Graduate Texts in Mathematics 155

Editorial Board
I.H. Ewing F.W. Gehring P.R. Halmos
Graduate Texts in Mathematics

TAKEUTIIZARING. Introduction to Axiomatic 33 HIRSCH. Differential Topology.

Set Theory. 2nd ed. 34 SPITZER. Principles of Random Walk. 2nd ed.
2 OXTOBY. Measure and Category. 2nd ed. 35 WERMER. Banach Algebras and Several
3 SCHAEFFER. Topological Vector Spaces. Complex Variables. 2nd ed.
4 HILTON/STAMMBACH. A Course in 36 KELLEy/NAMIOKA et al. Linear Topological
Homological Algebra. Spaces.
5 MAC LANE. Categories for the Working 37 MONK. Mathematical Logic.
Mathematician. 38 GRAUERT/FRITZSCHE. Several Complex
6 HUGHES/PIPER. Projective Planes. Variables.
7 SERRE. A Course in Arithmetic. 39 ARVESON. An Invitation to C*-Algebras.
8 TAKEUTI/ZARING. Axiomatic Set Theory. 40 KEMENy/SNELL/KNAPP. Denumerable Markov
9 HUMPHREYS. Introduction to Lie Algebra~ Chains. 2nd ed.
and Representation Theory. 41 ApOSTOL. Modular Functions and Dirichlet
10 COHEN. A Course in Simple Homotopy Series in Number Theory. 2nd ed.
Theory. 42 SERRE. Linear Representations of Finite
11 CONWAY. Functions of One Complex Groups.
Variable. 2nd ed. 43 GILLMAN/JERISON. Rings of Continuous
12 BEALS. Advanced Mathematical Analysis. Functions.
I3 ANDERSON/FULLER. Rings and Categories of 44 KENDIG. Elementary Algebraic Geometry.
Modules. 2nd ed. 45 LoEVE. Probability Theory I. 4th ed.
14 GOLUBITSKy/GUILEMIN. Stable Mappings and 46 LoEvE. Probability Theory II. 4th ed.
Their Singularities. 47 MOISE. Geometric Topology in Dimensions 2
15 BERBERIAN. Lectures in Functional Analysis and 3.
and Operator Theory. 48 SACHSlWu. General Relativity for
16 WINTER. The Structure of Fields. Mathematicians.
17 ROSENBLATT. Random Processes. 2nd ed. 49 GRUENBERGIWEIR. Linear Geometry. 2nd ed.
18 HALMos. Measure Theory. 50 EDWARDS. Fermat's Last Theorem.
19 HALMos. A Hilbert Space Problem Book. 51 KLINGENBERG. A Course in Differential
2nd ed. Geometry.
20 HUSEMOLLER. Fibre Bundles. 3rd ed. 52 HARTSHORNE. Algebraic Geometry.
21 HUMPHREYS. Linear Algebraic Groups. 53 MANIN. A Course in Mathematical Logic.
22 BARNES/MACK. An Algebraic Introduction to 54 GRAVERIWATKINS. Combinatorics with
Mathematical Logic. Emphasis on the Theory of Graphs.
23 GREUB. Linear Algebra. 4th ed. 55 BROWN/PEARCY. Introduction to Operator
24 HOLMES. Geometric Functional Analysis and Theory I: Elements of Functional Analysis.
Its Applications. 56 MASSEY. Algebraic Topology: An
25 HEWITT/STROMBERG. Real and Abstract Introduction.
Analysis. 57 CROWELL/Fox. Introduction to Knot Theory.
26 MANES. Algebraic Theories. 58 KOBUTZ. p-adic Numbers, p-adic Analysis,
27 KELLEY. General Topology. and zeta-Functions. 2nd ed.
28 ZARISKIISAMUEL. Commutative Algebra. 59 LANG. Cyclotomic Fields.
Vol.l. 60 ARNOLD. Mathematical Methods in Classical
29 ZARISKIISAMUEL. Commutative Algebra. Mechanics. 2nd ed.
Vol.lI. 61 WHITEHEAD. Elements of Homotopy Theory.
30 JACOBSON. Lectures in Abstract Algebra I. 62 KARGAPOLOVIMERLZJAKOv. Fundamentals of
Basic Concepts. the. Theory of Groups.
31 JACOBSON. Lectures in Abstract Algebra II. 63 BOLLOBAS. Graph Theory.
Linear Algebra. 64 EDWARDS. Fourier Series. Vol. I. 2nd ed.
32 JACOBSON. Lectures in Abstract Algebra III.
Theory of Fields and Galois Theory. continued after index
Christian Kassel

Quantum Groups
With 88 Illustrations

Springer-Science+Business Media, LLC

Christian Kassel
Institut de Recherche Mathematique Avancee
Universite Louis Pasteur-C.N.R.S.
67084 Strasbourg
France

Editorial Board
J.H. Ewing F. W. Gehring P.R. Halmos
Department of Department of Department of
Mathematics Mathematics Mathematics
Indiana University University of Michigan Santa Clara University
Bloomington, IN 47405 Ann Arbor, MI 48109 Santa Clara, CA 95053
USA USA USA

Mathematics Subject Classification (1991): Primary-17B37, 18DlO, 57M25, 81R50;

Secondary-16W30, 17B20, 17B35, 18D99, 20F36

Library of Congress Cataloging-in-Publication Data

Kassel, Christian.
Quantum groups/Christian Kassel.
p. cm. - (Graduate texts in mathematics; voI. 155)
Includes bibliographical references and index.
ISBN 978-1-4612-6900-7 ISBN 978-1-4612-0783-2 (eBook)
DOI 10.1007/978-1-4612-0783-2
1. Quantum groups. 2. Hopf algebras. 3. Topology.
4. Mathematical physics. 1. Title. II. Series: Graduate texts in
mathematics; 155.
QC20.7.G76K37 1995
512'.55-dc20 94-31760

Printed on acid-free paper.

1995 Springer Science+Business Media New York

Originally published by Springer-Verlag New York, Inc in 1995
Softcover reprint of the hardcover 1st edition 1995
All rights reserved. This work may not be translated or copied in whole or in part without the written
permission ofthe publisher (Springer Science+Business Media, LLC), except for brief excerpts in
connection with reviews or scholarly anaJysis. Use in connection with any form of information storage
and retrievaJ, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even
if the former are not especially identified, is not to be taken as a sign that such names, as
understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely
byanyone.

Production managed by Francine McNeill; manufacturing supervised by Genieve Shaw.

Photocomposed pages prepared using Patrick D.F. Ion's TeX files.

987654321

ISBN 978-1-4612-6900-7
Preface

{( Eh bien, Monsieur, que

pensez-vous des x et des y ?
Je lui ai repondu :
{( C'est bas de plafond.
V. Hugo [Hug51]

The term "quantum groups" was popularized by Drinfeld in his address to

the International Congress of Mathematicians in Berkeley (1986). It stands
for certain special Hopf algebras which are nontrivial deformations of the
enveloping Hopf algebras of semisimple Lie algebras or of the algebras of
regular functions on the corresponding algebraic groups. As was soon ob-
served, quantum groups have close connections with varied, a priori remote,
areas of mathematics and physics.
The aim of this book is to provide an introduction to the algebra behind
the words "quantum groups" with emphasis on the fascinating and spec-
tacular connections with low-dimensional topology. Despite the complexity
of the subject, we have tried to make this exposition accessible to a large
audience. We assume a standard knowledge of linear algebra and some
rudiments of topology (and of the theory of linear differential equations as
far as Chapter XIX is concerned).
We divided the book into four parts we now briefly describe. In Part I
we introduce the language of Hopf algebras and we illustrate it with the
Hopf algebras SLq(2) and Uq(.s((2)) associated with the classical group
8L 2 . These are the simplest examples of quantum groups, and actually the
only ones we treat in detail. Part II focuses on two classes of Hopf algebras
that provide solutions of the Yang-Baxter equation in a systematic way. We
review a method due to Faddeev, Reshetikhin, and Takhtadjian as well as
Drinfeld's quantum double construction, both designed to produce quan-
tum groups. Parts I and II may form the core of a one-year introductory
course on the subject.
Parts III and IV are devoted to some of the spectacular connections
alluded to before. The avowed objective of Part III is the construction of
isotopy invariants of knots and links in R 3 , including the Jones polynomial,
VI Preface

from certain solutions of the Yang-Baxter equation. To this end, we intro-

duce various classes of tensor categories that are responsible for the close
relationship between quantum groups and knot theory. Part IV presents
more advanced material: it is an account of Drinfeld's elegant treatment of
the monodromy of the Knizhnik-Zamolodchikov equations. Our aim is to
highlight Drinfeld's deep result expressing the braided tensor category of
modules over a quantum enveloping algebra in terms of the corresponding
semisimple Lie algebra. We conclude the book with the construction of a
"universal knot invariant". This is a nice, far-reaching application of the
algebraic techniques developed in the preceding chapters.
I wish to acknowledge the inspiration I drew during the composition of
this text from [Dri87] [Dri89a] [Dri89b] [Dri90] by Drinfeld, [JS93] by Joyal
and Street, [Tur89] [RT90] by Reshetikhin and Turaev. After having become
acquainted with quantum groups, the reader is encouraged to return to
these original sources. Further references are given in the notes at the end
of each chapter. Lusztig's and Turaev's monographs [Lus93] [Tur94] may
complement our exposition advantageously.
This book grew out of two graduate courses I taught at the Department
of Mathematics of the Universite Louis Pasteur in Strasbourg during the
years 1990-92. Part I is the expanded English translation of [Kas92]. It is a
pleasure to express my thanks to C. Bennis, R. Berger, C. Mitschi, P. Nuss,
C. Reutenauer, M. Rosso, V. Turaev, M. Wambst for valuable discussions
and comments, and to Raymond Seroul who coded the figures. lowe special
thanks to Patrick Ion for his marvellous job in preparing the book for
printing, with his attention to mathematical, English, typographical, and
computer details.

Christian Kassel
March 1994, Strasbourg

Notation. - Throughout the text, k is a field and the words "vector

space", "linear map" mean respectively "k-vector space" and "k-linear
map". The boldface letters N, Z, Q, R, and C stand successively for the
nonnegative integers, all integers, the field of rational, real, and complex
numbers. The Kronecker symbol l5ij is defined by l5 ij = 1 if i = j and is
zero otherwise. We denote the symmetric group on n letters by Sri' The
sign of a permutation u is indicated by c(u).
The symbol 0 indicates the end of a proof. Roman figures refer to the
numbering of the chapters.
Contents

Preface v

Part One Quantum 8L(2) 1

I
Preliminaries 3
1 Algebras and Modules . 3
2 Free Algebras . . . . . . 7
3 The Affine Line and Plane 8
4 Matrix Multiplication . . . 10
5 Determinants and Invertible Matrices 10
6 Graded and Filtered Algebras 12
7 Ore Extensions . . 14
8 Noetherian Rings 18
9 Exercises 20
10 Notes ....... 22

II
Tensor Products 23
1 Tensor Products of Vector Spaces 23
2 Tensor Products of Linear Maps 26
3 Duality and Traces . . . . . . . . 29
4 Tensor Products of Algebras .. 32
5 Tensor and Symmetric Algebras 34
6 Exercises 36
7 Notes ............... 38
viii Contents

III
The Language of Hopf Algebras 39
1 Coalgebras.. 39
2 Bialgebras . . . . . . . . . . . . 45
3 Hopf Algebras . . . . . . . . . . 49
4 Relationship with Chapter I. The Hopf Algebras GL(2)
and SL(2). . . . . . . . . . . 57
5 Modules over a Hopf Algebra. . . . . . . . . . . . . . . 57
6 Comodules......................... 61
7 Comodule-Algebras. Coaction of SL(2) on the Affine Plane 64
8 Exercises 66
9 Notes.............................. 70

IV
The Quantum Plane and Its Symmetries 72
1 The Quantum Plane . . . . . . . . . . . . . 72
2 Gauss Polynomials and the q- Binomial Formula 74
3 The Algebra Mq(2) . . . . . . . . . . 77
4 Ring-Theoretical Properties of Mq(2) . 81
5 Bialgebra Structure on Mq(2) . . . . . . 82
6 The Hopf Algebras GLq(2) and SLq(2) 83
7 Coaction on the Quantum Plane 85
8 Hopf *-Algebras 86
9 Exercises 88
10 Notes . . . . . . 90

V
The Lie Algebra of SL(2) 93
1 Lie Algebras . . . . . 93
2 Enveloping Algebras . . 94
3 The Lie Algebra .5[(2) . 99
4 Representations of .5[(2) 101
5 The Clebsch-Gordan Formula. 105
6 Module-Algebra over a Bialgebra. Action of .5[(2) on the
Affine Plane . . . . . . . . . . . . . . . . . . . . . . . 107
7 Duality between the Hopf Algebras U(.5[(2)) and SL(2) 109
8 Exercises 11 7
9 Notes............................ 119

VI
The Quantum Enveloping Algebra of .5[(2) 121
1 The Algebra Uq(.5[(2)) . . . . . . . . . . . . . 121
2 Relationship with the Enveloping Algebra of .5[(2) 125
3 Representations of Uq . . . . . . . . . . . . . . . . 127
4 The Harish-Chandra Homomorphism and the Centre of Uq 130
 Contents ix

5 Case when q is a Root of Unity. 134

6 Exercises 138
7 Notes . . . . . . . . . . . . . . . 138

VII
A Hopf Algebra Structure on Uis[(2)) 140
1 Comultiplication............. 140
2 Semi simplicity . . . . . . . . . . . . . . 143
3 Action of Uq(.s[(2)) on the Quantum Plane 146
4 Duality between the Hopf Algebras Uq(.s[(2)) and SLq(2) 150
5 Duality between Uq (.s[(2))-Modules and SL q(2)-Comodules 154
6 Scalar Products on Uq (.s[(2))-Modules 155
7 Quantum Clebsch-Gordan . 157
8 Exercises 162
9 Notes............ 163

Part Two Universal R-Matrices 165

VIII
The Yang-Baxter Equation and (Co)Braided Bialgebras 167
1 The Yang-Baxter Equation . . . . . . . . . . . . 167
2 Braided Bialgebras. . . . . . . . . . . . . . . . . . . . . 172
3 How a Braided Bialgebra Generates R- Matrices . . . . 178
4 The Square of the Antipode in a Braided Hopf Algebra 179
5 A Dual Concept: Cobraided Bialgebras 184
6 The FRT Construction . . . . . . 188
7 Application to GLq(2) and SLq(2) 194
8 Exercises 196
9 Notes................ 198

IX
Drinfeld's Quantum Double 199
1 Bicrossed Products of Groups . . . . . . 199
2 Bicrossed Products of Bialgebras . . . . . 202
3 Variations on the Adjoint Representation 207
4 Drinfeld's Quantum Double . . . . . . . . 213
5 Representation-Theoretic Interpretation of the
Quantum Double . . . . 220
6 Application to Uq(.s[(2)) . 223
7 R- Matrices for U q 230
8 Exercises 236
9 Notes . . . . . . . 238
x Contents

Part Three Low-Dimensional Topology and 239

Tensor Categories
x
Knots, Links, Tangles, and Braids 241
1 Knots and Links . . . . . . . . . . . 242
2 Classification of Links up to Isotopy 244
3 Link Diagrams . . . . . . . . . 246
4 The Jones-Conway Polynomial 252
5 Tangles. 257
6 Braids .. 262
7 Exercises 269
8 Notes .. 270
9 Appendix. The Fundamental Group 273

XI
Tensor Categories 275
1 The Language of Categories and Functors . 275
2 Tensor Categories . . . . . . . 281
3 Examples of Tensor Categories . . . . . . . 284
4 Tensor Functors . . . . . . . . . . . . . . . 287
5 Turning Tensor Categories into Strict Ones 288
6 Exercises 291
7 Notes . . . . . . . . . . . . . . . . . . . . . 293

XII
The Tangle Category 294
1 Presentation of a Strict Tensor Category 294
2 The Category of Tangles . . . . . . . . . 299
3 The Category of Tangle Diagrams . . . . 302
4 Representations of the Category of Tangles 305
5 Existence Proof for Jones-Conway Polynomial 311
6 Exercises 313
7 Notes . . . . . . . . . . . . . . . . . . . . . . . 313

XIII
Braidings 314
1 Braided Tensor Categories . . . . . 314
2 The Braid Category . . . . . . . . . 321
3 Universality of the Braid Category. 322
4 The Centre Construction . . . . . . 330
5 A Categorical Interpretation of the Quantum Double 333
6 Exercises 337
7 Notes........................... 338
 Contents xi

XIV
Duality in Tensor Categories 339
1 Representing Morphisms in a Tensor Category 339
2 Duality . . . . . . . . . . . . . . 342
3 Ribbon Categories . . . . . . . . 348
4 Quantum Trace and Dimension . 354
5 Examples of Ribbon Categories. 358
6 Ribbon Algebras 361
7 Exercises 365
8 Notes . . . . . . 366

XV
Quasi-Bialgebras 368
1 Quasi-Bialgebras . . . . . 368
2 Braided Quasi-Bialgebras 371
3 Gauge Transformations . 372
4 Braid Group Representations . 377
5 Quasi-Hopf Algebras. 379
6 Exercises 381
7 Notes . . . . . . . . . 381

Part Four Quantum Groups and Monodromy 383

XVI
Generalities on Quantum Enveloping Algebras 385
1 The Ring of Formal Series and h-Adic Topology 385
2 Topologically Free Modules . 388
3 Topological Tensor Product .. 390
4 Topological Algebras . . . . . 392
5 Quantum Enveloping Algebras 395
6 Symmetrizing the Universal R-Matrix 398
7 Exercises . . . . . . . . . 400
8 Notes . . . . . . . . . . . 401
9 Appendix. Inverse Limits 401

XVII
Drinfeld and Jimbo's Quantum Enveloping Algebras 403
1 Semisimple Lie Algebras . . . . . . 403
2 Drinfeld-Jimbo Algebras . . . . . . 406
3 Quantum Group Invariants of Links 410
4 The Case of s[(2) 412
5 Exercises 418
6 Notes....... 418
xii Contents

XVIII
Cohomology and Rigidity Theorems 420
1 Cohomology of Lie Algebras . . . . . . . . . . 420
2 Rigidity for Lie Algebras . . . . . . . . . . . . 424
3 Vanishing Results for Semisimple Lie Algebras 427
4 Application to Drinfeld-Jimbo Quantum Enveloping Algebras 430
5 Cohomology of Coalgebras . . . . . . . . . . . . . . . . .. 431
6 Action of a Semisimple Lie Algebra on the Cobar Complex 434
"1 Computations for Symmetric Coalgebras . . . . . . . . 435
8 Uniqueness Theorem for Quantum Enveloping Algebras 442
9 Exercises................. 446
10 N o t e s . . . . . . . . . . . . . . . . . . . 446
11 Appendix. Complexes and Resolutions. 447

XIX
Monodromy of the Knizhnik-Zamolodchikov Equations 449
1 Connections . . . . . . . . . . . . . . . . . . . . 449
2 Braid Group Representations from Monodromy . 451
3 The Knizhnik-Zamolodchikov Equations. 455
4 The Drinfeld-Kohno Theorem 458
5 Equivalence of Uh(g) and Ag,t . . . . . . 461
6 Drinfeld's Associator . . . . . . . . . . . 463
7 Construction of the Topological Braided Quasi-Bialgebra Ag,t 468
8 Verification of the Axioms 471
9 Exercises . . . . . . . . . . . 479
10 N o t e s . . . . . . . . . . . . . 479
11 Appendix. Iterated Integrals 480

XX
Postlude. A Universal Knot Invariant 484
1 Knot Invariants of Finite Type . . . . . 484
2 Chord Diagrams and Kontsevich's Theorem. 486
3 Algebra Structures on Chord Diagrams . . . 491
4 Infinitesimal Symmetric Categories. . . . . . 494
5 A Universal Category for Infinitesimal Braidings 496
6 Formal Integration of Infinitesimal Symmetric Categories 498
7 Construction of Kontsevich's Universal Invariant 499
8 Recovering Quantum Group Invariants 502
9 Exercises 505
10 Notes . . . . 505

References 506

Index 523
 Part One

Quantum 8L(2)
Chapter I
Preliminaries

The goal of this first chapter is the construction of polynomial algebras

GL(2) and SL(2) modelling the 2 x 2-matrices with invertible determinant
[resp. with determinant equal to 1]. The multiplication of matrices induces
an additional structure on these algebras. This structure is one of the basic
ingredients of what will be called a Hopf algebra in Chapter III. We com-
plete the chapter with various concepts of ring theory to be used in the
sequel. The ground field is denoted by k.

I.1 Algebras and Modules

We recall some facts on algebras and modules.
An algebra is a ring A together with a ring map rJA : k ...... A whose image
is contained in the centre of A. The map (A, a) ...... rJA(A)a from k x A to A
equips A with a vector space structure over k and the multiplication map
PA : A x A ...... A is bilinear.
A morphism of algebras or an algebra morphism is a ring map f : A ...... B
such that
(1.1 )
As a consequence of (1.1), f preserves the units, i.e., we have f(l) = 1.
The linear map rJA : k ...... A is a morphism of algebras. If i : A ...... B is an
injective algebra morphism, we say that A is a subalgebra of the algebra B.
Let us denote by HOmAlg(A, B) the set of algebra morphisms from A to
B. In general, this set has no further structure. Nevertheless, we shall soon
4 Chapter I. Preliminaries

see how to put a group structure on HOmAlg(A, B) when A and B satisfy

some additional hypotheses.
We give a few examples of algebras that will be used frequently in this
book.
l. Given an algebra A, we define the opposite algebra AOP as the algebra
with the same underlying vector space as A, but with multiplication defined
by
(l.2)
where T A,A is the flip switching the two factors of A x A. In other words,

(l.3)

An algebra A is commutative if and only if

(1.4)

2. The centre Z(A) of an algebra A is the subalgebra

{a E A I aa' = a' a for all a' E A}.

We have Z(A) = Z(AOP).

3. If I is a two-sided ideal of an algebra A, i.e., a subspace of A such that

then there exists a unique algebra structure on the quotient vector space
AI I such that the canonical projection from A onto AI I is a morphism of
algebras.
4. We endow the product set A = DiEI Ai of a family (Ai)iEI of algebras
with the unique algebra structure such that the canonical projection from
A to Ai is an algebra morphism for all i E I. The algebra A is called the
product algebra of the family (Ai)iEI.
5. Given an algebra A we can form the algebra A[x] of all polynomials
~~=o aix i where n is any non-negative integer and the algebra A[x, X-I]
n .
of all Laurent polynomials ~i=m aix~ where m, n E Z.
6. For any positive integer n we denote by Mn(A) the algebra of all
n x n-matrices with entries in A.
7. The space End(V) of linear endomorphisms of a vector space V is an
algebra with product given by the composition and unit by the identity
map id v of V.
Given an algebra A, a left A-module or, simply, an A-module is a vector
space V together with a bilinear map (a, v) f---> av from A x V to V such
that
a(a'v) = (aa')v and Iv = v (l.5)
 1.1 Algebras and Modules 5

for all a, a' E A and v E V. One similarly defines a right A-module using
a bilinear map from V x A to V. A right A-module is nothing else than a
left module over the opposite algebra AOP. Therefore we need only consider
left modules which shall for simplicity be called modules in the sequel.
If V and V' are A-modules, a linear map f : V --+ V'is said to be
A-linear or a morphism of A-modules if

f(av) = af(v) (1.6)

for all a E A and v E V.

An A-submodule V' of an A-module V is a subspace of V with an A-
module structure such that the inclusion of V' into V is A-linear.
The action of A on an A-module V defines an algebra morphism p from
A to End(V) by
p(a)(v) = avo (1.7)
The map p is called a representation of A on V.
Given A-modules VI' ... ' Vn , the direct sum VI EB EBVn has an A-module
structure given by

(1.8)

where a E A, VI E VI' ... ,Vn E Vn . These definitions lead us to the following

ones.

Definition 1.1.1. An A-module V is simple if it has no other submodules

than {O} and V. It is semisimple if it is isomorphic to a direct sum of
simple A-modules. It is indecomposable if it is not isomorphic to the direct
sum of two non-zero submodules.

In the language of representations, a simple module [resp. a semisimple

module] is an irreducible representation [resp. a completely reducible repre-
sentation]. The following well-known proposition will be used in Chapters
V-VII.
Proposition 1.1.2. The following statements are equivalent.
(i) For any pair V' C V of finite-dimensional A-modules, there exists
an A -module V" such that V s:! V' EB V".
(ii) For any pair V' C V of finite-dimensional A-modules where V'is
simple, there exists an A -module V" such that V s:! V' EB V".
(iii) For any pair V' C V of finite-dimensional A-modules, there exists
an A-linear map p : V --+ V' with p2 = p.
(iv) For any pair V' C V of finite-dimensional A-modules where V'is
simple, there exists an A-linear map p : V --+ V' with p2 = p.
(v) Any finite-dimensional A-module is semisimple.
6 Chapter 1. Preliminaries

PROOF. Clearly, (i) =} (ii) and (iii) =} (iv). We also have (i) =} (iii): it
suffices to define p as the canonical projection from Vi EB V" onto V'.
Similarly, (ii) =} (iv).
Assertion (iii) =} Assertion (i). Let V" = Ker (p); it is a submodule of V.
The relations v = p( v) + (v - p( v)) and p2 = P prove that V is the direct
sum Vi and V". Similarly, (iv) =} (ii).
Assertion (ii) =} Assertion (v). Assuming (ii), we have to prove that
any finite-dimensional A-module V is semisimple. We may also assume
that dim(V) > O. Consider a non-zero submodule VI of V of minimal
dimension; it has to be simple. By (ii) there exists a submodule VI such
that V ~ VI EB VI and dim(VI) < dim(V). Iterating this procedure, we
build a sequence (Vn)n>O of simple submodules and a sequence (vn )n>O of
submodules such that
V n ~ Vn+ I EB V n+ I and dim(Vn+1) < dim(Vn).
Since the dimension of vn is strictly decreasing, there exists an integer p
such that VP = {O}. The module V is a direct sum of simple modules:
V~VIEB"'EBVp'
It remains to be shown that Assertion (v) implies Assertion (i). Let
Vi C V be a pair of finite-dimensional A-modules. By (v)

V=EB1I;
iEI

is a direct sum over a finite index set I of simple submodules 11;. Let J be
a maximal subset of I such that
Vi n (EB Yj) = {o}. (1.9)
jEJ

If i tf- J, then
Vi n (11; EB EB Yj) =I {O},
jEJ
hence
11; n (V' + EB Yj) =I {O}.
jEJ
Since 11; is simple, this implies that

11; c Vi + EB Yj
jEJ

for all i tf- J. This holds also for all i E J. Consequently, for the sum V of
all 11; we must have
V = Vi + EB
Yj. (1.10)
jEJ
As a consequence of (1.9-1.10), we get V = Vi EB V" where V" is the
submodule EB jEJ Yj. 0
 I.2 Free Algebras 7

1.2 Free Algebras

Let X be a set. Consider the vector space k{ X} with basis the set of all
words xi, ... Xi p in the alphabet X, including the empty word 0. A word
will be called a monomial. Define the degree of the monomial Xi 1 ... X,"p as
its length p. Concatenation of words defines a multiplication on k{ X} by

Formula (2.1) equips k{X} with an algebra structure, called the free algebra
on the set X. The unit is the empty word: 1 = 0. In the sequel we shall
mainly consider free algebras on finite sets. If X = {Xl' ... ' Xn} we also
denote k{X} by k{xl, ... ,xn }.
Free algebras have the following universal property.

Proposition 1.2.1. Let X be a set. Given an algebra A and a set-theoretic

map f from X to A, there exists a unique algebra morphism 1 : k{ X} ----> A
such that 1(x) = f(x) for all X E X.

PROOF. It is enough to define Ion any word of X. For the empty word we
set 1(0) = 1. Otherwise, if x il ' ... , xi p are elements of X, we define

The rest of the proof follows easily. o

An equivalent formulation of Proposition 2.1 is: There exists a natural
bijection
HOmAlg(k{X},A) ~ Homset(X,A) (2.2)
where Homset(X, A) is the set of all set-theoretic maps from X to A. In
particular, if X is the finite set {Xl' ,xn }, then f ~ (f(x l ),, f(xn))
induces a bijection

(2.3)

Any algebra A is the quotient of a free algebra k{X}. It suffices to take

any generating set X for the algebra A (for instance X = A). Consequently,
A = k{X}jI where I is a two-sided ideal of k{X}. In this case, for any
algebra A' we have the natural bijection

HOmAlg(k{X} j I, A') ~ {f E Homset(X, A') 11(1) = O}. (2.4)

Example 1. Let I be the two-sided ideal of k{x l , ... ,xn } generated by all
elements of the form xix j - XjXi where i,j run over all integers from 1 to
n. The quotient-algebra k{ Xl' ... , Xn} j I is isomorphic to the polynomial
8 Chapter I. Preliminaries

algebra k[XI' ... ,xn ] in n variables with coefficients in the ground field k.
As a corollary of (2.4) we have

HOmAlg(k[xl> ... ,xn],A) ~ {(al' ,an) E An I aiaj = aja i for all (i,j)}
(2.5)
for any algebra A.
In the next sections we shall see more examples where families of elements
subject to "universal" algebraic relations are represented by quotients of
free algebras.

1.3 The Affine Line and Plane

Let us restrict to commutative algebras. As a consequence of (2.5) we have
the following proposition.

Proposition 1.3.1. Let A be a commutative algebra and f be a set-theoretic

map from the finite set {Xl' ... ,Xn } to A. There exists a unique morphism
of algebras! from k[XI' ... ,x n ] to A such that !(x i ) = f(xi) for all i.
In other words, giving an algebra morphism from the polynomial algebra
k[xI' ... ,xn] to a commutative algebra A is equivalent to giving an n-tuple
(aI' ... ,an) of elements of A:

(3.1)

Let us consider the special case n = 1 of (3.1). For any commutative

algebra A the underlying set A is in bijection with the set HOmAlg(k[x], A):

HOmAlg(k[x], A) ~ A. (3.2)

The algebra k[x] is called the affine line and the set HOmAlg(k[x], A) is
called the set of A-points of the affine line. Now A has an abelian group
structure. We wish to express it in a universal way using the affine line
k[x]. The abelian group structure of A consists of three maps, namely the
addition + : A2 -+ A, the unit 0 : {O} -+ A, and the inverse - : A -+ A,
satisfying the well-known axioms which express the fact that the addition
is associative and commutative, that it has 0 as a left and right unit and
that
(-a) + a = a + (-a) = 0
for all a E A. These laws do not depend on the particular commutative
algebra A. It will therefore be possible to express them universally.
To this end, let us introduce the affine plane k[x', x"] with the bijection

Hom Alg (k[x' , x"] , A) -c>< A2 (3.3)

 1.3 The Affine Line and Plane 9

obtained from (3.1) for n = 2. An element of Hom Alg (k [x' , x"], A) is called
an A-point of the affine plane. The set HOmAlg(k, A), reduced to the single
point rJ A' will be denoted by {O}.

Proposition 1.3.2. Let ~ : k[x] ---> k[x', x"], c : k[x] ---> k, S: k[x] ---> k[x]
be the algebra morphisms defined by

~(x) = x' + x", c(x) = 0, S(x) = -x.

Under the identifications (3.2-3.3), the morphisms
to the maps +, and - respectively.
~, c and S correspond

PROOF. Left to the reader. D

The morphisms ~, c and S are subject to further relations which express

the associativity, the commutativity, the unit and the inverse axioms of an
abelian group. They equip the affine line k[x] with what will be called a
cocommutative Hopf algebra structure in Chapter III.
In order to illustrate better the phenomenon we have just observed, we
give another example. For any algebra A denote by A x the group of in-
vertible elements of A. We represent the set A x by an algebra as above.
Consider the ideal I of k[x, y] generated by xy - 1. For any commutative
algebra A we have
(3.4)
The set {xkhEZ is a basis of the vector space k[x,y]/I. We denote this
algebra by k[x, X-I]; it is the algebra of Laurent polynomials in one variable.
One defines similarly the algebra

k[x', x", x'-I, X"-I] = k[x', y', x", Y"l/ (x' y' - 1, x" y" - 1)

of Laurent polynomials in two variables. We have a bijection

Hom Alg (k[x' , X,-I " x" X"-I] , <:Y A x x A x

A)- . (3.5)

Define algebra morphisms

A k[x x-I]
u ., ---> k[x' " x" X"-I]
X,-I" co k[x X-I] ---> k
<C., ,

by
~(x) = x'x", c(x) = 1, S(x) = X-I. (3.6)
Then the morphisms ~, c and S correspond respectively to the multipli-
cation in A x, to the unit 1 and to the inverse under the identifications
(3.4-3.5). Here again, the morphisms ~,c, S equip k[x, X-I] with a co com-
mutative Hopf algebra structure.
10 Chapter 1. Preliminaries

1.4 Matrix Multiplication

For any algebra A we denote by M 2 (A) the algebra of 2 x 2-matrices with
entries in A. As a set, M 2(A) is in bijection with the set A4 of 4-tuples of
A. By (3.1) we have a natural bijection

(4.1)

for any commutative algebra A where M(2) is defined as the polynomial

algebra k[a, b, c, d]. This bijection maps an algebra morphism f : M(2) --+ A
to the matrix
( f(a) f(b))
f(c) f( d) .
The multiplication of matrices is a map M 2(A) x M 2(A) --+ M 2(A) we
wish to represent universally on M(2), in the spirit of Section 3. The set
M 2 (A) x M 2(A) being in bijection with AS, we introduce the polynomial
algebra
M(2)@2 = k[a', a", b', b", c', c", d', d"]. (4.2)

Proposition 1.4.1. Let .6. : M(2) --+ M(2)@2 be the algebra morphism
defined by

.6.(a) = a'a" + b'c", .6. (b) = a' b" + b'd" ,

.6.(c) = c'a" + d'c", .6.(d) = c'b" + d'd" .

Then for any commutative algebra A, the morphism .6. corresponds to the
matrix multiplication in M 2 (A) under the identifications (4.1-4.2).

The proof is easy and left to the reader. It is convenient to rewrite the
formulas for .6. in Proposition 4.1 in the compact matrix form

a b) (.6.(a) .6. (b) ) (a' b') (a" b") (4.3)

.6. ( c d = .6. (c) .6. (d) = c' d' c" d" .

1.5 Determinants and Invertible Matrices

We keep the notations of the previous section. We now consider the group
GL 2(A) of invertible matrices of the matrix algebra M2(A). When A is
commutative, we know that a matrix is invertible if and only if its deter-
minant is invertible in A:

Define SL 2 (A) as the subgroup of GL 2 (A) of matrices with determinant

0.8 - /31 = 1.
 1.5 Determinants and Invertible Matrices 11

Proposition 1.5.1. Define the commutative algebras

GL(2) = M(2) [t]j((ad - be)t - 1)

and
SL(2) = GL(2)/(t - 1) = M(2)/(ad - be - 1).
For any commutative algebra A there are bijections

sending an algebra morphism f to the matrix

( f(a) f(b))
f(c) f(d) .

PROOF. We give it only for GL(2). Similar arguments work for SL(2). Let
(~ ~) be a matrix in GL 2 (A). Since A is commutative, there exists a
unique algebra morphism f : M(2)[tJ -+ A such that
f(a) = a, f(b) = (3, f(e) = "1, f(d) = 8 and f(t) = (a8 - (3"1)-1.

Now,

f((ad - be)t - 1) (i(a)f(d) - f(b)f(e))f(t) - f(l)

(a8 - (3'Y)(a8 - (3"1)-1 - 1
o.
This implies that the morphism f factors through the quotient algebra
GL(2). The rest ofthe proof is easy. 0

The next lemma follows from a straightforward computation using the

morphism Do of Proposition 4.1.
Lemma 1.5.2. We have Do(ad - be) = (a'd' - b'e')(a"d" - b"e").
We now lift the group structures of GL 2(A) and of SL 2(A) to the algebras
GL(2) and SL(2). Consider the commutative algebras
GL(2)2 = M(2)2 [t', t"J/ (( a' d' - b' e')t' - 1, (a" d" - b" e")t" - 1)

and

SL(2)2 = GL(2)2 /(t' -1, t" -1) = M(2)2/(a'd' -b' e' -1, a" d" -b" e" -1).

Proposition 1.5.3. The formulas of Proposition 4.1 define algebra mor-

phisms
Do: GL(2) -+ GL(2)2 and Do: SL(2) -+ SL(2)2.
12 Chapter I. Preliminaries

PROOF. The formulas of Proposition 4.1 define an algebra morphism ~

from M(2)[t] to GL(2),2)2 provided we set ~(t) = t't". In order to show
that ~ factors through GL(2) we have to check that ~((ad - bc)t - 1)
vanishes. Now, by Lemma 5.2 and by definition of GL(2)2, we have

~((ad - bc)t - 1) (a'd' - b'c')(a"d" - b"c")t't"-1

1.1 - 1 = o.
The proof for SL(2) is similar. D

In Section 4 we checked that the map ~ corresponded to matrix multi-

plication under the above identifications. Let us exhibit the algebra maps

c : GL(2) ----> k and c: SL(2) ----> k

corresponding to the units of the groups GL 2(A) and SL 2(A) and the
algebra morphisms

S: GL(2) ----> GL(2) and S: SL(2) ----> SL(2)

corresponding to the inversions in the same groups. They are defined by

the formulas

c(a) = c(d) = c(t) = 1, c(b) = c(c) = 0,

S(a) = (ad - bC)-1 d, S(b) = -(ad - bC)-1 b,
S(c) = -(ad - bc)-1 c, S(d) = (ad - bc)-1 a,
and S(t) = C 1 = ad - bc. We rewrite them in the more compact and
illuminating form

c ( ac db) = (1 0)
0 1 and S (ac db) = (ad - bc) -1 ( d
-c -b)
a .
(5.2)

1.6 Graded and Filtered Algebras

The remaining sections of this chapter are devoted to some concepts of ring
theory.

Definition 1.6.1. An algebra A is graded if there exist subspaces (Ai)iEN

such that
and

for all i,j EN. The elements of Ai are said to be homogeneous of degree i.
 I.6 Graded and Filtered Algebras 13

We always assume that the unit 1 of a graded algebra belongs to AD.

Example 1. Free algebras are graded by the length of words, i.e., the
subspace Ai of A = k{X} is defined as the subspace linearly generated by
all monomials of degree i. The elements of X are of degree 1.

Proposition 1.6.2. Let A = E9i>O Ai be a graded algebra and I be a two-

sided ideal generated by homogeneous elements. Then

I = EB InA;
i;:>O

and the quotient algebra AI I is graded with (AI I)i = Ad (I n Ai) for all i.
PROOF. It suffices to show that I = E9i>O I n Ai' First observe that the
sum has to be direct since the subspaces Ai form a direct sum. Therefore,
it remains to be checked that I = 2:i>O I n Ai' The ideal I is generated by
homogeneous elements xi of degree d~. Consequently, if x E I then

for some ai' b;EA. Now, a; = 2: j ai and bi = 2: j bi, where ai and bi are
homogeneous elements of degree j. It follows that

x = L 7
ai x i b
i,j,k

is a sum of homogeneous elements of degree d i + j + k in I. This implies

that I is a subspace of 2:i;:>O I n Ai' The converse inclusion is clear. 0

Example 2. The polynomial algebra k[XI' ... ,xnl is graded as the quotient
of the free algebra A = k{ Xl' ... ,xn } (graded as in Example 1) by the ideal
I generated by the degree-2 homogeneous elements xix j -XjXi where i and
j run over all integers between 1 and n. The generators Xl' ... 'X n are of
degree one.
The algebras M(2) and M(2)2 of Section 4 are graded as polynomial
algebras. On the contrary, the ideals defining the algebras GL(2) and 5L(2)
are not generated by homogeneous elements. Though not graded, GL(2)
and 5L(2) are filtered algebras in the sense of the following definition.

Definition 1.6.3. An algebra A is filtered if there exists an increasing se-

quence {O} C Fo(A) C ... C Fi(A) C ... C A of subspaces of A such
that
A = UFi(A) and Fi(A). Fj(A) C Fi+j(A).
i;:>O

The elements of FJA) are said to be of degree S; i.

14 Chapter 1. Preliminaries

For any filtered algebra A there exists a graded algebra S = gr(A) defined
by

We give a few examples of filtered algebras.

Example 3. Any algebra A has a trivial filtration given by Fi(A) = A for
all i.
Example 4. We filter any graded algebra A = EBi20 Ai by

Fi(A) = EB Aj
0:Scj:Sci
for all i E N. We have gr(A) = A.
Example 5. Let A:J ... :J Fl(A):J Fo(A) be a filtered algebra and I be a
two-sided ideal of A. The quotient-algebra A/lis filtered with

In this case we have

gr(A/1) = EB Fi(A)/(Fi_ (A) + Fi(A) n 1).

1
i2
As a special case, consider the algebra SL(2). It is filtered as the quotient
of the graded algebra M(2). We have

gr(SL(2)) ~ k[a, b, c, d, ]/(ad - bc).

1. 7 Ore Extensions
Let R be an algebra and R[t] be the free (left) R-module consisting of all
polynomials of the form

P = ant n + an_1t n - 1 + ... + aotO

with coefficients in R. If an -I=- 0, we say tha~ the degree deg(P) of P is
equal to n; by convention, we set deg(O) = -00. The aim of this section is
to find all algebra structures on R[t] compatible with the algebra structure
on R and with the degree. We need the following definition.
Let a be an algebra endomorphism of R. An a-derivation of R is a linear
endomorphism 8 of R such that

8(ab) = a(a)8(b) + 8(a)b (7.1)

for all a, bE R. Observe that (7.1) implies 8(1) = O.

 I. 7 Ore Extensions 15

Theorem 1.7.1. (a) Assume that R[t] has an algebra structure such that
the natural inclusion of R into R[t] is a morphism of algebras, and we have
deg(PQ) = deg(P) + deg(Q) for any pair (P, Q) of elements of R[t]. Then
R has no zero-divisors and there exist a unique injective algebra endomor-
phism a of R and a unique a-derivation 8 of R such that

ta = a(a)t + 8(a) (7.2)

for all a E R.
(b) Conversely, let R be an algebra without zero-divisors. Given an injec-
tive algebra endomorphism a of R and an a-derivation 8 of R, there exists
a unique algebra structure on R[t] such that the inclusion of R into R[t] is
an algebra morphism and Relation (7.2) holds for all a in R.

The algebra defined by Theorem 7.1 (b), denoted R [t, a, 8], is called the
Ore extension attached to the data (R, a, 8).
PROOF. (a) Let a, b be non-zero elements of R, hence of degree 0 in R[t].
We have deg(ab) = deg(a) + deg(b) = 0, which implies that ab =f. O. Conse-
quently, R has no zero-divisors.
Let us now prove the existence and the uniqueness of the endomorphisms
a and 8. Take any non-zero element a of R and consider the product tao
We have deg(ta) = deg(t) + deg(a) = 1. By definition of R[t] there exist
uniquely determined elements a(a) =f. 0 and 8(a) of R such that

ta = a(a)t + 8(a). (7.2)

This defines maps a and 8 in a unique fashion. The left multiplication by

t being linear, so are a and 8. Furthermore, a has to be injective. Let us
expand both sides of the equality (ta)b = t(ab) in R[t] using (7.2). Here a
and b are elements of R. We get

a(a)a(b)t + a(a)8(b) + 8(a)b = a(ab)t + 8(ab). (7.3)

Relation (7.3) implies that

a(ab) = a(a)a(b) and 8(ab) = a(a)8(b) + 8(a)b. (7.4)

Applying (7.2) to t1 = t yields a(l) = 1 and 8(1) = O. It follows that a is

an injective algebra endomorphism and 8 is an a-derivation.
(b) It clearly suffices to know the product ta for any a ERin order to
determine the product on R[t] completely. Thus, (7.2) defines the algebra
structure on R[t] uniquely.
Let us now prove the existence of the algebra structure. To this end, we
shall embed R[t] into the associative algebra M consisting of all infinite
matrices (fij)i,j?l with entries in the algebra End(R) of linear endo~or
phisms of R such that each row, as well as each column, has only fimtely
16 Chapter 1. Preliminaries

many non-zero entries. The unit of M is the infinite diagonal matrix I

with identities on the diagonal. Given an element a of R, we denote by
Ii E End(R) the left multiplication by a. The hypotheses made on a and 8
translate into the relations

81i = a(a)8 + 8(a)

~

ali = a(a)a and (7.5)

in End(R). Now, consider the infinite matrix

8 0 0 0
a 8 0 0
0 a 8 0
T= 0 0 a 8
0 0 0 a

in M. It allows one to define a linear map <P : R[t] ----+ M by

n n
<p(L aiti) = L (fiJ)T i . (7.6)
i=O i=O

Lemma I. 7.2. The map <P is injective.

PROOF. For any integer i ~ 1, let e i be the infinite column vector whose
entries are all zero, except for the i-th one which is equal to the unit 1 of
R. We may apply the matrix T of endomorphisms to ei . Since 8(1) = 0 and
a(l) = 1 we get
(7.7)
for all i ~ 1. Now, let P L:~=o ait i be an element of R[t] such that
<p(P) = O. We wish to show that all elements ao,"" an are zero. Apply
<p(P) to the vector column e1 . By (7.7) we get
n n
0= <p(P)(e 1 ) = L (iiiI)T i (e 1 ) = L iii ei+1'
i=O i=O

The set {eJ i>l being free, we have iii = 0 for all i. Since R has a unit, we
get a i = 0 for-all i. Hence, P = O. D

Relations (7.5) imply the following relation in M for all a E R.

Lemma 1.7.3. We have T(liI) = (a(a)I)T + (8(a)I).

We now complete the proof of Theorem 7.1 (b). Let S be the subalgebra
of M generated by the elements T and iiI where a runs over R. By Lemma
7.3 it is clear that S is the image of R[t] under the map <P. Since the latter
is injective, it induces a linear isomorphism from R[t] to the algebra S. This
 I. 7 Ore Extensions 17

allows one to lift the algebra structure of S to R[t]. Relation (7.2) holds in
R[t] in view of Lemma 7.3. 0

We draw a few consequences. First, we wish to give a general formula

for the product in R[t,a,8]. Consider P = 2:~=0 a/
and Q = 2::0 b/.
Set PQ = 2:~:; citi. Let Sn,k be the linear endomorphism of R defined
as the sum of all ( ~ ) possible compositions of k copies of 8 and of n - k
copies of a.

Corollary 1.7.4. Under the hypotheses of Theorem 7.1 (b), the following
holds.
(a) For all i with 0:::; i :::; m + n we have
i P

ci = Lap L Sp,k(bi - pH ) (7.8)

p=o k=O

and for all a E Rand n E N we have in R[t, a, 8]

n
tna = L Sn,k(a)t n - k . (7.9)
k=O
(b) The algebra R[t, a, 8] has no zero-divisors. As a left R-module, it is
free with basis {ti} iEN'
(c) If a is an automorphism, then R[t, a, 8] is also a right free R-module
with the same basis {tiLEN'

PROOF. (a) Relation (7.9) follows from (7.2) by induction on n. It implies

(7.8).
(b) This is a consequence of the existence of the degree and of the defi-
nition of R[t].
(c) Let us first prove that the set {ti L>o generates R[t, a, 8] as a right
R-module. This means that any element P of R[t, a, 8] can also be written
under the form P = 2:~=0 tia i where aD, ... , an E R. Let us prove this by
induction on the degree n of P. For n = 0, it is clear. For higher n we use
the relation
(7.10)
which makes sense once a is assumed to be invertible. It remains to be
proved that the set {ti L>o is free. Suppose it is not. Then there exists a
relation of the form -

with an i= O. Using (7.10) once again, we get another relation of the form
18 Chapter 1. Preliminaries

which, by Part (b), implies that an(a n ) = O. The map a being an isomor-
phism, we get an = 0, hence a contradiction. D

Example 1. Consider the special case a = id R . If b = 0, then the Ore

extension R[t, id R , 0] is clearly isomorphic to the polynomial algebra R[t].
In case of a general derivation b, the algebra R[t, id R , b] is an algebra of
polynomial differential operators (see Exercise 8). When R = k[x] and b
is the usual derivation dj dx of polynomials, then R[t, id R , b] is the Weyl
algebra which is generated by two variables x and b subject to the well-
known Heisenberg relation bx - xb = 1.

1. 8 Noetherian Rings
Proposition 1.8.1. Let A be a ring. The following two statements are
equivalent.
(i) Any left ideal I of A is finitely generated, i. e., there exist aI' ... , an
in I such that I = Aa 1 + ... + Aan .
(ii) Any ascending sequence II C 12 C 13 C ... c A of left ideals of A is
finite, i. e., there exists an integer r such that Ir+i = IT for all i ~ O.

PROOF. Let us first show that (i) implies (ii). Consider an ascending se-
quence II C 12 C 13 C ... of left ideals of A. The union of these ideals is
a left ideal I which, by (i), is generated by a finite number a 1 , . .. ,an of
elements of A. By definition of the union there exists an integer r such that
aI' ... ,an all belong to the ideal IT' It follows that I C IT C Ir+i C I for
all i ~ O.
We now establish the converse. Let I be a left ideal that is not finitely
generated and a 1 be an element of I. The left ideal II = Aa 1 is contained
in I and II =J I. Therefore, we can find an element a 2 E 1\ Aa 1 . We have
II C 12 = Aa 1 + Aa 2 C I and II =J 12 =J I. Proceeding inductively, we find
an infinite strictly ascending sequence II C ... In C 171 +1 C ... I of left
ideals. D

Any ring A satisfying the equivalent conditions of Proposition 8.1 is said

to be left Noetherian. The ring A is right Noetherian if the opposite ring A op
is left Noetherian. It is Noetherian if it is both left and right Noetherian.
Example 1. Any (skew-) field K is Noetherian, the only ideals being {O}
and K.
The property of being Noetherian is preserved by quotients and Ore
extensions, as will be seen next.

Proposition 1.8.2. Let ip : A ----+ B be a surjective morphism of rings. If

A is left Noetherian, then so is B.
 1.8 Noetherian Rings 19

PROOF. Let J be a left ideal of B. The left ideal I = '1'-1 (J) of A is

generated by elements al, ... ,an . Therefore, J = cp(cp-l(J)) is generated
by cp(a l ), ... , cp(a r,). 0

The following theorem is a non-commutative version of Hilbert's basis

theorem.

Theorem 1.8.3. Let R be an algebra, a be an algebra automorphism and 15

be an a-derivation of R. If R is left Noetherian, then so is the Ore extension
R[t, a, 15].

As a consequence of Proposition 8.2 and Theorem 8.3 applied to the case

a = id and 15 = 0, we have

Corollary 1.8.4. If R is left Noetherian, then so is R[Xl, ... ,Xn]/I for

any ideal I.

Proof of Theorem 8.3. Let I be a left ideal of the Ore extension R[t, a, 15].
We have to prove that I is finitely generated. Given an integer d ;::: 0, define
Id as the union of {O} and of all elements of R which appear as leading
coefficients of degree d elements of I. One checks easily that Id is a left
ideal of R.
On the other hand, if a is the leading coefficient of some polynomial P,
then a(a) is the leading coefficient of tF. Consequently, a(Id) is included
in I MI . We therefore have the ascending sequence

of left ideals in R. Since R is left Noetherian, there exists an integer n such

that In+i = ai(In) for all i ;::: O.
For any d with 0 :::; d :::; n choose generators ad,l" .. ,ad,p of I d. Let
Pd i be a degree d element of I whose leading coefficient is ad i' The set
{Pd,i}o<::d<::n,l<::i<::p is finite. Let us prove by induction on the degree that
any polynomial P in I belongs to the ideal I' = 2:d,i R[t, a, 15] Pd,i' This
will imply that I = I' is finitely generated, hence establish the theorem.
The induction hypothesis clearly holds in degree O. Suppose we have
proved that any element of degree < d in I is in I'. Let P be a degree d
element of I.
(a) If d:::; n, the leading coefficient a of Pis of the form a = 2:0<::i<::P riad,i
where r o,'" ,rp are elements of R. Consequently, Q = P - 2:0<::i<::p riPd,i
is an element of I of degree < d. By induction, Q, hence P, belong to I'.
(b) If d > n, the leading coefficient a of P belongs to Id = ad-n(In)' It
can be written a = 2:0<::i<::P riad-n(ad,i) for some r o,"" rp in R. Consider
the polynomial
Q=P- L ritd-nPd,i'
o<::i <::P
20 Chapter 1. Preliminaries

The coefficient of t d in Q is

a- L riad-n(ad,.J = 0,
O::;i::;p

which shows that the degree of Q is < d. We can therefore apply the
induction hypothesis and conclude as above. 0

I. 9 Exercises
1. (Schur's lemma) Prove that any A-linear map between simple A-
modules is either zero or an isomorphism. Deduce that the A-linear
endomorphisms of a simple A-module form a skew-field.

2. Let p = p2 be an A-linear idempotent endomorphism of an indecom-

posable A-module V. Show that p = 0 or p = id v .

3. Let AI' A2 be algebras. Let VI be an AI-module and V2 be an A 2-

module. Establish that (aI' a2)(VI' V2) = (a l VI' a 2v 2 ) (where a l E AI'
a 2 E A 2 , VI E VI' V 2 E V2 ) defines an Al x A 2 -module structure on
VI x V2 . Prove also that any Al x A 2 -module is of this form.

4. Let A be a filtered algebra and gr(A) the associated graded algebra.

Prove that if gr(A) is Noetherian without zero-divisors, then so is A.

5. (Rees algebra) Let A :=> .. :=> FI :=> Fo be a filtered algebra. Define

the Rees algebra R(A) as the sub algebra

of the polynomial algebra A[t]. Prove that

(i) there are algebra isomorphisms

R(A)/(t - 1) ~ A, R(A)/(t) ~ gr(A), R(A)[e l ] ~ A[t, ell,

(ii) if the algebra gr( A) is generated by homogeneous elements aI' ... ,

an of respective degrees d l , ... , d n , then R( A) is generated by the
elements t, a l t d1 , ... ,antdn where ai is a lift of ai in Fi for all i.

6. (Poincare series of a graded algebra) Let A = E9 i >o Ai be a graded

algebra such that the vector spaces Ai are all finite-dimensional. De-
fine the Poincare series of A as the formal series

P(A) = L dim(Ai) ti.

i?O
 1.9 Exercises 21

Prove that
1 1
P(k{x1,,x n }) =-- and P(k[x1,, Xn]) = (1 _ t)n
I - nt

7. Compute the Poincare series of the graded algebra associated to the

filtered algebra SL(2).

8. (Leibniz formula) Let 0 be an a-derivation of an algebra R. Prove

that if a 1, ... , an are elements of R, then

0(a 1 ... an) = 0(a 1)a 2 an

+ L:::::-21 a(a 1 ... ai-1)0(ai)ai+l'" an + a(a 1 an_1)0(a n )
and
n
On(a 1a2 ) = 2..= Sn,k(a 1) on-k(a 2 )
k=O
for n ;::: 1. The endomorphisms Sn,k were defined in Section 7.

9. Let R be an algebra with an algebra automorphism a and an a-

derivation O. Establish that oa -1 is an a -l-derivation of the opposite
algebra ROP and that we have an algebra isomorphism

Deduce that R[t, a, oj is right Noetherian if R is.

10. (Algebra of differential operators) Let R be an algebra over a field

of characteristic zero and let 0 be a derivation of R. The algebra of
differential operators associated to 0 is the Ore extension R[t, id R , 0],
which we simply denote by R[t, oj.
(a) Prove that for any integer n > 0 and any element a of R we have

(b) Show that any trace on R[t,o], i.e., any linear map T on R[t, oj
such that T(XY) = T(YX) for any pair (x, y) of elements of R[t, 0], is
zero.

11. (Algebra of pseudo-differential operators) Keep the hypotheses and

the notations of the previous exercise. Show that the formula

(2..= a/i) (2..= biti ) = 2..= cit i

i j, i
22 Chapter I. Preliminaries

where

" p(p - 1) ... (p - k + 1) bk (b )

~ k! ap '-p+k ,
kEN,pEZ

defines an algebra structure on the vector space R[t, b] [[C 1 ]] of formal

series of the form L~=-CXJ aiti. Check that R[t, b] is a subalgebra. De-
fine the non-commutative residue as the linear map from R[t, b][[C 1 ]]
to R/([R, R] + b(R)), sending the formal series L~=-CXJ ait i to the
class of the coefficient a_ 1 . Prove that the non-commutative residue
is a trace on the algebra R[t, b][[C 1 ]] of pseudo-differential operators.

1.10 Notes
Ore extensions were introduced by Ore in [Ore33]. They are also called
"skew polynomial rings" in [Coh71] [MR87] (see also [Cur52]). One of Ore's
motivations was to find a large class of non-commutative algebras that
are embeddable into a skew-field. As is well-known, this is possible for
any commutative integral domain, but not for a general non-commutative
algebra. Ore proved that any algebra obtained from a skew-field by iterated
Ore extensions can itself be embedded into some skew-field (see Proposition
0.8.4 in [Coh71]). For more details on Noetherian rings, we refer the reader
to [Lan65] and [MR87]. The examples given in [MR87], 2.11 show that the
non-commutative version of Hilbert's basis theorem is no longer true if the
endomorphism a is not assumed to be bijective.
Chapter II
Tensor Products

This chapter is devoted to a few facts on tensor products of vector spaces

and of algebras that will be needed in the sequel. We fix a field k once and
for all.

11.1 Tensor Products of Vector Spaces

Given vector spaces U and V, we denote by Hom(U, V) the space of linear
maps from U to V. In particular, define End(V) = Hom(V, V), the space
of linear endomorphisms of V. If W is a third vector space, we denote by
Hom(2) (U, V; W) the space of bilinear maps from U x V to W.
The tensor product U Q9 V of two vector spaces can be characterized as
follows.
Theorem 11.1.1. Given vector spaces U and V there exist a vector space
U Q9 V and a bilinear map CPo : U x V ---> U Q9 V such that, for all vector
spaces W, the linear map
Hom(U Q9 V, W) ---> Hom(2)(U, V; W)
given by f f-+ f 0 CPo is a linear isomorphism. The vector space U Q9 V is
called the tensor product of U and V. It is unique up to isomorphism.
For any u E U and v E V, set u Q9 v = CPo (u, v). Since CPo is bilinear, the
following relations hold in U Q9 V:
(u + u') Q9V = uQ9v + u' Q9V, (1.1 )
24 Chapter II. Tensor Products

U(v+v') =Uv+Uv', (1.2)

A(UV) = (AU) v = U(AV) (1.3)
where u, u' E U, V, v' E V and A E k. Moreover, as we shall see in the
subsequent proof, any element of U V is a finite sum of the form

(1.4)

where u I , ... ,up belong to U and VI' ... ,vp belong to V.

PROOF. We indicate the proof. Consider the vector space k[U x V] whose
basis is the set U x V. We define U Vas the quotient of k[U x V] by the
subspace generated by the elements

(U + u', v) - (u, v) - (u', v), (u, v + v') - (u, v) - (u, v'),

(AU, v) - A(U, v), (u, AV) - A(U, v)

where u,u' E U, v,v' E V and A E k. The class of (u,v) E U x V in U V
is denoted 'Po (u, v) = U v. By construction, the canonical map 'Po from
U x V to U V is bilinear. The rest of the proof follows easily. For details,
see [Bou70], Chap. 2 and [Lan65]. D

Corollary 11.1.2. For any triple (U, V, W) of vector spaces there is a

natural isomorphism

Hom(U V, W) ~ Hom(U, Hom(V, W)) .

PROOF. If 'P is a bilinear map from U x V to Wand U is any vector of

U, then 'P( u, -) is a linear map from V to W. This sets up the desired
isomorphism. D

The proof of the following easy proposition is left to the reader.

Proposition 11.1.3. Let U, V, W be vector spaces. There are isomor-

phisms
(U V) W ~ U (V W)
determined by (u v) W f--+ U (v w),

kV~V~Vk

determined by A v f--+ AV and v f--+ v 1, and

given by the flip Tv,W defined by TV,W(V w) = w v.

 11.1 Tensor Products of Vector Spaces 25

The tensor product also commutes with the direct sum of spaces. Let
(Ui)iEI be a family of vector spaces indexed by a set I. Recall that there
exists a vector space EBiEI Ui , called the direct sum of the family (Ui ), and
linear maps qi : Ui ---> EBiEI Ui such that for any vector space V, the linear
map
(1.5)
iEI iEI
given by f ---> (f 0 qi)i is an isomorphism.
Proposition 11.1.4. We have

(1.6)
iEI iEI
PROOF. By Corollary 1.2 and (1.5) we have the chain of isomorphisms

Hom((EB Ui ) Q9 V, W) ~
Hom( EB Ui , Hom(V, W))
iEI iEI
~
IIHom(Ui,Hom(V, W))
iEI
~
II Hom(Ui Q9 V, W)
iEI
~
Hom(EB (Ui Q9 V), W).
iEI
These hold for any vector space W. A classical argument given in full detail
in the second proof of Proposition 5.1 (c) allows one to conclude. 0

Recall also the notion of a direct product of vector spaces. Let (V,)iEI
be a family of vector spaces indexed by a set I. There exists a vector space
[LEI v" called the direct product of the family (V,)iEI' and linear maps
Pi : DiEI V, ---> V, such that for all vector spaces U, the map

Hom(U, II V,) II Hom(U, V,)

---> (1. 7)
iEI iEI
given by f f--+ (Pi 0 f)i is an isomorphism. As a set, DiEI V, may be real-
ized as the vector space of all families (vi)iEI such that Vi E V, for all i.
The direct sum EBiEI V, is then the subspace of DiEI V, consisting of the
families (Vi)iEI where all but finitely many Vi are zero. When the indexing
set I is finite, the direct product coincides with the direct sum. Otherwise,
the direct sum is a proper subspace of the direct product.

Corollary 11.1.5. Let {Ui}iEI be a basis of the vector space U and {Vj}jEJ
be a basis of V. Then the set {u i Q9 Vj}(i,j)EIXJ is a basis of the tensor
product U Q9 V. Consequently, we have dim(U Q9 V) = dim(U) dim(V).
26 Chapter II. Tensor Products

PROOF. By definition of the direct sum, we have

US;! EB kU i and V S;! EB kv j . (1.8)

iEI jEJ

Applying Propositions 1.3-1.4 and using k 0 k S;! k, we get

U 0 V S;! EB k (u i 0 V j)'
(i,j)ElxJ

o
Let us define the notion of a free module over an algebra A using the
tensor product. It is a module of the form A 0 V where V is a vector space
and A acts on A 0 V by

a(a' 0 v) = aa' 0 v

for a, a' E A and v E V. A basis of an A-module M is a subset {Xi LEI of

M such that the map
(ai)iEI f--t L aixi
iEI

from the direct sum EBiEI A to M is an isomorphism. By Propositions

1.3-1.4,
EBA S;! EB(A 0 k) S;! A 0 V
iEI iEI

where V = EBiEI k. It follows that an A-module has a basis if and only if

it is free.

II.2 Tensor Products of Linear Maps

Let f : U ........ U' and 9 : V ........ V' be linear maps. We define their tensor
product f 0 9 : U 0 V ........ U' 0 V' by

(f0g)(u0v) = f(u)0g(v) (2.1)

for all u in U and v in V. This gives rise to a linear map

>. : Hom(U, U') 0 Hom(V, V') ........ Hom(V 0 U, U' 0 V') (2.2)

defined by
(>'(f09))(v0u) = f(u) 0g(v). (2.3)

The reasons for the switch of U and V in (2.2) will become apparent in
III.5.2 and in Chapter XIV. The main result of this section is the following.
 11.2 Tensor Products of Linear Maps 27

Theorem 11.2.1. The map A is an isomorphism provided at least one of

the pairs (U, U' ), (V, V') or (U, V) consists of finite-dimensional vector
spaces.

PROOF. Assume that U and U' are finite-dimensional. We wish to show

that the map A of (2.2) is an isomorphism. We shall do this by reducing A to
simpler maps. We can write U = EBiEI kU i where {u i } iEI is a finite basis of
U. As a consequence of the isomorphism (1.5-1.6), the map A turns into a
map from (TIi Hom(ku i , U' )) (X) Hom(V, V') to TIi Hom(V (X) ku i , U' (X) V').
The set I being finite, we may replace TIi by EB i . Applying (1.6) again, it
remains to prove that the map

A : Hom( ku i , U' ) (X) Hom(V, V') ---+ Hom(V (X) ku i , U' (X) V')

is an isomorphism in the special case U = ku i .

Since kU i is one-dimensional, this amounts to checking that the map

A' : U' (X) Hom(V, V') ---+ Hom(V, U' (X) V') (2.4)

defined by
A'(U' (X) f)(v) = u' (X) f(v)
is an isomorphism. By assumption, we also have U' = EBiEI' ku~ for some
finite basis {uiLEI'. We again use (1.6-1.7) and the fact that the direct
product over the finite set I' is the same as the direct sum. We get

U' (9 Hom(V, V') ~ EB ku~ (X) Hom(V, V')

iEI'

and
Hom(V, U' (X) V') ~ II Hom(V, ku~ (X) V').
iEI'

This allows us to break A' into the direct product of the maps

A' : ku~ (X) Hom(V, V') ---+ Hom(V, ku~ (X) V').

In this special case, A' is given by A' (u~ (X) f) (v) = u~ (X) f (v), which is clearly
an isomorphism. Hence, so is the map A' of (2.4), which concludes the proof.
There are similar arguments in the remaining two cases. D

We deduce two corollaries involving the dual vector space V* = Hom(V, k)

of a vector space V. For the first one, we specialize Theorem 2.1 by taking
U ' = Vi = k.

Corollary 11.2.2. The map A : U* (X) V* ---+ (V (X) U)* is an isomorphism

provided U or V are finite-dimensional.

For the second corollary, we take U = Vi = k in Theorem 2.1.

28 Chapter II. Tensor Products

Corollary II.2.3. The map AU v : V U* ---) Hom(U, V) given for U E U,

v E V and a E U* by ,
AU,v(V a)(u) = a(u) v (2.5)
is an isomorphism if U or V are finite-dimensional. In particular, if V is
a finite-dimensional vector space, the map AV,v is an isomorphism
V V* ~ End(V).
We now wish to express the general map A of (2.2) in terms of the special
maps A defined in Corollaries 2.2-2.3 and of the flip. This is done in the
following lemma which will be useful later. Note that the map AU UI 16! AV VI
is invertible when either U or U I , and either V or Vi are finite-dir'nension'al.
Lemma II.2.4. The following diagram commutes:
AU,UI@AV,V I
U I U* Vi 16! V* Hom(U, U I ) Hom(V, VI)
lid@Tu*,v'@id

U I Vi U* 16! V*
lid@id@A

U I Vi 16! (V U)* Hom(V u, U I VI)

PROOF. Easy. o
There is another important operation on linear homomorphisms that we
have not yet discussed. It is the composition (g,1) I--> go f of two linear
maps. This operation is bilinear and leads, for any triple (U, V, W) of vector
spaces, to the map
Hom(V, W) 16! Hom(U, V)~ Hom(U, W).
Under some finite-dimensionality conditions, we can express the composi-
tion in simpler terms again involving the special maps A of Corollary 2.3
as well as the evaluation map
ev v : V* 16! V -+ k
which is defined as usual, namely by
evv(a 16! v) = < a, v> = a(v) (2.6)
for any linear form a and any vector v of V.
Lemma 11.2.5. The square
id@evv@id
WV* vu* WU*
1
---------+

AV,W@AU,V
o
1 AU,W

Hom(V, W) Hom(U, V) -----+ Hom(U, W)

commutes.
PROOF. Easy. o
 II.3 Duality and Traces 29

11.3 Duality and Traces

All vector spaces considered in this section are assumed to be finite-dimen-
simlal. If V is such a vector space, we denote a basis of V by {Vi} i using
the corresponding lower-case letter for vectors. The dual basis in the dual
vector space V* is denoted {vi L. Using these bases, the evaluation map
can be redefined by

(3.1)

Let us express the isomorphism AU.V : V U* ~ Hom(U, V) of Corollary

2.3 in terms of bases. Let f : U ---+ V be a linear map. Using bases for U
and V, we have
(3.2)

for some family (fJ)ij of scalars. It is easily checked that

f = AU,v (2: fJ Vi u j ). (3.3)

2J

In particular, taking for f the identity of V, we get

(3.4)

This allows us to define the coevaluation map of any finite-dimensional

vector space V as the linear map 8v : k ---+ V V* defined by

(3.5)

By its very definition, the map 8v is independent of the choice of a basis. We

now record some relations between the evaluation and coevaluation maps.
These relations will turn out to be fundamental when we define duality in
categories in Chapter XIV.

Proposition 11.3.1. The composition of the maps

ovidv
V ----+,VV * idvev v
V ,V

is equal to the identity of V. Similarly, the composition of the maps

V* id v * Ov , V* V V* ev v id v * , V*

is equal to the identity of V* .

PROOF. Immediate. o
30 Chapter II. Tensor Products

Let us recall the operation of transposition. For a linear map f : U -7 V,

define its transpose 1* : V* -7 U* as the linear map defined for all a E V*
and all u E U by
< 1*(a),u > = < a,f(u) >. (3.6)
In other words, 1* is the unique linear map such that the square
V* U f*i~
1id v * / (3.7)
V*V ~

commutes. The transposition may be recovered from the evaluation and

coevaluation maps as shown in the following result whose proof is left to
the reader.
Proposition 11.3.2. Let f : U -7 V be a linear map. Then the transpose
1* is equal to the composition of the maps
V* id v * Ou ) V* U U* id v * /id u * ) V* V U* ev v id u * ) U*.

