Grundlehren der

mathematischen Wissenschaften 347

A Series of Comprehensive Studies in Mathematics

Series editors

M. Berger P. de la Harpe F. Hirzebruch

N.J. Hitchin L. Hörmander A. Kupiainen
G. Lebeau F.-H. Lin S. Mori
B.C. Ngô M. Ratner D. Serre
N.J.A. Sloane A.M. Vershik M. Waldschmidt

Editor-in-Chief
A. Chenciner J. Coates S.R.S. Varadhan
For further volumes:
www.springer.com/series/138
Camille Laurent-Gengoux r Anne Pichereau r

Pol Vanhaecke

Poisson Structures
Camille Laurent-Gengoux Pol Vanhaecke
CNRS UMR 7122, Laboratoire CNRS UMR 7348, Lab. Mathématiques
de Mathématiques et Applications
Université de Lorraine Université de Poitiers
Metz, France Futuroscope Chasseneuil, France
Anne Pichereau
CNRS UMR 5208, Institut
Camille Jordan
Université Jean Monnet
Saint-Etienne, France

ISSN 0072-7830 Grundlehren der mathematischen Wissenschaften

ISBN 978-3-642-31089-8 ISBN 978-3-642-31090-4 (eBook)
DOI 10.1007/978-3-642-31090-4
Springer Heidelberg New York Dordrecht London

Library of Congress Control Number: 2012947315

Mathematics Subject Classification: 17B63, 53D17, 17B80, 53D55, 17B62

© Springer-Verlag 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
Publisher’s 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 pub-
lication, 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 acid-free 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. Symplec-
tic manifolds, Lie algebras, singularity theory, r-matrices, for example, all lead to a
certain type of Poisson structures, sharing several features, despite the distances be-
tween 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 con-
cept of a Poisson algebra, so we faced the challenge of presenting the concepts and
the results in detail, for the algebraic, algebraic-geometric and geometric contexts,
without copy-pasting large parts of the text two or three times. But as the writing
moved on, both the algebraic and geometric contexts kept imposing themselves; fi-
nally, 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, unex-
pectedly, 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 Mathématiques et Ap-
plications”. 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 dic-
tionary). When two of us moved away from Poitiers, we sometimes worked to-
gether 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

Saint-Etienne and at the CRM in Barcelona and the Max-Planck 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 Mi-
randa, Joana Nunes da Costa, Marco Pedroni, Michael Penkava, Claude Roger,
Pierre van Moerbeke, Yvette Kosmann-Schwarzbach, Hervé Sabourin, Mathieu
Stiénon, 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 Laurent-Gengoux, Anne Pichereau and Pol Vanhaecke

Contents

Part I Theoretical Background

1 Poisson Structures: Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1 Poisson Algebras and Their Morphisms . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Poisson Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Morphisms of Poisson Algebras . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.3 Hamiltonian Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Poisson Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Affine Poisson Varieties and Their Morphisms . . . . . . . . . . . . 8
1.2.2 The Poisson Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.3 The Rank of a Poisson Structure . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 Bivector Fields on Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Poisson Manifolds and Poisson Maps . . . . . . . . . . . . . . . . . . . 19
1.3.3 Local Structure: Weinstein’s Splitting Theorem . . . . . . . . . . . 23
1.3.4 Global Structure: The Symplectic Foliation . . . . . . . . . . . . . . 26
1.4 Poisson Structures on a Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2 Poisson Structures: Basic Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.1 The Tensor Product of Poisson Algebras and the Product
of Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.1 The Tensor Product of Poisson Algebras . . . . . . . . . . . . . . . . . 38
2.1.2 The Product of Poisson Manifolds . . . . . . . . . . . . . . . . . . . . . . 42
2.2 Poisson Ideals and Poisson Submanifolds . . . . . . . . . . . . . . . . . . . . . . 46
2.2.1 Poisson Ideals and Poisson Subvarieties . . . . . . . . . . . . . . . . . 46
2.2.2 Poisson Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3 Real and Holomorphic Poisson Structures . . . . . . . . . . . . . . . . . . . . . . 52

ix
x Contents

2.3.1
Real Poisson Structures Associated to Holomorphic
Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.3.2 Holomorphic Poisson Structures on Smooth Affine
Poisson Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.4 Other Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.4.1 Field Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.4.2 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.4.3 Germification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3 Multi-Derivations and Kähler Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.1 Multi-Derivations and Multivector Fields . . . . . . . . . . . . . . . . . . . . . . . 64
3.1.1 Multi-Derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.1.2 Basic Operations on Multi-Derivations . . . . . . . . . . . . . . . . . . 65
3.1.3 Multivector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Kähler Forms and Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.2.1 Kähler Differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.2 Kähler Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2.3 Algebra and Coalgebra Structure . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.4 The de Rham Differential and Cohomology . . . . . . . . . . . . . . 72
3.2.5 Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3 The Schouten Bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3.1 Internal Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.3.2 The Schouten Bracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.3 The Algebraic Schouten Bracket . . . . . . . . . . . . . . . . . . . . . . . 84
3.3.4 The (Generalized) Lie Derivative . . . . . . . . . . . . . . . . . . . . . . . 85
3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

4 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

5 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 Poisson–Dirac Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.3.1 Algebraic Poisson–Dirac Reduction . . . . . . . . . . . . . . . . . . . . . 134
5.3.2 Poisson–Dirac 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 Poisson–Dirac 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: Kähler Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

7 Linear Poisson Structures and Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . 179

7.1 The Lie–Poisson Structure on g∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
7.2 The Lie–Poisson Structure on g . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7.3 Properties of the Lie–Poisson Structure . . . . . . . . . . . . . . . . . . . . . . . . 185
7.3.1 The Symplectic Foliation: Coadjoint Orbits . . . . . . . . . . . . . . 185
7.3.2 The Cohomology of Lie–Poisson Structures . . . . . . . . . . . . . . 190
7.3.3 The Modular Class of a Lie–Poisson Structure . . . . . . . . . . . . 192
7.4 Affine Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
7.5 The Linearization of Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . 196
7.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
xii Contents

8 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 Nambu–Poisson Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
8.4 Transverse Poisson Structures to Adjoint Orbits . . . . . . . . . . . . . . . . . 227
8.4.1 S-triples in Semi-Simple Lie Algebras . . . . . . . . . . . . . . . . . . . 227
8.4.2 Weights Associated to an S-Triple . . . . . . . . . . . . . . . . . . . . . . 228
8.4.3 Transverse Poisson Structures to Nilpotent Orbits . . . . . . . . . 229
8.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

9 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 R-Brackets and r-Brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

10.1 Linear R-Brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
10.1.1 R-Matrices and the Yang–Baxter Equation . . . . . . . . . . . . . . . 270
10.1.2 R-Matrices and Lie Algebra Splittings . . . . . . . . . . . . . . . . . . . 272
10.1.3 Linear R-Brackets on g∗ and on g . . . . . . . . . . . . . . . . . . . . . . . 273
10.2 Linear r-Brackets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
10.2.1 Coboundary Lie Bialgebras and r-Matrices . . . . . . . . . . . . . . . 275
10.2.2 Linear r-Brackets on g . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
10.2.3 From r-Matrices to R-Matrices . . . . . . . . . . . . . . . . . . . . . . . . . 280
10.3 R-Brackets and r-Brackets of Higher Degree . . . . . . . . . . . . . . . . . . . . 283
10.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289

11 Poisson–Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

11.1 Multiplicative Poisson Structures and Poisson–Lie Groups . . . . . . . . 292
11.1.1 The Condition of Multiplicativity . . . . . . . . . . . . . . . . . . . . . . . 292
11.1.2 Basic Properties of Poisson–Lie Groups . . . . . . . . . . . . . . . . . 294
11.1.3 Poisson–Lie Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
11.1.4 Linear Multiplicative Poisson Structures . . . . . . . . . . . . . . . . . 297
Contents xiii

11.1.5 Coboundary Poisson–Lie Groups . . . . . . . . . . . . . . . . . . . . . . . 298

11.1.6 Multiplicative Poisson Structures on Vector Spaces . . . . . . . . 299
11.2 Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
11.2.1 Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
11.2.2 Lie Sub-bialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
11.2.3 Duality for Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
11.2.4 Coboundary Lie Bialgebras and r-Matrices . . . . . . . . . . . . . . . 308
11.2.5 Manin Triples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
11.2.6 Lie Bialgebras and Poisson Structures . . . . . . . . . . . . . . . . . . . 313
11.3 Poisson–Lie Groups and Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . . 314
11.3.1 The Lie Bialgebra of a Poisson–Lie Group . . . . . . . . . . . . . . . 314
11.3.2 Coboundary Poisson–Lie Groups and Lie Bialgebras . . . . . . 316
11.3.3 The Integration of Lie Bialgebras . . . . . . . . . . . . . . . . . . . . . . . 318
11.4 Dressing Actions and Symplectic Leaves . . . . . . . . . . . . . . . . . . . . . . . 321
11.5 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 Poisson’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
12.2.2 Noether’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
12.2.3 Bi-Hamiltonian Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . 333
12.2.4 Poisson Maps and Thimm’s Method . . . . . . . . . . . . . . . . . . . . 334
12.2.5 Lax Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
12.2.6 The Adler–Kostant–Symes Theorem . . . . . . . . . . . . . . . . . . . . 337
12.3 The Liouville Theorem and the Action-Angle Theorem . . . . . . . . . . . 341
12.3.1 Liouville Integrable Systems and Liouville’s Theorem . . . . . 341
12.3.2 Foliation by Standard Liouville Tori . . . . . . . . . . . . . . . . . . . . 343
12.3.3 Period Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
12.3.4 The Existence of Action-Angle Coordinates . . . . . . . . . . . . . . 346
12.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350

13 Deformation Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353

13.1 Deformations of Associative Products . . . . . . . . . . . . . . . . . . . . . . . . . 354
13.1.1 Formal and k-th Order Deformations . . . . . . . . . . . . . . . . . . . . 354
13.1.2 Hochschild Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
13.1.3 Deformations and Cohomology . . . . . . . . . . . . . . . . . . . . . . . . 359
13.1.4 The Maurer–Cartan 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 k-th Order Deformations . . . . . . . . . . . . . . . . . . . . 367
xiv Contents

13.2.2 Deformations and Cohomology . . . . . . . . . . . . . . . . . . . . . . . . 368

13.2.3 The Maurer–Cartan Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 371
13.3 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 Maurer–Cartan Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 376
13.3.4 Gauge Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
13.3.5 Path Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
13.3.6 Applications to Deformation Theory . . . . . . . . . . . . . . . . . . . . 385
13.4 L∞ -Morphisms of Differential Graded Lie Algebras . . . . . . . . . . . . . . 386
13.4.1 ΩΦ is Well-Defined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
13.4.2 Surjectivity of ΩΦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
13.4.3 Injectivity of ΩΦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
13.5 Kontsevich’s Formality Theorem and Its Consequences . . . . . . . . . . . 397
13.5.1 Kontsevich’s Formality Theorem . . . . . . . . . . . . . . . . . . . . . . . 398
13.5.2 Kontsevich’s Formality Theorem for Rd . . . . . . . . . . . . . . . . . 399
13.5.3 A Few Consequences of Kontsevich’s Formality Theorem . . 406
13.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408

A 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

B 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. Poisson’s classical bracket, which is
given for smooth functions F and G on R2r by
r  
∂F ∂G ∂G ∂F
{F, G} := ∑ − , (0.1)
i=1 ∂ qi ∂ pi ∂ qi ∂ pi

allows one to write Hamilton’s equations of motion

∂H ∂H
q̇i = , ṗi = − , i = 1, . . . , r , (0.2)
∂ pi ∂ qi

associated to a Hamiltonian H = H(q, p), in a coordinate-free manner, treating po-

sitions qi and momenta pi on an equal footing. Indeed, using (0.1), the equations of
motion (0.2) can be written as

q̇i = {qi , H} , ṗi = {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)) , (0.4)
dt
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 rôle which the Poisson structure plays in solving the problems of classi-
cal 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 integra-
tion of Hamilton’s equations, Poisson’s 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 integrat-
ing Hamilton’s equations by quadratures, under the hypothesis that these functions
commute for the Poisson bracket (are in involution). A yet further amplification is
the action-angle theorem, which states that in the neighborhood of a compact sub-
manifold, traced out by the flows of the (commuting!) vector fields XHi , there exist
(S1 )r × Rr -valued coordinates (θ1 , . . . , θr , ρ1 , . . . , ρr ) in which the Poisson bracket
takes the canonical form (0.1), namely
r  
∂F ∂G ∂G ∂F
{F, G} = ∑ − ,
i=1 ∂ θi ∂ ρi ∂ θi ∂ ρi

and such that the functions H1 , . . . , Hr (including the Hamiltonian H) depend on the
coordinates ρ1 , . . . , ρr only. According to (0.2), Hamilton’s equations take in these
coordinates the simple form

∂H ∂H
θ̇i = , ρ̇i = − =0, i = 1, . . . , r ,
∂ ρi ∂ θi
which entails that the ρi are constant and hence that the θi are affine functions of
time (since H does not depend on θ1 , . . . , θr ).
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 rôle
played by Poisson brackets, we explain the algebraic and geometric abstraction of
the classical Poisson bracket (0.1). It is well known that Poisson’s theorem is ex-
plained 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 an-
nounced his theorem. Taking arbitrary smooth functions xi j = −x ji defined on a
non-empty open subset of Rd , with coordinates x1 , . . . , xd , the skew-symmetric bi-
linear operation {· , ·}, defined on smooth functions by
d
∂F ∂G
{F, G} := ∑ xi j
∂ xi ∂ x j
, (0.6)
i, j=1

satisfies the Jacobi identity if and only if the functions xi j satisfy the identity
 
d
∂ xi j ∂ x jk ∂ xki
∑ k ∂ x i ∂ x  j ∂ x = 0 .
x + x + x
=1

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} = ∑ − ,
i=1 ∂ qi ∂ pi ∂ 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 (finite-dimensional)
Lie algebra g, where the bracket is directly inherited by the Lie bracket. Many im-
portant 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 rediscov-
ered 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 geomet-
rical terms, a Poisson structure on a smooth manifold M is a smooth bivector field π
on M, satisfying [π , π ]S = 0, where [· , ·]S stands for the Schouten bracket (a natural
extension of the Lie bracket of vector fields to arbitrary multivector fields). It im-
plies 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, πm  , (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 com-
mutative, 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, π ), 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 exam-

ple of Poisson structures making an unexpected appearance, and playing next a very
definite rôle 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 quanti-
zation”.
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
(0.9)
(a, b) → a t b

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 skew-symmetric bilinear map {· , ·} : A × A → A , which is defined

for a, b ∈ A by
d d
{a, b} := [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 prop-
erty (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 fam-
ily 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 Planck’s constant h̄, the
operators become in the limit h̄ → 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 de-
formations, 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 op-
erators.
A mathematical simplification of this problem is to ask whether the above pro-
cedure 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 [[ν ]] of for-
mal power series in one variable ν , whose coefficients belong to A . Let μ denote
an associative product on A [[ν ]], i.e., an F[[ν ]]-bilinear map from A [[ν ]] to itself,
which is associative. Given a, b ∈ A , we can write

μ (a, b) = ∑ μk (a, b) ν k ,
k∈N

which defines bilinear maps μi : A ×A → A for i ∈ N. By definition, μ is a formal

deformation of (A , ·, {· , ·}) if μ0 (a, b) = a · b and μ1 (a, b) − μ1 (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 μ .
xx Introduction

Consider for example R2r , with coordinates (x1 , . . . , x2r ) = (q1 , . . . , qr , p1 , . . . , pr ).

The Moyal product is the R[[ν ]]-bilinear product on C∞ (R2r )[[ν ]], which is given for
F, G ∈ C∞ (R2r ) by

∂ kF ∂ kG νk
μ (F, G) := ∑ ∑ Ji1 , j1 . . . Jik , jk
∂ xi1 . . . ∂ xir ∂ x j1 . . . ∂ x jr k!
,
k∈N 1i1 , j1 ,...,ik , jk 2r
 
0 1r
where J is the skew-symmetric matrix of size 2r, given by J := . The
−1r 0
Moyal product, also called the Moyal–Weyl product, defines an associative product
on C∞ (R2r )[[ν ]], whose leading terms satisfy μ0 (F, G) = FG and
r  
∂F ∂G ∂G ∂F
μ1 (F, G) − μ1 (G, F) = ∑ − ,
j=1 ∂qj ∂ pj ∂qj ∂ pj

where the latter right-hand 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, π ), 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 π .
Summarizing, when one deforms a commutative associative algebra, a Poisson
structure shows up and it plays a dominant rôle in the entire deformation process.

The examples. The main idea which led us to the writing of this book, and to its ac-
tual 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 mathe-
matics, 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 6–11, as follows.
• Constant Poisson structures, such as Poisson’s original bracket (0.1), are accord-
ing 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 Kähler 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 Chap-
ter 6.
Introduction xxi

• Linear Poisson structures are in one-to-one correspondence with Lie algebras and
many features of Lie algebras are most naturally approached in terms of the canon-
ical Poisson bracket on the dual of a (finite-dimensional) Lie algebra, the so-called
Lie–Poisson structure. The coadjoint orbits, for example, of a Lie algebra are the
symplectic leaves of the Lie–Poisson structure, showing on the one hand that they
are even dimensional, and on the other hand that they carry a canonical symplectic
structure, the Kostant–Kirillov–Souriau symplectic structure. Linear Poisson struc-
tures appear also on an arbitrary Poisson manifold (M, π ) at any point x where the
Poisson structure vanishes: the tangent space Tx M inherits a linear Poisson struc-
ture, the linearization of π at x. It leads to the linearization problem, which inquires
if π 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 struc-
tures 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 struc-
ture to an adjoint orbit in a semi-simple Lie algebra. Moreover, this Poisson structure
arises in the case of the subregular orbit from a Nambu–Poisson 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 non-trivial. For example, their local clas-
sification has up to now only been accomplished under quite strong regularity as-
sumptions 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 one-form), which eventually leads to the symplectic foliation. A surface in C3
which is defined by the zero locus of a polynomial ϕ inherits a Poisson structure
from the standard Nambu–Poisson structure on C3 , with ϕ 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 symplec-
tic 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 Lie-theoretical context, a linear Poisson structure, different from the canoni-
cal Lie–Poisson 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 struc-
ture, 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 orig-
inal Lie bracket on g is in the first case an element r ∈ g ⊗ g, called an r-matrix,
while it is in the second case a (vector space) endomorphism R of g, called an R-
matrix. These structures appeared first in the theory of integrable systems, but are
xxii Introduction

nowadays equally important in the theory of Lie–Poisson groups (see the next item)
and of quantum groups. In the context of Lie algebras which come from associative
algebras, r-matrices and R-matrices also lead to a quadratic Poisson structure on
the Lie algebra, which plays an important rôle in the theory of Lie–Poisson groups;
they also lead to a cubic Poisson structure, whose virtue seems at this point still very
mysterious. Poisson structures coming from r-matrices or R-matrices are the subject
of Chapter 10.
• A Poisson–Lie group is a Lie group G, equipped with a Poisson structure π for
which the group multiplication G × G → G is a Poisson map. The Poisson struc-
ture π 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 r-
matrix. Conversely, every finite-dimensional Lie bialgebra is obtained in this way
from a Poisson–Lie group, a result which extends Lie’s third theorem (every finite-
dimensional Lie algebra is the Lie algebra of a Lie group). Poisson–Lie groups have
many applications, for example in the description of the Schubert cells of a Lie
group. Poisson–Lie groups are discussed in Chapter 11.

Theoretical foundations. The examples (Chapters 6–11) are preceded by a first

part (5 chapters) with a systematic exposition of the general theory of Poisson struc-
tures. 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 construc-
tions and the vast majority of examples. In the algebraic context, we have a Poisson
algebra, whose underlying space is an F-vector 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 con-
texts, 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 alge-
braic 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 (commu-
tative 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 vec-
tor 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 famil-
iarity with these chapters should suffice for reading a good part of any other chapter
of the book. Chapter 3 deals with multi-derivations and their geometrical analog,
multivector fields, and with Kähler forms, which are the algebraic analogs of dif-
ferential 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 de-
formation 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 6–11) 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.
Quasi-Poisson structures and Poisson structures with background are not only math-
ematically speaking very interesting, see [112], they have in addition non-trivial ap-
plications 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 Drinfel’d, has several connections with the topics which are de-
veloped 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 in-
tegration 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 copy-and-paste proofs
from existing proofs, we did not think it was necessary to cite the proof in the lit-
erature 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 Weinstein’s splitting theorem or the Pois-
son and Poisson–Dirac 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 me-
chanical 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 Poisson’s 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 infinite-dimensional) vec-
tor space A two different algebra structures: (1) a commutative, associative multi-
plication, (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 man-
ifolds, 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 spe-
cialize the results of Sections 1.2 and 1.3 to the case of Poisson structures on a
finite-dimensional 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. Laurent-Gengoux et al., Poisson Structures, 3

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 1, © Springer-Verlag Berlin Heidelberg 2013
4 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 mor-
phism of Poisson algebras. The Hamiltonian operator, which associates to an ele-
ment 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 F-vector space A equipped with two mul-
tiplications (F, G) → F · G and (F, G) → {F, G}, such that
(1) (A , ·) is a commutative associative algebra over F, with unit 1;
(2) (A , {· , ·}) is a Lie algebra over F;
(3) 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 skew-symmetric bilinear map {· , ·} : A × A → A , which satisfies condi-
tion (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 formu-
las, (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 cycli-
cally 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 5

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 deriva-
tions. 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 skew-
symmetric bilinear map P : A × A → A ; we say then that P is a skew-symmetric
biderivation of A . For such a biderivation, it is convenient to enclose its arguments
in square brackets; thus, when P is a skew-symmetric biderivation, we write the first
equation in (1.4) as

P[FG, H] = F P[G, H] + G P[F, H] ;

by the skew-symmetry 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 skew-symmetric multi-derivation, 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 F-vector space of skew-symmetric biderivations of A is denoted by X2 (A ).
As recalled in Appendix A, a skew-symmetric bilinear map on A can be viewed as
a linear map on ∧2 A = A ∧ A ; we therefore often abbreviate ψ : A × A → A to
ψ : ∧2 A → A or ψ : A ∧ A → A . According to Definition 1.1, a Poisson bracket
on A is in particular a skew-symmetric 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 φ : A1 → A2 which satisfies, for all F, G ∈ A1 ,
(1) φ (F ·1 G) = φ (F) ·2 φ (G);
(2) φ ({F, G}1 ) = {φ (F), φ (G)}2 ,
is called a morphism of Poisson algebras.
It is clear that if φ : A1 → A2 is a morphism of Poisson algebras and φ is bijective,
then φ −1 : A2 → A1 is also a morphism of Poisson algebras. We say then that φ is
an isomorphism of Poisson algebras.
6 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 φ 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 and {B, B} ⊂ B , (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 restric-
tions of both multiplications to B and the inclusion map ı : 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 F-vector space of Hamiltonian derivations of A , so that we have an F-linear

map
1.2 Poisson Varieties 7

X : A → Ham(A , {· , ·})
(1.7)
H → XH := {· , H} .
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 F-vector 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) Cas(A ) is a subalgebra of (A , ·), which contains the image of F in A , under
the natural inclusion a → a · 1;
(2) If A has no zero divisors, then Cas(A ) is integrally closed in A ;
(3) Ham(A ) is not an A -module (in general); instead, XFG = FXG + GXF , for
every F, G ∈ A ;
(4) Ham(A ) is a Cas(A )-module;
(5) 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 );
(6) 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 proper-
ties (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
8 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 as-
sumed 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) alge-
bra, which can be considered as an algebra of functions on M, since the evalua-
tion of elements of F[x1 , . . . , xd ]/I at points of M is well-defined. 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 F|M , 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 func-
tions F and G in A , their Poisson bracket is given by
d   ∂F ∂G
{F, G} = ∑ xi , x j
∂ xi ∂ x j
. (1.8)
i, j=1

1 As we said in the introduction to this chapter, Sections 1.2 and 1.3 can be read independently.
1.2 Poisson Varieties 9

Proof. The proof is based on the standard argument that two biderivations (or p-
derivations, 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 right-hand side in (1.8) is obviously zero, but
also the left-hand 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 non-constant monomials, such that deg(F) + deg(G) = n + 1. By skew-
symmetry, 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   ∂ F2 ∂ G d   ∂ F1 ∂ G
= F1 ∑ xi , x j
∂ xi ∂ x j
+ F2 ∑ xi , x j
∂ xi ∂ x j
i, j=1 i, j=1
d   ∂F ∂G
= ∑ xi , x j
∂ xi ∂ x j
.
i, j=1

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 mor-
phism of varieties Ψ : M1 → M2 is called a Poisson morphism or a Poisson map if
the dual morphism
Ψ ∗ : 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) Ψ : M1 → M2 , the dual morphism Ψ ∗ is (well-)defined by Ψ ∗ (F) :=
F ◦Ψ , for all F ∈ F (M2 ), so that the first condition in Definition 1.2 is automatically
satisfied:

Ψ ∗ (FG) = (FG) ◦ Ψ = (F ◦ Ψ ) (G ◦ Ψ ) = (Ψ ∗ F) (Ψ ∗ G) .

In this case, the second condition, which says that

Ψ ∗ : (F (M2 ), {· , ·}2 ) → (F (M1 ), {· , ·}1 )

is a morphism of Lie algebras, can also be written as

{F, G}2 ◦ Ψ = {F ◦ Ψ , G ◦ Ψ }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 ı :

N → M is a morphism, then N is said to be a Poisson subvariety if N admits a
Poisson structure such that ı 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
space Fd (with its algebra of regular functions A := F[x1 , . . . , xd ]). The 2
 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 , (1.9)
d  ∂ x jk

∂ xi j ∂ xki
∑ xk ∂ x + xi ∂ x + x j ∂ x = 0 , (1.10)
=1

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 (skew-symmetric) 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 i-th component of [dF] is ∂∂ xFi . The matrix X ∈
Matd (A ) is called the Poisson matrix of (A , {· , ·}).
A natural question is: given a skew-symmetric matrix X = (xi j ) ∈ Matd (A ),
where A = F[x1 , . . . , xd ], does the corresponding skew-symmetric biderivation, de-
fined for F, G ∈ A by
d
∂F ∂G
{F, G} := ∑ xi j
i, j=1 ∂ xi ∂ x j
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
d
∂ xi j ∂ F ∂ G ∂ H
{{F, G} , H} +  (F, G, H) = ∑ xk
∂ x ∂ xi ∂ x j ∂ xk
+  (F, G, H) (1.11)
i, j,k,=1
1.2 Poisson Varieties 11
 
d d
∂ xi j ∂F ∂G ∂H
= ∑ ∑ k ∂ xx +  (i, j, k)
∂ xi ∂ x j ∂ xk
,
i, j,k=1 =1

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 skew-symmetric d × d matrix, with elements in
A = F[x1 , . . . , xd ], then
d
∂F ∂G
{F, G} := ∑ xi j (1.12)
i, j=1 ∂ xi ∂ x j
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 k
∂ x
+ x i
∂ x
+ x  j
∂ x
=0.
=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 skew-symmetric
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 skew-symmetric biderivation {· , ·}0 of F[x1 , . . . , xd ],
defined for F, G ∈ F[x1 , . . . , xd ] by
d
∂F ∂G
{F, G}0 := ∑ xi j
∂ xi ∂ x j
. (1.13)
i, j=1

We
 would like to define a skew-symmetric biderivation {· , ·} of A by setting
F, G := {F, G}0 , for all F, G ∈ F[x1 , . . . , xd ]. The problem is that, in general, this
is not well-defined. Namely, for every K ∈ I one has that K = 0, so that a necessary
condition for the bracket {· , ·} on A to be well-defined, is that for every K ∈ I and
1  i  d,
d
∂K
{xi , K}0 = ∑ xi j ∈I . (1.14)
j=1 ∂xj
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 well-defined. Also, {· , ·} is a skew-symmetric biderivation since {· , ·}0 is one.
Since {· , ·} is well-defined,
  
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 skew-symmetric 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 skew-symmetric biderivation {· , ·}0 on F[x1 , . . . , xd ]
by (1.13). Since {xi , ·} is a derivation of A , we have that

  d   ∂K d
∂K
0 = xi , K = ∑ xi , x j
∂xj
= ∑ xi j ∂ x j = {xi , K}0 ,
j=1 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 skew-symmetric matrix with coefficients in A . Picking representatives
xi j ∈ F[x1 , . . . , xd ] of xi j , with xi j = −x ji , define a skew-symmetric biderivation {· , ·}0
of F[x1 , . . . , xd ] by
d
∂F ∂G
{F, G}0 := ∑ xi j . (1.15)
i, j=1 ∂ xi ∂ x j
1.2 Poisson Varieties 13

If, for every 1  j  d and for every K ∈ I ,

d
∂K
∑ ∂ 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 representa-
tives 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

  d   ∂F ∂G
F, G = ∑ xi , x j
∂ xi ∂ x j
, (1.18)
i, j=1

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 infinite-dimensional, 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
   
X(m) := ( xi , x j (m))1i, jd and X  (m) := ( xi , x j (m))0i, 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),
d   ∂F
{xi , x0 } (m) = ∑ xi , x j (m)
∂xj
(m) ,
j=1

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
maxm∈M 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 well-defined skew-symmetric bilinear map:

I m Im F (M)
πm : × → F. (1.19)
Im2 Im2 Im

The bilinear map πm 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 πm in terms of the – possi-
bly linearly dependent – equivalence classes [xi − xi (m)] ∈ Im /Im2 , then the matrix
which we get has entries
   
πm ([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 πm 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) For every m ∈ M, the rank of {· , ·} at m is even;
(2) The rank of {· , ·} is at most equal to the dimension of M;
(3) 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.
Proof. Writing F (M) as F[x1 , . . . , xd   Poisson matrix of {· , ·} at m ∈ M is
]/I , the
the skew-symmetric 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→ gld
   (1.20)
m → xi , x j (m) 1i, jd

it is open; since the topology which is considered here is the Zariski topology, these
open subsets are dense as soon as they are non-empty. 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 previ-
ous 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 geomet-
ric analog of a skew-symmetric 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 Ψ : M → N be a map”, which the reader
may specialize to “Let M and N be two differentiable manifolds and let Ψ : M → N
be a smooth map” or to “Let M and N be two complex manifolds and let Ψ : 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,

δm (Fm Gm ) = F(m) δm Gm + G(m) δm 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 Ψ : M → N between manifolds leads for every m ∈ M
to a linear map TmΨ : Tm M → TΨ (m) N, called the tangent map of Ψ 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 non-empty 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
(1.21)
m → Vm Fm .

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 func-
tion, 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

Fm (M) Fm (M)
Bm : × →F
∼ ∼
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 skew-symmetric biderivations. Given two point-

wise derivations δm and εm at m, we construct a pointwise skew-symmetric bideriva-
tion at m by  
 
 δm Fm δm Gm 
 
(δm ∧ εm )(Fm , Gm ) :=  ,
 
 εm Fm εm Gm 
for all functions F and G, defined in a neighborhood of m. It follows easily from the
Hadamard lemma (Lemma B.2) that the basic pointwise skew-symmetric bideriva-
tions ∂∂xi ∧ ∂∂x j span the vector space of all pointwise skew-symmetric
m m
biderivations at m; therefore, we denote the latter space by ∧2 Tm M.
The passage from pointwise skew-symmetric biderivations to bivector fields is
similar to the passage from pointwise derivations to vector fields, detailed in Ap-
pendix B. Namely, we consider a map P which assigns to every point m of M an ele-
ment 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 coordi-
nate 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 ] ∧
∂ xi ∂ x j
. (1.23)
1i< jd

In differential geometric terms, one often refers to bivector fields as being (skew-
symmetric) 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
one-forms, 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
∂F ∂G
P[F, G](m) = ∑ P[xi , x j ](m)
∂ xi
(m)
∂xj
(m) , (1.24)
i, j=1

where ∂∂ xFi (m) is the i-th component of dm F, and similarly for ∂G

∂ x j (m): in view
of (1.22),  
∂ ∂ ∂F
dm F, = [F](m) = (m) .
∂ xi m ∂ xi ∂ xi
The fact that P[F, G](m) depends on dm F and dm G only, implies that we can define
a skew-symmetric F (M)-bilinear map from Ω 1 (M), the F (M)-module of differ-
ential one-forms on M, to F (M): simply put

P̂(dF, dG) := P[F, G] .

We will in the sequel usually make no distinction between the bivector field P and
the (2, 0)-tensor P̂, and denote both by the same letter P. Note also that the tensorial
interpretation, or equivalently (1.23), implies that we may specialize P to points
m ∈ M, giving a skew-symmetric, F-bilinear 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 Ψ∗ P of a bivector
field P by a diffeomorphism Ψ : M → N between two manifolds. The bivector field
Ψ∗ P on N is defined by setting3

(Ψ∗ P)Ψ (m) := ∧2 (TmΨ )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.
3 For a linear map φ : V → W between two vector spaces, the linear map ∧2 φ : ∧2V → ∧2W is
defined by (∧2 φ )(v ∧ w) := φ (v) ∧ φ (w), for all v, w ∈ V . See Appendix A for details and general-
izations.
1.3 Poisson Manifolds 19

Remark 1.14. Vector fields on a manifold M can be defined as sections of its tan-
gent 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 holomor-
phic) 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 π on M leads for every open subset U
of M to a skew-symmetric product on F (U), which is demanded to make F (U)
into a Poisson algebra.
Definition 1.15. Let π be a bivector field on a manifold M. We say that π is a Pois-
son structure on M if for every open subset U of M, the restriction of π to U makes
F (U) into a Poisson algebra. We then call (M, π ) a Poisson manifold.
The Poisson structure π of a Poisson manifold (M, π ) will often be denoted by {· , ·},
in particular the Poisson bracket π [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 π [F, G]
will be referred to as the bracket notation.
Let (U, x) be a coordinate neighborhood of a d-dimensional Poisson manifold
(M, π ). On U, the bivector field π can, according to (1.23), be written as
  ∂ ∂
π= ∑ xi , x j
∂ x i
∧
∂ xj
, (1.26)
1i< jd
 
so that the matrix X := xi , x j 1i, jd encodes the restriction of π to U. This
matrix X, whose elements belong to F (U), is called the Poisson matrix of π 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 skew-symmetric
bilinear map πm : Tm∗ M × Tm∗ M → F, defined by the Poisson structure at m.
Proposition 1.16. Let π = {· , ·} be a bivector field on a manifold M of dimension d.
Then π 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 ; (1.27)

(ii) 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 1 Poisson Structures: Basic Definitions

(iii) 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
∂ xi j ∂ x jk ∂ xki
∑ xk
∂ x
+ xi
∂ x
+ x j
∂ x
=0, (1.28)
=1
 
where xi j := xi , x j , for 1  i, j  d;
(iv) 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 π makes every F (U) into a Poisson algebra,
hence π 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)
d
∂ xi j ∂ F ∂ G ∂ H
= ∑ xk
∂ x ∂ xi ∂ x j ∂ xk
+  (F, G, H) (1.30)
i, j,k,=1
d  
d
∂ xi j ∂F ∂G ∂H
= ∑ ∑ xk +  (i, j, k) ,
i, j,k=1 =1 ∂ x  ∂ xi ∂ x j ∂ xk

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 func-
tions 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 Ψ : M → N between Poisson manifolds (M, π ) and (N, π  )
is called a Poisson map if for all open subsets U ⊂ M and V ⊂ N, with Ψ (U) ⊂ V , the
induced map Ψ ∗ : F (V ) → F (U), which is defined for all F ∈ F (V ) by Ψ ∗ (F) :=
F ◦ Ψ , is a morphism of Poisson algebras, i.e., for all functions F, G ∈ F (V ),

{F ◦ Ψ , G ◦ Ψ } = {F, G} ◦ Ψ , (1.31)

where {· , ·} := π and {· , ·} := π  .

1.3 Poisson Manifolds 21

The condition that Ψ : 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 Ψ : M → N be a map between two Poisson manifolds (M, π )
and (N, π  ). Then Ψ is a Poisson map if and only if

∧2 (T Ψ )π = π  , (1.32)

i.e., ∧2 (TmΨ )πm = πΨ (m) for every m ∈ M.

Proof. Let F and G be functions, defined on a neighborhood of Ψ (m) in N, and

let (U, x) be a coordinate chart for M around m, so that π can be written on U as
π = ∑i< j xi j ∂∂xi ∧ ∂∂x j .
     
∂ ∂
∧2 (TmΨ )πm [F, G] = ∑ ∧ (TmΨ ) xi j (m)
2
∧ [F, G]
∂ xi m ∂xj m
1i< jd    
∂ ∂
= ∑ xi j (m) TmΨ ∧ TmΨ [F, G]
∂ x i m ∂ xj m
1i< jd    
∂ ∂
= ∑ xi j (m) ∧ [F ◦ Ψ , G ◦ Ψ ]
1i< jd ∂ x i m ∂ xj m
= {F ◦ Ψ , G ◦ Ψ } (m) ,

where we have set, as usual, {· , ·} := π . Setting {· , ·} := π  we clearly have that

πΨ (m) [F, G] = {F, G} (Ψ (m)) ,

so that ∧2 (TmΨ )πm = πΨ (m) if and only if {F ◦ Ψ , G ◦ Ψ } (m) = {F, G} (Ψ (m)) for
all functions F and G, defined in a neighborhood of Ψ (m). This proves that Ψ is a
Poisson map if and only if ∧2 (T Ψ )π = π  .

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 Hamilto-
nian 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
d   ∂H ∂
XH = ∑ xi , x j
∂ x j ∂ xi
. (1.33)
i, j=1

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 d   ∂H
dt
= {xi , H} = ∑ xi , x j
∂xj
, i = 1, . . . , d . (1.34)
j=1

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, π ) and a point m ∈ M the integer Rk πm ,
which is also the rank of every Poisson matrix of π at m, is called the rank of π at m,
denoted Rkm π . One says that (M, π ) 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 π is said to be maximal at
m ∈ M, when it coincides with the dimension of M. The maximum maxm∈M Rkm π
is called the rank of π , denoted Rk π . A point m ∈ M is said to be a regular point
of (M, π ), if Rkm π = Rk π , otherwise it is said to be a singular point of (M, {· , ·}).
The set of singular points of (M, π ) is called the singular locus of (M, π ).

The main properties of the rank are given in the following proposition.
Proposition 1.21. Let (M, π ) be a Poisson manifold and let m ∈ M.
(1) Rkm π is even and is equal to dim Hamm (M);
(2) For every s ∈ N, the subset M(s) of M, defined by

M(s) := {m ∈ M | Rkm π  2s}

is open (and dense in the complex case); in particular, the subset of points
m ∈ M such that Rkm π = Rk π is open.

Proof. The rank of π at m is the rank of the bilinear map πm , so it is the dimension
of the image of the linear map Tm∗ M → Tm M, defined by ξ → πm (ξ , ·). In turn,
the latter equals the dimension of Hamm (M), since every ξ ∈ Tm∗ M can be written
as dm F, for some function F, defined in a neighborhood of m. Also, since πm is
skew-symmetric, 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 maxi-
mum on a dense open subset. This is shown in the following example.
Example 1.22. Let ϕ 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 ϕ ∂ /∂ 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, π ) be a Poisson manifold. The Lie derivative of π 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 π , which is the Lie derivative of π 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 := π = {· , ·} and
V := XH , with H ∈ F (U), this yields

LXH π [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 π = 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 , ·, π ) one says that a derivation V of (A , ·)

is a Poisson derivation if the Lie derivative of π 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: Weinstein’s Splitting Theorem

This section deals with Weinstein’s splitting theorem, which states that, in the neigh-
borhood of a point where the rank of the Poisson structure is 2r, the Poisson mani-
fold 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 (Weinstein’s splitting theorem). Let (M, π ) be a (real or com-
plex) Poisson manifold, let m ∈ M be an arbitrary point and denote the rank
24 1 Poisson Structures: Basic Definitions

of π at m by 2r. There exists a coordinate neighborhood U of m with coordi-

nates q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs , centered at m, such that, on U,
r
∂ ∂ ∂ ∂
π=∑ ∧ + ∑ φk (z) ∧ , (1.36)
i=1 ∂ q i ∂ p i 1k<s ∂ zk ∂ z

where the functions φk 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, π ) and for every point m such that the rank of π 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, π ) be a Poisson manifold and let
m ∈ M be a point for which Rkm π = 2r. We will show that the theorem holds for
(M, π ) at m.
Since Rkm π > 0, there exists a function p on a neighborhood of m, whose Hamil-
tonian 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 sys-
tem of coordinates (q, y2 , . . . , yd ) on a neighborhood U  of m, centered at m, such
that X p = ∂ /∂ q. Writing {· , ·} = π , it follows that

∂q
{q, p} = X p [q] = =1, [Xq , X p ] = X{p,q} = −X1 = 0 and X p [yi ] = 0 ,
∂q

for i = 2, . . . , d, on U  . Writing Xq in terms of these coordinates,

∂ d
∂
X q = ξ1 + ∑ ξi  ,
∂ q i=2 ∂ yi

we have that ξ1 = Xq [q] = {q, q} = 0, and all coefficients ξ2 , . . . , ξd are independent

of q, since Xq and X p = ∂ /∂ q commute. At m, we have

d
∂p
−1 = {p, q} (m) = Xq [p](m) = ∑ ξi (m) (m) ,
i=2 ∂ yi

so that the vector field Xq = ∑di=2 ξi ∂ /∂ yi is independent of q and does not van-
ish at m. Applying the straightening theorem once more, we may introduce a sys-
tem 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
∂
= − ∑ ξi  = −Xq .
∂p 
i=2 ∂ yi
1.3 Poisson Manifolds 25

Substituting p by p we consider (q, p, y3 , . . . , yd ) which is also a system of coordi-

nates 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 π is given, in terms of

the coordinates q, p, y3 , . . . , yd , by

∂ ∂ ∂ ∂
π= ∧ + ∑ {yk , y } ∧ . (1.37)
∂ q ∂ p 3k<d ∂ yk ∂ y

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 π . The Jacobi identity also yields that the second term in (1.37) defines
a Poisson structure π  on a neighborhood V of the origin o of Fd−2 . The Poisson
matrix of π with respect to q, p, y3 , . . . , yd has a diagonal block form, where the
lower block is the Poisson matrix of π  with respect to y3 , . . . , yd . Thus, π  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 π  takes on V the following
form:
r
∂ ∂ ∂ ∂
π = ∑ ∧ + ∑ φk (z) ∧ .
i=2 ∂ qi ∂ pi 1k<s ∂ zk ∂ z
In terms of the system of coordinates (q1 , q2 , . . . , qr , p1 , p2 , . . . , pr , z1 , . . . , zs ), which
is centered at m, π 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 Φ
26 1 Poisson Structures: Basic Definitions
 
where Φ ∈ Mats (F (M)) is given by Φi j := zi , z j and 1r denotes the identity
matrix of size r. It follows that Φ = 0 in a neighborhood of m, when the rank of
π 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, π ) be a (real or complex) Poisson
manifold of dimension d, and suppose that m is a point where the rank of π 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,
r
∂ ∂
π=∑ ∧ . (1.38)
i=1 ∂ qi ∂ pi

Moreover, π 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) d-
dimensional Poisson manifold (M, π ) is its rank Rk π .
To close this section we state a generalization of Weinstein’s splitting theorem
which we will use in Chapter 12. It provides a set of Darboux coordinates for the
Poisson structure π , which contains a given set p1 , . . . , pr of functions in involution
(i.e., functions which pairwise commute for the Poisson structure), whose Hamilto-
nian 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, π ) 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 , . . . , zd−2r , such that
(1) The d functions (q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd−2r ) form a system of coordi-
nates on U, centered at m;
(2) The Poisson structure π is given on U by
r
∂ ∂ d−2r
∂ ∂
π=∑ ∧ + ∑ gi j (z) ∧ , (1.39)
i=1 ∂ q i ∂ p i i, j=1 ∂ zi ∂ zj

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 π 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 struc-
ture; 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, π ) is an arbitrary Poisson manifold.
The idea which underlies the decomposition is based on the following two ob-
servations. 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 γ , defined on an open neigh-

borhood5 of [0, 1], with γ (0) = m and γ (1) = m , which is an integral curve of a
Hamiltonian vector field XF , where F is a function, defined on an open neighbor-
hood of γ ([0, 1]). More generally, when points m0 , . . . , mN in M are such that there
exists a Hamiltonian path from mi−1 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 well-defined 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 defini-
tion path-connected, 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 natu-
rally 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 ı : N → M is an immer-
sion, i.e., for every m ∈ N the tangent map Tm ı : 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, π ) whose rank Rkm π 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 neigh-
borhood 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
r
∂ ∂ ∂ ∂
π=∑ ∧ + ∑ φk (z) ∧ , (1.40)
i=1 ∂ q i ∂ p i 1k<s ∂ zk ∂ z

where the functions φk (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 πS . The map ı : (Sm (U, x), πS ) → (M, π ) 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} (ı(m )) = {F, G} (m ) = ∑
  (m ) (m ) − (m ) (m )
i=1 ∂ qi ∂ pi ∂ pi ∂ qi
= {F ◦ ı, G ◦ ı}S (m ) .

In the language of Section 2.2, (Sm (U, x), πS ) is a Poisson submanifold of (M, π ).
Notice that πS 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, π ) and let x =
(q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs ) be splitting coordinates at m, on a coordinate neigh-
borhood 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
s
∂F
XF [zi ](m ) = {zi , F} (m ) = ∑ φi (m ) ∂ z (m ) = 0 ,
=1

since the functions φi vanish at all points m of Sm (U, x). It follows that all Hamil-
tonian 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

⎧
⎪
⎪ dqi ∂H
⎪
⎨ = = qi (m ) ,
XH : dt ∂ p i
⎪
⎪ dp ∂H
⎪
⎩ 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

γ : t → (q1 (m )t, . . . , qr (m )t, p1 (m )t, . . . , pr (m )t, 0, . . . , 0) .

The image of γ is contained in U, because x(U) ⊂ F2r+s is a convex open subset.

Since γ is a Hamiltonian path from m = γ (0) to m = γ (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 holo-
morphic). Since the topology and the differential structure on these subsets Sm (U)
is intrinsically defined, as embedded submanifolds of M, these definitions yield a
well-defined manifold structure on Sm (M). In general, the topology which we de-
fined 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 well-defined Poisson structure of maximal rank (symplectic structure) on
Sm (M). The results of the present section, combined, prove the following funda-
mental theorem.

Theorem 1.30. Every Poisson manifold (M, π ) is the disjoint union of immersed
submanifolds, whose tangent spaces are spanned by the Hamiltonian vector fields
of (M, π ). 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, π ) be a Poisson manifold of dimension d and rank 2r.
(1) A function F on M is a Casimir function if and only if F is constant on each
symplectic leaf;
(2) The (non-empty open subsets of the) symplectic leaves are the smallest em-
bedded manifolds of M which are Poisson submanifolds (i.e., such that the
inclusion map is a Poisson map);
(3) 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 π is 2r.
Proof. The fact that the manifold Sm (M) admits locally (in the sense of the topol-
ogy 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 π at m is 2r, so that m is a regular point of π , then there exists local
coordinates q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zd−2r in a contractible neighborhood U of m
with respect to which the Poisson structure π is given by:
r
∂ ∂
π=∑ ∧ . (1.41)
i=1 ∂ q i ∂ pi

The vector fields ∂∂q , ∂∂p , . . . , ∂∂qr , ∂∂pr span the symplectic leaves of π on U. There-
1 1
fore, every vector field Y , which is tangent to the symplectic leaves of π , is of the
form
r
∂ r
∂
Y = ∑ Fi + ∑ Gi (1.42)
i=1 ∂ p i i=1 ∂ qi
for some smooth or holomorphic functions F1 , . . . , Fr , G1 , . . . , Gr , defined on U. Sup-
pose now that Y is a Poisson vector field, so that LY π = 0. Then one computes
easily from (1.35), (1.41) and (1.42) that

∂ Fi ∂ Fj ∂ Gi ∂Gj ∂ Fi ∂ Gi
= , = and =− ,
∂qj ∂ qi ∂ pj ∂ pi ∂ pj ∂qj

for i, j = 1, . . . , r. By the classical Poincaré lemma, there exists a function H, defined

on the contractible subset U, which satisfies
∂H ∂H
Fi = − and Gi = ,
∂ qi ∂ pi
for i = 1, . . . , r. Hence,
r
∂H r
∂H d−2r
∂H
XH = ∑ X qi + ∑ X pi + ∑ Xzk = Y ,
i=1 ∂ q i i=1 ∂ p i k=1 ∂ zk

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, π ) be a Poisson manifold of dimension d, let U be a non-
empty open subset of M and let F1 , . . . , Fs ∈ F (U), satisfying:
(1) The rank of π is constant on U and is equal to d − s;
(2) The functions F1 , . . . , Fs are Casimirs of the restriction of π to U;
(3) For every point m of U, the differentials dm F1 , . . . , dm Fs are independent.
Then the symplectic foliation of the restriction of π 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 π 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 /Λ , where Λ is a
lattice in R3 . Unless Λ 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 π1 and π2 be two Poisson structures on a manifold M. Sup-

pose 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 π1 is the same as the Poisson structure induced on S by π2 . Then
π1 and π2 are equal.

Proof. Let π1 = {· , ·}1 and π2 = {· , ·}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 ıS the inclusion map ıS : S → M. It
is assumed that the induced Poisson structures {· , ·}i,S on S coincide, so that
{F ◦ ıS , G ◦ ıS }1,S = {F ◦ ıS , G ◦ ıS }2,S for every F, G ∈ F (M). Since ıS is a
Poisson map, when M is equipped with π1 , or with π2 , we have for every m ∈ S ,

{F, G}1 (m) = {F ◦ ıS , G ◦ ıS }1,S (m) = {F ◦ ıS , G ◦ ıS }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 finite-dimensional F-vector space V , one can consider its algebra of polyno-

mial 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 dif-
ference 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 coordi-
nates.
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 func-
tions 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 poly-
nomial 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 deriva-
tions 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 , P= ∑ Pi j ∧ ,
i=1 ∂ xi 1i< jd ∂ 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 π = {· , ·} on V and that we wish to

express that π is a Poisson structure. When F (V ) is the algebra of polynomial
functions on V , then π is a Poisson structure if and only if π 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 holo-
morphic functions on V , precisely the same condition expresses that π 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 finite-dimensional vector space, with algebra of func-
tions F (V ) and let (x1 , . . . , xd ) be a system of linear coordinates on V . Given
a skew-symmetric matrix (xi j )1i, jd of elements of F (V ), consider the skew-
symmetric biderivation π of F (V ), which is defined, for F, G ∈ F (V ), by
d
∂F ∂G
{F, G} := ∑ xi j
∂ xi ∂ x j
,
i, j=1
 
which is the unique skew-symmetric biderivation of F (V ) such that xi , x j = xi j ,
for all 1  i, j  d. The following conditions are equivalent.
(i) π is a Poisson structure on V ;
(ii) π 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
∂ xi j ∂ x jk ∂ xki
∑ xk
∂ x
+ xi
∂ x
+ x j
∂ x
=0.
=1

In this case, the skew-symmetric matrix (xi j )1i, jd is the Poisson matrix of π 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 Tm∗V is canonically isomorphic to V ∗ . Therefore,
one usually thinks of the Poisson bivector at m, which is a skew-symmetric bilinear
map πm : Tm∗V × Tm∗V → F, as a skew-symmetric 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 non-empty 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 [F] = ∑ V [xi ] .
i=1 ∂ xi

Deduce from it an alternative proof of Proposition 1.6.

2. Two elements F, G of a Poisson algebra (A , ·, π ) are said to be in involution
if {F, G} = 0. Prove Poisson’s 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
Poisson’s theorem is that, since functions which are in involution with the Hamilto-
nian are constants of motion, taking the Poisson bracket of two constants of motion
produces a (a priori new) constant of motion.
3. Let π 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 , π1 ) and (M2 , π2 ) be two affine Poisson varieties, or two Poisson mani-
folds. Let Ψ : M1 → M2 be a Poisson map. Show that

Rkm M1  RkΨ (m) M2 ,

for every m ∈ M1 .
5. Given a skew-symmetric biderivation {· , ·} on a commutative associative alge-
bra A , define its Jacobiator J as the skew-symmetric tri-linear 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 Proposi-
tion 1.8.
6. Consider on F2 the bivector field π , 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 fi-
nite type, do there  exist  generators x1 , . . . , xd of A and representatives xi j = −x ji ∈
F[x1 , . . . , xd ] of xi , x j such that
 
d
∂ xi j ∂ x jk ∂ xki
∑ xk ∂ x + xi ∂ x + x j ∂ x = 0 ,
=1
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 consid-
ered 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 man-
ifold (see Cannas da Silva–Weinstein [34] or Dufour–Zung [63]) or as sheafs of
Poisson algebras on the manifold (see Przybylski [172]). Abstract Poisson algebras
are considered in Bhaskara–Viswanath [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 Weinstein’s 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 Dufour–Zung [63],
while a generalization to arbitrary Lie algebroids is given in Fernandes [75]. See
Laurent–Miranda–Vanhaecke [120] for the Carathéodory–Jacobi–Lie theorem for
Poisson manifolds, which was stated without proof in Theorem 1.28 above; for reg-
ular 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 construc-
tion 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 geomet-
rically 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 struc-
ture 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 Sec-
tion 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. Laurent-Gengoux et al., Poisson Structures, 37

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 2, © Springer-Verlag Berlin Heidelberg 2013
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 Pois-

son 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 (Propo-
sition 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 F-bilinear 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
(2.2)
F1 ⊗ F2 → V [F1 ] ⊗ F2 .

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
a derivation W of A2 , we can associate in a natural way a derivation W of A1 ⊗ A2 ,
such that W[j2 (F)] = j2 (W[F]), for every F ∈ A2 .
By extending this construction to the case of skew-symmetric biderivations, we
can construct a skew-symmetric biderivation of A1 ⊗ A2 (with values in A1 ⊗ A2 ),
starting from a skew-symmetric 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 Poisson-
commuting 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 = XH⊗1 ,

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 π1 : A1 × A1 → A1 be a skew-symmetric biderivation of A1 . Then it can
be verified, as in the case of derivations, that the skew-symmetric bilinear map π1 ,
defined by

π1 : (A1 ⊗ A2 ) × (A1 ⊗ A2 ) → A1 ⊗ A2

(2.4)
(F1 ⊗ F2 , G1 ⊗ G2 ) → π1 [F1 , G1 ] ⊗ (F2 G2 )

is a biderivation of A1 ⊗ A2 . If π1 satisfies moreover the Jacobi identity, so that π1

is a Poisson bracket on A1 , then π1 is a Poisson bracket on A1 ⊗ A2 , since
   
π1 π1 [F1 ⊗ F2 , G1 ⊗ G2 ], H1 ⊗ H2 = π1 π1 [F1 , G1 ], H1 ⊗ (F2 G2 H2 ) .

Notice also that π1 has the following properties:

π1 [F1 ⊗ 1, G1 ⊗ 1] = π1 [F1 , G1 ] ⊗ 1 ,

π1 [F1 ⊗ 1, 1 ⊗ G2 ] = 0 ,
π1 [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 π1 as a Poisson bracket on A1 ⊗
A2 ; according to the second formula, j1 and j2 have Poisson-commuting 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 π2 denote a Poisson bracket on A2 and taking the sum π := π1 +
π2 , we obtain a skew-symmetric biderivation of A1 ⊗ A2 , and the above formulas

lead to
40 2 Poisson Structures: Basic Constructions

π [F1 ⊗ 1, G1 ⊗ 1] = π1 [F1 , G1 ] ⊗ 1 ,
π [F1 ⊗ 1, 1 ⊗ G2 ] = 0 ,
π [1 ⊗ F2 , 1 ⊗ G2 ] = 1 ⊗ π2 [F2 , G2 ] .
By the biderivation property, these formulas can be summarized in the single for-
mula

π [F1 ⊗ F2 , G1 ⊗ G2 ] = π1 [F1 , G1 ] ⊗ (F2 G2 ) + (F1 G1 ) ⊗ π2 [F2 , G2 ] (2.5)

for every Fi , Gi ∈ Ai , (i = 1, 2). It follows that π satisfies the Jacobi identity, be-
cause (2.5), combined with the fact that π is a biderivation, implies that

π [π [F1 ⊗ F2 , G1 ⊗ G2 ], H1 ⊗ H2 ]+  (F, G, H)
= π1 [π1 [F1 , G1 ], H1 ] ⊗ (F2 G2 H2 ) + (F1 G1 H1 ) ⊗ π2 [π2 [F2 , G2 ], H2 ]+  (F, G, H) .

Therefore, π is a Poisson bracket on A1 ⊗ A2 , both maps ji : Ai → A1 ⊗ A2 are

morphisms of Poisson algebras and they have Poisson-commuting images. Notice
that in order to establish formula (2.5) we have used only that π is a biderivation
of A1 ⊗ A2 , such that canonical inclusions ji : Ai → A1 ⊗ A2 are Lie algebra ho-
momorphisms and have Poisson-commuting 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 = π1 [F1 , H] ⊗ F2 = π [F1 ⊗ F2 , H ⊗ 1]
= XH⊗1 [F1 ⊗ F2 ] ,


so that X H = XH⊗1 , 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 iso-
morphism, 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:
p∗i
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 p∗1 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 , πi ) be two affine Poisson varieties, where i = 1, 2. The
product variety M1 × M2 has a unique Poisson structure π = {· , ·} such that the
projection maps pi : M1 × M2 → Mi (i = 1, 2) are Poisson maps and such that

{p∗1 F1 , p∗2 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 , π ) is called the product Poisson variety of (M1 , π1 )
and (M2 , π2 ), while π 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. A function F ∈ F (M1 × M2 ) is of the form p∗i Fi for some Fi ∈ F (Mi ) if and
only if it is constant on all the fibers of pi ;
2. A vector field V on M1 × M2 satisfies V [p∗i 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) 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 ;
(ii) The Hamiltonian vector field of every function which is constant on all the
fibers of p1 is tangent to all the fibers of p2 ;
(iii) 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 struc-
ture 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 ), re-
spectively y1 , . . . , yd2 for F (M2 ). The Poisson matrix for the product Poisson struc-
ture {· , ·} on M1 ×M2 takes, in terms of the d1 +d2 generators p∗1 x1 , . . . , p∗1 xd1 , p∗2 y1 ,
. . . , p∗2 yd2 for F (M1 × M2 ), the form
d1
p∗1 xi j i, j=1
0
d . (2.7)
0 p∗2 yi j i,2j=1

Proof. Since p1 is a Poisson map we have that

 ∗   
p1 xi , p∗1 x j = p∗1 xi , x j 1 = p∗1 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

p∗1 xi , p∗2 y j of the Poisson matrix are zero, as follows from (2.6). 

By a slight abuse of notation, one usually simply writes xi for ∗ ∗
 p1 xi and y j for p2 y j .
X 0
Then the Poisson matrix (2.7) takes the simple form and the product Pois-
0Y
son structure π = {· , ·} can be explicitly written in the following form

∂ ∂ ∂ ∂
π= ∑ xi j ∧ + ∑ yi j
∂ xi ∂ x j 1i< ∂ y i
∧
∂ yj
.
1i< jd1 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 en-
dowed with the product topology. Second, an atlas of M1 × M2 is constructed from
an atlas of M1 and of M2 : let (Ui , x(i) )i∈I be coordinate charts covering M1 and let
(V j , y( j) ) j∈J be coordinate charts covering M2 , then (Ui ×V j )(i, j)∈I×J is an open cov-
ering 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 struc-

ture; 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 pro-

jection maps (i = 1, 2). For m1 ∈ M1 and m2 ∈ M2 , we will also use the inclusion
maps ım2 : M1 → M1 × M2 and ım1 : M2 → M1 × M2 , defined by ım2 (x) := (x, m2 )
and ım1 (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 ım2 (re-
spectively T ım2 ). 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 ım2 (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 ım2 (Vm1 )(F ◦ p1 )

= Vm1 (F ◦ p1 ◦ ım2 ) = 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 , π1 ) and (M2 , π2 ) be two Poisson manifolds. The product
manifold M1 ×M2 has a unique Poisson structure {· , ·} = π , 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 , π ) is called the product Poisson manifold of M1

and M2 , while π 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 F-linear map φ : V → W between vector spaces V and W , leads to a linear
map ∧2 φ : ∧2V → ∧2W . As recalled in Appendix A (Section A.2) it is defined for
v1 , v2 ∈ V by
∧2 φ (v1 ∧ v2 ) := φ (v1 ) ∧ φ (v2 ) .
Applied to the linear map

Tm1 ım2 : Tm1 M1 → Tm (M1 × M2 )

we can define a bivector field π1 on M1 × M2 by its value at m = (m1 , m2 ):

π1m := ∧2 (Tm1 ım2 )(π1 )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

π1 [F1 ◦ p1 , G1 ◦ p1 ] = π1 [F1 , G1 ] ◦ p1 ,

π1 [F1 ◦ p1 , G2 ◦ p2 ] = 0 ,
π1 [F2 ◦ p2 , G2 ◦ p2 ] = 0 .

It follows that the bivector field π := π1 + π2 , where π2 is constructed in the same
way as π1 , but using Tm2 ım1 instead of Tm1 ım2 , has the following properties:

π [F1 ◦ p1 , G1 ◦ p1 ] = π1 [F1 , G1 ] ◦ p1 ,
π [F1 ◦ p1 , G2 ◦ p2 ] = 0 , (2.10)
π [F2 ◦ p2 , G2 ◦ p2 ] = π2 [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 π satisfies the Jacobi
identity. In view of (iv) in Proposition 1.16, this implies that π 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 π 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 formu-
las for it, in local coordinates, which provides uniqueness of such a Poisson struc-
ture. 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 coordi-
nates on U ×V as x1 , . . . , xd1 , y1 , . . . , yd2 . According to (1.26) and (2.10), π is given
in terms of these coordinates by

∂ ∂ ∂ ∂
π= ∑ π [xi , x j ] ∧ + ∑ π [yi , y j ]
∂ xi ∂ x j 1i< ∂ y i
∧
∂ yj
1i< jd1 jd2
∂ ∂ ∂ ∂
= ∑ π1 [xi , x j ] ◦ p1 ∧ + ∑ π2 [yi , y j ] ◦ p2
∂ xi ∂ x j 1i< jd2
∧
∂ yi ∂ y j
,
1i< jd1

which is written, in the bracket notation, as

  ∂ ∂   ∂ ∂
{· , ·} = ∑ xi , x j 1
∧ + ∑
∂ xi ∂ x j 1i< jd2
yi , y j 2 ∧
∂ yi ∂ y j
,
1i< jd1
   
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 , π1 ) and (M2 , π2 ) be Poisson manifolds and let π denote
the product Poisson structure on M1 × M2 . Let F, G be functions on a non-empty
open subset U of M1 × M2 and let (m1 , m2 ) ∈ U. By definition of π , we have that

π(m1 ,m2 ) = π1 (m1 ,m2 ) + π2 (m1 ,m2 )

(2.11)
= ∧2 (Tm1 ım2 ) (π1 )m1 + ∧2 Tm2 ım1 (π2 )m2 ,

so that

π(m1 ,m2 ) (F, G) = (π1 )m1 (F ◦ ım2 , G ◦ ım2 ) + (π2 )m2 (F ◦ ım1 , G ◦ ım1 ) ,

which yields, in the bracket notation, the following formula for the product bracket
46 2 Poisson Structures: Basic Constructions
 
{F, G} (m1 , m2 ) = {F ◦ ım2 , G ◦ ım2 }1 (m1 ) + F ◦ ım1 , G ◦ ım1 2 (m2 ) , (2.12)

valid for all F, G ∈ F (U), where U is an open subset of M1 × M2 , which con-

tains (m1 , m2 ).

2.2 Poisson Ideals and Poisson Submanifolds

The simplest sub-objects of Poisson manifolds are Poisson submanifolds, which are
submanifolds, equipped with a Poisson structure, with respect to which the inclu-
sion 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 gen-
eral 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) There exists a derivation V0 of A /I such that V [F] = V0 [F], for every
F ∈A;
(ii) 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 ele-
ments F ∈ A /I by V0 [F] := V [F], which is well-defined 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 well-defined,
  
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) There exists a Poisson bracket {· , ·}0 on A /I such that p : A → A /I is
a Poisson morphism;
(ii) I is a Poisson ideal of (A , ·, {· , ·});
(iii) For every F ∈ A , the Hamiltonian derivation XF preserves I ;
(iv) 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 κ : 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 sec-
tion, 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 κ (V [F]) = V0 [κ (F)], for every
F ∈ F (M), which can be expressed as the commutativity of the following
diagram:
V
F (M) F (M)

κ κ

F (N) F (N)
V0

(ii) 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 ı : 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 κ : 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 subva-
riety of M, defined by a prime ideal IN ⊂ F (M). The following are equivalent:
(i) N is a Poisson subvariety of (M, {· , ·});
(ii) IN is a Poisson ideal of F (M);
(iii) For every F ∈ F (M), the Hamiltonian vector field XF is tangent to N;
(iv) 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 Sec-
tion 1.3.4).
First, let us recall the basic vocabulary. Suppose that ı : N0 → M is a map, where
both M and N0 are manifolds. If for every point n0 ∈ N0 the tangent map Tn0 ı :
Tn0 N0 → Tı(n0 ) M is injective, then ı is said to be an immersion. If the immersion is
injective one says that (N0 , ı) is an immersed submanifold, or submanifold for short;
one often refers to the subset N := ı(N0 ) of M as being an immersed submanifold,
keeping in mind that N comes with its own differential structure and topology. Since
Tn0 ı 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 ı(n0 ), with ı(V ) ⊂ U, and such that ı 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 ı(n0 ).
In the special case in which the topology on N which comes from N0 coincides
with the induced topology from M, i.e., ı 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 = ı(N0 ) is a submanifold of M, then for n := ı(n0 ) ∈ N, the subspace
Tn0 ı(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, π ) be a Poisson manifold. An (immersed or embedded)

submanifold (N0 , ı) of M is called a Poisson submanifold if there exists on N0 a
Poisson structure πN such that the inclusion map ı is a Poisson map. Viewing N :=
ı(N0 ) as a submanifold of M, we refer to πN 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, π ) be a Poisson manifold. For an (immersed or embed-
ded) submanifold N = ı(N0 ) of M, the following conditions are equivalent:
50 2 Poisson Structures: Basic Constructions

(i) N is a Poisson submanifold of (M, π );

(ii) For every open subset U ⊂ M and for every F ∈ F (U), the Hamiltonian
vector field XF is tangent to N;
(iii) For every n ∈ N, the intersection of N with the symplectic leaf Sn (M) con-
tains a neighborhood of n in Sn (M);
(iv) For every n ∈ N, the bivector πn 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 := ı(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
∂ d
∂
XF = ∑ {xi , F} + ∑ {x j , F} ,
i=1 ∂ xi j=s+1 ∂xj

where s is the dimension of N0 and {· , ·} := π . Let us

 denote
 the Poisson
 structure
 on
N by {· , ·}0 = π0 . Since ı is a Poisson morphism, x j , F (n) = x j ◦ ı, F ◦ ı 0 (n0 ),
which is zero when s + 1  j  d, since x j ◦ ı = 0, for such j. It follows that, for
n ∈ N ∩U,
 
s
∂ s
∂
(XF )n = ∑ {xi , F} (n) = ∑ {xi , F} (n) Tn0 ı ,
i=1 ∂ xi n i=1 ∂ xi n0

so that (XF )ı(n0 ) ∈ Tn0 ı 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 con-
tains n (see Section 1.3.4). Since ı is an immersion, there exists an open subset V of
N0 which contains n0 , where n = ı(n0 ), and an open subset U of M, such that ı(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) ⊂ ı(V ). To show this, it suffices to show that
every Hamiltonian path in U which starts in ı(V ), stays in ı(V ). Consider a Hamil-
tonian path γ : I → U, where I is an open neighborhood of 0 in F (which is R or C),
with γ (0) ∈ ı(V ). It is the integral curve of a Hamiltonian vector field XF , with
F ∈ F (U ), which is tangent to ı(V ), at all points of ı(V ), where U is an open
subset of U, containing γ (I). Suppose that γ (I) is not entirely contained in ı(V ).
Consider
I := {t ∈ I | γ (t) ∈ ı(V )} ,
which is an open subset of I, since γ is an integral curve of a vector field which is
tangent to ı(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 := γ (t ). Then
2.2 Poisson Ideals and Poisson Submanifolds 51

n ∈ ı(V ), since ı(V ) is a closed subset of U. Since γ is an integral curve, passing

through n , of a vector field which is tangent to ı(V ), we have that γ (t ) ⊂ ı(V ) for t
in a neighborhood of t , which contradicts the fact that t belongs to the boundary
of I0 . Therefore, all Hamiltonian paths γ : I → U, defined on a neighborhood I of 0,
with γ (0) ∈ ı(V ), stay in ı(V ) and ı(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 := ı(n0 ). Since ı is an
immersion,
Tn0 ı : 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 (πN )n0 ∈ ∧2 Tn0 N0 such that

∧2 (Tn0 ı)(πN )n0 = πn , (2.14)

as the latter belongs to ∧2 Tn Sn (M). We claim that the map

πN : N0 → ∧2 T N0
(2.15)
n0 → (πN )n0

is a Poisson structure on N0 , such that ı is a Poisson map. To see that πN 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 := ı(n0 ), and functions F, G ∈ F (U), such that F ◦ ı = F0 and
G ◦ ı = G0 on some neighborhood V of n0 in N0 . For every n = ı(n0 ) ∈ U ∩ ı(V ) we
have that

πN [F0 , G0 ](n0 ) = πN [F ◦ ı, G ◦ ı](n0 )

 
= (πN )n dn (F ◦ ı), dn (G ◦ ı)
0 0 0

= πn (dn F, dn G) = π [F, G](ı(n0 )) ,

where we have used the definition (2.14) of πN to obtain the last line. Since n0 →
π [F, G](ı(n0 )) is clearly a smooth/holomorphic function on V ⊂ N0 , it follows that
πN 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

πN [πN [F0 , G0 ] , H0 ] (n0 ) = π [π [F, G] , H] (ı(n0 )) ,

for all n0 in a neighborhood V of n0 in N0 . It leads to the Jacobi identity for πN , so

that πN is a Poisson structure on N0 .
According to Proposition 1.19, ı is a Poisson map; it is clear that πN is the unique
Poisson structure on N0 , such that ı 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 π as
52 2 Poisson Structures: Basic Constructions

∂ ∂ d d
∂ ∂
π= ∑ αi j ∧ + ∑ ∑ αi j ∧
∂ xi ∂ x j i=1 j=s+1 ∂ xi ∂ x j
,
1i< js

where all the coefficients αi j are functions, defined on U, with α ji = −αi j . If πn ∈

∧2 Tn N, then αi 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 mani-

fold, 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) H (M) denotes the C-algebra of all holomorphic functions on M,
(b) C∞ (MR , C) denotes the C-algebra of all C-valued R-differentiable functions
on MR ,
(c) C∞ (MR ) denotes the R-algebra of all R-valued R-differentiable functions on
MR ,
(d) H (M) denotes the C-algebra of all anti-holomorphic 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
√
H (U) → C∞ (UR , C)  C∞ (UR ) + −1C∞ (UR ) , (2.16)

where the latter isomorphism consists of writing a C-valued 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 =
∞
√where xk , yk ∈ C (UR ) are the real smooth functions obtained
(x1 , . . . , xd , y1 , . . . , yd ),
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 53

d
∂ 1 d ∂ √ ∂
V = ∑ V [zk ] ∂ zk
= ∑ V [zk ]
2 k=1 ∂ xk
− −1
∂ yk
. (2.17)
k=1

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 ℜ(V )
and ℑ(V ), by putting:

1 d ∂ ∂
ℜ(V ) := ∑ ℜ(V [zk ]) ∂ xk + ℑ(V [zk ]) ∂ yk
2 k=1
, (2.18)

1 d ∂ ∂
ℑ(V ) := ∑ ℑ(V [zk ]) ∂ xk − ℜ(V [zk ]) ∂ yk
2 k=1
. (2.19)

the R-algebra C∞ (UR ) extend uniquely to the C-algebra

These two derivations of √
C (UR , C)  C (UR ) + −1C∞ (UR ), by C-linearity; we will denote these ex-
∞ ∞

still by ℜ(V ) and ℑ(V ). Writing a holomorphic function F on U as

tensions √
F = G + −1 H and using the Cauchy–Riemann equations

∂G ∂H ∂G ∂H
= , =− ,
∂ xk ∂ yk ∂ yk ∂ xk

one verifies by a direct computation that, for all F ∈ H (U),

1 1
ℜ(V )[F] = V [F], ℜ(V )[F] = V [F] . (2.20)
2 2
These equations characterize ℜ(V ) since (2.18) can be derived from it. Similarly,
the imaginary part ℑ(V ) is characterized by the fact that
√ √
−1 −1
ℑ(V )[F] = − V [F], ℑ(V )[F] = V [F] ,
2 2
for all F ∈ H (U). These characterizations imply that ℜ(V ) and ℑ(V ) are inde-
pendent 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
∂ ∂
P= ∑ P[zk , z ] ∧
∂ zk ∂ z
1k<d
 
1 ∂ √ ∂ ∂ √ ∂
= ∑ P[zk , z ]
4 1k<d ∂ xk
− −1
∂ yk
∧
∂ x
− −1
∂ y
.

As above, decomposing P[zk , z ] into its real and imaginary parts, leads to the defi-
nition of two skew-symmetric biderivations ℜ(P) and ℑ(P) of C∞ (UR ), by putting,
54 2 Poisson Structures: Basic Constructions

1 ∂ ∂ ∂ ∂
ℜ(P) := ∑
4 1k<d
ℜ(P[zk , z ]) ∧ − ∧
∂ xk ∂ x ∂ yk ∂ y

1 ∂ ∂ ∂ ∂
+ ∑ ℑ(P[zk , z ]) ∂ xk ∧ ∂ y + ∂ yk ∧ ∂ x
4 1k<d
,

and similarly for ℑ(P). By definition, ℜ(P) and ℑ(P) are skew-symmetric bideriva-
tions of C∞ (UR ) and they uniquely extend to skew-symmetric biderivations of
C∞ (UR , C), by C-linearity. Then, ℜ(P) is, as in (2.20), characterized by the fol-
lowing properties: for all F, G ∈ H (U),

1
ℜ(P)[F, G] = P[F, G] ,
2
ℜ(P)[F, G] = 0 , (2.21)
1
ℜ(P)[F, G] = P[F, G] .
2
There is an analogous characterization for the imaginary part ℑ(P). Because of
these characterizations, we can conclude, as in the case of vector fields, that ℜ(P)
and ℑ(P) are well-defined 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 ℜ(P) is also Poisson structure. One shows similarly that ℑ(P) is a Poisson
structure.
Proposition 2.13. If (M, π ) is a complex Poisson manifold of dimension d, then the
bivector field ℜ(π ) is a Poisson structure on MR . In local coordinates, ℜ(π ) is given
by:

1 ∂ ∂ ∂ ∂
ℜ(π ) = ∑ ℜ(πk ) ∂ xk ∧ ∂ x − ∂ yk ∧ ∂ y
4 1k<d

1 ∂ ∂ ∂ ∂
+ ∑
4 1k<d
ℑ( π k ) ∧ + ∧
∂ xk ∂ y ∂ yk ∂ x
√
where zk = xk + −1yk and πk := π (zk , z ), for 1  k,   d.

Proof. According to the characterization of ℜ(π ), given above, and since π is a

holomorphic Poisson bracket, we have, for every coordinate chart (U, z) and for all
F, G, H ∈ H (U),

1
ℜ(π )[ℜ(π )[F, G], H] = {{F, G} , H} ,
4
ℜ(π )[ℜ(π )[F, G], H] = 0 ,
ℜ(π )[ℜ(π )[F, G], H] = 0 ,
1
ℜ(π )[ℜ(π )[F, G], H] = {{F, G} , H} ,
4
2.3 Real and Holomorphic Poisson Structures 55

where we have written {· , ·} for π . Since π satisfies the Jacobi identity, these equa-
tions imply that ℜ(π ) satisfies the Jacobi identity, when applied to arbitrary triples
of holomorphic or anti-holomorphic functions on U. In particular, for every coordi-
nate chart (UR , Z), the Jacobi identity will be satisfied by ℜ(π ) for the coordinate
functions xk and y , because they can be written as C-linear combinations of the
holomorphic and anti-holomorphic functions zk and zk . Then, according to Proposi-
tion 1.16, ℜ(π ) 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 s-dimensional 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)
∂xj 1it
1 jd

equals d − s is a (Zariski) open subset of N, called the smooth part of N, and de-
noted 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 com-
plex 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 skew-
symmetric 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

F̃ and G̃, defined on a neighborhood U of n in M. For n in a neighborhood of V in
N sm , contained in W ∩U, we define,

∂ F̃ ∂ G̃
{F, G}H (n ) := ∑ xi j (n )
∂ xi
(n )
∂xj
(n ).
1i< jd
   
It defines the unique bivector field on V such that xi |N , x j |N = xi |N , x j |N on V ,
H
for all 1  i, j  d, hence it is a well-defined holomorphic
 bivector field on N sm .
Since {· , ·} satisfies the Jacobi identity for all triples xi |N , x j |N , xk|N , 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 F-algebra and if F̂ is a field extension of F (think

of C, which is a field extension of R), then the tensor product F̂ ⊗F A carries a
natural structure of a commutative associative F̂-algebra, simply by putting, for all
a, b ∈ F̂ and for all F, G ∈ A ,

a(b ⊗ F) := (ab) ⊗ F ,
(2.22)
(a ⊗ F)(b ⊗ G) := (ab) ⊗ (FG) .

This very general algebraic procedure is known as “extension of scalars”. In fact,

an F-Lie algebra structure [· , ·] on A leads, in a similar way, to an F̂-Lie algebra
structure on F̂ ⊗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 F-Poisson algebra (A , ·, {· , ·}), leads
in a natural way to an F̂-Poisson algebra (F̂ ⊗F A , ·, {· , ·}). For example, it allows
2.4 Other Constructions 57

one to pass from a Poisson bracket on an F-algebra 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 commu-

tative associative algebra A with respect to a multiplicative system S in A (multi-
plicative 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 el-
ements 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 well-defined: 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 ]

− = . (2.26)
H1 H2 H1 H2
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 well-defined. 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
VS = VS + VS , (2.27)
HH H H H H

which says that VS is a derivation of A /S.

In the same way, every skew-symmetric biderivation P of A extends uniquely
to a skew-symmetric 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
   
1 1 P[H, G]
PS , G = −PS G, =− ,
H H H2
  (2.28)
1 1 P[H, K]
PS , = .
H K H 2K2

As above, it is easily checked that PS is a skew-symmetric 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 mor-
phism of Poisson algebras.
Proof. If a Poisson structure {· , ·}S on A/S, for which A → A /S is a Poisson mor-
phism, 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 gen-
erated 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 π on M will induce a Poisson structure πSm 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 , πSm ) as Poisson
subalgebras.
The second special case of localization consists in taking a non-constant func-
tion 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 in-
determinate ξ we have an algebra isomorphism

F (M)[ξ ]
ρ: → F (M)/SG
Gξ − 1
n n
∑ Fi ξ i → ∑ Fi /Gi .
i=0 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 F|U = G|U .
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 π [F, G] to U depends on the restrictions of F and G to U only, hence
the germ (π [F, G])m of π [F, G] at m is well-defined. Thus, a Poisson structure π ,
defined on a neighborhood of m ∈ M leads to a Poisson bracket πm 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 π ∼m π if
60 2 Poisson Structures: Basic Constructions

πm = πm . It amounts to saying that if F and G are functions which are defined in

a neighborhood of m in M, then π [F, G] and π [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 π at m by πm . In bracket notation,
the germ πm is given by
 
d
∂F ∂G
{Fm , Gm }m = ∑ (xi j )m
∂ xi ∂xj
, (2.29)
i, j=1 m m

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 skew-symmetric 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 πm . Although we have used the same notation πm 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 ρ : A → B ⊗ A be a Poisson
morphism, where the right-hand side is equipped with the product bracket. Prove
that
A B := {F ∈ A | ρ (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 , π1 ) and (M2 , π2 ) be two Poisson manifolds. Let Ψ : M1 → M2 be a
Poisson map. Show that if Ψ (M1 ) is a submanifold of M2 , then it is a Poisson sub-
manifold.
6. Let (M, π ) be a complex Poisson manifold. Show that every linear combination
of the Poisson structures ℜ(π ) and ℑ(π ), defined in Section 2.3.1, is a Poisson
structure on MR .
7. Let π 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 π has rank at most 2 at every point of U,
then F π 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 sym-
plectic manifold (see Section 6.3). Let (M, ω ) be a symplectic manifold, and denote
by π the canonical Poisson structure, associated to ω . Prove that the open subsets
of M are the only Poisson submanifolds of (M, π ).
9. The purpose of this exercise is to establish the existence of non-trivial 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. 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 ;
b. 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 ;
c. 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 ;
d. Using a coordinate chart, conclude that every manifold of dimension d  2,
admits a non-trivial Poisson structure; it can be chosen such that its rank, at a
given point, equals a given even integer 2s  d;
e. The following is an open problem: show that for every Poisson structure π on
Rd there exists a Poisson structure on Rd which coincides with π in a neigh-
borhood 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 Lichnerowicz’s 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, ex-
plained in Section 2.3.1, see Laurent–Stiénon–Xu [121].
Chapter 3
Multi-Derivations and Kähler 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 skew-symmetric multi-
derivations 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 Kähler forms Ω • (A )
(or on Ω • (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) Ω • (A ), Ω • (M)
Algebraic multi-derivation Kähler form
Geometric multivector field differential form
Wedge product ∧ : X• × X• → X• ∧ : Ω• ×Ω• → Ω•
Differential – d : Ω • → Ω •+1
Schouten bracket [· , ·]S : X• × X• → X•−1 –
Lie derivative LP : X• → X•+p−1 LP : Ω • → Ω •−p+1
Internal product – ıP : Ω • → Ω •−p

Since we are using mostly the algebraic properties of these operations, our con-
structions will be mainly algebraic, but they will always be reformulated in geo-
metrical terms. Multi-derivations are introduced in Section 3.1 and Kähler 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. Laurent-Gengoux et al., Poisson Structures, 63

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 3, © Springer-Verlag Berlin Heidelberg 2013
64 3 Multi-Derivations and Kähler Forms

3.1 Multi-Derivations and Multivector Fields

In this section, we generalize the notion of a derivation and of a skew-symmetric

biderivation, defined in Section 1.1.1, to the notion of a skew-symmetric multi-
derivation. We assemble all multi-derivations in one graded vector space, which
becomes a graded algebra, upon introducing the wedge product. Multivector fields,
which are the geometrical analogs of skew-symmetric multi-derivations, will be dis-
cussed in Section 3.1.3.

3.1.1 Multi-Derivations

Let A be a commutative associative F-algebra and let p  1. A skew-symmetric,

p-linear map P ∈ HomF (∧ p A , A ) is called a skew-symmetric p-derivation of A
(with values in A ), if P is a derivation in each of its arguments. By skew-symmetry,
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, intro-

duced in Chapter 1, to enclose the list of arguments of a skew-symmetric multi-
derivation 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 skew-symmetric p-derivations of A are

equal as soon as they are equal for all p-tuples, taken from a system of generators
of A ; in particular, a p-derivation of A vanishes as soon as it vanishes on all p-
tuples, 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 skew-symmetric p-derivations of A is denoted by X p (A ) and we introduce the
graded vector space
 
X• (A ) := X p (A ) ⊂ HomF (∧ p A , A ) ,
p∈N p∈N

whose elements are called skew-symmetric multi-derivations. 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 Multi-Derivations and Multivector Fields 65

We often think of A as being the algebra of regular functions on an affine vari-

ety 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 Multi-Derivations

There are two natural composition laws on the graded A -module X• (A ) of skew-
symmetric multi-derivations of a commutative associative F-algebra 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 permuta-
tion σ of the set {1, . . . , p + q}, such that σ (1) < · · · < σ (p) and σ (p + 1) < · · · <
σ (p + q). The set of all (p, q)-shuffles is denoted by S p,q . For a shuffle σ ∈ S p,q ,
we denote the signature of σ (as a permutation) by sgn(σ ). It is also convenient to
define S p,−1 := 0/ and S−1,q := 0, / for p, q ∈ N.
For P ∈ X (A ) and Q ∈ X (A ), their wedge product P ∧ Q ∈ X p+q (A ) is the
p q

skew-symmetric (p + q)-derivation of A , defined by

(P ∧ Q)[F1 , . . . , Fp+q ] :=
(3.3)
∑ sgn(σ ) P[Fσ (1) , . . . , Fσ (p) ] Q[Fσ (p+1) , . . . , Fσ (p+q) ] ,
σ ∈S p,q

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 skew-symmetrization of the
tensor product (of skew-symmetric multi-linear 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 com-

mutative: 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 skew-symmetric

biderivation, given by

(V ∧ W )[F, G] = V [F] W [G] − V [G] W [F] ,

66 3 Multi-Derivations and Kähler 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

ıF : X• (A ) → X•−1 (A ) ,

of degree −1, by taking F as the first element on which the multi-derivation is

evaluated: for P ∈ X p (A ), the skew-symmetric (p − 1)-derivation ıF P is defined by

ıF P[F2 , . . . , Fp ] := P[F, F2 . . . , Fp ] , (3.5)

for all F2 , . . . , Fp ∈ A . For a derivation V ∈ X1 (A ), this yields ıF V = V [F]; for

a function G (element of X0 (A )), the above definition should be understood as
ıF G := 0, since X p (A ) = {0} for p < 0. For every F ∈ A , the graded linear map
ıF is a graded derivation1 of degree −1 of X• (A ), i.e.,

ıF (P ∧ Q) = ıF P ∧ Q + (−1) p P ∧ ıF Q . (3.6)

This follows immediately from (3.3), upon using that for every shuffle σ ∈ S p,q , one
has that either σ (1) = 1 or σ (p + 1) = 1. Also, each derivation V ∈ X1 (A ) defines
a graded F-linear 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 deriva-
tive 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 skew-symmetric
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 Multi-Derivations 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 skew-symmetric biderivation, is a vector field,
respectively a bivector field. Similarly, multivector fields are the geometrical analogs
of skew-symmetric multi-derivations. 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 skew-
symmetric p-linear map  
Fm (M) p
Ψm : →F
∼
a skew-symmetric pointwise p-derivation at m, if for all functions F, G, H, . . . , K,
defined in a neighborhood of m in M,

Ψm (Fm Gm , Hm , . . . , Km )
= F(m) Ψm (Gm , Hm , . . . , Km ) + G(m) Ψm (Fm , Hm , . . . , Km ) .

The vector space of all pointwise, skew-symmetric p-derivations 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 p-derivations,
     
∂ ∂ ∂
∧ ∧···∧ ,
∂ 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
p-vector 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 p-vector field P
becomes a skew-symmetric p-derivation of the algebra of functions F (U) and P
admits the local coordinate expression

∂ ∂
P= ∑ P[xi1 , . . . , xi p ]
∂ x i1
∧···∧
∂ xi p
, (3.9)
1i1 <···<i p d

where d := dim M. It follows that a p-vector 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 Multi-Derivations and Kähler 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 p-vector fields on M by X p (M) and we de-
fine the graded F (M)-module X• (M) := ⊕ p∈N X p (M), which is the geometrical
analog of X• (A ). The operations which we have introduced in Section 3.1.2 for
skew-symmetric p-derivations, namely the wedge product ∧, the contraction ıF by
a function F and the Lie derivative LV , are introduced in the same way for p-vector
fields, upon replacing the functions on which they act by local functions. Clearly,
all properties of these operations can be repeated word-for-word, as they are purely
algebraic. However, in a geometrical approach, the formula (3.7) for the Lie deriva-
tive of a p-vector 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
d  
[V , W ]m := (LV W )m := TΦ−t (m) Φt WΦ−t (m) , (3.10)
dt |t=0

where Φt is the local flow of V . Notice that the tangent map TΦ−t (m) Φt of the map
Φt at Φ−t (m) sends tangent vectors at Φ−t (m) (here, the vector WΦ−t (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 p-vector 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 p-vector field,
whose value at m ∈ M is given by
d  
(LV P)m := ∧ p TΦ−t (m) Φt PΦ−t (m) . (3.11)
dt |t=0

One immediate consequence of this definition is that LV P = 0 if and only if P is pre-

served 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 Kähler Forms and Differential Forms

In this section, we introduce the objects which are, in a sense, dual to skew-
symmetric multi-derivations of a (commutative associative) algebra A : the Kähler
forms of A . In the case of a manifold M, the dual objects to multivector fields are
differential forms. The modules of Kähler forms and of differential forms will be
described in this section, together with their algebraic structure.
3.2 Kähler Forms and Differential Forms 69

3.2.1 Kähler Differentials

Let A be a commutative associative F-algebra. The module of Kähler differentials

of A (over F) is the A -module Ω 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 → Ω 1 (A ), defined by F → dF. The pair (Ω 1 (A ), d) satisfies the following
universal property: for every derivation V : A → A , there exists a unique A -linear
map Vˆ : Ω 1 (A ) → A , such that Vˆ ◦ d = V . In other words, there exists a unique
arrow Vˆ which makes the following diagram commutative:

d
A Ω 1 (A )

V Vˆ
A

In order to prove the universal property, notice that the commutativity of the diagram
amounts to saying that Vˆ (dF) = V [F], for all F ∈ A . It is then clear that Vˆ should
be defined as the (unique) A -linear map, such that

Vˆ (G dF) = G V [F] , (3.12)

for all F, G ∈ A . It is indeed easily verified that Vˆ , given by (3.12), is well-defined;

for example, if F, G ∈ A , then
 
Vˆ d(FG) − FdG − GdF = V [FG] − FV [G] − GV [F] = 0 ,

since V is a derivation of A . Given an arbitrary A -linear map Vˆ : Ω 1 (A ) → A , it

is also clear that Vˆ ◦ 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 (Ω 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 Multi-Derivations and Kähler Forms

3.2.2 Kähler Forms

In order to generalize (3.13) to skew-symmetric p-derivations, we introduce, for

p ∈ N, the A -module Ω p (A ) := ∧ p Ω 1 (A ), where ∧ is the wedge product over A ;
for p = 0, this should be read as Ω 0 (A ) := A . Together, the A -modules Ω p (A )
form the graded A -module

Ω • (A ) := Ω p (A ) .
p∈N

The elements of Ω p (A ), with p > 0, are called Kähler p-forms, or simply Kähler
forms. As a vector space, respectively as an A -module, Ω 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 F-linear map d : A → Ω 1 (A )
extends by functoriality of ∧ to an F-linear map ∧• d : ∧• A → Ω • (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 (Ω • (A ), ∧). For every skew-symmetric p-derivation P of A , there
exists a unique A -linear map P̂ : Ω p (A ) → A , such that the following triangle is
commutative:
∧pd
∧pA ∧ p Ω 1 (A ) = Ω p (A )

P P̂
A

The definition of P̂ for arbitrary p is the natural generalization of the defini-

tion (3.12) of Vˆ , which corresponds to the case of p = 1. It is given by

P̂(G dF1 ∧ · · · ∧ dFp ) := G P[F1 , . . . , Fp ] , (3.15)

for all G, F1 , . . . , Fp ∈ A . It leads to a natural isomorphism (of A -modules):

X p (A ) HomA (Ω p (A ), A ) . (3.16)

For some of the computations which we will do later, it is useful to view P̂ as an

A -linear map Ω • (A ) → A , so we extend P̂, defined in (3.15), trivially to Kähler
q-forms, with q = p:

P̂(G dF1 ∧ · · · ∧ dFq ) := 0, if q = p , (3.17)

for all G, F1 , . . . , Fq ∈ A . With this notation, the map P → P̂ yields a natural iso-
morphism
3.2 Kähler Forms and Differential Forms 71

X• (A ) → HomA (Ω p (A ), A ) .
p∈N

Remark 3.3. For p ∈ N, the natural pairing

Ψ : X p (A ) ⊗A Ω p (A ) → A
(3.18)
P⊗ω → ω , P := P̂(ω )

induces two A -linear maps

Ψ1 : X p (A ) → HomA (Ω p (A ), A ) ,
Ψ2 : Ω p (A ) → HomA (X p (A ), A ) .

As we have seen in (3.16), Ψ1 is an isomorphism. In general, Ψ2 is neither injective

nor surjective; however, if A is the algebra C∞ (M) of smooth functions on a real
manifold M, the map Ψ2 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 Ω • (A ), which are closely
related to the corresponding structures on ∧• A , where A is as before a commu-
tative associative F-algebra. 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 F-algebra, and that (∧• A , Δ ) is a coassociative F-
coalgebra. Since Ω • (A ) is an exterior algebra (over A ), we have, as we mentioned
in the previous section, a wedge product on Ω • (A ), also denoted by ∧, which
makes (Ω • (A ), ∧) into an associative, graded commutative A -algebra. Also, it
leads to an A -linear coassociative coproduct

Δ : Ω • (A ) → Ω • (A ) ⊗A Ω • (A ) ,

which is defined in the same way as in the case of ∧• A (see (A.17)). Namely, Δ p,q
is given, for p, q ∈ N, and F1 , . . . , Fp+q ∈ A by

Δ p,q (dF1 ∧ · · · ∧ dFp+q ) (3.19)

:= ∑ sgn(σ )(dFσ (1) ∧ · · · ∧ dFσ (p) ) ⊗ (dFσ (p+1) ∧ · · · ∧ dFσ (p+q) ) .
σ ∈S p,q

Recall that coassociativity means that

(1Ω • (A ) ⊗ Δ ) ◦ Δ = (Δ ⊗ 1Ω • (A ) ) ◦ Δ . (3.20)

It is clear, from its definition (3.14), that ∧• d is a homomorphism of the F-coalgebras

(∧• A , Δ ) and (Ω • (A ), Δ ).
72 3 Multi-Derivations and Kähler Forms

A useful application of the coproduct Δ of Ω • (A ), is that the wedge prod-

uct (3.3) of two multi-derivations P, Q ∈ X• (A ) can be expressed in terms of Δ ,
as follows:
P∧ Q = ν ◦ (P̂ ⊗ Q̂) ◦ Δ , (3.21)
where ν is the canonical isomorphism

ν : A ⊗A Ω • (A ) → Ω • (A )
(3.22)
F ⊗ω → Fω .

We used ν in (3.21) only for the case of ω ∈ Ω 0 (A ) = A , but we will need its
general form later on; in particular we will need the following properties of ν :

P̂ ◦ ν = ν ◦ (1A ⊗ P̂) ,
(3.23)
Δ ◦ ν = (ν ⊗ 1Ω • (A ) ) ◦ (1A ⊗ Δ ) ,

where P ∈ X• (A ). The proof of these properties follows at once from the definition
(3.22) of ν .

3.2.4 The de Rham Differential and Cohomology

The differential d : A → Ω 1 (A ) extends to a well-defined graded F-linear map

d : Ω • (A ) → Ω •+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 differ-
ential. It is a graded derivation, of degree 1, of (Ω • (A ), ∧), as

d(ω ∧ η ) = dω ∧ η + (−1) p ω ∧ dη , (3.25)

for ω ∈ Ω p (A ) and η ∈ Ω • (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

d d d
0 Ω 0 (A ) Ω 1 (A ) Ω 2 (A ) ···

is a complex, called the (algebraic) de Rham complex. A Kähler p-form ω for which
dω = 0 is called closed, while it is called exact if it is contained in the image of d. By
the above, exact Kähler forms are closed, which leads, for p ∈ N, to the cohomology
3.2 Kähler Forms and Differential Forms 73

space2
p Ker(d : Ω p (A ) → Ω p+1 (A ))
HdR (A ) := ,
Im(d : Ω p−1 (A ) → Ω p (A ))
where it is understood that HdR0 (A ) := Ker(d : A → Ω 1 (A )) = F. The F-vector
p
space HdR (A ) is called the p-th de Rham cohomology space and the graded vector
space  p
•
HdR (A ) := HdR (A )
p∈N

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 Kähler forms is closed; (2) the wedge product of an exact
Kähler form with a closed Kähler form is exact. Combining these two properties
shows that there is an induced product

• (A ) × H • (A ) → H • (A )
∧ : HdR dR dR
([ω ], [η ]) → [ω ∧ η ]

where [ω ] denotes the de Rham cohomology class of a closed Kähler form ω . It fol-
• (A ), ∧) is, just like (Ω • (A ), ∧), an associative, graded commutative
lows that (HdR
F-algebra.
To close this section, we wish to point out that Ω • leads to a covariant functor
from the category of commutative associative F-algebras to the category of graded
F-vector spaces (associative, graded commutative F-algebras, if you wish). Namely,
let φ : A → B be an algebra homomorphism, where A and B are commutative
associative algebras. For every p ∈ N, the induced linear map Ω p (φ ) : Ω p (A ) →
Ω p (B) is defined as follows:

Ω p (φ ) : Ω p (A ) → Ω p (B)
G dF1 ∧ · · · ∧ dFp → φ (G) dφ (F1 ) ∧ · · · ∧ dφ (Fp ) .

From this definition, it is easily verified that Ω • is indeed a covariant functor. Since
Ω • (φ ) and d commute, Ω • (φ ) induces a graded linear map of degree 0:
• • •
HdR (φ ) : HdR (A ) → HdR (B) .
• 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 F-linear.
74 3 Multi-Derivations and Kähler 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 p-form 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 skew-
symmetric, F (M)-p-linear 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 ωm ∈ ∧ p Tm∗ M, is called a differential
p-form if for every open subset U ⊂ M and for all vector fields V1 , . . . , V p ∈ X1 (U),
one has that ω (V1 , . . . , V p ) ∈ F (U), where ω (V1 , . . . , V p ) is the function on U,
defined by
ω (V1 , . . . , V p )(m) := ωm ((V1 )m , . . . , (V p )m ) ,
for all m ∈ U. Thus, on every coordinate chart (U, x), the differential p-form ω
becomes a skew-symmetric F (U)-p-linear map from F (U) to itself. It follows
that ω admits the local coordinate expression
 
∂ ∂
ω= ∑ ω ∂ xi , . . . , ∂ xi dxi1 ∧ · · · ∧ dxi p , (3.26)
1i <···<i p d
1 1 p

where d denotes the dimension of M. Just like a p-vector field, a differential p-form
is a tensorial object: for m ∈ M, the value of ω (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 p-forms.
The F (M)-module of differential p-forms on M is denoted by Ω p (M) and we
define the graded F (M)-module

Ω • (M) := Ω p (M) ,
p∈N

which is the geometrical analog of the A -module Ω • (A ), which was defined

in Section 3.2.2. If U is the domain of a coordinate chart on M, then Ω p (U) =
∧ p Ω 1 (U), just like in the case of Kähler differentials, although Ω p (M) = ∧ p Ω 1 (M),
in general. The fact that, locally, Ω p (M) is an exterior algebra, leads to a product,
called the wedge product of differential forms. It makes Ω • (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 : Ω 0 (M) → Ω 1 (M). It extends
to a unique graded derivation of Ω • (M) of degree 1, which means that

d(ω ∧ η ) = dω ∧ η + (−1) p ω ∧ dη , (3.27)

for ω ∈ Ω p (M) and η ∈ Ω • (M), just like in (3.25). In a coordinate chart (U, x), it
leads to the classical formula for the differential of a differential p-form, namely,
3.3 The Schouten Bracket 75

d ∑ ωi1 ,...,i p dxi1 ∧ · · · ∧ dxi p

1i1 <···<i p d
d ∂ ωi1 ,...,i p
= ∑ ∑ ∂xj
dx j ∧ dxi1 ∧ · · · ∧ dxi p ,
1i1 <···<i p d j=1

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
  Ker(d : Ω p (M) → Ω p+1 (M))
• p
HdR (M) := HdR (M) := ,
p∈N p∈N
Im(d : Ω p−1 (M) → Ω 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: dω is
given, for ω ∈ Ω p (M), by
p 
dω (V0 , V1 , . . . , V p ) = ∑ (−1)i Vi ω V0 , V1 , . . . , Vi , . . . , V p (3.28)
i=0

+ ∑ j , . . . , V p ,
(−1)i+ j ω [Vi , V j ] , V0 , . . . , Vi , . . . , 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 skew-symmetric multi-derivations and of Kähler forms on a commutative asso-
ciative algebra A . The first one is an action of X• (A ) on Ω • (A ), which gen-
eralizes the natural pairing between multi-derivations and Kähler forms, which we
encountered in Remark 3.3. The second operation is the Schouten bracket which ex-
tends the classical commutator of derivations to the case of skew-symmetric multi-
derivations. It will lead at the end of the section to the so-called Gerstenhaber al-
gebra structure on X• (A ) and to two generalizations of the Lie derivative, acting
in one case on the algebra of multi-derivations, in the other case on the algebra of
Kähler forms, where the derivative is in both cases taken with respect to a multi-
derivation. 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 al-
gebra of differential forms Ω • (M) and on the other hand a graded Lie bracket, the
Schouten bracket, on X• (M).
76 3 Multi-Derivations and Kähler Forms

3.3.1 Internal Products

The universal property of the graded A -module Ω • (A ) of Kähler differentials on a

commutative associative algebra A leads to a natural action of X• (A ) on Ω • (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,

ıP : Ω • (A ) → Ω •−p (A ) ,

which is defined for ω = dF1 ∧ · · · ∧ dFk ∈ Ω k (A ), by

ıP (dF1 ∧ · · · ∧ dFk ) := ∑ sgn(σ )P[Fσ (1) , . . . , Fσ (p) ] dFσ (p+1) ∧ · · · ∧ dFσ (k)
σ ∈S p,k−p
(3.29)
when k  p, and otherwise ıP ω := 0; for F ∈ X0 (A ) = A , (3.29) has to be under-
stood as ıF ω := F ω . The reader may wish to write the definition of ıP differently,
using  
P[Fσ (1) , . . . , Fσ (p) ] = P̂ dFσ (1) ∧ · · · ∧ dFσ (p) .
With this notation, it is obvious that the restriction of ıP to Ω p (A ) is just P̂, which
was defined in (3.15); the restriction to the spaces Ω q (A ), with q = p, is however
different, since P̂ acts trivially on them (see (3.17)). It also follows that ıP can be
written, for every P ∈ X• (A ), in terms of the coproduct Δ , defined in (3.19), as

ıP = ν ◦ (P̂ ⊗ 1Ω • (A ) ) ◦ Δ . (3.30)

As a consequence, ıP is the adjoint of taking the wedge product with P: for P ∈

X p (A ) and Q ∈ Xq (A ) and ω ∈ Ω p+q (A ), we have in terms of the notations
in (3.18),
ıP ω , Q = ω , P ∧ Q . (3.31)
Indeed, (3.30), (3.23) and (3.21) imply that, on Ω p+q (A ),

Q̂ ◦ ıP = Q̂ ◦ ν ◦ (P̂ ⊗ 1Ω • (A ) ) ◦ Δ
= ν ◦ (1A ⊗ Q̂) ◦ (P̂ ⊗ 1Ω • (A ) ) ◦ Δ
= ν ◦ (P̂ ⊗ Q̂) ◦ Δ
= P∧Q ,

which yields (3.31), when applied to ω ∈ Ω p+q (A ). Equation (3.31) explains the
notation ıP : as in the case of ıF (F ∈ A ), defined in (3.5), ıP provides ω with the
p-derivation P as its first p arguments. Yet, when considering ıF G for F, G ∈ A
one should decide from the context whether G is interpreted as a 0-derivation or as
a 0-form, since in the first case ıF G = 0 (as defined in (3.5)), while in the second
case ıF G = FG (as defined in (3.29)).
In the following proposition, we establish the main properties of ı.
3.3 The Schouten Bracket 77

Proposition 3.4. Let A be a commutative associative algebra. The following iden-

tities hold, for P ∈ X p (A ) and Q ∈ Xq (A ):
(1) ıP ◦ ıQ = ıQ∧P ;
(2) [ıP , ıQ ] = ıP ◦ ıQ − (−1) pq ıQ ◦ ıP = 0;
(3) For every F ∈ A and ω ∈ Ω • (A ):

ıP (dF ∧ ω ) = ı(ıF P) ω + (−1) p dF ∧ ıP ω . (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 )
Ω 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 ω ∈ Ω k (A ) and for homogeneous elements P, Q and
R of X• (A ), whose degrees sum up to k,

ıQ∧P ω , R = ω , (Q ∧ P) ∧ R = ω , Q ∧ (P ∧ R)
(3.33)
= ıQ ω , P ∧ R = ıP (ıQ ω ), R .

