Mathematical Sciences
Volume4
EditorinChief: R.V. Gamkrelidze
SpringerVerlag Berlin Heidelberg GmbH
V.I. Arnold S.P. Novikov (Eds.)
Dynamical Systems N
Symplectic Geometry and its Applications
Second, Expanded and Revised Edition
With 62 Figures
Springer
Consulting Editors of the Series: N.M. Ostianu, L.S. Pontryagin
Scientific Editors of the Series:
A.A. Agrachev, Z.A. Izmailova, V.V. Nikulin, V.P. Sakharova
Scientific Adviser: M.I. Levshtein
Title of the Russian edition:
Itogi nauki i tekhniki, Sovremennye problemy matematiki,
Fundamental'nye napravleniya, VoI. 4, Dinamicheskie sistemy 4
Publisher VINITI, Moscow 1985
Mathematics Subject Classification (2000):
58F05, 58F06,58F07, 70Hxx
ISSN 09380396
ISBN 9783642082979 ISBN 9783662067918 (eBook)
DOI 10.1007/9783662067918
This work is subject to copyright. AII rights are reserved, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of iIIustrations, recitation, broadcasting, reproduction on
microfilm or in any other way, and storage in data banb. 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 SpringerV erlag Berlin Heidelberg GmbH.
Violations are Iiable for prosecution under the
German Copyright Law.
http://www.springer.de
© SpringerVerlag Berlin Heidelberg 1990,2001
Originally published by SpringerVerlag Berlin Heidelberg New York in 2001
Softcover reprint of the hardcover 2nd edition 2001
Typesetting: Macmillan India Limited. Bangalore25
Printed on acidfree paper SPIN 10558411 46/3142LK  5 4 3 2 1 O
List of Editors, Authors and Translators
EditorinChief
R.V. Gamkrelidze, Russian Academy of Sciences, Steklov Mathematical Institute,
ul. Gubkina 8, 117966 Moscow; Institute for Scientific Information (VINITI),
ul. Usievicha 20a, 125219 Moscow, Russia; email: gam@ipsun.ras.ru
Consulting Editors
V .1. Arnold, Steklov Mathematical Institute, ul. Gubkina 8, 117966 Moscow,
Russia, email: arnold@genesis.mi.ras.ru; CEREMADE, Universite Paris 9
Dauphine, Place du Marechal de Lattre de Tassigny, 75775 Paris Cedex 16e,
France, email: arnold@ceremade.dauphine.fr
S.P. Novikov, Institute of Physical Sciences and Technology, University of Maryland,
College Park, MD 207422431, USA, email: novikov@ipst.umd.edu;
L.D. Landau Institute for Theoretical Physics, Russian Academy of Sciences,
2 Kosygina str., 117940 Moscow, Russsia, email: novikov@landau.ac.ru
Contributors
V .1. Arnold, Steklov Mathematical Institute, ul. Gubkina 8, 117966 Moscow,
Russia, email: arnold@genesis.mi.ras.ru; CEREMADE, Universite Paris 9
Dauphine, Place du Marechal de Lattre de Tassigny, 75775 Paris Cedex 16e,
France, email: arnold@ceremade.dauphine.fr
B.A. Dubrovin, International School for Advanced Studies (SISSA), Via Beirut 24,
34013 Trieste, Italy, email: dubrovin @sissa.it
A.B. Givental, Department of Mathematics, University of California at Berkeley,
Berkeley, CA 94720, USA, email: givental@math.berkeley.edu;
Dept. of Mathematics, California Institute of Technology, CA 91125, USA
A.A. Kirillov, Department of Mathematics, Pennsylvania State University,
Philadelphia, PA 191046395, USA, email:kirillov@math.upenn.edu
I.M. Krichever, Landau Institute for Theoretical Physics, 2 Kosygina str.,
117940 Moscow, email: igor!igor@cpd.landau.ac.ru; Department of Mathe
matics, Columbia University, 2990 Broadway, New York, NY 10027, USA,
email: krichev@m~th.columbia.edu
S.P. Novikov, lnstituteofPhysical Sciences and Technology, University of Maryland,
College Park, MD 207422431, USA, email: novikov@ipst.umd.edu;
L.D. Landau Institute for Theoretical Physics, Russian Academy of Sciences,
2 Kosygina str., 117940 Moscow, Russsia, email: novikov@landau.ac.ru
Translators
A. Dzhamay, Department of Mathematics, The University of Michigan, 2074 East
Hall, 525 East University Avenue, Ann Arbor, MI 48109, USA;
email: adzham @math.lsa.umich.edu
G. Wassermann, Fakultiit und Institut fur Mathematik, RuhrUniversitiit Bochum,
UniversitiitsstraBe 150, 44780 Bochum, Germany
Contents
Symplectic Geometry
V.I. Arnol'd, A.B. Givental'
1
Geometric Quantization
A.A. Kirillov
139
Integrable Systems. I
B.A. Dubrovin, I.M. Krichever, S.P. Novikov
177
Index
333
... . The Symplectic Group and its Lie Algebra.. . . . . . 12 3. .. . Linear Hamiltonian Systems . .. The Symplectic Group. the Lagrangian Grassmann Manifold. . .. Subgroups of the Symplectic Group . . . . . . .. . ... . .. .. . .. . . 7 2.. . . . . .. .. . ... .. . ..1. ....... . 19 4.. .. . . .... Wassermann Contents Foreword. . .. . .. . . . .. . 18 4. . . . . . . . . . . .. . . . .. . . Symplectic Geometry V. . 13 3. . 7 2. . . . . . . . . .3. . . . . ... . 5 1. .. . Linear Variational Problems. 8 2. . . . . .. .. . B. . 5 1. . .. .2. .3.. .. Families of Quadratic Hamiltonians .... .2. .. . . . .. .. ... .. 9 2. . . . .. Linear Symplectic Geometry ... . .. . . . . .. . . . ... . . . . .. . . . .. . . . .. .. . . .. . .. . . . ... . .. The Complex Classification of Hamiltonians.. . ... . . . Givental' Translated from the Russian by G. .. 11 § 3. .. 17 4.2. . . . . . .. . 6 § 2. . . .. .2.. . .. .1. .3. .. .. . . .. . The Exponential Mapping and the Cayley Parametrization. .5. . . . .... . . .. 12 3. . . . . . .. ... ... . .. . . . . . .. 4 Chapter 1. . . .. 10 2. .1.. 18 4. . . .. .. SignDefinite Hamiltonians and the Minimax Principle .4. . . . . . ..4.. . . . . . .. . .. . . Linear Hamiltonian Systems with Periodic Coefficients . . Generic Families. ... . . . 16 §4. . 17 4.. ... . 5 § 1. . . .. 14 3.. . .. .4. ...3. . .. . . . .. The Spectrum of a Symplectic Transformation. . ... . . A. . . ..5.. .. .. . ... . ...... 5 1. . .. .. . . .. .. .. . 19 . . . . . . . .. The Concept of the Miniversal Deformation. The SkewScalar Product .. .1. .. . .. . ..... . . . .. . . . . . . . The Topology of the Symplectic Group . Normal Forms of Real Quadratic Hamiltonians. . Arnol'd. . .. . . . . . . Subspaces . . . Bifurcation Diagrams . . . . . Miniversal Deformations of Quadratic Hamiltonians . . . . .. . . . .. . . . . .. . ... Symplectic Space . ... . .. . . . .. . . . . . .. . . .. . . . . . . I. . .. . . . .. . .
. .. . .. . .. .. . . . .6... .6. The Poisson Bracket . The Principle of Least Action . ... .1. . . . . .. .. . . . . .. . . . . .. . . .4. . . .. . . . . . . . . . . . . . The Linearization Problem . . . . . . .. .. .. Poisson Pairs. . . . . . . . 30 § 3. The Complex Case . .4. . .. . . . .. .. .. . 59 §3.. ... . . .. . . . . . . . . .. . The Shortest Way Around an Obstacle. . . . . . . .3. . .... . . . . .4. . ..1. . . . . . 42 § 1. . . . . . . . .. . 42 1. . . .... . ... . . Completely Integrable Systems. . . The Reduced Phase Space and Reduced Hamil~onians. . . . 33 3. . . . . . . 24 1. . ... .. . .. Symplectic Manifolds . . . . . . . 35 4. . . . .. .. Variational Problems with Higher Derivatives . .. . 45 1. .. . 63 . . ... . ... . . . . . . . . . . . . . . . . . 31 3. Examples of Lagrangian Manifolds. ....3.. . . .1. . . . . . . . .. Lagrangian Submanifolds and Fibrations . . . . .3. . . . .. . . . . . . . . The Degeneracies of Oosed 2Forms on R4 • • • • • • • • 23 1. .. . .. . .. . . . . .. . . . . .. 49 §2. . . . . . .. . ..... .. . . The Darboux Theorem . . . . . . .2. Elliptical Coordinates and Geodesics on an Ellipsoid . ... . . . ... . . .. . . . 26 1. . .. . . . . .. . Examples of Symplectic Manifolds... . . . . . . . . . . .. A. .. . . .1..... . . 28 2. .5.. . .. . Example. .. . ..2 V. Symplectic and Kahler Manifolds . .. . . . . .. . . . . . . . .. . Poisson Actions and Momentum Mappings. Integrability According to Liouville.3.. . . . . . . .. . 61 3.. The Exterior Geometry of Submanifolds... . . .. . ... .B. ... . . . . 54 2. Functions in Involution on the Orbits of a Lie Coalgebra .. . . . .. . . . Cotangent Bundles . . . . . . 46 1.5.. .... .. . Germs of Submanifolds of Symplectic Space. . .... . . . . .. . . . . .. . . . 51 2. . .. .. ... . . .. . . . . . . . . . . . . . .. . . . . . . .. . . Local Symplectic Geometry .2. . . Complex Projective Manifolds . . . . . . .3... .. . . . The ''ActionAngle" Variables. . . . ...2. . . . . . . . .. ... . . . . . Hidden Symmetries . . . . . . Givental' Chapter 2.. . . .. .. . .. 36 4. .1.. . 27 §2. . .. . 27 2. . ... . . . 44 1.. .. . . .. . .. . .2. . Lagrangian Fibrations.. . . . .. . ..4. ... .. . . . . . . .. ... . . . . . . . . . .. . . . .... . .. . . .2. . . . . . 48 J... .5.. . . .. .. . . . . . .. 22 1. .... . . . .. Hamiltonian Systems with Symmetries.. . . . Symplectic Geometry and Mechanics. 35 4. . . 32 3. . . . . 25 1. . . . . . . . .. . . . . . . .. 22 1. . Variational Principles..2.. 22 § 1. .. 58 2. Hamiltonian Mechanics. . .. .. . . . . . . . . .. .. . ... . . 27 2. . . . . . . . . ... . . . ... .. .. . . . . . .. Poisson Manifolds .. . . 51 2. . . .. . .. . . .I. . . . . . . .. Lagrangian Mechanics.. . . . . ... . . . . . . . .2. . . 43 1. . ..3. .4.. 61 3. . . The Lax Representation . .. .. .. . 38 Chapter 3... . ... . . . . 62 3. ... .. . ... .. . . . . . . . . .. . . . . .. . . . . . . The Orbits of the Coadjoint Action of a Lie Group . ...6. . . .. ... . . .. . . . . . . . . . .. . .. . .1. . . . . . . . . . . . . 31 3. .. .. .. . .. . . . . .1. . . . . . . .. .. .. . . 34 § 4. .. .. . The Lie Algebra of Hamiltonian Functions... . . . . . . The Classification of Submanifold Germs . . .. .. . .. . . . . . The Manifold of CharacteristicS . . . 53 2. .... .. . . . . . .. 57 2. .. . . Arnol'd. . . . 29 2. Intersections of Lagrangian Manifolds and Fixed Points of Symplectomorphisms. .. .3. . . ... Linear Poisson Structures . .. .. . . . .
Symplectic Geometry 3
3.4. Poisson Groups. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.5. Geodesics of LeftInvariant Metrics and the Euler Equation . . . 66
3.6. Relative Equilibria. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.7. Noncommutative Integrability of Hamiltonian Systems . . . . . . . 67
3.8. Poisson Actions of Tori. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Chapter 4. Contact Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
§ 1. Contact Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
1.1. Contact Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
1.2. Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
1.3. The Geometry of the Submanifolds of a Contact Space . . . . . . . 74
1.4. Degeneracies of Differential 1Forms on IR" . . . . . . . . . . . . . . . . 76
§ 2. Symplectification and Contact Hamiltonians . . . . . . . . . . . . . . . . . 77
2.1. Symplectification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.2. The Lie Algebra of Infinitesimal Contactomorphisms. . . . . . . . . 79
2.3. Contactification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.4. Lagrangian Embeddings in lR 2 ". • . • • • • • • • . • • • • • • • • • • • • • • • 81
§ 3. The Method of Characteristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.1. Characteristics on a Hypersurface in a Contact Space . . . . . . . . 82
3.2. The FirstOrder Partial Differential Equation. . . . . . . . . . . . . . . 83
3.3. Geometrical Optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4. The HamiltonJacobi Equation. . . . . . . . . . . . . . . . . . . . . . . . . . 85
Chapter 5. Lagrangian and Legendre Singularities . . . . . . . . . . . . . . . 87
§ 1. Lagrangian and Legendre Mappings . . . . . . . . . . . . . . . . . . . . . . . 87
1.1 Fronts and Legendre Mappings. . . . . . . . . . . . . . . . . . . . . . . . . . 87
1.2. Generating Families of Hypersurfaces. . . . . . . . . . . . . . . . . . . . . 89
1.3. Caustics and Lagrangian Mappings . . . . . . . . . . . . . . . . . . . . . . 91
1.4. Generating Families of Functions . . . . . . . . . . . . . . . . . . . . . . . . 92
1.5. Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
§ 2. The Classification of Critical Points of Functions . . . . . . . . . . . . . 94
2.1. Versal Deformations: An Informal Description. . . . . . . . . . . . . . 94
2.2. Critical Points of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
2.3. Simple Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.4. The Platonics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.5. Miniversal Deformations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
§3. Singularities of Wave Fronts and Caustics................... 99
3.1 The Classification of Singularities of Wave Fronts and Caustics
in Small Dimensions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.2. Boundary Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.3. Weyl Groups and Simple Fronts. . . . . . . . . . . . . . . . . . . . . . . . . 104
3.4. Metamorphoses of Wave Fronts and Caustics . . . . . . . . . . . . . . 106
3.5. Fronts in the Problem of Going Around an Obstacle.. . . . . . . . 109
4 V.I. Amol'd, A.B. Givental'
Chapter 6. Lagrangian and Legendre Cobordisms. . . . . . . . . . . . . . . . 113
§ 1. The Maslov Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
1.1. The Quasiclassical Asymptotics of the Solutions of the Schrodinger
Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
1.2. The Morse Index and the Maslov Index. . . . . . . . . . . . . . . . . . . 115
1.3. The Maslov Index of Closed Curves. . . . . . . . . . . . . . . . . . . . . . 116
1.4. The Lagrangian Grassmann Manifold and the Universal Maslov
Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
1.5. Cobordisms of Wave Fronts on the Plane . . . . . . . . . . . . . . . . . 119
§ 2. Cobordisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
2.1. The Lagrangian and the Legendre Boundary . . . . . . . . . . . . . . . 121
2.2. The Ring of Cobordism Classes . . . . . . . . . . . . . . . . . . . . . . . . . 122
2.3. Vector Bundles with a Trivial Complexification . . . . . . . . . . . . . 122
2.4. Cobordisms of Smooth Manifolds. . . . . . . . . . . . . . . . . . . . . . . . 123
2.5. The Legendre Cobordism Groups as Homotopy Groups . . . . . . 124
2.6. The Lagrangian Cobordism Groups. . . . . . . . . . . . . . . . . . . . . . 125
§ 3. Characteristic Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.1. Characteristic Classes of Vector Bundles ..... : . . . . . . . . . . . . . 126
3.2. The Characteristic Numbers of Cobordism Classes..... . . . . . . 127
3.3. Complexes of Singularities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.4. Coexistence of Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Foreword
Symplectic geometry is the mathematical apparatus of such areas of physics as
classical mechanics, geometrical optics and thermodynamics. Whenever the
equations of a theory can be gotten out of a variational principle, symplectic
geometry clears up and systematizes the relations between the quantities
entering into the theory. Symplectic geometry simplifies and makes perceptible
the frightening formal apparatus of Hamiltonian dynamics and the calculus of
variations in the same way that the ordinary geometry of vector spaces reduces
cumbersome coordinate computations to a small number of simple basic
principles.
In the present survey the simplest fundamental concepts of symplectic
geometry are expounded. The applications of symplectic geometry to mechanics
are discussed in greater detail in volume 3 of this series, and its applications to the
theory of integrable systems and to quantization receive more thorough review
in the articles of A.A. Kirillov and of B.A. Dubrovin, I.M. Krichever and S.P.
Novikov in this volume.
We would like to express our gratitude to Professor G. Wassermann for the
excellent and extremely careful translation.
Symplectic Geometry s
Chapter 1
Linear Symplectic Geometry
§ 1. Symplectic Space
1.1. The SkewScalar Product. By a symplectic structure or a skewscalar
product on a vector space we mean a nondegenerate skewsymmetric bilinear
form. The nondegeneracy of the skewsymmetric form implies that the space
must be evendimensional.
A symplectic structure on the plane is just an area form. The direct sum of n
symplectic planes has a symplectic structure: the skewscalar product of two
vectors is equal to the sum of the areas of the projections onto the n coordinate
planes of the oriented parallelogram which they span.
The Linear "Darboux Theorem". Any two symplectic spaces of the same
dimension are symplectically isomorphic, i.e., there exists a linear isomorphism
between them which preserves the skewscalar product.
Corollary. A symplectic structure on a 2ndimensional vector space has the
form p 1 1\ q1 + ... + Pn 1\ qn in suitable coordinates (p 1 , •.• , Pn• q 1 , • . . , qn).
Such coordinates are called Darboux coordinates, and the space IR 2 " with this
skewscalar product is called the standard symplectic space.
Examples. 1) The imaginary part of a Hermitian form defines a symplectic
structure. With respect to the coordinates zk = Pk + J""~qk on C" the imaginary
L
part of the Hermitian form zkz~ has the form  L
Pk " qk.
2) The direct sum of a vector space with its dual V = X*EBX is equipped with
a canonical symplectic structure w(e Et> x, '7 Et> y) = ~(y) 17(x). If (q 1 , • . . , qn) are
coordinates on X and (p 1 , . . • , Pn) are the dual coordinates on X*, then
(l) = LPk 1\ qk.
The standard symplectic structure on the coordinate space IR 2 " can be
expressed by means of the matrix n = (~n ~n). where En is the unit n x n
L
matrix: w(v, w) = (Qv, w ). Here (v, w) = vk wk is the Euclidean scalar product
on IR 2 ". Multiplication by n defines a complex structure on IR 2 ", since
Q2 = £2n•
1.2. Subspaces. Vectors v, we V for which w(v, w) = 0 are called skew
orthogonal. For an arbitrary subspace of a symplectic space the skeworthogonal
complement is defined, which by virtue of the nondegeneracy of the skewscalar
product does in fact have the complementary dimension, but, unlike the
6 V.I. Amol'd, A.B. Givental'
Euclidean case, may intersect the original subspace. For example, the skew
scalar square of any vector equals 0, and therefore the skeworthogonal
complement of a straight line is a hyperplane which contains that line.
Conversely, the skeworthogonal complement of a hyperplane is a straight line
which coincides with the kernel of the restriction of the symplectic structure to
the hyperplane.
While a subspace of a Euclidean space has only one invariantits dimension,
in symplectic geometry, in addition to the dimension, the rank of the restriction
of the symplectic structure to the subspace is essential. This invariant is trivial
only in the case of a line or a hyperplane. The general situation is described by
The Linear "Relative Darboux Theorem". In a symplectic space, a subspace of
rank 2r and dimension 2r+k is given in suitable Darboux coordinates by the
equations q,+k+ 1 = ... = qn = 0, Pr+ 1 = ... = Pn = 0.
The skeworthogonal complement of such a subspace is given by the equations
q 1 = ... = q, = 0, p 1 = ... = p,+k = 0, and it intersects the original subspace
along the kdimensional kernel of the restriction of the symplectic form.
Subspaces which lie within their skeworthogonalcomplements (i.e. which
have rank 0) are called isotropic. Subspaces which contain their skeworthogonal
complements are called coisotropic. Subspaces which are isotropic and co
isotropic at the same time are called Lagrangian. The dimension of Lagrangian
subspaces is equal to half the dimension of the symplectic space. Lagrangian
subspaces are maximal isotropic subspaces and minimal coisotropic ones.
Lagrangian subspaces play a special role in symplectic geometry.
Examples of Lagrangian Subspaces. 1) In X* ~ X, the subs paces {0} ffi X
and X* ffi { 0} are Lagrangian. 2) A linear operator X + X* is selfadjoint if and
only if its graph in X* ffi X is Lagrangian. To a selfadjoint operator A there
corresponds a quadratic form (Ax, x)/2 on X. It is called the generating function
of this Lagrangian subspace. 3) A linear transformation of a space V preserves
a symplectic form w exactly when its graph in the space Vffi Vis Lagrangian
with respect to the symplectic structure W = 7tfw7t!w, where 1t 1 and 1t 2 are
the projections onto the first and second summands (the area W(x, y) of a
parallelogram is equal to the difference of the areas of the projections).
1.3. The Lagrangian Grassmann Manifold. The set of all Lagrangian sub
spaces of a symplectic space of dimension 2n is a smooth manifold and is called
the Lagrangian Grassmann manifold An.
Theorem. An is diffeomorphic to the manifold of cosets of the subgroup On of
orthogonal matrices in the group un of unitary n X n matrices (a unitary frame in en
generates a Lagrangian subspace in C" considered as a real space).
Corollary. dim An = n( n + 1)/2.
On the topology of An, see chap. 6.
Symplectic Geometry 7
Example. A linear line complex. By a line complex is meant a three
dimensional family of lines in threedimensional projective space. Below we shall
give a construction connecting the socalled linear line complexes with the
simplest concepts of symplectic geometry. This connection gave symplectic
geometry its name: in place of the adjective "complex" (composed of Latin roots
meaning "plaited together"), which had introduced terminological confusion,
Hermann Weyl [75] in 1946 proposed using the adjective "symplectic", formed
from the equivalent Greek roots.
The construction which follows further on shows that the Lagrangian
Grassmann manifold A2 is diffeomorphic to a nonsingular quadric (of signature
(+ + +  )) in fourdimensional projective space.
The points of the projective space IP 3 = P( V) are onedimensional subs paces of
the fourdimensional vector ·space V. The lines in IP 3 are twodimensional
subspaces of V. Each such subspace uniquely determines up to a factor an
exterior 2form 4> of rank 2, whose kernel coincides with this subspace. In the
6dimensional space 1\ 2 V of all exterior 2forms, the forms of rank 2
form a quadratic cone with the equation 4> 1\ 4> = 0. Thus the manifold of all
lines in IP 3 is a quadric Q in IP 5 = P( 1\ 2 V). A linear line complex is
given by the intersection of the quadric Q with a hyperplane H in IP 5 •
A hyperplane in P( 1\ 2 V) can be given with the aid of an exterior 2form
w on V: H = P({ tf>e N VI w 1\ 4> = 0} ). Nondegeneracy of the form w is
equivalent to the condition that the linear line complex H n Q be nonsingular.
The equation w 1\ 4> = 0 for a form 4> of rank 2 means that its
kernel is Lagrangian with respect to the symplectic structure w. Therefore, a
nonsingular linear line complex is the Lagrangian Grassmann manifold A2 •
§ 2. Linear Hamiltonian Systems
Here we shall discuss the Jordan normal form of an infinitesimal symplectic
transformation.
2.1. The Symplectic Group and its Lie Algebra. A linear transformation G of
a symplectic space ( V, w) is called a symplectic transformation if it preserves the
skewscalar product: w(Gx, Gy) = w(x, y) for all x, ye V. The symplectic trans
formations form a Lie group, denoted by Sp(V) (Sp(2n, IR) or Sp(2n, C) for the
standard real or complex 2ndimensional symplectic space).
Let us consider a oneparameter family of symplectic transformations, and let
the parameter value 0 correspond to the identity transformation. The derivative
of the transformations ofthe family with respect to the parameter (at 0) is called a
Hamiltonian operator. By differentiating the condition for symplecticity of a
transformation, we may find the condition for an operator H to be Hamiltonian:
w(H x, y) + w(x, H y) = 0 for all x, y E V. A commutator of Hamiltonian operators
8 V.I. Amol'd, A.B. Givental'
is again a Hamiltonian operator: the Hamiltonian operators make up the Lie
algebra sp ( V) of the Lie group Sp ( V).
The quadratic form h(x) = w(x, Hx)/2 is called the Hamiltonian of the operator
H. A Hamiltonian operator can be reconstructed from its Hamiltonian out of the
equation h(x+ y)h(x)h(y) = w(y, Hx) for all x, y. We get an isomorphism of
the space of Hamiltonian operators to the space of quadratic forms on the
symplectic space V.
Corollary. dim Sp(V) = n(2n+ 1), where 2n =dim V.
The commutator of Hamiltonian operators defines a Lie algebra structure
on the space of quadratic Hamiltonians: {h 1 , h2 } (x) = w(x, (H 2 H 1  H 1 H 2 )x)/2
=w(H 1x,H 2x). The operation{·,·} is called the Poisson bracket. In Darboux
coordinates the Poisson bracket has the form {h 1,h2 }='f_(oh 1/opk·oh 2/oqk
oh2/opk · oh1/oqk).
The matrix of a Hamiltonian operator in Darboux coordinates H = ( ~ : )
satisfies the relations 1 B* = B, C* = C, D* = A. The corresponding Hamilton
ian his the quadratic form whose matrix is [h] = 21 OH =~( C D)·
2 A B
A linear Hamiltonian system of differential equations x = Hx can be written as
follows in Darboux coordinates: fJ = ohfoq, 4 = ohfop. In particular, the
Hamiltonian is a first integral of its own Hamiltonian system: Ji = ohfoq · 4
+ ohfop · p = 0. Thus we have conveyed the structure of the Lie algebra of the
symplectic group and its action on the space V, in terms of the space of quadratic
Hamiltonians.
Examples. 1) To the Hamiltonian h = w(p 2+ q 2 )/2 corresponds the system
of equations 4 = wp, p = wq of the harmonic oscillator. 2) The group of
symplectic transformations of the plane IR 2 coincides with the group SL(2, IR) of
2 x 2 matrices with determinant l. Its Lie algebra sl(2, IR) has three generators
X= q2f2, Y =  p2/2, H = pq with commutators [X, Y] = H, [H, X]= 2X,
[H, Y] = 2Y.
2.2. The Complex Classification of Hamiltonians. We shall consider two
Hamiltohian operators on a symplectic space V to be equivalent if they can be
transformed into one another by a symplectic transformation. The correspond
ing. classification problem is entirely analogous to the case of the Jordan normal
form of a linear operator. To put it in more erudite terms, it is a question of
classifying the orbits of the adjoint action of the symplectic group on its Lie
algebra. In the complex case the answer is given by
1 • = transposition.
where qk=xk. 3) the number of Jordan blocks of odd dimension with eigenvalue a = 0 is even. Linear Variational Problems. x. 3.4) with the quadratic Hamiltonian h= ±{poq1 +. As normaJ forms for linear Hamiltonian systems.• . Let us consider J the problem of optimizing the functional L(x0 .Under this identification the operator H*: V*+ V* dual to a Hamiltonian operator H: V+ Vis turned into H. 2) the Jordan blocks corresponding to the eigenvalues a and a have the same structure. On the other hand. + a 0 x~)/2...are equivalentif and only if they are similar (i. have the same Jordan structure). +(l)"a0 x 0 =0 is a linear homogeneous equation with constant coefficients involving only even order derivatives of the required function x. Let H: V+ V be a Hamiltonian operator. Therefore the Jordan structure of a Hamiltonian operator meets the restrictions 1) if a is an eigenvalue. sect. We remark that with this construction it is not possible to obtain a Hamiltonian system having a pair of Jordan blocks of odd order n with eigenvalue 0. then a is also an eigenvalue. the equation of the extremals is equivalent to the Hamiltonian system (see chap. The Hamiltonian ± (p 0 q 1 + . · · · aoq~)/2}..1 =xn. Symplectic Geometry 9 Williamson's Theorem [77].. Let x = x(t) be a function of the variable t.1 =akxkdPt!dt are Darboux coordinates on the 2ndimensional phase space of the equation of the extremals. Corollary. sect.2 q"_ d corresponds to such a system. The equation of the extremals of this functional X2 nan_ 1 x 2 n.. A Hamiltonian Jordan block of order 2n with eigenvalue 0 is obtained for L = x~/2. 1.3. 1. 3. on each of which the operator H has either two Jordan blocks of the same order with opposite eigenvalues. The symplectic form allows us to identify the space V with the dual space V* as follows: x~+w(x. one may take the equations of the extremals of special variational problems. ·). · · +Pn2qn1+(P~1an1q~1. + Pn.Pn. Then V decomposes as a direct skeworthogonal sum of symplectic subspaces. where the Hamiltonian formalism of variational problems with higher derivatives is described.) dt with the Lagrangian function L = (x~ +an_ 1 x~_ 1 + ..4.Pt. and let xk = dkx/dtk. The extremals in this case are the solutions of the equation . We assume that the reader is familiar with the simplest concepts of the calculus of variations. Apart from this.e. Hamiltonian operators on a complex symplectic space.2 + . and we shall make use of the formulas of chap. the Jordan structure of Hamiltonian operators is arbitrary. 2. or one Jordan block of even order with eigenvalue 0.. • .
I. 4) The case of a quadruple of Jordan blocks of order m = n/2 with eigenvalues ±a± bJ=T is represented by the Hamiltonian n1 ho + P~1/2. We shall give a list of the elementary normal forms of quadratic Hamiltonians in the Darboux coordinates (p 0 .4. the Williamson theorem of sect. 3) The case of a pair of Jordan blocks of order n with nonzero eigenvalues ±z is represented by a Hamiltonian of one of the two forms ± (ho + P~. b provided a =/= 0 and =/= 0. 2. . i. Givental' = 0. + a0 = em"(e + e1)m' . For quadratic Hamiltonians this is not always so.e. Amol'd.10 V. qn. 2. i. An obvious difference between the real case and the complex one is that the Jordan blocks of a Hamiltonian operator split into quadruples of blocks ofthe same dimension with eigenvalues ±a ± bJ=T.1 + .. in the way in which it is formulated there. •. In particular. For example. the Hamiltonians ± (p 2 + q 2 ) of the harmonic oscillator have the same eigenvalues ± 2J=l.. . Pnl• q 0 . A more essential difference lies in the following. .Jf.tf2. but for a purely imaginary z they are not equivalent). In general. . A. to the d 2"x/dt 2 " Lagrangian function L with characteristic polynomial ~n +an_ 1 ~n.. the polynomials x(t) of degree < 2n.d of the standard symplectic space IR 2 ". 2) The case of a Jordan block of even order 2n with eigenvalue 0 is represented by a Hamiltonian of exactly one of the two forms ± (ho + P~1/2).. they are equivalent over C. where . Normal Forms of Real Quadratic Hamiltonians.e. Two real matrices are similar in the real sense if they are similar as complex matrices. (e + edmk corresponds a Hamiltonian operator with one Jordan block of dimension 2m 0 and eigenvalue 0 and k pairs of Jordan blocks of dimensions mi with eigenvalues ± ..B. • . 1) The case of a pair of Jordan blocks of (odd) order n with eigenvalue 0 is represented by the Hamiltonian n2 ho = L k=O pkqk+ 1 (h 0 = 0 when n = 1). but they are not equivalent over IR: to these Hamiltonians correspond rotations in different directions on the phase plane oriented by the skewscalar product. L k=O AkqV2. "f c:z <nklqV2) k=O 2 (for real z these two Hamiltonians are equivalent to each other.2 does not carry over to the real case.
> 0. ). a skew symmetric bilinear form n can be written as A. 1} with respect to the form n.k~).c VN Lc vN+ 1. Invent. For the "frequencies" wk one has the following minimax principle. Math. ~ w 1 > 0. we obtain the minimax principle and analogous corollaries for the frequencies wk = 2/ ).M.k Corollary.! ~ ). ).. N. SignDefinite Hamiltonians and the Minimax Principle.. ~ ).. 27 (1983). 12941297 (English translation: Sov. 2) If we take a symplectic structure on ~ 2 " for n and a positive definite Hamiltonian h for the Euclidean structure.wk(pf + qf}/2. max S(L) = n).M+I· Lc wNM vNM c vN Lc vNM Remarks. 79 (1985). Math.: Semicontinuity of the singularity spectrum. J. Courant) for pairs of Hermitian forms in IC": U = I. Nauk SSSR 270 (1983).) . states that Ak= min max(U'/U)= max mm (U'/U)... . mm max S(L)=nAk. Symplectic Geometry II Theorem ([77]). 1 p 1 A q 1 + . In suitable Cartesian coordinates on the Euclidean space VN.. 2n :s. 1 Lc W Lc V and nA'1 = max S(L) ~ min max S(L) = n). Theorem.~~). nX1 = max S(L) :s. x) :s.. where w. . 2 In suitable Darboux coordinates. The invariants ).p.m=Ofor m>n). Arnol'd in 1977 in connection with the conjecture (now proved by Varchenko and Steenbrink) that the spectrum of a singularity is semicontinuous. a positive definite quadratic Hamiltonian has the form h = I._ 1 ~ ••. under an increase of the Hamiltonian the frequencies grow: w~ ~wk. 1) Courant's minimax principle (R. To an oriented plane L c v N let us associate the area S(L) of the unit disk D(L) = {xE L I <x. 1 :s.zkik.k+M (we set ).I.: On semicontinuity of the spectrum and an upper estimate for the number of singular points of a projective hypersurface. k= 1. N. ~ w.. . 2 The results in this section were obtained by V. From this it is easy to deduce C"cC" CcCk cn+tkc:C" CcC"+tk our theorem. 557565.. .~ of the restriction of the form n to a subspace wNMc VN satisfy the inequalities ). A.k· In particular. sA.. (See Varchenko. + A.5.H.2 ~ . 2.. Dokl. 735739) and Steenbrink. U' = LAkzkik. n. . A real symplectic space on which a quadratic Hamiltonian his given decomposes into a direct skeworthogonal sum of real symplectic subs paces such that the form h can be represented as a sum of elementary forms in suitable Darboux coordinates on these subspaces. 1\ q. Dokl. vN+ I lr. Akad. For example..
3. 1). Let M be the space of quadratic Hamiltonians on the standard symplectic space IR 2 n and let G = Sp(2n. The miniversal deformations introduced below are normal forms for families of operators which are safe from the deficiency mentioned. ±ck. Example. are the parameters) s H. x) is called versa/ if any deformation of the element x is equivalent to a deformation induced from ¢. The following deformation (A.e. x) are called equivalent if there is a deformation of the identity element g: ( V.I. IR) be the group of symplectic linear transformations on IR 2 n. The Versality Theorem. id) such that c/J(v) = g(v)ljl(v). in general. A deformation of an element x EM is a germ of a family ( V. x) (where 0 is the coordinate origin in V ~ !Rn). 0)> (M.1. 0)>(M. if ¢ = 1/1 ov.12 V. ±J=1 dd is nonmultiple.2kl)(Pzkl qzkq2kl Pzk) k=l r n + I (ck+A. 0)>(M. 0)>(M.B. Arnol'd. 0)>(M. The Concept of the Miniversal Deformation.k)Pkqk+ k=r+l k=2s+l L (dk+A. Corollary. A deformation of the point xEM is versa/ if and only if it is transversal to the orbit Gx of the point x in M._= L [(bk+A. x) under a smooth mapping of the base spaces v: (V.k)(pf+qn/2 (t) is a miniversal deformation of the Hamiltonian H 0 if its spectrum { ± ak ±j=tbk. Two points of Mare considered equivalent if they lie in one orbit. A versal deformation with the smallest basespace dimension possible for a versal deformation is called miniversal. One says that a deformation ¢: ( V. Let a Lie group G act on a smooth manifold M. A family with parameter space (base space) Vis a smooth mapping V> M. It has to do with the following abstract situation. Two deformations ¢. x) is induced from the deformation ljl: (W. Families of Quadratic Hamiltonians The Jordan normal form of an operator depending continuously on a parameter is. The number of parameters of a miniversal deformation is equal to the codimension of the orbit. if they go over into each other under the action of this group. a discontinuous function of the parameter. 0) >(W. A deformation¢: ( V. Definition. 0)>(G. A. 0)>(M. 0).One says that the deformation ¢: ( V. i. 1/J: (V. The germ of the manifold M at the point x is obviously a versal deformation for x. Let X c M be a submanifold. . Givental' § 3. x) of a point xEX is transversal to X if ¢*(T0 V)+ TxX = TxM (Fig. but generally speaking it is not miniversal.
Symplectic Geometry 13 Fig. Z)+( Y. 2). F)= tr (H F*). . x). 0)+(T. We note that the transposed matrix of a Hamiltonian matrix is again Hamiltonian.. From the properties of the trace we obtain: ([X.2. Now every deformation¢: (V. I. Let us choose a submanifold L transversal at the identity element of the group G to the isotropy subgroup Stx = {g Igx = x} of the point xE Manda submanifold T transversal at the point x to its orbit in M. 3.X g·T v / v v v Ld·T ~ X _. Let us introduce in the space of such matrices the elementwise scalar product <. ). 0)+(G. where * denotes transposition. t: (V. x) automatically takes on the form <jJ(v)=g(v)t(v). IR). Y]=XY. 0)+(M.YX is the commutator. Z])=O. Let M once again be the space of quadratic Hamiltonians in IR 2 " and G = Sp(2n. [X*. where g: (V. The action by the elements of L on the points of T gives a diffeomorphism of a neighbourhood of the point x in M to the direct product L x T. It can be represented in the form (H. Transversality L·../ t T Fig. id). 2. where [X. Y]. We shall identify M with the space of Hamiltonian matrices of order 2n x 2n. Miniversal Deformations of Quadratic Hamiltonians. The proof of the versality theorem Sketch of the proof of the theorem (see Fig.
Let us divide up the space of quadratic Hamiltonians into classes according to the existence of eigenvalues of different types (but not numerical values) and according to the dimensions of the Jordan blocks. The paper [29] gives the explicit form of the miniversal deformations for all normal forms of quadratic Hamiltonians. The codimension c of a given class is the smallest number of parameters of families in which Hamiltonians of this class are encountered unremovably. 2::mu and ml 2:: ..14 V. 2:: n. H): Xf4H +X* is a miniversal deformation of the quadratic Hamiltonian H. Generic Families. A.B.. Theorem ([29]). Givental' Lemma. a generic Hamiltonian has no multiple eigenvalues. i.3. it also does not have any preassigned spectrum of eigenvalues. The importance of studying generic phenomena is explained by the fact that in applications the object being investigated is often known only approximately or is subject to perturbations because of which exceptional phenomena are not observed directly.mk).. The dimension d of the base space of the miniversal deforma tion of a Hamiltonian H equals v u v + L [2(2jl)mi+l]+2I L j=l j=lk=l min(mi. and by m1 2:: . The deformation (Zn. Amol'd. [H*.(z) the dimen sions of the Jordan blocks with eigenvalue z#O. Such a classification. where the mi are even and the mi are odd (out of every pair of blocks of odd dimension only one is taken into account).. H*] =0} of the Hamiltonian H* in the Lie algebra of quadratic Hamiltonians.. Let us denote by v half the number of different nonzero eigenvalues of the Hamiltonians of the given class. in contrast to the classification by Gorbits. then (F. X] =0. 0 Corollary. One says that the Hamiltonians of a given class are not encountered in generic /parameter families if one can remove them by an arbitrarily small perturbation of the family. The orthogonal complement in M to the tangent space at the point H of the orbit of the Hamiltonian H coincides with the centralizer Zn• = {XEM I [X. and conversely. is discrete (even finite).I. F]. For a Hamiltonian H let us denote by n 1(z) 2:: n2 (z) 2:: . [H*. 2:: mv the dimensions of its Jordan blocks with eigenvalue 0. X) =0 for all FEM. Proof If ([H. .. XJ> =0. 0)+(M.e. For example. but nevertheless does have some other spectrum. 3.
F. in which a) for almost all t 0. q). be a smooth generic family of quadratic Hamiltonians which depends on one parameter. (±a) 2 (±b)2 . Symplectic Geometry IS Theorem. (±a) 2 denotes a pair of Jordan blocks of order 2 with eigenvalues a and a respectively). and on the fact that the number of parameters indexing the Gorbits of the given class is equal to v. 04. has one of the following forms (±a) 2 : P 1 Q 2 +P~/2(a 4 +JL 1 )Qf/2(a 2 +JL 2 )Q~+H. [9]). (±a±ib)2 . b)for isolated values of t 0 . F. (±ia)2(±ib)2 . In one and twoparameter families of quadratic Hamiltonians one encounters as irremovable only Jordan blocks of the following twelve types: c= 1: (±a) 2 . 3. 0 2 (here the Jordan blocks are denoted by their determinants. Let F. (±a)2 0 2 . c = d. (±a)2 (±ib)2 .v (so that the formula for c can be obtained out of the formula ford ofthe preceding theorem by diminishing each term of the form ~)2jl)ni(z) by one). q). depending smoothly on t. Corollary 2. (±ia) 3 .JLQ 2 /2] + H A(p. c=2: (±a) 3 .(p. Fig. (±ia)2 0 2 (the remaining eigenvalues are simple). 02 : ± [ P 2 /2. 3) (for more details on this see [ 4]. Generic families Corollary 1. q) . The proof of this theorem is based on the intuitively obvious fact that a generic family is transversal to every class (Fig. for example. (±ia) 2 : ± [P 1 Q2 +P~/2(a 4 + JL 1 )Qf/2+(a 2 + JL 2 )Qn + HA(p. has the form HA (see formula (1)). In the neighbourhood of an arbitrary value t = t 0 there exists a system of linear Darboux coordinates. (±ia)2 ..
L7 ili ~ Fig. Bifurcation diagrams of quadratic Hamiltonians. The bifurcation diagram of a deformation of a Hamiltonian is the germ of the partition of the parameter space into the preimages of the classes.t=O .) are smooth functions of the parameter t. Proof The formulas listed are miniversal deformations of representatives of the classes of codimension l. Givental' (here (P. Bifurcation Diagrams.4. see Fig. c = 1 ~ ~ \UY \.B.t=O l. 0)). 4.16 V. c = 2 . In Figs.=O Fig.I. Amol'd..(t 0 ). p. (A. Bifurcation diagrams of quadratic Hamiltonians. D 3.(t 0 )) = (0. Q. 5. 3) the class partition structure in the space of quadratic Hamiltonians itself. EBEBEB . The bifurcation diagrams of generic families reflect (in view of the condition of transversality to the classes. q) are Darboux coordinates on ~ 2 m. p. A. p. (A. 4 and 5 are presented the bifurcation diagrams of generic deformations for the classes of codimension 1 and the first four classes of codimension 2 in the order of their being listed in the statement of corollary 1.
w~ =w~ :#0. but without Jordan blocks. the defining relation shows that the matrices G and G..p~. The Hamiltonian operators with a nonmultiple purely imaginary spectrum form an open set in the space of Hamiltonian operators. to the points of which correspond the operators with Jordan blocks of dimension 2. In fact. 4. 6). The bifurcation diagram of the Hamiltonian Pt + qt. This leads to the invariance of the Jordan structure of a symplectic .AE is nilpotent) corresponding to symmetric eigenvalues have the same Jordan structure.1. then in suitable Darboux coordinates the corresponding Hamiltonian has the form h= [w 1 (p~ +qfl+ .e. IR) consists of the linear transformations of the space IR 2 n which preserve the standard symplectic structure w = LPt "q".+q.)]/2.q~ § 4. The symplectic group Sp(2n. say.J=lw"}. form a set of codimension 3 in the space of Hamiltonian operators. 6. The Hamiltonian operators with a purely imaginary spectrum with multiple eigenvalues. Ifthe invariants w 1 and w 2 of the Hamiltonian hare of the same sign. The spectrum of a real symplectic transformation is symmetric with respect to the unit circle and the real axis. The matrix G of a symplectic transformation in a Darboux basis therefore satisfies the defining relation G*OG=O. then the bifurcation diagram is a quadratic cone (Fig.1 are similar over C. If such an operator H has the spectrum { ±. The Symplectic Group The information we bring below on real symplectic groups is applied at the end of the section to the theory of linear Hamiltonian systems of differential equations with periodic coefficients.. Let us look at the following important example. If the invariants w 1 and w 2 are of different signs. Symplectic Geometry 17 Remark. +wn(P. Fig.. The Spectrum of a Symplectic Transformation. Theorem. Suppose. then the bifurcation diagram is a point (the class of h) in the space IR 3 all Hamiltonians near h have a purely imaginary spectrum and have no Jordan blocks. The eigenspaces in the weaker sense (i. where G. One should not think that the bifurcation diagram of a real quadratic Hamiltonian depends only on the Jordan structure.
onto the set of symplectic transformations Gall of whose eigenvalues are different from ± 1. the inter section Sp(V2")nO(V 2 ". h) is an ndimensional torus T" and is generated by transformations exp(AHk).G)k jk. The symplectic transformations ± E commute with all elements of the group Sp( V2 ") and form its center. If all the wk are different. All such tori are conjugate in the symplectic group.1 . isomorphic to the permutation group on n letters and acts on the torus by permuting the oneparameter subgroups exp (AHk). Every compact . Two elements of the torus are conjugate in the symplectic group if and only if they lie in the same orbit of this action. in suitable Darboux coordinates. where H is an elementary Hamiltonian operator from sect. Givental' transformation and its spectrum with respect to the symmetry AI+ A.exp and the results of§ 2. Theorem. the form ± exp (H).4. 2. 0.L{E.1 exp (H)G.18 V. Let us look at the normalizer N(T")= {geSp (V 2 ") IgT"g. we may obtain the following result. The expo nential of an operator gives the exponential mapping H~+exp(H)= "fHkjk! of the space of Hamiltonian operators to the symplectic group. ~) is neither injective nor surjective. Every compact commutative subgroup of Sp(V 2 ") lies in some torus T" of the kind described above.3. Every compact subgroup ofSp( V2 ") lies in the intersection of Sp(V 2 ") with the orthogonal group O(V 2 ") of mappings which preserve some positive definite quadratic Hamiltonian h = "fwk(P~ + qD/2. The mapping exp is a diffeomorphism of a neighbourhood of 0 in the Lie algebra onto a neighbourhood of the identity element in the group. h) coincides with the unitary group U" of this space.1 H G)= G. Using the mappings ca. The symplectic group acts by conjugation on itself and on its Lie algebra. The exponential mapping is invariant with respect to this action: exp (G. The factor group W = N(T")/T" is called the Weyl group. Therefore.I. Realness gives the second symmetry AI+ l 4. . ~)+Sp(2n. where Hk has the Hamiltonian (p~+qD/2. These formulas give a diffeomor phism ca of the set of Hamiltonian operators H all of whose eigenvalues are different from ± 1.B. Subgroups of the Symplectic Group. A symplectic space on which is given a symplectic transformation G splits into a direct skeworthogonal sum of symplectic subs paces on each of which the transformation G has. Amol'd. The mapping exp: sp(2n. H=(GE)(G+E). A. the Hamiltonian h together with the symplectic form endows V2 " with the structure of an ndimensional complex Hermitian space. 4. for the study of the symplectic group the Cayley parametrization is more useful: G=(E+H)(EH).1 .2. The inverse transformation is given by the series In G = . exp. It is finite. All the subgroups U" are conjugate. If all the wk are equal to each other. The Exponential Mapping and the Cayley Parametrization.1 . The intersection Sp(V 2 ")nO(V 2 ".1 = T"} of the torus T" in the symplectic group.
1 A.{zEC llzl = 1} gives the projection onto the second factor). IR) c Sp(2n.1 ). The group Sp 1 c Sp(2. 4. IR). As a maximal torus in Sp(2.H k) t+ exp(.S. IR) is connected. The group Sp(2. The fundamental group x 1(Sp(2n. the group SU" is connected and simply connected (this follows from the exact homotopy sequences of the fibrations sun~S 2 ". C) differs from the normalizer N(T") in Sp(2n. 4. the operators U turn out to be unitary. C). 1) The symplectic group Sp(2n. In particular. but its normalizer N c(T") in Sp(2n. The complex Weyl group We is isomorphic to 7Lf27L. The latter follows from the properties of the manifold Un: un~sun X S 1 (the function dete: Un. 2) The torus T" is a maximal torus in the complex symplectic group as well: T" c Sp(2n. C) a maximal compact subgroup is isomorphic to the compact symplectic group Sp" of transformations of an ndimensional space over the skew field of quaternions. IR)) is isomorphic to 7L. C) of unit quaternions coincides with the group SU(2). Let h be a quadratic Hamiltonian whose coefficients depend continuously on the timet and . In this case the maximal torus coincides with the maximal compact subgroup U 1 of Sp(2. Corollary.Hk). I) In the complex symplectic group Sp(2n.Hk) and reflections exp(A. IR) can be contracted to the unitary subgroup un• 2) The symplectic group Sp(2n. The Topology of the Symplectic Group Theorem. Remarks. The manifold Sp(2n. The key to the proof is given by the polar decomposition: an invertible operator A on a Euclidean space can be represented uniquely in the form of a product S· U of an invertible symmetric positive operator S =(AA *) 1' 2 and an orthogonal operator U = S. IR) is diffeomorphic to the product of an open disk with a circle. Linear Hamiltonian Systems with Periodic Coefficients [30]. For symmetric operators A acting on the underlying real space IR 2 " of the Hermitian space C".j=l). Its action on S0(2) is given by conjugation by means of the matrix diag (j=T. The Weyl group We= Nc(T")/T" acts on T" via compositions of permutations of the subgroups exp(A. Example. IR). and the logarithms InS of the operators S fill out the n(n + 1) dimensional space of symmetric Hamiltonian operators.. Symplectic Geometry 19 subgroup of Sp( V2 ") lies in some unitary subgroup of this type. a torus T" and its normalizer N(T") lie in a (unique) subgroup U". The Weyl group W is trivial. IR) is diffeomorphic to the Cartesian product of the unitary group U" with a vector space of dimension n(n + 1).4.A. C) one may take the group S0(2) of rotations of the plane. Exatnple.
e. Theorem A. The latter condition means that the closure of this group in Sp(2n. . and strongly stable if all nearby linear Hamiltonian systems with periodic coefficients are also stable (nearness is to be understood in the sense of the norm max llh(t)ll). . then G. The system (2) is strongly stable if and only if there are no relations of the form A. and the role of the integer parameter is played by the element of the fundamental group n 1(S p(2n. Here the number 2" shows up as the ratio of the orders of the Weyl groups We and W. We obtain a continuously differentiable curve G. and in questions of parametric resonance.kA.0 • The transformation G = G10 is called the monodromy operator of the system (2). Stability and strong stability of the system (2) are properties of its monodromy operator.G. and if t 0 is the period of the Hamiltonian h. with this initial condition.B. Amol'd. the stability of the system (2) is equivalent to the boundedness of the cyclic group {Gm} generated by the monodromy operator. i.k. IR).: IR 2 "+ IR 2 " which associates to an initial condition x(O) the value of the solution x(t) of equation (2). in automatic control theory. which uniquely determines the original system of equations. at the moment of timet. Let us consider the mapping G.. IR). IR) is compact. Theorem B. It turns out that the homotopy classes are naturally indexed by the 2" collections of n ± signs and by one more integer parameter.t> . in the symplectic group Sp(2n.20 V. A. In fact. The curve G. IR)). the monodromy operator lies in some torus T" c Sp{2n. begins at the identity element of the group: G0 =E.kl = 1. To the Hamiltonian h corresponds the linear Hamiltonian system with periodic coefficients (2) Such systems are encountered in the investigation of the stability of periodic solutions of nonlinear Hamiltonian systems.1= 1 between the numbers A. We shall call the system (2) stable if all of its solutions are bounded as t+ oo. The homotopy relation partitions all strongly stable systems (2) of order 2n into classes.+ 10 =G. Then the monodromy operator of a stable system takes the form G = diag(A.I.. The system (2) is stable if and only if its monodromy operator is diagonalizable and all of its eigenvalues lie on the unit circle. We may consider T" as the diagonal subgroup of the group of unitary transformations of the space C". IA. Givental' are periodic in t with a common period. An). I Two strongly stable Hamiltonian systems will be called homotopic if they can be deformed continuously into one another while remaining within the class of strongly stable systems of the form (2).
IR) . d. In the general case nearly the entire boundary of the set St" consists of nonstable operators. IR).1 = l to invariants of different signs: wk + w1 = 0.k = A. Under deformations ofthe system (2) within the class of strongly stable systems this sequence does not change: because of the relations A. 7 the set St 1 in the group Sp(2. a singularity like that of the quadratic cone in IR 3 . Theorem C. 7. The stationary points of the transformations belonging to the Weyl group Ware given by the equations A. t 0 ]. In Fig.1# l the eigenvalues can neither come down from the semicircle. nor change places if they have exponents of different signs.k = A.kA. 3. with its beginning at the identity element and its end at the point corresponding to the monodromy Fig. Symplectic Geometry 21 Remark. let us choose those which lie on the upper semicircle Im A. We obtain a welldefined sequence of n exponents ± l. Theorem D ([30]). 1 # ± l correspond in this connection to multiple invariants of the same sign: wk = w1 # 0.1= l.4 explains why strong stability is violated only in the second case. If the monodromy operator of a stable system has a nonmultiple spectrum. 5b. The monodromy operators of strongly stable systems (2) form an open set St" in the symplectic group Sp(2n.1• Using the Cayley parametrization one may verify that the partition into conjugacy classes in the neighbourhood of the monodromy operator GET" is organized in the same way as is the partition into equivalence classes of quadratic Hamiltonians in the neighbourhood of the Hamiltonian h= L. At such points the boundary has a singularity (in the simplest case.1 or A. The remark in sect. IR) is depicted. The stable but not strongly stable monodromy operators also lie on the boundary and form a set of codimension 3 in the symplectic group. Multiplicity of the spectrum means that A.wk(pf+qf}/2. Each component of the set St" is simply connected. Among the eigenvalues A.kA.k = A. The subset of stable operators in Sp(2. The singularities of the boundary of the set of strong stability along strata of codimension 2 can be seen in Fig. consisting of 2" connected components corresponding to the 2" different sequences of exponents. 6). These equations single out the stationary points of the transformations belonging to the Weyl group We on the torus T". Fig. and the relations A. The relations A.kA.! 1 of the monodromy operator of a strongly stable system. Under a homotopy of the system (2) the curve Gn tE[O. > 0. then the system is strongly stable.
Theorem E ([30]). where n 0 (Sp(2n.22 V. correspond homotopic systems (2). Fig. 8). The Darboux Theorem. Essentially we have described the relative homotopy "group" n=n 1(Sp(2n. is deformed continuously in the symplectic group. The homotopy classes of systems (2) whose monodromy operators lie in the same component of the set St" as the monodromy operator G ofa given system are in onetoone correspondence with the elements of the fundamental group n 1 (Sp(2n. A manifold equipped with a symplectic structure is called a symplectic manifold. Conversely. Stn) in terms of the exact homotopy sequence n 1 (Stn)+ n 1 (Sp(2n. n 1(St") = {0} (Theorem D).. A.I. Fig. Local Symplectic Geometry 1. IR)). Givental' operator. to homotopic curves G. IR)) = 71. IR))+ 7t+ n 0 (Stn)+ n 0 (Sp(2n.B. IR)) = {0} (the symplectic group is connected).1. A diffeomorphism of symplectic manifolds which takes the symplectic structure . IR). Arnol'd. Chapter 2 Symplectic Manifolds § 1. By a symplectic structure on a smooth even dimensional manifold we mean a closed nondegenerate differential 2form on it. IR)) = 7L (to a system (2) with the same monodromy operator G is associated in the fundamental group the class of the closed curve formed in Sp(2n. =II= n 0 (St") = 2" (Theorem C). IR). n 1(Sp(2n. Nonhomotopic systems with the same monodromy operator Remark. 8.
The closed ness condition in the definition of symplectic structure connects the skewscalar products in the tangent spaces at neighbouring points in such a way that the local geometry of symplectic manifolds turns out to be universal. d) The hyperbolic and elliptic sections of our curve are separated by parabolic points. . in suitable local coordinates p 1. a symplectic structure on a smooth manifold has the form dp 1 1\ dq 1 + . •• • . The field of kernels of the form w cuts out a field of directions on the threedimensional manifold. In a neighbourhood of such a point the form w can be reduced to one of the two forms d(xz 2 /2)" dy + d(xz ± tyz 3 f3) 1\ dt.. In the neighbourhood of an arbitrary point.. At a generic point of the curve the twodimensional kernel of the form w is tangent to the three dimensional manifold. Example. 9.. xk. 3 The tangent space at each point of a symplectic manifold is a symplectic vector space.. q 1 . Symplectic Geometry 23 of one over into the symplectic structure of the other is called a symplectic transformation or a symplectomorphism. At a generic point this submanifold is transversal to the twodimensional kernel of the form w. 1. • •• . q 1. [62]. • • . The Degeneracies of Closed 2Forms on R 4 • Let w be a generic closed differential 2form on a 4dimensional manifold. but transversal to this curve. [7]). at which the twodimensional kernel of the form w is tangent both to the threedimensional manifold and to the curve itself. . x 1 . Here is known only that there 3 In the literature the traditional name "canonical transformation" is also used. The field lines corresponding to the + sign in the normal form are depicted in Fig.2. 1\ dq. Corollary.. Its absence in the definition of a symplectic structure would make the local classification of such structures boundless. a) At a generic point of the manifold the form w is nondegenerate and can be reduced in a neighbourhood under a suitable choice of coordinates to the Darboux form dp 1 1\ dq 1 + dp 2 1\ dq 2 • b) At the points of a smooth threedimensional submanifold the form w has rank 2.. . under a suitable choice of local coordinates p 1 . qm. Pm. q. . + dp. The Darboux Theorem. p. + dpm 1\ dqm. The condition of nondegeneracy is worthy of special discussion. . In a neighbourhood of such a point w can be reduced to the form p 1dp 1 "dq 1 + dp 2 "dq 2 • c) The next degeneracy of the generic form w occurs at the points of a smooth curve on our threedimensional submanifold. With the sign the spiral rotation is replaced by a hyperbolic turning (see [51]. • • . • . Symplectic manifolds of the same dimension are locally symplectomorphic. the form dp 1 1\ dq 1 + . Nevertheless in the case of degeneracies of constant rank the answer is simple: a closed differential 2form of constant corank k has.
B. Let there be given a germ of a smooth submanifold at the coordinate origin of the space IR 2 n and two germs of symplectic structures w 0 and w 1 in a neighbourhood of the origin whose restrictions to this submanifold coincide. Germs of Submanifolds of Symplectic Space. A "magnetic field" connected with a degenerate structure on IR4 exists at least one modulusa continuous numeric parameter which dis tinguishes inequivalent degeneracies of the form in a neighbourhood of the parabolic point [35]. Thus we may consider that we have one submanifold germ and two symplectic structures in a neighbourhood of the submanifold which coincide upon restriction to it.I. we obtain the Darboux theorem of sect. Amol'd. Germs of submanifolds for which this is possible will be called equivalent. For equivalent germs these degeneracies are identical. The restriction of the symplectic structure of the ambient space to a submanifold defines on it a closed 2form. Givental' Fig. Proof We apply the homotopic method. Here we shall discuss under what conditions two germs of smooth submanifolds of symplectic space can be moved into one another by a local diffeomorphism of the ambient space which preserves the symplectic structure.24 V. e) All more profound degeneracies of the form w (for example. One may assume that the sub manifold is a linear subspace X and that the differential forms w 0 and w 1 coincide at the origin (the latter follows out of the linear "relative Darboux . their intrinsic geometry coincides. A. Two germs of submanifolds of Euclidean space with the same intrinsic geometry may have a different exterior geometry. in other words. 9. If this requirement is fulfilled. its turning to zero at isolated points) are removable by a small perturbation within the class of closed 2forms. In symplectic space this is not so. but not necessarily preserving this structure itself. Then there exists a germ ofa diffeomorphism of the space IR 2 n which is the identity on the submanifold and which takes w 0 over into w 1. 1. then there exists a local diffeomorphism of the ambient space taking two submanifold germs over into one another together with the restrictions onto them of the symplectic structure of the ambient space. possibly a degenerate one. 1. If the submanifold is a point.3. 1. The Relative Darboux Theorem I.
Theorem. q 1.) = oc. where ivw is the inner product of the field and the form and oc is a 1form defined by the condition doc= w 1 w0 uniquely up to addition of the differential of a function. in c') at points on a smooth curve. • . then oc = 2: (xioci + }.. where the oci are 1forms and the }.. this equation is uniquely solvable for an arbitrary 1form oc. . p 2 . In view of the nondegeneracy of the symplectic structures w. P3 = Ptq2 ± qtq3. Symplectic Geometry 25 theorem". sect. e. 0 1.w1 + (w 1 w 0 ) = 0 (here Lv is the Lie derivative). Let x 1 = . 1. and f are functions. Therefore it remains for us to show that the form oc can be taken to be zero at the points ofthe subspace X. Here we have run into the realization question: what is the smallest dimension of a symplectic space in which a given degeneracy of a closed 2form can be realized as the restriction of the symplectic structure onto a submanifold? The answer is given by The Extension Theorem. 1).2 can be realized as the restriction of the standard symplectic structure on ~ 6 to the germs at 0 of the following 4dimensional submanifolds (p 1. of vector fields equal to zero on X and satisfying the homological equation Lv. Then the w. to nearby forms one may associate nearby extensions). q 3 are the Darboux coordinates in ~ 6 ): a') P3 = q3 = 0.w1 + oc = 0.dxJ + df. P1 = p~/2. Thus. = (1. Consequently. and is unstable at isolated (parabolic) points. Since the form w 1 w0 is equal to zero on X.d(f + l:J. b') q 3 = 0. or.4.t)w 0 + tw 1 are symplectic structures in a neighbourhood of the origin for all tE[O. the degeneracies of closed 2forms on ~ 4 enumerated in sect.2 of chap. y 2 nk be the remaining coordinates on ~ 2 ". The Classification of Submanifold Germs. The germ of a generic smooth 4dimensional submanifold in 6 dimensional symplectic space can be reduced by a local symplectic change of coordinates to the form in a') at a general point. p 3 . = xk = 0 be the equations of X and let y 1 .. are closed. . what is equivalent. The operation ofextension can be made continuous in the coo topology (i. c') P2 = qtq2.xd. 1. in b') at the points of a smooth 3 dimensional submanifold. we may pass over to the equation iv. . A closed 2form on a submanifold of an even dimensional manifold can be extended to a symplectic structure in a neighbourhood of some point if and only if the corank of the form at this point does not exceed the codimension of the submanifold.fdepending only on y. 1]. The relative Darboux theorem allows us to transform information on the degeneracies of closed 2 forms into results on the classification of germs of submanifolds of symplectic space. for a family V.d}. q 2 . Since the forms w.q~/3. we may replace oc by the form l:xi(oci. equal to zero at the points of X. We shall look for a family of diffeomorphisms taking w. into w 0 and being the identity on X.
e. There exist examples which show that this condition can not be neglected. consequently. The homotopy exists if the symplectic structures w 0 and w 1 coincide not only on vectors tangent to N. The proof of the global theorem is completely analogous to the one cited above for the local version. i. . The Exterior Geometry ofSubmanifolds. on TNM. in some neighbourhood of NinM. even if one does not require the diffeomorphism g to be the identity on N. A. Arnol'd. but on all vectors tangent to M and applied at points of N. Let M be an evendimensional manifold.5.26 V.I. The question remains open of the dimension of a symplectic manifold Min which a manifold N given together with a closed 2form can be realized as a submanifold. i. In this case the linear combination tw 1 + (1. The Extension Theorem III. In this direction we cite the following result. on TN. over into w 0 luo: g*w 1 = w 0 • The distinctive difficulty in the global case consists in clarifying whether the structures w 0 and w 1 are homotopic in the sense indicated above: the linear combination tw 1 + (lt)w 0 can become degenerate along the way. Givental' 1.t)w 0 will for all t E [0. N a submanifold. Let us suppose that w 0 and w 1 can be continuously deformed into one another in the class of symplectic structures on M which coincide with them on N.one use the following lemma: The Relative Poincare Lemma. The Extension Theorem II. 1] be nondegenerate at the points of Nand.the coordinate argument in the concluding part of the proof. Then there exist neighbourhoods U 0 and U 1 of the submanifold N in M and a diffeomorphism g: U 0 + U 1 which is the identity on Nand which takes w 1 lu. The Relative Darboux Theorem II. A closed differential kform on M equal to 0 on TN can be represented in a tubular neighbourhood of N in M as the differential of a k1form equal to 0 on TNM. Then this field offorms can be extended to a symplectic structure on a neighbourhood of the submanifold N in M. We shall cite global analogues of the preceding theorems. It is only necessary that in place of the "integration by parts". and let w 0 and w 1 be two symplectic structures on M whose restrictions to N coincide. The basis of the proof of this lemma is the conical contraction of the normal bundle onto the zero section (see [73]).e.B. Any manifold N together with a closed differential 2form w can be realized as a submanifold in a symplectic manifold M of dimension 2dimN. Let N be a submanifold of M and on the fibres of the bundle TNM+ N let there be given a smooth field ofnondegenerate exterior 2 forms whose restriction to the subbundle TN defines a closed 2form on N.
The same applies to the content of sect. First we shall introduce on T* M the differential 11orm of action oc. 1. Let~ be a tangent vector to T*M applied at the point p (Fig. § 2. The definition of a symplectic structure and the Darboux theorem can be carried over verbatim to the case of complex analytic manifolds. complex projective manifolds and orbits of the coadjoint action of Lie groups 4 • 2.3 and extension theorem III. Let us now set 1Xp(~) = p(n. Symplectic Geometry 27 For the manifold M it is sufficient to take the cotangent bundle T*N. The projection n: T* N+ N determines a closed 2form n*w on T* N. It turns out that on the cotangent bundle there exists a canonical symplectic structure which is equal to 0 on the zero section and in sum with n*w is again nondegenerate.1. Cotangent Bundles.5 and 3. . n In local r*M X Fig. 1. 3. 10. Whether the remaining results of§ 1 are true in the complex analytic category is not known. 1. applied at the point X. Examples of Symplectic Manifolds In this section we discuss three sources of examples of symplectic manifolds cotangent bundles.2 of chap. manifolds of geodesics) see sects. The projection n: T* M+ M determines a tangent vector n. We shall begin the next section with the description of the canonical structure. The definition of the action !form 4 For other constructions which lead to symplectic manifolds (for example. Let us define a canonical symplectic structure on the space T* M of the cotangent bundle of an arbitrary (real or complex) manifold M. A point of the manifold T* M is defined by giving a linear function p E T"* M on the tangent space TxM toM at some point xEM. ~ to M. 10). but obviously a degenerate one.6. The Complex Case.
from which its nondegeneracy follows. 1. ) onto this orthogonal complement uniquely determines a Hermitian form on the tangent space to eOJ>n. any U ninvariant exterior 3form on the realification of the space en is equal to zero. In particular. Pn are coordinates on T: M dual to the coordinates dq 1 . The generalized momentum is the threedimensional vector of angular momentum of the body relative to the fixed point. An example: the configurations of a rigid body fixed at a point form the manifold S0(3). • •• . Therefore all that remains is to produce the promised Hermitian metric on eOJ>".zkik and let wk = zk/z 0 be an affine chart on eOJ>". Since the group U ncontains multiplication by . 1).1 of chap. Then the form w is proportional to 5 See volume 3 of the present publication. The points of complex projective space eOJ>n are onedimensional subs paces in en+ 1• A complex projective manifold is a nonsingular subvariety in eOJ>" consisting of the common zeroes of a system of homogeneous polynomial equations in the coordinates (zo. Arnol'd. there exists a symplectic structure. . It is the imaginary part of the restriction toM of the original Hermitian metric on eOJ>". from which it follows that the differential 2form w is closed.I. which acts on the tangent space to this point. p) on T* M. And just this is our canonical structure. The Hermitian metric constructed in this way on eOJ>" is invariant with respect to the unitary group Un+I of the space en+I. Cotangent bundles are constantly exploited by classical mechanics as phase spaces of Hamiltonian systems 5 . For this let us consider the Hermitian form ( . The base manifold Miscalled the configuration space. A... In explicit form. ) on the space en+ 1 • The tangent space at a point ofthe manifold eOJ>n can be identified with the Hermitian orthogonal complement of the corresponding line in en+ 1 uniquely up to multiplication by ei<P. Therefore the differential 2form w = doc gives a symplectic structure on T* M. It is constructed in this way. Therefore the restriction of the form ( . z) = L. dqn on TxM.B. the function p is called the generalized momentum of a mechanical system having the "configuration" x = n(p). ' zn) of en+ 1. 2. The restriction of this symplectic structure to a projective manifold M c eOJ>" gives a closed 2form on M. Complex Projective Manifolds. Therefore the imaginary part w of our Hermitian metric and its differential dw are Un+ 1  invariant.l. 28 V. • ••• . First a symplectic structure is introduced on eOJ>" as the imaginary part of a Hermitian metric on eOJ>" (compare sect. 'regarded as a 2kdimensional real manifold. It turns out that on an arbitrary kdimensional complex projective manifold. . the form dw is invariant with respect to the stabilizer Un of an arbitrary point in eOJ>".2. let (z. Givental' coordinates (q. where p 1 . the 1form oc has the form oc = LPkdqk.
On the other hand. • . qz)~+(pi + 1. di 1 . By a Kahler structure on a complex manifold is meant a Hermitian metric on it whose imaginary part is closed.e. qi.q. . Symplectic and Kibler Manifolds. a complex submanifold of a Kahler manifold is itself Kahler. q 2 the action of the group generated by the following symplectomorphisms: b: Pz~+Pz + 1. Kahler manifolds have characteristic geometric properties. Pz. Symplectic Geometry 29 where a. a symplectic structure on a manifold can always be strengthened to a quasiKahler structure. In another way this action may be described as the left translation in the group G of matrices of the form (1) by means of elements of the discrete subgroup G1 of integer matrices. The complex projective manifolds form a subclass of the class of Kahler manifolds. IR) of the tangent bundle of a symplectic manifold to the unitary group Un. the Hodge decomposition holds in the cohomology groups of a compact Kahler manifold M: Hk(M. Let us consider in the standard symplectic space IR 4 with coordinates p 1 .q consists of the classes representable by complexvalued closed differential kforms on M of type (p. This follows from the contractibility of the structure group Sp(2n. q 1 . qi.•• . d: (pi. [73]). . Thurston. i. Example (W. qz + Pz). q). a complex structure on the tangent bundle and a Hermitian metric whose imaginary part is closed. . . dzn. is a symplectic structure. i. The coefficient is chosen so that the integral over the projective line CIP' 1 c CIP'" will be equal to l. Just as in the preceding item. q = EB p+q=k Hp. 2. There exist compact symplectic manifolds which do not possess a Kahler structure.3.e. The latter means that in the basis dz 1 . Pz. p 2 . din ofthe complexified cotangent space the form appears as a linear combination of exterior products of p of th~ dzi and q of the dzj. where Hp. In particular. a are the differentials with respect to the holomorphic and anti holomorphic coordinates respectively.
therefore dime H 1 (M. Corollary.4. The differential of this action defines the adjoint representation Ad: G+ G L (Te G) of T: the group on its Lie algebra. where g~ = {x E g I ad:~= 0} is the annihilatorofthe function ~On the space gjg~ there i:. G1 ]. (ad:~)ly = ~([y. The fundamental group n 1(M) is isomorphic to G1 . [44]). therefore M is compact. The orbits of the coadjoint action are evendimensional. A. Givental' Therefore the quotient space M = G/G1 is a smooth symplectic manifold. The commutator group [Gz.:.x]). therefore the orbits of the adjoint action have a symplectic structure.30 V. The action of the group on itself by conjugations has the fixed point e E Gthe identity element of the group. and we get a new definition of the symplectic structure on complex projective space introduced in sect. ~Eg*. Theorem. The symplectic structure on the orbits of the coadjoint action plays an important role in the theory of Lie groups and their representations (see [ 41 ]. The Orbits of the Coadjoint Action of a Lie Group. Every orbit of the coadjoint action of a Lie group possesses a symplectic structure. Example. .G1 /[Gz. yEg. ] is the commutator in the Lie algebra g.I. The Lie algebra of the unitary group consists of the skew Hermitian operators on en+ 1 . Let G = U n + 1 • The adjoint representation is in this case isomor phic to the coadjoint one. C) = 3the dimension of the one dimensional cohomology space of M is odd! But from the Hodge decomposition it follows that for a Kahler manifold the cohomology spaces in the odd dimensions are evendimensional.. well defined the nondegenerate skewscalar product ~([x. It is constructed in the following manner. 2. Arnol'd.1 = a. 2.2. The corresponding adjoint representation ad: g+ End(g) and coadjoint re presentation ad*: g+ End(g*) of the Lie algebra are given explicitly by the formulas adxy = [x.y]). a") . The orbit of a skewHermitian operator of rank 1 is isomorphic to CIP>n.1 d. and the group H 1 (M. G1 ] is generated by the element bdb. X. The closedness of the thus obtained symplectic form on the orbit follows from the Jacobi identity on the Lie algebra. g = TeG its Lie algebra. yEg. The dual representation Ad*: G+ G L( G) on the dual space of the Lie algebra is called the coadjoint representation of the group. The functions q 1. Let G be a connected Lie group. The mapping x~+ad: ~ identifies the tangent space to the orbit of the coadjoint representation at the point ~ E g* with the space g/g~. X.B. p 2 (mod a") give a mapping M+ T 2 which is a fibration over the torus T 2 with fibre T 2 . where [ .y].
bilinearity holds.1 this construction is set forth and the properties of the "conservation laws" of classical mechanics are formulated in the language of the Poisson bracket. Corollary 2. Let H be a smooth function on M. as do anticommutativity {H. according to the rule w(·.e. .{ F. H}. {G. The vector field IdH is called the Hamiltonian field with Hamilton function (or Hamiltonian) H. a correspondence between differential 1 forms and vector fields on M. G}} =0. F 2 }. i. Let M be a symplectic manifold. This terminology is justified by the fact that in Darboux coordinates the field I dH has the form of a Hamiltonian system of equations: p= fJHjfJq.e. Je)= ~ (").fJFjfJp"fJHjfJq"). 3. In Darboux coordinates {H. Corollary (Liouville's Theorem). A w. Corollary 1. The phase flow of a smooth vector field on M preserves the symplectic structure if and only if the vector field is locally Hamiltonian. Theorem A. Let us define the Poisson bracket { . The Poisson Bracket A symplectic structure on a manifold allows one to introduce a Liealgebra structurethe Poisson bracketon the space of smooth functions on this manifold. which takes as its foundation the properties of the Poisson bracket. F} =w(ldH. The Leibniz formula is valid: {H. Ft}F 2 +F 1 {H... In sect. i. Theorem A signifies that the Lie subalgebra of vector fields whose flows preserve the symplectic structure goes over into the space of closed 1forms under the isomorphism I. The Lie Algebra of Hamiltonian Functions. The vector fields on a manifold form a Lie algebra with respect to the operation of commutation. IdF). The commutator of two locally Hamiltonian fields v1 and v2 is a Hamiltonian field with the Hamiltonian w(vh v2 ). H}} + {G. The commutator in the Lie algebra ofclosed 11orms on M has the form [a.1• Theorem B. q=fJHjfJp. {H. The Hamiltonian functions form a Lie algebra with respect to the Poisson bracket. . {F. Ip). F 1 F 2 }={H. The remainder of the section is devoted to an important generalization of the concept of a symplectic structure. } on the space of smooth functions on M by the formula {H. A Hamiltonian flow preserves the phase volume w A . The application of these formulas to Hamiltonian mechanics is based on the following obvious fact. F} = . and the Jacobi identity {F. .1. Symplectic Geometry 31 § 3. The skewscalar product w gives an isomorphism I: T* M+ TM of the cotangent and tangent bundles. F} = 'fJoHjfJp"fJF /oq". PJ =dw(Ia. F}} + {H. 3.
g)~+ W(f. From the Leibniz rule it follows that the Poisson bracket of an arbitrary function with a function having a second order zero at a given point vanishes at this point. like a symplectic one. W(g. Corollaries. If in a mechanical system two components M 1 . W] (f. depending smoothly on the point of application. .32 V. h))+ W(J. A smooth field W of exterior 2{orms on the cotangent spaces to a manifold gives a Poisson structure on it if and only if [ W. A Poisson structure on a manifold is a bilinear form { .. where the dots denote the terms with cyclically permuted J.) A (il/ilxi). the Jacobi identity and the Leibniz rule (see corollary 2 of the preceding item). 3. } on the space of smooth functions on it satisfying the requirement of anticommutativity. Poisson Manifolds. M 2 of the angular momentum vector are conserved. A. W] = 0. M 2 } will be conserved. g. F}. V(g. In coordinates a Poisson structure is given by a tensor Iwii(il/ilx. Noether: A Hamiltonian F whose flow preserves the Hamiltonian H is a first integral of the flow with Hamiltonian H. On a manifold let there be given two smooth fields V and W of exterior 2 forms on the cotangent bundle. j. that is. W] the smooth field of trilinear forms on the cotangent bundle defined by the following formula [ V. a) The law of conservation of energy: a Hamiltonian function is a first integral of its Hamiltonian flow. defines a homomorphism of the Lie algebra of smooth functions into the Lie algebra of vector fields on the manifold: . h))+ . g and h.2. where the wii are smooth functions satisfying the conditions: for all i. c) The Poisson theorem: The Poisson bracket of two first integrals of a H ami/ton ian flow is again a first integral. The derivative of a function F along a vector field with a Hamiltonian H is equal to the Poisson bracket {H. k wii= wii and A Poisson structure. therefore a Poisson structure defines an exterior 2form on each cotangent space to the manifold. Example.B. The first two properties of the Poisson bracket mean that it gives a Lie algebra structure on the space of smooth functions on the manifold. Conversely. Amol'd. the value of such a 2form on a pair df. then also the third one M 3 = {M 1. b) The theorem of E.. Givental' Theorem C. g) of the functions.I. h)= V(J. Lemma. dg of differentials of functions defines a skewsymmetric biderivation (J. We designate as their Schouten bracket [V. a bilinear skewsymmetric operation satisfying the Leibniz identity with respect to each argument. This form we shall still call a Poisson bracket.
The Splitting Theorem ([74]). the Casimir functions are the invariants of this action. The Foliation Theorem ([ 42]. The latter is uniquely determined up to isomorphism of germs of Poisson manifolds. The equivalence class of an arbitrary point of a Poisson manifold is a symplectic submanifold of a dimension equal to the rank of the Poisson structure at that point. A transversal to a symplectic leaf at any point intersects the neighbouring symplectic leaves transversally along symplectic manifolds and inherits a Poisson structure in a neighbourhood of the original point. At a point x of rank 0 on a Poisson manifold there is well defined the linear approximation of the Poisson structurea linear Poisson structure on the tangent space of this point (or a Lie algebra structure on the cotangent space): [dxJ. [74]). The following theorem aids in understanding the structure of Poisson manifolds and their connection with symplectic ones. Such fields are called Hamiltonian. which in aggregate determine the Poisson structure: the Poisson bracket of two functions can be computed over their restrictions to the symplectic leaves. Symplectic Geometry 33 the derivative of a function g along the field of the functionfis equal to {/. g}. In contrast to the nondegenerate symplectic case. the centre need not consist only of locally constant functions. A Poisson structure on a vector space is called linear if the Poisson bracket of linear functions is again linear. The germ of a Poisson manifold at any point is isomorphic (as a Poisson manifold) to the product of the germ of the symplectic leaf with the germ of the transversal Poisson manifold ofthis point. The vectors of Hamiltonian fields generate a tangent subspace at each point of the Poisson manifold. their flows preserve the Poisson structure. Its dimension is called the rank of the Poisson structure at the given point and is equal to the rank of the skewsymmetric 2form defined on the cotangent space. g}. This theorem reduces the study of Poisson manifolds in the neighbourhood of a point to the case of a point of rank zero. dxg] =dx{J. a Poisson manifold breaks up into symplectic leaves. 3. Let us call two points of a Poisson manifold equivalent if there exists a piecewise smooth curve joining them. Thus. Linear Poisson Structures. . each segment of which is a trajectory of a Hamiltonian vector field.3. Hamiltonians to which zero fields correspond are called Casimir functions and they form the centre of the Lie algebra of functions. the symplectic leaves of the linear structure are the orbits of the coadjoint action of this Lie algebra. A linear Poisson structure on a vector space is precisely a Lie algebra structure on the dual space.
34 V. The linear approximation of the transversal Poisson structure to the orbit of the coadjoint action at the point is e canonically isomorphic to the linear Poisson structure on the dual space of the annihilator g~. and the theorem follows from 2°. Corollary 1. r. We shall call a Lie algebra g (analytically. The Linearization Problem. A linear Hamiltonian system on IR 2 " has n linearly independent quadratic first integrals. Corollary l is obtained by an application of 3o to the annihilator 9x of the element x e g with respect to the adjoint action. The annihilator of a generic element E g* is commuta tive. The isomorphism is given by the mapping g* jad:e+ gt of the spaces of definition of the linear Poisson structures being regarded which is dual to the inclusion g~ <:+ g. while the answer to this question may depend on the category in which the linearization is carried out. let e g*. therefore the adjoint representation of a semisimple Lie algebra is isomorphic to the coadjoint one.). This form is invariant with respect to the adjoint action. sp(2n. Indeed. By the rank of a Lie algebra is meant the codimension of a generic orl>it of the coadjoint action (i.4. Duflo's Theorem. In connection with Poisson structures this problem can be stated thus: is a Poisson structure isomorphic in a neighbour hood of a point of rank 0 to its linear approximation at that point? All the previous resuits of the section were equally true both in the smooth and in the holomorphic case.B. IR) is a simple Lie algebra of rank n. C 00 ) sufficient if any Poisson structure whose linear approximation at a point of rank 0 is isomorphic to g* is itself (analytically. A. It follows from the annihilator theorem that the rank of the annihilator g~ of an e arbitrary E g* is not less than the rank of g. the co rank of its Poisson structure at a generic point). Each element of a semisimple Lie algebra of rank r is contained in an rdimensional commutative subalgebra. the symplectic leaves of the transversal Poisson structure at a generic pointe are zerodimensional. in view of the fact that x lies in the centre of the algebra gx. In fact. 4 o. D 3. . Let g be a Lie algebra. Amol'd. C 00 ) isomorphic to g* in a neighbourhood of this point. D Corollary 2.e. and let g~ = e {xeglad!e=O} be the annihilator of in g. y) = tr(adxo ad. Proof l 0 • One of the possible dennitions of a semisimple Lie algebra consists in the nondegeneracy of its Killing form (x. This definition is in keeping with the general approach to linearization problems in analysis: linearizability is treated as a property of the linear approximation. Givental' e The Annihilator Theorem. Indeed. Let us note that the dimension of such an annihilator is equal to the rank of the Lie algebra g.I. upon linearization the codimension of a generic symplectic leaf can only increase! e 3°.
A fibre of the cotangent bundle is also Lagrangian and corresponds to the "delta function" of the point of application. a submanifold of a symplectic manifold is called (co)isotropic 6 if its tangent spaces are so. Therefore linearizability modulo terms of sufficiently high order follows from the theorem of LeviMal'tsev. Symplectic Geometry 35 The description of sufficient Lie algebras is an open problem. We shall speak of the graph of the 1form to mean the totality of these covectors. Conversely. Lagrangian Submanifolds and Fibrations A submanifold of a symplectic manifold (M 2 ". If this 1form is exact. the restriction of the symplectic form w = d(pdq) to the graph of the 1form ~: qH p(q) is equal to d(p(q) dq) = d~ = 0 if~ is closed. Using the terminology of § 1 of Chapter 1. A smooth curve on a symplectic manifold is isotropic and a hypersurface is coisotropic. 6 Coisotropic submanifolds are also called involutive. It is easy to convince oneself that commutative Lie algebras are not sufficient in either of the two indicated senses. of a semisimple complement to its radicalthe nilpotent ideal consisting of the functions with a secondorder zero at the coordinate origin. A 1form on a manifold associates to a point of the manifold a covector at that point. A semisimple Lie algebra of a compact group is C"'sufficient. 4. . The graph of a closed 1form on X is a Lagrangian submanifold of M. More generally. The solvable twodimensional Lie algebra of the group of affine transformations of the line is sufficient in each of them. which asserts the existence of such a complement in the case of finitedimensional Lie algebras. a Lagrangian submanifold in M which projects onetoone onto the base is the graph of a closed 1form on X. w) is called Lagrangian if it has dimension nand the restriction to it of the symplectic form w is equal to 0. The semisimple Lie algebra si 2 (1R) of the group of symplectomorphisms of the plane is analytically sufficient. § 4. but is not C""sufficient. A semisimple (real or complex) Lie algebra is analytically sufficient. one may say that Lagrangianity of a submanifold is just Lagrangianity of its tangent spaces.1. let ex be the action 1form on M and w = dcx the canonical symplectic structure. 2) Let M = T* X. Indeed. This example suggests definition of a generalized function on X as an arbitrary Lagrangian submanifold in T* X. 1) A smooth curve on a symplectic surface is Lagrangian. Examples of Lagrangian Manifolds. Theorem ([17]). then its potentiai4J is called a generating function of the Lagrangian submanifold. in the Lie algebra of Hamiltonian functions. Let us observe that linearization of a germ of a Poisson structure with a semisimple linear approximation is equivalent to the selection. ~ = d4J.
2. The symplectic type of an isotropic submanifold Nk c M 2" in its tubular neighbourhood is determined in a oneone fashion by the equivalence class ofthe following 2(n. let us define on it the symplectic structure n 1 *w 1 n 2 *w 2 • Then the graph r c M 1 x M 2 of the symplectomorphism y is a Lagrangian submani fold. From the relative Darboux theorem II (sect. w 1 )+ (M 2 . the co vectors applied at the points of a submanifold Y c X and vanishing on vectors tangent to Y form a Lagrangian submanifold in T* X the "delta function" of the submanifold Y.k)dimensional symplectic vector bundle with base space Nk: the fibre of this bundle at a point x is the quotient space of the skeworthogonal complement to the tangent space T"N by the space T"N itself. Givental' More generally.5) we obtain the Corollary. Let us recall that an affine structure on a manifold is given by an atlas whose transition functions from one chart to another are affine transformations of the . A sufficiently small neighbourhood of a Lagrangian submanifold is symplectomorphic to a neighbourhood of the zero section in its cotangent bundle. If a Hamiltonian function is constant on an isotropic submanifold. 4) Let there be given a symplectomorphism y: (M 1 . 3) A locally Hamiltonian vector field on a symplectic manifold M is a Lagrangian submanifold in the tangent bundle TM. A. Properties of this kind will often be used in the sequel without special mention. Denoting by n 1 and n 2 the projections of the direct product M 1 x M 2 onto the first and second factors. A cotangent bundle is a Lagrangian fibration. Lagrangian Fibrations.5. 4. chap.1. 1. Conse quently (see sect. The corank of the restriction of the symplectic form to a (co)isotropic submanifold is equal to its (co)dimension and is therefore constant.I. Arnol'd. These assertions follow from the fact that the one dimensional kernel of the restriction of the symplectic form to the level hypersurface of a Hamilton function coincides with the direction of the vector of the Hamiltonian field. 1. chap. These examples illustrate a general principle: the objects of symplectic geometry are represented by Lagrangian manifolds. 1. 1.B. Example. By a Lagrangian fibration is meant a smooth locally trivial fibration of a symplectic manifold all fibres of which are Lagrangian.36 V. whose symplectic structure is given by its identification I: T* M+ TM with the cotangent bundle. The generating function of such a Lagrangian manifold is the Hamiltonian of the field. then its displacements by the flow of this function sweep out an isotropic submanifold. w 2 ). a germ of a (co)isotropic submanifold reduces in suitable Darboux coordinates to the linear normal form of sect. 1.2. This follows from the results of sect. 2). A coisotropic submanifold on the level surface of a Hamilton function is invariant with respect to its flow.
A Lagrangian fibration together with a polarization is determined uniquely by its action !form rx in a neighbourhood of a point where rx vanishes: the symplectic structure is drx. The Classification of Lagrangian Fibres.e. and the Euler field of homogeneous dilations in the fibres of the cotangent bundle is I rx. It is not hard to verify that this identification is a symplectomorphism. linear transformations or compositions of them). q) a Lagrangian fibration is given as the projection onto the qspace along the pspace. a Lagrangian fibration with a compact fibre is complete. finally. The Affine Structure Theorem. Proof The choice of a Lagrangian section of the Lagrangian fibration identifies it (by the construction out of the proof of the preceding theorem) with the cotangent bundle of the base space. First. 7 That is. isomorphic affine fibrations are not necessarily isomorphic as Lagrangian fibrations (i.1(x). second. The fibres of a Lagrangian fibration have a canonical affine structure. 0 The Darboux Theorem for Fibrations. In suitable local Darboux coordinates (p. affine lines defined locally can be indefinitely extended. Symplectic Geometry 37 coordinate space (parallel translations. A connected component ofa fibre ofa complete Lagrangian fibration is affinely isomorphic to the product of an affine space with a torus. Thus. A function with a zero differential at a point xEX defines a flow which is stationary on the fibre n. We shall call a Lagrangian fibration complete if its fibres are complete as affine manifolds 7 • In particular. Then the functions on X. 0 The assignment of a cotangent bundle structure on a Lagrangian fibration is called a polarization. fibrations with the same fibres can have different global structure. We have seen that in a neighbourhood of a point of the total space of the fibration a Lagrangian fibration is of standard structureequivalent to a cotangent bundle in a neighbourhood of a point of the zero section. with preservation of the symplectic structure). a neighbourhood of any point of the fibre is the local orbit of a fixedpoint free action of the additive group of the space T~X. The global picture is entirely different. Proof Let ·n: M+ X be a Lagrangian fibration. considered as Hamiltonians on M. . give commuting Hamiltonian flows there. the fibres of a Lagrangian fibration are not obliged to be isomorphic as manifolds with an affine structure. the zero section consists of the points where rx vanishes.
simply connected fibre is isomorphic to a twisted cotangent bundle. a" on X. Poincare to the following theorem. ••• . Proof The identification of the fibre T" over a point x of the base space X with an orbit of the group T~X gives on T~X a lattice of maximal rank. where ¢ is a closed 2form on the base space X. . Any complete Lagrangian fibration with a connected. The symplecticity of translations by the lattice vectors in T* X means that these forms are closed. The problem of the existence of periodic motions of dynamical systems led H. by the construction of the affine structure theorem we identify M with T* X. Theorem. therefore the class [¢]EH 2 (X. We have already encountered this construction at the end of§ t.3. Intersections of Lagrangian Manifolds and Fixed Points of Symplecto morphisms. ¢n): X+~" of this basis. ~) uniquely up to an isomorphism of Lagrangian fibrations (which is the identity on the base space). defines a chart of the desired atlas. The local potential (¢ 1. the completeness of the fibration means that every component of the fibre is a quotient space of the group of translations ~"by a discrete subgroup. Let us now describe the complete Lagrangian fibrations with an affine fibre.38 V. Arnol'd. A connected compact fibre of a Lagrangian fibration is a torus. A change of the zero section by s' changes ¢ by ds'. 0 We have already come across an example of a topologically nontrivial Lagrangian fibration with a compact fibre in sect. in some atlas on the base space the transition functions are compositions of translations and integral linear transformations on ~"). ~)is the only invariant of the twisted cotangent bundle. w)+X be a Lagrangian fibration whose fibres are affine spaces.It turns out the compactness ofthe fibre of a Lagrangian fibration imposes very stringent conditions on its base space. The base space of a Lagrangian fibration with a connected compact fibre has a canonical integral affine structure (in other words.I. 0 4. A twisted cotangent bundle is a cotangent bundle n: T* X +X with a symplectic structure on T* X equal to the sum of the canonical symplectic form and a form n*¢. A section s: X e+ M defines a closed form ¢ = s*w and w n*¢ is a symplectic structure on M in which s(X) is Lagrangian. d¢i=ai. The only such subgroups are the lattices z_k c ~". Proof. k ~ n. Taking s as the zero section. A continuous basis of such a lattice is a set of 1forms a 1. Theorem. a Lagrangian fibration over the torus T 2 with fibre T 2 . Givental' Indeed. A twisted cotangent bundle is determined by the cohomology class of the form ¢ in H 2(X. .3 on Kahler structures: we presented a symplectic manifold M 4 . 2.B. Let n: (M. Corollary. ••. not homeomorphic to T 2 x T 2 . A.
where X. A symplectomorphism which is homologous to and close (together with its first derivatives) to the identity has a singlevalued generating function. where u =(X+ x)/2. A fixed point of yis an intersection point of r with the graph A of the identity symplectomorphism. Y=y+g(x. Poincare's argument is based on the fact that the fixed points of a symplectomorphism of the annulus are precisely the critical points of the function F(u. The proof of this theorem is due to G. If the symplectomorphism y is sufficiently close to the identity. The symplectomorphisms homologous to the identity form the commutator subgroup of the identity component of the group of symplecto morphisms of the manifold M. Carrying over Poincare's arguments to the general symplectic situation leads to the following concepts and results. and the functions!. Conjecture A. v)fo(x.y). by a manyvalued one) of the Lagrangian section r c T* A. x are radial coordinates andY. Lemma. The boundary condition in the theorem means that the mapping has the form X=x+f(x. Their number is not les5 than the number of critical points which a smooth function on this manifold must have at the very least. A homeomorphism ofa plane circular annulus onto itself which preserves areas and which moves the boundary circles in different directions has at least two fixed points.dv. yare angular coordinates on the annulus. A symplectomorphism of a compact symplectic manifold which is homologous to and close to the identity has fixed points. Poincare succeeded in proving it only under certain restrictions.gdu). but his method is less special than Birkhoff's proof and lends itself to generalization. v = ( Y + y)/2. H. A symplectomorphism y: M+ M is given by a Lagrangian graph r c M X M. The preceding theorem is true without the condition of closeness of the symplectomorphism to the identity. Its critical points and the fixed points of ycoincide. and in the case of the sphere with handles not less than 3 fixed . and moreover g has different signs on the different boundaries of the annulus. Theorem A. y) is different from zero. Symplectic Geometry 39 Poincare's Geometric Theorem. This condition is automatically fulfilled if the symplectomorphism is not too far from the identity. Birkhoff. then it is given by a generating function (generally speaking. A tubular neighbourhood of A in M x M has the structure of the Lagrangian bundle T* A. The conjecture has been proved in the twodimensional case: a symplecto morphism of a compact surface homologous to the identity has for the sphere not less than 2.D. Let M be a symplectic manifold.y). g are continuous and 2n periodic in y. We shall say that a symplectomorphism y is homologous to the identity if it can be connected with the identity by a family of symplectomorphisms whose velocity field for each value of the parameter of the family is globally Hamiltonian. v) = J<f. true under the assumption that the Jacobian o(u.