Observe that if (3.2) holds, then

(3.8)

We thus see that transposition amounts to exchanging upper and lower

indices. We generalize this as follows. Let f be a linear map from V W
to X Y. Using bases on these spaces, we define the partial transposes

f+ : X* W -7 V* Y and fX: V Y* -7 X W*

by
f+(x i w j ) =L fk; v k Ye (3.9)
k,

and
r(v i y j ) = L fi~j xk w e (3.10)
k,e
if
f(v i Wj) = L fi~e xk Ye (3.11)
k,e
Lemma 11.3.3. The definitions of f+ and fX are independent of the choice
of bases. We also have

PROOF. Left to the reader. D

 11.3 Duality and Traces 31

The isomorphism AV v of Corollary 2.3 allows one to define the trace of

an endomorphism in a' finite-dimensional vector space V. The trace tr :
End(V) ---+ k is defined as the composition
A-1 T * ev
End(V)~V @ V*~~-.v* @ V~k. (3.12)

Proposition 11.3.4. Let f and g be endomorphisms of a finite-dimension-

al vector space V.
(a) The trace satisfies the relation
tr(f 0 g) = tr(g 0 f). (3.13)

(b) If (fJ)ij is the matrix of f in a basis of V, then

tr(f) =L ff (3.14)

(c) We also have

tr(f*) = tr(f). (3.15)
PROOF. (a) By linearity, it suffices to prove (3.13) for

f = AV,v(V @ a) and g = AV,v(W @ (3)

where v, wE V and a, (3 E V*. We have fog = Avv(a(w) v@(3) by Lemma

2.5. Consequently, tr(f 0 g) = a(w) (3(v) , which clearly equals tr(g 0 f).
(b) From Relations (3.2-3.3) we derive

tr(f) = L fj < vj,vi >= L ff.

ij

(c) Relation (3.15) follows from (3.8) and (3.14). o

The next result expresses the trace in terms of the evaluation and co-
evaluation maps and of the flip.
Proposition 11.3.5. The trace of f : V ---+ V is equal to the composition
of the maps

We close these generalities with the partial traces of an endomorphism f

of U@V. By Theorem 2.1 the map f@g 1--+ A(f@g)OTU,v is an isomorphism
>: from End(U) @ End(V) onto End(U @ V). We define trl and tr2 by the
following commutative diagram.
trl
End(V) f-- End(U@V) End(U)
r~ r~ id0tr
r~ (3.16)
tr0id
k@End(V) f--- End(U) @ End(V) ----> End(U) @k
32 Chapter II. Tensor Products

Lemma 11.3.6. If f(u i @ vj ) = L:k, fi~ Uk @ V on some bases of U and

V, then

(3.17)

PROOF. Left to the reader. D

II.4 Tensor Products of Algebras

Given algebras A and B, we put an algebra structure on the tensor product
A@Bby
(a@b)(a' @b') = aa' @bb' (4.1)
where a, a' E A and b, b' E B. We call A @ B the tensor product of the
algebras A and B. Its unit is 1@ 1. Defining iA(a) = a@ 1 and iB(b) = 1 @b,
we get algebra morphisms i A : A ---> A @ Band i B : B ---> A @ B. The
following relation holds in view of (4.1):

( 4.2)

for all a E A and b E B. The tensor product of algebras enjoys the following
universal property.

Proposition 11.4.1. Let f : A ---> C and g : B ---> C be algebra morphisms

such that, for any pair (a, b) E A x B, the relation f(a)g(b) = g(b )f(a) holds
in C. Then there exists a unique morphism of algebras f @ g : A @ B ---> C
such that (J @ g) 0 i A = f and (J @ g) 0 i B = g.

We can rephrase Proposition 4.1 by saying that Hom Alg (A @ B, C) is

the subset of HOmAlg(A, C) x HOmAlg(B, C) consisting of all pairs (J, g) of
morphisms whose images commute in C. In particular, if C is commutative
we have

HOmAlg(A@B, C) ~ HOmAlg(A, C) x HOmAlg(B, C). (4.3)

PROOF. Any element of A @ B is a finite sum of elements of the form a @ b.

Therefore, by (4.2), f @ 9 (if it exists) has to be of the form

(J@g)(a@b) = (J@g)(iA(a))(J@g)(iB(b)) = f(a)g(b).

This proves the uniqueness assertion. As for the existence of the map f @ g,
one checks that the previous formula defines an algebra morphism. This
 1I.4 Tensor Products of Algebras 33

uses the commutativity assumption as follows:

(f g)(a b)(f g)(a' b') f( a )g(b )f( a')g(b')

f( a )f( a')g(b )g(b')
f(aa')g(bb')
(f g)(aa' bb').

o
We apply Proposition 4.1 to a situation encountered in Chapter I.

Proposition 11.4.2. Let A = k{ X} / I be a quotient of the free algebra on a

set X. Take two copies X' and X" of X. Let I' and 1" be the corresponding
ideals in k{ X'} and k{ X"}. Then the tensor product algebra A A is
1:somorphic to the algebra

A0 2 = k{X' U X"}/(I',I",X'X" - X"X')

where X' U X" denotes the disjoint union of the two copies and where
X' X" - X" X' is the two-sided ideal generated by all elements of the form
x' x" - x" x' with x' E X' and x" E X".

PROOF. For any x E X we denote the corresponding copy in X' [resp. in

X"] by x' [resp. by x"]. Setting cp'(x) = x' and cp"(x) = x" defines algebra
morphisms cp', cp" : A --+ A 0 2. Since x' y" = y" x' by definition of A 02,
we have cp'(x)cp"(y) = cp"(y)cp'(x) for any pair (x,y) of elements of X. By
Proposition 4.1 there exists an algebra morphism cp : A A --+ A 02 such
that cp(xy) = x'y".
Conversely, we get an algebra morphism 'ljJ from A 0 2 to A A by setting
'ljJ(x') = xl and 'ljJ(x") = lx where x' E X' and x" E X". One easily
checks that cp and 'ljJ are inverse of each other. 0

We retain from the previous statement that one passes from A 02 to

A A by replacing the copy x' of x by x 1 and the copy x" by 1 x and
vice versa. Let us apply this recipe to the constructions of 1.4-5. Denoting
M(2), GL(2) or 8L(2) by G, we see that in all three cases the algebra G02
defined in 1.4-5 is isomorphic to the tensor product algebra G G. We can
thus rewrite the map ~ of Proposition I.4.1 as the algebra morphism from
G to G G determined by

~(a)=aa+bc, ~(b) = a b + b d,
~(c)=ca+dc, ~(d) = c b + d d.
34 Chapter II. Tensor Products

We rewrite these four relations in the compact matrix form

Lemma 1.5.2 implies that

6.(ad - be) = (ad - be) Q9(ad - be). (4.5)

II.5 Tensor and Symmetric Algebras

Let V be a vector space. Define TO (V) = k, TI(V) = V and Tn(v) = v@n
(the tensor product of n copies of V) if n > 1. The canonical isomorphisms

induce an associative product on the vector space T(V) = EBn>O Tn(v).

Equipped with this algebra structure, T(V) is called the tensor algebra of
V. The product in T(V) is explicitly given by

(Xl Q9 ... Q9 x n )(x n+1 Q9 . .. Q9 x n+m ) = Xl Q9 . .. Q9 Xn Q9 Xn+l Q9 . .. Q9 Xn+m

(5.1)
where Xl' ... ' Xn ' Xn+l , ... , Xn+m are elements of V. The unit for this prod-
uct is the image of the unit element 1 in k = TO (V). Let iv be the canonical
embedding of V = TI(V) into T(V). By (5.1) we have

(5.2)

which allows us to set

(5.3)

whenever Xl' ... 'X n are elements of V.

Proposition 11.5.1. (a) The algebra T(V) is graded such that Tn(V) is
the subspace of degree n homogeneous elements.
(b) For any algebra A and any linear map f : V ~ A, there exists a
unique algebra morphism f : T(V) ~ A such that f 0 iv = f. Consequently,
the map f f---> f 0 iv is a bijection

HOmAlg(T(V),A) ~ Hom(V, A). (5.4)

(c) Let I be an indexing set for a basis of the vector space V. Then the
tensor algebra T(V) is isomorphic to the free algebra k{ I}.
PROOF. Part (a) is clear. Let us prove Part (b). If f exists, it has to be of
the form
 11.5 Tensor and Symmetric Algebras 35

in view of (5.3). This proves the uniqueness of f. As for its existence, one
checks immediately that the previous formula defines an algebra morphism
from T(V) into A.
(c) By Corollary 1.5, if {eiLEI is a basis of V, then {ei1 ein }i 1 ,,,. ,inEI is
a basis of the vector space Tn (V). When n runs over the set of non-negative
integers we get a basis of T(V) which is clearly in bijection with Ii basis of
k{I}. This bijection induces an isomorphism between both vector spaces.
The product on T(V) corresponds to the concatenation in k{ I} under this
isomorphism.
Let us give another, less pedestrian, proof of Part (c). By (1.5), (1.8),
(5.4) and (1.2.2) we have the following chain of natural bijections:
HOmAlg(T(V) , A) ~
Hom(V,A)
~
Hom(EB ke i , A)
iEI
~
II Hom(kei,A)
iEI
~
Homset(I, A)
~
HOmAlg(k{I}, A).
Let a be the composition of these bijections. First, take A = T(V) and
define <p = a(idT(v)); this is an algebra morphism from k{I} to T(V). Now
take A = k{I} and define 'ljJ = a-l(idk{I}); this is an algebra morphism
from T(V) to k{I}. We claim that <p and 'ljJ are isomorphisms between T(V)
and k{I}. First, observe that the bijection a is natural, which means that
for any algebra morphism f : A --- A' we have
foa(w)=a(fow)

for any w E HOmAlg(T(V) , A). Let us now compose <p and 'ljJ. On the one
hand, we get

'ljJ 0 <p = 'ljJ 0 a(idT(V)) = a('ljJ 0 idT(v)) = a('ljJ) = idk{I} ,

whereas on the other hand, we have

a(<p 0 'ljJ) = <p 0 a('ljJ) = <p 0 idk{I} = <p,

whence cp 0 'ljJ = a-l(cp) = idT(v)' o
Let us define symmetric algebras. If V is a vector space, the symmetric
algebra S(V) is the quotient S(V) = T(V) /I(V) of the tensor algebra T(V)
by the two-sided ideal I(V) generated by all elements xy - yx where x and
y run over V. If Xl' ... ,X n are elements of V, we again denote by xl ... xn
the class of Xl ... xn in S(V). The image of Tn(V) under the projection
of T(V) onto S(V) is denoted sn(v). Let iv be the canonical map from
V = TI(V) to S(V).
36 Chapter II. Tensor Products

Proposition 11.5.2. (a) The algebra S(V) is commutative, and is graded

such that sn(v) is the subspace of degree n homogeneous elements.
(b) For any algebra A and any linear map f : V ---> A such that

f(x)f(y) = f(y)f(x)

for any pair (x, y) of elements of V, there exists a unique algebra morphism
f: S(V) ---> A such that f 0 iv = f
(c) If I is an indexing set for a basis of V, then the symmetric algebra
S(V) is isomorphic to the polynomial algebra k[I] on the set I.
(d) If V' is another vector space, we have an algebra isomorphism

S(V EEl V') ~ S(V) S(V'). (5.5)

Part (b) implies that the map f 1--+ f 0 iv is a bijection

HOmAlg(S(V),A) ~ Hom(V,A) (5.6)

when the algebra A is commutative.

PROOF. We leave (a)-(c) as an exercise. Let us give a short proof of (d).
Using (1.5), (4.3) and (5.6), we have the chain of natural bijections

HOmAlg(S(VEElV'),A) ~ Hom(VEElV',A)
~ Hom(V,A) x Hom (V' , A)
~ HOmAlg(S(V) , A) x HOmAlg(S(V'), A)
~ HOmAlg(S(V) S(V'), A).

We then successively take A to be S(V EEl V') and S(V) S(V'), which
produces isomorphisms between these algebras, as in the second proof of
Part (c) of Proposition 5.1. 0

II.6 Exercises
1. If f and l' [resp. g and g'] are composable linear maps, show that

(f' 0 1) (g' 0 g) = (f' g') 0 (f g).

2. Prove that if f is a surjective linear map, then so is f id v for any

vector space V. What about the kernel of f id v ?

3. Prove that the map oX of (2.2) is injective.

4. Let U, V be finite-dimensional vector spaces, f [resp. g] be an endo-

morphism of U [resp. of V]. Show that tr(f g) = tr(f) tr(g).
 II.6 Exercises 37

5. Let A = EBi>o Ai and A' = EBi>O A~ be graded algebras. Show that

the tensor p~oduct algebra A 0 A' is graded with

(A0A')n= E9 A i 0Aj.
i+j=n

6. (Exterior algebra) For any vector space V we define the exterior alge-
bra (or Grassmann algebra) A(V) as the quotient A(V) = T(V) / I' (V)
of T(V) by the two-sided ideal I' (V) generated by the elements x 0 x
where x runs over V. If Xl"" ,X n are elements of V, denote by
Xl 1\ .. . 1\ xn the class of Xl 0 ... 0 xn in A(V). The subspace of A(V)
generated by the elements Xl 1\ ... 1\ xn is denoted N'(V). Let iv
be the canonical map from V = TI (V) to A(V). Prove the following
statements.

(a) The algebra A(V) is graded such that An(v) is the subspace of
degree n homogeneous elements.
(b) For any algebra A and any linear map f : V -+ A satisfying
f(X)2 = 0 for all X E V, there exists a unique algebra morphism
! : A(V) -+ A such t.hat ! 0 iv = f.
(c) Let I be an ordered set indexing a basis {eJiEI of V. Then the
set {e i1 1\ ... 1\ einL,<...<inEl is a basis of N'(V).
(d) Assume V of finite dimension d. Prove that

L dim(N' (V)) t n = (1 + t)d.

n2:0

7. (Symmetric and antisymmetric tensors) The symmetric group Sn has

a left action on Tn(v) given by

where CJ E Sn and Xl'" ., Xn E V. Define two endomorphisms ~ (the

symmetrization operator) and A (the antisymmetrization operator)
of Tn(v) by

~(a) = L CJ(a) and A(a) = L c(CJ)CJ(a)

CTESn aESn

where c(CJ) is the sign of the permutation CJ. A tensor a of Tn(v) is

symmetric [resp. antisymmetric] if CJ(a) = a [resp. CJ(a) = c(CJ)CJ(a)]
for any permutation CJ. The subspace of symmetric [resp. antisym-
metric] tensors of Tn (V) is denoted S~(V) [resp. A~!(V)]. Prove that

(a) ~(Tn(v)) C S~(V) and A(Tn(V)) c A~(V),

38 Chapter II. Tensor Products

(b) if n! is invertible in the field k, the previous inclusions are equal-

ities and the composition ofthe inclusion s~(V) ---+ Tn(v) [resp.
of the inclusion A~(V) ---+ Tn(v)] with the canonical projection
Tn (V) ---+ sn (V) [resp. with the projection Tn (V) ---+ An (V)] is
an isomorphism.

8. Let A0 V be a free A-module. Prove that the space of A-linear maps

from A 0 V to any A-module W is isomorphic to Hom(V, W).

II.7 Notes
For more details on the tensor, symmetric and exterior algebras as well as
on the subspaces S~(V) and A~(V) of Exercise 7, see [Bou70], Chap. 3.
Chapter III
The Language of Hopf Algebras

In this chapter we introduce the fundamental concepts of coalgebras, bial-

gebras, Hopf algebras and comodules which we shall use extensively in the
sequel. We shall also prove that the algebras GL(2) and 8L(2) of Chapter
I are Hopf algebras.

IILl Coalge bras

The concept of a coalgebra is dual to the concept of an algebra in the
following sense. Paraphrasing the definition of an algebra in I.l, we can say
that an algebra is given by a triple (A, JL, 7]) where A is a vector space and
JL : A A -> A and 7] : k -> A are linear maps satisfying the following
axioms (Ass) and (Un).
(Ass): The square
I'id
AAA ~ AA
1id1' 11' (1.1)
I'
AA ----+ A
commutes.
(Un): The diagram
1)id id(1)
kA ------+ AA ~ Ak
'\.~ 11' /~ (1.2)
A
commutes.
40 Chapter III. The Language of Hopf Algebras

The axiom (Ass) expresses the requirement that the multiplication /1 is

associative whereas Axiom (Un) means that the element 1](1) of A is a left
and a right unit for /1. The algebra A is commutative if, in addition, it
satisfies the axiom
(Comm): The triangle

(1.3)
A
commutes, where T A A is the flip switching the factors: T A A (a0a') = a' 0a.
A morphism of algebras f : (A, /1, 1]) ---> (A', /1', 1]') is a linear map f from
A to A' such that
/1' 0 (J 0 J) = f 0 /1 and f 0 1] = 1]'. (1.4)
We now get the definition of a coalgebra by systematically reversing all
arrows in the previous diagrams.
Definition 111.1.1. (a) A coalgebra is a triple (C, 6., E) where C is a vector
space and 6. : C ---> C 0 C and E :. C ---> k are linear maps satisfying the
following axioms (Coass) and (Coun).
(Coass): The square

(1.5)

commutes.
(Coun): The diagram
k0C ~ C0C ~ C0k
""~ r~ /'~ (1.6)
C
commutes. The map 6. is called the coproduct or the comultiplication while
E is called the counit of the coalgebra. The squares (1.5-1.6) express that
the coproduct 6. is coassociative and counital.
If, furthermore, the triangle (Cocomm)
C
(1.7)
Te,c
~

commutes, where Te,e is the flip, we say that the coalgebra C is cocommu-
tative.
(b) Consider two coalgebras (C, 6., E) and (C', 6.', E'). A linear map f
from C to C' is a morphism of coalgebras or a coalgebra morphism if
(J 0 J) 0 6. = 6.' 0 f and E = E' 0 f. (1.8)
 IILl Coalgebras 41

It is easily checked that the composition of two morphisms of coalgebras

is again a morphism of coalgebras.
Let us give a few examples of coalgebras.
Example 1. (The ground coalgebra) The field k has a natural coalgebra
structure with ,0.(1) = 1 1 and E(1) = 1. Moreover, for any coalgebra
(C,,0.,c), the map 10: C --+ k is a morphism of coalgebras.

Example 2. (Opposite coalgebra) For any co algebra C = (C,,0., c) set

,0.0p = TC,C 0,0.. (1.9)

Then (C,,0. op, c) is a coalgebra which we call the opposite coalgebra and
denote by ccoP.
The next result relates algebras and coalgebras.

Proposition 111.1.2. The dual vector space of a coalgebra is an algebra.

PROOF. Let (C,,0., c) be a coalgebra. Recall the map A: C*C* --+ (CC)*
of Corollary II.2.2. Set); = A 0 T c* C*' Define A = C*, JL = ,0.* 0 ); and
7] = 10* where the superscript * on 'a linear map indicates its transpose.
Then (A, JL, 7]) is an algebra (use the commutative diagrams (1.1-1.2) and
(1.5-1.6)). D

Example 3. (Coalgebra of a set) Let X be a set and C = k[X] = EBxEX kx

be the vector space with basis X. We put a coalgebra structure on C by
defining
,0.(x) = x x and c(x) = 1 (1.10)
where x E X. The dual algebra C* is the algebra of functions on X with
values in k. Indeed, a linear form f on C is determined by its values on the
basis X. Let l' be another linear form. Then

(f 1')(x) = JL(f 1')(x) = );(f 1') (,0.(x)) = f(x)1'(x).

Finally, the unit of the algebra C* is given by the constant function c.

We shall later return to this example when X has, in addition, a group
structure.
In general, the dual vector space of an algebra does not carry a natural
coalgebra structure. Nevertheless, we have the following result in the finite-
dimensional case (see also Section 9).

Proposition 111.1.3. The dual vector space of a finite-dimensional alge-

bra has a coalgebra structure.

PROOF. Let (A, JL, 7]) be a finite-dimensional algebra. Then the map); from
A * A * to (A A) * is an isomorphism, which allows us to define ,0. by
42 Chapter III. The Language of Hopf Algebras

~= >:-1 op,*. We also set c = T]*. Using the commutative diagrams (1.1-
1.2) and (1.5-1.6), one checks that (A*,~,c) is a coalgebra. 0
Example 4. (The matrix coalgebra) Let A = Mn(k) be the algebra of n x n-
matrices with entries in k. Denote by Eij the matrix with all entries equal
to 0, except for the (i, j)-entry which is equal to 1. The set of matrices
Eij (1 S; i,j S; n) is a basis of Mn(k). Let {Xij} be the dual basis. Then
A * is the co algebra defined by
n
~(Xij) = :L>ik Xkj and C(Xij) = 8ij (1.11 )
k=l

Indeed, we have

C(Xij) = xij (T](l)) = xil2~Ekk) = L8ik 8kj = Dij

k k

and

xij(p,(Ekt Emn))
DtmXij(Ekn)
8tmDikDjn
L DikDtp8pmDjn
p

>:(LX iP Xpj) (Ekt Emn)'

p

Example 5. (Tensor product of coalgebras) The tensor product G G'

of two coalgebras (G,~, c) and (G',~', c') has a coalgebra structure with
comultiplication (id TC,G' id) 0 (~ ~') and counit c c'.
We return to Example 3.

Proposition 111.1.4. Let X and Y be two sets and X x Y be the product

set. There exists an isomorphism of coalgebras

k[X] k[Y] ~ k[X x Y].

PROOF. The isomorphism is given on the basis {x Y}CX,Y)EXxY of the

tensor product k[X] k[Y] by

1jJ(X y) = (x, y). (1.12)

 111.1 Coalgebras 43

It is clear that

('lj! 0 'lj!)(id 0 7 0 id)(A 0 A)(x 0 y) = (x, y) 0 (x, y) = A'lj!(x 0 y)

and e'lj!(x 0 y) = 1 = e(x)e(y), which shows that 'lj! is a morphism of
coalgebras. 0

We shall also need the following concept.

Definition III.1.5. Let (0, A, e) be a coalgebra. A subspace I of 0 is a
co ideal if A(I) c 100 + 001 and 10(1) = o.
When I is a coideal, then A factors through a map A from 0/1 to

000/(I00+00I) =0/100/1.

Similarly, the counit factors through a map ~ : 0/1 - t k. Then clearly, the
triple (0/1, A,~) is a coalgebra. It is called the quotient-coalgebra. We shall
give examples later.
Notation 1.6. We now present Sweedler's sigma notation which we shall
use continually in the sequel. If x is an element of a coalgebra (0, A, e), the
element A (x) of 0 0 0 is of the form

(1.13)

In order to get rid of the subscripts, we henceforth agree to write the sum
(1.13) in the form
A(x) = LX'
0x". (1.14)
(x)

Using (1.14) we may express the coassociativity of A, i.e., the commuta-

tivity of the square (1.5), by

L (L(X')' 0 (x')") 0 x" = LX' 0 (L (x")' 0 (x")"). (1.15)

(x) (x') (x) (x")

By convention again, we identify both sides of (1.15) with

L x' 0 x" 0 x"', (1.16)

(x)

also written L:(x) X(l) 0 X(2) 0 x(3). If we apply the comultiplication to

(1.16), we get the following three equal expressions

L A(x') 0 x" 0 XIII, LX' 0 A(x") 0 XIII, LX' 0 x" 0 A(x"')

(xl (x) (x)
44 Chapter III. The Language of Hopf Algebras

which we agree to write

LX' x" x'" x"" (1.17)

(x)

or L:(x) x(1) X(2) x(3) x(4). More generally, let D. (71) : C ----) c(n+1)

be defined inductively on n ;:: 1 by D. (1) = D. and

"(71)
u = ("d
u 1 e 0 (n-l) ) 0 ,,(71-1)
u = (d
1 e;s?(n-l) U") 0 U,,(71-1) . (1.18)

Then by convention, we write

D.(n)(x) = L x(1) ... x(n+l). (1.19)

(x)

These conventions and the coassociativity of D. imply for instance that

(ide D. id e02 ) ( L x(1) x(2) x(3) x(4)

(x)
= L XCI) x(2) x(3) X(4) x(5)
(x) (1.20)

Using the conventions (1.14), the condition (1.6) for counitality may be
reformulated for any x E C as

LE(X')X" = x = LX'E(X"). (1.21)

(x) (x)

As a consequence of (1.21) and of (1.19), we get identities such as

L x(l) E(X(2) x(3) x(4) x(5) = L x(1) X(2) x(3) X(4). (1.22)
(x) (x)

Indeed, the left-hand side may be rewritten as

L x(l) (E id)(D.(x(2)) x(3) x(4).

(x)

Then apply (1.21).

The co algebra C is co commutative if

LX' x" = LX" x' (1.23)

(x) (x)

for all x E c.
 111.2 Bialgebras 45

The left Relation (1.8) defining a coalgebra morphism can be reformu-

lated as
Lf(x') f(x") = L
f(x)' f(x)". (1.24)
(x) (f(x))

The comultiplication of the tensor product G G' of the coalgebras G

and G' (see Example 5) is given for x E G and y E G' by
~(x y) = L (x y)' (x y)" = L (x' y') (x" y"). (1.25)
(x<Sly) (x)(y)

We invite the reader to play with Sweedler's sigma notation in order to

acquire some familiarity with this most useful convention.

III. 2 Bialgebras
Let H be a vector space equipped simultaneously with an algebra structure
(H,fL,ry) and a coalgebra structure (H,~,c). Let us discuss two compati-
bility conditions between these two structures. We give H H the induced
structures of a tensor product of algebras (see II.4) and of a tensor product
of co algebras (see Section 1, Example 5).
Theorem 111.2.1. The following two statements are equivalent.
(i) The maps fL and ry are morphisms of coalgebras.
(ii) The maps ~ and care morphisms of algebras.
PROOF. It consists essentially in writing down the commutative diagrams
expressing both statements. The fact that fL is a morphism of coalgebras is
equivalent to the commutativity of the two squares
HH ....!!:... H HH
1(id<SlT<SIid) (ll.<SIll.) lll. l~
(HH)(HH) ~ HH H
whereas the fact that ry is a morphism of coalgebras is expressed by the
commutativity of the two diagrams
k ...!!..... H
~ id ,/ c o
k
Observe that these four commutative diagrams are exactly the same as the
following four diagrams whose commutativity express the fact that ~ and
care morphisms of algebras:
(H H) (H H)
1(~<SI~)(id<SlT<SIid)
HH
46 Chapter III. The Language of Hopf Algebras

and e<8le 7)
H0H ------+ k0k k --t H
l~ lid \, id ./e
e
H --t k k

This leads to the following definition.

Definition III.2.2. A bialgebra is a quintuple (H, {t, 'fl, A, c) where (H, {t, 'fl)
is an algebra and (H, A, c) is a coalgebra verifying the equivalent conditions
of Theorem 2.1. A morphism of bialgebras is a morphism for the underlying
algebra and coalgebra structures.

In the sequel, we shall mainly use Condition (ii) of Theorem 2.1 to define
a bialgebra structure. Using the conventions of 1.6, we see that the condition
A(xy) = A(x)A(y) is expressed for any pair (x, y) of elements in a bialgebra
by
L(xy)' 0 (xy)" = x'y' 0 x"y".L (2.1)
(xy) (x)(y)

We also have

A(l) = 101, c(xy) = c(x)c(y), c(l) = 1. (2.2)

The following proposition is easy to check.

Proposition III.2.3. Let H = (H, {t, 'fl, A, c) be a bialgebra. Then

HCOP = (H , f"'"I/. 71
'n
AOP
"
c)

and HOpcop = (H,{tP, 'fl,Aop, c) are bialgebras.

Example 1. By Propositions 1.2-1.3 the dual vector space H* of a finite-

dimensional bialgebra H has a natural bialgebra structure.
Example 2. In Example 3 of Section 1 we associated a coalgebra k[X] to
a set X. Assume now that X comes with a unital monoid structure, i.e.,
with an associative map {t : X X X --t X having a left and right unit e.
The map {t induces an algebra structure on k[X] with unit e. We have

A(xy) = xy 0 xy = (x 0 x)(y 0 y) = A(x)A(y)

and c(xy) = 1 = c(x)c(y), which implies that the maps A and care
morphisms of algebras. Thus k[X] becomes a bialgebra.
If, in addition, X is a finite set, then the dual of k[X] also is a bialgebra.
We have already observed that the algebra structure of the dual is the
usual algebra structure of the space of k-valued functions on X. An easy
 III.2 Bialgebras 47

computation shows that the comultiplication and the co unit on the algebra
of functions are given by

6. (f) (x y) = j(xy) and c(f) = j(e). (2.3)

Example 3. (The bialgebra M(n)) Let M(n) = k[xll, ... ,xnnl be the
polynomial algebra in n 2 variables {xijh:::;i,j:::;n' For all i,j, set
n

6.(x ij ) =L xik x kj and C(Xij) = Dij (2.4)

k=1

These formulas define morphisms of algebras 6. : M(n) -+ M(n) M(n)

and c : M(n) -+ k equipping M(n) with a bialgebra structure. When n = 2,
one recovers the bialgebra M(2) of 1.4.
We now endow the tensor algebra with a bialgebra structure.

Theorem 111.2.4. Given a vector space V, there exists a unique bialgebra

structure on the tensor algebra T(V) such that 6.( v) = 1 v + v 1 and
c( v) = 0 jar any element v oj V. This bialgebra structure is cocommutative
and jar all VI' ... , Vn E V we have

(2.5)

n-I
= 1 VI ... Vn + LL V(]"(l) ... V(]"(p) V(]"(p+l) ... V(]"(n) +V I ... Vn 1 (2.6)
p=1 (]"

where a runs over all permutations oj the symmetric gmup Sn such that

a(l) < a(2) ... < a(p) and a(p + 1) < a(p + 2) ... < a(n).
Such a permutation a is called a (p, n - p)-shuffie.
PROOF. By universality of the tensor algebra, there exist unique algebra
morphisms 6. : T(V) -+ T(V) T(V) and c : T(V) -+ k such that their
restrictions to V are given by the formulas of the theorem. Now consider
several elements vI' ... , vn in V. Formula (2.5) is a trivial consequence of
the multiplicativity of c.
Let us now compute 6.( vI ... v n ). We shall do this by induction on n.
Formula (2.6) holds for n = 1 by definition. Suppose it holds up to n-1 ~ 1.
Then we have the series of equalities

6.( VI ... vn )
= 6.(vI v n _ I )6.(v n )
48 Chapter III. The Language of Hopf Algebras

n-2
(10 VI' .. Vn - I +L L Vo-(I) ... v(J(p) 0 V(J(p+l) ... v(J(n-l)
p=1 (J

+ VI'" v n - I 01) (10 vn + vn 0 1)

n-2
10 VI'" Vn +L L V(J(I) ... v(J(p) 0 v(J(p+l) ... vo-(n-I)vn
p=1 (J
+v 1 ... v n - 1 0 vn + vn 0 VI ... v n- 1
n-2
+ L L(J
p=l
v(J(l) ... vo-(p)vn 0 Vo-(p+l) ... v(J(n-l) + VI'" vn 01

where (J runs over all (p, n -1- p)-shuffies of Sn-l' Let us rewrite the last
sum in the form
n-2
10 VI' .. Vn +L L Vp(l) ... Vp(p) 0 Vp(p+l) ... Vp(n_l)Vn
p=1 p

+ VI' .. Vn - 1 0 Vn + Vn 0 VI' .. Vn - 1
n-l
+ LL VT(l) ... VT(p_I)V n 0 VT(p) ... VT(n-l) + VI'" Vn 0 1
p=2 T

where p runs over all (p, n - 1 - p)-shuffies of Sn-I and T runs over all
(p - 1, n - p )-shuffies permuting the set {I, ... , n} \ {p}. Now observe that
if (J E Sn is a (p, n - p )-shuffie, then either d n) = n, hence the restriction
p of (J to Sn-l is a (p, n -1- p)-shuffie, or (J(p) = n, hence T = (J acting on
{l, ... , n} \ {p} is a (p -1, n - p)-shuffie. This completes the proof of (2.6).
It remains to prove the coassociativity, the counitality and the cocom-
mutativity of ~. The counitality results from an easy computation using
(2.5) and (2.6). The co commutativity is a consequence of the fact that the
permutation

(
p+1
1
p+2
2 ...
...
p
n
p+1
1
p+2
2
...
...
n)
P

switches (p, n - p )-shuffies and (n - p, p )-shuffies. As for the coassociativity,

one may check it directly using (2.6). But, we rather observe that ~ :
T(V) --> T(V) 0T(V) is induced by the diagonal map 8(v) = (v, v) from V
into V EB V. The coassociativity of ~ then results from the obvious relation
(80id)o8=(id08)o8. D

We now introduce the concept of a primitive element.

Definition 111.2.5. Let (C,~,c) be a coalgebra. An element x of C is
primitive if we have
~(x)=10x+x01.
 111.3 Hopf Algebras 49

We denote by Prim(C) the subspace of all primitive elements of c.

Proposition 111.2.6. If x is a primitive element of a bialgebra, then we

have c(x) = o. If y is another one, then the commutator [x, y] = xy - yx
is primitive too.

PROOF. By definition of the counit and of a primitive element we have

x = c(l)x + c(x)l = x + c(x)l.

The vanishing of c( x) follows immediately. As for the second assertion, we
have

6.(xy) = (1 x + x 1)(1 y + y 1) = 1 xy + x y + y x + xy l.
We deduce
6.([x, y]) = 1 [x, y] + [x, y] 1,
which implies that [x, y] is primitive. D

The generators v E V of the tensor algebra T(V) are primitive by The-

orem 2.4. Let H be a bialgebra and xl' ... ,xn be primitive elements of H.
Consider a vector space V with basis {vI' ... , v n }. There is a unique alge-
bra morphism f from the tensor algebra T(V) to H such that f( Vi) = Xi
for all i.
Proposition 111.2.7. The map f : T(V) ---> H is a morphism of bialge-
bras.

PROOF. We have to check that

c(f(~)) = c(O and (f f)6.(~) = 6.(f(0) (2.7)

for all ~ E T(V). Since all maps involved in (2.7) are algebra maps, it is
enough to check (2.7) when ~ = v E V. In this case (2.7) holds because Xi
is primitive and we have Proposition 2.6. D

As a consequence of Proposition 2.7, we see that for any set {Xl' ... ,x n }
of primitive elements in a bialgebra, 6.(XI' ... ,x n ) is given by Formula
(2.6) of Theorem 2.4 after replacing Vi by Xi.

III. 3 Bopf Algebras

Given an algebra (A,fJ.,ry) and a co algebra (C,6.,c) we define a bilinear
map, the convolution, on the vector space Hom( C, A) of linear maps from
C to A. By definition, if f, g are such linear maps, then the convolution
f * g is the composition of the maps
C~CC~AA~A. (3.1)
50 Chapter III. The Language of Hopf Algebras

Using Sweedler's sigma notation of 1.6, we have

U * g)(x) = :L f(x')g(x") (3.2)

(x)

for any element x E C. The convolution is clearly bilinear.

Proposition 111.3.1. (a) The triple (Hom(C, A), *, 7] 0 e:) is an algebra.
(b) The map AC A : A C* - t Hom( C, A) of Corollary II.2.3 is a mor-
phism of algebras !where A C* is the tensor product algebra of A and of
the algebra C* dual to the coalgebra C.

PROOF. (a) By (3.2), by the associativity of the product in A and by the

coassociativity of the coproduct in C we have

(U*g)*h)(x) =:L f(x')g(x")h(x"') =(!*(g*h))(x).

(x)

This proves that the convolution is associative. The map 7] 0 e: is a left unit
for the convolution in view of

((7]e:) * f)(x) =:L e:(x')f(x") =f(:L e:(x')x") =f(x),

(x) (x)

which results from (1.21). One proves similarly that 7] 0 e: is a right unit.
(b) Let a, bE A and a, (3 E C*. Then for x E C we have

( AC,A (a a) * AC,A (b (3) ) (x) :L a(x')(3(x") ab

(x)
(a(3) (x) ab
(Ac,A (ab a(3)) (x).

This proves that AC,A preserves the product. As for the unit, we have

(AC,A(1e:))(x) =e:(x)1 = (7] 0 e:)(x).

D

Example 1. When A = k the algebra structure (Hom(C, k), *, 7] oe:) on the

dual space C* is the same as the one defined in Proposition 1.2.
When (H, jL, 7],.6., e:) is a bialgebra we may consider the case C = A = H
and thus define the convolution on the vector space End(H) of endomor-
phisms of H.
Definition 111.3.2. Let (H, jL, 7],.6., e:) be a bialgebra. An endomorphism
S of H is called an antipode for the bialgebra H if

S * id H = id H *S = 7] 0 e:.
 111.3 Hopf Algebras 51

A Hopf algebra is a bialgebra with an antipode. A morphism of Hopf alge-

bras is a morphism between the underlying bialgebras commuting with the
antipodes.

A bialgebra does not necessarily have an antipode. But if it does, it has

only one. Indeed, if 8 and 8' are antipodes, then

A Hopf algebra with an antipode 8 will be denoted by (H, j.l, ry,~, c:, 8).
Using Sweedler's convention 1.6, we see that an antipode satisfies the
relations
L:: x'8(x") = c:(x)l = L:: 8(x')x" (3.3)
(x) (x)
for all x E H. In any Hopf algebra we have relations such as

L:: X(l) 0 X(2) 0 8(x(3) 0 x(4) 0 X(5) L:: X(l) 0 c:(x(2) 0 x(3) 0 X(4)
(x) (x)
L:: x(1) 0 X(2) 0 x(3) .
(x)

The first equality follows from (3.3), i.e., by definition of the antipode
while the second one follows from (1.21), i.e., from the Axiom (Coun).
Such computations will be performed later without further explanations.
We state the counterpart of Example 1 of Section 2.

Proposition 111.3.3. Let H be a finite-dimensional Hopf algebra with an-

tipode 8. Then the bialgebra H* is a Hopf algebra with antipode 8*.

PROOF. The endomorphism 8* of H* is the transpose of 8. Let us prove

the first equality in (3.3). For all 0 E H* and x E H we have

(L:: 0'8*(0")) (x) L:: 0'(x')8* (0") (x")

(a) (a)(x)

L:: 0'(x')0"(8x")
(a) (x)

0(L:: x'(8x"))
(x)
o(ryc:(x))
c:*ry*(o)(x).

One shows similarly that 2.:(a) 8*(0')0" = c:*ry*(o). o

Example 2. Let G be a monoid and k[G] the bialgebra of Section 2, Example
2. Then k[G] has an antipode if and only if any element x of G has an
52 Chapter III. The Language of Hopf Algebras

inverse, i.e., if and only if G is a group. Indeed, if S exists, by definition of

.6. we must have
xS(x) = S(x)x = c(x)l = 1
for any x E G. This implies that S(x) = x-I for x E G.
We state a few important properties of the antipode.

Theorem 111.3.4. Let (H, p" 7],.6.,10, S) be a Hopf algebra.

(a) Then S is a bialgebra morphism from H to HOP cOP, i. e., we have

S(xy) = S(y )S(x), S(l) = 1 (3.4)

for all x, y E Hand

(3.5)

(b) The following three statements are equivalent:

(i) we have S2 = id H ,
(ii) for all x E H we have L(x) S(x//)x' = c(x)l,
(iii) for all x E H we have L(x) x// S(x') = c(x) 1.
(c) If H is commutative or cocommutative, then S2 = id H .

The left relation in (3.5) can be reformulated under Sweedler's convention

1.6 as
L S(x)' S(x)" = L S(x//) S(x'). (3.6)
(S(x)) (x)

PROOF. (a) Let us start with (3.4). Define maps v, pin Hom(H H, H) by

v(x y) = S(y)S(x) and p(x y) = S(xy)

where x, y E H. We have to show that p = v. It is enough to prove that

p * p, = p, * v = 7]E. Now, by (1.21), (2.1) and (3.2)

(p*p,)(xy) = L p((xy)')p,((xy)//)
(xy)

L p(x' y')p,(x" y//)

(x)(y)

L S(x'y')x//y//
(x)(y)

L S((xy)')(xy)//
(xy)

7]E(XY)
 III.3 Hopf Algebras 53

On the other hand, we have

(/H v)(x y) = 2:= p,((x y)') v((x y)")

(x@y)

2:= x'y'S(yl)S(X")
(x)(y)

2:= x' (2:= y' S(yll)) S(X")

(x) (y)

2:= x'c(y)S(x")
(x)
ryc(x)ryc(y)
ryc(xy) ,
which is the same.
Applying (id * S)(x) = ryc(x) to x = 1, one gets S(l) = 1. This proves
(3.4).
Let us deal with (3.5). It is equivalent to prove t:J..oS = (SS)ot:J.. P. We
set p = t:J..oS and v = (SS)ot:J.. P. These are linear maps from H to HH.
We wish to show that p = v. This will follow from p*t:J.. = t:J..*v = (ryry)c,
which we prove now. On the one hand, by (1.21)

(p * t:J..)(x) 2:= t:J..(S(x') )t:J..(X") = t:J.. (2:= S(x')x")

(x) (x)
t:J..(ryc(x)) = ((ry ry)c) (x)
for all x E H. On the other hand, we have

(t:J..*v)(x) = 2:= t:J..(x')((SS)(t:J..OP(x")))

(x)
2:= (x' x") ( S(X"") S(x"'))
(x)
2:= X'S(X"") x" S(x"')
(x)
2:= x'S(x"') c(x")l
(x)

2:= x' c(x")S(x"') 1

(x)
2:= x'S(x") 1
(x)
c(x)l 1
(ry ry)(c(x)).
54 Chapter III. The Language of Bopf Algebras

The fourth and seventh equalities follow from (3.3), the sixth one from
(1.21).
We also derive

E(S(X)) = E( S(L: E(x')x ll )) = E(L: E(X')S(x")) = E(1JE(X)) = E(X)

(x) (x)
from (1.21). This completes the proof of (3.5).
(b) Let us prove that (ii) implies (i). By uniqueness of the inverse, it is
enough to show that S2 is a right inverse of S for the convolution, just as
is id H . Now, using (3.4) and Condition (ii), we get for all x E H

L: S(x')S2(X") = S(L: S(x")x')

(x) (x)
S(E(x)l) = E(x)S(l) = E(X)1.
This implies that S * S2 = 1JE, hence S2 = id H . Let us prove the converse
implication: if S2 = id H we have

L: S(x")x' S2(L: S(xll)x')

(x) (x)

S(L: S(X')S2(x ll ))
(x)

S(L: S(x')x ll )
(x)
S(E(x)l)
E(X) 1.
One proves that (i) is equivalent to (iii) in a similar fashion.
(c) Recall Relations (3.3): we have

L: X'S(X") = 1JE(x) = L: S(x')x"

(x) (x)
for all x E H. When H is commutative, the first equality becomes

L: S(x")x' = 1JE(x),
(x)

which implies S2 = id H by Part (b) (ii). When H is cocommutative, the

second equality becomes

1JE(x) = L: S(x")x'
(x)

which again implies S2 = id H in view of Part (b) (iii). D

As an immediate consequence of Theorem 3.4, we have the following.

 I11.3 Hopf Algebras 55

Corollary 111.3.5. Let H = (H, jL, 'fI,~, c:, 5) be a Hopf algebra. Then
HOPCOP = (H , ,...., ,." ~op " c: 5)
/lOP 'Yl

is another Hopf algebra and 5 : H ----t HOpcop is a morphism of Hopf

algebras. If, moreover, 5 is an isomorphism with inverse 5- 1 , then

are isomorphic Hopf algebras, the isomorphism being given by 5.

An endomorphism T of a bialgebra H such that

L T(x")x' = c:(x)l = L x"T(x') (3.7)

(x) (x)

for all x E H is sometimes called a skew-antipode for H. Alternatively, a

skew-antipode for H is an antipode for the bialgebras HOP and HCoP. By
Corollary 3.5 the inverse (if it exists) of an antipode is a skew-antipode.
It is not always easy to check the defining Relations (3.3) of an antipode
for every element of a bialgebra, but it may be simpler to check only for
some generators. It is convenient to have the following lemma.
Lemma 111.3.6. Let H be a bialgebra and 5 : H ----t HOP be an algebra
morphism. Assume that H is generated as an algebra by a subset X such
that
L x'5(x") = c:(x)l = L 5(x')x"
(x) (x)

for all x EX. Then 5 is an antipode for H.

PROOF. It is enough to check that if (3.3) holds for x and y, then it holds
for the product xy. Now, by (3.3-3.4)

L (xy)'5((xy)") L x'y' 5(XIY")

(xy) (x)(y)

L x'(Ly'5(yll))5(x")
(x) (y)

(L x'5(x"))c:(y)
(x)
c:(x)c:(y)
c:(xy).

One proves L(xy) 5((xy)')(xy)" = c:(xy) similarly. D

Use the previous lemma to show that the following provide examples of
Hopf algebras.
56 Chapter III. The Language of Bopf Algebras

Example 3. The tensor bialgebra H = T(V) is a Hopf algebra with an

antipode determined by S (1) = 1 and for all VI' V 2 ' ... 'V n E V by

Example 4. (The symmetric bialgebra S(V)) Let I be the kernel of the

projection of T(V) onto the symmetric algebra S(V). Let us show that I
is a coideal for the coalgebra structure put on T(V) in Theorem 2.4. Any
element of I is a sum of elements of the form xlv, w]y where x, y E T(V)
and v, wE V. By Theorem 2.4 we have

6.(x[V, w]y) = L (x'[v, w]y' 121 xl/yl/ + x'y' 121 xl/[v, w]yl/)
(x)(y)

which belongs to 1121 T(V) + T(V) 121 I and

c(x[v, w]y) = c(x)[c(v), c(w)]c(y) = 0,

which proves that I is a coideal. It follows that the bialgebra structure of

T(V) induces a bialgebra structure on S(V) for which the elements of V are
primitive. One checks that S(V) has an antipode which is the multiplication
by (_l)n on sn(V).
Another useful concept is the concept of a grouplike element of a coalge-
bra (H, 6., c), i.e., an element x -I- 0 such that

6.(x) =xx. (3.8)

The set of grouplike elements of H will be denoted by Q(H).

Proposition 111.3.7. Let H be a bialgebra. Then Q(H) is a monoid for

the multiplication of H with unit 1. If, furthermore, H has an invertible
antipode S, then any grouplike element x has an inverse in Q(H) which is
S(x). Consequently, Q(H) is a group.
PROOF. The first assertion is clear. As for the second, observe that (3.6)
and (3.8) imply 6.(S(x)) = S(x) 121 S(x). It follows that S(x) belongs to
Q(H). To complete the proof, one checks that c(x) = 1 when x is grouplike,
and one uses the computation in Example 2 in order to show that S (x) is
the inverse of x. D

Example 5. If k[G] is the Hopf algebra associated to a group G as in

Example 2, then the elements of G are the only grouplike elements of k[G].
In other words, we have
Q(k[G]) = G. (3.9)
 IlL5 Modules over a Hopf Algebra 57

III.4 Relationship with Chapter 1. The Hopf

Algebras GL(2) and 8L(2)
The aim of this section is to show that the algebras M(2), GL(2) and
SL(2) defined in 1.4 and 1.5 are bialgebras. We use Proposition II.4.2 in
order to identify M(2)2 with M(2) M(2), GL(2)2 with GL(2) GL(2)
and SL(2)2 with SL(2) SL(2). Let us show that the morphisms ~ of 1.4
and c: of 1.5 equip these algebras with a cocommutative bialgebra structure.
Recall (1.4.4): we have

(~~~j ~~~) = (~ !) (~ !) (4.1)

and ~(t) = t t. In order to prove that ~ is coassociative, it suffices to

check this on the generators a, b, e, d, and t, which results from the fact
that t is grouplike and from the matrix equality

Similarly, the counit axiom follows from c:(t) = 1 and from the matrix
equalities

(~ !) (~ ~) = (~ !) = (~ ~) (~ !). (4.2)

The algebra morphism S defined in (1.5.2) is an antipode for the bial-

gebras GL(2) and SL(2) which become Hopf algebras in this way. Indeed,
by Lemma 3.6, it is enough to check Relations (3.3) for the generators
a, b, e, d, t. For a, b, e, d it follows from

( ae db)( S(a)
S(e)
S(b)
S(d)
)=( S(a)
S(e)
S(b))( a b
S(d) c d
)=( c:(a)
c:(e)
C:(b))
c:(d) .
(4.3)
As for t, we have tS(t) = S(t)t = c:(t) = 1 since S(t) = C 1 = ad - be.
The antipode is an involution due to the fact that GL(2) and SL(2) are
both commutative. This can also be checked directly on Formula (1.5.2)
defining S.

IlL 5 Modules over a Hopf Algebra

Let A be an algebra. The tensor product U V of two A-modules is an
A A-module by
(aa')(uv) = aua'v (5.1)
58 Chapter III. The Language of Hopf Algebras

where a, a' E A, u E U and v E V. Now, if A possesses a bialgebra structure

(A, 1-", 'T],.6., c), then the algebra morphism .6. : A -+ A A enables us to
equip the A A-module U V with an A-module structure by

a(u v) = .6.(a)(u v) = L a'u a"v. (5.2)

(a)
The counit c equips any vector space V with a trivial A-module structure
by
av = c(a)v (5.3)
where a E A and v E V.
The following is the natural extension of Proposition II. 1.3 to the frame-
work of A-modules.
Proposition 111.5.1. If A is a bialgebra, U, V and Ware A-modules and
k is given the trivial A-module structure, then the canonical isomorphisms
of Proposition 11.1.3
(U V) W ~ U (V W) and k V ~ V ~ V k
are A-module isomorphisms. If, furthermore, A is cocommutative, then the
flip TV,W : V W ~ W V is an isomorphism of A-modules.
PROOF. The proof is easy and is left to the reader. o
Let us show how an antipode allows us to give a natural A-module struc-
ture to the vector space Hom(V, V') of linear maps from V to V' when V
and V' have A-module structures. We first observe that

((a al)f) (v) = af(a'v) (5.4)

puts an A AOP-module structure on Hom(V, V'). Indeed, we have

( (ab b' a' ) f) (v)

abf(b'a'v)
a((b bl)f) (a'v)

( (a a' ) ((b b') f) ) (v)

for a, a', b, b' E A, v E V and f E Hom(V, V'). Now, if A is a Hopf algebra

with antipode S, then the map (id S) 0.6. is a morphism of algebras from
A to AAoP. Pulling (5.4) back along this morphism, we get an A-module
structure on Hom(V, V'). Explicitly, if a E A, v E V and f E Hom(V, V'),
the action of A on Hom(V, V') is given by

(af)(v) = L a' f(S(a")v). (5.5)

(a)
 111.5 Modules over a Ropf Algebra 59

In particular, if V' = k is given the trivial A-module structure, then (5.5)

induces an A-module structure on the dual vector space V* which becomes

(af)(v) = f(S(a)v). (5.6)

Indeed, by (5.5) and (1.21), we get

(af)(v) = L c(a')f(S(a")v) = f( S(L c(a')a")v) = f(S(a)v).

(a) (a)

Proposition 111.5.2. Let (A, /-L, T/, A, c, S) be a Hopf algebra and U, U', V
and V' be A-modules such that, either U or U', and, either V or V', are
finite-dimensional vector spaces. Then the linear map

A : Hom(U, U') Hom(V, V') ---. Hom(V U, U' V')

of (II.2.2) is A-linear if, in addition, the flip TU* v' : U* V' ---. V' U*
is A-linear. In particular, the maps '

A : U* V* ---. (V U)* and AU,v : V U* ---. Hom(U, V)

are A-linear.

PROOF. (a) Let f: U ---. U', 9 : V ---. V', u E U, v E V and a E A. Let us

first compute A(a(f g)) using (II.2.2), (5.2) and (5.5). We have

Zl (A(a(fg)))(vu)

L A(a' f a" g) (v u)
(a)
L (a' f) (u) (a" 9) (v )
(a)
= L (a')' f(S((a')")u) (a")'g(S((a")")v)
(a)

La' f(S(a")u) a"'g(S(a"")v)

(a)

using Sweedler's sigma notation. On the other hand, aA(f g) is given by

Z2 (aA(fg))(vu)

L a'A(f g)(S(a")(v u))

(a)
La' A(f g) (S(a")'v S(a")"u)
(a)
60 Chapter III. The Language of Hopf Algebras

La' >..u 0 g) (S((a")")v 0 S((a")')u)

(a)
La' >..u 0 g) (S(a"')v 0 S(a")u)
(a)

L a' (f(S(a")u) 0 g(S(a"')v))

(a)
L (a')' f(S(a")u) 0 (a')"g(S(a"')v)
(a)
La' f(S(a"')u) 0 a" g(S(a"")v).
(a)

We used (3.6) for the fourth equality. Observe that Zl -I=- Z2 in general.
(b) Let V' = k be given the trivial action. Replacing a'" in Zl [resp. a"
in Z2] by c:(a"') [resp. by c:(a")] and using (1.21), we get

Zl = Z2 = L a' f(S(a")u) 0 g(S(a"')v),

(a)

which proves that>.. : Hom(U, U') 0 V* ---. Hom(V 0 U, U') is A-linear. We

get the two special cases of Proposition 5.2 with U' = k and with U = k.
For the general case, we use Lemma II.2.4 which expresses>.. in terms of
the special maps>.. and of the flip TU*,V" 0
As a corollary of Proposition 5.2, we see that the general map >.. of
Theorem II.2.1 is A-linear when A is cocommutative. This happens, for
instance, when A is a group algebra or an enveloping algebra.
As for the evaluation and the coevaluation maps, we have the following
result.
Proposition 111.5.3. Let V be an A-module. Then the evaluation map
ev v : V* 0 V ---. k is A -linear. If, moreover, the vector space V is finite-
dimensional, then the coevaluation map 8v : k ---. V 0 V* of II.3 and the
composition
Hom(V, W) o Hom(U, V)......':-..Hom(U, W)
are A-linear too.
PROOF. (a) Let a E A, v E V and a E V*. Then

evv(a(a 0 v)) = L evv(a' a 0 a"v)

(a)
L (a'a)(a"v)
(a)
a(L S(a')a"v)
(a)
 HI.6 Comodules 61

= O:(c(a)v)
= c(a)o:(v)
by the rightmost relation (3.3) and by (5.6). This implies that the evalua-
tion map is A-linear.
(b) The coevaluation map 0v is A-linear as the composition of the unit
TJ : k -; End(V) and of Av\;. The latter is A-linear by Proposition 5.2. So
is the map TJ : k -; End(V) following
(aTJ(l))(v) (aidv)(v)
2: a'idv(S(a")v)
(a)
2: a'S (a")v
(a)
c(a)v
(TJ(a1))(v)
for all v E V and a E A. Here we used the leftmost relation (3.3).
(c) For the composition map, one uses Lemma 11.2.5. 0

III. 6 Comodules
Algebras act on modules, coalgebras coact on comodules. This section is
devoted to the definition of the latter concept. Let A be an algebra. Recall
that an A-module is a pair (M, JLM) where M is a vector space and JLM :
A0M -; M is a linear map such that the following axioms (Ass) and (Un)
hold.
(Ass): The square

(6.1)

commutes.
(Un): The diagram
'T/0id
--+

(6.2)

commutes.
A morphism of A-modules f : (M, JLM) -; (M', JLM') is a linear map f
from M to M' such that
(6.3)
62 Chapter III. The Language of Hopf Algebras

The definition of a co module over a coalgebra is obtained by reversing

all arrows in the diagrams above.

Definition 111.6.1. Let (C, Ll, E) be a coalgebra.

(a) A C-comodule is a pair (N,LlN) where N is a vector space and
Ll N : N ---7 C N is a linear map, called the coaction of C on N, such that
the following axioms (Coass) and (Coun) are satisfied.
(Coass): The square

(6.4)
L>.0id
~

commutes.
(Co un): The diagram
c0id
kN f---

(6.5)
N

commutes.
(b) Let (N, Ll N) and (N', Ll NI) be C -comodules. A linear map f from N
to N' is a morphism of C-comodules if

(id f) 0 LlN = LlNI 0 f. (6.6)

(c) A subspace N' of a C-comodule (N,LlN) is a subcomodule of N if

LlN(N') c C N'.

Actually, the comodules we have just defined are left comodules. One
similarly defines a right C-comodule N, using a map N C ---7 N subject
to relations parallel to (6.4-6.5). A right C-comodule is the same as a (left)
comodule over the opposite coalgebra ccoP.
The composition of two morphisms of comodules is another morphism
of comodules. Similarly, the inclusion of a subcomodule into a co module is
a morphism of comodules. Let us give a few examples of comodules.
Example 1. Let C be a coalgebra. Then (C, Ll) is a C-comodule.
Example 2. Let C be a coalgebra and C* the dual vector space equipped
with the dual algebra structure of Proposition 1.2. If (N, LlN) is a C-
comodule, then the dual vector space N* has the structure of a right C*-
module given by the composition of the maps

N* C*~(C N)*~N*. (6.7)

 111.6 Comodules 63

Example 3. Let A be a finite-dimensional algebra and A * be the dual vector

space with the coalgebra structure given by Proposition 1.3. If (M, /-1M) is
a right A-module, then the dual vector space M* has a structure of A*-
comodule given by the composition of the maps

M*~(MQ9A)*~A* Q9M*. (6.8)

In order to put a structure of comodule on the tensor product of two

comodules, we need a bialgebra structure as in Section 5.
Example 4. (Tensor product of comodules) Let (H, /-1, 'T/, fl., s) be a bialgebra
and M and N be H-comodules. We define fl.M0N by

fl.M0N = (/-1 Q9 id M0N )(id H Q9 TM,H Q9 idN )(fl. M Q9 fl. N ) (6.9)

The map fl.M0N endows the tensor product M Q9 N with an H-comodule

structure.
Example 5. (Trivial comodule) Let (H, /-1, 'T/, fl., s) be a bialgebra and V be
a vector space. The linear map
V ~ k Q9 V 1)0id v lH Q9 V (6.10)

equips V with an H-comodule structure. Such a comodule is called a trivial

comodule.
Example 6. (Free co module) Let (C,fl.,s) be a coalgebra. The free C-
comodule on a vector space V is the comodule (C Q9 V, fl. Q9 id v ). This is a
generalization of Example 1.
Proposition 5.1 has the following counterpart for comodules. The proof
is left to the reader.
Proposition 111.6.2. If H is a bialgebra, M, N, Pare H-comodules and
k is given the trivial H -comodule structure of Example 5, then the canonical
isomorphisms of Proposition II.1.3

(M Q9 N) Q9 P ~ M Q9 (N Q9 P) and k Q9 M ~ M ~ M Q9 k

are isomorphisms of H -comodules. If, in addition, the bialgebra H is com-

mutative, then the flip TM,N : M Q9 N ~ N Q9 M is an isomorphism of
H -comodules too.

Notation 6.3. It is often convenient to use for comodules the same kind
of notation as was introduced for coalgebras in Section 1. Let (C, fl., s) be
a co algebra and (N, fl. N ) be a C-comodule. By convention we shall write

fl.N(X) =L Xc Q9 xN (6.11)
(xl
64 Chapter III. The Language of Hopf Algebras

for any x E N. Relation (6.4) is equivalent to

L (xc)' Q9 (xc)" Q9 xN = L Xc Q9 (xN)C Q9 (XN)N (6.12)

(x) (x)

for all x E N. Relation (6.5) is equivalent to

L E(Xc) Q9XN = X. (6.13)

(x)

A linear map f :N --t N' is a morphism of C-comodules if

L Xc Q9 f(xN) = L f(x)c Q9 f(X)NI. (6.14)

(x) (x)

IIL7 Comodule-Algebras. Coaction of SL(2) on

the Affine Plane
The aim of this section is to define a coaction of the bialgebra SL(2) on
the affine plane of Chapter 1. Before doing so, we introduce the following
concept.

Definition III. 7.1. Let (H, J-LH' TiH' 6. H , EH) be a bialgebra and (A, J-L A, TiA)
be an algebra. We say A is an H -comodule-algebra if
(a) the vector space A has an H-comodule structure given by a map
6. A : A --t H Q9 A, and
(b) the structure maps J-L A : A Q9 A --t A and TiA : k --t A are morphisms
of H -comodules, the tensor product AQ9A and the ground field k being given
the H -comodule structures described in Section 6.

We note the following useful characterization of comodule-algebra struc-

tures.

Proposition III. 7.2. Let H be a bialgebra and A be an algebra. Then A

is an H -comodule-algebra if and only if
(a) the vector space A has an H-comodule structure given by a map
6. A : A --t H Q9 A, and
(b) the map 6. A : A --t H Q9 A is a morphism of algebras.

PROOF. It is similar to the proof of Theorem 2.1. We first express the fact
that J-L A is a morphism of H-comodules with the commutative square

(7.1)
idMA
)
 III.7 Comoduie-Algebras. Coaction of SL(2) on the Affine Plane 65

where u = (/-LH @ id@ id) 0 (id @ TA.H @ id) 0 (D. A @ D. A ). The fact that 7)A
is a morphism of H-comodules is equivalent to the commutativity of the
square
k
I~ (7.2)
k@k
Now, Diagrams (7.1-7.2) are exactly the same as Diagrams (7.3) below
which express the fact that D.A is a morphism of algebras:

A@A (H @ A) @ (H @ A) k@k
I~A Iv IrlHT)A (7.3)
A H@A H@A

where v = (/-LH @ /-LA) (id @ TA,H @ id). Indeed, we have

(id @ /-LA) u = v (D. A @ D. A ).
D

U sing the conventions of Sections 1 and 6, we can rewrite Condition (b)

of Proposition 7.2 as D. A (l) = 1 @ 1 and

L (ab)H @ (ab)A = L aHbH @aAbA (7.4)

(ab) (a) (b)

for all a, b E A.
We now show that the affine plane k[x, y] defined in 1.3 possesses an
comodule-algebra structure over the bialgebras M(2) and 8L(2).

Theorem III. 7.3. There exists a unique M(2)-comodule-algebra struc-

ture and a unique 8L(2)-comodule-algebra structure on the affine plane
A = k[x, y] such that

D.A ( ~ ) = (~ ~) @ ( ~ ) .

This matrix notation is short for the two relations

D.A(x)=a@x+b@y and D.A(y)=c@x+d@y. (7.5)

PROOF. We use Proposition 7.2. First observe that Formulas (7.5) define a
morphism of algebras D.A : k[x, y] -* M(2)@k[x,y]. The projection of M(2)
onto 8L(2) being an algebra morphism too, so is the resulting composition
k[x, y] -* 8L(2) @ k[x, y].
It remains to be checked that D.A defines a comodule structure, i.e., that
for all Z E k[x, y] we have

([email protected])oD.A(z) = (D.@id)oD.A(z) and (c:@id)oD.A(z) = l@z (7.6)

66 Chapter III. The Language of Ropf Algebras

where ~ and c: are as in 1.4-1.5. As both sides of each equality to be proved

consist solely of algebra morphisms, it suffices to check (7.6) only for z = x
and z = y. The above matrix notation allows us to do this simultaneously.
We have

(~ !)Q9(~ !)Q9(~)
( (~ Q9 id) ~ A) ( ~ )
0

in view of (4.1). On the other hand, using (1.5.2) we get

Lemma III. 7.4. For all i, j ~ 0 we have

PROOF. Since ~A is an algebra morphism, we have

Next, apply the binomial formula. o

Let us denote by k[x, Y]n the subspace of homogeneous polynomials of
total degree n in A = k[x, y]. Lemma 7.4 implies that k[x, Y]n is a subco-
module of the affine plane due to the fact that

Actually, the M(2)-[resp. SL(2)-]comodule k[x, y] is the direct sum of the

comodules k[x, Y]n'
According to Section 6, Example 2, the dual vector space k[x, y]~ of the
comodule k[x, Y]n has a module structure over the algebra SL(2)*, the dual
of the coalgebra SL(2). We shall identify this module in V.7.

III.8 Exercises
1. (Tensor product of coalgebras) Let (0, ~, c) and (0', ~', c:') be coalge-
bras. Show that the linear maps 7T : 0 Q9 0' ---> 0 and 7T' : 0 Q9 0' ---> 0'
defined by 7T(CQ9C') = c:'(c')c and 7T'(CQ9C') = c(c)c' are morphisms
 III.8 Exercises 67

of coalgebras and that the coalgebra 0 0' satisfies the following

universal property: for any cocommutative coalgebra D and any pair
f : D --+ 0 and l' : D --+ 0' of coalgebra morphisms, there exists
a unique morphism of coalgebras f f' : D --+ 0 0' such that
7r 0 (f 1') = f and 7r' 0 (f 1') = 1'.

2. (Divided powers) Consider the vector space 0 = k[t] of polynomials

in one variable. Prove that there exists a unique coalgebra structure
(0, ~, c:) on 0 such that

~(tn) = L tP t q and c:(tn) = 0nO

p+q=n
for all n > O. Show that 0 becomes a bialgebra when given the
product

Find an antipode.
3. (Tensor coalgebra) Let V be a vector space.
(a) Show that the canonical isomorphisms v(n+m) ~ Vn Vm
endow T' (V) = EBn>O vn with a coalgebra structure, called
the tensor coalgebra of V.
(b) Let Pv be the canonical projection of T' (V) onto V. Prove that
for any co algebra 0 and any linear map f : 0 --+ V, there exists
a unique morphism of coalgebras f : 0 --+ T' (V) such that
f=Pvof.
(c) Using the notation of Chapter II, Exercise 7, define the subspace
8'(V) = EBn>O 8~(V) [resp. A'(V) = EBn>O A~(V)] of T'(V)
generated by all symmetric [resp. antisymffietric] tensors. Show
that 8' (V) and A' (V) are subcoalgebras of T' (V).
(d) Let 0 be a cocommutative coalgebra and f be a linear map from
o to V. Prove the existence and the uniqueness of a coalgebra
morphism f : 0 --+ 8' (V) such that f = Pv 0 f.
4. (Graded dual) The graded dual vector space of a graded vector space
V = EBn>O Vn is the graded vector space Vg*r = EBn>O V;. Let
W = EBn;'O Wn be another graded vector space. Show -that there
is a grading on the tensor product V W such that

(V W)n = ED Vi Wj .
i+j=n

Prove that Vg~ W;r ~ (V W);r if Vn is finite-dimensional for each

n 2': O.
68 Chapter III. The Language of Hopf Algebras

5. (Graded coalgebra) Keep the notation of the previous exercise. A coal-

gebra (O,~, c) is graded if there exist subspaces (On)n>O of C such
that 0 = ffin2:o On and ~(On) c ffii+j=n Ci OJ for all n ;::: 0 and
c(On) = {O} for all n > O.
(a) Prove that the graded dual vector space of a graded coalgebra
carries a natural graded algebra structure.
(b) Let A = ffin>O An be a graded algebra whose summands An are
all finite-dimensional. Prove that the graded dual vector space
of A carries a natural graded coalgebra structure.
(c) Check that the coalgebra 0 of Exercise 2 is the graded dual
vector space of the polynomial algebra k[t].
(d) (Shuffle bialgebra) Let V be a finite-dimensional vector space.
Show that the tensor coalgebra T' (V) of Exercise 3 is the graded
dual of the tensor algebra T(V). Deduce that T' (V) has a bial-
gebra structure whose multiplication is given by

(VI ... Vp)(Vp+1 ... Vp+q) = L Va(l) ... Va(p+q)

a
where VI' ... ,Vp+q are elements of V and where a runs over all
(p, q)-shufRes of the symmetric group Sp+q'
(e) Under the same hypotheses as before, show that S' (V) and
A' (V) are subbialgebras of T' (V) whose graded duals are the
bialgebras S(V) and A(V) respectively.

6. (Convolution algebra) Let G be a finite group. Equip the vector space

C( G) of complex-valued functions on G with the convolution product

(ff')(x) = L f(y)f'(y-Ix)
yEG

where x E G and f, f' E O(G). Show that O(G) has a Hopf algebra
structure such that the linear map f 1-+ 2:xEG f(x)x is a Hopf algebra
isomorphism from O(G) to the group Hopf algebra C[G]. Determine
the unit, the comultiplication, the counit and the antipode of O(G).

7. (An example of a non-commutative, non-cocommutative Hopf alge-

bra) Let H be the quotient of the free algebra k{ t, x} by the two-
sided ideal generated by t 2 - 1, x 2 , xt + tx. Prove that H is a four-
dimensional vector space and that

~(t) = t t, ~(x) = 1 x + x t,
c(t) = 1, c:(x) = 0, S(t) = t, S(x) = tx
endow H with a Hopf algebra structure whose antipode is of order 4.
 III.8 Exercises 69

8. (Convolution and composition) Consider a morphism of algebras f :

A -+ A' and a morphism of coalgebras 9 : C' -+ C. Prove that the
map h f--> f 0 hog from Hom( C, A) to Hom( C', A') is a morphism of
algebras for the convolution *.

9. Use the previous exercise to show that a morphism of bialgebras

between two Hopf algebras is necessarily a morphism of Hopf algebras
(Hint: prove 30 f = f 0 3 by applying left and right convolution with
f = f 0 id = id 0 j).
10. Let H = (H, /.1, Ti, 6., E, 3) be a Hopf algebra.

