; since we have normalized the equation of our curve as
2
Y = x6 + 2=3 + 1 we will focus in the sequel only at the level sets for which p 1. =
Fromdescription of the level sets as affine parts of the symmetric square of the
our
curve automorphism on these level sets comes from the order three
it is also clear that the
automorphisms on the curves. Explicitly our map S (defined in general by (111.2.3)) is given
by
U (/\) X2- (XI + X2)1\ + XIX27 =
((XI YI) (X2 Y2)) 1-+ Y1 -
Y2 -
X2YI (3.10)
i i i
V(,\) A +XIY2
X1 -
X2 X1 -
X2
191
and w(,\) is defined by the fundamental relation u(X)w(X) + v'(X) =

f (X), where f (x) is the
2
right hand side of our equation y = x6 + 2r=3 + 1 for the curve r. Since the action on the
curve given (x, y) -+ (ex, y) we see from (3.10) that it corresponds indeed with (3.9), hence
is
our automorphism extends the automorphism on the special Jacobians.
Now that have
explicit equations for an affine part of the Jacobian of ]Pr (i.e., (3.9)
we
with p =
1), we want to construct the
regular functions on this affine part which extend
to meromorphic functions on the Jacobian of rr. having a pole of order < 3 along a fixed
component of the divisor to be adjoined at infinity recall that two translates of the theta -
divisor need to be adjoined to each affine part in order to complete them into Jacobians. As
we explained in Section V.3 this is done by using the Laurent solution (the principal balances)
e.g. to the vector field X, of which we computed the first terms in Paragraph VI.4.2. From
these first terms one finds easily (i.e., by solving linear equations) a few extra terms, to wit
1
UO =
(2a - 2a 2t + dt2 + (2ac -
ad+ 2a 2b)t3
t
+ (2ae + 2ab 2+ 2bc -

bd -
2a3 b+a 2 d -
2f)t4 +
U1 =
(I + at+bt2 + Ct3 + et4 + ft5 +
t
Here a, , f.
denote the free parameters; the Laurent solutions for v, and vo follow from it
. .
by differentiation and those for w(A) upon using the invariants. The distinguishes between
the principal balances which correspond to the two different divisors. In order to have the
functions which descend to the quotient we look at those invariant polynomials (invariant
with respect to the action of -r) which have no poles upon substituting the Laurent solutions
when picking the sign and have a pole of order at most 3 when picking the + sign. Notice
-
that invariance just means here being weight homogeneous of weight 0 mod 3.
We arrive at the following list of functions.
zo =
1,
ZI =
UJUO -
VO,
Z2 =
2u,(uo + v, -
2),
ul
2uOv 12+ 2V20 2uovl(2uo U2) 2 3 (3.11)

Z3 = + _
1 + 2uvo(u 1 _
vi -
UO) + 2u 01
3
z4 = 2v -2 (U2I + 4UO)V21 + IOVO(UIVJ -
vo) + 2v, (7UOU21 _ U41 _

I 1U2)
0
+ 2vo(2r. + 15uluo -
5u 3)
I + 2(U21 _
UO)3 _ 1OU30 -
4nuluo.
To find the image of J(r) in p4 it suffices to eliminate the variables ui, vi and wi from (3.9)
and (3.11).
In fact, from the first three equations of (3.9) the variables wi are eliminated
linearly and the other equations reduce to
2n(uo _
U2)
1 + 3ulU20 _
UIV21 -
4uOU31 + 2v,vo + u51 =
0,
4-
(3.12)
-2nuluo + uOu I UOV2-
I
3U2U2
1 0
+ U30 + V20
so it suffices to eliminate ul, uo, v, and vo from (3.12) and (3.11) (we have already eliminated
the wi-variables in (3.11)). In the latter zi and z2 are solved linearly for v, and vo,
=
VO U1UO -
Zi,
2 Z2 (3-13)
V1 = U1 -
UO +
2u,
192
and the new equation for z3, obtained by substituting (3.13) in (3.11) is then solved linearly
for uo as
2 U!2
UO 2-1 (Z3 -
ZIZ2 -
2Z2).
1 (3.14)
Zi
After substitution of (3.13) and (3.14) in the last equation of (3.11) and in the equations
3
of (3.12), we are left with three linear equations in ul, which reflects the fact that i(r) will
be a 3: 1 cover of its image in p4. If we eliminate U3I we arrive at the following two equations:
8z3-
I
24NZ2I -
4 (2nz2 + 6Z3 + -4) ZI + 4mz3 -
2r,,z22 -
3z2z3 -
Z2Z4 :--
01
4
8z I _16r.Z3I -
4 (2 + 2KZ2 + 6Z3 + Z4 ) Z2I
+(8r. + 4NZ3 -
2rz22 -
4Z2 -
3z2z3 -
z2z4) zi
+ 2nz2z3 + 14Z3 + 2z4 -
2z22+ z3z4 + 542 0. =
Using the first equation, the second equation can be replaced by
8 (1 -
3n "2) Z21+4(-2r.+(1-2n"2) Z2-6KZ3-nz4)z1+2(1-n"2) Z2 2
2
5Z3 r.Z2 (5Z3 + z4) + 2z3 (2m"2 7) Z3Z4 2z4
- _
0, - - -
=
or equivalently by tZAZ =
0, where
0 -8r,. 0 r2 -7)
2(2" -2
-8K 16(l -
3 M2 ) 4(1 -
r2)
2" -24n -4r,.
A 0 4(1 - 2X2 )
t' 4(1 _
r2)
, -5n -K
2
2(2is -
7) -24n -5n -10 -1
-2 -4n -K -1 0
Although at this point equations for the quadric and cubic hypersurfaces which define S
these
as a subset of p4 identify in the sequel with S) may not seem very attractive,
(which we will
we will see that natural coordinates can be picked for p4 in Which these equations take a
very symmetric form. Indeed, the projective basis of p4 that we used is rather arbitraxy: for
example, the coordinates of the nine fixed points for r do not possess special coordinates in
terms of the present basis. One observes that if the five fixed points for r which do not lie
on E) are taken as base points for p4 then the coordinates of the four fixed points on 0 take
a simple form and are independent of n. It is read off from the incidence diagram that
1WO07 W11i W12) W21, W221
are the five points which do not lie on E). To find their coordinates, use a local parameter t
and take x =
t,
Y = :L (I + Mt3) + 0 (t6),
picking either sign around ol or 02, and in the same way, x == t-' and
r2
1-2"
Y
(t-3 + r +
2 t3) + 0 (t4)
193
for oo, and 002. Then a careful computation yields the following coordinates:
Woo :(0:0:0:0:1),
W117 W12 : (0: 0: 1 : I: T-3 -
2r.),
W21 W22 i
: (I : 1 : ::F2 -
2r. : T-2r. : 4N2 14n + 4).
We take the points

JW00i W121 W21i W12) W11
as base points for p4 (in that order), i.e., 0 (I : 0 : 0 = : 0 : 0), etc., with associated
coordinates yo, . . .
, y4. Then the four fixed points on 0 have as coordinates
W10 =
(1: 1: 1: 0: 0), W20 =
(1: 0: 0: 1: 1)1
WO, =
(1: 1: 0: 1: 0), W02 =
(1: 0: 1: 0: 1)7
we see that they lie on the (2-dimensional!) plane
YO -:=
Y2 + Y3 ==
Y1 + Y4i
and it is easy tosee that in fact 19 is contained in this plane. The translations twO and two,
correspond to projective transformations of the surface and take in terms of these coordinates
the simple form
I 1 0 0) -1 1 0 1
(-I 0)
-1 0 1 1 0 -1 0 1 1 0
twl" -1 1 0 0 1 and tw., 0 0 0 1 0
0 0 1 0 0 -1 1 0 0 1
0 1 0 0 0) 0 1 0 0 0)
from which the equations for the planes to which the other invariant curves belong, axe
obtained at This configuration of nine points in p4 is characterized by the fact that
once.
there exist nine planes with the property that each of these planes contains four of the nine
points and every point belongs to four of the planes. Thus we have recovered in a direct way
a configuration that has been studied in the work of Segre and Castelnuovo on nets of cubic
hypersurfaces in p4 (see [Cas] and [Seg]).

The equations of the quadric and cubic hypersurfarces Q and C take in terms of the new
coordinates the following symmetric form:
Q C(YI + Y4) (Y2 + Y3 + Y4 YO) + Z (Y2 + Y3) (YI + Y3 + V4 YO) CY42+Z Y32,
- -
3 2
Y2 (YIL + Y3 + Y4 -
YO) + eCY2 ((YI + Y4)(Y2 -
YO) + YOY3 + YIY4)

3 2
C YJ (Y2 + Y3 + Y4 -
YO) - EC2 Y1 ((Y2 + Y3) (Y1 -
YO) + YON + Y2V3) 0)
where c = I + n and Z I n. The cubic equation can be simplified in a significant

= -
way by
adding to it the equation for Q multiplied with c2y, E2 Y2. The result is _
C2YIY4(Y2 + Y3 -
YO) -
52 Y2Y3(YI + Y4 -
YO) = 0-
194
If we define
X1 "
YO -
Y2 -
Y3, X4 YI + Y4 -
Y07
X2 "
-Y1) X5 Y2,
X3 -::::::
-Y4, X6 Y3;
then S is given as an algebraic variety in P5 by
C c2XIX2X3 + F-2X4X5X6 --":

01
Q C(XIX2 + X2X3 + XIX3) + E(X4X5 + X5X6 + X4X6) --":
0: (3.15)
W XI + X2 + X3 + X4 + X5 + X6 :::::::
07
and the 94 configuration is presented in the form used by Segre and Castelnuovo. In fact,
the coordinates of the singular points W,, have a I on position r + s mod 3, a I on position -
3 + (r s mod 3) and zeros elsewhere; the nine planes they belong to are given by W n (xi
-
=
Xj+3 0) for i,j

=
1,...,3. Moreover the theta curves are mapped to the nine conics
=
Cij, (I < i, j < 3), given by Wij 3Cij. For example, C16 is given as
=
CX2X3 + eX4X5 =
0,
X2 + X3 + X4 + X5 = 0-
Notice that if one changes the sign of n in the equations (3.15) then an isomorphic surface
is obtained (interchange c ++ E and xi ++ Xi+3 for i 1, =
3), in agreement with the fact
that n2 is the modular parameter.
195
4. The Garnier potential
4.1. The Garnier potential and its integrability
It is shown in [CC] that for any A =(AI) ... iAn)i the potential
n n
V\ =
( 2)2
i=1
q, + 1: A,qi2,
j=1
(4.1)
defines an integrable Hamiltonian system on R2n =

J(qj.... iqniPli . . .
7 Pn) I qi I pi E Rj,
equipped with the standard symplectic structure W =
dqj A dpi, when the Hamiltonian is
taken as the total energy
H,\ == T+ V,\, T
2
Pi2
(T is the kinetic
energy). It was pointed out to me by A. Perelomov that the integrability of
thesepotentials was already known to Garnier; therefore we will call V,\ the Garnier potential.
We study here the case n 2 (two degrees of freedom) writing
=
2
V.,3 =
(q21 + q2 2)2 + aq, +,8q2'2.
It would be interesting to study also the higher-dimensional potentials from the point of view
of algebraic geometry.
Since only interested here in the complex geometry we will from now consider
we are
the potential being defined on the affine Poisson variety C4 with the standard Poisson
as
structure. However it is sometimes useful to extend it by the parameters a and 3, i.e., we

consider C' as the phase space with coordinates qj, pi, a and 8 and with jqj,pi I I and =
all other brackets between these coordinates are zero (see Proposition 11.3.24). The Poisson
structure on both C4 and C6 will be denoted by 1-, -1.
For a: -# we define A,9 =

C[F, G] where F and G are defined by
2
F =
(q,P2 _
q2 P1)2 + a) (Pi 2q4l + 2ql2 q22 + 2aq,2),

+
2 2
G =
(qIP2 _
q2 P1)2 + (a -
#)(P2 + 2q24 + 2ql q22 + 2,6q22).
Notice that the Hamiltonian
1 2
H =
2
(P21 + P2)
2 + (ql + q2 2)2 + aql2+,3 q2 2,
which corresponds to the potential V,,,3 belongs to A,,,3 since
F -
G =
2(,6 -
a)H.
We also define
C [F, G, H, a, 0]
A ==
idl(F -
G+ 2(a -,3)H)
196
It is easy to check that IF, GI = 0. This can for example be done by using the explicit form
of the vector field XH which is given by
2 2
41 =
Pi, -2q,(2q 1 + 2q2 + ce),
2
(4.2)
42 =
P21 fi2 -2q2(2q, + 2q52+,0).
When this vector field is taken on C' then of course & = = 0 are added. It follows
that both A,6 and A axe involutive and
clearly they have maximal dimension. All level
sets of the momentum map associated to Aq are two-dimensional and the general fiber is
irreducible hence A,6 is complete. It follows that A is also complete. Thus we have verified
that (C4, 1. J, Ao) and (C6, I-, J, A) are integrable Hamiltonian systems (for all a : - 0).
,
We will be interested in the level sets of the momentum map over closed points. Apart
from Paragraph 4.5 we will only be interested in those for which a =1-,6. We will denote these
by or Ff, when a 7 8 have been fixed. Explicitly Ff, is given as a subset of C4
by the following equations:
(qIP2 -
q2PI )2 + a)(,92I + 2q,4 + 2q2q22

1 + 2aq 12) =
(W2 -
q2PI )2 + (a -
0)(P22 + 2q 4 + 2q,2 q22 + 2,6qi2) =

g.
In order to simplify some of the formulas in the sequel we let, for given f and g, the constant
h be determined by f g 2(,6 a) h.
-
= -
Clearly this system has lot of automorphisms. For future reference we list them in the
a
table below. We also add two quasi-automorphisms which are of interest (i.e., 31 and 32)- r.2
is an automorphism of order three, e being a primitive cubic root of unity and 32 is not of
finite order, A being an arbitrary non-zero complex number.
q, q2 Pi P2 a 16 F G H
21 -qi q2 -Pi P2 a 0 F G H
72 q1 -q2 Pi -P2 a 0 F G H
31 qJ q2 -Pi -P2 a 0 F G H
32 Aq, Aq2 A2p, IN2P2 A2Ce A2# A6F A6G X4H
KI q2 q, P2 Pi 0 a G F H
K2 62 qi 62q2 epi 1EP2 ea 60 f 9 62h
Table 6
The involutions z1, 22 and 3 restrict to all level surfaces Ff, and their restriction will be
denoted by the same letter. Notice that the automorphisms in the table lead to many different
quotients and they will all play a special role in this text; they are not all the automorphism
but they are the ones which are seen "at once". In view of the automorphism 32 it is natural
to consider (a,fl, f g) as belonging to the weighted projective space 24 p(1,1,3,3).
,
24
A quick introduction to weighted projective spaces is given in an appendix to [AM7].
197
Chapter V11. Two-dimensional ax.i. systems
Notice that if a =,6 then F(= G) is just the square of the momentum
q =
qIP2 -
q2PI 1 (4.3)
which is obviously in involution with the energy corresponding to a central potential. What
is remarkable however is that if a =A 6 then the
equations defining Ff, can be rewritten
(birationally) in terms of q1, q2 and the momentum q, giving precisely the equations (IV.5.2)
of the octic 0 with
,XO =
4(a _,6)2 (a +,6) -
2(f + g), Yo =
A21 =
91 yi =
q, /2 (a
(4.4)
A22 =
2(a -
0)3, Y2 =
q1 Iffg,
-
'\23 =
f, Y3 =
q2 /2(a -,3)lg.
It follows that for general f,g the surface Yf, is birationally equivalent to the affine part
oo =o n Jyo -:A 01 of the octic 0 which is itself birationally equivalent to an Abelian surface
of type (1, 4). We show in the following proposition that Yf, actually is (isomorphic to) an
affine part of an Abelian surface of type (1, 4).
Proposition 4.1 Fixing any a 54,8 E C, the affine surface.Ff, C C4 defined by
(qIP2 _
q2p,)2 + (,6 -
a) (P21 2q,4 + 2q 21 q22 + 2aq 12)

+ =
2 2
(qIP2 _
q2 pl)2 + (a -
P2 + 2qi4 + 2ql q22 + 2,6q 2) =

g,
is for general25 f, g E C isomorphic to an affine part of an Abelian surface Tf2g, of type (1, 4),
obtained by removing a smooth curve Dfg of genus 5,
-Ffg = r
f"g \ Dfg,
and the vector field XH extends to a linear vector field on T2

fg.
Proof
(i) Let G be the group generated by the involutions *11, Z2, and 3. Our first aim is to show that
.Ff,IG is (isomorphic to) an affine part of a Kummer surface. Since f and g are general, we
may suppose that (1\0 : Al : 1\2 : A3) given by (4.4) do not belong to S. For these \i, let Q be
the quadric (Kummer surface)
2 2
X2ZO_,,
0 Z2Z3 + \2I (Z2
OZI + Z2Z32) + X2(Z2Z2
2 0 2 + Z2Z2)
1 3 + X2(Z2Z2
3 0 3 + Z2Z2)+
1 2
2,\1'\2(Z0Z1 + Z2Z3)(ZlZ3 ZOZ2) + 2/\11\3(ZO-'3 -'IZ2)(ZO-'l Z2-3)+

- - -
(4.5)
2A2A3(ZIZ2 + ZOZ3)(Zl-'3 + ZOZ2) --`
07
which is obtained from (IV.5.2) by setting zi =

yi2, i.e., there is an unramified 8 : I cover
0 -+ Q; this map restricts to a map po : 00 -+ Q0, where Q0 Q n Izo =A 01. Also the rational
=
25
Precise conditions will be given later (Proposition 4.6).
198
map Yf, -+ 00 given by (4.3) and (4.4) induces a birational map 0 : Yf, / G --+ Qo, giving
rise to a commutative diagram
77f.q 00
ir PO (4.6)
.Ff,q I G Q0
Since normal, it suffices to show that