Shnirel'man and N.I. Let us note that a smooth 8 The first is due to A. Amol'd. Theorem B. The proof of these theorems 8 bears an essentially twodimensional character. The latter is not difficult to achieve if one adds connective strips between the glued together annuli and extends the symplectomorphism so as not to have fixed points on these strips (here it will be necessary to use the rotation of the boundaries of the annulus in different directions). Thus.A. conjecture A has been proved for symplectomorphisms of the torus T 2 " = IR 2 " /7L 2 " with the standard symplectic structure. Givental' points. The sym plectomorphisms of the torus IR 2 j7L 2 homologous to the identity are precisely those which leave in place the centre of gravity of the unit square in IR 2 [2]. It turns out that the absence of intersection points of a compact Lagrangian submanifold with its Lagrangian perturbation means that this manifold is fibred over a circle.B. the second to Ya. . Nikishin. Let us observe that Poincare's theorem on the symplectomorphism of the annulus follows from the (proven) conjecture A for the twodimensional torus. By a new method progress was recently achieved in the proof of conjectures A and Bin the manydimensional case.I. Conjecture (theorem) A follows out of conjecture (theorem) B: the graph r c M X M of a symplectomorphism y: M> M homologous to the identity is a Lagrangian submanifold isotopic to the diagonal ~ Without the isotopy assumption intersection points of a compact Lagrangian submanifold with a Lagrangian perturbation of it (of which it is the question in theorem B) may be absent. Namely. A compact Lagrangian submanifold isotopic to and close to a given one has as many points of intersection with it as some smooth function on this manifold has critical points. Theorem B is true without the assumption of closeness of the isotopic Lagrangian submanifolds. just as conjecture B has been proved for the Lagrangian torus T" embedded in the standard symplectic torus T 2 " as a subgroup and for an arbitrary Lagrangian isotopy of the torus T" [ 19]. A. Nevertheless intersection points necessarily exist if the Euler characteristic of the Lagrangian manifold is different from 0 or if each closed 1form on this manifold is the differential of a function. We shall call two Lagrangian submanifolds of a symplectic manifold N (Lagrangianly) isotopic if one goes over into the other under the action of a symplectomorphism homologous to the identity.M. Ehliashberg.40 V. isotopic Lagrangian submanifolds are diffeomorphic. in the case of the sphere the proof is based on the fact that the index of an isolated singular point of a Hamiltonian vector field on the plane is equal to one minus onehalf the number of components into which the critical level curve of the Hamiltonian divides a neighbourhood of the singular point and hence does not exceed I. In particular. Out of two copies of the annulus one can glue together a torus. Conjecture B.
we shall give a sketch of the proof of the following assertion. Thus the function!. IR 2"+IR 2 " is the operator "multiplication by J(p. 2°. 4°. The spectrum of the operator A is 2nZ.J=l". is the 2ndimensional space of solutions with period l of the linear Hamiltonian system with Hamiltonian ). The second has the form Aw+PB(v+w)=O. the set of points where the value . where Pis the projector ofO onto W. I< 2D. let us consider the vector space 0 of loopssmooth mappings of R/Z + R. qdp)/2 . t) is the given periodic Hamiltonian with period 1. The equation VF=O splits into two: oF/oV=O and oFfoW=O. q)=( q. Theorem. The expansion of a loop IR/Z+IR 2 " by eigenfunctions of the operator A is simply its Fourier series expansion. Then the symplectic transformation of the torus after a period by the flow of this Hamiltonian has at least one . counting them with multiplicities [24]. i. where J: . The gradient VF: 0+0 has the form VF= JdfdtVH. We regard the mapping VF as a perturbation of the linear operator A= . the various fixed points of the symplectomorphism we are interested in. p). therefore the functionfis periodic on the subspace of such loops. JR/Z where H(p.(p 2 +q 2 )/2. correspond the various periodic trajectories with period 1 of the flow of the Hamiltonian H. Therefore the search for extrema of the functional F reduces to the determination of the critical points of the function f(v) = F(v. Dllwl wzll 30. q. On the standard symplectic 2ndimensional torus let there be given a Hamiltonian depending periodically on time. the second equation has for each v a unique solution w = w(v). ll).2n. and the eigenspace with the eigenvalue A. Symplectic Geometry 41 function on thendimensional torus has no fewer than n + 1 critical points and no fewer than 2". To the various critical points of the functional F : Q + R. and an infinitedimensional space W of "higher harmonics". from which the presence of critical points follows (for example.H dt].e. w(v)) on the finitedimensional space. Let us represent the space n in the form of a direct sum of a finite dimensional space V. which form the kernel of the operator A. correspond to the zero solution of the equation oF/oW=O.J d/dt.1 PB(v+w) is contracting. essentially. is defined on the manifold T 2 " x !Rf>o x !Rf<o· Its behaviour at infinity is determined by the unperturbed functional and has hyperbolic character (Fig.fixed point. The constant loops. Since the operator wH A . To demonstrate the essence of the method.with the vector space structure of pointwise addition and scalar multiplication and with the Euclidean structure J (p 2 + q 2 ) f2dt. spanned by the "lower harmonics" with frequency IA. The perturbation B=VH has a bounded image (IIB(w)II<C) and a bounded norm: IIB(wl)B(wz)ll:::. l Representing T 2" as the quotient space IR 2 " /Z 2 " of the standard symplectic 0 • space IR 2 ". On the space lR/Z Q let us introduce the action functional F = J [(pdq .
Novikov in this volume.M. and we shall describe a procedure for reducing the order of Hamiltonian systems having a continuous symmetry group.B. Givental' Fig. I. Krichever and S. A. we shall give a geometric introduction to the theory of completely integrable systems. in particular.I. Remark. II. Chapter 3 Symplectic Geometry and Mechanics Here we shall examine the connection of symplectic geometry with the variational calculus. Further analysis of this finitedimensional situation with tools developed in Morse theory ([55]) leads to an exact estimate of the number of fixed points ( ~ 2n + l ). Dubrovin. The fundamental difference of the argument cited to the standard formalism of the global variational calculus consists in the fact that the unperturbed quadratic functional on the loop space is not elliptic but hyperbolic.P. . and for the theory of integrable systems see the article by B. Variational Principles The motions of a mechanical system are the extremals of a suitable variational principle. any problem of the calculus of variations can be formulated in the language of symplectic geometry. For a systematic exposition of the questions of classical mechanics see volume 3. The change of structure of the level topology of the function f of the function is less than a given value undergoes a topological change of structure). Arnol'd. On the other hand.A.42 V. § 1. with Lagrangian mechanics.
A natural mechanical system is given by a kinetic and a potential energy. The tangent space to the configuration manifold at each point can be identified with the space ~ 3 : the direction of the vector we ~ 3 indicates the axis and the direction of the infinitesimal rotation of the body. Lagrangian Mechanics. the kinetic energy" is a Riemannian metric on the configuration manifold. Example. Thus the free rotation of a rigid body fixed at a point is described by the geodesic flow . Out of the kinetic and potential energies one composes the Lagrangian or Lagrange . for example.e. we obtain the definition of a Lagrangian mechanical system. admitting into consideration. i. For example. where I is the moment of inertia of the body with respect to the axis of rotation. 1. One may also assume an explicit dependence of the Lagrangian on time./lrl. Let us consider an absolutely rigid body one of whose points is fastened at the coordinate origin of the space ~ 3 • The configuration manifold of such a system is the rotation group S0 3 . q(t)) dt II (I) to More generally. the kinetic energy is given in the internal coordinates of the body by the inertia quadratic form.the configuration manifolds of such systems are no longer domains of coordinate spaces. A system of mass points in Euclidean space has the kinetic energy T='LmiV2 and the potential energy V='LVk 1(rkr 1). The kinetic energy of the rotation is T =I w 2 /2. i. Symplectic Geometry 43 I. I. A motion t f+q(t) of a natural system in the configur ation space is an extremal of the functional J L(q(t). At the same time. where rk is the radius vector of the kth point and vkl is the potential of the pairwise interaction of the mass points.U on the total space of the tangent bundle of the configuration manifold. Lagrangian mechanics generalizes Newtonian mechanics. systems of mass points with holonomic (rigid) constraints. if we consider an arbitrary Lagrange function L: TM+ ~ on the tangent bundle of the configuration manifold and if we designate as motions the extremals of the functional (l ). a positive definite quadratic form on each tangent space to the manifold of configurations depending smoothly on the point of application. the Lagrangian approach to mechanics permits considering it as a special case of the variational calculus. For a system of mass points in Euclidean space the EulerLagrange equations take the form of the system of Newton's equations mkrk= iJUjork. and the length of the vector indicates the angular velocity of the rotation.function L = T. the = problem of the "free" motion of a natural system ( U 0) is equivalent to the description of the geodesic flow on a configuration manifold with Riemannian metric T (see sect. Example. say the Newtonian gravitational potential Vkl(r)= ymkm.3). The extremals of the functional (I) are locally described by the system of EulerLagrange differential equations (d/dt)iJLjoq = oLjoq.e. Thus. The potential energy is a smooth function on the manifold of configurations (states) of the system.
44 V.L(q.B. chap. Arnol'd. 12. Givental' on the group S0 3 of a Riemannian metric (left)invariant with respect to the translations on the group.1.4=HP.f*(p)=max [ (p.I. 2). The system of EulerLagrange equations of a natural system with configuration manifold M. We shall suppose that the mapping TqM>T.1. v).[* of the dual argument p by the formula. In Darboux coordinates the system of Hamilton's equations has the form p= Hq. H (p. 1. Theorem. A.f(v)] . Hamiltonian mechanics generalizes Lagrangian mechanics. For example. 5). Hamiltonian Mechanics. q) = p4. chap.. The Legendre transformation . where 4 is determined from the equation p= L 4(q. 4 ). v )/2 is (p. Under the indicated identification of the total spaces of the tangent and cotangent bundles the mechanical system with Lagrangian . under the isomorphism of the tangent and cotangent bundles of the manifold M defined by the Riemannian metric 2T. (Fig. The motion in the Hamiltonian system is described by the phase flow of the corresponding Hamiltonian vector field (see sect.1 p )/2. Example l. 1. kinetic energy T and potential energy U is converted. 3. A Hamiltonian mechanical system is given by a smooth functionthe Hamiltonianon a symplectic manifold (the phase space). A Hamilton ian H depending explicitly on time gives a nonautonomous Hamiltonian system.function L goes over into the Hamiltonian system with Hamiltonian H. A. the Legendre transformation of the Euclidean form (Av. The Legendre transformation of a convex function f of the vector argument v is defined as a function. 12.M: 4~+p=L4 is a diffeo morphism for each qEM. In the general case let us define the Hamiltonian H: T* M> IR as the fibrewise Legendre transformation of the Lagrangian L: TM> IR.2. compare sect. 4). into a system of Hamilton's equations with Hamiltonian T + U with respect to the canonical symplectic structure on the total space of the cotangent bundle. Then the Hamiltonian His a smooth function on the total space of the cotangent bundle. Plf 1! Fig.
Let the Lagrangian be the sum of the Lagrangian of a natural system and a differential 1form A on the configuration manifold M. The proof of the principle of least action .Ja (Fig. The Principle of Least Action. The integral curves of the field of characteristic directions of a nonsingular hypersurface r c T* B transversal to the fibres of the cotangent bundle T* B+B are extremals of the action integral Jain the class of curves lying on rand joining the fibres T:0 B and T:. The fact that the problems of the calculus of variations have a Hamiltonian character is explained by the presence of a variational principle in the Hamiltonian formalism itself.U +A. Proof The increment of the action integral J a. 13. chap.3. then the phase space turns out to be a twisted cotangent bundle (sect. 2).pkdqk be the action 1form on M. Let the symplectic manifold M be polarized: M= T*B. and let a= '[. The corresponding system of EulerLagrange equations on TM is Hamiltonian with Hamiltonian function H = T + U with respect to the symplectic structure Q + dA. 13) is equal to the y' y symplectic area of the sheet joining two curves y andy'. 1.2. and in the case that y is an r r' r 8 Fig. Theorem (The principle of least action). B of the points q 0 and q 1 of the base space B. At the basis of this principle lies the following observation: the field of directions of a Hamiltonian vector field on a nonsingular level hypersurface of its Hamiltonian coincides with the field of characteristic directions of this hypersurfacethe field of skew orthogonal complements of its tangent hyperplanes. 4. where Q is the symplectic structure of example 1. Symplectic Geometry 45 Example 2. regarded as a function on TM linear with respect to the velocities: L = T. If the 1form A is a manyvalued "vectorpotential of a magnetic field" dA defined singlevaluedly on M. A natural system in a magnetic field.
U)ds 2 is defined on the whole compact configuration space S0 3 . One can show [55] that on a compact Riemannian manifold each element of the fundamental group is represented by a closed geodesic. The .B. q. the trajectories of a natural system are also geodesics of a certain Riemannian metric: in the region of the configuration space where U(q) < h the trajectories of a system with kinetic energy T=(dsjdt) 2 j2. Variational Problems with Higher Derivatives.forced to stay on a smooth Riemann manifold moves on geodesic curves (i. A rigid body in an arbitrary potentialfield has at least one periodic motion . A mass point . . In the case where the potential energy is different from zero. . For sufficiently large h the Riemannian metric (h. q(t). 1. on extremals of the length Jds).I.46 V. x<•+ 1 l)dt (2) a within the class of smooth curves x: IR>IR 1 with a given Taylor expansion at the ends of the interval [a. A. D Remark. extremals of the action integral J(pdq. and the action integral takes the form Jpdq = Jpljdt = J2Tdt=fo Jds. t) in the extended phase space IR" * x IR" x IR are . q(td=q 1 • Corollary.for each sufficiently large value of the total energy. We obtain the Corollary. potential energy U(q) and total energy h will be geodesic curves of the metric (h. in the case of free motion with kinetic energy T = (dsjdt) 2 /2 the parameter t ensuring a fixed value of the energy H = T = h must be proportional to the length. In the class of all noncontractible closed curves on S0 3 let us choose a curve of minimal length (this is possible [55]) with respect to the Riemann metric introduced above.. As an application let us consider the rotation of a rigid body around a fixed point in a potential field. . The space S0 3 is not simply connected (it is diffeomorphic to IRIP 3 and has a simply connected double covering by S 3 ).4.Hdt) in the class of curves t~>(p(t). t) to with the boundary conditions q(t 0 )=q 0 . The integral curves of a nonautonomous system of equations with the Hamiltonian function H(p. b] up to order n inclusive. In fact. Givental' integral curve. Let us describe the Hamiltonian formalism of the problem of minimizing the functional JL(x< h 0 l. From this one can obtain an analog of the preceding corollary for an arbitrary natural system with a compact nonsimply connected configuration space. . dt = dsj fo.e. Amol'd. where the Lagrangian L depends on the derivatives x<kl = dkxjdtk of the curve x(t) up to order n +I.U(q))ds 2 . it is infinitesimal to a higher order than the difference of the curves y andy'.
1 1 = TM. z. .1R on the manifold of n + !jets at 0 of curves x: IR __. TnU"lJ". . . .1 go over into their point of application: n.. y. y.o>. w)] dt. (3) which expresses the vanishing of the first variation of the functional (2). x<• + O) with respect to the variable x<• + o. under the condition of convexity of the Lagrangian L(x< 0 l. M. . the system of EulerPoisson equations (3) is equivalent to the Hamiltonian system (H. Pz. w) as a function of a point on a curve (x. 14) as the affine subbundle of the tangent bundle T1" consisting of those tangent vectors eETi. z.. .. z. .1" which under the differential n*: 1j...J"_.. + dpz Adz and with the aid of a Legendre transformation with respect to the variable w let us define the Hamiltonian function H(x. . .dt Lx'" + dt2 Lp. . Symplectic Geometry 47 extremals of the functional (2) satisfy the system of EulerPoisson equations d d2 d"+ I L..py. dz = wdt. Having thus written out the coordinate formulas. .. ..w)+ L(x. and let us put together the action form according to the Lagrange multiplier rule: a=px(dxydt) + . Upon replacement of the space IR 1 by an arbitrary /dimensional manifold Mit is natural to give the functional of type (2) by means of a Lagrange function L: J" + 1 _. +pzW . . " = 0. + p. Pz) is determined from the equation Pz = oLjow).1 of the projection n: 1"_. +pz(dzwdt)+ Ldt = [px(x.1.. .y)+ . z... In addition 1° = M. ... . Now let us regard the Lagrangian L(x. z. w): IR _. we shall now impart an invariant sense both to the variational problem (2) itself as well as to its Hamiltonian version. + (I)"+ I dt" +I L..Px)=PxY+ .y. W) ( W (x.. z... .jR<• + 211 satisfying the restrictions dx = ydt. ..... . .. OJ). The manifold J" + 1 is defined by induction together with the projection J" + 1 _.. Thus.. y. Let us introduce the symplectic form OJ= dpx A dx + ..(e) = j"E T 1"..(i. The extremals of the functional Ja satisfy the system of EulerLagrange equations in iR< 2 "+ 311 : (4) This system is equivalent to the system of EulerPoisson equations (3). Let the function L be convex on each affine fibre of the fibre .....L(x. .. y..1R is the total space T*1" of the cotangent bundle with the canonical symplectic structure.J" (Fig.1". The phase space of the Hamiltonian system corresponding to the Lagrangian L:1"+ 1 _. y. The system of Hamilton's equations with the Hamiltonian H and the symplectic structure OJ on iR< 2 "+ 211 together with the equation Pz=OLjow coincides with (4).
Let us suppose that the integral curves of the field of characteristic directions on a smooth hypersurface in a symplectic manifold form a smooth manifold (locally this is always so). The function H is constant on the characteristics of this hypersurface and defines a smooth function fi on M. Corollary 1. In the case of explicit dependence of the Lagrangian in the functional (2) on time the system of EulerPoisson equations is equivalent to the nonautonomous Hamiltonian system on the extended phase space ~ x T* J".~ as follows. is equivalent to the Hamiltonian system with phase space T* J" and Hamiltonian function H. where L is a smooth function. Let us set H(p)=max(p(w)L(w)). The definition of the jet spaces of curves bundle J"+ 1 . and let M be the manifold of characteristics of a hypersurface F = const.Let us construct a function H: T* J". weW Ostrogradskij's Theorem.]".TJn. Givental' JTII Tf"'. The manifold of characteristics has a symplectic structure (it is well defined by the condition: the skewscalar product of vectors tangent to the hypersurface is equal to the skewscalar product of their projections along the characteristics). 1.d 2 (Liw)>0. strictly convex 9 on each fibre of the fibre bundle J" + 1 +Jft. Remark. . The field of the Hamiltonian H on the hypersurface F = const defines. upon projection onto M. A first integral of a Hamiltonian system allows one to reduce its order by 2. 9 Thatis. 14.. The Manifold of Characteristics.5.1 /l/ l Fig. A. Theorem. We shall call it the manifold of characteristics. The system of EulerPoisson equations which is satisfied by the extremals of the Lagrangian L: r+ 1 +~. Let the Hamiltonian system with Hamiltbnian H have a first integral F.48 V..B. A covector p applied at the point jEJft defines a linear nonhomogeneous function on the fibre We 7Jr of the bundle r+ 1 +J".I. a Hamiltonian vector field on M with Hamiltonian fl. Amol'd.
The Shortest Way Around an Obstacle. 16) consists of straightline segments and a geodesic segment on its surface. Up to the sign of the symplectic structure it is symplecto morphic to the cotangent bundle of the unit sphere in IR".6. The ray space 1. Fig. If the oriented geodesics (the unparametrized ones) of a Riemann ian manifold form a smooth manifold. . The rays (i. a smooth hypersurface 1in L (l is an isotropic manifold) and a smooth hypersurface H in the ambient symplectic space. 1. the oriented straight lines) in Euclidean space IR" form a symplectic manifold. namely the phase space of the corresponding Hamiltonian system (for example.e. The length of the extremals is a manyvalued function of the point q 1 with singularities along the rays breaking loose from the obstacle surface in an asymptotic direction. The shortest path between two points q 0 and q 1 avoiding the obstacle (Fig. The rays on which the extremals issuing from the source break away from the obstacle surface form a Lagrangian variety with singularities in the symplectic manifold of all the rays of the space (compare sect. Example. p) = 1 in T*IR".the manifold of characteristics of the hypersurface (p. Figure 15 shows how to associate to a . The symplectic analysis of the problem of going around an obstacle leads to the notion of a triad in symplectic space. 15. T* J" for problem (2)). A triad (L. tangent to the Lagrangian manifold at the points of the isotropic one.5). Let us regard a smooth surface in space as the boundary of an obstacle. From the theorem we get Corollary 2. 1. Symplectic Geometry 49 The parametrized extremals of a variational problem (possibly a non autonomous one) form a symplectic manifold. H) consists of a smooth Lagrangian manifold L. The projection of the isotropic manifold along the characteristics of the hypersurface is a Lagrangian subvariety in the manifold of characteristics and has singularities at those places where the characteristics are tangent to 1. then it is symplectic.ray a (co)tangent vector to the sphere.
It is helpful to compare the equation of the hypersurface H with the quadratic J Hamiltonian of the functional (dmx/dtm) 2 dt (sect. 1.B. It is not hard to check (Fig.3.4 of chap. Let L consist of all possible extensions of co vectors eEA.50 V.. 17. The manifold A. 2. let us associate with a pencil of geodesics on the obstacle boundary a triad in the symplectic space T IR" = T* IR". Fig. Quadratic tangency of Hand L Let us denote by ta. 16. 2. c H of unit vectors tangent to the geodesics of the pencil is Lagrangian in TF = T* F (the length of the extremal is its generating function).4 of .1 (I) c T !R" be its unit level hypersurface. on F to co vectors '1 on IR" applied at the same point. let H = h.m the germ at 0 of the following triad in the symplectic space IR 2 " with Darboux coordinates L={p=p=O}. H={q!/2+pmqml + . 17) that H is strictly quadratically tangent to L along I. An extremal in the problem of going around an obstacle Returning to the problem of going around an obstacle in Euclidean (more generally.. Arnol'd. Let I= H n L. in a Riemannian) space. Motion along straight lines in IR" is given by the Hamiltonian h = (p. A. +p2q1 +P1 =0}. Givental' Fig. h) is a triad. i.e. p ). 1 or sect. The extremals issuing from the source form a pencil of geodesics on the boundary hypersurface F of the obstacle.I. I=Ln {qm=O}. I. (L.
Integrability According to Liouville. For more details on the problem of the shortest way around an obstacle see sect.m is diffeomorphic to the open swallowtail Lm the variety of polynomials of degree 2m. The curve 1: 2 on the parameter plane of the family of cubic polynomials t 3 + qt + p. Completely Integrable Systems The integrability of a Hamiltonian dynamical system is ensured by a sufficient supply of first integrals.n.A.1. 3). Therefore a natural symplectic structure arises in the space of polynomials.5. One says of functions whose Poisson bracket is equal to zero that they are in involution. The triad example shows that the symplectic version of variational problems can be nontrivial. In article 3. Theorem ([8]). 1) A Hamiltonian system with one degree of freedom (n= 1) is completely integrable.I with a fixed leading coefficient and zero root sum which have a root of multiplicity ~ m. 2) A linear Hamiltonian system is completely integrable. A function F on a symplectic manifold is a first integral of a Hamiltonian system with Hamiltonian H if and only if the Poisson bracket of H with F is equal to zero (see sect. For an investigation of actually integrated systems see the article by B.[7]. Corollary. Definition. § 2. A Hamiltonian system on a symplectic manifold M 2 " is called completely integrable if it has n first integrals in involution which are functionally independent almost everywhere on M 2 ". 2 we showed that each quadratic Hamiltonian on IR 2 " is contained in . 3. 2. and also [6]. Symplectic Geometry 51 chap. chap. .M. Novikov in this volume.m• m"5. The extremals of the functional are polynomials x(t).1. The singular Lagrangian variety of the triad rm. The germ of a generic triad at a point of quadratic tangency of the hypersurface with the Lagrangian man(fold is symplectomorphic to one of the germs '•. The germ of the Lagrangian variety of rays breaking loose from a generic pencil of geodesics on the boundary of a generic obstacle is symplecto morphic to the Cartesian product of a smooth manifold with the open swallowtail. [66]. 2). 3.P. I. [8]. Examples.3 of chap. The tangent at a point of simple inflection of a curve bounding an obstacle in the plane is a cusp point of the Lagrangian curve formed by the tangents to the obstacle boundary. Krichever and S. We discuss the geometric effects and causes of integrability. has the same kind of singularity. Dubrovin. formed by the polynomials with a multiple root. [54]. chap. Example. 5. The variety 1:m is Lagrangian in the natural symplectic structure on the space of polynomials.
Then (under the assumptions of Liouville's theorem) each connected component of M1 is anndimensional torus (see sect. . In fact the· subalgebra may be chosen so that its generators are functionally independent almost everywhere on IR 2 ". H=F 1 ). 1) The first integrals F 1 . M 1 = {xEM I F. n}. On the 2ndimensional symplectic manifold M let there be given n smooth functions in involution Let us consider a level set of the functions F. i... . in the affine coordinates ljJ = (ljJ 1.2.e. In the latter case the phase curves are parallel straightline windings of the torus (Fig.B. 2) M1 can locally be identified with a domain in the cotangent space of the base space IR" at the point!. Liouville's Theorem. H = H (/1 .• .. By the affine structure theorem (sect. Proof Under the premises of Liouville's theorem the mapping F = (F 1 . is constant. are independent (i. ••.(x)=.I. Fn): M+ IR" is a Lagrangian fibration in a neighbourhood of the manifold M1 .52 V. . A winding of the torus . goes over into the co vector d1 H under this identification.. since for the compactness of M1 it is sufficient that the energy level I I I 1/ _j '/ Fig. . Invariant tori are often encountered in mechanical integrable systems. 2). Arnol'd. chap. moreover the field on M 1 defined by the Hamiltonian F* H. 4. • . Let us suppose that on M 1 the n functions F.t. . 0 Remarks. 18).e. 2) Let M1 be compact. Fn) (say. Givental' an ndimensional commutative subalgebra of the Lie algebra of quadratic Hamiltonians. dF 1 A ••• A dF" oFO at each point of M1 ). 4. The Hamiltonian flow on such a torus is either periodic or conditionally periodic. A. i= 1.e. chap. 2) M1 has a canonical affine structure in which the phase fifJw straightens out. 18. . •.2. • • . Then: 1) M1 is a smooth manifold invariant with respect to the phase flow of the Hamiltonian H=H(F 1.• Jn). Fn are independent on M 1 for almost allfE IR" (the case is not excluded that M 1 might be empty here). This follows from Sard's theorem (see [9]). l/Jn) on M 1 one has ljJ = const. i. •.
The Lagrangian fibration itself can by a choice of a Lagrangian section be identified locally with (!R")*/(2n. and the basis cycles y1. In) on the factor IR" and angular coordinates (c/> 1. with the action coordinates (I 1. Theorem.2.dpk A dqk on IR 2 " has the form 'f. The Darboux coordinates on T*IR" turn into the actionangle coordinates after factorization. In sect. 4. for natural systems with a compact configuration space. The "ActionAngle" Variables. D Examples. 1) In the case of one degree of freedom the action is equal to the area divided by 2n of the region bounded by the closed component of the level line of the Hamiltonian. . Let us choose a basis (y 1 . • . This is the case. 19).. ••• .Z") x IR". . c/>n) on the torus IR"/2n. Then in a neighbourhood of M 0 the fibres M1 are n dimensional Lagrangian tori. up to a factor 2n and the addition of constants. In fact. for example. in which the symplectic structure 'f. 2. q)) are called the action variables. the action variables are just these affine coordinates. . ••• .Z".Z" acts on T*IR" by translations in each cotangent space).2. ••• . . The functions Ik=Ik(F(p. we constructed an integral affine structure on the base space of a Lagrangian fibration with fibre a torus: the identification of the tangent spaces to the affine torus M1 with the cotangent space of the base space TjiR" introduces an integral lattice .Zj c: TjiR" there. .. n. Let M = IR 2 " be the standard symplectic space and let the fibre M 1 for f=O be compact and satisfy the conditions of Liouville's theorem. Fig. k= 1. 19. Symplectic Geometry 53 manifolds H = const be compact. 2.dik A dcf>k· Proof. Yn) of one dimensional cycles on the torus M1 depending continuously onf(Fig. Families of cycles on the invariant tori We shall set 1 Ik(f)= 21t J yk(f) pdq. Yn give n differential 1forms on IR"the differentials of the affine coordinates on the base space.Z)" x IR"+ IR" (the group . In a neighbourhood of the torus M 0 one can introduce the structure of a direct product (IR"/2n. chap.
In such a case one says that one has managed to integrate the original system of equations by quadratures. Let E: V+ V * be a linear operator giving a Euclidean structure on the space V.4. for which the quadrics of a given confocal family pass through this point. Let us fix some ellipsoid in Euclidean space.E. A. By a confocal family of quadrics in Euclidean space is meant the family of quadrics dual to the quadrics of some Euclidean pencil. In) takes the form ik=O. x) = 2. A completely integrable system can be integrated by quadratures.. In actionangle variables the system of Hamilton's equations with Hamilton ian H(/ 1. Jacobi's Theorem.I. Until recent times. the only profound means of integration was the method of HamiltonJacobi (see sect. where A. Corollary. chap.e. ••• . not at all reflected in Liouville's theorem. essentially.A. i. 2. and let A: V+ V * be another symmetric operator. But it says nothing about how to find a full set of first integrals in involution. Example. 4. Liouville's theorem covers practically all problems of Hamiltonian mechanics which have been integrated to the present day. Givental' 2) For the linear oscillator H = 'f. After the discovery of infinitedimensional integrable Hamiltonian systems (starting with the Kortewegde Vries equation) many new integration mechanisms came to light. 4).3. Through each point of ndimensional Euclidean space pass n quadrics which are corifocal to the chosen ellipsoid. Arnol'd.. ee V*.(pf + wfqf)/2 the action variables have the form Ik = (pf + wfqf)j2wk (the ratio of the energy of the characteristic oscillation to its frequency). The plane curves confocal to a given ellipse are all the ellipses and hyperbolas with the same foci (Fig. 54 V. apart from differential and algebraic operations on functions only the inversion of diffeomorphisms and the integration of known functions"quadratures"were employed. ~k=aHjaJk and can immediately be integrated: In the construction of the actionangle variables.= A.B. 20). We shall cite below a number of illustrative examples. Elliptical Coordinates and Geodesics on an Ellipsoid. They are all connected with further algebraicgeometric properties of actually integrated systems.x. all of whose axes have unequal lengths. The smooth corifocal quadrics intersect at right angles. the family e. The elliptical coordinates of a point are the values of the parameter A. and the angular coordinates are the phases of the characteristic component oscillations. . By a Euclidean pencil of quadrics is meant the oneparameter family of degree two hypersurfaces (A. (AA" 1 0=2. A*= A.
x) 2 • 0 Chasles's Theorem. Proof The manifold of oriented straight lines in a Euclidean space V has a natural symplectic structure and up to the sign of this structure is symplecto morphic to the cotangent bundle of the unit sphere S (see sect.5). A generic straight line in ndimensional Euclidean space is tangent to n. Proof The visible contours of the quadrics of a confocal family under projection along a straight line form a family of quadrics dual to a family of quadrics in a hyperplane of the dual space passing through zero.I different quadrics of a family of confocal quadrics. Symplectic Geometry 55 Fig. Let F be a smooth hypersurface in V. x) = 2(1. Elliptical coordinates on the plane Proof In terms of the dual space the theorem signifies that the hyperplane <I. drawn at all points of the geodesic. The latter family is simply the section by a hyperplane of the original Euclidean pencil and therefore forms a Euclidean pencil in the hyperplane. and to the same ones for all points of the geodesic.!dimensional space of straight lines parallel to the given one. The mapping p which associates to a point of a geodesic curve on F its tangent line at that point takes the geodesics ofF over into the characteristics of the hypersurface P c T*S of straight lines tangent to F within the space of all straight lines. 1. The property stated follows from the fact that these vectors define the principal axes of the quadric (Ax. Thus. The tangent lines to a geodesic curve of a quadric in ndimensional space. 0 The JacobiChasles Theorem. the visible contours form a confocal family of quadrics in then.2 more quadrics confocal with it. . are tangent. where the radius vectors of the points of tangency are pairwise orthogonal. ton. x) = I is tangent to exactly n quadrics of the Euclidean pencil. Chasles's theorem now follows out of Jacobi's theorem applied to this family. Lemma A. apart from this quadric. moreover the planes tangent to the quadrics at their points of tangency with the straight line are pairwise orthogonal. 20.
G+ P take characteristics over into characteristics. Identifying V with V* with the aid of the Euclidean structure.56 V. G and P are determined only by the symplectic structures of the ambient spaces. Let us define an induced function on the space of straight lines.B. In the commutative diagram of Fig. Involutivity of the induced functions . Therefore the mapping p transfers the geodesics on F into the characteristics of P..I. The mappings G. the geodesics on F are the characteristics on the hypersurface G c T*F of all unit (co)vectors on F. The proof of Lemma A In the Euclidean space V let there now be given a smooth function and let some straight line be quadratically tangent to a level surface at some point. 22).P Fig. then the Poisson bracket of the induced functions is equal to zero at the point which represents the straight line under consideration (Fig. Givental' Indeed. 21. D T*V T*F y~~ G T*S 'J/\J G p . p) = l} of all unit vectors on V. equal to the value of the function at the point of tangency of the straight line with its level surface. q) I <P. and n 1 is the projection along the characteristics of the hypersurface {(p. 22. Amol'd. If two functions on Euclidean space are such that the tangent planes to their level surfaces at the points of tangency with some fixed straight line are orthogonal. Then nearby straight lines are tangent to nearby level surfaces of the function. q) Iq E F} of all vectors applied at points 11'1 1t2 of F. 21 n 2 is the projection along the characteristics of the hypersurface { (p. A. Fig. let us regard G as the submanifold G of codimension 3 in T* V of all unit vectors on V tangent to F. since the characteristics on G. Lemma B.
under movement along the geodesic of the first level surface which is tangent to our straight line. the theorem is proved. h on M. g) are coordinated via the identity V( W(f. 2) The coordinatefree nature of the arguments cited allows one to extend them to the infinitedimensional situation. continues to be tangent to the same level surface of the second function. In the following theorem we assume for simplicity's sake that M is simply connected and the Poisson structures V and W are everywhere non degenerate. 2) we find that two skewsymmetric bivector fields V. Let us construct in the neigh bourhood of the tangency points n.W are also Poisson structures. On the simply connected manifold M two symplectic structures v. w. g)h) + W( V(f. V + JJ. W] = [ V. the tangent line turns in the direction of the normal to this surface and by the same token. By Lemma B the induced functions on the space of straight lines are in involution. Let there be given on a manifold M two Poisson structures V and W (see sect. Insomuch as all the induced functions are constant on this characteristic.} on M such that a)fo is a Hamiltonian of the . Won M form a Poisson pair if and only if [ V. whose Poisson brackets V(f. Poisson Pairs. Remarks. up to secondorder small terms. g. On the manifold M let there be given a vector field v whose flow preserves both Poisson structures of the Poisson pair V. g) and W(f. we proved the JacobiChasles theorem for generic quadrics and straight lines.2. 3. Symplectic Geometry 57 Indeed. Using the Schouten bracket [ . chap. ] (sect. 0 Now let us consider a generic straight line in V. 2. Theorem ([20]).1 functions whose level surfaces are the quadrics of our family.2.field v . 3.4. Then there exists a sequence of smooth functions {J. g)h) +(cyclic permutations) =0 for any smooth functions/. Therefore the derivative of the second induced function along the Hamiltonian flow of the first is equal to zero at the point under consideration. One says that they form a Poisson pair if all of their linear combinations). V] = [ W. 1) Strictly speaking.1 are thereby defined. A geodesic flow on a quadric in Euclidean space is a completely integrable Hamiltonian system. but by continuity the result can easily be extended to degenerate cases.1 .1 quadrics of a confocal family. W.. A characteristic on the level surface of one of the induced functions consists (by Lemma A) of the tangent lines to some geodesic of the corresponding quadric. 2). 0 Corollary. By Chasles's theorem it is tangent to n. W] = 0. systems are for the concrete selfadjoint operators encountered in mathematical physics. chap. It would be interesting to clarify what these. We obtain a large stock of completely integrable systemsthe geodesic flows on the infinitedimensional ellipsoids defined by selfadjoint operators on Hilbert spaces.
Continuing by induction we obtain a sequence of functions {. .fie} satisfying a) and b).u. V + Jl W. where u = qk. W respectively (here it is assumed that the Poisson structures V.f. Let us set v = r. With the notation ok = ojouk. It describes the dy namics of N identical point masses with one degree of freedom each. the field v is Hamiltonian for both symplectic structures.J..fie+ 1 . generates the system of equations of the Toda lattice.. Going over to the system of variables uk = qk.P.B.58 V.5. vk = Ojopk.fie } are in involution with respect to both Poisson brackets.. :of first integrals in involution provides for the complete integrability of the Toda lattice (see the article by B.h = L[p:f3 + Pk(e"k1 + e"k)] etc. v is a Poisson pair. Then V(f. w + ). which proves c). The functions fv and Yw are in involution with respect to the Poisson structure A. At the end we shall obtain either V(J.. Givental' with respect to V. v is obtained from W by the translation Pk 1+ Pk + ). considered as a WHamiltonian..A.. The arising seriesfo. Another method of constructing functions in involution with respect to a Poisson pair consists in the following. Let_t. Let us consider a natural system on ~N with Hamiltonian H='LPN2+'Leqkqk+l. Amol'd.) or W(J.)= W(f. The system with WHamiltonian ft is VHamiltonian with the Hamiltonian ._t. A formal calculation in application of the identity [ V. A. As the flow preserving both = structures of the Poisson pair let us consider the flow of the field v 0.3. . One immediately checks that W is a Poisson structure on ~ 2 N. be a WHamiltonian of it. On the dual space g* there exists a linear Poisson structure (see sect.qk + 1 . S.fic coincides with the field of the WHamiltonian . by elastic bonds with a potential e". We shall apply this lemma in the next item.).e"k._ 1 . The function fo.J.J. Novikov in this volume). Toda). Proof By the condition. Let fo and ft be Hamiltonians of it with respect to V and W respectively. Let r > s. joined in a circle. Indeed. 0 Example.+d etc. Let g be a Lie algebra. W] = 0 shows that the flow of the VHamiltonian field with Hamiltonianft preserves the Poisson bracket of W.. c) the functions {.. we have the following equations of evolution of the Toda lattice: uk = Pk. b) the field of the VHamiltonian. this system is VHamiltonian with the Hamiltonianft = L(PU2+e"k). J.I. W. 2. Krichever.ft . Let fv.+d= V(f..ok1) + e"k Vk + 1 " Vk). The Toda lattice (M. qN+ 1=q 1. Pk = e"k'. Dubrovin. Lemma. Yw be Casimir functions of the Poisson structures V.Pk + 1 . W which form the Poisson pair are degenerateotherwise fv and Yw are necessarily constant)..ak.I).v k 1\ ( ak. let us set w = L(Ok 1\ ok+ I+ PkVk 1\ (ok. like a benzene molecule.qk + 1 is the difference of the coordinates of the coupled neighbours. The total momentum fo = LPk is a Casimir function for V and therefore fo is a V Hamiltonian of the field v.f. In accordance with the theorem.M. 3. Functions in Involution on the Orbits of a Lie Coalgebra.
x)+w([z. Let J.l~o) are in this case Casimir functions for the Poisson structures { . as it changes in time. Let w be an exterior 2form. and { . y.Jl and by continuity. 2): the Poisson bracket of two linear functions x.. One says that a Lax representation of a system of differential equations x = v(x) on a manifold Misgiven (vis a vector field on M). The constant Poisson structure on g* defined by the form w forms a Poisson pair with the linear Poisson structure on g* if and only if w is a 2cocycle on g. for arbitrary ). y on g* is equal to their commutator [x. z)+w([y. into a matrix algebra). ] by means of a translation in g* by). Jl E ~. the coefficients of the characteristic ~polynomial or the eigenvalues of L.. z].ding article. If we apply the lemma of the prect. Then thefunctionsf(e+. where eo E g* is a linear function on g. where Lis the derivative of L along the vector field v and [ .) is obtained from the linear Poisson structure [ .e. If the 2cocycle W is a co boundary. A] means that L. q) be a polynomial Hamiltonian on the standard symplectic space ~ 2 n with the singular point 0. the Casimir functions are the invariants of the coadjoint action. The symplectic leaves of this Poisson structure are the orbits of the coadjoint action of the Lie algebra gong*. Example 1. on each orbit of the coadjoint action..e. A : M + g of the manifold M into a Lie algebra g (for example. y]. Let us consider the following matrices of size (2n + 1) x (2n+ 1) . L.6. Jl E ~. The Lax Representation. The following method of constructing functions in involution on the orbits is called the method of translation of the argument. y] in g.on g. w(x. i. remains on the same orbit of the adjoint action of the Lie algebra g. = [·. }. Let us decompose H as a sum 'LHk of homogeneous components of degree k (k¥1) and let us set G ='I Hk/(k 1). x]. ¥. i. Let H(p. Lemma. then the Poisson structure { . if g is a matrix algebra) are first integrals of the system x = v(x). y) =eo ([X.] + ). 2) the Lax equation L = [L. L.w(. y ]). 0 2. we shall obtain the assertion of the theorem for ). g(e+Jleo) ofthe point e E g* are in involution for any A.. g: g* + ~ be invariants of the coadjoint action of the Lie e algebra g and let 0 Eg*. }. Theorem. if 1) there are given two mappings. respectively.e 0 ). A] holds. where L is an embedding. z E g w([x.eo). Vx. The Lax equation L = [L. The proof is based on the following lemma. eo· The functionsf(e + . Therefore the invariants of the orbit (for example.. y)=O. g(e + J. ] is the commutator in the Lie algebra g. Symplectic Geometry 59 chap.
z) +(a+ p). Usually inte grable systems are connected with nontrivial oneparameter families of Lax representations. Givental' (E is the unit matrix of size n): Then L = [L.1 nand making use of the identity w«. Now let S be the matrix of the quadratic Hamiltonian (Sz. n. )=det(A. lp} = (Q(w« + w_«)z.(wp + w_p)z) =(a. z) = 0. and let n = (~ ..p)wp!lw«. of our linear Hamiltonian system. Then A is a constant matrix.B.!lz)]. quadra tic in z.p).wp =(a. z) we get {1«.W_p)]z.1 ([(w«. .w_«)+(wp. where A. Then the characteristic polynomial of the matrix LA has the form: det(JIE 2 n+ 1 L . since the w11 . A] is the Lax representation for the system of Hamilton's equations with Hamiltonian H (we make use of the Euler formulas GPPP + Gpqq = HP. A. . GqpP + Gqqq = Hq). for the quadratic forms/«= (w«z.. Arnol'd. q).Jln).w_p)]z.SJI!l)[JI+((A. In example 1 let H be a quadratic Hamiltonian. and we obtain the Lax representation LA= [LA. are first integrals. z) and lp = (wpz. Example 2.l. ~) be the matrix of the symplectic form. They are in involution.SJI!l). In this example L 3 = 0. A. setting w11 = Q*().w _ 11 are skewsymmetric matrices.1 ([(w«..1 !lz.S. We may set LA= L +A. A] of the linear Hamiltonian system.I. The coef ficients in this polynomial of the Jl2k. and no first integrals arise at all.W_~)(Wp. k = 0. Indeed. .60 V. is a parameter. z)/2 in the Darboux coordinates z = (p.
. .* § 3. In the general case the function C(a. Then to each element of the Lie algebra H of the group G there corresponds a locally Hamiltonian vector field on M. w.D 2 . skewsymmetric and satisfies the identity C([a.1). on the Lie algebra so. Hamiltonian Systems with Symmetries The procedure described in sect.e. we obtain for the afigular momentum M(w) the Euler equation M = ad!M. b). . Denoting by M the operator of the inertia form. 11 = det (M + I. The inertia quadratic form on the algebra so.). • . ._ = w + I. and p(x) is the density of the body at the point x = (x 1 . w.). Symplectic Geometry 61 Practically all completely integrable systems known at the present day can be integrated with the aid of a suitable Lax representation in which L and A are matrices with coefficients which are polynomial in the parameter }..> so:._ = [M. a]. b]. In matrix form M(w) = Dw + wD. b) is bilinear. c]. Hb} = H 1a.. Definition.. 1. w]. We shall assume in the following that all these vector fields have singlevalued Hamiltonians. The system under consideration is equivalent to the geodesic flow of a particular left invariant Riemannian metric on the group SO.. w) by symplectomorphisms. This representation guarantees the complete integrability of the free rotation of an n dimensional rigid body about an immovable point. b)= 0. An action of a connected Lie group G by symplectomorphisms on a connected symplectic manifold is called a Poisson action if the basis Hamiltonians are chosen so that C(a. i. As we shall see in §3. that is._ = M + I. bE ~l Remark. The· involutivity of the first integrals H . it is a • See Editors· Remarks on page 136. The Poisson bracket {H a• H b} may differ from H 1a. Setting M .. of skewsymmetric n x n matrices has the form tr(wDw). and the Euler equation has the Lax form M= [M. !J D = diag(d 1 .. the investigation of such a system reduces to the study of Hamiltonian flows on the orbits of the coadjoint action on so: with a quadratic Hamilton function. Poisson Actions and Momentum Mappings. bJ by a constant: {Ha. we obtain a Lax representation with parameter for the Euler equation: M._. The free rotation of a multidimensional rigid body. • . x. . bJ + C(a. d. If we choose such Hamiltonians for a basis in g. Example 3. dk = p(x)x~dx. where wE so. c)+ C([b. M: so. a)+ C([ c.1.D. we get a linear mapping g> C"'' (M) which associates to an element a E g its Hamiltonian Ha._]. .5 for reducing the order of a Hamiltonian system invariant with respect to a Hamiltonian flow is generalized below to the case of an arbitrary Lie group of symmetries. Let the Lie group G act on the connected symplectic manifold (M. The metric is given by the inertia quadratic form "in the internal coordinates of the body" (see sect. b)= 0 for all a. 3. 1.D 2 + pE) can be proved using the theorem on translation of the argument from the previous section (see [24]).
The natural action ofthe group G on M is Poisson (with the choice of Hamiltonians indicated below). The corresponding Hamiltonians are the components of the angular momentum vector: M 1 = q2 p3 . A Poisson action defines a momentum mapping P: M + g* whose components are the basis Hamiltonians: to the point x eM corresponds the functional P(x)lae!l = H. {H.q 3 p2 etc. let v. A. Thus a symplectic action determines a cohomology class in H 2 (g. g] + R and gives a homomorphism of the Lie algebra g into the Lie algebra of Hamiltonian functions on M. equal. Let us suppose that the Hamiltonian H on the symplectic manifold M is invariant with respect . Givental' 2cocycle of the Lie algebra g. Amol'd. Theorem. In fact.I. 0 Corollary. V is equal to the value of the covector p on the velocity vector of the oneparameter subgroup of the element a e g at the point x.. Kirillov in the present volume). The value of the Hamiltonian H" on a covector p e T.62 V. is linear in the momenta. 3. Examples. and therefore Lva !X= div. b]). bJ are also linear and.. b)+ p([a. !X+ iva d!X = 0. 1) The group S0 3 of rotations of the Euclidean space R3 is generated by the oneparameter subgroups of the rotations with unit velocity about the q 1 . Hb} and H1.. Since the function H. q2 . Its value at the identity element of the group is just the momentum P(x) of the point x. The corresponding momentum mapping P: T* G + g* coincides with the right translation of covectors to the identity element of the group.... The action form !X on M is Gin variant. b)= 0 is determined uniquely up to the addition of a 1cocycle p: g/[g.2. Let us consider the mapping G + M defined by the action on some stipulated point xeM.. = iv.(x). This equality means that H. b)= C(a. 2) The action of the group by left translations on its cotangent bundle is Poisson. !X is the Hamiltonian of the field v. The preimage of the 1form !X on M under this mapping is a 1form on G.B. The Reduced Phase Spact and Reduced Hamiltonians. consequently. be the vector field on M of the oneparameter subgroup of the element a e G. P(x). leads to the replacement of the cocycle C by the cohomologous C'(a. The Poisson action of the group G on M goes over under the momentum mapping into the coadjoint action of the group G: P(gx) = Ad. A different choice of the constants in the Hamiltonians H. The momentum mapping in this case can be described as follows.. q3 coordinate axes. where pis a linear function on g.A. IR) and is a Poisson action if and only if this class is zero.... In the latter case the basis of Hamiltonians for which C(a. On this consideration is based the classification of homogeneous symplectic manifolds (see the article of A. Let the connected Lie group G act on the connected manifold V and let M = T* V be its cotangent bundle.
The reduced field on the reduced phase space is Hamiltonian with the reduced Hamilton function. c) the elements of the group GPact on M P without fixed points. The components of the momentum mapping are first integrals of such a Hamiltonian system. see [2]).3. The reduced phase space has a natural symplectic structure. x) the preimages of compacta are compact.e. b) GP is a compact Lie group. A generalization . Symplectic Geometry 63 to the Poisson action of the group G on M. 3. The Hamiltonian vector field on M corresponding to the function H is tangent to the fibre M P of the momentum mapping and is invariant with respect to the action of the group GPon M v· Therefore it defines a reduced vector field X Pon F v· Theorem. under the mapping GP x M P + M P x M P: (g. the action of a group on itself by translations is always proper. Let us suppose that the conditions we have formulated are fulfilled. Condition b) can be weakened: it is enough to suppose that the action of GP on M Pis proper. In order for F P to be a smooth manifold certain assumptions are necessary. In the case of the action of a Lie group by left translations on its cotangent bundle the fibre M P of the momentum mapping is the rightinvariant section of the cotangent bundle equal to p at the identity element of the group. Example. x) ~+ (gx. For example. Theorem (MarsdenWeinstein. One can show that the tangent space Tx M P to the fibre of the momentum mapping and the tangent space Tx(Gx) to the orbit of the group G are skeworthogonal complements of each other in the tangent space Tx M and intersect along the isotropic tangent space Tx(Gpx) to the orbit of the stabilizer Gv· From this follows the welldefinedness of the skewscalar product and its nondegeneracy. Let us denote by M P the fibre above the point pEg* of the momentum mapping P: M+ g*. i. so that Mv is a smooth manifold. One speaks of hidden symmetries when a Hamil tonian system possesses a nontrivial Lie algebra of. For example. it is sufficient to suppose that a) p is a regular value of the momentum mapping. The group G P acts on M p· The quotient space FP= M viG P is called the reduced phase space. The skewscalar product of vectors on FP is defined as the skewscalar product of their preimages under the projection M P + F P applied at one point of the fibre of the projection.first integrals not connected a priori with any action of a finitedimensional symmetry group. Hidden Symmetries. The stationary subgroup GP coincides with the stabilizer of the point p in the coadjoint representation. Let GP be the stabilizer of the point p in the coadjoint representation of the group G. The reduced phase space F P is symplectomorphic to the orbit of the point p. 0 An invariant Hamiltonian H defines a reduced Hamiltonfunction HP on FP.
. We have the following obvious Lemma. defines a reduced motion. Theorem ([74]). 0)+(IRn. The momentum mappings of the actions of a Lie group by left and right translations on its cotangent bundle furnish an important example of dual realizations. they form the subalgebra dnd'.I. isomorphic up to the sign of the Poisson bracket). The germs of the transversal Poisson structures to corre sponding symplectic leaves of dual realizations are antiisomorphic (i. Givental' of the momentum mapping in such a situation is the concept of a realization of a Poisson structure [74]a submersion M+ N of a symplectic manifold onto the Poisson manifold under which the Poisson bracket of functions on N goes over into the Poisson bracket of their pullbacks on M.e. The symplectic leaves of the latter are reduced phase spaces on which the Hamiltonian Hed'. Let us consider on M the distribution of skeworthogonal complements to the fibres of the submersion M+ N. There is a correspondence between the symplectic leaves inN and N'. 0) of a Poisson structure at a point of corank r possesses a realization P: (IRn+r. Let us define an equivalence of realizations M 1 + N. considered as a function on N'. A germ (IRn. The construction of the realization P cited in [74] is a nonlinear gen eralization of the momentum mapping T*G+g* of a Lie group G. and a stabilization of the realization M+ N as its composition with the projection onto a factor M x IR 2k+M of the product of symplectic manifolds. All such Hamiltonians form a Lie algebra d' connected with a realization M+ N' of a different Poisson structure. 0). In the general case there is a close connection between the Poisson manifolds N and N' of dual realizations. A submersion M+ N is a realization if and only if the inverse images of the symplectic leaves are coisotropic.B. Any realization of it is equivalent to a stabilization of P. Let the Hamiltonian H: M+ IR commute with the Lie algebra d of functions lifted from N under the realization. Therefore this distribution is integrable and is tangent to the coisotropic inverse images of the symplectic leaves. The projection M+ N' along its integral manifolds (which is defined at least locally) is (by the lemma) a realization of a Poisson structure arising on N'. Theorem ([74]). A. they share common Casimir functionsconsidered as functions on M.64 V.lt is generated by the fields of the Hamiltonians in the Lie algebra d. For example. Arnol'd. This realization is called dual to the original one and can be constructed as follows. M 2 + N as a sym plectomorphism M 1 + M 2 commuting with them. bijective if the inverse images of these leaves in M are connected: leaves with intersecting inverse images correspond to one another.
An example is the additive group of a Lie coalgebra. then the linear Poisson structure on its dual space is just the sought for Poisson structure on the orbit space G/H =(H. The same is true for the invariants of a connected subgroup H c G if the orthogonal complement l).4. Poisson groups have come to occupy an important place in the theory of completely integrable systems. Poisson Groups. Drinfel'd to generalize the procedure of reduction of Hamiltonian systems to a broader class of actions. On the Lie algebra g of a Poisson group G there is defined a linear Poisson structurethe linearization of the Poisson structure on G at the identity. An action of a Poisson group G on a Poisson manifold M is called Poisson if G x M+ M is a Poisson mapping. the smooth mappings which transform the Poisson bracket of two functions into the Poisson bracket of their pullbacks. Symplectic Geometry 65 3.L)*. Therefore a Lie algebra structure is defined on g* (the double structure arising here is a Lie bialgebra structure in the sense of [21 ]). that is. the example analyzed above is closely connected with the method of translation of the argument (sect. By a Poisson group one means a Poisson manifold endowed with a Lie group structure for which multiplication G x G+G is a Poisson mapping and inversion G+G is an antiautomorphism (changes the sign of the Poisson bracket). For such a subgroup there is on the manifold M/H (if it exists) a unique Poisson structure under which the projection M+ M I H is a Poisson mapping. This .G. Let us note that in the construction described the condition imposed on H does not mean that His a Poisson subgroupthe latter is true ifl). As an example let us consider the action of a connected subgroup H of the additive group of a Lie coalgebra G by translations on G. If H. The converse of this assertion is falsethe algebra of invariants' being closed under the Poisson bracket does not imply symplecticity of the action. Let us consider the category whose objects are the Poisson manifolds and whose morphisms are the Poisson mappings. In the case of an action of a group G with the trivial Poisson structure on a symplectic manifold M this condition is equivalent to the action's being symplectic (but not to its being Poisson in the old sense!).L c G* is a Lie subalgebra. A product M x N of Poisson manifolds is endowed with a Poisson structure for which the projections onto the factors are Poisson mappings and the pullbacks of functions from different factors are in involution.L c g* of its Lie algebra l) c g is a Lie subalgebra in g*. The Poisson bracket of two functions on a symplectic manifold which are invariant under a symplectic action of a Lie group is again an invariant function. In the example cited above it coincides with the original Lie algebra structure.L is an ideal in g*. This circumstance led V. On the symplectic leaves in M/H an Hinvariant Hamiltonian defines a reduced motion. Thus. 2. It is not hard to verify that the invariants of a Poisson action of a Poisson group on a Poisson manifold form a Lie subalgebra in the Lie algebra of functions on it.5).
A leftinvariant geodesic flow on the group defines a reduced Hamiltonian flow on each orbit of the coadjoint represen tationa reduced phase space. Let us note that the "inertia operator" v ~+rot v maps the space g bijectively onto the space of smooth divergencefree fields on D under certain restrictions on the domain D (it is sufficient that D be a contractible bounded domain with a smooth boundary). by a positive definite quadratic form Q on the space g*.A. Details can be found in the papers of M. i. incompressible. In these cases the group G is infinitedimensional. On the connected Lie group G let there be given a leftinvariant Riemannian metric. (Let us recall that the action of the group G P on M P is assumed to be free). where [.e. Theorem ([2]). The flow of an ideal fluid is a geodesic of this metric. . The Euler equation can be written down in the form orotvjot=[v. for example in [63].6. The equations of the hydro dynamics of an ideal fluid [2] and the system of MaxwellVlasov equations describing the dynamics of a plasma ( [74]) have the form of the Euler equation. The reduced Hamiltonian of this flow coincides with the restriction to the orbit of the quadratic form Q. Amol'd. In vector notation it has the form P= P x n. 3. Let us consider in greater detail the flow of an ideal (i. A phase curve of a system with a Ginvariant Hamiltonian function is a relative equilibrium if and only if it is the orbit of a oneparameter subgroup of the group Gin the original phase space. Let G be the group of volume preserving diffeomorphisms of the domain D. Geodesics of LeftInvariant Metrics and the Euler Equation. Let n: g*+g be the operator of the quadratic form Q. It is determined by its value at the identity of the group. in viscid) fluid in a domain D c ~ 3 . 3. where n is the angular velocity vector and P is the angular momentum vector. Givental' direction is developing rapidly. A.I. homogeneous. connected with the vector n by a linear transformationthe inertia operator of the body. The Lie algebra g consists of the smooth divergencefree vector fields on D which are tangent to the boundary of the domain D. rotv].]is the commutator of vector fields. For example.5. Then the reduced motion of a point Peg* is described by the Euler equation P= adl'i(P)P. the stationary rotations of a rigid body fixed at its centre of mass. 66 V. SemenovTyanShanskij. In the special case G = S0 3 we obtain the classical Euler equation describing the free rotation of a rigid body in the internal coordinates of the body. but also the rotations of a heavy rigid body with constant velocity about the vertical axis are relative equilibria. The phase curves of a system with a Ginvariant Hamiltonian function which project into an equilibrium position of the reduced Hamiltonian function on a reduced phase space are called relative equilibria.B. Relative Equilibria. The kinetic energy Jv 2 f2dx of the flow with velocity field v is a rightinvariant Riemannian metric on the group G.e.
A reduced phase space F P of the Poisson action of G on T* V is symplectomorphic to a twisted cotangent bundle of the factored configuration manifold V/G. The reduced configuration space in this case is the two dimensional sphere S 2 = S03/S 1 . 1. Symplectic Geometry 67 Now let G =~Ill be the circle. 3. has at least two stationary rotations (for each value of the angular momentum with respect to the symmetry axis). Both corollaries are based on the fact that a function on the spherethe potential of the reduced motionhas at least two critical points. 2). i. Remarks. Let an asymmetrical rigid body. chap. •• See Editors' Remarks on page 137.2). Let us suppose that the group G acts on the configuration manifold V without fixed points. An axissymmetrical rigid body fixed at a point on its symmetry axis has at least two stationary rotations (for each value of the angular momentum with respect to the symmetry axis) in an arbitrary potential force field. In this affine structure the phase flow of the invariant Hamiltonian H becomes straight. . If the dimension of the phase manifold is equal to the sum of the dimension of the algebra g and its rank.e. 1) We recall that the rank of a Lie algebra is the codimension of a generic orbit in the coadjoint representation. attached at a point on the axis of the field. then the level set M P of a generic regular level of the momentum mapping is nonsingular and has a canonical affine structure. fixed at a point. Proof The premise dim M =dim G + rkG together with the regularity of the generic value p of the momentum mapping implies that dim M P = rk G =dim GP' i. 3. Noncommutative Integrability of Hamiltonian Systems:• Let us suppose that the Hamiltonian H of a system is invariant with respect to the Poisson action of a Lie group G on the phase manifold M. the algebra gP (for a generic pEg*) is commutative. each connected component K of the level set M P is a quotient space of (a connected component of) the group GP by a discrete subgroup. By Duflo's theorem (sect. be subject to t~e action of the force of gravity or of another potential force which is symmetric with respect to the vertical axis. which is equal to zero only when p=O.3. and that pEg* is a regular value of the momentum mapping P: M+ g*. 2) The theorem just formulated generalizes Liouville's theorem on complete integrability: there the group G was a commutative group (~n) of rank n which acted on a symplectic manifold of dimension 2n =dim G + rk G.e. An asymmetrical rigid body in an axissymmetric potential field.7. Each compact connected component of the set M P is a torus on which the phase flow is conditionally periodic. Theorem. Corollary 2. Corollary 1. The reduction of a natural Hamiltonian system on T* V with a Gin variant potential and kinetic energy leads to a natural system in a magnetic field (see sect.
A. such systems are degenerate in comparison with general completely integrable systems. If this action is actually an action of the . It is evident from the formulation of the theorem (and from the example) that motion in a system which is integrable in the noncommutative sense takes place on tori of dimension less than onehalf the dimension of the phase space. its correctness for all Lie algebras would allow one to prove the analogous theorem for arbitrary and not only for compact phase manifolds. where r is the distance to the centre and p is momentum.rk g)/2 smooth functions in involution which are independent almost everywhere on generic orbits in g* (their dimension is equal to 2d). Therefore the Kepler problem is integrable in the noncommutative sense. that is. where dim g + rk g = 2n.ljr. If a Hamiltonian system on the compact symplectic manifold M 2 " possesses a Lie algebra g of almost everywhere independent first integrals. The same relates to an arbitrary natural system on the Euclidean space IR 3 with a spherically symmetric potential: the phase flow of such a system straightens out on the two dimensional combined level sets of the angular momentum vector and the energy. D Example.8. Then the Hamiltonian H is invariant with respect to the group of rotations so3 and its flow together with the action of the group S0 3 makes up a Poisson action of the fourdimensional group G = IR X so3 of rank 2 on the space T*IR 3of dimension 6=4+2. including the semisimple ones (see [25]). For k=n the geometry of such an action may be looked at like the geometry of completely integrable systems. A Hamiltonian H: M 2 ~1R on a compact symplectic surface gives a Poisson action of the additive group of IR. 2. The compact orbits of this action inevitably are tori. This statement has been proved (on the basis of the method of translation of the argument. Givental' K = IR"/lLk and in the compact case is a torus. then there exists another set of n almost everywhere independent integrals in involution. For a Lie algebra g of first integrals on an arbitrary symplectic manifold the assertion of the theorem follows from the statement: on the space g* there exist d =(dim g. The straightening of the flow is easily deduced from the invariance of the Hamiltonian H. A set of k functions in involution on a symplectic manifold gives a Poisson action of the commutative group IRk. Let us consider the Kepler problem of the motion of a mass point in the Newtonian gravitational potential of a fixed centre: H = p2 /2. Theorem ([26]). Amol'd.I. Example. although of a fairly special class. sect.B.5) for a broad class of Lie algebras. Geometrically this means that the invariant tori of the small dimension rk g can be united into tori of the half dimension. Poisson Actions of Tori. 3.68 V. Here we shall consider the case of a Poisson action of the torus Tk = IRk jlLk on a compact symplectic manifold M 2".
If this property of the function H is realized. Under the conditions of the theorem the fibres of the momentum mapping are connected. 24). and in particular. Remark. and the image of the whole manifold coincides with the convex hull of the set of vertices. What is more. Indeed. 23). and we may reckon that on the latter . The classical origin of this theorem are the Shur inequalities for Hermitian matrices: the vector of diagonal entries of a Hermitian matrix lies in the convex hull of the vectors obtained out of the set of its eigenvalues by permutations (see Fig. M 2 is a sphere (Fig. 23. has dimension rand lies in a subspace of dimension r parallel to the (integral) subspace of covectors in (~k)* which annihilate the tangent vectors to the stabilizer Tkr in the Lie algebra ~k of the torus Tk. The closure of each connected component of the union of the orbits of dimension r ~ k is a symplectic submanifold of M 2 " of codimension ~ 2(k. Then the image of the momentum mapping P: M 2 "+(~k)* is a convex polyhedron. Hamiltonian flows on surfaces Theorem ([ 10]). Example. on which the quotient group T' = Tk /Tk r of the torus Tk by the isotropy subgroup Tkr acts in a Poisson manner. the image of the set offixed points of the action of the group Tk on M 2 " consists of a finite number of points in (~k)* (called the vertices). Let there be given a Poisson action of the torus Tk = ~k/Zk on the compact connected symplectic manifold M 2 ". then its product with a suitable nonvanishing function is the Hamiltonian of a Poisson action of the group T 1 on the sphere. The image of this submanifold in (~k)* under the momentum mapping (a face of the polyhedron) is the convex hull of the image of its fixed points.r). let us consider the coadjoint action of the group sun+ I of special unitary matrices. The convexity of the image can be deduced from this by induction on the dimension of the torus. Symplectic Geometry 69 group T 1 = ~/Z. The space of such matrices can by multiplication with j=t be identified with the space of Hermitian (n + 1) x (n + 1) matrices with trace zero. then the function H necessarily has as its critical points only a nondegenerate maximum and minimum. It is isomorphic to the adjoint action on the Lie algebra of skewHermitian matrices with trace zero. Fig.
. of the quadratic part of the Hamiltonian Hat a critical point meH. 24.k<n. Newton polyhedra.B.. As another corollary of the integration formula we find an expression for the volume of the manifold M via the characteristics of the fixedpoint set of the J L action: wn = (. Xn) I:Lxt = 0}. . Then where the sum is taken over all critical values.70 V. In the simplest case of a Poisson action of the circle T 1 on a symplectic manifold (M 2 n.1(p). w) it has the following form. .I. The Shur inequalities space there is given an action of the group sun+ 1 whose orbits are compact symplectic manifolds. p Analogous results are true also for the actions of tori of greater dimension.1 on every interval of the set of regular values of the Hamiltonian H. . A. . [38]).. . and with the computation of characters of irreducible representations of Lie groups (see (II].•• .1.1tn! pnI E(p) and a series of relations on the critical values of M p the function H: LPk/E(p)=O for O::s. The fixed points of the action of the torus on the orbit are the diagonal matrices diag (. with characteristic classes. By means of the Fourier transformation one obtains from this that the function f(h)= J wn/dH (the H=h volume of the fibre over hEIR*) is a polynomial of degree :::. equal to the product of the nonzero eigenvalues. 0 . . ei"'")ILtPk = 0} of SUn+ 1 acts in a Poisson manner on each such orbit. wnn) in the space !Rn = {(x 0 . The momentum mapping associates to the Hermitian matrix (wkl) the vector of its diagonal entries (w00. . The maximal torus Tn = {diag(ei4>o.. the method of stationary phase. Another characteristic property of Poisson actions of tori is the integration formula [22]. . Let H: M 2 n+IR* be the Hamiltonian of the action. each divided by 2n. .. n. Givental' 'U+p=O Fig. An) of this orbit. . With each of its critical values p E IR* let us connect an integer E(p). The subject of this article has turned out to be connected with the theory of residues. . Amol'd.. toroidal embeddings. equivariant co homology.
in which the tangent space to the integral submanifold passing through x is isotropic. p 1 . . dz='[. is called a Legendre fibration. The meaning of the definition of a contact structure becomes clearer if one considers the problem of the existence of integral manifolds of the field of hyperplanes.pkdqk is a contact structure on 1Rl 2 "+ 1 . Integral submanifolds of dimension n in a 2n + !dimensional contact mani fold are called Legendre submanifolds. q. all of whose fibres are Legendre. q) the contact Darboux coordinates. gives a symplectic structure on the hyperplane of the field at the point x. Contact Manifolds l.l. Therefore a contact structure may be called a "maximally nonintegrable" field of hyperplanes. •. i. If an integral hypersurface passes through the point x. In fact. In the neighbourhood of each point ofa contact 2n +!dimensional manifold there exist coordinates (z.• . Pn) in which the contact structure has the form dz = LPk dqk. A field of hyperplanes on a 2n +!dimensional manifold is called a contact structure if the form a 1\ (drt.)" is nondegenerate. A smooth fibration of a contact manifold.. We shall call this structure the standard one. q 1 . § 1. This requirement's independence of the choice of the defining ! form a can instantly be verified. p. The connection between them is similar to the relation of projective and affine geometry. then :x A da lx = 0. Contact manifolds of the same dimension are locally contactomorphic. • . which does not vanish: :xlx = 0 is the equation of the hyperplane of the field at the point x. Corollary. One says that a field of hyperplanes is given on a smooth manifold if in the tangent space to every point a hyperplane is given which depends smoothly on the point of application. In fact the dimension of integral manifolds of a contact structure on a 2n + !dimensional manifold does not exceed n. Diffeomorphisms of contact manifolds which preserve the contact structure we shall call contactomorphisms. .e. For the proof let us observe that the form dxrt. . Contact Structure. A field of hyperplanes is defined locally by a differential 1form rt. submanifolds which at each of their points are tangent to the hyperplane of the field. Darboux's Theorem for Contact Manifolds. Symplectic Geometry 71 Chapter 4 Contact Geometry Contact geometry is the odddimensional twin of symplectic geometry. and the coordinates (z.
e. A. Indeed. By a contact element on a manifold M. + Pndqq =0. dx f). an isomorphism of P( V) and P( V*) can be lifted to an isomorphism of V and V*. i. Let us consider the set L(X) of contact elements on M which are applied at points of X and are tangent to X. P(V) is provided with a contact structure in the following way: the hyperplane of the contact structure at a point I e P( V) is given by the hyperplane P(H) c P( V) passing through l. .. Thus. In the neighbourhood of each point of the total space of a Legendre fibration there exist contact Darboux coordinates (z. Amol'd. p) in which the fibration is given by the projection (z. there exists a local contactomorphism of the total spaces of the fibrations which takes fibres into fibres. Legendre fibrations of contact manifolds of the same dimension are locally isomorphic. Conversely. the fibres (z.. p) 1+ (z. the incidence condition which was formulated is equivalent to the skew symmetry of this form. to a nondegenerate bilinear form on V. qn. . . Examples. qn. . Let X be a smooth submanifold of M. Let V be a 2n + 2dimensional symplec tic vector space. i. k ~ 1.e. p0 . .B. B. In Darboux coordinates (q 0 . All the contact elements on M form the total space PT* M of the projectivized cotangent bundle. where H is the skeworthogonal complement to the straight line l c V. The manifold of contact elements. The Legendre subspaces of P( V) are the pro jectivizations of the Lagrangian subspaces of V. is meant a hyperplane in the tangent space at that point. The 1jet of the smooth function fat the point x (notation j! f) is just (x. every isomorphism of IP 2 "+ 1 and IP* 2 "+ 1 with this incidence property is given by a symplectic structure on the underlying vector space and consequently defines a contact structure on IP 2 " + 1 . Indeed. • . . The following rule defines a contact structure on PT* M: the velocity vector of a motion of a contact element belongs to the hyperplane of the contact field if the velocity vector of the point of application of the contact element belongs to the contact element itself (Fig. L(X) is a Legendre submanifold of PT* M. Pn) on V and in the affine chart q0 = 1 on P( V) this structure has the form dp 0 = LPkdqk. .2. A. Corollary. .I. from which follows its maximal nonintegrability. . 1. q) = const are Legendre subspaces of the standard contact space. the bundle PT* M+ M is Legendrian. Projective space. f(x). . under which each point lies in the hyperplane corresponding to it. applied at a given point. In the special case when X is a point. Pn) on T* M and in the affine chart Po= 1 on PT* M this structure is given by the vanishing of the action form: dq 0 + p 1 dq 1 + . C. q. p0 .qkdpk. ••• .72 V. The contact structure introduced on P( V) gives an isomorphism between P( V) and the dual projective space P( V*) of hyperplanes in P( V). The space J 1 M = IR x T* M of 1jets . q ). q. The space of 1jets of functions. Givental' Darboux's Theorem for Legendre Fibrations. In Darboux coordinates (q 0 . 25). P(V) its projectivization. L(X) is the projective space of all contact elements on M which are applied at that point.
of a given body in any state. e and 17 are determined when the S$ate of the body is given.W. the entropy. so that between the five quantities v.. where u is the coordinate on the axis IR of values of the functions and IX= LPkdqk is the action 1 form on T* M. D. we have de= td17.. The 1graph of the function! (notationf f) consists of the 1jets of the function at all points. the (absolute) temperature.. p. and it may be permitted to call themfunctions of the state of the body. The definition of the contact structure on PT*M of functions on the manifold M has the contact structure du =IX. IR x M along the fibres of the cotangent bundle of M is a Legendre fibration.pdv. . e. t. the volume. j 1f is a Legendre submanifold of J 1 M. the energy. the heat received. 25.v. t. p. e and '1 there exist relations expressible by three finite equations. also W. different in general for different substances.d W. in the sense in which the term is used in the thermodynamics of fluids. d W = pdv. These are subject to the relations expressed by the following differential equations:. p. t. by the body in passing from one state to another. but always such as to be in harmony with the differential equation (1).. The projection J 1 M __. de= dH. Let us quote the beginning of the article of J. The phase space of thermodynamics. (1) The quantities v. Gibbs "Graphical Methods in the Thermodynamics of Fluids" [31]: "We have to consider the following quantities:. The state of a body." In our terminology the states of a body form a Legendre surface in the five dimensional phase space of thermodynamics equipped with the contact structure (1). Symplectic Geometry 73 PT*/'1 Fig. is capable of two independent variations. One can analogously define the contact structure and the Legendre fibration of the space of 1jets of sections of a onedimensional vector bundle over M (not necessarily the trivial one) over the total space of this bundle. Eliminating d Wand dH. and H. 17. the pressure. dH = td17. the work done.
The Relative Darboux Theorem for Contact Structures.. Let oc 0 and oc 1 be two contact forms on the manifold M which are transversal to the submanifold N. The Geometry of the Submanifolds of a Contact Space. A su bmanifold of a contact manifold carries an induced structure. Theorem A follows from theorem B..q = 0 and u = p. and lie in one connected component of the set of contact forms with these properties.B.3. . Then there exists a diffeomorphism of neighbourhoods of the submanifold N in M which is the identity on N and takes oc 1 over into oc 0 • Corollary..• . We shall call the contact form oc transversal to a submanifold if the field of kernels of the form doc is nowhere tangent to it. . U 1 of neighbourhoods of the point x in M which is the identity on N n U 0 and takes Yoluo over into y1 lu. • . Arnol'd. The thus defined induced structure is finer than simply the field of tangent subs paces cut out on the submanifold by the hyperplanes of the contact structure. •. A contact form oc defines a field of directionsthe field of kernels ofthe 2form doc. Thus to the form du.1. 2) In the neighbourhood of a generic point of an odddimensional sub manifold of a contact space the induced structure is a contact structure.LPkdqk corresponds the direction field ojou. xk. coincide on TN. +xkdyk=O [51]. 1. the contact structure du = pdq induces non diffeomorphic structures on the curves u = p.q 2 = 0 in a neighbourhood of the point 0. A differential 1form on a manifold which gives a contact structure on it we shall call a contact form.· In the special case N ={point} we get the Darboux theorem for contact manifolds of sect. Locally this structure is given by the restriction of the defining 1form to its tangent bundle. + xkdyk = 0. For example. The nongeneric points form a set of codimension ~ 2... . Lemma.LPkdqk.I. Examples. Givental' 1. Yk in which the induced structure has the form dy 1 + x 2 dy 2 + . A contact form reduces locally to the form du.74 V. Let us pass on to the proof of theorems A and B. Let N be a smooth submanifold of the manifold M and let Yo and y1 be two contact structures which coincide on TN.. A. B. The Relative Darboux Theorem for Contact Forms. The induced structure defines the "exterior" geometry of the submanifold at least locally: A. but in the neighbourhood of the points of some smooth hypersurface it reduces to one of the two (not equivalent) normal forms ± du 2 + (1 + xddy 1 + x 2 dJ2 + . Then for an arbitrary point x in N there exists a diffeomorphism U 0 _.. Forms obtained from one another by multiplication with a nonvanishing function give the same induced structure. 1) In the neighbourhood of a generic point of a generic even dimensional submanifold of a contact space there exists a coordinate system x 1 . y 1 .
au and is determined from the equation iwda + p = 0. equal to zero on N. where tX 1 is a smooth family of contact forms which are transversal to N and coincide on TN. moreover in such a way that nA IN= niN. 1]. and consequently also in some neighbourhood of it. Using the relative Poincare lemma out of sect.t)a 1 . This equation we want to solve with respect to the family of vector fields Jl'r.5. by a linear transformation A: M+ M we may take a 0 lx over into a 1 lx and kerxda 0 over into kerxda 1 .andyoand y1 are given by the contact forms a 0 and a 1 respectively. chap. where the field w does not depend on a. If a is transversal to N. Let us set eE u ff(u. x)de. we arrive at the equation Lv I tXr + oat! ot = 0. x)~+x by the integral curves of the field of = directions ker da. Symplectic Geometry 75 WemayassumethatM = JR2n+ 1 . then we may consider that a neighbourhood of the manifold N in M is the trivial fibration 11\1! x P+ P: (u. t E [0. We shall allow the reader to look after the smoothness in t of the subsequent contructions. N c {0} X P. Following the homotopic method (see sect. 11\1!. where n: M+ kerxa 1 is the projection along kerxda 1 . A contact form a gives a trivialization ofthe fibration ker da. where the coordinate u on the fibres is chosen so that ia1autX 1. Now ta 0 + (1. PITNM=O andfiN=O. D . 2. chap. After this it will be possible to set v = w. We want to represent the 1form oa/ot. They can be identified by means of a linear transformation which is the identity on kerxda 1 and on n(N) (compare §1. D Proof of Theorem B. where P' does not depend on du and P'ITNM =0. the resulting transformation possesses the required properties. The forms dxao and dxa 1 give two symplectic structures on kerxa 1 which coincide on n(N).(f + iva) a1au. There exists a linear transformation of the space M which is the identity on N and which takes a 0 lx over into a 1 lx and dxao over into dxa 1 • Indeed. Since N c kerxda 1 EB n(N) is the graph of the function a 1 IN. The lemma is proved. 2). 0 Then ffiN=O and oajot=P'+dff. X)= J[ia{au(iJajiJt)J(e. in the form of a sum P+df. where P and f = ff + ¢ satisfy the require ments stated above. Theorem B is proved. is a family of contact forms which coincide on TN and are transversal with respect toN at the point x. equal to zero on TN. 1.NisavectorsubspaceinM. 1. By multiplication of a 1 with an invertible function we may achieve the coincidence of a 0 and a 1 on TN and by multiplication ofthe forms a 0 and a 1 with the same invertible function we may make them transversal with respect to N at 0. where P does not depend on du. chap.x = O.3. we may represent P' in the form P' = P + d¢. 1).
+xmdYm (n= 2m+ 1) or (1 ± xf}dy 1 +x 2 dy 2 + .. +xmdYm (n=2m+ 1) or (1 +xddy 1 + x 2 dy 2 + . Martinet) ± du 2 +(1 +xddy 1 +x 2 dy 2 + .B. or its equivalence class is not determined by any finiteorder jet (i.76 V.4.. considered as a function of the parameter. On the integral curves y = const of this field let us consider the family of functions F(. Theorem ([78]). +xmdYm (n=2m). y) will be equal to 1. by its 2jet. In the neighbourhood of a generic point a generic differential 1form reduces by means of a diffeomorphism to the Darboux normal form du+x 1 dy 1 + .e. a finite number of coefficients of the Taylor series does not determine the equivalence class. 26. Then the difference of the critical values. Does there exist a symplecto morphism of a neighbourhood of a singular point of the field which takes v over into its linear part at that point? The connection of this problem with the original Fig. compare sect. For the investigation of !forms in the neighbourhood of singular points see [50]. y). 1.3). let v be a vector field such that Lvw = w. a differential1jorm either equivalent to one of the Darboux and Martinet normal is forms. y)dy and gives the field of directions dy = 0. On a symplectic space with the structure w...I. a generic differential 1form on the plane reduces to the form F(x. by a . Arnol'd.. If at the point under consideration two critical points of the functions of the family merge (Fig.. The equivalence class of the Darboux form is determined by its 1jet. + xmdYm (n =2m) ([51]. A. A functional modulus of a 1form on IR 2 . Remark. In particular.finite section of the Taylor series at the point under consideration). In the neighbourhood ofa point where it does not turn to zero.. we may choose the parameter y in such a way that the sum of the critical values of the functions F(·. and of the Martinet form. 26 ).. In the neighbourhood of a point where it does not turn to zero. but in the neighbourhood of a point on some smooth hypersurface it reduces to the Martinet normal form (J. Example. Degeneracies of DifferentiallForms on R". will be a functional invariant of the equivalence class of our 1form. Givental' 1. It leads to the following problem.
We note that both the vector field v and its linear part are a sum of the Euler field E=(1/2)Lxkofoxk and a Hamiltonian field with a singular point at the coordinate origin. The Hamiltonian Q can be taken in the normal form H 0 of sect. to be symplectically C 00 equivalent to V. where Q is a nondegenerate quadratic Hamiltonian. since the indicated contact structure is Sp(2n.qkdpd.mkEZ. in which the contact structure has the form dt = L(Pkdqk. Corollary. L·mk~3. Let M be a contact manifold. and having the property Lvw = w. in a neighbourhood of a point of tangency with the hyperplane of the contact field. it is necessary and sufficient that among the eigenvalues A. § 2. Symplectification. 1 . Therefore the spectrum of the field V is symmetric with respect to A. 27). Theorem ([50]). A generic hypersurface in a contact space. Let us consider the total space L of the onedimensional bundle L+ M whose fibre over a point x EM is formed by all nonzero linear functions on the tangent space TxM which vanish on the hyperplane of the contact field at the point x. to the normal form t = Q(p. . 2. Let M = PT *B be the projectivized cotangent bundle with the canonical contact structure. Example. In Darboux coordinates the 1form iEw has the form L (pkdqk. 0:$. Then L = T *B\B is the cotangent bundle with the zero section removed and IX is the action 1form on T *B. For the field v defined by the condition ivw =IX we get Lvw = divw = diX = w. IR)invariant. Symplectification and Contact Hamiltonians Symplectification associates to a contact manifold a symplectic manifold of dimension one greater. Remark. . We shall call such functions contact linear functions.qkdpk)/2. Symplectic Geometry 77 one is as follows. Giving L as a subbundle of the cotangent bundle T* M is equivalent to the introduction of a contact structure on M. 2 " of the field V there be no relations of the form LmkA. A. by means of a suitable choice of coordinates. • .= 1/2. w). chap. We shall bring a description of the Lie algebra of infinitesimal contactomorphisms based on the properties of this operation.1. q). The dual operation of contactification will be discussed. For an arbitrary vector field v on the symplectic space (IR 2 ".1. reduces. l. A generic differential lform IX on an evendimensional space gives a symplectic structure in a neighbourhood of a singular point. 3. On the manifold La diffentiallform a is canonically defined: the value of a on a tangent vector v applied at the point ~ E L is equal to the value of the contact linear function ~ on the image of the vector v under the projection L + M (Fig.k= 1. with a given linear part Vat a singular point. .
2. The inclusion L c. T *M and the projection L+ M establish a onetoone correspondence between the con tactomorphisms of the manifold M and the symplectomorphisms of the manifold L which commute with the action of the group IR x.B. The fibres of a Lagrangian fibration carry a canonical affine structure (see sect.. Amol'd. The symplectic manifold (L. Its nondegeneracy follows from the example cited. C. Using an IR xinvariant version of Darboux's theorem for Lagrangian fibrations. chap. Givental' o. The composition of the projection L+ M and a Legendre fibration M+ B defines an 1Rxinvariant Lagrangian fibration L+B. Together with the IR xaction on the space of the symplectifi cation this allows one to introduce a canonical projective structure on the fibres .=O v Fig. A. Conversely. 4. 27. da) is called the symplectification of the contact manifold M. The Properties of the Symplectification. Symplectification In the general case the !form a on the manifold L defines the symplectic structure da. a principal!R xfibration N+ M of a symplectic manifold with a homogeneous degree1 symplectic structure gives a contact structure on M for which N is the symplectification. This action turns L+ M into a principal bundle.I. D. The projection of the symplectification L+M gives a onetoone corre spondence between the IR xinvariant ("conical") Lagrangian submanifolds of L and the Legendre submanifolds of M. it is easy to deduce from this Darboux's theorem for Legendre fibrations. in view of the local uniqueness of the contact structure. 2). The symplectic structure da is homogeneous of degree 1 with respect to this action. Definition. A. and vice versa. The multiplicative group IR x of nonzero scalars acts on L by multiplication of the contact functionals with constants.78 V. B.
pdq (we drop the summation sign). Now let us suppose that the contact structure on the manifold is given by a globally defined differential 1form a. 2. This projective structure may be described explicitly thus. to the sphericalized cotangent bundle. The symplectification of contact vector fields gives an isomorphism of the Lie algebra of such fields with the Lie algebra of locally Hamiltonian IR x invariant vector fields on the symplectification of the contact manifold. Corollary. Theorem. Let IX= du. The contact Hamiltonian of a contact vector field on M is the function on M which at a point x is equal to the value of the homogeneous Hamiltonian of the symplectification of this vector field on the contact linear form aln considered as a point of the fibre above x in the bundle L + M. The contact form a defines a section of the bundle L+ M of contact functionals. if the dimension of the fibre is greater than one (in the case of a onedimensional fibre there is a countable number of different coverings). The Lie Algebra of Infinitesimal Contactomorphisms. As soon as a section has been chosen.e. it gives a onetoone correspondence between the homogeneous degree! Hamiltonians on L and the functions on M. Thus the existence of the form a is equivalent to the triviality of this bundle. We shall define in the following way the symplectification of a contact vector field: it is the vector field on the symplectification of the contact manifold whose flow is the symplectification of the flow of the original contact field. A Legendre fibration with a compact fibre is canonically con tactomorphic to the projectivized cotangent bundle of the base space or to its fibrewise covering. The Hamiltonian of such afield can be made homogeneous of degree 1 by the addition of a locally constant function. i. Such fields obviously form a Lie subalgebra of the Lie algebra of all vector fields on the contact manifold. The contact Hamiltonian K of a contact field V is equal to the value of the form a on this field: K =iva. We obtain a local contactomorphism M+ PT *B mapping the Legendre fibres into fibres. Let us cite the coordinate formulas for the contact field VK of a function K. applied at the point n(x). The hyperplane of the contact field at a point x EM contains the tangent space to the fibre of the fibration n: M+ B passing through x and consequently projects into a contact element on B. Definition. Then (with the notation dK =Kudu+Kpdp+ Kqdq) Corollaries. 1.2. Vector fields on a contact manifold whose local flows preserve the contact structure are called contact vector fields. . Symplectic Geometry 79 of a Legendre fibration.
x)f+(u + </J(x). ] on the space of smooth functions which gives a Lie algebra structure on this space and has the property of localness. The correspondence Vf+iva maps the space of contact fields bijectively onto the space ofsmoothfunctions. Since the symplectic structure on N is given as the differential of the canonical 1form a. multiplies the form a by an invertible function.B. 2. This result is a generalization of the theorem on symplectic leaves for Poisson manifolds. x) is a contactomorphism of (IR x N. Example. Then the Lagrange bracket [F. the contactification of N is defined. With the previous coordinate notations </JF=Fu. a 2 is exact (a 1 a 2 =d</J). from which. inasmuch as it does not satisfy the Leibniz rule. a) is the manifold IR x N with the contact structure du =a. the cited invariant formula for the Lagrange bracket follows. w) with an exact symplectic structure a different choice of the potential a (da = w) leads to different contact structures on IR x N. Contactification. Therefore we may associate to a contact Hamiltonian K a new function ¢K by the rule LvKa = ¢Ka. Nevertheless if the difference of two potentials a 1 . du a 1) . G] of two functions will take the form [F. The Lie algebra structure introduced in this manner on the space of smooth functions on M is called the Lagrange bracket. In the special case N = T* B the con tactification of the manifold N is the space of 1jets of functions on B. G] =(LvFG.80 V. and the like) for Lie structures have not been studied. It follows from the intrinsic definition of the Lagrange bracket that the expression on the righthand side satisfies the Jacobi identity (this is a non obvious formula). One can show [42] that a Lie manifold canonically breaks up into smooth symplectic and contact manifolds.e. of course. g and of their partial derivatives of arbitrary order at the point x. In particular. For a given symplectic manifold (N. By definition the contactification of the manifold (N. g] lx depends only on the values of the functions f. chap. linearization. Let N be the symplectification of a contact manifold. Amol'd. [F. A contact diffeomorphism.I. then the corresponding structures are equivalent in the following sense: the translation (u. It is defined in the case when a symplectic structure on a manifold N is given as the differential of a 1form a. preserving the contact structure. G] =FGuFuGp(FpGuFuGp)FpGq+FqGp. where u is the coordinate on IR. triviality ofthe bundle L+M implies the global Hamiltonicity of alliR x invariant locally Hamiltonian fields on L. [f. i. A. We shall denote as a Lie structure on a manifold a bilinear operation [ . Givental' 2.3. 2). The analogues of the other properties of Poisson structures (trans versal structures. The explicit description of this operation looks like this. The Lagrange bracket does not give a Poisson structure (§3.Lv 0 F + F</JaG</JF)/2.
If the closed form a. The situation described is typical. An embedded circle in the symplectic plane does not possess a contactification: the integral Jpdq is equal to the area of the region bounded by this circle and is different from zero. IA = d</J.a. The question of the existence of exact Lagrangian embeddings is nontrivial already for the twodimensional torus.)EIRxNIJ. A closed ndimensional manifold has no exact Lagrangian embeddings into the standard 2ndimensional symplectic space. The function <Pis defined uniquely up to the addition of a locally constant function on A. the sphere cannot be embedded Lagrangianly in IR 4 . and we see that the contactification is nonunique. A. Let there be given a contactification IR x N+ N. applied to the twodimensional torus. Proof The selfintersection number of an orientable submanifold in a tubular neighbourhood of it is the same as in the containing space. The con tactification exists only under certain conditions of topological triviality and may give a nonunique result. This theorem.EA}. If a. the projection of a Legendre circle in IR 3 onto the symplectic plane has selfintersection points. implies the existence of a symplectic manifold diffeomorphic to IR 4 but not symplectomorphic to any region in the standard symplectic space IR 4 . Symplectic Geometry 81 onto (IR x N. In other words. Theorem ([7]). There exist exact Lagrangian em beddings of the torus in the space IR 4 with a nonstandard symplectic structure. It is not difficult to convince oneself that a contactification of the Lagrangian manifold exists precisely in the case when the closed !form a. Gromov. 2. The selfintersection number in a tubular neighbourhood is equal to the Euler characteristic of the normal bundle. By a contactification of a Lagrangian manifold A c N is meant a Legendre submanifold L c IR x N which projects diffeomorphically onto A. 2 ). Lagrangian Embeddings in IR 2 ". 2 is not a total differential. du. The self intersection number in Euclidean space is zero. For a Lagrangian submanifold the normal bundle is isomorphic to the tangent bundle.4. Theorem (M. Here is yet another example of this sort. . Exact Lagrangian embeddings of the torus in the standard symplectic space do not exist. 1984). 1 . Lagrangian embeddings which admit a contactification will be called exact. The torus admits Lagrangian em beddings in IR 4 .a. D In particular. The symplectification of contact objects always exists and leads to a topologically trivial symplectic object. IA on A is exact. An orientable compact Lagrangian submanifold in the sym plectic space IR 2 " has vanishing Euler characteristic.). then one may set L={(</J(A. then these contact manifolds might not be contactomorphic.