(a) Set 1jJ0 = TiE, and 1jJrL = id;;' (convolution of n morphisms all
equal to the identity of H) if n > 0 and 1jJrL = 3*rL if n < O.
Prove that each map 1jJ" is an endomorphism of algebras [resp.
of coalgebras] when H is commutative [resp. co commutative]
and that, in both cases, we have 1jJrL o1jJm = 1jJrL+m for any pair
(n, m) of integers.
(b) Let H = k [G] be a group. Show that 1jJrL is the coalgebra endo-
morphism given by 1jJrL(g) = grL (g E G).
(c) Let H = 3(V) be a symmetric algebra. Then 1jJrL(x) = ndx for
any x E 3 d (V).
(d) Show that if H = 3L(2), then the algebra endomorphism 1jJrL is
determined by the matrix identity

if n > 0

and by

1jJrL(a) 1jJrL(b) ) _ ( d
(
1jJrL(c) - -c if n < O.
1jJrL(c)

11. Let H be a Hopf algebra, A be a commutative algebra and C be a

cocommutative coalgebra. Prove that the set HOmAlg(H, A) of algebra
morphisms (resp. the set HomCog(C, H) of coalgebra morphisms) is
a group for the convolution, the inverse of a morphism f being given
by f 0 3 [resp. by 30 fl

12. Let A be a commutative algebra.

(a) Let V be a finite-dimensional vector space. Consider the sym-

metric algebra 3(V) with the Hopf algebra structure described
in Section 3, Example 4. Prove that the group HOmAlg(3(V), A)
is isomorphic to the additive group underlying Adim(V).
70 Chapter III. The Language of Hopf Algebras

(b) Show that HOmAlg(k[Z], A) is isomorphic to the group of invert-

ible elements of A where k[Z] is the Hopf algebra of the group
of integers.
(c) Let 0 be the Hopf algebra of Exercise 2. Determine the group
HOmAlg(O,A).
13. Let (O,~, ) be a coalgebra and (0 @ V, ~ @ id v ) be a free comodule
(see Section 6, Example 6). Prove that for any comodule N the map
f f---7 ( @ id v ) 0 f is a linear isomorphism from the space of comodule
maps from N to 0 @ V to the space Hom(N, V).
14. Let 0 be a coalgebra and (N, ~N) be a O-comodule. Prove that ~N
is an injective morphism of comodules from N to the free comodule
(O@N,~@idN)

15. Let {XJiEI be a basis of a O-comodule (N, ~N) Define elements c1

ofthe coalgebra (O,~, ) by ~N(Xi) = ~jEI c1xj for all i E I.

(a) Prove that ~(c1) = ~kEI cf

@e{ and (c1) = 8ij for all i,j E I.
(b) Show that the subspace ON of 0 linearly generated by the
elements (c1)i,jEI is the smallest subspace 0' of 0 such that
~N(N) cO' @ N. Check that ON is a coalgebra.
(c) Assume that N is finite-dimensional. Prove that the element
t N = ~iEI c~ of ON is independent of the basis {xJ iEI
16. Prove the structure theorem for bimodules over a Hopf algebra as
stated in Section 9.

III. 9 Notes
The concept of a Hopf algebra was developed by algebraic topologists ab-
stracting the work of Hopf [Hop41] on manifolds admitting a product (such
as Lie groups). A basic reference is the famous article [MM65] by Milnor
and Moore. Hopf algebras also came up in the representation theory of Lie
groups and algebraic groups (see [Abe80] [DG70] [Hoc81] [Ser93]). For ab-
stract Hopf algebras, we refer to Abe's and Sweedler's monographs [Abe80]
[Swe69].
All examples of bialgebras given in this chapter turn out to be either
commutative or cocommutative, except for the Hopf algebra of Exercise 7
which is due to Sweedler. Not many examples of non-commutative, non-
cocommutative bialgebras were known before the "quantum group" era
(nevertheless, see [Par81]' [Rad76], [Swe69], pages 89-90, [Taf71]' [TW80J).
This has dramatically changed in the 1980's with the appearance of quan-
tum groups. For details on the order of the square of the antipode of a Hopf
algebra, see [Rad76][Taf71][TW80].
 III.9 Notes 71

(Restricted dual) We saw in Section 1 or in Exercise 5 how to put a

co algebra structure on the dual of an algebra A = (A, J-l, 'T]) which is either
finite-dimensional or graded. In the general case one can proceed as follows.
We know that the map A : A* 0 A* ---+ (A 0 A)* of Corollary II.2.2 allows
to identify A* 0 A* with a subspace of (A 0 A)*. Define

AO = {a E A* IJ-l*(a) E A* 0A*}.

If the algebra is finite-dimensional, then A is an isomorphism and A = A * .

One can show that A is the subspace of linear forms whose kernel contains
an ideal of finite co dimension in A. The vector space A enjoys the following
property: the embedding A induces an isomorphism

Consequently, J-l*(a) belongs to AO 0 AO whenever a is in AO. It results

that (Ao,J-liAo ,'T]*) defines a coalgebra structure on AO. If, in addition, A
has a bialgebra [resp. a Hopf algebra] structure, then so has AO. For more
details, read [Abe80] [Swe69] [Tak85].
(Restricted dual of a Hopf algebra and representations) Let H be a
Hopf algebra. Its restricted dual HO also has a Hopf algebra structure.
It has the following alternative definition based on representations. Let
P : H ---+ End(V) be a representation of H on a finite-dimensional vec-
tor space V. Consider the transpose map p* : End(V)* ---+ H*. Its image
1m (P*), called the coefficient space of the representation p, sits in the re-
stricted dual HO. Then the restricted dual may also be defined as the sum
of the coefficient spaces of all finite-dimensional representations. In the case
when all finite-dimensional H-modules are semisimple, HO is the direct sum
of the coalgebras 1m (Pi) where Pi runs over all finite-dimensional simple
H-modules up to isomorphism. (see [Abe80] [Ser93] [Swe69]).
(Bimodules) Let H be a bialgebra. Let M be a vector space equipped
with an H-module and an H-comodule structure given by maps

J-l M : H 0 M ---+ M and fl. M : M ---+ H 0 M.

Give H 0 M the induced module and comodule structures. Then J-l M is a

morphism of comodules if and only if fl.M is a morphism of modules. If
these equivalent conditions are satisfied, we say that M is an H-bimodule .
. Given such a bimodule M, define the subspace

N = {m E MIfl.M(m) = 10m}.
It turns out that N is a subcomodule, but not a submodule of M. Put the
induced comodule structure on the free H-module H 0 M. Then H 0 M
becomes a bimodule. The structure theorem for bimodules can be stated
as follows: if H is a Hopf algebra, then the map x 0 m 1---+ xm from H 0 N
to M is an isomorphism of H-bimodules. For details, see [Abe80] [Swe69].
Chapter IV
The Quantum Plane and Its
Symmetries

In Chapter I we defined the affine plane as the algebra freely generated by

two variables x and y subject to the trivial commutation relation yx = xy.
This corresponds to our classical perception of plane geometry. In this
chapter, we consider a modified commutation relation depending on a pa-
rameter q, namely
yx = q xy.

This new relation defines the quantum plane. In Section 2 we derive a few
identities well-known to combinatorialists and to the experts in the theory
of linear q-difference equations. Next, investigating the self-transformations
of the quantum plane, we build a bialgebra Mq(2) and Hopf algebras GLq(2)
and SLq(2), which are one-parameter deformations of the bialgebras 1v1(2) ,
G L(2), and S L(2) defined in Chapter 1. The bialgebras obtained in this
way are our first examples of quantum groups. They have the peculiarity
of being neither commutative nor cocommutative.

IV.l The Quantum Plane

Let q be an invertible element of the ground field k, and let Iq be the two-
sided ideal of the free algebra k{ x, y} generated by the element yx - qxy.
We define the quantum plane as the quotient-algebra

(1.1 )
 IV.1 The Quantum Plane 73

When q of- 1, the algebra kq[x,y] is non-commutative. If we give the free

algebra k{ x, y} its natural grading, then the ideal Iq is generated by a
homogeneous degree-two element. It follows that the quantum plane has a
grading such that the generators x and y are of degree 1. We denote by
kq[x, Y]n the vector subspace of all degree-n elements of kq[x, y].
Proposition IV. 1. 1. (a) If a is the automorphism of the polynomial ring
k[x] determined by a(x) = qx, then the algebra kq[x, y] is isomorphic to
the Ore extension k[x][y, a, 0]. Thus, kq[x, y] is Noetherian, has no zero
divisors, and the set of monomials {Xiyj} i,j2:0 is a basis of the underlying
vector space.
(b) For any pair (i, j) of nonnegative integers, we have

(1.2)

(c) Given any k-algebra R, there is a natural bijection

HOmAlg(kq[x, y], R) ~ {(X, Y) E R x R I YX = qXY}. (1.3)
A pair (X, Y) of elements of R subject to the relation Y X = qXY will
be called an R-point of the quantum plane.
PROOF. (a) We use the theory of Ore extensions as presented in 1.7-8.
Define an algebra morphism cp : k{x,y} ----; k[x][y,a,O] by cp(x) = x and
cp(y) = y. Since

cp(y)cp(x) - qcp(x)cp(y) = yx - qxy = a(x)y - qxy = 0,

the morphism cp vanishes on the ideal I q , thus defining a morphism of
algebras, still denoted cp, on kq[x, y]. The morphism cp is surjective because
the Ore extension k[x][y, a, 0] is generated by x and y. In order to show
that cp is an isomorphism, we only have to construct a linear map 'IjJ from
k[x][y, a, 0] to kq[x, y] such that 'IjJ 0 cp = id. We define 'IjJ on the basis
{X i yj}i,j2:0 of k[x][y, a, 0] by 'IjJ(xiyj) = xiyj. The rest of the proof of (i)
follows from 1.7 and 1.8.
Part (b) is proved by an easy induction. Part (c) is a consequence of the
universal property (1.2.4) and of the definition (1.1). 0

We give an example of an R-point of the quantum plane.

Example 1. Let A be the algebra of smooth complex functions on C \ {O}
and let q be a complex number different from 0 and from 1. Consider the
linear endomorphisms Tq and 8q in R = End(A) defined by

Tq(f)(X) = f(qx) and 8q(f)(x) = f(~X) - fix).

qx -x
The pair {Tq,8q} is an R-point of kq[x,y]. The "limit" of the operator 8q
when q tends to 1 is the usual derivative d/dx.
74 Chapter IV. The Quantum Plane and Its Symmetries

IV.2 Gauss Polynomials and the q-Binomial

Formula
We fix an invertible element q of the field k. For future applications, we
need to compute the powers of x + y in the quantum plane kq[x, y]. To this
end, we introduce the so-called Gauss polynomials which are polynomials
in one variable q whose values at q = 1 are equal to the classical binomial
coefficients.
Let us start with some notation. For any integer n > 0, set

n-l qn - 1
(
n ) =l+q+"'+q =--. (2.1)
q q- 1

Define the q-factorial of n by (O)!q = 1 and

(n )'q' = (1) qq'"

(2) () = (q - 1)(q2 - 1) ... (qn - 1)
n q (q _ l)n (2.2)

when n > O. The q-factorial of n is a polynomial in q with integral coef-

ficients and with value at q = 1 equal to the usual factorial n!. We define
the Gauss polynomials for 0 ~ k ~ n by

(2.3)
(k)!q (n - k)!q'

Proposition IV.2.1. Let 0 ~ k ~ n.

(a) ( ~ )q is a polynomial in q with integral coefficients and with value

at q = 1 equal to the binomial coefficient ( ~ ).

(b) We have

( ~ )q = ( n ~ k )q' (2.4)

(c) (q-Pascal identity) We also have

PROOF. Relations (2.4-2.5) follow from easy computations. For Part (a),
one proceeds by induction on n using (2.5). D

We return to the quantum plane of Section 1 and prove the q-binomial

formula.
 IV.2 Gauss Polynomials and the q-Binomial Formula 75

Proposition IV.2.2. Let x and y be variables subject to the quantum

plane relation yx = qxy. Then for all n > 0 we have

(x +y)n = L
O:Sk:Sn
PROOF. Because of the universal property of the quantum plane, it suffices
to prove the statement in kq[x, y]. Expanding (x+y)n and using (1.2), we see
that the monomials in the expansion are all scalar multiples of monomials
of the form xkyn-k. We therefore have

(x +y)n = L
O:Sk:Sn

where ( ~ )' is a polynomial with integral coefficients in q. Let us prove

by induction on n that we have

(2.6)

Relation (2.6) clearly holds for n = 1. It thus suffices to check that the

coefficients ( ~ )' satisfy (2.5). Using (1.2), we have

L (~)' xkyn-k
O:Sk:Sn

L ( n-l)'
k x
k+l n-l-k
y
O:Sk:Sn-l

=L (( n-l)'k-l +q
k (n-l )')
k
k n-k
xy.

We get (2.5) in view ofthe linear independence of the monomials {xkyn-kh

o
We now derive a few q-identities from the q-binomial formula. These
identities will not be needed in the sequel. The first one is the q-analogue
of the Chu- Vandermonde formula.
Proposition IV.2.3. For m :::: p :S n, we have

(m; n )q = o];;p q(m-k)(p-k) ( 7 )q ( p ~k )q

76 Chapter IV. The Quantum Plane and Its Symmetries

PROOF. Expand both sides of the identity (x + y)m+n = (x + y)ffi(X + y)n

using Proposition 2.2 and identify the terms corresponding to xpyffi+n- p.
D

We introduce a q-variant of the exponential. Let z be a variable (com-

muting with q). We define the q-exponential as the formal series

(2.7)

Observe that this series is well-defined provided q is not a root of unity,

which we assume until the end of this section.
Proposition IV.2.4. Let x and y be variables such that yx = qxy. Then

PROOF. By application of Proposition 2.2, we have

(L: (:)~ q) (L: (X)~ q ) "" 1 ("" (n)!q k e)

~ (n)! ~ (k)! ()! x y
k~O e~o n~O q k+e=n q q
(x + y)n
L:...:.....,,-.~
(n)!q .
n~O

D
The q-exponential is an invertible formal series, but, in contrast to the
case q = 1, we have eq(z)-l i= eq ( -z). In order to compute the inverse
of eq(z), we consider the algebra of formal series k[[zll and the algebra
End(k[[z]]) of linear endomorphisms of k[[zll. Define two elements Z and
Tq of End(k[[z]]) by (ZJ)(z) = zJ(z) and (TqJ)(Z) = J(qz). An easy com-
putation shows that (Z, Tq) is an End(k[[z]])-point of the quantum plane,
which is to say we have the following lemma.

Lemma IV.2.5. We have TqZ = q ZTq in End(k[[z]]).

If for any scalar a of the field k we apply the endomorphism ((a - Z)Tq)n
to the constant formal series 1, we get

In particular, for a = 0 we have

(_ZTq)n(l) = (_l)n qn(n-l)/2 zn. (2.9)
Proposition IV.2.6. The inverse of eq(z) is given by

eq(z)-l = L: (_l)n qn(n-l)/2 4.

n~O (n).q
 IV.3 The Algebra Mq(2) 77

PROOF. Lemma 2.5 implies (-ZTq)Z = qZ( -ZTq). Using Proposition 2.4,
we get the following identity in End(k[[z]]):
eq(Z(l- Tq)) = eq(Z) 0 eq(-ZTq). (2.10)
Let us apply both sides of (2.10) to the constant formal series 1. On the
one hand, we have eq(Z(l - Tq))(l) = 1 because (1 - Tq)(l) = O. On the
other hand, by (2.9) we get

eq(Z) (e q( -ZTq)(l)) = eq(z) (L (_l)n qn(n-l)/2 (~;'!

n2:0 q
).
o
Here are two more general q-identities.
Proposition IV.2.7. For any scalar a we have

(a-z)(a-qz) ... (a-qn-l Z ) = ~(_l)k (~ )q qk(k-l)/2 an-k z k

and

PROOF. One proceeds as in the proof of Proposition 2.6, but now with the
operator identity (aTq)( -ZTq) = q (-ZTq)(aTq). By Propositions 2.2 and
2.4, we get

and
eq((a - Z)Tq) = eq( -ZTq) 0 eq(a)
in End(k[[z]]). Applying again these identities to the constant formal series
1, we get the desired relations in view of eq( -ZTq)(l) = eq(z)-l, which was
~~d~~. 0

IV.3 The Algebra Mq(2)

From now on, we assume that q2 i- -1. Let us define a q-analogue of the
algebra M(2) of 1.4. In addition to variables x, y subject to the quantum
plane relation yx = qxy, consider four variables a, b, c, d commuting with
x and y. Define Xl, yl, x", and y" using the following matrix relations
78 Chapter IV. The Quantum Plane and Its Symmetries

Theorem IV.3.l. Under the previous hypotheses, there is an equivalence

between
(i) the two relations y' x' = qx' y' and y" x" = qx" y", and
(ii) the six relations
ba = qab, db = qbd, (3.2)

ca = qac, dc = qcd, (3.3)

bc = cb, ad - da = (q-l - q) bc. (3.4)
PROOF. Let us check that (i) implies (ii). By (3.1) we have

(cx + dy)(ax + by) = q (ax + by)(cx + dy).

Identifying the coefficients of x 2 , y2, and of xy, we obtain

ca = qac, db = qbd, cb + qda = qad + q2bc. (3.5)

Dividing the latter by q yields

ad - da = q-1cb - qbc. (3.6)

Using x" and y" in a similar fashion leads to three more relations obtained
from (3.5-3.6) by exchanging band c, namely

ba = qab, dc = qcd, ad - da = q-1bc - qcb. (3.7)

From (3.6-3.7) we derive (q-l + q)(bc - cb) = 0, which is equivalent to

bc = cb since q2 #- -1. We have proved that (i) implies (ii). The converse
implication follows from similar straightforward computations. 0

Definition IV.3.2. The algebra Mq(2) is the quotient of the free algebra
k{ a, b, c, d} by the two-sided ideal Jq generated by the six relations (3.2-3.4)
of Theorem 3.l.

When q = 1, the algebra Mq(2) is dearly isomorphic to the algebra

M(2) of 1.4. Since the ideal Jq is generated by quadratic elements, the
natural grading of the free algebra induces a grading on Mq(2) such that
the generators a, b, c, d are of degree 1.
Given an algebra R, we define an R-point of Mq(2) to be a quadruple
(A, B, C, D) E R4 satisfying the relations

BA = qAB, DB=qBD, (3.8)

CA = qAC, DC = qCD, (3.9)

BC=CB, AD-DA=(q-l_q)BC. (3.10)
By the very definition of Mq(2), the set of R-points of Mq(2) is in bijection
with the set HOmAIg(Mq(2), R) of algebra morphisms from Mq(2) to R. It
 IV.3 The Algebra Mq(2) 79

will sometimes be convenient and more enlightening to write an R-point

(A, B, C, D) of Mq(2) in the matrix form

(3.11)

Theorem 3.1 can be paraphrased using the language of R-points as follows:

a quadruple (~ ~) of elements of an algebra R is an R-point of Mq(2)
if and only if the following pairs (X', Y') and (X", ylI) are R'- points of the
quantum plane, where X', Y', X", yll are matricially defined by

and
A C
B D )( ~ )
and where R' is the tensor product algebra

R' = R kq[X, Y] = R{X, Y}/(YX - qXY).

We now introduce the quantum determinant detq as the following ele-

ment of the algebra Mq(2).

Proposition IV.3.3. The element detq = ad - q-1bc = da - qbc of Mq(2)

is central.

PROOF. It suffices to show that detq commutes with the generators a, b, c,

d. Now, by (3.2-3.4) we have

(ad - q-1bc)a = a(da - qbc), (ad - q-1bc)b = b(ad - q-1bc),

(ad - q-1bc)c = c(ad - q-1bc), (da - qbc)d = d(ad - q-1bc). 0

Given an R-point m = (~ ~) of Mq(2), the element

Detq(m) = AD - q-l BC = DA - qBC

of R is called the quantum determinant of m.

Proposition IV.3.4. Let R be an algebra and

A'
m=(~ ~) and m' = ( C' B' )
D'
be two R-points of Mq(2) such that the elements A, B, C, D commute with
the elements A', B', C', D'.
80 Chapter IV. The Quantum Plane and Its Symmetries

(a) The element m'm defined by the matrix product

,
m m =
(All
Gil
BII)
D" =
(A'
G'
B') ( A
D' G ~)
is an R-point of Mq (2).
(b) We have Detq(m'm) = Detq(m') Detq(m) in R.
(c) The quadruple
(_q~lG -!B)

is an R-point of M q-l(2) and an ROP-point of Mq(2).

PROOF. (a) We use the reformulation of Theorem 3.1 stated a few lines
ahead of Proposition 3.3. Let R' be the tensor product algebra

R' = R@ kq[X, Y] = R{X, Y}/(YX - qXY).

Define X', y', X", yll by

and ( XII)
y" -
(A'B' D'G') ( Y
X )
By definition, the elements X, Y of R' commute with the other variables A,
A', etc. It results from Theorem 3.1 that the pairs (X', Y') and (X", ylI)
are R'-points of the quantum plane. Now, by hypothesis, the elements A',
B', G', D' of R' commute with X' and y' and the elements A, B, G, D
commute with X" and yll. By a second application of Theorem 3.1,

A'
( G' B') ( X' ) = (All BII) ( X )
D' y' Gil D" Y

and
A" Gil ) ( X )
B" D" Y
are R'-points of the quantum plane. It follows that m'm is an R-point of
Mq(2).
(b) This follows from computations we leave to the reader. A more con-
ceptual method is suggested as an exercise at the end of this chapter.
(c) Define A' = D, B' = -qB, G' = _q-1G, and D' = A. Then Relations
(3.8-3.10) imply

A'B' = qB'A', B'D' = qD'B',

A'G' = qG'A', G'D' = qD'G',
G' B' = B' G', D' A' - A'D' = (q -1 - q) B' G',
which means precisely that (A', B', G', D') is an R-point of Mq-l (2) or an
ROP-point of Mq(2). D
 IV.4 Ring-Theoretical Properties of Mq(2) 81

IV.4 Ring-Theoretical Properties of Mq(2)

The aim of this section is to show that the algebra Mq(2), though non-
commutative, retains certain properties of the commutative algebra M(2).
We freely use the notations and the results of 1.7-8.

Theorem IV.4.1. The algebra Mq(2) is Noetherian and has no zero divi-
sors. A basis for the underlying vector space is given by the set of monomials
i . k C
{a l)1 cd} i,j,k,C?O .
We shall prove this theorem by building a tower

of algebras such that each Ai is an Ore extension of Ai-I' As a consequence

of Corollary 1.7.2, we conclude that the set {u(a)iu(b)ju(c)ku(d)C}i,j,k,c?o
is also a basis of Mq (2) for any permutation u of the set {a, b, c, d}. Define
the algebras Al = k[a], A2 = k{a, b}j(ba - qab), and

A3 = k{ a, b, e}j(ba - qab, ea - qae, eb - be).

The algebra Al is trivially an Ore extension of Ao. Let a l be the automor-

phism of Al determined by a l (a) = qa.

Lemma IV.4.2. There is an isomorphism between A2 and the Ore exten-

sion AI[b, aI' 0]. Furthermore, the set {aiblL,j?o is a basis of A 2 .

Observe that the algebra A2 is isomorphic to the quantum plane kq[x, y]

(the isomorphism sends a onto x and b onto y).
PROOF. Let us define 'PI : A2 ----> Adb,al,O] by 'PI(a) = a and 'PI(b) = b.
Since

'PI defines a morphism of algebras. This morphism is surjective since the

algebra Al [b, aI' 0] is generated by a and b. In order to show that it is an
isomorphism, we only have to build a linear map 'l/J I : Al [b, aI' 0] ----> A2
such that 'l/JI 0 'PI = id. We define 'l/J I on the basis {aiblL,j?o of Al [b, aI' 0]
by'l/JI(aibl)=aibl. D

It is easy to check that a 2 (a) = qa and a 2 (b) = b define an automorphism

a 2 of the algebra A 2 . We have the following result whose proof follows the
same lines as the proof of Lemma 4.2.

Lemma IV.4.3. The algebra A3 is isomorphic to the algebra A 2 [e, a 2 , 0];

the set {aiblekL,j,k?O is a basis of A 3 .
82 Chapter IV. The Quantum Plane and Its Symmetries

The last step consists in building A4 out of A 3 . This is the only step
involving a non-zero derivation. First, one checks that

define an algebra automorphism of A 3 . We define another endomorphism

8 of A3 on the basis {a i lJ1c k }i,j,k.::o:O by 8(1yc k ) = 0 and by

(4.1)

if i > O. The proof of the following result is left to the reader.

Lemma IV.4.4. The endomorphism 8 is an a 3 -derivation of A 3 .

We use this result to prove the next one.

Lemma IV.4.5. The algebra A4 = Mq(2) is isomorphic to the Ore exten-

sion A 3 [d,a3, 8], and {aib-ickdfL,j,k,c.::o:O is a basis of A 4 .

PROOF. Set if4(a) = a, if4(b) = b, if4(c) = c, if4(d) = d. This defines a sur-

jective morphism of algebras if4 from A4 onto the Ore extension A3 [d, a:l , 8],
provided we check that (if4(a), if4(b), if4(c), if4(d)) is an A 3 [d, a 3 , 8]-point
of Mq(2). This implies checking the six relations (3.8-3.10). Now the three
relations not involving d already hold in A 3 . As for the three remaining,
namely
db = qbd, dc = qcd, da = ad + (q - q-l)bc,
they hold in A 3 [d, a 3 , 8] by the very definition of a 3 and of 8. To complete
the proof, one constructs a linear map 1/J 4 such 1/J 4 0 if 4 = id as in the proof
of Lemma 4.2. D

Theorem 4.1 is now a consequence of Lemmas 4.2,4.3 and 4.5, of Corol-

lary 1. 7.2, and of Theorem 1.8.3.

IV.5 Bialgebra Structure on Mq(2)

We now endow the algebra Mq(2) with a bialgebra structure. The comul-
tiplication and the counit will be the same as the comultiplication and the
counit put on M(2) in 1.4 (see also III.4).

Theorem IV.5.l. There exist morphisms of algebras

uniquely determined by

L':l(a)=a@a+b@c, L':l(b)=a@b+b@d, (5.1)

 IV.6 The Hopf Algebras GLq(2) and SLq(2) 83

~ (c) = c 129 a + d 129 c, ~(d) = c 129 b + d 129 d, (5.2)

c(a) = c(d) = 1, c(b) = c(c) = O. (5.3)
Equipped with these morphisms, the algebra Mq(2) becomes a bialgebra that
is neither commutative nor cocommutative. Furthermore, we have

~(detq) = detq 129 detq and c(det q) = l. (5.4)

We may rewrite Relations (5.1-5.3) in the abridged matrix form

PROOF. In order to show that ~ is a morphism of algebras, it suffices to

check that (~(a), ~(b), ~(c), ~(d)) is an Mq(2) 129 M q(2)-point of Mq(2).
This follows from Proposition 3.4 (a). A simple computation shows that
(c(a), c(b), c(c), c(d)) is a k-point of Mq(2), thus proving that c also defines
an algebra morphism.
We now have to check the coassociativity and counit axioms. Let us start
with
(~ 129 id)~ = (id 129 ~)~. (5.6)
Since both sides of (5.6) are morphisms of algebras, it is enough to verify
it on the generators a, b, c, d. Using the matrix form, we have

((~ ~)129(~ ~))129(~ ~)

(~ ~)129((~ ~)129(~ ~))
((idl29~)~)(~ ~).
A similar argument shows that the counit axiom follows from the matrix
identity

As for the computation of ~(detq), it results from Proposition 3.4 (b). D

IV.6 The Hopf Algebras GLq(2) and SLq(2)

We proceed by analogy with I.5. Using the quantum determinant detq of
Proposition 3.3, we define the algebras
84 Chapter IV. The Quantum Plane and Its Symmetries

and
SLq(2) = Mq(2)/(det q - 1) = GLq(2)/(t - 1).
Given an algebra R, we define an R-point of GLq(2) [resp. of SL q(2)J as
an R-point m = (A, B, C, D) of Mq(2) whose quantum determinant

is invertible in R [resp. is equal to 1J. Denoting GLq(2) and SLq(2) by

G q , we see that the set of R-points of G q is in bijection with the set
HOmAlg(Gq,R) of algebra morphisms from G q to R.
Theorem IV.6.1. Relations (5.1-5.3) defining the comultiplication ~ and
the counit e of Mq(2) equip the algebras GLq(2) and SLq(2) with Hopf
algebra structures such that the antipode S is given in matrix form by

( Sea) S(b)) _ det -1 ( d -qb ) (6.1)

S(c) Sed) - q _q- 1c a .

PROOF. (a) We first have to show that ~ and e are well-defined on GLq(2)
and on SLq(2). For SLq(2) this results from the following computations:
by (5.4)
~(detq - 1) = (detq - 1) detq + 1 (detq - 1)

and e(detq -1) = o. A similar argument works for GL q(2) provided we set

~(t) = t t and e(t) = 1. (6.2)

The coassociativity and counit axioms hold for GLq(2) and for SLq(2) since
they already hold for Mq(2).
(b) It remains to check that GLq(2) and SLq(2) have antipodes. Set

S'(a) = d, S'(b) = -qb, S'(c) = _q-1 c, S'(d) = a. (6.3)

By Proposition 3.4 (c), the quadruple (S' (a), S' (b), s' (c), S' (d)) is a Mq(2)OP-
point of Mq(2). Consequently, S' defines a morphism of algebras from
Mq(2) to Mq(2)OP. Next, we extend S' to GLq(2) and to SLq(2) by setting
S'(t) = t. This is a well-defined algebra morphism because

S' (t)S' (detq) = (S' (d)S' (a) - q-1 S' (c)S' (b)) S' (t) = (ad - q- 1bc) t = 1.

Since the quantum determinant is invertible and central in G q = GLq(2)

and SLq(2), it is possible to define an algebra morphism S from G q to G~p
by Set) = r 1 and

( Sea) S(b)) = det -1 (S'(a) S'(b)) = det -1 ( d -qb )

S(c) Sed) q S'(c) S'(d) q _q- 1c a .
 IV.7 Coaction on the Quantum Plane 85

Finally, to check that S is an antipode, it suffices to work with the

generators a, b, c, d, according to Lemma III.3.6. Relations (III.3.3) are
equivalent to the matrix identities

~)

o
In contrast to the inversion in a group and to the antipode of GL(2)
and of SL(2), the antipode S of GLq(2) and of SLq(2) is in general not
involutive. Indeed, from (6.1) we derive

s2n(a)
(
s2n(c)

for any positive integer n. Fix such an n and let q be a root of unity of
order exactly n. Then we obtain two examples of Hopf algebras for which
the square of the antipode has order n. For results on the order of S2
previous to the quantum group era, see [Rad76] [Taf71] [TW80].

IV.7 Coaction on the Quantum Plane

We saw in III.7 that the affine plane k[x, y] was a comodule-algebra over
either one of the bialgebras M(2) and SL(2). We now develop a quantum
version of this.

Theorem IV.7.1. There exists a unique M q(2)-comodule-algebra struc-

ture and a unique SL q(2)-comodule-algebra structure on the quantum plane
A = kq[x,y] such that

6 A (x)=a0x+b0y and 6 A (y)=c0x+d0y.

We rewrite these formulas in the matrix form

(7.1)

PROOF. We use Proposition III. 7.2. We first check that (7.1) defines an
algebra morphism 6 A from A to Mq(2) 0 A. It is enough to verify that
86 Chapter IV. The Quantum Plane and Its Symmetries

in Mq(2) A. Now, by (3.2-3.4), we have

(e x + d y)(a x + b y)
qae x 2 + (be + qda) xy + qbd y2
q (ae x 2 + (q-1be + ad) xy + bd y2)
q (a x + b y)(e x + d y)
q~A(X)~A(Y)'

Since the projection of Mq(2) onto SLq(2) is a morphism of algebras, the

resulting map A --7 SLq(2) A is an algebra morphism too.
It remains to check that ~A defines a comodule structure on the quantum
plane. This is done as in the proof of Theorem III.7.3. D

We record the following quantum version of Lemma III.7.4.

Lemma IV.7.2. For i, j 2: 0 we have

PROOF. We first observe that ~A(xiyj) = ~A(x)i~A(y)j since ~A is an

algebra morphism. Next, we have

(by)(ax) =q2(ax)(by) and (dy)(ca) =q2(ca)(dy)

in the algebra Mq(2) A. This allows us to apply Proposition 2.2 to both

Denote by kq[x, Y]n the subspace of degree n elements of A = kq[x, y].

As a consequence of Lemma 7.2, we see that kq [x, Y]n is a subcomodule of
the quantum plane. Actually, the quantum plane is the direct sum of the
comodules kq[x, Y]n' By III.6, Example 2, the dual vector space kq[x, y]~ is
a module over the algebra SLq(2)* dual to the coalgebra SLq(2). We shall
identify this module in VII.5.

IV.8 Hopf *-Algebras

The standing assumption in this section is that the ground field k is the
field of complex numbers. Given a complex number z, we denote its complex
conjugate by z. Recall that an R-linear map u : V --7 Vi between complex
vector spaces is said to be antilinear if U(AV) = -Xv for all A E C and v E V.
 IV.8 Hopf *-Algebras 87

Definition IV.8.l. Let (H, IL, 71, il, c, S) be a complex Hopf algebra. We
say that H is a Hopf *-algebra if there exists an antilinear involution * on
H sat'isfying the two conditions
(i) the map * is an antimorphism of real algebras, i.e., an algebra mor-
phism from H into HOP, as well as a morphism of real coalgebras, and
(ii) we have S(S(x)*)* = x for all x E H.
Two Hopf *-algebra structures *1 and *2 on H are equivalent if there
exists a Hopf algebra automorphism r.p of H such that r.p(X*l ) = r.p(x )*2 for
all x in H.

We wish to show that the Hopf algebras GLq(2) and SLq(2) have natural
Hopf *-algebra structures given by matrix transposition. We shall need the
following equivalent formulation.

Lemma IV.8.2. A Hopf algebra H has a Hopf *-algebra structure if and

only 11 there exists an antilinear automorphism I of H such that
(i) the map I is a morphism of real algebras and an antimorphism of real
coalgebras, and
(ii) we have 12 = (SI)2 = id H .

PROOF. Suppose we have an involution * as in Definition 8.1. Define I by

I(X) = S-I(X*) for all x E H. It is clear that I is an antilinear algebra
automorphism. It is an antimorphism of coalgebras because so are the an-
tipode S and its inverse by Theorem III.3.4 (a). We have SI = *, which
shows that SI is an involution. Finally, I is an involution too, as can been
seen from
I 2 -_ (s-1 *)2 -- (( *S)2)-1 -
- 1'd- 1 - 'd
H - 1 H'

The second equality follows from * being an involution while the third
one follows from Definition 8.1 (ii). Conversely, define * = SI from an
automorphism I as in Lemma 8.2. It is an involution by Lemma 8.2 (ii).
Let us check Condition (ii) of Definition 8.1. We have

We now present the main result of this section. We freely use the notation
of the previous sections. Recall the inverse t of the element detq = ad-q- 1 bc
of GL q(2). In SLq(2) we have t = 1.

Theorem IV.8.3. There exist unique Hopf *-algebra structures on the

Hopf algebras GLq(2) and SLq(2) such that

a* = td, b* = -q te, c* = -q-l t b, d* = ta, t* = t- 1 .

88 Chapter IV. The Quantum Plane and Its Symmetries

PROOF. By Theorem 3.1, the transpose ( ab dc) of the matnx. (ac db)
is an Mq(2)-point of Mq(2). Consequently, there exists a unique antilinear
algebra endomorphism, of Mq (2) defined by the matrix identity

( ,(a) ,(b)) -_ (ab dc) .

,(c) ,(d) (8.1)

Since transposition is involutive, so is f. The map, is an antimorphism

of co algebras in view of the formula (5.5) giving the comultiplication on
Mq(2) and of the fact that matrix transposition reverses products.
We now extend, to GLq(2) by,(t) = t. Since ,(tdet q -1) = tdet q -1, it
defines an antilinear algebra automorphism both on GLq(2) and on 5L q(2).
Let us check that 5, is an involution. It is enough to verify this on the
generators a, b, c, d, and t. For t, this is clear. For the remaining generators,
we have
(5 , ) (a b) = t (
c d
~
-q lb
-qca ) .
Therefore,

(5,) (_q~lb -:c) (5,)(t)

t (~ ~) C 1

(~ ~).
We conclude the proof by recalling that * = 5f. o

IV.9 Exercises
1. (Gauss) Show that

if n is odd
"~ (_l)k ( nk ) = { (1- q)(l- q3)0 ... (1- qn-l)
if n is even.
O:Sk:Sn q

2. (Gauss) Show that

m+k
( n+m+1 )
m+1 q
m
 IV.9 Exercises 89

3. Let F be a finite field of order q.

(a) Show that ( ~ )q is equal to the number of k-dimensional sub-

spaces of a n-dimensional vector space over F.
(b) Prove Relations (2.4-2.5) using the previous assertion.

4. (q-differentiation) Consider the linear endomorphisms Z, Tq, and Oq of

the polynomial algebra k[z] and of the algebra k[[zlJ of formal series,
defined by

(Zf)(z) = zJ(z), (Tq(f))(Z) = J(qz), (0 (f))(z) = J(qz) - J(z).

q (qz-z)

(a) Check that

Oq Tq = q TqO q, [Oq, Z] = Tq, OqZ - q ZOq = 1.

(b) Prove that Tq is an algebra automorphism and that Oq is a Tq-

derivation.
(c) Show that any Tq-derivation 0 of k[z] is of the form 0 = POq for
some polynomial P. If, moreover, OTq = qTqO, then P has to be a
constant.
(d) Assume that q is not a root of unity. Check that

Oq(4) = zn-l,
(n).q (n - l).q
for all n ?: 1. Deduce that the q-exponential eq(z) is, up to a mul-
tiplicative constant, the only formal series solution of the equation
Oq(f) = J.
5. Let Aq[~, 17] be the algebra k{~, 17} /(e, 172, ~17 + q 17~). Set

where a, b, c and d are variables commuting with ~ and 17. Assume

that q2 -I- -1.
(a) Prove that Assertions (i) and (ii) of Theorem 3.1 are equivalent
to the relations
y'x' = qx'y' and (2 = 17,2 = ~17 + q17~ = O.

(b) Check that (a~ + b17)(C~ + d17) = detq ~17. Deduce Part (b) of
Proposition 3.4.
(c) Find a M q(2)-comodule-algebra structure on Aq[~,17]'
90 Chapter IV. The Quantum Plane and Its Symmetries

6. Show that the centre of Mq(2) is the subalgebra generated by detq

when q is not a root of unity.
. . k .. k
7. (Basis oj SLq(2)) Show that the set {a"lJ1c L,j,k20 U{b"dd L,j20,k>O
is a basis of SLq(2).
8. Let q be a root of unity of order d > 1. Prove that yx = qxy implies

9. Let H be a complex Hopf *-algebra whose counit is denoted c. Show

that c(x*) = c(x) for all elements x of H.

IV.I0 Notes
The content of Section 2 on q-identities, as well as Exercises 1-3, is classical.
We borrowed it from [And76]' Chap. 3 and from [Cig79].
The q-exponential is an example of a q-hypergeometric series or basic
hypergeometric series, i.e., of a formal series L:n>O anz n such that each
quotient an+Iia n is a rational function of qn (where q is a complex param-
eter different from 0 and from 1). Basic hypergeometric series first appeared
in a note published by Heine [Hei46] in 1846. Since, q-analogues of most
classical functions and identities have been found. F.H. Jackson [Jac10]
introduced the q-differentiation operator 8q and its inverse which is the q-
integration. Nowadays, q-series appear in combinatorics, in number theory,
in statistical mechanics, and in the theory of Lie algebras. There are many
monographs on this vast subject, e.g., [GR90] [Sla66].
The operator Tq introduced in Section 2 is fundamental in the theory of
linear q-difference equations with polynomial coefficients. Such an equation
is a functional equation of the form
n
L Pi(Z)J(qi z ) = Q(z)
i=O

where Po(z), ... , Pn(z), Q(z) are polynomials and J(z) is a function. Using
the operator Tq , one can rewrite the equation above as (L:~=o PiT~)(J) = Q.
The articles by Adams [Ada29] and by Trjitzinsky [Trj33] are two classical
references on the formalism of the q-difference equations.
Sections 3, 5 and 6 are taken from Manin's book [Man88]. With Section
3 we entered the heart of the subject of Part I of this book. The bialge-
bras Mq(2), GLq(2), and SLq(2) of Sections 5-6 depend on one parameter.
There also exist two-parameter versions such as the algebra M p ,q(2) gen-
erated by four generators a, b, c, d and the six relations
 IV.lO Notes 91

ba = pab, db = qbd,

ea = qae, de = ped,
be = pq-1eb, ad - da = (q-l - p) eb.
It has the same bialgebra structure as M(2). With the additional relation
ad - p-1be = 1, one gets the Hopf algebra SL p,q(2) of [AST91].
In higher dimension n > 2, Faddeev, Reshetikhin, Takhtadjian [RTF89]
defined the bialgebra Mq(n) generated by the generators (T/)l~i,j~n and
the relations

T:nT k
~J
= TkTm
J~'
rkTm _ TmT k
~J J~
= (q-l _ q) TmT k ~J

for i < j and k < m. The comultiplication and the counit are given by
n
b.(T/) =L Tik (8) T1 and (T/) = bij
k=l

The algebra Mq(n) is an iterated Ore extension and, like Mq(2), it possesses
a remarkable grouplike central element that is

detq = L (_q)-f(u) T;(l) ... T;:(n) ,

uESn

where ( a) is the length of a minimal decomposition of the permutation

a in product of transpositions. The quantum determinant detq allows one
to construct GLq(n) and SLq(n) as in the case n = 2 discussed in this
chapter. The bialgebra Mq(n) has two interesting comodule-algebras: the
first one
A~IO = k{ Xl' ... ,Xn } / (XjXi - qxix j for i < j)
generalizes the quantum plane whereas the second one

generalizes the algebra Aq[~, 17] of Exercise 5.

Both algebras A~lo and A~ln are examples of quadratic algebras, i.e., of
quotients of free algebras by ideals generated by degree-two homogeneous
elements. For authors like Manin, quadratic algebras form the starting point
of the theory of quantum groups. Manin assigns to every quadratic alge-
bra a universal Hopf algebra over which the given quadratic algebra is a
comodule-algebra. When applied to the quantum plane, Manin's construc-
tion yields GL q(2). For further reading, see [Man87] [Man88].
We have just mentioned the quantum groups SLq(n). There exist quan-
tum groups for all classical Lie groups and supergroups. For instance,
92 Chapter IV. The Quantum Plane and Its Symmetries

Takeuchi [Tak89] constructed quantum versions of the symplectic and or-

thogonal groups.
Woronowicz exhibited Hopf *-algebra structures on quantum groups in
the framework of C* -algebras. See [Wor87b] [Wor87a] [Wor88].
The reader will find more examples of and more details on quantum
groups in [AST91] [Ma190] [Ma193] [Man89] [PW91] [Res90] [Sud90]
[Tak92c].
Chapter V
The Lie Algebra of 8L(2)

In this chapter we investigate the enveloping Hopf algebra U = U(s[(2)) of

the Lie algebra s[(2) of traceless two-by-two matrices. This Hopf algebra
is in duality with SL(2). We also describe the finite-dimensional repre-
sentations of U. Chapter V prepares for Chapters VI-VII where we shall
construct a q-deformation Uq of U and study its finite-dimensional repre-
sentations. The statements and proofs for Uq will essentially be copied from
those of the present chapter. We start by recalling the classical concepts of
Lie algebras and enveloping algebras. As usual, we denote the ground field
by k.

V.I Lie Algebras

Definition V.1.1. (a) A Lie algebra L is a vector space with a bilinear
map [ , ] : L x L - t L, called the Lie bracket, satisfying the following two
conditions for all x, y, z E L:
(i) (antisymmetry)
[x, y] = -[y, x],
(ii) (Jacobi identity)

[x, [y, zll + [y, [z, xll + [z, [x, y]] = O.

(b) A Lie subalgebra L' of a Lie algebra L is a subspace L' of L such
that for any (x, y) E L' X L' we have [x, y] E L'. An ideal I of a Lie algebra
94 Chapter V. The Lie Algebra of 5L(2)

L is a subspace I of L such that for any element (x, y) E L x I we have

[x,y] E I.
(c) A morphism of Lie algebras f from the Lie algebra L into the Lie
algebra L' is a linear map f : L ---7 L' such that f([x, y]) = [f(x), f(y)] for
all X,y E L.
(d) A Lie algebra is abelian if its Lie bracket is zero.

Let us give a few examples of Lie algebras.

1. If Land L' are Lie algebras, we equip the direct sum L ffi L' with a
Lie bracket given by

[(x, x'), (y, y')] = ([x, y], [x', y'])

for x, y ELand x', y' E L'. The canonical injections of Land L' into L ffi L'
and the canonical projections of L ffi L' onto Land L' are morphisms of
Lie algebras.
2. Given a Lie algebra L, we define the opposite Lie algebra LOP as the
vector space L with Lie bracket [-, - p given by r
[x, yr p = [y, x] = -[x, y].
The linear map op(x) = -x is a Lie algebra isomorphism from L to LOP.
3. Let I be an ideal of a Lie algebra L. There exists a unique Lie al-
gebra structure on the quotient vector space L/ I such that the canonical
projection from L onto L/ I is a morphism of Lie algebras.
4. Let f : L ---7 L' be a morphism of Lie algebras. Its kernel Ker (1) is
an ideal of L, the image f(L) is a sub algebra of L', and the induced map
L /Ker (1) ---7 f (L) is an isomorphism of Lie algebras.
5. Let A be an (associative) algebra. Set [a, b] = ab - ba for a, bE A. It
is easy to show that this bilinear map is antisymmetric and satisfies the
Jacobi identity. We also have [a, bc] = [a, b]c+ bra, c] for all a, b, c E A. This
Lie algebra will be denoted by L(A).
For any vector space V, we denote the Lie algebra L(End(V)) of all
endomorphisms of V by g[(V). When V is of finite dimension n, then g[(V)
is isomorphic to the Lie algebra g[(n) = L(Mn(k)) of n x n-matrices with
entries in the field k. It is clear that the commutator of two matrices with
zero trace is of trace zero. Consequently, the vector space $[( n) of traceless
n by n matrices is a Lie subalgebra of g[(n).

V.2 Enveloping Algebras

To any Lie algebra L we assign an (associative) algebra U(L), called the
enveloping algebra of L, and a morphism of Lie algebras iL : L ---7 L(U(L)).
We define the enveloping algebra as follows. Let I (L) be the two-sided
 V.2 Enveloping Algebras 95

ideal of the tensor algebra T(L) generated by all elements of the form
xy - yx - [x, y] where x, yare elements of L. We define

U(L) = T(L)/I(L).

The above generators of I(L) are not homogeneous for the grading of T(L)
defined in II.5. Therefore there is no grading on the enveloping algebra
compatible with the grading of the tensor algebra. Nevertheless, U(L) is
filtered as a quotient algebra of T(L).
We define a map i L as the composition of the canonical injection of L
into T(L) and of the canonical surjection of the tensor algebra onto the
enveloping algebra. By definition of i L, we have i L([x, y]) = xy - yx, which
shows that i L is a morphism of Lie algebras.
Example 1. If L is an abelian Lie algebra, then U (L) coincides with the
symmetric algebra S(L). In particular, if L is the zero Lie algebra {O}, then
U({O}) = k. We also have U(LOP) = U(L)OP.
We now state the universal property of U (L).

Theorem V.2.1. Let L be a Lie algebra. Given any associative algebra

A and any morphism of Lie algebras f from L into L(A), there exists a
uniq'ue morphism of algebras l(! : U(L) ---> A such that l(! 0 iL = f.
If we denote by HomLie(L, L') the set of morphisms of Lie algebras from
L into L', we can express Theorem 2.1 by a natural bijection

HomLie(L, L(A)) ~ HOmAlg(U(L), A).

PROOF. By definition of the tensor algebra, f extends to a morphism of

algebras f from T(L) to A defined by f(x l x n ) = f(x l ) f(x n ) for
Xl' ... ,x n in L. The existence of l(! follows from f(1(L)) = {O}. In order to
prove this fact, we only have to show that f(xy - yx - [x, y]) vanishes for
any pair (x, y) of elements of L. Now,

f(xy - yx - [x, y]) = f(x)f(y) - f(y)f(x) - f([x, y]),

which is zero since f is a morphism of Lie algebras.

The uniqueness of l(! is due to the fact that L generates the algebra T( L),
hence U(L). D

We derive two corollaries from Theorem 2.1.

Corollary V.2.2. (a) For any morphism of Lie algebras f : L ---> L',
there exists a unique morphism of algebras U(f) : U(L) ---> U(L') such that
U(f) 0 i L = i L' 0 f. We also have U(id L ) = idu(L)'
(b) If l' : L ---> L" is another morphism of Lie algebras, then

U(f' 0 1) = U(f') 0 U(f).

96 Chapter V. The Lie Algebra of 8L(2)

PROOF. (a) Apply Theorem 2.1 to A = U(L') and to the morphism of Lie
algebras iu 0 f.
(b) We have

U(f') 0 U(f) 0 iL = U(f') 0 iu 0 f = iLl! 0 l' 0 f = U(f' 0 f) 0 i L.

One concludes by appealing to the uniqueness of U(f' 0 1) proved in Part

(a). The uniqueness assertion also implies that U(id L ) is the identity of
U(L). 0

Corollary V.2.3. Let Land L' be Lie algebras and L EB L' their direct
sum. Then
U(L EB L') ~ U(L) U(L').
PROOF. We first construct an algebra morphism cp from U(L EB L') to the
algebra U(L) U(L'). For any x ELand x' E L', set

f(x, x') = iL(x) 1 + 1 iu(x').

This formula defines a linear map f from L EB L' into U(L) U(L'). Let us
show that f is a morphi3m of Lie algebras. For x, y ELand x', y' E L' we
have

[J(x,x'),f(y,y')] = (iL(x) 1 + 1 iu(x'))(iL(y) 1 + 1 iu(Y'))

- (iL(Y) 1 + 1 iu(y'))(iL(x) 1 + 1 iu(x'))
[idx),iL(Y)] 1 + 1 [iu(x'),iu(Y')]
iL([x, y]) 1 + 1 iu([x', y'])
f([x, y], [x', y']) = f([(x, x'), (y, y')]).

Applying Theorem 2.1, we get an algebra morphism cp from U(L EB L') to

U(L) U(L').
We now use the universal property of the tensor product of two algebras
in order to build a morphism of algebras ' : U(L) U(L') ----+ U(L EB L').
The compositions of the canonical injections of L and of L' into L EB L' and
ofthe map iLtf)L' are morphisms of Lie algebras. By Theorem 2.1 there exist
morphisms of algebras 'l : U(L) ----+ U(LEBL') and '2 : U(L') ----+ U(LEBL')
such that, for any x ELand x' E L', we have

'l(X) = iLtf)u(x,O) and '2(x') = iLtf)u(O,x').

By Proposition H.4.1, the formula '(a a') = '1(a)'2(a') defines an al-

gebra morphism ' from U(L) U(L') into U(L EB L') provided we show
that '1(a)'2(a') = 'If'2(a')'1(a) for all a E U(L) and a' E U(L'). We prove
the latter by observing that it is enough to check that 'l (a) and '2 (a')
commute when a = x ELand a' = x' E L'. Now,
 V.2 Enveloping Algebras 97

[i L!J)L' (x, 0), i L!J)L' (0, x')]

i L!J)L' ([(x, 0), (0, x')])
i L!J)L' ([x, OJ, [0, x'D
O.
We claim that the morphisms tp and 'lj; are inverse of each other. Let
us consider the composition 'lj; 0 tp. It is an endomorphism of the algebra
U(L ffi L') restricting to the identity on the image of L ffi L'. Indeed, for all
x ELand x' E L'

'lj;( tp(x, x')) = 'lj;(x 1) + 'lj;(1 x') = iL!J)L' ((x, 0) + (0, x')) = i L!J)L' (x, x').
Consequently, 'lj; 0 tp = id. A similar argument shows that tp 0 'lj; = id. D

Corollaries 2.2 and 2.3 allow us to put a Hopf algebra structure on the
enveloping algebra U(L). Indeed, a comultiplication ~ on U(L) is defined
by ~ = tp 0 U (8), where 8 is the diagonal map x 1--+ (x, x) from L into L ffi L
and tp is the isomorphism U(L ffi L) ~ U(L) U(L) that was built in the
proof of Corollary 2.3. The counit is given by c = U(O) where 0 is the zero
morphism from L into the zero Lie algebra {O}. Finally, the antipode is
defined by S = U(op) where op is the isomorphism from L onto LOP of
Example 1. 2.

Proposition V.2.4. The enveloping algebra U (L) is a cocommutative H opf

algebra for the maps ~, c, and S defined above. For Xl' ... 'X n E L, we have
n-l
1 Xl .. Xn +L L Xa(l) ... Xa(p) Xa(p+l) ... Xa(n)
p=l a

+X I Xn 1
where (J runs over all (p, q)-shujjles of the symmetric group Sn' and

S(X I X 2 X n ) = (_1)nXnX2XI

PROOF. The coassociativity axiom (111.1.5) is satisfied as a consequence of

the commutativity of the square

LffiL
lid!J)8
8!J)id
-----> LffiLffiL

the counit axiom (111.1.6) because of the commutativity of the diagram

O!J)id id!J)O
<---- -----> LffiO
98 Chapter V. The Lie Algebra of 8L(2)

and the co commutativity (III.l. 7) thanks to the commutativity of the tri-

angle
L
T
----t

The formula for ~ results from Theorem III.2.4. The definition of Sand
Lemma III.3.6 imply that S is an antipode for U(L). 0
For the sake of completeness, we give two additional important properties
of enveloping algebras.
Theorem V.2.5. Let L be a Lie algebra.
(a) The algebra U(L) is filtered as a quotient of the tensor algebra T(L)
(graded as in II.5) and the corresponding graded algebra is isomorphic to
the symmetric algebra on L:

gr U(L) ~ S(L).

Hence, if {vJiEI is a totally ordered basis of L, {Vi, vinL,~ ... ~inEI,nEN

is a basis of U(L).
(b) When the characteristic of the field k is zero, the symmetrization
map 'rJ : S(L) ---+ U(L) defined by

(2.1)

for vI' ... 'V n E L, is an isomorphism of coalgebras.

Part (a) of the statement is known as the Poincare-Birkhoff- Witt The-
orem. For a proof of Theorem 2.5, we refer to [Bou60] [Dix74] [Hum72]
[Jac79].
We end this section by a few remarks on the representations of Lie al-
gebras. By definition, an L-module is a U(L )-module in the sense of I.l,
which is the same as a morphism of algebras p : U(L) ---+ End(V). In view
of the universal property of U (L) stated in Theorem 2.1, it is equivalent to
a morphism (still denoted p) of Lie algebras p : L ---+ g[(V). For x ELand
v E V, set xv = p(x)(v). We observe that (x, v) I---> xv is a bilinear map
from L x V to V such that

[x, y]v = x(yv) - y(xv) (2.2)

for x, Y ELand v E V. Conversely, any bilinear map from L x V to V such

that Relation (2.2) holds for all x, y ELand v E V, defines an L-module.
The L-module V is trivial in the sense of III.5 if we have xv = 0 for all
x ELand v E V. By definition of the coproduct in the enveloping algebra,
the structure of L-module on the tensor product of two L-modules V and
V'is given by
x( v Q9 v') = xv Q9 v' + v Q9 xv' (2.3)
 V.3 The Lie Algebra s[(2) 99

for x E L, v E V, and v' E V'. According to nI.5, the Lie algebra acts on
Hom(V, V') by
(xf)(v) = xf(v) - f(xv), (2.4)
which can also be expressed as p(x)(f) = [p(x), f] for f E Hom(V, V'). In
particular, if V'is the trivial module k, then L acts on the dual vector
space V* = Hom(V, k) by

(xf)(v) = - f(xv). (2.5)

Finally, L acts on itself by the so-called adjoint representation which is

given for x, x' E L by
XXi = [x, x']. (2.6)

V.3 The Lie Algebra 1[(2)

To simplify matters, we assume for the rest of this chapter that the ground
field k is the field of complex numbers. The Lie algebra 9[(2) = L(M2(k)) of
2 x 2-matrices with complex entries is four-dimensional. The four matrices

X=(~ ~), Y=(~ ~),

H = (~ ~1)' 1= (~ ~)
form a basis of 9((2). Their commutators are easily computed. We get

[X, Y] = H, [H, X] = 2X, [H, Y] = -2Y,

and
[I, X] = [I, Y] = [I, H] = O. (3.1)
The matrices of trace zero in 9[(2) form the subspace s[(2) spanned by
the basis {X, Y, H}. Relations (3.1) show that s[(2) is an ideal of 9[(2) and
that there is an isomorphism of Lie algebras

9[(2) ~ s[(2) EB kI,

which reduces the investigation of the Lie algebra 9[(2) to that of s((2).
The enveloping algebra U = U(s[(2)) of s[(2) is isomorphic to the algebra
generated by the three elements X, Y, H with the three relations

[X, Y] = H, [H, X] = 2X, [H, Y] = -2Y. (3.2)

We prove some relations in U.

100 Chapter V. The Lie Algebra of 8L(2)

Lemma V.3.1. The following relations hold in U for any p, q ~ 0:

XP Hq = (H - 2p)qXp, yP Hq = (H + 2p)qyP,

[X, yP] = pYP-l(H - p + 1) = p(H + p - l)yp-l,

[XP, Y] = pXP-l(H + p - 1) = p(H - p + l)Xp-l.
PROOF. One proves the first two relations by an easy double induction on
p and q using the relations XH = (H - 2)X and YH = (H + 2)Y, which
is another way of expressing the commutation relations (3.2).
We prove the third relation by induction on p. It trivially holds for p = 1.
When p > 1, we have

[X, YP-l]y + yp-l[X, Y]

(p -1)YP-2(H - p + 2)Y + yp-l H
yp-l ((p -l)(H - p) + H)
pYP-l(H - p + 1).

We conclude by letting yp-l jump over H according to the second relation.

As for the last relation, it can be obtained from the third one by applying
the automorphism a of s[(2) defined by

a(X) = Y, a(Y) = X, a(H) = -H. (3.3)

o
Proposition V.3.2. The set {Xiyj HkL,j,kEN is a basis of U(s[(2)).

PROOF. It is a consequence of the Poincare-Birkhoff-Witt Theorem 2.5.

Another proof can be given, using Ore extensions, along the lines of the
proof of Proposition VL1.4. 0

We close this section by a few remarks on the centre of U. Let us consider

the Casimir element defined as the element
H2
C=XY+YX+ T (3.4)

of the enveloping algebra U.

Lemma V.3.3. The Casimir element C belongs to the centre of U.

PROOF. It is enough to show that the Lie brackets of C with H, X, Y
vanish. Now,
1
[H,C] [H, X]Y + X[H, Y] + [H, Y]X + Y[H, X] + "2 [H, H2]
2XY - 2XY - 2Y X + 2Y X = O.
 V.4 Representations of 5[(2) 101

We also have
1 1
[X,C] X[X, Y] + [X, Y]X + "2 [X, H]H + "2 H[X, H]
XH+HX-XH-HX=O.

One shows [Y, C] = 0 in a similar fashion. o

Harish-Chandra constructed an isomorphism of algebras from the centre
of U to the polynomial algebra k[t]. This isomorphism sends C to the
generator t (see for instance [Bou60], Chap. 8 or [Dix74], Chap. 7). As a
consequence, the Casimir element generates the centre of the enveloping
algebra. We shall give full details in the quantum case (see VI.4).

V.4 Representations of -5((2)

We now determine all finite-dimensional U-modules. We start with the
concept of a highest weight vector.

Definition V.4.1. Let V be aU-module and A be a scalar. A vector v =t 0

in V is said to be of weight A E k if Hv = AV. If, in addition, we have
X v = 0, then we say that v is a highest weight vector of weight A.

Proposition V.4.2. Any non-zero finite-dimensional U -module V has a

highest weight vector.

PROOF. Since k is algebraically closed and V is finite-dimensional, the

operator H has an eigenvector w =t 0 with eigenvalue a: Hw = aw. If
X w = 0, then w is a highest weight vector and we are done. If not, let us
consider the sequence of vectors xnw. By Lemma 3.1, we have

H(Xnw) = (a + 2n)(X n w).

Consequently, (Xnw)n>O is a sequence of eigenvectors for H with distinct
eigenvalues. As V is finIte-dimensional, H can have but a finite number of
eigenvalues; consequently, there exists an integer n such that Xnw =t 0 and
xn+lw = O. The vector Xnw is a highest weight vector. 0

Lemma V.4.3. Let v be a highest weight vector of weight A. For PEN,

set vp = ~
p.
yPv. Then

PROOF. The third relation is trivial; the first two result from Lemma 3.1.
o
We now state the theorem describing simple finite-dimensional U-modules.
102 Chapter V. The Lie Algebra of 8(2)

Theorem V.4.4. ( a) Let V be a finite-dimensional U -module generated

by a highest weight vector v of weight A. Then
(i) The scalar A is an integer equal to dim(V) - 1.
(ii) Setting vp = lip! YPv, we have vp = 0 for p > A and, in addition,
{v = vo, VI' ... ' VA} is a basis for V.
(iii) The operator H acting on V is diagonalizable with the (A+1) distinct
eigenvalues {A, A - 2, ... , A - 2A = -A}.
(iv) Any other highest weight vector in V is a scalar multiple ofv and is
of weight A.
(v) The module V is simple.
(b) Any simple finite-dimensional U -module is generated by a highest weight
vector. Two finite-dimensional U -modules generated by highest weight vec-
tors of the same weight are isomorphic.
PROOF. (a) Accordir:g to Lemma 4.3, the sequence {v p }P2':o is a sequence of
eigenvectors for H with distinct eigenvalues. Since V is finite-dimensional,
there has to exist an integer n such that vn =f. 0 and v n+ 1 = O. The formulas
of Lemma 4.3 then show that vm = 0 for all m > nand Vm =f. 0 for all
m :::; n. We get n = A since we have 0 = XV n + 1 = (A - n)v n by Lemma
4.3. The family {v = vo, ... ,v A } is free, for it is composed of non-zero
eigenvectors for H with distinct eigenvalues. It also generates V; indeed,
the formulas of Lemma 4.3 show that any element of V, which is generated
by v as a module, is a linear combination of the set {v;} i. It results that
dim(V) = A + 1. We have thus proved (i) and (ii). The assertion (iii) is also
a consequence of Lemma 4.3.
(iv) Let Vi be another highest weight vector. It is an eigenvector for the
action of H; hence, it is a scalar multiple of some vector vi. But, again by
Lemma 4.3, the vector Vi is killed by X if and only i = O.
(v) Let Vi be a non-zero U-submodule of V and let Vi be a highest weight
vector of VI. Then Vi also is a highest weight vector for V. By (iv), Vi is a
non-zero scalar multiple of v. Therefore v is in VI. Since v generates V, we
must have V C Vi, which proves that V is simple.
(b) Let v be a highest weight vector of V; if V is simple, then the submod-
ule generated by v is necessarily equal to V. Consequently, V is generated
by a highest weight vector.
If V and Vi are generated by highest weight vectors v and Vi with the
same weight A, then the linear map sending vi to v; for all i is an isomor-
phism of U-modules. 0

Up to isomorphism, the simple U-modules are classified by the nonnega-

tive integers: given such an integer n, there exists a unique (up to isomor-
phism) simple U-module of dimension n+ 1, generated by a highest weight
vector of weight n. We denote this module by V(n) and the corresponding
morphism of Lie algebras by p(n) : s[(2) --* fj[(n + 1).
For instance, we have V(O) = k and p(O) = 0, which means that the
module V(O) is trivial, as is also the case for all modules of dimension 1.
 VA Representations of ,6[(2) 103

More generally, any trivial U-module is isomorphic to a direct sum of copies

of V(O).
Observe that the morphism p(l) : ,6[(2) ----- g[(2) is the natural embedding
of ,6[(2) into g[(2) and that the module V(2) is isomorphic to the adjoint
representation of ,6[(2) via the map sending the highest weight vector Vo
onto X, VI onto - Hand v 2 onto Y.
As for the higher-dimensional module V(n), the generators X, Y, and
H act by operators represented by the following matrices in the basis
{Vo,VI,,V n }:

0 n 0 0
0 0 n-l 0
p(n)(X) =
0 0 1
0 0 0 0

0 0 0 0
1 0 0 0
p(n)(Y) = 0 2 0 0

0 0 n 0
and
n 0 0 0
0 n-2 0 0
p(n)(H) =
0 0 -n+2 0
0 0 0 -n

Let us determine the action of the Casimir element on the simple module
V(n).
Lemma V.4.5. Any central element of U acts by a scalar on the sim-
ple module V(n). In particular, the Casimir element C acts on V(n) by
multiplication by the scalar n(n2+2) , which is non-zero when n > O.
PROOF. Let Z be a central element in U. It commutes with H which decom-
poses V (n) into a direct sum of one-dimensional eigenspaces. Consequently,
the operator Z is diagonal with the same eigenvectors {v = v o , ... , v n } as
H. In particular, there exist scalars aD, ... , an such that ZVp = apvp for
all p. Now

a p+1 Yvp = a p+1 (p + l)Vp+1 = (p + l)ZVp+I = ZYvp = Y ZVp = apYvp.

Consequently, all scalars a p are equal, which shows that Z acts as a scalar.
104 Chapter V. The Lie Algebra of 5L(2)

In order to determine the action of the Casimir element on V (n), we

have only to compute Cv for the highest weight vector v. By (3.4) and by
Lemma 4.3 we get

o
We finally show that any finite-dimensional U-module is a direct sum of
simple U-modules.

Theorem V.4.B. Any finite-dimensional U-module is semisimple.

PROOF. By Proposition 1.1.3, it suffices to show that for any finite-di-

mensional U-module V and any submodule Vi of V, there exists another
submodule V" such that V is isomorphic to the direct sum Vi E8 V". Set
L = s[(2).
1. We shall first prove the existence of such a submodule V" in the
case when Vi is of co dimension 1 in V. We proceed by induction on the
dimension of VI.
If dim(V/) = 0, we may take V" = V. If dim(V/) = 1, then necessarily
Vi and VjV I are trivial one-dimensional representations. Therefore there
exists a basis {v 1 E Vi, v 2 } of V such that LVl = 0 and LV2 C Vi = kv 1 .
Consequently, we have [L, L]vi = 0 for i = 1,2. Formulas (3.2) show that
the action of L on V is trivial. We thus may take for V" any supplementary
subspace of Vi in V.
We now assume that dim(V/) = p > 1 and that the assertion to be proved
holds in all dimensions < p. We have the following alternative: either Vi is
simple, or it is not.
La. Let us first suppose that Vi is not simple; then there exists a sub-
module V1 of Vi such that 0 < dim(Vl) < dim(V/) = p. Let 1f be the
canonical projection of V onto V = V jV1 . The module VI = 1f(V/) is a
submodule of V of codimension one and its dimension is < p. This allows
us to apply the induction hypothesis and to find a submodule V" of V such
that V ~ VI E8 V". Lifting this isomorphism to V, we get

Now, since dim(V") = 1, the vector space V1 is a submodule of codimension

one of 1f- 1 (V"). We again apply the induction hypothesis in order to find a
submodule V" of 1f- 1 (V") such that 1f- 1 (V") ~ V1 E8 V". Let us prove that
the one-dimensional submodule V" has the expected properties, namely
V ~ Vi E8 V". Indeed, the above argument implies that V = Vi + V1 + V";
now V1 is contained in VI, which shows that V is the sum of Vi and of
V". The formula dim(V) = dim(VI) + dim(V") implies that this is a direct
sum.
 V.5 The Clebsch-Gordan Formula 105

l.b. If the submodule V' is simple of dimension> 1, then Lemma 4.5 im-
plies that the Casimir element C acts on V' as a scalar a i- O. Consequently,
the operator Cia is the identity on V'. Now V IV' is one-dimensional, hence
a trivial module. Therefore C sends V into the submodule V', which means
that the map Cia is a projector of V onto V'. As Cia commutes with any
element of U, the map Cia is a morphism of U-modules. By Proposition
1.1.3, the submodule V" = Ker (Cia) is a supplementary submodule to V'.
2. General case. We are now given two finite-dimensional modules V' c V
without any restriction on the codimension. We shall reduce the situation
to the codimension-one case by considering vector spaces W' C W defined
as follows: W [resp. W'] is the subspace of all linear maps from V to V'
whose restriction to V' is a homothety [resp. is zero]. It is clear that W' is
of co dimension one in W. In order to reduce to Part 1, we have to equip
Wand W' with U-module structures. We give Hom(V, V') the U-module
structure defined by Relation (2.4). Let us check that Wand W' are U-
submodules. For fEW, let a be the scalar such that f(v) = av for all
v E V'; then for any x E L, we have

(xf)(v) = xf(v) - f(xv) = x(av) - a(xv) = o.