Under the above assumption, the fact that (3.33) is valid for all R, implies that
ıQ∧P ω = ıP (ıQ ω ), for all ω ∈ Ω • (A ), which leads to (1). Our second proof is
based on the fact that Δ is coassociative. To simplify the notation, we will write 1
for 1Ω • (A ) ; using (3.30), the properties (3.23) of ν , the coassociativity (3.20) of Δ
and (3.21) (in that order), we find

ıP ◦ ıQ = ν ◦ (P̂ ⊗ 1) ◦ Δ ◦ ν ◦ (Q̂ ⊗ 1) ◦ Δ
= ν ◦ (P̂ ⊗ 1) ◦ (ν ⊗ 1) ◦ (1A ⊗ Δ ) ◦ (Q̂ ⊗ 1) ◦ Δ
= ν ◦ (ν ⊗ 1) ◦ (1A ⊗ P̂ ⊗ 1) ◦ (Q̂ ⊗ 1 ⊗ 1) ◦ (1 ⊗ Δ ) ◦ Δ
= ν ◦ (ν ⊗ 1) ◦ (Q̂ ⊗ P̂ ⊗ 1) ◦ (Δ ⊗ 1) ◦ Δ
= ν ◦ ((ν ◦ (Q̂ ⊗ P̂) ◦ Δ ) ⊗ 1) ◦ Δ

= ν◦ Q  ∧P⊗1 ◦Δ
= ıQ∧P .

This terminates the second proof. For the third proof, we show by direct computation
that
(ıP ◦ ıQ ) (dF1 ∧ · · · ∧ dFk ) = ıQ∧P (dF1 ∧ · · · ∧ dFk ) (3.34)
for all F1 , . . . , Fk . Since ıP is A -linear for all P ∈ X p (A ), this will provide a third
proof of (1). Since ıP ω = 0 for ω ∈ Ω q (A ) when q < p, where P ∈ X p (A ), it
suffices to prove (3.34) when k  p + q. On the one hand,

ıP ◦ ıQ (dF1 ∧ · · · ∧ dFk )
⎛ ⎞

= ıP ⎝ ∑ sgn(σ )Q[Fσ (1) , . . . , Fσ (q) ]dFσ (q+1) ∧ · · · ∧ dFσ (k) ⎠

σ ∈Sq,k−q
78 3 Multi-Derivations and Kähler Forms

= ∑ sgn(τ )Q[Fτ (1) , . . . , Fτ (q) ]P[Fτ (q+1) , . . . , Fτ (q+p) ]

τ ∈Sq,p,k−q−p
dFτ (q+p+1) ∧ · · · ∧ dFτ (k) ,

where Sq,p,k−p−q is the set of all permutations σ 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

ıQ∧P (dF1 ∧ · · · ∧ dFk )

= ∑ sgn(ρ )(Q ∧ P)[Fρ (1) , . . . , Fρ (q+p) ]dFρ (q+p+1) ∧ · · · ∧ dFρ (k)
ρ ∈Sq+p,k−q−p

= ∑ sgn(τ )Q[Fτ (1) , . . . , Fτ (q) ]P[Fτ (q+1) , . . . , Fτ (q+p) ]

τ ∈Sq,p,k−q−p
dFτ (q+p+1) ∧ · · · ∧ dFτ (k) ,

which yields the third proof. The point of this computation is that an element of
Sq,p,k−p−q can be uniquely obtained from a shuffle σ ∈ Sq,k−q , followed by a shuffle
of the set σ (q + 1), . . . , σ (k), or from a shuffle ρ ∈ Sq+p,k−p−q , followed by a shuffle
of the set ρ (1), . . . , ρ (q + p).
The proof of (2) is immediate from (1), since the graded commutator is given by
[ıP , ıQ ] = ıP ◦ıQ −(−1) pq ıQ ◦ıP = ıQ∧P −(−1) pq ıP∧Q , and since P∧Q = (−1) pq Q∧ P
(graded commutativity, see (3.4)).
We now prove (3). Since, for every Q ∈ X• (A ), the map ıQ is A -linear and
the wedge product is also A -bilinear, we can suppose that ω is of the form ω =
dF1 ∧ · · · ∧ dFk , with k ∈ N∗ , k  p − 1 and F1 , . . . , Fk ∈ A . In this case,

ıP (dF ∧ dF1 ∧ · · · ∧ dFk )

= ∑ sgn(σ ) P[F, Fσ (1) , . . . , Fσ (p−1) ] dFσ (p) ∧ · · · ∧ dFσ (k)
σ ∈S p−1,k−p+1

+ ∑ (−1) p sgn(τ ) P[Fτ (1) , . . . , Fτ (p) ] dF ∧ dFτ (p+1) ∧ · · · ∧ dFτ (k) ,

τ ∈S p,k−p

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 ı(ıF P) (dF1 ∧ · · · ∧ dFk ) = ı(ıF P) ω ,
while the second sum is given by (−1) p dF ∧ ıP (dF1 ∧ · · · ∧ dFk ) = (−1) p dF ∧ ıP ω .
This yields a proof of Eq. (3.32).  

For P ∈ X p (A ), the operation ıP : Ω • (A ) → Ω •−p (A ) is closely related to the

natural extension of P : ∧ p A → A to a coderivation

P̃ : ∧• A → ∧•−p+1 (A )

of degree p − 1 of (∧• A , Δ ). It is defined by

3.3 The Schouten Bracket 79

P̃(F1 ∧ · · · ∧ Fk ) := ∑ sgn(τ ) P[Fτ (1) , . . . , Fτ (p) ] ∧ Fτ (p+1) ∧ · · · ∧ Fτ (k) , (3.35)

τ ∈S p,k−p

for k  p and P̃(F1 ∧ · · · ∧ Fk ) := 0 otherwise (see Example A.4 in Appendix A).

Comparing this formula with (3.29), it is clear that d ◦ ıP and P̃ correspond under
the map ∧• d : ∧• A → Ω • (A ), i.e., the following diagram is commutative:
P̃
∧• A ∧•−p+1 A

∧• d ∧• d

d◦ıP
Ω • (A ) Ω •−p+1 (A )

ıP d
•−p
Ω (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 Ω • (M), where we recall that X• (M) is
the F (M)-module of multivector fields on M and Ω • (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 ω ∈ Ω • (M) and F ∈ F (M).

3.3.2 The Schouten Bracket

We now introduce the Schouten bracket, which is a product of skew-symmetric

multi-derivations, based on the operation of composition. It is a family of maps

[· , ·]S : X p (A ) × Xq (A ) → X p+q−1 (A ) ,

for p, q ∈ N, which therefore only respects the grading up to a shift over 1, suggest-
ing to shift the degree of all multi-derivations by 1. Namely, we define X p−1 (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 Multi-Derivations and Kähler Forms

[P, Q]S [F1 , . . . , Fp+q+1 ]

 
= ∑ sgn(σ )P Q[Fσ (1) , . . . , Fσ (q) ], Fσ (q+1) , . . . , Fσ (q+p) (3.36)
σ ∈Sq,p
 
− (−1) pq ∑ sgn(σ )Q P[Fσ (1) , . . . , Fσ (p) ], Fσ (p+1) , . . . , Fσ (p+q) .
σ ∈S p,q

A priori, we have only that [P, Q]S ∈ Hom(A ⊗(p+q+1) , A ), but the reader will eas-
ily check that indeed [P, Q]S ∈ X p+q+1 (A ) = X p+q (A ), by showing that [P, Q]S is
skew-symmetric 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 elemen-
tary operations on functions, derivations and multi-derivations. First, let Q := F ∈ A
and P ∈ X p (A ), then [P, F]S = ıF 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 skew-symmetric 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 skew-
symmetric 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 Pois-
son brackets P and Q are compatible if and only if their sum (equivalently, an arbi-
trary linear combination with non-zero 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

skew-symmetric multi-derivations of A , then

[[ıP , d] , ıQ ] = ı[P,Q]S . (3.38)

This formula is called Cartan’s formula.

Proof. We prove Cartan’s formula for elements P ∈ X p (A ) and Q ∈ Xq (A ). Let

us first show that the linear map [[ıP , d], ıQ ] is A -linear. We take F ∈ A and ω ∈
Ω • (A ). Then, using (3.32),

[ıP , d](F ω ) = ıP (dF ∧ ω ) + FıP dω − (−1) p d(FıP ω )

3.3 The Schouten Bracket 81

= ı(ıF P) ω + FıP dω − (−1) p FdıP ω

= F[ıP , d]ω + ı(ıF P) ω ,

so that

[[ıP , d], ıQ ](F ω )

= [ıP , d](FıQ ω ) − (−1) pq ıQ [ıP , d](F ω )
= F[ıP , d](ıQ ω ) + ı(ıF P) ıQ ω − (−1) pq FıQ [ıP , d]ω − (−1) pq ıQ ı(ıF P) ω
= F[[ıP , d], ıQ ]ω ,
 
where we have used in the last step that ı(ıF P) , ıQ = 0 (Proposition 3.4). This shows
that
[[ıP , d] , ıQ ] : Ω • (A ) → Ω •−p−q+1 (A )
is A -linear. It follows that, in order to establish Cartan’s formula (3.38), we only
need to show that

ıP dıQ ω − (−1) p dıP ıQ ω − (−1) pq ıQ dıP ω = ı[P,Q]S ω , (3.39)

for ω of the form ω = 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 ω := 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 right-hand side of (3.39) (with ω = 1)
is given by ı[V ,F]S 1 = ıV [F] 1 = V [F], while the left-hand side is given by ıV dF +
dıV F − ıF dıV 1 = ıV dF = V [F], as ıV 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 left-
hand side of (3.39) reduces to ıF dG − d(FG) + ıG dF = FdG − d(FG) + GdF = 0.
This shows that (3.39) holds for ω of degree 0, for all P and Q.
We now assume that (3.39) holds for ω ∈ Ω k (A ) and we show that it holds
for dF ∧ ω , where F ∈ A is arbitrary. On the one hand, (3.32) and the induction
hypothesis imply that

ı[P,Q]S (dF ∧ ω )
= ı(ıF [P,Q]S ) ω − (−1) p+q dF ∧ ı[P,Q]S ω
= ı(ıF [P,Q]S ) ω − (−1) p+q dF ∧ ([[ıP , d] , ıQ ] ω )
= (−1)q ı[ıF P,Q]S ω + ı[P,ıF Q]S ω − (−1) p+q dF ∧ ([[ıP , d] , ıQ ] ω ) , (3.40)

where we have used

ıF [P, Q]S = (−1)q [ıF P, Q]S + [P, ıF Q]S

82 3 Multi-Derivations and Kähler 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 [[ıP , d] , ıQ ] (dF ∧ ω ), i.e.,
to
(ıP dıQ − (−1) p dıQ∧P − (−1) pq ıQ dıP )(dF ∧ ω ) . (3.41)
Using (3.32) twice and (3.25) once, we find

ıP dıQ (dF ∧ ω ) = ıP dı(ıF Q) ω − (−1)q ıP (dF ∧ dıQ ω )

= ıP dı(ıF Q) ω − (−1)q ı(ıF P) dıQ ω − (−1) p+q dF ∧ ıP dıQ ω .

Similarly, we compute, using (3.32) and (3.6)

dıQ∧P (dF ∧ ω ) = dı(ıF (Q∧P)) ω + (−1)q+p dF ∧ dıQ∧P ω

= dı(ıF Q∧P) ω + (−1)q dı(Q∧ıF P) ω + (−1)q+p dF ∧ dıQ∧P ω .

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:

(ıP dı(ıF Q) − (−1) p dı(ıF Q∧P) − (−1) pq ı(ıF Q) dıP ) ω = ı[P,ıF Q]S ω
(−1)q (ı(ıF P) dıQ + (−1) p dı(Q∧ıF P) − (−1) pq ıQ dı(ıF P) ) ω = (−1)q ı[ıF P,Q]S ω
 
dF ∧ −(−1) p+q ıP dıQ + (−1)q dıQ∧P − (−1) pq ıQ dıP ω =
−(−1) p+q dF ∧ ([[ıP , d] , ıQ ] ω ) .

This yields precisely the three terms in (3.40). 



Cartan’s 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 homoge-
neous elements P, Q and R in X• (A ), of respective shifted degree p, q and r,
(1) [P, Q]S ∈ X p+q (A );
(2) [P, Q]S = −(−1) pq [Q, P]S , (graded skew-symmetry);
(3) (−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:

[Q ∧ R, P]S = [Q, P]S ∧ R + (−1) pq Q ∧ [R, P]S , (3.42)

(graded Leibniz identity),

with P, Q and R as above. In view of the graded skew-symmetry 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 skew-symmetry also follows from Cartan’s for-
mula (3.38). Indeed, the graded Jacobi identity for the graded commutator [· , ·],

(−1) pq [[ıP , d] , ıQ ] + (−1) p [[d, ıQ ] , ıP ] + (−1)q [[ıQ , ıP ], d] = 0 ,

leads, since [ıQ , ıP ] = 0, to

ı[Q,P]S = [[ıQ , d], ıP ] = −(−1) pq [[ıP , d], ıQ ] = −(−1) pq ı[P,Q]S ,

which proves that [Q, P]S = −(−1) pq [P, Q]S , since P → ıP is injective. We next prove
the graded Jacobi identity for [· , ·]S . Let P ∈ X p (A ) and Q ∈ Xq (A ). We associate
to these multi-derivations, the coderivations P̃ (of degree p) and Q̃ (of degree q) of
the coalgebra (∧• A , Δ ), as in (3.35). We claim that


[P̃, Q̃] = [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

(P̃ ◦ Q̃)(F1 , . . . , Fk )

= P̃ ∑ sgn(σ ) Q[Fσ (1) , . . . , Fσ (q) ] ∧ Fσ (q+1) ∧ · · · ∧ Fσ (k)

σ ∈Sq,p
 
= ∑ sgn(σ )P Q[Fσ (1) , . . . , Fσ (q) ], Fσ (q+1) , . . . , Fσ (k) ,
σ ∈Sq,p

which leads, in view of (3.36), to


[P̃, Q̃](F1 , . . . , Fk ) = [P, Q]S [F1 , . . . , Fk ] = [P, Q]S (F1 , . . . , Fk ) .

This shows (3.44). As a consequence, for all skew-symmetric multi-derivations P, Q

and R,    

([P, [Q, R]S ]S )∼ = P̃, [Q, R]S = P̃, Q̃, R̃

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 Cartan’s for-
mula. Indeed, for P, Q and R as in (3.43), Cartan’s formula and (1) of Proposition 3.4
imply that

ı[P,Q∧R]S = [[ıP , d] , ıQ∧R ] = [[ıP , d] , ıR ◦ ıQ ]

84 3 Multi-Derivations and Kähler Forms

= [[ıP , d] , ıR ] ◦ ıQ + (−1)rp ıR ◦ [[ıP , d] , ıQ ]

= ıQ∧[P,R]S + (−1)rp ı[P,Q]S ∧R ,

which yields (3.43), and hence also (3.42). We have used the formula3

[ψ , φ1 ◦ φ2 ] = [ψ , φ1 ] ◦ φ2 + (−1)d1 d φ1 ◦ [ψ , φ2 ] , (3.45)

which is valid for graded linear maps φ1 , φ2 and ψ , 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 • = ⊕ p∈ZVp be a graded F-vector space and denote by p :=
p − 1 the shifted degree of P ∈ Vp . Suppose that V is equipped with two multiplica-
tions ∧ and [· , ·], such that
(1) (V • , ∧) is an associative, graded commutative algebra over F;
(2) (V • , [· , ·]) is a graded Lie algebra over F, with respect to the shifted grading;
(3) 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 asso-
ciative F-algebra A , the graded F-vector 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 F-vector 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 well-known 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 multi-derivations, in particular it only becomes a graded Lie algebra

structure when we shift the grading: for p ∈ N∗ , we let ∧ p−1 g := ∧ p g and we let
p := p − 1. The definition and main properties of the graded bracket on the graded
vector space ∧• g := ⊕k∈N ∧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 ]] (3.46)
p q
:= (−1) pq ∑ ∑ (−1)i+ j [xi , y j ] ∧ x1 ∧ · · · ∧ xi ∧ · · · ∧ 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
[[Y ∧ Z, X]] = [[Y, X]] ∧ Z + (−1) pqY ∧ [[Z, X]] , (3.47)
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 right-hand side of (3.47) are correct, which follows at once from the graded
skew-symmetry of [[· , ·]]. 

3.3.4 The (Generalized) Lie Derivative

We have already encountered the Lie derivative of a multi-derivation 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 be-
fore, 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
p-derivation 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 )
(3.48)
Q → [P, Q]S .
86 3 Multi-Derivations and Kähler 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) LP Q ∈ X p+q (A ) ;
(2) LP (Q ∧ R) = Q ∧ LP R + (−1) pr (LP Q) ∧ R ;
(3) 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 deriva-
tive with respect to an arbitrary p-vector field P; it associates to a q-vector 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 cal-
culus 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 ω be a differential p-form on M. Then LV ω is the differential
p-form on M, whose value at m ∈ M is given by
d  
(LV ω )m = Φ ∗ ωΦt (m) ,
dt |t=0 t

where Φt is the local flow of V and Φt∗ the pull-back of ω by Φt , so that Φt∗ (ωΦt (m) )
is a differential p-form on Tm M (depending on t). The classical4 Cartan formula says
that
LV = ıV ◦ d + d ◦ ıV , (3.49)
a formula which we can take as a definition of the Lie derivative in the alge-
braic setting. Writing (3.49) as LV = [ıV , 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 multi-derivation) on
a differential form (Kähler form), namely we define the Lie derivative with re-
spect to a skew-symmetric multi-derivation P ∈ X p (A ) as the graded F-linear map
Ω • (A ) → Ω •−p (A ) of degree −p, given by

LP := [ıP , d] ,

where we recall that ıP was defined in (3.29). Explicitly, this means that for ω ∈
Ω • (A ),
LP ω := ıP (dω ) − (−1) p d(ıP ω ) . (3.50)

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 Cartan’s 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 p-vector field, when p  2. In fact, a func-

tion F ∈ A can be viewed as an element of X0 (A ) or as an element of Ω 0 (A ).
According to (3.48), LP F = ıF (P) ∈ X p−1 (A ), which is a priori different from
zero; however, according to (3.50), LP (F) ∈ Ω 1−p (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 Def-
inition (3.50).
The main properties of the Lie derivative LP : Ω • (A ) → Ω •−p (A ) are summa-
rized 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) [LP , d] = 0 ;
(2) [LP , ıQ ] = ı[P,Q]S ;
(3) L[P,Q]S = [LP , LQ ] ;
(4) LP∧Q = ıQ ◦ LP + (−1) p LQ ◦ ıP .
Proof. Since d◦d = 0, we have that [LP , d] = −(−1) p d◦ıP ◦d+(−1) p d◦ıP ◦d = 0,
which leads to (1), while (2) is precisely Cartan’s formula (3.38). Combined with
(1) and (2), the graded Jacobi identity for [· , ·] yields

[LP , LQ ] = [LP , [ıQ , d]] = [[LP , ıQ ] , d] = ı[P,Q]S , d = L[P,Q]S ,

which proves (3). Finally, using ıP∧Q = ıQ ◦ ıP (see Proposition 3.4) and (3.45),

LP∧Q = [ıP∧Q , d] = [ıQ ◦ ıP , d]

= ıQ ◦ [ıP , d] + (−1) p [ıQ , d] ◦ ıP
= ıQ ◦ LP + (−1) p LQ ◦ ıP ,

which proves (4). 



3.4 Exercises

1. Prove the explicit formula (3.28) for a Kähler k-form ω on an arbitrary commu-
tative associative algebra A ; the arguments Vi are in this context arbitrary deriva-
tions of A .
2. Give an alternative proof of the graded Jacobi identity for the Schouten bracket
by using Cartan’s formula (3.38).
3. Let A be an arbitrary commutative associative algebra. Prove the following
formulas, for F ∈ A and ω ∈ Ω • (A ):

LFP ω = F LP ω − (−1) p dF ∧ ıP ω ,
LP (F ω ) = F LP ω + ı(ıF P) ω .
88 3 Multi-Derivations and Kähler Forms

These formulas show that LP ω is neither A -linear in P nor in ω .

4. Let A be a commutative associative algebra and let B be an A -module. A
linear map φ : A → B is called a derivation of A with values in B if for all
F1 , F2 ∈ A , one has φ (F1 F2 ) = F1 φ (F2 ) + F2 φ (F1 ).
a. 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 );
b. 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;
c. Interpret the de Rham differential d as a derivation of A with values in Ω 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 Ω 2 (A ) = 0.
7. The purpose of this exercise is to give an example of an algebra A , which
admits skew-symmetric 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

F[x, y, z]
A := .
yx, yz, y2 

∂ ∂
a. Show that P := y ∧ induces a non-zero skew-symmetric biderivation P
∂y ∂z
of A ;
b. 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 skew-

symmetric biderivation of A . Let A [ν ] denote the vector space of all polynomials
in ν with coefficients in A . Consider on A [ν ]/ν 2 the product, defined for ele-
ments F1 + ν F2 and G1 + ν G2 of A [ν ]/ν 2 by

(F1 + ν F2 )  (G1 + ν G2 ) := F1 G1 + ν (F1 G2 + F2 G1 + P[F1 , G1 ]) .

a. Show that  is F[ν ]-bilinear and associative;

b. Show that, if  can be extended to an associative F[ν ]-bilinear product on
A [ν ]/ν 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 Kähler forms, we refer to
Hartshorne [93].
Multivector fields are rarely explicitly treated in books on differential geome-
try; 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 Bhaskara–Viswanath [23]. Cannas da Silva–Weinstein [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 Schouten–Nijenhuis bracket, was intro-
duced by Schouten in [177, 178]; a generalization of the Schouten bracket, which
also generalizes the Fröhlicher–Nijenhuis 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
multi-derivations of A (multivector fields on M), and to homology spaces, derived
from the Kähler 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 co-
homology. 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 cohomol-
ogy. (A , ·, π ) 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 ) δLk HLk (g;V )
Poisson cohomology Xk (A ) δπk Hπk (A )
Lie homology V ⊗ ∧k g ∂kL HkL (g;V )
Poisson homology Ω k (A ) ∂kπ Hkπ (A )

In Sections 4.1 and 4.2, we construct the various complexes which lead to the ho-
mologies 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 Sec-
tion 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. Laurent-Gengoux et al., Poisson Structures, 91

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 4, © Springer-Verlag Berlin Heidelberg 2013
92 4 Poisson (Co)Homology

4.1 Lie Algebra and Poisson Cohomology

Before constructing the Poisson cohomology complex, and deriving the cohomol-
ogy spaces from it, we recall Lie algebra cohomology, which is formally very sim-
ilar 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 co-
homology 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 ρ : g → End(V ), where V is an F-vector space, called the
representation space and the Lie algebra structure on End(V ) is given by the com-
mutator: φ ◦ ψ − ψ ◦ φ , for φ , ψ ∈ End(V ). The vector space V , together with the
representation ρ , is called a g-module. We write ρx for ρ (x), where x ∈ g. We fix a
representation (ρ ,V ) of g and we introduce, for k ∈ N, the F-vector space

Ck (g;V ) := Hom(∧k g,V )

of skew-symmetric k-linear maps from g to V . Elements of Ck (g;V ) are called V -

valued k-cochains of g. The Lie coboundary map δLk : Ck (g;V ) → Ck+1 (g;V ) is
given, for c ∈ Ck (g;V ) and x0 , . . . , xk ∈ g, by
k
δLk (c)(x0 , . . . , xk ) = ∑ (−1)i ρxi c(x0 , . . . , xi , . . . , xk )
i=0
  (4.1)
+ ∑ (−1) i+ j
c [xi , x j ] , x0 , . . . , xi , . . . , xj , . . . , xk ,
0i< jk

where the symbol  over an argument means that we omit that argument. Ele-
ments of Ker δLk are called Lie k-cocycles, while elements in Im δLk−1 are called Lie
k-coboundaries. The Jacobi identity for [· , ·] and the fact that ρ is a Lie algebra ho-
momorphism permits one to show that δLk ◦ δLk−1 = 0, for every k ∈ N∗ . In words:
every coboundary is a cocycle. The converse is not true in general and the cohomol-
ogy spaces precisely measure the obstruction. Namely, for k ∈ N∗ , we define

HLk (g;V ) := Ker δLk / Im δLk−1 ,

and HL0 (g;V ) := Ker δL0 . Putting these spaces together, we obtain a graded F-vector
space 
HL• (g;V ) := HLk (g;V ) ,
k∈N

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 δL0 (v)(x) = ρx v, for
v ∈ V  C0 (g;V ) and x ∈ g. It follows that HL0 (g;V ) = Ker δL0 is given by

HL0 (g;V ) = Ker ρx ,
x∈g

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 (Whitehead’s lemma). Let g be a semi-simple Lie algebra and let ρ :
g → End(V ) be a representation of g. If V is finite-dimensional, then HL1 (g;V ) = 0
and HL2 (g;V ) = 0.
The following two examples of representations (ρ ,V ) are of particular interest
and will be our main cases.
Example 4.2. Let ρ : g → F be the trivial representation: ρx = 0 for every x ∈ g.
Then
Ck (g; F) = Hom(∧k g, F)
and (4.1) becomes:
 
δLk (c)(x0 , . . . , xk ) = ∑ (−1)i+ j c [xi , x j ] , x0 , . . . , xi , . . . , xj , . . . , xk . (4.2)
0i< jk

In this case, the Lie algebra cohomology of g is called the trivial Lie algebra coho-
mology of g and is simply denoted by HL• (g). It follows from (4.2) that Im δL0 = {0}
and that Ker δL1 = {c ∈ g∗ | ∀x, y ∈ g, c([x, y]) = 0} , so that
 ∗
HL1 (g)  g/ [g, g] .

For a semi-simple Lie algebra g, for example, it is well known that [g, g] = g (see
[190, Ch. 20]). Therefore, HL1 (g) = 0 when g is semi-simple, which is a special case
of Whitehead’s lemma (Lemma 4.1).
Example 4.3. Let ρ := ad, the adjoint representation of g on itself, which is given
by ρx = adx := [x, · ], for x ∈ g. Then

Ck (g; g) = Hom(∧k g, g)

and (4.1) becomes:

k
δLk (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 Chevalley–Eilenberg

cohomology of g. We denote it by HCE • (g). Let us give an interpretation of
94 4 Poisson (Co)Homology

1 (g). The 1-cocycles are the derivations of g, since

the space HCE

δL1 (c)(x, y) = [x, c(y)] + [c(x), y] − c([x, y])

for x, y ∈ g and c ∈ Hom(g, g). Since δL0 (x) = − adx , for x ∈ g  C0 (g; g), the space
of 1-coboundaries Im δL0 consists of all derivations of g of the form adx , where x ∈ g;
these derivations are called inner derivations. It follows that HCE1 (g) is the quotient

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.
The spaces HCE 2 (g) and H 3 (g) play an important rôle in the theory of formal
CE
deformations of Lie algebras. This fact is quite similar (by forgetting the multi-
derivation 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 π , so that π (F, G) = {F, G}, for F, G ∈ A .
For k ∈ N, the space of k-cochains of the Poisson cohomology complex is by
definition Xk (A ), the F-vector space of skew-symmetric k-derivations of A (see
Section 3.1.1). The Poisson coboundary operator is the Poisson analog of the
Chevalley–Eilenberg coboundary operator (4.3), where Hom(∧k g, g) has been re-
placed by Xk (A ), and where the Lie bracket [· , ·] is the Poisson bracket {· , ·}.
Namely, the graded F-linear map (of degree 1) δπ : X• (A ) → X•+1 (A ) is defined,
for Q ∈ Xq (A ), where q ∈ N, by
q
δπq (Q)[F0 , . . . , Fq ] := ∑ (−1)i i , . . . , Fq ]
Fi , Q[F0 , . . . , F (4.4)
i=0

+ ∑ (−1)i+ j Q i , . . . , Fj , . . . , Fq ,
Fi , Fj , F0 , . . . , F
0i< jq

for all F0 , . . . , Fq ∈ A . In order to establish that δπq (Q) is indeed a skew-symmetric

multi-derivation of A and that δπq+1 ◦ δπq = 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
δπq (Q) = −[Q, π ]S , (4.5)
for Q ∈ Xq (A ). Thus, δπq (Q) is the Schouten bracket of two skew-symmetric multi-
derivations, so it is itself a skew-symmetric multi-derivation. In order to see that
δπq+1 ◦ δπq = 0, we write down the graded Jacobi identity of the Schouten bracket
(Proposition 3.7) for the triple (Q, π , π ),
4.1 Lie Algebra and Poisson Cohomology 95

(−1)q−1 [Q, [π , π ]S ]S − [π , [Q, π ]S ]S + (−1)q−1 [π , [π , Q]S ]S = 0 , (4.6)

where we have used that the shifted degree q of a q-derivation is q − 1, in particular

the shifted degree of π is 1. Since [π , π ]S = 0 (because π is a Poisson bracket) and
since [Q, π ]S = −(−1)q [π , Q]S (graded skew-symmetry of [· , ·]S ), the identity (4.6)
reduces to [π , [π , Q]S ]S = 0, for every Q ∈ Xq (A ), showing that δπq+1 ◦ δπq = 0, for
all q ∈ N. Thus, we have a complex, the Poisson cohomology complex of A ,
q−1 q q+1
δπ δπ δπ
··· Xq−1 (A ) Xq (A ) Xq+1 (A ) ···

The elements of Ker δπq are called Poisson q-cocycles, while the elements of Im δπq−1
are called Poisson q-coboundaries. Elements of the q-th Poisson cohomology space
are Poisson q-cocycles modulo Poisson q-coboundaries,

Hπq (A ) := Ker δπq / Im δπq−1 ,

for q ∈ N∗ and Hπ0 (A ) := Ker δπ0 . The graded vector space


Hπ• (A ) := Hπq (A ) ,
q∈N

is called the Poisson cohomology of A .

In the case of a Poisson manifold (M, π ), one replaces in the above construction
the multi-derivations of A by the multivector fields on M, leading to the following
complex,
q−1 q q+1
δπ δπ δπ
··· Xq−1 (M) Xq (M) Xq+1 (M) ···

where the coboundary operator δπ is defined by (4.5), for Q ∈ Xq (M) (a q-vector

field on M). This complex is called the Poisson cohomology complex of M. The
q-th Poisson cohomology space of (M, π ), respectively the Poisson cohomology of
(M, π ), is denoted by Hπq (M) respectively Hπ• (M).

Remark 4.4. Let (A , ·, {· , ·}) be a Poisson algebra over F. Forgetting the asso-
ciative product, (A , {· , ·}) is simply a Lie algebra and it makes sense to con-
sider C• (A , A ), with the Chevalley–Eilenberg coboundary operator (4.3). This
yields precisely (4.4), so that the Poisson coboundary operator of (A , ·, {· , ·}) is
the Chevalley–Eilenberg coboundary operator, restricted to the multi-derivations
of (A , ·).

For small q, the Poisson coboundary operator δπq has a natural interpretation, which
yields a natural interpretation for the Poisson cohomology space Hπq (A ). One easily
reads off from (4.4) that for F ∈ X0 (A ) = A and for V ∈ X1 (A ),

δπ0 (F) = XF , δπ1 (V ) = −LV π , (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 ,

δπ2 (Q)[F, G, H] = {F, Q[G, H]} + Q[F, {G, H}]+  (F, G, H) .

It follows from (4.7) that Poisson 0-cocycles are Casimirs,

Hπ0 (A ) = Cas(A ) ,

Poisson 1-cocycles are Poisson derivations and Poisson 1-coboundaries are Hamil-
tonian derivations. Denoting the space of all Poisson derivations of A by P(A ), it
follows that
P(A )
Hπ1 (A ) = .
Ham(A )
Poisson 2-cocycles Q ∈ X2 (A ) are skew-symmetric biderivations which satisfy
[π , Q]S = 0, i.e., they are the ones which are compatible with {· , ·}, see Section 3.3.2;
Poisson 2-coboundaries are biderivations, obtained as a Lie derivative of the Poisson
structure. It follows that
skew-symmetric biderivations compatible with {· , ·}
Hπ2 (A ) = .
Lie derivatives of {· , ·}

As we will see in Section 13.2, Hπ2 (A ) and Hπ3 (A ) show up naturally in deforma-
tion 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

π  : Ω 1 (A ) → X1 (A )
(4.8)
G dF → GXF .

It is clear that this map is well-defined; for example, we have for all F, G ∈ A that

π  (d(FG) − FdG − GdF) = XFG − FXG − GXF = 0 ,

where we used (3) of Proposition 1.4 in the last step. The map π  extends naturally
to an A -linear map

∧• π  : Ω • (A ) → ∧• X1 (A ) → X• (A ) ,

which is explicitly given by

∧q π  (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 , π ).
Proposition 4.5. Let (A , ·, {· , ·}) be a Poisson algebra. The maps ∧q π  : Ω q (A ) →
Xq (A ) define a chain map from (Ω • (A ), d) to (X• (A ), δπ ), i.e., for every q ∈ N,
the following diagram is commutative:

d
Ω q (A ) Ω q+1 (A )

∧q π  ∧q+1 π 
q
δπ
Xq (A ) Xq+1 (A )

q
It leads for every q ∈ N to an F-linear map HdR (A ) → Hπq (A ).

Proof. Let ω := G dF1 ∧ · · · ∧ dFq ∈ Ω q (A ). On the one hand,

∧q+1 π  (dω ) = ∧q+1 π  (dG ∧ dF1 ∧ · · · ∧ dFq ) = XG ∧ XF1 ∧ · · · ∧ XFq .

On the other hand, (4.5) and (3.42) imply that

 
δπq (∧q π  (ω )) = δπq (GXF1 ∧ · · · ∧ XFq ) = − GXF1 ∧ · · · ∧ XFq , π S
= − [G, π ]S ∧ XF1 ∧ · · · ∧ XFq = XG ∧ XF1 ∧ · · · ∧ XFq ,

where we also used that each Hamiltonian vector field XFi is a Poisson 1-cocycle,
[XFi , π ]S = 0. This leads to the commutativity of the above diagram. It follows that
∧q π  sends q-cocycles to q-cocycles and sends q-coboundaries to q-coboundaries.
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 ϕ : A → B between two
Poisson algebras does not lead in general to a homomorphism between the spaces
of all multi-derivations 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 (ρ ,V ) is a representation of g. For k ∈ N,
we define
Ck (g;V ) := V ⊗ ∧k g ,
whose elements are called V -valued k-chains of g. Note that C0 (g,V )  V , since
∧0 g = F. The Lie boundary operator ∂kL : Ck (g;V ) → Ck−1 (g;V ) is the linear map
given, for all v ∈ V and x1 , . . . , xk ∈ g, by:
 
∂kL v ⊗ (x1 ∧ · · · ∧ xk )
k  
= ∑ (−1)i ρxi v ⊗ x1 ∧ · · · ∧ xi ∧ · · · ∧ xk
i=1
 
+ ∑ (−1)i+ j v ⊗ [xi , x j ] ∧ x1 ∧ · · · ∧ xi ∧ · · · ∧ xj ∧ · · · ∧ xk
1i< jk

and ∂0L := 0. We have that ∂k−1

L ◦ ∂ L = 0 as a consequence of the Jacobi identity
k
for [· , ·] and the Lie homomorphism property for ρ . The operators ∂kL lead to a
homology, called the V -valued Lie algebra homology of g. Namely, the k-th V -
valued Lie algebra homology space of g is denoted and defined by:

HkL (g;V ) := Ker ∂kL / Im ∂k+1

L
.

We denote by H•L (g;V ) the graded vector space


H•L (g;V ) := HkL (g;V ) .
k∈N

For v ∈ V and x ∈ g, we have that

∂1L (v ⊗ x) = −ρx 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 ρ = 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  
= ∑ (−1)i [xi , x0 ] ⊗ x1 ∧ · · · ∧ xi ∧ · · · ∧ xk (4.9)
i=1  
+ ∑ (−1)i+ j x0 ⊗ [xi , x j ] ∧ x1 ∧ · · · ∧ xi ∧ · · · ∧ xj ∧ · · · ∧ xk .
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 ρ : 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 π = {· , ·} on a commutative associative algebra A leads to

a homology, called its Poisson homology. The k-chains which define the Pois-
son homology complex are the Kähler 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 ∂ π : Ω • (A ) → Ω •−1 (A ) is given by

∂kπ (F0 dF1 ∧ · · · ∧ dFk )

k
= ∑ (−1)i {Fi , F0 } dF1 ∧ · · · ∧ dF
i ∧ · · · ∧ dFk
i=1

+ ∑ i ∧ · · · ∧ dF
(−1)i+ j F0 d Fi , Fj ∧ dF1 ∧ · · · ∧ dF j ∧ · · · ∧ dFk ,
1i< jk

where F0 , . . . , Fk ∈ A . For k = 0, this formula should be read as ∂0π (F0 ) := 0. The

main properties of ∂ π follow from the fact that ∂ π = Lπ = [ıπ , d], the Lie derivative
with respect to π , as stated in the following proposition.
Proposition 4.7. Let (A , ·, π ) be a Poisson algebra. Then the linear map ∂ π , de-
fined above, is given by the Lie derivative

∂ π = Lπ = [ıπ , d] = ıπ ◦ d − d ◦ ıπ .

It commutes (in the graded sense) with d and with ıπ ,

[∂ π , d] = ∂ π ◦ d + d ◦ ∂ π = 0 , (4.10)
π π π
[∂ , ıπ ] = ∂ ◦ ıπ − ıπ ◦ ∂ = 0 , (4.11)

and satisfies ∂ π ◦ ∂ π = 0 (i.e., it is a boundary operator).

100 4 Poisson (Co)Homology

Proof. For the biderivation π = {· , ·}, the internal product ıπ is, according to (3.29),
explicitly given by

ıπ (F0 dF1 ∧ · · · ∧ dFk )

= ∑ i · · · dF
(−1)i+ j+1 F0 Fi , Fj dF1 ∧ · · · ∧ dF j ∧ · · · ∧ dFk ,
1i< jk

where F0 , . . . , Fk ∈ A , so that

ıπ ◦ d(F0 dF1 ∧ · · · ∧ dFk )

= ∑ i . . . dF
(−1)i+ j+1 Fi , Fj dF0 ∧ · · · ∧ dF j ∧ · · · ∧ dFk ,
0i< jk

and

d ◦ ıπ (F0 dF1 ∧ · · · ∧ dFk )

= ∑ i . . . dF
(−1)i+ j+1 Fi , Fj dF0 ∧ · · · ∧ dF j ∧ · · · ∧ dFk ,
1i< jk

+ ∑ i . . . dF
(−1)i+ j+1 F0 d Fi , Fj ∧ dF1 ∧ · · · ∧ dF j ∧ · · · ∧ dFk .
1i< jk

Taking the difference of these two formulas leads at once to ∂ π = [ıπ , d] = Lπ . The
properties of the Lie derivative (Proposition 3.11) imply that ∂ π ◦ d + d ◦ ∂ π = 0,
and that
[∂ π , ıπ ] = [Lπ , ıπ ] = ı[π ,π ]S = 0 ,
where we used in the last equality that π is a Poisson structure. This yields (4.10)
and (4.11). Similarly, since Lπ has degree −1,

2 Lπ ◦ Lπ = [Lπ , Lπ ] = L[π ,π ]S = 0 ,

which shows that ∂ π = Lπ is a boundary operator.

According to the proposition, we have a complex

π π
∂k+1 ∂kπ ∂k−1
··· Ω k+1 (A ) Ω k (A ) Ω k−1 (A ) ···

whose homology is called the Poisson homology of (A , ·, π ). Namely, we define the

k-th Poisson homology space

Hkπ (A ) := Ker ∂kπ / Im ∂k+1

π
,

and we call elements of Ker ∂kπ Poisson k-cycles and elements of Im ∂k+1
π Poisson
k-boundaries. Putting the Poisson homology spaces together leads to the graded
vector space H•π (A ) := ⊕k∈N Hkπ (A ). In the case of a Poisson manifold (M, π ),
4.3 Operations in Homology and Cohomology 101

Poisson homology is defined similarly, using the differential forms on M, rather

than the Kähler forms on A . It is denoted by H•π (M).
For example, H0π (A ) is the abelianization of A : since Ker ∂0π = A while
Im ∂1π = π (A , A ) = {A , A }, we have that

A
H0π (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) func-

tor. Indeed, every algebra homomorphism ϕ : (A , ·) → (B, ·) extends to a degree
zero map Ω • (ϕ ) : Ω • (A ) → Ω • (B), which commutes with d. For k ∈ N, it is given
by Ω k (ϕ )(F0 dF1 ∧ · · · ∧ dFk ) = ϕ (F0 )dϕ (F1 ) ∧ · · · ∧ dϕ (Fk ), where F0 , . . . , Fk ∈ A .
If A and B are equipped with Poisson brackets πA and πB , and ϕ is a mor-
phism of Poisson algebras, then Ω • (ϕ ) ◦ ıπA = ıπB ◦ Ω • (ϕ ), so that Ω • (ϕ ) ◦ ∂ πA =
∂ πB ◦ Ω • (ϕ ). Thus, there is an induced map H•π (ϕ ) : H•π (A ) → H•π (B), which is
a homomorphism of graded vector spaces. Clearly, H π 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 prod-
ucts make (X• (A ), ∧, [· , ·]S ) into a Gerstenhaber algebra. Moreover, we have two
natural operations ı (the internal product) and L (the Lie derivative) of X• (A ) on
the module of Kähler differentials Ω • (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 = π , Q ∈ Xq (A ) and R ∈ Xr (A ), with δπ = − [·, π ]S ,

δπ (Q ∧ R) = δπ (Q) ∧ R + (−1)q Q ∧ δπ (R) ,

δπ ([Q, R]S ) = [Q, δπ (R)]S + (−1)r [δπ (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 well-defined 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 , ·, π ), (Hπ• (A ), ∧, [· , ·]S ) is a Ger-
stenhaber 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 ∂ π is the Lie derivative Lπ , we take from Proposi-
tion 3.11 the formulas which compute the graded commutator of a Lie derivative
with an internal product, or with another Lie derivative, namely,

[LP , ıQ ] = ı[P,Q]S and [LP , LQ ] = L[P,Q]S ,

which can be written, for P = π and ω ∈ Ω • (A ), as

(−1)q ∂ π (ıQ ω ) = ıQ ∂ π (ω ) − ıδπ (Q) ω ,

(−1)q ∂ π (LQ ω ) = LQ (∂ π (ω )) + Lδπ (Q) ω .

It follows from these formulas, as above, that ı 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 Hπ1 (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 λ on M,
preserved by all Hamiltonian flows?”
We call a Poisson manifold, which admits such a volume form, a unimodular Pois-
son 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, ω ), the answer to the above question is yes. In-
deed, since the symplectic structure ω is preserved by all Hamiltonian flows, the
same is true for the Liouville form ω d on M, where dim M = 2d. Thus, every sym-
plectic 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 λ is an arbitrary volume form on M. Recall that this means that
λ is a differential d-form 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 dλ = 0 and that dF ∧ λ = 0, for every F ∈
F (M). Combined with (3.32), this yields

ı(ıF P) λ = −(−1) p dF ∧ ıP λ , (4.12)

for all P ∈ X p (M). In particular,

V [F]λ = dF ∧ ıV λ , (4.13)

for every vector field V ∈ X1 (M). Also, if π is a Poisson structure on M, then (4.12)
specializes to
ıXF λ = dF ∧ ıπ λ . (4.14)
The Lie derivative of λ with respect to some vector field is again a top form, hence it
is a multiple of λ . For F ∈ F (M), let us denote by DF ∈ F (M) the smooth function
on M defined by LXF λ = DF λ . We show that the map F → DF is a derivation.
Indeed, for all F, G ∈ F (M), we have that

DFG λ = LXFG λ = LGXF +FXG λ

= GLXF λ + {G, F} λ + FLXG λ + {F, G} λ
= (GDF + FDG ) λ ,

where the identity

LHV λ = HLV λ + V [H]λ (4.15)
for all H ∈ F (M), V ∈ X1 (M) has been used (to prove (4.15), write LV = ıV ◦ d +
d ◦ ıV 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, π ) be a real Poisson manifold, equipped with a volume
form λ . The vector field Φλ , defined by

LXF λ = Φλ [F]λ , (4.16)

for all F ∈ F (M), is called the modular vector field of π with respect to λ .
When the chosen volume form is clear from the context, we simply write Φ for Φλ .
Notice that the modular vector field Φ can also be defined as the unique vector field
on M, satisfying
Lπ λ = ı Φ λ . (4.17)
This is seen by writing Φ [F]λ in two different ways. On the one hand, (4.13) tells
us that Φ [F]λ = dF ∧ ıΦ λ , for all F ∈ F (M). On the other hand, (4.16), (4.14) and
the formula Lπ = ıπ ◦ d − d ◦ ıπ imply that

Φ [F]λ = LXF λ = d(ıXF λ ) = d(dF ∧ ıπ λ ) = −dF ∧ dıπ λ = dF ∧ Lπ λ ,

for all F ∈ F (M). It follows that dF ∧ıΦ λ = dF ∧Lπ λ 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 λ on M,
with respect to which the modular vector field Φ 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 λ a volume form on M, the modular vector field Φ has the following properties:
(1) Φ is a Poisson vector field, i.e., it is a Poisson 1-cocycle;
(2) The Poisson cohomology class of Φ is independent of the chosen volume
form λ ;
(3) If Φ is a Hamiltonian vector field, i.e., if the Poisson cohomology class of Φ
is zero, then there exists a volume form λ  , with respect to which the modular
vector field is zero (and therefore λ  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 Φ [{F, G}] = {Φ [F], G} +
{F, Φ [G]}, for all F, G ∈ F (M). The basic properties of the Poisson bracket and
the Lie derivative yield:
 
Φ [{F, G}] λ = LX{F,G} λ = L[XG ,XF ] λ = LXG , LXF λ
= LXG (Φ [F]λ ) − LXF (Φ [G]λ )
= ({Φ [F], G} + {F, Φ [G]}) λ ,

where we have used in the last step that

Φ [F]LXG λ = Φ [F]Φ [G]λ = Φ [G]LXF λ .

This proves (1). Let us consider now another volume form μ on M and let us denote
by Φλ and Φμ the modular class of {· , ·} with respect to λ , respectively μ . There
exists a nowhere vanishing function ρ ∈ F (M) such that μ = ρ λ . Using that λ and
μ are both top forms and (4.13), we compute, for an arbitrary F ∈ F (M),

LXF (ρλ ) = XF [ρ ]λ + ρ LXF λ = −Xρ [F]λ + ρ Φλ [F]λ ,

to conclude that
1
Φμ [F] = Φλ [F] − Xρ [F] ,
ρ
for all F ∈ F (M), and hence that

1
Φμ = Φλ − Xρ = Φλ − Xln |ρ | .
ρ
This shows that the vector fields Φλ and Φμ differ by a Hamiltonian vector field,
i.e., a Poisson 1-coboundary. 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., Φ = XF for some F ∈ F (M),
then μ := exp(F)λ is a volume form whose modular vector field is zero.

Summarizing, we have a complete answer to our question:

Answer: “The Poisson manifold (M, π ) 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 ori-
entable. The cohomology class of the modular vector field (with respect to an ar-
bitrary 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 invari-
ant density on a unimodular Poisson manifold is unique.

Remark 4.13. In the case of a finite-dimensional vector space V , a natural volume

form is given, in terms of linear coordinates x1 , . . . , xd on V , by λ0 := dx1 ∧ · · · ∧ dxd ,
and every top form on V is of the form F λ0 , where F is an arbitrary function on V .
Up to a non-zero constant, λ0 is the only translation invariant volume form on V . If
F is assumed to be algebraically closed, then λ0 is, again up to a non-zero constant,
the unique volume form on V , because every non-constant polynomial F has a non-
empty zero locus, on which the top form F λ0 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 λ be
a volume form on M.
(1) The modular vector field Φ 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;
(2) 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
r
∂ ∂
{· , ·} = ∑ ∧ ,
i=1 ∂ qi ∂ pi

on U, where Rkm {· , ·} = 2r. We take on U as volume form


r 
s
λ := (dqi ∧ dpi ) dz j ,
i=1 j=1

and we denote
 
s
λk := (dqi ∧ dpi ) dz j ,
i=k 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
=d ∧ dpk − dqk ∧ d .
∂ pk ∂ qk

Therefore, for all k = 1, . . . , r,

 
∂ 2F ∂ 2F
λk ∧ LXF (dqk ∧ dpk ) = λ − =0,
∂ qk ∂ pk ∂ pk ∂ qk

from which it follows that LXF λ = 0, i.e., the modular vector field Φ 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
non-vanishing 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 Φ = ∂ y . In particular, the modular
class of this Poisson structure is not zero.

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 λ ). The divergence operator (with respect to λ ) is
the graded F-linear map (of degree −1),

Div : X• (M) → X•−1 (M) ,

which makes the following diagram commute

X• (M)

Ω d−• (M)

Div d (4.18)

X•−1 (M)  Ω d−•+1 (M)

where d := dim M. The diagram involves the star operator , which is the family of
F (M)-linear isomorphisms  : Xk (M) → Ω d−k (M), defined for Q ∈ Xk (M) by

 Q := ıQ λ . (4.19)
108 4 Poisson (Co)Homology

Since λ is a top form, we have for every vector field V on M that

LV λ = dıV λ = d  V =  Div(V ) = Div(V ) λ . (4.20)

For example, taking the standard volume form on Rd , i.e., λ = 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

d
V = ∑ (−1)i−1 Fi dx1 ∧ · · · ∧ dx
i ∧ · · · ∧ dxd ,
i=1

P = ∑ i ∧ · · · ∧ dx
(−1)i+ j−1 Pi j dx1 ∧ · · · ∧ dx j ∧ · · · ∧ dxd ,
1i< jd

so that Div(V ) is given by the classical formula

d
∂ Fi
Div(V ) = ∑ , (4.21)
i=1 ∂ xi

and Div(P) is given by

 
∂ Pi j ∂ ∂ Pi j ∂
Div(P) = ∑ ∂ x j ∂ xi
−
∂ xi ∂ x j
. (4.22)
1i< jd

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 λ 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 λ 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

dıV ∧W λ = −LV ∧W λ = LW ıV λ − ıW LV λ = ıV LW λ − ıW LV λ − ı[V ,W ] λ .

According to the definition of the divergence, this leads to

Div(V ∧ W ) = −1 dıV ∧W λ

= −1 ıV LW λ − −1 ıW LV λ − −1 ı[V ,W ] λ
= Div(W )V − Div(V )W − [V , W ] ,

where we used in the last step that ıV LW λ = ıV (Div(W )λ ) = ıDiv(W )V λ =

(Div(W )V ) and that  [V , W ] = ı[V ,W ] λ .
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, π ) be a real Poisson manifold, which is assumed ori-
entable. For every volume form λ on M,

Φ = − Div(π ) . (4.24)

In particular, the divergence of the modular vector field of π is zero.

Proof. Let λ be an arbitrary volume form on M and let F ∈ F (M). The definitions
of Φ and of , combined with (4.17), imply that

 Div(π ) = d  π = d(ıπ λ ) = −Lπ λ = −ıΦ λ = −  Φ . (4.25)

Since  is an isomorphism, this shows that the modular vector field is minus the
divergence of π . Since Div ◦ Div = −1 ◦ d ◦ d ◦  = 0, the modular vector field is
divergence-free.
According to (4.25), the modular vector field can also be defined by the formula

dπ = − Φ ;

forgetting the stars, the modular vector field is the differential of the Poisson bracket.
The string (4.25) also contains the identity Lπ λ =  Φ , which amounts to

∂ π (λ ) =  Φ . (4.26)

Thus,  Φ is always a Poisson 1-boundary, while Φ is a Poisson 1-coboundary 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, π ) be a d-dimensional Poisson mani-
fold. We first show how the Poisson boundary and coboundary operators are related
via the modular vector field Φ . To do this, we specialize Cartan’s formula (3.38)

[[ıP , d] , ıQ ] = ı[P,Q]S ,

valid for P ∈ X p (M) and Q ∈ Xq (M), to the case P = π , giving

∂ π ◦ ıQ − (−1)q ıQ ◦ ∂ π = (−1)q−1 ıδπ (Q) . (4.27)

If we apply (4.27) to λ and use (4.26), then we find

∂ π (Q) − (−1)q ıQ ◦ ıΦ λ = (−1)q−1  δπ (Q) ,

which we also write, using ıQ ◦ ıP = ıP∧Q , valid for P, Q ∈ X• (M) (see Proposi-
tion 3.4), as follows,
110 4 Poisson (Co)Homology

 (Φ ∧ Q) =  δπ (Q) + (−1)q ∂ π ( Q) . (4.28)

The latter formula says that the modular vector field Φ measures the non-commuta-
tivity of the following diagram

X• (M)

Ω d−• (M)

Φ ∧
δπ ∂π (4.29)

X•+1 (M)  Ω d−1−• (M)

The reader is invited to compare the diagrams (4.29) and (4.18), keeping in mind
that both depend on the volume form λ , but only the second one depends on the
Poisson structure π . 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:
π
Hπk (M)  Hd−k (M), 0kd. (4.30)

Proof. Let λ be an invariant density on M, so that Φ = 0. Then (4.28) implies that

the isomorphism  : X• (M) → Ω 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 struc-
ture, 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
∂ π (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 ad-invariant, 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 3-cocycle
in the trivial Lie algebra cohomology.
6. Consider the Poisson structure π on F2 given by

∂ ∂
π := x ∧ .
∂x ∂y
a. Show that the modular class of π is not trivial;
b. Compute the Poisson homology and the Poisson cohomology of π ;
c. Conclude from part b. that the conclusion of Proposition 4.18 does not hold for
π (F2 )
the Poisson manifold (F2 , π ), i.e., that the vector spaces Hπk (F2 ) and H2−k
are not isomorphic, for k = 0, 1, 2.

7. Let (A , ·, {· , ·}) be a Poisson algebra.

a. Show that the skew-symmetric biderivation {· , ·} is a Poisson 2-cocycle for the
Poisson cohomology of A ;
b. Assume that A is a polynomial algebra, A = F[x1 , . . . , xd ], and that the poly-
nomials xi , x j , for i, j = 1, . . . , d, are all homogeneous of the same degree k.
Show that if k = 2, then the skew-symmetric biderivation {· , ·} is a coboundary
(Hint: compute δπ1 (E ) where E := ∑di=1 xi ∂∂xi );
c. Assume now that A := C[x, y, z], and that

∂ ∂ ∂ ∂ ∂ ∂
π := x2 ∧ + y2 ∧ + z2 ∧ .
∂y ∂z ∂z ∂x ∂x ∂y

Show that π 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 δπ1 (V ) = π , then there exists such a derivation where F1 , F2
and F3 are linear functions).

4.6 Notes

Poisson cohomology was first introduced in Lichnerowicz’s paper [127], where it

was shown that the Poisson cohomology of a symplectic manifold is canonically
isomorphic to its de Rham cohomology. See Bhaskara–Viswanath [23] and Hueb-
schmann [96] for the algebraic cohomology of a general Poisson algebra. The re-
lation between Poisson cohomology and singularities is studied by Monnier [152]
and Roger–Vanhaecke [174] in the two-dimensional case and by Pichereau [166]
in the three-dimensional case. The definition of Poisson homology goes back to
Gelfand–Dorfman [81], Koszul [118] and Brylinski [28]. The modular vector field
of a Poisson manifold was first introduced by Koszul [118]. It was studied for gen-
eral Poisson manifolds and extended to Lie algebroids by Weinstein [201], Evans–
Lu–Weinstein [71] and Kosmann–Weinstein [116]. See Fernandes–Crainic [51, 52]
112 4 Poisson (Co)Homology

for applications of Poisson cohomology to the linearization and stability of Pois-

son 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 reduc-
tion. 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 construc-
tions 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) sub-
manifolds of Poisson manifolds. Poisson reduction is presented in Section 5.2. The
second construction, called Poisson–Dirac reduction, concerns Poisson structures
on submanifolds, which are not necessarily Poisson submanifolds; a typical exam-
ple is a submanifold which is transverse to a symplectic leaf, leading to the notion
of transverse Poisson structure. Poisson–Dirac reduction is discussed in Section 5.3.
As an application of these constructions, we consider in Section 5.4 Poisson va-
rieties and Poisson manifolds on which a group acts; we show how a Poisson struc-
ture is inherited on the quotient space, the subspace of fixed points of the action, and
on the reduced space of the so-called 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. Laurent-Gengoux et al., Poisson Structures, 113

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 5, © Springer-Verlag Berlin Heidelberg 2013
114 5 Reduction

5.1 Lie Groups and (Their) Lie Algebras

We recall in this section the natural correspondence between Lie groups and Lie al-
gebras, 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)
μ : G × G → G : (g, h) → gh
is assumed to be smooth (or holomorphic). Then the inverse map

inv : G → G : g → g−1

is also smooth (or holomorphic). The product of two elements g, h of G is denoted

by μ (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
Φ : 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 → ghg−1 .

The tangent maps of μ and inv can be expressed in terms of these diffeomorphisms:
since μ (g, h) = Lg (h) = Rh (g) for all g, h ∈ G, the tangent map of μ at (g, h) is given
by
T(g,h) μ (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 ◦ Lg−1 =
1G at e and μ (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 Lg−1 ◦ Tg Rg−1 . (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 ◦

Rg−1 , valid for g ∈ G, that the tangent map of the conjugation map Cg is given, at
the unit e, by
TeCg = Tg−1 Lg ◦ Te Rg−1 . (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
map Te Lg . For a fixed x ∈ Te G, this allows us to define a vector field ←
−
x on G by
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 left-invariant (respectively right-invariant) 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 left-invariant, since

Th Lg (←
−
x h ) = Th Lg (Te Lh (x)) = Te (Lg ◦ Lh )(x) = Te Lgh (x) = ←
−
x gh ,

for all g, h ∈ G. Similarly, each of the vector fields → −

x is right-invariant. It fol-
lows that there is a natural identification between Te G and the vector space of all
left-invariant vector fields on G (or the vector space of all right-invariant vector
fields on G). Since the commutator of two left-invariant vector fields on G is a left-
invariant 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 (Lie’s theorem). Let G and H be Lie groups, whose units are denoted
by e and e respectively, and let Φ : G → H be a Lie group homomorphism.
(1) The tangent space Te G is equipped with a natural Lie bracket [· , ·];
(2) The linear map Te Φ : 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 Φ : g → h is called the Lie
algebra homomorphism associated to Φ .
Conversely, let g and h be finite-dimensional Lie algebras and let φ : g → h be a
Lie algebra homomorphism.
(3) 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;
(4) 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 Φ : G → H, such that φ is the Lie algebra homomorphism
associated to Φ ;
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 neighbor-
hood 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 left-invariant 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 Lie’s 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 finite-dimensional representations are completely reducible; as we
will explain at the end of Section 5.1.2, such quotients are rather well-behaved.

5.1.2 Lie Group Actions

Given a Lie group G and a manifold M, a group action χ : G × M → M is called

a Lie group action, if χ is smooth (or holomorphic); since the only actions of Lie
groups on manifolds which are considered here are Lie group actions, we will sim-
ply call them group actions. When the action is clear, we write gm for χ (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(g−1 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 χ : G × V → V with the property that for every g ∈ G the map χg : V → V ,
defined for all x ∈ g by χg (x) := χ (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 χ 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 117

L : G × G → G : (g, h) → gh ,
R : G × G → G : (g, h) → hg−1 ,
C : G × G → G : (g, h) → ghg−1 .

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 so-called right action); we stress that it
amounts to R(g, h) = Rg−1 (h) for all g, h ∈ G.
For a fixed group action χ 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
xm [F] := F(χ (exp(−tx), m)) , (5.5)
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) = Rg−1 h = hg−1 ), this yields for all functions F,
defined in a neighborhood of g in G,
d d
xg [F] = 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
= Te Lg (x)[F] = ← −
x g [F] .

We conclude that xg = ← −
x g , for all g ∈ G, i.e., x = ←
−
x . In words: the left-invariant
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 = Rg−1 ◦ Lg ,
the fundamental vector field of conjugation on G, corresponding to x ∈ g, is given
by ←−
x −→ −x . The identity
inv ◦ Rg = Lg−1 ◦ inv
implies that a vector field V on G is right-invariant if and only if the vector field
on G, defined1 by inv∗ V , is left-invariant. It follows, using Te inv = −1Te G , that
inv∗ ←
−
x = −− →
x , for every x ∈ g.
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
p-vector field X is defined as
1Recall that for a diffeomorphism Φ : M → N between manifolds and for V a vector field on M,
we denote by Φ∗ V the vector field on N, which is defined by (Φ∗ V )Φ (m) := Tm Φ (Vm ), for all
m ∈ M, and similarly for multivector fields on M.
118 5 Reduction

X := x1 ∧ · · · ∧ x p

and is called the fundamental p-vector 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
fundamental multivector field, associated to X ∈ ∧ g. Equivalently, for X ∈ ∧ p g, the
p
←
− →
− ←
− →
−
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 left-invariant
→
−
multivector field and that X is a right-invariant multivector field.
←−
We show in the following proposition that the map, which sends X ∈ ∧• g to X ,
is a homomorphism of Gerstenhaber algebras. 2

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 , [[X,Y ]] = − X , Y S . (5.7)

Moreover, ←
−−→
X,Y S=0, (5.8)
for all X,Y ∈ ∧• 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
←
− →
−
general X,Y ∈ ∧• g∗ . Since inv∗ ← −x = −− →x , for all x ∈ g, we have inv∗ X = (−1) p X ,
for all X ∈ ∧ g; therefore, applying inv∗ to both equations in (5.6), we find the
p

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 ac-
tions on vector spaces (representations), such as the actions which we will con-
sider in the next section, taking the differential of the action, viewed as a map
χ : 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).
2 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 G-invariant 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 χ : G × M → M is called a proper action, if
the map
χ × p2 : G × M → M × M
(5.9)
(g, m) → (gm, m)
is a proper map, i.e., the inverse image of every compact subset of M × M is a
compact subset of G × M. Also, χ 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
ξ1 , . . . , ξ p ∈ g∗ and for all x1 , . . . , x p ∈ g by
 
 ξ1 [x1 ] · · · ξ1 [x p ] 
   . ..  .

ξ1 ∧ · · · ∧ ξ p , x1 , . . . , x p :=  .. .  (5.10)
 
 ξ p [x1 ] · · · ξ 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 Adg−1 is denoted by Ad∗g : for all ξ ∈ g∗ and for all x ∈ g,
   
Ad∗g ξ , x = ξ , Adg−1 x .

The resulting map (g, ξ ) → Ad∗g ξ is a group action of G on g∗ , called the coad-
joint 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)

Ad∗g (ξ1 ∧ · · · ∧ ξ p ) := Ad∗g ξ1 ∧ · · · ∧ Ad∗g ξ p , (5.12)

for all g ∈ G, all p ∈ N∗ , all x1 , . . . , x p ∈ g and all ξ1 , . . . , ξ p ∈ g∗ . It follows from

these definitions that  ∗   
Adg Ω , X = Ω , Adg−1 X ,

for all g ∈ G, all Ω ∈ ∧ 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 rep-
resentation 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 ad∗x is related to adx by

ad∗x Ω , X = − Ω , adx X = − Ω , [[x, X]] , (5.13)

for all Ω ∈ ∧ p g∗ and all X ∈ ∧ p g.

To finish this section, we shortly discuss invariant multivectors. An element X ∈
∧• g is said to be Ad-invariant (respectively ad-invariant) if Adg X = X, for all g ∈ G
(respectively adx X = 0, for all x ∈ g). The following lemma relates Ad and ad-
invariance 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) If X is Ad-invariant, then X is ad-invariant;
(2) If G is connected, then X is Ad-invariant if and only if X is ad-invariant;
←
− → −
(3) X is Ad-invariant if and only if X = X .
Proof. Let X ∈ ∧• g. If X is Ad-invariant, 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 Ad-invariant. 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 Rg−1 ) ◦ ∧ 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 Ad-invariant, 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
Ad-invariant, if for all g ∈ G and for all x, y ∈ g,
 
Adg x | Adg y = x | y .

If · | · is Ad-invariant, 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 ad-
invariant. When G is connected, every bilinear form which is ad-invariant is Ad-
invariant, so the two notions of invariance need not be distinguished.
We often use a non-degenerate symmetric bilinear form to identify g with its dual
vector space g∗ . Non-degeneracy of a bilinear form · | · means that the linear map
χ : g → g∗ , which sends x ∈ g to the linear form x | · : y → x | y, is an isomorphism.
Its inverse will be denoted by ψ . Explicitly, χ is given by

χ (x), y = x | y = χ (y), x ,

for all x, y ∈ g. If · | · is moreover Ad-invariant, then for every g ∈ G and for all
x, y ∈ g it follows from the definitions that
 ∗     
Adg (χ (x)), y = χ (x), Adg−1 y = x | Adg−1 y
   
= Adg x | y = χ (Adg x), y ,

so that the following diagram is commutative:

χ
g g∗

Adg Ad∗g (5.14)

g
χ
g∗

A Lie algebra g, equipped with a non-degenerate symmetric bilinear form · | ·,

which is ad-invariant is called a quadratic Lie algebra.
A function F on g is called Ad-invariant 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

ad∗dξ F ξ = 0 , (5.15)

for all ξ ∈ g∗ . Indeed, for z ∈ g and ξ ∈ g∗ ,

      
ad∗dξ F ξ , z = ξ , z, dξ F = − dξ F, ad∗z ξ = 0 ,

where we used in the last step that dtd |t=0 F(Ad∗exptz ξ ) = 0, itself a direct conse-
quence of the Ad∗ -invariance of F. If F : g∗ → F is an Ad∗ -invariant function and
if the bilinear form · | · is Ad-invariant, then the commutativity of (5.14) leads for
every g ∈ G to
F ◦ χ ◦ Adg = F ◦ Ad∗g ◦χ = F ◦ χ ,
so that F ◦ χ is Ad-invariant. Similarly, · | · is Ad-invariant and F ◦ χ is Ad-
invariant, then F is Ad∗ -invariant. Thus, χ establishes a one-to-one correspondence
5.2 Poisson Reduction 123

between Ad∗ -invariant functions on g∗ and Ad-invariant functions on g. For a func-

tion F : g → F, the Ad-invariance of · | · also leads to
 
ψ ad∗dξ (F◦ψ ) ξ = ∇ψ (ξ ) F, ψ (ξ ) , (5.16)

where ∇x F := ψ (dx F), for x ∈ g. It follows that a function F on g is Ad-invariant if

and only if [∇x F, x] = 0 for all x ∈ g. The proof of (5.16) follows at once from the
definitions of ψ , ad, ∇ and the ad-invariance of · | ·.

5.2 Poisson Reduction

In many geometrical situations of interest, one deals with a quotient of a subman-

ifold 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)

where M = (M, π ) is a Poisson manifold, N is a submanifold of M and p is a sub-

mersion. The example to keep in mind is that of a Poisson manifold on which a
group G acts, a submanifold N which is G-invariant 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 dia-
gram (5.17), we consider a Poisson algebra (A , ·, {· , ·}), an ideal I of (A , ·), the
canonical map κ : A → A /I and a subalgebra B of A /I , as in the following
diagram:
124 5 Reduction

A
κ (5.18)

B ⊂ A /I

We wouldlike to have a Poisson structure {· , ·}B on B, such that {F, G}B =

κ F̃, G̃ , for every F, G ∈ B, where F̃ and G̃ are arbitrary representatives in A
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

{F, G}B = κ F̃, G̃ , (5.19)

for all F, G ∈ B, where F̃, G̃ are arbitrary representatives of F and G in A . The

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) κ −1 (B) ⊂ N (I );
(2) κ −1 (B) is a Lie subalgebra of (A , {· , ·}).
For F, G ∈ B, their reduced Poisson bracket {F, G}B is then given by

{F, G}B = κ F̃, G̃ , (5.21)

where F̃ and G̃ are arbitrary representatives in A of F and G.

Proof. Suppose first that conditions (1) and (2) are satisfied. We wish to define a
Poisson bracket on B by using Eq. (5.21). Let F, G ∈ B and let F̃, G̃ be representa-
tives in A of F and G; both F̃ and G̃ are determined uniquely up to an element of I .
According to condition (1), F̃ and G̃ belong to N (I ), so their bracket F̃, G̃ is,
modulo I , independent of the chosen representatives F̃ andG̃ for F and G. Con-
dition (2) implies that F̃, G̃ ∈ κ −1 (B), so that κ F̃, G̃ ∈ B. It follows that
(5.21) defines a map {· , ·}B : B × B → B. Since κ is an algebra homomorphism,
5.2 Poisson Reduction 125

the fact that {· , ·} is a skew-symmetric biderivation of A implies that {· , ·}B is

a skew-symmetric biderivation of B. For the Jacobi identity, observe that, for all
F, G, H ∈ B, (5.21) yields that

{{F, G}B , H}B = κ F̃, G̃ , H̃

where F̃, G̃ and H̃ are arbitrary representatives of F, G and H. The Jacobi identity for
{· , ·}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 Pois-
son bracket {· , ·}B which satisfies (5.19). Since F̃ and G̃ are arbitrary
 elements of
κ −1 (B), the latter formula implies that κ κ −1 (B), κ −1 (B) ⊂ B, which yields
−1
 F̃ ∈ κ (B) and
condition (2). For arbitrary elements G̃ ∈ I we have from (5.19),
since κ (G̃) = 0, that κ F̃, G̃ = 0, so that κ −1 (B), I ⊂ I , which yields
condition (1).
Notice that since κ −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:
κ −1 (B) is a Poisson subalgebra of N (I ) .
Moreover, viewing the canonical map κ as a map κ : κ −1 (B) → B, the for-
mula (5.21) for the reduced Poisson bracket can be simply stated as κ being a
morphism of Poisson algebras, where κ −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 homomor-
phisms 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

{F, G}P (p(n)) = F̃, G̃ (n) ,

for all F, G ∈ F (P) and for all extensions F̃, G̃ ∈ F (M) of F ◦ p and G ◦ p. The
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, κ −1 (B) = κ −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) All Hamiltonian vector fields, which are associated to functions, whose re-
striction to N is of the form H ◦ p, for some function H ∈ F (P), are tangent
to N;
(2) For every two functions F, G ∈ F (P) and for all extensions F̃, G̃ ∈ F (M) of
F ◦ p and G◦ p to M, there exists a function H ∈ F (P), such that F̃, G̃ (n) =
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

{F, G}P (p(n)) = F̃, G̃ (n) ,

where F̃ and G̃ 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 F̃ ∈ F (M) the following conditions are equivalent:
5.2 Poisson Reduction 127

(i) The restriction of F̃ to N is of the form H ◦ p, for some H ∈ F (P);

(ii) The Hamiltonian vector field XF̃ 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
satisfied. For condition (2), let F, G ∈ F (P) and let F̃, G̃ ∈ F (M) be extensions of
F ◦ p and G ◦ p. According to (i) ⇒ (ii), XF̃ [IN ] ⊂ IN and XG̃ [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
   
X{F̃,G̃} [IN ] ⊂ XF̃ G̃, IN + XG̃ F̃, IN ⊂ XF̃ [IN ] + XG̃ [IN ] ⊂ IN ,

which implies, in view of (ii) ⇒ (i), that the restriction of F̃, G̃ to N is of the
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 defi-
nition 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 struc-
ture πP = {· , ·}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

{F, G}P (p(n)) = F̃, G̃ (n) , (5.25)

128 5 Reduction

at all points n of V , where F̃, G̃ ∈ F (U) are arbitrary extensions of F ◦ p|V and
G ◦ p|V . The Poisson structure πP 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
F, G ∈ F (p(U ∩ N)), one has (5.25), for all n ∈ U ∩ N, where F̃, G̃ ∈ F (U) are
arbitrary extensions of F ◦ p|U∩N and G ◦ p|U∩N .

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) For every function F̃ ∈ F (U), whose restriction to V is constant on the fibers
of p, the Hamiltonian vector field XF̃ is tangent to N at every point of V ;
(2) For every pair of functions F̃, G̃ ∈ F (U) whose restrictions to V are constant
on the fibers of p, their Poisson bracket F̃, G̃ , restricted to V , is constant on
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

{F, G}P (p(n)) = F̃, G̃ (n) , (5.26)

where F̃ and G̃ are arbitrary extensions to a neighborhood of n in M, of F ◦ p|V

and G ◦ p|V . In particular, the Hamiltonian vector field associated to F is given,
at p(n), by
(XF ) p(n) = dn p (XF̃ )n . (5.27)

Proof. We first point out that for an open subset V ⊂ N and a function F ∈ F (V )
the following conditions are equivalent:
(i) F is of the form H ◦ p for some function H ∈ F (p(V ));
(ii) F is constant on all the fibers of p.
The only thing which is not obvious in this equivalence is the fact that the func-
tion 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 sub-
sets, with V ⊂ U ∩ N, and suppose that F̃ ∈ F (U) is constant on the fibers of p,
when restricted to V . Then F̃|V = F ◦ p|V for some function F ∈ F (p(V )). For an
arbitrary n ∈ V and an arbitrary function G̃, defined on a neighborhood U  of n
in M, and vanishing on some neighborhood of n in N, we have in view of (5.25) that
F̃, G̃ (n) = {F, G}P (p(n)) = 0, since G = 0, in a neighborhood of p(n). There-
fore, XF̃ is tangent to N at n, leading to condition (1). Condition (2) follows also
5.2 Poisson Reduction 129

from (5.25), because the left-hand 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 skew-
symmetric biderivation (πP )w at w by: for all germs Fw , Gw at w,

(πP )w [Fw , Gw ] := πn [F̃n , G̃n ] = F̃, G̃ (n) , (5.28)

where n is an arbitrary point of N for which p(n) = w and F̃n , G̃n are germs at n
of arbitrary functions F̃, G̃, defined on a neighborhood of n in M, and satisfying
F̃|V = F ◦ p|V and G̃|V = G ◦ p|V for some neighborhood V of n in N. The right-hand
side in (5.28) is independent of the choice of n (with p(n) = w) because, according
to (2), the value of F̃, G̃ at points n of V depends on p(n) only. Similarly, it
is independent of the chosen F̃ and G̃ because, according to (1), if G̃ is such that
G̃|V = 0 and F̃ is such that F̃|V = F ◦ p|V , then F̃, G̃ (n) = 0. Thus πP is well-
defined at every point w ∈ P and it inherits from π the biderivation property at
each point. We need to show that the bivector field πP , defined on P by w → (πP )w
is smooth/holomorphic. Let w ∈ P, let n ∈ p−1 (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
{F, G}P ◦ p = F̃, G̃ | = H̃ ◦ p ,
V

for some function H̃ ∈ F (p(V )), in particular {F, G}P ∈ F (p(V )). This shows that
πP is a smooth/holomorphic bivector field in a neighborhood of w; since w was
arbitrary, πP 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

{{F, G}P , H}P (p(n)) = F̃, G̃ , H̃ (n) , (5.29)

for arbitrary functions F, G, H, defined in a neighborhood of p(n), and F̃, G̃, H̃ ex-
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 F̃ is

a function, defined on an open subset of M, whose restriction to an open sub-
set V of N vanishes, then the Hamiltonian vector field XF̃ 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, π ).
Definition 5.12. A submanifold N of a Poisson manifold (M, π ) is said to be a
coisotropic submanifold of M if for every open subset V of N and for every func-
130 5 Reduction

tion F̃, defined on a neighborhood of V in M, and such that F̃|N∩V = 0, the Hamilto-
nian vector field XF̃ is tangent to N, at every point of V .
The notion of a coisotropic submanifold generalizes the notion of a Poisson sub-
manifold, 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 πm :
Tm∗ M → Tm M, which corresponds3 to the bivector πm ∈ ∧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 := {ξ ∈ Tn∗ M | ∀v ∈ Tn N, ξ (v) = 0} .

By the implicit function theorem, every element of Tn⊥ N can be realized as the
differential at n of some function F̃, defined on a neighborhood of n in M, and whose
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, π ). The following
conditions on N are equivalent.
(i) N is a coisotropic submanifold;
(ii) For every n ∈ N, πn (Tn⊥ N, Tn⊥ N) = {0};
(iii) For every n ∈ N, πn (Tn⊥ N) ⊂ Tn N.
Similarly, the stronger condition that all Hamiltonian vector fields XF̃ are tangent to
the fibers pn := p−1 (p(n)) of the surjective map p : N → P can be stated as follows:
for every n ∈ N,
πn (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, π ) 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
πn 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.

3Our sign convention is that πm (dm H)[F] = πm [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 con-
ditions (1) and (2) in Proposition 5.11.
(1) The skew-symmetry of πn , combined with (5.30), yields that πn (Tn⊥ pn ) ⊂ Tn N,
for every n ∈ N. It implies that the Hamiltonian vector field of every function, de-
fined 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).
(2) Let U ⊂ M and V ⊂ N be non-empty open subsets, with V ⊂ U ∩ N. Let F̃, G̃ ∈
F (U) be two functions whose restrictions to V are constant on the fibers of p. We
show that the restriction of F̃, G̃ to V is constant on the fibers of p. Let n ∈ V
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 H̃ on a neighborhood U  ⊂ U of n in M, whose restriction
to V  := V ∩ U  is zero, with (XH̃ )n = −πn (dn H̃) = v. Since H̃ is zero on V  , its
Hamiltonian vector field XH̃ is indeed, again by (5.30), tangent to the fibers of p at
every point of V  . It follows that the function XH̃ [F̃] is zero on V  , because F̃|V is
constant on the fibers of p. By (5.30), XH̃ [F̃], G̃ vanishes on N (recall that G̃ is
also constant on the fibers of p). Similarly, XH̃ [G̃], F̃ (n) = 0. Using the Jacobi
identity, it follows that
   
v F̃, G̃ = XH̃ F̃, G̃ (n) = XH̃ [F̃], G̃ (n) + F̃, XH̃ [G̃] (n) = 0 .

Since n ∈ N and v ∈ Tn pn are arbitrary, the restriction of F̃, G̃ to each connected

component of pn ∩ V is constant. Notice that, since pn ∩ V is not necessarily con-
nected, we cannot conclude from it that the restriction of F̃, G̃ to every fiber of p
is constant. However, there is a function H on p−1 (p(V )), which agrees with F̃, G̃
at every point of V and which has the same property of being constant, when re-
stricted to each fiber pn , where n ∈ V ; since each fiber pn is connected, we can
conclude that F̃, G̃ is constant, when restricted to the fibers of p. In order to con-
struct H, let n be an arbitrary point of p−1 (p(V )). Since F̃ is constant on the fibers
of p, its restriction to V is of the form F ◦ p|V for some function F. On a small neigh-
borhood U  of n in M we can construct an extension F̃0 of F ◦ p|V  , where V  is a
neighborhood of n in N, with V  ⊂ U  ∩ N. Similarly, we construct G̃0 from G̃. It
follows from (5.30) that H := F̃0 , G̃0 is at points of V  independent of the chosen
extensions, hence leads to a well-defined function on p−1 (p(V )), which coincides
with F̃, G̃ 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, π ).

Assume that the dimension of πn (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 sub-
space πn (Tn⊥ N) of Tn M. This distribution is differentiable, because it is locally de-
fined 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, π ).

Remark 5.16. It is instructive to write down the conditions of Poisson reducibility

in terms of a well-chosen 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 subman-
ifold. 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 x̃i
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 (x̃1 , . . . , x̃s , 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 π be a Poisson structure on M. The Poisson matrix X of π with respect to the
coordinates x̃1 , . . . , x̃s , y1 , . . . , yt , z1 , . . . , zu (in that order) is given by the block matrix
⎛ ⎞
à B C
X = ⎝ −B D E ⎠
−C −E  F

where Ã, D and F are square matrices of size s, t and u respectively. Each of the
blocks Ã, B, C, . . . is a matrix-valued function on U, whose restriction to U ∩ N
is denoted by Ã|N , B|N , C|N , . . . 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 Hamil-
tonian 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 F|N = 0.
Let us explain the second line in the table. Condition (1) in Proposition 5.11
demands that the Hamiltonian vector field XG̃ is tangent to N for every function
G̃ 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 F|N = 0

Condition (1) in Proposition 5.11 C|N = F|N = 0

Condition (2) in Proposition 5.11 Ã|N depends only on x̃1 , . . . , x̃s

C|N = F|N = 0
(M, N, P) is Poisson reducible
Ã|N depends only on x̃1 , . . . , x̃s
C|N = F|N = 0
Condition (5.30) in Proposition 5.14
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 x̃1 , . . . , x̃s , z1 , . . . , zu are tangent to N, hence that C|N = F|N = 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 x̃i , x̃ j , restricted to U ∩ N depend on the functions x̃1 , . . . , x̃s only;
in terms of the Poisson matrix X, this condition amounts to saying that Ã|N depends
only on x̃1 , . . . , x̃s .
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 πn Tn⊥ N = Tn pn for every n ∈ N, implies that condition (1) in Proposition 5.11
is satisfied, so that C|N = F|N = 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, Ã|N depends only on the variables x̃1 , . . . , x̃s ,
so that there exists a matrix-valued function A, such that A ◦ p = Ã|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 Poisson–Dirac 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 Poisson–Dirac Reduction

Let (A , ·, {· , ·}) be a Poisson algebra and let I be an ideal of (A , ·). Consider the
composition of algebra homomorphisms

κ A
N (I ) A (5.31)
I

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 Poisson–Dirac ideal if one of the following equivalent conditions holds:
(i) The natural inclusion of algebras N (I )/(N (I ) ∩ I ) → A /I is an iso-
morphism;
(ii) The composition of algebra homomorphisms N (I ) → A → A /I is sur-
jective;
(iii) 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 Poisson–Dirac 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 Poisson–Dirac reduction.
When I is a Poisson–Dirac ideal, the different Poisson algebras which appear in
Poisson–Dirac reduction can be summarized in the following commutative diagram,
5.3 Poisson–Dirac Reduction 135

N (I ) A

κ
(5.32)

N (I )

A /I
N (I ) ∩ I

where all maps are morphisms of Poisson algebras, except (in general) for the
canonical projection κ : A → A /I .
We now consider Poisson–Dirac 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 func-
tions on M whose Hamiltonian vector field is tangent to N. Translating (ii) of Defi-
nition 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 Poisson–Dirac 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 Poisson–Dirac subvariety if and only if its
ideal IN is a Poisson–Dirac ideal. In view of Definition 5.17 this leads to other, al-
gebraic, characterizations of Poisson–Dirac subvarieties and Proposition 5.18 shows
that a Poisson–Dirac subvariety inherits a Poisson structure from its ambient Pois-
son variety. That is the content of the following proposition.
Proposition 5.20. Let N be a Poisson–Dirac 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

{F, G}N = F̃, G̃ N (5.33)

where F̃, G̃ are arbitrary extensions of F, G to M, whose Hamiltonian vector fields

XF̃ and XG̃ 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 Poisson–Dirac subvariety of an affine Poisson

variety (M, {· , ·}). Then the subalgebra N (IN ) of F (M) may not be finitely gen-
erated, 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 M

DN  N

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 Poisson–Dirac subvarieties a sim-
ilar, 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 Poisson–Dirac 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 nor-
malizer 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 Poisson–Dirac reduction. Assume that N is a Poisson–Dirac subvari-
ety 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 Poisson–Dirac subvariety, the functions
F1 , . . . , Fs , restricted to N, generate F (N). Also, the Poisson matrix of (N, {· , ·}N ),
5.3 Poisson–Dirac 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 differential-
geometrical context.

5.3.2 Poisson–Dirac 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 (F̃,U,V ) where
(1) U ⊂ M is a neighborhood of n in M;
(2) V ⊂ (W ∩U) ⊂ N is a neighborhood of n in N;
(3) F̃ ∈ F (U) is a function such that F̃|V = F|V .
The relation between the different subsets and maps is represented in the following
diagram:
N M

W V U
F|
V
F F̃

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 Poisson–Dirac submanifold of M,
if for every n ∈ N, every function F, defined in a neighborhood of n in N, admits a
local extension (F̃,U,V ) at n, such that XF̃ is tangent to N at every point of V .
Note that, in particular, a Poisson submanifold N of a Poisson manifold (M, {· , ·}) is
a Poisson–Dirac submanifold, since all Hamiltonian vector fields are tangent to N.
We show in the following proposition that every Poisson–Dirac submanifold in-
herits 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
that, for all open subsets V ⊂ N and U ⊂ M with V ⊂ U, for every function F̃, G̃ ∈
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
{F, G}N (n) = F̃, G̃ (n) , (5.34)
where F, G ∈ F (V ) are the restrictions of F̃, G̃ to V . In particular, locally, every
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 skew-symmetric pointwise biderivation (πN )n
by its values on germs Fn , Gn at n,

(πN )n (Fn , Gn ) := F̃, G̃ (n) , (5.35)

where the right-hand 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 (F̃,U,V ) and (G̃,U,V ) of F and G at n, such that the
Hamiltonian vector fields XF̃ and XG̃ are tangent to N, at every point of V .
We have to check that this definition makes sense: a priori, the right-hand 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 (F̃,U,V ) and (G̃,U,V ). The Hamiltonian vec-
tor field XG̃ is tangent to N at every point of the subset V of N, so that XG̃ [F̃](n)
depends only on the restriction of F̃ to V , and therefore depends only on the germ Fn
of F at n. Hence
F̃, G̃ (n) = XG̃ [F̃](n) = −XF̃ [G̃](n)
depends only on the germs Fn and Gn , as was to be checked.
We may now define a bivector field πN = {· , ·}N on N by its value on every pair
of functions F, G, defined on an open subset V of N, setting

{F, G}N (n) := (πN )n (Fn , Gn ) = F̃, G̃ (n) , (5.36)

for every n ∈ V . To check that the bivector field is smooth/holomorphic, notice that
the above extensions (F̃,U,V ) and (G̃,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 right-hand 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 πN is also smooth or holomorphic
(on V , hence on N).
Let, as above, F̃, G̃, H̃ be three functions, defined on an open subset U of M,
assume that their Hamiltonian vector fields are tangent to N at every point of an open
subset V of N, and let F, G, H denote their restrictions to V . The Hamiltonian vector
field of F̃, G̃ is tangent to N, at every point of V , since X{F̃,G̃} = − [XF̃ , XG̃ ]

(item (5) in Proposition 1.4). As a consequence, F̃, G̃ ,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 Poisson–Dirac Reduction 139

{{F, G}N , H}N (n) = F̃, G̃ , H̃ (n) , (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 Poisson–Dirac submanifolds

(Proposition 5.25 below), we express the Poisson bivector on a Poisson manifold
in terms of local coordinates which are adapted to the Poisson–Dirac submanifold.
To do this, let N be a Poisson–Dirac submanifold of a Poisson manifold (M, π ), and
let n ∈ N. We denote the dimension of M by d and the dimension of N by s. There ex-
ists a coordinate chart (U, x) of M, adapted to N, and centered at n, such that the vec-
tor 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 Poisson–Dirac 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 prop-
erty, 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 } = π [xi , x j ] = −Xxi [x j ] vanish on V , for 1  i  s < j  d. For
every point n ∈ V , the Poisson structure π , at n , is therefore given by:
       
∂ ∂ ∂ ∂
∑ xi j (n ) ∂ xi  ∧ ∂ x j  + ∑ xi j (n ) ∂ xi  ∧ ∂ x j  (5.38)
 
1i< js n n s<i< jd n n

where xi j (n ) := xi , x j (n ), for 1  i < j  d, and where the first sum is the re-
duced Poisson structure at n. Equation (5.38) implies that, in well-chosen local co-
ordinates, the Poisson matrix of π at points of N takes a block-diagonal form. We
show in the following proposition that this fact characterizes Poisson–Dirac sub-
manifolds.
Proposition 5.25. Let (M, π ) be a Poisson manifold of dimension d and let N be a
submanifold of M. Then N is a Poisson–Dirac submanifold of M if and only if there
exists for each n ∈ N a coordinate chart (U, x) of M, adapted at N, and centered
at n with the following property: if the Poisson matrix X := ( xi , x j )1i, jd of π 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 πN on N is
given, in terms of the above coordinates x1 |V , . . . , xs |V , on a neighborhood of n in N,
by A|V = xi , x j |V
.
1i, js
140 5 Reduction

Proof. We start with the stated characterization of a Poisson–Dirac submanifold. By

the above, every Poisson–Dirac submanifold admits indeed such an adapted coordi-
nate 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 block-diagonal form at every point n in a neigh-
borhood of n in N, as in (5.38). Let F be a function, defined in a neighborhood
of n in N and let F̃ := F ◦ p, where p : U → V is the projection map, associated
to (U, x). Then (F̃,U,V ) is an extension of F at n. Since the coordinate expression
of F̃ on (U, x) depends on the first s variables x1 , . . . , xs only, the Hamiltonian vector
field XF̃ , associated to F̃, takes at points n in a neighborhood of n in N the form
 
 ∂ F̃ ∂
s
(XF̃ )n = ∑ xi j (n ) 
(n ) ,
i=1 ∂xj ∂ xi n

so it is tangent to N at n . According to Definition 5.23, this shows that N is a

Poisson–Dirac submanifold.
We can use these coordinate charts to compute the Poisson matrix of the re-
duced 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
Poisson–Dirac submanifold.
Proposition 5.26. Let (N, πN ) be a Poisson–Dirac submanifold of a Poisson mani-
fold (M, π ) and let n ∈ N. Denoting by Sn (N), respectively Sn (M), the symplectic
leaf of (N, πN ), respectively of (M, π ), 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 γ for πN
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 (F̃,U,V ), whose Hamiltonian vector field XF̃ is equal
to XF on V . Since the image of γ is compact, it can be covered by finitely many such
local extensions, and γ is therefore a piecewise Hamiltonian path for π . 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 Poisson–Dirac Reduction 141

To do this, let (U, x) be local coordinates in a neighborhood of n ∈ N, which

are adapted to the Poisson–Dirac submanifold, as in (5.38). For every function
F̃ ∈ F (U), the Hamiltonian vector field XF̃ of F̃, with respect to {· , ·}, and
the Hamiltonian vector field XF of its restriction F = F̃|V ∈ F (V ), with respect
to {· , ·}N , are related at n ∈ V by
 
 ∂F  ∂
(XF̃ )n = (XF )n + ∑ xi j (n ) (n ) .
s<i< jd ∂ xi ∂ x j n

In particular, the tangent vector (XF̃ )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 non-empty 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 Poisson–Dirac reduction out of the Poisson matrix of the Poisson structure on the
ambient space. According to (5.34), if x̃1 , . . . , x̃s 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
x̃i , x̃ 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, π ) be a Poisson manifold and let N be an s-dimensional
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 π with respect to these coordinates.
142 5 Reduction

We write X in the block form

 
A B
X=
−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) V is a Poisson–Dirac submanifold of (M, π );
(2) 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     
x 1s −C x
= ,
y 0 1t y
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
t
= y j , xi (n ) − ∑ Cik (n ) y j , yk (n )
k=1
= (−B +CD)i j (n ) ,

which is zero, precisely if we choose C := BD−1 . According to Proposition 5.25, V

is a Poisson–Dirac submanifold of M and thePoisson
 matrix of the reduced Poisson
structure on V ⊂ N is given by the matrix xi , xj (n ) , whose entries we
1i, js
compute as above,
 
t
xi , xj (n ) = −Xxi [xj ](n ) = −Xxi [x j ](n ) = xi − ∑ Cik yk , x j (n )
k=1
t
= Ai j (n ) + ∑ Cik (n )B 
k j (n ) = A +CB

(n ) .
k=1 ij
5.3 Poisson–Dirac Reduction 143

Since C = BD−1 , the Poisson matrix of the reduced structure is given on V by A +

BD−1 B .

5.3.3 The Transverse Poisson Structure

Let (M, π ) be a Poisson manifold and let m ∈ M be a point where the rank of π 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 Weinstein’s splitting
theorem (Theorem 1.25), this means that, on U,
r
∂ ∂ ∂ ∂
π=∑ ∧ + ∑ φk (z) ∧ ,
i=1 ∂ q i ∂ p i 1k<s ∂ zk ∂ z

where the functions φk 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 Poisson–Dirac submanifold of M. According to Proposition 5.20, the reduced

Poisson structure πN on N is given by

∂ ∂
πN = ∑ φk (z) ∧
∂ zk ∂ z
,
1k<s

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 Poisson–Dirac submanifold
of M and is isomorphic, as a Poisson manifold, to a neighborhood of m in N.
Theorem 5.28. Let (M, π ) be a Poisson manifold and let m ∈ M. The rank of π at m
is denoted by 2r. Suppose that q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs are splitting coordi-
nates, defined on a neighborhood U of m in M, and centered at m. We denote by N0
the Poisson–Dirac submanifold of M, defined by

N0 := {m ∈ U | q1 (m ) = p1 (m ) = · · · = qr (m ) = pr (m ) = 0} .

Let N1 ⊂ M be an arbitrary s-dimensional submanifold of M, transverse at m to the

symplectic leaf Sm of π 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) V1 is a Poisson–Dirac submanifold of M;

(2) There exists a Poisson diffeomorphism between V0 and V1 , where V0 and V1
are both equipped with their reduced Poisson structure as Poisson–Dirac sub-
manifolds of M.

Proof. Let q1 , . . . , qr , p1 , . . . , pr , z1 , . . . , zs be splitting coordinates, centered at m, de-

fined on a neighborhood U of m in M. On U, the Poisson structure π is given by
r
∂ ∂
π=∑ ∧ + π , (5.41)
i=1 ∂ qi ∂ pi

where π  = ∑1k<s φk (z) ∂∂z ∧ ∂∂z is a Poisson structure on U, whose structure
k 
functions φk 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 s-dimensional submanifold N1 of M, transverse at m
to the symplectic leaf Sm of π , and of dimension s = dim N0 . Since N1 is of codi-
mension 2r in M, N1 is in a neighborhood of m in M given as the intersection of 2r
hypersurfaces
H1 (q, p, z) = · · · = H2r (q, p, z) = 0 . (5.42)
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 π 
vanishes at m,

qi , pj (m) = δi, j , qi , qj (m) = pi , pj (m) = 0 ,

where {· , ·} := π and δi, j stands for the Kronecker delta. As a consequence, the
matrix ⎛     ⎞
r r
qi , qj (m ) qi , pj (m )
⎜ i, j=1 i, j=1 ⎟
⎝   r   r ⎠
  
pi , q j (m )   
pi , p j (m )
i, j=1 i, j=1

is invertible for every m

in a neighborhood of m in M. It proves, in view of Propo-
sition 5.27, that some neighborhood V1 of m in N1 is a Poisson–Dirac submanifold
of M, which is the content of (1).
5.3 Poisson–Dirac 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 one-dimensional). The splitting coordinates which we have con-
structed on U  , plus the natural coordinate on F which we denote by τ , yield coor-
dinates on U  × F. We define on U  × F the following 2r functions

Qi := qi − τ Fi (z1 , . . . , zs ) and Pi := pi − τ Gi (z1 , . . . , zs ) ,

where i = 1, . . . , r. For τ ∈ F, their common zero locus is denoted by Nτ ,

Nτ := m ∈ U  | Qi (m , τ ) = Pi (m , τ ) = 0 for 1  i  r .

Letting τ vary between 0 and 1 allows us to relate the Poisson–Dirac submani-

folds 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 neigh-
borhood 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
Poisson–Dirac 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 = ∂∂τ + 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
r r
H := ∑ A j (τ , z1 , . . . , zs ) Q j + ∑ B j (τ , z1 , . . . , zs ) Pj , (5.43)
j=1 j=1

where the functions A j and B j are solutions to the linear system

⎧ r r
⎪
⎪
⎪
⎨ ∑ A j Q i , Q j + ∑ B j Qi , Pj = Fi ,
j=1 j=1
r r i = 1, . . . , r . (5.44)
⎪
⎪
⎪
⎩ ∑ A j Pi , Q j + ∑ B j Pi , Pj = Gi ,
j=1 j=1

Since

Qi , Q j (m, τ ) = Pi , Pj (m, τ ) = 0 , Qi , Pj (m, τ ) = δi, j ,

for all 1  i, j  r and for all τ ∈ 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. No-
tice 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 τ . Moreover, these functions

vanish at (m, 0) for all τ , 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 Φt of V is
well-defined on U  × {0} for all t ∈ W .
It remains to be shown that the time 1 flow Φ1 of V sends U  × {0} to a neigh-
borhood 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 , τ ) ∈ U  ×W | Qi (m , τ ) = Pi (m , τ ) = 0 for 1  i  r .

The vector field V is tangent to N , since,

∂ Qi 
V [Qi ] (m , τ ) = (m , τ ) + XH [Qi ](m , τ ) = Fi (m ) − Fi (m ) = 0 ,
∂τ
and similarly V [Pi ] (m , τ ) = 0, for all (m , τ ) ∈ 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 pre-
serves 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 s-dimensional submanifolds which are trans-
verse 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 struc-
ture.
We show in the following proposition that all s-dimensional submanifolds which
are transverse to a given symplectic leaf S of M are isomorphic as Poisson mani-
folds in a neighborhood of their intersection point with S , when they are equipped
with their reduced Poisson structure.

Proposition 5.29. Let (M, π ) be a Poisson manifold and let S be a symplectic leaf
of (M, π ). We denote the dimension of S by 2r and we let s := dim M − 2r. Assume
that we are given:
(1) Two points m, m which belong to S ;
(2) Two s-dimensional 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) V and V  are Poisson–Dirac submanifolds of (M, π );
(2) 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, π ). By definition, such points can be joined by a piecewise Hamiltonian
5.3 Poisson–Dirac 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 Φt of XH is defined for t in a
neighborhood of [0, 1] and for all points of U, with in particular Φ1 (m) = m . Since
Φt is the flow of a Hamiltonian vector field, Φ1 is a Poisson diffeomorphism be-
tween a neighborhood U of m and a neighborhood Φ1 (U) of Φ1 (m) = m . Since
Φ1 preserves the Poisson structure, it sends the symplectic leaf S through m to the
symplectic leaf S through Φ1 (m); moreover, since Φ1 is a diffeomorphism, it sends
every submanifold, transverse to S at m, to a submanifold transverse to S at m ,
and Φ1 realizes a Poisson diffeomorphism between them, when both are equipped
with the Poisson structure which they inherit from (M, π ) as Poisson–Dirac subman-
ifolds. 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, π ).

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, π ). Let N

be an arbitrary submanifold of M, transverse to S at some point m ∈ S , and let π 
denote the reduced Poisson structure, which is defined on a neighborhood of m in N.
The germ of π  at m is called the transverse Poisson structure to S .