Q0 is is bijective. Obviously is surjective.- if
(XI I
QO let (Y1 Y21 Y3) be such that yi2
X21 X3) E , I xi and let q1, q2, q be determined from (4.4).
Then these satisfy the condition under which P1,P2 exist such that (ql,q2,Pl P2) E -Ffg
i
and q =
q1P2 -
q2pl. Then (ql, q2, PI P2) i

=
(XI i X2 i X3). On the other hand, if ( o
7r) (ql I q2 7 PI 7 P2) =

( o ir) (q, q211 p, p'2)
I 7
then q, =
61 qJ q2 =
62q2' 7 q =
cq', (where q
ql'p'2 -
q2PI',) forfl , E21e E 1-1, 11. Then one sees that
1 q2, p ,, p'),
(ql, q2, PI P2) ,
= SCI
1
ZC2
2
Zf (q', 2
where i'll means %k in case Ek I and identity for Ck = 1. It follows that 7r(ql, q2, PI P2) 7
k
-7r(q]'L,q2'7p'17p'2),
I
and is injective. This shows that
0 is an isomorphism, hence TfglG is
isomorphic to the (affine) Kummer surface defined by Q0.
(ii) We proceed to show that Tf, is isomorphic to an affine part of an Abelian surface, more
precisely to the normalization A of 00 (the octic is singular along the coordinate planes).
This normalization can be obtained via the birational map 0'C : 7' -+ 0. In particulax, by
restriction of (IV.5.4) to an affine piece we get a commutative diagram
A Oo
POI IPO (4.7)
Ko Qo
010
where 0Ar2 is an isomorphism. If we combine both diagrams (4.6) and (4.7) we get
Ff 9 A
8:11 18:1
.Ff., / G K0
with p the birational map 0-10

1C
and 0 the isomorphism Now the two covers Ffg
.Ff, IGand A -+ Ko axe only ramified in discrete points; the same holds true if A and Ff,
are replaced by their closures: the closure of A is just 7-2 and the closure of Ffg is obtained
199
from the explicit embedding which will be given in 4.4-1. By Zaxiski's main Theorem the
normality of T' implies that the lifting W of 0 must also be an isomorphism and we get
'Ffg = 7-2
f9 \Dfg
,,
for some divisor Df., on a (1, 4)-polarized Abelian surface It is seen that E)fg is a 4
f9
1 unramified cover of a translate of the Riemann theta divisor of the canonical Jacobian,
hence Dfg is smooth and has genus 5; an equation for Dfg will be given in Paragraph 4.4.
2
(iii) Finally we show that XH extends to a linear vector field on V
fg Letting 00 1, 01 qI .
= =
and 03 q2, we have shown that an equation for the Kummer surface of the canonical
=
Jacobian associated to Ffg is a quartic in these variables. From (4.4) and (IV.5.2) the
2
leading term in 023 is given by ((a + MOO + 01 + 02) 4(a,800 +,301 + a02) or, in terms of -
the original variables,

2 2 2
( q, + q2 + C, + p)2 -
4(a,6 + aq2 + #ql2). (4.8)

We let x, and X2 be the roots of the polynomial
2
X 2+(q2+ q2+a+#)
, X + ap + aq22 +,8ql2
as suggested by the algorithm recalled in Section VA ("suggested" because we did not prove
yet that the system is a.c.i.). Explicitly, let
2 2
X1 + X2 -':--
-(ql + q2 + a + l + -' 2 =
-2(qlpl + q2P2)i
(4.9)
XIX2 = aO + aq22 + Oq12, Xl-' 2 + -' IX2 2(oqlpl + aW2),
then it is not hard to rewrite the equations F G g, defining Ff, in terms of
X1 7 X2; 6bl &2. This gives
8(xi + a) (xi +,8) (0 + (a + #)x? + (a,8 -

h)xi + (,3f -
ag)/2(a -,8))
(XI -
X2)2
so that
dxj dX2
- + -
=
0,
A/f (XI) NRf (X ) 2
(4.10)
xldxl X2dX2
- + -
= 2v/2dt,
NR (X1) Nrf (X2)
where
f (x) =
(x + a) (X + 16) ( 3+ (a
X + #)X2 + (afl -
h)x +
2(a -,3)
ag
) -
Integrating (4.10) we see that XH is a lineax vector field on Ff,, which obviously extends
to a linear vector field on
Vfg.
From this expression and (VI.4.9) explicit solutions for qk
(k =
1, 2) axe given by
-
O[J + qk] (At + D)
qk
0[ (At + D)
where 77, and 772 axe halfperiods which correspond to -a and -,0.
Notice that as a by-product we find an equation
2
X3 Pf -ag
Y =
(X + a)(X +,a) + (a + O)x2 + (a# -
h)x + (4.11)
2(a -,6)
for the curve whose Jacobian is the canonical Jacobian associated to
fg.
200
Alternatively, the linearizing variables (4.9) suggest a morphism 0 from the potential
V,,,5 on C4 to the genus 2 odd Mumford system, namely one defines O(qj,q2,P1,P2)
(u (A), v (X), w (A)) where
2
U(A) = A2 + (q,2 + q22 + a +,6)A + afl + aq2 + flq217
1
v(A) =
[(qlPl + W2))k + (NIPI + aq2P2)]
T2
2 (4.12)
w(A) = A3 + (a + q21 -
q22)A2 + (a+#) (ql + q2 2) a'O A

2
2
e2 + q
-
a# (2a
: !- +
2,6
2
1 + q2
2
) -
The genus 2 odd Mumford system which is considered here is the one on C7 and the Poisson
structure is a linear combination of the Poisson structures before26, namely we have considered
it corresponds to 2+ (a +,6)x-a,3. This morphism 0 is neither in ective nor surjective,
W = -x
however the latter is easily cured since the image is a level set of the Casimirs of the odd
Mumford system which carries an induced a.c.i. system. As for the injectivity of 0, clearly it
can be factorized through the quotient system obtained by dividing C' out by the group (of
order four) generated by zi and 22- It follows from Example 11.2.31 that the regular functions
on this quotient are generated by
Qj =
%2, Pi =
pj2, R, =
PjQj, (i =
1, 2)
with only two relations: Rj2 =
QjPj, (i =
1, 2). Then the quotient Poisson structure is given
by
0 4Rj 2Qj
T, 0
( T2)7
0
where Ti -4R,
-2Qi 2Pi
0 -2Pj
0
and the map via which 0 factors is
U(A) A2 + (Q1 + Q2 + a +,6),\ + a,6 + aQ2 + #Q17

1
V(A) [(Ri + R2)A + (#Rl + aR2)],
T2
P1 + P2
W(A) A3 + (a +,6 -
R, -
R2),\2 +(a+#)(Q,+Q2)-a,6 A
2
P, L2 + Q1
-
0 ( Ta +
20
+ Q2
Since everything in is linear, this map is a biregular map. Thus the quotient system is
isomorphic to a trivial subsystem of the genus 2 odd Mumford system and is a.c.i. We may
conclude as in Paragraph 6.1 that (for a 54 j8) (C4, 1-, -1, Aq) is an a.c.i. system of type
(1, 4). The same is of course true for (C', 1-, .1, A). As a by-product we get also a Lax
representation for the potentials V,8. For the vector field XH for example it is given by
d (X) U(A) V(A) U(A) 0
dt ( w(A)
v
-v(.X)
vl'2-
[( W(A) -V(A) A -
2(q,2 + q22 0
where u(A), v(A) and w(A) are given by (4.12).
26
These were given in 2.3, 2.4 and 2.5 on C6 and are easily rewritten on C7.
201
4.2. Some moduli spaces of Abelian surfaces of type (1,4)

In thisparagraph we describe a map,0 from the moduli space A(1,4) Of Polarized Abelian
surfaces of type (1, 4) into an algebraic cone M3 in some weighted projective space. To be
precise we recall from Section IV.5 that (1, 4)-polarized Abelian surfaces which are products
of elliptic curves (with the product polarization) are excluded from A(1,4). The map will be
bijective on the dense subset A(1,4) which is the moduli space of polarized Abelian surfaces
(T',,C)for which the rational map Oc : r _+ p3 is birational. Thus we construct a projective
model for the moduli space A(1,4). The main idea in this construction is to see how the Galois
group of the cover
AO(,,4) -+ A(1,4) acts on P and define M3 to be the quotient. This quotient
will be easy to calculate since it is a quotient of (a Zariski open subset of) p3 by a group
which acts linearly. The fact that this action is simple is surprising and was suggested
so
to us on the by the formulas (4.4) which show that the sign of the Xi does not
one hand
matter and on the other hand by the automorphism r,., which shows that the Abelian surface
which corresponds to a, #, f g is isomorphic to the one which corresponds to 6, a, g, f which
,
indicates that the modular parameters A, and X3 can be permuted (upon adding the proper
i's or signs for A0 and X2). A posteriori we can forget about the integrable Hamiltonian
system and proceed as follows.
IV.5 that
Recall from Paragraph A'(,,4) maps onto
P \ S
P =
AO -
_A0
U (three rational curves in S, each missing eight points),
bijectively on the first component (which is dense); the three rational curves are thought of
as lying in Pl/(Ao -,Xo) at the boundary of this component. AO(,, 4) is a 24 : 1 (ramified)
-
cover of A(1,4): let or and r be elements of order 4 such that K(L) =

(a) E) (-r), and define
K, 10, a, 2a, 3al, K4 =

10, a + 2T, 2o-, 3a + 2,rl,
K2 10,,T, 2T, 3,rl, K5 =
10, 2a + -r, 2T, 2o- + 3-rl,
K3 10, o- +,r, 2o- + 2-r, 3a + 3,rl, K6 =
10, or + 3-r, 2a + 2T, 3o- + -rJ.
These are the only cyclic subgroups of order 4 of K(L). It is easy to see that taking all
possible isomorphisms K(L) 5--- Z/4Z ED Z/4Z we find exactly the 24 decompositions
K(L) =
Ki ED Kj, (I < i, j :5 6, Ji -
jJ :A 0, 3).
We describe the cover
A 24:1
(1,4) A(1,4)
and we construct a 24: 1 cover p _+ M3 and a map
0 : A(,,4) _+ M3, where M3 is an algebraic variety (lying in weighted projective space
p(1,2,2,3,4))7 such that there results a commutative diagram
24:1
A0(1,4) ,,
A(1,4)
00
1 M3
(4.13)
-p _
24:1
M3 E)
202
in which the restriction OfO to A(1,4) is a bijection (D is a divisor on MI which will be

determined explicitly).
The group G =
GL(2, Z/4Z) acts transitively on (ordered!) bases as follows: if u,,r are
b
such that K(L) =
(a) E) (,r) and
(c d)
a
E G then
b
( d)
a
c
-
(a, T) =
(ao- + bT, ca + dr),
giving a new decomposition K(f-) (ao- + br) E) (co- + dr). We denote by H the normal
=
subgroup of G which consists of those elements of G which are congruent to the identity
matrix, modulo 2. Then H acts on the set of decompositions of K(L), thus H acts on AO(
1,4)7;
to determine the corresponding action on the isomorphic space P, it is sufficient to take any
element of H, act to obtain a new basis and determine the new coordinates (YO YI Y2 Y3)
according to (IV.5.1). Substituting these in (IV.5.2) the new parameters (AO Al A2 A3)
are found immediately. The result is contained in the following table (since diagonal matrices
act trivially, only one representative of each coset modulo diagonal matrices is shown):
basis K(,C) coo. for p3 moduli in P
(o-, -r) K, ED K2 (YO : YI : Y2 Y3) (AO : 1\1 : A2 :A3)

(o- + 2,r, -r) K4E)K2 (YO Y1 iY2 iY3) .
(AO -AI : A2 A3)
(a, 2a + 7-) K, E)K5 (YO iY1 Y2: iY3) (AO Al -A2 A3)
(a + 2-r, 2o- + -r) K4EDK5 (YO iY1 iY2
: -Y3) WO 1\1 A2 : -A3)
Table 7
The upshot of the table is that all (AO : Aj A2 : A3) correspond to the same
Abelian surface. The quotient space is given by
P
P/ =
(AO : 1\1 : 1\2 : A3) -

(AO : A1 : A2 : A3) (4.14)
(P3 \ S') U (three rational curves in Y, each missing three points),
upon defining pi = A? %
as coordinates for the quotient p3, from which in particular equations
for the three rational curves as well as for the three points are immediately obtained (the
fact that there are three missing points instead of
two is due to ramification of the quotient
map at two of the three points).
The divisors S and ffl will be calculated later. We will also
interpret this "intermediate" moduli space P.
Notice that GIH is isomorphic to the permutation group S3, so we have an action of S3
on P (which extends to all of p3 since it is linear). Choosing six representatives for GIH
we find as above the following table:
203
basis K(L) coo. for P3 moduli in P'
(U"r) KIEDK2 (YO:YI:Y2:Y3) (ILO: AI: A2: A3)

(,r, 3o-) K2EDKI (YO:Y2:Yl:iY3) (-AO: P2: AI: A3)
(o-, o-+-r) K, EDK3 (VITY2: YI: N/Z7YO: Y3) (AO: A3: -/42: ILI)
(o-+-r, -0 K3 E) K2 (YI: YO: N/ZY2: NrZY3) (AO: -ILI: /-13: A2)
(3-r, o-+,r) K2 ED K3 ( rVl: iY2: 07YO: YO (AO: -P3: ILI: -P2)
7
(o-+,r, 3o-) K3EDKI (NrZY2: VZYO: -Yl: -iY3) (/ZO: A2: -A3: -Al)
Table 8
The Tables 7 and 8 together show how to reconstruct explicitly the decomposition of
K(L) from the equation of the octic. More important, it allows us to construct the quotient
space M3 as is shown in the following proposition.
Proposition 4.2 There is a bijective map A43 1), where M3 is the cone
defined by
f42 =
f1 (4f23 -
27f32)
in weighted projective space p(1,2,2,3,4) (with coordinates (fo f4)) and D =
D, + D2 is
the divisor whose two irreducible components are cut off from.A43 by the hypersurfaces
DI f4 =
f, (f, -
3f2),
(4.15)
D2 512f4 = -
16 (16f22 + 72f, f2 -
27f2
1
-
48fo f3) + 3f02 (fO2 + 24f, -
32f2) .
In particular the moduli space A(1,4) has the structure of an affine variety. The map extends
in a natural way to a map
0 : 4(1,4) _+ m3,
the image of the (two-dimensional) boundary -4(1,4) \ A(1,4) being C \ fP, QJ, where C is the
rational curve (inside V) given by
C : V02 =
4(4f2 -
fl),
and P, Q E C are given by P =

(4: 0: 3: 2: 0), and Q =
(2: 1: 1: 0: -2). Moreover, apart
from its vertex (I : 0 : 0 : 0 : 0), all points in the cone M3 correspond to some level surface
for some a,fl, f and g, with a
Proof
we describe the quotient of p3 by the action of S3, and show that it is (isomorphic
First
to) algebraic variety M3 given by an equation f42
the f, (4f23 27f32) in weighted projective = -
space p(1,2,2,3,4). To do this we use the (induced) action of S3 on C3 which is given in terms
of affine coordinates xi pi/po for C3 by =
(1, 2) (XI X2 X3) =

(-X2; -XI -X3) i
(1,2,3) (XI X2 X3) =

(-X3 X1; -X2)
7
204
Since the action is orthogonal, it must be reducible, having an invariant line and an invariant
plane orthogonal to it. Indeed let
U1 X1 + X2 -
X31
U2 X1 -
X21 (4.16)
U3 X1 + X31
then u, is anti-invariant for (1, 2) and is invariant for (1, 2,3); U2 and U3 are chosen orthogonal
to ul. Then invaxiants
2
f2 = U22 U2U3 + U3;
f3 77--
U2U3 (U2 -
U3)
for the action of S3 are found. Also there is
A = U22(2U2 -
3U3) + U23 (2U3 -
3U2)
which is (1, 2)-anti-invariant and (1, 2, 3)-invaxiant, giving a new invariant f4 ul A. = Since f2
and f3 generate the invariants depending on U2 U3 the invariant A2 is expressible
i
in terms
of f2 and f3,
A2 =
4f23 -
27f32,
i.e., A2 is nothing else than the discriminant of the cubic polynomial X3 -
f2X + f3. It follows

that
f42 =
fl(4f23 -
27f32), (4.17)
where f, U2.
1
= Notice that (fl, f2, f3, f4) have degree (2, 2, 3, 4) so that the quotient of p 3
by the action of S3 is given by (4.17) viewed as an equation in weighted projective space
p(1,2,2,3,4) with respect to coordinates (fo : f, : f2 f3 : f4). In conclusion we have established
the cover -p _+ M 3 and there is an induced map A(1,4) _+ M3 which makes
AO 24:1
(1,4) A(1,4)
1PO
P M
24:1
into a commutative diagram (since the actions on

AO(14) ,
are the same by construction).
The reducible divisor D is easily computed once explicit equations for S (or S) are
known. Since we know of no easy direct way to determine S, we postpone the computation
of S to Paragraph 4.4, where the potentials will be used to compute S in a straightforward
way; will show there that Y breaks up in four irreducible pieces 11,
we -=
01 M2 01 /Z3 0
and disc(P3/(x)) 0 where P3 is the polynomial
=
3
P3 =
4142X _
(/to + 2M, + 6M2 + 2113 )X2 + (/to -
21t, + 2A2 -
61L3)X -
4P3i
and disc(P3"(x)) = 0 denotes its discriminant (in x). Granted this, we take pi =
0, let x, 0
and eliminate X2 and X3 from fl, f2 and f4. Then the relation
f4 =
f, (f, -
3f2)
205
is found at once;obviously the same equation is found for A2 01 P3 0. The computation == =
f6i disc(P3"(x))0 is longer but also straightforward. Namely, by a simple translation in x