A similar argument proves that W' is a submodule. Applying Part 1, we
get a one-dimensional submodule W" such that W ~ W' EB W". Let f be
a generator of W". By definition, it acts on V' as a scalar a i- O. It follows
that f I a is a projector of V onto V'. To conclude, it suffices to check that f
(hence f I a) is a morphism of modules. Now, since W" is a one-dimensional
submodule, it is trivial. Therefore, we have xf = 0 for all x E L, which by
(2.4) translates into xf(v) - f(xv) = 0 for all v E V. 0

V.5 The Clebsch-Gordan Formula

Given two finite-dimensional U-modules, we consider their tensor product
equipped with the module structure given by Relation (2.3). By Theorem
4.6 it can be decomposed in simple modules. By the distributivity of the
tensor product with respect to direct sums and by Theorems 4.4 and 4.6,
it is enough to decompose V(n) 0 V(m) into simple modules. This is the
object of the next assertion known as the Clebsch-Gordan formula.

Proposition V.5.l. Consider two nonnegative integers n ~ m. Then

there exists an isomorphism of U -modules

V(n)0V(m) ~ V(n+m)EBV(n+m-2)EB EBV(n-m+2)EBV(n-m).

PROOF. It is enough to prove that, for all p with 0 :S: p :S: m, the module
V (n) 0 V (m) contains a highest weight vector of weight n+m - 2p. In effect,
if so, there exists a non-zero morphism of modules from V (n + m - 2p) into
106 Chapter V. The Lie Algebra of SL(2)

V(n) V(m). The module V(n+m- 2p) being simple, the kernel of such a
morphism has to be zero, which means that the morphism is an embedding
ofV(n+m-2p) into V(n)V(m). Thesubmodules V(n+m-2p) being
simple and of distinct highest weights, their sum in V(n) V(m) is direct.
Thus, the right-hand side of the Clebsch-Gordan formula embeds into the
left-hand side. To conclude, it suffices to check that both sides have the same
dimension. Now the dimension of V(n+m) EEl V(n+m - 2) EEl EEl V(n - m)
equals
m
L (n +m - 2p + 1) (n+ l)(m+ 1)
p=o
dim(V(n)) dim(V(m))
dim(V(n) V(m)).

Proposition 5.1 will then be a consequence of the following lemma. 0

Lemma V.5.2. Let v be a highest weight vector ofV(n) and Vi be a highest

weight vector of V(m). Define vp = ~ YPv and v~ = ~ YPv l for p 2: O.
Then
~(_ )i (m-p+i)!(n-i)! I
~ 1 ( _ )" vi Vp~i
i=O m p .n.
is a highest weight vector of V(n) V(m) of weight n +m - 2p.

PROOF. Set
P
a
t
= (_l)i (m-p+i)!(n-i)!
(m - p)!n! and w = L aivi V~~i'
i=O

It is enough to check that Xw = 0 and Hw = (n + m - 2p)w. The latter

holds because the tensors Vi V~~i all are of weight n + m - 2p. Indeed,
by Lemma 4.3, we have

Let us compute X w. By Lemma 4.3 again, we have

P P
Xw = L a i X(v i ) V~~i +L a i Vi X(V~~i)
i=O i=O
P P
L ai(n - i + 1) Vi~l V~~i + L ai(m - P + i + 1) vi V~~i~l
i=O i=O
p

L (ai(n - i + 1) + ai~l (m - p + i)) Vi~l v~~i'

i=l
V.6 Module-Algebra over a Bialgebra. Action of .5[(2) on the Affine Plane 107

Now,

(ki(n - i + 1) + (ki_l (m - p + i)
(_l)i (m - p + i)!(n - i)! (n _ i + 1)
(m - p)!n!
i-l (m - p + i - l)!(n - i + I)! ( .)
+ (- l) (m _ p )'.n., m - p +z
o.
D

Remark 5.3. (a) One deduces from Proposition 5.1 that the adjoint repre-
sentation V(2) is related to V(O) and V(l) by

V(1)2 ~ V(2) 63 V(O).

(b) The dual module V(n)* is isomorphic to the simple module V(n)
(prove it). Consequently, we have the U-linear isomorphisms

Hom(V(n), V(m)) ~ V(m) V(n)* ~ V(m) V(n).

V.6 Module-Algebra over a Bialgebra. Action of

.5[(2) on the Affine Plane
We now introduce a concept that formalizes nicely many situations where
an algebra acts on another one.

Definition V.6.!. Let H be a bialgebra and A an algebra. We call A a

module-algebra over H if
(a) the vector space underlying A is an H -module, and
(b) the multiplication J.l : A A ---> A and the unit rJ : k ---> A of A are
morphisms of H -modules, the tensor product A A and the ground field k
being given the H -module structures described by Relations (III. 5. 2-5.3) .

In the literature, module-algebras over a bialgebra H are also called H-

algebras. By making explicit Condition (b) of Definition 6.1, we see that A
is a module-algebra over H if the action of H on A satisfies the following
two compatibility relations with the product and the unit of A:

x(ab) = L (x'a) (x" b) (6.1)

(x)

and
xl = c(x)l (6.2)
108 Chapter V. The Lie Algebra of 8L(2)

where x is an element of H and a, b are elements of A. Here we used

Sweedler's sigma notation (see IIL1.6). The map E: is the counit of the
bialgebra H while 1 is the unit of A.
It is not always convenient to check Relation (6.1) for all elements x of
H. The following result shows that it is enough to check it for a set of
generators.

Lemma V.6.2. Let H be a bialgebra and A be an algebra with a structure

of H -module such that Relation (6.2) holds. Assume that H is generated
as an algebra by a subset X whose elements x satisfy Relation (6.1) for all
elements a and b in A. Then A is a module-algebra over H.

PROOF. It suffices to check that if Relation (6.1) holds for x and y in H,

then it also holds for their product xy. Now, for all a, bE A, we have

(xy)(ab) x(y(ab))
x(L (y1a)(y"b))
(y)

L (x'(YI(a))) (x"(Y"(b)))
(x)(y)

L ((x'YI)a) ((x"y")b)
(x)(y)

L ((xy)la) ((xy)"b).
(xy)

o
The following examples show that module-algebra structures appear in
a number of situations.
Example 1. Let 'P be an automorphism of an algebra A. Consider the
algebra k[Z] of the group of integers with the bialgebra structure described
in IIL2, Example 2. If k[Z] acts on A by sending a generator of Z on 'P,
then A becomes a module-algebra over k[Z].
Let us describe module-algebras over enveloping algebras.

Lemma V.6.3. Let L be a Lie algebra. An algebra A is a module-algebra

over U(L) if and only if A has an L-module structure such that the elements
of L act on A as derivations.

PROOF. From Section 2 we know that a U(L)-module is an L-module and

conversely. Assume that A is a module-algebra over U(L). If x E L we have
~(x) = x Q9 1 + 1 Q9 x. For such an x, Relation (6.1) becomes

x(ab) = x(a)b + ax(b)

 V.7 Duality between the Hopf Algebras U(.5[(2)) and SL(2) 109

for all a, b E A, which shows that x acts as a derivation. The converse

statement results from Lemma 6.2. D

We now return to the Lie algebra .5[(2) and show how the affine plane
becomes a module-algebra over the enveloping algebra U (.5[(2)).
Theorem V.6.4. Define an action of the Lie algebra .5[(2) on the polyno-
mial algebra k[x, yl by
oP oP oP oP
X P = x oy , Y P = y ox ' H P = x ox - y oy

where P denotes any polynomial of k[x, yl.

(a) Then k[x, yl becomes a module-algebra over U(.5[(2)).
(b) The subspace k[x, Yln of homogeneous polynomials of degree n is a
submodule of k[x,yl isomorphic to the simple .5[(2)-module V(n).
We have thus succeeded in packing into a single module all simple finite-
dimensional U(.5[(2))-modules, thanks to the notion of module-algebra.
PROOF. (a) We shall first check that the above formulas define an action
of .5[(2) on k[x, Yl. We have

[X,YlP = x~
oy
(y OP)
ox
_y~
ox
(x OP)
oy
oP 02p oP 02p
x- + xy-- - y- - yx--
ox oyox oy oxoy
HP.
One similarly shows that [H,XlP = 2XP and [H, YlP = -2YP.
In order to conclude that we have a module-algebra structure, it is enough
in view of Lemma 6.3 to check that the generators X, Y, H act on k[x, yl
as derivations, which is clearly the case.
(b) Fix a non-negative integer n and set v = xn E k[x, Yln' Clearly, v is
a highest weight vector of weight n. For all p :::: 0 we have

v = ~ yPv = ( n ) xn-pyp
p p! P
if p ::; nand vp = 0 if p > n. Since the monomials {v p }p generate k[x, yln'
the latter is a .5[(2)-module generated by a highest weight vector of weight
n. Hence, by Theorem 4.4, it is isomorphic to the simple module V(n). D

V.7 Duality between the Hopf Algebras U(s((2))

and SL(2)
The main objective of this section is to relate this chapter to Chapter I by
building a duality between U = U(.5!(2)) and the Hopf algebra SL(2) de-
fined in 1.5. We start with the following definition due to Takeuchi [Tak81l
110 Chapter V. The Lie Algebra of S(2)

Definition V.7.1. Given bialgebras (U,p,,'T},Il,) and (H,p,,'T},Il,) and

a bilinear form < , > on U x H, we say that the bilinear form realizes a
duality between U and H, or that the bialgebras U and H are in duality, if
we have
< uv,X >= L < u,x' >< v,x" >, (7.1)
(x)

<u,xy>= L < u', x > < u", Y >, (7.2)

(u)
< 1,x > = (x), (7.3)
and
<u,l>=(u) (7.4)
for all u, v E U and x, y E H.
If, in addition, U and Hare Hopf algebras with antipodes S, then they
are said to be in duality if the underlying bialgebras are in duality and if,
moreover, we have
< S(u),x >=< u,S(x) > (7.5)
for all u E U and x E H.

Let us motivate this definition. Let 'P be the linear map from U to the
dual vector space H* defined by

'P(u)(x) = < u, x> .

Similarly, 'IjJ( x) (u) = < u, x > defines a linear map from H to U*. From
Proposition 111.1.2 we know that the dual spaces U* and H* carry natural
algebra structures. If, in addition, the vector space H is finite-dimensional,
then the dual space H* has a natural bialgebra structure induced by the one
on H (see 111.2, Example 1). We are now ready to state a characterization
for duality between bialgebras.

Proposition V.7.2. Given bialgebras U and H and a bilinear form < , >
on U x H, the bilinear form realizes a duality between U and H if and only
if the linear maps 'P and 'IjJ are morphisms of algebras.
If, moreover, H is finite-dimensional, then the bilinear form realizes a
duality if and only if'P is a morphism of bialyebras.

We shall say that the duality between U and H is perfect when both
maps 'P and 'IjJ are injective. In case U and H are finite-dimensional, a
perfect duality between them induces isomorphisms of bialgebras between
U and H* and between Hand U*.
PROOF. Let us express that 'P is a morphism of algebras. Recall that the
unit of H* is equal to the counit of H and that the product of two linear
 V.7 Duality between the Hopf Algebras U(s[(2)) and 5L(2) 111

forms a and f3 of H* is given by

(af3)(x) = L a(x')f3(x")
(xl

for all x E H. Then the relations 'P(1) = 1 and 'P(uv) = 'P(u)'P(v) imply
< 1, x> = 'P(l)(x) = E(X) and
< UV,X > 'P(UV)(x) = ('P(u)'P(v))(x)
L 'P(u)(x')'P(v)(x") = L < u, x' >< v, x" > .
(x) (x)

It results that Relations (7.1) and (7.3) of Definition 7.1 are equivalent
to the fact that 'P is a morphism of algebras. By symmetry, we see that
Relations (7.2) and (7.4) are equivalent to the fact that 7jJ is a morphism
of algebras.
Now assume that H is finite-dimensional. Then the dual space H* is
a bialgebra. We have already expressed the fact that 'P is a morphism of
algebras. Let us express that it is a morphism of coalgebras. On one hand,
the relation E'P = E expressing that 'P preserves the co unit reads

du) = (E'P)(U) = 'P(u)(l) = < u, 1 > .

On the other hand, if 'P preserves the comultiplication, we have

< u,xy > 'P(u)(xy) = ~('P(u))(x y)

L 'P(u')(x)'P(u")(y)
(u)

L < u',x >< u",y > .

(u)

Thus, the map 'P is a morphism of coalgebras if and only if Relations (7.2)
and (7.4) are satisfied. D

We return to the enveloping algebra U = U(s[(2)). We wish to set it

in duality with the Hopf algebra SL(2). Our first task is to construct a
morphism of algebras 7jJ from the algebra M(2) = k[a, b, c, d] (introduced in
1.4) to the dual algebra U*. We shall deduce a bilinear form on U x M(2)
defined by < u, x > = 7jJ(x)(u) and satisfying Relations (7.2) and (7.4).
Now, building 7jJ is equivalent to giving four pairwise commuting elements
A,B,C, D of U*.
The definitions of A, B, C, and D use the simple U-module V(l) with
the basis {vO'v 1 } described in Section 4. Given an element u in U, we set

A(u) B(U))
p(u) = ( C(u) D(u)
112 Chapter V. The Lie Algebra of 8L(2)

where p is the representation p(l) corresponding to V(I). This defines four

linear forms on U, hence four elements A, B, C, D of the dual space
U*. The comultiplication of U being cocommutative, the dual algebra U*
is commutative. Therefore, the quadruple (A, B, C, D) defines a unique
morphism of algebras 'ljJ : M(2) ----> U* such that

'ljJ(a) = A, 'ljJ(b) = B, 'ljJ(c) = C, 'ljJ(d) = D. (7.6)

Proposition V.7.3. The bilinear form < U,x >= 'ljJ(x)(u) realizes a du-
ality between the bialgebras U and M(2).
PROOF. It remains to check Relations (7.1) and (7.3). We start with (7.3).
The identity p(l) = 1 yields

( < 1, a> < l,b >

< l,c > < I,d> ) = (~g~ ~g~) = (~ ~)
= (c(a) c(b)) (7.7)
c(c) c(d)
by definition of the co unit in M(2). Now, from Relation (7.2) we get

< 1, xy > = < 1, x > < 1, Y > .

Both maps x f---+ < 1, x> and care morphisms of algebras and they coincide
on the generators a, b, c, d of M(2) by (7.7). Therefore, they have to be
equal, which proves Relation (7.3).
We now turn to the proof of Relation (7.1). Let us denote by C(x) the
following condition on an element x of M(2): For any pair (u, v) of elements
of U, we have
~
< uv,x >= ~ < u,x I >< v,x /I >.
(x)
Let us first show that C(I) is satisfied. Indeed, from (7.4) we get

< UV, 1 > = c(uv) = c(u)c(v) = < u, 1 >< v, 1 > .

We next prove that Conditions C(a), C(b), C(c), C(d) hold. By definition,
we have

p(u) =
( A(u)
C(u)
B(u) ) = ( < u, a>
D(u) <u,c>
< u,b >
< u,d > ).
Let us express that p(uv) = p(u)p(v). We have

( << uV,a >

UV,c >
< uV,b > )
< UV, d >

< u,a >

=( < U,c > < u,d >
< u,b >
)( < v,a >
< V,c >
< v,b >
< v,d > ). (7.8)
 V.7 Duality between the Hopf Algebras U(.s[(2)) and SL(2) 113

Expanding this matrix product, we get exactly the four desired conditions
since, as we know from Chapter I, the coproduct on M(2) is defined by the
matrix relation

In order to conclude the proof of (7.1), we need to check Condition C(x)

for an arbitrary element x of 1\11(2). To this end, we first observe that if
C(x) and C(y) are verified, then so is C(AX + y) for any scalar A; second,
we use the following lemma, which completes the proof of the proposition.
D

Lemma V.7.4. If Conditions C(x) and C(y) hold, then so does C(xy).

PROOF. Relation (7.2), and Conditions C(x) and C(y) imply that

< uv,xy > L < (uv)', x >< (uv)", y >

(nv)

L < u'v', x >< u"v", y >

(u)(v)

L < u', x' > < v', x" > < u", y' > < v", y" > .
(n)(v)(x)(y)

They also yield

L < u, (xy)' >< v, (xy)" >

(xy)

L < u, x'y' >< v, x"y" >

(x)(y)

L < u', x' > < u", y' > < v', x" > < v", y" >
(n)(v)(x)(y)
< uV,xy >.
D

The duality between M(2) and U is not perfect: the morphism 7j; is not
injective as the following lemma shows.

Lemma V.7.5. We have 7j;(ad - be) = 1.

Equivalently, < u, ad - be >= c(u) for all elements u of U.

PROOF. Lemma 1.5.2 as rephrased in (II.4.5) means that the element ad-be
is grouplike. Consequently, by (7.1) we have

< UV, ad - be >= < u, ad - be >< v, ad - be >

114 Chapter V. The Lie Algebra of S(2)

for any pair (u, v) of elements of U. On the other hand, by (7.3) we have
< 1, ad - be > = c (ad - be) = 1.
This implies that the linear map u f--+ < u, ad - be > is a morphism of
algebras from U to k. To show that this morphism coincides with the
co unit c, it suffices to check that both maps have the same values on the
generators X, Y and H. Now we have
< X,ad - be >
= c(a) < X, d > + < X, a> c(d) - c(b) < X, e > - < X, b > c(e)
= O=c(X).
Similarly, we get < Y, ad - be >= 0 = c(Y). Finally,
< H,ad - be >
c(a) < H,d > + < H,a > c(d) - c(b) < H,e > - < H,b > c(e)
= -1 + 1 = 0 = c(H).
o
As a consequence of the previous lemma, the morphism of algebras 1/J :
M(2) ---> U* factors through SL(2) = M(2)/(ad - be - 1). We still denote
by 1/J the induced morphism of algebras from SL(2) to U* and by < , >
the corresponding bilinear form.
Theorem V.7.6. The bilinear jorm < u,x >= 1/J(x)(u) realizes a duality
between the Hopj algebras U and SL(2).
PROOF. We already know that 1/J is a morphism of algebras. By Proposition
7.2 we are left with showing that cp : U ---> SL(2)* is a morphism of algebras
too. Now, the projection from M(2) onto SL(2) dualizes to an injective
morphism from SL(2)* into M(2)*. It is clear that, when composing the
latter with cp, we get the morphism of algebras cp : U ---> M(2)* investigated
earlier. Consequently, cp : U ---> SL(2)* is a morphism of algebras. This
shows that we have a duality between bialgebras.
It remains to examine the antipodes and to check Relation (7.5). Let us
start with the generators. In the abridged matrix form we have

< S(X), ( : !) > p(S(X)) = -p(X)

(~ ~1)
d
<X, ( -e ~b) >
S(a) S(b) ) >.
<X, ( S(e) S(d)
One proceeds similarly with Y, H, and 1.
 V.7 Duality between the Hopf Algebras U(.s[(2)) and SL(2) 115

For arbitrary elements of U and SL(2), we use the following result. 0

Lemma V.7.7. Letu,v be elements ofU. If

< S(u),x > = < u,S(x) > and < S(v),x > = < v,S(x) >
for all x E SL(2), then < S(uv), x> = < uv, S(x) >. Similarly, let x, y be
elements of SL(2). If

< S(u),x > = < u,S(x) > and < S(u),y > = < u,S(y) >
for all u E U, then < S(u), xy > = < u, S(xy) >.
PROOF. Theorem III.3.4 (a) and Definition 7.1 imply that

< S(uv), x> < S(v)S(u), x>

L < S(v), x' >< S(u), x" >
(x)

L < U, S(X") >< V, S(x') >

(x)

L < u, S(x)' >< V, S(X)" >

(S(x))

< uv, S(x) > .

The proof of the second assertion is similar. o
To the duality between U and SL(2) corresponds a duality between U-
modules and SL(2)-comodules. We now investigate this. In III.7 we showed
that the vector space k[x, Yln of homogeneous polynomials of total degree n
had a natural SL(2)-comodule structure. By duality, the dual vector space
k[x, Yl~ has a module structure over the algebra SL(2)*, hence over the
algebra U via the morphism ip : U ---+ SL(2)*. The following result gives
the structure of k[x, Yl~ as aU-module.
Theorem V.7.8. The U -module k[x, Yl~ is simple with highest weight n.

In other words, the SL(2)-comodule k[x, Yln corresponds by duality to

the U-module V(n).
PROOF. We shall show that the linear form on k[x, yln defined by

f(xiyn-i) = Oni

is a highest weight vector with weight n ofthe U-module k[x, Yl~, which will
imply that k[x, Yl~ contains a submodule isomorphic to the simple module
V(n). Since
dim(V(n)) = n + 1 = dim(k[x, Yl~),
we get k[x, Yl~ ~ V(n).
116 Chapter V. The Lie Algebra of SL(2)

In order to prove that f is a highest weight vector, we need the identity

(7.9)

for all u E U and for all i such that 0 :s: i :s: n. Indeed, by definition of f,
by III.6, Example 2 and by Lemma III.7.4 we have

(uf)(Xiyn-i)
(u f)(t:.A(xiyn-i))

ta~G)(n~i)
t~
r=Os=O
G) (n~i)
ta~ G) (n~i)
Let us apply Relation (7.9) to H. A straightforward computation using
(7.2-7.3) and the definition of the bilinear form yields

Consequently, we have (Hf)(xiyn-i) = n15ni , which implies that Hf = nf.

It remains to prove that X f = O. This is a consequence of Relation (7.9)
applied to X and of the fact that < X, aid >= 0 for all i and j. Let us
prove the latter. First, we have < X,l > = c(X) = O. Next, if i > 0, we
have by (7.2-7.3)

c(a) < X,a i - 1> + < X,a > c(ai - 1 )

< X, a i - 1 > = ... = < X, a> = O.
Similarly, if j > 0 we get

< x,d >= c(c) < x,d- 1 > + < X,C > c(d- 1 ) = o.
Consequently,

< X,aid > = c(a)i < x,d> + < X,a i > c(d) = o.
o
 V.8 Exercises 11 7

V.8 Exercises
1. Let L be a Lie algebra. Show that [L, L] is an ideal of L and that the
quotient Lie algebra Lab = L/[L, L] is abelian. Prove that if f is a
morphism of Lie algebras from L to any abelian Lie algebra V, then
r
there exists a unique linear map b from Lab into V such that f is
r
the composition of b and of the canonical projection from L onto
Lab.

2. For any Lie algebra L determine the group of grouplike elements of

the Hopf algebra U(L).
3. Let A be an algebra and Der(A) the vector space of all derivations
of A. Show that the commutator of any two derivations is again a
derivation and that Der(A) is a Lie subalgebra of g((A).
4. Show that any algebra A is a module-algebra over the enveloping
algebra of the Lie algebra Der(A) and over the bialgebra k[G] where
G is the group of algebra automorphisms of A.
5. Let L be a Lie algebra and p : L ----> g((V) a finite-dimensional repre-
sentation of L. Define a symmetric bilinear form on L by
< X,y >p = tr(p(x)p(y))
where tr denotes the trace of endomorphisms.

(a) Prove that this form is invariant, i.e., we have

< [x, y], z >p=< x, [y,z] >p

for all x,y,z E L.
(b) Let {XJl<i<d be a basis of L. Assume the form < , >p non-
degenerate. Define a new basis {xih::;i::;d of L by requiring that
< Xi' x j > P = 0ij' We get an element Op = Ll::;i::;d XiX i of
U(L). Show that Op belongs to the centre of the enveloping
algebra and that tr(p(Op)) = d = dim(L).
(c) (Whitehead Lemma) Let f : L ----> V be a linear map satisfying
the relation
f([x, y]) = xf(y) - yf(x)
for all X, y E L. Assume that the form < , > p is non-degenerate
and that Op is well-defined. Show that we have

Opf(x) = x( L xd(Xi)).
l::;i::;d
Deduce that, when p(Op) is invertible, there exists a vector v in
V such that f(x) = xv for all X in L.
118 Chapter V. The Lie Algebra of 8L(2)

6. Find all invariant symmetric bilinear forms of 5[(2) (as defined in the
previous exercise; assume that the field k is of characteristic zero).

7. Show that the enveloping algebra U(5[(2)) is Noetherian and has no

divisors of zero. Find its centre (Hint: proceed by analogy with VI.4).

8. Assume that k is a field of characteristic zero. Show that the Lie

algebra 5[(2) has no ideals but {O} and the algebra itself. Deduce
that 5[(2) = [5[(2),5[(2)].

9. Show that the dual of the U-module V(n) is isomorphic to V(n).

10. Determine all Hopf algebra automorphisms of U(5[(2)).

11. Check that there is an antiautomorphism T of algebras of U(5[(2))

such that T(X) = Y, T(Y) = X, and T(H) = H. Prove that T is a
morphism of coalgebras. Find all non-degenerate symmetric bilinear
forms ( , ) on the simple module V(n) such that (xv, v') = (v, T(x)v / )
for all x E U(5[(2)) and v, v' E V(n). Show that the basis of V(n)
consisting of the vectors v o ,"" vn (defined in Section 4) is orthogonal
for such a form.

12. (Bialgebra structure on the quantum plane) (a) Show that the formu-
las

.6(x)=x0x, .6(y)=x0y+y01, c(x)=I, c(y)=O

equip the free algebra k{x,y} and the quantum plane kq[x,y] with a
bialgebra structure.
(b) Prove that an algebra R is a module-algebra over the bialgebra
k{x,y} [resp. over kq[x,y]] if and only if R possesses an algebra
endomorphism 7 and a 7-derivation 0 [resp. 7 and 0 such that the
relation 07 = q70 holds].
(c) Find all kq[x, y]-algebra structures on the polynomial algebra k[z]
(consider only the ones for which 7 is an automorphism). In par-
ticular, show that, when 7 is the algebra automorphism 7 q of k[z]
considered in IV.2, then 0 is necessarily a scalar multiple of Oq (see
Exercise 4 in Chapter IV).

13. Show that any antilinear involution * on a complex Lie algebra L

such that [x, y]* = [y*, x*] for all x, y E L induces a Hopf *-algebra
structure on U(L).
14. Prove that there exists a unique Hopf *-algebra structure on U(5[(2))
such that X* = Y, Y* = X, and H* = -H.

15. Find all Hopf *-algebra structures on U(5[(2)) up to equivalence,

assuming that the ground field is the field of complex numbers.
 V.9 Notes 119

V.9 Notes
There exist numerous textbooks on the theory of Lie algebras. See, for
instance, [Bou60][Dix74][Hum72][Jac79][Ser65][Var74]. The content of this
chapter is essentially taken from these sources. We found the proof of The-
orem 4.6 in Serre's book [Ser65]. As for Definition 7.1, we took it from
[Tak81]. Let us supplement the content of this chapter with the following
remarks.
(Free Lie algebras) Let X be a set. Consider the smallest Lie sub algebra
(X) of the free algebra k{X} containing X. Denote by ix the injection
of X into (X). The free Lie algebra (X) enjoys the following universal
property: For any set-theoretic map f from X into a Lie algebra L, there
exists a unique morphism of Lie algebras J : (X) ----+ L such that f = JOi x'
It follows from this universal property, from Proposition 1.2.1, and from
Theorem 2.1 that there is an isomorphism of algebras

U((X)) ~ k{X}.

A description of bases for (X) may be found in [Bou60], Chap. 2. See also
[Reu93].
(Primitive elements of the enveloping bialgebra) Any Lie algebra L is
contained in the Lie algebra of primitive elements of its enveloping algebra.
In characteristic zero, this embedding is an equality:

L = Prim(U(L)).
When applied to free algebras, one gets (X) ~ Prim(k{ X}) (see [Bou60],
Chap. 2).
(Real forms) A real form of a complex Lie algebra L is a real Lie subal-
gebra LR of L such that the embedding of the complexification LR EB iLR
into L is an isomorphism of complex Lie algebras. Here i denotes a square
root of -1. To any real form of L, one associates its conjugation, which is
the antilinear involutive endomorphism of Lie algebras a given by

a(x + iy) = x - iy
for all x, y E L R . Conversely, given any such involution of L, we obtain a
real form by
LR = {x ELI a (x) = x}.
For any real form of L with conjugation a, we define a Hopf *-algebra
structure on the enveloping algebra U (L) by * = SoU (a). In other words,
we have 1* = 1 and

for all Xl' ... 'X n E L. Conversely, suppose we have a Hopf *-algebra struc-
ture on the enveloping algebra U (L). Since * is a coalgebra morphism, it
120 Chapter V. The Lie Algebra of 5L(2)

preserves the Lie sub algebra of primitive elements, which is L (we are in
characteristic zero). It is easy to check that the subspace of all elements
x of L such that x = -x* is a real form of L. We thus see that the real
forms on a complex Lie algebra L are in one-to-one correspondence with
the Hopf *-algebra structures on U(L).
For instance, the real Lie sub algebra su(2) of 2 x 2-matrices M in s((2)
such that M = _t M is a real form of s((2). The vectors A = ~(X - Y),
B = ~(X + Y), iH form a real basis of su(2) such that

[A,B] = C, [B,C] = A, [C,A] = B.

This proves that su(2) is isomorphic to the Lie algebra so(3) of real anti-
symmetric 3 x 3-matrices.
(Duality) Theorem 7.6 asserts the existence of a Hopf algebra morphism
from SL(2) to U(s((2))*. This morphism is actually an isomorphism from
SL(2) to the restricted dual U(s((2))o. This holds, more generally, for
any simply-connected algebraic group in characteristic zero (see [Abe80]
[Hoc8l] PS9lb] [Swe69]).
Chapter VI
The Quantum Enveloping Algebra
of 1[(2)

The aim of Chapters VI-VII is to construct a Hopf algebra Uq = Uq (5[(2))

which is a one-parameter deformation of the enveloping algebra of the Lie
algebra 5[(2) investigated in Chapter V, and which is in duality with the
Hopf algebra SLq(2) defined in Chapter IV. It will be our second main
example of a quantum group. When the parameter q is not a root of unity,
the algebra Uq has properties parallel to those of the enveloping algebra
of 5[(2). In the present chapter we classify the simple finite-dimensional
modules of Uq and determine its centre. We close the chapter with a few
considerations on the case when q is a root of unity.
We assume throughout this chapter that the ground field k is the field
of complex numbers.

VI. 1 The Algebra Uq (s[(2))

Let us fix an invertible element q of k different from 1 and -1 so that the
fraction ~
q-q
is well-defined. We introduce some notation.
For any integer n, set

[n] = q; ~ qq~ln = qn-l + qn-3 + ... + q-n+3 + q-n+l. (1.1)

These q-analogues are more symmetric than the ones defined in IV.2, as
shown by the relations
[-n] = -[n] and [m + n] = qn[m] + q-m[n]. (1.2)
122 Chapter VI. The Quantum Enveloping Algebra of .5[(2)

Observe that, if q is not a root of unity, then [n] =1= 0 for any non-zero
integer. This is not so when q is a root of unity. In that case, denote by d
its order, i.e., the smallest integer> 1 such that qd = 1. Since we assume
q2 =1= 1, we must have d > 2. Define also

d if d is odd
{ (1.3)
e = d/2 when d is even.

Let us agree that d =e= 00 when q is not a root of unity. Now it is easy
to check that
[n] = 0~n == 0 modulo e. (1.4)
We also have the following versions of factorials and binomial coefficients.
For integers 0 :::; k :::; n, set [OJ! = 1,

[k]! = [1][2] ... [k] (1.5)

if k > 0, and
[ n ] [n]! (1.6)
k - [k]![n - k]!
These q-analogues are related to those of IV.2 by

and
[ ~ ] = q-k(n-k) ( ~ ) q2 (1.8)

With this new notation we can rewrite Proposition IV.2.2 as follows. If x

and yare variables subject to the relation yx = q2 xy, then we have (n > 0)

(x + y)n = t
k=O
l(n-k) [ ~ ] xkyn-k. (1.9)

Definition VI. 1. 1. We define Uq = Ui5[(2)) as the algebra generated by

the four variables E, F, K, K- 1 with the relations

KK- 1 = K-1K = 1, (1.10)

KEK- 1 = q2E, KFK- 1 = q- 2F, (1.11)

and
[E,F] = K - K- 1 (1.12)
q _ q-l

The rest of the section is devoted to a few elementary properties of Uq .

The following lemma has an easy proof left to the reader.
 VI.l The Algebra Uq ('s!(2)) 123

Lemma VI.1.2. There is a unique algebra automorphism of Uq such that

w(E) = P, w(P) = E, w(K) = K- 1 .

The automorphism w is sometimes called the Cartan automorphism. We

now state a q-analogue of Lemma V.3.1.

Lemma VI.1.3. Let m 2: 0 and n E Z. The following relations hold in

Uq :

-(m-1) K _ m-1 K- 1
[m] p m - 1 q ~
q_ q 1

m-1 K _ -(m-1) K- 1
[m] q q -1 p m - 1,
q-q

-(m-1) K _ m-1 K- 1
[m] q q Em - 1
q _ q-1
m-1 K _ -(m-1) K- 1
[m] E m - 1 q q _
q_ q 1

PROOF. The first two relations result trivially from Relations (1.11). The
third one is proved by induction on musing

[E, pm] = [E, pm-1]p + pm-1 [E, P] = [E, pm-1]p + p m- 1K - ~~1

q-q
as in the proof of Lemma V.3.1. Applying the automorphism w to the third
relation, one gets the fourth one. D

We now describe a basis of Uq by showing that Uq is an iterated Ore

extension. We refer to 1.7-8 for information concerning Ore extensions.

Proposition VI.1.4. The algebra Uq is Noetherian and has no zero divi-

sors. The set {Ei pj KRL,jEN; fEZ is a basis of Uq .

PROOF. Define AD = k[K,K- 1 ]. We shall construct two Ore extensions

A1 C A2 such that A2 is isomorphic to Uq . First, observe that the algebra
AD has no zero divisors and is Noetherian as a quotient of a (Noetherian)
two-variable polynomial algebra. The family {KR} REZ is a basis of AD.
Consider the automorphism a 1 of AD determined by a 1 (K) = q2 K and
the corresponding Ore extension A1 = AD [P, a 1 ,0]: the latter has a basis
consisting of the monomials {pj KR}jEN,fEZ. An argument analogous to
the one used to prove Lemma IV.4.2 shows that A1 is the algebra generated
by P,K,K- 1 and the relation PK = q2KP.
124 Chapter VI. The Quantum Enveloping Algebra of 5((2)

We now build an Ore extension A2 = Al [E, aI' 8] from an automorphism

a 1 and an aI-derivation of AI' The automorphism a 1 is defined by

(1.13)

Let us take as given for a moment that there exists an aI-derivation 8 such
that
8(F) = K - ~-1 and 8(K) = O.
q_ q 1

Then the following relations hold in A 2 ;

EK = a 1 (K)E + 8(K) = q-2 KE

and
K _K- 1
EF=a 1 (F)E+8(F)=FE+ -1
q-q
From these one easily concludes that A2 is isomorphic to Uq . It then re-
sults from Corollary 1. 7.2 and from Theorem 1.8.3 that Uq has the required
properties. D

It remains to prove the following technical lemma in order to complete

the proof of Proposition 1.4.

Lemma VI.1.5. Denote by 8(F)(K) the Laurent polynomial ~=:-~' and ,

set 8(KR) = 0 and

j-l
8(Fj KR) = L Fj- 1 8(F)(q-2iK)KR (1.14)
i=O

when j > O. Then 8 extends to an a 1 -derivation of AI'

PROOF. We must check that, for all j, mEN and all , nEZ, we have

(1.15)

Let us compute the right-hand side of (1.15) using (1.11), (1.13), and (1.14).
We have

m-1

i=O
j-1
+ L pi-18(F)(q-2iK)KRFmKn
i=O
 VI.2 Relationship with the Enveloping Algebra of 5[(2) 125

L
m-l
q-2-2(m-l) FJ+m- 18(F)(q-2iK)KHn
i=O

L
j-l
+ q-2Rm F m+j- 18(F)(q-2i-2m K)KHn
i=O

L
m-l
q-2Rm F m+j- 18(F)(q- 2i K)KHn
i=O

L
j+m-l
+ q-2Rm Fm+j-18(F)(q-2iK)KHn
i=m
j+m-l
q-2Rm ( L FJ+m- 18(F) (q- 2i K)K Hn )
i=O
q-2m8(FJ+m KHn)
8(Fj K . F m Kn).

VI. 2 Relationship with the Enveloping Algebra of

.5[(2)
One expects to recover U = U(5[(2)) from Uq by setting q = 1. This is
impossible with Definition 1.1. So we first have to give another presentation
for Uq .

Proposition VI.2.1. The algebra Uq is isomorphic to the algebra U~ gen-

erated by the five variables E, F, K, K- 1 ,L and the relations

KK- 1 = K- 1K = 1, (2.1)

KEK- 1 = q2E, KFK- 1 = q- 2F, (2.2)

[E,Fj=L, (q-q-l)L=K-K- 1, (2.3)
[L,Ej=q(EK+K-1E), [L,Fj=-q-l(FK+K-1F). (2.4)
Observe that, contrary to Uq , the algebra U~ is defined for all values of
the parameter q, in particular for q = 1. In some sense, it would have been
better to proceed through the whole theory of the quantum enveloping
algebra of 5[(2) with U~ rather than with Uq , but the simpler presentation
given in Section 1 is sufficient for our purposes.
PROOF. Set
cp(E) = E, cp(F) = F, cp(K) =K
126 Chapter VI. The Quantum Enveloping Algebra of 5((2)

and
?/J(E) = E, ?/J(F) = F, ?/J(K) = K, ?/J(L) = [E, F].
It is clear that r.p gives rise to a well-defined morphism of algebras from Uq
to U~. Let us show that ?/J : U~ ----+ Uq is well-defined too. It suffices to check
that the images under ?/J of the defining Relations (2.1) hold in the algebra
Uq . This is clearly true for Relations (2.1 ~ 2.2) and for [E, F] = L. For the
remaining relation in (2.3) we have
(q - q~l)?/J(L) = (q - q~l)[E, F] = K - K~l.

For the first relation in (2.4) we get

[?/J(L),?/J(E)] = [[E,F],E] 1 [K _ K~l,E]

q_ q~l

(q2 _ l)EK + (q2 _ l)K~l E

q _ q~l
q (EK + K~l E).

One derives the last relation in a similar fashion.

The reader may now verify that r.p and ?/J are reciprocal algebra mor-
phisms by checking the necessary relations on the generators. 0

The relationship with the enveloping algebra U is given in the following

statement.
Proposition VI.2.2. If q = 1, we have
U{ ~ U[K]/(K2 - 1) and U ~ U{/(K - 1).
PROOF. It suffices to prove the first isomorphism. Now U{ has the following
presentation: it is generated by E, F, K, K~\ L and Relations (2.1 ~2.4) in
which q has been replaced by 1, namely
K K~l = K~l K = 1, (2.5)
KEK~l = E, KFK~l = F, (2.6)
[E, F] = L, K - K~l = 0, (2.7)
[L, E] = (EK + K~lE), [L, F] = -(F K + K~l F). (2.8)
Relations (2.5~2.6) imply that K is central. Relation (2.7) yields K2 = 1,
which allows one to rewrite the Relations (2.8) as
[L,E] = 2EK, [L,F] = -2FK. (2.9)
We then get an isomorphism from U{ to U[K]/(K2 - 1) by sending E to
X K, F to Y, K to K, and L to H K. 0

In particular, the projection of U{ onto U is obtained by sending E to

X, F to Y, K to 1, and L to H. One may use this projection to rederive
certain relations in U (for instance, Lemma V.3.1) from their q-analogues
. u'q'
In
 VI.3 Representations of Uq 127

VI. 3 Representations of Uq
We assume in this section that the complex parameter q is not a root of
unity. Our aim is to determine all finite-dimensional simple Uq-modules
under this assumption by closely following the methods of Section V.4.
For any Uq-module V and any scalar A =I=- 0, we denote by VA the subspace
of all vectors v in V such that K v = Av. The scalar A is called a weight of
V if VA =I=- {O}.

Lemma VI.3.1. We have EVA C Vq2A and FV A C Vq-2 A.

PROOF. For v E VA we have

o
Definition VI.3.2. Let V be a Uq-module and A be a scalar. An element
v =I=- 0 of V is a highest weight vector of weight A if Ev = 0 and if K v = Av.
A Uq-module is a highest weight module of highest weight A if it is generated
by a highest weight vector of weight A.
Proposition VI.3.3. Any non-zero finite-dimensional Uq-module V con-
tains a highest weight vector. Moreover, the endomorphisms induced by E
and F on V are nilpotent.
PROOF. Since k = C is algebraically closed and V is finite-dimensional,
there exists a non-zero vector wand a scalar a such that K w = aw. If
Ew = 0, the vector w is a highest weight vector and we are done. If not,
let us consider the sequence of vectors Enw where n runs over the non-
negative integers. According to Lemma 3.1, it is a sequence of eigenvectors
with distinct eigenvalues; consequently, there exists an integer n such that
Enw =I=- 0 and En+1w = O. The vector Enw is a highest weight vector.
In order to show that the action of E on V is nilpotent, it suffices to check
that 0 is the only possible eigenvalue of E. Now, if v is a non-zero eigen-
vector for E with eigenvalue A =I=- 0, then so is Knv with eigenvalue q-2n A.
The endomorphism E would then have infinitely many distinct eigenvalues,
which is impossible. The same argument works for F. 0

Lemma VI.3.4. Let v be a highest weight vector of weight A. Set Vo = v

and vp = [;J! FPv for p > O. Then

q-(p-1) A _ qP-1 A-I

Evp = -1 vp_ 1,
q-q
PROOF. These relations result from Lemma 1.3. o
We now determine all finite-dimensional simple Uq-modules.
128 Chapter VI. The Quantum Enveloping Algebra of -5[(2)

Theorem VI.3.5. (a) Let V be a finite-dimensional Uq-module generated

by a highest weight vector v of weight A. Then
(i) The scalar A is of the form A = c qn where c = 1 and n is the integer
defined by dim (V) = n + 1.
(ii) Setting vp = FPv/[p]!, we have vp = 0 for p > n and, in addition,
the set {v = v O ,v 1 , .. ,vn } is a basis ofV.
(iii) The operator K acting on V is diagonalizable with the (n+1) distinct
.
ezgenva Iues {n
cq ,cq n-2 , ... , cq -n+2 , cq -n} .
(iv) Any other highest weight vector in V is a scalar multiple ofv and is
of weight A.
( v) The module V is simple.
(b) Any simple finite-dimensional Uq-module is generated by a high-
est weight vector. Two finite-dimensional U -modules generated by highest
weight vectors of the same weight are isomorphic.

PROOF. (a) According to Lemma 3.4, the sequence {v p }P2:0 is a sequence of

eigenvectors for K with distinct eigenvalues. Since V is finite-dimensional,
there has to exist an integer n such that vn # 0 and v n+1 = O. The formulas
of Lemma 3.4 then show that vm = 0 for all m > nand vm # 0 for all
m :So n. By Lemma 3.4, we also have

Hence, q-n A = qn A-1, which is equivalent to A = qn. The rest of the

proof of (i)-(iii) is as in the classical case (see Theorem V.4.4).
(iv) Let Vi be another highest weight vector. It is an eigenvector for the
action of K; hence, it is a scalar multiple of some vector Vi' But, again by
Lemma 3.4, the vector Vi is killed by E if and only i = O.
(v) Let Vi be a non-zero Uq-submodule of V and let Vi be a highest
weight vector of V'. Then Vi also is a highest weight vector for V. By (iv),
Vi has to be a non-zero scalar multiple of v. Therefore v is in V'. Since v
generates V, we must have V c V', which proves that V is simple.
(b) The proof is the same as for Theorem V.4.4 (b). D

Theorem 3.5 implies that, up to isomorphism, there exists a unique sim-

ple Uq - module of dimension n + 1 and generated by a highest weight vector
of weight cqn. We denote this module by VE n and the corresponding mor-
phism of algebras Uq --+ End(~,n) by PE;,n-' Observe that the formulas of
Lemma 3.4 may be rewritten as follows for ~,n:

Kv = cqn-2Pv (3.1)
P P'

(3.2)
and
(3.3)
 VI.3 Representations of Uq 129

As a special case, we have ~,a = k. The morphism Pc,a is given by

Pc,a(K) = c, Pc,a(E) = Pc,a(F) = O.

We shall see in VII.I that Pc,a may be identified with the counit of a Hopf
algebra structure on Uq . It will imply that the module VI,a is trivial and
that any trivial Uq-module is isomorphic to a direct sum of copies of VI,a'
On the other hand, the module V_I a is not trivial.
On the (n + 1)-dimensional module Vc n' the generators E, F and K act
by operators that can be represented o~ the basis {va, vI' ... , v n } by the
matrices
0 [n] 0 0
0 0 [n-I] 0

Pc,n(E) = c
0 0 I
0 0 0 0
0 0 0 0
I 0 0 0

Pc,n(F) = 0 [2] 0 0

0 0 [n] 0
and
qn 0 0 0
0 qn-2 0 0
Pc,n(K) = c
0 0 q-n+2 0
0 0 0 q-n

So far, we have built Uq-modules generated by highest weight vectors

whose weights>. had special values. Let us now show that there exist highest
weight modules with arbitrary highest weights.
Let us fix a scalar>' #- O. Consider an infinite-dimensional vector space
V(>.) with denumerable basis {VJiEN' For p 2 0, set

(3.4)

and EVa = O.
Lemma VI.3.6. Relations (3.4-3.5) define a Uq-module structure on V(>.).
The element va generates V(>.) as a Uq-module and is a highest weight vec-
tor of weight >..
130 Chapter VI. The Quantum Enveloping Algebra of 5[(2)

PROOF. Immediate computations yield

KK-1v =
P
KEK-1v p =

We also have

[E,F]vp =

This proves that Relations (3.4-3.5) define a Uq-module structure on V('\).

Next, we have K Vo = .\ Vo and Evo = 0, which means that Vo is a highest
weight vector of weight .\. Finally, (3.5) implies that vp = FPvo/[p]! for all
p, which proves that V('\) is generated by vo' 0

By analogy with the classical case, the highest weight Uq-module V('\)
is called the Verma module of highest weight .\. It enjoys the following
universal property.

Proposition VI.3.7. Any highest weight Uq-module V of highest weight

.\ is a quotient of the Verma module V('\).

PROOF. Let v be a highest weight vector generating V. We define a linear

map f from V('\) to V by f(v p) = l/[p]! FPv. Lemma 3.4 implies that f is
Uq-linear. Since f(v o ) = v generates V, the map f is surjective. 0

In particular, the simple finite-dimensional module VE n described above

is a quotient of the Verma module V(cqn). As a conseq~ence, the module
V('\) cannot be simple when .\ is of the form qn where n is a nonnegative
integer.

VI.4 The Harish-Chandra Homomorphism and

the Centre of Uq
Our next objective is to describe the centre Zq of Uq in case q is not a root
of unity. We assume this throughout this section.
We start by introducing a special central element of Uq It is sometimes
called the quantum Casimir element.
 VI.4 The Harish-Chandra Homomorphism and the Centre of Uq 131

Proposition VI.4.1. The element

belongs to the centre of Uq .

PROOF. It suffices to check that C q commutes with the generators K, E, F.

The commutation with K is clear from KEFK- I = EF. As for E, we
have

Similar argument gives the result for F. D

Let U;: be the subalgebra of Uq of all elements commuting with K.

Lemma VI.4.2. An element of Uq belongs to U;: if and only if it is of

the form
(4.1)

where Po, PI' ... are elements of k[K, K- I ].

PROOF. This is a consequence of the fact that {FiKiEjL,jEN;iEZ is a

basis of Uq and that K(FiKiEj)K- I = q2(j-i) FiKiEj. D

Let us consider the left ideal 1= UqE n U;: of U;:.

Lemma VI.4.3. We have I = FUq n U;: and U;: = k[K, K- I ] EB I.

PROOF. Let u = Li>O Fi PiE i be an element of U;:. If u also lies in UqE,
then Po = O. Hence, ;, belongs to FUq n U;: and conversely. Since the form
(4.1) is unique for any element of U;:,
we get the desired direct sum. D

It results from I = FUq n U;: that I is a two-sided ideal and that the
projection 'P from U;: onto k[K, K- I ] is a morphism of algebras. The map
'P is called the Harish-Chandra homomorphism. It permits one to express
the action of the centre Zq on a highest weight module.

Proposition VI.4.4. Let V be a highest weight Uq -module with highest

weight A. Then, for any central element z of Uq and any v E V, we have

zv = 'P(z)(A)V.
132 Chapter VI. The Quantum Enveloping Algebra of .5[(2)

Recall that cp(z) is a Laurent polynomial in K and that cp(z)(>.) is its

value at >..
PROOF. Let Vo be a highest weight vector generating V and z a central
element of Uq . The element z can be written in the form

z = cp(z) + I: FiPiE i .
i>O

Since Evo = 0 and Kvo = >.vo, we get zVo = cp(z)(>.)v o. If v is an arbitrary

element of V, we have v = xVo for some x in Uq ; hence,
zv = zxvo = xzvo = cp(z) (>.)xvo = cp(z)(>.)v.
o
Example 1. The definition of the central element Cq shows that
qK - q-1K- 1
cp(Cq ) = (q _ q-l )2 (4.2)

Consequently, C q acts on a highest weight module of highest weight >. as

the multiplication by the scalar
q>. + q-l>.-l
(q_q-l)2 . (4.3)

Let us now prove that the restriction of the Harish-Chandra homomor-

phism to the centre Zq is injective.
Lemma VI.4.5. Let z E Zq. If cp(z) = 0, then z = o.
PROOF. Let z be an element in the centre such that cp(z) = o. Assume
", . .
z non-zero; it can be written as z = ui=k F' PiE' where 0 < k ::; are
integers and Pk , ... , P are non-zero Laurent polynomials in K. Consider
a Verma module V(>') whose highest weight is not a power of q. Then
Relations (3.4-3.5) show that Evp = 0 if and only if p = o. Let us apply
z to the vector vk of V(>'). On the one hand, Proposition 4.4 implies that
zV k = cp(z) (>')vk = 0; on the other, we get
zV k = Fk PkEkvk = c Pk(>.)v k ,
where c is a non-zero constant. It follows that Pk (>') = o. As a consequence,
we have a non-zero polynomial Pk with infinitely many roots; hence a
contradiction. 0
Verma modules will also allow us to prove a symmetry relation for the
polynomials cp(z). Before we state this, let us introduce the following nota-
tion. For any Laurent polynomial P in k[K,K- 1 ], denote by P the poly-
nomial defined by the change of variable
P(>.) = p(q-l >.).
 VI.4 The Harish-Chandra Homomorphism and the Centre of Uq 133

Lemma VI.4.6. For any element z in the centre Zq' we have

PROOF. For any integer n > 0, consider the Verma module V(qn-1). By
(3.5) we have

EV n =
q-(n-1)qn-1 - qn-1 q-(n-1)
-1
_
Vn -

.
q-q
Thus, Vn is a highest weight vector of weight qn-1-2n = q-n-1. By Propo-
sition 4.4, a central element z acts on the module generated by vn as the
mUltiplication by the scalar <p(z)(q-n-1); but, since vn is in V(qn-1), the
element z also acts as the scalar <p(z)(qn-1). In other words, we have

One concludes by observing that the powers of q form an infinite sequence

of distinct scalars. 0

We pause to record the following lemma.

Lemma VI.4.7. Any Laurent polynomial of k[K, K- 1] satisfying the re-
lation P(.>.) = P(.>. -1) is a polynomial in K + K- 1.
PROOF. We proceed by induction on the degree of the polynomial. If the
degree is 0, the statement holds trivially. Let us suppose that the lemma is
proved for all degrees < n and let P be a Laurent polynomial of degree n
such that P(.>.) = P(.>. -1). Then we may write P in the form

P(K) = c (Kn + K- n ) + (terms of degree < n).

Now,
K n + K- n = (K + K- 1)n + (terms of degree < n).
One concludes by applying the induction hypothesis. o
We are ready to state the main theorem.
Theorem VI.4.8. When q is not a root of unity, the centre Zq of Uq is
a polynomial algebra generated by the element C q . The restriction of the
Harish-Chandra homomorphism to Zq is an isomorphism onto the subalge-
bra of k[K, K- 1] generated by qK + q-1 K- 1.
PROOF. We already know that the restriction of <p to the centre is injective.
We are left with determining its image. By Lemmas 4.6 and 4.7, the latter
is contained in the subalgebra of k[K,K- 1] generated by qK + q- 1K- 1.
Consider the central element C q defined above. By (4.2) we know that
1
<p(Cq ) = (q-q _1)2(qK +q- 1K- 1),
134 Chapter VI. The Quantum Enveloping Algebra of ,5[(2)

which proves that the image of Zq is the whole subalgebra and that Cq
generates the centre. The latter is a polynomial algebra because the powers
of qK + q-l K- 1 are linearly independent for obvious reasons of degree.
o

VI. 5 Case when q is a Root of Unity

Our next aim is to find all finite-dimensional simple Uq-modules in the case
when the complex parameter q is a root of unity i= 1. As we shall quickly
see, the situation is much more complicated than in the generic case when
q is not a root of unity. Define the order d of q and the integer e as in (1.3).
Recall that [e] = O.
The following theorem asserts that the simple Uq-modules of sufficiently
low dimensions are the same as in the generic case.

Proposition VI.5.1. Any simple non-zero Uq-module of dimension < e

is isomorphic to a module of the form VE,n where c = 1 and 0 :S n < e-l.

The modules ~,n have been described in Section 3.

PROOF. The proof is exactly the same as the proof of Theorem 3.5. One
uses the fact that 1, q2, ... ,q2n are distinct scalars when n < e. 0

The first big difference with the generic case appears in the following
statement.

Proposition VI.5.2. There is no simple finite-dimensional Uq-module of

dimension > e.

Before we prove this proposition, we state two lemmas. The first one
implies that the centre of Uq is much bigger when q is a root of unity than
when it is not. The second one is a special case of a general statement on
finite-dimensional modules.

Lemma VI.5.3. The elements E e , Fe, and K e belong to the centre of Uq

PROOF. This is a consequence of Relation (1.1) and of Lemma 1.3. Indeed,

E e commutes with K because q2e = 1 and with F because [e] = O. Similar
arguments can be applied to Fe and to K e. 0

Lemma VI.5.4. Let z be a central element of Uq . Then z acts on any

finite-dimensional simple Uq-module V by multiplication by a scalar.

PROOF. Let u be the endomorphism induced by the action of z on V: it

is Uq-linear because z is central. Since V is finite-dimensional, the endo-
morphism u has an eigenvalue A. Consider the Uq-linear endomorphism
 VI.5 Case when q is a Root of Unity 135

u - >.id v . Its kernel K is a submodule of the simple module V. Since

K -=I- {O}, we must have K = V. 0

Proof of Proposition 5.2. Let us assume that there exists a simple finite-
dimensional module V of dimension > e. We shall prove that V has a
non-zero submodule of dimension::; e. Hence, a contradiction.
(a) Suppose there exists a non-zero eigenvector v E V for the action
of K such that Fv = O. We claim that the subspace V' generated by
v, Ev, ... ,Ee- 1v is a submodule of dimension::; e. It is enough to check
that V'is stable under the action of the generators E, F, K. This is clear
for K. Let us check that V'is stable under E. The vector E(EPv) = EP+1 V
belongs to V' if p < e - 1. If p = e - 1, we have

where c 1 is a scalar in view of Lemmas 5.3 and 5.4. Finally, V'is stable
under F thanks to Fv = 0 and Lemma 1.3.
(b) Now, suppose there is no non-zero eigenvector v E V for the action of
K such that Fv = O. Let v be a non-zero eigenvector for the action of K. We
have Fv -=I- O. We claim that the subspace V" generated by v, Fv, ... , F e- 1v
is also a submodule of dimension::; e. Again, V" is clearly stable under K.
It is also stable under F since the vector F(FPv) = FP+1 v belongs to V"
if P < e - 1. If p = e - I, we have

where c2 is another scalar, again in view of Lemmas 5.3 and 5.4. The scalar
c2 is not zero; otherwise, there would exist an integer p < e such that FP v
would be an eigenvector for K killed by F, which would contradict our
assumption.
In order to check that V" is stable under E, we use the central element
Cq defined in Section 4. By Lemma 5.4, it acts on V by multiplication by
a scalar c3 . By definition of Cq we get for p > 0

E(FPv) EF(FP-1 V )

(c q _ q-1(q_q-1)2
K + qK- 1 ) (FP-1 V )

F P-1 _ q-1 K + qK- (FP-1 )

1
c3 V (-1)2
q-q V ,

which show;:; that E(FPv) sits in V". When p = 0, we use the same argu-
ment after observing that v = C;-l Fev. 0

It remains now to find the simple Uq-modules in dimension e. We shall

content ourselves with their descriptions, omitting proofs. First, we give
two families of e-dimensional modules.
136 Chapter VI. The Quantum Enveloping Algebra of 5[(2)

The first one depends on three complex numbers A, a, and b. We assume

that A -I O. Consider an e-dimensional vector space equipped with a basis
{vo,'" ,Ve-I}' For O:S p < e -1, set

(5.1)

(5.2)

Fvp = v p+ I ' (5.3)

and Evo = aVe_I, FV e_ 1 = bvo , and KV e_ 1 = Aq-2(e-l)v e_ l . These formu-
las endow this vector space with a Uq-module structure, denoted V(A, a, b).
The second family depends on two scalars p, -I 0 and e. We let E, F, K
act on a vector space with basis {v o ,"" v e - I } by

(5.4)

(5.5)

(5.6)
if 0 :S p < e - 1 and by Fv o = 0, EV e_ 1 = ev o , and KV e_ 1 = p,q-2 ve _ 1
otherwise. These formulas determine another Uq-module, denoted V(p" e).
The following theorem which we admit without proof closes the list of
all simple finite-dimensional Uq-modules when q is a root of unity.
Theorem VI.5.5. Any simple Uq-module of dimension e is isomorphic to
a module of the following list:
(i) V(A, a, b) with b -I 0,
(ii) V(A, a, 0) where A is not of the form qj-I for any 1 :S j :S e - 1,
(iii) V(ql-j, e) with e -I 0 and 1 :S j :S e - l.

It should be added that all modules V (A, a, b) and V(p" e), including the
ones that are not in the list of Theorem 5.5, are indecomposable.
In the situation under investigation, the algebra Uq possesses an inter-
esting finite-dimensional quotient-algebra.
Definition VI.5.6. The algebra U q is the quotient of the algebra Uq by
the two-sided ideal generated by the central elements E e , Fe, and K e - 1.

It is not difficult to convince oneself that a finite-dimensional U q-module

is simple [resp. indecomposable] if and only if it is simple [resp. indecom-
posable] as a Uq-module. Therefore, in order to have a complete list of all
simple finite-dimensional U q-modules, it is enough to determine the simple
finite-dimensional Uq-modules on which E e , Fe and K e - 1 act by O. This
is done without any difficulty using Theorem 5.5 and Relations (5.1-5.6).
We get the following:
 VI.5 Case when q is a Root of Unity 137

Theorem VI.5.7. Any non-zero simple finite-dimensional U q-module is

isomorphic to a module of the form
(i) VI,n with 0 :::; n < e - 1, or V(q-I, 0, 0) if d = e is odd,
(ii) VI,n with n even < e - 1 if d and e are even,
(iii) VI,n with n even < e-1, or V-I,n with n odd < e-1, or V( _q-I, 0, 0)
if d is even and e is odd.
We shall need the following proposition in IX.6.
Proposition VI.5.S. The finite se~ {Ei Fj KRh:S;i,j,R:s;e-1 is a basis of U q'
PROOF. Thanks to the commutation relations between the generators,
we are reduced to showing that {Fj K Eih:S;i,j,:S;e-1 is a basis of U q'
By Proposition 1.4 it is clear that this set generates U q' It remains to
check that it is free. To this end, we introduce an intermediate quotient-
algebra Uq defined by Uq = Uq/(Ee,F e ) and we show first that the set
{Fj KR Ei }O:S;i,j:S;e-l; REZ is a basis of Uq Let us prove this claim. Again, it
is enough to prove that the set is free.
Let us consider a linear relation of the form
z= .
aijR FJ K E~
.
= O. (5.7)
O:S;i,j:S;e-l; r:S;:S;s

We let it act on the vectors vp of the canonical basis of the module V(A, 0, 0)
(check that this module is killed by E e and Fe, but in general not by K e -1).
We assume that A is neither zero, nor a root of unity. Since Evo = 0, we
have

(5.8)
Since v o, ... , v e - I are linearly independent, Relation (5.8) implies that
s-r
2: aO,j,Hr AR = 0 (5.9)
=0

for all j. Writing (5.9) for s - r + 1 distinct complex numbers A, we get a

linear system whose determinant is a non-zero Vandermonde determinant.
Consequently, a Oj = 0 for all j and f. Next, we apply Z to the vector
VI' The hypothesis made on A implies that EVI is a non-zero multiple of
Voi hence we get aIje = 0 for all j and f by the same argument as above.
Applying Z successively to the vectors v 2 up to v e - I , one shows that all
coefficients aij!' vanish.
Now that we have secured a basis for Uq , we prove Proposition 5.8. We
consider a linear relation of the form
j i
aijeF K E = O. (5.10)
138 Chapter VI. The Quantum Enveloping Algebra of g[(2)

in U q' Denoting by Z the element of Uq represented by the left-hand side of

(5.10), we see Z belongs to the two-sided ideal of Uq generated by K e - 1.
Hence, we have Z = (Ke -l)Y where Y = L:O:S:i,j:S:e-l;i'EZ fJiji'FjKfE i .
Since K e is central, we get
j f i
Z= fJiji'F K E. (5.11)
O:S:i,j:S:e-l; fEZ O:S:i,j:S:e-l; fEZ

Assume Z -=I- 0, hence Y -=I- O. Denote by d(Z) [resp. by 8(Z)] the degree in
K [resp. the degree in K- 1 ] of the non-zero element Z of Uq written in the
above-mentioned basis. Relation (5.11) implies that
d(Z) = d(Y) +e and 8(Z) = 8(Y). (5.12)
Now, by definition of Z, we have
o :::; 8(Z) :::; d(Z) < e. (5.13)
Combining (5.12-5.13), we get d(Y) < 0 :::; 8(Z) = 8(Y). This is impossible;
hence, Z = O. 0

VI.6 Exercises
1. Compute [Ei, Fj] in Uq .

2. (Simple Verma modules) Assume that q is not a root of unity. Show

that the Verma module V('x) is simple if and only if ,X is not of the
form ,X = qn with n E N.
3. Prove Theorem 5.5.
4. Prove Theorem 5.7.
5. Assume that q is of finite order d > 2. Let ,X be a non-zero scalar.
Consider the Verma module V('x). Show that Fevo generates a highest
weight submodule of weight ,X and that the quotient V('x) of V('x) by
this submodule is a simple Uq-module of dimension e.
6. Under what conditions on ,x, a, and b is the module V(,X, a, b) of Sec-
tion 5 a highest weight module?

VI. 7 Notes
The algebra Uq = Uq (g[(2)) is due to Kulish and Reshetikhin [KR81]. Drin-
feld [Dri85][Dri87] and Jimbo [Jim85] independently generalized this con-
struction by defining an algebra Uq(g) for any complex semisimple Lie
algebra (more generally, for any symmetrizable Kac-Moody Lie algebra) g.
 VI.7 Notes 139

A complex semis imp Ie Lie algebra is determined by its so-called Cart an

matrix (aij)1~i,.i9 (see [Bou60], chap. 8, [Hum72], [Ser65]). In case g is
of type A, D or E, the Cart an matrix (aij)l~i,j~ is symmetric, positive
definite with integral coefficients such that a ii = 2 and aij = 0, -1 if i -I- j.
Then Drinfeld-Jimbo's algebra Uq(g) can be presented as follows: it is the
algebra generated by (Ei' Fi , K i , Ki~l )l~i~ and the relations

K i Ej K~l
i = q aE
'J .i' K i Fj Ki~l = q
~aF
'J j'

K-K-:-l
[Ei' Fj ] = bij q-q
' ~'l'

EiEj = EjEi and FiFj = FjFi if aij = 0,

E; E j - [2] EiEjEi + EjE; = 0 and F; F j - [2] FiFjFi + FjF; = 0,

if aij = -1. When aij = 0 if Ii - j I > 1 and aij = -1 if Ii - j I = 1, we
obtain Uq(.5[({i + 1)). A presentation of Uq(g) corresponding to the algebra
U~ was given by Lusztig [Lus89].
The algebra U q (g) possesses a Poincare-Bir khoff-Witt-type basis ([Lus90a]
[Lus90b] [Ros89] [Yam89]) and a quantum Casimir element (see [Jim85]).
Lusztig [Lus88] and Rosso [Ros88] proved that, when q is not a root of
unity, any finite-dimensional simple I-module could be deformed into a
finite-dimensional simple Uq(g)-module. A quantum Harish-Chandra ho-
momorphism was constructed by [CK90] [JL92] [Ros90] [Tan90].
Numerous authors have investigated the algebras Uq(g) and their re-
presentations when q is a root of unity, for instance, [CK90] [CKP92]
[DJMM91] [Lus89] [Lus90b] [RA89] [SaI90] (see also [Ros92]). We refer to
[CK90] [CKP92] for a description of the centre of Uq : it is a finite extension
of the polynomial subalgebra generated by E e , Fe and K e . Contrary to the
generic case, there is a bound for the dimension of the finite-dimensional
simple Uq-modules. For I = .5[(2), this bound is e (see Proposition 5.2).
We owe the treatment of Section 5 (including statements and proofs) to
R. Berger.
Chapter VII
A Hopf Algebra Structure on
Uq (s[(2) )

We assume in this chapter that the field k is the field of complex numbers
and that q is not a root of unity. We now equip the algebra Uq = Uis[(2))
defined in Chapter VI with a Hopf algebra structure. Then we prove that
any finite-dimensional Uq-module is a direct sum of the simple modules de-
scribed in VI.3. We show later that Uq acts naturally on the quantum plane
of IV.1 and that it is in duality with the Hopf algebra SLq(2) of Chapter
IV. We shall also build scalar products on the simple finite-dimensional
Uq-modules. We describe the quantum Clebsch-Gordan formula and give
the main properties of the quantum Clebsch-Gordan coefficients.

VII. 1 Comultiplication
We resume the notation of the previous chapter. Set

/1(E) = 1 Q9 E + E Q9 K, /1(F) = K- 1 Q9 F + F Q9 1, (1.1)

/1(K) = K Q9 K, /1(K-l) = K- 1 Q9 K- 1 , (1.2)
c(E) = c(F) = 0, c(K) = c(K-l) = 1, (1.3)
and

S(E) = -EK-l, S(F) = -KF, S(K) = K- 1 , S(K- 1 ) = K. (1.4)

Proposition VILLI. Relations (1.1-1.4) endow Uq with a Hop! algebra

structure.
 VII.1 Comultiplication 141

PROOF. (a) We first show that 6. defines a morphism of algebras from Uq

into Uq I8l Uq . It is enough to check that

6.(K)6.(K-l) = 6.(K- 1 )6.(K) = 1, (1.5)

6.(K)6.(E)6.(K-l) = q26.(E), (1.6)

6.(K)6.(F)6.(K-l) = q-26.(F), (1.7)

[6.(E), 6.(F)] = 6.(K) - 6._(~-1) . (1.8)

q-q
Relations (1.5) are clear. As for (1.6), we have

(K I8l K)(l I8l E + E I8l K)(K- 1 I8l K- 1 )

1 I8l KEK- 1 + KEK- 1 I8l K
q2 (1 I8l E + E I8l K)
q26.(E).

Relation (1. 7) is proved in a similar way. Finally, for (1.8) we have

[6.(E),6.(F)]
(1 I8l E +E I8l K)(K- 1 I8l F +F I8l 1)
- (K- 1 I8l F +F I8l 1)(1 I8l E +E I8l K)
K- 1 I8l EF + F I8l E + EK- 1 I8l K F + EF I8l K
- K- 1 I8l FE - K- 1 E I8l F K - F I8l E - FE I8l K
K- 1 I8l [E, F] + [E, F] I8l K
K- 1 I8l (K - K- 1 ) + (K - K- 1 ) I8l K
q _ q-l
6.(K) - 6.(K-l)
q _ q-l

(b) Next, we check that 6. is coassociative. It suffices to do it on the four

generators. We give a sample calculation for E. On the one hand, we have

On the other hand, we have

(idl8l6.)6.(E) = (idl8l6.)(lI8lE+EI8lK) = 118l118lE+118lEI8lK +EI8lKI8lK,

which is the same.

(c) It is easy to check that c defines a morphism of algebras from Uq onto
k and satisfies the counit axiom.
142 Chapter VII. A Hopf Algebra Structure on Uq (.s[(2))

(d) It remains to see that S defines an antipode for Uq . We have first to

check that S is a morphism of algebras from Uq into U~P, namely that the
following four relations hold:

S(K- 1)S(K) = S(K)S(K-l) = 1, (l.9)

S(K-l)S(E)S(K) = q2S(E), (l.10)

S(K-l)S(F)S(K) = q- 2S(F), (1.11)

[S(F), S(E)] = S(K) - S~~-l). (l.12)

q-q
We give the computations for (l.1O) and (l.12). We have

and

[S(F), S(E)] KFEK- 1 - EK-1KF = [F,E]

K- 1 - K S(K) - S(K- 1)

To conclude that S is an antipode, we appeal to Lemma III.3.6. It suffices

to check that the relations

L X'S(X") = L S(x')x" = c(x)1

(x) (x)

hold when x is any of the generators E, F, K, K- 1 . This verification is left

to the reader. 0

We have thus defined a Hopf algebra that is neither commutative nor

cocommutative. Observe also that the square of the antipode is not the
identity (when q2 i= 1). Nevertheless, it is an inner automorphism, as ex-
pressed by the following statement.