We have seen two particular situations in which one constructs a (natural) rep-
resentative of the transverse Poisson structure to a symplectic leaf S in a Pois-
son manifold (M, π ). The first one is Weinstein’s splitting theorem (Theorem 1.25),
which allows us to write π in terms of splitting coordinates on a neighborhood of a
point m ∈ S as
r
∂ ∂ ∂ ∂
π=∑ ∧ + ∑ φk (z) ∧ , (5.45)
i=1 ∂ qi ∂ pi 1k<s ∂ zk ∂ z

where the functions φk 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 Weinstein’s 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 Poisson–Dirac submanifold of (M, π ) 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 Poisson–Dirac 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, π ) be an affine Poisson variety or
a Poisson manifold. We say that an action χ : G × M → M preserves the Poisson
structure when, for every g ∈ G, the map χg : M → M, defined for all m ∈ M by
χg (m) := χ (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 χ : 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 {· , ·}G×M at (g, m) ∈ G × M reduces, in view of (2.12), to

{F ◦ χ , G ◦ χ }G×M (g, m) = F ◦ χg , G ◦ χg (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 struc-
ture.
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 struc-

ture on the quotient space M/G and we give conditions on a G-invariant 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 χ : G × M → M be a group action, where either one of the

following is satisfied:
(1) (M, {· , ·}) and (G, {· , ·}G ) are affine Poisson varieties, with G being a reduc-
tive algebraic group;
(2) (M, {· , ·}) and (G, {· , ·}G ) are Poisson manifolds, with G being a Lie group,
and the action is proper and locally free.
If χ 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 G-invariant 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 χ is a Poisson action. To

do this, first notice that a function F ∈ F (M) belongs to F (M)G if and only if
χ ∗ F = p∗2 F where p2 : G × M → M is the projection onto the second component.
Thus, if F, G ∈ F (M)G , then the fact that χ is a Poisson map implies that

χ ∗ {F, G} = {χ ∗ F, χ ∗ G}G×M = {p∗2 F, p∗2 G}G×M = p∗2 {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 G-invariant 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 G-invariant. 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 G-invariant functions, defined on an arbitrary open subset of M,
is G-invariant. 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 G-invariant4 subvariety or

submanifold N of M and while N does not inherit a Poisson structure from M, the
4 Saying that N is G-invariant means that χ (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 Proposi-
tion 5.33, which is proved in exactly the same way as a direct application of Propo-
sition 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 χ : G × M → M be a group action and let N be a G-invariant
subset of M; denote by p : N → N/G the quotient map. Suppose that either one of
the following is satisfied:
(1) (M, {· , ·}) and (G, {· , ·}G ) are affine Poisson varieties, with G being a reduc-
tive algebraic group, and N is an affine subvariety of M;
(2) (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) χ is a Poisson action;
(2) For every function F, whose restriction to N is G-invariant, 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 ∈
p−1 (V ),
F̃, G̃ (n) = {F, G}N/G (p(n)) ,

where F̃ and G̃ are arbitrary extensions of F ◦ p and G ◦ p to an arbitrary open

subset of M.

5.4.3 Fixed Point Sets as Poisson–Dirac 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 Poisson–Dirac 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 χg (m) and gm for the action of g ∈ G on m ∈ M, while the induced
action on F ∈ F (M) is denoted by ψg (F), as in (5.4). We denote the algebra of
invariant functions by F (M)G ,

F (M)G = F ∈ F (M) | ψg (F) = F for all g ∈ G ,

and we denote the fixed point set of χ 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 = Lg−1 (Δ ), where Δ is the diagonal of M × M,
g∈G

Δ := {(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 func-
tions Lg∗ (IΔ ), where IΔ is the ideal of all regular functions which vanish on Δ ; for
i = 1, 2, if we denote by pi : M × M → M the canonical projection on the i-th compo-
nent, then IΔ is the ideal of F (M × M), generated by all functions p∗1 (F) − p∗2 (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 − ψg (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
F (M)G + I (M G ) = F (M) , (5.49)
then I (M G ) is a Poisson–Dirac ideal of F (M).

Proof. Let F ∈ F (M)G be a G-invariant function. For all G ∈ F (M) and all g ∈ G,
we have

F, G − ψg (G) = {F, G} − ψg (F), ψg (G) = {F, G} − ψg ({F, G}) ,

where we used in the last step that every map ψg is a morphism of Poisson algebras.
Since I (M G ) is generated by all functions of the form G − ψg (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 Poisson–Dirac 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 Poisson–Dirac 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) Poisson–Dirac subvariety of (M, {· , ·}). For
F, G ∈ F (M G ), their reduced bracket is given, at n ∈ M G , by
152 5 Reduction

1 1
{F, G}M G (n) =
|G|2 g ∑ ψg1 (F̃), ψg2 (G̃) (n) = ∑ ψg (F̃), G̃ (n)
|G| g∈G
1 ,g2 ∈G

where F̃, G̃ ∈ F (M) are arbitrary extensions of F, G.

Proof. By averaging the action over the group, one constructs a linear projection
ρ : F (M) → F (M)G ; explicitly, ρ (F) is given, for F ∈ F (M), by
1
ρ (F) := ∑ ψg (F) ,
|G| g∈G
(5.50)

where |G| stands for the order of the group G. Since

1
F − ρ (F) = ∑ (F − ψg (F)) ,
|G| g∈G

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 Poisson–Dirac 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) Poisson–Dirac subvariety. According to (5.33), the
Poisson–Dirac bracket of two functions F, G ∈ F (M G ) is given, at n ∈ M G , by
{F, G}M G (n) = F̂, Ĝ (n), where F̂ and Ĝ are arbitrary G-invariant extensions of
F and G to M. Let F̃ and G̃ be arbitrary extensions of F and G to M. Clearly, the
G-invariant functions ρ (F̃) and ρ (G̃) coincide with F and G on M G , so we can take
F̂ := ρ (F̃) and Ĝ := ρ (G̃), to compute

1
{F, G}M G (n) = ρ (F̃), ρ (G̃) (n) =
|G|2 g ∑ ψg1 (F̃), ψg2 (G̃) (n) ,
1 ,g2 ∈G

for all n ∈ M G . Since G̃ − ρ (G̃) ∈ I (M G ), the latter line can also be written as

1
{F, G}M G (n) = ρ (F̃), G̃ (n) = ∑ ψg (F̃), G̃ (n) ,
|G| g∈G

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, π ). If the action preserves the Poisson structure, then the fixed point set M G is
a (not necessarily irreducible) Poisson–Dirac subvariety of (M, π ).
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, π ). 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, π ). The
action is said to be a Hamiltonian action if there exists a Lie algebra homomorphism

μ̃ : g → F (M)
(5.51)
x→ μ̃x

such that the following diagram is commutative:

g X1 (M)

−X (5.52)
μ̃

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 μ̃ a co-
momentum map of the action. The dual map,

μ : M → g∗
(5.53)
m → μ (m) ,

where μ (m) ∈ g∗ is the linear map x → μ̃x (m), is said to be a momentum map of the
action.
We show in the following proposition that, under some assumptions, the orbit space
μ −1 (0)/G inherits a Poisson structure from M.

Proposition 5.39. Let G be a connected Lie group and let (M, π ) be a Poisson man-
ifold. We assume that G acts locally freely and properly on M and that the action is
Hamiltonian. Let μ denote a momentum map of the action. If 0 is a regular value of
μ , then
(1) μ −1 (0) is an embedded submanifold of M, invariant under the action of G;
(2) μ −1 (0)/G is a manifold;
(3) The triple (M, μ −1 (0), μ −1 (0)/G) is Poisson reducible.

The different spaces involved in the theorem are represented in the following dia-
gram:
154 5 Reduction

μ −1 (0) M

p (5.54)

μ −1 (0)/G

Proof. Let us denote N := μ −1 (0). It is an (embedded) submanifold of M since 0 is

a regular value of μ . We claim that, for every x ∈ g, the vector field x is tangent to N.
Indeed, N is precisely the zero locus of the vector space of functions μ̃y | y ∈ g ,

N= m ∈ M | μ̃y (m) = 0 ,
y∈g

and this vector space is invariant for all vector fields x, since,

x [μ̃y ] = −Xμ̃x [μ̃y ] = μ̃x , μ̃y = μ̃[x,y] ,

where we have used in the last step that μ̃ is a Lie algebra homomorphism. As
a corollary, since G is connected, N is G-invariant. 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

πn (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 μ , the space Tn⊥ N is, for
every n ∈ N, given by
Tn⊥ N = {dn μ̃x | x ∈ g} .

Since πn (dn μ̃x ) = − Xμ̃x n = xn , this means that

πn (Tn⊥ N) = {xn | x ∈ g} = Tn pn ,

which proves (5.55), and hence that (M, μ −1 (0), μ −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 funda-
mental vector field x for all x ∈ g, the Lie algebra of G. If M is equipped with a Pois-
son 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, π ). If the action is
Hamiltonian with momentum map μ , then μ −1 (0) is G-invariant and μ −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 struc-
ture on μ −1 (ξ )/Gξ for every regular value ξ ∈ g∗ of μ , where Gξ is the sub-
group of G, defined in terms of the adjoint action Ad : G × g → g, by Gξ :=
g ∈ G | ξ ◦ Adg = ξ . 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:
(i) →
−r is a Poisson structure on G;
(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. Show that r is a Poisson structure on M;
b. Prove that the action of (G, −
→
r −← −
r ) on (M, r) is a Poisson action;
c. 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 Pois-
son bracket. Show that for every m1 ∈ M1 (respectively m2 ∈ M2 ), the submanifold
{m1 } × M2 (respectively M1 × {m2 }) is a Poisson–Dirac submanifold of M1 × M2 .
4. Let M be a Poisson manifold, and let N be a Poisson–Dirac 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 φ : P → M be a submersive Poisson
map. Let N be a submanifold of M.
a. Show that if N is a coisotropic submanifold of M, then φ −1 (N) is a coisotropic
submanifold of P;
b. Show, by giving a counterexample, that if N is a Poisson–Dirac submanifold
of M, then φ −1 (N) need not be a Poisson–Dirac submanifold of P.
156 5 Reduction

6. We assume in this exercise that the reader is familiar with the notion of sym-
plectic manifold (see Section 6.3). Let (M, ω ) be a symplectic manifold, and denote
by π the canonical Poisson structure, associated to ω . Let N be a submanifold of M.
Show that the following are equivalent:
(i) N is a Poisson–Dirac submanifold of (M, π );
(ii) The restriction of ω to N is non-degenerate.
Conclude that for symplectic manifolds the notions of symplectic submanifold and
Poisson–Dirac submanifold coincide.
7. We assume in this exercise that the reader is familiar with the Lie–Poisson struc-
ture on the dual of a Lie algebra (see Chapter 7). Let g be a finite-dimensional
Lie algebra and let g = gξ ⊕ m be a vector space decomposition of g, where
gξ := {x ∈ g | ad∗x ξ = 0} denotes the centralizer of an element ξ ∈ g∗ . We consider
the affine subspace N := ξ + m⊥ of g∗ , which is isomorphic to g∗ξ , via the affine
map χ , which is defined, for all η ∈ N, by χ (η ) := (η − ξ )|g .
ξ

a. Show that, if m satisfies [gξ , m] ⊂ m, then N is a Poisson–Dirac submanifold of

g∗ , equipped with its Lie–Poisson structure, and the reduced Poisson structure
on N is isomorphic to the Lie–Poisson structure on g∗ξ , via χ .
b. As an example, consider g = gld (F), let x ∈ g and let ξ ∈ g∗ be the linear map,
defined by ξ (y) := Trace(xy), for all y ∈ g. Show that if x is diagonalizable, then
there exists a complementary subspace m of gξ in g, such that [gξ , m] ⊂ m.
8. Let (M, {· , ·}) be a Poisson manifold and let N be a submanifold of M. We
endow F (M)[ν ]/ν 2 with the structure of an F[ν ]-algebra, defined by

F ν G := FG + ν {F, G}

for all F, G ∈ F (M) (see Exercise 8 of Chapter 3, where this product is proven to
be associative). Suppose that

Φ : F (M)[ν ]/ν 2 × F (N)[ν ]/ν 2 → F (N)[ν ]/ν 2

defines the structure of a left module over the F[ν ]-algebra (F (M)[ν ]/ν 2 , ν ) on
F (N)[ν ]/ν 2 , satisfying Φ (F, H) = F|N H (mod ν ) for every F ∈ F (M) and for
every H ∈ F (N). Let σ : F (M) × F (N) → F (N) be the bilinear map, given for
all F ∈ F (M) and H ∈ F (N) by

Φ (F, H) − F|N H 
σ (F, H) =  .
ν 
ν =0

a. Show that σ satisfies the identity

{F, G}|N H = σ (F, G|N H) − σ (FG, H) + F|N σ (G, H) (5.56)

for all F, G ∈ F (M) and H ∈ F (N);

b. Deduce from it that N is a coisotropic submanifold of M.
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 Fulton–Harris [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
Noether’s theorem, which is the predecessor of the notion of momentum map and
the theory of Dirac constraints, which eventually led to the notion of Poisson–Dirac
reduction. The modern, geometrical development of these ideas first appears in the
seminal papers by Śniatycki–Tulczyjew [184], Meyer [148] and Marsden–Weinstein
[146]. See Kosmann–Schwarzbach [114], Weinstein [199] and the introduction of
Ortega–Ratiu [160] for more details on the history of the subject; in the latter in-
troduction 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 Poisson–Dirac subman-
ifolds, which are both particular cases of the so-called Marsden–Ratiu reduction
procedure [144]. The notion of momentum map, presented here, has also been gen-
eralized in many different ways, see for example Alekseev–Malkin–Meinrenken [9]
for Lie group valued momentum mappings and Lu [133] for Poisson–Lie group val-
ued 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

Poisson’s original bracket on R2r , given for smooth functions F and G by

r  
∂F ∂G ∂G ∂F
{F, G} := ∑ − , (6.1)
i=1 ∂ qi ∂ pi ∂ qi ∂ pi

is of fundamental importance in classical mechanics, in quantum mechanics and in

other areas of mathematical physics. In fact, locally, the Poisson structure of ev-
ery 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 struc-
ture 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 re-
lated 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 writ-
ten 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. Laurent-Gengoux et al., Poisson Structures, 161

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 6, © Springer-Verlag Berlin Heidelberg 2013
162 6 Constant Poisson Structures, Regular and Symplectic Manifolds

6.1 Constant Poisson Structures

Poisson’s original bracket on R2r , defined by

r  
∂F ∂G ∂G ∂F
{F, G} := ∑ − , (6.2)
i=1 ∂ qi ∂ pi ∂ qi ∂ pi

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 π on a finite-dimensional vector space V is
called a constant Poisson structure, if for each pair of linear functions F and G
on V , their Poisson bracket π [F, G], also denoted by {F, G}, is a constant function
on V .
Let X = (xi j ) ∈ Matd (F) be an arbitrary skew-symmetric 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 finite-dimensional F-vector 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 skew-symmetric biderivation of F (Fd ), defined for all F, G ∈
F (Fd ), by
d
∂F ∂G
{F, G} := ∑ xi j , (6.3)
i, j=1 ∂ xi ∂ x j

 x1 ,. . . , xd are the standard coordinates on F . Since

where d each
  of  the brack-
ets xi , x j = xi j is a constant function, we have that xi , x j , xk = 0, where
1  i, j, k  d, so that the Jacobi identity holds for every triple of coordinate func-
tions (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 struc-
ture on Fd , which we denote by π . Clearly, π is a constant Poisson structure and
every constant Poisson structure on Fd is of this form. Since every d-dimensional
vector space V can be identified with Fd , by choosing a basis of V , the matrix X
defines a Poisson structure on every d-dimensional vector space V , equipped with a
basis (equivalently, with a system of linear coordinates). Thus, the following propo-
sition holds.
Proposition 6.2. Let V be a finite-dimensional vector space, with a fixed system
of linear coordinates (x1 , . . . , xd ). There is a one-to-one correspondence between
skew-symmetric 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)
i, j=1 ∂ xi ∂ x j
6.1 Constant Poisson Structures 163

Example  6.3. Poisson’s  original bracket (6.2) corresponds to the skew-symmetric

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 π with respect to the co-
ordinates x1 , . . . , xd (see Section 1.2.2). Since the entries of X are constant, the rank
of π 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
(so-called Darboux coordinates). We show that in the case of a constant Poisson
bracket on a d-dimensional vector space V , linear coordinates which are global Dar-
boux coordinates can be constructed explicitly. Namely, since X is skew-symmetric,
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 π (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 , . . . , zd−2r ) := (x1 , . . . , xd )A
one obtains the following brackets
     
qi , p j = δi, 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 , . . . , zd−2r are Darboux coordinates for π . 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

∂H ∂H
q̇i = , ṗi = − , żk = 0 ,
∂ pi ∂ qi

where q̇i 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 Pois-
son structure on a finite-dimensional 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
1 In terms of internal products, as recalled in Appendix A, b (ξ ) = ıξ b, which shows that b , as
given by (6.5), is well-defined.
164 6 Constant Poisson Structures, Regular and Symplectic Manifolds
s
b (ξ ) = ∑ (ξ , vi  wi − ξ , wi  vi ) , (6.5)
i=1

for every ξ ∈ V ∗ , and we define a skew-symmetric bilinear form b̃ on V ∗ , by

s
b̃(ξ , η ) := η , b (ξ ) = ∑ (ξ , vi  η , wi  − η , vi  ξ , wi ) , (6.6)
i=1

for all ξ , η ∈ V ∗ , where · , · denotes the canonical pairing between ∗

 V  and V .
Both b and b̃ are skew-symmetric, which means for the former that η , b (ξ ) =

 
− ξ , b (η ) , for all ξ , η ∈ V ∗ . In terms of a basis (e1 , . . . , ed ) of V and the dual ba-
sis (e∗1 , . . . , e∗d ) of V ∗ , the bivector b can be written as b = ∑1i< jd b̃(e∗i , e∗j ) ei ∧ e j ,
so that b can be recovered from b̃ (and hence from b ).
To a bivector b ∈ V ∧ V one naturally associates a bivector field on V , by iden-
tifying 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) := b̃ (dm F, dm G) , (6.7)

where b̃ : V ∗ ×V ∗ → F is the bilinear form, associated to b, as in (6.6). Equivalently,

we can view b̃ under the same identification as a bilinear form b̃m on TmV ∗ , for
any m ∈ V , and the right-hand side of (6.7) can be written as b̃m (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[ξ ] = ξ , v for all linear functions ξ on V .
Geometrically, we think of v[·] as the directional derivative, defined by a vector v.
Equation (6.9) is a coordinate-free 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 finite-dimensional vector space V ,

equipped with a bivector b ∈ V ∧V .

By the above, we have the following abstract version and amplification of Proposi-
tion 6.2.

Proposition 6.5. For every finite-dimensional vector space V , there is a one-to-one

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 π = {· , ·} 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} = ∑ − ,
i=1 ∂ qi ∂ pi ∂ qi ∂ pi

for all F, G ∈ F (V ).

6.2 Regular Poisson Manifolds

In the case of a Poisson manifold (M, π ), 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 π 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, π ) is a
regular Poisson manifold.

Comparing the definitions, it is clear that every constant Poisson structure on a con-
nected 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
the rank of π is constant on U, since it is given by the rank ofthe Poisson matrix
of π with respect to x, whose entries are the Poisson brackets xi , x j ; since M is
connected, it follows that the rank of π takes the same value at every point of M,
i.e., π 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, π ) be a Poisson manifold, which is assumed to be con-
nected. Then (M, π ) is a regular Poisson manifold if and only if π is a constant
Poisson structure.
166 6 Constant Poisson Structures, Regular and Symplectic Manifolds

Proof. Let (M, π ) 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 brack-
ets xi , x j are constant (functions) on U. Since, by assumption, the rank of π 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 , . . . , zd−2r ), centered at m, in which the Poisson struc-
ture takes the canonical form
     
qi , q j = pi , p j = {qi , zk } = {pi , zk } = {zk , z } = 0 , qi , p j = δi, 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 , . . . , zd−2r 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, π )
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 differen-
tiable, 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, π ) are tan-
gent 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 skew-symmetric variant of an inner product space is a symplectic vector space.

Namely, let V be a finite-dimensional F-vector space and let ω ∈ Hom(V ∧ V, F)
be a skew-symmetric bilinear form on V . We say that ω is a symplectic structure
if ω has maximal rank, Rk ω = dimV . Then (V, ω ) is called a symplectic vector
space. If (V, ω ) is a symplectic vector space, then V is even-dimensional, since the
rank of every skew-symmetric bilinear form is even. Since the rank of a symplectic
structure ω is maximal, the linear map ω  , which is by definition ω , viewed3 as
an element of Hom(V,V ∗ ), is invertible. Its inverse (ω  )−1 ∈ Hom(V ∗ ,V ) is skew-
symmetric, hence is of the form −π  for some (unique) bivector π ∈ ∧2V . It fol-
lows that every symplectic vector space (V, ω ) is in a natural way a Poisson vector
space, and hence, according to Proposition 6.5, V is equipped with a constant Pois-
son structure π , called the canonical Poisson structure, associated to ω . Explicitly,
the Poisson bracket of F, G ∈ F (M) is computed at m ∈ M from −π  = (ω  )−1
using (6.7) and (6.6), namely
 
{F, G} (m) = π (dm F, dm G) = dm G, π  (dm F)
 
= − dm G, (ω  )−1 (dm F)
(6.10)
= ω ((ω  )−1 (dm F), (ω  )−1 (dm G))
= ω ((XF )m , (XG )m ) ,

where we used in the last line that (XF )m = −π  (dm F) (see (6.8)). The latter defi-
nition of (XF )m is, in view of the definition of π , equivalent to saying that (XF )m
is the unique element of TmV V such that

ω ((XF )m , ·) = dm F . (6.11)

When ω is viewed as a (constant) differential two-form on V , the formula for the

Poisson bracket is also written as

{F, G} = ω (XF , XG ) , (6.12)

and Eq. (6.11) for the Hamiltonian vector field is shortened to

ω (XF , ·) = dF .

The (constant) rank of π equals the rank of ω  , which is dimV .

The above construction is easily reversed: if π 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 ω  (v) = ω (v, ·), for

v ∈ V.
168 6 Constant Poisson Structures, Regular and Symplectic Manifolds

vector space (V, ω ), where ω ∈ 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 two-form ω on M is called an almost symplectic structure if its restric-
tion to each tangent space Tm M is non-degenerate, i.e., ωm is a symplectic struc-
ture on Tm M, for every m ∈ M, as defined in Section 6.3.1. Then (M, ω ) is called
an almost symplectic manifold. The globalization of (6.11) and (6.10) is naturally
given by
 −1
(XF )m := ωm dm F ∈ Tm M ,
(6.13)
{F, G} (m) := ωm ((XF )m , (XG )m ) ,
for m ∈ M and F, G ∈ F (M). In a compact form, these formulas are usually writ-
ten as
ω (XF , ·) = dF ,
(6.14)
{F, G} = ω (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 ω is closed, dω = 0, in which
case ω is called a symplectic structure and (M, ω ) is called a symplectic manifold.
This equivalence is shown in the following proposition.

Proposition 6.10. Let (M, ω ) be an almost symplectic manifold. Then π , as defined

by (6.14), is a Poisson structure on M if and only if (M, ω ) is a symplectic manifold.
In particular, every symplectic manifold (M, ω ) carries a canonical Poisson struc-
ture π , which makes (M, π ) 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 ω is closed. Notice that one important consequence
of this assumption is that ω is conserved by the flow of every Hamiltonian vector
field. To see this, use Cartan’s formula (3.50) for the Lie derivative LV ,

LV = ıV d + dıV (6.16)

when V is a (locally) Hamiltonian vector field, V = XF , and use that ıXF ω = dF

and ω are closed; it yields LXF ω = 0, for every function F, defined on an open
subset of M. It implies that the following formula,
6.3 Symplectic Manifolds 169

ı[V1 ,V2 ] ω = LV1 ıV2 ω − ıV2 LV1 ω ,

which is a special case of (2) in Proposition 3.11, simplifies for Hamiltonian vector
fields to

ı[XF ,XG ] ω = LXF ıXG ω = dıXF ıXG ω = −d (ω (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 ı[XF ,XG ] ω = −ıX{F,G} ω . Non-degeneracy of ω 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 ω is closed, so that (M, π ) is a symplectic manifold, then π is a Poisson
structure on M.
Assume now that ω is an almost symplectic structure, and that the bivector
field π , defined by (6.13), satisfies the Jacobi identity. Using the explicit for-
mula (3.28) for computing the de Rham differential, we obtain, for a triple of Hamil-
tonian vector fields as above,

dω (XF , XG , XH ) = XF [ω (XG , XH )] + ω (XF , [XG , XH ]) +  (F, G, H)

= 2 ({{F, G} , H} + {{G, H} , F} + {{H, F} , G})
= 0.

We have used that

XF [ω (XG , XH )] = {{G, H} , F} = ω (XF , [XG , XH ]) ,

as follows from (6.14), (6.15) and item (5) in Proposition 1.4. Since ωm is non-
degenerate for every m ∈ M, we have that Tm M = Hamm (π ) for every m ∈ M; the
above computation therefore shows that dω vanishes on every triple of vector fields,
i.e., that ω is closed, hence that (M, ω ) is a symplectic manifold.

Proposition 6.11. Suppose that (M, π ) is a regular Poisson manifold, whose rank is
equal to the dimension of M. There exists on M a (unique) symplectic structure ω ,
whose canonical Poisson structure is π .

Proof. We consider for m ∈ M the linear map πm : T ∗ M → T M, associated to π . By

assumption, πm is of maximal rank, hence πm is an isomorphism for every m ∈ M.
It follows that we can use πm to associate to each differential one-form on M, a vec-
tor field on M: given η ∈ Ω 1 (M), the corresponding vector field V is defined by its
value at m ∈ M as Vm := πm (ηm ). One often uses the compact4 notation V := π  (η ).
4 Some authors, such as [125] go one step further and write  η for π  (η ).
170 6 Constant Poisson Structures, Regular and Symplectic Manifolds

In particular, for local functions F and G on M, one has π  (FdG) = −FXG . Simi-
larly, one uses the inverse of πm to associate to each vector field on M a differential
one-form on M; in the compact notation, the differential one-form associated to
 −1
V ∈ X1 (M) is π  (V ). We use πm to define a non-degenerate differential two-
 −1
form ω on M: for vector fields V1 and V2 , let ω (V1 , V2 ) := − π  (V1 ), V2 .
Then
  
−1
ω (XF , XG ) := − π  (XF ), XG = dF, XG  = {F, G} ,

for F, G ∈ F (M). It follows that (M, ω ) is an almost symplectic manifold whose

associated bivector field is π . Since π is a Poisson structure, Proposition 6.10 implies
that (M, ω ) is actually a symplectic manifold.
The main properties of symplectic manifolds are listed in the following proposition.

Proposition 6.12. Let (M, ω ) be a symplectic manifold of dimension d and let π

denote its canonical Poisson structure.
(1) The dimension d of M is even;
(2) M is an orientable manifold;
(3) (M, π ) is a unimodular Poisson manifold;
(4) For every k ∈ N∗ , with 0  k  d, the de Rham cohomology space HdR
k (M), the
k
Poisson cohomology space Hπ (M) and the Poisson homology space Hd−k π (M)

are isomorphic,
k π
HdR (M) Hπk (M) Hd−k (M) .
Proof. Since every symplectic vector space is even-dimensional, every tangent
space of a symplectic manifold is even-dimensional, hence also the symplectic man-
ifold itself. In order to prove (2), we assume that M is a real manifold (complex
manifolds are always orientable). Since ω is non-degenerate (at every point) the
differential d-form Λ := ω d/2 is nowhere vanishing, hence it is a volume form and
M is orientable. The volume form Λ is called the Liouville volume form of (M, ω ).
It is preserved by the flow of every Hamiltonian vector field XF , since
d/2
LXF Λ = LXF ω d/2 = ∑ ω i−1 ∧ LXF ω ∧ ω d/2−i = 0 ,
i=1

where we have used in the last step that LXF ω = 0 (see the proof of Proposi-
tion 6.10). Thus, (M, π ) is unimodular; see Section 4.4.4, where it is shown that
for a unimodular Poisson manifold, the Poisson cohomology spaces Hπk (M) and the
Poisson homology spaces Hd−k π (M) (with 0  k  d) are isomorphic vector spaces,

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 isomor-
phisms πm : 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 πm is just ωm ; for a local section V of T M
 −1
(i.e., vector field), the inverse of π  is given by π  (V ) = ω (V , ·). The natural
•
extension of π to Ω (M) yields, for every k with 0  k  d, a map


∧k π  : Ω k (M) → ∧k X1 (M) → Xk (M) ,

which renders the following diagram commutative:

d
Ω k (M) Ω k+1 (M)

∧k π  ∧k+1 π 
−δπk
Xk (M) Xk+1 (M)

Indeed, for local functions F, G1 , . . . , Gk we have on the one hand that

 
∧k+1 π  d(F dG1 ∧ · · · ∧ dGk ) = π  (dF) ∧ π  (dG1 ) ∧ · · · ∧ π  (dGk )
= (−1)k+1 XF ∧ XG1 ∧ · · · ∧ XGk ,

while on the other hand

 
δπk ∧k π  (F dG1 ∧ · · · ∧ dGk ) = δπk (F XG1 ∧ · · · ∧ XGk )
= (−1)k XF ∧ XG1 ∧ · · · ∧ XGk ,

where we have used in the last step that XG1 , . . . , XGk are Poisson 1-cocycles. The
commutativity of the diagram implies that the Poisson complex (X• (M), [· , ·]S ) and
the de Rham complex (Ω • (M), d) are isomorphic. In particular, they have the same
cohomology.

Remark 6.13. A map Ψ : M1 → M2 between two symplectic manifolds (M1 , ω1 )

and (M2 , ω2 ) is called a symplectic map if Ψ ∗ ω2 = ω1 . Considering the Liouville
volume form on M1 , it is clear that dim M1  dim M2 and that Ψ is an immersion. In
particular, for every m ∈ M1 ,

RkΨ (m) π2 = dim M2  dim M1 = Rkm π1 .

The resulting inequality RkΨ (m) π2  Rkm π1 , is in sharp contrast with the inequality
RkΨ (m) π2  Rkm π1 , which is valid for a Poisson map Ψ (see Exercise 4 of Chap-
ter 1). Thus, a symplectic map Ψ : M1 → M2 between two symplectic manifolds
(M1 , ω1 ) and (M2 , ω2 ) is not necessarily a Poisson map, as a map Ψ : (M1 , π1 ) →
(M2 , π2 ), where πi is the canonical Poisson structure, associated to ωi , 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, ω ) be a symplectic manifold, N an (immersed or em-
bedded) submanifold of M with inclusion map ı : N → M, and let p : N → P be a
surjective submersion with connected fibers. If
 
πn 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
ωP on P. Also, ı∗ ω = p∗ ωP .

Proof. In view of Proposition 5.14, we know that (M, N, P) is Poisson reducible,

hence that P inherits a Poisson structure πP from (M, ω ). In order to show that πP
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 F̃ be an extension of F ◦ p to a neighborhood of n in M. According
to (5.27), dn p (XF̃ )n = 0, so that
 
(XF̃ )n ∈ Tn pn = πn Tn⊥ N .

Since M is a symplectic manifold, this means that dn F̃ ∈ Tn⊥ N, i.e., dn F̃|Tn N = 0,

so that dn (F ◦ p) = dn (F̃|N ) = 0. Since dn p is surjective (as p is a submersion), it
follows that d p(n) F = 0. This proves that πP is of maximal rank at p(n). Since p is
surjective, this shows that πP 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 ωP .
We show that ı∗ ω = p∗ ωP . 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 πn is invertible, (6.18) im-
 
plies that πn Tn⊥ pn = Tn N, so that we can write v as (XF̃ )n , where F̃ 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 F̃|N = F ◦ p, where F is a function, defined on a neighborhood of p(n)
in P. On the one hand,
   
(ı∗ ω )n (v, w) = ωn ((XF̃ )n , w) = dn F̃, w = p∗ d p(n) F, w , (6.19)

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 173

(p∗ ωP )n (v, w) = (ωP ) p(n) (Tn p(XF̃ )n , Tn p(w))

  (6.20)
= (ωP ) p(n) ((XF ) p(n) , Tn p(w)) = d p(n) F, Tn p(w) .

Comparing (6.19) and (6.20) we see that (ı∗ ω )n = (p∗ ωP )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 well-chosen coordinates on an open subset U of M, centered at a point n ∈ N,
the Poisson matrix of (M, π ) is given by
⎛ ⎞
à B C
X = ⎝ −B D E ⎠
−C −E  F

and that the condition (6.18) implies that C|N = F|N = 0 (see the last line of Ta-
ble 5.1). Recall also that the Poisson matrix of the reduced Poisson structure πP is
given by the matrix A, defined by Ã|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 π is of maximal rank. For every n ∈ N ∩U,

det X(n ) = (det E(n ))2 det Ã(n ) = (det E(n ))2 det A(p(n )) ,

so that det A(p(n )) = 0. This gives an alternative proof of the fact that A, and
hence πP , 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, ω ) 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 μ , then
(1) μ −1 (0) is an embedded submanifold of M, invariant under the action of G;
(2) μ −1 (0)/G is a manifold;
(3) The triple (M, μ −1 (0), μ −1 (0)/G) is Poisson reducible.
The reduced Poisson structure on μ −1 (0)/G is regular and of maximal rank, i.e.,
it is the canonical Poisson structure, associated to a symplectic structure ω0 on
μ −1 (0)/G. Also, ı∗ ω = p∗ ω0 , where p : μ −1 (0) → μ −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, ω ) be a symplectic C-vector 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 ω . Recall from Section 6.3.1 that the symplectic two-form ω
leads to a (constant) Poisson structure of rank 2d on V , which we called the canon-
ical Poisson of (V, ω ). 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 Sec-
tion 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 G-invariant 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, ω ) preserves the canoni-
cal Poisson structure of (V, ω ), the finite subgroup G ⊂ SP(V ) acts by Poisson iso-
morphisms. 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, ω ) be a symplectic C-vector 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 proposi-
tion implies that V /G is as symplectic as it can be, i.e., {· , ·}V /G is of maximal rank
at every non-singular point of V /G.
Proposition 6.18. Let (V, ω ) be a symplectic C-vector 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 (non-
empty) Zariski open subset of all x ∈ V for which the stabilizer Gx is trivial.
(1) A point in V /G is a non-singular point if and only if it belongs to p(U);
(2) The rank of {· , ·}V /G is equal to 2d at every non-singular 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 holo-
morphic two-form 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 non-singular 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 non-singular if and only if
Gx = {e}. This proves (1).
The open subset U ⊂ V is a G-invariant subset, the action of G on U is free
since the stabilizer of every point of U is by definition trivial and, according to Sec-
tion 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 struc-
ture 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 λ ∈ C∗ , the map x → λ −1 x leads to a regular map of V /G
to itself. Since x → λ −1 x multiplies the canonical Poisson structure on V by λ 2 , this
induced automorphism also transforms the quotient Poisson structure {· , ·}V /G into
λ 2 {· , ·}V /G . As a consequence, for every non-zero μ ∈ C∗ , the Poisson varieties
(V /G, {· , ·}V /G ) and (V /G, μ {· , ·}V /G ) are isomorphic as Poisson varieties.
The case where the dimension of V is 2 will be studied in more detail in Sec-
tion 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 C-vector 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 sub-
group G of SL(V ) is of finite order, hence is diagonalizable. Clearly, a diagonal-
izable 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 sym-
plectic 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 Kähler 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 lin-
ear coordinates on the fibers of ρ : T ∗ M → M yield canonical coordinates on phase
space. To make this precise, let us first show how to build natural local coordi-
nates 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 ◦ ρ . For the other half, let pi denote the function on T ∗U, which is defined
by    
∂
pi (ξm ) := ξm , , (6.21)
∂ xi m
where ξm ∈ T ∗U is in the fiber over m, i.e., ρ (ξm ) = 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 , = δi, j .
∂ xi m

On T ∗ M there is a natural differential one-form, called the Liouville form, which

can in terms of the above local coordinates be defined as
d
θ := ∑ pi dqi .
i=1

More intrinsically, θ ∈ Ω 1 (T ∗ M) is defined by

  
θ , V  (ξm ) := ξm , Tξm ρ Vξm , (6.22)

where ξm ∈ T ∗ M and V ∈ X1 (T ∗ M). The canonical pairing in this formula is be-

tween Tm∗ M and Tm M: indeed, Tξm ρ : Tξm (T ∗ M) → Tm M. To check that θ , as defined
by (6.22), is given in terms of the above coordinates qi , pi by ∑di=1 pi dqi , take an ar-
bitrary vector field V = ∑di=1 (βi ∂ /∂ qi + γi ∂ /∂ pi ) on T ∗ M and compare
   
   d
∂
θ , V  (ξm ) = ξm , Tξm ρ Vξm = ξm , ∑ βi (ξm ) ,
i=1 ∂ xi m

which was computed from (6.22), using qi = xi ◦ ρ , with

   
d 
d d
∂ ∂
∑ pi dqi , V (ξm ) = ∑ pi dqi , ∑ β j ∂ q j + γ j ∂ p j (ξm )
i=1 i=1 j=1
d
= ∑ p i ( ξm ) βi ( ξm )
i=1
6.3 Symplectic Manifolds 177
   
d
∂
= ∑ i m m ∂ xi
β ( ξ ) ξ , ,
i=1 m

which was computed from (6.21).

The canonical symplectic structure on T ∗ M is by definition the closed differential
two-form ω := dθ , which is locally given by ∑di=1 dpi ∧dqi , so that the coordinates qi
and pi are canonical coordinates. Notice that the differential two-form ω 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
associate a function F̃, respectively V˜ , on T ∗ M by

F̃ := F ◦ ρ , V˜ (ξm ) := ξm , V (m)

for every ξm ∈ Tm∗ M. The reader will easily verify that the canonical Poisson bracket
of ω satisfies the following formulas: if F1 , F2 ∈ F (M) and V1 , V2 ∈ X1 (M), then
     
F̃1 , F̃2 = 0, 
V˜1 , V˜2 = [V1 , V2 ] and

V˜1 , F̃1 = V1 [F1 ] .

In fact, these formulas can be used to define ω (see [3, Ch. 2]).

6.3.6 Example 2: Kähler 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 non-degenerate differential 2-form on MR , given by
√
−1
ω :=
2 1i,∑ hi j dzi ∧ dz j .
jd

One says that the metric ds2 is a Kähler metric if ω is closed, i.e., if ω is a symplectic
structure. Then (MR , ω ) is called a Kähler manifold and ω is called the associated
differential 2-form of the metric. Notice that the metric can be recovered from its
associated differential 2-form.
Two facts underlie the importance of Kähler manifolds. The first one is that the
complex projective space PN admits a Kähler metric. The second fact is that every
178 6 Constant Poisson Structures, Regular and Symplectic Manifolds

submanifold of a Kähler manifold is itself a Kähler manifold. This means that every
non-singular complex projective variety has a Kähler structure.

6.4 Notes

The study of symplectic manifolds, initiated by its importance in classical mechan-

ics, 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 con-
nected 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 Kähler manifolds is treated in detail. For an introduction to the
subject, largely motivated by classical mechanics, see Arnold’s 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 Libermann–Marle and [92]
by Guillemin–Sternberg, 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 Ortega–Ratiu [160] should also be consulted. Symplectic vector bundles and
contact manifolds are treated in Libermann–Marle [125], symplectic homogeneous
spaces are an additional topic in Guillemin–Sternberg [92], the Hamilton–Jacobi
method and perturbation theory are discussed in both Abraham–Marsden [1] and
Arnold [15].
Chapter 7
Linear Poisson Structures and Lie Algebras

Together with symplectic manifolds, considered in the previous chapter, Lie alge-
bras provide the first examples of Poisson manifolds. Namely, the dual g∗ of a finite-
dimensional Lie algebra g admits a natural Poisson structure, called its Lie–Poisson
structure, which provides new insights and technical tools in the study of Lie al-
gebras. For example, the coadjoint orbits in g∗ are precisely the symplectic leaves
of the Lie–Poisson structure, showing in particular that coadjoint orbits are even-
dimensional.
Lie–Poisson 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 finite-dimensional vector space) is a Lie–Poisson 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 non-
degenerate Ad-invariant 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 so-called
Lax form.
The Lie–Poisson 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 Lie–Poisson 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 in-
troduction 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. Laurent-Gengoux et al., Poisson Structures, 179

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 7, © Springer-Verlag Berlin Heidelberg 2013
180 7 Linear Poisson Structures and Lie Algebras

7.1 The Lie–Poisson Structure on g∗

Suppose that g is a finite-dimensional 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
ξ → ξ , e := ξ (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 := e∗i . We use it to
build the skew-symmetric 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 skew-symmetric bideriva-
tion {· , ·} of F[x1 , . . . , xd ], defined by setting, for F, G ∈ F[x1 , . . . , xd ],
d
∂F ∂G
{F, G} := ∑ xi j
∂ xi ∂ x j
. (7.1)
i, j=1
 
It is the unique skew-symmetric biderivation of F[x1 , . . . , xd ] such that e∗i , e∗j =
[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 ξ is a linear map dξ F : Tξ g∗ → F,
and since Tξ g∗ is naturally isomorphic with g∗ , we can think of dξ 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 finite-dimensional; 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 Lie–Poisson Structure on g∗ 181

is given, for e ∈ g, by e → e∗ ). Thus, we can identify dξ F with an element of g,

which we will also denote by dξ F. With this notation, the differential of F ∈ F (g∗ )
at ξ ∈ g∗ , is given by
d
∂F
dξ F = ∑ ( ξ ) ei , (7.3)
∂
i=1 xi

and (7.1), evaluated at ξ ∈ g∗ , can be written in the compact, coordinate-free form,

 
{F, G} (ξ ) = ξ , [dξ F, dξ G] . (7.4)

Definition 7.1. Let (g, [· , ·]) be a finite-dimensional Lie algebra and let F (g∗ ) de-
note the algebra of polynomial, smooth or holomorphic functions on g∗ . The Poisson
structure on g∗ , defined for F, G ∈ F (g∗ ), at ξ ∈ g∗ by
 
{F, G} (ξ ) := ξ , [dξ F, dξ G] (7.5)

is called the Lie–Poisson structure on g∗ , or the canonical Poisson structure on g∗ .

As we will show in Proposition 7.3 below, Lie–Poisson structures are in one-to-
one correspondence with linear Poisson structures, where the latter are defined as
follows.
Definition 7.2. A Poisson structure {· , ·} on a finite-dimensional vector space V is
called a linear Poisson structure if for each pair of linear functions ξ and η on V ,
their Poisson bracket {ξ , η } is a linear function on V .
Proposition 7.3. For every finite-dimensional vector space V , there is a natural one-
to-one correspondence between linear Poisson structures on V ∗ and Lie algebra
structures on V .
Proof. As we have explained, when V = g is a finite-dimensional 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 Lie–Poisson
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 Lie–Poisson 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 Lie–Poisson structure
(on g∗i ) is denoted by {· , ·}i . If φ : (g1 , [· , ·]1 ) → (g2 , [· , ·]2 ) is a Lie algebra ho-
momorphism, then the transpose map φ : (g∗2 , {· , ·}2 ) → (g∗1 , {· , ·}1 ) is a Poisson
map. It is easily checked that this leads to a contravariant functor from the cate-
gory of finite-dimensional 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 Lie–Poisson 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

  d
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
{F, G} := ∑ ckij xk
∂ xi ∂ x j
.
i, j,k=1

In tensorial terms, the algebra of polynomial functions on g∗ is canonically iso-

morphic to Sg, the symmetric algebra of g, and the construction of the Lie–Poisson
structure on Sg can be given without passing via g∗ , which has the advantage that its
construction does not require g to be finite-dimensional. Explicitly, a typical mono-
mial z1 z2 . . . z p ∈ Sg is viewed as a polynomial function on g∗ by setting, for ξ ∈ g∗ ,
 
(z1 z2 . . . z p )(ξ ) := ξ (z1 ) ξ (z2 ) . . . ξ (z p ) = ξ , z1  ξ , z2  · · · ξ , z p . (7.6)

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

  p q
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 Lie–Poisson Structure on g

Upon choosing an isomorphism between the finite-dimensional Lie algebra g and

its dual g∗ , the Lie–Poisson structure on g∗ can be transferred to a linear Poisson
structure on g. This is usually done by choosing a non-degenerate symmetric bi-
linear form · | · on g. Non-degeneracy of the bilinear form · | · means that the
linear map χ : g → g∗ , which sends x ∈ g to the linear form x | · : y → x | y, is an
isomorphism. Explicitly, χ is given by
7.2 The Lie–Poisson Structure on g 183

χ (x), y = x | y = χ (y), x , (7.7)

for all x, y ∈ g. The inverse map to χ is denoted by ψ : g∗ → g. As in (7.7),

ξ , ψ (η ) = ψ (ξ ) | ψ (η ) = η , ψ (ξ ) , (7.8)

for all ξ , η ∈ g∗ . To a function on g∗ , we can associate a function on g and vice versa,

simply by composing it (on the right) with χ or with ψ . Similarly, every Poisson
structure on g∗ can be transferred to g and vice versa. In particular, the Lie–Poisson
structure on g∗ , introduced in Section 7.1, leads to a unique Poisson structure {· , ·}g
on g with respect to which χ is an isomorphism of Poisson manifolds. We call {· , ·}g
the Lie–Poisson structure on g with respect to · | ·.
Explicitly, let F, G ∈ F (g) and compute their Lie–Poisson bracket {F, G}g (with
respect to · | ·) at a point x ∈ g by using (7.5) as follows.

{F, G}g (x) = {F ◦ ψ , G ◦ ψ } (χ (x))

 
= χ (x), dχ (x) (F ◦ ψ ), dχ (x) (G ◦ ψ ) (7.9)
= x | [ψ (dx F), ψ (dx G)] .

We have used that dχ (x) (F ◦ ψ ), viewed as an element of g, is precisely ψ (dx F). To

check the latter, first use the chain rule and the fact that ψ is a linear map to find that

dχ (x) (F ◦ ψ ) = dx F ◦ ψ ,

so that, for every ξ ∈ g∗ ,

 
dχ (x) (F ◦ ψ ), ξ = dx F, ψ (ξ ) = ξ , ψ (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 = ψ (dx F). Since F is a
function on a vector space, (7.11) can also be written in the following form,
d
∇x F | y = F(x + ty) , (7.12)
dt |t=0

which is the most useful form for explicit computation.

The Lie–Poisson 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 finite-dimensional Lie algebra, equipped with a
non-degenerate, 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
 
{F, G}h⊥ (x) = F̃, G̃ g (x) , (7.14)

where F̃ and G̃ 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
d
∇x F | y = F(x + ty) = 0 ,
dt |t=0

for every y ∈ h⊥ . Since · | · is non-degenerate, 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 writ-
ten down in a simple form, if · | · is ad-invariant (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 ẋ := (XH )x , the Hamiltonian vector field XH at x, is given by the Lax
equation
ẋ = [∇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 Lie–Poisson Structure 185

Example 7.6. The basic example of a finite-dimensional quadratic Lie algebra is

given by the F-vector 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 ] = δ j,k Ei −
δi, Ek j , for all 1  i, j, k,   d. A natural symmetric bilinear form on g is given
 x | y:= Trace(xy). It is non-degenerate, thanks to the orthogonality relations
by
Ei j | Ek = δ j,k δi, . Moreover, ad-invariance follows at once from the basic prop-
erty Trace(xy) = Trace(yx) of the trace form. Thus, g is a quadratic Lie algebra. Let
us denote by ξi j the linear  function
 on g which picks the entry at position (i, j) of a
matrix: ξi j (x) := xi j = E ji | x . For all x, y ∈ g,
       
∇x ξi j | y = dx ξi j , y = ξi j , y = E ji | y ,

so that ∇x ξi j = ∇ξi j is independent of x and is given by ∇ξi j = E ji . It follows that

the Poisson bracket of two such linear functions on g is given, at x ∈ g, by
     
ξi j , ξk g (x) = x | [E ji , Ek ] = x | δi, E jk − δ j,k Ei
= (δi, ξk j − δ j,k ξi )(x) ,
 
so that ξi j , ξk g = δi, ξk j − δ j,k ξi . 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  
∇x H p | y = H p (x + ty) = Trace x p−1 y = x p−1 | y .
dt |t=0

It follows that ∇x H p = x p−1 , 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 Lie–Poisson Structure

In this section we first study the symplectic foliation of a Lie–Poisson structure and
show how it is related to the coadjoint action. We then explain how the Poisson
cohomology of a Lie–Poisson structure is related to Lie algebra cohomology. In
Section 7.3.3, we relate the modular class of a Lie–Poisson 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 Lie–Poisson 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 finite-dimensional Lie algebra and let G be an
arbitrary connected Lie group whose Lie algebra is g. The symplectic leaves of the
Lie–Poisson 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 ξ ∈ g∗ the vector space of all fundamental
vector fields of the coadjoint action of G on g∗ at ξ coincides with Hamξ (g∗ , {· , ·}),
the vector space of all Hamiltonian vector fields of the Lie–Poisson structure on g∗ ,
at ξ . To show it, rewrite (7.5) as
 
(XG )ξ [F] = ξ , − addξ G dξ F = ad∗dξ G ξ , dξ F

to find that the Hamiltonian vector field XG is given, at ξ ∈ g∗ , by

(XG )ξ = ad∗dξ G ξ .

For every x ∈ g, the linear function x∗ , defined by x∗ (ξ ) := ξ , x for all ξ ∈ g∗ ,

satisfies dξ x∗ = x∗ , for every ξ ∈ g∗ . It follows that
 
Hamξ (g∗ ) = ad∗dξ G ξ | G ∈ F (g∗ ) = {ad∗x ξ | 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 Lie–Poisson 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 Lie–Poisson structure on g∗ .

Suppose now, as in Section 7.2, that we have a non-degenerate symmetric bilin-

ear form · | · on g, and let χ : g → g∗ be the resulting isomorphism (see (7.7)).
Since χ realizes an isomorphism between the Lie–Poisson structure on g∗ and the
Lie–Poisson structure on g with respect to · | ·, Proposition 7.7 leads to the follow-
ing result.
Proposition 7.8. Let (g, [· , ·]) be a finite-dimensional 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 Lie–Poisson structure on g with respect
to · | · are the adjoint orbits of G and the Casimir functions are the Ad-invariant
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 Lie–Poisson Structure 187

Fig. 7.1 For r  0 the sphere x2 + y2 + z2 = r2 is a symplectic leaf of the Lie–Poisson structure
on (so3 (R))∗ .

ture, the symplectic foliation can be rather complicated. This is illustrated in the
following examples of (real) three-dimensional Lie algebras.

Example 7.9. We start with the Lie–Poisson 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

⎛ ⎞
0 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 non-zero. 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 sym-
plectic 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 Lie–Poisson 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 Lie–Poisson 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 non-zero,
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) The one-sheeted hyperboloids 4xy + z2 = c, for c > 0;
(2) The two components of the two-sheeted hyperboloids 4xy + z2 = c, for c < 0;
(3) The two components of the cone z2 = −4xy minus the origin;
(4) 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 , [z, x] = 0 .

The Poisson matrix is now given by

⎛ ⎞
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 two-
dimensional, 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 , non-trivial 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 0 y⎟ (7.16)
⎝ ⎠.
−x −y 0
7.3 Properties of the Lie–Poisson Structure 189

Fig. 7.3 For the open book foliation, which is the symplectic foliation of the Lie–Poisson structure,
given by the Poisson matrix (7.16) the leaves are the points of the Z-axis and the half-planes,
obtained by removing the Z-axis 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 Z-axis, while the symplectic leaves of dimension 2 are the half-
planes, obtained by removing the Z-axis from all the planes which pass through the
Z-axis, as follows again from Proposition 1.32. This symplectic foliation is called
the open book foliation. Taking an arbitrary point on the Z-axis, it is clear that on no
neighborhood of it is there a non-constant Casimir function. See Fig. 7.3.

Example 7.13. We consider, in the last example, for a fixed α ∈ C, the basic Lie
brackets
[x, y] = 0 , [y, z] = y , [z, x] = α x ,
which lead to the following Poisson matrix for the corresponding Lie–Poisson struc-
ture on C3 , ⎛ ⎞
0 0 −α x
⎜ ⎟
⎜ 0 0 y ⎟.
⎝ ⎠
α x −y 0

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
= αx −y =0,
∂z ∂x ∂y

on U. A solution of these equations is easily found, giving F = cxyα + d, where

c and d are integration constants. If α ∈
/ Q, then the level sets of F are not even
algebraic varieties.
190 7 Linear Poisson Structures and Lie Algebras

7.3.2 The Cohomology of Lie–Poisson Structures

We show in this section that the Poisson cohomology of a Lie–Poisson structure

is naturally related to Lie algebra cohomology, leading under some conditions to
a fairly explicit description of it. Let g be a finite-dimensional Lie algebra, whose
Lie–Poisson structure (on g∗ ) is denoted by π or by {· , ·}. The algebra of func-
tions F (g∗ ) which we consider can be taken indifferently as the algebra of polyno-
mial, 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∗ , ξ  :=
ξ , x for all ξ ∈ g∗ . Consider the map g → X1 (g∗ ), which associates to x ∈ g the
Hamiltonian vector field of −x∗ with respect to the Lie–Poisson 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 con-
sider 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 q-cochains 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
q  
δLq (c)(x0 , . . . , xq ) = ∑ (−1)i xi∗ , c(x0 , . . . , xi , . . . , xq ) (7.17)
i=0
 
+ ∑ (−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 q-th Poisson cohomology space
of (g∗ , π ).
Proposition 7.14. Let g be a finite-dimensional Lie algebra. The dual g∗ of g is
equipped with its algebra F (g∗ ) of polynomial, holomorphic or smooth functions.
The Lie–Poisson bracket on F (g∗ ) is denoted by π . For every q ∈ N there is a
canonical isomorphism
HLq (g, F (g∗ ))  Hπq (g∗ ) .
Proof. We show that the F (g∗ )-valued Lie algebra cohomology complex of g is
isomorphic to the Poisson cohomology complex of (g∗ , π ). Let Q be a Poisson q-
cochain of (F (g∗ ), π ), i.e., Q ∈ Xq (g∗ ). We associate to Q the element Ψq (Q) ∈
Cq (g; F (g∗ )) = Hom(∧q g, F (g∗ )), defined by

Ψq (Q) : ∧q g → F (g∗ )
(7.18)
x1 ∧ · · · ∧ xq → Q[x1∗ , . . . , xq∗ ] .
7.3 Properties of the Lie–Poisson Structure 191

It is easy to see that for every q ∈ N the resulting linear map Ψq : Xq (g∗ ) →
Hom(∧q g, F (g∗ )) is anisomorphism. Moreover, a direct comparison of (4.4)
and (7.17), using xi∗ , x∗j = [xi , x j ]∗ , shows that the following diagram is commu-
tative. q
δπ
Xq (g∗ ) Xq+1 (g∗ )

Ψq Ψq+1
q
δL
Hom(∧q g, F (g∗ )) Hom(∧q+1 g, F (g∗ ))

The isomorphisms (Ψq )q∈N therefore induce isomorphisms in cohomology.

For a representation (ρ ,V ) of g, recall that V g stands for the space of all g-
invariant elements, i.e., elements v ∈ V such that ρ (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

ıq : HLq (g) ⊗ Cas(g∗ ) → Hπq (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∗ , π ). Recall from Propo-
sition 7.7 that these functions are the Ad∗ -invariant functions. For a semi-simple Lie
algebra, a non-trivial property on its Lie algebra cohomology (with coefficients in a
finite-dimensional representation) implies that ıq is an isomorphism, for every q ∈ N,
as we show in the following proposition.
Proposition 7.15. Let g be a semi-simple Lie algebra. We denote by F (g∗ ) the al-
gebra of polynomial functions on g∗ and by Cas(g∗ ) ⊂ F (g∗ ) the subalgebra of
polynomial Casimir functions. For every q ∈ N, the map ıq , given by (7.20), is an
isomorphism:
ıq : HLq (g) ⊗ Cas(g∗ )  Hπq (g∗ ) .
Proof. We denote by Fk (g∗ ) the vector space of homogeneous polynomials of de-
gree k on g∗ , where k ∈ N. It is a subrepresentation of F (g∗ ), and

F (g∗ ) = Fk (g∗ ) ,
k∈N

which implies that the F (g∗ )-valued Lie algebra cohomology of g decomposes as
a direct sum,  q
HLq (g, F (g∗ )) = HL (g, Fk (g∗ )) ,
k∈N
192 7 Linear Poisson Structures and Lie Algebras

for every q ∈ N. Since the Lie–Poisson structure on g∗ is linear, a function F ∈

F (g∗ ) is a Casimir function if and only if for every k ∈ N its homogeneous com-
ponent 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∗ ) ,
k∈N

for every q ∈ N. Moreover, for every q, k ∈ N, the linear map ıq restricts to yield a
map
ıq,k : HLq (g) ⊗ Cask (g∗ ) → HLq (g, Fk (g∗ )) .
Now, according to [95, Th. 11], for every finite-dimensional representation V of a
semi-simple 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, ıq,k is an iso-
morphism, which, in turn, implies that ıq is an isomorphism, as was to be shown.

7.3.3 The Modular Class of a Lie–Poisson Structure

The modular form of a finite-dimensional Lie algebra g is the linear form μg on g,

defined for x ∈ g by
μg (x) := Trace(adx ) .
A Lie algebra is said to be unimodular when its modular form vanishes. For x, y ∈ g,

μg ([x, y]) = Trace(ad[x,y] ) = Trace([adx , ady ]) = 0 , (7.21)

so that μg vanishes on [g, g]. Every Lie algebra which satisfies [g, g] = g is therefore
unimodular; in particular, if g is semi-simple, 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, solv-
able Lie algebras are not unimodular in general, for instance, the two-dimensional
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 μg of g with the modular
vector field Φ of the Poisson manifold (g∗ , π ), where π 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). Φ 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 finite-dimensional Lie algebra. Interpreted as an ele-
ment of g∗ , the modular vector field of (g∗ , π ) with respect to any translation invari-
ant volume form is, up to a sign the modular form of g.

Proof. Let (e1 , . . . , ed ) be a basis of g and let (ξ1 , . . . , ξd ) be the dual basis of g∗ .
We use the functions x1 = e∗1 , . . . , xd = e∗d as linear coordinates on g∗ . According
7.4 Affine Poisson Structures 193

to (4.22), the divergence of the Lie–Poisson structure π on g∗ is given by

∂   ∂   ∂
Div(π ) = ∑ xi , x j = ∑ [ei , e j ]∗ , ξ j .
1i, jd ∂ x j ∂ xi 1i, jd ∂ xi

Since Φ = − Div(π ) (see Proposition 4.17), it follows that

   
Φ (x) = − ∑ [ei , e j ]∗ , ξ j xi = − ∑ ξ j , [x, e j ] = − Trace(adx ) ,
1i, jd 1 jd

for every x = ∑di=1 xi ei ∈ g. This shows that Φ = −μg .

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 finite-dimensional 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 finite-dimensional vector space V , there is a one-to-
one correspondence between affine Poisson structures on V and pairs of compatible
Poisson structures (π0 , π1 ) on V , where π0 is a constant Poisson structure and π1 is
a linear Poisson structure.

Proof. Suppose that π is a skew-symmetric biderivation of F (V ) such that for every

pair of linear functions F and G on V , one has that π [F, G] is an affine function on V .
For such functions, let π0 [F, G] and π1 [F, G] denote the constant, respectively linear
parts of π [F, G]. We view π0 and π1 as bilinear maps on V ∗ and we extend them to
biderivations of F (V ), still denoted by π0 and π1 , so that π = π0 + π1 . According to
Chapter 6, π0 is a constant Poisson structure. In terms of the Schouten bracket [· , ·]S ,
one has that π is a Poisson structure if and only if [π , π ]S = 0, which is equivalent
to the vanishing of [π , π ]S [F, G, H] for all linear functions F, G and H on V . Since
[π0 , π0 ]S = 0, because π0 is a Poisson structure, we have that

[π , π ]S [F, G, H] = [π1 , π1 ]S [F, G, H] + 2 [π0 , π1 ]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, [π , π ]S = 0 if and only if [π1 , π1 ]S = 0
(i.e., π1 is a linear Poisson structure) and [π0 , π1 ]S = 0 (i.e., π0 and π1 are compati-
ble). It follows that the map, which sends an affine Poisson structure π to its constant
and linear parts (π0 , π1 ), yields the announced one-to-one 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 π is an affine Poisson structure on V ∗ , then V ∗
is the dual of a Lie algebra V and π0 , restricted to V , is a skew-symmetric bilinear
form on V .
Proposition 7.19. For every finite-dimensional vector space V , there is a one-to-one
correspondence between affine Poisson structures on V ∗ and pairs ([· , ·], c) where
[· , ·] is a Lie algebra structure on V and c is a 2-cocycle in the trivial Lie algebra
cohomology of (V, [· , ·]).
Proof. According to Proposition 7.18, we need to show that there is a one-to-
one correspondence between pairs (π0 , π1 ) of compatible Poisson structures on V ∗ ,
where π0 is constant and π1 is linear, and pairs ([· , ·], c), where [· , ·] is a Lie algebra
structure on V and c is a 2-cocycle in the trivial Lie algebra cohomology of (V, [· , ·]).
Let (π0 , π1 ) be a pair of Poisson structures on V ∗ , where π0 is constant and π1 is lin-
ear. Let [· , ·] be the Lie algebra structure on V , defined2 by [v, w]∗ := π1 [v∗ , w∗ ], for
all v, w ∈ V , and let c(v, w) := π0 [v∗ , w∗ ], which defines a linear map c : V ∧V → F.
We show that π0 and π1 are compatible if and only if c is a 2-cocycle in the trivial
Lie algebra cohomology of g := V ; since the Poisson structures π0 and π1 can be
reconstructed from the pair ([· , ·], c), this will prove the proposition. We have for
all u, v, w ∈ V = g,

[π0 , π1 ]S (u∗ , v∗ , w∗ ) = π0 [π1 [u∗ , v∗ ], w∗ ]+  (u, v, w)

= c([u, v] , w)+  (u, v, w)
= −δL2 (c)(u, v, w) ,

see Example 4.2, in particular formula (4.2) for the Lie algebra coboundary opera-
tor δL , in the case of a trivial representation. It follows that [π0 , π1 ]S = 0 if and only
if c is a Lie 2-cocycle, δL2 (c) = 0.
In view of the above proposition, affine Poisson structures are also called modified
canonical Poisson structures or modified Lie–Poisson structures.
If g is semisimple, then HL2 (g) is trivial (see Lemma 4.1), so that, if c is a Lie 2-
cocycle, then c is a Lie 2-coboundary, c = δL1 (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 Lie–Poisson 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 Lie–Poisson
structures in a different way. Namely, given a Lie algebra (g, [· , ·]) and a Lie 2-
cocycle 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

2As in the case of Lie algebras, if v ∈ V , then v∗ denotes its image under the canonical isomor-
phism V  (V ∗ )∗ .
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 Lie–Poisson structure {· , ·}c on (g × F)∗ , restricted to a hyperplane,
is isomorphic to the affine Poisson structure {· , ·} on g, defined by the 2-cocycle c.
Proposition 7.20. Let (g, [· , ·]) be a finite-dimensional Lie algebra and let c be a 2-
cocycle in the trivial Lie algebra cohomology of g. Denote the Lie–Poisson 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) The linear function 1∗ : (g × F)∗ → F, defined by φ → φ (0, 1), is a Casimir
function of {· , ·}c ;
(2) The hyperplane N := {φ ∈ (g × F)∗ | φ (0, 1) = 1} is a Poisson submanifold
of (g × F)∗ ;
(3) The isomorphism Ψ : N → g∗ , defined for φ ∈ N by φ → φ (·, 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 (ξ , a) ∈ g∗ × F, it follows that d(ξ ,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 (ξ , a) = (ξ , a), [d(ξ ,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 Ψ is an isomorphism between the affine space N and the vec-
tor space g∗ . In order to prove that Ψ is a Poisson morphism, it is sufficient to
prove that {x∗ ◦ Ψ , y∗ ◦ Ψ }c = {x∗ , y∗ } ◦ Ψ , for every x, y ∈ g, where, as above,
stars stand for the corresponding linear functions on g∗ . By a slight abuse of no-
tation, we will also consider x∗ as a linear function on (g × F)∗ , where we put
x∗ (φ ) := φ (x, 0), for all φ ∈ (g × F)∗ . With this abuse of notation, x∗ ◦ Ψ = x∗ ,
and {x∗ , y∗ } ◦ Ψ = [x, y]∗ + c(x, y). For {x∗ ◦ Ψ , y∗ ◦ Ψ }c , which is now written
as {x∗ , y∗ }c , let φ ∈ N and compute
 
{x∗ , y∗ }c (φ ) = φ , dφ x∗ , dφ y∗ c = φ , [(x, 0), (y, 0)]c 
= φ , ([x, y], c(x, y)) = [x, y]∗ (φ ) + c(x, y) ,

which is {x∗ , y∗ } ◦ Ψ , evaluated at φ . This shows that Ψ is a Poisson morphism;

since Ψ 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 2-cocycle which is a coboundary, is isomorphic to the
196 7 Linear Poisson Structures and Lie Algebras

original Lie–Poisson 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 2-cocycle is not a coboundary,
these symplectic foliations may be quite different.

Example 7.21. Consider on R3 the linear and affine Poisson structures π1 and π ,
whose Poisson matrices are given by
⎛ ⎞ ⎛ ⎞
0 0 −x 0 1 −x
⎜ ⎟ ⎜ ⎟
⎜0 0 y ⎟ and ⎜ −1 0 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 2-cocycle is given by
c(x, y) = 1 and c(x, z) = c(y, z) = 0. Notice that π is a regular Poisson structure,
of rank two, while the rank of π1 vanishes at the origin. Clearly, xy is a Casimir
function of π1 , while xy + z is a Casimir function of π . The symplectic leaves of π1
are the points on the Z-axis, the two half-planes obtained by removing the Z-axis
from the plane x = 0, the two half-planes obtained by removing the Z-axis from
the plane y = 0, and the connected components of the hyperbolic cylinders xy = c,
with c ∈ R∗ . For the affine Poisson structure π , the symplectic leaves are the parallel
hyperboloids xy + z = c, with c ∈ R.

7.5 The Linearization of Poisson Structures

According to Weinstein’s splitting theorem (Theorem 1.25), every Poisson manifold

(M, π ) 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 Darboux’s theorem (Theorem 1.26), the symplectic Poisson
structure can be written in a canonical form in well-chosen 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 associ-
ated to the Poisson structure at m. This linear Poisson structure is introduced in the
following proposition.

Proposition 7.22. Let (M, π ) be a real or complex Poisson manifold, and suppose
that m ∈ M is a point at which π vanishes. There exists a unique linear Poisson
structure {· , ·}1 = π1 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 Ψ : (M, π ) → (M , π ) is a Poisson map, where (M , π ) is also a

Poisson manifold. Then π vanishes at Ψ (m) and dmΨ : (Tm M, π1 ) → (TΨ (m) M , π1 )
is a Poisson map, where the linear Poisson structure π1 on TΨ (m) M is defined as
in (7.23).

We call the linear Poisson structure π1 on Tm M the linearized Poisson structure of π

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 π1 on Tm M,
satisfying (7.23). In order to establish the existence of this Poisson structure, we
use (7.23); namely for ξ , η ∈ Tm∗ M, which we view as linear functions on Tm M, we
define {ξ , η }1 := dm {F, G}, where F and G are arbitrary local functions, whose
differentials at m are ξ and η 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)
i, j=1 ∂ xi ∂ x j
 
where xi j := xi , x j for all 1  i, j  d, the assumption that π 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
dm {F, G} = ∑ (m) (m) dm xi j .
i, j=1 ∂ xi ∂xj
This gives the required independence of the chosen functions F and G, since
       
∂F ∂ ∂
(m) = dm F, = ξ, ,
∂ xi ∂ xi m ∂ xi m

and similarly for the partial derivatives of G at m. The Jacobi identity for the Poisson
structure π implies the Jacobi identity for π1 since it follows at once from (7.23) that

{ξ , {η , ζ }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 ξ , η and ζ , respectively. This shows that there exists a (unique)
Poisson structure π1 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, π ) and (M , π ) are two Poisson manifolds and that Ψ :
M → M is a Poisson map. If πm = 0, then πΨ (m) = 0 and we can consider the
linearized Poisson structures at Tm M and at TΨ (m) M ; we denote them by {· , ·}1
and {· , ·}1 respectively. We show that dmΨ : (Tm M, {· , ·}1 ) → (TΨ (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 TΨ (m) M ,
{L1 , L2 }1 ◦ dmΨ = {L1 ◦ dmΨ , L2 ◦ dmΨ }1 . (7.25)
Given such functions L1 and L2 , there exist functions F and G, defined on a neigh-
borhood of Ψ (m) in M , such that dΨ (m) F = L1 and dΨ (m) G = L2 . Since Ψ is a
Poisson map, {F, G} ◦ Ψ = {F ◦ Ψ , G ◦ Ψ } , which we differentiate at m, giving

dΨ (m) {F, G} ◦ dmΨ = dm {F ◦ Ψ , G ◦ Ψ } .

Written in terms of the linearized Poisson structures, this becomes

 
dΨ (m) F, dΨ (m) G 1
◦ dmΨ = {dm (F ◦ Ψ ), dm (G ◦ Ψ )}1
 
= dΨ (m) F ◦ dmΨ , dΨ (m) G ◦ dmΨ 1 ,

which is (7.25), since L1 = dΨ (m) F and L2 = dΨ (m) G.

Definition 7.23. Let (M, π ) be a Poisson manifold, suppose that m is a point of M
at which π vanishes and denote the linearized Poisson structure of π at m by π1 . We
say that π is linearizable at m, if there exists a Poisson diffeomorphism Ψ , between
a neighborhood of m in (M, π ) and of the origin o in (Tm M, π1 ), with Ψ (m) = o.
Example 7.24. Let m be a point of a Poisson manifold (M, π ). Suppose that π van-
ishes at m and that the linearized Poisson structure of π at m is the trivial Poisson
structure. Then π is linearizable at m if and only if π is zero in a neighborhood of m.
Example 7.25. If π is a linear Poisson structure on a finite-dimensional vector
space V (over R or C), then π is linearizable at the origin o. In fact, the canoni-
cal isomorphism V  ToV is a Poisson isomorphism, where V is equipped with π
and where ToV is equipped with the linearized Poisson structure of π at o.
Example 7.26. Let π = {· , ·} be a non-trivial Poisson structure on a finite-dimen-
sional 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 lan-
guage of Section 8.1, π is a homogeneous Poisson structure of degree k). Then π
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 π is non-

trivial and is homogeneous, there is no neighborhood of o on which π is the trivial
Poisson bracket. It follows from Example 7.24 that π is not linearizable at o.
Example 7.27. Let F be a smooth or holomorphic function, defined on a neighbor-
hood 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 struc-
ture on U, defined in terms of the standard coordinates x, y of F2 by {x, y} = F. We
claim that π 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, π is in this case not
linearizable at o. Assume now that do F = 0 and consider the vector field

∂F ∂ ∂F ∂
V := − ,
∂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 π is linear in terms of these coordinates. According to Example 7.25, π is
linearizable at o.
Remark 7.28. For i = 1, 2, let (Mi , πi ) be a Poisson manifold, whose Poisson struc-
ture vanishes at mi ∈ Mi . Let Ψ : M1 → M2 be an isomorphism of Poisson manifolds
with Ψ (m1 ) = m2 . Then the tangent map TmΨ : Tm1 M1 → Tm2 M2 is an isomorphism
of Poisson manifolds, where both tangent spaces are equipped with the linearized
Poisson structures of π1 and π2 .
In particular, if two finite-dimensional vector spaces V1 and V2 , equipped with a
linear Poisson structure, are isomorphic as Poisson manifolds via a map Ψ which
sends the origin o of V1 to the origin of V2 , then the linear map doΨ : V1  ToV1 →
ToV2  V2 is a Poisson isomorphism. In other words, if there exists (in the neighbor-
hood 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 π 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 Ψ between U and a neighborhood of the origin o in Tm M, with
Ψ (m) = o, such that
{ ξ ◦ Ψ , η ◦ Ψ } = {ξ , η }1 ◦ Ψ ,

for all ξ , η ∈ Tm∗ M. Denoting for  ∈ N by Im

(+1)
(U) the ideal of F (U), consisting
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.
Definition 7.29. Let (M, π ) be a Poisson manifold, and suppose that m ∈ M is a
point at which π vanishes. Then π is said to be linearizable at m up to order  ∈ N∗
if there exists a neighborhood U of m in M and a diffeomorphism Ψ between U and
a neighborhood of the origin o in Tm M, with Ψ (m) = o, such that
(+1)
{ξ ◦ Ψ , η ◦ Ψ } − {ξ , η }1 ◦ Ψ ∈ Im (U) ,

for all ξ , η ∈ Tm∗ M.

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 π on R2 by {x, y} := f (x), where f (x) := e−x for x = 0 and
f (0) := 0. The fact that f vanishes at 0, together with all its derivatives at 0, implies
(+1)
that {x, y} ∈ Io (R2 ), for all  ∈ N, where o denotes the origin of R2 . It follows
on the one hand that the linearized Poisson structure at o is trivial, and on the other
hand that π is linearizable up to order , for all  ∈ N∗ . However, the Poisson struc-
ture π is not linearizable at o, since there is no neighborhood of o on which π is
zero.

In order to clarify and compare the different notions of linearizability at a point m

in a Poisson manifold (M, π ), 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
xi , x j − ∑ ckij xk ∈ Im (U) ,
(2)
(7.26)
k=1
  d
x̃i , x̃ j 1
= ∑ ckij x̃k , (7.27)
k=1
  d
yi , y j = ∑ ckij yk , (7.28)
k=1
  d
yi , y j − ∑ ckij yk ∈ Im
(+1)
(V ) . (7.29)
k=1

The inclusion (7.26) holds for all 1  i, j  d if and only if π vanishes at m. In this
case, the linearized Poisson structure π1 at m is given by (7.27), where x̃1 , . . . , x̃d are
the linear coordinates on Tm M, defined by x̃i := dm xi , for i = 1, . . . , d. Still assuming
that π 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 π is
linearizable at m (respectively π is linearizable at m up to order ).
According to Proposition 7.3, there is a one-to-one correspondence between lin-
ear 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 π
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 Conn’s linearization results, which
we formulate in the following theorem.
7.5 The Linearization of Poisson Structures 201

Theorem 7.31 (Conn’s linearization theorem). Let (M, π ) be a Poisson manifold,

and let m ∈ M be a point at which π vanishes. Let [· , ·] denote the Lie bracket
on Tm∗ M, associated to the linearized Poisson structure at m.
(1) In the holomorphic case, if (Tm∗ M, [· , ·]) is a semi-simple complex Lie algebra,
then π is linearizable at m;
(2) In the real case, if (Tm∗ M, [· , ·]) is a compact semi-simple Lie algebra, then π
is linearizable at m;
(3) In both cases, if (Tm∗ M, [· , ·]) is a semi-simple Lie algebra, then π is lineariz-
able at m up to order , for all  ∈ N∗ .

Proof. We do not prove Conn’s results (1) and (2), as the proof is long and rather
technical (see [45, 46]); instead, we prove the analogous (but easier to prove) re-
sult (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 π is linearizable up to order . According to Definition 7.29, there exists a
neighborhood U of m in M and a diffeomorphism Ψ between U and a neighborhood
of the origin o in Tm M, with Ψ (m) = o, such that
(+1)
{x ◦ Ψ , y ◦ Ψ } − [x, y] ◦ Ψ ∈ 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 ◦ Ψ of F (U) and we denote
(+1) (+2)
I := Im (U) and I := Im (U). Notice that I /I is finite-dimensional,
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 π vanishes at m, we have that {I , F (U)} ⊂ I and {I , F (U)} ⊂ I . It

follows that we can define a map

χ : g × (I /I ) → I /I
(7.32)
(x, p(F)) →
 x · p(F) := p({Fx , F}) ,

where p : I → I /I denotes the canonical projection. We claim that χ is a (finite-

dimensional) 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)) . (7.33)

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,

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 π . Since g is semi-simple

and I /I is finite-dimensional, 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 2-cocycle is a Lie 2-coboundary. The Lie 2-cocycle c which we con-
sider is constructed from (7.31):

c : g∧g → I /I 
(7.34)
(x, y) → p( Fx , Fy − F[x,y] ) .

We have that c is a Lie 2-cocycle, 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 Ψ+1 from U to a neighborhood of o in Tm M, such that Fx =
x ◦ Ψ+1 , for every x ∈ g. Since dmΨ = dmΨ+1 , there exists a neighborhood U ⊂ U
7.6 Notes 203

of m in M on which Ψ+1 is a diffeomorphism. In terms of Ψ+1 , (7.36) is written as

(+2)
{x ◦ Ψ+1 , y ◦ Ψ+1 } − [x, y] ◦ Ψ+1 ∈ Im (U ) , (7.38)

for all x, y ∈ g, which means that π 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 Ψ 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 Lie–Poisson structures are in one-to-one correspondence with Lie al-
gebra structures is implicit in Lie’s 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 Kir-
illov [105, 106], Kostant [117] and Souriau [185], who clearly exhibit the symplectic
structure on the coadjoint orbits which is inherited from the Lie–Poisson structure;
this structure is nowadays known as the Kostant–Kirillov–Souriau symplectic struc-
ture. 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 one-to-one correspondence with linear
Poisson structures, Lie algebroids are in one-to-one 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 Dufour–Zung [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) ho-
mogeneous Poisson structures on a finite-dimensional 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, homo-
geneity of a Poisson structure permits one to split the determination of its Poisson
cohomology into its homogeneous parts. Also, weight homogeneous Poisson struc-
tures 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 struc-
tures on F2 , with a simple singularity.
Among distinguished classes of weight homogeneous Poisson structures, one
finds – besides constant and linear Poisson structures – quadratic Poisson struc-
tures (i.e., homogeneous Poisson structures of degree two), rank 2 Poisson structures
which arise from Nambu–Poisson structures and the transverse Poisson structures
to adjoint orbits in a semi-simple Lie algebra.
Section 8.1 is devoted to the study of general properties of (weight) homoge-
neous Poisson structures, while we study in Section 8.2 a few special properties and
characterizations of quadratic Poisson structures: their multiplicativity, their mod-
ular 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 semi-simple Lie algebra.
Unless otherwise stated, F denotes an arbitrary field of characteristic zero.

C. Laurent-Gengoux et al., Poisson Structures, 205

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 8, © Springer-Verlag Berlin Heidelberg 2013
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 struc-
tures.
Throughout the section, we consider Poisson structures and, more generally, mul-
tivector fields on a finite-dimensional vector space V , which is equipped with its al-
gebra of functions, denoted by F (V ). If F is an arbitrary field, this algebra is the al-
gebra 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 polyno-
mial 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 p-
vector fields on V (skew-symmetric p-derivations 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 finite-dimensional 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 p-vector field P on V (p ∈ N∗ ) is
said to be a polynomial p-vector 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 π on a finite-dimensional 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 p-vector field P on V
can be written as
∂ ∂
P= ∑ P[xi1 , . . . , xi p ]
∂ x i1
∧···∧
∂ xi p
, (8.1)
1i1 <···<i p d

so that P is a polynomial p-vector 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
∂ ∂
π= ∑ xi j ∧
∂ xi ∂ x j
,
1i< jd
 
where, for every 1  i < j  d, the function xi j := xi , x j is a polynomial function
on V , and the degree of π 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 finite-dimensional vector space. A function F ∈ F (V ) is said to be

homogeneous if there exists an integer r ∈ N, such that for every λ ∈ F and for
every v ∈ V , we have
F(λ v) = λ r F(v) . (8.2)
The integer r, which is unique if F = 0, is then called the degree of F, de-
noted 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(λ x1 , . . . , λ xd ) = λ r F(x1 , . . . , xd ) , (8.3)
for every λ ∈ F. A homogeneous function of degree zero is constant, since the condi-
tion F(λ v) = F(v) for all λ ∈ 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 finite-dimensional real (respectively complex) vector
space. If F is a smooth (respectively holomorphic) function on V and F is homoge-
neous 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 λ ∈ F and every v ∈ V . By differentiating, for a fixed λ , 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 “homo-

geneous 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 finite-dimensional vector space V
is said to be a homogeneous vector field on V if, for every F ∈ F (V ) which is ho-
mogeneous, the function V [F] ∈ F (V ) is also homogeneous. More generally, a
(smooth or holomorphic) p-vector 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 ho-

mogeneous. Since every polynomial function is the sum of homogeneous functions,
Proposition 8.3 implies that every homogeneous p-vector field on V is a polynomial
p-vector 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
∂ ∂
E = x1 + · · · + xd . (8.4)
∂ x1 ∂ xd
Clearly, E is a homogeneous vector field on V . Notice that, by differentiating (8.3)
with respect to λ , at λ = 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 non-zero p-vector field on a finite-dimensional vector
space V , where p ∈ N. The following conditions on P are equivalent.
(i) P is homogeneous;
(ii) 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;
(iii) 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 func-
tion of degree deg(F1 ) + · · · + deg(Fp ) + r − p;
(iv) 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 ϖ (P).

Proof. Let P be a non-zero p-vector 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 non-zero 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 p-vector 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 p-vector
field, we adopt the convention that both the degree and weight of the zero p-vector 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
 
p
= 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 ele-
ments F1 , . . . , Fp of F (V ), we have that P[F1 , . . . , Fp ] is zero, or is homogeneous of
p
degree ∑=1 deg(F ) − (r − p). This shows the equivalence (iii) ⇔ (iv).  

As a corollary, every polynomial p-vector field P on V can be written uniquely

as a finite sum of homogeneous p-vector 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

ϖ (P[F1 , . . . , Fp ]) = ϖ (F1 ) + · · · + ϖ (Fp ) + ϖ (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 finite-dimensional vector space. A Poisson structure π
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 π on V can be written in the
following form:
∂ ∂
π = ∑ xi j ∧ ;
1i< jd ∂ x i ∂ xj
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 π =
(r − 2)π .
Example 8.6. Let π be a Poisson structure on a finite-dimensional vector space V .
Then π is homogeneous of degree zero (equivalently, of weight −2) if and only if π
is a constant Poisson structure. Also, π is homogeneous of degree one (equivalently,
of weight −1) if and only if π is a linear Poisson structure.
210 8 Higher Degree Poisson Structures

Example 8.7. Consider the bivector field π := V1 ∧ V2 , where V1 and V2 are two
homogeneous vector fields on a finite-dimensional vector space V . Then π is homo-
geneous of weight ϖ (V1 ) + ϖ (V2 ), because

LE (π ) = LE (V1 ) ∧ V2 + V1 ∧ LE (V2 ) = (ϖ (V1 ) + ϖ (V2 )) V1 ∧ V2 .

Since
[π , π ]S = [V1 ∧ V2 , V1 ∧ V2 ]S = −2 V1 ∧ V2 ∧ [V1 , V2 ] ,
we have that π is a (homogeneous) Poisson structure on V if and only if the vec-
tor 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 ar-
bitrary homogeneous vector field, since then [E , V ] = LE V = ϖ (V ) V , which is
a multiple of V . The Poisson structure E ∧ V is homogeneous of weight ϖ (V ),
since ϖ (E ) = 0.

To finish this section, we give a proposition whose main content is that an isomor-
phism between two homogeneous Poisson structures on a vector space can always
be realized by a linear isomorphism.
Proposition 8.8. Let π1 and π2 be two homogeneous Poisson structures on a finite-
dimensional vector space V . Then the following are equivalent:
(i) There exists a Poisson isomorphism L : (V, π1 ) → (V, π2 ) which is a linear
map;
(ii) There exists a Poisson isomorphism Ψ : (V, π1 ) → (V, π2 ) with Ψ (o) = o;
(iii) There exists a Poisson isomorphism Ψ : (U1 , π1 ) → (U2 , π2 ) with Ψ (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 Ψ : (U1 , π1 ) → (U2 , π2 ) be a Poisson isomorphism with Ψ (o) = o. We are
going to prove that the differential L = doΨ of Ψ at the origin is a linear Poisson
isomorphism between (V, π1 ) and (V, π2 ). 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 ◦ Ψ = {F ◦ Ψ , G ◦ Ψ }1
where {· , ·}1 = π1 and {· , ·}2 = π2 are the Poisson brackets corresponding to π1
and π2 . Since π1 and π2 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 respec-
tively. 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 = doΨ is invert-
ible, 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 (non-homogeneous) Poisson structures on a vector space

Fd become homogeneous Poisson structures upon assigning specific (positive inte-
ger) weights to the natural coordinates x1 , . . . , xd on Fd . This means that many prop-
erties and techniques of homogeneous Poisson structures carry over to this more
general class of Poisson structures, which we will call weight homogeneous Pois-
son structures.
We first introduce the notion of weight homogeneity for functions and for multi-
vector fields on Fd . A d-tuple of integers ϖ := (ϖ1 , . . . , ϖd ) will be called a weight
vector (for Fd ); unless otherwise stated, we will always assume that all integers ϖi
are positive. Generalizing (8.3), a function F on Fd is called weight homogeneous
with respect to ϖ if there exists an integer r ∈ N, such that

F(λ ϖ1 x1 , . . . , λ ϖd xd ) = λ r F(x1 , . . . , xd ) , (8.8)

for all λ ∈ F. In this case, the integer r, which is unique if F = 0, is called the weight
of F and is denoted by ϖ (F). It is easy to see that, if F, G ∈ F (Fd ) are weight
homogeneous with respect to ϖ , then their product FG is also weight homogeneous
with respect to ϖ , of weight ϖ (FG) = ϖ (F) + ϖ (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 ϖ = (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 ϖi , so that we often refer to ϖi 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 func-
tion on Cd ) and F is weight homogeneous with respect to some weight vector, then F
is a polynomial function.
Proof. Let ϖ = (ϖ1 , . . . , ϖd ) be a weight vector for Fd (F = R or F = C) and assume
that F ∈ F (Fd ) satisfies (8.8), for all λ ∈ F. Let us consider s ∈ N∗ , such that, for all
1  j1 , . . . , js  d, we have that ϖ j1 + · · · + ϖ 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
functions ∂ xi ∂...F∂ xi have negative weight, hence are zero on Fd . This implies that F
s

1 s
is necessarily a polynomial function on Fd . 

212 8 Higher Degree Poisson Structures

A p-vector field P on Fd is said to be weight homogeneous (with respect to ϖ ),

if for all F1 , . . . , Fp ∈ F (Fd ), which are weight homogeneous with respect to ϖ ,
the function P[F1 , . . . , Fp ] ∈ F (Fd ) is weight homogeneous with respect to ϖ . 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 ϖ = (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 Eϖ on Fd , which is defined by

∂ ∂
Eϖ := ϖ1 x1 + · · · + ϖ d xd . (8.9)
∂ x1 ∂ xd
Notice that the weighted Euler vector field Eϖ is weight homogeneous with re-
spect to ϖ . It is clear from (8.8) that a function F on Fd is weight homogeneous of
weight r if and only if
Eϖ [F] = LEϖ F = r F . (8.10)
This formula, which generalizes the classical Euler formula (8.5), is called the
weighted Euler formula.
Proposition 8.12. Let ϖ = (ϖ1 , . . . , ϖd ) be a weight vector for Fd and let P be a
non-zero p-vector field on Fd , where p ∈ N. The following conditions on P are
equivalent.
(i) P is weight homogeneous with respect to ϖ ;
(ii) 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 + ϖi1 +
· · · + ϖi p ;
(iii) 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 homoge-
neous of weight r + ϖ (F1 ) + · · · + ϖ (Fp );
(iv) There exists an integer r ∈ Z, such that LEϖ P = rP.
The (possibly negative) integer r, which appears in (ii)–(iv) is the same and is
called 2 the weight of P, denoted ϖ (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 p-vector field, we

adopt, as for homogeneous multivector fields, the convention that both the degree and weight of
the zero p-vector field are −∞.
8.2 Quadratic Poisson Structures 213

homogeneous, then the integers ϖ (V [xi ]) − ϖi and ϖ (V [x j ]) − ϖ j are equal. In or-


ϖ
der to prove this property, we consider for t ∈ F the function V (1 − t)xi j + txϖj i ,
ϖ
which is weight homogeneous, because (1 − t)xi j + txϖj i and V are weight homo-
   
ϖ
geneous. As in the proof of Proposition 8.4, it follows that V xi j and V xϖj i
ϖ ϖ j −1
have the same weight. Since V [xi j ] = ϖ j xi V [xi ] and V [xϖj i ] = ϖi xϖj i −1 V [x j ],
we have that

ϖi (ϖ j − 1) + ϖ (V [xi ]) = ϖ j (ϖi − 1) + ϖ (V [x j ]) ,

and hence ϖ (V [xi ]) − ϖi = ϖ (V [x j ]) − ϖ j , which is what we wanted to show. 



It follows, as in the case of homogeneous multivector fields, that every polyno-

mial p-vector field P ∈ X p (Fd ) can be written in a unique way as a finite sum of
weight homogeneous p-vector 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

ϖ (P[F1 , . . . , Fp ]) = ϖ (P) + ϖ (F1 ) + · · · + ϖ (Fp ) . (8.11)

Since LEϖ Eϖ = [Eϖ , Eϖ ] = 0, the weighted Euler vector field is weight homoge-
neous of weight zero.
Finally, we specialize the definition of weight homogeneity to the case of Poisson
structures.
Definition 8.13. A Poisson structure π on Fd is said to be weight homogeneous if for
all F, G ∈ F (Fd ) which are weight homogeneous with respect to ϖ , their Poisson
bracket {F, G} is weight homogeneous with respect to ϖ .
Writing a Poisson structure π on Fd in the form

∂ ∂
π= ∑ xi j ∧
∂ xi ∂ x j
,
1i< jd

it follows that π is weight homogeneous(of weight  ϖ (π )) with respect to ϖ if and

only if each one of the functions xi j := xi , x j is a weight homogeneous polyno-
mial, of weight ϖ (π ) + ϖi + ϖ j ; this is also equivalent to the equality

LEϖ [π ] = ϖ (π ) π .

8.2 Quadratic Poisson Structures

A bivector field on a finite-dimensional 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 π takes the form

∂ ∂
π= ∑ xi j ∧
∂ xi ∂ x j
, (8.12)
1i< jd
 
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 π 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 skew-symmetric d × d ma-

trix. A quadratic bivector field π = {· , ·} is defined on Fd by setting

∂ ∂
π := ∑ ai j xi x j ∧
∂ xi ∂ x j
. (8.13)
1i< jd

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 skew-symmetric. This proves the Jacobi identity for every
triple of coordinate functions, which shows, in view of Proposition 1.36, that the
bivector field π 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 π at m is equal to the rank of the
matrix, obtained by removing from the matrix A the k-th line and the k-th column
for all k for which xk (m) = 0.

Example 8.15. If V is a linear vector field on a finite-dimensional 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, π )
is a Poisson vector field if the Lie derivative of π with respect to V is zero.
Proposition 8.16. For a Poisson structure π on a finite-dimensional vector space V ,
the following conditions are equivalent:
(i) The Poisson structure π is quadratic;
(ii) The Euler vector field E is a Poisson vector field;
8.2 Quadratic Poisson Structures 215

(iii) 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 π = {· , ·} 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 π is quadratic. This shows that (i) and (iii) are
equivalent.  

The projective space P(V ), which is the variety of 1-dimensional 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 π is a quadratic Pois-
son 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 finite-dimensional vector space and let μ : V ×V → V be a bilinear map.

We refer in this section to μ as a product on V , even if we do not demand that μ
is associative. We say that μ is a product with unit, if V admits an element e, such
that μ (v, e) = v = μ (e, v) for all v ∈ V . A Poisson structure π on V is said to be
multiplicative with respect to μ if μ : 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 finite-dimensional vector space and let μ : V ×V → V

be a product with unit. Every polynomial Poisson structure on V which is multi-
plicative with respect to μ , is quadratic.

Proof. Let π be a polynomial Poisson structure on V , which is multiplicative. In

view of (2.12), multiplicativity of {· , ·} = π implies that, for all F, G ∈ F (V ), and
all v, w ∈ V :

{F, G} (μ (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) := μ (u, w) and Lv (u) := μ (v, u) respectively. By bilinearity of μ , for every
a ∈ F and v ∈ V , the equalities μ (v, ae) = av = μ (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} μ (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 π 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. Special-
izing (8.15), we obtain therefore that

{F, G} (av) = {F, G} (μ (v, ae)) = a2 {F, G} (v) ,

for all a ∈ F and v ∈ V . This shows that {F, G} is a homogeneous polynomial of de-
gree 2. Since this holds for all linear functions F and G on V , the Poisson structure π
is quadratic.  

Proposition 8.18 takes in the smooth or holomorphic context the following form.

Proposition 8.19. Let V be a finite-dimensional vector space over R or over C and

let μ : V × V → V be a product with unit. Every smooth or holomorphic Poisson
structure on V , multiplicative with respect to μ , is quadratic.

Proof. The proof of the proposition goes along the same lines as the proof of Propo-
sition 8.18; the only difference is that one needs to reprove Lemma 8.17 in that
context, namely, to prove that there exists no non-zero smooth or holomorphic func-
tion 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 there-
fore 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 bi-differential operator π of the form

∂ ∂
π= ∑ xi j ∧
∂ xi ∂ x j
,
1i< jd

where all the functions xi j are of class Ck on Rd . Notice that, if π 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 π on Rd is a Poisson structure of class Ck if π 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) μ which corresponds

to the natural product on R[h]/h2 , via the vector space isomorphism, defined by
(a1 , a2 ) → a1 + ha2 . Explicitly, μ is given by

μ ((a1 , a2 ), (b1 , b2 )) = (a1 b1 , a1 b2 + a2 b1 ) ,

and the unit for μ is (1, 0). It follows that x1 ◦ μ = y1 z1 and x2 ◦ μ = 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 π of class C1 on R2 is defined by {x1 , x2 } := x12 ln(|x1 |). We
denote the product Poisson structure on R4 = R2 × R2 by π . In order to show that π
is multiplicative with respect to μ , notice that {x1 ◦ μ , x2 ◦ μ } = {x1 , x2 } ◦ μ , which
follows from

{x1 , x2 } ◦ μ = (x12 ln(|x1 |)) ◦ μ = (y1 z1 )2 ln(|y1 z1 |)

and

{x1 ◦ μ , x2 ◦ μ } = {y1 z1 , y1 z2 + z1 y2 }
= z21 {y1 , y2 } + y21 {z1 , z2 }
= z21 y21 ln(|y1 |) + y21 z21 ln(|z1 |) .

However, π 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 ξi j (x) := xi j . These functions provide a set of linear coordinates on V . For i and j
as above, let εi, j := 1 if i > j and εi, j := −1 if i < j and εi, j := 0 if i = j. Then the
brackets, defined for 1  i, j, k,   d, by
218 8 Higher Degree Poisson Structures
 
ξi j , ξk := (εi,k + ε j, )ξi ξk 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 finite-dimensional 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 λ
on V . Recall from (4.24) that the modular vector field Φπ of a Poisson structure π
on V is, up to a sign, the divergence Div(π ) of π , i.e., Φπ = − Div(π ). 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 λ be an arbitrary translation invariant volume form on a
finite-dimensional vector space V . The modular vector field of every quadratic Pois-
son structure on V , with respect to λ , is a linear vector field on V .
In the study of quadratic Poisson structures, the modular vector field plays a par-
ticular rôle, 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 Φπ of the diagonal Poisson
structure π on Fd , introduced in Example 8.14 with respect to the translation invari-
ant volume form λ = dx1 ∧ · · · ∧ dxd . By using the formula Φπ = − Div(π ) and the
explicit expression of the divergence of a bivector field (4.22), we obtain
 
d d
∂
Φπ = −2 ∑ ∑ ai j xi . (8.17)
i=1 j=1 ∂ xi

Example 8.25. Let V be a d-dimensional vector space and let V be a linear vector
field on V . As we have seen in Example 8.15, π := 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 Φπ of the Poisson structure π = E ∧ V is Φπ =

d V − Div(V ) E .
We show in the following proposition that every quadratic Poisson structure π 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 π , while the other term is unimodu-
lar. 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 λ on V , the star operator  is the family of isomorphisms

 : Xq (V ) → Ω d−q (V ) ,

defined for Q ∈ Xq (V ) by  Q = ıQ λ as in (4.19), and that Div is defined in terms of

the de Rham differential d by

Div = −1 ◦ d ◦  .

Since Φπ = − Div(π ), it follows that π is unimodular if and only if the differential

(d − 2)-form  π is closed.
Proposition 8.26. Let V be a d-dimensional vector space and let λ be a translation
invariant volume form on V . Every quadratic Poisson structure π on V decomposes
as follows
1
π = −1 α + E ∧ Φπ ,
d
where Φπ is the modular vector field of π with respect to λ , and where α is a closed
differential (d − 2)-form which satisfies LΦπ α = 0. Also, −1 α has the following
properties: it is a quadratic unimodular Poisson structure, compatible with π , and
it satisfies LΦπ (−1 α ) = 0.

Proof. Since Φπ is a linear vector field, E ∧ Φπ is a quadratic Poisson structure (see

Example 8.15). This Poisson structure is compatible with π , since

[E ∧ Φπ , π ]S = (LE π ) ∧ Φπ − E ∧ (LΦπ π ) = 0 ,

where we have used in the last step that both E and Φπ are Poisson vector fields
(with respect to π ). The modular vector field Φπ of the Poisson structure π :=
E ∧ Φπ is given by

Φπ = − Div(E ∧ Φπ ) = d Φπ − Div(Φπ ) E = d Φπ , (8.19)

where we have used (8.18), together with the fact that Div(Φπ ) = 0, which holds
because Φπ is itself the divergence of a bivector field. Since the quadratic Poisson
structures π and π = E ∧ Φπ are compatible, and since their modular vector fields
are proportional (see (8.19)), there exists a linear combination of π and E ∧ Φπ
which is a unimodular Poisson structure: R := π − d1 E ∧ Φπ is a Poisson structure
on V which is quadratic, compatible with π and unimodular.
Let us denote by α the differential (d − 2)-form on V , defined by α :=  R = ıR λ .
It is a closed form since R is unimodular. It remains to be checked that LΦπ R = 0
and that LΦπ α = 0. First, we have that LΦπ π = 0 (since Φπ is a Poisson vector
field), and we also have that LΦπ (E ∧ Φπ ) = 0 (since the vector fields E and Φπ
commute). We therefore have LΦπ R = 0. It follows from this, upon using (2) of
220 8 Higher Degree Poisson Structures

Proposition 3.11, that

LΦπ α = LΦπ (ıR λ ) = ıR (LΦπ λ ) + ı[Φπ ,R]S λ = 0 ,

where we used in the last step that [Φπ , R]S = LΦπ R = 0 and that LΦπ λ =
Div(Φπ ) λ = 0 (see (4.20)). This finishes the proof of the announced properties
of α and of R = −1 α . 

The quadratic Poisson structures whose modular vector field satisfies a genericity
assumption, stated below, are diagonal, when expressed in a system of well-chosen
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 d-dimensional
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 ∗
(8.20)
F → V [F] .

Definition 8.27. Let V be a linear vector field on a finite-dimensional vector spa-

ce V . We say that V is diagonalizable if the endomorphism V (1) of V ∗ is diago-
nalizable. 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
∂
V = ∑ ai xi . (8.21)
i=1 ∂ xi
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 diag-
onal 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 π on a finite-dimensional vector
space V is said to be diagonalizable, if there exist linear coordinates x1 , . . . , xd on V
in terms of which π is diagonal, i.e.,

∂ ∂
π= ∑ ai j xi x j ∧
∂ xi ∂ x j
, (8.22)
1i< jd

where (ai j ) ∈ Matd (F) is a skew-symmetric matrix. The system of linear coordinates
(x1 , . . . , xd ) is then said to be log-canonical for π .
8.2 Quadratic Poisson Structures 221

The above-mentioned converse, which yields a sufficient condition for the diago-
nalizability of a quadratic Poisson structure is given in the following proposition.

Proposition 8.29. Let π be a quadratic Poisson structure on a d-dimensional vector

space V . Suppose that
(1) The modular vector field Φπ of π is diagonalizable, with eigenvalues a1 , . . . ,
ad , which
 belong to F;
(2) The d+12 sums ai + a j with 1  i  j  d are all different.
Then π is a diagonalizable Poisson structure on V . Moreover, every system of linear
coordinates on V , consisting of eigenvectors of Φπ , is log-canonical for π .

Proof. Since the modular vector field Φπ is linear, it restricts to a linear endomor-
(2)
phism Φπ of the space A2 of all homogeneous polynomial functions of degree 2
on V ,
(2)
Φπ : A2 → A2
F → Φπ [F] .
Suppose that Φπ is diagonalizable and let x1 , . . . , xd be independent eigenvectors
of Φπ , with eigenvalues a1 , . . . , ad . We use x1 , . . . , xd as linear coordinates on V .
The d+1 2 polynomial functions {xi x j | 1  i  j  d} form a basis of A2 , and
(2)
each one of them is an eigenvector of Φπ , because Φπ is a derivation: xi x j is an
(2)
eigenvector of Φπ , corresponding to the eigenvalue ai + a j , for all 1  i  j  d.
  