=
the monic polynomial P3'(x)/(4/12) can be written as X3 -ax+b, with discriminant 4a 3- 27b2.
When this discriminant (depending on pi) is written in terms of ui using the inverse of (4.16),
the equation (4.15) for V2 is read off immediately.
As for the curve to be added to (A(1,4)) to obtain 10 (A(1,4)), notice that the action
ofS3 identifies the three rational curves in (4.14), leading to a single curve. To compute its
equation (as a subvariety of VI) in terms of the coordinates fi, let, according to (IV.5.5),
pi = 0 and po =
2(/-12 + M3). Then in terms of po and /-Z2 we get
A =
mo,
fi =
(2/12 -
po/2)2,
POIL2 M20
f2 = 1122 _
+
2 4
leading to
3f02 =
4(4f2 -
fi),
by elimination of po and /12. As for the two special points P and Q on this curve, it is easy
to check that picking Mi 0, P2 A3 and po
=
2(P2 + P3) leads to the point (4 : 0 : 3 : 2 : 0)
= =
and alternatively taking p, /-12 09 AO 2P3 leads to the point (2 : 1 : I : 0 : -2). This
= = =
gives explicit equations for all these spaces and proves the announced result in (4.13).
Finally, let (fo : ...
: f4) E M3 be any point different from the vertex (I : 0 : 0 : 0 : 0)
of this cone. Then JL2 :4 0 for at least one of the six points (po : p, : p2 : jL3) lying over this
point. Define a, 0, f, g by
a =
po + 2p, + 2A2 + 2/431
=
po + 2p, -
2/-12 + 2/13,
(4.18)
f = 128 112P3,
2
2
g =
128P2/-111
then 54,3 and a, 0, f and g satisfy (4.4).
a This shows that, apart from the vertex, all points
in the M3 correspond to some level
cone surface for some a =7 #, f aald g. This
concludes the proof of the proposition.
4.3. The precise relation with the canonical Jacobian
In this paragraph we want to show that a (1, 4)-polarized Abelian surface T2 E

A(1,4) is
intimately related to its canonical Jacobian J(T2) (introduced in Paragraph IV.5), hence also
to some curve of genus two, denoted IP(T2). In fact there is more: at the level of the Jacobian,
let J(T2) be represented as C2/A, then T2 induces a non-degenerate decomposition of the
lattice A and at the level of the curve, 7-2 induces a decomposition of the set of Weierstrass
points of r(T2) which in turn corresponds to an incidence diagram for the 16r, configuration
on its Kummer surface; moreover, the Abelian surface can be reconstructed from either of
these data (Proposition 4.3).
206
Recall that the canonical Jacobian of a (1, 4)-polarized Abelian surface (T2, L) E
A(1,4) is defined as (irreducible principally polaxized) Abelian surface J(TI)
the TI1K, =
where K is the (unique) subgroup of two-torsion elements of K(,C). We have seen that such
an Abelian surface is the Jacobian of a smooth curve r of genus two, i.e., it is given as C2/A,
where A is the period lattice
A 0 H,.(r, z)
t
consisting of all periods of 0 (W1 W2), the wi being (independent) holomorphic differentials
=
i
on r. The Abelian group HI (r, Z) has an (alternating) intersection form and HI (17, Z)
can be decomposed into non-degenerate planes (in many different ways),
HI (17, Z) = HI (D H2, 0(-)H, and 0(-)H2 non-degenerate.

Such a decomposition leads to a decomposition A =
A, E) A2 upon defining
Ai =
fy c I y E HiI ; (4.19)
both H1(17, Z) =
HI (D H2 and A ==
A, E) A2 will be called non-degenerate decompositions.
They called in addition simple if each Hi is generated by cycles which
are come from simple
closed curves (Jordan curves) in P I under some (hence any) double cover 7r : r -* P 1.
The relevance ofsimple, non-degenerate decompositions and incidence diagrams (re-

called in Paragraph IV.4.4) for (1, 4)-polaxized Abelian surfaces is seen from the following
proposition.
Proposition 4.3 There is a natural correspondence between the following four (isomor-
phism, classes of) data:
(1) a (1, 4) -polarized Abelian surface r E
(2) a Jacobi surface j =
C2/A + a simple, non-degenerate decomposition A = A, ED A2
of A,
(3) a smooth genus two curve r + a decomposition )IV =
W, U W2, #WI =
#W2 =
3,
of its Weierstrass Points.
(4) a smooth genus two curve r + an incidence diagram for the 16r, configuration on
its corresponding Kummer surface.

The correspondence (i) ++ (2) is established in two ways, namely J may be taken as the
quotient of T2 using A2 or as a cover of T' using A, (or WI). Moreover, interchanging the
components of the decomposition in (2) amounts to taking the dual P of T2 in (:L). J is the
Jacobian of the curve r which appears in (3) and (4) and interchanging A, and A2 in (2)
amounts to interchanging W, and W2 in (3) and taking the transpose of both square diagrams
in the incidence diagram in (4).
Summarizing we have the following commutative diagram, determined by T2 (only),

A2
J _P
A,
1 \2, 1 A, (4.20)
T2 J
A2
where 2j denotes multiplication by 2 in J and a Ai labeling an arrow means that a projection

is considered on the quotient torus that is obtained by doubling the sublattice Ai.
207
Proof
(3)-+(2) genus two curve rand a decomposition W

Given a WUI/V2 of its Weierstrass =
points, with #Wi 3, let -7r : r -+ P1 be any two-sheeted cover of P'. It is well known
=
that ir has branch points exactly at W; the points in W as well as their projections under 7r
will be denoted by Wl,...' Wr,, also 7r(Wi) will just be written as I/Vi. If P1 is covered with
connected open subsets U, and U2 for which I/Vi C Ui and U, n U2 n W 0 then H, (r, Z)
decomposes as H, ED H2 where H, and H2 are defined as
Hi =
1-y (=- H, (17, Z) I 7r.-( E H, (Ui \ I/Vi, Z) 1.
Among the cycles in Hi there are those which come from simple closed curves in Ui \ I/Vi
encircling two points in Wi and these generate Hi. Since any (different) of these intersect
(once) the restriction 0(.)H, is non-degenerate, hence leads (upon using (4.19)) to a non-
degenerate simple decomposition A Ai ED A2 for the period lattice. Thus C2 /A and A
=
A, ED A2 provide the corresponding data.

We now only depend (up to isomorphism) on the iso-
show that the constructed data
morphism r, W
class of the data W, U W2. Let o- : r -+ IP be an automorphism which
=
permutes the Weierstrass points (such an automorphism only exists for special curves r).
Then o- extends lineaxly to Jac(r) e_- C2 /A, hence also to the lattice A, giving a new decom-
position A o-Al ED o-A2. The lattice aAi contains the periods corresponding to the points
=
aWi (with respect to the same basis of the space of holomorphic differential forms), hence
A = o-Ai ED oA2 corresponds to the decomposition W o-W, U o-W2. =
(2) -* (3) By the classical Torelli Theorem, r can be reconstructed from its Jacobian,
actually in dimension two, r is isomorphic to the theta divisor of Jac(r). The lattice A c C2
is the period lattice of IP with respect to some basis 0 JW1 W21 of the space of holomorphic I
differentials on 1P, which determines an isomorphism A -+ H, (17, Z), which in turn leads
to a decomposition H, (IF, Z) H, ED H2 upon defining Hi
=
O(Ai). =
by
If -we denote points by 7r r --+ P, any two-sheeted
1/V the set of Weierstrass of r and -
cover as above, then Hi has a system of generators Xil AQ Where 7r* Aij is a simple closed i ,
curve in P1 \ W, encircling an even number of branch points Wi, which reduces to two in
this case (there axe only six points Wi and encircling four points amounts to the same as
encircling the other two points). Since the decomposition is non-degenerate, ?r.Ail and 7r*,\i2
encircle a common point, so we may take
Wi =
ir-11points in W encircled by ir.Ail or 7r*,Xi2J.
Then #W1 = #W2 = 3 and it is easy to see that w, n w2 = 0.

again that the constructed data are independent of the choice of the basis
We show
JW1 W21 and are well-defined up to isomorphism. To do this notice first that when the choice
i
of basis 0 I
(wi, w2) is not unique, say V is another basis producing A, then c
= AOY for =
some A E GL(2, C), hence
i O=A
it 0
for any y G H, (r, Z). We find that A AA, i.e., A has a non-trivial symmetry group. Then
=
Jac(]P) C2 /A has a non-trivial automorphism group and the data (C2 /A, A
= A, ED A2) =
and (C2 /A, A AA, ED AA2) are isomorphic. Thus it suffices to show that the constructed
=
data are well-defined up to isomorphism. This follows (as in the first part of the proof) at
208
once from the property that if Jac(]P) has a non-trivial automorphism a, then it is induced
by automorphism on r. To see this property (which is paxticular for the case in which the
an
genus of r is 2) let 19 be a generic translate of the Riemann theta divisor passing through the
origin 0 of Jac(]P). Then a(()) is another translate passing through 0 (since every curve in
Jac(]P) which is isomorphic to IP is a translate of 0) hence composing a with this translate
determines an automorphism of r. This shows the constructed data are well-defined.
(2) -+ (1) Given J =

C2/A and A = A, ED A2 we form the complex torus
C2/Al with A! = IA, E) A21

2
(i.e., the first lattice is doubled in both

directions) and equip this torus with the polarization
induced by the principal polaxization on J. We claim that r is a (1, 4)-polaxized Abelian
surface which belongs to A(1,4). To show this, first notice that the cycles J/\11 7'X211 A121 A22}
introduced above, form a symplectic basis of H, (1P, Z), i.e., 0 (Ali AN) 07 0 (Ail Ai2) 17 *
= ' =
hence these cycles lead to a period matrix of the form (see [GH])
1 0 b
(0 1
a
b c)-
satisfying the Riemann conditions. Since H, is spanned by All and A12 (which correspond
to the first and third columns of this matrix) A' has in terms of slightly different coordinates
the period matrix
1 0 2b
(0 4 2b
a
4c)
which leads immediately to the result that 7-2 is a (1, 4)-polarized Abelian surface, 4: 1 isoge-
neous to J (notice that the right block
of this matrix is positive definite). Since the original
j = C2 /A is the canonical Jacobian of T2, we axe in the generic case of Paxagraph IV.5
which implies 7' E -4(1,4)-
Dually the surface is (up to isomorphism) also constructed by taking
r = C2 /A" with A" =

A, ED 2A2i
but this decomposition induces a 4: 1 isogeny from J to (this) 71.

To show that the correspondence is well-defined, observe that
(C2 /A, A =
A, ED A2) t-
(C2 /A, A =
A, E) A!2)
implies
0
/(21A, (D A2 C2
/(2IA' A2) 1 ED and
C2/ (A, E) 2A2) 0 / (Ai (D 2A!2),
the last two isomorphisms being isomorphism of polarized Abelian surfaces.

(3.) -+ For given r E AO
(2) let J be its canonical Jacobian J(7-2). Then 7-2 -+ J
-"(14), ,
is part of the isogeny 2j : J -+ J hence there is a unique complementaxy

isogeny J -+ 71
with kernel Z/2Z ED Z/2Z. Writing J as j C2/A, the latter isogeny induces an injective
=
lattice homomorphism 0 : A -+ A whose cokernel is isomorphic to Z/2Z E)

Z/2Z. Then 0
determines a unique decomposition A, E) A2 of A for which
OJA2 is an isomorphism and OIA,
209
ismultiplication by 2. We have seen that such a decomposition is simple. It is also non-

degenerate, since otherwise 7-2 would not have an induced (1, 4)-polarization (see below).
Observe that in the exceptional case that 71 -+ J is another part of the isogeny 2j, the
two isogenies combine to an automorphism of J, leading to isomorphic data in W-
(3) ++ (4) This is classical (see [Hud]); we prove it as follows. Given a decomposition of
)/V, say W =
JW1, W2, W31 U JW4, W5, W61 the corresponding incidence diagram is taken as
W11 W12 W23 W13 ip,, r12 IP23 IP13

W45 W36 W16 W26 IP45 IP36 1P.6 1726
W46 W35 W15 W25 IP46 IP35 171, 1725
W56 W34 W14 W24 IP56 1734 IP14 1724
and obviously the decomposition of W is reconstructed from it at once. To show that every
incidence diagram is of this form, notice at first that we have the freedom to permute the
rows as well as the columns, so that we can put W11 W66 in the upper left corner.
=
...
=
The curves ]Fij this point W11 belongs to are the entries in the first row and the first column
(except r1l) of the square diagram on the right. If the origin belongs to rij n ipjk, (j =,A k),
then it also belongs to rik. Then ril is easily identified as the image of the map P -* Jac(]P)
defined by
P Wh
P -+
fW" f4 +
V
W
4 (mod A),
and the other three curves axe and rl,, with Ji, j, k, 1, m, n} 11, 2,3,4,5,61. Hence
=
the incidence diagram takes the above form from which the decomposition of W can be read
Off.
If the curve has non-trivial automorphisms, we define diagrams which correspond to such
automorphisms as being isomorphic, so as to obtain the equivalence (3) ++ (4) at the level of
isomorphism classes.
Finally we concentrate on the dual 1' of T' and its relation with the canonical Jacobian
of T. At first recall from [GH] that the period matrices of 71 and f' relate as
0 0 4a 0
2b)
(1 0 4
a
2b 2b)
4c (4
0 1 2b 2b) (1
c
_
0 4
c
2b 4a J
showing that 'P is constructed from J by taking A, ED IA2

2
instead of taking !A,
2
ED A2 when
constructing 7-2 from J. It follows that the isogeny 21 can be factorized via 11 as well and
that taking the dual of 7-2 corresponds to interchanging the components of the decomposition
of A. This finishes the proof of the proposition. N
Remarks 4.4
1. If in one considers simple degenerate decompositions (instead of non-
(2) above
degenerate) then the
decomposition in (3) is altered into W IN, U W2 U W3, #I/Vi 2 and = =
the order of the components in the decomposition of IN is now irrelevant. The corresponding
object in (3.) is then a Jacobi surface (different from the one in (2)) from which the original
Jacobi surface (or the curve) cannot be reconstructed.
210
2. Since (6)
3
20, there axe 20 different incidence diagrams
= and 20 possible decompo-
sitions of the isogeny 2j : J -4 J, some of which are isomorphic if and only if J (hence r)
has a non-trivial automorphism group (i.e., different from Z2). It follows from the above
proposition that the 20 intermediate Abelian surfaces appear in 10 groups of dual pairs.
3. Let C(2) denote the moduli space of all smooth curves of genus two. Then we have
the following isomorphisms
A(I,4) W1 W2 W3 i i W4, W5 W6 7
WiEP ii 6i=*WiA jJ1 mod PGL(2, C),
W
C(2) JJW1 i w2i W37 W47 W5, W61 I Wi E P17 i 34 i = Wi 0 WjJ / mod PGL(2, C);
and both spaces are related by an obvious unrainified cover A(1,4) -+ C(2). We have seen that
A(1,4) has a natural structure of affine variety which

compactified in a natural way into
an is
its projective closure, which is the (singular) algebraic variety M3. On the other hand, C (2)
also has a natural compactification (the Mumford-Deligne compactification). It would be
interesting to figure out how both compactifications are related.
4. Among the different ways to define (and chaxacterize) the canonical Jacobian J(TI)
of T', here is a final one: J J(T2) is the
=
only Jacobian. for which the diagram
4:1
1 \\ 77-
-2
4:1
7
commutes (2T is multiplication by 2 on 71). The proof is easy using the ideas of the above
proof. Observe that this diagram is (4.20) with 7-2 and J interchanged; we could drop a
superfluous triangle since i J. =
4.4. The relation with the canonical Jacobian made explicit
We have shown in Paragraph 4.3 that there is associated to an Abelian surface of

type (1, 4) the Jacobi surface of a genus two curve IP and some additional data. Also we
have seen (in Paxagraph 4.1)

that these Abelian surfaces appear as level of the integrable
Hamiltonian system defined by one of the potentials V,8. This allows us to make this relation
very explicit (in two different ways) and to calculate precisely the locus S in p3 for which the
associated quartic fails to be a Kummer surface (and hence the associated (1, 4)-polarized
Abelian surface fails to be birational to an octic). We know of no direct method (i.e., without
using the theory of integrable systems) to do this.
Our calculationsrely on the explicit construction of an embedding for 7-2 in projective

space, which is found by using the Laurent solutions to the differential equations (4.2). Since
we know that the potential V,,,6 is a.c.i. (for a the vector field XH has a coherent tree
of Laurent solutions (see Section V.3), in particular it has Laurent solutions depending on
dim C4 1 3 free parameters (principal balances). Moreover, since the divisor
_ =
Df, to be
adjoined to a (general) fiber.Ff, of the momentum map is irreducible, there is only one such
family. Also q1, q2 and q qIP2 q2p, have a simple pole along Df, since their squares
= -
211
descend to Jac(r) with a double pole along (some translate of) its theta divisor. With this
information the principal balance is given by
q1 =
1[a +2((,
t
+a
2
-b2)a + 2aV O)t2 + bd
3
+ O(t4) 1 (4.21)
q2 =
I[b 2((,
t
+
3
+ b2 -
a 2)b + 2ba2a)t2 -
ad
3
+ (9(t4) 1 1
where 2a 2 + 2b 2 + I =
0; the series for p., and P2 are found by differentiation. Using the
Laurent solutions it is easy to find an embedding of Tf2,g in projective space: since 2Dfg
induces a polarization of type (2,8), it is very ample and this can be done using the sixteen
functions with a double pole along Dfg, to wit,
2
ZO =
1, z8 =
%,
=
z, =
q1, z9 q1q,
z2 =
q27 Z10 =
q2q,
2
Z3 =
q =
q1P2 Wli zil =
(q 1 +q22)q + aqIP2 -
&2PIi
(4.22)
Z4 =P1, Z12 =
Jq1, qJ,
Z5 =
P2j Z13 =
Jq2,qJ,
2 2
Z6 =
q17 Z14=2qlq2(ql+q22) +PIP2i
=
2
Z7 q1q2, Z15 =
q 7
where If,, f2l jJ2 -