However Gromov's arguments are different from the variational methods described there and are based on the study of quasiKiihler structures on a symplectic manifold. The possibilities of the methods he developed extend far beyond the scope of the theorem stated above. The geometric meaning of Lagrangian selfintersections § 3.I. Fig. . 1t turns out that an arbitrary firstorder partial differential equation F(u. x) = 0 in the contact space of 1jets of functions of x. This result is closely connected with the generalizations of Poincare's geo metric theorem which were discussed in sect.1dimensional) intersection II(x) n Txr of the hyperplane of the contact field II with the tangent space to the hypersurface. Givental' The theorem may be reformulated as follows (see [7]): a compact hypersurjace of afront 10 in J 0 1R" with an everywhere nonvertical tangent space has a vertical chord with parallel tangent spaces at the ends (Fig. 2. 28). By the characteristic direction 1(x) at a point x E r is meant the kernel of the restriction of the differential dxa of the contact 1form a to the (generally 2n. x)=O admits a reduction to a system of ordinary differential equations on the hypersurface F(u.B. oujox.. 28. p. It has singular points where r is tangent to the hyperplanes of the contact field II.ax. An equivalent definition: on the inverse image of the hypersurface r under the symplectification L 2 " + 2 + M 2 " + 1 there is defined an IR x invariant field of directionsthe skeworthogonal complements to the tangent hyperplanes.(x) oujox.1.82 V. Arnol'd. A. 5. ° For the definition of a front see sect. 4. Let r c M 2 " +I be a hypersurface in a contact manifold. The Method of Characteristics The partial differential equation L a.1 1 of chap. 3. The projection of this field to M 2 "+ I defines the field of characteristic directions on r. 1. = 0 expresses the fact that the soughtfor function u is constant on the phase curves of the vector field Iai a. Characteristics on a Hypersurface in a Contact Space.3 of chap.
Symplectic Geometry 83 The integral curves of the field of characteristic directions are called the characteristics of the hypersurface r.2. if at this point the intersection of 'I' with r is transversal (see Fig. just like any submanifold of the base space of a Legendre fibration. Let D be a hypersurface in B and ¢ a smooth function on D. Corollary 2. The FirstOrder Partial Differential Equation. 1. x)lxeD. Let N c r be an integral submanifold ofthe contact structure. A point of the initial manifold is called noncharacteristic. u=¢(x). ¢)is meant a solution of the equation r which coincides with¢ on D. The solution of the Cauchy problem by the method of characteristics . The intersection N ='I' n r is called the initial manifold of the Cauchy problem. Proposition. A solution of the equation r is a smooth function on B whose 1graph (see sect. the 1graph of a solution consists of characteristics of the hypersurface r. By corollary 1 of the preceding item.1. We note that the points of tangency of the characteristics with the initial manifold do not satisfy this requirement. x)lxe D} in the space 1° B = IR x B. p.2) lies on r. If N has dimension n. then a Legendre submanifold of r containing a neighbourhood of the point x in N exists and is locally unique. Corollary 1. 29. which consists of all possible extensions of the 1jets of the function ¢ on D to 1jets of functions on B: 'l'={(u. Fig. This property of characteristics may also be taken as their definition. defines a Legendre submanifold 'I' c J1 B. If N is Legendrian then the characteristics passing through N lie inN. not tangent to the characteristic of the hypersurface r passing through the point X EN. We note that the initial condition defines an (n1)dimensional submanifold «<> = {(¢(x). This submanifold. Then the union of the characteristics ofr passing through N in the neighbourhood of the point x is again an integral slibmanifold. By a solution of the Cauchy problem for the equation r with the initial condition (D. 3. 29). Such an equation on an ndimensional manifold B is given by a hypersurface r in the space J1 B of 1jets of functions on B. Plrxn=dx¢}.
The equivalence wellknown in geometric optics of descriptions of the propagation of light in terms of rays and of fronts served as the prototype of the method of characteristics.B. p = . Let r c J 1 (IR") be given by the equation F (u. Givental' Theorem. p. Then the equation of the characteristics has the form x = FP. x) not be tangent to D at the point x. x) of the initial manifold is equivalent to the condition that the vector Fp(u.I. u= p FP' Noncharacteristicity of a point (u. A. p Fu. The fronts (a) and the caustic (b) of an elliptical source . A solution of the Cauchy problem in the neighbourhood of a noncharacteristic point ofthe initial manifold exists and is locally unique. Coordinate Formulas. p. Arnol'd. q)= p2 • The eikonal equation (aujaq) 2 = 1 describes the propagation of short light waves: the solution of it equal to zero on a hypersurface Din IR" is the optical length of the shortest path from the light source D to the point q. Geometrical Optics.Fx . The projections to IR" of the characteristics composing the 1graph of the function u are the normal lines (rays) to the level surfaces of the function u (the fronts).3. 30. The 1graph of the solution consists of the characteristics which intersect the initial manifold in a neighbourhood of this point. In other words. b Fig. noncharacteristicity permits one to find from the equation the derivative of the desired function at the points of D along the normal to D. x) = 0. 3. after the derivatives in the tangent directions and the value of the function have been determined by the initial condition ¢. The movement of "light cor puscules" along straight lines in IR" is described by the Hamiltonian H(p.84 V. p.
Then the functions Q(p. 30a). Let r 1 . If in the second equation variables again separate etc. 2 )+ .A. 1)+u 2 (q 2 . A. Integration of the equations of the characteristics reduces essentially to the integration of the Hamiltonian system with Hamiltonian H(p. 1 . 30b). q 1 ). The Hamiltonian of the . Indeed. then we finally arrive at a solution of the original equation of the form udq 1 . r 2 be the distance from the moving point to the centres 0 1 .A. qn)=h. if ouji}q 1 and q 1 enter into H only in the form of a combination ¢(oujoq 1 . Let us suppose that the equation ou(Q. 1 . . . . A. 0 2 (Fig. . Let us illustrate this approach with the example of Euler's problem of the attraction of a point on the plane by two fixed centres. that the Hamiltonian of the system is a function of Q. h) and we shall be able to apply Jacobi's theorem. H(A.). ••• . . 31). It was by just such a method that Jacobi integrated the equation of the geodesics on a triaxial ellipsoid. of integration of Hamilton ian systems by a reduction to the solution of a HamiltonJacobi equation has proved to be very effective in mechanics. q)joq}. Symplectic Geometry 85 The solutions of the eikonal equation may be manyvalued. It differs from the general firstorder equation in that it does not contain the required function u explicitly. The caustic may be defined as the set of the centres of curvature of the source or as the envelope of the family of rays. q)joq = p is solvable for Q. the Lagrangian manifolds AQ with the generating functions u{Q.nd and on then variables q. 5. . 1 . in particular. q) I p = ou(Q. 1 . and the fronts may have singularities. A. q 1 )=A. q) are n involutive first integrals of the Hamiltonian H. . q)=h the variable q 1 is separable.4. upon propagation of light inside an elliptical source on the plane the front acquires semicubical cusp points (Fig. A HamiltonJacobi equation is an equation of the form H(oujox. q 2 . ••• . The HamiltonJacobi Equation. x)=O.qn). ••• . D Success in applying Jacobi's theorem is always tied up with a felicitous choice of the system of coordinates in which the separation of the variables in the HamiltonJacobi equation takes place. For example. inverse to the method of characteristics. +un(q"' A. The HamiltonJacobi equation means that the restriction of H onto AQ is equal to h. we arrive at the system ¢(oujoq 1. in trying to find a solution in the form u=u 1 (q 1 )+U(q 2 . A.. q) be a solution of the HamiltonJacobi equation H(oujoq.. are the fibres of a Lagrangian fibration over the space of parameters Q. that det(o 2 ujoqoQ) # 0. q).e. The path. i.. Singularities of wave fronts and caustics will be studied in chap. One says that in the equation H(oujoq. oujoqn.n_ 1 . 3. Then. oujoq 2 . The eikonal equation is a special case of the HamiltonJacobi equation. Let u(Q. . 1.q)=h depending on n parameters Q=(h. In the neighbourhood of the caustic the light becomes concentrated. Jacobi's Theorem. . AQ = {(p. Upon movement of the front its singularities slide along a caustic (Fig.
.hrJ 2 4c2'72 drJ.>f2k(l/r 1 + l/r 2 ). Let us pass over to elliptical coordinates (~.'7 2 )+ 2k~ we may separate the variables.B. (ou/orJ) 2 (4c 2 . A. p.A. Amol'd. ~.+p. '1 In the HamiltonJacobi equation (oujo~) 2 (~ 2 4c 2 ) + (ouforJ) 2 (4c 2 .~. . 31. '1 are mutually orthogonal families of curvesdlipses and hyper bolas with the foci 0 1 .'7 2 ) = h(~ 2 .'7 2 ) + hrJ 2 = A..22· rJ '1 . whose successful solution constitutes today one of the best achievements of algebraic geometry. setting (oujo~) 2 (~ 2 4c 2 )2k~ h~ 2 =J.rJ)= I A. Integration of the Euler problem problem has the form (p.86 V. ~2 2 .I. From this we find a twoparameter family of solutions of the HamiltonJacobi I equation in the form u(h.. The endeavour to extract explicit expressions for the trajectories of Hamilton ian systems from a similar kind of formulas led Jacobi to the problem of the inversion of hyperelliptic integrals. rJ=r 1 r 2 • The level lines of the functions ~. '7) on T*IR 2 the Hamiltonian (after some computations) takes on the form e4c 2 4c 2 rJ 2 4k~ H =2p~ ~2 2 +2p. '7) on the plane: ~=r 1 +r 2 .. Givental' Fig. 0 2 • In the canonical coordinates (p~..+h~ 2 +2k~ e4c2 d~+ ).
too. p*) satisfying the incidence condition: the . The flow of the contact Hamiltonian H =. The image of the equidistant mapping is called the equidistant (compare sect. to the base space B is an equidistant of the submanifold X the set of free ends of segments of geodesics (extremals of the Lagrangian I p II) of length t sent out from X along normals. 4). The image of a Legendre mapping is called its front. Lagrangian and Legendre Mappings These are constructions which formalize in the language of symplectic geometry the concepts of a caustic and of a wave front of geometrical optics. The classification of their singularities is connected with the classification of regular polyhedra.3. Let us consider in the product iP' x IP'* of the projective space and its dual the submanifold F of pairs (p. A. and from Lie groups and Lie algebras. 1. A Legendre mapping is a diagram consisting of an embedding of a smooth manifold as a Legendre submanifold in the total space of a Legendre fibration. By abuse of language we shall call the composition of these maps the Legendre mapping. from reflection groups.II p I in the contact space ST* B of transversally oriented contact elements moves (in time t) the Legendre submanifold A 0 of elements tangent to X into a Legendre submanifold A1 • The projection of A. Perhaps this explains why the final results. and the projection of the total space of the Legendre fibration onto the base. § 1. The proofs take contributions from the theory of critical points of functions. The tangential mapping is the mapping which associates to each point of a hypersurface in a projective space the hyperplane tangent at this point. More generally. were not already obtained in the XIXth century. Thus. Fronts and Legendre Mappings. The equidistant mapping is the mapping which associates to each point of a cooriented hypersurface in Euclidean space the end point of the unit vector of the normal at this point. B. elementary in their form. chap. the equidistant mapping is Legendrian and the equidistant is its front. Symplectic Geometry 87 Chapter 5 Lagrangian and Legendre Singularities In this section we set forth the foundations of the mathematical theory of caustics and wave fronts. let B be a Riemannian manifold and X c B a smooth submanifold. Projective duality.1. 3. Examples.
1. The two contact structures on F=F* coincide. Amol'd. Its front X* in !PI* is called the dual hypersurface. 3°. q)r+(pqu. The problem of classifying the singularities of fronts reduces to the study of Legendre singular ities (i. 0 • The projection F+ !PI (F*+ !PI*) is a Legendre fibration of the manifold of contact elements F = PT*P (F* = PT*P* respectively). 3). (X*)*= X. 32). p. p. The Legendre transformation. Corollary. In the case that the function S is convex. oSfoq).88 V. the front of this mapping is again the graph of a convex function v = S*(p)the Legendre transform of the functionS (compare sect. formed by the contact elements of the submanifold X of !PI. 32. The fronts of Legendre mappings are in general not smooth. A. singularities of Legendre mappings). Therefore the tangential mapping of a smooth hypersurface is Legendrian. I The two incidence conditions coincide: F* =F. q)r+(u.2. Let us consider the two Legendre fibrations of the standard contact space IR 2 n+t of 1jets of functions on !Rn: (u. Thus the projection of a generic space curve into the plane has as its singularities only selfintersection points. Generic Legendre singularities are different from the singularities of generic mappings of ndimensional manifolds into n + 1dimensional ones. C. PT*IP = PT*IP* Fig.e. a Legendre manifold is determined by its projection to the base space of a Legendre fibration. The Legendre submanifolds A and A* coincide (Fig. Projective duality Indeed. The projection of the 1graph of a function u =S(q) onto the base of the second fibration gives a Legendre mapping qr+(qoSfoqS(q). Givental' point pE Plies in the hyperplane p* c !PI. to the base space of the second Legendre fibration PT*iPI*+iPI*. chap. p). while the projection of a generic Legendre curve in a Legendre fibration has cusp points as well. This follows from I o and the definition of the contact structure on P T* B.I. 0 The tangential mapping is just the projection of the Legendre submanifold A of PT*P. and also the submanifold F* picked out by the dual condition: the point p* E !PI* lies in the hyperplane p c !PI*. q) and (u. 2°.B. .
In this sense the effect of a Legendre equivalence is reduced to the effect of the diffeomorphism of the base on the front. Remarks. . q) = (u. 2. This remark is also applicable to singular fronts whose set of points of regularity is dense in the original Legendre manifold. A Legendre submanifold of PT*IRk+t is called regular if it is transversal to the mixed space P. The same may be asserted for the singularities of hypersurfaces projectively dual to smooth ones.. Let us give the proof of the second assertion of the lemma. so that the Legendre sub manifold germ being investigated projects injectively to the space of the coordinates (pJ. Let us consider an auxiliary trivial fibration 1Rk+ 1 + IR 1 of a "big space" IRk+' over the base space IR 1• The contact elements in IRk+t which are tangent to the fibres of the fibration form the mixed submanifold P c PT* IRk+ 1 of codimension k. Every germ of a Legendre submanifold of PT*IR 1 can be obtained by this construction from some regular Legendre submanifold. PJ> q1 . p. of a suitable auxiliary fibration. In the fibra tion PT* IR 1 + IR 1 let us introduce contact Darboux coordinates (u. I.. JuJ={1. q1 ) along the (u. 1. The last condition is violated only for germs of Legendre mappings forming a set of infinite codimension in the space of all germs. The mixed manifold is fibred over the manifold of contact elements of the base space (Fig. Lemma. 2. A smooth front of a Legendre mapping uniquely determines the original Legendre submanifold. Then from the relations . An arbitrary Legendre mapping can be given by a generating family of hypersurfaces. or for the singularities of Legendre transforms of graphs of smooth functions. 11}. . Symplectic Geometry 89 By a Legendre equivalence of two Legendre mappings is meant a con tactomorphism of the corresponding Legendre fibrations which takes the Legendre manifold of the first Legendre mapping over into the Legendre manifold of the second. qJ).2. generated by a generating hypersurface. 33). qJ)space. so that the generic singularities for fronts of Legendre mappings are the same as for equidistants). Generating Families of Hypersurfaces. p1 . The image of the projection of the intersection of a regular Legendre manifold with the mixed space Pinto the space of contact elements of the base is an immersed Legendre submanifold. One can show that (up to equivalence) all Legendre singularities can already be realized in the case of equidistants of hypersurfaces in Euclidean space.p 1 . 1. A nonsingular Legendre map ping is determined by its frontthe generating hypersurface of the Legendre manifold. In1=0. In this sense the investigation of Legendre singularities coincides with the investigation of equidistants (one can show that to nearby Legendre singularities correspond equidistants of nearby hypersurfaces and conversely. A contactomorphism of Legendre fibrations is uniquely deter mined by a diffeomorphism of the base spaces.
oF(x. Arnol'd. q) the propagation time of this wave to the point q of the space IR 1• Then the condition that the least time of motion of the light from the source X to the point q is equal to u gives the equations: 3x eX: u = F(x. Let us denote by F(x. p=F9 } with the mixed space {y=O}.f(v). through which the germ of the Legendre mapping is given by the construction described in the lemma. (1) But this is just the equation of the front for the projection to IR 1 of the intersection of the Legendre manifold {u=F. The "physical meaning" of the generating family consists in the following. where k= 111 and F(x. q)=xq1 +S(x. u e IR.I. cp{q)) which transfers r.). and r 2 in the total space of the fibration IRk+ 1 .. q 1 ). is called a generating family of hypersurfaces of this Legendre mapping (the elements of the family are in general singular intersections of the hypersurface with the fibres of the fibration !Rk+I_. A. qJ= oSjopJ. q)t+(h(x. q). Let us consider the propagation of light in IR 1 from a source X c IR1 of dimension k.IR'). u=pJqJ+S(p}o q. 33.y=Fx. each point x ofthe source radiates a spherical wave. as the generating hypersurface of the Legendre manifold u=F. Remark. y=Fx. q) in the "big" space IRk+'. q). 0 The hypersurface of the big space. The doubling of M with branching along r is the hypersurface in the direct product IR X M with the equation u2 =. By a fibred equivalence of generating families ofhypersurfaces r.90 V. The mixed space du=pdq=d(p 1 q1 )q1 dp 1 +p1dq 1 we get: there exists a germ of a smooth function S(pJo q 1 ) so that our Legendre submanifold is given by the equations PI=oSjoq. q)jox =0.!R' is meant a fibred diffeomorphism (x. into r 2 • Let r c M be a smooth hypersurface with a simple equation f =0. Now let us consider the hypersurface u=F(x.B. p=F9 • An application of the construction of the first part of the lemma leads to the original Legendre germ in PT*IR 1• The regularity condition has the form det( Fxq) # 0 and is fulfilled. Givental' Fig. In the complex case the doubling is a double covering of M . According to H uygens' principle.
its image is a Lagrangian submanifold in the space of all straight lines in IRn. . A. one must lay out along each normal the respective radii of the principal curvatures. 1.. B.. the point v of the space itself. is Lagrangian.families qf hypersurfaces are fibred stably equivalent.. The second is the Lagrangian projection T*snt+sn. which associates to a vector Iii! of a normal to a submanifold in Euclidean space.1 . Caustics and Lagrangian Mappings. The first associates to a point of the hypersurface the oriented normal to the hypersurface at that point. 35).3. The Gauss mapping of a transversally oriented hypersurface in Euclidean space IRn to the unit sphere is Lagrangian. A Lagrangian mapping is a diagram consisting of an embedding of a smooth manifold as a Lagrangian submanifold in the total space of a Lagrangian fibration and the projection to the base space of this fibration. 34).iJSjiJq is Lagrangian. Examples._. In fact. it is a composition of two maps. Example. Two germs of generating families (?fhypersurfaces give equiv alent germs qf Legendre mappings if and only if these . The caustic of the normal mapping of a submanifold in Euclidean space is the set of its centres of curvature: to construct the caustic. The normal mapping. Two families of hypersurfaces in auxiliary fibrations with a common base space are called stably fibredequivalent if they become fibredequivalent after a series of fibrewise doublings. Theorem ([9]). A gradient mapping q. Symplectic Geometry 91 ramified along r. The reason for the appearance of stable equivalence here will become clear in §2. which is isomorphic to the (co)tangent bundle of the sphere (Fig. The set of critical values of a Lagrangian mapping is called a caustic (Fig. The real type of the equation depends on the choice of the side of r. applied at the point u. Fig. 34. The Gauss mapping is Lagrangian C.
An automorphism of the Lagrangian bundle T* M+ M factors into the product of an automorphism induced by a diffeomorphism of the base space and a translation automorphism (p. The mixed manifold is fibred over T*IR 1 with kdimensional isotropic fibres. 1.92 V. All germs of Lagrangian mappings which are close to a given gradient (Gauss. Givental' Fig. any germ of a Lagrangian submanifold of T*IR' can be obtained as a projection of the intersection with the mixed submanifold Q of a regular Lagrangian submanifold A which is a section of the cotangent bundle of the "big space". Conversely. Therefore the generic local phenomena in these classes of Lagrangian mappings are the same as in the class of all Lagrangian mappings. A generating family offunctions of a germ of a Lagrangian mapping is a germ of a function on a "big" space which generates a regular Lagrangian section of the cotangent bundle of the "big" space and by means of the construction described above defines the given Lagrangian mapping. Remarks.a gradient (Gauss. The caustic of a Lagrangian mapping A Lagrangian equivalence of Lagrangian mappings is a symplectomorphism of the Lagrangian fibrations which takes the Lagrangian manifold of the first mapping over into the Lagrangian manifold of the second. where cf> is a closed !form on M. Amol'd. A Lagrangian submanifold of T*IRk+' transversal to Q is called regular. 35. 3. A. I. q)~+(p+ cf>(q). is given by its generating functiona potential of this !form. Let us consider an auxiliary fibration !Rk+ 1 +IR 1• The covectors in T*IRk+' which vanish on the tangent space to the fibre at their point of application form the mixed submanifold Q c T*!Rk+l. a closed !form on the base space. 2. The projection to T*IR' of the intersection An Q of a regular Lagrangian manifold is an immersed Lagrangian submanifold in T*IR'. The converse is in general not true.4. Generating Families of Functions. (The function on IRk+' is here . The caustics of equivalent Lagrangian mappings are diffeomorphic. Every germ of a Lagrangian mapping is equivalent to a germ of . normal) mapping are themselves gradient (Gauss. q). i. normal) mapping. for a suitable choice of the auxiliary fibration.e.I.B. normal). A Lagrangian section of a cotangent bundle.
qJ) for the Lagrangian fibration T*IR! 1 + IR! 1 such that the Lagrangian submanifold of T*IR 1 projects nonsingularly along the (p 1 . Two families of functions F 1 (x 1 . q)ly=Fx. Then the family of functions F(x. q) = 0. b) the intersection of A with the mixed space Q={y=O}: AnQ={(p. qJ)space. l/J(q)) + 'l'(q). q) is generating for some Lagrangian mapping if and only if rk(Fxxo Fxq)=k. In the general case the family of functions F(x. Symplectic Geometry 93 regarded as a family of functions on the fibres of the projection JR!k+ 1 +IR! 1. q)=O. q) fibred R +equivalent if there exists a fib red diffeomorphism (x. p=Fq}. p=Fq}. d) the caustic K in IR! 1: K={ql3x: F:x(x. q). F 2 (x. q) 1+ (h(x. Q2 (z 2 ) in new variables Example: the family x 3 + yz + qx (q is the parameter) IS stably fibred + qx. q). q)=xqJ+S(x. q). "q) }. such that F 2 (x. detiF":x(x. q)I=O}. q). X. The formulas ( 1) and (2) mean that the front of a generating family of hypersurfaces consists of those points of the parameter space for which the hypersurface of the family is singular. q1 ) may be taken as a generating family of the Lagrangian mapping of our Lagrangian submanifold of T*IR 1 to the base space IR'. q)IFx=O. F 2 (x 2 . qJ= oSjopJ> where S(pJ> q1 ) is some smooth function. 1. p = Fq(x. p. Summary.5. R +equivalent to the family x 3 Theorem ([9]). In this case it defines a) a regular Lagrangian submanifold A in T*JR!k+t: A={(y. q) 13x: F:x(x. The investigation of singularities of caustics and wave fronts led to the study of generating families of functions and hypersurfaces. let us give this submanifold by the equations p1 =oS/oq 1 . x. Two germs of Lagrangian mappings are Lagrangianly equiv alent if and only if the germs of their generating families are stably fibred R +equivalent. (2) We shall call two families of functions F 1 (x. q) = F 1 (h(x. depending on the point of the base space as a parameter). q). PJ> q1 . l/J(q)) and a smooth function 'l'(q) on the base space. and the caustic of a generating family of functions consists of . in general of a different number of variables. In Darboux coordinates (p 1 . c) a Lagrangian submanifold L c T*IR! 1the projection of An Q: L = {(p. will be called stably fib red R +equivalent if they become fibred R + equivalent after adding to them nondegenerate quadratic forms Q 1 (z 1 ).
l for quadratic Hamiltonians. A generating family of hypersurfaces of a Legendre mapping is the family of zero levels of some family of smooth functions.B. points at which the differential ofthe function turns to zero and the quadratic form of the second differential is degenerate. The functions with a singular zero level form a set of codimension one in this space. We regard the family of functions as a mapping of the (finite dimensional) base space of the family into the infinitedimensional space of smooth functions. Amol'd. i. Versal Deformations: An Informal Description. developed in § 3 of chap. The space of functions \?"\ Fig. The front of a Legendre mapping is just the inverse image of the set of such functions in the base space of the generating family (Fig.94 V. A generating family of functions of a Lagrangian mapping can by subtraction of a family of constants (this is an R +equivalence!) be turned into a family of functions equal to zero at the coordinate origin. § 2.I. Givental' those points of the parameter space for which the function of the family has degenerate critical points. 2. The inverse image of this set in the base is the caustic of the Lagrangian mapping being generated by the family.e. The Classification of Critical Points of Functions The theory considered below of deformations of germs of functions and hypersurfaces is analogous in principle to the finitedimensional theory of deformations. A.1. Therefore a Lagrangian mapping can be given by a mapping of the base space into the space of such functions. The functions with degenerate critical points form a set of codimension one in this space. The front as a bifurcation diagram . 36. 36). All the definitions and results of this section can be carried over verbatim to the holomorphic or realanalytic case.
The stratification of the space of functions the germ of a generic Lagrangian (Legendre) mapping with a base space of dimension less than I = 6 (resp.1. . are transversal to this set and to each orbit out of the finite list. Fib red equivalence is just the equivalence of families of points of the function space with respect to the action on it of a suitable (pseudo)group (see sect. This means that almost all of the space of functions is filled out by a finite number of orbits (Fig. 37). ___ Fig. Critical Points of Functions. considered as mappings of the base space into the function space. 37. Symplectic Geometry 95 Our programme is as follows. of the singularities of caustics (wave fronts) in spaces of fewer than I = 6 (resp. Finally. we obtain for Lagrangian and Legendre mappings implications such as: l=Z I I I l. 3. We shall call a germ of a Lagrangian (Legendre) mapping stable. We obtain the following result: the germ of a Lagrangian (Legendre) mapping at a point is stable if and only if the germ of its generating family at that point is versal with respect to the equivalence in the corresponding function space 11 . We shall need the following equivalence relations on the space of germs at 0 of holomorphic (smooth) functions on IC" (resp. Further on we shall cite results on the classification of germs of functions and we shall see that the beginning part of this classification is discrete. 1). In the language of generating families this means that germs near to a given generating family are fibredequivalent to it. on ~"). see sect. 2. we shall obtain an explicit description. 11 For the definition of a (mini)versal deformation. I. chap.2. chap. Since generic families of functions. when we have studied the miniversal deformations of representatives of the finite number of orbits. and the continuous families of orbits form a set of positive codimension I in the space of functions.1. if all nearby germs of Lagrangian (Legendre) mappings are equivalent to it. up to a diffeomorphism. 7) dimensions. 7) is stable and equivalent to one of the germs of a finite list. 3.
If the number of the latter is finite for an arbitrary small deformation. The germ of a function at a critical point of finite multiplicity is Requivalent to its Taylor polynomial of sufficiently high order. where 4> = O(lxl 3 ). ± x. in the neighbourhood of a nondegenerate critical point the function is Requivalent to a function ± xf ± . + const. For the construction of minimal generating families it is only necessary that in the constructions of sects. By the corank of a critical point is meant the corank of the second differential of the function at that point. Givental' Requivalent germs are taken into each other by the germ at 0 of a diffeomorphism of the preimage space. The Morse Lemma with Parameters ([9]).. The number of nondegenerate Fig.) ± x~+ 1 ± . This lemma explains the appearance of the concept of stable equivalence in the theorems on generating families: in fact the germ of a Lagrangian (Legendre) mapping at a point may be given by a germ of a generating family with zero second differential of the function at that point. At a critical (or singular) point of a function its differential turns to 0. R+equivalent germs become Requivalent after addition of a suitable con stant to one of them.I. A.. V)equivalent to sums of the same germ of rank 0 with nondegenerate quadratic forms of the appropriate number of additional variables. 1. Germs of functions (possibly of a different number of variables) are called stably R(R +. .2 and 1. A degenerate critical point falls apart into nondegenerate ones upon deform ation (Fig. then the critical point is said to be of finite multiplicity. 38). By the Morse lemma [9]. x.B.4 one choose the number of "pathological" variables pJ to be minimal..96 V. ••. Fibred equivalence of in this sense minimal generating families means the same as the equivalence of the original mappings. In the holomorphic case finite multiplicity is equivalent to isolatedness of the critical point. V)equivalent if they are R (R +. Vequivalent germs become Requivalent after multiplication of one of them with the germ of a nonvanishing function ( Vequivalence of germs of functions is just equivalence of the germs of their zerolevel hypersurfaces). ± x. 38. Morsification . The critical point is nondegenerate if the quadratic form of the second differential of the function at that point is so. . . Arnol'd. The germ of afunction at a critical point of corank r is Requivalent to the germ at 0 of a function of the form const + tj>(x 1 .
El :f(x. except for the + R . A5. Remarks. £ 7 : /(x. The nonsimple germs form a set of codimension 6 in these spaces. the critical value is a modulus with respect to R equivalence on the space of germs of functions at a given point. A7. Figure 39 depicts the adjacencies of the simple classes and the unimodal classes bordering on them in the space of functions.. A6 . but is not a modulus for R +equivalence on this space. 2. Adjacencies of simple singularities of functions . Examples: the modality of an arbitrary quadratic Hamiltonian on IR 2 " with respect to the action of the symplectic group is equal ton.J1 ~ l:f(x) = ±x 11 + 1 . 2) The enumerated germs are pairwise stably inequivalent.A 2. y) = x 3 + y5. The simplicity concept depends in general on the equivalence relation and is applicable to an arbitrary Lie group action on a manifold. The number of parameters (moduli) which are needed for the parametrization of the orbits in the neighbourhood of a given point of the manifold is called the modality of the point. D 2 k+lDlk+l• E 6 E6. E 8 :f(x. + V _ + V + V + R following cases: A 2 kA 2 k> A 11 A 11 ..y) = x 2 y±y11 . is stably R + (resp.  " o4 " Ds '0 Dso7os "" \"  '0 ~ '\ Es. Di. A germ of a function at a critical point is called simple if a neighbourhood of it in the space of germs of functions at this point can be covered by a finite number of equivalence classes. 3) In the holomorphic case the germs which differ only in the sign ± are equivalent among themselves.J1 ~ 4:f(x. y) = x 3 + xy 3 .A 3 . Simple Singularities. A 1 Ai"" (stably). V)equivalent to one of the following germs at zero Ai. The germs of infinite multiplicity form a set of infinite codimension in the space of germs of functions. Symplectic Geometry 97 critical points into which such a point falls apart upon deformation does not depend on the deformation and is called the multiplicity or the Milnor number J1 of the critical point. A4 . which is simple in the space ofgerms of smooth functions (with value zero at that point). A8 .E7E6 \ \ \ J10 Fig. 1) The index J1 is equal to the multiplicity of the critical point. y) = x 3 ± Y4 . A germ of a function at a critical point.. R. 1. A1. Theorem ([9]). 39.3.
In the theory of deformations of germs of functions one can prove a versality theorem [9]: germs of finite multiplicity have versal deformations (with a finite number of parameters). Miniversal Deformations. Arnol'd. Then the deformation F(x.. Let /(0) = 0. 2) The quotient algebra Q = IR {x} /(afjax) of the algebra of all germs of functions at 0 by the gradient ideal has dimension Jl. in the case of the cyclic subgroup r of order n generated by the unitary transformation of the plane (u..2"i'"v). the cube and the icosahedron. The relation f(x. It is the inverse image of this group in SU 2 which is the binary group of the polyhedron. E11 was already known in the XIXth century. y. critical points of the function/which have merged at the point 0). The definition of the binary group is as follows. the algebra of invariants is generated by the monomials x = uv.. The group of rotations of a regular polyhedron in space is a finite subgroup of S0 3 . . z. the cube and the icosahedron respectively). e 1 (x). E 8 (for the tetrahedron. the dihedra (regular polygons in space). monomials) which represent a basis of the space Q. er 1 (x) be functions (for example. y = u".axi (hajax is the germ at 0 of a vector field. 3) Let e0 (x) = 1. D 11 .I. consisting of all function germs of the form 'Lhi(x)aj.98 V. one may conceive of it as the algebra of functions on the set {x 1aflax = 0} of the Jl. v)~+(e 2 "ifnu. not necessarily equal to zero at the coordinate origin). (and is called the local algebra of the germ/. Theorem ([43]).B. All the surfaces C 2 /f for finite subgroups r c SU 2 have singularities of the types A11 (for polygons). E6 . They may be described as the binary subgroups of the regular polygons. E 7 . z) = 0 between them is just the equation of the surface C 2/f in C 3 • For example. Let us consider the finite subgroups of the group su2. the tetrahedron. e. An Rminiversal deformation of a germ of finite multiplicity (with respect to the pseudogroup of local changes of the independent variables) can be construc ted in the following manner. The group SU 2 maps epimorphically onto the rotation group S0 3 with kernel { ± l }. The Platonics. z = v" with the relation x" = yz. y.q) =f(x)+q11 _ 1 e11 .. +q 1 e 1 (x)+q 0 is Rminiversal for the germf(F is transversal to the tangent space of the orbit of the germ/). A.1 (x)+ . To the regular ngon there corresponds by definition a cyclic subgroup of order n in SU 2 • A finite subgroup r c SU 2 acts (together with SU 2 ) on the plane C 2 • The quotient space C 2/f is an algebraic surface with one singular point.4. . l) The tangent space to the orbit of the germfisjust its gradient ideal (aflax). D11 (for dihedra). Let us consider the germ/of a function at the critical point 0 of multiplicity Jl.. They are dependent. Givental' 2.5. . 2. In another context the list of the singularities A 11 . The algebra of rinvariant polynomials on C 2 has three generators X.
. Table I A.J. . +q 1 x is an R+miniversa1 deformation of the germ A~' and F(x. . xy 3 c. x. From this it follows that Rminiversal deformations of simple germs of functions are Vminiversal. y. . .3 oft he simple singularities offunctions a monomial basis ofthe algebra Q is listed in table 1. {1.. x. . q) + q0 is a Vminiversal deformation of it: indeed. I. 2. .1 E6 l.J..ax.1 F4 l. y 2 . 3.x.a. D~' orE~' is meant the hypersurface germ in the J. y. Therefore/= 'f.2.1 + . The Classification of Singularities of Wave Fronts and Caustics in Small Dimensions. y 2 ... Indeed.J. For the normal forms of sect... sect. x. X £7 I.y. y. .f 5) The analogous construction for the local algebra Q = IR{x}/(f. Symplectic Geometry 99 4) Throwing away the constant term q0 yields an R+miniversal deformation of the germ . whose points correspond to the functions with a singular zero level. 1.. .x..q)=x~'+ 1 +qr 1 x~'.l + 1 with a fixed leading coefficient and zero sum of the roots.. 3. F(x. x. an for the variables x 1 . Example.xy § 3. 1.1)).1 E8 I. ay = 1/(J. y". . . x.. xy. D~'' E~' lie in their own gradient ideal:fe(. I.e. .. . x•. x"'. ..6). By a front of type A~'.3 are quasihomogeneous (i. xy 2 o.2). The simple germs A~'. I. .1 } is a basis of the space IR{x}/(x~'). caustics. . y. y 3 . The front of type A~' is just the set of polynomials in one variable with multiple roots.1 is quasihomogeneous with the weights ax = (J. 40 are depicted the fronts A 2 and A3 . In Fig. 'y•. x. x 2 . x~'. We shall discuss generalizations of the theory of generating families for fronts and caustics originating from a source with boundary and in the problem of going around an obstacle (compare chap. the normal forms of the theorem of sect.1. xy..dimensional base space of a Vminiversal deformation of the correspond ing simple function germ.. Singularities of Wave Fronts and Caustics Classificational results will be cited for the singularities of wave fronts. in the space of polynomials of degree J..2)/(2J. .J. an example: the function x 2 y + y~'.fx) gives a Vminiversal deformation of the germf(hfx + 4Jfis the general form of a tangent vector to the class of equations of diffeomorphic hypersurfaces). x 2 y B. xn. Example. y. ...fx). . xy 2 . and their metamorphoses in time. xy. homo geneous of degree 1 after a choice of positive fractional degrees a 1 .at. y 2 . 2.
. is meant the hypersurface germ in the JJ.B. 41). Transversal fronts (caustics) By a caustic of type A. Example. Theorem.100 V. whose derivative (i. E.I. E~< with JJ. the projection of the cuspidal edge A2 of the front of the same name to the base space of the R + miniversal deformation).. A generic wave front in a space of I ~ 6 dimensions is stable and in the neighbourhood ofany of its points is diffeomorphic to the Cartesian product of a front of type A~<. A.! dimensional base space of an R + miniversal deformation of the corresponding simple function germ. 41. Singularities of wave fronts Theorem.) has a multiple root. Fig. + 1.. Amol'd. a polynomial of degree JJ. .e. A generic caustic in a space of I ~ 5 dimensions is stable and in the neighbourhood of any of its points is diffeomorphic to the Cartesian product of a . Givental' Fig. whose points correspond to the functions with degenerate critical points (i.D. The caustic of type A~< is diffeomorphic to the front of type A~<_ 1 : to a point of this caustic corresponds a polynomial of degree JJ.e. 40. or to a union of such fronts which are transversal (Fig. D~<.JJ. ~ I and a nonsingular manifold ofdimension 1.
D~'' E~' with jl. 12 and for I~ 10 (I~ 6) they may have functional moduli. The existing classification of unimodal and bimodal critical points of functions [9] carries considerable information about the singularities of generic fronts (caustics) in spaces of I:::. Generic fronts (caustics) in spaces of dimension lz 7 (lz 6) may be unstable 12 . 3. The diffeomorphisms which enter into the definition of equivalence are required to preserve the boundary (in the real case. Let us suppose that the source of a radiation is a manifold with boundary (for example. E~' are stable in all dimensions. or to a union of such caustics which are transversal.. A point is considered singular for a function on the manifold with boundary if it is critical either for the function itself or for its restriction to the boundary. Boundary Singularities.2. since the functions may have singularities also outside the boundary. A boundary should be taken to mean a nonsingular hypersurface on a manifold without boundary. Singularities of caustics in IR 3 Fronts and caustics of type A~'. The theory of boundary singularities of functions includes the usual one. I and a nonsingular manifold of dimension 1. In particular. Nonetheless a classification of the singularities of generic caustics in IR 6 . is lacking for the present.1 :::. Symplectic Geometry 101 caustic of type A~'.. each half space of the complement of the boundary). 11 (I:::. This is connected with the existence of nonsimple singularities of functions. 43). umbra and penumbra. The caustic in this case has three components in generalthe boundaries of light and umbra. 41. The corresponding theory of Lagrangian and Legendre mappings and their generating families leads to the theory of singularities of functions on a manifold with boundary. Swallowtail Pyramid Purse (elliptic umbilic) (hyperbolic umbilic J ~~~ A~ Fig. even only up to homeomorphisms.. . D~'.42. 39). In this situation the wave front has two componentsthe front of the radiation of the boundary and the front from the source proper (Fig. the first of which is P 8 (see Fig. 42.J1 + 1. and also of penumbra and light. 10) dimensions. the solar disk). a generic caustic in space is locally diffeomorphic to one of the surfaces of Figs..
6 and reduces to finding a monomial basis of the local algebra IR {x. ~ 2: ± x~'. D~<. F f: ± x 2 + y 3 • Remarks. Yn} j(xafjax.. except for the cases: Atk ~ A:lk. i5.3 on the boundary. Ei l±l: ± x + f(y. z) is the germ Ai. Givental' Fig. i5~<. Di. or of Bi. so that the singularities All. 43. F 4 such a basis is indicated in table l. germs which differ in the signs ± are equivalent in the holomorphic and Vclassification.. F 4 in space and on the plane. A source with a boundary Theorem. aflay). Ei of sect. For the germs B~<. C~<.~ Bj: (stably).. ±y. y 1 .. the second that of the front of the source itself (A 1 ). D~<. ~ 2: xy ± y~'. 2) The role of the nondegenerate singularities is played in the theory with boundary by the germs A1 : ±x±yf± . . . A. Ell may be considered as the stabilizations of the A~<. . 2.. The caustics of B~< and C~< are diffeomorphic to the fronts of B~< + 1 and C~< + 1 • The fronts and the caustics of the germs A~<. . The construction ofminiversal deformations of boundary singularities of finite multiplicity is analogous to that described in sect.I. D. 2. E~< singularities on the boundary. ±y.B. In the case of B~< the first component has the meaning of the front of the radiation from the boundary of the source (A 1 ). Ew Figure 44 depicts the caustics of B~<. 4) The theory of boundary singularities is equivalent to the theory of critical points of functions which are even in the variable u on the double covering x = u 2 . and A 1 : ±(xx0 ) 2 ±yf ± . l±l. z). but in the case of C ~<everything is the other way around. except for the case C2 k + 1 . i5~<. but also. l) The germs enumerated are pairwise inequivalent.Ci. The fronts of the miniversal families of B~< and C~< are diffeomorphic to each other and consist of two irreducible hypersurfaces in the space of polynomials of degree fJ. Cj: . . 3) The simple singularities outside the boundary are Ai. E~< are the same as for A~<. in which the hypersurfaces Cj:k+ 1 are realinequivalent. C~<.102 V. Ei .. Amol'd. where f(y. fJ. with a fixed leading coefficientthe set of polynomials with a zero root and the set of polynomials with a multiple root. ramified along the boundary x = 0. fJ. The simple germs of functions on a manifold with boundary are stably equivalent to the germs at 0 (the boundary is x = 0) either of Ai1±).
: Bz Fig. ·Az_.. F4 with Jl. 4 (I :::. 1 :::... (1JJ.... Symplectic Geometry 103 ... ··?': . . C.. 3) dimensions is stable and is diffeomorphic to the Cartesian product of the front (caustic) of a miniversal family of one of the germs A.. E. or to a union of such fronts (caustics) which are transversal.. I) and a germ of a nonsingular manifold of dimension 1JJ. 44.. A germ of a generic wave front (caustic) from a "source with boundary" in a space of I :::. B.. D. Unstable generic fronts (caustics) are encountered in spaces of dimension I~ 5 (I~ 4)...+ 1). Boundary caustics Theorem.. :::. I (JJ...
oo Es ~ n. In Euclidean space IR 3 let there be given a generic surface with boundary... Weyl Groups and Simple Fronts. The vertices of the Dynkin diagram correspond to the basis vectors of the lattice 7LJJ.. This duality generalizes the duality ofthe Band C series of simple boundary singularities and clears up the classification of unimodal and bimodal critical points of functions on a manifold with boundary [53]. 45. The caustic of such an object consists of three parts: the caustics of both Lagrangian manifolds and the projection of their intersection to the base space of the Lagrangian fibration. Dynkin diagrams 5 2 o~o . lattice) are classified by Dynkin diagrams [14].. the normals to the surface at the points of the boundary (B 2 )form the caustic F 4 at the centre of curvature of the surface. two nontangent Lagrangian manifolds which intersect along a hypersurface of each of them. 45.104 V. see Fig.oo E? B~ ~ .. Givental' Example([64]).3.<>0 ~ c. The theory of generating families of such objects reduces to the theory of singularities of functions on a manifold with boundary [58]. the edges give the scalar product of the basis vectors according to a definite rule (the absence of an edge signifies their orthogonality).oo F~t ~ Fig. the focal points of the boundary (A2 ). a>=x><>... A Weyl group is a finite group of orthogonal transformations of a Euclidean space V which is generated by reflections in hyperplanes and preserves some full dimensional integral lattice in V.. One of them is the classification of symmetry groups of regular integral polyhedra in multidimensional spaces.. Where the boundary is tangent to a line of curvature of the surface. r. ..B. The classification of simple germs of functions on a manifold with boundary is parallel to many other classifications of "simple" objects. The Weyl group corresponding to the diagram is generated by the reflections in the Es ~ AI' oo. Interchanging the two Lagrangian manifolds corresponds to interchanging the stable equivalence classes of the function itself and its restriction to the boundary [65]. The symplectic version of the theory of caustics from a source with boundary leads to the following object: in the total space of a Lagrangian fibration. The irreducible pairs (Weyl group. 3. Arnol'd. A..I. three things togetherthe focal points of the surface (A 2 ).
The Weyl group A 1 is just the group 71. The complex front of a simple boundary germ is diffeomorphic to the discriminant of the irreducible Weyl group of the same name. The Weyl groups on the plane (besides A 1 $Ad are just the symmetry groups of the regular triangle. .Ldimensional cube. Corollary.l + 1 in one variable with a given leading coefficient and zero sum of the roots. Weyl groups on IR 2 Let us consider the action of a Weyl group W on the complexified J. Az oo B2 = c2 o~o Fig. Example. It turns out that the quotient manifold vc1W is nonsingular and diffeomorphic to C~'. Examples. Fig. Theorem. The strata of the complex front of a simple boundary germ are in onetoone correspondence with the subdiagrams of the corresponding Dynkin diagram (a subdiagram is obtained by discarding a certain number of vertices of the diagram together with the edges adjoining them. 46). so that the corresponding Weyl groups coincide. 2 . 46.L dimensional space V c. Let us consider all reflections in hyperplanes (mirrors) which lie in the Weyl group W. square and hexagon (Fig. but the lattices connected with them are different. Example. the J. The discriminant of the Weyl group A~' is the variety of poly nomials with multiple roots in the J.Ldimensional space of polynomials of degree J.Ldimensional "octahedron". Symplectic Geometcy lOS hyperplanes orthogonal to the basis vectors of the lattice. For the group A~' of permutations of the roots of a polynomial of degree J. which acts by reflection on the line.L + 1 (with zero root sum) this is a fundamental theorem about symmetric polynomials: every symmetric polynomial can be uniquely represented as a polynomial in the elementary symmetric functions.w(the Vieta map) is a singular hypersurface and is called the discriminant of the Weyl group. and to B~' that of its dual. Any Weyl group is isomorphic to a direct product of irreducible ones. The image of the mirrors under the projection vc+ vc. To the diagram C~' corresponds the symmetry group of the J. 47).
The investigation of such metamorphoses leads to the problem of generic singularities in a family of Legendre mappings. F. C}it corresponds to the diagram Cw Direct correspondences between the various such classifications have not been established completely. for example. Lemma. We have already pointed out this connection in chap.I. D. B. q) = 0 of hypersurfaces in the xspace with the (q. sect.L. Givental' Y. Indeed. Metamorphoses of Wave Fronts and Caustics.106 V. 2) Besides Weyl groups. A spreading wave front will not at all moments of time be a generic front: at isolated moments of time it bifurcates. 1) The singularity with the front G2 can be obtained as a simple singularity in the theory of functions on the plane invariant with respect to the symmetry group of a regular triangle. The general reason for the universality of the A. Let us consider a family of Legendre mappings depending on one· parameter tthe time. platonics. Thus.4. 4. . 47. 1. it is given by the generating family F. although many of them are known. fronts and caustics other objects too can be classified by Dynkin diagrams. complex simple Lie groups [24]. The stratification of the front 0 4 Remarks.(x. E. Arnol'd. 3.~ I \ 0 0 0 oo / Fig. D An equivalence of metamorphoses of fronts is just a diffeomorphism of space time which takes the big fronts over into each other and preserves the time function on this space up to an additive constant: t 1+ t + const. from a simple complex Lie algebra one can construct a simple singularity of surfaces in C 3 together with its mini versa! deformation [ 15]. t)spacetime as the base space. C. The germ of the big front at each point is a germ of a front of a Legendre mapping in spacetime. We shall call the union in spacetime (the product of the base space by the time axis) of the fronts corresponding to the different values oft the big front. Gclassification remains rather mysterious.B. A.3 with regard to the example of the symplectic group Sp(2J.
Symplectic Geometry 107 Example.. Metamorphoses of wave fronts . + m = I+ 1 (Fig.. The metamorphoses in generic oneparameter families of fronts in spaces of dimension I < 6 are locally equivalent to the germs of special meta morphoses at 0......... A'l. or t = r1 • Theorem.. where f. Let us denote by (r 1 . A'l.. Special metamorphoses.... 48.. +qP_ 2 e 1 (x)+qr 1 . 48)..1 + .. . + q~< _ 1 ). for AP: x~' + 1 + q 0 x~'.J. In the space IRm x IR~' let us consider a big front 1: which is the product of IRm with the front of a simple germ of multiplicity f. ± r. Let us choose as a miniversal deformation of the simple germ fa monomial deformation of the form f(x)+q 0 eP_ 1 (x)+ ...J. A3 0  . o~~~w ~ l ~ + + o~R~~ l ~ ~ [I)tl]Ef ~~ Fig. • ' • ~ /(~ ~ A ~ nt n. ••. ~ .. where e~< _ 1 (x) represents the class of the highest quasihomogeneous degree in the local algebra of the germ (for example. <m) the coordinates on IRm and let us give a special metamorphosis by a time function on IRm x IR~' of the form t = ± q 0 ± ri ± ...
The general case can be obtained analogously by using the Vieta map vc+ vc I W of the other Weyl groups (and for m > 0. with Jl2+m=l.. A holomorphic function on Ck. a .I..2).1. xjk • • • A generic time function t(q) satisfies the requirement c = otjoq 0 j 0 ¥. [9]. ± r!.. One can show that such a diffeomorphism can be lowered to a diffeomorphism of the space of polynomials.. ..I.. The list of normal forms of the time function has been computed in the cases A._ 1 + aq 1 ± ri ± . can be reduced to its quadratic part by means of a local diffeomorphism which commutes with the action of G._ 2 x in the case of A. the Morse lemma with parameters of sect.B. A. even for simple big caustics.1 + ._ 2 x 1 in the case of D. The big caustic is given by the generating family F= ±x"+ 1 +q 0 x".) I Lxi = 0}.. then the resulting topological classification of generic meta morphoses turns out to be finite (V. x. like the bifurcations of fronts. can be described by the dissections of a big causticthe union of the momentary caustics in spacetimeby the level surfaces of the time function._ 1 (qeiC") by the space of their complex roots {(x 1 . 50 these metamorphoses are depicted for 1=3. case to the form t = t 1 or t = ± q 0 ± ri ± .. q E IR". Arnol'd. t E !Rm. and with a nondegenerate critical point at 0. the time function has a nondegenerate L L quadratic differential c · dxidx i• dx i = 0. case.. Bakhtin): for D4 t = q 0 + q 1 .. for D t t=q 0 ±q1 or t=q 0 +q 3 . and D.. +q. and in the D. Givental' The idea of the proof Let us consider the (branched) covering of the space of polynomials of the form x"+ 1 +q 0 x". these metamorphoses.. Here the spacetime is !Rm x IR"._ 3 x 2 +q. By means of an equivalence of metamorphoses the germ of a generic time function may be reduced in the A.. form= 0...0.1.. with the lack of an analogue of the Vieta map for caustics. . 49... +q. The coefficients qk then turn out to be elementary symmetric polynomials in the roots: qk = (. do not have such a universal normal form as the metamorphoses of fronts. +q..finite) group G on Ck..2 + . invariant with respect to a linear representation of a compact (for example. to the form t = t 1 or t = ± q0 q. If in the caseD. and F=xix 2 ±x~.1 + ..1 +q 0 x~. 2. .. and D.108 V. for D2 k and k ~ 3 t=q 0 ±q1 . D The metamorphoses of caustics in generic oneparameter families. Now in the case of the meta morphosis of a holomorphic front of type A 11 the proof of the theorem is completed by The Equivariant Morse Lemma [3]. • . In Figs. for D2 k+l t= ±q0 • In general oneparameter families of caustics in spaces of dimension l ~ 3 one only encounters metamorphoses equivalent to the enumerated ones of types A.. This means that as a function on the space of roots. if one also allows diffeomorphisms of the value axis of the time function. one allows diffeomorphisms of a punctured neighbourhood of the origin in spacetime which can be continuously extended to this point.1)k~:Xio . However.. ± r. in reducing the time function to normal form..
<? Q)C1)~CJ) ~~~ ~00~ A'I(+) CJDE 8 ~BB As ee~ Fig.f . the extremals are rays breaking away along tangent directions to the pencil of geodesics on the surface of the obstacle.5. the A series Not all metamorphoses of caustics can be realized in geometrical optics. 3 the .5). 3.6 of chap. the caustic of the pencil of straight rays from a smooth source on the plane has no inflection points (this will become clear in sect. Thus. Symplectic Geometry 109 . 3. In sect. 49. Metamorphoses of caustics. Fronts in the Problem of Going Around an Obstacle. The "flying saucer" of the metamorphosis A 3 ( +. In the problem of the quickest way around an obstacle bounded by a smooth surface in Euclidean space. +) also can not be realized as the caustic of a pencil of geodesics of a Riemannian metric on a threedimensional manifold. 1.
the D series singularities of a system of rays tangent to a geodesic pencil on a generic surface were studied.B.P. Here we shall describe. and the multiplicity of a more complicated critical point is equal to twice the number of points of type A 2 into which it breaks up under perturbation of the parameter q. In connection with our problem Huygens' principle is to the effect that each point x of the obstacle surface radiates into space along all rays close to the direction of the pencil geodesic on the surface at the point x.~Q J 2W !!S J liD Fig. q) is the distance between x and q in space. 52: to a generic extremal corresponds a critical point of type A 2 .110 V. What is decisive for our subsequent considerations is the fact that all the critical points of the generating family F are of even multiplicity. Givental' o ' &m £1. in supposition of the single valuedness of the time function on the obstacle surface. 50. the singularities of the time function and of its levelsthe fronts. The optical length of the path consisting of the segment of the pencil geodesic from the source to the point x and the segment of the ray from the point x to a point q of the space (Fig. q) = </J(x) + G(x. Shcherbak. 51) is F(x. A. ~~ . Metamorphoses of caustics.I. The discriminant of the symmetry group of the icosahedron appears among the normal forms in this problem. following O. The extremals of the problem of going around the obstacle which pass through the point q break away from the obstacle surface at critical points of the function F (·. The proof is depicted in Fig. q). . Amol'd. where </J(x) is the time function on the surface and G(x. q).
E~. in which all the functions have critical points only of even multiplicity. The graph of the time function in fourdimensional spacetime (the big front in the terminology of sect. D8. Such deformations of the simple germs are enumerated below.2 + . D~.. E8. D~. E8: X3 +Y 5 +q1Y 3 +q2Y 2 +q3y+q4. +qkzu+x)2du+qk1x+qk.. 0 The front of the family consists of the points in the parameter space corresponding to functions with a critical point on the zero level. J X A~k: y 2 + (uk+q 1uk.d 2du+qk. A~.. 51. 52.4) is in the neighbourhood of any point diffeomorphic to the Cartesian product of the front of one of the families A~. q) to normal form amounts to enumerating the maximal subfamilies (with a nonsingular base space) of Rminiversal families belonging to function germs with even multiplic ity. 0 E~: x 3 + Y 4 +q1Y 2 +q2y+q3. 3. The functions ofa generating family in the problem of going around a generic obstacle in space and in the case of a generic pencil of geodesics on the obstacle surface have only simple critical points. y E~: x 3 + J(u 2 +q 1x+q 2)2 du+q 3x+q4. E~ by a nonsingular manifold. . The even multiplicity of the critical points The reduction of the generic generating families F(x. Symplectic Geometry Ill Fig.. + qk. A~. 0 y D~k: f (uk1 +qluk2+ . The generating family in the problem of going around an obstacle Fig. Theorem.
a b Fig. 53. . 1) To the family A~ correspond the nonsingular points of the front. The metamorphoses of the front in the neighbourhood of a nonsingular point.B. A. Amo1'd. E~ and E~. 53a is shown the metamorphosis of a front in the neighbourhood of an inflection point of an obstacle on the plane. The discriminant of the symmetry group H 4 of this polytope is diffeomorphic to the front of the 13 A different description: the union of the tangents to the curve (1. 53 b) is diffeomorphic to the discriminant of the symmetry group H 3 of the icosahedron 13 . A4 by addition of the cube of a new variable. Going around an obstacle in IR 2 and the icosahedron discriminant 4) After breaking away from the obstacle surface in a nonasymptotic direc tion. 5) In the classification of regular polytopes (see [14]). A3 . The Lie group SU 2 of quaternions of unit length contains the binary group of the icosahedron (see sect.I. It is also diffeomorphic to the front in the neighbourhood of the points on an obstacle surface in space at which the pencil geodesic has an asymptotic direction. forming a caustic. or a swallowtail vertex of such a caustic are described by the families D~ (x 3 . 2) The selfintersection line of the swallowtailthe latter is the caustic of an R + versal deformation of the germ Accorresponds to the functions with two critical points of type A 2 . but does not change its front. 3) In Fig. 2. The discriminant of H 3 is the front of the family D~.4). This operation turns all critical points of the functions of the original family into ones of even multiplicity. The graph of the time function (Fig. 112 V. r 5 ) in IR 3 . a point on a cusp ridge. Givental' Examples. just after the pentagon and the icosahedron comes a body with 120 vertices in fourdimensional space. Its front q 2 ~ ± qi 12 is diffeomorphic to the discriminant of the symmetry group H 2 of the regular pentagon. This is how one obtains the family A~ of the preceding list.y 3 + q 1 y + q2 ). These families can be obtained from the Rversal families of the germs A 2 . r3 . the rays may become focused far away from it. Its 120 elements are just the vertices of our polytope in the space of quaternions. It can be described as follows.
Symplectic Geometry 113 family E~. 1 5 Fig. Cobordism § 1. The Maslov Index A Lagrangian submanifold of a phase space describes the phase of a short wave oscillation. the global properties of wave fronts which are preserved under metamorphoses.§ I includes a completely elementary exposition of the theory of cobordisms of wave fronts on the plane. 54. 54). we are compelled to abandon all concern for the inexperienced reader. The corresponding cobordism theories reflect. Chapter 6 Lagrangian and Legendre Co bordisms Cobordism theory studies the properties of a smooth manifold which do not change when it is replaced by another manifold of the same dimension which together with the first forms the boundary of a manifold of dimension one greater (Fig. The Maslov index associates integers to the curves on the 14 Such a ray breaks away from the obstacle at a parabolic point. In this chapter the manifolds and the sheets bounded by them will be Lagrangian or Legendre submanifolds. In the problem of going around an obstacle this front is encountered as the graph of the time function in the neighbourhood of a point of intersection of an asymptotic ray with a cusp ridge of the caustic far away from the obstacle surface 14 . . 15 Beginning with § 2. If we now turn to the classification of the irreducible Coxeter groups~finite groups generated by reflections but not necessarily preserving an integral lattice (see [14]). for example. By way of compensation. we have encountered the discriminants of all such groups except for the symmetry groups of the regular ngons with n z 6. then we will discover that among the wave fronts in the various problems of geometrical optics.
[q }. Then for h+0 · 'P(q. 114 V. and Jl.+ ~"on this traectory for 0 < r < t. the number of I 0 critical points of the Lagrangian projection L. project noninjcctively into the configuration space ~" (Fig. Qt(q. To equation (I) corresponds the classical mechanical system with Hamilton ian H = p 2 /2 + U (q) on the standard symplectic space ~ 2 ". Givental' Lagrangian submanifold.i is the Morse index of the trajectory g'xi. To an initial condition of the form '¥(q. Fig. unlike L 0 .t)=f(qi)+ J(pdq. chap. i./oqlq=qj are different from zero.1.joqi l. which is satisfied by the probability amplitude '¥(q.B.i) + O(h). TA'¥ + U(q)'¥ (1) (h is the Planck constant. where Si(Q. in.e.t) is the action along the trajectory g'xi: Si(Q. In the following sections the Maslov index will appear in the role of the simplest characteristic class of the theory of Legendre and Lagrangian cobordisms. A. The flow g' of the Hamiltonian H defines a family of Lagrangian manifolds L. 1. 1. 55). 55. A is the Laplace operator on ~"). The family of Lagrangian mappings .I. Amol'd.Hdr).112 exp Si(~ T) . t) of a quantummechanical particle moving in the potential U. The Quasiclassical Asymptotics of the Solutions of the SchrOdinger Equation. which for large t may. p 2_ g. There arises a family of Lagrangian mappings (see sect.) q. The following asymptotic formula holds for the solution of equation (l) with the given initial condition [52].t I I q.: IR"+ IR".f: L = { (p. Let us consider the Schrodinger equation a'¥ h2 ihat =. q) E IR 2" I p = . t) = (i ~ cf>(qi) loQ.2. = g' L.(qi) = Q and let xi(Pi• qi) be the corresponding points of L 0 • Let us suppose that the Jacobians'joQ. These numbers enter into an asymptotic expression for the solutions of the wave equations at the shortwave limit. 5) of the configuration space into itself Q. 0) = ¢(q) exp(if(q)/h) (¢ is a function of compact support) there corresponds the function ¢ on the Lagrangian manifold L in ~ 2 " with generating function . Let qi be the points in ~"such that Q.
56) shows that this co orientation can be extended in a welldefined manner to the points of type A3• 0 The Morse index may be interpreted as a Maslov index. q). An immediate check (Fig. For an arbitrary curve.2.q) run through the Lagrangian manifold L. p0 = H(p. Let us consider the phase space ~ 2 "+ 2 with coordinates (p 0 .+ ~" to the Maslov index of the phase curve passing through this point is determined by the sign of the derivative o 3 Sjov 2 ot. Indeed. This definition needs to be made more precise. Lemma. whose ends do not lie on r. in ~ 2 ". minus the number of opposite crossings.e. therefore the third differential is well defined. where (p. Proof of the lemma. Symplectic Geometry liS 1. For the coorientation of the hypersurface r let us consider the J action integral pdq. p. According to the classification of Lagrangian singulari ties (chap. 56). and by the lemma it does not depend on this perturbation. At a singular point of type A 2 the kernel of the differential of the Lagrangian projection L+X is onedimensional and is a tangent line transversal tor. at each of its points (outside the set of codimension 3) one may state which side ofr is "positive" and which is "negative". The Maslov index of an oriented curve on Lis its intersection number with the cycle of singular points ofthe Lagrangian projection L+ X. On this line the first and second differentials of the function S vanish.1fwe set q 0 =r. A study of the normal form of the germ A 3 shows that at the points of type A 3 the hypersurface r is nonsingular (Fig. the germ of the Lagrangian mapping L+ X at a generic point has type A 1 .. The Maslov index of a generic oriented curve on L can now be defined as the number ofits crossings from the "negative" side ofr to the ''positive". smooth outside a set of codimension 3 in L. 5). The Morse index is a special case of the Maslov index. We take as "positive" that side of the hypersurface r in the direction of which the third differential of the function S decreases. In the total space of the cotangent bundle T* X of a configuration manifold X let there be given a generic Lagrangian submanifold L. The Morse Index and the Maslov Index. The hypersurface r possesses a canonical coorientation. The Maslov index of such a curve on L coincides 'with the Morse index of the original phase curve in ~ 2 ". the contribution of a critical point of type A 2 of the Lagrangian projection L. where its index is equal to 2. The set r of singular points of the Lagrangian projection L+X is a hypersurface in L. and at the points of a variety of codimension 2. Locally on the manifold L it defines uniquely up to a constant summand a smooth function Sa generating function for L. and we make the point (p. type A 2 (the generic points of r). then a:s r changes from 0 to t we obtain an n + 1 dimensional Lagrangian manifold Lin ~ 2 "+ 2 • The phase curves of the flow of the Hamiltonian H which begin in L 0 may be considered as curves on L. type A 3 . and the preceding formula becomes the socalled . q). It is different from zero. q)E~ 2 ". at the points of some hypersurface. its Maslov index may be taken to be equal to the Maslov index of a perturbation of it.?: 3. Singularities of r begin with the stratum D4 and form a set of codimension . i. q0 .
116 V.I. Amol'd, A.B. Givental'
Fig. 56. The coorientation of the cycle r
J
S = (pdq H dr) is a generating function for L, and v is a vector from the kernel
of the differential of the projection L e+ IR 2 " + 2 + IR" + 1 at the critical point under
consideration. This sign is always negative in view of the convexity of the
function H = p2 /2 + U (q) with respect to the momenta. Therefore each critical
point lying on the phase curve gives a contribution of + I both to its Maslov
index and to the Morse index.
1.3. The Maslov Index of Closed Curves. The intersection number of a closed
curve on a Lagrangian manifold L c T* X with the cooriented hypersurface r of
singular points of the projection L+X does not change upon replacement of the
curve by a homologous one. Therefore r defines a M aslov class in the
cohomology group H 1 (L, Z).
The Maslov indices of closed curves enter into the asymptotic formulas for the
solutions of stationary problems (characteristic oscillations) [52]. Let us suppose
that on the level manifold H = E of the Hamiltonian H = p2 /2 + U(q) there lies a
Lagrangian submanifold L. If a sequence of numbers Jl.N+ oo satisfies the
conditions
2Jl.N §pdq =indy (mod 4)
1l: y
for all closed contours yon L (for the existence of the sequence Jl.N the existence
of at least one such number JJ.=!O is sufficient), then the equation !l.'P/2=
A. 2(U(q)E)'P has a series of eigenvalues A.N with the asymptotic behaviour
A.N= Jl.N+ O(Jl.N 1 ).
In the onedimensional case the Lagrangian manifold is an embedded circle,
its index is equal to 2, and the preceding formula becomes the socalled
Symplectic Geometry 117
"quantization condition" (see the article by A.A. Kirillov in this volume)
11 f
H=E
pdq=2n(N+t).
For example, in the case H=p 2 /2+q 2 /2 with the given Planck constant
h = l/11 we obtain the exact values for the characteristic energy levels
EN=h(N +t), N =0, l, ... of the quantum harmonic oscillator.
The Maslov class of Lagrangian submanifolds of the standard symplectic
space IR1 2 n is the inverse image of a universal class of the Lagrangian Grassmann
manifold An under the Gauss mapping. The Gauss mapping G: L+An associates
to a point of lhe Lagrangian submanifold the tangent Lagrangian space at that
point, translated to 0. The cohomology group H 1 (An, £')is isomorphic to £'.
Theorem. The generator of the group H 1 (An, £') goes over into the M aslov
class under the homomorphism G*: H 1 (An, Z)+H 1 (L, £').
Corollary. The Maslov class of a Lagrangian submanifold in IR1 2 n does not
depend on the Lagrangian projection IR1 2 n+ IR1n.
Below we shall cite two descriptions of the generators in the group
H 1(A.n, Z). We will identify JR2n = T*JRn with C" by z = q + ip.
1.4. The Lagrangian Grassmann Manifold and the Universal Maslov Class.
The Lagrangian Grassmann manifold An is the manifold of all Lagrangian
linear subspaces of the 2ndimensional symplectic space. The manifold of
all oriented Lagrangian subspaces in IR1 2 n is called the oriented Lagrangian
Grassmann manifold and is denoted by A: .
Obviously A:
is a double
covering of An.
Examples. l) A 1 =1R11P' 1 ~A{=S 1 .
2) The manifold A 2 is isomorphic to the quadric x 2 + y 2 + z 2 = u 2 + v 2 in IR11P' 4
as a projective algebraic manifold (see sect. 1.3, chap. l). The manifold A; is
diffeomorphic to S 2 X S 1• The covering A; +Az is the factorization of S 2 X S 1 by
the antipodal involution (x, y)t+( x, y) (in [37] A 2 is found incorrectly).
3) dim An= n(n + l)/2: a generic Lagrangian subspace in !R 2 n is given by a
generating quadratic form in n variables.
Theorem. The Grassmann manifolds of Lagrangian subspaces in en are
homogeneous spaces: An= Un/On, An+= Un/SOn, where Vn, On, SOn are the
unitary, orthogonal and special orthogonal groups respectively.
Indeed, an orthonormal basis of a Lagrangian subspace in C" is a unitary basis
in en and conversely, the real linear span of a unitary frame in en is a Lagrangian
subspace. Unitary bases which generate the same Lagrangian subspace can be
obtained from each other by an orthogonal transformation of this subspace. 0
Corollary. n 1 (An)=H 1 (An, Z)~H 1 (An, £')~£'.
118 V.I. Arnol'd, A.B. Givental'
Let us consider the mapping detl: Un+C x, which associates to a unitary
matrix the square of its determinant. The mapping det 2 welldefines a fibration of
the Lagrangian Grassmann manifold An over the circle S 1 = {ei<ll} of complex
numbers of modulus 1. The fibre SUn/SOn of this fibration is simply connected,
and therefore the differential 1form oc=(l/2n)(det 2 )*dc/J on An represents a
generator of the onedimensional cohomology group H 1 (An, Z). We shall call
this generator the universal M aslov class.
Example. detl: A 1 +S 1 is a diffeomorphism, and oc=dO/n, where 0 is the
angular coordinate of the line leA 1 on the plane IR 2 •
Let us define an inclusion j: Anl e+ An in the following manner. Let H be a
hyperplane in IR 2 n. The projection H+ HI H l_ into the 2n 2dimensional
symplectic space of characteristics of the hyperplane H establishes a one~ toone
correspondence between the n1dimensional Lagrangian subspaces in H/Hl_
and thendimensional Lagrangian subspaces lying in H.
Lemma. The inclusion j induces an isomorphism of the fundamental groups.
Indeed, the sequence of inclusions A 1 e+ A 2 e+ ..• e+ An maps the circle A 1 into
a circle S over which the integral of the 1form oc on An is equal to 1, i.e. the
inclusion A 1 e+ An induces an isomorphism n 1 (A.)+n 1 (An). 0
Let us now present the cycle which is dual to the generator of the group
H 1 (An, Z)=Z. Let us fix a Lagrangian subspace Lc IR 2 n and let us denote by~
the hypersurface in An formed by the Lagrangian subspaces in IR 2 n which are not
transversal to L. The Lagrangian spaces in IR 2 n which intersect L along subspaces
of dimension 2 or more form a set~' of codimension 3 in An (and 2 in~). At
points A.e~\I:' the hypersurface I: is smooth. Let us coorient I: by choosing as a
vector of the positive normal at the point A. the velocity vector of the curve ei8 A. (it
is transversal to ~).
Theorem. The intersection number of a 1cycle in An with the cooriented cycle
~ is equal to the value of the universal M aslov class on this 1cycle.
Indeed, the intersection number of the generatorS of the group H 1 (An,.Z) with
the cycle ~ is equal to 1. 0
The cycle r of singularities of the projection of a Lagrangian submanifold in
IR 2 nalong
the Lagrangian subspace Lis the inverse image ofthe cycle~ under the
Gauss map. From this the theorem of sect. 1.3 follows. 0
Remark. The Maslov index has found application in the theory of represen
tations [ 49]. In this context a series of Maslov indices is usedsymplectic
invariants of chains of k Lagrangian subspaces in IR 2 n. The simplest of them is the
triple index t(A. 1, A. 2, A. 3), equal to the signature of the quadratic form
Q(x 1 Ef)x 2 Ef)x 3)=w(x 1, x 2)+w(x 2, x 3)+w(x 3, x.) on the direct sum of the
Lagrangian subspaces A. 1Ef) A. 2 Ef) A. 3, where w is the symplectic form on IR 2n. It has
the cocycle property: t(A. 1, A. 2, A. 3) t(A. 1, A. 2, A. 4 ) + t(A.,, ). 3, A.4) r(A.2, A.3, A.4) = 0.
Symplectic Geometry 119
With the aid of the Maslov index of a quadruple of subs paces r(A. 1 , A. 2 , A. 3 ) +
r(A. 1 , ). 3 , A. 4 ) one may define on a Lagrangian submanifold in the total space of
the cotangent bundle of an arbitrary manifold a Cech cocycle corresponding to
the Maslov class of sect. 1.3 (see [37]).
1.5. Cobordisms of Wave Fronts on the Plane. The simplest example of a
Legendre cobordism is the relation between the traces of a wave front spreading
in a threedimensional medium on its boundary at the various moments of time.
These traces are not necessarily homeomorphic, but their being cobordant
imposes a restriction on the types of the singularities (see Corollary 1 below).
Two compact wave fronts F0 , F 1 on the plane are called cobordant, if in the
direct product of the plane with the interval 0 :s; t :s; 1 there exists a compact front
K transversal to the planes t = 0 and t = 1 whose intersection with the first of
these is F0 and with the second, F 1 (Fig. 57). The front K is called a cobordism. We
distinguish the cases of oriented and nonoriented, armed (cooriented) or un
armed fronts and cobordisms. An armament of a cobordism K induces an
armament of the boundary oK = F0 u F 1 , which must coincide with the own
armament of the fronts F0 , F 1 (i.e. fronts which coincide but which are spreading
to different sides are considered different and may be incobordant). The same
also relates to orientations. The addition of fronts is defined as their disjoint
union. This operation provides the set of classes of cobordant fronts with the
structure of a commutative semigroup. In the cases being considered it turns out
to be a group. The class of the empty front serves as the zero element.
Fig. 57. Cobordism of fronts in IR 2
Theorem ( [ 5] ). The group of cobordism classes of armed oriented fronts on the
plane is free cyclic (the generator is the class of the "bow", Fig. 58a), of armed
nonoriented ones is trivial, and of unarmed ones is finite: 7L 2 Ef> 7L 2 for oriented and
7L 2 for nonoriented fronts (as generators one may take the classes of the "drops",
Fig. 58b, which differ in the orientation in the oriented case).
What serves as the unique invariant of the cobordism class of an armed
oriented front on the plane is its indexthe number of cusp points (or the
number of inflection points) with signs taken into account, Fig. 58c. The index of
a compact front in IR 2 is even.
120 V.I. Arnol'd, A.B. Givental'
oc>
a
b
66
Fig. 58. a) the bow b) the drop c) the Maslov index of an armed front
Corollary 1. The algebraic number of cusp (inflection) points of the compact
trace on the plane of an oriented wave front spreading in space is even and does not
change with time.
The index of a front is connected with the Maslov index in the following way.
An armed front in IR 2 defines the conical Lagrangian surface L in T*IR 2 of
co vectors equal to zero on a contact element tangent to the front and positive on
the arming normal. For a generic front L is smoothly immersed in T*IR 2 • The
scalar product on IR 2 defines an immersion of the front into L (to a point of the
front corresponds the covector equal to 1 on the unit vector of the arming
normal). The index of an oriented front in IR 2 is equal to the Maslov index of the
constructed curve on L.
The computation of the cobordism groups described in the theorem is based
on the information on metamorphoses of wave fronts from sect. 3.4 of chap. 5.
Thus, the nonoriented cobordance to zero of the armed bow follows from the
series of restructurings of Fig. 59, using the local metamorphoses A 2 and A3 of
Fig. 48.
Fig. 59. The unoriented bow is cobordant to zero
Symplectic Geometry 121
A cobordism K of fronts defines the immersed Legendre manifold of its
contact elements in the manifold of all contact elements of the product of the
pla.ne with the interval.
Corollary 2. The Klein bottle admits a Legendre immersion into IRs.
This immersion is induced by the metamorphoses of Fig. 59 together with the
reverse series: the fronts being bifurcated in this drawing have a nonvertical
tangent everywhere and their union glues the "bow" by means of a Mobius
strip. D
Theorem ([5], compare sect. 2.4, chap. 4). The compact connected two
dimensional manifolds with an even Euler characteristic admit a Legendre immer
sion into the contact space IRs, but those with an odd Euler characteristic do not
even admit a Lagrangian immersion into IR 4 •
§ 2. Cobordisms
In [5] about a score of different Lagrangian and Legendre cobordism theories
are defined and the corresponding groups of classes of cobordant curves are
computed. Here we shall consider in the main the cobordisms of exact
Lagrangian immersionsthe theory in which the most complete results have
been obtained.
2.1. The Lagrangian and the Legendre Boundary. In the total space of the
cotangent bundle of a manifold with boundary let there be given an immersed
Lagrangian submanifold L c; T* M, transversal to the boundary o( T* M). Under
the mapping o(T*M).T*(oM) (to a covector at a point of the boundary is
associated its restriction to the boundary) the intersection L r'l o( T* M) projects
to an immersed Lagrangian submanifold oL of the total space of the cotangent
bundle of the boundary. oL is called the Lagrangian boundary of the manifold L.
The Legendre boundary oL of a Legendre manifold L c; J 1 M immersed in the
space of 1jets of functions on a manifold with boundary and transversal to the
boundary o(Jl M) is defined analogously with the aid of the projection
o(J 1 M).J 1 (oM) (to the 1jet of a function at a point of the boundary is
associated the 1jet of the restriction of the function to the boundary). In a similar
way one can define the boundary of a Legendre submanifold of the space of
(cooriented) contact elements on a manifold with boundary.
A Lagrangian cobordism of two compact Lagrangian immersed sub
manifolds L 0 , L 1 c; T*M is an immersed Lagrangian submanifold of the space
T *(M x [0, 1]), the cotangent bundle of the cylinder over M, whose Lagrangian
bo~ndary is the difference of L 1 x 1 and L 0 x 0 (for oriented cobordisms
changing the orientation of a manifold changes its sign). The manifolds L 0 , L 1
2. if there exists a Lagrangian (oriented) cobordism between them. Lagrangian (Legendre) immersions of manifolds L 1 .I. Givental' are called Lagrangianly (orientedly) cobordant.3. The Ring of Cobordism Classes. Vector Bundles with a Trivial Complexification. A. [12]. chap. .1..k of all kdimensional subspaces in a space IRN of growing dimension N (serving as the fibre of the tautological bundle over a point is the kdimensional subspace which answers to that point). 2. 5. . L 2 <:+ T*IRm as the (exact) Lagrangian immersion of the direct product L 1 x L 2 in T*IR"+m = T*IR" x T*IRm The corresponding cobordism classes form a skewcommutative graded ring with respect to the operations introduced. We are far from being able to prove this theorem. 4). called the sum of the original immersions. Legendre cobordisms are defined analogously. x 9 . The classes of Lagrangianly (Legendrianly) cobordant immer sions into T* M (J 1 M. k establishes a onetoone correspondence between the equivalence classes of kdimensional vector bundles . The theory of cobordisms of Legendre immersions in the space of 1jets of functions is equivalent to the theory of cobordisms of exact Lagrangian immersions: under the projection J 1 M+ T* M the Legendre immersed submanifolds go over into Lagrangian immersed submanifolds on which the action 1form is exact. L 2 of the same dimension into a symplectic (contact) manifold give an immersion of their disjoint union into this manifold... Every kdimensional vector bundle with a finite C W base space X can be induced from a universal classifying bundle ~k· As the latter one may take the tautological bundle over the Grassmann manifold Goo.3. 2.. Amol'd. [19]).. Lemma ([5]).] of polynomials with coefficients in 1': 2 in generators xk of odd degrees k #. Theorem ([7]. and conversely (see sect. The induction of a bundle over X under a continuous mapping X+ G x. 2. xll.122 V.. 2) The graded ring IL* = EB ILk k oriented Legendre cobordism classes in the spaces of 1jets offunctions on IRk is isomorphic (after tensor multiplication with Q) to the exterior algebra over Q with generators of degrees 1. .. . In the simplest and most important case M = IR" let us define the product of two (exact) Lagrangian immersions L 1 <:+ T*IR". 1) The graded ring 9liL* = E89l1Lk ofnonoriented k Legendre cobordism classes in the spaces of 1jets offunctions on IRk is isomorphic to the graded ring l' 2[x 5 . but we shall cite the fundamental results on the way to its proof. PT* M or ST*M)form an abelian group with respect to the summation operation.B. 9.2'. In this case in place of cobor dism of Legendre manifolds one may speak directly of cobordism of fronts. 4n + 1.
we get a mapping of X into the Grassmann manifold of kdimensional real subspaces L of Ck for which LniL=O. The Thorn space . Theorem ([27]). In the computation of the corresponding cobordism groups the key role is played by the following construction.k· In the category of oriented vector bundles the role of the classifying spaces is played by the Grassmann manifolds G!. An example: the complexification of a tangent space L to a Lagrangian manifold immersed in the realification JR 2 k of the Hermitian space Ck is canonically isomorphic to Ck = L Ef> iL.k <4) over the (oriented) Lagran gian Grassmann manifold Ak (Ak+) is classffying in the category of kdimensional (oriented) vector bundles with a trivialized complexijication. which takes the distinguished point (oo) over into the distinguished point. Symplectic Geometry 123 with the base space X and the homotopy classes of mappings of X into the classifying space G oo. Cobordisms of Smooth Manifolds. 60). In this classical theory two closed manifolds (immersed nowhere) are called cobordant. Let the compact ndimensional manifold M be embedded in a sphere of large dimension n + k. Let the complexification of a real kdimensional vector bundle over X be trivial and let a trivialization be fixed. Collapsing the complement of a tubular neighbourhood of M in X 00 T~ Fig. The tautological bundle i. For an induction of a bundle~ from a bundle 17 under a mapping of the base spaces X+ Y there is a corresponding mapping Te+ TYf of the Thorn spaces. if their difference is the boundary of some compact manifold with boundary. By the Thorn space Te of a vector bundle ~ with a compact base space is meant the onepoint compactification of the total space of this bundle (Fig.k of oriented k dimensional subspaces. If we associate to a point in X the subspace ofCk with which the fibre over it is identified under the trivialization.4. This Grassmann manifold is homotopy equivalent to the Lagrangian Grassmann manifold Ak. 60. 2.
The group 91.+k(Tek. Givental' s•+k to a point gives a mapping S"+k+ Tv of the sphere to the Thorn space of the normal bundle of the manifold M. Theorem ( [36]. is isomorphic modulo torsion to the ring of polynomials over Z with generators of degrees 4k.3.B. 2. 1) The ring 91* = ffi91.5. A mapping S"+k+ Tek can be suspended to a mapping s•+k+ 1 + T(ek EB 1) of the sphere to the Thorn space of the sum of the bundle ek with the onedimensional trivial bundle. the inverse image of the zero section Gex>. just as in the theory of cobordisms of smooth manifolds. i. Let us associate to a Legendre immersion of a manifold L in J 1 IR" a trivialization of the complexified tangent bundle TC L. .M.I. k = 1. Amol'd. Here the symbol lim denotes the following. [ 46] ).124 V. I]+ nk transversal to it is a cobordism of such manifolds. which one can succeed in solving to a significant degree. applied to a cobordism of manifolds which is embedded in S" +k x [0. is isomorphic to the ring Z 2 [y 2 .) of(oriented) cobordism classes of closed n dimensional manifolds is isomorphic to the stable homotopy group lim n. I]. ••• ] of polynomials over Z 2 in generators Yn of degrees n "# 2' I. Theorem ([69]). The Legendre Cobordism Groups as Homotopy Groups. This theorem reduces the computation of cobordism groups to a purely homotopic problem.k <:+ Tek under a mapping of the sphere S"+k+ Tek transversal to it is a smooth ndimensional submanifold in s•+k. as explained in the example of sect. . and the inverse image of the zero section under a homotopy s• + k X [0. The induction of the normal bundle from the classifying bundle ek furnishes a mapping Tv+ Tek of Thorn spaces. y 5 . which in composition with the first defines a mapping S"+k+ Tek· An analogous construction. 2. The groups of Legendre cobordism classes are isomorphic to the stable homotopy groups of the Thorn spaces of the . Which gives a mapping sn+k+ 1 + nk+ 1· It is OVer the SO arising Sequence of homomorphisms 1tn+k( Tek)+1tn+k+ 1 ( nk+ 1) that the limit is taken. y4 . Conversely. This sum can be induced from ek+ 1. This mapping is a onetoone correspondence between the set of homotopy classes of Legendre immersions L <:+ J 1 1R" and the set of homotopy classes of trivializations of the bundle Tc L. shows that for a cobordism class there is a corresponding homotopy class of mappings S" +k+ Tek. From this. one can deduce Theorem (Ya. (Q. Ehliashberg. an element of the homotopy group n. 2) The ring Q* = ffin. . 2. Theorem ([69]). A. .+k( Tek) of the Thorn spaces of the classifying (oriented) vector bundles. see [19]).e. oo).
and the second is the cohomology class of the restriction to L of the Liouville 1form pdq on the cotangent bundle T*R". These groups are as a rule too large to be readily visible. 2. The Lagrangian Cobordism Groups. but these homotopy groups have at present not been computed. (Note added in translation). 1)) = IR.k leads to the following refinement of the theorem of sect. 2. Ehliashberg's article in [19]. then the space H 1 (L. defined by the homomorphism of the fundamental groups n 1 (L)+H 1 (L. T*!R".6.. . 17The first of these two maps is the Hurewicz homomorphism.4. The point is that the action integral over a closed curve on a Lagrangian cobordismmanifold depends only on the homology class of the curve on it. ]..C l ). 1. Symplectic Geometry 125 tautological bundles over the Lagrangian Grassmann manifolds: k+ 00 k+ 00 The passage to the limit is in correspondence with the sequence of inclusions of Lagrangian Grassmann manifolds described in sect. it is true) holds also for the other Lagrangian and Legendre cobordism theories 16 . 18 The group of oriented Lagrangian cobordism classes in T*!R" is lim 1tn+k(Tk). = (}(911L*)® Z2 [J2. 0)+ R may be thought of as being given by integration of this !form along the !cycles. Only the Lagrangian cobordism groups of curves on surfaces have been computed [5]. Q) together with the linear realvalued function on it defined by the cohomology class of the action form is an invariant of the Lagrangian cobordism class of the manifold L.. Y2k• . Ehliashberg's article in [19]. which on Lis closed because Lis Lagrangian. nk(K (!R. Theorem.Q)+R 17 LetT" be the Thorn space of the bundle over K(!R. The homomorphism H 1 (L.. A Lagrangian immersion L c. The nonoriented Legendre cobordism class ofa Legendre immersion L G J 1 !R" depends only on the manifold L. l)) = 0 for k :. If the action integrals over a basis of 1cycles on a Lagrangianly immersed closed manifold L are rationally independent. An analogous expression for the cobordism groups through homotopy groups (of more unwieldy spaces. The computation of the stable homotopy groups of the spaces TA. 18 See Ya. 1) into the EilenbergMacLane space of the additive group of real numbers (n dK (IR.M. along with the Gauss mapping L+An.M.2. . gives a mapping L+ K (!R. l) x An+ induced from the tautological bundle by the projection onto the second factor. . The mapping 0: 9ll*+ 91* which associates to the cobordism class of an immersed Legendre manifold the nonoriented cobordism class of this manifold is an inclusion of graded rings and m. Theorem ([12]). Corollary. k+ 00 16See Ya.
3) H*(A +.n is the ideal generated by the squares [28]..I. § 3.. called a characteristic class of the bundle. Z 2 ) over the field Z 2 is equal to the number of partitions of k into a sum of different naturalnumber summands. Z 2 ) defined by the inclusions An<:+ G2 n..Z 2 )+ H*(A. 4) H*(A. [27]). The original cohomology class of the classifying space is called the universal characteristic class. G. Remarks.. ••• .n'+ Goo.n <:+ G oo. . . The graded stable cohomology rings of the Grassmann manifolds are as follows: 1) H*(G +.. They arise from cohomology classes of the Lagrangian Grassmann manifolds.. ] is the ring of polynomials over the . . . 1' 2..Whitney classes. Arnol'd. wk. . 2) H*(G. 1) The kernel of the epimorphism H*(G.Z 2 ) = . Z 2 ) is the exterior algebra over Z 2 in the Stiefel.Z 2 [w 1 .n+l <:+ · · ·. This circumstance sheds light on the algebraic nature of the classification of critical points of functions.1. The sequences of homomorphisms of the cohomology groups corresponding to the sequences of inclusions Goo. 0!) is an exterior algebra over 0! with integral generators of degrees 4k + 1' k = 0... Characteristic Numbers Here we shall describe discrete invariants of the Lagrangian (Legendre) cobordism class.Zthe only (and independent) invariants of the Lagrangian cobordism class of a closed curve on the symplectic plane are its Maslov index and the area of the region bounded by the curve [5]. 3. 0!) = O![p 1.Whitney classes wk of degree k. but they receive a geometric interpretation in the calculation of Lagrangian and Legendre singularities.n'+ a. We note that the dimension of the space Hk(A. . Pk• .. Characteristic Classes of Vector Bundles. An '+ An+ I '+ · · • • A: '+ An\ 1 '+ • · · • define the stable cohomology groups of the corresponding Grassmann mani folds. . If two bundles of the same dimension with a common base space have different characteristic classes obtained from one universal class. Givental' For n= I this group is IR EB . then these bundles are inequivalent. Theorem ([56]..n+1 <:+ ..B. ••• . ] is the ring of polynomials with rational coefficients in the integral Pontryagin classes Pk of degree 4k.field Z 2 in the Stiefel. A.126 V. When a vector bundle is induced by a mapping of the base space into a~classifying space a cohomology class of the classifying space determines a cohomology class of the base space.
the characteristic number w1 = 0. Theorem ([12]).2. characteristic numbers of a Lagrangian (Legendre) immersion into T* IR" ( J 1 1Rl") are defined by the n dimensional characteristic classes of a trivialization of the complexified tangent bundle of the immersed manifold and they are invariants of the Lagrangian (Legendre) cobordism class. . 2) The aiwlogous homomorphism lln+ H n(A +. 1) The group homomorphism 911L. since this cycle is homologous to zero. 3. Corollary.+ Hn(A.e. I) An analogous theorem is true for the cobordism theory of closed manifolds. Analogously. Z) is a monomorphism and becomes an isomorphism after tensoring with Q. The nonoriented Legendre cobordism class of a Legendre immersion in J 1 1Rl" is determined by the Stiefel.. the restriction of the tangent bundle of a manifold to its boundary is isomorphic to the sum of the onedimensional trivial bundle and the tangent bundle of the boundary. Symplectic Geometry 127 2) Every nonzero element of the cohomology rings enumerated in the theorem defines a nontrivial universal characteristic class which distinguishes the stable equivalence classes of (oriented) vector bundles and of (oriented) vector bundles with a trivialized complexification respectively. the characteristic number of a sum of cobordism classes is equal to the sum of the characteristic numbers of the summands. 0 Thus. For example. 3) Between the StiefelWhitney numbers of Legendre immersions there are relations.. each stable universal ndimensional characteristic class defines a characteristic number of a closed ndimensional manifold~an invariant of the cobordism class of this manifold. it is stably equivalent to the latter. Remarks. The value of an ndimensional stable characteristic class of the tangent bundle of an ndimensional closed manifold on its fundamental cycle depends only on the cobordism class of this manifold. ). The Characteristic Numbers of Cobordism Classes Lemma. Z 2 ) given by the StiefelWhitney characteristic numbers is a monomorphism. Obviously. Indeed. i.e. The value of a stable characteristic class of the tangent bundle of the manifold on the fundamental cycle of the boundary is equal to zero. i. 2) The number of partitions of a natural number k into a sum of odd summands~the dimension of the subspace Hk of the graded ring EfJHk of polynomials over Z 2 in generators of odd degree~is equal to the number of partitions of k into distinct summands (we split each even summand into a sum of 2' identical odd ones . the Maslov index of a closed Legendre curve in J 1 1Rl is even.Whitney numbers of the immersed manifold.