If we assume that the d+1
2 sums ai + a j | 1  i  j  d are all different (which
(2)
is assumption (2)), then all eigenvalues of Φπ are different, hence the eigenspace
(2)
decomposition of A2 with respect to Φπ is given by

A2 = Cxi x j . (8.23)
1i jd

We show that this implies that π is diagonalizable. To do this, we use that the mod-
ular vector field Φπ is a Poisson vector field: writing {· , ·} for π and specializing

Φπ [{F, G}] = {Φπ [F], G} + {F, Φπ [G]} ,

to F = xi and G = x j , where 1  i, j  d, we obtain that

       
Φπ [ xi , x j ] = Φπ [xi ], x j + xi , Φπ [x j ] = (ai + a j ) xi , x j .
  (2)
Therefore, xi , x j is zero or it is an eigenvector of Φπ with respect to the eigen-
value 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 π is of the diagonal
form (8.22) in the system of coordinates (x1 , . . . , xd ).  
222 8 Higher Degree Poisson Structures

8.3 Nambu–Poisson Structures

Let V be a d-dimensional vector space, equipped with linear coordinates x1 , . . . , xd

and denote the algebra of functions on V by F (V ), as before. Let χ , ϕ1 , . . . , ϕd−2
be d − 1 functions on V and define, for F, G ∈ F (V ), a function on V by

∂F ∂ G ∂ ϕ1 ∂ ϕd−2
···
∂ x1 ∂ x1 ∂ x1 ∂ x1
∂ (F, G, ϕ1 , . . . , ϕd−2 ) .. .. ..
{F, G} := χ =χ . . . . (8.24)
∂ (x1 , . . . , xd )
∂F ∂ G ∂ ϕ1 ∂ ϕd−2
···
∂ xd ∂ xd ∂ xd ∂ xd
It leads to a map F (V ) × F (V ) → F (V ), which is a skew-symmetric biderivation
of F (V ) and which, up to a non-zero constant, does not depend on the chosen linear
coordinates x1 , . . . , xd , since (8.24) can also be written as
dF ∧ dG ∧ dϕ1 ∧ · · · ∧ dϕd−2
{F, G} = χ . (8.25)
dx1 ∧ dx2 ∧ · · · ∧ dxd
Moreover, it defines a Poisson structure on V , a fact which we will prove in this sec-
tion. When the functions ϕ1 , . . . , ϕd−2 are polynomial or weight homogeneous, then
so is the corresponding Poisson structure. In order to prove that {· , ·} defines a Pois-
son structure on V , we briefly introduce the notion of a Nambu–Poisson structure,
which generalizes the notion of a Poisson structure.
Definition 8.30. For q  2, a q-vector field {· , . . . , ·} on a finite-dimensional vector
space V is called a Nambu–Poisson structure of order q on V if it satisfies the so-
called fundamental identity:

{F1 , . . . , Fq−1 , {G1 , . . . , Gq }} =

q (8.26)
∑ {G1 , . . . , Gi−1 , {F1 , . . . , Fq−1 , Gi }, Gi+1 , . . . , Gq } ,
i=1

for all functions F1 , . . . , Fq−1 , G1 , . . . , Gq ∈ F (V ).

Remark 8.31. The definition is easily generalized to arbitrary (real or complex) man-
ifolds by demanding that the q-vector field satisfies the fundamental identity for lo-
cal functions, which yields the notion of a Nambu–Poisson manifold. Replacing the
algebra of functions F (V ) by an arbitrary commutative associative algebra, leads
to the abstract notion of a Nambu–Poisson 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

Nambu–Poisson structure of order 2 is a bivector field {· , ·} which satisfies

{F, {G1 , G2 }} = {{F, G1 } , G2 } + {G1 , {F, G2 }} ,

8.3 Nambu–Poisson Structures 223

for all F, G1 , G2 ∈ F (V ), which is precisely the Jacobi identity for {· , ·}. Therefore,
a Nambu–Poisson structure of order 2 is the same as a Poisson structure. In this
sense, Nambu–Poisson structures are a generalization of Poisson structures.
Given a Nambu–Poisson structure {· , . . . , ·} of order q  2 on V , one can associate
to q − 1 elements F1 , . . . , Fq−1 of F (V ), a vector field XF1 ,...,Fq−1 on V by defining,
for G ∈ F (V ):
XF1 ,...,Fq−1 [G] := {G, F1 , . . . , Fq−1 } .
It is called the Hamiltonian vector field associated to F1 , . . . , Fq−1 . The fundamental
identity (8.26) says that all Hamiltonian vector fields XF1 ,...,Fq−1 are derivations of
(F (V ), {· , . . . , ·}).
We show in the following proposition how Nambu–Poisson structures of or-
der q − 1 can be constructed from Nambu–Poisson structures of order q.
Proposition 8.33. Let {· , . . . , ·} be a Nambu–Poisson structure of order q  3 on a
finite-dimensional vector space V . Given H ∈ F (V ), define a (q − 1)-vector field
on V by
{· , . . . , ·}H := {· , . . . , ·, H} .
Then {· , . . . , ·}H is a Nambu–Poisson structure of order q − 1 on V .
Proof. The fundamental identity (8.26) for {· , . . . , ·}H takes the form

{F1 , . . . , Fq−2 , {G1 , . . . , Gq−1 }H }H =

q−1 (8.27)
∑ {G1 , . . . , Gi−1 , {F1 , . . . , Fq−2 , Gi }H , Gi+1 , . . . , Gq−1 }H ,
i=1

for all F1 , . . . , Fq−2 , G1 , . . . , Gq−1 ∈ F (V ). To prove (8.27), substitute Fq−1 = H and

Gq = H in the fundamental identity (8.26), to find

{F1 , . . . , Fq−2 , H, {G1 , . . . , Gq−1 , H}} =

q−1
∑ {G1 , . . . , Gi−1 , {F1 , . . . , Fq−2 , H, Gi }, Gi+1 , . . . , Gq−1 , H} (8.28)
i=1
+ {G1 , . . . , Gq−1 , {F1 , . . . , Fq−2 , H, H}} ,

which is precisely (8.27), since the skew-symmetry of {· , . . . , ·} implies that the last
term in (8.28) is equal to zero. 
We show in the following proposition that for d-vector fields on a d-dimensional
vector space V , the fundamental identity is automatically satisfied.
Proposition 8.34. Let V be a d-dimensional vector space, with d  2. Every d-
vector field on V is a Nambu–Poisson structure of order d on V .
Proof. Let {· , . . . , ·} be a d-vector field on V . We need to prove that {· , . . . , ·} sat-
isfies the fundamental identity (8.26). To do this, we consider the multilinear map
N : F (V )2d−1 → F (V ), which is defined by
224 8 Higher Degree Poisson Structures

N (F1 , . . . , Fd−1 , G1 , . . . , Gd )
:= {F1 , . . . , Fd−1 , {G1 , . . . , Gd }}
d
− ∑ {G1 , . . . , Gi−1 , {F1 , . . . , Fd−1 , Gi } , Gi+1 , . . . , Gd } ,
i=1

for F1 , . . . , Fd−1 , G1 , . . . , Gd ∈ F (V ). Obviously, the vanishing of N is equivalent to

the fundamental identity (8.26) for {· , . . . , ·}. Moreover, since for all F1 , . . . , Fd−1 ∈
F (V ), the map

F (V )d → F (V )
(8.29)
(G1 , . . . , Gd ) → N (F1 , . . . , Fd−1 , G1 , . . . , Gd )

is a skew-symmetric d-derivation of F (V ), as is easily checked directly from the

definition of N , the vanishing of N is equivalent to

N (F1 , . . . , Fd−1 , x1 , . . . , xd ) = 0 , (8.30)

for all F1 , . . . , Fd−1 ∈ F (V ), where (x1 , . . . , xd ) is an arbitrary system of linear co-

ordinates on V . If we denote χ := {x1 , . . . , xd }, then the d-vector field {· , . . . , ·} can
be written, according to (8.1), as

∂ ∂
{· , . . . , ·} = χ ∧···∧ , (8.31)
∂ x1 ∂ xd
so that (8.30), which we need to prove, takes the form
d
∂
{F1 , . . . , Fd−1 , χ } = χ ∑ ∂ xi {F1 , . . . , Fd−1 , xi } . (8.32)
i=1

In order to prove (8.32) we compute its left-hand side from (8.31):

∂ F1 ∂ Fd−1 ∂χ
···
∂ x1 ∂ x1 ∂ x1
{F1 , . . . , Fd−1 , χ } = χ .
.. .. ..
. .
∂ F1 ∂ Fd−1 ∂χ
···
∂ xd ∂ xd ∂ xd
dF1 ∧ · · · ∧ dFd−1 ∧ dχ
=χ
dx1 ∧ · · · ∧ dxd
d χ dF1 ∧ · · · ∧ dFd−1
= (−1)d−1 χ
dx1 ∧ · · · ∧ dxd
d i ∧ · · · ∧ dxd
d Ci dx1 ∧ · · · ∧ dx
= (−1) d−1
χ ∑ dx1 ∧ · · · ∧ dxd
i=1
8.3 Nambu–Poisson Structures 225

d
∂ Ci
= χ ∑ (−1)d−i ,
i=1 ∂ xi

where
 
∂ ∂ ∂
Ci := χ (dF1 ∧ · · · ∧ dFd−1 ) ,..., ,...,
∂ x1 ∂ xi ∂ xd
 
∂ ∂ ∂ ∂
= χ (dF1 ∧ · · · ∧ dFd−1 ∧ dxi ) ,..., ,..., ,
∂ x1 ∂ xi ∂ xd ∂ xi
 
∂ ∂
= (−1) χ (dF1 ∧ · · · ∧ dFd−1 ∧ dxi )
d−i
,...,
∂ x1 ∂ xd
= (−1)d−i {F1 , . . . , Fd−1 , xi } .

We thus have obtained (8.32), so that {· , . . . , ·} satisfies the fundamental identity

and is a Nambu–Poisson 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 d-dimensional vector space and let χ , ϕ1 , . . . , ϕd−2 ∈
F (V ) be d − 1 functions on V . A Poisson structure π is defined on V by

∂ (F, G, ϕ1 , . . . , ϕd−2 )
{F, G} := χ , (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 ϕ1 , . . . , ϕd−2 are Casimir functions for π and the rank
of π is 2 at all points of V , except at the zeros of χ and at the points m ∈ V where
dm ϕ1 , . . . , dm ϕd−2 are linearly dependent (in those points the rank is zero).
Proof. Repeated use of Proposition 8.33 on the d-vector field (Nambu–Poisson
structure of order d) {· , . . . , ·} := χ ∂∂x ∧ · · · ∧ ∂∂x on V , yields a Nambu–Poisson
1 d
structure π of order 2, which is given by

∂ (F, G, ϕ1 , . . . , ϕd−2 )
{F, G} := {F, G, ϕ1 , . . . , ϕd−2 } = χ . (8.34)
∂ (x1 , . . . , xd )

Since π is a Nambu–Poisson structure of order 2, it is a Poisson structure (see Exam-

ple 8.32). The skew-symmetry of {· , . . . , ·} implies that {F, ϕi } = 0 for all F ∈ F (V )
and 1  i  d − 2, so that each one of ϕ1 , . . . , ϕd−2 is a Casimir function of π .
If m ∈ V with χ (m) = 0 or dm ϕ1 ∧ · · · ∧ dm ϕd−2 = 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 χ (m) = 0 and dm ϕ1 ∧· · ·∧dm ϕd−2 = 0.
Then
 there exist i, j such that dm xi ∧ dm x j ∧ dm ϕ1 ∧ · · · ∧ dm ϕd−2 = 0, so that

xi , x j (m) = 0. This shows that Rkm π  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 χ , ϕ1 , . . . , ϕd−2 are polyno-
mial functions on V , then the Poisson structure π is a polynomial Poisson struc-
ture on V . If, moreover, for some fixed weights ϖ1 , . . . , ϖd of the linear coordinates
x1 , . . . , xd on V , the functions χ , ϕ1 , . . . , ϕd−2 are weight homogeneous, then it also
follows from this formula that the Poisson structure π is a weight homogeneous
Poisson structure on V , of weight
d−2 d
ϖ (π ) = ϖ ( χ ) + ∑ ϖ (ϕi ) − ∑ ϖ i .
i=1 i=1

Remark 8.36. In the case of a d-dimensional manifold M, every d-vector field on M

defines a Nambu–Poisson structure of order d and hence, upon fixing d − 2 functions
ϕ1 , . . . , ϕd−2 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 d-vector field {· , . . . , ·} takes the form

∂ ∂
{· , . . . , ·} = χ ∧···∧ ,
∂ x1 ∂ xd
and the formula for the Poisson structure is given by

∂ (F, G, ϕ1 , . . . , ϕd−2 )
{F, G} := χ ,
∂ (x1 , . . . , xd )

so it is in local coordinates formally given by the same expression as in (8.33).

A coordinate-free description of the Poisson structure can be given when M is an
orientable manifold, so that there exists a volume form λ on M. Then a Nambu–
Poisson structure {· , . . . , ·} of order d is defined on M by letting

dF1 ∧ · · · ∧ dFd
{F1 , . . . , Fd } := χ ,
λ
for all (local) functions F1 , . . . , Fd on M. In particular, if we fix d − 1 functions
χ , ϕ1 , . . . , ϕd−2 on M and a volume form λ , then a Poisson structure π on M is
defined by
dF ∧ dG ∧ dϕ1 ∧ · · · ∧ dϕd−2
{F, G} := χ ,
λ
for all (local) functions F, G on M. The functions ϕ1 , . . . , ϕd−2 are Casimir functions
of π and the rank of π is two at all points of M, except at the zeros of χ and at
the points where the differentials of the latter Casimir functions are dependent. In
this case, a simple proof of the fact that π is a Poisson structure can be given by
extending, at points where the rank is two, the Casimir functions ϕ1 , . . . , ϕd−2 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 semi-simple
Lie algebra is weight homogeneous (of weight −2).

8.4.1 S-triples in Semi-Simple Lie Algebras

In this section, we recall a few facts which we will use about complex semi-simple
Lie algebra (see [99, 190] for details and proofs). Let (g, [· , ·]) be a semi-simple Lie
algebra over F = C; we denote the algebra of polynomial functions or of holomor-
phic 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
Jacobson–Morozov 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 non-zero triple (h, e, f ), satisfying (8.35), is called an S-triple 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 ,
i∈Z

where gi is the eigenspace of adh which corresponds to the eigenvalue i. For exam-
ple, e ∈ g2 and f ∈ g−2 . 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 s-module:

s
g= Vn j , (8.37)
j=1
228 8 Higher Degree Poisson Structures

where each Vn j is a simple s-module, 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 S-Triple

We show in this section that every S-triple (h, e, f ) of g leads to a system of linear
coordinates on g, centered at e, such that the Lie–Poisson structure on g is weight
homogeneous with respect to some weight vector ϖ , which has a natural Lie alge-
braic interpretation: up to a shift of 2, it consists of the weights of g as an s-module.
For every t ∈ C∗ , choose a complex number λt such that exp(−λt ) = t. Such a
choice cannot be made in such a way that λt 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
λ : C∗ → G
t → exp(λt h) .
For a fixed t, the map Adλ (t) : g → g is well-defined (independent of the choice made
for λt ) and it acts on each of the eigenspaces gi of adh as a homothecy with ratio t −i ,

Adλ (t) x = t −i x , (8.39)

for all x in gi . Indeed, for all t ∈ C∗ and x ∈ gi ,

Adλ (t) x = Adexp (λt h) x = exp(adλt h )x = exp(λt adh )x

∞
= ∑ (iλt )k x = exp(iλt )x = t −i x ;
k=0
√
also, replacing in this computation λt by λt + 2π n −1, where n ∈ Z, clearly leads
to the same result, which shows that the above choice of λt is irrelevant. It follows
that t → Adλ (t) is a (linear) action of C∗ on g. Since e ∈ g2 , the action ρ of C∗ on g,
defined for t ∈ C∗ and for y ∈ g by ρt · y := t 2 Adλ (t) y, fixes e; the action ρ is known
as Slodowy’s 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 Ad-invariance of the Killing form imply that

 
(ρt∗ Fx ) (z) = ρt −1 · z − e | x = t −2 Adλ (t −1 ) (z − e) | x
   
= t −2 z − e | Adλ (t) x = t −2 z − e |t −i x = t −i−2 Fx (z) .

We claim that, if we define for x ∈ gi the weight ϖ (Fx ) of Fx to be equal to i + 2,

then the Lie–Poisson 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 (ϖ (F1 ), . . . , ϖ (Fd )) = (i1 + 2, . . . , id + 2). To prove this, first recall
from Section 7.2 that the Lie–Poisson 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
 
ϖ ( Fx , Fy g ) − ϖ (Fx ) − ϖ (Fy ) = ϖ (F[x,y] ) − ϖ (Fx ) − ϖ (Fy )
= i + j + 2 − (i + 2) − ( j + 2) = −2 .

This result extends to the case i + j = −2, since then ϖ (F[x,y] ) = i + j + 2 = 0, which
is the weight of the constant function e | [x, y], and then ϖ ({Fx , Fy }g ) − ϖ (Fx ) −
ϖ (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 semi-simple 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
Lie–Poisson 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 Poisson–Dirac manifold

which is transversal to the symplectic leaf O at e. We show in the following propo-
sition 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 Lie–Poisson structure on g. Recall from Section 8.4.2 that the weight
homogeneity of the Lie–Poisson structure on g was determined by the choice of an
S-triple (h, e, f ) of g which contains e, that this S-triple 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 s-module.
Proposition 8.37. Let O = G · e be a nilpotent orbit in a semi-simple Lie algebra
g and let (h, e, f ) be an S-triple 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) N := e + n⊥ is transverse to O at e;
(2) N is a Poisson–Dirac submanifold of g;
(3) 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 s-module, 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) corre-

sponds 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 ν , 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) = ,
1i, js+2r −B(z) D(z)
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, 1  m  2r ;
Dm (z) = z | [xs+ , xs+m ] 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 in-
vertible for z in a neighborhood V ⊂ N of e in g. According to Proposition 5.27, V is
a Poisson–Dirac 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 ϖ (qi ) := ni + 2, for
i = 1, . . . , s. Notice that these weights are all positive.
We prove that the determinant of D is a non-zero 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,α1 . . . D2r,α2r , where {α1 , . . . , α2r } = {1, . . . , 2r}, so that its weight is
given by
2r 2r 2r
∑ ϖ (D,α ) = ∑ (ν + να + 2) = 2 ∑ (ν + 1) .
=1 =1 =1

Since =1 ν
∑2r + ∑si=1 ni
is the sum of all adh -weights, it is zero, which yields in
combination with (8.38)
2r s
∑ (ν + 1) = 2r − ∑ ni = 2r − (dim g − s) = 0 .
=1 i=1

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 trans-
verse Poisson structure is polynomial.
To finish, we show that the Poisson structure {· , ·}N is weight homogeneous of
weight −2 with respect to ϖ = (n1 + 2, . . . , ns + 2). According to (8.42) we need to
show that for all 1  i, j  s the functions Ai j and (BD−1 B )i j are weight homoge-
neous of degree ϖ (qi ) + ϖ (q j ) − 2 = ni + n j + 2. For Ai j this is clear, since A is part
of the Poisson matrix of the Lie–Poisson structure on g, which we have seen to be
weight homogeneous of weight −2. Similarly, we have that ϖ (Bim ) = ni + νm + 2.
Since
ϖ (Bim D−1 −1
m B j ) = ni + n j + νm + ν + 4 + ϖ (Dm ) ,
232 8 Higher Degree Poisson Structures

it means that we need to show that

ϖ (D−1
m ) = −νm − ν − 2 . (8.43)

Dαβ
A typical term of the element D−1 m of the inverse matrix D
−1 is of the form
det(D) ,
where Dαβ = Dα1 β1 . . . Dα2r−1 β2r−1 , with

{α1 , . . . , α2r−1 } = {1, . . . , 2r} \ {} and {β1 , . . . , β2r−1 } = {1, . . . , 2r} \ {m} .

Since det(D) has weight zero, we need to show that Dαβ has weight −νm − ν − 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 lin-
ear 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 skew-symmetric r-matrix on an associative algebra leads
to multiplicative quadratic Poisson structure, which makes the set of invertible el-
ements into a Poisson–Lie 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. Pois-
son structures coming from r-matrices or R-matrices are the subject of Chapter 10;
see Section 10.3.
See Dufour–Zung [63] for an extensive account on Nambu–Poisson structures.
Transverse Poisson structures to general adjoint orbits in a semi-simple Lie alge-
bra, with special emphasis to transverse Poisson structures to subregular orbits, are
studied in Damianou–Sabourin–Vanhaecke [56].
Chapter 9
Poisson Structures in Dimensions Two and Three

Low-dimensional 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 unex-
pected complexity. In the two-dimensional 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 non-trivial, 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 non-trivial 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 one-form, 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 Pois-
son structures in dimension two: their rank is equal to two (except when the Poisson
structure is trivial). Many results about three-dimensional Poisson manifolds are es-
sentially 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. Laurent-Gengoux et al., Poisson Structures, 233

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 9, © Springer-Verlag Berlin Heidelberg 2013
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 dimen-

sion two. Since a Poisson structure is a bivector field or a biderivation, the smallest
dimension which one can consider for studying non-trivial Poisson structures is di-
mension two. In this case, the Jacobi identity is always satisfied, so that every bivec-
tor field or skew-symmetric 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 π on M is a Pois-
son structure because [π , π ]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 π at m is zero or
two. Discarding the case of the trivial Poisson structure, it follows that the singular
locus of (M, π ) coincides with the set of points of M where π 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 skew-symmetric biderivation π of F (M) is a Poisson structure of M.
To do this, we prove that every skew-symmetric triderivation of F (M) is zero;
since [π , π ]S is a skew-symmetric triderivation of F (M), it follows that π is a Pois-
son structure on M. Let us denote by M sm the set of all smooth points of M. Let P
be a skew-symmetric triderivation on M and suppose that P is different from zero at
some point m ∈ M. Since P can be written as

∂ ∂ ∂
P= ∑ Pi jk ∧ ∧
∂ xi ∂ x j ∂ xk
,
1i< j<kd

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 non-zero 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 skew-symmetric triderivation
of F (M) corresponds on it to a trivector field. Since M sm is a two-dimensional
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 skew-symmetric biderivation π of F (M) is a Poisson structure on M. As
in the case of two-dimensional Poisson manifolds, the rank of π can only take the
9.1 Poisson Structures in Dimension Two 235

∂
Fig. 9.1 The singular locus of the Poisson structure π = χ ∂x ∧ ∂∂y is the zero locus of χ , i.e., the
curve defined by χ = 0.

values zero and two at the points of M, and the singular locus of (M, π ) coincides
with the set of points where π vanishes (assuming that π is non-trivial).
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 skew-symmetric biderivation on F (F2 )
or bivector field on F2 (i.e., every Poisson structure on F2 ) is of the form

∂ ∂
π=χ ∧ , where χ ∈ F (F2 ) , (9.1)
∂x ∂y

and by the above, every such π is a Poisson structure on F2 . Hence, there is a one-
to-one 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 π coincides with the zero locus
of χ , which is the curve {χ = 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

∂ ∂
π=χ ∧ , where χ = π [x, y] ∈ F (U) . (9.2)
∂x ∂y
We have seen in Section 9.1.1 that every bivector field on U is a Poisson structure.
The rank of π is two at all points m ∈ U where χ (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, π ) be a two-dimensional Poisson manifold and let m be
an arbitrary point of M. We write π in a neighborhood of m in M as

∂ ∂
π=χ ∧ ,
∂x ∂y

where x, y are local coordinates, vanishing at m, and where χ := π [x, y]. We distin-
guish the following three cases:
(1) If χ (m) = 0, which means that the rank of π at m is two, then there exists,
according to Weinstein’s splitting theorem (Theorem 1.25), a coordinate neighbor-
hood U of m in M, with coordinates x, y, centered at m, such that, on U,

∂ ∂
π= ∧ .
∂x ∂y

(2) If χ (m) = 0 but dm χ = 0, Example 7.27 tells us that π 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
∂ ∂
π =x ∧ .
∂x ∂y
(3) If χ (m) = 0 and dm χ = 0, then the zero locus of the function χ has a sin-
gularity at m. Let us consider the case of quadratic Poisson structures on C2 . Two
examples are given by

∂ ∂ ∂ ∂
π1 := (x2 − y2 ) ∧ and π2 := x2 ∧ .
∂x ∂y ∂x ∂y
Since the singular locus of π1 consists of the pair of lines y = ±x, while the singular
locus of π2 consists of the (double) line x2 = 0, it is clear that these Poisson struc-
tures 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
non-isomorphic Poisson structures on the plane, as there are non-isomorphic alge-
braic curves in the plane. In fact there are more. In order to see this, consider the
following two Poisson structures:

∂ ∂ ∂ ∂
π1 := (x2 − y2 ) ∧ , and π3 := 2(x2 − y2 ) ∧ .
∂x ∂y ∂x ∂y
Since they are proportional, they have the same zero locus, but they are not isomor-
phic as Poisson structures, a fact which can easily be checked by using Proposi-
tion 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 di-
mension two. Since the Jacobi identity is trivially satisfied in dimension two, the
classification amounts to the simpler problem of classifying bivector fields in di-
mension two, yet only partial results are known. The classification theorem which
we will present below has the following restrictions:
(1) The Poisson structures and the coordinate transformations which are consid-
ered are formal;
(2) The Poisson structures which are considered, are assumed to have a singular
locus which admits a simple singularity at the origin;
(3) 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 α ∂∂x ∧ ∂∂y , with α ∈ 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 isomor-
phism Φ : C[[x, y]] → C[[x, y]] such that the formal power series Φ (x) and Φ (y) have
no constant term. Two formal power series α , β ∈ C[[x, y]] are said to be equivalent
if there exists a formal coordinate transformation Φ of C2 , such that Φ (α ) = β . The
following formal version of the inverse function theorem holds: if Φ is an algebra
homomorphism such that the formal power series Φ (x) and Φ (y) have no constant
term and whose Jacobian
⎛ ⎞
∂ Φ (x) ∂ Φ (x)
⎜ ∂x ∂y ⎟
Jac Φ := det ⎜⎝ ∂ Φ (y) ∂ Φ (y) ⎠
⎟

∂x ∂y
is invertible, i.e., its constant term is different from zero, then Φ is a formal coordi-
nate transformation.
238 9 Poisson Structures in Dimensions Two and Three

Let Φ be a formal coordinate transformation of C2 . Given a formal Poisson struc-

ture π on C2 , there exists a unique formal Poisson structure on C2 , denoted Φ∗ π ,
such that Φ : C[[x, y]] → C[[x, y]]

Φ : (C[[x, y]], Φ∗ π ) → (C[[x, y]], π )

is a morphism (hence isomorphism) of Poisson algebras. Explicitly, Φ∗ π is given

by
(Φ∗ π ) [β , γ ] := Φ −1 (π [Φ (β ), Φ (γ )]) ,
for all β , γ ∈ C[[x, y]]. Since

∂ ∂
Φ∗ π = (Φ∗ π )[x, y] ∧ ,
∂x ∂y

it follows that, if we write π in the form π = α ∂∂x ∧ ∂∂y , then Φ∗ π takes the form

∂ ∂
Φ∗ π = Φ −1 (α Jac(Φ )) ∧ . (9.3)
∂x ∂y

It leads to the notion of equivalence of formal Poisson structures (on C2 ).

Definition 9.2. Two formal Poisson structures π1 and π2 on C2 are said to be for-
mally isomorphic if there exists a formal coordinate transformation of C2 , such
that π2 = Φ∗ π1 .
We consider formal Poisson structures on C2 , with a simple singularity at the origin,
i.e., formal Poisson structures of the form α ∂∂x ∧ ∂∂y , where α ∈ 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 α is equivalent to a weight ho-
mogeneous polynomial χ ∈ 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
∂ ∂
χβ ∧ , (9.4)
∂x ∂y
where χ ∈ C[x, y] is a weight homogeneous polynomial and β ∈ C[[x, y]] is a for-
mal 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 se-
ries β in (9.4) can be replaced, modulo a non-zero additive constant, by a weight
homogeneous polynomial.
Proposition 9.3. Let π = χβ ∂∂x ∧ ∂∂y be a formal Poisson structure on C2 , where
χ ∈ C[x, y] is a non-constant polynomial, which is weight homogeneous with respect
to the weight vector ϖ = (ϖ1 , ϖ2 ), and β ∈ C[[x, y]] is a formal power series, with
constant term 1. Then π is formally isomorphic to a Poisson structure of the form
9.1 Poisson Structures in Dimension Two 239

∂ ∂
χ (1 + ψ ) ∧ ,
∂x ∂y

where ψ ∈ C[x, y] is a weight homogeneous polynomial of weight ϖ (χ ) − ϖ1 − ϖ2 .

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 non-zero multiplicative constant, to a Poisson structure of the form

∂ ∂
χ (1 + ψ ) ∧ ,
∂x ∂y
where χ is one of the weight homogeneous polynomials which appear in Arnold’s
classification theorem of formal power series with a simple singularity (Theo-
rem 9.7) and ψ ∈ C[x, y] is an arbitrary weight homogeneous polynomial of weight
ϖ (χ ) − ϖ1 − ϖ2 . For each χ in Arnold’s list, one easily determines all possibili-
ties for ψ , 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 π = α ∂∂x ∧ ∂∂y be a formal Poisson structure on C2 , where
α ∈ C[[x, y]] is a formal power series, which has a simple singularity at the ori-
gin. Then, up to a non-zero multiplicative constant, π is formally isomorphic to a
Poisson structure of the form η ∂∂x ∧ ∂∂y , where η ∈ C[x, y] is one the polynomials
which appear in the third column of the following table.

Type k η
A2k k1 x2 + y2k+1
 
A2k−1 k1 x2 + y2k 1 + λ yk−1
 
D2k k2 x2 y + y2k−1 1 + λ x + μ yk−1
 2
D2k+1 k2 x y + y2k (1 + λ x)
E6 x 3 + y4
 3 
E7 x + xy3 1 + λ y2
E8 x 3 + y5

Remark 9.5. The Poisson structures which appear in the classification theorem are
not claimed to be all non-isomorphic. Special care has to be taken with respect to the
phrase “up to a non-zero multiplicative constant”, which appears in its statement. In
general, a formal Poisson structure π on C2 is formally isomorphic to every Poisson
structure of the form λ π , with λ ∈ C∗ , but there are exceptions. We already gave
in Section 9.1.2 an example of two proportional quadratic Poisson structures which
are non-isomorphic. In order to make a more general statement, let χ be a weight
homogeneous polynomial of C[x, y], with respect to the weight vector (ϖ1 , ϖ2 ) and
240 9 Poisson Structures in Dimensions Two and Three

of weight ϖ (χ ). Let us consider the Poisson structure on C2 corresponding to χ ,

namely π := χ ∂∂x ∧ ∂∂y . Define, for arbitrary a ∈ C∗ , the formal coordinate transfor-
mation of C2 , which amounts to rescaling x and y as follows:

Φa (x) := aϖ1 x, Φa (y) := aϖ2 y .

According to (8.8) and (9.3), we have

∂ ∂
(Φa )∗ π = aϖ1 +ϖ2 −ϖ (χ ) χ ∧ .
∂x ∂y

This formula permits us to conclude that, if ϖ (χ ) = ϖ1 + ϖ2 , then every Poisson

structure of the form λ π , with λ ∈ C∗ is formally isomorphic to π . However, if
ϖ (χ ) = ϖ1 + ϖ2 , then each of the formal coordinate transformations Φa acts triv-
ially on π ; in fact, as can be checked by using Proposition 8.8, the homogeneous
Poisson structures π and λ π are, for λ = ±1, non-isomorphic.

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 α ∈ C[[x, y]] can be written in a unique way as α =
∑i, j∈N αi j xi y j , so the datum of a family (αi j )i, j∈N in C is equivalent to the datum
of a formal power series in C[[x, y]]. This elementary fact is important when deal-
ing with the question whether certain operations yield well-defined power series: a
formal power series is well-defined if for every i, j ∈ N the coefficient ai j is well-
defined. For example, the product of two power series is well-defined, because each
term of the product is given by a finite sum. Similarly, if α ∈ C[[x, y]] is a formal
power series without constant term, then

αi
exp(α ) := ∑
i∈N i!

is a well-defined power series. A formal power series α ∈ C[[x, y]] can be evaluated
at (x, y) = (0, 0), which yields the constant term of the series, denoted α (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 α for which α (0, 0) = 0.
We define the order of a formal power series α ∈ C[[x, y]] as the minimum of
the total degrees of each of its terms, denoted ord(α ). When we are in a weight
homogeneous context, we will also consider the order of a power series α with
respect to a weight vector ϖ = (ϖ1 , ϖ2 ), where ϖ1 and ϖ2 are the weights of the
9.1 Poisson Structures in Dimension Two 241

variables x and y (see Section 8.1.3). Then ordϖ (α ), the order of α with respect to
ϖ , is by definition the minimum of the weights of each of the terms of α .
As we said previously, a formal coordinate transformation of C2 is an algebra iso-
morphism Φ : C[[x, y]] → C[[x, y]] so that the constant terms of Φ (x) and of Φ (y) are
zero. Also, recall that two formal power series α , β ∈ C[[x, y]] are said to be equiva-
lent if there exists a formal coordinate transformation Φ of C2 such that Φ (α ) = β .
Notice that if α and β are two equivalent formal series of C[[x, y]], they have the
same constant term.
Definition 9.6. A formal power series α ∈ C[[x, y]] satisfying α (0, 0) = 0 is said to
have a simple singularity at the origin, if a small enough neighborhood of α (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 Arnold’s 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 (Arnold’s theorem). Let α ∈ C[[x, y]] be a formal power series with-
out constant term, which has a simple singularity at (0, 0). Then α is equivalent to
one of the weight homogeneous polynomials χ which appear in the third column of
the following table.

Type k χ
Ak k1 x2 + yk+1
Dk k4 x2 y + yk−1
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 al-

gebra C[[x, y]] of formal power series in two variables. Since a Poisson bracket is
in particular a skew-symmetric biderivation, we first study the algebra of skew-
symmetric multi-derivations 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, ∂x ∧ ∂y yields a skew-symmetric biderivation of this algebra.
More general examples are constructed by taking α ∂∂x + β ∂∂y , respectively α ∂∂x ∧ ∂∂y ,
242 9 Poisson Structures in Dimensions Two and Three

where α , β ∈ C[[x, y]] are arbitrary. We show in the following proposition that there
are no other skew-symmetric k-derivations of C[[x, y]] (k  1).
Proposition 9.8. The spaces of skew-symmetric multi-derivations of C[[x, y]] are
given by

∂ ∂
X1 C[[x, y]] = α +β | α , β ∈ C[[x, y]] ,
∂x ∂y
∂ ∂
X2 C[[x, y]] = α ∧ | α ∈ C[[x, y]] ,
∂x ∂y
Xk 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 skew-symmetric biderivations and skew-symmetric k-derivations
of C[[x, y]] (for k  3) being analogous. To do that, we first point out that ev-
ery derivation W of C[[x, y]] satisfies the following property: if γ ∈ C[[x, y]], then
ord(W [γ ]) is either −∞ (i.e. W [γ ] = 0) or ord(W [γ ])  ord(γ ) − 1. This can be
done easily by recursion on ord(γ ), first in the particular case where γ ∈ C[[x]]
or γ ∈ C[[y]], and secondly for an arbitrary γ ∈ C[[x, y]], using the fact that every
γ ∈ C[[x, y]], with ord(γ )  1, can be written in at least one of the following two
forms:
• γ = xγ1 + γ2 , with γ1 ∈ C[[x, y]], ord(γ1 ) = ord(γ ) − 1 and γ2 ∈ C[[y]] ;
• γ = yγ1 + γ2 , with γ1 ∈ C[[x, y]], ord(γ1 ) = ord(γ ) − 1 and γ2 ∈ C[[x]] .
Now, let V ∈ X1 (C[[x, y]]) and consider the formal power series α := V [x] and
β := V [y]. We prove that V = α ∂∂x + β ∂∂y , by showing that W := V − α ∂∂x − β ∂∂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 γ ∈ C[[x, y]] be an arbitrary formal power
series. Suppose that W [γ ] = 0 and let s := ord(W [γ ]). We decompose the formal
power series γ as follows: γ = γs + γ̂s , where γs ∈ C[x, y] is a polynomial of total
degree at most s + 1 and γ̂s ∈ C[[x, y]] is a formal power series of order at least s + 2.
Since W vanishes on all polynomials, W [γ ] = W [γs ] + W [γ̂s ] = W [γ̂s ], so that

ord(W [γ ]) = ord(W [γ̂s ])  ord(γ̂s ) − 1  s + 1 = ord(W [γ ]) + 1 ,

which is impossible. Therefore, W [γ ] = 0 for all formal power series γ and the
equality V = α ∂∂x + β ∂∂y follows. 

It follows from Proposition 9.8 that every formal Poisson structure π on C2 is of

the form:
∂ ∂
π =α ∧ , (9.5)
∂x ∂y
for some formal power series α ∈ C[[x, y]]. Moreover, X3 (C[[x, y]]) = {0}, so that
the Schouten bracket of every skew-symmetric 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 struc-
ture on C2 . Thus, there is a one-to-one 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 skew-symmetric biderivations of C[[x, y]] and the set of
all formal Poisson structures on C2 . Notice that the series α in (9.5) is given by
α = π [x, y], so that a formal Poisson structure π on C2 is completely determined by
its value on the variables x and y. Given the above one-to-one 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 π = α ∂∂x ∧ ∂∂y be a formal Poisson structure of C2 , satisfying
α (0, 0) = 0. If α admits a simple singularity at the origin, then π is said to admit a
simple singularity at the origin.
To finish this subsection, we introduce a certain class of formal coordinate trans-
formations 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 β ∈ C[[x, y]] the series V [β ] is ei-
ther zero or of order larger than the order of β , with respect to a weight vector ϖ .
For every k  1 and for every β ∈ C[[x, y]], the formal power series V k [β ] is either
zero or of order at least k + ordϖ (β ). Therefore, for every i, j ∈ N, the coefficient
of xi y j in the infinite sum
1
exp (V ) (β ) := ∑ k! V k [β ]
k∈N

is determined by a finite number of terms, and we have a well-defined map

exp (V ) : C[[x, y]] → C[[x, y]] .

Clearly, the formal power series exp (V ) (x) and exp (V ) (y) vanish at the ori-
gin. We show that exp (V ) is an algebra homomorphism; i.e., that exp(V )(β γ ) =
exp(V )(β ) exp(V )(γ ), for all β , γ ∈ C[[x, y]]. Since V is a derivation of C[[x, y]],
we have for all β , γ ∈ C[[x, y]],
 i
i
V (β γ ) = ∑
i
V i−k (β ) V k (γ ) ,
k=0 k

so that
i  
1 i 1 i
exp(V )(β γ ) = ∑ V (β γ ) = ∑ ∑ V i−k (β ) V k (γ )
i∈N i! i∈N k=0 i! k
i
1
= ∑ ∑ (i − k)! k! V i−k (β ) V k (γ )
i∈N k=0
1
= ∑ j! k!
V j (β ) V k (γ ) = exp(V )(β ) exp(V )(γ ) .
j,k∈N
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 β and γ ); 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 α Eϖ , where α is a formal power series without constant term and ϖ
is a weight vector. In this case, we have for every non-constant formal power series β
that ordϖ (α Eϖ [β ]) = ordϖ (α ) + ordϖ (β ), so that exp(α Eϖ ) is a formal coordinate
transformation of C2 . Fixing a formal power series α without constant term, we will
refer to the family (exp (t α Eϖ ))t∈C as the formal flow of the derivation α Eϖ . For
fixed t, the formal coordinate transformation exp (t α Eϖ ) of C2 is called the formal
flow of α Eϖ 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,
ϖ = (ϖ1 , ϖ2 ) 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 ϖ . As
before, Eϖ denotes the weighted Euler derivation and we denote |ϖ | := ϖ1 + ϖ2 .
We start with three technical lemmas.

Lemma 9.10. Let π = χβ ∂∂x ∧ ∂∂y be a formal Poisson structure on C2 , where χ

is a (non-zero) weight homogeneous polynomial and β is a formal power series,
satisfying β (0, 0) = 1. Let α ∈ C[[x, y]] be a formal power series without constant
term and let (Φt )t∈C denote the formal flow of α Eϖ . For every t ∈ C, the polynomial
χ ∈ C[x, y] divides the skew-symmetric biderivation (Φt )∗ π .

Proof. According to (9.3), (Φt )∗ π = γ ∂∂x ∧ ∂∂y , where γ ∈ C[[x, y]] is given by

γ = Φt−1 (χβ Jac Φt ) = Φt−1 (β Jac Φt ) Φt−1 (χ ) . (9.6)

It therefore suffices to show that χ divides Φt−1 (χ ) for every t ∈ C. To do this, we

will show that there exists a one-parameter family (λt )t∈C of formal power series in
C[[x, y]], such that
Φt−1 (χ ) = λt χ ,
and such that moreover each coefficient of the power series λt depends polynomially
on t ∈ C. Suppose that (λt )t∈C is a one-parameter family of formal power series in
C[[x, y]], which satisfies λ0 = 1 and λt (0, 0) = 0, for every t ∈ C. Then we have the
following equivalences:

Φt−1 (χ ) = λt χ ⇐⇒ μt χ = Φt (χ ) ,
9.1 Poisson Structures in Dimension Two 245

1
where μt := is a formal power series in C[[x, y]] (depending on t), since
Φt (λt )
Φt (λt )(0, 0) = λt (0, 0) = 0 for all t. Our problem is therefore reduced to constructing
a one-parameter family (μt )t∈C of formal power series in C[[x, y]], such that Φt (χ ) =
μt χ , such that each coefficient of the power series μt depends polynomially on t, and
such that μ0 = 1 and μt (0, 0) = 0 for all t ∈ C. Under the latter hypothesis on μt , we
have the following equivalences:1
 ·
1
μt χ = Φt (χ ) ⇐⇒ Φt (χ ) = 0
μt
μ̇t 1
⇐⇒ − 2 Φt (χ ) + Φt (α Eϖ [χ ]) = 0
μt μt
 
−1 μ̇t
⇐⇒ α Eϖ [χ ] = Φt χ
μt
 
μ̇t
⇐⇒ ϖ (χ ) α = Φt−1 ,
μt

where we have used in the second equivalence that (Φt )t∈C is the formal flow of
α Eϖ , and the weighted Euler formula (8.10) in the last equivalence. Our problem is
thereby reduced to solving the linear differential equation

μ̇t = ϖ (χ )Φt (α )μt , (9.7)

with initial condition μ0 = 1. As in calculus,

⎛ ⎞
t
μt := exp ⎝ϖ (χ ) Φτ (α )dτ ⎠
0

is the unique solution. It is for each t a well-defined formal power series (since the
t
formal power series Φτ (α )dτ does not have a constant term) whose coefficients
0
depend polynomially on t, and it satisfies μt (0, 0) = 1, for all t ∈ C. This proves the
announced existence of the family (μt )t∈C , and hence of the family (λt )t∈C , which
proves that χ divides (Φt )∗ π , for all t. 
∂ ∂
Lemma 9.11. Let π = α ∂x ∧ ∂y be a formal Poisson structure on C2 and let us
write α as
α = χs0 +|ϖ | + · · · + χsr +|ϖ | + α̂ ,
where s0 < s1 < · · · < sr ∈ Z and where χs0 +|ϖ | , . . . , χsr +|ϖ | are the first r + 1
non-zero weight homogeneous terms of the formal power series α , of weight
s0 + |ϖ |, . . . , sr + |ϖ | and α̂ is a formal power series in C[[x, y]] of order at least

 in C[[x, y]], whose coefficients are polynomial functions

1 For a family (F )
t t∈C of power series
of t, we denote by Ḟt , respectively by 0t Fτ dτ , 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 + |ϖ | + 1. Assume that there exists a weight homogeneous derivation V =

V1 ∂∂x + V2 ∂∂y of C[[x, y]], such that
 
∂ ∂ ∂ ∂
LV χs0 +|ϖ | ∧ = χsr +|ϖ | ∧ . (9.8)
∂x ∂y ∂x ∂y

Let Φ be a formal coordinate transformation of C2 , defined by Φ := exp(V ). Then

Φ∗ π = β ∂∂x ∧ ∂∂y , where β is of the form

β = χs0 +|ϖ | + · · · + χsr−1 +|ϖ | + β̂ , (9.9)

with β̂ ∈ C[[x, y]] and of order at least sr + |ϖ | + 1. Stated briefly, Φ eliminates the
term χsr +|ϖ | without modifying the terms of lower weight.

Proof. Let π = α ∂∂x ∧ ∂∂y and suppose that there exists a weight homogeneous
derivation V , satisfying (9.8), where χs0 +|ϖ | and χsr +|ϖ | are the parts of α of weight
s0 + |ϖ |, respectively of weight sr + |ϖ |. Then V1 and V2 are weight homogeneous
polynomials of weight ϖ (V1 ) = sr − s0 + ϖ1 and ϖ (V2 ) = sr − s0 + ϖ2 . It implies
that Φ := exp(V ) is a well-defined formal coordinate transformation of C2 ; explic-
itly, Φ (x) and Φ (y) are of the form

Φ (x) = x + V1 + Vˆ1 ,
(9.10)
Φ (y) = y + V2 + Vˆ2 ,

where Vˆi is a formal power series in C[[x, y]] with ordϖ Vˆi > sr − s0 + ϖi for i =
1, 2, so that

   
∂ V1 ∂ V2
ϖ =ϖ = sr − s0 ,
∂x ∂y
   
∂ Vˆ1 ∂ Vˆ2
ordϖ > sr − s0 , ordϖ > sr − s0 ,
∂x ∂y

and  
 
 
 1 + ∂ V1 + ∂ V1 ∂ V1 ∂ Vˆ1
ˆ
+ 
 ∂x ∂x ∂y ∂y 
Jac Φ =  .

 ∂ V2 ∂ Vˆ2 ∂ V2 ∂ V2 
ˆ
 1+
 ∂x + ∂x ∂y
+
∂y 
It follows that the terms of α Jac Φ , whose weights are at most sr +|ϖ |, are precisely
the following:

∂ V1 ∂ V2
χs0 +|ϖ | , . . . , χsr−1 +|ϖ | , χsr +|ϖ | + χs0 +|ϖ | + χ .
∂x ∂ y s0 +|ϖ |
9.1 Poisson Structures in Dimension Two 247

Let β be the formal power series, defined by Φ∗ π = β ∂∂x ∧ ∂∂y . We denote its term
of weight i by ψi . According to (9.3), Φ (β ) = α Jac Φ , with Φ of the form (9.10).
The terms of Φ (β ), and hence of β , of weight i, smaller than sr + |ϖ |, coincide with
the corresponding terms of α Jac Φ . Specifically, all are zero except for

χs0 +|ϖ | = ψs0 +|ϖ | , . . . , χsr−1 +|ϖ | = ψsr−1 +|ϖ | . (9.11)

It follows that the term of weight sr + |ϖ | of Φ (β ) is given by

   
ψsr +|ϖ | + V ψs0 +|ϖ | = ψsr +|ϖ | + V χs0 +|ϖ | .

Equating these terms with the terms of weight sr + |ϖ | in α Jac Φ , we obtain

 
∂ V1 ∂ V2  
χsr +|ϖ | + + χs0 +|ϖ | = ψsr +|ϖ | + V χs0 +|ϖ | . (9.12)
∂x ∂y

Recall that V satisfies (9.8). Applied to x and y, this condition means that
 
  ∂ V1 ∂ V2
V χs0 +|ϖ | − + χs0 +|ϖ | = χsr +|ϖ | ,
∂x ∂y

which leads, according to equation (9.12), to ψsr +|ϖ | = 0. Combined with (9.11),
this shows that β is indeed of the desired form (9.9). 
The following lemma is an analog of Euler’s weighted formula for bivector fields
on C2 .
Lemma 9.12. Let ψ ∈ C[x, y] be a polynomial and let π = χ ∂∂x ∧ ∂∂y be a Poisson
structure on C2 . If ψ and π are weight homogeneous with respect to some weight
vector ϖ = (ϖ1 , ϖ2 ), then

Lψ Eϖ π = (ϖ (χ ) − |ϖ | − ϖ (ψ )) ψ π . (9.13)

Proof. Under the stated assumptions on ψ and π , we need to show that


Lψ Eϖ π [x, y] = (ϖ (χ ) − |ϖ | − ϖ (ψ )) ψ χ .

Using several times the weighted Euler formula (8.10), we compute


Lψ Eϖ π [x, y] = ψ Eϖ [χ ] − π [ψ Eϖ [x], y] − π [x, ψ Eϖ (y)]
= ψϖ (χ )χ − ϖ1 π [ψ x, y] − ϖ2 π [x, ψ y]
 
∂ψ ∂ψ
= ( ϖ ( χ ) − ϖ 1 − ϖ 2 )ψ χ − ϖ 1 x + ϖ2 y χ
∂x ∂y
= (ϖ (χ ) − |ϖ | − ϖ (ψ )) ψ χ ,

which yields the desired result. 

Using the above three lemmas, we prove Proposition 9.3.
248 9 Poisson Structures in Dimensions Two and Three

Proof (of Proposition 9.3). Let π = χβ ∂∂x ∧ ∂∂y be a formal Poisson structure on C2 ,
where χ ∈ C[x, y] and β ∈ C[[x, y]] are as stated. Since β verifies β (0, 0) = 1, we
can write β = 1 + ψs + βs , where ψs ∈ C[x, y] is a weight homogeneous polynomial
of weight s and βs ∈ 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 ϖ (χ ) − |ϖ |. Assume that
s < ϖ (χ ) − |ϖ | and consider the Lie derivative
 
∂ ∂ ∂ ∂
Lψs Eϖ χ ∧ = (ϖ (χ ) − |ϖ | − s) χψs ∧ ,
∂x ∂y ∂x ∂y

which follows easily from Lemma 9.12. Letting V := ψs Eϖ /(ϖ (χ ) − |ϖ | − s),

we have that (9.8) is satisfied, with r := 1 and χs0 +|ϖ | := χ and χsr +|ϖ | := χψs .
We can therefore apply Lemma 9.11, which shows that π is formally isomorphic
to (χ + α̂ ) ∂∂x ∧ ∂∂y , where α̂ is a formal power series, of order at least ϖ (χ ) + 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 (χ + α̂ ) ∂∂x ∧ ∂∂y is a multiple of χ . In conclusion, π is for-
mally isomorphic to χ (1 + βs ) ∂∂x ∧ ∂∂y , where βs is a formal power series of order
at least s + 1. A repeated use of this procedure shows that π is formally isomorphic
to χ (1 + β ) ∂∂x ∧ ∂∂y , where β is a formal power series of order at least ϖ (χ ) − |ϖ |.
Let us now show that, using the same technique, the formal power series β in
χ (1 + β ) ∂∂x ∧ ∂∂y can be replaced by its terms of weight ϖ (χ ) − |ϖ |. To do this,
let us write β = ψ + ϕ + β̂ , where ψ is a weight homogeneous polynomial of
weight ϖ (χ ) − |ϖ |, where ϕ is a weight homogeneous polynomial of weight at
least ϖ (χ ) − |ϖ | + 1, and where β̂ is a formal power series, with ordϖ β̂ > ϖ (ϕ ).
We first show how to remove the term ϕ from β , without changing ψ . Setting
r := 2 and χs0 +|ϖ | := χ and χs2 +|ϖ | := χϕ it is verified, as above, that the vector
field V := ϕ Eϖ /(ϖ (χ ) − |ϖ | − ϖ (ϕ )) satisfies (9.8), so Lemma 9.11 applies again,
showing as above that π is formally isomorphic to χ (1 + ψ + α̂ ) ∂∂x ∧ ∂∂y , where ψ
has weight ϖ (χ ) − |ϖ | and the order of α̂ is at least ϖ (χ ) − |ϖ |.
Repeating this procedure, we can make the order of α̂ arbitrarily large. To make
α̂ disappear, the procedure needs to be repeated an infinite number of times, i.e.,
we need to consider the algebra homomorphism Φ : C[[x, y]] → C[[x, y]] which
is obtained as the composition of an infinite number of the above formal coordi-
nate 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 Φ (x) and in Φ (y) will only depend on a finite num-
ber of the formal coordinate transformations. Therefore, π is formally isomorphic
to χ (1 + ψ ) ∂∂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 π = α ∂∂x ∧ ∂∂y be a formal Poisson structure on C2 ,
where α ∈ C[[x, y]] is a formal power series, which has a simple singularity at the
origin. According to Arnold’s theorem (Theorem 9.7) there exists a formal coordi-
nate transformation Φ , such that Φ −1 (α ) = χ , where χ is a weight homogeneous
polynomial, with respect to some weight vector ϖ = (ϖ1 , ϖ2 ). According to (9.3),
Φ∗ π = χβ ∂∂x ∧ ∂∂y , where β ∈ C[[x, y]] is a formal power series, with β (0, 0) = 0.
Using Proposition 9.3, we can conclude that, up to a constant, π is isomorphic to
χ (1 + ψ ) ∂∂x ∧ ∂∂y , where χ is one of the weight homogeneous polynomials which
appear in Theorem 9.7 and where ψ is a weight homogeneous polynomial of weight
ϖ (χ ) − |ϖ |. It suffices now to consider χ in the list of weight homogeneous poly-
nomials of Theorem 9.7 and construct the most general weight homogeneous poly-
nomial ψ ∈ C[x, y] of weight ϖ (χ ) − |ϖ |, which amounts to finding all monomials
xi y j , with i, j ∈ N satisfying

ϖ1 i + ϖ2 j = ϖ ( χ ) − ϖ1 − ϖ2 . (9.14)

The result is displayed in Table 9.1. 

Table 9.1 The different types of formal simple singularities which appear in Theorem 9.7, with
the corresponding weight vector ϖ = (ϖ1 , ϖ2 ), the integers ϖ1 and ϖ2 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 k χ (ϖ1 , ϖ2 ) Equation (9.14) xi y j

A2k k  1 x2 + y2k+1 (2k + 1, 2) (2k + 1)i + 2 j = 2k − 1 –
A2k−1 k  1 x2 + y2k (k, 1) ki + j = k − 1 yk−1
D2k k  2 x2 y + y2k−1 (k − 1, 1) (k − 1)i + j = k − 1 x, yk−1
D2k+1 k  2 x2 y + y2k (2k − 1, 2) (2k − 1)i + 2 j = 2k − 1 x
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 two-

dimensional 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 two-dimensional 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 al-
gebra 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 holo-
morphic) or polynomial functions on F3 , depending on the context. The standard
coordinates on F3 are denoted by (x, y, z). Every bivector field π on F3 decom-
poses as
∂ ∂ ∂ ∂ ∂ ∂
π = P1 ∧ + P2 ∧ + P3 ∧ , (9.15)
∂y ∂z ∂z ∂x ∂x ∂y
where P1 := {y, z}, P2 := {z, x} and P3 := {x, y}. It follows from item (iii) in Propo-
sition 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
P1 − + P2 − + P3 − =0. (9.16)
∂y ∂z ∂z ∂x ∂x ∂y

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
product on F and by curl the curl operator. Then (9.16) can be rewritten as
3

→
− −
→
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 well-known properties of vector
analysis. For instance, recall the classical identity
9.2 Poisson Structures in Dimension Three 251
 −
→ →
− →
−
curl χ P = χ curl P + grad χ × P , (9.18)
→
−
which is valid for every function χ ∈ 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 π be a Poisson structure on F3 . For every function χ ∈
F (F3 ), the bivector field χπ is a Poisson structure on F3 .
→
− →
−
Proof. We denote, as above, by P the vector field on F3 , associated to π , so that χ P
is the vector field associated to χπ . One computes, with the help of (9.18),
 −→  −→  −
→ →
− →
− −
→
χ P | curl χ P = χ 2 P | curl P + χ P | grad χ × P .

By (9.17), the assumption that π is a Poisson structure gives the vanishing of the
first term on the right-hand 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 χ P , so that the bivector field χπ is a Poisson
structure on F3 . 
Another classical identity which involves the curl operator is

curl grad ϕ = 0 ,

valid for every function ϕ ∈ F (F3 ). It implies, in view of (9.17), that when the
→
− →
−
vector field P , associated to a bivector field π , is the gradient of a function, P =
grad ϕ , then π is a Poisson structure. Substituted in (9.15), this means that

∂ϕ ∂ ∂ ∂ϕ ∂ ∂ ∂ϕ ∂ ∂
πϕ := ∧ + ∧ + ∧ (9.19)
∂x ∂y ∂z ∂y ∂z ∂x ∂z ∂x ∂y

is a Poisson structure on F3 , for every function ϕ ∈ F (F3 ). Let λ = dx ∧ dy ∧ dz

denote the standard volume form on F3 . Then it is clear from (9.19) that πϕ and dϕ
are related by
ıπ ϕ λ = d ϕ , (9.20)
which yields an alternative definition for πϕ . Also, this Poisson bracket comes from
a standard Nambu–Poisson structure of order 3 on F3 , upon using the function ϕ ,
as described in Proposition 8.35. Therefore, the Poisson bracket {F, G}ϕ of two
arbitrary functions F, G ∈ F (F3 ) can also be written as

dF ∧ dG ∧ dϕ
{F, G}ϕ = .
dx ∧ dy ∧ dz
In particular, ϕ is a Casimir function for πϕ . Proposition 9.13 implies the following
statement.
Proposition 9.14. Let ϕ ∈ F (F3 ) be an arbitrary function on F3 and consider the
bivector field πϕ on F3 , defined by (9.19). For every function χ ∈ F (F3 ), the bivec-
tor field χπϕ is a Poisson structure on F3 and it admits ϕ 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 π = {· , ·} (not assumed to be a Poisson
structure) every function ϕ ∈ F (F3 ) such that {F, ϕ } = 0 for all F ∈ F (F3 ).

Proposition 9.15. Let π be a bivector field on F3 and suppose that π admits a

Casimir function ϕ .
(1) If the differential of ϕ is different from zero in a dense subset of F3 , then π is
a Poisson structure on F3 ;
(2) If the bivector field π is a polynomial Poisson structure, and if the Casimir
function ϕ is a polynomial function for which the partial derivatives ∂∂ϕx , ∂∂ϕy
and ∂∂ϕz are relatively prime, then there exists a polynomial function χ ∈
F (F3 ), such that π = χπϕ , where πϕ is the Poisson structure defined in (9.19).

Proof. The bivector field π is a Poisson structure on F3 if and only if the Schouten
bracket [π , π ]S is zero. Since

1
[π , π ]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 ϕ is a Casimir function of π , then
[π , π ]S [F, G, ϕ ] = 0, for all F, G ∈ F (F3 ). Let us show that this implies that, if ϕ is a
Casimir function of π and dϕ is non-zero on a dense subset of F3 , then [π , π ]S = 0,
so that π is a Poisson structure on F3 . Since [π , π ]S is a trivector field on F3 , it can
be written as
∂ ∂ ∂
[π , π ]S = ψ ∧ ∧ , (9.21)
∂x ∂y ∂z
where ψ ∈ F (F3 ). Plugging F := x and G := y in [π , π ]S [F, G, ϕ ] = 0 we obtain,
in view of (9.21), ψ ∂∂ϕz = 0. Similarly, one obtains ψ ∂∂ϕx = ψ ∂∂ϕy = 0, by plugging
in the two other pairs of variables. Since, by assumption, dϕ is non-zero on a dense
subset of F3 , it follows that ψ = 0 on F3 , hence that the trivector field [π , π ]S is zero
on F3 . This proves (1).
We next prove (2). We assume that π is a polynomial Poisson structure on F3 and
we define polynomial functions P1 , P2 , P3 ∈ F (F3 ) by

∂ ∂ ∂ ∂ ∂ ∂
π = P1 ∧ + P2 ∧ + P3 ∧ .
∂y ∂z ∂z ∂x ∂x ∂y
Suppose that ϕ is a polynomial Casimir function of π . We apply the equation
{F, ϕ } = 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 :

∂ϕ ∂ϕ ∂ϕ ∂ϕ ∂ϕ ∂ϕ
P2 = P3 , P3 = P1 , P1 = P2 . (9.22)
∂z ∂y ∂x ∂z ∂y ∂x
9.2 Poisson Structures in Dimension Three 253

∂ϕ ∂ϕ ∂ϕ
Let us assume that the partial derivatives ∂ x , ∂ y , ∂ z are relatively prime, and let Q
denote the greatest common divisor of ∂ x and ∂∂ϕy . Then it follows from the third
∂ϕ

equation in (9.22) that there exists χ  ∈ F (F3 ), such that

∂ϕ ∂ϕ
QP1 = χ  and QP2 = χ  . (9.23)
∂x ∂y
In view of the first (or the second) equality in (9.22), this implies that

∂ϕ
QP3 = χ  . (9.24)
∂z

Since Q and ∂∂ϕz are relatively prime, Q divides χ  . Define χ by Qχ = χ  . We obtain

in view of (9.23) and (9.24):

∂ϕ ∂ϕ ∂ϕ
P1 = χ , P2 = χ and P3 = χ .
∂x ∂y ∂z
This proves (2). 

Remark 9.16. When ϕ is a non-constant weight homogeneous polynomial which is

square-free, then the partial derivatives ∂∂ϕx , ∂∂ϕy , ∂∂ϕz are relatively prime, so that the
assumptions about ϕ in item (2) of Proposition 9.15 are satisfied. Let us prove this
claim. If ϕ is weight homogeneous of weight ϖ (ϕ ) = 0, it satisfies the weighted
Euler formula (8.10),

∂ϕ ∂ϕ ∂ϕ
ϖ (ϕ )ϕ = ϖ 1 x + ϖ2 y + ϖ3 z , (9.25)
∂x ∂y ∂z
where ϖ1 , ϖ2 and ϖ3 are the weights of the variables x, y and z. Suppose that Q ∈
F (F3 ) is a non-constant irreducible polynomial function which divides the partial
derivatives ∂∂ϕx , ∂∂ϕy and ∂∂ϕz . It follows from (9.25) that Q divides ϕ . Since Q is non-
∂Q ∂Q ∂Q ∂Q
constant, at least one of the partial derivatives ∂x , ∂y and ∂z , say ∂x , is different
∂ϕ
from zero. Since the irreducible polynomial Q divides ϕ and its square Q2
∂x ,
also divides ϕ , so that ϕ is not square-free. Therefore, if ϕ is non-constant, weight
homogeneous and square-free, then ∂∂ϕx , ∂∂ϕy and ∂∂ϕz are relatively prime.

We finish this section with an example of Poisson structure on F3 which is not

of the form χπϕ , and which in fact does not admit Casimir functions defined on F3 ,
except constant functions. Consider the bivector field on F3 , defined by
 
∂ ∂ ∂
π := 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 π is a Poisson structure on F3 . A function ϕ ∈ F (M) is a

Casimir function for π if and only if

∂ϕ ∂ϕ ∂ϕ
= 0 and x +y =0.
∂z ∂x ∂y
The first equation means that ϕ does not depend on the variable z, while the second
one implies, in view of the Euler formula (8.5), that ϕ 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 π is constant. It
follows from Proposition 9.14 that π is not of the form χπϕ , with χ , ϕ ∈ F (F3 ).

9.2.2 Poisson Manifolds of Dimension Three

We now turn to Poisson structures on three-dimensional 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 three-dimensional 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 π be a bivector field on a manifold M. For every m ∈ M, we consider the asso-

ciated linear map πm : Tm∗ M → Tm M. As in the case of Poisson structures, the rank
of πm , which is always even, is called the rank of π at m, and π is said to be a
bivector field of rank two if the rank of π 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 π is two. If π is a bivector field of rank two on M, then the subset
{πm (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 π is equivalent to
the involutivity of this distribution.
Proposition 9.17. Let π = {· , ·} 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 π 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} π = XF ∧ XG ; (9.26)
(2) The bivector field π is a Poisson structure on M if and only if the distribution
{πm (Tm∗ M) | m ∈ M(2) } on M(2) is involutive.
9.2 Poisson Structures in Dimension Three 255

Proof. Suppose that π is a bivector field of rank two on M. Let F and G be arbitrary
functions, defined on a non-empty open subset U of M. Let m be an arbitrary point
of U. We show that

{F, G} (m) πm = (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 right-hand 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 left-hand 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 π at m is two, these tangent vectors span πm (Tm∗ M), so that for
every function H, defined on a neighborhood of m ∈ M, there exist a, b ∈ F such
that
(XH )m = a (XF )m + b (XG )m . (9.28)
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 pro-
portional, thereby proving (1).
We now prove (2). If π is a Poisson structure of rank 2 on M, then {πm (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
π be a bivector field of rank two on M and assume that the distribution {πm (Tm∗ M) |
m ∈ M(2) } is involutive. Let m be a point of M. If πm is zero, then the Schouten
bracket [π , π ]S , evaluated at m, vanishes as well. If πm 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), π = V ∧ XG on U, where V := {F,G} .A
direct computation, with the help of the graded Leibniz identity (3.42), gives
1
[π , π ]S = V ∧ [V , XG ] ∧ XG .
2

Since XG and V are tangent to the distribution {πm (Tm∗ M) | m ∈ M(2) }, assumed to
be involutive, the bracket [V , XG ] is also tangent to the latter distribution. Hence,
([π , π ]S )m ∈ ∧3 πm (Tm∗ M), for every m ∈ M(2) . But the latter space is zero, since
256 9 Poisson Structures in Dimensions Two and Three

πm (Tm∗ M) is a two-dimensional vector space. In conclusion, the Schouten bracket

of π with itself vanishes at every point in M, so that π is a Poisson structure. 

As a corollary, we obtain the following analog of Proposition 9.13.

Proposition 9.18. Let π be a Poisson structure of rank two on a manifold M. For

every function ϕ ∈ F (M), the bivector field ϕπ is a Poisson structure.

Proof. A proof can be given by picking local coordinates and using Proposi-
tion 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 π , respectively of ϕπ , is two. It
M(2)
 ⊂M 
is clear that M(2) (2) and that for every point m ∈ M(2) , the vector spaces

(ϕπ )m (Tm∗ M) and πm (Tm∗ M)

 defined by the latter vector spaces is involutive,
coincide. The distribution on M(2)
in view of item (2) in Proposition 9.17. Hence, the distribution on M(2) defined by

the former is also involutive, which in view of the same proposition, implies that ϕπ
is a Poisson structure on M. 

If in Proposition 9.18 the function ϕ is nowhere vanishing, the vector spaces

(ϕπ )m (Tm∗ M) and πm (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 π1 and π2 be two regular Poisson structures of rank 2 on
a manifold M. If π1 and π2 have the same symplectic leaves, then there exists a
(unique) nowhere vanishing function χ ∈ F (M), such that π1 = χπ2 .

Proof. Suppose that π1 and π2 are Poisson structures which have the same sym-
plectic 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
two-dimensional vector space, so that ∧2 Tm Sm is one-dimensional, while on the
other hand both (π1 )m and (π2 )m are non-zero elements of ∧2 Tm Sm , so there exists
a unique χ (m) ∈ F∗ , such that π1 (m) = χ (m)π2 (m). Thus, under the above assump-
tions, there exists a nowhere vanishing function χ on M, such that π1 = χπ2 ; since χ
is given locally as the quotient of two non-vanishing functions, χ belongs to F (M).


9.2.2.2 B – Poisson Manifolds of Dimension Three and One-Forms

Let M be a real orientable manifold of dimension three and let λ be a vol-

ume form on M. Recall from (4.19) that there is a natural F (M)-linear isomor-
phism  : Xk (M) → Ω 3−k (M), defined for Q ∈ Xk (M) by

 Q := ıQ λ .
9.2 Poisson Structures in Dimension Three 257

In particular, to a bivector field π on M corresponds a differential one-form α :=

π = iπ λ on M. In this subsection we wish to express the properties of π , for ex-
ample the condition that π be a Poisson structure on M, in terms of the differential
one-form α . Of course, this differential one-form depends on the chosen volume
form: replacing λ by F λ , with F a nowhere vanishing function, amounts to replac-
ing α by F α .
Proposition 9.20. Let M be a real orientable manifold of dimension three. Suppose
that π is a bivector field on M.
(1) Let λ be an arbitrary volume form on M, and let α := ıπ λ . Then π is a Poisson
structure on M if and only if

α ∧ dα = 0 . (9.29)

In this case, for every m ∈ M at which the rank of π is two, the kernel of αm is
the tangent space of the symplectic leaf of π through m;
(2) The bivector field π is a unimodular Poisson structure if and only if there
exists a volume form λ on M, such that the differential one-form α := ıπ λ is
closed.

Proof. Recall from Cartan’s formula (3.38) that if P and Q are two multivector fields
on M, then
[[ıP , d] , ıQ ] = ı[P,Q]S .
Specializing to P := π and Q := π , and applied to the volume form λ , it yields

2ıπ dıπ λ = ı[π ,π ]S λ . (9.30)

Multiplied by the volume form λ , and expressed in terms of α , (9.30) becomes

2(ıπ dα )λ = (ı[π ,π ]S λ )λ . (9.31)

Now, for every bivector field P and for every differential 2-form β , the fact that λ is
a top form implies that2
(ıP β )λ = β ∧ ıP λ . (9.32)
In particular, (ıπ dα )λ = dα ∧ ıπ λ = dα ∧ α , so that (9.31) gives eventually:

2 dα ∧ α = ı[π ,π ]S λ λ .

As a consequence, dα ∧ α = 0 if and only if [π , π ]S = 0, i.e., if and only if π 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 π at m is two, then
the spaces Ker(αm ) and Tm Sm coincide, where Sm is the symplectic leaf of π ,
passing through m. Since both vector spaces are two-dimensional, 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 λ = dx ∧ dy ∧ dz.
258 9 Poisson Structures in Dimensions Two and Three

the inclusion Tm Sm ⊂ Ker αm . 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,

α (XF ) = (ıπ λ ) (XF ) = λ (π ∧ XF ) ,

and π ∧ XF = 0 because the rank of π is 2 (see (9.26)), so that (XF )m ∈ Ker αm 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 Φ of π with
respect to a given volume form λ , can be expressed in terms of the divergence of π
as Φ = − Div(π ). According to the definition of Div (see (4.18)) and the definition
α := ıπ λ , we have that

 Φ = −  Div(π ) = −d  π = −dıπ λ = −dα ,

so that Φ = 0 if and only if α is closed. Since a Poisson structure is unimodular if

and only if there exists a volume form λ with respect to which the modular vector
field Φ is zero, this means that π is unimodular if and only if there exists a volume
form λ such that α := ıπ λ 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 λ the standard volume
form
λ := dx ∧ dy ∧ dz .
Recall from Section 8.2.2 that the modular vector field, with respect to λ , 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 λ is zero.
Specializing and adapting Proposition 8.26 to the three-dimensional case, yields
the next proposition. By definition, a cubic polynomial function on Fd is a homoge-
neous polynomial function of degree three on Fd .
Proposition 9.21. There is a one-to-one correspondence between quadratic Poisson
structures on F3 and pairs (Φ , ϕ ), with Φ a linear vector field on F3 , and ϕ a cubic
polynomial function on F3 , satisfying

Div(Φ ) = 0 and Φ [ϕ ] = 0 . (9.33)

This correspondence assigns to such a pair (Φ , ϕ ) the Poisson structure, defined by

9.2 Poisson Structures in Dimension Three 259

1
πΦ ,ϕ := πϕ − E ∧ Φ , (9.34)
3
where E is the Euler vector field, and πϕ is the Poisson structure defined in (9.19).

Proof. To start with, let us show that, given a linear vector field Φ on F3 and a cubic
function ϕ on F3 , which satisfy (9.33), the bivector field πΦ ,ϕ defined by (9.34) is
a Poisson structure. First, both E ∧ Φ and πϕ are Poisson structures according to
Example 8.15 and Proposition 9.14 respectively. As a consequence,
  2  2 2
π Φ , ϕ , π Φ , ϕ S = − π ϕ , E ∧ Φ S = − LE π ϕ ∧ Φ + L Φ π ϕ ∧ 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 πϕ = 0. The vanishing of the sec-
ond term follows at once by applying the following identity, valid for every vector
field V on F3 ,
LV πϕ = Div(V ) πϕ + πV [ϕ ] (9.35)
to V := Φ , upon using (9.33). We prove the identity (9.35) by showing that

ıLV πϕ λ = Div(V ) ıπϕ λ + ıπV [ϕ ] λ . (9.36)

First, by item (2) in Proposition 3.11 and by (4.20), the left-hand side of (9.36) is
given by

ıLV πϕ λ = ı[V ,πϕ ] λ = LV ıπϕ λ + ıπϕ LV λ = LV ıπϕ λ + Div(V ) ıπϕ λ . (9.37)

S

Therefore, it remains to be shown that LV ıπϕ λ = ıπV [ϕ ] λ , which means, in view

of (9.20), that LV dϕ = d(V [ϕ ]); but this is clear since LV and d commute. This
proves (9.35), and hence that the bivector field πΦ ,ϕ defined by (9.34) is a Poisson
structure.
Next, we show that the assignment (Φ , ϕ ) → πΦ ,ϕ , defined by (9.34), is surjec-
tive. Let π be a quadratic Poisson bracket, and let Φπ denote its modular vector field.
Proposition 8.26 implies that there exists a closed differential one-form α such that:
1
π = −1 α − E ∧ Φπ .
3
By the algebraic Poincaré lemma, there exists a function ϕ such that dϕ = α ; since
α is a quadratic differential one-form, ϕ can be chosen (in a unique way) to be a
cubic function. In view of (9.20), we have −1 α = −1 dϕ = πϕ . Hence, π is given
as in (9.34). We still have to check that the pair (Φπ , ϕ ) satisfies (9.33). According to
Proposition 4.17, the divergence of the modular vector field Φπ is zero. According
to Proposition 8.26, we have that LΦπ α = 0, so that d(Φπ [ϕ ]) = 0. Since Φπ [ϕ ] is
a cubic function, this implies Φπ [ϕ ] = 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 πϕ vanishes and according to Example 8.25,
 
1 1
Div(πΦ ,ϕ ) = Div − E ∧ Φ = Φ − Div(Φ ) E = Φ ,
3 3

so that we can recover Φ as the modular vector field of πΦ ,ϕ . Since the cubic poly-
nomial ϕ can also be uniquely recovered from πϕ , both Φ and ϕ can be uniquely
recovered from πΦ ,ϕ , 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 π be a quadratic Poisson structure on C3 . There exist linear
coordinates on C3 in which π takes the form
1
π = πϕ − E ∧ Φ ,
3
where (Φ , ϕ ) corresponds to one of the types I–VII in Table 9.2, where πϕ is the
Poisson structure
∂ϕ ∂ ∂ ∂ϕ ∂ ∂ ∂ϕ ∂ ∂
πϕ := ∧ + ∧ + ∧ ,
∂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 divergence-free linear vector fields on C3 . Recall that in this section the di-
vergence is always taken with respect to the standard volume form λ = dx ∧ dy ∧ dz
on C3 .
Lemma 9.23. Let Φ be a divergence-free linear vector field on C3 . There exist lin-
ear coordinates x, y, z on C3 such that Φ 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 finite-dimensional vec-
tor space V are in one-to-one correspondence with endomorphisms of the dual vec-
tor space V ∗ : more precisely, to a linear vector field Φ on V corresponds, as in
(8.20), the endomorphism V (1) of V ∗ defined by

Φ (1) : V ∗ → V ∗
(9.38)
F → Φ [F] .

If we write Φ in terms of linear coordinates x1 , . . . , xd on V as

d
∂
Φ= ∑ ai j xi
∂xj
,
i, j=1
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 πΦ ,ϕ ,
where Φ and ϕ are given in the second and third columns.

Type Φ ϕ Parameters

a1 , a2 , a3 ∈ C∗
all ai different
I a1 x ∂∂x + a2 y ∂∂y + a3 z ∂∂z bxyz
a1 + a2 + a3 = 0;
b∈C

a ∈ C∗
II ax ∂∂x + ay ∂∂y − 2az ∂∂z z G(x, y)
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

V x ∂∂y + y ∂∂z bx3 + cx(y2 − 2xz) b, c ∈ C

VI x ∂∂z G(x, y) G cubic function

VII 0 G(x, y, z) G cubic function

then A := (ai j ) ∈ Matd (C) is the matrix of Φ (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
Div(Φ ) = ∑ (ai j xi ) = ∑ aii ,
i, j=1 ∂ x j i=1

so that the divergence of Φ is a constant function, whose value is the trace of the
endomorphism Φ (1) . In particular, a linear vector field Φ on C3 is divergence-free
if and only if the corresponding endomorphism Φ (1) of C3 is traceless. There are
seven types of such endomorphisms:
• If all eigenvalues of Φ (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 well-chosen 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

a1 , a2 , a3 ∈ C∗
I diag(a1 , a2 , a3 ) all ai different
a1 + a2 + a3 = 0

II diag(a, a, −2a) a ∈ C∗

III diag(a, −a, 0) a ∈ C∗

⎛ ⎞
a1 0
⎜ ⎟
IV ⎝0 a 0 ⎠ a ∈ C∗
0 0 −2a

⎛ ⎞
010
⎜ ⎟
V ⎝0 0 1⎠ −
000

⎛ ⎞
001
⎜ ⎟
VI ⎝0 0 0⎠ −
000

VII 0 −

• If Φ (1) has precisely two different eigenvalues, then they are both different from
zero, leading to types II or IV, according to whether Φ (1) is diagonalizable or
not;
• If all eigenvalues of Φ (1) are equal, then they are all zero, so that Φ (1) is a
nilpotent endomorphism. This leads to three types, the types V–VII.
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 cor-
respond 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 classifica-
tion of the set of quadratic Poisson structures on C3 .
9.2 Poisson Structures in Dimension Three 263

Proof (of Theorem 9.22). Let π be a quadratic Poisson structure on C3 . According

to Proposition 9.21, there exists a pair (Φ , ϕ ), with Φ a divergence-free linear vector
field on C3 , and ϕ a cubic polynomial function on C3 , satisfying Φ [ϕ ] = 0, such that
π is of the form (9.34). According to Lemma 9.23, there exist linear coordinates
x, y, z on C3 in which Φ takes the form corresponding to one of the types I–VII in
Table 9.3. It therefore suffices to determine for each of these types the vector space
of all cubic functions ϕ ∈ F (C3 ), satisfying Φ [ϕ ] = 0. It is easy to see, by direct
computation, that for each line of the table, the given function ϕ satisfies indeed
Φ [ϕ ] = 0. Let us show that if Φ is of one of the types I–V, there are no other such
ϕ (for Φ of the other types VI and VII this is obvious).
For the types I–III, Φ is the weighted Euler operator, with (not necessarily inte-
gral) 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 Φ is spanned by
all monomials F satisfying Φ [F] = 0. Taking a cubic monomial F = xα1 yα2 zα3 ,
this leads for type I to the equation α1 a1 + α2 a2 + α3 a3 = 0, which has (1, 1, 1) as
its unique solution, under the constraint α1 + α2 + α3 = 3 (because deg(F) = 3),
with all ai ∈ C∗ being different. This leads to the monomial F = xyz. Similarly, the
equation Φ [F] = 0 leads for type II to the equation a(α1 + α2 ) − 2aα3 = 0, under
the constraint α1 + α2 + α3 = 3, which has as solution all triples (α1 , α2 , 1), with
α1 + α2 = 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 (α1 , α2 , α3 ),
satisfying α1 − α2 = 0, which has (1, 1, 1) and (0, 0, 3) as its only solutions under
the constraint α1 + α2 + α3 = 3; it corresponds to the monomials xyz and z3 .
For Φ of type IV, one computes easily the images under Φ of the 10 monomials
xα1 yα2 zα3 , with α1 + α2 + α3 = 3, by using the fact that Φ is the sum of the weighted
Euler operator (with weights (a, a, −2a)) and the operator x ∂∂y . One sees that, be-
sides for Φ (x2 z) (which is zero), these images are linearly independent, proving
that the kernel of Φ (restricted to the space of cubic polynomials) is spanned by
x2 z. Similarly, for Φ of type V, one easily checks that the images of the eight cubic
monomials xα1 yα2 zα3 , with (α1 , α2 , α3 ) = (3, 0, 0) and (α1 , α2 , α3 ) = (2, 0, 1) are
linearly independent. The kernel of Φ is therefore two-dimensional, 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) 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).
(2) Let ϕ be an irreducible polynomial in three variables. The Poisson structure
πϕ = {· , ·}ϕ on C3 , introduced in (9.19), given by
264 9 Poisson Structures in Dimensions Two and Three

∂ϕ ∂ ∂ ∂ϕ ∂ ∂ ∂ϕ ∂ ∂
πϕ := ∧ + ∧ + ∧ , (9.39)
∂x ∂y ∂z ∂y ∂z ∂x ∂z ∂x ∂y

admits ϕ as a Casimir, hence the singular affine surface Σϕ ⊂ C3 , given by

ϕ = 0, is a (singular) Poisson surface.
The purpose of the present section is to show how these two constructions are re-
lated. Namely, we will show that every Poisson surface C2 /G of the first type is
also of the second type Σϕ , for a polynomial ϕ which will be given explicitly (for
each G).
Before doing this, let us recall a few facts about the classification of finite sub-
groups 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
2π −1 2π −1
in terms of generators in the second column of the table, where ε := e 8 and η := e 5 .

Type of G Generators of G ⊂ SL2 (C)

⎛ √ ⎞
2 −1π
⎜e k+1 0 ⎟
Ak k1 ⎝ √
−2 −1π
⎠
0 e k+1

⎛ √ ⎞ ⎛ ⎞
−1π
⎜e k 0 ⎟ ⎜ 0 1 ⎟
Dk k4 ⎝ √
− −1π
⎠ and ⎝ ⎠
0 e k −1 0

⎞ ⎛
⎜ ε 7 ε7
⎟
E6 generators of D4 and √12 ⎝ ⎠
ε5 ε

⎞ ⎛
⎜ε 0 ⎟
E7 generators of E6 and ⎝ ⎠
0 ε7

⎛ ⎞ ⎛ ⎞
⎜ −η ⎜η +η
3 0 4 1
⎟ 1 ⎟
E8 ⎝ ⎠ and η 2 −η 3 ⎝ ⎠
0 −η 2 1 −η − η 4

Up to conjugation, every subgroup G of SL2 (C) is conjugated to one of these sub-

groups A1 , A2 , . . . , E8 . We then say that G is of type A1 , A2 , . . . , E8 . The quo-
tient 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 Σϕ , where ϕ 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.

Type of G ϕ Weights

Ak k1 x2 + y2 − zk+1 (k + 1, k + 1, 2)

Dk k  4 x2 + y2 z + zk−1 (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 G-invariant 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 ϕ . The isomorphism C2 /G  Σϕ is induced by the map C2 → C3 , given
by (P1 , P2 , P3 ), as expressed in the following commutative diagram:
(P1 ,P2 ,P3 )
C2 C3

p ı (9.40)

C2 /G 
Σϕ

Recall from Section 5.1.2 that F (C2 )G  F (C2 /G) (since G is a finite group). On
the other hand, since Σϕ is the affine surface of C3 , defined by the polynomial ϕ ,
its algebra of functions F (Σϕ ) is given by C[x, y, z]/ ϕ . It follows that we have an
algebra isomorphism
F (C2 )G  C[x, y, z]/ ϕ . (9.41)
We show in the following proposition that the isomorphism C2 /G  Σϕ is actu-
ally 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
(Σϕ , {· , ·}ϕ ), where ϕ is the polynomial, associated to G, as indicated in Table 9.5.

Proof. Consider the isomorphism Ξ : C2 /G → Σϕ , given in (9.40). We show that

Ξ 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 Σϕ , it follows that it is the image of p(o) under Ξ and that it is the only
singular point of Σϕ (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 {· , ·}ϕ is given, in terms of the generators x, y and z
of F (Σϕ ) as
⎛ ⎞
⎜ 0 ∂ϕ ∂ϕ ⎟
⎜ −
⎜ ∂z ∂y ⎟⎟
⎜ ⎟
⎜ ⎟
⎜ ∂ϕ ∂ϕ ⎟ .
⎜− 0 ⎟
⎜ ∂z ∂x ⎟
⎜ ⎟
⎜ ⎟
⎝ ∂ϕ ∂ϕ ⎠
− 0
∂y ∂x
∂ϕ ∂ϕ
The rank of this matrix is two, except at the point(s) of Σϕ for which ∂x = ∂y =
∂ϕ
∂z = 0, i.e., at the singular points of Σϕ . Since the origin of Σϕ is its only singular
point, the rank of {· , ·}ϕ is two at all points, except at the singular point of Σϕ (where
the rank is zero).
Using the isomorphism Ξ : C2 /G → Σϕ , we can transport the Poisson structure
{· , ·}C2 /G on C2 /G to a Poisson structure on Σϕ , which we denote by {· , ·}ϕ . On the
smooth surface Σϕ \ {p(o)} both Poisson structures {· , ·}ϕ and {· , ·}ϕ are of rank 2,
hence according to Proposition 9.19 there exists a nowhere vanishing holomorphic
function χ on Σϕ \ {p(o)} such that {· , ·}ϕ = χ {· , ·}ϕ . For arbitrary F, G ∈ F (Σϕ )
such that {F, G}ϕ = 0, we have that

{F, G}ϕ
χ= ,
{F, G}ϕ

so that χ is actually a rational function on Σϕ . Since its divisor of zeros and its di-
visor of poles are both discrete (they contain at most the point p(o)), χ is constant.
This implies that (Σϕ , {· , ·}ϕ ) and (C2 /G, μ {· , ·}C2 /G ) are isomorphic Poisson va-
rieties for some μ ∈ C∗ . In view of Remark 6.19, this constant μ 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 Arnold’s 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
non-formal, 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 classi-
fication 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 topo-
logically 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 Dufour–Haraki [60] and
Liu–Xu [129]. For the relation between Poisson structures and differential forms, in
the presence of a volume form, see [88].
Chapter 10
R-Brackets and r-Brackets

The theory of integrable systems gave birth to interesting constructions, which yield
new Poisson structures on a given (finite-dimensional) 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 R-matrix, or as an element
r ∈ g ⊗ g, in which case r is called an r-matrix. In their basic form, the new Poisson
structures are Lie–Poisson structures, so it will be assumed that the reader is familiar
with the basic theory of Lie–Poisson structures, as developed in Chapter 7.
For a given linear map R : g → g, which satisfies a condition which will be ex-
plained 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 R-bracket on g∗ . Upon identifying g with g∗ , which is usu-
ally done by using a non-degenerate symmetric bilinear form, the R-bracket on g∗
becomes a Lie–Poisson structure on g.
A tensor r ∈ g ⊗ g can be viewed as a zero-chain on g, with values in g ⊗ g, so
its coboundary δL0 (r) : g → g ⊗ g admits as transpose a linear map γr : g∗ ⊗ g∗ → g∗ .
When r satisfies a condition, which will also be explained in detail, γr defines a Lie
algebra structure on g∗ , hence leads directly to a Lie–Poisson bracket on g, i.e., no
identification of g with g∗ is needed to define this Lie–Poisson structure on g. This
Lie–Poisson bracket is called an r-bracket.
Section 10.1 is devoted to the construction of the R-bracket 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 r-brackets and we show how R-brackets and
r-brackets are related. The construction of higher degree brackets, from an R-matrix
or r-matrix, is given in Section 10.3. For the connection with Poisson–Lie groups,
we refer to Chapter 11.
In this chapter, F is an arbitrary field of characteristic zero, g is a finite-
dimensional Lie algebra over F and F (g) denotes the algebra of polynomial func-
tions on g, but it may also be taken as the algebra of smooth, respectively holomor-
phic, functions on g, when F = R, respectively when F = C.

C. Laurent-Gengoux et al., Poisson Structures, 269

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 10, © Springer-Verlag Berlin Heidelberg 2013
270 10 R-Brackets and r-Brackets

10.1 Linear R-Brackets

As we have seen in Chapter 7, a Lie algebra structure on a finite-dimensional vector

space leads to a Poisson structure on its dual vector space. We consider in this sec-
tion Lie algebras which are equipped with a second Lie algebra structure, obtained
from the original one by using an R-matrix, as explained below. In this case, the dual
vector space comes equipped with a second Lie–Poisson structure, whose proper-
ties are in general quite different from the properties of the original Lie–Poisson
structure.

10.1.1 R-Matrices and the Yang–Baxter Equation

We first give the general definition of an R-matrix and discuss its relation with the
Yang–Baxter equation.
Definition 10.1. Let (g, [· , ·]) be a Lie algebra and let R : g → g be a linear map. If
the skew-symmetric bilinear map [· , ·]R : g × g → g, defined for x, y ∈ g by

1
[x, y]R := ([Rx, y] + [x, Ry]) , (10.1)
2
defines a Lie bracket on g, then R is called an R-matrix for g. Then g, equipped with
the two Lie algebra structures [· , ·] and [· , ·]R , is called a double Lie algebra.
The standard terminology R-matrix 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 R-matrix 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 R-
matrix 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]) . (10.3)

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)
10.1 Linear R-Brackets 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 R-matrix 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 Yang–Baxter 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 Yang–Baxter
equation. Sometimes one adds the adjective classical, to distinguish these equations
from the quantum Yang–Baxter equation. We warn the reader that the terminol-
ogy Yang–Baxter equation, with or without adjectives, is used for many equations
which are closely related to, but not equivalent to, the above Yang–Baxter 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 R-matrix for g, starting from a given R-matrix for g, as given in the following
proposition.
Proposition 10.3. Let (g, [· , ·]) be a Lie algebra and let R be an R-matrix for g. If
T : g → g is an intertwining operator of g, then the linear map R ◦ T is an R-matrix
for g.

Proof. Suppose that R is an R-matrix 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 BR◦T (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 equiva-
lent to
T [BR (T x, Ty) , z] +  (x, y, z) = 0 , (10.6)
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 α ∈ F and replace in (10.6) T by
T + α 1g . The identity

(T + α 1g ) [BR (T x + α x, Ty + α y) , z] +  (x, y, z) = 0 (10.7)

holds because T + α 1g is also an intertwining operator. Since (10.7) is a cubic poly-

nomial in α and since it holds for all α ∈ F, all of its coefficients as a polynomial
in α are equal to zero. This gives the following equations:
272 10 R-Brackets and r-Brackets

[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 Yang–Baxter equa-

tion, then the R-matrix R ◦ T is also a solution of the Yang–Baxter equation, since
BR◦T (x, y) = BR (T x, Ty) for all x, y ∈ g. However, the corresponding property for
solutions of the modified Yang–Baxter equation does not hold in general.

10.1.2 R-Matrices and Lie Algebra Splittings

The main, and historically first, example of an R-matrix 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 R-matrix 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 Yang–Baxter 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 R-matrix for g.

Proof. Let g = g+ ⊕ g− be a Lie algebra splitting and let R := P+ − P− be the dif-

ference 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 R-Brackets 273

Notice that, when R = P+ − P− comes from a Lie algebra splitting, then an alterna-
tive formula for the R-bracket [· , ·]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 sec-
tion. 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 Yang–Baxter equation (10.4), with c = 1.

10.1.3 Linear R-Brackets on g∗ and on g

If we assume that the Lie algebra (g, [· , ·]) is finite-dimensional, every R-matrix for g
leads to a Lie–Poisson structure on g∗ , called an R-bracket; the trivial case R = 1g
is usually excluded, because it corresponds to the canonical Lie–Poisson structure
on g∗ . According to (7.5), it is given for all F, G ∈ F (g∗ ) at ξ ∈ g∗ by
   
{F, G}R (ξ ) := ξ , dξ F, dξ G R
1       
= ξ , R dξ F , dξ G + dξ F, R dξ G ,
2
where we recall that dξ 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 (ξ ) = ξ , dξ F + , dξ G + − ξ , dξ F − , dξ G − . (10.10)

Upon using, as in Section 7.2, a non-degenerate symmetric bilinear form · | · on g,

we can transfer the R-bracket to g, which leads according to (7.10) for a general
R-matrix R to
274 10 R-Brackets and r-Brackets

1
{F, G}R,g (x) = 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 R-bracket 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 finite-dimensional Lie algebra, equipped with
a non-degenerate symmetric bilinear form · | ·, and let R be the R-matrix 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 F̃ + , ∇x G̃ + , (10.12)

where F̃ and G̃ 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 R-bracket 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 R-brackets.
Example 10.7. Consider the Lie algebra splitting of g := gld (F), given by

g = Δd ⊕ Δd< = g+ ⊕ g− ,

where g+ := Δd (respectively g− := Δd< ) denotes the Lie algebra of upper trian-
gular matrices (respectively of strictly lower triangular matrices). Letting x | y :=
 < ⊥
Trace(xy), we have that g⊥ − = Δd = Δd , the vector space of all lower triangular
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 non-zero entry is a 1 at position ( j, i). According
to (10.12), the Poisson bracket on Δd is therefore given by

xi j , xk − (x) = x | [E ji , Ek ] ,

where j  i and   k.
10.2 Linear r-Brackets 275

This example is easily generalized to arbitrary semi-simple 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) =
Δd ⊕ g− , where g− stands for the Lie subalgebra of skew-symmetric matrices. Us-
ing 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 r-Brackets

In the previous section, we have seen that a linear map R : g → g leads, under certain
conditions, to a new Lie–Poisson 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
Lie–Poisson structure on g.

10.2.1 Coboundary Lie Bialgebras and r-Matrices

Let r be an element of g ⊗ g, where (g, [· , ·]) is a finite-dimensional 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 R-Brackets and r-Brackets

adx Y = 0, for all x ∈ g, are called ad-invariant; for example, elements of degree one
are ad-invariant 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 rep-
resentation 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 skew-symmetric p-linear maps on g, with values in
the representation space g ⊗ g. According to Section 4.1.1, C• (g) carries a cobound-
ary operator δL : C• (g) → C•+1 (g). Applied to r ∈ g ⊗ g  C0 (g), we obtain a linear
map γr := δL0 (r) : g → g ⊗ g, which is given by

γr (x) = adx r ,

for all x ∈ g. The transpose to γr is a linear map γr : g∗ ⊗ g∗  (g ⊗ g)∗ → g∗ , which

we view as a bilinear map g∗ × g∗ → g∗ , still denoted by γr . Explicitly,
 
γr (ξ , η ), x = (ξ ⊗ η )(adx r) (10.14)

for all x ∈ g and ξ , η ∈ g∗ . We will be interested in the case in which the bilinear
map γr defines a Lie algebra structure on g∗ .
Definition 10.9. Let (g, [· , ·]) be a finite-dimensional Lie algebra. A bivector r ∈
g ⊗ g is called an r-matrix for g if the transpose γr of γr = δL0 (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
[ξ , η ]r , x := ξ ∧ η , adx r (10.15)
for all ξ , η ∈ 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
Poisson–Lie groups, see Chapter 11.
We will give necessary and sufficient conditions on r ∈ g⊗g so that (g, [· , ·], [· , ·]r )
is a coboundary Lie bialgebra. Our proof will be based on the following proposition.

Proposition 10.10. Let φ : V → V ∧ V be a linear map, where V is a finite-

dimensional vector space. Denote by φ̃ : ∧•V → ∧•+1V the extension of φ to a
graded derivation of ∧•V of degree 1 and let φ  : V ∗ ∧ V ∗ → V ∗ denote the trans-
pose map to φ . Then φ  satisfies the Jacobi identity (so that φ  defines a Lie algebra
structure on V ∗ ) if and only if φ̃ ◦ φ̃ = 0.
Proof. Recall from Section A.5 in Appendix A, in particular Example A.3, that φ
extends uniquely to a derivation φ̃ 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 ξ1 , ξ2 ∈ g∗ and for all x1 , x2 ∈ g, by ξ1 ∧ ξ2 , x1 ∧ x2  =

ξ1 (x1 )ξ2 (x2 ) − ξ1 (x2 )ξ2 (x1 ).
10.2 Linear r-Brackets 277

φ̃ (X ∧Y ) = φ̃ (X) ∧Y + (−1) p X ∧ φ̃ (Y ) , (10.16)

for X ∈ ∧ pV and Y ∈ ∧•V . It follows from (10.16) that φ̃ ◦ φ̃ = 0 if and only if
φ̃ (φ (v)) = 0 for all v ∈ V . Now, for all ξ , η , ζ ∈ V ∗ ,
 (∗)
 
φ̃ (φ (v)), ξ ∧ η ∧ ζ = φ (v), φ  (ξ ∧ η ) ∧ ζ +  (ξ , η , ζ )
   
= v, φ  φ  (ξ ∧ η ) ∧ ζ +  (ξ , η , ζ ) ,

so that φ̃ ◦ φ̃ = 0 if and only if φ  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 ξ , η , ζ ∈ V ∗ , that

φ̃ (X), ξ ∧ η ∧ ζ
= φ (v) ∧ w − φ (w) ∧ v, ξ ∧ η ∧ ζ 
= φ (v), ξ ∧ η  w, ζ  − φ (w), ξ ∧ η  v, ζ  +  (ξ , η , ζ )
   
= v, φ  (ξ ∧ η ) w, ζ  − w, φ  (ξ ∧ η ) v, ζ  +  (ξ , η , ζ )
 
= v ∧ w, φ  (ξ ∧ η ) ∧ ζ +  (ξ , η , ζ )
 
= X, φ  (ξ ∧ η ) ∧ ζ +  (ξ , η , ζ ) .

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 skew-symmetric) 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 finite-dimensional Lie algebra and let r be an
element of g ⊗ g. Then r is an r-matrix for g if and only if r satisfies the following
two conditions:
(1) r+ is ad-invariant;
(2) [[r− , r− ]] is ad-invariant.
In this case, the Lie bracket associated with r is given by

[ξ , η ]r , x = ξ ∧ η , adx r− , (10.17)

for all ξ , η ∈ g∗ and x ∈ g.

Proof. Let ξ , η ∈ g∗ and let x ∈ g. Then
 
γr (ξ , η ) + γr (η , ξ ), x = (ξ ⊗ η )(adx r) + (η ⊗ ξ )(adx r) = 2(ξ ⊗ η )(adx r+ ) ,
(10.18)
278 10 R-Brackets and r-Brackets

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 γr is skew-symmetric
if and only if r+ is ad-invariant.
Suppose now that r+ is ad-invariant. Then γr (x) = adx r = adx r− = γr− (x) for
all x ∈ g, so that γr depends only on a := r− . Thus we are given a skew-symmetric
element a ∈ g ∧ g and we wish to prove that the Jacobi identity for γa is equivalent
to the ad-invariance of [[a, a]]. In view of Proposition 10.10, it suffices to show that
the ad-invariance of [[a, a]] is equivalent to γ̃a ◦ γ̃a = 0 (on ∧• g; equivalently, on g).
Combining (10.13) and (3.46), we find that

γa (x) = adx a = [[x, a]] = −[[a, x]] ,

for all x ∈ g. Since γ̃a and −[[a, ·]] are both graded derivations of degree 1 of ∧• g, it
follows that γ̃a (X) = −[[a, X]], for all X ∈ ∧• g. Therefore, γ̃a ◦ γ̃a = 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 ad-invariance of [[a, a]]. 

The simplest way to satisfy both conditions in Proposition 10.11 is to demand that
r is skew-symmetric and satisfies [[r, r]] = 0.

10.2.2 Linear r-Brackets on g

Suppose that r ∈ g ⊗ g is an r-matrix for a finite-dimensional Lie algebra (g, [· , ·]).

Recall from Definition 10.9 and Proposition 10.11 that this means that the bilinear
map, defined for x ∈ g and ξ , η ∈ g∗ by

[ξ , η ]r , x := ξ ∧ η , adx r = ξ ∧ η , 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 finite-dimensional Lie algebra and let r be an
r-matrix for g. Let (e1 , . . . , ed ) be a basis for g and let (ξ1 , . . . , ξd ) be the dual basis.
The Poisson matrix of {· , ·}r at x ∈ g, with respect to the coordinates ξ1 , . . . , ξd , is
the matrix of adx r− in the basis (e1 , . . . , ed ).
10.2 Linear r-Brackets 279

Proof. By definition, the matrix of adx r− ∈ g ∧ g in the basis (e1 , . . . , ed ) is the

skew-symmetric 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 com-
puted by 
ai j = ξi ∧ ξ j , adx r− = ξi , ξ j r (x) ,
as was to be shown. 

A compact formula for the collection of Poisson brackets ξi , ξ 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

 
Ξ ⊗, Ξ r (x) = 1 ⊗ x + x ⊗ 1, r− , (10.21)

where the left-hand side is the bivector, defined by

Ξ ⊗, Ξ r
(x) := ∑ ξi , ξ j r
(x) ei ∧ e j .
1i< jd

In this formula, (e1 , . . . , ed ) is a basis of g and (ξ1 , . . . , ξd ) is its dual basis. Equa-
tion (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 non-zero 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  ) = δ j,i δ,k Ei j ⊗ Ek ,
while
EiN+k, jN+ Ei N+k , j N+ = δ j,i δ,k EiN+k, j N+ .
2Since g is finite-dimensional, 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 R-Brackets and r-Brackets

10.2.3 From r-Matrices to R-Matrices

In this section, we show how R-matrices and r-matrices on a finite-dimensional

quadratic Lie algebra g are related. Denoting the bilinear form of g by · | ·, we
have an isomorphism χ : g → g∗ , defined by χ (x), y := x | y, for all x, y ∈ g. This
leads also to an isomorphism

χ ⊗ 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 skew-symmetric). Obviously, R can be decomposed uniquely as R =
R+ + R− , where R+ is symmetric and R− is skew-symmetric.
Proposition 10.13. Let (g, [· , ·]) be a finite-dimensional quadratic Lie algebra and
let χ : 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(χ ⊗ 1g )(r) (10.22)

(or, equivalently, r = 14 (χ −1 ⊗ 1g )(R)). Then, R is a skew-symmetric R-matrix for g

if and only if r is a skew-symmetric r-matrix for g. In this case, χ : (g, [· , ·]R ) →
(g∗ , [· , ·]r ) is a Lie algebra isomorphism, where [· , ·]R is the Lie bracket on g, asso-
ciated with R, and [· , ·]r the Lie bracket on g∗ , associated with r.

Proof. We decompose r = ∑α sα ⊗ tα , which we write in this proof as sα ⊗ tα in

order to simplify the formulas: in the sequel of the proof, a sum over α is implicit
in all formulas which contain α . Using this notation, it follows from (10.22) that
Rx = 4 sα | xtα for x ∈ g. It implies that R∗ x = 4 tα | x sα and hence that r is skew-
symmetric if and only if R∗ = −R, i.e., R is also skew-symmetric. Let us suppose
that R (and hence r) is skew-symmetric. We show that

χ ([x, y]R ) = [χ (x), χ (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 = [χ (x), χ (y)]r , z . (10.24)
2
With the above formulas for R, the left-hand side in (10.24) can be written as
10.2 Linear r-Brackets 281

2 sα | x [tα , y] + tα | y [sα , x] | z = 2(sα | x [tα , y] | z + tα | y [sα , x] | z) ,

(10.25)
while the right-hand side in (10.24) is given, in view of (10.19), by

[χ (x), χ (y)]r , z = χ (x) ∧ χ (y), adz r−
= χ (x) ∧ χ (y), [z, sα ] ⊗ tα + sα ⊗ [z,tα ]
= x | [z, sα ] y |tα  + x | sα  y | [z,tα ]
− y | [z, sα ] x |tα  − y | sα  x | [z,tα ] ,

which is the same as (10.25), because · | · is ad-invariant and r is skew-symmetric.