f1j2, the Wronskian of f2 and fl. Since the embedding variables
depend regularly on the base space (i.e., on a, fl, f and g) it follows that this a.c.i. system is
completable.
We compute the correspondence between the data by using the cover J -+ T2; this can
also be done using the cover T2 -+ J (see [Van3]). RecaJ1 from Paragraph 4.3 that given
there is a unique Jacobian J J(7-2) such that =
P1
I\ J
T J
P2
yields a factorization of the map 2j (multiplication by 2). This implies the existence of a
singular divisor in T2 whose components are birational equivalent to r r(7-2) as is shown =
in the following proposition.
Proposition 4.5 The image pl(k) of Kummer's 16r, configuration IC consists of four
curves, all passing through the half periods of 7-2; these points are the images of
the six-
teen points in the configuration and each of the four image curves has an ordinary three-fold
point at one of these points, with tangents at this point, which are different from the tangents
to the other curves. Each curve is birational equivalent to r and induces a (1, 4)-polarization
on r. The image P2 (PI (IQ is one single curve, birational equivalent to
r with an ordinary
six-fold point.
212
Proof
The map pi identifies all half-periods which appear in a row in the first square diagram
of the incidence diagram which corresponds to 71. Therefore p, also identifies the curves
which appear in a row in the second square diagram of this incidence diagram and we obtain
four curves passing through the four image points, every curve having a three-fold point at
the image of the three points in the same row (but not the same column) of the first square
diagram. Since IC induces a (16,16)-polarization on J, pi(IC) induces a (4,16)-polarization
on 7, hence each component induces a (1, 4)-polarization. The virtual genus of each compo-
nent is thus five, and since each is obviously birational to r via pl, the threefold point must
be ordinary and there axe no other singular points.
The intersection of two of these components is the self-intersection of one of them (since
they are translates of each other), hence is by Theorem IV.3.7 equal to 2(5 1) 8; on the - =
other hand, since each passes through the three-fold point of the other and since they have
two simple points in common, this gives already 3 + 3 + 1 + I 8 so all tangents must be =
different and there are no other intersection points. The fact that P2 (PI (IC)) has an ordinary
six-fold point and is birational equivalent to r is shown in a similar way. 0
The image 2 j (0) is a divisor A with a six-fold point, first studied in [Van2] (where it was
an essential ingredient in the construction of linearizing variables for integrable Hamiltonian
systems) nothing but p*2 A. We have also shown there that this divisor is the
and p, (K) is
zero locus of the leading term in the equation of the Kummer surface of J (when normalized)
as in the algorithm in Section VA
To apply this in the present case, we use the leading term (4.8) of the equation of
the Kummer surface of J
(rf*g,) (which is expressed in terms of the original variables), and
investigate its zero locus, i.e.,
2 2
(q2I + q2 + Ce +p)2 -
4(a,6 + Pq 1 + aq2)
2
= 0.
This can be factorized completely as
fi=l
[q2 -
el V4a -
13 -
62iql] = 0.
reflecting the fact that p*A

2
is reducible. In order to find an equation for r(Tf2g), let q2
61 A/a + CAI in the equations for Ffg. Eliminating P2 one finds an equation for the curve
2
ACI 62 : pjQ(qj)(qj -
6162i ,Fa -
#)ql +p2( qI )=O,
where
Q(X) =
IE162i(a -
6)312X3 + (a -#)(2a _
#)X2 +6162iVfa j8(h+a(,3-a))x-2'

P is some polynomial of degree 3. This curve is clearly isomorphic to the curve
2
Z ==
X(X -
iftf2A/a 6)Q(X)- (4.23)
213
In order to decide to which decomposition of the Weierstrass points this corresponds, let
Pl'...' P4 be the following points in P15
P, =
(0 : ...
: 0 -iVa-_-O VGa--# -
1 +i(a -#)),
P2 =
(0: ... 0 +i-.,fa---,3 +V/-a--,3 1 +i (a -
P3 =
(0: ... 0 +iV a -Vfa---#
-
1 -i(a -,3)),
P4 =
(0: ... 0 -iVa-_-O +V/a---# 1 -i(a -,8)),
and let q6 denote the three roots of Q(x). Then it is easily checked by picking local parameters
around the points at infinity of A,1,2 that the incidence relation of the Pi on the A,,,, is
given by the following table:
q, -+ 0 q, --+ oo q1 -+ q6 q1 6162iVa P
A+I,+l P, P4 3P3 P2
A-1'+1 P2 P3 3P4 P,
A+I,-l P3 P2 3P, P4
A_1'_1 P4 P, 3 P2 P3
Table 9
The table is in agreement with the fact that each curve has a three-fold point and passes
the other singulaxities. Moreover it shows that the three points q6 were identified
through
under the map pi when going from J to 7-2, hence these form the subset I/V1 in Proposition 4.3
and W2 10,00,611E2iNFa 711- If we substitute
x+a
X 1-+ 7'
a-
in the equation (4.23) for the curves 6,1,2 then we find the equation (4.11),
Of
-,8))
ag
-
Y2 =
(x + a)(x + 0) X3 + (a + O)X2 + (aO -
h)x + ' (4.24)

2(a
Then the decomposition of W is given as follows: W, contains the roots of x3 + (a + t3)X2 +
(a# -
h)x + (,8f -
ag)/(2a -
20), and W2 =
foo, -a, -,31.
Suppose that (T2, f-) E A(1,4) and let the surface be represented by a surface JF(,p,f,,),
for some a :A # (using (4.18)). Then the curve IP(7) corresponding to it under the basic
bijection explained in Paragraph 4.3 must be smooth. Since we know from (4.24) that an
equation for r(T2) is given by
ag
Y
2
=
(X + Cl)(X + p)p3(X), p3(X) = X3 + (a + O)X2 + (aO -
h)x + , (4.25)
2(a -
0)
we conclude that disc(P3 (x)) 7 0 and P3 (-a) : k 0, P3 (-0) = - 0, the last condition meaning
just that f : 0 and g =,A 0. Conversely, both conditions together are sufficient to guaranty
214
that the curve is smooth and the corresponding Abelian surface is in A(1,4). In order to state
this result in terms of the coordinates yj for p3, use (4.18) to rewrite (4.25) in the simple
form y2 X(X I)pl'(X)
=
3
where
-
P3"(x) =
41Z2X3 -
(po + 2p, + 6/-42 + 2/t3 )X2 + (po -,2p, + 2p2 + 61-13)X -
4131
(x and y are slightly resealed);

representation W2 in this
10, 1, ool and W, contains the =
roots of P31'(x). P1 : P2 : P3) to correspond to a surface in 4-(1,4) is

The condition for (po
now that /11A2/-t3 0 and disc(P3"(x)) 0. It shows that the locus S' is given by the four
divisors MIA2,A3 0 and disc(P31'(x))
= 0 and the exceptional locus S is found immediately
=
from it by substituting A? for pi in these equationS27. Combining this with Proposition 4.1
we have shown the following proposition.
Proposition 4.6 The surface F(, is (isomorphic to) an affine part T2 D of an

Abelian surface (T2, [D]) E A (1,4) if and only if a 6, f 54 0, g : : 0 and disc(P3 (x)) -A 0.
Equivalently (MO : p, :/12 : A3) E p3 are moduli coming from the birational Map28 O'C : r2 _+
p3 With (7-2, L) E A(1,4) if and only if Al/Z2/13 74- 0 and disc(P3" (x)) :A 0. The curve ]P(r)
corresponding to the canonical Jacobian of T2 is then written as
y2 =X(X -
1) (4P2X3 -
(MO + 2p, + 6/42 + 2P3 )X2 + (po -
2pi + 2A2 + 6P3)X -
4/-t3)
when the coordinates x for P' is taken such that W2
the equation of =
10, 1, ool. Conversely
the octic (IV.5.2) is written down at when
of the genus two curve and
once giving the equation
a decomposition W W1 U W2 of its set of Weierstrass points: the coefficients of the octic
=
are Aj Vp-i where pi are essentially the symmetric functions of W2 when the coordinate x
=
for P1 is taken such that W2 10, 1, ool. =
Taking also the non-generic case into account, there is an Abelian surface _77(Q,#j,g)
corresponding to each point in the image 0 (A(1,4) ) (M3 \ D) U (C \ fp7Qj). =
The following corollary follows at once from this proposition.
Corollary 4.7 For any Abelian surface (T2, [D]) E A(1,4) the offine variety 7-2 \ E) is
(isomorphic to) a complete intersection of two quartics in C4.
Remarks 4.8
1. Recalling the description Of A(1,4) from Remark 4.3.2 one has the following description
of the moduli space A(1,4):
A(, ,4) -
I(IWI, W2, W317 JW41 W5, W61) I Wi E P11 i 76 j = , Wi 7 Wj mod PGL(2, C)
=
jjW4j W57 W61 I Wi E C \ 10, 1}7 i 6j = , Wj i4 Wjj 14
27
These equations for S principle be found purely algebraic, but the calculations
can in
are very tedious and some cases areeasily overlooked. In fact it is claimed (without proof)
in [BLS] that the only condition is lLlA2lJ,3 : 0, the more subtle condition disc(p3Nx)) 5A 0
being overlooked.
213
Recall that pi =
X?, where Ai are taken from (IV.5.2).
215
where the action of S3 consists of permuting 0, 1 and oo in the equation y2 =

X(X _
1)(X
W4) (x W5) (x W6), i.e., it is generated by replacing x by 11x and I
- -
equation. -
x in this
Obviously the ring of invariants of the symmetric functions of W4, W5 and W6 is just the
cone M3, which explains why A(1,4) has such a nice structure. Using Tables 7 and 8, this
leads to a geometric interpretation of the "intermediate" moduli space p3 \ S, namely
P3 \ 9 5--' JJW41 W5, W61 I Wi E C \ 10, 11, i:74- i = , Wi 7 Wj I -
To explain this, notice that taking the basis vectors mod 2 in the first column of Table 8
determines an ordering for the 4 half-periods on the canonical Jacobian which correspond to
the lattice A2, which in turn induce an ordering in the points in W2; on the other hand, all
elements in the first column of Table 7 are the same mod 2.
2. In the classical literature one defines a Rosenhain tetrahedron for a Kummer surface
as a tetrahedron in p3 with singular planes of the surface as faces and singular points of it
as vertices. In [Hud] it is shown that the equation for the Kummer surface with respect to a
Rosenhain tetrahedron is written as the quartic (4.5). It then follows from Proposition 4.6 how
to read off from the equation of a Kummer surface with respect to a Rosenhain tetrahedron,
an equation for the curve corresponding to this Kummer surface and vice versa. It seems
that this result is not known in the classical or recent literature.
4.5. The central Garnier potentials
In this finalpaxagraph we concentrate on the potentials V,,,,, which were excluded up

to now. interesting to compare the classical linearization of the central potential V,,,,
It is
which uses polar coordinates with the a 0 limit of the linearization of the perturbed
=
potential V,,,3 (a :A 3): they will be seen to coincide. We will also construct a Lax pair for
this limiting case and discuss the geometry of its level manifolds.
At first, consider for general values of h, k the level surface Yhk defined by
1
h =
2
(p21 + p2) 2 + (qJ2 + q22)2 + a (ql2+ q22),
Yhk
k =
qIP2 -
Wli
which in terms of polax coordinates (p, 0) becomes
1
h =
2 (A2 P2 2)
+ + #4 + ap 2,
k =
p 26,
leading to
1 k
P2 2 =
p6 + 04 -
hp2 +
2 2
This suggests setting 0. p2, yielding
6,2 3 2
k2
--=a +au -ha+ (4.26)
8 2
216
Secondly the transformation (4.9) reduces for a =,3 to
X1 +X2 = -
(q 21 + q22 + 2a)
2 2 2
(4.27)
XIX2 = a + aq, + aqi
and (4.10) becomes
8(xi + a)2 (2S_ + 2ax? (a2- h)xi

%
+ -
(ha + f/2))
&?% -
(4.28)
(XI X2 )2 -
The equivalence of (4.26) and (4.28) becomes clear after the simple translation xi xi + a
on the curve; indeed (4.27) becomes
51 +52 (q 2+ q2),
1 2
8182 =-
01
so that only one of the si differs from zero, say 0 =A s, (q21 + q2)
2
=
_s, (the last equality is
a definition), which matches the linearizing variable or introduced above. In terms of s (4.28)
is reduced to one equation which reads
h2 3 2 f
--=s +as -hs+-
8 2'
which is exactly (4.26) since f =

(qlP2 -
q2pj )2 = k2.
It is also interesting that the Lax pair gives in the limit a a Lax pair for the
potential V,,,,. The polynomials u(A), v(A) and w(A) axe now all divisible by (A + a),
u(A) =
(A + a) (A + q2+
I q22 + a),
I
V(A) =
(A + a) (qlpl + q2P2)
,/2-
2) A_'2 (pi +p2)

2 2 2
w(A) =
(A+ a) V + (a -
q, -
q2 2 -a (q, 2)
+ q2
which leads to a simpler Lax pair by canceling the factor (A + a).

Finally we describe the level surfaces for the central potentials V,,". These turn out to
be C*-bundles over the elliptic curves (4.26), as described in the following proposition.
Proposition 4.9 For any k, h E C, let J7hk denote the affine surface defined by
1
h =
(P21 + p2)
2 + (qJ2 + q22) 2+ a (q2+
1 q2), 2
T,hk 2 (4.29)
k =
q1P2 -
q2PI-
If k - 4 0 then Yhk is a C* -bundle over the elliptic curve
2 V
T 3 2
Ehk : - = or + aor -
hor +
_Y* (4.30)
2
Moreover the C*-action on Yhk is a Hamiltonian action, the Hamiltonian function corre-
sponding to it being the momentum qIP2 -
Wl-
217
Proof
The linearizing variables, calculated above suggest to consider the map C4 -+ C2
given by
(q, I q2 i P1 i P2) (q2I + q22,qlpl + q2P2)-
Our first claim is that the image (,Fhk) is given by the plane elliptic curve (4.30). Indeed,
one easily obtains for q,2 + q22 : L 07
q2k -
qiT
Pi -
2 7
q]2L + qi
qik + q2*T
P2 -
2 2
q, + q2
which leads by direct substitution in the first equation of (4.29) immediately to
T k
- == or3+ aor2 -
ha +
T T*
For q2 + q22 =
0, i.e., q2 =
+iql one gets
h (p21 + p2),
2
2
k q, (P2 T- iPI);
T qi(p, iP2)i
from which we deduce r =

ik, giving the point (o-, -r) =
(0, :Lik) on Ehk, proving the first
claim.
Secondly, we determine the fiber -l (a, -r) over each point on Ehk. To do this, observe
that the multiplicative group of non-zero complex numbers,
b
C* -5-- SO (2, C)
a
a) I a2+ b2 = I
acts on Fhk by
a b q, pi aq, + bq2 aP1 + bP2

-b a q2 P2 aq2 -
bql aP2 -
bP1
and the surjective map is C*-invariant. It is proved by direct calculation that the action
isfree, hence each fiber of consists of one or more C's. If (a,,r) E Ehk then p, and P2 are
determined from q, and q2 (at least if q 21 + q22 4 0), which themselves are determined (up to
the action of C*) by q2+ 1 q22 p, so exactly one C* lies over each point (qi,q2iP11P2) for
=
which q12 + q22 :? 0; in the special case that q12 + q22 0, the same is true, since pi and P2 =
are determined (up to the action of C*) by p2 1 +p2 2h, and qj, q2 are uniquely determined =
2
from p, and P2 It follows that Fhk is a C*-bundle over the elliptic curve Ehk.
-
Finally, observe that the Hamiltonian vector field corresponding to the momentum qIP2 -
q2PI is given by
41 =
-q2i P1 =
-P21
42 =
qj, P2 =
Pli
218
from which it is seen that the complex flow of this vector field is given by the C*-action,
proving the last claim in the proposition. 0
Let us define (and calculate) the moduli (in p(1,2,2,3,4)) corresponding to a level sur-
face Thk of a central potential for k 0 as the limit29
-2 2
f=k
Then an computation shows that this limit exists, is independent of f :A 0, h and a

easy
and is exactly equal to the special point P at the boundary of '0 (A(i,4)) defined
moreover
in
Proposition 4.2. Namely for f -+ g and a -+,6 one finds
(PO : PI : A2 : /13) =
(-4 : I : 0 : 1)
so that
(fo, fl, f2, f3, f4) =
(-4: 0: 3: -2: 0)
hence by weight homogeneity the associated moduli correspond to P. Notice that the point
is independent of a # as well as of f
=
g, so the map 0 does not distinguish between any
=
of the level surfaces of any central potential V,,,,,.
21
Recall that f -
g =
2(j6 -
a)h.
219
5. An integrable geodesic flow on SO(4)

