Grundlehren der
mathematischen Wissenschaften
A Series of Comprehensive Studies in Mathematics
Series editors
M. Berger P. de la Harpe F. Hirzebruch
N.J. Hitchin L. Hrmander A. Kupiainen
G. Lebeau F.H. Lin S. Mori
B.C. Ng M. Ratner D. Serre
N.J.A. Sloane A.M. Vershik M. Waldschmidt
EditorinChief
A. Chenciner J. Coates S.R.S. Varadhan
347
For further volumes:
www.springer.com/series/138
Camille LaurentGengoux r Anne Pichereau
Pol Vanhaecke
Poisson Structures
Camille LaurentGengoux
CNRS UMR 7122, Laboratoire
de Mathmatiques
Universit de Lorraine
Metz, France
Anne Pichereau
CNRS UMR 5208, Institut
Camille Jordan
Universit Jean Monnet
SaintEtienne, France
Pol Vanhaecke
CNRS UMR 7348, Lab. Mathmatiques
et Applications
Universit de Poitiers
Futuroscope Chasseneuil, France
ISSN 00727830 Grundlehren der mathematischen Wissenschaften
ISBN 9783642310904 (eBook)
ISBN 9783642310898
DOI 10.1007/9783642310904
Springer Heidelberg New York Dordrecht London
Library of Congress Control Number: 2012947315
Mathematics Subject Classification: 17B63, 53D17, 17B80, 53D55, 17B62
SpringerVerlag Berlin Heidelberg 2013
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publishers location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.
Printed on acidfree paper
Springer is part of Springer Science+Business Media (www.springer.com)
For Clio
Preface
Poisson structures naturally appear in very different forms and contexts. Symplectic manifolds, Lie algebras, singularity theory, rmatrices, for example, all lead to a
certain type of Poisson structures, sharing several features, despite the distances between the mathematics they originate from. This observation motivated us to bring
the different worlds in which Poisson structures live together in a single volume,
providing a multitude of entrances to the book, and hence to the subject. Thus, the
idea of the body of the book, Part II, was born.
It is well known, i.e., it is commonly agreed upon by the experts, that results and
techniques which are valid for one type of Poisson structure ought to apply, mutatis
mutandis, to other types of Poisson structures. When starting to write Part I of the
book, we soon realized that not everything could be derived from the general concept of a Poisson algebra, so we faced the challenge of presenting the concepts and
the results in detail, for the algebraic, algebraicgeometric and geometric contexts,
without copypasting large parts of the text two or three times. But as the writing
moved on, both the algebraic and geometric contexts kept imposing themselves; finally, each found its proper place, appearing as being complementary and dual to
the other, rather than a consequence or rephrasing, one of the other. It added, unexpectedly, an extra dimension to the book.
It was pointed out by one of the referees that the main applications of Poisson
structures should be present in a book which has Poisson structures as its main
subject. This was quite another challenge, giving rise to the third and final part of
the book, Part III, undoubtedly an important addition.
Several years were necessary for this project, a big part was done at Poitiers,
when the three of us were appointed to the Laboratoire de Mathematiques et Applications. We were spending long hours together in what our colleagues called
The Aquarium (for an explanation of the name, look up poisson in a French dictionary). When two of us moved away from Poitiers, we sometimes worked together in other places, which always offered us a pleasant and stimulating working
atmosphere. Thus, we are happy to acknowledge the hospitality of our colleagues
from the mathematics departments at the universities of Poitiers, Antwerp, Coimbra,
vii
viii
Preface
SaintEtienne and at the CRM in Barcelona and the MaxPlanck Institute in Bonn.
Each one of us was partially supported by an ANR contract (TcChAm, DPSing and
GIMP, respectively), which made it possible to keep working on a blackboard, rather
than seeing each other on a computer screen.
We learned Poisson structures from our teachers, friends and collaborators: Mark
Adler, Paulo Antunes, Pantelis Damianou, Rui Fernandes, Benoit Fresse, Eva Miranda, Joana Nunes da Costa, Marco Pedroni, Michael Penkava, Claude Roger,
Pierre van Moerbeke, Yvette KosmannSchwarzbach, Herve Sabourin, Mathieu
Stienon, Friedrich Wagemann, Alan Weinstein, Ping Xu and Marco Zambon. The
multitude of different points of view on the subject, which we learned from them,
have given a quite specific flavor to the book. Even if a book cannot replace what one
learns through lectures, discussions and collaborations, it is our hope that this book
may further transfer what we learned from them and that it invites other researchers
to explore the subject of Poisson structures, which turns out to be as diverse and rich
as the colorful fauna and flora which inhabit the bottom of our oceans.
April 1, 2012
Camille LaurentGengoux, Anne Pichereau and Pol Vanhaecke
Contents
Part I Theoretical Background
1
Poisson Structures: Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Poisson Algebras and Their Morphisms . . . . . . . . . . . . . . . . . . . . . . . .
1.1.1 Poisson Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.2 Morphisms of Poisson Algebras . . . . . . . . . . . . . . . . . . . . . . . .
1.1.3 Hamiltonian Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Poisson Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 Affine Poisson Varieties and Their Morphisms . . . . . . . . . . . .
1.2.2 The Poisson Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.3 The Rank of a Poisson Structure . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 Bivector Fields on Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.2 Poisson Manifolds and Poisson Maps . . . . . . . . . . . . . . . . . . .
1.3.3 Local Structure: Weinsteins Splitting Theorem . . . . . . . . . . .
1.3.4 Global Structure: The Symplectic Foliation . . . . . . . . . . . . . .
1.4 Poisson Structures on a Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
4
4
5
6
7
8
10
13
15
16
19
23
26
32
34
35
Poisson Structures: Basic Constructions . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1 The Tensor Product of Poisson Algebras and the Product
of Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1 The Tensor Product of Poisson Algebras . . . . . . . . . . . . . . . . .
2.1.2 The Product of Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . .
2.2 Poisson Ideals and Poisson Submanifolds . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Poisson Ideals and Poisson Subvarieties . . . . . . . . . . . . . . . . .
2.2.2 Poisson Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Real and Holomorphic Poisson Structures . . . . . . . . . . . . . . . . . . . . . .
37
38
38
42
46
46
49
52
ix
Contents
2.3.1
Real Poisson Structures Associated to Holomorphic
Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 Holomorphic Poisson Structures on Smooth Affine
Poisson Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Other Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1 Field Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.2 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.3 Germification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
55
56
56
57
59
60
61
MultiDerivations and Kahler Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1 MultiDerivations and Multivector Fields . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 MultiDerivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.2 Basic Operations on MultiDerivations . . . . . . . . . . . . . . . . . .
3.1.3 Multivector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Kahler Forms and Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Kahler Differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.2 Kahler Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.3 Algebra and Coalgebra Structure . . . . . . . . . . . . . . . . . . . . . . .
3.2.4 The de Rham Differential and Cohomology . . . . . . . . . . . . . .
3.2.5 Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 The Schouten Bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.1 Internal Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.2 The Schouten Bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.3 The Algebraic Schouten Bracket . . . . . . . . . . . . . . . . . . . . . . .
3.3.4 The (Generalized) Lie Derivative . . . . . . . . . . . . . . . . . . . . . . .
3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
64
64
65
67
68
69
70
71
72
74
75
76
79
84
85
87
88
Poisson (Co)Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.1 Lie Algebra and Poisson Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.1.1 Lie Algebra Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.1.2 Poisson Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.2 Lie Algebra and Poisson Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.2.1 Lie Algebra Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.2.2 Poisson Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3 Operations in Homology and Cohomology . . . . . . . . . . . . . . . . . . . . . 101
4.4 The Modular Class of a Poisson Manifold . . . . . . . . . . . . . . . . . . . . . . 102
4.4.1 The Modular Vector Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.4.2 The Modular Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.4.3 The Divergence of the Poisson Bracket . . . . . . . . . . . . . . . . . . 107
4.4.4 Unimodular Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 109
4.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Contents
xi
Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.1 Lie Groups and (Their) Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.1.1 Lie Groups and Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.1.2 Lie Group Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.1.3 Adjoint and Coadjoint Action . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.1.4 Invariant Bilinear Forms and Invariant Functions . . . . . . . . . . 121
5.2 Poisson Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.2.1 Algebraic Poisson Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.2.2 Poisson Reduction for Poisson Manifolds . . . . . . . . . . . . . . . . 127
5.3 PoissonDirac Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.3.1 Algebraic PoissonDirac Reduction . . . . . . . . . . . . . . . . . . . . . 134
5.3.2 PoissonDirac Reduction for Poisson Manifolds . . . . . . . . . . 137
5.3.3 The Transverse Poisson Structure . . . . . . . . . . . . . . . . . . . . . . . 143
5.4 Poisson Structures and Group Actions . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.4.1 Poisson Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.4.2 Poisson Actions and Quotient Spaces . . . . . . . . . . . . . . . . . . . 149
5.4.3 Fixed Point Sets as PoissonDirac Submanifolds . . . . . . . . . . 150
5.4.4 Reduction with Respect to a Momentum Map . . . . . . . . . . . . 153
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Part II Examples
6
Constant Poisson Structures, Regular and Symplectic Manifolds . . . . 161
6.1 Constant Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.2 Regular Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.3 Symplectic Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.3.1 Symplectic Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.3.2 Symplectic Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.3.3 Symplectic Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.3.4 Quotients by Finite Groups of Symplectomorphisms . . . . . . . 174
6.3.5 Example 1: Cotangent Bundles . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.3.6 Example 2: Kahler Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
Linear Poisson Structures and Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . 179
7.1 The LiePoisson Structure on g . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
7.2 The LiePoisson Structure on g . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7.3 Properties of the LiePoisson Structure . . . . . . . . . . . . . . . . . . . . . . . . 185
7.3.1 The Symplectic Foliation: Coadjoint Orbits . . . . . . . . . . . . . . 185
7.3.2 The Cohomology of LiePoisson Structures . . . . . . . . . . . . . . 190
7.3.3 The Modular Class of a LiePoisson Structure . . . . . . . . . . . . 192
7.4 Affine Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
7.5 The Linearization of Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . 196
7.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
xii
Contents
Higher Degree Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8.1 Polynomial and (Weight) Homogeneous Poisson Structures . . . . . . . 206
8.1.1 Polynomial Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . 206
8.1.2 Homogeneous Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . 207
8.1.3 Weight Homogeneous Poisson Structures . . . . . . . . . . . . . . . . 211
8.2 Quadratic Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8.2.1 Multiplicative Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . 215
8.2.2 The Modular Vector Field of a Quadratic Poisson Structure . 218
8.3 NambuPoisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
8.4 Transverse Poisson Structures to Adjoint Orbits . . . . . . . . . . . . . . . . . 227
8.4.1 Striples in SemiSimple Lie Algebras . . . . . . . . . . . . . . . . . . . 227
8.4.2 Weights Associated to an STriple . . . . . . . . . . . . . . . . . . . . . . 228
8.4.3 Transverse Poisson Structures to Nilpotent Orbits . . . . . . . . . 229
8.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
Poisson Structures in Dimensions Two and Three . . . . . . . . . . . . . . . . . . 233
9.1 Poisson Structures in Dimension Two . . . . . . . . . . . . . . . . . . . . . . . . . . 234
9.1.1 Global Point of View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
9.1.2 Local Point of View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
9.1.3 Classification in Dimension Two . . . . . . . . . . . . . . . . . . . . . . . 237
9.2 Poisson Structures in Dimension Three . . . . . . . . . . . . . . . . . . . . . . . . 250
9.2.1 Poisson Structures on F3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
9.2.2 Poisson Manifolds of Dimension Three . . . . . . . . . . . . . . . . . . 254
9.2.3 Quadratic Poisson Structures on F3 . . . . . . . . . . . . . . . . . . . . . 258
9.2.4 Poisson Surfaces in C3 and Du Val Singularities . . . . . . . . . . 263
9.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
10 RBrackets and rBrackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
10.1 Linear RBrackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.1.1 RMatrices and the YangBaxter Equation . . . . . . . . . . . . . . . 270
10.1.2 RMatrices and Lie Algebra Splittings . . . . . . . . . . . . . . . . . . . 272
10.1.3 Linear RBrackets on g and on g . . . . . . . . . . . . . . . . . . . . . . . 273
10.2 Linear rBrackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
10.2.1 Coboundary Lie Bialgebras and rMatrices . . . . . . . . . . . . . . . 275
10.2.2 Linear rBrackets on g . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
10.2.3 From rMatrices to RMatrices . . . . . . . . . . . . . . . . . . . . . . . . . 280
10.3 RBrackets and rBrackets of Higher Degree . . . . . . . . . . . . . . . . . . . . 283
10.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
11 PoissonLie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
11.1 Multiplicative Poisson Structures and PoissonLie Groups . . . . . . . . 292
11.1.1 The Condition of Multiplicativity . . . . . . . . . . . . . . . . . . . . . . . 292
11.1.2 Basic Properties of PoissonLie Groups . . . . . . . . . . . . . . . . . 294
11.1.3 PoissonLie Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
11.1.4 Linear Multiplicative Poisson Structures . . . . . . . . . . . . . . . . . 297
Contents
11.2
11.3
11.4
11.5
xiii
11.1.5 Coboundary PoissonLie Groups . . . . . . . . . . . . . . . . . . . . . . . 298
11.1.6 Multiplicative Poisson Structures on Vector Spaces . . . . . . . . 299
Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
11.2.1 Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
11.2.2 Lie Subbialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
11.2.3 Duality for Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
11.2.4 Coboundary Lie Bialgebras and rMatrices . . . . . . . . . . . . . . . 308
11.2.5 Manin Triples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
11.2.6 Lie Bialgebras and Poisson Structures . . . . . . . . . . . . . . . . . . . 313
PoissonLie Groups and Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . 314
11.3.1 The Lie Bialgebra of a PoissonLie Group . . . . . . . . . . . . . . . 314
11.3.2 Coboundary PoissonLie Groups and Lie Bialgebras . . . . . . 316
11.3.3 The Integration of Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . 318
Dressing Actions and Symplectic Leaves . . . . . . . . . . . . . . . . . . . . . . . 321
Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Part III Applications
12 Liouville Integrable Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
12.1 Functions in Involution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
12.2 Constructions of Functions in Involution . . . . . . . . . . . . . . . . . . . . . . . 332
12.2.1 Poissons Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
12.2.2 Noethers Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
12.2.3 BiHamiltonian Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 333
12.2.4 Poisson Maps and Thimms Method . . . . . . . . . . . . . . . . . . . . 334
12.2.5 Lax Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
12.2.6 The AdlerKostantSymes Theorem . . . . . . . . . . . . . . . . . . . . 337
12.3 The Liouville Theorem and the ActionAngle Theorem . . . . . . . . . . . 341
12.3.1 Liouville Integrable Systems and Liouvilles Theorem . . . . . 341
12.3.2 Foliation by Standard Liouville Tori . . . . . . . . . . . . . . . . . . . . 343
12.3.3 Period Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
12.3.4 The Existence of ActionAngle Coordinates . . . . . . . . . . . . . . 346
12.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
13 Deformation Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
13.1 Deformations of Associative Products . . . . . . . . . . . . . . . . . . . . . . . . . 354
13.1.1 Formal and kth Order Deformations . . . . . . . . . . . . . . . . . . . . 354
13.1.2 Hochschild Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
13.1.3 Deformations and Cohomology . . . . . . . . . . . . . . . . . . . . . . . . 359
13.1.4 The MaurerCartan Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 361
13.1.5 Star Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
13.1.6 A Star Product for Constant Poisson Structures . . . . . . . . . . . 363
13.1.7 A Star Product for Symplectic Manifolds . . . . . . . . . . . . . . . . 365
13.2 Deformations of Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
13.2.1 Formal and kth Order Deformations . . . . . . . . . . . . . . . . . . . . 367
xiv
Contents
13.3
13.4
13.5
13.6
13.2.2 Deformations and Cohomology . . . . . . . . . . . . . . . . . . . . . . . . 368
13.2.3 The MaurerCartan Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 371
Differential Graded Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
13.3.1 The Symmetric Algebra of a Graded Vector Space . . . . . . . . 371
13.3.2 Differential Graded Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . 372
13.3.3 The MaurerCartan Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 376
13.3.4 Gauge Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
13.3.5 Path Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
13.3.6 Applications to Deformation Theory . . . . . . . . . . . . . . . . . . . . 385
L Morphisms of Differential Graded Lie Algebras . . . . . . . . . . . . . . 386
13.4.1 is WellDefined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
13.4.2 Surjectivity of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
13.4.3 Injectivity of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Kontsevichs Formality Theorem and Its Consequences . . . . . . . . . . . 397
13.5.1 Kontsevichs Formality Theorem . . . . . . . . . . . . . . . . . . . . . . . 398
13.5.2 Kontsevichs Formality Theorem for Rd . . . . . . . . . . . . . . . . . 399
13.5.3 A Few Consequences of Kontsevichs Formality Theorem . . 406
Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
Multilinear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
A.1 Tensor Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
A.2 Exterior and Symmetric Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
A.3 Algebras and Graded Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
A.4 Coalgebras and Graded Coalgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
A.5 Graded Derivations and Coderivations . . . . . . . . . . . . . . . . . . . . . . . . . 425
Real and Complex Differential Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 427
B.1 Real and Complex Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
B.2 The Tangent Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
B.3 Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
B.4 The Flow of a Vector Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
B.5 The Frobenius Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
List of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
Introduction
Poisson structures appear in a large variety of different contexts, ranging from string
theory, classical/quantum mechanics and differential geometry to abstract algebra,
algebraic geometry and representation theory. In each one of these contexts, it turns
out that the Poisson structure is not a theoretical artifact, but a key element which,
unsolicited, comes along with the problem which is investigated and its delicate
properties are in basically all cases decisive for the solution to the problem.
Hamiltonian mechanics and integrable systems. A first striking example of this
phenomenon appears in classical mechanics. Poissons classical bracket, which is
given for smooth functions F and G on R2r by
r
F G G F
{F, G} :=
,
(0.1)
qi pi
i=1 qi pi
allows one to write Hamiltons equations of motion
qi =
H
,
pi
pi =
H
,
qi
i = 1, . . . , r ,
(0.2)
associated to a Hamiltonian H = H(q, p), in a coordinatefree manner, treating positions qi and momenta pi on an equal footing. Indeed, using (0.1), the equations of
motion (0.2) can be written as
qi = {qi , H} ,
pi = {pi , H} ,
i = 1, . . . , r .
(0.3)
For every solution t (q(t), p(t)) to these differential equations, it follows from
(0.1) and (0.3) that
d
F(q(t), p(t)) = {F, H} (q(t), p(t)) ,
dt
(0.4)
for all smooth functions F on R2r .
xv
xvi
Introduction
In differential geometrical terms, the Poisson bracket defines an operator which
associates to every smooth function H on R2r a vector field XH on R2r . In terms
of local coordinates, the vector field XH is given by (0.3) and the variation of a
smooth function F on R2r along any integral curve (q(t), p(t)) of the vector field
XH is given by (0.4). In fact, when the function H is the Hamiltonian (the energy)
of the system, the integral curves of XH describe the time evolution of the system.
Thus, the equations of motion of classical mechanics come along with an associated
Poisson bracket; for simple systems, such as systems of particles with a classical
interaction, the Poisson structure is the above one, but for constrained and reduced
systems, the Poisson bracket takes a more complicated form.
The role which the Poisson structure plays in solving the problems of classical mechanics was known partly to Poisson, who pointed out that the vanishing of
{F, H} and {G, H} implies the vanishing of {{F, G} , H}. In mechanical terms, this
means in view of (0.4) that the Poisson bracket of two constants of motion is again a
constant of motion. Since constants of motion are a great help in the explicit integration of Hamiltons equations, Poissons theorem brings to light the Poisson bracket
as a tool for integration.
A strong amplification of this fact is the Liouville theorem, which states that r
independent constants of motion H1 = H, H2 , . . . , Hr of (0.3) suffice for integrating Hamiltons equations by quadratures, under the hypothesis that these functions
commute for the Poisson bracket (are in involution). A yet further amplification is
the actionangle theorem, which states that in the neighborhood of a compact submanifold, traced out by the flows of the (commuting!) vector fields XHi , there exist
(S1 )r Rr valued coordinates (1 , . . . , r , 1 , . . . , r ) in which the Poisson bracket
takes the canonical form (0.1), namely
r
F G G F
{F, G} =
,
i i
i=1 i i
and such that the functions H1 , . . . , Hr (including the Hamiltonian H) depend on the
coordinates 1 , . . . , r only. According to (0.2), Hamiltons equations take in these
coordinates the simple form
H
i =
,
i
i =
H
=0,
i
i = 1, . . . , r ,
which entails that the i are constant and hence that the i are affine functions of
time (since H does not depend on 1 , . . . , r ).
Thus, in a nutshell, the equations of motion of classical mechanics come with a
Poisson bracket, the Poisson bracket is decisive for their integrability and it permits
us to construct coordinates in which the problem (including the Poisson bracket)
takes a very simple form, from which the solutions and their characteristics can be
read off at once.
Introduction
xvii
Abstraction and generalization. Before giving another example of the key role
played by Poisson brackets, we explain the algebraic and geometric abstraction of
the classical Poisson bracket (0.1). It is well known that Poissons theorem is explained by the Jacobi identity
{{F, G} , H} + {{G, H} , F} + {{H, F} , G} = 0 ,
(0.5)
valid for arbitrary smooth functions F, G and H on R2r . It is however noteworthy
to point out that Jacobi discovered this identity only thirty years after Poisson announced his theorem. Taking arbitrary smooth functions xi j = x ji defined on a
nonempty open subset of Rd , with coordinates x1 , . . . , xd , the skewsymmetric bilinear operation { , }, defined on smooth functions by
{F, G} :=
i, j=1
xi j
F G
,
xi x j
(0.6)
satisfies the Jacobi identity if and only if the functions xi j satisfy the identity
x jk
xi j
xki
x
+
x
+
x
k x i x j x = 0 .
=1
d
This fact was observed by Lie, who also pointed out that under the assumption of
constancy of the rank (which he assumes implicitly), there exist local coordinates
q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs , where d = 2r + s, such that (0.6) takes the form
r
F G G F
{F, G} =
,
qi pi
i=1 qi pi
which is formally the same as (0.1), but on a d = 2r + s dimensional space, with s of
the variables being totally absent from the bracket. An important class of examples
of brackets (0.6), satisfying (0.5), is defined on the dual g of a (finitedimensional)
Lie algebra g, where the bracket is directly inherited by the Lie bracket. Many important facts about the Poisson bracket on g , such as the relation to the coadjoint
representation of the adjoint group G of g, were observed by Lie and were rediscovered and further expanded by Berezin [21], Kirillov [105, 106], Kostant [117] and
Souriau [185].
It was pointed out by Lichnerowicz in [126] that, in abstract differential geometrical terms, a Poisson structure on a smooth manifold M is a smooth bivector field
on M, satisfying [ , ]S = 0, where [ , ]S stands for the Schouten bracket (a natural
extension of the Lie bracket of vector fields to arbitrary multivector fields). It implies that for every open subset U of M, the bilinear operation { , }U , defined for
F, G C (U) by
{F, G}U (m) := dm F dm G, m ,
(0.7)
for all m M, satisfies the Jacobi identity, hence defines a Lie algebra structure
on C (U).
xviii
Introduction
Moreover, this Lie algebra structure on C (U) and the associative structure
on C (U) (pointwise multiplication of functions) are compatible in the sense that
{FG, H}U = F {G, H}U + G {F, H}U ,
for all F, G, H C (U).
It is from here easy to make the algebraic abstraction of the notion of a Poisson
structure, leading to the notion of a Poisson algebra. For concreteness, we give the
definition in the case of a vector space A over a field F of characteristic zero (one
may think of R or C). Two products (bilinear maps from A to itself) are assumed
to be given on A , one denoted by , which is assumed to be associative and commutative, while the other one, denoted by { , }, is assumed to be a Lie bracket. The
triple (A , , { , }) is called a Poisson algebra if the two products are compatible in
the sense that
{F G, H} = F {G, H} + G {F, H} ,
(0.8)
for all F, G, H A . It is clear that in the case of a Poisson manifold (M, ), for
each open subset U of M, the triple (C (U), , { , }U ) is a Poisson algebra, where
stands for the pointwise product of functions and { , }U stands for the Lie bracket
defined by (0.7).
Deformation theory and quantization. We now come to a second striking example of Poisson structures making an unexpected appearance, and playing next a very
definite role in the statement of the problem and in its final solution. We give a purely
algebraic, but very simple and natural, introduction to the problem. In this approach,
one speaks of deformation theory. A physical interpretation of the problem will
be given afterwards; one then speaks of quantization or of deformation quantization.
We start with a commutative associative algebra A = (A , ). The reader may
think of the example of the polynomial algebra A = F[x1 , . . . , xd ]. We suppose that
the product in A is part of a family of associative (not necessarily commutative)
products, say we have a family t of bilinear maps
t : A A A
(a, b) a t b
(0.9)
such that, for each fixed t F, the product t is associative, with 0 being the original
associative commutative product on A , i.e., (A , ) = (A , 0 ). We refer to the family
of associative products t on A as a deformation of A . It is assumed that the family
depends nicely on t, in a way which we do not make precise here. We consider, for
every t F, the commutator of t , which is defined for all a, b A by
[a, b]t := a t b b t a .
Introduction
xix
Since t is associative, the commutator [ , ]t satisfies for each t F the following
identities, valid for all a, b, c A :
[a t b, c]t = a t [b, c]t + [a, c]t t b ,
(0.10)
[[a, b]t , c]t + [[b, c]t , a]t + [[c, a]t , b]t = 0 .
Consider the skewsymmetric bilinear map { , } : A A A , which is defined
for a, b A by
{a, b} :=
d
d
[a, b]t =
(a t b b t a) .
dt t=0
dt t=0
Then the first equation in (0.10) implies that { , } satisfies the Leibniz property (0.8), while the second equation in (0.10) implies that { , } satisfies the Jacobi
identity. Combined, it means that (A , , { , }) is a Poisson algebra. Thus, a family of deformations of a given commutative associative algebra leads naturally to a
Poisson bracket on this algebra! In the particular case of the associative algebras of
operators of quantum mechanics, in which t is a multiple of Plancks constant h , the
operators become in the limit h 0 commuting classical variables and the above
procedure yields a Poisson bracket on the algebra of classical variables.
The main problem of deformation theory is to invert this procedure; in the context
of physics, this is called the quantization problem or, more precisely, the problem
of quantization by deformation. In its algebraic version, the first question is whether
given a Poisson algebra (A , , { , }) there exists a family of associative products t
on A , such that (A , ) = (A , 0 ) and such that { , }, obtained from t by the above
procedure, is a Poisson bracket on A . The second question is to classify all such deformations, for a given (A , , { , }). The physical interpretation of these questions
is that, starting from the algebra of classical variables, which inherits from phase
space a Poisson bracket, as explained above in the context of mechanical systems,
one wishes to (re)construct the corresponding algebra of quantum mechanical operators.
A mathematical simplification of this problem is to ask whether the above procedure can be inverted formally. Stated in a precise way, this means the following.
Let (A , , { , }) be a Poisson algebra and consider the vector space A [[ ]] of formal power series in one variable , whose coefficients belong to A . Let denote
an associative product on A [[ ]], i.e., an F[[ ]]bilinear map from A [[ ]] to itself,
which is associative. Given a, b A , we can write
(a, b) =
k (a, b) k ,
kN
which defines bilinear maps i : A A A for i N. By definition, is a formal
deformation of (A , , { , }) if 0 (a, b) = a b and 1 (a, b) 1 (b, a) = {a, b}, for
all a, b A . The main question, stated formally, is then if every Poisson algebra
(A , , { , }) admits a formal deformation .
xx
Introduction
Consider for example R2r , with coordinates (x1 , . . . , x2r ) = (q1 , . . . , qr , p1 , . . . , pr ).
The Moyal product is the R[[ ]]bilinear product on C (R2r )[[ ]], which is given for
F, G C (R2r ) by
(F, G) :=
kN 1i1 , j1 ,...,ik , jk 2r
Ji1 , j1 . . . Jik , jk
kF
kG
k
,
xi1 . . . xir x j1 . . . x jr k!
0 1r
. The
1r 0
Moyal product, also called the MoyalWeyl product, defines an associative product
on C (R2r )[[ ]], whose leading terms satisfy 0 (F, G) = FG and
r
F G G F
,
1 (F, G) 1 (G, F) =
qj pj qj pj
j=1
where J is the skewsymmetric matrix of size 2r, given by J :=
where the latter righthand side is the Poisson bracket (0.1). This shows that the
classical Poisson bracket admits a formal deformation.
It was shown by De Wilde and Lecomte in [57] that the Poisson algebra of an
arbitrary symplectic manifold (regular Poisson manifolds of maximal rank) admits
a formal deformation. A geometrical proof of their result, which also works in the
case of arbitrary regular Poisson manifolds, was given by Fedosov [73]. Finally, it
was Kontsevich [107] who proved, using ideas which come from string theory, that
the algebra of functions on any Poisson manifold admits a formal deformation. He
proved in fact a quite stronger result, which says that for a given Poisson manifold
(M, ), the equivalence classes of formal deformations of the algebra of smooth
functions on M are classified by the equivalence classes of formal deformations of
the Poisson structure .
Summarizing, when one deforms a commutative associative algebra, a Poisson
structure shows up and it plays a dominant role in the entire deformation process.
The examples. The main idea which led us to the writing of this book, and to its actual structure, is that Poisson structures come in big classes (families) where roughly
speaking each class has its own tools and is related to a very definite part of mathematics, hence leading to specific questions. As a consequence, the main part of the
book (about half of it) is Part II, dedicated to six classes of examples. They make
their appearance in six different chapters, namely Chapters 611, as follows.
Constant Poisson structures, such as Poissons original bracket (0.1), are according to the Darboux theorem the model (normal form) for Poisson structures on real
or complex manifolds, in the neighborhood of any point where the rank is locally
constant. Regular Poisson manifolds and symplectic manifolds (such as cotangent
bundles and Kahler manifolds) are the main examples, which exhibit even in the
absence of singularities some of the main phenomena in Poisson geometry, which
distinguish it from Riemannian geometry and show its pertinence for classical and
quantum mechanics. Constant and regular Poisson structures are studied in Chapter 6.
Introduction
xxi
Linear Poisson structures are in onetoone correspondence with Lie algebras and
many features of Lie algebras are most naturally approached in terms of the canonical Poisson bracket on the dual of a (finitedimensional) Lie algebra, the socalled
LiePoisson structure. The coadjoint orbits, for example, of a Lie algebra are the
symplectic leaves of the LiePoisson structure, showing on the one hand that they
are even dimensional, and on the other hand that they carry a canonical symplectic
structure, the KostantKirillovSouriau symplectic structure. Linear Poisson structures appear also on an arbitrary Poisson manifold (M, ) at any point x where the
Poisson structure vanishes: the tangent space Tx M inherits a linear Poisson structure, the linearization of at x. It leads to the linearization problem, which inquires
if and its linearization at x are isomorphic, at least on a neighborhood of x. Linear
Poisson structures and their relation to Lie algebras are the subject of Chapter 7.
Many Poisson structures of interest are neither constant nor linear (or affine). In
fact, while the basic properties of the latter Poisson structures can be traced back to
(known) properties of bilinear forms and of Lie algebras, new phenomena appear
when considering quadratic Poisson structures (i.e., homogeneous Poisson structures of degree two) and higher degree Poisson structures, which often turn out to
be weight homogeneous. A prime example of this is the transverse Poisson structure to an adjoint orbit in a semisimple Lie algebra. Moreover, this Poisson structure
arises in the case of the subregular orbit from a NambuPoisson structure, defined
by the invariant functions of the Lie algebra. Higher degree Poisson structures are
studied in Chapter 8.
In dimension two, every bivector field is a Poisson structure, yet many questions
about Poisson structures on surfaces are nontrivial. For example, their local classification has up to now only been accomplished under quite strong regularity assumptions on the singular locus of the Poisson structure. In dimension three, the
Jacobi identity can be stated as an integrability condition of a distribution (or of
a oneform), which eventually leads to the symplectic foliation. A surface in C3
which is defined by the zero locus of a polynomial inherits a Poisson structure
from the standard NambuPoisson structure on C3 , with as Casimir. In the case
of singular surfaces C2 /G, where G is a finite subgroup of SL2 (C), this Poisson
structure coincides with the Poisson structure obtained from the canonical symplectic structure on C2 by reduction. Thus, rather than being trivial, Poisson structures in
dimensions two and three form a rich playground on which a variety of phenomena
can be observed. Poisson structures in dimensions two and three are discussed in
Chapter 9.
In a Lietheoretical context, a linear Poisson structure, different from the canonical LiePoisson structure, often pops up. This Poisson structure appears on the Lie
algebra g at hand, so that the dual g of g comes equipped with a Lie algebra structure, or it appears on g , so that g is equipped with a second Lie algebra structure.
The underlying operator, which relates either of these Lie structures with the original Lie bracket on g is in the first case an element r g g, called an rmatrix,
while it is in the second case a (vector space) endomorphism R of g, called an Rmatrix. These structures appeared first in the theory of integrable systems, but are
xxii
Introduction
nowadays equally important in the theory of LiePoisson groups (see the next item)
and of quantum groups. In the context of Lie algebras which come from associative
algebras, rmatrices and Rmatrices also lead to a quadratic Poisson structure on
the Lie algebra, which plays an important role in the theory of LiePoisson groups;
they also lead to a cubic Poisson structure, whose virtue seems at this point still very
mysterious. Poisson structures coming from rmatrices or Rmatrices are the subject
of Chapter 10.
A PoissonLie group is a Lie group G, equipped with a Poisson structure for
which the group multiplication G G G is a Poisson map. The Poisson structure leads to a Lie algebra structure on the dual of the Lie algebra g of G, making g
into a Lie bialgebra, a structure generalizing the structure which comes from an rmatrix. Conversely, every finitedimensional Lie bialgebra is obtained in this way
from a PoissonLie group, a result which extends Lies third theorem (every finitedimensional Lie algebra is the Lie algebra of a Lie group). PoissonLie groups have
many applications, for example in the description of the Schubert cells of a Lie
group. PoissonLie groups are discussed in Chapter 11.
Theoretical foundations. The examples (Chapters 611) are preceded by a first
part (5 chapters) with a systematic exposition of the general theory of Poisson structures. We used two principles in the writing of these chapters. The first one is that we
wanted to develop both the algebraic and geometric points of view of the theory. In
fact, there has for the last decade been an increasing interest in the algebraic aspects
of Poisson structures, while of course geometrical intuition underlies all constructions and the vast majority of examples. In the algebraic context, we have a Poisson
algebra, whose underlying space is an Fvector space, where F is an arbitrary field
of characteristic zero. In the geometric context, we have a Poisson manifold, where
the manifold is either real or complex. In between the algebraic and geometric contexts, Poisson varieties appear, whose underlying object can both be viewed as a
variety or as a finitely generated algebra. Throughout this first part, the reader will
experience that although most of the results are formally very similar in the algebraic and in the geometric setting, their concrete implementation (including proofs
of the propositions) is quite different and needs to be worked out in detail in both
settings.
The second principle which we used was to keep the prerequisites in (commutative and Lie) algebra and in (differential and algebraic) geometry to a minimum.
Thus, for example, when we describe Poisson structures geometrically as bivector
fields, we explain the notion of a bivector field by first recalling the notion of a vector field, rather than simply saying that a bivector field is a section of the exterior
square of the tangent bundle to the manifold. Similarly, while the basic facts about
Lie groups and Lie algebras are supposed to be known, the interplay between Lie
groups and Lie algebras is recalled in detail, as it is further amplified in the presence
of Poisson structures. We added an appendix at the end of the book with some facts
on multilinear algebra (tensor products, wedges, algebra and coalgebra structures)
and an appendix which recalls the basic facts about differential geometry. The ob
Introduction
xxiii
jects and properties which are recalled in these appendices are used throughout the
book.
Chapters 1 and 2 contain the basic definitions and constructions. A good familiarity with these chapters should suffice for reading a good part of any other chapter
of the book. Chapter 3 deals with multiderivations and their geometrical analog,
multivector fields, and with Kahler forms, which are the algebraic analogs of differential forms. Chapter 4 deals with Poisson cohomology and Chapter 5 is devoted
to reduction in the context of Poisson structures. At the end of each of these five
chapters, we give a list of exercises, so that the reader can test his understanding of
the theory.
Applications. The third part of the book contains the two major applications, which
we discussed above: integrable systems are discussed in Chapter 12, while deformation quantization is detailed in Chapter 13. They are however not the only
applications which are given in the book. In fact, we end each example chapter
(Chapters 611) with an application of the class of Poisson structures, studied in
that chapter.
What is absent from this book. The main topic about Poisson structures which is
absent from this book is what should be called Poisson geometry. By this we mean
the global geometry of Poisson structures, which involves the integration of Poisson
brackets, Poisson connections, stability of symplectic leaves and related topics. For
this, we refer to [51, 53, 74].
Equally absent from this book are the many generalizations of Poisson structures.
QuasiPoisson structures and Poisson structures with background are not only mathematically speaking very interesting, see [112], they have in addition nontrivial applications in physics, see [119, 162, 176], and are themselves a particular case of
Dirac structures [48, 89]. The theory of quantum groups, initiated in the seminal
ICM talk [59] by Drinfeld, has several connections with the topics which are developed in this book, but adding several chapters to the book would not have been
sufficient to give a fair account of this subject; we simply refer to the excellent books
[40, 103].
Equally absent from this book is the theory of Lie algebroids, introduced by
Pradines [171], which are both a particular case and a generalization of Poisson
manifolds. In particular, after the pioneering work of [47], the problem of the integration of Poisson manifolds, which claims that symplectic Lie groupoids are to
Poisson manifolds what Lie groups are to Lie algebras, has gained a long attention.
For a definite solution of this problem, given by Cranic and Fernandes, see [49, 50].
The notes and the references. When writing the history of the theory of Poisson
brackets, giving each of the main results in the theory and examples the right
credit is a challenge which is beyond the scope of this book. We claim no originality
in this book, but since none of the proofs which are given are copyandpaste proofs
from existing proofs, we did not think it was necessary to cite the proof in the literature which is closest to the proof that we give. Also, many properties of Poisson
xxiv
Introduction
structures, and many examples, have been through a long series of transformations
and generalizations, think for example of Weinsteins splitting theorem or the Poisson and PoissonDirac reduction theorems; structuring the long list of important
intermediate results which led to the final results is certainly interesting from the
epistemological point of view, but this was not our main focus when writing the
book.
However, at the end of each chapter, we have put a section called Notes, in
which we indicate some references, including the main references which we know
of, for further reading on what has been treated in the chapter. These notes are also
used to situate, a posteriori, the treated subjects in the scientific literature and to trace
some of their historical developments. Finally, the notes are also used to indicate
some further connections to other fields of mathematics or physics. Since Poisson
structures are a very active field of research, it is clear that the list of connections
with other fields is not exhaustive.
Part I
Theoretical Background
Chapter 1
Poisson Structures: Basic Definitions
In this chapter, we give the basic definitions of a Poisson algebra, of a Poisson
variety, of a Poisson manifold and of a Poisson morphism.
Geometrically speaking, a Poisson structure on a smooth manifold M associates
to every smooth function H on M, a vector field XH on M. In the context of a mechanical system, this vector field yields the equations of motion, when H is taken
as the Hamiltonian. The Poisson bracket is demanded to be a Lie bracket, which
amounts to demanding that Poissons theorem is valid, namely that the Poisson
bracket of two constants of motion is itself a constant of motion.
Algebraically speaking, one considers on a (typically infinitedimensional) vector space A two different algebra structures: (1) a commutative, associative multiplication, (2) a Lie bracket. It results in the following definition:
Poisson algebra := Comm. assoc. algebra + Lie algebra + Compatibility.
The compatibility between the two algebra structures is precisely what allows us
to associate to elements of A , derivations of A , derivations being the algebraic
analogs of vector fields.
Poisson algebras and their morphisms are defined in Section 1.1. In Section 1.2,
respectively Section 1.3, we introduce Poisson varieties, respectively Poisson manifolds, and their morphisms. Sections 1.2 and 1.3 can be read independently and
in either order; the reader is invited to discover, by comparing them, up to which
point Poisson varieties and Poisson manifolds can be treated uniformly, and how
quickly the techniques and results diverge, past this point. In Section 1.4 we specialize the results of Sections 1.2 and 1.3 to the case of Poisson structures on a
finitedimensional vector space; such a space can indeed be viewed as an affine
variety or as a manifold, depending on the choice of algebra of functions on it.
Throughout the present chapter, we fix a ground field F of characteristic zero,
which the reader may think of as being R or C, especially in the context of varieties.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 1, SpringerVerlag Berlin Heidelberg 2013
1 Poisson Structures: Basic Definitions
1.1 Poisson Algebras and Their Morphisms
In this section we introduce the general notions of a Poisson algebra and of a morphism of Poisson algebras. The Hamiltonian operator, which associates to an element of the Poisson algebra a derivation of the Poisson algebra, will be introduced
and its basic algebraic properties will be given.
1.1.1 Poisson Algebras
We start with the algebraic notion of a Poisson algebra over a field F which, as we
have already said, is always assumed to be of characteristic zero.
Definition 1.1. A Poisson algebra is an Fvector space A equipped with two multiplications (F, G) F G and (F, G) {F, G}, such that
(1)
(2)
(3)
(A , ) is a commutative associative algebra over F, with unit 1;
(A , { , }) is a Lie algebra over F;
The two multiplications are compatible in the sense that
{F G, H} = F {G, H} + G {F, H} ,
(1.1)
where F, G and H are arbitrary elements of A .
The Lie bracket { , } is then called a Poisson bracket.
Poisson brackets are often constructed on a given commutative associative algebra
A = (A , ). The algebra of regular functions on an (affine algebraic) variety is a
typical example of this, leading to the notion of Poisson variety (Section 1.2); see
Section 1.3 for the case of real or complex Poisson manifolds. On A , one considers
then a skewsymmetric bilinear map { , } : A A A , which satisfies condition (1.1). It will be a Poisson bracket on A as soon as the Jacobi identity
{F, {G, H}} + {G, {H, F}} + {H, {F, G}} = 0
(1.2)
is verified for all triples (F, G, H) of elements of A . In order to shorten the formulas, (1.2) is often written as
{F, {G, H}} + (F, G, H) = 0 ,
and similarly for other formulas which involve sums over three terms with cyclically permuted variables. The algebraic terminology which expresses (1.1) is that
for every H A the linear map F {F, H} is a derivation of A . Namely, a linear
map V : A A is called a derivation of A (with values in A ) if
V (FG) = FV (G) + G V (F) ,
(1.3)
1.1 Poisson Algebras and Their Morphisms
for every F, G A . We have written FG for F G, a convenient shorthand which
we will use freely in the sequel. We denote by X1 (A ) the Lie algebra of derivations
of A , where the Lie bracket [ , ] on X1 (A ) is given by the commutator of derivations. A bilinear map P : A A A is called a biderivation of A (with values
in A ) if
P(FG, H) = F P(G, H) + G P(F, H) ,
(1.4)
P(H, FG) = F P(H, G) + G P(H, F) ,
for every F, G, H A . Our biderivations will always be defined by a skewsymmetric bilinear map P : A A A ; we say then that P is a skewsymmetric
biderivation of A . For such a biderivation, it is convenient to enclose its arguments
in square brackets; thus, when P is a skewsymmetric biderivation, we write the first
equation in (1.4) as
P[FG, H] = F P[G, H] + G P[F, H] ;
by the skewsymmetry of P, this formula is of course equivalent with the second
equation in (1.4). Square brackets will also be used to enclose the arguments of
a general skewsymmetric multiderivation, a notion which will be introduced in
Section 3.1; in particular, we will also use square brackets to enclose the argument
of a derivation, writing V [F] for V (F) when V is a derivation of A .
The Fvector space of skewsymmetric biderivations of A is denoted by X2 (A ).
As recalled in Appendix A, a skewsymmetric bilinear map on A can be viewed as
a linear map on 2 A = A A ; we therefore often abbreviate : A A A to
: 2 A A or : A A A . According to Definition 1.1, a Poisson bracket
on A is in particular a skewsymmetric biderivation of A .
1.1.2 Morphisms of Poisson Algebras
We now turn to the notion of a morphism between two Poisson algebras (defined
over the same field F). Since a Poisson algebra is an algebra in two different ways,
it is natural to demand that a morphism of Poisson algebras be a morphism with
respect to both algebra structures.
Definition 1.2. Let (Ai , i , { , }i ), i = 1, 2, be two Poisson algebras over F. A linear
map : A1 A2 which satisfies, for all F, G A1 ,
(1)
(2)
(F 1 G) = (F) 2 (G);
({F, G}1 ) = { (F), (G)}2 ,
is called a morphism of Poisson algebras.
It is clear that if : A1 A2 is a morphism of Poisson algebras and is bijective,
then 1 : A2 A1 is also a morphism of Poisson algebras. We say then that is
an isomorphism of Poisson algebras.
1 Poisson Structures: Basic Definitions
As we will see in the case of Poisson manifolds and of Poisson varieties, one is
often led to morphisms of the associative algebras which underlie the two Poisson
algebras. In this case, we only have to consider condition (2) in the above definition,
which says that is a morphism of Lie algebras.
When one deals with subalgebras or ideals of Poisson algebras, it is important
to distinguish which multiplication (maybe both) is considered. Our convention is
that subalgebra and ideal refer to the associative multiplication, Lie subalgebra and
Lie ideal refer to the Lie bracket, and Poisson subalgebra and Poisson ideal refer to
both. This means that a vector subspace B A is a Poisson subalgebra of A =
(A , , { , }) if
{B, B} B ,
BB B
and
(1.5)
while a vector subspace I A is a Poisson ideal of A if
I A I
and
{I , A } I .
(1.6)
In the first case, B becomes itself a Poisson algebra, when equipped with the restrictions of both multiplications to B and the inclusion map : B A is a morphism of
Poisson algebras. In the second case, A /I inherits a Poisson bracket from A , such
that the canonical projection p : A A /I is a morphism of Poisson algebras. We
will come back at length to the notions of Poisson subalgebra and Poisson ideal in
Chapter 2, when we have developed sufficient concepts and tools to illustrate these
notions geometrically.
In view of Definitions 1.1 and 1.2 we get, for a fixed field F, a category whose
objects are the Poisson algebras over F and whose morphisms are the morphisms of
Poisson algebras.
1.1.3 Hamiltonian Derivations
The biderivation property of the Poisson bracket leads to a fundamental operation
which allows one to associate to elements of A , derivations of A ; in the case of a
real Poisson manifold M, it allows us to associate to a function on M, a vector field
on M (see Section 1.3). This operation, which we introduce now, corresponds in the
Hamiltonian formulation of classical mechanics to writing the equations of motion
for a given Hamiltonian (see Section 6.1).
Definition 1.3. Let (A , , { , }) be a Poisson algebra and let H A . The derivation
XH := { , H} of A is called a Hamiltonian derivation and we call H a Hamiltonian,
associated to XH . We write
Ham (A , { , }) := {XH  H A }
for the Fvector space of Hamiltonian derivations of A , so that we have an Flinear
map
1.2 Poisson Varieties
X : A Ham(A , { , })
H XH := { , H} .
(1.7)
An element H A whose Hamiltonian derivation is zero, XH = 0, is called a
Casimir and we denote
Cas (A , { , }) := {H A  XH = 0}
for the Fvector space of Casimirs. When no confusion can arise, we write Ham(A )
for Ham(A , { , }) and Cas(A ) for Cas(A , { , }). It is clear that Cas(A ) is the
center of the Lie algebra (A , { , }).
The defining properties of the Poisson bracket lead to the following proposition.
Proposition 1.4. Let (A , , { , }) be a Poisson algebra.
(1)
(2)
(3)
(4)
(5)
(6)
Cas(A ) is a subalgebra of (A , ), which contains the image of F in A , under
the natural inclusion a a 1;
If A has no zero divisors, then Cas(A ) is integrally closed in A ;
Ham(A ) is not an A module (in general); instead, XFG = FXG + GXF , for
every F, G A ;
Ham(A ) is a Cas(A )module;
The map A X1 (A ) which is defined by H XH is a morphism of Lie
algebras; as a consequence, Ham(A ) is a Lie subalgebra of X1 (A );
The Lie algebra sequence
X
0 Cas(A ) A Ham(A ) 0
is a short exact sequence.
Proof. Properties (1)(4) follow from the biderivation property (1.1), while properties (5) and (6) follow from the Jacobi identity (1.2) for { , }. We only show (2).
Let F be integral over Cas(A ), that is F A and there exists a monic polynomial
p(X) Cas(A )[X], such that p(F) = 0. We need to show that F Cas(A ). Let p
be the monic polynomial of smallest degree such that p(F) = 0. If deg(p) = 1, then
F Cas(A ) and we are done. Let us suppose therefore that r := deg(p) > 1. For
every element G A we have by the derivation property that 0 = {p(F), G} =
p (F) {F, G}, where p denotes the derivative of p. Now p (F) = 0, as p /r would
otherwise be a monic polynomial of degree r 1, such that p (F)/r = 0. Since A is
without zero divisors, {F, G} = 0. As G is arbitrary, this shows that F is a Casimir
of A .
1.2 Poisson Varieties
When A is an algebra of functions on some variety or manifold, the properties of
a Poisson bracket on A get a geometrical meaning, leading to many interesting
1 Poisson Structures: Basic Definitions
constructions, which can often be generalized to arbitrary Poisson algebras. In this
section, we consider affine varieties, see the next section for the case of (real or
complex) manifolds.1 The reader, unfamiliar with affine varieties, may on his first
reading of this section think of the affine variety M as being the affine space Fd ,
where d N , equipped with the algebra of polynomial functions F[x1 , . . . , xd ].
1.2.1 Affine Poisson Varieties and Their Morphisms
We use the convention that an affine variety is an irreducible algebraic subset M
of an affine space Fd , where algebraic subset means that M is the zero locus of
a family of polynomials (in d variables). It conforms to the American convention,
see e.g. [93, Ch. I.1]; in the French literature, affine varieties are usually not assumed to be irreducible. Given an algebraic subset M Fd , one considers the prime
ideal I of F[x1 , . . . , xd ] which consists of all polynomial functions, vanishing on M.
Then F[x1 , . . . , xd ]/I becomes a finitely generated (commutative associative) algebra, which can be considered as an algebra of functions on M, since the evaluation of elements of F[x1 , . . . , xd ]/I at points of M is welldefined. This algebra of
functions on M is denoted by F (M) and is called the affine coordinate ring of M.
For F F[x1 , . . . , xd ] we denote its projection in F (M) = F[x1 , . . . , xd ]/I by F or
by FM , the latter notation being justified by the fact that F can be viewed as the
restriction of F to M. Since M is irreducible, F (M) has no zero divisors. Elements
of F (M) are called regular functions on M, although the name polynomial function
is also used, especially when M = Fd .
Thus, the vector space of regular functions on an affine variety is naturally
equipped with a first algebra structure: the structure of a commutative associative
algebra. The extra datum of a Poisson bracket on this algebra leads to the notion of
an affine Poisson variety.
Definition 1.5. Let M be an affine variety and suppose that F (M) is equipped with
a Lie bracket { , } : F (M) F (M) F (M), which makes (F (M), , { , }) into a
Poisson algebra. Then we say that (M, { , }) is an affine Poisson variety, or simply
a Poisson variety. The Poisson bracket on F (M) is usually referred to as a Poisson
structure on M.
In the case of a Poisson structure on Fd , the Poisson bracket of two functions can be
computed from the following standard formula; see Section 1.2.2 for the case of a
general affine Poisson variety.
Proposition 1.6. Let { , } be a Poisson bracket on A = F[x1 , . . . , xd ]. For all functions F and G in A , their Poisson bracket is given by
{F, G} =
i, j=1
1
xi , x j
F G
.
xi x j
(1.8)
As we said in the introduction to this chapter, Sections 1.2 and 1.3 can be read independently.
1.2 Poisson Varieties
Proof. The proof is based on the standard argument that two biderivations (or pderivations, in general) of some commutative associative algebra A are equal as
soon as they agree on a system of generators for A . Since we will use this argument
several times, we spell it out in detail. Both sides of (1.8) are bilinear in F and G,
so it suffices to show (1.8) when F and G are monomials in x1 , . . . , xd . If F or G is a
monomial of total degree 0, then the righthand side in (1.8) is obviously zero, but
also the lefthand side is zero, because constant functions are Casimirs (item (1) in
Proposition 1.4). Also, the fact that (1.8) holds when F and G are both monomials of
degree 1, is clear. Suppose now that (1.8) holds when deg(F)+deg(G) n, for some
n 2; we show that it holds for all F and G such that deg(F) + deg(G) = n + 1. Let
F and G be nonconstant monomials, such that deg(F) + deg(G) = n + 1. By skewsymmetry, we may assume that deg(F) > 1. There exist monomials F1 , F2 A ,
with deg(F1 ) < deg(F) and deg(F2 ) < deg(F), such that F = F1 F2 . Since { , } is a
biderivation and in view of the recursion hypothesis, we have that
{F, G} = {F1 F2 , G} = F1 {F2 , G} + F2 {F1 , G}
d
= F1
i, j=1
xi , x j
d
F2 G
F1 G
+ F2 xi , x j
xi x j
xi x j
i, j=1
F G
xi , x j
.
xi x j
i, j=1
d
This proves (1.8) for arbitrary polynomials F and G.
We now turn to the notion of a Poisson morphism of Poisson varieties.
Definition 1.7. Let (M1 , { , }1 ) and (M2 , { , }2 ) be two Poisson varieties. A morphism of varieties : M1 M2 is called a Poisson morphism or a Poisson map if
the dual morphism
: F (M2 ) F (M1 )
is a morphism of Poisson algebras.
Recall (see e.g. [182, Ch. I.2.3]) that for a morphism of varieties (also called a
regular map) : M1 M2 , the dual morphism is (well)defined by (F) :=
F , for all F F (M2 ), so that the first condition in Definition 1.2 is automatically
satisfied:
(FG) = (FG) = (F ) (G ) = ( F) ( G) .
In this case, the second condition, which says that
: (F (M2 ), { , }2 ) (F (M1 ), { , }1 )
is a morphism of Lie algebras, can also be written as
{F, G}2 = {F , G }1 ,
for all F, G F (M2 ) .
10
1 Poisson Structures: Basic Definitions
When N is a subvariety of M, which amounts to saying that the inclusion map :
N M is a morphism, then N is said to be a Poisson subvariety if N admits a
Poisson structure such that is a Poisson map. See Section 2.2.
Combining Definitions 1.5 and 1.7 we get, for a fixed F, a category whose objects
are (affine) Poisson varieties and whose morphisms are Poisson maps.
1.2.2 The Poisson Matrix
In this section, we show how a Poisson structure on an affine variety can be encoded
in a matrix. We first consider the case of a Poisson structure { , } on the affine
2
space Fd (with its algebra of regular functions A := F[x1 , . . . , xd ]). The
d structure
functions xi j A , which are defined for 1 i, j d by xi j := xi , x j , satisfy
xi j = x ji ,
d
x jk
xi j
xki
xk x + xi x + x j x = 0 ,
=1
(1.9)
(1.10)
for all 1 i, j, k d. Formula (1.10) is obtained by writing the Jacobi identity (1.2)
for the triple (xi , x j , xk ) in the form
xi j , xk + x jk , xi + xki , x j = 0 ,
and by using (1.8).
If we view the structure functions xi j as the elements of a (skewsymmetric) d d
matrix X, then the Poisson bracket may, according to (1.8), also be expressed in a
compact form by
{F, G} = [dF]
X [dG] ,
where [dF] is the column vector which represents dF, the differential of F, in the
natural basis (dx1 , . . . , dxd ), i.e., the ith component of [dF] is xFi . The matrix X
Matd (A ) is called the Poisson matrix of (A , { , }).
A natural question is: given a skewsymmetric matrix X = (xi j ) Matd (A ),
where A = F[x1 , . . . , xd ], does the corresponding skewsymmetric biderivation, defined for F, G A by
d
F G
{F, G} := xi j
xi x j
i, j=1
define a Poisson structure on A , i.e., does it satisfy the Jacobi identity? For F, G
and H in A , the biderivation property and the fact that second order derivatives
commute, imply that
{{F, G} , H} + (F, G, H) =
i, j,k,=1
xk
xi j F G H
+ (F, G, H) (1.11)
x xi x j xk
1.2 Poisson Varieties
11
xi j
F G H
x
+
(i,
j,
k)
,
k x
xi x j xk
i, j,k=1 =1
d
so that the Jacobi identity is satisfied for all triples of elements of A = F[x1 , . . . , xd ]
if and only if (1.10) holds, i.e., if and only if the Jacobi identity holds for every triple
(xi , x j , xk ), with 1 i < j < k d.
Summarizing, we have the following proposition.
Proposition 1.8. If X = (xi j ) is a skewsymmetric d d matrix, with elements in
A = F[x1 , . . . , xd ], then
d
F G
{F, G} := xi j
(1.12)
xi x j
i, j=1
defines a Poisson bracket on A (with Poisson matrix X) if and only if every triple
(xi , x j , xk ), with 1 i < j < k d, satisfies the Jacobi identity
xi , x j , xk + x j , xk , xi + {xk , xi } , x j = 0 ,
which amounts to the condition
d
x jk
xi j
xki
x
=0.
+
x
+
x
k
i
j
x
x
x
=1
An analogous result is valid when A is the algebra of regular functions on an affine
variety,
A := F[x1 , . . . , xd ]/I ,
where I is a prime ideal of F[x1 , . . . , xd ]. For F F[x1 , . . . , xd ], let us denote the
coset F + I by F. Then x1 , . . . , xd are generators of A and I is the set of all
relations between these generators. We suppose that we are given a skewsymmetric
d d matrix X, whose elements belong to A . We can write X = (xi j ), where the
elements xi j F[x1 , . . . , xd ], which we choose such that xi j = x ji , are of course
not unique. Let us consider the skewsymmetric biderivation { , }0 of F[x1 , . . . , xd ],
defined for F, G F[x1 , . . . , xd ] by
{F, G}0 :=
i, j=1
xi j
F G
.
xi x j
(1.13)
We
like to define a skewsymmetric biderivation { , } of A by setting
would
F, G := {F, G}0 , for all F, G F[x1 , . . . , xd ]. The problem is that, in general, this
is not welldefined. Namely, for every K I one has that K = 0, so that a necessary
condition for the bracket { , } on A to be welldefined, is that for every K I and
1 i d,
d
K
{xi , K}0 = xi j
I .
(1.14)
xj
j=1
12
1 Poisson Structures: Basic Definitions
Notice that this condition is independent of the chosen elements xi j F[x1 , . . . , xd ],
which represent xi j .
When condition (1.14) holds for every K I and for every i with 1 i d,
then {F[x1 , . . . , xd ], I }0 I , since { , }0 is a biderivation. It follows that { , }
is welldefined. Also, { , } is a skewsymmetric biderivation since { , }0 is one.
Since { , } is welldefined,
F, G , H = {{F, G}0 , H}0 ,
for all F, G, H F[x1 , . . . , xd ], so that { , } is a Poisson bracket on A if and only if
{{F, G}0 , H}0 + {{G, H}0 , F}0 + {{H, F}0 , G}0 I ,
for all F, G, H F[x1 , . . . , xd ]. Since { , }0 is a skewsymmetric biderivation, this
amounts to checking that
xi , x j 0 , xk 0 + x j , xk 0 , xi 0 + {xk , xi }0 , x j 0 I ,
for every 1 i < j < k d. Again, notice that this condition is independent of the
chosen elements xi j F[x1 , . . . , xd ], which represent xi j .
Suppose that { , } is a Poisson bracket on A = F[x1 , . . . , xd ]/I . We can
choose elements
xi j F[x1 , . . . , xd ] such that x ji = xi j for all i and j, and such
that xi j = xi , x j . Define a skewsymmetric biderivation { , }0 on F[x1 , . . . , xd ]
by (1.13). Since {xi , } is a derivation of A , we have that
0 = xi , K =
K
xi , x j
=
xj
j=1
d
xi j x j = {xi , K}0 ,
j=1
for every K I , so that condition (1.14) holds. This means that { , } , defined
for all F, G A by F, G := {F, G}0 , is a biderivation of A , which coincides
with { , } on a system of generators (x1 , . .. , xd ) of A . Thus, { , } = { , } and
we can compute the Poisson bracket F, G by using (1.13), as in the case of a
polynomial algebra.
We summarize these results in the following proposition.
Proposition 1.9. Let M be an affine variety in Fd , with algebra of regular functions
A = F[x1 , . . . , xd ]/I ,
where I is the ideal of F[x1 , . . . , xd ] of polynomial functions vanishing on M, and let
X = (xi j ) be a skewsymmetric matrix with coefficients in A . Picking representatives
xi j F[x1 , . . . , xd ] of xi j , with xi j = x ji , define a skewsymmetric biderivation { , }0
of F[x1 , . . . , xd ] by
d
F G
{F, G}0 := xi j
.
(1.15)
xi x j
i, j=1
1.2 Poisson Varieties
13
If, for every 1 j d and for every K I ,
xi xi j I ,
(1.16)
i=1
and, for every 1 i < j < k d,
xi , x j 0 , xk 0 + x j , xk 0 , xi 0 + {xk , xi }0 , x j 0 I ,
(1.17)
then a Poisson bracket { , } on A is defined by setting F, G := {F, G}0 , for all
F, G A . Conditions (1.16) and (1.17) are independent of the chosen representatives xi j of the elements of the matrix X. Also, the Poisson bracket { , } on A is
independent of this choice of representatives.
Conversely, given a Poisson structure { , } on A , the Poisson bracket F, G of
any two elements F and G of A can be computed from (1.15), namely
F, G =
i, j=1
xi , x j
F G
,
xi x j
(1.18)
where F and G are arbitrary polynomials which represent F and G.
As in the case of A = F[x1 , . . . , xd ], the matrix X = ( xi , x j ) is called the Poisson
matrix of (M, { , }), with respect to x1 , . . . , xd . Notice that, even if the Lie algebra
(F (M), { , }) is in general infinitedimensional, the Poisson bracket of arbitrary
elements of F (M) is, in the case of an affine variety M Fd , completely determined
by the Poisson matrix X, i.e., by the brackets between all pairs of elements of an
arbitrary set of generators of F (M), as is seen from the explicit formula (1.18).
1.2.3 The Rank of a Poisson Structure
In this section, we introduce the rank of a Poisson structure at a point of a Poisson
variety. Our definition is based on the fact that for a Poisson variety (M, { , }) the
rank of its Poisson matrix at a point does not depend on the chosen generators for
the algebra of regular functions on M.
Lemma 1.10. Let (M, { , })be anaffine Poisson variety and let m M. The rank
of the Poisson matrix X = ( xi , x j ) evaluated at m, is independent of the chosen
generators x1 , . . . , xd of F (M).
Proof. It suffices to show that if x1 , . . . , xd are generators of F (M) and x0 is an
arbitrary element of F (M), then the matrices
and
X (m) := ( xi , x j (m))0i, jd ,
X(m) := ( xi , x j (m))1i, jd
14
1 Poisson Structures: Basic Definitions
have the same rank. Since x0 can be written as a polynomial in the generators
x1 , . . . , xd of F (M), say x0 = F(x1 , . . . , xd ), we have in view of (1.18),
{xi , x0 } (m) =
j=1
F
xi , x j (m)
(m) ,
xj
and so the zeroth column of X (m) is a linear combination of the other columns of
X (m), i.e., of the columns of X(m). Thus, X (m) and X(m) have the same rank.
Definition 1.11. For a Poisson variety (M, { , }) and a point m M, the rank of the
Poisson matrix of { , } with respect to an arbitrary system of generators of F (M),
evaluated at m, is called the rank of { , } at m, denoted Rkm { , }. The maximum
maxmM Rkm { , } is called the rank of { , }, denoted Rk { , }. A point m M is
said to be a regular point of (M, { , }), if Rkm { , } = Rk { , }, otherwise it is said
to be a singular point of (M, { , }). The set of singular points of (M, { , }) is called
the singular locus of (M, { , }).
A more intrinsic definition of the rank of a Poisson structure can be given in terms of
the (Zariski) cotangent space to M at m. Recall that this vector space is intrinsically
defined as Im /Im2 , where Im denotes the ideal of F (M), consisting of all functions
which vanish at m; its dual space is the (Zariski) tangent space to M at m. We
denote the tangent space to M at m by Tm M, while the cotangent space is denoted
by Tm M. Taking generators of F (M) which vanish at m, the ideal Im corresponds
to elements of F (M) without constant term, so that F (M)/Im F, in a natural
way (i.e., by evaluation at m). A point m of M where the dimension of Tm M attains
its minimal value is called a smooth point of M. The smooth points of M form a
Zariski open subset of M and the dimension of Tm M in a smooth point m of M is
called the dimension of M, denoted dim M.
By the biderivation property (1.1), we have that Im2 , Im Im . Therefore, the
bracket {, } induces for every m M a welldefined skewsymmetric bilinear map:
m :
I m Im
F (M)
F.
Im2 Im2
Im
(1.19)
The bilinear map m is closely related to the Poisson matrix of { , } at m.
Namely, let x1 , . . . , xd be generators of F (M), so that x1 x1 (m), . . . , xd xd (m)
generate the ideal Im . If we express the bilinear map m in terms of the possibly linearly dependent equivalence classes [xi xi (m)] Im /Im2 , then the matrix
which we get has entries
m ([xi xi (m)], [x j x j (m)]) = xi xi (m), x j x j (m) (m) = xi , x j (m) ,
i.e., the resulting matrix is nothing but the Poisson matrix of { , }, with respect to
x1 , . . . , xd , evaluated at m. It follows that the rank of { , } at m can also be defined
as the rank of the intrinsically defined bilinear form m on Tm M.
The main properties of the rank are given in the following proposition.
1.3 Poisson Manifolds
15
Proposition 1.12. Let (M, { , }) be an affine Poisson variety.
(1)
(2)
(3)
For every m M, the rank of { , } at m is even;
The rank of { , } is at most equal to the dimension of M;
For every s N, the subset M(s) of M, defined by
M(s) := {m M  Rkm { , } 2s}
is open; in particular, the subset of points m M such that Rkm { , } =
Rk { , } is open and dense in M.
]/I , the
Proof. Writing F (M) as F[x1 , . . . , xd
Poisson matrix of { , } at m M is
the skewsymmetric matrix X(m) = ( xi , x j (m))1i, jd , whose rank is even. This
shows (1). Consider the open subset Rs gld of all d d matrices of rank greater
than or equal to 2s. Since M(s) is the inverse image of Rs by the continuous map
X :M
m
gld
xi , x j (m) 1i, jd
(1.20)
it is open; since the topology which is considered here is the Zariski topology, these
open subsets are dense as soon as they are nonempty. This yields the proof of (3).
Finally, let m be an arbitrary point of M and denote the rank of { , } at m by 2r.
Consider a point m of M(r) , which is a smooth point of M; such a point exists
because M(r) and the set of smooth points of M are both dense subsets of M. At
such a point m , the dimension of the (co)tangent space coincides with the dimension
of M, so that
Rkm { , } Rkm { , } dim Im /Im2 = dim M .
Since, in this formula, m M is arbitrary, Rk { , } dim M, which is the content
of (2).
A more geometric interpretation of the rank will be given in the next section, when
we consider Poisson structures on manifolds.
1.3 Poisson Manifolds
In this section, we introduce the notion of a Poisson structure on a (real or complex)
manifold. Our definitions and constructions are more geometric than in the previous section. A geometric approach to Poisson structures is in the case of manifolds
not only more natural, but in the case of complex manifolds it is even obligatory,
because a complex manifold cannot be replaced by its algebra of (global) functions,
as the latter may consist of constant functions only. Thus we will use the geometric analog of a skewsymmetric biderivation, which is a bivector field. We will at
the beginning of this section only shortly recall the basic terminology and notations
16
1 Poisson Structures: Basic Definitions
from differential geometry which we will use, referring the reader who needs more
details to Appendix B at the end of the book.
1.3.1 Bivector Fields on Manifolds
We consider both real manifolds and complex manifolds, since many constructions
in Poisson geometry apply in the same way if all manifolds are considered real or
if they are all considered complex; we will therefore make simple statements such
as Let M and N be two manifolds and let : M N be a map, which the reader
may specialize to Let M and N be two differentiable manifolds and let : M N
be a smooth map or to Let M and N be two complex manifolds and let : M N
be a holomorphic map. Similarly, dim M stands for the real or complex dimension,
according to the context, F (M) stands for the appropriate algebra of functions, F
stands for R or C, and so on. In both cases, the coordinates, real or complex, will be
denoted by x = (x1 , . . . , xd ), where d = dim M.
For m M the tangent space to M at m is denoted by Tm M. Elements of Tm M
are by definition pointwise derivation at m, i.e., they are linear forms on the vector
space of all function germs at m, satisfying, for all functions F and G, defined on a
neighborhood of m in M,
m (Fm Gm ) = F(m) m Gm + G(m) m Fm ;
in this formula, Fm stands for the germ of F at m. The dual space to Tm M is the
cotangent space, denoted Tm M. The canonical pairing between Tm M and Tm M is
denoted by , . A map : M N between manifolds leads for every m M
to a linear map Tm : Tm M T (m) N, called the tangent map of at m. Upon
identifying the tangent spaces to F with F, the tangent map leads to the differential
of a function; specifically, if F is a function on M, then we view the differential of
F at m, denoted dm F as a linear function on Tm M, that is, as an element of Tm M.
We denote by X1 (M) the F (M)module of vector fields on M and for V X1 (M)
and m M we denote by Vm the value of V at m (which is an element of Tm M). We
often use (and define) vector fields on M through their action on (local) functions:
when U is a nonempty open subset of M and V is a vector field on M (or on U)
then V [F] denotes the function on U, defined by
V [F] : U F
m Vm Fm .
(1.21)
We also write Vm [F] for Vm Fm , so that
Vm [F] = V [F](m) = dm F, Vm .
(1.22)
1.3 Poisson Manifolds
17
Notice that we use square brackets to denote the action of a vector field on a function, as we did in the case of a derivation; this notation will be extended to bivector
fields in this section, and to multivector fields in Chapter 3.
We now come to the definition of a bivector field on a manifold. To do this, we
first introduce the notion of a pointwise biderivation.
Definition 1.13. Let M be a manifold and let m M. A bilinear map
Bm :
Fm (M) Fm (M)
is a pointwise biderivation of F (M) at m, if for all functions F, G and H, defined in
a neighborhood of m in M,
Bm (Fm Gm , Hm ) = F(m) Bm (Gm , Hm ) + G(m) Bm (Fm , Hm ) ,
Bm (Hm , Fm Gm ) = F(m) Bm (Hm , Gm ) + G(m) Bm (Hm , Fm ) .
We will usually consider pointwise skewsymmetric biderivations. Given two pointwise derivations m and m at m, we construct a pointwise skewsymmetric biderivation at m by
m Fm m Gm
(m m )(Fm , Gm ) :=
,
m Fm m Gm
for all functions F and G, defined in a neighborhood of m. It follows easily from the
Hadamard
B.2) that the basic pointwise skewsymmetric bideriva
(Lemma
lemma
x j
span the vector space of all pointwise skewsymmetric
tions xi
m
biderivations at m; therefore, we denote the latter space by 2 Tm M.
The passage from pointwise skewsymmetric biderivations to bivector fields is
similar to the passage from pointwise derivations to vector fields, detailed in Appendix B. Namely, we consider a map P which assigns to every point m of M an element Pm of 2 Tm M. We say that P is a (smooth) bivector field on M, if for every open
subset U of M, and for all functions F, G F (U), one has that P[F, G] F (U),
where P[F, G] is the function on U, which is defined by
P[F, G](m) := Pm (Fm , Gm ) ,
for every m U. We also write Pm [F, G] for Pm (Fm , Gm ). Clearly, P is a (smooth)
bivector field as soon as P[xi , x j ] F (U) for some collection of coordinate charts
(U, x) which cover M, and for all i, j, with 1 i < j d. Notice that, as we did
for vector fields, we enclose the arguments of a bivector field in square brackets.
Just like vector fields, bivector fields can be restricted to open subsets; the same
notation will be used for P and each of its restrictions to open subsets. The coordinate expression for a bivector field on a coordinate chart (U, x) takes the following
18
1 Poisson Structures: Basic Definitions
form:2
P=
P[xi , x j ]
1i< jd
.
xi x j
(1.23)
In differential geometric terms, one often refers to bivector fields as being (skewsymmetric) tensors (of type (2, 0)), which refers to the fact that they are F (M)linear
objects. In view of the biderivation property for a bivector field, this may seem
absurd, but the point is that, as a tensor, a bivector field acts on a pair of differential
oneforms, rather than on a pair of functions. To explain this, notice that the value
of a bivector field P on two elements F and G of F (M), at m M, depends on the
differentials dm F and dm G only. Indeed, (1.23) yields
d
P[F, G](m) =
i, j=1
P[xi , x j ](m)
F
G
(m)
(m) ,
xi
xj
where xFi (m) is the ith component of dm F, and similarly for
of (1.22),
F
dm F,
=
[F](m) =
(m) .
xi m
xi
xi
(1.24)
G
x j (m):
in view
The fact that P[F, G](m) depends on dm F and dm G only, implies that we can define
a skewsymmetric F (M)bilinear map from 1 (M), the F (M)module of differential oneforms on M, to F (M): simply put
P(dF,
dG) := P[F, G] .
We will in the sequel usually make no distinction between the bivector field P and
and denote both by the same letter P. Note also that the tensorial
the (2, 0)tensor P,
interpretation, or equivalently (1.23), implies that we may specialize P to points
m M, giving a skewsymmetric, Fbilinear form
Pm : Tm M Tm M F ,
(1.25)
which is the differential geometric analog of (1.19). As in the algebraic case, Pm
totally determines the bivector field P at m.
Just like for vector fields, we can also define the pushforward P of a bivector
field P by a diffeomorphism : M N between two manifolds. The bivector field
P on N is defined by setting3
( P) (m) := 2 (Tm )P ,
for all m M.
2
See (B.8) for the coordinate expression of a vector field, of which (1.23) is a direct generalization.
For a linear map : V W between two vector spaces, the linear map 2 : 2V 2W is
defined by (2 )(v w) := (v) (w), for all v, w V . See Appendix A for details and generalizations.
3
1.3 Poisson Manifolds
19
Remark 1.14. Vector fields on a manifold M can be defined as sections of its tangent bundle, T M, the vector bundle over M, whose fiber over m M is the vector
space Tm M. Similarly, bivector fields on M can be defined as (smooth or holomorphic) sections of 2 T M M, where the latter vector bundle has as fiber over m M
the vector space 2 Tm M.
1.3.2 Poisson Manifolds and Poisson Maps
We are now ready to define the notion of a Poisson structure on a (real or complex)
manifold M. The idea is that a bivector field on M leads for every open subset U
of M to a skewsymmetric product on F (U), which is demanded to make F (U)
into a Poisson algebra.
Definition 1.15. Let be a bivector field on a manifold M. We say that is a Poisson structure on M if for every open subset U of M, the restriction of to U makes
F (U) into a Poisson algebra. We then call (M, ) a Poisson manifold.
The Poisson structure of a Poisson manifold (M, ) will often be denoted by { , },
in particular the Poisson bracket [F, G] of two functions F, G, defined on an open
subset U of M, will usually be denoted by {F, G}. The notation {F, G} for [F, G]
will be referred to as the bracket notation.
Let (U, x) be a coordinate neighborhood of a ddimensional Poisson manifold
(M, ). On U, the bivector field can, according to (1.23), be written as
xi , x j
xj
i
1i< jd
(1.26)
so that the matrix X := xi , x j 1i, jd encodes the restriction of to U. This
matrix X, whose elements belong to F (U), is called the Poisson matrix of with
respect to the coordinates x1 , . . . , xd . It is the differential geometric analog of the
Poisson matrix, introduced in Section 1.2.2 in the context of affine Poisson varieties.
Evaluating the Poisson matrix X at m M, we get the matrix of the skewsymmetric
bilinear map m : Tm M Tm M F, defined by the Poisson structure at m.
Proposition 1.16. Let = { , } be a bivector field on a manifold M of dimension d.
Then is a Poisson structure on M if and only if one of the following equivalent
conditions holds.
(i)
For every open subset U of M and for all functions F, G, H F (U), the
Jacobi identity holds:
{{F, G} , H} + {{G, H} , F} + {{H, F} , G} = 0 ;
(ii)
(1.27)
For some collection of coordinate charts (U, x) of M which cover M and for
all functions F, G, H F (U), the Jacobi identity (1.27) holds;
20
(iii)
1 Poisson Structures: Basic Definitions
For some collection of coordinate charts (U, x) of M which cover M and for
all i, j, k with 1 i < j < k d the following equality holds:
d
=1
(iv)
xk
x jk
xi j
xki
+ xi
+ x j
x
x
x
=0,
(1.28)
where xi j := xi , x j , for 1 i, j d;
For some collection of coordinate charts (U, x) of M which cover M and for
all i, j, k with 1 i < j < k d, the Jacobi identity holds:
xi , x j , xk + x j , xk , xi + {xk , xi } , x j = 0 .
(1.29)
Proof. Clearly, (i) holds if and only if makes every F (U) into a Poisson algebra,
hence is a Poisson structure if and only if (i) holds. (ii) is a consequence of (i), by
specialization. The equivalence of (ii) and (iii) follows from the fact that
{{F, G} , H} + (F, G, H)
xi j F G H
+ (F, G, H)
x xi x j xk
i, j,k,=1
d
d
xi j
F G H
+ (i, j, k)
,
= xk
xi x j xk
i, j,k=1 =1
d
xk
(1.30)
where we used (1.26) to compute the Poisson brackets. Notice that this computation
shows that the value of (1.30) at a point m U depends only on F, G and H in a
neighborhood of m, in other words on the germs Fm , Gm and Hm . Therefore, if the
Jacobi identity holds on a coordinate neighborhood (U, x), then it will hold on every
open subset V U, which leads to the implication (ii) (i). The equivalence of
(iii) and (iv) is clear because (1.28) is just (1.29) written out explicitly, using (1.26).
Remark 1.17. In the case of a real manifold, every germ is the germ of a globally
defined function, hence it suffices to verify the Jacobi identity for all triples of functions on M. For complex manifolds, this is not sufficient, see Remark B.4.
We now turn to the notion of a Poisson map between two Poisson manifolds.
Definition 1.18. A map : M N between Poisson manifolds (M, ) and (N, )
is called a Poisson map if for all open subsets U M and V N, with (U) V , the
induced map : F (V ) F (U), which is defined for all F F (V ) by (F) :=
F , is a morphism of Poisson algebras, i.e., for all functions F, G F (V ),
{F , G } = {F, G} ,
where { , } := and { , } := .
(1.31)
1.3 Poisson Manifolds
21
The condition that : M N be a Poisson map can be written directly in terms
of the bivector field, without reference to local functions on N, as shown in the
following proposition.
Proposition 1.19. Let : M N be a map between two Poisson manifolds (M, )
and (N, ). Then is a Poisson map if and only if
2 (T ) = ,
(1.32)
i.e., 2 (Tm )m = (m) for every m M.
Proof. Let F and G be functions, defined on a neighborhood of (m) in N, and
let (U, x) be a coordinate chart for M around m, so that can be written on U as
= i< j xi j xi x j .
[F, G]
= (Tm ) xi j (m)
xi m
xj m
1i< jd
= xi j (m) Tm
Tm
[F, G]
xj m
i m
1i< jd
[F , G ]
= xi j (m)
xj m
i m
1i< jd
2 (Tm )m [F, G]
= {F , G } (m) ,
where we have set, as usual, { , } := . Setting { , } := we clearly have that
(m) [F, G] = {F, G} ( (m)) ,
so that 2 (Tm )m = (m) if and only if {F , G } (m) = {F, G} ( (m)) for
all functions F and G, defined in a neighborhood of (m). This proves that is a
Poisson map if and only if 2 (T ) = .
In the context of Poisson manifolds, one calls the Hamiltonian derivation XH ,
associated to a function H F (M), the Hamiltonian vector field, associated to H.
One says then that H is a Hamiltonian function, corresponding to this vector field.
More generally, a vector field V is called a locally Hamiltonian vector field, if every
point m M belongs to a neighborhood U on which V is Hamiltonian, i.e., there
exists H F (U) such that V = XH on U; such a function is called a local Hamiltonian of V . According to (1.26), the (local) Hamiltonian vector field XH , associated
to H F (U), is given in a coordinate chart (U, x) by
XH =
H
xi , x j
.
x j xi
i, j=1
d
(1.33)
It follows that the first order differential equation, which describes the integral
curves of XH in a coordinate chart (U, x), is given by
22
1 Poisson Structures: Basic Definitions
dxi
= {xi , H} =
dt
H
xi , x j
,
xj
j=1
d
i = 1, . . . , d .
(1.34)
Besides the Cas(M)module Ham(M) of Hamiltonian vector fields on M, we will
also consider the vector space Hamm (M), which consists of all tangent vectors of
the form (XH )m , where H is an arbitrary function, defined on some neighborhood
of m. Elements of Hamm (M) are called Hamiltonian vectors at m.
Definition 1.20. For a Poisson manifold (M, ) and a point m M the integer Rk m ,
which is also the rank of every Poisson matrix of at m, is called the rank of at m,
denoted Rkm . One says that (M, ) is a regular Poisson manifold when its rank is
constant (independent of m M), and that the rank is locally constant at m, when it
is constant on some neighborhood of m in M. The rank of is said to be maximal at
m M, when it coincides with the dimension of M. The maximum maxmM Rkm
is called the rank of , denoted Rk . A point m M is said to be a regular point
of (M, ), if Rkm = Rk , otherwise it is said to be a singular point of (M, { , }).
The set of singular points of (M, ) is called the singular locus of (M, ).
The main properties of the rank are given in the following proposition.
Proposition 1.21. Let (M, ) be a Poisson manifold and let m M.
(1)
(2)
Rkm is even and is equal to dim Hamm (M);
For every s N, the subset M(s) of M, defined by
M(s) := {m M  Rkm 2s}
is open (and dense in the complex case); in particular, the subset of points
m M such that Rkm = Rk is open.
Proof. The rank of at m is the rank of the bilinear map m , so it is the dimension
of the image of the linear map Tm M Tm M, defined by m ( , ). In turn,
the latter equals the dimension of Hamm (M), since every Tm M can be written
as dm F, for some function F, defined in a neighborhood of m. Also, since m is
skewsymmetric, its rank is even. This shows (1). The set Rs of d d matrices of
rank greater than or equal to 2s is an open subset of gld . Let m be a point in M(s) and
let (x1 , . . . , xd ) be local coordinates on a neighborhood U of m. The restriction to U
of M(s) is open since it is the inverse image of Rs by the continuous map X : U gld
defined by m xi , x j (m) 1i, jd . This yields the proof of (2).
In the real case, the rank of a Poisson structure does not necessarily attain its maximum on a dense open subset. This is shown in the following example.
Example 1.22. Let be a smooth function on R2 which is positive on the interior
of the unit disk and which is zero elsewhere. Then the rank of the Poisson structure
on R2 , which is defined by the bivector field / x / y, is of rank 2 only on
the inside of the disk.
1.3 Poisson Manifolds
23
To finish this section, we show that the flow of a (locally) Hamiltonian vector field
leaves the Poisson structure invariant.
Proposition 1.23. Let (M, ) be a Poisson manifold. The Lie derivative of with
respect to every (locally) Hamiltonian vector field is zero. As a consequence, the
flow of each (locally) Hamiltonian vector field preserves the Poisson structure.
Proof. Let U be an arbitrary open subset of M and let F, G F (U). We need to
show that the bivector field LV , which is the Lie derivative of with respect
to V , vanishes on the pair (F, G). To do this, we use the classical formula for the
Lie derivative of a tensor; in the case of a bivector field P, this formula is given by
LV P[F, G] := V [P[F, G]] P[V [F], G] P[F, V [G]] ,
(1.35)
for all F, G F (U) (see Section 3.3.4 for generalizations). For P := = { , } and
V := XH , with H F (U), this yields
LXH [F, G] = XH [{F, G}] {XH [F], G} {F, XH [G]}
= {{F, G} , H} {{F, H} , G} {F, {G, H}} ,
whose vanishing is equivalent to the Jacobi identity (1.2).
A vector field V is called a Poisson vector field if taking the Lie derivative with
respect to V kills the Poisson structure, i.e., if LV = 0. Proposition 1.23 implies
that all Hamiltonian vector fields are Poisson vector fields. This property will be
reformulated in cohomological terms in Section 4.1.
Remark 1.24. For a Poisson algebra (A , , ) one says that a derivation V of (A , )
is a Poisson derivation if the Lie derivative of with respect to V vanishes. One
has as in Proposition 1.23 that all Hamiltonian derivations are Poisson derivations;
there is, however, no interpretation of Poisson derivations in terms of flows, since
the notion of a flow does not make sense for derivations of algebras, in general.
1.3.3 Local Structure: Weinsteins Splitting Theorem
This section deals with Weinsteins splitting theorem, which states that, in the neighborhood of a point where the rank of the Poisson structure is 2r, the Poisson manifold is a product of a symplectic manifold of dimension 2r, and a Poisson manifold
which has rank zero at the origin. It implies the Darboux theorem, which states that
all symplectic manifolds of the same dimension are locally isomorphic, there are
no local invariants in symplectic geometry, as well as a generalization to regular
Poisson manifolds.
Theorem 1.25 (Weinsteins splitting theorem). Let (M, ) be a (real or complex) Poisson manifold, let m M be an arbitrary point and denote the rank
24
1 Poisson Structures: Basic Definitions
of at m by 2r. There exists a coordinate neighborhood U of m with coordinates q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs , centered at m, such that, on U,
+ k (z)
z
i
i
k
i=1
1k<s
r
(1.36)
where the functions k are (smooth or holomorphic) functions, which depend
on z = (z1 , . . . , zs ) only, and which vanish when z = 0. Such local coordinates
q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs are called splitting coordinates, centered at m.
Proof. We use induction on r. For the case r = 0, it is clear that for every Poisson
manifold (M, ) and for every point m such that the rank of at m is zero, an
arbitrary system of local coordinates (z1 , . . . , zd ), centered at m, works (d := dim M).
Let r N and assume that Theorem 1.25 holds true for every Poisson manifold, at
every point where the rank is 2(r 1). Let (M, ) be a Poisson manifold and let
m M be a point for which Rkm = 2r. We will show that the theorem holds for
(M, ) at m.
Since Rkm > 0, there exists a function p on a neighborhood of m, whose Hamiltonian vector field X p does not vanish at m M; we may suppose that p(m) = 0.
Since X p (m) = 0, there exists by the straightening theorem (Theorem B.7) a system of coordinates (q, y2 , . . . , yd ) on a neighborhood U of m, centered at m, such
that X p = / q. Writing { , } = , it follows that
{q, p} = X p [q] =
q
=1,
q
[Xq , X p ] = X{p,q} = X1 = 0 and X p [yi ] = 0 ,
for i = 2, . . . , d, on U . Writing Xq in terms of these coordinates,
X q = 1
+ i ,
q i=2 yi
we have that 1 = Xq [q] = {q, q} = 0, and all coefficients 2 , . . . , d are independent
of q, since Xq and X p = / q commute. At m, we have
d
1 = {p, q} (m) = Xq [p](m) = i (m)
i=2
p
(m) ,
yi
so that the vector field Xq = di=2 i / yi is independent of q and does not vanish at m. Applying the straightening theorem once more, we may introduce a system of coordinates (q, p , y3 , . . . , yd ) on a neighborhood of m, centered at m, where
p , y3 , . . . , yd depend on y2 , . . . , yd only, with
d
= i = Xq .
p
yi
i=2
1.3 Poisson Manifolds
25
Substituting p by p we consider (q, p, y3 , . . . , yd ) which is also a system of coordinates on a neighborhood U of m, since
p
= Xq [p] = {q, p} = 1 ,
p
in a neighborhood of m. Since p , y3 , . . . , yd depend on y2 , . . . , yd only, / q has the
same meaning in both coordinate systems, so that the Poisson brackets take in the
new coordinates the following form,
{q, p} = 1 ,
{q, yi } = Xq [yi ] = yi / p = 0 ,
{p, yi } = X p [yi ] = yi / q = 0 ,
for i = 3, . . . , d, and we conclude that the Poisson structure is given, in terms of
the coordinates q, p, y3 , . . . , yd , by
+ {yk , y }
.
q p 3k<d
yk y
(1.37)
To show that {yk , y } is independent of p and q, for all values of k, , we just have
to check that {{yk , y } , p} = {{yk , y } , q} = 0, an easy consequence of the Jacobi
identity for . The Jacobi identity also yields that the second term in (1.37) defines
a Poisson structure on a neighborhood V of the origin o of Fd2 . The Poisson
matrix of with respect to q, p, y3 , . . . , yd has a diagonal block form, where the
lower block is the Poisson matrix of with respect to y3 , . . . , yd . Thus, has
rank 2(r 1) at o, and by the induction hypothesis there exist local coordinates
q2 , . . . , qr , p2 , . . . , pr , z1 , . . . , zs centered at o, such that takes on V the following
form:
r
=
+ k (z)
.
pi 1k<s
zk z
i=2 qi
In terms of the system of coordinates (q1 , q2 , . . . , qr , p1 , p2 , . . . , pr , z1 , . . . , zs ), which
is centered at m, takes the required form (1.36), where we have set q1 := q and
p1 := p.
For a given point m in a Poisson manifold M, splitting coordinates are not unique.