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 χ 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 R-matrix (respectively r is an r-matrix). 

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) = χ (x) ∧ χ (y) ∧ χ (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 Yang–Baxter equation.

Example 10.14. Let (g, [· , ·]) be a finite-dimensional Lie algebra, equipped with a
non-degenerate 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
R-matrix. 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 skew-symmetric. Let (ε1 , . . . , εd ) be a basis

of g+ . There exists
 a (unique) basis (e1 , . . . , ed ) of g− which is dual to it with respect
to · | ·, i.e., ei | ε j = δi, j for i, j = 1, . . . , d; indeed, both g+ and g− are isotropic
and · | · is non-degenerate. The endomorphism R can be described, for all x ∈ g, by
d d
R(x) = ∑ ei | x εi − ∑ εi | x ei ,
i=1 i=1

so that R = ∑di=1 χ (ei ) ⊗ εi − ∑di=1 χ (εi ) ⊗ ei , as an element of g∗ ⊗ g. It follows that

the r-matrix r := 14 (χ −1 ⊗ 1g )(R), associated with R, is given by
282 10 R-Brackets and r-Brackets

1 d
r= ∑ e i ∧ εi .
2 i=1
(10.26)

Example 10.15. The above example can be easily generalized to the case of a de-
composition 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 con-
tained in the normalizer of g+ and of g− . Moreover, it is assumed that g comes
equipped with a non-degenerate 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 projec-
tions of x on g+ , g0 and g− respectively, is a solution of the modified Yang–Baxter
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 skew-symmetric. Let (ε1 , . . . , εd ) 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
d d
R= ∑ χ (ei ) ⊗ εi − ∑ χ (εi ) ⊗ ei ,
i=1 i=1

1 d
r = ∑ e i ∧ εi .
2 i=1

Example 10.16. Starting from a particular case of Example 10.15, we construct an

R-matrix for sud and we compute the corresponding r-matrix. 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 R-matrix√associated to the above de-
composition, given in Example 10.15 and let R := 2 −1R, which is an R-matrix
for sld (C), because R is a multiple of R. Notice that R leaves sud invariant, since
√ √
x+x = 0 implies that x+ +x−  = 0, which in turns yields −1Rx+ −1(Rx) = 0.

It follows that R can be restricted to sud and is an R-matrix for sud . We con-
sider the symmetric bilinear form on sud , given by (x, y) → x | y := ℜ(Trace(xy)).
It is non-degenerate, hence induces an isomorphism χ between sud and its dual,
and R is skew-symmetric with respect to · | ·. Define, for all 1  i < j  d,
the skew-symmetric matrices Ai j := √12 (Ei j − E ji ), the symmetric matrices Si j :=
√ √
√−1 (Ei j
+ E ji ) and the diagonal matrices Hi := √−1 (Eii − Ei+1,i+1 ). These matri-
2 2
ces form a basis of sud , with the set of all Si j and Ak forming an orthogonal set,
10.3 R-Brackets and r-Brackets 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 ∑ χ (Si j ) ⊗ Ai j − 2 ∑ χ (Ai j ) ⊗ Si j .
1i< jd 1i< jd

As a consequence, the corresponding r-matrix r is given by

r = ∑ Si j ∧ Ai j .
1i< jd

10.3 R-Brackets and r-Brackets of Higher Degree

In this section we show that, under some assumptions, an R-matrix 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 finite-dimensional 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 non-degenerate symmetric bilinear map · | ·, which
satisfies, for all x, y, z ∈ g,
xy | z = x | yz . (10.27)
Notice that this property implies that · | · is ad-invariant.

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 R-bracket.

Proposition 10.18. Let (g, [· , ·]) be a finite-dimensional Lie algebra, associated
with an associative algebra, equipped with a non-degenerate symmetric bilinear
map · | ·, which satisfies xy | z = x | yz, for all x, y, z ∈ g. Let R : g → g be an ar-
bitrary linear map. If R and R− are solutions of the modified Yang–Baxter 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)
(10.28)
− 12 [x, ∇x G] | R(x ∇x F + ∇x F x) ,
284 10 R-Brackets and r-Brackets

is a quadratic Poisson structure on g. If H is an Ad-invariant function on g, then its

Hamiltonian vector field with respect to this Poisson structure is given by

ẋ = [R(x ∇x H), x] . (10.29)

Proof. It is clear that {· , ·}Q

R , as defined by (10.28), is a quadratic skew-symmetric
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
R , F3 (x)+  (1, 2, 3) = 0 . (10.30)
R

In order to organize this verification, it is useful to write R = R+ + R− , where R+

is symmetric and R− is skew-symmetric. 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 ,

(10.31)
{F, G}− (x) = R− (∇x F x) | ∇x G x − R− (x ∇x F) | x ∇x G ,

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 ,
(10.32)
∇x {F1 , F2 }− = R− ( f1 x) f2 − R− ( f2 x) f1 + f1 R− (x f2 ) − f2 R− (x f1 ) .

Using in two subsequent steps that R− is skew-symmetric and then that R− satisfies
the modified Yang–Baxter 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 R-Brackets and r-Brackets 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 B∗R : g ⊗ g → g by requiring that for all
x, y, z ∈ g,
BR (x, y) | z = x | B∗R (y, z) . (10.36)
Explicitly, B∗R is given, for x, y ∈ g, by

B∗R (x, y) = R∗ [Rx, y] − R∗ [x, R∗ y] − [Rx, R∗ y] .

Notice that

B∗R (y, x) − B∗R (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) + B∗R (y, x) − B∗R (x, y)
2 2 (10.37)
= [R+ x, R+ y] − R+ [R− x, y] − R+ [x, R− y] ,
which is the left-hand 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 Yang–Baxter equation (with the same con-
stant c) which, combined with (10.36) yields that B∗R (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 R-Brackets and r-Brackets

Finally, let H be an Ad-invariant function on g. We show that its Hamiltonian

vector field is given by the Lax equation (10.29). To do this, recall from Sec-
tion 5.1.4 that the Ad-invariance of H implies that [∇x H, x] = 0 for all x ∈ g, hence
that 12 (x∇x H + ∇x Hx) = x∇x H. Therefore, if F ∈ F (g) and x ∈ g, then

1
dx F, (XH )x  = {F, H}Q
R (x) = [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] ,

which proves (10.29). 

If, in addition to the hypotheses of Proposition 10.18, R is assumed to be skew-

symmetric, then the formula (10.28) for the quadratic bracket takes the simpler form

R (x) = R(∇x F x) | ∇x G x − R(x ∇x F) | x ∇x G ,

{F, G}Q (10.38)

for all F, G ∈ F (g). In the particular case where R comes from an r-matrix 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− = ∑α sα ⊗ tα and use, as in the proof of
Proposition 10.13, the convention that a sum over α is implicit in all formulas which
contain α . Recall that with this notation,

Rx | y = 4 sα | x tα | y , (10.39)

for all x, y ∈ g. For F, G ∈ F (g), the bracket (10.38) can be rewritten, using (10.39),
as

R (x) = 4 sα | ∇x F x tα | ∇x G x − 4 sα | x ∇x F tα | x ∇x G

{F, G}Q
= 4 ∇x F | xsα  ∇x G | xtα  − 4 ∇x F | sα x ∇x G |tα x
= 4 dx F, xsα  dx G, xtα  − 4 dx F, sα x dx G,tα x
= 2 dx F ∧ dx G, xsα ⊗ xtα − sα x ⊗ tα x
= 2 dx F ∧ dx G, [x ⊗ x, sα ⊗ tα ]
  
= 2 dx F ∧ dx G, x ⊗ x, r− .

We choose a basis (e1 , . . . , ed ) of g, with dual basis (ξ1 , . . . , ξd ) of g∗ . Then

Q  Q
ξi , ξ j R (x) = 2 ξi ∧ ξ j , [x ⊗ x, r− ] , so that ξi , ξ j R (x) is twice the coefficient
of ei ⊗ e j in [x ⊗ x, r− ]. It follows that, if we define

∑
Q Q
Ξ ⊗, Ξ r
(x) := ξi , ξ j R
(x) ei ∧ e j ,
1i< jd

then we find the following compact formula

 
(x) = 2 x ⊗ x, r−
Q
Ξ ⊗, Ξ r
, (10.40)
10.3 R-Brackets and r-Brackets 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 R-bracket, the context is the same as in the case of the quadratic
R-bracket, but the assumptions are weaker.

Proposition 10.19. Let (g, [· , ·]) be a finite-dimensional Lie algebra, associated

with an associative algebra, equipped with a non-degenerate 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 Yang–Baxter 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 Ad-invariant function on g, then its

Hamiltonian vector field with respect to this Poisson structure is given by

ẋ = [R(x∇x H x), x] .

Proof. The proposition can be proved in the same way as in the case of the quadratic
R-bracket (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 skew-symmetric 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 func-
tions 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 x∇x {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 R-brackets 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 R-Brackets and r-Brackets

(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 non-degenerate 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 skew-symmetric part R−
is also a solution of the modified Yang–Baxter equation, with the same constant c.
Then the linear, quadratic and cubic Poisson structures on g, associated with R,
namely {· , ·}R,g , {· , ·}Q C
R and {· , ·}R , are related by Lie derivatives, as indicated in
the following commutative diagram:
−L2 −L2 L2
{· , ·}R,g {· , ·}Q
R {· , ·}CR 0

−L1 L1 (10.42)

1L 1L
L0 2 0 2 0
0 {· , ·}R,g {· , ·}Q
R {· , ·}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
R . Let F1 , F2 be linear functions on g and let f 1 := ∇F1
that −L2 {· , ·}R,g = {· , ·}Q
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
R = −V2 {F1 , F2 }R,g + {V2 [F1 ], F2 }R,g + {F1 , V2 [F2 ]}R,g . (10.43)

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,
d
∇x (V2 [F]) | y = 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 for-
mulas are only needed for the verification of the other cases. Since {F1 , F2 }R,g is a
linear function on g, the right-hand 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 Sec-
tion 3.3 that, in terms of the Schouten bracket, two Poisson structures π and π  are
compatible if [π , π  ]S = 0, while the Lie derivative with respect to a vector field V
is expressed as LV π = [V , π ]S . Clearly, [LV π , π ]S = [[V , π ]S , π ]S = 0, since [·, π ]S
is a coboundary operator (since π is a Poisson structure, see Section 4.1.2). Thus,
on the one hand, {· , ·}R,g and {· , ·}Q 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 πR , πRQ and πRC and compute, using the
graded Jacobi identity for the Schouten bracket (Proposition 3.7),
 
πR , πRC S
= − πR , V2 , πRQ = − V2 , πRQ , πR − πRQ , [πR , V2 ]S =0,
S S S S S

where we used in the last step that πRQ and πR are compatible, and that [V2 , πR ]S =
−πRQ (which is a Poisson structure!). 

10.4 Notes

R-matrices and r-matrices have their origins in the theory of integrable systems,
but also play a fundamental rôle in the theory of quantum groups. They first ap-
peared in the contexts of the Adler–Kostant–Symes theorem, exhibiting the Lie al-
gebraic structures which underlie the integrability of both the Toda lattices and the
Korteweg–de Vries equation, see Adler [2].
The term “quantum group” is a generic name which is used for several types
of non-commutative 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 R-Brackets and r-Brackets

general constructed from solutions of the quantum Yang–Baxter equation, just as

Poisson–Lie 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 r-matrix as its first order term, while Poisson–Lie 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 r-matrix has been generalized to the general context of Lie
algebroids. The resulting object is called a (classical) dynamical r-matrix; it depends
on parameters, turning some of the conditions which it has to satisfy into differential
equations. See Etingof–Varchenko [69, 70] and Liu–Xu [130].
Chapter 11
Poisson–Lie Groups

A Poisson–Lie group (G, π ) is a Lie group G which is equipped with a Poisson

structure π . 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 mor-
phism 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 (left-invariant) 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 Poisson–Lie group (G, π ), 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 Poisson–Lie group, there corresponds a Lie bialgebra. Moreover,
to every homomorphism of Poisson–Lie groups there naturally corresponds a homo-
morphism of Lie bialgebras, hence the correspondence between Poisson–Lie groups
and Lie bialgebras is functorial. Most importantly, the converse also holds. As we
know from Lie’s third theorem, every finite-dimensional Lie algebra is the Lie alge-
bra of some Lie group, which can be chosen to be connected and simply connected.
This Lie group can be made into a Poisson–Lie group if the Lie algebra is a Lie
bialgebra.
Poisson–Lie 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 Poisson–Lie groups is discussed in Section 11.3. Us-
ing dressing actions, we study the symplectic leaves of Poisson–Lie groups in Sec-
tion 11.4.
All Lie groups in this chapter are real (F = R) or complex (F = C).

C. Laurent-Gengoux et al., Poisson Structures, 291

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 11, © Springer-Verlag Berlin Heidelberg 2013
292 11 Poisson–Lie Groups

11.1 Multiplicative Poisson Structures and Poisson–Lie 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 π on a (real or complex) Lie group G is said
to be multiplicative if the product map μ : G × G → G is a Poisson map, where
G × G is endowed with the product Poisson structure. The pair (G, π ) is then called
a Poisson–Lie group.
A map Φ : G → H between two Poisson–Lie groups (G, π ) and (H, π  ) is called
a Poisson–Lie group homomorphism if it is both a Poisson map and a group homo-
morphism.

11.1.1 The Condition of Multiplicativity

We give in the following proposition two useful characterizations of the multiplica-

tivity 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 π on G, the follow-
ing three conditions are equivalent:
(i) π is multiplicative;
(ii) For all g, h ∈ G:

πgh = ∧2 (Tg Rh ) πg + ∧2 (Th Lg ) πh ; (11.1)

(iii) The map Ψ : G → ∧2 g, which is defined for all g ∈ G by

Ψ (g) := ∧2 (Tg Rg−1 ) πg , (11.2)

is a cocycle of G, with respect to the adjoint representation of G on ∧2 g, i.e.,

Ψ (gh) = Ψ (g) + Adg Ψ (h) , (11.3)

for all g, h ∈ G.

Proof. We first prove that (i) and (ii) are equivalent, which amounts to proving that
μ : 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 Π ; according to (2.11), the value
of Π at (g, h) ∈ G × G is given by the bivector

Π(g,h) = ∧2 (Tg ıh ) πg + ∧2 (Th ıg ) πh , (11.4)

11.1 Multiplicative Poisson Structures and Poisson–Lie Groups 293

where ıh and ıg are the inclusions of G into G × G given by ıh : g → (g, h) and
ıg : h → (g, h) respectively. According to (1.32), μ is a Poisson map if and only if

πgh = ∧2 (T(g,h) μ ) Π(g,h) , (11.5)

for all g, h ∈ G. According to (11.4), and using the identities μ ◦ ıh = Rh and

μ ◦ ıg = Lg , the right-hand side of (11.5) is given by

∧2 (T(g,h) μ ) Π(g,h) = ∧2 (T(g,h) μ ) (∧2 (Tg ıh ) πg ) + ∧2 (T(g,h) μ ) (∧2 (Th ıg ) πh )

= ∧2 (Tg (μ ◦ ıh )) πg + ∧2 (Th (μ ◦ ıg )) πh
= ∧2 (Tg Rh ) πg + ∧2 (Th Lg ) πh ,

which is precisely the right-hand 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 ) πgh

= ∧2 (Tgh R(gh)−1 ) ∧2 (Tg Rh ) πg + ∧2 (Tgh R(gh)−1 ) ∧2 (Th Lg ) πh ,

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 Ψ and from the identity
R(gh)−1 ◦ Rh = Rg−1 , namely

∧2 (Tgh R(gh)−1 ) πgh = Ψ (gh) ,

∧2 (Tgh R(gh)−1 )(∧2 (Tg Rh ) πg ) = ∧2 (Tg Rg−1 ) πg = Ψ (g) .

For the third and final term, recall that the map Cg is conjugation by g, so that
Cg = Rg−1 ◦ Lg , leading to the identity R(gh)−1 ◦ Lg = Cg ◦ Rh−1 , which in turn yields

∧2 (Tgh R(gh)−1 )(∧2 (Th Lg ) πh ) = ∧2 (Th (R(gh)−1 ◦ Lg )) πh

= ∧2 (Th (Cg ◦ Rh−1 )) πh = ∧2 (TeCg )(∧2 (Th Rh−1 ) πh )
= Adg Ψ (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 1-cocycle)
and proving the one result on group cocycles which we will use.
294 11 Poisson–Lie Groups

Definition 11.3. Let ρ : G ×V → V be a representation of a Lie group G on a finite-

dimensional vector space V . A cocycle of G in V is a smooth (or holomorphic)
function Ψ : G → V , satisfying

Ψ (gh) = Ψ (g) + ρ (g, Ψ (h)) , (11.6)

for all g, h ∈ G.

Proposition 11.4. Let G be a connected Lie group, let ρ : G ×V → V be a represen-

tation of G on a finite-dimensional vector space V and let W ⊂ V be a G-invariant
subspace. Suppose that Ψ : G → V is a cocycle of G in V .
(1) If TeΨ (x) = 0 for all x ∈ g, then Ψ = 0;
(2) If TeΨ (x) ∈ W for all x ∈ g, then Ψ (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

TgΨ ◦ Te Lg = ρg ◦ TeΨ ,

so that the vanishing of TeΨ implies the vanishing of TgΨ ◦ Te Lg , and therefore
of TgΨ , for all g ∈ G. Since G is assumed to be connected, this implies that Ψ is
a constant function. Since Ψ (e) = 0, as follows from (11.6), Ψ vanishes at each
point, which shows (1). If W is a G-invariant subspace, then we have an induced
representation G × V /W → V /W and Ψ 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 TeΨ (x) ∈ W for all x ∈ g, which proves (2).
See [181] for more information on group cohomology.

11.1.2 Basic Properties of Poisson–Lie Groups

We start with an interesting feature of Poisson–Lie groups. Let (G, π ) be a Poisson–

Lie group with Lie algebra g. The Poisson structure π 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 rôle in the sequel of this chapter.
Proposition 11.5. Let (G, π ) be a Poisson–Lie group.
(1) π vanishes at e;
(2) The linearized Poisson structure of π at e is the linear Poisson structure π1 on
g = Te G, whose value at x ∈ g is given by (π1 )x = TeΨ (x), where Ψ : G → ∧2 g
is the cocycle, defined as in (11.2);
(3) The inverse map inv : G → G is an anti-Poisson map.

Proof. Substituting e for g and h in (11.1) amounts to πe = πe + πe , hence πe = 0,

which yields (1). According to Proposition 7.22, g = Te G inherits from π = {· , ·}
11.1 Multiplicative Poisson Structures and Poisson–Lie Groups 295

a linear Poisson structure π1 = {· , ·}1 . For functions F, G on g, defined in a neigh-

borhood 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, (π1 )x , (11.7)
where de F, de G and de {F, G} are viewed as (linear) functions on g = Te G, and (π1 )x
is identified with an element of ∧2 g. Alternatively, in terms of the map Ψ : 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, πg = dg F ∧ dg G, ∧2 (Te Rg )Ψ (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, TeΨ (x) , (11.8)
where we have used that Ψ (e) = 0. Comparing (11.7) with (11.8) leads to the desired
equality
(π1 )x = TeΨ (x) , (11.9)
proving (2). To say that inv is an anti-Poisson map means that inv : (G, π ) →
(G, −π ) is a Poisson map: for every g ∈ G, ∧2 (Tg inv) πg = −πg−1 . To prove that
the latter holds, let g ∈ G, substitute g−1 for h in (11.1), and use πe = 0, to find that

0 = ∧2 (Tg Rg−1 ) πg + ∧2 (Tg−1 Lg ) πg−1 .

Combined with Tg inv = −Te Lg−1 ◦ Tg Rg−1 (see (5.2)), this leads to

∧2 (Tg inv) πg = ∧2 (Te Lg−1 )(∧2 (Tg Rg−1 ) πg )

= − ∧2 (Te Lg−1 )(∧2 (Tg−1 Lg ) πg−1 )
= − ∧2 (Tg−1 (Lg−1 ◦ Lg )) πg−1
= −πg−1 ,

which proves that inv is an anti-Poisson map.

Comparing Definitions 5.32 and 11.1, we see that a Poisson structure π 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, π ) be a Poisson–Lie group. For every g ∈ G, the following
conditions are equivalent:
(i) πg = 0;
(ii) The map Lg : G → G, left translation by g, is a Poisson map;
1 We use here and below the natural pairing · , · between ∧ p g∗ and ∧ p g, introduced in (5.10).
296 11 Poisson–Lie Groups

(iii) 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,

πgh = ∧2 (Th Lg ) πh . (11.10)

In view of (11.1), which says that π is multiplicative, (11.10) is tantamount to

∧2 (Tg Rh ) πg = 0, which is equivalent to πg = 0, because the linear map Tg Rh is
an isomorphism. This proves the equivalence of (i) and (ii). The proof of the equiv-
alence of (i) and (iii) is similar.

11.1.3 Poisson–Lie Subgroups

We consider in this section the notion of a Poisson–Lie subgroup of a Poisson–Lie

group (G, π ). Recall that a Lie subgroup of G need not be a closed subset of G and
that there is associated to π a map Ψ : G → ∧2 g, defined for g ∈ G by Ψ (g) :=
∧2 (Tg Rg−1 ) πg .
Proposition 11.7. Let (G, π ) be a Poisson–Lie group. For a Lie subgroup H of G,
with Lie algebra h, the following conditions are equivalent:
(i) H is a Poisson submanifold of (G, π );
(ii) For every h ∈ H, the bivector Ψ (h) belongs to ∧2 h.
Under these conditions, the pair (H, π  ) is a Poisson–Lie group, where π  denotes
the restriction of π to H. One then says that H is a Poisson–Lie 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 TeΨ (x) belongs to ∧2 h.
Proof. In view of Proposition 2.12, H is a Poisson submanifold of G if and only if
πh ∈ ∧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 Poisson–Lie group.
Suppose now that H is connected. It is clear that (ii) implies (iii). Let us show
the inverse implication. Since Ψ : G → ∧2 g is a cocycle for the adjoint repre-
sentation of G in ∧2 g, its restriction to H is a cocycle for the adjoint represen-
tation of H in ∧2 g, which admits ∧2 h ⊂ ∧2 g as an H-invariant subspace. By as-
sumption, TeΨ (x) ∈ ∧2 h for all x ∈ h. According to Proposition 11.4, this implies
that Ψ (h) ∈ ∧2 h for all h ∈ H, which is condition (ii).
We show in the following proposition that the datum of a Poisson–Lie group and of a
Poisson–Lie subgroup leads, under a topological assumption, to a Poisson structure
on the (left) quotient space.
11.1 Multiplicative Poisson Structures and Poisson–Lie Groups 297

Proposition 11.8. Let (G, π ) be a Poisson–Lie group and let (H, π  ) 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, π ) such that
the canonical projection p : G → H\G is a Poisson map. Since the product map
μ : 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 finite-dimensional 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 finite-dimensional 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 π = {· , ·} 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, π 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, πv+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, πv
= dv (F ◦ Rw ) ∧ dv (G ◦ Rw ), πv (11.13)
 
= dv+w F ∧ dv+w G, ∧2 (Tv Rw ) πv .
298 11 Poisson–Lie Groups

Substituting (11.12) and (11.13) in (11.11) we conclude that π is a linear Poisson

structure on V if and only if

πv+w = ∧2 (Tv Rw ) πv + ∧2 (Tw Lv ) πw ,

for all v, w ∈ V . Since the latter condition is precisely condition (ii) in Proposi-
tion 11.2, this shows that π is linear if and only if π is multiplicative.

Since the dual of a finite-dimensional Lie algebra admits a canonical linear Pois-
son structure (the Lie–Poisson structure), it follows from the above proposition that
the dual of a finite-dimensional Lie algebra is in a canonical way a Poisson–Lie
group.

11.1.5 Coboundary Poisson–Lie Groups

An important class of Poisson–Lie groups are those constructed out of r-matrices.

Recall from Proposition 10.11 that if g is a Lie algebra and r is an r-matrix for g,
then its skew-symmetric part a := r− ∈ ∧2 g is such that [[a, a]] is ad-invariant. We
show in the following proposition that this condition admits a natural connection
with Poisson–Lie 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 right-invariant (re-
spectively left-invariant) 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 π , defined by π := ← −
a −→−a . The pair (G, π ) is
a Poisson–Lie group if and only if [[a, a]] is an Ad-invariant element. In particular,
if G is connected, then (G, π ) is a Poisson–Lie group if and only if a is an r-matrix.

Proof. It follows immediately from Proposition 5.2 that the Schouten bracket of
π := ←−
a −→ −
a with itself is given by

[π , π ]S = [←−
a −→ −a ,←−
a −→ −a ]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 π , as defined above, is
a Poisson structure on G if and only if [[a, a]] is Ad-invariant. It remains to be shown
that π is moreover multiplicative. The map Ψ : G → ∧2 g associated with π as in
item (iii) in Proposition 11.2 (i.e. defined, for all g ∈ G, by Ψ (g) := ∧2 (Tg Rg−1 ) πg )
is given by
Ψ (g) = Adg a − a , (11.14)
11.1 Multiplicative Poisson Structures and Poisson–Lie Groups 299

as one computes using the following description of the bivector field ←

−
a −−
→
a at a
point g ∈ G:
πg = ∧2 (Te Lg )a − ∧2 (Te Rg )a .
For every g, h ∈ G, we can therefore check that

Ψ (gh) = Adgh a − a = Adgh a − Adg a + Adg a − a = Ψ (g) + Adg Ψ (h) , (11.15)

which, in view of the equivalence of items (i) and (iii) in Proposition 11.2, implies
that π is multiplicative. Therefore, (G, π ) is a Poisson–Lie group if and only if
[[a, a]] is Ad-invariant. If G is connected, then this is, according to Proposition 5.3,
equivalent to [[a, a]] being ad-invariant, i.e., to a being an r-matrix.
The previous proposition leads to the following definition.
Definition 11.11. A Poisson–Lie group (G, π ) is said to be a coboundary Poisson–
Lie group if π is of the form π = ←
−
a −→
−
a for some element a ∈ ∧2 g.
For a given coboundary Poisson–Lie group (G, π ) there may exist two different
elements a1 , a2 ∈ ∧2 g such that π = ←a−1 − −
→
a1 and π = ←
a−2 − −
→
a2 . However, in such a
case, we have
a−−−−→ ←−−−−
1 − a2 = a 1 − a2

so that a1 and a2 differ by an Ad-invariant element in ∧2 g. Therefore, for a

given coboundary Poisson–Lie group, the element in ∧2 g which appears in Defi-
nition 11.11 is defined only up to addition of an Ad-invariant element in ∧2 g.
Remark 11.12. The expression (11.14) of Ψ in terms of a, namely that Ψ (g) =
Adg a − a for all g ∈ G, has the following cohomological interpretation: it says that
Ψ is a 1-coboundary (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
Poisson–Lie group, introduced above. Also, since every 1-coboundary is a cocycle,
the cohomological interpretation of Ψ yields an alternative proof of (11.15), which
shows that Ψ 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
finite-dimensional vector space V , equipped with a product μ , a notion which was
introduced in Section 8.2.1. For the present discussion, we will assume that the
product μ is associative and admits a unit e; we will therefore refer to (V, μ ) 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 μ (v, w) = e.
Proposition 11.13. Let (V, μ ) be a finite-dimensional associative algebra with unit,
and let Vinv denote the set of all invertible elements in V .
300 11 Poisson–Lie Groups

(1) Vinv is a (non-empty) Zariski open subset of V and the restriction of μ 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] = μ (x, y) − μ (y, x) for all x, y ∈ V ;
(2) If π is a multiplicative Poisson structure on (V, μ ), in the sense of Sec-
tion 8.2.1, then the restriction π  of π to Vinv is a multiplicative Poisson struc-
ture on the Lie group Vinv , making (Vinv , π  ) into a Poisson–Lie group;
(3) Let a ∈ ∧2V be a skew-symmetric r-matrix for V . The quadratic bivector
field π on V , whose value at x ∈ V is given by

πx := 2 [x ⊗ x, a] (11.16)

is a multiplicative Poisson structure on V . The Poisson–Lie group (Vinv , π  ),

where π  denotes the restriction of π to Vinv , is the coboundary Poisson–Lie
group (Vinv , 2(←
−
a −→−a )).

Proof. Since μ 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) := μ (v, w),
is an isomorphism, i.e.,

Vinv = {v ∈ V | det Lv = 0} .

Since μ is a bilinear map, v → Lv is a linear map, and hence v → det Lv is a poly-

nomial function on V (it is homogeneous of degree dimV ). It follows that Vinv is a
Zariski open subset of V ; it is non-empty since it contains the unit e of μ . Defining
Rv : V → V by Rv (w) := μ (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
μ (v, w) = e, then we have, in view of the associativity of μ ,

μ (μ (w, v), w) = μ (w, μ (v, w)) = w ,

so that Ker Rv = {0}, i.e., Rv is an isomorphism (recall that V is finite-dimensional).

Therefore, the invertible elements of (V, μ ) form a group, with e as unit. As an
open subset of V , the group (Vinv , μ ) inherits a manifold structure from V ; since μ
is a bilinear map, it restricts to a smooth map μ  on Vinv , hence (Vinv , μ  ) is a Lie
group. For x, y ∈ V  TeV , we need to show that their Lie bracket [x, y] is given by
μ (x, y) − μ (y, x). Under the canonical isomorphism V  TeV we have, by definition,
←−−−−−−−−−−−−
[x, y] = [←
−
x ,←−y ]e and (μ (x, y) − μ (y, x))e = μ (x, y) − μ (y, x). It therefore suffices to
show that
←−−−−−−−−−−
[←
−x ,←
−
y ] = μ (x, y) − μ (y, x) . (11.17)
Let F : V → F be a linear function. For v ∈ V we have that

←
− F(v + t μ (v, x)) − F(v)
x [F](v) = xe [F ◦ Lv ] = lim = F(μ (v, x)),
t→0 t
so that ←
−
x [F] = F ◦ Rx . It follows that
11.1 Multiplicative Poisson Structures and Poisson–Lie Groups 301

[←
−
x ,←
−
y ] [F] = ←
−
x [←
−
y [F]] − ←
−
y [←
−
x [F]]
= F ◦ (Ry ◦ Rx − Rx ◦ Ry )
= F ◦ Rμ (x,y)−μ (y,x)
←−−−−−−−−−−−−
= (μ (x, y) − μ (y, x))[F],

for all linear functions F on V , leading to (11.17). This shows (1).

Let π be a multiplicative Poisson structure on (V, μ ). Recall from Section 8.2.1
that this means that the product map μ is a Poisson map. Since Vinv is an open subset
of V , the restriction π  of π to Vinv is a Poisson structure and the restriction μ  of μ
to Vinv , i.e., the product map of the Lie group (Vinv , μ  ), is a Poisson map. This shows
that (Vinv , π  ) is a Poisson–Lie group, which is the content of (2).
Let a be a skew-symmetric r-matrix for V , which is the Lie algebra of the
connected Lie group (Vinv , μ ). According to Proposition 11.10, (Vinv , ← −
a −→−
a ) is
a coboundary Poisson–Lie group. We show that the quadratic bivector field2 π
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 left-invariant vector
field ←z of z ∈ V is given at x ∈ Vinv , by ←
− −
z x = xz. Similarly −→z x = zx. It follows that,
for a bivector z1 ∧ z2 ∈ ∧ V , 2

←−−− −−−→
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)

This shows our claim. Since π and 2(← −

a −− →
a ) have the same value at every point
of Vinv , the restriction of π to Vinv is multiplicative, hence π itself is multiplicative,
showing (3).
Example 11.14. We show that the bivector field on V = Matd (F), which we dis-
cussed in Example 8.22 is associated with an r-matrix, which proves that it is a mul-
tiplicative Poisson structure. Let V be equipped with the non-degenerate ad-invariant
bilinear form x | y = Trace(xy). We denote the elementary matrices of V by Ei j ,
and we introduce
  linear coordinates ξi j on V by defining, for a matrix x ∈ Matd (F),
ξi j (x) := E ji | x , so that ξi 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
r := ∑ Eab ∧ Eba
2 1a<bd
(11.19)

2We know from Section 10.3, in particular from the second Russian formula (10.40), that π is a
Poisson structure, but this fact is not used in the proof.
302 11 Poisson–Lie Groups

is an r-matrix on the Lie algebra Matd (F). With the help of this r-matrix, a Pois-
son structure π on Matd (F) can be constructed as in Proposition 11.13. Let us ex-
press the coefficients of the Poisson matrix of π with respect to the linear coordi-
nates ξi 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,
   
ξi j , ξk r (x) = 2 ξi j ∧ ξkl , [x ⊗ x, r]
 
= ∑ ξi j ∧ ξk , [x ⊗ x, Eab ∧ Eba ]
a<b
 
= ∑ ξi j ∧ ξk , xEab ∧ xEba − Eab x ∧ Eba x
a<b
= ∑ (δ j,b δ,a − δ j,a δ,b − δi,a δk,b + δi,b δk,a ) ξi (x)ξk j (x)
a<b
= (ε j, + εi,k ) ξi (x)ξk j (x) ,

where εi, 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 finite-dimensional 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, ξ ) are often abbreviated
to x and ξ . 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),
  
ξ , [x, y]g = − ad∗x ξ , y and [ξ , η ]g∗ , x = − η , ad∗ξ x , (11.20)

for all x, y ∈ g and for all ξ , η ∈ 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 δ : g → g ∧ g. Precisely, δ is defined by

ξ ∧ η , δ (x) = [ξ , η ]g∗ , x , (11.21)

for all x ∈ g, and for all ξ , η ∈ g∗ . We will use the natural bilinear form · | · d on d,
defined for all x, y ∈ g and for all ξ , η ∈ g∗ by
11.2 Lie Bialgebras 303

(x, ξ ) | (y, η ) d := ξ , y + η , x . (11.22)

It is clear that · | · d is symmetric and non-degenerate.

11.2.1 Lie Bialgebras

The following proposition will justify the definition of a Lie bialgebra.

Proposition 11.15. Let g be a finite-dimensional vector space. Suppose that [· , ·]g
and [· , ·]g∗ are Lie brackets on g and g∗ respectively. On d := g × g∗ , consider the
non-degenerate symmetric bilinear form · | · d , given by (11.22). Then the following
conditions are equivalent:
(i) There exists a Lie bracket on d, with respect to which · | · d is ad-invariant,
such that the natural inclusions g → d and g∗ → d are Lie algebra homo-
morphisms;
(ii) The skew-symmetric bilinear map [· , ·]d : d × d → d given, for all x, y ∈ g and
for all ξ , η ∈ g∗ , by

[(x, ξ ), (y, η )]d := ([x, y]g +ad∗ξ y−ad∗η x, [ξ , η ]g∗ +ad∗x η −ad∗y ξ ) , (11.23)

satisfies the Jacobi identity;

(iii) The transpose δ : g → ∧2 g of the Lie bracket [· , ·]g∗ : ∧2 g∗ → g∗ , defined by
(11.21), satisfies for x, y ∈ g the derivation property

δ [x, y]g = [[δ (x), y]] + [[x, δ (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 finite-dimensional vector space, equipped with Lie brack-
ets [· , ·]g and [· , ·]g∗ on g, respectively on g∗ . The skew-symmetric 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 ; (11.25)

(2) [· , ·]d satisfies the Jacobi identity if and only if for all x, y ∈ g → d, and for all
ξ , η ∈ g∗ → d:

[x, y]d | [ξ , η ]d d+ [y, ξ ]d | [x, η ]d d+ [ξ , x]d | [y, η ]d d =0. (11.26)

Proof. Let di = (xi , ξi ), 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 Poisson–Lie Groups
 
d1 | [d2 , d3 ]d d = ξ1 , [x2 , x3 ]g + [ξ1 , ξ2 ]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 ϕ denote the 4-linear map from d to itself, defined for all
d1 , . . . , d4 ∈ d by

ϕ (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
non-degenerate, ϕ = 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 ξ , η ∈ g∗ , if and only if ϕ = 0. Since Jd (d1 , d2 , d3 ) = 0 when
d1 , d2 , d3 all belong to g or all belong to g∗ , we have that ϕ (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
skew-symmetry of Jd , combined with (11.25), implies that ϕ is skew-symmetric.
Therefore, ϕ (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 ϕ is 4-linear, that ϕ = 0 if and only if
ϕ (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 ξ , η ∈ 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 sat-
isfied. 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 ex-
ists 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
[x, η ] = (− ad∗η x, ad∗x η ) , (11.28)
for all x ∈ g and all η ∈ g∗ . For (z, γ ) ∈ g × g∗ , we have in view of (11.25), combined
with the properties of [· , ·], assumed in (i), that

[x, η ] | (z, γ ) = [(x, 0), (0, η )] | (z, 0) + (0, γ ) d

d
 
= [z, x]g , 0 | (0, η ) + 0, [η , γ ]g∗ | (x, 0)
  d d

= η , [z, x]g + [η , γ ]g∗ , x

11.2 Lie Bialgebras 305
 
= (− ad∗η x, ad∗x η ) | (z, γ ) d .

Since · | · d is non-degenerate, 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
ξ , η ∈ 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
ξ , η ∈ g∗ :
  
ξ ∧ η , [[δ (x), y]] = ad∗y ξ , ad∗η x − ad∗y η , ad∗ξ x . (11.29)

Indeed, using (11.21 ), (11.29) and the definitions of [· , ·]d and · | · d , we find that

ξ ∧ η , δ ([x, y]g ) − [[δ (x), y]] − [[x, δ (y)]]
   
= [ξ , η ]g∗ , [x, y]g − ad∗y ξ , ad∗η x + ad∗y η , ad∗ξ x
  
+ ad∗x ξ , ad∗η y − ad∗x η , ad∗ξ y
= [x, y]d | [ξ , η ]d d+ [y, ξ ]d | [x, η ]d d+ [ξ , x]d | [y, η ]d d ,

for all x, y ∈ g, and for all ξ , η ∈ g∗ . In order to show (11.29), use (5.13) and the
derivation property of the coadjoint action of g on ∧2 g∗ :
   
ξ ∧ η , [[δ (x), y]] = − ξ ∧ η , ady δ (x) = ad∗y (ξ ∧ η ), δ (x)
 
= ad∗y ξ ∧ η + ξ ∧ ad∗y η , δ (x)

= ad∗y ξ , η g∗ + ξ , ad∗y η g∗ , x
  
= ad∗y ξ , ad∗η x − ad∗y η , ad∗ξ 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 finite-dimensional 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 (finite-dimensional)
Lie bialgebras. A linear map φ : g → h is called a homomorphism of Lie bialgebras
if both φ : (g, [· , ·]g ) → (h, [· , ·]h ) and its transpose φ  : (h∗ , [· , ·]h∗ ) → (g∗ , [· , ·]g∗ )
are Lie algebra homomorphisms.
306 11 Poisson–Lie Groups

Example 11.18. Let (g, [· , ·]g ) be a finite-dimensional 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 δ = 0.
Dually, let V be any finite-dimensional 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 bial-
gebra. 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 one-dimensional because g is two-
dimensional. 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 δ (x) = 0 and δ (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∗ = λ x∗ , (11.30)

for some λ ∈ F∗ . In this case δ (x) = λ x ∧ y and δ (y) = 0, so that we arrive at the
same conclusion.
Remark 11.20. For every λ , μ ∈ F and every Lie bialgebra (g, [· , ·]g , [· , ·]g∗ ), the
triple (g, λ [· , ·]g , μ [· , ·]g∗ ) is a Lie bialgebra. It is clear that it is isomorphic to the
original one if λ = μ ∈ F∗ , under the isomorphism λ 1g : g → g. However, in general,
it is not the case when λ = μ , even if both λ and μ are different from 0. For instance,
for two different values of the parameter λ , the Lie bialgebra structures described
by (11.30) are not isomorphic.

11.2.2 Lie Sub-bialgebras

The notion of a homomorphism of Lie bialgebras, given in Definition 11.17, leads

naturally to the notion of a Lie sub-bialgebra.
Definition 11.21. Let (g, [· , ·]g , [· , ·]g∗ ) be a finite-dimensional Lie bialgebra. A sub-
space h of g is called a Lie sub-bialgebra 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 deter-
mined 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 finite-dimensional Lie bialgebra. A

subspace h of g is a Lie sub-bialgebra of (g, [· , ·]g , [· , ·]g∗ ) if and only if the following
two conditions are satisfied:
(1) The vector space h is a Lie subalgebra of (g, [· , ·]g );
(2) The annihilator h⊥ of h is a Lie ideal of (g∗ , [· , ·]g∗ ).

Proof. For h a subspace of g, let ı : h → g denote the inclusion map, with transpose
ı : g∗ → h∗ . If a Lie sub-bialgebra 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, ı : 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 δ  of [· , ·]h∗ is the restriction of δ to h, item (iii) in Proposition 11.15
implies that these brackets define a Lie bialgebra structure on h. By construction, ı
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 finite-
dimensional 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, ξ ) ∈ g × g∗ by S(x, ξ ) := (ξ , 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 rôle, hence the first part of the proposition is clear. The isomor-
phism S : g × g∗ → g∗ × g which sends (x, ξ ) to (ξ , x) is then clearly a Lie alge-
bra 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 (finite-dimensional) 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 sub-bialgebra 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 Poisson–Lie Groups

11.2.4 Coboundary Lie Bialgebras and r-Matrices

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 r-matrix when the transpose of the map δ : 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 r-matrix 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 r-matrix
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 r-
matrix 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
δ : g → g ∧ g is the linear map, defined by

ξ ∧ η , δ (x) = [ξ , η ]r , x ,

for all x ∈ g and all ξ , η ∈ g∗ . According to (10.17),

 
[ξ , η ]r , x = ξ ∧ η , adx r− ,

so that δ (x) = adx r− , where r− denotes the skew-symmetric part of r. Using the
fact that ad is a Lie algebra representation of g (on g ∧ g),

δ ([x, y]g ) = ad[x,y]g r− = adx ady r− − ady adx r− = adx δ (y) − ady δ (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 Poisson–Lie group.
Proposition 11.25. Let (g, [· , ·]g , [· , ·]g∗ ) be a finite-dimensional Lie bialgebra. On
its double (d, [· , ·]d ), there exists a skew-symmetric r-matrix a ∈ ∧2 d, with corre-
sponding Lie bracket [· , ·]a on d∗ , such that (g, [· , ·]g , [· , ·]g∗ ) is a Lie sub-bialgebra
of (d, [· , ·]d , [· , ·]a ). In terms of a pair of dual bases (e1 , . . . , ed ) and (ε1 , . . . , εd ) for
g and for g∗ respectively, a is given by a = 12 ∑di=1 ei ∧ εi .
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 R-matrix of d is given by R(x, ξ ) = ξ − x, for (x, ξ ) ∈ d. It
is skew-symmetric because both g and g∗ are isotropic subspaces of d with re-
spect to the bilinear form · | · d (see (11.22)). Using the isomorphism between d
and its dual d∗ , induced by · | · d , the R-matrix becomes an element a ∈ g ⊗ g,
which is according to Proposition 10.13 a skew-symmetric r-matrix, whose asso-
ciated 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 (ε1 , . . . , εd ) for g and for g∗ respectively, by a = 12 ∑di=1 ei ∧ εi . We show that
(d, [· , ·]d , [· , ·]a ) contains (g, [· , ·]g , [· , ·]g∗ ) as a Lie sub-bialgebra. The natural inclu-
sion ı : g  g × {0} → d is, by definition of the bracket on d, a Lie algebra homo-
morphism (g, [· , ·]g ) → (d, [· , ·]d ). Under the natural identification between d∗ and
g∗ × g, the transpose ı to ı is given by ı (ξ , x) = ξ , for (ξ , x) ∈ g∗ × g. Therefore,
showing that ı is also a Lie algebra homomorphism, amounts to showing that

[(ξ , y), (η , z)]a , (x, 0) = [ξ , η ]g∗ , x ,

for all (ξ , y), (η , z) ∈ g∗ × g and for all x ∈ g. This is done by explicitly computing
the left-hand side of the latter equation, namely

[(ξ , y), (η , z)]a , (x, 0)

 
d
1
= (ξ , y) ∧ (η , z), ad(x,0) ∑ ei ∧ εi
2 i=1

1 d
= ∑ (ξ , y) ∧ (η , z), [(x, 0), (ei , 0)]d ∧ (0, εi ) + (ei , 0) ∧ [(x, 0), (0, εi )]d
2 i=1
1 d 
= ∑ (ξ , y) ∧ (η , z), ([x, ei ]g , 0) ∧ (0, εi ) + (ei , 0) ∧ (− ad∗εi x, ad∗x εi )
2 i=1
1 d  
= ∑ ξ , [x, ei ]g εi , z − η , [x, ei ]g εi , y
2 i=1
1 d       
+ ∑
2 i=1
ξ , ei ad∗x εi , z − η , ad∗εi x − η , ei ad∗x εi , y − ξ , ad∗εi x

1 d     
=− ∑
2 i=1
ξ , ei η , ad∗εi x − η , ei ξ , ad∗εi x

= [ξ , η ]g∗ , x .

It follows that ı is a Lie algebra homomorphism, so that ı is a Lie bialgebra homo-

morphism.
Proposition 11.26. Let (g, [· , ·]g , [· , ·]g∗ ) be a Lie bialgebra. If (g, [· , ·]g ) is semi-
simple, then (g, [· , ·]g , [· , ·]g∗ ) is a coboundary Lie bialgebra.
310 11 Poisson–Lie Groups

Proof. Condition (iii) in Proposition 11.15 means that the map δ : g → ∧2 g is a 1-

cocycle 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 Whitehead’s lemma
(Lemma 4.1), δ is a coboundary, which, in the present case means that δ (x) = adx a
for some a ∈ ∧2 g. Since δ is the transpose of a Lie algebra bracket, a is a (skew-
symmetric) r-matrix, 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 finite-dimensional 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) E = V ⊕W ;
(2) V and W are Lie subalgebras of E;
(3) V and W are isotropic with respect to · | · .

We show in the following proposition that there is a natural one-to-one correspon-

dence between Manin triples and Lie bialgebras.
Proposition 11.28.
(1) Let (g, [· , ·]g , [· , ·]g∗ ) be a finite-dimensional 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.
(2) Conversely, let ((E, [· , ·]E , · | · E ),V,W ) be a Manin triple, where E is as-
sumed to be finite-dimensional. Denote by [· , ·]V and [· , ·]W the restriction
of [· , ·]E to V and W respectively, and by χ∗ ([· , ·]W ) the Lie algebra bracket
on V ∗ obtained by transporting [· , ·]W under the isomorphism χ : V ∗  W ,
induced by · | · E . Then (V, [· , ·]V , χ∗ ([· , ·]W )) is a Lie bialgebra.

Proof. Let (g, [· , ·]g , [· , ·]g∗ ) be a finite-dimensional 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 finite-
dimensional. 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 be-
cause (E, · | · E ) is quadratic, (V, [· , ·]V , χ∗ ([· , ·]W )) is a Lie bialgebra. It is easily
verified that both constructions are inverse to each other.
3 The triple is usually simply written as (E,V,W ).
11.2 Lie Bialgebras 311

Notice that, inverting the rôles 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 Exam-
ple 10.15. Thus, we assume that (g, [· , ·]) is a finite-dimensional 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 non-degenerate symmetric bilinear form · | · with respect to which both g+ and
g− are isotropic and orthogonal to g0 . Recall that the endomorphism R of g, de-
fined 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 skew-symmetric solution to the modified
Yang–Baxter equation (10.4), with c = 1, hence leads to a skew-symmetric r-matrix
for g, making (g, [· , ·], [· , ·]r ) into a Lie bialgebra. Let E := g × g, equipped with
the product Lie bracket, denoted by [· , ·] , and with the ad-invariant symmetric non-
degenerate bilinear form · | ·  , which is defined, for all x1 , x2 , y2 , y2 ∈ g, by

(x1 , y1 ) | (x2 , y2 )  := x1 | x2 − y1 | y2 . (11.31)

Let Δ be the subspace of E, defined by Δ := {(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 = Δ ⊕ W . Moreover, the assumed properties on the factors of g

imply at once that Δ and W are both subalgebras of E and are both isotropic with
respect to · | ·  . It follows that (E, Δ ,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 Δ , respectively to W , yields the Lie brackets [· , ·]
and [· , ·]r upon identifying Δ 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 Δ , coincides under this identification with [· , ·]. Consider the map
ψ : W → g, defined for (x + z, y − z) ∈ W by ψ (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 ψ , the Lie algebra
structure on W corresponds to the R-bracket [· , ·]R on g, showing our claim, since the
r-bracket on g∗ corresponds to the R-bracket 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 :

ψ (x + z, y − z), ψ (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 Poisson–Lie Groups

= x, x − y, y + x, z + z, x + y, z + z, y
= x + z, x + z − y − z, y − z
= ψ ( x + z, x + z , y − z, y − z )

= ψ ( (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 Exam-
ple 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 ℑ := ℑ 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 ℑ = ℑ (Trace(xy)) = 0. This shows that sud is isotropic with respect

to · | · ℑ . For every x, y ∈ t+ , it is clear that xy ∈ t+ , hence Trace(xy) has vanish-
ing imaginary part and t+ is also isotropic with respect to · | · ℑ . It follows that
(sld (C), sud , t+ ) is a Manin triple. We show that it is the Manin triple correspond-
ing to the Lie bialgebra discussed in Example 10.16. To do this, recall from the latter
example that we consider on sud the non-degenerate symmetric bilinear form · | · ,
defined for x, y ∈ sud by x | y := ℜ(Trace(xy)). Let ψ : t+ → sud be the composi-
tion of the isomorphisms t+  su∗d  sud , where√the first one is induced by · | · ℑ
and the second one by · | · . Explicitly, ψ (x) = − 2−1 (x + x ), for x ∈ t+ , as is com-
puted from x | y ℑ = ψ (x) | y , for all y ∈ sud . For x, y ∈ t+ , one checks by direct
computation that ψ ([x, y]) = [ψ (x), ψ (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 finite-
dimensional. Let E × E be equipped with the product Lie bracket, which we denote
by [· , ·] , and with the non-degenerate ad-invariant 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} ,
(11.33)
W  := W ×V .
11.2 Lie Bialgebras 313

Then ((E × E, [· , ·] , · | ·  ),V  ,W  ) is a Manin triple and the Lie bialgebra associ-
ated 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 Proposi-
tion 11.25, d is a coboundary Lie bialgebra, where the Lie bracket on d∗ is the
r-bracket 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 struc-
ture: according to Proposition 7.3, to a Lie bracket [· , ·]g∗ on g∗ corresponds a linear
Poisson structure π on g (and vice versa).
Proposition 11.32. Let (g, [· , ·]g ) be a finite-dimensional Lie algebra and let π be a
linear Poisson structure on g. Then the following conditions are equivalent:
(i) Denoting by [· , ·]g∗ the Lie bracket on g∗ , corresponding to π , the triple
(g, [· , ·]g , [· , ·]g∗ ) is a Lie bialgebra;
(ii) For every x, y ∈ g, the bivector πx ∈ ∧2 Tx g, considered as an element of ∧2 g,
satisfies
π[x,y]g = adx πy − ady πx . (11.34)

Proof. Let [· , ·]g∗ be the Lie bracket on g∗ which corresponds to the linear Pois-
son structure π = {· , ·} 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, δ (x) ,

where we recall that in this formula dx F and dx G are viewed as elements of g∗ . It

follows that δ (x) = πx 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 finite-dimensional Lie bialgebras,
with corresponding Poisson structures {· , ·}g and {· , ·}h on g and h respectively,
then a linear map ψ : g → h is a homomorphism of Lie bialgebras if and only if
314 11 Poisson–Lie Groups

ψ : (g, [· , ·]g ) → (h, [· , ·]h ) is a homomorphism of Lie algebras and ψ : (g, {· , ·}g ) →
(h, {· , ·}h ) is a Poisson map. Indeed, the latter condition is equivalent to the fact that
ψ  : (h∗ , [· , ·]h∗ ) → (g∗ , [· , ·]g∗ ) is a Lie algebra homomorphism. It follows that Lie
sub-bialgebras can be characterized as follows:

Proposition 11.33. Let (g, [· , ·]g , [· , ·]g∗ ) be a Lie bialgebra, and denote by π the
linear Poisson structure on g corresponding to [· , ·]g∗ . A subspace h ⊂ g is a Lie
sub-bialgebra if and only if the following two conditions are satisfied:
(1) h is a Lie subalgebra of the Lie algebra (g, [· , ·]g );
(2) h is a Poisson submanifold of the Poisson manifold (g, π ).

11.3 Poisson–Lie 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 Poisson–Lie groups and
finite-dimensional Lie bialgebras. We first show that the tangent space to the unit
of a Poisson–Lie group inherits the structure of a Lie bialgebra (Section 11.3.1).
We then show that every finite-dimensional Lie bialgebra is the Lie bialgebra
of a Poisson–Lie 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 finite-
dimensional Lie bialgebra can be embedded in a coboundary Lie bialgebra (its dou-
ble, see Proposition 11.25).

11.3.1 The Lie Bialgebra of a Poisson–Lie Group

Let (G, π ) be a Poisson–Lie 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 π vanishes at e, leading to a linear Poisson structure π1 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, π ). We show in the
following proposition that the Lie structures [· , ·]g and [· , ·]g∗ make g into a Lie bial-
gebra, so that we can naturally associate to every Poisson–Lie group a Lie bialgebra.

Proposition 11.34. Let (G, π ) be a Poisson–Lie group, with Lie algebra (g, [· , ·]g ),
and with linearized bracket [· , ·]g∗ .
(1) The triple (g, [· , ·]g , [· , ·]g∗ ) is a Lie bialgebra;
(2) Let Φ be a Poisson–Lie group homomorphism from (G, π ) to a Poisson–Lie
group (H, π  ), with corresponding Lie bialgebra (h, [· , ·]h , [· , ·]h∗ ). The tan-
gent map Te Φ of Φ at the unit e of G is a Lie bialgebra homomorphism from
(g, [· , ·]g , [· , ·]g∗ ) to (h, [· , ·]h , [· , ·]h∗ ).
11.3 Poisson–Lie Groups and Lie Bialgebras 315

Proof. Let (G, π ) be a Poisson–Lie group with Lie algebra (g, [· , ·]g ) and let Ψ
denote the cocycle G → ∧2 g, defined by (11.2). We show that the cocycle con-
dition (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, π ). Consider the identity

Ψ (ghg−1 ) = Ψ (g) + Adg Ψ (h) − Adghg−1 Ψ (g) , (11.35)

which is valid for all g, h ∈ G, as follows from the cocycle condition (11.3), upon
using that Ψ (e) = 0. Differentiating (11.35) leads to an identity for the linearized
Poisson bracket π1 , since
(π1 )x = TeΨ (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

(π1 )Adg y = Adg (π1 )y − adAdg y Ψ (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

(π1 )[x,y]g = adx (π1 )y − ady (π1 )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 π1 is a linear
Poisson structure, the map g → ∧2 g, defined by x → (π1 )x is a linear map. It follows
that, for every function z : F → g,

d 
(π1 )z(t) = (π1 ) d | z(t) .
dt t=0 dt t=0

Applied to the left-hand side of (11.37), with g = exp(tx), this yields the left-hand
side of (11.38), in view of the definition of ad. Also,

d 
Ad (π1 )y = adx (π1 )y ,
dt t=0 exp(tx)

which yields the first term in the right-hand side of (11.37). Finally, since Ψ (e) = 0,
 
d  d 
ad Ψ (exp(tx)) = ad Ψ (exp(tx))
dt t=0 dt t=0
Adexp(tx) y y

= ady TeΨ (x) = ady (π1 )x ,

which is the second term in the right-hand side of (11.37). This completes the proof
of (11.38), and hence of item (1).
316 11 Poisson–Lie Groups

We turn to the proof of (2). Since Φ : G → H is a Lie group homomorphism,

the tangent map Te Φ : g → h is a Lie algebra homomorphism (Theorem 5.1). Simi-
larly, since Φ : G → H is a Poisson map, with π vanishing at e and π  vanishing at
Φ (e), the linear map Te Φ : g → h is a Poisson map (Proposition 7.22). According
to Section 11.2.6, this shows that φ is a Lie bialgebra homomorphism.

The second item of the previous proposition implies that the Lie bialgebra of a
Poisson–Lie subgroup of (G, π ) is a Lie sub-bialgebra of the bialgebra of (G, π ).
This is stated, together with its converse, in the following proposition.

Proposition 11.35. Let (G, π ) be a Poisson–Lie group, whose Lie bialgebra is de-
noted by (g, [· , ·]g , [· , ·]g∗ ). For a connected Lie subgroup H of G, with Lie algebra h,
the following conditions are equivalent:
(i) H is a Poisson–Lie subgroup of (G, π );
(ii) h is a Lie sub-bialgebra 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, π ), by π  . Then (h, [· , ·]h , [· , ·]h∗ ) is the
Lie bialgebra of (H, π  ).

Proof. The implication (i) ⇒ (ii) follows at once from Proposition 11.34, since
the inclusion map of a Poisson–Lie subgroup is a Poisson–Lie group homomor-
phism. Conversely, assume that h is a Lie sub-bialgebra of the Lie bialgebra
(g, [· , ·]g , [· , ·]g∗ ) of (G, π ). Then the transpose δ : 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,
δ (x) = TeΨ (x) for all x ∈ g, where Ψ is defined as usual by (11.2). It follows that
TeΨ (x) ∈ ∧2 h for all x ∈ h. According to Proposition 11.7, this shows that the con-
nected Lie subgroup H of G is a Poisson–Lie subgroup. Thus, condition (ii) implies
condition (i).

11.3.2 Coboundary Poisson–Lie Groups and Lie Bialgebras

We show in this section that the Lie bialgebra of a coboundary Poisson–Lie group
is a coboundary Lie bialgebra, which justifies why the same adjective “cobound-
ary” is used for both structures. We also show the following converse: every
finite-dimensional coboundary Lie bialgebra is the Lie bialgebra of a coboundary
Poisson–Lie group.
We first show that there corresponds to every coboundary Poisson–Lie group a
coboundary Lie bialgebra. Let (G, π ) be a coboundary Poisson–Lie group. By defi-
nition, there exists a ∈ ∧2 g such that π = ← −
a −−→a , where (g, [· , ·]) is the Lie algebra
of G. According to Proposition 11.10, [[a, a]] is Ad-invariant, hence is ad-invariant.
Proposition 10.11 implies that a is a skew-symmetric r-matrix, i.e., the associated
bracket [· , ·]a is a Lie bracket on g∗ , making (g, [· , ·], [· , ·]a ) into a coboundary Lie
11.3 Poisson–Lie 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 Poisson–Lie group (G, π ).
Proposition 11.36. Let (G, π ) be a coboundary Poisson–Lie group, with π = ← −
a −→−
a
for some r-matrix a ∈ ∧2 g, where (g, [· , ·]g ) denotes the Lie algebra of G. The Lie
bialgebra of (G, π ) is the coboundary Lie bialgebra (g, [· , ·]g , [· , ·]a ).

Proof. According to (11.14), the map Ψ : G → ∧2 g, defined by g → ∧2 (Tg Rg−1 ) πg ,

is given by Ψ (g) = Adg a − a. The tangent map of Ψ at e is therefore given by
TeΨ (x) = adx a for all x ∈ g. According to item (2) in Proposition 11.5, the linearized
Poisson structure π1 of π at the unit e of G is given by (π1 )x = TeΨ (x), for all x ∈ g,
so that the linearized Poisson structure π1 is given at x ∈ g by (π1 )x = adx a, whose
transpose is precisely the Lie bracket [· , ·]a . This shows that (g, [· , ·]g , [· , ·]a ) is the
Lie bialgebra of (G, π ).

Corollary 11.37. Let (g, [· , ·]g , [· , ·]a ) be a finite-dimensional coboundary Lie bial-
gebra. If G is a connected Lie group, with Lie algebra (g, [· , ·]g ), then G ad-
mits the structure of a coboundary Poisson–Lie group, whose Lie bialgebra is
(g, [· , ·]g , [· , ·]a ). In particular, every finite-dimensional coboundary Lie bialgebra
is the Lie bialgebra of a coboundary Poisson–Lie group.

Proof. We only need to prove the first part of the corollary, since the last state-
ment 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 finite-
dimensional coboundary Lie bialgebra, with a ∈ ∧2 g so that [[a, a]] is ad-invariant.
Let G be a connected Lie group whose Lie algebra is g. Since G is assumed to be
connected, [[a, a]] is Ad-invariant. Setting π := ←
−
a −−
→
a , Propositions 11.10 and 11.36
imply that (G, π ) is a Poisson–Lie group, whose Lie bialgebra is (g, [· , ·]g , [· , ·]a ), as
was to be shown.

Corollary 11.38. Let (g, [· , ·]g , [· , ·]g∗ ) be a finite-dimensional 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 πD , making (D, πD ) into
a coboundary Poisson–Lie group, whose Lie bialgebra is (d, [· , ·]d , [· , ·]a ). In terms
of a pair of dual bases (e1 , . . . , ed ) and (ε1 , . . . , εd ) for g and for g∗ respectively, the
skew-symmetric r-matrix a of d is given by a = 12 ∑di=1 ei ∧ εi .

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 Poisson–Lie group, whose Lie bial-
gebra 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 Poisson–Lie Groups

11.3.3 The Integration of Lie Bialgebras

We show in this section that every finite-dimensional Lie bialgebra can be inte-
grated. We mean by this that there exists for every finite-dimensional Lie bialgebra
(g, [· , ·]g , [· , ·]g∗ ) a (not necessarily unique) Poisson–Lie group (G, π ) 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 finite-dimensional Lie bialgebra and
let G be a connected Lie group with Lie algebra (g, [· , ·]g ).
(1) There exists at most one Poisson structure π on G such that (G, π ) is a
Poisson–Lie group, whose Lie bialgebra is (g, [· , ·]g , [· , ·]g∗ );
(2) If G is simply connected, then there exists on G a unique Poisson struc-
ture π , such that (G, π ) is a Poisson–Lie group, whose Lie bialgebra is
(g, [· , ·]g , [· , ·]g∗ );
(3) In particular, every finite-dimensional Lie bialgebra is the Lie bialgebra of a
Poisson–Lie group.
Proof. Let π (1) and π (2) be two multiplicative Poisson structures on the connected
Lie group G, leading to the same Lie bialgebra, i.e., their linearized Poisson struc-
tures at e coincide. In view of (2) in Proposition 11.5, TeΨ (1) (x) = TeΨ (2) (x), for all
(i)
x ∈ g, where Ψ (i) : G → ∧2 g is defined by Ψ (i) (g) := ∧2 (Tg Rg−1 )πg , for i = 1, 2. It
follows that Ψ (1) − Ψ (2) , which is a cocycle (since both Ψ (1) and Ψ (2) are cocycles),
has zero tangent map at e. By Proposition 11.4, this implies that Ψ (1) − Ψ (2) = 0,
and hence that π (1) = π (2) . This shows item (1).
Let (g, [· , ·]g , [· , ·]g∗ ) be a finite-dimensional Lie bialgebra. According to Propo-
sition 11.25, (g, [· , ·]g , [· , ·]g∗ ) is a Lie sub-bialgebra of a coboundary Lie bialgebra,
which we denote by (d, [· , ·]d , [· , ·]a ), because the underlying Lie algebra is the dou-
ble of (g, [· , ·]g , [· , ·]g∗ ). In view of Proposition 11.37, (d, [· , ·]d , [· , ·]a ) can be inte-
grated into a Poisson–Lie group (D, πD ), which is the only fact about d which we
will use in this proof. We denote by Ĝ the unique connected Lie subgroup of D,
whose Lie algebra is g. According to Proposition 11.35, the subgroup Ĝ inherits a
multiplicative Poisson structure π̂ from πD , whose corresponding Lie bialgebra is
(g, [· , ·]g , [· , ·]g∗ ). This shows (3).
We show now that (Ĝ, π̂ ) can be replaced by the connected and simply con-
nected Lie group G whose Lie algebra is g, which is the content of (2). In view
of Lie’s theorem (see item (4) of Theorem 5.1), there exists a Lie group homomor-
phism Φ : G → Ĝ, whose induced Lie algebra homomorphism is the identity map.
We claim that there is a unique Poisson structure π on G which turns Φ : G → Ĝ into
a Poisson map, and we show that π is multiplicative. Then (G, π ) is a Poisson–Lie
group and Φ is a homomorphism of Poisson–Lie 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 Φ = RΦ (g) Φ Rg−1 at g ∈ G
shows that the tangent maps of Φ at g ∈ G and at e are related by Tg Φ = Tê RΦ (g) ◦
11.3 Poisson–Lie Groups and Lie Bialgebras 319

Te Φ ◦ Tg Rg−1 (in this formula, e and ê are the units of the Lie groups G and Ĝ
respectively). Since the three linear maps which appear in the right-hand side of
this formula are invertible, so is Tg Φ , hence Φ is a local diffeomorphism. We can
therefore define π for all g ∈ G by setting

πg := ∧2 (Tg Φ )−1 π̂Φ (g) . (11.39)

According to Proposition 1.19, if there exists a Poisson structure on G which makes

Φ into a Poisson map, then it must be given by (11.39), which gives the desired
uniqueness of π . Also, the vertical maps of the following commutative diagram are
local diffeomorphisms (here μ and μ̂ stand for the group products of G and Ĝ).
μ
G×G G

Φ ×Φ Φ

Ĝ × Ĝ μ̂
Ĝ

Combined with the fact that μ is a Poisson map, this shows that μ̂ is Poisson map.
As a consequence, π 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 Poisson–Lie groups.
Theorem 11.40. Let (g, [· , ·]g , [· , ·]g∗ ) and (h, [· , ·]h , [· , ·]h∗ ) be two finite-dimensi-
onal Lie bialgebras and let (G, π ) and (H, π  ) be Poisson–Lie groups which in-
tegrate them, where G is chosen connected and simply connected. For every Lie
bialgebra homomorphism φ : g → h, there exists a unique Poisson–Lie group homo-
morphism Φ : G → H whose induced Lie bialgebra homomorphism is φ , i.e., such
that Te Φ = φ , where e denotes the unit element of G.

Proof. Suppose that φ : g → h is a homomorphism of Lie bialgebras. In view of

Lie’s theorem (item (4) of Theorem 5.1), there exists a unique Lie group homomor-
phism Φ : G → H such that Te Φ = φ . It remains to be shown that Φ is a Poisson
map. To do this, we consider the map Ξ : G → ∧2 h, defined for all g ∈ G by

Ξ (g) := ∧2 φ (Ψ (g)) − Ψ  (Φ (g)) ,

where Ψ : G → ∧2 g and Ψ  : H → ∧2 h denote the group cocycles, associated to

(G, π ) and to (H, π  ) respectively (see item (iii) in Proposition 11.2). The map Ξ is
a cocycle of G with respect to the adjoint representation of G on ∧2 h, i.e.,

Ξ (gh) = Ξ (g) + AdΦ (g) Ξ (h) ,

for all g, h ∈ G. We first show that Ξ = 0. In view of item (1) of Proposition 11.4,
it suffices to show that Te Ξ (x) = 0 for all x ∈ g. To prove the latter, recall from
320 11 Poisson–Lie Groups

item (2) in Proposition 11.5 that TeΨ (x) = (π1 )x , for all x ∈ g, where π1 denotes the
linearized Poisson structure of π at e, and similarly for Ψ  and π1 . It allows us to
compute

Te Ξ (x) = ∧2 φ (TeΨ (x)) − TeΨ  (Te Φ (x))

= ∧2 φ ((π1 )x ) − (π1 )φ (x) = 0 ,

where the last equality, which states that φ is a Poisson map, is tantamount to the
fact that the transpose map φ  : (h∗ , [· , ·]h∗ ) → (g∗ , [· , ·]g∗ ) is a Lie algebra homo-
morphism (see Section 11.2.6). It follows that Ξ = 0, leading to the equality

∧2 φ (Ψ (g)) = Ψ  (Φ (g)) ,

valid for all g ∈ G, which can be written, using Te Φ = φ and using the defini-
tion (11.2) of the cocycle Ψ , as

∧2 (Tg (Φ ◦ Rg−1 ))πg = ∧2 TΦ (g) RΦ (g)−1 πΦ (g) . (11.40)

Now Φ is a group homomorphism, so that

Φ ◦ Rg−1 = RΦ (g−1 ) ◦ Φ ,

which, substituted in (11.40), leads to ∧2 (Tg Φ )πg = πΦ (g) , since in a Lie group,
right translation by an element of the group is a diffeomorphism. In view of Propo-
sition 1.19, the latter equality implies that Φ 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 Poisson–Lie groups.

Definition 11.41. Two Poisson–Lie groups (G, π ) and (G∗ , π ∗ ) are said to be dual
to each other if their Lie bialgebras are dual to each other.

Let (G, π ) be a Poisson–Lie group and let (g, [· , ·]g , [· , ·]g∗ ) denote its Lie bialge-
bra, whose dual bialgebra is (g∗ , [· , ·]g∗ , [· , ·]g ). Every Poisson–Lie group (G∗ , π ∗ )
which integrates the latter Lie bialgebra is dual to (G, π ). 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∗ there exists a unique
Poisson structure π ∗ such that (G∗ , π ∗ ) integrates (g, [· , ·]g , [· , ·]g∗ ). This Poisson–
Lie group is called the dual of (G, π ). For (G, π ) a connected and simply connected
Poisson–Lie group, taking twice the dual gives back the Poisson–Lie group (G, π ).

Example 11.42. Let G be a connected, simply connected Lie group, with Lie algebra
(g, [· , ·]g ). Consider g∗ as a Lie group, where the group operation is addition. In
view of Corollary 11.9, g∗ becomes a Poisson–Lie group, when it is equipped with
the linear Poisson structure π on g∗ corresponding to the Lie bracket [· , ·]g . Its Lie
11.4 Dressing Actions and Symplectic Leaves 321

bialgebra is (g∗ , 0, [· , ·]g ), with dual Lie bialgebra (g, [· , ·]g , 0). It follows that the
(connected and simply connected) dual to the Poisson–Lie group (g∗ , π ) is the Lie
group G, equipped with the trivial Poisson structure.

11.4 Dressing Actions and Symplectic Leaves

We show in this section that the symplectic leaves of a Poisson–Lie group can be
described by using dressing actions, a notion which we first define in the general
context of Lie groups.
Let G be a connected Lie group and let G1 and G2 be two Lie subgroups of G.
The triple (G, G1 , G2 ) is said to be complete if the map G1 × G2 → G given by
(g1 , g2 ) → g1 g2 is a diffeomorphism. Since (g1 g2 )−1 = g−1 −1
2 g1 , for all g1 , g2 ∈ G,
the triple (G, G1 , G2 ) is complete if and only if the triple (G, G2 , G1 ) is complete. It
follows that, when (G, G1 , G2 ) is complete, there exists for every (g1 , g2 ) ∈ G1 ×G2
a unique pair (g1 , g2 ) ∈ G1 × G2 , such that g1 g2 = g2 g1 . It is customary to denote
the elements g1 and g2 by gg12 and gg21 respectively.
The associativity of the group structure on G implies that the assignments
−1
(g, h) → hg and (g, h) → gh are group actions. Indeed, the relation (g1 h1 )g2 =
g1 (h1 g2 ) implies, for g1 , h1 ∈ G1 and g2 ∈ G2 , that
h
(g 1 )
gg21 h1 (g1 h1 )g2 = g1 gh21 hg12 = (gh21 )g1 g1 2 hg12 ,

so that, by uniqueness of the decomposition, we can derive the relation

gg21 h1 = (gh21 )g1 , (11.41)

valid for all g1 , h1 ∈ G1 and g2 ∈ G2 . This shows that the assignment (g1 , g2 ) → gg21
defines a group action of G1 on (the manifold) G2 . Similarly, one derives from the
relation g1 (g2 h2 ) = (g1 g2 )h2 , for g1 ∈ G1 and g2 , h2 ∈ G2 , the relation

gg12 h2 = (gg12 )h2 , (11.42)

g−1
which says that the assignment (g2 , g1 ) → g12 defines a group action of G2 on (the
manifold) G1 . These two group actions are respectively called the dressing action
of G1 on G2 and the dressing action of G2 on G1 .

Example 11.43. According to a standard theorem of linear algebra (essentially the

Gram–Schmidt orthogonalization process), every complex invertible matrix of de-
terminant 1 can be decomposed in a unique way as the product of an element in
SUd and an element in T+ , the group of upper triangular matrices of determinant 1,
admitting only positive entries on the diagonal. It follows that (SLd (C), SUd , T+ )
is complete.
322 11 Poisson–Lie Groups

In the case of Poisson–Lie groups, the notion of dressing action leads to the
following definition.

Definition 11.44. Let (G, π ) be a connected Poisson–Lie group with Lie bialgebra
(g, [· , ·]g , [· , ·]g∗ ). A Lie group D is called a dressing group of (G, π ) if the following
three conditions are satisfied.
(1) The Lie algebra of D is the double (d, [· , ·]d ) of (g, [· , ·]g , [· , ·]g∗ );
(2) G is the connected Lie subgroup of D corresponding to the Lie subalgebra
g ⊂ d;
(3) (D, G, G∗ ) is complete, where G∗ stands for the connected Lie subgroup of D
corresponding to the Lie subalgebra g∗ ⊂ d.
The Lie subgroup G∗ of D is then called the dual of (G, π ) with respect to the
dressing group D. The Poisson structure on G∗ as a Poisson submanifold of the
Poisson–Lie group (D, πD ) is denoted by π ∗ .

Example 11.45. A Lie bialgebra structure on g := sud was given in Example 10.16,
We showed in Example 11.30 that its double d = g ⊕ g∗ is isomorphic to the Lie
algebra sld (C) of traceless complex matrices, identification under which the Lie
algebras g and g∗ become the Lie algebras sud and t+ , where t+ is the Lie alge-
bra of T+ , i.e., the vector space of upper triangular matrices of trace zero with real
coefficients on the diagonal. Consider the Poisson bracket π on SUd which makes
(SUd , π ) into a Poisson–Lie group, integrating the latter bialgebra structure on sud .
Since (SLd (C), SUd , T+ ) is complete, as we have seen in Example 11.43, condi-
tions (1)–(3) in the above definition are verified, so that SLd (C) is a dressing group
of (SUd , π ).

Proposition 11.46. Let (G, π ) be a connected Poisson–Lie group and suppose

that D is a dressing group of (G, π ), and denote by (G∗ , π ∗ ) the dual of (G, π )
with respect to D.
(1) The diffeomorphism G × G∗ → G∗ × G, defined by (g, h) → (hg , gh ) is a Pois-
son map;
(2) The dressing action of G on G∗ is a Poisson action;
(3) The dressing action of G∗ on G is also a Poisson action, provided that we
replace the Poisson structure π ∗ by its opposite −π ∗ ;
(4) For α ∈ g∗ , the fundamental vector field α of the dressing action of G∗ on G
is given, at g ∈ G, by
  
α g = −Te Rg Pg Adg α , (11.43)

where Pg : d → g is the projection from d = g⊕g∗ onto g, and where Ad stands

for the adjoint action of D on its Lie algebra d.

Proof. The inclusion map G × G∗ → D × D is a Poisson map because G and G∗

are Poisson submanifolds of D. Since the product map D × D → D is a Poisson map
(because D is a Poisson–Lie group), the diffeomorphism G × G∗ → D, given by
11.4 Dressing Actions and Symplectic Leaves 323

(g, h) → gh, is a Poisson diffeomorphism. Since the same is true for the diffeomor-
phism G∗ × G → D, given by (h, g) → hg, the map G × G∗ → G∗ × G, defined by
(g, h) → (hg , gh ) is a Poisson diffeomorphism. This proves (1). As a corollary, the
map G×G∗ → G∗ , defined by (g, h) → hg , is a Poisson map, meaning that the dress-
ing action of G on G∗ is a Poisson map, proving (2). Since the inverse map h → h−1
is a Poisson map from (G∗ , −π ∗ ) to (G∗ , π ∗ ) (see Proposition 11.5, item (3)), the
−1
assignment (g, h) → gh is also a Poisson map, provided that G × G∗ is endowed
with the product Poisson structure (π , −π ∗ ), which means precisely that the dress-
ing action of G∗ on G is Poisson, provided that we replace the Poisson structure
π ∗ of G∗ by its opposite. This proves (3).
We now prove (4). For all g ∈ G and t ∈ F we have, by definition of the dressing
action
−1
g exp(−t α ) = h(t)gexp(−t α ) = h(t)gexp(t α ) , (11.44)
where h(t) = exp (−t α )g . Note that h(0) = e, and that h(t) is a path in G∗ . Multi-
plying (11.44) by g−1 and taking the derivative with respect to t at t = 0, we obtain
 
d  exp(t α )−1 −1 d 
− Adg α =  g g +  h(t)
dt t=0 dt t=0

d 
= Tg Rg−1 α g +  h(t) , (11.45)
dt t=0

where the last equality follows from the definition of the fundamental vector field
of a Lie group action. Since h(t) ∗
∗
 is apath in G , (11.45) is the decomposition of
− Adg α in g ⊕ g , so that −Pg Adg α = Tg Rg−1 α g , which amounts to (11.43).
We can now describe the symplectic leaves of a Poisson–Lie group which admits
a dressing group, in terms of the dressing action.
Theorem 11.47. Let (G, π ) be a connected Poisson–Lie group. Assume that D is a
dressing group of (G, π ), and let G∗ ⊂ D be the dual of G with respect to D. Then
the symplectic leaves of (G, π ) are the orbits of the dressing action of G∗ on G.
Proof. Recall from Corollary 11.38 that D admits a Poisson structure πD , making
(D, πD ) into a coboundary Poisson–Lie group. This Poisson structure is associated
to the skew-symmetric r-matrix a := 12 ∑di=1 ei ∧ εi ∈ ∧2 d, where (e1 , . . . , ed ) and
(ε1 , . . . , εd ) are dual bases of g and g∗ . In view of (11.14), one has for every h ∈ D,

∧2 (Th Rh−1 ) (πD )h = Adh a − a .

Since (G, π ) is a Poisson–Lie subgroup of (D, πD ), it follows for every g ∈ G that

1 d 1 d
∧2 (Tg Rg−1 ) πg = ∑
2 i=1
Adg ei ∧ Adg εi − ∑ ei ∧ εi .
2 i=1

The left-hand side of this equation belongs to ∧2 g, hence is fixed by the linear map
∧2 Pg (with Pg the projection of d = g ⊕ g∗ onto g). It follows that
324 11 Poisson–Lie Groups

1 d
∧2 (Tg Rg−1 ) πg = ∑ Adg ei ∧ Pg(Adg εi ) ,
2 i=1
(11.46)

which, in turn, gives the following expression of the Poisson bivector π , evaluated
at the point g,
1 d −
πg = ∑ ( ← ei )g ∧ εi g , (11.47)
2 i=1
where we used item (3) of Proposition 11.46 as well as the identity Te Rg (Adg x) =
Te Lg (x) = ←
−x g for every x ∈ g.
We use the explicit expression (11.47) of πg to show that the symplectic leaf of π
through g coincides with the orbit through g of the dressing action of G∗ on G.
Since G∗ is connected and since (ε1 , . . . , εd ) is a basis of g∗ , it amounts to showing
that the image of the linear map πg : Tg∗ G → Tg G, defined by πg , is the vector sub-
space of Tg G generated by ε1 g , . . . , εd g . To show this, it suffices according to (11.47)
to show that the matrix (ai j )i, j=1,...,d expressing the elements εi g , i = 1, . . . , d, in
terms of the basis of tangent vectors (← e−1 )g , . . . , (←
e−
d )g is skew-symmetric, since
these elements are then the columns of the Poisson matrix of π at g, expressed
in terms of the latter basis. Notice that, in view of (11.46), this matrix is also the
matrix which expresses the elements Pg (Adg εi ), i = 1, . . . , g, in terms of the vec-
tors Adg e1 , . . . , Adg ed ,

d
Pg (Adg εi ) = ∑ ai j Adg e j ,
j=1

so that    
ai j = Ad∗g ε j , Pg (Adg εi ) = Pg∗ (Adg ε j ), Pg (Adg εi ) ,
where Pg∗ is the projection of d = g ⊕ g∗ onto g∗ ; we also used the fact that
ad∗x ε j = Pg∗ ([x, (y, ε j )]d ) for all x, y ∈ g, which follows from the definition (11.23)
of [· , ·]d . Hence, to show that the matrix (ai j )i, j=1,...,d is skew-symmetric, we need
to show that
   
Pg∗ (Adg εi ), Pg (Adg ε j ) + Pg∗ (Adg ε j ), Pg (Adg εi ) = 0 (11.48)

for 1  i, j  d. According to the

 definition (11.22)
 of · | · d , the left-hand side of
this equation can be written as Adg εi | Adg ε j d . Now
   
Adg εi | Adg ε j d
= εi | ε j d = 0 ,

in view of the Ad-invariance of · | · d . This proves (11.48), hence yields that the
matrix (ai j )i, j=1,...,d is skew-symmetric, which was to be shown.
11.5 Notes 325

11.5 Notes

Poisson structures can be seen as the semi-classical limit of quantum structures, i.e.,
as being what remains of quantization when h̄2 (but not h̄) is considered to be
small enough to be ignored (see Chapter 13). This is especially true for Poisson–Lie
groups which first appeared as semi-classical limits of quantum groups, in the pio-
neering work [58] of Drinfel’d. A second motivation for the introduction of Poisson–
Lie groups was the study of integrable systems associated to infinite-dimensional
Lie algebras, such as the Korteweg–de Vries equation, where Poisson–Lie groups
provide the hidden symmetry, see Semenov–Tian–Shansky [180].
Poisson–Lie groups have proved to be interesting on their own. First, they natu-
rally appear when studying the singularities of several highly singular spaces asso-
ciated to semi-simple Lie groups, like Bruhat and Schubert cells, see Lu–Weinstein
[132, 136]. Also, a Poisson–Lie group is the first instance of a more general ob-
ject, namely a Poisson–Lie groupoid, see Mackenzie–Xu [140], which appears natu-
rally when studying the dynamical Yang–Baxter equation. Exactly like Poisson–Lie
groups and Lie bialgebras are in one-to-one correspondence, Poisson–Lie groupoids
are in one-to-one correspondence with Lie bialgebroids, see Mackenzie–Xu [140].
One of the first systematic studies of Poisson–Lie groups is the PhD thesis of
Lu [132], which remains an excellent introduction to the subject, especially to those
who want to learn more about the relation with Lie groupoids and enlarged defi-
nitions of moment maps. It also deals with a slightly more general object: affine
Poisson structures on Lie groups. Another useful reference is the review article by
Kosmann–Schwarzbach [110], which insists in particular on the link with Lie bial-
gebras.
 Part III
Applications
Chapter 12
Liouville Integrable Systems

In this chapter we present the main application of Poisson structures: the theory of
integrable Hamiltonian systems.
Poisson’s theorem, which states that the Poisson bracket of two constants of mo-
tion (of a Hamiltonian system) is a constant of motion, is from the modern point of
view a direct consequence of the formalism, but was in Poisson’s time considered
as a fundamental step towards the explicit integration of Hamilton’s equations, as
for doing this by the known methods, a large number of constants of motion was
required.
The fundamental rôle of Poisson structures in integrability was emphasized by
Liouville. As he showed, having half of the dimension of phase space of independent
constants of motion is sufficient for integrating the equations of motion, under the
assumption that these constants of motion are in involution (Poisson commute). Pre-
cisely, integration means here integration by quadratures. A geometrical counterpart
to this result states that, under some topological assumptions, the motion of such a
Hamiltonian system evolves on tori (the so-called Liouville tori), whose dimension
is half the dimension of phase space, and the motion on them is quasi-periodic. A
more elaborate version of this theorem is the action-angle theorem, which yields a
canonical model for Liouville integrable systems, in the neighborhood of a Liouville
torus.
Liouville’s theorems and the action-angle theorems will be presented here in
the context of general Poisson manifolds. These manifolds generalize the phase
spaces, considered in the classical results, namely the cotangent bundle of the con-
figuration space, equipped with its canonical symplectic structure. We give the ba-
sic definitions and properties of functions in involution and of the map, associ-
ated to them, in Section 12.1. Several constructions of functions in involution are
given in Section 12.2: Poisson’s theorem, the Hamiltonian form of Noether’s theo-
rem, bi-Hamiltonian vector fields, Thimm’s method, Lax equations and the Adler–
Kostant–Symes theorem. The action-angle theorem for Poisson manifolds is stated
and proved in Section 12.3.

C. Laurent-Gengoux et al., Poisson Structures, 329

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 12, © Springer-Verlag Berlin Heidelberg 2013
330 12 Liouville Integrable Systems

12.1 Functions in Involution

In this section we give the basic definitions and the (geometrical) properties of
functions in involution. Throughout the section, all manifolds considered are real
(F = R) or complex (F = C). For such a manifold M, we denote its algebra of
smooth or holomorphic functions by F (M).
Definition 12.1. Let (M, π ) be a Poisson manifold and let F1 and F2 be two elements
of F (M). We say that F1 and F2 are in involution if their Poisson bracket is zero,
{F1 , F2 } = 0. More generally, let F = (F1 , . . . , Fs ) be an s-tuple of elements of F (M).
We say that F is involutive if for all 1  i, j  s, the functions Fi and Fj are in
involution.
Classically, a function F ∈ F (M) which is constant on all the integral curves of a
Hamiltonian vector field XH , where H is a given function on a Poisson manifold
(M, π ), is called a constant of motion (for H). Since XH [F] = {F, H}, we have that
F is a constant of motion for H if and only if F and H are in involution. Therefore,
the notion of involutive s-tuples generalizes the notion of constant of motion.
An s-tuple F = (F1 , . . . , Fs ) of elements of F (M) defines a map M → Fs , which
will be denoted by the same letter F. We say that F is independent if the open
subset UF of M, defined by

UF := {m ∈ M | dm F1 ∧ · · · ∧ dm Fs = 0} , (12.1)

is a dense subset of M. Besides the open subset UF , defined by (12.1), we will in

this chapter also make extensive use of the open subset

M(r) := {m ∈ M | Rkm π = 2r} ,

where 2r stands for the rank of π . Notice that, if F is independent, then the open
subset UF ∩ M(r) of M is non-empty.
When dealing with involutive functions, the reader should have the geometrical
picture in mind, described by the following proposition (see also Fig. 12.1).
Proposition 12.2. Let (M, π ) be a Poisson manifold and assume that F = (F1 , . . . , Fs )
is an involutive and independent s-tuple of elements of F (M). The Hamiltonian vec-
tor fields XFi , 1  i  s, commute and they are tangent to the smooth fibers of the
map F : M → Fs .
Proof. The Hamiltonian vector fields, which are associated to functions in involu-
tion, commute, because [XF , XG ] = −X{F,G} for all F, G ∈ F (M). Assume that
 
F = (F1 , . . . , Fs ) is involutive and let 1  i  s; since XFi [Fj ] = Fj , Fi = 0 for all
1  j  s, the vector field XFi is tangent to the smooth level sets of F1 , . . . , Fs , that
is, to the smooth fibers of F.  
The number of independent functions on a manifold M is bounded by the dimension
of M. In the case of a Poisson manifold, the upper bound for the number of indepen-
dent functions in involution is much lower, as given by the following proposition.
12.1 Functions in Involution 331

Fig. 12.1 The Hamiltonian vector fields of the components of F = (F1 , . . . , Fs ) are tangent to the
smooth fibers F−1 (c) of F and they commute.

Proposition 12.3. Let (M, π ) be a Poisson manifold of dimension d and rank 2r and
suppose that F = (F1 , . . . , Fs ) is an independent s-tuple of elements of F (M).
(1) If F1 , . . . , Fs are Casimirs, then s  d − 2r;
(2) If F is involutive, then s  d − r;
(3) If F is involutive, with s = d − r, then
 
dim span (XF1 )m , . . . , (XFs )m = r

for every m ∈ UF ∩ M(r) .

Proof. For m ∈ M, let πm denote the linear map Tm∗ M → Tm M, which corresponds
to the Poisson structure. Explicitly, for H ∈ F (M),

πm (dm H) = − (XH )m .

Recall that the rank of πm is Rkm π . For every H ∈ Cas(M) the cotangent vector
dm H belongs to Ker πm , whose dimension is d − Rkm π . Let F = (F1 , . . . , Fs ) be
independent and let m be an element of the non-empty (open) subset UF ∩ M(r)
of M. Suppose first that each element of F is a Casimir. Since dm F1 , . . . , dm Fs are
independent elements of Ker πm , we have that

s  dim Ker πm = d − 2r

and (1) follows. Next, suppose that F is involutive and consider Fm , the fiber of F,
passing through m. Since m ∈ UF ∩ M(r) , the restriction of Fm to a neighborhood U
of m is a submanifold of dimension d − s of U, passing
 through m. This dimension
is an upper bound for the dimension dm of span (XF1 )m , . . . , (XFs )m , because
these s vectors are tangent to that fiber at m. Moreover, dm  s − dim Ker πm =
s + 2r − d, because the differential one-forms dF1 , . . . , dFs are independent at m.
Combining the two inequalities for dm , we get
332 12 Liouville Integrable Systems

s + 2r − d = s − dim Ker πm  dm  d − s , (12.2)

leading to (2). Third, suppose that s = d − r. For m ∈ UF ∩ M(r) we deduce

from (12.2) that
r = s − dim Ker πm  dm  r ,
 
so that dim span (XF1 )m , . . . , (XFs )m = dm = r. 


The extreme case r + s = d will be discussed in detail in Section 12.3. Notice

that, in this case, the commuting Hamiltonian vector fields XF1 , . . . , XFs define an
integrable distribution of rank r on the open subset UF ∩ M(r) of M.

12.2 Constructions of Functions in Involution

In this section we will give a list of constructions of functions in involution. The

first two of them (Poisson’s theorem and Noether’s theorem (in its Hamiltonian
form)) are very classical, the other ones were discovered quite recently. All mani-
folds which are considered in this section are real or complex; accordingly, F stands
for R or for C.

12.2.1 Poisson’s Theorem

Poisson’s theorem yields new constants of motion (of a given Hamiltonian H), start-
ing from two given constants of motion.

Theorem 12.4 (Poisson’s theorem). Let (M, π ) be a Poisson manifold and let F, G
and H be elements of F (M). If F and G are constants of motion for H, then their
Poisson bracket {F, G} is also a constant of motion for H.

Proof. The proof is an easy consequence of the Jacobi identity, valid for all
F, G, H ∈ F (M):

{F, {G, H}} + {G, {H, F}} + {H, {F, G}} = 0 . (12.3)

Namely, if F and G are constants of motion for H, then the first two terms in (12.3)
vanish. Then the remaining term also vanishes, that is, {F, G} is a constant of motion
for H.  

In practice, Poisson’s theorem often does not yield a function which is independent
of the given constants of motion; as we will see, it happens in many important cases
that the Poisson bracket of two constants of motion is actually equal to zero.
12.2 Constructions of Functions in Involution 333

12.2.2 Noether’s Theorem

A second classical theorem which leads to constants of motion is Noether’s theo-

rem, which we give in a Hamiltonian form (the classical version is in a Lagrangian
form). Let G be a Lie group, whose Lie algebra is denoted by g. Suppose that G
acts on a Poisson manifold (M, π ) and suppose that the action is Hamiltonian
(Definition 5.38). Recall that this means in particular that one can construct a co-
momentum map, which is a linear map μ̃ : g → F (M), having the property that for
every x ∈ g the function −μ̃x ∈ F (M) is a Hamiltonian for the fundamental vector
field x, i.e.,
x = −Xμ̃x .
Noether’s theorem, in its Hamiltonian form, then states that the co-momentum map
yields constants of motion for every G-invariant Hamiltonian.
Theorem 12.5 (Noether’s theorem). Let G be a group which acts on a Poisson
manifold (M, π ) and assume that the action is Hamiltonian, with co-momentum
map μ̃ . If H is a G-invariant function on M, then for every x ∈ g the function μ̃x is
a constant of motion for XH .

Proof. Let x ∈ g and m ∈ M be arbitrary. We denote for every g ∈ G by χg : M → M

the diffeomorphism which is induced by the action and we write {· , ·} for π . By G-
invariance of H, that is, H ◦ χg = H for every g ∈ G, we have that
 
{H, μ̃x } (m) = Xμ̃x m [H] = −xm [H]
d   d
=− H χexp(−tx) (m) = − H(m) = 0 .
dt |t=0 dt |t=0

Since m ∈ M is arbitrary, this shows that the function μ̃x is a constant of motion
for H. 

12.2.3 Bi-Hamiltonian Vector Fields

A bi-Hamiltonian manifold (M, π1 , π2 ) is a manifold M, equipped with two com-

patible Poisson structures πi = {· , ·}i , i = 1, 2. Let V be a vector field on a bi-
Hamiltonian manifold (M, π1 , π2 ) and suppose that there exist functions F and G
on M for which
V = {· , F}1 = {· , G}2 . (12.4)
Then V is called a bi-Hamiltonian vector field. Equation (12.4) implies that F and G
are in involution with respect to both Poisson structures since {F, G}2 = {F, F}1 = 0
and {G, F}1 = {G, G}2 = 0. They are in involution with respect to every linear com-
bination of these Poisson structures, which is itself a Poisson structure, because π1
and π2 are compatible.
334 12 Liouville Integrable Systems

More generally, suppose that we have a bi-Hamiltonian hierarchy on a bi-

Hamiltonian manifold (M, π1 , π2 ), i.e., a sequence of functions F = (Fi )i∈Z such
that
{· , Fi+1 }1 = {· , Fi }2 , (12.5)
for every i ∈ Z. In this case one has, for all i < j ∈ Z,
   
Fi , Fj 1 = Fi , Fj−1 2
 
= Fi+1 , Fj−1 1
= ...
 
= Fj , Fi 1 ,
 
so that Fi , Fj 1 = 0 by skew-symmetry. It follows that F is involutive with respect
   
to π1 . Also, F is involutive with respect to π2 , because Fi , Fj 2 = Fi , Fj+1 1 .
Such bi-Hamiltonian hierarchies may be constructed, under some assumptions,
as follows. Given two compatible Poisson structures π1 and π2 , one first consid-
ers, for arbitrary λ ∈ F, the Poisson structure πλ := π1 − λ π2 . We set, as before,
{· , ·}λ := πλ . The one-dimensional family of Poisson structures (πλ )λ ∈F is called a
Poisson pencil. Suppose that there exists a Laurent polynomial in λ ,

Fλ = ∑ λ i Fi ,
i=−k

such that {· , Fλ }λ = 0 for all λ ∈ F. Such a Laurent polynomial is called a Casimir

for the Poisson pencil. Setting Fi := 0 for i >  and for i < −k, we can write

{· , Fλ }λ = ∑ λ i+1 ({· , Fi+1 }1 − {· , Fi }2 ) .

i∈Z

It follows that, since Fλ is a Casimir for the Poisson pencil, the sequence of functions
(Fi )i∈Z satisfies (12.5) for all i ∈ Z, hence is a bi-Hamiltonian hierarchy. By the
above, (Fi )i∈Z is involutive.

12.2.4 Poisson Maps and Thimm’s Method

Thimm’s method [191] for constructing large families of functions in involution is

based on a few elementary observations, cleverly combined. The first one is that if
Ψ : (M1 , {· , ·}1 ) → (M2 , {· , ·}2 ) is a Poisson map (between two Poisson manifolds),
then the pull-back of every involutive s-tuple on M2 yields an involutive s-tuple
on M1 . This is simply because if F and G are functions on M2 and {F, G}2 = 0, then

{F ◦ Ψ , G ◦ Ψ }1 = {F, G}2 ◦ Ψ = 0 .
12.2 Constructions of Functions in Involution 335

The second one is that Lie algebra homomorphisms between finite-dimensional Lie
algebras induce Poisson maps: let φ : g1 → g2 be a Lie algebra homomorphism,
where (g1 , [· , ·]1 ) and (g2 , [· , ·]2 ) are finite-dimensional Lie algebras and consider
the transpose map φ : g∗2 → g∗1 , which assigns to ξ2 ∈ g∗2 the linear form φ (ξ2 ) :
g1 → F, defined for all x ∈ g1 by
 
φ (ξ2 ), x := ξ2 , φ (x) ,

where we recall that · , · stands for the canonical pairing between a vector space
and its dual. Then φ is a Poisson map, when both g∗1 and g∗2 are equipped with their
Lie–Poisson structure (denoted {· , ·}i , for i = 1, 2). To check this, it is sufficient to
check that if x, y ∈ g1  (g∗1 )∗ , then

x◦φ ,y◦φ = {x, y}1 ◦ φ . (12.6)
2

Recall that under the canonical isomorphism g1  (g∗1 )∗ , the Lie–Poisson bracket
[x, y]1 corresponds to {x, y}1 , and similarly for g2 . Also, under the isomorphism
(g∗2 )∗  g2 , the linear map x ◦ φ : g∗2 → F corresponds to φ (x), for all x ∈ g1 .
Therefore, (12.6) is equivalent to saying that φ is a Lie algebra homomorphism.
Combining both observations, it is clear that Lie algebra inclusions lead to func-
tions in involution (on the dual of the larger Lie algebra). It has the following non-
trivial consequence: if g0 is a Lie subalgebra of g, then the Ad∗ -invariant functions
on g∗0 provide involutive functions on g∗ ; notice that while the Hamiltonian vector
fields of these functions are trivial on g∗0 , they are non-trivial, in general, on g∗ .
Applied to a sequence of subalgebras, a family of functions in involution can be
constructed on the dual of a Lie algebra (equipped with its Lie–Poisson structure).
One usually refers to this technique as Thimm’s method. Explicitly, let g be a finite-
dimensional Lie algebra and suppose that

g1 ⊂ g2 ⊂ · · · ⊂ gk ⊂ gk+1 = g

where g1 , . . . , gk are Lie subalgebras of g. By the above, the functions on g∗ , obtained

by pulling back to g∗ all Ad∗ -invariant functions on each of the spaces g1 , . . . , gk+1 ,
are in involution. Notice that to this involutive family, the pull-back to g∗ of an
arbitrary involutive set on g∗1 can be added.
Thimm [191] also analyzes the case of two Lie subalgebras g1 , g2 of the same
Lie algebra g (the above case corresponds to g1 ⊂ g2 ).
Proposition 12.6. Let g be a finite-dimensional Lie algebra and let g1 and g2 be
two Lie subalgebras of g, with inclusion maps ıi , whose transpose maps g∗ → g∗i are
denoted by ıi (i = 1, 2). Suppose that

[g1 , g2 ] ⊂ g2 . (12.7)

Then for every function F1 on g∗1 and for every Ad∗ -invariant function F2 on g∗2 , the
functions F1 ◦ ı1 and F2 ◦ ı2 on g∗ are in involution.
336 12 Liouville Integrable Systems

Proof. Let us denote the Lie–Poisson brackets on g, respectively on gi , by {· , ·},

respectively {· , ·}i (i = 1, 2). For all ξ ∈ g∗ we have that
 
F1 ◦ ı1 , F2 ◦ ı2 (ξ ) = ξ , dξ (F1 ◦ ı1 ), dξ (F2 ◦ ı2 )
  
= ξ , dı (ξ ) F1 , dı (ξ ) F2 .
1 2

In view of (12.7), the latter Lie bracket belongs to g2 , so we can replace ξ by ı2 (ξ ),

which is the restriction of ξ to g2 . It follows that
   
F1 ◦ ı1 , F2 ◦ ı2 (ξ ) = ı2 (ξ ), dı (ξ ) F1 , dı (ξ ) F2 = 0 ,
1 2

where we used Ad-invariance of F2 in the last step (see (5.15)). 



12.2.5 Lax Equations

Let M ⊆ gld (F) be an affine subspace of the Lie algebra of (d × d)-matrices with
entries in F and let V be a vector field on M. Suppose that there exists a matrix-
valued function φ : M → gld (F) such that

Vx = [x, φ (x)] (12.8)

for all x ∈ M. Then V is said to be written in Lax form and (12.8) is called a Lax
equation. As we have seen in Section 7.2, if g is a quadratic Lie algebra, identified
with its dual using its bilinear form, and equipped with its Lie–Poisson structure,
then every Hamiltonian vector field on g is of this form. We claim that all coefficients
of the characteristic polynomial det(λ 1d − x), which we view as functions (of x)
on M, are constants of motion of V . To see this, let x(t) be an integral curve of V ,
defined for t in a neighborhood of 0, and denote by y(t) the value of φ at x(t) ∈ M.
Take an arbitrary i > 0 and use Trace(AB) = Trace(BA) to compute
 
d dx
Trace xi (t) = i Trace xi−1 (t) (t)
dt dt
 i 
= i Trace x (t)y(t) − xi−1 (t)y(t)x(t) = 0 .