Proposition VII.1.2. We have S2(U) = KuK- 1 for any u E Uq

PROOF. In effect, we have

S2(E) = q2E = KEK-l, S2(F) = q- 2F = KFK-l,

and S2(K) = K. 0

We thus get, just as in Chapter IV, examples of Hopf algebras whose

antipodes have a finite order 2N for any integer N > 1; it suffices to take
any primitive 2N-th root of unity as the parameter q.
 VII.2 Semisimplicity 143

The algebra U~ of VI.2 can be endowed with a Hopf algebra structure

such that the isomorphism rp : Uq - t U~ of Proposition VI.2.1 preserves the
Hopf algebra structures. In addition to Relations (1.1-1.4), it suffices to set

6.(L) = K- 1 @ L + L @ K, c(L) = 0, S(L) = -L. (1.13)

It follows easily that the isomorphism U(st(2)) ~ U{j(K -1) is an isomor-

phism of Hopf algebras. In other words, the Hopf algebra structure of Uq
extends the Hopf algebra structure of the enveloping algebra U(st(2)).
We end this section by expressing the comultiplication of Uq in the basis
described in Proposition VI.1.4.

Proposition VII.1.3. For all i,j EN and . E Z we have

tt
r=Os=O
qr(i-r)+s(j-s)-2(i-r)(j-s) [ ~ ] [ ~ ]

PROOF. First observe that

6.(~)i6.(F)j6.(K)R
(1 @ ~ + ~ @K)i(K-l @F+F@ 1)1(KR @K R).
Now,
(~@ K)(1 @~) = q2 (1 @ ~)(~ @ K)
and
(K- 1 @ F)(F @ 1) = q2 (F @ 1)(K- 1 @ F).
Applying Relation (VI.1.g), we get

6.(~)i = t
r=O
qr(i-r) [ ; ] ~i-r @ ~r K i - r

and
6.(F)j = ts=o
qs(j-s) [ ~ ] F S K-(j-s) @ Fj-s.

One concludes with (VI. 1. 11). D

VII.2 Semisimplicity
In this section we shall prove that any finite-dimensional Uq-module is the
direct sum of simple Uq-modules when q is not a root of unity, which we
assume in this chapter. Let us start with a technical lemma on the simple
modules Vc:.n of VI.3.
144 Chapter VII. A Hopf Algebra Structure on Uq (s[(2))

Lemma VII.2.1. There exists an element C of the centre of Uq acting by

o on ~,o and by a non-zero scalar on ~I,n when n is an integer> 0 and
c,c' = l.

PROOF. Define
q + q-l
C = Cq - c (q-q -1 )2

where Cq is the central element introduced in VI.4. By (VI.4.3), C acts on

~,o by
q+q-l q+q-l
c (q-q -1)2 - c (q-q -1)2 = 0,

and on ~I,n by

We have to show that the latter is not O. If it were, we would have

or, equivalently,
(qn+2 _ cc/)(qn - cc/) = 0,
which would be contrary to the assumptions. D

We now state a quantum version of Theorem V.4.6.

Theorem VII.2.2. When q is not a root of unity, any finite-dimensional

Uq-module is semisimple.
PROOF. We follow the proof of Theorem V.4.6 step by step. Recall that it
is enough to prove that if V is any finite-dimensional Uq-module and V'is
any submodule of V, then there exists another submodule V" such that V
is isomorphic to the direct sum V' E8 V" as a module.
1. We shall first prove the existence of such a submodule V" in the
case when V'is of co dimension one in V. We proceed by induction on the
dimension of V'.
If dim(V') = 0, we may take V" = V. If dim(V') = 1, then necessarily
V' and V/V ' are simple one-dimensional mo(iules of respective weights cl
and c2. If the weights c 1 and c2 differ, there exists a basis {VI' v 2 } of V in
which K acts diagonally. Since EVi is an eigenvector for K with eigenvalue
ciq2 i=- Cj' we must have EVi = 0 for i = 1,2. Similarly, F acts trivially on
V. Hence, the module V is the direct sum of the submodules V' = kVl and
V" = kv 2 .
Otherwise, there exists a basis {VI' v 2 } with V' = kVl such that we have
K VI = cv 1 and K V2 = cV 2 + (xVI. Again, EVI is an eigenvector for K with
 VII.2 Semisimplicity 145

eigenvalue cq2 -I=- c, hence it is zero. Let us prove that EV2 is zero too.
Indeed, writing EV2 = >'v 1 + /-w 2, we have

c>'v 1 +/-l(cv2+av 1) = KEv2 = q2 EKv2 = q2 E(Cv 2+av 1) = cq2(>.v 1+/-lV2) ,

which implies /-lc(q2 - 1) = 0 and >.c(q2 - 1) = /-la. Thus, >. = /-l = O. One
proves in a similar way that F acts as 0 on V. Since [E, F] acts as 0, we
have K = K- 1 on V. In particular, since K- 1 V2 = cV 2 - av 1 , we have
a = -a, hence a = O. In this situation K is also diagonalizable and we
reach the same conclusion as before.
We now assume that dim (V') = p > 1 and that the assertion to be
proved holds in dimension < p. There is the following alternative: either
V'is simple, or it is not.
La. If V'is not simple, one uses the same argument as in Part La of the
proof of Theorem V.4.6.
Lb. Suppose now that the submodule V'is simple of dimension> 1. The
one-dimensional quotient module V/V' has weight c = 1. Let us consider
the operator C of Lemma 2.1; it acts by 0 on V/V'. Consequently, we have
CV c V'. On the other hand, C acts on V' as multiplication by a scalar
a -I=- O. It follows that Cia is the identity on V'. Therefore the map Cia is
a projector of V onto V'. This projector is Uq-linear since C is central. By
Proposition 1.1.3, the submodule V" = Ker (C / a) meets the requirements.
2. General case. We are now given finite-dimensional modules V' c V
without any restriction on the codimension. We shall reduce to the codimen-
sion-one case by considering vector spaces W' C W defined as follows: W
[resp. W'] is the subspace of all linear maps from V to V' whose restriction
to V'is a homothety [resp. is zero]. It is clear that W' is of co dimension one
in W. In order to reduce to Part 1, we have to equip Wand W' with Uq -
module structures. We give Hom(V, V') the Uq-module structure defined
in II1.5. Let us check that Wand W' are submodules of Hom(V, V'). For
fEW, let a be the scalar such that f (v) = av for all v E V'; then for all
x E Uq and v E V', we have

(xf)(v) = 2: x'f(S(x")v) = a (2: X'S(X"))V = ac(x)v.

(xl (xl

A similar argument proves that W' is a submodule too. Applying Part 1,

we get a one-dimensional submodule W" such that W ~ W' E9 W". Let
f be a generator of W". By definition, it acts on V' as a scalar a -I=- O. It
follows that f /a is a projector of V onto V' and that V" = Ker (I) is a
supplementary subspace of V'. To conclude, it suffices to check that V" is
a Uq-submodule of V. Now, since W" is a one-dimensional submodule, it
is simple of weight 1. Therefore, for all x E Uq we have xf = c(x)f. In
particular, if v belongs to V", we have

K- 1 f(Kv) = (K- 1f)(v) = c(K-1)f(v) = 0,

146 Chapter VII. A Hopf Algebra Structure on Uq (s((2))

which implies f(K v) = O. This proves that KV" c V". Similarly, V" is
stable under K- 1 . On the other hand, we have for v, hence for K v in V",

o c(E)f(Kv) = (Ef)(Kv
f(S(E)Kv) + Ef(K- 1 Kv) = - f(Ev) + Ef(v).
Consequently, f(Ev) = 0, which implies that V" is stable under the action
of E. A similar computation shows that FV" c V". The subspace V" is
therefore a submodule. D

VII.3 Action of Uq (s((2)) on the Quantum Plane

This section is the quantum version of V.6. We start with a few generalities
on skew-derivations of an algebra A. For a E A, denote by a [resp. aT] the
left [resp. right] multiplication by the element a. If cr is an automorphism
of the algebra A, we have

(3.1)

Given two automorphisms cr and T of an algebra A, a linear endomor-

phism 8 of A is called a (cr, T) -derivation if

8(aa') = cr(a)8(a') + 8(a)T(a') (3.2)

for all a, a' in A. Relation (3.2) is equivalent to

(3.3)

or to
(3.4)
It is well-known that, if 8 is a derivation of a commutative algebra, then
aR8 is a derivation too. In a non-commutative situation, this is no longer
the case. Nevertheless, the following assertion holds.
Lemma VII.3.1. Let 8 be a (cr, T)-derivation of A and a be an element of
A. If there exist algebra automorphisms cr' and T' of A such that

then the linear endomorphism aR8 is a (cr', T)-derivation and aT8 is a (cr, T')-
derivation.
PROOF. This follows from straightforward computations. D

We now return to the quantum plane A = kq[x, y] ofIV.I. Let us consider

its algebra automorphisms cr x and cry defined by
 VII.3 Action of Uq (s[(2)) on the Quantum Plane 147

When q = 1, we have ax = a y = id. We define q-analogues Oq/ox and

aq / oy of the classical partial derivatives by

aq (xmyn)
= [m 1 x m-l y n and (3.6)
ax
for all m, n 2: O. Let us describe all commutation relations between the
endomorphisms xc' Xr , YRl Yr' ax' ay' Oq/ox, Oq/oy. We say that a commu-
tation relation between two endomorphisms u and v is trivial if uv = vu.

Proposition VII.3.2. (a) Within the algebra of linear endomorphisms of

kq[x, y], all commutation relations between the above six endomorphisms
are trivial, except the following ones:

ayYe,r = qYe,ray'
Oq Oq
oY a y = qa y oY ,

Oq Oq
oY xr = qXr oY ,

We also have

(b) The endomorphism Z~ is a (a;;;lay,ax)-derivation and, similarly, Z~

is a (a y , a xa:;/) -derivation.

PROOF. (a) This part results from easy, but fastidious computations.
(b) First observe that, if Relation (3.3) holds for two elements a, a' of A,
then it holds for their product aa'. Indeed, we have

8aa~
a(a)8a~ + 8(a)Ta~
a(a)a(a')8 + a(a)8(a')T + 8(a)T(a')T
a(aa')8 + 8(aa')T.
148 Chapter VII. A Hopf Algebra Structure on Uq (s[(2))

We are reduced to checking Relation (3.3) for Oq/ox and Oq/oy in the case
when a = x and a = y. For Oq/ox we have

-1 Oq (OqX) -1 Oq Oq
(0' x 0' y)( X) C ox + ox / x = q Xc ox + 0' x = ox Xf

by Proposition 3.2(a). We also have

-1 Oq (Oqy) Oq Oq
(O'x O'y)(Y)c ox + ox fO'x = qypox = ox Yp .

Similar computations can be carried out for Oq/oy. o

We now show how the "quantum partial derivatives" ~~ and ~~ endow

the quantum plane with the structure of a module-algebra (as defined in
V.6.1) over the Hopf algebra Uq .

Theorem VII.3.3. For any P E kq[x, y], set

(3.7)
(a) Formulas (3.7) define the structure of a Uq-module-algebra on kq[x, y].
(b) The subspace kq[x,Yln of homogeneous elements of degree n is a Uq -
submodule of the quantum plane. It is generated by the highest weight vector
xn and is isomorphic to the simple module VLn .

Theorem 3.3 is the quantum version of Theorem V.6.4. It shows that the
quantum plane contains all finite-dimensional simple Uq-modules.
PROOF. (a) We first show that the formulas (3.7) equip kq[x, yl with a
Uq-module structure. In other words, we have to check Relations (VI.l.lO-
l.12). We use Proposition 3.2.
Relation (l.10) is trivially verified. For Relation (l.11) we have

One proves KFK- 1 = q- 2 F in a similar fashion. As for (l.12), we have

 VII.3 Action of Uq (,5[(2)) on the Quantum Plane 149

Oq Oq Oq Oq
[E,F] XfoyYrOx -YrOxXeoy
-1 Oq Oq Oq -1 Oq Oq Oq
q XeYr oyox +XeUyOX -q YrXCoxoy -YrUx oy
Oq Oq
XCUy Ox - YrUx OY
Uy(U x - u;l) - Ux(U y - u;l)
q _ q-1
UxU y-1 - UyU x-1
q _ q-1
K-K-1
q _ q-1 .

We now prove that the quantum plane is a Uq-algebra. By Lemma V.6.2,

it is enough to check that for any U E Uq , we have

ul = c(u)l, (3.8)

and
K(PQ) = K(P)K(Q), (3.9)
E(PQ) = + E(P)K(Q),
PE(Q) (3.10)
F(PQ) = K- 1(P)F(Q) + F(P)Q, (3.11)
for any pair (P, Q) of elements of the quantum plane. Relation (3.8) fol-
lows easily from (3.5-3.7) and Relation (3.9) from the fact that K acts as
an algebra automorphism. By Lemma 3.1 and by Proposition 3.2(b), the
endomorphism Xc ~~ is a (id, u x u;l )-derivation and Yr ~~ is a (u;lu y , id)-
derivation, which implies Relations (3.10-3.11).
(b) We have Exn = 0, Kxn = qnxn, and

1 P( n) _ -P [n]! n-p p
lP]! F x - q [p]![n _ p]! x y.

Consequently, xn is a highest weight vector of weight qn and generates the

submodule kq[x, Y]n' D

Observe that [E, F] acts on the quantum plane as the operator

Its "limit when q tends to I" is the operator xO / Ox - yO / oY by which the

element H of ,5[(2) acts on the affine plane (see Theorem V.6.4).
150 Chapter VII. A Hopf Algebra Structure on Uq (s((2))

VIl.4 Duality between the Hopf Algebras

Uq(s((2)) and SLq(2)
We now relate this chapter to Chapter IV by showing that Uq is in duality
with the Hopf algebra SLq(2) defined in IV.6. We use the concept of duality
introduced in V.7.
As in V.7, our first task is to construct an algebra morphism 1jJ from the
algebra Mq(2) (defined in IV.3) into the dual algebra U;. We shall deduce
a bilinear form on Uq x Mq (2) defined by < u, x > = 1jJ( x) (u) and satisfying
Relations (V.7.2) and (V.7.4). Giving the morphism 1jJ is equivalent to
giving four elements A, B, C, D in U; satisfying the six defining relations
of Mq(2) (see IV.3).
The definitions of A, B, C, D use the simple Uq-module V1,1 of highest
weight q and with basis {vo, Vl}' The matrix representations of the gener-
ators E, P and K in this basis have been given in VI.3. Setting p = Pl.l'
we have .

p(E) = (~ ~), p(P) = (~ ~), p(K) = (6 q~l)' (4.1)

More generally, for any element u of Uq , define

A(u) B(u) )
p(u) = ( C(u) D(u) . (4.2)

We thus get four linear forms on Uq , hence four elements A, B, C, D of U;.

Lemma VII.4.1. The quadruple (A,B,C,D) is a U;-point of Mq(2).

PROOF. This is done by a direct, but laborious checking. First, one has to
compute in U; the twelve products AB, BA, AC, CA, ... formed by all
pairs of distinct elements of the set {A, B, C, D}. Recall that the product
of any two elements x, y of U; is given by

(xy)(u) = L x(u')y(u"). (4.3)

(u)

It suffices to evaluate (xy)(u) on the basis {EipjKR} of Uq . Let us set

u = Ei pj K C When i > 2 or when j > 2, we see from Proposition 1.3 that
in the sum I:(u) u'@u", either u' or u" contains powers of E or of P with
exponents> 1. Now, by (4.1), p(Ei pj KR) = p(E)ip(p)j p(K)f vanishes
when i > 1 or j > 1. Consequently, if x, y E {A, B, C, D}, we have

whenever i > 2 or j > 2. It therefore remains to evaluate the products on

the elements Ei pj K C where 0 ::; i ::; 2 and 0 ::; j ::; 2.
 VII.4 Duality between the Hopf Algebras Uq (s[(2)) and SLq(2) 151

(i) If u = K R, we have 6.(KR) = KR KR and all products evaluated on

u vanish, except that

(AD)(u) = (DA)(u) = l. (4.4)

(ii) If u = FK, we have 6.(FKR) = KR-l FKR + FKE KR and all

products evaluated on u vanish, except that

(CA)(u) = q (AC)(u) = q2R and (DC)(u) = q (CD)(u) = q. (4.5)

(iii) If u = F2 K R, we have

and all products evaluated on u vanish.

(iv) If u = EKE, we have 6.(EKR) = EKR KHl + KR EKR and all
products evaluated on u vanish, except that

(BA)(u) = q (AB)(u) = q and (DB)(u) = q (BD)(u) = q-2. (4.6)

(v) If u = EFK R, we have

6.(EFK) = K-l EFK + EFK KHl

+ FK EKR + q-2 EK c- 1 FKHl
and all products evaluated on u vanish, except that

(BC)(u) = (CB)(u) = 1, (DA)(u) = q, and (AD)(u) = q-l. (4.7)

(vi) Ifu = EF2K, we have

6.(EF2 KR) = 00 2 (F K C- 1 EF KR + q-2 EF K R- 1 F KH1)
+ (terms of degree > 2 in F)
and all products evaluated on u vanish, except

(4.8)

(vii) If u = E2 K C, we have

6.(E2 KR) = 00 3 EKR EKHl + (terms of degree > 2 in E)

and all products evaluated on u vanish.

(viii) If u = E2 FK R, we have

6.(E2 F KR) = 00 4 (EF K EKHl + q-2 EK R- 1 EF KH1)

+ (terms of degree > 2 in E)
152 Chapter VII. A Hopf Algebra Structure on Uq (s[(2))

and all products evaluated on u vanish, except

(BA)(u) = q (AB)(u) = 0<4. (4.9)

(ix) If u = E2 F2 K R, we have

/:l(E2 F2 Kf) = 0<5 EF K R- l EFKf+l + (terms of degree > 2 in E and F)

and all products evaluated on u vanish.

In Cases (iii) and (vi-ix) we denoted by 0<1' 0<2' 0<3' 0<4' and 0<5 scalars
that are well-defined, but about which we need not be explicit. From this
case-by-case analysis, it is easy to check that A, B, G, D satisfy the six
defining relations of Mq(2). As a sample calculation, we check the most
involved relation, namely

DA - AD = (q - q-l) BG.

From the above observations, we see that it is enough to perform the check-
ing for u = Kf, which is trivial, and for u = EF KR. In the latter case, (4.7)
implies
(DA - AD)(u) = q - q-l = (q _ q-l) (BG)(u).
D
As a consequence of Lemma 4.1 and of IV.3, there exists a unique mor-
phism of algebras 'Ij; from Mq(2) into U; such that

'Ij;(a) = A, 'Ij;(b) = B, 'Ij;(c) = G, 'Ij;(d) = D.

Proposition VII.4.2. The bilinear form < U,x >= 'Ij;(x)(u) realizes a
duality between the bialgebras Uq and Mq(2).
PROOF. The comultiplication and the counit of Mq(2) being the same as
those of M(2), the proof follows along the same lines as in the proof of
Proposition V.7.3. D

The duality between Mq(2) and Uq is not perfect, just as in the classical
case.

Lemma VII.4.3. For the quantum determinant detq = da-qbc of Mq(2),

we have 'Ij;( det q) = l.
Equivalently, < u, detq > = c:( u) for all elements u of Uq
PROOF. By Theorem IV.5.1, the element detq is grouplike, i.e., we have
/:l(detq) = detq det q. It results that the map u f--+< u,det q > is a
morphism of algebras from Uq to k. To show that this morphism coincides
with the counit c:, it suffices to check that both maps take the same values
 VIl.4 Duality between the Hopf Algebras Uq(.sl(2)) and SLq(2) 153

on the generators E, F, K and K- 1 . Using (V.7.2-7.3) and (1.1), we get

for E:

< E,det q >

< E, da > -q < E, be>
s(d) < E,a > + < E,d >< K,a >
-qs(b) < E,c > -q < E,b >< K,c >
o = s(E).
For K we have

< K,det q > < K, da > -q < K, be >

< K, d >< K, a > -q < K, b >< K, c >
q-lq = 1 = c(K).
Similar computations can be carried out for F and K- 1 . o
As a consequence of Lemma 4.3, the algebra morphism 't/J from Mq(2)
to U; factors through SLq(2) = Mq(2)j(detq - 1). We still denote by 't/J
the induced morphism of algebras from SLq(2) into U; and by < , > the
corresponding bilinear form.

Theorem VII.4.4. The bilinear form < u, x> = 't/J(x)(u) realizes a dual-
ity between the Hopf algebras Uq and SLq(2).

PROOF. We use the same argument as in the proof of Theorem V.7.6. The
only difference lies with the antipodes. We first check Relation (V.7.5) for
the generators. Using the condensed matrix form, we have

< S(E), (~ ~) > = p(S(E)) = -p(E)p(K-l) = (~ ~q)

d -qb ) (s(a) S(b))
= < E, ( _q-1c a > = < E, S(c) S(d) >.

For F we have

< S(F), (~ ~) > = p(S(F)) = -p(K)p(F) = (_~-l ~)

d -qb ) (S(a) S(b))
= < F, ( _q-1c a > = < F, S(c) S(d) > .

One proceeds with K and K- 1 similarly. To conclude, one appeals to

Lemma V.7.7. 0
154 Chapter VII. A Hopf Algebra Structure on Uq (.s((2))

VII.5 Duality between Uq (.s[(2) )-Modules and

SL q (2)-Comodules
Exactly as in the classical case considered in V.7, there is a duality between
Uq-modules and SL q(2)-comodules. We have seen in IV.7 that the vector
space kq[x, Y]n of homogeneous elements of degree n of the quantum plane
has a natural structure as an SL q(2)-comodule. By duality, the dual vector
space kq [x, y]~ has a module structure over the algebra SLq(2)*, hence
over the algebra Uq via the morphism c.p : Uq ---+ SLq(2)*. The following
statement gives the structure of kq[x, y]~ as a Uq-module.
Theorem VII.5.l. The Uq-module kq[x, y]~ is isomorphic to the simple
module VI,n of highest weight qn.
Thus, the SL q(2)-comodule kq[x, Y]n corresponds by duality to the Uq-
module VI,n'
PROOF. We shall show that the linear form on kq[x,Y]n defined by
f(xiy n - i ) = bni
is a highest weight vector, with weight qn, of the Uq-module kq[x, Y];" which
implies that kq[x, y]~ contains a sub module isomorphic to the simple mod-
ule VI,n' Since
dim(VI,n) = n + 1 = dim(kq[x, y]~),
we get kq[x, y]~ ~ VI,n'
In order to prove that f is a highest weight vector, we need the relation
(5.1)
for all u E Uq and for all i such that 0 :::; i :::; n. But this is so since,
by definition of f, by III.6, Example 2, by Lemma IV.7.2, and using the
abbreviation
n-i
s
to shorten the formulas, we have

1, n-i

r=Os=O
i n-i

r=Os=O
i n-i

LL
r=Os=O
Cr,s
 VII.6 Scalar Products on Uq (s[(2))-Modules 155

Let us apply Relation (5.1) to K. A straightforward computation yields

< K,aid >=< K,a >i< K,c >j = Djoqi.
Consequently, we have (K 1) (xiyn-i) = Dni qi = Dni qn, which implies that
Kf = qnf
It remains to prove that Ef = O. This is a consequence of Relation (5.1)
applied to E and of the fact that < E, aid> = 0 for all i and j. Let us
prove the latter. First, we have < E,l > = c(E) = O. Next, if i > 0 we
have by (V.7.2-7.3)

< E,a i > c(a)< E,a i - 1 > + < E,a >< K,a i - 1 >
< E, a i- 1 > = ... = < E, a > = O.
Similarly, if j > 0 we get
< E,d >= c(c) < E,d- 1 > + < E,c >< K,d- 1 >= O.
Consequently,
< E,aid > = c(a)i < E,d > + < E,a i >< K,d > = O.
o

VII.6 Scalar Products on Uq (sr(2))-Modules

In this section, given any finite-dimensional Uq-module V, we construct a
scalar product, i.e., a non-degenerate symmetric bilinear form ( , ) on V
such that
(XV,V') = (v,T(x)v' ) (6.1)
for all x E Uq and v, v' E V. The linear map T is the algebra antiautomor-
phism of Uq defined as follows.
Proposition VII.6.!. There exists a unique algebra antiautomorphism T
of Uq such that T(E) = K F, T(F) = EK- 1 , and T(K) = K. The auto-
morphism T is also a morphism of coalgebras.
PROOF. Left to the reader. o
By Theorem 2.2, it is enough to construct a scalar product on any simple
Uq-module of the form Vc,n' This is done in the following theorem.
Theorem VII.6.2. On the simple Uq-module ~,n generated by the highest
weight vector v, there exists a unique scalar product such that (v, v) = 1. If
we define the vectors Vi for all i 2: 0 by Vi = Fiv/[i]!, then they are pairwise
orthogonal and we have

(Vi,v i ) -_ q-(n-i-l)i [ ni ] .
156 Chapter VII. A Hopf Algebra Structure on Uq (.sl(2))

PROOF. Let us first assume that there exists a scalar product on VE n such
that (v, v) = 1. Let us show that (vi' Vj ) is necessarily of the pre~cribed
form. By definition and by (6.1) we have
1 1 . 1 1
(Vi' Vj ) = [ill (Pv, vj ) = [ill (v, T(F)'vj ) = [ill (v, (EK- )'vj ).

An easy induction on i shows that (EK-1)i = qi(H1) K- i Ei for any i > O.

Consequently, the vector T(F)i Vj is a scalar multiple of Eiv j which vanishes
as soon as i > j. Therefore (Vi' Vj ) = 0 if i > j. By symmetry, we also have
(vi,vj)=Oifi<j.
We need the formula
Ei _ i [n - j + ill
Vj - e [n _ j]! vj _i

to compute (Vi' Vi)' We have

~qi(H1) (v K- i Eiv.)
[i]! ' ,
e i qi(H1) [n]! (v K-iv)
[i]![n - ill '
qi(H1)-ni [ 7](v, v).
This proves the uniqueness of the scalar product. Let us now prove its
existence.
Clearly, there exists a non-degenerate symmetric bilinear form such that

.
( v" v).) =
q-(n-i-1)i [ ni ] J: ..
U,). (6.2)

We have to check that it satisfies Relation (6.1). It is enough to check this

for x = E, F, K and K- 1 . We shall do this for x = E leaving all other
computations to the reader. On the one hand, we have

(Ev. v.) = e[n - i

" )
+ 1](v.
,-1')
v.) = co.,-I,).q-(n-i)(i-1) [n]!
[i-l]![n-i]!'
On the other hand, by (VI.3.1-3.3) and by (6.2), we have
(vi,T(E)vj) (vi,KFvj)
eqn-2(j+1)[j + 1] (vi,vj+l)
eO.. q-(n-i-1)Hn-2(j+l) [j + 1] [n]!
',)+1 [i]![n - ill
o
c i,j+1 q
-(n-i)(i-1) [n]!
[i _ 1]![n _ ill -
- (E )
vi' Vj .
o
 VII.7 Quantum Clebsch-Gordan 157

VII.7 Quantum Clebsch-Gordan

We now prove a quantum Clebsch-Gordan formula for the finite-dimensional
simple Uq-modules. Since

(7.1)

we need give this formula only for the modules V1,n, henceforth denoted
for simplicity by Vn .

Theorem VII.7.1. Let n ::::: m be two nonnegative integers. There exists

an isomorphism of Uq-modules

One proves Theorem 7.1 in the same way as Proposition V.5.l. It suffices
to check that the module Vn 129 V m contains a highest weight vector of weight
qn+m-2 p for any integer p such that 0 :::; p :::; m.

Lemma VII.7.2. Let v(n) be a highest weight vector of weight qn in Vn

and v(m) be a highest weight vector of weight qm in V m . Let us define
p = ...l.FPv(n)
v(n) [p]! and v(m)
p = ...l.FPv(m)
[p]! for all p >
J' _
O. Then ,

v(n+m-2p) = '""'
p
(_l)i [-
m p + ~']'[. n -
~.
']'
q-i(m-2 p+i+1) v(n) 129 v(m)
~ [m - p]![n]! 'p-'
,=0

is a highest weight vector of weight qn+m-2 p in Vn 129 Vm

PROOF. It is clear that v~n) 129 v~~~ has weight qn-2i+m-2(p-i) = qn+m-2P.
Let us prove that Ev(n+m-2p) = O. Recall that .6.(E) = 1129 E + E 129 K. It
follows that
Ev(n+m-2p)

~ (_l)i [m - p + i]![n - ill -i(m-2p+i+l) (n) 129 E (m)

~ [m _ p]![n]! q v, v p_,
,=0
+ ~ (_l)i [m - p + i]![n - ill q-i(m-2p+i+ 1) Ev(n) 129 Kv(m)
~ [m - p]![n]! 'p-'
,=0
'""'
p
(_l)i [m _ + i + 1] [ p+
m '- ] ~ .n
'[
- ~']'. q-i(m-2 p+i+l)
~ p [m-p]![n]!
,=0
xv(n) to. v(m)
, VY p-,-l

+ ~ (_l)i [n _ i + 1] [m - p + i]![n - ill q-i(m-2 p+i+l)+(m-2 p+2i)

~ [m - p]![n]!
,=0
158 Chapter VII. A Hopf Algebra Structure on Uq (.sl(2))

xv(n) Q9v(m)
,-1 p-'

~ (_I)i ([m - p + i]![n - i + I]! q-(i-1)(m-2p+i)

L..-
,=0
[m - p]![n]!
[m - p + i]![n -i + I]! -(i-1)(m-2P+i)) (n) Q9 (m)
- [
m_ p.n.
]'[ ]' q Vi - 1 vp_ i
o.
o
This concludes the proof of Theorem 7.1. We wish to go one step further
and address the following problem. We now have two bases of Vn Q9 Vm at
our disposal. They are of different natures: the first one, adapted to the
tensor product, is the set

{ (n) (m)} .
Vi Q9 Vj O~i~n, O~j~m ,

the second one, formed by the vectors

(n+m-2p) _ ~ Fk (n+m-2p)
Vk - [k]! v

with 0 :S p :S m and 0 :S k :S n + m - 2p, is better adapted to the Uq -

module structure. Comparing both bases leads us to the so-called quantum
Clebsch- Gordan coefficients

n m n+m- 2p ]
[
i j k

defined for 0 :S p :S m and 0 :S k :S n +m - 2p by

vin+m - 2P ) = L [~n: n +~- 2p ] v~n) Q9 vJ m ). (7.2)

O~i~n, O~j~m Z J

The remainder of this section is devoted to a few properties of these coef-

ficients, also called quantum 3j-symbols in the physics literature.

Lemma VII.7.3. Fix p and k. The vector vin+m - 2P ) is a linear combina-

tion of vectors of the form v~n) Q9 v~:I+k' Therefore, we have

[7 ; n +~- 2p ] =0 (7.3)

when i + j #- p + k. We also have the induction relation

[
n m n + m - 2p ] = [j + I]q-(n-2i) + [i] [n m n+m- 2P ]
i j +1 k+1 [k + 1] i j k .
(7.4)
 VII.7 Quantum Clebsch-Gordan 159

PROOF. This goes by induction on k. The assertion holds for k = 0 thanks

to Lemma 7.2. Supposing
(n+m~2p) _ ""' (n) rcA (m)
vk - ~ (Yi vi YY Vp~i+k'

we get
[k + 1]Vk~~m~2p) = FVkn+m~2p)

= ""' . (K~I v,(n)

~ (Y,
M
'01
F Vp~'+k
(m) + F v,(n) M
'C)
(m)
Vp~'+k
)

= ""'
~
(Y.
, ([p - i + k + 1]q~(n~2i)v(n)
, v(m) p~,+k+1
i

+ [~. + 1] Vi+1
(n) (m))
Vp~i+k

= ""'
~ ,
(Y. ([p - i + k + 1]q~(n~2i) + [i])v(n)
, v(m) p~,+k+I'
i

The rest follows easily. o

We now prove some orthogonality relations for the quantum Clebsch-
Gordan coefficients, which will allow us to express the basis {v}n) vjm)L,j
in terms ofthe basis {vkn+m~2p)}p,k' Let us equip v;., and Vm with the scalar
product ( , ) defined in Section 6. Consider the symmetric bilinear form on
Vn Vm given by
(VI V~'V2 v;) = (vI,V2)(V~,V;) (7.5)
where v I ,v2 E Vn and v~,v~ E Vm
Lemma VII.7.4. The symmetric bilinear form (7.5) is non-degenerate
and the basis {v;n) vJ rn ) h,j is orthogonal. Furthermore, for all x E Uq
and all WI' W 2 E Vn Vrn , we have

PROOF. The first two assertions are clear. Let us prove the last one. If
WI = VI v~ and W 2 = V 2 v~, we have
(~(X)(VI v~), V2 v~)

L (x' VI' V2) (x" V~, V~)

(x)
L (VI> T(x')v 2)( V~, T(x")v~)
(x)
L (vI,T(x)'V2)(V~,T(x)"v;)
(T(x))

(w l ,T(x)w2),
160 Chapter VII. A Hopf Algebra Structure on Uq (.s[(2))

using the fact that T is an automorphism of coalgebras (see Proposition

6.1). 0

The second basis of Vn Q9 Vm is orthogonal too.

. { (n+m-2 p )}
Proposition VII.7.5. () aTe h baszs vk O~p~m, O~k~n+m-2p ZS
orthogonal.
(b) Fix integers p, q, k, R. We have the following orthogonality relations:

o L q-i(n-i-l)-j(m-j-l) [ 7] [7]
',J
X [7 ; n +~- 2p ] [7 7 n +~- 2q ]

when p i= q or k i= R, and

L q-i(n-i-l)-j(m-j-l) m
j
n +m-
k
2p ]2
i,j

= q-k(n+m-2 p -k-l) [ n +~- 2p ] .

(c) Given i and j, we have

(n)
vi
(m) _
Q9 Vj - q
-i(n-i-l)-j(m-j-l) [ ~." ] [mJ. ]

m n+m-2p [n n: n+ ~ - 2p ]
x L L qk(n+m-2 p -k-l) Z J vin + m - 2P ).

p=O k=O [ n +~- 2p ]

PROOF. (a) Arguing as in the proof of Theorem 6.2, one shows that
( V (n+m-2p) ,V(n+m-2 P)) -- 0
k

whenever k i= R. Let us examine the case when p i= q. Let us first show

that the highest weight vectors v(n+m-2p) and v(n+m-2q) are orthogonal.
In fact, Lemma 7.2 implies that (v(n+m-2 p ),v(n+m-2 q )) can be written
""'" j3 ((n) 'Vj(n))( Vp_i,V
~ oc i j Vi
(m) (m))
q_ j
i,j

which is zero because p - i i= q - i. It remains to show that

( v (n+m-2p) (n+m-2 q )) - 0
k ,~ -

when k,R > O.

 VII.7 Quantum Clebsch-Gordan 161

By symmetry, it is enough to consider the case k 2: f. We have

( v (n+rn-2p) (n+rn-2 q )) ,(pk V (n+rn-2 p ), v~n+m-2q))
k ,vc
,'(v(n+m-2 p ), Ekv~n+m-2q))

for some scalars, and ,'. Now, if k 2: f, the vector Ekv~n+rn-2q) is zero or
is a scalar multiple of the highest weight vector v(n+m-2 q ), which brings us
back to a previous case.
(b) Let us compute (vkn+rn-2p),v~n+rn-2q)). It is equal to

m
J
n+m - 2p
k ][~ m
s
n+m - 2q
f

x (v(n) v(n))(v(m) v(m))

~ 'r .J''''

L n ; n + ~ - 2p ] [ 7 mj n+m - 2q
f
H.i=p+k

x (v~n), v.}n)) (v;m) ,vjm))

q-2(n-1-1)-J(rn-J -l) [ n ] [ ; ] [ n m n +m - 2p ]
j k
Hj=p+k

n m n +m - 2q ]
x [ i J f .
On the other hand, we have

( (n+rn-2 p )
Vk
(n+m-2 q )) _ " s:
,vR - upqukR q
-k(n+m-2p-k-l) [ n + mk
- 2p ]

(c) We have v(n)

1
vern)
J
= ,\,m_ ,\,n~m-2p 'Y v(n+m-2 p )
L.-p_O L.-k_O Ipk k
for some coef-
ficients 'pk. Therefore,
( , (.n+rn-2P) (n+rn-2 P))
'pk uk ,vk

( Vi(n) rv. (rn)

vy Vj ,vk
(n+m-2 P))

[ ni m n+m-2P ] ((17.) (n))( (m) (rn))

j k v 2 ,v, V] ,vJ .

Applying (6.2), one gets the desired explicit expression for 'pk. o
For more details on the quantum Clebsch-Gordan coefficients, see [KR89]
[KK89] [Vak89] where they are expressed in terms of q-Hahn polynomi-
als, i.e., of certain orthogonal q-hypergeometric series (see also [GR90],
Chap, 7). Koelink-Koornwinder and Vaksman showed that the orthogonal-
ity relations of the q- Hahn polynomials were equivalent to the orthogonality
relations of the quantum Clebsch-Gordan coefficients. The corresponding
property for the classical Clebsch-Gordan coefficients was known already
(see [Koo90]).
162 Chapter VII. A Hopf Algebra Structure on Uq (s((2))

VII.8 Exercises
1. Compute S(EiFjKR) in Uq .

2. Let x be an element of Uq . Prove successively that

(a) x is group like if and only if x is of the form x = Kn;

(b) if 6.(x) = 1 @ x + x @ K and c(x) = 0, then x is a linear
combination of E and of K F;
(c) if 6.(x) = K- I @ X + X @ 1 and c(x) = 0, then x is a linear
combination of F and of EK- I ;
(d) if6.(x)=l@x+x@K-l,thenx=O.

3. Use Exercise 2 to show that there exists an isomorphism of Hopf

algebras from Uq onto Uq , if and only if q' = ql, and that any
Hopf algebra automorphism r.p of Uq is of the form

r.p(E) = aE, r.p(F) = a-I F, r.p(K) = K

where a is a non-zero scalar.

4. (Hop! *-algebra structures on Uq ) We use the concepts introduced in

IV.S.

(a) Prove that Uq is a Hopf *-algebra if and only if q2 is a real

number or q is a complex number of modulus 1.
(b) Check that the following formulas determine five Hopf *-algebra
structures on Uq :
(i)E* = E, F* = F, and K* = K if Iql = 1;
(ii)E* = KF, F* = EK- 1 , and K* = Kif q is real> 0;
(iii)E* = -KF, F* = -EK- 1 , and K* = K if q is real < 0;
(iv) E* = iKF, F* = iEK-l, and K* = K if q = Ai with A
real> 0;
(v) E* = -iKF, F* = -iEK- 1 , and K* = K if q = Ai with A
real < O.
(c) Show that any Hopf *-algebra structure on Uq is equivalent to
one of the previous five ones (Hint: use Exercise 2).

5. Given a Hopf *-algebra structure on Uq and a Uq-module V, define a

Hermitian scalar product as a definite positive Hermitian form ( , )
such that (xv,v') = (v,x*v') for all x E Uq and v,v' E V. Determine
all Hermitian scalar products on the simple module Vo,n'

6. Prove that there exists a Uq-linear isomorphism between the simple

module Vc,n and its dual module.
 VII.9 Notes 163

VII.9 Notes
The Hopf algebra structure of Uq (s((2)) is due to Sklyanin [Sk185]. The
Drinfeld-Jimbo algebras Uq(g) also have a non-commutative, non-cocom-
mutative Hopf algebra structure. In the cases A, D, E considered in VI. 7,
it is given on the generators (Ei' Fi , Kih~i~R by

D.(Ei) = lEi +EKi , D.(Fi) = K i- 1 Fi +Fi 1, D.(Ki) = KiKi'

C(Ei) = c(Fi) = 0, c(Ki) = 1,
and
S(Ei) = -EiK;-l, S(FJ = -KiFi' S(Ki) = K i- 1 .
In this chapter we adopted the conventions of Takeuchi [Tak92c] rather
than those of Drinfeld and Jimbo. In the special case g = s((2), Takeuchi's
conventions allow Uq to act on the quantum plane of Chapter IV. Following
Drinfeld [Dri87J, Takeuchi [Tak92c] [Tak92b] also showed the existence of a
duality between Uq(s((n)) and the Hopf algebra SLq(n) ofIV.9, embedding
the latter into the restricted dual of Uq(s((n)).
The semisimplicity of the finite-dimensional Uq-modules is due to Rosso
[Ros88]. We followed his proof closely.
For more details on quantum Clebsch-Gordan coefficients, read [KR89]
[KK89] [Koo90] [Vak89]. For the Hopf *-algebra structures on Uq (deter-
mined in Exercise 4), see [MMN+90].
 Part Two

Universal R-Matrices
Chapter VIII
The Yang-Baxter Equation and
(Co )Braided Bialgebras

Part II is centered around the now famous Yang-Baxter equation whose so-
lutions are the so-called R-matrices. We introduce the concept of braided
bialgebras due to Drinfeld. These are bialgebras with a universal R-matrix
inducing a solution of the Yang-Baxter equation on any of their mod-
ules. This provides a systematic method to produce solutions of the Yang-
Baxter equation. There is a dual notion of cobraided bialgebras. We show
how to construct a cobraided bialgebra out of any solution of the Yang-
Baxter equation by a method due to Faddeev, Reshetikhin and Takhtadjian
[RTF89]. We conclude this chapter by proving that the quantum groups
GLq(2) and SLq(2) of Chapter IV can be obtained by this method and
that they are cobraided.

VIlLI The Yang-Baxter Equation

Definition VIII.1.1. Let V be a vector space over a field k. A linear
automorphism c of V V is said to be an R-matrix if it is a solution of
the Yang-Baxter equation
(c idv)(id v c)(c id v ) = (id v c)(c id v ) (id v c)
that holds in the automorphism group of V V V.
Finding all solutions of the Yang-Baxter equation is a difficult task, as
will appear from the examples given below. Let {viL be a basis of the
vector space V. An automorphism c of V V is defined by the family
168 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

(C~f);,j,k' of scalars determined by

Then c is a solution of the Yang-Baxter equation if and only if for all

i, j, k, , m, n, we have

which is equivalent to

(1.1)

for all i, j, k, , m, n. Solving the non-linear equations (1.1) is a highly non-

trivial problem. Nevertheless, numerous solutions of the Yang-Baxter equa-
tion have been discovered in the 1980's. Let us list a few examples.
Example 1. For any vector space V we denote by TV,V E Aut(V 16> V) the
flip switching the two copies of V. It is defined by

for any VI' V 2 E V. The flip satisfies the Yang-Baxter equation because of
the Coxeter relation (12)(23)(12) = (23)(12)(23) in the symmetry group
53'
Here is a way to generate new R-matrices from old ones.

Lemma VIII.1.2. If c E Aut(V 16> V) is an R-matrix, then so are AC, c- 1

and TV,v 0 co TV,v where A is any non-zero scalar.

PROOF. This follows from the identities

(AC 16> id v ) = A(C 16> id v ), (id v 16> AC) = A(id v 16> c),

(c- 1 16> id v ) = (c 16> idv)-l, (id v 16> c- 1 ) = (id v 16> C)-I,

(c' 16> id v ) = cr(id v 16> c)cr- 1 , (id v 16> c') = cr(c 16> id v )cr- 1 ,
where c' = TV V 0 COTV v and cr is the automorphism of V 16> V 16> V defined
by cr( VI 16> v 2 v 3 ) = V; 16> v 2 16> VI for vI' V 2 , V3 E V. 0

Example 2. Let us solve the Yang-Baxter equation when V = VI = Vilis

the 2-dimensional simple module over the Hopf algebra Uq = Uq (s[(2)) ~on
sidered in Chapters VI-VII. More precisely, let us find all Uq-automorphisms
of VI 16> VI that are R-matrices. We freely use the notation of the above-
mentioned chapters. Recall that if va is a highest weight vector of VI' then
 VIII.1 The Yang-Baxter Equation 169

the set {vo, VI = Fv} is a basis of VI' By the Clebsch-Gordan Theorem

VII.7.1 we have VI Q9 VI ~ V2 E9 Vo. Lemma VII.7.2 implies that the vectors

Wo = Vo Q9 Vo and t = Vo Q9 VI - q-1V 1 Q9 Vo

are highest weight vectors of respective weights q2 and 1. We complete the

set of linearly independent vectors {wo, t} into a basis for V Q9 V by setting

1 2
and w2 = [2] F Wo = VI Q9 VI

where [2] = q + q-l.

Proposition VIII.1.3. Any Uq-linear automorphism cp of VI Q9 VI is dia-
gonalizable and of the form cp( Wi) = AWi (i = 0,1,2) and cp(t) = I.d where
A and p are non-zero scalars. The automorphism cp is an R-matrix if and
only if

PROOF. Since cp is Uq-linear, the image under cp of a highest weight vector

is a highest weight vector of the same weight. Now, Wo and t have different
weights (we still assume that q2 =F 1); therefore, there exist A and p such
that cp(w o) = AWo and cp(t) = pt.
As for the remaining basis vectors, we have

for i = 1,2. This completes the proof of the first assertion in Proposition
1.3.
The second assertion results from tedious computation. Let us give some
details. We first observe that the matrix <I> of cp with respect to the basis
{vo Q9 vo, Vo Q9 VI' VI Q9 Vo, VI Q9 vd is given by

~~ (
n
A 0 0
0 a 'Y
0 'Y (3
0 0 0

where
q-l A + qp qA+q-lp A-P
a= (3= , 'Y = [2]'
[2] [2]
The automorphisms cp Q9 id and id Q9 cp can be expressed, respectively, by
the 8 x 8-matrices <1>12 and <1>23 in the basis consisting of the elements
VoQ9VoQ9Vo, VO Q9VOQ9Vl' VOQ9Vl Q9Vo, VOQ9Vl Q9V 1, VI Q9VoQ9Vo, VI Q9VOQ9Vl'
VI Q9 VI Q9 vo, and VI Q9 VI Q9 VI of V Q9 V Q9 V where
170 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

A 0 0 0 0 0 0 0
0 A 0 0 0 0 0 0
0 0 a 0 'Y 0 0 0
0 0 0 a 0 'Y 0 0
<1>12 = 0 0 0 0 0 0
'Y /3
0 0 0 'Y 0 /3 0 0
0 0 0 0 0 0 A 0
0 0 0 0 0 0 0 A
and
A 0 0 0 0 0 0 0
0 a 'Y 0 0 0 0 0
0 'Y /3 0 0 0 0 0
0 0 0 A 0 0 0 0
<1>23 = 0 0 0 0 A 0 0 0
0 0 0 0 0 a 'Y 0
0 0 0 0 0 'Y /3 0
0 0 0 0 0 0 0 A
Now, <1>12<1>23<1>12 - <1>23<1>12<1>23

0 0 0 0 0 0 0 0
0 K -a/3'Y 0 0 0 0 0
0 -a/3'Y L 0 a/3'Y 0 0 0
0 0 0 -K 0 a/3'Y 0 0
0 0 a/3'Y 0 M 0 0 0
0 0 0 a/3'Y 0 -L a/3'Y 0
0 0 0 0 0 a/3'Y -M 0
0 0 0 0 0 0 0 0

where K = a((A - a)A - 'Y 2 ), L = a/3(a - (3) and M = /3b 2 + A(/3 - A)).
Suppose that we have proved that K, Land M are multiples of a/3'Y. Then

where W is a non-zero matrix. It follows that <I> is an R-matrix if and only

if a/3'Y = 0, which would complete the proof of Proposition 1.3.
It remains to show that K, Land M are multiples of a/3'Y. An easy
computation proves that

A - a = q'Y, A - /3 = q- 1'Y, q-1 A - 'Y = q-1 a , qA - 'Y = q/3

and /3 - a = (q - q-1),,(. Therefore,

K = a'Y(qA - 'Y) = q a/3'Y, L = -(q - q-1) a/3'Y

and M = /3'Yb - q-1 A) = _q-1 a /3'Y. 0

 VIIl.l The Yang-Baxter Equation 171

To sum up, the R-matrices of the Uq-module VI @ VI belong to the

following three types depending on a parameter A =I- 0:

l. If A = /1, <p is a homothety.

2. If qA + q-l/1 = 0, then

~~qA (y 0
q-I _ q
1
0
0
1
0
0
0
0
0
q-I
)
3. If q-I A + q/1 = 0,

U
then

~ ~q-'A
n
0 0
0 1
1 q _ q-I
0 0

It is clear that Cases 2 and 3 are equivalent within a change of basis after
exchanging q and q-I. As we shall see in the next example, the minimal
polynomial of <I> is of degree:::; 2.
Example 3. We now give an important class of R-matrices with quadratic
minimal polynomial. Such R-matrices will be used in Chapter XII to con-
struct isotopy invariants of links in R 3 .
Let V be a finite-dimensional vector space with a basis {e l , ... , eN}' For
two invertible scalars p,q and for any family {rij}l:S;i,j:S;N of scalars in k
such that r ii = q and rijr ji = P when i =I- j, we define an automorphism c
ofV@V by

c(ei @ ei )
if i < j
c(e i @ ej )
if i > j.
Proposition VIII.1.4. The automorphism c is a solution of the Yang-
Baxter equation. Moreover, we have

(c - qidv~w )(c + pq-I idv~w) = 0,

or, equivalently, c2 - (q - pq-I)C - pid v0V = O.
PROOF. (a) We first show that c is an R-matrix. In order to simplify the
proof, let us introduce the following notation. The symbol (ijk) will stand
for the vector e i @ ej @ ek' and [i > j] for the integer 1 if i > j and for 0
otherwise. Then c can be redefined as

c(ei @ ej ) = rjie j @ ei + [i > j],Bei @ ej

where ,B =q- pq-I.
172 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

An immediate computation yields

(c Q9 id)(id Q9 c)(c Q9 id)((ijk))

rjirkirkj(kji) + rjirki[j > k](3(jki)
+rkjrki[i > j](3(kij) + rkj[i > j][j > k](32(ikj)
+rji([j > i][i > k] + [i > j][j > k])(32(jik)

+ (rjirij[i > k](3 + [i > j][j > k](33) (ijk)

and

(id Q9 c)(c Q9 id)(id Q9 c)((ijk))

rjirkirkj(kji) + rjirki[j > k](3(jki)
+rkjrki[i > j](3(kij) + rji[i > k][j > k](32(jik)
+rkj ([i > k][k > j] + [i > j][j > k])(32(ikj)

+ (rjkrkj[i > k](3+ [i > j][j > k](33) (ijk).

We have to prove that these expressions are equal for all i, j, k. This is
clearly the case if i = j = k. If i, j, k are distinct indices, they are equal in
view of relations of the type

[i > j][i > k] = [i > j][j > k] + [i > k][k > j]

which express the fact that for distinct indices, we have i > j and i > k
if and only if i > j > k or i > k > j. If exactly two indices are equal, say
i = j -I- k, then the desired equality is equivalent to rTi = (3rii + p, which
holds since r ii = q and (3 = q - pq-l.
(b) One computes c2 - (3c - p id v0v on any vector of the form ei Q9 ej . If
i -I- j, one immediately obtains o. If i = j, one gets (q2 - (3q - p) (e i Q9 ei ),
which is zero because of the value given to (3. D

Consider the following two special cases:

(i) If p = q2 and r ij = q for all i,j, then c is a homothety.
(ii) Take p = 1 and r ij = 1 for i -I- j. Then c takes the form shown in
Case 3 of Example 2 when V is two-dimensional. Thus, Example 2 turns
out to be a special case of Example 3.

VIII. 2 Braided Bialgebras

The aim of this section is to define the concept of a braided bialgebra. The
importance of this concept comes from the fact proved in Section 3 that
braided bialgebras generate solutions of the Yang-Baxter equation.
 VIII.2 Braided Bialgebras 173

Definition VIII.2.1. Let (H, /-L, 'TI, 6., E) be a bialgebra. We call it quasi-
cocommutative if there exists an invertible element R of the algebra H H
such that for all x E H we have

(2.1)

Here 6. 0p = TH H 0 6. denotes the opposite coproduct on H. An element

R satisfying thi~ condition is called a universal R-matrix. It is part of the
data of a quasi-cocommutative bialgebra. Any cocommutative bialgebra is
quasi-cocommutative with universal R-matrix equal to R = 1 1. Thus
we can look upon a quasi-co commutative bialgebra as a bialgebra whose
non-co commutativity is controlled by its universal R-matrix.
If we set R = 2: i si t i , then Relation (2.1) can be expressed, for all
x E H, by
" " ' XII si x 't i = '""'
'~ ~ SiX I t iX II (2.2)
(x),i (x),i

using Sweedler's sigma notation introduced in IlL 1. We also define a quasi-

cocommutative Hopf algebra as a Hopf algebra whose underlying bialgebra
has a universal R-matrix.
Convention. The following notation will be used extensively in the sequel.
Let H be an algebra and X = 2: i X~1) ... x~p) E HQ9P (p> 1). For any
p-tuple (k 1 , ... , kp ) of distinct elements of {I, ... ,n} (n ?': p), we denote
by X k, ... kp the element of HQ9n given by

'""' (1) (n)

X k, ... kp = ~ Yi "'Yi
i

where yjkj) = x~j) for any j ::::; p and y;k) = 1 otherwise. For instance, if
R = 2:i Si t i , then R31 will be the element of HQ93 given by

R31 = L ti 1 Si'
i

We now introduce the main concept of this section.

Definition VIII.2.2. A quasi-cocommutative bialgebra (H, /-L, 'TI, 6., E, R)

or a quasi-cocommutative Hopf algebra (H, /-L, 'TI, 6., E, S, S-1, R) is braided
if the universal R-matrix R satisfies the two relations

(2.3)

and
(2.4)
174 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

Braided bialgebras are central in the theory of quantum groups and R-

matrices. In the literature, notably in Drinfeld's papers [Dri87] [Dri89a]
where this concept was defined for the first time, braided bialgebras are
called quasi-triangular bialgebras. We call them braided because their cat-
egories of modules are braided in a sense that will be explained in Chapter
XIII.
If R = 2:: i si rgyt i , Relations (2.3) and (2.4) can be expressed respectively
as
(2.5)
i,(s;) i,j

and
L si rgy (td rgy (t i )" = L SiSj rgy tj rgy t i (2.6)
i,(t;) i,j
Example 1. Cocommutative bialgebras are braided with universal R-matrix
R=lrgyl.
Here is a non-trivial example.
Example 2. (Sweedler's four-dimensional Hopf algebra) Let H be the al-
gebra generated by two elements x, y and relations

x 2 = 1, y2 = 0, yx + xy = 0.
The set {I, x, y, xy} forms a basis of the under lying vector space. There is
a unique Hopf algebra structure on H such that

.6.(x) = x rgy x, c(x) = 1, 5(x) = x,

.6.(y) = 1 rgy Y + y rgy x, c(y) = 0, 5(y) = xy.

Observe that the antipode 5 is of order 4 and that, for any a E H, we have
S2(a) = xax- 1 . Set

R,\ = ~ ( 1 rgy 1 + 1 rgy x + x rgy 1 - x rgy x) + ~ (y rgy y + y rgy xy + xy rgy xy - xy rgy y)

where), is any scalar. It is easy to show that R,\ satisfies the conditions
of Definition 2.2, thus endowing H with the structure of a braided Hopf
algebra for any scalar ),. Observe that R-;.l = TH,H(R,\).
We now investigate a few properties of universal R-matrices. The follow-
ing lemma will be useful later. It shows how to form new quasi-cocommut-
ative Hopf algebras from a given one.

Lemma VIII.2.3. (a) If (H, fL, 7),.6., c, 5, 5-1, R) is a quasi-cocommut-

ative Hopf algebra whose antipode 5 is bijective, then so are

(H, fLOP, 7),.6., c, 5- 1 ,5, R- 1 ), (H, fL, 7), .6. oP, c, S-l, 5, R- 1 )

and (H, fL, 7),.6. oP, c, S-l, s, TH,H(R)).

 VIII.2 Braided Bialgebras 175

(b) If, furthermore, (H, /-L, 1), 11, c:, S, s-1, R) is braided, then so is
(H, /-L, 1), 11oP, C:, s-1, s, TH,H(R)).

PROOF. (a) As a result of Corollary III.3.5, we see that (H, /-Lop, 1), 11, C:, S-l)
and (H,/-L,1},ll oP ,c,S-l) are Hopfalgebras. In (H,/-LP,1), 11,10, S-l), Rela-
tion (2.1) reads ll oP (x) = R- 1 1l(x)R, whereas it becomes

Il(x) = R- 1 Il op (x)R and Il(x) = TH,H(R)ll oP (X)TH,H(R)-l

in (H,/-L,1},ll op ,c,S-l), which proves Part (a).

(b) According to (a), the Hopf algebra (H,/-L,1},ll oP ,c,S-1,S,T(R)) is
quasi-commutative. We now have to check Relations (2.3) and (2.4).
Let us start with (11 lSi id H )(R) = R 13 R 23 and let us apply the transpo-
sition (12) to it. We get

We now use the circular permutation (123) to obtain

Similarly, one shows that Relation (2.4) for R implies Relation (2.3) for
TH,H(R). 0

Theorem VIII.2.4. Let (H, /-L, 1), 11, c:, R) be a braided bialgebra.
(a) Then the universal R-matrix R satisfies the equation

(2.7)
and we have
(10 lSi idH )(R) = 1 = (idH lSi c)(R). (2.8)
(b) If, moreover, H has an invertible antipode, then

(S lSi id H )(R) = R- 1 = (id H lSi S-l )(R) (2.9)

and
(S lSi S)(R) = R. (2.10)
Using the above conventions, in any braided Hopf algebra H whose uni-
versal R-matrix is of the form R = ~i Si lSi t i , Relations (2.7-2.9) are
equivalent to

i,j,k i,j,k

(2.12)
176 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

and

PROOF. (a) Relation (2.3) and the definition of R imply

R12(b. id)(R)
(b. 0P id)(R)R12
(TH ,H id)(b. id)(R)R12
(TH ,H id)(R13R23)R12

From (c id)b. = id and from (2.3), we get

R = (c id id)(b. id)(R) = (c id id)(R13R23) = (c id)(R)c(l)R.

Since c(l) = 1 and R is invertible, we obtain (cid)(R) = 1. Similarly, we

use the relation (id c)b. = id and (2.4) to derive (id c)(R) = 1.
(b) Now suppose that H has an invertible antipode S. We know that the
antipode verifies J-l(S id)b.(x) = c(x)l for all x E H. This implies

(J-l id)(S id id)(b. id)(R) = (c id)(R) = 1

from (2.8). Consequently,

1 = (J-l id)(S id id)(R13R23) = (S id)(R) S(l)R.

Since S(l) = 1, we get
(S id)(R) = R- 1 . (2.14)
Replace (H, J-l, TI, b., c, S, S-1, R) by the braided Hopf algebra

(H, J-l, TI, b.0P , c, S-1, s, TH,H(R))

of Lemma 2.3 (b). Then Relation (2.14) becomes

(S-1 id)(TH,H(R)) = TH,H(R)-l,

which is clearly equivalent to (id S-1 )(R) =~ R- 1. Finally, we have

(S S)(R) (id S)(S id)(R)

(id S)(R- 1 )
(id S)(id S-1 )(R)
(id id)(R)
R.

o
 VIII.2 Braided Bialgebras 177

In Chapter XIII we shall give a categorical interpretation of Relations

(2.3) and (2.4). Here we give another one in terms of algebra and coalgebra
maps. Indeed, with the universal R-matrix R we can build two linear maps
RA and AR from the dual vector space H* into H. They are defined by

(2.15)

where R = L; si Q9 tj and a is any linear form on H. We endow the

dual space H* with its canonical algebra structure and, if H is finite-
dimensional, with its canonical coalgebra structure.

Proposition VIII.2.5. Let (H, fl, 7],~, c, R) be a braided bialgebra. Then

RA is an algebra map and AR is an algebra antimorphism. Moreover, if H
is finite-dimensional, then AR is a coalgebra map and RA is a coalgebra
antimorphism.

PROOF. We first prove that RA is an algebra map. Let us compute RA(c).

From (2.12) we get RA(c) = Li c(Si)ti = 1, which shows that RA sends the
unit of H* to the unit of H. Now, let a and {3 be linear forms on H. Then
by (2.3), or its equivalent form (2.5), we have

l: (a{3)(si) t; = l: (a (3)(~(Si)) ti
Q9

l: a(si){3(Sj) titj = RA(a) RA({3),

i,j

which proves that RA preserves the multiplications. One may show in an

analogous way that AR is an algebra antimorphism using (2.4).
Now assume that H is finite-dimensional. Then the dual space H* has a
coalgebra structure. Its comultiplication ~ satisfies

Q(XY) = ~(a)(x Q9 y) = l: a'(x)al/(y).

( CY)

In order to prove that AR is a co algebra map, we first have to check that

Now, we have

so using (2.5) we get ~(AR(a)) = Li,j a(tjtj) si Q9 Sj' On the other hand,
178 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

L AR(a') AR(a//)
(a)

L Sia'(t i ) sja//(tj)
i,j,(a)

L a(titj) Si Sj
i,j

We next prove that AR preserves counits. Using (2.12), we get

E AR(a) = E(L Sia(t i )) = a(L E(si)ti) = a(l) = E(a).

(a),i (a),i

One similarly proves that RA is a coalgebra antimorphism using (2.4). 0

VIII.3 How a Braided Bialgebra Generates

R-Matrices
We now prove the existence of a solution of the Yang-Baxter equation on
every module over a braided bialgebra (H, j1, Tj,~, E, R).
Let V and W be two H-modules. The universal R-matrix R in H H
allows us to build a natural isomorphism c~, w of H - mod ules between V W
and W V. This isomorphism generalizes the flip TV W between the factors
V and Wand is defined for all v E V and W E W by

c~,w(vw) = Tv,w(R(vw)) = L tiwSiV (3.1)

where R = 2: i Si t i . By (2.13) c~,w is an isomorphism with inverse given

by

(c~,w )-l(w v) = R- 1(v w) = L S(si)V tiw = L siv S-l(t i )W.

(3.2)
The latter two equalities hold only when H has an invertible antipode.
Proposition VIII.3.1. Under the previous ."t,ypotheses,
(a) the map c~w is an isomorphism of H-modules, and
(b) for any triple (U, V, W) of H -modules, we have
c~v,w = (c~,w id v )(iduc~,w), c~,vw = (idvc~,w )(c~,v id w )
and
 VIll.4 The Square of the Antipode in a Braided Hopf Algebra 179

PROOF. (a) We have to prove that c~w is H-linear. Now, by (2.1), for any
x E H we have '
c~w(x(v
, Q9 w)) TV,W (R~(x)(v Q9 w))
TV,W (~OP(x)R(v Q9 w))
~(X)TV,W (R(V Q9 w))
x(c~,w(v Q9 w)).
(b) We prove the second and the last relations, leaving the first one to
the reader. For U E U, v E V and W E W we get using (2.6)

i,j

c~,VW(U Q9 v Q9 w).
As for the last relation in Part (b) of Proposition 3.1, we have

(c~,wQ9idu)(idvQ9c~,w)(c~,vQ9idw)(uQ9vQ9w) = L tktjWQ9SktiVQ9SjSiU
ifj,k

and
(idwQ9c~,v )(c~,wQ9idv )(iduQ9c~,w )(UQ9VQ9w) = L tjtiWQ9tksiVQ9SkSju,
i,j,k

Both right-hand sides are equal in view of (2.11). An alternative proof will
be given in XIII.l. 0

Setting U = V = W in Part (b) of Proposition 3.1, we conclude that

c~v is a solution of the Yang-Baxter equation for any H-module V. This
efficient way of producing R-matrices explains why the element R is called
the universal R-matrix of H. Observe that if R = 1 Q9 1, then c~w = TVW
is the flip. We have already remarked in Proposition IIL5.1 th'at the flip
was an isomorphism of H-modules for cocommutative H.

VIll.4 The Square of the Antipode in a Braided

Hopf Algebra
As we observed in Theorem IIL3.4, the antipode 8 of a cocommutative
Hopf algebra is an involution: 8 2 = id H . In the quasi-cocommutative case,
180 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

S2 is in general not equal to the identity. Nevertheless, as we shall see in

this section, it is an inner automorphism.
Let (H, f.i, 7/, t., 10, S, S-l, R) be a quasi-co commutative Hopf algebra with
an invertible antipode. Consider the element u of H given by

(4.1)

where R = I:i Si 0 t i Set R- 1 = I:i si 0ti'

Proposition VIII.4.1. Under the previous hypothesis, the element u is
invertible in H with inverse given by

(4.2)

and for all x E H we have

(4.3)
PROOF. Let us first show that S2 (x)u = ux for all x. If y belongs to H 0 H,
Relation (2.1) implies the equality
(t. 0P 0 id)(y)(R (1) = (R (1)(t. 0 id)(y)
in H 0 H 0 H. When y = t.(x) for some x E H, we get

L" ' XII si 0 X't i 0 X"' = ""'

" L SiX ' 0 t i XII 0 X" ' .
i,(x) i,(x)
To the latter relation we apply the linear map from H 0 H 0 H to H which
is id H 0 S 0 S2 composed with the multiplication from right to left. This
yields
L S2(x l ')S(x\)x" si = L S2(XIll)S(tiX")S;X',
i,(x) i,(x)
or, equivalently,

L S2(XI')S(ti)S(X')x" si = L S2(XI')S(X")S(ti)SiX', (4.4)

i,(x) i,(x)
since the antipode is an antiautomorphism of algebra. Let us first evaluate
the left-hand side of (4.4). By definition of th2 antipode and of the counit,
we have
L
S(x')x" 0 XIII = L
E(x')1 0 x" = 10 x.
(x) (x)
Hence, I:(x) S(x')x" 0 S2(x"') = 10 S2(X). Multiplying both sides on the
right by I:i Si 0 S(ti)' we get

L S(x')x" Si 0 S2(XIll)S(ti) = L Si 0 S2(x)S(ti)'

(x) ,i
 VIll.4 The Square of the Antipode in a Braided Hopf Algebra 181

Consequently, for the left-hand side of (4.4), we have

2:= S2(XI')S(ti)S(X')X"Si = 2:= S2(x)S(ti)Si = S2(X)u. (4.5)

i.(x)

The relation S2(X)U = ux will then be a consequence of (4.4-4.5) and of

2:= S2(XI')S(X")S(ti)SiX' = ux. (4.6)

i,(x)

Let us prove (4.6).

2:= x' @ S(X"S(X"')) = LX' @ S(c(X") 1) = L X'c(X") @ S(l) = x @ 1.

(:r) (x) (x)

Multiplying by u @ 1 on the left, we get

L S(ti)SiX' @ S2(XI')S(X") = UX @ 1,
i,(x)

which implies (4.6) after applying the multiplication in H.

It remains to show that u is invertible. Set

(4.7)

from the first part of the proof. Consequently,

UV = L S(tii ) SjS; = S(l)l = 1

i,j

since Li,j sisi @ tjti = RR- 1 = 1 @ 1. It follows that 1 = uv = S2(V)U,

which implies that u is left and right invertible with inverse v. D

Observe that S2(u) = u and S2(u- 1 ) = u- 1 .

Corollary VIIIA.2. Under the hypotheses of Proposition 4.1, we have

uS(u) = S(u)u. This element is central in H.
PROOF. Let x be any element in H. Applying S to ux = S2(x)u implies
S(x)S(u) = S(u)S3(x). Replacing x by S-l(X), we get

xS(u) = S(u)S2(x) = S(u)uxu- 1 ,

182 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

hence xS(u)u = S(u)ux. This proves that S(u)u is central in H. For x = u,

this formula leads to uS(u) = S(u)u. D

As we already know, any module V over a Hopf algebra H with invertible

antipode has two duals V* and *V. As vector spaces, both coincide with
the vector space of linear forms on V. However, the H-actions are different:
On V* an element a of H acts on a linear form a by

< aa,- >=< a,S(a)- >

whereas on *V it acts by

< aa,- >=< a,S-l(a) - >.

Using the defining property of the antipode we observe that the evaluation
maps V* V --> k and V *V --> k are H-linear (notice the precise order
of the tensorands). The element u induces an isomorphism between both
duals as recorded in the next proposition.
Proposition VIII.4.3. If H is a quasi-cocommutative Hopf algebra, then
the map a r--; a( u?) from V* to * V is an isomorphism of H -modules.
PROOF. By a(u?) we mean the linear form v r--; a(uv). Set j(a) = a(u?).
The map j is bijective because u is invertible. Let us show that j is H-linear.
For any v E V, Relation (4.3) implies

< j(aa), v> < a(S(a)u?), v>

< a, S(a)uv >
< a, S2(S-1(a))uv >
< a,uS-1(a)v >
< j(a), S-l(a)v >
< aj(a),v >.
D

Define the biduals V** and **V by V** = (V*)* and **V = *(*V). The
reader is invited to prove the following proposition.
Proposition VIII.4.4. Under the hypotheses of Proposition 4.3, the map
v r--; < -, uv > [resp. the map v r--; < -, u-1v >] from V to V** [resp. to
** V] is an H -linear injective map.