We will see however in Section 5.3.3 that the Poisson structure, which is defined
in a neighborhood of z = 0 in Fs by the second term in (1.36), is unique, up to
isomorphism.
In terms of the splitting coordinates q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs , the Poisson
structure (1.36) has the block form
0 1r 0
1r 0 0
0 0
26
1 Poisson Structures: Basic Definitions
where Mats (F (M)) is given by i j := zi , z j and 1r denotes the identity
matrix of size r. It follows that = 0 in a neighborhood of m, when the rank of
is locally constant (= 2r) at m. This leads to the following strengthening of the
splitting theorem for points at which the rank is locally constant.
Theorem 1.26 (Darboux theorem). Let (M, ) be a (real or complex) Poisson
manifold of dimension d, and suppose that m is a point where the rank of is
locally constant and equal to 2r. There exists a coordinate neighborhood U of m
with coordinates (q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs ) such that, on U,
.
pi
i=1 qi
r
(1.38)
Moreover, is locally of the form (1.38), in terms of arbitrary splitting coordinates
on M. Such coordinates are called Darboux coordinates.
Corollary 1.27. The only local invariant of a regular (smooth or complex) ddimensional Poisson manifold (M, ) is its rank Rk .
To close this section we state a generalization of Weinsteins splitting theorem
which we will use in Chapter 12. It provides a set of Darboux coordinates for the
Poisson structure , which contains a given set p1 , . . . , pr of functions in involution
(i.e., functions which pairwise commute for the Poisson structure), whose Hamiltonian vector fields are assumed to be independent at a point m M. See [120] for a
proof and see Section 12.3 below for an application.
Theorem 1.28. Let m be a point of a Poisson manifold (M, ) of dimension d. Let
p1 , . . . , pr be r functions in involution, defined on a neighborhood of m, which vanish
at m and whose Hamiltonian vector fields are linearly independent at m. There exist,
on a neighborhood U of m, functions q1 , . . . , qr , z1 , . . . , zd2r , such that
(1)
(2)
The d functions (q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd2r ) form a system of coordinates on U, centered at m;
The Poisson structure is given on U by
d2r
+ gi j (z)
zj
i
i
i
i=1
i, j=1
r
(1.39)
where each function gi j (z) is a smooth function on U and is independent of
q1 , . . . , q r , p 1 , . . . , p r .
The rank of at m is 2r if and only if all the functions gi j (z) vanish for z = 0.
1.3.4 Global Structure: The Symplectic Foliation
We now come to the global structure of a Poisson manifold. As we will explain in
this section, a Poisson manifold naturally decomposes into immersed submanifolds,
1.3 Poisson Manifolds
27
where each submanifold inherits a Poisson structure from the ambient Poisson structure; since this induced Poisson structure is of maximal rank, it is symplectic,4 so
one usually refers to this decomposition as the symplectic foliation of the Poisson
manifold. We combine two approaches, which are both due to Weinstein [199], to
obtain the symplectic foliation; both have their advantage, a fact that we exploit to
get at the decomposition in a natural way and to prove its properties with minimal
effort. For another approach, using the theory of generalized distributions, we refer
to [125]. Throughout this section (M, ) is an arbitrary Poisson manifold.
The idea which underlies the decomposition is based on the following two observations. First, recall from Proposition 1.23 that the flow of every Hamiltonian
vector field preserves the Poisson structure, hence its rank at each point. Second,
the rank of the Poisson structure at a point is precisely the dimension of the space
of Hamiltonian vector fields, at that point. Thus, the flow of all Hamiltonian vector
fields, starting from a given point m M where the rank of the Poisson structure
is 2r, should trace out locally a submanifold of dimension 2r, on which the Poisson
structure has constant rank 2r. This leads, for m M, to the following definition of
a subset of M,
Sm (M) := m M  a piecewise Hamiltonian path in M from m to m .
By a Hamiltonian path from m to m we mean a curve , defined on an open neighborhood5 of [0, 1], with (0) = m and (1) = m , which is an integral curve of a
Hamiltonian vector field XF , where F is a function, defined on an open neighborhood of ([0, 1]). More generally, when points m0 , . . . , mN in M are such that there
exists a Hamiltonian path from mi1 to mi , for i = 1, . . . , N, we say that there exists
a piecewise Hamiltonian path from m0 to mN .
It is clear that the subsets Sm (M) define a welldefined global decomposition
of M: the subsets Sm (M), with m M fill up M, and two subsets Sm (M) and
Sm (M) are disjoint, unless they coincide. Notice however that some care has to be
taken when passing to open subsets: if U is an open subset of M, then Sm (U) is,
in general, not the intersection of Sm (M) and U: the subset Sm (U) is by definition pathconnected, while Sm (M) U does not have to be connected, even if U is
connected.
The differential geometric properties of Sm (M) are best understood by relating
Sm (M) to an immersed submanifold of M, passing through m, which appears naturally in the splitting theorem (Theorem 1.25). Recall that a subset N of a manifold
M is called an immersed submanifold if the inclusion map : N M is an immersion, i.e., for every m N the tangent map Tm : Tm N Tm M is injective. We are
viewing here N as a manifold in its own right, i.e., with a topology, which is not
necessarily the induced topology from M (one speaks of an embedded submanifold
4
Symplectic manifolds will be introduced and discussed in Chapter 6. For now, the reader may
think of a symplectic manifold as being a Poisson manifold (M, ) whose rank Rkm at each point
is equal to the dimension of M.
5 The neighborhood is taken in R when M is a real manifold and is taken in C when M is a complex
manifold.
28
1 Poisson Structures: Basic Definitions
when the manifold topology on N is the induced topology). However, locally, with
respect to the manifold topology on N, every point in N has a neighborhood, which
sits in M in the same way an open subset of Fd sits in Fd (d = dim N d = dim M).
To construct this immersed submanifold of M, through m, we choose on a neighborhood U of m, splitting coordinates (at m) which we denote by x = (q1 , . . . , qr ,
p1 , . . . , pr , z1 , . . . , zs ) and we define
Sm (U, x) := m U  z1 (m ) = = zs (m ) = 0 .
It is clear that Sm (U, x) is an embedded submanifold of M of dimension 2r, and
that (q1 , . . . , qr , p1 , . . . , pr ) (restricted to Sm (U, x)) forms a system of coordinates
on Sm (U, x). Since the Poisson structure has on U the form
+ k (z)
z
i
i
k
i=1
1k<s
r
(1.40)
where the functions k (z) vanish for z = 0, it follows that at points m Sm (U, x),
the Poisson structure takes the form ri=1 qi pi , which defines a Poisson structure
on Sm (U, x), which we denote by S . The map : (Sm (U, x), S ) (M, ) is a
Poisson map, since for all functions F, G F (U), and for every m Sm (U, x), we
have that
r
F G
F G
{F, G} ((m )) = {F, G} (m ) =
(m )
(m )
(m )
(m )
pi
pi
qi
i=1 qi
= {F , G }S (m ) .
In the language of Section 2.2, (Sm (U, x), S ) is a Poisson submanifold of (M, ).
Notice that S has constant rank 2r = dim Sm (U, x). In the following proposition
we show that, for each m M, the subsets Sm (U) and Sm (U, x) coincide, for small
enough U.
Proposition 1.29. Let m be a point in a Poisson manifold (M, ) and let x =
(q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs ) be splitting coordinates at m, on a coordinate neighborhood U, with x(U) convex. Then Sm (U) = Sm (U, x).
Proof. For F F (U) and m Sm (U, x) we have, in view of (1.40), that
XF [zi ](m ) = {zi , F} (m ) =
i (m ) z (m ) = 0 ,
=1
since the functions i vanish at all points m of Sm (U, x). It follows that all Hamiltonian vector fields of U are tangent to Sm (U, x) at points m Sm (U, x). Thus,
a (piecewise) Hamiltonian path in U which starts at m must stay in Sm (U, x). It
follows that Sm (U) Sm (U, x). In order to show the other inclusion, suppose that
m Sm (U, x). We show that there exists a Hamiltonian path from m to m in U.
Consider the (linear) function H on U, given by
1.3 Poisson Manifolds
29
r
H := (qi (m )pi pi (m )qi ) ,
i=1
whose (constant) Hamiltonian vector field XH is given by
H
dqi
=
= qi (m ) ,
dt
p
i
XH :
H
dp
i =
= pi (m ) .
dt
qi
Consider the integral curve of XH , defined for t in a small neighborhood of [0, 1]
(in R or C) by
: t (q1 (m )t, . . . , qr (m )t, p1 (m )t, . . . , pr (m )t, 0, . . . , 0) .
The image of is contained in U, because x(U) F2r+s is a convex open subset.
Since is a Hamiltonian path from m = (0) to m = (1) in U, it follows that
m Sm (U).
We use Proposition 1.29 to build a topology and a differential structure on each
one of the subsets Sm (M). A subset V Sm (M) is by definition an open subset if
and only if for all splitting coordinates (U, x) at m V , with x(U) convex, V U is
an open subset of Sm (U, x) = Sm (U). Similarly, an algebra of functions is defined
on Sm (M) by saying that F : Sm (M) F is smooth (holomorphic in the complex
case), when its restriction to every such Sm (U, x) = Sm (U) is smooth (or holomorphic). Since the topology and the differential structure on these subsets Sm (U)
is intrinsically defined, as embedded submanifolds of M, these definitions yield a
welldefined manifold structure on Sm (M). In general, the topology which we defined on Sm (M) is different from (finer than) the induced topology which Sm (M)
inherits as a subset of M: it is an immersed submanifold, which is not necessarily
an embedded submanifold. The Poisson structure on the submanifolds Sm (U, x)
yields a welldefined Poisson structure of maximal rank (symplectic structure) on
Sm (M). The results of the present section, combined, prove the following fundamental theorem.
Theorem 1.30. Every Poisson manifold (M, ) is the disjoint union of immersed
submanifolds, whose tangent spaces are spanned by the Hamiltonian vector fields
of (M, ). The Poisson structure, restricted to each of these submanifolds yields
a Poisson structure of maximal rank (symplectic structure). This decomposition is
called the symplectic foliation of M and the immersed submanifolds are called the
symplectic leaves of M. For m M the symplectic leaf which contains m is given by
Sm (M) = m M  a piecewise Hamiltonian path in M from m to m .
The first part of the theorem can also be restated by saying that the (singular)
distribution, defined by the Hamiltonian vector fields, is integrable (admits an inte
30
1 Poisson Structures: Basic Definitions
gral manifold through each point), see [125, Appendix 3]. The theorem leads to the
following description of the symplectic leaves of a Poisson manifold.
Proposition 1.31. Let (M, ) be a Poisson manifold of dimension d and rank 2r.
(1)
(2)
(3)
A function F on M is a Casimir function if and only if F is constant on each
symplectic leaf;
The (nonempty open subsets of the) symplectic leaves are the smallest embedded manifolds of M which are Poisson submanifolds (i.e., such that the
inclusion map is a Poisson map);
If Y is a Poisson vector field on M which is tangent to every symplectic leaf
of M, then Y is Hamiltonian in the neighborhood of every point m M where
the rank of is 2r.
Proof. The fact that the manifold Sm (M) admits locally (in the sense of the topology on Sm (M) which we have constructed) the description as a submanifold
Sm (U, x), leads at once to the proof of (1) and (2). We proceed to the proof of (3).
If the rank of at m is 2r, so that m is a regular point of , then there exists local
coordinates q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd2r in a contractible neighborhood U of m
with respect to which the Poisson structure is given by:
pi
i
i=1
r
(1.41)
The vector fields q , p , . . . , qr , pr span the symplectic leaves of on U. There1
1
fore, every vector field Y , which is tangent to the symplectic leaves of , is of the
form
r
r
Y = Fi
+ Gi
(1.42)
qi
i
i=1
i=1
for some smooth or holomorphic functions F1 , . . . , Fr , G1 , . . . , Gr , defined on U. Suppose now that Y is a Poisson vector field, so that LY = 0. Then one computes
easily from (1.35), (1.41) and (1.42) that
Fj
Fi
=
,
qj
qi
Gj
Gi
=
pj
pi
and
Fi
Gi
=
,
pj
qj
for i, j = 1, . . . , r. By the classical Poincare lemma, there exists a function H, defined
on the contractible subset U, which satisfies
Fi =
H
qi
and
Gi =
H
,
pi
for i = 1, . . . , r. Hence,
r
d2r
H
H
H
X qi +
X pi +
Xzk = Y ,
p
i
i
i=1
i=1
k=1 zk
r
XH =
which shows that Y is a Hamiltonian vector field on U.
1.3 Poisson Manifolds
31
Locally, the regular leaves (i.e., the leaves which pass through a point where the
rank is maximal, hence locally constant) are given as the level sets of the (local)
Casimir functions. This is shown in the following proposition.
Proposition 1.32. Let (M, ) be a Poisson manifold of dimension d, let U be a nonempty open subset of M and let F1 , . . . , Fs F (U), satisfying:
(1)
(2)
(3)
The rank of is constant on U and is equal to d s;
The functions F1 , . . . , Fs are Casimirs of the restriction of to U;
For every point m of U, the differentials dm F1 , . . . , dm Fs are independent.
Then the symplectic foliation of the restriction of to U coincides with the foliation
which is defined on U by the map F := (F1 , . . . , Fs ) : M Fs .
Proof. Under the stated assumptions, the restriction of to U is a regular Poisson
structure, hence the symplectic foliation is a regular foliation of U, where every
leaf has dimension d s. Since F := (F1 , . . . , Fs ) has constant, maximal rank, F
also defines a foliation F on U, whose leaves are (d s)dimensional. In view
of (1) in Proposition 1.31, F is constant on every symplectic leaf in U, hence every
symplectic leaf is contained in a leaf of F . Since all symplectic leaves and all leaves
of F have the same dimension (d s), it follows that both foliations coincide.
In good cases most or all of the symplectic leaves are level sets of the Casimir
functions, but this is not true in general.
Example 1.33. Consider the Poisson structure / x / y on R3 , where x, y, z are
the natural coordinates on R3 . Since translations over a constant vector are obviously
Poisson maps, it descends to a Poisson structure on every torus R3 / , where is a
lattice in R3 . Unless is very special, all symplectic leaves are dense on the torus,
hence they cannot be the level sets of a (Casimir) function.
Proposition 1.34. Let 1 and 2 be two Poisson structures on a manifold M. Suppose that both Poisson structures define the same symplectic foliation on M and that
for every symplectic leaf S (of both structures), the Poisson (symplectic) structure
induced on S by 1 is the same as the Poisson structure induced on S by 2 . Then
1 and 2 are equal.
Proof. Let 1 = { , }1 and 2 = { , }2 be two Poisson structures on M which
determine the same symplectic foliation on M. Let m M and let S be the leaf
which passes through m; we denote by S the inclusion map S : S M. It
is assumed that the induced Poisson structures { , }i,S on S coincide, so that
{F S , G S }1,S = {F S , G S }2,S for every F, G F (M). Since S is a
Poisson map, when M is equipped with 1 , or with 2 , we have for every m S ,
{F, G}1 (m) = {F S , G S }1,S (m) = {F S , G S }2,S (m) = {F, G}2 (m) .
This applies to every point m of M, so that {F, G}1 (m) = {F, G}2 (m), for all m M,
which was to be shown.
32
1 Poisson Structures: Basic Definitions
1.4 Poisson Structures on a Vector Space
On a finitedimensional Fvector space V , one can consider its algebra of polynomial functions, which is canonically isomorphic to the symmetric algebra SV , but
when F = R one may also consider the algebra of smooth functions on V , while
when F = C one may consider the algebra of holomorphic functions on V . The difference corresponds to viewing V as an affine variety or as a real/complex manifold.
What makes a vector space V special, from either point of view, is that it admits a
basis, say (e1 , . . . , ed ), and hence a dual basis, consisting of linear forms xi : V F,
for i = 1, . . . , d.
(1) From the affine variety point of view, SV = F[x1 , . . . , xd ], so every function can
be written uniquely as a polynomial in x1 , . . . , xd . More generally, every derivation V
of SV can be uniquely written as V = di=1 Vi xi , where Vi := V [xi ], and similarly
for biderivations, such as Poisson structures.
(2) From the manifold point of view, x1 , . . . , xd are global coordinates on V , i.e., V
is covered by a single coordinate chart. Functions, vector fields, Poisson structures,
and so on, admit a global coordinate representation in terms of these global coordinates.
Many results about Poisson structures on a vector space take formally the same
form when the Poisson structures are considered on the algebra of polynomial functions or on the algebra of smooth/holomorphic functions. Therefore we will often
refer to F (V ) as the algebra of functions on V , independently of whether the algebra
of polynomial, smooth or holomorphic functions is considered; however, when F is
different from R and C, then F (V ) always stands for the algebra of polynomial
functions on V . To stress that we consider a Poisson bracket on the algebra of polynomial functions on V , we often use the term polynomial Poisson structure.
Vector fields on a real or complex vector space V can be identified with derivations of the algebra of smooth or holomorphic functions on V , so we will simply
write X1 (V ) for the Lie algebra of vector fields on V ; even when F is different
from R and C, we refer to elements of X1 (V ) as vector fields on V . Similarly, we
refer to elements of X2 (V ) as bivector fields on V . Vector fields and bivector fields
admit in this general context the following description.
Proposition 1.35. Let V be a vector space and let F (V ) be its algebra of functions.
Let V be a vector field on V and let P be a bivector field on V . In terms of linear
coordinates x1 , . . . , xd for V , the elements V and P can be written (uniquely) as
d
V = Vi
i=1
,
xi
P=
1i< jd
Pi j
,
xi x j
where Vi F (V ) and Pi j F (V ), for all 1 i, j d. In particular, every vector
field on V is determined by its values on a system of linear coordinates on V , and
every bivector field on V is determined by its values on all pairs of functions, taken
from a system of linear coordinates on V .
1.4 Poisson Structures on a Vector Space
33
Suppose now that we have a bivector field = { , } on V and that we wish to
express that is a Poisson structure. When F (V ) is the algebra of polynomial
functions on V , then is a Poisson structure if and only if satisfies the Jacobi
identity,
{F, {G, H}} + {G, {H, F}} + {H, {F, G}} = 0 ,
(1.43)
for all F, G, H F (V ); however, when F (V ) is the algebra of smooth of holomorphic functions on V , precisely the same condition expresses that is a Poisson
structure, because V can be covered with a single coordinate chart. We formulate
this result in the following proposition, which is a transcription and a specialization
of Propositions 1.8 and 1.16 to the case of a vector space V .
Proposition 1.36. Let V be a finitedimensional vector space, with algebra of functions F (V ) and let (x1 , . . . , xd ) be a system of linear coordinates on V . Given
a skewsymmetric matrix (xi j )1i, jd of elements of F (V ), consider the skewsymmetric biderivation of F (V ), which is defined, for F, G F (V ), by
{F, G} :=
xi j
i, j=1
F G
,
xi x j
which is the unique skewsymmetric biderivation of F (V ) such that xi , x j = xi j ,
for all 1 i, j d. The following conditions are equivalent.
(i)
(ii)
is a Poisson structure on V ;
satisfies the Jacobi identity: for all F, G, H F (V ),
{F, {G, H}} + {G, {H, F}} + {H, {F, G}} = 0 ;
(iii)
For all i, j, k, with 1 i < j < k d,
xi , x j , xk + x j , xk , xi + {xk , xi } , x j = 0 ;
(iv)
For all i, j, k, with 1 i < j < k d,
d
=1
xk
x jk
xi j
xki
+ xi
+ x j
x
x
x
=0.
In this case, the skewsymmetric matrix (xi j )1i, jd is the Poisson matrix of with
respect to the coordinates x1 , . . . , xd .
At every point m V , the tangent space TmV is in a canonical way isomorphic to V ,
and similarly the cotangent space TmV is canonically isomorphic to V . Therefore,
one usually thinks of the Poisson bivector at m, which is a skewsymmetric bilinear
map m : TmV TmV F, as a skewsymmetric bilinear map V V F, as an
element of V V , or as a linear map V V . The rank of the latter linear map is the
rank of the Poisson structure at m.
It is clear that the results in this section are also valid when the vector space V is
replaced by a nonempty open subset of V .
34
1 Poisson Structures: Basic Definitions
1.5 Exercises
1. Suppose that V is a derivation of A := F[x1 , . . . , xd ]. Show that, for every
F A,
d
F
V [xi ] .
V [F] =
i=1 xi
Deduce from it an alternative proof of Proposition 1.6.
2. Two elements F, G of a Poisson algebra (A , , ) are said to be in involution
if {F, G} = 0. Prove Poissons theorem: if F and G are in involution with a third
element H A , then {F, G} is in involution with H. The original motivation of
Poissons theorem is that, since functions which are in involution with the Hamiltonian are constants of motion, taking the Poisson bracket of two constants of motion
produces a (a priori new) constant of motion.
3. Let be a polynomial Poisson structure on Fd , with Poisson matrix X, and
suppose that X vanishes at the origin o Fd (i.e., all entries xi j of X vanish at o).
Let Y be the matrix whose entry yi j is the linear part of xi j , for 1 i, j d. Show
that Y is a Poisson matrix.
4. Let (M1 , 1 ) and (M2 , 2 ) be two affine Poisson varieties, or two Poisson manifolds. Let : M1 M2 be a Poisson map. Show that
Rkm M1 Rk (m) M2 ,
for every m M1 .
5. Given a skewsymmetric biderivation { , } on a commutative associative algebra A , define its Jacobiator J as the skewsymmetric trilinear map A 3 A ,
given by
J (F, G, H) := {F, {G, H}} + {G, {H, F}} + {H, {F, G}} .
Show that J is a derivation in each of its arguments (as in (1.4); in the language of
Section 3.1, it is a triderivation) and derive from it an alternative proof of Proposition 1.8.
6. Consider on F2 the bivector field , defined by {x, y} := x2 , and let Y denote the
vector field on F2 , defined by Y [x] := 0 and Y [y] := x. Show that Y is a Poisson
vector field, but that there exists no neighborhood of o = (0, 0) on which Y is a
Hamiltonian vector field (see item (3) of Proposition 1.31).
7. The following is an open problem: given a Poisson algebra (A , , { , }) of finite type, do there
generators x1 , . . . , xd of A and representatives xi j = x ji
exist
F[x1 , . . . , xd ] of xi , x j such that
x jk
xi j
xki
xk x + xi x + x j x = 0 ,
=1
d
1.6 Notes
35
for every 1 i, j, k d; in other words, such that the matrix X := (xi j )1i, jd is a
Poisson matrix?
1.6 Notes
In the mathematics and physics literature, Poisson structures are most often considered in the case of smooth manifolds. In this context, Poisson structures are usually
defined as sections of the exterior square of the tangent bundle of the manifold (see
Vaisman [194]), as Lie algebra structures on the algebra of functions on the manifold (see Cannas da SilvaWeinstein [34] or DufourZung [63]) or as sheafs of
Poisson algebras on the manifold (see Przybylski [172]). Abstract Poisson algebras
are considered in BhaskaraViswanath [23]; see also Huebschmann [96]. For more
information on the general theory of real manifolds, we refer to Warner [198] or to
Spivak [187]; for complex manifolds, see Wells [203]. For a gentle introduction to
algebraic geometry, in particular the link between commutative associative algebras
and affine varieties, see Perrin [165] or Shafarevich [182].
The main result in this chapter is Weinsteins splitting theorem (Theorem 1.25),
from which we derived the symplectic foliation. This general theorem admits further
generalizations: for example, an equivariant version is given in DufourZung [63],
while a generalization to arbitrary Lie algebroids is given in Fernandes [75]. See
LaurentMirandaVanhaecke [120] for the CaratheodoryJacobiLie theorem for
Poisson manifolds, which was stated without proof in Theorem 1.28 above; for regular Poisson manifolds, this theorem goes back to Lie and Engel [128].
Chapter 2
Poisson Structures: Basic Constructions
In this chapter we give a few basic, general constructions which allow one to build
new Poisson structures from given ones. These constructions are fundamental and
will be used throughout the book. More advanced constructions, which all fall under
the general concept of reduction, will be discussed in detail in Chapter 5, while
several constructions which are specific to a particular class of examples will be
given in Part II.
As in the previous chapter, all Poisson algebras will be defined over an arbitrary
field F of characteristic zero, our Poisson varieties will be affine varieties over F
and our Poisson manifolds will be either real smooth manifolds (F = R) or complex
manifolds (F = C). We will adhere to our habit of reformulating all our algebraic
constructions in geometrical terms, or, conversely, present the geometrical construction in general algebraic terms. We stress that although the geometrical and algebraic
constructions are based on the same idea, their concrete implementation is usually
quite different.
Section 2.1 deals with the tensor product of Poisson algebras, which geometrically corresponds to the construction of a Poisson structure on the product of
two Poisson manifolds. We investigate in Section 2.2 the notion of a Poisson ideal,
whose geometrical counterpart is that of a Poisson submanifold. A Poisson structure cannot be restricted to any submanifold; a necessary and sufficient conditions
for this is that all Hamiltonian vector fields be tangent to the submanifold. In Section 2.3, we show on the one hand how real and complex Poisson structures are
related, and on the other hand how complex algebraic and holomorphic Poisson
structures are related. In Section 2.4, we assemble a few constructions which are of
a different nature, namely changing the base field, localization and germification.
Unless otherwise stated, F stands throughout this chapter for an arbitrary field of
characteristic zero.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 2, SpringerVerlag Berlin Heidelberg 2013
37
38
2 Poisson Structures: Basic Constructions
2.1 The Tensor Product of Poisson Algebras and the Product
of Poisson Manifolds
The product of two Poisson algebras/varieties/manifolds is in a natural way a Poisson algebra/variety/manifold. We first give the algebraic construction and we make
precise what we mean by a natural way (Proposition 2.1); it yields by a direct
translation the corresponding construction for Poisson varieties, where we give an
explicit formula for the product structure in terms of the Poisson matrix (Proposition 2.4). Finally, we give a purely geometric construction, in terms of bivector
fields, for the product of two Poisson manifolds (Proposition 2.5).
2.1.1 The Tensor Product of Poisson Algebras
Let (A1 , 1 ) and (A2 , 2 ) be two commutative associative algebras over F, with unit.
Their tensor product A1 A2 is itself a commutative associative algebra with unit,
with respect to the Fbilinear product, which is defined for F1 , G1 A1 and for
F2 , G2 A2 by:
(F1 F2 ) (G1 G2 ) := (F1 G1 ) (F2 G2 ) .
(2.1)
In this formula, and in the formulas which follow, we use the obvious abbreviations
F1 G1 for F1 1 G1 and similarly for F2 G2 . The algebras A1 and A2 are naturally
identified with subalgebras of A1 A2 via the inclusion maps ji : Ai A1 A2
(i = 1, 2), given by j1 (F) = F 1 for every F A1 , and similarly j2 (F) = 1 F
for every F A2 . Also, to a derivation V of A1 we can associate in a natural way a
derivation V of A1 A2 (with values in A1 A2 ), such that V [j1 (F)] = j1 (V [F]),
for every F A1 . It is defined by
V : A1 A2 A1 A2
F1 F2 V [F1 ] F2 .
(2.2)
Notice that V [j2 (F)] = 0, for every F A2 . To see that V is indeed a derivation of
A1 A2 , use the fact that V is a derivation of A1 , combined with (2.1). Similarly, to
of A1 A2 ,
a derivation W of A2 , we can associate in a natural way a derivation W
[j2 (F)] = j2 (W
[F]), for every F A2 .
such that W
By extending this construction to the case of skewsymmetric biderivations, we
can construct a skewsymmetric biderivation of A1 A2 (with values in A1 A2 ),
starting from a skewsymmetric biderivation of A1 (or of A2 ). We use this in the
following proposition to construct a Poisson structure on A1 A2 , starting from a
Poisson structure on A1 and a Poisson structure on A2 .
2.1 Tensor Products and Products
39
Proposition 2.1. Let (Ai , i , { , }i ) be two Poisson algebras over F, where i = 1, 2.
The tensor product A1 A2 admits a unique Poisson bracket { , }, making the
canonical inclusions ji : Ai A1 A2 into Poisson morphisms with Poissoncommuting images, {j1 (A1 ), j2 (A2 )} = 0. For Fi , Gi Ai , this Poisson bracket is
given by
{F1 F2 , G1 G2 } = {F1 , G1 }1 (F2 G2 ) + (F1 G1 ) {F2 , G2 }2 .
(2.3)
For every Hamiltonian derivation XH of A1 , with H A1 , X
H is a Hamiltonian
derivation of A1 A2 , more precisely,
X
H = XH1 ,
and similarly for the Hamiltonian derivations of A2 .
We call (A1 A2 , , { , }) the Poisson tensor product of A1 and A2 , while { , } is
called the product Poisson bracket.
Proof. Let 1 : A1 A1 A1 be a skewsymmetric biderivation of A1 . Then it can
be verified, as in the case of derivations, that the skewsymmetric bilinear map 1 ,
defined by
1 : (A1 A2 ) (A1 A2 ) A1 A2
(F1 F2 , G1 G2 )
1 [F1 , G1 ] (F2 G2 )
(2.4)
is a biderivation of A1 A2 . If 1 satisfies moreover the Jacobi identity, so that 1
is a Poisson bracket on A1 , then 1 is a Poisson bracket on A1 A2 , since
1 1 [F1 F2 , G1 G2 ], H1 H2 = 1 1 [F1 , G1 ], H1 (F2 G2 H2 ) .
Notice also that 1 has the following properties:
1 [F1 1, G1 1] = 1 [F1 , G1 ] 1 ,
1 [F1 1, 1 G2 ] = 0 ,
1 [1 F2 , 1 G2 ] = 0 ,
for every Fi , Gi Ai , (i = 1, 2). The first formula tells us that j1 : A1 A1 A2
is a morphism of Poisson algebras, when we take 1 as a Poisson bracket on A1
A2 ; according to the second formula, j1 and j2 have Poissoncommuting images;
however, the third formula indicates that j2 : A2 A1 A2 is only a morphism of
Poisson algebras when A2 is equipped with the trivial Poisson bracket.
However, letting 2 denote a Poisson bracket on A2 and taking the sum := 1 +
2 , we obtain a skewsymmetric biderivation of A1 A2 , and the above formulas
lead to
40
2 Poisson Structures: Basic Constructions
[F1 1, G1 1] = 1 [F1 , G1 ] 1 ,
[F1 1, 1 G2 ] = 0 ,
[1 F2 , 1 G2 ] = 1 2 [F2 , G2 ] .
By the biderivation property, these formulas can be summarized in the single formula
[F1 F2 , G1 G2 ] = 1 [F1 , G1 ] (F2 G2 ) + (F1 G1 ) 2 [F2 , G2 ]
(2.5)
for every Fi , Gi Ai , (i = 1, 2). It follows that satisfies the Jacobi identity, because (2.5), combined with the fact that is a biderivation, implies that
[ [F1 F2 , G1 G2 ], H1 H2 ]+ (F, G, H)
= 1 [1 [F1 , G1 ], H1 ] (F2 G2 H2 ) + (F1 G1 H1 ) 2 [2 [F2 , G2 ], H2 ]+ (F, G, H) .
Therefore, is a Poisson bracket on A1 A2 , both maps ji : Ai A1 A2 are
morphisms of Poisson algebras and they have Poissoncommuting images. Notice
that in order to establish formula (2.5) we have used only that is a biderivation
of A1 A2 , such that canonical inclusions ji : Ai A1 A2 are Lie algebra homomorphisms and have Poissoncommuting images, which yields uniqueness of a
Poisson bracket on A1 A2 with the stated properties, and it provides us with an
explicit formula for it. In the bracket notation, (2.5) is (2.3).
Finally, let us show that the linear map V V sends Hamiltonian derivations
of A1 to Hamiltonian derivations of A1 A2 . To do this, it suffices to show that
for every H A1 , the derivation X
H is the Hamiltonian derivation of A1 A2 ,
associated to H 1. Let F1 A1 and F2 A2 be arbitrary elements. According
to (2.2) and (2.5)
X
H [F1 F2 ] = XH [F1 ] F2 = 1 [F1 , H] F2 = [F1 F2 , H 1]
= XH1 [F1 F2 ] ,
so that X
H = XH1 , as was to be shown.
Given two affine varieties M1 and M2 , their product M1 M2 is an affine variety,
with algebra of functions F (M1 M2 ) F (M1 ) F (M2 ). Under the latter isomorphism, the natural projection maps pi : M1 M2 Mi , with i = 1, 2, are dual
to the inclusion maps ji : F (Mi ) F (M1 ) F (M2 ); specifically, the following
triangle of algebra homomorphisms is commutative:
pi
F (Mi )
F (M1 M2 )
ji
F (M1 ) F (M2 )
Translated in the language of varieties, Proposition 2.1 yields the following result.
2.1 Tensor Products and Products
41
Fig. 2.1 On the product of two affine varieties M1 and M2 , functions which are constant on the
fibers of the canonical projection map p1 : M1 M2 M1 are of the form p1 F = F p1 , for some
F F (M1 ). Every vector field on M1 M2 which annihilates all these functions is tangent to the
fibers of p1 .
Proposition 2.2. Let (Mi , i ) be two affine Poisson varieties, where i = 1, 2. The
product variety M1 M2 has a unique Poisson structure = { , } such that the
projection maps pi : M1 M2 Mi (i = 1, 2) are Poisson maps and such that
{p1 F1 , p2 F2 } = 0 ,
(2.6)
for all functions F1 F (M1 ) and F2 F (M2 ). Each Hamiltonian vector field on M1
(or M2 ) extends to a Hamiltonian vector field on M1 M2 , tangent to the fibers of
p2 (or p1 ).
The Poisson variety (M1 M2 , ) is called the product Poisson variety of (M1 , 1 )
and (M2 , 2 ), while is called its product Poisson structure.
Remark 2.3. For i = 1, 2, the fibers of the projection maps pi : M1 M2 Mi have
the following properties (see Fig. 2.1):
1.
2.
A function F F (M1 M2 ) is of the form pi Fi for some Fi F (Mi ) if and
only if it is constant on all the fibers of pi ;
A vector field V on M1 M2 satisfies V [pi Fi ] = 0 for every function Fi
F (Mi ) if and only if it is tangent to all the fibers of pi .
It follows that the condition given by (2.6) can be rephrased in either of the following
equivalent ways.
(i)
(ii)
(iii)
Every function which is constant on all the fibers of p1 is in involution with
every function which is constant on all the fibers of p2 ;
The Hamiltonian vector field of every function which is constant on all the
fibers of p1 is tangent to all the fibers of p2 ;
The Hamiltonian vector field of every function which is constant on all the
fibers of p2 is tangent to all the fibers of p1 .
To finish this section, we describe the Poisson matrix of the product Poisson structure on M1 M2 , when M1 and M2 are affine Poisson varieties. We show that, in
42
2 Poisson Structures: Basic Constructions
terms of natural generators of F (M1 M2 ), coming from generators of F (M1 ) and
F (M2 ), the Poisson matrix takes a block diagonal form.
Proposition 2.4. Suppose that (M1 , { , }1 ) and (M2 , { , }2 ) are affine Poisson va
rieties, with Poisson matrices X = (xi j )di,1j=1 = ( xi , x j 1 )di,1j=1 , respectively Y =
(yi j )di,2j=1 = ( yi , y j 2 )di,2j=1 , with respect to generators x1 , . . . , xd1 for F (M1 ), respectively y1 , . . . , yd2 for F (M2 ). The Poisson matrix for the product Poisson structure { , } on M1 M2 takes, in terms of the d1 +d2 generators p1 x1 , . . . , p1 xd1 , p2 y1 ,
. . . , p2 yd2 for F (M1 M2 ), the form
p1 xi j
0
d1
i, j=1
0
p2 yi j
d2
(2.7)
i, j=1
Proof. Since p1 is a Poisson map we have that
p1 xi , p1 x j = p1 xi , x j 1 = p1 xi j ,
which yields the first diagonal block of the Poisson matrix of { , }. Similarly, the
fact that p2is a Poisson map yields the second diagonal block. The other entries
p1 xi , p2 y j of the Poisson matrix are zero, as follows from (2.6).
By a slight abuse of notation, one usually simply writes
p1 xi and y j for p2 y j .
xi for
X 0
and the product PoisThen the Poisson matrix (2.7) takes the simple form
0Y
son structure = { , } can be explicitly written in the following form
1i< jd1
xi j
+ yi j
.
xi x j 1i<
yj
i
jd2
2.1.2 The Product of Poisson Manifolds
The product of two manifolds M1 and M2 has a natural structure of a manifold,
whose construction we briefly recall. First, as a topological space, M1 M2 is endowed with the product topology. Second, an atlas of M1 M2 is constructed from
an atlas of M1 and of M2 : let (Ui , x(i) )iI be coordinate charts covering M1 and let
(V j , y( j) ) jJ be coordinate charts covering M2 , then (Ui V j )(i, j)IJ is an open covering of M1 M2 and the homeomorphisms
(x(i) , y( j) ) : Ui V j x(i) (Ui ) y( j) (V j ) Fd1 +d2
define compatible coordinate charts, hence endow M1 M2 with a manifold structure; in this formula, d1 and d2 are the dimensions of M1 and M2 respectively; as
before, F = R or F = C, depending on whether the manifolds which are considered
are real or complex.
2.1 Tensor Products and Products
43
Fig. 2.2 A vector field on M1 can be naturally lifted to a vector field on M1 M2 , tangent to the
fibers of p2 : M1 M2 M2 .
As in the case of varieties, we denote by pi : M1 M2 Mi the canonical projection maps (i = 1, 2). For m1 M1 and m2 M2 , we will also use the inclusion
maps m2 : M1 M1 M2 and
m1 : M2 M1 M2 , defined by m2 (x) := (x, m2 )
and
m1 (y) := (m1 , y), for x M1 and y M2 .
To a vector field on an open subset of M1 (or of M2 ), we can associate a
vector field on an open subset of M1 M2 by using the tangent maps T m2 (respectively T m2 ). Namely, to a vector field V X1 (U1 ), where U1 is an open
subset of M1 , we associate the vector field V X1 (U1 M2 ), whose value at
m = (m1 , m2 ) U1 M2 is by definition
Vm := Tm1 m2 (Vm1 ) Tm (M1 M2 ) .
This means that for every F F (U1 ) and G F (U2 ), where U2 is an open subset
of M2 ,
V[F p1 ] = V [F] p1
and
V[G p2 ] = 0 ,
(2.8)
since for all m = (m1 , m2 ) U1 U2 ,
V[F p1 ](m) = Vm (F p1 ) = Tm1 m2 (Vm1 )(F p1 )
= Vm1 (F p1 m2 ) = Vm1 (F) = V [F](p1 (m)) ,
and similarly for the second equality in (2.8). These equalities can be used to
write V in terms of local coordinates, showing in particular that V is a vector field
(i.e., it is smooth/holomorphic). Similarly, every vector field W on an open subset
on M1 U2 . See Fig. 2.2.
U2 of M2 leads to a vector field W
By extending this construction to the case of bivector fields, we can construct
a bivector field on open subsets of M1 M2 , starting from a bivector field defined
on open subsets of M1 (or M2 ). We will use this in the following proposition to
construct a Poisson structure on M1 M2 , starting from a Poisson structure on M1
and a Poisson structure on M2 .
44
2 Poisson Structures: Basic Constructions
Proposition 2.5. Let (M1 , 1 ) and (M2 , 2 ) be two Poisson manifolds. The product
manifold M1 M2 has a unique Poisson structure { , } = , such that the projection
maps pi : M1 M2 Mi (i = 1, 2) are Poisson maps and such that, for all open
subsets Ui Mi and for every Fi F (Ui ) with i = 1, 2, the functions F1 p1 and
F2 p2 are in involution (on U1 U2 ),
{F1 p1 , F2 p2 } = 0 .
(2.9)
The Poisson manifold (M1 M2 , ) is called the product Poisson manifold of M1
and M2 , while is called its product Poisson structure.
Proof. We have seen how we can associate in a natural way to a vector field on M1
a vector field on M1 M2 . In order to generalize this to bivector fields, we use that
every Flinear map : V W between vector spaces V and W , leads to a linear
map 2 : 2V 2W . As recalled in Appendix A (Section A.2) it is defined for
v1 , v2 V by
2 (v1 v2 ) := (v1 ) (v2 ) .
Applied to the linear map
Tm1 m2 : Tm1 M1 Tm (M1 M2 )
we can define a bivector field 1 on M1 M2 by its value at m = (m1 , m2 ):
1m := 2 (Tm1 m2 )(1 )m1 2 Tm (M1 M2 ) .
For open subsets U1 M1 and U2 M2 and for functions F1 , G1 F (U1 ) and
F2 , G2 F (U2 ) we have, as in (2.8), that
1 [F1 p1 , G1 p1 ] = 1 [F1 , G1 ] p1 ,
1 [F1 p1 , G2 p2 ] = 0 ,
1 [F2 p2 , G2 p2 ] = 0 .
It follows that the bivector field := 1 + 2 , where 2 is constructed in the same
way as 1 , but using Tm2
m1 instead of Tm1 m2 , has the following properties:
[F1 p1 , G1 p1 ] = 1 [F1 , G1 ] p1 ,
[F1 p1 , G2 p2 ] = 0 ,
(2.10)
[F2 p2 , G2 p2 ] = 2 [F2 , G2 ] p2 .
Since every point in M1 M2 admits local coordinates of the form xi p1 and y j
p2 , where xi and y j are coordinate functions on M1 , respectively on M2 , it follows
from (2.10) that every point admits local coordinates in which satisfies the Jacobi
identity. In view of (iv) in Proposition 1.16, this implies that is a Poisson structure.
2.1 Tensor Products and Products
45
Moreover, the first and third equations in (2.10) express that p1 and p2 are Poisson
maps, while the second one says that satisfies (2.9).
If a Poisson structure on M1 M2 satisfies the requirements of Proposition 2.5,
then it satisfies the equations (2.10), from which one can obtain explicit formulas for it, in local coordinates, which provides uniqueness of such a Poisson structure. In order to show how these explicit formulas are obtained from (2.10), let
(U, x) and (V, y) be coordinate charts of M1 and M2 , where x = (x1 , . . . , xd1 ) and
y = (y1 , . . . , yd2 ). By a slight abuse of notation, we write the corresponding coordinates on U V as x1 , . . . , xd1 , y1 , . . . , yd2 . According to (1.26) and (2.10), is given
in terms of these coordinates by
[xi , x j ]
1 [xi , x j ] p1
1i< jd1
+ [yi , y j ]
xi x j 1i<
yj
i
jd2
1i< jd1
+ 2 [yi , y j ] p2
,
xi x j 1i< jd2
yi y j
which is written, in the bracket notation, as
{ , } =
1i< jd1
xi , x j
1
yi , y j 2
,
xi x j 1i< jd2
yi y j
where xi , x j 1 and yi , y j 2 are taken as functions on U V .
Remark 2.6. Note that for all open subsets U1 M1 and U2 M2 , the restriction
of the Poisson structure { , } on M1 M2 to F (U1 ) F (U2 ) F (U1 U2 ) is
precisely equal to the product Poisson bracket, introduced in the previous section,
of the Poisson algebras F (U1 ) and F (U2 ).
Remark 2.7. Remark 2.3 applies also to manifolds, upon replacing the functions
and vector fields which appear in that remark by arbitrary local functions and vector
fields, i.e., functions and vector fields which are defined in the neighborhood of a
point.
To finish this section, we give a useful formula for computing the product Poisson
bracket. As above, let (M1 , 1 ) and (M2 , 2 ) be Poisson manifolds and let denote
the product Poisson structure on M1 M2 . Let F, G be functions on a nonempty
open subset U of M1 M2 and let (m1 , m2 ) U. By definition of , we have that
(m1 ,m2 ) = 1 (m1 ,m2 ) + 2 (m1 ,m2 )
= 2 (Tm1 m2 ) (1 )m1 + 2 Tm2
m1 (2 )m2 ,
(2.11)
so that
(m1 ,m2 ) (F, G) = (1 )m1 (F m2 , G m2 ) + (2 )m2 (F
m1 , G
m1 ) ,
which yields, in the bracket notation, the following formula for the product bracket
46
2 Poisson Structures: Basic Constructions
{F, G} (m1 , m2 ) = {F m2 , G m2 }1 (m1 ) + F
m1 , G
m1 2 (m2 ) ,
(2.12)
valid for all F, G F (U), where U is an open subset of M1 M2 , which contains (m1 , m2 ).
2.2 Poisson Ideals and Poisson Submanifolds
The simplest subobjects of Poisson manifolds are Poisson submanifolds, which are
submanifolds, equipped with a Poisson structure, with respect to which the inclusion map is a Poisson map. For a submanifold, several conditions will be given,
which are equivalent to the fact that it is a Poisson submanifold, including the well
known condition that all Hamiltonian vector fields (of locally defined functions) be
tangent to the submanifold. The algebraic condition, which corresponds to it, is that
the ideal of functions which vanish on the submanifold, is a Poisson ideal. In general algebraic terms, for an ideal I of a Poisson algebra (A , , { , }), one has the
following fact: A /I has a Poisson bracket which makes the canonical projection
into a morphism of Poisson algebras if and only if I is a Poisson ideal; if such a
Poisson bracket on A /I exists, then it is unique.
We give in Section 2.2.1 the algebraic construction, and we reformulate the result
in geometrical terms for subvarieties of affine Poisson varieties. In Section 2.2.2, we
give a purely geometrical construction for (immersed or embedded) submanifolds of
a Poisson manifold. For the latter, we also relate the Poisson submanifold condition
to the intersection of the submanifold with the symplectic leaves.
2.2.1 Poisson Ideals and Poisson Subvarieties
Let (A , ) be a commutative associative algebra and let I be an ideal of A . The
quotient A /I inherits from A the structure of an associative algebra. We denote
the quotient map by p : A A /I and we use, for F A , the convenient notation
F := p(F). For a derivation V of A , the following conditions are equivalent:
(i)
(ii)
There exists a derivation V0 of A /I such that V [F] = V0 [F], for every
F A;
V maps I to itself: for every F I , V [F] I .
The proof (i) (ii) is immediate; for the other implication, we define V0 on elements F A /I by V0 [F] := V [F], which is welldefined in view of (ii), and which
is then easily shown to be a derivation of A /I .
Similarly, if A comes equipped with a Poisson bracket, i.e., (A , , { , }) is a
Poisson algebra, it is a natural question to ask what are the conditions on the ideal
I which imply that A /I carries a Poisson bracket { , }0 , which comes from
2.2 Poisson Ideals and Poisson Submanifolds
47
{ , }, i.e., such that p is a Poisson morphism. The fact that p is surjective implies
that, if such a bracket exists, then it is unique.
Suppose that such a Poisson bracket { , }0 exists. The morphism property
F, G
0
= {F, G}
(2.13)
implies that {F, G} = 0 for every F I and G A , since F = 0 if F I . This
means that I is a Lie ideal of (A , { , }), i.e., {I , A } I . Thus I is an ideal for
both multiplications of the Poisson algebra: one calls such an ideal a Poisson ideal.
Conversely, suppose that I is a Poisson ideal of A . Then (2.13) can be taken as
the definition of a bilinear map { , }0 from A /I into itself. Since it is welldefined,
F, G 0 , H 0 = {{F, G} , H} ,
for all elements F, G and H of A , so that { , }0 defines a Lie bracket on A /I .
Since p is a morphism for both algebra structures on A , { , }0 is also a biderivation.
Thus, { , }0 is a Poisson bracket on A which makes p into a Poisson morphism.
The condition {F, G} I for all F I , G A , rewritten as XG (F) I
means that every Hamiltonian derivation of A maps I to itself, and rewritten as
XF (G) I means that every Hamiltonian derivation of F I maps A to I .
Summarizing, the following proposition holds.
Proposition 2.8. Let (A , , { , }) be a Poisson algebra and let I be an ideal
of (A , ). The following conditions are equivalent:
(i)
(ii)
(iii)
(iv)
There exists a Poisson bracket { , }0 on A /I such that p : A A /I is
a Poisson morphism;
I is a Poisson ideal of (A , , { , });
For every F A , the Hamiltonian derivation XF preserves I ;
For every F I , the Hamiltonian derivation XF is I valued.
Moreover, if one of these conditions holds, then the Poisson bracket { , }0 on A /I ,
satisfying (i), is unique and it is defined by
F, G 0 := {F, G} ,
for all F, G A .
We now turn to the case of affine Poisson varieties, which will bring us to the notion
of a Poisson subvariety. We first recall that if N and M are affine varieties which
belong to the same affine space Fd , then N is said to be an affine subvariety of M if
N M. Then N is the zero locus of a prime ideal IN in F (M) and F (N) is in a
natural way isomorphic to F (M)/IN , with which it will in the sequel be identified.
We denote by : F (M) F (N) the canonical projection, which we think of as a
restriction map.
48
2 Poisson Structures: Basic Constructions
Translating the equivalent algebraic conditions, given at the beginning of this section, to the case of varieties, we have that for a derivation V of F (M), the following
conditions are equivalent:
(i)
There exists a derivation V0 of F (N) such that (V [F]) = V0 [ (F)], for every
F F (M), which can be expressed as the commutativity of the following
diagram:
V
F (M)
F (M)
F (N)
(ii)
V0
F (N)
V maps the prime ideal IN of all functions which vanish on N to itself: for
every F IN , V [F] IN .
As we will see in the next section, these two equivalent conditions on V correspond
in the context of manifolds to the fact that V , viewed as a vector field on M, is
tangent to N, and V0 is the restriction of V to N. In the case of varieties, we therefore
also think of V0 as being the restriction of V to N and, by a slight abuse of language,
we say that V is tangent to N (at points of N), if V satisfies V [IN ] IN .
Definition 2.9. Let (M, { , }) be a Poisson variety and let N M be a subvariety.
We say that N is a Poisson subvariety of M if there exists a Poisson structure on N
which turns the inclusion map : N M into a Poisson map.
This means that, if N is a subvariety of M, and IN F (M) denotes the prime
ideal of all functions on M which vanish on N, then N is a Poisson subvariety of
(M, { , }) if and only if F (M)/IN has a Poisson bracket such that the quotient
map : F (M) F (M)/IN is a morphism of Poisson algebras. It leads, for affine
Poisson varieties, to the following geometric analog of Proposition 2.8.
Proposition 2.10. Let (M, { , }) be an affine Poisson variety and let N be a subvariety of M, defined by a prime ideal IN F (M). The following are equivalent:
(i)
(ii)
(iii)
(iv)
N is a Poisson subvariety of (M, { , });
IN is a Poisson ideal of F (M);
For every F F (M), the Hamiltonian vector field XF is tangent to N;
For every F F (M) which vanishes on N, the restriction of the Hamiltonian
vector field XF to N is zero.
Moreover, if one of these conditions holds, then the Poisson structure on N which
makes the inclusion map into a Poisson map is unique.
2.2 Poisson Ideals and Poisson Submanifolds
49
2.2.2 Poisson Submanifolds
Another approach is needed in the case of submanifolds of a Poisson manifold, as
submanifolds are, in general, not defined by ideals. Moreover, important examples
of submanifolds in Poisson geometry are not embedded submanifolds, but immersed
submanifolds; even locally they are, in general, not defined by an ideal. Think, for
example, of the leaves of the symplectic foliation of a Poisson manifold (see Section 1.3.4).
First, let us recall the basic vocabulary. Suppose that : N0 M is a map, where
both M and N0 are manifolds. If for every point n0 N0 the tangent map Tn0 :
Tn0 N0 T(n0 ) M is injective, then is said to be an immersion. If the immersion is
injective one says that (N0 , ) is an immersed submanifold, or submanifold for short;
one often refers to the subset N := (N0 ) of M as being an immersed submanifold,
keeping in mind that N comes with its own differential structure and topology. Since
Tn0 is injective we can, by the implicit function theorem, find a coordinate chart
(V, y = (y1 , . . . , ys )) of N0 , containing n0 , and a coordinate chart (U, x = (x1 , . . . , xd ))
of M, containing (n0 ), with (V ) U, and such that is given in terms of these
coordinates by
(y1 , . . . , ys ) (x1 , . . . , xd ) = (y1 , . . . , ys , 0, . . . , 0) ,
where s is the dimension of N0 and d is the dimension of M. The coordinate chart
(U, x) for M is then said to be adapted to N at (n0 ).
In the special case in which the topology on N which comes from N0 coincides
with the induced topology from M, i.e., is a homeomorphism between N0 and
N (equipped with the induced topology from M), one says that N is an embedded
submanifold. Open subsets of N are then restrictions to N of open subsets of M and
every coordinate chart for N, can be extended to a coordinate chart of M, adapted
to N.
When N = (N0 ) is a submanifold of M, then for n := (n0 ) N, the subspace
Tn0 (Tn0 N0 ) of Tn M is denoted by Tn N and is called the tangent space to N at n.
A vector field V , defined on an open subset U M, is said to be tangent to N if
Vn Tn N for every n U N.
Definition 2.11. Let (M, ) be a Poisson manifold. An (immersed or embedded)
submanifold (N0 , ) of M is called a Poisson submanifold if there exists on N0 a
Poisson structure N such that the inclusion map is a Poisson map. Viewing N :=
(N0 ) as a submanifold of M, we refer to N as a Poisson structure on N and we say
that N is an (immersed or embedded) Poisson submanifold of M.
We give in the following proposition a characterization of Poisson submanifolds
in terms of (locally defined) Hamiltonian vector fields, as well as in terms of the
symplectic leaves of the ambient Poisson manifold.
Proposition 2.12. Let (M, ) be a Poisson manifold. For an (immersed or embedded) submanifold N = (N0 ) of M, the following conditions are equivalent:
50
(i)
(ii)
(iii)
(iv)
2 Poisson Structures: Basic Constructions
N is a Poisson submanifold of (M, );
For every open subset U M and for every F F (U), the Hamiltonian
vector field XF is tangent to N;
For every n N, the intersection of N with the symplectic leaf Sn (M) contains a neighborhood of n in Sn (M);
For every n N, the bivector n belongs to 2 Tn N.
Moreover, if one of these conditions holds, then the Poisson structure on N, which
makes the inclusion map into a Poisson map, is unique.
Proof. (i) = (ii). Let n := (n0 ) N, where n0 N0 is an arbitrary point. Let
(U, x) be a coordinate chart of M, adapted to N at n. In terms of the coordinates
x = (x1 , . . . , xd ), where d is the dimension of M, the Hamiltonian vector field of
F F (U) is given, according to (1.33), by
s
XF = {xi , F}
i=1
+ {x j , F}
,
xi j=s+1
xj
where s is the dimension of N0 and { , } := . Let us
structure
denote
the Poisson
on
N by { , }0 = 0 . Since is a Poisson morphism, x j , F (n) = x j , F 0 (n0 ),
which is zero when s + 1 j d, since x j = 0, for such j. It follows that, for
n N U,
s
s
(XF )n = {xi , F} (n)
= {xi , F} (n) Tn0
,
xi n i=1
xi n0
i=1
so that (XF )(n0 ) Tn0 Tn0 N0 = Tn N, for all n0 N0 , which means that XF is
tangent to N. This shows that all (locally defined) Hamiltonian vector fields are
tangent to N.
(ii) = (iii). Let n N and consider the symplectic leaf Sn (M) which contains n (see Section 1.3.4). Since is an immersion, there exists an open subset V of
N0 which contains n0 , where n = (n0 ), and an open subset U of M, such that (V ) is
a closed, embedded submanifold of U. We also consider the symplectic leaf Sn (U),
which is an open subset of Sn (M) (recall that it is, in general, strictly smaller than
Sn (M) U). We claim that Sn (U) (V ). To show this, it suffices to show that
every Hamiltonian path in U which starts in (V ), stays in (V ). Consider a Hamiltonian path : I U, where I is an open neighborhood of 0 in F (which is R or C),
with (0) (V ). It is the integral curve of a Hamiltonian vector field XF , with
F F (U
), which is tangent to (V ), at all points of (V ), where U
is an open
subset of U, containing (I). Suppose that (I) is not entirely contained in (V ).
Consider
I
:= {t I  (t) (V )} ,
which is an open subset of I, since is an integral curve of a vector field which is
tangent to (V ). Let I0
denote the connected component of I
which contains 0. Let t
be an arbitrary element of the (topological) boundary of I0
and let n
:= (t
). Then
2.2 Poisson Ideals and Poisson Submanifolds
51
n
(V ), since (V ) is a closed subset of U. Since is an integral curve, passing
through n
, of a vector field which is tangent to (V ), we have that (t
) (V ) for t
in a neighborhood of t
, which contradicts the fact that t
belongs to the boundary
of I0
. Therefore, all Hamiltonian paths : I U, defined on a neighborhood I of 0,
with (0) (V ), stay in (V ) and (V ) Sn (U). Since the latter is an open subset
of Sn (M), property (iii) follows.
(iii) = (i). Let n0 N0 be an arbitrary point and let n := (n0 ). Since is an
immersion,
Tn0 : Tn0 N0 Tn N Tn M
is an isomorphism, while (iii) implies that Tn Sn (M) Tn N. It follows that there
exists a unique bivector (N )n0 2 Tn0 N0 such that
2 (Tn0 )(N )n0 = n ,
(2.14)
as the latter belongs to 2 Tn Sn (M). We claim that the map
N : N0 2 T N0
n0 (N )n0
(2.15)
is a Poisson structure on N0 , such that is a Poisson map. To see that N is a
(smooth/holomorphic) bivector field on N0 , let n0 be an arbitrary point and let
F0 , G0 be arbitrary functions, defined on a neighborhood of n0 in N0 . There exists a
neighborhood U of n := (n0 ), and functions F, G F (U), such that F = F0 and
G = G0 on some neighborhood V of n0 in N0 . For every n
= (n
0 ) U (V ) we
have that
N [F0 , G0 ](n
0 ) = N [F , G ](n
0 )
= (N )n
dn
(F ), dn
(G )
0
0
0
= n
(dn
F, dn
G) = [F, G]((n
0 )) ,
where we have used the definition (2.14) of N to obtain the last line. Since n
0
[F, G]((n
0 )) is clearly a smooth/holomorphic function on V N0 , it follows that
N is a (smooth/holomorphic) bivector field on N0 . The same computation, with
three functions F0 , G0 and H0 , defined on a neighborhood of n0 in N0 , yields
N [N [F0 , G0 ] , H0 ] (n
0 ) = [ [F, G] , H] ((n
0 )) ,
for all n
0 in a neighborhood V of n0 in N0 . It leads to the Jacobi identity for N , so
that N is a Poisson structure on N0 .
According to Proposition 1.19, is a Poisson map; it is clear that N is the unique
Poisson structure on N0 , such that is a Poisson map.
(ii) (iv). As in the first step of the proof, let (U, x) be a coordinate chart
of M, adapted to N at n N. On U we can write as
52
2 Poisson Structures: Basic Constructions
1i< js
i j
d
d
+ i j
,
xi x j i=1 j=s+1 xi x j
where all the coefficients i j are functions, defined on U, with ji = i j . If n
2 Tn N, then i j (n) = 0 for all i, j with 1 i d and s + 1 j d, which is
equivalent to saying that every Hamiltonian vector field on U is tangent to N at n.
This shows that (ii) and (iv) are equivalent.
2.3 Real and Holomorphic Poisson Structures
This section is devoted to three constructions which are intimately related to the real
or holomorphic nature of Poisson structures.
2.3.1 Real Poisson Structures Associated to Holomorphic Poisson
Structures
Our aim is to construct from a holomorphic Poisson structure on a complex manifold, real Poisson structures on the underlying real manifold. Let M be a complex
manifold of (complex) dimension d N , whose underlying real manifold will be
denoted by MR . We fix some notations for the different algebras of functions which
will be considered on M or MR :
(a)
(b)
(c)
(d)
H (M) denotes the Calgebra of all holomorphic functions on M,
C (MR , C) denotes the Calgebra of all Cvalued Rdifferentiable functions
on MR ,
C (MR ) denotes the Ralgebra of all Rvalued Rdifferentiable functions on
MR ,
H (M) denotes the Calgebra of all antiholomorphic functions on M, i.e. the
functions of the form F, with F H (M).
We also use analogous notations for the functions defined on an open subset U of M
(or UR of MR ). We then have
(2.16)
H (U) C (UR , C) C (UR ) + 1C (UR ) ,
where the latter isomorphism consists of writing a Cvalued function as the sum
of its real and its imaginary parts. It is clear that every coordinate chart (U, z)
of M, with z = (z1 , . . . , zd ) leads to a coordinate chart (UR , Z) of MR , with Z =
(x1 , . . . , xd , y1 , . . . , yd ),
where xk , yk C (UR ) are the real smooth functions obtained
by writing zk = xk + 1 yk under the decomposition (2.16).
Let V be a holomorphic vector field on M. According to (B.8), in a coordinate
chart (U, z), we can write
2.3 Real and Holomorphic Poisson Structures
V =
V [zk ]
k=1
53
1 d
.
= V [zk ]
1
zk
2 k=1
xk
yk
(2.17)
Decomposing V [zk ] into its real and imaginary parts, leads to the definition of
two derivations of C (UR ), the real and imaginary parts of V , denoted by (V )
and (V ), by putting:
1 d
(V
[z
,
])
+
(V
[z
])
k
k
2 k=1
xk
yk
1 d
.
(V [zk ])
(V ) := (V [zk ])
2 k=1
xk
yk
(V ) :=
(2.18)
(2.19)
These two derivations of
the Ralgebra C (UR ) extend uniquely to the Calgebra
C (UR , C) C (UR ) + 1C (UR ), by Clinearity; we will denote these extensions
still by (V ) and (V ). Writing a holomorphic function F on U as
F = G + 1 H and using the CauchyRiemann equations
G
H
=
,
xk
yk
G
H
=
,
yk
xk
one verifies by a direct computation that, for all F H (U),
1
(V )[F] = V [F],
2
1
(V )[F] = V [F] .
2
(2.20)
These equations characterize (V ) since (2.18) can be derived from it. Similarly,
the imaginary part (V ) is characterized by the fact that
1
1
V [F], (V )[F] =
(V )[F] =
V [F] ,
2
2
for all F H (U). These characterizations imply that (V ) and (V ) are independent of the choice of the coordinates and therefore, define (smooth) vector fields
on MR .
An analogous construction works in the case of bivector fields. Indeed, let P be a
bivector field of M. According to (1.23), we can write P in a coordinate chart (U, z)
as
zk z
1k<d
1
=
P[zk , z ] xk 1 yk x 1 y .
4 1k<d
P=
P[zk , z ]
As above, decomposing P[zk , z ] into its real and imaginary parts, leads to the definition of two skewsymmetric biderivations (P) and (P) of C (UR ), by putting,
54
2 Poisson Structures: Basic Constructions
1
(P) :=
(P[zk , z ]) xk x yk y
4 1k<d
1
+
(P[zk , z ]) xk y + yk x ,
4 1k<d
and similarly for (P). By definition, (P) and (P) are skewsymmetric biderivations of C (UR ) and they uniquely extend to skewsymmetric biderivations of
C (UR , C), by Clinearity. Then, (P) is, as in (2.20), characterized by the following properties: for all F, G H (U),
1
P[F, G] ,
2
(P)[F, G] = 0 ,
1
(P)[F, G] = P[F, G] .
2
(P)[F, G] =
(2.21)
There is an analogous characterization for the imaginary part (P). Because of
these characterizations, we can conclude, as in the case of vector fields, that (P)
and (P) are welldefined bivector fields of MR . We show in the next proposition
that, if the bivector field P has the additional property of being a Poisson structure,
then (P) is also Poisson structure. One shows similarly that (P) is a Poisson
structure.
Proposition 2.13. If (M, ) is a complex Poisson manifold of dimension d, then the
bivector field ( ) is a Poisson structure on MR . In local coordinates, ( ) is given
by:
1
( ) =
(k ) xk x yk y
4 1k<d
1
+
(
k
4 1k<d
xk y yk x
where zk = xk + 1yk and k := (zk , z ), for 1 k, d.
Proof. According to the characterization of ( ), given above, and since is a
holomorphic Poisson bracket, we have, for every coordinate chart (U, z) and for all
F, G, H H (U),
1
{{F, G} , H} ,
4
( )[( )[F, G], H] = 0 ,
( )[( )[F, G], H] =
( )[( )[F, G], H] = 0 ,
1
( )[( )[F, G], H] = {{F, G} , H} ,
4
2.3 Real and Holomorphic Poisson Structures
55
where we have written { , } for . Since satisfies the Jacobi identity, these equations imply that ( ) satisfies the Jacobi identity, when applied to arbitrary triples
of holomorphic or antiholomorphic functions on U. In particular, for every coordinate chart (UR , Z), the Jacobi identity will be satisfied by ( ) for the coordinate
functions xk and y , because they can be written as Clinear combinations of the
holomorphic and antiholomorphic functions zk and zk . Then, according to Proposition 1.16, ( ) is a bivector field of MR , satisfying the Jacobi identity, i.e., it is a
Poisson structure on MR .
2.3.2 Holomorphic Poisson Structures on Smooth Affine Poisson
Varieties
Let N be an sdimensional complex affine variety, N Cd . It is the zero locus of a
prime ideal I of C[x1 , . . . , xd ] and the algebra of regular functions on N is related
to I by F (N) = C[x1 , . . . , xd ]/I (see Section 1.2.1). Let H1 , . . . , Ht be a system of
generators of I . The set of points n of N where the rank of the Jacobian matrix
Hi
(n)
1it
xj
1 jd
equals d s is a (Zariski) open subset of N, called the smooth part of N, and denoted N sm . The points of N sm are called smooth points of N. By the implicit function
theorem, N admits in the neighborhood of every smooth point the structure of a complex manifold, which is the unique structure for which the functions x1 N , . . . , xd N
are holomorphic. In particular, N sm is a complex manifold. For every point n in N sm
we can choose s functions among x1 N , . . . , xd N , which constitute local coordinates
of N in a neighborhood of n in N sm and the other functions xi N are then expressible
as holomorphic functions in terms of these local coordinates.
Suppose now that N comes equipped with a Poisson structure, i.e., (N, { , }) is
a (complex) affine Poisson variety, N Cd . Since { , } associates to two regular
functions1 on N a regular function on N, we may choose,
forevery i, j, with 1
i, j d, a representative xi j of the Poisson bracket xi N , x j N , i.e., an element of
C[x1 , . . . , xd ] such that xi j N = xi N , x j N . Since all xi j are polynomials, the skewsymmetric biderivation
xi j xi x j
1i< jd
defines a holomorphic bivector field on Cd . We use it to construct a holomorphic
bivector field on N sm . Let n be an arbitrary point of N sm and let F, G be holomorphic
1 Recall that a regular function on N is, by definition, the restriction to N of a polynomial function
on Cd .
56
2 Poisson Structures: Basic Constructions
functions, defined on a neighborhood W of n in N sm . We extend F and G to functions
defined on a neighborhood U of n in M. For n
in a neighborhood of V in
F and G,
sm
N , contained in W U, we define,
{F, G}H (n
) :=
1i< jd
xi j (n
)
F
G
(n )
(n ).
xi
xj
= xi N , x j N on V ,
It defines the unique bivector field on V such that xi N , x j N
H
for all 1 i, j d, hence it is a welldefined holomorphic
bivector field on N sm .
Since { , } satisfies the Jacobi identity for all triples xi N , x j N , xkN , with 1 i <
j < k d, the same holds true for { , }H , which means that the Jacobi identity
holds for { , }H , since in a neighborhood of every point, a subset of x1 , . . . , xd
yields a system of local coordinates. This shows that { , } leads to a holomorphic
Poisson structure { , }H on N sm . In particular, if N is a smooth Poisson variety, so
that N sm = N, the above construction yields a holomorphic Poisson structure on N,
starting from the Poisson bracket on the algebra of regular functions on N.
2.4 Other Constructions
We finish this chapter with a few isolated constructions, which are of a different
nature.
2.4.1 Field Extension
If A is a commutative associative Falgebra and if F is a field extension of F (think
of C, which is a field extension of R), then the tensor product F F A carries a
natural structure of a commutative associative Falgebra,
simply by putting, for all
a, b F and for all F, G A ,
a(b F) := (ab) F ,
(a F)(b G) := (ab) (FG) .
(2.22)
This very general algebraic procedure is known as extension of scalars. In fact,
an FLie algebra structure [ , ] on A leads, in a similar way, to an FLie
algebra
structure on F F A : simply replace the second line in (2.22) by
[a F, b G] := (ab) [F, G] .
(2.23)
Combining (2.22) and (2.23) shows that every FPoisson algebra (A , , { , }), leads
in a natural way to an FPoisson
algebra (F F A , , { , }). For example, it allows
2.4 Other Constructions
57
one to pass from a Poisson bracket on an Falgebra A to a Poisson bracket on
F F A , where F is the algebraic closure of F.
2.4.2 Localization
A classical construction in algebra and in algebraic geometry is to localize a commutative associative algebra A with respect to a multiplicative system S in A (multiplicative means that S, which is a subset of A \ {0}, is closed under multiplication).
When A is without zero divisors, which we assume in the sequel of this section,
the localization A /S may be defined as the smallest algebra which contains the elements of A and the inverses of elements of S. Formally this means that one forms
the quotient of A S by the equivalence relation , defined by
(F, H) (F
, H
) FH
= F
H .
(2.24)
The class of (F, H) is denoted by F/H or by HF and A /S is made into an algebra by
defining the sum and product of such elements in the way one adds and multiplies
fractions. We think of A as a subalgebra of A /S by using the inclusion F F/1,
where the latter element is also simply written as F.
A derivation V of A extends to a unique derivation VS of A /S. To show this, we
first prove that a derivation VS of A /S is entirely determined by its values VS [F] =
V [F] on elements F = F/1 of A . For F A and H S, applying the derivation VS
to the equality F = HF H, we find the familiar formula from calculus,
F
H V [F] F V [H]
VS
=
,
(2.25)
H
H2
showing that VS is indeed determined by its values on A . In order to show that every
derivation V of A extends to a derivation of A /S one takes (2.25) as a definition
of VS . One needs to check that (2.25) is welldefined: to do this, suppose that F1 H2 =
F2 H1 , with Fi A and Hi S. First, notice that the equality F1 H2 = F2 H1 implies
that V [F1 H2 F2 H1 ] = 0, which we write as
V [F1 ] V [F2 ] F2 V [H1 ] F1 V [H2 ]
=
.
H1
H2
H1 H2
(2.26)
The definition (2.25) of VS , combined with (2.26) and with the equality F1 H2 =
F2 H1 , yields
58
2 Poisson Structures: Basic Constructions
VS [F1 /H1 ] VS [F2 /H2 ]
= F1 V [H1 ]/H12 + V [F1 ]/H1 + F2 V [H2 ]/H22 V [F2 ]/H2
= F1 V [H1 ]/H12 + F2 V [H2 ]/H22 + F2 V [H1 ]/(H1 H2 ) F1 V [H2 ]/(H1 H2 )
= (F2 H1 F1 H2 )V [H1 ]/(H12 H2 ) + (F2 H1 F1 H2 )V [H2 ]/(H1 H22 )
= 0.
This shows that VS is welldefined. It is easy to verify that (2.25) and the fact that V
is a derivation, imply that, for all F, F
A and H, H
S, one has that
F F
F
F
F
F
,
(2.27)
VS
= VS
+
VS
HH
H
H
H
H
which says that VS is a derivation of A /S.
In the same way, every skewsymmetric biderivation P of A extends uniquely
to a skewsymmetric biderivation PS of A /S. As above, one obtains that the basic
formulas which allow one to compute PS , are given as follows: if F, G A /1 A
and H, K S, then PS [F, G] = P[F, G], and
P[H, G]
1
1
, G = PS G,
=
,
PS
H
H
H2
1 1
P[H, K]
,
=
.
PS
H K
H 2K2
(2.28)
As above, it is easily checked that PS is a skewsymmetric biderivation of A /S. If P
is a Poisson bracket on A , then PS is a Poisson bracket on A /S. That is the content
of the following proposition.
Proposition 2.14. Let (A , , { , }) be a Poisson algebra without zero divisors and
let S be a multiplicative system of A . The localization algebra A /S carries a unique
Poisson structure { , }S which makes the natural inclusion A A /S into a morphism of Poisson algebras.
Proof. If a Poisson structure { , }S on A/S, for which A A /S is a Poisson morphism, exists, then { , }S is the extension of { , } to A /S, as constructed above.
Therefore, we only need to prove the Jacobi identity for { , }S . Since A /S is generated by all elements F, where F A and 1/H, where H S, we only need to
verify the Jacobi identity for triples consisting of such elements. Let us do this for a
triple (F, 1/H, 1/K), where F A and H, K S. From (2.28) we compute,
{{1/H, 1/K}S , F}S + {{1/K, F}S , 1/H}S + {{F, 1/H}S , 1/K}S
= {H, K} /(HK)2 , F S {K, F} /K 2 , 1/H S {F, H} /H 2 , 1/K S
= ({{H, K} , F} + (H, K, F))/(HK)2 2 {K, H} {K, F} /(H 2 K 3 )
2 {H, K} {HK, F} /(HK)3 2 {F, H} {H, K} /(H 3 K 2 )
2.4 Other Constructions
59
{{H, K} , F} + (H, K, F)
,
(HK)2
which is zero, in view of the Jacobi identity for { , }.
In the case of an affine variety M there are two important cases of localization. The
first one consists in taking for S the set Sm of all functions F F (M) which do not
vanish at a given point m. Then the localization F (M)/Sm is the algebra of rational
functions F/G, with F, G F (M) and G(m) = 0. By the above, every Poisson
structure on M will induce a Poisson structure Sm on this algebra of functions. Of
course, the algebra of all rational functions on M also inherits a Poisson structure
from M, leading to a Poisson algebra which contains all (F (M)/Sm , Sm ) as Poisson
subalgebras.
The second special case of localization consists in taking a nonconstant function G A and considering for S the set SG := {Gn  n N}. Then F (M)/SG is
the algebra of rational functions F/Gn , with F F (M) and n N. Picking an indeterminate we have an algebra isomorphism
F (M)[ ]
F (M)/SG
G 1
n
Fi i
i=0
Fi /Gi .
i=0
It follows that the localization F (M)/SG is the algebra of regular functions on an
affine variety, which we may think of as being obtained by removing the zero locus
G = 0 from M. Such an open subset of M is sometimes called a principal open
subset. By the above, every Poisson structure on M will lead to a Poisson structure
on this open subset, making it into an affine Poisson variety.
2.4.3 Germification
Recall that a function germ at a point m of a manifold M is an equivalence class
of functions, defined on a neighborhood of m, where the equivalence is defined by
F G if there exists an open neighborhood U of m in M, such that FU = GU .
Recall also that we denote the germ of F at m by Fm . Suppose that M (or at least a
neighborhood of m in M) is equipped with a Poisson structure and let Fm and Gm
be two germs at m, represented by functions F and G, which are both defined on
a neighborhood U of m. For every open subset U
which is contained in U, the
restriction of [F, G] to U
depends on the restrictions of F and G to U only, hence
the germ ( [F, G])m of [F, G] at m is welldefined. Thus, a Poisson structure ,
defined on a neighborhood of m M leads to a Poisson bracket m on the algebra
of germs of functions at m. We may now define an equivalence relation on the set
of all Poisson structures, defined in a neighborhood of m by saying that m
if
60
2 Poisson Structures: Basic Constructions
m = m
. It amounts to saying that if F and G are functions which are defined in
a neighborhood of m in M, then [F, G] and
[F, G] agree on some neighborhood
of m in M. A germ of a Poisson structure at m is then an equivalence class of m ; by
a slight abuse of notation we denote the germ of at m by m . In bracket notation,
the germ m is given by
{Fm , Gm }m =
i, j=1
(xi j )m
F
xi
m
G
xj
,
(2.29)
where F and G are arbitrary representatives of Fm and Gm ; also x1 , . . . , xd are local
coordinates, in a neighborhood of m and xi j := xi , x j . Evaluating {Fm , Gm }m at m
yields a skewsymmetric bilinear map Tm M Tm M F, which is precisely the
pointwise biderivation at m, associated to the Poisson structure, which we denoted
earlier by m . Although we have used the same notation m for the germ of the
Poisson structure at m and for the Poisson bivector at m, one should not forget that
the germ at m contains more information than the Poisson bivector at m.
2.5 Exercises
1. Determine the algebra of Casimirs of the Poisson tensor product of two Poisson
algebras.
2. Let A and B be two Poisson algebras and let : A B A be a Poisson
morphism, where the righthand side is equipped with the product bracket. Prove
that
A B := {F A  (F) = 1 F}
is a Poisson subalgebra of A .
3. Prove Proposition 2.14 in the case in which A may have zero divisors (in
this case, the definition (2.24) of the equivalence relation is replaced by (a, s)
(a
, s
) u S such that u(as
a
s) = 0).
4. Reprove Proposition 2.14 by first showing that a biderivation of A /S vanishes,
as soon as it vanishes on A .
5. Let (M1 , 1 ) and (M2 , 2 ) be two Poisson manifolds. Let : M1 M2 be a
Poisson map. Show that if (M1 ) is a submanifold of M2 , then it is a Poisson submanifold.
6. Let (M, ) be a complex Poisson manifold. Show that every linear combination
of the Poisson structures ( ) and ( ), defined in Section 2.3.1, is a Poisson
structure on MR .
7. Let be a Poisson structure on an open subset U of Rd , where d 2, and let F
be a smooth function on U. Show that if has rank at most 2 at every point of U,
then F is a Poisson structure on U.
2.6 Notes
61
8. We assume in this exercise that the reader is familiar with the notion of symplectic manifold (see Section 6.3). Let (M, ) be a symplectic manifold, and denote
by the canonical Poisson structure, associated to . Prove that the open subsets
of M are the only Poisson submanifolds of (M, ).
9. The purpose of this exercise is to establish the existence of nontrivial Poisson
structures on every (smooth real) manifold of dimension at least 2. Throughout the
exercise we consider Rd with its algebra of smooth functions.
a.
b.
c.
d.
e.
Show that, if d 2, then there exists a Poisson structure on Rd which has
constant rank 2 in a neighborhood of 0 and which has rank zero outside some
compact subset of Rd ;
Show that, if d 4, then there exists a Poisson structure on Rd which has
constant rank 4 in a neighborhood of 0 and which has rank at most 2 outside
some compact subset of Rd . Using Exercise 6, conclude that there exists such
a Poisson structure, which vanishes outside some compact subset of Rd ;
Generalizing parts a. and b., show that for every s N, such that 2s d, there
exists a Poisson structure on Rd which has constant rank 2s in a neighborhood
of 0 and which vanishes outside some compact subset of Rd ;
Using a coordinate chart, conclude that every manifold of dimension d 2,
admits a nontrivial Poisson structure; it can be chosen such that its rank, at a
given point, equals a given even integer 2s d;
The following is an open problem: show that for every Poisson structure on
Rd there exists a Poisson structure on Rd which coincides with in a neighborhood of the origin of 0 and which vanishes outside some compact subset
of Rd .
2.6 Notes
The constructions which were given in this chapter are so basic that it is difficult
to give the original references. Most of the geometrical constructions are at least
implicit in Lichnerowiczs seminal paper [126], in which Poisson manifolds (of
constant rank) were introduced. The algebraic formulation of these constructions
follows easily. For the link between real and holomorphic Poisson structures, explained in Section 2.3.1, see LaurentStienonXu [121].
Chapter 3
MultiDerivations and Kahler Forms
The Jacobi identity, which is a key element in the definition of a Poisson bracket on
an algebra A (or a Poisson structure on a manifold M), can be naturally formulated
in terms of the Schouten bracket on the space X (A ) of all skewsymmetric multiderivations of A (or on the space X (M) of all multivector fields on M). In a sense,
the Schouten bracket is dual to the differential on the space of Kahler forms (A )
(or on (M), the space of differential forms on M). See Table 3.1 for an overview
of the algebraic structures which will be explained in detail in this chapter, with
special emphasis on their connections with Poisson structures.
Table 3.1 A summary of the operations which will be considered in this chapter. In the table,
P X p , where p > 0.
X (A ), X (M)
(A ), (M)
Algebraic
multiderivation
Kahler form
Geometric
multivector field
differential form
Wedge product
: X X X
d : +1
Differential
Schouten bracket
[ , ]S
: X X
X1
Lie derivative
LP : X X+p1
LP : p+1
Internal product
P : p
Since we are using mostly the algebraic properties of these operations, our constructions will be mainly algebraic, but they will always be reformulated in geometrical terms. Multiderivations are introduced in Section 3.1 and Kahler forms in
Section 3.2. The Schouten bracket and the closely related notion of (generalized)
Lie derivative will be introduced in Section 3.3.
Unless otherwise stated, F stands throughout this chapter for an arbitrary field of
characteristic zero.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 3, SpringerVerlag Berlin Heidelberg 2013
63
64
3 MultiDerivations and Kahler Forms
3.1 MultiDerivations and Multivector Fields
In this section, we generalize the notion of a derivation and of a skewsymmetric
biderivation, defined in Section 1.1.1, to the notion of a skewsymmetric multiderivation. We assemble all multiderivations in one graded vector space, which
becomes a graded algebra, upon introducing the wedge product. Multivector fields,
which are the geometrical analogs of skewsymmetric multiderivations, will be discussed in Section 3.1.3.
3.1.1 MultiDerivations
Let A be a commutative associative Falgebra and let p 1. A skewsymmetric,
plinear map P HomF ( p A , A ) is called a skewsymmetric pderivation of A
(with values in A ), if P is a derivation in each of its arguments. By skewsymmetry,
this is equivalent to demanding that P be a derivation in its first argument, i.e.,
P(FG, F2 , . . . , Fp ) = F P(G, F2 , . . . , Fp ) + G P(F, F2 , . . . , Fp ) ,
(3.1)
for arbitrary elements F, G, F2 , . . . , Fp of A . We keep using our convention, introduced in Chapter 1, to enclose the list of arguments of a skewsymmetric multiderivation in square brackets; in this notation, (3.1) becomes
P[FG, F2 , . . . , Fp ] = F P[G, F2 , . . . , Fp ] + G P[F, F2 , . . . , Fp ] .
(3.2)
In view of the derivation property (3.2), two skewsymmetric pderivations of A are
equal as soon as they are equal for all ptuples, taken from a system of generators
of A ; in particular, a pderivation of A vanishes as soon as it vanishes on all ptuples, taken from a system of generators of A (a detailed argument of this fact was
given in the case of derivations in the proof of Proposition 1.6). The vector space of
all skewsymmetric pderivations of A is denoted by X p (A ) and we introduce the
graded vector space
X (A ) :=
pN
X p (A )
HomF ( p A , A ) ,
pN
whose elements are called skewsymmetric multiderivations. By convention, the
first term in this sum, X0 (A ), is just A , and X p (A ) := {0} for p < 0. Every X p (A )
has a natural A module structure, where for F A and P X p (A ), the product
FP X p (A ) is defined by setting
(FP)[F1 , . . . , Fp ] := F P[F1 , . . . , Fp ] ,
for all F1 , . . . , Fp A . Thus, X (A ) is actually a graded A module.
3.1 MultiDerivations and Multivector Fields
65
We often think of A as being the algebra of regular functions on an affine variety M; then P is an operator which associates to p regular functions on M a regular
function on M.
3.1.2 Basic Operations on MultiDerivations
There are two natural composition laws on the graded A module X (A ) of skewsymmetric multiderivations of a commutative associative Falgebra A . The first
one, introduced here, is the wedge product. The second one is a graded Lie bracket,
the Schouten bracket, which will be introduced in Section 3.3.
We first recall the notion of a shuffle. For p, q N, a (p, q)shuffle is a permutation of the set {1, . . . , p + q}, such that (1) < < (p) and (p + 1) < <
(p + q). The set of all (p, q)shuffles is denoted by S p,q . For a shuffle S p,q ,
we denote the signature of (as a permutation) by sgn( ). It is also convenient to
/ for p, q N.
define S p,1 := 0/ and S1,q := 0,
p
q
For P X (A ) and Q X (A ), their wedge product P Q X p+q (A ) is the
skewsymmetric (p + q)derivation of A , defined by
(P Q)[F1 , . . . , Fp+q ] :=
S p,q
sgn( ) P[F (1) , . . . , F (p) ] Q[F (p+1) , . . . , F (p+q) ] ,
(3.3)
for all F1 , . . . , Fp+q A . Notice that, in particular, F P = P F = FP when F
X0 (A ) = A . Stated briefly, the wedge product is the skewsymmetrization of the
tensor product (of skewsymmetric multilinear maps). A priori, we have only that
P Q Hom( p+q A , A ), but the reader will easily check that indeed P Q
X p+q (A ), by showing that P Q verifies the derivation property (3.2). The wedge
product has the following properties.
Proposition 3.1. Let A be a commutative associative algebra. The wedge product
: X (A ) X (A ) X (A )
makes X (A ) into an associative graded A algebra. Moreover, it is graded commutative: for P X p (A ) and Q Xq (A ), one has
P Q = (1) pq Q P .
(3.4)
Example 3.2. If V and W are derivations, then V W is the skewsymmetric
biderivation, given by
(V W )[F, G] = V [F] W [G] V [G] W [F] ,
66
3 MultiDerivations and Kahler Forms
for F, G A . More generally, if V1 , . . . , Vk are derivations of A and F1 , . . . , Fk are
elements of A , then
V1 [F1 ] V1 [Fk ]
.. .
(V1 Vk )[F1 , . . . , Fk ] = ...
.
Vk [F1 ] Vk [Fk ]
Each element F of A defines a graded A linear map
F : X (A ) X1 (A ) ,
of degree 1, by taking F as the first element on which the multiderivation is
evaluated: for P X p (A ), the skewsymmetric (p 1)derivation F P is defined by
F P[F2 , . . . , Fp ] := P[F, F2 . . . , Fp ] ,
(3.5)
for all F2 , . . . , Fp A . For a derivation V X1 (A ), this yields F V = V [F]; for
a function G (element of X0 (A )), the above definition should be understood as
F G := 0, since X p (A ) = {0} for p < 0. For every F A , the graded linear map
F is a graded derivation1 of degree 1 of X (A ), i.e.,
F (P Q) = F P Q + (1) p P F Q .
(3.6)
This follows immediately from (3.3), upon using that for every shuffle S p,q , one
has that either (1) = 1 or (p + 1) = 1. Also, each derivation V X1 (A ) defines
a graded Flinear map of degree 0,
LV : X (A ) X (A ) ,
which is called the Lie derivative with respect to V . For P X p (A ), its Lie derivative LV P X p (A ) is defined as follows:
p
LV P[F1 , . . . , Fp ] := V [P[F1 , . . . , Fp ]] P[F1 , . . . , V [Fi ], . . . , Fp ] ,
(3.7)
i=1
for all F1 , . . . , Fp A . It is easy to check that indeed LV P X p (A ) by showing that
LV P, as defined in (3.7), satisfies (3.2). Notice that, in particular, LV F = V [F],
for F A , and LV W = [V , W ], for all V , W X1 (A ); for a skewsymmetric
biderivation P, (3.7) is precisely (1.35). It follows at once from the definition that
the Lie derivative is a graded derivation of degree 0 of X (A ):
LV (P Q) = LV P Q + P LV Q ,
(3.8)
for all P, Q X (A).
1
See Appendix A (in particular, Section A.5) for the basic definitions and properties of graded
derivations and coderivations.
3.1 MultiDerivations and Multivector Fields
67
3.1.3 Multivector Fields
We have already pointed out in Section B.2 that, in the context of manifolds, the
analog of a derivation, respectively skewsymmetric biderivation, is a vector field,
respectively a bivector field. Similarly, multivector fields are the geometrical analogs
of skewsymmetric multiderivations. Since the basic definitions and properties of
multivector fields are a direct generalization of what we have seen for bivector fields
in Section B.2, the explanations and justifications which are given in this section are
kept to a minimum.
Let M be a manifold (a real manifold when F = R, a complex manifold when
F = C) and let p N and m M. Generalizing Definition 1.13, we call a skewsymmetric plinear map
Fm (M) p
m :
F
a skewsymmetric pointwise pderivation at m, if for all functions F, G, H, . . . , K,
defined in a neighborhood of m in M,
m (Fm Gm , Hm , . . . , Km )
= F(m) m (Gm , Hm , . . . , Km ) + G(m) m (Fm , Hm , . . . , Km ) .
The vector space of all pointwise, skewsymmetric pderivations at m is denoted
by p Tm M, since, in a coordinate chart (U, x) around m, a basis of this vector space
is given by the pointwise pderivations,
,
x i1 m
x i2 m
xi p m
where 1 i1 < i2 < < i p dim M; in this formula, the wedge product is the
wedge product of (linear) maps, as in the previous section (see for example Eq. 3.3).
A map, which assigns to every point m M an element Pm of p Tm M is called a
pvector field on M if for every open subset U M and for all functions F1 , . . . , Fp
F (U), one has that P[F1 , . . . , Fp ] F (U), where P[F1 , . . . , Fp ] is the function on U,
defined for all m U by P[F1 , . . . , Fp ](m) := Pm ((F1 )m , . . . , (Fp )m ), which we also
write as Pm [F1 , . . . , Fp ]. Thus, on every coordinate chart (U, x), the pvector field P
becomes a skewsymmetric pderivation of the algebra of functions F (U) and P
admits the local coordinate expression
P=
1i1 <<i p d
P[xi1 , . . . , xi p ]
,
x i1
xi p
(3.9)
where d := dim M. It follows that a pvector field is a tensorial object: for m M,
the value of P[F1 , . . . , Fp ](m) depends on the differentials dm F1 , . . . , dm Fp only. This
is a useful fact for constructing multivector fields, since it implies that it is sufficient
68
3 MultiDerivations and Kahler Forms
to define a multivector field for all local functions, defined on open subsets which
cover the manifold M.