Thus, the functions Hi : x → Trace xi are constants of motion of V . The same is true
for each coefficient of the characteristic polynomial of x, viewed as a function on M,
because each such coefficient is a polynomial in the functions Hi . In particular, if
V = XH is a Hamiltonian vector field, with Hamiltonian H, then H is in involution
with each Hi , and with each coefficient of the characteristic polynomial of x.
12.2 Constructions of Functions in Involution 337

12.2.6 The Adler–Kostant–Symes Theorem

The Adler–Kostant–Symes theorem is a fundamental theorem in the modern theory

of integrable systems, which allows one to construct, from Lie algebra splittings, on
the one hand large families of independent functions in involution and on the other
hand explicit solutions to Hamilton’s equations, with either one of these functions
as Hamiltonian. We will focus on the case in which the Lie algebra is quadratic, be-
cause in the known examples where the Adler–Kostant–Symes theorem is applied,
the Lie algebra is usually quadratic.
Let g be a finite-dimensional Lie algebra. We recall from Section 10.1 that a
linear map R : g → g is called an R-matrix of g if the skew-symmetric bilinear map
[· , ·]R : g × g → g, defined for x, y ∈ g by

1
[x, y]R := ([Rx, y] + [x, Ry]) ,
2
defines a Lie bracket on g. If g is quadratic, i.e., g is equipped with an ad-invariant
non-degenerate symmetric bilinear form · | · , then the original Lie bracket [· , ·] and
the R-bracket [· , ·]R lead to two Lie–Poisson structures {· , ·} and {· , ·}R on g. In
order to describe these, recall that to a function F ∈ F (g) and an element x ∈ g, an
element ∇x F of g is associated by setting ∇x F | · = dx F, that is,

∇x F | y = dx F, y ,

for all y ∈ g. Using this notation, the two Lie–Poisson structures on g take the form

{F, G} (x) = x | [∇x F, ∇x G] ,

(12.9)
{F, G}R (x) = x | [∇x F, ∇x G]R ,

for F, G ∈ F (g) at x ∈ g (see Section 7.2). The involutivity theorem yields functions
in involution, coming from the R-bracket; moreover, their Hamiltonian vector fields
can be written in a Lax form.
Theorem 12.7 (Involutivity theorem). Let (g, [· , ·]) be a finite-dimensional quadra-
tic Lie algebra, whose bilinear form is denoted by · | · . Let R be an R-matrix
for g and denote the Lie–Poisson bracket on g, which corresponds to the R-bracket,
by {· , ·}R . If F and H are Ad-invariant functions on g, then
(1) {F, H}R = 0;
(2) The Hamiltonian vector field of H with respect to {· , ·}R is given, at x ∈ g, by

1
(XH )x = − [x, R(∇x H)] . (12.10)
2
Proof. Recall that the Ad-invariance of H means that [∇x H, x] = 0 for all x ∈ g (see
Section 5.1.4). For functions F and H on g, their R-bracket is given, at x ∈ g, by
338 12 Liouville Integrable Systems

1 1
{F, H}R (x) = x | [R (∇x F) , ∇x H] + x | [∇x F, R (∇x H)]
2 2
1 1
= R (∇x F) | [∇x H, x] + R (∇x H) | [x, ∇x F] ,
2 2
which is zero when F and H are both Ad-invariant; if only H is assumed Ad-
invariant, only the first term vanishes, leading to
1 1
{F, H}R (x) = ∇x F | [R (∇x H) , x] = dx F, [R (∇x H) , x] ,
2 2
which proves (12.10). 


We will now refine the involutivity theorem to obtain the (historically older) Adler–
Kostant–Symes theorem, in which the R-matrix comes from a Lie algebra splitting.
Recall from Section 10.1.2 that if (g, [· , ·]) is a finite-dimensional 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, and that such a splitting
leads to a natural R-matrix. Denoting the projections of x ∈ g on g+ and g− by x+ ,
respectively x− , the endomorphism R : g → g is defined by

Rx := x+ − x− ,

for all x ∈ g. As in the case of the involutivity theorem, we assume that g is a

quadratic Lie algebra, with bilinear form · | · . The Lie–Poisson bracket {· , ·}R on
g can then also be written as

{F, G}R (x) = x | (∇x F)+ , (∇x G)+ − x | (∇x F)− , (∇x G)− .

Non-degeneracy of · | · implies that g admits another (vector space) direct sum

decomposition, namely as g = g⊥ ⊥ ⊥ ⊥
+ ⊕ g− where g+ (respectively g− ) is the orthogonal
complement (with respect to · | · ) of g+ (respectively g− ).
The Adler–Kostant–Symes theorem (AKS theorem), which we formulate and
prove next, yields functions in involution on certain translates of g⊥
− (items (1)–(3)
below) and the explicit integration of their Hamiltonian vector fields (item (4)).
Theorem 12.8 (AKS theorem). Suppose that g = g+ ⊕ g− is a Lie algebra split-
ting, with associated R-matrix R, and that · | · is an Ad-invariant non-degenerate
symmetric bilinear form on g. Let F and H be Ad-invariant functions on g and
suppose that ε ∈ g satisfies

[ ε , g+ ] ⊂ g ⊥
+, [ ε , g− ] ⊂ g ⊥
−. (12.11)

The restrictions of F and H to ε + g⊥

− are denoted by Fε and Hε .

(1) The affine subspace ε + g⊥ − of g is a Poisson submanifold of (g, {· , ·}R ); the

restricted Poisson structure on ε + g⊥− will be denoted by {· , ·}ε ;
(2) {Fε , Hε }ε = 0;
12.2 Constructions of Functions in Involution 339

(3) The Hamiltonian vector field of Hε with respect to {· , ·}ε is given by

1
XHε (y) = − [y, R(∇y H)] = ±[y, (∇y H)∓ ] ,
2
where y ∈ ε + g⊥ −;
(4) Let G be a Lie group, whose Lie algebra is g, and let G+ and G− denote the
Lie subgroups of G, corresponding to g+ , respectively g− . For x0 ∈ g and for
|t| small, let g+ (t) and g− (t) denote the smooth curves in G+ respectively G−
which solve the factorization problem1

exp(−t∇x0 H) = g+ (t)−1 g− (t) , g± (0) = e . (12.12)

Then the integral curve of XH := {· , H}R , which starts at x0 ∈ g (or in par-

ticular of XHε , starting at x0 ∈ ε + g⊥
− ), is given, for |t| small, by

x(t) = Adg+ (t) x0 = Adg− (t) x0 . (12.13)

Proof. We first show that ε + g⊥ ⊥

− is a Poisson submanifold of (g, {· , ·}R ). Let x ∈ g−
and let H be a function, defined in a neighborhood of ε +x in g. We need to show that
(XH )ε +x ∈ Tε +x g⊥ ⊥
−  g− . Let F be an arbitrary function, defined on a neighborhood
of ε + x in g, which is constant on ε + g⊥− . Then ∇ε +x F ∈ g− since

d
∇ε +x F | y = F(ε + x + ty) = 0 ,
dt |t=0

for all y ∈ g⊥
− . It follows from ad-invariance of · | · and from the assumptions (12.11)
on ε that

(XH )ε +x [F] = {F, H}R (ε + x)

= ε + x | (∇ε +x F)+ , (∇ε +x H)+ − (∇ε +x F)− , (∇ε +x H)−
= x | (∇ε +x F)+ , (∇ε +x H)+ − x | (∇ε +x F)− , (∇ε +x H)− .

Both of the above terms are equal to zero: the first one because ∇ε +x F ∈ g− and
the second one because x ∈ g⊥ − , combined with the fact that g− is a Lie subalge-
bra (of g). This proves that all Hamiltonian vector fields XH are tangent to ε + g⊥ −,
hence that ε +g⊥ − is a Poisson submanifold of (g, {· , ·}R ), which is the content of (1).
Since (ε + g⊥ − , {· , ·}ε ) is a Poisson submanifold of (g, {· , ·}R ), the involutivity of
two Ad-invariant functions F and H on g (Theorem 12.7) implies the involutivity
of Fε and Hε on ε + g⊥ − , which proves (2); similarly, the first equality in (3) fol-
lows from (1) and from the involutivity theorem (Theorem 12.7), while the second
equality is a direct consequence of the obvious formulas

R (∇x H) = (∇x H)+ − (∇x H)− = 2(∇x H)+ − ∇x H = ∇x H − 2(∇x H)− .

1 For g in a neighborhood of the identity e ∈ G, the factorization g = g+ g− , with g± ∈ G± exists
and is unique.
340 12 Liouville Integrable Systems

We now turn to (4), the integral curves of XH . We first show that Adg+ (t) x0 =
Adg− (t) x0 (the second equality in (12.13)). Since Ad is a group homomorphism, the
factorization (12.12) implies that

Adexp(−t∇x H) x0 = Adg+ (t)−1 Adg− (t) x0 ,

0

for all x0 ∈ g. We have for every x ∈ g that

   
Adexp(−t∇x H) x0 | x = x0 | Adexp(t∇x H) x
0 0
   
= x0 | exp t ad∇x0 H x
= x0 | x + t x0 | ∇x0 H, 
= x0 | x + t x0 , ∇x0 H | 
= x0 | x ,

where the value of  (which depends on t) is irrelevant, in view of the Ad-invariance

of H and · | · . This shows that Adexp(−t∇x H) x0 = x0 , and hence that Adg+ (t) x0 =
0
Adg− (t) x0 .
We now show that x(t) := Adg+ (t) x0 is (for small |t|) a solution to (12.10), which
amounts to proving that
d    
Adg+ (t) x0 = − x(t), ∇x(t) H + . (12.14)
dt
In order to simplify the proof (mainly the notation), we will assume that G is a linear
group, so that Ad is just given by conjugation and we will write ġ± (t) for dgdt± (t).
Then the left-hand side of (12.14) is given by ġ+ (t)g+ (t)−1 , x(t) , so it suffices to
show that
ġ+ (t)g+ (t)−1 = (∇x(t) H)+ .
Differentiating the identity g+ (t) exp(−t∇x0 H) = g− (t) with respect to t and multi-
plying both sides of the result by g− (t)−1 , we get
 
ġ+ (t)g+ (t)−1 − g− (t) ∇x0 H g− (t)−1 = ġ− (t)g− (t)−1 .

Taking the + part of both sides of this equation, we find

     
ġ+ (t)g+ (t)−1 = g− (t) ∇x0 H g− (t)−1 + = ∇x(t) H + ,

where we have used Ad-invariance of H and of · | · in the last step. 


12.3 The Liouville Theorem and the Action-Angle Theorem 341

12.3 The Liouville Theorem and the Action-Angle Theorem

In this section, we formulate and prove two basic theorems in the theory of inte-
grable systems: the Liouville theorem and the action-angle theorem. Classically,
these theorems are considered for Liouville integrable systems on symplectic man-
ifolds; we consider these theorems here in the more natural context of Poisson
manifolds. We follow [3] in Section 12.3.1 and [120] in Sections 12.3.2 to 12.3.4.
Throughout the section, all manifolds are real and F (M) stands for the algebra of
smooth functions on the (real) manifold M.

12.3.1 Liouville Integrable Systems and Liouville’s Theorem

We first give the definition of a Liouville integrable system on a real Poisson mani-
fold.
Definition 12.9. Let (M, π ) be a real Poisson manifold of rank 2r and of dimen-
sion d. Let F = (F1 , . . . , Fs ) be an s-tuple of smooth functions on M and suppose
that
(1) F is independent;
(2) F is in involution;
(3) r + s = d.
Then (M, π , F) is called a Liouville integrable system of dimension d and rank 2r.
Viewed as a map, F : M → Rs is called the momentum map of (M, π , F).
Suppose that (M, π , F) is a Liouville integrable system, where (M, π ) is a Poisson
manifold of rank 2r. We recall that M(r) denotes the open subset of M where the rank
of π is equal to 2r, and UF denotes the dense open subset of M, which consists of
all points of M where the differentials of the elements of F are linearly independent.
We consider on the non-empty open subset UF ∩ M(r) of M two (regular) foliations:
(1) The Hamiltonian vector fields XF1 , . . . , XFs define, according to Proposi-
tions 12.2 and 12.3, on the non-empty open subset UF ∩ M(r) of M an inte-
grable distribution D of rank r. The integral manifolds of D are the leaves
of a foliation, which we denote by F ; the leaf of F , passing through
m ∈ UF ∩ M(r) , is denoted by Fm , and is called the invariant manifold of F,
through m;
(2) Since, by definition, the restriction F of F to UF ∩ M(r) is a submersion, its
fibers define a foliation on UF ∩ M(r) , which is denoted by F . Thus, the leaves
of F are the connected components of the fibers of the map

F = (F1 , . . . , Fs ) : UF ∩ M(r) → Rs .

We show in the following proposition that both foliations F and F coincide, lead-
ing to a description of the invariant manifolds of F.
342 12 Liouville Integrable Systems

Proposition 12.10. Let (M, π , F) be a Liouville integrable system and let F and F
denote the foliations, defined above.
(1) The foliations F and F coincide;
(2) For m ∈ UF ∩ M(r) , the invariant manifold Fm of F is the connected compo-
nent of the fiber of F which contains m. In particular, Fm is an (embedded)
submanifold of M.

Proof. We prove item (1); item (2) is an immediate consequence of it. Since all
leaves of F and of F are r-dimensional (and are, by definition, connected), it suf-
fices to show for every m ∈ UF ∩ M(r) the following inclusion of tangent spaces:
Tm F ⊂ Tm F . According to Proposition 12.2, the fact that F1 , . . . , Fs are pairwise
in involution implies that each of the Hamiltonian vector fields XF1 , . . . , XFs , at m,
belong to Tm F . Since Tm F is spanned by these vector fields, the required inclusion
follows.  

We now come to the Liouville theorem for Liouville integrable systems on Poisson
manifolds.

Theorem 12.11 (Liouville’s theorem). Let (M, π , F) be a Liouville integrable sys-

tem of dimension d and rank 2r. For m ∈ UF ∩ M(r) , let Fm denote the invariant
manifold of F = (F1 , . . . , Fs ) which passes through m.
(1) If Fm is compact, then there exists a diffeomorphism from Fm to the torus
Tr = (R/Z)r , under which the vector fields XF1 , . . . , XFs are mapped to con-
stant (i.e., translation-invariant) vector fields;
(2) If Fm is not compact, but the flow of each of the vector fields XFi (i = 1, . . . , s)
is complete on Fm , then there exists a diffeomorphism from Fm to a cylin-
der Rr−q × Tq (0  q < r), under which the vector fields XFi are mapped to
constant vector fields.

Proof. We suppose that the flow Φ (i) of each of the integrable vector fields XFi is
complete on Fm , i.e., is defined for all t ∈ R. We may order the functions Fi such
that the r vector fields XF1 , . . . , XFr are independent at m. These vector fields are
then independent at every point of Fm . Indeed, since the vector fields XFi pairwise
commute,
r
LXF j (XF1 ∧ · · · ∧ XFr ) = ∑ XF1 ∧ · · · ∧ XFj , XFi ∧ · · · ∧ XFr = 0 ,
i=1

for j = 1, . . . , s. It means that XF1 ∧ · · · ∧ XFr is conserved by the flow of each one of
the vector fields XF1 , . . . , XFs . In particular, since this r-vector field is non-vanishing
at m, it is non-vanishing on the entire invariant manifold Fm . As a consequence, Fm
is a leaf of the distribution, defined by the first r vector fields XF1 , . . . , XFr , in a
neighborhood of Fm .
The completeness and commutativity of the vector fields XF1 , . . . , XFr on Fm
imply that we can define an action Rr × Fm → Fm by
12.3 The Liouville Theorem and the Action-Angle Theorem 343
  (r)  
(t1 , . . . ,tr ), m → Φt1 ◦ Φt2 ◦ · · · ◦ Φtr m .
(1) (2)

Since Fm is the integral manifold through m of the distribution defined by the first
r integrable vector fields, the action is transitive on Fm and Fm becomes a homo-
geneous space. The action is also locally free, because the vector fields XFi are
independent at every point of Fm . Therefore the stabilizer is a discrete subgroup
Hm of Rr and Fm is diffeomorphic to Rr /Hm . If Fm is compact, then Hm must be a
lattice, so Rr /Hm is a torus, smoothly embedded into M. Otherwise Hm is a discrete
subgroup whose rank q is at most r − 1 and Rr /Hm is isomorphic to Rr−q × Tq . By
construction, the vector fields XFi are mapped to translation-invariant vector fields
in both cases.  

For what follows, we will be uniquely interested in the case in which Fm is

compact. Then Fm is a torus; we call such a torus a standard Liouville torus.

12.3.2 Foliation by Standard Liouville Tori

We show in this section that, in some neighborhood of a standard Liouville torus,

the invariant manifolds of a Liouville integrable system (M, π , F) of dimension d
and rank 2r form a trivial torus fibration over an open ball Bd−r .

Proposition 12.12. Suppose that Fm is a standard Liouville torus of a Liouville in-

tegrable system (M, π , F) of dimension d and rank 2r. There exists an open sub-
set U ⊂ UF ∩ M(r) , containing Fm , and there exists a diffeomorphism φ : U →
Tr × Bd−r , which takes the foliation F to the foliation defined by the fibers of the
canonical projection pB : Tr × Bd−r → Bd−r , leading to the following commutative
diagram.
φ
F m = Fm U  Tr × Bd−r

F|U
pB

Bd−r
Moreover, U and φ can be chosen such that the last d − 2r components of φ are
Casimirs of π , restricted to U.

Proof. Suppose that Fm is a standard Liouville torus. Setting s := d − r, as before,

we first show that there exists a neighborhood U of Fm and a diffeomorphism φ :
U → Fm × Bs , which sends the foliation F (which, according to Proposition 12.10,
coincides with F ), restricted to U, to the foliation defined by pB on Fm × Bs . The
proof of this statement depends only on the fact that Fm is a compact component
of a fiber of a submersion (namely F; see Proposition 12.10). Notice that since F
is a submersion, every point m ∈ F m = Fm has a neighborhood Um in M, which
344 12 Liouville Integrable Systems

is diffeomorphic to the product of a neighborhood Vm of m in Fm times an open

ball Bsm , centered at F(m ) = F(m) in Rs ; such a diffeomorphism φm , as provided
by the implicit function theorem, is a lifting of F, that is, it leads to the following
commutative diagram:
φm
Um Vm × Bsm

pB
F

Bsm

Since Fm is compact, it is covered by finitely many of the sets Vm , say Vm1 , . . . ,Vm .
Thus, if every pair of the diffeomorphisms φm1 , . . . , φm agrees on the intersection
of their domain of definition (whenever non-empty), we can define a global dif-
feomorphism on a neighborhood U of Fm , whose image is the intersection of the
concentric balls Bsm1 , . . . , Bsm . In order to ensure that these diffeomorphisms agree,
we need to chose them in a more specific way. This is done by choosing an arbitrary
Riemannian metric on M. Using the exponential map, defined by the metric, we
can identify a neighborhood of the zero section in the normal bundle of Fm , with
a neighborhood of Fm in M; in particular, for every m ∈ Fm there exist neighbor-
hoods Um of m in M and Vm of m in Fm , with smooth maps ψm : Um → Vm , which
have the important virtue that they agree on the intersection of their domains. Upon
shrinking the open subsets Um , if necessary, the maps φm := ψm × (F1 , . . . , Fs ) are a
choice of diffeomorphisms, defined on a neighborhood U of Fm , with the required
properties. This leads to the diffeomorphism φ : U → Fm × Bs , which we view as a
diffeomorphism φ : U → Tr × Bs , using the fact that Fm is diffeomorphic with Tr ,
as stated in the Liouville theorem.
It remains to be shown that φ can be chosen such that its last d − 2r = s − r com-
ponents are Casimirs. To do this, we consider on U the integrable distribution D 
defined by all Hamiltonian vector fields on U; it has rank 2r and its leaves are the
symplectic leaves of (U, π ). It is clear that D ⊂ D  , where we recall that D is the
distribution defined by the Hamiltonian vector fields XF1 , . . . , XFs . Consider the
submersive map
pB ◦ φ : U → Tr × Bs → Bs ,
whose fibers are by assumption the leaves of F , that is, the integral manifolds of D
(restricted to U), so that the kernel of the tangent map T (pB ◦ φ ) is precisely D. The
image of D  by T (pB ◦ φ ) is therefore a (smooth) distribution D  of rank r on Bs ,
which is integrable, since D  is integrable. The foliation defined by the integral
manifolds of D  is, in the neighborhood of the point pB (φ (m)), defined by s − r =
d − 2r independent functions z1 , . . . , zd−2r . Pulling them back to M, we get functions
z1 , . . . , zd−2r on a neighborhood V ⊂ U of Fm which are Casimir functions, because
they are constant on the leaves of D  , which are the symplectic leaves of (U, π ).
12.3 The Liouville Theorem and the Action-Angle Theorem 345

Since they have independent differentials on U, they can be chosen as d − 2r of the

last d − r components of φ , upon shrinking Bs into a smaller ball, if needed. 


12.3.3 Period Normalization

In this section, we study Liouville integrable systems of the form (Tr × Bs , π , F),
where F = pB : Tr × Bs → Bs is the projection on the second factor and π is a
regular Poisson structure of rank 2r on Tr × Bs . Given such a Liouville integrable
system, there is no reason why the Hamiltonian vector fields XFi , which are tangent
to the fibers of F (which are tori), would be constant on these fibers, and even if they
were constant on these fibers, they might vary from one fiber to another in the sense
that their flow may not come from an action of the torus Tr on Tr × Bs . We show
in the following proposition that vector fields which satisfy the latter property can
be constructed by taking well-chosen linear combinations of the Hamiltonian vector
fields XFi , with as coefficients F-basic functions, i.e., functions of the form λ ◦ F,
where λ ∈ F (Bs ); equivalently, smooth functions on Tr × Bs which are constant on
the fibers of F.
Proposition 12.13. Let (Tr × Bs , π , F) be a Liouville integrable system, where π
has constant rank 2r and F = (F1 , . . . , Fs ) is the projection on the second compo-
nent. Suppose that among the s commuting vector fields XF1 , . . . , XFs , the first r
are independent at every point of Tr × Bs . There exists a ball Bs0 ⊂ Bs , concentric
with Bs , and there exist F-basic functions λij ∈ F (Bs0 ), such that the vector fields
V1 , . . . , Vr , defined by
r
Vi := ∑ λi j X F j ,
j=1

are the fundamental vector fields of a torus action of Tr on Tr × Bs0 .

(i)
Proof. The fibers of F = (F1 , . . . , Fs ) are compact, so for i = 1, . . . , r, the flow Φti
of the Hamiltonian vector field XFi is complete and we can define a map,

Φ : Rr × (Tr × Bs ) → Tr × Bs
(1) (r)
((t1 , . . . ,tr ), m) → Φt1 ◦ · · · ◦ Φtr (m) .

(i)
Since the vector fields XFi are pairwise commuting, the flows Φti pairwise com-
mute and Φ is an action of Rr on Tr × Bs . Since the vector fields XF1 , . . . , XFr are
independent at all points of Tr × Bs , the fibers of F, which are r-dimensional tori,
are the orbits of the action. For c ∈ Bs , let Λc denote the lattice of Rr , which is the
isotropy group of every point in F−1 (c); it is the period lattice of the action Φ , re-
stricted to F−1 (c). In order to simplify the notation, we assume that Bs is centered
at o.
Let mo be an arbitrary point of F−1 (o) and choose a basis (λ1 (o), . . . , λr (o)) of
the lattice Λo . For a fixed i, with 1  i  r, for m in a neighborhood of mo in Tr × Bs
346 12 Liouville Integrable Systems

and for Li in a neighborhood of λi (o) in Rr , consider the equation Φ (Li , m) = m.

Since F(Φ (Li , m)) = F(m) for all Li and m, it is meaningful to write Φ (Li , m) −
m and solving the equation Φ (Li , m) = m locally for Li amounts to applying the
implicit function theorem to the map
Φ (Li ,m)−m
Rr × (Tr × Bs ) Tr × Bs Tr .

Since the action is locally free, the Jacobian condition is satisfied and we get by
solving for Li around λi (o) a smooth Rr -valued function λi (m), defined for m in
a neighborhood Wi of mo . Doing this for i = 1, . . . , r and setting W := ∩ri=1Wi , we
have that W is a neighborhood of mo , and on W we have functions λ1 (m), . . . , λr (m),
with the property that Φ (λi (m), m) = m for all m ∈ W and for all 1  i  r. Thus,
λ1 (m), . . . , λr (m) belong to the lattice ΛF(m) for all m ∈ W and they form a basis
when m = mo ; by continuity, they form a basis of ΛF(m) for all m ∈ W .
The functions λi can be extended to a neighborhood of the torus F−1 (o). In
fact, the functions λi are F-basic, hence extend uniquely to F-basic functions on
F−1 (F(W )). We will use in the sequel the same notation λi for these extensions and
we write F−1 (F(W )) simply as W . Using these functions, we define the following
smooth map:
Φ̃ : Rr ×W →  W 
r
(12.15)
((t1 , . . . ,tr ), m) → Φ ∑ ti λi (m), m .
i=1

Since the functions λi are F-basic, the fact that Φ is an action implies that Φ̃ is an
action. This action has the extra feature that the stabilizer of every point in W is Zr .
Thus, Φ̃ induces an action of Tr on W , which we still denote by Φ̃ . By shrinking W ,
if necessary, we may assume that W is of the form F−1 (Bs0 ), where Bs0 is an open
ball, concentric with Bs , and contained in it. Thus we have a torus action

Φ̃ : Tr ×W →  W 
r
((t1 , . . . ,tr ), m) → Φ ∑ ti λi (m), m .
i=1

Looking at the action on a single point m ∈ W , it is clear that the fundamental

vector fields of Φ̃ are expressible as linear combinations of the Hamiltonian vector
fields XFi , with F-basic functions as coefficients. 


12.3.4 The Existence of Action-Angle Coordinates

We formulate and prove in this section the action-angle theorem. The following
lemma plays a key rôle in the proof.
12.3 The Liouville Theorem and the Action-Angle Theorem 347

Lemma 12.14. Let M be a manifold, equipped with a vector field V and a bivector
field P. Suppose that V is complete and that each one of its integral curves has
period 1. If LV2 P = 0, then LV P = 0.
Proof. Let Q be the bivector field on M, defined by Q := LV P. We pick an arbitrary
point m and we show that Qm = 0. Denoting the flow of V by Φt , we have for all t
that
d 
(Φt )∗ PΦ−t (m) = (Φt )∗ (LV P)Φ−t (m) = (Φt )∗ QΦ−t (m) = Qm , (12.16)
dt
where we used in the last step that the bivector field Q satisfies LV Q = 0. By inte-
grating (12.16),
(Φt )∗ PΦ−t (m) = Pm + tQm .
Evaluated at t = 1 this yields Qm = 0, since Φ1 = Φ−1 = 1M , as V has period 1.



Theorem 12.15 (Action-angle theorem). Let (M, π , F) be a Liouville integrable

system, where (M, π ) is a Poisson manifold of dimension d and rank 2r. Sup-
pose that Fm is a standard Liouville torus, where m ∈ UF ∩ M(r) . Then there ex-
ist R-valued smooth functions (p1 , . . . , pd−r ) and R/Z-valued smooth functions
(θ1 , . . . , θr ), defined in a neighborhood U of Fm , such that
(1) The map, defined by (θ1 , . . . , θr , p1 , . . . , pd−r ), is a diffeomorphism between U
and Tr × Bd−r ;
(2) The Poisson structure π can be written in terms of these coordinates as
r
∂ ∂
π=∑ ∧ ,
i=1 ∂ θi ∂ pi

in particular the functions pr+1 , . . . , pd−r are Casimirs of π (restricted to U);

(3) The leaves of the surjective submersion F = (F1 , . . . , Fd−r ) are given by the
projection onto the second component Tr × Bd−r , in particular, the functions
F1 , . . . , Fd−r depend on the functions p1 , . . . , pd−r only.
The functions θ1 , . . . , θr are called angle coordinates, the functions p1 , . . . , pr are
called action coordinates and the remaining functions pr+1 , . . . , pd−r are called
transverse coordinates.
Proof. We denote s := d − r, as before. The hypotheses imply, according to Proposi-
tion 12.12 that we can assume that (M, π , F) is of the form (Tr ×Bs , π , pB ), where Bs
is a ball, centered at the origin of Rs . We do this, but we still write F for pB , denoting
its components as ( f1 , . . . , fr , z1 , . . . , zs−r ), since according to the latter proposition,
the last components of F can be assumed to be Casimir functions. According to
Proposition 12.13 there exist F-basic functions λij ∈ F (Tr × Bs ), such that the r
vector fields
r
Vi := ∑ λi j X f j , (12.17)
j=1
348 12 Liouville Integrable Systems

where i = 1, . . . , r, are the fundamental vector fields of an action of Tr on Tr × Bs .

We first prove that these vector fields are Poisson vector fields, i.e., that LVi π =
[Vi , π ]S = 0 for i = 1, . . . , r. Since each Vi has period 1 and is complete, it suffices
to show, according to Lemma 12.14, that LV2i π = [Vi , [Vi , π ]] = 0, for i = 1 . . . , r.
Since each of the coefficients λij which appear in (12.17) is an F-basic function,
it suffices to show that if F, F  , G and G are F-basic functions on Tr × Bs , then
[FXG , [F  XG , π ]S ]S = 0. In fact, Hamiltonian vector fields leave π invariant, F-
basic functions are in involution and their Hamiltonian vector fields commute, so
that the graded Leibniz identity for the Schouten bracket yields

FXG , F  XG , π S S
= [FXG , XF  ∧ XG ]S
= F [XG , XF  ∧ XG ]S + XG [F, XF  ∧ XG ]S = 0 ,

as was to be shown.
We proceed to construct the action coordinates: we show that the Poisson vector
fields Vi are actually Hamiltonian vector fields (upon shrinking Br , if necessary),
where the Hamiltonians can be taken as F-basic functions; these Hamiltonians will
yield the action coordinates. According to Theorem 1.28, we can construct on a
neighborhood U  of m in Tr × Bs functions g1 , . . . , gr such that

( f1 , . . . , fr , g1 , . . . , gr , z1 , . . . , zs−r )

is a system of coordinates on U  , in which the Poisson structure takes the form

r
∂ ∂
π=∑ ∧ .
i=1 ∂ f i ∂ gi

In what follows, we make no notational distinction between the F-basic functions

f1 , . . . , fr , z1 , . . . , zs−r , λij , considered as functions on F(U  ) ⊂ Bs , and the functions
f1 , . . . , fr , z1 , . . . , zs−r , λij themselves, defined on U  . With this notation, the vector
fields Xλ j and Vi can be written as
i

r
∂ λi j r r
∂
Xλ j = ∑ X fk , and Vi = ∑ λi j X f j = − ∑ λi j ∂ g j .
i
k=1 ∂ f k j=1 j=1

Expressing that Vi is a Poisson vector field, [Vi , π ]S = 0, then takes the explicit form
 
r
∂ ∂ ∂ r
∂ λi j ∂ ∂
− ∑ λi j , ∧
∂ g j ∂ f k ∂ gk
= ∑ ∂ f k ∂ g j ∧ ∂ gk = 0 ,
j,k=1 S j,k=1

which is rewritten in terms of the Hamiltonian vector fields X f j as

 
∂ λij ∂ λik
∑ ∂ fk
−
∂ fj
X f j ∧ X fk = 0 . (12.18)
1 j<kr
12.3 The Liouville Theorem and the Action-Angle Theorem 349

Since the vector fields X f1 , . . . , X fr are linearly independent at all points of U  , all
coefficients of (12.18) vanish and we get, for every i, j, k ∈ {1, . . . , r}:

∂ λi j ∂ λik
= . (12.19)
∂ fk ∂ fj

As in the proof of the classical Poincaré lemma, the smooth functions p1 , . . . , pr

on F(U  ), defined by
r  1
pi = pi ( f1 , . . . , fs ) := ∑ λij (t f1 , . . . ,t fr , z1 , . . . , zs−r ) f j dt (12.20)
j=1 t=0

satisfy
∂ pi
λi j = , (12.21)
∂ fj
for all 1  i, j  r. As above, we view these functions as functions on U  . They are
Hamiltonians of the vector fields Vi , because
r
∂ pi r
X pi = ∑ ∂ fj Xfj = ∑ λij X f j = Vi . (12.22)
j=1 j=1

This shows that each one of the vector fields Vi is a Hamiltonian vector field on U  .
Since the functions pi and z j are F-basic, each admits a unique extension to an
F-basic function on F(F−1 (U  )), a neighborhood of Fm which we call U  in the
sequel. On U  , we still have Vi = X pi .
We proceed to the construction of the angle coordinates. In view of Theo-
rem 1.28, there exist on a neighborhood U  ⊂ U  of m in Tr × Bs , smooth functions
θ1 , . . . , θr such that
r
∂ ∂
π=∑ ∧ . (12.23)
j=1 ∂ θ j ∂ pj

On U  , X p j = ∂∂θ j , for j = 1, . . . , r; since each of these vector fields has period 1, it

is natural to view the functions θ1 , . . . , θr as R/Z-valued functions, which we will do
without changing the notation. Notice that the functions θ1 , . . . , θr are independent
and pairwise in involution on U  , as a trivial consequence of (12.23). In particular,
θ1 , . . . , θr , p1 , . . . , ps define local coordinates on U  . In these coordinates, the action
of Tr is given by

(t1 , . . . ,tr ) · (θ1 , . . . , θr , p1 , . . . , ps ) = (θ1 + t1 , . . . , θr + tr , p1 , . . . , ps ) , (12.24)

so that the functions θi uniquely extend to smooth R/Z-valued functions satisfy-

ing (12.24), on U := F−1 (F(U  )), which is an open neighborhood  of Fm in M; the
extended functions are still denoted by θi . It is clear that θi , p j = δi, j on U, for
all i, j = 1, . . . , r. Combined with the Jacobi identity, this leads to
350 12 Liouville Integrable Systems
        
X pk [ θi , θ j ] = θi , θ j , pk = θi , δ j,k − θ j , δi,k = 0 ,
 
which shows that the Poisson brackets θi , θ j are invariant under the Tr -action;
but the latter vanish on U  , hence these brackets vanish at every point of U. We
may conclude that on U, the functions (θ1 , . . . , θr , p1 , . . . , ps ) have independent dif-
ferentials, so they define a diffeomorphism to Tr × Bs where Bs is a (small) ball
with center 0, and that the Poisson structure takes in terms of these coordinates the
canonical form (12.23), as required.  
In terms of action-angle coordinates, the Hamiltonian vector fields X fi take a
particularly simple form. In fact, since the functions Fi depend on the coordinates pi
only, the equations of motion, associated to the Hamiltonian Fi are given by

  ∂ Fi
ṗ j = p j , Fi = − =0, j = 1, . . . , r ,
∂θj
  ∂ Fi
θ̇ j = θ j , Fi = = cj , j = 1, . . . , r ,
∂ pj
żk = 0 , k = 1, . . . , s − 2r ,

where c j depends on the p and z coordinates only. Integrating the first and third
equations, we find that all p j and zk are constant (which is not surprising since the
Fj are constants of motion and the zk are Casimirs), so that the c j are constant,
which yields upon using the second equation that the angle coordinates evolve lin-
early, θ j (t) = c j t, for j = 1, . . . , r. Thus, action-angle coordinates not only exhibit
an integrable system locally as a Hamiltonian system with linear flow(s) on a family
of tori, they are also the coordinates in which the equations of motion take their
simplest possible form.
The action-angle theorem admits a natural generalization to the case of non-
commutative integrable systems (see [120]). The classical examples of Liouville
integrable systems can be found in [204]. For a modern treatment of Liouville inte-
grability, including recently discovered examples, we refer to [3, 164].

12.4 Notes

The term “integrable system” appears in a large variety of mathematical and physi-
cal contexts, referring often more to phenomena than to a precise general definition.
The notion of a Liouville integrable system, detailed in this chapter, is a notable
exception. Liouville’s observation that “Liouville integrable systems” can be solved
by quadratures has been a main impulse to the development of Poisson geometry,
for a long time restricted to symplectic geometry (the geometry of the phase space
of a mechanical system). For an excellent account of what was classically known
about integrable systems, the reader should consult Whittaker [204]. The books by
Abraham–Marsden [1], Arnold [15] and Libermann–Marle [125] are all devoted
12.4 Notes 351

partly to integrability in classical mechanics, in the setting of symplectic manifolds,

although Lie groups and Lie algebras are present throughout these books, just as
Poisson brackets are. See also the book by Kosmann–Schwarzbach [114], which
focuses on the Noether theorems, with special emphasis on their historical develop-
ments, as well as on their influence in mathematics and physics.
Shortly before 1970, the remarkable discovery was made that the Korteweg–de
Vries equation can be solved by the inverse spectral method, has an infinite num-
ber of constant of motion, has a (in fact many) Hamiltonian structure(s) and can be
described as an isospectral flow of a Sturm–Liouville operator (as a Lax equation).
It has very much revived the interest in integrable systems, in particular in Liou-
ville integrable systems, on which the newly discovered infinite-dimensional tools
(inverse spectral method, Lax equations, recursion operators), were successfully ap-
plied, thereby shedding a new light on some classical results, leading to the solution
of some classical problems and to the discovery of some new integrable systems.
See the books by Perelomov [164] and Fomenko [78], which together cover a large
spectrum of examples and phenomena. From the Poisson point of view, the im-
portant observation by Adler that the Poisson structure of the Korteweg–de Vries
equation comes from a Lie algebra splitting led to the notion of an R-bracket, which
in turn was at the basis of a better understanding of the symmetries of quantum sys-
tems, culminating in the discovery of quantum groups, see Faddeev–Takhtajan [72]
and Chari–Pressley [40].
The presence of algebraic geometry in integrable systems, already present in
the classical works of Euler, Jacobi, Painlevé and Kowalevskaya, was reaffirmed
by Its–Matveev, who showed that the theta functions of hyperelliptic curves lead
to solutions of the Korteweg–de Vries equation, and by Adler–van Moerbeke, who
showed upon building upon Kowalevskaya’s intuition that basically all classical ex-
amples of integrable systems are algebraically integrable, see Adler–van Moerbeke–
Vanhaecke [3, 196]. The underlying Lie and Poisson geometry of these systems is
always very rich, but often only partly understood.
Chapter 13
Deformation Quantization

In this chapter we present a second application of Poisson structures: the theory of

deformations of commutative associative algebras, also called deformation quanti-
zation. Geometric quantization and deformation quantization are two approaches to
the process of forming a quantum mechanical system from a given classical sys-
tem. Both are based on Dirac’s observation that the natural quantization rules for
Euclidean-invariant systems yield for the simplest variables (position and momen-
tum) differential operators which obey commutation relations analogous to the Pois-
son bracket for the corresponding classical variables. We will only discuss deforma-
tion quantization here, which consists of considering for a given Poisson manifold
(M, {· , ·}) all associative products  on the vector space C∞ (M)[[ν ]], which are of
the form:
1
F  G = FG + {F, G} ν +C(F, G)ν 2 + · · ·
2
Kontsevich’s formality theorem implies that for every Poisson manifold, such a de-
formation quantization exists.
The problem of describing for a given commutative associative algebra (A , ·)
all associative structures on A [[ν ]], which deform the product “·” is also interesting
from the mathematical point of view. Since even in this more general context, any
such deformation makes a Poisson bracket appear on (A , ·), the question is most
naturally asked when this Poisson bracket is fixed in advance, i.e., when (A , ·, {· , ·})
is a Poisson algebra. As we will see, the problem of describing all Poisson brackets
on (A [[ν ]], ·) which deform the Poisson bracket {· , ·} is related to the deformation
quantization problem.
Formal deformations of associative products and of Poisson structures are intro-
duced in Sections 13.1 and 13.2. Section 13.3 is devoted to the basics on differential
graded Lie algebras, while the more advanced theory of L∞ -morphisms of differ-
ential graded Lie algebras is discussed in Section 13.4. Combined, these sections
provide the material which allows the reader to understand and appreciate Kontse-
vich’s formality theorem, which is stated, together with some of its consequences,
in the final Section 13.5.
Unless stated otherwise, F is an arbitrary field of characteristic zero.
C. Laurent-Gengoux et al., Poisson Structures, 353
Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4 13, © Springer-Verlag Berlin Heidelberg 2013
354 13 Deformation Quantization

13.1 Deformations of Associative Products

In this section, we introduce the notion of a formal deformation and of a k-th order
deformation of a commutative associative algebra and we show how these defor-
mations are related to Hochschild cohomology. We pick an indeterminate ν and we
denote by Fν the ring F[[ν ]] of formal power series in ν with coefficients in F.

13.1.1 Formal and k-th Order Deformations

Let A be a commutative associative algebra over F with unit. The product on A will
be denoted in its functional form by μ , or by a simple juxtaposition of the factors,
so we write FG for μ (F, G) for all F, G ∈ A . Associated to A , we consider the Fν -
algebra A [[ν ]] of formal power series in ν , with coefficients in A ; we also denote
this algebra by A ν . Thus, an element of A ν is a formal power series ∑i∈N Fi ν i ,
where all Fi ∈ A . The product which A ν inherits from A by “extension of scalars”,
as in Section 2.4.1, is given by
   
∑ Fi ν i · ∑ G jν j := ∑ Fi G j ν i+ j , (13.1)
i∈N j∈N i, j∈N

which is well-defined, since for every k ∈ N, the coefficient of ν k in the right-hand

side of (13.1) is a finite sum. This product is associative, Fν -bilinear and reduces to
the original product μ on A when setting ν = 0. It is therefore a formal deformation
of μ in the sense of the following definition.
Definition 13.1. Let (A , μ ) be a commutative associative algebra. An Fν -bilinear
map μ : A ν × A ν → A ν is called a formal deformation of μ if μ is associative
and μ = μ (mod ν ), i.e., if μ (F, G) − μ (F, G) ∈ ν A ν for all F, G ∈ A .
The question which we will address later in this chapter is the construction of all
formal deformations of a commutative associative product μ . The above deforma-
tion (13.1), which is commutative, is a particular example, in fact a rather trivial
one. We stress that the formal deformations which we consider, are not required to
be commutative.
The notion of equivalence of deformations of a given commutative associative
product μ is given in the following definition.
Definition 13.2. Let (A , μ ) be a commutative associative algebra and let μ and μ
be two formal deformations of μ . Then μ and μ are said to be equivalent if there
exists an Fν -linear map Φ : A ν → A ν , such that
(1) Φ (F) = F (mod ν ), for all F ∈ A ;
(2) Φ (μ (F, G)) = μ (Φ (F), Φ (G)), for all F, G ∈ A .
The map Φ is then called an equivalence of formal deformations.
13.1 Deformations of Associative Products 355

Condition (1) in the above definition implies that the map Φ is invertible, so the
notion of equivalence of formal deformations defines an equivalence relation on the
set of all formal deformations of a product μ . Also, if μ is a formal deformation
of μ and Φ : A ν → A ν is an Fν -linear map such that Φ (F) = F (mod ν ), for
every F ∈ A , then the map μ : A ν × A ν → A ν , defined for F, G ∈ A ν by
 
μ (F, G) := Φ μ (Φ −1 (F), Φ −1 (G)) ,

is a formal deformation of μ , equivalent to μ .

Remark 13.3. A formal deformation μ of μ can be seen as an element of Hom(A ×

A , A )[[ν ]], because there is for every k ∈ N∗ a natural isomorphism of Fν -modules
 
HomF (A k , A )[[ν ]]  HomFν (A ν )k , A ν .

We show this in the case of k = 1, the case of arbitrary k being only notationally
more complicated. Suppose that Φ : A ν → A ν is an Fν -linear map. For F ∈ A ,
we can write Φ (F) in a unique way as

Φ (F) = Φ0 (F) + Φ1 (F)ν + · · · + Φk (F)ν k + · · · , (13.2)

where every Φi (F) belongs to A . It follows that we can associate to every ele-
ment Φ of HomFν (A ν , A ν ) a sequence of linear maps (Φi )i∈N from A to A , that
is, an element of HomF (A , A )[[ν ]]. Conversely, given such a sequence of F-linear
maps (Φi )i∈N , the map Φ , defined by (13.2), extends in a unique way to an Fν -linear
map. By a slight abuse of notation, we keep the notation Φi for the Fν -linear exten-
sions of the maps Φi to A ν , which has the effect that (13.2) remains valid when F
belongs to A ν .

Formal deformations are often constructed term by term, which is done rigor-
ously by considering k-th order deformations. For k ∈ N, we consider the ring
Fνk := Fν /ν k+1 . Elements of this ring can be represented by polynomials in ν
of degree at most k and the product in this ring is the ordinary multiplication of
polynomials, followed by stripping off the terms whose degree is higher than k.
Starting from a commutative associative algebra A , we consider similarly

Akν := A ν /ν k+1  ,

which we can view as an Fν -algebra, or as an Fνk -algebra. We have that A0ν  A ,

in a natural way; these algebras will be identified in the sequel. Notice also that in
Remark 13.3, one can replace Fν by Fνk and A ν by Akν .
Definition 13.4. Let (A , μ ) be a commutative associative algebra. For k ∈ N, an
Fνk -bilinear map μ(k) : Akν × Akν → Akν ,

μ(k) = μ0 + μ1 ν + · · · + μk ν k , (13.3)
356 13 Deformation Quantization

where each μi is an F-bilinear map from A to itself, is called a k-th order deforma-
tion of μ if μ0 = μ and μ(k) is associative.
Under the natural quotient maps A ν → Akν and Ak+1 ν → A ν , formal deformations
k
(respectively (k + 1)-th order deformations) are mapped to k-th order deformations.
By a slight abuse of language, we say that an Fν -bilinear map μ : A ν × A ν → A ν
such that μ = μ (mod ν ) defines a k-th order deformation of μ if the induced
Fνk -bilinear map μ(k) : Akν × Akν → Akν is a k-th order deformation of μ . Clearly,
this is equivalent to saying that

μ (μ (F, G), H) = μ (F, μ (G, H)) (mod ν k+1 )

for all F, G, H ∈ A .
The notion of equivalence of k-th order deformations of μ is obtained by replac-
ing Fν by Fνk and A ν by Akν in Definition 13.2, which deals with the notion of
equivalence of formal deformations. Again, an Fν -linear map A ν → A ν which in-
duces an equivalence of k-th order deformations will be said to define an equivalence
of k-th order deformations.
Let (A , μ ) be a commutative associative algebra and suppose that μ(1) = μ0 +
μ1 ν defines a first order deformation of μ = μ0 . For a, b, c ∈ A , the coefficient
in ν in the associativity equation μ(1) (μ(1) (a, b), c) = μ(1) (a, μ(1) (b, c)) (mod ν 2 )
is given by
μ1 (ab, c) + μ1 (a, b)c = aμ1 (b, c) + μ1 (a, bc) .
It follows that

μ1− (ab, c) + μ1− (a, b)c − aμ1− (b, c) − μ1− (a, bc) = 0 ,

where μ1− is the skew-symmetric part of μ1 , i.e.,

1
μ1− (a, b) := (μ1 (a, b) − μ1 (b, a)) , (13.4)
2
for all a, b ∈ A . Performing a cyclic permutation of a, b, c in this equation leads to
the following three equations.

μ1− (ab, c) + μ1− (a, b)c − μ1− (b, c)a − μ1− (a, bc) = 0 ,
μ1− (bc, a) + μ1− (b, c)a − μ1− (c, a)b − μ1− (b, ac) = 0 ,
μ1− (ac, b) + μ1− (c, a)b − μ1− (a, b)c − μ1− (c, ab) = 0 ,

whose alternating sum simplifies to

μ1− (ab, c) = μ1− (a, c)b + μ1− (b, c)a .

It follows that if μ(1) is a first order deformation of μ = μ0 , then μ1− is a skew-

symmetric biderivation of (A , μ ). Suppose now that μ(2) = μ0 + μ1 ν + μ2 ν 2 is a
second order deformation of μ = μ0 . The associativity of μ(2) (up to ν 2 ) implies
13.1 Deformations of Associative Products 357

that the commutator

[a, b] := μ (a, b) − μ (b, a)

= 2ν μ1− (a, b) (mod ν 2 ) .

satisfies the Jacobi identity up to order ν 2 ,

[[a, b] , c] +  (a, b, c) = 0 (mod ν 3 ) ,

for all a, b and c in A . It follows that

μ1− (μ1− (a, b), c)+  (a, b, c) = 0 ,

for all a, b and c in A , i.e., μ1− satisfies the Jacobi identity. In conclusion, if μ(2) =
μ0 + μ1 ν + μ2 ν 2 is a second order deformation of an associative algebra (A , μ ),
then (A , μ , μ1− ) is a Poisson algebra!

13.1.2 Hochschild Cohomology

Deformations of associative products are governed by Hochschild cohomology. In

the present section, we introduce this cohomology and we show in the next section
how it is related to deformation theory. Throughout this section, we fix a commuta-
tive ring R.
Let V be a module over R. For k ∈ N∗ , we denote by HCk (V ) := Hom(V k ,V )
the space of all k-linear maps V k → V , while HC0 (V ) := V . We will define
a bracket [· , ·]G , called the Gerstenhaber bracket, on

HC• (V ) := HCk (V ) ,
k∈N

which is a graded Lie bracket, just like the Schouten bracket, after a shift of degree
(see Section 3.3.2). Thus, we denote k := k −1, and we define HCk−1 (V ) := HCk (V )
for k ∈ N so that HCk (V ) = HCk (V ). Elements of the latter space have degree k and
shifted degree k. With respect to the shifted degree, the Gerstenhaber bracket is a
graded Lie bracket, i.e., for all k,  ∈ N∗ ,

[· , ·]G : HCk (V ) × HC (V ) → HCk+ (V )

and [· , ·]G is graded skew-symmetric and satisfies the graded Jacobi identity (see
Table A.2 in Appendix A).
Explicitly, the bracket [· , ·]G is defined, for φ ∈ HCk (V ) and ψ ∈ HC (V ), by the
following formula,
358 13 Deformation Quantization

[φ , ψ ]G (x1 , . . . , xk++1 ) (13.5)

k    
:= ∑ (−1)i  φ x1 , . . . , xi−1 , ψ xi , . . . , xi+ , xi++1 , . . . , xk+
i=1
    
− (−1)k  ∑ (−1)i k ψ x1 , . . . , xi−1 , φ xi , . . . , xi+k , xi+k+1 , . . . , x+k ,
i=1

where x1 , . . . , xk++1 ∈ V . In particular, if μ ∈ HC2 (V ), then

[μ , μ ]G (x1 , x2 , x3 ) = 2μ (μ (x1 , x2 ), x3 ) − 2μ (x1 , μ (x2 , x3 )) ,

for all x1 , x2 , x3 ∈ V , so that μ defines an associative product on V if and only if

[μ , μ ]G = 0.
Suppose now that (A , μ ) is an associative (not necessarily commutative) alge-
bra with unit, over R. Combined with the graded Jacobi identity, the associativity
condition [μ , μ ]G = 0 implies that

δμk := − [·, μ ]G : HCk (A ) → HCk+1 (A ) (13.6)

is a coboundary operator, i.e., δμk+1 ◦ δμk = 0 for every k ∈ N. This operator is called
the Hochschild coboundary operator and the associated complex
δμk−1 δμk δμk+1
··· HCk−1 (A ) HCk (A ) HCk+1 (A ) ···

is called the Hochschild complex of A . The cohomology of this complex,


HH•μ (A ) := HHkμ (A )
k∈N

is called the Hochschild cohomology of A , where

HHkμ (A ) := Ker δμk / Im δμk−1 ,

for k ∈ N∗ , while HH0μ (A ) := Ker δμ0 . The elements of Ker δμk are called Hochschild
k-cocycles and the elements of Im δμk−1 are called Hochschild k-coboundaries.
An explicit formula for the Hochschild coboundary operator δμ is obtained by
substituting μ for ψ in (13.5); writing FG for μ (F, G), it takes, for φ ∈ HCk (A ),
and F0 , . . . , Fk ∈ A , the simple form

δμk (φ )(F0 , . . . , Fk ) := F0 φ (F1 , . . . , Fk ) + (−1)k+1 φ (F0 , . . . , Fk−1 )Fk

k (13.7)
+ ∑ (−1)i φ (F0 , . . . , Fi−1 Fi , . . . , Fk ) .
i=1

Since
δμ0 (F)(G) = GF − FG ,
13.1 Deformations of Associative Products 359

for F, G ∈ A , one has that HH0μ (A ) = Ker δμ0 is the center of the associative alge-
bra (A , μ ). Since moreover

δμ1 (φ )(F, G) = F φ (G) − φ (FG) + φ (F)G , (13.8)

for F, G ∈ A and φ ∈ HC1 (A ), the kernel of δμ1 consists of linear maps φ : A → A

which satisfy
φ (FG) = F φ (G) + φ (F)G ,
for all F, G ∈ A , which is the non-commutative analog of the derivation property. It
follows that the first Hochschild cohomology space associated to (A , μ ) consists of
the quotient of the space of all the derivations of (A , μ ), by the space of all inner
derivations of (A , μ ) (i.e., derivations of (A , μ ) of the form [F, ·], for F ∈ A ,
where [F, G] := FG − GF, for G ∈ A ). In particular, if (A , μ ) is commutative,
then HH1μ (A ) is the R-module X1 (A ) of all derivations of A .

13.1.3 Deformations and Cohomology

We now explain how the Hochschild cohomology of a commutative associative al-

gebra (A , μ ) is related to the deformations of its product μ . As we already pointed
out, a formal deformation (respectively, an -th order deformation) defines for ev-
ery k ∈ N (respectively for every k  ) a k-th order deformation. Conversely, let
μ1 , . . . , μk ∈ HC2 (A ) and suppose that μ(k) := μ0 + μ1 ν + · · · + μk ν k defines a
k-th order deformation of μ = μ0 , so that μ(k) , μ(k) G = 0 (mod ν k+1 ), where1
[· , ·]G denotes the Fν -linear extension of [· , ·]G to A ν . Let μk+1 be a bilinear map
A × A → A . The bilinear map μ(k+1) , defined by:

μ(k+1) := μ(k) + μk+1 ν k+1

defines a (k + 1)-th order deformation of μ , if and only if μ(k+1) , μ(k+1) G = 0

(mod ν k+2 ). Since μ(k) , μ(k) G = 0 (in Akν ), all that remains in the latter equation is

∑ [μi , μ j ]G = 0 ,
i+ j=k+1
i, j0

which we can write in terms of the Hochschild coboundary operator (13.6), associ-
ated to μ = μ0 , as
1
δμ2 (μk+1 ) = −[μk+1 , μ0 ]S =
2 ∑ [ μi , μ j ] G . (13.9)
i+ j=k+1
i, j1

1 In other words, [· , ·]G denotes the Gerstenhaber bracket on the Fν -module HC• (A ν ).
360 13 Deformation Quantization

This means that μ(k) can be extended to a deformation of μ of order k +1, if and only
if the right-hand side of (13.9) is a 3-coboundary for the Hochschild coboundary
operator δμ2 . Notice that it is a 3-cocycle, since
⎛ ⎞
⎜ ⎟
δμ3 ⎜
⎝ ∑ [μi , μ j ]G ⎟
⎠= ∑ μ , [μi , μ j ]G G
i+ j=k+1 i+ j=k+1
i, j1 i, j1
 
=− ∑ μi , [μ j , μ ]G G
+ [μ j , [μ , μi ]G ]G
i+ j=k+1
i, j1

=2 ∑ μi , δμ2 (μ j ) G
i+ j=k+1
i, j1

= ∑ μi , [μ j , μ ]G G
i+ j+=k+1
i, j,1
= 0,

where we have used the graded Jacobi identity for [· , ·]G to obtain the second and the
last lines. It follows that the obstruction to extending a deformation of μ of some
order to the next order lies in the third Hochschild cohomology space HH3μ (A ).
This statement is the content of the following proposition.
Proposition 13.5. Let (A , μ ) be a commutative associative algebra and suppose
that μ(k) = ∑ki=0 μi ν i defines a k-th order deformation of μ . Then μ(k) can be ex-
tended to a (k + 1)-th order deformation of μ if and only if the Hochschild 3-cocycle
∑ki=1 [μi , μk+1−i ]G is a Hochschild 3-coboundary.
It is clear from (13.9) that the term μk+1 ν k+1 which makes a k-th order deformation
of μ into a (k + 1)-th order deformation of μ is not unique (if it exists): one can
always add a Hochschild 2-cocycle to μk+1 and this is the only freedom in the choice
of μk+1 . If this 2-cocycle is a coboundary, then the two extended (k + 1)-th order
deformations are essentially the same, in the sense that they are equivalent (see
Definition 13.2). More generally, we have the following proposition.
Proposition 13.6. Let (A , μ ) be a commutative associative algebra and suppose
that μ(k) = ∑ki=0 μi ν i defines a k-th order deformation of μ , where k ∈ N∗ . Let   k,
where  ∈ N∗ . Let φ : A → A be an F-linear map, which we extend to an Fν -linear
map A ν → A ν , still denoted by φ , and let

μ() = μ0 + μ1 ν + μ2 ν 2 + · · · + μ−1 ν −1 + (μ + δμ1 (φ )) ν  . (13.10)
 extends to a k-th order deformation μ  of μ , which is equivalent to μ .
Then, μ() (k) (k)

Proof. Consider the Fν -linear map Φ : A ν → A ν , defined by

13.1 Deformations of Associative Products 361

Φ : Aν → Aν
F → F − φ (F) ν 

whose inverse is given by Φ −1 (F) = ∑i∈N φ i (F) ν i . The Fν -bilinear map μ(k)
 , de-

fined for all F, G ∈ A by


μ(k) (F, G) := Φ (μ(k) (Φ −1 (F), Φ −1 (G)))

is a k-th order deformation of μ , which is equivalent to μ(k) . Up to order , it is

given by

 (F, G) = Φ ( μ (F + φ (F) ν  , G + φ (G) ν  ))

μ() ()

= Φ (μ() (F, G) + φ (F)G ν  + F φ (G) ν  )

= μ() (F, G) + (φ (F)G − φ (FG) + F φ (G)) ν 
= μ() (F, G) + δμ1 (φ )(F, G)ν  ,

 = μ + δ 1 (φ ) ν  , as
where we have used (13.8) in the last step. It follows that μ() () μ
in (13.10).

It follows from the proposition that in order to construct all possible deforma-
tions of μ , up to equivalence, one only has to consider, at every step k, as many
possibilities as there are elements in HH2μ (A ).

13.1.4 The Maurer–Cartan Equation

Let μ = ∑∞ ν
i=0 μi ν be an element of HC (A ), where μ0 = μ . As above, μ is
i 2

associative if and only if [μ , μ ]G = 0, which is in turn equivalent to demanding

that μ − μ ∈ ν HC2 (A ν ) is a solution of the following equation

1
δμ2 (x) − [x, x]G = 0 , (13.11)
2
where, by a slight abuse of notation, δμ2 stands for the Fν -linear extension of δμ2
to HC2 (A ν )  HC2 (A )[[ν ]]. For future use, we stress that we consider only so-
lutions x of this equation which are formal series without constant term. Precisely,
x ∈ ν HC2 (A ν ). In terms of the terminology of Section 13.3, equation (13.11) is
called the Maurer–Cartan equation associated to the differential graded Lie algebra
( HC• (A ), [· , ·]G , δμ ).
362 13 Deformation Quantization

13.1.5 Star Products

As a particular case, we consider the algebra A = C∞ (M) of smooth functions on

a real manifold M, where the associative product μ is the usual product of func-
tions. In this case, one usually considers a particular class of formal deformations
of μ , where all Hochschild cochains are bi-differential operators, a notion which we
define first.
A bi-differential operator on M is a map Φ : C∞ (M) × C∞ (M) → C∞ (M), com-
patible with restrictions to open subsets and such that, on every coordinate chart
(U, (x1 , . . . , xd )) and for all smooth functions F, G on U, one can write:
 r  s 
∂ F ∂ G
Φ (F, G) = ∑ φr,s , (13.12)
r,s∈Nd
∂ x r ∂ xs

where for r, s ∈ Nd the coefficients φr,s are smooth functions on U, which van-
ish except for a finite number of (r, s). When all coefficients φr,s are constant,
we speak of a constant coefficient bi-differential operator. For each multi-index
r = (r1 , . . . , rd ) ∈ Nd , we used in (13.12) the notation

∂ rF ∂ r1 +···+rd F
:= r .
∂ xr ∂ x1r1 · · · ∂ xdd

The generalization of the definition of a bi-differential operator to a multi-differential

operator is clear. Notice that μ is a bi-differential operator on M.
Definition 13.7. Let M be a real manifold.
(1) A star product on C∞ (M) is a formal deformation μ = ∑i∈N μi ν i of the prod-
uct μ of C∞ (M), where each μi is a bi-differential operator on M;
(2) Two star products μ and μ on C∞ (M) are said to be equivalent star products
if they are equivalent as formal deformations, via an equivalence of the form
Φ = 1C∞ (M) + φ1 ν + · · · + φk ν k + · · · , where each one of the F-linear maps
φi : C∞ (M) → C∞ (M) is a differential operator;
(3) If π is a Poisson structure on M, then (M, π ) is said to admit a deformation
quantization if there exists a star product μ = ∑i∈N μi ν i on C∞ (M), such that
μ1 = π2 . Then μ is said to be a star product for (M, π ).
We also use the infix notation “” for a star product. Thus, we often write F  G
instead of μ (F, G), for F, G ∈ C∞ (M).
Several authors demand that the constant power series “1” is a unit for the star
product; this condition is not very essential, since it can be shown that every star
product as defined above is equivalent to a star product which has “1” as its unit
(see [39, Cor. 3.4.5]).
The description of star products in terms of cohomology is the same as for de-
formations, except that one considers as cochains only the differential Hochschild
13.1 Deformations of Associative Products 363

cochains of C∞ (M), i.e., the Hochschild cochains which are given by multi-differen-
tial operators. The space of all differential Hochschild cochains of C∞ (M) is denoted

by HC•diff (M) := k∈N HCkdiff (M). Notice that the Gerstenhaber bracket can be re-
stricted to such cochains because the Gerstenhaber bracket of two multi-differential
operators is a multi-differential operator.

13.1.6 A Star Product for Constant Poisson Structures

In this section we consider the case of a constant Poisson structure π on a finite-

dimensional vector space V . We describe the Moyal–Weyl product, which is an
explicit star product for (V, π ).
In order to describe the Moyal–Weyl product, we first show how a constant
Poisson structure leads naturally to an endomorphism of F (V ) ⊗ F (V ), where
F (V ) := C∞ (V ). Given two endomorphisms V1 and V2 of F (V ), an endomorphism
V1 ⊗ V2 of F (V ) ⊗ F (V ) is defined by letting

(V1 ⊗ V2 )(F ⊗ G) := V1 (F) ⊗ V2 (G) ,

for all F, G ∈ F (V ). Given endomorphisms V1 , . . . , Vk , V1 , . . . , Vk of F (V ) leads by

linear extension to an endomorphism ∑ki=1 Vi ⊗ Vi  of F (V ) ⊗ F (V ). In particular,
given a constant Poisson structure on V , which we write as π = ∑ki=1 Vi ∧ Vi  , with
V1 , . . . , Vk , V1 , . . . , Vk constant vector fields on V , the previous construction yields
an endomorphism of F (V ) ⊗ F (V ), which we also denote π . We can then write

π [F, G] = μ (π (F ⊗ G)) , (13.13)

where μ denotes the linear map μ : F (V ) ⊗ F (V ) → F (V ), which simply sends

F ⊗ G to the product FG, for functions F, G ∈ F (V ). The interest of writing the
Poisson structure, and the other differential operators which will come up in the con-
struction, as in the right-hand side of (13.13) is that although π cannot be reapplied
to π [F, G], it can be reapplied to π (F ⊗ G). It allows us to define, for F, G ∈ F (V )
(or, by ν -linearity, for F, G ∈ F (V )ν ) the following formal power series:

νi
eνπ /2 (F ⊗ G) := F ⊗ G + ∑∗ 2i i! π i (F ⊗ G) . (13.14)
i∈N

Proposition 13.8. Let π be a constant Poisson structure on a finite-dimensional vec-

tor space V and let μ denote the linear map μ : F (V )ν ⊗F (V )ν → F (V )ν , defined
by μ (F ⊗ G) := FG, for all F, G ∈ F (V )ν . The bilinear map F (V )ν × F (V )ν →
F (V )ν , defined by  
F  G := μ eνπ /2 (F ⊗ G) , (13.15)

is a star product for (V, π ).

364 13 Deformation Quantization

Proof. We first prove that  is associative: for all F, G, H ∈ F (V ), we show that

F  (G  H) = (F  G)  H. This can be done by direct computation, upon using
the classical trick of representing a bi-differential operator as a composition of two
differential operators: if Φ is a bi-differential operator, which we consider here with
constant coefficients, then there exists a differential operator Φ̂xy , such that

Φ (F, G) = Φ̂xy (F(x)G(y))|y=x .

Explicitly, when Φ is of the form (13.12), with all φr,s constant, then Φ̂xy is given by

∂ r+s
Φ̂xy = ∑ φr,s
∂ x r ∂ ys
.
r,s∈Nd

With this notation, we can write (13.15) as

F  G = eν π̂xy /2 F(x)G(y)|y=x .

The associativity of  then follows from the following computation:

(F  G)  H = eν π̂xz /2 (F  G)(x)H(z)|z=x
= eν π̂xz /2+ν π̂yz /2 eν π̂xy /2 F(x)G(y)H(z)|z=y=x
= eν (π̂xz +π̂yz +π̂xy )/2 F(x)G(y)H(z)|z=y=x
= F  (G  H) .

The computation demands some extra justification. We have used in the third equal-
ity that if φ1 and φ2 are commuting endomorphisms, then eφ1 +φ2 = eφ1 eφ2 ; this prop-
erty follows at once from the definition given in (13.14). We used in the second
equality that if V = Vx is a derivation of F (V ), then
 
V F(x, y)|y=x = Vx F(x, y)|y=x + Vy F(x, y)|y=x = (Vx + Vy )F(x, y)|y=x ,

which follows from the chain rule. This proves that  is associative, hence is a formal
deformation of μ ; since for F, G ∈ F (V ) the coefficient of ν in F  G is 12 π [F, G], it
follows that  is a star product for (V, π ).
When the rank 2r of the constant Poisson structure π is equal to the dimension of V ,
there exists a system of linear coordinates (q1 , . . . , qr , p1 , . . . , pr ) on V , in which the
Poisson bracket of functions F, G ∈ F (V ) takes the canonical form
r  
∂F ∂G ∂F ∂G
{F, G} = ∑ − ;
i=1 ∂ qi ∂ pi ∂ pi ∂ qi

see Proposition 6.5 for the existence of such coordinates. For such a choice of Pois-
son structure and such coordinates, the explicit formula (13.14) is referred to as the
Moyal–Weyl product, or simply the Moyal product.
13.1 Deformations of Associative Products 365

13.1.7 A Star Product for Symplectic Manifolds

In this section we explain how the Moyal–Weyl product can be used to prove that
every symplectic manifold (M, ω ) admits a star product; more precisely, the alge-
bra of smooth functions on (M, ω ), equipped with its canonical Poisson structure,
admits a star product. The construction presented here is due to Fedosov; we refer
to his original paper [73] for proofs and details.
We first need to elaborate on the star product in the case of a symplectic vector
space (the Moyal–Weyl product, explained in Section 13.1.6 above). Let (V, ω ) be
a finite-dimensional symplectic real vector space and let π = {· , ·} denote the (con-
stant) Poisson structure on V which corresponds to ω (see Section 6.3.1). We denote
by Fˆ (V ) the space of all formal power series in the variables that constitute a basis
for V ∗ and we denote by Fˆ  j (V ) the subspace of Fˆ (V ), consisting of those formal
power series whose valuation is at least j ∈ N. We also consider, for k ∈ N,

E k (V ) := ∑ ν i Fˆ  j (V ) .
i, j∈N
2i+ jk

It is easy to see from the explicit formula (13.15) that the Moyal–Weyl product
on C∞ (V )ν extends to Fˆ (V )ν := Fˆ (V )[[ν ]]; moreover, for all k,  ∈ N, we have
that
E k (V )  E  (V ) ⊂ E k+ (V ) , (13.16)
 k 
which follows immediately from the fact that Fˆ (V ), Fˆ  (V ) ⊂ Fˆ k+−2 (V ),
for all k,  ∈ N∗ . To V we associate the associative algebra A(V ), defined by

A(V ) := E 0 (V ) ⊗ ∧•V ∗ ,

equipped with the associative product  given by:

(F ⊗ ξ )  (G ⊗ η ) := (F  G) ⊗ (ξ ∧ η ) ,

for all F, G ∈ E 0 (V ) and ξ , η ∈ ∧•V ∗ . Note that there is a decreasing sequence of

vector spaces
Ak (V ) := E k (V ) ⊗ ∧•V ∗
with A0 (V ) = A(V ). Note also that A(V )/Ak (V ) is a finite-dimensional vector
space, for each k ∈ N, and that there is on A(V ) a natural linear form P0 : A(V ) → R
defined by mapping F ⊗ ξ to 0 if ξ ∈ ⊕i1 ∧i V ∗ and by mapping F ⊗ 1 to F(0)
for all F ∈ E 0 (V ), where 1 stands for the unit element of the algebra ∧•V ∗ , and
where F(0) stands for the constant term of the formal power series F.
We now get back to the case of a symplectic manifold (M, ω ). The canonical
Poisson structure associated to ω is denoted by π and the algebra of smooth func-
tions on M is denoted by F (M). Each tangent space Tm M to M is a symplectic
vector space, associated to the Poisson structure πm , so that we can consider as
above a star product m on Fˆ (Tm M)ν and an associative product m on A(Tm M).
366 13 Deformation Quantization

We say that a section of the canonical projection ∪m∈M A(Tm M) → M mapping each
element to its base point is smooth if its projection on the vector bundle
 A(Tm M)
k (T M)
→M
m∈M
A m

is smooth, for every k ∈ N. For k ∈ N, the space of all smooth sections which take
values in Ak (Tm M), for every m ∈ M, is denoted by A k (M); we use the shorthand
A (M) for A 0 (M). The product of two sections a ∈ A k (M) and b ∈ A  (M) is
the section of A k+ (M) whose value at m ∈ M is a(m) m b(m). Combined, the
linear maps P0 : A(Tm M) → R defined above lead to a linear map A (M) → F (M)ν ,
which we also denote by P0 .

Theorem 13.9. Let (M, ω ) be a symplectic manifold and let (A (M), ) be the al-
gebra constructed as above. There exists a derivation D : A (M) → A (M) with the
following property: for every F ∈ F (M)ν , there exists a unique aF ∈ A (M) such
that D(aF ) = 0 and P0 (aF ) = F.

Without going into details, we mention briefly how D is constructed, thereby

sketching a proof of the theorem; see [73] for details. D is an operator of the follow-
ing form:
D := d∇ + [ω̂ + γ , ·] . (13.17)
The first term, d∇ , depends on the choice of a symplectic connection, i.e. a torsion-
free connection ∇ on T M which satisfies

ıZ dω (V , W ) = ω (∇Z V , W ) + ω (V , ∇Z W ) ,

for arbitrary vector fields V ,W , Z on M. Such a symplectic connection induces

a connection on the bundle m∈M Fˆ (Tm M) → M, which is strictly speaking not
a vector bundle (because the vector spaces Fˆ (Tm M) are infinite-dimensional), yet
it leads to a covariant derivative, which is d∇ : A (M) → A (M). The bracket [· , ·]
in (13.17) is the Lie bracket on A (M) induced from the bracket [· , ·]m on the spaces
A(Tm M), with m ∈ M, where [· , ·]m is defined by

[a ⊗ ξ , b ⊗ η ]m := (a  b − b  a) ⊗ ξ ∧ η ,

for all a, b ∈ Fˆ (Tm M) and ξ , η ∈ ∧• Tm∗ M. Third, ω̂ is the symplectic struc-

ture ω , viewed as an element of A (M). Finally, γ ∈ A 3 (M) is a section of
 ν ∗
m∈M F (Tm M) ⊗ Tm M, which is chosen such that D = 0. Using the latter prop-
ˆ 2

erty, one proves the existence and uniqueness of aF , as stated in Theorem 13.9.
It is clear that D is a derivation, because both the adjoint action and the covariant
derivatives are derivations.

Corollary 13.10. Let (M, ω ) be a symplectic manifold, let (A (M), ) be the al-
gebra constructed above and let D be a derivation of A (M) as in Theorem 13.9.
Denote by  the bilinear map on F (M)ν defined for all F, G ∈ F (M)ν by
13.2 Deformations of Poisson Structures 367

F  G := P0 (aF  aG ) , (13.18)

where aF is the unique element in the kernel of D, for which P0 (aF ) = F, and simi-
larly for aG . Then  is a star product for (M, π ), where π is the canonical Poisson
structure on M, associated to ω .
Proof. We first prove that the product, defined by (13.18), is associative. Let
B(M) ⊂ A (M) be the kernel of D; it is a subalgebra of A (M) since D is a
derivation. By Theorem 13.9, the assignment F → aF is a bijective linear map
from F (M)ν to B(M) with inverse map the restriction of P0 to B(M). It fol-
lows that the product  on F (M)ν is obtained from the associative product 
on B(M), via a bijection, hence is associative. We prove that  is a star product
for (M, π ). Let F and G be smooth functions on M (in particular, they are inde-
pendent of ν ). A closer look at the construction of aF and aG in [73] implies that
aF = (F + dm F + Hessm F) ⊗ 1 + bF and aG = (G + dm G + Hessm G) ⊗ 1 + bG for
some bF , bG ∈ A 3 (M), where dm F and dm G are viewed as linear functions on
Tm M, and Hessm F and Hessm G are viewed as quadratic functions on Tm M. An
explicit computation then yields P0 (aF  aG ) = FG + ν2 {F, G} (mod ν 2 ), which
completes the proof.

13.2 Deformations of Poisson Structures

The deformation theory of commutative associative algebras, described in Sec-

tion 13.1, can quite easily be transcribed for Lie algebras. For Poisson algebras,
a few extra adaptions are necessary, since Poisson algebras have not one but two
algebra structures; in this section, we will deform only one of them, the Poisson
bracket, but we need to keep in mind that the deformed Poisson bracket should be
compatible with the associative product. As we will see, the deformation theory of
Poisson algebras is, despite this constraint, formally very similar to the deformation
theory of commutative associative algebras.

13.2.1 Formal and k-th Order Deformations

Let (A , ·, π ) be a Poisson algebra. On A ν we consider, as in (13.1), the Fν -linear

extension of the product “·”, which makes A ν into a commutative associative alge-
bra (over the ring Fν ).
Definition 13.11. Let (A , ·, π ) be a Poisson algebra. An Fν -linear skew-symmetric
biderivation π of (A ν , ·) is called a formal deformation of π if π satisfies the
Jacobi identity and π = π (mod ν ).
When π is a formal deformation of a Poisson structure π , the triple (A ν , ·, π ) is a
Poisson algebra over the ring Fν . Strictly speaking, our Definition 1.1 of a Poisson
368 13 Deformation Quantization

algebra is valid only for algebras over a field, but it is clear that one can define
Poisson algebras over an arbitrary commutative ring (with unit) R, by replacing in
Definition 1.1 “F-vector space A ” by “R-module A ”. It is also clear that the Jacobi
identity for π is satisfied if and only if [π , π ]S = 0, where the Schouten bracket
[· , ·]S is still given by (3.36), but for Fν -multi-linear maps, rather than F-multi-linear
maps; this is the main fact which we will use about this more general context.
Definition 13.12. Let (A , ·, π ) be a Poisson algebra and let π and π be two formal
deformations of π . Then π and π are said to be equivalent deformations if there
exists an Fν -linear map Φ : A ν → A ν , satisfying for all F, G ∈ A the following
conditions:
(1) Φ (F) = F (mod ν );
(2) Φ (FG) = Φ (F)Φ (G);
(3) Φ (π [F, G]) = π [Φ (F), Φ (G)].
Condition (1) implies that Φ is invertible; then (2) means that Φ is an automorphism
of the associative algebra (A , ·), hence of (A ν , ·) and (3) means that Φ is a Lie
algebra isomorphism (A ν , π ) → (A ν , π ).
It is clear that, if π is a formal deformation of π and Φ : A ν → A ν is an
ν
F -linear map satisfying items (1) and (2) of Definition 13.12, then the map π :
A ν × A ν → A ν , defined for F, G ∈ A ν by
 
π [F, G] := Φ π [Φ −1 (F), Φ −1 (G)] , (13.19)

is a formal deformation of π , equivalent to π .

As in the associative case, we define the notion of a k-th order deformation of a
Poisson structure.
Definition 13.13. Let (A , ·, π ) be a Poisson algebra. For k ∈ N∗ , an Fνk -bilinear
skew-symmetric biderivation π(k) of Akν ,

π(k) = π0 + π1 ν + · · · + πk ν k , (13.20)

is called a k-th order deformation of π , if π0 = π and if π(k) satisfies the Jacobi

identity.
One defines the notion of equivalence of k-th order deformations by replacing Fν
by Fνk and A ν by Akν in Definition 13.12.

13.2.2 Deformations and Cohomology

As in the associative case, it is clear that a formal deformation can be truncated

to a k-th order deformation, for every k ∈ N. Similarly, an -th deformation can be
truncated to a k-th order deformation, for every k  . This suggests to approach also
in this case the construction and the study of all (formal) deformations of a Poisson
13.2 Deformations of Poisson Structures 369

bracket, by analyzing the extensibility of a k-th order deformation to a (k + 1)-th

order deformation.
Therefore, let (A , ·, π ) be a Poisson algebra and suppose that π(k) = π0 + π1 ν +
· · · + πk ν k defines a k-th order deformation of π , where π1 , . . . , πk are elements of
X2 (A ) and π0 = π . Given a skew-symmetric bilinear map πk+1 : A × A → A , the
skew-symmetric Fνk+1 -bilinear map π(k+1) , defined by π(k+1) := π(k) + πk+1 ν k+1 , is
a (k +1)-th order deformation of π if and only if π(k+1) , π(k+1) S = 0 (mod ν k+2 ),
i.e.,
1
δπ2 (πk+1 ) = −[πk+1 , π0 ]S = ∑ [πi , π j ]S ,
2 i+ j=k+1
(13.21)
i, j1

where we have used the Poisson coboundary operator (4.5), associated to the Pois-
son bracket π = π0 . As in the associative case (see Section 13.1.3), one checks by
using the graded Jacobi identity for [· , ·]S , that the right-hand side in (13.21) is a
Poisson 3-cocycle, so that the obstruction for extending a deformation of the Pois-
son bracket π , of some order to the next order lies in the third Poisson cohomology
space Hπ3 (A ). This shows the following proposition.
Proposition 13.14. Let (A , ·, π ) be a Poisson algebra and suppose that π(k) =
∑ki=0 πi ν i is a k-th order deformation of π . Then π(k) can be extended to a (k + 1)-th
order deformation of π if and only if the Poisson 3-cocycle ∑ki=1 [πi , πk+1−i ]S is a
Poisson 3-coboundary.
It is clear from (13.21) that the term πk+1 ν k+1 which makes a k-th order deforma-
tion of π into a (k + 1)-th order deformation of π is not unique (if it exists): one
can always add an arbitrary Poisson 2-cocycle (for π ) to πk+1 . If this cocycle is a
coboundary, then the two extended (k + 1)-th order deformations are essentially the
same, in the sense that they are equivalent (see Definition 13.12). A stronger state-
ment is given in the following proposition, which is the analog of Proposition 13.6.
Proposition 13.15. Let (A , ·, π ) be a Poisson algebra and suppose that π(k) =
∑ki=0 πi ν i is a k-th order deformation of π , where k ∈ N∗ and let   k, where  ∈ N∗ .
Let ϕ ∈ X1 (A ), which we extend to an Fν -linear map A ν → A ν , still denoted by ϕ ,
and let

π() = π0 + π1 ν + π2 ν 2 + · · · + π−1 ν −1 + (π + δπ1 (ϕ )) ν  . (13.22)
 extends to a k-th order deformation π  of π , which is equivalent to π .
Then, π() (k) (k)

Proof. Let us consider the map

Φ : Aν → Aν
(−1)k k k (13.23)
F → e−ν ϕ (F) = ∑

ν ϕ (F) .
k∈N k!

It is clear that Φ is Fν -linear and that Φ (F) = F (mod ν ), for all F ∈ A . We show
that Φ (FG) = Φ (F)Φ (G) for all F, G ∈ A . To do this, we use that for all k ∈ N,
370 13 Deformation Quantization
 
k
ϕ (FG) = ∑
k
ϕ r (F) ϕ s (G) , (13.24)
r+s=k
s
r,s0

which is a consequence of the fact that ϕ is a derivation of A . Then,

(−1)k k k
Φ (FG) = e−ν ϕ (FG) = ∑

ν ϕ (FG)
k∈N k!
 
(−1)k k k r
= ∑ ∑ k! s
ν ϕ (F) ϕ s (G)
k∈N r+s=k
r,s0

(−1)r+s (r+s) r
= ∑∑ s!r!
ν ϕ (F) ϕ s (G)
r∈N s∈N
  
(−1)r r r (−1)s s s
= ∑ ν ϕ (F) ∑ ν ϕ (G)
r∈N r! s∈N s!
= Φ (F) Φ (G) .
 , which is defined for all F, G ∈ A ν by
As in (13.19), we have that π(k)


 
π(k) [F, G] := Φ π(k) Φ −1 (F), Φ −1 (G) ,

is a k-th order deformation of π which is equivalent to π(k) . Using the formula for
the inverse of Φ , namely

ν k k
Φ −1 (F) = eν ϕ (F) = ∑

ϕ (F) ,
k∈N k!

we can make the following computation in the algebra Aν (i.e., all equalities are
modulo ν +1 ):

 (F, G) = Φ π

π() () F + ϕ [F]ν , G + ϕ [G]ν
 
 
= Φ π() [F, G] + (π [ϕ [F], G] + π [F, ϕ [G]])ν 
= π() [F, G] + (π [ϕ [F], G] + π [F, ϕ [G]] − ϕ [π [F, G]])ν 
= π() [F, G] + δπ1 (ϕ )[F, G] ν  .

 = π + δ 1 (ϕ ) ν  , which is the content of (13.22).

It shows that π() () π

It follows from the proposition that in order to construct all possible deformations
of π , up to equivalence, one only has to consider, at every step k, as many possibili-
ties as there are elements in Hπ2 (A ). Notice that the associative analog of the above
proposition, which was given in Proposition 13.6, is easier to prove; the reason for
13.3 Differential Graded Lie Algebras 371

this is that in the Poisson case the map Φ which defines the equivalence needs to be
an automorphism of the associative algebra (Akν , ·).

13.2.3 The Maurer–Cartan Equation

Let (A , ·, π ) be a Poisson algebra and let π = ∑∞i=0 πi ν be an element of X (A ),

i 2 ν

where π0 = π . As above, π is a Poisson structure if and only if [π , π ]S = 0. Denot-

ing by δπ2 the Fν -linear extension of δπ2 to X2 (A ν )  X2 (A )[[ν ]], it follows that π
is a Poisson structure if and only if π − π is a solution of the following equation
1
δπ2 (x) − [x, x]S = 0 . (13.25)
2
In the terminology of Section 13.3, this equation is called the Maurer–Cartan equa-
tion associated to the differential graded Lie algebra (X• (A ), [· , ·]S , δπ ).

13.3 Differential Graded Lie Algebras

Differential graded Lie algebras are graded Lie algebras which are equipped with
a derivation of degree one whose square is zero. A prime example, which we have
already encountered several times in this book, is the algebra of skew-symmetric
multi-derivations of a Poisson algebra (or multivector fields on a Poisson manifold),
where the graded Lie bracket is the Schouten bracket and the differential is (up to
a sign) the Poisson coboundary operator. Since in the study of a differential graded
Lie algebra its symmetric algebra plays an important rôle we first give the basic
definitions and properties of the symmetric algebra of a graded vector space.

13.3.1 The Symmetric Algebra of a Graded Vector Space


Let V = i∈Z Vi be a graded vector space. We denote the degree of a homoge-
neous element x ∈ V by d 0 (x). The symmetric algebra of V is the associative al-
gebra S•V , which is obtained by dividing the tensor (graded associative) algebra

T •V = ∈N V ⊗ of V by the two-sided ideal generated by all elements of the form
0 0
x ⊗ y − (−1)d (x)d (y) y ⊗ x, with x and y homogeneous. The induced product is de-
noted by juxtaposition of elements. With this notation,
0 (x)d 0 (y)
xy = (−1)d yx, (13.26)

for all homogeneous elements x, y of V . Note that if x is a homogeneous element

of V of odd degree, then it follows from (13.26) that x2 = − x2 , so that x2 = 0. We
372 13 Deformation Quantization

call an element of S•V of the form x1 . . . x with all xi homogeneous, a homogeneous

monomial of S•V . Let i1 , . . . , i be integers for which there exists a non-zero homo-
geneous monomial X = x1 . . . x where x1 ∈ Vi1 , . . . , x ∈ Vi . Then we can define the
sign sgn(σ ; i1 , . . . , i ) ∈ {−1, 1}, associated to (i1 , . . . , i ) and σ ∈ S , by the equality

xσ (1) . . . xσ () = sgn(σ ; i1 , . . . , i ) x1 . . . x , (13.27)

since it only depends on the -tuple (i1 , . . . , i ) and not on the choice of x1 , . . . , x .
With X as above, we often write sgn(σ ; X) as a shorthand for sgn(σ ; i1 , . . . , i ). For
indices i1 , . . . , i such that Vi1 Vi2 . . .Vi = 0 we define sgn(σ ; i1 , . . . , i ) := 1 for all
σ ∈ S .
In what follows, the graded vector space of which we consider the symmetric
algebra will always be a graded Lie algebra, but in which we shift the grading by

one. Thus, if g = ∈N g is a graded Lie algebra, we consider on g two gradings:
for x ∈ g we define |x| :=  and d 0 (x) :=  − 1 = |x| − 1. Then g is a graded Lie
algebra with respect to the grading | · |, but when the grading d 0 is used, it is only
a graded vector space; when we use the latter grading, we will write g instead of g.
The symmetric algebra which plays an important rôle in what follows is the graded
commutative algebra S• g, where the grading on S• g is given, for a homogeneous
monomial x1 . . . x , by
 
d 0 (x1 . . . x ) := ∑ |xi | −  = ∑ d 0 (xi ) . (13.28)
i=1 i=1

Of course, S• g inherits also a grading from the tensor algebra, which assigns to a
monomial x1 . . . x the degree ; we will not use this grading, except that we denote,
for  ∈ N, by S g the span of all monomials of the form x1 . . . x , where x1 , . . . , x ∈ g.

13.3.2 Differential Graded Lie Algebras

In this section, we introduce the notion of a differential graded Lie algebra and of a
morphism of differential graded Lie algebras. The basic examples, given below, are
the Poisson complex, associated to a Poisson algebra, and the Hochschild complex,
associated to an associative algebra.
Definition 13.16. A differential graded Lie algebra is a triple (g, [· , ·], D), where
(g, [· , ·]) is a graded Lie algebra and D : g → g is a differential of g, i.e., D is a
graded linear map of degree one from g to g, D satisfies D ◦ D = 0 and D is a graded
derivation of degree one of (g, [· , ·]).

Expressed in terms of formulas, the fact that (g = k∈Z gk , [· , ·]) is a graded Lie
algebra means that [· , ·] satisfies, for all homogeneous elements x, y and z of g and
for all i, j ∈ Z, the following properties:
(1) [gi , g j ] ⊂ gi+ j ;
13.3 Differential Graded Lie Algebras 373

(2) [x, y] = −(−1)|x||y| [y, x];

(3) (−1)|x||z| [[x, y] , z] +  (x, y, z) = 0;
while the fact that D is a differential on g means that D satisfies, for all homogeneous
elements x and y of g and for all i ∈ Z, the following properties:
(4) D(gi ) ⊂ gi+1 ;
(5) D ([x, y]) = [D(x), y] + (−1)|x| [x, D(y)];
(6) D ◦ D = 0.
Let (g, [· , ·]) be a graded Lie algebra and let z be an element of degree one, z ∈ g1 .
The graded Jacobi identity for [· , ·], i.e., the above identity (3), implies that the
degree one map Dz , defined by

Dz (y) := [z, y]

for all y ∈ g, is a graded derivation of g of degree one. Moreover, if z satisfies

[z, z] = 0, the graded Jacobi identity implies that Dz ◦ Dz = 0, hence (g, [· , ·], Dz )
is a differential graded Lie algebra. We then call the triple (g, [· , ·], z) a pointed dif-
ferential graded Lie algebra. The differential of a pointed differential graded Lie
algebra (g, [· , ·], z) will, without further mention, always be denoted by Dz .
Example 13.17. Let (A , ·, π ) be a Poisson algebra and consider the graded Lie al-
gebra (X• (A ), [· , ·]S ), where we recall that X• (A ) denotes the graded vector space
of skew-symmetric multi-derivations of A , with shifted degree, and [· , ·]S denotes
the Schouten bracket on X• (A ) (see Chapter 3). Since the shifted degree of π is one
and since [π , π ]S = 0 because π is a Poisson bracket, (X• (A ), [· , ·]S , π ) is a pointed
differential graded Lie algebra.
Example 13.18. The previous example can also be considered in the smooth cate-
gory. Namely, let (M, π ) be a real Poisson manifold and recall that we denote by
(X• (M), [· , ·]S ) the graded Lie algebra of multivector fields on M, equipped with
the Schouten bracket [· , ·]S . Then (X• (M), [· , ·]S , π ) is a pointed differential graded
Lie algebra. Notice that this example can be considered for any real manifold M by
taking the trivial Poisson structure on M, i.e., for every real manifold M, the triple
(X• (M), [· , ·]S , 0) is a pointed differential graded Lie algebra.
Example 13.19. Let (A , μ ) be a commutative associative algebra and consider the
graded Lie algebra ( HC• (A ), [· , ·]G ), where we recall from Section 13.1.2 that
HCk (A ) = Hom(A k+1 , A ) is the space of all (k + 1)-linear maps from A to itself
and the graded Lie bracket [· , ·]G is the Gerstenhaber bracket (defined in (13.5)).
Since μ is associative, [μ , μ ]G = 0 and hence ( HC• (A ), [· , ·]G , μ ) is a pointed dif-
ferential graded Lie algebra.
Example 13.20. The previous example can also be considered when (A , μ ) is the
algebra of smooth functions on a real manifold M, equipped with the usual product
of functions μ . Consider the graded Lie algebra ( HC•diff (M), [· , ·]G ), which is the
subalgebra of ( HC• (C∞ (M)), [· , ·]G ) which consists of the differential Hochschild
cochains (see Section 13.1.5). Then ( HC•diff (M), [· , ·]G , μ ) is a pointed differential
graded Lie algebra.
374 13 Deformation Quantization

Example 13.21. Every differential graded Lie algebra (g, [· , ·], D) leads to a coho-
mology space, which is itself a pointed differential graded Lie algebra. To see this,
consider Dk : gk → gk+1 , the restriction of the differential D to gk . Since D ◦ D = 0,
we have a complex
Dk−1 Dk Dk+1
··· gk−1 gk gk+1 ···

The cohomology of this complex is denoted by HD• (g) = k
k∈Z HD (g), where, for
all k ∈ Z,
HDk (g) := Ker Dk / Im Dk−1 .
D is a graded derivation of degree one of (g, [· , ·]), hence [· , ·] induces a Lie bracket
on HD• (g), also denoted by [· , ·] and the triple (HD• (g), [· , ·], 0) is a pointed differential
graded Lie algebra.

We now show that the differential and the bracket of a differential graded Lie algebra
(g, [· , ·], D) can be encoded in a single differential Λ on S• g. Let g be a graded vector
space, let [· , ·] : g × g → g be a graded skew-symmetric bilinear map of degree zero
and let D : g → g be a graded linear map of degree one. Two linear maps from S• g
to itself are defined by setting for every homogeneous monomial X = x1 . . . x ∈ S g,

ΛD (X) := ∑ sgn(σi ; X)D(xi )X̂i ,
i=1 (13.29)
∑
0 (x
Λ[· ,·] (X) := sgn(σi j ; X)(−1)d i) [xi , x j ] X̂i j ,
1i< j

where X̂i and X̂i j are obtained by removing from X the element xi , respectively the
elements xi and x j . For example, X̂i = x1 . . . xi−1 xi+1 . . . x . The permutation σi ∈ S
is the permutation which sends 1 to i and subtracts one from the integers 2, . . . , i and
fixes the other integers, i.e., it is the cycle (i, i − 1, . . . , 1); similarly, σi j is the per-
mutation which sends 1 to i, sends 2 to j, removes two from the integers 3, . . . , i + 1,
removes one from the integers i + 2, . . . , j and fixes the other integers. We leave it as
an exercise to the reader to check that the maps ΛD and Λ[· ,·] are well-defined. We
define Λ to be the sum of the two linear maps in (13.29),

Λ := ΛD + Λ[· ,·] . (13.30)

Taking a homogeneous monomial X ∈ S• g, it is easy to see that d 0 (ΛD (X)) =

d 0 (Λ[· ,·] (X)) = d 0 (X) + 1, so that Λ has degree 1. Notice however that ΛD : S g →
S g, while Λ[· ,·] : S g → S−1 g.

Proposition 13.22. Let (g, [· , ·], D) be a triple, where g = k∈Z gk is a graded vector
space, D : g → g a graded linear map of degree one and [· , ·] : ∧2 g → g a graded
skew-symmetric bilinear map of degree zero. Let ΛD , Λ[· ,·] and Λ denote the linear
maps defined in (13.29) and (13.30). Then the following statements are equivalent.
(i) (g, [· , ·], D) is a differential graded Lie algebra;
13.3 Differential Graded Lie Algebras 375

(ii) ΛD2 = Λ[·2,·] = 0 and ΛD ◦ Λ[· ,·] + Λ[· ,·] ◦ ΛD = 0;

(iii) Λ 2 = 0.
For a given differential graded Lie algebra (g, [· , ·], D), we call the linear map Λ :
S• g → S• g the total differential of (g, [· , ·], D).
Proof. As we already pointed out, for X ∈ S g, we have that ΛD (X) ∈ S g, while
Λ[· ,·] (X) ∈ S−1 g. Therefore, ΛD2 (X) ∈ S g, while Λ[·2,·] (X) ∈ S−2 g and (ΛD ◦ Λ[· ,·] +
Λ[· ,·] ◦ ΛD )(X) ∈ S−1 g. Since

Λ 2 = ΛD2 + ΛD ◦ Λ[· ,·] + Λ[· ,·] ◦ ΛD + Λ[·2,·] ,

the equivalence of (ii) and (iii) follows. Let X = x1 . . . x be a homogeneous mono-

mial of S g. Then

ΛD2 (X) = ∑ sgn(σi ; X) D2 (xi )X̂i ,
i=1

so that ΛD2 = 0 if and only if D2 = 0. Similarly,

Λ[·2,·] (X) = ∑ sgn(σi jk ; X)(−1)|xi ||xk |+|x j | αi jk (X)X̂i jk ,

1i< j<k

where σi jk and X̂i jk are defined similarly to σi j and X̂i j , defined earlier in this section,
and
αi jk (X) := (−1)|xi ||xk | [[xi , x j ] , xk ] +  (i, j, k) .
It follows that Λ[·2,·] = 0 if and only if [· , ·] satisfies the graded Jacobi identity. Finally,

(ΛD ◦ Λ[· ,·] + Λ[· ,·] ◦ ΛD )(X) = − ∑ sgn(σi j ; X)(−1)|xi | βi j (X)X̂i j ,

1i< j

where
βi j (X) := D([xi , x j ]) − [D(xi ), x j ] − (−1)|xi | [xi , D(x j )] ;
as a consequence, ΛD ◦ Λ[· ,·] + Λ[· ,·] ◦ ΛD = 0 if and only if D is a derivation of [· , ·].
The last three equivalences, combined, prove the equivalence of (i) and (ii).
We finish this section by giving the definition of a morphism of differential
graded Lie algebras.
Definition 13.23. Let (g, [· , ·]g , Dg ) and (h, [· , ·]h , Dh ) be two differential graded Lie
algebras. A graded linear map φ : g → h of degree zero is said to be a morphism of
differential graded Lie algebras, if φ is a homomorphism of graded Lie algebras
and φ ◦ Dg = Dh ◦ φ .
Let us denote by Λg and Λh the total differentials of (g, [· , ·]g , Dg ), respectively
of (h, [· , ·]h , Dh ). It follows easily from the definitions that a graded linear map of
degree zero φ : g → h is a morphism of differential graded Lie algebras if and only
if S• φ ◦ Λg = Λh ◦ S• φ , where S• φ : S• g → S• h is defined by
376 13 Deformation Quantization

S• φ (x1 . . . x ) := φ (x1 ) . . . φ (x ) (13.31)

for all x1 , . . . , x ∈ g.

13.3.3 The Maurer–Cartan Equation

In this section, we define the Maurer–Cartan equation associated to a general dif-

ferential graded Lie algebra. We have already met this equation in the case of the
pointed differential graded Lie algebras ( HC• (A ), [· , ·]G , μ ) and (X• (A ), [· , ·]S , π )
(see (13.11) and (13.25)); as we explained, these equations govern deformations of
associative products and of Poisson brackets. We will come back to these particular
cases in Section 13.3.6 below.
As in the case of deformations of associative products and of Poisson brack-
ets, we consider formal power series with coefficients in a (differential graded
Lie) algebra. Thus, given a differential graded Lie algebra (g, [· , ·], D), we consider
gν := g[[ν ]], the Fν -module of formal power series with coefficients in g. Every
element x of gν is of the form x = ∑∞ k=0 xk ν ; even without writing explicitly x in
k

this form, we will often write xk (or (x)k in the case of ambiguity) for the coefficient
of ν k in x, a notation which comes in particularly handy when picking a particular
coefficient of the series of a complex expression.
We keep the same notations [· , ·] and D for the Fν -linear extension of [· , ·] and
of D to gν . Similarly, when Λ denotes the total differential on S• g, then Λ will still
denote its extension to (S• g)ν := (S• g)[[ν ]]. Consider the subspace ν gν1 := ν g1 [[ν ]]
of gν , which consists of all formal power series in ν without constant term and with
coefficients in g1 , the degree one component of g. On ν gν1 we consider the equation

1
D(x) + [x, x] = 0 , (13.32)
2
which is called the Maurer–Cartan equation associated to (g, [· , ·], D) (or to g). We
denote by MC(g) the set of all elements x of ν gν1 which are solutions of the Maurer–
Cartan equation associated to g. We will give in Proposition 13.26 below two char-
acterizations of the solutions of the Maurer–Cartan equation. To do this, we define,
for x ∈ ν gν1 and for k ∈ N∗ ,

1
Obsk (x) := D(xk ) +
2 ∑ [xi , x j ] . (13.33)
i+ j=k
i, j1

Then (13.32) can be written out as the infinite list of equations Obsk (x) = 0,
where k ∈ N∗ . If x satisfies Obsk (x) = 0 for k = 1, . . . ,  we say that x is a solu-
tion of the Maurer–Cartan equation up to order  and we denote by MC (g) the set
of all such elements. Clearly, x ∈ MC (g) if and only if x satisfies (13.32) modulo
13.3 Differential Graded Lie Algebras 377

ν +1 and x ∈ MC(g) if and only if x ∈ MC (g) for all  ∈ N∗ . Moreover, we have
the following proposition.
Proposition 13.24. Let (g, [· , ·], D) be a differential graded Lie algebra. Suppose
that x ∈ MC (g) for some  ∈ N∗ . Then Obs+1 (x) is a cocycle.
Proof. Since D is a graded derivation, satisfying D2 = 0, and since the Lie bracket
is graded skew-symmetric, we have that
1
D (Obs+1 (x)) =
2 ∑ ([D(x j ), xk ] − [x j , D(xk )]) = ∑ [D(x j ), xk ] .
j+k=+1 j+k=+1
j,k1 j,k1

Since x satisfies the Maurer–Cartan equation up to order , we obtain

1
D (Obs+1 (x)) = −
2 ∑ [[xi , x j ] , xk ] = 0 ,
i+ j+k=+1
i, j,k1

where we used the graded Jacobi identity in the last step.

For the second characterization, we use the exponential map. For x ∈ ν gν , we

define ex to be the following formal power series

x2 xk
ex := 1 + x + + · · · + + · · · ∈ (S• g)ν . (13.34)
2 k!
Notice that this infinite sum is a well-defined formal power series in ν with co-
efficients in S• g since at most the first k + 1 terms of the infinite sum in (13.34)
contribute to the coefficient of ν k in ex (recall that x is a formal power series in
ν without constant term). The total differential Λ , applied to ex , can be explicitly
computed, as is shown in the following lemma.
Lemma 13.25. Let (g, [· , ·], D) be a differential graded Lie algebra with total differ-
ential Λ : S• g → S• g. For x ∈ ν gν1 , one has
 1 
Λ (ex ) = D(x) + [x, x] ex . (13.35)
2
Proof. One computes directly that

k D(x)xk−1 k(k − 1) [x, x] k−2

Λ (ex ) = ∑ k!
+∑
2 k!
x
k1 k2
 1 
= D(x) + [x, x] ex .
2
This completes the proof.
The lemma and the remarks made about Obsk lead at once to the proof of the fol-
lowing proposition.
378 13 Deformation Quantization

Proposition 13.26. Let (g, [· , ·], D) be a differential graded Lie algebra with total
differential Λ : S• g → S• g. For x ∈ ν gν1 , the following statements are equivalent.
(i) x satisfies the Maurer–Cartan equation associated to g;
(ii) Λ (ex ) = 0;
(iii) Obsk (x) = 0 for every k ∈ N∗ .
In the case of a pointed differential graded Lie algebra (g, [· , ·], z), these conditions
are also equivalent to
(iv) [z + x, z + x] = 0.

13.3.4 Gauge Equivalence

We have introduced in Sections 13.1.1 and 13.2.1 the notion of equivalence of de-
formations in the case of associative products and of Poisson brackets. We now
introduce the corresponding notion for formal power series (without constant term)
with coefficients in the degree one component of an arbitrary pointed differential
graded Lie algebra.
Definition 13.27. Let (g, [· , ·], z) be a pointed differential graded Lie algebra. Two
formal power series without constant term x, y ∈ ν gν are said to be gauge equivalent
denoted by x ∼ y, if there exists ξ ∈ ν gν0 such that

z + y = eadξ (z + x) .