The Lagrang ian projection into Mn defines a stratification of the manifold L according to the types of the singularities of the Lagrangian mapping L+ Mn in correspondence with the stratification of the space of germs of functions. Complexes of Singularities. Theorem (V. the substitution x 1+ .3.x changes the orientation of the transversal. 5) The corollary has no analogue in the oriented case: the Pontryagin classes of the tangent bundle of a manifold which admits a Lagrangian immersion in T*IRn are zero. A.A. We have used the classification of critical points of functions on the line only as an illustration. and therefore the operator o is welldefined and o2 = 0. Givental' 4) The multiplication of characteristic classes together with the multiplication of the objects dual to themthe cobordism classesgives a Hopf algebra structure on H = IL* ® IQ: the comultiplication H+ H ® H is an algebra homomorphism. To each (cooriented) cycle of the universal complex v (ro) there corresponds a (cooriented) cycle of the closed (oriented) manifold L with coefficients in 7L 2 (7L). Vasil'ev [71]). In reality we need universal complexes ro and v defined analogously with respect to a discrete stratification of the spaces of germs of functions of an arbitrary number of variables into nonsingular strata which are invariant with respect to stable Requivalence of germs. A homology class of the complex v (ro) defines a cohomology class of the manifold L with coefficients in 7L 2 (7L)the intersection number of cycles on L with the corresponding (cooriented) cycle of singularities.I. 2 x 2 . the transversal to the stratum A.•• and the class containing the zero function. We shall call a stratum of finite codimension coorientable if its normal bundle possesses an orientation invariant with respect to the action of the group of germs of diffeomorphisms on the space of germs of functions. Af. For example. which does not take into account the coorientation of the strata. . 1 x 3 + A. We remark that a noncoorientable stratum enters into the boundary of a coorientable one with coefficient zero. Those which prove to be noncoorientable are the strata Afk _ 1 and only they. 3.B. Changing the orientation changes the sign of the stratum. Let there be given a generic Lagrangian immersion L G T* Mn. consists of the formal sums with coefficients in the field 7L 2 of all strata and is provided with an operator of taking the boundary of a stratum modulo 2. .128 V. The complex v. chap. The co boundary operator{) is defined using the adjacencies of strata like the usual operator of taking the boundary of chains. Let us define a complex w whose group of kchains consists of the formal integral linear combinations of the cooriented strata of codimension k. at the point x 4 may be taken of the form x 4 + }. Amol'd. Let us partition the space of germs of smooth functions of one variable at a critical point 0 with critical value zero (more preciselythe space of jets of such functions of sufficiently high order) into nonsingular stratathe Requivalence classes (see §2. A 2 . 5) of At. A4 .
Similarly to the way that the Maslov class of a generic Lagrangian immersion L c. £ 7 . The corresponding cycle of singularities is just simply a set of points with signs determined by whether or not the coorientation of the point coincides with the orientation of the manifold. E classes are simple (see sect.+). The corresponding stable Requivalence classes of germs of functions are the following: Afk _ 1 . P 8 . There exist natural homomorphisms H*(ro)+ H*(A +. 2. let us consider the space J N of jets of a high order N of germs at 0 of (oriented) Lagrangian submanifolds in T*IR". see [72]). This construction generalizes the construction of the Maslov class of sect. £ 6 . A2k (k = I. 3). 2. can be induced from suitable cohomology classes of the Lagrangian Grassmann manifolds under the Gauss mapping. Z).e. The results of the computation of the cohomology groups are listed in Table 2. The cohomology groups of the universal complexes ro and v have been computed for the strata of codimension ~ 6. taking these signs into account in the oriented case. for the theory of Legendre cobordisms in P T* M" (ST* M" or J 1 M" respectively). or of Legendre immersions in J 1 1R".3. The space J 1 is the Lagrangian Grassmann manifold A.3. and P 8 is the unimodal class of germs x 3 + ax 2 z ± xz 2 + y 2 z (a is a modulus.4. Symplectic Geometry 129 The cohomology classes on L defined by the homology classes of the universal complexes ro and v are called the characteristic classes of the Lagrangian immersion. Indeed. chap. invariant with respect to the group of germs of diffeomorphisms of the containing space. 7). and modulo 2 in the nonoriented case. There are similar constructions [71] of characteristic classes and numbers in the Legendre cobordism theory. The characteristic classes of Lagrangian immersions in T* IR". and the characteristic number is the quantity of such points. i. defined by the homology classes of the universal complexes. (A. Audio [ 19]). The fibration J N+ J 1 has a contractible fibre. H*(v)+ H*(A. 1. The Lagrangian projection T* IR"+ IR" allows one to stratify the space· J N by types of singularities of germs of Lagrangian mappings. . J N is homotopy equivalent to J 1 . 5. a 2 # 4 in the case of the + sign. The value of a characteristic class of highest dimension on the fundamental cycle of the manifold L defines a characteristic numberan invariant of the Lagrangian cobordism class. 6. 1' 2 ). 1.: T*IR" is defined as the inverse image of a cycle I: c J 1 (see sect.4) under the Gauss mapping. where the A. Theorem (M. D. Dk± (k = 4. 5). The corresponding universal complexes ro and v of cooriented and noncooriented singularities of Legendre mappings are defined according to a nonsingular stratification of the spaces of germs of (cooriented) hypersurfaces at a singular point. Coexistence of Singularities. each cycle of singularities on L is the inverse image of a corresponding cycle on the space J N under the mapping which associates to a point on L the Njet of the Lagrangian immersion at that point. The strata Afk _ 1 and Dk± turn out to be noncoorientable. 3.
on an arbitrary closed oriented Lagrangian manifold of appropriate dimension. P 8 u P 8 = 0 modulo torsion. the numbers of A4 + D4 . Hl(w) z 0 0 z2 l l JIM generators A2 As A 6 or E 6 Pa or zeros 2As A6E6 E. P 8 The cohomological product (cup product) u of characteristic classes is also a characteristic class. For example. A 3 uA 5 =E 7 =P 8 and the remaining products of dimension :::. A 2 u A 7 =E 8 =D 8 =A 8 .. A8 + D8 and A8 + E 8 are even.B. the parities of the quantities of (A 1 A 2 ). For example. the number of A 3 points on a generic closed Legendre surface in J 1 M 2 is always even. no new characteristic classes arise in this way. as are also the numbers of A 6 . for noncooriented classes A 2 u A 3 = A4 = D4 . 3P8 +E 1 Hl(v) z2 z2 z22 z2 l~ z~ generators A2 A3 A4. Not all cohomology classes of the universal complexes generate nontrivial characteristic classes of Lagrangian or Legendre immersions. D 5 . E 7 + P 8 . since the A 3 points of a generic closed wave front are pairwise joined by selfintersection lines of the wave front. A 6 + E 6 . (A 1 A4 ). D6 .130 V. The examination of points of the intersections of different strata of wave fronts leads to new characteristic numbers. Givental' Table 2 Theory k 2 3 4 5 6 T*M. +3P 8 ST*M Hl(v) l2 z2 l2 l2 l2 l~ generators A2 A3 A4 or D4 As A 6 or D6 A 1 . E1 or P 8 PT*M Hl(w) 0 0 0 l2 0 z €f> Z2 generators As P 8 . On an arbitrary closed Lagrangian immersed manifold. the following multiplicative relations between characteristic classes are valid: for cooriented classes A 2 u A 2 = 0. E6 A 1 . (A 2 A4 ). D4 As A 6. issuing from them. A 2 u A 6 = E 7 = P 8 . A.E6 and E 7 + 3P 8 . Thus. A 2 u A 6 = 3P 8 . and (A 1 A 2 A4 ) points on generic fronts of appropriate dimension turn out to be Legendre characteristic numbers. the number of A 5 points taken with regard for signs is equal to zero. D6 . E.I. of type (A 1 A 1 ). D 7 . Theorem. brought into general position with respect to the Lagrangian projection. (A 1 A6 ). The relations between cycles of singularities in the cohomology of the universal complexes impose a restriction on the coexistence of singularities. Arnol'd.. and on a nonoriented one. The . 6 of the generators out of table 2 are zero. As it turns out.
[31]. Symplectic Geometry 131 enumerated classes are cocycles of the universal complex of multisingularities defined in [72]. A 6 .1 4. * For the convenience of the reader. For example. A 5 . [23]. and Jahrbuch iiber die Fortschritte der Mathematik (FdM. . the characteristic classes A 2 . [39]. As was already noted in§ 1. the sum is the codimension of the orbit. The number of A 6 singularities. on a generic Lagrangian oriented manifold in T*IR 5 is a multiple of three.2 Ps 4.1 6. on a generic closed front in J 0 M 2 .1 5. the number of(A 1 A 2 ) points where the front is pierced by its cusp ridge is even. A 3 . A 4 . From this the assertion we made follows. references to reviews in Zentralblatt fiir Mathematik (Zbl. In conclusion let us note the almost complete parallelism between the beginnings of the hierarchies of degenerate critical points of functions and of Steenrod's "admissible sequences" [69] (x 1 ~ 2x 2 ~ "4x 3 ~ ••• ): A 2 A 3 A4 A 5 A6 A 1 A 8 2 3 4 5 6 7 D4 D5 D6 D7 D8 2.). Corollary. perhaps. w2 w3 .1 E6 E1 Es ? 4. a few contemporary monographs [1]. compiled using the MATH database. [61]. since for the open threedimensional manifold J 0 M 2 this product is zero. Theorem. References* Besides the references which were quoted.2. and on a nonoriented one.) have. as far as possible. From computations in this complex many implications may be obtained about the coexistence of multisingularities. E 1 coincide with the StiefelWhitney classes w 1 . w 1 w 2 . by the relation A 6 "' E6 in the complex co). [60]. For the manydimensional generalization of this result see [7].1 (the number of terms of the sequence is equal to the corank of the singularity. w 1 w3 . [45]. On an oriented compact Lagrangian submanifold of T* IR" the characteristic classes A 6 and P 8 coincide modulo torsion with the classes induced from three times the generator of the group H 5 (A +) and from the product of the Maslov class with this generator respectively. [48]. the Maslov class A 2 of a Lagrangian immersion in T*IR" is induced under the Gauss mapping from a generator of the group H 1 (A +). been included in this bibliography. w 1 w 2 w 3 respectively. the deficiency of E6 is explained. w2 . the list includes classical works and textbooks on dynamics [16].2 5. [76].1 3. taken with regard for sign. It turns out that the product of the cohomology classes of the space containing the front which are dual to the front and to the closure of its cusp ridge is dual to the closure of the (A 1 A 2 ) stratum.
521. 3. 186. Francaviglia. (English translation: Graduate Texts in Mathematics 60. Anal. VINITI. pp.I. 343. and for isolated parts of it in [4]. 87176 (1983)).F. Math. Anal. Moscow 1983. Topology 23. Benenti.58007 8. [67]. Zbl.): Modern Developments in Analytical Mechanics. No. in: ltogi Nauki Tekh. Turin 1983. Zbl. Vols.).: Convexity and commuting Hamiltonians.: Supplementary Chapters of the Theory of Ordinary Differential Equations. 304 p. 382 p. GuseinZade. Lichnerowicz. 1978. Paris 1987. Grundlehren der mathematischen Wissenschaften 250.58001 Zbl.I. Moscow 1974.: Singularities of Differentiable Maps. Nauka. 817 (1980) (English translation: Funct. 545.57019. 431 p. 336 p. Zbl. 858 p. Benjamin/Cummings.: Groupes et algebres de Lie. M. J. Math.70001 3. Nauka. Proc. IUTAMISIMM Symp. M. V. S. 77147 (1983) (English translation: Russ.).. Zbl. I.2.58025 12. Bott. Paris 1968. of evolvents and of fronts at an obstacle. Volume II 1987. Funct. V. 525. 537. Arnol'd.58001 Amol'd V. lIS (1982). (English translation: Geometrical Methods in the Theory of Ordinary Differential Equations. 29. S. No. Zbl. 513.I.: The moment map and equivariant cohomology.I. 35 Nl. Travaux en Cours. Varchenko. Lond.132 V.I. [47].: Lagrange and Legendre cobordisms. 472. 14. Zbl. Appl. or almost. 304 p. Zbl. Arnol'd. Nauk 38. Zbl. 46.57017 Zbl.57001. Among the bibliographical sources on our topic let us note [I].. Mass. 198 p. 806 p. Anal. 557582 (1976).. 288 p. No. Moscow 1978. Amol'd. Anal. 38. 113 (1980) and /4. Hermann.70001 2. A.B. /4.. Fourier 35. I. Bourbaki. Accademia delle Scienze di Torino. Bennequin in [19]). Cobordismes d'immersions lagrangiennes et legendriennes. [68]. Commun. 167177 and 252260 (1980)).. 522. (English translation: Birkhiiuser. [44]. Arnol'd. (eds. Funct. 462 p. Atiyah. New York HeidelbergBerlin 1978. V. Soc.I. 26792713 (1984)). Arnol'd. 559. R. Probl. Arnol'd. SpringerVerlag.. 368 (1995) Amol'd V. (Contemporary Problems of Mathematics) 22. Zbl. lnst. [9].N. II. Arnol'd.. I and II. Nauka. Surv. Volume I 1985. Zbl. 3. Bull. Ann. Zbl. untouched upon in the survey but are connected with it by their subject[18]. Arnol'd.4. Sov. SpringerVerlag. N. Math.: Singularities of Legendre varieties.: Foundations of Mechanics.00013 14. Detailed expositions of the foundations of symplectic geometryfrom various points of viewcan be found in [2] and [37]. [59]. M. 507. 159194 (1985). Audin. 20.I. 27. On the characteristic class entering in quantization conditions. 386.F. The collections [13] and [19] give a good idea of the directions of present research. Appl. Funct. A.I. 615.. V. 2 (230).M. 34 Nl.: Mathematical Methods of Classical Mechanics. V. Prilozh. 397. Ergodic Theory Dyn. 13. Atiyah. 572. 400 p. Math. and also papers which relate to questions which were completely. Anal.55002 6. 301309 (1982). Mat.: Singularities of systems of rays. New YorkHeidelbergBerlin 1983. [24]. No. 482.). Moscow 1982. Surveys 50 Nl. The quatemionic Lagrangian Grassmannian. ch.I. V. Volume I. [70] (for an important advance in contact topology we refer to D. Zbl. 355 (English translation: J. Givental' [33]. Usp. Torino/ltaly 1982. M. BostonBaselStuttgart. Volume II.53051 7. 14. Nauka. No. Reading.58012 9. Russ. [32].58013 II. 448.: Wave front evolution and equivariant Morse lemma. Appl. (2001) 10. 2. Hermann. 334 p. 128 (1984). V.: Singularities in variational calculus. (2000) Amol'd V. Arnol'd. A. Zbl. 1 Nl.I.330 . Zbl.58003 4. Elements de mathematique 34.I. 114 (1967) Amol'd V.34003 5.: Quelques calculs en cobordisme lagrangien.. Abraham. Geometry of spherical curves and algebra of quaternions. V. Math. Moscow 1984. Mat. Appl. Syst. Zbl. Pure Appl. Funkts. R. Sovrem.. The complex Lagrangian Grassmannian. [54]. [73].I.I. Ser. Marsden. [37].
16. 8393 (1982)).W. GauthierVillars. Zbl. I. E. Zbl.58025 Is. Ann.58017 22. Zbl. 175. Mishchenko.203 28. J. Akad. Semin. Godbillon.T. 2. Ser. A. Izv. 565593 (1985). Lie bialgebras and the geometric meaning of the classical YangBaxter equations. Prilozh. 383. DesolneuxMoulis. Golubitsky. Prilozh. Zbl. 19 (1982)(English translation: Funct.. 6273 (1 970) (English translation under the title: Cohomo1ogies of the group COS mod 2.. 223. Mishchenko. 285287 (1983) (English translation: Sov. 64. I. Dokl. II. Vektorn. Zbl. Fun ct.S. Mat. No. Appl.22012 16. Societatis Scientiarum Naturalium Helveticae. Carlan. Zbl. 20. 27.: On the local stability of differential forms.: Closed curves on convex hypersurfaces and periods of nonlinear oscillations. C.219225 (1977). Semin.585 32. II. Tomes IIII. Paris 1922.58004 and: Normal forms for smooth Poisson structures. 528. D. Golubitsky. Usp.M. Paris 1971.: Hamiltonian operators and the classical YangBaxter equation.I. 205221 (1976). Am. Duistermaat. Zbl.. 303306 (1968) (English translation: Sov. 503. G. (eds. 121. Anal. 340 (1955). Math.: Graphical methods in the thermodynamics of fluids.00006. 720 (1982) (English translation under the title: Mapping of periods and intersection form. Funkts. Ser. Funct. 4. Gel'fand.B. 143151 (1970)). 9699 (1968)). 2. Tischler. Zbl. Hermann.N. 1765 (reprinted in: Leonhardi Euleri Opera Omnia II 8. FdM 5. 200235).G. Fuks. Galin.58015 23. 222. 343.58015 27.: On the structure of the domains of stability of linear canonical systems of differential equations with periodic coefficients (in Russian).6871 (1983)). I (63). pp. D. Appl.. Ziirich 1964. 345. Gibbs. II: Zbl. Zbl.Ya. Akad. in: Actes du Congres International des Mathematiciens 1970. Zbl. Anal.. Moscow State University. A. Appl. pp. Zbl. 210 p.: Euler equations on finitedimensional Lie groups. Givental'. 526. Brieskorn. Nauk SSSR 268. 48. Math. 6374(1975) (English translation: Trans!.32008 33. 473. Tischler.: MaslovArnol'd characteristic classes. Im.·G. D.. Funkts.538 17.: Geometric differentielle et mecanique analytique. JJ9.. Zbl. Zbl.00016 20. Geom.M. Invent. 471.. Paris 1984. Ann. Math. 183 p.53001 25. Memoires de l'academie des sciences de Berlin /4. No. France 1983. A. Math.: Singular elements of semisimple algebraic groups. Prilozh. Zbl. Math. Soc. 309342 (18713).: Differential Geometry and Topology. Drinfel'd. P. Zbl. Mat.. Anal. Invent. 216 p. 69. FdM. Journees lyonnaises de Ia Soc. Nauk 10. V. Croke. Anal. Math. 553. D. Moscow 1983.58006 26. 497.89 31.. I /8. Zbl.: The period mapping and the intersection form.: Du mouvement de rotation des corps solides au tour d'un axe variable. A. Math. Akad. M. 521. Fomenko. /6. Anal. Lidskij.241248 (1982)). Tr. math. L. 592.58002 .. Anal.T. Gel'fand.: The cohomology of the braid group mod 2.J.4.B. /6. Paris 1969. 396415 (1978) (English translation: Math.. Hermann. Math.. Weinstein. M. 259268 (1982). Ser. Conn. A. 371389 (1978)). USSR. Zbl. 125. Transactions of the Connecti cut Academy of Arts and Sciences II. Tome 2. 199202 (1981). Dokl. Dokl. Funkts.00017. 517. 112 (1982)).S. C.T. Am. 9. /6.. J. 4. Math. 223. Euler. A.): Seminaire sudrhodanien de geometric. 1: Zbl.: On the variation in the cohomology of the symplectic form of the reduced phase space. Prilozh. No. 279284. Fomenko. 64. Tenzorn. Supplementary Chapters (in Russian). Trans.: Le<. Hermann. 577601 (1984). Nauk SSSR 178. 339. Dokl. E. 174. Mekh. Nauk SSSR. Anal.58018 21. Soc. 521. 527. 154193 (1758). V. Petrov skogo /.58003 35.B. A.B. No. Symplectic Geometry 133 15.J.70020 19. Tr.58006 30.: Hamiltonian structures on Lie groups. Fomenko. Fiz. 554 (1981). Invent. Heckman.57031 29. Fuks. Zbl.2 24. III: Zbl.: Versa! deformations of linear Hamiltonian systems.F.. 42. Dazord. 38.ons sur les invariants integraux. D. Zbl. Izv.I. J. A. II. Dorfman.: The integration of Hamiltonian systems with noncommuta tive symmetries (in Russian). Varchenko. 12.: An example of moduli for singular symplectic forms.: Normal forms for analytic Poisson structures. Ser..M.246 34. N.
IRMA de Strasbourg 1978 (revised and translated into English: Lagrangian Analysis and Quantum Mechanics.: A topological technique for the construction of solutions of differential equations and inequalities. 364. Prog. 483. Zbl. Math. Jahrhundert.58018 39. Maslov Index and Theta Series.). in: Actes du Congres International des Mathematiciens 1970. Semin. Teil II 1927.: Geometric quantization and multiplicities of group representations. 264.G. (reprint: Chelsea Publishing Company. 14. C. OxfordNew YorkToronto 1976. Im. J. J.: Local Lie algebras. 308.: Morse Theory.B. 108. 474 p. N. G. No.V. Ann. Mass.R. 300 p. Surv. Nauk 31. Lie.22011 42. Tr. Maslov. LeipzigBerlin 1930).57019 37. 315 p.: Quantization and unitary representations..: Unimodal and bimodal germs of functions on a manifold with boundary. B. No. V. Cambridge University Press.: On the classification of Lagrange immersions. L. 174189 (1981) (English translation: J.: Vorlesungen iiber Dynamik.: Vorlesungen iiber die Entwicklung der Mathematik im 19.58005 54. I. Duke Math. Math. BerlinHeidelbergNew York 1970. 237.). Zbl. Teil I und II. BostonBaselStuttgart 1980. No.J. J. Teubner. Fourier 20. 271p. 217224 (1976). 43. in: Lectures in Modern Analysis and Applications III. No. Paris 1972. Klein. Math. Dunod GauthierVillars. Inst... Nauka. Vol. 134 V. S. 656 p. Berlin 1884. 37. Zbl.363 40. Lifshitz [Lifshits]. 4 (190). Cambridge 1940. Usp. Zbl. Kirillov. 165191 (1976). A. Jacobi. I. Vol. Sternberg. V. 503. Nr. The MIT Press.: Analyse lagrangienne et mecanique quantique. Princeton. (English translation: Pergamon Press. Givental' 36. Zbl. Martinet. SpringerVerlag. Paris 1971. Moscow State University. V. No. Math.P. Berlin 1926. Cambridge. Sternberg. B. E.I.53028 45.: Geometric Asymptotics. 30.100 52. Tome 2. 67.). 2: Dynamics. Moscow 1972.A. Reimer. (English translation: Grundlehren der mathematischen Wissenschaften 220. 337 p. G.35002 48.53033 55. Zbl. 5776 (1976) (English translation: Russ. Berlin HeidelbergNew York 1976. Lychagin. Teubner. Math.: Equivalence of glancing hypersurfaces. American Mathematical Society. 14. 170. 384p. Landau. Zbl. R. (reprint of both parts in a single volume: SpringerVerlag. and Recherche cooperative sur programme CNRS No. 554 p.31933205 (1985)).53011 38 Guillemin. V.. Ann. 95178 (1970). F. Sov.). pp. 25.: Theorie der Transformationsgruppen. Birkhiiuser. I.C. 208 p. 169 p. Mat. Providence.M.58014 43. Exp. 49.A. 87208. SpringerVerlag. Petrovskogo 7. 114. Nauka.: Local classification of nonlinear first order partial differential equations.4. Zweiter Abschnitt. Surv.: Course ofTheoretical Physics. Zbl. Invent. M. 444. A. Mechanics (Third edition).47010 53. Boston 6. (French translation with 2 additional articles by V. I..G. 515538 (1982). 51. 208 p. Zbl. Zbl. I. Vergne. 354. 352. Zbl.. (reprint: B.35018 51. Guillemin. Zbl. Paris 1977. Zbl. 1963. Zbl. 385 p. No. 303 p. New York 1969) 41.London 1981.J.B. J. A. Invent.l. FdM 52. Bouslaev [Buslaev]: Etudes mathematiques. 329.: Surles singularites des formes differentielles. 189. GauthierVillars. S. 105175 (1975)). 101171 (1975) (English translation: Russ. Leray. 5575 (1976)). J. Gromov. Verlag von Julius Springer. Moscow 1965. Hamilton.336 p.22 FdM 53. 559 p.. Milnor. 221225. Lect. Seminaire sur les Equations aux Derivees Partielles (19761977). Mat. 24. Part 1: Prequantization.: Theorie des perturbations et methodes asymptotiques. BerlinHeidelbergNew York 1979). 3/. Kirillov. College de France. V. Usp.D.148 46.L.24 44. Princeton University Press.: Elements of the Theory of Representations. Mathematical Surveys No. Math.I.22005 50. Kostant. Lion. Zbl. 1977. I (181). Cunningham Mem. Moscow 1973.I.G.I. Leipzig 1890. A. Arnol'd. Arnol'd and V.58006 47. S. Nauk 30. Notes Math. 223. 497.G. 153 p.: The Weil Representation.: The Mathematical Papers of Sir William Rowan Hamilton. M. Zbl. Lees. 31. W.104 . Matov. pp. Melrose. 247. Zbl. Math. R. Stud.
C. Reg.58001 71. wl 0 #0. J. Appl.M. R.: Duality of boundary singularities.: Les methodes nouvelles de Ia mecanique celeste. 79.: A Treatise on the Analytical Dynamics of Particles and Rigid Bodies (Fourth edition). Zbl.48 p. I. Stasheff. Anal. 382/479/414 p. 1974. Math. Math. 534. Mat. Paris. Milnor.53030 60. Geom. Souriau. 17. J. Zbl.: Modeles locaux de champs et de formes. Vasil'ev. O. Cl and F 4 • Funkts. M. Pnevmatikos.58031 64.: What is a classical rmatrix?. J. 1977. J. Whittaker. J. Surv. Zbl.A. N. 523557 (1983). 17. and the group H 3 generated by reflections.: Lectures on Symplectic Manifolds.: Characteristic classes of Lagrangian and Legendre manifolds dual to singularities of caustics and wave fronts. No. Zhitomirskij. 12. J. Inc.58006 66. 3. Zbl.: Finitely determined !forms w. 195196 (1984)). 7072 (1983) (English translation: Funct. Anal. Math. Nauk 39. 576. R. 1733 (1983) (English translation: Funct. The planar nbody problem. Paris 1892. Vasil'ev.232.57017 63. Roussarie.155 70 Tischler. A 289. N. Williamson.Ya. Anal. Geom.238 61. Anal.T. Anal. S.: Traite de mecanique (2• edition). New York 1957. Prilozh.284 78. 39. Princeton University Press. 524..: On the algebraic problem concerning the normal forms of linear dynamical systems.58007 73..2. Zbl. Ser. 4564 (1970).. Cambridge University Press. Ser. R. Appl. Usp. Funkts. and II. 76.: The local structure of Poisson manifolds. 181 p. 120. Funkts. No.: Structure des systemes dynamiques. Differ. 18. Bachelier.58011 67. Anal. Sci. 441. Zbl.: Closed 2forms and an embedding theorem for symplectic manifolds. Appl. Zbl. American Mathema tical Society.. Zbl.58011 75. Appl. 581. Prilozh.: Structures hamiltoniennes en presence de contraintes. 414 p.D. 556. Nguy~n h3u Due. Zbl. I.A..: Singularities of families of evolvents in the neighborhood of an inflection point of the curve. A 291. Tomes I. 18. D. Zbl. R. 259272 (1983)). 493. Appl. Paris 1832.: Topology and mechanics I. H. Weinstein. 10. Am. Shcherbak. (This edition has been reprinted many times with new publication dates).J. 28. Zbl. Differ. Soc.G. 6869 ( 1982) (English translation: Funct.261 68. Princeton. Ann.4. Zbl. A. Anal. 1833 62. No. 17. No. Zbl. Stud. Funkts. Conf.580 69. SemenovTyanShanskij. Math. Zbl. Zbl. Paris. 58.4. J. Invent. 9091 (1984) (English translation: Funct. 286294 (1965). 327.P. Princeton. 7172 (1985) (English translation: Funct. Helv.: Characteristic Classes. 19. I. Math.206 76.57008 72. Asterisque 30 (1975). Acad. Zbl. Moser. Math.57008 57. 301303 (1983)).58002 65. 1893. S. V. 16. Zbl. Providence. C.G. 57. Nguy~n tien Dai: Stabilite de !'interaction geometrique entre deux compos antes holonomes simples. Math.: On the volume elements on a manifold. No. I. Zbl. No. II. 544. 141163 (1936). 16.: Selfintersections of wave fronts and Legendre (Lagrangian) characteristic numbers. S. Zbl. 5961 (1985)).: The Classical Groups. 29. 8485 (1984)).2. 186. 1022 (1981)(English translation: Funct. 456 p. Comment. Poisson. 2. GauthierVillars. 141. 20.164 77. Tomes 13. A.). Anal. Smale. 298. Funkts. 113116 (1980). 320 p.I. J. and caustics of groups generated by reflections Bl.A. Anal. Prilozh. 386. Shcherbak. 19. E. Thorn. 131133 (1982)). Funkts. Trans. M. Appl. 1786 (1954). Symplectic Geometry 135 56.: Focal set of a surface with boundary. 305331 (1970) and I I. 799802 (1979). Dunod. Zbl. Ser. 330 p. Weinstein. Anal. Am. 164173 (1981)). Prilozh. are exhausted by the Darboux and Martinet models. 15. Prilozh.58004 59. Acad. 1899 (reprint: Dover Publications. R. 17. 18. Poincare. Zbl.53031 74.58001 . Sci. Anal. Anal. Zbl. Shcherbak. Their Invariants and Representations. 535. No. V. Princeton University Press and University of Tokyo Press. 454.J. 91. H. Paris 1970. Weyl. 406. 13. No. Prilozh. Zbl. 203.194 58. 1946. 207208 (1984) (English translation: Russ. 202. 229235 (1977).: Quelques proprietes globales des varietes differentiables. Cambridge 1937. 15.
. V. Soc. AMS.B. Acad. Anal. O. J. Sur un propi~te topologique des applications globalement canoniques de Ia meca nique classique. 24. 50. New York 1988. A. Fields Institute Communications. Vassilyev [Vasil'ev]. 251259 (1985)) On Poisson structures: Arnol'd V. 169). Mathematics and physics: parent and child or sisters? Uspekhi Phys. Arnol'd V. Remarks on the Poisson structures on the plane and on other powers of the vol ume forms.I. Vol. 705727 (1988) On contact singularities: 5. No..I.136 V. Vol. No. N2.I. Bennequin. Soc. Invariants and perestroikas ofplance fronts. Vol.3954 7.B. Arnol'd.: Singular Lagrangian varieties and their Lagrangian mappings (in Russian).I. On Lagrangian intersections: Arnol'd V. A.60 pp. 101104 (1987) 4. 261.Uspekhi") 42 NI2 (vol. D. A.I. 194. Zametki 44. 32463278 (1990)) On cobordism theory: 10.I. Gordon and Breach. Bourbaki No. Mat. Arnol'd. Singularities of caustics and wavefronts. Sci. Mat. 489497 (1988)) 6. 52. Seminaire N. l999. 16. 18 (1999) Amol'd V. VINITI. 209.: Wavefronts and reflection groups. Topological problems in wave propagation theory and topological economy prin ciple in algebraic geometry. 260 pp. USSR.I. I.I. Proceedings ("Tzudy") of the Steklov Mathematical Institute. Ser. Arnol'd V. 125160 (1988) (English translation: Russ. 5. Math. R. Math. 149194 (1988)) 9.: Lagrange and Legendre characteristic classes. 13111323 Arnol'd V. Weinstein. First steps oflocal symplectic algebra.: Caustique mystique. Trans!. AMS.: On the interior scattering of waves defined by the hyperbolic variational principles. University Lectures Series.3. Mat. 28. Shcherbak. Nauk 43. Kluwer. Japan 40. 1990. (Contemporary Problems of Mathematics) 33. Soc. 37193722 (1965). 51 N6. Karasev. Usp.: Analogues of the objects of Lie group theory for nonlinear Poisson brackets. Surv. Anal. Funkts. 4. No. C. Math. Vol. 278 p. M. Akad. Math.I. Bull. pp. 110 (1985) (English translation: Funct. 2. 43. ser. Ser. 318 (1988) (English translation: Math. 19. 5 N3.I. Givental'. 1994. Amer. 12 (1987). Izv. . V. Prilozh. Sov.I. Nauk (trans lated as "Physics.I.: Surfaces defined by hyperbolic equations.V. No. Geom. Canadian Journal of Mathematics. Izv. Givental' Added in proof: Additional list of new publications On linear symplectic geometry: l. In: "The Arnoldfest".. 634 (1984). 497527 (1987)) 3. Am. New Ser.3.: The Sturm theorems and symplectic geometry. Probl. First steps of local contact algebra. 508538 (1986) (English translation: Math. Phys. 1956 (1986) 8. 19. in: Itogi Nauki Tekh. Sovrem. Weinstein. Proceedings Petrovski Seminar. Arnol'd V.: Coisotropic calculus and Poisson groupoids. Math. 55112(English translation: J. pp. 3746 2. Topological invariants of plane curves and caustics. 1156 (1995) Amo1'd V. Asterisque 133134. Arnol'd V.I.: Symplectic groupoids and Poisson manifolds. Nauk SSSR. 305315 (1988) Amol'd V.pp. Appl. Mat. Notes 44. Moscow 1988. 11231134 (1999) On geometrical optics: Arnol'd V.A. Paris.P. A. J.
Singularity theory. First steps of symplectic topology. Preprint Nr.. Funkts. 73.they cannot reproduce integration procedure even for so4. 361390 (1987) 16. 82. Henri Poincare.. Anal. Paragraph 2. Moreover. Hofer.. Pier. We don't know any natural and general elementary tech nique except ISTM. Fakultiit und Institut fiir Mathematik der RuhrUniversitiit Bochum. where our matrix functions Land A depend on the additional parameter >. Math. Invent. Nauk (translated as "Russian Math. Palermo 33.JL) = P(>..: Lagrangian embeddings and critical point theory. No. Math. 23. Duke Math.. 110 (1989) (English translation: Funct. 115. 349357 (1985) 13. complete algebrogeometrical integration of these sys r tems through the 0functions of follows from >. Symplectic Geometry 137 Arnol'd V. 37033743 (2000) Arnol'd V. Inst.. Non Lineaire 2.dependent Laxtype representations". Invent. C.. Ser. Sikorav. R. In: "Development of Mathematics. The BirkhoffLewis fixed point theorem and a conjecture of V. Math. 19502000" (J. Uspekhi Mat. Viterbo. Acad. Rend. Bochum 1987 15. 298 Nl7. A] implies that all coefficients of the Riemann surface r = {det(L(>. Math.6. which leads to the integration in terms of 0functions and Rie mann surfaces. Sur un theoreme de geometrie.: Persistance d'intersection avec Ia section nulle au cours d'une isotopie hamiltonienne dans un fibre cotangent. The authors of [22. Zehnder E. J.representation as a standard pro cedure of the Theory of Solitons (Inverse Spectral Transform Method. Laudenbach. Sci. 23 (1989)) Remarks Page 61. pp.) .ISTM}. 24] extracted only integrals from Manakov's work . 53. Invent. Bull. Mat. I. In this case representation L = [L. C. J. Appl. Birkhiuser. . F.. Paris. Given tal'.B. Soc. Anal. Journal of Mathematical Physics 41 N6. [71] of the article of DubrovinKrichever Novikov below. H. Basel 2000.P. A. 6395 Conely C. proof of the Arnol'd conjecture and generalizations to certain Kaehler manifolds. fonctionelles d'action et indice des systemes hamiltoniens. 132 (1986) Chaperon M. Floer. Prilozh. Ed.: Pseudo holomorphic curves in symplectic manifolds. Symplectic geometry and topology. Fr. JL) = 0} are the integrals. where corresponding integrals had been found many years before by Mishchenko in ref. We can see that all attempts to avoid ISTM here remained unsucces full until now. Arnol'd. Example 3 discussed in fact some byproducts of the so called ">. Anal.: Intersection de sousvarietes lagrangiennes. 108/1987. 375407 (1912) I I.I. 407462 (1985) 14. The LaxManakov >.}. Gromov. Surveys'') 41 N6. Une idee du type "geodesiques brisees" pour les systemes hamiltoniens. A. Ann. 293296 (1984) Poincare H.. 82.I. as polynomials or rational functions. 307347 (1985) 12. 121 (1986) Arnol'd V. representation for the EulerArnold rigid body system on the Lie algebras SOn and their generalizations was found in ref.: Intersections of Lagrangian submanifolds of symplectic manifolds. M.C.: Periodic mappings in symplectic topology (in Russian).I. Math. I.I. Circ. [74] of the article ofDubrovin KricheverNovikov below. 3349 (1983) Floer A.
v. 2n .P. j =I.k and (/1 •. The authors of [20] axiomatized this remark in the language of Lie algebras. h} + g{f.. i =I./. Nauk. fk (who are functions on K1 •. Liouvilltype integration ap pears if we can find k elements in this subalgebra !I. S. pp. K2nk : {gj. . . = const. v. Novikov [#) Nekhoroshev. h} = f{g. g2nk) such that {f. . . . generated by the inte grals g 1.. Two theorems on the actionangle variable. Givental' Page 67. It follows (like in the standard Liouville case) from the action of commutative Poisson group generated by I::=l a. fk) are independent in the generic points.. which is a special case only. a.. A.. This remark has been first made by Nekhoroshev in[#]. not only linear combinations of basis vectors as it is in Lie Algebras).B. The tenninology like "Noncommuatative Integration" and language of Lie algebras look here artificial. gj} = 0. . .I. 24. k. H} = 0. . 181198. Arnol'd. 26. g} (which is equal to some function from generators. Liouvilletype integrability is always commu tative. Actioangle variables and generalizations.. .. j = I. The algebra of C 00 functions on the phase space M is a Poisson algebra with 2 operations: coordinate multiplication of functions fg and Poisson bracket {f..... 2n. Nekhoroshev. Uspekhi Math.138 V. . such that {fg. Trudy MMO (1972)..k. . ... . N5. h}.. Let dim M = 2n and let system with hamiltonian function H E C 00 (M) be given with big Poisson subalgebra.
. . . . . . . . . . . . . Statement of the Problem . . . . . . . . . . . 155 3. . . 140 1. Hermitian Bundles with a Connection. .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Space of a Quantization . . . . . . . . . . . .4. . . . . . . . . . . . . . 161 4. . . . . . . . . . . The Definition of a Polarization . 149 2. .1. . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . .2. . . . . . . . . . . . . . . . . 147 §2. . . . . . . . . . Polarizations . . . . . .2. . . . .1. . . . . Kirillov Translated from the Russian by G. . . . . . 167 4. . . . . . . . . 169 . . . . . . . . . . . Wassermann Contents Introduction . . . . . . . . . . . . . . . .3. . . . . . . . 159 4. . . . . . . . . . . . . . 148 2. . . . . 140 §1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 4. . . Polarizations on Homogeneous Manifolds . 140 1. . . . . . . . . The Mathematical Model of Quantum Mechanics . . . . . . . . . . . . . . . . . . 153 2. . . . . . Geometric Quantization A. . . . . . . .2. . . The KoopmanVan HoveSegal Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . The SouriauKostant Prequantization . . . . Prequantization . . . A. . . Examples. . . . . . . Quantization . . 157 §4. . . . . Quantization of a Flat Space . . . . The Statement of the Quantization Problem. . . . . . . . . . 148 2. . . . The Connection with the Method of Orbits in Representation Theory . . . . . 155 §3. The General Scheme of Geometric Quantization . . . . . . . . . .2. The Connection with the Maslov Index and with the Weil Representation . . 155 3. . . Prequantization of Symplectic Supermanifolds . . . . . . . The Mathematical Model of Classical Mechanics in the Hamiltonian Formalism . . Prequantization of the TwoDimensional Sphere and the TwoDimensional Torus . . . . .3. . . . . . . . . . . . . 145 1. .
which led to the orbit method. on the one hand. Geometric quantization sets as its goal the construction of quantum objects using the geometry of the corresponding classical objects as a point of departure.A. cohomology). The Quantization Operators 170 Supplement to the Second Edition 172 References 174 Introduction The word "quantization" is used both in physical and in mathematical works in many different senses. "geometric" quanti zation. In the simplest . In asymptotic quantization one assumes that the observables and the states can be expanded as series in a small parameter h. have emerged.irillov 4. The common basis of all these theories is the premise that classical and quantum mechanics are just different realizations of the same abstract scheme. etc. § 1.140 A. and on the other hand. The merging of these two sources took place at the end of the sixties and turned out to be useful both for physicists and for mathematicians. Deformational quantization investigates the algebraic structure of the algebra of observables and regards the quantum situation as a deformation of the classical one. Although the possibilities of geometric quantization are still far from having been exhausted. and the firstorder terms give the so called quasiclassical approximation. The physicists added to their arsenal a new mathematical apparatus (fibre bundles.1. it is possible even now to give a description of the foundations of this method and to indicate approximately the scope of its applicability. The Mathematical Model of Classical Mechanics in the Hamiltonian Formalism (see [2] and volume 3 of the present publication). "deformational".. The origins of geometric quantization lie. The constant terms of these series correspond to classical mechanics. in the attempts of physicists to extend the known quantization procedures for simple mechanical systems to more general configurations and phase spaces. The fundamental components of this scheme are the algebra of observables (physical quantities) and the space of states (phase space). K. connections. In recent times this has come to be reflected explicitly in the terminology: the terms "asymptotic". Statement of the Problem 1.5. the mathematicians were enriched by new heuristic considerations and formulations of problems. in the development by mathematicians of the theory of unitary representations.
. The coordinates q 1 . i<j i<j with mutually inverse skewsymmetric matrices llwiill and llciill. 1\ dxi. In addition to this. in this case are defined only locally and map a part U of the manifold N onto a region V in IR". these canonical coordinates are far from being uniquely defined. p.I. a Poisson bracket is defined in C 00 (M). The form w is obviously closed and k =I nondegenerate on M.1) k=l However. On M = T* N the !form 8= I" pkdqk is welldefined. (1. identifying T*U with Vx (IR")*.p. run through the space (IR. In more complicated systems the phase space M is the cotangent bundle over a smooth manifold N (the configuration space). In a general system of coordinates the form wand the bivector care written in the form w= I wiidx. • . Arnol'd and A.1) holds. The inverse isomorphism is given by a bivector c. which describe the position and the velocity of the particles composing the system. ••• . For the formulation of the Hamiltonian formalism it is sufficient to have a smooth manifold M (not necessarily of the form T* N) with a closed nondegener ate form w on it.B.3) . which has the form a a L n c= ~f\~ (1. Givental') asserts that locally in suitable coordinates the form w can always be written in the form W= I" dpk 1\ dqk. Such a manifold is called symplectic. p 1 .. . Geometric Quantization 141 mechanical systems the phase space is the usual2ndimensional real vector space with coordinates q 1 . • • • . and their division into "positional" and "momentum" coordinates bears the character of a convention. giving a Lie algebra structure: (1. The Darboux theorem (compare [2] and the article of V. q.")* dual to IR" and give a trivialization of the cotangent bundle over U. • .2) k=laPk aqk in the same system of coordinates in which equality (1. With respect to the usual multiplication C'"'(M) forms a commutative associative algebra. c= I ciiai 1\ a. The form w sets up an isomorphism between the tangent and cotangent spaces at each point of M. The corresponding coordinates p 1 . • . and hence so is k=l n its differential w = d(] = I dpk 1\ dqk.q. Physical quantities or observables are identified with the smooth functions on M and form the space C'0 (M).
mixed classical states which satisfy the classical analogue of the uncertainty principle are always limits of pure quantum states. Kirillov The Jacobi identity for the Poisson bracket { {F. If such a polarization does exist. Weinstein's suggestion [46] to consider as elementary classical states <>functions concentrated on Lagrangian submani folds in M. can be identified with an open subset in a cotangent bundle T* N.4) is equivalent to the condition dOJ=O. [24]) [c. we obtain a partition of the space M into Lagrangian submanifolds. By changing the constants ck in the equations (1.142 A. cJik=OLcimomcik. F}+ {{H. (1.7) For a system which represents a mass point in IR 3 this summand can be interpreted as an external magnetic field.R. By a pure state is meant an extremal point of the set of states. the manifold is called Lagrangian.Fi}=O. The dimension of an isotropic manifold does not exceed n. F}. i.(M) which takes nonnegative values on nonnegative functions and equals 1 on the function which is identically equal to 1. or some covering of it. H} + { {G. As a rule a manifold M does not admit global polarizations. whose role can be played by an arbitrary function HE C 00 (M). Of interest in connection with this is A. This is a measure of the <>function type. (As we shall see below. G}=O (1. Such a partition is called a (global) real polarization if it is a smooth fibration.6).j. We note that under this identification the form OJ does not generally go over into the canonical form (1.) The dynamics of a system is determined by the choice of a Hamiltonian function (W. on M.. A subman!fold N c M is called isotropic if the form OJ vanishes on TN. If it is equal to n. k. H}.6) where {Fd is a system of functions which are in involution. which generate a commutative subalgebra with respect to the Poisson bracket: {Fi. where N is the base space of the polarization. G}. then the manifold M. Locally a Lagrangian manifold is given by a system of equations (1.5) ijk m where the sign O denotes the sum over the cyclic permutations of the indices i. Hamilton) or energy. The general form of such a functional is a probability measure Jl.e. and also to the vanishing of the Schouten bracket (see e.A. but differs from it by a summand of the form (1. concentrated at one point of M. A state of the system is a linear functional on coc.1) on T* N.g. Two equivalent ways' of describing the dynamics .
p}. . A function F is called afirst integral of the system if it is in involution with H. The existence of several first integrals which generate a finitedimensional Lie algebra g with respect to the Poisson bracket (S. . It is easy to verify that a mixed state with the density p(p. ..F}.D.. (1. . we obtain Hamilton's equations (1. n =idim M. It is easy to verify that this condition is equivalent to saying that the functions F 1. . Both pictures are equivalent. Liouville) and is usually used in statistical mechanics. Small oscillations. functions on M x IR. . F m locally separate points almost everywhere on M (that is. by applying ( 1. •. The other method of describing the motion consists in considering the physical quantities to be functions on M not depending on time and having the states of the system change with time in such a manner that a pure state with coordinates p(t).H}dpdq. that is {F. that is. Fm of physical quantities is called complete if from the conditions {Fi. An example of a complete set are the coordinates and the momenta in the case M = T*IR".• .1. G} = 0. A set F 1 .8) to the canonical variables Pk• qk.j Hamiltonian flow which commutes with the evolution of the system. there are enough of these functions for the construction of local coordinate systems on M).p)=f {H.1 0) This description of the motion is called the Liouville picture (J. i = 1.9) Here the dot denotes the derivative with respect to time. In the first of themthe socalled Hamiltonian picturethe states do not depend on time.F}pdpdq=f F{p. H} = {H.9). and the physical quantities are functions of the point of the phase space and of time. q. Example 1. t)= {p.ciioiFoj generates a i. In particular. ( 1. Geometric Quantization 143 are possible. H} =0. . The equations of motion have the form: F={H. t) changes here according to the law p(p. t M M The latter equality is true because the Hamiltonian flow preserves the canonical measure dpdq = al'.8) where F is an arbitrary observable. The oscillator. q. m it follows that G = const. q(t) obeys the Hamilton equations (1. Poisson) leads to a realization of the corresponding Lie group G = exp g as a group of symmetries of the system under consideration. since the mean value of the quantity Fin the state p changes with time in the same way: d d (F. In this case the vector field c(F)= L.
then the potential energy has the form: V= mgx 3 = ±mgJr 2 qfq~. y 2 . In the coordinates p and q the kinetic energy is expressed by the formula T__l_( m ..144 A.A.• 2 2 1 (pk+qk). y 3 . Global coordinates q 1 . the phase space with IR" x (IR")*.2 2+ 2 (plql +p2q2) 2 ) P1 P2 2 r . The mathematical pendulum. x 3 . that is to say. that the Poisson brackets between the fundamental observables have the form: {H.q}=l. directed along the x 3 axis. As local coordinates on None may take q 1 = x 1 . • •• . where m is the mass of the pendulum. p 1 . every manifold is flat and every function in the neighbourhood of an extremum is quadratic.2. (In the language of the Hamiltonian formalism a union of systems without interaction corresponds to multiplication of the phase spaces and addition of the Hamiltonians. q 2 =x 2 • With respect to the quadratic form T= (m/2)( yf + y~ + y~).q.) We shall therefore assume that n = I. The phase space M = T* N is the cotangent bundle of the sphere S 2 • One may identify the manifold M with the subset of IR 6 with coordinates X 1 . q. given by the equations xf+x~+x~=r 2 . The Hamiltonian H = T + V possesses a first integralthe angular momentum with respect to the x 3 axis: . y 1 . x 1y 1+x 2 y 2 +x 3 y 3 =0. with a circle p 2 + q 2 = const./. The configuration space N is the twodimensional sphere S 2 . the dual coordinates in the fibres have the form p 1 =m(y 1 x 3 x 1 y 3 )/x 3 . The trajectory of a point coincides with a level curve of the function H. Here the configuration space can be identified with IW. If the pendulum is considered to be in a constant gravitational field. Example 1.p}=. This explains the large role of a particular example of a mechanical systemthe socalled oscillator.q}=p. Pn are defined. {p.x 2 y 3 )/x 3 . This system may be regarded as a union of noninteracting systems with one degree of freedom. In this case the motion of the system consists of a uniform rotation of the phase plane with unit angular velocity. given by the equation xi+ x~ + x~ = r 2 . x 2 .. which gives the kinetic energy of the system. • •• . {H. p 2 = m(y 2 x 3 . The Hamiltonian of the system has the form: 1 ~ H=2. Let us note also. Kirillov To a first approximation.
) As we shall see below. Namely.3. in quantum mechanics consists ofthe socalled "density matrices".12) where h is Planck's constant (M. With respect to the first operation the set of observables forms a commutative. but in the form of a "growth" over the usual phase space. which allows one to construct a quantization of this system under the condition that the surface area of M is an integer. As the Hamiltonian of this system one usually considers the linear function H =a 1 x 1 +a 2 x 2 +a 3 x 3 . whose coefficients can be interpreted as the projections of the field strength vector of a magnetic field. This exotic system turns out to be the classical analogue of a quantum system with one spin degree of freedom. In applications it is encountered not all by itself as a rule.11) 2) the commutator (1. The state space. Geometric Quantization 145 The joint level lines of the functions H and P give a polarization with singularities on M. which in the local coordinates u=x 1 .2. but the remaining fibres are two dimensional tori on which a conditionally periodic motion takes place (see [2]). but not associative algebra. with respect to the second it forms a Lie algebra. As the phase manifold M let us take the sphere x~ + x~ + x~ =r 2 . the nonnegative definite operators S with the property that tr S = 1. In quan tum mechanics the physical quantities or observables are selfadjoint linear operators on some complex Hilbert space :Yt'. and for the form w let us take the usual area. that is. Planck). H = ±mgr degenerate to a point (if H = . The Mathematical Model of Quantum Mechanics (see [13]). or phase space. v=x 2 has the form: w=r(du A dv)/Jr 2 u 2 v 2 • (In this case M cannot be represented as a cotangent bundle T*Nthe configuration space does not exist! A wellknown theorem of smooth topology asserts that there are no nonvanishing (continuous) vector fields on M.) .2 .mgr) or to a pinched torus (if H = + mgr). (1.2 2 )2. 1.(AB (A+B . By the same token there are also no real polarizations. These two operations are the quantum analogues ofthe usual multiplication and the Poisson bracket in classical mechanics. the notion of spin as a "hidden rotational degree of freedom" gets an exact classical interpretation here. Thus. The pure states (the extremal points of the set of states) are . the fibres P=O. The motion of the system consists in a uniform rotation of the sphere. They form a linear space on which two bilinear operations are defined: 1) Jordan multiplication 1 AoB=2(AB+BA)= . Example 1. in this example there exists a complex polarization.
The integrals of the system are all the operators which commute with fi. given by a function 1/J..Pto then in the state 1/1 it takes on the value A. in which the operators corresponding to physical quantities do not change. where functions which differ by a numeric multiplier give the same state. The mean value of the quantity A in the stateS is by definition equal to tr AS..14) called the Schriidinger equation.k with probability Pk = 1(1/1. is a multiple of the identity.. The eigenfunctions of the SchrOdinger operator give the stationary states of the system. 1/J). (Remember that functions differing by a scalar multiplier define the same state). We shall call a set of quantum physical quantities A1 . The other description. just as in classical mechanics.6). Example 1. Heisenberg) (1. The Hilbert space Jt' consists of the complex functions 1/J(x) on the real line with a squareintegrable modulus. 1/J). I/Jk)l 2 • The dynamics of the system is defined by the energy operator fi. Am complete. Am. where EA is the spectral family (resolution of the identity) of the operator A. i = 1.13) which is the exact analogue of Hamilton's equation (1. is not obliged to have an exactly defined value. The basic physical quantities are represented .146 A. is either equal to {0} or to Jt'. Its probability distribution is given by the monotone function p(l)=(EAI/J. . if the operator A has a simple discrete spectrum with eigenvalues A. Schrodinger).A. (It is customary to place a caret above a letter to distinguish a quantummechanical quantityan operator on Jt'from the corresponding classical quantitya function on M).4.k and eigenfunctions . two means of description are possible. The motion is described by Heisenberg's equation (W. . but the quantities change. In particular. then we obtain the Heisenberg picture. Usually Jt' can be realized as a space of squareintegrable functions. is called the Schro'dinger picture (E. but the states change. in this case a pure state is given by a function 1/J with unit norm (a socalled wave function). One can show that this condition is equivalent to the irreducibility of the set AI> ••• . That means that any closed subspace of Jt' which is invariant with respect to all the A.. Here. m. this quantity is equal to (AI/I. the energy operator itself does not change with time (the law of conservation of energy in quantum mechanics). In quantum mechanics an observable A. For a pure state. 1 ~i~m. if any operator fj which commutes with A. ••• . A mass particle in a potential force field on the line. If the states do not depend on time. Kirillov the onedimensional projection operators on Jt'. In particular. It is easy to check that for the equivalence of these two ways of describing the motion it is necessary that a pure state 1/1 should vary according to the law (1. even in a pure state 1/J.
~ liz dz the energy operator H = . Geometric Quantization 147 by the following operators: the coordinate operator x consists in multiplication of the function 1/J(x) by x. Therefore there is a priori no hope that to each classical symmetry there should correspond a quantum symmetry. w). In concrete situations one or another finitedimensional symmetry group might be preserved while others might be broken. The problem of geometric quan tization is. w) is the infinite dimensional group Symp(M. then the irreducible representations of the group G must be connected with homogeneous symplectic Gmanifolds. These two groups are not isomorphic. 1. In particular. starting from the geometry of a symplectic manifold (M. The maximal symmetry group of the symplectic manifold (M. Such systems have no Ginvariant subsystems. in the state 1/1 the coordinate of the particle has a probability distribution with density II/J(x)l 2 . If the thesis is true that every quantum system with a symmetry group G can be obtained by quan tization of a classical system with the same symmetry group. w) of all symplectomorphisms. which acts on the space n* dual to the Lie algebra of the group G. the momentum operator p=iii :x. For the quantum system the maximal symmetry group is the infinite dimensional group PV(Yf) of all projective unitary transformations. That means that on the space Yf there should act a unitary representation (possibly a projective one. on which some Lie group G acts transitively. In the physical literature one speaks in the latter case of "quantum anomalies" in the commutation relations. or canonical trans formations of M which preserve the form w. A particular interest is presented by the homogeneous symplectic manifolds (M. The method of orbits in the theory of unitary representations of Lie groups ties together the unitary representations of a Lie group G with the orbits of this group in the coadjoint representation. The Connection with the Method of Orbits in Representation Theory. If the initial classical system had a symmetry group G it is natural to require that the quantum model obtained should also possess this symmetry. where ~(k)= 1/J(x)eikxdx is the Fourier transform of the function 1/J. w) which gives the model of a classical mechanical system. but its momentum is distributed with density J l~(k!liW. see below) of the group G.2m dx 2 + V(x). Here V(x) is the given potential. li = h/2n is the reduced Planck constant.3. The Statement of the Quantization Problem. . Therefore in the corresponding quantum systems irreducible representations of the group G must arise. to construct a Hilbert space Yf and a set of operators on it which give the quantum analogue of this system.
who generalized the results of D. For the case M = T*N.e. It has the form: h F=F + v 2 . and the t on the right is the identity operator). Segal [37]. Following P.A. • . Kirillov The connection between quantization and the method of orbits is founded on the following remarkable circumstance. Van Hove [45]. the space C 00 (M) equipped with the Poisson bracket (1.1) where c(F) is the Hamiltonian vector field on M with generating function F. 1. [ 42]). The procedure of geometric quantization is the natural generalization to the nonhomogeneous situation of the procedure for constructing an irreducible unitary representation of a group G starting from a Gorbit in the coadjoint representation. Koopman [25] and L. and the asterisk on the right denotes the transition to the adjoint operator). .1) a prequantization was con structed in 1960 by I. The KoopmanVan HoveSegal Representation.c(F)O(c(F)). finally. Theorem 1. § 2. Dirac. 2) {F. i. By 8li(M.3).1 ([22]. a quantization is a linear mapping F+F of the Poisson algebra (or some subalgebra of it) into the set of operators on some (pre)Hilbert space. More precisely. Every Gorbit in g* is a homogeneous symplectic Gmanifold and conversely.<i]h( =~(F<i<it)} 3) ~=(F)* (the asterisk on the left denotes complex conjugation.M. w=d8 (see sect. having the properties: t) I= t (here the t on the left denotes the function on M which is identically equal to t. [26]. Let (M. .•• .A. .148 A. w) be a sym plectic manifold. 7rl (2. Prequantization 2. A linear mapping F+ F which possesses the first three properties is called a prequantization. the concept of a complex polarization arises from the construction of a holomorphically induced representation.G}~=[F. cohomology representations have also recently found an analogue in the method of geometric quantization (see §4). Pm also form a complete set. w) we shall denote the Poisson algebra on M. 4) for some complete set of functions F 1 . quantization with the aid of a choice of a real polarization (see below § 3) generalizes the construction of an induced represen tation. .1.Fm the operators P1 . every homogeneous symplectic Gmanifold is locally isomorphic to an orbit in the coadjoint representation of the group G or a central extension of it.
In particular. If the field c(F) is complete (that is. The space of the prequantization consists of the smooth functions of compact support on M with the scalar product where v = wn = dnpdnq is the Liouville measure on M.2. Therefore. In this case the coordinate and momentum operators have the form: on the space of smooth functions on the plane with coordinates p. That condition 2) is fulfilled follows from the wellknown identities [3] [c(F). c(G)) = c(F)O(c(G)). The point is that the possibility of representing the operators of the prequantization by means of formula (2. We shall consider a generalization of this method below in §3. Condition 3) is equivalent to the condition that the operators F be selfadjoint for real functions FE C 00 (M). if the form w is not exact (for example this will be so for an arbitrary compact manifold M ). then the representation (2. The prequantization (2. equipped with a connection and a Hermitian structure on the fibres.1) is not a quantization. q and measure dp 1\ dq. c(G)]). G}) and { F. generates a oneparameter group of transformations of M).0( [c(F). 2.2 shows that in the present case one can construct a quantization if one restricts the action of q and p to the space of functions not depending on p. A comparison with example 1. Geometric Quantization 149 considered as an operator on C 00 (M). c(G)] =c( {F. G} = w(c(F). However a closed form w is always exact locally. then the operator F is essentially selfadjoint. That the mapping is linear and condition 1) is fulfilled is obvious. That these operators are symmetric follows from the fact that the field c(F) is Hamiltonian. Hermitian Bundles with a Connection.c(G)O(c(F)).1) is impossible. if the function F has compact support. The SouriauKostant Pre quantization. So one can cover the manifold . condition 3) is fulfilled. It is easy to check that the operators ()j()p and ()j()q + (2ni/h)p commute with q and p. as can be seen in the simplest example M = T*IR. Therefore q and p do not form a complete set.1) is based on the equality w =dO. The attempt to carry over the KoopmanVan HoveSegal construction to general symplectic manifolds leads to complex line (=one dimensional vector) bundles over M.
The verification of condition 2) requires some calculations. Under this identification the operator V~ takes on the form: (2.2) Here (s 1 . h 21[l. (2.er(L. The form 0.3) with (2. We shall suppose that on L a Hermitian structure ( ·. Vc(FJ· F=F+. but on the sections of some line bundle Lover M. which are compatible in a natural fashion: (2. with the additional condition that the cohomology class given by the form w be integral. s2 )= J(s 10 s2 )dv (2. U.) by the formula C"'(U.(e)·s.4) A comparison of (2. s 2 ) denotes the function on M which at a point xeM is equal to the scalar product of s1(x) and s2 (x) in the sense of the Hermitian structure.. can be "glued together" to one global operator F. (2.. Let L be a complex vector bundle over M with a onedimensional fibre. U. ·) and a connection V are defined.).. if one defines a scalar product on the space of sections of L by the formula (s 1 .. then the space of sections r(L..· By the same token we obtain the possibility of defining operators F...= To. is defined out of the equation 2ni v~s. holds for a suitable 1 form 0. on Ua.1) suggests the following fonnula of the SouriauKostant prequantization: .A..150 A. these local operators F.3) where the form 0.=F+(h/2ni)c(F)O..5) That condition 1) is fulfilled is obvious. It turns out that.· Let us pass over to exact formulations... Kirillov M by open sets U. such that in each Ua..) can be identified with C"'(U..).5)..)3l/J ++ l/J •s. Condition 3) is fulfilled for F with compact support by virtue of(2.... Let us recall that the curvature form of the connection V is defined as the 2form n on M given by .6) M and observes that the volume dv = w" = dnp A d"q is invariant with respect to the Hamiltonian flow generated by the field c(F). c M admits a non vanishing section s. V~ is the operator of covariant differentiation of a section along the field eon M.(c(F)) on C"'(U~~. If the bundle L over the domain U.. the equality w =dO.. However this operator acts not on the functions. can be interpreted here as the local expression for a connection on the trivialization of L over the domain Ua..
c(G)). then from (2.7) From (2.p). A Hermitian structure on L compatible with V exists if and only if the form n is real.p = U a.p satisfy the relations ca.1. and when is it possible to define a Hermitian structure on L which is compatible with the connection.p E Coo ( U a.1 w. = 1 on U a.pCp 1 c 1a. n Up the equality holds. The SouriauKostant formula (2. Thus the following theorem is true: Theorem 2. where Ca.3) one can derive the following local expression for the curvature form: (2.· Namely.py = U a. G} =hQ(c(F). if one writes ca.p + bp 1 + b 1a. (2.2. The answer to these questions is given by Theorem 2.10) . The question arises of which 2forms on M can serve as curvature forms for a connection on some line bundle Lover M.py there will hold the equality ba.4) it follows that (2. n UP n U 1 • Therefore.8) On the other hand. =const E 7L. condition 2) can be rewritten in our case in the form {F. The proof of the first assertion of the theorem follows from the relations which connect the transition functions of the bundle L with the forms (}a. the integral of the form Q over an arbitrary 2cycle in M is an integer). Geometric Quantization 151 the formula (2.5) gives a prequantization &(M.9) The transitiOn functions ca.p in the form then in the domain U a. if on the intersection U a. A form Q is the curvature form of ~orne line bundle L over M with connection V if and only if the cohomology class defined by the form Q is integral (that is to say. w) if and only if the curvature form n of the connection V coincides with h.
In the general case we have Theorem 2.12) Let us define a Hermitian structure on L by setting lis~~ II= (sll. we obtain the transition func tions of the desired bundle L.. L = M x C. we obtain yet another proof that the class of n is integral.2) follows from (2.11) If a Vinvariant Hermitian structure exists. sll//2 =exp P~~· (2. . can be chosen so that the functions bripEC 00 (U11p). let n define an integral cohomology class.152 A.4).f. let the form n be real. de Rham) the cohomology class of this cocycle coincides with the class defined by the form n. given (up to a constant summand) by the equations dbii{J = h l((J{J. Kirillov Thus we have constructed an integral Cech 2cocycle (E. (As the unique open set U11 one must take the whole manifold M = T* N and set 011 = 0.f.1 0). The set of prequantizations of a symplectic manifold (M.. which is defined up to a summand ofthe form d. Setting C11p = exp(2nib11p).A.11) can easily be generalized to arbitrary curves r = oD not necessarily lying in a U 11 • If we apply this formula to a closed 2cycle D. For simply connected manifolds the answer is positive. from which it follows that n is real..13) It is easily verified that this structure is compatible with the connection V: (2. Then the form (}II. Conversely.13) and (2.eC 00 (U 11 ). Souriau8. Now let r be a piecewise smooth closed path on the manifold M. P~~Pp=lnlcapl in U11 p· (2. Kostant) is uniquely determined by the symplectic manifold (M.1) arises as a special case of the construction described. w) for which the form n = h. bounding a twodimensional surface D c U 11 • From (2. (Let us note that formula (2. Cech) of the manifold M with the covering U11 • By the de Rham theorem (G.(}II).) The natural question also arises whether the SouriauKostant prequantiz ation (J..) Conversely. w).4) it follows that under parallel translation along r the section S11 is multiplied by the numeric coefficient (2. then this coefficient must be 1 in modulus. It is recommended to the reader that he retrace how the prequantization (2. Then the forms Im 011 are closed and one can find real functions Pain C 00 (UII) such that 21t dp 11 =hlm011. satisfy condition (2. S11 = l.1 w defines an integral cohomology class is a principal homogeneous space for the character group n* of the fundamental group n of the manifold M.3.M.
For a twodimensional manifold a symplectic structure is just simply an area element. In these coordinates the form QN takes on the form: n _!!. We shall construct explicitly a bundle LN over S 2 with a connection V.( w_dw_ . and a Hermitian structure <·.dw_ nc. for which the form QN = h. Moser [32].. 0.2ni () {} . Q. The area of the sphere will then be equal to Nh. . 2. ·).1 wN will be the curvature form. Examples. +)N 1+1w1 2 1+1w+l 2 N~. In the neighbourhood U ±let us introduce a complex coordinate x±iy W±=·.2m 1+1w±l 2 Then for the transition function c on U + n U _ we obtain the equation di . the area is the unique invariant of a compact twodimensional symplectic manifold (if its topological genus is given). [r] is its class in the group n._ dw± 1\ dw± N2ni(l+lw±f) 2 • As the 1form () ± on the domains U ± let us take ()+ = N~ w±dw± . V 2 ) one can construct a character x12 Ell* given by the formula (2. V 1 ) and (L 2 .(r) is the coefficient by which a section of the bundle L. For this Jet us cover the sphere by two coordinate neighbourhoods U ± = S 2 \ { P ±}. We shall analyze the two simplest examples here: the sphere S 2 and the torus T 2 =S 1 xS 1 • On the standard sphere x 2 + y 2 + z 2 = 1 in IR 3 let us introduce the surface element wN=(Nhj4n)(xdy 1\ dz+ ydz 1\ dx+zdx 1\ dy).w+dw+ ). when measured in units of h. which go over into each other under a suitable diffeomorphism of L 1 onto L 2 • From the two prequantizations corresponding to (L 1 . . The condition for the existence of a prequantization reduces to the condition that the area be an integer. with a connection V. ± 1). Prequantization of the TwoDimensional Sphere and the Two Dimensional Torus. According to a wellknown result of J.h (. Geometric Quantization 153 Here two prequantizations are considered to be the same if they are defined with the aid of equivalent line bundles L.14) where r is a closed path on M. 2.11) and the fact that the curvature forms for V 1 and V 2 coincide guarantee that the righthand side in (2. is multiplied upon parallel translation along the path r. Formula (2.where P + =(0.3. 1+z On the intersection U + n U _ the relation w +w _ = 1 holds. i = 1.14) depends only on the class [r] of the path r.
(2. Kirillov from which we have c=(w _)N =(w +)N. m. b) one can define a mapping l/Ja. m. y+n)=F(x. It is easy to check that l/J:. the bundle LN is the Nth tensor power of the Hopf bundle (H. Under the isomorphism with 9'(R) described above they go over into d/dx and 2nix respectively. y)dy..17) kel The prequantization operator corresponding to a doubly periodic function H(x.. y)= L l/J(x+k)e2•iky. Now let us consider the twodimensional torus T 2 = R 2 /7l.f(x..A.)dy.154 A. y + n) = e2"iNm'f(x. For each pair ofreal numbers (a.b so that l/J:.b(x. Let us note that the operators ojax and ojoy. when N = 1 the space of all smooth sections of LN admits an isomorphism onto the Schwartz space (L. Namely. Schwartz) 9'(R).Oct. Thus.b(Oa. As the area form on T 2 let us take wN = Nhdx 1\ dy.15) into itself.) = Nh(c. y+b)mod 71.b(wN)= Nhdx 1\ dy. n E 71. Hopf)..b) = Nhxdy. to each functionf(x. Thus. y) having the property (2. y) can be recovered from l/J(x): f(x.b of the open unit square K c R2 into the torus: lPa. Therefore the sections of the bundle LN can be identified with the functions on the plane having the property f(x + m. y). (2. (2. Let us choose the 1form Ba. 2 • Then l/J:.d.2nix take the space offunctions satisfying condition (2. nE 71.15) The bundles LN have many interesting properties.b(O. . 2 • We shall identify the space of functions on the torus with the space of doubly periodic functions on the plane: F(x+m.d.16) 0 Conversely.aH ~) H=H+.c.. y)=(x+a. (2.18) 2ni ax oy ay ax .. y) has the form: 1 (fJH(~2nix). y).15) one may associate a function l/J(x) on the line: 1 l/J(x)= Jf(x.
. A detailed exposition of this construction is given in [28]. and special odd coordinates ~ 1 .2 are valid. Let (M. {G. Here we shall indicate only the principal changes which must be brought in. xn. Prequantization of Symplectic Supermanifolds. Geometric Quantization 155 2.1 can be retained entirely if one merely replaces the commutator on the righthand side by the supercommutator [A. . The Poisson bracket looks as follows: {F. . {F. called even coordinates.1 and 2. (2.19) In suitable local coordinates the symplectic form w reduces to the canonical form n m w= I k= I dpk 1\ dqk + I j= I sj(d~j) 2 . H} +( l)a<Fla<Gl{G. § 3. . w) be a symplectic manifold. 2. ~j are the odd coordinates.G}= I [OF oG _oF oG]+(Jt<F) k= 1 opk oqk oqk opk s/F oG_ j= 1 o~j o~j f (2_22 ) The Poisson bracket operation defines on c~(M) a Lie superalgebra structure. H}} = {{F. It has the properties { F.4. The analogues of theorems 2. pia y a part. The Hamiltonian vector field corresponding to a function F has the form c(F)= I [oF~.1. ~m• which take values in some Grassmann algebra and which satisfy the anticommutation relations (2. As local coordinates on a supermanifold M both ordinary numerical co ordinates x 1 . . and t:j = ± I.21) where a( F) is the parity of the function F. The Definition of a Polarization. B] = AB ( l)a(A)a(Bl BA. The KostantSouriau prequantization scheme described here can be carried over mutatis mutandis to symplectic supermanifolds M (see [7].20) where Pk• qk are the even coordinates. [8]) with an even form w. F}.( J)a(F)a(G) {G. (2. G} = . o~j o~j (2. . (2. A subbundle P c rc M is called a polarization if it fulfills the conditions . H}}. rc M the complexification of the tangent bundle over M.oF ~]+(l)a(F) k= 1 opk oqk oqk opk f j= 1 t:j oF~.23) {F. . Polarizations 3. The definition of the quantization and prequantization of sect. . G}.24) The tie between !forms and connections on onedimensional bundles is also preserved.
in the neighbourhood of an arbitrary point of M one can introduce local coordinates x 1 . . according to the wellknown Frobenius criterium.. In the general case let us denote by E(x) and D(x) the subspaces in Tx(M) for which £C(x) = P(x) + P(x). Each leaf is a Lagrangian submanifold of M equipped with a canonical affine structure [2]. b) Let us call a differential form eon M ?admissible if the form O(x) vanishes when one of its arguments belongs to P(x).1). the polarization is called real. Y) (3..A. in such a way = = that their sum is always equal to 2n. vk so that D(x) is generated by the fields ofoxj and P(x) by the fields ojoxj and Ofozm. The NirenbergNew lander theorem [33]. If dim E(x) n + k..1) is a nondegenerate Hermitian form on P(x). xnk. dim D(x) n. . the local description of the polarization is not known at present.. This question is closely tied up with the problem of classifying real submanifolds of a complex manifold up to holo morphic changes of variables. Usually one requires in addition that this foliation be a smooth fibration (compare sect. The manifold M admits a foliation of half its dimension such that the space pR(x) coincides with the tangent space of the leaf passing through the point x. .. .k. Yn k. • .2) The dimensions of E(x) and D(x) can change from point to point. In the case where Eisa nonintegrable distribution. 'ofoin. u1.. .. . the polarization Pis called pseudoKdhlerian. Obviously. 2) The distribution x~P(x) is integrable. ••• . If P = P. then the complex conjugate subbundle Pis also a polarization. Dc(x) = P(x) n P(x). (3. . The polarization is called Kiihlerian if this form is positive definite. Where Zm = Um + ivm. It is clear that if the subbundle Pis a polarization. The distribution Pis integrable if and only if the differential of an arbitrary ?admissible form is ?admissible.. In this case P is the complexification of some real integrable sub bundle pn c TM. and if E is an integrable bundle (let us note that Dis automatically integrable). can be stated in two equivalent forms: a) Let us call a vector field ~ on M ?admissible if ~(x)EP(x) for all xEM. zn in the neighbourhood of an arbitrary point of M so that P is generated by the fields ofoi. In this case b(X. v1. The distribution P is integrable if and only if the commutator of any two ?admissible fields is ?admissible. Y)=iw(X.156 A.. . The latter condition. If P(x)nP(x)= {0} for all xEM. . In this case. [ 17] states that for an arbitrary pseudoKiihler polarization P one can introduce local complex coordinates z 1 . 1. y1. the ?admissible forms form an ideal. Kirillov I) The space P(x) is a Lagrangian (=maximal isotropic) subspace of THM) for each x EM. uk. then the polarization P is called admissible. .
which we denote by c(X. The Hamiltonian vector fields with generating functions in §"(P) commute pairwise.!r} can differ from the function . t. then P 1 and P2 are transverse. are tangent to the leaves and are constant with respect to the canonical affine structure on the leaves. The preimage of g(P) in &>(M. w) which preserve the polarization P.fx. 3. One may also assume that the mapping X 1+fx from g to C 00 (M) is linear. (3. Geometric Quantization 157 Polarizations P 1 and P 2 are called transverse if P 1 (x) n P 2 (x) = {0} for all xEM. For a real polar ization P the algebra §"(P) consists of the functions which are constant on the leaves of a certain Lagrangian foliation. Y) is a 2cocycle on the Lie algebra g.3) This cocycle defines a onedimensional central extension g of the Lie algebra g. Polarizations on Homogeneous Manifolds. For a pseudoKahler polarization P the algebra §"(P) consists of the holomorphic functions with respect to a certain complex structure. If P 1 and P2 are arbitrary Kahler polarizations. it admits a transitive action of a Lie group G which preserves the form w. Y).4) The mapping X$ tX I+ fx$a = fx + tX (3. Y].fix. Let g be the Lie algebra of the group G. w) coincides with the normalizer of the commutative subalgebra §"(P). Y). In the general case an explicit description of . and g* the dual space to g. X). c([X. (3. Y)=O.~(P) is for the present not known. w) is homogeneous. Let us note first of all that without loss of generality one may consider the group G to be simply connected. Let us suppose that the symplectic manifold (M. It is clear that §"(P) is an algebra with respect to the usual multiplication offunctions and a maximal abelian Lie subalgebra with respect to the Poisson bracket. Y)= c(Y. Y] Ee c(X. As a linear space g=gEe IR. For a real polarization P the algebra g( P) consists of the vector fields whose generating functions are affine on the leaves of the polarization P. that is. Y Ee PJ =[X. Let us denote by G(P) the group of automorphisms of (M. Passing if necessary to a covering M of the manifold M. For an arbitrary polarization P let us denote by §"(P) the space of functions on M which are annihilated by all Padmissible vector fields.2. it has the properties c(X. Z)+c([Y. The Lie algebra g(P) of this infinitedimensional group admits of a simple description. X)+c([Z. X]. that is. Let us briefly call to mind the classification of these manifolds which was formulated above in theorem t. we may assume that the field ~x possesses a generating function. To each element X Eg corresponds a Hamiltonian vector field ~x on M. but the multiplication law in g has the form: [X Ee a. It is easily verified that c(X. Z].5) . The Poisson bracket Ux. YJ only by a constant.
[ 46]).7) The momentum mapping is equivariant (commutes with the action of the group G) and is a local homeomorphism.158 A. Y) = b([X.eg* defined by the identity F. ro). c(X. and let G = Sp(2n. In this case G is a nontrivial central extension of the group G. G to M. called the Heisenberg group (W. Heisenberg). G allows one to consider the group action to be Poisson. B] = AwB. One can identify the space 9* with g with the aid of a nondegenerate bilinear form on g: The momentum mapping has the form Xt+W(X.. ·)+ 1 and takes M over into the orbit consisting of all affine functions equal to 1 at the point 0. polynomials of degree ~ 1) with the operation of the Poisson bracket..1. If the cocycle c is trivial. In this case the momentum mapping is defined. Let G be the simply connected Lie group with Lie algebra g..6) for some be 9*. then the algebra 9 is the direct sum of 9 and lit The group Gwill in this case be the direct product of G and IR. Kirillov is a homomorphism of the Lie algebra g into the Poisson algebra aP(M. that is. Example 3. (3.A.2.(X)=fx(x). IR) be the group of linear symplectic transformations. Its image is one of the Gorbits in g* with respect to the coadjoint representation.BwA. The Lie algebra 9 can be realized in the form of the space Sym(2n. where the second factor acts on M trivially. It acts transitively on M and this action is Poisson (see [3]. The space g* may be identified with g just as above. the transition from M. IR) of symmetric matrices with the bracket [A. Thus. Let M = IR 2 " \ {0} with the standard symplectic structure ro. The algebra 9 may be realized in the form of the space of affine functions on IR 2 " (that is. and let G = IR 2 " be the group of parallel translations on M. Let M = IR 2 " with the standard symplectic structure ro.. Y]) (3. but also as the space of quadratic forms on IR 2 " with the operation of the Poisson bracket. Example 3. The momentum mapping has the form: XI+ XX' . which takes a point xeM over into the functional F.
2. Y)= (F. [34].1) to be fulfilled. [36] and also volume 16 of this edition). Theorem 3. For a Ginvariant distribution P on M to be a polarization. [14]. B) is a maximal isotropic subspace of V with respect to B. A convenient way of constructing polarizations has been proposed by Michele Vergne (see the reference in [23]). c: Vn = V be a chain of linear spaces. and the form B has the shape B(X. Let us recall that a subalgebra a c: g is called subordinate to a functional FE g* if F vanishes on the derived algebra [a.1. Quantization 4. Y]) for some FE V*. and I) its Lie algebra.1. [15]. Examples are also known of orbits which do not admit invariant polar izations. Geometric Quantization 159 (where xis a column vector. n Let us set W(s. the space of a prequantization is too large for the completeness condition (condition 4 of sect. It is known [23] that for solvable Lie groups invariant polarizations exist for all orbits in the coadjoint representation. . This construction is based on the following simple assertion. a G invariant polarization P is determined by the space P(x 0 ) c: T~0 (M) for any point x 0 E M. it is necessary and sufficient that the space p be a subalgebra of gc subordinate to the functional FE g* which is the image of x 0 EM under the momentum mapping. Theorem 3. In particular. As was already noted above. then W(s. means transposition) and it takes IR 2 n over into the cone of nonnegative matrices of rank I in Sym(2n. b) If Vis a Lie algebra. the ~ are ideals in V. [12]. dim Vk = k. Let us suppose that on V a bilinear skewsymmetric form B is given and that Bk is the restriction of B to Vk. In ordinary quantum mechanics the wave functions depend on only half of the coordinates of the classical phase space: in the . . The Space of a Quantization.2 above). Let H be the isotropy group of the point x 0 . Let us denote by p the subspace of gc containing I)C and such that pji)C = P(x 0 ). Then: k =I a) W(s. the orbits of minimal dimension for the symplectic groups of rank ~ 2 are of this kind (see example 3. The space Tx 0 (M) can be identified with g/1) and T~0 (M) with gc/I)C. Let s: 0 = V0 c: V1 c: . B) is a subalgebra of V. IR). B)= L ker Bk. § 4. For a Ghomogeneous symplectic manifold M it is natural to ask about the existence of Ginvariant polarizations on M. Since G acts transitively. This is true also for generic orbits in an arbitrary complex Lie algebra. 2. [X. Recently interest in invariant polarizations on homogeneous manifolds has grown in connection with the socalled group approach in the theory of completely integrable Hamiltonian systems (see [1]. a] (see [20]).
have the property that V~s = 0 for any Padmissible field ~.A. Kirillov coordinate representation the wave functions do not depend on the momenta. There arc at least three obstructions to the construction of the quantization space out of the elements of r(L. In place of f(L.D. M."(L. This corroborates L. M. n E Z. rc p). I) vanishes on the leaf under consideration.160 A. CCp) leads to the same quantization space as is obtained by the method of generalized sections. The mapping f~+Q(f) defines a character of the fundamental group of the leaf (compare sect. and for the path r which goes once around the circle r=const we get Q(f)=exp(2n 2 r 2 /ih). Example4.ydx) as the tr form 0. CC p) = f(L.1) We shall denote the space of such sections by r(L. If the leaves are not . M. and let the polarization P have as its leaves the circles x 2 + y 2 = const. P) one may consider the set of generalized solutions of equation (4. BohrA. that is. P) and its higher cohomology groups Hk(M. Therefore in the general case it is natural to try to construct the space of the quantization out of those elements of the prequantization space (i. 2. If r can not be contracted to a point. (4.1. In the language of symplectic geometry one may say that the space of the quantization in both cases consists of the functions which are constant along the leaves of a certain real polarization. M. Sommerfeld) in M. P). Thus. Let M=~ 2 \{(0. They all are concen trated on M 0 and they form a space which we shall denote by f 8. M. the sections of the bundle Lover M) which are covariantly constant along some polarization P. It is clear that in this case any solution of equation (4. Thus.1). the number Q(f) is generally different from I. Another method consists in considering the sheaf CCP of germs of sections in f(L. In [39] it is shown that in the case of example 4.1 the transition from H 0 (M. in the momentum representation they do not depend on the coordinates. The second obstacle is that the scalar square of a section sE f(L. P) to H 1(M.e. the BohrSommerfeld subvariety consists in this case of the circles which bound an area nr 2 == nh. Let us denote by M 0 the union of those leaves for which this character is trivial. but with singularities". In polar coordinates r. P). P) is a function which is constant on the leaves of the polarization P. one way of circumventing the obstacle which arises lies in the passage from ordinary sections to generalized ones. ljJ the form(} has the form 2 d4J. P). M. The first obstacle arises in the case when the polarization is real and its leaves are not simply connected. Let us choose t(_xdy.2 above). Parallel translation of a section along a closed path r lying in a leaf results in multiplication of the section by a number Q( f).w=dx "dy. Faddeev's wellknown thesis: "cohomology means the same functions.0)}. M. The subset M 0 is called the BohrSommerfeld subvariety (N.
on which there is no measure a priori. In modern language the elements of this space may be described as halfdensities (or densities of weight 1/2) on M. V(c(Fl. P).2.5).1. 7) and theorem 2.W. is the polarization into which P goes over under the diffeomorphism cp. that is. P) over into r(L. then the integral of this function over the manifold M diverges and one must replace it by the integral over the set of leaves.=exp(tc(F)). where xis the character of the fundamental group of the leaf corresponding to the natural flat connection on the bundle of halfforms over this leaf. The way out of this difficulty is to construct a canonical isomorphism between the spaces r(L. we have Theorem 4. nEZ. The proof follows immediately from the commutation relation between F and V~: . sections of a bundle with a onedimensional fibre over M whose transition functions are square roots of the absolute values of Jacobian determinants. introduced by Blattner and Kostant. of the "intrinsic Hilbert space" L 2 (M).). it will be helpful to consider first the model situation of a flat space M with a constant polarization P.~l' [F.1 x(r) = .M. We shall consider only constant polarizations on M. M. (2. (4. 4. This will be done in sect. More precisely. M. It turns out this can also be achieved (at least in the flat case) by the introduction of Lvalued halfforms. M.P) for various P. Returning to the first obstacle. Finally. Let M = IR 2 n with the standard symplectic n form w = L dpk " dqk. (The origins of this concept may be seen in the construction. A way out of this difficulty is given by the concept.2. which could be applied to an arbitrary smooth manifold M. where P. h 2 m. The set of such polarizations forms the complex . This result (in contrast to the one obtained above) agrees with the wellknown structure of the spectrum of the energy operator for the harmonic oscillator.2) which in turn follows from (2. 4.! for which P(x) c C 2 n. Geometric Quantization 161 compact. In the case of example 4. of Lvalued halfforms on M normal to a given polarization P. let us observe that the transition from sections of L to Lvalued halfforms leads to a modification of the BohrSommerfeld conditions. =P k. in place of the condition Q(r) =I the condition Q(r) · x(r) = I must be fulfilled. If the field c(F) is complete. Therefore the modified BohrSommerfeld condition takes the form nr 2 =(n+!)h. Quantization of a Flat Space. Namely.) Before giving the exact definition of Lvalued halfforms and of the scalar product for them. proposed by G.1. P. a third obstacle is that the prequantization operators F generally do not preserve the space r(L. then the operator exp(tF) takes r(L. V~]=.I for a generating cycle r. Mackey about 30 years ago.
One can introduce suitable coordinates on the set A +(C 2 ") which map it onto a bounded region in C"1"+ 1" 2 • Namely. let zi= pi+ iqi. IR) the set A+(C 2 ") splits into n+ I orbits. zi as a block matrix of the form 1 (.. The group G = Sp(2n.js. Let us consider the subspace P. . A~(C 2 ) is the interior of the hemisphere and A~ (C 2 ) is the equator.n. c) The polarization P. Kirillov Lagrangian Grassmann manifold A(C 2 ").j 5. let us single out the subset A+(C 2 ") of positive polarizations P.4) In the t coordinates the action of this group on A+(C 2 ") has the form tt+(At +B}(Bt + A). IR) oflinear symplectic transformations can be written in the coordinates zi. 1 s. is positive if and only if the matrix 1 . For n =I A(C 2 ) is the Riemann sphere. the S\lbset A~ (C 2 ") is closed and consists of the real polarizations. AB'=BA'. (4.A.n. One can identify its sections with functions on M if one fixes a nonvanishing section s 0 . numbered by the rank of the form (3. One can show that A~ (C 2 ") is a skeleton of the boundary of A+(C 2 ")..1) on P: A+(C 2 ")= " A\ (C U 2 "). 1 with respect to the standard Hilbert structure on C"). Under the action of Sp(2n. 1) on the unit disk. In it. IR)~SU(1. AA*BB*= 1. b) The polarization P. is real if and only if the matrix tis unitary (an equivalent condition: t has n linearly independent real eigenvectors). for which the restriction of the form (3.t*t is nonnegative definite (an equivalent condition: lit II 5. (4. k=O The subset A"+(C 2") is open and consists ofthe Kahler polarizations. A+(C 2 ) is a closed hemisphere.• a I s.2.162 A. be complex coordinates on M and let t =II t Jk II be a complex matrix of order n x n. The prequantization bundle over M = IR 2 " is trivial. IR) of linear symplectic transformations of M acts on A(C 2 ") and preserves A+(C 2 "). ~). Let us construct the quantization space . The group Sp(2n.5) For n= 1 this action turns into the wellknown fractional linear action of the group Sp(2.. is a polarization if and only if the matrix tis symmetric.1• (4.TfP corresponding to a polarization PeA+(C 2 ").3) uzi k= 1uzk Theorem 4. By virtue 1 Here ' denotes transposition. of C 2 " generated by the vector fields ej=~+ a L" tjk. a) The subspace P. This is a compact complex mani fold of complex dimension tn(n + I).1) to P is nonnegative.
(4. .4) the choice of s0 up to a constant factor is equivalent to the choice of a 1form () giving the connection on L. M. M. by (2. . . . Gd=()ik• {Gi.8) 0 In the flat case we may take as Spa suitable quadratic form on M. P) which interests us. For the construction of the quantization space . IR). . In the notation of sect. It turns out this can be achieved if one defines the scalar product in . Fn are complex linear functions on M for which the fields c(F 1 ). Let us define the action function Sp on M by the condition dSp=O on P. The choice of the scalar product is dictated (up to a factor) by the following circumstance. • . zk) which are holomorphic in z 1 . P) can be identified with .fi'(P) consists of the functions f(x 1 . and equivalent to one another. irreducible.fi'(P) (that is. M.f1.see sect. 3. Xnk• z 1. G" are defined by the conditions {Fi.6) (This is the only !form on M having the property dO= w and invariant with respect to the linear symplectic group Sp(2n. f(L. Condition (4. Geometric Quantization 163 of (2. • . . The choice of this form becomes uniquely determined by the additional condition of linearity along P. .fi'(P). . Namely. . 3. 2ni ) ·s . a function which is annihilated by all Padmissible vector fields. With the aid of the action function it is easy to describe the space r(L. c(F") give a basis of P. ( (. Therefore () is a closed. With a fixed action function Sp the space r(L. .ff'(P)·exp ( hSp (4. . M. since dO= w vanishes on P.Yf P to be unitary.?(P).fi'(P).7) defines the function Sp up to a summand in . The prequantization operators F for FE . then (4. z) exp  4nlmSp) nk k k h d xd ud v. c/1(x. and if the linear functions G 1 .1 ).1 the space . 1 on M.7) This definition makes sense.Yfp a suitable scalar product remains to be introduced on . Clearly f!JJ 1(M) is a Lie subalgebra of. ••• . If F 1. Gd=O. . (4. P) and hence give a representation of the Heisenberg algebra on each .9) Let PEA\(C 2 "). zk. P)=.) Since 0 is a real form. and hence also an exact form on P.. (4.10) . z)/2 (x. Let us denote by :?J> 1 (M) the set of all real polynomials of degree ::. It is convenient to set J n 0=2kf:l (pkdqkqkdpk). It is called the Heisenberg algebra and is a nilpotent algebra with a onedimensional centre :!J> 0 (M) consisting of the constants. w).f2 )p=CpUl". .?(P) for P E A\(C 2") by the formula: JJ .13) the section s0 (times the proper constant factor) has unit norm at all points of M.9 1 (M) preserve each of the spaces r(L. It is natural to require the corresponding representations on . the Poisson algebra :?J>(M. • • .
having the property JAiP®JAiP'.nm. where sk =. we need Lemma 4. where siE r(L.10). cD 2 c E 2 c that is. [29].js. and Jt'p 2 for an arbitrary pair of polarizations P 1.12) where '. P) and<·. and Jt'p 2 has the form: (4..s0 er(L. Kirillov Let us note that the function under the integral sign can be written in the form (s 1 . s 2 ). 1 ~js. Then a) P 1 +P2 =E~ 2 for subspace E 12 c IR 2 ".. 1 ~l<:~m. To describe this pairing.A. s 2 ). i = 1. 12. The group Sp(2n. [30]). 3. Just as in sect. IR) acts in a natural way on 1\ "P.164 A. Furthermore. denotes isomorphism of Mp(2n. Df = Ei. see [16].·) is the Hermitian structure on the fibres of L. Let us investigate the "geometric meaning" of the expression cl'. (4.1. w) is a bundle with onedimensional fibre over A(C 2 "). M. and E 12 is generated by the fields ojoyi. we may choose particular coordinates on IR 2 " such that D 12 is generated by the fields ofoyi. Let us note that the function under the integral sign in (4.. l. together with an action on it of the metaplectic group Mp(2n. 1s. . denoted by JAiP. Let P 1 and P 2 be positive polarizations.l' 2 d"mxdmudmv. In these coordinates the pairing between Jt'p.11) When P1 = P2 = P this formula goes over into (4. the condition of equivalence of the representations defines (up to a factor) a pairing between Jt'p.!!. b) P 1 11 P2 = D~ 2 for some subspace D 12 c IR 2 ". P 2 eA+(C 2 "). 2.IR) (the connected double covering of Sp(2n. Let us denote by 1\ "P the vector bundle with onedimensional fibre over A(C 2 ") which is the nth exterior power of the "tautological" bundle over A(C 2 ") (the fibre of this bundle over a point P E A(C 2 ") is the space P c C 2" itself). A metaplectic structure on M =(IR 2 ". M. IR)spaces.nm.A"P. Pk).. IR). w)) diagram of inclusions cD 1 cE 1 c 2 Oc D 12 E 12 c IR ".11) has the form (s 1.!. and by ofouk and ojovk. c) there obtains a selfdual (with respect to passage to the orthogonal comp lement in (IR 2 ".h · exp [(2ni/h)Spk].