5.1. The geodesic flow on SO(4) for metric II
It was shown by Adler and van Moerbeke (unpublished proof) that there exist three
classes of left-invariant metrics SO(4) for on which the
geodesic flow reduces to an algebraic
completely integrable system (a.c.i. system) on its Lie algebra 50(4). In the sequel, we will
consider the second case, known as the case of metric II. In suitable coordinates, the first
vector field X1 of this a.c.i. system is given by the differential equations
il :--
2Z5Zr,, i2 ==
2z3Z4, i3 Z5 (ZI + Z4),
i4=2Z2Z3, i5=z3(zl+z4), i6=2zlz5_
The second vector field X2, commuting with X1, is given by the differential equations
ZI =
Z24i z/2 =
z4(2Z3 -
Z6), ZI3 =
Z4Z5,
(5.2)
z4' =
z2(2z3 -
z6), Z5' =
Z3,Z47 Z6' =
ZIZ2;
the vector fields X, and X2 admit four independent quadratic invariants, given by the follow-
ing functions:
2
F, = z
3
_
z251
F2 _,2I -'i2,
2 2 (5.3)
F3 z zi
F4 =
(z, + Z4 )2 + 4(_,23 -
Z2Z5 -
Z3Z6)-
It is easy to verify that there exist precisely three
linearly independent linear Poisson struc-
tures on C6 with respect to which X, and X2 axe Hamiltonian; moreover, these Poisson
structures are compatible, implying that the integrable system admits a tri-Hamiltonian
structure. Explicitly, for any (a,,6, -y) E C3, the matrix
0 aZ6 -OZ5 0 -,6Z3 -
2,fz6 6(Z2 -
2z5)
-aZ6 0 2,yz4 a(Z6 -
2z3) 0 -aZl -
PZ4
OZ5 -2-yz4 0 -aZ5 -
2,yz2 -Y(ZI + Z4) 0

0 a(2Z3 -
ZO az5 + 2^(Z2 0 aZ3 -J6Z2

#Z3 + 2^(Z6 0 'Y(Z1 + Z4) -az3 0 2,yzl
\,6(2Z5 -
Z2) azi +J6Z4 0 6Z2 -2-yzi 0
is the Poisson matrix of a Poisson structure P,6, on C'. If (a,,6,'y) 5A (0, 0, 0) then P"'6,
generates the Hamiltonian vector fields X, and X2 as described in the following table; a
system of generators of the algebra of Casimirs of these structures P,3, also follow from the
table.
F, F2 F3 F4
Ploo 0 0 2 X2 -2XI
Polo 0 2(XI -
X2) 0 2 X,
Pool 2XI 0 0 8 X2
Table 10
220
5. Geodesic flow on SO(4)
It was shown by Adler and van Moerbeke in [AM7] that, for any f =
(fl, f2, f3, A)
which belongs to some3o Zariski open subset -H of C4, the affine surface
Af =
Jz E C6 I Fi(z) =
gi, i =
4}
is isomorphic to an affine part of the Jacobian of a compact Riemann surface rf of genus

two (which depends on f e R), Af 5--- Jac(Pf) \ Df and that the vector fields X, and X2 are
linear, thereby proving that the above system is algebraic completely integrable. The affine
the divisor Df and the Rlemann surface rf be described as follows. First notice
part Af, can
that the group T of involutions, generated by
011 (Z1 I... I Z6) =

(-Zli -Z21 Z3, -Z4i -Z5, -6),
(5.4)
U2 (Z1 7... , Z6) =
(-ZI Z2 -Z3 -Z4 Z5 -4),
7 i 7 i 1
commutes with the vector fields X, and X2 and leaves the affine surfaces Af invariant; in
for any f E W, a group Tf of translations half periods in the

fact they generate, over
tori Jac(:Pf). As a consequence, the divisors Df are also stable under these translations.
For a more precise description of the divisors Df one applies Painlev6 analysis to the vector
four principal balances,
field X, (or any combination of X, and X2). It has has precisely
labeled by el = +
1, 162 = 1 , whose first few terms are explicitly given as follows (a, b, e .
are the free parameters)-
(a -
1)el 2
4
(1 -
bt + (b -d- e )t2 + O(t3))
'El 'E2
Z2 =
(a -
abt + ((a -
1)(ae -
c -
ab2) + a2d)t2 + 0 (t3))
t
62
Z3 =
(1 + bt -
((a -
1)e + ad -
c -
ab 2)t2 + O(t3))
2t
(5.5)
61
Z4 =
T (-a + abt + Ct2 + O(t3))

IE162
Z5 =
2t
(I + bt + dt2 + O(t3))
(a 1)62 et2
Z6 =
-,. (-1 + bt -
+ O(t3))
When any of these families of Laurent solutions is substituted in the equations Fi(z) fi, =
i 1,...,4, the resulting expressions are independent of t. This leads to four algebraic
=
equations in the five free parameters, giving explicit equations for an affine part rf of Pf
Each of these equations is easily rewritten as
Y2 =
X(I _
X) [4X3f, -
(4f, + f4 )X2 + y4 _
f3 _
f2 )X + f31 . (5.6)
In what will refer to the in C2, given by (5.6), as the curve ]Pf. In order to
follows, we curve
recover the Riemann surface Pf from it one has to adjoin one point which we by oof.
denote
Since there axe four families of Laurent solutions (5.5), the divisor Df consists of four copies
Vf(1E1i62)i 621 = 62
:2
=
1, of the curve Pf, i.e.,
E)f =
Pf (1, 1) + Pf (1, -1) + rf (-1, 1) + TPf (-1, -1).
30
Explicit equations for R will be given in the next paragraph.
221
The Laurent solutions can also be used to compute an explicit embedding of the tori Jac(rf
in P'5: the sections of the line bundle on Jac (r- f ), defined by Df correspond to the meromor- ,
phic functions on Jac(Pf) with a simple pole (at worst) at the divisor Df and, in turn, these
are found by constructing those polynomials on C6 which have a simple pole in t (at worst)
when any of the four families of Laurent solutions are substituted in them (see Chapter V).
Apart from the constant function zo I and the functions zi, i=
1, 6, one easily finds =
. .
the following independent functions with this property:
Z7 =
z5(2z3 -
Z6) -
Z2-'3,
Z12 =
ZIZ2Z3 -
4Z64,
z8 =
zi (2z3 -
z6) -
z4z6,
Z13 =
Z2Z3Z6 -
ZIZ4Z5i
z9 =
z4(2z5 -
Z2) -
ZIZ27 (5.7)
Z14 =
Z2Z5Z6 -
Z1Z3Z41
z1o =
(2z5 _
Z2)2 _
Z62'
Z15 =
ZIZ2Z5 -
Z3Z4Z6-
--1.1 =
(2z3 _
Z6 )2 _
Z227
The embedding of Jac(Pf) in P15 is given on the affine part Af by the map
0: Af __ p15: p =
(_,l '... , ZO 4 (I : ZI (P) : ...
: Z15(P))-
These functions will be used later to construct two maps which are similar to 0 and which
map two different quotients of Af birationally into p3.
5.2. Linearizing variables
In this paragraph we show that from the point of view of moduli, the family of affine
surfaces E W, can be replaced by a family of polarized Abelian surfaces of type (1, 4).
Af, f
In order to do this we will first construct an explicit map from the affine surface Af (f e W)
to an affine part of Jac(:Pf). Following [BV4] we do this by following the algorithm which
was outlined in Section VA.

We define W to be the set of those f =
(fl, f2, f3, f4) E C4 for which the curve (5.6) is a
non-singular of genus two, i.e., that its right hand side is of degree 5 and has no multiple
curve
roots; notice that this entails in partic-ular that flf2f3 0 for all f E W. It will follow from
our construction that, for every f E W, Af is indeed an affine part of the Jacobian, thereby
justifying the notation W. In order to apply the procedure described in SectionVA, we fix an
arbitrary element f E W and we choose one component, say C Pf (1, -1), of the divisor Df =
on Jac(ff). The meromorphic functions on Jac(ff) which have at worst a double pole along
the divisor C can be obtained by constructing those polynomials on C' which have at worst
a double pole in t when the Laurent solutions (5.5) corresponding to el 1, 62 -1 are = ::--
substituted into them (and no poles when the other solutions are substituted). It is easily
computed that the space of such polynomials is spanned by
X0 =
11 X1 =
(-2 + -4) (Z3 + Z5), X2 =
(Z3 + Z5) (ZI + Z6) i X3 =
(ZI + Z6) (Z2 +,4), (5.8)
where we think of these polynomials as being restricted to Af. The mapping 0, given on
Jac(ff) \ C by
0: Jac(rf) \ c -+ P3 '
P =
(ZliZ2i ...
I Z6) '-+ (XO (P) : X1 (P) : X2 (P) : X3 (P))
222
maps the surface Jac(:rf) to its Kummer surface, which is a singular quartic in P3. An
equation for this quartic surface can be computed by eliminating the variables zl,...,ZC,
from the equations (5.3) and (5.8): solving the equations (5.8) and the first three equations
in (5.3) for the variables zi, z2, z6 and substituting these values in the remaining equation,
..
.'
the equation for the Kummer surface of Jac(:Pf) can be written in the form
X32((XI +X2-2f, )2 +8flXl)+f3(XIIX2)X3+f4(XliX2):--Oi (5-9)
where f3 (respectively f4) is a polynomial of degree three (respectively four) in X, and X2-
It follows from (5.9) that asystem of linearizing variables (xi, X2) is given by the equa-
tions
-2f,(x,+X2):--':XI+X2-2f,, -2flxIX2:--Xl- (5.10)
This is checked in the present case as follows. First make use of (5.8), to rewrite the equa-
tions (5.10) as
(Z3 + Z5)(Z2 + Z4) =

-2fjxjX2i W + ZO (ZI + ZO =
2f, (xi -
1) (X2 -
1) - (5.11)
Since f C W the variables x, and X2 are both different from I and from 0 so that below we
can divide by xi and

by xi 1 as necessary. Deriving
-
the equations (5.11) with respect to
the vector field X, given by (5.1) we find that
I
JxiI + Lb2X2 =
z, + z4 + 2Z37
(5.12)
Lb&l -
1)-1 + &2(X2 -
1)-1 =
z, + z4 + 2z5.
Then we can solve the first three equations of (5.3), together with (5.11) and the difference
of the two equations in (5.12) for zj, I Z6 Substituting these values in the second equation
... -
of (5.12) we find that
2
1 f2 f3
&1
(xl(xl 1)) -
X2(X2
&2
-
1) )2 = -
X1 -
X2
[4f, +
(X1 -
WX2
-
1)
+ -
X1X2
(5.13)
Notice that this equation is linear in &21 and &22 Finally we substitute the values for zi,
. Z6 . . .
,
in the fourth equation of (5.3) to find another equation which is linear in &21 and &2, 2 leading
to
&?
Xxi)
(i =
1, 2),
(XI X2) -21
-
where
g(X) =
X(I _
X)[4fI X3 -
(4f, +f)X2+(f
4 4 _f2
_
f3 )X+fl.
3
(We note that the curve y2 =

g(X) is precisely the curve ]Pf given by (5.6).) It follows that,
in terms of the coordinates X1 7 X2 given by (5. 10), the differential equations (5. 1) reduce to
the Jacobi form
&I X2 XIXI X2X2

-
-
+ - =
0, - + -
=
1,
Nrg(xl) Vrg(X2) Nrg(xl) Nrg(X2)
so that x, and X2 are indeed linearizing vaxiables.
223
The construction of these linearizing vaxiables leads to an explicit map into the Jacobian
Jac(Pf) by defining
Z2 + Z4
U(X) = X 2+
(ZI+Z2+Z4+Z6
2(z3 -
z,, 2(z3 -
z,5)
(5.14)
and by defining a polynomial v of degree at most 1 as follows
v(0)=u(0)(z1+z4+2z3), v(1)=u(1)(z1+z4+2Z5)- (5.15)

2
Indeed, g(x) -
v (X) is divisible by u(x), as can be checked by a direct computation, so that
the above formulas indeed define a point of Jac(:Pf) \ Of, where Of is (a translate of) the
theta divisor of Jac(Pf). Notice that as such this does not define a map to the odd Mumford
system because g is not monic. Since f E R, fl 0 and hence Z3 -
Z5 :A 0, showing that the

above map is regular; moreover it is birational because (5.15) gives
I (V(O) V(1)
Z3 -
Z5
-
2 i(O)
-
U(1) ) I (5.16)
while, using (5.14), z2 + z4 and z, + z6 can be rewritten as follows:
V(1) V(O)
Z2 + Z4 ::
Nl) U(O) .) U(0)' (5.17)
V(O) V(1)
Z, +Z6 =
NO) U(1) ') U(1).
Using the invariants F1, F2 and F3 one easily finds formulas for Z3 + Z5, z2 z4 and z, Z6 - -
showing that the map is birational. On the one hand this proves that when f E 71, i.e.,
when l7f is a non-singular curve of genus two, then Af is isomorphic to an affine part of
Jac(rf). On the other hand it leads to explicit solutions for (5.1) with respect to initial
conditions which correspond to a point f E W, in terms of theta functions,
2 2
V[60](At + B) V[61](At + B)
U(O) -
'0
( 0[ (At + B) ) , U(1) =
C,
( 0[ (At + B) ) '
as follows from (VI.4.9). The expressions for v(O) and v(1) in terms of theta functions follow
from it because they are the derivatives of u(O) and u(1) with respect to t.
We see that the inverse map, given by (5.16) and (5.17), is holomorphic away from the
divisors u(O) =
0, u(1) 0 and u(l)v(O)
= -
u(O)v(l) = 0. When u(O) = 0 then 0 is one of the

roots of u so that the corresponding divisors are of the form Wo + P, where Wo stands for
the Weierstrass point over 0, x(WO) 0 and P E Pf Similarly, u(1)
= 0 corresponds to the -
=
divisors W, + P, where W, stands for the Weierstrass point over 1. In order to avoid a rather
involved explicit computation for the third divisor we appeal to the fact that the divisor at
infinity Df is invariant for the group Tf. Knowing that Df consists of the theta divisor
(consisting divisors oof + P) besides the two divisors that we have just determined we can
of
identify the elements of Tf as translations over [W1 Wo], [oof Wil and [Wo oof ]. Thus, - - -
the divisor u(1)v(0)-u(0)v(1) corresponds to the effective divisors in [Wo+W1+P-oof].

= 0
It is now easy to see that the four points 2oof, oof + Wo, oof + W, and Wo + W, (which
constitute a single Tf orbit) each belong to exactly three of the four curves and that these
224
four curves have no other intersection points. Thus, as a byproduct, we have recovered3l the
following intersection pattern of the components of the divisor Df.
Figure 8
We will now use the above results to study the moduli space M defined by
M =
JAf I f E WI/isomorphism,
where isomorphism. means isomorphism of affine algebraic surfaces. We will relate this moduli
space to the cone M', introduced in Section 4. In the following two propositions we show
how M and M(1,4) are related.
Proposition 5.1 For any f E R the quotient Af ITf is an affine part of an Abelian
surface Tf The line bundle Cf
.
=
[Df ITf I induces a polarization of type (1, 4) on Tf and the
induced map OCf : Tf -+ P3 is birational to its image.
Proof
We have shown in Proposition 4.3 that the quotient of a by a group of
Jacobi surface
translations of the form 10, [W2 -
Wj], [Wi -
Wo], [Wo -
W2]1,WO, W, and W2 are

where
Weierstrass points on the underlying curve, is an Abelian surface of type (1, 4), more precisely
it belongs to -4(1,4). The divisor Df descends to the irreducible divisor Df ITf which has a
triple point which corresponds to the singular points of Df. Since Df induces a polarization
of type (4,4) on Jac(Pf), DfITf induces a polarization of type (1,4) on 7-f. In order to
see that the induced map Ocf is birational onto its image one considers TflKf where Kf
is the group of two-torsion elements inside the kernel of the natural isogeny from Tf to its
dual Abelian surface 'ff. Since Tf Jac(:Pf)/Tf the map Jac(rf) -+ TflKf is an isogeny
=
whose kernel consists of the sixteen half periods of Jac(Pf). This means that this isogeny is
multiplication by 2 in Jac(:Pf) and hence that 7f' lKf is a Jacobi surface. This implies that
the map OCf : 7-f _+ p3 is birational to its image. 0
31 the Laurent solutions

This intersection pattern was first determined in [AM9] by using
to the vector field Xi.
225
Proposition 5.2 The above correspondence between affine surfaces Af and Abelian sur-
faces T induces a bijection X : M
Proof
For f E ?i we know that,Tf is a group of four translations of Jac(rf) over half periods
leaving Df invariant. Since the group of translations over half periods acts transitively on
the set of theta curves this property characterizes Tf. It follows that isomorphic surfaces Af
and Ak lead to isomorphic quotients Af ITf and AkITk and hence to isomorphic polarized
Abelian surfaces (7-f ,Cf ) and (Tk, Lk). This shows that the given correspondence between
affine surfaces Af and Abelian surfaces T induces a map X : M -+ A(1,4).
Starting from polarized Abelian surface (T, Oc) of type (1, 4) for which the in-
any
duced map is birational there exists a Riemann surface V and a partition W-= W1 U W2 =
JWo, W1, W21 U JW3, W4, W51 of its Weierstrass points such that T Jac(]P)/T, where 'I =
is the group of translations, given by T 10, [Wo Wj], [WI W2], [W2 WO] I. Moreover
= - - -
the triple (f W1, W2) is uniquely determined up to isomorphism (see Proposition 4.3). Let
,
us pick one particular triple (P, W1, W2) and let us choose coordinates for PI such that the
.mage of W, under the natural double cover P1 is given by 0, 1 and oo (in some order).
Then we find an equation of the form
Y2 =
x(I -
x)(AX3 + BX2 + Cx + D)
in which the right hand side has no double roots.