The definition requires some explanation. As the notation suggests,2 eadξ stands for
the endomorphism of gν , defined for x ∈ gν by

∞ adkξ x 1
e adξ
(x) := ∑ k!
= x + [ξ , x] + [ξ , [ξ , x]] + · · · .
2
(13.36)
k=0

The latter infinite sum is a well-defined formal power series in ν with coefficients
in g since at most the first k + 1 terms of the sum contribute to the coefficient of ν k .
Notice that it follows easily from (13.36) that

d adt ξ  
e (x) = ξ , eadt ξ (x) , (13.37)
dt
for ξ ∈ ν gν0 and x ∈ gν . When we work in a fixed pointed differential graded Lie
algebra (g, [· , ·], z), we will make extensive use of the notation
∞
1 k
ξ  x := eadξ (z + x) − z = x + ∑ adξ (z + x) ,
k=1 k!
2The endomorphism eadξ of gν , defined here, should not be confused with the exponential eξ ,
which is the element of (S• g)ν , defined in (13.34).
13.3 Differential Graded Lie Algebras 379

where x and ξ are as above. In terms of this notation, x, y ∈ ν gν are gauge equivalent
if and only if there exists ξ ∈ ν gν0 such that y = ξ  x.
In order to show that the relation ∼ is an equivalence relation, we use the
Campbell–Hausdorff formula

exp(u) exp(v) = exp(CH(u, v)) , (13.38)

where CH(u, v) is an infinite sum, each of whose terms consists only of repeated
brackets of u and v. The first few terms of CH(u, v) are given by

1 1
CH(u, v) = u + v + [u, v] + ([u, [u, v]] + [v, [v, u]]) + · · · . (13.39)
2 12
In particular, CH(u, −u) = 0 for all u. This formula is usually given for u and v in
a finite-dimensional Lie algebra, with exp being the exponential map between the
Lie algebra and its Lie group (see for example [29]), but the Campbell–Hausdorff
formula is essentially a formal identity, in particular it is equally valid when for
example u and v are formal power series (without constant term) with coefficients
in an arbitrary Lie algebra. In the present case, we apply it for u and v of the form
adξ , where ξ is a formal power series without constant term and the Lie bracket is
the graded commutator (of graded linear maps, such as adξ ). For ξ , η ∈ ν gν0 and
x ∈ ν gν , the Campbell–Hausdorff formula and the fact that ad is a representation
imply that
eadη eadξ (x) = eCH(adη ,adξ ) (x) = eadCH(η ,ξ ) (x) ,
so that
η  (ξ  x) = CH(η , ξ )  x . (13.40)
The fact that ∼ is an equivalence relation follows at once from it.
To finish this section, we show that the equivalence relation ∼ is compatible with
the Maurer–Cartan equation, i.e., that if x ∼ y, then x is a solution of the Maurer–
Cartan equation if and only if y is a solution of the Maurer–Cartan equation. To do
this, let x, y ∈ ν gν1 be two equivalent solutions of the Maurer–Cartan equation and let
ξ ∈ ν gν0 be such that y = ξ  x. Since x is a solution of the Maurer–Cartan equation
if and only if [z + x, z + x] = 0, we need to show that
 
eadξ (z + x), eadξ (z + x) = 0 .

For s,t ∈ gν , the fact that adξ is a derivation of the Lie bracket implies that
 
∞
1 j ∞ j
1 j  i 
e adξ
[s,t] = ∑ adξ [s,t] = ∑ ∑ adξ s, adξj−i t
j=0 j! j=0 i=0 j! i
∞ j   ∞  
1 i 1 1 k 1 
= ∑∑ adξ s, adξj−i t = ∑ adξ s, adξ t
j=0 i=0 i! ( j − i)! k,=0 k! !
380 13 Deformation Quantization
 
= eadξ s, eadξ t .
 
We may conclude that eadξ (z + x), eadξ (z + x) = eadξ [z + x, z + x] = 0, since x is a
solution of the Maurer–Cartan equation.
The upshot is that gauge equivalence defines an equivalence relation on the set of
solutions of the Maurer–Cartan equation associated to g. For x ∈ MC(g), we denote
by cl(x) its equivalence class modulo gauge equivalence.

13.3.5 Path Equivalence

We introduce in this section another notion of equivalence of solutions of the

Maurer–Cartan equation. This notion, which turns out to be equivalent to the no-
tion of gauge equivalence (see Proposition 13.31 below) seems both less intuitive
and technically more complicated, but we will see that it is easier to work with the
latter notion when dealing with L∞ -morphisms, introduced in the next section.
We first define the notion of a polynomial path in V ν := V [[ν ]], where V is an
arbitrary F-vector space. By definition, a polynomial path in V ν is a map from F to
V ν whose component in V ν k is for every k ∈ N a polynomial in t; it means that a
polynomial path x(t) in V ν can be written as
∞
x(t) = ∑ xk (t)ν k ,
k=0

where each one of the functions xk (t) is a polynomial function, with values in V . We
will often consider such paths satisfying x0 (t) = 0; they will be referred to as poly-
nomial paths in ν V ν . Polynomial paths in V ν admit well-defined derivatives with
respect to t, which are themselves also polynomial paths in V ν . Also, polynomial
paths in a (graded) algebra, such as (S•V )ν , form an algebra and, for every polyno-
mial path γ (t) in ν V ν , its exponential eγ (t) (defined as in (13.34)) is a polynomial
path in (S•V )ν .
Definition 13.28. Let (g, [· , ·], z) be a pointed differential graded Lie algebra, and
let Λ denote the total differential on (S• g)ν , as in (13.30). We say that two solutions
x, y ∈ ν gν1 of the Maurer–Cartan equation are path equivalent if there exists two
polynomial paths γ (t), η (t) in ν gν1 and ν gν0 respectively, such that γ (0) = x, γ (1) = y
and  
d γ (t)
e = Λ η (t)eγ (t) . (13.41)
dt
We will show in Proposition 13.31 below that two solutions of the Maurer–Cartan
equation are gauge equivalent if and only if they are path equivalent. We will do this
by using the following two lemmas, the first one of which can be viewed as giving
yet another (equivalent!) notion of equivalence of solutions of the Maurer–Cartan
equation.
13.3 Differential Graded Lie Algebras 381

Lemma 13.29. Let (g, [· , ·], z) be a pointed differential graded Lie algebra and
suppose that x, y ∈ ν gν1 are two solutions of the Maurer–Cartan equation asso-
ciated to g. Let γ (t) and η (t) be polynomial paths in ν gν1 and ν gν0 respectively,
with γ (0) = x and γ (1) = y. Then condition (13.41) is equivalent to the following
condition:
d
γ (t) = [z + γ (t), η (t)] . (13.42)
dt
Proof. If condition
 (13.41)
 holds, then the component in gν = S1 gν ⊂ (S• g)ν of
d γ (t) γ
dt e − Λ η (t)e (t) vanishes. Since the components of γ (t) ∈ ν gν1 commute, this
component is given by
d d
γ (t) − ΛDz (η (t)) − Λ[· ,·] (η (t)γ (t)) = γ (t) − [z + γ (t), η (t)] ,
dt dt
condition (13.41) implies condition (13.42).
Conversely, suppose that γ (t) and η (t) satisfy condition (13.42), where γ (0) = x
is a solution of the Maurer–Cartan equation and γ (1) = y. We first show that γ (t)
is a solution of the Maurer–Cartan equation for all t, which we do by showing that
[z + γ (t), z + γ (t)] = 0 for all t. This is in turn done by showing that the polynomial
path F(t) := [z + γ (t), z + γ (t)] is the unique solution of a linear differential equa-
tion, with initial condition F(0) = 0. Taking the derivative of F and using (13.42)
and the graded Jacobi identity we find
   
d d d
F(t) = γ (t), z + γ (t) + z + γ (t), γ (t)
dt dt dt
= [[z + γ (t), η (t)] , z + γ (t)] + [z + γ (t), [z + γ (t), η (t)]]
= [[z + γ (t), z + γ (t)] , η (t)]
= [F(t), η (t)] ,

which yields the required linear differential equation. We write F(t) = ∑k1 Fk (t)ν k
and η (t) = ∑k1 ηk (t)ν k , where all coefficients Fk (t) and ηk (t) depend polynomi-
ally on t. Then the differential equation can be written out as an infinite list of
equations
k−1
d
Fk (t) = ∑ [Fk− (t), η (t)] and Fk (0) = 0 ,
dt =1

where k ∈ N∗ . An easy recursion on k implies that Fk (t) = 0 for all k ∈ N∗ . In

conclusion, γ (t) is a solution of the Maurer–Cartan equation for all t ∈ F. Next, we
sum up, for k ∈ N, each one of the following equalities,
 
ΛDz η (t)γ k (t) = ΛDz (η (t))γ k (t) + ΛDz (γ k (t))η (t),
   
Λ[· ,·] η (t)γ k (t) = −k [η (t), γ (t)] γ k−1 (t) + Λ[· ,·] γ k (t) η (t) ,

to find that
382 13 Deformation Quantization
   
Λ η (t)eγ (t) = [z + γ (t), η (t)] eγ (t) + Λ eγ (t) η (t)
= [z + γ (t), η (t)] eγ (t) ,

where we have used in the last step that γ (t) ∈ MC(g) for all t, combined with
item (ii) of Proposition 13.26. It follows that

d γ (t)   d 
e − Λ η (t)eγ (t) = γ (t) − [z + γ (t), η (t)] eγ (t) ,
dt dt

showing that condition (13.42) implies condition (13.41).

The second lemma which we will use to prove Proposition 13.31 below can
be seen as a generalization of the classical formula, giving the derivative of e f (t) ,
when f (t) takes values in a non-commutative algebra.
Lemma 13.30. Let (g, [· , ·]) be a graded Lie algebra and let ξ (t) be a polynomial
path in ν gν0 . Then

dead−ξ (t)
eadξ (t) ◦ = − ad 1 dξ (t) . (13.43)
dt ∑ (k+1)! adkξ (t) ( dt )
k∈N

Proof. Given a polynomial path ξ (t) in ν gν0 , we consider for fixed t the polynomial
path in End(g1 )ν , defined for s ∈ F by

dead−sξ (t)
Gs := eadsξ (t) ◦ + ad sk+1 dξ (t) . (13.44)
dt ∑ (k+1)!
adkξ (t) dt
k∈N

It is clear that G0 = 0. We will show that

d
Gs = adξ (t) ◦ Gs − Gs ◦ adξ (t) . (13.45)
ds
It follows, as in the proof of Lemma 13.29, by writing Gs and ξ as formal series in
ν that all coefficients of G as a series in ν vanish, so that Gs = 0 for all s ∈ F. In
particular, G1 = 0, which is tantamount to (13.43).
In order to show (13.45), we write the two terms in the right-hand side of (13.44)
as As and Bs , so that Gs = As + Bs . Explicitly, As (x) and Bs (x) are given, for x ∈ gν1 ,
by
 
d ad−sξ (t)
As (x) := eadsξ (t) e (x) ,
 dt  (13.46)
sk+1
Bs (x) := ∑ adξ (t) ξ̇ (t), x ,
k
k∈N (k + 1)!

where ξ̇ (t) is an abbreviation for dt ξ (t).

d
Using (13.37) twice, we compute
13.3 Differential Graded Lie Algebras 383
 
d d
As (x) = [ξ (t), As (x)] + eadsξ (t) −ξ (t), ead−sξ (t) (x)
ds dt
d  ad−sξ (t) 
= [ξ (t), As (x)] − eadsξ (t) e ([ξ (t), x])
dt  
= [ξ (t), As (x)] − As ([ξ (t), x]) − ξ̇ (t), x . (13.47)

For the derivative of Bs , it follows at once from the second equation in (13.46) that
d  
Bs (x) = eadsξ (t) ξ̇ (t), x ,
ds
so that, in view of the graded Jacobi identity,
 
sk+1
[ξ (t), Bs (x)] = ∑ ξ (t) ξ̇ (t), x + Bs ([ξ (t), x])
adk+1
k∈N (k + 1)!
d  
= Bs (x) − ξ̇ (t), x + Bs ([ξ (t), x]) ,
ds
giving the following formula for the derivative of Bs :
d  
Bs (x) = [ξ (t), Bs (x)] − Bs ([ξ (t), x]) + ξ̇ (t), x . (13.48)
ds
Equations (13.47) and (13.48), combined, yield the proof of (13.45), and hence
of (13.43).
Using the two lemmas, we show that gauge equivalence and path equivalence are
the same (for solutions of the Maurer–Cartan equation).
Proposition 13.31. Let (g, [· , ·], z) be a pointed differential graded Lie algebra. Two
solutions x, y ∈ ν gν1 of the Maurer–Cartan equation associated to g are gauge equiv-
alent if and only if they are path equivalent.
Proof. Let us first assume that x and y are gauge equivalent. Then there exists ξ ∈
ν gν0 such that y = ξ  x. For t ∈ F, define

γ (t) := (t ξ )  x = eadt ξ (z + x) − z ,

and let η (t) := −ξ , which is a constant polynomial path. Then γ (0) = x, γ (1) = y
and it follows from (13.37) that γ (t) and η (t) satisfy condition (13.42). In view of
Lemma 13.29, x and y are path equivalent.
Conversely, let us assume that x and y are path equivalent, and let γ (t), η (t) be
polynomial paths in ν gν1 and ν gν0 respectively, such that γ (0) = x, γ (1) = y and such
that the equivalent conditions (13.41) and (13.42) are satisfied. In order to establish
that x, y are gauge equivalent, we will construct a polynomial path ξ (t) in ν gν0 such
that γ (t) = ξ (t)  x, i.e.,

e− adξ (t) (z + γ (t)) = z + x , (13.49)

384 13 Deformation Quantization

for all t ∈ F. It implies that y = γ (1) = ξ (1)  x, so that x and y are gauge equivalent.
In order to satisfy (13.49) for t = 0, we require ξ (0) = 0. With this additional as-
sumption, the condition which we wish to impose on ξ (t) is dtd e− adξ (t) (z + γ (t)) = 0,
or equivalently:
d  − adξ (t) 
eadξ (t) e (z + γ (t)) = 0 .
dt
We can compute the left-hand side of the previous equation with the help of (13.43)
and (13.42) as follows:

d  − adξ (t)  de− adξ (t) d

eadξ (t) e (z + γ (t)) = eadξ (t) (z + γ (t)) + γ (t)
dt dt dt
d
= − ad 1 dξ (t) (z + γ (t)) + γ (t)
∑ (k+1)! adkξ (t) dt dt
k∈N
 
1 dξ (t)
= z + γ (t), ∑ k
adξ (t) + η (t) .
k∈N (k + 1)! dt

It suffices therefore to find a polynomial path ξ (t) in ν gν0 which satisfies

1 dξ (t)
∑ (k + 1)! adkξ (t) dt
= −η (t) (13.50)
k∈N

and ξ (0) = 0. To see that such a polynomial path ξ (t) exists, substitute ξ (t) =
∑∈N∗ ξ (t)ν  and η (t) = ∑∈N∗ η (t)ν  in (13.50) and observe that the equation
decomposes into an infinite number of equations

dξ (t) −1

1 dξ j (t)
= −η (t) − ∑ ∑ adξi (t) ◦ · · · ◦ adξik (t) , (13.51)
dt k=1 (k + 1)! i1 +···+ik + j=
1 dt
i1 ,...,ik , j1

indexed by  ∈ N∗ , which express dξdt (t) as a multi-linear expression in η (t) and

ξ1 (t), . . . , ξ−1 (t). By integration, we find a unique polynomial ξ (t), satisfying the
initial condition ξ (0) = 0. Thus, the coefficients of ξ (t) can be constructed recur-
sively. It leads to a polynomial path ξ (t) in ν gν0 which at t = 1 realizes the gauge
equivalence between x and y.

Remark 13.32. Let x, y ∈ ν gν1 be solutions of the Maurer–Cartan equation, as in

Proposition 13.31. Suppose that x and y are gauge equivalent, y = ξ  x, where
ξ ∈ ν  gν0 . Then it is clear from the first part of the proof of the proposition that the
constant polynomial path η (t) belongs to ν  gν0 . Alternatively, suppose that x and y
are path equivalent, where the polynomial path η (t) belongs to ν  gν0 . Then the in-
tegration of (13.51) gives a polynomial path ξ (t) in ν  gν0 , since ηk (t) = 0 for k < .
Then ξ (1) ∈ ν  gν0 , so that x and y are gauge equivalent via an element of ν  gν0 .
13.3 Differential Graded Lie Algebras 385

13.3.6 Applications to Deformation Theory

We pointed out in Example 13.17 that for every Poisson algebra (A , ·, π ), the triple
(X• (A ), [· , ·]S , π ) is a pointed differential graded Lie algebra. Similarly, accord-
ing to Example 13.19, for every commutative associative algebra (A , μ ), the triple
( HC• (A ), [· , ·]G , μ ) is a pointed differential graded Lie algebra. Up to a sign, the
differentials Dπ and Dμ coincide with the Poisson and Hochschild coboundary op-
erators, namely

δπ (P) = (−1) p Dπ (P) , δμ (ψ ) = (−1)q Dμ (ψ ) ,

for P ∈ X p (A ) and ψ ∈ HCq (A ). Thus, the cohomology of (X• (A ), [· , ·]S , π ) is the

Poisson cohomology of (A , ·, π ) and the cohomology of ( HC• (A ), [· , ·]G , μ ) is the
Hochschild cohomology of (A , μ ). Since for elements of shifted degree 1, δπ and
Dπ have opposite sign, the Maurer–Cartan equation (13.32) of (X• (A ), [· , ·]S , π )
is precisely the Maurer–Cartan equation (13.25) of (A , ·, π ), and similarly for the
Maurer–Cartan equation (13.11) of (A , μ ).
The fact that the notion of gauge equivalence of solutions of the Maurer–Cartan
equation for pointed differential graded Lie algebras coincides in the cases of
(HC• (A ), [· , ·]G , μ ) and (X• (A ), [· , ·]S , π ) with the notion of equivalence of formal
deformations of associative products and of Poisson brackets is also true, but it is
less obvious. We prove it in the case of formal deformations of associative products,
the proof in the case of Poisson brackets being completely analogous. Thus, we con-
sider the pointed differential graded Lie algebra ( HC• (A ), [· , ·]G , μ ). Let x and y be
solutions of the Maurer–Cartan equation associated to this differential graded Lie al-
gebra, which is equivalent to saying that μ + x and μ + y are formal deformations of
μ . Let us suppose that x and y are gauge equivalent, which means that there exists
an element ξ ∈ ν HC0 (A )ν = ν Hom(A , A )[[ν ]], such that μ + y = eadξ (μ + x),
where adξ = [ξ , ·]G . Let Φ : A ν → A ν be the Fν -linear map, defined by Φ := eξ .
We show that Φ realizes an equivalence between the formal deformations μ + x and
μ + y (see Definition 13.2). Since Φ (F) = F (mod ν ) for all F ∈ A , we only need
to show that
Φ ((μ + x)(Φ −1 (F), Φ −1 (G))) = (μ + y)(F, G)
for all F, G ∈ A , which amounts to showing that

eξ ((μ + x)(e−ξ (F), e−ξ (G))) = (eadξ (μ + x))(F, G)

for all F, G ∈ A . Writing x for μ + x, the left-hand side of the above equation is
given by
  
eξ x e−ξ (F), e−ξ (G)
1 1 1 r  s 
= x (F, G) + ∑∗ ∑ (−1)s+t
r! s! t!
ξ x (ξ (F), ξ t (G))
k∈N r,s,t∈N
r+s+t=k
386 13 Deformation Quantization

1
= x (F, G) + ∑∗ k! adkξ (x )(F, G) = eadξ (x )(F, G)
k∈N

where the second equality is easily proved by induction on k ∈ N∗ . This shows

that if two solutions of the Maurer–Cartan equation are gauge equivalent, then the
corresponding deformations are equivalent. The same computation shows that if two
deformations are equivalent, then the corresponding solutions of the Maurer–Cartan
equation are gauge equivalent: one constructs in this case ξ ∈ ν Hom(A , A )[[ν ]]
from Φ by taking the logarithm (using the series expansion of log(1 + x)). Summing
up, we have shown the following proposition.
Proposition 13.33. Let (A , μ ) be a commutative associative algebra and let x, y ∈
ν HC2 (A )ν .
(1) x is a solution of the Maurer–Cartan equation associated to HC• (A ) if and
only if μ + x is a formal deformation of the associative product μ ;
(2) If x and y are solutions of the Maurer–Cartan equation associated to HC• (A ),
then x and y are gauge equivalent if and only if the formal deformations μ + x
and μ + y are equivalent.
The above proposition and its proof are easily transcribed to the case of Poisson
brackets. The statement is the following.
Proposition 13.34. Let (A , ·, π ) be a Poisson algebra and let x, y ∈ ν X2 (A )ν .
(1) x is a solution of the Maurer–Cartan equation associated to X• (A ) if and
only if π + x is a formal deformation of π ;
(2) If x and y are solutions of the Maurer–Cartan equation associated to X• (A ),
then x and y are gauge equivalent if and only if the formal deformations π + x
and π + y are equivalent.
Both propositions are easily specialized to the case of smooth manifolds; in the case
of Proposition 13.33 one can also restrict oneself to star products on the algebra of
smooth functions on a manifold M, rather than considering arbitrary deformations
of the standard product on C∞ (M). Also, taking all formulas in the above proof
modulo ν k+1 shows that two k-th order deformations μ + x and μ + y are equivalent
if and only if x and y are gauge equivalent up to order k.

13.4 L∞ -Morphisms of Differential Graded Lie Algebras

In this section we introduce the notion of an L∞ -morphism and of an L∞ -quasi-

isomorphism of differential graded Lie algebras and we show that an L∞ -quasi-
isomorphism between two pointed differential graded Lie algebras (g, [· , ·]g , zg )
and (h, [· , ·]h , zh ) induces a bijection between the sets MC(g)/ ∼ and MC(h)/ ∼ of
gauge equivalence classes of solutions of the Maurer–Cartan equation of (g, [· , ·]g , zg )
and of (h, [· , ·]h , zh ).
13.4 L∞ -Morphisms of Differential Graded Lie Algebras 387
 
Let g = ∈Z g and h = ∈Z h be two graded Lie algebras. In what follows we
will be mainly interested in graded linear maps Φ : S• g → h of degree 0. We recall
that g is the same vector space as g, but the degree of its homogeneous elements is
lowered by one; it is this grading which is extended to S• g (see (13.28)). Denoting
for k ∈ N by Φk the restriction of Φ to Sk g, the fact that Φ is graded of degree 0
means that Φk maps gi1 . . . gik to hi1 +···+ik −k+1 for all i1 , . . . , ik ∈ Z. In particular, Φk
maps g1 . . . g1 to h1 , a fact which will be useful when we consider solutions of the
Maurer–Cartan equations associated to g and h.
We extend Φ to a graded linear map Φ  : S• g → S• h, whose component in
•
h ⊂ S h is Φ . First, the restriction of Φ  to S g  F is by definition the identity map
0

onto S0 h  F. For k,  ∈ N∗ and for a homogeneous monomial X = x1 . . . xk ∈ Sk g,

we define the component in S h of Φ  (X) to be given by

1 
∑ ∑ sgn(σ ; X)
! ∏
Φi j (xσ (i1 +···+i j−1 +1) . . . xσ (i1 +···+i j ) ) , (13.52)
i1 ,...,i ∈N∗ σ ∈Si1 ,...,i j=1
i1 +···+i =k

where Si1 ,...,i is the set of (i1 , . . . , i )-shuffles, i.e.,

⎧ # ⎫
⎪ # σ (1)<···<σ (i1 ), ⎪
⎪
⎨ # ⎪
⎬
# σ σ
Si1 ,...,i := σ ∈ Si1 +···+i ##
(i 1 +1)<···< (i 1 +i 2 ),
.
⎪
⎪ # ··· ⎪
⎪
⎩ # σ (i1 +i2 +···+i−1 +1)<···<σ (i1 +i2 +···+i ) ⎭

Remark 13.35.
(1) Φ  is the (unique) extension of Φ as a coderivation of S• g with values in S• h,
but this fact will not be used in what follows.
(2) It is clear that, when Φ is identically zero on Sk g for all k  2, i.e., when Φ is
simply a graded linear map of degree 0 from g to h, then Φ  = S• Φ (see (13.31)).
 (x1 . . . xk ) is zero for all  > k.
(3) It is also clear that the component in S h of Φ
(4) The order of the terms in the product in (13.52) is important: the meaning of
the product is that the terms are ordered as follows: Φi1 Φi2 . . . Φi .

Definition 13.36. Let (g, [· , ·]g , Dg ) and (h, [· , ·]h , Dh ) be two differential graded Lie
algebras, whose total differentials are denoted by Λg and by Λh respectively. A
graded linear map Φ : S• g → h of degree 0 is called an L∞ -morphism between g
and h if
 ◦ Λg = Λh ◦ Φ
Φ . (13.53)

 (x) = Φ1 (x). Therefore, if Φ : S• g → h

For x ∈ g we have that Λg (x) = Dg (x) and Φ
is an L∞ -morphism, then
Φ1 ◦ Dg = Dh ◦ Φ1 , (13.54)
388 13 Deformation Quantization

as maps3 from g to h, so that Φ1 : g → h induces a morphism in cohomology,

Φ1• : HD• g (g) → HD• h (h) .

Notice that Φ1 is not a morphism of differential graded Lie algebras, in general.

However, if a graded linear map φ : g → h is a morphism of differential graded
Lie algebras (see Definition 13.23), then one can extend φ to a graded linear map
Φ : S• g → h, with Φ1 = φ and Φk = 0 for all k  2. According to item (2) of
Remark 13.35, the map Φ is then an L∞ -morphism. In this sense, morphisms of
differential graded Lie algebra are particular cases of L∞ -morphisms.
Definition 13.37. Let (g, [· , ·]g , Dg ) and (h, [· , ·]h , Dh ) be two differential graded Lie
algebras and let Φ : S• g → h be an L∞ -morphism. Then Φ is called an L∞ -quasi-
isomorphism if the morphism Φ1• : HD• g (g) → HD• h (h), induced by the restriction
Φ1 : g → h of Φ to g, is an isomorphism.
We have seen in (13.54) that the equation Φ  , which is an equality of
 ◦ Λg = Λh ◦ Φ
• •
maps S g → S h yields, when restricted to S g = g, the equality Φ1 ◦ Dg = Dh ◦ Φ1 .
1

Similarly, we can consider the restriction to Sk g, which yields an equality of maps,

involving Φ1 , . . . , Φk . In the example which follows, we show how this equality is
obtained for k = 2.
Example 13.38. Let (g, [· , ·]g , Dg ) and (h, [· , ·]h , Dh ) be two differential graded Lie
algebras and let Φ : S• g → h be an L∞ -morphism. We compute the restriction of the
compact equation (13.53) to S2 g. To do this, let x, y ∈ g, with x homogeneous, and
recall that, by definition
0 (x) 0 (x)
Λg (xy) = Dg (x)y + (−1)d xDg (y) + (−1)d [x, y]g ,
 (xy) = Φ2 (xy) + Φ1 (x)Φ1 (y) ,
Φ

 (x) = Φ1 (x). It follows that

while Λg (x) = Dg (x) and Φ

 ◦ Λg (xy) = Φ1 (Dg (x))Φ1 (y) + Φ2 (Dg (x)y)

Φ
0
 
+ (−1)d (x) Φ1 (x)Φ1 (Dg (y)) + Φ2 (xDg (y)) + Φ1 ([x, y]g )

and
 (xy) = Dh (Φ2 (xy)) + Dh (Φ1 (x))Φ1 (y)
Λh ◦ Φ
0
 
+ (−1)d (x) [Φ1 (x), Φ1 (y)]h + Φ1 (x)Dh (Φ1 (y)) .

Equating the right-hand sides of these two equations yields, upon using that Φ1 ◦
Dg = Dh ◦ Φ1 , the following equation, involving both the maps Φ1 and Φ2 :
3 Strictly speaking, Φ1 is a map from S1 g → S1 h, but we rather consider it as a map from g to h,
i.e., we consider g and h with their original grading, as differential graded Lie algebras; with this
convention Φ1 induces maps in cohomology with the following labeling: Φ1k : HDk g (g) → HDk h (h).
13.4 L∞ -Morphisms of Differential Graded Lie Algebras 389

[Φ1 (x), Φ1 (y)]h − Φ1 ([x, y]g )

0 (x)   (13.55)
= Φ2 (xDg (y)) + (−1)d Φ2 (Dg (x)y) − Dh (Φ2 (xy)) .

Let Φ : S• g → h be a graded linear map of degree 0. The Fν -linear extension of

 . To Φ , one can associate
Φ to (S• g)ν will also be denoted by Φ , and similarly for Φ
a map Ω̃Φ : ν gν → ν hν , defined for x ∈ ν gν by:
1
Ω̃Φ (x) := ∑∗ k! Φk (xk ) . (13.56)
k∈N

The infinite sum makes sense because, for every k ∈ N∗ , at most the first k terms
contribute to the coefficient of ν k in Ω̃Φ (x). Notice that if x ∈ ν gν1 , then all elements
xk are formal power series whose coefficients are elements of degree 0 in S• g, since
xk is of degree 0 when x ∈ g1 (or x ∈ gν1 ). It follows that Φ̃k (xk ) ∈ ν hν1 for all k ∈ N∗ ,
hence Ω̃Φ (x) ∈ ν hν1 .
Sections 13.4.1 to 13.4.3 below are devoted to the proof of the following theorem.
Theorem 13.39. Let (g, [· , ·]g , zg ) and (h, [· , ·]h , zh ) be two pointed differential gra-
ded Lie algebras. If Φ : S• g → h is an L∞ -morphism, then

ΩΦ : MC(g)/ ∼ → MC(h)/ ∼
 
cl(x) → cl Ω̃Φ (x)

is a well-defined map, which is a bijection if Φ is an L∞ -quasi-isomorphism.

Theorem 13.39 is true more generally for L∞ -morphism of arbitrary (not necessarily
pointed) differential graded Lie algebras, but we will not use it in that generality.

13.4.1 ΩΦ is Well-Defined

We start with a proposition which will be useful for showing that ΩΦ is well-defined.

Proposition 13.40. Let (g, [· , ·]g , Dg ) and (h, [· , ·]h , Dh ) be two differential graded
Lie algebras. Let Φ : S• g → h be a graded linear map of degree 0 and let Ω̃Φ be the
associated map, defined as in (13.56).
(1) If x ∈ ν gν1 , then
 (ex ) = eΩ̃Φ (x) .
Φ (13.57)
(2) If x ∈ ν gν1 and ω ∈ ν gν , then
 
 (ω e ) = ∑ 1
Φ x
Φk (ω x ) eΩ̃Φ (x) .
k−1
(13.58)
k∈N∗ (k − 1)!
390 13 Deformation Quantization

Proof. Let x ∈ ν gν1 . It follows from the definition of Ω̃Φ , that

1 1
eΩ̃Φ (x) = 1 + ∑∗ ! ∑ Φi1 (xi1 ) . . . Φi (xi ) . (13.59)
∈N i1 ,...,i ∈N∗ 1 . . i !

i ! .

For every k ∈ N∗ and for every permutation σ ∈ Sk , one has sgn(σ ; xk ) = 1, because
x ∈ ν gν1 . Hence, by definition of Φ̂ , we have:

1
 (ex ) =
Φ ∑ k! Φ (xk )
k∈N
1 1
= 1+ ∑∗ k! ∑∗ ∑ ! σ ∈S∑
Φi1 (xi1 ) . . . Φi (xi )
k∈N ∈N i1 +···+i =k i ,...,i
1 
i1 ,...,i 1
1 1
= 1+ ∑∗ ! ∑ ∗ (i + · · · + i )! ∑ Φi1 (xi1 ) . . . Φi (xi )
∈N i1 ,...,i ∈N

1  σ ∈S i ,...,i
1 

which is the same as the right-hand side of (13.59), since #Si1 ,...,i = (i1i+···+i
1 !...i !
 )!
. This
proves (1).
In order to prove (2), we introduce a formal parameter t and we compute the
coefficient in t of both sides of the equality

 (ex+t ω ) = eΩ̃Φ (x+t ω )

Φ (mod t 2 ) . (13.60)

The proof that this equality holds is the same as the proof of (13.57) since everything
is computed modulo t 2 , so that all products of elements of odd degree in h, which
are precisely the ones which do not commute, are discarded. For the same reason,
the coefficient of t in Φ  (ex+t ω ) is given by Φ  (ω ex ), which is the left-hand side
of (13.58), and the coefficient of t in e Φ Ω̃ (x+t ω ) is given by the coefficient of t of

eΩ̃Φ (x) eZ1 (t) , where Z1 (t) is the linear term in t of Ω̃Φ (x + t ω ), which is given by
k−1
1 1
Z1 (t) = ∑∗ k! ∑ Φk (x (t ω )xk−−1) = ∑∗ (k − 1)! Φk (ω xk−1 )t ,
k∈N =0 k∈N

so that the coefficient of t in eΩ̃Φ (x+t ω ) is given by the right-hand side of (13.58),
which establishes (2).

Proposition 13.41. Let (g, [· , ·]g , zg ) and (h, [· , ·]h , zh ) be two pointed differential
graded Lie algebras. Let Φ : S• g → h be an L∞ -morphism and let Ω̃Φ : ν gν → ν hν
be the associated map, defined as in (13.56).
(1) If x ∈ ν gν1 is a solution of the Maurer–Cartan equation associated to g, then
Ω̃Φ (x) is a solution of the Maurer–Cartan equation associated to h;
(2) If x, y ∈ ν gν1 are gauge equivalent solutions of the Maurer–Cartan equation
associated to g, then Ω̃Φ (x) and Ω̃Φ (y) are gauge equivalent;
13.4 L∞ -Morphisms of Differential Graded Lie Algebras 391

(3) If x ∈ ν gν1 and ξ ∈ ν k gν0 for some k ∈ N, then there exists s ∈ ν k hν0 , such that
Ω̃Φ (ξ  x) = s  Ω̃Φ (x). Moreover, s = Φ1 (ξ ) (mod ν k+1 ).

Proof. We denote by Λg and Λh the total differential of g, respectively of h. Accord-

ing to Proposition 13.26, if x ∈ ν gν1 , then x ∈ MC(g) if and only if Λg (ex ) = 0, and
Ω̃Φ (x) ∈ MC(h) if and only if Λh (eΩ̃Φ (x) ) = 0. The fact that Φ is an L∞ -morphism
and (13.57) imply that

Λh (eΩ̃Φ (x) ) = Λh ◦ Φ̂ (ex ) = Φ̂ ◦ Λg (ex ) , (13.61)

and (1) follows.

Let x, y ∈ ν gν1 be two gauge equivalent solutions of the Maurer–Cartan equation.
As we saw in Proposition 13.31, x and y are also path equivalent, which means, by
definition, that there exist a polynomial path γ (t) in ν gν1 and a polynomial path η (t)
in ν gν0 , such that γ (0) = x, γ (1) = y and

d γ (t)  
e = Λg η (t)eγ (t) .
dt

Applying Φ̂ to both sides of this equation, and using the definition of an L∞ -

morphism, we obtain
d  γ (t)   
Φ̂ e = Λh ◦ Φ̂ η (t)eγ (t) ,
dt
which gives, in view of both items of Proposition 13.40,
d Ω̃Φ (γ (t))  
e = Λh χ (t)eΩ̃Φ (γ (t)) , (13.62)
dt
where
1
χ (t) := ∑∗ ( − 1)! Φ (η (t) γ −1 (t)) . (13.63)
∈N

Now Ω̃Φ (γ (t)) is a polynomial path in ν hν1 and χ (t) is a polynomial path in ν hν0 .
Equation (13.62) says that Ω̃Φ (x) = Ω̃Φ (γ (0)) and Ω̃Φ (y) = Ω̃Φ (γ (1)) are path
equivalent, hence gauge equivalent (by Proposition 13.31), which is the content
of (2).
According to Remark 13.32, if x ∈ ν gν1 and ξ ∈ ν k gν0 , then the polynomial
path η (t) which is used in establishing the path equivalence between x and ξ  x
can be chosen in ν k gν0 , namely we can choose η (t) := −ξ . Then the polynomial
path χ (t), given by (13.63), belongs to ν k hν0 so that, again by Remark 13.32, the
gauge equivalence between Ω̃Φ (ξ  x) and Ω̃Φ (x) can be realized by an element
s ∈ ν k hν0 . Since χ (t) = Φ1 (η (t)) (mod ν k+1 ), we have s = Φ1 (ξ ) (mod ν k+1 ),
which proves (3).
392 13 Deformation Quantization

13.4.2 Surjectivity of ΩΦ

We show in this section that if Φ is an L∞ -quasi-isomorphism, then ΩΦ is surjective,

which is part of the statement of Theorem 13.39. We will use the following propo-
sition, which describes how the operators Obs behave under an L∞ -morphism (see
Section 13.3.3 for the definition of Obs and for its relation to the Maurer–Cartan
equation).
Proposition 13.42. Suppose that Φ is an L∞ -morphism between two differential
graded Lie algebras (g, [· , ·]g , Dg ) and (h, [· , ·]h , Dh ). Let x ∈ ν gν1 and let  ∈ N∗ . If
x ∈ MC (g), then Ω̃Φ (x) ∈ MC (h). Moreover, the following equality holds:
 
Φ1 (Obs+1 (x)) = Obs+1 Ω̃Φ (x) . (13.64)

Proof. As before, we denote by Λg and Λh the total differentials on S• g and S• h

respectively. For x ∈ ν gν1 , we have according to (13.61) that
 
Λh eΩ̃Φ (x) = Φ (Λg (ex )) . (13.65)

Both sides of this equation can be computed from (13.35), namely

   1

Ω̃Φ (x)
Λh e = Dh (Ω̃Φ (x)) + Ω̃Φ (x), Ω̃Φ (x) h eΩ̃Φ (x) , (13.66)
2
  
  1
Φ (Λg (e )) = Φ
x
Dg (x) + [x, x]g ex . (13.67)
2

Let  ∈ N∗ and suppose that x ∈ MC (g). Then Dg (x) + 12 [x, x]g = 0 (mod ν +1 ),
hence the right-hand side of equation (13.67) is zero, modulo ν +1 . In view
of (13.65), the right-hand side of (13.66) is also zero, modulo ν +1 , so that
1
Dh (Ω̃Φ (x)) + Ω̃Φ (x), Ω̃Φ (x) h
=0 (mod ν +1 ) .
2
This shows that Ω̃Φ (x) ∈ MC (h), which proves the first part of the proposition.
For x ∈ MC (g), the first part of the proof and (13.66) imply that
    
Λh Φ  (ex ) = Λh (eΩ̃Φ (x) ) = Obs+1 (Ω̃Φ (x)) . (13.68)
+1 +1

Moreover, letting ω := Dg (x) + 12 [x, x]g ∈ ν +1 gν2 we obtain, using (13.67) and
Proposition 13.40,
 
  1
Λh Φ x  (ω e ) = ∑
 (e ) = Φ x
Φk (ω x ) eΩ̃Φ (x) .
k−1
(13.69)
k∈N ∗ (k − 1)!

Therefore,
13.4 L∞ -Morphisms of Differential Graded Lie Algebras 393
  
 (ex )
Λh Φ = (Φ1 (ω ))+1 = Φ1 (ω+1 ) = Φ1 (Obs+1 (x)) ,
+1

which leads, combined with (13.68), to Eq. (13.64).

We now prove that if Φ is an L∞ -quasi-isomorphism between two pointed differen-
tial graded Lie algebras (g, [· , ·]g , zg ) and (h, [· , ·]h , zh ), then ΩΦ is surjective, which
is part of the statement of Theorem 13.39.
Proof. First, let us point out a fact which will be used several times in this proof. Let
(g, [· , ·], D) be a differential graded Lie algebra, let  ∈ N∗ and let x = ∑k∈N∗ xk ν k
and x = ∑k∈N∗ xk ν k . If x = x (mod ν k ), then

Obsk (x) − Obsk (x ) = D(xk − xk ) , (13.70)

in particular Obsk (x) − Obsk (x ) is a coboundary. This fact follows immediately
from the definition (13.33) of the operator Obsk .
Suppose that y ∈ MC(h) (in particular, all non-zero coefficients yk of y belong to
h1 ). We show that there exist formal power series x ∈ ν gν1 and ξ ∈ ν hν0 , such that
y = ξ  Ω̃Φ (x). These series will be constructed term by term. Thus, we assume that
we have constructed the first k terms x1 , . . . , xk of x and the first k terms ξ1 , . . . , ξk
of ξ , such that
(ak ) xk := x1 ν + · · · + xk ν k ∈ MCk (g);
(bk ) ξ k := ξ1 ν + · · · + ξk ν k and xk satisfy the following equation:

y = ξ k  Ω̃Φ (xk ) (mod ν k+1 ) .

We show how to construct an extra term xk+1 for x and ξk+1 for ξ , such that (ak+1 )
and (bk+1 ) are satisfied. Notice that (a0 ) and (b0 ) are trivially satisfied, since by
construction x0 = 0 and ξ 0 = 0, while the constant term of y is zero.

First we construct xk+1 ∈ g1 such that xk+1 := xk + xk+1 ν k+1 ∈ MCk+1 (g). In
view of Proposition 13.42, (bk ) and (13.70), there exists u ∈ h1 such that
 
Φ1 (Obsk+1 (xk )) = Obsk+1 (Ω̃Φ (xk )) = Obsk+1 (−ξ k )  y + D(u) = D(u) ,

where we used in the last step that y, and hence (−ξ k )  y, belongs to MC(h). Since
xk ∈ MCk (g), Proposition 13.24 implies that Obsk+1 (xk ) is a cocycle, so it defines
a cohomology class in HD2 zg (g); by the above computation, the image under Φ1 of
this cohomology class is trivial. Now Φ12 : HD2 zg (g) → HD2 z (h) is injective, since Φ
h
is an L∞ -quasi-isomorphism, hence the cohomology class of Obsk+1 (xk ) is trivial,

i.e., there exists xk+1 
∈ g1 satisfying Obsk+1 (xk ) = xk+1 , zg g . Letting xk+1 := xk +

xk+1 ν k+1 , we have that

Obsi (xk+1 ) = Obsi (xk ) = 0, for all 1  i  k ,

while (13.70) implies that

394 13 Deformation Quantization

Obsk+1 (xk+1 ) = Obsk+1 (xk ) + Dzg (xk+1

 
) = Obsk+1 (xk ) + zg , xk+1 g
=0,

so that xk+1 ∈ MCk+1 (g), i.e., xk+1 satisfies property (ak+1 ).

We now prove that there exist a cocycle xk+1  ∈ g1 and an element ξk+1 ∈ h0 , such
 
that xk+1 := xk+1 + xk+1 ν k+1  
= xk + (xk+1 + xk+1 )ν k+1 and ξ k+1 := ξ k + ξk+1 ν k+1
satisfy property (bk+1 ); property (ak+1 ) remains satisfied when we replace xk+1

by xk+1 because xk+1 is a cocycle.
Property (bk ) and the fact that xk+1 = xk (mod ν k+1 ) imply that

y = ξ k  Ω̃Φ (xk+1 ) (mod ν k+1 ) .

Writing ξ k  Ω̃Φ (xk+1 ) = ∑k∈N∗ Xk ν k , with Xk ∈ h1 for all k, we have according

to (13.70) that
 
Obsk+1 (y) − Obsk+1 ξ k  Ω̃Φ (xk+1 ) = Dzh (yk+1 − Xk+1 ) . (13.71)

As xk+1 ∈ MCk+1 (g), Proposition 13.42 implies that Ω̃Φ (xk+1 ) ∈ MCk+1 (h), hence
 
ξ k  Ω̃Φ (xk+1 ) ∈ MCk+1 (h), so that Obsk+1 ξ k  Ω̃Φ (xk+1 ) = 0. Moreover, since
y ∈ MC(h), we also have Obsk+1 (y) = 0, so that (13.71) gives Dzh (yk+1 −Xk+1 ) = 0.

Since Φ11 : HD1 zg (g) → HD1 z (h) is surjective, there exists a cocycle xk+1 ∈ g1 and an
h
element ξk+1 ∈ h0 , such that
 
yk+1 − Xk+1 = Φ1 (xk+1 ) − Dzh (ξk+1 ) = Φ1 (xk+1 ) + ξk+1 , zh h
,

in other words, we obtain

   
y − ξ k  Ω̃Φ (xk+1 ) = Φ1 (xk+1

) + ξk+1 , zh h ν k+1 (mod ν k+2 ) . (13.72)

With the help of (13.56) and (13.72), we compute modulo ν k+2 :

ξ k+1  Ω̃Φ (xk+1 ) = (ξ k + ξk+1 ν k+1 )  Ω̃Φ (xk+1 + xk+1

 ν k+1 )
 
= ξ k  Ω̃Φ (xk+1 ) + Φ1 (xk+1
 ) + ξ
k+1 h h ν
, z k+1 = y .

Thus, xk+1 and ξ k+1 satisfy (ak+1 ) and (bk+1 ).

Remark 13.43. Notice that the fact that Φ1• is an isomorphism is not fully used in
the proof. In fact, we only used that Φ12 : HD2 zg (g) → HD2 z (h) is injective and that
h
Φ11 : HD1 zg (g) → HD1 z (h) is surjective.
h
13.4 L∞ -Morphisms of Differential Graded Lie Algebras 395

13.4.3 Injectivity of ΩΦ

In this section we prove that ΩΦ is injective, which is also part of the statement of
Theorem 13.39. To do this, we use the following lemma. Recall that, for a formal
power series x (in ν ), we denote by xk its coefficient of ν k .
Lemma 13.44. Let (g, [· , ·], z) be a pointed differential graded Lie algebra. Suppose
that x ∈ ν gν1 and that ξ ∈ ν k gν0 .
(1) (ξ  x − x)k = −Dz (ξk );
(2) If s ∈ g0 is a cocycle, then (ξ + sν k )  x = ξ  x (mod ν k+1 );
(3) If ξ  x = x (mod ν k+1 ), then ξk is a cocycle;
(4) If ξk is a coboundary and x is solution of the Maurer–Cartan equation, then
there exists ξ  ∈ ν k+1 gν0 , such that ξ  x = ξ   x.

Proof. Items (1)–(3) follow from the fact that, since ξ ∈ ν k gν0 , the term in ν  in
ξ  x is x when 1   < k and is xk + [ξk , z] = xk − Dz (ξk ) when  = k.
For item (4), assume that ξk is a coboundary, ξk = Dz (η ) for some η ∈ g−1 , and
that x ∈ MC(g), so that [z + x, z + x] = 0. For w ∈ ν gν−1 , the graded Jacobi identity
implies that ad[w,z+x] (z + x) = − 12 [[z + x, z + x] , w] = 0, so that

1
[w, z + x]  x = x + ∑∗ k! adk[w,z+x] (z + x) = x .
k∈N

In particular, if w ∈ ν k gν−1 , then

ξ  x = ξ  ([w, z + x]  x) = CH(ξ , [w, z + x])  x , (13.73)

and it follows from (13.39) that (CH(ξ , [w, z + x]))k = ξk + [wk , z] = Dz (η − wk ).

Therefore, taking w := ην k , we have that ξ  := CH(ξ , [w, z + x]) ∈ ν k+1 gν0 , while
ξ   x = ξ  x, according to (13.73). This proves (4).

We can now prove that if Φ is an L∞ -quasi-isomorphism between two pointed dif-

ferential graded Lie algebras (g, [· , ·]g , zg ) and (h, [· , ·]h , zh ), then ΩΦ is injective,
which is part of the statement of Theorem 13.39.
Proof. Suppose that x, y ∈ MC(g) have the same image under ΩΦ , which means
that Ω̃Φ (x) and Ω̃Φ (y) are gauge equivalent. We need to show that x and y are
gauge equivalent. To do this, we construct three sequences: a sequence (x(k) )k∈N
in ν gν1 with x(0) = x, a sequence (ξ (k) )k∈N in g0 and a sequence (u(k) )k∈N in hν0 ,
with u(k) ∈ ν k+1 hν0 for all k. These sequences will be constructed such that their
elements satisfy, for all k ∈ N∗ , the following four properties:
(Ak ) x(k) = (ξ (k) ν k )  x(k−1) ;
(Bk ) x(k) = y (mod ν k+1 );
(k−1)
(Ck ) uk + Φ1 (ξ (k) ) is a coboundary;
(Dk ) Ω̃Φ (x(k) ) = u(k)  Ω̃Φ (y).
396 13 Deformation Quantization

We first show that the existence of these sequences satisfying the properties (Ak )
and (Bk ) proves that x and y are gauge equivalent; properties (Ck ) and (Dk ) will
turn out to be useful for constructing the sequences satisfying the former properties.
Consider the formal power series Ξ := limk→∞ Ξ (k) , where the polynomials Ξ (k) are
recursively defined by Ξ (k) := CH(ξ (k) ν k , Ξk−1 ) for k ∈ N∗ and Ξ0 := 0. Notice that
the limit is well-defined, because Ξ (k) = Ξ (k−1) (mod ν k ) for k ∈ N∗ , and that it
belongs to ν gν0 . In view of (A1 ),. . . ,(Ak ), (Bk ), and (13.40),

Ξ (k)  x = x(k) = y (mod ν k+1 ) ,

for all k ∈ N, so that Ξ  x = y.

We proceed to the construction of the three sequences. First, we set ξ (0) := 0 and
we choose for u(0) an element in ν hν0 for which u(0)  Ω̃Φ (y) = Ω̃Φ (x), which can
be done because, by assumption, Ω̃Φ (x) and Ω̃Φ (y) are gauge equivalent. In order
(−1)
to satisfy (A0 ), (B0 ), (C0 ) and (D0 ), we define x(−1) := x(0) = x and u0 := 0. Let
us assume that x(0) , . . . , x(k) and u(0) , . . . , u(k) and ξ (0) , . . . , ξ (k) are constructed, and
satisfy the above conditions (A ), (B ), (C ) and (D ), for  = 0, . . . , k.
We first construct x(k+1) and ξ (k+1) satisfying (Ak+1 ) and (Bk+1 ). In view of (Bk ),
x and y agree modulo ν k+1 . As a first consequence, since both are solutions of the
(k)
(k)
Maurer–Cartan equation, (13.70) implies that D(xk+1 − yk+1 ) = 0, i.e., (x(k) − y)k+1
is a cocycle. As a second consequence, Φ (y ) = Φ ((x(k) ) ) (mod ν k+2 ) for   2,


so that    
(k)
Ω̃Φ (x(k) ) − Ω̃Φ (y) = Φ1 xk+1 − yk+1 . (13.74)
k+1
The left-hand side of (13.74) is a coboundary, in view of (Dk ) and item (1) of
Lemma 13.44. Since Φ11 : HD1 zg (g) → HD1 z (h) is injective, (x(k) − y)k+1 is also
h
(k)
a coboundary, i.e., there exists ξ (k+1) ∈ g0 such that Dzg (ξ (k+1) ) = xk+1 − yk+1 .
 
(k)
Now ((ξ (k+1) ν k+1 )  x(k) )k+1 = xk+1 + ξ (k+1) , zg , so that (ξ (k+1) ν k+1 )  x(k) = y
(mod ν k+2 ). Defining x(k+1) := (ξ (k+1) ν k+1 )  x(k) , it follows that x(k+1) and ξ (k+1)
satisfy (Ak+1 ) and (Bk+1 ).
(k)
The constructed element ξ (k+1) does not satisfy (Ck+1 ), namely uk+1 + Φ1 (ξ (k+1) )
need not be a coboundary, but it is a cocycle. In fact, in view of (Ak+1 ) and (Bk+1 ),

(ξ (k+1) ν k+1 )  x(k) = y (mod ν k+2 ) ,

so that, in view of (Dk ), we have the following equalities, modulo ν k+2 :

 
u(k)  Ω̃Φ (ξ (k+1) ν k+1 )  x(k) = u(k)  Ω̃Φ (y) = Ω̃Φ (x(k) ) . (13.75)

Since Φ is an L∞ -morphism, item (3) of Proposition 13.41 implies that

Ω̃Φ (ξ (k+1) ν k+1  x(k) ) = Φ1 (ξ (k+1) ν k+1 )  Ω̃Φ (x(k) ) (mod ν k+2 ) ,
13.5 Kontsevich’s Formality Theorem and Its Consequences 397

so that, still modulo ν k+2, the left-hand side of (13.75) is CH(u(k) , Φ1 (ξ (k+1) ν k+1 ))
Ω̃Φ (x(k) ). According to (13.39),
(k)
(CH(u(k) , Φ1 (ξ (k+1) ν k+1 )))k+1 = uk+1 + Φ1 (ξ (k+1) ) ,

(k)
so that we can apply item in (3) of Lemma 13.44 to (13.75), to conclude that uk+1 +
Φ1 (ξ (k+1) ) is a cocycle.
We now show how ξ (k+1) can be modified, so that (Ck+1 ) is satisfied, i.e., so that
(k)
uk+1 + Φ1 (ξ (k+1) ) is a coboundary. We will do this without much altering x(k+1) .
(k)
Since uk+1 + Φ1 (ξ (k+1) ) is a cocycle it defines a cohomology class in HD0 z (h);
h
since Φ10 : HD0 zg (g) → HD0 z (h) is surjective, this cohomology class is the image of
h
a cocycle κ ∈ g0 . Upon replacing ξ (k+1) by ξ (k+1) − κ , the new x(k+1) , defined by
(Ak+1 ), still satisfies (Bk+1 ), in view of item (2) of Lemma 13.44, i.e., removing a
cocycle from ξ (k+1) does not alter x(k+1) modulo ν k+1 . However, the cohomology
(k)
class of uk+1 + Φ1 (ξ (k+1) ) is now trivial, i.e., the latter element is a coboundary.
The last step of the construction is the construction of uk+1 , such that (Dk+1 ) is
satisfied. Item (3) in Proposition 13.41 yields the existence of s ∈ ν k+1 hν0 such that
Ω̃Φ (ξ (k+1) ν k+1  x(k) ) = s  Ω̃Φ (x(k) ); by the same proposition, s = Φ1 (ξ (k+1) )
(mod ν k+2 ). Using (Ak+1 ) and (Dk ), it follows that
 
Ω̃Φ (x(k+1) ) = s  Ω̃Φ (x(k) ) = s  u(k)  Ω̃Φ (y) = CH(s, u(k) )  Ω̃Φ (y) .

(k)
In view of (13.39), (CH(s, u(k) ))k+1 = Φ1 (ξ (k+1) ) + uk+1 , which is a coboundary,
according to (Ck+1 ). Item (4) in Lemma 13.44 implies that there exists u(k+1) in
ν k+2 hν0 such that

u(k+1)  Ω̃Φ (y) = CH(s, u(k) )  Ω̃Φ (y) = Ω̃Φ (x(k+1) ) ,

which is (Dk+1 ).

Remark 13.45. The fact that Φ1• is an isomorphism is not fully used in the proof of
the injectivity of ΩΦ . Indeed, we only used that Φ11 : HD1 zg (g) → HD1 z (h) is injec-
h
tive and that Φ10 : HD0 zg (g) → HD0 z (h) is surjective (see Remark 13.43 for a similar
h
remark concerning the proof of the surjectivity of ΩΦ ).

13.5 Kontsevich’s Formality Theorem and Its Consequences

The present section is devoted to a fundamental result by Kontsevich, whose state-

ment and implications are accessible for the readers who have become sufficiently
familiar with the concepts developed in the previous sections of this chapter. In-
398 13 Deformation Quantization

stead of giving a full proof of Kontsevich’s theorem, which is beyond the scope of
this book, we present some of the main ingredients, and we concentrate on a few of
its consequences, which have deep implications to both Poisson geometry and Lie
theory. The most popular consequence can be stated in a few words: the algebra of
smooth functions on a real Poisson manifold admits a formal deformation quantiza-
tion. This result will be proved in Theorem 13.51, as a consequence of Kontsevich’s
formality theorem (Theorem 13.46).

13.5.1 Kontsevich’s Formality Theorem

Let M be a real manifold and denote by μ the usual product on C∞ (M). Recall that
there are two pointed differential graded Lie algebras, which are naturally associated
to M:
• (X• (M), [· , ·]S , 0) is the graded vector space of multivector fields on M, with
shifted degree, equipped with the Schouten bracket and with the zero differen-
tial (see Example 13.18);
• ( HC•diff (M), [· , ·]G , μ ) is the graded vector space of differential Hochschild
cochains of C∞ (M), with shifted degree, equipped with the Gerstenhaber bracket
and with the Hochschild differential associated to μ (see Example 13.20).
First, we show how these differential graded Lie algebras are related. Let P be a
p-vector field on M, where p  1. A p-linear map φ (P) : (C∞ (M)) p → C∞ (M) is
naturally associated to P by setting, for all F1 , . . . , Fp ∈ C∞ (M),

1
φ (P)(F1 , . . . , Fp ) := P[F1 , . . . , Fp ] . (13.76)
p!
It leads for every p to a graded linear map of degree 0,

φ : X p (M) → HCdiff
p
(M) , (13.77)

which is injective, because P can be recovered from φ (P). We have, according to

(13.7), for all F0 , . . . , Fp ∈ C∞ (M),

δμp (p! φ (P))(F0 , . . . , Fp ) = F0 P[F1 , . . . , Fp ] + (−1) p+1 P[F0 , . . . , Fp−1 ]Fp

p
+ ∑ (−1)i P[F0 , . . . , Fi−1 Fi , . . . , Fp ]
i=1
= 0,

where we used in the last equality that P is a derivation in each of its arguments.
It follows that the map φ takes values in the Hochschild cocycles of (C∞ (M), μ ).
In particular, φ induces a linear map H • φ between the cohomologies of the pointed
differential graded Lie algebras (X• (M), [· , ·]S , 0) and ( HC•diff (M), [· , ·]G , μ ). Ac-
cording to the Hochschild–Kostant–Rosenberg theorem (see [131]), H • φ is an iso-
13.5 Kontsevich’s Formality Theorem and Its Consequences 399

morphism of differential graded Lie algebras. The map φ itself is however not a
morphism of (differential) graded Lie algebras, since [P, Q]S = [P, Q]G for general
multivector fields P, Q as one easily checks from the definitions.
The following result, due to Kontsevich, states that, even if φ is not a mor-
phism of differential graded Lie algebras, it extends to an L∞ -morphism Φ between
(X• (M), [· , ·]S , 0) and ( HC•diff (M), [· , ·]G , μ ); since the induced map in cohomology
H • Φ1 is H • φ , which is according to the Hochschild–Kostant–Rosenberg theorem
an isomorphism, Φ is an L∞ -quasi-isomorphism (see Definition 13.37).
Theorem 13.46 (Kontsevich’s formality theorem). Let M be a real manifold and
denote by μ the usual product on C∞ (M). There exists an L∞ -quasi-isomorphism Φ
from (X• (M), [· , ·]S , 0) to ( HC•diff (M), [· , ·]G , μ ), such that the map Φ1 : X• (M) →
HC•diff (M) is the natural inclusion map φ , defined in (13.77).
For a proof of this theorem, which is quite involved and is beyond the scope of
this book, we refer to [38, 39, 107, 189] and [14]. We will give in the following
section some more details about the construction of the L∞ -quasi-isomorphism in
the case of M = Rd . But first, we say a few words about the origin of the name
of the theorem. Recall from Example 13.21 that the cohomology of a differen-
tial graded Lie algebra is itself also a (pointed) differential graded Lie algebra. In
the case of (X• (M), [· , ·]S , 0), the differential is trivial hence the latter differential
graded Lie algebra and its cohomology are isomorphic differential graded Lie al-
gebras. According to the Hochschild–Kostant–Rosenberg theorem, cited above, this
cohomology and the cohomology of ( HC•diff (M), [· , ·]G , μ ) are isomorphic differen-
tial graded Lie algebras. Thus, Kontsevich’s theorem says that there exists an L∞ -
quasi-isomorphism between ( HC•diff (M), [· , ·]G , μ ) and its cohomology. A differen-
tial graded Lie algebra which admits an L∞ -quasi-isomorphism to its cohomology
is called formal, a terminology which is borrowed from topology. Thus, Kontse-
vich’s theorem says that ( HC•diff (M), [· , ·]G , μ ) is formal, which explains the name
“formality theorem”.

13.5.2 Kontsevich’s Formality Theorem for Rd

Without going into technical details, we present in this section Kontsevich’s ex-
plicit construction of the L∞ -quasi-isomorphism Φ between (X• (M), [· , ·]S , 0) and
( HC•diff (M), [· , ·]G , μ ) in the particular case where M = Rd . In order to do this, we
first need to explain how multi-differential operators are naturally coded by graphs,
how weights are associated to these graphs and how they all sum up to yield a for-
mula for the L∞ -quasi-isomorphism Φ .
400 13 Deformation Quantization

13.5.2.1 A – Kontsevich’s Graphs

We describe in this subsection the graphs which are used in Kontsevich’s construc-
tion. Recall that an oriented graph is a pair (V, A), where A is a subset of V × V .
The elements of V are called vertices and the elements of A arrows. For an arrow
a = (x, y) ∈ A, we call y its head, denoted h(a) and x its tail, denoted t(a). We will
assume that Γ contains no loops, which means that if (x, y) ∈ A, then x = y. For given
integers k,  ∈ N, the vertex set of the graphs which are considered is the set Vk, ,
defined by
Vk, := {1, . . . , k} ∪ {1, . . . , } .
For reasons which will be clear later, it is demanded that these integers satisfy 2k +
 − 2  0, that is the cases (k, ) = (0, 0) and (k, ) = (1, 0) are excluded. The k
vertices 1, . . . , k of Vk, are called vertices of the first type, while the  vertices 1, . . . , 
of Vk, are called vertices of the second type. Another piece of data added to the
graphs which are considered here is an ordering of each of the sets of arrows with a
common tail, i.e., of each of the sets

Star(s) := {a ∈ A | t(a) = s} ,

for s ∈ {1, . . . , k}. Denoting the number of elements in Star(s) by ps , such an order-
ing amounts to a numbering σs of the elements of Star(s) by the integers 1, . . . , ps ;
using this numbering we write Star(s) as Star(s) = {σs1 , . . . , σsps } .
Definition 13.47. Let k,  ∈ N, such that 2k +  − 2  0. A triplet Γ = (Vk, , A, σ ) is
said to be a Kontsevich graph if
(1) The pair (Vk, , A) is an oriented graph without loops, called the underlying
graph of Γ ;
(2) For every arrow a ∈ A, its tail t(a) is a vertex of the first type;
(3) For every s ∈ {1, . . . , k}, σs is an ordering of Star(s).
The set of Kontsevich graphs Γ = (Vk, , A, σ ) is denoted by Gk, . The underlying
graph of some element G3,4 is depicted in Fig. 13.1.

13.5.2.2 B – The Weight of a Kontsevich Graph

Let k,  ∈ N, such that 2k +  − 2  0. To a Kontsevich graph Γ ∈ Gk, we associate

a number ϖΓ , called its weight. The definition involves angles in the Poincaré half-
plane and integration over a configuration space. We recall first that the Poincaré
half-plane is the open subset
 
H := (x, y) ∈ R2 | y > 0 ,

equipped with the metric

13.5 Kontsevich’s Formality Theorem and Its Consequences 401

Fig. 13.1 The elements in Gk, are triplets (V, A, σ ), where the set of vertices consists of k vertices
of the first type, denoted 1, . . . , k and  vertices of the second type, denoted 1, . . . , . All arrows start
from vertices of the first type and are ordered by σ . Depicted here is the graph (V, A) that underlies
some element of G3,4 . For reasons that will become clear later, the graph is drawn in the upper
half-plane, with the vertices of the second type on its boundary.

Fig. 13.2 For p and q, points in the Poincaré half-plane H , θ (p, q) denotes the angle θ between
the vertical geodesic through p and the geodesic through p and q.

(dx)2 + (dy)2
(ds)2 = ,
y2
whose geodesics are the vertical half-lines, starting from the horizontal axis (the
line y = 0) and half-circles, with center on the horizontal axis. For p, q ∈ H :=
H ∪ {y = 0}, with p = q, we denote by θ (p, q) the angle (at p) between the verti-
cal geodesic through p and the geodesic through p and q (see Fig. 13.2). The angle
function θ leads, for each arrow a of Γ , to an analytic function θa on a configu-
ration space of points, whose construction we briefly recall (see the appendix by
A. Bruguières in [39] for details). With k,  as above, consider Confk, , the space
of (k + )-tuples of distinct points, the first k points belonging to H and the  re-
maining points belonging to the boundary of H . We write a point of Confk, as a
(k + )-tuple (z1 , . . . , zk , z1 , . . . , z ), since we think of such a point as corresponding
to the vertices of a graph of Gk, , drawn in H . The group G of affine transforma-
tions of C of the form z → λ z + μ , with λ ∈ R∗+ and μ ∈ R, acts in a natural way
on Confk, . This action is free and the quotient manifold
402 13 Deformation Quantization

Fig. 13.3 The graph Γ belongs to G1, . It has one vertex of the first type and  vertices of the
second type. Its weight is 1/(!)2 .

Ck, := Confk, /G

has dimension 2k +  − 2; the class of a point (z1 , . . . , zk , z1 , . . . , z ) ∈ Confk, in Ck,

is denoted by [(z1 , . . . , zk , z1 , . . . , z )]. The manifold Ck, admits a natural compact-
ification as a manifold with corners; the connected component which contains the
points [(z1 , . . . , zk , z1 , . . . , z )] for which z1 < z2 < · · · < z is denoted by Ck, . To an
arrow a = (t(a), h(a)) of Γ , we can associate an analytic function on Ck, , by defin-
ing θa ([(z1 , . . . , zk , z1 , . . . , z )]) := θ (zt(a) , zh(a) ). When Γ has p arrows, we obtain in
this way p functions, and hence a p-form on the configuration space, which leads
upon integrating to the weight associated to Γ = (Vk, , A, σ ), namely

k ' ( dθa
1
ϖΓ := ∏ .
s=1 (# Star(s))! a∈A
2π
Ck,

We do not discuss here the technical issues, related to the fact that Ck, has corners
and concerning the choice of ordering in the above wedge product. We simply point
out that, for dimensional reasons, ϖΓ = 0 whenever the number of arrows in Γ is
different from the dimension 2k +  − 2 of Ck, .

Example 13.48. Up to the ordering of its arrows, there is a single graph Γ in G1,
which has  arrows. It is depicted in Fig. 13.3. The Kontsevich weight of Γ is given
by

' (
 '
1 1 1
ϖΓ = dθ(1,t) = dθ1 ∧ · · · ∧ dθ = .
! (2π ) t=1
! (2π ) (!)2
C1, 0θ1 <···<θ 2π
(13.78)
13.5 Kontsevich’s Formality Theorem and Its Consequences 403

Fig. 13.4 To each vertex of a graph one associates a multivector field on Rd (a 0-vector field, i.e.,
a function, when the vertex is of the second type). Depicted is the graph of Fig. 13.1, with the extra
data.

13.5.2.3 C – The Differential Operator Associated to a Kontsevich Graph

Let Γ = (Vk, , A, σ ) ∈ Gk, be a Kontsevich graph, with 2k +  − 2  0, as before. We

show how there is naturally associated to Γ (equipped with some extra data) a multi-
differential operator on Rd ; the natural coordinates on Rd are denoted by x1 , . . . , xd .
We add the following data to the graph:
(1) To each vertex s of the first type, a ps -vector field Qs on Rd ;
(2) To each vertex t of the second type, a smooth function Ft on Rd .
See Fig. 13.4, which is Fig. 13.1 with the extra data. Let γ : A → {1, . . . , d} be any
map. Using the above data and γ , we can associate to each vertex t of the second
type a function DΓγ,t (Ft ), namely
⎛ ⎞
⎜ ∂ ⎟
DΓγ,t (Ft ) := ⎜
⎝ ∏ ⎟ Ft .
∂ xγ (a) ⎠
a∈A
h(a)=t

Similarly, we associate a function DΓγ, s (Qs ) to each vertex s of the first type, by
setting
⎛ ⎞
⎜ ∂ ⎟   
DΓγ, s (Qs ) := ⎜
⎝ ∏ ⎟ Qs x 1 , . . . , x ps
∂ xγ (a) ⎠ γ ( σs ) γ ( σs ) ,
a∈A
h(a)=s

 
where we recall that σs1 , . . . , σsps = Star(s) is the set of arrows whose tail is s,
ordered by σs . When the above products are taken over the empty set, i.e., if t,
respectively s, is not the head of an arrow in A, then the derivatives reduce to a
zero-th order derivative, i.e., no derivatives are taken. Taking the product of the thus
404 13 Deformation Quantization

Fig. 13.5 The graph Γ with some extra data.

constructed k +  functions and summing over all maps γ : A → {1, . . . , d}, leads to
the function
k 
BΓ (Q1 , . . . , Qk )(F1 , . . . , F ) := ∑ ∏ DΓγ, s (Qs ) ∏ DΓγ,t (Ft ) .
γ :A→{1,...,d} s=1 t=1

The right-hand side of this equation is clearly a multi-differential operator in the

functions F1 , . . . , F . In summary, we have associated to each Kontsevich graph Γ =
(Vk, , A) and to each choice of multivector fields (of the right degree) an -differential
operator BΓ (Q1 , . . . , Qk ), whose action on functions is defined as above. If at least
one of the multivector fields Qs is not of the right degree (i.e., its degree is different
from ps , the number of arrows of Γ whose tail is s), we define BΓ (Q1 , . . . , Qk ) to be
equal to the zero -differential operator. Extending these definitions by linearity, we
get for each Kontsevich graph Γ = (Vk, , A) ∈ Gk, a linear map BΓ : Sk g → h, where
we recall that g and h are respectively the pointed differential graded Lie algebras
•
(X• (Rd ), [· , ·]S , 0) and ( HCdiff (Rd ), [· , ·]G , μ ).
Example 13.49. We write down BΓ , where Γ is the graph in G1, which we intro-
duced in Example 13.48. Let Q be an -vector field on Rd and let F1 , . . . , F be
functions on Rd . See Fig. 13.5, where these data have been added to the graph. We
label the arrows of the graph from 1 to , so that the arrow labeled i connects 1 to i ,
for all i = 1, . . . , . For a map γ : A = {1, . . . , } → {1, . . . , d} we write γ (i) as γi ;
then, summing over all such maps γ simply amounts to summing over all γ1 , . . . , γ ,
ranging from 1 to d. With these notations,

∂ F1 ∂ F2 ∂ F
BΓ (Q)(F1 , . . . , F ) = ∑ Q[xγ1 , . . . , xγ ]
∂ xγ1 ∂ xγ2
···
∂ xγ
1γ1 ,...,γ d
= Q[F1 , . . . , F ],

so that BΓ (Q) = Q.

13.5 Kontsevich’s Formality Theorem and Its Consequences 405

Example 13.50. Similarly, we write down BΓ , where Γ is the graph of Fig. 13.4.
The arrows of the graph are labeled 1, . . . , 7, starting with the arrows whose tail is
Q1 ∈ X2 (Rd ), then those whose tail is Q2 ∈ X3 (Rd ), finally the ones whose tail is
Q3 ∈ X2 (Rd ). It leads to the following formula:

∂ Q̃1 ∂ F1 ∂ 2 F2 ∂ 2 F3 ∂ F4
BΓ (Q1 , Q2 , Q3 )(F1 , . . . , F4 ) = ∑ Q̃2 Q̃3 ,
1γ1 ,...,γ7 d ∂ xγ6 ∂ xγ1 ∂ xγ2 ∂ xγ3 ∂ xγ4 ∂ xγ7 ∂ xγ5

where
Q̃1 := Q1 [xγ1 , xγ2 ], Q̃2 := Q2 [xγ3 , xγ4 , xγ5 ], Q̃3 := Q3 [xγ6 , xγ7 ] .

13.5.2.4 D – Kontsevich’s Formula for the L∞ -Quasi-Isomorphism

We have introduced in the previous subsections all the ingredients which are
used in Kontsevich’s explicit construction of the L∞ -quasi-isomorphism Φ between
•
(X• (Rd ), [· , ·]S , 0) and ( HCdiff (Rd ), [· , ·]G , μ ). As above, we denote these pointed
differential graded Lie algebras by g and h respectively, and we denote the natu-
ral coordinates on Rd by x1 , . . . , xd . Recall that Φ : S• g → h is completely deter-
mined by the family of maps (Φk )k∈N , where for each k ∈ N, Φk is the restric-
tion of Φ to Sk g. They are defined as follows: for all p1 , . . . , pk ∈ N and for all
Q1 Q2 . . . Qk ∈ g p1 −1 · · · g pk −1 , the -differential operator Φk (Q1 Q2 . . . Qk ) ∈ h−1 ,
where  := ∑ki=1 pi − 2k + 2, is defined by:

Φk (Q1 Q2 . . . Qk ) := ∑ ϖΓ BΓ (Q1 , . . . , Qk ) . (13.79)

Γ ∈Gk,

In this definition, the sum is taken over all Kontsevich graphs Γ = (Vk, , A) ∈ Gk,
and for each such graph Γ , ϖΓ and BΓ are the weight and the -differential operator
associated to Γ , as defined in Subsections 13.5.2.2 and 13.5.2.3.
A few comments about this definition are in order:
(1) As we have seen, ϖΓ = 0 when Γ is a graph in Gk, whose number of arrows
is different from 2k +  − 2. This explain why the sum in (13.79) is only taken
over Gk, , where  := ∑ki=1 pi − 2k + 2. Recall also that BΓ is zero when Γ has
a vertex i for which the number of arrows which start from i is not equal to pi .
(2) With respect to the grading which we introduced on S• g, the monomial
Q1 Q2 . . . Qk in (13.79) is of degree ∑ki=1 pi − 2k, while BΓ (Q1 Q2 . . . Qk ) is of
weight  − 2, with  as above. It follows that Φ is a graded map (of degree 0).
(3) Each Φk is well-defined: the order of Q1 , . . . , Qk in the right-hand side of
(13.79) is irrelevant, since both ϖΓ and BΓ are independent of the ordering
of the vertices of the first type.
(4) The above formula for Φk is particularly simple for k = 1. In fact, there are !
graphs Γ = (V1, , A, σ ) in G1, with  arrows, differing only by their ordering
(as given by σ ). As we have seen in Example 13.48, the weight of these graphs
is 1/(!)2 ; according to Example 13.49, BΓ (Q) = Q. It follows that
406 13 Deformation Quantization

1
Φ1 (Q) = ∑ ϖΓ BΓ (Q) =
!
Q.
Γ ∈G1,

In view of (13.76), it follows that Φ1 = φ , as stated in Kontsevich’s formality

theorem (Theorem 13.46).
The main result is that Φ is an L∞ -quasi-isomorphism between g and h. As a con-
sequence, Kontsevich’s formality theorem (Theorem 13.46) holds for Rd and the
L∞ -quasi-isomorphism which proves this theorem can be constructed quite explic-
itly. For a proof that Φ is an L∞ -quasi-isomorphism we refer to [39, 107].

13.5.3 A Few Consequences of Kontsevich’s Formality Theorem

The following theorem is a first important corollary of Kontsevich’s formality theo-

rem.

Theorem 13.51. Let M be a real manifold and denote by μ the usual product
on C∞ (M).
(1) Every Poisson structure π on M admits a star product, i.e., there exists a star
product μ = μ + ∑i∈N∗ μi ν i on C∞ (M), with μ1 = π2 ;
(2) There is a one-to-one correspondence between the set of equivalence classes
of formal deformations of the trivial Poisson structure on M and the set of
equivalence classes of star products on C∞ (M);
(3) The correspondence in (2) is natural in the following sense: if the equivalence
class of a formal deformation π = ∑i∈N∗ πi ν i of the zero Poisson structure
on M and the equivalence class of a star product μ = μ + ∑i∈N∗ μi ν i on M
correspond under (2), then π1 is the skew-symmetric part of 2μ1 .

Proof. Let π be a Poisson structure on M. It is an element of degree one of

(X• (M), [· , ·]S , 0) and since it satisfies the Jacobi identity, νπ is a solution of
the Maurer–Cartan equation associated to (X• (M), [· , ·]S , 0). According to Theo-
rem 13.46, there exists an L∞ -quasi-isomorphism Φ between the differential graded
Lie algebras (X• (M), [· , ·]S , 0) and ( HC•diff (M), [· , ·]G , μ ), with Φ1 = φ . According
to the first item of Proposition 13.41, Ω̃Φ (νπ ) is a solution of the Maurer–Cartan
equation associated to ( HC•diff (M), [· , ·]G , μ ), so that μ + Ω̃Φ (νπ ) is a star product
on C∞ (M) (Proposition 13.33). Moreover, by definition of Ω̃Φ (see (13.56)), we can
write
1 νk
Ω̃Φ (νπ ) = ∑ k! Φk ((νπ )k ) = ∑ Φk (π k ) , (13.80)
k1 k1 k!

so that the coefficient in ν of the star product μ + Ω̃Φ (νπ ) is Φ1 (π ) = φ (π ) = π2 ,

according to Theorem 13.46. This shows (1).
13.5 Kontsevich’s Formality Theorem and Its Consequences 407

Theorems 13.46 and 13.39, combined, show that there is a one-to-one corre-
spondence between the set of gauge equivalence classes of solutions of the Maurer–
Cartan equation associated to the differential graded Lie algebra (X• (M), [· , ·]S , 0)
and the set of gauge equivalence classes of solutions of the Maurer–Cartan equa-
tion associated to ( HC•diff (M), [· , ·]G , μ ). According to Proposition 13.33, the first
set is the set of equivalence classes of deformations of the trivial Poisson structure,
while the second set, after a translation by μ , is the set of equivalence classes of star
products on C∞ (M). This yields the one-to-one correspondence in (2).
Let π = ∑k∈N∗ πk ν k be a formal deformation of the trivial Poisson structure on
M and consider μ := μ + Ω̃Φ (π ), which is a star product on C∞ (M). Recall that
the equivalence class of π and the equivalence class of μ are related by the one-to-
one correspondence in (2). We have, as in (13.80), that μ = μ + π21 ν (mod ν 2 ).
Let μ = μ + ∑k∈N∗ μk ν k be an arbitrary star product on C∞ (M) which is equivalent
to μ , which means that μ − μ and μ − μ are gauge equivalent. Then there exists
ξ = ∑k∈N∗ ξk ν k in ν HC0diff (M)ν such that μ = eadξ (μ ), so that
π 
μ = μ +
1
+ [ξ1 , μ ]G ν (mod ν 2 ) .
2
The bilinear map [ξ1 , μ ]G ∈ HC1diff (M) is symmetric, because

[ξ1 , μ ]G (F, G) = ξ1 (μ (F, G)) − μ (ξ1 (F), G) − μ (F, ξ1 (G)) ,

for all F, G ∈ C∞ (M). This shows that π1 is the skew-symmetric part of 2μ1 , which
is the content of (3).
If (M, π ) is a real Poisson manifold, one can also consider the formal deformations
of π rather than of the trivial Poisson structure on M. We show in the following
theorem how these deformations are related to star products on C∞ (M)M.
Theorem 13.52. Let (M, π ) be a real Poisson manifold and let μ denote the usual
product on C∞ (M). There is a one-to-one correspondence between the set of equiva-
lence classes of formal deformations of the Poisson structure π and the set of equiv-
alence classes of star products μ = μ + ∑k∈N∗ μk ν k on C∞ (M), for which π is the
skew-symmetric part of 2μ1 .
Proof. Let us denote by D0 the set of equivalence classes of formal deformations
of the trivial Poisson structure on M and by Dπ the set of equivalence classes of
formal deformations of the Poisson structure π on M. Let us also denote by Sμ the
set of equivalence classes of star products on C∞ (M) and finally by Sμ ,π the set of
equivalence classes of star products μ = μ + ∑k∈N∗ μk ν k on C∞ (M), such that the
skew-symmetric part of 2μ1 is π . Of course, one has Sμ ,π ⊆ Sμ and according to
item (2) of Theorem 13.51, there exists a bijection ψ : D0 → Sμ .
An element π = π + ∑k∈N∗ πk ν k ∈ X1 (M)ν is a formal deformation of π if
and only if [π , π ]S = 0, which is also equivalent to [π ν , π ν ]S = 0, meaning that
π ν = πν + ∑k∈N∗ πk ν k+1 is a formal deformation of the trivial Poisson structure.
Moreover, saying that two formal deformations π and π of π are equivalent means
408 13 Deformation Quantization

that π − π and π − π are two gauge equivalent solutions of the Maurer–Cartan
equation associated to the pointed differential graded Lie algebra (X• (M), [· , ·]S , π ),
i.e., that there exists ξ ∈ ν X0 (M)ν satisfying

π = eadξ (π ) .

Since this equation is equivalent to π ν = ν eadξ (π ) = eadξ (π ν ), two formal de-
formations π and π of π are equivalent if and only if π ν and π ν are equivalent
as formal deformations of the trivial Poisson structure. This fact permits us to iden-
tify ν Dπ with a subset of D0 . Now, according to item (3) of Theorem 13.51, we
have ψ (ν Dπ ) ⊆ Sμ ,π . Moreover, if μ = μ + ∑k∈N∗ μk ν k is an element of Sμ ,π ,
one has 2μ1− = π and, once more according to item (3) of Theorem 13.51, there
exists a formal deformation π = ∑k∈N∗ πk ν k of the trivial Poisson structure, whose
equivalence class is sent by ψ to μ and such that π = 2μ1− = π1 . This implies in
particular that ν1 π is a formal deformation of the Poisson structure π and that μ
lies in ψ (ν Dπ ). This shows that ψ (ν Dπ ) = Sμ ,π , so that ψ induces a bijection
between ν Dπ and Sμ ,π , hence between Dπ and Sμ ,π .

Example 13.53. In the case of Rd , the explicit construction of the L∞ -quasi-iso-

morphism which we have given in Subsection 13.5.2.4 leads to an explicit formula
for the star product on C∞ (Rd ), equipped with an arbitrary Poisson structure π .
Specializing (13.80) to (13.79), gives the following formula for the star product
on C∞ (Rd ), for F, G ∈ C∞ (Rd ),

νk
μ (F, G) = μ (F, G) + ∑ Φk (π k )(F, G)
k1 k!
νk
= μ (F, G) + ∑ ∑ ϖΓ BΓ (π , . . . , π )(F, G) .
k1 k! Γ ∈Gk,2

In the above sum, only those graphs for which every vertex of the first type is the
tail of exactly two arrows contribute, since BΓ (π , . . . , π ) is equal to zero for all other
graphs.

13.6 Notes

Deformation quantization was proposed by Bayen, Flato, Fronsdal, Lichnerowicz

and Sternheimer in [19] as an approach to reverse the standard operation of tak-
ing the classical limit of a quantum system, which is loosely speaking done by
letting Planck’s constant h̄ tend to zero. They raise the question whether a defor-
mation quantization of every Poisson manifold exists. For symplectic manifolds,
an affirmative answer to this question was given by De Wilde–Lecomte [57], who
use the Moyal–Weyl product, which answers the question locally (for symplec-
tic manifolds). A geometric proof of the latter result was given by Fedosov [73].
13.6 Notes 409

The fact that every Poisson manifold admits a deformation quantization has been
shown by Kontsevich [107]. An alternative proof of Kontsevich’s result was given
by Tamarkin [189].
In a purely algebraic context, the problem of (formally) deforming commuta-
tive associative algebras was considered by Gerstenhaber in [82], who establishes
the link with Hochschild cohomology. Deformations of Poisson structures were in-
troduced in [126]. Deformations of Poisson manifolds, in which both the associa-
tive product and the Poisson bracket are deformed, are considered in Ginzburg–
Kaledin [85].
Appendix A
Multilinear Algebra

In this appendix, we recall some basic definitions and properties related to multi-
linear algebra, including (graded) algebra and coalgebra structures, derivations and
coderivations. We need them in two particular cases, namely for vector spaces over
a field F, and for A -modules, where A is a commutative associative algebra over a
field F. Notice that the latter modules can also be thought of as vector spaces over F,
a fact which we often use, since many operations which we consider are F-linear,
rather than A -linear. In order to cover both cases, we consider in this appendix
the structures which we need on modules over an arbitrary commutative ring R with
unit; the reader may find it useful to keep in mind the example of the C∞ (M)-module
of vector fields or differential forms over a manifold M.
Throughout the appendix, F denotes a field of characteristic zero and R denotes
an arbitrary commutative ring with unit, denoted by 1.

A.1 Tensor Algebra

For R-modules V and W , the set of linear1 maps V → W is denoted by HomR (V,W ),
or by Hom(V,W ) when it is clear that V and W are considered as R-modules.
Hom(V,W ) is itself an R-module in a natural way. When V is a free R-module (for
example when R = F, so that V is an F-vector space) and B is a basis of V , every
map B → W extends to a unique element of Hom(V,W ) by linearity; when both V
and W are finite-dimensional F-vector spaces, and bases for V and W have been
fixed, we often think of elements of Hom(V,W ) as matrices (with dimW rows and
dimV columns).
The dual of an R-module V is the R-module V ∗ := Hom(V, R). For a finite-
dimensional F-vector space V , the dual V ∗ is an F-vector space, isomorphic to V ;
the isomorphism is however not canonical, since it depends on the choice of a basis
1 We rarely use the word R-linear, to avoid confusion with the terminology for multilinear maps,
i.e., k-linear maps, with k ∈ N∗ .

C. Laurent-Gengoux et al., Poisson Structures, 411

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4, © Springer-Verlag Berlin Heidelberg 2013
412 A Multilinear Algebra

for V . The natural pairing (evaluation map)

· , · : V ∗ ×V → R
(A.1)
(ξ , v) → ξ , v := ξ (v) ,

leads for fixed v ∈ V to a linear map ·, v : V ∗ → R, i.e., to an element of (V ∗ )∗ =

Hom(V ∗ , R). The R-module (V ∗ )∗ is called the bidual of V . The resulting linear map
V → (V ∗ )∗ is, in general, neither injective nor surjective, but there are two important
particular cases in which it is an isomorphism:
• When V is a finite-dimensional vector space over R = F;
• When R = C∞ (M), the algebra of smooth functions on a smooth manifold M,
with V the space of smooth differential k-forms on M; in this case, V ∗ can be
identified with the R-module of smooth vector fields on M.

We will often deal with bilinear maps, sometimes with more general multilinear
maps between R-modules, where multilinear means (R-)linear in each of its argu-
ments, keeping the other arguments fixed. The language of tensor products, which
we introduce now, is very useful for this. Let V1 , V2 ,V and W be R-modules and
let p : V1 ×V2 → V be a bilinear map. If we compose p with a linear map V → W ,
then we obtain a bilinear map V1 × V2 → W ; one may wonder if, given V1 and V2 ,
there exists an R-module V , such that, for every R-module W , every bilinear map
V1 × V2 → W can be obtained in this way from a linear map on V → W . In fact,
there is a unique (up to isomorphism) such R-module, called the tensor product of
V1 and V2 , denoted by V1 ⊗R V2 , or V1 ⊗V2 , and it comes with a natural bilinear map
p : V1 × V2 → V1 ⊗ V2 . Formally, the stated property means that every bilinear map
φ : V1 × V2 → W factors uniquely via p, meaning that there exists a unique linear
map φ , such that φ ◦ p = φ . This property is displayed in the following diagram:
p
V1 ×V2 V1 ⊗V2

φ φ

In practice we do not make a distinction between the bilinear map φ and the linear
map φ , so we simply write φ ∈ Hom(V1 ⊗ V2 ,W ). A natural construction of the
tensor product V1 ⊗V2 is as the quotient of the free R-module which is generated by
all formal expressions v1 ⊗ v2 , with v1 ∈ V1 and v2 ∈ V2 , divided by the equivalence
relation defined by
(v1 + v1 ) ⊗ v2 = v1 ⊗ v2 + v1 ⊗ v2 ,
v1 ⊗ (v2 + v2 ) = v1 ⊗ v2 + v1 ⊗ v2 ,
a(v1 ⊗ v2 ) = (av1 ) ⊗ v2 = v1 ⊗ (av2 ),
where v1 , v1 ∈ V1 and v2 , v2 ∈ V2 and a ∈ R. One says that v1 ⊗ v2 is the ten-
sor product of v1 and v2 . The maps p and φ are in this notation simply given by
A.1 Tensor Algebra 413

p(v1 , v2 ) = v1 ⊗ v2 and φ (v1 ⊗ v2 ) = φ (v1 , v2 ). In view of the above three properties,

p is a bilinear map.
As an application, consider the bilinear map V1 ×V2 → V2 ⊗V1 , which is given by
(v1 , v2 ) → v2 ⊗ v1 . It factors via p to yield an R-module isomorphism S : V1 ⊗V2 →
V2 ⊗V1 , given by S(v1 ⊗ v2 ) = v2 ⊗ v1 , and called the twist map.
One easily shows that for arbitrary R-modules V1 , V2 and V3 , one has a natural
isomorphism (V1 ⊗ V2 ) ⊗ V3 V1 ⊗ (V2 ⊗ V3 ). The latter is also denoted as V1 ⊗
V2 ⊗ V3 as it can also be described, like in the case of bilinear maps, as a universal
object for trilinear maps, defined on V1 ×V2 ×V3 . Under the natural isomorphisms,
(V1 ⊗V2 ) ⊗V3 V1 ⊗ (V2 ⊗V3 ) V1 ⊗V2 ⊗V3 , one has that

(v1 ⊗ v2 ) ⊗ v3 ↔ v1 ⊗ (v2 ⊗ v3 ) ↔ v1 ⊗ v2 ⊗ v3 , (A.2)

so that, in the sequel, we will not make a notational distinction between the elements
in (A.2). The extension to several R-modules is clear. For a given R-module V , we
obtain a natural sequence of R-modules T kV := V ⊗k , where k = 0, 1, 2, . . . , which is
defined, as the notation suggests, by V ⊗k := V ⊗ · · · ⊗V (k factors), when k  1 and
V ⊗0 := R. Equipped with the product (X,Y ) → X ⊗Y , where X ∈ V ⊗k and Y ∈ V ⊗ ,
the R-module
∞
 ∞

T •V := T kV = V ⊗k
k=0 k=0

becomes a graded R-algebra, called the tensor algebra of V . See Section A.3 below
for the basic definitions on graded algebras.
Fixing one R-module Z, it is useful to think of taking the tensor product with Z
as a functor, which means on the one hand that a linear map φ ∈ Hom(V,W ) yields,
in a natural way, a linear map

φ̃ ∈ Hom(V ⊗ Z,W ⊗ Z) , (A.3)

simply by putting φ̃ (v ⊗ z) := φ (v) ⊗ z, for v ∈ V and z ∈ Z, which is well-defined;

the map φ̃ is usually denoted by φ ⊗ 1Z , where 1Z stands for the identity map on Z.
On the other hand, it means that taking the tensor product with Z has the usual
functorial properties, which make it into a covariant functor. In formulas, this is
written as follows:

1V ⊗ 1Z = 1V ⊗Z , (φ ◦ ψ ) ⊗ 1Z = (φ ⊗ 1Z ) ◦ (ψ ⊗ 1Z ) ,

where φ ∈ Hom(V,W ) and ψ ∈ Hom(U,V ). Above, we tensored with Z on the right,

but we could also have tensored with Z on the left.
For R-modules Vi and Wi , with i = 1, 2, there is also a natural injective morphism
(which is not surjective, in general)

Hom(V1 ,W1 ) ⊗ Hom(V2 ,W2 ) → Hom(V1 ⊗V2 ,W1 ⊗W2 ) ,

414 A Multilinear Algebra

where φ1 ⊗ φ2 ∈ Hom(V1 ,W1 ) ⊗ Hom(V2 ,W2 ) is, as a linear map from V1 ⊗ V2 to

W1 ⊗W2 , given by

(φ1 ⊗ φ2 )(v1 ⊗ v2 ) := φ1 (v1 ) ⊗ φ2 (v2 ) ,

for all v1 ∈ V1 and v2 ∈ V2 . This justifies the notation φ ⊗ 1Z introduced above.

For R-modules V1 , V2 and W , bilinear maps V1 × V2 → W are also in natural
correspondence with linear maps V1 → Hom(V2 ,W ), or with linear maps V2 →
Hom(V1 ,W ). We use this usually in the form of the two natural isomorphisms

Hom(V1 ⊗V2 ,W ) Hom(V1 , Hom(V2 ,W )) Hom(V2 , Hom(V1 ,W )) ,

which allows one to use a single bilinear map V1 × V2 → W to associate to each

element of V1 (respectively V2 ) a linear map V2 → W (respectively a linear map
V1 → W ). For example, for arbitrary R-modules V and W ,

(V ⊗W )∗ Hom(V,W ∗ ) Hom(W,V ∗ ) . (A.4)

For given R-modules V and W , there is also a natural linear map Ψ : V ∗ ⊗ W →

Hom(V,W ), which associates to an element ξ ⊗ w, with ξ ∈ V ∗ and w ∈ W , the
linear map
Ψ (ξ ⊗ w) : V → W
v → ξ (v) w = ξ , v w .
The map Ψ is always injective, but is in general not surjective. When R = F, then
Ψ is an isomorphism if and only if V or W is finite-dimensional. Combining the
injection V ∗ ⊗W ∗ → Hom(V,W ∗ ) with the first isomorphism in (A.4), we obtain a
natural inclusion
V ∗ ⊗W ∗ → (V ⊗W )∗ ,
which is an isomorphism when R = F and V or W is finite-dimensional.
When φ1 ∈ Hom(V1 , A ) and φ2 ∈ Hom(V2 , A ), where A is an R-algebra, then
φ1 ⊗ φ2 is often implicitly combined with the multiplication map A ⊗ A → A ,
yielding φ1 ⊗ φ2 ∈ Hom(V1 ⊗V2 , A ), given for v1 ∈ V1 and v2 ∈ V2 , by

(φ1 ⊗ φ2 )(v1 ⊗ v2 ) := φ1 (v1 )φ2 (v2 ) , (A.5)

where the latter product is the multiplication in A . When this multiplication has
extra properties, they yield similar properties for the product of maps: for example,
if A is commutative, then

φ1 ⊗ φ2 = (φ2 ⊗ φ1 ) ◦ S ,

as elements of Hom(V1 ⊗V2 , A ). Associativity of A implies that

(φ1 ⊗ φ2 ) ⊗ φ3 = φ1 ⊗ (φ2 ⊗ φ3 ) ,

as elements of Hom((V1 ⊗V2 ) ⊗V3 , A ) Hom(V1 ⊗ (V2 ⊗V3 ), A ).

A.2 Exterior and Symmetric Algebra 415

A.2 Exterior and Symmetric Algebra

Our multilinear maps are usually skew-symmetric k-linear maps V k → V , where V is

an R-module, so we recall here the corresponding tensorial notions. In T •V , consider
the submodule N which is generated by all -tensors ( ∈ N∗ ) of the form vi1 ⊗ · · · ⊗
vi , where vis = vit for some 1  s < t  , and let Nk := N ∩ T kV for k ∈ N∗ and
N0 := {0}. Then we may consider
∞
 ∞

∧•V := T •V /N = T kV /Nk = ∧k V ,
k=0 k=0

where ∧kV := T kV /Nk , for k ∈ N. Notice that ∧kV = {0} as soon as k is bigger
than the (minimal) number of generators of V . We denote the quotient maps by
p : T •V → ∧•V and pk : T kV → ∧kV . Since N is a two-sided ideal of T •V , we have
that the associative product ⊗ on T •V induces an associative product on ∧•V , which
is denoted by ∧. Thus p(v1 ⊗ · · · ⊗ vk ) = p(v1 ) ∧ · · · ∧ p(vk ), which we also write as
v1 ∧· · ·∧vk , because p1 (i.e., the restriction of p to V ) is injective. One easily verifies
that, if X ∈ ∧iV and Y ∈ ∧ jV , then X ∧Y ∈ ∧i+ jV and

X ∧Y = (−1)i jY ∧ X . (A.6)

In the language of the next section, this property is called graded commutativity.
The associative, graded commutative R-algebra (∧•V, ∧) is called the exterior alge-
bra of V and elements of ∧•V are called multivectors. One similarly constructs the
symmetric algebra (S•V, ·) which is the associative commutative graded R-algebra
obtained as T •V /N , where N is the two-sided ideal of T •V , generated by all
v ⊗ w − w ⊗ v, where v, w ∈ V .
It is clear that every skew-symmetric k-linear map φ ∈ Hom(V ⊗k ,W ) vanishes
on Nk . Therefore, for R-modules V and W , we have that every skew-symmetric
k-linear map φ ∈ Hom(V ⊗k ,W ) corresponds in a canonical way to a linear map
φ : ∧kV → W , as in the following commutative diagram:
pk
V ⊗k ∧k V

φ φ

In formulas, φ (v1 ∧ · · · ∧ vk ) = φ (v1 , . . . , vk ). From now on, we do not distinguish

notationally between the maps φ and φ : we write φ ∈ Hom(∧kV,W ) and we simply
say that φ is a skew-symmetric k-linear map.
One usually thinks of elements of ∧kV as skew-symmetric tensors: the permuta-
tion group Sk defines a natural linear action on V ⊗k which is defined for σ ∈ Sk and
v1 ⊗ · · · ⊗ vk ∈ V ⊗k by
416 A Multilinear Algebra

σ (v1 ⊗ · · · ⊗ vk ) := vσ (1) ⊗ · · · ⊗ vσ (k) .

An element X ∈ V ⊗k is called a symmetric tensor when σ (X) = X for all σ ∈ Sk ,

while it is called a skew-symmetric tensor when σ (X) = sgn(σ )X for all σ ∈ Sk ,
where sgn(σ ) denotes the signature of σ . In order to identify the skew-symmetric
k-tensors with ∧kV , one defines a linear map ρk− , the skew-symmetrization map, by

ρk− : ∧k V → T kV
1
k! σ∑
v1 ∧ · · · ∧ vk → sgn(σ )vσ (1) ⊗ · · · ⊗ vσ (k) . (A.7)
∈S k

It is easy to see that this map is well-defined and injective, and that its image consists
precisely of all skew-symmetric k-tensors. One similarly identifies the symmetric k-
tensors with SkV by using the symmetrization map

ρk+ : SkV → T kV ,

whose definition is formally the same as the above definition (A.7) of the skew-
symmetrization map, except that one leaves out the factor sgn(σ ).
There are two types of internal products related to the exterior algebra. Let V
and W be arbitrary R-modules. For X ∈ ∧ jV , the internal product ıX yields, for
every i ∈ N, a linear map

ıX : Hom(∧iV,W ) → Hom(∧i− jV,W ) ,

which is given by
ıX φ (Z) := φ (X ∧ Z)
where φ : ∧iV → W and Z ∈ ∧i− jV , assuming i  j; otherwise, i.e., when i < j, then
ıX φ := 0. It is easily verified that

ıX∧Y = ıY ◦ ıX ,

for X ∈ ∧ jV and Y ∈ ∧kV . For φ ∈ Hom(∧iV, R) the internal product ıφ , is the

family of linear maps, indexed by j ∈ N,

ıφ : ∧ jV → ∧ j−iV ,

defined for all v1 ∧ · · · ∧ v j ∈ ∧ jV by

ıφ (v1 ∧ · · · ∧ v j ) := ∑ sgn(σ ) φ (vσ (1) , . . . , vσ (i) ) vσ (i+1) ∧ · · · ∧ vσ ( j)

σ ∈Si, j−i

when i  j, and ıφ is the zero map otherwise. In this formula, Si,k denotes the set
of all (i, k)-shuffles, i.e., all permutations σ ∈ Si+k for which σ (1) < · · · < σ (i) and
σ (i + 1) < · · · < σ (i + k); sgn(σ ) is the signature of σ as a permutation.
A.2 Exterior and Symmetric Algebra 417

To finish this section, we define a product on skew-symmetric multilinear maps

which take values in a commutative associative R-algebra A . We write FG for the
product of elements F and G in A . As above, V denotes an arbitrary R-module;
often, in the body of the book, this product will be considered for V = A , viewed
as an F-vector space (R = F). For φ ∈ Hom(∧iV, A ) and ψ ∈ Hom(∧ jV, A ), where
i, j > 0, we define φ ∧ ψ ∈ Hom(∧i+ jV, A ) by

(φ ∧ ψ )(v1 , . . . , vi+ j ) := ∑ sgn(σ ) φ (vσ (1) , . . . , vσ (i) ) ψ (vσ (i+1) , . . . , vσ (i+ j) ) ,

σ ∈Si, j

for all v1 , . . . , vi+ j ∈ V . We also write

 
φ ∧ ψ , v1 ∧ · · · ∧ vi+ j := (φ ∧ ψ )(v1 , . . . , vi+ j ) ,

thereby extending the notation of (A.1).

Notice that, if A = R, so that φ ∈ (∧iV )∗ and ψ ∈ (∧ jV )∗ , then we can alterna-
tively define φ ∧ ψ ∈ (∧i+ jV )∗ as being the linear map ψ ◦ ıφ . In this particular case,
the definition implies the following relation between ı and ∧:

ıφ ∧ψ = ıψ ◦ ıφ ,

for φ , ψ ∈ ⊕k∈N (∧kV )∗ . Indeed, let φ ∈ Hom(∧iV, A ) and ψ ∈ Hom(∧ jV, A ),

where i, j > 0, and let v1 , . . . , vk ∈ V , with k  i + j. Then,

ıφ ∧ψ (v1 , . . . , vk )
= ∑ sgn(σ )(φ ∧ ψ )(vσ (1) , . . . , vσ (i+ j) ) vσ (i+ j+1) ∧ · · · ∧ vσ (k)
σ ∈Si+ j,k−i− j

= ∑ sgn(τ ) φ (vτ (1) , . . . , vτ (i) ) ψ (vτ (i+1) , . . . , vτ (i+ j) )

τ ∈Si, j,k−i− j
(A.8)
vτ (i+ j+1) ∧ · · · ∧ vτ (k)
 
= ıψ ∑ sgn(ρ )φ (vρ (1) , . . . , vρ (i) ) vρ (i+1) ∧ · · · ∧ vρ (k)
ρ ∈Si,k−i

= (ıψ ◦ ıφ )(v1 , . . . , vk ) ,

where we have used the notation Si, j,k−i− j for the set of all permutations of {1, . . . , k}
that are increasing on the intervals 1, . . . , i and i + 1, . . . , i + j and i + j + 1, . . . , k.
We return to the general case for which A is not necessarily R, but an arbitrary
commutative associative R-algebra. We point out that, if φ ∈ Hom(∧iV, A ), ψ ∈
Hom(∧ jV, A ) and χ ∈ Hom(∧kV, A ), where i, j, k  0, then

(φ ∧ ψ ) ∧ χ = φ ∧ (ψ ∧ χ ) ,

i.e., ∧ is an associative product on ⊕∈N Hom(∧ , A ). This is obtained by exactly

the same computation as in (A.8), with k =  − i − j and with χ acting on the last
418 A Multilinear Algebra

part (i.e., on the last vs that appear out of the arguments of φ and ψ ). Moreover, for
φ ∈ Hom(∧iV, A ) and ψ ∈ Hom(∧ jV, A ), we have

φ ∧ ψ = (−1)i j ψ ∧ φ .