Namely. is a section of the bundle ~ of the form (4. (4. which we shall denote by JI(A. These sections are called Lvalued half forms on M with respect to P.1w in the isomorphism (4. 2 of the bundles~ and~ there corresponds a differential form of highest degree on M/D 12 .15). to speak of its being "constant along P". we see that NP1 ® Nf52 ~ (An+m£~2)2 ~ (A"+m(IR2"/Ddc)2.11) in place of the expression cP.13) (With the aid of w we identify the 1form dFi with the vector field c(Fi).16) Thus. where s E f(L. the form w restricted to E 12 has D12 as its kernel and defines a symplectic structure on E 12 /D 12 . From the exact sequence of linear spaces 0+ P 1 n P2 + P 1 ~P2 + P 1 +P2 + 0 II II DC12 it follows that there is an isomorphism A"P 1 ® Nf5 2 ~ A"mD~ 2 ® N+mE~ 2 .13). the quantization space . It is convenient to write them as s ® ). where FiE~(P). for FE ~(P) the field c(F) will be Padmissible. From this it follows that there is an isomor phism (4. just as for s. For a section~ it makes sense.P 2 dnm xdmudmv. Geometric Quantization 165 The sections of the bundle ~ may be written formally in the form JS.14) Furthermore. 1 ® l 2 ). There is a natural isomorphism ~ ® ~ ~ A"+m(IR2"/Ddc. Now let P 1 and P 2 be two positive polarizations. (4. M. ). (4. if one takes a system of units in which the Planck constant h = 1. where sis a section of A"P.13). the introduction of a metaplectic structure on M is equivalent to a consistent choice of the sign of the square root in the expressions (4.. Lemma 4. It turns out one can "take the square root" of the latter equality.) In other words.15) Combining (4. but the formil=h. An example of such a section is given by A. (In the general case.15). one must use not the form w. to a pair of sections )..YI'p consists of the sections of the bundle L ® ~ which are constant along P.1t turns out that just this form must appear in (4. P) and ). 1 .=JdF 1 " ••• "dF.14) and (4.. Namely.2. .) Now we can sum up and define the quantization space in purely geometric terms.
. Example 4.dz. ak >: (4. This fact to~ether with the property F = F* allows one to compute Ffor all FE&' 1 (M) in the case of a Kahler polarization. The quantization of a flat space constructed in this section gives simple explicit formulas for the quantization operators Fin the case where F is a polynomial of degree no greater than two.2). is a Kahler polarization.1. Let us note that the space &' 2 (M) of such polynomials forms a Lie subalgebra of &'(M. • •• . a" are called creation operators. (4. fJjfJz = l/2(fJjfJp.1).17) The expression under the integral sign in (4. is given by a complex number • with the condition 1•1:::. Its generalization to the nonlinear situation will be considered in sect. We have: w={i/2)dz.. It is generated by the field e = fJjfJz +•o/fJz. S. and the adjoint operators af...4.* • ] . If 1•1 < 1 then P. KostantS.3).=(z't'z)/)l1•1 2 . .z. w) in which&' dM) is a nilpotent ideal. The quotient algebra&' 2 (M )/&' 1 (M) is isomorphic to the Lie algebra of the symplectic group Sp(2n. wherefis a holomorphic function.dz.lz. BlattnerB. if FE~ (P). .18) 2 In the quantization space there exists a unique vector (up to a factor) which is annihilated by all annihilation operators. .17) has been given the name BlattnerKostantSternberg kernel (R. Adz. The quantization operators look the simplest for FE~(P).. Sternberg). Let P +be a fixed Kahler polarization and a 1 . e (n/2h)lz. The corresponding operators generate the Weil representation (A. consists of the expressions of the form 1/1 = f(z. The operators a1 .0. ··] a. a: are called annihilation operators. 2 has the form: (s1®A.).* [ a..1 ®I2). 2 = J M/D12 (s1. where z = p + iq. and £'.hn u.A.. A constant positive polarization P. 4.s2®A.l 2 .. ak . where M = ~ 2 is the ordinary plane with coordinates p.k. ~). [.166 A.. ••• .. . q and symplectic form w = dp A dq.. a" an orthonormal basis of P + with respect to the form (3. lJ=(i/4)(z. Namely. is generated by the field fJjfJz. a. 1. Weil) (see sect. then F coincides with the operator of multiplication by F. 4. Let us introduce a complex coordinate z.s2)Jl(A. Then P. They are tied together by the commutation relations ••ak ] . Kirillov The pairing between £'.=(i/4)1z. So® fo.2. It is called the vacuum vector. Henceforth we shall identify 1/J withfand £'.[. .ifJ/fJq). We shall cite explicit quantization formulas for all constant polarizations in the case n = l. . •.. The quantization space £'.).
t/Jz)P..with the space L 2(1R. then this pairing gives the wellknown Bargmann isomorphism (V. 2 The vacuum vector is proportional to exp (.. If on the other hand Irl = I. + Jt<'p. In this realization the coordinate and momentum operators have the form: q= h.. I +f (l + r)h d . then the polarization P. . t/Jz)p. The pairing (4.e"(P'+q'ldpdq).ap«. is real.q«dp«). established .P. A p= ~+ 2j1 lrl 2 2niJI lrl dz. Geometric Quantization 167 with the space of holomorphic functions with the norm 11!/12 = [lf(z.3. m q« m q« The vacuum vector is proportional to exp( nq. fJ = t(p«dq«. 2j1 lrl 2 2niJI lrl 2 dz..19) These operators have the properties (4.tp«q«.p sin a:+ q cos a:.20) where c(P 1 .q« SIO !X + 2 h a ~ COS !X a. P 2 .by the pairing (4. 4. (4. The coordinate and momentum operators have the form: A I.P. and £. let us introduce the coordinates p« = p cos a:+ qsin a: and q« = . v = Im z. is generated by the field a.We(x/h)lz. a A f' • = . We have: w = dp« "dq«.r)h d q= ~. The reader will find it a useful exercise to check that the isomorphism between £. where u = Re z.q. The quantization space £. q« COS !X + 2 . Henceforth t/J will be identified withfand £. = . S.17) turns into the ordinary Fourier transforminthecaselr 1 1= lr 2 1 = l. SIO iX a.So ® Jdi.rz. Then P. P./h).t/11. dq«). = (UP. The Connection with the Maslov Index and with the Weil Representation.17) generates a set of unitary operators UP.r 1 = r 2 . P 3 ) is a complex number equal to 1 in absolute value.f (I ... Let r = e 2 i«...: Jt<'p..1fontheotherhandr 1 =land r 2 = 0./2h). If the . Bargmann) between L 2 (1R.dq) and the Fock space L~01 (C. for which (t/11. consists of the expressions of the form t/1 = f(q«)e(xi/h)p.l'du ~ dv.
p 2 for real polarizations are closely connected with the Weil representation of the metaplectic group Mp(2n.3. One may regard the Weil representation as a projective representation of the symplectic group Sp(2n. The corresponding 2cocycle on Sp(2n. ~)there corresponds an automorphism of the bundle L ® Jf\"P. Kashiwara). Kirillov = polarizations P 1 . (4. ~) is connected with the Maslov index by the equality ([29].22) Let A(~ 2 ") = A~ (C 2 ") be the real Lagrangian Grassmann manifold ( = the set of real polarizations of ~ 2 "). then c(P 1 . If one lifts the LerayKashiwara index m(P 1 . Theorem 4. Pacts on the space . if these polarizations are real.P.A. the metaplectic representation). ShaleA.1 o U9cPJ. (4.(~ 2 ") we shall denote the universal covering. P 2 . The correspondence g ~+ Wp(g) is a unitary representation of the group Mp(2n. The composition Wp(g) = T(g. [29].P. P 3 ).Yt'p+ Yf'9 cPJ (compare theorem 4. P 2 . x 1 ). The operators UP. P3 ) + Jl(P 3 . n ~ 2 ") given by the formula Q(x 1 Eex 2 Eex 3 ) = w(x 1 . then.Yt'p. P 2 . P 2 . Precisely this function was originally introduced by V. By A. x 2 )+w(x 2 .. [30]). It corresponds to an orbit of the minimal dimension 2n in the coadjoint representation of the symplectic group Sp(2n. P 3 ) to A 00 (~ 2 ") then it admits a decomposition m(P 1 . the spinor representation. P 2 .. x 3 )+w(x 3 . P . as is shown in [29] and [30]. P3 ) = J.. P 3 ) is a skewsymmetric integral function. P 2 ) is a skewsymmetric function of the pair of points on Aoo (~ 2 "). P): . ~)(see [16]. P 2 . [30] ): (4. ~).24) . The equivalence class of the representation Wp does not depend on the choice of PEA~(C 2 "). (4. This index is defined as the signature of the quadratic form Q on P~ EB P~ EB P~ (where Pr = P.21) where m(P 1 . Maslov in [31] as the index of a path leading from P 1 to P 2 on A (~ 2 " ). and hence a unitary operator T(g.168 A. Maslov) in the LerayKashiwara form (J.Yt'p. ni c(P 1 . P2 ) + Jl(P 2 . Namely. ~)on . called the Maslov index (V. P 3 ) = exp4m(P 1 . P 3 ) l. LerayM. the ShaleWeil represent ation (D. On the other hand. taking P over to g(P).). P).23) where Jl(P 1 . Weil).L(P 1 . P 2 . ~). This representation is called the Weil representation (other names are: the oscillator representation. P 3 are Kahlerian. to each element gEMp(2n.1). It is well known that the fundamental group of A(~ 2 ") is isomorphic to Z and therefore for each natural number q a qfold covering Aq(~ 2 ") is defined.
2).25) holds for all S.. w). The starting material for quantization consists of l) a symplectic manifold (M. Geometric Quantization 169 The group Sp(2n.. but also on its double covering A 2 (1R 2 ") (the elements of A 2 (IR 2 ") are the oriented Lagrangian subs paces in IR 2 "). One can check that the Maslov index J. This is equivalent to a lifting of the structure group of the bundle TM from Sp(2n. wj) = bij. P2 ) mod 2q is well defined on Aq(IR 2 "). and Yfp 2 in the "curvilinear" case runs into the following obstruction. be two positive polarizations on the symplectic manifold (M. A metaplectic structure on a curvilinear manifold M consists of a consistent assignment of metaplectic structures on the tangent spaces TAM) (see sect. ). IR) to Mp(2n. ••• . 4.26) i= 1 where k=dimP 1 nP2 and v1 .l(P 1 .Yfp out of the sections of the bundle L ® JI\"P which are covariantly constant along P can be carried out just as in the flat case (see sect. IR) acts not only on the real Lagrangian Grassmann manifold A(IR 2 ").) and e(x)ED 12 (x) if and only if the covector field XP. w) and let L be a prequantization bundle with a connection V.. w. 2) a prequantization bundle L over M with a connection V and a Hermitian structure on the fibres. (4.Er(L. The equality (4.2) for all x EM. Theorem 4. it is necessary and sufficient that the first Chern class of the bundle TM (more precisely. vJ = w(w. then the set of metaplectic structures on M is a principal homogeneous space for H 1 ( M.vk. wj) = 0.P2 = L w([v.23) on the group Mp(2n. £').Yfp.. Theorem 4. Therefore the cocycle (4.wk are (P 1 +P2 )admissible fields normalized by the conditions w(v. The General Scheme of Geometric Quantization. 3) an admissible positive polarization P on M.P.. The first three items were discussed in §§13 respectively. w(v. acting on A 4 (1R 2 "). discovered by Blattner.24) becomes trivial by virtue of (4. 4. IR) is isomorphic to U(n)) be divisible by 2 in H 2 (M. The attempt to define the pairing between the spaces .]. and finally 4) a metaplectic structure on M.5 ([9]). . p 2 E T* M /(P 1 + P2 ) defined by the following formula vanishes: k XP. s 2 ) need not be constant along D 12 .M.. £' 2 ). In order that there exist a metaplectic structure on M. 4.4.. of the principal U (n)bundle associated with it. If all the structures enumerated are on hand. IR). Let P. recall that a maximal compact subgroup ofSp(2n. The value of the Hermitian form <s 1 . IR).If this condition is fulfilled. w 1 . the construction of the quantization space .4 ([26]). ••• .
P.A. the mapping (l/J.. In the case when XP.(x) = (l/J.1 (x)). p 2 =1. )* P(l/J. takes the polarization P over into a polarization P. this is true for all constant polarizations)... The investigation of the possibilities which arise here presents itself as being of extraordinary interest. there arises a family of operators ~ l/J. just as in the fiat case. M. M.A. Kirillov The simplest example of two polarizations for which XP. 4.P. since A. p 2 = 0.30) where sEr(L. 4.. is a covariantly constant section of .j7\"P with respect to the affine structure on P.1 ). the operator F reduces to multiplication by F. P..=exp(2ni/h)tF of transformations of r(L.)..E.: .. If XP. generates a oneparameter group l/J. then the desired operator F can be defined on the space Jl'p as..170 A. In particular. s®A. and Lc(Fl.TI'p.. p 2 vanishes if P 1 + P2 is an integrable distribution (in particular. 3. = exp tc(F) of transfor mations of M. to define the pairing (4.· General theorems which guarantee that this operator is unitary are still not known at present.0 can be construc ted for M = IR 4 . if the function F is constant along P. then it is possible. M ). If c(F) is complete. c(F) E g(P) in the terminology of sect.29) By theorem 4. If a Hamiltonian field c(F) preserves some positive polarization P (that is. (4.). M. P). then the field c(F) is Padmissible.. In addition. i.)* takes sections of the bundle . P) over into r(L.17) and the BlattnerKostantSternberg integral operator UP. The Quantization Operators.28) where by F is denoted the prequantization operator (2. a} {alapl PI= +a2ap2 ' (4.5.j7\"P constant along P over into sections constant along P. that is. when P is an admissible positive polarization) the operators l/J. w = dp 1 A dq 1 + dp 2 A dq 2 : a . The spectrum of the operator f is defined in this case by the modified BohrSommerfeld variety (see sect.~+exp (2ni htF) s®(l/J. then one can define a corresponding group Ci>. = 0. are unitary with respect to the Hilbertspace . (4. = 0~ (for example.: P.1 the mapping exp(tF) takes r(L.: Jl'p+. Now suppose the field c(F) does not preserve the polarization P. The transformation l/J. follows (4. Thus.27) P2 {a(p~a! 1 +P2a!2)+b(P2a! p~a! 2 ) }· = 1 Let us note that XP.ji\"P.1).5).
In the nonhomogeneous situation general theorems are lacking for the time being. 2).)ir=o· (4. Also popular is the quantum Kepler problem. Woodhouse [47].17). N is a Riemannian manifold.. besides.32) is obtained.31) This is the most general definition being used at present in the theory of geometric quantization. defined by the pairing (4. The case of the harmonic oscillator has been analyzed in many works (see. the operator UP. where A denotes the "twisted" LaplaceBeltrami operator. where Q is the quadric in ~IJl>(n + 1) associated with the cone of isotropic vectors in ~" + 1• 1• The various affine forms of this quadric correspond to the three different cases of the Kepler problem (see [35]). . [16]. but a number of concrete examples have already been studied. [10]. Thus in [ 40] the case is analyzed of M = T* N where. where A is the LaplaceBeltrami operator on Nand R is the scalar curvature of the manifold N. In the homogeneous situation it turns out to be adequate in order to construct a complete set of irreducible unitary represen tations for a broad class of Lie groups (see [4]. [11].Pis defined and unitary. [38]).Pa</J. . which describes the motion of a particle in a Coulomb field ([35]. For the quantum energy operator the expression fi =A+ V(q)+!R (4. [38]. in which the usual covariant derivatives \\are replaced by "long" derivatives Vk + (2nie/h)Ak which include the electromagnetic potential A. The connection of geometric quantization with the twistor approach of Penrose is discussed in the book of N. for example.32). w = dp " dq. [41]) and the standard result on the structure of the energy levels has been obtained. The most natural realization of the phase space of this system is an orbit of minimal dimension (equal to 2n) in the coadjoint representation of the group G = SO(n + 1. If. The quantum energy operator is given by the previous formula (4. h d  F = 2ni dt(UP. In a number of works thendimensional generalization of this problem is considered. then one can define the quantization operator by the equality . Geometric Quantization 171 structures on Jt'p and Jt'p. The manifold M in this case is a "twisted" cotangent bundle over Minkowski space: where F is the electromagnetic field tensor on N and n is the projection of T* N onto N. As a symplectic Gspace this orbit is isomorphic to T*Q. The example of a relativistic system representing a charged particle in an external electromagnetic field is analyzed in the same work. [23]). H = p 2 + V(q).
(See e. In the case of an infinite number of degrees of freedom the method of geometric quantization cannot be applied literally and demands essentially new ideas. All these formulas are not convenient when the weight 11. introduced by J. the geometric quantization method suggests that there should be a transparent geometric interpretation of the multiplicity in terms of the symplectic volume of some "reduced manifold" (cf. which has proved very useful in the theory of representations of infinitedimensional groups. The main technical tool here is the localization formula proved in [BV].E. E.A. The explicit conjecture was formulated by V.M.in the irre ducible module with the highest weight A.g. It asserts that the geomet ric quantization procedure commutes with the procedure of symplectic reduction. Souriau. A geometrical approach to thermodynamical problems is being advocated by one of the founders of geometric quantization. is the examination by G.I. he suggests axiomatizing the class of smooth mappings from M to the space which interests us [44]. of a diffeology.(/L) of the weight 11. which does not exist in many concrete examples (because of the presence of singularities. The proof of the conjecture was obtained by several authors under slightly differ ent conditions (See [G]. [DH]). the multiplicity of the zero weight is not easy to get from them. [KT]. Souriau [ 43]. J. Another idea. Quantization and Reduction. Ol'shanskij of the semigroup of contraction operators lying in the closure of the image of the group under a unitary irreducible representation of it. [H]. nonseparability. or infinite dimension). One such idea is the concept. [K3]). [JK]. [GS].is far from the boundary of the convex hull of extreme weights. [W]. In place of a smooth manifold structure. On the other hand.M. Sternberg [GS].172 A.g. Guillemin and S. Hurt [18]. This subject is related to the problem of com puting multiplicities of weights of a finitedimensional irreducible representation rr of a semisimple group G. [V]). [S]. Supplement to the Second Edition During the ten years passed after this paper was written the idea of geometric quantization was thoroughly investigated and widely exploited both in physics and in mathematics. I am not able to survey all this activity and restrict myself to some topics which have attracted the most attention. It is the classical part of the representation theory and several formulas are known for the multiplicity m~. 1. Kirillov An interesting direction which connects the ideas of geometric quantization with the Selberg trace formula has been under development recently by N.. It' is worth mentioning that in .
Berline and M. The corresponding moment map sends a connection to its curvature. This manifold has several remarkable realiza tions including the space of univalent functions on the unit disc. So. Guillemin. One of the most important examples is the manifold M = Diff+(S 1)/Rot(S 1).F. 2. Brylinski. Vol. [KY]). Bott. Path integral quantization of the coadjoint orbits of the Virasoro group and 2D gravity. [BY] N. Namely. :E a Riemann surface (without boundary but possibly with some marked points) and M(:E. This subject became rather popular in the last ten years. the classical object of complex analysis (See [K]. The equivariant index and Kiril/ov's character formula. Inventiones Mathematicae 69 (1982). The YangMills equations over Riemann surfaces. Let G be a compact lie group. that this example is related to the space of knots in M. Lond. J. In: Progress in Mathematics. [G] V. Remarkable exam ples of infinitedimensional homogeneous Kahler manifolds arise as coadjoint or bits for infinitedimensional Lie groups. Math. Alexeev and S. G) can be easily explained if one observes that this moduli space is a reduced manifold. The quantization of this space would give the geometric construction of the discrete series ofunirreps (=unitary irreducible repre sentations) of the so called VirasoroBott group. characteristic classes and geometric quantization. Shatashvili.. G) the moduli space of flat connections on Gbundles over :E without marked points. 523615. G) of all connections with the natural Poisson structure and take into account that the action of the gauge group GM preserves this structure. On the variation in the cohomology of the symplectic form of the reduced phase space. Phil. Duistermaat and G. Nuclear Phys. We refer to [AB] for the introduction to this beautiful theory. We refer to the book [B] for the details.J. On the physical level this was done in [AS] but the rigorous mathematical theory still does not exist. Vergne.L. Soc. Birkhiiuser (1994) 1150 . Loop spaces. Geometric Quantization 173 the framework of equivariant Ktheory one can define the quantization so that the statement is automatically true but the problem is when (and whether) this definition agrees with other possible ones. G). Poisson Structure on the Moduli Space of Flat Connections. 3. From the other examples we mention the coadjoint orbits for the group of the vol ume preserving diffeomorphisms of a 3dimensional manifold M. [DH] J. Birkhiiuser (1993) 1300. Progress in Mathematics 122. 8323 ( 1989). Heckman. 259268. the preimage of the zero point of the moment space consists of flat connections and the standard procedure of reduction gives us the space M(:E. Moment maps and combinatorial invariants of Hamiltonian Tnspaces. consider the space A(:E. 719733. A 308 (1983). Trans. Atiyah and R. It shows how infinitedimensional ob jects can be used to study finitedimensional ones. the classical object of topology. 11591190. The point is that the presence of the natural symplectic (or Poisson) structure on M (:E.J. References [AB] M. [AS] A. Amer. 107. It turns out. [B] J. 107 (1985). R. Quantization of Some InfiniteDimensional Manifolds.
Two dimensional gauge theories revisited.A. Fundamental'nye Napravleniya (Contemporary Problems of Mathe matics. [V] M. V. Kirillov [GS] V. F. pp. Mat. Berezin. Math. Geom. Localization for non abelian group actions. Ser. (English translation: Graduate Texts in Mathematics 60. Atiyah.22001 5. . 303368. Homogeneous quantization and multiplicities of group represen tations. Nauka. [Kl] A. 14. 386. Holomorphic slices. Moscow 1983..58030 4. [KY] A.V. Zbl. 187214 (1961).70001 3. symplectic reduction and multiplicities of representations. [JK] L. Acad. pp.). Mat.15020 * For the convenience of the reader. 9 (1992). 42. 284294. Kirillov.I. 1136). 1togi Nauki Tekh. 5139 (English translation: see present volume. 424 p. V. 1156. Ser. references to reviews in Zentralblatt fiir Mathematik (Zbl. [KT] Y. D. Arnol'd. F. SpringerVerlag. Probl. References* I. Kirillov. Napravleniya 22 (1988).). Sovrem. Commun. compiled using the MATH database.F. Ser.J. An nals of Math. Sjamaar. Phys.: Funct. Zbl.: A geometric construction of the discrete series for semisimple Lie groups.4 (1987). Akad. Introduction to the Theory of Representations and Noncommutative Harmonic Analysis.: Symplectic geometry. Karshon and S.. Inventiones Mathematicae 67 (1982). in the collection: ltogi Nauki Tekh. Anal. 107. 38. Moscow State University. Dordrecht BostonTokyo 1987. [W] E. Invent. 11161175 (1974) (English translation: Math. Projections orbits and asymptotic behavior of multiplicities for compact con nected Lie group.I. Guillemin and S. [K3] A. J. 141.).53049 7. Schmid. M. V.174 A. Heckman.A. (S] R. Prilozh. No. in: Representation Theory and Noncommutative Harmonic Analysis I. No. 393. Contemporary Mathematics 145 (1993). USSR. Springer. Paris 319 (1994). New York HeidelbergBerlin 1978. Pure Appl. 219248 (1979). 21. 465484. 344380. R. Appl.. as far as possible. The moment map and line bundles over presymplectic toric manifolds. J. J. M. Nauk SSSR. Sternberg. 462 p.2 (1987)...: Quantization. Vergne. Anal. Jeffrey and F. Moscow 1985. 11091165 (1974)). II. Zbl. Bargmann. 4245. Zbl. Anal. Kirillov and D. 208 p. 327 332. Givental'. [K2] A. Funct. Tolman. lzv. 592. 527.: Mathematical Methods of Classical Mechanics.. II: Infinitedimensional Lie groups and Lie algebras. Fundam. 312. Topology 34. Zbl. 431 p.2 (1995) 291327. 87129. Sci. I (1995). 50. Enlish trans!. Probl. 21. En cyclopaedia of Mathematical Sciences 22. Part I. Yuriev. Quantification geometrique et multiplicites. Appl. Reidel. 21. Mathematical Physics and Applied Mathematics 9.2 (1987). Kirwan. Funct. Kirillov. 163. English trans!. Moscow 1974. 47 (1982). Mat. Anal. Zbl. Math.A. 162 (1977). I: Geometric quantization. Kahler structures on Korbits of the group of diffeomorphisms of the circle. Kiihler geometry of the infinitedimensional space M = Diff(S 1)/ Rot(S 1 ). W. Berezin. Math.A. The orbit method.A. 8. No. Sovrem. Arnol'd. have. Diff.91 6. 122 125.: On a Hilbert space of analytic functions and an associated integral transform. C.A. 1994. Invent. 373. Adler. No. A.B. Witten. (English translation: Part I of: Introduction to Superanalysis. 333356. [H] G. Zbl. Ser. lzv.35058 2. been included in this bibliography. Funkts.: Introduction to Algebra and Analysis with Anticommuting Variables. VINITI. Fundamental Directions) 4. No.A. 5 162.: On a trace functional for formal pseudodifferential operators and the symplectic structure of the KortewegDevries type equations. Geom. 38 (1993).
Notes Phys. SpringerVerlag. Kirillov. Zap. Lect. Guillemin.53028 27. Nauchn. 149. Nat!. p. 529. Notes Math.S. 356 (1984) (English translation: Russ. A. 2.. Zbl. Moscow State University. 331. 1980. Zbl. Berezin. Lect. Tenzorn. 12.A. Tr. Mishchenko.: Symplectic spinors.: Lectures on Quantum Mechanics for Students of Mathemat ics (in Russian). 87208. Blattner. A.250 21. Otd. Berlin HeidelbergNew York 1970. Inst. Mat.V. Dokl. 167 (1984)). USA 17. Zbl. BerlinHeidelbergNew York 1977. SpringerVerlag. 139152. A. No. 138. 505. L.2. Zbl. D.530ll 17.57 26.4.A. pp.: Geometric Quantization in Action. 170. A. Zbl. di Alt. 505. Math. in: Lectures in Modern Analysis and Applications III. in: Differential Geometrical Methods in Mathematical Physics. N. 474 p. Fiz.17003 20. pp. Fomenko. (Contemporary Problems of Mathematics) 16. Kirillov.: Lectures on the Theory of Group Representations.53027 10. 343 p. VINITI. in the collection: ltogi Nauki Tekh. Sov. SpringerVerlag. L.: Liouville integrability of Hamiltonian systems on Lie algebras. 53104 (1962)). DordrechtBostonLondon 1983. 315 p. 364.A. Nauk 39.2. Zbl. Mekh. Proc.58005 9. Sov. Sov.T. 4151 (1970) (English translation: Mosc. Princeton 1966. A. Surv. F. in:. 129221 12.: The metalinear geometry of nonreal polarizations. Mat.. A. Semin. Surv. pp. 394 (1979) (English translation: Sel.. Fourth International Colloquium. Faddeev. Bull. Zbl. No. 17. D. Usp.T.. Moscow 1971 23. B. LondonNew York 1974. M.J.. Hormander. Naples 1982. Proceedings of the Symposium Held at the University of Bonn. 25 (1970). pp. Lejtes. 153213 (1982). 336 p.: Construction de representations unitaires d'un groupe de Lie.: Invariant operators on geometric quantities.A.A. Joseph. 121 (1982)). in: Group Theoretical Methods in Physics.58015 15. Zbl. 208 p. Roma. Math. Mosk. Nijmegen 1975. Fomenko. A. Kirillov. V. BerlinHeidelbergNew York 1976. (English translation (of the first edition): Grundlehren der mathematischen Wissenschaften 220.: Supermanifolds. Mathematics and Its Applications 8. Hurt.D. Zbl. Nauk SSSR 224. No. Usp. 12181222 (1975)).: Geometric Asymptotics. Yakubovskij. Symp.A. 800806 (1985)). No. Van Nostrand. Zbl. Univ. Leningrad 1980. Cours d'ete du CIME.35082 13. L. Notes Math. Acta Math. Nr.: Construction of unitary irreducible representations of Lie groups. 4. Zbl. 346. Mat. 18. 505508 (1975) (English translation: Sov.: Integrability of Euler equations on semisimple Lie algebras. B. Mat. Koopman.: Quantization and unitary representations. 16. 223. Math.22007 19. 28.53008 25. Leningrad State University. M. Sternberg. Istituto Naz.: Theorie de Mackey pour les groupes de Lie algebriques. Vestn..22011 22. Dufto. Kostant. Ser. Providence. A. O. Part 1: Prequantization. Group Represen tations and Mechanics (in Russian). Akad. July 14. 2.: Elements of the Theory of Representations (Second edition). A. 1977. Nauk 17. 329 (English translation: J. Semin.D. Probl. 50.: Unitary representations of nilpotent Lie groups. Zbl.. Geom. Math. 570. Nauka. Zbl. 1975.I. B. Geometric Quantization 175 8. Lect. 1145. V. Sovrem. Zbl. 95109.A.: Simple connection between the geometric and the Hamil tonian representations of integrable nonlinear equations. 264273 (1982) (English translation: J. Mathematical Surveys 14. Math. L. Part IV.A. American Mathematical Society. 321.22001 24. Moscow 1980. in: Harmonic Analysis and Group Representations. BerlinHeidelbergNew York 1976.58015 . Acad.62 18. 6976 (1972)).2. R. Kirillov. Kostant.: Hamiltonian systems and transformations in Hilbert space.. Prilozh. Univ. S. Kirillov.A. 374. Dufto. Faddeev.: The algebraic method in representation theory (enveloping algebras). 39. Reidel. 19. Academic Press.220ll 11. No. Anal. Zbl. Ser.58024 16. SpringerVerlag. Leningr.O. 342. pp. 14. I 25. 452. Math. 249. 200 p.). Zbl. Dokl.: An Introduction to Complex Analysis in Several Variables. Cortona. Trofimov. Liguori Editore. 453. 207291 (1982)). Math. Steklova 115. Sci. 14. 549. Moscow 1978. 57110 (1962) (English translation: Russ. 315318 (1931) Zbl. Takhtadzhyan. Math.. 106. Vektorn. Mat. R.
/7.. in: Differential Geometrical Methods in Mathematical Physics. Trans. Sniatycki. SemenovTyanShanskij. SpringerVerlag.. 483. 102 p.M.: Sur certaines representations unitaires d'un groupe infini de transformations. Lect. 6668 (1977)).M. Notes Math. 401408 (1976) 36. Moscow 1965 (French translation. 7576 (1917)(English translation: Funct. 99. Bull. Simms. BerlinHeidelbergNew York 1980. in: Differential Geometrical Methods in Mathematical Physics. Belg.M. 21.194 33. Am. July 1316. R. Phys. 303 p. Anal.). Kirillov 28. with two additional articles by V. J. Math.. D.. Zbl. in: Differential Geometrical Methods in Mathematical Physics. G.58013 47. Notes Pbys. Dokl.161 34.: Tbeorie des perturbations et methodes asymptotiques. N. 570. graded Lie theory.: Groupes differentiels. Weinstein. Zbl 14l. Lect.. 271 p. 1917. Segal.J.: Quantification geometrique. Sniatycki. Moscow State University. 230 p.176 A. Appl. L. Zbl. J. I. Zbl. 337 p. B. Nirenberg. Nauk SSSR 251.J.: On the volume elements on a manifold. Lect. Fourth International Colloquium. pp. Woodhouse. SpringerVerlag. Math. Souriau.: Lectures on Geometric Quantization. Zbl. 166 p. The MIT Press. 676.53019 40. Ser.58007 41. J. 113 (1981).. J.22005 31. 1975. Newlander.: Thermodynamique et geometric..: Graded manifolds.A. Leray. Zbl. Ol'shanetskij. /. Woodhouse. Proceedings.58010 45. 26. Phys. pp. 351. 186. Zbl. 79. Perelomov. 13101314 (1980) (English translation: Sov. Sci. Zbl. BostonBaselStuttgart 1980. Dokl. New Ser. Commun. University of Bonn. Arnol'd and V. Jl. Paris 1977. pp. pp. 286294 (1965). N. Zbl. Rejman. 6 (1951). 414 p. 358. 397. Anal.J. V. July 14. Lect.A. M. J. Boston 6. 444.M. BerlinHeidelbergNew York 1976.: Explicit solutions of some completely integrable Hamil tonian systems. 177306.387 46.M. 391404 (1957). A. Nijmegen 1975. Maslov. Proceedings of the Symposium Held at the University of Bonn. Van Hove. pp. 5. Moser. BerlinHeidelbergNew York 1976.: Symplectic geometry. Cambridge. Math. Proceedings of the Symposium Held at the University of Bonn. in: Differential Geometrical Methods in Mathematical Physics II. 65. A. Souriau. SpringerVerlag. No. September 1014. Prog. Zbl. Vergne. Exp. J. and prequantization. Am. 384 p. 364. New YorkHeidelbergBerlin 1980. Paris 1970. Zbl. Jl. No. Phys. Notes Math. Lect. 374398 (1966) 42. BerlinHeidelbergNew York 1978. 343. Acad. 1979. Math. Ann. SpringerVerlag. Maslov Index and Theta Series. no.: Structure des systemes dynamiques. Math.A. Akad. Mass..P.: Current algebras and nonlinear partial differ ential equations. SpringerVerlag. 120. Zbl. Maitrises de mathematiques. 468488 (1960). A. Applied Mathematical Sciences 30.35002 30.I. 45. /. Kostant. Zbl. 91128. Zbl.C. Zbl. Notes Phys. II. 1975. I. 429. Funkts. J. Soc. IRMA de Strasbourg 1978 (revised and translated into English: Lagrangian Analysis and Quantum Mechanics.34 (1980)). 4666.E. Birkhiiuser. 836.58001 35. Souriau. SpringerVerlag. J. Onofri. BerlinHeidelbergNew York 1977. 53. Lion. Souriau..: The Weil Representation. M. Paris 1972. SC:minaire sur les Equations aux Derivees Partielles (19761977). Collect. J. Prilozh. L.London 1981. 465.M. Zbl. 353.: Quantization of nonlinear systems. 501. Octavo.). 501. BerlinHeidelbergNew York 1917.: Geometric Quantization and Quantum Mechanics.58018 37. Cl. Math. Notes Math. 247. Math. 369397.: Twistor theory and geometric quantization. 6306.: Analyse lagrangienne et mecanique quantique. and Recherche cooperative sur programme CNRS No. July 14. Bouslaev [Buslaev]: Dunod GauthierVillars. Zbl. A.53023 39.M. J. I.53015 . I. Dunod. College de France.53048 44. 149163. Lect. in: Group Theoretical Methods in Physics. M. Math. Proceedings of the Conferences Held at AixenProvence..: Dynamical quantization of the Kepler manifold. Notes Math.: Complex analytic coordinates in almost complex manifolds. 25.224 38.47010 32. Zbl. 1979 and Salamanca. September 37.: On cohomology groups appearing in geometric quantization. 50. SpringerVerlag.53024 29. Soc.. Mem. 570.580 43.G. E.
265 §6. 210 §4. I. . . . . . Classical Methods of Integration 180 § 1. 180 §2. . . Dubrovin. . . . Novikov Translated from the Russian by G. . . . . . . I B. . . . . 216 §2. . . . . . . . . . . Commutational Representations ofEvolution Systems . . . . . . . . . . .M. . . . . . . . . . . . . . . . . ActionAngle Variables . Modem Ideas on the Integrability ofEvolution Systems . . . . . . . . . . . . . . . . . . 254 §5. AlgebraicGeometric Integrability of FiniteDimensional A. . . . . . . .A. . Liouville's Theorem. . Integrals and Reduction ofthe Order of Hamiltonian Systems. . . . . . Pole Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .P. Systems with Symmetry . . . . . . . . . . . . . . . . . . . . . . The Method of Separation of Variables. . Integrable Systems. The Most Important Examples of Systems Integrable by Two Dimensional Theta Functions . . . . . . . . . . . . . . . . . . . . The General Concept of the Poisson Bracket. . . 230 §3. . . . . . . 213 Chapter 2. 198 §3. . . . . . . . . . . . . . . . . . . .The Classical Method of Integration and of Finding ActionAngle Variables . . . . . . . . . . . . . . . . . 179 Chapter 1. . . . . . . .Families . . . . . .Families . . . . . 247 §4. . . . . . S. . Hamiltonian Systems. . . . . Krichever. . . . . . . . . . . . . . . . . The Hamilton Jacobi Equation. . . . . Integrable Systems and the AlgebraicGeometric Spectral Theory of Linear Periodic Operators . . Wassermann Contents Introduction . . . . . . . . . . . . . . . . . . 276 . . . . . . . . The Hamiltonian Theory ofHyperelliptic A. The Principal Examples . . . 216 §1. . . . . . . . . .
. . . . . . 295 Appendix C. . . . . . . . . . . . . . 319 References . Hamiltonian Systems and Spectral Theory . . . . . . . . . .A. . . . . . . . . . The NonLinear WKB Method . . . . . . . . . . . . . . AlgebraicGeometrical Integration of (2 + 1)Systems . . . . . . .P. . Dubrovin. . . . 289 Appendix A. . . . . . . Integrability of Systems of Hydrodynamic Type. . . . . . . . . . . . . . . . . . . . . . . . . 287 Appendices . . . . . . . Krichever. . . . . TwoDimensional SchrOdinger Operators and Integrable Systems . . 305 Appendix D.M. . . . . . . . I. . . . . . Spectral Theory of TwoDimensional Periodic Operators 311 List of Additional References . . . . . . . . . . . . . . . 289 Appendix B. S. . . . . . . . . . . . . . .178 B. . . . . . . . . . . . . . . . . . . . . . Novikov Historical Remarks Concerning Algebraic Geometry. . . . . . . 323 . . . . . .
The most striking and important of them are connected with the concept of solitons and their periodic analogues (which will be the topic of discJ. as became clear in the midsixties. the construction for them of variables of the actionangle type turn out to be essential during the following stages in the construction of a theory of perturbations and of diverse versions of the averaging methods. Just as in classical analytical mechanics. the sineGordon equation and many others. basically. ideas about the structure of nontrivial integrable systems did not exert any real influence on the development of physics. the theory of elementary particles. be ginning with the results of Poincare and Bruns at the end of the last century.lssion to a significant degree later on). The situation changed radically with the discovery of the inverse scattering method. have been perceived as something exotic. [59] it subsequently became clear that the equations to which the inverse scattering method is applicable are Hamiltonian and. this article is devoted (and among . in the construction of the quantum analogue of the inverse scattering method. The inverse scattering method allowed people for the first time to discover and to understand a number of principally new effects which had not become apparent in any way in the theory of perturbations. The very insignificant list of such examples practically did not change until the 1960's. The goal of the present survey is the presentation of the modern theory of integrable systems as a constituent part ofthe inverse scattering method. possess a remarkable universality property. They arise in the description (in the simplest approxi mation after the linear one) ofthe most diverse phenomena in plasma physics. special emphasis is laid on finitedimensional systems. The finitedimensional dynamical systems to which the inverse scattering method is applicable and to which. Although after the papers [54]. The concept of solitons has become one of the fundamental ones in contempor ary nonlinear physics. Integrable Systems l 179 Introduction Integrable systems which do not have an "obvious" group symmetry. The Hamiltonicity of these equations. These sections remain outside the scope of the present article. The evergrowing interest in this method is connected with the fact that it has proved to be applicable to a number of nonlinear equations of mathemat ical physics which. Although a number of fundamental methods of mathematical physics were based essentially on the perturbation theory analysis of the simplest integrable examples. the integration of these equations within the framework of the inverse scattering method does not make use of the Hamiltonian theory. the theory of superconductivity. what is more. in nonlinear optics and in a number of other problems which are reducible to spatially one dimensional ones. the nonlinear Schrodinger equation. Among the equations referred to are the Kortewegde Vries equation. are the field analogues of completely integrable Hamiltonian systems.
.A. Let yi. The second chapter is the nucleus of the present survey.P. Novikov them are contained all the known classical completely integrable systems) are finitedimensional in their original physical formulation or they arise during the construction of particular classes of exact solutions of the fieldtheoretic equations as restrictions of the latter to finitedimensional invariant sub manifolds. . Dubrovin.A. going back to the classical ones. .A. Perelomov. The Principal Examples From the modern point of view. the concept of the Poisson bracket (S. and also its numerous applications. of integrating Hamiltonian systems which have an explicit symmetry or which admit of separation of the variables. Ol'shanetskij. In the first chapter of the survey the modern views of the Hamiltonian formalism of both finitedimensional and fieldtheoretic systems are presented.D. are presented in the second chapter. I. in terms of special classes of functions. Classical Methods of Integration § 1. One should especially stress the significantly greater effectiveness of the inverse scattering method as compared with the classical methods of integrating Hamiltonian systems. A scheme based on the application of the methods of classical algebraic geometry has proven to be the most fruitful one in the theory of integrable finitedimensional systems. i = 1. in contrast to the ineffective integration procedure given by Liouville's theorem. Krichever. M. as well as canonical actionangle variables.M. SemenovTyanShanskij. the inverse scattering method allows one to explicitly produce solutions of the equations of motion. Also set forth are the methods. Chapter 1 Hamiltonian Systems. which is the starting point of all the integration schemes which are unified by the ideas of the inverse scattering method. and M. It needs to be noted that naturally abutting on the present survey there will be a survey "Integrable systems II" by A. This scheme. The General Concept of the Poisson Bracket. For completely integrable systems. which will be published in one of the following volumes of the present series. N be .. Its first chapter is devoted to grouptheoretic methods of integration of some special finitedimensional systems.180 B. The second chapter is devoted to geometric quantization of the open Toda lattice and its generalizations.M. In it the paramount concept of the commutation representation of evolution systems is introduced. Poisson) lies at the basis of the Hamiltonian formalism. S.
{J. Jl=const. uy' (l. of og J.9) en(Y)h (y)a:YJ. H} = e'n:. The commutator of two Hamiltonian fields is connected with the Poisson bracket by the relation (1. h}. The vector field en=(ek) corresponding to the Hamiltonian system (1.J.5) Let us note that (1. {g. h}} + {h.l indices is implied). (1.G. Integrable Systems I 181 local coordinates on a manifold Ythe phase space.. Leibniz) {fg.2) and skewsymmetry { g. g} = oyi oyi y'. called the Hamiltonian..3) b) the Leibniz identity (G.6) and the definition (l. (1. Such vector fields are called Hamiltonian.8) where H =H(y) is an arbitrary function. .y. f}} =0. The Poisson bracket of two functions f(y) and g(y) is given by a tensor field hii(y). { .{J. Jacobi) {J. yl (1. {h.ll) .10) It is clear that the derivative of an arbitrary function f = f(y) by means of the Hamiltonian system (1.4) c) the Jacobi identity (C..W. . N.8) has the form · .l) (here and further on the summation over repeate<. (1. A.7) Hamiltonian systems by definition have the form: (1.l) can be written in the form of og { . Here it is required that the following properties be fulfilled: a) bilinearity {A!+ Jlg. h} + Jl{g. h} = lc{J. (1. h}. h}=g{J. h}+f{g. H(y)}._ (1. 1l. g} = h'l(y) oyi oyi (l.'. g}.8) has the form: i _ ij oH(y) { i .} {J. of J = {J. g}} + {g.f} = . .
) (l.. Constant brackets and Lagrangian variational problems. Given tal'). The restriction of the tensor hqr to this surface is no longer degenerate.(y) have been found.B. Arnol'd and A. I. ahq.13) locally always exist. S. where the Hamiltonian is possibly defined locally [ 42] (see also the second chapter of the article by V.dyq A dy' has constant coefficients and is therefore .8).l2) (Transformations which preserve the Poisson bracket are called canonical. = o.P. 2. (l. d:~. 0 .5) it follows that the form n is closed. The Jacobi identity is automatically fulfilled in this case: on a plane where the matrix hq' becomes nondegenerate the corresponding 2form n = hq.· Let us examine the basic types of phase spaces.(O). Let zq be coordinates on the level surface (1..q + ah. Novikov The flow of ( 1. In this case the Poisson bracket is called degenerate: the matrix hii(y) is degenerate. yi(O) }. then the closedness condition (1. (l.I.e. yi(t)} = {y.M.[q. Any oneparameter group of canonical transformations for nondegenerate brackets det (hii) :J: 0 has the form ( 1.l4) the Poisson bracket no longer stays degenerate.15) The inverse matrix defines a 2form (l.182 B. It is possible that there are nontrivial functions..[q(y) of ( 1..g}=O (1.13) for any function g(y).5) [ 42]. . Krichever.) If all such quantities j. then on their common level surface j. (For a degenerate matrix hii(y) of constant rank the functions . and there is an inverse matrix (1. Thus phase spaces with a nondegenerate Poisson bracket are symplectic manifolds. Type I. Dubrovin.l6) From property (1.[q(y) (maybe given locally on the manifold) such that {..8) preserves the Poisson bracket: { y. + ah.l7) az' azP azq If the Poisson bracket was nondegenerate right from the start..14)..(t). t. .(y)=const (I= 1."= .17) turns out to be equivalent to the Jacobi identity (1.A. Let the matrix hii be constant and skewsymmetric.
( 1. where S= JL(x. In the nondegenerate case coordinates (y) = (x 1 . n (1.' These are the canonical Hamilton equations (W. p 1 . .M. (1. i = 1. n. p) are called canonical ones. By a linear transformation constant brackets can be reduced to the form 0 1 1 0 0 0 1 (1. Integrable Systems I 183 closed.. The equations (1. xn. . iJH i'=. i)dt.. Darboux)). Legendre) iJL Pi = iJii.R. . if one performs a Legendre transformation (A. They can be obtained from the variational principle bS = 0. Hamilton). i) is the Lagrangian.. . i = 1.8) take on the form: . ..18) 1 0 0 0 0 Locally any Poisson brackets of constant rank can be brought into this form (the Darboux theorem (G.22) .20) iJp. ••• .21) (x) belongs to the configuration space and L(x. Pn) may be introduced such that 0 0 0 1 ( 1.19) 0 0 0 1 The coordinates (x. • . .
p).1 k· x(JO ' i . j = 1. .28) Let us consider the set L of all linear functions on the phase space. not the coordinates of x as in ( 1.25) can be solved uniquely in the form x = x(q.f0 .24) can be brought into the form (1. but depends linearly 2 on the coordinates (y) (1. For the basis linear formsthe coordinates yithe bracket defines a "commutation" operation [yi.A. (1.' .. It was rediscovered by F.dt•oxU+s)' .. X. p)). 2 The third case in order of complexity.yi] = diyk = {yi.20) with the aid of M.A. when the tensor h'i(y) depends quadratically on y. (1. · • • Pktqkpkx(k). V.p) = Lptq 2 . ·' ' ll(q..184 B. Conversely. p).) defined from the equations oll(x. if ll(x. X.28). l.3). (1. then we have a Lagrangian L(x.P. ki d• oL q 1 . is also very interesting and has recently begun to be studied [129]. L(x. x(kl)dt (1. S..27) Type II. X.29) turns the linear space L into a Lie algebra (S.M. which we shall denote by L *. Kirillov and B. . Krichever. Ostrogradskij's transformation .. . p). LiePoisson brackets.A. when the tensor hii is not constant.20}{ 1.. The bracket of this form was first examined by Lie [93].23) opi It is assumed that one can solve the equations xi = olljopi for the variables Pi· Similar variational problems with higher derivatives [ 42] bS = 0. . Lie) whose dual space L * is the phase space for the Poisson bracket (1. p . .ollfoqi. Novikov (it is assumed that the equations Pi = oLjoxi can be solved in the form xi = xi(x.. Dubrovin.. . p) is a Hamiltonian.) = pixi ll(x.. x(zkt) = x( 2ktl(q. Berezin [14] and used by A. . k. J S = L(x. X. Pi= . p).yi}.5) imply that the operation (1. (1. .26)1 qi = oH/opj. (1.29) The requirements (1.= x(q." ( 1)• 1·. p) xi = . Kostant [67] (in the less convenient language of symplectic manifolds) in the theory of 1 Note that in these equations the qi and Pi are vectors. Let us now consider the second case in order of complexity. if the equations (1. and j essentially indexes the derivatives of different orders of the vector x.23).
{xi. (1.}=0.2.32) (globally) nondegenerate Poisson brackets on the space T* M which satisfy the following requirement: any two functions f. k).j. g on the base space M (not depending on the variables on the fibre.34) where the signum of the permutation (i. It is useful to consider also a Poisson bracket of the form (1. {xi. (1. i=l. Pi}= 0. if _ ii _ { i. In a region where F = dA the bracket (1.pi}=Jj.3.36) . The corresponding 2form n has the form: (1.xi}=O. 3) the equations of motion of a charged particle in the external magnetic field (n = 2. Let us turn now to examples connected with the LiePoisson brackets.31) Example 2. The function M 2 = I. ck . g} = 0 (see [112]). j. The Poisson bracket of the basis functions Mi on L * has the form: (1. Integrable Systems I 185 infinitedimensional representations of Lie groups. 3) Fii (or electromagnetic field for n = 4). p) and the Poisson bracket (1. Example 1. j. 0.32) can be reduced to the standard form (1. k. xi}= {pi.Fii(x): {xi. The fundamental example of the Hamiltonian formalism of type 1 is the phase space T* Mthe space of covectors (with lowered indices) on a manifold M (the configuration space). pi}= Jj (1.Pi}=Fii(x).M. As a rule one can reduce to the form (1. k are all different eiik ..30) which in addition is distorted by an "external field" Fii = . if there is a pair of coinciding (1.30) and with the form (1. dF = 0.33) The equations of motion with a Hamiltonian H(x.30). Let L be the Lie algebra of the rotation group S0(3).35) indices i. The bracket (1. which consists of all the covectors) have a vanishing Poisson bracket: {f.32) represent (for n = 2.Mi2 is such that {M 2 .32) where the 2form F ==. The Killing metric (W. On T* M there are local coordinates xi (on M) and conjugate momenta Pi (on the fibre) with Poisson brackets {xi.28) is in general degenerate. Example 3. Killing) on L is Euclidean and it allows us not to distinguish between Land L *(all indices will be considered to be lower ones). {pi. Fiidxi" dxi is closed.
41) is no longer degenerate (when p :1= 0). i = 1.+qj.39) where g is an element of the Lie group.A. Krichever. p2 .37) reduce to the form of the "Euler equations" (L. M. On the phase space L * there are 6 coordinates {M~o M 2 .i). Arnol'd.M. (We identify the tangent bundle with the cotangent bundle by means of the standard Riemannian metric on the sphere.{q. such that {.42) On the level surfaces J..M. Dubrovin. Such systems on the groups SO(N) are called the "manydimensional analogue of a rigid body". I. Euler) M = [M. if the Hamiltonian has the aspect of a quadratic form on the space of skewsymmetric matrices M = (M. (When H = t(a 1 Mf +a 2M~+ a 3 Mn the equations (1. and Lis the Lie algebra. Novikov On the level surfaces M 2 = const (spheres) the bracket (1. = {M. (1.aM. in accordance with V. S. With the Lie algebra L of the group E(3) of motions of three dimensional Euclidean space some important systems arising in hydrodynamics are connected.} = 0. h. The substitution CT.32). The Hamiltonian systems on L * have the form: M.} = {. LCT. = ps the bracket (1.186 B. (1. 3.. 2. h._ = LPl. = 0.40) Example 4. M 3 .j=q. This algebra is no longer semisimple.w]. q = 1. The appropriate substitution . (1. The equations (1. 2.37) Let wi = aH.37) coincide with the equations of motion of a rigid body fixed at its centre of gravity).38) where the square brackets denote the commutator in L.(sfp) Pi sets up an isomorphism of these level surfaces with the tangent bundle T*S 2 of the sphere. P. (1.>O.{q._ = p2 .34) becomes non degenerate.H(M)}. the Killing metric allows us not to distinguish between upper and lower indices. It turns out that the brackets which arise on T*S 2 can be reduced globally to the form (1. p 3 } and the Poisson brackets The bracket (1.P. The derivation of the equations (1. = LP. = M. p 1 . q.41) is nondegenerate.) On these level surfaces the restriction ofthe Poisson bracket (1.41) possesses two independent functions J. where H(M)= L dijMi~• i < j d.P.I.38) is valid for all compact (and semisimple) Lie groups on which there is a Killing metrica Euclidean (pseudoEuclidean) metric on the Lie algebra which is invariant with respect to inner automorphisms (1..
p 2 = pcos()sini/J. p).1 Pi· It is easy to deduce from formula (1. ui = Mi. can be given in the form 2H = l.33).. (1. Let us introduce the notation ui = oHjopi. In this case H is the energy. p and positive definite. The equations (1.49) . ():::. 2n. The Hamilton equations will assume the form of "Kirchhoff's equations" (P. The corresponding 2form n takes on the form (1. quadratic in M.aiMi2 + W(l'pJ. . has the form HF = Hn = 4ns = 4nj.sp.pe} = {1/J. e2 = pt/1.aiMf + "L.ciiPiPi. Integrable Systems I 187 (see [112]) has the form: p 1 = pcos()cosi/J.47) coincide (for quadratic Hamiltonians H(M.47) (the square brackets denote the vector product). where .l/f} = {Pe.44) {p 6 . u 2 = p"'tan()sini/Jp6 cosi/J. When s # 0 the effective magnetic field is always different from zero and represents a "Dirac monopole" (nonquan tized). p)) with Kirchhoff's equations for the motion of a rigid body in a fluidin a fluid which is perfect. M and p are the total angular momentum and the momentum of the bodyfluid system in a moving coordinate system rigidly connected with the body.bii(PiMi + Mip) + l.48) One can reduce to the form (1. (1.43) (J3 = pt/1. 112 . The integral of the form F (and Q) over the basis cycle [S 2 ] eH 2 (T*S 2 ) = 7L. (1. F = scos()d() 1\ di/J.e} =0. u 1 = pt/ltan()cosi/Jp6 sini/J. 0:::. supplementarily distorted by an effective magnetic field F. The motion of the fluid itself is considered to be potential.46) S2 [S2] Thus we obtain the standard Poisson bracket on T*S 2 . (1.45) where y' = ().p"'} = 1. The energy H(M. and at rest at infinity [1 07]. e' = p6 .. p 3 = psin().. w] + [p. incompressible.n/2:::.47) the equations of motion of a rigid body with a fixed point in an axially symmetric force field with a potential W(z). ai = oHjoMi..l/l} = {p"'. pt/1} = s cos().J. p) be a Hamiltonian. 1/1:::.43) that {e.. (1. u] (1. (1. {e. Let H(M. n = dp 6 1\ d() + dp"' 1\ di/J + s cos e de 1\ di/J = dei 1\ dyi + F. The corresponding Hamiltonian has the form: H = 21 "L.. n/2. w]. y 2 = 1/J. Kirchhoff) p = [p. M = [M.
[19].1 Ui)+ PLbii(UiPiP. therefore JJ F12d0" dt/1 = JJ F = 4ns S2 S2 (1. . for example.h = p2 = 1.46).52) 2V=s 2Laipfp.2+ 2psLbiiPiPiP. They are the direction cosines of a unit vector.49) can be written as follows in the variables {y.1). (1.53) In view of homogeneity. one always has . Remark. (1. = M. I.2· (1. The quantities P.e.49). (1. Krichever.51) Aa~a=S(LaiPiP.1+ UiPiP. in a potential field U(y).58) by virtue of (1. ~): l H= 2gab{y)~a~b + Aa(y)~a + V(y).55) Deduction ([112]).1 • For the top (1. Steklov [133] and have been rediscovered in a number of modern papers (see. [138]).50) Here for the Hamiltonian (1. The equations of the Kirchhoff type reduce to a system which is mathematically equivalent to a classical charged particle moving on the sphere S2 with the Riemannian metric gab(y). gabgbc = ~~.A. here are dimensionless and do not have the physical meaning of momenta.). but Aa~a=siaiUiPi> (1. On the surface / 1 = p2. the Hamiltonian can also be written on the level surface / 1 = 1. 1 U(y) = V(y) 2 gabAa Ab.188 B. Aa=gabAb.56) and also in an effective magnetic field Fab(y).M.2 + P2LCiiPiPiP. i.P.47) (see [112]). s=fd1 112 . It has recently become clear that on S0(4) there arise systems which in certain cases describe the motion of a rigid body with cavities filled with a fluid.50). Novikov where [i is the constant vector giving the position of the centre of mass relative to the principal axes and the point of attachment.54) 2 V = s2Laipf + 2 W(lip. where the metric gab again has the form (1.51). The equations of the dynamics of the spin in the Aphase of superfluid 3 He can also be reduced to the form (1.48) or (1. S.57) The form Aadya is globally defined on the sphere S 2 .sp 1 Pi• (1. The integrable cases here were found by V. f 2= ps the Hamiltonians H of the form (1. the Hamiltonian H depends only on sp. F12=scosOa 1A2+a2A1.A. (1.uf = gab~a~b• U.. (1.48) we shall have L a. Dubrovin.f2 = s in the form (1.
63) Hamilton's equations can be written in the form .. . The Poisson bracket of two functionals F and G has the form: ( 1. { .n. u"(x)) of some type. p 1 (x). Pi(y)} = 0. . i.. where x=(x 1. (1.. . They arise. from the field variational principle (1.61) Example 1. The Poisson bracket is given by a matrix (1. .64) b£ Pi(x)= {Pi(x).. q. £' = bplx)' (1. } b£ q'(x)= q'(x). bG bui(x) h'l(x.1.. y) bui(y) dmxdmy. .) is the density of the Lagrangian. . ( 1. The local fieldtheoretic brackets of Lagrangian variational problems. Bogoyavlenskij together with the integrable cases in these problems (see [18]). . Pn(x)) with pairwise Poisson brackets of the form {qi(x). defined by the equalities bF = f bF . G[u] the Poisson bracket can be computed by the formula {F.60) Here bF/bui(x) are variational derivatives. Now let us consider infinitedimensional examples of phase spacesspaces of fields u=(u 1(x). For two "functions" (functionals) F[u]. qi(y)} = {Pi(x). xm) is one of the indices in the formulas. .66) where A(q. y) (generalized functions).pj(y)}=b~.b(xy).bqi(x)' where £' = Jt'[p. with the aid of the field . in particular. q"(x). q] is the Hamiltonian.62) {qi(x). .. Integrable Systems I 189 A number of other applications of Euler equations on Lie algebras in problems of mathematical physics have been found just lately by 0.59) of functions hii(x. . G} = If bF . ( 1. which in general depend on the fields.. . bui(x) bu'(x)dmx. There are two sets of fields u = (q 1 (x). ..j=1. qx.. £'} =.65) (1.
.. . has the form: (1.fields. qx. q.190 B...· (1. quadratic in the derivatives of the mapping J. On M" let us fix a closed (q + 1)form w (the "external field"). .. If the manifold N 4 is presented in the form of a product N 4 = pq. that the equations P.62) by a "magnetic field"a closed 2form on the space of fields q(x). Novikov theoretic version of the Legendre transformation (1. S. . . Thus the standard "chiral Lagrangian" for a principal chiral field.P.68) where gil• is the metric of N 4.. can be solved for q:l. just as above in the finitedimensional case..A.)/oq.. As is wellknown (see. r (1.. · One can also consider. for example. i~w is the inner product of the form w=(w. where Ill is the axis of the time t and y = (x. Let us note beforehand that an arbitrary differential form w of degree q + r on the manifold I M" defines a differential rform n. The . N 4 is the Minkowski space N 4 = lll4 . let there be defined a functional S0 (f) on the mappingsf: N 4 + M"..68) can be brought into the Hamiltonian form by means of the Legendre transformation (1.67). then the form n.. Let us analyze an example connected with the inclusion of "external fields" in the theory of chiral.. 1 ).. then the EulerLagrange equations for the action ( 1.1 x Ill. dw = 0.1(y)of(y)foyll. the distortion of the brackets (1. = oA(q. Then it defines a I closed 1form 0 1 on the space of mappings N 4 + M" in the way cited above. (• ) ]:i I~W iz .. Krichever. .)with the vector ~=(~i). Dubrovin.71) If the form w is closed. where M" = G is a Lie group with a twosided invariant metric. possibly with some additional terms.M. Now let us define the procedure for including an external field.70) are "tangent vectors" to the space of mappings (vector fields on M" at the points f(y)). by analogy with the finitedimensional case. [115]). = '> Wuz .69) where ~k(y)=~d(y). on the infinitedimensional space of mappings is also closed [ll2]. t) (for example. I. on the space of mappings {N 4 + M"} via the formula (1.. the definition of a nonlinear chiral field is as follows: one has arbitrary Riemannian manifolds N 4 and M".67) (it is assumed. All =f.1... The functional S0 (f) has the form of a Dirichlet functional. k= 1. . q(x)=f(x).
. y. Lagrange) <5S=O. For these brackets and for local Hamiltonians of the form Yf= Jh(u.73) It turns out that for N~ = p~. ( l. KortewegG. 77) J On a subspace I_ 1 = udx = c (for example. the functional I _ 1 = Judx has vanishing bracket with any other functional F: {F.74) the Hamilton equations u= {u. Example 2. 79) ..75). .LI}=O. de Vries) (KdV) {u(x).75) The Poisson bracket of two functionals has the form: { F.E. Important Example.72) where the functional S 0 is of the type (1. 78) 0 <5Yf u. ( l. defines a socalled "multivalued functional" S of the chiral field fin the external field w [112]. u!•l)dmx (1. 76) is no longer degenerate.y). (1. from the EulerLagrange equations (L.D. The case m = 1. (1.76) is obvious. the correctness of the Jacobi identity follows from the fact that the ''tensor" hii is constant here (it does not depend on the field variables). EulerJ. 76) The skewsymmetry of the brackets (1.59) are called local if the generalized functions hii(x..J. y) present themselves as finite sums of the delta function <5(x. the incl)<lsion of an external field is equivalent to distortion of the Poisson brackets by a "magnetic field" Fa closed 2form on f the space of fields { pq. More generally. This 2form F = 0 2 can be defined via ( 1. The bracket (1.1 + M"} without changing the Hamiltonian. ( l.y) and its derivatives with coefficients which depend on the values of the field variables and their derivatives at the points x. (1. c = 0) the bracket ( 1. (l. n = l. Faddeev bracket) which arises in the theory ofthe Kortewegde Vries equation (D. The KdV equation itself is given by the Hamiltonian I 1 =Yf= f(~ +u 3 ) dx. The extremals ofthis functional can be determined. u(y)} = b'(x.69) with Nq replaced by pq. ux.75) is degenerate.1 . . the fieldtheoretic brackets ( 1. Here one has a bracket (the C.= ox <5u(x) =6uuxuxxx· ( l.75).1 x IR. ZakharovL.68). Integrable Systems I 191 closed 1form (1. GardnerV.Yf'} can be written in the form of partial differential equations.L. G} = f <5F o <5G <5u(x) ox <5u(x) dx. as usual.
J axJ (1.zJ f>(zx)dmz=.81) d3 d d ) A= . z)vi(y)wk(z). S. w]'(x)= vJ(x) axi .BJ(~).z) f>~f>(y. see chap.80) plays the role of the momentum (the generator of the translations in x). This means that the variables Pi(x) are covector densities.A.83) should be written in terms of "structure constants" in the form [v.85) a<. It is curious that one of the manifestations of the integrability of the KdV equation (by the inverse scattering method.P. Dubrovin. is an arbitrary constant). 10 } =ux(x). ovi(x) [v. ( 1. which under changes of variables are additionally multiplied by the Jacobian determinant (we shall call them momentum densities). . dx 3 + 2( u dx + dx u · There is even a family of brackets: one can replace the operator A by A+ A. .83) The role of the index is played here by pairs (x.y). J axJ Jf(z)a<.83) with (1. must be such that the quantity (1.82) Example 3. (1. y. connected with infinitedimensional Lie algebras L. owi(x) . The commutator of two fields vi (x).x'=~. =A f>u(x).84) By comparing (1. i)a point x and an index i.x) a~z'f>(z. Now let us consider continuous examples of the LiePoisson brackets. The operation (1. {u(x). w]i(x)= Jdmydmzc}k(x.87) is scalar with respect to change of variables. (1. (1. Novikov The quantity 10 = f2 u2 dx. (1. The starting point for the subsequent constructions will be the Lie algebra L of vector fields on an mdimensional space. I.wJ(x) axi .84) we obtain c~t(x. on the dual space L * to the vector fields vi (x). G} = f f>F f>G f>u(x) A f>u(x) dx. In the new Hamiltonian structure the KdV itself has the form Mo u. z) = f>~f>(z. wi (x) has the form: .86) The variables Pi(x) conjugate to the velocity components. y. (1. 2 below) is the presence of another local bracket [94] of the form { F.M. Krichever. (d/dx) (A.192 B.x)B~' f>(y.
Pi (z)o~· 1 !5(z. which is considered to be constant) a subalgebra L 0 of divergencefree fields (1.94) O. momentum densities P. Integrable Systems I 193 By (1. (1.(x) give trivial linear forms on L 0 if P. it represents a special case of the Hamiltonian formalism for fluids with internal degrees of freedom. p=const. In the algebra L of vector fields on Euclidean space (where there is a distinguished Euclidean metric and the volume element is the mass density. { (1.85) the Poisson bracket has the form: {pi(y)..95) P. [112] on the space L~ with the Hamiltonian H= f pv 2 dmx 2 ' p=const' ov'=O l • .91) In other words.=pv._= {P.p(z)!5'(z. z)p. where quantities of the form o.89) The substitution p=u 2 reduces this bracket to the bracket (1.P have arisen because of the transition from L~ to the space L*. The Hamiltonian formalism for a perfect compressible fluid cannot be realized on the algebra L..(x)dm x = Pk(y)oY' !5(yz). (1. (1.90) is given.t/> ).92) The Euler equations for the hydrodynamics of a perfect incompressible fluid can be written as a Hamiltonian system [7].(x) = O. . which is equivalent to the space of velocities in the given case pv. The pressure pis only defined up to a constant here. The Poisson bracket on the space L~ may be written in the form (1. H} +o.88).t/>(x): (1. p(z)} = p(y)!5'(yz).p.93) and the Poisson brackets (1. Pk(z)} = f c~k(x. p=pv' l • (1.88) In the important special case m = 1 we get {p(y). The dual space L~ can be obtained by factoring by the gradients L~ = L */(o.y). y. The terms O.V'=O. One always writes these equations on the full space L*.75).p are equivalent to zero.y). Even the ordinary compressible fluid has such internal degrees of freedomthe mass density p and the entropy density s.
s· The corresponding variables in the dual space L.y). S. we shall add another pair of fields vP and v• with commutators of the form [(v. ••• . (w.99) are firstorder quasilinear equations u.y) {u'(x). Besides the vector fields vi. wP. p (the velocities are here the covectors vi= PiP. Other examples of LiePoisson brackets which arise in hydrodynamics can be found in [112].. (1.1 ).100) The most general form for Poisson brackets which lead to the equations (1. General brackets of hydrodynamic type...(oivioivJc5(x. Dubrovin. H} (1. p(y)} = p(x)oic5(y.98) where the densities h(u) depend only on the fields u=(u 1. w].. . (1. s)]dmx is just the energy. v"). s(y)} = s(x)oic5(y. i= 1. vi(y)} =.194 B. s(y)} = 0.. I.97) 1 { vi(x). have the form: { Pi(x).• we shall denote by p (the mass density) and s (the entropy density).x).M. oc5(x. { Pi(x). w")] =([v. . . 2) The Hamilton equations u!(x)= { ui(x). Essentially the Poisson brackets (1. p(y)} = {s(x). The quantities M = pdmx and S= sdmx have J vanishing Poisson brackets (the trivial conservation laws).97) were appropriately chosen so that mass and entropy would be transported together with the particles. vioiwP wioivP..96) by Lp.100) for Hamiltonians (1. =v~(u)~. } . n. in contrast to the energy. vP.A.x).96) We shall denote the algebra (1. uJ(y) = g'J"(u(x)) oy" + W"(u(x)) 0~uX1 ~> c5(x · y)~ {1. (1.vioiw". (1. Novikov whose inclusion requires the extension of the Lie algebra L of vector fields. Example 4.101) .P. u") and not on their derivatives. which is conserved only as a whole. { p(x). The Hamiltonian H = J[p 2 /2p J +e(p. s(y)} = { p(x). The Poisson brackets and Hamiltonians considered in the previous example have the following properties: 1) The Hamiltonians have the form: J H = h(u)dm x. Krichever. The Poisson brackets in L.wioiv").98) is as follows: .
i. The set of relations in [49] is incomplete and incorrect for N :::. For nondegenerate metrics gii« the conditions under which the expression (1. in particular. 2.l02) like Christoffel symbols (see [48]). (1. .88) are always degener ate. to a constant one: (1. This means that by local coordinate changes ( 1.a are linearly depen dent on the coordinates (u.1 02).lOO) and the brackets (1.b~ifJ glia. The complete set of relations can be found in [Mo].. . rx = 1.) (see [49] for N > 2 and [Mo] for N = 2). the connection to zero.105) It should be noted that the natural "physical" variables ui in which the equations (l.102) like a metric (with upper indices) on the space of fields u. For example.e.I is nondegenerate and is reduced to a constant metric. defined by the equalities (1.103) In this case the symmetric matrix gii(u)=gii(u) behaves under the changes (1. However all the metrics gii« cannot as a rule be reduced to a constant form by a single transformation. then the quantities qk.103) gives a Poisson bracket if and only if the conn~ction qk is symmetric. Integrable Systems I 195 The form of equations (1. then the connections rj~. where W« = .l03).88) when m > 1 such a reduction is impossible.101) gives a Poisson bracket can be written as a set of relations on the tensors Tiik«fJ that we shall not discuss here (see [49]). In the multidimensional case m > 1 a family of metrics gii«. the Hamiltonians (1. These questions are discussed below. by the same token. If it is nondegenerate. and have zero curvature. symmetric.102) Let us consider here the onedimensional case m = 1: (1. If they are nondegenerate. It turns out [48] that the expression (1. In any case.101) is invariant with respect to local changes of field variables u=u(w). the metric gii(u) is nontrivial in the coordinates ui. The obstruction to such a reduction are the tensors Tiik«P = W« glkfJ.1 04) transform under the changes (l. compatible with the metric gii and has zero curvature. for the brackets (1. are compatible with these metrics. .100). Poisson brackets of hydrodynamic type play a fundamental role in the theory of the Hamiltonian systems of hydrodynamic type and. and the bracket (l. the classification of Poisson brackets of hydrodynamic type is reduced to certain ques tions in the theory of special infinite dimensional Lie algebras.. Thus. m actually arises.98) and the brackets (1. the following theorem is true: if the metric gii. Let us also note that for m > 2 the metrics gii« = p 5 ((ji•(ji« + (ji•(ji«) corresponding to the brackets (1.gisan~. then all other metrics gii. in the study of the effective integrability of such systems in the case of one spatial variable (see Appendix below).103) arise are essentially "curvilinear". the metric gii can be reduced to the Euclidean (or a pseudoEuclidean) one.
i. This is the only case . In the case B = lR there is only one nontrivial "GelfandFuks" cocycle for k=3 (see [GF]).t.M. p(x). Such finitedimensional algebras B over lR or C cannot be simple for N > 1 (see [Z]). if the metric gii is nondegenerate at the point u = uo. We now restrict our attention to the spatially onedimensional case (m = I) and. to x). Dubrovin. it seems that the only nontrivial cocycles are cocycles of order 3 at some point (uo). where (k) is the number of differentiations w. Let B be an algebra with a basis e1.. but their general theory has not been developed. In the general case. if the coefficients of the Poisson brack ets are linearly dependent on the "fieldscoordinates" {u(x)}. S. 2.q(x)] = p'q q'p. q(x) E $8. The Jacobi identity for $ 8 is equivalent to the Jacobi identity for the Poisson bracket of hydrodynamic type (1. following [BN]. 1. depending only on the first derivatives w. x and y(jl = const.103) C ij  k  bij k + bjik' b1 = const.P.103) with m = 1 and gii = c!iuk.e. then the "small" Lie algebras form a background of a "big" algebra of Poisson brackets. then this metric is flat. y(jl = C~i u~..first order (i. . we obtain commutative associative algebras B. In particular.(x). Lie algebras c:£8 have !arge families of central extensions 0 ~ lR ~ $8 ~ $8 ~ 0 defined by cocycles of the form < p. there are many nontrivial cocycles fork = 0. In this case. . p.e. In the important case b~ = b{.q. i.(x). we obtain Frobenius algebras ([BN]).r. Such brack ets are called the LiePoissonBeresinKirillovKostant brackets and the systematic development of their theory was initiated in [GDo] and [BN]. eN and multiplication Then the structure of an infinite dimensional Lie algebra $ 8 on the space of B valued vectorfunctions of a single variable x is defined as follows: [p(x). the Jacobi identity is equivalent to the following two identities in the algebra B: where La(b) = Rb(a) = ab.A. consider hydrodynamic type Poisson brackets of the form (1. This is a general form of spatially onedimensional translation invariant Lie algebras of the . q(x) = e. Moreover. and the multiplication law is induced by the multiplication in B. X E S 1.r. 3 ([BN]). q >= i sJ ij (k) Y<klPi (x)qi(x)dx. Novikov As was mentioned at the beginning of§ 1.196 B. If det gii =I= 0. where p(x) = e.. I. Krichever.e. det(gii (u 0 )) =I= 0.t..
then the curvature is constant.k = gi. y) L + :L<w")iu!a.y)]. + [e~u! .81). If L = I and (w 1 )~ = coj.75) and the second "Magri bracket" (1. 75) has the form [34]: 1 {u.y). we have the following result: if det(gii (u )) =1= 0. Such brackets are important for the theory of integrable systems of hydrodynamic type (see Appendix).y). then the form giidu.y) + b~ (u)u!o(x .(x).y) + . Recall that these brackets appear in the study of the KdV hierarchy.t' •=0 k=O The notation ( >+ will be explained below (Insertion 2) . Also see [F] for general theorems.y) + [c~u!. b..(x). A dui is closed if and only if the Poisson bracket can be reduced to a constant bracket. a= I Here the metric gii is no longer required to be flat.ei' has zero curvature and zero torsion (see the result of Potemin in [N2]. + di1u!u~Wk. ui(y)} 1 = Mii o8(x. There is also a higher orQer analogue of Poisson brackets of hydrodynamic type. where the functions g. c. Integrable Systems I 197 where the flat coordinates for the metric gii = cf uk have been effectively con structed. where Mii is defined by the relation n1 :L MijTJja.(x).1 [(w")!(u)u~8(x. We would also like to make the following remark about the Important Example (above).j=O for n1 n1 X~ = "a•1 ~ X Q TJso L = an + "uk ak X ~ . ui(y)} = gii(u(x))o<k>(x. this problem is solved only in a number of special cases con nected with topological quantum field theories [Dl]. The theory of such brackets for k = 2 was constructed in [PI].. d depend only on u.. The theory of integrable systems led to the generalizations ofthe Gardner ZakharovFaddeev bracket (1. also see [D]). The connection IJ.. =([X~. [F]): {u.2 >(x. where the theory of Poisson brackets and systems of hydrodynamic type recently began to play an interesting role (see [Dl] and Appendix to [Nl]).. In general. The higher order analogue of the bracket ( 1.y) + b~(u)u!o<kJ>(x. + · · ·]8(x. Nonlocal Poisson brackets of hydrodynamic type have the form ([MF]. Fork = 2. ui (y)} = gii (u(x))8'(x. These brackets are called "homogeneous differentialgeometric brackets of order k" and they have the following form (see [N2]): {u. LD+ i...
1 = tk=l+J1 [(k{I)u*a. A summary of the results of [D2] can also be found in [DNI].8) if its bracket with the Hamiltonian H(y) is equal to zero: .. They can be interpreted as classical limits of "Zamolodchikov W algebras" [FL].ij1_(k~I)ca. Let the tensor r V ® V ~ V ® V satisfy the "linearized YangBaxter equation": [r12.(LX~)+L.P. Because of this connection with twodimensional confonnal quantum field theories. The connection between the linearized YangBaxter equation and another class of Poisson brackets .y). I. i. Example 5. Systems with Symmetry A function F(y) is called an integral of the Hamiltonian system (1. S. The higher order analogue of the bracket (1. r23 are defined similarly. Novikov and is equal to M.i=O and this must be satisfied for any 1J = (7Jo. rz3] = 0.1 o o(x. rn] + [r12. rz3] + [rn.M. Dubrovin.198 B. Integrals and Reduction of the Order of Hamiltonian Systems. Then on a space with coordinates tj one can define a Poisson bracket by the following fonnula: {tk'i tj} I i}tatb tit} ab = rab k I . Some useful infonnation and references can be found in [DNI]. See [FT] for the detailed discussion on the connection between the theory of the linearized YangBaxter equation and the soliton theory and for a large list of references on the subject.ij1)ouk]. 1Jnd· It is remarkable that the higher order analogues of the second bracket turned out to play an important role in the confonnal field theory.. where n1 L(Kijl]j)O~ = L(X~L)+. these Poisson algebras are often called "classical W algebras". (x). Krichever.The YangBaxter brackets. where r12 = r : V ® V ® 1 ~ V ® V ® I and rn.difference analogues of Poisson brackets of hydrodynamic type was studied in [D2]. . Consider one more fundamental class of Poisson brackets related to the theory of integrable systems .A. ui (y) } 2 = K.a brkl · These brackets are very useful for the integration of certain integrable systems in the soliton theory. § 2. .81) has the fonn [2]: { u.
we get: the quantity F is conserved along the trajectories of the Hamiltonian system (1.8) by two all at once.H}=O.8). 3 of the current publication and chap.8) with Hamiltonian H permutes the trajectories of the flow (1. Let us give the appropriate construction (see also vol. zN.2 (y) satisfy (1. ••• . which commute with any Hamiltonian. F} = l.108) The flow (1. but the Hamiltonian of the reduced system will depend explicitly on time [7].108) lying on the level Me are "indexed" by the points of a surface M~ (in general defined locally) transversal to these trajectories.. {zP. Let F(y) be an integral of a Hamiltonian system with Hamiltonian H.11) into account. 3.2 give local coordinates on the transverse surface M~. the Hamiltonian H itself (if it does not depend on time) is always a conserved quantity. Arnol'd and A. The trajectories of the flow (1. . .110) Therefore we may impose on the choice of coordinates z 1. (1. .I. If the Poisson bracket is degenerate. Let us define the "reduction operation" of the original Poisson bracket hii(y) onto M~.8) lie entirely on a level surface F = const. { F(y). {r. zN. z'} = f4(z. We have looked at an example of the reduction of the Hamiltonian formalism with the help of trivial integrals in § l in connection with equations of the Kirchhoff type.108). They are constant along the trajectories ofthe flow (1. F(y)}. Let us consider a level surface Me: F(y)=c. where the vector (~~(y)) = (hii(y)(oF(y)joyi) is independent of ~ 8 . The trajectories of the system (1.2 the useful 3 With the aid of the energy integral H the order of the system can also be reduced by two. Integrable Systems I 199 {F.106) Taking (1. z(y)} = 0. and therefore defines a dynamical system on the set of trajectories of the flow (1. N. Givental'). F). .107) and on it the Hamiltonian flow defined by the Hamiltonian F(y). In particular. then there are always "trivial" integrals (1. Hamiltonian systems with one degree of freedom (N = 2) with a timeindependent Hamiltonian can always be integrated by quadratures. By the same token..106).109) and not functionally depend on F(y). We obviously have {r.13). § 3 of the article by V.108) by virtue of the commutation (1. (1.108) and together with F(y) and r they define local coordinates in a neighbourhood of the transverse surface M~. Let us consider the subalgebra of all functions z(y) which commute with F(y). i = l. (1.B. (1. zN. F). z'} = hPf(z.. (1. ••• . .109) Let the independent functions z 1 (y). the quantities z 1.. The presence of nontrivial integrals when N > 2 which do not depend on the "energy" H 3 allows one to reduce the order of the Hamiltonian system (1. y~ = {yi.
115) (by one quadrature). . Dubrovin. .19).P. oH(z. p) is invariant with respect to "spatial translations" (1. Pn) of the form (1.. .M. N 2.. It is sufficient.113) and therefore the original Hamiltonian system has a welldefined restriction to o. and does not depend on the choice of the surface M~. ••• . H(x. zq}red = {zP(y).111) The reduced Poisson bracket on M~ has by definition the form {zP. . F).106) the Hamiltonian has the form H(y)=H(zl. p. H(z. F)}= oF (1.pn. ..200 B.q= 1. S. p 1 . .p n 1. the righthand side depends only on the coordinates on M~ (and on c).112) Obviously. the quantity ("total momentum") n F= L Pi i= 1 (1. c). After this the dependence of the coordinate r on time can be determined from the equation (taking (1. of translations along the trajectories of the system (1. (1.=O. M c.A.111) into account) .. . to suppose that c is a regular value of the function F(y) and the oneparameter group G. obviously. (1.•• . (1. By virtue of(l. x". p) be a Hamiltonian on the phase space ~ 2 " with canonical coordinates (x 1.. Let us suppose that H(x.) (1. (1.. (1.110).2 • Example 1. zN. for example..114) Thus. N 2.114). Novikov additional conditions {r.X. q= 1. F) r = {r.xn1 x. integration of the original system (1.116) i. Let H.2. .117) For this it is sufficient that the Hamiltonian have the form H( x 1. zq} =0. (1. ..119) . I.. In the practical realization of the procedure described above the main difficulty lies in the construction of the "transversal" coordinates z 1 . zN. .e. .. . . Krichever. . n iJH La i=1 X . ••• .. zq(y)} =hpq(z. ..8) is reduced to the integration of the Hamiltonian system (1. Whether it is possible to carry out the reduction procedure globally requires a supplementary investigation.Pl>···•Pn n ) _ H~( 1 n X x.108) is compact and has no fixed points. . whose order has been lowered by two.118) Then.
...xn1 . .122) q=l The reduced brackets are the canonical ones: {xq. H} =0.121) has the form . Example 2. then the presence of n pairwise commuting integrals for the Hamiltonian system allows one. {xq. We shall discuss the properties and examples of such systems in§ 3. . .. z 2 ".x=(x 1 ..2 (1. (1.Pn. F} =0. Here one has the three . {H.p.n1.c) on the surface p.x 3 ). (1. The proof is obvious from the Jacobi identity (1. . Integrable Systems I 201 commutes with H. A set of noncommuting integrals also allows one to reduce the order of the original Hamiltonian system.. (1.124) The dependence of the quantity r = x" on the time t can be found from the equation (1. •. N = 2n. according to the scheme described above.. x'}red = {pq.. The coordinates z=(z 1.p=(p 1 ... +pn=C (1.. .5). however.p 3 ).j= 1. . Let H(x.. Let us note.126) allows one.2 )=(xq.8) form a subalgebra with respect to the Poisson bracket.c..p. p. to integrate this system by quadratures. •••  .. pq) on the reduced phase space have the form Xq=Xqx". . first of all. if the initial Poisson bracket was non degenerate.1 .c)H x. p. . q= 1. in principle. and the original Hamiltonian system reduces to the system on IR 2 ".} =b~..1 .n1. The presence of nontrivial pairwise commuting integrals F 1 (y)..r=1. . L n1  Pq ) . here the reduction algorithm is more complicated. (1. p) =(lpl 2/2m) + U(lxl) be a spherically symmetric Hamiltonian on IR 6 . _ ~ ( 1 H(x.125) Now let us suppose that the Hamiltonian system (1.}red =0.8) with the Hamiltonian H possesses several integrals. Let us first analyze a simple example.. q=1. .. .120) pq=pq.123) q.. {F. Fk(y). n1.+ . Fi} =0. p. to reduce the order of the Hamiltonian system by 2k.. a simple but important assertion: the integrals of the system (1. k. i. {F.p 2 . . In particular.x 2 . The reduced Hamiltonian H(x.
serve as canonical coordinates on the reduced phase space. M3=X1P2X2P1 (1. Let us fix a value of the angular momentum M=m#O. If we factor by this flow. thus acts on the threedimensional surface x 1 = p 1 = 0.2  m 2 +U(r). putting x 2 = rcos¢. and the conditions (1.P.133) Then r.) Here there are three integrals.128) (In fact. from which we get (1. i. Dubrovin. the motion takes place in the (x 2. p 3) planes by the same angle. M2=x3p1x1p3. (1.130) can be written in the form (1. x 3) and (p 2 .202 B.131) From the last two equations it follows when 11#0 that x 1 = p 1 =0.P. 0. For the factorization it is most convenient to use polar coordinates r.cos ¢. Mi} =eiikMk. ¢ in the (x 2 . sin¢. the reduced Hamiltonian has the form . we obtain the desired reduced phase space.129) is conserved.p] (1. x 3 = rsin¢ (1.135) . Novikov "angular momentum integrals" M1 =x2p3X3p2. I.130) Without loss of generality we may assume that m = (Jl.132) and introducing the conjugate momenta p 2 = p.e. x 3) plane. the whole angular momentum vector M=[x.p. S. (1.)=... but because of their non commutativity the reduced phase space will have dimension 2. x 2 p 3 . r (1.M. p. x 3) plane. which represents a rotation in the (x 2.127) with pairwise brackets {M.134) The dependence of¢ on time is obtained separately from the equation P. = Jl. 0). mr (1.. Krichever. p. Jl2 2 + H(r.x 3 p2 = Jl. The flow with the Hamiltonian M 1.A.
(1.128) of example 2). Integrable Systems I 203 The construction of example 2 admits of obvious generalizations. . j= 1.(y)}. .. a') is a vector in L. of the Lie algebra L are the Hamiltonian flows y{= {yj. . .(y). (1.r. .) E L * and let us consider the momentum level surface (the simultaneous level surface of the integrals F 1 . . .. Thus the space of linear combinations L={a. whose pairwise brackets can be expressed as linear combinations of these same functions. F. then F(y)(a) =a.8) possesses integrals F 1 . . F. Then G acts locally on the phase space by canonical transformations (ones which preserve Poisson brackets): the oneparameter subgroups of G which correspond to the basis vectors F. compare the formulas (1. .F. . which go back to Jacobi and H. .(y))=F(y) may be considered as the coordinates of a linearform on the Lie algebra L: if (a 1 .142) Such vectors a form a Lie subalgebra Le c L. I.}.. a basis of it is constituted by the functions fs(y)=a~F. Given tal'). a') satisfies the linear relations {fa.) (1. .B.(y). . § 3 of the article by V. . F. Suppose the Hamiltonian system ( 1.141) Let us suppose that this surface Me is a manifold. . . .. a' are constants. F. 3 of the present publication and chap. • .. (1.. . Let I be the dimension of this subalgebra.I. ..FJ1Mc=aic7jck=0. (1. s= 1. . and the c7j are the structure constants.137) is closed with respect to the Poisson bracket and for this reason forms a finite dimensional Lie algebra.(y) form a basis of L.F. (l. .140) Let us fix some element c = (c 1 . .138) For a fixed y the collection of numbers (F 1 (y).. .(y) preserves the level surface Me if the vector a= (a 1.. N. a 1.136) the coefficients c7j being constants (this is the next case in order of complexity after commuting integrals. . 3. F. . (1. The functions F 1 (y). F. (1.143) .. . . Arnol'd and A. c. The Hamiltonian flow corresponding to the Hamiltonianfa(y) =a. Poincare and have been formulated in the language of symplectic manifolds in a number of works in the last decades (see also vol. ...139) Thereby we have defined the momentum mapping y~>F(y)EL*. . Let G be the corresponding Lie group. j=1.
149) The functions Fi(Y) are integrals of the Hamiltonian system (1. must generate the whole tangent space to Me). Fi(y)}. let e 1 . In example 2 (above) the subgroup Ge coincided with the rotations about the c = (Jl. A canonical Lie group action is called Poisson if the functions Fi(Y) are defined globally and their Poisson brackets have the form (1. . . We obtain the reduced Hamiltonian system (1. where the c~i are the structure constants ( 1. (The subgroup Ge c G with the Lie algebra Lc c Lis just the isotropy subgroup ofthe element c E L * in the coadjoint representation Ad*. e. j = 1.. the transformations y~+exp(reJy are translations along the trajectories of the system y{ = {yi.e.148) of the Lie algebra L.ft. (1.e.142).j..) If we factor Me by the action of the flows with the Hamiltonians .fa out of the subalgebra Le. I= r and the order of the system is reduced by 2r.146) It is clear that in the commutative case c~i = 0 the subalgebra Le coincides with the whole Lie algebra L..204 B. gEG.. .e.A. As coordinates on M? one may take functions zq = zq(y) such that (1. I. Indeed. As above.144) which do not depend on the functions .143) (the gradients of the functions zq andf. we obtain the reduced phase space M? (of codimension r + l). N.P. 0) axis.136). i. 0.147).M. Novikov where (a!) is a fundamental system of solutions of the equations (1. (1.148) Each oneparameter subgroup exp(tei) of transformations of the phase space has a Hamiltonian Fi(Y) (defined perhaps locally). i. We have not yet discussed the mechanisms by which integrals of Hamiltonian systems arise. Fi(y)} = dr H(exp(reJy)r=o = 0. (All Me is fibred (locally) over M? with fibre Ge). The functions Fi(y) are called generators of the canonical action of G. (1.147) Let L be the Lie algebra of the group G. preserving the Hamiltonian: H(gy)=H(y). we define the reduced brackets on M? by the equality (1. S. Krichever. d {H(y). Dubrovin. . ••• . the presence of a continuous group G of canonical transformations of the phase space. and let the commutators in L have the form: (1.145) The Hamiltonian also restricts to M? in a welldefined way. be a basis of L. The best known of these mechanisms is a symmetry of the Hamiltonian system. . i.
In the state ofhydrodynamical rest . 17]. have the form ( 1. are welldefined.151) The functions F.153) (a consequence of the Jacobi identity (1. without fixed points.. and the bij = .(y). ~]. for example. the reduced phase space has the form T*(M/G). even if the Poisson bracket is nondegenerate.136). ~)=0 (1. Example 3 ([112]).5)). where y=(x. if of course the quotient manifold MjG is defined. Their Poisson brackets. the isotropy subgroup G. the integrals F.bii are certain constants.(y) might not be globally defined (and singlevalued). In the first place. Leggett's equations for the dynamics of the "order parameters" in the Bphase of superfluid 3 He.140) (i. (]. O+B([(.153') (the cocycle bij is co homologous to zero). .e. The skewsymmetric matrix bij defines a bilinear form on the Lie algebra L. The corresponding action of G on T* M is canonical. for the case T* M. Let us suppose further that the functions F. p) are the canonical coordinates on T* M (locally). Let the phase space have the form of the cotangent bundle T* M of a smooth ndimensional manifold M with the standard brackets (1.)x). In this case.. 17) = b. where the group G acts on M.150) Let us put (1. which is a (twodimensional) cocycle: B([~. Let k X. 1J)+B([1J.=o• = dr k k =I. and its elements act on M. It is not difficult to show that then their Poisson brackets have the form: ( 1. as is easy to see.152) where the c~i are the structure constants of the Lie algebra L. is a manifold). If the action of the group G on the phase space is Poisson. Thus the action of G on T* M is Poisson..30) and let G act on Mas a group of diffeomorphisms.(x) d (exp(re. Integrable Systems I 205 Example. For the action of the group G to be Poisson it is necessary that the matrix bij should have the form (flk are certain constants): (1.(x. Thus. by substituting Fi ~ Fi + fli we obtain a Poisson action. A. to suppose that cis a regular value for the momentum mapping (1. p) are defined globally on T* M and are generators of the action of G. B( ~. M. . then the reduction procedure described above for the Hamiltonian formalism can be carried out globally under certain additional restrictions. For an arbitrary phase space a canonical action of G might not be Poisson. Let us construct the functions F. It is sufficient. n. only their differentials dF.i~i1Ji. ( 1. are defined globally (up to a constant). of the element cE L * with respect to the coadjoint representation Ad* is compact.J.
represent coordinates on the dual space to the Lie algebra of the group S0(3).1 58) After the substitution (1. (1.. The variables s. S.). V(cos O)=const(!+ 2 cos 0) 2 . Rik) the standard Poisson brackets on T*S0(3) are written thus: {s.} =eiikRkl· (1.ni +sin Oeiiknk.. 3. Novikov and with nonzero spin a state in the Bphase is defined by a paira rotation matrix R = (R. where g belongs to the group of rotations around the axis of the field F. (1. {s. (1. 0). With zero flux F = 0 the system admits the group S0(3) of transformations (1. i = 1.F. In the variables (s. s 1+ gs. further.M. Suppose. Ai} =eiikAk.. bare constants. !as 2 + V(cos 0)} = 0.1.162) The variables s2 and 0 which enter into the Hamiltonian when F = 0 generate the . si} =eiiksk. (1.i) eS0(3) and s = (s. Krichever.).. 2. l. 0.159) we will obtain a Lagrangian system in the variables (Rii• Rii) on T*S0(3). (1.. {Rii• Rk. Ri. F=(F.154) The Hamiltonian of the Leggett system in the Bphase and in an external magnetic field has the form: H =!as 2 + b ~). conserved when F=O: (1.A.160) If the field F=(F.206 B. (1.160).161) where the Poisson brackets are the same as for the ordinary angular momentum {A.1 55) where a. {A.the "magnetic moment".} =0.156) here Rii is the rotation by the angle 0 about the axis of the vector (n. Lnf = 1: Rii =cos Ob. analogously to the angular momenta M.160) and is completely integrated in [95]. and the potential V(cos 0) is invariant with respect to inner automorphisms R 1+ gRg. F=(F.P. where the kinetic energy is defined by a twosided invariant Killing metric. Dubrovin.i + (1 cos O)n. then the whole Hamiltonian is invariant with respect to the oneparameter group of transformations (1.1 57) 1+2cos0=Rii=tr R. The transformations (1.) is the external field.160) generate the vector.. g E S0(3).+ V(cos 0).) is constant.