We denote the F (M)module of pvector fields on M by X p (M) and we define the graded F (M)module X (M) := pN X p (M), which is the geometrical
analog of X (A ). The operations which we have introduced in Section 3.1.2 for
skewsymmetric pderivations, namely the wedge product , the contraction F by
a function F and the Lie derivative LV , are introduced in the same way for pvector
fields, upon replacing the functions on which they act by local functions. Clearly,
all properties of these operations can be repeated wordforword, as they are purely
algebraic. However, in a geometrical approach, the formula (3.7) for the Lie derivative of a pvector field should be viewed as a property of the Lie derivative, not as its
definition. To give the geometrical definition of the Lie derivative, let us first recall
that if V and W are vector fields on a manifold M, then their commutator [V , W ] is
the vector field on M, whose value at m M is given by
[V , W ]m := (LV W )m :=
d
Tt (m) t Wt (m) ,
dt t=0
(3.10)
where t is the local flow of V . Notice that the tangent map Tt (m) t of the map
t at t (m) sends tangent vectors at t (m) (here, the vector Wt (m) ) to tangent
vectors at m, so taking the derivative in (3.10) amounts to differentiating a curve
in Tm M. More generally, for P a pvector field on a manifold M and V a vector
field on M, the Lie derivative of P with respect to V is defined as the pvector field,
whose value at m M is given by
(LV P)m :=
d
p Tt (m) t Pt (m) .
dt t=0
(3.11)
One immediate consequence of this definition is that LV P = 0 if and only if P is preserved by the local flow of the vector field V . The graded derivation property (3.8)
of the Lie derivative, and hence also (3.7), follow easily from (3.11).
3.2 Kahler Forms and Differential Forms
In this section, we introduce the objects which are, in a sense, dual to skewsymmetric multiderivations of a (commutative associative) algebra A : the Kahler
forms of A . In the case of a manifold M, the dual objects to multivector fields are
differential forms. The modules of Kahler forms and of differential forms will be
described in this section, together with their algebraic structure.
3.2 Kahler Forms and Differential Forms
69
3.2.1 Kahler Differentials
Let A be a commutative associative Falgebra. The module of Kahler differentials
of A (over F) is the A module 1 (A ), which is the free A module generated by
the set {d(F)  F A }, modulo the submodule, generated by all elements of either
one of the following forms:
d(F + G) d(F) d(G) ,
d(FG) F d(G) G d(F) ,
d(a) ,
where a F and F, G A . We write dF rather than d(F) and we consider d as a
map A 1 (A ), defined by F dF. The pair ( 1 (A ), d) satisfies the following
universal property: for every derivation V : A A , there exists a unique A linear
map V : 1 (A ) A , such that V d = V . In other words, there exists a unique
arrow V which makes the following diagram commutative:
d
A
V
1 (A )
V
A
In order to prove the universal property, notice that the commutativity of the diagram
amounts to saying that V (dF) = V [F], for all F A . It is then clear that V should
be defined as the (unique) A linear map, such that
V (G dF) = G V [F] ,
(3.12)
for all F, G A . It is indeed easily verified that V , given by (3.12), is welldefined;
for example, if F, G A , then
V d(FG) FdG GdF = V [FG] FV [G] GV [F] = 0 ,
since V is a derivation of A . Given an arbitrary A linear map V : 1 (A ) A , it
is also clear that V d : A A is a derivation of A . This fact, combined with the
universal property, implies the existence of a natural isomorphism (of A modules)
X1 (A ) HomA ( 1 (A ), A ) ,
(3.13)
which allows us to think of derivations of A as A linear maps on some A module,
intimately related to A .
70
3 MultiDerivations and Kahler Forms
3.2.2 Kahler Forms
In order to generalize (3.13) to skewsymmetric pderivations, we introduce, for
p N, the A module p (A ) := p 1 (A ), where is the wedge product over A ;
for p = 0, this should be read as 0 (A ) := A . Together, the A modules p (A )
form the graded A module
(A ) :=
p (A ) .
pN
The elements of p (A ), with p > 0, are called Kahler pforms, or simply Kahler
forms. As a vector space, respectively as an A module, p (A ) is generated by
elements of the form GdF1 dFp , respectively of the form dF1 dFp , where
G, F1 , . . . , Fp A . Being, by definition, an exterior algebra, it is an associative,
graded commutative A algebra with product . The Flinear map d : A 1 (A )
extends by functoriality of to an Flinear map d : A (A ). Explicitly,
it is given by
d(F1 Fp ) := dF1 dFp ,
(3.14)
where F1 , . . . , Fp A . It is clear that d is a homomorphism between the algebras
( A , ) and ( (A ), ). For every skewsymmetric pderivation P of A , there
exists a unique A linear map P : p (A ) A , such that the following triangle is
commutative:
pd
pA
p 1 (A ) = p (A )
The definition of P for arbitrary p is the natural generalization of the definition (3.12) of V , which corresponds to the case of p = 1. It is given by
dF1 dFp ) := G P[F1 , . . . , Fp ] ,
P(G
(3.15)
for all G, F1 , . . . , Fp A . It leads to a natural isomorphism (of A modules):
X p (A ) HomA ( p (A ), A ) .
(3.16)
For some of the computations which we will do later, it is useful to view P as an
defined in (3.15), trivially to Kahler
A linear map (A ) A , so we extend P,
qforms, with q = p:
dF1 dFq ) := 0, if q = p ,
P(G
(3.17)
for all G, F1 , . . . , Fq A . With this notation, the map P P yields a natural isomorphism
3.2 Kahler Forms and Differential Forms
X (A )
71
HomA ( p (A ), A ) .
pN
Remark 3.3. For p N, the natural pairing
: X p (A ) A p (A ) A
)
, P := P(
P
(3.18)
induces two A linear maps
1 : X p (A ) HomA ( p (A ), A ) ,
2 : p (A ) HomA (X p (A ), A ) .
As we have seen in (3.16), 1 is an isomorphism. In general, 2 is neither injective
nor surjective; however, if A is the algebra C (M) of smooth functions on a real
manifold M, the map 2 is also an isomorphism (see Section 3.2.5).
3.2.3 Algebra and Coalgebra Structure
We now consider the algebra and coalgebra structures on (A ), which are closely
related to the corresponding structures on A , where A is as before a commutative associative Falgebra. We assume that the reader is familiar with the facts,
recalled in Appendix A (in particular, in Sections A.3 and A.4), that ( A , ) is an
associative, graded commutative Falgebra, and that ( A , ) is a coassociative Fcoalgebra. Since (A ) is an exterior algebra (over A ), we have, as we mentioned
in the previous section, a wedge product on (A ), also denoted by , which
makes ( (A ), ) into an associative, graded commutative A algebra. Also, it
leads to an A linear coassociative coproduct
: (A ) (A ) A (A ) ,
which is defined in the same way as in the case of A (see (A.17)). Namely, p,q
is given, for p, q N, and F1 , . . . , Fp+q A by
p,q (dF1 dFp+q )
(3.19)
:= sgn( )(dF (1) dF (p) ) (dF (p+1) dF (p+q) ) .
S p,q
Recall that coassociativity means that
(1 (A ) ) = ( 1 (A ) ) .
(3.20)
It is clear, from its definition (3.14), that d is a homomorphism of the Fcoalgebras
( A , ) and ( (A ), ).
72
3 MultiDerivations and Kahler Forms
A useful application of the coproduct of (A ), is that the wedge product (3.3) of two multiderivations P, Q X (A ) can be expressed in terms of ,
as follows:
,
(3.21)
P
Q = (P Q)
where is the canonical isomorphism
: A A (A ) (A )
F
(3.22)
F .
We used in (3.21) only for the case of 0 (A ) = A , but we will need its
general form later on; in particular we will need the following properties of :
,
P = (1A P)
(3.23)
= ( 1 (A ) ) (1A ) ,
where P X (A ). The proof of these properties follows at once from the definition
(3.22) of .
3.2.4 The de Rham Differential and Cohomology
The differential d : A 1 (A ) extends to a welldefined graded Flinear map
d : (A ) +1 (A ) ,
by putting
d(G dF1 dFp ) := dG dF1 dFp ,
(3.24)
for all G, F1 , . . . , Fp A , where p N. It is called the (algebraic) de Rham differential. It is a graded derivation, of degree 1, of ( (A ), ), as
d( ) = d + (1) p d ,
(3.25)
for p (A ) and (A ), an easy consequence of (3.24). Since da = 0 for
every a F, it follows from (3.24) that d d = 0, so that d is a coboundary operator
and the sequence
0
0 (A )
1 (A )
2 (A )
is a complex, called the (algebraic) de Rham complex. A Kahler pform for which
d = 0 is called closed, while it is called exact if it is contained in the image of d. By
the above, exact Kahler forms are closed, which leads, for p N, to the cohomology
3.2 Kahler Forms and Differential Forms
space2
p
HdR
(A ) :=
73
Ker(d : p (A ) p+1 (A ))
,
Im(d : p1 (A ) p (A ))
0 (A ) := Ker(d : A 1 (A )) = F. The Fvector
where it is understood that HdR
p
(A ) is called the pth de Rham cohomology space and the graded vector
space HdR
space
p
HdR
(A ) :=
HdR (A )
pN
is called the (algebraic) de Rham cohomology of A .
The derivation property (3.25) of d implies the following two facts: (1) the wedge
product of two closed Kahler forms is closed; (2) the wedge product of an exact
Kahler form with a closed Kahler form is exact. Combining these two properties
shows that there is an induced product
(A ) H (A ) H (A )
: HdR
dR
dR
([ ], [ ])
[ ]
where [ ] denotes the de Rham cohomology class of a closed Kahler form . It fol (A ), ) is, just like ( (A ), ), an associative, graded commutative
lows that (HdR
Falgebra.
To close this section, we wish to point out that leads to a covariant functor
from the category of commutative associative Falgebras to the category of graded
Fvector spaces (associative, graded commutative Falgebras, if you wish). Namely,
let : A B be an algebra homomorphism, where A and B are commutative
associative algebras. For every p N, the induced linear map p ( ) : p (A )
p (B) is defined as follows:
p ( ) :
p (A )
p (B)
G dF1 dFp (G) d (F1 ) d (Fp ) .
From this definition, it is easily verified that is indeed a covariant functor. Since
( ) and d commute, ( ) induces a graded linear map of degree 0:
( ) : HdR
(A ) HdR
(B) .
HdR
is also a (covariant) functor.
It follows that HdR
2
Notice that it is only a vector space and not an A module, as the differential d is only Flinear.
74
3 MultiDerivations and Kahler Forms
3.2.5 Differential Forms
There are several equivalent ways of defining the notion of a differential form on a
manifold. The most economical definition of a differential pform on a real manifold
is that it is an F (M)linear map from X p (M) to F (M). In other words, it is a skewsymmetric, F (M)plinear map from X1 (M) to F (M). This definition does not
work in the case of a complex manifold since, as we have already said, the space of
holomorphic functions on M may consist of constant functions only. To remedy this,
one considers local functions and local vector fields. Namely, given a manifold M, a
map which assigns to every m M an element m p Tm M, is called a differential
pform if for every open subset U M and for all vector fields V1 , . . . , V p X1 (U),
one has that (V1 , . . . , V p ) F (U), where (V1 , . . . , V p ) is the function on U,
defined by
(V1 , . . . , V p )(m) := m ((V1 )m , . . . , (V p )m ) ,
for all m U. Thus, on every coordinate chart (U, x), the differential pform
becomes a skewsymmetric F (U)plinear map from F (U) to itself. It follows
that admits the local coordinate expression
=
(3.26)
xi , . . . , xi dxi1 dxi p ,
p
1
1i <<i p d
1
where d denotes the dimension of M. Just like a pvector field, a differential pform
is a tensorial object: for m M, the value of (V1 , . . . , V p )(m) depends only on the
values at m of the vector fields V1 , . . . , V p , which is a useful fact for constructing
differential pforms.
The F (M)module of differential pforms on M is denoted by p (M) and we
define the graded F (M)module
(M) :=
p (M) ,
pN
which is the geometrical analog of the A module (A ), which was defined
in Section 3.2.2. If U is the domain of a coordinate chart on M, then p (U) =
p 1 (U), just like in the case of Kahler differentials, although p (M) = p 1 (M),
in general. The fact that, locally, p (M) is an exterior algebra, leads to a product,
called the wedge product of differential forms. It makes (M) into an associative,
graded commutative algebra. The differential of a function, which we introduced in
Section 1.3.1, is naturally viewed as a linear map d : 0 (M) 1 (M). It extends
to a unique graded derivation of (M) of degree 1, which means that
d( ) = d + (1) p d ,
(3.27)
for p (M) and (M), just like in (3.25). In a coordinate chart (U, x), it
leads to the classical formula for the differential of a differential pform, namely,
3.3 The Schouten Bracket
75
1i1 <<i p d
i1 ,...,i p dxi1 dxi p
i1 ,...,i p
dx j dxi1 dxi p ,
xj
j=1
d
1i1 <<i p d
and d is called the de Rham differential. It has formally the same properties as the
algebraic de Rham differential, in particular it leads to a complex, known as the de
Rham complex, whose cohomology
HdR
(M) :=
p
HdR
(M) :=
pN
Ker(d : p (M) p+1 (M))
pN
Im(d : p1 (M) p (M))
is called the de Rham cohomology of M. In the manifold case, the properties of the
de Rham differential are usually derived from the following explicit formula: d is
given, for p (M), by
d (V0 , V1 , . . . , V p ) =
i
V
(1)
V
,
V
,
.
.
.
,
V
,
.
.
.
,
V
i
i
p
0 1
i=0
(3.28)
j , . . . , V p ,
(1)i+ j [Vi , V j ] , V0 , . . . , V
i , . . . , V
0i< jp
where V0 , . . . , V p are arbitrary local vector fields on M.
3.3 The Schouten Bracket
In this section we introduce a few extra natural operations on the graded algebras
of skewsymmetric multiderivations and of Kahler forms on a commutative associative algebra A . The first one is an action of X (A ) on (A ), which generalizes the natural pairing between multiderivations and Kahler forms, which we
encountered in Remark 3.3. The second operation is the Schouten bracket which extends the classical commutator of derivations to the case of skewsymmetric multiderivations. It will lead at the end of the section to the socalled Gerstenhaber algebra structure on X (A ) and to two generalizations of the Lie derivative, acting
in one case on the algebra of multiderivations, in the other case on the algebra of
Kahler forms, where the derivative is in both cases taken with respect to a multiderivation. In the geometrical setting, this yields for every manifold M on the one
hand an action of the graded algebra of multivector fields X (M) on the graded algebra of differential forms (M) and on the other hand a graded Lie bracket, the
Schouten bracket, on X (M).
76
3 MultiDerivations and Kahler Forms
3.3.1 Internal Products
The universal property of the graded A module (A ) of Kahler differentials on a
commutative associative algebra A leads to a natural action of X (A ) on (A ),
often referred to as the internal product. To introduce it, we associate to P X p (A ),
the graded A linear map of degree p,
P : (A ) p (A ) ,
which is defined for = dF1 dFk k (A ), by
P (dF1 dFk ) :=
S p,kp
sgn( )P[F (1) , . . . , F (p) ] dF (p+1) dF (k)
(3.29)
when k p, and otherwise P := 0; for F X0 (A ) = A , (3.29) has to be understood as F := F . The reader may wish to write the definition of P differently,
using
P[F (1) , . . . , F (p) ] = P dF (1) dF (p) .
which
With this notation, it is obvious that the restriction of P to p (A ) is just P,
was defined in (3.15); the restriction to the spaces q (A ), with q = p, is however
different, since P acts trivially on them (see (3.17)). It also follows that P can be
written, for every P X (A ), in terms of the coproduct , defined in (3.19), as
P = (P 1 (A ) ) .
(3.30)
As a consequence, P is the adjoint of taking the wedge product with P: for P
X p (A ) and Q Xq (A ) and p+q (A ), we have in terms of the notations
in (3.18),
P , Q = , P Q .
(3.31)
Indeed, (3.30), (3.23) and (3.21) imply that, on p+q (A ),
Q P = Q (P 1 (A ) )
(P 1 (A ) )
= (1A Q)
= (P Q)
= P
Q ,
which yields (3.31), when applied to p+q (A ). Equation (3.31) explains the
notation P : as in the case of F (F A ), defined in (3.5), P provides with the
pderivation P as its first p arguments. Yet, when considering F G for F, G A
one should decide from the context whether G is interpreted as a 0derivation or as
a 0form, since in the first case F G = 0 (as defined in (3.5)), while in the second
case F G = FG (as defined in (3.29)).
In the following proposition, we establish the main properties of .
3.3 The Schouten Bracket
77
Proposition 3.4. Let A be a commutative associative algebra. The following identities hold, for P X p (A ) and Q Xq (A ):
(1)
(2)
(3)
P Q = QP ;
[P , Q ] = P Q (1) pq Q P = 0;
For every F A and (A ):
P (dF ) = (F P) + (1) p dF P .
(3.32)
Proof. We give three different proofs of (1). The first one is a quick proof, but it
is not valid for general algebras, because we assume that HomA (X p (A ), A )
p (A ), for all p N (see Remark 3.3 at the end of Section 3.2.2). Applying (3.31)
three times, we find for every k (A ) and for homogeneous elements P, Q and
R of X (A ), whose degrees sum up to k,
QP , R = , (Q P) R = , Q (P R)
(3.33)
= Q , P R = P (Q ), R .
Under the above assumption, the fact that (3.33) is valid for all R, implies that
QP = P (Q ), for all (A ), which leads to (1). Our second proof is
based on the fact that is coassociative. To simplify the notation, we will write 1
for 1 (A ) ; using (3.30), the properties (3.23) of , the coassociativity (3.20) of
and (3.21) (in that order), we find
P Q = (P 1) (Q 1)
= (P 1) ( 1) (1A ) (Q 1)
= ( 1) (1A P 1) (Q 1 1) (1 )
= ( 1) (Q P 1) ( 1)
) 1)
= (( (Q P)
= Q
P1
= QP .
This terminates the second proof. For the third proof, we show by direct computation
that
(3.34)
(P Q ) (dF1 dFk ) = QP (dF1 dFk )
for all F1 , . . . , Fk . Since P is A linear for all P X p (A ), this will provide a third
proof of (1). Since P = 0 for q (A ) when q < p, where P X p (A ), it
suffices to prove (3.34) when k p + q. On the one hand,
P Q (dF1 dFk )
= P
Sq,kq
sgn( )Q[F (1) , . . . , F (q) ]dF (q+1) dF (k)
78
3 MultiDerivations and Kahler Forms
Sq,p,kqp
sgn( )Q[F (1) , . . . , F (q) ]P[F (q+1) , . . . , F (q+p) ]
dF (q+p+1) dF (k) ,
where Sq,p,kpq is the set of all permutations of the set {1, . . . , k} which are
increasing on the intervals 1, . . . , q and q + 1, . . . , q + p and q + p + 1, . . . , k. On the
other hand, (3.29) and (3.3) lead to
QP (dF1 dFk )
=
sgn( )(Q P)[F (1) , . . . , F (q+p) ]dF (q+p+1) dF (k)
Sq+p,kqp
Sq,p,kqp
sgn( )Q[F (1) , . . . , F (q) ]P[F (q+1) , . . . , F (q+p) ]
dF (q+p+1) dF (k) ,
which yields the third proof. The point of this computation is that an element of
Sq,p,kpq can be uniquely obtained from a shuffle Sq,kq , followed by a shuffle
of the set (q + 1), . . . , (k), or from a shuffle Sq+p,kpq , followed by a shuffle
of the set (1), . . . , (q + p).
The proof of (2) is immediate from (1), since the graded commutator is given by
[P , Q ] = P Q (1) pq Q P = QP (1) pq PQ , and since PQ = (1) pq Q P
(graded commutativity, see (3.4)).
We now prove (3). Since, for every Q X (A ), the map Q is A linear and
the wedge product is also A bilinear, we can suppose that is of the form =
dF1 dFk , with k N , k p 1 and F1 , . . . , Fk A . In this case,
P (dF dF1 dFk )
S p1,kp+1
S p,kp
sgn( ) P[F, F (1) , . . . , F (p1) ] dF (p) dF (k)
(1) p sgn( ) P[F (1) , . . . , F (p) ] dF dF (p+1) dF (k) ,
since every (p, k p + 1)shuffle of {0, . . . , k} has 0 as the first element in its first
part or in its second part. The first sum is exactly (F P) (dF1 dFk ) = (F P) ,
while the second sum is given by (1) p dF P (dF1 dFk ) = (1) p dF P .
This yields a proof of Eq. (3.32).
For P X p (A ), the operation P : (A ) p (A ) is closely related to the
natural extension of P : p A A to a coderivation
P : A p+1 (A )
of degree p 1 of ( A , ). It is defined by
3.3 The Schouten Bracket
1 Fk ) :=
P(F
79
S p,kp
sgn( ) P[F (1) , . . . , F (p) ] F (p+1) F (k) , (3.35)
1 Fk ) := 0 otherwise (see Example A.4 in Appendix A).
for k p and P(F
Comparing this formula with (3.29), it is clear that d P and P correspond under
the map d : A (A ), i.e., the following diagram is commutative:
P
p+1 A
d
dP
(A )
p+1 (A )
(A )
As before, the algebraic constructions in this section, in particular the internal
product, can easily be transcribed for the case of a manifolds. For a manifold M, the
internal product is an action of X (M) on (M), where we recall that X (M) is
the F (M)module of multivector fields on M and (M) is the F (M)module of
differential forms on M. This action is defined as in (3.29), the functions F1 , . . . , Fk
being now arbitrary local functions on M. The formulas in Proposition 3.4 remain
valid, with P, Q X (M), with (M) and F F (M).
3.3.2 The Schouten Bracket
We now introduce the Schouten bracket, which is a product of skewsymmetric
multiderivations, based on the operation of composition. It is a family of maps
[ , ]S : X p (A ) Xq (A ) X p+q1 (A ) ,
for p, q N, which therefore only respects the grading up to a shift over 1, suggesting to shift the degree of all multiderivations by 1. Namely, we define X p1 (A ) :=
X p (A ), for p N, and we call p := p 1 the shifted degree of an element P of
X p (A ) = X p (A ). Thus, elements of A have shifted degree 1, derivations of A
have shifted degree 0, and so on. Notice that p + q 1 = p + q, so that the Schouten
bracket is a family of maps
[ , ]S : X p (A ) Xq (A ) X p+q (A ) ,
for p, q N {1}, hence respects the (new) grading. It is defined, for P X p (A )
and Q Xq (A ), and for F1 , . . . , Fp+q+1 A , by
80
3 MultiDerivations and Kahler Forms
[P, Q]S [F1 , . . . , Fp+q+1 ]
= sgn( )P Q[F (1) , . . . , F (q) ], F (q+1) , . . . , F (q+p)
Sq,p
(1) pq
S p,q
(3.36)
sgn( )Q P[F (1) , . . . , F (p) ], F (p+1) , . . . , F (p+q) .
A priori, we have only that [P, Q]S Hom(A (p+q+1) , A ), but the reader will easily check that indeed [P, Q]S X p+q+1 (A ) = X p+q (A ), by showing that [P, Q]S is
skewsymmetric and verifies the derivation property (3.2). In the case of a manifold,
the Schouten bracket of multivector fields is defined as in (3.36), but replacing the
functions Fi by local functions.
The Schouten bracket can be seen as a generalization of many classical elementary operations on functions, derivations and multiderivations. First, let Q := F A
and P X p (A ), then [P, F]S = F P (see (3.5)). Second, let P := V X1 (A ) and Q
Xq (A ), then [V , Q]S = LV Q, the Lie derivative of Q with respect to V (see (3.7)).
Third, the Schouten bracket of two skewsymmetric biderivations P, Q X2 (A ) is
given by
[P, Q]S [F1 , F2 , F3 ] = P[Q[F1 , F2 ], F3 ] + Q[P[F1 , F2 ], F3 ]+ (F1 , F2 , F3 ) ,
(3.37)
where F1 , F2 , F3 A . This leads to the following result.
Proposition 3.5. Let A be a commutative associative algebra. If P is a skewsymmetric biderivation of A , i.e., P X2 (A ), then P defines a Poisson bracket
on A if and only if [P, P]S = 0.
Similarly, in the case of a manifold M, a bivector field P on M defines a Poisson
structure on M if and only if [P, P]S = 0.
We say that P, Q X2 (A ) are compatible if [P, Q]S = 0. It follows that two Poisson brackets P and Q are compatible if and only if their sum (equivalently, an arbitrary linear combination with nonzero coefficients) is a Poisson bracket.
In the following proposition, we express the Schouten bracket in terms of the de
Rham differential.
Proposition 3.6. Let A be a commutative associative algebra. If P and Q are two
skewsymmetric multiderivations of A , then
[[P , d] , Q ] = [P,Q]S .
(3.38)
This formula is called Cartans formula.
Proof. We prove Cartans formula for elements P X p (A ) and Q Xq (A ). Let
us first show that the linear map [[P , d], Q ] is A linear. We take F A and
(A ). Then, using (3.32),
[P , d](F ) = P (dF ) + FP d (1) p d(FP )
3.3 The Schouten Bracket
81
= (F P) + FP d (1) p FdP
= F[P , d] + (F P) ,
so that
[[P , d], Q ](F )
= [P , d](FQ ) (1) pq Q [P , d](F )
= F[P , d](Q ) + (F P) Q (1) pq FQ [P , d] (1) pq Q (F P)
= F[[P , d], Q ] ,
where we have used in the last step that (F P) , Q = 0 (Proposition 3.4). This shows
that
[[P , d] , Q ] : (A ) pq+1 (A )
is A linear. It follows that, in order to establish Cartans formula (3.38), we only
need to show that
P dQ (1) p dP Q (1) pq Q dP = [P,Q]S ,
(3.39)
for of the form = dF1 dFk , where F1 , . . . , Fk A and where k N.
We do this by induction on k, starting with k = 0, so that we take := 1 A .
Notice that both sides in (3.39) have degree k p q + 1, so that when k = 0,
both sides of (3.39) are zero as soon as p + q > 1. If p + q = 1, we may suppose
that p = 1 and q = 0 (the case p = 0 and q = 1 is obtained by symmetry): then
P := V X1 (A ) and Q := F A and the righthand side of (3.39) (with = 1)
is given by [V ,F]S 1 = V [F] 1 = V [F], while the lefthand side is given by V dF +
dV F F dV 1 = V dF = V [F], as V F = 0 for degree reasons. If p + q = 0, then
p = q = 0 so that P := F and Q := G, with F, G A . Then [F, G]S = 0, while the lefthand side of (3.39) reduces to F dG d(FG) + G dF = FdG d(FG) + GdF = 0.
This shows that (3.39) holds for of degree 0, for all P and Q.
We now assume that (3.39) holds for k (A ) and we show that it holds
for dF , where F A is arbitrary. On the one hand, (3.32) and the induction
hypothesis imply that
[P,Q]S (dF )
= (F [P,Q]S ) (1) p+q dF [P,Q]S
= (F [P,Q]S ) (1) p+q dF ([[P , d] , Q ] )
= (1)q [F P,Q]S + [P,F Q]S (1) p+q dF ([[P , d] , Q ] ) ,
where we have used
F [P, Q]S = (1)q [F P, Q]S + [P, F Q]S
(3.40)
82
3 MultiDerivations and Kahler Forms
to obtain the last line, which is similar to (3.6), and is a direct consequence of the
definition of [P, Q]S . Equation (3.40) has to be compared to [[P , d] , Q ] (dF ), i.e.,
to
(P dQ (1) p dQP (1) pq Q dP )(dF ) .
(3.41)
Using (3.32) twice and (3.25) once, we find
P dQ (dF ) = P d(F Q) (1)q P (dF dQ )
= P d(F Q) (1)q (F P) dQ (1) p+q dF P dQ .
Similarly, we compute, using (3.32) and (3.6)
dQP (dF ) = d(F (QP)) + (1)q+p dF dQP
= d(F QP) + (1)q d(QF P) + (1)q+p dF dQP .
If we substitute these results in (3.41), then we find nine terms which can, by using
the induction hypothesis twice, be written as follows:
(P d(F Q) (1) p d(F QP) (1) pq (F Q) dP ) = [P,F Q]S
(1)q ((F P) dQ + (1) p d(QF P) (1) pq Q d(F P) ) = (1)q [F P,Q]S
dF (1) p+q P dQ + (1)q dQP (1) pq Q dP =
(1) p+q dF ([[P , d] , Q ] ) .
This yields precisely the three terms in (3.40).
Cartans formula leads to a simple proof of the main properties of the Schouten
bracket, as given in the following proposition.
Proposition 3.7. Let A be a commutative associative algebra. The Schouten bracket
[ , ]S : X (A ) X (A ) X (A ) ,
defines a graded Lie algebra structure on X (A ), meaning that, for all homogeneous elements P, Q and R in X (A ), of respective shifted degree p, q and r,
(1)
(2)
(3)
[P, Q]S X p+q (A );
(graded skewsymmetry);
[P, Q]S = (1) pq [Q, P]S ,
(1) pr [P, [Q, R]S ]S + (P, Q, R) = 0
(graded Jacobi identity).
Moreover, the associative product and the Lie bracket [ , ]S are compatible in the
sense that [, P]S is a graded derivation of (X (A ), ) of degree p:
(3.42)
[Q R, P]S = [Q, P]S R + (1) pq Q [R, P]S ,
(graded Leibniz identity),
with P, Q and R as above. In view of the graded skewsymmetry of [ , ]S , one also
has, for such elements,
3.3 The Schouten Bracket
83
[P, Q R]S = Q [P, R]S + (1) pr [P, Q]S R .
(3.43)
Proof. (1) and (2) follow at once from the explicit formula for the Schouten bracket
(Eq. (3.36)). Notice that the graded skewsymmetry also follows from Cartans formula (3.38). Indeed, the graded Jacobi identity for the graded commutator [ , ],
(1) pq [[P , d] , Q ] + (1) p [[d, Q ] , P ] + (1)q [[Q , P ], d] = 0 ,
leads, since [Q , P ] = 0, to
[Q,P]S = [[Q , d], P ] = (1) pq [[P , d], Q ] = (1) pq [P,Q]S ,
which proves that [Q, P]S = (1) pq [P, Q]S , since P P is injective. We next prove
the graded Jacobi identity for [ , ]S . Let P X p (A ) and Q Xq (A ). We associate
to these multiderivations, the coderivations P (of degree p) and Q (of degree q) of
the coalgebra ( A , ), as in (3.35). We claim that
= [P,
Q]
[P,
Q]S .
(3.44)
Since both members are coderivations of A of degree p + q 2, it suffices to
verify that they agree in degree k := p + q 1 (see Example A.4 in Appendix A).
For F1 , . . . , Fk A , we have that
1 , . . . , Fk )
(P Q)(F
= P
Sq,p
Sq,p
sgn( ) Q[F (1) , . . . , F (q) ] F (q+1) F (k)
sgn( )P Q[F (1) , . . . , F (q) ], F (q+1) , . . . , F (k) ,
which leads, in view of (3.36), to
1 , . . . , Fk ) = [P, Q]S [F1 , . . . , Fk ] = [P,
Q](F
[P,
Q]S (F1 , . . . , Fk ) .
This shows (3.44). As a consequence, for all skewsymmetric multiderivations P, Q
and R,
R
Q,
[Q,
([P, [Q, R]S ]S ) = P,
R]S = P,
and the graded Jacobi identity for the Schouten bracket follows from the graded
Jacobi identity for the commutator of coderivations.
The Leibniz property of the Schouten bracket follows easily from Cartans formula. Indeed, for P, Q and R as in (3.43), Cartans formula and (1) of Proposition 3.4
imply that
[P,QR]S = [[P , d] , QR ] = [[P , d] , R Q ]
84
3 MultiDerivations and Kahler Forms
= [[P , d] , R ] Q + (1)rp R [[P , d] , Q ]
= Q[P,R]S + (1)rp [P,Q]S R ,
which yields (3.43), and hence also (3.42). We have used the formula3
[ , 1 2 ] = [ , 1 ] 2 + (1)d1 d 1 [ , 2 ] ,
(3.45)
which is valid for graded linear maps 1 , 2 and , of respective degrees d1 , d2
and d.
We have equipped X (A ) with two graded algebra structures: the wedge product ,
which is associative and graded commutative, and the Schouten bracket [ , ]S , which
is a graded Lie bracket (with respect to a different grading). Moreover, we have seen
that these two products are, in a sense, compatible. Formalizing these properties
leads to the notion of a Gerstenhaber algebra.
Definition 3.8. Let V = pZVp be a graded Fvector space and denote by p :=
p 1 the shifted degree of P Vp . Suppose that V is equipped with two multiplications and [ , ], such that
(1)
(2)
(3)
(V , ) is an associative, graded commutative algebra over F;
(V , [ , ]) is a graded Lie algebra over F, with respect to the shifted grading;
The two algebra structures are compatible in the sense that
[Q R, P] = [Q, P] R + (1) pq Q [R, P] ,
where P, Q and R are arbitrary homogeneous elements of V , of degree p, q
and r.
Then (V , , [ , ]) is called a Gerstenhaber algebra (over F).
Comparing the three items in Definition 3.8 to the ones in Definition 1.1, one sees
that the notion of a Gerstenhaber algebra is a natural graded analog of the notion of
a Poisson algebra.
Proposition 3.7 can be reformulated by saying that for every commutative associative Falgebra A , the graded Fvector space X (A ), equipped with the wedge
product and the Schouten bracket, is a Gerstenhaber algebra (over F). In the case of a
manifold M, the graded Fvector space X (M) of multivector fields on M, equipped
with the wedge product and the Schouten bracket, is a Gerstenhaber algebra.
3.3.3 The Algebraic Schouten Bracket
For any Lie algebra (g, [ , ]) there is a graded Lie algebra structure on g, which is
the natural extension of the Lie bracket [ , ] on g. It is very similar to the Schouten
3
The formula is a graded version of the wellknown matrix identity [x, yz] = [x, y]z + y[x, z]; also,
it is half of the Jacobi identity (A.13).
3.3 The Schouten Bracket
85
bracket on multiderivations, in particular it only becomes a graded Lie algebra
structure when we shift the grading: for p N , we let p1 g := p g and we let
p := p 1. The definition and main properties of the graded bracket on the graded
vector space g := kN k g are given in the following proposition.
Proposition 3.9. Let (g, [ , ]) be a Lie algebra and let [[ , ]] denote the bracket
on g, defined for all pairs of monomials by
[[x1 x p , y1 yq ]]
p
(3.46)
:= (1) pq (1)i+ j [xi , y j ] x1 x
i x p y1 y
j yq .
i=1 j=1
Then ( g, , [[ , ]]) is a Gerstenhaber algebra. The graded Lie bracket [[ , ]] on g
is called the algebraic Schouten bracket.
Proof. The proof of each one of the properties of ( g, , [[ , ]]) is either immediate
from the definitions or follows from a direct computation. Using the graded Leibniz
identity
(3.47)
[[Y Z, X]] = [[Y, X]] Z + (1) pqY [[Z, X]] ,
valid for all X p g, Y q g and Z g, the graded Jacobi identity can be
proven easily by recursion, since for three elements of shifted degree zero, it is just
the Jacobi identity in g, while if at least one element is a monomial of higher shifted
degree, say X, it can be written as X = X x, with x g and X g, of lower
degree. For a quick proof of (3.47), one clearly only needs to check that the signs
in the righthand side of (3.47) are correct, which follows at once from the graded
skewsymmetry of [[ , ]].
3.3.4 The (Generalized) Lie Derivative
We have already encountered the Lie derivative of a multiderivation with respect
to a derivation in Section 3.1.2; it is the algebraic analog of the Lie derivative of a
multivector field with respect to a vector field, as briefly discussed in Section 3.1.3.
As we have shown in Section 3.3.2, the Lie derivative can be written in terms of the
Schouten bracket, namely LV Q = [V , Q]S , for V X1 (A ) and Q Xq (A ) (as before, A is a commutative associative algebra). Since [ , ]S is a graded derivation of
degree one in its first argument (see (3.42)), taking the Schouten bracket with a given
pderivation P, is a natural generalization of the Lie derivative, called generalized
Lie derivative, or simply Lie derivative. Explicitly, for P X p (A ), the generalized
Lie derivative LP is given by the graded linear map (of degree p = p 1)
LP : X (A ) X+p (A )
Q [P, Q]S .
(3.48)
86
3 MultiDerivations and Kahler Forms
The properties of the Schouten bracket, given in Proposition 3.7, lead at once to the
following properties for this generalized Lie derivative.
Proposition 3.10. Let A be a commutative associative algebra and let P, Q and R
be homogeneous elements of X (A ) of shifted degrees p, q and r. Then
(1)
(2)
(3)
LP Q X p+q (A ) ;
LP (Q R) = Q LP R + (1) pr (LP Q) R ;
LP [Q, R]S = [LP Q, R]S + (1) pq [Q, LP R]S .
In the geometrical context, the generalized Lie derivative becomes a Lie derivative with respect to an arbitrary pvector field P; it associates to a qvector field a
(q + p 1)vector field, i.e., it raises the degree of a multivector field by p = p 1.
Since the calculus of differential forms has been more popularized than the calculus of multivector fields, the notion of the Lie derivative of a differential form
(rather than of a multivector field) with respect to a vector field is better known.
We recall this definition and generalize it to the Lie derivative of a differential form
with respect to a multivector field. We start from the geometrical definition, which
is similar to the geometrical definition (3.11). Namely, let V be a vector field on a
manifold M and let be a differential pform on M. Then LV is the differential
pform on M, whose value at m M is given by
(LV )m =
d
t (m) ,
dt t=0 t
where t is the local flow of V and t the pullback of by t , so that t (t (m) )
is a differential pform on Tm M (depending on t). The classical4 Cartan formula says
that
(3.49)
LV = V d + d V ,
a formula which we can take as a definition of the Lie derivative in the algebraic setting. Writing (3.49) as LV = [V , d], where we recall that [ , ] denotes
the graded commutator (see (A.12)), leads to the following natural generalization
of the Lie derivative to an action of a multivector field (or multiderivation) on
a differential form (Kahler form), namely we define the Lie derivative with respect to a skewsymmetric multiderivation P X p (A ) as the graded Flinear map
(A ) p (A ) of degree p, given by
LP := [P , d] ,
where we recall that P was defined in (3.29). Explicitly, this means that for
(A ),
(3.50)
LP := P (d ) (1) p d(P ) .
Comparing the Definitions (3.48) and (3.50) of the generalized Lie derivative,
one notices that there is some ambiguity in the definition of the generalized Lie
4
The classical Cartan formula should not be confused with Cartans formula (3.38), which relates
the Schouten bracket to the de Rham differential.
3.4 Exercises
87
derivative of a function with respect to a pvector field, when p 2. In fact, a function F A can be viewed as an element of X0 (A ) or as an element of 0 (A ).
According to (3.48), LP F = F (P) X p1 (A ), which is a priori different from
zero; however, according to (3.50), LP (F) 1p (A ) = {0}. Therefore, when
applying the generalized Lie derivative to functions, one should decide from the
context and/or from the problem at hand, whether Definition (3.48) is used, or Definition (3.50).
The main properties of the Lie derivative LP : (A ) p (A ) are summarized in the following proposition.
Proposition 3.11. Let A be a commutative associative algebra. For P X p (A )
and Q Xq (A ), the following formulas hold:
(1)
(2)
(3)
(4)
[LP , d] = 0 ;
[LP , Q ] = [P,Q]S ;
L[P,Q]S = [LP , LQ ] ;
LPQ = Q LP + (1) p LQ P .
Proof. Since dd = 0, we have that [LP , d] = (1) p dP d+(1) p dP d = 0,
which leads to (1), while (2) is precisely Cartans formula (3.38). Combined with
(1) and (2), the graded Jacobi identity for [ , ] yields
[LP , LQ ] = [LP , [Q , d]] = [[LP , Q ] , d] = [P,Q]S , d = L[P,Q]S ,
which proves (3). Finally, using PQ = Q P (see Proposition 3.4) and (3.45),
LPQ = [PQ , d] = [Q P , d]
= Q [P , d] + (1) p [Q , d] P
= Q LP + (1) p LQ P ,
which proves (4).
3.4 Exercises
1. Prove the explicit formula (3.28) for a Kahler kform on an arbitrary commutative associative algebra A ; the arguments Vi are in this context arbitrary derivations of A .
2. Give an alternative proof of the graded Jacobi identity for the Schouten bracket
by using Cartans formula (3.38).
3. Let A be an arbitrary commutative associative algebra. Prove the following
formulas, for F A and (A ):
LFP = F LP (1) p dF P ,
LP (F ) = F LP + (F P) .
88
3 MultiDerivations and Kahler Forms
These formulas show that LP is neither A linear in P nor in .
4. Let A be a commutative associative algebra and let B be an A module. A
linear map : A B is called a derivation of A with values in B if for all
F1 , F2 A , one has (F1 F2 ) = F1 (F2 ) + F2 (F1 ).
a.
b.
c.
Interpret the Hamiltonian operator X , associated to a Poisson bracket on A
(see Definition 1.3), as a derivation of A with values in X1 (A );
For A = F (M) an algebra of functions on a variety or manifold, interpret
pointwise derivations at a point m M as derivations with values in F, where F
is viewed as an A module, via F a := F(m)a, for F A and a F;
Interpret the de Rham differential d as a derivation of A with values in 1 (A ).
5. Suppose that A is a commutative associative algebra, that { , } is a Poisson
bracket on A and that V is a derivation of A . Show that { , } and the Lie derivative
of { , } with respect to V are compatible (see Section 3.3).
6. Let A := F[x, y]/x2 + y2 . Show that X2 (A ) = 0 and that 2 (A ) = 0.
7. The purpose of this exercise is to give an example of an algebra A , which
admits skewsymmetric biderivations which cannot be written as the sum of wedge
products of derivations of A . In formulas, this means that X2 (A ) = 2 X1 (A ).
Consider the quotient algebra
A :=
F[x, y, z]
.
yx, yz, y2
induces a nonzero skewsymmetric biderivation P
y z
a.
Show that P := y
b.
of A ;
Show that P does not belong to the exterior algebra 2 X1 (A ).
8. Let A be a commutative associative algebra, and let P X2 (A ) be a skewsymmetric biderivation of A . Let A [ ] denote the vector space of all polynomials
in with coefficients in A . Consider on A [ ]/ 2 the product, defined for elements F1 + F2 and G1 + G2 of A [ ]/ 2 by
(F1 + F2 ) (G1 + G2 ) := F1 G1 + (F1 G2 + F2 G1 + P[F1 , G1 ]) .
a.
b.
Show that is F[ ]bilinear and associative;
Show that, if can be extended to an associative F[ ]bilinear product on
A [ ]/ 3 , then P satisfies the Jacobi identity.
3.5 Notes
The calculus of differential forms on manifolds is explained in detail in all basic
books on differential geometry, see for example Spivak [186, 187] or Warner [198].
3.5 Notes
89
For the algebraic analog, which is the calculus of Kahler forms, we refer to
Hartshorne [93].
Multivector fields are rarely explicitly treated in books on differential geometry; even if those fields themselves are particular cases of tensor fields, everywhere
treated in detail, their algebraic structure does not derive from the general tensor
formalism. The calculus of multivector fields is presented in detail in Vaisman [194]
and in BhaskaraViswanath [23]. Cannas da SilvaWeinstein [34] concentrate on
the structure of the space of multivector fields as a Gerstenhaber algebra, which
they present as a particular case of the Gerstenhaber algebra of a Lie algebroid.
The Schouten bracket, also called the SchoutenNijenhuis bracket, was introduced by Schouten in [177, 178]; a generalization of the Schouten bracket, which
also generalizes the FrohlicherNijenhuis bracket on vector valued differential
forms, is given in Vinogradov [197].
Chapter 4
Poisson (Co)Homology
A Poisson bracket on a commutative associative algebra A , or a Poisson structure
on a manifold M, leads in a natural way to cohomology spaces, derived from the
multiderivations of A (multivector fields on M), and to homology spaces, derived
from the Kahler differentials of A (differential forms on M). These spaces give
information on the derivations, normal forms, deformations and several invariants
of the Poisson structure. In some specific, but important, cases they are related to
classically known cohomology spaces, as de Rham cohomology or Lie algebra cohomology. In general, Poisson cohomology is finer, but is also more difficult to
compute. See Table 4.1 for a concise overview of the spaces and the (co)boundary
operations which will be introduced in this chapter.
Table 4.1 A summary of the notations which we use for Lie and Poisson homology and cohomology. (A , , ) is an arbitrary Poisson algebra over F, and g is a Lie algebra with a representation
space V .
k(co)chains (co)boundary (co)homology spaces
Lie cohomology
Hom(k g,V )
Lk
HLk (g;V )
Poisson cohomology
Xk (A )
Hk (A )
Lie homology
V k g
kL
HkL (g;V )
Poisson homology
k (A
Hk (A )
In Sections 4.1 and 4.2, we construct the various complexes which lead to the homologies and cohomologies which we will consider. In Section 4.3 we describe a
few natural operations in Poisson cohomology and homology and we show in Section 4.4 that the Poisson cohomology and homology spaces of a Poisson manifold
are, under certain conditions, isomorphic to each other.
Throughout this chapter, F is an arbitrary field of characteristic zero.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 4, SpringerVerlag Berlin Heidelberg 2013
91
92
4 Poisson (Co)Homology
4.1 Lie Algebra and Poisson Cohomology
Before constructing the Poisson cohomology complex, and deriving the cohomology spaces from it, we recall Lie algebra cohomology, which is formally very similar to Poisson cohomology, and more widely known (see [156]). The basic notions
which we recall will also be useful when we compare Poisson and Lie algebra cohomology in the case of the canonical Poisson structure, defined on the dual of a
Lie algebra (see Chapter 7).
4.1.1 Lie Algebra Cohomology
Let (g, [ , ]) be a Lie algebra over F. We recall that a representation of g is a Lie
algebra homomorphism : g End(V ), where V is an Fvector space, called the
representation space and the Lie algebra structure on End(V ) is given by the commutator: , for , End(V ). The vector space V , together with the
representation , is called a gmodule. We write x for (x), where x g. We fix a
representation ( ,V ) of g and we introduce, for k N, the Fvector space
Ck (g;V ) := Hom(k g,V )
of skewsymmetric klinear maps from g to V . Elements of Ck (g;V ) are called V valued kcochains of g. The Lie coboundary map Lk : Ck (g;V ) Ck+1 (g;V ) is
given, for c Ck (g;V ) and x0 , . . . , xk g, by
Lk (c)(x0 , . . . , xk ) =
(1)i xi c(x0 , . . . , xi , . . . , xk )
i=0
(1)
i+ j
(4.1)
c [xi , x j ] , x0 , . . . , xi , . . . , xj , . . . , xk ,
0i< jk
where the symbol over an argument means that we omit that argument. Elements of Ker Lk are called Lie kcocycles, while elements in Im Lk1 are called Lie
kcoboundaries. The Jacobi identity for [ , ] and the fact that is a Lie algebra homomorphism permits one to show that Lk Lk1 = 0, for every k N . In words:
every coboundary is a cocycle. The converse is not true in general and the cohomology spaces precisely measure the obstruction. Namely, for k N , we define
HLk (g;V ) := Ker Lk / Im Lk1 ,
and HL0 (g;V ) := Ker L0 . Putting these spaces together, we obtain a graded Fvector
space
HLk (g;V ) ,
HL (g;V ) :=
kN
which is called the V valued Lie algebra cohomology of g.
4.1 Lie Algebra and Poisson Cohomology
93
For example, for k = 0, the coboundary operator is given by L0 (v)(x) = x v, for
v V C0 (g;V ) and x g. It follows that HL0 (g;V ) = Ker L0 is given by
HL0 (g;V ) =
Ker x ,
xg
the subspace of all invariant elements in V , also denoted by V g . The following
lemma is fundamental in Lie theory; we will use it several times to illustrate the
phenomena which we encounter. For a proof, see [102].
Lemma 4.1 (Whiteheads lemma). Let g be a semisimple Lie algebra and let :
g End(V ) be a representation of g. If V is finitedimensional, then HL1 (g;V ) = 0
and HL2 (g;V ) = 0.
The following two examples of representations ( ,V ) are of particular interest
and will be our main cases.
Example 4.2. Let : g F be the trivial representation: x = 0 for every x g.
Then
Ck (g; F) = Hom(k g, F)
and (4.1) becomes:
(1)i+ j c [xi , x j ] , x0 , . . . , xi , . . . , xj , . . . , xk .
Lk (c)(x0 , . . . , xk ) =
(4.2)
0i< jk
In this case, the Lie algebra cohomology of g is called the trivial Lie algebra cohomology of g and is simply denoted by HL (g). It follows from (4.2) that Im L0 = {0}
and that Ker L1 = {c g  x, y g, c([x, y]) = 0} , so that
HL1 (g) g/ [g, g] .
For a semisimple Lie algebra g, for example, it is well known that [g, g] = g (see
[190, Ch. 20]). Therefore, HL1 (g) = 0 when g is semisimple, which is a special case
of Whiteheads lemma (Lemma 4.1).
Example 4.3. Let := ad, the adjoint representation of g on itself, which is given
by x = adx := [x, ], for x g. Then
Ck (g; g) = Hom(k g, g)
and (4.1) becomes:
Lk (c)(x0 , . . . , xk ) =
(1)i [xi , c(x0 , . . . , xi , . . . , xk )]
i=0
(4.3)
(1)i+ j c [xi , x j ] , x0 , . . . , xi , . . . , xj , . . . , xk .
0i< jk
In this case, the Lie algebra cohomology of g is called the ChevalleyEilenberg
(g). Let us give an interpretation of
cohomology of g. We denote it by HCE
94
4 Poisson (Co)Homology
1 (g). The 1cocycles are the derivations of g, since
the space HCE
L1 (c)(x, y) = [x, c(y)] + [c(x), y] c([x, y])
for x, y g and c Hom(g, g). Since L0 (x) = adx , for x g C0 (g; g), the space
of 1coboundaries Im L0 consists of all derivations of g of the form adx , where x g;
1 (g) is the quotient
these derivations are called inner derivations. It follows that HCE
of the space of all derivations of g, by the space of all inner derivations of g. The
elements of this quotient are called the exterior derivations of g.
2 (g) and H 3 (g) play an important r
ole in the theory of formal
The spaces HCE
CE
deformations of Lie algebras. This fact is quite similar (by forgetting the multiderivation aspect) to what happens in the case of Poisson algebras. Therefore, we
will not make it explicit in the Lie algebra case, but in the Poisson algebra case, see
Section 13.2.
4.1.2 Poisson Cohomology
We now define Poisson cohomology for a Poisson algebra (A , , { , }); the Poisson
bracket on A will also be denoted by , so that (F, G) = {F, G}, for F, G A .
For k N, the space of kcochains of the Poisson cohomology complex is by
definition Xk (A ), the Fvector space of skewsymmetric kderivations of A (see
Section 3.1.1). The Poisson coboundary operator is the Poisson analog of the
ChevalleyEilenberg coboundary operator (4.3), where Hom(k g, g) has been replaced by Xk (A ), and where the Lie bracket [ , ] is the Poisson bracket { , }.
Namely, the graded Flinear map (of degree 1) : X (A ) X+1 (A ) is defined,
for Q Xq (A ), where q N, by
q (Q)[F0 , . . . , Fq ] :=
(1)i
i , . . . , Fq ]
Fi , Q[F0 , . . . , F
i=0
(1)i+ j Q
(4.4)
i , . . . , Fj , . . . , Fq ,
Fi , Fj , F0 , . . . , F
0i< jq
for all F0 , . . . , Fq A . In order to establish that q (Q) is indeed a skewsymmetric
multiderivation of A and that q+1 q = 0, for q N, we relate the coboundary
operator to the Schouten bracket. In fact, the explicit formula (3.36) for the Schouten
bracket [ , ]S , implies that
q (Q) = [Q, ]S ,
(4.5)
for Q Xq (A ). Thus, q (Q) is the Schouten bracket of two skewsymmetric multiderivations, so it is itself a skewsymmetric multiderivation. In order to see that
q+1 q = 0, we write down the graded Jacobi identity of the Schouten bracket
(Proposition 3.7) for the triple (Q, , ),
4.1 Lie Algebra and Poisson Cohomology
95
(1)q1 [Q, [ , ]S ]S [ , [Q, ]S ]S + (1)q1 [ , [ , Q]S ]S = 0 ,
(4.6)
where we have used that the shifted degree q of a qderivation is q 1, in particular
the shifted degree of is 1. Since [ , ]S = 0 (because is a Poisson bracket) and
since [Q, ]S = (1)q [ , Q]S (graded skewsymmetry of [ , ]S ), the identity (4.6)
reduces to [ , [ , Q]S ]S = 0, for every Q Xq (A ), showing that q+1 q = 0, for
all q N. Thus, we have a complex, the Poisson cohomology complex of A ,
q1
Xq1 (A )
Xq (A )
q+1
Xq+1 (A )
The elements of Ker q are called Poisson qcocycles, while the elements of Im q1
are called Poisson qcoboundaries. Elements of the qth Poisson cohomology space
are Poisson qcocycles modulo Poisson qcoboundaries,
Hq (A ) := Ker q / Im q1 ,
for q N and H0 (A ) := Ker 0 . The graded vector space
H (A ) :=
Hq (A ) ,
qN
is called the Poisson cohomology of A .
In the case of a Poisson manifold (M, ), one replaces in the above construction
the multiderivations of A by the multivector fields on M, leading to the following
complex,
q1
Xq1 (M)
Xq (M)
q+1
Xq+1 (M)
where the coboundary operator is defined by (4.5), for Q Xq (M) (a qvector
field on M). This complex is called the Poisson cohomology complex of M. The
qth Poisson cohomology space of (M, ), respectively the Poisson cohomology of
(M, ), is denoted by Hq (M) respectively H (M).
Remark 4.4. Let (A , , { , }) be a Poisson algebra over F. Forgetting the associative product, (A , { , }) is simply a Lie algebra and it makes sense to consider C (A , A ), with the ChevalleyEilenberg coboundary operator (4.3). This
yields precisely (4.4), so that the Poisson coboundary operator of (A , , { , }) is
the ChevalleyEilenberg coboundary operator, restricted to the multiderivations
of (A , ).
For small q, the Poisson coboundary operator q has a natural interpretation, which
yields a natural interpretation for the Poisson cohomology space Hq (A ). One easily
reads off from (4.4) that for F X0 (A ) = A and for V X1 (A ),
0 (F) = XF ,
1 (V ) = LV ,
(4.7)
96
4 Poisson (Co)Homology
where we recall that XF is the Hamiltonian derivation associated to F A (see
Definition 1.3), and LV denotes the Lie derivative with respect to V (see (3.7)).
Also, for Q X2 (A ) and F, G, H A ,
2 (Q)[F, G, H] = {F, Q[G, H]} + Q[F, {G, H}]+ (F, G, H) .
It follows from (4.7) that Poisson 0cocycles are Casimirs,
H0 (A ) = Cas(A ) ,
Poisson 1cocycles are Poisson derivations and Poisson 1coboundaries are Hamiltonian derivations. Denoting the space of all Poisson derivations of A by P(A ), it
follows that
P(A )
.
H1 (A ) =
Ham(A )
Poisson 2cocycles Q X2 (A ) are skewsymmetric biderivations which satisfy
[ , Q]S = 0, i.e., they are the ones which are compatible with { , }, see Section 3.3.2;
Poisson 2coboundaries are biderivations, obtained as a Lie derivative of the Poisson
structure. It follows that
H2 (A ) =
skewsymmetric biderivations compatible with { , }
.
Lie derivatives of { , }
As we will see in Section 13.2, H2 (A ) and H3 (A ) show up naturally in deformation theory.
There is a natural morphism from de Rham cohomology to Poisson cohomology,
which we explain here in the algebraic context; this morphism is defined in the same
way in the case of a smooth manifold. We first construct an A linear map
: 1 (A ) X1 (A )
G dF
GXF .
(4.8)
It is clear that this map is welldefined; for example, we have for all F, G A that
(d(FG) FdG GdF) = XFG FXG GXF = 0 ,
where we used (3) of Proposition 1.4 in the last step. The map extends naturally
to an A linear map
: (A ) X1 (A ) X (A ) ,
which is explicitly given by
q (FdG1 dGq ) = FXG1 XGq ,
4.2 Lie Algebra and Poisson Homology
97
for all F, G1 , . . . , Gq A . It leads to a natural morphism between the de Rham
cohomology and the Poisson cohomology of (A , ).
Proposition 4.5. Let (A , , { , }) be a Poisson algebra. The maps q : q (A )
Xq (A ) define a chain map from ( (A ), d) to (X (A ), ), i.e., for every q N,
the following diagram is commutative:
q (A )
q+1 (A )
q
q+1
q
Xq (A )
Xq+1 (A )
q
It leads for every q N to an Flinear map HdR
(A ) Hq (A ).
Proof. Let := G dF1 dFq q (A ). On the one hand,
q+1 (d ) = q+1 (dG dF1 dFq ) = XG XF1 XFq .
On the other hand, (4.5) and (3.42) imply that
q (q ( )) = q (GXF1 XFq ) = GXF1 XFq , S
= [G, ]S XF1 XFq = XG XF1 XFq ,
where we also used that each Hamiltonian vector field XFi is a Poisson 1cocycle,
[XFi , ]S = 0. This leads to the commutativity of the above diagram. It follows that
q sends qcocycles to qcocycles and sends qcoboundaries to qcoboundaries.
Thus, it induces a map in cohomology.
Remark 4.6. Contrary to what we will see for Poisson homology in Section 4.2,
Poisson cohomology is not a functor: a homomorphism : A B between two
Poisson algebras does not lead in general to a homomorphism between the spaces
of all multiderivations of A and B, not even between their corresponding Poisson
cohomology spaces. In the case of manifolds, for example, it is well known that
vector fields cannot be transferred, in either direction, by a smooth map, which is
not a diffeomorphism.
4.2 Lie Algebra and Poisson Homology
A Poisson bracket on a commutative associative algebra A also leads to homology
spaces. In special cases they are isomorphic to the Poisson cohomology spaces, but
in general they define new invariants for a Poisson algebra. The Poisson homology
spaces have fewer immediate applications than the Poisson cohomology spaces, but
have the advantage of being sometimes easier to compute and have better functorial
98
4 Poisson (Co)Homology
properties, as we will see. As in the previous section, we start with the Lie algebra
case.
4.2.1 Lie Algebra Homology
The basic setup of Lie algebra homology is the same as in the case of Lie algebra
cohomology: (g, [ , ]) is a Lie algebra and ( ,V ) is a representation of g. For k N,
we define
Ck (g;V ) := V k g ,
whose elements are called V valued kchains of g. Note that C0 (g,V ) V , since
0 g = F. The Lie boundary operator kL : Ck (g;V ) Ck1 (g;V ) is the linear map
given, for all v V and x1 , . . . , xk g, by:
kL v (x1 xk )
k
= (1)i xi v x1 xi xk
i=1
(1)i+ j v [xi , x j ] x1 xi xj xk
1i< jk
L L = 0 as a consequence of the Jacobi identity
and 0L := 0. We have that k1
k
for [ , ] and the Lie homomorphism property for . The operators kL lead to a
homology, called the V valued Lie algebra homology of g. Namely, the kth V valued Lie algebra homology space of g is denoted and defined by:
L
.
HkL (g;V ) := Ker kL / Im k+1
We denote by HL (g;V ) the graded vector space
HL (g;V ) :=
HkL (g;V ) .
kN
For v V and x g, we have that
1L (v x) = x v .
Therefore, the subspace Im 1L of V is denoted by gV and H0L (g;V ) = V /gV , since
0L = 0. When V = g and = ad, as in Example 4.3, the boundary operator takes
the following form
4.2 Lie Algebra and Poisson Homology
99
kL x0 (x1 xk )
k
x1 xi xk
(1)i+ j x0 [xi , x j ] x1 xi xj xk .
(1)i [xi , x0 ]
i=1
(4.9)
1i< jk
In this case, gV = [g, g], so that H0CE (g) := H0L (g; g) = g/[g, g]. This space also
appears when one computes H1L (g; F), where : g F is the trivial representation,
as in Example 4.2. Indeed, in this case,
Ker 1L = C1 (g; F) = {a x  a F and x g} g ,
Im 2L = {a [x, y]  a F and x, y g} [g, g] .
This leads to H1L (g) := H1L (g; F) g/[g, g].
4.2.2 Poisson Homology
A Poisson bracket = { , } on a commutative associative algebra A leads to
a homology, called its Poisson homology. The kchains which define the Poisson homology complex are the Kahler differentials on A (the differential forms
when A = F (M), the algebra of smooth functions on a manifold M). In analogy
with (4.9), the Poisson boundary operator : (A ) 1 (A ) is given by
k (F0 dF1 dFk )
k
i dFk
(1)i {Fi , F0 } dF1 dF
i=1
i dF
j dFk ,
(1)i+ j F0 d Fi , Fj dF1 dF
1i< jk
where F0 , . . . , Fk A . For k = 0, this formula should be read as 0 (F0 ) := 0. The
main properties of follow from the fact that = L = [ , d], the Lie derivative
with respect to , as stated in the following proposition.
Proposition 4.7. Let (A , , ) be a Poisson algebra. Then the linear map , defined above, is given by the Lie derivative
= L = [ , d] = d d .
It commutes (in the graded sense) with d and with ,
[ , d] = d + d = 0 ,
(4.10)
[ , ] = = 0 ,
(4.11)
and satisfies = 0 (i.e., it is a boundary operator).
100
4 Poisson (Co)Homology
Proof. For the biderivation = { , }, the internal product is, according to (3.29),
explicitly given by
(F0 dF1 dFk )
i dF
j dFk ,
(1)i+ j+1 F0 Fi , Fj dF1 dF
1i< jk
where F0 , . . . , Fk A , so that
d(F0 dF1 dFk )
=
i . . . dF
j dFk ,
(1)i+ j+1 Fi , Fj dF0 dF
0i< jk
and
d (F0 dF1 dFk )
i . . . dF
j dFk ,
(1)i+ j+1 Fi , Fj dF0 dF
1i< jk
i . . . dF
j dFk .
(1)i+ j+1 F0 d Fi , Fj dF1 dF
1i< jk
Taking the difference of these two formulas leads at once to = [ , d] = L . The
properties of the Lie derivative (Proposition 3.11) imply that d + d = 0,
and that
[ , ] = [L , ] = [ , ]S = 0 ,
where we used in the last equality that is a Poisson structure. This yields (4.10)
and (4.11). Similarly, since L has degree 1,
2 L L = [L , L ] = L[ , ]S = 0 ,
which shows that = L is a boundary operator.
According to the proposition, we have a complex
k+1 (A )
k+1
k (A )
k1 (A )
k1
whose homology is called the Poisson homology of (A , , ). Namely, we define the
kth Poisson homology space
,
Hk (A ) := Ker k / Im k+1
Poisson
and we call elements of Ker k Poisson kcycles and elements of Im k+1
kboundaries. Putting the Poisson homology spaces together leads to the graded
vector space H (A ) := kN Hk (A ). In the case of a Poisson manifold (M, ),
4.3 Operations in Homology and Cohomology
101
Poisson homology is defined similarly, using the differential forms on M, rather
than the Kahler forms on A . It is denoted by H (M).
For example, H0 (A ) is the abelianization of A : since Ker 0 = A while
Im 1 = (A , A ) = {A , A }, we have that
H0 (A ) =
A
.
{A , A }
For the higher Poisson homology spaces, no simple interpretation is known.
Remark 4.8. Unlike Poisson cohomology, Poisson homology is a (covariant) functor. Indeed, every algebra homomorphism : (A , ) (B, ) extends to a degree
zero map ( ) : (A ) (B), which commutes with d. For k N, it is given
by k ( )(F0 dF1 dFk ) = (F0 )d (F1 ) d (Fk ), where F0 , . . . , Fk A .
If A and B are equipped with Poisson brackets A and B , and is a morphism of Poisson algebras, then ( ) A = B ( ), so that ( ) A =
B ( ). Thus, there is an induced map H ( ) : H (A ) H (B), which is
a homomorphism of graded vector spaces. Clearly, H has all properties which are
required for a (covariant) functor.
4.3 Operations in Homology and Cohomology
We have introduced on the graded vector space X (A ) (where A is a commutative
associative algebra) two multiplications, the wedge product , which is associative,
graded commutative and the Schouten bracket [ , ]S , which is a graded Lie bracket
(with a grading, shifted by 1). As we discussed in Section 3.3.2, these two products make (X (A ), , [ , ]S ) into a Gerstenhaber algebra. Moreover, we have two
natural operations (the internal product) and L (the Lie derivative) of X (A ) on
the module of Kahler differentials (A ). We will show in this section that these
four operations induce similar operations, with the same algebraic properties, in
(co)homology. The reader will have no difficulty in transcribing these operations
(and their properties) for the case of (real or complex) manifolds.
We first define the wedge product and the Schouten bracket in cohomology. The
graded Leibniz identity and the graded Jacobi identity for the Schouten bracket,
[Q R, P]S = [Q, P]S R + (1) pq Q [R, P]S ,
(1) pr [P, [Q, R]S ]S + (1)qp [Q, [R, P]S ]S + (1)rq [R, [P, Q]S ]S = 0 ,
become, for P = , Q Xq (A ) and R Xr (A ), with = [, ]S ,
(Q R) = (Q) R + (1)q Q (R) ,
([Q, R]S ) = [Q, (R)]S + (1)r [ (Q), R]S .
102
4 Poisson (Co)Homology
First, it follows from these formulas that if Q and R are Poisson cocycles, then Q R
and [Q, R]S are Poisson cocycles. This means that both operations can be restricted to
cocycles. Second, it follows from the same formulas that either product of a Poisson
cocycle with a Poisson coboundary is a Poisson coboundary. This means that, for
Poisson cocycles Q and R, the cohomology class of Q R and the cohomology
class of [Q, R]S are independent of the representatives Q and R of their cohomology
classes. Thus, we have two welldefined products in Poisson cohomology, which we
also denote by and [ , ]S . Obviously, their algebraic properties are the same as on
X (A ) and we obtain:
Proposition 4.9. For every Poisson algebra (A , , ), (H (A ), , [ , ]S ) is a Gerstenhaber algebra.
We now consider the action of Poisson cohomology on Poisson homology, which
comes from the internal product and from the Lie derivative. Remembering that
the Poisson boundary operator is the Lie derivative L , we take from Proposition 3.11 the formulas which compute the graded commutator of a Lie derivative
with an internal product, or with another Lie derivative, namely,
[LP , Q ] = [P,Q]S
and
[LP , LQ ] = L[P,Q]S ,
which can be written, for P = and (A ), as
(1)q (Q ) = Q ( ) (Q) ,
(1)q (LQ ) = LQ ( ( )) + L (Q) .
It follows from these formulas, as above, that and L induce indeed actions of the
Poisson cohomology of A on the Poisson homology of A .
4.4 The Modular Class of a Poisson Manifold
The modular class of a Poisson manifold M is a Poisson cohomology class (element
of H1 (M)) which is canonically associated to the Poisson structure. We will not
give the description in the most general case and restrict ourselves to the case of
an orientable (real) Poisson manifold. The definitions and arguments remain valid
in the case of Cd , equipped with its algebra of holomorphic functions, but some
minor adaptations are needed for the affine space Fd , equipped with its algebra
of polynomial functions: see Remark 4.13 below. Our exposition is based on the
following:
Question: Let M be a Poisson manifold. Does there exist a volume form on M,
preserved by all Hamiltonian flows?
We call a Poisson manifold, which admits such a volume form, a unimodular Poisson manifold; the Poisson structure is then called a unimodular Poisson structure.
4.4 The Modular Class of a Poisson Manifold
103
For a symplectic1 manifold (M, ), the answer to the above question is yes. Indeed, since the symplectic structure is preserved by all Hamiltonian flows, the
same is true for the Liouville form d on M, where dim M = 2d. Thus, every symplectic manifold is a unimodular Poisson manifold. Many properties of symplectic
manifolds generalize to the case of unimodular Poisson manifolds. For example, for
unimodular Poisson manifolds the Poisson cohomology and the Poisson homology
spaces are isomorphic to each other (Theorem 4.18).
4.4.1 The Modular Vector Field
We first list a few basic facts about volume forms on a real orientable manifold
M. Suppose that is an arbitrary volume form on M. Recall that this means that
is a differential dform on M, where d := dim M, vanishing at no point of M,
and that such a form exists on a manifold if and only if the manifold is orientable;
moreover, two such forms differ only by a factor, which is a nowhere vanishing
smooth function on M. We have that d = 0 and that dF = 0, for every F
F (M). Combined with (3.32), this yields
(F P) = (1) p dF P ,
(4.12)
for all P X p (M). In particular,
V [F] = dF V ,
(4.13)
for every vector field V X1 (M). Also, if is a Poisson structure on M, then (4.12)
specializes to
(4.14)
XF = dF .
The Lie derivative of with respect to some vector field is again a top form, hence it
is a multiple of . For F F (M), let us denote by DF F (M) the smooth function
on M defined by LXF = DF . We show that the map F
DF is a derivation.
Indeed, for all F, G F (M), we have that
DFG = LXFG = LGXF +FXG
= GLXF + {G, F} + FLXG + {F, G}
= (GDF + FDG ) ,
where the identity
LHV = HLV + V [H]
(4.15)
for all H F (M), V X1 (M) has been used (to prove (4.15), write LV = V d +
d V and use (4.13)). Since D : F
DF is a derivation of F (M), it corresponds to
a vector field.
1
See Section 6.3.
104
4 Poisson (Co)Homology
Definition 4.10. Let (M, ) be a real Poisson manifold, equipped with a volume
form . The vector field , defined by
LXF = [F] ,
(4.16)
for all F F (M), is called the modular vector field of with respect to .
When the chosen volume form is clear from the context, we simply write for .
Notice that the modular vector field can also be defined as the unique vector field
on M, satisfying
(4.17)
L = .
This is seen by writing [F] in two different ways. On the one hand, (4.13) tells
us that [F] = dF , for all F F (M). On the other hand, (4.16), (4.14) and
the formula L = d d imply that
[F] = LXF = d(XF ) = d(dF ) = dF d = dF L ,
for all F F (M). It follows that dF = dF L for every F F (M), which
leads to (4.17).
In view of Definition 4.10, the question which we posed at the beginning of this
section now becomes:
Question: Let M be a Poisson manifold. Does there exist a volume form on M,
with respect to which the modular vector field is zero?
By giving a cohomological interpretation of the modular vector field, we will be
able to answer the question in the next section.
4.4.2 The Modular Class
The modular vector field which we constructed in the previous section depends on
the choice of volume form on M. We will now show that when the volume form is
changed, the modular vector field will be modified by a Hamiltonian vector field.
We also show that the modular vector field is a Poisson vector field, so that it defines
a Poisson cohomology class which, by the above, is independent of the choice of
volume form.
Proposition 4.11. Let (M, { , }) be a real Poisson manifold, which is orientable.
For a volume form on M, the modular vector field has the following properties:
(1)
(2)
(3)
is a Poisson vector field, i.e., it is a Poisson 1cocycle;
The Poisson cohomology class of is independent of the chosen volume
form ;
If is a Hamiltonian vector field, i.e., if the Poisson cohomology class of
is zero, then there exists a volume form , with respect to which the modular
vector field is zero (and therefore is preserved by all the Hamiltonian flows).
4.4 The Modular Class of a Poisson Manifold
105
Proof. In order to prove (1) we need to verify that [{F, G}] = { [F], G} +
{F, [G]}, for all F, G F (M). The basic properties of the Poisson bracket and
the Lie derivative yield:
[{F, G}] = LX{F,G} = L[XG ,XF ] = LXG , LXF
= LXG ( [F] ) LXF ( [G] )
= ({ [F], G} + {F, [G]}) ,
where we have used in the last step that
[F]LXG = [F] [G] = [G]LXF .
This proves (1). Let us consider now another volume form on M and let us denote
by and the modular class of { , } with respect to , respectively . There
exists a nowhere vanishing function F (M) such that = . Using that and
are both top forms and (4.13), we compute, for an arbitrary F F (M),
LXF ( ) = XF [ ] + LXF = X [F] + [F] ,
to conclude that
1
[F] = [F] X [F] ,
for all F F (M), and hence that
1
= X = Xln   .
This shows that the vector fields and differ by a Hamiltonian vector field,
i.e., a Poisson 1coboundary. The cohomology class of the modular form is therefore
independent of the chosen volume form, which proves (2). The previous calculation
shows that if this cohomology class is trivial, i.e., = XF for some F F (M),
then := exp(F) is a volume form whose modular vector field is zero.
Summarizing, we have a complete answer to our question:
Answer: The Poisson manifold (M, ) admits a volume form, preserved by all
Hamiltonian flows if and only if its modular class is zero.
It leads to the following definition.
Definition 4.12. Let (M, { , }) be a real Poisson manifold, which is assumed orientable. The cohomology class of the modular vector field (with respect to an arbitrary volume form on M) is called the modular class of the Poisson manifold. If
this cohomology class is trivial, then we say that (M, { , }) is a unimodular Poisson
manifold. A volume form on M, whose modular vector field is zero, is called an
invariant density.
106
4 Poisson (Co)Homology
Notice that if one multiplies an invariant density by a nowhere vanishing Casimir
function, then one obtains another invariant density; up to this ambiguity, the invariant density on a unimodular Poisson manifold is unique.
Remark 4.13. In the case of a finitedimensional vector space V , a natural volume
form is given, in terms of linear coordinates x1 , . . . , xd on V , by 0 := dx1 dxd ,
and every top form on V is of the form F 0 , where F is an arbitrary function on V .
Up to a nonzero constant, 0 is the only translation invariant volume form on V . If
F is assumed to be algebraically closed, then 0 is, again up to a nonzero constant,
the unique volume form on V , because every nonconstant polynomial F has a nonempty zero locus, on which the top form F 0 fails to be a volume form. Notice that,
since the modular vector field does not change when the volume form is multiplied
by a constant, the modular vector field of a Poisson structure is independent of
the chosen volume form, when only translation invariant volume forms on V are
considered.
We show in the following proposition that the obstruction to the existence of an
invariant density is intimately related to the existence of points where the rank of
the Poisson structure drops: in the neighborhood of each regular point, the modular
vector field is Hamiltonian, hence an invariant density exists on this neighborhood.
Proposition 4.14. Let (M, { , }) be a real orientable Poisson manifold and let be
a volume form on M.
(1)
(2)
The modular vector field is a Hamiltonian vector field in a neighborhood
of every point at which the rank is locally constant. In particular, it is tangent
to all symplectic leaves of maximal dimension;
The modular class is preserved under Poisson diffeomorphisms M M.
Proof. Let m M be a point where the rank of { , } is locally constant. According
to the Darboux theorem (Theorem 1.26), there exist splitting coordinates q1 , . . . , qr ,
p1 , . . . , pr , z1 , . . . , zs on a neighborhood U of m, such that
,
pi
i=1 qi
r
{ , } =
on U, where Rkm { , } = 2r. We take on U as volume form
:=
r
(dqi dpi )
i=1
and we denote
k :=
i=k
s
dz j ,
j=1
(dqi dpi )
s
dz j ,
j=1
for k = 1, . . . , r. For such k and for F F (M) we have
4.4 The Modular Class of a Poisson Manifold
107
LXF (dqk dpk ) = LXF dqk dpk + dqk LXF dpk
F
F
dpk dqk d
.
=d
pk
qk
Therefore, for all k = 1, . . . , r,
k LXF (dqk dpk ) =
2F
2F
qk pk pk qk
=0,
from which it follows that LXF = 0, i.e., the modular vector field is zero on U.
Then (2) of Proposition 4.11 implies that all modular vector fields are Hamiltonian
on U, hence they are tangent on U to all symplectic leaves. The second property is
obvious, since the modular class depends on { , } only.
Example 4.15. If m is a point where the rank of the Poisson structure is not locally
constant, then it may happen that there does not exist a neighborhood of m on which
the modular vector field is Hamiltonian. In fact, the modular vector field may be
nonvanishing at a point where the rank of the Poisson structure vanishes. Consider
for example the Poisson structure { , } = x x y on R2 , equipped with the volume
form dx dy. The rank at the origin is zero, while =
class of this Poisson structure is not zero.
y . In particular, the modular
4.4.3 The Divergence of the Poisson Bracket
We now show that the modular vector field is (up to a sign) the divergence of the
Poisson structure (with respect to ). The divergence operator (with respect to ) is
the graded Flinear map (of degree 1),
Div : X (M) X1 (M) ,
which makes the following diagram commute
X (M)
d (M)
Div
X1 (M)
(4.18)
d+1 (M)
where d := dim M. The diagram involves the star operator , which is the family of
F (M)linear isomorphisms : Xk (M) dk (M), defined for Q Xk (M) by
Q := Q .
(4.19)
108
4 Poisson (Co)Homology
Since is a top form, we have for every vector field V on M that
LV = dV = d V = Div(V ) = Div(V ) .
(4.20)
For example, taking the standard volume form on Rd , i.e., = dx1 dxd
the star of a vector field V = di=1 Fi / xi , respectively of a bivector field P =
i< j Pi j / xi / x j is given by
V =
i dxd ,
(1)i1 Fi dx1 dx
i=1
P =
i dx
j dxd ,
(1)i+ j1 Pi j dx1 dx
1i< jd
so that Div(V ) is given by the classical formula
Fi
,
i=1 xi
d
Div(V ) =
(4.21)
and Div(P) is given by
Div(P) =
1i< jd
Pi j
Pi j
x j xi
xi x j
.
(4.22)
In general, the divergence of a bivector field, written in terms of vector fields, can
be computed using the following proposition.
Proposition 4.16. Let be a volume form on an orientable manifold M and let V
and W be two vector fields on M. The divergence of V W with respect to is
given by
Div(V W ) = Div(W )V Div(V )W [V , W ] .
(4.23)
Proof. According to the definition of the (generalized) Lie derivative (3.50) and its
properties (4) and (2) in Proposition 3.11, we have that
dV W = LV W = LW V W LV = V LW W LV [V ,W ] .
According to the definition of the divergence, this leads to
Div(V W ) = 1 dV W
= 1 V LW 1 W LV 1 [V ,W ]
= Div(W )V Div(V )W [V , W ] ,
where we used in the last step that V LW = V (Div(W ) ) = Div(W )V =
(Div(W )V ) and that [V , W ] = [V ,W ] .
We now show that the modular vector field is, up to a sign, the divergence of the
Poisson structure.
4.4 The Modular Class of a Poisson Manifold
109
Proposition 4.17. Let (M, ) be a real Poisson manifold, which is assumed orientable. For every volume form on M,
= Div( ) .
(4.24)
In particular, the divergence of the modular vector field of is zero.
Proof. Let be an arbitrary volume form on M and let F F (M). The definitions
of and of , combined with (4.17), imply that
Div( ) = d = d( ) = L = = .
(4.25)
Since is an isomorphism, this shows that the modular vector field is minus the
divergence of . Since Div Div = 1 d d = 0, the modular vector field is
divergencefree.
According to (4.25), the modular vector field can also be defined by the formula
d = ;
forgetting the stars, the modular vector field is the differential of the Poisson bracket.
The string (4.25) also contains the identity L = , which amounts to
( ) = .
(4.26)
Thus, is always a Poisson 1boundary, while is a Poisson 1coboundary if
and only if the underlying Poisson manifold is unimodular.
4.4.4 Unimodular Poisson Manifolds
We now prove that, for unimodular Poisson manifolds, the Poisson homology and
cohomology spaces are isomorphic. Let (M, ) be a ddimensional Poisson manifold. We first show how the Poisson boundary and coboundary operators are related
via the modular vector field . To do this, we specialize Cartans formula (3.38)
[[P , d] , Q ] = [P,Q]S ,
valid for P X p (M) and Q Xq (M), to the case P = , giving
Q (1)q Q = (1)q1 (Q) .
(4.27)
If we apply (4.27) to and use (4.26), then we find
(Q) (1)q Q = (1)q1 (Q) ,
which we also write, using Q P = PQ , valid for P, Q X (M) (see Proposition 3.4), as follows,
110
4 Poisson (Co)Homology
( Q) = (Q) + (1)q ( Q) .
(4.28)
The latter formula says that the modular vector field measures the noncommutativity of the following diagram
X (M)
X+1 (M)
d (M)
(4.29)
d1 (M)
The reader is invited to compare the diagrams (4.29) and (4.18), keeping in mind
that both depend on the volume form , but only the second one depends on the
Poisson structure . Equation (4.28) leads at once to the following duality theorem.
Proposition 4.18. For every unimodular Poisson manifold (M, { , }) of dimension
d, the Poisson cohomology spaces and Poisson homology spaces are isomorphic:
(M),
Hk (M) Hdk
0kd.
(4.30)
Proof. Let be an invariant density on M, so that = 0. Then (4.28) implies that
the isomorphism : X (M) d (M) is a bijection between the Poisson cocycles
(respectively coboundaries) of M and the Poisson cycles (respectively boundaries)
of M. It follows that induces isomorphisms, as indicated in (4.30).
Notice that the isomorphism in (4.30) is not canonical, since there is no canonical
invariant density (in general; symplectic manifolds form an important exception).
Remark 4.19. The proposition is also valid for Fd , equipped with a Poisson structure, in case it is a unimodular Poisson variety (see Remark 4.13).
4.5 Exercises
1. Show that does not induce a product in Poisson homology (Hint: compute
(dF dG) for F, G A ).
2. Show that the Poisson cohomology spaces are modules over the algebra of
Casimirs.
3. Find a formula for the Poisson cohomology of the tensor product of two Poisson
algebras.
4. Prove property (2) of Proposition 4.11 by using the star operator (4.19).
5. Suppose that g is a Lie algebra, equipped with a symmetric bilinear form 
which is adinvariant, i.e., [x, y]  z = x  [y, z] for all x, y, z g. Consider the ele
4.6 Notes
111
ment C Hom(3 g, F), defined by C(x, y, z) := [x, y]  z. Show that C is a 3cocycle
in the trivial Lie algebra cohomology.
6. Consider the Poisson structure on F2 given by
:= x
.
x y
a.
b.
c.
Show that the modular class of is not trivial;
Compute the Poisson homology and the Poisson cohomology of ;
Conclude from part b. that the conclusion of Proposition 4.18 does not hold for
(F2 )
the Poisson manifold (F2 , ), i.e., that the vector spaces Hk (F2 ) and H2k
are not isomorphic, for k = 0, 1, 2.
7.
Let (A , , { , }) be a Poisson algebra.
a.
Show that the skewsymmetric biderivation { , } is a Poisson 2cocycle for the
Poisson cohomology of A ;
Assume that A
is a polynomial algebra, A = F[x1 , . . . , xd ], and that the polynomials xi , x j , for i, j = 1, . . . , d, are all homogeneous of the same degree k.
Show that if k = 2, then the skewsymmetric biderivation { , } is a coboundary
(Hint: compute 1 (E ) where E := di=1 xi xi );
Assume now that A := C[x, y, z], and that
b.
c.
:= x2
+ y2
+ z2
.
y z
z x
x y
Show that is a Poisson bracket on A := C[x, y, z] which is not a coboundary
(Hint: show first that if there exists a derivation V = F1 x + F2 y + F3 z
X1 (A ) satisfying 1 (V ) = , then there exists such a derivation where F1 , F2
and F3 are linear functions).
4.6 Notes
Poisson cohomology was first introduced in Lichnerowiczs paper [127], where it
was shown that the Poisson cohomology of a symplectic manifold is canonically
isomorphic to its de Rham cohomology. See BhaskaraViswanath [23] and Huebschmann [96] for the algebraic cohomology of a general Poisson algebra. The relation between Poisson cohomology and singularities is studied by Monnier [152]
and RogerVanhaecke [174] in the twodimensional case and by Pichereau [166]
in the threedimensional case. The definition of Poisson homology goes back to
GelfandDorfman [81], Koszul [118] and Brylinski [28]. The modular vector field
of a Poisson manifold was first introduced by Koszul [118]. It was studied for general Poisson manifolds and extended to Lie algebroids by Weinstein [201], Evans
LuWeinstein [71] and KosmannWeinstein [116]. See FernandesCrainic [51, 52]
112
4 Poisson (Co)Homology
for applications of Poisson cohomology to the linearization and stability of Poisson structures; for applications of Poisson cohomology to deformation theory, see
Chapter 13 below.
Chapter 5
Reduction
In Chapter 2, we have given a few basic constructions which allow one to build
new Poisson structures from given ones. In the present chapter, we explain a few
more advanced constructions, which all fall under the general concept of reduction. Roughly speaking, reduction means that the object under study (here a Poisson
structure), is replaced by an object of the same type, but on a manifold of smaller
dimension, in classical terms with fewer degrees of freedom. These constructions
are based on geometrical considerations, yet their algebraic formulation highlights
some key elements of the reduction mechanism. We therefore present all constructions in both algebraic and geometrical terms.
We present two constructions for reducing Poisson structures. Geometrically
speaking, the first one, called Poisson reduction, deals with Poisson structures on
quotients of Poisson manifolds, or, more generally on quotients of (coisotropic) submanifolds of Poisson manifolds. Poisson reduction is presented in Section 5.2. The
second construction, called PoissonDirac reduction, concerns Poisson structures
on submanifolds, which are not necessarily Poisson submanifolds; a typical example is a submanifold which is transverse to a symplectic leaf, leading to the notion
of transverse Poisson structure. PoissonDirac reduction is discussed in Section 5.3.
As an application of these constructions, we consider in Section 5.4 Poisson varieties and Poisson manifolds on which a group acts; we show how a Poisson structure is inherited on the quotient space, the subspace of fixed points of the action, and
on the reduced space of the socalled momentum map. For the convenience of the
reader, and in order to fix both our notations and conventions, the basic facts on Lie
groups, and their relation with Lie algebras are recalled in the first section of this
chapter.
As in the previous chapters, all Poisson algebras are defined over an arbitrary
field F of characteristic zero, all Poisson varieties are affine varieties over F and the
Poisson manifolds are real or complex manifolds.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 5, SpringerVerlag Berlin Heidelberg 2013
113
114
5 Reduction
5.1 Lie Groups and (Their) Lie Algebras
We recall in this section the natural correspondence between Lie groups and Lie algebras, we give the basic definitions and properties of group actions and we explain
construction of invariant (multi)vector fields on a Lie group.
5.1.1 Lie Groups and Lie Algebras
A Lie group is a set G endowed with two structures, a group structure and a manifold
structure, which are required to be compatible, in the sense that the product map (the
group structure)
: G G G : (g, h) gh
is assumed to be smooth (or holomorphic). Then the inverse map
inv : G G : g g1
is also smooth (or holomorphic). The product of two elements g, h of G is denoted
by (g, h) or simply by gh, and the unit of G is denoted by e; as a manifold, G
may be either real or complex. The identity map on G is denoted by 1G . Given two
Lie groups G and H, a Lie group morphism from G to H is a group homomorphism
: G H, which is a smooth (or holomorphic) map. If G is contained in H and the
inclusion map is a Lie group morphism, then G is said to be a Lie subgroup of H.
The multiplication in G leads to three, a priori different, sets of diffeomorphisms
of G: given g G, we define left translation, right translation and conjugation by
g G respectively as follows:
Lg : G G : h
gh ,
Rg : G G : h
hg ,
Cg : G G : h ghg1 .
The tangent maps of and inv can be expressed in terms of these diffeomorphisms:
since (g, h) = Lg (h) = Rh (g) for all g, h G, the tangent map of at (g, h) is given
by
T(g,h) (v, w) = Tg Rh (v) + Th Lg (w)
(5.1)
for all (v, w) T(g,h) (G G) Tg G Th G. Differentiating the identities Lg Lg1 =
1G at e and (g, inv(g)) = e at g G, leads, upon using (5.1), to the following
expression for the tangent map of inv at g G,
Tg inv = Te Lg1 Tg Rg1 .
(5.2)
5.1 Lie Groups and (Their) Lie Algebras
115
Notice that, in particular, Te inv = 1Te G . Similarly, one computes from Cg = Lg
Rg1 , valid for g G, that the tangent map of the conjugation map Cg is given, at
the unit e, by
TeCg = Tg1 Lg Te Rg1 .
(5.3)
Since for a fixed g G, left translation by g is a diffeomorphism of G, which
sends e to g, we can identify the tangent spaces Te G and Tg G by using the linear
x on G by
map Te Lg . For a fixed x Te G, this allows us to define a vector field
setting, for all g G:
x g := Te Lg (x) .
The same can be done using right translation by g, yielding a vector field
x on G,
defined for all g G by
x g := Te Rg (x) .
It is clear that the vector fields x and
x are determined by their value, x, at e.
This fact is closely related to an invariance property which we now define. A vector
field V on G is said to be leftinvariant (respectively rightinvariant) if Th Lg (Vh ) =
Vgh , for all g, h G (respectively, Tg Rh (Vg ) = Vgh , for all g, h G). Each one of the
vector fields
x , where x Te G, is leftinvariant, since
x h ) = Th Lg (Te Lh (x)) = Te (Lg Lh )(x) = Te Lgh (x) =
x gh ,
Th Lg (
for all g, h G. Similarly, each of the vector fields
x is rightinvariant. It follows that there is a natural identification between Te G and the vector space of all
leftinvariant vector fields on G (or the vector space of all rightinvariant vector
fields on G). Since the commutator of two leftinvariant vector fields on G is a leftinvariant vector field on G, the tangent space Te G at the unit of a Lie group G has a
natural Lie algebra structure. It leads to the following basic correspondence between
Lie groups and Lie algebras (for a proof, see [66]).