In the language of the next section, the product ∧ makes ⊕k∈N Hom(∧kV, A ) into a
graded R-algebra which is associative and graded commutative. Notice that, in the
notation which we use, if φ1 , . . . , φk ∈ Hom(V, A ) and v1 , . . . , vk ∈ V , then
 
φ1 ∧ · · · ∧ φk , v1 ∧ · · · ∧ vk  = det φi , v j 1i, jk . (A.9)

A.3 Algebras and Graded Algebras

In this section we recall the basic definitions of algebras and graded algebras. These
definitions will be dualized in the next section, to obtain the notions of a coalgebra
and of a graded coalgebra.
Let V be an R-module. An algebra structure on V is a bilinear map μ : V ×V → V ,
called a product. We also view μ as an element of Hom(V ⊗V,V ), and we say that
(V, μ ) is an R-algebra. Usually, μ is assumed to have additional properties; the typ-
ical extra properties that μ may be supposed to have are summarized in Table A.1.

Table A.1 A product μ on an R-module, which makes it into an R-algebra, is usually assumed to
have one or two additional properties, taken from the list which appears in this table. We write the
properties in their usual form (with u, v, w ∈ V ) and in their functional form; the latter is useful for
obtaining the “co”-version (see Section A.4). S is the twist map u ⊗ v → v ⊗ u and S is the cycle
map u ⊗ v ⊗ w → v ⊗ w ⊗ u.

Property Usual / functional form

commutative μ (v, u) = μ (u, v)

μ ◦S = μ

skew-symmetric μ (v, u) = −μ (u, v)

μ ◦ S = −μ

μ (u, μ (v, w)) = μ (μ (u, v), w)

associative
μ ◦ (1V ⊗ μ ) = μ ◦ (μ ⊗ 1V )

μ (u, μ (v, w))+  (u, v, w) = 0

Jacobi identity
∑2=0 μ ◦ (1V ⊗ μ ) ◦ S  = 0

The usual combinations of adjectives are the following: (1) If μ is skew-symmetric

and satisfies the Jacobi identity, then (V, μ ) is called a Lie algebra, and μ is called
a Lie bracket on V . We employ the standard custom of using brackets, such as [· , ·]
and {· , ·}, for the product. (2) An equally important combination of properties that μ
A.3 Algebras and Graded Algebras 419

may have are commutativity and associativity; some authors call (V, μ ) in this case
simply an algebra (or R-algebra), but we will not use this convention here since
our modules will usually have two algebra structures, one of which is a Lie algebra
structure, and the other one is associative and commutative.

Example A.1. A simple example of a Lie algebra structure is given by the vector
space Hom(W,W ), where W is an arbitrary vector space, equipped with the commu-
tator
[φ1 , φ2 ] := φ1 ◦ φ2 − φ2 ◦ φ1 ,
where φ1 , φ2 ∈ Hom(W,W ). The Jacobi identity for [· , ·] is a direct consequence of
the associativity of the composition of (linear) maps.

A linear map φ : V → W between R-algebras (V, μ ) and (W, μ ) is called an

algebra homomorphism if φ (μ (v1 , v2 )) = μ (φ (v1 ), φ (v2 )), for all v1 , v2 ∈ V , which
is written in functional form as φ ◦ μ = μ ◦ (φ ⊗ φ ), and which corresponds to the
commutativity of the following diagram.
μ
V ⊗V V

φ⊗φ φ (A.10)
μ
W ⊗W W

In the case of Lie algebras (V, [· , ·]) and (W, [· , ·] ), such a linear map φ is called
a Lie algebra homomorphism. The homomorphism property then takes the form
φ ([v1 , v2 ]) = [φ (v1 ), φ (v2 )] for all v1 , v2 ∈ V .
We now turn to the graded version of these definitions. For this, it is assumed
that V is a graded R-module, 
V= Vi , (A.11)
i∈Z

where each of the subspaces Vi is invariant under the action of R. The notation V•
(or V • , when the subspaces are indexed by superscripts) is also used for V . In many
cases, one has V = ⊕i∈NVi , or even V = ⊕ki=0Vi , the other Vi being undefined, but
one easily arrives at the form (A.11) by defining Vi := {0}, for those values of i
where Vi was undefined. An element of Vi is called a homogeneous element of V of
degree i. A linear map φ : V → W between graded R-modules is said to be graded
of degree r if φ (Vi ) ⊂ Wi+r for every i ∈ Z. When V and W are written as V• and W• ,
the suggestive notation φ : V• → W•+r is also used. We denote the R-module of all
graded linear maps from V to W of degree r by Homr (V,W ). A graded product on V
is a product μ on V such that

μ (Vi ⊗V j ) ⊂ Vi+ j , for all i, j ∈ Z .

Then V , equipped with μ , becomes a graded R-algebra and we have induced maps
μi, j : Vi ⊗V j → Vi+ j for all i, j. A graded linear map of degree zero φ : V → W which
420 A Multilinear Algebra

is a (Lie) algebra homomorphism is called a graded (Lie) algebra homomorphism.

The graded analog of Table A.1 is given by Table A.2.

Table A.2 The graded analog of Table A.1 is displayed. The only difference between the graded
and ungraded notions lies in the signs; in fact, as there are no signs in the case of graded associa-
tivity, the notion of associativity and graded associativity coincide. For the graded Jacobi identity,
it is understood that when one sums over the three cyclic permutations of (i, j, k), the exponent 
takes the consecutive values 0, 1 and 2. The elements u, v and w are assumed to be homogeneous
of respective degrees i, j and k.

Property Usual / functional form

graded commutative μ (v, u) = (−1)i j μ (u, v)

μ j,i ◦ S = (−1)i j μi, j

graded skew-symmetric μ (v, u) = −(−1)i j μ (u, v)

μ j,i ◦ S = −(−1)i j μi, j

μ (u, μ (v, w)) = μ (μ (u, v), w)

(graded) associative
μi, j+k ◦ (1Vi ⊗ μ j,k ) = μi+ j,k ◦ (μi, j ⊗ 1Vk )

(−1)ik μ (u, μ (v, w))+  (u, v, w) = 0

graded Jacobi identity
(−1)ik μi, j+k ◦ (1Vi ⊗ μ j,k ) ◦ S  +  (i, j, k) = 0

As in the ungraded case, the combination of graded skew-symmetric and the

graded Jacobi identity leads to the notion of a graded Lie bracket and of a graded
Lie algebra. The combination of graded commutativity and associativity leads to
the notion of an associative, graded commutative algebra.

Example A.2. To give a simple example of a graded Lie algebra, we define for
graded linear maps φi ∈ Homri (V,V ), where i = 1, 2, their graded commutator
[φ1 , φ2 ] as the graded linear map of degree r1 + r2 , given by

[φ1 , φ2 ] := φ1 ◦ φ2 − (−1)r1 r2 φ2 ◦ φ1 . (A.12)

The graded R-module ⊕r∈Z Homr (V,V ), equipped with this bracket, is a graded
Lie algebra. For graded linear maps φ1 , φ2 , φ3 of degree r1 , r2 , r3 , the graded Jacobi
identity takes the following form

(−1)r1 r3 [φ1 , [φ2 , φ3 ]] + (−1)r2 r1 [φ2 , [φ3 , φ1 ]] + (−1)r3 r2 [φ3 , [φ1 , φ2 ]] = 0 , (A.13)

which can also be written, in view of the graded skew-symmetry of [· , ·], as

[φ1 , [φ2 , φ3 ]] = [[φ1 , φ2 ] , φ3 ] + (−1)r1 r2 [φ2 , [φ1 , φ3 ]] .

In the language of Section A.5, this means that [φ1 , ·], which is a graded linear map
of degree r1 , is a graded derivation of [· , ·].
A.3 Algebras and Graded Algebras 421

We have in this appendix already met the following three examples of graded
algebras, associated to an R-module V .
(T •V, ⊗) is a graded R-algebra, which is associative;
(∧•V, ∧) is a graded R-algebra, which is associative and graded commutative;
(S•V, ·) is a graded R-algebra, which is associative and commutative.
The grading on T •V comes from the natural decomposition T •V = ⊕i∈NV ⊗i ;
for ∧•V and S•V , the induced grading is used. The commutativity of the graded
algebra (S•V, ·) should not be confused with the graded commutativity of the
graded algebra (∧•V, ∧): in the former the arguments commute, but in the latter
they only commute up to a sign, see (A.6). We think of T • as a functor: given
a linear map φ ∈ Hom(V,W ), we obtain a homomorphism of graded algebras
T • φ : T •V → T •W, whose restriction to T kV is the linear map T kV → T kW , de-
fined by T k φ := φ ⊗ φ ⊗ · · · ⊗ φ , i.e.,

T k φ (v1 ⊗ v2 ⊗ · · · ⊗ vk ) := φ (v1 ) ⊗ φ (v2 ) ⊗ · · · ⊗ φ (vk ) .

T • is a covariant functor: for φ ∈ Hom(V,W ) and ψ ∈ Hom(W, Z) one obtains

T • (1V ) = 1T •V , T • (ψ ◦ φ ) = T • (ψ ) ◦ T • (φ ) .

Similarly, ∧• and S• are covariant functors.

Associated to a graded vector space V = i∈Z Vi there is a graded exterior al-
gebra which takes into account the grading on V . This algebra, denoted by ∧•V
is the associative, graded commutative algebra, obtained by dividing the tensor al-
gebra T •V = k∈N T kV of V by the ideal generated by the elements of the form
x ⊗ y + (−1)i j y ⊗ x, with x ∈ Vi and y ∈ V j . Denoting by ∧ the product in ∧•V , one
then has
x ∧ y = −(−1)i j y ∧ x ,
for all x ∈ Vi and y ∈ V j . For integers i1 , . . . , ik and for σ ∈ Sk an arbitrary permu-
tation of {1, . . . , k}, let sgn(σ ; i1 , . . . , ik ) ∈ {1, −1} denote the sign, defined by the
equality
x1 ∧ · · · ∧ xk = sgn(σ ; i1 , . . . , ik ) xσ (1) ∧ · · · ∧ xσ (k) ,
where x is an arbitrary homogeneous element of V of degree i , for  = 1, . . . , k. As
in the ungraded case, if φ : V ⊗k → W is a linear map, satisfying

φ (x1 ⊗ · · · ⊗ xk ) = sgn(σ ; i1 , . . . , ik ) φ (xσ (1) ⊗ · · · ⊗ xσ (k) ) ,

for all homogeneous elements x1 , . . . , xk , with i denoting the degree of x for  =
1, . . . , k, then φ descends to a linear map ∧•V → W , which we also denote by φ .
422 A Multilinear Algebra

A.4 Coalgebras and Graded Coalgebras

We now consider the “co”-versions of the concepts which were introduced in the
previous section. This is done in the customary way, namely we obtain the “co”-
versions by dualizing the definitions of the previous section, written in their func-
tional forms (see Tables A.1 and A.2), which is done by reversing all arrows and
switching the order of their composition. Let us first dualize the definition of an al-
gebra: a coalgebra structure on an R-module V is an element Δ of Hom(V,V ⊗V ),
called a coproduct. The most important additional properties which Δ may have, are
summarized in Table A.3.

Table A.3 Dualizing the usual properties of the product of an algebra, written in functional form,
as in Table A.1, we obtain the usual properties which the coproduct Δ , defining a coalgebra struc-
ture on an R-module, can have. As before, S is the twist map and S is the cycle map.

Property Functional form

cocommutative S ◦Δ = Δ
co-skew-symmetric S ◦ Δ = −Δ
coassociative (1V ⊗ Δ ) ◦ Δ = (Δ ⊗ 1V ) ◦ Δ
co-Jacobi identity ∑2=0 S  ◦ (1V ⊗ Δ ) ◦ Δ = 0

In order to define a homomorphism of coalgebras, we just reverse the arrows

in (A.10): for given coalgebras (V, Δ ) and (W, Δ ) we call a linear map φ : W → V
a homomorphism if the following diagram is commutative.
Δ
V ⊗V V

φ⊗φ φ

W ⊗W W
Δ

We now consider the graded versions of the above “co”-concepts. Let V be a graded
R-module, V = ⊕i∈ZVi . A graded coproduct is a coproduct Δ on V such that for
every k ∈ Z, 
Δ (Vk ) ⊂ Vi ⊗V j ,
i+ j=k

is finitely supported, i.e., Δ (Vk ) has a non-trivial intersection with only a finite
number of Vi ⊗ V j . Notice that Δ (Vk ) is automatically finitely supported when
Vi = {0} for all i < 0. A graded R-module V , equipped with a graded coprod-
uct Δ , is called a graded R-coalgebra. For fixed i, j ∈ Z, we will need the linear map
Δi, j : Vi+ j → Vi ⊗V j , which is obtained by composing Δ , restricted to Vi+ j , with the
A.4 Coalgebras and Graded Coalgebras 423

natural projection V ⊗V → Vi ⊗V j . The usual properties of a graded coproduct are

displayed in Table A.4.

Table A.4 This last table deals with the case of graded coalgebras. We recall that S is the twist
map and S is the cycle map (see Table A.2).

Property Functional form

graded cocommutative S ◦ Δ j,i = (−1)i j Δi, j
graded co-skew-symmetric S ◦ Δ j,i = −(−1)i j Δi, j
graded coassociative (1Vi ⊗ Δ j,k ) ◦ Δi, j+k = (Δi, j ⊗ 1Vk ) ◦ Δi+ j,k
graded co-Jacobi identity (−1)ik S  ◦ (1Vi ⊗ Δ j,k ) ◦ Δi, j+k +  (i, j, k) = 0

We have seen in the previous section that the graded R-modules T •V, ∧•V
and S•V have a natural (graded) algebra structure. We now show that they also have
a graded coalgebra structure; the latter structure is important at a few places in this
book. The coalgebra structure Δ on T •V is called de-concatenation and is defined,
for k ∈ N and for v1 , . . . , vk ∈ V , by
k
Δ (v1 ⊗ · · · ⊗ vk ) := ∑ (v1 ⊗ · · · ⊗ vi ) ⊗ (vi+1 ⊗ · · · ⊗ vk ) , (A.14)
i=0

which means that the linear maps Δi, j are given by

Δi, j : V ⊗(i+ j) → V ⊗i ⊗V ⊗ j
(A.15)
v1 ⊗ · · · ⊗ vi+ j → (v1 ⊗ · · · ⊗ vi ) ⊗ (vi+1 ⊗ · · · ⊗ vi+ j ) .

Graded coassociativity of Δ is an immediate consequence of the associativity of ⊗:

((Δi, j ⊗ 1Vk−i− j ) ◦ Δi+ j,k−i− j )(v1 ⊗ · · · ⊗ vk )

= ((v1 ⊗ · · · ⊗ vi ) ⊗ (vi+1 ⊗ · · · ⊗ v j )) ⊗ (v j+1 ⊗ · · · ⊗ vk )

= (v1 ⊗ · · · ⊗ vi ) ⊗ ((vi+1 ⊗ · · · ⊗ v j ) ⊗ (v j+1 ⊗ · · · ⊗ vk ))

= ((1Vi ⊗ Δ j,k−i− j ) ◦ Δi,k−i )(v1 ⊗ · · · ⊗ vk ) .

We now turn to the natural coalgebra structure of the exterior algebra ∧•V . Let us
denote by δ the diagonal map V → V × V : v → (v, v). By functoriality of ∧• , it
induces a linear map
∧• δ : ∧•V → ∧• (V ×V ) ,
which we view as a linear map

Δ := ρ ◦ ∧• δ : ∧•V → ∧•V ⊗ ∧•V ,

424 A Multilinear Algebra

where ρ : ∧• (V ×V ) → ∧•V ⊗ ∧•V is the natural isomorphism, given by

 
ρ (v1 , 0) ∧ · · · ∧ (vi , 0) ∧ (0, vi+1 ) ∧ · · · ∧ (0, vk ) :=
(v1 ∧ · · · ∧ vi ) ⊗ (vi+1 ∧ · · · ∧ vk ) ,

where v1 , . . . , vk ∈ V . We denote the natural product on ∧•V ⊗ ∧•V which makes ρ

into a homomorphism of graded algebras by  (not to be confused with Δ ). It
follows easily from the graded commutativity of ∧ that, if vi ∈ ∧ri V for i = 1, . . . , 4,
then
(v1 ⊗ v2 )(v3 ⊗ v4 ) = (−1)r2 r3 (v1 ∧ v3 ) ⊗ (v2 ∧ v4 ) . (A.16)
The commutative diagram
δ
V V ×V

δ δ ×1V

V ×V V ×V ×V
1V ×δ

leads to a commutative diagram

∧• δ
∧•V ∧• (V ×V )

∧• δ ∧• (δ ×1V )

∧• (V ×V ) ∧• (V ×V ×V )
∧• (1V ×δ )

which, in terms of Δ , becomes the coassociativity property of the coproduct Δ .

Δ
∧•V ∧•V ⊗ ∧•V

Δ Δ ⊗1∧•V

∧•V ⊗ ∧•V 1∧•V ⊗Δ

∧•V ⊗ ∧•V ⊗ ∧•V

The graded algebra structure on ∧•V ⊗ ∧•V ⊗ ∧•V is defined as in (A.16). An ex-
plicit formula for Δ is given by

Δ (v1 ∧ v2 ∧ · · · ∧ vk ) = ρ (∧• δ (v1 ∧ · · · ∧ vk )) = ρ (δ (v1 ) ∧ · · · ∧ δ (vk ))

 
= ρ ((v1 , 0) + (0, v1 )) ∧ · · · ∧ ((vk , 0) + (0, vk ))
= (v1 ⊗ 1 + 1 ⊗ v1 ) · · · (vk ⊗ 1V + 1V ⊗ vk )
A.5 Graded Derivations and Coderivations 425

= ∑ ∑ sgn(σ )(vσ (1) ∧ · · · ∧ vσ (i) ) ⊗ (vσ (i+1) ∧ · · · ∧ vσ (k) ) .

i+ j=k σ ∈Si, j

In particular, Δi, j is given by

Δi, j (v1 ∧ · · · ∧ vi+ j ) = ∑ sgn(σ )(vσ (1) ∧ · · · ∧ vσ (i) ) ⊗ (vσ (i+1) ∧ · · · ∧ vσ (i+ j) ) ,
σ ∈Si, j
(A.17)
when i, j ∈ N and Δi, j = 0 when i < 0 or j < 0. The above formula is similar to the
de-concatenation formula (A.14) which we have introduced in the case of the tensor
algebra.

A.5 Graded Derivations and Coderivations

Let (V, μ ) be a (not necessarily associative) graded algebra, where V = ⊕i∈ZVi is a

graded R-module. A graded linear map φ ∈ Homr (V,V ) of degree r is said to be a
graded derivation of degree r if

φ (vw) = φ (v)w + (−1)rp vφ (w) ,

for all v ∈ Vp and w ∈ Vq , where vw is a shorthand for μ (v, w). In functional notation
this means that for every p, q ∈ Z,

φ ◦ μ p,q = μ p+r,q ◦ (φ ⊗ 1Vq ) + (−1)rp μ p,r+q ◦ (1Vp ⊗ φ ) , (A.18)

as linear maps Vp ⊗ Vq → Vp+r+q . We denote the R-module of all graded deriva-

tions of degree r of V by Derr (V ). It is easily verified by direct computation that
the graded commutator of two graded derivations of degrees r1 and r2 is a graded
derivation of degree r1 + r2 . This implies that ⊕r∈Z Derr (V ) is a graded Lie algebra,
with the graded commutator as Lie bracket. It is a Lie subalgebra of the graded Lie
algebra ⊕r∈Z Homr (V,V ), equipped with the graded commutator, which we consid-
ered in Section A.3. Clearly, a derivation is completely determined by its values on
an arbitrary set of elements of V , which generates V as a (graded) algebra.
Example A.3. Let V be an R-module and consider the graded algebra ∧•V . Equipped
with the graded commutator, the R-module ⊕r∈Z Derr (∧•V ) is a graded Lie algebra.
Particular elements of Derr (∧•V ) can be constructed from linear maps V → ∧r+1V :
every linear map
φ : V → ∧r+1V
extends to a derivation of degree r of the graded algebra (∧•V, ∧) by putting

φ̃ : ∧•V → ∧•+rV
k
v1 ∧ · · · ∧ vk → ∑ (−1)i−1 φ (vi ) ∧ v1 ∧ · · · ∧ vi ∧ · · · ∧ vk .
i=1
426 A Multilinear Algebra

Indeed, for X ∈ ∧ pV and Y ∈ ∧qV ,

φ̃ (X ∧Y ) = φ̃ (X) ∧Y + (−1) pr X ∧ φ̃ (Y ),

which is an easy consequence of the fact that φ will either be applied to a factor
which appears in X, leaving Y untouched, or vice versa.
We now formulate the notion of a derivation for the case of a graded coalgebra.
As before, (V, Δ ) is a graded coalgebra, where V = ⊕i∈ZVi is a graded R-module.
Dualizing (A.18), a linear map φ : V → V of degree −r is called a graded coderiva-
tion of degree r if for every p, q ∈ Z,

Δ p,q ◦ φ = (φ ⊗ 1Vq ) ◦ Δ p+r,q + (−1)rp (1Vp ⊗ φ ) ◦ Δ p,r+q ,

as maps from Vp+r+q to Vp ⊗Vq . We denote the R-module of all graded coderivations
of degree r of V by CoDerr (V ). Again, it follows by direct computation that the
graded commutator of two graded coderivations of degrees r1 and r2 is a graded
coderivation of degree r1 + r2 . This implies that ⊕r∈Z CoDerr (V ) is also a graded
Lie algebra, with the graded commutator as Lie bracket. Like ⊕r∈Z Derr (V ), it is
a Lie subalgebra of the graded Lie algebra ⊕r∈Z Homr (V,V ), equipped with the
graded commutator. The following example is the “co”-version of Example A.3.
Example A.4. Let V be an R-module and consider the graded coalgebra ∧•V .
Equipped with the graded commutator, the R-module ⊕r∈Z CoDerr (∧•V ) is a graded
Lie algebra. Particular elements of CoDerr (∧•V ) can be constructed from linear
maps ∧r+1V → V : every linear map

φ : ∧r+1V → V

extends to a coderivation of degree r of the graded R-coalgebra (∧•V, Δ ) by putting

φ̃ : ∧ •V → ∧•−rV
v1 ∧ · · · ∧ vk → ∑ sgn(τ ) φ (vτ (1) , . . . , vτ (r+1) ) ∧ vτ (r+2) ∧ · · · ∧ vτ (k) .
τ ∈Sr+1,k−1−r

It is understood that the above definition means that φ̃ = 0 on ∧kV for k  r. It

follows that, for p, q ∈ N we have that

Δ p,q ◦ φ̃ = (φ̃ ⊗ 1∧qV ) ◦ Δ p+r,q + (−1)rp (1∧ pV ⊗ φ̃ ) ◦ Δ p,r+q . (A.19)

All coderivations of ∧•V are obtained in this way: if Φ : ∧•V → ∧•−rV is a coderiva-
tion of degree r of (∧•V, Δ ), then Φ = φ̃ , where φ : ∧r+1V → V is the restriction of
Φ to ∧r+1V . Indeed, since Φ and φ̃ agree on ∧kV , for k  r + 1 (they are both zero
when k  r, for degree reasons), they also agree, in view of (A.19), on ∧r+2V (take
p = q = 1 in (A.19)), and similarly for the higher exterior powers of V .
Appendix B
Real and Complex Differential Geometry

In this appendix we recall the basic notions of differential geometry: the definition
of a real manifold, of a complex manifold and of a vector field on such a manifold.
We also recall briefly the main properties of vector fields on manifolds: the existence
of integral curves of a vector field, the flow of a vector field, the bracket of vector
fields and the straightening theorem, which says that a vector field takes, in well-
chosen coordinates, a simple form. Our definition of vector fields on a manifold is
based on the concept of a pointwise derivation. This approach easily generalizes to
the introduction of the concept of a bivector field on a manifold, a crucial element
in the (geometrical!) definition of the notion of a Poisson structure on a (real or
complex) manifold (see Section 1.3).

B.1 Real and Complex Manifolds

We adopt the following geometric point of view: a differentiable manifold is a (sec-

ond countable, Hausdorff) topological space which is covered by a family of co-
ordinate charts (U, x), where U is an open subset of M, called the domain of the
chart, and x = (x1 , . . . , xd ) is a homeomorphism from U to an open subset of Rd .
The functions x1 , . . . , xd are called local coordinates; they are said to be centered at
m if x(m) = o, where o stands for the origin of Rd . The coordinate charts are de-
manded to be compatible in the sense that, if (U, x) and (V, y) are coordinate charts,
with U ∩V = 0,
/ then the homeomorphism

y ◦ x−1 x(U∩V )
: x(U ∩V ) → y(U ∩V )

is a smooth map, see Fig. B.1. This homeomorphism is called a transition map. A
collection of compatible coordinate charts of M, whose domains cover M, is called
an atlas of M. For connected differentiable manifolds, the integer d is independent
of the coordinate chart; it is called the dimension of M, a terminology which we also
use in the non-connected case, when d is independent of the coordinate chart; it is

C. Laurent-Gengoux et al., Poisson Structures, 427

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4, © Springer-Verlag Berlin Heidelberg 2013
428 B Real and Complex Differential Geometry

Fig. B.1 A manifold comes equipped with an atlas, a collection of coordinate charts (U, x), where
the transition maps y ◦ x−1 between coordinate charts (U, x) and (V, y) (with U ∩ V = 0)
/ are de-
manded to be smooth.

denoted by dim M. When the integer d is even, we may interpret the homeomor-
phisms x as taking values in Cd/2 ; in this case, if all transition maps are complex
analytic (holomorphic), then M is called a complex manifold and d/2 is called the
(complex) dimension of M. It is a trivial, but important, fact that every non-empty
open subset of a real or complex manifold is itself, in a natural way, a real or com-
plex manifold. In particular, every open subset of a (real or complex) vector space
is a (real or complex) manifold.
The main virtue of manifolds is that we can do calculus on them, hence also
analytic geometry. Roughly speaking, the coordinate charts allow us to identify ob-
jects on the manifold, locally, with standard objects on open subsets of Fd (F = R
or F = C) and the transition maps allow us to compare these standard objects. The
first object one thinks of is that of a smooth function: a function F : M → F on a
real (respectively complex) manifold M is called a smooth function (respectively a
holomorphic function) if for every coordinate chart (U, x) of M the function

F̃ = F ◦ x−1 : x(U) → F

is smooth (respectively holomorphic). The function F̃ is called the coordinate ex-

pression of F in the coordinate chart (U, x). See Fig. B.2. The (commutative asso-
ciative) algebra of all such functions F on M will be denoted by F (M). For an open
subset U, viewing U itself as a manifold, we have an algebra of functions F (U) and
there are obvious restriction maps F (M) → F (U), which are in general neither in-
jective nor surjective; the restriction of F ∈ F (M) to U will be denoted by F|U .
Notice that F (M) may consist only of the constant functions, for example when M
is a compact complex manifold.
B.2 The Tangent Space 429

Fig. B.2 Smooth functions on a manifold M are functions on M which are smooth in terms of local
coordinates.

One similarly defines the notion of a smooth map between real manifolds and a
holomorphic map between complex manifolds. It leads to two categories: the cat-
egory of real manifolds, whose objects are real manifolds with smooth maps as
morphisms, and the category of complex manifolds, whose objects are complex
manifolds and whose morphisms are holomorphic maps. Since complex manifolds
are in a natural way also real manifolds, and since holomorphic maps are smooth,
there is a natural forgetful functor from the latter category to the former.

B.2 The Tangent Space

Let M be a manifold and let m ∈ M be an arbitrary point. We define the tangent space
Tm M of M at m. To do this, we consider the set Fm (M) of all pairs (F,U), where U is
an open subset of M which contains m, and F is an element of F (U). Two elements
(F,U) and (G,V ) of Fm (M) are defined to be equivalent, denoted (F,U) ∼ (G,V ), if
there exists a pair (H,W ) ∈ Fm (M), such that W ⊂ U ∩V and H = F|W = G|W . For
(F,U) ∈ Fm (M), we denote its equivalence class by Fm and we call it the germ of F
(or of (F,U)) at m. It is clear that the quotient set Fm (M)/ ∼ of all function germs
at m inherits from F (M) the structure of an associative F-algebra. For example,
in the complex case, the algebra Fm (M)/ ∼ is isomorphic to the algebra of power
series in d variables, whose radius of convergence is positive. Notice that a function
germ Fm has a well-defined value at m, which is simply F(m).
Definition B.1. Let M be a manifold and let m ∈ M. A pointwise derivation δm of
F (M) at m is a linear function
430 B Real and Complex Differential Geometry

Fm (M)
δm : →F,
∼
satisfying, for all functions F and G, defined on a neighborhood of m in M,

δm (Fm Gm ) = F(m) δm Gm + G(m) δm Fm . (B.1)

The vector space of all pointwise derivations of F (M) at m is denoted by Tm M, and

is called the tangent space of M at m, while the dual space Tm∗ M of linear forms
Tm M → F is called the cotangent space of M at m. The canonical pairing between
the dual vector spaces Tm M and Tm∗ M, which amounts to evaluating elements of Tm∗ M
on elements of Tm M, is denoted by · , ·.
One easily deduces from (B.1) that if δm is a pointwise derivation at m, and F is
constant in a neighborhood of m, then δm Fm = 0. For a given function F, defined on
a neighborhood of m in M, consider the function dm F on Tm M, defined by

dm F : Tm M → F
(B.2)
δm → δm Fm .

Clearly, dm F is a linear function, hence it is an element of the cotangent space Tm∗ M.

It is called the differential of F at m. The differential of a function admits a natural
generalization to the case of maps between manifolds. Let M and N be two mani-
folds and let Ψ be a map, defined on a neighborhood of a point m ∈ M, with values
in N. The linear map TmΨ : Tm M → TΨ (m) N, called the tangent map of Ψ at m, asso-
ciates to a pointwise derivation δm of F (M) at m, the pointwise derivation TmΨ (δm )
of F (N) at Ψ (m), defined for every germ GΨ (m) at Ψ (m) by

(TmΨ )(δm ) GΨ (m) := δm (G ◦ Ψ )m ,

see Fig. B.3. This is well-defined, because the germ (G ◦ Ψ )m is independent of the
function G which represents the germ GΨ (m) . When F is a function, defined on a
neighborhood of m, the tangent map at m is a linear map Tm F : Tm M → TF(m) F,
and we can recover the differential dm F, upon composing Tm F with the canonical
isomorphism TF(m) F F, which will be explained below.
The tangent map obeys the usual rules of calculus: for example, if M, N and P
are manifolds, Ψ : M → N and Ξ : N → P are maps and m ∈ M, then

Tm (Ξ ◦ Ψ ) = (TΨ (m) Ξ ) ◦ TmΨ .

In particular, if Ψ is a diffeomorphism (at least in the neighborhood of m), then TmΨ

is invertible and
(TmΨ )−1 = TΨ (m)Ψ −1 .
Consider the vector space Fd , viewed as a d-dimensional manifold, and let m be a
point of Fd . There is a natural isomorphism between the vector spaces Tm Fd and Fd .
Namely to a vector v ∈ Fd we can associate a pointwise derivation vm at m by setting,
B.2 The Tangent Space 431

Fig. B.3 A map Ψ between two manifolds M and N leads for every point m ∈ M to a tangent map
TmΨ , which is a linear map between the tangent space Tm M and TΨ (m) N.

for every germ Fm at m,

d F(m + tv) − F(m)

vm Fm := F(m + tv) = lim .
dt |t=0 t→0 t

It is clear that v → vm defines an injective linear map from Fd to Tm Fd . The surjec-

tivity of this map follows from the Hadamard lemma.
Lemma B.2 (Hadamard’s lemma). Let F ∈ F (U), where U is an open subset
of Fd and let m ∈ U. On a small neighborhood V ⊂ U of m in Fd ,
d
F = F(m) + ∑ (xi − xi (m)) F (i) , (B.3)
i=1

where each of the functions F (i) belongs to F (V ). In particular,

F (i) (m) = (ei )m Fm ,

where (e1 , . . . , ed ) denotes the natural basis of Fd .

For a proof of this lemma, which is essentially a first-order form of Taylor’s theorem,
see [155, p. 17].
Let F ∈ F (U) and consider (B.3), germified at m ∈ U,
432 B Real and Complex Differential Geometry

d
Fm = F(m) + ∑ ((xi )m − xi (m))Fm .
(i)
(B.4)
i=1

Let δm be a pointwise derivation at m. Applying δm to (B.4) we find, using (B.1), that

d d
δm Fm = ∑ δm (xi )m F (i) (m) = ∑ δm (xi )m (ei )m Fm ,
i=1 i=1

for all germs Fm at m, so that

d
δm = ∑ δm (xi )m (ei )m . (B.5)
i=1

This shows that Tm Fd is spanned by the pointwise derivations (ei )m , so that the map,
defined by v → vm , is an isomorphism between Fd and Tm Fd .
If (U, x) is a coordinate chart of a manifold M, centered at m, then each of
the d vectors of the natural basis (e1 , . . . , ed ) of Fd leads to a pointwise derivation
of F (M) at m, defined by

∂  
:= To x−1 (ei )o .
∂ xi m

For a function F, defined on an open neighborhood of m in M, this means that

∂
Fm = ∂i F̃(x(m)) ,
∂ xi m

where F̃ : x(U) → F is the coordinate expression of a representative F ∈ F (U) of

the germ Fm (see Fig. B.2), and 1  i  d; also, ∂i F̃ denotes the derivative of F̃ with
respect to its i-th variable (real or complex). Since To Fd is spanned by the point-
wise derivations −1
 (e1 )o , . .. , (ed)o and since To x is an isomorphism, the pointwise
derivations ∂∂x , . . . , ∂∂x span Tm M.
1 m d m

Remark B.3. When the tangent space of M at m is viewed as an equivalence class

of curves, passing through m, as is done in a more analytic approach to elementary
differential geometry, then the differential dm F of a function F : M → F at m, is
d
dm F(γ̃m ) := F(γ (t))
dt |t=0

where γ : I → M is a curve, defined on a neighborhood I of 0 in F, with γ (0) = m,

whose equivalence class has been denoted by γ̃m . As we have seen, in our approach
to defining the tangent space, the definition of the differential takes the more alge-
braic form
dm F, δm  := δm Fm ,
B.3 Vector Fields 433

for all δm ∈ Tm M. Thus, in our setup, we view tangent vectors as objects which act
on equivalence classes of functions, rather than viewing functions as objects which
define linear forms on equivalence classes of curves, although both points of view
are equivalent (see [198, Ch. 1]).

B.3 Vector Fields

Let us consider a manifold M and a map V , which assigns to every m ∈ M an

element Vm of Tm M. To each function F ∈ F (U), where U is an open subset of M,
we can associate a function V [F] on U by defining, for all m ∈ U,

V [F](m) := Vm Fm ∈ F . (B.6)

We also write Vm [F] for Vm Fm , so that Vm [F] = V [F](m). We say that V is a smooth
vector field (respectively holomorphic vector field) on M if for every open subset
U ⊂ M and for every function F ∈ F (U), the function V [F], defined by (B.6),
belongs to F (U) (i.e., it is a smooth, respectively holomorphic function on U).
When the type of manifold which is considered is irrelevant or is clear from the
context, we simply say vector field for smooth or holomorphic vector field. Notice
that we use square brackets to denote the action of a vector field on a function.
With respect to pointwise multiplication, the vector fields on M form an F (M)-
module, which is denoted by X1 (M). Viewed as a vector space, X1 (M) is a Lie
algebra, where the Lie bracket is the commutator of vector fields, defined as follows.
Let V and W be vector fields on M, and let m ∈ M. For every function F, defined
in a neighborhood of m, letting

[V , W ]m (Fm ) := Vm (W [F])m − Wm (V [F])m ,

leads to a well-defined linear map [V , W ]m : Fm (M)/ ∼ → F, which is eas-

ily shown to be a pointwise derivation at m. For given vector fields V and W
on M, the map which assigns to every m ∈ M the element [V , W ]m of the tan-
gent space Tm M is a (smooth or holomorphic) vector field, hence we have a map
[· , ·] : X1 (M) × X1 (M) → X1 (M). Clearly, [· , ·] is a skew-symmetric bilinear map,
which satisfies the Jacobi identity, hence it defines a Lie algebra structure on X1 (M);
it is called the Lie bracket on vector fields.
We have seen that a map Ψ : M → N leads for every m ∈ M to a linear map TmΨ :
Tm M → TΨ (m) N. However, since Ψ is in general neither injective nor surjective, this
collection of linear maps cannot be used to associate to a vector field V on M, a
vector field on N. Nevertheless, when Ψ is bijective, so that Ψ is a diffeomorphism
(or biholomorphism), we get a vector field Ψ∗ V on N by setting

(Ψ∗ V )Ψ (m) := (TmΨ )Vm ,

for all m ∈ M. The vector field Ψ∗ V is called the pushforward of V by Ψ .

434 B Real and Complex Differential Geometry

It is clear that vector fields can be restricted to open subsets; we usually do not
make a notational distinction between a vector field on M and its restriction to some
open subset of M. It is also clear from (B.6) that V defines, for every open subset U
of M, a derivation of F (U), i.e., we have

V [FG] = F V [G] + G V [F] ,

for all F, G ∈ F (U). In particular, a vector field on M defines a derivation of F (M).

Remark B.4. It is shown in standard books on differential geometry that for a real
manifold M, the above natural correspondence between (smooth) vector fields on M
and derivations of F (M) is bijective. For complex manifolds however, this is not
true in general: think of a compact complex torus Cd /Z2d , which has non-trivial
holomorphic vector fields, but whose algebra of holomorphic functions consists of
constant functions only, so that all its derivations are trivial. The same phenomenon
occurs for skew-symmetric biderivations and bivector fields (e.g., Poisson struc-
tures), introduced in Chapter 1.

Remark B.5. The set of all tangent vectors at m, for m ranging through M, has a
natural vector bundle structure over M, denoted T M → M. The fiber over m is the
vector space Tm M and the vector fields on M can be defined as the (smooth, holo-
morphic) sections of T M → M. In abstract geometrical constructions, it is the latter
point of view on vector fields which is often the most appropriate.

On a coordinate chart (U, x) of M,  thereare d distinguished vector fields ∂ /∂ xi , i =

1, . . . , d, which are defined by m → ∂∂xi , for all m ∈ U, which amounts to defining
m

∂
[F](m) := ∂i F̃(x(m)) , (B.7)
∂ xi

for all F ∈ F (M), where F̃ : x(U) → F is the coordinate expression of F, as in

Fig. B.2. By a slight abuse of notation, we usually write ∂∂ xFi (m) instead of either
expression in (B.7). If V is a map which assigns to every m ∈ U an element Vm ∈
  in a unique way as V = ∑i=1 V ∂ /∂ xi , since the
Tm M, then V can be written d (i)

∂
pointwise derivations ∂ xi form a basis of Tm M, for every m ∈ U. Then V is
m
a (smooth) vector field on U if and only if the coefficients V (i) in this expression
belong to F (U). It is clear that these coefficients V (i) are given by V (i) = V [xi ].
By a slight abuse of language, we often refer to the expression
d
∂
V = ∑ V [xi ] (B.8)
i=1 ∂ xi

as a coordinate expression of V in the coordinate chart (U, x). It is very useful for
explicit computations, as it allows us to compute with vector fields on a manifold,
locally, in the same way as on Fd .
B.4 The Flow of a Vector Field 435

Fig. B.4 For a given vector field on a manifold, there passes through every point of the manifold a
unique integral curve.

B.4 The Flow of a Vector Field

Let V be a vector field on a manifold M and let m ∈ M. The fundamental theo-

rem on the existence and uniqueness of solutions of first order ordinary differential
equations with initial conditions, tells us that there exists a connected open neigh-
borhood I of 0 in F, and there exists a map φ : I → M, such that φ (0) = m and such
that
d
(Tt φ ) = Vφ (t ) , (B.9)
dt t
for all t ∈ I; the map is unique in the sense that if φ1 : I1 → M and φ2 : I2 → M are
two such maps, then they coincide on (I1 ∩ I2 )0 , the connected component of their
common intersection, which contains 0,

φ1 | = φ2 | .
(I1 ∩I2 )0 (I1 ∩I2 )0

The map φ or the pair (I, φ ) is called an integral curve of V , passing through m, see
Fig. B.4. By a slight abuse of notation, the left-hand side in (B.9) is often denoted
by ddtφ (t ); using this notation (B.9) takes the more familiar form

dφ
(t ) = Vφ (t ) .
dt
The integral curves of a vector field depend smoothly on the initial data; this is stated
in a precise way in the following theorem (see [187, Ch. 5] for a proof).
Theorem B.6. Let V be a vector field on a manifold M and let m ∈ M. There exists
a neighborhood U of (0, m) in F × M and there exists a map Φ : U → M such that,
for every (0, m ) ∈ U, the restriction

Φ |I : Im → M , (B.10)
m

is an integral curve of V , passing through m ; the subset Im in (B.10) is the

connected component of U ∩ (F × {m }) which contains (0, m ). Moreover, Φ has
the following property: for every (t , m) in the connected component of (0, m) in
436 B Real and Complex Differential Geometry

U ∩ (F × {m}), there exists a neighborhood Ut of (t , m) in U, such that the restric-

tion
Φ |U : Ut → M ,
t

is a diffeomorphism (biholomorphism) between Ut and its image Φ (Ut ).

We refer to such a map Φ as being the (local) flow of the vector field V in a neigh-
borhood of m. The diffeomorphism Φ |U is called the local flow at t and is usually
t
denoted by Φt (omitting its domain of definition, which is all of M in good cases,
for example when M is a compact real manifold).
An important and useful consequence of the (local) existence of the flow of a
vector field, is the straightening theorem, which says that the coordinate expression
of a vector field on a manifold takes a particularly simple form, at points where the
vector field does not vanish.
Theorem B.7 (Straightening theorem). Let M be a manifold and let V be a vector
field on M. If m ∈ M is such that V (m) = 0, then there exist local coordinates
x1 , . . . , xd on a neighborhood U of m, such that V = ∂ /∂ x1 on U.

B.5 The Frobenius Theorem

Instead of having a vector at every point of a manifold M, as is the case of a vector

field on M, one may have a one-dimensional subspace of the tangent space to M,
at every point of M. This is what is called a 1-dimensional distribution on M; a
k-dimensional distribution D on M is then the datum of a k-dimensional subspace
D(m) of Tm M for every m ∈ M. One says that D is smooth (or holomorphic) if
there exist for every m ∈ M smooth (or holomorphic) vector fields V1 , . . . , Vk , on a
neighborhood U of m, such that

D(m) = span {(V1 )m , . . . , (Vk )m } ,

for every m ∈ U. When it is clear from the context, we often simply say distribution
for smooth (or holomorphic) distribution. A vector field V , defined on an open
subset U of M, is said to be adapted to D on U if Vm ∈ D(m) for every m ∈ U. A
distribution D on M is said to be involutive if for every open subset U of M, and for
every pair of vector fields on U, which are adapted to D on U, their Lie bracket is
also adapted to D on U. Involutivity of a distribution is a very strong condition, as
is plain from Frobenius’ theorem.
Theorem B.8 (Frobenius’ theorem). Let M be a d-dimensional manifold and sup-
pose that D is a k-dimensional distribution on M. If D is involutive, then every point
m ∈ M admits a coordinate chart (U, x), with m ∈ U, such that
 
∂ ∂
D(m ) = span ,..., , for every m ∈ U .
∂ x1 m ∂ xk m
B.5 The Frobenius Theorem 437

For a short and elementary proof, which is immediately adapted to the holomorphic
case, we refer to [41] or [138]. It is clear that Frobenius’ theorem can be seen as a
generalization of the straightening theorem (Theorem B.7).
References

1. R. Abraham and J. Marsden. Foundations of mechanics. Benjamin/Cummings Publishing

Co. Inc. Advanced Book Program, Reading, MA, 1978. Second edition, revised and enlarged,
With the assistance of T. Ratiu and R. Cushman.
2. M. Adler. On a trace functional for formal pseudo differential operators and the symplectic
structure of the Korteweg-de Vries type equations. Invent. Math., 50(3):219–248, 1978/79.
3. M. Adler, P. van Moerbeke, and P. Vanhaecke. Algebraic integrability, Painlevé geometry and
Lie algebras, volume 47 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A
Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd
Series. A Series of Modern Surveys in Mathematics]. Springer-Verlag, Berlin, 2004.
4. A. Alekseev. A new proof of the convexity theorem for the Poisson-Lie moment map. In
Geometry and physics (Aarhus, 1995), volume 184 of Lecture Notes in Pure and Appl. Math.,
pages 231–236. Dekker, New York, 1997.
5. A. Alekseev. On Poisson actions of compact Lie groups on symplectic manifolds. J. Differ-
ential Geom., 45(2):241–256, 1997.
6. A. Alekseev and Y. Kosmann-Schwarzbach. Manin pairs and moment maps. J. Differential
Geom., 56(1):133–165, 2000.
7. A. Alekseev, Y. Kosmann-Schwarzbach, and E. Meinrenken. Quasi-Poisson manifolds.
Canad. J. Math., 54(1):3–29, 2002.
8. A. Alekseev and A. Malkin. Symplectic structure of the moduli space of flat connection[s]
on a Riemann surface. Comm. Math. Phys., 169(1):99–119, 1995.
9. A. Alekseev, A. Malkin, and E. Meinrenken. Lie group valued moment maps. J. Differential
Geom., 48(3):445–495, 1998.
10. A. Alekseev and E. Meinrenken. Clifford algebras and the classical dynamical Yang-Baxter
equation. Math. Res. Lett., 10(2–3):253–268, 2003.
11. A. Alekseev and E. Meinrenken. On the Kashiwara-Vergne conjecture. Invent. Math.,
164(3):615–634, 2006.
12. A. Alekseev, E. Meinrenken, and C. Woodward. Linearization of Poisson actions and singular
values of matrix products. Ann. Inst. Fourier (Grenoble), 51(6):1691–1717, 2001.
13. J. Andersen, J. Mattes, and N. Reshetikhin. The Poisson structure on the moduli space of flat
connections and chord diagrams. Topology, 35(4):1069–1083, 1996.
14. D. Arnal, D. Manchon, and M. Masmoudi. Choix des signes pour la formalité de M. Kontse-
vich. Pacific J. Math., 203(1):23–66, 2002.
15. V. Arnol’d. Mathematical methods of classical mechanics. Springer-Verlag, New York,
1978. Translated from the Russian by K. Vogtmann and A. Weinstein, Graduate Texts in
Mathematics, 60.
16. V. Arnol’d. Remarks on Poisson structures on a plane and on other powers of volume ele-
ments. J. Soviet Math., 47(3):2509–2516, 1989.

C. Laurent-Gengoux et al., Poisson Structures, 439

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4, © Springer-Verlag Berlin Heidelberg 2013
440 References

17. V. I. Arnol’d, S. M. Guseı̆n-Zade, and A. N. Varchenko. Singularities of differentiable maps.

Vol. I, volume 82 of Monographs in Mathematics. Birkhäuser Boston Inc., Boston, MA,
1985. The classification of critical points, caustics and wave fronts, Translated from the Rus-
sian by Ian Porteous and Mark Reynolds.
18. M. Audin. Torus actions on symplectic manifolds, volume 93 of Progress in Mathematics.
Birkhäuser Verlag, Basel, revised edition, 2004.
19. F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, and D. Sternheimer. Deformation theory
and quantization. I. Deformations of symplectic structures. Ann. Physics, 111(1):61–110,
1978.
20. F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, and D. Sternheimer. Deformation theory
and quantization. II. Physical applications. Ann. Physics, 111(1):111–151, 1978.
21. F. A. Berezin. Quantization. Izv. Akad. Nauk SSSR Ser. Mat., 38:1116–1175, 1974.
22. R. Bezrukavnikov and V. Ginzburg. On deformations of associative algebras. Ann. of Math.
(2), 166(2):533–548, 2007.
23. K. H. Bhaskara and K. Viswanath. Poisson algebras and Poisson manifolds, volume 174 of
Pitman Research Notes in Mathematics Series. Longman Scientific & Technical, Harlow,
1988.
24. F. Bottacin. Poisson structures on moduli spaces of sheaves over Poisson surfaces. Invent.
Math., 121(2):421–436, 1995.
25. F. Bottacin. Poisson structures on Hilbert schemes of points of a surface and integrable sys-
tems. Manuscripta Math., 97(4):517–527, 1998.
26. F. Bottacin. Poisson structures on moduli spaces of parabolic bundles on surfaces.
Manuscripta Math., 103(1):31–46, 2000.
27. N. Bourbaki. Lie groups and Lie algebras. Chapters 1–3. Elements of Mathematics (Berlin).
Springer-Verlag, Berlin, 1998. Translated from the French, Reprint of the 1989 English trans-
lation.
28. J.-L. Brylinski. A differential complex for Poisson manifolds. J. Differential Geom.,
28(1):93–114, 1988.
29. D. Bump. Lie groups, volume 225 of Graduate Texts in Mathematics. Springer-Verlag, New
York, 2004.
30. F. Butin. Poisson homology in degree 0 for some rings of symplectic invariants. J. Algebra,
322(10):3580–3613, 2009.
31. M. Cahen, S. Gutt, and J. Rawnsley. Nonlinearizability of the Iwasawa Poisson Lie structure.
Lett. Math. Phys., 24(1):79–83, 1992.
32. M. Cahen, S. Gutt, and J. Rawnsley. Some remarks on the classification of Poisson Lie
groups. In Symplectic geometry and quantization (Sanda and Yokohama, 1993), volume 179
of Contemp. Math., pages 1–16. Amer. Math. Soc., Providence, RI, 1994.
33. A. Cannas da Silva. Lectures on symplectic geometry, volume 1764 of Lecture Notes in Math-
ematics. Springer-Verlag, Berlin, 2001.
34. A. Cannas da Silva and A. Weinstein. Geometric models for noncommutative algebras, vol-
ume 10 of Berkeley Mathematics Lecture Notes. American Mathematical Society, Provi-
dence, RI, 1999.
35. A. Cattaneo. On the integration of Poisson manifolds, Lie algebroids, and coisotropic sub-
manifolds. Lett. Math. Phys., 67(1):33–48, 2004.
36. A. Cattaneo. Deformation quantization from functional integrals. In Déformation, quantifi-
cation, théorie de Lie, volume 20 of Panoramas et Synthèses [Panoramas and Syntheses],
pages 123–164. Société Mathématique de France, Paris, 2005.
37. A. Cattaneo and G. Felder. A path integral approach to the Kontsevich quantization formula.
Comm. Math. Phys., 212(3):591–611, 2000.
38. A. Cattaneo, G. Felder, and L. Tomassini. From local to global deformation quantization of
Poisson manifolds. Duke Math. J., 115(2):329–352, 2002.
39. A. Cattaneo, B. Keller, Ch. Torossian, and A. Bruguières. Déformation, quantification,
théorie de Lie, volume 20 of Panoramas et Synthèses [Panoramas and Syntheses]. Société
Mathématique de France, Paris, 2005.
References 441

40. V. Chari and A. Pressley. A guide to quantum groups. Cambridge University Press, Cam-
bridge, 1995. Corrected reprint of the 1994 original.
41. S. Chern and J. Wolfson. A simple proof of Frobenius theorem. In Manifolds and Lie groups
(Notre Dame, IN, 1980), volume 14 of Progr. Math., pages 67–69. Birkhäuser, Boston,
MA, 1981.
42. C. Chevalley. Invariants of finite groups generated by reflections. Amer. J. Math., 77:778–
782, 1955.
43. N. Chriss and V. Ginzburg. Representation theory and complex geometry. Birkhäuser Boston
Inc., Boston, MA, 1997.
44. D. Collingwood and W. McGovern. Nilpotent orbits in semisimple Lie algebras. Van Nos-
trand Reinhold Mathematics Series. Van Nostrand Reinhold Co., New York, 1993.
45. J. Conn. Normal forms for analytic Poisson structures. Ann. of Math. (2), 119(3):577–601,
1984.
46. J. Conn. Normal forms for smooth Poisson structures. Ann. of Math. (2), 121(3):565–593,
1985.
47. A. Coste, P. Dazord, and A. Weinstein. Groupoı̈des symplectiques. In Publications du
Département de Mathématiques. Nouvelle Série. A, Vol. 2, volume 87 of Publ. Dép. Math.
Nouvelle Sér. A, pages i–ii, 1–62. Univ. Claude-Bernard, Lyon, 1987.
48. Th. Courant. Dirac manifolds. Trans. Amer. Math. Soc., 319(2):631–661, 1990.
49. M. Crainic and R. Fernandes. Integrability of Lie brackets. Ann. of Math. (2), 157(2):575–
620, 2003.
50. M. Crainic and R. Fernandes. Integrability of Poisson brackets. J. Differential Geom.,
66(1):71–137, 2004.
51. M. Crainic and R. Fernandes. Stability of symplectic leaves. Invent. Math., 180(3):481–533,
2010.
52. M. Crainic and R. Fernandes. A geometric approach to Conn’s linearization theorem. Ann.
of Math. (2), 173(2):1121–1139, 2011.
53. M. Crainic and R. Fernandes. Lectures on integrability of Lie brackets. In Lectures on
Poisson geometry, volume 17 of Geom. Topol. Monogr., pages 1–107. Geom. Topol. Publ.,
Coventry, 2011.
54. P. Damianou and R. Fernandes. From the Toda lattice to the Volterra lattice and back. Rep.
Math. Phys., 50(3):361–378, 2002.
55. P. Damianou and R. Fernandes. Integrable hierarchies and the modular class. Ann. Inst.
Fourier (Grenoble), 58(1):107–137, 2008.
56. P. Damianou, H. Sabourin, and P. Vanhaecke. Transverse Poisson structures to adjoint orbits
in semisimple Lie algebras. Pacific J. Math., 232(1):111–138, 2007.
57. M. De Wilde and P. Lecomte. Existence of star-products and of formal deformations
of the Poisson Lie algebra of arbitrary symplectic manifolds. Lett. Math. Phys., 7(6):487–
496, 1983.
58. V. Drinfel’d. Hamiltonian structures on Lie groups, Lie bialgebras and the geometric meaning
of classical Yang-Baxter equations. Dokl. Akad. Nauk SSSR, 268(2):285–287, 1983.
59. V. Drinfel’d. Quantum groups. In Proceedings of the International Congress of Mathemati-
cians, Vol. 1, 2 (Berkeley, CA, 1986), pages 798–820, Providence, RI, 1987. Amer. Math.
Soc.
60. J.-P. Dufour and A. Haraki. Rotationnels et structures de Poisson quadratiques. C. R. Acad.
Sci. Paris Sér. I Math., 312(1):137–140, 1991.
61. J.-P. Dufour and M. Zhitomirskii. Classification of nonresonant Poisson structures. J. London
Math. Soc. (2), 60(3):935–950, 1999.
62. J.-P. Dufour and M. Zhitomirskii. Singularities and bifurcations of 3-dimensional Poisson
structures. Israel J. Math., 121:199–220, 2001.
63. J.-P. Dufour and N. T. Zung. Poisson structures and their normal forms, volume 242 of
Progress in Mathematics. Birkhäuser Verlag, Basel, 2005.
64. J.-P. Dufour and N.T. Zung. Linearization of Nambu structures. Compositio Math.,
117(1):77–98, 1999.
442 References

65. J. Duistermaat. On global action-angle coordinates. Comm. Pure Appl. Math., 33(6):687–
706, 1980.
66. J. Duistermaat and J. Kolk. Lie groups. Universitext. Springer-Verlag, Berlin, 2000.
67. Ch. Ehresmann. Les connexions infinitésimales dans un espace fibré différentiable. In
Colloque de topologie (espaces fibrés), Bruxelles, 1950, pages 29–55. Georges Thone,
Liège, 1951.
68. P. Etingof and V. Ginzburg. Noncommutative del Pezzo surfaces and Calabi-Yau algebras.
J. Eur. Math. Soc. (JEMS), 12(6):1371–1416, 2010.
69. P. Etingof and A. Varchenko. Geometry and classification of solutions of the classical dy-
namical Yang-Baxter equation. Comm. Math. Phys., 192(1):77–120, 1998.
70. P. Etingof and A. Varchenko. Solutions of the quantum dynamical Yang-Baxter equation and
dynamical quantum groups. Comm. Math. Phys., 196(3):591–640, 1998.
71. S. Evens, J.-H. Lu, and A. Weinstein. Transverse measures, the modular class and a coho-
mology pairing for Lie algebroids. Quart. J. Math. Oxford Ser. (2), 50(200):417–436, 1999.
72. L. Faddeev and L. Takhtajan. Hamiltonian methods in the theory of solitons. Classics in
Mathematics. Springer, Berlin, English edition, 2007. Translated from the 1986 Russian orig-
inal by A. Reyman.
73. B. Fedosov. A simple geometrical construction of deformation quantization. J. Differential
Geom., 40(2):213–238, 1994.
74. R. Fernandes. Connections in Poisson geometry. I. Holonomy and invariants. J. Differential
Geom., 54(2):303–365, 2000.
75. R. Fernandes. Lie algebroids, holonomy and characteristic classes. Adv. Math., 170(1):119–
179, 2002.
76. R. Fernandes and Ph. Monnier. Linearization of Poisson brackets. Lett. Math. Phys., 69:89–
114, 2004.
77. V. V. Fock and A. A. Rosly. Poisson structure on moduli of flat connections on Riemann
surfaces and the r-matrix. In Moscow Seminar in Mathematical Physics, volume 191 of Amer.
Math. Soc. Transl. Ser. 2, pages 67–86. Amer. Math. Soc., Providence, RI, 1999.
78. A. Fomenko. Integrability and nonintegrability in geometry and mechanics, volume 31 of
Mathematics and its Applications (Soviet Series). Kluwer Academic Publishers Group, Dor-
drecht, 1988. Translated from the Russian by M. V. Tsaplina.
79. J.-P. Françoise, G. L. Naber, and T. S. Tsun, editors. Encyclopedia of mathematical physics.
Vol. 1, 2, 3, 4, 5. Academic Press/Elsevier Science, Oxford, 2006.
80. W. Fulton and J. Harris. Representation theory, volume 129 of Graduate Texts in Mathemat-
ics. Springer-Verlag, New York, 1991. A first course, Readings in Mathematics.
81. I. Gel’fand and I. Dorfman. Hamiltonian operators and the classical Yang-Baxter equation.
Funktsional. Anal. i Prilozhen., 16(4):1–9, 96, 1982.
82. M. Gerstenhaber. On the deformation of rings and algebras. Ann. of Math. (2), 79:59–103,
1964.
83. V. Ginzburg. Momentum mappings and Poisson cohomology. Internat. J. Math., 7(3):329–
358, 1996.
84. V. Ginzburg. Coisotropic intersections. Duke Math. J., 140(1):111–163, 2007.
85. V. Ginzburg and D. Kaledin. Poisson deformations of symplectic quotient singularities. Adv.
Math., 186(1):1–57, 2004.
86. V. Ginzburg and A. Weinstein. Lie-Poisson structure on some Poisson Lie groups. J. Amer.
Math. Soc., 5(2):445–453, 1992.
87. J. Grabowski, G. Marmo, and P. W. Michor. Construction of completely integrable systems
by Poisson mappings. Modern Phys. Lett. A, 14(30):2109–2118, 1999.
88. J. Grabowski, G. Marmo, and A. M. Perelomov. Poisson structures: towards a classification.
Modern Phys. Lett. A, 8(18):1719–1733, 1993.
89. M. Gualtieri. Generalized complex geometry. PhD. Thesis, Oxford, 2004.
90. L. Guieu and C. Roger. L’algèbre et le groupe de Virasoro. Les Publications CRM, Montreal,
QC, 2007. Aspects géométriques et algébriques, généralisations. [Geometric and algebraic
aspects, generalizations], With an appendix by V. Sergiescu.
References 443

91. V. Guillemin. Moment maps and combinatorial invariants of Hamiltonian T n -spaces, vol-
ume 122 of Progress in Mathematics. Birkhäuser Boston Inc., Boston, MA, 1994.
92. V. Guillemin and S. Sternberg. Symplectic techniques in physics. Cambridge University
Press, Cambridge, second edition, 1990.
93. R. Hartshorne. Algebraic geometry. Springer-Verlag, New York, 1977. Graduate Texts in
Mathematics, No. 52.
94. N. Hitchin. Instantons, Poisson structures and generalized Kähler geometry. Comm. Math.
Phys., 265(1):131–164, 2006.
95. G. Hochschild and J.-P. Serre. Cohomology of Lie algebras. Ann. of Math. (2), 57:591–603,
1953.
96. J. Huebschmann. Poisson cohomology and quantization. J. Reine Angew. Math., 408:57–113,
1990.
97. J. Huebschmann. Poisson structures on certain moduli spaces for bundles on a surface. Ann.
Inst. Fourier (Grenoble), 45(1):65–91, 1995.
98. J. Huebschmann. Symplectic and Poisson structures of certain moduli spaces. I. Duke Math.
J., 80(3):737–756, 1995.
99. J. Humphreys. Introduction to Lie algebras and representation theory, volume 9 of Graduate
Texts in Mathematics. Springer-Verlag, New York, 1978. Second printing, revised.
100. C. Jacobi. Sur le mouvement d’un point et sur un cas particulier du problème des trois corps.
Compt. Rend., 3:59–61, 1836.
101. C. Jacobi. Note von der geodätischen linie auf einem ellipsoid und den verschiedenen an-
wendungen einer merkwürdigen analytischen substitution. J. Reine Angew. Math., 19:309–
313, 1839.
102. N. Jacobson. Lie algebras. Dover Publications Inc., New York, 1979. Republication of the
1962 original.
103. Ch. Kassel. Quantum groups, volume 155 of Graduate Texts in Mathematics. Springer-
Verlag, New York, 1995.
104. S. Khoroshkin, A. Radul, and V. Rubtsov. A family of Poisson structures on Hermitian sym-
metric spaces. Comm. Math. Phys., 152(2):299–315, 1993.
105. A. A. Kirillov. Unitary representations of nilpotent Lie groups. Uspehi Mat. Nauk,
17(4(106)):57–110, 1962.
106. A. A. Kirillov. Local Lie algebras. Uspehi Mat. Nauk, 31(4(190)):57–76, 1976.
107. M. Kontsevich. Deformation quantization of Poisson manifolds. Lett. Math. Phys.,
66(3):157–216, 2003 (q-alg/9709040, 1997).
108. Y. Kosmann-Schwarzbach. Quantum and classical Yang-Baxter equations. Modern Phys.
Lett. A, 5(13):981–990, 1990.
109. Y. Kosmann-Schwarzbach. From Poisson algebras to Gerstenhaber algebras. Ann. Inst.
Fourier (Grenoble), 46(5):1243–1274, 1996.
110. Y. Kosmann-Schwarzbach. Lie bialgebras, Poisson Lie groups and dressing transformations.
In Integrability of nonlinear systems (Pondicherry, 1996), volume 495 of Lecture Notes in
Phys., pages 104–170. Springer, Berlin, 1997.
111. Y. Kosmann-Schwarzbach. Odd and even Poisson brackets. In Quantum theory and symme-
tries (Goslar, 1999), pages 565–571. World Sci. Publ., River Edge, NJ, 2000.
112. Y. Kosmann-Schwarzbach. Quasi, twisted, and all that. . .in Poisson geometry and Lie al-
gebroid theory. In The breadth of symplectic and Poisson geometry, volume 232 of Progr.
Math., pages 363–389. Birkhäuser Boston, Boston, MA, 2005.
113. Y. Kosmann-Schwarzbach. Poisson manifolds, Lie algebroids, modular classes: a survey.
SIGMA Symmetry Integrability Geom. Methods Appl., 4:Paper 005, 30, 2008.
114. Y. Kosmann-Schwarzbach. The Noether theorems. Sources and Studies in the History of
Mathematics and Physical Sciences. Springer, New York, 2011. Invariance and conserva-
tion laws in the twentieth century, Translated, revised and augmented from the 2006 French
edition by Bertram E. Schwarzbach.
115. Y. Kosmann-Schwarzbach and F. Magri. Poisson-Nijenhuis structures. Ann. Inst. H. Poincaré
Phys. Théor., 53(1):35–81, 1990.
444 References

116. Y. Kosmann-Schwarzbach and A. Weinstein. Relative modular classes of Lie algebroids.

C. R. Math. Acad. Sci. Paris, 341(8):509–514, 2005.
117. B. Kostant. Orbits, symplectic structures and representation theory. In Proc. U.S.-Japan Sem-
inar in Differential Geometry (Kyoto, 1965), page 71. Nippon Hyoronsha, Tokyo, 1966.
118. J.-L. Koszul. Crochet de Schouten-Nijenhuis et cohomologie. Astérisque, (Numero Hors
Serie):257–271, 1985. The mathematical heritage of Élie Cartan (Lyon, 1984).
119. A. Kotov, P. Schaller, and Th. Strobl. Dirac sigma models. Comm. Math. Phys., 260(2):455–
480, 2005.
120. C. Laurent-Gengoux, E. Miranda, and P. Vanhaecke. Action-angle coordinates for integrable
systems on Poisson manifolds. Int. Math. Res. Not. IMRN, (8):1839–1869, 2011.
121. C. Laurent-Gengoux, M. Stiénon, and P. Xu. Holomorphic Poisson manifolds and holomor-
phic Lie algebroids. Int. Math. Res. Not. IMRN, pages Art. ID rnn 130, 30, 2008.
122. A. Le Blanc. Quasi-Poisson structures and integrable systems related to the moduli space of
flat connections on a punctured Riemann sphere. J. Geom. Phys., 57(8):1631–1652, 2007.
123. P. Lecomte and C. Roger. Modules et cohomologies des bigèbres de Lie. C. R. Acad. Sci.
Paris Sér. I Math., 310(6):405–410, 1990.
124. L.-C. Li and S. Parmentier. On dynamical Poisson groupoids. I. Mem. Amer. Math. Soc.,
174(824):vi+72, 2005.
125. P. Libermann and C.-M. Marle. Symplectic geometry and analytical mechanics, volume 35
of Mathematics and its Applications. D. Reidel Publishing Co., Dordrecht, 1987. Translated
from the French by B. E. Schwarzbach.
126. A. Lichnerowicz. Les variétés de Poisson et leurs algèbres de Lie associées. J. Differential
Geometry, 12(2):253–300, 1977.
127. A. Lichnerowicz. Variétés de Poisson et feuilletages. Ann. Fac. Sci. Toulouse Math. (5), 4(3–
4):195–262 (1983), 1982.
128. S. Lie and F. Engel. Theorie der transformationsgruppen. Teubner Verlag, Leipzig, 2, 1890.
129. Z.-J. Liu and P. Xu. On quadratic Poisson structures. Lett. Math. Phys., 26(1):33–42, 1992.
130. Z.-J. Liu and P. Xu. Dirac structures and dynamical r-matrices. Ann. Inst. Fourier (Grenoble),
51(3):835–859, 2001.
131. J.-L. Loday. Cyclic homology, volume 301 of Grundlehren der Mathematischen Wis-
senschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin,
1992. Appendix E by M. O. Ronco.
132. J.-H. Lu. Multiplicative and Affine Poisson structures on Lie groups. PhD. Thesis, University
of California at Berkeley, 1990.
133. J.-H. Lu. Momentum mappings and reduction of Poisson actions. In Symplectic geometry,
groupoids, and integrable systems (Berkeley, CA, 1989), volume 20 of Math. Sci. Res. Inst.
Publ., pages 209–226. Springer, New York, 1991.
134. J.-H. Lu. Coordinates on Schubert cells, Kostant’s harmonic forms, and the Bruhat Poisson
structure on G/B. Transform. Groups, 4(4):355–374, 1999.
135. J.-H. Lu. Classical dynamical r-matrices and homogeneous Poisson structures on G/H and
K/T . Comm. Math. Phys., 212(2):337–370, 2000.
136. J.-H. Lu and A. Weinstein. Poisson Lie groups, dressing transformations, and Bruhat decom-
positions. J. Differential Geom., 31(2):501–526, 1990.
137. J.-H. Lu and M. Yakimov. Group orbits and regular partitions of Poisson manifolds. Comm.
Math. Phys., 283(3):729–748, 2008.
138. A. Lundell. A short proof of the Frobenius theorem. Proc. Amer. Math. Soc., 116(4):1131–
1133, 1992.
139. K. Mackenzie. General theory of Lie groupoids and Lie algebroids, volume 213 of London
Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2005.
140. K. Mackenzie and P. Xu. Lie bialgebroids and Poisson groupoids. Duke Math. J., 73(2):415–
452, 1994.
141. F. Magri. A simple model of the integrable Hamiltonian equation. J. Math. Phys.,
19(5):1156–1162, 1978.
References 445

142. S. Majid. Foundations of quantum group theory. Cambridge University Press, Cambridge,
1995.
143. Ch.-M. Marle. The Schouten-Nijenhuis bracket and interior products. J. Geom. Phys., 23(3–
4):350–359, 1997.
144. J. Marsden and T. Ratiu. Reduction of Poisson manifolds. Lett. Math. Phys., 11(2):161–169,
1986.
145. J. Marsden and T. Ratiu. Introduction to mechanics and symmetry, volume 17 of Texts in
Applied Mathematics. Springer-Verlag, New York, second edition, 1999. A basic exposition
of classical mechanical systems.
146. J. Marsden and A. Weinstein. Reduction of symplectic manifolds with symmetry. Rep. Math-
ematical Phys., 5(1):121–130, 1974.
147. J. McKay. Graphs, singularities, and finite groups. In The Santa Cruz Conference on Finite
Groups (Univ. California, Santa Cruz, CA, 1979), volume 37 of Proc. Sympos. Pure Math.,
pages 183–186. Amer. Math. Soc., Providence, RI, 1980.
148. Kenneth R. Meyer. Symmetries and integrals in mechanics. In Dynamical systems (Proc.
Sympos., Univ. Bahia, Salvador, 1971), pages 259–272. Academic Press, New York, 1973.
149. E. Miranda and N.T. Zung. A note on equivariant normal forms of Poisson structures. Math.
Res. Lett., 13(5-6):1001–1012, 2006.
150. Ph. Monnier. Une cohomologie associée à une fonction. Applications aux cohomologies de
Poisson et de Nambu-poisson. PhD. Thesis, Université de Montpellier, 2001.
151. Ph. Monnier. Formal Poisson cohomology of quadratic Poisson structures. Lett. Math. Phys.,
59(3):253–267, 2002.
152. Ph. Monnier. Poisson cohomology in dimension two. Israel J. Math., 129:189–207, 2002.
153. Ph. Monnier and N. T. Zung. Levi decomposition for smooth Poisson structures. J. Differen-
tial Geom., 68(2):347–395, 2004.
154. D. Mumford, J. Fogarty, and F. Kirwan. Geometric invariant theory, volume 34 of Ergebnisse
der Mathematik und ihrer Grenzgebiete (2) [Results in Mathematics and Related Areas (2)].
Springer-Verlag, Berlin, third edition, 1994.
155. J. Nestruev. Smooth manifolds and observables, volume 220 of Graduate Texts in Mathe-
matics. Springer-Verlag, New York, 2003. Joint work of A. M. Astashov, A. B. Bocharov,
S. V. Duzhin, A. B. Sossinsky, A. M. Vinogradov and M. M. Vinogradov, Translated from
the 2000 Russian edition by Sossinsky, I. S. Krasil’schik and Duzhin.
156. A. Nijenhuis and R. Richardson. Deformations of Lie algebra structures. J. Math. Mech.,
17:89–105, 1967.
157. A. Odesskiı̆ and V. Rubtsov. Polynomial Poisson algebras with a regular structure of sym-
plectic leaves. Teoret. Mat. Fiz., 133(1):1321–1337, 2002.
158. P. Olver. Applications of Lie groups to differential equations, volume 107 of Graduate Texts
in Mathematics. Springer-Verlag, New York, second edition, 1993.
159. J.-P. Ortega and T. Ratiu. The optimal momentum map. In Geometry, mechanics, and dy-
namics, pages 329–362. Springer, New York, 2002.
160. J.-P. Ortega and T. Ratiu. Momentum maps and Hamiltonian reduction, volume 222 of
Progress in Mathematics. Birkhäuser Boston Inc., Boston, MA, 2004.
161. V. Ovsienko and K. Rozhe. Deformations of Poisson brackets and extensions of Lie algebras
of contact vector fields. Russian Math. Surveys, 47(6):135–191, 1992.
162. J.-S. Park. Topological open p-branes. In Symplectic geometry and mirror symmetry (Seoul,
2000), pages 311–384. World Sci. Publ., River Edge, NJ, 2001.
163. S. Pelap. Poisson (co)homology of polynomial Poisson algebras in dimension four:
Sklyanin’s case. J. Algebra, 322(4):1151–1169, 2009.
164. A. M. Perelomov. Integrable systems of classical mechanics and Lie algebras. Vol. I.
Birkhäuser Verlag, Basel, 1990. Translated from the Russian by A. G. Reyman.
165. D. Perrin. Algebraic geometry. Universitext. Springer-Verlag London Ltd., London, 2008.
An introduction, Translated from the 1995 French original by C. Maclean.
166. A. Pichereau. Poisson (co)homology and isolated singularities. J. Algebra, 299(2):747–777,
2006.
446 References

167. A. Pichereau. Formal deformations of Poisson structures in low dimensions. Pacific J. Math.,
239(1):105–133, 2009.
168. A. Pichereau and G. Van de Weyer. Double Poisson cohomology of path algebras of quivers.
J. Algebra, 319(5):2166–2208, 2008.
169. S. Poisson. Mémoire sur la variation des constantes arbitraires dans les questions de
mécanique. J. Ecole Polytec., 8:266–344, 1809.
170. A. Polishchuk. Algebraic geometry of Poisson brackets. J. Math. Sci. (New York),
84(5):1413–1444, 1997. Algebraic geometry, 7.
171. J. Pradines. Théorie de Lie pour les groupoı̈des différentiables. Calcul différentiel dans
la catégorie des groupoı̈des infinitésimaux. C. R. Acad. Sci. Paris Sér. A-B, 264:A245–
A248, 1967.
172. B. Przybylski. Complex Poisson manifolds. In Differential geometry and its applications
(Opava, 1992), volume 1 of Math. Publ., pages 227–241. Silesian Univ. Opava, Opava, 1993.
173. O. Radko. A classification of topologically stable Poisson structures on a compact oriented
surface. J. Symplectic Geom., 1(3):523–542, 2002.
174. C. Roger and P. Vanhaecke. Poisson cohomology of the affine plane. J. Algebra, 251(1):448–
460, 2002.
175. V. Roubtsov and T. Skrypnyk. Compatible Poisson brackets, quadratic Poisson algebras
and classical r-matrices. In Differential equations: geometry, symmetries and integrability,
volume 5 of Abel Symp., pages 311–333. Springer, Berlin, 2009.
176. P. Saksida. Lattices of Neumann oscillators and Maxwell-Bloch equations. Nonlinearity,
19(3):747–768, 2006.
177. J. A. Schouten. Ueber Differentialkomitanten zweier kontravarianter Grössen. Nederl. Akad.
Wetensch., Proc., 43:449–452, 1940.
178. J. A. Schouten. On the differential operators of first order in tensor calculus. Rapport ZA
1953-012. Math. Centrum Amsterdam, 1953.
179. M. Semenov-Tian-Shansky. What a classical r-matrix is. Functional Anal. Appl., 17(4):259–
272, 1983.
180. M. Semenov-Tian-Shansky. Dressing transformations and Poisson group actions. Publ. Res.
Inst. Math. Sci., 21(6):1237–1260, 1985.
181. J.-P. Serre. Local fields, volume 67 of Graduate Texts in Mathematics. Springer-Verlag, New
York, 1979. Translated from the French by M. J. Greenberg.
182. I. Shafarevich. Basic algebraic geometry. 1. Springer-Verlag, Berlin, second edition, 1994.
Varieties in projective space, Translated from the 1988 Russian edition and with notes by
M. Reid.
183. P. Slodowy. Simple singularities and simple algebraic groups, volume 815 of Lecture Notes
in Mathematics. Springer, Berlin, 1980.
184. J. Śniatycki and W. M. Tulczyjew. Generating forms of Lagrangian submanifolds. Indiana
Univ. Math. J., 22:267–275, 1972/73.
185. J.-M. Souriau. Quantification géométrique. Comm. Math. Phys., 1:374–398, 1966.
186. M. Spivak. Calculus on manifolds. A modern approach to classical theorems of advanced
calculus. W. A. Benjamin, Inc., New York-Amsterdam, 1965.
187. M. Spivak. A comprehensive introduction to differential geometry. Vol. I. Publish or Perish
Inc., Wilmington, DE, second edition, 1979.
188. T. A. Springer. Invariant theory. Lecture Notes in Mathematics, Vol. 585. Springer-Verlag,
Berlin, 1977.
189. D. E. Tamarkin. Operadic proof of M. Kontsevich’s formality theorem. ProQuest LLC, Ann
Arbor, MI, 1999. Thesis (Ph.D.)–The Pennsylvania State University.
190. P. Tauvel and R. Yu. Lie algebras and algebraic groups. Springer Monographs in Mathemat-
ics. Springer-Verlag, Berlin, 2005.
191. A. Thimm. Integrable geodesic flows on homogeneous spaces. Ergodic Theory Dynamical
Systems, 1(4):495–517 (1982), 1981.
References 447

192. W. M. Tulczyjew. Geometric formulations of physical theories, volume 11 of Monographs

and Textbooks in Physical Science. Lecture Notes. Bibliopolis, Naples, 1989. Statics and
dynamics of mechanical systems.
193. I. Vaisman. Remarks on the Lichnerowicz-Poisson cohomology. Ann. Inst. Fourier (Greno-
ble), 40(4):951–963 (1991), 1990.
194. I. Vaisman. Lectures on the geometry of Poisson manifolds, volume 118 of Progress in Math-
ematics. Birkhäuser Verlag, Basel, 1994.
195. I. Vaisman. Lectures on symplectic and Poisson geometry. Textos de Matemática.
Série B [Texts in Mathematics. Series B], 23. Universidade de Coimbra Departamento de
Matemática, Coimbra, 2000. Notes edited by Fani Petalidou.
196. P. Vanhaecke. Integrable systems in the realm of algebraic geometry, volume 1638 of Lecture
Notes in Mathematics. Springer-Verlag, Berlin, second edition, 2001.
197. A. M. Vinogradov. The union of the Schouten and Nijenhuis brackets, cohomology, and
superdifferential operators. Mat. Zametki, 47(6):138–140, 1990.
198. F. Warner. Foundations of differentiable manifolds and Lie groups, volume 94 of Gradu-
ate Texts in Mathematics. Springer-Verlag, New York, 1983. Corrected reprint of the 1971
edition.
199. A. Weinstein. The local structure of Poisson manifolds. J. Differential Geom., 18(3):523–
557, 1983.
200. A. Weinstein. Affine Poisson structures. Internat. J. Math., 1(3):343–360, 1990.
201. A. Weinstein. The modular automorphism group of a Poisson manifold. J. Geom. Phys.,
23(3-4):379–394, 1997.
202. A. Weinstein. Poisson geometry. Differential Geom. Appl., 9(1-2):213–238, 1998.
203. R. Wells. Differential analysis on complex manifolds, volume 65 of Graduate Texts in Math-
ematics. Springer, New York, third edition, 2008. With a new appendix by O. Garcia-Prada.
204. E. Whittaker. A treatise on the analytical dynamics of particles and rigid bodies. Cambridge
Mathematical Library. Cambridge University Press, Cambridge, 1988. With an introduction
to the problem of three bodies, Reprint of the 1937 edition, With a foreword by W. McCrea.
205. P. Xu. Poisson cohomology of regular Poisson manifolds. Ann. Inst. Fourier (Grenoble),
42(4):967–988, 1992.
206. P. Xu. Gerstenhaber algebras and BV-algebras in Poisson geometry. Comm. Math. Phys.,
200(3):545–560, 1999.
207. P. Xu. Dirac submanifolds and Poisson involutions. Ann. Sci. École Norm. Sup. (4),
36(3):403–430, 2003.
208. P. Xu. Momentum maps and Morita equivalence. J. Differential Geom., 67(2):289–333,
2004.
209. M. Yakimov. Symplectic leaves of complex reductive Poisson-Lie groups. Duke Math. J.,
112(3):453–509, 2002.
Index

action Poisson, 4, 368

adjoint, 120 symmetric, 371, 415
coadjoint, 120 tensor, 413
coordinates, 347 algebraic group, 116
dressing, 321 reductive, 116
Hamiltonian, 153 angle coordinates, 347
locally free, 119 anti-Poisson map, 295
Poisson, 148 Arnold’s theorem, 241
proper, 119 arrow, 400
action-angle theorem, 347 associative product, 418
adapted vector field, 436 atlas, 427
Ad-invariant
bilinear form, 121 basic function, 345
function, 122 bi-Hamiltonian
multivector, 121 hierarchy, 334
Ad∗ -invariant function, 122 manifold, 333
ad-invariant vector field, 333
bilinear form, 122 bialgebra
tensor, 276 Lie, 305
adjoint biderivation, 5
representation, 120 pointwise, 17
adjoint action, 120 skew-symmetric, 5
Adler–Kostant–Symes theorem, 337 bidual, 412
affine bilinear form
coordinate ring, 8 Ad-invariant, 121
Poisson variety, 8 ad-invariant, 122
subvariety, 47 invariant, 121
variety, 8 non-degenerate, 122, 182
algebra, 418 bivector, 163
exterior, 415 field, 16, 17
Gerstenhaber, 84 boundary operator
graded, 419, 420 Lie, 98
graded Lie, 420 Poisson, 99
homomorphism, 419 bracket
graded, 420 associated, 276
Lie, 114, 115, 418 Gerstenhaber, 357
Nambu–Poisson, 222 graded Lie, 420

C. Laurent-Gengoux et al., Poisson Structures, 449

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4, © Springer-Verlag Berlin Heidelberg 2013
450 Index

Lie, 418, 433 constants

linearized, 314 structure, 182
notation, 19 coordinate
Poisson, 4 affine
Schouten, 79 ring, 8
bundle chart, 427
tangent, 19 adapted, 49
expression, 428, 434
Campbell–Hausdorff formula, 379 coordinates
canonical coordinates, 166 action, 347
Cartan’s formula, 80 angle, 347
Casimir, 7 canonical, 166
centered coordinates, 427 Darboux, 26
centralizer, 227 local, 427
closed form, 72 log-canonical, 220
coadjoint splitting, 24
representation, 120 transverse, 347
coadjoint action, 120 coproduct, 422
coalgebra, 422 graded, 422
graded, 422 cotangent
homomorphism, 422 bundle, 175
coboundary space, 14, 16, 430
Hochschild, 358 cubic polynomial, 258
Lie, 92 curve
Lie bialgebra, 276 integral, 435
Poisson, 95
cocycle Darboux
group, 294 coordinates, 26
Hochschild, 358 theorem, 26
Lie, 92 de Rham
Poisson, 95 cohomology, 73, 75
coderivation differential, 72, 75
graded, 426 de-concatenation, 423
cohomology deformation
Chevalley–Eilenberg, 93 equivalent, 354, 368
de Rham, 73, 75 order k, 356
group, 294 formal, 354, 367
Hochschild, 358 order k, 356, 368
Lie algebra, 92 quantization, 362
Poisson, 95 degree, 206
coisotropic submanifold, 129 of homogeneous element, 419
commutator, 419 of linear map, 419
graded, 420 shifted, 79
compatible, 80 density
complete triple, 321 invariant, 105
complex manifold, 428 derivation, 4, 88
complex of de Rham, 72, 75 p-derivation, 64
conjugation, 117 exterior, 94
Conn’s theorem, 201 graded, 425
connection Hamiltonian, 6
symplectic, 366 inner, 94, 359
constant of motion, 330 pointwise, 429
constant Poisson structure, 165 Poisson, 23
Index 451

diagonal Poisson structure, 214 exponential map, 116, 377

diagonalizable extension
Poisson structure, 220 field, 56
vector field, 220 local, 137
differential, 16, 372, 430 exterior algebra, 415
de Rham, 72, 75
form, 74 field
graded Lie algebra, 372 bivector, 16, 17
morphism, 375 extension, 56
pointed, 373 multivector, 67
Kähler, 69 vector, 433
total, 375 flow
dimension, 14, 16, 427, 428 formal, 244
distribution, 436 of vector field, 436
holomorphic, 436 foliation
integrable, 29 open book, 189
involutive, 436 symplectic, 26, 29
singular, 29 form
smooth, 436 associated, 177
divergence differential, 74
of Poisson structure, 108 exact, 72
operator, 107 Kähler, 70
domain, 427 Liouville, 176
double modular, 192
Lie algebra, 270 volume, 103
of Lie bialgebra, 305 formal
dressing coordinate transformation, 237
action, 321 deformation, 354, 367
group, 322 flow, 244
Du Val singularity, 265 isomorphism, 238
dual Poisson structure, 237
module, 411 power series
of Lie bialgebra, 307 equivalent, 237
of Poisson–Lie group, 320 formality theorem, 399
Poisson–Lie groups, 320 formula
w.r.t. dressing, 322 Campbell–Hausdorff, 379
Frobenius’ theorem, 436
eigenvalue, 220 function
eigenvector, 220 basic, 345
embedded submanifold, 27, 49 Hamiltonian, 21
equation holomorphic, 428
Lax, 184 invariant, 121
Maurer–Cartan, 361, 371, 376 polynomial, 8
equivalent regular, 8
deformation, 354, 368 smooth, 428
order k, 356 structure, 10
gauge, 378 functions
path, 380 in involution, 26, 330
star product, 362 independent, 330
Euler involutive, 330
formula, 208 fundamental
vector field, 208 identity, 222
weighted, 212 vector field, 117
exact form, 72
452 Index

gauge holomorphic
equivalence, 380 distribution, 436
equivalent, 378 function, 428
generalized Lie derivative, 85 map, 429
germ, 429 vector field, 433
germification, 59 homogeneous
Gerstenhaber element, 419
algebra, 84 function, 207
bracket, 357 monomial, 372
graded Poisson structure, 207
algebra, 419, 420 homology
coalgebra, 422 Lie algebra, 98
coderivation, 426 Poisson, 99
commutator, 420 homomorphism
coproduct, 422 algebra, 419
derivation, 425 coalgebra, 422
Lie algebra, 82, 420 Lie algebra, 419
Lie bracket, 420 Lie bialgebra, 305
linear map, 419 Poisson–Lie group, 292
module, 419
product, 419 ideal, 6
gradient, 183 Lie, 6
graph Poisson, 6
Kontsevich, 400 Poisson–Dirac, 134
underlying, 400 identity
weight, 400 fundamental, 222
group Jacobi, 4
algebraic, 116 immersed submanifold, 49
cocycle, 294 immersion, 49
cohomology, 294 independent functions, 330
dressing, 322 inner derivation, 94, 359
Lie, 114 integrable distribution, 29
Poisson–Lie, 292 integrable system
representation, 116 Liouville, 341
integral curve, 435
Hadamard lemma, 431 internal product, 76, 416
Hamiltonian, 6 intertwining, 271
action, 153 invariant
derivation, 6 bilinear form, 121
function, 21 density, 105
local, 21 element, 93
path, 27 function, 121
piecewise, 27 left, 115
vector, 22 manifold, 341
field, 6, 21, 223 right, 115
head, 400 submanifold, 149
Hermitian metric, 177 subset, 149
hierarchy subvariety, 149
bi-Hamiltonian, 334 involution, 34
Hochschild functions in, 26, 330
coboundary, 358 involutive
cocycle, 358 distribution, 436
cohomology, 358 functions, 330
complex, 358
Index 453

isomorphism morphism, 114

formal, 238 ideal, 6
Poisson algebra, 5 module, 92
isotropic subspace, 281 sub-bialgebra, 306
subalgebra, 6
Jacobi identity, 4 subgroup, 114
Jacobiator, 34 theorem, 115
Jacobson–Morozov theorem, 227 Lie–Poisson structure
modified, 194
Kähler on g, 183
differential, 69 on g∗ , 181
form, 70 linear map
closed, 72 graded, 419
manifold, 177 linearizable Poisson structure, 198
metric, 177 order , 199
Kleinian singularity, 265 linearized
Kontsevich bracket, 314
formality theorem, 399 Poisson structure, 197
graph, 400 L∞ -morphism, 387
L∞ -quasi-isomorphism, 388
Lax Liouville
equation, 184, 336 form, 176
form, 336 integrable system, 341
left standard
invariant, 115 torus, 343
translation, 117 theorem, 342
lemma volume, 170
Hadamard, 431 local
Whitehead, 93 coordinates, 427
Lie extension, 137
algebra, 114, 115, 418 Hamiltonian, 21
cohomology, 92 localization, 57
differential graded, 372 locally free action, 119
differential graded pointed, 373 locus
double, 270 singular, 14, 22
graded, 82, 420 log-canonical coordinates, 220
homology, 98
homomorphism, 419 manifold
quadratic, 122 bi-Hamiltonian, 333
splitting, 272, 338 complex, 428
trivial cohomology, 93 invariant, 341
bialgebra, 305 Kähler, 177
coboundary, 276 Nambu–Poisson, 222
double of, 305 Poisson, 15, 19
dual of, 307 real, 434
homomorphism, 305 symplectic, 168
bracket, 418, 433 Manin triple, 310
graded, 420 map
coboundary, 92 co-momentum, 153
cocycle, 92 exponential, 377
derivative, 66, 68, 86 holomorphic, 429
generalized, 85 momentum, 153
group, 114 Poisson, 9, 20
dual, 320 smooth, 429
454 Index

symplectic, 171 path

tangent, 16, 430 equivalent, 380
twist, 413 polynomial, 380
matrix pencil
Poisson, 10, 13, 19, 33 Poisson, 334
Maurer–Cartan point
equation, 361, 371, 376 regular, 14, 22
metric singular, 14, 22
Hermitian, 177 smooth, 14, 55
Kähler, 177 pointed differential graded Lie
modular algebra, 373
class, 105 pointwise
form, 192 biderivation, 17
vector field, 104 derivation, 429
module multi-derivation, 67
Lie, 92 Poisson
momentum map, 153, 341 action, 148
monomial algebra, 4, 368
homogeneous, 372 isomorphism, 5
morphism morphism, 5
differential graded Lie algebra, 375 boundary operator, 99
Lie group, 114 bracket, 4
Poisson, 9 reduced, 124
Poisson algebra, 5 coboundary, 95
Moyal product, 364 cocycle, 95
Moyal–Weyl product, 364 cohomology, 95
multi-derivation, 64 derivation, 23
pointwise, 67 homology, 99
multiplicative, 292 ideal, 6, 47
Poisson structure, 215 manifold, 15, 19
system, 57 regular, 22, 165
multivector, 415 unimodular, 102, 105
Ad-invariant, 121 map, 9, 20
field, 67 matrix, 10, 13, 19, 33
homogeneous, 207 morphism, 9
polynomial, 206 pencil, 334
weight homogeneous, 212 product
bracket, 39
Nambu–Poisson manifold, 44
algebra, 222 structure, 41, 44
manifold, 222 variety, 41
structure, 222 rank, 14, 19, 22
nilpotent, 227 reducible, 124, 125, 127
orbit, 229 reduction, 123
Noether’s theorem, 333 algebraic, 123
normalizer, 124, 273 structure, 3, 8, 19
affine, 193
operator canonical on g∗ , 181
bi-differential, 362 constant, 162, 165
multi-differential, 362 degree of, 206
orbit diagonal, 214
nilpotent, 229 diagonalizable, 220
order of power series, 240 divergence of, 108
orthogonal, 183 formal, 237
Index 455

germ of, 60 proper action, 119

homogeneous, 207, 209 pushforward, 18, 433
linear, 181
linearizable, 198 quadratic
linearizable order , 199 Lie algebra, 122
linearization, 196 Poisson structure, 213
linearized, 197 quantization
modified canonical, 194 deformation, 362
multiplicative, 215 quotient
polynomial, 32, 206 Poisson structure, 174
quadratic, 213 Poisson variety, 174
quotient, 174
reduced, 123, 126, 128, 135 rank
regular, 165 locally constant, 22
stable, 267 maximal, 22
transverse, 147 Poisson, 14, 19, 22
unimodular, 102 R-bracket, 273
weight homogeneous, 213 real manifold, 434
subalgebra, 6 reduced
submanifold, 49 Poisson structure, 123, 126, 128, 135
subvariety, 10, 48 Poisson bracket, 124
tensor product, 39 reducible
theorem, 34, 332 Poisson, 124, 125, 127
variety, 7, 8 reduction
affine, 8 Poisson, 123
quotient, 174 algebraic, 123
vector field, 23 Poisson–Dirac, 134
vector space, 165 algebraic, 134
Poisson–Dirac reductive algebraic group, 116
ideal, 134 regular
reduction, 134 function, 8
algebraic, 134 point, 14, 22
submanifold, 137 Poisson manifold, 22, 165
subvariety, 135 Poisson structure, 165
Poisson–Lie representation, 92
group, 292 adjoint, 120
dual, 320 coadjoint, 120
dual w.r.t. dressing, 322 group, 116
homomorphism, 292 space, 92
subgroup, 296 right
polynomial invariant, 115
cubic, 258 translation, 117
function, 8 R-matrix, 270
path, 380 r-matrix, 276
Poisson structure, 32, 206 Russian formula
power series first, 279
order of, 240 second, 287
product, 418
graded, 419 S-triple, 227
internal, 76, 416 Schouten bracket, 79
Moyal–Weyl, 364 algebraic, 85
star, 362 shifted degree, 79
tensor, 412 shuffle, 65, 416
wedge, 65, 74
456 Index

singular symmetric
distribution, 29 algebra, 371, 415
locus, 14, 22 tensor, 416
point, 14, 22 symmetrization, 416
singularity symplectic
Du Val, 265 connection, 366
Klein, 265 foliation, 26, 27, 29
simple, 241, 243 leaf, 29
skew-symmetric manifold, 168
biderivation, 5 almost, 168
tensor, 416 exact, 177
skew-symmetrization, 416 map, 171
smooth structure, 167, 168
distribution, 436 almost, 168
function, 428 vector space, 167
map, 429 symplectic connection, 366
point, 14, 55
vector field, 433 tail, 400
space tangent, 48, 49
cotangent, 14, 16, 430 bundle, 19
representation, 92 map, 16, 430
tangent, 14, 16, 49, 430 space, 14, 16, 49, 430
splitting tensor
coordinates, 24 algebra, 413
Lie algebra, 272, 338 product, 39, 412
theorem, 23 skew-symmetric, 416
star product, 362 symmetric, 416
equivalent, 362 theorem
straightening theorem, 436 action-angle, 347
structure Adler–Kostant–Symes, 337
constants, 182 Arnold, 241
function, 10 Conn’s linearization, 201
Nambu–Poisson, 222 Darboux, 26
Poisson, 8, 19 Frobenius, 436
symplectic, 167, 168 Jacobson–Morozov, 227
subalgebra, 6 Kontsevich’s formality, 399
Lie, 6 Lie, 115
Poisson, 6 Liouville, 342
subgroup Noether, 333
Lie, 114 Poisson, 34, 332
Poisson–Lie, 296 splitting, 23
submanifold, 49 straightening, 436
coisotropic, 129 Weinstein, 23
embedded, 27, 49 Thimm’s method, 335
immersed, 27, 49 total differential, 375
Poisson, 49 transition map, 427
Poisson–Dirac, 137 translation
submersion, 127 left, 117
subspace right, 117
isotropic, 281 transverse
subvariety coordinates, 347
affine, 47 Poisson structure, 147
Poisson, 10, 48 submanifolds, 143
Poisson–Dirac, 135 triple
Index 457

complete, 321 smooth, 433

twist map, 413 Hamiltonian, 22
type, 264 space
Poisson, 165
unimodular symplectic, 167
Lie algebra, 192 vertex, 400
Poisson manifold, 102, 105 volume
Poisson structure, 102 form, 103
Liouville, 170
variety
affine, 8 wedge product, 65, 74
Poisson, 7, 8 weight, 208, 212
vector graph, 400
field, 433 of function, 211
adapted, 436 vector, 211
bi-Hamiltonian, 333 weight homogeneous
diagonalizable, 220 function, 211
Euler, 208 multivector field, 212
Euler weighted, 212 Poisson structure, 213
flow of, 436 weighted
fundamental, 117 Euler formula, 212
Hamiltonian, 6, 21, 223 Euler vector field, 212
holomorphic, 433 Weinstein’s theorem, 23
locally Hamiltonian, 21 Whitehead’s lemma, 93
modular, 104
Poisson, 23 Yang–Baxter equation, 271
polynomial, 206 modified, 271
List of Notations

F, 3 Pm [F, G], 17
{· , ·}, 4 T M, 19
{F, G}, 4 Rkm π , 22
, 4 Rk π , 22
X1 (A ), 5 LV , 23
[V , W ], 5 o, 25
P[F, G], 5 1r , 26
V [F], 5 Sm (M), 27
X2 (A ), 5 F (V ), 32
XH , 6 X1 (M), 32
Ham(A , {· , ·}), 6 X2 (M), 32
Ham(A ), 6 J , 34
Cas(A , {· , ·}), 7 A1 ⊗ A2 , 39
Cas(A ), 7 N sm , 55
F (M), 8 [· , ·], 56
Ψ ∗, 9 [F, G], 56
dF, 10 P[F1 , . . . , Fp ], 64
Matd (A ), 10 X p (A ), 64
Rkm {· , ·}, 14 X• (A ), 64
Rk {· , ·}, 14 S p,q , 65
Tm M, 14 sgn, 65
Tm∗ M, 14 P ∧ Q, 65
πm , 14 ıF , 66
M(s) , 15 LV , 66
dim, 16 P[F1 , . . . , Fp ], 67
F (M), 16 Ω 1 (A ), 69
Tm M, 16 dF, 69
Fm , 16 Ω p (A ), 70
Tm∗ M, 16 Ω • (A ), 70
· , ·, 16 ω ∧ η , 71
ξ , v, 16 HdR p
(A ), 73
TmΨ , 16 •
HdR (A ), 73
dm F, 16 [ω ], 73
X1 (M), 16 Ω p (M), 74
Vm , 16 Ω • (M), 74
V [F], 16 d, 74
p
Vm [F], 17 HdR (M), 75

C. Laurent-Gengoux et al., Poisson Structures, 459

Grundlehren der mathematischen Wissenschaften 347,
DOI 10.1007/978-3-642-31090-4, © Springer-Verlag Berlin Heidelberg 2013
460 List of Notations

HdR • (M), 75 mod ν k , 354

ıP , 76 Fνk , 355
[· , ·]S , 79 Akν , 355
[P, Q]S , 79 μ(k) , 355
X p (A ), 79 HCk (V ), 357
p, 79 HC• (V ), 357
[[· , ·]], 85 HCk (V ), 357
LP , 85 HC• (V ), 357
δLk , 91 [· , ·]G , 357
δπk , 91 δμk , 358
∂kL , 91 δμk , 358
∂kπ , 91 HH•μ (A ), 358
HLk (g;V ), 91 HHkμ (A ), 358
Hπk (A ), 91 sgn(σ ; i1 , . . . , ik ), 372
HkL (g;V ), 91 d 0 (x), 372
Hkπ (A ), 91 g, 372
δLk , 92 Dz , 373
HLk (g;V ), 92 HC•diff (M), 373
HL• (g;V ), 92 HD• (g), 374
V g , 93 HDk (g), 374
HL• (g), 93 ΛD , 374
HCE • (g), 93 Λ[· ,·] , 374
δπk , 94 Λ , 374
Hπq (A ), 95 MC(g), 376
Hπ• (A ), 95 Obsk (x), 376
Hπq (M), 95 MCk (g), 376
Hπ• (M), 95 ex , 377
π  , 96 x ∼ y, 378
∂kL , 98 eadξ , 378
HkL (g;V ), 98 ξ  x, 378
H•L (g;V ), 98 CH(u, v), 379
∂kπ , 99 cl(x), 380
Hkπ (A ), 100 Φ (X), 387
H•π (A ), 100 Si1 ,...,i , 387
Φλ , 104 Ω̃Φ (x), 389
Div, 107 ΩΦ (cl(x)), 389
P, 107 h(a), 400
1X , 114 t(a), 400
F (M)G , 119 Gk, , 400
g∗ , 122 HomR (V,W ), 411
πm , 130 Hom(V,W ), 411
Tn⊥ N, 130 V ∗ , 411
δi, j , 144 V ⊗R W , 412
SP(V ), 174 V ⊗W , 412
ϖ (F), 211 v ⊗ w, 412
ϖ (P), 212 S, 413
d, 302 V ⊗k , 413
G∗ , 322 T •V , 413
π ∗ , 322 φ ⊗ 1V , 413
F, 330 φ ⊗ ψ , 414
UF , 330 ∧•V , 415
Fν , 354 X ∧Y , 415
A ν , 354 S•V , 415
List of Notations 461

Sk , 415 δ , 423
sgn(σ ), 416 , 424
ıX , 416 Derr (V ), 425
ıφ , 416 CoDerr (V ), 426
Si, j , 416 F (M), 428
φ ∧ ψ , 417 Fm , 429
Si, j,k , 417 Tm M, 430
S , 418
Tm∗ M, 430
V • , 419
· , ·, 430
V• , 419
V• → W•+r , 419 ξ , v, 430
Homr (V,W ), 419 dm F, 430
μi, j , 419 TmΨ , 430
[· , ·], 420 V [F], 433
T • φ , 421 Vm [F], 433
Δ , 422 X1 (M), 433
Δi, j , 422 Ψ∗ V , 433