(1. With the condition A 1 =const. n} =Jt(ln )A 2 2 iAi. 0.164).167) with a constraint of purely geometric origin s 2 r 2 =(s 2 siXs 2 s~)(s 2 n 1 s 1 su) 2 • (1. The quantity A 1 . where g is a rotation around the first axis. 0. has vanishing bracket with all the variables 0= {AI> s 2 } = {A 1 . sm The quantity A 2 =LAf=(1cosBXs 2 s~) has zero Poisson bracket with everything in this subalgebra { A2 .0}=1. that of n=(l.8=1. 11 11 ). Variables which are invariant with respect to this subgroup are the following: (1.n}=O. su} =0. (1.169) Therefore. (1.155) and has functional dimension 5. s 8} [112]. if we impose the condition A 1 =const it is possible as before to make use of the formulas for the Poisson brackets of the quantities (1.168) It is not hard to work out that the variables (1.160).167) form a closed Poisson bracket algebra which contains the Hamiltonian H (1. {O.8=2s { } s . 8} =0. su} = {A 1 .164) 11 . 2 B (ss 2 (1.167) which follow from (1.170) Their brackets have the form: {su. si} ={AI> t}. {A 2 .165) In a nonzero magnetic field (F. The integral A1 generates a group (1. 0} = {A 1 . ni} = {A 1 . s 2 } = {A 2 . s 11 } = {A 2 . 0) there remains only one integral (besides the energy) (1. which lies in this algebra. where 11 .171) {A 2 . 0). we shall choose the following as basis variables: (1. { 2 } 1 +cosO s.163) 2 } { s. . (1.su=. 0} = {A2 . Integrable Systems I 207 closed Poisson bracket algebra {s 2 .166) In the present case it proves to be possible to carry through the reduction procedure for the Hamiltonian formalism to the end (globally) and to reduce the system to two degrees of freedom. s 2 } = {A 2 .
2 2 . y2 = ¢.174) and let us pass over to the Lagrangian formalism. the canonical variables can be chosen in the form X 1 =0. E=.208 B. When A1 .A. (1. there hold laws of conservation of the total energy E=O. A2 =8bFcos¢sin 3 xcosx. n. n=n 1 =sin¢ (1.176) and of the total momentum vector (1. Thus we have obtained a system on a region of the sphere S2 with the usual metric. 2 2SID SID Xcos '+' X ~b 2 F 2 (sin 2 ¢ + 4cos 2 ¢ sin 3 x cos x). +bF . t.65). I. Novikov Thus..175) where i = x.U(y). Now let us consider examples of continuous systems with spare integrals. (1.~. qx. . P2 =pn= 1n2 (1n2)2" The Hamiltonian takes on the form: Now let us introduce spherical coordinates 0=2x. We will obtain L=2a(i 2 +sin 2 x~ 2 )A 1 y 1 A 2 y 2 . where there is an effective magnetic field and a scalar potential. described in the Hamiltonian form with the aid of the transformation ( 1. since it has a singularity for ¢ = 0.) does not depend explicitly on the space time coordinates x«..172) 2A 2 Af x2=n. aAf A1 (1sin 2 xcos 2 x) U= V(cos0)+ 4 . Let us first consider the case of Lagrangian fieldtheoretic systems (1.60 this system cannot be extended onto the whole sphere.67).. In the case when the density of the Lagrangian A(q. tX = 1.A) (1.177) . Dubrovin. Krichever. . S. m.. Pt=Pe=sn.Yf=Jdmx(piq. A1 =2bsin¢. q.M.P.
. . X 0 = const...179) Let us stress that the generators of the spatial translations are local field integrals.n. x 0 = t) which are invariant with 0 ).q. x 0 =t. (1. .m. The quantity (1. We have [142] J E= Tgdmx. m.(x) . {q (x). . . respect to more general oneparameter groups of transformations G.180) where a.. S=JA(x.m. 1..m. by introducing the energymomentum tensor (1.182) More generally.184) do not affect time. .. (1.. Integrable Systems I 209 The functionals Pa are generators of the translations in the spatial variables. i. l.Pp}=O. . whose generator is a local field integral with the density (1. i. one can consider variational problems of the form JS=O.186) (Noether's theorem (E. .e. rx..184) q~=Qi(x... op.m. as above. If the transformations (1. (1. .181) oT'b=o axa ' b = o' 1' . a=O. To each such group there corresponds a "conserved current" Ja=AXa+ oi\(Qiqi Xb) (1. (1. (1.q).P=l.. Pa} = oxa ..177) are often written in infinitesimal form. then they define a family of canonical transformations on the space of fields (p(x). b=O. The conservation laws (1. J Pa= T~dmx.185) (} i qa x• ' (1.(x).. q) of the form x~=X 0 (x). . Noether) [17]).183) (here. (1. Pa} =~.e.188) The second Noether theorem concerns variational problems which admit .176).qx)dm+ix (1. i=1. oqi(x) {P.(x. . we put x = (x a= 0.187) is conserved.. l. q(x)).178) {Pa.
. F1} =O. . 1 . F 2 .: 'f. on a 2ndimensional phase space) where there are exactly n functionally independent integrals F 1 = H.210 B.n. Novikov symmetries with functional parameters (as.A. In this case the level surfaces of the integrals (1. c) in the coordinates (s. This subgroup is discrete and is therefore isomorphic to a lattice spanned by k vectors of IR". A level surface Me ofthe form (1. the independence of their "gradients" (~. acting on x 0 . F n whose pairwise Poisson brackets are equal to zero.<2n) ("actionangle" variables) such that: a) (1. I=. . i= 1.Fn (i.. (1. . ..=O.e. ..189) are quotient groups of IR" by lattices of finite rank ~ n.191) l/J.=bf~1 . Dubrovin. Now let us construct the actionangle variables. A subgroup {d} c IR" arises.. (1. . The important theorem of Liouville (J. S. Fn acts on this manifold. . . • .. but satisfy some system of differential relations. If the level surface (1. The equations for the extremals (1..• . ••• . Krichever. .190) b) s. On the given level surface Me one may put together linear combinations of the fields ~. where k ~ n. . [ 42] ). Liouville's Theorem. .(S1.M. then in a neighbourhood of it one can introduce coordinates st> .189) is compact.. sn.(F 1 ..65) are not independent in this case.) 1 =hi"(oFJoy").=s.. t~J1 are coordinates on the levelsurfaces (1. Let us give the idea of the proof ofthis theorem (see.e. for example.192) such that the coordinates introduced with their help on the group IR" acting on . I. {F.n. only for k = n will we obtain a compact manifold (a torus T"). . compact nonsingular level surfaces are ndimensional tori. . yields x 0 again. in particular. People often call such systems completely integrable. in the theory of gauge fields [53]). Liouville) studies the case of Hamiltonian systems with n degrees of freedom (i.P.j= 1. ActionAngle Variables In this section we shall restrict ourselves to the consideration of phase spaces with a nondegenerate Poisson bracket. for example.189) is a smooth manifold by virtue ofthe independence of the integrals F 1 . n.189). •. = W. Fn). . § 3.. l/J) the initial Hamiltonian system has the form: s. . ••• .l/J 1. l/Jn (O~l/J. The group IR" ofthe flows with the Hamiltonians F" . We shall not discuss this theorem here. Sn). Let us choose an initial point x 0 = x 0 (c) e Me and let us pick out a lattice in IR": a vector dE IR" belongs to the lattice if d. Obviously. } .
yi: 05.193) This introduces coordinates ~ 1 .197) On an arbitrary phase space with a form 0. Integrable Systems I 2ll the torus T" =Me coincide with the angles 05...=dw.199) Let us select the bi according to the condition {c/Jio c/Ji} = 0.200) The equations of motion will have the form (1. ~i} ~0. analogous to (1. . ~i=const for j~i.2n. F". In this region we have coordinates (F 1 . . ••• . p 1 .190). ••• . ~" chosen earlier. The coefficients bf will depend on the collection c 1 . This is conditionally periodic .195) Here Yi is the ith basis cycle of the torus T".191). ••• . Therefore on some neighbourhood of the given torus T"=Me this form is exact: O. • . c/Jn) and· a non degenerate matrix of Poisson brackets (1. .•• . Thus we have (1. (1. . c/Ji< 2n (c/Ji=O is just the point x 0 ). On each level surface Me the coordinates c/J 1 .= hiidyi A dyi of the type (1. ••• . The action variables have the form. c" in a neighbourhood of the chosen level surface.198) 2n 7• Now let us set (1.194) where det {Fi. This can always be done by virtue of (1. Pn) the action variables have the form: (1. ••• . x". (1.196) We get: (1. n.17) one must do the following: the form n vanishes on the tori T"=Me. i=l. c/J 1 . The matrix of Poisson brackets takes on the form (1. and the Hamiltonian H = F 1 can be written in the form (1. . For the phase space IR 2 " with the canonical coordinates (x 1 . c/Jn coincide up to a translation with the angles ~ 1 . ~"on a whole re~on arou!_!d the given Me.~i5. .195): 1 si=§w.197). . Now let us introduce the action variables. . .
'Sn . in the form of a Hamiltonian system on £(3) with the Hamiltonian Mf M~ M~ (1. S. a) The Euler case: y1 =y 2 =y 3 =0. The equations of the rotation of a heavy rigid body with a fixed point can be represented. p)=E of a system with one degree of freedom be compact. b 11 = b22 . c 11 =c 22 . Example 3. there is an axial symmetry. in accordance with § l. y2 . it is sufficient for "complete integrability" to know one integral not dependent on the energy H.ff case.• sn) • Wi S1. (1. V. y1 =y 2 =0.41). cii=O for i=Fj. y1 . Here the appearance of the spare integral (1. Here the coefficients of the Hamiltonian (1. y3 =0. with a 1 =a 2 . • .48). but the rank of the matrix of Poisson brackets is equal to 4.204) is not connected with a symmetry of the system (see chap. Here. a) The Clebsch case (R. The Poisson brackets have the form (1. .. Example 2. Kovalevskaya's case I 1 =I 2 =I 3 /2. The simplest of them is the K irchho.M. I 2 . and the extra integral is M 3 • More complicated integrable cases (with a "hidden symmetry") have the following form.. where the Hamiltonian has the form (1.oii(s1. Novikov motion along an ndimensional torus with the frequencies ( ). Clebsch). the origin is at the point of attachment.202) H=E Now let us consider some examples of completely integrable systems with two degrees of freedom. The phase space is sixdimensional here. Let the level surface H(x. The extra integral is the square of the total angular momentum M 2 = M f + M ~ + M ~. 2 below). Therefore for integrability according to Liouville it is enough to know one integral (besides the energy integral). </>} = 1. I 3 are the principal moments of inertia of the body. Here. y 3 are the coordinates of the centre of mass.212 B. {s. I.203) H(M.205) . Well known are the following cases of integrability. p)= 211 + 2I 2 + 2I 3 +Y1P1 +YzPz +Y3P3· Here the axes of the coordinate system coincide with the principal axes of the body. Here there is an axial symmetry (with respect to the third axis). This gives the extra integral M 3 = const. I 1 . Krichever. Dubrovin. The problem of the motion of a rigid body in a perfect fluid (see above § 1) is far richer in integrable cases. b1i=0 for i=Fj.!3. c) S. b) The Lagrange case: I 1 =I 2 .201) us1 Example 1.A. just as in the Lagrange case.P.48) are like this: (1. according to Liouville's theorem. Then we have the canonical actionangle coordinates s(E)= § pdx. (1.
The Method of Separation of VariablesThe Classical Method of Integration and of Finding ActionAngle Variables The theory of completely integrable Hamiltonian systems set forth in the preceding section arose as a generalization of the HamiltonJacobi method of integrating the canonical equations. • . S=S(x..19)).210) Let us consider a canonical transformation (i. (1. p 1 .212) . c 1 =J1 2 a 1 (a 2 a 3 ) 2 +v'.. . where bj=J1(a 1 a2 a 3 )aj. v. The Hamiltonian system with the Hamiltonian H = H(x. X'=. v'. p) to coordinates (X. . We shall consider here the phase space IR 2 " with the canonical coordinates (x 1 . (1. . (1. p) has the form: (1. The HamiltonJacobi Equation. P).. .209) j (the parameters v. Pn) (see (1.208) (Jl.1 +v.211) ux' uPi (1. The supplementary integral is ~JMJ2J1(aj+ v)MjpJ + J1 2 [(a 2 a 3 ) 2 + v"]pf + . P) of the form as .pf +a 2 p~ +a 3 p~). . v" are unessential). as Pi=:. (1. .. one preserving the Poisson brackets (1. §4. Integrable Systems I 213 where the coefficients ai and cj satisfy the relation (1. v' = const).:'"..e.19)) of the coordinates (x.206) The supplementary integral has the form Mi + M~ + M~ (a. .207) b) The LyapunovSteklovKolosov case: bij = bi~ij• cij = cAj.. x".
The only method of integrating the HamiltonJacobi equation employed with success in classical analytical mechanics is the method of separation of variables.. :~)=K. ' sn will be written in the form . .. ••• .) _(.cd+S 2 (x 2 ..cn). Namely.=O Thus. (1.214) The idea of the HamiltonJacobi method consists in choosing the transformation (1. n.n. S.. p.191) xi= aK(P) } .ax. i = 1. ••. aK X'=ap. . .218) where the equations for the functions s1' . aP.p(X.. P) has the form K(X. P) satisfying the HamiltonJacobi equation H(x. . P)=H(x(X. (1.M. the system (1.214 B.216) can be written in the form (1.. +Sn(x". Pn) be a general integral of equation (1. . suppose the HamiltonJacobi equation (1.213) can be written in the form (1. i.ax. the problem of integrating the canonical equations (1. Pn will obviously be variables of action type. ) ··=~K i=1. I.(xi. (1. that det(a 2 s. and the conjugate variables X 1. ( x'.216) depending on n parameters P 1 . .c 2 )+ . X" will be the corresponding angles. where the Hamiltonian K = K(X. ax. Novikov In the new coordinates the system (1.217) where the/..• ..e. P 1> ••• . x".214) appropriately so that in the new coordinates the Hamiltonian K does not depend on X: K = K (P).aPj)#O [7]).P. •. In this case its general integral may be sought in the form S=S 1 (x 1 . (1.210) can be written in the form .210) reduces to finding a function S(x. In this case the variables P 1 .213) P. i. ( 1. .219) .n.e. (1. ••• . Dubrovin. (1. .215) P. P)..216). =c. P)). as. Krichever.) are certain functions. i=1.211).. P" (it is necessary that the function S = S(x 1 .A. .
A3 in space are defined as the roots of the equation 2 2 2 ~+~+~=1 (1. . Integrable Systems I 215 and in an obvious manner can be integrated by quadratures.222) The elliptical coordinates At. . These equations were integrated in 1861 by K. . The corresponding variables of angle type can be computed by formulas (1. en) is determined by the equation h(ct. Cn.227) and a3 <ex <at is an arbitrary constant.224) where (1. (1. Weierstrass in theta functions of two variables.220) The variables (1. K)=O. It is not hard to show [120] that integration of the equations of motion reduces to the hyperelliptic quadratures (of genus 2) dAt dt dA2 dt (1. The solution of the . The ellipsoid (1. 2] (1.223) atA a2A a3A ' where A3<a 3 < A2 <a 2 <At <at. Let the ellipsoid have the form: (1.226) ft{IJ At l/ ft{IJ A2 At' where (1. will be variables of action type. A2 . . The Hamiltonian ofthe free motion of a unit point mass on the surface ofthe ellipsoid coincides with the kinetic energy (the metric) and has the form  H. The dependence of the Hamiltonian on the new variables H (x..225) The variables have been separated.211). Example 1. 1839). . Pt At.221) if they are globally defined. p) = K (c t. . 2 [(at Ad(At a2)(At a3) 2 .A2 At + (at A2)(a2A2)(A2a3) A2 P2 .. Geodesics on an ellipsoid (Jacobi.222) is obtained for A3 =0.
and actionangle variables for it were found very recently (see below chap. 2 . Krichever.229) l/JIZ=aex' lPE=aE· The change of variables (1..j < k::. E). The question of separation of variables for Hamiltonian systems was studied intensively in the second half of the last century (see the bibliography in [92]). (1.a 3 ) (of genus 2. n (there is no summation over repeated indices). Chapter 2 Modern Ideas on the Integrability of Evolution Systems § 1. 1.230) 1 ::. _JE S(). 2 below). S.V.(}. [32]. 2).11 v2 fv ex ).M.a 1 )(). I.ex)(). To conclude this section let us note that the system of S. see chap. 2)f+(l/JIZ.P. From this the variables of angle type are found by the formulas (1. )..a 2 )(}. LeviCivita.211) as as (1. R(A.A. Therefore the invariant tori here can be extended to the complex domain and are abelian. Novikov HamiltonJacobi equation has the form: .1 !Dii\d).216 B. v 2 v R(}. Dubrovin. Commutational Representations of Evolution Systems The algebraic mechanism lying at the foundation of the procedure for integrating the KdV equation with initial conditions rapidly decreasing in x . p) is integrable by the method of separation of variables in a given coordinate system if and only if the function H satisfies the following system of equations (1.!Dii\d).228) where E = H is the energy. ex. ). The application ofthis criterion to the investigation of the integrability (via HamiltonJacobi) of Hamiltonian systems is a nontrivial problem..z) 2 ex • 2. 1.229) (). [91]): the system with Hamiltonian H(x. corresponding to the hyperelliptic Riemann surface of the root J.d 1 + 11 JE f A. Kovalevskaya mentioned above cannot be integrated by the method of separation of variables. The following criterion was established (by T.. advances in certain special classes of Hamiltonians were obtained in [27]. l/JE) is an Abel map..
1 one can always achieve that u:~ 1 =0. The first and most natural step is the generalization of equation (2. [126] and G. [127]) which by one method or another realize a reduction of the general equation (2.5) then (2. all schemes for producing new equations to which the inverse scattering method is applicable were based on various generalizations of the commutation equation (2.2) for the auxiliary linear differential operators ()2 ()3 3 () 3 L= iJx2 +u(x. if(/= 1) L = o 2 + u. It turns out that from the first m equations. n+m1. the matrix entries of which are differential poLynomials in the matrix entries uf 11 (x. J. parametrized by the constants hj.3) Beginning with this paper.L. one can successively find the vi(x. . v:. t). It was observed that this equation 4u. .1) is equivalent to the commutation condition [ a LA • ot J oL =0 ¢>=[A ot • L] (2. t) ~ UX (2. t).(X. we obtain a system of evolution equations only in the coefficients of the operator L.. Kruskal and R. (2.E.M. Shabat [125].4) with matrix (I x I) or scalar coefficients. for example.. [126]. Green. Let un and vm be constant nondegenerate diagonal matrices with distinct entries on the diagonals.2) to the case when L and A are arbitrary differential operators n ~ m ~ L= I i='O U. [75]. New physically important equations to which the inverse scattering method is applicable were discovered along just this path in the papers of V. t) and certain constants hj. [34]. Lamb [88]. I. [1]. A= iJx3 +2u ox +4ux.2) represents a family of Lax equations. A= o3 + v1 o + v 2.. The equations (2. t) ~.B. t): L>gLg. Integrable Systems I 217 which was proposed in the famous paper of C. o = ojox. For example. IX. and these are called equations of Lax type. (2. The system (2. [115].. IX= I. If we substitute the expressions obtained into the remaining n equations. [106].2) form a system of n + m matrix equations in the coefficients of the operators Land A.(X. ZakharovA. M. .2) to equations in the coefficients of the operator L.O. Miura [60] was cleared up in P. = 6uux + Uxxx (2.6) . There exist a great number of schemes (see.2). Gardner. Lax's paper [89]. fJ = 1.1 and A>gAg. . I. ."_ 1 =0... . By conjugating with a suitable matrix function g(x. obtained by equating to zero the coefficients of the iJkjiJx\ k=n.S. UX A= I i=O V.
A= o 2 + u(x.+w(x. (2.+u(x. + u. S. An important example of such systems is the "twodimensionalized" KdV equationthe KadomtsevPetviashvili (KP) equation {~u.... (2.7) Let us give a few simplest examples of equations of Lax type.8) then equation (2.u.13) Remark.2) leads to the system (2. t) at the Boussinesq equation (J. For example. we arrive for u(x. spectral theory of twodimensional operators and their applications is given in the Appendices. A review of twodimension. y. +~ uu. [43]. Novikov and equation (2. Krichever. for the twodimensional Schrodinger operator L = o~ +a. [126]. A =o~ +~u(x..2) were first discovered in [37].10) If we eliminate w from these equations...2w.P.+~u.15) . If (1= 1) L= iP +~u(x. t).+o. I..14) The operators Land A in this case are matricial and are equal to L=(~ o) a (o r) .. [117].. Example 2 ([125]).=O.V.....12) w. Boussinesq) (2. In this case L= o. (2. y) there is not a single nontrivial equation of Lax type. t)o + w(x.M.al integrable systems.11) Twodimensional systems which admit a representation of the Lax type (2.9) (2..218 B. t). y.1 ox + q 0 ' q = ± f.. (2. = r "" ± lrl 2 r.+ u(x. t). Example 1 ([142]).2) is equivalent to the family of equations 4u. The nonlinear Schrodinger equation (NLS±) ir.. (2.. Dubrovin. y.. A number of authors have shown that the usual Lax representation for twodimensional systems is possible only for operators which involve differentiation of no higher than first order with respect to one of the variables. The correct (nontrivial) analogue of the equations of Lax type for 2+ 1systems was found later in the works [98].. + 4h 1u. t). (2. = u.A..w.xx + 6uu. their algebraicgeometrical so lutions. t)o. =0.
+ 1 >a. Integrable Systems I 219 Example 3. •.~ii• a. (2..23) which.=c. coincide with the equations of motion of the socalled Toda latticethe Hamiltonian system of particles on the line with the Hamiltonian H =t2: p. [109].2) leads to the equations 2c.c.x. + 1 . Jii=b. The inverse scattering method is also applicable to certain differentialdifference systems.20) leads to the difference analogue of the KdV equation dt . + L exp(x. uii=uii(x.18) Qii=Qi.= iuii~' j>i.+ 1 . = c.+v 23 u23 x=ieu 13 u12 .26) . vii=const.)).17) The operators L and A are matricial (3 x 3) and are equal to a a L=l ax +[I. [56]) and the difference analogue of the KdV equation [96].). (2. (2.22) v.=c.=c . (2.Q].1).25) then (2. d c.~ii• (2.(c•+l c. a. A=J ax +[J.= exp(t(x.v.b ib. The equations of threewave interaction u23 . v.ai vii= ._l.2 (2. The Toda lattice ((M.Q]. =0 and for A one chooses Al/l.c. lii=a.+ 1x.21) then equation (2.19) a.20) one sets v.+ll/ln+ICn1Cn21/1n2J. t). (2. Toda) [96]. If L and A are difference operators of the form (2. (2..20) (2.24) • • If in (2.).=t[c. c. [55]. = x.. if one sets c. .(v.ai Example 4.
(2.29) 0 0 0 A.M.. In the paper [ 111] a representation of a different type than (2.)= 0. t. we obtain that on the space of solutions of (2. A.P. n1. One of the most important facts in the theory of integrable equations is the commutativity of all the equations which enter into the general hierarchy. . They have the form n L.30) is equivalent to an equation .)=O (2.. u(21t+ '') . Dubrovin. The equation Ly=A. t.220 B.y (2.U L(x. A. which are the reduction to equations in the coefficients of L of the equations (2. Un2 Un1 If we act with the operator A on the coordinates of the vector Y and use (2. . and the matricial coefficients of L are l x 1). The vector Yis the column composed of the vectors (oifoxi)y(x. Novikov With each operator L there is connected a whole hierarchy of Laxtype equations.uo u..representation can be constructed in the following manner. hItQIt (u. A. t. I..).A.) J Y(x. .2) with operators A of different orders. For the KdV equation the corresponding equations are called the "higher order KdV equations".27) the equation (:t A )y(x. lt=l and are the commutativity condition for the SturmLiouville operator with the operators ofotA. The matrix U L has the following block structure 0 1 0 0 0 0 0 0 0 UL=u.27) is equivalent to the firstorder matrix equation [ :x."' u. .t. ..2) for the higher order KdV equations was used for the first timea representation of Lax type in matrix functions depending on an additional spectral parameter.27) to express the (o"fox")y in terms oflowerorder derivatives and the parameter A.A. where A has order 2n+ 1. i=O. S. Krichever. For the general equation (2.28) where U L is an nl x nl matrix (n is the order of L.2) such a A. 1 (2.
. t. t)+ L L v. k=ls=l m dr (2. t. The compatibility of equations (2.30) implies the compatibility of (2.)=O.) and V(x.31). A. was proposed in [ 127] as a general scheme for the production of onedimensional integrable equations.. t.V(x. (2. The beginning of this program goes back to the paper [1]. A.. (2.VA(x. t. t) (A.d•.. This equation is equivalent to a system of . A. t. r=l s=l The condition of compatibility of the linear problems ( :x.(x. A. t.)= 0 is represented by the equation of zero curvature U.)) Y(x. A.Vx+[U. 2 2 4 Subsequently.)=O. (2. t)+ L L uk. V]=O.31) The matrix entries of VA depend polynomially on A. A. Integrable Systems I 221 ( :t.(x. A. in which for the integration of the sine Gordon equation an example of a rational family was first introduced. t.)=v 0 (x.A. VA J= 0.~. Hence [:X .U L• :t. (2. t). A. Let U(x. t.) be arbitrary matrix functions depending rationally on the parameter k n hk U(x.)=u 0 (x.33) Ux ( 4 VA= 2 (2.28) and (2.)) 'l'(x. and their derivatives..34) 12 UA U Uxx II.37) which must be fulfilled for all A. t)(A. the matrix entries of the ui(x.U(x.35) V(x. equation (2. A.)) 'l'(x.32) For the KdV equation the matrices UL> VA have the form [111]: (2. A.)•.32). where U and V were now arbitrary rational functions of the parameter A. t.27) and (2. t.36) ( :t.
37). in the coefficients only of U(x. [52].o (2.v0 " + [u 0 .40) 4 It is interesting that this example was first brought to the open together with the notion of zero curvature in a remarkable (but forgotten) classical paper by Rene Garnier [61]. If the poles of U and V coincide. which do not coincide.P.1+gVg1. t). [36].39) 1A.). t). Krichever..37) and destroying the gauge symmetry is called a setting of the gauge. Novikov 1 + L hk + L d. Here. In singling out some particular equations from (2. . t.37) (after the substitution x+e = x'.37) over into solutions of the same equation. t).= p. t). by changing the multiplicity of the poles of V. A choice of conditions on the matrices U (x. v] =0.. This problem reduces to describing the orbits of the coadjoint representation of the current algebra[2].' A.k> A. which may be found in the papers [2]. t. ).field ([106]. I. equal to u01 .M. t)=v 0 (x. (2. an important issue is singling out the invariant submanifolds for the equation (2. [106]. If g(x. v0 (x. [122]) 4 u~+Hu. ).= ). then these equations can be reduced to a family of equations. which are parametrized by arbitrary constants. This underdeterminacy is connected with a "gauge symmetry" ofthe equations (2. u = u(e. t+'f = x' + t') lead to the equations of a principal chiral. v0 ]. V(x. the equations (2. t.A. ).+ I ' then the equations (2. (x.222 B.t'. v]. The number of equations is one matrix equation fewer than the number of unknown matrix functions.). The simplest gauge is the pair of conditions u0 (x. Dubrovin. If U and V have one pole each. t) is an arbitrary nondegenerate matrix function. ). we will obtain a hierarchy of commuting flows associated with U(x.37) are essentially generating equations for a whole family of integrable systems. S.). [126]. Example S.38) V+g 1g. [36]. Just as in the aboveconsidered case of commutation equations for differential operators. '1) V=v(e. and also the absolute term. [124]). called a "gauge transformation". matrix equations in the unknown functions uks(x. v~=Hu. let us cite the two simplest examples. [52]. then the transformation U+g"g1 +gUg1. k r u 0 (x. in the framework of which the Hamiltonian theory of equations of zero curvature can naturally be introduced (see [30]... Leaving aside the further analysis of the questions of reduction and gauge equivalence of systems. v. (2. takes the solutions of(2.37). t)=O. These equations arise when one equates to zero all the singular terms on the lefthand side of (2.) compatible with the equations (2.37) at the points A. t.37).
(2.68) (the currents A. in the theories of quasionedimensional conductors.68) correspond. ). As in the general case of a principal chiral field.) has been constructed by one means or another. after the function 'P(e. v=G~G.37)). ~s =~So+ dl1.44) which arose in the theory of surfaces of constant negative curvature. (2. A.1 • (2.e.72) one introduces a coupling constant K. the representation (2. 11) = 'P(e. Example 6. v( e.41). of filmformation processes on crystalline substrates. i. O~U=2[v.43) In the inverse scattering·method. the desired solutions of the equations of motion are defined by the formula c(e.39). in the theory of fields) leading to equation (2. u]. 17). K ). etc.44).40) yields (2. the matrices U and V have one pole each . the road to constructing solutions of the equations (2.37) goes via the con struction with the aid of various schemes ("dressing".1 . where U and V are given by the formulas (2. Integrable Systems I 223 Here u( e. 17) are the currents of the chiral field u=G~G. u]. If in the definition of the corresponding Lagrangian (1. 17. [1] u~~=4 sin u. In the case under consideration. simultaneously gives the solutions of the equations of motion of a principal chiral field with a "multivalued additional term".37). l'f. As was remarked in [ 135].42) The last equations are Lagrangian with Lagrangian (1. As another example let us cite the SineGordon equation[I34].36) with some U. in the notation of (1. then the equations of motion can be written in terms of the currents in the form l +K lK a~v=2[v.) of functions 'P(e. as is evident from (2. to u and v). this equation belongs to the most important in contemporary mathematical physics. algebraic geometric schemes. 11.4 1) Equation (2.) which by their construction satisfy the equations (2. as will be stressed repeatedly in the following. By the number of applications (in the theory of superconductivity. V (which will automatically satisfy (2.
A. let U(x.224 B..P. An example of such an equation is (2. Example 7.M. The commutator [U. The matrix U equals . as is easy to verify. V] has poles of total multiplicity N + M.y)+((c+ y) and does not depend. I. Roch).A)= ( 2 (2. One of them was proposed in the papers [84].45) . ). ).47) (socalled KricheverNovik:ov equation). Hence the equations (2. Indeed.) and V (x. y)=(( 2c)+((c. (2.B. It was shown that here the number of equations with gauge equivalence taken into account is equal to the number of the independent variables.e. on y ((is Weierstrass's (function [11]).el") V(~._1 0 >. t. There are two ways of circumventing the obstacle mentioned. Novikov iu~ U(e.F.37) to the case of matrices U and V whose spectral parameter is defined on an algebraic curve r of genus greater than zero (the case of rational families corresponds to g=O) is obstructed by the RiemannRoch theorem (G. S. where the matrices U and V were allowed to have.) be merom orphic functions on r having poles of total multiplicity Nand M. Dubrovin. [86]. the dimension of the space of matrix functions with the same poles as U and V) equals 12 (N g+ 1) for U and 12 (M g+ 1) for V.46) It should be noted that an automatic generalization of equations (2. [86]. With gauge equivalence taken into account the number of equations is always greater than the number of variables. RiemannG. besides poles stationary with respect to x and t. This equation together with the following pair with an elliptic spectral parameter (an elliptic family) was obtained in [85]. 1]. gl poles depending on x and t in a definite fashion. t. By the RiemannRoch theorem [131] the number of independent variables (i. Here the quantity Q = iJtl>fiJc + tf> 2 is defined through tf>(c.) = ( elu 0 .. which (just as in the rational case) are the singular parts of U and Vat the stationary poles. Krichever.37) are equivalent to F(N + M g+ 1) equations. ).
where For V we have the formulas where a2a! au. where y 1 =c(x. Each solution of equation (2. wi. we shall arrive finally at equation (2. t)=(Cxxl)cx 2 2 +8<1>cxx+4cx2(o<l> &<I> ) 2cxxxCx1 (2. 2(a 1 a 2 ) (p(y.48) a rank 2 genus 1 solution of the KadomtsevPetviashvili equation (KP).48)). u2 W2 = W!x2+ p(yd+ P(Y2). t)+ y. t)+c 0 .)+A 1 ((yd+B 1 ((y 2 )+( ~ ~). . Integrable Systems I 225 U=A 1 ((Ay 1 )+B1 ((Ay 2 )+(~ ~)((). y.47) defines by the formula 8U(X. y2 = yc(x.47). and afterwards the ai (the formulas for which we shall omit here) and u (see (2. which equal 1 ux w. u.37) are equivalent to a system of equations in the functions Yi• ai. By successively eliminating from this system the variables wi. ).p(y 2 ))4. ) 2(a 2 ad and The equations (2.
). Krichever.p(yd)) (2. and p dn(A. Dubrovin.fJr.)S. Pauli). The physically most interesting example of such equations is the LandauLifshitz equation The second way of introducing a nonrational spectral parameter is based on the choice of a special form for the matrices U and V and has been successfully realized only in certain examples on elliptic curves r (g = 1)..SpSyxB"fJY...=S X sxx+S X JS.52) V= .A.. Such an operator of order 4 has the form: d d . (2. S. The second way of introducing a nonrational spectral parameter is based on the choice of a special form for the matrices U and V and has been success fully realized only in certain examples on elliptic curves r{g = 1).51) is the compatibility condition for the linear equations (2..M. ex. k)' w3 = f? sn(A.49) Equation (2.=4Vxxx+ 8vx (vxxPiv))..47) also describes the deformation of commuting linear differ ential operators of orders 4 and 6.i ~~y L w. lSI= 1.(A.47). where the u...35)..2i L a.(x. I.. [128]. and J.. k)' ..51) where Sis a threedimensional vector of unit length. k) w1 = sn(}.(A.47) reduces to the algebraic form 1 3 2 v.)S.. Novikov Equation (2. By the substitution v=((c) equation (2.226 B.. «=1 (2.p=J. .2" =(o2 + u)2 +cx(p(y2). (2. where the (2 x 2) matrices U and V are 3 U = i L w... As has been shown in the papers [22]. k)' w2 = p sn(A..50) Example 8.. equation (2. The physically most interesting example of such equations is the LandauLifshitz equation S. as is shown in [130].p(yl)) dx + dx (cx(p(y2). are the Pauli matrices (W. . is the only one of the equations of the form possessing a "hidden symmetry" which cannot be reduced to the ordinary KdV by transformations of "Miura type" w = w(c.. (2.. k) cn(A.(A. t)u.p is a diagonal matrix.u.P.)u.36)..
. [KZ). For this system L and M are matrices generally not depending on a spectral parameter. lo1'i (2.. where U and V are closely connected with the pair for the LandauLifshitz equation and. L] (2. dn(A. This is a system of particles x. z(~)=2¢' c/JW=sn(~. v~=[v. cn(A.58) Mu=( L z(x.j+(1<5u) ¢(x.. For a long time the commutational representation L=[M. . (2.57) where SJ is the Weierstrass function.:x). Another interesting example are the equations of an anisotropic 0(3)field [29] u~=[u. Integrable Systems I 227 (where sn(A.1 )dx.54) where u and v are currents: u=g~g1. k) are Jacobi's elliptic functions [11] ).Jv]. (2. Calogero) appeared to be a special case.<5. just as this latter pair.x. Example 10. k). contain an elliptic dependence on the spectral parameter. [KLWZ]. k). The parameters J a are given by the relations (2.56) of Lax type for the equations of motion of the MoserCalogero system • (J.1 Jg~g.53) Example 9. on the line with the Hamiltonian (2. (2.55) The equations (2.and fourdimensional gauge theories and relations with quantum groups [KBBT].k) • Recently this system and its discrete versions has attracted a special interest due to its deep re lations to two.59) ¢" 1 y(~)=¢'(~).x).37). MoserF.1 • The Lagrangian of this model is equal to !i'= 1 Jtr(g~g. v=g~g.54) are equivalent to (2.))bij(1bu )Y(X.Ju]. and have the form: Lu=P.
. The commutativity of the flows allows one to restrict the original system to the set of stationary points of any other flow which enters into the same hierarchy. [41]. This discovery made by S. Novikov in 1974 be came an initial point of the periodic soliton theory. Krichever. as was shown in [ 111 ]. [113]. The stationary points of the "nth higherorder analogue of the KdV equation" are described by the ordinary differential equation n "£. to conclude this section let us underline once more those basic features which are characteristic for systems to which the inverse method is applicable. [83]. An important corollary of this result is that the generic solution of the purely algebraic problem (2. As will be shown in § 5.P. [86]. t. S.37) for the pair of linear problems (2. equal to Jk=(1/k)tr Lk. It defines "finitegap" Schrodinger Operators in the sense of Spectral Theory on the line. but to have an exponential essential singularity.) and V (x. ).60) k~l which is equivalent to the condition of commutation of the SturmLiouville operator L and an operator L 1 of order 2n + 1 (2. each such equation is included in a whole hierarchy of flows which commute with it. but here the matrix entries turn out not to be meromorphic functions.57). [76]. 228 B. [115].56) are sufficient for the construction of integrals of the system (2. .M. First. Second. hk Q2k + I (U. Restriction of the KdV equation to the stationary points of the "nth higher order analogue" of the KdV equation served as the starting point in the construction of the theory of finite gap SturmLiouville operators and in the further development of algebraicgeometric integration methods which are applicable to all systems admitting commutational representations [111]. ' (2. Surveys of the various stages in the development of the theory of finite gap or "algebraicgeometric'' integration can be found in [40].) are meromorphic functions of the "spectral parameter ).P.A.61) turns out to be periodic and quasiperiodic in the variable x. all such (1 + 1) systems are equivalent to the compatibility condition (2. [45]. Meanwhile. are a completely integrable Hamilton ian system. A series of important examples will be cited and analyzed in detail in the following sections." (defined in the basic examples on a rational or elliptic curve).61) These equations. the matrices L and M admit the introduction of a spectral parameter on an elliptic curve. t.36). l. • • • ' U (2k+ ll)0 .. The examples which have been brought by far do not exhaust all systems to which the inverse scattering method is applicable. where U(x. Dubrovin. ). Novikov The representations (2. but they are insufficient for the explicit construction of angletype variables and the integration of the equations of motion in terms of theta functions.
for the KdV equation) are everywhere dense in the space of periodic solutions.. although finite formulas were not obtained. which in a number of cases (for example. calculation of coefficients) new method of "the KP deformations ofTyurin parameters" was developed in [84][86]. rank 2 and 3 (see [86]..63) The logarithmic derivative t/Jxt/1. [75]. Jl)=O there corresponds a joint eigenfunction 1/J(x. explicit expressions in terms of Riemann theta functions were found in these papers which showed that the general solutions of the equations (2. P)=Jlt/J(x. [75]. Burchnall and T. P). In the paper [26] it was shown that commuting operators of relatively prime orders are uniquely determined by the polynomial Q and the assignment of the generic points y1(x 0 ). [GR?]. For the ef fective classification (i. . Baker [9]. i.62) In the case of operators of relatively prime orders.61) describe the invariant sub manifolds of equations of Lax type. y9 (x). The problem of classifying commuting linear differential operators with scalar coefficients was considered from a purely algebraic point of view in the papers of J.4) which enters into the original Lax pair and L 1 is an auxiliary operator.t/J(x.. Lt/J(x. •. No thetafunctions! Investigation of the higher rank problem was started in [35] as a continuation of[74].61) describes an invariant submanifold of the initial equation (2. For the coefficients of commuting scalar operators of relatively prime order. (2.• . J.e. [MO?]) . P)=). where Lis the operator (2. As has already been said. When we increase the order of L 1 we obtain an everascending family of such submanifolds. Remark. unique up to proportionality. L 1 1/J(x. These equations were considered from this point of view in the papers [74]. P). (2.A. where g is the genus of the curve r (the remaining poles do not depend on x).. Integrable Systems I 229 For general Lax equations (2. It was shown by them that for such operators there can be found a polynomial Q (>. Jl).2). Unfortu nately.2) the condition which picks out the algebraic geometric solutions of these equations also has the form (2.1 is a meromorphic function on r which generically has g poles y1 (x). P = (}. of the operators L and L 1 . in which the results of the twenties were made significantly more effective and were generalized to the case of operators with matrix coefficients. Baker's program was never realized and in the course of a long time these papers were undeservedly forgotten. In 60s Dixmier wrote explicitly a pair of commuting operators of the orders 4 and 6 with polyno mial coefficients connected by the equation of the nonsingular elliptic curve. A program for effectivization of these results was proposed by H. A classification was obtained in [78]. to each point of the curve r given by the equation Q(. P). Equation (2.61).e.61) in this case are quasiperiodic functions. the equations (2.Ld=O..L) in two variables such that Q(L. Chaundy [26]. y9(x 0 ).. . All problems were solved for genus 1.
t.). where r is a divisor of the orders of the operators Land L 1 • It has been shown that a ring of commuting operators .).64). We shall speak of ''finite gap" or algebraicgeometric solutions of equations admitting a commutational representation to mean solutions for which a matrix valued function W(x.49)). such that J [ :x.Families The basic goal of this section is the presentation of a procedure for integrating equations (2. i.64) and (2.) = 0.U(x. (2.)) (2. t.66) V(x. S.) = I. t. t. 'l'(x. t.)=0.)) 'l'(x.fit is determined by a curve r and a matrix divisor of rank r.U(x. ). (which is defined on the same curve as the parameter in U and V).67) . t. Dubrovin.230 B. Let us denote by 'l'(x. i. t. which generalizes condition (2.65).) can be found.65) The equations (2. it is possible to eliminate the necessity of solving a Riemann problem and to obtain explicit formulas for the operators L and L 1 (the formula for an operator L of order 4 which commutes with an operator of 6th order was quoted above (2. t. AlgebraicGeometric Integrability of FiniteDimensional J. In individual cases.63) form an rdimensionallinear space. normalized by the condition '1'(0. are also of independent interest.) the fundamental solution matrix of the equation (. W(x. § 2. (2. ). The reconstruction of the coefficients of the operators from these data comes down to a linear Riemann problem. I. (2.A. ). ).P.M. depending meromorphically on the parameter A. t.~. Let us give a general definition of finite gap solutions for the equations of zero curvature (2.V(x.65). Novikov In the general case the solutions of (2. t.61) in a natural way. A. (2. as will be evident in the sequel.) = 0.37). ). W(x. i. A. 0.64) J [ :t. Krichever.37). which in a definition like this one play only an auxiliary role in the process of integrating the original equation (2.) = 0. i. To them one can reduce practically all interesting examples of finitedimensional Hamiltonian systems which are integrable by the inverse scattering method. (:t. as was shown in [85].
. where G does not depend on (x. which has led to the construction of finite gap solutions of rank r > 1 for twodimensional equations of the Kadomtsev Petviashvili type depending on functional parameters (see [84]. Any solution of(2.74) The curve r ifoldly covers the curve f on which the parameter A.)h(x. i. where W is an I xI matrix) unconditionally have nonzero codimension. t. t. i.)) 1/J(x. t. can be found here in the·example 7 above for the special case r = 2.(x.)'l'(x.64) for which the eigenvalues of Ware rfold degenerate identically in i. the coefficients of the characteristic equation Q(i. JL) E r. i.).70) are integrals of the equations (2.69) and (2. of the matrices U and V the matrix 'l'(x. A. t. (2.). The notion of genericity in the question at hand is not completely trivial. JL)=det(W(x. Let us normalize the vector h by requiring. If 'l'i(x. (2. They are polynomials in the matrix entries of W. Although the levels of the integrals of equation (2. t. g = 1 where this problem was solved effectively. y) = JLh(x. (2.e. Taking the normalization condition (2. t. of the matrix W W(x. in the procedure for reconstructing the matrices U and V from these data additional functional parameters appear.65).. unique up to proportionality.) (2. t. A.. pair {i.72) j simultaneously satisfies the equations ( :x . y) = 0.68) is also a solution of equation (2. t.69) Consequently. i. t. i. 73) W(x.)G{i. 0. y)=JLI/J(x. t. as is shown by the results of the papers [36]. Integrable Systems I 231 It follows from (2.). that h 1(x. y =(A. we get that W(x. A.64) that W(x. y). y) = I.66) is uniquely determined by its initial conditions and has the form 'l'(x. i.). t.the matrix entries of U(x. A. (r is obliged to be a divisor of/. y)= l or ~). t. is defined. t. A. y) are rational functions of A.).U(x. [78].e. t. a compact Riemann surface). i. r).(x. t. as was explained in the preceding section.e. t. y) = L hiO. (2.) 'l'(x. and JL. i. y = (i. Generically equation (2. t.66). In both cases all ofthe coordinates h. then it follows from (2. t.70) defines a nonsingular algebraic curve r (i. i. [85]. t. to each point of which (i. V(x.. y). t.)w 1)=0 (2. The elaboration of this direction. A.71) that the vector function 1/J(x. or. Outside the poles A. (2.71) Remark.) are the columns of the matrix 'l'(x.) is a . they are differential polynomials of the basic phase variables.) W(O. 0. t. i. A. t. (2. meromorphic functions on the curve r. t. for example. t. [86]). y) = ( :t.)) 1/J(x. t.67) into account.64). JL). y) 'l'i(x.) = 'l'(x. i. t. JL}) there corresponds an eigenvector h.)I/J(x.
).) is a rational function of A.80) We have U(x.. y) and therefore do not depend on x. (2.71) and (2. the curve r may branch). t. t.)='ii. y).M.'P. t. (2.)=(detH(x. ). t..H.W (2. t.77) Consequently. t..) the diagonal matrix Fii(x. Dubrovin. t. The poles of r/J(x..)+A. is defined on the ordinary complex plane).(x. t. The function r(x. Then one can write (2. y) =f(x. t. I.}. y)h(x. Ji. r(x. The number of poles of r is equal to the number of zeroes. A. Let us consider the matrix H(x. t. ).P.i.. From this we get that FJ. ). A. t. the vector function r/J(x. t.). 0.). t.) are the preimages of the point A.81) V(x.)Jii..)='iixqlt =HxH. t. A.the preimages of the poles A.=(A. y) are meromorphic on r. on r under the natural projection r +C. y) and hence also of r/1 is equal to N=g+ll. ).1 H. t.79) Let us denote by 'i'(x. t.J.) F(x. Since the h.F. ).1 +HF. 76) where vis the number of branch points with multiplicities counted.. It has double poles at the images of the poles of h(x. of the matrices U and V. Consequently. (We observe that above the points A. t. It follows from (2.1 +HF. t. Yi). y) is meromorphic on r outside the points P". ). (A. ). Let us further restrict ourselves to the case of a rational family (i. Novikov holomorphic function of the parameter A. 79) in the form 'il(x. welldefined function of A. t. t. (2. ).the preimages of the poles of U(x. V(x. A. Yi = (). ).where the }'. t.).) the matrix whose columns are the vectors r/J(x. t.. JJ. t. y) and zeroes at the points above which r branches. t. and is therefore a. (2. t. A formula is wellknown which connects the genus g of a smooth curve r 1foldly covering C with the number of branch points [131] 2g2=v2l.e. JJ. t.1 • Without Joss of generality we may suppose that H is regular and non degenerate at the points ). }'. t. 78) Now Jet us find the behaviour of r/J(x. (2.75) does not depend on the numbering of the }'. A.) whose columns are the vectors h(x. (2.1 =H. Hence 2N=v.) =f(x. y) coincide with the poles of h(O. y) in the neighbourhood of the points P". S.232 B. y. the number of poles of h(x. Krichever.74) that the vectors hand r/1 are proportional r/J(x.1 coincides modulo 0(1) .).) = H(x.1.1 H. and by F(x.A.
(2.) such that (2.j for which aii:t=O in (2.1 = k". A basis of the holomorphic differentials (of the first kind) on a nonsingular curve has the form . t. ). t. ka)). As has already been said.). 2°.a(X. [86]. which is what one calls the convex hull of the integer points with the coordinates i.64). ka)). t. t. t.1 (P) is a local parameter in the neighbourhood of Pa.37). (2. (2. (2. where g is the genus of the curve r. t. and qa(x. Summing up. The number of poles of 1/1 (counting multiplicity) is equal to g + 1. t. y) from the enumerated analytic properties.83) where the . If w is nondegenerate. The vector function 1/J(x. k) is a polynomial in k.65). t. Theorem 2.. t)k"s) exp (qa(X. The genus of this curve can be found conveniently with the aid of the socalled Newton polygon. The basic idea of the algebraicgeometric version of the inverse problem consists in reconstructing the vector 1/J(x. t. )')=exp(q"(. Integrable Systems I 233 with the eigenvalues of the singular part of U(x. and similarly for F. Before going over to the procedure for reconstructing 1/J. A.) at i. in a neighbourhood of Pa f(x. t. ).84). t. and. An arbitrary compact Riemann surface can be given by an equation (2.fa is a regular function in the neighbourhood of Pa. The specific nature of these properties guarantees the existence of U(x. [83].1• Thus. y) = (JO e.69) hold. [ 113]. [ 41 ]. the development of the fundamental stages of the theory of finite gap integration is mirrored in detail in [ 115] and in the surveys [ 40]. is meromorphic on r outside the points P".82) Here k".84) where i. k) is polynomial ink.1. we come to the following assertion. The genus of the curve is equal to the number of integer points lying within the Newton polygon.) and V(x. Its divisor of poles does not depend on X.1(Pa)=0.x. t. then generically the curve r is nonsingular. A consequence of the compatibility of these systems are the equations (2.fa(x. y) has the form: 1/J(X. q/x. t..1. t. y) 1°. i. in a neighbourhood of the points Pa the function 1/J(x. [ 45]. j run through some finite set of integers.1(y) of a holomorphic vector. [76]. W(x.66) and (2.F. [ 44]. ")'). t.first factor is the expansion with respect to the local parameter ka. ka. Generically this curve will be nonsingular. let us quote what we need to know from the classical algebraic geometry of Riemann surfaces and the theory of theta functions.
I. IX. 1X)))8(z+fJ+B1X). N) = LBiiNiNi. called the Jacobi variety (or Jacobian) of the curve r.9.. and odd otherwise. NeZ" Z =(z I. ••• . Novikov (2. The unit basis vectors in IC 9 and the vectors B. with the coordinates B. It is symmetric and has a positive definite imaginary part. j belong to the interior of the Newton polygon.92) Characteristics [IX.=bik• i.234 B.. fJ] is even if 4(1X. ME 7L 9 • (2. .86) By taking suitable linear combinations.85) where the i. S. M)))O(z). +N9 z9 . . b 1 . (2. a9 .91) Also often used are theta functions with characteristics 8[1X. On the curve r one can choose a basis of cycles a 1 . Dubrovin. z)=N 1 z 1 + .90) (N. M) + 2(z. fJ] for which all the coordinates of IX.z)). (BN. (2. fJ equal 0 or 1/2 are called halfperiods. k= 1. b9 with the following intersection numbers (2. fJ](z)=exp(ni((BIX. g. the quotient by which is a 2g dimensional torus T 29 =J(r).88) Uk The matrix (2..87) normalized by the conditions §W. N.t generate a lattice in 1(. ••• . This function is entire. /J) =0 (mod 2). .P. z9 ). Krichever.M. •• .89) is called the period matrix of the Riemann surface r. A halfperiod [IX. 1X)+2(z+fJ. we obtain a canonical basis of the holomorphic differentials (2. /JEIR 9 • (2. The Riemann theta function of the surface r is constructed in terms of the matrix B O(z)= L exp(ni(BN.A. Under translation of the argument by a vector of the lattice it is transformed according to the Ia~ O(z + N + BM) = exp ( ni((BM.N)+2ni(N..
95) (such divisors are called principal). if for the vector ( = (( 1 . . Abel) (2.0 is not identically equal to zero on r. Integrable Systems I 235 The Abel map of a Riemann surface r into its Jacobi variety A(P) = (A 1 (P). of the gth symmetric power of r (2. ql. .f) of its zeroes P 1 . • . niEZ. Qm (with multiplicities p 1 . dim l(DkdegDg+ 1.94) For any function f merom orphic on r there is defined the divisor (. . i. . . The corresponding divisors are called nonspecial. . . A divisor for which all ni~O is called a positive divisor D~O (or an effective divisor). Its solution (Riemann) can be given in the language of theta functions. p. . The dimension of this space is given by the RiemannRoch theorem [131].e. Let us consider the Abel map 5 of unordered sets P 1 . The divisors form an abelian group. For any divisor D the linear space /(D) associated with it is the space of meromorphic functions f on r such that (f)+D~O. ( 9) the function B(A(P). A 9 (P)) is given in the following way (2.97) is an equality. then it has on r exactly g zeroes P 1 . P 9 of points of r. . .93) can be extended linearly to the group of all divisors. For a divisor of degree greater than or equal to g. Namely..H.niPi. (2. . P.96) The Abel map (N..97) For generic divisors (2. The degree of a divisor is the number (2.93) where Q is a fixed point on r. ••• . A divisor on r is a formal integer combination of points on r. and poles Q1 . . D='[. • • . .98) The problem of inverting this map is known as the Jacobi inversion problem. . giving the solution of the inversion problem 5 Also often called the Jacobi map (translator's note). ••• . . qm respectively) (2.. P9 . .. (2.
It has the form: .. ••• .100) Here Q 11 .Jf'"=(%" 1 . 101) "' . y) of the operators of (2. The fundamental algebraicgeometric tool in the theory of finite gap linear operators and in the algebraicgeometric version of the inverse scattering method are the socalled BakerAkhiezer functions. the choice of the basis of cycles on it. and let D be a divisor on r.2.99) where . Dubrovin. . ' p n be points on a Riemann surface r of genus g. t. . (2. a= I. An npoint BakerAkhiezer function given by these data is a function: a) meromorphic on r outside the points P~. Akhiezer [5] stated examples of interpretation of such functions in the spectral theory of operators on the halfline. . The thetafunctional formula for this special case was obtained first by A. was given in [75] on the basis of a generalization of the analytic properties of Bloch eigenfunctions of operators with periodic and almost periodic coefficients [38]..g + l.65) from its analytic properties. Now we are ready to pass over to the solution of the inverse problem of reconstructing the "eigen"vector 1/J(x.I. let q 1(k). The connection with periodic problem was not known untill974.. Krichever.64). and the initial point of the Abel map. l?f the form dq~(kiP)) (normalization means (2. and such that b) for P+P~ the product 1/J(P) exp (. Novikov (2. [45].! nq. The general definition of these functions. Let p I• ••. Multi point functions are functions which have essential singularities of exponential type at several points. I. n. then 1/1 is uniquely determined up to a factor.A. [64]. is a normalized abelian dffferential of the second kind with a principal part at the point P..236 B. Singlepoint function of this kind (of the exponential type) appears as a joint eigenfunction of a pair of scalar commuting operators of relatively prime order as it was pointed out by A. !f D is a generic set of g points. (2. k. N. with the divisor of its poles and zeroes (1/1) satisfying the condition (1/!)+D~O. For a nonspecial divisor D of degree N the dimension of the linear space l?{functions with the enumerated properties is equal to N. Theorem 2.P.q~(k~(P)) is analytic. In particular. S. He mentioned that it can be computed through the thetafunctions using material of the book [8] but his program never has been carried out. Jf'"g) is the vector of Riemann constants [40].. p o(A(P)+~v<q.J_c) 1/J(P)=c exp(~. Definition. including the multipoint ones. Its in 1976 (see Appendix in [45]). 1(P ~> = 0.M. Baker in the work [9]. . let k. which depend only on the Riemann surface.) O(A(P)0 . 1(P) be local parameters in the neighbourhood of these points. qn(k) be a set of polynomials.
37).ib ij· By using the path of the proof of equation (2. there is associated a Riemann surface r. (o. P) constructed above. rational in ). t. t. W(x.). D By virtue of theorem 2. t. Changing the path of integration from Q to P leads to a translation of the arguments of the theta functions by a vector of the period lattice.103) For the proof of the theorem it is enough to consider the matrix 'i'(x. a set of polynomials qlx. Let t/J(x. M). Pi=(). in which ( has been set equal to g1 (i= L s= 1 A(P.( ~UI"· 1..2 one may choose for the t/li the functions given by formula (2.ij = Jl.2 for the construction of the inverse mapping. to each finite gap solution of rank 1 of the equations (2. So. of the points Pi).. t. From this it follows that U and V . however. do not depend on this numbering) and by virtue of the analytic properties of t/1 they are rational functions of).1. P) be the vector function whose coordinates are the t/li(x. solution of the system (2. t. t. t.91) it follows that the value of t/J{f) does not depend on the choice of the path of integration. P) (the polynomials qa(x.102) Theorem 2. There exist unique matrix functions U(x. The proof of formula (2. such that (2.e. The exponent of the exponential is translated by 2n. Integrable Systems I 237 with this condition nq. ). t.1 00) it follows that the function possesses all the necessary analytic properties. V(x. 'i'p'i'.1 00) amounts to checking that it correctly defines a function on r. N + BM. k) and a nonspecial divisor of degree g + /1.65). where g is the genus of the curve r. (=A(D)+f.)+A(Pg!+i)+X. W respectively.1. t. which generically can be considered to be nonsingular.e. From (2.) whose columns are the vectors t/l(x. t.100). From (2.). Jl.).l is the vector of b periods l?{the d(fferential ilq. This matrix depends on the numbering of the columns (i.e.3.). Here 12 is the diagonal matrix equal to p... (2. By theorem 2. the matrices (ox'il)'il'. is not a branch point. we get that det 'il #0 if).'i')'i'. the vector 2niU!'I. k) depend on x and t as parameters. i. V. In the linear space of BakerAkhiezer functions corresponding to these data let us choose an arbitrary basis t/li(x.64). let there be given the set of data enumerated above. A. exists and is unique). Let us make use of theorem 2.78) in the opposite direction.1 (2:104) are already welldefined (i. (2. the t/li also will obviously depend on the same parameters). ). These matrices are designated by U. A. Pi).
t.65) (the matrix W goes over under such a transformation into (2. I. W constructed by the formulas (2. (2. W can be carried out from the requirement that in the neighbourhoods of the P~ there should hold the congruences: aA. The formulas (2.M. ~(x. Krichever.(x. P)=g(x. In the construction of the vector t/J(x.64). Hence t/J(x. especially those corresponding to reductions of equations. The set of finite gap solutions (considered up to gauge equival ence) corresponding to a nonsingular curve r is isomorphic to a torusJ (r)the Jacobi variety qf this curve.37).~(x. corresponding to two equivalent divisors D and D. P).83) of the regular part of t/J at the point P~. S.38). P)(mod 0(1)expq~(. (2. This path will be traced in detail later on in examples of the construction of finite gap solutions of equations of the Lax type (see [ 40]. (2. P).1 05) and the definition of r (2. From the definition it follows that the components off~ possess the same analytic properties as the components of the vector function t/J(x. Here the matrix entries of U. U(x.A. t. . P) from the set of data given before theorem 2.64). where g is a nondegenerate matrix. However in a majority of cases. t.105). P) goes over into g(x.37). (2.1 04) give the most economical way of proving the theorem in general. V.x.64). (2. (2. Dubrovin. From (2.65) and their solutions up to gauge transformations (2. t.)t/J(x.65). t)t/J(x.3 there is an arbitrariness connected with the possibility of choosing different bases t/1 i in the linear space of BakerAkhiezer functions corresponding to the polar divisor D. under which t/J(x. t. P). t.P. t)f(P)~(x. P). [74]. t. To this arbitrariness. t.106) and the functions t/J and ~ define gaugeequivalent solutions. Remark. k~)) (and the analogous congruences for V and W). P)=.238 B. t. there corresponds a gauge symmetry (2.4. t. t) of the expansion (2. The matrices U. W turn out to be differential polynomials in the e. Theorem 2. We shall consider both the equations (2. t. t.104) satisfy the equations (2. V. V.70) it follows that the gauge transformations leave the curves r invariant. the explicit computation of the matrices U. ). P). relating to arbitrary rational families. Corollary. [75]. The equivalence of these divisors means that there exists a meromorphic function f(P) such that its poles coincide with D and its zeroes with D.38) of the equations (2. Novikov have poles only at the projections of the distinguished points P ~· and W only at the projections of the points on r where /l has poles. (2. [76]). (2.105) Let us consider two veCtor functions t/J(x.37).
The curve r may be represented as being glued together out of two copies of the A. where U L has the form (2. u<ln+ I) and. They satisfy an ordinary differ ential equation equivalent to the operator equation [L. for which the surfacer has singularities. (2. this equation admits the ). and in the case of singularities of the "cusp" type rational solutions are gotten.WA(x. . ). . A.109) gives a hyperelliptic curve r. which is the product of a torus with a linear space.R 2 . To the different singularity types there also correspond different solution types.33).+ 1 and the point at infinity E = oo. in the proved fashion. A (D)= A (D). As was shown above. The congruence sign means congruence modulo periods of the Jacobian of the curve r. [115]).+ 1 (A.107).] =0.32). (2. integrals of the equation (2.37) correspond rational curves with singularities.) =0. Jl) are integrals of the equations (2. in the case of singularities of the selfintersection type multisoliton solutions are obtained (see. plane along cuts joining the Eithe zeroes of the polynomial R 2 . The coefficients of the polynomial 2n+ I R2n+ I _A1 2n + I + "£.. Let us consider at greater length a number of examples connected with hyperelliptic curves.r i A2n + I .108) Hence the characteristic equation (2.. For example. The theorem just formulated means that the level set of these integrals is generically a torus. As has already been said. t..37). i i= I are polynomials in u. 70). . is a differential operator of order 2n + 1. for example. 0 •n1 +0(). A. The coefficients of the polynomial Q(). (2.))= J1 2 .. and "") ).107) where A.64). Integrable Systems I 239 The assertion of the theorem follows from the fact that by virtue of the well known theorem of Abel two divisors are equivalent if and only if deg D = deg f>. . finite gap solutions of the KdV equation are the restriction of this equation to the stationary points of one of the higherorder analogues of the KdV equation. for the KdV equation§ 3. u'. To multisoliton and rational solutions of the equations (2. (2. det(w 1. For special values of the integrals. the level manifold of these integrals is isomorphic to the generalized Jacobian of the curve.representation (2.
(x.112) The function i[J(x. I. ••• ..113) where (2. P) = ekx+k''O(k.) As the acycles it is convenient to choose the cycles situated above the forbidden bands. t.~1 ~. n are the forbidden bands 6 . equal to the lefthand side of (2.t)kJ (2.I).240 B. Let u(x. Analogously. Sturm J. Its expansion (2.C.112). P). t) = 2~'1 . Dubrovin.. Novikov Fig.F. Then a straight substitution of (2. the allowed and forbidden bands are often called the stable and unstable bands respectively (translator's note). E 2 i + 1].+ u(x. Let t/J(x. and as the bcycles. (For more details about the spectral theory of finite gap operators see § 6. . t) such that At/Jo.1 + .t/J=O(k. t. . t) and v2 (x.P. Krichever. (2.110) the segments [E 2 i_ 1 .P)=exp(kx+k 3 t)( 1+... there exist unique functions v1 (x. E2i] of the real axis. E 2 J are the allowed bands of the spectrum of L on the entire axis. cycles encompassing the segment [£ 1 . and the [E 2 i. P) be the BakerAkhiezer function having n poles (n is the genus of r) y1 .111) k=fi. t) + A.114) 6 In English.dxz +u(x).. Liouville) d2 L=. Yn and in a neighbourhood of P 0 = oo the form tjl(x.t. (2. (2.M.1 )ekx+k'r. S. i = 1.A. From the uniqueness of t/1 it follows that i[J = 0. I For real u(x) the Ei are the simple points of the spectrum of the periodic and antiperiodic problems for the SturmLiouville operator (J. t.111) gives (a.)t/J(x.111) in the neighbourhood of P 0 begins with ~ 1 k. satisfies all the requirements defining t/J except one.2 it exists and is unique. . By theorem 2.
t) = . then after an explicit calculation of the vector of Riemann constants we get (2.118) we arrive finally at the MatveevIts formula [64] for the finite gap solutions of the equation u(x..120) E. The compatibility of (2.118) where f Q(2). (2.()+canst. If we choose E 1 as the initial point for the Abel mapping. u + t E.121) In the neighbourhood of P0 we have A(P)= Uk.2 . n. .113) and (2. t. (2.).." + t st A.112) is equivalent to the KdV equation.122) and by expanding (2..St= IE. f. P)= C exp ( X E.( fP n(2l Vt() 'I' X. i= I i= I The coefficients r.)= 2n+ I n 2n +I (A. (2.1.113) we have VI= .. . Integrable Systems I 241 From (2."+ f. (2. E2k n:iVk= (2.A.123) . Q(4)= i=l d). The normalized abelian differentials of the second kind having poles at P 0 of second and fourth order are representable in the form: n A.fP un(4l)O(A(P)+Vx+ O (A(P)() ' c=c(x. (2. 3~'1 = ! U.115) jR2n+ tV) I n ). (2.2 ) (2...116) jR2n+l().2a~ In e (U x + Vt. f E2k n:iVk = Q(4)."+ I r. are determined by the normalization conditions f f E2i+ I E2i+ 1 Q(2)= Q(4l=O.A.117) E2i E2i By theorem 2." ."i 0(2)= i=l d). i= 1.119) E.E..1 +0(k.) whereR2n+I(A. t).
1) in the neighbourhood of P 0 . t.) and t/1 x(x. t) of the Bloch function t/J(x. .jR (for example.(x. t))=jR(yi(x. Hence P(x..t) Comparing the preceding equalities. Hence it is representable unambiguously in the form t/lx JRW+P(x. t.(P)t/J(x.= Yi only for one sign of . Novikov Remark 1.yi(x.. (JJ.3 associates with the algebraicgeometric spectral data r. fJ..Yi (x. A." . .WA..A. t) = . we finally find that 2jR(yi(x. Let us also indicate that theorem 2.) t/1 fi (A.2 let us indicate only that the components of the vector t/1 which figures in their formulation are given by t/J(x. Remark 2. S.242 B. Yi(x.)=P. which were first obtained in [38]. .. Its coefficients can be determined uniquely from the congruence Ant/J=JJ. t)) .1 • The function t/1 xt/1.1 we have t/lx y. for the sign plus). The divisor of poles of t/1 coincides with the divisor of poles of t/J(x. t.32) in (2 x 2) matrices. t.1 07). . 2..1 has poles at the points Yi(x. P).. t.(x. A. =the A. P).representation of the operator An of (2. t. a matrix W which is . A.t/1. besides the operators Land A.). a Yi= ox Yi· t/J A.l ). In the neighbourhood of a pole of t/Jxt/1. In a number of applications the equations of motion of the zeroes Yi(x. t)) y. . t)) i= 1 where P(x. t)).P. From the congruence there follows the exact equality Ant/J(x.. t) and the form k + O(k.. The above construction of finite gap solutions of the KdV equation was carried out with the aid of the original commutational represen tation without an explicit transition to the A.=pn+t+atpnt+ .t) +0(1). For a comparison with theorems 2.1 has a pole over A.1 and t/1 . A.representation (2...1.) is a polynomial of degree (n.) together with the point P 0 .. y9 . I. y 1 . t. t. t)yi(x.t/J(mod ekx+k''O(kt)). 0 ji'i (yi(x. turn out to be useful. It is uniquely determined by the fact that t/1 xt/J. For this let us consider the functions t/1 xt/1. Dubrovin. A. A. Krichever. P)=JJ. t.. The operator An itself can be recovered from these data in an analogous way to the construction of the operator L.M.
(2.Pi =ek(x±t> k±1 (Jo xA<e.+l/ 2 • The functions 1/1. Without retracing word for word the course of the proof of theorem 2.= 1.98) linearizes these equations on the Jacobian J (r).124) . Integrable Systems I 243 Analogously 1/1. (2. A.. Hence the hyperelliptic curve r corresponding to a finite gap solution of the sine Gordon equation has branching at the points A.. are determined uniquely by the normalization xri. . P) have n poles Y.1.) ~= n <"Y.(e.64).125) ~=x+t. '1)k±"). W(e.126) xo2 Analogously.65) that W(e. It follows from (2. t)) i and. '1)k±•). So they are proportional. outside the branch points P + and P _.p2 =eiu"'1· (2. we get The Abel isomorphism (2.(X.. In a neighbourhood of these points 1/1 f = ek(x±r> ( Jo x.= oo. a. repeating the derivation of the equations for the yj..I/J 2 have the same analytic properties.}RW+P 1(x.~ <e. k± =A. situated above A.= 0.. A. (2.. let us give the form of the BakerAkhiezer vector functions for this equation.= oo. A. 17. . oo) commutes with the singular part of Vat the point A. which were first obtained in [68].44).=O.) It follows from the definitions of 1/1 1 and 1/1 2 that a.1 ' 2 in the expansions of these functions at P +· We have eiu=Xo1_ (2.= oo. (The divisor D of degree n+ll=n+l is equal to y 1 + . 0) commutes with the singular part of U at A=0.. t.p 1 and A. A.127) In the same way it can be shown that ... 11=xt. As a second example let us consider the construction of finite gap solutions of the sineGordon equation (2. For the comp'utation of the constant of proportionality one must compare the coefficients of the term A. The components 1/J. +Yn+P+..
P)=r. Novikov (2. It can be shown analogously to theorem 2.2 that t/l . '/)·exp ( eJn<..128) (2. (e.129) Corollary.A. since by virtue of the Riemann relations and Abel's theorem 2V=2(A(P +)A(P _)):::0 (the last congruence holds inasmuch as the divisors 2P + and 2P _ are equivalent. (2. (e.130) X O(A(P)() .133) . (e. Krichever.. Picking out the real nonsingular solutions among them turns out to be comparatively easy in those cases in which the auxiliary linear problem (2. Let us find its explicit appearance.244 B.A."=const O(U+ e+ U 11( + V)O(u+ e + u'1(. J7. or (:x +U(x.131) e.132) for the Lax representation. I. The function u(e.  The factor r.126) is a solution ofthe sine Gordon equation.M..11 + Vn) (2.P. '/)is chosen via the condition that the multiplier in front of the exponential be equal to one at the point P +·Then Xo.V)" Let us note that the vector V is equal to a halfperiod. Dubrovin.. Until now we have been talking about the construction of complex solutions of nonlinear equations which admit a commutational representation of one of the enumerated forms. Here the n<f> are normalized abelian differentials with poles of second order at the points P ±• n+ _is a differential of the third kind with the residues ± 1 at P +• 2niU ±.>+'l Jn<!>+n JQ+ _ p p p ) Q Q Q O(A(P)C + u+e+ u. on r).))t/1=0. S. equals the value of this multiplier at P · After simple computations we finally arrive at the following expression for the finite gap solutions 02 (u+ e+ u'1C> (2. being the zeroes and the poles of A. t) deflned out o/(2.t. 2niV are the vectors of bperiods of these differentials.
. However for almost all nonlinear equations (the nonlinear Schrodinger equation. A.) is an antiholomorphic involution of the l plane (e. From this it follows at once that the corresponding finite gap solution u(x. . t. The description of real curve types and.135) where the cross denotes the Hermitian adjoint. is selfadjoint. A. t. A.. Since one can also subject the matrix W(x. i. of BakerAkhiezer functions leading to real solutions. (2. ). t. (2. a().. [41].)=JU(x. Integrable Systems I 245 in the case ofthe general representation (2.)) r 1. ji) and 1/J(x.)= JU + (x. the distribution on them of the poles Y.136) since both the right and lefthand sides have the same analytic properties and are equal to each other by virtue of the uniqueness of 1/J.37). pose problems of real algebraic geometry which until comparatively recently had not been worked out at all.) the corresponding linear problems are not selfadjoint. given by equation (2. The typical conditions which select physically interesting real solutions have one of the following types U(x. there is an antiholomorphic involution defined on them r: r . t. t. 70).134) U(x. or permutes them in a welldefined fashion. Here let us describe on the basis of the two examples analyzed above the two basic types of involutions on the set of divisors. a curve given by equation (2. t. although the results obtained in these papers are far from being effective).r. t. (The first serious progress in the solution of these problems in connection with the nonlinear SchrOdinger equation and the sineGordon equation was made in [68] and [28]. what is the most difficult and interesting part. whose various combinations give all the realness conditions known at present. • . P) is the BakerAkhiezer function corresponding to them. the sineGordon equation. A detailed exposition of recent achievements in real finite gap integration is given in [13]. a( A. Jl)t( I. . [44]. (2. r(P))=I/J(x. which leaves the distinguished points P11 fixed. P). the equations of the nonlinear interaction of wave packets etc. ). i. Yn is invariant with respect to the antiholomorphic involution r: (A.e.e. t) of the KdV equation is real. [46].I.109) with a real polynomial R 2 n+ 1 . Let r be a real hyperelliptic curve. t.g. If the set of points y 1 . then lf(x. a(A. .1) and J is a diagonal matrix with entries ek = ± l. the curves r.I.) to the same realness conditions. which arise in the construction of real finite gap solutions are real.))F 1..
the zero divisor of a holomorphic differential on r. Yf. . (2.246 B. Since there are n zeroes in all.140) Analogously. as is evident from the construction of 1/J.M. Yf. i.138) with poles at the points P ±. P) satisfy the condition D+t(D)=K +P + +P _. . Yf. . then xo1 · xo1 = 1. they are separate from P 0 . ji). t(P))w. on one of which the point P 0 = oo lies. t) arises only when one of then zeroes ofl/l hits P 0 (here O(Vx+ VtC)=O).e.A. Indeed. Since 1/1 has a pole on each ai it also has at least one zero. Condition (2. . Let us consider the differential 1/11(~. (2. P)ifi2<e. if we consider the differential A. Let the polar divisor of 1/Jn(~. t(D) are the zeroes of a differential ofthe third kind (2. Novikov Now let us suppose that t has n+ 1 fixed ovals a 1 .. . P)t/i 1 (~. Dubrovin. (2. Mcurves are those with g+ I components. S.137) where K is the canonical class. If the points Yi are distributed one on each oval. 7 For some authors. JL)+( I. . which need not all be ovals (translator's note). then u(x.136) 1/1 is real on real ovals.) In the case under consideration this means that all the branch points Ei are real. YiEai. Krichever. On it let us consider the antiinvolution t: {). For the finite gap solutions of the sineGordon equation to be real it is necessary that the hyperelliptic curve r be real [68]. Yf. (In real algebraic geometry the curves of genus g with g+ 1 real ovals are called Mcurves 7 .139) From (2.137) it follows that this is a meromorphic differential with its only poles at the points P ±·The residue of this differential at P + is equal to I.124) and (2. t) has no singularities. Since the sum of the residues of any meromorphic differential is equal to zero.t/12(~. a pole of u(x.137) means that D. t(P))w. But by virtue of (2. an+ 1 on r. The action of this antiinvolution on the local parameters k± 1 is such that t*(k±)= f±. Yf. I.P.
the sineGordon equation.. In the preceding section an isomorphism was set up between this space and the space (r. (2. • •• . For example.(P) without indicating ei= ±. h2 . §3. P 1 .the number of the sheet of the surface r.65). The initial "physical" coordinates on the finitedimensional space of their solutions are the values of the matrix entries of U and W at some initial point x = x 0 (or rather the values of the matrix entries of the singular parts of U and W at their poles).A.) 2g+ 1 i=l (as for the KdV equation. following [116]. by (2. we shall present the Hamiltonian theory of systems which are connected with hyperelliptic curves (see the examples of the preceding and the following sections). the Toda lattice etc. given in the form w2 = n (A. . where the constants hi arise in expressing the coefficients of the operator An in terms of u and its derivatives. . for the "higherorder KdV equations" the space of solutions of the commutativity equations for the SturmLiouville operator L and an operator An of order 2n + 1 has dimension 3n + 1. • •• . Coordinates on it are given by u(x 0 ). In the following (when this does not give rise to misunderstanding) the points Pi will be denoted as yi=A.Families In this section. hn. or in the form (the nonlinear Schrodinger equation. where A. 1'/) is real.64). Pk)=N"+k.). These systems usually come from systems of the form (2. where the matrices are (2 x 2). where r is a hyperelliptic curve. u<ln+ l) (x 0). [118]. . The Hamiltonian Theory of Hyperelliptic A. plane. • A generalisation of this theory for general systems which are connected with arbitrary curves and its relations N = 2 supersymmetric gauge theories was obtained in [KP]. Integrable Systems I 247 we get Hence.64) are ordinary differential equations. • . Coordinates in the neighbourhood of Pi are given by A. .(P)the projections of the points onto the A. and u(e." The equations (2. 1 = 0).126).
S. are any coordinates on the manifold MA then it follows from (2.141) The derivatives of Q(r) in all directions of the base space tangent to the manifolds M A are required to be globally defined meromorphic differential forms on the Riemann surface r itself (and not on the covering). is birationally isomorphic to a complex torusthe Jacobian of the curve.142) is given by the properties {Q(y. In the fundamental examples k = g (KdV. In local notation Q(r)=Q(r. then it will be said that an analytic Poisson bracket with annihilator A is given on an open region of N"+k. The dimension of NA in this case must be equal to 2k. Dubrovin. sineGordon) or k = g + 1 (NLS). i. Q) has the properties a). If the closed 2form (2. In the first case the fibre. This at once implies the proposition: . J (r). c). a) Let A be some set of functions on N"+k which depend only on the point of the base space M". c) In all of the major examples the form Q has turned out either to be meromorphic on r right from the start.248 B. The fibre of this fibration is skrthe kth symmetric power of the curve r. A)dA. on the hyperelliptic curve. the Poisson bracket of (2.e. By definition. where the image of n 1 (f}+n 1 (r) is generated by a set of cycles with vanishing pairwise intersection numbers. (2. (In the sequel A will play the role of the annihilator of the Poisson bracket. A dyi in its expansion. N r+ M A• MAc M". I.Q(yk)}=O. Let us define analytic Poisson brackets on the phase space N" +k of our systems. Krichever.yi}={J.) b) Let a merom orphic 1form Q(r) be given on the Riemann surface r or on a covering of it f+ r.).).142) that QQ contains only terms of the form dA. If A.M.A. feA.P. Coordinates on M" are given by the A. by virtue of Abel's theorem. Q(y)} =0. Yi} =Oii.143) {J.142) is nondegenerate at a "general" point of a region of the N A• where the pair (A. Novikov The space N"+k is fibred over M". or to be meromorphic on a regular covering f with an abelian monodromy group. {Q(y. which becomes nondegenerate on the manifolds N A given by the equations f = const for all f e A. the manifold of hyperelliptic curves.. b). Definition. (2.
be as indicated above.147) The transformation (2. Theorem 2.146) allows one on the basis of an analytic Poisson bracket (A..Q is a meromorphic differential. h which depend only on rEM" are in involution {g(r).101)) differentials W. which by (2. have been chosen so that thew.'). (2.86)) as a sum of a holomorphic differential w.144) Lett 1> ••• . Without loss of generality. Q) satisfying the requirements enumerated above to construct a fibration whose fibre is the quotient of ct by the lattice generated by the vectors (2. and normalized (see (2. The definition of the 1/1i (2.144) commute with each other.147).. basis of the holomorphic differentials.145) where roil has a pair of firstorder poles at points ( P. V. .5. The proof can be obtained entirely analogously to the standard Liouville procedure. h(r)} =0. and ro. of the second and third kinds respectively: (2. Like any other differential. for i ~g form a normalized § wi=bu a. Integrable Systems I 249 Any two functions g.146) are independent and have dynamics linear with respect to time. P. Let the coordinates t.146) depends on the choice of the paths between P0 and the Yi· iherefore these quantities are determined up to the lattice in ct generated by the 2g + 1 periods eq. Yt• · · · 'Yt) in general position the complex variables (2. '1s of the gradient VQ: (2. Then at a point Wo.. tt be the tangent directions toM A at a "point in general position". and for j>g wi=O. Let us consider the flows on N" H generated by Hamiltonians of the form H(r). when k ~ g one may assume that locally the coordinates t. it can be decomposed uniquely (if a basis of acycles is fixed (2. By the definition of an analytic Poisson bracket. e~.
M.. Q in all directions tangent to M A give a basis of the holomorphic differentials on r. actually already appeared earlier in the description of the real nonsingular finite gap solutions of the KdV equation. varying from 0 to 2n. For Mcurves and k=g the admissible sets of yieai form g+ 1 connected components isomorphic to the real torus T'. corresponding to the differentiation of Q with respect to socalled "action variables" canonically conjugate to the "angles" on tori. (2. which may be called the "elementary" ones for the Hamiltonian systems which interest us.93). ForMcurves and k=g+ 1 or for (M I)curves and k=g there is only one connected componenta real torus T 9 + 1 or T 9 • . H( u(r)) = H(f). ••• . we get the following theorem. Theorem 2. Let us consider real hyperelliptic curves r.Q be holomorphic. I. ur in a natural way: a) u1Q=Q.P.6. K ~I. Lemma. Only for the real theory does it make sense to discuss a special choice of the vectors ri. (Such curves are called Mcurves or M. Comparing (2. In general.1curves). (2. By no means will one always get an abelian torus. which is induced by an antiholomorphic involution u on the space M" of all hyperelliptic curves.146) with the definition of the Abel map (2.149) b) u*A=A.148) The variables t/1 i form a compact torus T&k only in the case when 2g + K= 2k. The simplest example of real structures. if and only if the derivatives V. ... Q). For this it is necessary and sufficient that k=g and that all the forms V. Novikov Let K be the number of functionally independent (modulo A) residues of the form Q.149) then sets of points Yi (i = l. The Abel transformation S9 r+ J(r) linearizes the dynamics of all Hamiltonians of the form H(r) for Poisson brackets given by generic pairs (A.A. S. 2g+K ~ 2k. Krichever. If the Poisson bracket (A. Let ur have g+ 1 or g fixed ovals on r. g + 1 or i = i 1 . The form Q and the annihilator A must also be compatible with u. i9 ) lying on pairwise distinct fixed ovals ai of the antiinvolution ur are invariant with respect to the dynamics generated by the real Hamiltonians H(f). In the second case k =g. 250 B. Q) has the properties (2. In the first case k = g or k = g + 1. These are curves with an anti holomorphic involution Ur: r+ r.. Dubrovin. .
. canonically conjugate to the coordinates on the tori Tk varying from 0 to 2n.152) represents the quantity 1 Jj=2§pdq. Theorem 2. A) + M A On the fibres of J Q ( r) there must be a superposition of a translation and an automorphism of JQ(r) as a real commutative group. the action variables Ji. Z). Their computation for the sineGordon equation was done for the first time in [46].. Integrable Systems I 251 An example of a nonelementary real structure arose in the description of the real solutions of the sineGordon equation (§ 2).151) The case (2.6 gives necessary and sufficient conditions on the bracket (Q. By a nonelementary real structure will be meant an antiinvolution r: N 1Q. For analytic Poisson brackets satisfying the elementary and non elementary realness conditions.. In particular.146). The real submanifolds in the phase space are picked out by the conditions (2.150) or (2. where P is the set of poles of the form Q. (2. Theorem 2. The effective assignment of such a structure is possible only in terms of coordinates on N connected with y by the transformation (2.151) is realized for the sineGordon and theNLS (2. A) which guarantee the linearization by the Abel mapping of the Hamiltonian flows generated by Hamiltonians H(r). A).153) 7r iii where the ai are the basis 1cycles on the tori Tk. for example.152) The proof of the theorem follows in essence from the course of the proof of Liouville's theorem and from the fact that (2. For the class of Poisson brackets under study the action variables Ji (2. For a given curve the vector '1o may take on a finite number of values.14).7. until [116] no explicit con struction of action variables for the Kovalevskaya case (see below) was known. A) which is compatible with the fibration JQ(f) N(Q. since the only method for constructing them known to the classical workers was the method of separation of the variables in the HamiltonJacobi equation. This results in a significantly greater effectiveness in the construction of action variables than in Liouville's theorem.153) acquire an important interpretation as integrals over the elements ai of the group H 1 (qP. N 1Q. are given by the formula (2.
The GardnerZakharovFaddeev bracket.)=O.)= ktO (~J qk(f). (2. ••• . In conclusion let us enumerate a number of major examples.. T+oo J Tg.. the Abel mapping linearizes all the higherorder KdVs. The remaining coefficients qk belong to the annihilator A. The coefficients of the expansion Q(f. S.A.g.156) The periods 1j of the quasiperiodic potential u(x) are defined as 1 . Tg. where (2.) is a differential ofthe second kind with a unique pole at ).J}. In this bracket the higher order KdV equations have the form ( u= al+bax ~· a)oH 1 a3 a a l=2.. To (2. Krichever.§dp=1j. ••• .159) Forb= 0 the annihilator is A={T1 .M. Magri) [94]. Example 1. and its extremals are given by the finite gap solutions constructed over r. Dubrovin. (2.154) are such that the h1(f) = q21 + 3 (f) are the Hamiltonians of the higherorder KdVs with the number l ~ 0.158) Here we have (2. A= { T1 . where dp().155) p().)d). (2. § dp().252 B. . Example 3.) is the quasimomentum. . Novikov As is well known. The Hamiltonian formalism of the stationary problem for the higherorder KdV. (2. .P.ax 3 +uax + axu. Theorem 2. ii = lim _!_ udx}. I. = 0.).157) 21t1 bj Example 2. The Magri bracket (F.8. Let us express the Hamiltonians corresponding to these flows in terms of the form Q. From [56] one can extract Q = 2ip(). j= 1.160) is a linear combination of Kruskal integrals.
The integrable case of GoryachevChaplygin in the dynamics of a rigid body with a fixed point [71].). 2 A. ••• .A 29 }. Jl. Integrable Systems I 253 The commutativity equation (2. is a parameter.162) A={A0 . By integrating Q over the "real" cycles ai. 2 A where R 5 is given by (2.161) where J is the same as in (2. Example 5. The annihilator of the bracket is generated by the first (g + 1) symmetric polynomials in the Ai. we obtain the action variables Ji.160). G is the GoryachevChaplygin integral.R(A)dA. on which the Yi = si liethe Kovalevskaya variables. The Hamiltonian structure generated by the "hidden isomor phism of Moser and Trubowitz" [136].183).61) (see [18]. This representation naturally gives rise to the Hamiltonian formalism of the system (2.184): (2. A)= y'=IIn(y'=I(A6h) 2 k 4 )+ y'=I(A81 2 )+J R 5 (A. Example 4.A2 .181). In the notation of [71] (and of the next section.). [20]).174). (2. [110] (for more details on which see the examples of the next section) between the dynamics of the KdV on the space of finite gap potentials and the Neumann systems (2. (2. The curve r is given by the equation y 2 = R 5 (). . (2.181)) we have: 1 v2 Q(r. From [6] one can extract Q = j. Example 6. In the wellknown Kovalevskaya case the action variables formerly could not be calculated. Here where H is the energy of the top. see (2. The curve r is given by the equation yz =4ttz Az (AJ HA4G)z.61) may be presented in the form (jJ =0.
i) (2. q2). J2 = pf + 2q1P1P2 + (2q2.166) (2.h2)P~ + D(ql. Novikov § 4. The integrals Ji define a curve (2.165) The commutativity equation (2. the hyperelliptic curve r of genus 2 5 Y2 = Rs(A. Dubrovin.163).254 B.)= 0 i= 1 (A. The Most Important Examples of Systems Integrable by TwoDimensional Theta Functions By the example analyzed in § 2.Krichever.163) generates a pair of commuting operators a2 L= ax2 + u(x) (2. The integrals of the system (2.P.168) in involution have the form J 1 = H.169) (by a substitution u+ u + const the constant h1 has been made zero in equations (2. (2. I.167) According to [115].164) as ( a3 a3 ) a ( a a ) As=16axs+20 uax3+ax3u +30uaxu5 u"ax +axu" (2.M.166) is equivalent to a Hamiltonian system with two degrees of freedom and with the Hamiltonian (2.4q 1q~ + 2h 2q 1q2 + 2h 3q 2. D = q~ + h 2 q~. the equation (2.A.170) . The corresponding polynomial Rs is equal to (2.168) (2.168).A. .169)). S.107) for the operators Land As on the function u may be written in the Lagrangian form 15 JAdx = 0 (2.
172) Y1 y2 These equations.174). by twodimensional theta functions.Y 2)2 (Kovalevskaya's integral).177) ) Under fulfillment of the constraint (2. Kovalevskaya's Problem. . V. = const. are linearized by the Abel transformation.) The equations (2. 1Y2=PY3ry1.123) j]2 u(x) = 2 ox 2 1n 0( U x. (2.y 3. Integrable Systems I 255 The results of § 2 indicate that coordinates on the level manifold J 1 = const.176) Let us consider the combined level surface of these integrals L=21. S. L= 2(py 1 + qy 2) + ry 3 (the angular momentum). as was already noted in § 2.Y2• YJ = qy1. (2.q 2 + Jl. the form (see § 2) I 2ijRTYJ Y1 = • (2.2Jl. (A representation of the Lax type for this system was found in [121]. Their connection with the initial variables is given by means of the socalled trace formulas (2. are the zeroes of the polynomial (2. 1~q= prJl. (2. The equations of motion of a heavy rigid body with a fixed point in Kovalevskaya's case have the form: 2p=qr. The equations on theY.170). y2 the locations of the poles of the corresponding BakerAkhiezer function.175) K = (p 2 . (2. In addition there is fulfilled the constraint condition Yi + Y~ + Y~ = l.173) Later a number of examples will be cited of systems leading to twogap potentials and integrable. The twogap potential u(x) equals (2.Y 1)2 + (2pq + Jl. j1=ry2qy3.PY2• J1. J 2 = const are given by y 1 .() + const.Y 1 (the energy). (2.170) have the following integrals H = 2(p 2 + q 2 ) + r 2 .174) r=Ji. which are equivalent to the original system have.171) where the A.176) these equations give a twodimen sional invariant submanifold of the original system (2. in the given case. as a consequence.
Novikov The Kovalevskaya variablescoordinates on this surfaceare defined in the following manner (2. I. by the results of§ 2.256 B. When n = 3 this system bears the name Neumann system. Krichever.183). As for the variables s. 2 = p ± iq. they. The Neumann system may be obtained from the Hamiltonian flow on IR 6 with the Hamiltonian (2. y2 .181).A.P. The expressions for the original variables p.186) .172). y 1 .=const. Consequently. (2. Dubrovin. (2. 1 n U(x) = a. y3 in terms of the Kovalevskaya variables are cited in [63].179) An easy computation shows that in the variables s. The equations of motion of a particle on the (n I)dimensional sphere n x2 = L i= I xf =1 (2.xf. the equations (2. The Neumann and Jacobi Problems. S..185) where u(t) is a Lagrange multiplier arising because of the imposition of the constraints (2. q.163) with R 5 equal to (2. can be defined as solutions of the equations O(A(sJ+ Ut0=0.184) have the form: (2. (2.174) have the form: (2.f:1 a. The General Garnier System. r.183) under the action of a quadratic potential 2. they will be linearized by the Abel substitution.180) (2.178) where X 1.M.182) Here A: r >l(r) and r is given by equation (2.181) These equations coincide up to a factor with the equations (2.
) i= I .) is a point of r.F. P) is regular at the point at inifinity P0 • Besides.184) was proved in [ 136]. 1 <). Let us remark that although these systems are trajectorially isomorphic. The problem of geodesics on a triaxial ellipsoid is called the Jacobi problem. > 0) 3 2 L X.186). The functions i (2.k.). <). 2 n+l· In §2 it was shown that it satisfies the equation 1/J"(x.  y= X.191) TI (}. P) = .Y. .189) takes the constructed Hamiltonian flow over into a geodesic flow on the triaxial ellipsoid (when a. The Hamiltonian H itself has the form: 1 3 H=2 L a.183).184). u(P)). (2. and those of the Neumann system (and the solutions of the Jacobi problem) in twodimensional theta functions. In the work [ 110] a trajectory isomorphism was established between the equations in x for the periodic ngap potentials of the SturmLiouville operator and the equations in t+x for the periodic trajectories of the system (2. P)= _i=~. P) = 1/J(x.183). P)t/l+(x. the corresponding Hamiltonian structures (as was shown in the preceding section) are different.188) i=l The transformation x=y. = l..jii. Hence the function 1/J(x. n (i. P) + u(x)t/J(x. (2.187) are a system of integrals in involution for (2... (2. i= 1 a. where u is the involution which exchanges the sheets of r. and hence is a rational function of . it does not depend on the choice of the sheet of r. 2 < .(x)) n 1/J(x. P) associated with a hyper elliptic curve r with real branch points ). Let us consider a BakerAkhiezer function 1/J(x.)'.1/J(x. P).. (2. Integrable Systems I 257 by restriction to the surface x 2 =1 (here xy =I X. By the same token the general solutions of the latter system can be expressed in ndimensional theta functions.).190) where P = ().. Let us denote 1/J + (x. P)t/1 + (x.)'. The full phase isomorphism of the systems for ngap potentials and the system (2. Its operation on the local parameter is u*(k) = . (2. (2.