Theorem 5.1 (Lies theorem). Let G and H be Lie groups, whose units are denoted
by e and e respectively, and let : G H be a Lie group homomorphism.
(1)
(2)
The tangent space Te G is equipped with a natural Lie bracket [ , ];
The linear map Te : Te G Te H is a morphism of Lie algebras.
The Lie algebra (Te G, [ , ]) is called the Lie algebra of G and is denoted by (g, [ , ])
(or g, for short). Denoting h := Te H, the linear map Te : g h is called the Lie
algebra homomorphism associated to .
Conversely, let g and h be finitedimensional Lie algebras and let : g h be a
Lie algebra homomorphism.
(3)
(4)
There exists a connected and simply connected Lie group, whose Lie algebra
is isomorphic to g; up to isomorphism, such a Lie group is unique;
Let G be a connected and simply connected Lie group with Lie algebra g,
and let H be a Lie group with Lie algebra h. There exists a unique Lie group
homomorphism : G H, such that is the Lie algebra homomorphism
associated to ;
116
5 Reduction
(5)
Let G be a connected and simply connected Lie group with Lie algebra g, and
let h be a Lie subalgebra of g. There exists a unique connected Lie subgroup H
of G with Lie algebra h.
In a different vein, a Lie group and its Lie algebra are related as follows: there exists
a diffeomorphism exp between a neighborhood of the origin o in g and a neighborhood of the unit e in G, having the fundamental property that for every x g, the
integral curve of
x which starts from o is given (for small t) by t exp(tx). In
fact, one possible construction of the diffeomorphism exp is by integrating, for x in
a neighborhood of o in g, the leftinvariant vector fields
x . We refer to exp as the
exponential map.
We end this section with a few comments on algebraic groups, which for us will
always be complex (defined over C). By definition, a (complex) algebraic group is
an algebraic variety with a group structure, such that the product map and the inverse
map are morphisms (of algebraic varieties). Much of what we discussed in this
section carries over to algebraic groups, for example there is canonically associated
to every algebraic group G a Lie algebra, whose underlying vector space g is the
(Zariski) tangent space to G at the unit e of G. However, there is in general no
algebraic analog of the exponential map and parts (3) and (4) of Lies theorem do
not hold in general. When dealing with quotients of algebraic varieties by algebraic
group actions, we will often assume that the algebraic group is reductive, which
means that all its finitedimensional representations are completely reducible; as we
will explain at the end of Section 5.1.2, such quotients are rather wellbehaved.
5.1.2 Lie Group Actions
Given a Lie group G and a manifold M, a group action : G M M is called
a Lie group action, if is smooth (or holomorphic); since the only actions of Lie
groups on manifolds which are considered here are Lie group actions, we will simply call them group actions. When the action is clear, we write gm for (g, m),
where (g, m) G M. Clearly, a group action of G on M induces a group action
of G on F (M), the algebra of smooth (or holomorphic) functions on M, by defining
for g G and F F (M)
(g F)(m) := F(g1 m) ,
(5.4)
for all m M. The virtue of the latter group action is that it is a representation
of G (on F (M)); a representation of a Lie group G on a vector space V is a group
action : G V V with the property that for every g G the map g : V V ,
defined for all x g by g (x) := (g, x), is a linear map (an endomorphism of V ).
The action of an algebraic group on an algebraic variety is defined in an analogous
way, demanding now that the map is a morphism of algebraic varieties.
The main group actions which we will consider in this chapter are the group
actions of G on itself, given by
5.1 Lie Groups and (Their) Lie Algebras
L : G G G : (g, h)
117
gh ,
R : G G G : (g, h) hg1 ,
C : G G G : (g, h) ghg1 .
These actions are called left translation, right translation and conjugation (in that
order). Notice that taking the inverse of g in the definition of R is essential to make
it into an action (otherwise it would be a socalled right action); we stress that it
amounts to R(g, h) = Rg1 (h) for all g, h G.
For a fixed group action of G on a manifold M, the fundamental vector field
associated to x Te G, denoted x, is the vector field on M, whose value at m M is
given by
d
F( (exp(tx), m)) ,
(5.5)
xm [F] :=
dt t=0
for all functions F, defined on a neighborhood of m in M. Writing g = exp(tx),
one sees that Eq. (5.5) is the infinitesimal version of (5.4). Applied to the case of
right translation (recall that R(g, h) = Rg1 h = hg1 ), this yields for all functions F,
defined in a neighborhood of g in G,
d
d
F(R(exp(tx), g)) =
F(g exp(tx))
dt t=0
dt t=0
d
=
((F Lg ) exp(tx)) = x[F Lg ]
dt t=0
x g [F] .
= Te Lg (x)[F] =
xg [F] =
x g , for all g G, i.e., x =
x . In words: the leftinvariant
We conclude that xg =
vector field on G, whose value at e is x, is the fundamental vector field of right
translation, corresponding to x. Similarly, the fundamental vector field of left trans
lation on G, corresponding to x g, is given by
x . Also, since Cg = Rg1 Lg ,
the fundamental vector field of conjugation on G, corresponding to x g, is given
by
x
x . The identity
inv Rg = Lg1 inv
implies that a vector field V on G is rightinvariant if and only if the vector field
on G, defined1 by inv V , is leftinvariant. It follows, using Te inv = 1Te G , that
x =
x , for every x g.
inv
Given a group action of a Lie group G on a manifold M, the map g X1 (M),
which associates, as above, to x g the fundamental vector field x on M, admits
as a natural generalization a map g X (M), which associates to a multivector
X of g a multivector field X on M. For X p g of the form X = x1 x p , the
pvector field X is defined as
Recall that for a diffeomorphism : M N between manifolds and for V a vector field on M,
we denote by V the vector field on N, which is defined by ( V ) (m) := Tm (Vm ), for all
m M, and similarly for multivector fields on M.
118
5 Reduction
X := x1 x p
and is called the fundamental pvector field, corresponding to X. In the case of right
and left translations on a Lie group G, we write X , respectively (1) p X for the
p
fundamental multivector field, associated to X g. Equivalently, for X p g, the
vector fields X and X can be defined by X g := p (Te Lg )X and X g := p (Te Rg )X,
for all g G. As above, it follows from this definition that X is a leftinvariant
multivector field and that X is a rightinvariant multivector field.
We show in the following proposition that the map, which sends X g to X ,
2
is a homomorphism of Gerstenhaber algebras.
Proposition 5.2. Let G be a Lie group with Lie algebra g. For all X,Y g, we
have
X Y = X Y ,
[[X,Y ]] = X , Y S ,
(5.6)
so the map X X preserves the wedge product and the Schouten bracket. Up to a
sign, the same is true for the map X X : for all X,Y g, we have
X Y = X Y ,
Moreover,
for all X,Y
[[X,Y ]] = X , Y S .
X,Y S=0,
(5.7)
(5.8)
g.
Proof. Since every element of g is a sum of elements of the form x1 x p ,
where 1 p dim g, and x1 , . . . , x p g, it suffices to prove (5.6)(5.8) for X and Y of
this form. Then the first equation in (5.6) follows from the associativity of on g
and on X (G). The second equation in (5.6) clearly holds when X and Y belong
to g; upon using the graded Leibniz identities (3.42) and (3.47) of the Schouten
brackets [[ , ]] on g and [ , ]S on X (G), it follows that this formula holds for
x =
x , for all x g, we have inv X = (1) p X ,
general X,Y g . Since inv
p
for all X g; therefore, applying inv to both equations in (5.6), we find the
equations in (5.7). We now prove (5.8). In view of the graded Leibniz rule for the
Schouten bracket, it suffices to prove (5.8) when X and Y are elements of g, which
we write as x and y. The vector fields
x and
y are the fundamental vector fields
of the actions of right, respectively left, translation of G on itself. These actions
commute, hence the flows of
x and
y also commute. The commutator [
x ,
y ] is
therefore zero.
For algebraic groups there is, in general, no exponential map, but for linear actions on vector spaces (representations), such as the actions which we will consider in the next section, taking the differential of the action, viewed as a map
: G End(V ), yields a linear vector field on V , which is the analog (up to a
sign) of the fundamental vector field, defined in (5.5).
Recall that ( g, , [[ , ]]) is a Gerstenhaber algebra, just like (X (M), , [ , ]S ), where M is an
arbitrary manifold (see Proposition 3.9).
5.1 Lie Groups and (Their) Lie Algebras
119
To finish this section, we recall some rather general conditions under which the
quotient space M/G of a group action G M M is of the same type as M, i.e., is
an affine variety or a manifold.
(1) Suppose that M is an affine variety on which a reductive algebraic group G acts.
Then the algebra of Ginvariant functions F (M)G is finitely generated, hence is the
algebra of regular functions on an affine variety, which can be identified with the
orbit space M/G. Since, in particular, every finite group G is reductive, the quotient
of an affine variety by a finite group is an affine variety.
(2) Suppose that M is a manifold and that G is a Lie group, which acts properly
and locally freely on M. Then the orbit space M/G has a unique manifold structure
for which the canonical map p : M M/G is smooth. Interesting particular cases
include free actions of a finite group and locally free actions of a compact group on
a manifold. We recall that an action : G M M is called a proper action, if
the map
p2 : G M M M
(g, m) (gm, m)
(5.9)
is a proper map, i.e., the inverse image of every compact subset of M M is a
compact subset of G M. Also, is called a locally free action if, for every m M,
there exists a neighborhood U of the identity in G such that the restriction to U of
the map g gm is injective.
5.1.3 Adjoint and Coadjoint Action
We next consider the adjoint action of a Lie group G on its Lie algebra g, and more
generally on the space g; we will also consider its infinitesimal version, which is
a Lie algebra action of g on itself, and more generally on g. Dually, we will also
consider the corresponding group action of G and the corresponding Lie algebra
action of g on its dual vector space g and on g . The formulas are most naturally
written in terms on the natural pairing
, between p g and p g, given for all
1 , . . . , p g and for all x1 , . . . , x p g by
1 [x1 ] 1 [x p ]
.
.. .
1 p , x1 , . . . , x p := ..
(5.10)
.
p [x1 ] p [x p ]
For g G, the conjugation map Cg : G G is a Lie group automorphism of G.
According to Theorem 5.1, the tangent map to Cg at e is a Lie algebra automorphism
of g. It is customary to denote this map by Adg . Thus, Adg := TeCg : g g and
Adg ([x, y]) = [Adg x, Adg y] ,
120
5 Reduction
for all x, y g. For g, h G, we have that
Adgh = TeCgh = Te (Cg Ch ) = (TeCg ) (TeCh ) = Adg Adh ,
so that the assignment (g, x) Adg x defines a group action of G on g. It is called
the adjoint action of G on g, denoted Ad. Since the maps Adg are linear maps, Ad is
actually a representation of G, the adjoint representation of G on g. The transpose
map of Adg1 is denoted by Adg : for all g and for all x g,
Adg , x = , Adg1 x .
The resulting map (g, ) Adg is a group action of G on g , called the coadjoint action of G on g , denoted Ad . As in the case of the adjoint action, Ad is a
representation of G, the coadjoint representation of G on g.
These adjoint and coadjoint actions extend to group actions of G on g and
on g , also denoted by Ad and Ad , upon setting
Adg (x1 x p ) := Adg x1 Adg x p ,
(5.11)
Adg (1 p ) := Adg 1 Adg p ,
(5.12)
for all g G, all p N , all x1 , . . . , x p g and all 1 , . . . , p g . It follows from
these definitions that
Adg , X = , Adg1 X ,
for all g G, all p g and all X p g.
The tangent map of Ad : G End(g) at e is the linear map
ad := Te Ad : g End(g) ,
called the adjoint representation of g. Writing adx for the image of x g by ad,
it is a fundamental fact that the endomorphism adx is given by adx (y) = [x, y],
for all y g. It follows that ad is indeed a Lie algebra representation: ad[x,y] =
adx ady ady adx , for all x, y g. More generally, taking the tangent map of
Ad : G End( g) at e we obtain the adjoint representation of g on g, still
denoted by ad. Equation (5.11) implies that
p
adx (x1 x p ) = x1 [x, xi ] x p ,
i=1
for all x1 , . . . , x p g hence that adx X = [[x, X]], for all X g. In view of the graded
Jacobi identity for the algebraic Schouten bracket [[ , ]], we have for all x, y g and
for all X g that
ad[x,y] X = [[[x, y] , X]] = [[x, [[y, X]]]] [[y, [[x, X]]]] = adx (ady X) ady (adx X) ,
5.1 Lie Groups and (Their) Lie Algebras
121
so that ad[x,y] = adx ady ady adx , which shows that ad is indeed a Lie algebra representation of g (on g). Similarly, we obtain the coadjoint representation by taking
the tangent map of Ad : G End( g ) at e, which yields ad : g End( g ).
As above, ad is for all x g a derivation of g and adx is related to adx by
adx , X =
, adx X =
, [[x, X]] ,
(5.13)
for all p g and all X p g.
To finish this section, we shortly discuss invariant multivectors. An element X
g is said to be Adinvariant (respectively adinvariant) if Adg X = X, for all g G
(respectively adx X = 0, for all x g). The following lemma relates Ad and adinvariance and gives a useful characterization in terms of the associated infinitesimal
multivector fields.
Proposition 5.3. Let G be a Lie group with Lie algebra g, and let X g.
(1)
(2)
(3)
If X is Adinvariant, then X is adinvariant;
If G is connected, then X is Adinvariant if and only if X is adinvariant;
X is Adinvariant if and only if X = X .
Proof. Let X g. If X is Adinvariant, then the map G g, defined for
g G by g Adg X is a constant map, hence its tangent map is zero, adx X =
Te Ad(x)(X) = 0, for all x g. Conversely, the latter map is locally a constant map
on G when adx X = 0 for all x g; if G is connected, this implies that g Adg X is
a constant map, hence that X is Adinvariant. This proves (1) and (2). For X g,
say X p g, and for g G, the equality X g = X g , which can also be written as
p (Te Lg )X = p (Te Rg )X ,
is equivalent to p Adg X = X, because
p Adg = p (TeCg ) = p (Tg Rg1 ) p (Te Lg ) = p (Te Rg )1 p (Te Lg ) .
Therefore, X g = X g for all g G if and only if X is Adinvariant, which is the
content of (3).
5.1.4 Invariant Bilinear Forms and Invariant Functions
Let G be a Lie group and let g be its Lie algebra. A bilinear form
 on g is called
Adinvariant, if for all g G and for all x, y g,
Adg x  Adg y =
x  y .
If
 is Adinvariant, then
[z, x]  y +
x  [z, y] = 0 ,
122
5 Reduction
for all x, y, z g; a bilinear form on g satisfying the latter condition is called adinvariant. When G is connected, every bilinear form which is adinvariant is Adinvariant, so the two notions of invariance need not be distinguished.
We often use a nondegenerate symmetric bilinear form to identify g with its dual
vector space g . Nondegeneracy of a bilinear form
 means that the linear map
: g g , which sends x g to the linear form
x  : y
x  y, is an isomorphism.
Its inverse will be denoted by . Explicitly, is given by
(x), y =
x  y =
(y), x ,
for all x, y g. If
 is moreover Adinvariant, then for every g G and for all
x, y g it follows from the definitions that
Adg ( (x)), y = (x), Adg1 y = x  Adg1 y
= Adg x  y = (Adg x), y ,
so that the following diagram is commutative:
g
g
Adg
Adg
(5.14)
A Lie algebra g, equipped with a nondegenerate symmetric bilinear form
 ,
which is adinvariant is called a quadratic Lie algebra.
A function F on g is called Adinvariant if it is invariant for the adjoint action
of G on g, i.e., if F(Adg x) = F(x) for all g G and x g. Similarly, a function F
on g is called Ad invariant if it is invariant for the coadjoint action of G on g .
The Ad invariance of a function F on g is characterized infinitesimally by
add F = 0 ,
(5.15)
for all g . Indeed, for z g and g ,
add F , z = , z, d F = d F, adz = 0 ,
where we used in the last step that dtd t=0 F(Adexptz ) = 0, itself a direct consequence of the Ad invariance of F. If F : g F is an Ad invariant function and
if the bilinear form
 is Adinvariant, then the commutativity of (5.14) leads for
every g G to
F Adg = F Adg = F ,
so that F is Adinvariant. Similarly,
 is Adinvariant and F is Adinvariant, then F is Ad invariant. Thus, establishes a onetoone correspondence
5.2 Poisson Reduction
123
between Ad invariant functions on g and Adinvariant functions on g. For a function F : g F, the Adinvariance of
 also leads to
add (F ) = ( ) F, ( ) ,
(5.16)
where x F := (dx F), for x g. It follows that a function F on g is Adinvariant if
and only if [x F, x] = 0 for all x g. The proof of (5.16) follows at once from the
definitions of , ad, and the adinvariance of
 .
5.2 Poisson Reduction
In many geometrical situations of interest, one deals with a quotient of a submanifold of a Poisson manifold, and while the submanifold itself does not inherit a
Poisson structure from its ambient space (i.e., it is not a Poisson submanifold, as
in Section 2.2), the quotient does inherit a Poisson structure. The construction of
such a Poisson structure is called Poisson reduction, while the Poisson structure
itself is called a reduced Poisson structure. We give in this section necessary and
sufficient conditions for Poisson reduction: we formulate them first algebraically
(Section 5.2.1) and then for manifolds (Section 5.2.2). The geometrical picture to
keep in mind, reading this section, consists of the following manifolds and maps:
N M
p
(5.17)
P
where M = (M, ) is a Poisson manifold, N is a submanifold of M and p is a submersion. The example to keep in mind is that of a Poisson manifold on which a
group G acts, a submanifold N which is Ginvariant and the quotient P = N/G. We
will come back to this example in Section 5.4.2.
5.2.1 Algebraic Poisson Reduction
We first describe Poisson reduction in purely algebraic terms. Dualizing the diagram (5.17), we consider a Poisson algebra (A , , { , }), an ideal I of (A , ), the
canonical map : A A /I and a subalgebra B of A /I , as in the following
diagram:
124
5 Reduction
(5.18)
B A /I
We
would
like to have a Poisson structure { , }B on B, such that {F, G}B =
G , for every F, G B, where F and G are arbitrary representatives in A
F,
of F and G.
Definition 5.4. Let (A , , { , }) be a Poisson algebra, let I be an ideal of (A , )
and let B be a subalgebra of A /I . We say that the triple (A , I , B) is Poisson
reducible if there exists a Poisson bracket { , }B on B, such that
G ,
{F, G}B = F,
(5.19)
G are arbitrary representatives of F and G in A . The
for all F, G B, where F,
Poisson bracket { , }B on B is called the reduced Poisson bracket.
We give in the following proposition a necessary and sufficient condition for a
triple (A , I , B) to be Poisson reducible. For the ideal I of (A , ) we denote its
normalizer in (A , { , }) by N (I ),
N (I ) = {F A  {I , F} I } .
(5.20)
Clearly, N (I ) is a Poisson subalgebra of A : since { , } is a biderivation, N (I )
is a subalgebra of (A , ), while the Jacobi identity for { , } implies that N (I ) is
a Lie subalgebra of (A , { , }).
Proposition 5.5. Let (A , , { , }) be a Poisson algebra, let I be an ideal of (A , )
and let B be a subalgebra of A /I . Then (A , I , B) is Poisson reducible if and
only if the following conditions are satisfied:
(1)
(2)
1 (B) N (I );
1 (B) is a Lie subalgebra of (A , { , }).
For F, G B, their reduced Poisson bracket {F, G}B is then given by
G ,
{F, G}B = F,
(5.21)
where F and G are arbitrary representatives in A of F and G.
Proof. Suppose first that conditions (1) and (2) are satisfied. We wish to define a
G be representaPoisson bracket on B by using Eq. (5.21). Let F, G B and let F,
tives in A of F and G; both F and G are determined uniquely up to an element
of
I .
G is,
According to condition (1), F and G belong to N (I ), so their bracket F,
and G for F and G. Conmodulo I , independent
of
the
chosen
representatives
F
G B. It follows that
G 1 (B), so that F,
dition (2) implies that F,
(5.21) defines a map { , }B : B B B. Since is an algebra homomorphism,
5.2 Poisson Reduction
125
the fact that { , } is a skewsymmetric biderivation of A implies that { , }B is
a skewsymmetric biderivation of B. For the Jacobi identity, observe that, for all
F, G, H B, (5.21) yields that
G , H
{{F, G}B , H} =
F,
B
G and H are arbitrary representatives of F, G and H. The Jacobi identity for
where F,
{ , }B therefore follows from the Jacobi identity for the bracket { , }. We conclude
that (A , I , B) is Poisson reducible, in particular the reduced Poisson bracket on
B is given by (5.21).
Suppose now that (A , I , B) is Poisson reducible, so that B admits a Poisarbitrary elements of
son bracket { , }B which satisfies (5.19).Since F and G are
1 (B), the latter formula implies that 1 (B), 1 (B) B, which yields
1
condition (2). For arbitrary
G I
we have from (5.19),
elements
F (B)
and
1
since (G) = 0, that F, G = 0, so that (B), I I , which yields
condition (1).
Notice that since 1 (B) is a subalgebra of (A , ) and N (I ) is a Poisson
subalgebra of (A , , { , }), the two conditions (1) and (2) in Proposition 5.5 can be
summarized in the single condition:
1 (B) is a Poisson subalgebra of N (I ) .
Moreover, viewing the canonical map as a map : 1 (B) B, the formula (5.21) for the reduced Poisson bracket can be simply stated as being a
morphism of Poisson algebras, where 1 (B), which is a Poisson subalgebra
of N (I ), and hence of A , is equipped with the Poisson bracket which it inherits
from A .
We next consider Poisson reduction in the setting of affine varieties. We do this
by replacing each of the algebras in (5.18) by an affine variety, keeping in mind that
under the basic correspondence between algebras of finite type and affine varieties,
surjective algebra homomorphisms are replaced by injective maps between varieties
(with the direction of the arrow being reversed) while injective algebra homomorphisms are replaced by dominant maps between varieties, i.e., maps whose image
is (Zariski) dense in the target variety. We are thus led to consider the following
diagram,
N M
p
(5.22)
P
where M, N and P are affine varieties and p is a dominant map.
Definition 5.6. Let N be a subvariety of an affine Poisson variety (M, { , }) and let
p : N P be a dominant map, where P is also an affine variety. We say that the
triple (M, N, P) is Poisson reducible if there exists a Poisson structure { , }P on P,
such that, for every n N,
126
5 Reduction
G (n) ,
{F, G}P (p(n)) = F,
G F (M) of F p and G p. The
for all F, G F (P) and for all extensions F,
Poisson structure { , }P is called a reduced Poisson structure.
It is clear from the definitions that a triple of affine varieties (M, N, P) is Poisson
reducible if and only if the triple of algebras (A , I , B) := (F (M), IN , F (P)) is
Poisson reducible, where IN denotes the ideal of N. The geometric interpretation
of the normalizer N (IN ) of IN , defined in (5.20), is that
N (IN ) = {F F (M)  XF is tangent to N at every point of N} .
(5.23)
Also, 1 (B) = 1 (F (P)) consists of all functions on M, whose restriction to N
is of the form H p, for some function H F (P). It follows that Proposition 5.5
leads for affine varieties to the following result.
Proposition 5.7. Let (M, { , }) be an affine Poisson variety, let N be a subvariety
of M and let P be an affine variety. Suppose that p : N P is a dominant map, as
in diagram (5.22). Then (M, N, P) is Poisson reducible if and only if the following
conditions are satisfied:
(1)
(2)
All Hamiltonian vector fields, which are associated to functions, whose restriction to N is of the form H p, for some function H F (P), are tangent
to N;
G F (M) of
For every two functions F, G F (P) and for all extensions F,
G (n) =
F p and G p to M, there exists a function H F (P), such that F,
H(p(n)), for every n N.
In this case, the Poisson bracket of F, G F (P) at p(n), where n N, is given by
G (n) ,
{F, G}P (p(n)) = F,
where F and G are arbitrary extensions to M, of F p and G p.
There are two extreme cases of Proposition 5.7. The first one occurs when N = P.
Comparing Definitions 2.9 and 5.6, one has that (M, N, N) is defined to be Poisson
reducible if and only if N is a Poisson subvariety of M; Proposition 5.7 says that this
is the case precisely when all Hamiltonian vector fields are tangent to N. Thus the
equivalence (i) (iii) in Proposition 2.10 is a particular case of Proposition 5.7.
For the second extreme case of Proposition 5.7, suppose now that N = M. Then
Proposition 5.7 says that (M, M, P) is Poisson reducible if and only if the Poisson
bracket of every pair of functions which are of the form H p, with H F (P), is
also of that form. We will see an important particular case of this in Section 5.4,
where these functions are the invariant functions for some group action. Another
particular case of Proposition 5.7 is given in the following corollary.
Corollary 5.8. Let (M, { , }) be an affine Poisson variety, let N be a subvariety
of M and let P be an affine variety. Suppose that p : N P is a dominant map and
that for a function F F (M) the following conditions are equivalent:
5.2 Poisson Reduction
(i)
(ii)
127
The restriction of F to N is of the form H p, for some H F (P);
The Hamiltonian vector field XF is tangent to N.
Then (M, N, P) is Poisson reducible.
Proof. If (i) and (ii) are equivalent, then condition (1) in Proposition 5.7 is clearly
G F (M) be extensions of
satisfied. For condition (2), let F, G F (P) and let F,
F p and G p. According to (i) (ii), XF [IN ] IN and XG [IN ] IN , where,
as before, we denote the ideal of all functions on M which vanish on N by IN . Using
the Jacobi identity for { , } in the form of Proposition 1.4 (5), it follows that
IN + X F,
IN XF [IN ] + XG [IN ] IN ,
G,
X{F,
G } [IN ] XF
G
G to N is of the
which implies, in view of (ii) (i), that the restriction of F,
form H p, for some H F (P).
The sufficient condition for Poisson reducibility, given in Corollary 5.8, is not a
necessary one. For example, let P and Q be affine Poisson varieties and consider the
triple (P Q, P Q, P), where P Q is equipped with the product Poisson structure.
If Q is of positive dimension (i.e., is not a point), then conditions (i) and (ii) in the
corollary are not equivalent. However, the triple is reducible, since condition (1) in
Proposition 5.7 is trivially satisfied, while condition (2) is a consequence of the fact
that the projection map p2 : P Q Q is a Poisson map (see Proposition 2.2).
5.2.2 Poisson Reduction for Poisson Manifolds
We now turn to the case of Poisson manifolds. The basic setup and the basic definition are very similar to the case of affine Poisson varieties (Section 5.2.1). First,
recall that, by definition, a map p : N P between manifolds is a submersion if the
tangent map Tn p : Tn N Tp(n) P is surjective for every n N; in particular, every
submersion is an open map (it maps open subsets of N to open subsets of P).
Definition 5.9. Let (M, { , }) be a Poisson manifold, N an (immersed or embedded)
submanifold of M, P a manifold and p : N P a surjective submersion, as in the
following diagram:
N M
p
(5.24)
P
We say that the triple (M, N, P) is Poisson reducible if there exists a Poisson structure P = { , }P on P, such that for all open subsets V N and U M, with
V U N, and for all functions F, G F (p(V )), one has
G (n) ,
{F, G}P (p(n)) = F,
(5.25)
128
5 Reduction
G F (U) are arbitrary extensions of F p and
at all points n of V , where F,
V
G pV . The Poisson structure P on P is called the reduced Poisson structure.
Remark 5.10. Recall from Section 2.2.2 that an immersed submanifold comes with
its own topology and differential structure, so the open subset V of N need not be
of the form U N for some open subset U of M. However, when N is an embedded
submanifold of M, it carries the induced topology from M and the condition in the
above definition simplifies to: for every open subset U M and for all functions
G F (U) are
F, G F (p(U N)), one has (5.25), for all n U N, where F,
arbitrary extensions of F pUN and G pUN .
Proposition 5.11. Let (M, { , }) be a Poisson manifold, let N be an (immersed or
embedded) submanifold of M and let p : N P be a surjective submersion onto a
manifold P. Then (M, N, P) is Poisson reducible if and only if for all open subsets
V N and U M, with V U N, the following conditions are satisfied:
(1)
(2)
For every function F F (U), whose restriction to V is constant on the fibers
of p, the Hamiltonian vector field XF is tangent to N at every point of V ;
G F (U) whose restrictions to V are constant
For every pair of functions F,
G , restricted to V , is constant on
on the fibers of p, their Poisson bracket F,
the fibers of p.
In this case, for every open subset V N and for all n V , the Poisson bracket of
F, G F (p(V )) at p(n) P is given by
G (n) ,
{F, G}P (p(n)) = F,
(5.26)
where F and G are arbitrary extensions to a neighborhood of n in M, of F pV
and G pV . In particular, the Hamiltonian vector field associated to F is given,
at p(n), by
(5.27)
(XF ) p(n) = dn p (XF )n .
Proof. We first point out that for an open subset V N and a function F F (V )
the following conditions are equivalent:
(i)
(ii)
F is of the form H p for some function H F (p(V ));
F is constant on all the fibers of p.
The only thing which is not obvious in this equivalence is the fact that the function H, constructed in (ii) (i), is smooth/holomorphic; it is a consequence of the
implicit function theorem, applied to the surjective submersion p.
Suppose that (M, N, P) is Poisson reducible. Let V N and U M be open subsets, with V U N, and suppose that F F (U) is constant on the fibers of p,
when restricted to V . Then FV = F pV for some function F F (p(V )). For an
defined on a neighborhood U of n
arbitrary n V and an arbitrary function G,
in
M, and
vanishing on some neighborhood of n in N, we have in view of (5.25) that
G (n) = {F, G}P (p(n)) = 0, since G = 0, in a neighborhood of p(n). ThereF,
fore, XF is tangent to N at n, leading to condition (1). Condition (2) follows also
5.2 Poisson Reduction
129
from (5.25), because the lefthand side of (5.25) depends only on p(n), i.e., on the
fiber which contains n, rather than on n itself.
Suppose now that conditions (1) and (2) hold. For every w P, we define a skewsymmetric biderivation (P )w at w by: for all germs Fw , Gw at w,
G (n) ,
(P )w [Fw , Gw ] := n [Fn , G n ] = F,
(5.28)
where n is an arbitrary point of N for which p(n) = w and Fn , G n are germs at n
defined on a neighborhood of n in M, and satisfying
G,
of arbitrary functions F,
FV = F pV and GV = G pV for some neighborhood V of n in N. The righthand
side in (5.28) is independent
of the choice of n (with p(n) = w) because, according
G at points n of V depends on p(n) only. Similarly, it
to (2), the value of F,
is independent of the chosen F and G because, according
to (1), if G is such that
GV = 0 and F is such that FV = F pV , then F, G (n) = 0. Thus P is welldefined at every point w P and it inherits from the biderivation property at
each point. We need to show that the bivector field P , defined on P by w (P )w
is smooth/holomorphic. Let w P, let n p1 (w) and let V be a neighborhood
of n in N, having the property that every function in F (V ) extends to a function
in F (U), where U is a neighborhood of V in M. Such neighborhoods V and U exist
because N is an immersed submanifold of M. For every pair of functions F, G
F (p(V )), we then have, using the equivalence (i)(ii), given at the beginning of
the proof, that
G
{F, G}P p = F,
= H p ,

V
for some function H F (p(V )), in particular {F, G}P F (p(V )). This shows that
P is a smooth/holomorphic bivector field in a neighborhood of w; since w was
arbitrary, P is a smooth/holomorphic bivector field on P. In the bracket notation, it
is given by (5.26), for arbitrary points p(n) in P and for arbitrary functions F and G,
defined in a neighborhood of p(n) in P. Notice that (5.26) implies that
G , H (n) ,
{{F, G}P , H}P (p(n)) = F,
(5.29)
H ex G,
for arbitrary functions F, G, H, defined in a neighborhood of p(n), and F,
tensions as before. It follows from (5.29) that the Jacobi identity for { , } implies
the Jacobi identity for { , }P . Also, (5.27) is an immediate consequence of (5.26).
According to (1) in Proposition 5.11, if (M, N, P) is Poisson reducible and F is
a function, defined on an open subset of M, whose restriction to an open subset V of N vanishes, then the Hamiltonian vector field XF is tangent to N at every
point of V . This motivates the following definition, from which it follows that if
(M, N, P) is Poisson reducible, then N is a coisotropic submanifold of the Poisson
manifold (M, ).
Definition 5.12. A submanifold N of a Poisson manifold (M, ) is said to be a
coisotropic submanifold of M if for every open subset V of N and for every func
130
5 Reduction
defined on a neighborhood of V in M, and such that F
tion F,
= 0, the HamiltoNV
nian vector field XF is tangent to N, at every point of V .
The notion of a coisotropic submanifold generalizes the notion of a Poisson submanifold, since the Hamiltonian vector fields, associated to (local) functions which
vanish on a Poisson submanifold, are not only tangent to the Poisson submanifold,
but they actually vanish on it.
The condition that a submanifold N of a Poisson manifold is coisotropic can be
stated pointwise. To do this, we introduce, for every m M, the linear map m :
Tm M Tm M, which corresponds3 to the bivector m 2 Tm M, and for n N, the
subspace Tn N of Tn M, which consists of the annihilator of Tn N in Tn M,
Tn N := { Tn M  v Tn N, (v) = 0} .
By the implicit function theorem, every element of Tn N can be realized as the
defined on a neighborhood of n in M, and whose
differential at n of some function F,
restriction to N vanishes. It leads to the following two pointwise characterizations
of a coisotropic submanifold.
Lemma 5.13. Let N be a submanifold of a Poisson manifold (M, ). The following
conditions on N are equivalent.
(i)
(ii)
(iii)
N is a coisotropic submanifold;
For every n N, n (Tn N, Tn N) = {0};
For every n N, n (Tn N) Tn N.
Similarly, the stronger condition that all Hamiltonian vector fields XF are tangent to
the fibers pn := p1 (p(n)) of the surjective map p : N P can be stated as follows:
for every n N,
n (Tn N) Tn pn .
In the following proposition we show that for a triple (M, N, P) as above, if all the
tangent spaces Tn pn are spanned by Hamiltonian vector fields which come from
functions which vanish on N, then (M, N, P) is Poisson reducible.
Proposition 5.14. Let (M, ) be a Poisson manifold, N an (immersed or embedded)
submanifold of M and let p : N P be a surjective submersion with connected
fibers. If
n Tn N = Tn pn
(5.30)
for every n N, then N is a coisotropic submanifold of M and (M, N, P) is Poisson
reducible.
Our sign convention is that m (dm H)[F] = m [Hm , Fm ] = {H, F} (m), for all functions F and H,
defined on a neighborhood of m.
5.2 Poisson Reduction
131
Proof. Notice first that in view of (iii) (i) in Lemma 5.13, N is a coisotropic
submanifold of M. To show that (M, N, P) is Poisson reducible, we verify both conditions (1) and (2) in Proposition 5.11.
(1) The skewsymmetry of n , combined with (5.30), yields that n (Tn pn ) Tn N,
for every n N. It implies that the Hamiltonian vector field of every function, defined on a neighborhood of n in M, whose restriction to N is constant on the fibers
of p, is tangent to N, which is condition (1).
G
(2) Let U M and V N be nonempty open subsets, with V U N. Let F,
F (U) be two functions whose
restrictions
to
V
are
constant
on
the
fibers
of
p. We
G to V is constant on the fibers of p. Let n V
show that the restriction of F,
and v Tn pn be arbitrary. Then v can be extended, on a neighborhood of n in M,
to a Hamiltonian vector field, which is tangent to the fibers of p. Indeed, by (5.30),
there exists a function H on a neighborhood U U of n in M, whose restriction
= v. Since H is zero on V , its
to V := V U is zero, with (XH )n = n (dn H)
Hamiltonian vector field XH is indeed, again by (5.30), tangent to the fibers of p at
is zero on V , because F is
every point of V . It follows that the function XH [F]
V
is
constant on the fibers of p. By (5.30), XH
[F], G vanishes
on
N
(recall
that
G
F (n) = 0. Using the Jacobi
also constant on the fibers of p). Similarly, XH [G],
identity, it follows that
(n) = 0 .
G (n) = XH [F],
G (n) + F,
XH [G]
G = XH F,
v F,
G to each connected
Since n N and v Tn pn are arbitrary, the restriction of F,
component of pn V is constant. Notice that, since pn V is not
necessarily con G to every fiber of p
nected, we cannot conclude from it that the restriction of F,
G
is constant. However, there is a function H on p1 (p(V )), which agrees with F,
at every point of V and which has the same property of being constant, when restricted to each
fiber
pn , where n V ; since each fiber pn is connected, we can
G is constant, when restricted to the fibers of p. In order to conconclude that F,
struct H, let n be an arbitrary point of p1 (p(V )). Since F is constant on the fibers
of p, its restriction to V is of the form F pV for some function F. On a small neighborhood U of n in M we can construct an extension F0 of F pV , where V is a
It
neighborhood of n in N, with V U
N. Similarly, we construct G 0 from G.
follows from (5.30) that H := F0 , G0 is at points of V independent of the chosen
1
extensions,
hence leads to a welldefined function on p (p(V )), which coincides
with F, G on V . Moreover, it follows as earlier in this proof that H is constant on
the fibers of p. This proves the existence of H, with the announced properties, hence
achieves the proof.
Remark 5.15. Let N be a coisotropic submanifold of a Poisson manifold (M, ).
Assume that the dimension of n (Tn N) is independent of n N. According to
Lemma 5.13, a distribution is defined on N by assigning to every n N the subspace n (Tn N) of Tn M. This distribution is differentiable, because it is locally defined by the Hamiltonian vector fields which correspond to local functions on M
132
5 Reduction
which vanish on N. For functions F, G, defined on an open subset U of M, and
vanishing on N, their commutator [XF , XG ] = X{F,G} is the Hamiltonian vector
field of the function {F, G}, which vanishes on N; indeed, {F, G} = XG [F] vanishes
on N because F is constant on N (in fact, it vanishes on N) and XG is tangent to N
(since N is coisotropic). Thus, [XF , XG ] belongs to the distribution and the latter is
integrable in the sense of Frobenius. If the space P of leaves of this distribution is a
manifold, then the conditions of Proposition 5.14 hold, so that (M, N, P) is Poisson
reducible and P inherits a Poisson structure from (M, ).
Remark 5.16. It is instructive to write down the conditions of Poisson reducibility
in terms of a wellchosen system of local coordinates for the ambient manifold. It
leads to an expression for the Poisson matrix of the reduced Poisson structure in
terms of the Poisson matrix of the initial Poisson structure.
Let (M, N, P) be three manifolds with N M a submanifold of M, and p : N P
a surjective submersion. For simplicity, we assume that N is an embedded submanifold. We define s,t and u by
s := dim P,
s + t := dim N,
s + t + u := dim M .
Let n be an arbitrary point of N and let (x1 , . . . , xs ) be an arbitrary coordinate system,
defined on a neighborhood of p(n) in P. For i = 1, . . . , s, choose an extension xi
of xi p to a neighborhood of n in M. Then there exist on a neighborhood of n
in M functions y1 , . . . , yt and z1 , . . . , zu , such that (x1 , . . . , xs , y1 , . . . , yt , z1 , . . . , zu ) is a
coordinate system for M in a neighborhood U of n, and such that N U is given by
the equations z1 = = zu = 0.
Let be a Poisson structure on M. The Poisson matrix X of with respect to the
coordinates x1 , . . . , xs , y1 , . . . , yt , z1 , . . . , zu (in that order) is given by the block matrix
A
B C
X = B D E
C E F
D and F are square matrices of size s, t and u respectively. Each of the
where A,
B, C, . . . is a matrixvalued function on U, whose restriction to U N
blocks A,
is denoted by A N , BN , CN , . . . We summarize the conditions related to Poisson
reducibility, expressed in terms of the Poisson matrix X, in Table 5.1.
The proof of each one of the lines of the table is a direct consequence of the
definitions. To start with, N U is coisotropic in U M if and only if every Hamiltonian vector field XG , with G a function on U which vanishes on N U, is tangent
to N at every point of N U. Since N U is the zero locus of the local coordinates
z1 , . . . , zs , this happens if and only if the Hamiltonian vector fields Xzi are tangent
to N U at every point of N U for all i = 1, . . . , u, i.e. if and only if Xzi [z j ](n ) = 0
for all i, j = 1, . . . , s and n N U. This is equivalent to saying that FN = 0.
Let us explain the second line in the table. Condition (1) in Proposition 5.11
demands that the Hamiltonian vector field XG is tangent to N for every function
G whose restriction to N is of the form G p for some function G defined in a
5.2 Poisson Reduction
133
Table 5.1 A summary of the conditions related to Poisson reducibility, expressed in terms of the
Poisson matrix X.
Condition
Poisson matrix
N is a coisotropic submanifold
FN = 0
Condition (1) in Proposition 5.11
CN = FN = 0
Condition (2) in Proposition 5.11
A N depends only on x1 , . . . , xs
(M, N, P) is Poisson reducible
Condition (5.30) in Proposition 5.14
CN = FN = 0
A N depends only on x1 , . . . , xs
CN = FN = 0
Rk(E(n )) = t
for every n N U
neighborhood of p(n) in P. It implies that the Hamiltonian vector fields associated
to the functions x1 , . . . , xs , z1 , . . . , zu are tangent to N, hence that CN = FN = 0. The
inverse implication follows from the fact that if the restriction of a function G to N
is constant on the fibers of p, then this restriction is independent of the coordinates
y1 , . . . , yt .
For the third
line,
condition (2) in Proposition 5.11 is tantamount to saying that
the functions xi , x j , restricted to U N depend on the functions x1 , . . . , xs only;
in terms of the Poisson matrix X, this condition amounts to saying that A N depends
only on x1 , . . . , xs .
The fourth line follows from the second and third lines, because (M, N, P) being
Poisson reducible means precisely that conditions (1) and (2) in Proposition 5.11
are satisfied.
For the last line in the table, condition (5.30) in Proposition 5.14, which states
that n Tn N = Tn pn for every n N, implies that condition (1) in Proposition 5.11
is satisfied, so that CN = FN = 0. It also implies that, at every point n N U, the
u columns of the matrix E, which represent the Hamiltonian vector fields of the
functions z1 , . . . , zu , generate a space of dimension t, namely the tangent space of
the fiber of p through n .
When (M, N, P) is Poisson reducible, A N depends only on the variables x1 , . . . , xs ,
so that there exists a matrixvalued function A, such that A p = A U . In view of
(5.26), the Poisson matrix of the reduced Poisson structure on P, expressed in terms
of the local coordinates x1 , . . . , xs , is precisely the matrix A.
134
5 Reduction
5.3 PoissonDirac Reduction
We saw in Section 2.2 that if (A , , { , }) is a Poisson algebra and I is a Poisson
ideal of (A , ), then A /I inherits a Poisson algebra structure from A . As we will
show in this section, A /I may inherit a Poisson structure from A under weaker
conditions; we then say that the Poisson structure on A /I is obtained by Poisson
Dirac reduction. For Poisson manifolds it amounts to obtaining a Poisson structure
on a submanifold, which is not a Poisson submanifold.
5.3.1 Algebraic PoissonDirac Reduction
Let (A , , { , }) be a Poisson algebra and let I be an ideal of (A , ). Consider the
composition of algebra homomorphisms
N (I )
A
I
(5.31)
where we recall from Section 5.2.1 that N (I ) is the normalizer of I ,
N (I ) := {F A  {F, I } I } ,
which is a Poisson subalgebra of A . The kernel of the composite map in (5.31) is the
Poisson ideal N (I ) I of N (I ). It follows that N (I )/(N (I ) I ) is itself
a Poisson algebra. Notice that it is isomorphic to a subalgebra of A /I , although
the latter is not a Poisson algebra, in general. It leads to the following definition.
Definition 5.17. Let (A , , { , }) be a Poisson algebra. An ideal I of (A , ) is said
to be a PoissonDirac ideal if one of the following equivalent conditions holds:
(i)
(ii)
(iii)
The natural inclusion of algebras N (I )/(N (I ) I ) A /I is an isomorphism;
The composition of algebra homomorphisms N (I ) A A /I is surjective;
I + N (I ) = A .
The three conditions in the definition are clearly equivalent. Condition (i) implies
that A /I is a Poisson algebra, isomorphic to N (I )/(N (I ) I ), as stated in
the following proposition.
Proposition 5.18. Let I be a PoissonDirac ideal of a Poisson algebra (A , , { , }).
Then A /I has a unique Poisson bracket which makes the surjective morphism
N (I ) A A /I into a morphism of Poisson algebras. This bracket is said
to be obtained by PoissonDirac reduction.
When I is a PoissonDirac ideal, the different Poisson algebras which appear in
PoissonDirac reduction can be summarized in the following commutative diagram,
5.3 PoissonDirac Reduction
135
N (I )
N (I )
N (I ) I
(5.32)
A /I
where all maps are morphisms of Poisson algebras, except (in general) for the
canonical projection : A A /I .
We now consider PoissonDirac reduction for affine Poisson varieties. Suppose
that N M is a subvariety of an affine Poisson variety (M, { , }). We denote the
ideal of N (in F (M)) by IN . Recall from (5.23) that N (IN ) consists of all functions on M whose Hamiltonian vector field is tangent to N. Translating (ii) of Definition 5.17 into geometrical terms, leads to the following definition.
Definition 5.19. Let (M, { , }) be an affine Poisson variety. A subvariety N M is
said to be a PoissonDirac subvariety if every function on N is the restriction of a
function on M whose Hamiltonian vector field is tangent to N.
Thus, a subvariety N is, by definition, a PoissonDirac subvariety if and only if its
ideal IN is a PoissonDirac ideal. In view of Definition 5.17 this leads to other, algebraic, characterizations of PoissonDirac subvarieties and Proposition 5.18 shows
that a PoissonDirac subvariety inherits a Poisson structure from its ambient Poisson variety. That is the content of the following proposition.
Proposition 5.20. Let N be a PoissonDirac subvariety of an affine Poisson variety
(M, { , }). There exists a unique Poisson structure { , }N on N such that, for every
F, G F (N), their Poisson bracket is given by
G
{F, G}N = F,
(5.33)
N
G are arbitrary extensions of F, G to M, whose Hamiltonian vector fields
where F,
XF and XG are tangent to N at all points of N. In particular, every Hamiltonian
vector field on (N, { , }N ) is the restriction, to N, of a Hamiltonian vector field
on (M, { , }). The Poisson structure { , }N is called a reduced Poisson structure.
Remark 5.21. Suppose that N is a PoissonDirac subvariety of an affine Poisson
variety (M, { , }). Then the subalgebra N (IN ) of F (M) may not be finitely generated, and therefore may not be the algebra of functions on some affine variety.
However, when N (IN ) is finitely generated, then the commutative diagram (5.32)
leads to the following commutative diagram of affine varieties
136
5 Reduction
QN
DN
where QN , respectively DN , is the affine variety, whose algebra of regular functions
is N (IN ), respectively N (IN )/ (N (IN ) IN ). All maps in this diagram are
Poisson maps, except (in general) for the inclusion N M.
We have seen in Section 2.2 that if N is a Poisson subvariety of a Poisson variety
(M, { , }), then for every function F on M, which vanishes on N, the Hamiltonian
vector field XF vanishes at every point of N. For PoissonDirac subvarieties a similar, but weaker, property holds.
Proposition 5.22. Let (M, { , }) be an affine Poisson variety and let N be a Poisson
Dirac subvariety, whose ideal is denoted by IN . Let F IN . If the Hamiltonian
vector field XF is tangent to N at some point n N, then it vanishes at that
point, (XF )n = 0. In other words, Hamn (IN ) Tn N = {0} for every n N, where
Hamn (IN ) = {(XF )n  F IN } is the vector space of Hamiltonian vectors at n,
which come from elements of IN .
Proof. Let F IN and let n N. Suppose that XF is tangent to N at n, which
means that XF [G](n) = 0, for every G IN . We wish to show that (XF )n = 0. If
H F (M), then H can be written as H1 + H2 with H1 IN and H2 N (IN ),
since IN is a PoissonDirac ideal (condition (iii) in Definition 5.17). By the above,
XF [H1 ](n) = 0, while XF [H2 ](n) = {H2 , F} (n) = 0, since H2 belongs to the normalizer of IN . Therefore, XF [H](n) = 0 for every H F (M), which means that
(XF )n = 0.
To finish this section, we write a Poisson matrix for the reduced Poisson structures,
obtained by PoissonDirac reduction. Assume that N is a PoissonDirac subvariety of an affine Poisson variety (M, { , }). As before, the ideal of N is denoted
by IN and its normalizer by N (IN ). There exist generators F1 , . . . , Fs , G1 , . . . , Gt
of F (M) with every Fi belonging to N (IN ) and every Gi belonging to IN ; to
construct such generators one can for example start from generators H1 , . . . , Hs for
F (M) = N (IN ) + IN and decompose every Hi as Fi + Gi , with Fi N (IN )
and Gi IN . In terms of the generators F1 , . . . , Fs , G1 , . . . , Gt , the Poisson matrix of
(M, { , }) is a 2 2 block matrix
A B
B D
where
Ai j = Fi , Fj , Bik = {Fi , Gk } , Dk = {Gk , G } ,
for 1 i, j s and 1 k, t. Since N is a PoissonDirac subvariety, the functions
F1 , . . . , Fs , restricted to N, generate F (N). Also, the Poisson matrix of (N, { , }N ),
5.3 PoissonDirac Reduction
137
in terms of these generators, is given by the matrix A, with entries Ai j = Fi , Fj  ,
N
where 1 i, j s; indeed, the Hamiltonian vector fields of F1 , . . . , Fs are tangent
to N at every point ofN, hencewe can use these functions in (5.33) to compute
the Poisson brackets Fi N , Fj N . Note also that all entries of B belong to IN ,
N
hence vanish on N; see Proposition 5.25 below for a converse in the differentialgeometrical context.
5.3.2 PoissonDirac Reduction for Poisson Manifolds
Let N be an (immersed or embedded) submanifold of a manifold M. Given a point
n in N and a function F, defined on a neighborhood W of n in N, a local extension
of F at n is a triple (F,U,V
) where
(1)
(2)
(3)
U M is a neighborhood of n in M;
V (W U) N is a neighborhood of n in N;
F F (U) is a function such that FV = FV .
The relation between the different subsets and maps is represented in the following
diagram:
N
M
V
F
F
U
F
F
Every function F, defined on an open subset of N, admits a local extension at every
point of its domain of definition.
Definition 5.23. Let (M, { , }) be a Poisson manifold and let N be an (immersed or
embedded) submanifold of M. We say that N is a PoissonDirac submanifold of M,
if for every n N, every function F, defined in a neighborhood of n in N, admits a
local extension (F,U,V
) at n, such that XF is tangent to N at every point of V .
Note that, in particular, a Poisson submanifold N of a Poisson manifold (M, { , }) is
a PoissonDirac submanifold, since all Hamiltonian vector fields are tangent to N.
We show in the following proposition that every PoissonDirac submanifold inherits a Poisson structure from its ambient Poisson manifold.
Proposition 5.24. Let (M, { , }) be a Poisson manifold and let N be a Poisson
Dirac submanifold of M. There exists on N a unique Poisson structure { , }N such
G
that, for all open subsets V N and U M with V U, for every function F,
138
5 Reduction
F (U) whose Hamiltonian vector fields are tangent to N at every point of V , and for
every n V , we have
G (n) ,
{F, G}N (n) = F,
(5.34)
G to V . In particular, locally, every
where F, G F (V ) are the restrictions of F,
Hamiltonian vector field on N is the restriction to N of a Hamiltonian vector field
on M.
Proof. For every n N, we define a skewsymmetric pointwise biderivation (N )n
by its values on germs Fn , Gn at n,
G (n) ,
(N )n (Fn , Gn ) := F,
(5.35)
where the righthand side is constructed as follows:
Choose functions F and G, defined in a neighborhood of n in N, with germs
Fn , Gn at n;
Choose local extensions (F,U,V
) and (G,U,V
) of F and G at n, such that the
Hamiltonian vector fields XF and XG are tangent to N, at every point of V .
We have to check that this definition makes sense: a priori, the righthand side
of (5.35) depends on the choice of representatives F and G of the germs Fn and Gn ,
and on the choice of local extensions (F,U,V
) and (G,U,V
). The Hamiltonian vec
tor field XG is tangent to N at every point of the subset V of N, so that XG [F](n)
depends only on the restriction of F to V , and therefore depends only on the germ Fn
of F at n. Hence
G (n) = XG [F](n)
= XF [G](n)
F,
depends only on the germs Fn and Gn , as was to be checked.
We may now define a bivector field N = { , }N on N by its value on every pair
of functions F, G, defined on an open subset V of N, setting
G (n) ,
{F, G}N (n) := (N )n (Fn , Gn ) = F,
(5.36)
for every n V . To check that the bivector field is smooth/holomorphic, notice that
the above extensions (F,U,V
) and (G,U,V
) are local extensions of F and G at
every n V , with Hamiltonian vector fields tangent to N at every point of V , so
that (5.36) is not only valid for n, but for every n V . Since the righthand side of
(5.36) is a smooth or holomorphic function of n, and since F and G are arbitrary
functions, we can conclude that the bivector field N is also smooth or holomorphic
(on V , hence on N).
H be three functions, defined on an open subset U of M,
G,
Let, as above, F,
assume that their Hamiltonian vector fields are tangent to N at every point of an open
subset Vof N,
and let F, G, H denote their restrictions to V . The Hamiltonian vector
G is tangent to N, at every point of V , since X = [XF , XG ]
field of F,
{F,G}
(item (5) in Proposition 1.4). As a consequence, F, G ,U,V is a local extension
of {F, G}N at n, whose Hamiltonian vector field is tangent to N at every point of V ,
hence
5.3 PoissonDirac Reduction
{{F, G}N , H}N (n) =
139
G , H (n) ,
F,
(5.37)
for every n V , so that the Jacobi identity for the bracket { , }N follows from the
Jacobi identity for { , }.
In order to give a geometrical characterization of PoissonDirac submanifolds
(Proposition 5.25 below), we express the Poisson bivector on a Poisson manifold
in terms of local coordinates which are adapted to the PoissonDirac submanifold.
To do this, let N be a PoissonDirac submanifold of a Poisson manifold (M, ), and
let n N. We denote the dimension of M by d and the dimension of N by s. There exists a coordinate chart (U, x) of M, adapted to N, and centered at n, such that the vector fields Xx1 , . . . , Xxs are tangent to N, at every point of an open subset V of U N,
which contains n. In order to construct such local coordinates, one may start from an
arbitrary coordinate chart (U, y) of M, adapted to N at n, and centered at n; since N is
a PoissonDirac submanifold, each of the functions y1 N , . . . , ys N can be extended in
a neighborhood of n, to produce the above functions x1 , . . . , xs , with the stated property, and supplemented with the remaining functions xs+1 := ys+1 , . . . , xd := yd they
form the required coordinate chart, after possibly shrinking the neighborhoods U
and V of n.
Since the vector fields Xx1 , . . . , Xxs are tangent to N, at every point of V , all the
functions {xi , x j } = [xi , x j ] = Xxi [x j ] vanish on V , for 1 i s < j d. For
every point n V , the Poisson structure , at n , is therefore given by:
xi j (n ) xi x j + xi j (n ) xi x j (5.38)
n
n
n
n
1i< js
s<i< jd
where xi j (n ) := xi , x j (n ), for 1 i < j d, and where the first sum is the reduced Poisson structure at n. Equation (5.38) implies that, in wellchosen local coordinates, the Poisson matrix of at points of N takes a blockdiagonal form. We
show in the following proposition that this fact characterizes PoissonDirac submanifolds.
Proposition 5.25. Let (M, ) be a Poisson manifold of dimension d and let N be a
submanifold of M. Then N is a PoissonDirac submanifold of M if and only if there
exists for each n N a coordinate chart (U, x) of M, adapted
N, and centered
at
at n with the following property: if the Poisson matrix X := ( xi , x j )1i, jd of in
terms of these coordinates is written in block form
A B
,
X=
B D
where A has size s := dim N, then there exists a neighborhood V of n in U N,
such that B(n ) = 0 for every n V . Moreover, when these equivalent conditions
are satisfied, then the Poisson matrix of the reduced Poisson structure N on N is
given, in terms of the above coordinates x1 V , . . . , xs V , on a neighborhood of n in N,
.
by AV = xi , x j 
V
1i, js
140
5 Reduction
Proof. We start with the stated characterization of a PoissonDirac submanifold. By
the above, every PoissonDirac submanifold admits indeed such an adapted coordinate chart, for each n N. For the inverse implication, let n N and assume that a
coordinate chart (U, x) of M is given, adapted at N, and centered at n. Let V denote
the corresponding open subset of N which contains n, and on which x1 V , . . . , xs V
are local coordinates for N. Suppose that the Poisson matrix of { , } in terms of
the coordinate chart (U, x) takes a blockdiagonal form at every point n in a neighborhood of n in N, as in (5.38). Let F be a function, defined in a neighborhood
of n in N and let F := F p, where p : U V is the projection map, associated
to (U, x). Then (F,U,V
) is an extension of F at n. Since the coordinate expression
of F on (U, x) depends on the first s variables x1 , . . . , xs only, the Hamiltonian vector
takes at points n in a neighborhood of n in N the form
field XF , associated to F,
s
F
(XF )n = xi j (n )
(n )
,
xj
xi n
i=1
so it is tangent to N at n . According to Definition 5.23, this shows that N is a
PoissonDirac submanifold.
We can use these coordinate charts to compute the Poisson matrix of the reduced Poisson structure on N. Indeed, on a neighborhood V of n in N,
the functions
x1 V , . . . , xs V are local coordinates on N and their Poisson brackets xi V , x j V
N
can, according to Proposition 5.24, be computed as xi , x j V . It follows that, in
terms of these coordinates, the Poisson matrix of the reduced Poisson structure is
the matrix A, restricted to V .
We give in the following proposition a description of the symplectic leaves of a
PoissonDirac submanifold.
Proposition 5.26. Let (N, N ) be a PoissonDirac submanifold of a Poisson manifold (M, ) and let n N. Denoting by Sn (N), respectively Sn (M), the symplectic
leaf of (N, N ), respectively of (M, ), which contains n, the symplectic leaf Sn (N)
is the connected component of N Sn (M) which contains n.
Proof. Let n N and let n Sn (N). By the definition of the symplectic leaf Sn (N),
passing through n, this means that there exists a piecewise Hamiltonian path for N
with endpoints n and n (recall from Section 1.3.4 that a Hamiltonian path is by
definition an integral curve of a Hamiltonian vector field). Since N is a Poisson
Dirac submanifold of M, every function F on N admits at every point n in its domain
of definition a local extension (F,U,V
), whose Hamiltonian vector field XF is equal
to XF on V . Since the image of is compact, it can be covered by finitely many such
local extensions, and is therefore a piecewise Hamiltonian path for . It follows
that
Sn (N) Sn (M) N .
(5.39)
Let C be the connected component of Sn (M) N containing n N. Since Sn (N) is
a connected set, (5.39) implies the stronger inclusion Sn (N) C . We need to prove
that Sn (N) = C .
5.3 PoissonDirac Reduction
141
To do this, let (U, x) be local coordinates in a neighborhood of n N, which
are adapted to the PoissonDirac submanifold, as in (5.38). For every function
with respect to { , }, and
F F (U), the Hamiltonian vector field XF of F,
the Hamiltonian vector field XF of its restriction F = FV F (V ), with respect
to { , }N , are related at n V by
F
(n )
.
(XF )n = (XF )n + xi j (n )
xi
x j n
s<i< jd
In particular, the tangent vector (XF )n in Tn M belongs to Tn N if and only if it is
equal to (XF )n . Since, by Theorem 1.30, the tangent space of a symplectic leaf at
a given point n is the space of all Hamiltonian vectors (XF )n at n , an element in
Tn Sn (M) belongs to Tn N if and only if it belongs to Tn Sn (N). In short, we have,
for all n Sn (N),
Tn Sn (N) = Tn Sn (M) Tn N .
(5.40)
In general geometrical terms, (5.39) and (5.40) say that we have three submanifolds
M0 , M1 and M2 of M, such that
M0 M1 M2
and
Tm0 M0 = Tm0 M1 Tm0 M2 ,
for all m0 M0 . According to the implicit function theorem, there exist in this case
neighborhoods U0 , U1 and U2 of m0 in M0 , M1 and M2 respectively such that U0 =
U1 U2 . In particular, there is a neighborhood of m0 in M1 M2 , which is contained
in M0 . Applied to (5.39) and (5.40), this means that Sn (N) is an open subset in C .
Let us show that Sn (N) is also a closed subset of C . Let n be a point in the closure
of Sn (N) in C . There is, for the above reasons, an open subset W of C contained in
Sn (N). Since W has a nonempty intersection with Sn (N), the symplectic leaves
Sn (N) and Sn (N) coincide, so that in particular n Sn (N), i.e., Sn (N) is a closed
subset of C . In conclusion, Sn (N) is, for every n N, an open and closed subset
of C , the connected component of Sn (M) N, which contains n.
We now explain how to compute the Poisson matrix of a Poisson structure obtained
by PoissonDirac reduction out of the Poisson matrix of the Poisson structure on the
ambient space. According to (5.34), if x1 , . . . , xs are functions whose Hamiltonian
vector fields are tangent to N, at every point in a neighborhood V of n in N, and
such that their
restrictions x1 , . . . , xs to V are local coordinates for N, then the matrix
xi , x j 
is the Poisson matrix of { , }N with respect to the coordinates
V
1i, js
x1 , . . . , xs . We show in the following proposition how to construct such functions and
we derive from it an explicit formula for the latter Poisson structure.
Proposition 5.27. Let (M, ) be a Poisson manifold and let N be an sdimensional
submanifold of M. Suppose that (U, x) is a coordinate chart of M, adapted to N
and centered at some point
n N. LetV denote an open subset of N, containing n,
having the property that x1 V , . . . , xs V is a system of local coordinates on V . Let X
denote the Poisson matrix xi , x j 1i, jd of with respect to these coordinates.
142
5 Reduction
We write X in the block form
X=
A B
B D
where A and D are square matrices, of size s, respectively d s, while B has size
s (d s). If the matrix D(m) is invertible for every m U, then
(1)
(2)
V is a PoissonDirac submanifold of (M, );
The Poisson matrix XN of the reduced Poisson structure on V is given, at
n V , by
XN (n ) = A(n ) + B(n )D(n )1 B(n ) .
Proof. We write the above coordinates x1 , . . . , xd as x1 , . . . , xs , y1 , . . . , yt , where t :=
d s, and we consider on U alternative coordinates (x , y ) = (x1 , . . . , xs , y1 , . . . , yt ),
defined by
1s C
x
x
=
,
y
y
0 1t
where C is a matrix which we will determine. These alternative coordinates are
also adapted to N, since xi xi vanishes on V , for i = 1, . . . , s. We determine C by
demanding that the s Hamiltonian vector fields Xxi are tangent to N at every point
of V . This means that dn y j , Xxi = 0 for all 1 i s, 1 j t and n V ,
n
where
, stands for the canonical pairing between Tn M and Tn M. Using the fact
that, for all F, G F (V ), we have that (XFG )n = F(n ) (XG )n at every point n
of N where the function G vanishes, we find that
t
dn y j , Xxi = dn y j , (Xxi )n Cik (n ) dn y j , Xyk n
n
k=1
= y j , xi (n ) Cik (n ) y j , yk (n )
t
k=1
= (B +CD)i j (n ) ,
which is zero, precisely if we choose C := BD1 . According to Proposition 5.25, V
is a PoissonDirac submanifold of M and the
matrix
of the reduced Poisson
Poisson
structure on V N is given by the matrix
xi , xj (n )
1i, js
, whose entries we
compute as above,
xi , x j (n ) = Xxi [xj ](n ) = Xxi [x j ](n ) =
t
xi Cik yk , x j (n )
k=1
= Ai j (n ) + Cik (n )B
k j (n ) = A +CB
k=1
ij
(n ) .
5.3 PoissonDirac Reduction
143
Since C = BD1 , the Poisson matrix of the reduced structure is given on V by A +
BD1 B .
5.3.3 The Transverse Poisson Structure
Let (M, ) be a Poisson manifold and let m M be a point where the rank of is 2r.
Suppose that q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs are splitting coordinates, defined on a
neighborhood U of m in M, and centered at m. According to Weinsteins splitting
theorem (Theorem 1.25), this means that, on U,
+ k (z)
z
i
i
k
i=1
1k<s
r
where the functions k are (smooth or holomorphic) functions, which depend on
z = (z1 , . . . , zs ) only, and which vanish when z = 0. In view of Definition 5.23, the
embedded submanifold N of M, defined by
N := {m U  q1 (m ) = p1 (m ) = = qr (m ) = pr (m ) = 0}
is a PoissonDirac submanifold of M. According to Proposition 5.20, the reduced
Poisson structure N on N is given by
N =
1k<s
k (z)
,
zk z
with the understanding that (z1 , . . . , zs ) stands for the restriction of (z1 , . . . , zs ) to N.
The submanifold N of M which appears in the latter example is transverse at m
to the symplectic leaf Sm through m, since Tm N Tm Sm = Tm M. Recall that two
submanifolds N and O of a manifold M are said to be transverse at a point m M if
m N O and Tm N + Tm O = Tm M; when N and O have complementary dimension
in M, this is equivalent to Tm N Tm Om = Tm M. The following theorem states that,
in a neighborhood of m in M, every submanifold of M of dimension dim N, which
is transverse to Sm at m, is in a neighborhood of m a PoissonDirac submanifold
of M and is isomorphic, as a Poisson manifold, to a neighborhood of m in N.
Theorem 5.28. Let (M, ) be a Poisson manifold and let m M. The rank of at m
is denoted by 2r. Suppose that q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs are splitting coordinates, defined on a neighborhood U of m in M, and centered at m. We denote by N0
the PoissonDirac submanifold of M, defined by
N0 := {m U  q1 (m ) = p1 (m ) = = qr (m ) = pr (m ) = 0} .
Let N1 M be an arbitrary sdimensional submanifold of M, transverse at m to the
symplectic leaf Sm of which contains m. Then there exists a neighborhood V0 of m
in N0 and a neighborhood V1 of m in N1 , such that
144
5 Reduction
(1)
(2)
V1 is a PoissonDirac submanifold of M;
There exists a Poisson diffeomorphism between V0 and V1 , where V0 and V1
are both equipped with their reduced Poisson structure as PoissonDirac submanifolds of M.
Proof. Let q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs be splitting coordinates, centered at m, defined on a neighborhood U of m in M. On U, the Poisson structure is given by
+ ,
pi
i=1 qi
r
(5.41)
where = 1k<s k (z) z z is a Poisson structure on U, whose structure
k
functions k depend only on the variables z = (z1 , . . . , zs ) and vanish for z = 0.
These coordinates are adapted to the symplectic leaf Sm , since Sm is defined on U
by z = 0. Consider an arbitrary sdimensional submanifold N1 of M, transverse at m
to the symplectic leaf Sm of , and of dimension s = dim N0 . Since N1 is of codimension 2r in M, N1 is in a neighborhood of m in M given as the intersection of 2r
hypersurfaces
(5.42)
H1 (q, p, z) = = H2r (q, p, z) = 0 .
Since N1 is transverse to Sm at m, these hypersurfaces are transverse to Sm at m
and their tangent spaces have Tm N1 as their common intersection. By the implicit
function theorem, applied to (5.42), there exist functions F1 , . . . , Fr , G1 , . . . , Gr , such
that N1 is, in a neighborhood of m, given by
qi = Fi (z1 , . . . , zs ) ,
pi = Gi (z1 , . . . , zs ) ,
for i = 1, . . . , r. Letting
qi := qi Fi (z1 , . . . , zs ) ,
pi := pi Gi (z1 , . . . , zs ) ,
it follows that (q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs ) is in a neighborhood U U of m
a coordinate system for M, adapted to both submanifolds N1 and Sm . Since
vanishes at m,
qi , p j (m) = i, j ,
qi , q j (m) = pi , pj (m) = 0 ,
where { , } := and i, j stands for the Kronecker delta. As a consequence, the
matrix
qi , pj (m )
qi , qj (m )
i, j=1
i, j=1
pi , q j (m )
pi , p j (m )
i, j=1
m
i, j=1
in a neighborhood of m in M. It proves, in view of Propois invertible for every
sition 5.27, that some neighborhood V1 of m in N1 is a PoissonDirac submanifold
of M, which is the content of (1).
5.3 PoissonDirac Reduction
145
The proof of (2) uses some of the objects which were constructed in the first
part of the proof, namely the open neighborhood U of m in M and the functions
F1 , G1 , . . . , Fr , Gr which are defined on U . Consider the product manifold U F,
which we equip with its product Poisson structure (the Poisson structure on F is
trivial because F is onedimensional). The splitting coordinates which we have constructed on U , plus the natural coordinate on F which we denote by , yield coordinates on U F. We define on U F the following 2r functions
Qi := qi Fi (z1 , . . . , zs ) and Pi := pi Gi (z1 , . . . , zs ) ,
where i = 1, . . . , r. For F, their common zero locus is denoted by N ,
N := m U  Qi (m , ) = Pi (m , ) = 0 for 1 i r .
Letting vary between 0 and 1 allows us to relate the PoissonDirac submanifolds N0 and N1 ; precisely, N0 = (N0 U ) {0} and N1 = (N1 U ) {1}, so
that N0 and N1 can be identified as Poisson manifolds with open neighborhoods
of m in N0 and in N1 respectively. We construct a Poisson vector field V on U F,
whose time 1 flow yields a diffeomorphism between N0 and N1 , at least in a neighborhood of m{0}; under the above identification, this diffeomorphism is a Poisson
diffeomorphism between a neighborhood of m in N0 and a neighborhood of m in N1 ,
equipped with their reduced structure, because the reduced Poisson structure on a
PoissonDirac submanifold (of U , or equivalently of U F) is uniquely defined.
The construction of the Poisson vector field V with the stated property will therefore
prove (2).
The vector field V is chosen of the form V = + XH , where H is a function
which will be specified below. This choice ensures that V is a Poisson vector field
on U F, since XH is a Hamiltonian vector field on U . The Hamiltonian H is
chosen as
H :=
j=1
j=1
A j ( , z1 , . . . , zs ) Q j + B j ( , z1 , . . . , zs ) Pj ,
(5.43)
where the functions A j and B j are solutions to the linear system
r
r
Q
+
A
,
Q
B j Qi , Pj = Fi ,
j
i
j
j=1
j=1
i = 1, . . . , r .
r
r
P
+
A
,
Q
B j Pi , Pj = Gi ,
j
i
j
j=1
(5.44)
j=1
Since
Qi , Q j (m, ) = Pi , Pj (m, ) = 0 ,
Qi , Pj (m, ) = i, j ,
for all 1 i, j r and for all F, this linear system has a unique solution, which is
defined on a neighborhood U W of m [0, 1] in U F, and which is smooth. Notice that since all coefficients in (5.44) are independent of the coordinates pi and qi ,
146
5 Reduction
the functions A j and B j depend only on z1 , . . . , zs and . Moreover, these functions
vanish at (m, 0) for all , since the functions Fi and Gi vanish at m. The integral
curve of V , starting from (m, 0) is therefore defined for all t W , being given by
t (m,t). It follows that by eventually shrinking U to a smaller neighborhood of m
and W to a smaller neighborhood of [0, 1], we may assume that the flow t of V is
welldefined on U {0} for all t W .
It remains to be shown that the time 1 flow 1 of V sends U {0} to a neighborhood of (m, 1) in N1 . Consider the submanifold N of U W , defined by the
common zeros of the functions Qi and Pi ,
N := (m , ) U W  Qi (m , ) = Pi (m , ) = 0 for 1 i r .
The vector field V is tangent to N , since,
V [Qi ] (m , ) =
Qi
(m , ) + XH [Qi ](m , ) = Fi (m ) Fi (m ) = 0 ,
and similarly V [Pi ] (m , ) = 0, for all (m , ) N and all i = 1, . . . , r. The time 1
flow of V maps U {0} to a neighborhood of (m, 1) in U {1}. Since it preserves N , it maps (N0 U ) {0} to a neighborhood of (m, 1) in N1 {1}, as was
to be shown.
It follows from Theorem 5.28 that all sdimensional submanifolds which are transverse to the symplectic leaf which contains m, are isomorphic as Poisson manifolds
in a neighborhood of m, when they are equipped with their reduced Poisson structure.
We show in the following proposition that all sdimensional submanifolds which
are transverse to a given symplectic leaf S of M are isomorphic as Poisson manifolds in a neighborhood of their intersection point with S , when they are equipped
with their reduced Poisson structure.
Proposition 5.29. Let (M, ) be a Poisson manifold and let S be a symplectic leaf
of (M, ). We denote the dimension of S by 2r and we let s := dim M 2r. Assume
that we are given:
(1)
(2)
Two points m, m which belong to S ;
Two sdimensional submanifolds N and N of M, which are transverse to S
at m and at m respectively.
Then there exist neighborhoods V and V of m and m in N and N respectively, such
that:
(1)
(2)
V and V are PoissonDirac submanifolds of (M, );
Equipped with their reduced Poisson structure, V and V are isomorphic as
Poisson manifolds.
Proof. Let m and m be two points of M which belong to the same symplectic leaf
S of (M, ). By definition, such points can be joined by a piecewise Hamiltonian
5.3 PoissonDirac Reduction
147
path. It is therefore sufficient to prove the proposition for points m and m which
can be joined by a Hamiltonian path. Assume therefore that U is an open subset
of M, that H is a function on U and that the flow t of XH is defined for t in a
neighborhood of [0, 1] and for all points of U, with in particular 1 (m) = m . Since
t is the flow of a Hamiltonian vector field, 1 is a Poisson diffeomorphism between a neighborhood U of m and a neighborhood 1 (U) of 1 (m) = m . Since
1 preserves the Poisson structure, it sends the symplectic leaf S through m to the
symplectic leaf S through 1 (m); moreover, since 1 is a diffeomorphism, it sends
every submanifold, transverse to S at m, to a submanifold transverse to S at m ,
and 1 realizes a Poisson diffeomorphism between them, when both are equipped
with the Poisson structure which they inherit from (M, ) as PoissonDirac submanifolds. Since we know from item (2) of Theorem 5.28 that the transverse Poisson
structure is independent of the chosen transversal through a given point, this shows
in addition that the transverse Poisson structure is independent of the chosen point
on a given symplectic leaf of (M, ).
Theorem 5.28 and Proposition 5.29 motivate and justify the following definition.
Definition 5.30. Let S be a symplectic leaf of a Poisson manifold (M, ). Let N
be an arbitrary submanifold of M, transverse to S at some point m S , and let
denote the reduced Poisson structure, which is defined on a neighborhood of m in N.
The germ of at m is called the transverse Poisson structure to S .
We have seen two particular situations in which one constructs a (natural) representative of the transverse Poisson structure to a symplectic leaf S in a Poisson manifold (M, ). The first one is Weinsteins splitting theorem (Theorem 1.25),
which allows us to write in terms of splitting coordinates on a neighborhood of a
point m S as
+ k (z)
,
pi 1k<s
zk z
i=1 qi
r
(5.45)
where the functions k are (smooth or holomorphic) functions, which depend on
z = (z1 , . . . , zs ) only, and which vanish when z = 0; in this case, the second sum in
(5.45) is a Poisson structure, which is a representative for the transverse Poisson
structure to the symplectic leaf S , while the first sum in (5.45) yields the canonical
Poisson structure on the symplectic leaf S . This shows that the decomposition of
Poisson structures, which appears in Weinsteins splitting theorem, is unique. The
second situation is when one selects a submanifold N of M, which is transverse to
the symplectic leaf S ; in this case, Theorem 5.28 says that N is in a neighborhood V
of the intersection point {m} := S N a PoissonDirac submanifold of (M, ) and
the reduced Poisson structure on V is a representative for the transverse Poisson
structure to the symplectic leaf S .
148
5 Reduction
5.4 Poisson Structures and Group Actions
Poisson manifolds often come with symmetry groups which lead, in favorable cases,
to new Poisson manifolds. If a Lie group G acts on a Poisson manifold, preserving
the Poisson structure, then there are three natural constructions which lead to new
Poisson manifolds:
By Poisson reduction, the quotient M/G, assumed to be a manifold, inherits a
Poisson structure;
By PoissonDirac reduction, the set of fixed points of the action, assumed to be
a manifold, inherits a Poisson structure;
By Poisson reduction, the action yields a Poisson structure on certain quotients
of the fibers of the momentum map, assuming that the latter map exists.
Similarly, such new Poisson structures arise in the case of an affine Poisson variety,
when an algebraic group is acting. As indicated in the three items above, this section
is an application of the previous sections, in the particular case of a Poisson manifold
(or affine Poisson variety) upon which a group is acting.
5.4.1 Poisson Actions
There are two natural types of actions of a group on a Poisson manifold or on a
Poisson variety, as given in the following two definitions.
Definition 5.31. Let G be a group and let (M, ) be an affine Poisson variety or
a Poisson manifold. We say that an action : G M M preserves the Poisson
structure when, for every g G, the map g : M M, defined for all m M by
g (m) := (g, m), is a Poisson map.
Definition 5.32. Let G be a group acting on M, where M and G are either one of the
following:
(M, { , }) and G = (G, { , }G ) are affine Poisson varieties and G is an affine
algebraic group;
(M, { , }) and G = (G, { , }G ) are Poisson manifolds and G is a Lie group.
Then we say that the action is a Poisson action when the action map : G M M
is a Poisson map, where G M is equipped with the product Poisson structure.
When G M M is a Poisson action and the Poisson structure on G is trivial, then
the product bracket { , }GM at (g, m) G M reduces, in view of (2.12), to
{F , G }GM (g, m) = F g , G g (m) ,
so that a Poisson action is in this case precisely an action preserving the Poisson
structure. In general, however, a Poisson action does not preserve the Poisson structure.
5.4 Poisson Structures and Group Actions
149
5.4.2 Poisson Actions and Quotient Spaces
In this section, we show that a Poisson action G M M leads to a Poisson structure on the quotient space M/G and we give conditions on a Ginvariant subvariety
N of M, which lead to a Poisson structure on N/G. The following proposition gives
sufficient conditions for M/G to inherit a Poisson structure from M and G.
Proposition 5.33. Let : G M M be a group action, where either one of the
following is satisfied:
(1)
(2)
(M, { , }) and (G, { , }G ) are affine Poisson varieties, with G being a reductive algebraic group;
(M, { , }) and (G, { , }G ) are Poisson manifolds, with G being a Lie group,
and the action is proper and locally free.
If is a Poisson action, then M/G carries a unique Poisson structure such that the
canonical projection p : M M/G is a Poisson map.
Proof. In the case of an affine Poisson variety, M/G is an affine variety. According
to Proposition 5.7, in which (1) is automatically satisfied, (M, M, M/G) is Poisson
reducible as soon as the Ginvariant functions on M form a Lie subalgebra of F (M),
in formulas
F (M)G , F (M)G F (M)G .
(5.46)
We show that (5.46) is a consequence of the fact that is a Poisson action. To
do this, first notice that a function F F (M) belongs to F (M)G if and only if
F = p2 F where p2 : G M M is the projection onto the second component.
Thus, if F, G F (M)G , then the fact that is a Poisson map implies that
{F, G} = { F, G}GM = {p2 F, p2 G}GM = p2 {F, G} ,
(5.47)
where we used in the last step that p2 is a Poisson map (see Proposition 2.2). This
shows, in view of the above characterization of Ginvariant functions, that {F, G}
F (M)G . Hence, F (M)G is indeed a Lie subalgebra of (M, { , }), which achieves
the proof for the case of affine Poisson varieties.
In the manifold case, M/G is a manifold since the action is proper and locally
free. Functions which are defined on open subsets of M are constant on the fibers
of p : M M/G if and only if they are Ginvariant. Therefore, (M, M, M/G) is,
in view of Proposition 5.11, Poisson reducible if and only if the Poisson bracket
of every pair of Ginvariant functions, defined on an arbitrary open subset of M,
is Ginvariant. This is checked precisely as in (5.47), taking for F and G arbitrary
functions, which are defined on an open subset of M.
In applications, one often considers the quotient of a Ginvariant4 subvariety or
submanifold N of M and while N does not inherit a Poisson structure from M, the
4
Saying that N is Ginvariant means that (g, n) N for every (g, n) G N.
150
5 Reduction
quotient N/G does. This is the content of the following generalization of Proposition 5.33, which is proved in exactly the same way as a direct application of Proposition 5.7 in the case of an affine Poisson variety, and of Proposition 5.11 in the case
of a Poisson manifold.
Proposition 5.34. Let : G M M be a group action and let N be a Ginvariant
subset of M; denote by p : N N/G the quotient map. Suppose that either one of
the following is satisfied:
(1)
(2)
(M, { , }) and (G, { , }G ) are affine Poisson varieties, with G being a reductive algebraic group, and N is an affine subvariety of M;
(M, { , }) and (G, { , }G ) are Poisson manifolds, with G being a Lie group,
N is a submanifold of M and the action of G on N is proper and locally free.
Suppose that, in addition, the following conditions hold:
(1)
(2)
is a Poisson action;
For every function F, whose restriction to N is Ginvariant, the Hamiltonian
vector field XF is tangent to N; it suffices that this property holds for F
F (M) in the affine variety case, but it is required to hold for F defined on
arbitrary open subsets in the manifold case.
Then N/G carries a unique Poisson structure { , }N/G such that for every open
subset V of N/G and for every pair of functions F, G F (V ), and for all n
p1 (V ),
G (n) = {F, G}
(p(n)) ,
F,
N/G
where F and G are arbitrary extensions of F p and G p to an arbitrary open
subset of M.
5.4.3 Fixed Point Sets as PoissonDirac Submanifolds
In this section we show that, under some general assumptions, if a group action on
a Poisson variety or manifold leaves the Poisson structure invariant, then the fixed
point set of the action is a PoissonDirac subvariety or submanifold. We do this in
the case of a finite or linear algebraic group, acting on an affine variety; see the end
of this section for the case of Lie groups acting on manifolds.
Let G be a finite group or linear algebraic group, acting on an affine variety M.
We write both g (m) and gm for the action of g G on m M, while the induced
action on F F (M) is denoted by g (F), as in (5.4). We denote the algebra of
invariant functions by F (M)G ,
F (M)G = F F (M)  g (F) = F for all g G ,
and we denote the fixed point set of by M G ,
5.4 Poisson Structures and Group Actions
151
M G := {m M  gm = m for all g G} .
In order to describe the algebraic structure of M G , consider for a fixed g G the
regular map
Lg : M M M
(5.48)
m (m, gm) .
Then M G =
gG
Lg1 ( ), where is the diagonal of M M,
:= {(m, m)  m M} .
Thus, the group action leads to a natural ideal I (M G ) of F (M), consisting of
functions which vanish on M G , namely the ideal of F (M), generated by the functions Lg (I ), where I is the ideal of all regular functions which vanish on ; for
i = 1, 2, if we denote by pi : M M M the canonical projection on the ith component, then I is the ideal of F (M M), generated by all functions p1 (F) p2 (F),
where F F (M). It follows that the ideal I (M G ) is generated by the set of all
functions of the form F g (F), where g G and F F (M). Notice that although
the algebraic subset which is defined by (the common zeros of the elements of)
I (M G ) coincides with M G , the latter ideal is in general neither prime nor reduced.
Proposition 5.35. Let G be a finite group or an algebraic group, acting on an affine
Poisson variety (M, { , }). If the group action preserves the Poisson structure, then
F (M)G is contained in the normalizer N (I (M G )) of I (M G ). As a consequence,
if
(5.49)
F (M)G + I (M G ) = F (M) ,
then I (M G ) is a PoissonDirac ideal of F (M).
Proof. Let F F (M)G be a Ginvariant function. For all G F (M) and all g G,
we have
F, G g (G) = {F, G} g (F), g (G) = {F, G} g ({F, G}) ,
where we used in the last step that every map g is a morphism of Poisson algebras.
Since I (M G ) is generatedby all functions of
the form G g (G), with g G and
G F (M), the inclusion F (M)G , I (M G ) I (M G ) follows, i.e., F (M)G is
contained in the normalizer of I (M G ). In view of (iii) in Definition 5.17, I (M G )
will be a PoissonDirac ideal of F (M) if F (M)G + I (M G ) = F (M).
We consider two particular cases of the above proposition, which both lead to the
result that the fixed point set of the group action is a PoissonDirac subvariety. We
first consider the simpler case in which G is a finite group.
Proposition 5.36. Let G be a finite group, acting on an affine Poisson variety
(M, { , }). If the action preserves the Poisson structure, then the fixed point set
M G is a (not necessarily irreducible) PoissonDirac subvariety of (M, { , }). For
F, G F (M G ), their reduced bracket is given, at n M G , by
152
5 Reduction
{F, G}M G (n) =
1
G2 g
1 ,g2 G
(n) = 1 g (F),
g2 (G)
G (n)
g1 (F),
G gG
G F (M) are arbitrary extensions of F, G.
where F,
Proof. By averaging the action over the group, one constructs a linear projection
: F (M) F (M)G ; explicitly, (F) is given, for F F (M), by
(F) :=
1
g (F) ,
G gG
(5.50)
where G stands for the order of the group G. Since
F (F) =
1
(F g (F)) ,
G gG
it is an element of I (M G ), which shows that F (M)G + I (M G ) = F (M). In view
of Proposition 5.35, it follows that I (M G ) is a PoissonDirac ideal of F (M).
Since G is finite, the (dual) action of G on F (M) is completely reducible, so that
the ideal I (M G ) is the ideal of all regular functions, vanishing on M G , so that M G
is a (not necessarily irreducible) PoissonDirac subvariety. According to (5.33), the
G
G
PoissonDirac bracket
of two functions F, G F (M ) is given, at n M , by
G (n), where F and G are arbitrary Ginvariant extensions of
{F, G}M G (n) = F,
F and G to M. Let F and G be arbitrary extensions of F and G to M. Clearly, the
coincide with F and G on M G , so we can take
and (G)
Ginvariant functions (F)
to compute
and G := (G),
F := (F)
(n) = 1
(G)
{F, G}M G (n) = (F),
G2 g
1 ,g2 G
(n) ,
g2 (G)
g1 (F),
I (M G ), the latter line can also be written as
for all n M G . Since G (G)
G (n) = 1 g (F),
G (n) ,
{F, G}M G (n) = (F),
G gG
for all n M G .
As a general principle in invariant theory, results which hold for the action of a
finite group, acting on an affine variety, can be extended to the action of a reductive
group, acting on an affine variety, by replacing the above averaging operator with
the Reynolds operator (see [190]). It leads to the following result.
Proposition 5.37. Let G be a reductive group, acting on an affine Poisson variety
(M, ). If the action preserves the Poisson structure, then the fixed point set M G is
a (not necessarily irreducible) PoissonDirac subvariety of (M, ).
For a differential geometric analog of the proposition, see [54, Th. 2.1].
5.4 Poisson Structures and Group Actions
153
5.4.4 Reduction with Respect to a Momentum Map
Let G be a Lie group, with Lie algebra g. Suppose that G is acting on a Poisson
manifold (M, ). It is natural to demand that the fundamental vector fields x, with
x g, are Hamiltonian. It leads to the following definition.
Definition 5.38. Let G be a Lie group acting on a Poisson manifold (M, ). The
action is said to be a Hamiltonian action if there exists a Lie algebra homomorphism
: g F (M)
x
(5.51)
such that the following diagram is commutative:
X1 (M)
(5.52)
F (M)
where the horizontal arrow is the map x x, and where we recall that X is a
morphism of Lie algebras (see (5) of Proposition 1.4 and (1.7)). We call a comomentum map of the action. The dual map,
: M g
m (m) ,
(5.53)
where (m) g is the linear map x x (m), is said to be a momentum map of the
action.
We show in the following proposition that, under some assumptions, the orbit space
1 (0)/G inherits a Poisson structure from M.
Proposition 5.39. Let G be a connected Lie group and let (M, ) be a Poisson manifold. We assume that G acts locally freely and properly on M and that the action is
Hamiltonian. Let denote a momentum map of the action. If 0 is a regular value of
, then
(1)
(2)
(3)
1 (0) is an embedded submanifold of M, invariant under the action of G;
1 (0)/G is a manifold;
The triple (M, 1 (0), 1 (0)/G) is Poisson reducible.
The different spaces involved in the theorem are represented in the following diagram:
154
5 Reduction
1 (0)
(5.54)
1 (0)/G
Proof. Let us denote N := 1 (0). It is an (embedded) submanifold of M since 0 is
a regular value of . We claim that, for every x g, the vector field x istangent to
N.
Indeed, N is precisely the zero locus of the vector space of functions y  y g ,
m M  y (m) = 0 ,
N=
yg
and this vector space is invariant for all vector fields x, since,
x [ y ] = X x [ y ] = x , y = [x,y] ,
where we have used in the last step that is a Lie algebra homomorphism. As
a corollary, since G is connected, N is Ginvariant. The quotient space N/G is a
manifold because the action of G on M is locally free and proper. This proves (1)
and (2).
In order to prove (3), we check that the assumptions of Proposition 5.14 are
satisfied. The fibers of the canonical projection p : N N/G are connected, since G
is connected. Therefore, we only need to verify that (5.30) holds, i.e., that
n (Tn N) = Tn pn
(5.55)
for every n N, where we recall that pn is the fiber of p which passes through n.