Obviously we can find then at least one
f E W such that this above curve corresponds to the
rf, given by (5.6). By construction
curve
(the isomorphism class of) the affine surface Af is contained in the fiber X- 1 (TC), showing
the surjectivity of X. Finally, a triple (V, Wl, W2) which is isomorphic to (TP, W1, W2) leads
to an isomorphic surface Ak because Af is intrinsically described in terms of the triple
(r, W1, W2) as being the affine part of the Jacobian of I-', obtained by removing the translates
of the theta divisor, corresponding to the half periods 10, [Wo Wj], [Wi W2], [W2 WO] - - -
where W, :--:
IWO W1 W2 I
7 ,
5.3. The map M -+ M1
It follows from Section 5.2 that for any f E W the line bundle Lf which corresponds to
Df ITf defines a birational map 0,cf from Tf to an octic surface in P3. We will compute an
equation of this octic because the coefficients of this equation, which depend on f, will allow
us to construct explicitly the map M _+ M3. Since Tf Jac(ff)/Tf the = vector space of
functions which provide this map consists of the Tf-invariant functions on Jac(rf) with a
simple pole along Df (at worst), i.e., the T-invariant functions in the span of jz07 Z151- ...
1
Using (5.4) and (5.7) one finds the following four independent invariant functions:
00 =
Zo =
1,
01 =
z1o =
(z2 -
2z,5 )2 _
Z62'
(5.18)
02 =
Z11 =
(2Z3 _
Z6 )2_Z221
03 =
Z12 =
Z1Z2Z3 -
Z4Z5Z6-
In order to compute an equation for the octic it suffices -

in principle to eliminate the
-
variables z, Z6 from the equations (5.3) and (5.18). In practice, doing the calculation in a
226
5. Geodesic flow on SO (4)
straightforward way leads to disastrous results, even when using a computer algebra package
such as MuPad or Maple. Therefore we will describe in some detail how this computation
can be done. As a first step we notice that the octic which we want to compute is isomorphic
to the variety defined by the following equations:
h X3 -
X5, 0 X1X4 _
Z2,
1
2
f2 XI -
x6i 0 X2X5 -
Z2 i
f3 X2 -
x4i 0 X3X6 -
Z2,
3
f4 = X, + X4 + 2ZI + 4X3 -
4Z2 -
4Z3, (5-19)
01 =
4X5 -
4Z2 + X2 -
X6,
02 =
4X3 -
4Z3 + X6 -
X2i
02=
3 XIX2X3 + X4X5X6 -
2ZIZ2Z3.
To see this, we consider a regular map V from the variety given by (5.3) and (5.18) to the
variety given by (5.19). The map V is given by Xj zj2 and Zj Zi Zj+3, where i 1, 6 = = =
. ..
,
and j 1,
= 2,3. On the one hand V is constant on the orbits of T because all Xi and Zj are
T-invariant; on the other hand it is easy to check that every fiber of W contains precisely four
points, hence the degree of W is four. This shows that (5.19) represents the image of Af /Tf
in projective space, obtained by using the sections of the line bundle associated to Df ITf -
Six of the equations in (5.19) are linear and we can use these equations to eliminate
X21 X3) X5, X6, Z2 and Z3 from the four non-linear equations. Apart from X,X4 Z12, this =
leaves us with the following three equations (we have used X,X4 Z12 to simplify them) =
2 2
2 [f3XI -
f2X4 -
(f2 -
f3 -
01 -
02)ZI21 -
2(4f, + f2 -
f3 -
f4 + 0I)X4ZI
-
2(f2 -
f3 -
A + 02)XIZ1 + 2f3(4f, -
A + 01 + 02)XI + 2f2(f4 -
01 -
02)X4
(h + f3 -
f4 + 02)(4f, -
f2 -
f3 -
f4 + 01)Z1 -
8 023 =
0,
4f3X, -
4(f2 01) (f3 + X4) + 4XI X4 -
(f2 -
f3 -
f4 + 02 + 2ZI )2 =
0,
4(f3 + 02)(XI f2) -
4X4f2 + 4XIX4 -
(4f, + f2 -
f3 -
f4 + 01 + 2ZI )2 = 0.
(5.20)
The first trick that we use computation feasible stems from the
to make the rest of the
following observation. If we multiply the second equation by X, and the third equation by
X4 to remove from the first equation in (5.20) those terms which contain X21 and X '2, then
the resulting equation is a linear equation in XI, X4 and ZI (the relation XJX4 Z12 is again =
used to simplify this expression) so that (5.20) is equivalent to a linear system of equations in
XI, X4 and ZI, which is solved at once. An equation for the octic is then given by substituting
the expressions for XI, X4 and ZI in the only remaining equation X,X4 Z12. =
The resulting equation is monstrous (it has 2441 terms), in contrast with the equation
(IV.5.2) for the octic, corresponding to an Abelian surface of type (1, 4). By a recaling of
some of the coordinates by roots of -1 we write the octic. in the following more symmetric
form:
2 2
IL YO YI Y2
2
0 1 + Y4Y4)
2V32 + P2(V4Y4
1 2 3 + t42(y4V4
2 4Y4)
1 3 + Yo 2 + P2( 41,43 + VI4V4)+
3 Yo 2
2 2
-
21A, 02 (Yo YI + 2Z/32)(y2V2

V2 0 2 2t,2)
Y3 1 (Y2Y2
2/12P3 0 2 +
_
Y2y2)(Y2Y32
3 1 0 Y2Y2)
-
1 2
_
(5.21)
-
2P3P1 (y2y20 3
+ Y2y22)(y2Y2
1 0 1 y2y2)
2 3
0. _ =
227
The difference between these two equations lies of course in the choice of coordinates. In order
to compute the coordinate transformation which reduces our equation to the symmetric form
(5.21) we use the

following geometric farct. Since the octic that we obtained has the form
A043 + B023 + C2
0 the octic has a singular point of order four at (0 0 0 : 1) and such
= - -
a singular point necessarily comes from four of the sixteen half periods on Jac(rf). Clearly,
(5.2 1) also has a singular point of order four at (0 : 0 : 0 : 1). On the other hand, we see that
the tangent cone to (5.2 1) at (0 : 0 : 0 : 1), is the union of four hyperplanes because the zero
locus of the coefficient of y34 in (5.21) has the form
(YO + Y1 + Y2) (YO -
Y1 + Y2) (YO + Y1 -
Y2) (YO -
Y1 -
Y2)
Where Yo =
V//_13YOi Y1 V/P-2YI and Y2
= =
i,//-IIY2 (the particular choices made for each
square root are irrelevant). The coefficient A of 04
3
in our equation for the octic must also be
the product of four linear factors, but these are harder to determine because this can only
be done by passing to an extension field of the field C[f3., f2, f3, f4]. However, if one uses the
sections of a symmetric line bundle to map a Jacobian in projective space, then symmetric
equations for the image are usually obtained by explicitly introducing the Weierstrass points
on the curve, rather than working with the coefficients of a polynomial that defines the
underlying curve (see [PV2]). In view of the equation (5.6) for ]Pf we are therefore led to
32
defining
4x 3f, -
(4f, + f4)X2 + y4 _
f3 _
f2)X + f3 :--
(X -
Al) (X -
A2) (X -
A3)
Indeed, in terms of the Ai one finds the following factorization for A,
A =
00 JJ[AAi(Ai -
1)00 -Ai01 -
(Ai -
1)021-
i=1
In order to find the required coordinate transformation, we can now use the following ansatz-
YO + Y1 + Y2 =
00i
-YO -
Y1 + Y2 =
K1 (AINI (AI -
1)00 -A101 -
(Al -
1)02)1
(5.22)
-YO + Y1 -
Y2 =
r1,2(AA2(A2 -
1)00 -A201 -
(A2 -
1)02)1
YO -
Y1 -
Y2 =
r-3(A/\3(,X3 -
1)00 -
A301 -
(A3 -
1)02)-
The coefficients ni are uniquely determined by the compatibility equations, which stem from
the vanishing of the sum of the left hand sides of these four equations. If we denote A(x) =
A(X Al)(X A2)(X A3) then the solution to the compatibility equations
- - -
is given by rvi =
-1/A'(Ai), (i 1,...,3). Substituting these values for r,,i in (5.22) we

= can rewrite our
equation for the octic in terms of the coordinates Yo,...'Y3. Putting Yi piyj we can =
determine the pi such that we obtain precisely (5.21). It gives the following values for
Ai /Ili /12 7 P3:
N? =
Ai(I -
Ai)(Ai+1 -
Ai+2)3i
(5.23)
P2 =
12(or22 _
0.20.3)
1 + 2(c2 -
al)(U10'2 + 9(73)1
32
The final result will be symmetric in Al, A21 A3, hence does not depend on the order of
these parameters.
228
where ai is the i-th symmetric function Of /\I i 1\2 7 1\3 and A4 1\1 i \5 1\2 This determines
-
the parameters pi explicitly in terms of the Weierstrass points of the curve ff. The sign of
the parameters p and pi is not important. Indeed, the coefficients (A,A1,/-12,A3) are only
intermediate moduli for Abelian surfaces of type (1, 4), the moduli themselves being given by
the following expressions which realize the moduli space as the cone C : f42 f, (4f23 27f32) = -
in weighted projective space p(1,2,2,3,4):
f0 =
P2,
f1 =
(/'121 + /122 + tZ2)2'
3
f2 =
1141 + P42 + /.143 P2P21 2
_
P2P2
2 3
_
/12t,2
3 11
f3 =
(P22 P2)(/,2
_
1 3 _/,2)(/
2
_,2
1
_
112),
3
(/t21 + t,22 + /'Z2)(P21 + 142 2-2 P2)3 (/12

2 + P2 2/.z2)(/.A2
3 + P2
f4 =
3 3
-
1 1
-2 A2).
2
2 2 2
The standaxd action of the symmetric group S3 Oil C [1\1 A2 1\31 induces on C [/.tl 'P2 /.13] an
i i
action which is determined by (1, 2) (j12'

1 /,42,
2 p2)
3
.
2 -p2,
(-M2, 1 -/.12)
=
3
and (1, 2, 3) -(/42, 2, /-, 32)
1 it 2
2 /,12,
(112, 3 p2) Therefore, every symmetric fimetion in C[p2' 1 p2'
2 p2]
1 .
3
is either invariant or anti-
invariant with respect to this induced action. It follows that the above polynomials fo, f4 . . .
,
axe symmetric in Al /\2 A3. They axe easily expressed in terms of f

7 i (fl f2 f3 f4) giving
=
i i i 7
an explicit formula for the map M -+ M3 C p(1,2,2,3,4).
229
6. The H,6non-Heiles hierarchy
6.1. The cubic H6non-Heiles potential
On C' we take qj, q2, pi and P2 as coordinates and the standard Poisson bracket given
by jqi, q2J :--
JPI) P2} =:
0; jqi, pj I =
Jij. We denote by A3 the algebra C[K3, L,3] where
1 3
K3 (P21 + P2)
2 + 8q2 + 4q2q2'
1
2
2 2
L3 _
q2 P2I + qlpIP2 + q, (q, + 4q22).
They are in involution and determine the following two (commuting) vector fields:
41 Pi, qI =
-2q2pi + q021
42 P27 q'2 =
qipj,
2
j -8qlq2, P11 =
-PIP2 -
4qj (q1 + 2q22)7

2 2
2 -4q I
-
24 q22' P12 ==P21 -
8qlq2.
Clearly, for any closed Spec A3 its corresponding level set has dimension two and is
point in
irreducible hence A3 is complete(by Proposition 11.3.7). It follows that A3 is integrable.
We now define a regular map 03 : C4 _+ C6, where C6 is the phase space of the genus 2
odd Mumford system, which we view again as the affine space of Lax operators (2.1). The
map 03 is given by
2
1 U(A) = '\2 2q2A -
q,
V (A) =
(P2 A + q1PI)
(qj, q2 i P1 i P2) vf2-
p2
--i
w(A) = A' + 2q2 A2 + (q2 +4q 2),X _
It is verified by direct substitution that
0*3 (_U21 + U0 + WI) ==

0,
0*3 (_UIUO + U, W, + Wo + V2)
1
= _K3,
0*(ulwo
3 + uow, + 2vivo) =
-L31
0*3 (UOWO + V2)
0
= 0.
In terms of the system of generators H3 of the integrable algebra A for the genus 2 odd
Ho, . . .
,
Mumford system, this means 0*H0 3 0, O*H2

3 -K3, OW,
=
3 O*H3
3 -L3- It follows that
= = =
03*A A3 and the Poisson structure on C6 which makes 0,3 into a morphism. (of integrable
=
Hamiltonian systems) is the one given by (2.4), denoted here by 1'7 *12-
This morphism is neither injective nor surjective. It is however finite and its image is
given by Ho H3 =0. Thus we know from Corollary 11.2.16 that 0 is the composition
=
of a surjective and an injective morphism. Indeed, since the Poisson structure 1, *12 has 1
Ho and H3 as Casimirs, it restricts to a Poisson structure on the level over any point of
CaS(I','12), in particular to the level Y defined by Ho H3 0. Since .97 is irreducible = =
230
6. The Mnon-Heiles hierarchy
it is an affine Poisson variety and since it is four-dimensional (37, A) is an integrable

Hamiltonian system (here 1* 12 and A denote their restrictions to F). Thus we can see 03
as a surjective morphism,
03 : (C4, 1, A3) -+ (-T-77 J* *}27 A)

i
to a trivial subsystem of the genus 2 odd Mumford system.

We decompose further this surjective morphism. 03- Consider the action T Of Z2 on C4
given by
T (ql, q2) P1 P2)I (-ql 7 q2 j -PI 7 P2)
It isan automorphism of the 116non-Heiles system since it is a Poisson map and 7-*A3 A3; =
in fact even more is true: r*f f for all f E A3. By Proposition 11.3.25 we have a quotient
=
integrable Hamiltonian system which we denote as (C4/T, f_, '10 A3) (Since all elements of A3 ,
go to the quotient). Notice that in this case the level sets are quotients of the original level
sets. Moreover the action leaves 03 invariant and 03 can be factorized via the quotient C4/7-,
i.e., we have a morphism (between affine varieties) 3 : C41-r -+.F which makes the following
diagram commutative.
c4
ir
I \\
C4/,r -
0 jr
03
Since 03 and 7r axe Poisson the same is true for 3 and *3A =
A3 by surjectivity of 7r. Thus
3 : (C4/Tj 1, 1, 10 A3) 5
-+ (-177 1 *
1 '12 A)
1
is a surjective morphism of integrable Hamiltonian systems.

It follows from 11.2.31 that Q,
Example q,2, Q2 Q) P1 p2'
1
p2=
P2, R qlpl = = =
generate the algebra of invariant functions on C4 for the action of r; they generate the algebra
of regular fimetions on the quotient, which therefore is a cone with equation R2 QJPJ. The
map 3 is then given by
U(,X) = A2 -
2Q2/\ -
Q1;
(Q1 Q2, P1 P2 R)
v (A) =
(P2 A+ R),
, i -vF2
P1
W(A) = A3 + 2Q2 A2 + (Q1 +4 Q2)'X
2
_
and it is important to note here that this map is biregular. The first conclusion is that the
116non-Heiles potential admits an automorphism of order two whose quotient is isomorphic
to a trivial subsystem of the genus 2 even Mumford system. It is easy to see that the quotient
map is even unramified.