198) Yi' = Yi(Lxiyi + aJ On the invariant plane xi= aiyi we exactly get the Neumann system on the sphere..192) that the functions tl>i(x)=t/J(x. it follows from (2. v1 = [(l/2.192) Since at the branch points t/J(x.194) The equalities (2. l/2)]. l/2)]. (0. n+ I. there holds the simple identity (2. For the Neumann system we get.196) and the characteristics [vJ of the theta functions equal v0 = [(l/2.258 B. (2. The system (2.193) satisfy the identity n L t/>f(x)= I. Dubrovin. . 2j+1 (2. 0).194) coincide (after renaming x to t) with the equations (2. V3=[(0.184). A. 11.197) v2 = [(0. P) which were obtained in § 2 thereby give the solutions of the system (2. 0).185) and (2. .i)=t/l+(x. J (. which is obtained from (2. (2. (2.191) and (2.M.i). i= l. (2. l/2)]. (l/2..i.184) can be obtained from a more general system.195) where (2. 0).2i+d0 A. (0. A. S.183). . Krichevcr.183). (2.183).2i+1Yi )1~ 2i+1 A. Novikov For an arbitrary polynomial P(A. l/2). Another interesting case is the system of anharmonic oscillators. The expressions in terms of theta functions for t/J(x. (l/2.A.P. discovered by Garnier [61] xi' =xi(LXiYi+ai).190) and (2. i= 1 (2. l/2)]. A. in particular.198) by restricting to the plane xi= Yit [62].) of degree n (whose coefficients may depend on parameters) and for arbitrary points Jl. I.
M = {M I• M2. 2 .209) ). The commutational representation for the Clebsch system was found in [119].204) (2. . The Clebsch case is the limit under contraction of the group S0{4) to E{3) of the integrable tops obtained in [97] as it was observed first in [112].202) = {:ZJ ~~ = {~~} Below we shall give the commutational representation of equations (2. If one fixes .) =[A(a).201) where H = H(M.bij.200) The Motion of a Body in an Ideal Fluid. :z (2. (2.201) in the integrable cases of Clebsch and of LyapunovSteklovKolosov (see (1. p= {Pt• P2• PJ}. A(). Integration of the Clebsch Case.203) (2.)]j(). The matrix L has the form: L =).1 P.205) Cij=C.A + L 0 . ).199) where the matrix A= (A.) has the form: All =).48). iJH · oH oH p=px iJM' M=px op +Mx oM' (2. p) is the Hamiltonian (1. As was already said in chap. 1. the equations of motion of a rigid body in an ideal fluid have the form: .a).205){ 1. Integrable Systems I 259 The Garnier system is equivalent (under a suitable choice of the parameter r) to the commutation conditions d~!). (2. M3}. LXiYi• (2.
The LyapunovSteklovKolosov Case. although its integrals hold out under this passage and coincide after it with the integrals of the Clebsch case. the Kirchhoff equations for the Clebsch case go over into the Manakov equations for the algebra so(4) (see below) after a suitable linear change of variables. [132]. An analogous linear change takes the integrable case of LyapunovSteklov Kolosov (see below) over into the integrable case of Steklov [133] for the rotation of a rigid body with an ellipsoidal cavity filled with a fluid [15]. .(b «=1 1 +b 2 +b 3 b. «=1 (2.203)(2.208) The fourth integral has the form 3 21= L b...205) will not endure the deformation of so(4) to e(3)... Kotter's representation for the equations of motion had the form: (the remaining equations have the same appearance up to a cyclic permutation of the indices). S.. under which The Lax pair (2.+B «=1 «=1 L p. F. It is of interest to remark that not only the tie indicated above exists between these systems.260 B.207) Let us set (2.(M.)up. has the form: 3 3 3 2H= L b. On the other hand.. I. [140].206) then this contraction corresponds to a passage to a limit.A. a commutational representation with an elliptic spectral parameter. with (1. (2...M.) 2 +A L p.p. Novikov a basis of the algebra so(4) with the commutation relations (2. Krichever. Dubrovin. the pair (2.M.209) The LyapunovSteklovKolosov case was integrated in the paper [70].212) for tops on so(4) diverges under the contraction.P.208) taken into account. As was found in [10]. The Clebsch case was integrated in [69]. between the lines. It is curious that in this paper for this system there was practically used. [15].. In this case the Hamiltonian.
B=J.J(A.e. where (2.f. . the system (2. for the equations of motion (with the exception of the Clebsch case) a splitting of the separatrices occurs. Let us set ei=bit(b 1 +b 2 +b 3 ). V]=il. M are elliptic functions of A. A=J2. (2. Then L. (2.211) and Jii=Jl>ii is the inertia operator 8 of the rigid body. V]w 1)=0. defined on r.. V]].f. where . as was shown in [73]. Integrable Systems I 261 Heres is the "spectral paramet~r". [B. By the results of§ 2.201) with the Hamiltonian (1. These equations. they are nonintegrable.b 1 z.21 1) has the form: [:t [B. VJ=[[A. s= . are equivalent to the Lax equation L = [L.A. V].J is the Weierstrass function corresponding to the elliptic curve r with the branch points ei.213) The commutational representation for (2. p) [1 19]. V] ]=0.210) 0 Jc. as is simple to verify..48) has a fourth integral quadratic in (M.212) [B.214) 8 Often called the inertia ten$Or (translator's note).211) is equivalent to the system [A. A Multidimensional Free Rigid Body.B. M].A[A. the general solutions can be expressed via theta functions of the Riemann surface r of the form det(A.)+(b 1 +b 2 +b 3 )/3. V]+A. ci=sbi. i. M].l Here vi=zi+aspi. 9 For n=4 in [108]. Let us note that for general diagonal metrics. The complete integra bility of this system for all n was proved in [97] 9 • As was remarked in this paper. (2. The equations of a multidimensional rigid body have the form [7]: M=[Q.A[A. M=Jil+QJ (2. The Clebsch and LyapunovSteklovKolosov cases exhaust all the possibil ities when the system (2..
[ v] is an e[ arbitrary nondegenerate (grad v] (0) # 0) odd halfperiod. .. were devoted to the transfer of this technique to some other Lie algebras. j U (P) is the period vector of the differential QVl with a double pole at P.. The solutions of the general equations (2.51 ).j)p lne(P. a...212) coincide with the equations of a motion on SO(N) with the diagonal metric which for N =4 go over under the contraction ofS0(4) to E(3) into the integrable case of Clebsch.. Krichever. are the points at infinity of this curve. P.212) were integrated (for arbitrary A and B) in the paper [39].214)).A.! A. where Jl.215) k a C. = . t)=q(xat) for the LandauLifshitz equation (2. Novikov The rigorous and abstract exposition of the Manakov theory and the direct verification of the independence of the Manakov integrals constructed (the coefficients of the characteristic polynomial (2.)iP=P. of which a survey is given in [58]. when P.? are arbitrary nonzero constants. were realized in the paper [57]. Waves in the LandauLifshitz Equation. without relying on the spectral theory of operators. The equations (2. An investigation of the dynamics of these systems not only on the Lie algebra. the vector U has the form: V= L_biU(P).212): (2.. let us look at solutions of the travellingwave type S(x. Dubrovin. the &function is constructed with respect to a curve of the form (2.M. for general diagonal matrices A and B the equations (2.214). A series of subsequent papers. As was remarked in [97]. Following [137]. I.P. but also on the whole Lie group was given in the paper [105].262 B. P. S. P.· Here the A.
=(q. A.216) turns out to be equivalent to the already analyzed Clebsch system (2. Then equation (2. They correspond to the Hamiltonian system on E(3) + E(3) with the Hamiltonian (2. q=qxM.204) M=MxJq.=vx(vxx+Ju). The pairs p. Jq)q 2 • Let us introduce the variable M=qxq+aq. M and q. (2. (2. A generalization of the LandauLifshitz equations is given by the equations u. which were considered in [139] and which describe a twosublattice system. . q(e)=v(e) and M =u x u~+au.218) have the block form Here Mand N are skewsymmetric 3 x 3 matrices corresponding in a canonical way to the vectors {Md.52) for the LandauLifshitz equations (2. N satisfy the commutation relations of E(3).q+aq x q. v=v(xat). can be integrated.216) Taking the vector product of this equality with q and using the condition q 2 = 1. N =v x v~+av.217) v. starting from the Lax pair (2. As a special case they also contain solutions of the travellingwave type.51).+Jv). The matrices L and A which enter into the commutational representation for the equations of motion of the system of (2. In [16] finite gap solutions are constructed in terms of Prym theta functions. we get ti"+Jq=A.=u x (uxx. In [139] it is shown that the equations describing the travelling wave u=u(xat). Integrable Systems I 263 We have aq=q x (q+Jq). {Nd.218) Here p(e)=u(e).
M and w are the angular momentum and angular velocity vectors. in the gravitational field being created by an arbitrary body V. Let p(x) be the mass density of the body V at its point x. Bi=I1I2I3Ii.220) are integrable. where L(A.w 2 w 1. [ '] M=I' 'I w+w . (2. the problem of the rotation of a top. the matrix u(x) = J3Gp(x)R. [w.3(x)y(x) x Iy(x) d 3 x. A]. setting yii(x) = eiikYk(x). C 3). [u 1. u2 ] = 0. From this it follows that the system (2.1 u.221) U= U.221) can be integrated in theta functions of the . A(A. where w is a skewsymmetric matrix and u is a symmetric matrix.219). . In this approximation the equations of the rotation of the top can be written in the form M= J M x w + 3Gp(x)R. i.C. has the form L = [L.B3). fixed at its centre of gravity.222) B=diag(B1. obtained in [19]. and the commutators have the form: [w 1.) = w+ A. u). (2.uw.220) can be written in the form of the system { M=[M. . where to simplify the formulas we assume that I 1+I 2 +I 3 = 0. Let us supplement the equations (2. Krichever. Let us consider. C 2 .P. Here C = diag (C 1. let R(x) be the distance from the point x to the fixation point of the top.219) with the obvious relation }'(x) = y(x) x w. and the integration procedure does not depend on the body V.220) We shall show that the equations (2.B2. following [19]. Let us write the equations in a coordinate system S rigidly connected with the top. Let us introduce.M. M = (MJ. Dubrovin.B +A.e. The system (2. v The equations (2. Let us associate to the vectors y(x) = (yi(x)).264 B. (2. and so on.C].1. the inertia operator I of the top is diagonal.W.3 (xW(x)d 3 x. I= (Iic5i)· We shall suppose the dimensions of the top to be small in comparison with the distances R(x) to the body V. (2. u] = wu.) = M+ A. where we shall orient the axes of this system along the principal axes of the inertia operator. I.w]+[u. (2. G is the gravitational constant. M= (Mii).219).221).A. w= (roii). w = (wi) skew symmetric matrices )i(x) = (yii(x)). S. The Lax representation for the system (2. The Hamiltonian has the form H = tr(tMw+uC). (2.219) v where y(x) is the unit vector of the direction going from the point x of the body to the top (written in the system S!). Novikov A Top in a Gravitational Field. further. w 2 ] = w 1w 2 .221) is Hamiltonian on the Lie algebra whose elements are pairs of 3 x 3 matrices (w.
Then u = QTaQ. Here the equations of motion can be written in the form (2.2 xit was shown that the coordinates of the particles xi(t) are the eigenvalues of a matrix depending linearly on t.). = 0 and A. corresponding to ).2 or sinh.x 2(t)) + 2f. was developed without the use of a direct connection with solutions of partialdifferential wave equations of the KdV type to which the inverse scattering method is applicable. For the system (2.J1. • 1) = 0. integrable by the method of L. where the matrix u is constructed as follows.= oo. § 5. Another application of systems of the form (2. t) = 2.221). Let us remark that elliptic solutions of the KdV equation of the form u(x.222) to a top in the field of a quadratic potential was noted in [123]).J(x.x 1 (t))+ 2~o(x.x 3 (t)) were first constructed without any connection to finitedimensional systems in the paper [47]. In the case of a finite number of vortices the corresponding system turns out to be a finitedimensional Hamiltonian system. I) j (2. JJ.221) can be expressed via the Prym theta functions (of three variables) of this covering.57) was first discovered in the paper [ 4]. A pairs.J functionpotentials x. On the surface r of genus 4 an obvious involution of the form (A. i.t. Integrable Systems I 265 Riemann surfacer given by the equation det (L().57).J(x. The construction of solutions of the equations of motion of these systems was based on the theory of Lie algebras. In twodimensional hydrodynamics the poles of the solutions correspond to the dynamics of vortices.. which will be discussed in detail in the second part of this work.1 aiixixi [19] (the possibility of applying LA pairs of type (2. where a= (aii).) acts with six fixed points.57) in degenerate cases of the Weierstrass f..JJ. and of their generalizations. const X n(xxj(t)) = det(At+ Bx. .223) (the matrix entries of A and B can be expressed explicitly in terms of the initial coordinates and momenta of the particles).221) is the proof of the integrability of the problem ofthe rotation of a rigid body about a fixed point in a Newtonian field with an arbitrary quadratic potential U = 2.e. The connection of the dynamics of poles of rational and elliptic solutions of the KdV equation to the equations of motion of the system (2. Let Q be the transition matrix from the Ssystem to the fixed system. Originally the theory of the MoserCalogero systems (2. and the phase variables of the system (2.). Pole Systems The program for research on the dynamics of poles of solutions of equations to which the inverse scattering method is applicable goes back to the article [87].) ~ (. . Therefore this surface doubly covers an elliptic curve.
58). I. where L is given in (2. twoparticle case). However till [80] both problemsthe construction of angletype variables for the system (2. Moreover .A. (2. restricted to the fixed points of the original system. Krichever. Using this connection in [77]. Novikov But in the elliptic case only the involutivity and independence of the integrals 1 Jk =ktr Lk (2. In the paper [ 4] already mentioned above it was shown that the dynamics of the poles xi(t) of solutions of the KdV equation rational in x. which may have only the form N = d(d + l)/2.224) was known. (2. Rational multisoliton solutions for the KP equation were constructed within the framework of the inverse scattering method in [99].M.2 • Here the j= I dynamics of the poles of xi(y. but also the problem of constructing elliptic solutions of the KP equationremained completely un solved (except for the simplest. r)).57) and the integration of its equations of motion in terms of theta functions. where these problems were solved. grad H = 0. At the basis of the paper [80]. The connection between the rational MoserCalogero system and rational solutions of nonlinear equations proves to be more natural in the case of two dimensional systems. J 3 (2.225) coincides with the Hamiltonian flow generated by the integral J 3 . Dubrovin.266 B. In the paper [31] the isomorphism of the two problems indicated in [77] was carried over to the elliptic case as well. J 2 =H.224).59)) involves a spectral parameter defined on the elliptic curve r. The number N is arbitrary. The necessity of restricting the flow to the stationary points grad H = 0 leads also to a restriction on the number of particles.227) of the system (2.57). t) in y and t correspond to the two commuting flows J 2 = H. S. all solutions rational in x of the KP equation (2. it was shown that the construction of [77] gives all rational solutions of the KP equation. which are obliged to have the form N u(x.t)=2 L (xxi(t)) j= I 2. This commutational representation (in contrast to (2. · As was remarked in [77]. lay the commutational representation found for the equations of motion (2.P.58).226) N which subside as lxl+ oo have the form u = 2 L: (xxi(y.
· a(A.. rp. t)) (2.A.= 0. ). and the variables on it are variables of the angle type.Jij)<t>'(xij• .235) i=O are elliptic functions with poles at the point ).232) that the function R(k. . U]. Let us define matrices vii= xi<5ij+2(1Jij)<l>(xij• A. can be represented in the form U(A. = 0.)~the coefficients of the expression n R(k.)= 0 an algebraic curve rn which nfoldly covers the original elliptic curve r. t) = g(). (2. which has essential singularities for A.) = det (2k + V().) are representable as a linear combination of the pfunction and its derivatives..)a(z) e ' . Integrable Systems I 267 with respect to this parameter the matrix entries of U and Vare BakerAkhiezer functions. ).) = oza <l>(z. t)U(A.).232) with the existence of solutions of a special form for the nonstationary Schrodinger equation with an elliptic potential.) ki. ~().229) k*i where <I>( ) ) = a(z..)xJ Consequently.t>) + 20 . The functions ri(A. t)g.. The coefficients of such an expansion are integrals of the system (2. The equations (2. The matrix U. (2. A. t).57).)z (2.P (.227) uses the connection of equation (2. Proposition.230) z..t).234) where 0 does not have an essential singularity at). A further putting to good effect of the solution of equations (2. (2. and gii = Jii exp (((A.231) and xii(t) = xi(t)xit).232) It follows from (2. = [V.. Each set of fixed values for these integrals gives by means of the equation R(k.228) vij = Jij (2 I P (xik>. A. (2..) = L ri(A.1 (A. Generically the genus of the curve which arises is equal to n.= 0.227) are equivalent to the commutational equation V. The Jacobian of the curve rn is isomorphic to the level manifold of the integrals rn. (2. A.).) (2.233} does not depend on t. <I> (z.
we get that t/J satisfies (2. Here <l>(z.9. (2. Let us formulate the final assertion.1 to zero.) is given by formula (2.229). Krichever.10.)<l>(xxi.240) has n roots in the fundamental cell with periods 2w.P.239) where U and V are the same as in (2. The explicit expressions for t/J(x. and U( 2 l. Dubrovin.2 and (xxi).238) (:t + V(). • .227). Here 0 is a Riemann theta function corresponding to the surfacer". S.230).. .9. I. The analytic properties of a on the Riemann surface rn can be clarified analogously to§ 2. we finally get Theorem 2. as a function of the variable x. t)a = 2ka. The equation (2.227).. Theorem 2. Let r n be given by the equation R(k.236) if and only if the vector a= (a 1 . where R is defined in (2. A. = 0.237) i = 1 if and only if the xi(t) satisfy equations (2.) = 0.233).236) and equating the coefficients of(xxi). y) of the nonstationary Schrodinger equation (2. Substituting it into (2.1 (x.236) is defined on the nfold covering r n of the original elliptic curve.)ekx+klr (2. These quantities can be expressed in terms . A. U (3 ) are the periods of differentials of the second kind with poles of the second and third order at the distinguished point P 0 . Novikov Theorem 2. an) satisfies the equations U().11. A function t/1 of the form (2.236) has a solution t/J of the form L n t/1 = ai(t.M. y) which were obtained in § 2 give that the poles of t/J in x coincide with the zeroes of the function O(U( 2 lx+ U( 3 lt+(). (2. The eigenfunction t/J(x.x 1 (0)) + n2 ). has simple poles at the points xi(t). Then the equation in x 0( U (2 lx + U (3)t + 0 = 0 (2.228).237). The function t/1 is a BakerAkhiezer function with a unique essential singularity of the form exp (n). t.A. Comparing with theorem 2.2 t) at an isolated preimage p 0 on r n of the point ).t))a=O. (2. t. ). . 2w' the xi(t) which satiify equation (2. . A.268 B. k..
R) of the latter are defined by the solution f(x. k 2 =A.5). the simplest example being moving wave type solutions ("solitons") decreasing as x ~ oo: 4a 2 u(x. ). = [A. The entries (T.(O) and x. R = b/a are "transition" and "reflection" coefficients respectively.)l/1+ + t22(A.. 1): • (a(A.) = b(A. t) =.).) a(A. E IR. Potentials for which b(A. such that 1/1± ~ exp{±ikx}. A. ~±(X. Such solutions have the property that they can be written as a rational combination of exponen tial functions. We introduce the following monodromy matrix T(). x ~ oo ~± ~ exp{±ikx}.) · The matrix T(A. The vector ' in (2. Here T = Ija. k 2 =A. For realvalued functions u(x) and A.): X~ 00. we have~. Up= A. E 1R are called "reflectionless potentials"..240) is arbitrary and corresponds to variables of the angle type.= ~+• 1/1.)1/f_ ~.)eikX. the famous multisoliton potentials [60] are of this type.) =/= 0 for A. = t21 (A.) = {tij (A.6)) combined with the KdV equation implies . A.) from quantum mechanics. All of the parameters in (2.) b(A. x ~ +oo Then. we consider first the rapidly decreasing case.)) T(A. In this case u(x) ~ 0 fast enough as lxl ~ oo.~.l/1. f(x. In order to explain the ideas of [60] and show the effectiveness of the algebraic Lax representation (2.)1/f_ 0 Note that det(f{A.)l/1+ + tl2(A. by definition. Integrable Systems I 269 of the integrals r.227). = 1/1+· Thus.)~ T(A. For example.) should not be confused with the unitary scattering matrix S(A.240) can be expressed by quadratures in terms of x. A.) be solutions of the equations Ll/1 = A. Let 1/f±(x.a2t)) In the general case b(A.A. 2 • ch (2a(x . L] (see (2. x ~ +oo.)) = I.)eikx + R(A.)) is the transition matrix from the basis 1/1± to the basis ~± : ~+ = tu (A. in this case f(A. the monodromy matrix f (A. of the equations (2. Their evolu tion with t generates multisoliton solutions of the KdV equation.) E SU(l.(O). E R The Lax equation L.) = 0 for A.
By definition. A= a..1 (recall that k 2 = '). Novikov da =O dt. 2. Krichever. These formulae are not reduced.+ 2(uax + axu). u'. I.6): . are purely imaginary and are total derivatives. + cklo + Ck+) Lt ). construct exact multi soliton solutions. 8u(x) 8u(x) As it was shown in [89]. Consider a solution x of the Ricatti equation as a series in k.. these aspects of the soliton theory are beyond the scope of this paper. 0. A general higherorder KdV equation has the form u. 89]): A~= ax 8/k+t. 1. Integrals of motion (K..ruskal integrals) can be obtained in the following way [60]. . S. A more compact definition can be given with the help of the resolvent [34].270 B.A. The right hand sides of the KdV equation and its higherorder analogues can be written in the following form (the Gardner form): Lt = J udx. Here Xi are polynomials in u. u". 59]. Dubrovin. e C.. . put const/k = J X2k+3(x)dx. From this. all higherorder KdV equations admit the Lax type rep resentations (2.. k = 1.75). this form leads to the Hamiltonian structure described in the Important Example of the Poisson brackets (1.' for all ').P. . u") that are total derivatives. one can obtain (using the GelfandLevitanMarchenko integral equation for inverse scattering problems as in [ 115]) a procedure for solving rapidly decreasing Cauchy problems for the KdV equation. investigate asymptotic behavior of solutions. since one can omit parts of Xi(u. u'. and so on. However.M.): The quantities X2. As it was noted in [54. The right hand sides of the higherorder KdY equations can be recursively ob tained from each other with the help of the following relation ([59.k =ax 8u~x) {Ik +C) ht + .
+ a2(x)a..~")+ The Hamiltonian property of such systems was proved from two different points of view in [2] and [34]. This operator can be factored L + ao = (ax+ v)(a". I.+ u(x) the operators At have the form k = 0. n) = I. Since it is essentially the same for operators of different order. j=l Therefore. Following [34].order operator L to an arbitrary scalar differential operator L. .. + am(X) + :~::::aj(x)a..6)): Ao = a". we consider here the case of the second order operator L =a. + 3(ua" + axu).. =a.j = (Q)+ + Laj(x)a. There is one more interesting feature of the theory of integrable systems induced by scalar operators L.v). A1 = 4a. .. Ak. Then one obtains analogues of the KdV hierarchy connected with the scalar opera tors of order greater than 2.n• L. In [34].(x). GD]. this approach was generalized by changing the second .+ u(x). but different operators At. Integrable Systems I 271 with the same operator L = a.2 + · · · + a. it was shown that for L =a. We have (see (2.. Qnfm (n. (k..].j j=l j=l one can uniquely put in correspondence its mth root Q 1 : 00 Q= Qj.i. one can define fractional powers = Qj..n = (L.l + .+ a! (x)a. To any formal operator given by a series 00 00 Q =a.+ u.m) =I and their ''positive" parts (Qj)+ = (Q"'m>+· In [34. Q1 =ax+bo(x)+ Lbi(x)a. a~ ark = [Ak. This gives a higherorder KdV equation for each value of k.. See the end of§ 1 for the discussion on two different Poisson brackets generalizing the Important Example of§ 1. 2. it is convenient to write down the general form of the operator At using fractional powers of operators on the real line.
Consider the following 2 x 2 matrix operator M = ax + ( 0v v 0 ) + ( A 0 1) 0 It is easy to see that the operator L. I. but the problem of constructing finitegap so lutions (whose theory will be considered below) cannot be reduced to anything sim pler.1 Vx.M. there exists a matrix V (x) such that (~ ~V) = y. N=2K. t) ofthe KdV equations are obtained from the zero potential (solution) by iteration of such transfonnations.272 B. Krichever. It is also possible to obtain multisoliton solutions (potentials) on the background of the finitegap or algebro geometric solutions (potentials) [Krl].v)(ax + v) is called the BacklundDarboux transfonnation depending on the parameter a 0 • In the soliton theory. is gauge equivalent to M. i=O If L.. If we put ao = 0.A. this class is called "ShabatDrinfeldSokolov" reduction (SDSreduction). in [W] it was suggested to consider cyclic chains of BacklundDarboux transfonnations.81 ). when written in the matrix fonn L = ax + ( u ~ A ~) .e. In the set of matrix operators of the first order. Also. BacklundDarboux transfonnations Ba0 : L ~ L allow us to construct new solu tions of the KdV equation from the known ones. S. All finitegap potentials can be obtained from the following cyclic condition on the oddlength chains (proof of the conjecture of Weiss): N a= Lai = 0. Novikov The transfonnation Ba0 : L ~ L =(ax.P. the change of variables from u(x) to v(x) is called the Miura transfonnation. Main results were obtained in [SV]. it transfonns the KdV equation into the MKdV equation which drastically simplifies the second Poisson bracket ( 1. from the algebraic point of view. For example. the Miura transfonnation is a transfor mation from the scalar operator L to M in the class of 2 x 2 matrix operators of the first order (see [SS] for the case of n x n matrix systems and [DS] for the generaliza tions from the Lie algebras point of view). all multisoliton po tentials u(x) and solutions u(x..~ ~) V  1 ( y. then solutions of the above cyclic conditions are oscillatortype potentials u(x) with asymptotic behavior .=O ai =I= 0. i. Therefore. Dubrovin.
••• ... Denote by T the monodromy operator. A = ax + Tax + aax +b. some basis in the space of solutions of the equation L 1{! = 'Al{!. lxl+ oo. This is also true for difference and multidimensional linear periodic operators. and with discrete spectrum which consists of the union of N + I arithmetic pro gressions with the same difference. but since at that time there was no relationship with Bloch or Hill spectral theory (in ~(IR)). j=l The spectral theory of such operators in the Hilbert space ~(IR) is based on the notion ofBlochFloquet functions (or "Bloch waves"). Below we show that BlochFloquet functions always appear in the spectral theory of periodic linear differential operators. Now let us consider the periodic case which focuses on the theory of finitegap (algebrogeometric) operators and corresponding solutions of the KdV equations. these results remained unknown to the specialists. This operator commutes with L. as a multivalued function on J.L) = 0. 'A).. let us define the monodromy matrix T('A) w. In the periodic case. This function is called a "dispersion relation". q .k= l. we can consider ).t. 'A)l{!j = L tjq(xo. . .n.L. Lame potentials)..+ u and satisfy the algebraic relation [L.+ L::>j(x)a. Integrable Systems I 273 ax)2 u(x) "' ( l + O(x).L must satisfy the relation</>(). The two dimensional analogues of these results will be discussed in the Appendix. For example.r. 1{!. By definition. Tl{!(x) = l{!(x + T). To begin with.) to be the standard basis satisfying j. Ll{!(x) = 'Al{!(x).. The monodromy operator T. Before the appearance of the KdV theory. For N = 2. xo. 'A)l{!q(x. we can take 1{! = (1{! 1 . in verse problems were not systematically considered. A]= aA. There were some examples of completely integrable potentials (for example. such operators have the form L = a. LT = T L.Ll/l(x).j. The eigenvalues 'A. We begin with general remarks. J.. the spectral theory is completely different and bears no re semblance to the scattering theory.. BlochFloquet functions are common eigenvectors of the operators L and T: T1/f(x) = 1/f(x + T) = J. Consider a linear ordinary differential operator with the periodic coefficients n L = a. J. Tl{!(x) = l{!(x + T). 2 3 ax 2 L = ax + U. Hence it maps solutions to solutions and its matrix f in the basis 1{! is defined by T(xo. In particular. commutes with L.
An important and wellknown class of examples is given by the second order operators.+ u. It includes the differential Schrodinger operator of the form L =a. difference operators. 2 x 2systems of first order operators. In some cases the Riemann surface of a Bloch.J.) of the matrix t(x 0 .). In the case of quasiperiodic coefficients ai(x}. and A. n. is a quasiperiodic function with the same group of quasiperiods as the coefficients of the operator L. In this case r was a ramified double cover of an algebraic curve of genus 1 and L was a scalar differential operator of order 4. If the complete BlochFioquet solution exists for all A E C. For a selfadjoint SchrOdinger operator L = a.+ u(x).).A. 1/1 is called a BlochFioquet solution if (log 1/1 ).i(A.l/1.) are independent of x0 and we obtain an nvalued function f. .. we have the following property: away from branching points there is a solution 1/1 such that . Then on r we have a complexvalued meromorphic function 1/f(x... r is a ramified double cover of a Riemann sphere CP 1 = S2 = CU {oo}). f.) can be easily computed. Denote by r the corresponding Riemann surface.).. Dubrovin. The eigenvalues /l.. . tben there is a Riemann surface r with the same analytic properties as in the periodic case. S. Definition 1. A.274 B. axo where the matrix Q(xo. A.P. j = l. xo.Flo quet function is a double cover of a more complicated algebraic curve of a spectral parameter A E r (the coefficients of L depend on x and A. this definition is valid only for operators with periodic coeffi cients. Strictly speaking. Novikov The matrix f satisfies the following equation (in xo): at  A = [Q(xo. u(x + T) = u(x) u(x) E IR. A.pencils whose Rie mann surfaces r are hyperelliptic (i. A) =1.) satisfying the following relations: Ll/1 = A. An operator L is called a finitegap or algebrogeometric operator if the genus of the Riemann surface r of its BlochFloquet function 1/1 is finite. The Riemann surface r together with the set of poles of 1/1 is called inverse spectral data.)]. T(xo.1.. Krichever. Tl/1 = f.e.J. A. I. Definition 2. In this case the number of poles of 1/1 is equal to the genus of r. A.M. 1/f(xo.l/1. This is always true for surfaces of finite genus.i(A. = ll. One interesting example of such behavior for genus g = 1 was considered in [GN1] in connection with the theory of commuting operators of rank 2.
and also from solid state quantum physics where the state of an electron in the lattice is determined by Bloch waves and corresponding zones are called allowed and forbidden energy bands. these zones are called stability and instability zones. Finitegap potentials generate finitegap conditionally periodic solutions of the KdV equation. It helps to justify general properties.explicit theta function formulae have a reasonable form only for very small N. where g is the number of forbidden zones of finite length (any real potential will also have one forbidden zone of infinite length). Classics considered only the case x e R where the spectrwn is discrete. u (x) = n (n + l )&:> (x) 2 . all but the first 2n + l states of a periodic problem are doubly degenerate. a typical SchrOdinger operator L with TrT = 2 cos(pT) = IL+ + IL will have infinitely many forbidden zones satisfying ITrTI > I. but it is not very useful in practice . it was not until around 1940 that some interesting features of their spectra were noticed ([11]). A difference operator L is called essentially finitegap if the number of forbidden zones is much less than 2 p(x) is the Weierstrass pfunction. For real e the quantity p ("quasimomentum") can be real or purely imaginary. But there are some cases (for example.generically. This is all that was known before the development of the KdV theory. There is a large class of periodic or quasiperiodic finitegap (algebrogeometric) potentials u(x) such that all but finitely many of their forbidden zones have length zero. 102. In general. for the Lame poten tials. The periodic KdV theory is based on the following fact.e. Integrable Systems I 275 L1/f = e1/f. This motivates the following definition. This "finitegap" property is not very interesting. Regions with p real are called allowed zones (or spectral zones) and regions with purely imaginary p are calledforbidden zones (or gaps). For example. u = const) where the lengths of all for bidden zones are contracted to zero. This generalization is essentially a ''topological closure" of the corre sponding formulae for the finitegap potentials. to the period). N]). . Then the Riemann surface r of the corresponding BlochFloquet function has genus g. Although it was known for a long time that the Lame potentials can be exactly solved formally. Here x e R + i(lJ2. This terminology comes from the spectral theory of operators in the Hilbert space ~(R). but with a very large number of zones . (ITrTI ~ 1). It is possible to generalize the theta function formulae for the infinitegenus case [ l 03]. 90.. where 2ilU2 is an imaginary period. 38. In me chanics and stability theory (with x being a time variable). where IL± = exp(±ipT}. Bel~w we sketch the theory in a somewhat simpler case of periodic difference operators L. Finitegap potentials are everywhere dense in the space of all periodic functions [MO]. In this case L is always a finitegap operator. where periodic operators are sometimes called Hill operators. This theory was created in ([Ill. and Floquet function is meaningless. the genus of the corresponding Riemann surface is equal to the number N of lattice points (i. 47.
Dubrovin.24] to the "Peierls jelly model" (a wellknown model in solid state physics originating in the 1930's). Then finitegap potentials are periodic functions satisfying the Euler Lagrange equation ~=0.e. and the problem of finding these extrema is closely connected with the theory of finitegap potentials. admits an infinitedimensional symmetry group generated by the KdV system and its higher analogues. [Ill]. (2. This analogue was developed by physicists in 1970's for the study of "charge density waves". )")be solutions of the equations of zero curvature depending periodically on x.P. t.1 (x. It is easy to see that any functional of the potential u(x). 3 the theta function formulae are very complicated and they started being used in applications only after the development of this theory. t. )")='I' (x + T. § 6. Such waves were experimentally observed in the quasionedimensional materials. This idea was first applied in [12. t.241) . However.). Let U(x. Let us consider the matrix W (x. [ 45]. Note the following property of finitegap potentials (this is important for the PeierlsFrohlich model below). This difference model is discussed below. ou(x) It is important that the functionals I actually depend only on the spectrum of L. and it was successfully generalized in [51. even in cases g = 2.M. We now return to the notation of the previous sections. [I 00]. [90]. [115] ). )")'I'. The term "finite gap solutions" is connected with just this approach. Finitegap potentials are extrema for the Kruskal functionals (defined above). t. and are independent of the remaining spectral data. Then it is natural to expect that the extrema of such functionals can be found explicitly. [64]. t. Krichever. eigenvalues of L and periodic and antiperiodic boundary conditions on the boundaries of the forbidden zones.81] for a very important difference analogue of the Peierls model. Let us briefly point out the interconnection between this approach and the algebraicgeometric one which was set forth in§ 2.In1 + · · · + cnlo. [102]. I. A.276 B. i. Integrable Systems and the AlgebraicGeometric Spectral Theory of Linear Periodic Operators The original approach to the construction of finite gap solutions of the KdV equations. S. depending only on the spectrum of L = a. Novikov the period N. )")and V(x.+ u(x). Let I = In+ c. This is indeed the case. the nonlinear Schrodinger equation and a number of others was based on the spectral theory of linear operators with periodic coefficients (see [38]. where ci are some constants.A.
t. (2. which is equivalent to the existence of a solution W (x. (2. A.20) (2.65) have the property that their corresponding Bloch function is defined on a Riemann surface of finite genus and coincides with the BakerAkhiezer jUnction. W]=[or. Thus.242) (en= en+ N # 0. vn = vn + N ). From the fact that 'l'(x + T. The finite gap periodic solutions are singled out by the condition that the genus of the surface r is finite. [I 15]) were briefly cited in §2. t. In recent years there have been discovered new remarkable applications of the algebraicgeometric spectral theory to PeierlsFrohlich problems. the periodic solutions of the equations (2. which enters into the Lax representation for the equations of the Toda lattice and for the KdV difference equation (when vn 0). (2.64). y) defined by the equations (2. In the theory of operators with periodic coefficients such functions are called Bloch functions. where it generically has essential singularities.37). and we arrive at equations (2. The spectral properties of a SturmLiouville operator with finite· gap potentials (properties obtained in the work presented in [45].) is analytic outside the poles of U and V. ). y)=Jt(y)l/t(x. Integrable Systems I 277 where Tis the period and 'I' is a solution of the equations (2.64). t.65).) is also a solution of equations (2. which are among the most fundamental ones in the theory of quasi onedimensional conductors.) irrespective of nonlinear equations. t.) which is rational in A. ).65).V. The corresponding matrices U are called finite gap potentials.36) it follows that [oxU. It is clear that the finite gap notion can be carried over verbatim to an arbitrary linear operator ox. [24]. The matrix W (x. t. This matrix is called the monodromy matrix (describing the translation by a period of the solutions of the linear equations (2.U (x.36)). W]=O. The Riemann surface r on which a Bloch function becomes single valued has infinite genus in the general case (its branch points accumulate at the poles of U and V ). = Remark. Below we shall describe these properties in greater detail and shall give sketches of the proofs of the fundamental assertions using the example of the spectral theory of the Schrodinger difference operator (2. This theory (the formulas for the variational derivatives of Kruskal integrals. In the continuous limit this model was investigated in the papers [12]. ). (2. variation with respect to the period group) was first applied to a full . y). The vector function 1/t(x.66H2. for equations (2. t.64).36). where indeed a connection between the Peierls model and the theory of.finite gap SturmLiouville operators was discovered for the first time.72) is an eigenfunction of the periodtranslation operator: 1/t(x+ T.
in which these results were carried over to the discrete Peierls model and considerably developed. The electronic energy levels are defined as the points £ 1 <£ 2 ~ ••• ~EN of the spectrum of the periodic problem for an operator L which has the form (2.. H. In the paper [33] the expressions for v. [24].278 B. were written in the form oI O(Un+ Vt+Z) v =. Date [33].n +const. Frohlich) describes the self consistent behaviour of a lattice of atoms with coordinates X 11 <x11 + 1 and electrons. corresponding to a onegap operator L. I. and the elastic energy of the lattice: Herem is the number of electrons and <l>(c11 . formulas were obtained for V11 • In [ 45] the symmetric case V11 = 0 was also studied.+ dt/1. 3. Novikov extent in [12].) = L [K(v.242) was first begun by S.. In [23] the case was considered of L<l>(c II 11 . which proves that in these papers the ground state was found. Novikov [ 45. + 2c. There are two models. In the first case it was shown that H(c11 . In addition. the speed of sound and of a charge density wave were found. The latter papers served as a starting point for the subsequent investigations [23].P. This theory was carried through to the finish in [ 45]. chap. (2.242). The PeierlsFrohlich model (R.E.P.). V11 =0. S.A.) has a unique extremal... [51]. [81].PIn II C11 ] and the more general one of where the Ik are integrals of the Toda lattice or the Langmuir lattice (J. Peierls.X11 + d. TanakaE. In the first the atom at each lattice site also possesses an internal degree of freedom: V11 • In the second model.= t/1. which at absolute zero occupy them lowest levels. Krichever. C11 =C11 +N• V11 = V11 +N· The energy of the system consists ofthe energy of the electrons. but only in the elliptic case. where C11 =exp(x11 . V11 ) is the elastic energy potential.M. v. In the continuous limit this extremal goes over into the extremals obtained in [ 13]..243) II ot O(U(n+ 1)+ Vt+Z) . [50]. § 1] and by S. The systematic construction of an algebraicgeometric BlochFloquet spectral theory on !l'2 (Z) for the Schrodinger difference operator (2. Dubrovin. Langmuir) (v 11 =0). v. For the more general models the stability of the extremals was investigated and the ground state was found. With the aid of the trace formulas for the function x..
cf>N + 1ON= cf>olJI.243) determines x"(t) up to a choice of the numbers xn(O). by virtue of the condition . the expression (analogue of the Wronskian) Cn(cf>n(}n+lcf>n+l(}n) (2. analogously to the continuous case.. cf> 1 =0. one can find the remaining values t/Jn in a recursive manner. . where. The standard basis cf>n(E) and lJ"(E) is given by the conditions cf>o= I..248) . ..oo < n < oo.247) does not depend on n.244) that for any two solutions of this equation. among them also complex ones. 2("tl vk)E". of the parameter E.X"= v" formula (2. which was rectified in the book [115].. (2. Having given arbitrary values to t/1 0 and t/1 1 .. Integrable Systems I 279 (In [45] an insignificant error was committed.244) is twodimensional. in which explicit expressions were obtained for the x" and the solutions of the difference KdV. 01 =I. trace formulas for the Xn were used. as has already been said. Cn1 (E". The idea of [79] consists in using explicit expressions for the t/1" in terms of theta functions and analogues of the "trace identities" for the"'"' in contrast to [45].242) with periodic coefficients) for all values.Oocf>t = 1.3+ · · · )• k=2 (Jn(E)= 1 L vk ) ( En1_ ( n1 En2 C1 . 00 =0.246) It easily follows from (2. Cn 1 k= 1 (2. [33]. From the recursive procedure for computing cf>n(E) and (}"(E) it follows that (for n > 0) they are polynomials in E cf>n(E)= Co C1 . The basic contemporary approach to spectral problems for periodic operators is the analysis of the analytic properties of the solutions of the equation (2.. These investigations received their completion in [79].245) The matrix W(E) of the monodromy operator f: Yn+Yn+N in the basis cl>n and On has the form: (2. in particular for cf> and lJ.244) (here L is the operator (2. Yn) were explicitly obtained whose existence had been ineffectively proved in [ 45]. Since c0=cN. we have det W = cf>N(JN+ 1. In the later paper [82] "local trace identities" en= cn(Y 1.) In the case of the Toda lattice. . For any E the space of solutions of equation (2. The difference KdV was not considered in [33].
280 B.A. Dubrovin, I.M. Krichever, S.P. Novikov
The eigenvalues w of the monodromy operator can be determined from the
characteristic equation
(2.249)
The polynomial Q has degree N and its highestorder terms have the form
2Q(E)=
Co···
1
CN1
(£N(Nr.1 V~~:)EN1
k=O
(2.250)
The spectra E l of the periodic and antiperiodic problems for L can be
determined from the equations Q(El )= ± 1, since when this holds w= ± 1.
Let us denote by E;, i= 1, ... , 2q+2, qs;,N 1, the simple points of the
spectrum of the periodic and antiperiodic problems for L, i.e. the simple roots of
the equation
(2.251)
For a pointE in general position the equation (2.249) has two roots wand w 1 •
To· each root there corresponds a unique eigenvector normalized by the
condition 1/1 0 = 1
(2.252)
This solution is called a Bloch solution.
Theorem 2.12. The twovalued function 1/1! (E) is a singlevalued meromorphic
function 1/Jn(P) on the hyperelliptic curve r, Per, corresponding to the Riemann
surface of the function jR(E}
2q+2
R(E)= fl (EE;). (2.253)
i= 1
Outside the points at infinity it has q poles y1 , .•• , Yq· In the neighbourhood of the
points at infinity
1/Ji(E)=e±""£±"(1+.~1 e.±(n)E•). (2.254)
Here the ± signs correspond to the upper and lower sheets of the surfacer (by
the upper sheet will be meant the one on which at infinity JR.""'
Eq+ 1 ).
A Bloch solution, like any other solution of equation (2.244), has the form
l/ln=l/loc/Jn+l/l 1 0n. The vector (1/1 0 , 1/1 1 ) is an eigenvector for the matrix W.
wcpN
Hence 1/1 0 =1, 1/1 1 =e;; or
(2.255)
Integrable Systems I 281
Let ej• j = 1, ... , N q 1 be the double roots of the equation
Q2 (E)= 1, i.e.
Nq1
Q2 (E)1 =C 2 r2 (E)R(E), r(E)= fl (Eej), (2.256)
j=1
c1=Co ... CN1·
At the points ej the matrix of the operator T with respect to a Bloch basis is
equal to ± 1. So it is equal to ± 1 in any other basis. Hence
(}N(E) = r(E)ON(E), <f>N + 1 (E)= r(E){j)N + 1 (E),
</>N(ej) =eN+ de)= w(ej) = ± 1. (2.257)
From (2.257) it follows that Q(E) <f>N(E) = r(E)Q(E).
Here ON, (j)N+ 1 , Qare polynomials in E. Substituting w= Q+ CrJR in (2.255)
and using the preceding equalities, we get
(2.258)
And this equality means in fact that the doublevalued function ljJ .± is a single
valued meromorphic function of"the point of r. The poles of ljJ lie at points
y 1 , •• • , yq disposed one above each of the roots of the polynomial ON( E). Indeed,
ifBN(E)=O then the two roots wu are equal to <f>N(E) and (}N+ 1 (E). In addition
</>N(E) =f. (}N + 1 (E). Consequently, for one of the roots w (i.e. on one of the sheets of
rover the root of ON(E) =0) the numerator of the fraction in (2.258) vanishes. The
pole of 1/1. lies on the second sheet.
To complete the proof of the theorem it remains to consider the behaviour of
1/1 "±(E) when E+ oo. From (2.258) it follows that 1/1 1 has a simple pole at P +. We
immediately get from (2.244) that 1/1" has a pole at P + of nth order for all n > 0.
Similarly, 1/1 n has a pole of nth order at p. This, together with the fact that w
has a pole of Nth order at P + and a zero of multiplicity N at P, implies
equation (2.254), where the x. are such that x 0 = 0, c, = exp(x. x. + 1 ). 0
The parameters y;, or rather their projections onto the E plane (which, as
earlier, we shall for brevity denote the same) have a natural spectral meaning.
Lemma 2.2. The set of points ej (the double points of the spectrum of the
periodic and antiperiodic problems for L) and {Y;} are the spectrum for the problem
(2.244) with zero boundary conditions.
Proof The surface r has two sheets above the points ej, on each of which w
takes on the same value I or  1.
As ~.one may take
 + _ 2C~ (2.259)
1/J.(ej)=l/l. (e)1/1. (ej)=  e.(eJ
ON(ej)
282 B.A. Dubrovin, I.M. Krichever, S.P. Novikov
The points )I; are zeroes of ON(E). As was already said above, when E = Y; then
for one of the signs in front of .JR
in (2.258) the numerator of the second term
vanishes. Hence for the second it is different from zero. Let this, for example, be
the plus sign. Then
(2.260)
is a nontrivial solution of equation (2.244), E = yi• with zero boundary
conditions. D
Let us consider the inverse problem. Let arbitrary distinct points E; be given,
i= 1, ... , 2q+2, and points y1 , ••• , yq on the Riemann surface r of the
function JR(E}, whose projections to the E plane are all different. In difference
problems the analogue of theorem 2.2 is the RiemannRoch theorem (131]. In
the given case it states that there exists a meromorphic function t/Jn(P) on r,
unique up to proportionality, having poles at the points y 1 , •.• , yq, an nth
order pole at p+ and an nth order zero at p. The function t/Jn(P) can be
normalized up to sign by requiring that the coefficients of E±" on the upper and
lower sheets at infinity be reciprocal. Having fixed the signs arbitrarily, we shall
denote the corresponding coefficients by e±x•. With this t/ln will have the form
(2.254) in a neighbourhood of infinity.
Lemma 2.3. The constructed functions t/1 n(P) satisfy equation (2.244), where
the coefficients of the operator L equal
(2.261)
Proof. Let us consider the function ~n=L.Pn(P)EI/In(P). It has poles at
the points y 1 , ••• , Yq· From (2.254), (2.261) it follows that~" has an (n1)st
order pole at p+ and an nth order zero at the point p. By the RiemanriRoch
theorem ~n = 0.
The method of obtaining explicit formulas for the t/ln and the coefficients of Lis
completely analogous to the continuous case. As before, let us fix a canonical set
of cycles on r. Let us denote by idp the normalized abelian differential of the
third kind with its only singularities at infinity
Eq + L
q1
a.;Eqi1
h(E)dE
idp= i=O dE (2.262)
JR(Ej JR(E)"
The coefficients !Xi are determined from the normalization conditions
§ dp=O, i= 1, ... , q. (2.263)
a,
Integrable Systems I 283
Lemma 2.4. The function 1/J.(P) has the form
• P )B(A(P)+nU +()
1/J.(P)=r.exp ( zn J. dp O(A(P)+() , (2.264)
where Uk=(1/2n) ~ dp, r. is a constant.
b,
In a neighbourhood of the point at infinity on the upper sheet we have
exp(i l dp )=Ee 10 (11 1 £ 1 + ... ), P=(E,jR)Er.
It follows from (2.254) that e2x" + 210" equals the ratio of the values of
the multipliers attached to the exponential in (2.264) taken at the images
A(P ±) = ± z0 . From (2.264) and the fact that by the Riemann bilinear relations
2z 0 = U, we get
(2.265)
where [ =' z 0 .
In a neighbourhood of p+ we have
where the coordinates Vk of the vector V are defined by the equality
wk=(Vk+O(E'))dE 1•
By expanding (2.264) in a series in E 1 , we get from (2.261)
_ !!__ O((n 1)U + f + Vt)
v.d 1n _ +I 1 .
I (2.266)
· t O(nU+(+ Vt) •=o
Theorem 2.13. The.formulas (2.265), (2.266) recover the coefficients of Lfrom
the parameters E; and rj·
It is important to note that in general formulas (2.265), (2.266) define
quasiperiodic functions c,. and vn For c. and v. to be periodic it is necessary and
sufficient that for the corresponding differential dp the conditions be fulfilled:
I I. mk .
U k =  r dp =, the mk bemg integers. (2.267)
2n b• N
As follows from the definition of the 1/1"' the parameters E;. tj determine them
up to sign.
Changes of the signs of the 1/1. result in a change of the signs of the c•.
Operators differing only in the signs of the c. need not be distinguished, since
their eigenfunctions can be trivially transferred into one another.
284 B.A. Dubrovin, I.M. Krichever, S.P. Novikov
So far we have been talking about operators L with arbitrary complex
coefficients. Now let c" and vn be real; then all of the polynomials fUE), ¢n(E),
Q(E) introduced above will be real. In addition, the periodic and the antiperiodic
problems for L are selfadjoint. Hence there are N real points in the spectrum for
each of these problems, i.e. the polynomial Q 2  1 has 2N real roots. Hence at the
extrema of the polynomial Q(E), dQ/dE = 0, one has that IQ(E)I ::0: I. The graph of
the polynomial Q has the form:
Fig. 2
The intervals [Ezit, E 2 J, on which IQ(E)I:::::; 1, are called the allowed bands 10
In these intervals Iw I= 1 and the manyvalued function p(E) defined from the
equality w = eipN is real. It is called the quasimomentum. Its differential coincides
with (2.262), where in (2.263) the ai are the cycles situated above the forbidden
bands 11 [Ezi• Ezi+ tl
Lemma 2.5. The poles Yi of the Bloch function 1/Jn(P) of a real operator L are
distributed one in each of the finite forbidden bands, E zi $; Yi::::: E zi + 1 •
Proof The poles Yi are the zeroes of the polynomial {}N(E). At these points
I =det W=¢N(yJ(JN+t(YJ
Since ¢N and (}N+ 1 are real, we have
IQ(y;)l =ti¢N(yJ+8N+ I(YJI ::0: I
and Yi lies either in a forbidden band or in one of the collapsed bands the
points ei. At the latter 1/JiP), as was shown above, has no singularities. D
Theorem 2.14. It the points E 1 , . . . , E 2q+z are real and the points y 1, ... , yq
of the corresponding Riemann surface lie one above each of the forbidden bands
[E 2 i, E 2 i+ 1], then the coefficients V11 and c" of the operator L determined by them by
virtue of theorem 2.13 are real.
Proof The necessity of the conditions of the theorem within the class of
periodic operators is given by lemma 2.5.
10 Or the stable bands (translator's note).
11 Or unstable bands (translator's note).
Integrable Systems I 285
Let the E; be real. Complex conjugation induces an antiinvolution r on
the curve r, r: P = (E, jR)+r(P) = (E, jR(E}). The fixed ovals of this anti
involution are the cycles disposed above the intervals [£ 2 ;, E 2 ;+ 1 ] and above the
infinite band which joins through infinity the points E 2 q+ 2 , £ 1 .
Let us consider ~,.(r(P)). This function possesses all the analytic properties of
t/1,.. Since t/1,. is determined by these properties up to sign, we get
li/,.(r (P)) = ± t/I,.(P). (2.268)
From (2.262) it follows that the v,. are real, and the c,. are either real or pure
imaginary (i.e. c; is real).
Let us prove that under the assumptions of the theorem c,.#O, c,.#oo. The
negation of this assertion is equivalent to one or several of the zeroes Y;(n) of the
function t/J,.(P) finding themselves at infinity on the upper or the lower sheet of r.
From (2.268) it follows that on the cycles disposed above [E 2 ;, £ 2 ;+ 1 ] t/1,. is either
real or pure imaginary. On each cycle there is one pole Y;; therefore there is also
at least one zero. Since there are q zeroes in all, the Y;(n) are distributed, like theY;,
one above each [£ 2 ;, E 2 ;+ 1 ], and hence are separate from infinity.
By virtue of what has been proved, the sign of c;
does not change under
continuous deformations of E; andY; for which the conditions of the theorem are
fulfilled. Let us deform them so that all the forbidden bands close up. Here it is
easy to check that the operator L is deformed into an operator L 0 which has
v,. = 0 and c; = const > 0. The theorem is proved. 0
To conclude the section let us examine the conditions which pick out
operators L for which v,. = 0, i.e.
(2.269)
Theorem 2.15 ([45], chap. 3, § 1). Necessary and sufficient conditions for the
operator L reconstructed by virtue()[ theorem 2.13 from the data E; and l'i to have
the .form (2.269), i.e. to have v,. = 0, are symmetry oft he points E; relative to zero and
invariance (}[the set b'J with respect to the involution on r
q+l
(E, jR(E}).( E, jR(E}), R(E)= fl (£ 2  Ef).
i= 1
The necessity of the conditions follows from the fact that if t/J ,.(P) is a Bloch
solution for the operator (2.269), then for (i;,.(E)=( l)"t/1,.(£) we have
L(i;,. =  E(i;,., (i;n+N = ( ltw(i;,..
The sufficiency of the conditions can be proved analogously to the proof of
theoreem 2.14. 0
Let us define a function t/l,.(t, P) which is meromorphic on r outside P ±, has
poles y 1 , . • . , /'q, and in a neighbourhood of p± has the form:
t/1! (t, E)=e±xn£±"(1 + Jl ~s±(n, t)E')e+'/ 2 . (2.270)
Analogously to lemma 2.) The parameters of the theta function.271) where Land A have the form (2.17.273) k=O where p(E) is the quasimomentum. the vectors U. (2. 11 =N L n=l Vn.functional H. I 0 is the mean distance between particles. Let us define the functionals lk=lk{cn. q+l H = Iq + 2 + L k=O I a.286 B.k k. Consequently. in(O). (Here I 1 is the average momentum. let us cite on the basis of this example one more aspect of the theory of finite gap integration. Dubrovin.16.LIkE\ (2.M. I. To conclude the chapter. the Xn=Xn(t) satisfy the equations of the periodic Toda lattice. These functionals have the form: where the local densities rk are polynomials.4. (2. (can be expressed by quadratures in terms of the initial data xn(O).21).20). The operator L is qgap if and only if its coefficients are extremals of the .272) satisfy the equations of the Toda lattice. S.274) . From (2. Theorem 2.its connections with variational principles for functionals of the Kruskal type (M. Theorem 2.P. L N n=l Inc". Kruskal).A. The functions O(Un+ Vt+O x"(t)=ln O(U(n+ 1)+ Vt+O +I 1 tnl 0 (2. P) and to find explicit expressions for the xn(t). V. vn} by the formula 00 ip{E)=InE. it is possible to write out an explicit formula for the 1/Jn(t. etc.250) we have 1 N 1 N 10 =. Novikov It can be proved in the standard way that such a function satisfies the linear equations (2. Krichever.
Historical Remarks Concerning Algebraic Geometry.. Its Matveev (1975). Mum ford ( 1978). 1 1 =M 1 +~M 0 . The proof of formula (2.275) in the neighbourhood of p+ and comparing with the coefficients of (2.. we get that JH=O. . Commuting difference operators. Charge Density Waves problem. McKeanvan Moerbeke ( 1975). Dzyaloshinsky. First discovery for KdV and periodic !dimensional Schrodinger operator.Henon. /2 = .275) In fact. Gordunin.276) From the first q + I equalities fhe coefficients lk will be expressed via the M k• k5oq+ I.Novikov (1974). applications in quantum solid state physics .) (jJ O• s2 = i~i E.kM. hyperelliptic algebraic curves. Novikov. Lax (1975). Brazovsky. Date ( 1976).275). (2. Toda lattice and periodic difference Schri:idinger operators .2 in the expansion (2.M z + s~ (jJ 1 +CIs. Tanaka.Krichever. Equating the coefficients of Eq.273). Flashka. k1 lk= Mk+ L /3. Krichever ( 19801982).273) and in (2.274) can be obtained in an entirely analogous way to the proof of its continuous version [38].Belokolos.277) where the ak are the symmetric polynomials in the E. Integrable Systems I 287 This assertion follows from the formula (2. Peierls functionals. Manakov (1974). i=O (2. by expanding (2.. we get l0 =M 0 .Ei. Hamiltonian Systems and Spectral Theory I. s 1 =~E. DubrovinNovikov (1974).
Krichever. Theory of special solvable operators. Einolsky. Wiegmann ( 19961997).Krichever. Novi kovVeselov hierarchy and deformations of surfaces in 3dim space preserving Wiimore functional. Veselov ( 1981 ).Hitchin (1988). Purely potential SchrOdinger operators and Prym Varieties. algebraic geometry and 2dim Schrodinger operators. Periodic Darboux chains. Potemin (1989). Chodnoovski G. Veselov. Talon. 5. commuting ordinary differential op erators . Taimanov (1986). Modified NovikovVeselov hierar chy. Riemanian Geometry and integrabil ity. Lipan. Results of Dubrovin ( 1981 ).Krichever ( 19751976). 7. Krichever. Dubro vin. Mokhov ( 1983).Novikov. 6.P. Extension to 2dimensional system. Periodic Schrodinger operators and approximation of the generic operators by the algebraicgeometrical (finitegap) operators: MarchenkoOstrovsky (1977). purely algebraic construc tion. Biley. YangMills fields and their deformations. 3a. Krichever (1988). McKean (1977).M.Krichever.Krichever (1985).Krichever. 3. Solutions of rank I ~ 2. Phong ( 1996). Com pactification of BlochFlouque manifolds of eigenfunctions for all energy levels  Feldman. Veselov (19821984). Arbarello. commuting ordinary differential operators .Dubrovin. Novikov (1996). Segal.Krichever (1989). Veselov (1984). Miwa. elliptic cases.Dubrovin. Hamiltonian theory of the integrable systems (finitedimensional and field the oretical) . connection with quantum groups . Dubrovin. Grinevich ( 1982).Airault. Novikov. Krichever (1995). Zabrodin. Mokhov (19881990).Sato. Belokolos. Spectral theory of one energy level and approximation of the generic 2d potential SchrOdinger operators by algebraicgeometrical. Etingof (1997) 4. Elliptic finitegap potentials . Smirnov. Zabrodin (1995). Soliton theory. Choodnovski D. Grassmanians and Representation theory . Topological quantum field the .Weiss( 1988). (1978). Novikov 1a. Styrkas (1993). Trubowitz (1989). Hitchin's hierarchy of deformations of stable holomorphic vector bundles on Riemann surfaces. Elliptic case. Veselov. Novikov ( 19781980). all algebraic curves.BogoyavlenskiNovikov ( 1975). 2a. Novikov ( 1974). Grinevich ( 19851986). ShabatVeselov ( 1992). Ferapontov (1990). Tsarev (1985). KdV and KP theory and poles systems . Connection to SeibergWitten solution of N = 2 supersymmetric gauge theories . Novikov. Babelon. Knorrer. Man inLebedev ( 1979).Bogdanov (1987).288 B. Taimanov (19911993)..Cha 1ikh. Dubrovin. Krichever. S. Novikov. KP hierarchy. Taimanov (19961997). 2. KP hierarchy.Man akov (1976).Taimanov (19951996). Jimbo ( 1979). Novikov (1976). FlashkaMcLughlin ( 1976).Kriche ver (1980). Moser. Krichever's solutions and Novikov's conjecture concerning classi cal RiemannShottky problem. Hamiltonian hydrodynamic type systems. Difference 2dim operators. Wilson ( 1980).A. Adler ( 1978). Novikov (19831984). de Conchini ( 1984) and final solution by Shiota ( 1986). Krichever (1978). KP hierarchy and deformations of semi stable vector bundles over algebaric curves. Gelfand Dikii ( 19771979). McKeanTrubowitz ( 1977). TreibichVerdier (1990). I.
Mokhov. Sadov (1990). KricheverNovikov Fourier type basises and almost graded algebras. bosonic strings and harmonic analysis on the Riemann surfaces. the KdV equation (2. Novikov (19871990). Dubrovin (19901992). For example. Soliton theory.2) At the same time.22).Krichever ( 1996). Nutku ( 1995). algebraicgeometrical solutions .1) is obtained from (2. It corresponds to solutions of KP that do not depend on t. Genus 1 surfaces (tori) of constant mean curvature in Euclidian 3dim space and finitegap solutions .Krichever (19901992). the sineGordon equation may be considered as another reduction of the same 2d. (A. Bobenko ( 1991 1992) Appendices Appendix A AlgebraicGeometrical Integration of (2 + 1)Systems Strange as it may seem. . This slightly exaggerated statement reflects the fol lowing interrelations between spatially two.Wente. the algebraicgeometrical construction of periodic and quasiperiodic solutions of multidimensional (2 + I)integrable systems is to some extent "simpler" than the construction of the finitegap solutions of onedimensional evolution integrable equations. Another example is the Toda lattice (2. All onedimensional evolution integrable equations can be considered as a reduc tion of their twodimensional analogs.11) is the reduction of the KP equation as well. Schti helmeier (1990).Krichever.) The Boussi nesq equation (2. it corresponds to N = 2 periodic (/Jn+N = (/Jn solutions of (A.1) corresponding to the case where there is no dependence on the y variable.7) is the reduction of the KP equation 3 4Uyy = ( Ut.Toda lattice equations.12) by elimination of w.2). Hitch in. Jaffe.23) which is the reduction of the 2dToda lattice equations (A. Pinkal. Ferapontov.and onedimensional integrable sys tems. northogonal coordinate systems in Euclidean spaces and inverse scattering Zakharov (1996). 9. Sonora (1988). Namely. Solvable cases of the associativity equations . (Note that the equation (A. (2. Integrable Systems I 289 ory and geometry. Valter (1986). Klimek (1988).2UUx 3 1 + 4Uxxx )x . 8.
y. Q) = exp ( 7.)) function 1/l(t. lead to solutions of "unrestricted" twodimensional integrable systems...5) corresponding to the unique normalized J.} (that are the coefficients of the polynomials qa).. ....a = 0. . i ~ 1. S.ail... (if all of them are distinct).a + Z) 1/l(t. 1( Q) in neighborhoods of the punctures.a• (A.. dil. y.4) where: a) O(z) = O(ziB) is the Riemann thetafunction corresponding to the matrix B of bperiods of normalized holomorphic differentials dw. Krichever.a U. b) !1. (A.....290 B. Let r be a nonsingular algebraic curve of genus g with N punctures Pa and fixed local parameters k. such that: (i) the function 1/1 (as a function of the variable Q which is a point of f) is meromorphic everywhere except for the points Pa and has at most simple poles at the points Yl· .ati.. In these new variables the thetafunctional formula (2.1 00) becomes "" ) O(A(P) + L. . I.. ..P. . tN. Q)...a(P) is an abelian integral f1. g on r. Dubrovin. For any set of points YI· .a(P) O(A(P) + Z) • (A. that are constructed for the ar bitrary Riemann surface with fixed local coordinates at neighborhoods of the punc tures.A.t. •. in general position there exists a unique (up to constant factor c<ta. t = <ta. We begin the presentation of the general algebraicgeometrical construction with the definition of the most basic multipoint and multivariable BakerAkhiezer func tion...a(P) = JP dili.. .6) ht meromorphic differential on r with the only pole of the form ..M. all these interrelations are a corol lary of the fact that the BakerAkhiezer functions. N. Novikov From the "finitegap" theory point of view. i = 1.). The algebraicgeometrical solutions of the equations that are the reductions of these two dimensional equations correspond to reductions or specifications of the algebraic geometrical data.. . a= 1. (ii) in a neighborhood of the point Pa the function 1/1 has the form Note that this is the same set of functions that was introduced in section 2 of Chapter 2 but with a special choice of a set ofthe external parameters t = {t~.
. For any formal series of the form (A.IO) The idea of the proof of the theorems of this type which was proposed in [74. Indeed. Q).13) are gauge invariant.8) d) Z is an arbitrary vector (it corresponds to the divisor of poles of the Baker Akhiezer function).• )l/t(t. The expansion ofthis function in the neighborhood of Pa starts from O(k.i) such that (aa.a (A.• = a.l2) It has the same analytical properties as 1/t.1 ). From the uniqueness of the BakerAkhiezer function it follows that for each pair (a.. From the uniqueness ofthe BakerAkhiezer function it follows that 1/11 = 0 and the equality (A.9).k~).l3) Remark. (A.ll) becomes an equality.La. For any function g(t) opera tors .a· They can be found after substitution of the series (A.l' (A. It turns out that if the series (A.9) such that 00 (aa.3) is not formal but is an expansion of the Baker Akhiezer function in the neighborhood of Pa the congruence ( A.n. 75] is universal. n) there exists a unique operator La. (A.n of the form (A.1 )exp(L::>a.i.n)l/t(t.i.La . II) i=l The coefficients of La. (A.9) i=l (where aa. Q) = 0(k. Q) = 0. c) 2rr i uj. Corollary Al.i = a.n satisfy the compatibility conditions (A.La .a is the vector of bperiods of the differential d ilj.• )l/t(t.3) into (A.7) at the puncture Pa. The equations (A.3) there exists a unique operator La. Integrable Systems I 291 (A. with one exception.n are differential polynomials with respect to ~s. 1 + L:>~"·">(r)a~. let us consider the function 1/11 1/11 = (aa.l 0) is proved. The operators La.• of the form n1 La.ata.
292 B.A. Dubrovin, I.M. Krichever, S.P. Novikov
(A.l4)
have the same form (A.9) and satisfy the same operator equations (A.l3). The
gauge transformation (A.l4) corresponds to the gauge transformation of the Baker
Akhiezer function
1/IJ (to Q) = g(t)l/f(t 1 Q) (A.l5)
Example. Onepuncture BakerAkhiezer function.
In the onepuncture case the BakerAkhiezer function has an exponential singu
larity at a single point Pt and depends on a single set of variables. Let us choose
the normalization of the BakerAkhiezer function with the help of the condition
~ 1.0 = I, i.e. an expansion of 1/f in the neighborhood of P1 equals
(A.l6)
In this case, the operator Ln has the form
n2
L nan+'"'
t L.,iu<n>a;t· (A.l7)
i=O
For example, for n = 2, 3 after redefinition x = t 1 we have L 2 = a L, L3 = A,
where L and A are differential operators (2.12) and
(A.l8)
Therefore, if we define y = a 1t2, t = t3, then u(x, y, t, t4, .. . ) satisfies the KP
equation (A. I).
It follows from (A.l8) that in order to get the solution of the KP equation, it is
enough to take the derivative of the first coefficient of the expansion at the punc
ture of the ratio of thetafunctions in the formula (A.4). The final formula for the
algebraicgeometrical solutions of the KP hierarchy has the form
u(tt, t2, ... ) = 28f !nO (t. + Z) +
U;t; const, (A.l9)
(see details in [75]).
The formula (A.19) that was derived in [75] has lead to one of the most important
pure mathematical applications of the theory of nonlinear integrable systems. This
is the solution of the famous RiemannShottky problem.
According to the Torrelli theorem the matrix of bperiods of normalized holomor
phic differentials uniquely defines the corresponding algebraic curve. The Riemann
Shottky problem is to describe symmetric matrices with positive imaginary part that
are the matrices of bperiods of normalized holomorphic differentials on algebraic
curves. Novikov conjectured that the function
Integrable Systems I 293
u(x, y, t) = 28f ln9(Ux + Vy + Wt + ZIB) (A.20)
is a solution of the KPequation iff the matrix B that defines the thetafunction is the
matrix of bperiods of nonnalized holomorphic differentials on an algebraic curve
and U, V, W are vectors of the bperiods of corresponding nonnalized meromor
phic differentials that have one pole at a point of this curve. This conjecture was
proved in [Shi].
Let us make a few commments about the multipuncture case. For each a the
equation (A.l3) up to gauge transfonnations, is equivalent to the KP hierarchy cor
responding to each set of variables {t,.,;}. One could ask, "What is the interaction
between two different KP hierarchies?"
As it was found in [43], for the twopuncture case a full set of equations can be
represented in the following fonn
(A.21)
where H"·f1 is the twodimensional SchrOdinger operator in a magnetic field
(A.22)
and operators D~~ are differential operators in the variables t,.,t, t11.t.
The sense of (A.21) is as follows. For the given operator H"·f1 any differential
operator D in the variables t,., 1, t11 , 1 can be uniquely represented in the fonn
(A.23)
where D2 is a differential operator with respect to the variable t,.,t only and D3 is
a differential operator with respect to the variable t11 .t only. The equation (A.21)
implies that the second and the third tenns in the corresponding representation for
the left hand side of (A.21) are equal to zero. This implies n + m  I equations
on n + m unknown functions ( the coefficients of operators L,.,n and L 11 .m ). The
equations (A.21) are gauge invariant. That's why the number of equations is equal
to the number of unknown functions. Therefore, the operator equation (A.21) is
equivalent to the welldefined system of nonlinear partial differential equations.
We shall discuss multipoint case at a greater length in the next Appendix. In this
section we only consider the 2d Toda lattice as a basic twopoint example.
Let l/fn (t, Q) be the BakerAkhiezer type function that is defined by the following
analytical properties: a) l/1, as a function of the variable Q E r, is meromorphic
on r outside two punctures P1, P2 and has at most simple poles at the points
y1 , ••• , y8 ; b) at the point P,. the function l/1 has the fonn
(A.24)
294 B.A. Dubrovin, I.M. Krichever, S.P. Novikov
Here a = l, 2 and t1.o = n, t2.o = n. This function has the same representation
(A.4) through theta functions if we add one term in the arguments of the exponential
factor and thetafunction. Namely,
1{1 (t, Q) =ex (nil + "t· Q. (P)) O(A(P) + nU + L;,a U;,al;,a + Z)
n p ± ~ ,,a ,,a O(A(P) + Z) •
1,a
(A.25)
where dil± is the normalized differential with simple poles at the points P 1 and
P2 with residues ± l and 2;r iU is a vector of bberiods of this differential. Let us
normalize this function with the help of the condition ~0 • 1 = l and denote ~o. 2 (n, t)
by ~0.2 = exp(cpn(t)). From (A.25) it follows that
O(A(P2) + nU + L;.a U;,ati,a + Z)e(A(Pt) + Z) (A.26)
~=~
O(A(Pt) + nU + L;,a U;.ati.a + Z)e(A(P2) + Z)
0
If we denote the first "times" corresponding to the punctures by ~ = tu and 17 =
t2.1, then the usual arguments prove that
8t1frn(~, 17, Q) = 1/fn+l(~, '7, Q) + Vn(~, 11)1/fn(~, '7, Q), (A.27)
8q1/fn(~, 17, Q) = Cn(~, 17)1/ln1(;, 17, Q). (A.28)
where
(A.29)
The compatibility conditions of (A.27) and (A.28) are equivalent to (A.2). Hence,
the formula (A.26) gives solutions of the 2d Toda lattice equations.
Now let us consider the reduction problem. Let us assume that for the algebraic
curve rand fixed local coordinate k 1 in the neighborhood of the puncture P 1 there
r
exists a meromorphic function E(Q) that is holomorphic on outside the puncture
and has the form
(A.30)
in the vicinity of Pt. Then for the corresponding BakerAkhiezer function the fol
lowing identity is fulfilled
(A.3l)
For the proof of (A.31) it is enough to note that the right and the left hand sides of
it have the same analytical properties. Then the uniqueness of the BakerAkhiezer
function implies that both sides coincide.
The equality ( A.31) implies that the corresponding solution of the KP hierarchy
does not depend on the variable tn and the corresponding linear equation (A.l 0)
becomes
Ln1/f(t, Q) = E(Q)1/f(t, Q). (A.32)
Therefore, if r is a hyperelliptic curve that is defined by the equation
Integrable Systems I 295
2g+l
l= n(EE;) (A.33)
i=l
and the puncture is the "infinity" E = oo, then the formula (A.l9) defines the
solution of the KdV equation. If we choose the curve given by the equation
l+e+ L a;dEj =0
ni+3j:<On2
then the formula (A.19) gives solutions of the Boussinesq equation.
Now let us consider, as another example, the periodic reductions of the 2d Toda
lattice. Let w(Q) be a function on the curve r
such that it has a pole of order N
at the point P1 and a zero of order N at the puncture P2 and is holomorphic on r
except at P1• In that case the corresponding BakerAkhiezer function satisfies the
relation
Vtn+N(t, Q) = W(Q)1/fn(f, Q). (A.34)
Again in order to prove (A.34) it is enough to check that the right and the left hand
sides of it have the same analytical properties. This equality implies that
(A.35)
For N = 2 we have that the corresponding curve has to be a hyperelliptic curve that
can be represented in the form
2g
y2 =wn<ww;). (A.36)
i=l
r
Therefore, we conclude that if is defined by the equation (A.36) and two punc
tures are two branch points, P 1 = oo and P2 = 0, then the formula (A.26) defines
the solution of the SineGordon equation iu = qJ 1  qJo (this formula coincides with
(2.131)).
Appendix B
TwoDimensional Schrodinger Operators and Integrable Systems
Consider the general twodimensional Schrodinger operator L for the electric and
magnetic fields on the Euclidean plane JR 2 . After the standard identification of JR2
with C, z = x + iy, L can be written in the following complex form
296 B.A. Dubrovin, I.M. Krichever, S.P. Novikov
where a = ax iay. a= ax+ iay. and A, B, A), A2. v. u are functions of X, y
such that
2iA, =A+ B, 2A2 =A B, 2U=A,B,2V
The function U (x, y) is called the potential and the function H such that
is called the magnetic field. In the sequel, by the potential we mean the function
V(x, y). The operator Lis defined up to the gauge transformations
The only invariants of L are the potential V and the magnetic field H. For real V
and H we usually choose a gauge in which
B =A, A,x + A2y = 0
(i.e., A,, A2 are real and the Lorenz condition is satisfied).
In the twodimensional case, there are no nontrivial Lax type equations L,
[ Q, L] for differential operators L. The correct twodimensional generalization was
first developed in [43, 98]. In [98], it was suggested to consider the equation
L, = [ Q, L] + P L,
where P and Q are differential operators. Then it was shown in [43] that the inverse
spectral problem for L can be solved from the data obtained from eigenfunctions of
a single energy level,
L'l/1 = 0.
Using the methods of algebraic geometry, it is possible to find a large class of
exactly solvable systems. This theory was developed in [30, 86, 117] for the periodic
case and in [GN2, GM, G] for the rapidly decreasing case. Certain related problems,
for example the theory of pole systems, were solved in [KZ, FN"]. Below we outline
the main ideas of the theory in the periodic case and consider some interesting two
dimensional systems that can be integrated by these methods.
The simplest and the most interesting example of a nonlinear system is the fol
lowing twodimensional analogue of the KdV equation from the so called commu
tative NovikovVeselov hierarchy [NVl]:
 aw
at
= (a +a
3
8
3
+
u,a+u2a) w
2 •
where in the real case we put u 1 = u 2 •
This equation is equivalent to the Laxtype representation
Integrable Systems I 297
aL
 = [Q, L] + PL
at
with L = Ll + W, and where
Note that this system is more fundamental than the usual KdV or KP equations,
since its yindependent subsystem coincides with the usual KdV equation, while the
KP system can be obtained as a result of a special limiting procedure. There is also
an analogue of the Miura transformation [Bog]. The substitution
Uz =3( a
 1
au/)
 at) ,
i2
U1 a
 1
au/)

= 3 i2
(
af) ,
 at
W=tti2
transforms this equation into the modified Novikov  Veselov equation
Interesting classes of solutions of this system can be found by algebrogeometric
methods. These include doublyperiodic solutions, solutions which are quasiperi
odic w.r.t. x andy variables, and classes of exactly solvable twodimensional S.chro
dinger operators associated with a single energy level L 1{! = 0. Let all the coeffi
cients of the operator L be periodic in x and y with periods T1 and Tz respectively.
We call this case topologically trivial, since the flux of the magnetic field through
the elementary cell is zero:
[H] = 1T 1T' H(x, y)dxdy = 0,
2
where 2H = B,  A,.
The Bloch waves 1{! such that
Ll{!=O,
where
T11{!(x, y) = l{!(x + T1, y),
are parameterized by onedimensional complex manifolds r:
1{! = l{!(x, y, .9), .<?JE r.
for some differential operators Bkl· Clearly. the coefficients of L are only quasiperiodic..r. Krichever. In the general nonsingular rank one case we have the following analytic proper r ties.9'+f?7'_. where Qo = L. .A.j~O satisfying the relations k.. 1/J(x.298 B. Any algebra of this type. In general.y. 97i. .. Then l) the function 1/J(x. Definition. 9f'i E Let W± = k±_ 1 be local parameters in the neighborhood of 9f'i.m. S. z)a) + BL. QJ. . @)is a meromorphic function on r away from 9f'i. where L = aa + (lnc)z8 + V(x. Then the operator L. A commutative ring of operators "mod L" is given by the collection of operators Qo. 3) f71 are independent of x. is isomorphic to the algebra of functions on an algebraic curve r. . annihilates 1/1: Ll/J =0.9'+ ~. Dubrovin.I. Qk = L:a~(z. the genus of r is finite).9') "'exp{k+z}(I + O(k~ 1 )) as ..l=O. I.. i.y)exp{k_z}(I+O(k= 1)) as .M. the operators Qk are defined up to the additive part Therefore. Novikov Definition..P.9') "'c(x. 2) 1/1 has exactly g poles of the first order at the points 91. A differential operator L is called finitegap or algebrageometric w.. the level Ll/J = 0 if r is an algebraic curve (i.. 4) at 9"i the function 1/J has essential singularities of the special form: 1/J(x. y.t.e. whose coefficients can be expressed with the help of the theta function of the curve r. y. y. if it is complete enough and has rank one. Suppose is an algebraic curve of genus g with two marked "infinity" points r. which are merom orphic away from at most two points ~ E r. f71 f. y)..z)aiaj. . . and correspond to the Schrodinger operator L . ~. Q . The theory of twodimensional algebrogeometric Schrodinger operators is close ly connected with the twodimensional generalization of commutative rings of dif ferential operators.. . any operator Qk can be reduced to the following form: Qk"" ( ~aJ<z. z)a) + ( ~bJCz.
y. We now explain the conditions necessary to obtain a purely potential selfadjoint operator L=aa+W(x. ·~2 1/f(x. t).y). Here 9 is the standard Prym theta function with zero characteristics. Integrable Systems I 299 [84]. t). The antiinvolution r must have a maximal number 2g + I fixed cycles . Namely. . ti.. ti. The effective approach for such for mulae via nonlinear equations (similar to [46] for SineGordon and KP equations) was considered in [Tai]. W(x. a). a(~)=~. .. Under a these conditions we obtain a selfadjoint purely potential operator L = a+ W with W = 2aa log9(VJZ + V2z + Vo) + C(r. For such algebras of functions and. Consider the curve r whose group of involutions is isomorphic to z2 Xz2. ·~2 In the real case.t{. r let have a holomorphic involution a with two fixed points a: r+ r.Fourier bases. . but with the different asymptotic behavior near ~: 1/f(x. these bases give a structure of an almost graded algebra [KN l]. Nat2] and sufficient [NVI. ti. y) E R. more generally. . .) "'exp lk+z + Lk~ttj. of arbitrary order tensors on r. the following conditions are necessary [Natl. ar = ra. and V1. These Krichever . C(r) is some constant.! . . r(~) = ~.) "'exp lkz + L k~t. The divisor of poles !YJ = !?If + · · · + g. y. In the scalar case (as above).. To study nonlinear systems one must consider 1/J with the same analytic properties. ti. and antiholomorphic involution r r: r+ r. t{. it is possible ([KNI]) to construct analogues of the Laurent..Novikov bases are necessary for the operator quantization of the bosonic string. V2 are vectors of normalized periods of abelian differentials ofthe second kind corresponding to the points ~.. .. must satisfy r where K is the canonical divisor of and "' denotes the linear equivalence. GN2].. the following conditions must be satisfied: To obtain a strictly positive purely potential operator L. .
finitegap operators is everywhere dense in the space of doublyperiodic potentials W (x. The correct generalization of this theory for the g = oo case was developed in [Kr3]. If Ll/1 = 0. As was mentioned before. Hk+l = Hk + !Llfk· . we have A= A (In V). . S. The rapidly decreasing analogue of this theory was studied in [GN*.. was described in [G]. the theory of the dynamical evolution for the pole systems of the rational solutions of the above equation was constructed in [FN*]. Novikov where g is the genus of the curve r0 = r 1a..2. Consider the infinite chain of Laplace transformations for the potential and the magnetic field: ef<+. Llln V H=H+.300 B. and it is possible to drop the standard assumption. and !¥ = £:11+ + £:11. l. then we put {it = (8 + A)l/1. were developed in [KT] for exactly the same class of curves as considered in [Kr3]. It was shown that the class of. where Eo < 0 is below the spectrum.+ . .P.. Dubrovin. y). the case of a positive operator L = 88 + W is considered in [GN2]. In other words. a : Cg+l + Cg+l. G].. The example of a simple rational decreasing potential. =elk + Hk+J. L = (B + B)(8 +A)+ 2V.. . In particular. Laplace Transformations The Laplace transformation for solutions of the Schrodinger equation is a two dimensional analogue of the BackluildDarboux transformation. j =I. v = v +If. + ~+ + ~ g'j± C C1±. In this case. L= V(8+A)v. Let V = exp(f). whose corresponding operator L has a scattering matrix which is trivial on one level. for example the generalization of the theory of theta functions. 'fie = 1. of the small norm of the potential.M. for this theory and the corresponding KP theory.1 (B+B)+2V. g a : C1+ + Ci.A. GN2. GM. Some generalizations of recent results. a(g'j+) = g'j. the data is obtained from the level Ll/1 = e0 1/f. Krichever. 2. iJ = B.
the dependence of the inverse spectral problem data (i.t.efo. n =3 !Llfo = ef•. we can replace constants C by arbitrary harmonic functions. this equation and its Lax representation were first obtained in [Mi] as a twodimensional analogue (via the ZakharovShabat approach) of the usual Toda chain.e. 4 = Hn = 0). for n = 4 there is a reduction fo + /1 = C 1 to the sinhGordon equation I Llfo = C2efo.efo.. n" boundary con ditions.t.efo. In the Toda chain theory. and nonsingular.efo. it was shown that the algebrogeometric solutions of the twodimensional Toda lattice. whose approach was based on the Laplace trans formations (as above).efo. Also. Tsitseika [T]). and the topology of T 2 (see the review [Bo. Tail]). It is worth mentioning that in the doublyperiodic case any algebrogeometric operator L is topologically trivial.e. we obtain a wellknown two dimensional Toda lattice This equation and some of its reductions were already known to the classical ge ometers (Darboux. the above result was already known in connection with the theory of surfaces of the constant mean curvature in JR. then the corresponding Laplace operator Lk is always algebrogeometric (finitegap on a single energy level Lk 1/t = 0). a Riemann surface with two marked points and a divisor of degree g) on the parameter k can be described as follows: the Riemann surface and the marked points are fixed and the divisor of poles is shifting. then we have: If Lo is a potential operator (i.) = ce3/o211. x ~ n. i. Therefore. Integrable Systems I 301 Making a change of variables fk = gk . !Ll(2fo + j. obtained in [Kr2] using the methods of the soliton theory.f = exp(f).efo = cefo. 4 In [NV]. the variable x.. determine for each value of the discrete parameter k the algebrogeometric (from the spectral theory of the periodic operators point of view) SchrOdinger operator Lk. Ho !Llfo = ce2/o.3 . the periodicity condition has a solution in the sinhGordon form I Llfo =ell. Moreover. For n = 2. n =4. In the soliton theory.r. the magnetic flux is zero: . Under certain "open w. if all ft are smooth. doublyperiodic. the connection with Lie algebras was first noted and used in [LS]. In general. it is a difference version of the KP equation w.gkl.. in this case. In this case. this equation is an analogue of the Liouville equation /J. corre sponding to the simple Lie algebras.r. We note that the n = 2 case corresponds to the sinhGordon equation and.e. Periodic chains of Laplace transfor mations fn+k = fk were studied by Darboux and Tsitseika [T].
Novikov [H] = ji H(x. y)dxdy = 0. . The levels A.P. DN3. the spectrum in $ 2 (JR 2 ) is discrete and forms an arithmetic progression so= H.o = 0. This level (i. In the sim plest case H = const we obtain the Landau case.. if H satisfies the integrality condition [H] = 2rrm > 0. They are infinitely degenerate. for the n = 2 case. Dubrovin. 2e 10 ' [Ho] > 0. A2 are real and the Lorenz condition is satisfied). the whole Hilbert space) is canonically isomorphic to the ground Landau level with the same magnetic flux and H = const. In the periodic case ([DN2.• .A. A2 (or A. I. It is convenient to choose a gauge in which B = A (i.• .. where rp = ~ ji H(x'. a 1 .aj)e"'. Also note that for nonsingular smooth periodic ft.e. then the magnetic Bloch eigenfunctions of the ground level are given by 1/1 = e"' n m j=l a(z. m EZ.. •. am are arbitrary constants. In [AC. where K is the fundamental domain for the action of the group of periods 'Z} on the plane JR2 . DN2. A 1. Already. it was shown that.302 B. Ln of Laplace transformations are given by the relations Ho = Vo. in the rapidly decreasing and periodic cases. The quasicyclic chains Lo. this fact implies the following interesting equation: 1 lJ. DN3. These chains satisfy the strong integrability properties.fo = c2 . N3]). the con dition V = H implies that the operator L has a strongly degenerate "ground" (basic) level A. N3]. the Laplace transfor mation preserves the magnetic flux: [Hk+J] = [Hd [Vk+J] = [Vd + [Hk+ll· The case in question is called doublyperiodic topologically nontrivial if H and V are periodic with possibly nonperiodic potentials A 1. B). The Landau operator can be considered as a natural twodimensional analogue of the harmonic oscillator whose spectrum forms an arithmetic progression.z')l dx'dy'. and the constant a can be expressed in terms of a 1. 2 .k are called kth Landau levels. In this case. am. y') In la(z. S. •.. .e. Krichever.M. Z =X+ iy.
2 =. n1 E Z. In the hyperbolic case we begin with the equation L 1{1 = 0. 0). in the discrete case they are formally different.C2 and. Such are the twodimensional analogues of the theory of cyclic BacklundDarboux chains. DN3. {f.= + v. put L{f = 0. there is a ground level A. TJ) + s.T1) + l] The representation of the form L =g.T2)w. n2). where 1{1 = e'P n. C 2). I.e8 appears in many different problems [dV]. 1). = 1/ln +anTt1/ln +bnT21/In +cnTtT21/In. a+ ay. Integrable Systems I 303 Let C2 > 0. [0 + v. the solu tions of this equation correspond to interesting onedimensional "oscillatorytype" potentials whose spectral properties are very different from the spectral properties of the cyclic chains [NV]. where 0 are shifts on the basis vectors of the lattice. Since for the operator L 2 + C2f2 we have H = V ([DN2. (l L = 1 :~. The operator L 2 + C2/2 has two strongly degenerate energy levels. The second level corresponds to A. The operator L admits the representation Similarly to the continuous case. 10 + u. where L1{1.z')l d 2z Ho = efo. the eigenfunctions have the form $= efll2(a + At)etol2(a + Ao)1{1. It seems that in the doublyperiodic case with nonzero magnetic flux (excluding the Landau case). Now we consider the discrete analogue ofthe Laplace transformations. . 11(/J = ~ JJK Ho(z') In la(z. The nonlinear equation 11g = I . Both levels 0 and C2 are strongly degenerate as Landau levels and there must be a nontrivial spectrum in the interval (0. T1 = (l. T2 = (0. there are at most two strongly degenerate levels. after the choice of real gauge. [(I + Pn T2)(l + q. N3]).] generates the inverse representation L 1{1 = 0.aj)eaz.T2)1{1. Although in the continuous case hyperbolic and elliptic operators are formally related by the substitution a + ax.=l a(z. [Ho] = 2nm > 0. Even in the onedimensional case. It is not integrable from the soliton theory point of view. but it has large families of nonsingular doublyperiodic solutions. and n = (nt.o = 0.
+r2b. 2. are invariant under the gauge transfonnations L+.304 B.bn+T1 e"= a.I J J ' where the lattice consists of equilateral triangles. every vertex has six neighbors. The infinite chain of Laplace transfonnations gives a complete discrete ana logue of the twodimensional Toda lattice for the variables w~k>.Tt + c. H~kt>) by the Laplace transfonnation. From the second equation above it follows that the pair (w~k>.1) + Wn 1 2 implies that for any solution L l/1 = 0 we can put in correspondence the solution Llit = 0. 1(C + Wn+T ){C + Wn+T ){1 + Wn+T )I (1 + Wn+T2 )I. after the reduction w<t> = C(w<O>)t n n ' Wn+T +T = w. S.1 + Znr T2.A. Krichever. The factorization L = (Xn + Yn Tt + Zn T2)(x. Therefore. 1 2 1 2 1 we obtain the discrete analogue of the sinhGordon equation. Novikov The potential and the magnetic field H a. It is interesting to compare this sys tem with the system constructed in [Kr7] and [KLWZ] using the theory of the Yang Baxter equation. (1 + Wn+Tl). The Laplace transfonnation for the selfadjoint real difference operators L is considered only for the operators of the fonn L =a. H~k>) can be obtained from (w~k1>. + Ynr Tl.M.P.gLg.1• The Laplace transfonnation is given by the following fonnulae: eii• = eH"Wn+TlWn+TI(WnWn+TI+Tl)1 1+ Wn+TI = eii. Dubrovin.. Then. eH•. for N = 2. where .+ b.T2 + dn+TI Tt Ti 1 + bnTI TIl+ CnTl T2l + dn+TlTIl T2 L + = L' T+ = r. It is also possible to impose the following cyclicity condition on chains. I.
TI + Zn T2) + l 1 As usual. Kr4. where Mm is some "manifold of compo nents". m. DN3. . Notice that for both classes of operators .. . Seiberg and E.. + YnT TIl + ZnTz T2. = v. The new applications of these ideas to the twodimensional quantum field theories are discussed in Lecture Notes [D l] and the Appendix to [N l ]. SW2]) on the supersymmetric YangMills theory. 1 (x. 2. um) are special local coordinates on the manifold Mm which diagonalize the matrix of velocities: . P2] and summarized· in [DN l]. where one has to solve the Ricattitype equations or their discrete analogues. . The factorization of the operator L can be described by local algebraic formulae in the coefficients of L. + y.• + Mm. p. In this equation. the systems of hydrodynamic type have the form u. N3]. see [AC.hyperbolic in the square lattice. In the spatially onedimensional case n = I. ••• . This is different from the onedimensional case. Ts. Integrable Systems I 305 {r. q = I. we can consider infinite chains of Laplace transformations and the cyclicity conditions. and it leads to new discrete analogues of the twodimensional Toda lattice and its reductions. AN. = (Xn + YnT TIl+ Znr T2.1)1/t. there is an important notion of Riemann invariants. Witten ([SWI. and it recently attracted a lot of attention after the interesting work of N. a= l. This is an active area of current research. Appendix C Integrability of Systems of Hydrodynamic Type. Riemann invariants (u 1. 2. = 0) or (w: = 0) correspond to the theory of zero modes. discussed below. was developed in [48. and real selfadjoint in the equilateral triangle lattice .1)w. the alternative factorization corresponds to the inverse Laplace transformation. 1 2 i = (x. n.·"(u(x))u:. . The NonLinear WKB Method The circle of ideas.. u (x) is a map JR.we have the following result. As before. DN2.. As was discussed at the end of §2. The conditions (w.