Since 0 g is a regular value for the momentum map , the space Tn N is, for
every n N, given by
Tn N = {dn x  x g} .
Since n (dn x ) = X x n = xn , this means that
n (Tn N) = {xn  x g} = Tn pn ,
which proves (5.55), and hence that (M, 1 (0), 1 (0)/G) is Poisson reducible.
Remark 5.40. Let G be a reductive algebraic group, acting on an affine variety M.
Then one can define as in the case of a Lie group, acting on a manifold, the fundamental vector field x for all x g, the Lie algebra of G. If M is equipped with a Poisson structure, one can define the notion of a momentum map as in Definition 5.38.
The algebraic version of Proposition 5.39 takes then the following form: let G be a
reductive algebraic group, acting on an affine Poisson variety (M, ). If the action is
Hamiltonian with momentum map , then 1 (0) is Ginvariant and 1 (0)/G is
5.5 Exercises
155
a Poisson variety with respect to a Poisson structure obtained by Poisson reduction
from the diagram (5.54). In order to prove this, one uses Proposition 5.7 instead of
Proposition 5.14.
Remark 5.41. Also, one can generalize Proposition 5.39 and obtain a Poisson structure on 1 ( )/G for every regular value g of , where G is the subgroup of G, defined in terms of the adjoint action Ad : G g g, by G :=
g G  Adg = . See [144] for this and for other generalizations.
5.5 Exercises
1. Let G be a Lie group, with Lie algebra g. For r 2 g, show that the following
are equivalent:
r is a Poisson structure on G;
(i)
(ii)
r is a Poisson structure on G;
(iii)
[[r, r]] = 0.
Show that, when these equivalent conditions are satisfied, the Poisson structures
r
and r are compatible. For r of the form r = x y, where x, y g, show that these
conditions are satisfied if and only if x and y span a Lie subalgebra of g.
2. Let M be a manifold on which a Lie group G acts. For x g, the Lie algebra
of G, denote by x the fundamental vector field, associated to x. Let r = ki=1 xi yi
2 g, satisfying [[r, r]] = 0 and consider the bivector field r := ki=1 xi yi on M.
a.
b.
c.
Show that r is a Poisson structure on M;
Prove that the action of (G,
r
r ) on (M, r) is a Poisson action;
Suppose that M is a vector space and the action of G on M is a representation.
Show that r is a quadratic Poisson structure (see Section 8.2).
3. Let M1 and M2 be Poisson manifolds and endow M1 M2 with the product Poisson bracket. Show that for every m1 M1 (respectively m2 M2 ), the submanifold
{m1 } M2 (respectively M1 {m2 }) is a PoissonDirac submanifold of M1 M2 .
4. Let M be a Poisson manifold, and let N be a PoissonDirac submanifold of M.
Suppose that V is a Poisson vector field on M, which is tangent to N at every point
of N. Show that the restriction of V to N is a Poisson vector field.
5. Let P and M be Poisson manifolds and let : P M be a submersive Poisson
map. Let N be a submanifold of M.
a.
b.
Show that if N is a coisotropic submanifold of M, then 1 (N) is a coisotropic
submanifold of P;
Show, by giving a counterexample, that if N is a PoissonDirac submanifold
of M, then 1 (N) need not be a PoissonDirac submanifold of P.
156
5 Reduction
6. We assume in this exercise that the reader is familiar with the notion of symplectic manifold (see Section 6.3). Let (M, ) be a symplectic manifold, and denote
by the canonical Poisson structure, associated to . Let N be a submanifold of M.
Show that the following are equivalent:
(i)
(ii)
N is a PoissonDirac submanifold of (M, );
The restriction of to N is nondegenerate.
Conclude that for symplectic manifolds the notions of symplectic submanifold and
PoissonDirac submanifold coincide.
7. We assume in this exercise that the reader is familiar with the LiePoisson structure on the dual of a Lie algebra (see Chapter 7). Let g be a finitedimensional
Lie algebra and let g = g m be a vector space decomposition of g, where
g := {x g  adx = 0} denotes the centralizer of an element g . We consider
the affine subspace N := + m of g , which is isomorphic to g , via the affine
map , which is defined, for all N, by ( ) := ( )g .
a.
b.
Show that, if m satisfies [g , m] m, then N is a PoissonDirac submanifold of
g , equipped with its LiePoisson structure, and the reduced Poisson structure
on N is isomorphic to the LiePoisson structure on g , via .
As an example, consider g = gld (F), let x g and let g be the linear map,
defined by (y) := Trace(xy), for all y g. Show that if x is diagonalizable, then
there exists a complementary subspace m of g in g, such that [g , m] m.
8. Let (M, { , }) be a Poisson manifold and let N be a submanifold of M. We
endow F (M)[ ]/ 2 with the structure of an F[ ]algebra, defined by
F G := FG + {F, G}
for all F, G F (M) (see Exercise 8 of Chapter 3, where this product is proven to
be associative). Suppose that
: F (M)[ ]/ 2 F (N)[ ]/ 2 F (N)[ ]/ 2
defines the structure of a left module over the F[ ]algebra (F (M)[ ]/ 2 , ) on
F (N)[ ]/ 2 , satisfying (F, H) = FN H (mod ) for every F F (M) and for
every H F (N). Let : F (M) F (N) F (N) be the bilinear map, given for
all F F (M) and H F (N) by
(F, H) FN H
(F, H) =
.
=0
a.
Show that satisfies the identity
{F, G}N H = (F, GN H) (FG, H) + FN (G, H)
b.
for all F, G F (M) and H F (N);
Deduce from it that N is a coisotropic submanifold of M.
(5.56)
5.6 Notes
157
5.6 Notes
Lie groups and Lie algebras are standard tools in geometry and in physics; in turn,
the study of their structure and their representations is strongly inspired by ideas
which come from geometry and physics. A quick introduction to Lie groups is given
in Warner [198]; see Bump [29] for a more extensive treatment. For the theory of
Lie algebras, we refer to Humphreys [99]. A great introduction to both Lie groups
and Lie algebras and their representation theory is given in FultonHarris [80].
The starting point of reduction theory comes from the fact that integrals of motion
permit to eliminate degrees of freedom in Hamiltonian systems. One may cite here
Noethers theorem, which is the predecessor of the notion of momentum map and
the theory of Dirac constraints, which eventually led to the notion of PoissonDirac
reduction. The modern, geometrical development of these ideas first appears in the
seminal papers by SniatyckiTulczyjew
[184], Meyer [148] and MarsdenWeinstein
[146]. See KosmannSchwarzbach [114], Weinstein [199] and the introduction of
OrtegaRatiu [160] for more details on the history of the subject; in the latter introduction one also finds a detailed list of topics in the theory of moment maps and
reduction.
In our presentation of reduction, we have separated the Poisson reduction of
coisotropic submanifolds from the theory of reduction of PoissonDirac submanifolds, which are both particular cases of the socalled MarsdenRatiu reduction
procedure [144]. The notion of momentum map, presented here, has also been generalized in many different ways, see for example AlekseevMalkinMeinrenken [9]
for Lie group valued momentum mappings and Lu [133] for PoissonLie group valued momentum mappings. These generalizations and their associated reduction can
be nicely understood with the help of Lie groupoids, see Xu [208].
We have not discussed the geometry of the image of a momentum map (which is,
under some assumptions, a convex polytope). See Guillemin [91] and the references
therein.
Part II
Examples
Chapter 6
Constant Poisson Structures, Regular
and Symplectic Manifolds
Poissons original bracket on R2r , given for smooth functions F and G by
r
F G G F
{F, G} :=
,
qi pi
i=1 qi pi
(6.1)
is of fundamental importance in classical mechanics, in quantum mechanics and in
other areas of mathematical physics. In fact, locally, the Poisson structure of every symplectic manifold of dimension 2r is of this form, and symplectic manifolds
appear as phase spaces of many mechanical systems. The rank of the Poisson structure defined by (6.1), is maximal at every point; if we replace maximality of the
rank by constancy of the rank (i.e., the rank is independent of the point) we arrive at
the notion of a regular Poisson structure. Locally the Poisson structure of a regular
Poisson manifold is also given by (6.1), but 2r denotes now the rank of the Poisson
structure, and not the dimension d of the manifold, which may be larger: there are
d 2r local coordinates which are absent in the formula for the bracket. Notice that
the Poisson bracket (6.1), applied to a pair of coordinate functions, always yields
a constant function: this will be the starting point of the algebraic definition of a
constant Poisson structure. In fact, the Darboux theorem (Theorem 1.26) shows that
regularity of the Poisson structure corresponds locally exactly to constancy of the
Poisson structure (in an appropriate coordinate chart).
Constant Poisson structures are introduced in Section 6.1, and they will be related to regular Poisson manifolds in Section 6.2. The particular class of symplectic
manifolds will be discussed in Section 6.3. Entire (excellent!) books have been written on symplectic manifolds, so we will restrict ourselves here to the most common
classes of examples of symplectic and regular Poisson manifolds, and our focus will
be on their properties from the Poisson point of view.
Unless stated otherwise, F is an arbitrary field of characteristic zero.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 6, SpringerVerlag Berlin Heidelberg 2013
161
162
6 Constant Poisson Structures, Regular and Symplectic Manifolds
6.1 Constant Poisson Structures
Poissons original bracket on R2r , defined by
r
F G G F
{F, G} :=
,
qi pi
i=1 qi pi
(6.2)
has the property that the Poisson bracket {F, G} of two linear functions F and G is a
constant function on R2r . This property leads to the following definition, which will
cover a first class of basic examples of Poisson structures.
Definition 6.1. A Poisson structure on a finitedimensional vector space V is
called a constant Poisson structure, if for each pair of linear functions F and G
on V , their Poisson bracket [F, G], also denoted by {F, G}, is a constant function
on V .
Let X = (xi j ) Matd (F) be an arbitrary skewsymmetric matrix, where d N . We
show that X defines a Poisson structure on Fd , i.e., a Poisson bracket on F (Fd ),
where we recall that, for a finitedimensional Fvector space V , the notation F (V )
stands for the algebra of polynomial functions on V , in general, but may also be
chosen as the algebra of smooth functions on V , when F = R, or as the algebra
of holomorphic functions on V , when F = C. In order to construct this Poisson
structure, consider the skewsymmetric biderivation of F (Fd ), defined for all F, G
F (Fd ), by
d
F G
{F, G} := xi j
,
(6.3)
xi x j
i, j=1
d
where
each
of
the brack
x1 ,. . . , xd are the standard coordinates on F . Since
xi , x j , xk = 0, where
ets xi , x j = xi j is a constant function, we have that
1 i, j, k d, so that the Jacobi identity holds for every triple of coordinate functions (xi , x j , xk ) for F (Fd ). It implies that the Jacobi identity holds for all triples
of functions in F (Fd ) (see Proposition 1.36), so that (6.3) defines a Poisson structure on Fd , which we denote by . Clearly, is a constant Poisson structure and
every constant Poisson structure on Fd is of this form. Since every ddimensional
vector space V can be identified with Fd , by choosing a basis of V , the matrix X
defines a Poisson structure on every ddimensional vector space V , equipped with a
basis (equivalently, with a system of linear coordinates). Thus, the following proposition holds.
Proposition 6.2. Let V be a finitedimensional vector space, with a fixed system
of linear coordinates (x1 , . . . , xd ). There is a onetoone correspondence between
skewsymmetric matrices X Matd (F) and constant Poisson structures on V . It is
defined by assigning to X = (xi j ) Matd (F) the Poisson structure, defined for all
F, G F (V ) by
d
F G
{F, G} := xi j
.
(6.4)
xi x j
i, j=1
6.1 Constant Poisson Structures
163
Example
original bracket (6.2) corresponds to the skewsymmetric
6.3. Poissons
0 1r
matrix
, where the standard coordinates on R2r are ordered as follows:
1r 0
q 1 , . . . , qr , p 1 , . . . , p r .
The matrix X in Proposition 6.2 is the Poisson matrix of with respect to the coordinates x1 , . . . , xd (see Section 1.2.2). Since the entries of X are constant, the rank
of is constant. According to the Darboux theorem (Theorem 1.26), if the rank of
a Poisson structure on a manifold is locally constant at a point m, then there exist
local coordinates around m in which the Poisson structure takes a canonical form
(socalled Darboux coordinates). We show that in the case of a constant Poisson
bracket on a ddimensional vector space V , linear coordinates which are global Darboux coordinates can be constructed explicitly. Namely, since X is skewsymmetric,
there exists an invertible matrix A Matd (F) such that
0 1r 0
A XA =
1r 0 0 ,
0
0 0
where 2r is the rank of X and hence of the Poisson bracket (at every point). It
means that, in terms of new linear coordinates on V , defined in terms of x1 , . . . , xd
by
(q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd2r ) := (x1 , . . . , xd )A
one obtains the following brackets
qi , p j = i, j ,
qi , q j = pi , p j = {qi , zk } = {pi , zk } = {zk , z } = 0 ,
where 1 i, j r and 1 k, d 2r. As a consequence, the d coordinates
q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd2r are Darboux coordinates for . Thus, by simple
linear algebra, we can construct (global) Darboux coordinates for every constant
Poisson structure on V . In terms of these Darboux coordinates, the Hamiltonian
vector field XH which corresponds to H F (V ) takes the well known form
qi =
H
,
pi
pi =
H
,
qi
zk = 0 ,
where qi is a commonly used shorthand for XH [qi ] = {qi , H}, and similarly for the
other variables.
In order to obtain a more intrinsic formulation of the notion of a constant Poisson structure on a finitedimensional vector space V , we consider the vector space
2V = V V , whose elements are called bivectors of V . A bivector b of V is often
viewed as a linear map, or as a bilinear form, on V . This is done as follows: writing
b = si=1 vi wi , where vi , wi V , we define1 the linear map b : V V , by
In terms of internal products, as recalled in Appendix A, b ( ) = b, which shows that b , as
given by (6.5), is welldefined.
1
164
6 Constant Poisson Structures, Regular and Symplectic Manifolds
s
b ( ) = ( , vi wi , wi vi ) ,
(6.5)
i=1
for every V , and we define a skewsymmetric bilinear form b on V , by
s
, ) := , b ( ) = ( , vi , wi , vi , wi ) ,
b(
(6.6)
i=1
for all , V , where , denotes the canonical pairing between
V and V .
Both b and b are skewsymmetric, which means for the former that , b ( ) =
, b ( ) , for all , V . In terms of a basis (e1 , . . . , ed ) of V and the dual ba , e ) ei e j ,
sis (e1 , . . . , ed ) of V , the bivector b can be written as b = 1i< jd b(e
i
j
so that b can be recovered from b (and hence from b ).
To a bivector b V V one naturally associates a bivector field on V , by identifying all cotangent spaces to V with V , which can be done thanks to the affine
structure of V . If F F (V ) and m V then we can view dm F as an element of V ,
hence we can define a bivector field on V , by setting, for F, G F (V ) and m V ,
{F, G} (m) := b (dm F, dm G) ,
(6.7)
where b : V V F is the bilinear form, associated to b, as in (6.6). Equivalently,
we can view b under the same identification as a bilinear form b m on TmV , for
any m V , and the righthand side of (6.7) can be written as b m (dm F, dm G), where
the pairing now takes place between 2 TmV (2 TmV ) and 2 TmV , rather than
between 2V (2V ) and 2V .
For linear functions F, G on V , we have that {F, G} is independent of m, i.e., is
constant; first, this implies that the bivector field defined by (6.7) satisfies the Jacobi
identity, so that it is a Poisson structure, and second that it is a constant Poisson
structure. For H F (M), the Hamiltonian vector field XH at m is given, as an
element of V , by
(XH )m = b (dm H) .
(6.8)
Writing b as b = si=1 vi wi , where vi , wi V , the formula for the Poisson bracket
can also be written as
s
{F, G} = (vi [F]wi [G] vi [G]wi [F]) .
(6.9)
i=1
The square brackets in this formula have the following meaning: for v V , v[] is the
unique derivation of F (V ) such that v[ ] = , v for all linear functions on V .
Geometrically, we think of v[] as the directional derivative, defined by a vector v.
Equation (6.9) is a coordinatefree analog of (6.4).
The fact that every bivector on V defines a Poisson structure on V , motivates the
following definition.
6.2 Regular Poisson Manifolds
165
Definition 6.4. A Poisson vector space (V, b) is a finitedimensional vector space V ,
equipped with a bivector b V V .
By the above, we have the following abstract version and amplification of Proposition 6.2.
Proposition 6.5. For every finitedimensional vector space V , there is a onetoone
correspondence between bivectors on V (making V into a Poisson vector space)
and constant Poisson structures on V . If b 2V is a bivector of rank 2r, then the
Poisson structure = { , } on V which corresponds to b, has rank 2r and V admits
a system of linear coordinates (q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs ) in which the Poisson
structure is given by
r
F G G F
{F, G} =
,
qi pi
i=1 qi pi
for all F, G F (V ).
6.2 Regular Poisson Manifolds
In the case of a Poisson manifold (M, ), there is no notion of linear functions on M,
so we cannot use Definition 6.1 as it stands, to define constant Poisson structures on
manifolds. Two natural generalizations of the latter definition are the following.
Definition 6.6. A Poisson structure on a connected manifold M is said to be a
constant Poisson structure if for every point m M,thereexists a coordinate chart
(U, x), centered at m, such that all Poisson brackets xi , x j are constant on U.
Definition 6.7. A Poisson structure on a manifold M is said to be a regular Poisson
structure if it has the same rank at every point m M. We say then that (M, ) is a
regular Poisson manifold.
Comparing the definitions, it is clear that every constant Poisson structure on a connected manifold is regular: taking around
an arbitrary
point m M a coordinate
chart (U, x) such that all Poisson brackets xi , x j are constant on U, we see that
matrix
the rank of is constant on U, since it is given by the rank ofthe Poisson
of with respect to x, whose entries are the Poisson brackets xi , x j ; since M is
connected, it follows that the rank of takes the same value at every point of M,
i.e., is a regular Poisson structure. It is a fundamental fact that the converse is also
true, leading to the following proposition.
Proposition 6.8. Let (M, ) be a Poisson manifold, which is assumed to be connected. Then (M, ) is a regular Poisson manifold if and only if is a constant
Poisson structure.
166
6 Constant Poisson Structures, Regular and Symplectic Manifolds
Proof. Let (M, ) be a regular Poisson manifold and let m M. We need to show
thatthere
exists a coordinate chart (U, x), centered at m, such that all Poisson brackets xi , x j are constant (functions) on U. Since, by assumption, the rank of is
constant at m, this is an immediate consequence of the Darboux theorem, which
we proved in Section 1.3.3 (Theorem 1.26). According to this theorem, if the rank
of a Poisson structure is constant (say, equal to 2r) in the neighborhood of a point
m M, then there exists in a neighborhood of m a coordinate chart (U, x), with
x = (q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd2r ), centered at m, in which the Poisson structure takes the canonical form
qi , q j = pi , p j = {qi , zk } = {pi , zk } = {zk , z } = 0 ,
qi , p j = i, j ,
where 1 i, j r and 1 k, d 2r. In particular, all these Poisson brackets
are constant. The Darboux coordinates q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd2r are often
referred to as canonical coordinates, especially in the physics literature.
The study of regular Poisson manifolds is in general much simpler than the study of
arbitrary Poisson manifolds. Being locally the product of a symplectic manifold and
Rs or Cs with the zero Poisson structure, their (local) study can often be reduced
to the case of symplectic manifolds. But sometimes, it is the regularity itself which
can be exploited, as we show in the following example.
Example 6.9. According to Proposition 1.21, the rank of a Poisson manifold (M, )
at a point m is equal to the dimension of the vector space of Hamiltonian vector
fields at that point, Rkm M = dim Hamm (M). Therefore, if the rank is constant,
the Hamiltonian vector fields define a distribution of rank Rk M. It is differentiable, because it is defined by smooth vector fields. According to Proposition 1.4,
[XF , XG ] = X{F,G} , so the distribution is integrable in the sense of Frobenius, so
through every point there passes a unique integral manifold of the distribution, by
Frobenius theorem. By definition, all Hamiltonian vector fields of (M, ) are tangent to this integral manifold, at each of its points. By Proposition 2.10, this integral
manifold carries a (unique) Poisson structure of maximal rank. In this way we have
recovered, by simple means, the symplectic foliation2 of a Poisson manifold, in the
regular case.
6.3 Symplectic Manifolds
The most important class of regular Poisson manifolds consists of the symplectic
manifolds (real or complex). We first consider the particular case of a symplectic
vector space.
2
In this case it is indeed a foliation, and not a singular foliation.
6.3 Symplectic Manifolds
167
6.3.1 Symplectic Vector Spaces
The skewsymmetric variant of an inner product space is a symplectic vector space.
Namely, let V be a finitedimensional Fvector space and let Hom(V V, F)
be a skewsymmetric bilinear form on V . We say that is a symplectic structure
if has maximal rank, Rk = dimV . Then (V, ) is called a symplectic vector
space. If (V, ) is a symplectic vector space, then V is evendimensional, since the
rank of every skewsymmetric bilinear form is even. Since the rank of a symplectic
structure is maximal, the linear map , which is by definition , viewed3 as
an element of Hom(V,V ), is invertible. Its inverse ( )1 Hom(V ,V ) is skewsymmetric, hence is of the form for some (unique) bivector 2V . It follows that every symplectic vector space (V, ) is in a natural way a Poisson vector
space, and hence, according to Proposition 6.5, V is equipped with a constant Poisson structure , called the canonical Poisson structure, associated to . Explicitly,
the Poisson bracket of F, G F (M) is computed at m M from = ( )1
using (6.7) and (6.6), namely
{F, G} (m) = (dm F, dm G) = dm G, (dm F)
= dm G, ( )1 (dm F)
= (( )1 (dm F), ( )1 (dm G))
(6.10)
= ((XF )m , (XG )m ) ,
where we used in the last line that (XF )m = (dm F) (see (6.8)). The latter definition of (XF )m is, in view of the definition of , equivalent to saying that (XF )m
is the unique element of TmV V such that
((XF )m , ) = dm F .
(6.11)
When is viewed as a (constant) differential twoform on V , the formula for the
Poisson bracket is also written as
{F, G} = (XF , XG ) ,
(6.12)
and Eq. (6.11) for the Hamiltonian vector field is shortened to
(XF , ) = dF .
The (constant) rank of equals the rank of , which is dimV .
The above construction is easily reversed: if is a constant Poisson structure
on F (V ) of maximal rank, then it is the canonical Poisson structure of a symplectic
3 This can be done in two ways, which differ by a sign. Our convention is that (v) = (v, ), for
v V.
168
6 Constant Poisson Structures, Regular and Symplectic Manifolds
vector space (V, ), where Hom(V V, F). We leave this as an exercise to the
reader; a globalization of this statement will be worked out in the next section.
6.3.2 Symplectic Manifolds
Let M be a real or complex manifold (F = R or F = C). A (smooth or holomorphic)
differential twoform on M is called an almost symplectic structure if its restriction to each tangent space Tm M is nondegenerate, i.e., m is a symplectic structure on Tm M, for every m M, as defined in Section 6.3.1. Then (M, ) is called
an almost symplectic manifold. The globalization of (6.11) and (6.10) is naturally
given by
1
dm F Tm M ,
(XF )m := m
(6.13)
{F, G} (m) := m ((XF )m , (XG )m ) ,
for m M and F, G F (M). In a compact form, these formulas are usually written as
(XF , ) = dF ,
(6.14)
{F, G} = (XF , XG ) .
As in the case of symplectic vector spaces, these definitions imply that
{F, G} = dF(XG ) = XG [F] ,
(6.15)
so that { , } defines a bivector field on V and XF , as defined by (6.14), is the
Hamiltonian vector field of F with respect to this bivector field. However, the Jacobi
identity does not need to hold. It holds if and only if is closed, d = 0, in which
case is called a symplectic structure and (M, ) is called a symplectic manifold.
This equivalence is shown in the following proposition.
Proposition 6.10. Let (M, ) be an almost symplectic manifold. Then , as defined
by (6.14), is a Poisson structure on M if and only if (M, ) is a symplectic manifold.
In particular, every symplectic manifold (M, ) carries a canonical Poisson structure , which makes (M, ) into a Poisson manifold; it is a regular Poisson structure
whose rank is equal to the dimension of M.
Proof. Let us first assume that is closed. Notice that one important consequence
of this assumption is that is conserved by the flow of every Hamiltonian vector
field. To see this, use Cartans formula (3.50) for the Lie derivative LV ,
LV = V d + dV
(6.16)
when V is a (locally) Hamiltonian vector field, V = XF , and use that XF = dF
and are closed; it yields LXF = 0, for every function F, defined on an open
subset of M. It implies that the following formula,
6.3 Symplectic Manifolds
169
[V1 ,V2 ] = LV1 V2 V2 LV1 ,
which is a special case of (2) in Proposition 3.11, simplifies for Hamiltonian vector
fields to
[XF ,XG ] = LXF XG = dXF XG = d ( (XF , XG )) = d {F, G} , (6.17)
where F and G are functions, defined on an open subset U of M. By the definition
of X , Eq. (6.17) means that [XF ,XG ] = X{F,G} . Nondegeneracy of allows
us to conclude that [XF , XG ] = X{F,G} . Applying this equation to an arbitrary
function H F (U), we find
XF XG [H] XG XF [H] + X{F,G} [H] = 0 ,
which is precisely the Jacobi identity, as follows from (6.15). Thus, we have shown
that if is closed, so that (M, ) is a symplectic manifold, then is a Poisson
structure on M.
Assume now that is an almost symplectic structure, and that the bivector
field , defined by (6.13), satisfies the Jacobi identity. Using the explicit formula (3.28) for computing the de Rham differential, we obtain, for a triple of Hamiltonian vector fields as above,
d (XF , XG , XH ) = XF [ (XG , XH )] + (XF , [XG , XH ]) + (F, G, H)
= 2 ({{F, G} , H} + {{G, H} , F} + {{H, F} , G})
= 0.
We have used that
XF [ (XG , XH )] = {{G, H} , F} = (XF , [XG , XH ]) ,
as follows from (6.14), (6.15) and item (5) in Proposition 1.4. Since m is nondegenerate for every m M, we have that Tm M = Hamm ( ) for every m M; the
above computation therefore shows that d vanishes on every triple of vector fields,
i.e., that is closed, hence that (M, ) is a symplectic manifold.
Proposition 6.11. Suppose that (M, ) is a regular Poisson manifold, whose rank is
equal to the dimension of M. There exists on M a (unique) symplectic structure ,
whose canonical Poisson structure is .
Proof. We consider for m M the linear map m : T M T M, associated to . By
assumption, m is of maximal rank, hence m is an isomorphism for every m M.
It follows that we can use m to associate to each differential oneform on M, a vector field on M: given 1 (M), the corresponding vector field V is defined by its
value at m M as Vm := m (m ). One often uses the compact4 notation V := ( ).
4
Some authors, such as [125] go one step further and write for ( ).
170
6 Constant Poisson Structures, Regular and Symplectic Manifolds
In particular, for local functions F and G on M, one has (FdG) = FXG . Similarly, one uses the inverse of m to associate to each vector field on M a differential
oneform on M; in the compact notation, the differential oneform associated to
1
V X1 (M) is
(V ). We use m to define a nondegenerate
two
differential
1
(V1 ), V2 .
form on M: for vector fields V1 and V2 , let (V1 , V2 ) :=
Then
1
(XF , XG ) :=
(XF ), XG = dF, XG = {F, G} ,
for F, G F (M). It follows that (M, ) is an almost symplectic manifold whose
associated bivector field is . Since is a Poisson structure, Proposition 6.10 implies
that (M, ) is actually a symplectic manifold.
The main properties of symplectic manifolds are listed in the following proposition.
Proposition 6.12. Let (M, ) be a symplectic manifold of dimension d and let
denote its canonical Poisson structure.
(1)
(2)
(3)
(4)
The dimension d of M is even;
M is an orientable manifold;
(M, ) is a unimodular Poisson manifold;
k (M), the
For every k N , with 0 k d, the de Rham cohomology space HdR
k
(M)
Poisson cohomology space H (M) and the Poisson homology space Hdk
are isomorphic,
k
(M) Hk (M) Hdk
(M) .
HdR
Proof. Since every symplectic vector space is evendimensional, every tangent
space of a symplectic manifold is evendimensional, hence also the symplectic manifold itself. In order to prove (2), we assume that M is a real manifold (complex
manifolds are always orientable). Since is nondegenerate (at every point) the
differential dform := d/2 is nowhere vanishing, hence it is a volume form and
M is orientable. The volume form is called the Liouville volume form of (M, ).
It is preserved by the flow of every Hamiltonian vector field XF , since
d/2
LXF = LXF d/2 = i1 LXF d/2i = 0 ,
i=1
where we have used in the last step that LXF = 0 (see the proof of Proposition 6.10). Thus, (M, ) is unimodular; see Section 4.4.4, where it is shown that
for a unimodular Poisson manifold, the Poisson cohomology spaces Hk (M) and the
(M) (with 0 k d) are isomorphic vector spaces,
Poisson homology spaces Hdk
which proves part of (4).
In the case of a symplectic manifold we show that, in addition, the de Rham and
Poisson cohomology spaces are isomorphic. To do this, we consider the isomorphisms m : Tm M Tm M, which we used in the proof of Proposition 6.11. Notice
6.3 Symplectic Manifolds
171
that at every point m, the inverse of m is just m ; for a local section V of T M
1
(i.e., vector field), the inverse of is given by
(V ) = (V , ). The natural
extension of to (M) yields, for every k with 0 k d, a map
k : k (M) k X1 (M) Xk (M) ,
which renders the following diagram commutative:
k (M)
k
Xk (M)
k+1 (M)
k+1
Xk+1 (M)
Indeed, for local functions F, G1 , . . . , Gk we have on the one hand that
k+1 d(F dG1 dGk ) = (dF) (dG1 ) (dGk )
= (1)k+1 XF XG1 XGk ,
while on the other hand
k k (F dG1 dGk ) = k (F XG1 XGk )
= (1)k XF XG1 XGk ,
where we have used in the last step that XG1 , . . . , XGk are Poisson 1cocycles. The
commutativity of the diagram implies that the Poisson complex (X (M), [ , ]S ) and
the de Rham complex ( (M), d) are isomorphic. In particular, they have the same
cohomology.
Remark 6.13. A map : M1 M2 between two symplectic manifolds (M1 , 1 )
and (M2 , 2 ) is called a symplectic map if 2 = 1 . Considering the Liouville
volume form on M1 , it is clear that dim M1 dim M2 and that is an immersion. In
particular, for every m M1 ,
Rk (m) 2 = dim M2 dim M1 = Rkm 1 .
The resulting inequality Rk (m) 2 Rkm 1 , is in sharp contrast with the inequality
Rk (m) 2 Rkm 1 , which is valid for a Poisson map (see Exercise 4 of Chapter 1). Thus, a symplectic map : M1 M2 between two symplectic manifolds
(M1 , 1 ) and (M2 , 2 ) is not necessarily a Poisson map, as a map : (M1 , 1 )
(M2 , 2 ), where i is the canonical Poisson structure, associated to i , for i = 1, 2. It
follows that the map, which associates to a symplectic manifold the corresponding
Poisson manifold, is not functorial.
172
6 Constant Poisson Structures, Regular and Symplectic Manifolds
6.3.3 Symplectic Reduction
In this section, we specialize Poisson reduction, in particular Proposition 5.14, to
the case of a symplectic manifold. The main extra result is that the reduced Poisson
manifold is also a symplectic manifold.
Proposition 6.14. Let (M, ) be a symplectic manifold, N an (immersed or embedded) submanifold of M with inclusion map : N M, and let p : N P be a
surjective submersion with connected fibers. If
n Tn N = Tn pn
(6.18)
for every n N, then N is a coisotropic submanifold of M, the triple (M, N, P) is
Poisson reducible and the reduced Poisson structure on P is regular and of maximal
rank, i.e., it is the canonical Poisson structure, associated to a symplectic structure
P on P. Also, = p P .
Proof. In view of Proposition 5.14, we know that (M, N, P) is Poisson reducible,
hence that P inherits a Poisson structure P from (M, ). In order to show that P
is of maximal rank (equal to the dimension of P) at every point of P, let n N
and suppose that F is a function, defined in a neighborhood of p(n) in P, such that
(XF ) p(n) = 0. Let F be an extension of F p to a neighborhood of n in M. According
to (5.27), dn p (XF )n = 0, so that
(XF )n Tn pn = n Tn N .
Since M is a symplectic manifold, this means that dn F Tn N, i.e., dn FTn N = 0,
so that dn (F p) = dn (FN ) = 0. Since dn p is surjective (as p is a submersion), it
follows that d p(n) F = 0. This proves that P is of maximal rank at p(n). Since p is
surjective, this shows that P is of maximal rank at every point of P; it is therefore
the canonical Poisson structure, associated to a symplectic structure on P, which we
denote by P .
We show that = p P . To do this, let n N and let v, w be arbitrary elements
of Tn N, which we identify with elements of Tn M. Since n is invertible, (6.18) im
plies that n Tn pn = Tn N, so that we can write v as (XF )n , where F is defined in
a neighborhood of n in M and is constant on the fibers of p, so its restriction to N is
of the form FN = F p, where F is a function, defined on a neighborhood of p(n)
in P. On the one hand,
w = p d p(n) F, w ,
(6.19)
( )n (v, w) = n ((XF )n , w) = dn F,
where we have used in the last step that w is tangent to N. On the other hand, using
(5.27) it follows that
6.3 Symplectic Manifolds
(p P )n (v, w) = (P ) p(n) (Tn p(XF )n , Tn p(w))
= (P ) p(n) ((XF ) p(n) , Tn p(w)) = d p(n) F, Tn p(w) .
173
(6.20)
Comparing (6.19) and (6.20) we see that ( )n = (p P )n for every n N, which
we needed to show.
Remark 6.15. The fact that the reduced Poisson structure on P is of maximal rank
(symplectic) can also be read off from its Poisson matrix. Recall from Remark 5.16
that in wellchosen coordinates on an open subset U of M, centered at a point n N,
the Poisson matrix of (M, ) is given by
A
B C
X = B D E
C E F
and that the condition (6.18) implies that CN = FN = 0 (see the last line of Table 5.1). Recall also that the Poisson matrix of the reduced Poisson structure P is
given by the matrix A, defined by A N = A p. Notice that, in the present case, E is
a square matrix, because the dimension t of the fibers of p : N P is equal to the
codimension of N in M, a consequence of condition (6.18), combined with the fact
that is of maximal rank. For every n N U,
) = (det E(n ))2 det A(p(n )) ,
det X(n ) = (det E(n ))2 det A(n
so that det A(p(n )) = 0. This gives an alternative proof of the fact that A, and
hence P , is of maximal rank at every point of P. Notice that the argument also
reproves (in the symplectic case) that Rk(E(n )) = t for every n N U (see again
the last line of Table 5.1).
Combining the above proposition with Proposition 5.39 leads to the following result.
Corollary 6.16. Let G be a connected Lie group and let (M, ) be a symplectic
manifold. We assume that G acts locally freely and properly on M and that the
action is Hamiltonian. If 0 is a regular value of its momentum map , then
(1)
(2)
(3)
1 (0) is an embedded submanifold of M, invariant under the action of G;
1 (0)/G is a manifold;
The triple (M, 1 (0), 1 (0)/G) is Poisson reducible.
The reduced Poisson structure on 1 (0)/G is regular and of maximal rank, i.e.,
it is the canonical Poisson structure, associated to a symplectic structure 0 on
1 (0)/G. Also, = p 0 , where p : 1 (0) 1 (0)/G is the quotient map.
174
6 Constant Poisson Structures, Regular and Symplectic Manifolds
6.3.4 Quotients by Finite Groups of Symplectomorphisms
Let (V, ) be a symplectic Cvector space of dimension 2d. We denote by SP(V ) the
group of all linear symplectomorphisms of V , i.e., the group of all endomorphisms
of V which preserve . Recall from Section 6.3.1 that the symplectic twoform
leads to a (constant) Poisson structure of rank 2d on V , which we called the canonical Poisson of (V, ). Let G be a finite subgroup of SP(V ). The purpose of the
present section is to study the singular affine variety V /G, both as an affine variety
and as a Poisson variety. The case where the dimension of V is equal to 2 will be
studied in more detail in Section 9.2.4.
Let us specialize to the present setting what has been said at the very end of Section 5.1.2 about the quotient of an affine variety by a finite group. Since G is a finite
subgroup of SP(V ), it acts on V , and the quotient space V /G is an affine variety,
whose algebra of regular functions F (V /G) is naturally identified with the algebra
F (V )G of all Ginvariant polynomials on V . The canonical projection p : V V /G
corresponds, dually, to the inclusion of algebras F (V /G) F (V )G F (V ).
Moreover, since every linear symplectomorphism of (V, ) preserves the canonical Poisson structure of (V, ), the finite subgroup G SP(V ) acts by Poisson isomorphisms. As a consequence, according to the first item of Proposition 5.33, V /G
carries a unique Poisson structure with respect to which p : V V /G is a Poisson
map. This justifies the following definition.
Definition 6.17. Let (V, ) be a symplectic Cvector space and let G be a finite
subgroup of SP(V ). The unique Poisson structure { , }V /G on V /G with respect
to which the natural projection map p : V V /G is a Poisson map, is called the
quotient Poisson structure associated to G. The pair (V /G, { , }V /G ) is called the
quotient Poisson variety associated to G.
Since the quotient space V /G is, in general, singular, it makes no sense to ask
whether it is a symplectic manifold. However, the second item of the next proposition implies that V /G is as symplectic as it can be, i.e., { , }V /G is of maximal rank
at every nonsingular point of V /G.
Proposition 6.18. Let (V, ) be a symplectic Cvector space of dimension 2d and
let G = {e} be a finite subgroup of SP(V ), the group of all symplectomorphisms
of V . Denote by p the natural projection map p : V V /G and by U the (nonempty) Zariski open subset of all x V for which the stabilizer Gx is trivial.
(1)
(2)
A point in V /G is a nonsingular point if and only if it belongs to p(U);
The rank of { , }V /G is equal to 2d at every nonsingular point of V /G.
As a consequence, the restriction of { , }V /G to p(U) is a holomorphic Poisson
structure, which is the canonical Poisson structure associated to a symplectic holomorphic twoform on p(U).
Proof. It is a classical result (valid for the action of an arbitrary finite group on a
vector space) that, for every x V , the point p(x) is a nonsingular point of V /G
6.3 Symplectic Manifolds
175
if and only if its stabilizer Gx is a complex reflection group, i.e., a group generated
by elements for which the space of fixed points is a hyperplane (see [42]). But
the set of fixed points of a linear symplectomorphism cannot be a hyperplane. As
a consequence, in the present case, the point p(x) is nonsingular if and only if
Gx = {e}. This proves (1).
The open subset U V is a Ginvariant subset, the action of G on U is free
since the stabilizer of every point of U is by definition trivial and, according to Section 2.3.2, U is equipped with a holomorphic Poisson structure. This implies on the
one hand that the projection map p : U U/G = p(U) is a local biholomorphism
(when both sets are considered as holomorphic manifolds), and on the other hand,
since p is moreover a Poisson map between Poisson manifolds, that p preserves the
rank of the Poisson structure at each point, so that the rank of { , }V /G is 2d at every
point of p(U). Therefore, { , }V /G restricts to a regular holomorphic Poisson structure of maximal rank on p(U). According to Proposition 6.11, it is the canonical
Poisson structure, associated to a symplectic structure on the holomorphic manifold
p(U). This completes the proof of (2).
Remark 6.19. For every C , the map x 1 x leads to a regular map of V /G
to itself. Since x 1 x multiplies the canonical Poisson structure on V by 2 , this
induced automorphism also transforms the quotient Poisson structure { , }V /G into
2 { , }V /G . As a consequence, for every nonzero C , the Poisson varieties
(V /G, { , }V /G ) and (V /G, { , }V /G ) are isomorphic as Poisson varieties.
The case where the dimension of V is 2 will be studied in more detail in Section 9.2.4. We give here an immediate corollary of Proposition 6.18, which will
be useful then.
Corollary 6.20. Let V be a symplectic Cvector space of dimension 2 and let G =
{e} be a finite subgroup of SP(V ). Then the image of the origin o of V through the
canonical projection p : V V /G is the only point of V /G which is singular. It is
also the only point where the rank of the quotient Poisson structure is zero.
Proof. For a vector space V of dimension 2, the group SP(V ) coincides with the
group SL(V ) of all linear maps of determinant +1. Every element of a finite subgroup G of SL(V ) is of finite order, hence is diagonalizable. Clearly, a diagonalizable element of SL(V ), different from the unit, admits o V as its unique fixed
point, hence U = V \ {o}. Clearly, the rank of { , }V /G at p(o) is zero. According
to Proposition 6.18, the rank of { , }V /G is 2 at every other point of V /G.
6.3.5 Example 1: Cotangent Bundles
It is shown in Section 1.3.4 that every Poisson manifold admits a (singular) foliation
whose leaves carry a symplectic structure. Thus, the number of examples of symplectic manifolds is abundant. There are however two large classes of symplectic
176
6 Constant Poisson Structures, Regular and Symplectic Manifolds
manifolds which need special attention: cotangent bundles and Kahler manifolds.
The first class, considered in this section, is of fundamental importance in classical
mechanics, see e.g. [1, 15] and [125].
The idea that cotangent bundles are symplectic manifolds comes from classical
mechanics: local coordinates on M and the corresponding momenta, which are linear coordinates on the fibers of : T M M yield canonical coordinates on phase
space. To make this precise, let us first show how to build natural local coordinates on T M from local coordinates on M. Suppose that x1 , . . . , xd are coordinates
on U M; it yields half of a system of coordinates on T U T M by letting
qi := xi . For the other half, let pi denote the function on T U, which is defined
by
,
(6.21)
pi (m ) := m ,
xi m
where m T U is in the fiber over m, i.e., (m ) = m. The functions pi and xi are
dual in the following sense: for m U and 1 i, j d,
pi (dm x j ) = dm x j ,
= i, j .
xi m
On T M there is a natural differential oneform, called the Liouville form, which
can in terms of the above local coordinates be defined as
d
:= pi dqi .
i=1
More intrinsically, 1 (T M) is defined by
, V (m ) := m , Tm Vm ,
(6.22)
where m T M and V X1 (T M). The canonical pairing in this formula is between Tm M and Tm M: indeed, Tm : Tm (T M) Tm M. To check that , as defined
by (6.22), is given in terms of the above coordinates qi , pi by di=1 pi dqi , take an arbitrary vector field V = di=1 (i / qi + i / pi ) on T M and compare
d
, V (m ) = m , Tm Vm = m , i (m )
,
xi m
i=1
which was computed from (6.22), using qi = xi , with
d
d
d
pi dqi , V (m ) = pi dqi , j q j + j p j (m )
i=1
i=1
j=1
d
p i ( m ) i ( m )
i=1
6.3 Symplectic Manifolds
177
,
i m m xi
m
i=1
d
which was computed from (6.21).
The canonical symplectic structure on T M is by definition the closed differential
twoform := d , which is locally given by di=1 dpi dqi , so that the coordinates qi
and pi are canonical coordinates. Notice that the differential twoform is actually
exact, making T M into an exact symplectic manifold.
To a function F F (M), respectively to a vector field V X1 (M), one can
respectively V , on T M by
associate a function F,
F := F ,
V (m ) := m , V (m)
for every m Tm M. The reader will easily verify that the canonical Poisson bracket
of satisfies the following formulas: if F1 , F2 F (M) and V1 , V2 X1 (M), then
F1 , F2 = 0,
V1 , V2 = [V
1 , V2 ] and
V1 , F1 = V
1 [F1 ] .
In fact, these formulas can be used to define (see [3, Ch. 2]).
6.3.6 Example 2: Kahler Manifolds
We now turn to a second class of examples, which is important from the point of
view of complex geometry. Let M be a complex manifold and let us denote the
underlying real manifold by MR . Suppose that ds2 is a Hermitian metric on M. This
means that, in terms of a system of local (holomorphic) coordinates z1 , . . . , zd ,
ds2 =
hi j dzi dz j ,
1i, jd
where (hi j ) is a positive definite Hermitian matrix. The imaginary part (up to a sign)
of such a metric is a nondegenerate differential 2form on MR , given by
1
:=
hi j dzi dz j .
2 1i,
jd
One says that the metric ds2 is a Kahler metric if is closed, i.e., if is a symplectic
structure. Then (MR , ) is called a Kahler manifold and is called the associated
differential 2form of the metric. Notice that the metric can be recovered from its
associated differential 2form.
Two facts underlie the importance of Kahler manifolds. The first one is that the
complex projective space PN admits a Kahler metric. The second fact is that every
178
6 Constant Poisson Structures, Regular and Symplectic Manifolds
submanifold of a Kahler manifold is itself a Kahler manifold. This means that every
nonsingular complex projective variety has a Kahler structure.
6.4 Notes
The study of symplectic manifolds, initiated by its importance in classical mechanics, has become a subject of its own. Many books have symplectic manifolds as their
central theme, yet there are significant variations in how much the theory is connected to classical, and eventually quantum mechanics, as well as in the advanced
topics which are treated. For an excellent introduction, see Cannas da Silva [33],
where the example of Kahler manifolds is treated in detail. For an introduction to the
subject, largely motivated by classical mechanics, see Arnolds classical book [15].
Another classical book is the mechanics book [1] by Abraham and Marsden, where
symplectic geometry is used as a tool, rather than a subject of itself; in it, the utility
of symplectic structures is plain. The books [125] by LibermannMarle and [92]
by GuilleminSternberg, two books of opposite style, are great complements to the
above references.
All the above books treat additional topics. Group actions and the momentum
map are studied in varying detail in all of them; for this, the books of Audin [18]
and OrtegaRatiu [160] should also be consulted. Symplectic vector bundles and
contact manifolds are treated in LibermannMarle [125], symplectic homogeneous
spaces are an additional topic in GuilleminSternberg [92], the HamiltonJacobi
method and perturbation theory are discussed in both AbrahamMarsden [1] and
Arnold [15].
Chapter 7
Linear Poisson Structures and Lie Algebras
Together with symplectic manifolds, considered in the previous chapter, Lie algebras provide the first examples of Poisson manifolds. Namely, the dual g of a finitedimensional Lie algebra g admits a natural Poisson structure, called its LiePoisson
structure, which provides new insights and technical tools in the study of Lie algebras. For example, the coadjoint orbits in g are precisely the symplectic leaves
of the LiePoisson structure, showing in particular that coadjoint orbits are evendimensional.
LiePoisson structures are linear Poisson structures, in the sense that they are
Poisson structures on a vector space V , for which the Poisson bracket of every pair
of linear functions on V (elements of V ) is a linear function on V . Also, every linear
Poisson structure (on a finitedimensional vector space) is a LiePoisson structure
and we have a functorial equivalence between linear Poisson structures and Lie
algebra structures.
In many cases, one considers the Poisson structure on the Lie algebra g itself,
rather than on g . This is usually done by identifying g with g , using a nondegenerate Adinvariant symmetric bilinear form on g, which exists for example
for every semisimple Lie algebra. The coadjoint orbits are then identified with the
adjoint orbits and the Hamiltonian vector fields on g take a natural form, a socalled
Lax form.
The LiePoisson structure on g is introduced in Section 7.1, while the induced
Poisson structure on g is discussed in Section 7.2. The main properties of the Lie
Poisson structure are given in Section 7.3. A variant of LiePoisson structures,
namely affine (= linear + constant) Poisson structures and their Lie theoretical
interpretation is discussed in Section 7.4. We finish this chapter with a short introduction to the linearization of Poisson structures (in the neighborhood of a point
where the rank is zero).
Unless otherwise stated, F denotes an arbitrary field of characteristic zero.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 7, SpringerVerlag Berlin Heidelberg 2013
179
180
7 Linear Poisson Structures and Lie Algebras
7.1 The LiePoisson Structure on g
Suppose that g is a finitedimensional Lie algebra over F, with Lie bracket [ , ]. We
denote the dual vector space to g by g . We associate to each element1 e of g, a
linear function e : g F, defined by
e : g F
, e := (e) .
In words, e is evaluation at e. Taking a basis (e1 , . . . , ed ) of g, we obtain a
system of linear coordinates (x1 , . . . , xd ) on g by letting xi := ei . We use it to
build the skewsymmetric matrix X = (xi j )1i, jd , whose elements are defined by
xi j := [ei , e j ] ; each entry of X is a function on g, which can be expressed as a linear
combination of the coordinates x1 , . . . , xd . Consider the skewsymmetric biderivation { , } of F[x1 , . . . , xd ], defined by setting, for F, G F[x1 , . . . , xd ],
{F, G} :=
i, j=1
xi j
F G
.
xi x j
(7.1)
It is the unique skewsymmetric biderivation of F[x1 , . . . , xd ] such that ei , ej =
[ei , e j ] , for all 1 i < j d. By bilinearity of { , } and [ , ],
{e , f } = [e, f ] ,
(7.2)
for all e, f g. On the one hand, this implies that Definition (7.1) is independent
of the chosen basis (e1 , . . . , ed ). On the other hand, it yields, for all i, j, k with
1 i, j, k d,
xi , x j , xk = [[ei , e j ], ek ] ,
so that the Jacobi identity for [ , ] implies that xi , x j , xk + (i, j, k) = 0 for all
1 i < j < k d. According to Proposition 1.8, this shows that { , } is a Poisson
structure on F[x1 , . . . , xd ]. Similarly, according to (iii) in Proposition 1.36, Eq. (7.1)
also defines a Poisson structure on the algebra of smooth functions on g , when
F = R, or on the algebra of holomorphic functions on g , when F = C; in either
case, it is the unique Poisson structure which satisfies (7.2). Thus, F (g ) inherits a
Poisson structure from the Lie bracket on g, irrespective of whether we take F (g )
to be the algebra of polynomial, smooth, or holomorphic functions on g .
An intrinsic formula for {F, G} can be written down in terms of the differentials
of F and G. Since the differential of F F (g ) at is a linear map d F : T g F,
and since T g is naturally isomorphic with g , we can think of d F as being an
element of (g ) , i.e., as an element of g, since g and its bidual are canonically
isomorphic (recall that g is finitedimensional; the canonical isomorphism g (g )
1
We use in this section letters e and f to denote elements of a Lie algebra g, to reserve the letters
x, y, z for elements of the bidual (g ) . After identification of g with its bidual, the letters x, y, z
become elements of g, just like everywhere else in the book.
7.1 The LiePoisson Structure on g
181
is given, for e g, by e e ). Thus, we can identify d F with an element of g,
which we will also denote by d F. With this notation, the differential of F F (g )
at g , is given by
d
F
( ) ei ,
(7.3)
d F =
i=1 xi
and (7.1), evaluated at g , can be written in the compact, coordinatefree form,
{F, G} ( ) = , [d F, d G] .
(7.4)
Definition 7.1. Let (g, [ , ]) be a finitedimensional Lie algebra and let F (g ) denote the algebra of polynomial, smooth or holomorphic functions on g . The Poisson
structure on g , defined for F, G F (g ), at g by
{F, G} ( ) := , [d F, d G]
(7.5)
is called the LiePoisson structure on g , or the canonical Poisson structure on g .
As we will show in Proposition 7.3 below, LiePoisson structures are in onetoone correspondence with linear Poisson structures, where the latter are defined as
follows.
Definition 7.2. A Poisson structure { , } on a finitedimensional vector space V is
called a linear Poisson structure if for each pair of linear functions and on V ,
their Poisson bracket { , } is a linear function on V .
Proposition 7.3. For every finitedimensional vector space V , there is a natural onetoone correspondence between linear Poisson structures on V and Lie algebra
structures on V .
Proof. As we have explained, when V = g is a finitedimensional Lie algebra, its
dual vector space g carries a Poisson structure. If x, y are linear functions on g ,
then they are of the form x = e and y = f , where e, f g, and it follows from (7.2)
that their Poisson bracket is a linear function on g . This shows that the LiePoisson
structure on g is a linear Poisson structure. On the other hand, a linear Poisson
structure { , } can be restricted to linear functions on V and the restriction is a
Lie bracket on (V ) V . The LiePoisson structure on V , associated to this Lie
bracket, is the original Poisson bracket { , }, since both Poisson brackets coincide
for all linear functions on V .
Remark 7.4. For i = 1, 2, let (gi , [ , ]i ) be a Lie algebra whose LiePoisson structure
(on gi ) is denoted by { , }i . If : (g1 , [ , ]1 ) (g2 , [ , ]2 ) is a Lie algebra homomorphism, then the transpose map : (g2 , { , }2 ) (g1 , { , }1 ) is a Poisson
map. It is easily checked that this leads to a contravariant functor from the category of finitedimensional Lie algebras to the category of Poisson manifolds (affine
Poisson varieties if F is different from R and C).
It is often convenient to represent the LiePoisson structure by its Poisson matrix
(see Section 1.2.2). To do this, consider, as above, a basis (e1 , . . . , ed ) of g, which
182
7 Linear Poisson Structures and Lie Algebras
leads to a system of linear coordinates (x1 , . . . , xd ) on g . Let us denote by ckij the
structure constants of the Lie algebra g in terms of the above basis:
d
[ei , e j ] =
ckij ek ,
1 i, j d .
k=1
Then, the entries of the Poisson matrix ( xi , x j )1i, jd are given by
xi , x j = [ei , e j ] =
ckij xk .
k=1
In this notation, Eq. (7.1) for the Poisson structure takes the form
d
{F, G} :=
ckij xk
i, j,k=1
F G
.
xi x j
In tensorial terms, the algebra of polynomial functions on g is canonically isomorphic to Sg, the symmetric algebra of g, and the construction of the LiePoisson
structure on Sg can be given without passing via g , which has the advantage that its
construction does not require g to be finitedimensional. Explicitly, a typical monomial z1 z2 . . . z p Sg is viewed as a polynomial function on g by setting, for g ,
(7.6)
(z1 z2 . . . z p )( ) := (z1 ) (z2 ) . . . (z p ) = , z1 , z2 , z p .
For monomials of degree 1, the Poisson bracket { , } on Sg is just the usual Lie
bracket, {z, z } := [z, z ]. For monomials of higher degree, one uses the biderivation
property, which gives
z1 . . . z p , z1 . . . zq =
z1 . . . zk . . . z p z1 . . . z . . . zq
zk , z .
k=1 =1
From this formula, one easily gives a direct proof of the Jacobi identity for the
bracket { , } on Sg.
7.2 The LiePoisson Structure on g
Upon choosing an isomorphism between the finitedimensional Lie algebra g and
its dual g , the LiePoisson structure on g can be transferred to a linear Poisson
structure on g. This is usually done by choosing a nondegenerate symmetric bilinear form  on g. Nondegeneracy of the bilinear form  means that the
linear map : g g , which sends x g to the linear form x  : y x  y, is an
isomorphism. Explicitly, is given by
7.2 The LiePoisson Structure on g
183
(x), y = x  y = (y), x ,
(7.7)
for all x, y g. The inverse map to is denoted by : g g. As in (7.7),
, ( ) = ( )  ( ) = , ( ) ,
(7.8)
for all , g . To a function on g , we can associate a function on g and vice versa,
simply by composing it (on the right) with or with . Similarly, every Poisson
structure on g can be transferred to g and vice versa. In particular, the LiePoisson
structure on g , introduced in Section 7.1, leads to a unique Poisson structure { , }g
on g with respect to which is an isomorphism of Poisson manifolds. We call { , }g
the LiePoisson structure on g with respect to  .
Explicitly, let F, G F (g) and compute their LiePoisson bracket {F, G}g (with
respect to  ) at a point x g by using (7.5) as follows.
{F, G}g (x) = {F , G } ( (x))
= (x), d (x) (F ), d (x) (G )
= x  [ (dx F), (dx G)] .
(7.9)
We have used that d (x) (F ), viewed as an element of g, is precisely (dx F). To
check the latter, first use the chain rule and the fact that is a linear map to find that
d (x) (F ) = dx F ,
so that, for every g ,
d (x) (F ), = dx F, ( ) = , (dx F) ,
where we have used (7.8) in the last step. Equation (7.9) is usually written in the
following form,
{F, G}g (x) = x  [x F, x G] ,
(7.10)
where x F, the gradient of F at x (with respect to  ) is defined, for F F (g)
and x g by
x F  y = dx F, y ,
(7.11)
for every y g, which is equivalent to saying that x F = (dx F). Since F is a
function on a vector space, (7.11) can also be written in the following form,
x F  y =
d
F(x + ty) ,
dt t=0
(7.12)
which is the most useful form for explicit computation.
The LiePoisson structure on g is often restricted to a subspace of g which is
defined by a Lie ideal of g. If A is a subset of g, the orthogonal to A, with respect
to  , is the subspace A g, defined as
184
7 Linear Poisson Structures and Lie Algebras
A := {x g  y A, x  y = 0} .
(7.13)
Proposition 7.5. Let (g, [ , ]) be a finitedimensional Lie algebra, equipped with a
nondegenerate, symmetric bilinear form  , and suppose that h g is a Lie ideal
of g. Then its orthogonal, h , is a Poisson submanifold of g. For F, G F (h ),
their Poisson bracket {F, G}h is given, at x h , by
G (x) ,
{F, G}h (x) = F,
g
(7.14)
where F and G are arbitrary extensions of F and G to a neighborhood of x in g.
Proof. We first show that if h is a Lie ideal of g, then h is a Poisson submanifold
of g. Let x h and let G be an arbitrary function, defined in a neighborhood of x
in g. According to Proposition 2.12, we need to show that (XG )x Tx h h . To
do this, let F be an arbitrary function, defined in a neighborhood of x in g, which is
constant on h . For such a function, we have that x F h; indeed, F(x +ty) = F(x)
for every y h and every t F, so that
x F  y =
d
F(x + ty) = 0 ,
dt t=0
for every y h . Since  is nondegenerate, we may conclude that x F h.
Since h is a Lie ideal of g, we have for every x h that
(XG )x [F] = {F, G}g (x) = x  [x F, x G] = 0 .
This shows that XG is tangent to h at every point x h . Proposition 2.12 also
leads to the formula (7.14) for the Poisson structure { , }h on h .
We now show that the Hamiltonian vector fields XH on (g, { , }) can be written down in a simple form, if  is adinvariant (i.e., when g is a quadratic Lie
algebra). Recall from Section 5.1.4 that this means that x  [y, z] = [x, y]  z for
all x, y, z g. Then we have that (7.10) can be written, for H = G, as
{F, H}g (x) = x F  [x H, x] = dx F, [x H, x] .
Comparing this to
{F, H}g (x) = (XH )x [F] = dx F, (XH )x ,
we find that x := (XH )x , the Hamiltonian vector field XH at x, is given by the Lax
equation
x = [x H, x] .
(7.15)
Lax equations are very important in the theory of integrable systems, as one easily
shows that their flows are isospectral, hence admit many constants of motion (see
Section 12.2.5 below).
7.3 Properties of the LiePoisson Structure
185
Example 7.6. The basic example of a finitedimensional quadratic Lie algebra is
given by the Fvector space g := gld (F) of d d matrices with coefficients in F.
The Lie bracket on g is the commutator [x, y] := xy yx. If Ei j denotes the d d
matrix with a 1 at position (i, j) and zeros elsewhere, then [Ei j , Ek ] = j,k Ei
i, Ek j , for all 1 i, j, k, d. A natural symmetric bilinear form on g is given
by
x  y:= Trace(xy). It is nondegenerate, thanks to the orthogonality relations
Ei j  Ek = j,k i, . Moreover, adinvariance follows at once from the basic property Trace(xy) = Trace(yx) of the trace form. Thus, g is a quadratic Lie algebra. Let
on g which picks the entry at position (i, j) of a
us denote by i j the linear
function
matrix: i j (x) := xi j = E ji  x . For all x, y g,
x i j  y = dx i j , y = i j , y = E ji  y ,
so that x i j = i j is independent of x and is given by i j = E ji . It follows that
the Poisson bracket of two such linear functions on g is given, at x g, by
i j , k g (x) = x  [E ji , Ek ] = x  i, E jk j,k Ei
= (i, k j j,k i )(x) ,
so that i j , k g = i, k j j,k i . Let us show that for every p N , the function
H p : g F, defined by H p (x) := 1p Trace x p , is a Casimir function for { , }g . For all
x, y g,
d
H p (x + ty) = Trace x p1 y = x p1  y .
x H p  y =
dt t=0
It follows that x H p = x p1 , which implies that [x H p , x] = 0, for every x g. In
view of (7.15), this means that the functions Hp are Casimir functions of { , }g .
7.3 Properties of the LiePoisson Structure
In this section we first study the symplectic foliation of a LiePoisson structure and
show how it is related to the coadjoint action. We then explain how the Poisson
cohomology of a LiePoisson structure is related to Lie algebra cohomology. In
Section 7.3.3, we relate the modular class of a LiePoisson structure to the modular
form of a Lie algebra.
7.3.1 The Symplectic Foliation: Coadjoint Orbits
We have seen in Section 1.3.3 that a Poisson manifold naturally decomposes into
symplectic manifolds. We now show that in the case of a LiePoisson structure
186
7 Linear Poisson Structures and Lie Algebras
on g , these symplectic manifolds are the coadjoint orbits of G in g , where G is
any connected Lie group whose Lie algebra is g; we also show that the Casimir
functions are the Ad invariant functions. See Section 5.1.3 for the definition and
the basic properties of the coadjoint action and of Ad invariant functions on g .
Proposition 7.7. Let (g, [ , ]) be a finitedimensional Lie algebra and let G be an
arbitrary connected Lie group whose Lie algebra is g. The symplectic leaves of the
LiePoisson structure on g are the coadjoint orbits of G and the Casimir functions
are the Ad invariant functions.
Proof. We first show that at each point g the vector space of all fundamental
vector fields of the coadjoint action of G on g at coincides with Ham (g , { , }),
the vector space of all Hamiltonian vector fields of the LiePoisson structure on g ,
at . To show it, rewrite (7.5) as
(XG ) [F] = , add G d F = add G , d F
to find that the Hamiltonian vector field XG is given, at g , by
(XG ) = add G .
For every x g, the linear function x , defined by x ( ) := , x for all g ,
satisfies d x = x , for every g . It follows that
Ham (g ) = add G  G F (g ) = {adx  x g} ,
as we needed to show. Since the coadjoint orbits of G are connected, this shows that
the leaves of the symplectic foliation of the LiePoisson structure on g are precisely
the coadjoint orbits. It follows that the Ad invariant functions are precisely the
functions which are constant on every leaf of the symplectic foliation. According to
Proposition 1.31, they are the Casimir functions of the LiePoisson structure on g .
Suppose now, as in Section 7.2, that we have a nondegenerate symmetric bilinear form  on g, and let : g g be the resulting isomorphism (see (7.7)).
Since realizes an isomorphism between the LiePoisson structure on g and the
LiePoisson structure on g with respect to  , Proposition 7.7 leads to the following result.
Proposition 7.8. Let (g, [ , ]) be a finitedimensional quadratic Lie algebra, whose
bilinear form is denoted by  . Let G be an arbitrary connected Lie group whose
Lie algebra is g. The symplectic leaves of the LiePoisson structure on g with respect
to  are the adjoint orbits of G and the Casimir functions are the Adinvariant
functions.
The fact that the symplectic foliation of the dual of a Lie algebra is given by the
orbits of the coadjoint action shows that, even in the case of a linear Poisson struc
7.3 Properties of the LiePoisson Structure
187
Fig. 7.1 For r 0 the sphere x2 + y2 + z2 = r2 is a symplectic leaf of the LiePoisson structure
on (so3 (R)) .
ture, the symplectic foliation can be rather complicated. This is illustrated in the
following examples of (real) threedimensional Lie algebras.
Example 7.9. We start with the LiePoisson structure of (so3 (R)) , which is defined
by the following brackets
[x, y] = z ,
[y, z] = x ,
[z, x] = y .
The Poisson matrix is given by
z y
z 0 x ,
y x 0
and C := x2 + y2 + z2 is a Casimir function. The rank of the Poisson structure is
two at every point, different from the origin o, and at such a point the differential of
the Casimir function C is nonzero. According to Proposition 1.32, the symplectic
leaves in R3 \ {o} are the connected components of the fibers of C, i.e., the spheres
x2 + y2 + z2 = r2 , with r = 0. Of course, {o} is itself a symplectic leaf, so the symplectic leaves are the concentric spheres x2 + y2 + z2 = r2 , with r R0 , where the
origin appears as a sphere of radius r = 0. See Fig. 7.1.
Example 7.10. We next consider the LiePoisson structure of (sl2 (R)) , where the
standard sl2 (R) brackets
[x, y] = z ,
[y, z] = 2y ,
[z, x] = 2x ,
lead to the following Poisson matrix
0 z 2x
z 0 2y .
2x 2y 0
188
7 Linear Poisson Structures and Lie Algebras
Fig. 7.2 Every point of the plane x = 0 is a symplectic leaf for the LiePoisson structure on the
dual of the Heisenberg algebra. The planes x = c, with c = 0, are the other symplectic leaves.
Clearly, C := 4xy + z2 is a Casimir function and the rank of the Poisson structure
is two at every point, except at the origin; also, the differential of C is nonzero,
except at the origin. As in the previous example, Proposition 1.32 implies that the
symplectic leaves are the connected components of the fibers of C, namely,
(1)
(2)
(3)
(4)
The onesheeted hyperboloids 4xy + z2 = c, for c > 0;
The two components of the twosheeted hyperboloids 4xy + z2 = c, for c < 0;
The two components of the cone z2 = 4xy minus the origin;
The origin.
Notice that the previous example and the present example are geometrically
very different, yet the underlying Lie algebras have the same complexification,
namely sl2 (C).
Example 7.11. Our third example is the Heisenberg algebra, with Lie brackets
[x, y] = 0 ,
[y, z] = x ,
The Poisson matrix is now given by
[z, x] = 0 .
0 0 0
0 0 x .
0 x 0
It is immediate that the Casimir functions are precisely the functions which are
independent of y and of z. The Poisson structure has now rank zero on the plane
x = 0, so that each point of this plane is a symplectic leaf; together, these points
form the zero locus of the Casimir function x. The other symplectic leaves are twodimensional, they are the planes x = c, with c R . See Fig. 7.2.
Example 7.12. In the fourth example we show that, for a linear Poisson structure
on R3 , nontrivial smooth Casimir functions may even not exist locally (at some
points). Take on R3 , with respect to the system of coordinates (x, y, z), the following
Poisson matrix,
0
0 x
0
(7.16)
0 y
.
x y 0
7.3 Properties of the LiePoisson Structure
189
Fig. 7.3 For the open book foliation, which is the symplectic foliation of the LiePoisson structure,
given by the Poisson matrix (7.16) the leaves are the points of the Zaxis and the halfplanes,
obtained by removing the Zaxis from the planes passing through it.
On R3 , minus the plane x = 0, y/x is a Casimir function, while on R3 , minus the
plane y = 0, x/y is a Casimir function. The symplectic leaves of dimension 0 are
the points on the Zaxis, while the symplectic leaves of dimension 2 are the halfplanes, obtained by removing the Zaxis from all the planes which pass through the
Zaxis, as follows again from Proposition 1.32. This symplectic foliation is called
the open book foliation. Taking an arbitrary point on the Zaxis, it is clear that on no
neighborhood of it is there a nonconstant Casimir function. See Fig. 7.3.
Example 7.13. We consider, in the last example, for a fixed C, the basic Lie
brackets
[x, y] = 0 , [y, z] = y , [z, x] = x ,
which lead to the following Poisson matrix for the corresponding LiePoisson structure on C3 ,
0 0 x
0 0 y .
x y
A holomorphic function F, defined on an open subset U C3 , is a Casimir function
for this Poisson structure if and only if
F
F
F
= x
y
=0,
z
x
y
on U. A solution of these equations is easily found, giving F = cxy + d, where
/ Q, then the level sets of F are not even
c and d are integration constants. If
algebraic varieties.
190
7 Linear Poisson Structures and Lie Algebras
7.3.2 The Cohomology of LiePoisson Structures
We show in this section that the Poisson cohomology of a LiePoisson structure
is naturally related to Lie algebra cohomology, leading under some conditions to
a fairly explicit description of it. Let g be a finitedimensional Lie algebra, whose
LiePoisson structure (on g ) is denoted by or by { , }. The algebra of functions F (g ) which we consider can be taken indifferently as the algebra of polynomial, holomorphic or smooth functions.
For x g, we will denote as in Section 7.1, by x the element x, viewed as an
element of the bidual of g, i.e., x is the linear function on g , defined by x , :=
, x for all g . Consider the map g X1 (g ), which associates to x g the
Hamiltonian vector field of x with respect to the LiePoisson structure. It is a Lie
algebra homomorphism, since
X[x,y] = X{x ,y } = [Xx , Xy ] ,
for all x, y g. Equivalently, the assignment g F (g ) F (g ) given by (x, F)
Xx [F] = {x , F} defines a representation of g on F (g ). We can therefore consider HL (g, F (g )), the F (g )valued Lie algebra cohomology of g. Recall from
Section 4.1.1 that for every q N, the space of F (g )valued qcochains of g is
given by
Cq (g; F (g )) = Hom(q g, F (g )) ,
while the differential is given, for all c Cq (g; F (g )) and all x0 , . . . , xq g, by
Lq (c)(x0 , . . . , xq ) =
(1)i
i=0
xi , c(x0 , . . . , xi , . . . , xq )
(7.17)
(1)i+ j c [xi , x j ] , x0 , . . . , xi , . . . , xj , . . . , xq .
0i< jq
We show in the following proposition that, for every q N, the cohomology space
HLq (g, F (g )) is canonically isomorphic to the qth Poisson cohomology space
of (g , ).
Proposition 7.14. Let g be a finitedimensional Lie algebra. The dual g of g is
equipped with its algebra F (g ) of polynomial, holomorphic or smooth functions.
The LiePoisson bracket on F (g ) is denoted by . For every q N there is a
canonical isomorphism
HLq (g, F (g )) Hq (g ) .
Proof. We show that the F (g )valued Lie algebra cohomology complex of g is
isomorphic to the Poisson cohomology complex of (g , ). Let Q be a Poisson qcochain of (F (g ), ), i.e., Q Xq (g ). We associate to Q the element q (Q)
Cq (g; F (g )) = Hom(q g, F (g )), defined by
q (Q) :
q g
F (g )
x1 xq Q[x1 , . . . , xq ] .
(7.18)
7.3 Properties of the LiePoisson Structure
191
It is easy to see that for every q N the resulting linear map q : Xq (g )
Hom(q g, F (g )) is anisomorphism. Moreover, a direct comparison of (4.4)
and (7.17), using xi , xj = [xi , x j ] , shows that the following diagram is commutative.
q
Xq (g )
Xq+1 (g )
q+1
Hom(q g, F (g ))
q
L
Hom(q+1 g, F (g ))
The isomorphisms (q )qN therefore induce isomorphisms in cohomology.
For a representation ( ,V ) of g, recall that V g stands for the space of all ginvariant elements, i.e., elements v V such that (x, v) = 0 for every x g. Recall
also that we denote by HL (g) the trivial Lie algebra cohomology (see Example 4.2).
Since V g is a subrepresentation of V , on which g acts trivially, we have linear maps
HLq (g) V g HLq (g,V g ) HLq (g,V ) ,
(7.19)
where the first map is an isomorphism, but the second map is in general neither
injective nor surjective. In the case of the above representation of g on F (g ), it
leads for every q N to the linear map
q : HLq (g) Cas(g ) Hq (g ) ,
(7.20)
since the invariant elements of F (g ) are the functions F F (g ) such that
{x , F} = 0 for every x g, i.e., the Casimir functions of (g , ). Recall from Proposition 7.7 that these functions are the Ad invariant functions. For a semisimple Lie
algebra, a nontrivial property on its Lie algebra cohomology (with coefficients in a
finitedimensional representation) implies that q is an isomorphism, for every q N,
as we show in the following proposition.
Proposition 7.15. Let g be a semisimple Lie algebra. We denote by F (g ) the algebra of polynomial functions on g and by Cas(g ) F (g ) the subalgebra of
polynomial Casimir functions. For every q N, the map q , given by (7.20), is an
isomorphism:
q : HLq (g) Cas(g ) Hq (g ) .
Proof. We denote by Fk (g ) the vector space of homogeneous polynomials of degree k on g , where k N. It is a subrepresentation of F (g ), and
F (g ) =
Fk (g ) ,
kN
which implies that the F (g )valued Lie algebra cohomology of g decomposes as
a direct sum,
q
HL (g, Fk (g )) ,
HLq (g, F (g )) =
kN
192
7 Linear Poisson Structures and Lie Algebras
for every q N. Since the LiePoisson structure on g is linear, a function F
F (g ) is a Casimir function if and only if for every k N its homogeneous component of degree k is a Casimir function. Setting Cask (g ) := Cas(g ) Fk (g ), it
follows that
q
HLq (g) Cas(g ) =
HL (g) Cask (g ) ,
kN
for every q N. Moreover, for every q, k N, the linear map q restricts to yield a
map
q,k : HLq (g) Cask (g ) HLq (g, Fk (g )) .
Now, according to [95, Th. 11], for every finitedimensional representation V of a
semisimple Lie algebra g, the natural linear map HLq (g) V g HLq (g,V ) described
in (7.19) is bijective for every q N. In particular, for every k N, q,k is an isomorphism, which, in turn, implies that q is an isomorphism, as was to be shown.
7.3.3 The Modular Class of a LiePoisson Structure
The modular form of a finitedimensional Lie algebra g is the linear form g on g,
defined for x g by
g (x) := Trace(adx ) .
A Lie algebra is said to be unimodular when its modular form vanishes. For x, y g,
g ([x, y]) = Trace(ad[x,y] ) = Trace([adx , ady ]) = 0 ,
(7.21)
so that g vanishes on [g, g]. Every Lie algebra which satisfies [g, g] = g is therefore
unimodular; in particular, if g is semisimple, then g is unimodular. If g is nilpotent,
then g is unimodular, since adx is nilpotent, hence traceless, for every x g. For
a different reason, quadratic Lie algebras are unimodular as well. However, solvable Lie algebras are not unimodular in general, for instance, the twodimensional
Lie algebra defined, on a basis (e, f ) of g by [e, f ] := f , is not unimodular, since
Trace(ade ) = 1.
In the following proposition, we relate the modular form g of g with the modular
vector field of the Poisson manifold (g , ), where stands for the canonical Lie
Poisson structure on g . The volume form which we choose on g is any translation
invariant volume form (see Remark 4.13). is a constant vector field on g , hence
corresponds in a canonical way to an element of g .
Proposition 7.16. Let g be a finitedimensional Lie algebra. Interpreted as an element of g , the modular vector field of (g , ) with respect to any translation invariant volume form is, up to a sign the modular form of g.
Proof. Let (e1 , . . . , ed ) be a basis of g and let (1 , . . . , d ) be the dual basis of g .
We use the functions x1 = e1 , . . . , xd = ed as linear coordinates on g . According
7.4 Affine Poisson Structures
193
to (4.22), the divergence of the LiePoisson structure on g is given by
Div( ) =
xi , x j
= [ei , e j ] , j
.
xi 1i, jd
xi
1i, jd x j
Since = Div( ) (see Proposition 4.17), it follows that
(x) = [ei , e j ] , j xi = j , [x, e j ] = Trace(adx ) ,
1i, jd
1 jd
for every x = di=1 xi ei g. This shows that = g .
7.4 Affine Poisson Structures
We now turn to a combination of constant and linear Poisson structures, which we
call affine Poisson structures.
Definition 7.17. A Poisson structure { , } on a finitedimensional vector space V is
called an affine Poisson structure if, for every pair of affine functions F and G on V ,
their Poisson bracket {F, G} is an affine function on V .
Equivalently, for every pair of linear functions F and G on V , their Poisson bracket
{F, G} is an affine function on V .
Proposition 7.18. For every finitedimensional vector space V , there is a onetoone correspondence between affine Poisson structures on V and pairs of compatible
Poisson structures (0 , 1 ) on V , where 0 is a constant Poisson structure and 1 is
a linear Poisson structure.
Proof. Suppose that is a skewsymmetric biderivation of F (V ) such that for every
pair of linear functions F and G on V , one has that [F, G] is an affine function on V .
For such functions, let 0 [F, G] and 1 [F, G] denote the constant, respectively linear
parts of [F, G]. We view 0 and 1 as bilinear maps on V and we extend them to
biderivations of F (V ), still denoted by 0 and 1 , so that = 0 + 1 . According to
Chapter 6, 0 is a constant Poisson structure. In terms of the Schouten bracket [ , ]S ,
one has that is a Poisson structure if and only if [ , ]S = 0, which is equivalent
to the vanishing of [ , ]S [F, G, H] for all linear functions F, G and H on V . Since
[0 , 0 ]S = 0, because 0 is a Poisson structure, we have that
[ , ]S [F, G, H] = [1 , 1 ]S [F, G, H] + 2 [0 , 1 ]S [F, G, H] .
In view of (3.37), the first term in this expression is a linear function on V , while the
second term is a constant function on V . Thus, [ , ]S = 0 if and only if [1 , 1 ]S = 0
(i.e., 1 is a linear Poisson structure) and [0 , 1 ]S = 0 (i.e., 0 and 1 are compatible). It follows that the map, which sends an affine Poisson structure to its constant
and linear parts (0 , 1 ), yields the announced onetoone correspondence.
194
7 Linear Poisson Structures and Lie Algebras
In view of Proposition 7.18, it is natural to look for a Lie algebraic interpretation of
affine Poisson structures. Indeed, if is an affine Poisson structure on V , then V
is the dual of a Lie algebra V and 0 , restricted to V , is a skewsymmetric bilinear
form on V .
Proposition 7.19. For every finitedimensional vector space V , there is a onetoone
correspondence between affine Poisson structures on V and pairs ([ , ], c) where
[ , ] is a Lie algebra structure on V and c is a 2cocycle in the trivial Lie algebra
cohomology of (V, [ , ]).
Proof. According to Proposition 7.18, we need to show that there is a onetoone correspondence between pairs (0 , 1 ) of compatible Poisson structures on V ,
where 0 is constant and 1 is linear, and pairs ([ , ], c), where [ , ] is a Lie algebra
structure on V and c is a 2cocycle in the trivial Lie algebra cohomology of (V, [ , ]).
Let (0 , 1 ) be a pair of Poisson structures on V , where 0 is constant and 1 is linear. Let [ , ] be the Lie algebra structure on V , defined2 by [v, w] := 1 [v , w ], for
all v, w V , and let c(v, w) := 0 [v , w ], which defines a linear map c : V V F.
We show that 0 and 1 are compatible if and only if c is a 2cocycle in the trivial
Lie algebra cohomology of g := V ; since the Poisson structures 0 and 1 can be
reconstructed from the pair ([ , ], c), this will prove the proposition. We have for
all u, v, w V = g,
[0 , 1 ]S (u , v , w ) = 0 [1 [u , v ], w ]+ (u, v, w)
= c([u, v] , w)+ (u, v, w)
= L2 (c)(u, v, w) ,
see Example 4.2, in particular formula (4.2) for the Lie algebra coboundary operator L , in the case of a trivial representation. It follows that [0 , 1 ]S = 0 if and only
if c is a Lie 2cocycle, L2 (c) = 0.
In view of the above proposition, affine Poisson structures are also called modified
canonical Poisson structures or modified LiePoisson structures.
If g is semisimple, then HL2 (g) is trivial (see Lemma 4.1), so that, if c is a Lie 2cocycle, then c is a Lie 2coboundary, c = L1 (c ), written out, c(xi , x j ) = c ([x j , xi ])
and we see that the affine Poisson structure, corresponding to ([ , ], c), is nothing but
the LiePoisson structure associated to [ , ], with xi replaced by xi +c (xi ), i.e., both
Poisson structures are the same up to an affine change of variables (a translation).
Affine Poisson structures are, in general, also closely related to LiePoisson
structures in a different way. Namely, given a Lie algebra (g, [ , ]) and a Lie 2cocycle c, as in Proposition 7.19, we consider on g F the Lie bracket
[(x, a), (y, b)]c := ([x, y] , c(x, y)) .
(7.22)
For every (xi , ai ) g F, where i = 1, 2, 3, one has that
As in the case of Lie algebras, if v V , then v denotes its image under the canonical isomorphism V (V ) .
2
7.4 Affine Poisson Structures
195
[[(x1 , a1 ), (x2 , a2 )]c , (x3 , a3 )]c = [([x1 , x2 ] , c(x1 , x2 )), (x3 , a3 )]c
= ([[x1 , x2 ] , x3 ] , c([x1 , x2 ] , x3 )) ,
so that the Jacobi identity for [ , ] and the cocycle condition for c imply the Jacobi
identity for [ , ]c , i.e., (g F, [ , ]c ) is a Lie algebra. In the following proposition we
show that the LiePoisson structure { , }c on (g F) , restricted to a hyperplane,
is isomorphic to the affine Poisson structure { , } on g, defined by the 2cocycle c.
Proposition 7.20. Let (g, [ , ]) be a finitedimensional Lie algebra and let c be a 2cocycle in the trivial Lie algebra cohomology of g. Denote the LiePoisson structure
on (g F) , associated to (g F, [ , ]c ), by { , }c , and denote by { , } the affine
Poisson structure on g , with constant term c.
(1)
(2)
(3)
The linear function 1 : (g F) F, defined by (0, 1), is a Casimir
function of { , }c ;
The hyperplane N := { (g F)  (0, 1) = 1} is a Poisson submanifold
of (g F) ;
The isomorphism : N g , defined for N by (, 0), is a Poisson
isomorphism between (N, { , }c ) and (g , { , }).
Proof. We first show that 1 is a Casimir function of { , }c . Notice that 1 = (0, 1),
as an element of g F. For ( , a) g F, it follows that d( ,a) 1 = 1 = (0, 1),
which belongs to the center of [ , ]c (see (7.22)). It follows that for every function
F on (g F) ,
{F, 1 }c ( , a) = ( , a), [d( ,a) F, (0, 1)]c = 0 .
This shows (1). Since the hyperplane N, as defined in (2), is a level set of the
Casimir function 1 , it is a Poisson submanifold of (g F) , which yields (2).
It is clear that is an isomorphism between the affine space N and the vector space g . In order to prove that is a Poisson morphism, it is sufficient to
prove that {x , y }c = {x , y } , for every x, y g, where, as above,
stars stand for the corresponding linear functions on g . By a slight abuse of notation, we will also consider x as a linear function on (g F) , where we put
x ( ) := (x, 0), for all (g F) . With this abuse of notation, x = x ,
and {x , y } = [x, y] + c(x, y). For {x , y }c , which is now written
as {x , y }c , let N and compute
{x , y }c ( ) = , d x , d y c = , [(x, 0), (y, 0)]c
= , ([x, y], c(x, y)) = [x, y] ( ) + c(x, y) ,
which is {x , y } , evaluated at . This shows that is a Poisson morphism;
since is an isomorphism, it is a Poisson isomorphism, which is the content of (3).
As we have shown earlier in this section, every affine Poisson structure which is
constructed by using a Lie 2cocycle which is a coboundary, is isomorphic to the
196
7 Linear Poisson Structures and Lie Algebras
original LiePoisson structure, by an affine change of variables, which implies that
both Poisson manifolds have the same symplectic foliation, up to a translation. We
show in the following example that when the Lie 2cocycle is not a coboundary,
these symplectic foliations may be quite different.
Example 7.21. Consider on R3 the linear and affine Poisson structures 1 and ,
whose Poisson matrices are given by
0 0 x
0
1 x
0 0
and 1 0
.
y
y
x y 0
x y 0
The underlying Lie algebra of both Poisson structures is defined by the brackets
[x, y] = 0, and [y, z] = y and [z, x] = x. The corresponding Lie 2cocycle is given by
c(x, y) = 1 and c(x, z) = c(y, z) = 0. Notice that is a regular Poisson structure,
of rank two, while the rank of 1 vanishes at the origin. Clearly, xy is a Casimir
function of 1 , while xy + z is a Casimir function of . The symplectic leaves of 1
are the points on the Zaxis, the two halfplanes obtained by removing the Zaxis
from the plane x = 0, the two halfplanes obtained by removing the Zaxis from
the plane y = 0, and the connected components of the hyperbolic cylinders xy = c,
with c R . For the affine Poisson structure , the symplectic leaves are the parallel
hyperboloids xy + z = c, with c R.
7.5 The Linearization of Poisson Structures
According to Weinsteins splitting theorem (Theorem 1.25), every Poisson manifold
(M, ) is, in the neighborhood of each of its points, isomorphic to the product of a
symplectic manifold and a Poisson manifold whose Poisson structure vanishes at
a point. According to Darbouxs theorem (Theorem 1.26), the symplectic Poisson
structure can be written in a canonical form in wellchosen coordinates. For Poisson
structures which vanish at a point m, there is no canonical form; we will explain
in this section that, under some conditions, such a Poisson structure is isomorphic
in a neighborhood of m to a linear Poisson structure, which is canonically associated to the Poisson structure at m. This linear Poisson structure is introduced in the
following proposition.
Proposition 7.22. Let (M, ) be a real or complex Poisson manifold, and suppose
that m M is a point at which vanishes. There exists a unique linear Poisson
structure { , }1 = 1 on Tm M, such that, for every neighborhood U of m in M and
for all F, G F (U),
{dm F, dm G}1 = dm {F, G} ,
(7.23)
where the elements dm F, dm G and dm {F, G} of Tm M are viewed as (linear) functions
on Tm M.
7.5 The Linearization of Poisson Structures
197
Suppose that : (M, ) (M , ) is a Poisson map, where (M , ) is also a
Poisson manifold. Then vanishes at (m) and dm : (Tm M, 1 ) (T (m) M , 1 )
is a Poisson map, where the linear Poisson structure 1 on T (m) M is defined as
in (7.23).
We call the linear Poisson structure 1 on Tm M the linearized Poisson structure of
at m.
Proof. Every linear function on Tm M is an element of Tm M, and is therefore of the
form dm F for some function F, defined in a neighborhood of m in M. Since a Poisson
structure on a vector space (here Tm M) is uniquely determined by its value on linear
functions, it follows that there exists at most one Poisson structure 1 on Tm M,
satisfying (7.23). In order to establish the existence of this Poisson structure, we
use (7.23); namely for , Tm M, which we view as linear functions on Tm M, we
define { , }1 := dm {F, G}, where F and G are arbitrary local functions, whose
differentials at m are and respectively. In order to show that dm {F, G} does not
depend on the choice of F and G, we use a system of local coordinates (x1 , . . . , xd ),
centered at m: writing
d
F G
{F, G} := xi j
,
(7.24)
xi x j
i, j=1
where xi j := xi , x j for all 1 i, j d, the assumption that vanishes at m amounts
to xi j (m) = 0 for all i, j, so that the differential of both sides of (7.24) at m is given
by
d
F
G
(m)
(m) dm xi j .
dm {F, G} =
xj
i, j=1 xi
This gives the required independence of the chosen functions F and G, since
F
= ,
,
(m) = dm F,
xi
xi m
xi m
and similarly for the partial derivatives of G at m. The Jacobi identity for the Poisson
structure implies the Jacobi identity for 1 since it follows at once from (7.23) that
{ , { , }1 }1 = dm {F, {G, H}} ,
where F, G and H are arbitrary functions, defined in a neighborhood of m, whose
differentials at m are , and , respectively. This shows that there exists a (unique)
Poisson structure 1 on Tm M, satisfying (7.23) for all open subsets U of M which
contain m and for all F, G F (U).
Suppose now that (M, ) and (M , ) are two Poisson manifolds and that :
M M is a Poisson map. If m = 0, then (m) = 0 and we can consider the
linearized Poisson structures at Tm M and at T (m) M ; we denote them by { , }1
and { , }1 respectively. We show that dm : (Tm M, { , }1 ) (T (m) M , { , }1 ) is
a Poisson map. To do this, it suffices to show that for all linear functions L1 , L2
198
7 Linear Poisson Structures and Lie Algebras
on T (m) M ,
{L1 , L2 }1 dm = {L1 dm , L2 dm }1 .
(7.25)
Given such functions L1 and L2 , there exist functions F and G, defined on a neighborhood of (m) in M , such that d (m) F = L1 and d (m) G = L2 . Since is a
Poisson map, {F, G} = {F , G } , which we differentiate at m, giving
d (m) {F, G} dm = dm {F , G } .
Written in terms of the linearized Poisson structures, this becomes
d (m) F, d (m) G
1
dm = {dm (F ), dm (G )}1
= d (m) F dm , d (m) G dm 1 ,
which is (7.25), since L1 = d (m) F and L2 = d (m) G.
Definition 7.23. Let (M, ) be a Poisson manifold, suppose that m is a point of M
at which vanishes and denote the linearized Poisson structure of at m by 1 . We
say that is linearizable at m, if there exists a Poisson diffeomorphism , between
a neighborhood of m in (M, ) and of the origin o in (Tm M, 1 ), with (m) = o.
Example 7.24. Let m be a point of a Poisson manifold (M, ). Suppose that vanishes at m and that the linearized Poisson structure of at m is the trivial Poisson
structure. Then is linearizable at m if and only if is zero in a neighborhood of m.
Example 7.25. If is a linear Poisson structure on a finitedimensional vector
space V (over R or C), then is linearizable at the origin o. In fact, the canonical isomorphism V ToV is a Poisson isomorphism, where V is equipped with
and where ToV is equipped with the linearized Poisson structure of at o.