We now assume that H is braided. Then by (2.13) and by Proposition

4.1, the inverse of u is given by

(4.8)

In the braided case, we have the following additional relations for u.

 VIll.4 The Square of the Antipode in a Braided Hopf Algebra 183

Proposition VIII.4.5. If H is a braided Hopf algebra, then the element

u satisfies the relations
c(u) = 1, ~(u) = (R21R)-1(UU) = (UU)(R21R)-1,
~(S(u)) = (R 21 R)-1 (S(u) S(u)) = (S(u) S(u)) (R 21 R)-1

and for the central element uS (u) we have

~(uS(u)) = (R 21 R)-2 (uS(u) uS(u)) = (uS(u) uS(u)) (R21R)-2.

PROOF. (a) The relation c(u) = 1 follows from (2.12).

(b) Let us compute ~(u). Applying the flip TH,H to (2.1), we get

~(a) = R21~OP(a)Ki} (4.9)

for all a E H. Relations (2.1) and (4.9) imply

(4.10)
for all a E H. In view of (4.10) it is enough to show that ~(U)R21R = uu.
By (4.10) again and by Theorem III.3.4 we have

L ~(S(ti))~(si)R21R
L (S S)(~OP(ti))~(Si)R21R
L (S S)(~OP(ti))R21R~(si)
We now let the algebra H4 act on H H on the right by

(ab) (AB) = (SS)(B)(ab)A

where a, bE H and A, BE H H. We can rewrite the previous equalities
as
~(U)R21R = R21 . (R12R13R23R14R24)
By (2.7) this equals R21 . (R23R13R12R14R24)' which we now evaluate.
Using Relation (2.13), which gives the inverse of R, we get

R21 . R 23 = L S(tj) ti SiSj

i,j

Z,]

(S id)(R;-/ R 21 )
(S id)(l 1)
1 1.
184 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

Hence,

Next,

and

(u 1)(id 5)(R- 1 R)
(u 1)(id 5)(1 1)
(u 1).

Finally, we have

which is what we wished to prove.

(c) The formula for ~(5(u)) is an easy consequence of the formula for
~(u) and of (5 5) 0 ~ = ~op 0 5, which was proved in Theorem 111.3.4.
(d) The last relation follows from (b), (c) and the centrality of u5(u).
D

VIII. 5 A Dual Concept: Cobraided Bialgebras

Just as braided bialgebras induce R-matrices on their modules, there are
bialgebras inducing R-matrices on their comodules. These are the cobraided
bialgebras which we now define.

Definition VIII.5.l. A cobraided bialgebra (H,fL,TJ,~,E,r) is a bialgebra

H together with a linear form r on H H satisfying the conditions
(i) there exists a linear form r on H H such that

r * r = r * r = E, (5.1)

(ii) we have
fLOP = r * fL * r, (5.2)
(iii) and
 VIII.5 A Dual Concept: Cobraided Bialgebras 185

where * is the convolution operation on linear forms, and the linear forms
r 12 , r 23 and r 13 are defined by

r 12 =r c, r 23 = c r, r 13 = (c r)(TH,H id H ).
The linear form r is called the universal R-form of H. A Hopf algebra is
cobraided if the underlying bialgebra is.
This definition is dual to Definition 2.2. More precisely, Relation (5.2)
is dual to Relation (2.1), whereas Relations (5.3) correspond to Relations
(2.3-2.4). Conditions (5.1-5.3) can be reexpressed in the following way. For
any triple (x, y, z) of elements of H we have
(i)
l..:: r(x' y')r(x" y") = l..:: r(x' y')r(x" y") = c(x)c(y), (5.4)
(x)(y) (x)(y)

(ii)
yx = l..:: r(x' y')x"y"r(x lll y"') , (5.5)
(x)(y)

(iii)

r(xy z) = l..:: r(x' z')c(y')c(x")r(y" Zll) = l..:: r(x z')r(y Zll)

(x)(y)(z) (z)
(5.6)
and
r(xyz) = l..:: r(x' z')c(y')c(z")r(x" y") = l..:: r(x' z)r(x" y).
(x)(y)(z) (x)
(5.7)
A bialgebra satisfying only Conditions (i) and (ii) of Definition 5.1 may
be called quasi-commutative by analogy with the quasi-cocommutative case
of Section 2.
We now show how the universal R-form r of a cobraided bialgebra H
induces a solution of the Yang-Baxter equation on any H-comodule. The
map c~,w defined in (3.1) for a braided bialgebra H with a universal R-
matrix Rand H-modules V, W is the composition of the maps
R""d
VW '0'1 V0W )HHVW---7
idH07H,v0idw )HVHW 11V0p,w )VW~WV

where /-Lv and /-Lw are the actions of H on V and W respectively and where
we have identified R with the linear map from k to H H, sending 1 to R.
Let H be a cobraided bialgebra with universal R-form r. Given the H-
comodules V and W with respective coact ions Av : V ~ H V and
Aw : W ~ H W, we define the linear map

cv,w: VW ~ WV
186 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

by analogy with the above definition of c~,w as the composition of the

maps
VQ9W~WQ9V 6.w06.v )HQ9WQ9HQ9V~
id H 0rw,H0 id v
------------~)
H Q9 H Q9 W Q9 V r0 id wv )W Q9 V (5.8)

(this was obtained by reversing the arrows and interchanging V and W).
Using the conventions of nI.6 we can rewrite this definition for any v E V
and W E W as

cv ,w(VQ9W) = L r(wHQ9V H)WWQ9 Vv (5.9)

(v)(w)

Proposition VIII.5.2. (a) Under the previous hypotheses, the map cvw
is an isomorphism of H -comodules. '
(b) If U is a third H -comodule, we have

and
CU,v0W = (id v Q9 Cu,w)(cu,v Q9id w )
Moreover, we have

(cv,w Q9 id u )(id v Q9 cu,w)(cu,v Q9 id w )

= (id w Q9 cu,v ) (cu,w Q9 idv)(id u Q9 cv,w)
Setting U = V = W in the last relation, we see that cv,v is a solution of
the Yang-Baxter equation.
PROOF. (a) We use Condition (5.1) to prove that
a linear map cv ,w from W Q9 V to V Q9 W by
Cv,w is invertible. Define
cv,w(w Q9 v) = L r(vH Q9 wH) Vv Q9 ww
(v)(w)

We claim that cvw is an inverse to cvw. Let us show that it is a left

inverse. We have ' ,

v, cv,w )(v Q9 w)
(c w 0

L r(wH Q9v H )r((vV )H Q9 (WW)H) (vv)v Q9 (ww)w

(v)(w)

L r((w H )' Q9 (VH)') r((w H )" Q9 (VH)") Vv Q9Ww

(v)(w)

L (WH)(VH)VV Q9W W
(v)(w)
v Q9w.
 VIII.5 A Dual Concept: Cobraided Bialgebras 187

The second equality follows from the coassociativity of the coactions while
the third one is a consequence of Relation (5.1) and the last one follows from
v
the counitarity of the coactions. One proves that C W is a right inverse to
Cv W in a similar way. '
We now prove that Relation (5.2) implies that cv,w is a map of comod-
ules, namely we have

This is equivalent to

r(wH 0 VH) (ww )H(VV)H 0 (ww)w 0 (vv)v

L r((WW)H 0 (VV)H) vHw H 0 (ww)w 0 (vv)v
(v)(w)

for any v E V and W E W. Now by the coassociativity of the coactions, the

previous relation can be rewritten as

L r((wH)'0(VH)') (w H )"(vH )"0ww 0vv

(v)(w)

= L(v)(w) (VH)'(wH)'r((w H)" 0 (VH)") 0 Ww 0 vv'

The latter is a consequence of r * /-l = /-lop *r, which is equivalent to Relation

(5.2) after convolution with r.
(b) Let us prove that cuv,w = (cu,w 0 idv)(id u 0 cv,w). We have

(cu,W 0 id v ) ((id u 0 cv,w )(u 0 v 0 w))

L r(wH 0VH)r((WW)H 0 UH) (ww)w 0uu 0vv

(u)(v)(w)

= L r((w H )' 0 VH) r((w H )" 0 UH) Ww 0 Uu 0 Vv

(u)(v)(w)

L r(wH0uHVH)Ww0uu0Vv
(u)(v)(w)

cuv,w(u 0 v 0 w).

The second equality follows from the co associativity, and the third one from
Relation (5.7). One proves that cu,vw = (id v 0 cu,w)(cu,v 0 id w ) in a
similar way.
The last relation of Proposition 5.2 is a consequence of the previous
relations and of the naturality of the maps cr. We leave the proof to the
reader. A proof in a more general context will be given in XIII.l. 0
188 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

VIII.6 The FRT Construction

We have just seen that a cobraided bialgebra gives rise to an R-matrix on
each comodule. Conversely, Faddeev, Reshetikhin and Takhtadjian showed
in [RTF89] that any R-matrix c in Aut(V V) on a finite-dimensional
vector space V can be obtained as in Section 5 from a cobraided bialgebra
A(c) coacting on V. The Faddeev-Reshetikhin-Takhtadjian construction
(FRT construction for short) is based on the following theorem.

Theorem VIII.6.1. Let V be a finite-dimensional vector space and c an

endomorphism ofV V. There exists a bialgebra A(c) together with a linear
map D. v : V ----* A(c) V such that
(i) the map D. v equips V with the structure of a comodule over A(c),
(ii) the map c becomes a comodule map with respect to this structure,
(iii) if A' is another bialgebra coacting on V via a linear map D.~ such
that Condition (ii) is satisfied, then there exists a unique bialgebra mor-
phism f : A(c) ----* A' such that

The bialgebra A(c) is unique up to isomorphism.

The proof will be given in several steps.

1. In the first one, we define A(c) as an algebra. Let {vih<i<N be a basis
of V and let the coefficients c'?t
be defined by - -

C(Vi vj ) = L crt Vm Vn-

lsm,nsN

Pick a family of indeterminates T!, where i and j both run over the set
{I, ... ,N}.
Definition VIII.6.2. The algebra A(c)is the quotient of the free algebra
F generated by the family (T!hsi,jSN by the two-sided ideal I(c) generated
by all elements err where
e:nJ. n = ~
~
ckTmT n _
ij k
~
~
TkTcmn
i j k (6.1)
lSk,SN lSk,SN
and i, j, m and n run over the indexing set.

2. We put a bialgebra structure on A(c).

Lemma VIII.6.3. There is a unique bialgebra structure on A(c) such that

D.(T!) = L Tik Tk and c(T!) = 8ij (6.2)

lSkSN
 VIII.6 The FRT Construction 189

PROOF. It is clear that the above formulas define unique algebra maps

A :F -+ F &; F and c: F -+ k.

To check the coassociativity and the counitarity, it is enough to check these

on the generators T!, which is easy and done as in IIl.4 and in IV.5. We
also have to prove that J(c) is a coideal, i.e., that

A(J(c)) C J(c) &; F + F &; J(c) and c(J(c)) = o.

We have

k,,p,q k,,p,q

p,q k,,p,q

p,q k,,p,q

p,q p,q
and

c(c::;n)
k, k,
L C~fOkmOn - L 0ikOjcrt
k, k,

crr - cfr = o.
o
3. We now let A(c) coact on V. Define a linear map Av from V to
A(c) &; Von the basis {vih:5i:5N by

AV(vi ) = L T! &; Vj. (6.3)

l:5j:5N
It is an easy exercise to check that this endows V with a left comodule
structure over the bialgebra A(c).
4. We prove that the endomorphism c of V &; V is a comodule map for
the coaction we have just defined. The coaction Av induces on V &; V a
coaction Av cw defined by

AV@v(vi &; Vj) = L TikTf &; vk &; vi

l:5k,:5N
190 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

Then c is a comodule map if and only if

b.. VlZiV (C(Vi Vj)) - (idA(c) c) (b.. V0V (V i v))

L Tk'TF c~Jvm vn - L TikTJ ckinvm vn

k,R.,m,n k,i,m,n

vanishes in A(c) V. Now, it is clear that the last expression is equal to

m,n
which is zero by definition of A(c).
5. We now establish the universality of A(c). Let (A', b..~) be a pair satis-
fying the conditions of Theorem 6.1. Then there exists a family (U{)l$;i,j$;N
of elements of A' uniquely determined by

b..~(Vi) = L u{ Vj.
l$;j$;N

The coassociativity and the counitarity of b..~ imply that

b..(u{) = L u~ u{ and c(u{) = 8ij

ls;k$;N

Condition (ii) of Theorem 6.1 is equivalent to the vanishing of

for all i and j, in other words to the vanishing of

'~
" ck1umu n _ ' " uku1cmn
ij k l ~ i j kl
l$;k,l$;N l$;k,l$;N

for all i, j, m and n. From this it is clear that the map! from F to A'
defined by !(T!) = u{ for all i and j extends to a bialgebra map factoring
through A(c). Let us check the relation b..~ = (f idv)b.. v . For any i we
have

l$;j$;N l$;j$;N

Conversely, the relation b..~ = (f id v )b.. v necessarily implies !(T!) = u{,

which proves the uniqueness of ! along with the fact that the family (T!)
generates the algebra A(c). This completes the proof of Theorem 6.1.
 VIII.6 The FRT Construction 191

Theorem VIII.6.4. Assume in addition to the hypotheses of Theorem 6.1

that the endomorphism C ofV0V is a solution of the Yang-Baxter equation.
Then there exists a unique linear form r on A(c) 0 A(c) turning A(c) into
a cobraided bialgebra such that cv,v = c. We have

(6.4)
for all i,j,m and n.

The rest of this section is devoted to the proof of Theorem 6.4.

(a) Suppose A(c) is cobraided with a universal R-form r such that the
automorphism Cv v coincides with the given R-matrix c. By (5.9) and (6.3)
we have '

rn,n

On the other hand, we have c(v j 0v i ) = L:m,n cjtvm 0v n . It follows that

r(Tim 0TJ') = cjln for all i,j, m, n. Relations (5.6~5.7) imply the uniqueness
assertion in Theorem 6.4.
(b) We now prove the existence of r. We first have to define r on the
whole space A(c) 0 A(c). Let W be the vector subspace of F spanned by
the set {Tlh:'Oi,j:'ON. We define r : W 0 W ----> k by (6.4). Conditions (5.3)
and
r(l 0 Tn = r(Tl 01) = c(Tf) = (jij
allow one to extend r into a linear form, still denoted r, on F 0 F.
In order to prove that r defines a form on A(c) 0 A(c), we have to prove
the following lemma.
Lemma VIII.6.5. We have r(F 0 I(c)) = r(I(c) 0 F) = o.
PROOF. First, we observe that

r(l 0 I(c)) = r(I(c) 01) = c(I(c)) = o.

Using Conditions (5.3), we see that it is now enough to show that the
images r(T% 0 eft) and r( eft 0 T%) vanish for all i, j, m, n, p, q. We have

"6 ckf
'J
r(Tq
P
0 TmT
k f
n) - "
6 r(Tqp 0 rkTf)
'J
cmn
k
k,f k.f

L c~f r(T; 0 TP) r(Ti 0 Tk')

k,f,r
-L r(T; 0 TJ) r(Ti 0 Tik ) c~n
k,,r
"6 kJI cfp
cij rn cqm - "
6 rcjpc
f qk ckJI
mn ,
kr ir
k,,r k,P,r

which is zero in view of (1.1), i.e., of the Yang-Baxter equation. 0

192 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

(c) Now that r is defined, we have to check the conditions of Definition

5.1. This will be done in several steps.
1. Conditions (iii) are satisfied by definition of r.
2. Condition (i): we have to prove that r is invertible with respect to the
convolution, namely that there exists a linear form r on A(c) (9 A(c) such
that r * r = r * r = c. We define r on the generators T! by

where the coefficients (c-1)iT are defined in terms of the inverse c- 1 of c

by

m,n

Lemma VIII.6.6. The above formulas define a unique linear form r on

A(c) (9 A(c) such that for all x, y in A(c) we have

r(xy (9 z) =L r(y (9 z') r(x (9 z") (6.5)

(z)

and
r(x (9 yz) =L r(x' (9 y) r(x" (9 z). (6.6)
(x)
PROOF. The proof is similar to the proof of Lemma 6.5. Use the fact that
c- 1 is also a solution of the Yang-Baxter equation. 0

We now check Relation (5.4). Let us prove that

L r(x' (9 y') r(x" (9 y") = c(x)c(y) (6.7)

(x)(y)

by induction on the degrees of x and y. If x or y is of degree zero, this is

immediate. If both x and yare of degree 1, this follows from the subsequent
computation. For x = ~m and y = Tp we have

p,q p,q

The second equality results from the fact that c- 1 is the inverse of c. The
general case follows from the next lemma.

Lemma VIII.6.7. If Relation (6.7) is verified by the couples (x,y), (x,z)

and (y, z), then it also holds for the couples (x, yz) and (xy, z).
 VIII.6 The FRT Construction 193

PROOF. We give the proof for the couple (x, yz). The proof for (xy, z) is
similar. In view of Relation (5.7) and Relations (6.6-6.7) we have

L r(x' @ y' z')r(x" @ y" z")

(x)(y)(z)

L r(x' @ z')r(x" @ y')r(x'" @ y")r(x"" @ z")

(x)(y)(z)

L r(x' @ z')c(x")c(y)r(x'" @ z")

(x)(z)

L c(y)r(x' @ z')r(x" @ z")

(x)(z)

c(y)c(x)c(z)
c(x)c(yz).

o
The relation L(x),(y) r(x' @y')r(x" @y") = c(x)c(y) is proved similarly.
3. Condition (ii): We have to check that for any x and y in A(c) we have

L r(x' @y')x"y" = L y'x'r(x" @y"). (6.8)

(x),(y) (x),(y)

We proceed as for Condition (i), namely we first check (6.8) in case x = 1

or Y = 1 when it is trivial and in case x = and y = Tr TT,
then show that
if (6.8) is true for (x, V), (x, z) and (y, z), then it is for (x, yz) and (xy, z).
Firstly, we have

p,q p,q

p,q

p,q

because of the defining relations of A(c).

We continue with the following analogue of Lemma 6.7.
Lemma VIII.6.8. 1/(6.8) is verified by the couples (x, V), (x, z) and (y, z),
then it is by the couples (x, yz) and (xy, z).
PROOF. Suppose (6.8) is true for (x, y) and for (x, z). Then for (x, yz) we
have

L r(x' @ y' z')x"y" z" L r( x' @ z')r( x" @ y')x'" y" z"
(x)(y)(z) (x)(y)(z)
194 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

L r(x' c>9 z')y' x" r(x"' c>9 y")ZI

(x)(y)(z)

L y'r(x' c>9 z')x" zllr(x lll c>9 y")

(x)(y)(z)

L y' z' x'r(x" c>9 z")r(x"' c>9 y")

(x)(y)(z)

L y' z' x' r(x" c>9 y" Zll).

(x)(y)(z)

The other cases are proved similarly. o

This completes the proof of Theorem 6.4.

VIII. 7 Application to GLq(2) and SLq(2)

In this section we show that the bialgebra Mq(2) and the Hopf algebras
GLq(2) and SLq(2) defined in Chapter IV are cobraided.
Let V be a two-dimensional vector space with basis {VI' v 2 } and let c
be the automorphism of V c>9 V whose matrix with respect to the basis
{vI c>9 VI' V 2 c>9 V 2 , VI c>9 V 2 , V 2 c>9 vd is

q 0 0 0 )
-1/2
( 0 q 0 0 (7.1)
q 000 1
o 0 1 q _ q-l

where ql/2 is an invertible scalar. This matrix has been displayed in Section
1 where we proved it was an R-matrix. The FRT construction associates
to e a cobraided bialgebra A(e) which we now describe.
Proposition VIII.7.1. The bialgebra A(e) associated to the R-matrix (7.1)
is isomorphic to the bialgebra Mq(2) of Definition IV.3.2.

PROOF. Let Tl = a, Tf = b, Ti = c and Ti = d. By the FRT construction,

A( e) is the algebra generated by a, b, e, d and the sixteen relations which
can be written in the following compact matrix form
o
6~ ~
(0 o ba
de )
00 1 be
001 q _ q-l da

)
o 0 o
( ~~!~~!
ae
ea
bd
db
ad
eb
~~)(60
be
da 0
q
o
o
0
0
1
o
1
q _ q-l
 VIII.7 Application to GLq(2) and SLq(2) 195

An easy computation shows that these relations are equivalent to the six
relations
ba qab, db qbd,
ca qac, dc qcd,
cb bc, da-ad (q - q-l)bc,

defining the algebra Mq(2). This identifies A(c) and Mq(2) as algebras.
The corresponding comultiplications are clearly the same (compare (6.2)
and Theorem IV.5.1). 0

From this and from Theorem 6.4, we deduce the following important
result on Mq(2).
Corollary VIII.7.2. The bialgebra Mq(2) has a unique structure as a co-
braided bialgebra with universal R-form r determined by

r(aa) r(bb) r(a b) o 0

r(ba) ) (q
( r(cc) r(dd) r(c d) r(dc) = oX 0 q 0
r(ac) r(bd) r(a d) r(b c) 0 o I
r(ca) r(db) r(c b) r(da) 0 o 0

where oX = q-l/2.
It is easy to check that the coaction of A(c) on the two-dimensional vector
space V coincides with the coaction of Mq(2) on the elements of degree I
of the quantum plane kq[x,y] (see IV.7).
We now show that GLq(2) and SLq(2) are cobraided with the same
universal R-form. Since SLq(2) is a quotient of GLq(2), it is enough to
prove this for SLq(2). We start with the following lemma.
Lemma VIII.7.3. For all x E Mq(2) we have

r(x det q) = r(detq x) = c(x).

Recall that detq = da - qbc is the quantum determinant introduced in
Chapter IV.
PROOF. Suppose we have proved that the relations in Lemma 7.3 hold for
two elements x and y. Since detq is grouplike by Theorem IV.5.1, we deduce
from (5.6) that

r(xy det q) = r(x det q) r(y det q) = c(x)c(y) = c(xy),

which reduces a proof of Lemma 7.3 to checking it for x = a, b, c, d.
For x = a we have

r(a det q) = r(a a)r(a d) + r(b a)r(d d)

-qr(a c)r(a b) - qr(b c)r(d b)
196 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

since ,6,(a) = a a + b d. Using Corollary 7.2, we get

r(a det q ) = q-l/2 qq -l/2 +0- 0 - 0 = 1 = c(a).

We leave the other verifications to the reader. o

Corollary VIII.7.4. The Hopf algebras GLq(2) and SLq(2) are cobraided
with universal R-form r.

PROOF. Recall that SLq(2) is the quotient of Mq(2) by the ideal I generated
by the element detq -1. Now, Lemma 7.3 is equivalent to the statement that

r((detq-1)x) =r(x(detq-1)) =0

for all x E Mq(2). Therefore r vanishes on I Mq(2) and on Mq(2) I,

which proves that r defines a bilinear form on SLq(2). 0

Remark 7.5. The normalization constant q-l/2 in front of the R-matrix in

(7.1) has been introduced precisely so as to have r vanish on the ideal I
defining SLq(2).

VIII. 8 Exercises
1. Consider a matrix of the form

Show that it is a solution of the Yang-Baxter equation if and only if

the following conditions are satisfied:

adb = adc = ad(a - d) = 0,

p 2 a = pa 2 + abc, q2 a = qa 2 + abc,
p2 d = pd 2 + dbc, q2 d = qd 2 + dbc.
2. Consider the Hopf algebra H of Section 2, Example 2. Show that
there exists an automorphism lP of the Hopf algebra H such that
(lP lP) (R)J = R)..' if and only if there exists a non-zero scalar fL such
that )..' = fL2)...

3. Find all (co )braided bialgebra structures on the group bialgebra of a

finite group.
4. Let H be a finite-dimensional bialgebra and H* be the dual bialgebra.
Show that H* is cobraided if and only if H is braided.
 VIII.8 Exercises 197

5. Let H be a cobraided bialgebra with universal R-form r. Show that

6. Let (H, fL, ry,~, c, 8, 8-1, r) be a cobraided Hopf algebra with invert-
ible antipode 8.

(a) Show that r 0 (ry id H ) = r 0 (id H ry) = c and

r 0 (8 id H ) = f, f 0 (id H 8) = r, r 0 (S 8) = r.
(b) Define a linear form u on H by u = r 0 (id H 8) 0 ~op. Show
that u is invertible as an element of H* and that 8 2 = u*idH*u
where u denotes the inverse of u for the convolution.

7. Let (H, fL, ry,~, c, r) be a cobraided bialgebra. Define linear maps rA

and Ar from H to H* by rA(X) = r(- x) and Ar(X) = r(x -).
Show that rA is an algebra antimorphism and Ar is an algebra map
and, in case H is finite-dimensional, rA is a coalgebra map and Ar a
coalgebra antimorphism.

8. Let A be the algebra k{s, t, e 1}/(S2, st+ts). Show that the following
formulas define a unique cobraided Hopf algebra structure on A:

~(t) = t t, ~(s) = s 1 + e 1 s,

c(t) = 1, c(s) = 0, S(t) =e 1, 8(s) = st,

r(tt) =-1, r(st) =r(ts) =r(ss) =0.
Check that the antipode 8 is of order 4.

9. Let c E Aut(V V) be a solution of the Yang-Baxter equation and

let c' = T v,v 0 COTv,v. Show that we have the following isomorphisms

10. Let c be the R-matrix of Proposition 1.4. Prove that A(c) is the
algebra generated by (T!)l$.i,j$.N and the relations

rji TjTr = r nm TimTj, r ji TrT;: - r mn T;:Tr = (q - pq-1) TimTj

where i, j, m, n run over all positive integers :S N such that i < j and
m>n.
11. Use the description of the universal R-form on 8Lq(2) to find an
R-matrix on the SL q(2)-comodule kq[x, y] described in IV.7.
198 Chapter VIII. The Yang-Baxter Equation and (Co)Braided Bialgebras

VIII. 9 Notes
The Yang-Baxter equation first came up in a paper by Yang [Yan67] as a
factorization condition of the scattering S-matrix in the many-body prob-
lem in one dimension and in work of Baxter [Bax72] [BaxS2] on exactly solv-
able models in statistical mechanics. It also played an important role in the
quantum inverse scattering method created around 1975-79 by Faddeev,
Sklyanin, Takhtadjian [FadS4] for the construction of quantum integrable
systems. Attempts to find solutions of the Yang-Baxter equation in a sys-
tematic way have led to the theory of quantum groups (see [DriS7]). Many
papers in the literature are devoted to the construction of R-matrices, e.g.,
[DriS5] [DriS7] [JimS6a] [JimS6b] [KSSO], to quote but a few.
The concept of a quasi-co commutative and of a braided (or quasi-tri-
angular) Hopf algebra is due to Drinfeld [DriS7] [DriS9a]. For a review, see
[Maj90b]. The four-dimensional Hopf algebra of Example 2 of Section 2
is due to Sweedler. The universal R-matrices R). were found by Radford
[Rad93a].
The dual concept of cobraided bialgebras appears in [Hay92] [LT91]
[Maj91b] [Sch92]. Cobraided bialgebras have properties dual to braided
bialgebras. We gave some of them in Exercises 5-7.
The FRT construction is due to Faddeev, Reshetikhin and Takhtadjian
[RTFS9]. The bialgebras M p,q(2) and Mq(n) of IV.I0 can be obtained by
this method (see Exercise 10). In Sections 5-6 we followed the treatment
proposed by [LT91].
Exercise 1 is taken from [Kau91] and Exercise 2 from [Rad93a]. The
cobraided Hopf algebra of Exercise S was found by Pareigis [ParSl] before
the advent of quantum groups.
Chapter IX
Drinfeld's Quantum Double

In the previous chapter we showed that braided Hopf algebras provided

solutions of the Yang-Baxter equation. The problem is now to find enough
such Hopf algebras. Drinfeld [Dri87] devised an ingenious method, the
"quantum double construction" , which builds a braided Hopf algebra out of
any finite-dimensional Hopf algebra with invertible antipode. It is the goal
of this chapter to describe this construction in full detail, and to show how
to apply it to the finite-dimensional quotient of the Hopf algebra Uq (.s1(2))
considered in VI.5. We also give a characterization of the modules over the
quantum double in Section 5.

IX.1 Bicrossed Products of Groups

The quantum double construction is a special case of the bicrossed product
construction. Since the latter is rather involved, we start with an analo-
gous construction for groups, namely the bicrossed product of groups due
to Takeuchi [Tak81]; it generalizes the notion of a semidirect product of
groups.
Let G be a group with subgroups Hand K. We assume that for any
element x in G there exists a unique pair (y, z) E H x K satisfying

x =yz. (1.1)

This allows us to attach to any pair (y, z) E H x K a unique element z . y

200 Chapter IX. Drinfeld's Quantum Double

in H and a unique element zY in K such that

zy = (z . y) zY. (1.2)

Let y, y' be elements of Hand z, z' be elements of K. Expanding the

associativity relations

(zz')y = z(z'y) and z(yy') = (zy)y'

gives us the relations

(zz') . y z (z' . y), (1.3)

(zz')Y zZ'Y z'Y , (1.4)
z (yy') (z y)(zY . y'), (1.5)
zYY' (zY)Y' . (1.6)

Similarly, we expand z = z1 and y = 1y, which implies

1, (1.7)
z, (1.8)
1 y y, (1.9)
1Y 1. (1.10)

Relations (1.3) and (1.9) mean precisely that the map a : K x H ~ H

defined by
a(z,y)=zy
is a left action of the group K on the set H. Similarly (1.6) and (1.8) mean
that the map {3 : K x H ~ K defined by

(3(z, y) = zY

is a right action of the group H on the set K. We make the following

definition.

Definition IX.1.1. A pair (H, K) of groups is said to be matched if there

exist a left action a of the group K on the set H and a right action {3 of
the group H on the set K, such that for all y, y' E Hand z, z' E K we have
,
(zz')Y = Zz Y z'Y, (1.4)

z (yy') = (z y)(zY . y'), (1.5)

z1 = 1, (1.7)
1Y = 1, (1.10)
where a(z,y) = z y and (3(z,y) = zY.
 IX.1 Bicrossed Products of Groups 201

Proposition IX.1.2. (a) Let (H, K) be a matched pair of groups. There

exists a unique group structure, denoted H I><l K, on the set-theoretic pro-
duct H x K with unit (1,1) such that

(y, z)(y', z') = (y(z y'), zy' z')

for all y, y' E Hand z, z' E K. This group structure is wlled the bicrossed
product of Hand K. Furthermore, the groups Hand K can be identified
respectively with the subgroups H x {I} and {I} x K of H I><l K, and every
element (y, z) in H I><l K can be written uniquely as the product of an
element of H x {I} and an element of {I} x K:

(y,z) = (y,l)(l,z)

where y E Hand z E K.
(b) Conversely, let G be a group and H, K be subgroups of G such that
the multiplication on G induces a set-theoretic bijection from H x K onto
G. Then the pair (H, K) is necessarily matched and the previous bijection
induces a group isomorphism from the bicrossed product H I><l K onto G.

PROOF. (a) It is easy to check that the above-defined product on H I><l K

is associative with (1,1) as unit. Details are left to the reader.
To prove that (y, z) is invertible in the bicrossed product, let us first look
for elements y' E Hand z' E K such that

(y, z)(y', z') = (1,1).

By definition of the product, this is equivalent to the following two relations:

y(z . y') =1 and zy' z' = 1.

From the first one we derive

Y , =Z -1 . ( zy ') =z -1 .y -1 ,

and then from the second one we get

z' = (ZZ-1.y-')-1.

Set (y', z')(y, z) = (Y, Z) where y' and z' are given the above values. We
have to show that (Y, Z) = (1,1). Multiplying the last identity by (y, z) on
the left, we get

(y, z) = (y, z)(Y, Z) = (y(z Y), zY Z).

This implies that

Y = z-1 . (z Y) = Z-1 . 1 = 1 and Z = zY z-1 = z1 Z-1 = 1.

202 Chapter IX. Drinfeld's Quantum Double

Thus, the element (y, z) is invertible with inverse equal to

( y,z ) -1 -_ ( -1.
z -1 ( z-1.y-l)_1)
y , z .

It is easy to check that

(y, l)(y', 1) = (yy', 1), (1, z)(l, z') = (1, zz') and (y, 1)(1, z) = (y, z),
which proves the remaining assertions of Part (a).
For the proof of Part (b), it suffices to review the arguments that led us
to Definition 1.1. D

Example 1. (Product of groups) Let Hand K be groups. We let each one

act trivially on the other, which means, using the above notation, that

z . y = y and zY = Z.

Then (H, K) is a matched pair, and the bicrossed product H [XI K is

isomorphic to the usual product of groups H x K.
Example 2. (Semidirect product of groups) Let Hand K be groups. We
suppose that H acts trivially on K, which means that zY = z, and that K
acts on H by group automorphisms, which means that

z (yy') = (z y)(z y') and z 1 = 1

for all y, y' E Hand z E K. Then (H, K) is a matched pair and the
bicrossed product H [XI K is isomorphic to the semidirect product of K
by H. In this case, the identity (1, z)(y, 1)(1., z)-1 = ((z y), 1) proves that
H x {I} is a normal subgroup of H [XI K and that the action of K on H
corresponds to the conjugation in the bicrossed product.

IX.2 Bicrossed Products of Bialgebras

We observed in Chapter III that the algebra of a group has a natural Hopf
algebra structure. The question we raise now is this: Given a matched pair
(H, K) of groups, can we build the algebra ofthe bicrossed product H [XI K
out of the group algebras k[H] and k[K]? In order to answer this question,
we first give a group algebra version of the action of a group on a set. Let
us consider the case of a group G acting on a set X via a map

a : G x X ---> X.

Linearizing, we get a morphism of coalgebras

a : k[G x X] ---> k[X]

 IX.2 Bicrossed Products of Bialgebras 203

for the coalgebra structures introduced in IIL1, Example 3. Composing

with the natural isomorphism

k[G] k[X] ~ k[G x X],

which is a co algebra isomorphism by Proposition IIL1.4, we see that the

group action of G on the set X gives rise to an action of the Hopf algebra
k[G] on the coalgebra k[X] such that the structural map

k[G] k[X] -> k[X]

is a morphism of coalgebras. The coalgebra k[X] is thus a module-coalgebra

on the Hopf algebra k[G] in the sense of the following definition.

Definition IX.2.1. Let H be a bialgebra and C be a coalgebra. We say

that C is a module-coalgebra over H if there exists a morphism of coalgebras
H C -> C inducing an H -module structure on C.

We are ready to give the definition of a matched pair of bialgebras.

Definition IX.2.2. A pair (X, A) of bialgebras is matched if there exist

linear maps a : A X -> X and {3 : A X -> A turning X into a module-
coalgebra over A, and turning A into a right module-coalgebra over X, such
that, if we set

a(ax)=ax and {3(ax)=a x ,

the following conditions are satisfied:

a . (xy) = ~
~ (a I . x I ) (a , ,x" . y), (2.1)
(a)(x)

a1=c(a)1, (2.2)
(ab)X = L ab'.x' b" X" , (2.3)
(b)(x)
F = c(x)l, (2.4)
L a'x' a" . x" = L a" x" a' . x' (2.5)
(a) (x) (a) (x)

for all a,b E A and x,y E X.

Observe that Condition (2.5) is automatically satisfied when both bial-

gebras A and X are cocommutative. We also remark that Definition 2.2
is an immediate generalization of Definition 1.1. As a basic example of a
matched pair of bialgebras, we may take the pair (k[Hl, k[K]) of group
bialgebras where (H, K) is a matched pair of groups.
204 Chapter IX. Drinfeld's Quantum Double

The maps a and f3 being morphisms of coalgebras, we have

uA( ax ) = "'"
L-t a I X I r59a " X " and c(a x) = c(a)c(x)1 (2.6)
(a)(x)

in X, and

~(aX) = L a'x ' r59 a"x" and c(a X) = c(a)c(x)1 (2.7)

(a) (x)

in A. We state the main result of this section; it is a natural extension of

Proposition 1.2.

Theorem IX.2.3. Let (X, A) be a matched pair of bialgebras. There exists

a unique bialgebra structure on the vector space X r59 A, with unit equal to
1 r59 1, such that its product is given by

(x r59 a)(yr59 b) = L x(a l 'y')r59a"Y"b,

(a)(y)

its coproduct by

~(x r59 a) = L (x' r59 a' ) r59 (x" r59 a"),

(a)(x)

and its co unit by

c(x r59 a) = c(x)c(a)
for all x, y E X and a, bE A. Equipped with this bialgebra structure, X r59 A
is called the bicrossed product of X and A and denoted X ~ A. Further-
more, the injective maps i x (x) = x r59 1 and i A (a) = 1 r59 a from X and from
A into X ~ A are bialgebra morphisms. We also have

x r59 a = (x r59 1)(1 r59 a)

foraEAandxEX.
If the bialgebras X and A have antipodes, respectively denoted S x and
SA' then the bicrossed product is a Hopf algebra with antipode S given by

S(x r59 a) = L SA(a"). Sx(x") r59 SA(a')Sx(x').

(x)(a)

PROOF. The above formulas show that we equipped the bicrossed product
with the coalgebra structure of the tensor product of coalgebras X and A.
It is then clear that i x and i A are co algebra morphisms. It remains to be
proved that X ~ A has an algebra structure and that the coproduct and
the counit, as well as the embeddings ix and i A , are algebra morphisms.
Let us start with the associativity of the product. An easy but tedious
computation using Relations (2.1) and (2.3) and the fact that both a and
 IX.2 Bicrossed Products of Bialgebras 205

(3 are coalgebra morphisms, shows that if x, y, z E X and a, b, c E A, then

both
((xa)(yb))(zc) and (xa)((yb)(zc))
are equal to

L x( a' . y') (( a"Y" b') . z') a"'Y'" (b"z") b"'z'" c.

(a)(b)(y)(z)

For the unit we get, using (2.2) and (2.4),

(ll)(xa) = L (l.x')l X " a = L x'c(x")a = L x'c(x")a = xa

(x) (x) (x)
and

(xa)(ll) = L x(a'l)a"l = L xc(a')a" = L xc(a')a" = xa.

(a) (a) (a)
Let us prove that the counit is a morphism of algebras. We have to check
that
c((x a)(y b)) = c(x a)c(y b) = c(x)c(a)c(y)c(b).
Now the left-hand side is equal to

c( L x(a'. y') a"Y"b) L c( x )c( a' . y')c( a"Y" )c(b)

(a)(y) (a)(y)
c(x )c(b) (L c( a')c(y')c( a")c(y"))
(a)(y)
c(x)c(b)c(a)c(y)

in view of (2.6) and (2.7). To conclude, we show that Relation (2.5) implies
that the coproduct is a morphism of algebras. We have

.6. (x a).6.(y b) = L x'(a' y') a"y"b' x"(a"' y",) a""Y""b",

(x)(a)(y)(b)

and, on the other hand,

.6. ( (x a) (y b)) = L x' (a' . y') a"'Y'" b' x" (a" . y") a""Y"" b".
(x)(a)(y)(b)

Both expressions are equal in view of Relation (2.5).

Now suppose that A and X have antipodes. We have to check that the
formula
S(x a) = L SA(a") Sx(x") SA(a')sx(x')
(x)(a)
206 Chapter IX. Drinfeld's Quantum Double

defines an antipode on the bialgebra X [Xl A. Using the fact that SA and
Sx are antipodes, we get

L (x' a')S(x" a")

(x)(a)

L (x' a' ) (SA (alii) . Sx(x lll ) SA(a")Sx(x lf ))

(x)(a)

L x' ((aISA(a llll )) . SX(X"")) (aIlSA(a'Il))Sx(xlf)

(x)(a)

L x' ((aISA(a lll )) . SX(X IIl )) c(a l )c(Sx(x"))l

(x)(a)

L x' ((aISA(a")) . Sx(x ll ) ) 1

(x)(a)

c(a)(L X'SX(X") 1)
(x)
c(a)c(x)l 1
c(xa)ll.

Similarly, we have

L S(X' a')(x" a")

(x)(a)

L (SA(a") Sx(x") SA(al)sx(x'))(x 'll a'II)

(x)(a)

L (SA(a'Il ). Sx (X"I)) (SA (a")SX (Xlf) . XliII)

(x)(a) ' S ( ') Iflf' 1111
SA ( a) x x x a

L SA(a"). (SX(X")X'") SA(a')Sx(x')x lflf a'II

(x)(a)

L c(x") (SA (a") . 1) SA (a')Sx(x')x lf ' alii

(x)(a)

L C(SA(a")) 1 SA(a')Sx(x')x lf alii

(x)(a)

c(x)L lSA(a' )la"

(a)
c(x)c(a)l 1
c(xa)ll.
D
 IX.3 Variations on the Adjoint Representation 207

We conclude this section with three examples.

Example 1. As already observed, if (H, K) is a matched pair of groups,
then the pair (k[H], k[K]) is a matched pair of bialgebras. Furthermore,
the group algebra of the bicrossed product is isomorphic to the bicrossed
product of the group algebras
k[H [Xl K] ~ k[H] [Xl k[K].

Example 2. (Tensor product of bialgebras) Let X and A be bialgebras. We

let each one act trivially on the other one by
a x = c(a)x and aX = c(x)a
for all a E A and x EX. It is easy to see that these trivial actions satisfy
the conditions of Definition 2.2. In particular, both sides of Relation (2.5)
are equal to x0a. The formulas given in Theorem 2.3 show that in this case
the bicrossed product X [Xl A is isomorphic to the tensor product bialgebra
X0A.
Example 3. (Crossed product of bialgebras) This notion is parallel to the
semidirect product of groups. Let X and A be bialgebras. Suppose firstly
that X acts trivially on A as in Example 2, namely that aX = c(x)a for
all a E A and x EX, secondly that A acts on X via a map a which turns
X not only into a module-coalgebra, but also into a module-algebra, and
thirdly that we have the compatibility relation

L a' 0 a" . x = L a" 0 a' . x, (2.8)

(a) (a)

which is satisfied, for instance, when A is cocommutative. Then X and A

are matched bialgebras, and the corresponding bicrossed product is called
the crossed product of A by X. The multiplication in the crossed product
is given by
(x 0 a)(y 0 b) = L
x(a' . y) 0 a"b. (2.9)
(a)

IX.3 Variations on the Adjoint Representation

Let (H, /1, 'T], fl, c, S) be a Hopf algebra. If a and x are elements of H, we
set
L
a x = a'xS(a") and x a = L
S(a')xa". (3.1)
(a) (a)

Proposition IX.3.1. The map (a, x) f---+ a x endows H with the structure
of a left module-algebra on the bialgebra H. We denote by adH the thus
208 Chapter IX. Drinfeld's Quantum Double

defined H -module, and we call this action the left adjoint representation
of H. Similarly, the map (x, a) f-+ x a endows H with the structure of a
right module-algebra on the bialgebra H. We denote by Had the H-module
defined this way, and we call this action the right adjoint representation of
H.

PROOF. We give the proof for the left adjoint representation. We first check
that (a,x) f-+ a x puts an H-module structure on H. Indeed, we have
1 x = x and

b (a x) = L b'a'xS(a")S(b") = L (ba)'xS((ba)") = (ba).r

(a) (b) (ba)

for all a, b, x E H. Let us show it is a module-algebra over H. We have

a 1 = L a'S(a") = c(a)1
(a)

and

L (a' . x) (a" . y) L a'xS(a")a///yS(a"")

(a) (a)
L a'xc(a")yS(a///)
(a)
L a'xyS(a")
(a)
a (xy).

o
Example 1. (Conjugacy in a group) Let G be a group and k[G] be the
corresponding Hopf algebra. The left adjoint representation of k[G] is given
by the formula
a x = axa- 1
for a,x E G.
Example 2. (Adjoint representation of a Lie algebra) Let L be a Lie alge-
bra and U(L) be its enveloping algebra equipped with its canonical Hopf
algebra structure (see V.2). The left adjoint representation of U(L) is given
by the formula
a x = ax - xa
for a, x E L. The corresponding representation of L is called the adjoint
representation of the Lie algebra L.
We now wish to deduce the so-called coadjoint representations of H on
the dual vector space H* from the above-defined adjoint representations.
We use the following lemma.
 IX.3 Variations on the Adjoint Representation 209

Lemma IX.3.2. Consider a Hopf algebra H with invertible antipode S

and an algebra A that is a left [resp. right] module-algebra over H. Let us
put on the dual vector space A * the left [resp. right] H -module structure
given by
< a, xf > = < S-l(x)a, f > [resp. < a,fx > = < as-1(x), f > ]
for all a E A, x E H, and f E A*. If A is finite-dimensional, then the
coalgebra (AOP)* is a module-coalgebra over H.
The comultiplication on the finite-dimensional coalgebra (AOP)* is the
opposite comultiplication of the dual coalgebra A *; in other words,

< ab, f > = L < b, f' > < a, f" > (3.2)
(f)

whenever a, bE A and f E (AOP)*.

PROOF. Checking that A* is a left H-module is easy. Let us show that the
left action of H on A* defines an H-module-coalgebra structure on A*. It
suffices to check that the map from H 0 A * to A * which defines the action
of H on A * is a co algebra morphism. More precisely, we should have
c(xf) = c(x)c(f)
and
L (xf)' 0 (xf)" = L x' f' 0 x" r (3.3)
(x!) (x)(f)
Now,
c(xf) = (xf)(l) = f(S-l(X)I) = c(S-l(x))f(l) = c(x)c(f)
since xl = c(x)l. Let us check (3.3) by evaluating both sides on an element
a 0 b in A 0 A. We have

< a 0 b, L (xf)' 0 (xf)" > L < a, (xf)' >< b, (xf)" >

(x!) (x!)
< ba,xf >
< S-l(x)(ba), f >
L < (S-l(x)'b)(S-l(x)"a),f >
(x)
L < S-l(x")b, r >< S-l(x')a, f' >
(x)(f)

L < a,x'f' >< b,x"f" >

(x)(f)

< a 0 b, L x' f' 0 x" f" > .

(x)(!)

One proceeds in a similar fashion for the right action. D

210 Chapter IX. Drinfeld's Quantum Double

As an immediate consequence of Proposition 3.1 and of Lemma 3.2, we

deduce
Corollary IX.3.3. Let H = (H, J.L, TJ, A, c, S, S-l) be a finite-dimensional
Hopf algebra with invertible antipode S. There is a unique left [resp. right]
H -module-coalgebra structure on the dual of the opposite Hopf algebra, that
is, on (HOP)* = (H*, A *, c*, (J.L 0P ) * ,TJ*, (S-l)*, S*), given for a, x E Hand
f E H* by
< a,x f > L < S-l(x")ax', f >
(x)
[resp. < a,r > = L < x"aS- 1 (x'),j >].
(x)

These actions will be called the left and right coadjoint representations
of H. Applying Corollary 3.3 to the Hopf algebra

(HCOP)* = (H*, (AOP)*, c*, J.L*,TJ*, (S-l)*,S*)

and using the natural identification between the bidual H** and H, we get
a right (HCOP)* -module-coalgebra structure on the Hopf algebra

H = (H, J.L, TJ, A, c, S, S-l).

By Corollary III.3.5, the Hopf algebra (HCOP)* is isomorphic via the map
S* to the Hopf algebra (HOP)* = (H*, A *, c*, (J.L 0P ) * ,TJ*, (S-l)*, S*). This
isomorphism induces a right action of (HOP)* on H. We summarize this
with the following statement.
Proposition IX.3.4. Under the hypotheses of Corollary 3.3, there exists
a unique right (HOP)* -module-coalgebra structure on H given for a E H
and f E H* by
af = L
f(S-l(a"')a')a".
(a)
PROOF. Let f, g E H* and a E H. By Corollary 3.3, the action of (HCOP)*
on H is given by

< af,g >= L < a,f"gS*(j') >.

(f)

Computing in (HCOP)*, we get

<af,g> = L <alll,j"><a",g><a',S*(j'
(f)(a)

L < S(a'),f' >< alll,f" >< a",g >

(f)(a)
 IX.3 Variations on the Adjoint Representation 211

L S(a')a''', f > a", g >

(a)
L < f(S(a')a"')a", g > .
(a)

Therefore, the right coadjoint action of (Heop)* on H is given by

af = L f(S(a')a"')a".
(a)

Composing with (S-l)*, we get a right action of (HOP)* on H given by

a(S-')*(f) = afos-' = L f(S-l(S(a')a"'))a" = L f(S-l(a lll )a')a".

(a) (a)

o
We now state the main result of this section. It will allow us to construct
Drinfeld's quantum double in the next section.

Theorem IX.3.5. Let (H, IL, 7),,0., E, S, S-l) be a finite-dimensional Hopf

algebra with invertible antipode. Consider the Hopf algebra

Let a : H @ X ---> X and (3 : H @ X ---> H be the linear maps given by

a(a @ f) = a f = L f(S-l(a")?a')
(a)

and
(3(a@ f) = af = L f(S-l(a"')a')a"
(a)
where a E Hand f EX. Then the pair (H, X) of Hopf algebras is matched
in the sense of Definition 2.2.

PROOF. In this proof we systematically use Sweedler's sigma notation (de-

fined in 111.1) as well as the definitions of a and (3, the counitality of E and
relations of the form L(a) a"S-l(a') = E(a). The question mark? serves
as a mute variable. Corollary 3.3 and Proposition 3.4 show that a and (3
endow each Ropf algebra with the structure of a module-coalgebra over the
other one. We yet have to check Relations (2.1-2.5) of Definition 2.2.
212 Chapter IX. Drinfeld's Quantum Double

Relation (2.1): For x E H, we have

< x, L (a'. f')(a"f" . g) >

(a)(f)

L (a' f')(x')(a"f" . g)(X")

(a)(f)(x)

L f'(S-1(a(2))x'a(1))!"(S-1(a(5))a(3))(a(4). g)(X")
(a)(f)(x)

L f' (S-l (a (2) )x' a(1))!" (S-l (a(6))a (3) )g(S-l (a(5) )x" a(4))
(a)(f)(x)

L 1(S-1 (a(6) )a(3) S-l (a(2) )x' a(1) )g(S-l (a(5) )x" a(4))
(a)(x)

L c( a (2))1(S-1 (a(5))x' a(1))g(S-l (a(4) )x" a(3))

(a) (x)
L 1(S-l (a'II')x' a')g(S-l (a"') x" a"),
(a)(x)

which proves Relation (2.1).

Relation (2.2): We have

a c = L c(S-l(a")?a')c(a)c L c(a')c(a")c = = c(a)c.

(a) (a)

Relation (2.3): We have to show that

( ab ) f = '""""
~ ab't' bIIf" .
(b)(f)

Now,

L a b'f b" f" L f' (S-l (b")S-l (a'")a' b')!" (S-l (b" lll )bill) a"b""
(b)(f) (a)(b)(f)

= L 1(S-1 (b""')b'" S-l (b")S-l (alll)a'b')a"b""

(a) (b)

= L c(bl)J(S-l(bll)S-l(a'l)alb')alb"1
(a) (b)

= L 1 (S-l (b lll )S-l (a'lI)a'b') a"b"

(a) (b)

= L J(S-l((ab)II/)(ab)') (ab)"
(ab)

= (ab)f.
 IX.4 Drinfeld's Quantum Double 213

Relation (2.4): We have If = f(l) = E(f)1.

Relation (2.5): We have to check that

L a'l' 161 a" . f" = L a"f" 161 a' . 1'. (3.4)

(a)(f) (a)(f)

For the left-hand side of (3.4) we get

L a' I' 161 a" . f" L 1'(S-l(a ll )a' )a" 161 fl(S-l(alll)?a"l)
(a)(f) (a)(f)

L a" 161 f(S-l(alll)?a"1 S-l(a lll )a' )

(a)

L E(a'")a" 161 f(S-l(a" I )?a' )

(a)

L a" 161 f(S-l(a lll )?a' )

(a)(.f)

whereas for the right-hand side we have

'\"'
L a
II f" I
a fI L f"(S-l(alll)all)a"1 161 1'(S-l(a")?a' )
(a)(f) (a)(f)

L a"" 161 f(S-l(a"ll)a'l S-l(a")?a ' )

(a)

L E(a")a'" 161 f(S-l(a" I )?a' )

(a)
L a" 161 f(S-l(a lll )?a' ),
(a)

which proves (2.5). o

IX.4 Drinfeld's Quantum Double

Let (H, fL, TI,~, E, S, S-l) be a finite-dimensional Hopf algebra with invert-
ible antipode S. Let X = (HOP)* = (H*, ~ *, E, (fLO P)*, TI, (S-l)*, S*) be the
dual Hopf algebra. We have just proved that (H, X) is a matched pair of
Hopf algebras.

IX.4.1 The quantum double as a Hopf algebra

Definition IX.4.1. The quantum double D(H) of the Hopf algebra H is
the bicrossed product of H and of X = (HOP)*:

D(H) = X l><I H = (HOP)* l><I H.

214 Chapter IX. Drinfeld's Quantum Double

We first give a more explicit description of D(H), then in the next sub-
section we prove that the quantum double is a braided Hopf algebra in the
sense of VIII.2.
As a vector space, we have D(H) = X 0 H. The unit of D(H) is 10 l.
Its counit and its comultiplication are given by

c(f 0 a) = c(a)l(l) (4.1)

and
6.(f 0 a) = L (f' 0 a') 0 (f" 0 a") (4.2)
(a)(J)

where 1EX and a,b E H.

Lemma IX.4.2. The multiplication in D(H) is given by

(f 0 a)(g 0 b) = L 1 g(S-l(a"')?a') 0 a"b (4.3)

(a)
where l,g E X and a,b E H.
Here g(S-l(a lll )?a') means the map x f--+ g(S-l(a"')xa').
PROOF. By definition of the bicrossed product, the product of D(H) is
given by
(f 0 a)(g 0 b) = L 1 (a' . g') 0 a"gll b.
(a)(g)

Computing the right-hand side using the formulas of Theorem 3.5, we get

L 1 g'(S-l(a")?a') 0 g"(S-l(a"lII)a"')a""b
(a) (g)

L 1 g(S-l(a""')a'" S-l(a")?a') 0 a""b

(a)
L c(a") 1 g(S-l(a"")?a') 0 a"'b
(a)
L 1 g(S-l(a lll )?a') 0 a"b.
(a)

The quantum double D(H) contains H and X as Hopf subalgebras via

the embeddings i Hand i x given by

iH(a) = 10 a and ix(f) = 10l.

Formula (4.3) implies that
(4.4)

for all 1E X and a E H.

 IX.4 Drinfeld's Quantum Double 215

We shall use Relation (4.4) in order to simplify our notations and write
f a instead of f Q9 a = i x (f)i H (a) whenever a is an element of Hand f is
a linear form on H. Under this convention, the multiplication in D(H) is
determined by the straightening formula

af = L f(S-l(a"')?a')a" (4.5)
(a)

where f E X and a E H.
When H is cocommutative, the bicrossed product construction of the
double of H can be reduced to the crossed product construction of Section
2, Example 3, as shown in the following statement.

Proposition IX.4.3. Let H be a cocommutative finite-dimensional Hopf

algebra with invertible antipode. Then the quantum double D(H) is iso-
morphic, as a Hopf algebra, to the crossed product of H with (HOP)*, the
first algebra acting on the second one by the left coadjoint representation of
Corollary 3.3.

PROOF. We first have to prove that we are in the situation of Example 3

of Section 2, namely that (HOP)* acts trivially on H and that (HOP)* is a
module-algebra over H for the left coadjoint representation. The compati-
bility condition (2.8) is trivially satisfied since H is cocommutative.
Resuming the notations of Proposition 3.4 and using the cocommutativ-
ity of H, we have

af L f(S-l(a"')a')a"
(a)

L f(S-l(a"')a")a'
(a)

L f(l)s(a")a'
(a)

s(f)a,

which proves that (HOP)* acts trivially on H.

In order to prove that (HOP)* is a module-algebra over H, we have to
check that

a 1 = s(a)l and a (fg) = L (a f)(a g)

(a)

for a E Hand f, g E (HOP)*. This is left to the reader.

Let us now prove that the multiplication in D(H) coincides with the
multiplication of the crossed product given in (2.9). For a E Hand f E H*,
216 Chapter IX. Drinfeld's Quantum Double

we have the following equalities in the quantum double:

af = L f(S-l(al')?a')a"
(a)

L f(S-l(a'I)?a")a'
(a)

L (a" f)a'
Ca)
L (a' . f)a".
Ca)

Here we used the cocommutativity of H in the second and fourth equalities

as well as the definition of the coadjoint representation (see Corollary 3.3).
The last term of this series of equalities is the multiplication formula given
in (2.9) for the crossed product algebra. The coalgebra structures coincide
for both constructions. 0

IX.4.2 Description of the universal R-matrix

Let us consider the map AH H : HX --> End(H) defined in II.2 for a, bE H
and f E X by AH,H(a f)(b) = f(b)a. Since H is finite-dimensional, the
map AH,H is an isomorphism, which allows us to set

We define the universal R-matrix of the quantum double as the element

We get a more explicit formula for R by choosing a basis {e i LEI of the

vector space H together with its dual basis {e i LEI in X. Then

iEI iEI

We state the main theorem of this section.

Theorem IX.4.4. Under the previous hypotheses, the Hopf algebra D(H)
equipped with the element R = ~iEI (1 e i ) (e i 1) E D(H) D(H) is
braided.

PROOF. We have to prove that R satisfies the conditions of Definitions

VIII.2.1-2.2. More precisely, we must prove
(1) that R is invertible in D(H) D(H),
(2) that (.6. id)(R) = R 13 R 23 and (id .6.)(R) = R 13 R 12 , and
 IX.4 Drinfeld's Quantum Double 217

(3) that for all f E X and a E H, we have

(1) We claim that R is invertible with inverse R equal to

R= L(10ei)0((eioS)01).
iEI
Consider an element ~ = b 0 U 0 C 0 v in H 0 X 0 H 0 X. Let us pair it
with RR using the duality between H and X. We get

i,jEI
c(b)v(l) L u( (L ei(c')) (L e (S(c j ll
))) eiej)
(c) iEI jEI

c(b)V(l)U(L c'S(c"))
(c)
c(b)v(l)c(c)u(l)
< 1 0 1 0 1 0 1, ~ > .
Consequently, RR = 1 0 1 0 1 0 1. One proves that R is a left inverse of R
in a similar way.
(2) We now check that (Do 0 id)(R) = R13R23 or, equivalently, that

L 10 e~ 010 e~' 0 ei 01 = L 10 ei 010 e j 0 eie j 01.

iEI,(ei) i,jEI

Let us evaluate the left-hand side on an element f) = a 0 t 0 b 0 U 0 C 0 v

of the tensor product D(H) 0 D(H) 0 D(H). We have

< (Do 0 id)(R), f) > = < L 10 e~ 010 e~' 0 e i 0 1, f) >

iEI,(ei)
c(a)c(b)v(l)( L ei(c)t(e~)u(e~')).
iEI,(ei)

We now remark that

La' 0 a" = L ei(a) e~ 0 e~' (4.7)

(a) (a), iEI,(ei)

by application of the coproduct of H to a = LiEI ei(a) ei . Using (4.7), we

obtain
L ei(c)t(e~)u(e~') = L t(c')u(c").
iEI (c)
218 Chapter IX. Drinfeld's Quantum Double

Therefore,

< (~ id)(R), () >= L c(a)c(b)v(l)t(e')u(e").

(c)

On the other hand, we have

i,jEI

c(a)c(b)v(1) L t(L ei(e')e i ) u(L ej(ell)e j )

(c) iEI JEI

c(a)c(b)v(l) L t(e')u(e")
(c)

< (~ id)(R), () > .

One checks similarly that (id ~)(R) = R 13 R 12 .

(3) Let us evaluate ~oP(f a)R on ~ = b u e v. We have
< ~oP(fa)R,~ >
L < (f" a")(l ei ) (f' a')(e i 1), ~ >
(a)(f),iEI

L < f" a""e i 1'ei (S-l(a"')?a') a",~ >

(a)(f), iEI

L 1" (b )u( a"" ei )l' (e')e i (S-l (a"') e" a')v( a")
(a)(c)(f), iEI

L f(be')u( a"" ei (S-l (alii) e" a')ei)v( a")

(a)(c), iEI

L f(be')u( a"" S-l (a"')e" a')v( a")

(a) (c)

L c(a"')f(be')u(e"a')v(a")
(a)(c)

L f(be')u(e"a')v(a").
(a) (c)
On the other hand, we have
< R~(f a),~ >
L < (1 ei)(f' a') (e i 1)(f" a"), ~ >
(a)(f),iEI

L < 1'(S-l(en?e~) e~'a' ei1" a",~ >

(a)(f),iEI,(ei)
 IX.4 Drinfeld's Quantum Double 219

L l' (S-1 (et)be~)u( e~' a')e i (e')J" (e")v( a")

(a)(c)(f), iEI,(ei)

L f(c" S-1(enbeDu(e~' a')ei(c')v(a").

(a)(c), iEI,(e,)

L e' e" e'" = L ei(c) e~ e~' et (4.8)

(c) iEI,(ei)

Using (4.8), we obtain

< R/1(fa),~ > = L f (e''''S-1 (e"') be') u(e"a')v(a")

(a)(c)

L c( e"')f(be')u( e" a')v( a")

(a)(c)

= L f(be')u(e"a')v(a")
(a)(c)

< /1oP(f a)R, ~ > .

o

IX.4.3 Quantum double of a group algebra

We end this section by applying the quantum double construction to the
finite-dimensional co commutative Hopf algebra k[G] where G is a finite
group. By Proposition 4.3 we know that D(k[G]) is a crossed product.
Let {e9}9EG be the dual basis of the basis {g}9EG of k[G]. It is easy to
check that the dual algebra (k[G]OP)* is the algebra kG with multiplication
given by
egeh = Ogh eg (4.9)
for all g, h E G and with unit L:gEG eg = 1. The comultiplication /1, the
counit c, and the antipode S of (k[G]OP)* are defined by

/1(eg) = L ev eu ' c(eg) = Og1' S(eg) = eg-l (4.10)

uv=g

for each element g of the group.

The above description of a quantum double shows the set {egh hg,h)EGXG
is a basis of D(k[G]). The product of the quantum double is determined by

(4.11)
220 Chapter IX. Drinfeld's Quantum Double

which proves again that D(k[G]) is the crossed product of k[G] by itself,
where the algebra acts on itself by conjugation. Its universal R-matrix is

R= L g0 e g . (4.12)
gEG

Despite the fact that the quantum double is not cocommutative when G is
not abelian, its antipode is involutive, which implies that the element

u = L eg_,g (4.13)
gEG

introduced in VIlI.4 is central in D(k[G]). We also have

S(u) = u. (4.14)

IX.5 Representation-Theoretic Interpretation of

the Quantum Double
Let H = (H, /1, 'TJ,~, E:, S) be a finite-dimensional Hopf algebra with invert-
ible antipode. Again we choose a finite basis {aJiEI of H along with its
dual basis {aiLEI' The purpose of this section is to characterize modules
over the quantum double D(H). In view of Relation (4.5), a D(H)-module
is nothing but a vector space V with a left H-module structure as well as
a left H* -module structure such that for all a E H, f E H*, and v E V we
have
a(Jv) = L f(S-l(a lll )?a') (aI/v). (5.1)
(a)
We wish to rephrase such data purely in terms of H without any reference
to the dual algebra H*. We first introduce the following concept.
Definition IX.5.l. A crossed H -bimodule is a vector space V together
with linear maps /1v : H 0 V ---7 V and ~v : V ---7 V 0 H such that
(i) the map /1v [resp. ~v] turns V into a left H-module [resp. into a
right H -comodule], and
(ii) the diagram
fl.0id v
H0H0V
lfl.0fl. v 1idH0!lv

H0H0V0H H0V
1 idH0TH,v0idH lTH,v

H0V0H0H V0H
1!lV0!l 1fl.vidH

V0H V0H0H
commutes.
 IX.5 Representation-Theoretic Interpretation of the Quantum Double 221

Set ILv(a Q9 v) = av and ~v(v) = I:(v) Vv Q9 vH for a E H and v E V.

Then, according to the conventions set up in III.1 and III.5, the commuta-
tivity of the above diagram is equivalent to

'\'""
~ a I Vv Q9 a /I VH = '\'""
~
(
a 1/)
v v Q9 (1/)
a v H aI (5.2)
(a) (v) (a)(v)

where a runs over all elements in H and v over all elements in V.

We state the main result of this section.

Theorem IX.5.2. Let H be a finite-dimensional Hopf algebra with invert-

ible antipode. Any left D(H)-module has a natural structure as a crossed
H -bimodule. Conversely, any crossed H -bimodule has a natural structure
as a left module over the quantum double D (H).

PROOF. (a) Let V be a left D(H)-module. As we mentioned before, the

space V is a left H-module as well as a left H* -module satisfying Relation
(5.1). We wish to show that V can be endowed with a crossed bimodule
structure.
Given a basis {aJ i of H and its dual basis {a i } i' we use the left action
of H* on V in order to define a map ~v : V -+ V Q9 H by

(5.3)

for any v E V. Let us show that ~v defines a right coaction of H on V. We

have to check that ~v is coassociative and counitary. Rather than verify
this directly, we observe that ~v is the transpose of the associative, unitary
right action V* Q9 H* -+ V* of H* on V* given by

< aj, v > = < a, jv >

for a E V*, v E V and f E H*. Indeed, we have

2: a(aiv)j(ai )
< a, (2: j(ai)ai)v >
< a, fv >
< aj,v >
since j = I:i j(ai)a i . Incidentally, this observation implies that ~v is
independent of the choice of the basis.
In order to complete the proof that V is a crossed H-bimodule, we have
to check Relation (5.2) using (5.1). If a E H, v E V and j E H*, then
222 Chapter IX. Drinfeld's Quantum Double

(id (9 f) ( L a'vv (9 a"v H) (id (9 j) (L a'(aiv) (9 a" ai )

(a) (v) (a),i

L a'(aiv)f(a"a i )
(a),i

L 1'(ai )1"(a")a'(a iv)

(a)(f),i

L 1"(a")a'((L f'(ai)ai)v)
(a) (f) .

L 1" (a")a' (I'v )

(a)(f)

L 1" (a"") l' (S~ 1 (a"')? a') (a" v)

(a)(f)

L f(a""S~l(a"')?a')(a"v)
(a)
L c(a"')f(?a') (a"v)
(a)
L f(?a') (a"v)
(a)
L 1'(a') f"a"v
(a) (f)

L ai (a" v)1" (ai)1' (a')

(a)(f),i

L ai(a"v)f(aia')
(a),i

(id (9 j) (L ai(a"v) (9 aia')

(a),i

(id (9 j) ( L (a"v)v (9 (a"v)H a').

(a)(v)

This, being true for any linear form f, implies (5.2). In the previous series
of equalities, we used the comultiplication on H*, Relation (5.1), the fact
that S~l is a skew-antipode, that c is a counit, and that f = Li f( ai)a i.
(b) Conversely, let V be a crossed H-bimodule. We show that V can be
given a D(H)-module structure. Observe that if (V, ~v : V ---> V (9 H) is
a right H-comodule, then V becomes a left module over the dual algebra
X = H* by
 IX.6 Application to Uq (.s1(2)) 223

where ev H is the evaluation map. In other words, a linear form f E H*

acts on an element v E V by

f v = L < f,vH > vv' (5.4)

(v)

In view of this observation, we see that a crossed bimodule has a left H-

action as well as a left H* -action. In order to prove V is a D(H}-module,
it is enough to check Relation (5.1). We have

L f(S-1(a"'}?a'} . (a"v) L < f, S-1 (a"') (a"v}H a' > (a"v}v

(a) (a)(v)

L < f, S-1(a"'}a" vH > a'vv

(a)(v)

L c(a") < f, vH > a'vv

(a)(v)

L < f,v H > avv

(v)

aU v}

for any a E H, f E H* and v E V. The second equality is a consequence

of Relation (5.2). The third one follows from the fact that S-1 is a skew-
antipode. 0

Remark 5.3. Formula (5.3) defining the coaction ~v may be rewritten as

(5.5)

where R21 is obtained from the universal R-matrix of D(H} by applying

the flip. We shall use Relation (5.5) in order to determine the universal
R-matrix of Uq (.s1(2}} in XVII. 4.

IX.6 Application to Uq (s((2))

We now return to the Hopf algebra Uq = Uq (.s1(2}} studied in detail in
Chapters VI-VII. We wish to show that it has a universal R-matrix using
the quantum double construction of Section 4. However, we only gave this
construction for finite-dimensional Hopf algebras, which is not the case of
Uq . Therefore we postpone the construction of the universal R-matrix of
Uq to Chapter XVII. Instead, we now work with the finite-dimensional
quotient U q introduced in VI.5.
224 Chapter IX. Drinfeld's Quantum Double

We assume until the end of this chapter that q is a root of unity of order
d in the field k where d is an odd integer > 1. Let us resume the notation
of VL1. Recall
qn _ q-n
[n] = q _ q-l '
which is defined for any integer n, and the corresponding q-factorials [n]!.
We have [n] 1= 0 if 0 < n < d and [d] = O.
In VL5 we defined the algebra U q as the quotient of Uq by the two-
sided ideal generated by the three elements Ed, F d, Kd - 1. We proved in
Proposition VL5.7 that the finite set {EiFjK}o::;i,j,e::;d_l was a basis of
the underlying vector space of U q . We endow the algebra U q with a Hopf
algebra structure.