Since 3 is an seen that (C4/,r,
isomorphism we have A3) is an a-c.i. system (whose
general fiber is an affine part of
Jacobian). To conclude that the H6non-Heiles system is
a
a.c.i. we use the explicit expressions (VI.4.9) of the variables ui in terms of theta functions. It
follows from that formula that uo is the square of a quotient of two theta functions, hence qj,
which is its square root, is a quotient of two theta functions. It follows that the generators
of the ideal which define the level sets axe expressible in terms of theta functions, hence they
define an affine part of an Abelian variety.
231
6.2. The quartic Mnon-Heiles potential
We now turn to the H6non-Heiles potential V4. It is defined on C4 with the same Poisson
structure as the cubic H6non-Helles potential, the integrable algebra now being given by
A4 =
C[K4, L4] where
1
K4 =
(p21 + P2)
2 + q41 + 12q2q
1 2
2+ 16q42,
2
2 2 3 2
L4 =
q2P1 + qIPIP2 + q, (8q2 + 4q, q2
A4 is also contained in the algebra of invariant functions (for T) and leads also to a quotient
system (C4/,r, I., .1o, A4). The analog of the map 03 is in this case the map 04 : C4 _+ C7
defined by
2
U(A) = A2 2q2A -
qII
V(A) =
(P2A + qlpl),
(ql, q2, PI P2) i N12-
2
W (A) = 'X4 + 2q2 A3 + (q1

2
+ 4q22) A2 +
2
4q2 (q 1 + 2q22) -Pi
2
Here C7
interpreted as the phase space of the genus 2 even Mumford system. The map 04
is
is a morphism when the Poisson structure (2.8) is chosen for it. As in the previous
Poisson
paragraph it follows that the quotient system is isomorphic to a trivial subsystem of the
genus 2 even Mumford system, the level now being defined by H4 H3 Ho 0. In = = =
particular the quotient system is an a.c.i. system (with an affine part of a two-dimensional
Jacobian as the general fiber of its momentum map).
Our argument which showed that the H6non-Heiles potential is a.c.i. is not valid in this
case. To see this let us fix one Jacobian from the trivial subsystem of the genus 2 even
Mumford system (or from the isomorphic quotient system) and recall from Section VIA that
the divisor to be adjoined to the general fiber of the momentum map consists of two translates
of the theta divisor and that uo has on each of these a simple pole. If the quotient map,
restricted to general level set, extends to an unramified map 7r over the whole Jacobian
a
then ir*uo has simple pole on the inverse images of the two translates of the theta divisor;
a
since 7r is unramified these inverse images are reduced. But then 7r*uo cannot be the square
of a meromorphic function on the inverse image of the level set (which is a level set of the
quartic potential). It follows that -7r must be ramified and since it is unramified on the affine
part it must be ramified at the divisor at infinity. Thus the completed general fiber of the
momentum map of the quaxtic potential is a ramified cover of a two-dimensional Jacobians.
A first example of an integrable Hamiltonian system with this property was given by
Bechlivanidis and van Moerbeke in [BM].The aJgebraic invariants of the surfaces which
appeax as ramified covers of Jacobians in the context of integrable Hamiltonian systems were
computed by L. Piovan (see [Pio2]).
232
6. The Mnon-Heiles hierarchy
6.3. The H6non-Heiles hierarchy
For the higher potentials the situation is even worse. We can still construct a quotient
system, but these quotient systems are not a.c.i. They are however still trivial subsystems
even
of the (two-dimensional) hyperelliptic systems which we considered in Paragraph 111.2.4; these

hyperelliptic. systems correspond to higher genus curves and are not a.c.i. (recall that only if
g = d they are a.c.i.; here d 2 and g > 3). =
We still considering the standard Poisson structure on C4 and consider now the
axe
integrable algebra A,, [Kn, Ln] which generalizes the algebras A3 and A4. The polynomials
=
Kn and Ln are given by

Kn =
(P21 +P2)
2 + Vn,
2
2
Ln =
-q2p, + qipIP2 +qI2Vn-17
where Vn is defined by
[n/21
n k
) 2kqn-2k.
-
n-2k
Vn 2 q, 9,
2
k
k=O
and satisfies the recursion relation
Vi+2 =
2q2Vi+l + q211
We still have 7- defining an automorphism of order two and we can consider the quotient
system. The analog of the morphisms 03 and 04 is the map On defined by
2
U(,\) = A2 _
2q2A -
qj I
V(A) =
(P2A + qlPI)7
(qj, q2 i P1 i P2)
V2-
n-1 2
1
w(A) =
1: VjAn-i -
2
i=O
Using the recursion relation it is easy to compute that
U(A)W(A) + V2 (A) = An+2 _

A2Kn -
ALn.
This can be seen as in the previous cases as a morphism. to a trivial subsystem of one of
the hyperelliptic systems discussed in Paragraph 111.2.4 (namely the one associated to the
family of polynomials F(x, y) y2 Xn+2 + aX2 with parameter a). These are also multi-
= -
Hamiltonian and the Poisson structure which has to be taken (in order to have a Poisson
2
morphism) is again the one for which all but the coefficients of A and A are Casimirs; this
determines the brackets of the coefficients of w(A) in terms of the brackets of the coefficients
of u(A) and v (A), which are 0 zero except Jul, vi I luo, vo I =
-uo (which corresponds
to W = x in the notation of Chapter III) -
We are merely interested here in the level sets of the momentum map of the potential
Vn (n > 4). Clearly these surfaces are unramified 2 : I covers of the level set FF,2 (defined
in (111.3.1)) where F y2 Xn+2
= A2Kn ALn- We determine these level sets for F of the
- _ -
2
more general form F
y f=
(x) and of any dimension lower than the genus of the curve rp
-
(which is assumed non-singular) in the following proposition.
233
Proposition 6.1 In the hyperelliptic case F(x, y) yl f (x), the level set Fpd is for
= -
d < g biholomorphic to a (smooth) affine part of a distinguished d-dimen8ional subvariety

Wd of Jac(Pp), namely
TF, d Wd Wd -I deg f (x) odd,
TF,d Wd (Wd-1 U (e + Wd- 1)) deg f (x) even,
where 6 E Jac(P.F) is given by 6 =

Ab(ool -
002) 10 mod Ap,. Also
W, =
Jac(fp),
W,-i = theta divisor E) C Jac(f.F),
Wi = curve PF embedded in Jac(PF),

Wo =
origin of Jac(rF).
Proof
We prove the proposition only for the case in which deg f (x) is odd. In this case IFF
is compactified by adding one point which we call oo. We choose this point as the base
point for the Abel-Jacobi map (on the symmetric product) and define Wk for k 1,...,g =
as Wk Abk (Sym" VF). By Jacobi's Theorem Wg

=
Jac(:PF) and by Riemann's Theorem,
=
W,_1 is (a translate of) the Pdemann theta divisor. Clearly for each k < g, Wk-1 is a divisor
in Wk and, Wk \ Wk-1 is smooth. We claim that
d
Abd(SYM IPF \ DFd) =
Wd \ Wd-1,
more precisely Abd realizes a holomorphic bijection between these smooth varieties. Namely,
(P Pd) C SyMd ]pp \ DFd

Pi =
Pj and Pi is not
Vi Pi :,4- oo and 3i : jx (Pi) = x (Pj)
a ramification point of x
Abd(Pli... Pd) 0 Wd 7 Wd-li
where we used Abel's Theorem in the last step. It follows that SYMdr. \ DPd and Wd \ Wd-1
are biholomorphic, hence by Proposition 111.3.3, Abd OOFd is a biholomorphism and the
manifolds -7Fd and Wj \ Wd- I are biholomorphic. 0
Applied to the case of the H6non-Heiles hierarchy the proposition says that the general
fiber of the momentum map of the n-th potential of this hierarchy is an unramified 2 : 1
cover of an affine part of the W2 stratum of a hyperelliptic curve of genus ["+'];
2
this affine
part is completed into the W2 by adding one copy of the curve if n is odd, otherwise
stratum
two copies need to be added. As before, the morphism. on to the Mumford systems lead to
a Lax representation (with spectral parameter) for the H6non-Heiles hierarchy. Notice that
although we have a Lax equation with spectral parameter the system is not a.c.i.: its general
level surface corresponds to a non-linear subvariety of a higher-dimensional hyperelliptic
Jacobian.
234
7. The Tbda lattice
7. The Toda lattice
In this paragraph we look at the st(3) periodic Toda lattice and some of its variants. For
a generalization to other Lie algebras, Lax equations and a physical interpretation see [OP2]
and [AM8]. For the non-periodic case, see [FH].
7.1. Different forms of the Toda lattice
Consider the following Poisson matrix on C:
0 -t1L tI
0
-tT T)
0
where T
-t3
t2 0
t3
-t2
0
we denote the corresponding Poisson structure C6 by 1-, -1. One easily find that t1t2t3
on
and t4 + t5 + t6 are Casimirs. We show that the algebra of Casimirs is generated by these
two elements, in particular it is a polynomial algebra.
Lemma7.1 Cas(C6,1.,.I)=C[tlt2t3,t4+t5+t6l-
Proof
Let denote a
us t1t2t3 and b
=
t4 + t5 + t6. Then F E Cas(M, 1., -1) can be written in
=
terms of tj I... I t5 and the Casimir b by replacing t6 with b t4 t5- We call the resulting - -
polynomial Fl. Since F is a Casimir the same holds true for F, and we find ftom It,, F, I
It2: F1 I 0 that
=
oViL OF,
= = 0'
ji5 54
hence F, depends on t1, t2 and t3 only. If F, is symmetric in these variables it is a polynomial
in VI tl +t2 +t3 and V2
=
t1t2 +t2t3 +t3t1 and V3
=
t1t2t3, so FI(tl,t2, W F2(vl, V2, V3)- =
Since F2 and v3 are Casimirs we find from 14, F21 0 that
OF2 9F2
Yv l (9V2
Since both derivatives are polynomials in vi only it follows from this that these derivatives
are actually zero, hence F, is a polynomial in t1t2t3 only. If F, is not symmetric then one
symmetrizes it,
F2(Vli V27 V3) =

Fl(tli t21 t3) + Fl(t21 t31 t1) + Fl(tO17 Wi
and as above one finds that F2 depends On V3 only. This implies that F, is a polynomial in
t1t2t3 only and we are done. M
Recall that we computed the invariant polynomial of this Poisson structure in Exam-
ple 11.2.56. We define
T1 =
t1t2t3i
T2 =
t4 + t5 + t61
1
T3 =
2
(t24 + t25 + t2)
6 + t, + t2 + t31
T4 =
t4t5t6 -
t1t4 -
t2t5 -
W6
235
and A = C [TI, T2, T3, T4]. Since T1 and T2 axe Casimirs and JT3, T41 = 0 the algebra A
is involutive. We do not show completeness here because it will follow automatically from
Paragraph 7.2.
Let us look at the level sets of the Casimirs over closed points. They are given by
t1t2t3 =
a,
t4 + t5 + t6 =
b,
where a, b E C are arbitrary; we denote this level set by ab If a :A 0 then -

1,r_7
ab is irreducible,
four-dimensional and the Poisson structure has rank four (even at every If a 0 point). =
then Yab has three irreducible components, each of which is a four-dimensional plane and
the Poisson structure has rank four. It follows as in Proposition 11.2.42 that if we restrict
the Poisson structure to any of these levels then its algebra of Casimirs is still maximal.
Similarly the integrable algebra leads to an integrable Hamiltonian system on (the irreducible
components) of these level sets, as in Proposition 11.3.19. The original Toda lattice corrsponds
to a = I and b =
0, but often it is just as easy to work on the larger space C6, or on the
hyperplane W C C' defined by t4 + t,9 + t6 0. One may however also want to consider
=
all possible levels of the Casimirs at once, except the reducible ones, i.e., the ones for which
a = 0. This is easily done by using Proposition 11.3.29. Then the phase space is taken as the
affine variety defined by t0t1t2t3 = I in C7. Since t1t2t3 is a Casimir the same holds true
for to. The algebra of Casimirs is now given by
CaS(C7) =
C[TI0, T1, T2]/idl(TOTI -
1)
and the corresponding integrable algebra is the tensor product of it with C [T3, T4]. We will
come back to this form of the Toda lattice in Paragraph 7.2.
As is even apparent from the physical origin of the problem, the Toda lattice has an
automorphism of order three, which is given by the map -T : C6 _+ C6,
7' : (t1i t21 t31 t41 t5i t6) =

(t21 t31 t1i t57 t67 t4)-
As preparation of the computation of the quotient by this automorphism (in Paragraph 7.3)
a
we a simple linear transformation giving coordinates which are diagonal for the action of r.
do
Let c denote a fixed cubic root of unity and define
X1 =
tl + t2 + t37 Y1 =
t4 + t5 + t67
2
X2 =
t1 + d2 + 'E t31 Y2 =
t4 + d5 + 1E2t6l
2
X3 =
tl + 6 t2 + 'Et37 Y3 =
t4 + 62t,5 + ft6-
Then -T is in diagonal form and is given by
'r(XI, X2, X31 Y11 Y2, Y3) =

(XI, 62X20EX3 Yl,62Y21fY3)-
The integrable algebra A is now given by A = C [XI I X2, X3 X4] where

7
3 3
X, = X3I + X2 + X3 -
3xlX2X3,
X2 =
Yli
X3 =
3xi + YMi
X4 =
Y23 + Y33 + 9(X2Y3 + X3Y2)-
236
T. The Tbda lattice
The Poisson structure is (up to a factor vr--3) given in these coordinates by
-X2
0
-tx
X
0 ) where X=
(0 0
0
-X3
-XI
X3)
X1
X2
and the algebra of Casimirs reduces to C[XI, X2].
7.2. A morphism to the genus 2 even Mumford system
We now morphism from the sl(3) periodic Toda lattice to the Bechlivanidis-
describe a
van Moerbeke system which we have shown to be isomorphic to the genus 2 even Mumford
system (on C7). For a generalization, giving a morphism. from the sr(g + 1) periodic Toda
lattice to the genus g even Mumford system, see [FV]. For explicitness we recall that we
consider the Toda lattice as being defined on the hyperplane W defined by 4 + t5 + t6 0 =
in C6. If we do not suppose this then we do not have a morphism to the Bechlivanidis-van
Moerbeke system, but still to the genus 2 even Mumford system, which we now have to take
as being defined on C', namely we cannot assume anymore that the coefficient of X5 in the
equation y2 f (x) vanishes. Since the formulas are simpler in this case and since there
= is
no phenomenologic difference we will not consider this more general case.
Consider the map 0: -H C C6 -+ C7 defined by
81 =
W6 -
t1i
S4,5 =
(t2 + W/8,
82 =
-4/2,
86,7 =
(tA t2t5)/8.
83 =
-t2t3/16,
Then
0*(s, -
4s 2-
2 8S4) =
-T3i
(3182 + 436) =
-T4/2,
(,q24 _ '925 + SO =
07 (7-1)
0*(S2-53 + 84S6 + 307) =
01
0*(827 _ 826 -
3183) =
-T1116.
Thus 0 is a regular map and O*A! A where X denotes the Bechlivanidis-van Moerbeke
=
system. Moreover it is easy to check that the Poisson structure which has to be taken on
the latter system in order for 0 to be a Poisson map is the one given by (2.14) and we have
obtained a morphism of integrable Hamiltonian systems.
As in the H6non-Heiles example this morphism is again neither injective nor surjective.
As first attempt to cure the non-surjectivity we restrict the Bechlivanidis-van Moerbeke
a
system to the level set Y (of the Casimirs) given by
824 _
82+
5 53 =
07
3233 + -9486 + 85-97 =
01
which is irreducible and of codimension two, hence it is an affine Poisson variety; clearly
it contains the image of 0. Let us compute the invariant polynomial p(Y) of the Poisson
bracket on F. The matrix 2.14 has rank 4 for X3 :A 0, hence by irreducibility of F and since
237
Chapter VII. Two-dimensional a.c.i. systems
dim.F =
5, the leading term of p(.F) is RIS5. The rank of (2.14) caai only go down at those
points where 83 = 0 and (at the same time) the rank of the matrix
-4(s586 + 507) 2s7 -286 2sis5 + 482S7 4S2S6 -

2sIS4
0 -S5 -S4
(7.2)
S7 86
2
is less than 2, i.e., all 2 x 2 determinants must vanish. Since on.F S4
,
_
S25 + 83 = 0 it Splits
up in the cases S5 = =LS4 and we are left with
4S4(87 =L S6) 287 -2s6 4S2S7 + 2s,S4 432S6

( 2s,S4
-
0 T84 -S4 S7 S6
If S7 S6 = 0 then the rank of this matrix is smaller than 2 if and only if
-2s6 :F2(2S236 SIS4)

( )
-
det 2(S26 -
S4(2S2S6 -
SIS4)) = 0-
-84 T86
This means that have found two irreducible components which are defined (in C7) by 53
we
2
0, 85 =
54, s7 =
-54(2S286 SIS4); each of these components has dimension 3.
zFs6 and 36 _ -
If S7 :L S6 :A 0 then the rank of (7.2) can only be smaller than 2 if S4 = S5 86 = s7 0, a = =
locus which is contained in both irreducible components, hence this one does not contribute
to p(,F). Finally the rank is zero if and only if S3 = S4 S5 = S6 S7 0, a two-dimensional = = =
plane inside.F. The upshot is that p(.F) is given by
p(.F) = R 2S5 + 2RS3 + S2.
This should be compared with the invariant polynomial for the Poisson structure of the Toda
lattice W, which we denote by p(W). The invariant polynomial for the Poisson structure of
on
the Toda lattice on C6 was found in Example 11.2-56 and computed as R2S6 + 3RS4 + S3; it
follows at once (devide by S) that its restriction to the hyperplane W (given by t4 + t5 +t6 0) =
equals
R2S5 + 3RS3 + S2.
The small difference between p(Y) and p(W) suffices to conclude that 0 is not a biregular
map (although it is regular and dominant). The polynomials actually indicate that something
goes wrong at the rank 2 level. Let us denote the subsets of.F (resp. W) were the rank is at
most 2 by.Fj (resp. WI). Thus 0 restrictS33 to a regular map 01 : W1 -+.Tj and W., has three
irreducible components whileY, has two. Thus either (at least) two irreducible components
are identified or (at least) one of them is mapped completely inside 770, the locus inside.F of
points where the rank is zero. On the component t2 =

t3 = 0 Of W1 the map 0 is given by
81 =
W6 -
t1i
S2 =
-4/2,
33 =
-94 =
85 =
86 =
57 =
07
34
hence the latter case occurs and we conclude that 0 is not a finite map.
33
Recall that the rank at an image point is never larger than the rank at the point, hence
restricts indeed.
34
The other two components of W, are mapped neatly to the two components of F1.
238
7. The Tbda lattice
Our way to deal with this is to cut away the bad piece: from the Toda side we remove the
(reducible) divisor of T, tIt2t3 and from the Bechlivanidis-van Moerbeke side we remove
:
2_
the zero locus of S5 =
87 862_813,3: it follows from (7. 1) that these correspond under 0. Since
both T, and S,5 axe Casimirs we obtain from Proposition 11.3.29 two integrable Hamiltonian
systems (MI, 1-, .11, AI) and (M2 J* C12 A2)
i 7
where MI and M2 are given by
M, =
J(toltll ...
I W I t0t1t2t3 =
Ii t4 + t5 + t6 =
017
2_ 2_
M2 =
J(50i 811 ... 787) 1 50(87 86 SISO =
11 824 _ '925 +S3 =
8253 + 8486 + 3587 =
017
and I-, .1i and Ai obtained

accordingly. The morphism MI -- M2 which corresponds to
are
0 (and which is also

morphism of integrable Hamiltonian systems) will be denoted by the
a
same letter 0. Now both invariant polynomials are the same (being given by R 2S5), however
0 : MI -+ M2 is not surJective, since it is obviously missing the points of M2 where s3

vanishes. Since 83 does not belong to A2 and (M2 I' *12) A2) satisfies the conditions of i 1
Proposition 11.3.7, removing the zero locus of s3 leads toanother integrable Hamiltonian
system(M3 1" 1'13 A3) where M3 is given by
, 1
M3 I(SO 7 50 1 SI 7 ... ISO I SO(S 2_s2_