Example 7.26. Let = { , } be a nontrivial Poisson structure on a finitedimensional vector space V , such that for every pair of linear functions on V , their Poisson
bracket is either zero or it is a homogeneous function of degree k 2 (in the language of Section 8.1, is a homogeneous Poisson structure of degree k). Then
vanishes at the origin o V and the linearized Poisson structure { , }1 is trivial,
since for all linear functions F, G on V ,
{do F, do G}1 = do {F, G} = 0 ,
because {F, G} is either zero or is homogeneous of degree k 2. Since is nontrivial and is homogeneous, there is no neighborhood of o on which is the trivial
Poisson bracket. It follows from Example 7.24 that is not linearizable at o.
Example 7.27. Let F be a smooth or holomorphic function, defined on a neighborhood U of the origin o of F2 . We assume that F vanishes at o, but that the restriction
of F to every neighborhood of o is different from zero. Consider the Poisson structure on U, defined in terms of the standard coordinates x, y of F2 by {x, y} = F. We
claim that is linearizable at o if and only if do F = 0.
7.5 The Linearization of Poisson Structures
199
Let us prove the claim. First, if do F = 0, then {do x, do y}1 = do F = 0, so that the
linearized Poisson structure is trivial; in view of Example 7.24, is in this case not
linearizable at o. Assume now that do F = 0 and consider the vector field
V :=
F
F
,
x y y x
which does not vanish at o. By the straightening theorem (Theorem B.7), there exists
a function G, defined in a neighborhood of o, such that V [G] = 1. Since
F G F G
= V [G] = 1 ,
x y
y x
we have on the one hand that do F and do G are linearly independent, and on the other
hand that {F, G} = F. Thus, (F, G) is a system of coordinates on a neighborhood
of o and is linear in terms of these coordinates. According to Example 7.25, is
linearizable at o.
Remark 7.28. For i = 1, 2, let (Mi , i ) be a Poisson manifold, whose Poisson structure vanishes at mi Mi . Let : M1 M2 be an isomorphism of Poisson manifolds
with (m1 ) = m2 . Then the tangent map Tm : Tm1 M1 Tm2 M2 is an isomorphism
of Poisson manifolds, where both tangent spaces are equipped with the linearized
Poisson structures of 1 and 2 .
In particular, if two finitedimensional vector spaces V1 and V2 , equipped with a
linear Poisson structure, are isomorphic as Poisson manifolds via a map which
sends the origin o of V1 to the origin of V2 , then the linear map do : V1 ToV1
ToV2 V2 is a Poisson isomorphism. In other words, if there exists (in the neighborhood of the origin) a Poisson isomorphism between two linear Poisson structures,
then a linear Poisson isomorphism between them also exists.
The condition that a Poisson structure which vanishes at m M is linearizable
at m can be expressed as the existence of a neighborhood U of m in M and of a
diffeomorphism between U and a neighborhood of the origin o in Tm M, with
(m) = o, such that
{ , } = { , }1 ,
(U) the ideal of F (U), consisting
for all , Tm M. Denoting for N by Im
of all functions which vanish, together with all their partial derivatives up to order ,
at the point m, leads to the following weaker notion of linearizability.
(+1)
Definition 7.29. Let (M, ) be a Poisson manifold, and suppose that m M is a
point at which vanishes. Then is said to be linearizable at m up to order N
if there exists a neighborhood U of m in M and a diffeomorphism between U and
a neighborhood of the origin o in Tm M, with (m) = o, such that
(+1)
{ , } { , }1 Im
for all , Tm M.
(U) ,
200
7 Linear Poisson Structures and Lie Algebras
Example 7.30. It is clear from the definition that every Poisson structure which is
linearizable at a point is at that point linearizable up to order , for all N .
However, the converse is not true, in general. To show this, we define a (smooth)
2
Poisson structure on R2 by {x, y} := f (x), where f (x) := ex for x = 0 and
f (0) := 0. The fact that f vanishes at 0, together with all its derivatives at 0, implies
(+1)
(R2 ), for all N, where o denotes the origin of R2 . It follows
that {x, y} Io
on the one hand that the linearized Poisson structure at o is trivial, and on the other
hand that is linearizable up to order , for all N . However, the Poisson structure is not linearizable at o, since there is no neighborhood of o on which is
zero.
In order to clarify and compare the different notions of linearizability at a point m
in a Poisson manifold (M, ), we consider a coordinate
chart
(U, x) centered at m,
we define, for 1 i, j, k d, the constants ckij := x xi , x j (m) and we consider
k
the following four statements.
d
(2)
xi , x j ckij xk Im (U) ,
(7.26)
k=1
xi , x j
yi , y j =
ckij xk ,
(7.27)
k=1
d
ckij yk ,
(7.28)
k=1
d
(+1)
yi , y j ckij yk Im
(V ) .
(7.29)
k=1
The inclusion (7.26) holds for all 1 i, j d if and only if vanishes at m. In this
case, the linearized Poisson structure 1 at m is given by (7.27), where x1 , . . . , xd are
the linear coordinates on Tm M, defined by xi := dm xi , for i = 1, . . . , d. Still assuming
that vanishes at m, there exists a coordinate chart (V, y) for M, centered at m M,
such that (7.28) (respectively (7.29)) holds for all 1 i, j d, if and only if is
linearizable at m (respectively is linearizable at m up to order ).
According to Proposition 7.3, there is a onetoone correspondence between linear Poisson structures on Tm M and Lie algebra brackets on Tm M. It follows that there
corresponds to the linearized Poisson structure at m (where m is a point at which
vanishes) a Lie algebra structure [ , ] on Tm M. It satisfies, by construction,
[dm F, dm G] = {dm F, dm G}1 = dm {F, G} ,
for all functions F and G, defined on an arbitrary neighborhood of m in M. The
structure of this Lie algebra is a key element in Conns linearization results, which
we formulate in the following theorem.
7.5 The Linearization of Poisson Structures
201
Theorem 7.31 (Conns linearization theorem). Let (M, ) be a Poisson manifold,
and let m M be a point at which vanishes. Let [ , ] denote the Lie bracket
on Tm M, associated to the linearized Poisson structure at m.
(1)
(2)
(3)
In the holomorphic case, if (Tm M, [ , ]) is a semisimple complex Lie algebra,
then is linearizable at m;
In the real case, if (Tm M, [ , ]) is a compact semisimple Lie algebra, then
is linearizable at m;
In both cases, if (Tm M, [ , ]) is a semisimple Lie algebra, then is linearizable at m up to order , for all N .
Proof. We do not prove Conns results (1) and (2), as the proof is long and rather
technical (see [45, 46]); instead, we prove the analogous (but easier to prove) result (3). We do this by recursion on . Every Poisson structure which vanishes at m
is linearizable up to order 1, so that the statement holds true for = 1. Assume
now that is linearizable up to order . According to Definition 7.29, there exists a
neighborhood U of m in M and a diffeomorphism between U and a neighborhood
of the origin o in Tm M, with (m) = o, such that
(+1)
{x , y } [x, y] Im
(U) ,
(7.30)
for all x, y Tm M; we write here, and in the rest of the proof, the elements of
Tm M with latin letters x, y, z, because we view Tm M = g as a Lie algebra. For
x g, we write Fx as a shorthand for the element x of F (U) and we denote
(+1)
(+2)
(U) and I := Im
(U). Notice that I /I is finitedimensional,
I := Im
because I /I is isomorphic to the space of homogeneous polynomials of degree
+ 1 in dim M variables. With these notations, (7.30) can be written as
Fx , Fy F[x,y] I .
(7.31)
Since vanishes at m, we have that {I , F (U)} I and {I , F (U)} I . It
follows that we can define a map
: g (I /I ) I /I
(x, p(F)) x p(F) := p({Fx , F}) ,
(7.32)
where p : I I /I denotes the canonical projection. We claim that is a (finitedimensional) representation of g. To prove this, we need to show that for all x, y g
and for every F I ,
[x, y] p(F) = x (y p(F)) y (x p(F)) .
On the one hand, (7.31) and the inclusion {I , I } I imply that
[x, y] p(F) = p( F[x,y] , F ) = p( Fx , Fy , F ) ,
while, on the other hand,
(7.33)
202
7 Linear Poisson Structures and Lie Algebras
x (y p(F)) y (x p(F)) = p( Fx , Fy , F ) p( Fy , {Fx , F} ) ,
so that (7.33) is a consequence of the Jacobi identity for . Since g is semisimple
and I /I is finitedimensional, the cohomology space HL2 (g; I /I ) is trivial (for
a proof, see [102, Th. III.13]; Lie algebra cohomology is recalled in Section 4.1.1),
so every Lie 2cocycle is a Lie 2coboundary. The Lie 2cocycle c which we consider is constructed from (7.31):
c : gg I
/I
(x, y) p( Fx , Fy F[x,y] ) .
(7.34)
We have that c is a Lie 2cocycle, which means that for all x, y, z g,
x c(y, z) + c(x, [y, z])+ (x, y, z) = 0,
because, using (7.31),
x c(y, z) + c(x, [y, z]) = p( Fx , Fy , Fz ) p(F[x,[y,z]] ) .
Since HL2 (g; I /I ) is trivial, there exists a linear map b : g I /I , such that c
is the coboundary of b,
c(x, y) = x by y bx b[x,y]
(7.35)
for all x, y g. For every x g, let Fx := Fx bx , where the linear map b : g I is
an arbitrary lift of b. We claim that
Fx , Fy F[x,y]
I ,
(7.36)
for all x, y g. Since
Fx , Fy = Fx , Fy Fx , by bx , Fy + bx , by ,
(2+1)
(+2)
and since bx , by is an element of Im
(U) Im
(U) = I , as 1, (7.36)
amounts to saying that
Fx , Fy F[x,y] + b[x,y] Fx , by bx , Fy ,
(7.37)
which is an element of I , belongs to the kernel of p. According to the definition
of c, the image of (7.37) under p is given by
p Fx , Fy F[x,y] + p(b[x,y] ) x p(by ) + y p(bx )
= c(x, y) + b[x,y] x by + y bx ,
which is zero in view of (7.35). This proves (7.36). Since x Fx is a linear map,
there exists a map +1 from U to a neighborhood of o in Tm M, such that Fx =
x +1 , for every x g. Since dm = dm+1 , there exists a neighborhood U U
7.6 Notes
203
of m in M on which +1 is a diffeomorphism. In terms of +1 , (7.36) is written as
(+2)
{x +1 , y +1 } [x, y] +1 Im
(U ) ,
(7.38)
for all x, y g, which means that is linearizable up to order + 1 at m.
Remark 7.32. It is clear from the above proof that the open subsets on which the
diffeomorphisms are defined form a sequence which decreases with , and that
the intersection of all these open subsets may be reduced to a point. Notice that,
even if this intersection is not reduced to a point, the Poisson structure may not be
linearizable at m (see Example 7.30).
7.6 Notes
The fact that LiePoisson structures are in onetoone correspondence with Lie algebra structures is implicit in Lies works, in which one also finds the relation to the
coadjoint representation of the adjoint group of the Lie algebra. In the first half of
the twentieth century, the theory of Lie algebras and Lie groups became a subject of
its own, with focus on applications to theoretical physics and differential geometry.
The cited observations of Lie were rediscovered and reinforced in the works of Kirillov [105, 106], Kostant [117] and Souriau [185], who clearly exhibit the symplectic
structure on the coadjoint orbits which is inherited from the LiePoisson structure;
this structure is nowadays known as the KostantKirillovSouriau symplectic structure. For general information about Lie algebras and their representation theory, we
refer to Jacobson [102] or to the first chapter of Bourbaki [27].
Exactly as Lie algebra structures are in onetoone correspondence with linear
Poisson structures, Lie algebroids are in onetoone correspondence with fiberwise
linear Poisson structures [64], i.e., Poisson structures on a vector bundle which have
the property that the bracket of any two fiberwise linear functions is a fiberwise
linear function, while the bracket of a fiberwise linear function with a basic function
is a basic function. See [53] for an introduction to Lie algebroids.
The linearization problem was first posed and studied by Weinstein [199]. The
main results, which give conditions under which a Poisson structure is linearizable,
were obtained by Conn [45, 46]; a geometrical proof of these results is given in [52].
See DufourZung [63] for a detailed account on the linearization problem.
Chapter 8
Higher Degree Poisson Structures
Constant and linear Poisson structures, studied in the previous two chapters, are
particular cases of a bigger class of Poisson structures, namely the (weight) homogeneous Poisson structures on a finitedimensional vector space V : constant and
linear Poisson structures are the homogeneous Poisson structures of degree zero and
one, respectively. Like in the constant or linear case, (weight) homogeneous Poisson
structures are easier to study than general Poisson structures. For example, homogeneity of a Poisson structure permits one to split the determination of its Poisson
cohomology into its homogeneous parts. Also, weight homogeneous Poisson structures turn out to be useful for studying general Poisson structures. We will see an
example of this in Section 9.1.3 of the next chapter, when classifying Poisson structures on F2 , with a simple singularity.
Among distinguished classes of weight homogeneous Poisson structures, one
finds besides constant and linear Poisson structures quadratic Poisson structures (i.e., homogeneous Poisson structures of degree two), rank 2 Poisson structures
which arise from NambuPoisson structures and the transverse Poisson structures
to adjoint orbits in a semisimple Lie algebra.
Section 8.1 is devoted to the study of general properties of (weight) homogeneous Poisson structures, while we study in Section 8.2 a few special properties and
characterizations of quadratic Poisson structures: their multiplicativity, their modular class and the decomposition of a quadratic Poisson structure, using a related
unimodular Poisson structure. In Section 8.3 we define the notion of a Nambu
Poisson structure and we explain the construction which leads to a large class of
(weight homogeneous) rank 2 Poisson structures. Finally, Section 8.4 deals with the
transverse Poisson structure to a nilpotent orbit in a semisimple Lie algebra.
Unless otherwise stated, F denotes an arbitrary field of characteristic zero.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 8, SpringerVerlag Berlin Heidelberg 2013
205
206
8 Higher Degree Poisson Structures
8.1 Polynomial and (Weight) Homogeneous Poisson Structures
In this section, we define and study polynomial Poisson structures on a vector space,
with special emphasis on homogeneous and weight homogeneous Poisson structures.
Throughout the section, we consider Poisson structures and, more generally, multivector fields on a finitedimensional vector space V , which is equipped with its algebra of functions, denoted by F (V ). If F is an arbitrary field, this algebra is the algebra of polynomial functions on V ; however, if F = R (respectively F = C), F (V )
may stand either for the algebra of smooth (respectively holomorphic) or polynomial functions on V , depending on the context. It is clear that, in either case, F (V )
contains all polynomial functions on V . For p N, the corresponding space of pvector fields on V (skewsymmetric pderivations of F (V )) is denoted by X p (V ).
The origin of V will be denoted by o.
8.1.1 Polynomial Poisson Structures
Let V be a vector field on a finitedimensional vector space V . Then, V is said to
be a polynomial vector field on V if, for every polynomial function F on V , V [F]
is a polynomial function on V . More generally, a pvector field P on V (p N ) is
said to be a polynomial pvector field on V if P[F1 , . . . , Fp ] is a polynomial function
on V , for all polynomial functions F1 , . . . , Fp on V . In the case of Poisson structures
on V , this leads to the following definition.
Definition 8.1. A Poisson structure on a finitedimensional vector space V is said
to be a polynomial Poisson structure if for all polynomial functions F and G on V ,
their Poisson bracket {F, G} is a polynomial function on V .
In terms of arbitrary linear coordinates x1 , . . . , xd of V , every pvector field P on V
can be written as
P=
1i1 <<i p d
P[xi1 , . . . , xi p ]
,
x i1
xi p
(8.1)
so that P is a polynomial pvector field if and only if P[xi1 , . . . , xi p ] is a polynomial
function, for all 1 i1 < < i p d. In this case, the maximum of the degrees of
all P[xi1 , . . . , xi p ] is called the degree of P. In particular, every polynomial Poisson
structure on V is of the form
1i< jd
xi j
,
xi x j
where, for every 1 i < j d, the function xi j := xi , x j is a polynomial function
on V , and the degree of is the maximum of the degrees of the polynomials xi j .
8.1 Polynomial and (Weight) Homogeneous Poisson Structures
207
Example 8.2. Constant Poisson structures (see Chapter 6) are polynomial Poisson
structures of degree zero, while linear and affine Poisson structures (see Chapter 7)
are polynomial Poisson structures of degree one.
It is clear that when F (V ) is the algebra of polynomial functions on V , then all
Poisson structures on V are polynomial Poisson structures on V .
8.1.2 Homogeneous Poisson Structures
Let V be a finitedimensional vector space. A function F F (V ) is said to be
homogeneous if there exists an integer r N, such that for every F and for
every v V , we have
F( v) = r F(v) .
(8.2)
The integer r, which is unique if F = 0, is then called the degree of F, denoted deg(F). If F is expressed in terms of a system of linear coordinates (x1 , . . . , xd )
for V , which we write as F = F(x1 , . . . , xd ), then condition (8.2) is equivalent to the
fact that
F( x1 , . . . , xd ) = r F(x1 , . . . , xd ) ,
(8.3)
for every F. A homogeneous function of degree zero is constant, since the condition F( v) = F(v) for all F and v V implies that F(v) = F(0) for every v V .
The following proposition states that smooth or holomorphic functions which are
homogeneous, are polynomial functions.
Proposition 8.3. Let V be a finitedimensional real (respectively complex) vector
space. If F is a smooth (respectively holomorphic) function on V and F is homogeneous of degree r, then F is a polynomial function on V , whose degree is r.
Proof. Let F F (V ) be a smooth or holomorphic function on V , satisfying (8.2)
for every F and every v V . By differentiating, for a fixed , both sides of
this equation in the direction of a vector u V , we obtain that, for every u V , the
directional derivative of F along the vector u is a homogeneous function of degree
r 1. By iterating this procedure, we obtain that the directional derivatives of F of
order r are all homogeneous of degree 0, i.e., are constant functions. This implies
that F is a polynomial function on V , of degree less than or equal to r. Then, because
of equation (8.2), we can conclude that the degree of F is r.
According to Proposition 8.3, the notions of homogeneous function and homogeneous polynomial function do not have to be distinguished and the notion of
degree coincides with the usual notion of degree of a polynomial.
A (smooth or holomorphic) vector field V on a finitedimensional vector space V
is said to be a homogeneous vector field on V if, for every F F (V ) which is homogeneous, the function V [F] F (V ) is also homogeneous. More generally, a
(smooth or holomorphic) pvector field P X p (V ) is said to be homogeneous if,
208
8 Higher Degree Poisson Structures
for all F1 , . . . , Fp F (V ) which are homogeneous, the function P[F1 , . . . , Fp ] is homogeneous. Since every polynomial function is the sum of homogeneous functions,
Proposition 8.3 implies that every homogeneous pvector field on V is a polynomial
pvector field on V .
In the following proposition, we give a few characterizations of homogeneous
multivector fields. For one of them, we use the Euler vector field on V , which is
denoted by E and is defined as the unique vector field on V which acts as the identity
on all linear functions on V ; in terms of linear coordinates x1 , . . . , xd on V , it is
given by
+ + xd
.
(8.4)
E = x1
x1
xd
Clearly, E is a homogeneous vector field on V . Notice that, by differentiating (8.3)
with respect to , at = 1, we obtain that a function F on V is homogeneous of
degree r N if and only if
E [F] = LE F = r F .
(8.5)
This formula is called the Euler formula.
Proposition 8.4. Let P be a nonzero pvector field on a finitedimensional vector
space V , where p N. The following conditions on P are equivalent.
(i)
(ii)
(iii)
(iv)
P is homogeneous;
There exists an integer r N, such that for all linear functions F1 , . . . , Fp on V ,
one has that P[F1 , . . . , Fp ] is zero, or is a homogeneous function of degree r;
There exists an integer r N, such that for all homogeneous functions
F1 , . . . , Fp on V , one has that P[F1 , . . . , Fp ] is zero, or is a homogeneous function of degree deg(F1 ) + + deg(Fp ) + r p;
There exists an integer r N, such that LE P = (r p)P.
The integer r, which appears in (ii)(iv) is the same and is called 1 the degree of P,
denoted deg(P), while r p is called the weight of P, denoted (P).
Proof. Let P be a nonzero pvector field, P X p (V ). We assume that p 1, since
the proposition is clear for p = 0. Condition (ii) implies obviously condition (i), in
view of (8.1). We only show that (i) implies (ii) for p = 1, since the generalization of
the proof to arbitrary p 1 will be clear. Thus, we assume that P is a homogeneous
vector field, and we consider two linear functions F and G on V , satisfying P[F] = 0
and P[G] = 0. For a generic t F the polynomial P[(1 t)F + tG] is nonzero and
homogeneous, with a degree r which is independent of t; in particular, since P[F]
and P[G] are different from zero, they have the same degree r, which proves (ii).
The equivalence between conditions (ii) and (iii) follows immediately from the fact
that a pvector field P is a derivation in each of its arguments. Finally, let F1 , . . . , Fp
be homogeneous elements of F (V ) and let r N. Using (3.7) and (8.5), we have
1
In order that the formulas which involve the degree or weight are valid also for the zero pvector
field, we adopt the convention that both the degree and weight of the zero pvector field are .
8.1 Polynomial and (Weight) Homogeneous Poisson Structures
209
(LE P (r p) P) [F1 , . . . , Fp ]
p
= E [P[F1 , . . . , Fp ]] P[F1 , . . . , E [F ], . . . , Fp ] (r p) P[F1 , . . . , Fp ]
=1
= E [P[F1 , . . . , Fp ]]
deg(F ) (r p)
P[F1 , . . . , Fp ] .
=1
Using (8.5), we conclude that there exists r N such that LE P = (r p) P if and
only if there exists r N such that the following holds: for all homogeneous elements F1 , . . . , Fp of F (V ), we have that P[F1 , . . . , Fp ] is zero, or is homogeneous of
p
deg(F ) (r p). This shows the equivalence (iii) (iv).
degree =1
As a corollary, every polynomial pvector field P on V can be written uniquely
as a finite sum of homogeneous pvector fields on V , whose degrees are pairwise
different.
Notice that, according to (iii) of Proposition 8.4, if F1 , . . . , Fp F (V ) and
P X p (V ) are homogeneous, then the polynomial P[F1 , . . . , Fp ] is either zero, or
is homogeneous of degree
deg (P[F1 , . . . , Fp ]) = deg(F1 ) + + deg(Fp ) + deg(P) p ,
(8.6)
which, in terms of weights, becomes
(P[F1 , . . . , Fp ]) = (F1 ) + + (Fp ) + (P) .
(8.7)
Since LE E = [E , E ] = 0, the Euler vector field is weight homogeneous of degree
one, i.e., of weight zero.
Specializing the definition of homogeneous bivector fields to Poisson structures
on V , we obtain the following definition.
Definition 8.5. Let V be a finitedimensional vector space. A Poisson structure
on V is said to be homogeneous if for all F, G F (V ), which are homogeneous,
their Poisson bracket {F, G} F (V ) is also homogeneous.
According to Proposition 1.35, every Poisson structure on V can be written in the
following form:
= xi j
xj
i
1i< jd
according to Proposition
8.4, it is homogeneous (of degree r) if and only if all the
functions xi j := xi , x j are homogeneous of the same degree r; in this case, LE =
(r 2) .
Example 8.6. Let be a Poisson structure on a finitedimensional vector space V .
Then is homogeneous of degree zero (equivalently, of weight 2) if and only if
is a constant Poisson structure. Also, is homogeneous of degree one (equivalently,
of weight 1) if and only if is a linear Poisson structure.
210
8 Higher Degree Poisson Structures
Example 8.7. Consider the bivector field := V1 V2 , where V1 and V2 are two
homogeneous vector fields on a finitedimensional vector space V . Then is homogeneous of weight (V1 ) + (V2 ), because
LE ( ) = LE (V1 ) V2 + V1 LE (V2 ) = ( (V1 ) + (V2 )) V1 V2 .
Since
[ , ]S = [V1 V2 , V1 V2 ]S = 2 V1 V2 [V1 , V2 ] ,
we have that is a (homogeneous) Poisson structure on V if and only if the vector fields V1 , V2 , [V1 , V2 ] are linearly dependent at all points of V . This condition is
automatically satisfied when V1 is the Euler vector field E and V2 = V is an arbitrary homogeneous vector field, since then [E , V ] = LE V = (V ) V , which is
a multiple of V . The Poisson structure E V is homogeneous of weight (V ),
since (E ) = 0.
To finish this section, we give a proposition whose main content is that an isomorphism between two homogeneous Poisson structures on a vector space can always
be realized by a linear isomorphism.
Proposition 8.8. Let 1 and 2 be two homogeneous Poisson structures on a finitedimensional vector space V . Then the following are equivalent:
(i)
(ii)
(iii)
There exists a Poisson isomorphism L : (V, 1 ) (V, 2 ) which is a linear
map;
There exists a Poisson isomorphism : (V, 1 ) (V, 2 ) with (o) = o;
There exists a Poisson isomorphism : (U1 , 1 ) (U2 , 2 ) with (o) = o,
where U1 ,U2 are neighborhoods of the origin o in V .
Proof. The implications (i) = (ii) = (iii) are clear. Assume now that (iii) holds,
and let : (U1 , 1 ) (U2 , 2 ) be a Poisson isomorphism with (o) = o. We are
going to prove that the differential L = do of at the origin is a linear Poisson
isomorphism between (V, 1 ) and (V, 2 ). This can be done by comparing the Taylor
expansions at o of both sides of the following identity, which is valid for all functions
F, G F (V2 )
{F, G}2 = {F , G }1
where { , }1 = 1 and { , }2 = 2 are the Poisson brackets corresponding to 1
and 2 . Since 1 and 2 are homogeneous of degree k, the Taylor expansions at
the origin o of both sides of this equality vanish for degrees 0, . . . , k 1, while the
components of degree k are equal to {do F, do G}2 L and {do F L, do G L}1 respectively. Since every linear function on V is the differential at the origin of a function
in F (V ), we obtain that for all linear functions F , G F (V ),
F , G 2 L = F L, G L 1
which, in turn, implies that L is a Poisson map. Moreover, since L = do is invertible, L is in fact a Poisson isomorphism.
8.1 Polynomial and (Weight) Homogeneous Poisson Structures
211
We will use the above proposition in Section 9.2.3, when we classify the set of
all quadratic Poisson structures on C3 .
8.1.3 Weight Homogeneous Poisson Structures
Many known examples of (nonhomogeneous) Poisson structures on a vector space
Fd become homogeneous Poisson structures upon assigning specific (positive integer) weights to the natural coordinates x1 , . . . , xd on Fd . This means that many properties and techniques of homogeneous Poisson structures carry over to this more
general class of Poisson structures, which we will call weight homogeneous Poisson structures.
We first introduce the notion of weight homogeneity for functions and for multivector fields on Fd . A dtuple of integers := (1 , . . . , d ) will be called a weight
vector (for Fd ); unless otherwise stated, we will always assume that all integers i
are positive. Generalizing (8.3), a function F on Fd is called weight homogeneous
with respect to if there exists an integer r N, such that
F( 1 x1 , . . . , d xd ) = r F(x1 , . . . , xd ) ,
(8.8)
for all F. In this case, the integer r, which is unique if F = 0, is called the weight
of F and is denoted by (F). It is easy to see that, if F, G F (Fd ) are weight
homogeneous with respect to , then their product FG is also weight homogeneous
with respect to , of weight (FG) = (F) + (G).
Remark 8.9. It is clear that a function F on Fd is homogeneous if and only if it is
weight homogeneous with respect to = (1, . . . , 1), and in this case, the weight
of F is equal to the degree of F.
Taking F := xi in (8.8), we see that each coordinate function xi , with 1 i d is
weight homogeneous of weight i , so that we often refer to i as being the weight
of xi .
Similarly to the case of homogeneous functions (see Proposition 8.3), we show
in the following proposition that a smooth or holomorphic function which is weight
homogeneous, is a polynomial function.
Proposition 8.10. If F is a smooth function on Rd (respectively a holomorphic function on Cd ) and F is weight homogeneous with respect to some weight vector, then F
is a polynomial function.
Proof. Let = (1 , . . . , d ) be a weight vector for Fd (F = R or F = C) and assume
that F F (Fd ) satisfies (8.8), for all F. Let us consider s N , such that, for all
1 j1 , . . . , js d, we have that j1 + + js > r. By differentiating both sides of
equation (8.3) with respect to xi1 , . . . , xis , for all 1 i1 , . . . , is d, we obtain that all
s
functions xi ...F xi have negative weight, hence are zero on Fd . This implies that F
1
is necessarily a polynomial function on Fd .
212
8 Higher Degree Poisson Structures
A pvector field P on Fd is said to be weight homogeneous (with respect to ),
if for all F1 , . . . , Fp F (Fd ), which are weight homogeneous with respect to ,
the function P[F1 , . . . , Fp ] F (Fd ) is weight homogeneous with respect to . One
shows, as in the case of homogeneous multivector fields, that a weight homogeneous
multivector field on Fd is a polynomial multivector field.
Remark 8.11. It is clear that a multivector field on Fd is homogeneous if and only if
it is weight homogeneous with respect to = (1, . . . , 1).
We will give a few characterizations of the weight homogeneity of a multivector
field, which generalize the characterizations of the homogeneity of a multivector
field, given in Proposition 8.4. To do this, we introduce the weighted Euler vector
field E on Fd , which is defined by
E := 1 x1
+ + d xd
.
x1
xd
(8.9)
Notice that the weighted Euler vector field E is weight homogeneous with respect to . It is clear from (8.8) that a function F on Fd is weight homogeneous of
weight r if and only if
(8.10)
E [F] = LE F = r F .
This formula, which generalizes the classical Euler formula (8.5), is called the
weighted Euler formula.
Proposition 8.12. Let = (1 , . . . , d ) be a weight vector for Fd and let P be a
nonzero pvector field on Fd , where p N. The following conditions on P are
equivalent.
(i)
(ii)
(iii)
(iv)
P is weight homogeneous with respect to ;
There exists an integer r Z, such that for all 1 i1 < < i p d, the
function P[xi1 , . . . , xi p ] is zero, or is weight homogeneous of weight r + i1 +
+ i p ;
There exists an integer r Z, such that for all weight homogeneous elements
F1 , . . . , Fp of F (Fd ), one has that P[F1 , . . . , Fp ] is zero, or is weight homogeneous of weight r + (F1 ) + + (Fp );
There exists an integer r Z, such that LE P = rP.
The (possibly negative) integer r, which appears in (ii)(iv) is the same and is
called 2 the weight of P, denoted (P).
Proof. Since the proof is very similar to the proof of Proposition 8.4, we will not
repeat it here; we only focus on the implication (i)(ii), which requires some extra
explanation. Suppose that V is a vector field on Fd and that for some 1 i < j d
the functions V [xi ] and V [x j ] are different from zero. We claim that if V is weight
2 In order that the formulas which involve the weight are valid also for the zero pvector field, we
adopt, as for homogeneous multivector fields, the convention that both the degree and weight of
the zero pvector field are .
8.2 Quadratic Poisson Structures
213
homogeneous, then the integers (V [xi ]) i and (V [x j ]) j are equal. In or
der to prove this property, we consider for t F the function V (1 t)xi j + txj i ,
which is weight homogeneous, because (1 t)xi j + txj i and V are weight homo
geneous. As in the proof of Proposition 8.4, it follows that V xi j and V xj i
j 1
have the same weight. Since V [xi j ] = j xi
we have that
V [xi ] and V [xj i ] = i xj i 1 V [x j ],
i ( j 1) + (V [xi ]) = j (i 1) + (V [x j ]) ,
and hence (V [xi ]) i = (V [x j ]) j , which is what we wanted to show.
It follows, as in the case of homogeneous multivector fields, that every polynomial pvector field P X p (Fd ) can be written in a unique way as a finite sum of
weight homogeneous pvector fields, whose weights are pairwise different. Also, if
P X p (Fd ) and F1 , . . . , Fp F (F d ) are weight homogeneous, then the polynomial
P[F1 , . . . , Fp ] is weight homogeneous, of weight
(P[F1 , . . . , Fp ]) = (P) + (F1 ) + + (Fp ) .
(8.11)
Since LE E = [E , E ] = 0, the weighted Euler vector field is weight homogeneous of weight zero.
Finally, we specialize the definition of weight homogeneity to the case of Poisson
structures.
Definition 8.13. A Poisson structure on Fd is said to be weight homogeneous if for
all F, G F (Fd ) which are weight homogeneous with respect to , their Poisson
bracket {F, G} is weight homogeneous with respect to .
Writing a Poisson structure on Fd in the form
xi j
1i< jd
,
xi x j
it follows that is weight homogeneous(of weight
( )) with respect to if and
only if each one of the functions xi j := xi , x j is a weight homogeneous polynomial, of weight ( ) + i + j ; this is also equivalent to the equality
LE [ ] = ( ) .
8.2 Quadratic Poisson Structures
A bivector field on a finitedimensional vector space V is said to be quadratic, if it is
homogeneous of degree 2. A quadratic Poisson structure on V is a Poisson bivector
214
8 Higher Degree Poisson Structures
field on V which is quadratic. In terms of an arbitrary system of linear coordinates
(x1 , . . . , xd ) on V , every quadratic Poisson structure takes the form
xi j
1i< jd
,
xi x j
(8.12)
where all xi j := xi , x j , with 1 i < j d, are homogeneous polynomial functions
of degree 2 on V . Notice that a Poisson structure on V is quadratic if and only if the
Poisson bracket of every pair of linear functions on V is a homogeneous polynomial
function of degree 2 on V .
To start with, we give two important classes of quadratic Poisson structures.
Example 8.14. Let A = (ai j ) Matd (F) be an arbitrary skewsymmetric d d matrix. A quadratic bivector field = { , } is defined on Fd by setting
:=
1i< jd
ai j xi x j
.
xi x j
(8.13)
A direct computation from (8.13) gives, for all 1 i < j < k d,
xi , x j , xk + (i, j, k) = ai j (aik + a jk ) xi x j xk + (i, j, k) ,
which is zero since A is skewsymmetric. This proves the Jacobi identity for every
triple of coordinate functions, which shows, in view of Proposition 1.36, that the
bivector field is a Poisson structure on Fd . A quadratic Poisson structure of the
form (8.13) is called a diagonal Poisson structure.
Notice that the rank of this Poisson structure is equal to the rank of the matrix A.
More precisely, for every m Fd , the rank of at m is equal to the rank of the
matrix, obtained by removing from the matrix A the kth line and the kth column
for all k for which xk (m) = 0.
Example 8.15. If V is a linear vector field on a finitedimensional vector space V ,
then E V is a quadratic bivector field, where E is the Euler vector field (8.4). Since
[E V , E V ]S = 2 E V [E , V ] = 0 ,
the bivector field E V is a quadratic Poisson structure. It is a particular case of the
class of homogeneous Poisson structures, constructed in Example 8.7.
In the following proposition, we give two characterizations of quadratic Poisson
structures. We recall from Section 1.3.2 that, by definition, a vector field V on (V, )
is a Poisson vector field if the Lie derivative of with respect to V is zero.
Proposition 8.16. For a Poisson structure on a finitedimensional vector space V ,
the following conditions are equivalent:
(i)
(ii)
The Poisson structure is quadratic;
The Euler vector field E is a Poisson vector field;
8.2 Quadratic Poisson Structures
(iii)
215
For every a F , the scalar multiplication by a is a Poisson automorphism
of V .
Proof. The equivalence between (i) and (ii) is a particular case of the equivalence
between (i) and (iv) in Proposition 8.4 with p = 2. Let = { , } be a Poisson
structure on V . The multiplication by a F is a Poisson map if and only if
{F, G} (av) = {aF, aG} (v) = a2 {F, G} (v) ,
for all linear functions F, G V and all v V , which is equivalent to saying that
{F, G} is a homogeneous polynomial function of degree 2, for all F, G V . By
Proposition 8.4, the latter means that is quadratic. This shows that (i) and (iii) are
equivalent.
The projective space P(V ), which is the variety of 1dimensional subspaces of V ,
is naturally identified with the quotient space (V \{o})/F of the free action of F
by scalar multiplication. By Propositions 5.33 and 8.16, if is a quadratic Poisson structure on V , then P(V ) admits a (unique) Poisson structure, such that the
canonical projection map V \ {o} P(V ) is a Poisson map.
8.2.1 Multiplicative Poisson Structures
Let V be a finitedimensional vector space and let : V V V be a bilinear map.
We refer in this section to as a product on V , even if we do not demand that
is associative. We say that is a product with unit, if V admits an element e, such
that (v, e) = v = (e, v) for all v V . A Poisson structure on V is said to be
multiplicative with respect to if : V V V is a Poisson map, where V V is
endowed with the product Poisson structure.
We show in this section that Poisson structures, which are multiplicative with
respect to a product with unit, are quadratic. We will do this by using the following
elementary lemma.
Lemma 8.17. The only polynomial p in one variable, satisfying
p(ab) = a2 p(b) + b2 p(a) ,
(8.14)
for all a, b F, is the zero polynomial.
Proof. Let p be a polynomial which satisfies (8.14), for all a, b F. The degree r of
p is at most 2, because (a, b)
p(ab) is a polynomial function (in two variables)
of total degree 2r, while the total degree of a2 p(b) + b2 p(a) is at most r + 2. But
(8.14) implies that p(0) = p(1) = p(1) = 0, so that p is the zero polynomial.
We now establish the link between multiplicative and quadratic Poisson structures.
216
8 Higher Degree Poisson Structures
Proposition 8.18. Let V be a finitedimensional vector space and let : V V V
be a product with unit. Every polynomial Poisson structure on V which is multiplicative with respect to , is quadratic.
Proof. Let be a polynomial Poisson structure on V , which is multiplicative. In
view of (2.12), multiplicativity of { , } = implies that, for all F, G F (V ), and
all v, w V :
{F, G} ( (v, w)) = {F Rw , G Rw } (v) + {F Lv , G Lv } (w) ,
(8.15)
where Rw : V V and Lv : V V are the endomorphisms of V , defined for u V
by Rw (u) := (u, w) and Lv (u) := (v, u) respectively. By bilinearity of , for every
a F and v V , the equalities (v, ae) = av = (ae, v) hold, as well as the dual
equalities F Rae = aF = F Lae , valid for all linear functions F on V and all a F.
Let F, G be two linear functions on V . For all a, b F, (8.15) specializes to
{F, G} (abe) = {F, G} (ae, be) = a2 {F, G} (be) + b2 {F, G} (ae) .
(8.16)
The map p : F F, defined for all a F by p(a) := {F, G} (ae), is a polynomial
function, since is a polynomial Poisson structure, and (8.16) states that p satisfies
condition (8.14) of Lemma 8.17. Therefore, {F, G} (ae) = 0 for all a F. Specializing (8.15), we obtain therefore that
{F, G} (av) = {F, G} ( (v, ae)) = a2 {F, G} (v) ,
for all a F and v V . This shows that {F, G} is a homogeneous polynomial of degree 2. Since this holds for all linear functions F and G on V , the Poisson structure
is quadratic.
Proposition 8.18 takes in the smooth or holomorphic context the following form.
Proposition 8.19. Let V be a finitedimensional vector space over R or over C and
let : V V V be a product with unit. Every smooth or holomorphic Poisson
structure on V , multiplicative with respect to , is quadratic.
Proof. The proof of the proposition goes along the same lines as the proof of Proposition 8.18; the only difference is that one needs to reprove Lemma 8.17 in that
context, namely, to prove that there exists no nonzero smooth or holomorphic function p satisfying (8.14). Deriving condition (8.14) three times with respect to a,
yields that if p satisfies (8.14) for all a, b F, it also satisfies bp (ab) = p (a), for
all a, b F. Plugging b = 0 in it, we obtain that p (a) = 0 for all a F, and therefore that p is a polynomial. According to Lemma 8.17, p is the zero polynomial.
Remark 8.20. It is clear that the only differentiability condition which we use on
the function p in Proposition 8.19 is that p is of class C3 in a neighborhood of o
in V . Therefore, Proposition 8.19 can be easily generalized to multiplicative bivector
8.2 Quadratic Poisson Structures
217
fields of class Ck , with k 3, where for k N , we call bivector field of class Ck on
Rd every bidifferential operator of the form
1i< jd
xi j
,
xi x j
where all the functions xi j are of class Ck on Rd . Notice that, if is of class Ck on
Rd (with k N ), then for every pair of smooth functions (F, G) on V , the bracket
{F, G} is a function of class Ck , so one can still consider the Jacobi identity for
smooth functions and define Poisson structures of class Ck , with k N , by saying
that a bivector field on Rd is a Poisson structure of class Ck if satisfies the Jacobi
identity for all triples of smooth functions F, G, H on V .
For Poisson structures of class C2 , it can be shown that Proposition 8.19 still
holds, but for Poisson structures of class C1 this is not the case (in general), as is
shown in the following example.
Example 8.21. On R2 we consider the product (with unit) which corresponds
to the natural product on R[h]/h2 , via the vector space isomorphism, defined by
(a1 , a2 )
a1 + ha2 . Explicitly, is given by
((a1 , a2 ), (b1 , b2 )) = (a1 b1 , a1 b2 + a2 b1 ) ,
and the unit for is (1, 0). It follows that x1 = y1 z1 and x2 = y1 z2 + z1 y2 ,
where (x1 , x2 ), respectively (y1 , y2 , z1 , z2 ), are the natural systems of coordinates
on R2 , respectively on R2 R2 R4 .
A Poisson structure of class C1 on R2 is defined by {x1 , x2 } := x12 ln(x1 ). We
denote the product Poisson structure on R4 = R2 R2 by . In order to show that
is multiplicative with respect to , notice that {x1 , x2 } = {x1 , x2 } , which
follows from
{x1 , x2 } = (x12 ln(x1 )) = (y1 z1 )2 ln(y1 z1 )
and
{x1 , x2 } = {y1 z1 , y1 z2 + z1 y2 }
= z21 {y1 , y2 } + y21 {z1 , z2 }
= z21 y21 ln(y1 ) + y21 z21 ln(z1 ) .
However, is not quadratic.
Example 8.22. Let V := Matd (F) the vector space of square matrices of size d. For
1 i, j d, we consider on V the linear function, defined for all x = (xi j )di, j=1 V
by i j (x) := xi j . These functions provide a set of linear coordinates on V . For i and j
as above, let i, j := 1 if i > j and i, j := 1 if i < j and i, j := 0 if i = j. Then the
brackets, defined for 1 i, j, k, d, by
218
8 Higher Degree Poisson Structures
i j , k := (i,k + j, )i k j
define a multiplicative quadratic Poisson structure on V . This can be checked by
direct computation; see Example 11.14 below for a more conceptual proof.
8.2.2 The Modular Vector Field of a Quadratic Poisson Structure
On a finitedimensional vector space V , which is equipped with a Poisson structure,
there is a vector field, which is naturally associated to the Poisson structure: the
modular vector field, with respect to an arbitrary translation invariant volume form
on V . Recall from (4.24) that the modular vector field of a Poisson structure
on V is, up to a sign, the divergence Div( ) of , i.e., = Div( ). As a first
application, we have, in view of the explicit formula (4.22) for the divergence of a
bivector field, that the modular vector field is homogeneous of degree r 1, if the
Poisson structure is homogeneous of degree r; in the particular case of quadratic
Poisson structures, this leads to the following proposition.
Proposition 8.23. Let be an arbitrary translation invariant volume form on a
finitedimensional vector space V . The modular vector field of every quadratic Poisson structure on V , with respect to , is a linear vector field on V .
In the study of quadratic Poisson structures, the modular vector field plays a particular role, which we now underline, and which will lead in Section 9.2.3 to a
classification of all quadratic Poisson structures on C3 .
Example 8.24. We compute the modular vector field of the diagonal Poisson
structure on Fd , introduced in Example 8.14 with respect to the translation invariant volume form = dx1 dxd . By using the formula = Div( ) and the
explicit expression of the divergence of a bivector field (4.22), we obtain
d
d
= 2 ai j xi
.
(8.17)
xi
i=1 j=1
Example 8.25. Let V be a ddimensional vector space and let V be a linear vector
field on V . As we have seen in Example 8.15, := E V is a quadratic Poisson
structure on V . In order to compute its divergence, notice that Div(E ) = d, which
follows from (4.21), and that [E , V ] = 0, because the weight of a linear vector field
is 0. Substituted in equation (4.23) for the divergence of a wedge product, one finds
Div(E V ) = Div(V ) E d V ,
(8.18)
so that the modular vector field of the Poisson structure = E V is =
d V Div(V ) E .
We show in the following proposition that every quadratic Poisson structure can
be decomposed as a sum of two compatible quadratic Poisson structures: one of the
8.2 Quadratic Poisson Structures
219
terms involves only the modular vector field of , while the other term is unimodular. This decomposition is very useful, as we will see in Section 9.2.3.
We first recall two more facts from Section 4.4.3, namely that, given a volume
form on V , the star operator is the family of isomorphisms
: Xq (V ) dq (V ) ,
defined for Q Xq (V ) by Q = Q as in (4.19), and that Div is defined in terms of
the de Rham differential d by
Div = 1 d .
Since = Div( ), it follows that is unimodular if and only if the differential
(d 2)form is closed.
Proposition 8.26. Let V be a ddimensional vector space and let be a translation
invariant volume form on V . Every quadratic Poisson structure on V decomposes
as follows
1
= 1 + E ,
d
where is the modular vector field of with respect to , and where is a closed
differential (d 2)form which satisfies L = 0. Also, 1 has the following
properties: it is a quadratic unimodular Poisson structure, compatible with , and
it satisfies L (1 ) = 0.
Proof. Since is a linear vector field, E is a quadratic Poisson structure (see
Example 8.15). This Poisson structure is compatible with , since
[E , ]S = (LE ) E (L ) = 0 ,
where we have used in the last step that both E and are Poisson vector fields
(with respect to ). The modular vector field of the Poisson structure :=
E is given by
= Div(E ) = d Div( ) E = d ,
(8.19)
where we have used (8.18), together with the fact that Div( ) = 0, which holds
because is itself the divergence of a bivector field. Since the quadratic Poisson
structures and = E are compatible, and since their modular vector fields
are proportional (see (8.19)), there exists a linear combination of and E
which is a unimodular Poisson structure: R := d1 E is a Poisson structure
on V which is quadratic, compatible with and unimodular.
Let us denote by the differential (d 2)form on V , defined by := R = R .
It is a closed form since R is unimodular. It remains to be checked that L R = 0
and that L = 0. First, we have that L = 0 (since is a Poisson vector
field), and we also have that L (E ) = 0 (since the vector fields E and
commute). We therefore have L R = 0. It follows from this, upon using (2) of
220
8 Higher Degree Poisson Structures
Proposition 3.11, that
L = L (R ) = R (L ) + [ ,R]S = 0 ,
where we used in the last step that [ , R]S = L R = 0 and that L =
Div( ) = 0 (see (4.20)). This finishes the proof of the announced properties
of and of R = 1 .
The quadratic Poisson structures whose modular vector field satisfies a genericity
assumption, stated below, are diagonal, when expressed in a system of wellchosen
coordinates. We devote the end of this section to explain and to prove this. First, we
introduce the required notions. Let V be a linear vector field on a ddimensional
vector space V . Since V sends linear functions on V (elements of V ) to linear
functions on V , we can restrict V to V , which yields an endomorphism V (1) of V ,
V (1) : V V
F
V [F] .
(8.20)
Definition 8.27. Let V be a linear vector field on a finitedimensional vector space V . We say that V is diagonalizable if the endomorphism V (1) of V is diagonalizable. In this case, the eigenvalues, respectively eigenvectors of V (1) are called
eigenvalues, respectively eigenvectors of V .
It is clear that, if (x1 , . . . , xd ) is a basis of eigenvectors of a diagonalizable linear
vector field V , with corresponding eigenvalues a1 , . . . , ad F, then (x1 , . . . , xd ) is
a system of linear coordinates on V , and that V takes, in these coordinates, the
following form:
d
.
(8.21)
V = ai xi
xi
i=1
It is also clear that, conversely, every vector field on V , which is of the form (8.21)
is a diagonalizable linear vector field.
Comparing (8.17) and (8.21), we see that the modular vector field of every diagonal quadratic Poisson structure is diagonalizable. As we will see, the converse is
almost true: up to a technical assumption on the eigenvalues of the modular vector
field, a quadratic Poisson structure is diagonal (in some system of linear coordinates)
if its modular vector field is diagonalizable. It motivates the following definition of
a diagonalizable Poisson structure.
Definition 8.28. A quadratic Poisson structure on a finitedimensional vector
space V is said to be diagonalizable, if there exist linear coordinates x1 , . . . , xd on V
in terms of which is diagonal, i.e.,
1i< jd
ai j xi x j
,
xi x j
(8.22)
where (ai j ) Matd (F) is a skewsymmetric matrix. The system of linear coordinates
(x1 , . . . , xd ) is then said to be logcanonical for .
8.2 Quadratic Poisson Structures
221
The abovementioned converse, which yields a sufficient condition for the diagonalizability of a quadratic Poisson structure is given in the following proposition.
Proposition 8.29. Let be a quadratic Poisson structure on a ddimensional vector
space V . Suppose that
(1)
(2)
The modular vector field of is diagonalizable, with eigenvalues a1 , . . . ,
ad , which
belong to F;
sums ai + a j with 1 i j d are all different.
The d+1
2
Then is a diagonalizable Poisson structure on V . Moreover, every system of linear
coordinates on V , consisting of eigenvectors of , is logcanonical for .
Proof. Since the modular vector field is linear, it restricts to a linear endomor(2)
phism of the space A2 of all homogeneous polynomial functions of degree 2
on V ,
(2)
: A2 A2
F
[F] .
Suppose that is diagonalizable and let x1 , . . . , xd be independent eigenvectors
of , with
eigenvalues a1 , . . . , ad . We use x1 , . . . , xd as linear coordinates on V .
polynomial functions {xi x j  1 i j d} form a basis of A2 , and
The d+1
2
(2)
each one of them is an eigenvector of , because is a derivation: xi x j is an
(2)
eigenvector of , corresponding to the eigenvalue ai + a j , for all 1 i j d.
sums ai + a j  1 i j d are all different (which
If we assume that the d+1
2
(2)
is assumption (2)), then all eigenvalues of are different, hence the eigenspace
(2)
decomposition of A2 with respect to is given by
A2 =
Cxi x j .
(8.23)
1i jd
We show that this implies that is diagonalizable. To do this, we use that the modular vector field is a Poisson vector field: writing { , } for and specializing
[{F, G}] = { [F], G} + {F, [G]} ,
to F = xi and G = x j , where 1 i, j d, we obtain that
[ xi , x j ] = [xi ], x j + xi , [x j ] = (ai + a j ) xi , x j .
(2)
Therefore, xi , x j is zero or it is an eigenvector of with respect to the eigenvalue ai +a j . Inview of the eigenspace decomposition (8.23), it follows, in either
case, that xi , x j is proportional to xi x j , i.e., there exists a unique constant ai j F,
such that xi , x j = ai j xi x j . By construction, ai j = a ji , and is of the diagonal
form (8.22) in the system of coordinates (x1 , . . . , xd ).
222
8 Higher Degree Poisson Structures
8.3 NambuPoisson Structures
Let V be a ddimensional vector space, equipped with linear coordinates x1 , . . . , xd
and denote the algebra of functions on V by F (V ), as before. Let , 1 , . . . , d2
be d 1 functions on V and define, for F, G F (V ), a function on V by
F
x1
(F, G, 1 , . . . , d2 )
{F, G} :=
= ...
(x1 , . . . , xd )
F
xd
G 1
x1 x1
..
.
G 1
xd xd
d2
x1
..
.
d2
xd
(8.24)
It leads to a map F (V ) F (V ) F (V ), which is a skewsymmetric biderivation
of F (V ) and which, up to a nonzero constant, does not depend on the chosen linear
coordinates x1 , . . . , xd , since (8.24) can also be written as
{F, G} =
dF dG d1 dd2
.
dx1 dx2 dxd
(8.25)
Moreover, it defines a Poisson structure on V , a fact which we will prove in this section. When the functions 1 , . . . , d2 are polynomial or weight homogeneous, then
so is the corresponding Poisson structure. In order to prove that { , } defines a Poisson structure on V , we briefly introduce the notion of a NambuPoisson structure,
which generalizes the notion of a Poisson structure.
Definition 8.30. For q 2, a qvector field { , . . . , } on a finitedimensional vector
space V is called a NambuPoisson structure of order q on V if it satisfies the socalled fundamental identity:
{F1 , . . . , Fq1 , {G1 , . . . , Gq }} =
q
{G1 , . . . , Gi1 , {F1 , . . . , Fq1 , Gi }, Gi+1 , . . . , Gq } ,
(8.26)
i=1
for all functions F1 , . . . , Fq1 , G1 , . . . , Gq F (V ).
Remark 8.31. The definition is easily generalized to arbitrary (real or complex) manifolds by demanding that the qvector field satisfies the fundamental identity for local functions, which yields the notion of a NambuPoisson manifold. Replacing the
algebra of functions F (V ) by an arbitrary commutative associative algebra, leads
to the abstract notion of a NambuPoisson algebra, which generalizes the notion of
a Poisson algebra, as given in Definition 1.1.
Example 8.32. Specializing Definition 8.30 to the case of q = 2, we find that a
NambuPoisson structure of order 2 is a bivector field { , } which satisfies
{F, {G1 , G2 }} = {{F, G1 } , G2 } + {G1 , {F, G2 }} ,
8.3 NambuPoisson Structures
223
for all F, G1 , G2 F (V ), which is precisely the Jacobi identity for { , }. Therefore,
a NambuPoisson structure of order 2 is the same as a Poisson structure. In this
sense, NambuPoisson structures are a generalization of Poisson structures.
Given a NambuPoisson structure { , . . . , } of order q 2 on V , one can associate
to q 1 elements F1 , . . . , Fq1 of F (V ), a vector field XF1 ,...,Fq1 on V by defining,
for G F (V ):
XF1 ,...,Fq1 [G] := {G, F1 , . . . , Fq1 } .
It is called the Hamiltonian vector field associated to F1 , . . . , Fq1 . The fundamental
identity (8.26) says that all Hamiltonian vector fields XF1 ,...,Fq1 are derivations of
(F (V ), { , . . . , }).
We show in the following proposition how NambuPoisson structures of order q 1 can be constructed from NambuPoisson structures of order q.
Proposition 8.33. Let { , . . . , } be a NambuPoisson structure of order q 3 on a
finitedimensional vector space V . Given H F (V ), define a (q 1)vector field
on V by
{ , . . . , }H := { , . . . , , H} .
Then { , . . . , }H is a NambuPoisson structure of order q 1 on V .
Proof. The fundamental identity (8.26) for { , . . . , }H takes the form
{F1 , . . . , Fq2 , {G1 , . . . , Gq1 }H }H =
q1
{G1 , . . . , Gi1 , {F1 , . . . , Fq2 , Gi }H , Gi+1 , . . . , Gq1 }H ,
(8.27)
i=1
for all F1 , . . . , Fq2 , G1 , . . . , Gq1 F (V ). To prove (8.27), substitute Fq1 = H and
Gq = H in the fundamental identity (8.26), to find
{F1 , . . . , Fq2 , H, {G1 , . . . , Gq1 , H}} =
q1
{G1 , . . . , Gi1 , {F1 , . . . , Fq2 , H, Gi }, Gi+1 , . . . , Gq1 , H}
(8.28)
i=1
+ {G1 , . . . , Gq1 , {F1 , . . . , Fq2 , H, H}} ,
which is precisely (8.27), since the skewsymmetry of { , . . . , } implies that the last
term in (8.28) is equal to zero.
We show in the following proposition that for dvector fields on a ddimensional
vector space V , the fundamental identity is automatically satisfied.
Proposition 8.34. Let V be a ddimensional vector space, with d 2. Every dvector field on V is a NambuPoisson structure of order d on V .
Proof. Let { , . . . , } be a dvector field on V . We need to prove that { , . . . , } satisfies the fundamental identity (8.26). To do this, we consider the multilinear map
N : F (V )2d1 F (V ), which is defined by
224
8 Higher Degree Poisson Structures
N (F1 , . . . , Fd1 , G1 , . . . , Gd )
:= {F1 , . . . , Fd1 , {G1 , . . . , Gd }}
d
{G1 , . . . , Gi1 , {F1 , . . . , Fd1 , Gi } , Gi+1 , . . . , Gd } ,
i=1
for F1 , . . . , Fd1 , G1 , . . . , Gd F (V ). Obviously, the vanishing of N is equivalent to
the fundamental identity (8.26) for { , . . . , }. Moreover, since for all F1 , . . . , Fd1
F (V ), the map
F (V )d F (V )
(G1 , . . . , Gd )
N (F1 , . . . , Fd1 , G1 , . . . , Gd )
(8.29)
is a skewsymmetric dderivation of F (V ), as is easily checked directly from the
definition of N , the vanishing of N is equivalent to
N (F1 , . . . , Fd1 , x1 , . . . , xd ) = 0 ,
(8.30)
for all F1 , . . . , Fd1 F (V ), where (x1 , . . . , xd ) is an arbitrary system of linear coordinates on V . If we denote := {x1 , . . . , xd }, then the dvector field { , . . . , } can
be written, according to (8.1), as
,
x1
xd
{ , . . . , } =
(8.31)
so that (8.30), which we need to prove, takes the form
{F1 , . . . , Fd1 , } =
xi {F1 , . . . , Fd1 , xi } .
i=1
In order to prove (8.32) we compute its lefthand side from (8.31):
F1
x1
.
{F1 , . . . , Fd1 , } = ..
F
1
xd
Fd1
x1
..
.
Fd1
xd
x1
..
.
xd
dF1 dFd1 d
dx1 dxd
d dF1 dFd1
= (1)d1
dx1 dxd
i dxd
d d Ci dx1 dx
= (1)d1
dx1 dxd
i=1
=
(8.32)
8.3 NambuPoisson Structures
225
d
= (1)di
i=1
where
Ci
,
xi
,...,
,...,
Ci := (dF1 dFd1 )
x1
xi
xd
= (dF1 dFd1 dxi )
,...,
,...,
,
x1
xi
xd xi
di
,...,
= (1) (dF1 dFd1 dxi )
x1
xd
= (1)di {F1 , . . . , Fd1 , xi } .
We thus have obtained (8.32), so that { , . . . , } satisfies the fundamental identity
and is a NambuPoisson structure on V .
Combining Propositions 8.33 and 8.34, we find the following proposition, which
yields the Poisson structures of rank 2, which we introduced in (8.24).
Proposition 8.35. Let V be a ddimensional vector space and let , 1 , . . . , d2
F (V ) be d 1 functions on V . A Poisson structure is defined on V by
(F, G, 1 , . . . , d2 )
,
{F, G} :=
(8.33)
(x1 , . . . , xd )
for all F, G F (V ), where (x1 , . . . , xd ) is an arbitrary system of linear coordinates
on V . The given functions 1 , . . . , d2 are Casimir functions for and the rank
of is 2 at all points of V , except at the zeros of and at the points m V where
dm 1 , . . . , dm d2 are linearly dependent (in those points the rank is zero).
Proof. Repeated use of Proposition 8.33 on the dvector field (NambuPoisson
structure of order d) { , . . . , } := x x on V , yields a NambuPoisson
1
d
structure of order 2, which is given by
(F, G, 1 , . . . , d2 )
.
{F, G} := {F, G, 1 , . . . , d2 } =
(8.34)
(x1 , . . . , xd )
Since is a NambuPoisson structure of order 2, it is a Poisson structure (see Example 8.32). The skewsymmetry of { , . . . , } implies that {F, i } = 0 for all F F (V )
and 1 i d 2, so that each one of 1 , . . . , d2 is a Casimir function of .
If m V with (m) = 0 or dm 1 dm d2 = 0, then it is easy to see from (8.25)
that {F, G} (m) = 0 for all F, G F (V ), so that the Poisson structure vanishes at m.
Suppose now that m is a point of V for which (m) = 0 and dm 1 dm d2 = 0.
Then
there exist i, j such that dm xi dm x j dm 1 dm d2 = 0, so that
xi , x j (m) = 0. This shows that Rkm 2; since we have d 2 independent
Casimir functions at m, the rank is at most 2 at m, so it is precisely 2.
226
8 Higher Degree Poisson Structures
It follows at once from Eq. (8.33) that, if the functions , 1 , . . . , d2 are polynomial functions on V , then the Poisson structure is a polynomial Poisson structure on V . If, moreover, for some fixed weights 1 , . . . , d of the linear coordinates
x1 , . . . , xd on V , the functions , 1 , . . . , d2 are weight homogeneous, then it also
follows from this formula that the Poisson structure is a weight homogeneous
Poisson structure on V , of weight
d2
i=1
i=1
( ) = ( ) + (i ) i .
Remark 8.36. In the case of a ddimensional manifold M, every dvector field on M
defines a NambuPoisson structure of order d and hence, upon fixing d 2 functions
1 , . . . , d2 on M, a Poisson structure of rank two on M (assuming that there is at
least one point where the differentials of these functions are independent). In local
coordinates x1 , . . . , xd , a dvector field { , . . . , } takes the form
{ , . . . , } =
,
x1
xd
and the formula for the Poisson structure is given by
(F, G, 1 , . . . , d2 )
,
{F, G} :=
(x1 , . . . , xd )
so it is in local coordinates formally given by the same expression as in (8.33).
A coordinatefree description of the Poisson structure can be given when M is an
orientable manifold, so that there exists a volume form on M. Then a Nambu
Poisson structure { , . . . , } of order d is defined on M by letting
{F1 , . . . , Fd } :=
dF1 dFd
,
for all (local) functions F1 , . . . , Fd on M. In particular, if we fix d 1 functions
, 1 , . . . , d2 on M and a volume form , then a Poisson structure on M is
defined by
dF dG d1 dd2
{F, G} :=
,
for all (local) functions F, G on M. The functions 1 , . . . , d2 are Casimir functions
of and the rank of is two at all points of M, except at the zeros of and at
the points where the differentials of the latter Casimir functions are dependent. In
this case, a simple proof of the fact that is a Poisson structure can be given by
extending, at points where the rank is two, the Casimir functions 1 , . . . , d2 to a
system of local coordinates.
8.4 Transverse Poisson Structures to Adjoint Orbits
227
8.4 Transverse Poisson Structures to Adjoint Orbits
In this section, we show that, in terms of a system of natural coordinates which we
will construct, the transverse Poisson structure to a nilpotent orbit in a semisimple
Lie algebra is weight homogeneous (of weight 2).
8.4.1 Striples in SemiSimple Lie Algebras
In this section, we recall a few facts which we will use about complex semisimple
Lie algebra (see [99, 190] for details and proofs). Let (g, [ , ]) be a semisimple Lie
algebra over F = C; we denote the algebra of polynomial functions or of holomorphic functions on g by F (g). The Killing form of g is the symmetric bilinear form
on g, defined for all x, y g by
x  y := Trace (adx ady ) ,
where we recall that adx : g g is the linear map, defined by adx z := [x, z], for
all z g. For a subspace n of g, the orthogonal complement of n in g with respect to
the Killing form is denoted by n . For x g, we denote g(x) := {y g  [x, y] = 0},
the centralizer of x in g. An element x g is said to be nilpotent if the endomorphism
adx of g is nilpotent, i.e., if there exists a k N, such that adkx = 0. According to the
JacobsonMorozov theorem, if e g is nilpotent, then there exist h, f g such that
[h, e] = 2e ,
[h, f ] = 2 f ,
[e, f ] = h .
(8.35)
Every nonzero triple (h, e, f ), satisfying (8.35), is called an Striple of g. In view
of (8.35), the vectors h, e and f generate a Lie subalgebra of g, isomorphic to sl2 (C),
which we denote by s. It leads to two decompositions of g:
(1) A decomposition of g into eigenspaces relative to adh , making g into a graded
Lie algebra. Since each eigenvalue of adh is an integer, we have
g=
gi ,
iZ
where gi is the eigenspace of adh which corresponds to the eigenvalue i. For example, e g2 and f g2 . For x gi and y g j , we have an inclusion adx ady (gk )
gi+ j+k , for every k, so that
gi  g j = 0 if i + j = 0 .
(8.36)
(2) A decomposition of g as an smodule:
g=
j=1
Vn j ,
(8.37)
228
8 Higher Degree Poisson Structures
where each Vn j is a simple smodule, with adh eigenvalues (also called adh weights)
n j , n j 2, n j 4, . . . , n j , so that dimVn j = n j + 1. It follows that
s
n j = dim g s .
(8.38)
j=1
Moreover, s = dim g(e), since the centralizer g(e) of e is generated by the vectors of
highest adh weight of each Vn j .
8.4.2 Weights Associated to an STriple
We show in this section that every Striple (h, e, f ) of g leads to a system of linear
coordinates on g, centered at e, such that the LiePoisson structure on g is weight
homogeneous with respect to some weight vector , which has a natural Lie algebraic interpretation: up to a shift of 2, it consists of the weights of g as an smodule.
For every t C , choose a complex number t such that exp(t ) = t. Such a
choice cannot be made in such a way that t depends in a continuous way on t, but
that will be irrelevant for what follows. We denote the adjoint group of g by G and
we consider the map
: C G
t
exp(t h) .
For a fixed t, the map Ad (t) : g g is welldefined (independent of the choice made
for t ) and it acts on each of the eigenspaces gi of adh as a homothecy with ratio t i ,
Ad (t) x = t i x ,
(8.39)
for all x in gi . Indeed, for all t C and x gi ,
Ad (t) x = Adexp (t h) x = exp(adt h )x = exp(t adh )x
(it )k x = exp(it )x = t i x ;
k=0
also, replacing in this computation t by t + 2 n 1, where n Z, clearly leads
to the same result, which shows that the above choice of t is irrelevant. It follows
that t
Ad (t) is a (linear) action of C on g. Since e g2 , the action of C on g,
defined for t C and for y g by t y := t 2 Ad (t) y, fixes e; the action is known
as Slodowys action. Let us denote for x g by Fx the affine function on g, defined
by
Fx (z) := z e  x ,
for all z g. We consider a basis (x1 , . . . , xd ) of g, where each xk belongs to some
eigenspace gik , and we denote the function Fxk which corresponds to xk simply by Fk .
For k = 1, . . . , d, the functions Fk define linear coordinates on g, centered at e. For
8.4 Transverse Poisson Structures to Adjoint Orbits
229
x gi , (8.39) and Adinvariance of the Killing form imply that
(t Fx ) (z) = t 1 z e  x = t 2 Ad (t 1 ) (z e)  x
= t 2 z e  Ad (t) x = t 2 z e t i x = t i2 Fx (z) .
We claim that, if we define for x gi the weight (Fx ) of Fx to be equal to i + 2,
then the LiePoisson structure { , }g , which is an affine Poisson structure in terms
of the coordinates F1 , . . . , Fd , is weight homogeneous of degree 2 with respect to
the weight vector ( (F1 ), . . . , (Fd )) = (i1 + 2, . . . , id + 2). To prove this, first recall
from Section 7.2 that the LiePoisson structure { , }g on g is given, for F, G
F (g), by
{F, G}g (z) = z  [z F, z G] ,
(8.40)
where we recall from (7.11) that z F, the gradient of F at z (with respect to  )
is defined, for F F (g) and z g by
z F  u = dz F, u ,
for all u g. Since z Fx = x, for all x, z g, we have, for all x, y, z g,
Fx , Fy g (z) = z  [x, y] = F[x,y] (z) + e  [x, y] .
(8.41)
If x gi and y g j , with i + j = 2, then (8.36) implies that e  [x, y] = 0, so that
( Fx , Fy g ) (Fx ) (Fy ) = (F[x,y] ) (Fx ) (Fy )
= i + j + 2 (i + 2) ( j + 2) = 2 .
This result extends to the case i + j = 2, since then (F[x,y] ) = i + j + 2 = 0, which
is the weight of the constant function e  [x, y], and then ({Fx , Fy }g ) (Fx )
(Fy ) = (i + 2) ( j + 2) = 2. This shows, according to (8.11), that the Lie
Poisson structure on g is weight homogeneous of degree 2 with respect to the
weight vector (i1 + 2, . . . , id + 2). Notice that this weight vector contains, in general,
both positive and negative integers.
8.4.3 Transverse Poisson Structures to Nilpotent Orbits
In this section we consider the transverse Poisson structure to a nilpotent orbit O
in a semisimple Lie algebra g. Such an orbit is by definition the orbit G e of a
nilpotent element e g under the adjoint action of G (the adjoint group of g) on g;
in this case, all elements of O are nilpotent. Since O is a symplectic leaf of the
LiePoisson structure on g (see Section 7.3.1), the transverse Poisson structure at
e O is independent of the chosen point on the orbit; we speak of the transverse
structure to the orbit. Recall from Theorem 5.28 that the transverse Poisson struc
230
8 Higher Degree Poisson Structures
ture at e is the reduced Poisson structure on an arbitrary PoissonDirac manifold
which is transversal to the symplectic leaf O at e. We show in the following proposition that, in terms of coordinates which we will construct, the transverse Poisson
structure to a nilpotent orbit is polynomial and weight homogeneous of weight 2,
just like the LiePoisson structure on g. Recall from Section 8.4.2 that the weight
homogeneity of the LiePoisson structure on g was determined by the choice of an
Striple (h, e, f ) of g which contains e, that this Striple generates a Lie subalgebra
s of g, isomorphic to sl2 (C) and that up to a shift of 2, the weight vector consists of
the weights of g, as an smodule.
Proposition 8.37. Let O = G e be a nilpotent orbit in a semisimple Lie algebra
g and let (h, e, f ) be an Striple of g which contains e. If n is a complementary
subspace to g(e) in g, which is adh invariant, [h, n] n, then
(1)
(2)
(3)
N := e + n is transverse to O at e;
N is a PoissonDirac submanifold of g;
The transverse Poisson structure on N is a polynomial Poisson structure which
is weight homogeneous of weight 2, with respect to the weight vector (n1 +
2, . . . , ns + 2).
The integers n1 , . . . , ns are the highest weights of g as an smodule, where s denotes
the Lie subalgebra of g, generated by h, e and f .
Proof. First, let us point out that an adh invariant complementary subspace n to g(e)
exists, because the centralizer g(e) is generated by the highest weight vectors of
each Vn j in the decomposition (8.37) of g. Also, it is clear that
Te g = Te O Te N ,
since under the natural isomorphism Te g g, the subspace Te O = Te (G e) corresponds to adg e = g(e) and Te N corresponds to n . In particular, N is transverse
to O at e, which is (1).
In order to show (2) and (3), we use a natural basis of g = g(e) n. As we already
pointed out, the centralizer g(e) of e in g admits a basis (x1 , . . . , xs ), consisting of the
highest weight vectors of Vn1 , . . . ,Vns in the decomposition (8.37) of g; recall also
that the adh weight of x j is n j . Since n, which has dimension 2r := dim O, is adh invariant, we can also choose a basis (xs+1 , . . . , xs+2r ) for n, where each vector xs+
is an eigenvector of adh ; we denote the adh weight of xs+ by , where 1 2r.
The basis (x1 , . . . , xs+2r ) of g leads to linear coordinates F1 , . . . , Fs+2r on g, centered
at e, defined by Fk (z) := z e  xk , for k = 1, . . . , s + 2r and z g. Since z Fk = xk
for k = 1, . . . , s + 2r, it follows from (8.40) that the Poisson matrix of { , }g at z g
is given by
A(z) B(z)
,
=
Fi , Fj g (z)
B(z) D(z)
1i, js+2r
where
8.4 Transverse Poisson Structures to Adjoint Orbits
231
Ai j (z) = z  [xi , x j ]
1 i, j s ;
Bim (z) = z  [xi , xs+m ]
1 i s,
Dm (z) = z  [xs+ , xs+m ]
1 , m 2r .
1 m 2r ;
At e, both matrices A and B vanish, so the rank of D(e) is the rank of { , }g at e,
which equals dim O = 2r. It follows that D(e) is invertible and hence that D(z) is invertible for z in a neighborhood V N of e in g. According to Proposition 5.27, V is
a PoissonDirac submanifold of g and the Poisson matrix XN of the reduced Poisson
structure { , }N is given, for n N, by
XN (n) = A(n) + B(n)D(n)1 B(n) .
(8.42)
The functions F1 , . . . , Fs , restricted to N, will be denoted by q1 , . . . , qs ; they yield
linear coordinates (centered at e) for N and all other functions Fk vanish on N, so
that N can also be defined as the submanifold defined by Fs+1 = = Fs+2r = 0.
Since the weight of Fi is ni + 2, we define the weight of qi to be (qi ) := ni + 2, for
i = 1, . . . , s. Notice that these weights are all positive.
We prove that the determinant of D is a nonzero constant function on N, which
shows that the transverse Poisson structure { , }N is polynomial, in view of (8.42).
Since the restriction of the polynomial function det(D) on g to N is obtained by
setting Fs+1 = = Fs+2r = 0 and since the remaining coordinates q1 , . . . , qs have
positive weight, it suffices to show that the restriction of det(D) to N is a weight
homogeneous function of weight 0. A typical term of det(D) is, up to a sign, of
the form D1,1 . . . D2r,2r , where {1 , . . . , 2r } = {1, . . . , 2r}, so that its weight is
given by
2r
2r
2r
=1
=1
=1
(D, ) = ( + + 2) = 2 ( + 1) .
2r
=1
+ si=1 ni
is the sum of all adh weights, it is zero, which yields in
Since
combination with (8.38)
2r
=1
i=1
( + 1) = 2r ni = 2r (dim g s) = 0 .
This shows that every term of det(D), restricted to N, is of weight zero, hence det(D)
is constant; since det(D(e)) = 0, this constant is different from zero and the transverse Poisson structure is polynomial.
To finish, we show that the Poisson structure { , }N is weight homogeneous of
weight 2 with respect to = (n1 + 2, . . . , ns + 2). According to (8.42) we need to
show that for all 1 i, j s the functions Ai j and (BD1 B )i j are weight homogeneous of degree (qi ) + (q j ) 2 = ni + n j + 2. For Ai j this is clear, since A is part
of the Poisson matrix of the LiePoisson structure on g, which we have seen to be
weight homogeneous of weight 2. Similarly, we have that (Bim ) = ni + m + 2.
Since
1
(Bim D1
m B j ) = ni + n j + m + + 4 + (Dm ) ,
232
8 Higher Degree Poisson Structures
it means that we need to show that
(D1
m ) = m 2 .
1 is of the form
A typical term of the element D1
m of the inverse matrix D
where D = D1 1 . . . D2r1 2r1 , with
(8.43)
D
det(D) ,
{1 , . . . , 2r1 } = {1, . . . , 2r} \ {} and {1 , . . . , 2r1 } = {1, . . . , 2r} \ {m} .
Since det(D) has weight zero, we need to show that D has weight m 2,
which is done in precisely the same way as the above proof that det(D) has weight
zero. This gives us (8.43).
8.5 Notes
For higher order Poisson structures, starting with quadratic Poisson structures, there
is no general theory and there is no immediate interpretation, as there is in the
case of constant Poisson structures (in terms of bivectors) and in the case of linear Poisson structures (in terms of Lie brackets). For a few subclasses of quadratic
Poisson structures there is, however, a quite general theory and an interpretation.
Namely, multiplicative quadratic Poisson structures are closely related to Poisson
Lie groups: the datum of a skewsymmetric rmatrix on an associative algebra leads
to multiplicative quadratic Poisson structure, which makes the set of invertible elements into a PoissonLie group; see Section 11.1.6 for details. It also leads to a
cubic Poisson structure, whose virtue seems at this point still very mysterious. Poisson structures coming from rmatrices or Rmatrices are the subject of Chapter 10;
see Section 10.3.
See DufourZung [63] for an extensive account on NambuPoisson structures.
Transverse Poisson structures to general adjoint orbits in a semisimple Lie algebra, with special emphasis to transverse Poisson structures to subregular orbits, are
studied in DamianouSabourinVanhaecke [56].
Chapter 9
Poisson Structures in Dimensions Two and Three
Lowdimensional Poisson manifolds are often used as toy models, to obtain a better
understanding of the theory of Poisson manifolds, as well as to illustrate their unexpected complexity. In the twodimensional case, for example, describing all Poisson
structures on the affine space F2 is deceivingly simple, because the Jacobi identity is
satisfied for all bivector fields; however, their local classification is nontrivial, and
has up to now only been accomplished under quite strong regularity assumptions
on the singular locus of the Poisson structure, which can be identified with the zero
locus of a local function on the manifold. As a result, the study of Poisson structures
in two dimensions already takes us to the nontrivial singularity theory of functions
of two variables!
Dimension three is the smallest dimension in which the Jacobi identity for a
bivector field is not always satisfied. In this dimension, the Jacobi identity can be
stated as an integrability condition of a distribution, which eventually leads to the
symplectic foliation, or as the integrability condition of a differential oneform, dual
to the Poisson structure with respect to a volume form (assuming that the manifold
is orientable). Despite the extra complexity which comes from the extra dimension,
Poisson structures in dimension three have one key property in common with Poisson structures in dimension two: their rank is equal to two (except when the Poisson
structure is trivial). Many results about threedimensional Poisson manifolds are essentially true because the Poisson structure is of rank two. A key example of such a
result is the characterization of a regular Poisson structure of rank two in terms of
its symplectic foliation.
Poisson structures in dimensions two and three are presented in different sections
(Sections 9.1 and 9.2). In both cases, we pay special attention to the global/local and
geometrical/algebraic aspects of these Poisson structures and we prove at the end of
each section a (partial) classification result.
Unless otherwise stated, F is an arbitrary field of characteristic zero.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 9, SpringerVerlag Berlin Heidelberg 2013
233
234
9 Poisson Structures in Dimensions Two and Three
9.1 Poisson Structures in Dimension Two
In this section, we study Poisson structures on manifolds or affine varieties of dimension two. Since a Poisson structure is a bivector field or a biderivation, the smallest
dimension which one can consider for studying nontrivial Poisson structures is dimension two. In this case, the Jacobi identity is always satisfied, so that every bivector field or skewsymmetric biderivation is a Poisson structure. We present Poisson
structures in dimension two first from the global point of view, then from the local
point of view and finally we discuss their formal classification.
9.1.1 Global Point of View
First of all, let M be a manifold (real or complex, depending on whether F = R
or F = C) of dimension two and let F (M) be the algebra of smooth or holomorphic
functions on M. According to Proposition 3.5, every bivector field on M is a Poisson structure because [ , ]S is a trivector field on M, hence is zero. We conclude
that the space of all Poisson structures on M coincides with the vector space X2 (M)
of all bivector fields on M. In particular, two Poisson structures on M are always
compatible (see Section 3.3.2). For every point m of M, the rank of at m is zero or
two. Discarding the case of the trivial Poisson structure, it follows that the singular
locus of (M, ) coincides with the set of points of M where vanishes.
Next, let us consider a complex affine surface M Cd of dimension two,
equipped with its algebra F (M) = C[x1 , . . . , xd ]/I of regular functions. We show
that every skewsymmetric biderivation of F (M) is a Poisson structure of M.
To do this, we prove that every skewsymmetric triderivation of F (M) is zero;
since [ , ]S is a skewsymmetric triderivation of F (M), it follows that is a Poisson structure on M. Let us denote by M sm the set of all smooth points of M. Let P
be a skewsymmetric triderivation on M and suppose that P is different from zero at
some point m M. Since P can be written as
P=
1i< j<kd
Pi jk
,
xi x j xk
where Pi jk := P[xi , x j , xk ], this means that Pi jk (m) = 0, for some i, j, k. It follows
that P is different from zero in a neighborhood of m in M. Since M sm is dense
in M, this means that P is nonzero in the neighborhood of some smooth point of M.
However, as we have seen in Section 2.3.2, the set of smooth points of M carries
a natural structure of a complex manifold, and every skewsymmetric triderivation
of F (M) corresponds on it to a trivector field. Since M sm is a twodimensional
manifold, every trivector field on M sm is zero, hence P is zero on M sm , and we
arrive at a contradiction. This shows that P vanishes at every point of M, hence
that every skewsymmetric biderivation of F (M) is a Poisson structure on M. As
in the case of twodimensional Poisson manifolds, the rank of can only take the
9.1 Poisson Structures in Dimension Two
235
Fig. 9.1 The singular locus of the Poisson structure =
curve defined by = 0.
y is the zero locus of , i.e., the
values zero and two at the points of M, and the singular locus of (M, ) coincides
with the set of points where vanishes (assuming that is nontrivial).
Finally, consider F2 , with standard coordinates (x, y), equipped with its algebra
of functions F (F2 ); recall that F (F2 ) denotes the algebra of polynomial functions
on F2 if F is an arbitrary field, but can also be the algebra of smooth or holomorphic
functions on F2 if F = R or F = C. Every skewsymmetric biderivation on F (F2 )
or bivector field on F2 (i.e., every Poisson structure on F2 ) is of the form
,
x y
where F (F2 ) ,
(9.1)
and by the above, every such is a Poisson structure on F2 . Hence, there is a onetoone correspondence between the following three objects: the algebra F (F2 ) of
all functions on F2 , the vector space X2 (F2 ) of all bivector fields on F2 and the set
of all Poisson structures on F2 . The singular locus of coincides with the zero locus
of , which is the curve { = 0} F2 . See Fig. 9.1.
9.1.2 Local Point of View
Let M be a (real or complex) manifold of dimension two and let F (M) denote its
algebra of smooth or holomorphic functions. Let (U, (x, y)) be a coordinate chart
of M. According to (1.23), every bivector field on U can be written as
,
x y
where = [x, y] F (U) .
(9.2)
We have seen in Section 9.1.1 that every bivector field on U is a Poisson structure.
The rank of is two at all points m U where (m) = 0 and is zero at all other
points of U.
236
9 Poisson Structures in Dimensions Two and Three
Let us show that, although the Jacobi identity is trivially satisfied in dimension
two, and although all Poisson structures in dimension two are locally given by the
simple formula (9.2), two Poisson structures in dimension two may, both from the
algebraic and geometric points of view, be very different in the neighborhood of
their singularities.
To illustrate this, let (M, ) be a twodimensional Poisson manifold and let m be
an arbitrary point of M. We write in a neighborhood of m in M as
,
x y
where x, y are local coordinates, vanishing at m, and where := [x, y]. We distinguish the following three cases:
(1) If (m) = 0, which means that the rank of at m is two, then there exists,
according to Weinsteins splitting theorem (Theorem 1.25), a coordinate neighborhood U of m in M, with coordinates x, y, centered at m, such that, on U,
.
x y
(2) If (m) = 0 but dm = 0, Example 7.27 tells us that is linearizable at m,
i.e., there exists a coordinate chart for M, still denoted by (U, (x, y)) and centered
at m, such that
.
=x
x y
(3) If (m) = 0 and dm = 0, then the zero locus of the function has a singularity at m. Let us consider the case of quadratic Poisson structures on C2 . Two
examples are given by
1 := (x2 y2 )
x y
and 2 := x2
.
x y
Since the singular locus of 1 consists of the pair of lines y = x, while the singular
locus of 2 consists of the (double) line x2 = 0, it is clear that these Poisson structures are (even locally, in the neighborhood of a point of their singular locus) not
isomorphic. This simple example illustrates the fact that there are at least as many
nonisomorphic Poisson structures on the plane, as there are nonisomorphic algebraic curves in the plane. In fact there are more. In order to see this, consider the
following two Poisson structures:
1 := (x2 y2 )
,
x y
and 3 := 2(x2 y2 )
.
x y
Since they are proportional, they have the same zero locus, but they are not isomorphic as Poisson structures, a fact which can easily be checked by using Proposition 8.8.
9.1 Poisson Structures in Dimension Two
237
As we will see in the next section, there is a partial local classification of Poisson
structures on C2 (or R2 ), under the strong hypothesis that the singular locus of the
Poisson structure admits a simple singularity (at the origin).
9.1.3 Classification in Dimension Two
In this section, we consider the classification problem for Poisson structures in dimension two. Since the Jacobi identity is trivially satisfied in dimension two, the
classification amounts to the simpler problem of classifying bivector fields in dimension two, yet only partial results are known. The classification theorem which
we will present below has the following restrictions:
(1)
(2)
(3)
The Poisson structures and the coordinate transformations which are considered are formal;
The Poisson structures which are considered, are assumed to have a singular
locus which admits a simple singularity at the origin;
The ground field is C.
As is made explicit in the notes at the end of the chapter, some of these restrictions
can be weakened.
We first give an overview of the results and the ideas which will be presented in
this section; full proofs will be given in Subsections A, B, C and D below. We begin
by giving the definition of a formal Poisson structure on C2 .
Definition 9.1. A formal Poisson structure on C2 is a Poisson bracket on the algebra
of all formal power series in two variables C[[x, y]].
As we show in Subsection 9.1.3.1, every formal Poisson structure on C2 is of the
form x y , with C[[x, y]], and every such biderivation is a formal Poisson
structure on C2 .
By definition, a formal coordinate transformation of C2 is an algebra isomorphism : C[[x, y]] C[[x, y]] such that the formal power series (x) and (y) have
no constant term. Two formal power series , C[[x, y]] are said to be equivalent
if there exists a formal coordinate transformation of C2 , such that ( ) = . The
following formal version of the inverse function theorem holds: if is an algebra
homomorphism such that the formal power series (x) and (y) have no constant
term and whose Jacobian
(x) (x)
x
y
Jac := det
(y) (y)
x
y
is invertible, i.e., its constant term is different from zero, then is a formal coordinate transformation.
238
9 Poisson Structures in Dimensions Two and Three
Let be a formal coordinate transformation of C2 . Given a formal Poisson structure on C2 , there exists a unique formal Poisson structure on C2 , denoted ,
such that : C[[x, y]] C[[x, y]]
: (C[[x, y]], ) (C[[x, y]], )
is a morphism (hence isomorphism) of Poisson algebras. Explicitly, is given
by
( ) [ , ] := 1 ( [ ( ), ( )]) ,
for all , C[[x, y]]. Since
= ( )[x, y]
,
x y
it follows that, if we write in the form = x y , then takes the form
= 1 ( Jac( ))
.
x y
(9.3)
It leads to the notion of equivalence of formal Poisson structures (on C2 ).
Definition 9.2. Two formal Poisson structures 1 and 2 on C2 are said to be formally isomorphic if there exists a formal coordinate transformation of C2 , such
that 2 = 1 .
We consider formal Poisson structures on C2 , with a simple singularity at the origin,
i.e., formal Poisson structures of the form x y , where C[[x, y]] is a formal
power series, without constant term, which admits a simple singularity at the origin
(see Subsection 9.1.3.1 below for a precise definition of this notion). According to
a theorem by Arnold, such a formal power series is equivalent to a weight homogeneous polynomial C[x, y] (see Theorem 9.7 below). Combined with (9.3),
this implies that every formal Poisson structure on C2 is formally isomorphic to a
Poisson structure of the form
,
(9.4)
x y
where C[x, y] is a weight homogeneous polynomial and C[[x, y]] is a formal power series whose constant term is different from zero. The next step in the
classification is the following proposition, which states that the formal power series in (9.4) can be replaced, modulo a nonzero additive constant, by a weight
homogeneous polynomial.
Proposition 9.3. Let = x y be a formal Poisson structure on C2 , where
C[x, y] is a nonconstant polynomial, which is weight homogeneous with respect
to the weight vector = (1 , 2 ), and C[[x, y]] is a formal power series, with
constant term 1. Then is formally isomorphic to a Poisson structure of the form
9.1 Poisson Structures in Dimension Two
239
,
x y
(1 + )
where C[x, y] is a weight homogeneous polynomial of weight ( ) 1 2 .
See Subsection 9.1.3.3 below for a proof of this proposition. It follows that a Poisson
structure which admits a simple singularity at the origin is formally isomorphic, up
to a nonzero multiplicative constant, to a Poisson structure of the form
,
x y
(1 + )
where is one of the weight homogeneous polynomials which appear in Arnolds
classification theorem of formal power series with a simple singularity (Theorem 9.7) and C[x, y] is an arbitrary weight homogeneous polynomial of weight
( ) 1 2 . For each in Arnolds list, one easily determines all possibilities for , which leads to the following classification theorem of formal Poisson
structures on C2 , which admit a simple singularity at the origin.
Theorem 9.4. Let = x y be a formal Poisson structure on C2 , where
C[[x, y]] is a formal power series, which has a simple singularity at the origin. Then, up to a nonzero multiplicative constant, is formally isomorphic to a
Poisson structure of the form x y , where C[x, y] is one the polynomials
which appear in the third column of the following table.
Type
x2 + y2k+1
A2k
k1
A2k1
k1
D2k
k2
D2k+1
k2
x2 + y2k 1 + yk1
x2 y + y2k1 1 + x + yk1
2
x y + y2k (1 + x)
E7
x 3 + y4
3
x + xy3 1 + y2
E8
x 3 + y5
E6
Remark 9.5. The Poisson structures which appear in the classification theorem are
not claimed to be all nonisomorphic. Special care has to be taken with respect to the
phrase up to a nonzero multiplicative constant, which appears in its statement. In
general, a formal Poisson structure on C2 is formally isomorphic to every Poisson
structure of the form , with C , but there are exceptions. We already gave
in Section 9.1.2 an example of two proportional quadratic Poisson structures which
are nonisomorphic. In order to make a more general statement, let be a weight
homogeneous polynomial of C[x, y], with respect to the weight vector (1 , 2 ) and
240
9 Poisson Structures in Dimensions Two and Three
of weight ( ). Let us consider the Poisson structure on C2 corresponding to ,
namely := x y . Define, for arbitrary a C , the formal coordinate transformation of C2 , which amounts to rescaling x and y as follows:
a (x) := a1 x,
a (y) := a2 y .