Proposition IX.6.1. The algebra U q has a unique Hopf algebra structure

such that the canonical projection from Uq to U q is a morphism of H opf
algebras.
In other words, the comultiplication, the counit and the antipode of
U q are determined by Formulas (VIL 1. 1-1.4) defining the Hopf algebra
structure of Uq .
PROOF. The proof proceeds as for Proposition VILLI. We still have to
check that
b.(E)d = b.(F)d = b.(K)d - 1 = 0,
c:(E)d = c:(F)d = c:(K)d - 1 = 0,
S(E)d = S(F)d = S(K)d - 1 = o.
The only non-trivial computations concern the vanishing of b.(E)d and of
b.(F)d. Following Proposition VIL1.3, we get
r=d-l
b.(E)d = Ed Kd + ~ qr(d-r) [ ; ] E d- r E r K d- r + 1 Ed = 0

because Ed = 0 on one hand and

[ d ] _ [d]! _ 0
r - [r]![d - r]! -

on the other. One proves that b.(F)d = 0 in a similar way. D

The goal of this section is to establish that U q is a braided Hopf algebra.

To this end, we shall present U q as a quotient of the quantum double of
a Hopf subalgebra Bq of U q. We define Bq as the subspace of U q linearly
generated by the set {EmKnJo::;m,n::;d_l. Formulas (VIL1.1-1.4) show that
 IX.6 Application to Uq (s((2)) 225

Bq is a Hopf subalgebra of U q. The reader may check that, as an algebra,

B q is generated by E and K and the relations

(6.1)
We now apply the quantum double construction of Section 4 to the
Hopf algebra H = B q . We first determine X = (B~P)* as a Hopf al-
gebra. Consider the linear forms a and ry on Bq defined on the basis
{EmKnh:S:m,n:S:d_I by
< a, Em K n > = DmO q2n and < ry, Em K n > = DmI . (6.2)
Proposition IX.6.2. The following relations hold in the Hopf algebra X:
ad = 1, ryd = 0, arya- I = q-2ry,

~(a) =a a, ~(71) = 1 71 + 71 a,
c:(a) = 1, c:(ry) = 0,
8(a) = ad-I, 8(71) = _rya d- I .
Moreover, the set {ryiaj}O:S:i,j:S:d-I forms a basis of X.
PROOF. We start with the following lemma.
Lemma IX.6.3. For all integers i, j, m, n, we have
< ryia j , Em K n > = 8mi (i)!q2 q2 j (i+n).
PROOF. By Proposition VII.1.3, if a and j3 are linear forms on H, then the
product a(3 in X is given by

< aj3, Em K n > = rf ( r; )

r=O q
2 < a, E m- r K n >< (3, E r K m+n - r >.
(6.3)
One uses (6.3) to show that

< 71i , EmK n >-

-8mi (')'
Z 'q2

by induction on i, and that

< a j , Em K n >= 8mO q2 j n
by induction on j. Then

r~ r; ) ( q2

%(r; q2 ) 8m-r,t8rO (i)''q2 q2 j (m+n-r)

= 8mz. (,;)' n
'q2 q2 (i+ )
j o
226 Chapter IX. Drinfeld's Quantum Double

Let us prove Proposition 6.2.

(1) Using the previous lemma and qd = 1, we get

Therefore, ad = c is the unit of X. Analogously,

since (d)q2 = (q2d - 1)j(q2 - 1) = O. As for arJ, we have

In view of Lemma 6.3 we can write

whence rJa = q2arJ

(2) Let us deal with the comultiplication of X. By definition, if a is a
linear form on H, then ~(a) is given by ~(a)(x y) = a(yx) for x, y E H.
Therefore

q2ni < a, EHm Kj+n >

D. q2niq2(j+n)
.+m,O
D..0 DmO q2 j q2n
< aa,EiKj EmK n >,

which implies that ~(a) =a a.

Similarly, we have

q2ni < rJ, E Hm Kj+n >

DHm,l q2ni
DiO Dml + Dil DmO q2n
< 1 rJ+rJ a,EiKj EmK n > .
Consequently, ~(rJ) = 1 rJ + rJ a.
(3) Concerning the counit, we have

c(a) =< a, 1 >= 1 and c(rJ) =< rJ, 1 >= O.

The computation of 8(a) and of 8(rJ) is left to the reader.

 IX.6 Application to Uq (st(2)) 227

(4) Let us prove the last assertion of Proposition 6.2. As the dimension
of X is d 2 , it is enough to show that the set {1]i o:1}O::;i,j::;d_l is linearly
independent. Suppose there exists a relation of the form

L Aij1]i a j = O.
O::;i,j::;d-l

Applying it to the vector Em Kn, we get

L AijDmi(i)!q2q2j(Hn) = (m)!q2 ( L Amjq2j(m+n)) = O.

O::;i,j::;d-l O::;j::;d-I

Letting m fixed and running n over the d integers between 0 and d - 1, we

obtain a system of d linear equations with unknowns AmO' AmI"'" Am d-I'
The determinant of this system is the determinant of the matrix (A~l)kf!
defined by Akf! = (q2(m+))k. It is a Vandermonde determinant which does
not vanish because q2(m+) ! q2(m+') whenever 0 ::::; .e ! .e' ::::; d - 1.
Therefore, the system has 0 as its unique solution; in other words, we have
Amj = 0 for all j. 0

We now construct the quantum double D = D(Bq). By definition, the

set {1] i a j 129 Ek Kflo::;i,j,k,::;d_I is a basis of D. To simplify notation, we
identify an element x of H = Bq with its image 1129 x in D and an element
a of the dual X with its image a 129 1. Under the convention already set
up in Section 4, the elements of the previous basis can be rewritten in the
form
1]i a j 129 Ek Kf = 1]i a j Ek Kf.

To determine the multiplication of the double D, it is enough to know how

the generators a, 1], E, K in D multiply together.

Proposition IX.6.4. The following relations hold in D = D(Bq):

Ea = q-2 aE, E1] = _q-2(1 -1]E - aK).

PROOF. By (4.5) the product xa in D of x E H of a E X is given by

xa =L a(S-l(x"')?x') x".
(x)

Let us apply this formula to the generators. First, we have S-I(K) = K- I

and (Lll29id) (Ll(K)) = KI29KI29K. Consequently, for any linear form (3 E X
we have
(6.4)
228 Chapter IX. Drinfeld's Quantum Double

Next, S-l(E) = _K- 1E and

(ll @ id)(ll(E)) = 1 @ 1 @ E +1@ E @ K +E @ K @ K.

Hence,
E(3 = -(3(K- 1E?) + (3(K- 1?) E + (3(K- 1?E) K. (6.5)
Proposition 6.4 is then a consequence of (6.4-6.5) and of the following
lemma. 0

Lemma IX.6.5. We have

a(K- 1?K) a, a(K-1E?)
a(K-1?) q-2 a, a(K- 1 ?E)
T)(K- 1?K) q-2 T), T)(K- 1E?)
T)(K- 1?) q-2 T), T)(K- 1?E)

PROOF. Left to the reader. o

We now relate the quantum double D(B q) and the Hopf algebra U q'

Theorem IX.6.6. Let X : D(Bq) -7 U q be the linear map determined by

where 0 <::: i, j, k, ji <::: d - 1. Then X is a surjective Hop! algebra morphism.

PROOF. The surjectivity of X follows from the fact that the image of the
basis {T)ia J Ek Ki} generates U q'
In order to show that X is a map of algebras, it is enough to check that
the images under X of the generators E, K, a, T) satisfy the relations of
Proposition 6.4. Observe that (6.6) implies

X(E) = E, X(K) = K,
q _ q-1
x(a) = K, X(T)) = 2 FK.
q
Now, by definition of U q we have

x(K)x(a) x(a)x(K),
q _ q-l
X(K)x(T)) ~C::-2-KFK=q-2 X(T))X(K),
q
x(E)x(a) q-2 x(a)x(E).
 IX.6 Application to Uq (s (( 2) ) 229

Finally, we get
-1
X(E)x(TJ) q-q EFK
q2
q - q-1 FEK + ~ (K _ K-1)K
q2 q2

'1
-1
_q-2(1_ q - q FKE-K 2)

_q-2 (1 - X(TJ)x(E) - x(a)x(K)).

This proves that X is a morphism of algebras.

Again, to show that X respects the comultiplication and the antipode, it
is enough to check on the generators. For E, K, and a, this is clear. We
still have to examine the case of TJ for which we have
-1
~(X(TJ)) = q - q2q ~(FK)

-1
q- ; (1 Q9 F K + FK Q9 K)
q
+ X(TJ) Q9 x(a)
X(l) Q9 X(TJ)
(XQ9X)(~(TJ)).
Similarly,

X(S(TJ)) -X(TJa- 1 )
q _ q-1
q2 F
-1
q- ; S(K)S(F)
q
-1
q-q S(FK)
q2
S(X(TJ))
D

We draw the following consequence which was our main goal.

Corollary IX.6.7. The Hop! algebra U q is braided.

PROOF. The Hopf algebra D = D(B q ) is braided by Theorem 4.3. Let

RD ED Q9 D be its universal R-matrix. Define the invertible element R of
U q Q9U q by
(6.7)
230 Chapter IX. Drinfeld's Quantum Double

Since X is a surjective morphism of Hopf algebras, it is clear that R satisfies

Conditions (VIII.2.1-2.3). 0
We shall compute the universal R-matrix R of fJ q in the next section.

IX.7 R- Matrices for U q

We keep the hypotheses and the notations of Section 6.

Theorem IX.7.1. The universal R-matrix of U q is given by

(q -l)k
- q k(k-l)/2+2k(i-j)-2ij Ek Ki Fk Kj.
[k]! q
OSi,j.kSd-l

PROOF. According to Section 4, we have RD = LiEI e i e i where {e i LEI is

any basis of the vector space Bq and {ei}iEl is its dual basis. Consequently,
by (6.7)
R= L
x(e i ) x(e i ). (7.1)
iEI

As above, we take the set {EiKj}OSi,jSd-l as a basis of B q . Denote by

{;3i j }OSi,jSd-l the dual basis. By Proposition 6.2 we know there exist
scalars {t.l~RloSi,j,k,RSd-l such that

(7.2)

Apply Relation (7.2) to the vector Em K n: using Lemma 6.3, we obtain the
linear system of equations

lI ij
"'kR Ukm
I: (m)''q2 q2C(k+n)
OSk,CSd-l

(m)!q2 ( L f.L~R q2C(m+n)).

OSRSd-l

An argument similar to the one that proved the linear independence of the
family {r]i o:1}OSi,jSd-l in Proposition 6.2 shows that f.L~R = 0 for m i' i.
Computing the coefficients f.L~~ that are solutions of the linear system

lI ij q2R(Hn) = __1_ /5
"'~R (m)!q2 In

requires inverting a Vandermonde matrix. We shall not do this since we are

interested in R, not in RD' Instead, we shall use a simpler and more direct
method.
 IX.7 R-Matrices for U q 231

Indeed, from the above arguments we know that Ii is a tensor of the

form
R=
0<:;i,j,<:;d-1

Now x(r/c/) is a scalar multiple of piKiH. Therefore, R has the more

precise form

O<:;i,j,k<:;d-1

We now determine the coefficients Ci,j,k' Theorem 7.1 will follow from
Lemma 7.2. 0

Lemma IX.7.2. For all i, j, k we have

~ ~ (q - q-1)k qk(k-1)/2+2k(i- j )-2ij

Ci,j,k ~ d [k]! .

PROOF. (a) We first express Ci,j,k in terms of C = co,o,o using the relations

~OP(x)R = R~(x)

for x = E and x = P. We have

~OP(E)R = (E 1 + K E) ( L Ci,j,k Ek Ki pk Kj)

O<:;i,j,k<:;d-1

0<:;i,j,k<:;d-1

+
O<:;i,j,k<:;d-1

and

R~(E)

0<:;i,j,k<:;d-1

+
O<:;i,j,k<:;d-1

"L-t q2j C.
t,],k
Ek Ki Epk Kj
O<:;i,j,k<:;d-1
2j-(k-1)
"[k]q
o -1
c. EkKipk-1Kj+1
',],k
+
k<d-1
_ ,], -
O<i . q - q
232 Chapter IX. Drinfeld's Quantum Double

2j+(k-1)
+ '"
~
[k] q -1
c
,,),k
Ek Ki Q9 F k - 1 Kj-l
OSi,j,kSd-1 q - q

+
OSi.,j,kSd-1

Identifying the coefficients of Ek Ki Q9 EFk Kj, we get

_ 2(k-j)
Ci,j,k - q ci-l,j,k' (7.3)

Starting all over again with F, we get

OSi,j,kSd-1

+
OSi,j,kSd-1

OSi,j,kSd-1
2k+(k-1)
'~
" [k] q -1
c
',],k
E k - 1 K i + 1 Q9 Fk Kj-1
OSi,j,ksd-1 q - q
2k-(k-1)
+ '"
~
[k] q -1
C
',],k
E k - 1 K i - 1 Q9 Fk Kj-1
OSi,j,kSd-1 q - q

+ L
OSi,j,ksd-1

On the other hand, we have

R6.(F)

OSi,j,kSd-1

+ q -2i C. Ek F Ki rV. Fk Ki
,t,],k '6l

OSi,j,kSd-1

We identify the coefficients of Ek F Ki Q9 Fk Kj-1, which yields

C.
',],k
= q-2(k+i) C .
I,)-l,k' (7.4)

and the coefficients of Ek Ki Q9 Fk+1 Kj, which leads to

 IX.7 R-Matrices for U q 233

q2(k+1)+k
-2j
q CH1 ,j,k Ci,j,k - [k + 1J q _ q-1 Ci - 1,j+1,k+1

q2(k+1)-k
+[k + 1J q _ q-1 Ci +l,j+1,k+1' (7.5)

From (7.3) and (7.4) we get

Ci,j,O = q -2ij co,o,o = q -2ij c. (7.6)

Combining (7.3) and (7.5), we obtain
q 4j -k+2
q -2j+2(k- j ) c k
1,),
-- c1,),. k - [k + 1J-----:- . 1 ,k + 1
q _ q-1 C1+ 1 ,)+

qk+2
+[k + 1J q _ q-1 CH1 ,j+1,k+l'

hence
-1
q - q k-4j-2
CH1 ,j+1,k+1 = [k + 1J q Ci,j,k'

or, equivalently,
-1
q - q k-4j+l
Ci,j,k = [kJ q Ci - 1,j-1,k-1'

Therefore, we get
( -1)k
q - q qk(k+ 1)/2-4kj+2k(k-1)+k C
Ci,j,k = [kJ! i-k,j-k,O

( -1)k
q - q k(k-1)/2+2k(k-2j) -2(i-k)(j-k)
~! q q C

by (7.6). In other words, we have

( -1)k
c.. = C q - q qk(k-1)/2+2k(i- j )-2ij
1,),k [kJ! .
(b) It is now enough to prove that C = lid. From Part (a) we know that
R is of the form
R= C L q-2i j Ki Kj + ...
O~i,j<d

where + ... stands for a sum of monomials containing only positive pow-
ers of E or of F. We now use (.6. id)(R) = R 13 R 23 , which is Relation
(VIII.2.3). We have

(.6. id)(R) = C L q-2i j Ki Ki Kj +... (7.7)

O~i,j<d
234 Chapter IX. Drinfeld's Quantum Double

whereas

O~i,e,m,j<d

O~i,e,m,j<d

c2 L q-2m j (L q(2m-2i)) Ki 0 K m 0 Kj + ....

O~i,m,j<d o~e<d

Now, 2:0d
_J
qNj vanishes except when N is a multiple of d, in which case
the sum equals d. Therefore,

R 13 R23 = dc 2 L q-2i j Ki 0 Ki 0 Kj + .... (7.8)

O~i,j<d

We deduce from (7.7-7.8) that c = dc 2 . Since R is invertible, c does not

vanish. This implies the announced value for c. D

We conclude this section by deriving a few R-matrices from Theorem 7.l.

Let 0 < n < d. Consider the simple U q-module Vn = V1,n defined in Chap-
ter VI. As a module, it is generated by a highest weight vector v~n) of weight
qn. Recall that the action of Uq on the canonical basis {v6 n ), vi n ) , ... ,v~n)}
of Vn is given by the relations

Kv(n)
p = qn-2 p v(n)
p' Ev(n)
p = [n - p + 1] v(n)
p-l' Fv(n)
p = [p + 1] v(n)
p+ 1 .

We use them to deduce the form of the R-matrix

obtained from R via the construction of (VIII.3.1). Recall that c~n,v", is a

solution of the Yang-Baxter equation.

Corollary IX.7.3. The isomorphism c~n, v'" : Vn 0 Vm --+ Vm 0 Vn is the

U q-linear map determined by

cR (v(n) 0 v(m))
Vn,VTn P r

'"' (q - q-l)k [n - p + k]![r + k]! nm(k ) (m) (n)

L.t [k]! [n _ p]![r]! qpr ,a vr+k 0 v p_ k
O~k~d-l

where a is any integer such that m + ad is even and

q;;'(k, a) = qk(k-l)/2+k(m-n)-pm-rn-2(k- p)(k+r)+(m+ad)n/2.
 IX.7 R-Matrices for U q 235

PROOF. By definition of e~n, v:rn and in view of Theorem 7.1, we have

eR
Vn,Vm
(v(n)
p
v(m))
r
1 (q_q-I)k [n-p+k]![r+k]! k(k-I)/2
q
d [k]! [n - p]![r]!
O"Si,j,k"Sd-1
x Qnm(k) v(m) v(n)
pr r+k p-k

where
Q;;:'(k) = L q2(i-j)k-2ij+i(n-2p)+j(m-2r) ,
a"Si,j<d
which we can rewrite as

Q;;:'(k) = L q2ik+i(n-2 p) (L q(m-2r-2i-2k)j).

a"Si<d O"Sj<d

Again, L.:O"Sj<d qNj vanishes except when N is a multiple of d. Thus,

where i runs over the set of all integers in [0, d - 1] such that

2i = m - 2r - 2k + ad.
As 2 is invertible modulo d, there exists only one integer i satisfying these
conditions. Therefore,
Q;;:'(k) = d q 2ik+i(n-2 p) = d q k(m-n)-pm-rn-2(k- p)(k+r)+(m+ad)n/2. 0

Application 7.4. Consider the case n m 1. We may take a 1.

Corollary 7.3 implies that

e~,,v, (va va) )..qva Va,

e~,,v, (va VI) )..V I Vo,
e~,v, (VI Va) ).. (Va VI + (q - q-I)V I Va),
e~,v, (VI VI) )..qv I VI

where).. = q(d-I)/2, va = vb l ) and VI = v~I). The reader is invited to

compare these formulas with the R-matrices of VIlLI, Example 2.
236 Chapter IX. Drinfeld's Quantum Double

IX.8 Exercises
1. Let H be a bialgebra and C a co algebra. Prove that C is a module-
coalgebra over H if and only if there exists an H-module structure
on C such that the comultiplication ,0. : C ~ C @ C and the counit
c: C ~ k of Care H-module maps for the tensor product H-module
structure on C @ C and for the trivial H-module structure on k.

2. Let H be a bialgebra and C a coalgebra. Then C is a comodule-

coalgebra over H if there exists an H-comodule structure on C such
that the coproduct,0. : C ~ C@C and the counit c: C ~ k of Care
morphisms of H-comodules for the tensor product H-comodule struc-
ture on C@C and for the trivial H-comodule structure on k. Draw the
commutative diagrams expressing an H-coalgebra-comodule struc-
ture on C. Deduce that C is a comodule-coalgebra over H if and only
if there exists a linear map,0.e : C ~ H@C inducing an H-comodule
structure on C and satisfying for all x E Hand c E C the relations

L cH @ (cd' @ (cd" = L c~c'iI @ (c')e @ (c")e

(~ (~

and 2:(c) cHc(cd = c(c)l where ,0.c(c) = 2:(c) cH @ ce'

3. Let H be a bialgebra and C a coalgebra equipped with a comodule-
coalgebra structure on H. Show that the dual algebra C* can be given
a comodule-algebra structure on H.

4. Let H be a bialgebra and C a co algebra equipped with a module-

coalgebra structure on H. Show that if C is finite-dimensional, then
the dual algebra C* can be given a module-algebra structure on H.

5. (Adjoint corepresentation) Let H be a Hopf algebra. Define a linear

map ,0.ad from H to H @ H by

,0.ad(a) =L a'S(a"') @ a".

(a)

Prove that ,0.ad endows H with a comodule-coalgebra structure over

itself.

6. (Coadjoint corepresentation) Let H be a finite-dimensional Hopf alge-

bra with invertible antipode. Prove that the adjoint corepresentation
induces an H-comodule-algebra structure on the dual vector space
H*.

7. Let G be a finite group. (a) Show that a left module over the quantum
double D(k[G]) is a left G-module V with a decomposition of the form
V = EB9EG Vg such that hVg C Vhgh-l for all g, hE G.
 IX.8 Exercises 237

(b) Let W = EB9EG Wg be another left D(k[G])-module. Show that

the automorphism c R defined in VIII.3 sends Vg W h to W h Vhgh-l
as the map v W f---> W hv.

8. (Tensor product of crossed bimodules) Use Theorem 5.2 to define the

tensor product of two crossed bimodules.

9. Compute the central element uS(u) (defined in VIII.4) for U q .

10. Determine C~,v2 associated to the simple Uq-module V2 under the

form of a 9 x 9 matrix.

11. (A cobraided Hopf algebra structure on End(H)) Suppose given a

finite-dimensional Hopf algebra (H, jL, 'f},~, c, S, S-l) with bijective
antipode. Let E = End(H), and identify E E with End(H H).

(a) Prove that there exists a Hopf algebra structure on E for which
the product is the convolution of I1L3, the unit is 'f} 0 c, the
coproduct ~', the counit c', and the antipode S' are given by

~'(f)(x y) = L (1 x')~(f(yx"))(1 S(x"')),

(x)

c'(f) = c(f(I)), S'(f)(x) = L S(x')(SfS- 1 )(x")x'"

(x)
for all x, y E Hand fEE.
(b) Identifying E with H H* via the map ,\ H H of Corollary 11.2.3,
define maps PH : E ---+ Hand PH* : E ---+ iI*cop by

PH(X a) = a(l)x and PH*(X a) = c(x)a

where x E H and a E H*. Prove that PH and PH* are morphisms
of Hopf algebras such that the composition of the maps

EE E PH@PH* IH H*

. \-1
IS /lH,H.

(c) Check that the linear form r on E E defined by

for f, gEE equips E with a cobraided Hopf algebra structure.

(d) Show that the dual braided Hopf algebra E* is isomorphic to
Drinfeld's quantum double D(H).
238 Chapter IX. Drinfeld's Quantum Double

IX.9 Notes
The quantum double construction is due to Drinfeld [Dri87]. Our pre-
sentation is inspired from [Maj90a] [Tak81] (see also [RSTS88]). Radford
[Rad93a] proved that the quantum double is a minimal braided Hopf al-
gebra, i.e., it has no proper braided Hopf subalgebras. Conversely, any
minimal braided Hopf algebra is finite-dimensional and is a quotient of the
quantum double of some Hopf algebra. More generally, if H is braided with
universal R-matrix R, consider the subspace A of H generated by all ele-
ments of the form (idH 0 o:)(R) where 0: is any linear form on H. Radford
showed that the subspace A can be given the structure of a Hopf subalge-
bra, and that there exists a map of braided Hopf algebras from D(A) to H
whose image is a minimal braided Hopf subalgebra of H.
In Section 4 we proved that the quantum double of H was isomorphic to
a crossed product when H is cocommutative. This is true more generally
when H is braided. For more details, see [Maj91a].
Exercise 11 presents a construction dual to Drinfeld's quantum double,
yielding cobraided Hopf algebras. We took it from Takeuchi [Tak92a] where
a dual version of Theorem 5.2 is also given (see also [PW90] [RSTS88]).
The term "crossed bimodule" is due to [Yet90]. It was called a "quantum
Yang-Baxter module" in [Rad93b].
The Hopf algebra Uq has been considered in [Lus90a][Lus90b]. A compu-
tation of its universal R-matrix was performed in [KM91] using a different
method. It also appears in work by Reshetikhin-Turaev [RT91] constructing
quantum invariants for 3-dimensional manifolds.
 Part Three

Low-Dimensional
Topology and
Tensor Categories
Chapter X
Knots, Links, Tangles, and Braids

We now embark into a topological digression which will lead us into the
world of knots. The reason for the presence of this chapter in a book de-
voted to quantum groups is the close relationship between the newly dis-
covered invariants of links (such as the celebrated Jones polynomial) and
R-matrices. This relationship will become more precise in Chapter XII. In
this one we proceed to describe several classes of one-dimensional subman-
ifolds of the three-dimensional space R 3 , such as knots, links, tangles, and
braids. Since there are excellent textbooks on knot theory, we shall not
prove all assertions that can be found elsewhere. Nevertheless, all results
pertaining to the matter of this book, namely those connecting topological
problems with the algebra of quantum groups, will be proved in detail.
After defining knots and links in R 3 , we recall the classical problem of
their classification up to isotopy. Traditionally, one approaches this problem
by constructing algebraic isotopy invariants. One major step in this direc-
tion was undertaken in the 1920's by Alexander, who associated a polyno-
mial to each isotopy class of oriented links. The Alexander polynomial was
used to distinguish many links and has been a powerful tool in knot theory
since.
In the summer of 1984 Vaughan Jones found a different one-variable
polynomial which distinguished knots that the Alexander polynomial could
not distinguish [Jon85]. Shortly after, a new invariant appeared, the so-
called Jones-Conway polynomial, which is a two-variable generalization of
both the Alexander and the Jones polynomials. Another aim of this chapter
is to establish the existence and the main properties of the Jones-Conway
polynomial.
242 Chapter X. Knots, Links, Tangles, and Braids

X.I Knots and Links

Let us start with some vocabulary from general topology. The only topo-
logical spaces considered here are the real Euclidean vector spaces R n with
their standard topology as well as their subsets and quotients with the
induced topologies.
A continuous map f from a subset U of R m to a subset X of R n
is piecewise-linear if there exists a finite partition (Ui)i of U such that
the n components of the restriction of f to any Ui are maps of the form
(zl' ... , zm) f-+ a o + a 1 ZI + ... am zm where a o , aI' ... ,am are real numbers.
Let X be a convex topological subspace of the Euclidean space R 3 . In
the sequel, X will be either R 3 , R2, R2 X [0,1]' or R x [0,1]. Given a
finite sequence (MIl" ., MrJ of points in X, we denote by [M1 ,, Mn]
[resp.]Ml"" ,Mn[] their closed [resp. open] convex envelope, i.e., the set
of all points of the form Al Ml + ... + AnMn where (AI' ... , An) is a sequence
of real numbers 2': 0 [resp. > 0] such that Al + ... + An = l.

Definition X.LL A polygonal arc L in X is the union

n-l
L = U [Mi' Mi+l]
i=1

of a finite number of segments such that ]Mi ,Mi+l[ n ]Mj ,Mj + 1 [ = 0 if

i cJ j. The points M 1 , ... ,Mn are called the vertices of the polygonal arc
and the segments [Mi' M i +1] are its edges. We say that the polygonal arc is
simple if the points M 1 , ... ,Mn - 1 are all distinct. The polygonal arc L is
closed if Ml = Mn; in this case, we say that the boundary 8L is empty. If
Ml cJ M n , we set 8L = {Ml' Mn}; the point Ml is the origin of the simple
polygonal arc Land Mn is its endpoint.

By ordering the vertices of L we define an orientation on L. It will be

materialized in the figures by arrows on the edges such that on the edge
[Mi' Mi+l] the arrow points to M i + 1

Definition X.L2. A link L in X is the union of a finite number m of

pairwise disjoint simple closed polygonal arcs in X. The closed arcs are
called the connected components of L. The integer m is called the order of
the link. A knot is a link of order 1.

A link is oriented by giving an orientation to each of its connected compo-

nents. In the sequel we consider only oriented links. Following Reidemeister
[Rei32], we define a combinatorial operation ~ on links. We assume X to
be R 3 until the end of this section.

Definition X.L3. (a) Let L be a link in X and M i , Mi+l two consecutive

vertices in a connected component of L. Given a point N in X such that
 X.I Knots and Links 243

N 1:. L, Mi 1:. [N, MHIL MHI 1:. [Mi' NL and

[Mi' N, M H1 ] n L = [Mi , M H1 ],
we denote by L' the link
L' = (L \ [Mi' M H1 ]) U [Mi , N] U [N, M H1 ].
We say that L' is obtained from L by a f::l.-operation.
(b) Two links Land L' are combinatorially equivalent - we write this
L '" c L' - if there exist links L = L o, L 1 , ... ,L k = L' such that for
all i, one of the two links L i , LHI is obtained from the other one by a
f::l.-operation.
The relation", c is an equivalence relation: it is the equivalence relation
generated by the f::l.-operations.

Figure 1.1. A t:.-operation

It is also possible to deform links continuously. This leads us to the

concept of isotopy.
Definition X.lo4. (a) An isotopy of X is a piecewise-linear map h from
[0,1] x X to X such that, for any t E I, the mapping h(t, -) is a homeo-
morphism of X, and h(O, -) is the identity of X.
(b) Two links Land L' are isotopic - we write L '" i L' - if there exists
an orientation-preserving isotopy h of X such that h(l, L) = L'.
Lemma X.lo5. Isotopy is an equivalence relation for links.
PROOF. Let L, L', and L" be links. (a) Set h(t, -) = id x for all t E I. It is
clear that h is an isotopy between L and itself: L "'i L.
(b) Let us suppose that L '" i L' via an isotopy h. Let h' (t, -) = h( t, _ )-1
be the inverse homeomorphism. It is an isotopy between L' and L. Hence
L' "'i L.
(c) If, moreover, L' '" i L" via an isotopy h', then

"(
h t,-
)
=
{h(2t,-)
h'(2t-1,-)oh(1,-)
if
if
~ t ~ 1/2
1/2 ~ t ~ 1
defines an isotopy between Land L". In other words, the relation '" i is
transitive. 0

We now have two equivalence relations on links. The following proposi-

tion identifies them with each other.
244 Chapter X. Knots, Links, Tangles, and Braids

Proposition X.I.6. Let Land L' be links in R3. Then

L '"" c L' ~ L '"" i L'.

The reader will find a proof of this result in [BZ85], Prop. 1.10. As a
consequence, we shall suppress the subscripts i and c from the symbols '""i
and,"" c and henceforth speak of isotopic or equivalent links.
We end this section with a definition of a trivial link.

Definition X.I. 7. A link of order m in R3 is trivial if it is isotopic to

the union of m disjoint triangles in a plane. A trivial knot is a trivial link
of order 1.

We denote a trivial link of order m by

om = 00 ... 0 (m times).

Trivial links of the same order, but with different orientations, are always
isotopic. Therefore we need not specify the orientation of a trivial link.

X.2 Classification of Links up to Isotopy

The fundamental problem in knot theory is to classify all links in R3 up
to isotopy. In particular, one would like to have convenient criteria for two
links to be isotopic or for a link to be trivial. This is a difficult problem.
A classical way of appro ching this problem is to assign to each link L an
algebraic object h such that I L = I L' whenever Land L' are equivalent.
Such a function I is called an isotopy invariant for links. Let us give some
examples.
(a) (The order) It is clear that the number of connected components of
a link is preserved by an isotopy or a b.-operation. Therefore the order of a
link, i.e., the number of its connected components, is an isotopy invariant.
However, this invariant is weak since it is clearly insensitive to how much
a link is "knotted". Indeed all knots have the same order and nevertheless
there exists non-trivial knots such as the right-handed trefoil knot drawn
in Figure 2.1.

Figure 2.1. A right-handed trefoil knot

 X.2 Classification of Links up to Isotopy 245

(b) (The linking number) This is a more refined invariant which dates
back to Gauss. Let us consider two connected components L1 and L2 of
a link L. Consider a diagram of L (to be defined in Section 3). It shows
crossings of L1 and of L 2 . We associate to each crossing P an integer
E(P) = 1 defined as in Figure 2.2.

x x
E(P)=+l E(P) =-1

Figure 2.2. The linking number

Then the linking number of the components L1 and L2 is the integer

where P runs over all crossings of L1 and L 2 . This number does not de-
pend on the projection and is an isotopy invariant for links of order 2. For
instance, we have lk( 00) = 0 for the trivial link with two components,
and lk(H) = 1 for the Hopf link H drawn in Figure 2.3. It follows that the
Hopf link is not trivial.

Figure 2.3. The Hopi link

(c) (The fundamental group of a link) Define 7r( L) = 7r1 (R3 \ L) as the
fundamental group of the complement of the link in R3 (the definition
of the fundamental group is given in the Appendix to this chapter). For
the trivial knot, the group 7r( 0) is isomorphic to Z. More generally, the
group of the trivial link of order m is isomorphic to the free group F m on
m generators. By the very definition of isotopy, the fundamental group of
a link is an isotopy invariant. It is a very powerful invariant as one can
see from a theorem of Dehn's which asserts that a link L of order m is
trivial if and only if 7r( L) = Fm' In general, the group of a link is non-
abelian. Though it is possible to give a presentation of 7r(L) by generators
and relations from a plane projection of L, it is very difficult to use this
246 Chapter X. Knots, Links, Tangles, and Braids

presentation, for instance, to decide whether L is isotopic to another given

link. For more details, see [Bir74][BZ85].
(d) (Alexander and Conway polynomials) In 1928 Alexander [Ale28] con-
structed for each link L a polynomial6. L E Z[t, ell defined only up to a
power of t, which he proved to be an isotopy invariant. This invariant was a
very efficient tool for distinguishing links that were not equivalent. In 1970
Conway [Con70] showed that a suitable normalization of the Alexander
polynomial was of the form 6. L(t) = \7 L(t - e l ) where \7 L(z) is a poly-
nomial, now called the Conway polynomial, in Z[z]. Moreover, the Conway
polynomial has a simple characterization in terms of the skein relations
that will be described in Section 4.

X.3 Link Diagrams

The simplest way to describe a link in R3 is to represent it by a planar
diagram. We have already used this technique for the figures of Sections
1-2. We now give a definition of what we mean by a link diagram. We first
need the notion of a regular projection.

Definition X.3.l. (a) A link projection II is the union of a finite number

of closed polygonal arcs in R 2 such that no vertex lies in the interior of
any edge. A crossing point of II is a point of the link projection lying in
the interior of at least two edges. The order of a crossing point P is the
number of distinct edges in the interior of which Plies.
(b) A link projection is regular if each crossing point is of order exactly 2.
It is not difficult to see that a crossing point cannot be a vertex, and that
a link projection has only finitely many crossing points. The ordering of
each component will be represented by arrows on the edges of the projection
of the link following the rule given in Section l.
Let II be a regular link projection in the plane. Given a crossing point P
we may consider the set Ep consisting of the two edges on which P lies. A
priori, the set E p is unordered. This brings us to the following definition.

Definition X.3.2. A link diagram is a regular link projection in R2 for

which all sets E p (indexed by the crossing points P) are ordered. Given a
crossing point P, the first edge of the set Ep with respect to the ordering
is called the overcrossing edge whereas the other edge is called the under-
crossing edge.

Observe that an overcrossing edge for a crossing point may be under-

crossing for another crossing point. Changing the ordering in some sets
E p will be called a change of crossings. If a regular link projection has
m crossing points, then clearly there are 2m link diagrams with the same
underlying link projection.
 X.3 Link Diagrams 247

We represent a link diagram by a drawing of the regular link projection

in which the undercrossing edges are interrupted in the neighbourhood of
the crossings (as in Figures 2.1 and 2.3). From such a picture we observe
that any link diagram defines a link in R 3 by letting any undercrossing edge
pass under the corresponding overcrossing edge in the neighbourhood ofthe
crossing point. This link is defined only up to isotopy. There is no reason -
and in general it is false - why two link diagrams differing by a change of
crossings should define equivalent links. Nevertheless, the following should
be noted.

Lemma X.3.3. Any link diagram may be turned after appropriate changes
of crossings into a link diagram representing a trivial link in R3.

PROOF. Consider a link diagram. Pick a vertex and start moving along the
link, leaving a trail of red paint on the edges. At each crossing point, make
the red edge into an overcrossing edge unless the other edge is already red,
in which case the first edge is made into an undercrossing one. Apply this
procedure to each connected component. The resulting link diagram (ob-
tained from the original ones by a series of changes of crossings) represents
a trivial link. 0

The obvious question is now: Can any link in R 3 be represented by a

link diagram, at least up to isotopy? The answer is yes and provided by
the following proposition where we fix a linear projection 7ro of the space
R3 on the plane R2.

Proposition X.3.4. Any link in R3 is equivalent to a link L whose image

7r 0 (L) is a regular link projection.

PROOF. We sketch the proof. For details, see [BZ85]. Let L be a link in
R3. Consider the set S of all possible linear projections of R3 onto a fixed
plane. Given a projection 7r of S, there exists a homeomorphism h of R3
such that 7ro(h(L)) = 7r(L). It is therefore enough to show that the subset
Sreg of those projections 7r of S such that 7r(L) is a regular link projection
is not empty. Now S is in bijection with R2. Therefore we can transport
the topology of R2 onto S. What we shall actually prove, is that Sreg is
dense in S for this topology.
Let 7r be an element of S \ Sreg. Then in the projection 7r(L) we may
have the following singularities: some crossing point may be of order 2:: 3
or some vertex may sit in the interior of some edge. This happens when the
direction of the projection 7r passes through three edges or when it passes
through a vertex and an edge. In the first case, the direction sweeps over a
portion of a quadric; this projects to a part of a conic. In the second case,
it determines by projection a segment of the plane. Identifying S with R 2 ,
we see that S \ Sreg is contained in a finite number of straight lines and
conics of the plane. Therefore Sreg is dense in S. 0
248 Chapter X. Knots, Links, Tangles, and Braids

Having expressed the problem of classification of links in R:' in purely

two-dimensional terms, we now ask: When do two link diagram8 represent
isotopic link8? Before we answer thi8 important question, let us again fol-
low Reidemeister by introducing the four transformations on link diagrams
shown in Figures 3.1-3.4. These transformations are also called Reidemeis-
ter moves.

Figure 3.1. Reidemeisier move (0)

Figure 3.2. Reidemeisier move (I)

Figure 3.3. Reidemeister move (II)

Figure 3.4. Reidemeisier move (III)

Applying Transformation (0) to a link diagram means that one modifies it

locally by substituting one of the drawings of the figure of Transformation
(0) by another one without touching the rest of the link diagram. Similarly
for the other Transformations. Figures 3.5-3.8 show that Transformations
(0), (I), (II), and (III) are obtained by projection of ~-operations. Conse-
quently, applying these transformations to link diagrams doe8 not change
the i80topy class of the link in R 3 .
 X.3 Link Diagrams 249

Figure 3.5. (0 projected)

Figure 3.6. (I projected)

Figure 3.7. (II projected)

Figure 3.8. (III projected)

Reidemeister Transformations are sufficient in a sense we shall make

precise below after we defined the following additional concepts. Two link
diagrams II, II' are isotopic if there exists an isotopy h of R2 (see Defi-
nition 1.4) such that h(1, II) = II'. By this we mean that the underlying
projections are isotopic in the plane and that the orders of the sets Ep are
preserved in the course of the isotopy. Two isotopic link diagrams represent
isotopic links in R 3 .
Introducing the height as the second projection of R 2 onto R, we define a
generic link diagram to be a link diagram for which any two distinct vertices
250 Chapter X. Knots, Links, Tangles, and Braids

have different heights. In particular, a generic link diagram cannot have a

horizontal edge, i.e., an edge parallel to R x {O}. A generic isotopy between
two generic link diagrams II, II' is an isotopy of R2 such that h(1, II) = II'
and such that h( t, II) is a generic diagram for all t E [0, 1]. The following
statement is a criterion for generic diagrams to be isotopic as diagrams.

Lemma X.3.5. Two generic diagrams are isotopic if and only if they are
obtained from one another by a finite number of operations belonging to the
following set:
(A) a generic isotopy,
(B) an isotopy interchanging the order of the vertices with respect to the
height,
(C) a Reidemeister Transformation (0), and
(D) an isotopy in the neighbourhood of a local maximum as depicted in
Figure 3.9 and its images under reflection in the plane of the page, in a
horizontal line and in a vertical line.

Figure 3.9. An isotopy in the neighbourhood of a local maximum

We replace Transformation (D) of the previous lemma by another set of

operations that will turn out to be more adequate in Chapter XII.

Lemma X.3.6. Two generic diagrams are isotopic if and only if they are
obtained from one another by a finite number of the following operations:
(A) a generic isotopy,
(B) an isotopy interchanging the order of the vertices with respect to the
height,
(C) a Reidemeister Transformation (0), and
(E) an isotopy in the neighbourhood of a crossing point as shown in
Figure 3.10 and its images under reflection in the plane of the page.
 X.3 Link Diagrams 251

XL\l
Figure 3.10. An isotopy in the neighbourhood of a crossing point

PROOF. It is clear that the operations in (E) are obtained by isotopies of

diagrams. By Lemma 3.5 they follow from (A), (B), (0), and (D).
It remains to prove that Transformation (D) is a consequence of Trans-
formations (A), (B), (0), and (E). Figures 3.11-3.12 below give a proof of
this fact for the two operations represented in Figure 3.9.

Figure 3.11.

In Figure 3.11 the first operation is of type (0), the second one of type (A)
and (B) while the third one is of type (E).

Figure 3.12.

In Figure 3.12 the first and fourth operations are of type (E), the second
one of type (0), and the third one of type (A) and (B). Reflecting the
previous transformations with respect to the plane of the paper or with
respect to a vertical line takes care of their images under these reflections.
As for the reflections with respect to a horizontal line (they involve local
252 Chapter X. Knots, Links, Tangles, and Braids

minima), our assertion follows from the set of operations of Figure 3.13
where the first and last ones are Transformations (C), the middle one is
authorized by Figure 3.11, and the remaining ones are of type (A) and (B).
o

Figure 3.13.

We return to the problem of representing links in R 3 by (generic) link

diagram,;. Since moving vertices up and down locally allows us to turn any
link diagram into a generic one, we see that Proposition 3.4 implies that
any link in R3 is equivalent to a link L whose projection 7r o(L) is a generic
diagram.
Reidemeister [Rei32] proved the following important theorem which ex-
presses isotopy classes of links in R 3 in terms of purely two-dimensional
link diagrams.

Theorem X.3.7. Two generic link diagrams represent equivalent links in

R3 if and only if one is obtained from the other by a finite sequence of
Reidemeister Transformations (I), (II), (III), and of isotopies of diagrams.

X.4 The Jones-Conway Polynomial

We now construct the Jones-Conway polynomial. This is an isotopy invari-
ant of oriented links satisfying what knot theorists call "skein relations". In
order to formulate these relations, we introduce the concept of a Conway
triple. This concept already appears in [Ale28] p. 301, but Conway [Con70]
was the first one to observe that it could characterize knot invariants such
as the Alexander or the Conway polynomials.

Definition X.4.1. A triple (L+, L_, Lo) of oriented links in R3 is a Con-

way triple if they can be represented by link diagrams D +, D _, Do which
coincide outside a disk in R 2 and which are respectively isotopic to X+, X_
and 11 inside the disk.

The diagrams X+ and X_ are represented in Figure 4.1.

 X.4 The Jones-Conway Polynomial 253

X X
X+
Figure 4.1.
X

We now state the main theorem of this chapter.

Theorem X.4.2. There exists a unique map L 1-+ P L from the set of all
oriented links in R3 to the ring Z[x, x-I, y, y-l] of two-variable Laurent
polynomials such that
(i) if L '" L', then P L = PL"
(ii) the value of P on the trivial knot is 1, and
(iii) whenever (L+, L_, Lo) is a Conway triple, we have

(4.1)

The invariant P L is called the Jones-Conway polynomial, or the two-

variable Jones polynomial, or the Homfiy polynomial (after the initials of
the six authors of [FYH+85]) of the link L. Relations (4.1) are called skein
relations. The polynomial \7 L E Z[z] that Conway [Con70] devised as a
variant of the Alexander polynomial is characterized by Properties (i)-(ii)
of Theorem 4.2 along with the skein relation

(4.2)

Similarly, the polynomial VL E Z[tl/2, rl/2] which was discovered in

1984 by Vaughan Jones [Jon85][Jon87] is characterized by Properties (i)-
(ii) and by the skein relation

(4.3)

As a consequence of Theorem 4.2, the Conway polynomial and the Jones

polynomial exist, and they are related to the two-variable Jones-Conway
polynomial by

The Jones-Conway polynomial can help distinguish a link L from its

mirror image I, i.e., its image under a reflection with respect to a plane in
R3. Theorem 4.2 has the following corollary.
Corollary X.4.3. We have PZ(x, y) = PL(X- 1 , -y).
254 Chapter X. Knots, Links, Tangles, and Braids

PROOF. The mirror image of the Conway triple (L+,L_,Lo) is the triple
(L_, L+, Lo). Consequently, we have

One concludes by appealing to the characteristic properties of P. 0

Let us prove Theorem 4.2. Consider the ring A = Z[x,x-l,y,y-l], the

set /C of equivalence classes of all oriented links in R 3 , and A[/C] the free
A-module generated by /C. We denote by Y the quotient of A[/c] by the
A-submodule generated by

(4.4)

where (L+,L_,Lo) runs over all Conway triples of /C. The A-module Y is
called the skein module of R 3 .

Proposition X.4.4. Let Q : A ----., Y be the A-linear map sending 1 on the

class [0] of the trivial knot. Then Q is an isomorphism.

Consequently, the skein module Y is a free A-module of rank one gen-

erated by [0]. Proposition 4.4 implies Theorem 4.2. Indeed, let L be an
oriented link and [L] its class in Y. Set

It is clear that P satisfies all three conditions of Theorem 4.2. It remains

to establish Proposition 4.4. The proof of the latter divides into two parts
consisting in proving successively that the map Q is surjective, then injec-
tive.
Surjectivity of Q. This is purely topological and is essentially independent
of the nature of the ring A. It is enough to check that the A-module Y is
generated by the class [0] of the trivial knot. This will be shown in two
steps.

Lemma X.4.5. The A-module Y is generated by the family {[on]}n>O

of isotopy classes of all trivial links.

PROOF. Let Y m be the A-submodule generated by the isotopy classes of

links representable by link diagrams with S; m crossing points. Clearly, Y m
maps to Y m+l and Y is the union of all Y m' It is therefore enough to prove
Lemma 4.5 for each Y m' This is done by induction on m. The case m = 0
holds by definition of trivial links. Suppose the assertion is proved for all
integers < m. Let [L] be the class of a link L in Y m' It may be represented
by a link diagram with m crossing points. Consider one of them. Then
there exists a Conway triple (L+,L_,Lo) such that L = L+ or L = L_
and the diagram Lo has less than m crossing points. It follows from (4.4)
 X.4 The Jones-Conway Polynomial 255

that [L+l == x-2[L_l modulo Ym-l. In other words, a change of crossings

changes the class of L modulo Y m-l by multiplication by an invertible
element of A. By Lemma 3.3 this implies that the class of L belongs to
the submodule generated by the trivial links and Y m-l. The latter is also
generated by trivial links in view of the induction hypothesis. 0

The second step in the proof of the surjectivity of Q is the following

lemma which, incidentally, shows the necessity for y to be invertible (unless
x = 1 as in the case of the Conway polynomial).

Lemma X.4.6. For any integer n > 1, we have

PROOF. Figure 4.2 implies that (Qi8l n , O'81 n , QiSIn+l) is a Conway triple for
all n 2: 1. By definition of Y we get

One concludes by induction on n. o

Figure 4.2. A Conway triple

DO
Injectivity of Q. This part of the proof is algebraic in contrast with the
surjectivity part. We use the following proposition whose proof will be
given in Chapter XII.

Proposition X.4.7. Let q =/:- 0 be a complex number that is not a root of

unity and let m be an integer> 1. There exists a unique map <Pm,q from
the set of all oriented links in R3 to the field C of complex numbers such
that
(i) if L rv L', then <pm,q(L) = <Pm,q(L'),
(ii) the value of <P m,q on the trivial knot is
qm _ q-m
<Pmq(O)
, = q-q -1 =/:-0,

(iii) and, whenever (L+, L_, Lo) is a Conway triple, we have

256 Chapter X. Knots, Links, Tangles, and Braids

Assuming Proposition 4.7 and using the ring map (q,m : A ----> C defined
by (q,m(x) = qm and (q,m(y) = q-q-l, which furnishes C with a A-module
structure, we see that IJ>m,q extends by linearity to a unique A-linear map
IJ>~,q from A[KJ to C. Now for any Conway triple (L+,L_,L o), we have

IJ>~,q (x[L+J - X-I [L-l - y[LoJ)

(q,m(x)lJ>m,q(L+) - (q,m(x-I)lJ>m,q(L_) - (q,m(y)lJ>m,q(Lo)
qmlJ>m,q(L+) - q-mlJ>m,q(L_) - (q - q-I)lJ>m,q(Lo)
o
by Proposition 4.7. Consequently, IJ>~,q factors through a unique A-linear
map, denoted IJ>~,q, from T into C such that IJ>~,q([L]) = IJ>m,q(L) for any
oriented link L.
We are now ready to prove the injectivity of the map Q from A into T,
which will complete the proof of Proposition 4.4, hence of Theorem 4.2.
Let f(x, y) E A be a two-variable Laurent polynomial chosen such that
Q(J(x,y)) = f(x,y)[OJ vanishes in T. Then, using the above-defined map
IJ>~,q : T ----> C, we have

for any integer m > 1 and any complex number q that is not a root of
unity. Since IJ>m,q(O) =f:. 0, we have f(qm,q_q-I) = O. Since this is true for
an infinite number of distinct powers of q, the polynomial f is divisible by
the polynomial y - (q - q-I). The latter assertion holds for infinitely many
complex numbers q, which is possible only if the polynomial f is zero. This
proves the injectivity of Q. D
Application 4.8. We end this section with the computation of the Jones-
Conway polynomial of the right-handed trefoil knot K and of the Hopf
link H. Figures 4.3-4.4 show that (H, 00, 0) and (K, 0, H) are Conway
triples.

Figure 4.3.
 X.5 Tangles 257

Figure 4.4.

By (4.1) we have
-1
xPH = X
-1
Poo + yPo = x
-1 X - X
y
+ y.
Hence,
PH = (x- 1 - X- 3 )y-1 + x- 1 y.
A similar computation yields

for the right-handed trefoil knot. By Corollary 4.3, we see that the Jones-
Conway polynomial of the mirror image K is given by

This proves that the trefoil knot is not isotopic to its mirror image, a fact
already observed by Dehn [Deh14] in 1914.

X.5 Tangles
This section is devoted to the concept of tangles which generalizes the no-
tion of links. Tangles will be used extensively in Chapter XII, in particular
for the proof of Proposition 4.7.
For any integer n > 0, we set [n] = {1, 2, ... ,n}. When n = 0, we agree
that [0] is the empty set. We denote by I the closed interval [0, 1] and by
R 2 the real plane.

Definition X.5.l. Let k and be nonnegative integers. A tangle L of type

(k, ) is the union of a finite number of pairwise disjoint simple oriented
polygonal arcs in X = R2 X I such that the boundary 8L of L satisfies the
condition

8L = Ln (R2 x {0,1}) = ([k] x {O} x {O}) U ([] x {O} x {1}).

258 Chapter X. Knots, Links, Tangles, and Braids

The boundary condition in Definition 5.1 means that the tangle intersects
the two boundary planes of R2 x I transversally. Observe that a link in
R2 x I is a tangle of type (0,0). Figure 5.1 shows an example of a tangle
that is not a link.

-~--y-~---

~
___Y' ___
Figure 5.1. A tangle

Given a tangle L of type (k, f), we define two finite sequences s(L) and
b(L) consisting of + and - signs. If k = 0, then s(L) = 0 is the empty set
by convention. Similarly if = 0, we set b(L) = 0. In the general case, we
define
S(L)=(Cl"",Ck) and b(L) = ('rfl""''rfc)
where ci = + [resp. 'rfi = + ] if the point (i, 0, 0) [resp. the point (i, 0,1)]
is an endpoint [resp. an origin] of L. We have ci = - and 'rfi = - in the
remaining cases.
Let us give a few examples of tangles that shall be used in the sequel.
1. We denote the polygonal arcs [(1,0,1), (1,0,0)] and [(1,0,0), (1,0,1)] by
1 and i respectively. We have s(1) = (+), b(1) = (+), s(i) = (-), and
b(i) = (-).
2. The tangles X+ and X_ of Figure 4.1 can be defined by

where M 1 , Mt, M 3, Nu Nt, N3 are the points whose coordinates in R2 x I

are
Ml (2,0,1), (1,0,1),
Mt = (3/2, =fl, 1/2), (3/2, 1, 1/2),
M3 = (1,0,0), (2,0,0).
We have s(X) = b(X) = (+, +).
3. The tangles nand n of type (2,0) are defined by
n = [(1,0,0), (3/2,0,1/2)] U [(3/2,0,1/2), (2,0,0)] (5.2)

and
n= [(2,0,0), (3/2,0,1/2)] U [(3/2,0,1/2), (1,0,0)]. (5.3)
 X.5 Tangles 259

We have s(n) = (-, +), b(n) = 0, s(n) = (+, -), and b(n) = 0 (see
Figure 5.2).

Figure 5.2. The tangles nand n

4. Similarly, we define tangles U and U of type (0,2) by
U = [(1,0,1), (3/2,0,1/2)] U [(3/2,0,1/2), (2,0,1)] (5.4)
and
U = [(2,0,1), (3/2,0,1/2)] U [(3/2,0,1/2), (1,0,1)]. (5.5)
We have s(U) = 0, b(U) = (+, -), s(U) = 0, and b(U) = (-, +) (see

v
Figure 5.3).

Figure 5.3. The tangles U and U

We have the same equivalence relations for tangles as for links. Let us
adapt their definition to the case of tangles. We start with the combinatorial
relation "'c'
Definition X.5.2. (a) Let L be a tangle in X and M i , Mi+1 be two con-
secutive vertices of L. We are also given a point N in R2 x ]0, 1[ such that
N r:J- L, Mi r:J- [N, Mi+1], Mi+1 r:J- [Mi' N], and
[Mi' N, Mi+1] n L = [Mi' Mi+1]'
We define L' as the tangle

We say that L' is obtained from L by a tl-operation.

(b) Two tangles Land L' are combinatorially equivalent - we write this
L rv c L' - if there exist tangles L = La, L 1 , ... ,Lk = L' such that for all i,
one of the tangles L i , Li+l is obtained from the other one by a tl-operation.
260 Chapter X. Knots, Links, Tangles, and Braids

Clearly, if Land L' are combinatorially equivalent, they have the same
boundaries and are of the same type. Isotopies are defined as follows.

Definition X.5.3. (a) An isotopy of X = R2 X I is a piecewise-linear

map h : I x X ~ X such that for all t E I, the mapping h(t, -) is a
homeomorphism of X restricting to the identity map on the boundary ax =
R2 x {O, l} and such that h(O, -) is the identity of X.
(b) Two tangles Land L' are isotopic - we write L rv i L' - z! there exists
an isotopy h of X such that h(l, L) = L'.

Again if Land L' are isotopic, they have the same boundaries and are
of the same type. The isotopy is shown to be an equivalence relation for
tangles in the same way as it was for links (see Lemma 1.5). We have the
following counterpart of Proposition 1.6.

Proposition X.5.4. Let Land L' be two tangles. Then

L rve L' ~ L rv.; L'.

As for links, we shall suppress the subscripts i and c from the symbols
and rv c and we shall henceforth speak of isotopic or equivalent tangles.
rv i

Tangles can also be represented by planar diagrams. We adapt the fol-

lowing concepts and results from Section 3 without bothering to give un-
necessary details.

Definition X.5.5. (a) A tangle projection rr is the union of a finite num-

ber of (not necessarily closed) polygonal arcs in R2 such that no vertex sits
in the interior of any edge and such that the boundary arr of rr satisfies
the condition

arr = rr n (R x {O, l}) = ([k] x {O}) U Of] x {l}).

A crossing point ofrr is a point of the tangle projection sitting in the inter'ior
of at least two edges. The order of a crossing point P is the number of
distinct edges in the interior of which P sits.
(b) A tangle projection is regular if each crossing point is of order exactly
2.

Let rr be a regular tangle projection in the plane. Given a crossing point

P we may again consider the unordered set E p consisting of the two edges
on which P sits.

Definition X.5.6. A tangle diagram is a regular tangle projection in R x I

for which all the sets E p (indexed by the crossing points P) are ordered.
Given a crossing point P, the first edge of the set E p with respect to the
ordering is called the overcrossing edge whereas the other edge is called the
undercrossing edge.
 X.5 Tangles 261

Replacing R2 by R x [0,1]' one defines the concepts of isotopy of tangle

diagrams, of generic tangle diagrams, and of generic isotopy as for links.
Similar results hold. We record here the counterpart of Lemma 3.6.

Lemma X.5.7. Two generic tangle diagrams are isotopic if and only if
they are obtained from one another by a finite number among the following
operations:
(A) a generic isotopy,
(B) an isotopy interchanging the order of the vertices with respect to the
height,
(C) a Reidemeister Transformation (0), and
(E) an isotopy in the neighbourhood of a crossing point as shown in
Figure 3.10 and their images under reflection in the plane of the page.

As in the case of links, any (generic) tangle diagram defines a tangle in

R2 x I which is unique up to isotopy. Fix a linear projection
KO of the space
R2 x I on the strip R x I.

Proposition X.5.8. Any tangle in R2 xl is equivalent to a tangle L whose

projection K 0 (L) is a generic tangle diagram.

When do two tangle diagrams represent isotopic tangles? The answer to

this question is the same as in the case of links. It uses the Reidemeister
moves already defined in Section 3.

Theorem X.5.9. Two generic tangle diagrams represent equivalent tan-

gles in R 2 X I if and only if one 'is obtained from the other by a finite
seq'uence of Reidemeister Transformations (I), (II), (III), and of isotopies
of diagrams.

We close this section by defining a partial binary operation on tangles.

Consider the piecewise-linear mappings a 1 , a 2 from the topological space
R 2 X I into itself defined by

a 1 (p, z) = (p, z/2) and a2 (p, z) = (p, (z + 1)/2)

where p E R2 and z E I. When Land L' are oriented tangles such that
b(L) = s(L'), then
L' 0 L = a 1 (L) U a2 (L')
is an oriented tangle with

s(L' 0 L) = s(L) and b(L' 0 L) = b(L').

The tangle L' 0 L is called the composition of Land L'. It is obtained by
placing L' on top of L, by glueing their middle ends together and by squeez-
ing the whole into R2 x [0,1]. Let us prove that composition is compatible
with the equivalence of tangles.
262 Chapter X. Knots, Links, Tangles, and Braids

Lemma X.5.10. Let L 1, L 2 , L 3, L4 be oriented tangles with b(L 1) = 8(L2)

and b(L3) = 8(L4)'
(a) If L1 rv L3 and L2 rv L 4, then L2 0 L1 rv L4 0 L 3
(b) If, furthermore, b(L 2 ) = 8(L 3), then (L3 0 L 2) 0 L1 rv L3 0 (L2 0 L1)'

PROOF. (a) Use Reidemeister Transformations.

(b) The tangles (L3 0 L 2) 0 L1 and L3 0 (L2 0 L 1) are isotopic through
the isotopy h(t,p,z) = (p,'Pt(z)) wherep E R2, t,z E [0,1] and 'P is the
continuous mapping from I x I into I defined by

z(l - ~) if O:S z :S 1/2,

'Pt(z) = { z - ~ if 1/2:S z :S 3/4,
(1 + t)z - t if 3/4:S z :S 1.
o
The composition has partial left and right units up to isotopy. Indeed, for
any finite sequence c of signs of length n, define the tangle idE as the union
of the n intervals {1, ... ,n} x {O} x [0, 1], their origins and endpoints being
uniquely determined by the requirement s(idE ) = b(id E ) = c. If the sequence
c is empty, take id 0 to be the empty tangle. An immediate application of
A-operations proves the following lemma.

Lemma X.5.ll. For any tangle L, we have

idb(L) 0 L rv L rv L 0 ids(L)'

X.6 Braids
We now consider a special class of tangles, called braids. Fix an integer
n'?1.

Definition X.6.l. A braid L with n strands is a tangle of type (n, n) such

that
(i) s(L) = b(L) = (+,+,00',+),
(ii) L contains no closed are, and
(iii) for all z E I, the intersection of L with theplane R2 x {z} consists
of exactly n distinct points.

In other words, a braid with n strands is the union of n pairwise disjoint

simple polygonal arcs, relating the set [n] x {O} x {1} to the set [n] x {O} x {O}
and having no local maximum or minimum with respect to the "height"
projection R2 x I -+ I. Figure 6.1 shows a braid with 5 strands.
 X.6 Braids 263

Figure 6.1. A braid

By definition, two braids are equivalent if they can be obtained from

each other by a finite sequence of D.-operations performed within the set
of braids. Up to equivalence, a braid with n strands can be represented by
what may be called a braid diagram, i.e., a tangle diagram such that for all
z E I the intersection of the diagram with R x {z} consists of exactly n
distinct points. There is also a notion of isotopy of braid diagrams which
is the restriction to braid diagrams of the corresponding notion for tan-
gle diagrams. Braid diagrams are isotopic if and only if they are obtained
from each other by moving vertices up and down. Concerning Reidemeister
moves, Transformations (0) and (I) are clearly forbidden for braid dia-
grams. Only Transformations (II) and (III) may occur. They are sufficient
to generate the braid equivalence, as witnessed by the following proposition
whose proof is similar to (and simpler than) the proof of the corresponding
Theorem 5.9 for tangles.

Proposition X.6.2. Two braid diagrams represent equivalent braids if and

only if they are obtained from each other by a finite sequence of Reidemeis-
ter Transformations (II), (III), and of isotopies of braid diagrams.

X.6.1 The braid group En

In Section 5 we defined the composition L' 0 L for tangles L, L' such that
b(L) = s(L'). We see from the definitions that the composition of two braids
with n strands is still a braid with n strands. A special braid with n strands
is ide; (as defined at the end of Section 5) where c is the sequence consisting
of n signs +. We denote its equivalence class by In' Given a braid L we
define the inverse braid L -1 as the image of L under the reflection through
the plane R2 x {1/2}.
Denote the set of equivalence classes of braids with n strands by Bn'
The set Bo has one single element, namely the empty braid. Restricted to
braids, Lemma 5.10 implies that the composition of braids is compatible
264 Chapter X. Knots, Links, Tangles, and Braids

with braid equivalence. Therefore the composition induces a product on

En. As a matter of fact, we have the following result.
Proposition X.6.3. The composition of braids induces a group structure
on En with In as a unit.
PROOF. Associativity of the product follows from Lemma 5.10 (b) whereas
Lemma 5.11 implies that In is a left and right unit for the product on En.
Repeated use of Reidemeister Transformation (II) shows that
L- 1 oLrvI n rvLoL- 1 ,
which implies that the equivalence class of L -1 is an inverse for the equiv-
alence class of L. D

The group En was introduced by E. Artin [Art25]. It is called the braid

group (on n strands). The groups Eo = E1 are isomorphic to the trivial
group {l}.
We now give a presentation by generators and relations of En. First, we
define special elements 0"1,0"2' ... ,0"n-1 in En. A braid diagram of the braid
O"i is shown in Figure 6.2.

1 i i +1 n

X\
Figure 6.2. The braid O"i

Using the tangle X+ of Section 5, we see that O"i is equivalent to the

braid
11 X+ 1 1
exchanging the i-th and the (i + I)-st strands and leaving the other ones
untouched. Its inverse 0";1 is equivalent to the braid
11 X_ 1 1
with the opposite crossing (see Figure 6.3).

1 i+I n

><
Figure 6.3. The braid 0";1
 X.6 Braids 265

Lemma X.6.4. (a) The group Bn is generated by 0"1"'" O"n-l'

(b) When n ?: 3 and 1 ::; i, j ::; n - 1, we have the following relations in
the braid group Bn:
(6.1)
if Ii - jl > 1 and
(6.2)

Relations (6.1-6.2) are called the braid group relations.

PROOF. (a) Represent a braid by a braid diagram. Move crossing points
up or down so that one can find 0 < t1 < ... < tr < 1 such that for all i
there is only one crossing point in R x [ti' ti+1J. This means that the braid
is equivalent to a braid whose restriction to R x [ti' ti+lJ is equivalent to
some O"k or its inverse. Using the definition of the product in the braid
group, we see that the given braid can be expressed in Bn as the product
of the elements O"k and their inverses.
(b) If Ii - jl > 1, then clearly O"iO"j and O"jO"i are equivalent (draw a
picture). Both sides of Relation (6.2) are represented in Figure 3.4. One
passes from one diagram to the other by Reidemeister Transformation (III).
D

We now state an important theorem due to E. Artin [Art25J [Art47J.

Theorem X.6.5. Given a group G and elements cl , ... , cn - l (n> 2) such

that for all i, j we have cic j = cjci if Ii - jl > 1 and

then there exists a unique group morphism from Bn to G mapping O"i to ci .

Corollary X.6.6. The group Bn is isomorphic to the group generated by
0"1'0"2""'O"n_1 and the braid group relations (6.1-6.2).
PROOF. Let G be the latter group. By Theorem 6.5, there exists a unique
group morphism p : Bn ---t G such that P(O"i) = O"i for all i. Now by
definition of a group given by generators and relations, there exists a unique
group morphism p' : G ---t Bn sending O"i onto O"i for all i. Then p' is inverse
to p. D
Proof of Theorem 6.5. The uniqueness of the group morphism follows from
the fact proven in Lemma 6.4 (a) that 0"1"'" O"n-1 generate Bn'
Now, let us establish the existence of a group morphism p : Bn ---t G such
that P(O"i) = ci for i = 1, ... , n-l. We sketch a geometric proof. Consider a
braid L represented by a generic diagram as in the proof of Lemma 6.4 (a),
namely for which two different crossing points have different heights. To
such a diagram we can assign a unique braid word w in the generators O"i
and their inverses as in the proof of Lemma 6.4 (a). Define p(w) to be the
266 Chapter X. Knots, Links, Tangles, and Braids

element of G obtained by replacing (Ji by Ci and (Ji l by C;l in the word

w. Let us show that p(w) depends only on the initial braid.
According to Proposition 6.2, we have to check that p( w) does not change
when we perform an isotopy of diagram or Reidemeister Transformations
of type (II) and (III). In the first case, moving crossing points up and
down amounts to changing the braid word by products of commutators
of the form (Ji(Jj(Jil(Jjl for Ii - jl > 1. This leaves p(w) unchanged be-
cause Cicjc;lcjl = 1 for Ii - jl > 1. Under a Reidemeister Transforma-
tion (II), braid words differ by (Ji(Ji l or by (Jil(Ji whose images under
p is 1. Under a Reidemeister Transformation (III), braid words differ by
(Ji (Ji +1 (Ji(J~l (Jil(J~l or its inverse. Their images under p again is 1 because
of the relation c i Ci+ 1 Ci = Ci+ 1 Ci Ci+ l' 0
We still have to investigate the case of braids with two strands. An adap-
tation of the proofs of Lemma 6.4 and Theorem 6.5 proves the following.

Proposition X.6.7. The group B2 is generated by (J 1 and is isomorphic

to the group Z of integers.

X.6.2 Braid group representations from R-matrices

We show how Theorem 6.5 allows us to construct braid group representa-
tions from any solution of the Yang-Baxter equation.
Let V be a vector space, C a linear automorphism of V 0 V, and n > 1
an integer. Then for 1 ~ i ~ n -1, define a linear automorphism Ci of Vi)SIn
by

C 0 id v @(n-2) if i = 1,
idV@(i-l) 0 C 0 idv@(n-i-l) if 1 < i < n -1, (6.3)
id v @(n-2) 0 C if i=n-1.

Clearly cic j = cjci if Ii - jl > 1. It is easy to check the following lemma.

Lemma X.6.S. Under the previous hypothesis, we have

for all i if and only C is a solution of the Yang-Baxter equation.

The Yang-Baxter equation of VIII. 1 can be expressed with the present

notation as the equation c l c 2 Cl = c 2 c l c 2 holding in Aut(V 0 V 0 V). The
following is a consequence of Theorem 6.5 and Lemma 6.8.

Corollary X.6.9. Let C E Aut(V 0 V) be a solution of the Yang-Baxter

equation. Then, for any n > 0, there exists a unique group morphism p~ :
Bn ---+ Aut(V0 n ) such that P~((Ji) = Ci for i = 1, ... , n - 1.
 X.6 Braids 267

Consequently, by this procedure, any linear automorphism c of V Q9 V

that is a solution of the Yang-Baxter gives rise to a representation of the
braid group Bn on the tensor power vn where n is any integer;:: 2.

X.6.3 Relation with the symmetric group Sn

Given a braid L, there exists a unique permutation CJ(L) of the set {I, ... ,n}
such that for all k E {I, ... , n} the endpoint (k, 0, 0) lies in the same con-
nected component as the origin (CJ(k), 0,1). The permutation CJ(L) is called
the permutation of the braid L.