7 6 SIS3) --::: SI0 S3 7--
17 s24 _ S25 +S3 82S3+S4S6+S5S7 =
0},
and f 13 and A3 derive at once from f' 12 and A2 -
Finally we have a biregular map! It is explicitly given by
so =
-16to, to =
-so116,
s'0 =
-16toti, t, = s'0 (s 2_
7 82)
6 -81,
81 =
t5t6 -
t1i t2 =
4(84 + 807
82 =
-4/2, t3 =
4(84 -
85)7
33 -t2t3/16, t4 =
-2S2)
S4,5 (t2 t3)/8, t5 =
-80(S4 -
35)(86 -
S7)7
85,6 (t3t6 t2t5)/8 t6 =
-SO(S4 + 35)(86 + 87)-
Now it follows at once that the Toda lattice (on W) is a.c.i.: since the Bechlivianidis-van
Moerbeke system is a.c.i. the same is true for the system we constructed on M3 since we
just removed a divisor (with two components), however the level sets have changed since we
removed the divisor of a belong to the integrable algebra. Thus our

function which does not
system on MI is a.c.i. and since it contains the general level set of the original Toda lattice
(on W) as its general level set, the Toda lattice is a.c.i. and its general level set is an affine part
of a Jacobian; using the order three automorphism one easily recovers the well-known fact
that this affine part is obtained from the Jacobian by removing three translates of a genus
two curve, earch pair of which is tangent at their intersection point. Each of these curves
induces a principal polarisation on its Jacobian, hence the Toda lattice is an a.c.i. system of
polarization type (3,3).
239
7.3. Toda and Abelian surfaces of type (1,3)
Having shown that the Toda lattice is a.c.i. of polarization type (3,3) we are now ready
to construct a new a.c.i. system of polarization type (1, 3). Using Proposition V.2.5 it is
obtained as follows: the order three automorphism r fixes all fibers of the momentum map
(since r*f f for all f E A). Therefore the general level set of the quotient a.c.i. system
=
is isogeneous to a Jacobian and the isogeny is a 3 : 1 (unramified) map, showing that this
level set carries a polarization of type (1,3). It is also easy to determine the divisor which
is to be adjoined to the general fiber (of the momentum map) in the quotient system: it
is the quotient of the Toda divisor by T which acts as a translation permuting the three
components of this divisor (and the three intersection points). Hence the quotient is an
irreducible divisor, birationally equivalent to the genus two curve and having one singular
point which is a tacnode. It follows from IV.3.4 that its virtual genus equals 4 which is by
Theorem IV.3.7 consistent with the fact that it induces a polarization of type (1,3) on the
surface.
If one an explicit realization of the quotient system, then the main object
wants to have
to be computed C6 1-r (or Wft): the Poisson structure and the involutive algebra are
is
obtained at once. Thus we need to construct a system of generators of for O(C61-r) (or
O(Rl-r)) as well as a generating set of relations between these generators. To do this, we use
the coordinates xi, yi, constructed in Paragraph 7.1, which are diagonal with respect to the
action of -r. Obviously the following elements are invariant with respect to r.
X1 X1, Y1 =
Y1,
X2 X2X3i Y2 -=
M3i X3 =
X021 Y3 =
X2Y37
X4 =
3, y4
X2 =
Y23, X5 = X3,
3
y5 =
Y33,
2
X6 = X22Y2i Y6 =
X2Y22, X7 = X23Y3 Y7 ,
=
X3Y3
We claim that these fourteen elements generate O(C61-r). To show this, let F be an invariant
polynomial. Writing it as a polynomial in x, and y, it suffices to show that an invariant poly-
1101nial in X27 X31 y2 and Y3 can be written as a polynomial in the above candidate generators.
Because of the elements X4, Y4, X5 and Y,5 it suffices to check this for a polynomial of degree
less than three in X2, X31 Y2 and Y3- Since the action is diagonal it suffices to check it for
monomials; there are 27 of these, namely the monomials
2 Yi4
XilXi2Yi3 (mod 3).
with ii + N2 + i3 + 24 = 0
2 3 3
It is easy to check that they all depend on the fourteen ones above.
Finding all relations requires more work: there are quite a lot of them. We give half
of the explicit list, the other half is found by interchanging X and Y in our list; our list is
ordered by the degree of the monomials (in the xi, yi) they come from. In order four there is
just one,
X2Y2 =
X3y3i
in order five there are (two times) six,
X3X4 =
X2X6i X2Y4 =
X3y6i
X2X7 =
x5y3i X6Y2 =
Y3Y61
X2Y5 =
Y3Y77 X7Y2 =
X3Y71
240
7. The Toda lattice
finally, in order six there are (ten plus) twelve,
X32 3
X4X51 X20Y3 X4X7i
2
X33 X5Y41 X2Y --X4y7i
X4Y4:-` X6Y6; X A2X3 X6 X6
X5Y5 X7y7i X2X3Y3 X6X71
2
X6' Y4 Y6) X2Y2Y3 X6Y7)
X27 X5y7, X2 X2=yy
5 '6-
3
More important than this list is how to prove that it is complete; we do this as before
by
reducing it to a finite list. Every relation must come from all identity in the variables xi, yi,
since x, and yj are invariant themselves we may forget about these as before. Second, since
the variables xi and yj are diagonal for the action these identities come from identities between
monomials, i.e., from identities which are written in terms of the xi, yj as
il ii ki 11 Xi2 j2 k2 12 i3 j3 k3 13 4 j4 k4 14
X2 X3 Y2 Y3 .
2 X3 Y2 Y3
:==
X2 X3 Y2 Y3 *
X2 X3 Y2 Y3 (7.3)
with i, + 2j, + k, + 21, 0 (mod 3) by invariance (s 1, All powers in this equation

, 4).
= =
. . .
may be supposed smaller than three; to check this for the powers Of X2, suppose that X2
appears on the left (hence also on the right) with a power at least three, then this must come
from one of the following terms (or a multiple of it):
2 2 2
x3'
2 X2'Y35 X2'y6i X2Y 2, X2y3y6i X2 y2'
6 Yj3, Y Y6 Y3 Yd2, y3.
r i
Now check that by using the relations one can always factorize X4 (on both sides) thereby
reducing the order, for example
2
X22y
',6 =
X2X3X6 ==
X3X4-
Since now all exponents in (7.3) may be supposed smaller than three we have a finite list to
check. In this way it is verified that all relations are a consequence of the relations which we
have given and we have an explicit description of the ax.i. system.
241
Index
Abel's Theorem, 119 Chern class, 110
Abelian variety, 108 closed point, 37

-
dual, 110 compatible
-
(principally) polarized, 108 -
Poisson brackets, 25
-
reducible, 109 -
integrable Hamiltonian systems, 62
Abel-Jacobi map, 119 completable a.c.i. system, 131

adjunction formula, Ill complete algebra, 47
affine Lie-Poisson group, 30 completion, 48
a.c.i. system, 131 complex torus, 108

-
algebraic completely integrable system, 131
-
completable, 131 Darboux
-
irreducible, 132 -
coordinates, 66
-
(polarization) type of, 132 -

Theorem, 66
analytic Poisson manifold, 65 deformation property, 25
ample line bundle, 105 degrees of freedom, 49
dimension, 37, 49
base space, 49 divisor
Ber-hlivanidis-van Moerbeke system, 181 -

canonical, 103
bi-Hamiltonian -
elementary, 108
-
hierarchy, 63 -
degree of, 99
-
integrable system, 62 -
group, 99
-
vector field, 62 -
pole, 99
branch point, 106 -
zero, 99
double Lie algebra, 140
canonical dual Abelian variety, 110

-
brackets, 66 Garnier potential, 196

-
coordinates, 66 general
-
divisor, 103 -
fiber, 37
-
coordinates, 66 - level set, 37
- line bundle, 103 -
point, 37
Casimir
-
decomposition, 38 Hamiltonian, 19
-
function, 19, 65 -
derivation, 19
- level set of, 38 -

vector field, 19, 65
253
Index
H6non-Heiles K-3 surface, 121

-
hierarchy, 233 Kodaira
-
potential, 230, 232, 233 -
Serre duality, 103
hyperelliptic -
vanishing Theorem, 113
-
case, 83 Kummer
curve, 106 -
generalized surface, 121
-
involution, 107 -
intermediate surface, 121
hierarchy -
surface, 121
-
bi-Hamiltonian, 63 -
variety, 121
-
H6non-Heiles, 233 Krull dimension, 37
Hodge form, 105
Laurent solutions, 136

indicial equations, 136 Lax
integrable -
equation, 141
algebra, 49, 69 -
equation with spectral parameter, 141
bi-Hamiltonian system, 62 -
representation, 142
Hamiltonian system, 49, 69 -

type, 141
multi-Hamiltonian system, 62 level set
vector field, 49, 69 -

general, 37
integral - of the Casimirs, 38
-
closure, 47 - of the integrable system, 50
-
curve, 68 Lie derivative, 20
-
first, 1 Lie-Poisson
invariant of affine Poisson variety - affine group, 30
-
polynomial, 41 -
structure, 23
-
matrix, 41 - modified structure, 25
involutive line bundle
-
algebra, 47 -
ample, 105
- Hamiltonian system, 47 -
canonical, 103
irreducible a.c.i. system, 132 -

holomorphic, 100
isogeny, 109 -
very ample, 105

lower balances, 136
Jacobian, 114
Jacobi maximal
- inversion Theorem, 120 -

algebra of Casimirs, 39
- Mumford system, 143 -

spectrum, 37
-
surface, 121 momentum map, 50
variety, 114
254
Index
morphism -
ideal, 29
-
of affine Poisson varieties, 26 -
isomorphism, 27, 68
- of integrable Hamiltonian systems, 54, 70 -
manifold, 65
- of Poisson spaces, 68 -
matrix, 21
multipliers, 109 -
morphism, 26, 68
multi-Hamiltonian -
product bracket, 30
-
integrable system, 62 -
reducible, 32
-
vector field, 62 -
space, 65
Mumford. system -
standard structure, 23
-
even, 163, 179 -

structure, 19, 65
-
odd, 161, 177 -
subalgebra, 31
-
trivial structure, 22
node, 122 -
vector field, 20
Noether's formula, 113 Poincax6
-
dual, 111
Painlev6 analysis, 136 -

reducibility theorem, 109
parameter pseudo-differential operator

-
map, 38 -
algebra of, 145
-
space, 38 -
monic, 145
phase space, 49 -
normalized, 145
period -
order of, 145
-
half, 121 polarized Abelian vaxiety, 108
-
matrix of A periods, 115 polarization
- matrix of B periods, 115 -
principal, 108
-
i-th period, 115 -
type, 108
-
lattice, 114 -
type of a.c.i. system, 132

Picard group, 100 principal balance, 136
Poisson principally polarized Abelian variety, 108
-
action, 31
-
algebra, 19 quasi-automorphism, 61
- affine brackets, 25
- affine subvariety, 27 ramification point, 87

-
algebra, 19 rank
-
analytic manifold, 65 - of Poisson structure, 22, 69
-
bracket, 19, 65 - rank decomposition, 41
- canonical structure, 23 reciprocity laws, 115

-
cohomology, 20 real level sets, 85
-
constant structure, 23 reducible Abelian variety, 109
255
Index
regular Poisson structure, 22 Toda, lattice
Pdemann -
generafized, 141
-
conditions, 108 -
three body, 235
-
constant, 120 trope, 122
-
Hurwitz formula, 106 trivial
-
Roch Theorem, 103, 112 -
Poisson structure, 22
-
theta divisor, 110 -
subsystem, 58
- theta function, 110
vector field
Sato Grassmannian, 145 -
bi-Hamiltonian, 62
Schouten bracket, 20 -
Hamiltonian, 19, 65
Schotky problem, 118 -
integrable, 49, 69
seven-dimensionaJ system, 181 -

KP, 152
spectral -
linear, 131
-
parameter, 141 -
multi-Hamiltoinian, 62
-
curve, 142 -
Poisson, 20
symplectic -
super-integrable, 49
-
basis, 115 virtual genus, 112

-
manifold, 65 Volterra group, 145
-
two-form, 65
-
decomposition, 67 Weierstrass point, 107
-
foliation, 67 weight homogeneous, 191
super-integrable vector field, 49
Yang-Baxter equation
type of a.c.i. system, 132 -
classical, 140
theta -
modified classical, 140
-
curve, 122
-
function, 110
- function with characteristics, 118

-
divisor, 110
256

Integrable Systems in The Realm OfAlgebraic Geometry

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Integrable Systems in The Realm OfAlgebraic Geometry

Uploaded by

Copyright:

Available Formats

Lecture Notes in Mathematics 1638

Cataloging-in-Publication Data applied for

Die Deutsche Bibliothek -

ed.. Berlin ; Heidelberg New York ; Barcelona

; Hong Kong ; London

Mathematics Subject Classification (2000): 14K20, 14H70, 17B63, 37J35

ISSN 0075- 8434

Springer-Verlag Berlin Heidelberg New York

publication does not

for general use.

Typesetting: Camera-ready TEX output by the author

For useful comments on the manuscript I am indebted to Mich6le Audin, an anonymous

IT. Integrable Hamiltonian systems on affine Poisson varieties . . . . . .

2. Affine Poisson varieties and their morphisms . . . . . . . . . . . . . . . .

2.2. Morphisms of affine Poisson varieties . . . . . . . . . . . . . . . . .

3. Integrable Hamiltonian systems and their morphisms . . . . . . . . . . . .

3.1. Integrable Hamiltonian systems on affine Poisson varieties . . . . . . . .

3.3. Constructions of integrable Hamiltonian systems . . . . . . . . . . . .

4. Integrable Hamiltonian systems on other spaces . . . . . . . . . . . . . . .

4.1. Poisson spaces . . . . . . . . . . . . . . . . . . . . . . . . . . .

111. Integrable Hamiltonian systems and symmetric products of curves .

2. The systems and their integrability . . . . . . . . . . . . . . . . . . . .

2.3. PolynomiaJs in involution for I., -1d' 78.

2.4. The hyperelliptic case . . . . . . . . . . . . . . . . . . . . . . . .

3. The geometry of the level manifolds . . . . . . . . . . . . . . . . . . . .

3.1. The real and complex level sets . . . . . . . . . . . . . . . . . . . .

3.2. The structure of the complex level manifolds . . . . . . . . . . . . . .

3.3. The structure of the real level manifolds . . . . . . . . . . . . . . . .

3.4. Compactification of the complex level manifolds . . . . . . . . . . . .

3.5. The significance of the Poisson structures -j'Pd . .

2. Divisors and line bundles . . . . . . . . . . . . . . . . . . . . . . . .

2.2. Line bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.3. Sections of line bundles . . . . . . . . . . . . . . . . . . . . . . .

2.4. The Riemann-Roch Theorem . . . . . . . . . . . . . . . . . . . . .

2.5. Line bundles and embeddings in projective space . . . . . . . . . . . .

2.6. Hyperelliptic curves . . . . . . . . . . . . . . . . . . . . . . . . .

3.1. Complex tori and Abelian varieties . . . . . . . . . . . . . . . . . .

3.2. Line bundles on Abelian varieties . . . . . . . . . . . . . . . . . . .

3.3. Abelian surfaces . . . . . . . . . . . . . . . . . . . . . . . . . .

4.1. The algebraic Jacobian . . . . . . . . . . . . . . . . . . . . . . .

4.2. The analytic/transcendental Jacobian . . . . . . . . . . . . . . . . .

4.3. Abel's Theorem and Jacobi inversion . . . . . . . . . . . . . . . . .

4.4. Jacobi and Kummer surfaces . . . . . . . . . . . . . . . . . . . . .

5. Abelian surfaces of type (1,4) . . . . . . . . . . . . . . . . . . . . . . .

5.1. The generic case . . . . . . . . . . . . . . . . . . . . . . . . . .

5.2. The non-generic case . . . . . . . . . . . . . . . . . . . . . . . .

V. Algebraic completely integrable Hamiltonian systems . . . . . . . .

3. Painlev6 analysis for a.c.i. systems . . . . . . . . . . . . . . . . . . . .

4. The linearization of two-dimensional a.c.i. systems . . . . . . . . . . . . .

VI. The Mumford systems . . . . . . . . . . . . . . . . . . . . .

2.1. The algebra of pseudo-differential operators . . . . . . . . . . . . . .

2.2. The matrix associated to two commuting operators . . . . . . . . . . .

2.3. The inverse construction . . . . . . . . . . . . . . . . . . . . . . .

2.4. The KP vector fields . . . . . . . . . . . . . . . . . . . . . . . .

3.1. The loop algebra 91, . . . . . . . . . . . . . . . . . . . . .

3.2. Pteducing the R-brackets and the vector field V . . . . . . . . . . . 157.

4. The odd and the even Mumford systems . . . . . . . . . . . . . . .

4.1. The (odd) Mumford system . . . . . . . . . . . . . . . . . . . . .

4.3. Algebraic complete integrability and Laurent solutions . . . . . . . . . .

VII. Two-dimensional a.c.i. systems and applications . . . . . . . 175.

2. The genus two Mumford. systems . . . . . . . . . . . . . . . 177.

2.1. The genus two odd Mumford, system . . . . . . . . . . . . . . .

2.2. The genus two even Mumford system . . . . . . . . . . . . . . . . .

the hat denoting the natural pairing (,g)