We now consider the Tsarev method for integrating diagonal Hamiltonian systems ofthe form u~ = vP(u)u:. and they are connected with the velocities by the relation kjp (warning: no summation). in the theory of the dynamics of gases (m = 3). Dubrovin. This theory is based on the Riemannian geometry lying in the foundation of the Hamiltonian approach. In the case m = 2. . The exact solvability of the twocomponent spatially onedimensional sys tems of hydrodynamic type is described by the following theorem (attributed to Riemann): if we do the "hodograph transformation". According to the Novikov hypothesis. For example. would also be diagonal: gpq(u) = gP(u)o. Riemann showed that if the eigenvalues of the matrix V: are real and distinct in the given domain. The Whitham method is described in [Wh I] and the existence of Riemann invariants in this case is discussed in [Wh2.M. The Christoffel symbols rk~ for the standard (LeviCivita) connection are obtained by the formula a rkkP =21 8uP . In fact. u 2). Riemann invariants exist only for a special (and noninteresting from the physical point of view) value of the Poisson adiabatic exponent.t. In general. I.. These are systems which are Hamiltonian w. Krichever. This method is based on the fact that in this case the pseudoriemannian metric. S.P. This hypothesis was proved in [Ts.I n lgk(u)l (warning: no summation). it is possible to find Riemann invariants for some important systems of hydrodynamic type. F]. this theory can be extended slightly to the class of semiHamiltonian diagonal systems. which is used to define the Poisson bracket of hydrodynamic type.r. These systems arise in the theory of asymptotic methods for the soliton equations of the KdV type and they are a nonlinear analogue of the WKB approximation method or the method of the slow modulation of the parameter (this method is also called the BogolubovWhitham method). for m > 2. the existence of Riemann invariants together with the Hamiltonian properties of the systems of hydrodynamic type im ply their strong integrability. we cannot always find the set of Riemann invariants. As was mentioned at the end of §I of Chapter I. the Hamiltonian approach for the systems of hydrodynamic type was developed in [48.. i. 56]. rewrite the equation in terms of x(u 1. t(u 1. nonlocal Poisson brackets of the type considered in [MF.e. then it is always possible to find such a coordinate system. 49. Tsl. where the beautiful theory of integration for the diagonal Hamiltonian systems ofhydrody namic type was constructed. Mo] (see also Insert for §I). Novikov A system written down in Riemann invariants is called diagonal. DNI]. However.A.306 B. u 2). then the new equation would be linear.
m form a solution to the original system. Integrable Systems I 307 Therefore.. We conclude this review with the following general theorem. /JuP such that vP is independent of uP. = wP(u)u~ (symmetry) Then a theorem says that the functions uP(x. This technique was developed in [Kr4. p = I. Although expressed in different terms.. One of the examples in which it is possible is pro vided by the so called weaklylinear systems. In certain cases. semiHamiltonian systems are actually Hamiltonian w. The systems of hydrodynamic type which arise in the soliton theory (for exam ple.. . each symmetry corresponds to an integral.. the generalized nonlocal Poisson brackets of hydrodynamic type (see Insert for §1). which commute with each other and with the given system. If functions u 1 (x). t)) = vP(u(x. Diagonal Hamiltonian and semiHamiltonian systems have ample families of in tegrals of hydrodynamic type. = vP(u)u~ (the original system) u. . but this fact is not proved in general. Note that the met rics considered in the theory of the local Poisson brackets of hydrodynamic type are always flat (see §I).r. it is hard to apply these beautiful general differentialgeometric theorems to concrete problems. The method for integrat ing Hamiltonian and semiHamiltonian systems of hydrodynamic type is called the generalized hodograph method or the Tsarev procedure. .t. we have the following important Tsarev relation: This relation is taken as a definition of the semiHamiltonian diagonal systems of hydrodynamic type. i. in the KdV theory) from the nonlinear WKB method are strongly nonlinear. but this (diagonal) metric can have a nonzero curvature tensor. It can be shown (though the proof is not effective) that there exists a large family of"symmetries". then the symmetries of hydrodynamic type generate the complete set of .. The implementation of the Tsarev method requires the use of algebraic geometry.. .. the integrals of motion and symmetries are in onetoone corre spondence. . Consider the two systems: u.2. YaR]. PM] and is discussed below. systems of hydrodynamic type. We can always recover the metric from the above formulae. um(x) are convex. Weakly linear systems are the systems of the form /JvP =0. p = 1. . t))t + x. t). this result for the two component systems has been known for a long time [La. m (no summation). 2. determined from the equations wP(u(x. In the Hamiltonian case. each integral generates a symmetry.e. Unfortunately.
The systems of hydrodynamic type appear in the following context. For simplic ity. the vari ables q 1. 1/ko and K vectors U. . u m) . u 1. um). ••• . assume that 1/li = KP(1fJ. in a neighborhood of convex functions u 1(x).e. t) be "slow" functions.. T = et. ••• . .A. We are looking for a solution 1/J.. which is asymptotic w. In other words. FFM]).. ••• . and is of the form 1/1= tP(S(X. This function must satisfy the socalled "averaged system" auP = VP (u(X T) auq) ar q • ax ' . S. where the functions tP(q 1.P. Let 1/JP(x. T). Then there is a dispersionless limit of this system: 1/li = v:<t>t:. AB. . um) . Tlk) are periodic with period 2rr w. ••• . commuting with the given diagonal Hamiltonian system.r.. . (a) (b) Each next derivative is of a smaller order.308 B. DNl]).r. more sophisticated method for obtaining systems of hydrodynamic type is a nonlinear analogue of the WKB method ([Whl. it is possible to show that the neces sary condition for the existence of such solutions is the following condition on the function uP(X. ••• . La.... Another. where u = (uP(X. This means that the symmetries form a basis for the tangent space to the generic level surface of the pairwise commuting integrals of hydrodynamic type. their linear combinations are everywhere dense. V( aT= u ' . T). Krichever. T)). 1/Jx. ..t e.>) is an evolution system such that all1/f = const are solutions. 1/lc.. as I In certain cases (see the review [DNI]). and it is a system of hydrodynamic type. e+ 0. t. X=ex..M. Wh2. i. u 1. . um. um) + O(e). um(x). This method requires the existence of a family of exact solutions of the form 1/f(x. I.t. u) = t/J(Ux + Vt. Ma. Novikov flows. The evolution equation has the form where GP = 0 and all omitted terms are small by assumption. V can be expressed in terms of m parameters u 1. the system is completely Liouville integrable ([Ts. Thus. Dubrovin. . and axas _ U(u 1.
rm are the boundaries of the Bloch spectral zones in $2 (lR) of the SchrOdinger operator with the finitegap quasiperiodic potential U = (/> ([FFM.I. ·} = 0 for all q = 3. Vrx• . Here we average over the above solutions.•.r. ... um) can be taken in the form of the average densities: Then uq are said to be physical variables. {/q. the physical variables u I •... k + I) For Hamiltonian evolutionary systems with a local translationinvariant Poisson bracket {.t. )dx. Bohr.• um: gPq(u) = yPq(u) + yqP(u) aypq bPq(u) = . . in the KdV case. .. and the value of an integral on the space of almost periodic functions is defined as the average of its density in the sense of H. DNl]).. be annihilators. . s au• . Finally.. m. we can use the method developed in [48]. For these systems. . . Integrable Systems I 309 which is a system of hydrodynamic type. q = 1. We also require that the parameters uq for the family of the exact quasiperiodic solutions 1/f = (/J(Ux + Vt.e... the Riemann invariants exist for all k 2:: 1. . The Poisson bracket can be written in the "Liouville form" w. )dx. The sufficiency condition is known only for certain cases with k = I.2.. The fundamental principle of the conservation of the Hamiltonicity under averaging claims that the averaged system is also Hamiltonian with the Poisson bracket of hydrodynamic type. ••• . }0 .. for the KdV equation such equations of hydrodynamic type (Whitham systems) were obtained in [Wh2] fork = I and in [FFM] fork > I from the family of the finitegap quasiperiodic solutions with m = 2k + I. This method requires the existence of a family of local integrals lq =I Pq(l/1. h = P be the momentum. the Riemann invariants r 1. i.. . These systems are Hamiltonian [48].. A similar statement holds for the momentum. The pseudoRiemannian metric gPq is flat and has signature (k. and / 3 . Its Hamiltonian coincides with the usual energy: u I  = P1.. s. . . The dispersionless limit corresponds to the case k = 0.m. . we require the integrals lq to be in involution: p. Namely. In particular. .q = 1. H =I P1(1/r. Let / 1 = H be the Hamiltonian of our original system. u 1 .
Dubrovin. We denote their densities by uq. define an infinite family of integrals of of hydrodynamic type. The formulae for the Poisson bracket of hydrodynamic type for these cases can be found in [48].1{1x. [DNI]. For k ~ 1 we have to add 2k families. This result was proved in complete generality only recently by A. Maltsev (to appear in Izvestia. Novikov Since {IP.)) · Then we have ([DNl]): yPq = QPq + const. u 2 . lq} = 0. For the usual KdV we have V(l{l) = 1{1 3 . one has to integrate a certain form il = p d). . one has to add two more families of integrals.M.. 2. u 1. I. Ya. .)}o = dx (QPq(1{1.P.A. u 3 • These parameters are the average values of the following integrals: ul = 1/13 + V(l{l) energy density 2 u2 = 1{12 momentum density u3 = 1{1 annihilator (or "Casimir") density. their description can be found in [Ti]. To obtain these families. 0.l{l(y)}o = 8'(x.. = dx 81{1(x) · The family of exact solutions of the form l{l(x . the original Poisson bracket has the form d {/p. + V(l{l)) dx This system can be written in the Gardner form (see §l): d 8H 1{1. · · . ser math) (the review [DNl] contains a mistake. q = 1.310 B. .1{1x.. For k = 1. S.ri). u2 • u 3). depends on three parameters u 1 . For the KdV equation this result was obtained earlier in [48]. j=O . Krichever. Pq(1{1. over basic cycles on a hyperelliptic Riemann surface n 2k r: ~L 2 = R2k+l (A)= (A. 1{1 = <P(Ux + Vt. 1. and the averaged densities of the Kruskal integrals lq. The existence of this family is not sufficient to have complete integrability. Consider the KdV equation and its deformations via the Gardner ZakharovFaddeev bracket {1/l(x).vt).. its proof for some special cases can be found in [NM]. Example I.y) H = j (~. see [N4]).
. AppendixD Spectral Theory of TwoDimensional Periodic Operators The basic algebraicgeometrical construction allows one to obtain periodic and quasiperiodic solutions for twodimensional integrable systems but leaves abso lutely open the following basic question: "How many algebraicgeometrical solu tions are there? And what is their role in the solution of the periodic Cauchy problem for twodimensional equations of the KP type?" For finite dimensional (0+ I) systems that have the Lax representation of the form a. In particular. r2k) is a socalled quasimomentum connected with the spectrum of the finitegap Schrodinger operator L. (D. One of the most important problems in the theory of twodimensional inte . the integral/4 corresponds to a very interesting solution. Any integral of hydrodynamic type generates an exact solution of the averaged Whitham system via the Tsarev procedure (see above). . the answer to the question is as follows: all the general solutions are algebraicgeometrical and can be represented in terms of the Riemann thetafunctions. The actual solution was found in [P2] using the procedure from [Kr4].) rational (or sometimes elliptic) matrixvalued functions of the spectral parameter ). It seems.U(t. A. This solution is discussed in great detail in [DNl].. For spacial onedimensional evolution equations ofthe KdV type (1+1systems) the existence of direct and inverse spectral transforms allow one to prove (though it is not always the rigorous mathematical statement) that algebraicgeometrical solu tions are dense in the space of all periodic (in x) solutions. It turns out that the situation for twodimensional integrable equations is much more complicated. and it is very important in physi cal applications. I) with U(t.) and V(t. . The existence of this solution and some of its qualitative properties were established in [GP]. This solution is called a dispersive analogue of a shock wave (or the GurevichPitaevskii solution). ). A. A. that the same statement for the KP1 equation (0' = i) is false... ro. The averaged Kruskal inte grals generate selfsimilar solutions [Kr4]. For one of the real forms of the KP equation that is called the KP 2 the algebraicgeometrical solutions are dense in the space of all periodic (in x and y) solutions [Kr3]. The main family of the Kruskal integrals and their averages is obtained from the quasimomentum expansion at the "infinity" point. although it was not clear that it was even C 1smooth. V(t. Integrable Systems I 311 where p = p().)= [U(t.)].).. ).
trans fers the solutions of (D. WI. WI.13) and for which a wide class of periodic solutions has been constructed. Wio W2).4) with a periodic potential u(x. where ayh(y) is a periodic function. W2) = W21/I(X.S) 1/I(X. Wio W2) =0 (D. Consequently. T(E)) (D. eh<Y>l/1. W2) = Wtl/I(X. The gauge transformation 1/1 . for which there exists such a solution is called the Floque set and will be denoted by r. a.+ U(X. S. Krichever. We would like to mention that despite its application to the theory of nonlinear equations and related topics.+ u(x. y. The set of pairs Q = (w 1. y)}l/I(X. The solutions 1/l(x. w2).a. y + 12) = u(x. y) = u(x.4) into solutions of the same equation but with another po u tential = u uayh. We would like to emphasize that this is the only possible representation that can be used in two dimensions. The multivalued functions p(Q) and E(Q) such that are called quasimomentum and quasienergy. E) = det(w . I. The proof of the integrability of the periodic problem for the KP2 equation is based on the spectral Floque theory of the parabolic operator M = ay. 1/I(X + a1.P. Novikov grable systems which are still unsolved is "in what sense" the KP1 equation that has the operator representation (2. i. w~o w2) of the nonstationary SchrOdinger equation (ua1 . respectively. y).2). W2). WI. the spectral sets corresponding to the potentials . y) = u(x.6) The Bloch functions will always be assumed to be normalized so that 1/1(0. y. Dubrovin. The characteristic equation R(w. y. w1. In [Kr3] another represen tation of the Riemann surface of the Bloch solutions was proposed.e.3) where T(E) is the monodromy operator for periodic ordinary differential operator L defines the Riemann surfacer of the Bloch solutions of this operator as aN = nl sheet covering of the complex plane of the spectral parameter E (here n is the order of L and I is the matrix dimension of its coefficients).M. y. (D. y. (D. the structure of the Riemann surface of Bloch solutions ofthe corresponding linear equation that was found in [Kr3] has been used as a starting point for the abstract definition ofthe Riemann surfaces of infinite genus [KT]. y +a2) are called the Bloch solutions if they are eigenfunctions of the monodromy operators. Let us consider as an example a nonstationary Schrodinger operator (D. y) = u(x +a1. (D.312 B. w 2) = I. is a "nonintegrable" system. 0. y + a2.A.2) with periodic potential u(x + l1. y).
9) it follows that k _ k' = 2rrN . y) = 0.12) The formulae (0.4) as a formal series . (0.10) The functions l{l+(x.9) define the map k E C ~ ( w?. Therefore. W~) E C 2 . Hence.. The Floque set of this operator is parametrized by the points ofthe complex plane of the variable k (0. in what follows we restrict ourselves to the case of periodic potentials such that 1 41 u(x. i = I. of the equation (0. 2.15) Q) k2 _ (k') 2 = u2rriM . Integrable Systems I 313 u u and are isomorphic. The selfintersections correspond to the pairs k =I= k' such that w?(k) = w?(k'). k = kN. (0. (0. For any ko =I= kN.17) a1 Na2 The basic idea of the construction of the Riemann surface of Bloch solutions of the equation (0. y) is formally small. (0. (0.8) with zero potential u(x. k) = eikx+a•k2y.. (D. k = kN.M· (0. (0.14) From (0.4) as a perturbation of the free operator (0.4) that w. y. It is the Riemann surface with selfintersections. 13) Its image is the Floque set for the free operator Mo..as proposed in [Kr3] is to consider (0.8).11) are Bloch solutions of the formal adjoint operator (ua.7) To begin with let us consider as a basic example the "free" operator Mo = ua. all the resonant points have the form rrN uiMa1 . (0. assuming that the potential u(x.16) a2 where N and M are intergers.M it is easy to construct a formal Bloch solution if.a. N =I= 0..9) and the Bloch solutions have the form (0.>t+ = o.M = .. + a. y)dx = 0.
Wzo = w~(ko) (0. w10 = w?(ko).). y. ko) = wwl/t(x. y. = ko + a! ..25) 2Jrn 1/1. = 1/f. ko) = (0.24) it follows that the formula . Krichever. ko) cp. y))l/t(x. y. Novikov =L 00 1/t ijJ.20) has a formal solution of the form =L 00 1/t(x. ko) = F(y.26) (0.a. = 1/1 + (x. i{Jo(x. cpo = 1/Fo. y.M then there exists a unique formal series L F. y.22) (here and below (f (x)) x stands for the mean value in x of the corresponding peri odic function f) 1/t(x + a 1 . k. y.314 B..) . k.25 D.23. y) = 1/f(x. ko). ko). Dubrovin.(x. 0. y. (0.29).(x. (0.27) (0. ko) (0.24) The corresponding solution is unique and is given by the recursion formulae (D. ko)l/t(x. y.+ u(x. 0.(x.18) s=O This series describes a "perturbation" of the Bloch solution 1/fo of the nonperturbed equation. ko) = 1/Fo. y.21) s=O satisfying the conditions (0. ko). y. S. I. ko) 00 F(y.P. ko) = 1/f(x.M. (0. (0. Lemma Dl. k.20. Jfko =/= kN.23) 1/t(x. ko). y. y. (0. + 1/1.29) From (0.A. ko) = Wzol/t(x.28) (0.19) •=I such that the equation (aay. y + az.(y.
+ u)l/la(x.37) fJ . WJ) = L F. The discreteness of the resonant points in the last case is crucial for the extension of If to a "resonant" domain ( and for the proof of the approximation theorem ).fco)dy' (D.34) is an analog of simplicity of an eigenvalue of an operator. ko).4): lfr(x + a1. y. WJ)I/Ifl(x. In the case Re a =I= 0 there is only a finite number of the resonant points (D. ko) 1/1(0. y. (D. The principle distinction between the cases Rea = 0 and Re a =I= 0 is revealed under an attempt to extend 1f to a "resonant" domain.33) For sufficiently small u (x. y) it is not too hard to show that the above constructed series of the perturbation theory converges outside some neighborhood of the reso nant points (D. The condition (D. ko) = wwl/l(x. a. k ) 0 = 1/!(x. ko) = w2o1/J(x.I7) and therefore detennines a function If (x. it is necessary to proceed along the same lines as in the perturbation theory of multiple eigenvalues.35) Lemma 02. when u does not depend on y.31) lfr(x. WJ) (D.(y. y. In the stationary case. y. y.i. In the case Re a = 0 the resonant points are dense on the real axis.(x 'Y 'y. Integrable Systems I 315 . As the set of indices corresponding to the resonances we can take an arbitrary set of integers I E Z such that (D. the preceding fonnulae turn out to be the usual fonnulae of the perturbation theory of eigenfunctions corre sponding to simple eigenvalues. There are unique formal series L F% (y. y + a2. (D.32) where the corresponding Bloch multiplier is equal to (D. In cases when it is vio lated. 0. y.17) in any finite domain of the complex plane. ko) which is analytic in ko. y. ko) e ()'' J: F(y'. WJ) = 5 WJ) (D. This is true for any a. X Fp(y. ko).36) s~l such that the equations (a8y.30) defines the fonnal Bloch solution of the equation (D.
Novikov have unique formal Bloch solutions of the form L cp:(x.y. Let us define the matrix T = r. y. WI).40) such that (0. y.A.316 B.47) a then (0. w2) be an eigenvector of the matrix r. ka). (0. ~ Let ha(WI. wt) = 0 cp = Vra = 1/r(x. (y.y.wi) = ""' a ~TfJ(y. Krichever.4). y.28) (see details in [Kr3]). and 1/1" are the matrix generalizations of the formulae {0.48) a .(wi)w2. while under the translation by the period in y they are transformed as follows (0. 00 1/ta(x. wt). (0. Under the translation by the period in x they are multiplied by WI. (0.T+TF=O. WI)= WII/t"(x. y.39) 1/t"(x.43) {J are solutions of (0. y.wt) (0. wt). Dubrovin. (0. y + a2. I.260.38) s=O 1/t"(x + ai. The characteristic equation (0.45) It is natural to call a finite set of the formal solutions ifta quasiBloch. T(O.41) The corresponding formulae for F.M.wi)=l. WI) by the equation ao.fJ normalized so that (0.44) where (0. WI)= w2alft"(x.P.42) The functions 1ft~" (x.46) is an analog of the "secular equation" in the ordinary perturbation theory of multiple eigenvalues. y. since it remains invariant under the translation by the periods in x and y. S.wi)Ift {J • (x.
M and RN.M can be naturally identified with the bound aries of RN.M• but it turns out that the series of the lemma 3.M and RN. If the potential u (x.M. From the topological point of view this surgery is a gluing of a "handle" between two resonant points. The boundary of RN.25 D. the structure (local) of the Riemann surface r of the Bloch functions looks as follows. The remarkable thing is that the perturbation approach works even when u (x.29) of the nonresonant perturbation theory diverge in RN. In [Kr3] it was proved that if the potential u(x. 17) can be analytically extended into the domain (D. Structure of the "Global" Riemann Surface of Bloch Solutions. Let us cut out RN. respectively. y) of the equation (D.49) and defines a twosheet covering RN. The last statement means that the Bloch solutions are defined (locally) on the Riemann surface (D.4) are defined. Of course.4) with multipliers w 1 and w2 .M on which the Bloch solutions of (D. in that case the estimations of the perturbation theory series are much more complicated.M and in this domain defines quasiBloch solutions of (D.M from the complex plane and glue instead of them a corresponding piece of the Riemann sur face RN.M and kN.4) which are analytic in w1• The characteristic equation (D. Let us consider some neighborhoods RN. Theorem Dl. In the central domain Ro we have to glue some finite genus piece of the corresponding Riemann surface Ro instead of a disc Ro. 50) then the Riemann sureface r of the Bloch solu tions of this equation is a result of the abovedefined glueing of the three types of "pieces": 1o.46) in this case has the form (D. The function w 1 (k) (D.M. A complex plane of the variable k without small neighborhoods of the finite or infinite set ofpoints kN. r 2 then the perturbation series for the nonresonant case converges outside some central finite domain Ro and outside RN.50) for some r1.M ~ Ro).M ~ Ro.M· Hence.M over RN. we shall give here the explanation of the structure of a "global" Riemann surface of Bloch solutions in the case of small u. y) is not small.M for kN.9) identifies them with some neighborhood RN.M• kN.4) with Rea =F 0.M) = WJ(kN.M and RN.M of the point WJ(kN. The series(D. As a result we obtain the global Riemann surface r of the Bloch solutions of the equation (D.2 converges in RN. To begin with. Outside Ro we again have to perform a surgery of the previous type ("glue" handles between kN. normalized in the standard way.M and without some central domain lkl > Ko: .M and RN.M and kN.46). y) can be analytically extended into a domain jim xi < <1.M) on the complex plane ofthe variable WJ. jim yj < <2 (D. Integrable Systems I 317 is the formal Bloch solution of (D.M for kN.M of the resonant pair of the points kN.
Q E r.. The function ljr(x. y) for a = 1 the final form of the Floque set can be represented in the following form [Kr3].Ps.51) al and the intervals [Ps. 2] do not intersect.2] and [.r 1).M: 3°.17). If there is a finite number of handles that are glued then the corresponding curve is compactifed by one point and the corresponding Bloch function is the Baker Akhiezer function on the compacti. M).) Let us define the Riemann surface r (Jr) for any admissible set 7f. IPs.M. S. that are normalized by the con dition ljr(O. if JrN Re Ps. Theorem D2.P. Novikov 2o. The following theorem states that the finitegap potentials are dense in the space of all periodic smooth functions in two variables ([Kr3]).fied by only one point P1 (k = oo).Ps. If the admissible set 7f contains only a finite number of pairs.I. s = (N. .I.2] and by sewing after that the left side of the first cut with the right side of the second cut and vice versa. The set of pairs of complex numbers Jr = {Ps. 0. Outside these domains the function 1jr is holomorphic and has no zeros. The potentials u for which r (Jr) has finite genus are called finitegap and as it follows from the last statement of the theorem they coincide with the algebraic geometrical potentials.318 B. the Bloch solutions of the equation (D. It has one simple pole in each of the domains R1N. I. Dubrovin.M• kN.i = .I. . i = I. y)ks). on r(Jr).MI· In the domain Ro it has go poles. y) which can be analytically extended into some neighborhood of the real values x. 49) as the twosheets covering of the small neighborhoods of the pairs kN.2] there corresponds a nontrivial cycle a. 2.4) ljr(x. (Here ks are resonant points (0. Krichever.I. Q) which is normalized by the condition ljr(O.A.4) with a = 1 are parametrized by points Q of the Riemann surface r(Jr) corresponding to some admissible set Jr.l = o(lk. y.k. Q) = 1 are meromorphic on r. In the case of real and smooth u (x. y. where go in general position (when Ro is smooth) equals the genus of Ro). 1 . Ps. A Riemann suiface Ro (with boundary) of a finite genus go. y.. M) .MI that are defined by the equations of the form (D. Their poles do not depend on x. It is obtained from the complex plane of the variable k by cutting it along the intervals [p•. For any real periodic potential u(x.(x. Let us fix some finite or infinite subset S of integer pairs (N > 0. y. ( After this surgery for any cut [Ps. (0. Ps.2} where s E S would be called "admissible". The Bloch solutions of(D. in the neighborhood of which the Bloch function 1/r has the form: 1/r = ekx+k2y (1 + ~. Q). 0. Ps. A set of "handles" R1N.fied Riemann suiface. then r(Jr) has finite genus and is compacti. Q) = 1 is meromorphic on rand has a simple pole Ys on each cycle a•. Ps.i.
Usp. Zbl. Nauk 46. Mat. 145 (1991)). Usp.: Evolution of Whitham zone in the KdV theory.580 18 [Bo] Bobenko A. R. [D2] Dubrovin B. .. Mat.A. Sci. 10361039 (1985) (English translation: Sov. Phys. Novikov S. 2.712.P. Anal. 225238 (1970). AN SSSR (English translation: Sov. 49.V.58022 [DI] Dubrovin B. Math. Mat. Math.J. Teor.. Each smooth periodic potential u(x.655.345349 (1987). No. Dokl.Dokl.4. Zbl. y can be approxi mated by finitegap potentials uniformly with any number of derivatives. 1620. Math.35072 [Bogo] Bogoyavlenskij 0. Math. Fiz. ( 1987)). Zbl. 51. Paris 3/4. Novikov S.240244 (1980)). Studies Appl. Ser. 564566.. Krichever I. Serie I.: The evolution of multiple modes for nonlinear dispersive waves.: Rational solutions of KdV type equations in dimension 2+ I and mbody problems on the line.: Evolution of Whitham zone in the Kortewegde Vries Theory. Shaposnik F.Fiz 70. Poisson brackets of hydrodynamic type. Acad. 11()(}. No. Commentarii Math. 79. 32. No. Phys. JETP 52.G. Phys. No..58032 [DN2) Dubrovin B. [DS] Drinfeld V. Phys.: Classical vortex solution of the Abelian Higgs model. No. 5759 (I 989). 109113. Dokl. Phys. Zbl. y) of the equation (D.P. Dokl.: On perturbations of periodic Toda lattice. 32. Frobenious algebras and Lie algebras. D 14. AN SSSR 295.: Surfaces of constant mean curvature and integrable equations. 2998 (1989) (English translation: Russian Math Surveys 44. Novikov S. Teor. Magnetoc Bloch functions and vector bundles. No 3.. Prilozh 23.. AN SSSR 253. (Contemporary problems in Mathematics) 24. 10061016 (1980) (English translation: Sov.A.: Hydrodynamics of weakly defonned soliton lattices. [D] Doyle P. List of Additional References [AC) Aharonov A.4. Novikov S. A (3) 19. 19752036 (1985)).219223 (1987)).P. AN SSSR 283..Dokl.606.Math..: Ground state of a spin1/2 charged particle in a twodimensional mag netic field. 13141338 (1993). 2.. 35124 (1989)). 309314 (1987) (English translation: Theor. No. Lecture Notes in Math.. 22.A.4) with Re a i= 0 analytically extendable to a neighbourhood of real x.778. [AN] Avilov V.: Lie Algebras and equations of the Kortewegde Vries type. C. 12931297 (1980) (English translation: Sov.A. Phys. Dokl. 34'2 (1991) (English translation: Russian Math Surveys 46. Probl. [DN I] Dubrovin B. Casher A. Integrable Systems I 319 Theorem D3.A.65132 [BN] Balinskij A. Dokl. Poisson brackets of Hydrodynamic type and Differential Geometry.639.: Ground states of twodimensional electron in a periodic mag netic field.V. Rev.4. 32.P. No. ltogi Nauki Tekh.: Differential geometric Poisson bivectors in one space variable. SpringerVerlag.1. 24612462 (1979). Novikov S..P. Sov..: Lectures on 2D Topological Field Theory.Mat. Phys. Math. 70. Moscow 1984 English translation: J. Eksp. [AB] Ablowitz M.M.P.V.G. 120 348 (1996).A. 1992. Dokl. Novikov S. [ANK] Avilov V.A. Zh. [DN3] Dubrovin B. 366368 (1987)). 6.W.. Novikov R.. 2.1106 (1976).: On the differentially geometric Poisson brackets on the lattice. No.: Ground states in a periodic field. [Bog] Bogdanov L.6.P. Funkts. Sokolov V. J. 201209 (1976). I.. 34. Phys. Sovrein. [dV] DeVega H.V. Nauk 44. VINITI.: NovikovVeselov equation as a natural twpdimensional generalization of the Kortewegde Vries equation. Sov. [FN*] Fran~oise J. (English translation. 228231 ( 1985)).. 30. Rev.511516 (1980)). Zbl. Benney D.
: Rational solutions of the VeselovNovikov equations are reftectinless two dimensional potentials at fixed energy.699. Anal.P. [GD] Gelfand I. 15.2. Funkts. Modern Phys. [GDo] Gelfand I.35188 (Kr4] Krichever I. Appl. Zbl.. Novikov S. 9. Moscow.Novikov) [FFM] Flaschka H. 2340 (1981) (English translation: Funct.276285 (1985)). 3358 (1996). No. J. 9. 20.M. Nauk 44.617. Comrnun. Anal.L.657.: Twodimensional periodic difference operators and algebraic geometry. 2.V. Soc..415429 (1992). No. Appl. No. iimethod and nonlinear equations. I. Zbl.M. 200213 (1988)). Anal. 4. A 7.2. No.: PoissonLie groups and classical Walgebras. 259274 (1976)). Akad.P. Ser. No. 65. Pitaevskii L. No.: "Inverse scattering problem" for the two dimensional Schrooin ger operator at a fixed negative enery and generalized analytic finctions. 47. Funkts. 69. Zbl. Nauk 36. Usp.. V. Prilozh.B. Funkts.2. Intemat. Fiz.94103 (1986)). 22.G. Zbl. (G] Grinevich P. Surv.2. Manakov S. Anal.: Hamiltonian Methods in the Theory of Solitons. Anal. Knorrer H. Zbl. Differential geometry and applications. matrix modles and topo logical field theories. Anal..• Forrest M. Pure Appl. Usp...47035 [ON2] Orinevich P. No. (ed by S. Appl.G. 307310 (1986) (English translation: Theor. 19. 853876 (1992).. 3. 19. I.35121 [GN•] Grinevich P. Prilozh 2.35088 [Kr5] Krichever I.: Spectral theory of commuting operators of rank two with pe riodic coeffitients. 121184 (1989). 14525 (1989)). 170. No. . S. Funkts.Ya.M.811. Anal.. Prilozh.P. Dokl. Comrn. 1987).: The potentials with zero reflection coefficient on the background of the finite gap potentials.O. Anal. 7277 (1981) (English translation: Russian Math. [F] Ferapontov E.1172 (1986)).: Cohomology of Lie algebra of vector fields on the circle. Math. No.34022 [Kr2] Krichever I. 15.L. Zbl.P. I.: Riemann surfaces of the infinite genus.A.M.: Fractional powers of operators and Hamiltonian systems.A.A.D.: Method of averaging for twodimensional integrable equations. 16.: The Taufunction of the universal Whitham hierarchy. 3752 (1988) (English translation: Funct. 1986 (English translation: SpringerVerlag. Math.514.. [FKT] Feldman J. Ehksp. Phys 143. Anal.M.617. Math.333. 10.M. Dorfman I.320 B.: Spectral theory of twodimensional periodic operators and its applications. Funct.. 1927 (1988) [GP] Gurevich A. Appl. Anal. Anal. Anal. Funkts. 2526 ( 1982) (English translation: Funct. Phys. 161163 (1975)). Teor.4. I.3. Surveys 36. No. 2. Anal. Amer.: The periodic nonabelian Toda lattice and twodimensional generalization. Prilzh.M.35121 [GM] Grinevich P. [Krl] Krichever I.G. (En glish translation: Funct. No. No.437475 (1994).. No. Appl. Appl. Appl.: Hamiltonian operators and infinite dimensional Lie algebras. 173187 (1981). Trubowitz E. No. Zbl.. Novikov S. [GF] Gelfand I. Appl. No.M. Mat. Dikii L.G. 1170. 739784 (1980).?? (1981)) [Kr3] Krichever I. Krichever. 1424 ( 1986). Translations. [FL] Fateev V. New York.: Nonlocal hamiltonian operators of hydrodynamic type. Heidelberg. Funkts.G. Takhtadzhyan A. 3. Nauk SSSR 285. 7778 (1975) (English translation: Funct. 3242 (1985) (English translation: Funct. Funkts. Mat. 22. 3136 (1985). Teor. 2. Funkts. Fuks D. No.M. Pure Appl. Fiz. 590604 (1973). 9293 (1968). Zh.M. Math.35122 [GNI] Orinevich P. Anal.: Multiphase averaging and the inverse spectral solution of the Kortewegde Vries equation. Math. 22..2. Novikov R. (English translation: Russ. No.M. No.688. 33.: Analog of multisoliton potentials for the twodimensional Schrodinger operator. Math. Anal.O.: Inverse scattering for the twodimensional SchrOdinger opera tors. and McLaughlin D. Novikov [IT] Faddeev L. Mat.. 4. 1920 (1982)). Comrnun.: Nonstationary structure of the collisionless shock wave.753. 10. Prilozh. No. Prilozh. Prilozh.V. Nauka. 44. Lukyanov S. Berlin.. Zbl. preprint ETH Zurich. Zbl. 3.5. Dubrovin. 69. 20.35095 [Kr6] Krichever I. Prilozh.: The dispersionless Lax equations and topological minimal models.58064 [Kr7] Krichever I. 16. I. Anal. 1328 (1976) (English translation: Funct.
4.A. Zbl. Surv.. No. Mat. No. No.. 298303 (1981). 245246 (1976). 809830 (I983) [MO] Marchenko V. The method of av eraging method for fieldtheoretical systems. 11011150 (1995)).. JETP) [Mo] Mokhov 0.43357 [KT] Knorrer H. No. Hilley E. H. Math. Dokl. 21. Soc..: Nonlocal hamiltonian operators of hydrodynamic type asso ciated with constant curvature metrics. No.. Ehksp. Saveliev M. Pure Appl.: SturmLiouville operator and applications.2. 22.V.: Magnetic Bloch functions and vector bundles. 67.: The geometry of conservative systems of hydrodynamic type.M. Novikov S. No. 12.1. Levermore C. Math.5. No. [Mi] Mikhailov A. 21. SpringerVerlag. 6. No. No. 336337 (1988)) . 489494 (1979). [L) Laplace P.: Limit for h + 0 of Heisenberg equation.659. No. Mos cow 1986 (English translation: Encyclopaedia Math. [KN I) Krichever I..960. Anal. Kiev. 188. Riemann surfaces and strings in Minkov sky space. Zbl.: A directional compactification of the complex Bloch variety. Funct. 1997. Surv.P. 8598 (1985)) Zbl. [N] Novikov S.: Representation of zero curvature for the system of nonlinear partial differential equation Xa.: Periodic and conditionaly periodic analogues of the manysoliton solutions of the Kortewegde Vries equation.P. 3. Ostrovski 1. 3. Fund.2.P. Usp. Integrable Systems I 321 [KBBT) Krichever 1. 23. Mat.58069 [KLWZ] Krichever 1.1.zz = exp {K XI a· Lett. Phys. [Nat2) Natanzon S. Wiegmann P.48601 [KZ] Krichever 1. Zh. II. 191192 (1990) (English translation: Russian Math Surveys 45.. Zbl. Commun. Teor. Anal. Nauk 50. Zh. 1320 ( 1992).443448 (1979) (English translation: Sov. Riemann surfaces and soliton theory. 2) 170. Helvetici 65. Math. III.3.. Appl.M. Mat.: The small dispersion limit of the KortewegdeVries equation. JETP 40. Trubowitz E. 83119 (1995). Amer. Phys. Math.M.: Topology. Phys. Teor.: The method of inverse scattering problem and twodimensional evolution equa tions. S. 21. Mat. 3.: Recherches sur le calcul integral aux differences partielles. (La) Lax P.P. Appl.4. Usp. Trans!. Com ment. 22. 4.P. 4. Babelon 0. preprint Landau Intitute. Chernogolovka. 267304 (1997). 378383 (1969). Geom. 218219 (1990)) [MP) Maltsev A. J.: Spin generalization of the RuijsenaarsSchneider model.: Real nonsingular finitezone solutions of soliton Equations. Usp. Probl. Math. ltogi Nauki Tekh. [Mas] Maslov V. 1996) (N2] Novikov S. Prilozh.: On the Poisson brackets of the DubrovinNovikov type (ONbrackets). Zabrodin A. 153184 (1996) (ed. No. Zbl. Funkts.. . No. 4761 (1987) (English translation: Funct.. Sciences. Ferapontov E.M. Nauk 45. No. Commun. 36. Nauk 40.: Virasorotype algebras.: Quantum integrable systems and discrete classical Hirota equations. Histore de I' Academie de Sciences (1977).. Fiz. 1. No.: Averaged Poisson brackets. [MF] Mokhov 0. (Natl] Natanzon S.. Anal. 571593.M. (Ser. 114149 (1990). 294307 (1987)).6.A. Zabrodin A.2. I. Anal. Naukova Dumka (I 977) (in Russian) [Ma] Manakov S. Prilozh. 349389 (1997). Teor. Lipan 0. Novikov S.: Vrrasorotype algebras.V. Usp. 213F2144 (1974) (English transla tion: Sov. Diff. 356 (1995) (English translation: Russ. Appl.. Math. Math. 12. nonabelian 20 Toda chain and representations of the Sklyanin algebra.4.P. Vol. No. 26. Phys. [KN2) Krichever I. Typical dispertion relations and their quantum numbers. 40. (ser. AMS Trans!. Sov.. Socr. Mat.. Funkts.17012 [KP) Krichever I.970.: see in the book Marchenko V. 7989 ( 1985) (English translation: Russ. 10551063 (1974)).: On the integrable geometry of soliton equations and N = 2 su persymmetric gauge theories. 50. Phong D. Anal. Talon M.843. Mat. Fiz.P.: On the integrability of the twodimensional generalization of Toda lattices. 30.: Differential equation for Prym fuction. Funkts. VINITI. Anal. Fiz. Ekspr. A criterion for twodimensional finite gap purely potential SchrOdinger operator to be real. 2) 170. Prilozh.: Spin generalization of the CalogeroMoser system and the Matrix KP equation. 45. 6. Nauk 31.654. 4663 (1987). (Nl] Novikov S. [LS] Leznov A... Math. 9293 (1988) (English translation: Funct. Novikov). V.76004 [N3] Novikov S. Napr. 253290.
No.: Laplace transformations and exactly solvable twodimensional SchrOdinger operators. Moscow.33 (1986)).. Math. 2.P. 43. Mat. 10481068 (1990). 27. 121 (1993). 83. 1952 (1994). Dubrovin.: Modified NovikovVeselov equation and differential geometry of surfaces. Dokl. 7172 (1995) (English translation: Russian Math Surveys.B. S. No. [SV] Shabat A. monopole condensation.: Electricmagnetic duality. I.: Liouville form of the averaged Poisson brackets.: Geometrie Differentielle projective des reseans. Novikov [N4] Novikov S. [SWI] Seiberg N. Commun. [NV] Novikov S.. Nauka.. 33. Generalized Hodpgraph method. I. Dokl.R. 1977 [YaR] Yanenko N. Math. Dokl.. 238261 (1965).2.M. Witten E. Rozhdestvenskij B.P. Dokl. Usp. Anal.A. New York.B..5. Pri lozh 14. A) pairs and Ricatti type transformations. [Wh2] Whitham G. Zbl. Witten E. Nauk 43. 2. and confinement in N = 2 supersymmetric YangMills theory. Nuclear Phys. 534537 (1987).606. Sokolov V.B. [Ts] Tsarev S. [T] Tsiseika G. Math. I. Math. [Tail] Taimanov I.: Poisson brackets and onedimensional systems of hydrodynamic type.) 54.484550 (1994). Veselov A. J. translations.629 .621.5. Nauk 50. AN SSSR 286. I. Prilozh. Invent. Math.: On a class of local translation invariant Lie algebras.: Geometry of the Hydrodynamic type systems. Zbl. [Ti] Tian F. 1924. and chiral symmetry breaking in N = 2 supersym metric QCD.A. No. Ser. Mat. 211212 (1988) (English translation: Russian Math. No. John Wiley Interscience Publ. Nuclear Phys. No.I. Math.: Characterization of Jacobian varieties in terms of soliton equations.: On Poisson brackets of differential geometric type. AN SSSR (ser. Notes 50. duality. Pisa 1992 (published in 1994) [NM] Novikov S. 34.P.: Prym theta functions and hierarhies of nonlinear equations.: Linear and nonliear waves. Amer. Nauk 48. 216218 (1987)).: Nonlinear dispersive waves. Doklady 35. Phys.: Monopoles. Sov. [Tai] Taimanov I. Lezzio ni Fermiane. 12941297 (1987) (English translation: Sov.: Periodic fixed points of Backlund transformations and the KdV equation. lzvest.P. 133151 (1997). [Whl] Whitham G. B 426. Math.: The initial value problem for the Whitham averaged system.: Systems of quasilinear equations and their applications in gas dynamics.322 B. [Ts I] Tsarev S. ParisBudapest. Funkts. 155156 (1993). No. 26472676 (1986). Mat. Mat. No.B.. 7980 (1980).6.35097 [SS] Shabat A. 3942 (1986) (English translation: Sov. 1986 (second edition) [Z] Zehnanov E. 30. K. 50 (1991) (English translation: Math. Maltsev A. [W] Weiss J.. Phys. 723730 (1991).P. mat.: A dressing chain and the spectral theory of the SchrOdinger opera tor.A. No.. 179.short notes) [PI] Potemin G.P. Proc Royal Society A 283..53010 [P2] Potemin G. Zbl. No.: Solitons and Geometry. 27. Usp. AN SSSR 292. Veselov A.P. Funkts. 333382 (1986). . 79115 (1994). Academia dei Lincei. Zametki. B 431. Anal.richever.: (L. 5.V. Soc. Usp..P. [SW2] Seiberg N. Scuola Normale Superiore. 252253 (1988)) [Shi] Shiota T.: Algebrageometrical construction of selfsimilar solution to Whitham equation.17002 . Surv.
reference to reviews in Zentralblatt fiir Mathematik (Zbl. in which the mechanism for integrating the KdV equation which had been proposed in the pioneering paper [60] was cleared up.V. have. The use of the symmetry of Hamiltonian systems to construct their integrals and to reduce their order also goes back to the classical works of Jacobi. In the papers [84]. [17]). to Poisson. [103] appeared. 5.). Hamilton. [59]. [39]. The fundamental material of section 4 (chap. and an example of a rational family was met with in the paper [I]. and also in the theory of the Korteweg<ie Vries equation [54]. The question of whether one can approximate an arbitrary periodic potential by * For the convenience of the reader. [7]. [1]. [96]. Noether. 4. Beginning with the paper [89]. For the anisotropic LandauLifshitz equation. The concept of a completely integrable Hamiltonian system arose in the works of Bour and Liouville (see the textbooks [7]. 2. Integrable Systems I 323 References* I. as far as possible. Jacobi. [47]. This line received a further development in the papers [30]. proposed in [127]. Poincare and others. for example. appendix 2). More complicated examples arose in the hydrodynamics of a perfect incompressible fluid (see (7]. 6. [ 42]) and surveys (see [112]). the text [7]). compiled using the MATH database. included in a natural way all examples known up till then. appendix 5). [25]). been included in this bibliography. The construction of integrals for fieldtheoretic Lagrangian systems with symmetry was given by E. We do not discuss here degenerate completely integrable systems with a larger number of integrals than the number of degrees of freedom (see. Different versions of the presentation of these classical concepts are to be found in quite a number of textbooks (see. [29]. [42]). (Later the papers [102]. all schemes for producing integrable evolution equations have been based on representing them in the form of a compatibility condition for the auxiliary linear problems. (These examples and a number of others are presented together with the history of the development of the first stages of the inverse scattering method in the books [115]. . [64]. [56]. The program for integration of the periodic problem for the KdV equation was initiated by the paper [Ill] (somewhat later and in a less effective form it was considered in [90]). 3. The general concept of a Poisson bracket of hydrodynamic type was introduced and studied in [48]. In the article [80] (for greater detail see [83]) a representation for the MoserCalogero system was proposed in which the dependence of the matrix entries on a parameter defined on an elliptic curve contained essential singularities of a special form. the papers [22]. for example. [57] and V. [128] first used a zerocurvature representation for families of operators with a spectral parameter on an elliptic curve. for example. The modern formalism of Poisson brackets in application to . in particular such key stages in the development of the method as [54]. The scheme based on the equations of"zero curvature for rational families of operators". for example.) For the sineGordon equation finite gap solutions were constructed in [68]. Infinitedimensional analogues of the Hamiltonian formalism until recently were considered only for Lagrangian field systems in connection with the needs of quantum field theory (see. for a modern presentation see [101] (see also [7]. A representation of the KdV equation in the form of the zero curvature equation for polynomial families of operators was first proposed in [111]. [86] another way was proposed of generalizing the zerocurvature equations for rational families to the case where the spectral parameter is defined on an algebraic curve of nonzero genus. Kozlov's survey article [72]). [88]. 1) is contained in the classical works of Hamilton and Jacobi (see. [59]. and Jahrbuch iiber die Fortschritte der Mathematik (FdM.infinitedimensional (field theoretic) systems was systematically developed in [112]. [45]. The basic concepts of the Hamiltonian formalism go back to the classical work in analytical mechanics. Lie. The employment of the methods of algebraic geometry for the construction of periodic and quasiperiodic solutions of the KdV and nonlinear Schrodinger equations was begun in the articles [38].).
P.l. The construction of variables of the angle type for the elliptic MoserCalogero system and the construction of all elliptic solutions of the KP equation were obtained in [80]. The algebraicgeometric Floquet spectral theory of linear operators with periodic coefficients was developed in the publications [38]. Without any connection with finitedimen sional systems. [75]. The program of the research on the dynamics of the poles of solutions of equations to which the inverse scattering method is applicable goes back to the paper [87]. [118]. [51].A. A construction of actionangle variables for the Hamiltonian systems connected with finite gap SturmLiouville operators was first obtained in [6]. The definition of such functions. 7. In the literature list which follows. 9. This theory made it possible to examine from a single point of view and to unify not only the Hamiltonian structure itself of diverse systems. [56]. [45]. Translator's Remark. [44]. In these cases I have supplied my own translation of the Russian title and have indicated how the title of the English translation differs. [24]. for equations for which the auxiliary linear problem is not selfadjoint. [100]. The relation of the stationary and nonstationary Hamiltonian formalisms for these systems was obtained in [18]. including the' multipoint ones. whenever a Russian work has been translated into English a reference to the translation has been included. elliptic solutions of the KdV equation with three poles were constructed in [47]. [81]. unless (as is not infrequent!) the title of the translation is incorrect or differs significantly from the Russian title. The concept of the ClebschGordanBakerAkhiezer function became the central concept of this scheme. but also to give a unified construction of variables of the actionangle type. Novikov finite gap potentials of a SturmLiouville operator with conservation of the period was settled positively in [100]. References to works devoted to the algebraicgeometric integration of a number of classical systems of mechanics and hydrodynamici are cited in sections 3 and 4 of Chapter 2 in the course of the analysis of a series of prime examples. [41]. The starting point for these works was the problem of constructing periodic solutions of equations of the KdV type. The construction of the algebraicgeometric spectral theory of the Schrodinger difference operator was begun in the articles [ 45].M.1! also included in a natural way the constructions of solutions of onedimensional evolution equations which were proposed in the works cited above. In [31] this result was carried over to the elliptic case. [50]. Their connection with a joint eigenfunction of a pair of commuting operators of relatively prime orders was first noted in [9] by Baker. A general Hamiltonian theory of systems whose integration is connected with hyperelliptic curves was proposed in [116]. in which the discrete Peierls model was integrated and its perturbations were investigated. The first stage of the theory of finite gap integration was presented in [ 45]. I. [ 115]. For Kovalevskaya's system a construction of variables of the action type was obtained for the first time in just these articles. Akhiezer indicated examples of the interpretation of such functions in the spectral theory of operators on the halfline. Krichever. 10. [46]. The isolation of the real nonsingular solutions within the framework of the general scheme. N. [115]. A general scheme for integrating twodimensional equations of the type of the Kadomtsev Petviashvili equation with the aid of the methods of algebraic geometry was proposed in [74]. . [90]. was given in [75] on the basis of a generalization of the analytic properties of Bloch functions of finite gap periodic and quasiperiodic operators. 8. was begun in [28] and was earnestly pushed forward in [13]. The connection of the dynamics of the poles of rational and elliptic solutions of the KdV equation with the rational and elliptic MoserCalogero systems was first discovered in [4]. Dubrovin. [33] and received its completion in [79]. The isomorphism of the rational MoserCalogero system and the polar system of rational solutions of the KP equation was established in [77]. These results were used in the papers [23]. "Singlepoint" functions of this kind were introduced as a formal generalization of the concept of exponentials in the 19th century by Gordan and Clebsch (see [8]). [64]. S. The possibilities for applying the algebraicgeometric spectral theory to the continuous PeierlsFrohlich model were discovered in [12]. [34].324 B. and the title I have given is then simply the title of the English translation. [104]. [20]. [Ill].
Berezin. 45. 219248 (1979). 455. 431 p. 64. LOMI. Moscow 1976. Kaup. FdM 28. No. 2. 10. Pure Appl.70001 8. Zbl. Zbl. I. 45. A 118..... John Wiley & Sons. 48. D. 0. Proc. Belokolos. Math. D.331 9. Funkts.: Abelian Functions (the full title is: Abel's Theorem and the Allied Theory Including the Theory of the Theta Functions). Prepr.. 6466 (1986) (English translation: Funct. I have not corrected one "mistake" which is nearly universal in translations of the subject matter treated in this article.. V. Math. Ablowitz. 51. (English translation: Graduate Texts in Mathematics 60.P.584593 (1928). II: Teor. M.I. New YorkTorontoLondon 1953/1953/1955. It is the lazy translation "finitezone" (a literal translation of the Russian term) for what English writers generally call . N. 38. ITF84128 R. Bar'yakhtar.: Some remarks on the universal enveloping algebra of a Lie algebra.. Ser. etc. Phys.2. Shirkov.35075 7. Zbl. V. McKean. Leningrad 1983." by J. Nauk SSSR /41. Zbl. I. A. Zbl. potentials.: Euler equations in the algebras e(3) and so(4). 34. Appl. Burchnall and J. II: Theor.A. 28 p.: A continuous analogue of orthogonal polynomials on a system of intervals.). 507. Zbl. Euclidean Lie algebras and curves. E.J. Nauka. 30. Dokl. 12621264 (1973). H. E883. Akad. Fiz. Third Edition.. I: Teor. 50. H. Belokolos. 6069 (1981) (English translation: I: Theor. Moser. M. No.l. H. Prilozh.: Completely integrable systems. F. Anal.: Note on the foregoing paper "Commutative ordinary differential operators. Al'ber. 302/396/292 p. 267317 (1980).34023 . 20. R. Integrable Systems I 325 However. 53.63 12. Zbl. McGraw Hill Book Company. Appl. (English translation: WileyInterscience. Appl.D. Anal. Phys. J. 479 p. Zbl. Pure Appl. Zbl. Nauka.F. Math. 30. 109. 912 (1976) (English translation: Funct. Lett. Prilozh.: Mathematical Methods of Classical Mechanics.).: On a trace functional for formal pseudodifferential operators and the symplectic structure of the KortewegDevries type equations.W.58017 4. Prepr. Lond. Mat. 11201127 (1982)). Adler. Funkts. Moscow 1974.: Onedimensional magnetic structures arid the higherorder LandauLifshitz equations (in Russian). 227. 17. 10.296 6. Phys..35024 5. 456. 604610 (1981)) 13.2. Funct. Bikbaev. New YorkHeidelbergBerlin 1978. 1. 268275 (1980). 259212 (1981). Airault. 9295 (1976)). Anal. Anal. 27 p. S. Bateman.1. 271282 (1982) (English translation: Theor.439 10. Common.: On stationary problems for equations of Kortewegde Vries type. P. H. Math. 338. Invent.35058 3. Because "finitezone" is so frequent (although it is found almost exclusively in translations from the Russian) I have left it unchanged in the English titles but wish to draw the reader's attention to it here. Fiz. Fiz. 393.: Rational and elliptic solutions of the Kortewegde Vries equation and a related manybody problem.I. 1. Teor.22020 15. Cambridge 1897. No. No. 620 p. P.58010 16. Soc. Math. Belokolos.F. Funkts. 386.C. editor): Higher transcendental functions (in three volumes). A.J. 684 p.: On finite gap integration of the LandauLifshitz equation. The XYZ case (in Russian). Zbl. A. H. Adler.: PeierlsFrohlich problem and potentials with finite number of gaps.. Kiev 1984. Bobenko. Ehnol'skij. Akhiezer. 52. 2.L.V.G.. 622. Math. 91102 (1967)). l. V. Bobenko. E.D.: Integrals of higherorder stationary KdV equations and eigenvalues of the Hill operator.I. Baker. 14091412 (1961)). Arnol'd.35060 14. Zbl. E. Baker. Isomorphisms of integrable cases. Golod. SpringerVerlag. N. 5356 (1986)). 2. 263266 (1961) (English translation: Sov. 114 (1967) (English translation under the title: Some remarks about the associated envelope of a Lie algebra. Anal. Math. Adv. A.. Phys. 54.) 18. Dokl. II.. Prilozh. No. Bogolyubov (Bogoliubov). Cambridge University Press. H. van Moerbeke.295 Zbl. New YorkChichesterBrisbaneToronto 1980. 20. M. Rev.: Method for solving the sineGordon equation.I. Math. Anal. 11. 1.: Generalized Lamb ansatz.finite gap (operators.: Introduction to the Theory of Quantized Fields. Bogoyavlenskij.2.. FdM. 462 p.95148 (1977). Common. No. Mat. (Erdelyi. 359. 48. Newell. Chaundy. 10221026 (1980).303 Zbl. R.I. Segur. Mat. 53.Z.D.N.F.
II. Zbl. Zbl. 10. 0. Trans!. 285287 (1983) (English translation: Sov..N.58017 37. Springer Verlag.J. Math. Math.G. BerlinHeidelbergNew York 1980. Caudrey. LV.(English translation: Funct.V.58012 20.M. 38.M.P. Math. Dubrovin. 753755 (1974) (English translation: JETP Lett. Phys.. Phys. Nauk SSSR 268.58006 31. D. Dokl.E. Nauk SSSR. l. Prilozh. Novikov. Ehksp. Anal.: An exact solution of the Peierls model with an arbitrary number of electrons in the unit cell.: Reality conditions in "finitezone integration". Funkts. I. I. ll04ll08 (1980) (English translation: Sov. Math. Nauk 30. I. F. Bogoyavlensky. Funkts. Mat. Ehksp. Zbl.A.A. I.: Nsoliton solutions of the nonlinear LandauLifshitz equation. Dubrovin. Anal.5. Bogoyavlenskij.: Commutative ordinary differential operators.: Discrete Peierls models with exact solutions. Prilozh.P. Pis'ma Zh. Zbl. note. No.. 28.)). S.JETP 56. (editors): Solitons. 338. 31. Zbl. Dzyaloshinskij. Phys.486491 (1980) (English translation: JETP Lett. 212225 (1982)) 24. Teor. Bogoyavlenskij. Lond. 629632 (1978) (English translation: JETP Lett.: On perturbations of the periodic Toda lattice. R. I. Commun. R. Krichever. Drinfel'd. II. 3. T. Burgatti. E. Izv. 51. Ser. Math. Usp.: Integrability of the equation of a twodimensional asymmetric chiral 0(3)field and of its quantum analog. S. No. Anal. Mat. I. 201209 (1976) 22. P. Chaundy. Kirova. Soc. Cherednik.A.00010 26. Phys.326 B. A.. Teor.35022 . V. Choodnovsky. what is being referred to is in fact a parametrized family of functions. 10. 48. 33..278282 (1981) (English translation: Sov. Dryuma. 19. 9.. Akad. 144145 (1981)) 30. No. Funkts.58007 35. J.439 27. 339353 (1977) 32. Math. 27. Ann. G. Nauk SSSR 252. Surv.: Analogue of inverse scattering theory for the discrete Hill's equation and exact solutions for the periodic Toda lattice. Theor. S.. 457465 (1976) 34. Appl. Zh. 31. P.420440 (1923). Appl.L. 9.: On commutative subrings of certain noncommutative rings.3. 30. FdM 49. 912 (1977)). 913 (1976) (English translation: Funct. 321 (1983) (English translation: Russ. 428. 11..3. 122 (1983) (The title of the English translation speaks of YangBaxter bundles..: Integrable Euler equations on Lie algebras arising in problems of mathematical physics.1.V. Fiz. Ser.A. 207257 (1985)). 108ll1 (!" semestre 1911) 28. Lie bialgebras and the geometric meaning of the classical YangBaxter equations. No. 77113 (1975)). Anal. 28. Gel'fand. Brazovskij. Gordyunin. V. 359. J.. 215223 (1975)). Phys. S. Brazovskij. 55. Teor. 25. 66. 450452 (1980)). B.A.V. Dokl. No..557583 (1928).1.: The abstract Hamiltonian formalism for the classical YangBaxter families. 1: Proc.: The relationship between Hamiltonian formalisms of stationary and nonstationary problems. Fiz. Nuovo Cimento B 40. Lond. A 118.3ll FdM 54.G.S. 536.. Prog.. Anal.: Hamiltonian structures on Lie groups. S.: Pole expansions of nonlinear partial differential equations. No. Borovik. Akad. Dikij. Zbl. Soc. USSR.: Sulla integrazione delle equazioni di HamiltonJacobi per separazione di variabili. Novikov 19. 21.: Asymptotic behaviour of the resolvent of SturmLiouville equations and the algebra of the Korteweg<le Vries equations.A. Burchnall. 4151 (1975).14011 36. S. Nauk 38. I. Dokl. Aka d. Tanaka. No. 345.700ll 21. Usp. 583. Cherednik. 398415 (1909) 33. Yad. 389 p. Cherednik. Dokl.1. 83.K.A. 25. Krichever. Rend. translating the corresponding Russian word literally but inappropriately. Mat. Zbl. 526. 33. Zbl. 883938 ( 1984) (English translation: Math. 6871 (1983)). Dall'Acqua. Fiz. Ser.. 581584 (1978)) 23. Appl. Zh. Zbl.: Determinazione dell'equazioni di HamiltonJacobi integrabili mediante Ia separ azione delle variabili. 0. Topics in Current Physics 17. 389415 (1982) (English translation: Sov. Fiz. Izv.. Ehksp.W. I. Ehksp. 67100 (1975) (English translation: Russ. 811 (1976)). Pis'ma 19. 11. Accad.1. V. 491. L. Drinfel'd. Nucl. Choodnovsky.M. R.: Periodic problems for the Korteweg<le Vries equation in the class of finite band potentials.M. II: Proc. 456461 (1980)) 25. 0. 387388 (1974)) 38.5. Bullough. Prilozh. Fiz. Anal. Date.V.: Analytic solution of the twodimensional Kortewegde Vries (KdV) equation. Lincei 20. Pis'ma Zh. Surv. 334.E.35044 29. 1114 (1977) (English translation: Funct. Gel'fand. N. Teor.
. Springer Verlag.. Phys. Mat..57001 43.: Completely integrable Hamiltonian systems associated with matrix operators and Abelian varieties. Sci. Akad. Dubrovin. II.P. Phys. Saclay 1982. 19241925 ( 1974). Moscow 1979. 326. 554. 441.A. 10311040 (1983)) 52..M. Moscow 1978. Fiz. Funkts.2. 591.D. No.P. 16. Sov. L. Anal. Sov. Phys. A. Anal. No. The geometry and topology of manifolds. Slavnov.D. B.4. Dubrovin. in: Itogi Nauki Tekh.: Sound and chargedensity wave in the discrete Peierls model. Novikov. 561. Ehksp. 27. and the BogolyubovWhitham averaging method..: The Schrodinger equation in a periodic field and Riemann surfaces. Dubrovin.58043 42. 486. 51. Appl. Dokl.T. 2050 (1985)). Dokl. Fomenko. Zbl. 55136 (1976) (English translation: Russ.P. Zbl. Prog. I. Novikov. Nauk SSSR 229. Sovrem.58008 40.53052 54. Zbl.: On Poisson brackets of hydrodynamic type. 781785 (1983)(English translation: Sov. Dokl. No. Ph. 1827 (1971) (English translation: Funct. Prilozh.JETP 58.A. 464 p. 665669 (1983)(in the translation.P. Dubrovin.: Hamiltonian formalism of onedimensional systems of hydro dynamic type. 257. 294297 (1984) (English translation: Sov.A. 17. Matveev. B. Math. Zbl. Usp. B.: Integrable models in I+ I dimensional quantum theory. Funkts.. 82/76.E. Zh. L.: Theta functions and nonlinear equations... I. Dubrovin. Krichever. 97 p.D. the name Whitham is given incorrectly as "Whitman")). Zbl.58012 50. B 9..E. 3. 232 p. S. Fiz. Dokl. and Abelian varieties. 534.). 239 p. B. B.. Nauka.53002. Mat. Faddeev.M.. 17711789 (1983) (English translation: Sov. Zh.: Matrix finitezone operators. Mat. B. Math. B.759 p. (English translation: Gauge Fields: Introduction to Quantum Theory.35100 47.A. V. V. (Contemporary Problems of Mathematics) 23. Math. 2743 (1982)(English translation: Funct.908913 (1982)) 51.35011 46..35011 49. S. New YorkBerlin HeidelbergTokyo 1985. B. Moscow 1983. Krichever. No. Novikov. Krichever. S. 31. Natanzon. Akad..430 p. Part II..: Modem GeometryMethods and Appli cations. Akad.M. 703716 (1974) . Novikov. S. Graduate Texts in Mathematics 104. Flaschka.: Introduction to the Quantum Theory of Gauge Fields. Zbl. 3378 (English trans lation: J.P. S. Nauk SSSR 270. Zbl. Zh.A.A. A. Anal.: Nonlinear equations of Kortewegde Vries type. I.A. C. Graduate Texts in Mathematics 93..JETP 56.. L. Appl. Nauk SSSR 279. finitezone linear operators. 59146 (1976)). Krichever.. Math. 529. Anal. Faddeev. Rev.: Real twozone solutions of the sineGordon equation. Anal. Math. 36. Teor.A.B. 947951 (1976)). Faddeev.). II. 16. transformation groups. Nauk 31. Reading. VINITI. 2841 (1977) (English translation: Funct.35021 44. No. 1150 (1982). Frontiers in. 30. Fiz. Teor. Phys. Anal. and fields.JETP 40. Zbl. 651654 (1984)). Zbl.M. Surv. Nauka. 85. 377. I. 1. Theor. (English translation in two parts: Part 1: The geometry of surfaces. 5. Prilozh. Zbl.E.inversescattering solution.: The Toda lattice. 53. Dubrovin. Novikov.. CEN Saclay. 565.. Dubrovin. Phys. Surv. Appl. and On the Toda Lattice.. Math. 15761586 (1982)(English translation: Sov. Dokl.P.4. B. Ser. 553. Dokl.: Periodic and conditionally periodic analogs of the manysoliton solutions of the Kortewegde Vries equation. No. Dzyaloshinskij. Preprint S. SpringerVerlag. Novikov.58038 41. Dzyaloshinskij. S. 28. 280287 (1971)). T. Novikov. No.. 265277 (1977)). 67. I. 1180 (1981) (English translation: Russ. S.58002 45. Usp. B. Existence of integrals. I.2133 (1982)). 83. Ehksp. II. Zbl.. H. Rev.. Zbl. 5. Probl. Nauk 36. S. New YorkBerlinHeidelbergTokyo 1984.: Kortewegde Vries equation: a completely integrable Hamiltonian system.2. 1518 (1976) (English translation: Sov.35074 55. Mass. Math. 1192 (1981)). Funkts. BenjaminCummings. 1980. II.P. Dubrovin.A.1. Teor. Integrable Systems I 327 39. Ehksp. 1. Dubrovin.A. Dubrovin.A. Zbl. Dubrovin. II. Sect.: Topological and algebraic geometry methods in contemporary mathematical physics. B. 10581063 (1974)) 48. Zakharov. Prilozh.A. 21312144 (1974) (English translation: Sov. 478. Rev.M.Physics 50.: Commensurability effects in the discrete Peierls model. Phys.
525. Kirchhoff. Zbl. Math. lzv. /09. 12981300 (1982) (English translation: Sov. Kozlov. Anal. 337.A. Mat.: Methods of Qualitative Analysis in the Dynamics of a Rigid Body (in Russian). Gardner.on finitedimensional Lie groups.328 B. (English translation: Grundlehren der mathematischen Wissenschaften 220. Akad. Wiss.: Algebraicgeometric construction of the ZakharovShabat equations and their periodic solutions.. Moscow 1953. Moscow 1972.: Sur une classe de systemes differentiels abeliens deduits de Ia theorie des equations lineaires. No. M. Palermo 43. J. A. 167 (1984)).W. Zbl.. Garnier. Preuss. Prilozh. Fiz.A. Leipzig 1876. V. No.R. Nauk 38. IV. Kotter. 466 p. Nauk Ukr. V. 5181. A. Zbl. 371389 (1978)). Dokl. 336 p.35050 63. FdM 8. R.I.V. Zbl. Math. Fomenko. I. No. 12. 346. Mishchenko. I. Philadelphia 1953. 26.70009 72. F...: Integrability and nonintegrability in Hamiltonian mechanics. C. D. 42..: Die von Steklow und Liapunow entdeckten integrablen Faile der Bewegung eines starren Korpers in einer Fliissigkeit.. Trofimov. No. Its. Zbl. SpringerVerlag. 1226 (1977)). Moscow 1980.977 70. 315 p. Akad. 367 (1983) (English translation: Russ. J. Martin. 1531 (1977)(English translation: Funct. Theor.35021 60. 155191 (1919). Mat. 240 p.: Euler equations . 71. Krichever. Math. FdM 47. Moscow State University. Krichever. 197212 (1978).70023 73.58006 58. Mat. Nauka. Nauk SSSR 227. Dokl. 11. Circ. Zbl.404 62.: Uber die Bewegung eines festen Korpers in einer Fliissigkeit. C. Leningrad State University.: Almost periodic solutions of the equation u.: Bounds on the number of eigenvalues of the Schrodinget operator. Mechanik.G. 223. lzv.: Quantization and unitary representations. Zbl. McLaughlin. 51.. Mat.V.A. Leningrad 1976.. Funkts. Nauk SSSR 266. Anal. 7987.: Elements of ths: Theory of Representations. Kozel. Lett. Teubner. SSR. in: Probl.. D. 356 (1984) (English translation: Russ.A.: A class of solutions of the Kortewegde Vries equation (in Russian). No. Green. 394397 (1976)). A. Zbl.58024 59.V. Grosse.: Method for solving the Kortewegde Vries equation.M. Dokl. 361. Phys. 59. Math.S. V. 89111 (1892). 55. 878881 (1976).P. Nauk 39.: Nonintegrability of Kirchhoff's equations.: Liouville integrability of Hamiltonian systems on Lie algebras. Matveev. 65. 287 p. 291294 (1976) (English translation: Sov. Krichever. /9. A. Flaschka. A. Nauk SSSR.2.35007 75. Dokl. Kotter. 12. Novikov 56. Kruskal. (English translation: Coronet Books. 176 (1983)). A no.M.. Ser. Notes Math.: Vorlesungen iiber mathematische Physik. A. 230 p. Sitzungsber. USSR. Phys. Miura.. Part 1: Prequantization. in: Lecture Notes in Modern Analysis and Applications Ill. Glaser. Appl. 170. (Problems in Mathematical Physics) 8. Kozlov. G. SpringerVerlag. 549. Commun. Reine Angew.495498 (1982)~ Zbl.: Lectures on Integration of the Equations of Motion of a Rigid Body about a Fixed Point. Akad. Rend. V. Ser. Dokl.70009 74. Prog.542 66. 383. Phys. 10951097 (1967) 61. B. 396415 (1978) (English translation: Math. V.P.uxx+sin u=O (in Russian). H. Kozlov.B. 38. JJ. Math. V. Zbl. Rev. Gostekhizdat. I.: Integration of nonlinear equations by the methods of algebraic geometry. 15481551 (1971). Zbl.. Zbl. Kirillov.22011 67. H. Akad. K. 7092.. 17.151 64. R. V. Kostant. Lect.). 283.53028 68.S. F.). Usp. B.2.35028 . Surv.T.. Golubev. Fomenko.V. 541. Phys. Onishchenko. Mat. 438456 (1976) 57. V. Dubrovin.: Kortewegde Vries equation and generalizations.T.M. The Korteweg<le Vries equation as a Hamiltonian system. Akad. Zbl. Surv. Kotlyarov. 87208.S. I. V. Math. Math. J. S. 557. FdM 23. 264. BerlinHeidelbergNew York 1970.: Canonically conjugate variables for the Kortewegde Vries equation and the Toda lattice with periodic boundary conditions.V. /0. BerlinHeidelbergNew York 1976. 373. Usp.35003 69. I. Berlin 1900. 39. Gardner.
I. Bologna 1950/1951/1952. Probl. 15. Potsdam. 501. Appl.34008 19. Prilozh.M. 32. Funkts. 408. 276286 (1978)). Rev. No. Anal. Magri. Zbl. 6183.35017 88. LeviCivita.: Holomorphic fiberings and nonlinear equations.. 382. 4..70008 78. Zbl. Surv. Abschnitt. 2031 (1978)(English translation: Funct. Providence 1974. Krichever. Lect. 59. Nuova edizione riveduta e corretta. N. Integrable Systems I 329 16. Commun. 183208 (1977) (English translation: Russ. 7678 (1978) (English translation: Funct. 14. (unter Mitwirkung von F.35065 . N. 255256 (1978)).: Algebrogeometric spectral theory of the Schriidinger difference operator and the Peierls model. (reprint: B. 1. 393. 14. 28. FdM 35. No. 248263 (1982)) 82. 6.123. Krichever. Zbl. No. Finite zone solutions of rank 2. Krichever. No. 21. Funkts. Appl..: Periodic solutions of the KdV equation.35061 85. Anal. Novikov. Ann. Nauk SSSR 247. Zbl. T. American Mathematical Society. I. Phys.368 94. Anal. Math. 28.G. Math. No. 1. Krichever. I. No. Zbl. Anal.: The Korteweg{je Vries equation and related evolution equations. II Parte 1. Funkts. G. Commun.: The Peierls model.35087 84.35004 91. Moscow 1983. Krichever. S. 99124 (1971) 89. Prilozh. 16. Mat. Math. Math. 141188 (1975).G. Zbl. 4. Krichever. Akad. Prilozh. in: ltogi Nauki Tekh.M. Zbl. LeipzigBerlin 1930). Akad.3337 (1979) (English translation: Sov.J2.: Lezioni di Meccanica Razionale.D. 462. Math. Funkts. 35.6. 194198 (1982)). Prilozh.: Algebraic curves and nonlinear difference equations. J. Zbl. 20. Zbl. 43. Teubner.: Elliptic solutions of the KadomtsevPetviashvili equation and integrable systems of particles. pp. Prilozh. Anal. Amaldi. 4768 (1980) (English translation: Russ. Krichever.: Integrals of nonlinear equations of evolution and solitary waves. Math. I. 11561162 (1978).D. I. LeviCivita.: Methods of algebraic geometry in the theory of nonlinear equations. Appl. Nauk 32.4. Mod. I.: Rational solutions of the KadomtsevPetviashvili equation and integrable systems of N particles on a line. No. I. I. 26. 1026 (1982) (English translation: Funct. in: Non linear Wave Motion (Proc.Y.M. 374. Nauk 33. of Technology.M.39001 83.. Anal. Pure Appl..M. 5379 (1980)). Engel): Theorie der Transformationsgruppen.. Krichever. 10541058 (1982) (English translation: Sov. Lie. 12.: A simple model of the integrable Hamiltonian equation. Zbl. Lamb. B. S.: Holomorphic bundles over algebraic curves and nonlinear equations. 12. 19.L: Analytical descriptions of ultrashort optical pulse propagation in a resonant medium. Anal. Phys. 5190 (1985)). Zbl./2. 632/554/830 p. 295. Krichever. Krichever. 4554 (1980) (English translation: Funct. Mat. 4. 215216 (1978) (English translation: Russ.39003 80. 1912). 175185 (1978)). Zbl.111.3. 650{. Kruskal. Nauk 35.411 90. Zbl. Surv. 16. S. I. Lax. 282290 (1980)). Zbl. Appl. Leipzig 1888/1890/1893. I. No. Anal. 372. Math. U.: Commutative rings of ordinary linear differential operators. (Contemporary Problems of Mathematics) 23. 434. Math. Pure Appl.35071 87. Anal. Anal. I. 816/510/673 p. 12. Dokl.54 (1979)). 79136 (English translation: J. 383397 (1904).M. 292. 45. 162. Usp. M.M.: Holomorphic bundles over Riemann surfaces and the KadomtsevPetviashvili (KP) equation.: Sulla integrazione della equazione di HamiltonJacobi per separazione di variabili. Mat.35002 11. AMSSIAM Summer Sem.6. P. Math.35080 81. No. P. Ser. Novikov. Math. 1: FdM 20. 12. Novikov.1 (1978)). I.362 92. Surv.6. Clarkson Coli..M. Mat. II Parte 2. Zbl. 527. Math. Sovrem.P. F. Usp. T. Teubner. 185213 (1977)). Krichever. Zbl. 59{. 33. I. 47. Dokl. Appi. Lax. 595. 4. 383. Sov. 467490 (1968). S. Nauk SSSR 265. VINITI.M.173 93. No. Zanichelli.P. Funkts. Appl.P.M.: Nonlinear equations and elliptic curves. Dokl. 4152 (1978) (English translation: Funct. No.D. Dokl.M. Usp.35078 86.
S. Probl.: Various aspects of integrable Hamiltonian systems. 89. Ser. Zbl.5. Grinevich. Moser. Sovrem.3. Usp. Manakov. Bressanone.P.34022 101. Its. /17 (159). Adv. Martin's Press. 120 (1985)). Novikov. S. New York 1960. Phys. I. Anal. S.. Bordag. Trubowitz.: Reduction of symplectic manifolds with symmetry. No. Funkts. S. 299. No. B 13. H. Ehksp.217274 (1975). Appl. 834848 (1977) 105.58028 106. Zh. 10. Phys. Fiz. Zbl. Marsden.: Integrals of geodesic flows on Lie groups.: The method of the inverse scattering problem and twodimensional evolution equations (in Russian). Prilozh.19531973 (1978) (English translation: Sov. Maki. H. 660 p. Nauk 31. No. Zbl. Weinstein.V.R.P. Fiz. Math.V. 506. Appl. 143226 (1976). Macmillan. Prog. Zbl.34019 110. 197220 (1975).M. M. 8. MilneThompson.. Manakov. Zakharov. 22. 564.: Relativistically invariant twodimensional models of field theory which are integrable by means of the inverse scattering problem method. A. 343. V..A. 28. 30. Funct. 205206 (1977) 100. Math. Zbl.. 3.JETP 47. Anal. Nov. VINITI. Ser.22022 109. Teor. Zbl. 5466 (1974) (English translation: Funct. 339. H. Appl.V. Novikov. 16. Matveev. Mat. 473485 (1983)). in: Dynamical Systems.35083 114.. Manakov. 332 (English translation: J. Funkts.P. (English translation: Operator Theory: Advances and Applications. Anal. Zakharov.: Theoretical Hydrodynamics (Fourth Edition).. Rev. 7377 (1970) (English translation under the incorrect title: Integral geodesics of a flow on Lie groups. Boston 8.70003 98.P. Mat. Nauk 37. Prilozh. V. 399. Zbl. Naukova Dumka. S. Sb. BostonBasel Stuttgart 1980. June 1978. Zbl. S. 328329 (1976)). I. /8. Marchenko. Krichever. V. Math. Lett.330 B.: Exact magnetic ringing solutions in superfluid 3 HeB.58011 Ill.. 319.: The periodic problem for the Korteweg<ie Vries equation. S. Math. 241. 571. 29242930 (1976) 96. L. 10171027 (1978)) 107. Math. Mat.: Twodimensional Schrodinger operators in periodic fields. Moscow 1983. No. 245246 (1976). Mat. Funkts. P. No. Invent. Novikov. 156 (1982)).B. J.A. 481493 (1982) (English translation: Math. Surv.. 16..E. Manakov. Mishchenko.P. A. Anal. Anal. Kiev 1977. Commun. Phys. N.E.A. Appl.35017 112.: Complete integrability and stochastization of discrete dynamical systems. USSR.V. 4.34024 103. 236246 (1974)). J. Sov. Lectures. Zbl. 16. 543555 (1974) (English translation: Sov. Novikov.I..: The spectrum of Hill's equation.P. Italy. 2526 (l982)(English translation: Funct. Zbl.). P. van Moerbeke. in: Itogi Nauki Tekh. E. Ebisawa.580 ll 113. Zbl.E. 5. A 63. Math. 327. A.426 108. Rep. Usp.34024 104. Phys. Zbl.G.. 514. 4. J. Anal.. McKean. 9394 (1976) (English translation: Funct. Zh. 8. Zbl. S. Birkhauser. S. 233289. Birkhauser. Dubrovin. (Contemporary Problems of Mathematics) 23. 349 (1982)(English translation: Russ.M.. Anal. J. C. 232235 (1970)).: Twodimensional solitons of the KadomtsevPetviashvili equation and their interaction. No. McKean.: Hill's operator and hyperelliptic function theory in the presence of infinitely many branch points.N.47035 .: Note on the integration of Euler's equations of the dynamics of an n dimensional rigid body. 37. 1920 (1982)). Vol. Math. V.V. Ehksp.58005 102. pp. Moser. Math. 4.5..: Spectral theory of commuting operators of rank two with periodic coefficients. Meiman.V.: The Hamiltonian formalism and a manyvalued analogue of Morse theory. L. Funkts. Prilozh. Anal. Mikhajlov. 45.5. 367 p. Novikov 95. Teor. Pure Appl. London/St.269274 (1974)) 97. Meshcheryakov. Prilozh.: The integration of the equations for geodesics of leftinvariant metrics on simple Lie groups using special functions. Sb. Phys. 10.M. 468. 29..P.: Three integrable Hamiltonian systems connected with isospectral deformations.: The theory of onedimensional Schrodinger operators with a periodic potential. Zbl.35055 99.JETP 40. A. Zbl. No. 331 p. Phys. BaselBostonStuttgart 1986. K. 345. 67.: SturmLiouville Operators and Applications.S.. 121130 (1974). 74. 303.
1957. 226235 (1974)). Steklov Inst.P. E. 46. 613. Inst.. Phys. 26.58022 119.: Integrable Hamiltonian systems connected with graded Lie algebras. Prilozh. 8687 (1982) (English translation: Funct.P. 8.P.: Integrable systems of classical mechanics and Lie algebras. Zbl. V. Integrable Systems I 331 115. Ehksp... A. Anal. Inst. Zbl. Manakov. A. Ehksp. twodimensional. Veselov. A. Anal.276 p... Prepr. 488. Zbl. Steklova 95. Ser. Teor. Math. Anal. Ehksp.: Exact theory of twodimensional selffocusing and onedimen sional selfmodulation of waves in nonlinear media. Zbl.. Zbl. Zakharov. Anal. Phys. 565. Leningrad 1979. Perelomov.A. Anal.. Mass.M. Steklov. Dokl. Fac. Zbl.: On complete integrability of the LandauLifshitz equation (in Russian). Perelomov. Fiz. Mat. SemenovTyanShanskij. Math.B.4. 118134 (1971) (English translation: Sov.: On Poisson brackets compatible with algebraic geometry and Kortewegde Vries dynamics on the set of finitezone potentials.G. Zakharov. 64 p. 13. E379.P. Nauk SSSR 266. 555.: Integrable Hamiltonian systems and interactions through quadratic con straints.35089 129. Appl. 3. Inst.P. Semin. Zbl. Moscow 1983. A. Shabat. 535. S. Commun. Zakharov. /7.: Finitezone.: Poisson brackets and complex tori. Perelomov.: Evolution equations with nontrivial conservative laws. Khar'kov 1893. Anal.35072 131.V.K. Steklova 165. 17. Consultants Bureau [Plenum].E.M. Leningr. 78. II. I. 207221 (1976) 123. K. 145256 (1909) .. 263270 (1982)). No. 147. 513. 32 p. 1983. 165. Zakharov.70016 120.E. The motion of a rigid body about a fixed point (in Russian). Shabat.2024 (1984) (English translation: Sov.E. 62{i9 (1972)) 126. M. /5. 357362 (1982)). V. Prilozh.58028 130. W. 4961 (1984) (English translation: Proc. Dokl. 2734 (1982) (English translation: Funct.: Integration of nonlinear equations of mathematical physics by the method of inverse scattering.35020 118. Rejman... (Stekloff. Zbl. Funkts. Nauk SSSR 279. Sci. Veselov. /6. III. Zap.. Dokl. Math.: Some remarks on the integrability of the equations of motion of a rigid body in an ideal fluid. Fiz. Fonkts.P. Ann.A.Phys.B. No. L. 4353 (1974) (English translation: Funct. 495.. Zbl. Sokolov. 122. Svinolupov.35090 128. Zbl.P. Anal. No. Sklyanin.4. Appl. Springer. Akad. Steklov. Math.: A scheme for integrating the nonlinear equations of mathematical physics by the method of the inverse scattering problem. 1733 (1983) (English translation: Funct. No. Math. /6. 508. E. 121. 3. potential Schrodinger operators. 5365 (1985: 3)).: Theory of Solitons. A. Moskva. Zbl. No. (English translation: Contempor ary Soviet Mathematics. Sklyanin. V.. 319 p.35024 127. Sov. Appl. Prilozh. Prepr. Novikov. Tr. 354 (1980) (English translation: J.66 132. Zh. 598.P. Zbl. Math. Mat. No. Shabat. Inst. 166174 (1979)). Zbl. Prepr. Moscow 1980. S.. Prilozh. V. 30. Prilozh. Novikov.35003 116. 61. Toulouse. 259272 (1983)).A. 16.): Sur le mouvement d'un corps solide ayant une cavite de forme ellipsoidale remplie par un liquide incompressible et sur les variations des latitudes. 2. Nauka.. 63 p.V. Reading. Funkts. Dokl. AddisonWesley.35106 117. Pohlmeyer. 133. 13. Teor. Moscow 1983. S.: Integrable systems of classical mechanics and Lie algebras.: What is a classical rmatrix? Funkts.. Appl. Systems with constraints (in Russian). Teor. Funkts. Novikov. Akad. S. /. A. 15.: Some algebraic structures connected with the YangBaxter equation. A. Anal. Anal.: On the motion of a rigid body in a fluid (in Russian). 1322 (1979) (English translation: Funct. New YorkLondon 1984. 15071545 (1982)). The Inverse Scattering Method. 533537 (1982) (English translation: Sov. A. Nauchn. No.M.58031 125. Explicit formulas and evolution equations.JETP 34. 98 p. Moskva. V. Anal. 8385 ( 1981) (English translation: Funct. 8. Anal.K.4. Anal. A.). A.70013 124. 144146 (1981)). 448. V.E. V. 307 p. Appl. 1983. 317319 (1982)). S. 303. ll6.. Fiz. LOMI. Novikov. /6. 449. Prilozh. Pitaevskij (Pitaevskii). S.l.B. Appl.: Introduction to Riemann Surfaces. Funkts. G. No. Veselov. Otd. 588591 (1984)). 19.
1286 142.P. Nauk SSSR 279. Takhtadzhyan. Zbl.34001 137. Ehksp.: On integrability conditions for the Euler equations on so(4).: Integrability of the Novikov equations for principal chiral fields with a multivalued Lagrangian. A. Dokl. Funkts. 108110 (1974)) . Veselov. A. Cambridge 1937. Phys.A.P.: Finitezone potentials and integrable systems on a sphere with quadratic potential. I. Phys. Zh. Math. 3139 (1980)).P..JETP 39. H. 393395 (1984)) 140. Zbl.JETP 38. Veselov. Nauk SSSR 270. L. 10971100 (1984) (English translation: Sov. 29.58013 139. Nauk SSSR 276.: On stochastization of onedimensional chains of nonlinear oscillators. A. Veselov. Krichever. 14. 458459 (1983)) 138. Weber.A. 66. E. Dokl. No. 740742 (1983)). Nauk SSSR 270.T. Dokl. Dokl. 14. 27. Phys.: Cnoidal solutions of the LandauLifshitz equation for twosublattice magnetic materials. Dokl. 574. 4850 (1980)(English translation: Funct. Veselov.P. Dokl.. Veselov. Zakharov. Zh. Whittaker. 1. Akad. Anal. Fiz. Dokl. Dokl.E. Ehksp. Phys. 173206 (1879) 141. A. 476489 (1974) (English translation: Sov. Prilozh.: Exact theory of propagation of ultrashort optical pulses in twolevel media. Appl. Akad. 456 p.590593 (1984) (English translation: Sov. 10941097 (1983) (English translation: Sov. Akad.. 28. Cambridge University Press.332 B. Takhtadzhyan.: Anwendung der Thetafunctionen zweier Veninderlicher auf die Theorie der Bewegung eines festen Korpers in einer Flussigkeit. Math.P. 228233 (1974)) 135. 65.M.A. 539. Sov. S. Novikov 134. Dubrovin. 29. Teor. 12981300 (1983) (English translation: On integrability conditions for the Euler equations on S0(4). Akad. 14. L. V.. Phys. Teor.: The LandauLifshits equation and integrable systems of classical mechanics. 994996 (1984)) 136. Anal.P. Ann.: A Treatise on the Analytical Dynamics of Particles and Rigid Bodies (Fourth Edition). A. Fiz..219225 (1973) (English translation: Sov. FdM 63.
233 .vector field 79 .o f quantum physical quantities 146 Fock space 167 completely integrable Hamiltonian system 51.operator 228. Index Abel map 235 form 74 .operator 146 characteristic class 126 equation of zero curvature 221 number 127 equidistant mapping 87 characteristics of a hypersurface in a contact Euler case 212 manifold 83 equation 61..structure 71 action form 27 . npoint 236 theorem for contact manifolds 71 Bargmann isomorphism 167 for Lagrangian fibrations 37 BlattnerKostantSternberg kernel 166 . 66.for Legendre fibrations 72 Bloch function 242.. 182 .BakerAkhiezer function 277 even coordinates 15 5 coadjoint representation of a Lie group 30 exact Lagrangian embedding 81 cobordance 119 cobordism of fronts 119 field of characteristic directions of a hyper coisotropic submanifold 35 surface 45 commutation equation 217 finite gap integration 228. 277 divisor 235 BohrSommerfeld condition 161 dual hypersurface 88 BohrSommerfeld subvariety 160 .momentum tensor 209 caustic 91 .Lagrange system of equations 43 Clebsch case 212. front 87 210 functions in involution 142 configuration manifold 43 contact Darboux coordinates 71 Garnier system 258.. 186 chiral fields 190.solution 229 complete set of physical quantities 143 first integral 143 .representation 180 . 210 . 277 compact symplectic group 19 . 213 Poisson system of equations 47 .function 163 contactification 80 algebraic curve 231 contactomorphism 71 analytic Poisson bracket 248 cotangent bundle 27 armed front 121 Darboux coordinates 5 BakerAkhiezer function.function 142 Casirnir function 33 .realization of a Poisson structure 64 canonical Poisson bracket 183 elliptic family 224 . 223 .Hamiltonian 79 abelian differentials of the second kind 241 manifold 71 "actionangle" variables 53. 259 element 72 Gauss mapping 117 .symplectic structure 27 energy 142 transformation 147..transformation 255 .
of translation of the argument 59 case 212 modified BohrSommerfeld condition 161 Lagrangian boundary 123 momentum mapping 62.. 20 I . 44..of quadratic Hamiltonians 8 vector field 31. 217 group 203 .equivalence 92 multivalued fimctional 191 . 214 . 206 . 118 Jacobi's theorem 85 index 115.o n JR2n 164.mechanical system 43 .cobordism 123 MoserCalogero system 227. 31. 265 .hypersurface 89 .submanifold 71 HamiltonJacobi method 213 transformation 44. 181 Leggett's equations 205.variety 234 Maslov class 116.of hypersurfaces 90 mapping 91 generating fimction 6.fimction 31. 184 Hamiltonian 8. 158.Poisson structure 33 hyperelliptic curve 239.structure on a manifold 169 Kortewegde Vries equation 20 I .of configurations 43 map 235 .system 8. 142 Leibniz identity 181 operator 7 Lie algebra of first integrals 63 picture 143 ..of contact elements 72 .fimction 43 npoint BakerAkhiezer fimction 236 . 119 natural mechanical system 43 . 148 .fibration 36 . 260 involutive submanifold 35 isotropic submanifold 35. 142 Magri bracket 252 manifold of characteristics 48 Jacobi identity 31.Darboux theorem 5 hierarchy of equations of Lax type 220 .on integrable systems 52 integration formula 70 local fieldtheoretic brackets 189 inverse scattering method 177 LyapunovSteklovKolosov case 213. 179 . 35. 247 Liouville picture 143 Liouville's theorem 31.picture 146 . 253 method of HamiltonJacobi 213 of separation of variables 214 Lagrange bracket 80 .superalgebra 155 equation 146 linear approximation of a Poisson group 158 structure 33 . 142 .of infinitesimal contactomorphisms 79 .cobordism 124 equivalence 89 halfdensity 161 . 88. 203 .Grassmann manifold 6. 201 LiePoisson bracket 183 Hamilton's equations 143 Lie structure 80 Heisenberg algebra 163 .fibration 71 halfform with values in a bundle 165 mapping 87 HamiltonJacobi equation 85.334 Index generating family of fimctions 92 manifold 35. 210 integral of a Hamiltonian system 194 . 168 mean value of a physical quantity 146 Kiihler polarization 156 metamorphoses of caustics 108 kinetic energy 43 of wave fronts 106 Kirchhoff case 212 metaplectic group 164 Kirchhoff's equations 186 . 183. 44..submanifold 35. 142 generators of a canonical action of a Lax equation 59. 169 Kovalevskaya case 212..representation 59 geodesics on an ellipsoid 215 Legendre boundary 123 gradient mapping 91 .
state of a system 142 158.. 204 state space 145 bracket 8.manifold 32 supermanifolds 155 pair 57 symmetry of a Hamiltonian system 204 . 156 universal Maslov class 118 realization of a Poisson structure 64 reduced Hamilton function 63 vacuwn vector 166 . 181 StUrmLiouville operator 240 . 201. 180.of EulerPoisson equations 47 pure state of a system 142. 142 oriented Lagrangian Grassmann manifold 117 SchrOdinger equation 146 Ostrogradskij transformation 184 picture 146 Ostrogradskij's theorem 48 skeworthogonal complement 5 skeworthogonal vectors 5 pairing 164 skewscalar productS PeierlsFrohlich model 278 space of !jets of functions 72.of E. 141. 78 . 168 . 232 twisted cotangent bundle 38 real hyperelliptic curve 247. 210 space 165 . 147 principle of least action 45 .theta function 228 observable quantity 142 odd coordinates 155 Schouten bracket 32. 73 phase space 44. Index 335 Newnann system 247 relative Darboux theorem 24 E. 251 sphericalized cotangent bundle 79 physical quantity 141 standard contact structure 71 Planck's constant 145 .. 209 .of hydrodynamic type 194 subordinate subalgebra 159 . 145. 209 ..symplectic space 5 Poisson action of a connected Lie group 61. 141.vector field 63 Williamson's theorem 9 .phase space 63 Planck constant 147 Weil representation 166. 23.equilibriwn 66 normal mapping 91 Riemann surface 233 . 155.structure 32 symplectic group 17 Poisson's theorem 32 leaf33 polarization 37. 182 prequantization 148 structure 5. 145. 32. 156 . 22 of the sphere 153 symplectification 77.homologous to the identity 39 projectivized cotangent bundle 72 system of EulerLagrange equations 43 pseudoKaJlier polarization !56 . 250 polarization 142.of Poisson 32 rank of a Poisson structure 33 transverse polarizations !57 rational family 221. 146 tangential mapping 87 quadratic Hamiltonian 8 theorem of Jacobi 85 quantization 148 of Liouville 31.on integrable systems 52 quantwn anomaly 147 . Noether 32. Noether's theorem 32. 31.linear transformation 7 potential energy 43 manifold 22.of the torus 154 symplectomorphism 22.