According to (8.8) and (9.3), we have
(a ) = a1 +2 ( )
.
x y
This formula permits us to conclude that, if ( ) = 1 + 2 , then every Poisson
structure of the form , with C is formally isomorphic to . However, if
( ) = 1 + 2 , then each of the formal coordinate transformations a acts trivially on ; in fact, as can be checked by using Proposition 8.8, the homogeneous
Poisson structures and are, for = 1, nonisomorphic.
9.1.3.1 A Formal Power Series and Simple Singularities
In this subsection, we recall some basic notions about formal power series (in two
variables) and we introduce some concepts which are needed to state the theorem of
Arnold about the classification of formal power series with a simple singularity.
A formal power series C[[x, y]] can be written in a unique way as =
i, jN i j xi y j , so the datum of a family (i j )i, jN in C is equivalent to the datum
of a formal power series in C[[x, y]]. This elementary fact is important when dealing with the question whether certain operations yield welldefined power series: a
formal power series is welldefined if for every i, j N the coefficient ai j is welldefined. For example, the product of two power series is welldefined, because each
term of the product is given by a finite sum. Similarly, if C[[x, y]] is a formal
power series without constant term, then
exp( ) :=
i
iN i!
is a welldefined power series. A formal power series C[[x, y]] can be evaluated
at (x, y) = (0, 0), which yields the constant term of the series, denoted (0, 0), but
can in general not be evaluated at other values of x, y. With the standard addition and
product of power series, it is clear that C[[x, y]] forms an associative algebra, whose
invertible elements are the series for which (0, 0) = 0.
We define the order of a formal power series C[[x, y]] as the minimum of
the total degrees of each of its terms, denoted ord( ). When we are in a weight
homogeneous context, we will also consider the order of a power series with
respect to a weight vector = (1 , 2 ), where 1 and 2 are the weights of the
9.1 Poisson Structures in Dimension Two
241
variables x and y (see Section 8.1.3). Then ord ( ), the order of with respect to
, is by definition the minimum of the weights of each of the terms of .
As we said previously, a formal coordinate transformation of C2 is an algebra isomorphism : C[[x, y]] C[[x, y]] so that the constant terms of (x) and of (y) are
zero. Also, recall that two formal power series , C[[x, y]] are said to be equivalent if there exists a formal coordinate transformation of C2 such that ( ) = .
Notice that if and are two equivalent formal series of C[[x, y]], they have the
same constant term.
Definition 9.6. A formal power series C[[x, y]] satisfying (0, 0) = 0 is said to
have a simple singularity at the origin, if a small enough neighborhood of (for the
Whitney topology, see [17]) intersects only a finite number of orbits associated to
the equivalence of formal power series.
We are now able to state Arnolds classification in the case of formal power series in
C[[x, y]], with a simple singularity. The names of the singularity types are borrowed
from the classification theory of simple Lie algebras; see [147] for an explanation.
Theorem 9.7 (Arnolds theorem). Let C[[x, y]] be a formal power series without constant term, which has a simple singularity at (0, 0). Then is equivalent to
one of the weight homogeneous polynomials which appear in the third column of
the following table.
Type
Ak
k1
x2 + yk+1
Dk
k4
x2 y + yk1
E6
x 3 + y4
E7
x3 + xy3
E8
x 3 + y5
9.1.3.2 B Formal Poisson Structures on C2
We have defined a formal Poisson structure on C2 as a Poisson bracket on the algebra C[[x, y]] of formal power series in two variables. Since a Poisson bracket is
in particular a skewsymmetric biderivation, we first study the algebra of skewsymmetric multiderivations of C[[x, y]]. As we will see, the results are very similar
to the case of the algebra of polynomial or holomorphic functions on C2 .
The formal partial derivatives x and y clearly yield derivations of C[[x, y]].
Also, their wedge,
yields a skewsymmetric biderivation of this algebra.
More general examples are constructed by taking x + y , respectively x y ,
242
9 Poisson Structures in Dimensions Two and Three
where , C[[x, y]] are arbitrary. We show in the following proposition that there
are no other skewsymmetric kderivations of C[[x, y]] (k 1).
Proposition 9.8. The spaces of skewsymmetric multiderivations of C[[x, y]] are
given by
+
 , C[[x, y]] ,
X1 C[[x, y]] =
x
y
 C[[x, y]] ,
X2 C[[x, y]] =
x y
k
X C[[x, y]] = {0}, for all k 3 .
Proof. Let us prove this proposition for the case of the derivations of C[[x, y]],
the case of the skewsymmetric biderivations and skewsymmetric kderivations
of C[[x, y]] (for k 3) being analogous. To do that, we first point out that every derivation W of C[[x, y]] satisfies the following property: if C[[x, y]], then
ord(W [ ]) is either (i.e. W [ ] = 0) or ord(W [ ]) ord( ) 1. This can be
done easily by recursion on ord( ), first in the particular case where C[[x]]
or C[[y]], and secondly for an arbitrary C[[x, y]], using the fact that every
C[[x, y]], with ord( ) 1, can be written in at least one of the following two
forms:
= x1 + 2 , with 1 C[[x, y]], ord(1 ) = ord( ) 1 and 2 C[[y]] ;
= y1 + 2 , with 1 C[[x, y]], ord(1 ) = ord( ) 1 and 2 C[[x]] .
Now, let V X1 (C[[x, y]]) and consider the formal power series := V [x] and
:= V [y]. We prove that V = x + y , by showing that W := V x y
is zero. First, W is a derivation of C[[x, y]] and W [x] = W [y] = 0, hence W vanishes
on every polynomial in C[x, y]. Let now C[[x, y]] be an arbitrary formal power
series. Suppose that W [ ] = 0 and let s := ord(W [ ]). We decompose the formal
power series as follows: = s + s , where s C[x, y] is a polynomial of total
degree at most s + 1 and s C[[x, y]] is a formal power series of order at least s + 2.
Since W vanishes on all polynomials, W [ ] = W [s ] + W [s ] = W [s ], so that
ord(W [ ]) = ord(W [s ]) ord(s ) 1 s + 1 = ord(W [ ]) + 1 ,
which is impossible. Therefore, W [ ] = 0 for all formal power series and the
equality V = x + y follows.
It follows from Proposition 9.8 that every formal Poisson structure on C2 is of
the form:
,
(9.5)
=
x y
for some formal power series C[[x, y]]. Moreover, X3 (C[[x, y]]) = {0}, so that
the Schouten bracket of every skewsymmetric biderivation of C[[x, y]] with itself
9.1 Poisson Structures in Dimension Two
243
is zero, which implies that every such biderivation defines a formal Poisson structure on C2 . Thus, there is a onetoone correspondence between the following three
objects: the algebra C[[x, y]] of all formal power series in two variables, the vector
space X2 (C[[x, y]]) of all skewsymmetric biderivations of C[[x, y]] and the set of
all formal Poisson structures on C2 . Notice that the series in (9.5) is given by
= [x, y], so that a formal Poisson structure on C2 is completely determined by
its value on the variables x and y. Given the above onetoone correspondence, it is
natural to transport the property that a function has a simple singularity, to the case
of a formal Poisson structure on C2 .
Definition 9.9. Let = x y be a formal Poisson structure of C2 , satisfying
(0, 0) = 0. If admits a simple singularity at the origin, then is said to admit a
simple singularity at the origin.
To finish this subsection, we introduce a certain class of formal coordinate transformations of C2 , which are constructed from derivations of C[[x, y]]. Let V be a
derivation of C[[x, y]] and assume that for every C[[x, y]] the series V [ ] is either zero or of order larger than the order of , with respect to a weight vector .
For every k 1 and for every C[[x, y]], the formal power series V k [ ] is either
zero or of order at least k + ord ( ). Therefore, for every i, j N, the coefficient
of xi y j in the infinite sum
exp (V ) ( ) :=
k! V k [ ]
kN
is determined by a finite number of terms, and we have a welldefined map
exp (V ) : C[[x, y]] C[[x, y]] .
Clearly, the formal power series exp (V ) (x) and exp (V ) (y) vanish at the origin. We show that exp (V ) is an algebra homomorphism; i.e., that exp(V )( ) =
exp(V )( ) exp(V )( ), for all , C[[x, y]]. Since V is a derivation of C[[x, y]],
we have for all , C[[x, y]],
i
V ik ( ) V k ( ) ,
V ( ) =
k
k=0
i
so that
i
1 i
1 i
V ik ( ) V k ( )
exp(V )( ) = V ( ) =
k
i!
i!
iN
iN k=0
i
(i k)! k! V ik ( ) V k ( )
iN k=0
j,kN
1
V j ( ) V k ( ) = exp(V )( ) exp(V )( ) .
j! k!
244
9 Poisson Structures in Dimensions Two and Three
The equality which is proven in the above computation is an equality of formal
power series (for a given and ); since at every step, each term of the series is
determined by a finite number of terms, the above manipulations of infinite sums
are valid. Since the Jacobian of exp(V ) equals 1 at (0, 0), exp(V ) defines a formal
coordinate transformation of C2 .
An important class of derivations V which satisfy the above property, consists of
the derivations E , where is a formal power series without constant term and
is a weight vector. In this case, we have for every nonconstant formal power series
that ord ( E [ ]) = ord ( ) + ord ( ), so that exp( E ) is a formal coordinate
transformation of C2 . Fixing a formal power series without constant term, we will
refer to the family (exp (t E ))tC as the formal flow of the derivation E . For
fixed t, the formal coordinate transformation exp (t E ) of C2 is called the formal
flow of E at t.
9.1.3.3 C Proof of Proposition 9.3
The aim of this subsection is to prove Proposition 9.3. Throughout the subsection,
= (1 , 2 ) is a fixed weight vector; weight homogeneity of a polynomial and the
order of a formal power series are to be understood as being with respect to . As
before, E denotes the weighted Euler derivation and we denote   := 1 + 2 .
We start with three technical lemmas.
Lemma 9.10. Let = x y be a formal Poisson structure on C2 , where
is a (nonzero) weight homogeneous polynomial and is a formal power series,
satisfying (0, 0) = 1. Let C[[x, y]] be a formal power series without constant
term and let (t )tC denote the formal flow of E . For every t C, the polynomial
C[x, y] divides the skewsymmetric biderivation (t ) .
Proof. According to (9.3), (t ) = x y , where C[[x, y]] is given by
= t1 ( Jac t ) = t1 ( Jac t ) t1 ( ) .
(9.6)
It therefore suffices to show that divides t1 ( ) for every t C. To do this, we
will show that there exists a oneparameter family (t )tC of formal power series in
C[[x, y]], such that
t1 ( ) = t ,
and such that moreover each coefficient of the power series t depends polynomially
on t C. Suppose that (t )tC is a oneparameter family of formal power series in
C[[x, y]], which satisfies 0 = 1 and t (0, 0) = 0, for every t C. Then we have the
following equivalences:
t1 ( ) = t t = t ( ) ,
9.1 Poisson Structures in Dimension Two
245
1
is a formal power series in C[[x, y]] (depending on t), since
t (t )
t (t )(0, 0) = t (0, 0) = 0 for all t. Our problem is therefore reduced to constructing
a oneparameter family (t )tC of formal power series in C[[x, y]], such that t ( ) =
t , such that each coefficient of the power series t depends polynomially on t, and
such that 0 = 1 and t (0, 0) = 0 for all t C. Under the latter hypothesis on t , we
have the following equivalences:1
1
t = t ( )
t ( ) = 0
t
t
1
2 t ( ) + t ( E [ ]) = 0
t
t
t
1
E [ ] = t
t
t
,
( ) = t1
t
where t :=
where we have used in the second equivalence that (t )tC is the formal flow of
E , and the weighted Euler formula (8.10) in the last equivalence. Our problem is
thereby reduced to solving the linear differential equation
t = ( )t ( )t ,
with initial condition 0 = 1. As in calculus,
t := exp ( )
t
(9.7)
( )d
is the unique solution. It is for each t a welldefined formal power series (since the
formal power series
t
( )d does not have a constant term) whose coefficients
depend polynomially on t, and it satisfies t (0, 0) = 1, for all t C. This proves the
announced existence of the family (t )tC , and hence of the family (t )tC , which
proves that divides (t ) , for all t.
Lemma 9.11. Let =
write as
be a formal Poisson structure on C2 and let us
= s0 +  + + sr +  + ,
where s0 < s1 < < sr Z and where s0 +  , . . . , sr +  are the first r + 1
nonzero weight homogeneous terms of the formal power series , of weight
s0 +  , . . . , sr +   and is a formal power series in C[[x, y]] of order at least
1 For a family (F )
t tC of power series
in C[[x, y]], whose coefficients are polynomial functions
of t, we denote by Ft , respectively by 0t F d , the formal power series obtained by differentiating,
respectively integrating, each coefficient with respect to t.
246
9 Poisson Structures in Dimensions Two and Three
sr +   + 1. Assume that there exists a weight homogeneous derivation V =
V1 x + V2 y of C[[x, y]], such that
LV
s0 + 
x y
= sr + 
.
x y
(9.8)
Let be a formal coordinate transformation of C2 , defined by := exp(V ). Then
= x y , where is of the form
= s0 +  + + sr1 +  + ,
(9.9)
with C[[x, y]] and of order at least sr +   + 1. Stated briefly, eliminates the
term sr +  without modifying the terms of lower weight.
Proof. Let = x y and suppose that there exists a weight homogeneous
derivation V , satisfying (9.8), where s0 +  and sr +  are the parts of of weight
s0 +  , respectively of weight sr +  . Then V1 and V2 are weight homogeneous
polynomials of weight (V1 ) = sr s0 + 1 and (V2 ) = sr s0 + 2 . It implies
that := exp(V ) is a welldefined formal coordinate transformation of C2 ; explicitly, (x) and (y) are of the form
(x) = x + V1 + V1 ,
(9.10)
(y) = y + V2 + V2 ,
where Vi is a formal power series in C[[x, y]] with ord Vi > sr s0 + i for i =
1, 2, so that
V1
V2
=
= sr s0 ,
x
y
V1
V2
> sr s0 , ord
> sr s0 ,
ord
x
y
and
1 + V1 + V1
x
x
Jac =
V2 V2
x + x
.
V2 V2
1+
+
y
y
V1 V1
+
y
y
It follows that the terms of Jac , whose weights are at most sr + , are precisely
the following:
s0 +  , . . . , sr1 +  , sr +  +
V1
V2
s0 +  +
.
x
y s0 + 
9.1 Poisson Structures in Dimension Two
247
Let be the formal power series, defined by = x y . We denote its term
of weight i by i . According to (9.3), ( ) = Jac , with of the form (9.10).
The terms of ( ), and hence of , of weight i, smaller than sr +  , coincide with
the corresponding terms of Jac . Specifically, all are zero except for
s0 +  = s0 +  , . . . , sr1 +  = sr1 +  .
(9.11)
It follows that the term of weight sr +   of ( ) is given by
sr +  + V s0 +  = sr +  + V s0 +  .
Equating these terms with the terms of weight sr +   in Jac , we obtain
V1 V2
+
(9.12)
sr +  +
s0 +  = sr +  + V s0 +  .
x
y
Recall that V satisfies (9.8). Applied to x and y, this condition means that
V1 V2
V s0 + 
+
s0 +  = sr +  ,
x
y
which leads, according to equation (9.12), to sr +  = 0. Combined with (9.11),
this shows that is indeed of the desired form (9.9).
The following lemma is an analog of Eulers weighted formula for bivector fields
on C2 .
Lemma 9.12. Let C[x, y] be a polynomial and let = x y be a Poisson
structure on C2 . If and are weight homogeneous with respect to some weight
vector = (1 , 2 ), then
L E = ( ( )   ( )) .
Proof. Under the stated assumptions on and , we need to show that
L E [x, y] = ( ( )   ( )) .
Using several times the weighted Euler formula (8.10), we compute
L E [x, y] = E [ ] [ E [x], y] [x, E (y)]
= ( ) 1 [ x, y] 2 [x, y]
= ( ( ) 1 2 ) 1 x
+ 2 y
x
y
= ( ( )   ( )) ,
which yields the desired result.
Using the above three lemmas, we prove Proposition 9.3.
(9.13)
248
9 Poisson Structures in Dimensions Two and Three
Proof (of Proposition 9.3). Let = x y be a formal Poisson structure on C2 ,
where C[x, y] and C[[x, y]] are as stated. Since verifies (0, 0) = 1, we
can write = 1 + s + s , where s C[x, y] is a weight homogeneous polynomial
of weight s and s C[[x, y]] is a formal power series of order at least s + 1. Let
us first show that s can be assumed to be at least equal to ( )  . Assume that
s < ( )   and consider the Lie derivative
= ( ( )   s) s
,
Ls E
x y
x y
which follows easily from Lemma 9.12. Letting V := s E /( ( )   s),
we have that (9.8) is satisfied, with r := 1 and s0 +  := and sr +  := s .
We can therefore apply Lemma 9.11, which shows that is formally isomorphic
to ( + ) x y , where is a formal power series, of order at least ( ) + s + 1.
The formal coordinate transformation (9.10) is in this case of the form exp V ,
with V a polynomial times the weighted Euler vector field. Therefore, Lemma 9.10
applies, showing that ( + ) x y is a multiple of . In conclusion, is for
mally isomorphic to (1 + s ) x y , where s is a formal power series of order
at least s + 1. A repeated use of this procedure shows that is formally isomorphic
to (1 + ) x y , where is a formal power series of order at least ( )  .
Let us now show that, using the same technique, the formal power series in
(1 + ) x y can be replaced by its terms of weight ( )  . To do this,
let us write = + + , where is a weight homogeneous polynomial of
weight ( )  , where is a weight homogeneous polynomial of weight at
least ( )   + 1, and where is a formal power series, with ord > ( ).
We first show how to remove the term from , without changing . Setting
r := 2 and s0 +  := and s2 +  := it is verified, as above, that the vector
field V := E /( ( )   ( )) satisfies (9.8), so Lemma 9.11 applies again,
showing as above that is formally isomorphic to (1 + + ) x y , where
has weight ( )   and the order of is at least ( )  .
Repeating this procedure, we can make the order of arbitrarily large. To make
disappear, the procedure needs to be repeated an infinite number of times, i.e.,
we need to consider the algebra homomorphism : C[[x, y]] C[[x, y]] which
is obtained as the composition of an infinite number of the above formal coordinate transformations. However, at each step, the coordinate transformation is of the
form (9.10), with V1 and V2 of higher order than the V1 and V2 which were used in
the previous step, each term in (x) and in (y) will only depend on a finite number of the formal coordinate transformations. Therefore, is formally isomorphic
to (1 + ) x y .
9.1 Poisson Structures in Dimension Two
249
9.1.3.4 D Proof of the Classification Theorem
We are finally ready to give the proof of the classification theorem of formal Poisson
structures on C2 , which admit a simple singularity.
Proof (of Theorem 9.4). Let = x y be a formal Poisson structure on C2 ,
where C[[x, y]] is a formal power series, which has a simple singularity at the
origin. According to Arnolds theorem (Theorem 9.7) there exists a formal coordinate transformation , such that 1 ( ) = , where is a weight homogeneous
polynomial, with respect to some weight vector = (1 , 2 ). According to (9.3),
= x y , where C[[x, y]] is a formal power series, with (0, 0) = 0.
Using Proposition 9.3, we can conclude that, up to a constant, is isomorphic to
(1 + ) x y , where is one of the weight homogeneous polynomials which
appear in Theorem 9.7 and where is a weight homogeneous polynomial of weight
( )  . It suffices now to consider in the list of weight homogeneous polynomials of Theorem 9.7 and construct the most general weight homogeneous polynomial C[x, y] of weight ( )  , which amounts to finding all monomials
xi y j , with i, j N satisfying
1 i + 2 j = ( ) 1 2 .
The result is displayed in Table 9.1.
(9.14)
Table 9.1 The different types of formal simple singularities which appear in Theorem 9.7, with
the corresponding weight vector = (1 , 2 ), the integers 1 and 2 being the weights of the
variables x and y. For each type, equation (9.14) and the monomials corresponding to the solutions
of this equation are given.
Type
A2k
(1 , 2 )
Equation (9.14)
k 1 x2 + y2k+1 (2k + 1, 2) (2k + 1)i + 2 j = 2k 1
A2k1 k 1
D2k
x2 + y2k
(k, 1)
k 2 x2 y + y2k1 (k 1, 1)
D2k+1 k 2
x2 y + y2k
xi y j
ki + j = k 1
yk1
(k 1)i + j = k 1
x, yk1
(2k 1, 2) (2k 1)i + 2 j = 2k 1
E6
x 3 + y4
(4, 3)
4i + 3 j = 5
E7
x3 + xy3
(3, 2)
3i + 2 j = 4
y2
E8
x 3 + y5
(5, 3)
5i + 3 j = 7
250
9 Poisson Structures in Dimensions Two and Three
9.2 Poisson Structures in Dimension Three
We now consider Poisson structures on manifolds of dimension three. In the twodimensional case, we have seen that every bivector field is a Poisson structure,
because the Jacobi identity is trivially satisfied. This is not the case in dimension
three. However, like in the twodimensional case, the rank of a bivector field, in
particular of a Poisson structure, is at most two (hence zero or two), at every point.
Section 9.2.1 is devoted to Poisson structures on F3 , while Section 9.2.2 enlarges
the discussion to real or complex manifolds of dimension three. Quadratic Poisson
structures on C3 will be classified in Section 9.2.3. In Section 9.2.4, we will consider
singular Poisson surfaces in C3 .
9.2.1 Poisson Structures on F3
In this section, we study Poisson structures on the affine space F3 , which is equipped
with its algebra of functions, denoted by F (F3 ). If F is an arbitrary field, this algebra is the algebra of polynomial functions on F3 ; however, if F = R (respectively
F = C), then F (F3 ) may stand either for the algebra of smooth (respectively holomorphic) or polynomial functions on F3 , depending on the context. The standard
coordinates on F3 are denoted by (x, y, z). Every bivector field on F3 decomposes as
+ P2
+ P3
,
(9.15)
= P1
y z
z x
x y
where P1 := {y, z}, P2 := {z, x} and P3 := {x, y}. It follows from item (iii) in Proposition 1.16 that this bivector field satisfies the Jacobi identity, and is therefore a
Poisson structure on F3 , if and only if
{x, {y, z}} + {y, {z, x}} + {z, {x, y}} = 0 ,
which is equivalent to
P3 P2
P1 P3
P2 P1
+ P2
+ P3
=0.
P1
y
z
z
x
x
y
(9.16)
Condition (9.16) can be compactly restated in vector notation. To do this, define a
vector field P on F3 by P := (P1 , P2 , P3 ), and denote by 
the standard inner
3
product on F and by curl the curl operator. Then (9.16) can be rewritten as
P  curl P = 0 .
(9.17)
Condition (9.17) provides an elementary way to produce examples and to establish
results about Poisson structures on F3 by using wellknown properties of vector
analysis. For instance, recall the classical identity
9.2 Poisson Structures in Dimension Three
251
curl P = curl P + grad P ,
(9.18)
which is valid for every function F (F3 ) and for every vector field P on F3 ; in
this formula, stands for the vector product in F3 and grad is the gradient operator.
We use (9.18) to prove the following proposition.
Proposition 9.13. Let be a Poisson structure on F3 . For every function
F (F3 ), the bivector field is a Poisson structure on F3 .
Proof. We denote, as above, by P the vector field on F3 , associated to , so that P
is the vector field associated to . One computes, with the help of (9.18),
P  curl P = 2 P  curl P + P  grad P .
By (9.17), the assumption that is a Poisson structure gives the vanishing of the
first term on the righthand side of this equation, while the vanishing of the second
term follows from the identity A  B A
= 0, valid for all A, B F3 . This shows
that condition (9.17) is satisfied for P , so that the bivector field is a Poisson
structure on F3 .
Another classical identity which involves the curl operator is
curl grad = 0 ,
valid for every function F (F3 ). It implies, in view of (9.17), that when the
vector field P , associated to a bivector field , is the gradient of a function, P =
grad , then is a Poisson structure. Substituted in (9.15), this means that
:=
x y z y z x z x y
(9.19)
is a Poisson structure on F3 , for every function F (F3 ). Let = dx dy dz
denote the standard volume form on F3 . Then it is clear from (9.19) that and d
are related by
(9.20)
= d ,
which yields an alternative definition for . Also, this Poisson bracket comes from
a standard NambuPoisson structure of order 3 on F3 , upon using the function ,
as described in Proposition 8.35. Therefore, the Poisson bracket {F, G} of two
arbitrary functions F, G F (F3 ) can also be written as
{F, G} =
dF dG d
.
dx dy dz
In particular, is a Casimir function for . Proposition 9.13 implies the following
statement.
Proposition 9.14. Let F (F3 ) be an arbitrary function on F3 and consider the
bivector field on F3 , defined by (9.19). For every function F (F3 ), the bivector field is a Poisson structure on F3 and it admits as a Casimir function.
252
9 Poisson Structures in Dimensions Two and Three
This proposition motivates the next theorem. By a slight abuse of terminology, we
call Casimir function of a bivector field = { , } (not assumed to be a Poisson
structure) every function F (F3 ) such that {F, } = 0 for all F F (F3 ).
Proposition 9.15. Let be a bivector field on F3 and suppose that admits a
Casimir function .
(1)
(2)
If the differential of is different from zero in a dense subset of F3 , then is
a Poisson structure on F3 ;
If the bivector field is a polynomial Poisson structure, and if the Casimir
function is a polynomial function for which the partial derivatives x , y
and z are relatively prime, then there exists a polynomial function
F (F3 ), such that = , where is the Poisson structure defined in (9.19).
Proof. The bivector field is a Poisson structure on F3 if and only if the Schouten
bracket [ , ]S is zero. Since
1
[ , ]S [F, G, H] = {{F, G} , H} + {{G, H} , F} + {{H, F} , G} ,
2
for all functions F, G, H F (F3 ), we have that, if is a Casimir function of , then
[ , ]S [F, G, ] = 0, for all F, G F (F3 ). Let us show that this implies that, if is a
Casimir function of and d is nonzero on a dense subset of F3 , then [ , ]S = 0,
so that is a Poisson structure on F3 . Since [ , ]S is a trivector field on F3 , it can
be written as
,
(9.21)
[ , ]S =
x y z
where F (F3 ). Plugging F := x and G := y in [ , ]S [F, G, ] = 0 we obtain,
in view of (9.21), z = 0. Similarly, one obtains x = y = 0, by plugging
in the two other pairs of variables. Since, by assumption, d is nonzero on a dense
subset of F3 , it follows that = 0 on F3 , hence that the trivector field [ , ]S is zero
on F3 . This proves (1).
We next prove (2). We assume that is a polynomial Poisson structure on F3 and
we define polynomial functions P1 , P2 , P3 F (F3 ) by
= P1
+ P2
+ P3
.
y z
z x
x y
Suppose that is a polynomial Casimir function of . We apply the equation
{F, } = 0, which is valid for all F F (F3 ), successively to F := x, F := y and
F := z. It gives the following relations between the functions P1 , P2 and P3 :
P2
= P3
,
z
y
P3
= P1
,
x
z
P1
= P2
.
y
x
(9.22)
9.2 Poisson Structures in Dimension Three
253
x , y , z are relatively prime, and let Q
denote the greatest common divisor of x and y . Then it follows from the third
equation in (9.22) that there exists F (F3 ), such that
Let us assume that the partial derivatives
QP1 =
and QP2 =
.
y
(9.23)
In view of the first (or the second) equality in (9.22), this implies that
QP3 =
.
z
(9.24)
Since Q and z are relatively prime, Q divides . Define by Q = . We obtain
in view of (9.23) and (9.24):
P1 =
This proves (2).
,
x
P2 =
and
P3 =
.
z
Remark 9.16. When is a nonconstant weight homogeneous polynomial which is
squarefree, then the partial derivatives x , y , z are relatively prime, so that the
assumptions about in item (2) of Proposition 9.15 are satisfied. Let us prove this
claim. If is weight homogeneous of weight ( ) = 0, it satisfies the weighted
Euler formula (8.10),
( ) = 1 x
+ 2 y
+ 3 z
,
x
y
z
(9.25)
where 1 , 2 and 3 are the weights of the variables x, y and z. Suppose that Q
F (F3 ) is a nonconstant irreducible polynomial function which divides the partial
derivatives x , y and z . It follows from (9.25) that Q divides . Since Q is nonconstant, at least one of the partial derivatives
Q Q
x , y
and
Q
z ,
say
Q
x ,
is different
x ,
its square Q2
from zero. Since the irreducible polynomial Q divides and
also divides , so that is not squarefree. Therefore, if is nonconstant, weight
homogeneous and squarefree, then x , y and z are relatively prime.
We finish this section with an example of Poisson structure on F3 which is not
of the form , and which in fact does not admit Casimir functions defined on F3 ,
except constant functions. Consider the bivector field on F3 , defined by
.
:= x + y
x
y
z
Note that it appeared already in Example 7.12. The associatedvector field is given
by P : (y, x, 0), whose curl is (0, 0, 2), so that P  curl P = 0, which means,
254
9 Poisson Structures in Dimensions Two and Three
in view of (9.17), that is a Poisson structure on F3 . A function F (M) is a
Casimir function for if and only if
= 0 and x
+y
=0.
z
x
y
The first equation means that does not depend on the variable z, while the second
one implies, in view of the Euler formula (8.5), that is homogeneous of degree 0
in the two remaining variables. According to Proposition 8.3, every homogeneous
function of degree zero is constant, so every Casimir function of is constant. It
follows from Proposition 9.14 that is not of the form , with , F (F3 ).
9.2.2 Poisson Manifolds of Dimension Three
We now turn to Poisson structures on threedimensional manifolds. Since the rank
of a Poisson structure is an even number, smaller than or equal to the dimension of
the manifold, the rank of every Poisson structure on a threedimensional manifold
is equal to two (except when the trivial Poisson structure is considered, in which
case the rank is zero). Many properties of such Poisson structures hold true because
their rank is two, rather than because the dimension of the manifold is three. We
will therefore, in Subsection 9.2.2.1, first study the properties of rank two Poisson
structures on manifolds of arbitrary dimension.
9.2.2.1 A Poisson Structures of Rank Two
Let be a bivector field on a manifold M. For every m M, we consider the associated linear map m : Tm M Tm M. As in the case of Poisson structures, the rank
of m , which is always even, is called the rank of at m, and is said to be a
bivector field of rank two if the rank of is at most two at all points of M and is
two at least one point of M. We denote by M(2) the open subset of M of points at
which the rank of is two. If is a bivector field of rank two on M, then the subset
{m (Tm M)  m M(2) } of the tangent bundle is a distribution of rank two on M(2) .
We show in the following proposition that the Jacobi identity for is equivalent to
the involutivity of this distribution.
Proposition 9.17. Let = { , } be a bivector field of rank two on a manifold M,
and let M(2) M denote the open subset of points at which the rank of is two.
(1)
Let U be an open subset of M and let F and G be functions on U. Then, on U,
we have
{F, G} = XF XG ;
(9.26)
(2)
The bivector field is a Poisson structure on M if and only if the distribution
{m (Tm M)  m M(2) } on M(2) is involutive.
9.2 Poisson Structures in Dimension Three
255
Proof. Suppose that is a bivector field of rank two on M. Let F and G be arbitrary
functions, defined on a nonempty open subset U of M. Let m be an arbitrary point
of U. We show that
{F, G} (m) m = (XF )m (XG )m .
(9.27)
Suppose first that the tangent vectors (XF )m and (XG )m are proportional, say
(XG )m = a (XF )m for some a F. Then the righthand side of (9.27) vanishes
at m, while
{F, G} (m) = (XG )m [F] = a(XF )m [F] = a {F, F} (m) = 0 ,
so the lefthand side of (9.27) also vanishes at m. Thus, equation (9.27) is satisfied.
Assume now that the tangent vectors (XF )m and (XG )m are not proportional.
Since the rank of at m is two, these tangent vectors span m (Tm M), so that for
every function H, defined on a neighborhood of m M, there exist a, b F such
that
(9.28)
(XH )m = a (XF )m + b (XG )m .
Applying (9.28) to F yields
(XH )m [F] = a (XF )m [F] + b (XG )m [F] = b {F, G} (m) ,
and we obtain that {F, H} (m) = b {F, G} (m). Similarly, applying (9.28) to G yields
{G, H} (m) = a {F, G} (m). Substituted in (9.28), we obtain
{F, G} (m)(XH )m = {F, H} (m)(XG )m {G, H} (m)(XF )m .
Since H is arbitrary, this proves (9.27) also when (XF )m and (XG )m are not proportional, thereby proving (1).
We now prove (2). If is a Poisson structure of rank 2 on M, then {m (Tm M) 
m M(2) }, the distribution on M(2) , defined by all Hamiltonian vector fields, is
involutive because [XF , XG ] = X{F,G} for all functions F and G. Conversely, let
be a bivector field of rank two on M and assume that the distribution {m (Tm M) 
m M(2) } is involutive. Let m be a point of M. If m is zero, then the Schouten
bracket [ , ]S , evaluated at m, vanishes as well. If m is different from zero, so that
m M(2) , then there exist functions F, G, defined in a neighborhood U of m, such
XF
that {F, G} = 0 on U. According to (1), = V XG on U, where V := {F,G}
.A
direct computation, with the help of the graded Leibniz identity (3.42), gives
1
[ , ]S = V [V , XG ] XG .
2
Since XG and V are tangent to the distribution {m (Tm M)  m M(2) }, assumed to
be involutive, the bracket [V , XG ] is also tangent to the latter distribution. Hence,
([ , ]S )m 3 m (Tm M), for every m M(2) . But the latter space is zero, since
256
9 Poisson Structures in Dimensions Two and Three
m (Tm M) is a twodimensional vector space. In conclusion, the Schouten bracket
of with itself vanishes at every point in M, so that is a Poisson structure.
As a corollary, we obtain the following analog of Proposition 9.13.
Proposition 9.18. Let be a Poisson structure of rank two on a manifold M. For
every function F (M), the bivector field is a Poisson structure.
Proof. A proof can be given by picking local coordinates and using Proposition 9.13. We give a more geometrical proof. Consider the open subsets M(2) and
of M consisting of all points where the rank of , respectively of , is two. It
M(2)
M
is clear that M(2)
(2) and that for every point m M(2) , the vector spaces
( )m (Tm M) and m (Tm M)
defined by the latter vector spaces is involutive,
coincide. The distribution on M(2)
defined by
in view of item (2) in Proposition 9.17. Hence, the distribution on M(2)
the former is also involutive, which in view of the same proposition, implies that
is a Poisson structure on M.
If in Proposition 9.18 the function is nowhere vanishing, the vector spaces
( )m (Tm M) and m (Tm M) coincide for all m M, so that both Poisson structures
have the same symplectic foliation. For regular Poisson structures of rank 2, the
following inverse statement holds.
Proposition 9.19. Let 1 and 2 be two regular Poisson structures of rank 2 on
a manifold M. If 1 and 2 have the same symplectic leaves, then there exists a
(unique) nowhere vanishing function F (M), such that 1 = 2 .
Proof. Suppose that 1 and 2 are Poisson structures which have the same symplectic leaves. For m M, denote by Sm their common symplectic leaf through m.
If both Poisson structures are regular of rank 2, then on the one hand, Tm Sm is a
twodimensional vector space, so that 2 Tm Sm is onedimensional, while on the
other hand both (1 )m and (2 )m are nonzero elements of 2 Tm Sm , so there exists
a unique (m) F , such that 1 (m) = (m)2 (m). Thus, under the above assumptions, there exists a nowhere vanishing function on M, such that 1 = 2 ; since
is given locally as the quotient of two nonvanishing functions, belongs to F (M).
9.2.2.2 B Poisson Manifolds of Dimension Three and OneForms
Let M be a real orientable manifold of dimension three and let be a volume form on M. Recall from (4.19) that there is a natural F (M)linear isomorphism : Xk (M) 3k (M), defined for Q Xk (M) by
Q := Q .
9.2 Poisson Structures in Dimension Three
257
In particular, to a bivector field on M corresponds a differential oneform :=
= i on M. In this subsection we wish to express the properties of , for example the condition that be a Poisson structure on M, in terms of the differential
oneform . Of course, this differential oneform depends on the chosen volume
form: replacing by F , with F a nowhere vanishing function, amounts to replacing by F .
Proposition 9.20. Let M be a real orientable manifold of dimension three. Suppose
that is a bivector field on M.
(1)
Let be an arbitrary volume form on M, and let := . Then is a Poisson
structure on M if and only if
d = 0 .
(2)
(9.29)
In this case, for every m M at which the rank of is two, the kernel of m is
the tangent space of the symplectic leaf of through m;
The bivector field is a unimodular Poisson structure if and only if there
exists a volume form on M, such that the differential oneform := is
closed.
Proof. Recall from Cartans formula (3.38) that if P and Q are two multivector fields
on M, then
[[P , d] , Q ] = [P,Q]S .
Specializing to P := and Q := , and applied to the volume form , it yields
2 d = [ , ]S .
(9.30)
Multiplied by the volume form , and expressed in terms of , (9.30) becomes
2( d ) = ([ , ]S ) .
(9.31)
Now, for every bivector field P and for every differential 2form , the fact that is
a top form implies that2
(P ) = P .
(9.32)
In particular, ( d ) = d = d , so that (9.31) gives eventually:
2 d = [ , ]S .
As a consequence, d = 0 if and only if [ , ]S = 0, i.e., if and only if is a
Poisson structure on M. This proves the first part of item (1).
Let us prove the second part of (1), i.e., that if the rank of at m is two, then
the spaces Ker(m ) and Tm Sm coincide, where Sm is the symplectic leaf of ,
passing through m. Since both vector spaces are twodimensional, it suffices to show
2
In the actual case where the dimension of M is 3, a simple proof of (9.32) can be given by taking
local coordinates in which = dx dy dz.
258
9 Poisson Structures in Dimensions Two and Three
the inclusion Tm Sm Ker m . The tangent space at m to Sm is spanned by the
Hamiltonian vector fields XF at m, where F is an arbitrarily function, defined in a
neighborhood of m. Now,
(XF ) = ( ) (XF ) = ( XF ) ,
and XF = 0 because the rank of is 2 (see (9.26)), so that (XF )m Ker m for
every m M and for every function F, defined in a neighborhood of m in M. This
completes the proof of (1).
We now prove (2). Recall from (4.24) that the modular vector field of with
respect to a given volume form , can be expressed in terms of the divergence of
as = Div( ). According to the definition of Div (see (4.18)) and the definition
:= , we have that
= Div( ) = d = d = d ,
so that = 0 if and only if is closed. Since a Poisson structure is unimodular if
and only if there exists a volume form with respect to which the modular vector
field is zero, this means that is unimodular if and only if there exists a volume
form such that := is closed. This shows (2).
9.2.3 Quadratic Poisson Structures on F3
In this section, we present a classification of quadratic Poisson structures on the
affine space C3 . Recall from Proposition 8.8 that two quadratic Poisson structures
on F3 are isomorphic if and only if there exists a linear isomorphism of F3 , which
sends one of the quadratic Poisson structures to the other one. As before, we denote
the standard coordinates on F3 by x, y, z, and we denote by the standard volume
form
:= dx dy dz .
Recall from Section 8.2.2 that the modular vector field, with respect to , of a
quadratic Poisson structure is a linear vector field (Proposition 8.23), and recall
from Proposition 4.17 that its divergence with respect to is zero.
Specializing and adapting Proposition 8.26 to the threedimensional case, yields
the next proposition. By definition, a cubic polynomial function on Fd is a homogeneous polynomial function of degree three on Fd .
Proposition 9.21. There is a onetoone correspondence between quadratic Poisson
structures on F3 and pairs ( , ), with a linear vector field on F3 , and a cubic
polynomial function on F3 , satisfying
Div( ) = 0
and
[ ] = 0 .
(9.33)
This correspondence assigns to such a pair ( , ) the Poisson structure, defined by
9.2 Poisson Structures in Dimension Three
1
, := E ,
3
259
(9.34)
where E is the Euler vector field, and is the Poisson structure defined in (9.19).
Proof. To start with, let us show that, given a linear vector field on F3 and a cubic
function on F3 , which satisfy (9.33), the bivector field , defined by (9.34) is
a Poisson structure. First, both E and are Poisson structures according to
Example 8.15 and Proposition 9.14 respectively. As a consequence,
2
2
2
, , , S = , E S = LE + L E .
3
3
3
It suffices to prove that both of the remaining terms vanish. For the first term, by
Proposition 8.4, the Lie derivative with respect to the Euler vector field of every
quadratic Poisson structure is zero, so that LE = 0. The vanishing of the second term follows at once by applying the following identity, valid for every vector
field V on F3 ,
(9.35)
LV = Div(V ) + V [ ]
to V := , upon using (9.33). We prove the identity (9.35) by showing that
LV = Div(V ) + V [ ] .
(9.36)
First, by item (2) in Proposition 3.11 and by (4.20), the lefthand side of (9.36) is
given by
LV = [V , ] = LV + LV = LV + Div(V ) .
S
(9.37)
Therefore, it remains to be shown that LV = V [ ] , which means, in view
of (9.20), that LV d = d(V [ ]); but this is clear since LV and d commute. This
proves (9.35), and hence that the bivector field , defined by (9.34) is a Poisson
structure.
Next, we show that the assignment ( , ) , , defined by (9.34), is surjective. Let be a quadratic Poisson bracket, and let denote its modular vector field.
Proposition 8.26 implies that there exists a closed differential oneform such that:
1
= 1 E .
3
By the algebraic Poincare lemma, there exists a function such that d = ; since
is a quadratic differential oneform, can be chosen (in a unique way) to be a
cubic function. In view of (9.20), we have 1 = 1 d = . Hence, is given
as in (9.34). We still have to check that the pair ( , ) satisfies (9.33). According to
Proposition 4.17, the divergence of the modular vector field is zero. According
to Proposition 8.26, we have that L = 0, so that d( [ ]) = 0. Since [ ] is
a cubic function, this implies [ ] = 0.
260
9 Poisson Structures in Dimensions Two and Three
We are left with the task of showing that the assignment, defined by (9.34) is
injective. Since the divergence of vanishes and according to Example 8.25,
1
1
Div( , ) = Div E = Div( ) E = ,
3
3
so that we can recover as the modular vector field of , . Since the cubic polynomial can also be uniquely recovered from , both and can be uniquely
recovered from , , which proves injectivity, hence bijectivity of the assignment,
defined by (9.34). This completes the proof.
Proposition 9.21 leads to an explicit description of the family of all quadratic
Poisson structures on F3 , as given in the following theorem.
Theorem 9.22. Let be a quadratic Poisson structure on C3 . There exist linear
coordinates on C3 in which takes the form
1
= E ,
3
where ( , ) corresponds to one of the types IVII in Table 9.2, where is the
Poisson structure
:=
,
x y z y z x z x y
and where E is the Euler vector field on C3 .
In order to prove this theorem, we first establish, in the following lemma, the list
of all divergencefree linear vector fields on C3 . Recall that in this section the divergence is always taken with respect to the standard volume form = dx dy dz
on C3 .
Lemma 9.23. Let be a divergencefree linear vector field on C3 . There exist linear coordinates x, y, z on C3 such that takes one of the forms, given in the second
column of Table 9.3.
Proof. Recall from Section 8.2 that linear vector fields on a finitedimensional vector space V are in onetoone correspondence with endomorphisms of the dual vector space V : more precisely, to a linear vector field on V corresponds, as in
(8.20), the endomorphism V (1) of V defined by
(1) : V V
F [F] .
If we write in terms of linear coordinates x1 , . . . , xd on V as
i, j=1
ai j xi
,
xj
(9.38)
9.2 Poisson Structures in Dimension Three
261
Table 9.2 Up to linear isomorphism, every quadratic Poisson structure on C3 is of the form , ,
where and are given in the second and third columns.
Type
Parameters
a1 x x + a2 y y + a3 z z
bxyz
a1 , a2 , a3 C
all ai different
a1 + a2 + a3 = 0;
bC
II
ax x + ay y 2az z
z G(x, y)
a C
G quadratic function
III
ax x ay y
bxyz + cz3
a C ; b, c C
IV
ax x + (ay + x) y 2az z
bx2 z
a C ; b C
x y + y z
bx3 + cx(y2 2xz)
b, c C
VI
x z
G(x, y)
G cubic function
VII
G(x, y, z)
G cubic function
then A := (ai j ) Matd (C) is the matrix of (1) in terms of the basis (x1 , . . . , xd )
for V . Using the classical formula (4.21) for the divergence of a vector field, we
find
d
d
(ai j xi ) = aii ,
Div( ) =
i, j=1 x j
i=1
so that the divergence of is a constant function, whose value is the trace of the
endomorphism (1) . In particular, a linear vector field on C3 is divergencefree
if and only if the corresponding endomorphism (1) of C3 is traceless. There are
seven types of such endomorphisms:
If all eigenvalues of (1) are different, then there are two possibilities, according
to whether one or zero of these eigenvalues are zero, leading to types I and III
in Table 9.3;
262
9 Poisson Structures in Dimensions Two and Three
Table 9.3 There are, up to isomorphism, seven types of traceless endomorphisms of C3 . For each
type, its matrix takes in terms of a wellchosen basis the form indicated in the second column of the
table. The range of the parameters which appear in the matrices is indicated in the third column.
Type
Matrix
Parameters
diag(a1 , a2 , a3 )
a1 , a2 , a3 C
all ai different
a1 + a2 + a3 = 0
II
diag(a, a, 2a)
a C
III
diag(a, a, 0)
a C
a1 0
0 a 0
0 0 2a
a C
IV
010
0 0 1
000
001
0 0 0
000
VI
VII
If (1) has precisely two different eigenvalues, then they are both different from
zero, leading to types II or IV, according to whether (1) is diagonalizable or
not;
If all eigenvalues of (1) are equal, then they are all zero, so that (1) is a
nilpotent endomorphism. This leads to three types, the types VVII.
The conditions on the parameters in Table 9.3 are chosen such that every traceless
endomorphism of C3 appears exactly once in the table.
It is clear that the matrices which are given in the second column of Table 9.3 correspond to the linear vector fields on C3 , listed in the second column of Table 9.2.
Combining Proposition 9.21 and Lemma 9.23 leads to the announced classification of the set of quadratic Poisson structures on C3 .
9.2 Poisson Structures in Dimension Three
263
Proof (of Theorem 9.22). Let be a quadratic Poisson structure on C3 . According
to Proposition 9.21, there exists a pair ( , ), with a divergencefree linear vector
field on C3 , and a cubic polynomial function on C3 , satisfying [ ] = 0, such that
is of the form (9.34). According to Lemma 9.23, there exist linear coordinates
x, y, z on C3 in which takes the form corresponding to one of the types IVII in
Table 9.3. It therefore suffices to determine for each of these types the vector space
of all cubic functions F (C3 ), satisfying [ ] = 0. It is easy to see, by direct
computation, that for each line of the table, the given function satisfies indeed
[ ] = 0. Let us show that if is of one of the types IV, there are no other such
(for of the other types VI and VII this is obvious).
For the types IIII, is the weighted Euler operator, with (not necessarily integral) weight vector (a1 , a2 , a3 ) with a1 + a2 + a3 = 0 for type I, with weight vector
(a, a, 2a) for type II and weight vector (a, a, 0) for type III. Therefore, for each
of these types, the vector space of all polynomials annihilated by is spanned by
all monomials F satisfying [F] = 0. Taking a cubic monomial F = x1 y2 z3 ,
this leads for type I to the equation 1 a1 + 2 a2 + 3 a3 = 0, which has (1, 1, 1) as
its unique solution, under the constraint 1 + 2 + 3 = 3 (because deg(F) = 3),
with all ai C being different. This leads to the monomial F = xyz. Similarly, the
equation [F] = 0 leads for type II to the equation a(1 + 2 ) 2a3 = 0, under
the constraint 1 + 2 + 3 = 3, which has as solution all triples (1 , 2 , 1), with
1 + 2 = 2, corresponding to all monomials of the form zG(x, y), where G is a
quadratic polynomial. Similarly, for type III, one looks for all triples (1 , 2 , 3 ),
satisfying 1 2 = 0, which has (1, 1, 1) and (0, 0, 3) as its only solutions under
the constraint 1 + 2 + 3 = 3; it corresponds to the monomials xyz and z3 .
For of type IV, one computes easily the images under of the 10 monomials
x1 y2 z3 , with 1 + 2 + 3 = 3, by using the fact that is the sum of the weighted
Euler operator (with weights (a, a, 2a)) and the operator x y . One sees that, besides for (x2 z) (which is zero), these images are linearly independent, proving
that the kernel of (restricted to the space of cubic polynomials) is spanned by
x2 z. Similarly, for of type V, one easily checks that the images of the eight cubic
monomials x1 y2 z3 , with (1 , 2 , 3 ) = (3, 0, 0) and (1 , 2 , 3 ) = (2, 0, 1) are
linearly independent. The kernel of is therefore twodimensional, and is spanned
by x3 and x(y2 2xz).
9.2.4 Poisson Surfaces in C3 and Du Val Singularities
We have seen two different ways of constructing (singular) Poisson surfaces.
(1)
(2)
Let G be a finite subgroup of SL2 (C). The standard action of G on C2 induces
on the quotient surface C2 /G a Poisson structure { , }C2 /G , making it into a
(singular) Poisson surface (see Section 6.3.4).
Let be an irreducible polynomial in three variables. The Poisson structure
= { , } on C3 , introduced in (9.19), given by
264
9 Poisson Structures in Dimensions Two and Three
:=
,
x y z y z x z x y
(9.39)
admits as a Casimir, hence the singular affine surface C3 , given by
= 0, is a (singular) Poisson surface.
The purpose of the present section is to show how these two constructions are related. Namely, we will show that every Poisson surface C2 /G of the first type is
also of the second type , for a polynomial which will be given explicitly (for
each G).
Before doing this, let us recall a few facts about the classification of finite subgroups of SL2 (C) (see [183] for proofs and details). Consider the list of subgroups
Ak , for k 1, Dk for k 4, or E6 , E7 , E8 , whose generators are given in Table 9.4.
Table 9.4 Every finite subgroup of SL2 (C) is conjugated to precisely one
of the subgroups,
given
in terms of generators in the second column of the table, where := e
2 1
5
k1
k4
2 1
k+1
1
k
2 1
k+1
Dk
and := e
Generators of G SL2 (C)
Type of G
Ak
2 1
8
0
e
1
k
0
1
and
1 0
7 7
generators of D4 and 12
E6
0
generators of E6 and
0 7
E7
E8
3 0
and
2
0
1
2 3
4
1
+
4
1
Up to conjugation, every subgroup G of SL2 (C) is conjugated to one of these subgroups A1 , A2 , . . . , E8 . We then say that G is of type A1 , A2 , . . . , E8 . The quotient surface C2 /G is singular at the image in C2 /G of the origin o of C2 , and at
9.2 Poisson Structures in Dimension Three
265
no other point; a singularity obtained in this way is called a Du Val singularity or
Kleinian singularity, depending on the reference. It is a classical result [183, 188]
that the affine variety C2 /G is, as an affine variety, isomorphic to , where is
the polynomial, depending on the type of G, which appears in the second column of
Table 9.5.
Table 9.5 For each of the finite subgroups G of SL2 (C), the quotient surface C2 /G is the zero
locus of a weight homogeneous polynomial, given in the second column of the table. The third
column indicates the weight vector with respect to which each of these polynomials is weight
homogeneous.
Weights
x2 + y2 zk+1
(k + 1, k + 1, 2)
Type of G
Ak
k1
Dk
k 4 x2 + y2 z + zk1 (k 1, k 2, 2)
E6
x 2 + y 3 + z4
(6, 4, 3)
E7
x2 + y3 + yz3
(9, 6, 4)
E8
x 2 + y 3 + z5
(15, 10, 6)
Note that this means that, for each group G which appears in the table, the algebra
of Ginvariant polynomials is an algebra with three weight homogeneous generators
P1 , P2 , P3 , and that every polynomial R C[X,Y, Z], satisfying R(P1 , P2 , P3 ) = 0, is a
multiple of . The isomorphism C2 /G is induced by the map C2 C3 , given
by (P1 , P2 , P3 ), as expressed in the following commutative diagram:
C2
(P1 ,P2 ,P3 )
C2 /G
C3
(9.40)
Recall from Section 5.1.2 that F (C2 )G F (C2 /G) (since G is a finite group). On
the other hand, since is the affine surface of C3 , defined by the polynomial ,
its algebra of functions F ( ) is given by C[x, y, z]/
. It follows that we have an
algebra isomorphism
F (C2 )G C[x, y, z]/
.
(9.41)
We show in the following proposition that the isomorphism C2 /G is actually an isomorphism of Poisson varieties.
Proposition 9.24. Let G be a finite subgroup of SL2 (C) and let { , }C2 /G denote
the Poisson structure on C2 /G, induced by the action of G on C2 . The Poisson sur
266
9 Poisson Structures in Dimensions Two and Three
face (C2 /G, { , }C2 /G ) is isomorphic, as a Poisson variety, to the Poisson surface
( , { , } ), where is the polynomial, associated to G, as indicated in Table 9.5.
Proof. Consider the isomorphism : C2 /G , given in (9.40). We show that
is a Poisson isomorphism, with respect to the stated Poisson structures (up to
a constant). According to Corollary 6.20, C2 /G has a unique singular point, the
point p(o), where o is the origin of C2 . Since the origin of C3 is clearly a singular
point of , it follows that it is the image of p(o) under and that it is the only
singular point of (the latter fact can also easily be verified by direct computation).
According to the latter corollary, the Poisson structure { , }C2 /G on C2 /G has rank
two at every point of C2 /G, except for the singular point p(o), where its rank is
zero. The Poisson matrix of { , } is given, in terms of the generators x, y and z
of F ( ) as
z
y
0
z
x
0
y
x
The rank of this matrix is two, except at the point(s) of for which
= 0, i.e., at the singular points of . Since the origin of is its only singular
point, the rank of { , } is two at all points, except at the singular point of (where
the rank is zero).
Using the isomorphism : C2 /G , we can transport the Poisson structure
{ , }C2 /G on C2 /G to a Poisson structure on , which we denote by { , } . On the
smooth surface \ {p(o)} both Poisson structures { , } and { , } are of rank 2,
hence according to Proposition 9.19 there exists a nowhere vanishing holomorphic
function on \ {p(o)} such that { , } = { , } . For arbitrary F, G F ( )
such that {F, G} = 0, we have that
{F, G}
{F, G}
so that is actually a rational function on . Since its divisor of zeros and its divisor of poles are both discrete (they contain at most the point p(o)), is constant.
This implies that ( , { , } ) and (C2 /G, { , }C2 /G ) are isomorphic Poisson varieties for some C . In view of Remark 6.19, this constant can be chosen
equal to 1.
9.3 Notes
267
9.3 Notes
The classification in dimension two given in this chapter follows Monnier [150, 152],
a variant of Arnolds proof [16]. In these references, the proof is more general than
the proof given here, since the case when R is taken as the ground field is also
considered, and with some extra work it is shown that the classification is actually
nonformal, i.e., the formal coordinate transformations can be realized by smooth
(or analytic) local coordinate transformations, when the given Poisson structure is
smooth (or analytic). The question of having a global rather than a local classification in the context of varieties or manifolds remains a real challenge; for an
important contribution in this direction, see [173], in which a classification of topologically stable Poisson structures on real surfaces is given.
The classification of quadratic Poisson structures in dimension three, detailed in
Section 9.2.3 above, has been obtained independently by DufourHaraki [60] and
LiuXu [129]. For the relation between Poisson structures and differential forms, in
the presence of a volume form, see [88].
Chapter 10
RBrackets and rBrackets
The theory of integrable systems gave birth to interesting constructions, which yield
new Poisson structures on a given (finitedimensional) Lie algebra g, upon using
some extra structure on the Lie algebra, roughly speaking a matrix. There are two
distinct formalisms for doing this, depending on whether this matrix is viewed as
a linear map R : g g, in which case R is called an Rmatrix, or as an element
r g g, in which case r is called an rmatrix. In their basic form, the new Poisson
structures are LiePoisson structures, so it will be assumed that the reader is familiar
with the basic theory of LiePoisson structures, as developed in Chapter 7.
For a given linear map R : g g, which satisfies a condition which will be explained in detail, a new Lie bracket on g is defined by letting [x, y]R := 12 ([Rx, y] +
[x, Ry]), for all x, y g. The new Lie bracket on g leads to a Poisson structure on g ,
which is referred to as an Rbracket on g . Upon identifying g with g , which is usually done by using a nondegenerate symmetric bilinear form, the Rbracket on g
becomes a LiePoisson structure on g.
A tensor r g g can be viewed as a zerochain on g, with values in g g, so
its coboundary L0 (r) : g g g admits as transpose a linear map r : g g g .
When r satisfies a condition, which will also be explained in detail, r defines a Lie
algebra structure on g , hence leads directly to a LiePoisson bracket on g, i.e., no
identification of g with g is needed to define this LiePoisson structure on g. This
LiePoisson bracket is called an rbracket.
Section 10.1 is devoted to the construction of the Rbracket on g and on g , with
special emphasis on the main example, which comes from Lie algebra splittings. In
Section 10.2, we discuss the case of rbrackets and we show how Rbrackets and
rbrackets are related. The construction of higher degree brackets, from an Rmatrix
or rmatrix, is given in Section 10.3. For the connection with PoissonLie groups,
we refer to Chapter 11.
In this chapter, F is an arbitrary field of characteristic zero, g is a finitedimensional Lie algebra over F and F (g) denotes the algebra of polynomial functions on g, but it may also be taken as the algebra of smooth, respectively holomorphic, functions on g, when F = R, respectively when F = C.
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 10, SpringerVerlag Berlin Heidelberg 2013
269
270
10 RBrackets and rBrackets
10.1 Linear RBrackets
As we have seen in Chapter 7, a Lie algebra structure on a finitedimensional vector
space leads to a Poisson structure on its dual vector space. We consider in this section Lie algebras which are equipped with a second Lie algebra structure, obtained
from the original one by using an Rmatrix, as explained below. In this case, the dual
vector space comes equipped with a second LiePoisson structure, whose properties are in general quite different from the properties of the original LiePoisson
structure.
10.1.1 RMatrices and the YangBaxter Equation
We first give the general definition of an Rmatrix and discuss its relation with the
YangBaxter equation.
Definition 10.1. Let (g, [ , ]) be a Lie algebra and let R : g g be a linear map. If
the skewsymmetric bilinear map [ , ]R : g g g, defined for x, y g by
[x, y]R :=
1
([Rx, y] + [x, Ry]) ,
2
(10.1)
defines a Lie bracket on g, then R is called an Rmatrix for g. Then g, equipped with
the two Lie algebra structures [ , ] and [ , ]R , is called a double Lie algebra.
The standard terminology Rmatrix and the notation Rx instead of R(x) are used
because one often thinks of R (and also x) as being a matrix.
We show in the following proposition that the conditions which express that a
linear map R : g g is an Rmatrix for g, can be rewritten in a symmetric form,
using a bilinear map on g, which depends quadratically on R.
Proposition 10.2. Let (g, [ , ]) be a Lie algebra. A linear map R : g g is an Rmatrix for g if and only if
x, y, z g :
[BR (x, y) , z] + (x, y, z) = 0 ,
(10.2)
where BR Hom(g g, g) is defined, for x, y g, by
BR (x, y) := [Rx, Ry] R ([Rx, y] + [x, Ry]) .
Proof. For x, y, z g, we have
4 [[x, y]R , z]R + (x, y, z)
= 2 [[Rx, y] + [x, Ry] , z]R + (x, y, z)
= [R ([Rx, y] + [x, Ry]) , z] + [[Rx, y] + [x, Ry] , Rz] + (x, y, z)
= [R ([Rx, y] + [x, Ry]) , z] + [[Ry, z] , Rx] + [[z, Rx] , Ry] + (x, y, z)
(10.3)
10.1 Linear RBrackets
271
= [R ([Rx, y] + [x, Ry]) [Rx, Ry] , z] + (x, y, z) ,
where we have used the Jacobi identity for [ , ] in the last step. This shows that the
Jacobi identity for [ , ]R is equivalent to (10.2).
It follows from the proposition that, if there exists a constant c F such that
BR (x, y) = c [x, y] for all x, y g, then R is an Rmatrix for g. Given c F, the
equation
x, y g : BR (x, y) = c [x, y] ,
(10.4)
viewed as an equation in R, is called the modified YangBaxter equation. When g is
a complex (respectively real) Lie algebra and c
= 0, then a rescaling of R in (10.4)
yields c = 1 (respectively c = 1). When c = 0, one speaks of the YangBaxter
equation. Sometimes one adds the adjective classical, to distinguish these equations
from the quantum YangBaxter equation. We warn the reader that the terminology YangBaxter equation, with or without adjectives, is used for many equations
which are closely related to, but not equivalent to, the above YangBaxter equation
(see [108]).
Given a Lie algebra (g, [ , ]), a linear map T : g g is called an intertwining
operator of g if T [x, y] = [T x, y] = [x, Ty], for all x, y g, where we have written T x
as a shorthand for T (x). An intertwining operator of g can be used to construct a
new Rmatrix for g, starting from a given Rmatrix for g, as given in the following
proposition.
Proposition 10.3. Let (g, [ , ]) be a Lie algebra and let R be an Rmatrix for g. If
T : g g is an intertwining operator of g, then the linear map R T is an Rmatrix
for g.
Proof. Suppose that R is an Rmatrix for g and that T is an intertwining operator
of g. According to Proposition 10.2, we need to show that
[BR (T x, Ty) , z] + (x, y, z) = 0 ,
(10.5)
for all x, y, z g, where we have used that BRT (x, y) = BR (T x, Ty), which is valid
because T is an intertwining operator of g. If T is invertible, then (10.5) is equivalent to
(10.6)
T [BR (T x, Ty) , z] + (x, y, z) = 0 ,
which holds for all x, y, z g, because T is an intertwining operator of g and in view
of (10.2). If T is not invertible, pick an arbitrary F and replace in (10.6) T by
T + 1g . The identity
(T + 1g ) [BR (T x + x, Ty + y) , z] + (x, y, z) = 0
(10.7)
holds because T + 1g is also an intertwining operator. Since (10.7) is a cubic polynomial in and since it holds for all F, all of its coefficients as a polynomial
in are equal to zero. This gives the following equations:
272
10 RBrackets and rBrackets
[BR (x, y), z] + (x, y, z) = 0 ,
[BR (T x, y) + BR (x, Ty), z] + T [BR (x, y), z] + (x, y, z) = 0 ,
[BR (T x, Ty), z] + T ([BR (T x, y) + BR (x, Ty), z])+ (x, y, z) = 0 ,
T [BR (T x, Ty), z] + (x, y, z) = 0 .
In view of the first line, the second line becomes
[BR (T x, y) + BR (x, Ty), z] + (x, y, z) = 0 ;
next, substituting this in the third line yields (10.5).
If T is an intertwining operator of g and R is a solution of the YangBaxter equation, then the Rmatrix R T is also a solution of the YangBaxter equation, since
BRT (x, y) = BR (T x, Ty) for all x, y g. However, the corresponding property for
solutions of the modified YangBaxter equation does not hold in general.
10.1.2 RMatrices and Lie Algebra Splittings
The main, and historically first, example of an Rmatrix comes from a Lie algebra
splitting. If (g, [ , ]) is a Lie algebra and g is written as a vector space direct sum
g = g+ g of two Lie subalgebras g+ and g of g, then g+ g is called a Lie
algebra splitting. It leads to two projections P+ : g g+ and P : g g . For
x g, we abbreviate P+ (x) to x+ and P (x) to x . A Lie algebra splitting leads to
an Rmatrix for g, as described by the following proposition.
Proposition 10.4. Let (g, [ , ]) be a Lie algebra and let g = g+ g be a Lie algebra
splitting, with corresponding projections P : g g . The involutive endomorphism
R := P+ P of g is a solution of the modified YangBaxter equation (10.4), with
c = 1, namely
[Rx, Ry] R ([Rx, y] + [x, Ry]) = [x, y] ,
(10.8)
for all x, y g. In particular, R is an Rmatrix for g.
Proof. Let g = g+ g be a Lie algebra splitting and let R := P+ P be the difference of the corresponding projection maps. For x, y g, we have
[Rx, Ry] + [x, y] = 2([x+ , y+ ] + [x , y ]) ,
while
R([Rx, y] + [x, Ry]) = R(2 [x+ , y+ ] 2 [x , y ]) = 2([x+ , y+ ] + [x , y ]) ,
where we have used in the last step that g+ and g are Lie subalgebras of g. This
proves that R satisfies (10.8).
10.1 Linear RBrackets
273
Notice that, when R = P+ P comes from a Lie algebra splitting, then an alternative formula for the Rbracket [ , ]R is given, for x, y g, by
[x, y]R = [x+ , y+ ] [x , y ] .
(10.9)
This elementary fact has important consequences, as we will see in the next section. Notice that (10.9) can be used to give an elementary direct proof of the Jacobi
identity for [ , ]R , without using Propositions 10.2 and 10.4.
Remark 10.5. More generally, suppose that g = g+ g0 g is a vector space
decomposition of a Lie algebra g, where g+ and g are Lie subalgebras, g0 is
abelian and is contained in the normalizer of g+ and of g , i.e., [g0 , g ] g . Let
R End(g) be defined for x g by Rx := x+ x , where x+ , x0 and x stand for
the projections of x on g+ , g0 and g respectively. Then one finds, as in the proof of
Proposition 10.4, that
[Rx, Ry] + [x, y] = 2 [x+ , y+ ] + 2 [x , y ] + [x0 , y] + [x, y0 ]
= R([Rx, y] + [x, Ry]) ,
for all x, y g. It follows that, as in the case of a Lie algebra splitting, R is a solution
of the modified YangBaxter equation (10.4), with c = 1.
10.1.3 Linear RBrackets on g and on g
If we assume that the Lie algebra (g, [ , ]) is finitedimensional, every Rmatrix for g
leads to a LiePoisson structure on g , called an Rbracket; the trivial case R = 1g
is usually excluded, because it corresponds to the canonical LiePoisson structure
on g . According to (7.5), it is given for all F, G F (g ) at g by
{F, G}R ( ) := , d F, d G R
1
, R d F , d G + d F, R d G
,
=
2
where we recall that d F is viewed in this formula as an element of g, under the
canonical isomorphism between g and its bidual. Also, , stands for the canonical
pairing between g and g. When R = P+ P comes from a Lie algebra splitting
g = g+ g , this formula can, according to (10.9), also be written as
{F, G}R ( ) = , d F + , d G + , d F , d G . (10.10)
Upon using, as in Section 7.2, a nondegenerate symmetric bilinear form  on g,
we can transfer the Rbracket to g, which leads according to (7.10) for a general
Rmatrix R to
274
10 RBrackets and rBrackets
{F, G}R,g (x) =
1
x  [R (x F) , x G] + [x F, R (x G)] ,
2
for F, G F (g) and x g. When R comes from a Lie algebra splitting g = g+ g ,
this bracket becomes
{F, G}R,g (x) = x  (x F)+ , (x G)+ x  (x F) , (x G) , (10.11)
for F, G F (g) and x g.
An important property of the latter Rbracket is that it restricts to two natural
subspaces of g, namely the orthogonal complement g
+ of g+ , respectively g of g ,
with respect to  (see (7.13)). It is given in the following proposition.
Proposition 10.6. Let (g, [ , ]) be a finitedimensional Lie algebra, equipped with
a nondegenerate symmetric bilinear form  , and let R be the Rmatrix for g,
associated with a Lie algebra splitting g = g+ g . Then the orthogonal comple
ments g
+ and g are Poisson submanifolds of (g, { , }R,g ). Anexplicit formula for
the restricted Poisson bracket on g
is given, for F, G F g at x g , by
{F, G} (x) = x  x F + , x G + ,
(10.12)
where F and G are arbitrary extensions of F and G to g. An analogous formula
holds for the restricted Poisson bracket on g
+.
Proof. It is clear from formula (10.9) for the Rbracket which comes from a Lie
algebra splitting that [g+ , g]R g+ and that [g , g]R g . Therefore, g+ and g
are both Lie ideals of (g, [ , ]R ) and we may conclude, according to Proposition 7.5,
that g
+ and g are both Poisson submanifolds of (g, { , }R,g ). Equation (10.12)
follows at once upon combining Eqs. (7.14) and (10.11).
We finish this section by giving two simple, but important, examples of Rbrackets.
Example 10.7. Consider the Lie algebra splitting of g := gld (F), given by
g = d d< = g+ g ,
where g+ := d (respectively g := d< ) denotes the Lie algebra of upper triangular matrices (respectively of strictly lower triangular matrices). Letting x  y :=
<
= d , the vector space of all lower triangular
Trace(xy), we have that g
= d
matrices in g. If F is a linear function on g, then x F = F is independent of x g
and it follows from (7.12) that it is given by F  y = F(y), for all y g. Denoting
by xi j the linear function which picks the element (i, j) of a matrix, it follows that
xi j = E ji , the matrix whose only nonzero entry is a 1 at position ( j, i). According
to (10.12), the Poisson bracket on d is therefore given by
xi j , xk (x) = x  [E ji , Ek ] ,
where j i and k.
10.2 Linear rBrackets
275
This example is easily generalized to arbitrary semisimple Lie algebras: upon
choosing a Cartan subalgebra h, such a Lie algebra g can be decomposed as
g = n h n+ , where each one of the four subspaces n and b := n h is
a Lie subalgebra of g. Thus, g = b+ n is a Lie algebra splitting. With respect to
the Killing form, n
= b and we have on the latter vector space a linear Poisson
structure.
Example 10.8. Another natural Lie algebra splitting of gld (F) is given by gld (F) =
d g , where g stands for the Lie subalgebra of skewsymmetric matrices. Using again x  y := Trace(xy), we have that g
is the vector space of all symmetric
d d matrices, which inherits a Poisson structure from the Lie algebra splitting.
10.2 Linear rBrackets
In the previous section, we have seen that a linear map R : g g leads, under certain
conditions, to a new LiePoisson structure on g , and hence on g, when g and
g have been identified. In the present section, we describe a similar construction,
starting from an element r g g, which gives, under certain conditions, directly a
LiePoisson structure on g.
10.2.1 Coboundary Lie Bialgebras and rMatrices
Let r be an element of g g, where (g, [ , ]) is a finitedimensional Lie algebra.
In this section, we explain how there is naturally associated with r a linear map
g g g, whose transpose defines, under some conditions, a Lie algebra structure
[ , ]r : g g g . To do this, we need a natural representation of g on the tensor
algebra T g of g and we need to make explicit the Lie algebra cohomology of g,
with values in this representation; see Section A.1 in Appendix A and Section 4.1.1
for the basic definitions of tensor algebra, respectively of Lie algebra cohomology.
A natural representation of a Lie algebra g on its tensor algebra T g is obtained
by extending, for every x g, the linear map adx : g g to a (graded) derivation (of
degree zero) of T g. Explicitly, for a monomial y1 y2 y p T g,
p
adx (y1 y2 y p ) := y1 [x, yi ] y p .
(10.13)
i=1
The Jacobi identity for [ , ] implies that the linear map ad : g T g T g, defined
by (10.13), is a representation of g on T g. Combined with the natural projection
map T g g, we obtain the adjoint representation of g on g which, as we have
seen in Section 5.1.3, can be written in terms of the algebraic Schouten bracket:
adx Y = [[x,Y ]] for all x g and all Y g. Elements Y of T g or of g for which
276
10 RBrackets and rBrackets
adx Y = 0, for all x g, are called adinvariant; for example, elements of degree one
are adinvariant if and only if they belong to the center of g.
It is clear from (10.13) that adx can be restricted to g g, which yields a representation of g on g g. It allows us to consider, for p N, the vector space
C p (g) := Hom( p g, g g), of skewsymmetric plinear maps on g, with values in
the representation space g g. According to Section 4.1.1, C (g) carries a coboundary operator L : C (g) C+1 (g). Applied to r g g C0 (g), we obtain a linear
map r := L0 (r) : g g g, which is given by
r (x) = adx r ,
for all x g. The transpose to r is a linear map r : g g (g g) g , which
we view as a bilinear map g g g , still denoted by r . Explicitly,
r ( , ), x = ( )(adx r)
(10.14)
for all x g and , g . We will be interested in the case in which the bilinear
map r defines a Lie algebra structure on g .
Definition 10.9. Let (g, [ , ]) be a finitedimensional Lie algebra. A bivector r
g g is called an rmatrix for g if the transpose r of r = L0 (r) : g g g is a
Lie bracket on g . In this case, r leads to a Lie bracket on g , denoted by [ , ]r , and
defined by1
[ , ]r , x := , adx r
(10.15)
for all , g and x g. This bracket is called the Lie bracket, associated with r,
and (g, [ , ], [ , ]r ) is called a coboundary Lie bialgebra.
As the terminology suggests, coboundary Lie bialgebras are a particular case of
Lie bialgebras, but we will only consider this more general class when we discuss
PoissonLie groups, see Chapter 11.
We will give necessary and sufficient conditions on r gg so that (g, [ , ], [ , ]r )
is a coboundary Lie bialgebra. Our proof will be based on the following proposition.
Proposition 10.10. Let : V V V be a linear map, where V is a finitedimensional vector space. Denote by : V +1V the extension of to a
graded derivation of V of degree 1 and let : V V V denote the transpose map to . Then satisfies the Jacobi identity (so that defines a Lie algebra
structure on V ) if and only if = 0.
Proof. Recall from Section A.5 in Appendix A, in particular Example A.3, that
extends uniquely to a derivation of V of degree 1, where the derivation property
means that
1 We use here and below the natural pairing , between p g and p g, introduced in (5.10).
When p = 2, it is given, for all 1 , 2 g and for all x1 , x2 g, by 1 2 , x1 x2 =
1 (x1 )2 (x2 ) 1 (x2 )2 (x1 ).
10.2 Linear rBrackets
277
(X Y ) = (X) Y + (1) p X (Y ) ,
(10.16)
V .
and Y
It follows from (10.16) that = 0 if and only if
for X
( (v)) = 0 for all v V . Now, for all , , V ,
pV
()
( (v)), = (v), ( ) + ( , , )
= v, ( ) + ( , , ) ,
so that = 0 if and only if satisfies the Jacobi identity. Step () requires a
proof. Let X 2V , which we suppose to be of the form X = v w, with v, w V
(in general, an element of 2V is a linear combination of such monomials). In view
of the graded derivation property (10.16) we have, for all , , V , that
(X),
= (v) w (w) v,
= (v), w, (w), v, + ( , , )
= v, ( ) w, w, ( ) v, + ( , , )
= v w, ( ) + ( , , )
= X, ( ) + ( , , ) .
This proves ().
We are now ready to give necessary and sufficient conditions on r, implying that
(g, [ , ], [ , ]r ) is a coboundary Lie bialgebra. They will be expressed in terms of
the algebraic Schouten bracket [[ , ]] on g (see Section 3.3.3). We denote for s
g g by s+ (respectively s ) its symmetric (respectively skewsymmetric) part. For
example, if x, y g, then (x y) = x y, viewed as an element of 2 g.
Proposition 10.11. Let (g, [ , ]) be a finitedimensional Lie algebra and let r be an
element of g g. Then r is an rmatrix for g if and only if r satisfies the following
two conditions:
(1)
(2)
r+ is adinvariant;
[[r , r ]] is adinvariant.
In this case, the Lie bracket associated with r is given by
[ , ]r , x = , adx r ,
(10.17)
for all , g and x g.
Proof. Let , g and let x g. Then
r ( , ) + r ( , ), x = ( )(adx r) + ( )(adx r) = 2( )(adx r+ ) ,
(10.18)
278
10 RBrackets and rBrackets
where we have used that (adx r) = adx r ; to verify the latter, write r as a sum of
elements of the form y z z y. It follows from (10.18) that r is skewsymmetric
if and only if r+ is adinvariant.
Suppose now that r+ is adinvariant. Then r (x) = adx r = adx r = r (x) for
all x g, so that r depends only on a := r . Thus we are given a skewsymmetric
element a g g and we wish to prove that the Jacobi identity for a is equivalent
to the adinvariance of [[a, a]]. In view of Proposition 10.10, it suffices to show that
the adinvariance of [[a, a]] is equivalent to a a = 0 (on g; equivalently, on g).
Combining (10.13) and (3.46), we find that
a (x) = adx a = [[x, a]] = [[a, x]] ,
for all x g. Since a and [[a, ]] are both graded derivations of degree 1 of g, it
follows that a (X) = [[a, X]], for all X g. Therefore, a a = 0 if and only if
[[a, [[a, x]]]] = 0, for all x g. In view of the graded Jacobi identity for [[ , ]],
[[x, [[a, a]]]] + [[a, [[a, x]]]] [[a, [[x, a]]]] = 0 ,
so that [[a, [[a, x]]]] = 0, for all x g, is equivalent to [[x, [[a, a]]]] = 0, for all x g,
which is precisely the adinvariance of [[a, a]].
The simplest way to satisfy both conditions in Proposition 10.11 is to demand that
r is skewsymmetric and satisfies [[r, r]] = 0.
10.2.2 Linear rBrackets on g
Suppose that r g g is an rmatrix for a finitedimensional Lie algebra (g, [ , ]).
Recall from Definition 10.9 and Proposition 10.11 that this means that the bilinear
map, defined for x g and , g by
[ , ]r , x := , adx r = , adx r ,
(10.19)
defines a Lie algebra structure on g . Then g inherits a linear Poisson structure { , }r
from [ , ]r , since g (g ) is then the dual of a Lie algebra. According to (7.4), the
Poisson bracket of two functions F, G F (g) is given by
{F, G}r (x) = [dx F, dx G]r , x = dx F dx G, adx r ,
(10.20)
for all x g. This leads to the following proposition.
Proposition 10.12. Let (g, [ , ]) be a finitedimensional Lie algebra and let r be an
rmatrix for g. Let (e1 , . . . , ed ) be a basis for g and let (1 , . . . , d ) be the dual basis.
The Poisson matrix of { , }r at x g, with respect to the coordinates 1 , . . . , d , is
the matrix of adx r in the basis (e1 , . . . , ed ).
10.2 Linear rBrackets
279
Proof. By definition, the matrix of adx r g g in the basis (e1 , . . . , ed ) is the
skewsymmetric matrix (ai j )1i, jd , defined by
adx r =
ai j ei e j .
1i< jd
Combined with (10.20), this means that the coefficients of the matrix can be computed by
ai j = i j , adx r = i , j r (x) ,
as was to be shown.
A compact formula for the collection of Poisson brackets i , j r can be given
when g is the Lie algebra of an associative algebra (or is a subalgebra of such a Lie
algebra2 ). To show this, let us assume that we have an associative product on g and
that [x, y] = xy yx for all x, y g. Then
adx r = [[x, r ]] = 1 x + x 1, r ,
where the latter bracket stands for the commutator in g g. Then
, r (x) = 1 x + x 1, r ,
(10.21)
where the lefthand side is the bivector, defined by
, r (x) :=
i , j r (x) ei e j .
1i< jd
In this formula, (e1 , . . . , ed ) is a basis of g and (1 , . . . , d ) is its dual basis. Equation (10.21) is known as the first Russian formula.
The first Russian formula is very useful for explicit computations, in particular
when g is realized as a Lie subalgebra of glN (F) MatN (F). Then this formula
can be thought of as an equality of matrices of size N 2 . Indeed, there is a standard
algebra isomorphism between MatN (F) MatN (F) and MatN 2 (F); to describe it,
we use the classical elementary matrices Ei j , whose only nonzero entry is a 1 at
position (i, j), but we use the convention that numbering starts from zero, in order
to simplify the formula. Then the N 2 elements Ei j with 0 i, j < N form a basis
of MatN (F) and the N 4 elements Ei j Ek , with 0 i, j, k, < N are a basis of
MatN (F) MatN (F). The isomorphism MatN (F) MatN (F) MatN 2 (F) is given
by mapping Ei j Ek to EiN+k, jN+ . It is an isomorphism of associative algebras,
since
(Ei j Ek )(Ei j Ek ) = j,i ,k Ei j Ek ,
while
2
EiN+k, jN+ Ei N+k , j N+ = j,i ,k EiN+k, j N+ .
Since g is finitedimensional, g is isomorphic to a subalgebra of glN (F), for some N, which is the
Lie algebra of the associative algebra Matn (F) of square matrices of size N.
280
10 RBrackets and rBrackets
10.2.3 From rMatrices to RMatrices
In this section, we show how Rmatrices and rmatrices on a finitedimensional
quadratic Lie algebra g are related. Denoting the bilinear form of g by  , we
have an isomorphism : g g , defined by (x), y := x  y, for all x, y g. This
leads also to an isomorphism
1g : g g g g End(g) ,
which allows us to associate with every element of g g a linear map g g. We
recall that, for a given linear map R : g g, we denote its adjoint with respect to
 by R . In formulas, this means that
Rx  y = x  R y ,
for all x, y g. If R = R (respectively R = R), then R is said to be symmetric
(respectively skewsymmetric). Obviously, R can be decomposed uniquely as R =
R+ + R , where R+ is symmetric and R is skewsymmetric.
Proposition 10.13. Let (g, [ , ]) be a finitedimensional quadratic Lie algebra and
let : g g be the isomorphism which is induced by the bilinear form on g. Let
r g g and R g g End(g) be two elements related to each other by
R = 4( 1g )(r)
(10.22)
(or, equivalently, r = 14 ( 1 1g )(R)). Then, R is a skewsymmetric Rmatrix for g
if and only if r is a skewsymmetric rmatrix for g. In this case, : (g, [ , ]R )
(g , [ , ]r ) is a Lie algebra isomorphism, where [ , ]R is the Lie bracket on g, associated with R, and [ , ]r the Lie bracket on g , associated with r.
Proof. We decompose r = s t , which we write in this proof as s t in
order to simplify the formulas: in the sequel of the proof, a sum over is implicit
in all formulas which contain . Using this notation, it follows from (10.22) that
Rx = 4 s  xt for x g. It implies that R x = 4 t  x s and hence that r is skewsymmetric if and only if R = R, i.e., R is also skewsymmetric. Let us suppose
that R (and hence r) is skewsymmetric. We show that
([x, y]R ) = [ (x), (y)]r ,
(10.23)
for all x, y g, which is, in view of the definition (10.1) of [ , ]R , equivalent to
showing that, for all x, y, z g,
1
[Rx, y] + [x, Ry]  z = [ (x), (y)]r , z .
2
(10.24)
With the above formulas for R, the lefthand side in (10.24) can be written as
10.2 Linear rBrackets
281
2 s  x [t , y] + t  y [s , x]  z = 2(s  x [t , y]  z + t  y [s , x]  z) ,
(10.25)
while the righthand side in (10.24) is given, in view of (10.19), by
[ (x), (y)]r , z = (x) (y), adz r
= (x) (y), [z, s ] t + s [z,t ]
= x  [z, s ] y t + x  s y  [z,t ]
y  [z, s ] x t y  s x  [z,t ] ,
which is the same as (10.25), because  is adinvariant and r is skewsymmetric.
This proves (10.23), which yields at the same time that [ , ]R is a Lie bracket if and
only if [ , ]r is a Lie bracket, and that is then an isomorphism of Lie algebras. This
completes the proof, since saying that [ , ]R (respectively [ , ]r ) is a Lie bracket is
tantamount to saying that R is an Rmatrix (respectively r is an rmatrix).
We leave it as an exercise for the reader to check that, under the assumptions of
Proposition 10.13, the operator BR , defined in (10.3), is expressible in terms of the
bracket [[r , r ]], defined in (3.46), as
z  BR (x, y) = (x) (y) (z), [[r , r ]] ,
for all x, y, z g. It implies that [[r , r ]] = 0 if and only if BR (x, y) = 0 for all x, y g,
i.e., if and only if R is a solution of the YangBaxter equation.
Example 10.14. Let (g, [ , ]) be a finitedimensional Lie algebra, equipped with a
nondegenerate symmetric bilinear form  . Suppose that g = g+ g is a Lie
algebra splitting, where each of the factors is isotropic: g+  g+ = g  g = {0} .
Let R be the linear endomorphism of g, given by R := P+ P , where P+ and P are
the projections onto g+ and g respectively. According to Proposition 10.4, R is an
Rmatrix. For x g, we denote x+ := P+ (x) and x := P (x). Then the Lie bracket
associated with R is given, for all x, y g, by
[x, y]R = [x+ , y+ ] [x , y ] .
Since g+ and g are isotropic, R is skewsymmetric. Let (1 , . . . , d ) be a basis
a (unique) basis (e1 , . . . , ed ) of g which is dual to it with respect
of g+ . There exists
to  , i.e., ei  j = i, j for i, j = 1, . . . , d; indeed, both g+ and g are isotropic
and  is nondegenerate. The endomorphism R can be described, for all x g, by
d
i=1
i=1
R(x) = ei  x i i  x ei ,
so that R = di=1 (ei ) i di=1 (i ) ei , as an element of g g. It follows that
the rmatrix r := 14 ( 1 1g )(R), associated with R, is given by
282
10 RBrackets and rBrackets
r=
1 d
e i i .
2 i=1
(10.26)
Example 10.15. The above example can be easily generalized to the case of a decomposition g = g+ g0 g , as in Remark 10.5. It is assumed, as in the cited
remark, that g+ and g are Lie subalgebras of g and that g0 is abelian and is contained in the normalizer of g+ and of g . Moreover, it is assumed that g comes
equipped with a nondegenerate symmetric bilinear form  with respect to which
both g+ and g are isotropic and orthogonal to g0 . Recall that the endomorphism R
of g, defined for x g by Rx := x+ x , where x+ , x0 and x stand for the projections of x on g+ , g0 and g respectively, is a solution of the modified YangBaxter
equation (10.4), with c = 1. In view of the above assumptions,
Rx  y = x+ x  y = x+  y x  y+ = x  y y+ = x  Ry ,
for all x, y g, so that R is skewsymmetric. Let (1 , . . . , d ) be a basis of g+ . There
exists in view of the above assumptions a (unique) basis (e1 , . . . , ed ) of g which
is dual to it with respect to  ; a basis of g0 is not needed in the formulas which
follow. As in the previous example, R and the corresponding r are explicitly given by
R=
i=1
i=1
(ei ) i (i ) ei ,
1 d
r = e i i .
2 i=1
Example 10.16. Starting from a particular case of Example 10.15, we construct an
Rmatrix for sud and we compute the corresponding rmatrix. We consider sld (C)
decomposed as sld (C) = g+ g0 g , where g+ (respectively g ) consists of the
strictly upper (respectively lower) triangular matrices in sld (C) and g0 consists of
the diagonal matrices in sld (C). Viewed as a real vector space, sld (C) contains
sud := x sld (C)  x = x
as a Lie subalgebra. Consider on sld (C) the Rmatrixassociated to the above decomposition, given in Example 10.15 and let R := 2 1R, which is an Rmatrix
for sld (C), because R is a multiple of R. Notice that R leaves sud invariant, since
= 0, which in turns yields 1Rx+ 1(Rx) = 0.
x+x = 0 implies that x+ +x
It follows that R can be restricted to sud and is an Rmatrix for sud . We consider the symmetric bilinear form on sud , given by (x, y) x  y := (Trace(xy)).
It is nondegenerate, hence induces an isomorphism between sud and its dual,
and R is skewsymmetric with respect to  . Define, for all 1 i < j d,
the skewsymmetric matrices Ai j := 12 (Ei j E ji ), the symmetric matrices Si j :=
1 (Ei j
2
+ E ji ) and the diagonal matrices Hi := 1
(Eii Ei+1,i+1 ). These matri2
ces form a basis of sud , with the set of all Si j and Ak forming an orthogonal set,
10.3 RBrackets and rBrackets of Higher Degree
283
with Si j  Si j = Ai j  Ai j = 1. The restriction of R to sud vanishes on diagonal
matrices, and satisfies
R (Ai j ) = 2Si j and R (Si j ) = 2Ai j ,
so that it is given by
R = 2
(Si j ) Ai j 2
1i< jd
(Ai j ) Si j .
1i< jd
As a consequence, the corresponding rmatrix r is given by
r =
Si j Ai j .
1i< jd
10.3 RBrackets and rBrackets of Higher Degree
In this section we show that, under some assumptions, an Rmatrix leads not only
to a linear Poisson structure (see Section 10.1), but also to a quadratic and a cubic
Poisson structure. For each of these Poisson structures, the finitedimensional Lie
algebra g, which underlies the construction, is assumed to be the Lie algebra of an
associative algebra: we suppose that we have an associative product on g and the Lie
bracket on g is defined, for all x, y g, by [x, y] := xy yx. It is moreover assumed
that g comes equipped with a nondegenerate symmetric bilinear map  , which
satisfies, for all x, y, z g,
xy  z = x  yz .
(10.27)
Notice that this property implies that  is adinvariant.
Example 10.17. The key example which the reader should keep in mind is that of
g := gld (F), which is the Lie algebra of Matd (F), the associative algebra of d d
matrices with coefficients in F. The symmetric bilinear form  , which is defined
by x  y := Trace(xy), for all x, y g, satisfies (10.27).
We start with the construction of the quadratic Rbracket.
Proposition 10.18. Let (g, [ , ]) be a finitedimensional Lie algebra, associated
with an associative algebra, equipped with a nondegenerate symmetric bilinear
map  , which satisfies xy  z = x  yz, for all x, y, z g. Let R : g g be an arbitrary linear map. If R and R are solutions of the modified YangBaxter equation
(10.4), with the same constant, then the bivector field on g, defined for F, G F (g)
and x g by
{F, G}Q
R (x) :=
1
2
[x, x F]  R(x x G + x G x)
12 [x, x G]  R(x x F + x F x) ,
(10.28)
284
10 RBrackets and rBrackets
is a quadratic Poisson structure on g. If H is an Adinvariant function on g, then its
Hamiltonian vector field with respect to this Poisson structure is given by
x = [R(x x H), x] .
(10.29)
Proof. It is clear that { , }Q
R , as defined by (10.28), is a quadratic skewsymmetric
biderivation of F (g), i.e., it defines a bivector field on g. We need to show that
{ , }Q
R satisfies the Jacobi identity, which amounts to verifying that, if F1 , F2 and F3
are linear functions on g, then for all x g,
Q
{F1 , F2 }Q
,
F
(x)+ (1, 2, 3) = 0 .
3
R
R
(10.30)
In order to organize this verification, it is useful to write R = R+ + R , where R+
is symmetric and R is skewsymmetric. Since (10.28) depends linearly on R, we
have that { , }Q
R = { , }+ + { , } , where { , }+ and { , } are the bivector fields,
defined as in (10.28), but using R instead of R. Since R+ = R+ and R = R ,
these two bivector fields can be written as
{F, G}+ (x) = R+ (x x F)  x G x R+ (x F x)  x x G ,
{F, G} (x) = R (x F x)  x G x R (x x F)  x x G ,
(10.31)
for all F, G F (g). For the linear functions Fi (i = 1, 2, 3), their gradient x Fi g is
independent of x; we therefore simply denote x Fi by fi . When computing (10.30),
we will need x {F1 , F2 }+ and x {F1 , F2 } . Since {F1 , F2 }+ and {F1 , F2 } depend
quadratically on x, these gradients are easily computed from (10.31), giving
x {F1 , F2 }+ = R+ (x f1 ) f2 + f1 R+ ( f2 x) f2 R+ ( f1 x) R+ (x f2 ) f1 ,
x {F1 , F2 } = R ( f1 x) f2 R ( f2 x) f1 + f1 R (x f2 ) f2 R (x f1 ) .
(10.32)
Using in two subsequent steps that R is skewsymmetric and then that R satisfies
the modified YangBaxter equation, we compute
{F1 , F2 } , F3 (x)+ (1, 2, 3)
= [R (x f1 ), R (x f2 )]  x f3 [R ( f1 x), R ( f2 x)]  f3 x + (1, 2, 3)
= BR (x f1 , x f2 )  x f3 BR ( f1 x, f2 x)  f3 x
= c [x f1 , x f2 ]  x f3 + c [ f1 x, f2 x]  f3 x
= 0,
which shows that { , } is a (quadratic) Poisson structure. In particular, when
we write (10.30) in terms of R+ and R , only three terms survive: the Jacobiator
of { , }+ ,
10.3 RBrackets and rBrackets of Higher Degree
285
{F1 , F2 }+ , F3 + + (1, 2, 3)
= [R+ ( f1 x), R+ ( f2 x)]  x f3 [R+ (x f1 ), R+ (x f2 )]  f3 x + (1, 2, 3) ,
(10.33)
and the following two other terms:
{F1 , F2 }+ , F3 + {F1 , F2 } , F3 + + (1, 2, 3)
(10.34)
= 2 R+ [x f1 , x f2 ]R  f3 x 2 R+ [ f1 x, f2 x]R  x f3 + (1, 2, 3) ,
where the Lie bracket [ , ]R is defined as in (10.1), using R instead of R: for
x, y g,
1
[x, y]R := ([R x, y] + [x, R y]) .
2
Clearly, (10.33) and (10.34) sum up to zero (which yields (10.30)), if
1
R+ : (g, [ , ]R ) (g, [ , ])
2
is a homomorphism of Lie algebras. Written out, this means that it suffices to prove
that
[R+ x, R+ y] R+ [R x, y] R+ [x, R y] = 0 ,
(10.35)
for all x, y g. We define a linear map BR : g g g by requiring that for all
x, y, z g,
BR (x, y)  z = x  BR (y, z) .
(10.36)
Explicitly, BR is given, for x, y g, by
BR (x, y) = R [Rx, y] R [x, R y] [Rx, R y] .
Notice that
BR (y, x) BR (x, y)
= 2([R+ x, R+ y] [R x, R y] R [R x, y] R [x, R y]) ,
so that
1
1
BR (x, y) + BR (y, x) BR (x, y)
2
2
= [R+ x, R+ y] R+ [R x, y] R+ [x, R y] ,
(10.37)
which is the lefthand side of (10.35). Since R is a solution of the modified Yang
Baxter equation (with constant c), we have that BR (x, y) = c [x, y], for all x, y g.
But R is also a solution of the modified YangBaxter equation (with the same constant c) which, combined with (10.36) yields that BR (x, y) = c [x, y], for all x, y g.
Substituted in (10.37), this yields (10.35). Thus, (10.33) and (10.34) sum up to zero,
which proves (10.30), hence { , }Q
R is a (quadratic) Poisson structure.
286
10 RBrackets and rBrackets
Finally, let H be an Adinvariant function on g. We show that its Hamiltonian
vector field is given by the Lax equation (10.29). To do this, recall from Section 5.1.4 that the Adinvariance of H implies that [x H, x] = 0 for all x g, hence
that 12 (xx H + x Hx) = xx H. Therefore, if F F (g) and x g, then
1
[x, x F]  R(x x H + x H x)
2
= [x, x F]  R(x x H) = x F  [R(x x H), x]
= dx F, [R(x x H), x] ,
dx F, (XH )x = {F, H}Q
R (x) =
which proves (10.29).
If, in addition to the hypotheses of Proposition 10.18, R is assumed to be skewsymmetric, then the formula (10.28) for the quadratic bracket takes the simpler form
{F, G}Q
R (x) = R(x F x)  x G x R(x x F)  x x G ,
(10.38)
for all F, G F (g). In the particular case where R comes from an rmatrix r g g,
as explained in Section 10.2.3, this formula takes a particularly simple form. In
order to establish it, we decompose r = s t and use, as in the proof of
Proposition 10.13, the convention that a sum over is implicit in all formulas which
contain . Recall that with this notation,
Rx  y = 4 s  x t  y ,
(10.39)
for all x, y g. For F, G F (g), the bracket (10.38) can be rewritten, using (10.39),
as
{F, G}Q
R (x) = 4 s  x F x t  x G x 4 s  x x F t  x x G
= 4 x F  xs x G  xt 4 x F  s x x G t x
= 4 dx F, xs dx G, xt 4 dx F, s x dx G,t x
= 2 dx F dx G, xs xt s x t x
= 2 dx F dx G, [x x, s t ]
= 2 dx F dx G, x x, r .
We choose a basis (e1 , . . . , ed ) of g, with dual basis (1 , . . . , d ) of g . Then
Q
Q
i , j R (x) = 2 i j , [x x, r ] , so that i , j R (x) is twice the coefficient
of ei e j in [x x, r ]. It follows that, if we define
Q
r
(x) :=
1i< jd
i , j
Q
R
(x) ei e j ,
then we find the following compact formula
Q
, r (x) = 2 x x, r ,
(10.40)
10.3 RBrackets and rBrackets of Higher Degree
287
which is known as the second Russian formula. As in the case of the first Russian
formula, this formula can be thought of in terms of matrices in MatN 2 (F), upon using
the standard algebra isomorphism between MatN (F) MatN (F) and MatN 2 (F).
For the cubic Rbracket, the context is the same as in the case of the quadratic
Rbracket, but the assumptions are weaker.
Proposition 10.19. Let (g, [ , ]) be a finitedimensional Lie algebra, associated
with an associative algebra, equipped with a nondegenerate symmetric bilinear
map  , which satisfies xy  z = x  yz, for all x, y, z g. Suppose that R : g g
is a linear map, which is a solution of the modified YangBaxter equation (10.4).
The bivector field on g, defined for F, G F (g) and x g by
{F, G}CR (x) := [x, x F]  R(x x G x) [x, x G]  R(x x F x) ,
(10.41)
is a cubic Poisson structure on g. If H is an Adinvariant function on g, then its
Hamiltonian vector field with respect to this Poisson structure is given by
x = [R(xx H x), x] .
Proof. The proposition can be proved in the same way as in the case of the quadratic
Rbracket (Proposition 10.18). Actually, the proof of the Jacobi identity for { , }CR
requires less computation, as one does not need to split up R into its symmetric
and skewsymmetric parts and as the formula for the cubic bracket (10.41) has only
half as many terms as the formula for the quadratic bracket (10.28). For linear functions F1 , F2 , F3 F (g), whose gradients (at any point) are denoted by f1 , f2 , f3 , first
compute from (10.41) that at x g one has
x {F1 , F2 }CR = R [x, f1 ] x f2 + f2 xR [x, f1 ] + [ f1 , R(x f2 x)] ( f1 f2 ) ,
which yields, with some extra work,
C
{F1 , F2 }CR , F3 + (1, 2, 3)
R
= x, x {F1 , F2 }CR  R(x f3 x) [x, f3 ]  R xx {F1 , F2 }CR x
+ (1, 2, 3)
= [x, f1 ]  BR (x f2 x, x f3 x) + (1, 2, 3)
= c [x, f1 ]  [x f2 x, x f3 x] + (1, 2, 3)
= 0.
This completes the proof.
To finish this section, we show how the linear, quadratic and cubic Rbrackets are
related. The Lie algebra g is, as above, the Lie algebra of an associative algebra,
with unit e. On g, we can therefore consider the vector fields V0 , V1 and V2 , given at
all x g by
288
10 RBrackets and rBrackets
(V0 )x := e ,
(V1 )x := x ,
(V2 )x := x2 .
For a linear function F on g, this means that
(V0 )x [F] = F(e) ,
(V1 )x [F] = F(x) ,
(V2 )x [F] = F(x2 ) .
for every x g. In particular, V1 is the Euler vector field, which was defined in
Section 8.1.2. For i = 0, 1, 2, we write Li for the Lie derivative with respect to Vi .
Proposition 10.20. Let (g, [ , ]) be the Lie algebra of an associative algebra, with
unit e, which is equipped with a nondegenerate symmetric bilinear map  , which
satisfies xy  z = x  yz, for all x, y, z g. Let R be a solution of the modified Yang
Baxter equation, with constant c F, and suppose that its skewsymmetric part R
is also a solution of the modified YangBaxter equation, with the same constant c.
Then the linear, quadratic and cubic Poisson structures on g, associated with R,
C
namely { , }R,g , { , }Q
R and { , }R , are related by Lie derivatives, as indicated in
the following commutative diagram:
{ , }R,g
L2
{ , }Q
R
L2
L1
L0
{ , }CR
L1
{ , }R,g
1L
2 0
{ , }Q
R
1L
2 0
L2
0
(10.42)
{ , }CR
Moreover, for a fixed such R, the three Poisson structures { , }R,g , { , }Q
R and
C
{ , }R are compatible.
Proof. Since V1 is the Euler vector field, the Lie derivatives of the linear, quadratic
and cubic structures with respect to V1 give a multiple of these Poisson structures, up
to a constant, which is respectively 1, 0 and 1 (see Proposition 8.4). This yields the
two vertical arrows in the diagram. For each of the Lie derivatives L0 and L2 , the
computation is very similar, so we will only do it for one of them, namely we show
that L2 { , }R,g = { , }Q
R . Let F1 , F2 be linear functions on g and let f 1 := F1
and f2 := F2 , as before. According to the formula (3.7) for the Lie derivative of a
biderivation (p = 2), we need to show that
{F1 , F2 }Q
(10.43)
R = V2 {F1 , F2 }R,g + {V2 [F1 ], F2 }R,g + {F1 , V2 [F2 ]}R,g .
To do this, first notice that for a linear function F on g, one has x F  y
= F(y),
independently of x g (so we write f for x F) and V2 [F](x) = F(x2 ) = f  x2 . It
follows that, for all x, y g,
x (V2 [F])  y =
d
F((x + ty)2 ) = F(xy) + F(yx) = x f + f x  y .
dt t=0
10.4 Notes
289
Therefore, for all linear functions F on g, we have
x (V2 [F]) = x f + f x .
One computes similarly that x (V1 [F]) = f and that x (V0 [F]) = 0, but these formulas are only needed for the verification of the other cases. Since {F1 , F2 }R,g is a
linear function on g, the righthand side of (10.43), evaluated at x, is given by
x2  [ f1 , f2 ]R + x  [x (V2 [F1 ]), f2 ]R + x  [ f1 , x (V2 [F2 ])]R
= x2  [ f1 , f2 ]R + x  [x f1 + f1 x, f2 ]R + x  [ f1 , x f2 + f2 x]R
1
1
= [x, f1 ]  R(x f2 + f2 x) [x, f2 ]  R(x f1 + f1 x) ,
2
2
which is precisely {F1 , F2 }Q
R (see (10.28)).
The compatibility of the three Poisson structures follows easily from the relations
in the diagram (10.42) and the properties of the Schouten bracket. Recall from Section 3.3 that, in terms of the Schouten bracket, two Poisson structures and are
compatible if [ , ]S = 0, while the Lie derivative with respect to a vector field V
is expressed as LV = [V , ]S . Clearly, [LV , ]S = [[V , ]S , ]S = 0, since [, ]S
is a coboundary operator (since is a Poisson structure, see Section 4.1.2). Thus,
Q
on the one hand, { , }R,g and { , }Q
R are compatible, on the other hand, { , }R and
{ , }CR are compatible. In order to see that3 { , }R,g and { , }CR are also compatible,
let us denote the three Poisson structures by R , RQ and RC and compute, using the
graded Jacobi identity for the Schouten bracket (Proposition 3.7),
R , RC S = R , V2 , RQ
= V2 , RQ , R
RQ , [R , V2 ]S = 0 ,
S S
S S
where we used in the last step that RQ and R are compatible, and that [V2 , R ]S =
RQ (which is a Poisson structure!).
10.4 Notes
Rmatrices and rmatrices have their origins in the theory of integrable systems,
but also play a fundamental role in the theory of quantum groups. They first appeared in the contexts of the AdlerKostantSymes theorem, exhibiting the Lie algebraic structures which underlie the integrability of both the Toda lattices and the
Kortewegde Vries equation, see Adler [2].
The term quantum group is a generic name which is used for several types
of noncommutative algebras, usually Hopf algebras, i.e., algebras (rather than
groups!) endowed with an algebra structure on the dual. Quantum groups are in
3
In general, the notion is compatible with is not transitive.
290
10 RBrackets and rBrackets
general constructed from solutions of the quantum YangBaxter equation, just as
PoissonLie groups are constructed from solutions of the modified classical Yang
Baxter equation. These constructions are related, since a solution of the Yang
Baxter equation admits an rmatrix as its first order term, while PoissonLie groups
appear as limits of quantum groups. For the theory of quantum groups, we refer to
the classical books of Kassel [103] or Majid [142].
The notion of an rmatrix has been generalized to the general context of Lie
algebroids. The resulting object is called a (classical) dynamical rmatrix; it depends
on parameters, turning some of the conditions which it has to satisfy into differential
equations. See EtingofVarchenko [69, 70] and LiuXu [130].
Chapter 11
PoissonLie Groups
A PoissonLie group (G, ) is a Lie group G which is equipped with a Poisson
structure . The Poisson structure is demanded to be compatible with the group
structure, in the sense that one requires the product map G G G to be a morphism of Poisson manifolds, where G G is equipped with the product Poisson
structure.
To a Lie group G there corresponds a Lie algebra g, which is the tangent space
to G at e, where the Lie bracket [ , ]g on g is inherited from the Lie bracket on the
space of (leftinvariant) vector fields on G. If, in addition, G comes equipped with a
Poisson structure, the Lie algebra g inherits the structure of a Lie coalgebra, which
amounts to a Lie algebra structure [ , ]g on the dual vector space g . In the case
of a PoissonLie group (G, ), the fact that the multiplication in G and the Poisson
structure on G are compatible, implies a compatibility between the two Lie brackets
[ , ]g and [ , ]g : the transpose to [ , ]g is a derivation of the Lie bracket [ , ]g .
Formalizing this property leads to the notion of a Lie bialgebra.
Thus, to every PoissonLie group, there corresponds a Lie bialgebra. Moreover,
to every homomorphism of PoissonLie groups there naturally corresponds a homomorphism of Lie bialgebras, hence the correspondence between PoissonLie groups
and Lie bialgebras is functorial. Most importantly, the converse also holds. As we
know from Lies third theorem, every finitedimensional Lie algebra is the Lie algebra of some Lie group, which can be chosen to be connected and simply connected.
This Lie group can be made into a PoissonLie group if the Lie algebra is a Lie
bialgebra.
PoissonLie groups are introduced in Section 11.1, while their infinitesimal
analogs, Lie bialgebras, are introduced in Section 11.2. The above correspondence
between Lie bialgebras and PoissonLie groups is discussed in Section 11.3. Using dressing actions, we study the symplectic leaves of PoissonLie groups in Section 11.4.
All Lie groups in this chapter are real (F = R) or complex (F = C).
C. LaurentGengoux et al., Poisson Structures,
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/9783642310904 11, SpringerVerlag Berlin Heidelberg 2013
291
292
11 PoissonLie Groups
11.1 Multiplicative Poisson Structures and PoissonLie Groups
A Poisson structure on a Lie group G makes G into a Poisson manifold. For reasons
which will be given later in this chapter, one demands the following compatibility
relation between the Poisson structure on G and the group structure of G.
Definition 11.1. A Poisson structure on a (real or complex) Lie group G is said
to be multiplicative if the product map : G G G is a Poisson map, where
G G is endowed with the product Poisson structure. The pair (G, ) is then called
a PoissonLie group.
A map : G H between two PoissonLie groups (G, ) and (H, ) is called
a PoissonLie group homomorphism if it is both a Poisson map and a group homomorphism.
11.1.1 The Condition of Multiplicativity
We give in the following proposition two useful characterizations of the multiplicativity of a Poisson structure on a Lie group. Recall that for g G the maps of left
and right translation Lg and Rg in G are defined by Lg (h) := gh and Rg (h) := hg, for
all h G.
Proposition 11.2. Let G be a Lie group. For a Poisson structure on G, the following three conditions are equivalent:
(i)
(ii)
is multiplicative;
For all g, h G:
gh = 2 (Tg Rh ) g + 2 (Th Lg ) h ;
(iii)
(11.1)
The map : G 2 g, which is defined for all g G by
(g) := 2 (Tg Rg1 ) g ,
(11.2)
is a cocycle of G, with respect to the adjoint representation of G on 2 g, i.e.,
(gh) = (g) + Adg (h) ,
(11.3)
for all g, h G.
Proof. We first prove that (i) and (ii) are equivalent, which amounts to proving that
: G G G is a Poisson map if and only if (11.1) holds for all g, h G. Let us
denote the product Poisson structure on G G by ; according to (2.11), the value
of at (g, h) G G is given by the bivector
(g,h) = 2 (Tg h ) g + 2 (Th g ) h ,
(11.4)
11.1 Multiplicative Poisson Structures and PoissonLie Groups
293
where h and g are the inclusions of G into G G given by h : g (g, h) and
g : h (g, h) respectively. According to (1.32), is a Poisson map if and only if
gh = 2 (T(g,h) ) (g,h) ,
(11.5)
for all g, h G. According to (11.4), and using the identities h = Rh and
g = Lg , the righthand side of (11.5) is given by
2 (T(g,h) ) (g,h) = 2 (T(g,h) ) (2 (Tg h ) g ) + 2 (T(g,h) ) (2 (Th g ) h )
= 2 (Tg ( h )) g + 2 (Th ( g )) h
= 2 (Tg Rh ) g + 2 (Th Lg ) h ,
which is precisely the righthand side of (11.1). This proves that (i) and (ii) are
equivalent.
In order to show that (ii) and (iii) are equivalent, we fix g, h G and we apply
the isomorphism 2 (Tgh R(gh)1 ) to both sides of (11.1), giving
2 (Tgh R(gh)1 ) gh
= 2 (Tgh R(gh)1 ) 2 (Tg Rh ) g + 2 (Tgh R(gh)1 ) 2 (Th Lg ) h ,
and we show that the latter equation is, term by term, equation (11.3). For the first
two terms, this follows from the definition of and from the identity
R(gh)1 Rh = Rg1 , namely
2 (Tgh R(gh)1 ) gh = (gh) ,
2 (Tgh R(gh)1 )(2 (Tg Rh ) g ) = 2 (Tg Rg1 ) g = (g) .
For the third and final term, recall that the map Cg is conjugation by g, so that
Cg = Rg1 Lg , leading to the identity R(gh)1 Lg = Cg Rh1 , which in turn yields
2 (Tgh R(gh)1 )(2 (Th Lg ) h ) = 2 (Th (R(gh)1 Lg )) h
= 2 (Th (Cg Rh1 )) h = 2 (TeCg )(2 (Th Rh1 ) h )
= Adg (h) ,
where we have used in the last step that (TeCg )(x) = Adg x for all x g and g G.
This proves that condition (11.1) is equivalent to condition (11.3).
In item (iii) of Proposition 11.2, we used the term cocycle, which is borrowed from
the terminology used when introducing the concept of cohomology for Lie groups.
We will not develop Lie group cohomology in this book, we rather restrict ourselves
to giving the general definition of a cocycle of a group (strictly speaking a 1cocycle)
and proving the one result on group cocycles which we will use.
294
11 PoissonLie Groups
Definition 11.3. Let : G V V be a representation of a Lie group G on a finitedimensional vector space V . A cocycle of G in V is a smooth (or holomorphic)
function : G V , satisfying
(gh) = (g) + (g, (h)) ,
(11.6)
for all g, h G.
Proposition 11.4. Let G be a connected Lie group, let : G V V be a representation of G on a finitedimensional vector space V and let W V be a Ginvariant
subspace. Suppose that : G V is a cocycle of G in V .
(1)
(2)
If Te (x) = 0 for all x g, then = 0;
If Te (x) W for all x g, then (g) W for all g G.
Proof. The tangent map of the cocycle relation (11.6) with respect to the variable h
at the point h = e is given by
Tg Te Lg = g Te ,
so that the vanishing of Te implies the vanishing of Tg Te Lg , and therefore
of Tg , for all g G. Since G is assumed to be connected, this implies that is
a constant function. Since (e) = 0, as follows from (11.6), vanishes at each
point, which shows (1). If W is a Ginvariant subspace, then we have an induced
representation G V /W V /W and induces a cocycle of G in V /W . In view
of (1), the latter cocycle is zero as soon as its tangent map at e is zero, i.e., as soon
as Te (x) W for all x g, which proves (2).
See [181] for more information on group cohomology.
11.1.2 Basic Properties of PoissonLie Groups
We start with an interesting feature of PoissonLie groups. Let (G, ) be a Poisson
Lie group with Lie algebra g. The Poisson structure vanishes at the unit e of G,
hence leads to a linear Poisson structure on g, i.e., a Lie bracket on g . This structure
will play a mayor role in the sequel of this chapter.
Proposition 11.5. Let (G, ) be a PoissonLie group.
(1)
(2)
(3)
vanishes at e;
The linearized Poisson structure of at e is the linear Poisson structure 1 on
g = Te G, whose value at x g is given by (1 )x = Te (x), where : G 2 g
is the cocycle, defined as in (11.2);
The inverse map inv : G G is an antiPoisson map.
Proof. Substituting e for g and h in (11.1) amounts to e = e + e , hence e = 0,
which yields (1). According to Proposition 7.22, g = Te G inherits from = { , }
11.1 Multiplicative Poisson Structures and PoissonLie Groups
295
a linear Poisson structure 1 = { , }1 . For functions F, G on g, defined in a neighborhood of the origin o g, the linearized Poisson bracket {F, G}1 and the original
Poisson bracket {F, G} are, according to Proposition 7.22, related for every x g
by1
de {F, G} , x
= {de F, de G}1 (x) = de F de G, (1 )x
,
(11.7)
where de F, de G and de {F, G} are viewed as (linear) functions on g = Te G, and (1 )x
is identified with an element of 2 g. Alternatively, in terms of the map : G 2 g
defined in (11.2), the Poisson bracket {F, G} at g G can be written as
{F, G} (g) = dg F dg G, g = dg F dg G, 2 (Te Rg ) (g) ,
which yields, upon putting g := exp(tx) and taking the derivative with respect to t
at t = 0,
de {F, G} , x
= de F de G, Te (x)
,
(11.8)
where we have used that (e) = 0. Comparing (11.7) with (11.8) leads to the desired
equality
(1 )x = Te (x) ,
(11.9)
proving (2). To say that inv is an antiPoisson map means that inv : (G, )
(G, ) is a Poisson map: for every g G, 2 (Tg inv) g = g1 . To prove that
the latter holds, let g G, substitute g1 for h in (11.1), and use e = 0, to find that
0 = 2 (Tg Rg1 ) g + 2 (Tg1 Lg ) g1 .
Combined with Tg inv = Te Lg1 Tg Rg1 (see (5.2)), this leads to
2 (Tg inv) g = 2 (Te Lg1 )(2 (Tg Rg1 ) g )
= 2 (Te Lg1 )(2 (Tg1 Lg ) g1 )
= 2 (Tg1 (Lg1 Lg )) g1
= g1 ,
which proves that inv is an antiPoisson map.
Comparing Definitions 5.32 and 11.1, we see that a Poisson structure on a Lie
group G is multiplicative if and only if left translation L : G G G is a Poisson
action. We show in the following corollary of Proposition 11.2 that this condition
does not imply, in general, that the map Lg (left translation by a given element g G)
is, for every g G, a Poisson map.
Proposition 11.6. Let (G, ) be a PoissonLie group. For every g G, the following
conditions are equivalent:
(i)
(ii)
1
g = 0;
The map Lg : G G, left translation by g, is a Poisson map;
We use here and below the natural pairing ,
between p g and p g, introduced in (5.10).
296
(iii)
11 PoissonLie Groups
The map Rg : G G, right translation by g, is a Poisson map.
Proof. For g G, left translation by g is a Poisson map if and only if, for all h G,
gh = 2 (Th Lg ) h .
(11.10)
In view of (11.1), which says that is multiplicative, (11.10) is tantamount to
2 (Tg Rh ) g = 0, which is equivalent to g = 0, because the linear map Tg Rh is
an isomorphism. This proves the equivalence of (i) and (ii). The proof of the equivalence of (i) and (iii) is similar.
11.1.3 PoissonLie Subgroups
We consider in this section the notion of a PoissonLie subgroup of a PoissonLie
group (G, ). Recall that a Lie subgroup of G need not be a closed subset of G and
that there is associated to a map : G 2 g, defined for g G by (g) :=
2 (Tg Rg1 ) g .
Proposition 11.7. Let (G, ) be a PoissonLie group. For a Lie subgroup H of G,
with Lie algebra h, the following conditions are equivalent:
(i)
(ii)
H is a Poisson submanifold of (G, );
For every h H, the bivector (h) belongs to 2 h.
Under these conditions, the pair (H, ) is a PoissonLie group, where denotes
the restriction of to H. One then says that H is a PoissonLie subgroup of G.
When H is connected, the conditions (i) and (ii) are also equivalent to the following
condition:
(iii)
For every x h, the bivector Te (x) belongs to 2 h.
Proof. In view of Proposition 2.12, H is a Poisson submanifold of G if and only if
h 2 Th H for every h H. Since 2 Th H = 2 (Te Rh ) (2 h), the equivalence of (i)
and (ii) follows.
Suppose that H is a Poisson submanifold of G and consider the product map
on H. It is the restriction of the product map on G to H, hence it is a Poisson map.
It follows that H itself is also a PoissonLie group.
Suppose now that H is connected. It is clear that (ii) implies (iii). Let us show
the inverse implication. Since : G 2 g is a cocycle for the adjoint representation of G in 2 g, its restriction to H is a cocycle for the adjoint representation of H in 2 g, which admits 2 h 2 g as an Hinvariant subspace. By assumption, Te (x) 2 h for all x h. According to Proposition 11.4, this implies
that (h) 2 h for all h H, which is condition (ii).
We show in the following proposition that the datum of a PoissonLie group and of a
PoissonLie subgroup leads, under a topological assumption, to a Poisson structure
on the (left) quotient space.
11.1 Multiplicative Poisson Structures and PoissonLie Groups
297
Proposition 11.8. Let (G, ) be a PoissonLie group and let (H, ) be a Poisson
Lie subgroup of G. Suppose that H is a closed subgroup of G, so that the left coset
space H \G is a smooth manifold. There exists a unique Poisson structure on H \G,
such that the canonical projection p : G H\G is a Poisson map. Moreover, the
right action of G on H\G is a (right) Poisson action.
Proof. Consider the left action of H on G which is the restriction of the product
map on G. Since H is a Poisson submanifold of G, the product H G is a Poisson
submanifold of G G, hence the left action of H on G is a Poisson action. Since
the quotient space of this action is precisely H\G, according to Proposition 5.33,
the quotient space H\G inherits a unique Poisson structure from (G, ) such that
the canonical projection p : G H\G is a Poisson map. Since the product map
: G G G is a Poisson map, the induced map (H\G) G (H\G) is also a
Poisson map, hence the (right) action of G on H\G is a Poisson action.
11.1.4 Linear Multiplicative Poisson Structures
Let V be a finitedimensional vector space (over R or C). Addition of vectors in V
defines a group structure on V which makes V into an abelian Lie group. Using
Proposition 11.2, we characterize multiplicative Poisson structures on V as linear
Poisson structures on V .
Proposition 11.9. Let V be a finitedimensional vector space. A Poisson structure
on the Lie group (V, +) is multiplicative if and only if it is a linear Poisson structure.
Proof. Recall that a Poisson structure = { , } on V is linear if and only if the
Poisson bracket of every pair of linear functions on V is a linear function on V (see
Section 7.1). By continuity, a real smooth (or complex holomorphic) function H
on V , which satisfies
H(v + w) = H(v) + H(w) ,
for all v, w V , is a linear function on V . Therefore, is a linear Poisson structure
on V , if and only if
{F, G} (v + w) = {F, G} (v) + {F, G} (w) ,
(11.11)
for all v, w V and for all linear functions F, G V . We have on the one hand that
{F, G} (v + w) = dv+w F dv+w G, v+w
(11.12)
and on the other hand, since F and F Rw differ by a constant (namely F(w)), that
{F, G} (v) = dv F dv G, v
= dv (F Rw ) dv (G Rw ), v
= dv+w F dv+w G, 2 (Tv Rw ) v .
(11.13)
298
11 PoissonLie Groups
Substituting (11.12) and (11.13) in (11.11) we conclude that is a linear Poisson
structure on V if and only if
v+w = 2 (Tv Rw ) v + 2 (Tw Lv ) w ,
for all v, w V . Since the latter condition is precisely condition (ii) in Proposition 11.2, this shows that is linear if and only if is multiplicative.
Since the dual of a finitedimensional Lie algebra admits a canonical linear Poisson structure (the LiePoisson structure), it follows from the above proposition that
the dual of a finitedimensional Lie algebra is in a canonical way a PoissonLie
group.
11.1.5 Coboundary PoissonLie Groups
An important class of PoissonLie groups are those constructed out of rmatrices.
Recall from Proposition 10.11 that if g is a Lie algebra and r is an rmatrix for g,
then its skewsymmetric part a := r 2 g is such that [[a, a]] is adinvariant. We
show in the following proposition that this condition admits a natural connection
with PoissonLie groups. Recall from Section 5.1 that if G is a Lie group with
Lie algebra g, we denote for x g by
x (respectively
x ) the rightinvariant (respectively leftinvariant) vector field on G, whose value at e is x, and similarly for
elements X g.
Proposition 11.10. Let G be a Lie group with Lie algebra (g, [ , ]) and let a 2 g.
On G we consider the bivector field , defined by :=
a
a . The pair (G, ) is
a PoissonLie group if and only if [[a, a]] is an Adinvariant element. In particular,
if G is connected, then (G, ) is a PoissonLie group if and only if a is an rmatrix.
Proof. It follows immediately from Proposition 5.2 that the Schouten bracket of
:=
a
a with itself is given by
a
a ,
a
a ]S
[ , ]S = [
= [ a , a ]S [ a ,
a ]S [
a ,
a ]S + [
a ,
a ]S
= [ a , a ]S + [ a , a ]S
= [[a, a]] [[a, a]] .
In view of item (3) in Proposition 5.3, we can conclude that , as defined above, is
a Poisson structure on G if and only if [[a, a]] is Adinvariant. It remains to be shown
that is moreover multiplicative. The map : G 2 g associated with as in
item (iii) in Proposition 11.2 (i.e. defined, for all g G, by (g) := 2 (Tg Rg1 ) g )
is given by
(g) = Adg a a ,
(11.14)
11.1 Multiplicative Poisson Structures and PoissonLie Groups
299
as one computes using the following description of the bivector field
a
a at a
point g G:
g = 2 (Te Lg )a 2 (Te Rg )a .
For every g, h G, we can therefore check that
(gh) = Adgh a a = Adgh a Adg a + Adg a a = (g) + Adg (h) , (11.15)
which, in view of the equivalence of items (i) and (iii) in Proposition 11.2, implies
that is multiplicative. Therefore, (G, ) is a PoissonLie group if and only if
[[a, a]] is Adinvariant. If G is connected, then this is, according to Proposition 5.3,
equivalent to [[a, a]] being adinvariant, i.e., to a being an rmatrix.
The previous proposition leads to the following definition.
Definition 11.11. A PoissonLie group (G, ) is said to be a coboundary Poisson
a
a for some element a 2 g.
Lie group if is of the form =
For a given coboundary PoissonLie group (G, ) there may exist two different
a1
a1 and =
a2
a2 . However, in such a
elements a1 , a2 2 g such that =
case, we have
a
1 a2 = a 1 a2
so that a1 and a2 differ by an Adinvariant element in 2 g. Therefore, for a
given coboundary PoissonLie group, the element in 2 g which appears in Definition 11.11 is defined only up to addition of an Adinvariant element in 2 g.
Remark 11.12. The expression (11.14) of in terms of a, namely that (g) =
Adg a a for all g G, has the following cohomological interpretation: it says that
is a 1coboundary (in fact, of a) for the chain complex which underlies the group
cohomology of G with coefficients in 2 g. It explains the terminology coboundary
PoissonLie group, introduced above. Also, since every 1coboundary is a cocycle,
the cohomological interpretation of yields an alternative proof of (11.15), which
shows that is a cocycle.
11.1.6 Multiplicative Poisson Structures on Vector Spaces
In this section, we explain how the notion of a multiplicative Poisson structure on
a Lie group G is related to the notion of a multiplicative Poisson structure on a
finitedimensional vector space V , equipped with a product , a notion which was
introduced in Section 8.2.1. For the present discussion, we will assume that the
product is associative and admits a unit e; we will therefore refer to (V, ) as an
associative algebra with unit. An element v V is said to be invertible if it admits a
right inverse, i.e., there exists w V such that (v, w) = e.
Proposition 11.13. Let (V, ) be a finitedimensional associative algebra with unit,
and let Vinv denote the set of all invertible elements in V .
300
11 PoissonLie Groups
(1)
Vinv is a (nonempty) Zariski open subset of V and the restriction of to Vinv
defines the structure of a Lie group on Vinv , whose Lie algebra is isomorphic
to V , with Lie bracket given by [x, y] = (x, y) (y, x) for all x, y V ;
If is a multiplicative Poisson structure on (V, ), in the sense of Section 8.2.1, then the restriction of to Vinv is a multiplicative Poisson structure on the Lie group Vinv , making (Vinv , ) into a PoissonLie group;
Let a 2V be a skewsymmetric rmatrix for V . The quadratic bivector
field on V , whose value at x V is given by
(2)
(3)
x := 2 [x x, a]
(11.16)
is a multiplicative Poisson structure on V . The PoissonLie group (Vinv , ),
where denotes the restriction of to Vinv , is the coboundary PoissonLie
a
a )).
group (Vinv , 2(
Proof. Since is assumed to be associative, the invertible elements are precisely the
elements v V for which the linear map Lv : V V , defined by Lv (w) := (v, w),
is an isomorphism, i.e.,
Vinv = {v V  det Lv = 0} .
Since is a bilinear map, v Lv is a linear map, and hence v det Lv is a polynomial function on V (it is homogeneous of degree dimV ). It follows that Vinv is a
Zariski open subset of V ; it is nonempty since it contains the unit e of . Defining
Rv : V V by Rv (w) := (w, v), we have that Rv is an isomorphism if and only if Lv
is an isomorphism. Indeed, if Lv is an isomorphism, so that there exists w V with
(v, w) = e, then we have, in view of the associativity of ,
( (w, v), w) = (w, (v, w)) = w ,
so that Ker Rv = {0}, i.e., Rv is an isomorphism (recall that V is finitedimensional).
Therefore, the invertible elements of (V, ) form a group, with e as unit. As an
open subset of V , the group (Vinv , ) inherits a manifold structure from V ; since
is a bilinear map, it restricts to a smooth map on Vinv , hence (Vinv , ) is a Lie
group. For x, y V TeV , we need to show that their Lie bracket [x, y] is given by
(x, y) (y, x). Under the canonical isomorphism V TeV we have, by definition,
[x, y] = [
x ,
y ]e and ( (x, y) (y, x))e = (x, y) (y, x). It therefore suffices to
show that
(11.17)
[
x ,
y ] = (x, y) (y, x) .
Let F : V F be a linear function. For v V we have that
F(v + t (v, x)) F(v)
x [F](v) = xe [F Lv ] = lim
= F( (v, x)),
t0
t
so that
x [F] = F Rx . It follows that
11.1 Multiplicative Poisson Structures and PoissonLie Groups
301
[
x ,
y ] [F] =
x [
y [F]]
y [
x [F]]
= F (Ry Rx Rx Ry )
= F R (x,y) (y,x)
= ( (x, y) (y, x))[F],
for all linear functions F on V , leading to (11.17). This shows (1).
Let be a multiplicative Poisson structure on (V, ). Recall from Section 8.2.1
that this means that the product map is a Poisson map. Since Vinv is an open subset
of V , the restriction of to Vinv is a Poisson structure and the restriction of
to Vinv , i.e., the product map of the Lie group (Vinv , ), is a Poisson map. This shows
that (Vinv , ) is a PoissonLie group, which is the content of (2).
Let a be a skewsymmetric rmatrix for V , which is the Lie algebra of the
a
a ) is
connected Lie group (Vinv , ). According to Proposition 11.10, (Vinv ,
a coboundary PoissonLie group. We show that the quadratic bivector field2
on V , defined by (11.16), takes at every point of Vinv the same value as the bivec
tor field 2(
a
a ). To do this, we identify all tangent spaces to V with V , so that
the derivative of left translation by x Vinv is the linear map Te Lx : V V , given
for z V by Te Lx (z) = xz, since Lx is a linear map. Thus, the leftinvariant vector
z x = xz. Similarly
z x = zx. It follows that,
field
z of z V is given at x Vinv , by
2
for a bivector z1 z2 V ,
z1 z2 x = xz1 xz2 ,
z1 z2 x = z1 x z2 x ,
for all x Vinv , in particular
a x
a x = (x x)a a(x x) = [x x, a] .
(11.18)
a
a ) have the same value at every point
This shows our claim. Since and 2(
of Vinv , the restriction of to Vinv is multiplicative, hence itself is multiplicative,
showing (3).
Example 11.14. We show that the bivector field on V = Matd (F), which we discussed in Example 8.22 is associated with an rmatrix, which proves that it is a multiplicative Poisson structure. Let V be equipped with the nondegenerate adinvariant
bilinear form x  y
= Trace(xy). We denote the elementary matrices of V by Ei j ,
and we introduce
linear coordinates i j on V by defining, for a matrix x Matd (F),
i j (x) := E ji  x , so that i j (x) = xi j , when x is written as (xi j )1i, jd . An arbitrary
matrix can be decomposed as the sum of a strictly upper triangular, a diagonal, and
a strictly lower triangular matrix: this decomposition satisfies the requirements of
Example 10.15 so that
1
(11.19)
r :=
Eab Eba
2 1a<bd
We know from Section 10.3, in particular from the second Russian formula (10.40), that is a
Poisson structure, but this fact is not used in the proof.
302
11 PoissonLie Groups
is an rmatrix on the Lie algebra Matd (F). With the help of this rmatrix, a Poisson structure on Matd (F) can be constructed as in Proposition 11.13. Let us express the coefficients of the Poisson matrix of with respect to the linear coordinates i j . For all i, j, k, = 1, . . . , d, and all x Matd (F), we find using the explicit
formula (11.19) for r and (11.18) for the corresponding Poisson structure,
i j , k r (x) = 2 i j kl , [x x, r]
= i j k , [x x, Eab Eba ]
a<b
i j k , xEab xEba Eab x Eba x
a<b
( j,b ,a j,a ,b i,a k,b + i,b k,a ) i (x)k j (x)
a<b
= ( j, + i,k ) i (x)k j (x) ,
where i, j is 1 if i > j, is 1 if i < j and is 0 if i = j. It is the quadratic Poisson
structure which we met in Example 8.22.
11.2 Lie Bialgebras
In this section, we consider Lie algebra structures on a finitedimensional vector
space g, on its dual vector space g and on their product d := g g . We will often
view g and g as subspaces of d, via the natural identifications g g {0} and
g {0} g . By a slight abuse of notation, (x, 0) and (0, ) are often abbreviated
to x and . We will make a notational distinction between the three Lie brackets
which will be considered: we will write [ , ]g , [ , ]g , and [ , ]d for the Lie bracket
on g, on g and on d respectively. We will however use the same notation ad for the
coadjoint actions of g on g and of g on g. In view of the definition of the coadjoint
action (see Section 5.1.3),
, [x, y]g = adx , y
and
[ , ]g , x = , ad x ,
(11.20)
for all x, y g and for all , g .
We will find it often convenient to view the Lie algebra structure [ , ]g on g as
a structure on g. This is done by taking the transpose of [ , ]g : g g g , which
can be considered as a linear map : g g g. Precisely, is defined by
, (x)
= [ , ]g , x ,
(11.21)
for all x g, and for all , g . We will use the natural bilinear form 
d on d,
defined for all x, y g and for all , g by
11.2 Lie Bialgebras
303
(x, )  (y, )
d := , y
+ , x
.
(11.22)
It is clear that 
d is symmetric and nondegenerate.
11.2.1 Lie Bialgebras
The following proposition will justify the definition of a Lie bialgebra.
Proposition 11.15. Let g be a finitedimensional vector space. Suppose that [ , ]g
and [ , ]g are Lie brackets on g and g respectively. On d := g g , consider the
nondegenerate symmetric bilinear form 
d , given by (11.22). Then the following
conditions are equivalent:
(i)
(ii)
There exists a Lie bracket on d, with respect to which 
d is adinvariant,
such that the natural inclusions g d and g d are Lie algebra homomorphisms;
The skewsymmetric bilinear map [ , ]d : d d d given, for all x, y g and
for all , g , by
[(x, ), (y, )]d := ([x, y]g +ad yad x, [ , ]g +adx ady ) , (11.23)
(iii)
satisfies the Jacobi identity;
The transpose : g 2 g of the Lie bracket [ , ]g : 2 g g , defined by
(11.21), satisfies for x, y g the derivation property
[x, y]g = [[ (x), y]] + [[x, (y)]] .
(11.24)
If any of these three equivalent conditions is satisfied, the Lie bracket on d, which
satisfies the conditions demanded in (i), is unique and is given by (11.23); moreover,
it makes (d, 
d ) into a quadratic Lie algebra.
The proof of this proposition will make use of the following lemma.
Lemma 11.16. Let g be a finitedimensional vector space, equipped with Lie brackets [ , ]g and [ , ]g on g, respectively on g . The skewsymmetric bilinear map
[ , ]d : d d d, defined in (11.23), has the following properties:
(1)
For all d1 , d2 , d3 d:
d1  [d2 , d3 ]d
d = [d1 , d2 ]d  d3
d ;
(2)
(11.25)
[ , ]d satisfies the Jacobi identity if and only if for all x, y g d, and for all
, g d:
[x, y]d  [ , ]d
d + [y, ]d  [x, ]d
d + [ , x]d  [y, ]d
d = 0 .
(11.26)
Proof. Let di = (xi , i ), for i = 1, 2, 3. In view of the definition of [ , ]d , of 
d
and of the coadjoint actions of g on g and of g on g,
304
11 PoissonLie Groups
d1  [d2 , d3 ]d
d = 1 , [x2 , x3 ]g
+ [1 , 2 ]g , x3 + (1, 2, 3) .
Since the latter expression is invariant with respect to a cyclic permutation of the
indices 1, 2, 3, it follows that
d1  [d2 , d3 ]d
d = d3  [d1 , d2 ]d
d = [d1 , d2 ]d  d3
d ,
which proves (1). Let denote the 4linear map from d to itself, defined for all
d1 , . . . , d4 d by
(d1 , d2 , d3 , d4 ) := Jd (d1 , d2 , d3 )  d4
d ,
where Jd is the Jacobiator of [ , ]d , i.e., Jd is given, for all d1 , d2 , d3 d, by
Jd (d1 , d2 , d3 ) := [[d1 , d2 ]d , d3 ]d + [[d2 , d3 ]d , d1 ]d + [[d3 , d1 ]d , d2 ]d .
(11.27)
Since Jd = 0 if and only if [ , ]d satisfies the Jacobi identity, and since 
d is
nondegenerate, = 0 if and only if [ , ]d satisfies the Jacobi identity.
In order to prove (2), we therefore only need to show that (11.26) holds for all
x, y g and all , g , if and only if = 0. Since Jd (d1 , d2 , d3 ) = 0 when
d1 , d2 , d3 all belong to g or all belong to g , we have that (d1 , d2 , d3 , d4 ) = 0 when
d1 , d2 , d3 all belong to g or all belong to g , whatever the value of d4 d. The
skewsymmetry of Jd , combined with (11.25), implies that is skewsymmetric.
Therefore, (d1 , d2 , d3 , d4 ) = 0 as soon as at least three of the di all belong to g,
or all belong to g . It follows, since is 4linear, that = 0 if and only if
(d1 , d2 , d3 , d4 ) = 0, whenever two of the di belong to g and the two other belong
to g ; but that is, in view of (11.25), precisely the condition that (11.26) holds for
all x, y g and all , g . This establishes the proof of (2).
We are now ready to give the proof of Proposition 11.15.
Proof. We first prove that (i) and (ii) are equivalent. Suppose first that (ii) is satisfied. Then the Lie bracket, defined by the explicit formula (11.23) has in view
of (1) in Lemma 11.16 the properties announced in (i). Suppose next that there exists a Lie bracket [ , ] on d, with the properties announced in item (i). We show that
[ , ] = [ , ]d , i.e., [ , ] is given by the explicit formula (11.23). In view of (i), [ , ]
and [ , ]d agree when both arguments are taken either from g or from g , so we only
need to show that
(11.28)
[x, ] = ( ad x, adx ) ,
for all x g and all g . For (z, ) g g , we have in view of (11.25), combined
with the properties of [ , ], assumed in (i), that
[x, ]  (z, )
d = [(x, 0), (0, )]  (z, 0) + (0, )
d
=
[z, x]g , 0  (0, ) + 0, [ , ]g  (x, 0)
d
d
= , [z, x]g + [ , ]g , x
11.2 Lie Bialgebras
305
= ( ad x, adx )  (z, ) d .
Since 
d is nondegenerate, this shows (11.28), hence that if there exists a Lie
bracket on d, satisfying (i), then it is given by (11.23). In particular, it shows that (i)
and (ii) are equivalent.
We now prove the equivalence of items (ii) and (iii). In view of item (2) of
Lemma 11.16, we need to show that (11.26) holds for all x, y g, and for all
, g , if and only if (11.24) holds for all x, y g. We claim that, to do this,
it suffices to show that the following formulas hold, for all x, y g, and for all
, g :
(11.29)
, [[ (x), y]]
= ady , ad x ady , ad x .
Indeed, using (11.21 ), (11.29) and the definitions of [ , ]d and 
d , we find that
, ([x, y]g ) [[ (x), y]] [[x, (y)]]
= [ , ]g , [x, y]g ady , ad x + ady , ad x
+ adx , ad y adx , ad y
= [x, y]d  [ , ]d
d + [y, ]d  [x, ]d
d + [ , x]d  [y, ]d
d ,
for all x, y g, and for all , g . In order to show (11.29), use (5.13) and the
derivation property of the coadjoint action of g on 2 g :
, [[ (x), y]]
= , ady (x) = ady ( ), (x)
= ady + ady , (x)
= ady , g + , ady g , x
= ady , ad x ady , ad x .
This proves (11.29) and hence also the equivalence of (ii) and (iii).
Proposition 11.15 motivates the following definition.
Definition 11.17. Let g be a finitedimensional vector space, let [ , ]g be a Lie
bracket on g and let [ , ]g be a Lie bracket on g . The triple (g, [ , ]g , [ , ]g ) is
said to be a Lie bialgebra if it satisfies any one of the equivalent conditions given
in Proposition 11.15. In this case, the Lie algebra (d, [ , ]d ), where [ , ]d is the Lie
bracket on d := g g , given in (11.23), is called the double of the Lie bialgebra
(g, [ , ]g , [ , ]g ).
Suppose that (g, [ , ]g , [ , ]g ) and (h, [ , ]h , [ , ]h ) are two (finitedimensional)
Lie bialgebras. A linear map : g h is called a homomorphism of Lie bialgebras
if both : (g, [ , ]g ) (h, [ , ]h ) and its transpose : (h , [ , ]h ) (g , [ , ]g )
are Lie algebra homomorphisms.
306
11 PoissonLie Groups
Example 11.18. Let (g, [ , ]g ) be a finitedimensional Lie algebra. When g is
equipped with the trivial Lie bracket ([ , ]g := 0), then the triple (g, [ , ]g , [ , ]g ) is
a Lie biagebra, since item (iii) in Proposition 11.15 is trivially satisfied when = 0.
Dually, let V be any finitedimensional vector space, whose dual vector space V is
equipped with a Lie bracket [ , ]V . Making V in a trivial way into a Lie algebra by
putting [ , ]V := 0, the triple (V, [ , ]V , [ , ]V ) is a Lie bialgebra.
Example 11.19. Let g be a vector space of dimension two and let [ , ]g and [ , ]g
be Lie brackets on g and on g respectively. The triple (g, [ , ]g , [ , ]g ) is a Lie bialgebra. In view of Example 11.18, this is clear when at least one of the Lie brackets
[ , ]g or [ , ]g is the trivial bracket. Suppose therefore that none of these brackets is
trivial and denote by V , respectively W the image of the maps [ , ]g : 2 g g
and [ , ]g : 2 g g . Both V and W are onedimensional because g is twodimensional. If W is the annihilator of V , i.e., W,V
= 0, there exists a basis (x, y)
of g in which
[x, y]g = x and [x , y ]g = y ,
where (x , y ) is the basis of g which is dual to (x, y). Then (x) = 0 and (y) =
x y and one sees that condition (iii) in Proposition 11.15 is satisfied. Otherwise,
there exists a basis (x, y) of g in which
[x, y]g = x
and
[x , y ]g = x ,
(11.30)
for some F . In this case (x) = x y and (y) = 0, so that we arrive at the
same conclusion.
Remark 11.20. For every , F and every Lie bialgebra (g, [ , ]g , [ , ]g ), the
triple (g, [ , ]g , [ , ]g ) is a Lie bialgebra. It is clear that it is isomorphic to the
original one if = F , under the isomorphism 1g : g g. However, in general,
it is not the case when = , even if both and are different from 0. For instance,
for two different values of the parameter , the Lie bialgebra structures described
by (11.30) are not isomorphic.
11.2.2 Lie Subbialgebras
The notion of a homomorphism of Lie bialgebras, given in Definition 11.17, leads
naturally to the notion of a Lie subbialgebra.
Definition 11.21. Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra. A subspace h of g is called a Lie subbialgebra if h admits the structure of a Lie bialgebra,
such that the inclusion h g is a homomorphism of Lie bialgebras.
It is clear that the Lie bialgebra structure on a Lie subalgebra is completely determined by the Lie bialgebra structure on the ambient Lie bialgebra. The fact that the
transpose of the inclusion map h g needs to be a Lie algebra homomorphism,
admits a natural reformulation, which leads to the following proposition.
11.2 Lie Bialgebras
307
Proposition 11.22. Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra. A
subspace h of g is a Lie subbialgebra of (g, [ , ]g , [ , ]g ) if and only if the following
two conditions are satisfied:
(1)
(2)
The vector space h is a Lie subalgebra of (g, [ , ]g );
The annihilator h of h is a Lie ideal of (g , [ , ]g ).
Proof. For h a subspace of g, let : h g denote the inclusion map, with transpose
: g h . If a Lie subbialgebra structure on h exists, making the inclusion map
into a homomorphism of Lie bialgebras, then the Lie bracket on h is the restriction
of the Lie bracket on g (and condition (1) is satisfied); also, : g h is a Lie
algebra homomorphism, so that its kernel h is a Lie ideal of g (and condition (2)
is satisfied).
Conversely, if the two conditions are satisfied, the Lie brackets on g and on g
induce Lie brackets on h (by restriction) and on h (by taking the quotient with
respect to h ), hence equip h and h with two Lie brackets [ , ]h and [ , ]h . Since
the transpose of [ , ]h is the restriction of to h, item (iii) in Proposition 11.15
implies that these brackets define a Lie bialgebra structure on h. By construction,
is a homomorphism of Lie bialgebras.
11.2.3 Duality for Lie Bialgebras
We now come to the notion of duality for Lie bialgebras.
Proposition 11.23. Let (g, [ , ]g , [ , ]g ) be a Lie bialgebra, where g is a finitedimensional vector space. The triple (g , [ , ]g , [ , ]g ) is a Lie bialgebra, called the
dual of the Lie bialgebra (g, [ , ]g , [ , ]g ). The natural isomorphism S : g g
g g, given for all (x, ) g g by S(x, ) := ( , x) is a Lie algebra isomorphism
of the doubles of these two Lie bialgebras.
Proof. In item (i) in Proposition 11.15, the Lie algebras (g, [ , ]g ) and (g , [ , ]g )
play a symmetric role, hence the first part of the proposition is clear. The isomorphism S : g g g g which sends (x, ) to ( , x) is then clearly a Lie algebra isomorphism between the doubles of the Lie bialgebras (g, [ , ]g , [ , ]g ) and
(g , [ , ]g , [ , ]g ), which is the second part of the proposition. Notice that S arises
as the composition of the isomorphism g g (g g ) , induced by 
d , with
the natural isomorphism (g g ) g g.
It is clear that the dual of the dual of a (finitedimensional) Lie bialgebra is just the
Lie bialgebra itself. It is also obvious from Definition 11.17 that the transpose of a
homomorphism of Lie bialgebras is a Lie bialgebra homomorphism between their
dual Lie bialgebras. We leave it as an exercise to the reader to show that, if h is a
Lie subbialgebra of g, then there exists a (unique) Lie bialgebra structure on the
quotient space g /h such that the projection map g g /h is a Lie bialgebra
homomorphism.
308
11 PoissonLie Groups
11.2.4 Coboundary Lie Bialgebras and rMatrices
In Section 10.2.1, the notion of a coboundary Lie bialgebra was introduced. As
the terminology suggests, coboundary Lie bialgebras are Lie bialgebras, as we will
see in the next proposition. Recall that, given (g, [ , ]g ) a Lie algebra, an element
r g g is a called an rmatrix when the transpose of the map : x adx r, which
is a linear map g g g , defines a Lie bracket on g . Recall also that the Lie
bracket on g which is associated to an rmatrix r is denoted by [ , ]r and is explicitly
given by (10.15). The corresponding triple (g, [ , ]g , [ , ]r ) was called a coboundary
Lie bialgebra in Definition 10.9.
Proposition 11.24. Let (g, [ , ]g ) be a Lie algebra, and let r g g be an rmatrix
of g. Then the coboundary Lie bialgebra (g, [ , ]g , [ , ]r ) is a Lie bialgebra.
Proof. Consider the coboundary Lie bialgebra (g, [ , ]g , [ , ]r ), where r is an rmatrix of g. Since both [ , ]g and [ , ]r are Lie brackets, we only need to verify
that condition (iii) in Proposition 11.15 holds, where we recall from (11.21) that
: g g g is the linear map, defined by
, (x)
= [ , ]r , x
,
for all x g and all , g . According to (10.17),
[ , ]r , x
= , adx r ,
so that (x) = adx r , where r denotes the skewsymmetric part of r. Using the
fact that ad is a Lie algebra representation of g (on g g),
([x, y]g ) = ad[x,y]g r = adx ady r ady adx r = adx (y) ady (x)
for all x, y g. This proves that condition (iii) in Proposition 11.15 holds, hence that
(g, [ , ]g , [ , ]r ) is a Lie bialgebra.
Not every Lie bialgebra is a coboundary Lie bialgebra. When the Lie algebra g is
abelian, recall from Example 11.18 that (g , [ , ]g = 0, [ , ]g ) is a Lie bialgebra for
every Lie bracket [ , ]g on g . Such a Lie bialgebra cannot be a coboundary Lie
bialgebra, unless [ , ]g is trivial.
We show in the following proposition that the double of a Lie bialgebra is a
coboundary Lie bialgebra. Thus we can embed every Lie bialgebra in a coboundary
Lie bialgebra, a fact which will be useful when we prove, in the next section, that
every Lie bialgebra is the Lie bialgebra of a PoissonLie group.
Proposition 11.25. Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra. On
its double (d, [ , ]d ), there exists a skewsymmetric rmatrix a 2 d, with corresponding Lie bracket [ , ]a on d , such that (g, [ , ]g , [ , ]g ) is a Lie subbialgebra
of (d, [ , ]d , [ , ]a ). In terms of a pair of dual bases (e1 , . . . , ed ) and (1 , . . . , d ) for
g and for g respectively, a is given by a = 12 di=1 ei i .
11.2 Lie Bialgebras
309
Proof. There is a natural Lie algebra splitting of d, given by d = g g, where g and
g have been identified with Lie subalgebras of d via g g {0} and g {0} g .
The corresponding Rmatrix of d is given by R(x, ) = x, for (x, ) d. It
is skewsymmetric because both g and g are isotropic subspaces of d with respect to the bilinear form 
d (see (11.22)). Using the isomorphism between d
and its dual d , induced by 
d , the Rmatrix becomes an element a g g,
which is according to Proposition 10.13 a skewsymmetric rmatrix, whose associated Lie bracket [ , ]a makes (d, [ , ]d , [ , ]a ) into a (coboundary) Lie bialgebra.
As we have seen in Example 10.14, a is given in terms of dual bases (e1 , . . . , ed )
and (1 , . . . , d ) for g and for g respectively, by a = 12 di=1 ei i . We show that
(d, [ , ]d , [ , ]a ) contains (g, [ , ]g , [ , ]g ) as a Lie subbialgebra. The natural inclusion : g g {0} d is, by definition of the bracket on d, a Lie algebra homomorphism (g, [ , ]g ) (d, [ , ]d ). Under the natural identification between d and
g g, the transpose to is given by ( , x) = , for ( , x) g g. Therefore,
showing that is also a Lie algebra homomorphism, amounts to showing that
[( , y), ( , z)]a , (x, 0)
= [ , ]g , x ,
for all ( , y), ( , z) g g and for all x g. This is done by explicitly computing
the lefthand side of the latter equation, namely
[( , y), ( , z)]a , (x, 0)
d
1
=
( , y) ( , z), ad(x,0) ei i
2
i=1
1 d
( , y) ( , z), [(x, 0), (ei , 0)]d (0, i ) + (ei , 0) [(x, 0), (0, i )]d
2 i=1
1 d
= ( , y) ( , z), ([x, ei ]g , 0) (0, i ) + (ei , 0) ( adi x, adx i )
2 i=1
1 d
= , [x, ei ]g i , z
, [x, ei ]g i , y
2 i=1
1 d
, ei
adx i , z
, adi x , ei
adx i , y
, adi x
2 i=1
1 d
, ei
, adi x , ei
, adi x
2 i=1
= [ , ]g , x .
=
It follows that is a Lie algebra homomorphism, so that is a Lie bialgebra homomorphism.
Proposition 11.26. Let (g, [ , ]g , [ , ]g ) be a Lie bialgebra. If (g, [ , ]g ) is semisimple, then (g, [ , ]g , [ , ]g ) is a coboundary Lie bialgebra.
310
11 PoissonLie Groups
Proof. Condition (iii) in Proposition 11.15 means that the map : g 2 g is a 1cocycle in the complex computing the cohomology of the Lie algebra g with values
on 2 g, which is acted upon by adjoint action. According to Whiteheads lemma
(Lemma 4.1), is a coboundary, which, in the present case means that (x) = adx a
for some a 2 g. Since is the transpose of a Lie algebra bracket, a is a (skew
symmetric) rmatrix, hence (g, [ , ]g , [ , ]g ) is a coboundary Lie bialgebra.
11.2.5 Manin Triples
Making abstraction of the structure of the double d := g g of a Lie bialgebra
(g, [ , ]g , [ , ]g ) leads to the notion of a Manin triple, defined as follows.
Definition 11.27. Let (E, [ , ]) be a finitedimensional quadratic Lie algebra, whose
bilinear form is denoted by 
. Let V and W be vector subspaces of E. The triple3
((E, [ , ], 
),V,W ) is said to be a Manin triple if
(1)
(2)
(3)
E = V W ;
V and W are Lie subalgebras of E;
V and W are isotropic with respect to 
.
We show in the following proposition that there is a natural onetoone correspondence between Manin triples and Lie bialgebras.
Proposition 11.28.
(1)
(2)
Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra. Define d := g g ,
the double Lie algebra, with Lie bracket [ , ]d and with bilinear form 
d
given by (11.23) and (11.22) respectively. The triple ((d, [ , ]d , 
d ), g, g )
is a Manin triple.
Conversely, let ((E, [ , ]E , 
E ),V,W ) be a Manin triple, where E is assumed to be finitedimensional. Denote by [ , ]V and [ , ]W the restriction
of [ , ]E to V and W respectively, and by ([ , ]W ) the Lie algebra bracket
on V obtained by transporting [ , ]W under the isomorphism : V W ,
induced by 
E . Then (V, [ , ]V , ([ , ]W )) is a Lie bialgebra.
Proof. Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra, with double
d := g g . Identify, as before, g and g as Lie subalgebras of d, via the natural
identifications g g {0} and g {0} g , so that d = g g . Clearly, both g
and g are isotropic with respect to 
d , so that ((d, [ , ]d , 
d ), g, g ) is a Manin
triple. For the converse, let ((E, [ , ]E , 
E ),V,W ) be a Manin triple, with E finitedimensional. Then 
E induces an isomorphism between V and W , so we obtain
a Lie algebra structure on V and on V . In view of (i) in Proposition 11.15 and because (E, 
E ) is quadratic, (V, [ , ]V , ([ , ]W )) is a Lie bialgebra. It is easily
verified that both constructions are inverse to each other.
The triple is usually simply written as (E,V,W ).
11.2 Lie Bialgebras
311
Notice that, inverting the roles of V and W in a Manin triple, corresponds at the
bialgebra level to taking the dual.
Example 11.29. We construct a natural Manin triple which corresponds to Example 10.15. Thus, we assume that (g, [ , ]) is a finitedimensional Lie algebra and
that g = g+ g0 g is a Lie algebra decomposition, where g+ and g are Lie
subalgebras of g and that g0 is abelian and is contained in the normalizer of g+
and in the normalizer of g . Moreover, it is assumed that g comes equipped with
a nondegenerate symmetric bilinear form 
with respect to which both g+ and
g are isotropic and orthogonal to g0 . Recall that the endomorphism R of g, defined for x g by Rx := x+ x , where x+ , x0 and x stand for the projections
of x on g+ , g0 and g respectively, is a skewsymmetric solution to the modified
YangBaxter equation (10.4), with c = 1, hence leads to a skewsymmetric rmatrix
for g, making (g, [ , ], [ , ]r ) into a Lie bialgebra. Let E := g g, equipped with
the product Lie bracket, denoted by [ , ] , and with the adinvariant symmetric nondegenerate bilinear form 
, which is defined, for all x1 , x2 , y2 , y2 g, by
(x1 , y1 )  (x2 , y2 )
:= x1  x2
y1  y2
.
(11.31)
Let be the subspace of E, defined by := {(x, x)  x g} and let W denote the
subspace of E, defined by
W := {(x + z, y z)  x g+ , y g and z g0 } .
It is clear that E = W . Moreover, the assumed properties on the factors of g
imply at once that and W are both subalgebras of E and are both isotropic with
respect to 
. It follows that (E, ,W ) is a Manin triple. In order to show that this
Manin triple corresponds to the coboundary Lie bialgebra (g, [ , ], [ , ]r ) we need to
show that the restriction of [ , ] to , respectively to W , yields the Lie brackets [ , ]
and [ , ]r upon identifying and g, respectively W and g ; the former identification
is done by the diagonal map, which maps x g to (x, x) g g. It is clear that
[ , ] , restricted to , coincides under this identification with [ , ]. Consider the map
: W g, defined for (x + z, y z) W by (x + z, y z) := x y + 2z. It is
the transpose of the isomorphism g W , induced by 
, composed with the
isomorphism g g, induced by 
. We claim that, under , the Lie algebra
structure on W corresponds to the Rbracket [ , ]R on g, showing our claim, since the
rbracket on g corresponds to the Rbracket on g under the isomorphism induced
by 
. This follows from the following computation, where we use that both g+
and g are Lie subalgebras of g, that g0 is contained in the normalizer of g+ and
in the normalizer of g , and that g0 is abelian: for all (x + z, y z) W and (x +
z , y z ) W :
(x + z, y z), (x + z , y z ) R
= x y + 2z, x y + 2z R
1
= ( x + y, x y + 2z + x y + 2z, x + y )
2
312
11 PoissonLie Groups
= x, x y, y + x, z + z, x + y, z + z, y
= x + z, x + z y z, y z
= ( x + z, x + z , y z, y z )
= ( (x + z, y z), (x + z , y z ) ) .
Notice that, considering the particular case of g0 = {0}, the constructed Manin triple
corresponds to the case of a Lie algebra splitting (see Example 10.14).
Example 11.30. We also construct the Manin triple which corresponds to Example 10.16. To do this, we consider the complex Lie algebra sld (C) as a quadratic
Lie algebra over R, equipped
with the bilinear form, defined for x, y sld (C) by
x  y
:= Trace(xy) . We consider two Lie subalgebras: t+ , the space of upper
triangular matrices admitting only real coefficients on the diagonal, and sud . We
have that sld (C) = sud t+ . For all x, y sud ,
Trace(xy) = Trace(xy) = Trace x y = Trace(yx) = Trace(xy) ,
so that x  y
= (Trace(xy)) = 0. This shows that sud is isotropic with respect
to 
. For every x, y t+ , it is clear that xy t+ , hence Trace(xy) has vanishing imaginary part and t+ is also isotropic with respect to 
. It follows that
(sld (C), sud , t+ ) is a Manin triple. We show that it is the Manin triple corresponding to the Lie bialgebra discussed in Example 10.16. To do this, recall from the latter
example that we consider on sud the nondegenerate symmetric bilinear form 
,
defined for x, y sud by x  y
:= (Trace(xy)). Let : t+ sud be the composition of the isomorphisms t+ sud sud , wherethe first one is induced by 
and the second one by 
. Explicitly, (x) = 21 (x + x ), for x t+ , as is computed from x  y
= (x)  y
, for all y sud . For x, y t+ , one checks by direct
computation that ([x, y]) = [ (x), (y)]R , which confirms that the Lie bialgebra,
associated to the above Manin triple, is the one from Example 10.16.
We have seen in Proposition 11.25 that the double of a Lie bialgebra is also a Lie
bialgebra. We construct in the following proposition the Manin triple of this double
Lie bialgebra in terms of the Manin triple of the original bialgebra.
Proposition 11.31. Let ((E, [ , ], 
),V,W ) be a Manin triple, where E is finitedimensional. Let E E be equipped with the product Lie bracket, which we denote
by [ , ] , and with the nondegenerate adinvariant symmetric bilinear form 
,
given for (x1 , y1 ), (x2 , y2 ) E E by
(x1 , y1 )  (x2 , y2 )
:= x1  x2
y1  y2
.
(11.32)
Let
V := {(u, u) E E  u E} ,
W := W V .
(11.33)
11.2 Lie Bialgebras
313
Then ((E E, [ , ] , 
),V ,W ) is a Manin triple and the Lie bialgebra associated to it is the double of the Lie bialgebra associated to (E,V,W ).
Proof. Let ((E, [ , ], 
),V,W ) be a Manin triple and let (g, [ , ]g , [ , ]g ) be its
associated Lie bialgebra. According to Proposition 11.28, we may identify (E, [ , ])
with (d = g g , [ , ]d ); under this identification, (V, [ , ]V ) and (W, [ , ]W ) are
identified with (g, [ , ]g ) and (g , [ , ]g ) respectively; the inner product 
on
E corresponds to the natural inner product (11.22) on d. According to Proposition 11.25, d is a coboundary Lie bialgebra, where the Lie bracket on d is the
rbracket a which is associated to the Lie algebra splitting d = g g, i.e., the
Lie algebra splitting E = W V . According to Example 11.29, in the particular
case of g0 = {0}, the Manin triple of such a coboundary Lie algebra is given by
((E E, [ , ] , 
),V ,W ), where [ , ] is the product Lie bracket on E E and
where 
and V , W are given by (11.32) and (11.33).
11.2.6 Lie Bialgebras and Poisson Structures
We give in the following proposition a different description of the notion of a Lie
bialgebra, by interpreting one of the Lie algebra structures as a linear Poisson structure: according to Proposition 7.3, to a Lie bracket [ , ]g on g corresponds a linear
Poisson structure on g (and vice versa).
Proposition 11.32. Let (g, [ , ]g ) be a finitedimensional Lie algebra and let be a
linear Poisson structure on g. Then the following conditions are equivalent:
(i)
(ii)
Denoting by [ , ]g the Lie bracket on g , corresponding to , the triple
(g, [ , ]g , [ , ]g ) is a Lie bialgebra;
For every x, y g, the bivector x 2 Tx g, considered as an element of 2 g,
satisfies
[x,y]g = adx y ady x .
(11.34)
Proof. Let [ , ]g be the Lie bracket on g which corresponds to the linear Poisson structure = { , } on g. For F, G F (g), their Poisson bracket at x g is,
according to (7.4) and (11.21), given by
{F, G} (x) = [dx F, dx G]g , x = dx F dx G, (x)
,
where we recall that in this formula dx F and dx G are viewed as elements of g . It
follows that (x) = x for every x g, where the latter bivector is considered as an
element of 2 g. The equivalence between (i) and (ii) then follows from item (iii) in
Proposition 11.15 and Definition 11.17.
If (g, [ , ]g , [ , ]g ) and (h, [ , ]h , [ , ]h ) are two finitedimensional Lie bialgebras,
with corresponding Poisson structures { , }g and { , }h on g and h respectively,
then a linear map : g h is a homomorphism of Lie bialgebras if and only if
314
11 PoissonLie Groups
: (g, [ , ]g ) (h, [ , ]h ) is a homomorphism of Lie algebras and : (g, { , }g )
(h, { , }h ) is a Poisson map. Indeed, the latter condition is equivalent to the fact that
: (h , [ , ]h ) (g , [ , ]g ) is a Lie algebra homomorphism. It follows that Lie
subbialgebras can be characterized as follows:
Proposition 11.33. Let (g, [ , ]g , [ , ]g ) be a Lie bialgebra, and denote by the
linear Poisson structure on g corresponding to [ , ]g . A subspace h g is a Lie
subbialgebra if and only if the following two conditions are satisfied:
(1)
(2)
h is a Lie subalgebra of the Lie algebra (g, [ , ]g );
h is a Poisson submanifold of the Poisson manifold (g, ).
11.3 PoissonLie Groups and Lie Bialgebras
In this section, we relate the objects which appeared in the two previous sections,
namely we establish a natural correspondence between PoissonLie groups and
finitedimensional Lie bialgebras. We first show that the tangent space to the unit
of a PoissonLie group inherits the structure of a Lie bialgebra (Section 11.3.1).
We then show that every finitedimensional Lie bialgebra is the Lie bialgebra
of a PoissonLie group. This is first done in Section 11.3.2 for coboundary Lie
bialgebras and then in general in Section 11.3.3, using the fact that every finitedimensional Lie bialgebra can be embedded in a coboundary Lie bialgebra (its double, see Proposition 11.25).
11.3.1 The Lie Bialgebra of a PoissonLie Group
Let (G, ) be a PoissonLie group, whose unit is denoted by e. First, since G is a
Lie group, the tangent space g := Te G is equipped with a Lie algebra bracket [ , ]g
(see Theorem 5.1). Second, according to item (1) in Proposition 11.5, the Poisson
structure vanishes at e, leading to a linear Poisson structure 1 on Te G = g (the
linearized Poisson structure, see Section 7.5), hence a Lie bracket [ , ]g on g (see
Proposition 7.3), which we call the linearized bracket of (G, ). We show in the
following proposition that the Lie structures [ , ]g and [ , ]g make g into a Lie bialgebra, so that we can naturally associate to every PoissonLie group a Lie bialgebra.
Proposition 11.34. Let (G, ) be a PoissonLie group, with Lie algebra (g, [ , ]g ),
and with linearized bracket [ , ]g .
(1)
(2)
The triple (g, [ , ]g , [ , ]g ) is a Lie bialgebra;
Let be a PoissonLie group homomorphism from (G, ) to a PoissonLie
group (H, ), with corresponding Lie bialgebra (h, [ , ]h , [ , ]h ). The tangent map Te of at the unit e of G is a Lie bialgebra homomorphism from
(g, [ , ]g , [ , ]g ) to (h, [ , ]h , [ , ]h ).
11.3 PoissonLie Groups and Lie Bialgebras
315
Proof. Let (G, ) be a PoissonLie group with Lie algebra (g, [ , ]g ) and let
denote the cocycle G 2 g, defined by (11.2). We show that the cocycle condition (11.3) implies condition (ii) in Proposition 11.32, which guarantees that the
triple (g, [ , ]g , [ , ]g ) is a Lie bialgebra, where [ , ]g denotes the linearized bracket
of (G, ). Consider the identity
(ghg1 ) = (g) + Adg (h) Adghg1 (g) ,
(11.35)
which is valid for all g, h G, as follows from the cocycle condition (11.3), upon
using that (e) = 0. Differentiating (11.35) leads to an identity for the linearized
Poisson bracket 1 , since
(1 )x = Te (x)
(11.36)
for x g (see (2) in Proposition 11.5). In order to obtain this identity, replace h
by exp(ty) in (11.35) and take the derivative at t = 0. In view of (11.36), this leads to
(1 )Adg y = Adg (1 )y adAdg y (g) ,
(11.37)
for all y g and g G. We replace now in (11.37) g by exp(tx) and we take the
derivative at t = 0; we claim that we obtain
(1 )[x,y]g = adx (1 )y ady (1 )x ,
(11.38)
which in view of Proposition 11.32, shows that the triple (g, [ , ]g ), [ , ]g ) is a Lie
bialgebra.
Let us show in detail how (11.38) is obtained from (11.37). Since 1 is a linear
Poisson structure, the map g 2 g, defined by x (1 )x is a linear map. It follows
that, for every function z : F g,
d
(1 )z(t) = (1 ) d  z(t) .
dt t=0
dt t=0
Applied to the lefthand side of (11.37), with g = exp(tx), this yields the lefthand
side of (11.38), in view of the definition of ad. Also,
d
Ad
(1 )y = adx (1 )y ,
dt t=0 exp(tx)
which yields the first term in the righthand side of (11.37). Finally, since (e) = 0,
d
d
ad
(exp(tx))
=
ad
(exp(tx))
y
Adexp(tx) y
dt t=0
dt t=0
= ady Te (x) = ady (1 )x ,
which is the second term in the righthand side of (11.37). This completes the proof
of (11.38), and hence of item (1).
316
11 PoissonLie Groups
We turn to the proof of (2). Since : G H is a Lie group homomorphism,
the tangent map Te : g h is a Lie algebra homomorphism (Theorem 5.1). Similarly, since : G H is a Poisson map, with vanishing at e and vanishing at
(e), the linear map Te : g h is a Poisson map (Proposition 7.22). According
to Section 11.2.6, this shows that is a Lie bialgebra homomorphism.
The second item of the previous proposition implies that the Lie bialgebra of a
PoissonLie subgroup of (G, ) is a Lie subbialgebra of the bialgebra of (G, ).
This is stated, together with its converse, in the following proposition.
Proposition 11.35. Let (G, ) be a PoissonLie group, whose Lie bialgebra is denoted by (g, [ , ]g , [ , ]g ). For a connected Lie subgroup H of G, with Lie algebra h,
the following conditions are equivalent:
(i)
(ii)
H is a PoissonLie subgroup of (G, );
h is a Lie subbialgebra of (g, [ , ]g , [ , ]g ).
Assuming that these equivalent conditions hold, denote the Lie brackets of the Lie
bialgebra h by [ , ]h and [ , ]h , and denote the Poisson structure on H, which
makes H into a Poisson submanifold of (G, ), by . Then (h, [ , ]h , [ , ]h ) is the
Lie bialgebra of (H, ).
Proof. The implication (i) (ii) follows at once from Proposition 11.34, since
the inclusion map of a PoissonLie subgroup is a PoissonLie group homomorphism. Conversely, assume that h is a Lie subbialgebra of the Lie bialgebra
(g, [ , ]g , [ , ]g ) of (G, ). Then the transpose : g 2 g of [ , ]g maps h to
2 h. According to item (2) in Proposition 11.5 and the proof of Proposition 11.32,
(x) = Te (x) for all x g, where is defined as usual by (11.2). It follows that
Te (x) 2 h for all x h. According to Proposition 11.7, this shows that the connected Lie subgroup H of G is a PoissonLie subgroup. Thus, condition (ii) implies
condition (i).
11.3.2 Coboundary PoissonLie Groups and Lie Bialgebras
We show in this section that the Lie bialgebra of a coboundary PoissonLie group
is a coboundary Lie bialgebra, which justifies why the same adjective coboundary is used for both structures. We also show the following converse: every
finitedimensional coboundary Lie bialgebra is the Lie bialgebra of a coboundary
PoissonLie group.
We first show that there corresponds to every coboundary PoissonLie group a
coboundary Lie bialgebra. Let (G, ) be a coboundary PoissonLie group. By defi
a
a , where (g, [ , ]) is the Lie algebra
nition, there exists a 2 g such that =
of G. According to Proposition 11.10, [[a, a]] is Adinvariant, hence is adinvariant.
Proposition 10.11 implies that a is a skewsymmetric rmatrix, i.e., the associated
bracket [ , ]a is a Lie bracket on g , making (g, [ , ], [ , ]a ) into a coboundary Lie
11.3 PoissonLie Groups and Lie Bialgebras
317
bialgebra (see Definition 10.9). The next proposition shows that this coboundary Lie
bialgebra is the Lie bialgebra of the PoissonLie group (G, ).
a
a
Proposition 11.36. Let (G, ) be a coboundary PoissonLie group, with =
for some rmatrix a 2 g, where (g, [ , ]g ) denotes the Lie algebra of G. The Lie
bialgebra of (G, ) is the coboundary Lie bialgebra (g, [ , ]g , [ , ]a ).
Proof. According to (11.14), the map : G 2 g, defined by g 2 (Tg Rg1 ) g ,
is given by (g) = Adg a a. The tangent map of at e is therefore given by
Te (x) = adx a for all x g. According to item (2) in Proposition 11.5, the linearized
Poisson structure 1 of at the unit e of G is given by (1 )x = Te (x), for all x g,
so that the linearized Poisson structure 1 is given at x g by (1 )x = adx a, whose
transpose is precisely the Lie bracket [ , ]a . This shows that (g, [ , ]g , [ , ]a ) is the
Lie bialgebra of (G, ).
Corollary 11.37. Let (g, [ , ]g , [ , ]a ) be a finitedimensional coboundary Lie bialgebra. If G is a connected Lie group, with Lie algebra (g, [ , ]g ), then G admits the structure of a coboundary PoissonLie group, whose Lie bialgebra is
(g, [ , ]g , [ , ]a ). In particular, every finitedimensional coboundary Lie bialgebra
is the Lie bialgebra of a coboundary PoissonLie group.
Proof. We only need to prove the first part of the corollary, since the last statement follows from it at once upon using the fact that every Lie algebra is the Lie
algebra of a connected Lie group (see Theorem 5.1). Let (g, [ , ]g , [ , ]a ) be a finitedimensional coboundary Lie bialgebra, with a 2 g so that [[a, a]] is adinvariant.
Let G be a connected Lie group whose Lie algebra is g. Since G is assumed to be
connected, [[a, a]] is Adinvariant. Setting :=
a
a , Propositions 11.10 and 11.36
imply that (G, ) is a PoissonLie group, whose Lie bialgebra is (g, [ , ]g , [ , ]a ), as
was to be shown.
Corollary 11.38. Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra and
let D be a connected Lie group whose Lie algebra is the double (d, [ , ]d ) of
(g, [ , ]g , [ , ]g ). Then D admits a unique Poisson structure D , making (D, D ) into
a coboundary PoissonLie group, whose Lie bialgebra is (d, [ , ]d , [ , ]a ). In terms
of a pair of dual bases (e1 , . . . , ed ) and (1 , . . . , d ) for g and for g respectively, the
skewsymmetric rmatrix a of d is given by a = 12 di=1 ei i .
Proof. According to Proposition 11.25, the double (d, [ , ]d ) is a coboundary Lie
bialgebra; as before, we denote the corresponding Lie structure on d by [ , ]a . If D
is a connected Lie group with Lie algebra (d, [ , ]d ), then D admits, according to
Corollary 11.37, the structure of a coboundary PoissonLie group, whose Lie bialgebra is (d, [ , ]d , [ , ]a ). As we will see in Theorem 11.39, the Poisson structure
on D which integrates the given Lie bialgebra structure is unique.
318
11 PoissonLie Groups
11.3.3 The Integration of Lie Bialgebras
We show in this section that every finitedimensional Lie bialgebra can be integrated. We mean by this that there exists for every finitedimensional Lie bialgebra
(g, [ , ]g , [ , ]g ) a (not necessarily unique) PoissonLie group (G, ) whose Lie
bialgebra is isomorphic to (g, [ , ]g , [ , ]g ). For the particular case of a coboundary
Lie bialgebras, this was proved in Corollary 11.37.
Theorem 11.39. Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra and
let G be a connected Lie group with Lie algebra (g, [ , ]g ).
(1)
(2)
(3)
There exists at most one Poisson structure on G such that (G, ) is a
PoissonLie group, whose Lie bialgebra is (g, [ , ]g , [ , ]g );
If G is simply connected, then there exists on G a unique Poisson structure , such that (G, ) is a PoissonLie group, whose Lie bialgebra is
(g, [ , ]g , [ , ]g );
In particular, every finitedimensional Lie bialgebra is the Lie bialgebra of a
PoissonLie group.
Proof. Let (1) and (2) be two multiplicative Poisson structures on the connected
Lie group G, leading to the same Lie bialgebra, i.e., their linearized Poisson structures at e coincide. In view of (2) in Proposition 11.5, Te (1) (x) = Te (2) (x), for all
(i)
x g, where (i) : G 2 g is defined by (i) (g) := 2 (Tg Rg1 )g , for i = 1, 2. It
follows that (1) (2) , which is a cocycle (since both (1) and (2) are cocycles),
has zero tangent map at e. By Proposition 11.4, this implies that (1) (2) = 0,
and hence that (1) = (2) . This shows item (1).
Let (g, [ , ]g , [ , ]g ) be a finitedimensional Lie bialgebra. According to Proposition 11.25, (g, [ , ]g , [ , ]g ) is a Lie subbialgebra of a coboundary Lie bialgebra,
which we denote by (d, [ , ]d , [ , ]a ), because the underlying Lie algebra is the double of (g, [ , ]g , [ , ]g ). In view of Proposition 11.37, (d, [ , ]d , [ , ]a ) can be integrated into a PoissonLie group (D, D ), which is the only fact about d which we
the unique connected Lie subgroup of D,
will use in this proof. We denote by G
inherits a
whose Lie algebra is g. According to Proposition 11.35, the subgroup G
multiplicative Poisson structure from D , whose corresponding Lie bialgebra is
(g, [ , ]g , [ , ]g ). This shows (3).
) can be replaced by the connected and simply conWe show now that (G,
nected Lie group G whose Lie algebra is g, which is the content of (2). In view
of Lies theorem (see item (4) of Theorem 5.1), there exists a Lie group homomor whose induced Lie algebra homomorphism is the identity map.
phism : G G,
into
We claim that there is a unique Poisson structure on G which turns : G G
a Poisson map, and we show that is multiplicative. Then (G, ) is a PoissonLie
group and is a homomorphism of PoissonLie groups, so that the induced
Lie bialgebra homomorphism, which is the identity map, is a homomorphism of Lie
bialgebras, which means that the Lie bialgebra of G is (g, [ , ]g , [ , ]g ).
Let us prove the claim. Differentiating the identity = R (g) Rg1 at g G
shows that the tangent maps of at g G and at e are related by Tg = Te R (g)
11.3 PoissonLie Groups and Lie Bialgebras
319
Te Tg Rg1 (in this formula, e and e are the units of the Lie groups G and G
respectively). Since the three linear maps which appear in the righthand side of
this formula are invertible, so is Tg , hence is a local diffeomorphism. We can
therefore define for all g G by setting
g := 2 (Tg )1 (g) .
(11.39)
According to Proposition 1.19, if there exists a Poisson structure on G which makes
into a Poisson map, then it must be given by (11.39), which gives the desired
uniqueness of . Also, the vertical maps of the following commutative diagram are
local diffeomorphisms (here and stand for the group products of G and G).
GG
Combined with the fact that is a Poisson map, this shows that is Poisson map.
As a consequence, is multiplicative, as remained to be shown.
Having established the fact that every Lie bialgebra can be integrated into a Poisson
Lie group, we now show that every Lie bialgebra homomorphism can be integrated
into a morphism of PoissonLie groups.
Theorem 11.40. Let (g, [ , ]g , [ , ]g ) and (h, [ , ]h , [ , ]h ) be two finitedimensional Lie bialgebras and let (G, ) and (H, ) be PoissonLie groups which integrate them, where G is chosen connected and simply connected. For every Lie
bialgebra homomorphism : g h, there exists a unique PoissonLie group homomorphism : G H whose induced Lie bialgebra homomorphism is , i.e., such
that Te = , where e denotes the unit element of G.
Proof. Suppose that : g h is a homomorphism of Lie bialgebras. In view of
Lies theorem (item (4) of Theorem 5.1), there exists a unique Lie group homomorphism : G H such that Te = . It remains to be shown that is a Poisson
map. To do this, we consider the map : G 2 h, defined for all g G by
(g) := 2 ( (g)) ( (g)) ,
where : G 2 g and : H 2 h denote the group cocycles, associated to
(G, ) and to (H, ) respectively (see item (iii) in Proposition 11.2). The map is
a cocycle of G with respect to the adjoint representation of G on 2 h, i.e.,
(gh) = (g) + Ad (g) (h) ,
for all g, h G. We first show that = 0. In view of item (1) of Proposition 11.4,
it suffices to show that Te (x) = 0 for all x g. To prove the latter, recall from
320
11 PoissonLie Groups
item (2) in Proposition 11.5 that Te (x) = (1 )x , for all x g, where 1 denotes the
linearized Poisson structure of at e, and similarly for and 1 . It allows us to
compute
Te (x) = 2 (Te (x)) Te (Te (x))
= 2 ((1 )x ) (1 ) (x) = 0 ,
where the last equality, which states that is a Poisson map, is tantamount to the
fact that the transpose map : (h , [ , ]h ) (g , [ , ]g ) is a Lie algebra homomorphism (see Section 11.2.6). It follows that = 0, leading to the equality
2 ( (g)) = ( (g)) ,
valid for all g G, which can be written, using Te = and using the definition (11.2) of the cocycle , as
2 (Tg ( Rg1 ))g = 2 T (g) R (g)1 (g) .
(11.40)
Now is a group homomorphism, so that
Rg1 = R (g1 ) ,
which, substituted in (11.40), leads to 2 (Tg )g = (g) , since in a Lie group,
right translation by an element of the group is a diffeomorphism. In view of Proposition 1.19, the latter equality implies that is a Poisson map, as was to be shown.
The notion of duality for Lie bialgebra leads to a notion of duality for connected,
simply connected PoissonLie groups.
Definition 11.41. Two PoissonLie groups (G, ) and (G , ) are said to be dual
to each other if their Lie bialgebras are dual to each other.
Let (G, ) be a PoissonLie group and let (g, [ , ]g , [ , ]g ) denote its Lie bialgebra, whose dual bialgebra is (g , [ , ]g , [ , ]g ). Every PoissonLie group (G , )
which integrates the latter Lie bialgebra is dual to (G, ). A canonical dual is picked
by demanding that G be connected and simply connected. Indeed, there is up to
isomorphism a unique connected and simply connected Lie group G whose Lie
algebra is (g , [ , ]g ) (item (3) of Theorem 5.1) and on G the
Viel mehr als nur Dokumente.
Entdecken, was Scribd alles zu bieten hat, inklusive Bücher und Hörbücher von großen Verlagen.
Jederzeit kündbar.