Lemma X.6.10. The map f f-7 CJ(J) induces a surjective morphism of

groups from the braid group Bn onto the symmetric group Sn'

PROOF. First it is clear that equivalent braids have the same permutation.
Thus the map factors through Bn' It is a morphism of groups because we
have CJ(L' 0 L) = CJ(L') 0 CJ(L), CJ(1n) = id, and CJ(L-l) = CJ(L)-l. The
permutation of the braid CJi is the transposition (i, i + 1). The surjectivity
of the map follows from the fact that such transpositions generate the
symmetric group Sn' 0

This lemma is not surprising in view of Moore's theorem which gives the
following presentation in terms of generators and relations for the symmet-
ric group Sn: it is generated by the n - 1 transpositions Si = (i, i + 1) and
by Relations (6.1-6.2) where CJ i has been replaced by si' as well as by the
additional relations s; = 1 for i = 1, ... ,n - l.
One big difference between symmetric groups and braid groups is that
the former are finite groups while the latter are infinite groups when n > l.
Moreover, the group Bn has no torsion, that is to say, all elements i- In
have infinite order.

X.6.4 Representing braids as loops

We end this section by giving a function-theoretic definition of braids. Let
n be an integer ;:: 1. Consider the set

endowed with the subset topology of en. The symmetric group Sn acts on
Yn by permutation of the coordinates. Let Xn = Yn/ Sn be the quotient
space with the quotient topology. The space Xn is the configuration space
of n distinct points in e. Consider the following set p = {I, 2, ... ,n} of n
distinct points in X n .

Definition X.6.ll. A loop in Xn is a piecewise linear map

268 Chapter X. Knots, Links, Tangles, and Braids

such that for all tEl we have fi(t) "I fj(t) whenever i "I j and

1(0) = (1,2, ... ,n) and {ll(I),12(1), ... ,ln(1)} = {1,2, ... ,n}.

Loops in Xn and braids with n strands in R2 x [0,1] are equivalent

notions after we have identified e with R2. Indeed, given a loop f =
(fl,12, ... ,ln ) in X n , the union of the graphs of the maps Ii is a braid
with n strands. Conversely, for any braid L with n strands, we define Ii (z)
to be the projection onto R 2 = e of the intersection of the plane R 2 x {z}
with the connected component of L ending at the point (i, 0, 0). This defines
a loop I = (f1' 12'' In) in X n
For any loop I = (f1' ... ' In) in X n , define the permutation u(f) of the
set {l, ... ,n} by u(f)(k) = Ik(l) for all k. Check that, if I is the loop
corresponding to a braid L, then u(f) = u(L).
The equivalence of braids can be expressed on loops in Xn as follows.
Two loops 1= (f1' 12' ... ,in) and 9 = (gl' g2' ... , gn) in Xn are homotopic
in Xn - we write I '" 9 - if there exists a piecewise linear map, called an
isotopy,
H = (H 1 ,H2, ... ,Hn): I x I - t en
such that for all (s, t) E I x I and i "I j we have Hi(S, t) "I Hj(s, t), for all
s E I and k with 1 :::; k :::; n we have

for all tEl and 1 :::; k :::; n we have

Proposition X.6.12. Two braids with n strands are equivalent if and only
il the corresponding loops are isotopic in X n .
We can transpose the composition of tangles on the level of loops. Let I
[resp. 1'] be the loop in Xn corresponding to a braid L [resp. L']. It is easy
to see that the loop I l' = ((f I'h, ... , (f I')n) corresponding to the braid
L' 0 Lx, composed in the sense of tangles, (see Section 5) is given for all i
by
if 0:::; t :::; 1/2,
( ') () {Ii (2t)
II i t = 1~(i)(2t -1) if 1/2 :::; t :::; 1
where u = u(f) is the permutation of I. The loop I l' is called the product
of the loops I, and 1'. We have u(f1') = u(f') 0 u(f).
Given a loop I = (fl' ... ,in) corresponding to a braid L, the loop I-I
defined by
li-l(Z) = la--1(i)(I- z),
where again u = u(f), corresponds to the inverse braid L -1. The loop
corresponding to the braid In is the constant map z f--+ (1,2, ... , n).
 X.7 Exercises 269

We have the following lemma which is the counterpart for loops of Lem-
mas 5.10-5.11.

Lemma X.6.13. Given loops f,f',f",g,g' in X n ,

(a) if f '" 9 and f' '" g', then f f' '" gg',
(b) (If')!,, '" f(l'!"),
(c) Inf'" f '" fIn' and
(d) f r l '" In '" rlf
PROOF. See the Appendix to this chapter. D

We have the following important result as a consequence of the presen-

tation of braids by loops and of the definitions of the Appendix.

Proposition X.6.14. The braid group Bn is isomorphic to the fundamen-

tal group of the configuration space Xn of n distinct points in C:

where p is the set {I, ... , n}.

X.7 Exercises
1. (Centre of the braid group) Let n be an integer> 2. Show that the cen-
tre of the braid group Bn is generated by the element (0'1'" O'n_l)n.

2. For an integer n > 1 let Fn be the free group generated by Xl"" ,x n .

Define automorphisms Xl"" ,Xn-l of Fn by

if j = i,
if j = i + 1,
if j ~ i, i + 1.

Prove that there exists a morphism X of the braid group Bn into the
group of automorphisms of Fn such that X(O'i) = Xi for all i.

3. (Burau representations) (a) Let n > 1 be an integer and {v l ,, vn }

be a basis of a free Z[t, ell-module Vn of rank n. For any i such that
1 :::; i :::; n - 1 define an automorphism {3i of Vn by (3i(Vk) = vk if
k f. i, i + 1, and

Show that there exists a unique morphism {3 of the braid group Bn

into the group of automorphisms of Vn such that (3(O' i ) = {3i for all i.
(b) Let {el, ... ,en-d be a basis of a free Z[t,el]-module Vn - l of
rank n - 1. For any i such that 1 < i < n - 1 define an automorphism
270 Chapter X. Knots, Links, Tangles, and Braids

(3i of Vn- 1 by (3i(ek) = ek if k -=J i and (3i(e i ) = te i_1 - tei + ei + 1

with the c~vention eo = en = O. Show that there exists a unique
morphism (3 of the braid group Bn into the group of automorphisms
of Vn- 1 such that (3(O"i) = (3i for all i. Prove that

(3((O"l"'O"n_l)n) =tnid Vn _ 1 '

(c) When n = 3 and t = -1, prove that (3 induces a surjection of

groups from B3 onto the group SL 2 (Z) of integral 2 x 2-matrices
with determinant one.
4. Define the pure braid group Pn as the kernel of the map f f-+ O"(f)
from the braid group Bn to the symmetric group Sn' Show that P n is
isomorphic to the fundamental group of the space Y n defined in 6.4.

5. (Kauffman's bracket) Show that the Kauffman bracket as defined in

Section 8 is invariant under Reidemeister Transformations (0), (I'),
(II) and (III) (Transformation (1') is the variant of Transformation
(I) defined in Section 8).

X.8 Notes
Classical references on knot theory are [Bir74][BZ85][Kau87a][Rei32][RoI76].
The Jones polynomial VL was defined in [Jon85] [Jon87]. Its two-variable
extension PL appeared in a number of papers written almost simultane-
ously [FYH+85] [Hos86] [LM87] [PT87] (see also [HKW86] [Kau91]). For
Theorem 4.2 we followed the proof given by Thraev in [Tur89].
(Smooth tangles) There is a version of tangles and isotopies where piece-
wise-linear maps are replaced by C= maps and the boundary condition
of Definition 5.1 is replaced by a transversality condition. Such smooth
tangles project to smooth tangle diagrams. It may be shown that smooth
isotopy classes of smooth tangles are in bijection with isotopy classes of
tangles as defined in Section 5 (see [BZ85]).
(Framed tangles) Let us define a normal vector field on a smooth tangle L
as a C= vector field on L that is nowhere tangent to L and that is given by
the vector (0, -1,0) at all points of the boundary 8L. One may suggestively
think of a tangle with a normal vector field as a tangled ribbon defined as
follows: one edge of the ribbon is the tangle itself whereas the other one
is obtained from the first one by a small translation along the vector field.
A framing of the tangle L is a homotopy class of normal vector fields on
L where two normal vector fields are said to be homotopic if they can be
deformed into one another within the class of normal vector fields. One can
extend the concept of isotopy from tangles to tangles with framings. Isotopy
classes of tangles with framings are called framed tangles or ribbons. We
 x.s Notes 271

shall see in Chapter XIV that ribbons give rise to an interesting categorical
structure, the so-called ribbon categories.
Framed tangles can be represented by tangle diagrams in the sense of
Section 5 just as ordinary tangles are. Take a tangle diagram. By defini-
tion, it represents the following framed tangle: the underlying tangle is the
tangle represented by the diagram and the framing is determined by the
constant normal vector field (0, -1, 0) that is perpendicular to the plane of
the diagram and points to the reader. Any framed tangled may be repre-
sented by a planar diagram in such a way. We already know this for the
underlying unframed tangle. To represent a general framed tangle with a
vector field whirling around it, it is enough to know how to represent a ver-
tical tangle around which the vector field turns by an angle of 27T or of - 27T.
The corresponding ribbons appear in Figure 8.1 and may be represented
by the diagrams of Figure 8.2.

Figure 8.1.

Figure 8.2.

There is an analogue of Reidemeister theorem for framed tangles. For this

we need a variant (1') of Reidemeister Transformation (I). It is depicted in
Figure 8.3. Two tangle diagrams represent isotopic framed tangles if and
only if they can be obtained from one another by a finite sequence of
Reidemeister Transformations (1'), (II), (III), and of isotopies of diagrams.
272 Chapter X. Knots, Links, Tangles, and Braids

Figure 8.3. Reidemeister move (I')

(Kauffman's bracket) Shortly after Vaughan Jones's discovery of the link

polynomial VL , Louis Kauffman [Kau87b] found an isotopy invariant, now
called the Kauffman bracket, for framed links. It is with values in the ring
Z[x, X-I] of Laurent polynomials. The Kauffman bracket < L > can be
characterized as follows. Take any diagram representing the framed link
L. Single out a crossing. Define Lo [resp. L(XJ] to be the diagram where
the crossing has been replaced by II [resp. by ~ ]. Then the bracket is
determined by the rules

< L >= x < Lo > +x- I < L(XJ >

and < O';>9n >= (_1)n-1 (x 2 + x-2)n-l. The Jones polynomial can be
recovered from the Kauffman bracket (see [Kau87b]).
(Braid groups) The braid groups were defined by E. Artin in [Art25].
Their presentation, as in Corollary 6.6, is also due to him [Art25][Art47].
The representations described in Exercise 3 were found by Burau in 1936
[Bur36]. The Burau representation is faithful for the braid group B 3 . It has
long been conjectured that the general Burau representation was faithful
too. This was disproven recently by J.A. Moody [Moo91]. It is still an open
question whether the braid groups Bn (for n large) have faithful finite-
dimensional representations at all.

Figure 8.4. Closure of a braid

(Closure of a braid) For any braid (}" E Bn define a link (j by

n
(j = (}" U U ([(k, 0, 0), (k, 1, 1/2)] U [(k, 1, 1/2)]' (k, 0, 1)]).
k=1
 X.9 Appendix. The Fundamental Group 273

The link a, called the closure of (J, is isotopic to the one represented in
Figure 8.4. Alexander [Ale23] showed that any link in R3 was equivalent
to the closure of some braid. Non-equivalent braids may have equivalent
closures. Define an equivalence relation ~ on the set of all braids by: (J ~ (J'
if (J and (J' have equivalent closures. Then Markov's theorem ([Mar36]; for
a proof, see [Bir74]) states that ~ is the equivalence relation generated by
conjugation in the braid groups and by relations of the form (J ~ in ((J )(J;1
where (J E Bn and in is the morphism of Bn to Bn+l defined by in ((Ji) = (Ji
for i = 1, ... , n -1. As a consequence, any family (fn : Bn ----t C)n>O of set-
theoretic maps with values in a set C such that for all n and all (J, T E Bn

gives rise to a unique C-valued isotopy invariant f of links in R3 defined by

f(L) = fn((J) when L is equivalent to the closure of the braid (J E Bn' This
approach was used by V. Jones to construct the polynomial VL in [Jon85]
[Jon87]. It is to be observed that the approach using Markov's theorem is
less elementary than the one by Reidemeister moves.

X.9 Appendix. The Fundamental Group

We briefly recall the definition of the fundamental group of a topological
space. Set f = [0,1].
Let X be a topological space with a distinguished point *. A loop in X
at the point * is a continuous map f : f ----t X such that f(O) = f(l) = *.
Denote the set of such maps by .c*X. Given elements f, gin .c*X we define
their product f g by

f(2t) if 0::; t ::; 1/2,

(fg)(t) = { g(2t - 1) if 1/2::; t ::; 1.

The constant loop e is given by e(t) = *. The inverse f- 1 of f is defined

by f-l(t) = f(l - t) for t E f.
A homotopy from f to g is a continuous map h : f x f ----t X such that

h(O,-)=f, h(1,-)=g, h(s,O)=h(s,l)=*

for all s E f. If such a homotopy exists, we write f '" g. Homotopy is an

equivalence relation. Indeed, it is
(a) reflexive because (s, t) f-+ f(t) is a homotopy from f to itself;
(b) symmetric: if h is a homotopy from f to g, then h(l - s, -) is a
homotopy from g to f;
274 Chapter X. Knots, Links, Tangles, and Braids

(c) transitive: if hI and h2 are homotopies from fl to h and from f2 to

f3 respectively, then

h(s _) = { hl(2s,-) if 0:::; s :::; 1/2,

, h2(2s-1,-) if 1/2 :::; s :::; 1

is a homotopy from fl to f2
We define 7f 1 (X,*) as the set of homotopy classes in L*X. We have the
following lemma.

Lemma X.g.1. Let f, 1', f", g, g' be elements of L*X. Then

(a) f r-v 9 and l' r-v g' imply f l' r-v gg',
(b) (f 1')f" r-v f(f' f"),
(c) fe r-v f r-vef, and
(d) ff- 1 r-v e r-v r
1 f.

PROOF. (a) If h [resp. h'] is a homotopy from f to 9 [resp. from l' to g']'
then (s, t) f---t (h(s, - )h'(s, - ))(t) is a homotopy from ff' to gg'.
(b) A homotopy from (f 1') f" to f (f' f") is given by

f( S~I) if 0< -
t < 8+1
- 4 '
1'(4t - s - 1) if s+1 < < t s+2
4 - - 4 '
f" (4t2~;2) if 05+2
4 -
< < t -
1.

(c) The map

if O<t<s+1
- - 2 '
if s+1
2 -
<t <
-
1

is a homotopy from fe to f. One can also exhibit a homotopy from ef to f.

(d) A homotopy from e to f f- 1 is given by

f(2t) if 0 :::; 2t :::; s,

h(s, t) = { f(s) if s:::; 2t :::; 2 - 8,
f- 1 (2t - 1) if 2 - s :::; 2t :::; 2.

Exchange f and f- 1 to get an homotopy from e to f- 1 f. o

As a consequence, we see that the product vf loops equips 7f 1 (X, *) with
the structure of a group in which the unit is the homotopy class of the con-
stant loop e. This group is called the fundamental group of the topological
space X at the point *.
In the above definitions one may replace continuous maps by piecewise-
linear or by Coo maps when X is an open subset of R 3 or a quotient space
of it. One gets a piecewise-linear or a smooth version of the fundamental
group. These variants are isomorphic to the fundamental group defined
above.
Chapter XI
Tensor Categories

This is our first chapter on tensor categories. As will become apparent in

the sequel, tensor categories form the right framework for representations
of Hopf algebras as well as for the topological objects of Chapter X. They
provide a bridge between quantum groups and knot theory.

XLI The Language of Categories and Functors

We start with a few elementary definitions from category theory.

XL1.1 Categories
Definition XL1.1. A category C consists
(1) of a class Ob(C) whose elements are called the objects of the category,
(2) of a class Hom(C) whose elements are called the morphisms of the
category, and
(3) of maps

identity id Ob(C) ----. Hom(C),

source s Hom(C) ----.Ob(C),
target b Hom(C) ----.Ob(C),
composition o Hom(C) xOb(C) Hom(C) ----.Ob(C),

such that
276 Chapter XI. Tensor Categories

(a) Jor any object V E Ob(C), we have

s(id v ) = b(id v ) = V,
(b) Jor any morphism J E Hom( C), we have

idb(f) 0 J = J 0 ids(f) = J,
(c) Jor any morphisms J, g, h satisJying b(J) = s(g) and b(g) = s(h), we
have
(h 0 g) 0 J= h 0 (g 0 J).
Here Hom( C) x Ob(C) Hom( C) denotes the class of couples (J, g) of com-
posable morphisms in the category, i.e., such that b(J) = 8(g). The conven-
tional notation for the composition of J and 9 is go J or gJ. The object
s(J) is called the source of the morphism J and b(J) is its target. For the
identity morphism of an object V we write id v . We denote by Homc(V, W)
the class of morphisms of the category C whose source is the object V and
whose target is the object W. If J E Homc(V, W), we write

J: V ---+ W or v-.L..w.
A morphism from an object V to itself is called an endomorphism of V.
The class of all endomorphisms of V is denoted End(V). A morphism J
from V to W in the category is an isomorphism if there exists a morphism
g: W ---+ V such that go J = id v and Jog = id w .
Everybody knows (or at least uses) the category Set of sets and the
category Gr of groups. We have already made use of the category Vect(k)
[resp. of Vectf(k)] consisting of vector spaces [resp. of finite-dimensional
vector spaces] and of linear maps over a field k. In Chapter I we used the
category Alg of algebras and the category A-Mod of left A-modules where
A is an algebra. We have also considered the category Cog of coalgebras.
We define the product oj two categories C and V as the category C x V
whose objects are pairs of objects (V, W) E C x V and whose morphisms
are given by

Homcxv((V, W), (V', W')) = Homc(V, V') x Homv(W, W').

A subcategory C of a category V consists of a subclass Ob(C) of Ob(V)
and of a subclass Hom(C) of Hom(V) that rre stable under the identity,
source, target, and composition maps in V.
Let us present two examples of categories that are groupoids, i.e., cate-
gories in which all morphisms are isomorphisms.
Example 1. (Category associated to a Jamily oj groups) Let (Gi)iEI be a
family of groups indexed by a set I. We consider the category 9 defined by
Ob(Q) = I and
if i i= j,
Homg(i,j) = { 0
Gi if i = j,
 XLI The Language of Categories and Functors 277

the composition on Homg (i, i) being given by the multiplication of the

group G i :
hog = gh
where g, hE G i . Note that any morphism gin 9 is an isomorphism whose
inverse morphism is 9 -1.
As a special case, consider a set I reduced to a single element o. We get
a groupoid 9 with one object and with Go as the set of morphisms.
Example 2. (Category of isomorphisms in a category) Let C be a category.
If we set Ob(CiJ = Ob(C) and define Hom(CiJ as the subclass of isomor-
phisms of C, then Cis is a category called the groupoid of isomorphisms
of C.

XI. 1. 2 Functors and natural transformations

Definition XI.1.2. A functor F : C --> C' from the category C to the
category C' consists of a map F : Ob(C) --> Ob(C') and of a map F :
Hom( C) --> Hom( C') such that
(a) for any object V E Ob(C), we have F(id v ) = idF(v),
(b) fOT any morphism f E Hom(C), we have

s(F(f)) = F(s(f)) and b(F(f)) = F(b(f)),

(c) if f, g are composable morphisms in the category C, we have

F(g 0 1) = F(g) 0 F(f).

It is clear that if F : C --> C' and G : C' --> C" are functors, then the
composition G F is a functor from C to C". For any category C, there exists
a functor ide, called the identity functor of C, which is the identity on the
objects and on the morphisms in C. The inclusion of a subcategory in a
category is a functor.

Definition XI.1.3. Let F, G be functors from the category C to the cate-

gory C'. A natural transformation Tf from F to G ~ we write Tf : F --> G ~
is a family of morphisms Tf(V) : F(V) --> G(V) in C' indexed by the objects
V of C such that, for any morphism f : V --> W in C, the square

F(V) ~~ G(V)
IF(f) 1G(f)
F(W) 2~ G(W)
commutes.
If, furthermore, Tf(V) is an isomorphism of c' for any object V in C, we
say that Tf : F --> G is a natural isomorphism.
278 Chapter XI. Tensor Categories

If rt : F -+ G is a natural isomorphism, then the collection of all mor-

phisms rt(V)-l defines a natural isomorphism rt- 1 from G to F. We next
define the important concept of an equivalence of categories.

Definition XI.1.4. Let F : C -+ D be a functor. Then F is an equivalence

of categories if there exist a functor G : D -+ C and natural isomorphisms

rt : id v -+ FG and B: GF -+ idc

We now give a useful criterion for a functor F : C -+ D to be an equiva-

lence of categories. Let us first say that a functor F : C -+ D is essentially
surjective if, for any object W of D, there exists an object V of C such that
F(V) So' W in D. It is said to be faithful [resp. fully faithfUl] if, for any
couple (V, V') of objects of C, the map

F : Homc(V, V') -+ Homv(F(V), F(V'))

on morphisms is injective [resp. bijective].

Proposition XI.1.5. A functor F : C -+ D is an equivalence of categoTies

if and only if F is essentially surjective and fully faithful.

PROOF. (a) Suppose that F is an equivalence. Then there exist a functor

G : D -+ C and natural isomorphisms rt : id v -+ FG and e : GF -+ idc.
The first isomorphism shows that W So' F( G(W)) for any object W of
D. In other words, F is essentially surjective. Now consider a morphism
f: V -+ V/ in C. The square

GF(V) ~ V
1GF(f) 11
GF(V/) ~ v/
commutes. It results that if F(f) = F(f'), hence GF(f) = GF(f'), then
we have f = 1'. Therefore, the functor F is faithful. Using the natural
isomorphism rt in a similar way, we prove that G is faithful too. Now con-
sider a morphism 9 : F(V) -+ F(V/). Let us show that 9 = F(f) where
f = B(V') 0 G(g) 0 B(V)-l. Indeed,
e(V/) 0 GF(f) 0 B(V)-l = f = B(V/) 0 G(g) 0 B(V)-l.

Therefore GF(f) = G(g). As G is faithful, we get 9 = F(f). This proves

that F is fully faithful.
(b) Let F be an essentially surjective and fully faithful functor. For any
object Win D, we choose an object G(W) of C and an isomorphism rt(W) :
W -+ FG(W) in D. If 9 : W -+ W/ is a morphism of D, we may consider

rt(W/) 0 9 0 rt(W)-l : FG(W) -+ FG(W').

 XI.1 The Language of Categories and Functors 279

Since F is fully faithful, there exists a unique morphism G(g) from G(W)
to G(W') such that

FG(g) = 1](W') 0 go 1](W)-l : FG(W) ---> FG(W').

One checks easily that this defines a functor G from V into C and that
1] : id v ---> FG is a natural isomorphism. In order to show that F and G
are equivalences of categories, we have only to find a natural isomorphism
() : GF ---> idc. We define ()(V) : GF(V) ---> V for any object V E Ob(C) as
the unique morphism such that F(()(V)) = 1](F(V))-l. It is easily checked
that this formula defines a natural isomorphism. D

Corollary XI.1.6. Let C be a category and C' a subcategory of C such

that any object of C is isomorphic to an object of C' and such that we have
Homc'(V, V') = Homc(V, V') for all V, V' E Ob(C'). Then the inclusion of
C' into C is an equivalence of categories.
We deduce the following examples of equivalent categories.
Example 3. (The groupoid GL( k)) Let GL n (k) be the group of invertible
matrices of order n with entries in a field k. Set GLo(k) = {l}. By Example
1 we can associate to the family (GLn(k))n>O a groupoid denoted GL(k).
By the previous corollary, the category GL(kj is equivalent to the groupoid
(Vectf(k))is of all finite dimensional k-vector spaces whose morphisms are
the linear isomorphisms.
Example 4. (The groupoid S) Let Sn be the symmetric group of all per-
mutations of the finite set {I, 2, ... , n}. Set So = {I}. Again by the con-
struction of Example 1, we get a groupoid S. The category S is equivalent
to the groupoid (Setf)is of finite sets whose morphisms are bijective.

XL 1.3 Adjoint functors

We end these preliminaries on categories with the concept of adjoint func-
tors. As we may observe from Proposition 1.8, as well from the examples of
this section and the exercises of this chapter, the concept of adjoint func-
tors is nothing but the categorical translation of the idea of a universal
property.
Definition XI.1.7. Let F : C ---> V and G : V ---> C be functors. Then
F is right adjoint to G or G is left adjoint to F if there exist natural
transformations 1] : id v ---> FG and () : G F ---> idc such that the composite
maps
F(V) 7)(F(V))I(FGF)(V) F((}(V))IF(V)

and
G(W) G(7)(W))I(GFG)(W) (}(G(W))IG(W)

are identity morphisms for all objects V of C and W in V.

280 Chapter XI. Tensor Categories

The following characterizes adjoint functors in terms of natural bijec-

tions.
Proposition XI.1.8. Let F : C --> D and G : D --> C be functors. Then F
is right adjoint to G if and only if for all objects V in C and all objects W
in D there exists a natural bijection
<I>(V, W): Home(G(W), V) --> HOffi.D(W,F(V)),
i. e., such that, for all morphisms f in C and all morphisms 9 in D, the
diagram
Homc(G(W') , V) ip(V,W') Homv (W' , F (V))
1
Hom(G(g),v) 1 Hom(g,F(V))
Homc(G(W) , V) ip(V,W) Homv(W, F(V))
1Hom(G(W),f) 1Hom(W,F(f))
Homc(G(W) , V') ip(V',W) Homv(W, F(V'))

commutes where V = s(f), V' = b(f), W = s(g) and W' = b(g).

PROOF. The vertical maps of the above diagram are the obvious maps
obtained by composition with f, F(f), g, and G(g). We sketch the proof
of this proposition. For details, see [Mac71J, Chapter IV.
(a) Let F be a right adjoint to G. Set <I>(V, W)(f) = F(f) o1](W) for any
morphism f : G(W) --> V, and IJ!(V, W)(g) = O(V)oG(g) for any morphism
9 : W --> F(V). Using the definition of adjoint functors, one checks that
the map <I>(V, W) is bijective with inverse IJ!(V, W).
(b) Suppose given the bijections <I>(V, W). We have to construct natural
transformations 1] : id v --> FG and 0 : G F --> ide. They are defined by

1](W) = <I>(G(W), W)(idG(w)) and O(V) = <I>-1(V, F(V))(idF(v)).

The reader will easily check that 1] and 0 are natural transformations. 0

An equivalence of categories is always left and right adjoint to another

equivalence. We give two examples of adjoint functors already encountered
in this book.
Example 5. (Free algebra on a set) Let X be a set and k{X} be the free k-
algebra associated to X as in 1.2. Then X 1---+ k{X} is a left adjoint functor
to the forgetful functor which assigns to any algebra its underlying set.
Example 6. (Tensor products) Any vector space V determines two functors
F, G from the category of vector spaces into itself: F(U) = Hom(V, U) and
G(U) = U V. The natural isomorphism
Hom(U V, W) ~ Hom(U, Hom(V, W))
of Corollary 11.1.2 shows that G is left adjoint to F.
 XI.2 Tensor Categories 281

XI. 2 Tensor Categories

XI.2.1 Definitions
Let C be a category and 129 be a functor from C x C to C. This means that
(a) we have an object V 129 W associated to any pair (V, W) of objects of
the category,
(b) we have a morphism j 129 9 associated to any pair (f, g) of morphisms
of C such that

s(f 129 g) = s(f) 129 s(g) and b(f 129 g) = b(f) 129 b(g),

(c) if l' and g' are morphisms such that s(f') = b(f) and s(g') = b(g),
then
(f' 129 g') 0 (f 129 g) = (f' 0 1) 129 (g' 0 g), (2.1)

(d) and
idv@w = id v 129 id w . (2.2)
Relation (2.1) implies that

j 129 9 = (f 129 idb(g)) 0 (ids (f) 129 g) = (idb(f) 129 g) 0 (f 129 ids(g))' (2.3)

Example 1. Let C = Vect(k) be the category of vector spaces over a field

k. Then the tensor product of vector spaces (see II.1-2) defines a functor
from C x C to C.
Any functor 129 : C xC ---+ C will be called a tensor product by analogy with
Example 1. Let C be a category with a tensor product 129. An associativity
constraint for 129 is a natural isomorphism

a: 129(129 x id) ---+ Q9(id x 129).

This means that, for any triple (U, V, W) of objects of C, there exists an
isomorphism
auvw .
, : (U 129 V) 129 W ---+ U 129 (V 129 W) (2.4)
such that the square
aU,V,lV
(U 129 V) 129 W U 129 (V 129 W)
1
----t

(f@g)@h If@(g@h) (2.5)

au'.v',W'
(U' 129 V') 129 w' )
u' 129 (V' 129 W')
commutes whenever j, g, hare morphisms in the category.
282 Chapter XI. Tensor Categories

The associativity constraint a satisfies the Pentagon Axiom if the pen-

tagonal diagram
au,v,w<8Iidx
( U 0 (V 0 W)) 0 X ((U0V)0W) 0X
1 aU0V,W,X

luvwx (U 0 V) 0 (W 0 X) (2,6)
1 aU,V,W0X

U 0 ((V 0 W) 0 X) idu<8lav,w,x
U 0 (V 0 (W 0 X) )

commutes for all objects U, V, W, X of C.

Fix an object I in the category. A left unit constraint [resp. a right unit
constraint] with respect to I is a natural isomorphism

l : 0(I x id) -> id [resp. r : 0(id x I) -> id].

This means that for any object V of C there exists an isomorphism

lv : 10 V -> V [resp. rv : V 0 I -> V] (2.7)

such that
Iv TV
I0V ---+ V V0I ---+ V
1 idI <8I1 11 [resp . 1 / <81 id I 11 ] (2.8)
I0V' ~ V' V'0I ~ V'

commutes for any morphism f.

Given an associativity constraint a, and left and right unit constraints
l, r with respect to an object I, we say that they satisfy the Triangle Axiom
if the triangle
aV,I,W
(V01) 0 W ------+ V0 (10 W)
~ Tv<8Iidw .,/ idv<8llw (2.9)
V0W
commutes for all pairs (V, W) of objects.

Definition XI.2.1. A tensor category (C, 0, I, a, l, r) is a categoryC which

is equipped with a tensor product 0 : C x C -> C, with an object I, called the
unit of the tensor category, with an associativity constraint a, a left unit
constraint l and a right unit constraint r with respect to I such that the
Pentagon Axiom (2.6) and the Triangle Axiom (2.9) are satisfied.
The tensor category is said to be strict if the associativity and unit con-
straints a, l, r are all identities of the category.
Examples of tensor categories will be given in Section 3.
 XI.2 Tensor Categories 283

XI.2.2 Properties of the unit

Let (C, Q9, I, a, l, r) be a tensor category. We state a few properties of the
unit I.
Lemma XI.2.2. The triangles
(I Q9 V) Q9 W ~ I Q9 (V Q9 W)
'\., Iv<8Iidw ,/' IVQ1Iw
VQ9W

and
(V Q9 W) Q9 I ~ V Q9 (W Q9 1)
'\., rVQ1IW ,/' idv<8Irw
VQ9W
commute for any pair (V, W) of objects of C.
PROOF. Consider the diagram

( U Q9 (I Q9 V)) Q9 W
/ ' a<8lidw

( (U Q9 1) Q9 V) Q9 W
'\., (ru<8lid v) (id U <8lIVV
<8Iidw <8Iidw

I
(U Q9 V) Q9 W
a la a

U Q9 (V Q9 W)

/ ' idvQ1IW
ru<81 id u <81 "
(lv<8Iidw)"
(U Q9 1) Q9 (V Q9 W) id u Q9 lV<8IW U Q9 ((I Q9 V) Q9 W)
'\., a ,/' idu <8Ia

UQ9 (I Q9 (V Q9 W) )
Here we dropped the subscripts of the associativity constraint a for sim-
plicity. The outside hexagon commutes by the Pentagon Axiom (2.6). The
naturality (2.5) of a implies the commutativity of the two middle squares
whereas (2.9) implies the commutativity of the top square and of the lower
left triangle. Consequently, the lower right triangle commutes as well. Set-
ting U = I, we get
id J Q9 (lV<8IW 0 a) = id J Q9 (lv Q9 id w )

This relation, together with the naturality of the left unit constraint (2.8)
and the fact that l is an isomorphism, implies lV<8IWoa = lvQ9id w , which ex-
presses the commutativity of the upper triangle in the statement of Lemma
XI.2.2. A similar proof works for the other triangle. 0
284 Chapter XI. Tensor Categories

Lemma XI.2.3. Let I be the unit of a tensor category. For any object V
we have

PROOF. By naturality (2.8) of l, we have lv 0 lUi9V = lv 0 (id I lv). Since

lv is an isomorphism, we get the first equality of the lemma. The second
one is similarly a consequence of the naturality of r.
Let us prove that II = rio By Lemma 2.2 and the first equality of Lemma
2.3, we have
l I id I = l I 0I 0 a = (id I l I) 0 a.
From (2.9) we get rI id I = (id I lI) 0 a. Combining both relations yields
II id[ = r[ id I . This implies II = rI in view of the fact that r is a
natural isomorphism. 0

We are now ready to prove the main result of this subsection.

Proposition XI.2.4. The set End(I) of endomorphisms of the unit object

I is a commutative monoid for the composition. Moreover, for any pair
(f, g) of endomorphisms of I, we have

f 9 = 9 f = rIl 0 (f 0 g) 0 rI = rIl 0 (g 0 f) 0 rio

In other words, if we identify I I with I via r I = l [, then the tensor

product of morphisms coincides in End(I) with their composition.
PROOF. The composition equips End(I) with the structure of a monoid
whose unit is idI . Let us prove that it is commutative. By (2.8) we have

fidI=rIlofor[ and idIg=lIlogol[.

Combining r[ = l[ of Lemma 2.3 with Relation (2.3) implies that

f 9 = rIl 0 (f 0 g) 0 rI = rIl 0 (g 01) 0 r[ = 9 f.

It follows that fog = go f. o

XI. 3 Examples of Tensor Categories

In this book we shall be concerned with two main types of tensor cate-
gories. The first type is built on vector spaces and their tensor products as
introduced in Chapter II. The second one uses the I-dimensional objects of
Chapter X such as links, tangles and braids. We shall connect both types
in Chapter XII.
 XI.3 Examples of Tensor Categories 285

XI.3.1 Tensor categories of vector spaces

The most fundamental example of a tensor category is given by the category
C = Vect( k) of vector spaces over a field k. The latter is equipped with a
tensor structure for which (9 is the tensor product (defined in 11.1) of the
vector spaces over k, the unit object I is the ground field k itself, and the
associativity and unit constraints are the natural isomorphisms

a((u(9v)(9W) =u(9(v(9w) and l(1(9v)=v=r(v(91) (3.1)

of Proposition 11.1.3. The pentagon and triangle axioms are clearly satisfied.
There are some important examples of subcategories of Vect(k) preserv-
ing the tensor structure. For instance, if G is a group, then the category
k[G]-Mod of representations of Gover k, or, equivalently, of k[G]-modules,
is a subtensor category of Vect(k) where the tensor product U (9 V of two
G-modules and the field k are given the following G-structures:

g(u (9 v) = gu (9 gv and g)... = )..

for 9 E G, u E U, v E V and)", E k.
We know from Chapter III that the group algebra k[G] is an associative
algebra over k with a comultiplication and a counit. We now investigate
such types of algebras. Let A be an associative unital k-algebra with a
morphism of algebras ~ : A ----+ A (9 A, called the comultiplication, and a
morphism of algebras E : A ----+ k, called the counit. Let us denote by A-
Mod the category of left A-modules (alias, representations of A). If U, V
are left A-modules, the tensor product U (9 V is a left A (9 A-module. The
comultiplication allows to pull back this A (9 A-module structure into an
A-module structure. It is given by

a(u (9 v) = ~(a)(u (9 v) (3.2)

for a E A, u E U and v E V. We endow k with an A-module structure

given by
a)... = E(a) .... (3.3)
It is now clear that the tensor product in Vect(k) restricts to a functor

(9 : A-Mod x A-Mod ----+ A-Mod

for which I = k is a unit. The following characterizes bialgebras in terms

of their categories of modules.

Proposition XI.3.1. Let A = (A,~, E) be an algebra with comultiplica-

tion and co unit as above. It is a bialgebra if and only if the category A-Mod
equipped with the tensor product described above and the constraints a, l, r
of Vect(k) is a tensor category.
286 Chapter XI. Tensor Categories

PROOF. Let (A, 'P, r], L.\, c) be a bialgebra. It follows from Proposition III. 5. 1
that (A-Mod, 0,1 = k, a, l, r) is a tensor category.
Conversely, let (A, 'P, r], L.\, c) be an algebra with comultiplication and
counit. Suppose that (A-Mod, 0, 1= k, a, l, r) is a tensor category. Let us
prove that L.\ is coassociative and that 10 is a counit in the sense of Definition
III. 1. 1.
Let us start with the coassociativity of L.\. Consider the associativity
constraint aA A A- By hypothesis, it is A-linear, which means that for
a,u,v,w E A ~~ have

aA,A,A (a (( u 0 v) 0 W) ) = a aA,A,A ( (u 0 v) 0 w) .
By definition of the associativity constraint, this can be reexpressed as

(L.\ 0 id)(L.\(a))(u 0 (v 0 w)) = (id 0 L.\)(L.\(a))(u 0 (v 0 w)).

Setting u = v = w = 1 E A, we get

(L.\ 0 id)(L.\(a)) = (id 0 L.\)(L.\(a)).

Similarly, lA is A-linear if and only if (10 0 id)(L.\(a)) a, and r A is

A-linear if and only if (id 0 c) (L.\(a)) = a for all a E A. o

XI.3.2 Tensor categories built on groups

We now give examples of strict tensor categories. Let (Gi)iEN be a family
of groups indexed by the monoid N of nonnegative integers. We may form a
category 9 as in Section 1, Example 1. Suppose that Go = {I}, and for any
pair (n, m) of integers we have a group morphism Pn,m : G n x G m ---> G n+m .
We now define a tensor product on the category 9 by n 0 m = n + m and,
if 9 E G nand h E G m' we set

90 h = Pn,m(g, h) E G n+m

Check that (9,0, I = 0, a = id, 1 = id, r = id) is a strict tensor category

provided the morphisms Pn,m are subject to the relations

Pn+m,p 0 (Pn,m 0 id G ) = Pn,m+p 0 (id Gn 0 Pm,p) (3.4)

and PO,n = Pn,o = id Gn after natural identification. This construction can

be applied to the following families of groups.
(a) Consider the groupoid GL(k) of Section 1, Example 3 built from the
family of groups (GLn(k)). Define maps Pn,m by
 XI.4 Tensor Functors 287

Relations (3.4) are satisfied. The category GL(k) becomes a tensor category
in this way.
(b) A family of subgroups {Gn}n of GLn(k) preserved by the maps Pn m
gives also rise to a strict tensor category. For instance, take the family 'of
symmetric groups Sn' the latter being realized as a subgroup of GLn(k) via
the permutation matrices. The resulting category S is a tensor category.

XI. 4 Tensor Functors

Definition XI.4.1. (a) Let C = (C, 0, I, a, l, r) and V = (V, 0,1, a, l, r)
be tensor categories. A tensor functor from C to V is a triple (F, 'Po, '(2)
where F : C -7 V is a functor, 'Po is an isomorphism from I to F(I), and

'P2(U, V) : F(U) 0 F(V) -7 F(U 0 V)

is a family of natural isomorphisms indexed by all couples (U, V) of objects

of C such that the diagrams

(F(U) 0 F(V)) 0 F(W) aF(U),F(V),F(W) I F(U) 0 (F(V) 0 F(W))

1 '1'2 (U,v)<ZiidF(w) 1id F (u)<Zi'P2(V,W)

F(U 0 V) 0 F(W) F(U) 0 F(V 0 W)

1 '1'2 (U<ZiV,W) 1'1'2 (U,v<ZiW)

F(au,v,w)
F((U0V)0W) I F(U0 (V 0 W))

IF(u)
(4.1)
10 F(U) ----> F(U)

1 'Po<ZiidF(u) rF(lu) (4.2)

'1'2 (I,U)
F(I) 0 F(U) I F(I0 U)

and rF(U)
F(U) 0 I ~ F(U)

1 idF(u)<Zi'Po rF(ru) (4.3)

'1'2 (U,I)
F(U) 0 F(I) I F(U 01)

commute for all objects (U, V, W) in C. The tensor functor (F, 'Po, '(2) is
said to be strict if the isomorphisms 'Po and 'P2 are identities of V.
(b) A natural tensor transformation 'rJ : (F, 'Po, '(2) -7 (F', 'P~, 'P~) be-
tween tensor functors from C to V is a natural transformation 'rJ : F -7 F'
288 Chapter XI. Tensor Categories

such that the following diagrams commute for each couple (U, V) of objects
in C:

F(I) F(U) F(V)

1
/ 'Po

I 1rM) and 7J (UV)

' " 'PS

F'(I) F'(U) F'(V) 'P;(U~ F'(U V)
(4.4)
A natural tensor isomorphism is a natural tensor transformation that is
also a natural isomorphism.
(c) A tensor equivalence between tensor categories is a tensor functor
F : C ----+ V such that there exist a tensor functor F' : V ----+ C and natural
tensor isomorphisms rl : id v ----=-. F F' ande:
F' F ----=-'idc .

In case there exists a tensor equivalence between C and V, we say that

C and V are tensor equivalent. Observe that if (F, CPo, CP2) and (F', CP;j, cp;)
are tensor functors, then so is the composition (F' F, F'(cpo)cp~, F'(CP2)CP;)'
The identity functor is a strict tensor functor.
We denote by Tens(C, V) [resp. Tensstr(C, V]) the category whose objects
are the tensor functors [resp. the strict tensor functors] from C to V and
whose morphisms are the natural tensor transformations.
Example 1. Let A be a bialgebra. The forgetful functor associating to an
A-module its underlying vector space is a strict tensor functor.
Example 2. Let f : Al ----+ A2 be a map of bialgebras. Given an A 2 -module
V we can equip V with a A1-module structure given by a v = f(a)v for
a E Al and v E V. This construction yields a strict tensor functor 1* from
A2-Mod to A1-Mod.
We shall encounter our first examples of non-strict tensor functors in
Chapter XV devoted to quasi-bialgebras.

XI. 5 Turning Tensor Categories into Strict Ones

Since the tensor product in a tensor category is associative only up to iso-
morphism, one has to keep track of parentheses very carefully. This is rather
fastidious and should be avoided as much as possible. We now indicate a
way out of this problem: given a tensor category C = (C,,I,a,l,r), we
c
construct a strict tensor category str which is tensor equivalent to C. This
is done as follows.
Let S be the class of all finite sequences S = (V1, ... , Vk ) of objects of C,
including the empty sequence 0. The integer k is by definition the length
of the sequence S = (V1, ... , Vk)' The length of the empty sequence is 0 by
 XI.5 Thrning Tensor Categories into Strict Ones 289

convention. If S = (VI'.' Vk ) and S' = (Vk+I' ... ' Vk+n) are nonempty
sequences of S, we denote by S * S' the sequence
(5.1)
obtained by placing S' after S. We also agree that S * 0 = S = 0 * S. To
any sequence S of S, we assign an object F(V) of C defined inductively by

F(0) = I, F((V)) = V, F(S * (V)) = F(S) V.

In other words,

where all opening parentheses are placed on the left-hand side of VI.
c
We are now ready to define the category str : its objects are the elements
of S, i.e., the finite sequences of objects of C, and its morphisms are given
by
Homes,,(S, S') = Homc(F(S) , F(S')).
This defines a category whose identities and composition are taken from C.
c
The rest of the section is devoted to the proof that str is a strict tensor
category equivalent to C.
Proposition XI.5.1. The categories cstr and C are equivalent.
PROOF. The map F defined above on the objects of str extends to a c
c
functor F : str -+ C which is the identity on morphisms, hence fully
faithful. As any object in C is clearly isomorphic to the image under F of a
sequence of length one, we see that F is essentially surjective. This proves
the proposition in view of Proposition 1.5. Observe that G(V) = (V) defines
a functor G : C -+ cstr which is the inverse equivalence to F. Indeed, we
have FG = ide and 0 : GF -+ ides" via the natural isomorphism

O(S) = idF(s) : GF(S) -+ S.

o
c
We now equip str with the structure of a strict tensor category. Defining
the tensor product on objects of cstr is easy: we set S s' = S * S'. It is
clearly associative on objects.
c
In order to define the tensor product of two morphisms of str , we first
construct a natural isomorphism

cp(S, S') : F(S) F(S') -+ F(S * S')

c
for any pair (S, S') of objects in str . This isomorphism is defined by in-
duction on the length of the sequence S'. First, we set cp(0, S) = ls and
cp(S,0) = rs. Next,
cp(S, (V)) = idF(s)v : F(S) V -+ F(S (V))
290 Chapter XI. Tensor Categories

and
rp(5,5' * (V)) = (rp(5,5')@id v )oa-;;}S),F(S'),v' (5.3)
The following lemma will be used in the proof of Theorem 5.3.

Lemma XI.5.2. If 5, 5', 5" are objects on cst", we have

rp(5, 5' * 5") 0 (ids @ rp(5', 5")) 0 aF(S),F(S'),F(S")

= rp(5 * 5',5") 0 (rp(5, 5') @ ids")'

PROOF. We proceed by induction on the length of 5". If 5" = 0, we have

rp(5, 5')(id s @ rp(5', 0)) aF(S).F(S'),I
. rp(5, 5')(id F (s) @ rF(S') aF(S),F(S').I
rp(5, 5') rF(S)F(S')
r F(S*S') (rp(5, 5') @ id J )
rp(5 * 5', 0)(rp(5, 5') @ id J ).
The first and last equalities are by definition, the second one by Lemma
2.2, and the third one by naturality of r.
Let V be an object of the category. Let us prove that the equality of
Lemma 5.2 for the triple (5,5',5") implies the equality for (5,5',5" * (V)).
We have

rp(5, 5'* 5" * (V))(id s @ rp(5', 5" * (V))) aF(S),F(S'),F(S"*(V

(rp(5, 5' * 5") @id v ) a-;;'~S),F(S'*S"),v (ids @ rp(5', 5") @id v )

(ids @ a-;;'~S,),F(S"),v) aF(S),F(S'),F(S")V

(rp( 5,5' * 5") @ id v ) (ids @ rp(5', 5") @ id v ) a -;;'~S),F(S')F(S"),v

(ids @ a-;;'~S,),F(S"),v) aF(S),F(S'),F(S")V
(rp(5, 5' * 5") @ id v ) (ids @ rp(5', 5") @ id v )

(aF(S),F(S'),F(S") @ idv ) a-;;'~S)F(S,),F(S"),v

(rp(5 * 5',5") @ idv ) (rp(5, 5') @ ids" @ idv ) a-;;'~S)F(S,),F(S")'V

(rp(5 * 5',5") @idv ) a-;;'~S*S,),F(S"),v (rp(5, 5') @ids " @idv )
rp(5 * 5',5" * (V)) (rp(5, 5') @ ids"*(V).

The first and last equalities follow from (5.3), the second and fifth ones
from the naturality of the associativity constraint, i.e., from Relation (2.5),
the third from the Pentagon Axiom (2.6), and the fourth one from the
induction hypothesis. 0
 XI.6 Exercises 291

We can now define the tensor product f * f' of two morphisms f : S --; T
c
and f' : S' --; T' of str . By definition, f is a morphism from F(S) to F(T)
and f' is another one from F(S') to F(T') in C. We define the tensor
product f * g in cstr by the commutative square
<pcs,S')
F(S) F(S') )
F(S * S')
1 f @f'
<pCT,T')
1 f *f' (5.4)
F(T) F(T') )
F(T * T') .

c
Theorem XI.5.3. Equipped with this tensor product str is a strict tensor
c
category. The categories C and str are tensor equivalent.

PROOF. It is easy to check that the above-defined * is a functor. This

c
functor is strictly associative by construction. Therefore str is a strict
tensor category.
In order to prove that it is tensor equivalent to C, we have to exhibit
tensor functors and natural tensor isomorphisms. We first claim that the
c
triple (F, id I' ~) is a tensor functor from str to C where ~ is the natural
isomorphism defined above. Indeed, Lemma 5.2 is a reformulation of Rela-
tion (4.1) while Relations (4.2-4.3) follow from the definition of ~(S, 0) and
of ~(0, S). The functor G of the proof of Proposition 5.1 is a strict tensor
functor. Finally, the natural isomorphism e is a natural tensor isomorphism.
o
Theorem 5.3 implies Mac Lane's coherence theorem which states that
in a tensor category any diagram built from the constraints a, l, r, and
the identities by composing and tensoring, commutes. In other words, the
commutation of all such diagrams is equivalent to the commutation of the
pentagon (2.6) and of the triangle (2.9).

XI. 6 Exercises
1. Let f be a pre-ordered set, i.e., a set with a binary relation s:; such
that x s:; x, and (x s:; y and y s:; z) =? x s:; z. Set Ob(..1) = f,
Hom(..1) = {(x, y) E f x fix s:; y}, s(x, y) = x, b(x, y) = y, and
(y, z) 0 (x, y) = (x, z). Show that these data define a category ..1.
2. Prove that the class of all categories form a category Cat whose ob-
jects are the categories and whose morphisms are the functors.

3. Prove that the class of functors form a category Funet whose objects
are the functors and whose morphisms are the natural transforma-
tions between functors.

4. Express in terms of adjoint functors the following natural bijections:

292 Chapter XI. Tensor Categories

(a)
HOmAlg(k[G], A) ~ HomGl(G, A X)

where G is a group, k a commutative ring, A an algebra, and

A x the group of invertible elements in A.
(b)
HOmAlg(U(L), A) ~ HomLie(L, L(A))
where Lie is the category of Lie algebras, L an object of Lie,
A an algebra, L(A) the underlying Lie algebra, and U(L) the
enveloping algebra of L (see V.2).
(c)
HomCog(k[XJ, C) ~ Homset(X, G(C))
where X is a set, k[X] the corresponding coalgebra (see III.l,
Example 3), and G(C) the set of grouplike elements of C.

5. Let I be a set and Vect I the category whose objects are families
(Ui)iEI of vector spaces indexed by I. The set of morphisms in Vece
from (UJi to ell;)i is the product set TIiEI Hom(Ui , Vi). For any vec-
tor space U, we set 6.(U) = (Ui)i where Ui = U for all i E I. Show
that 6. defines a functor from Vect to Vect I and that the direct
sum EB and the direct product II of vector spaces define functors
EB, II : Vect I -+ Vect. Prove that the diagonal functor 6. is right
adjoint to the functor EB and left adjoint to the functor II.

6. Let R be the category of commutative rings without zero divisors and

F the category of fields. Show that the correspondence assigning to
any ring in R its field of fractions is a functor from R into F which
is left adjoint to the "forgetful" functor.

7. Let G be a group and Q be the corresponding category (as in Section

1, Example 1). For any x E G define a functor Ad x from Q to itself
by Ad x (g) = xgx -1. Show that there exists a natural isomorphism
from the functor Ad x to the identity functor.

S. Let Vect gr (k) be the category of nonnegatively graded vector spaces

over a field k with linear maps of degree zero. Equip it with the graded
tensor product (see Chapter III, Exercise 3). Define constraints a, I, r
by
a((u v) w) = a(m, n,p)u (v w),
1(1 v) = .>..(n)v, r(v 1) = p(n)v
where u, v, ware homogeneous vectors of respective degrees m, n,p
and where a,.>.., p are functions on N with values in k \ {a}. Show
 XI.7 Notes 293

that these constraints satisfy the Pentagon and the Triangle Axioms
if and only if a, A, p satisfy the functional equations

a(n,p, q)a(m + n,p, q)-la(m, n + p, q)a(m, n,p + q)-la(m, n,p) = 1

and a(m,O,p) = p(m)A(p)-l for all integers m,n,p,q.

9. Show that the subcategory Vectf(k)is of all finite-dimensional vector

spaces of Vect(k) with their isomorphisms is tensor equivalent to the
tensor category GL(k) of 3.2.

XI. 7 Notes
Tensor categories were introduced in 1963 by Benabou [Ben63]. See also
[Mac63] where constraints as well as the Pentagon and the Triangle Axioms
were defined. Tensor categories are also called monoidal categories in the
literature. Our terminology is taken from Joyal and Street [JS91a] [JS93].
Lemma 2.2 is due to Kelly [Ke164]. For a proof of Mac Lane's coherence
theorem, see [Mac63] [Mac71]. Exercise 8 was taken from [Ke164].
Chapter XII
The Tangle Category

The aim of this chapter is to set up a categorical construction of isotopy

invariants of links. To this end, we build a strict tensor category T out of
the tangles defined in X.5. Any strict tensor functor from T to a category of
finite-dimensional vector spaces gives rise to an isotopy invariant. Using a
presentation of T by generators and relations, we shall reduce in Section 4
the construction of such a functor to an algebraic data, called an enhanced
R-matrix, consisting of a finite-dimensional vector space, an R-matrix, and
a compatible automorphism. We shall apply this method in Section 5 to
exhibit explicit isotopy invariants that will allow us to complete the proof
of Theorem X.4.2 asserting the existence of the Jones-Conway polynomial.
We start with the notion of a presentation of a strict tensor category by
generators and relations.

XII. 1 Presentation of a Strict Tensor Category by

Generators and Relations
One of the most efficient way of comprehending a group G is to present it
by generators and relations. Recall the following facts: Let F be a subset
of G and R be a set of pairs of words in the alphabet F. Then (F, R) is a
presentation of the group G if the two following conditions are satisfied:
(i) the subset F generates G, and
(ii) two words a and b in the alphabet F represent the same element in G
 XII.l Presentation of a Strict Tensor Category 295

if and only if one may pass from a to b by operations replacing any subword
of the form c by a subword of the form d where (c, d) belongs to R.
Example 1. The abelian group Z2 has a presentation (F, R) where

F = {x,y} and R = {(xy,yx)}.

As an application of group presentations, we see we can define a group
morphism by its restriction to a generating subset. In fact, let (F, R) be a
presentation of a group G. Suppose we have a map fo : F ----> H with values
in another group H. It extends to a multiplicative map, still denoted fo,
from the set of words in the alphabet F to H. Then there exists a unique
morphism of groups f : G ----> H restricting to fo on F if and only if
fo(c) = fo(d) for any element (c, d) in R.
A similar formalism works for tensor categories. Its description is the
main objective of this section.
Let (C, Q9, J) be a strict tensor category and F be a collection of mor-
phisms of C. We wish to define certain symbols which we shall call words
in F. Any word a will possess subwords and will be assigned a morphism
Ii of C. We say that the word a represents the morphism Ii of C.
By definition, a word of rank 1 is a symbol of the form [J] where f is an
element of F or of the form [id v ] where V is an object of C. We define the
morphism of C represented by such symbols by [J] = f and [id v ] = id v .
By definition, a subword of a word of rank 1 is the word itself.
Suppose defined all words of rank:::; n where n ?: 1, the morphisms they
represent and their subwords. Define a word of rank n + 1 as a symbol of
the form a 0 b or of the form a Q9 b where a and b are words of rank :::; n.
We define the corresponding morphism by setting

(1.1)
where the symbols 0 and Q9 in the right-hand sides denote the composition
and the tensor product in the tensor category C respectively. The subwords
of a 0 b and those of a Q9 b consist of the word itself, the subwords of a and
those of b.
The class of words in F is the union of all words of positive rank. We
introduce an equivalence relation on words.
Definition XII.1.1. Two words a and a' in F are equivalent if there exist
words a o = a, a 1, ... ,ak = a' such that for all i, the word ai+l is obtained
from a i by replacing a subword x of one of them by a subword y of the other
where x and yare the two sides of any of the following relations:

([f] 0 [g]) 0 [h] rv [J] 0 ([g] 0 [h]), (1.2)

[idb(nl 0 [f] rv [f], [f] 0 [ids(f)] rv [f], (1.3)

[id v ] 0 [id v ] rv [id v ], (1.4)
296 Chapter XII. The Tangle Category

([J] Q9 [g]) Q9 [h] ~ [J] Q9 ([g] Q9 [h]), (1.5)

[id]] Q9 [J] ~ [J], [J] Q9 [id]] ~ [J], (1.6)
[id v ] Q9 [id w ] ~ [idvw], (1.7)
([f] Q9 [g]) 0 ([J'] Q9 [g']) ~ ([J] 0 [1']) Q9 ([g] 0 [g']) (1.8)
'Where V, Ware objects of C and f, 1', g, g', h are elements of F.
We write a ~ b if a and b are equivalent words. Observe that if a ~ b,
then a = b holds for the corresponding morphisms in C. The following
lemma gives examples of equivalent words.

Lemma XII.1.2. (a) If f, 9 E F, then

(b) If fl' f 2 , , h E Fare morphisms such that b(fJ = S(ji+l) for

all i, then

([id v ] Q9 [Jl] Q9 [id w ]) 0 ([id v ] Q9 [J2] Q9 [id w ]) 0 0 ([id v ] Q9 [Jk1 Q9 [id w ])

is equivalent to [id v ] Q9 ([Jl] 0 [J2] 0 0 [Jk]) Q9 [id w ]

(c) Any 'Word in :F is equivalent to a 'Word of the form [id v ] or of the
form

PROOF. (a) By (1.3) and (1.8) we have the equivalences

([J1 Q9 [idb(g)]) 0 ([ids(f)] Q9 [g]) ~ ([J1 0 [ids(f)D Q9 ([idb(g)] 0 [g])

~ [J1 Q9 [g]
~ ([idb(nl 0 [fD Q9 ([g] 0 [ids(g)D
~ ([idb(f)] Q9 [g]) 0 ([J1 Q9 [ids(g)])'

(b) We proceed by induction on k. For k = 1, the statement is clear. For

k > 1 the induction hypothesis, Relations (1.2), (1.4), (1.8), and Part (a)
imply that

([idv1 Q9 [fl] Q9 [idwD 0 ([idvl Q9 [J21 Q9 [idwD 0'" 0 ([idvl @ [fd @ [idwD
~ ([idv1 Q9 ([f l ] 0 [f2 10 0 [Jk-ID Q9 [id w ]) 0 ([id v ] @ [fk1 @ [idwD
~ (([id v ] Q9 ([Jl] 0 [J2] 0 0 [Jk-lD)@ [idw1)0 (([id v ] @ [h]) @ [id w ])

~ (([id v ] @ ([Jl] 0 [J2] 0 0 [Jk-l])) 0 ([id v ] Q9 [hD) @ ([id w ] 0 [id w ])

~ [id v ] @ ([f l ] 0 [f2] 0 ... 0 [Jk]) Q9 [id w ].
 XII.1 Presentation of a Strict Tensor Category 297

(c) We prove the assertion by induction on the rank of words. When a

word is of rank 1, then it is [f] where f = id v or f E :F. In both cases it is
equivalent to [id I ] [J] [idI ]. Suppose the assertion proved for all words
of rank ::; n. Let a be a word of rank ::; n + 1. If a = b 0 c, then, by the
induction hypothesis, the words band C are equivalent to identities or to a
word of the above form. In both cases, a is equivalent to an identity or to
a word of the desired form.
Now consider the case when a = b c. Let us restrict to the interesting
case where band c are not equivalent to identities. Then,

where the words bl ... bk , cl , . . . ,ce are of the form [id v ] 0 [f] 0 [id w ] for
some f E :F. Set S = s(b k ) and T = b(c l ). Then by (1.3) and Part(a) we
get

a = b c '" (b 0 [idsr fl ) ([idT]ok 0 c)

(b l [idT]) 0 ... 0 (b k [idT]) 0 ([ids] CI) 0 . . . 0 ([ids] cfI)'

which is equivalent to the desired form in view of

for some f, l' E:F. The last two equivalences follow from (1.7). D

Composing and tensoring words are operations that are compatible with
the above-defined equivalence relation. Denote by M(F) the class of equiv-
alence classes of words in F. We define a strict tensor category C(F) as
follows. The objects of C(F) are the objects of C whereas M(F) is the
class of morphisms in C(F). The identity, source, and target maps for C(F)
are given by

id v = [id v ], s(a) = s(O:), b(a) = b(O:).

The composition and the tensor product of words have already been de-
fined.
The map sending a word a to the morphism 0: of C is a strict tensor
functor from C(F) to C. When this functor is an equivalence of categories,
we say that the strict tensor category C is free on the class F. In view of
Proposition XL1.5, this is equivalent to

for any pair (a, b) of words in F.

We also say that F generates the strict tensor category C if any morphism
in C can be obtained from morphisms in F and from identities of C by
applying finitely many times the operations of composing and tensoring.
298 Chapter XII. The Tangle Category

If F generates C, then any morphism of C is of the form a where a is an

element of M (F).
We now wish to introduce further relations on M(F). In addition to F,
we also choose a collection R of pairs (c, d) of words in F such that c = d
in C. Using R we may put a new equivalence relation on M(F). Given two
elements a, a' of M(F), we say that a and a' are congruent modulo R -
we write a "'n a' - if there exist words a o = a, a 1 , ... ,ak = a' such that
for all i, ai+l is obtained from a i by replacing a subword c of one of them
by a subword d of the other one where (c, d) is an element of R.
We are now ready to define the presentation of a strict tensor category
by generators and relations.

Definition XII.1.3. The strict tensor category C is generated by :F and

by the relations R if
(a) the set F generates C, and
(b) for any pair (a, a') of elements of M (F) we have the equivalence

a "'n a' {==} a = a'.

The main interest of this definition lies in the following proposition stat-
ing under which conditions one can define a functor on C by its restriction
to the generating set F.

Proposition XII.1.4. Let C be a strict tensor category generated by the

family of morphisms F and the relations R. Suppose given a strict tensor
category V, a map Fo : Ob(C) --+ Ob(V) such that Fo(I) = I and

Fo(V V') = F()(V) Fo(V')

for all couples (V, V') of objects of C, and a morphism 9 1 from Fo (s(J)) to
Fo (b(J)) for any morphism f E F. Then there exists a unique strict tensor
functor F : C --+ V such that we have F(V) = Fo(V) for any object V of
C and F(J) = 91 for any morphism f in F, if and only if for any couple
(c, d) ofR we obtain equal morphisms in V after replacing any subword [fl
(J E F) of c and d by gl' and any subword [idvl by idFo(V)'
PROOF. The implication =} is clear since the words c and d represent the
same morphism in C. Therefore their images under F obtained after per-
forming the substitutions indicated above are identical.
Let us prove the reverse implication. The uniqueness of F follows from
the fact that the family F generates C. It remains essentially to define F
on the morphisms of C. Now any morphism in C can be represented by an
element a E M(F). We define Fl(a) by replacing any subword [fl(J E F)
of the word a by gl and any subword [idvl by idFo(v)' By definition of
the presentation of C, the words a and a' represent the same morphism in
C if and only if a and a' are congruent modulo R. The substitution law
stated in Proposition 1.4 implies that, if a and a' are congruent modulo R,
 XII.2 The Category of Tangles 299

then Fl(a) = Fl(a'). Setting F(a) = Fl(a) defines F unambiguously on

the morphisms of C. 0

Proposition 1.4 will be used as an essential tool in Section 4. We end this

section with a technical result. Suppose that the strict tensor category C is
generated by a set F and the relations n. Suppose also that there exists
a subset F' of F such that any f E F' is congruent modulo n to a word
aU) in M(Fo) where Fo = F \ P. Denote by no the collection of pairs
(c, d) of words in Fo obtained by replacing any f E P by the word aU) in
all pairs of words of n.
Lemma XII.lo5. Under the previous hypothesis, the tensor category C is
also generated by the set Fo and the relations no
PROOF. Clearly, the set Fo generates the category, and if a, a' E M(Fo) are
congruent modulo no, then they are congruent modulo n, which implies
a = a'. Conversely, if a = a', then by definition a and a' are congruent
modulo n. Now we may replace any f E F' by aU) E M(Fo) in these
congruences, which yields congruences in no between a and a'. 0

XII.2 The Category of Tangles

We defined the concept of tangles and of isotopy classes of tangles in X.5.
The reason why we introduced these one-dimensional objects is that tangles
form a strict tensor category T as follows. By definition, the objects of T
consist of finite sequences of signs, including the empty sequence 0, and
the morphisms of T are the isotopy classes of oriented tangles. For any
oriented tangle L, the sequences s(L) and b(L) defined in X.5 will be the
source and the target of L respectively. The identity id : Ob(T) - t Hom(T)
is defined by the following rules: id 0 is the empty set 0; if C is a finite
sequence of length n in Ob(T), we define id" as the isotopy class of the
tangle L formed by the union of intervals {I, 2, ... , n} x {O} x [0, 1J. The
orientation of these intervals is determined by the rule s(id,,) = b(id,,) = c.
The composition of tangles introduced in X.5 defines the composition in T.
Recall that L' oL is obtained by placing L' on top of L. Lemmas X.5.10-5.11
imply that T is a category with identity maps id".
We equip T with a tensor product. It is defined on objects by concatena-
tion of sequences, i.e., if C = (Cl' ... ' ck) and c' = (Ck+l' ... ' c) are objects
of T, then their tensor product is given by

We also set 0 C = C = C 0. This operation is clearly associative. Let

us now define the tensor product on the morphisms of T. If Land L' are
isotopy classes of oriented tangles, L L' is the isotopy class of the oriented
tangle obtained by placing L' to the right of L as in Figure 2.1.
300 Chapter XII. The Tangle Category

Figure 2.1. The isotopy class L Q9 L'

This operation is well-defined up to isotopy, is associative on isotopy

classes of tangles and defines a functor from TxT to T. This is summarized
in the next proposition.

Proposition XII.2.1. The tangle category T equipped with the tensor

product defined above is a strict tensor category in which the unit I is
the empty set 0.

Observe that the endomorphisms of the unit object 0 are exactly the
tangles without boundaries, i.e., the links in the space R2xlO, 1[. This ob-
servation will be crucial in Sections 4-5.
We now state the main theorem of this section. It involves the "ele-
mentary" tangles defined by (X.5.1-5.5). We shall also use the following
conventions: 1= id(+) and 1= id(_), and XY is short for X Y when X
and Yare elements of the generating set below.

Theorem XII.2.2. The strict tensor category T is generated by the six

morphisms

and the relations

(1 n) 0 (u 1) =1= (n 1) 0 (1 u), (2.1)

(l n) 0 (U l) =1= (n l) 0 (l u), (2.2)

(n n) 0 (I n 1n) 0 (n X n) 0 (n 1 U l) 0 (n u)
= (n n) 0 (n 1 n l) 0 (I U 1n) 0 (U n), (2.3)
(n X n) 0

X+ 0 X_ = X_ 0 X+ =11, (2.4)
(X+ 1) 0 (1 X+) 0 (X+ 1) = (1 X+) 0 (X+ 1) 0 (1 X+), (2.5)

(1 n) 0 (X l) 0 (1 u) =1, (2.6)

(n 1l) 0 (I X'f l) 0 (l! u) 0 (l! n) 0 (I X l) 0 (U !l) =1 l, (2.7)

(l! n) 0 (I X l) 0 (U !l) 0 (n 1l) 0 (I X'f l) 0 (l! u) =l! . (2.8)

 XII.2 The Category of Tangles 301

The proof of Theorem 2.2 will be given at the end of Section 3. Figures
2.2-2.9 illustrate Relations (2.1-2.8).

Figure 2.2. Relation (2.1) Figure 2.3. Relation (2.2)

Figure 2.4. Relation (2.3)

Figure 2.5. Relation (2.4) Figure 2.6. Relation (2.5)

Figure 2.7. Relation (2. 6)

J '"

(
)
Figure 2.8. Relation (2.7)

}
! n
Figure 2.9. Relation (2.8)
302 Chapter XII. The Tangle Category

These relations are valid in the tangle category as a consequence of Lemma

X.5.7 and Theorem X.5.9. More precisely, Relations (2.1-2.3) follow purely
from isotopies of diagrams, Relation (2.4), Relation (2.5), and Relation (2.6)
follow from isotopies of diagrams and from Reidemeister Transformations
(II), (III), and (I), respectively, whereas Relations (2.7-2.8) follow from
isotopies of diagrams and from Reidemeister Transformation (II).

XII.3 The Category of Tangle Diagrams

In order to prove Theorem 2.2, we introduce a strict tensor category D of
tangle diagrams and give a presentation of it by generators and relations.
The category D is defined as the tangle category T of Section 2, but with
tangles in R2 x [0,1] replaced by tangle diagrams in R x [0,1]. More pre-
cisely, the objects of D are the same as the objects of T, namely finite
sequences of signs. The morphisms of D are isotopy classes of tangle
diagrams in R x [0,1] as defined in Chapter X. Identity, source, target,
composition, and tensor product are defined as for the tangle category. We
thus obtain a strict tensor category D. The tangle category T is, roughly
speaking, the quotient of D by the Reidemeister Transformations (I-III).
Let us introduce more "elementary" tangle diagrams as in Figure 3.1.

xx xx xx
They differ from the tangles X only by their orientations.

Figure 3.1. Six elementary tangle diagrams

Lemma XII.3.l. The following relations hold in the category D:

y = (1'1 n) 0 (i X I) 0 (u 11), (3.1)

T = (n 11) 0 (1' X I) 0 (il U), (3.2)
Z = (n Ii) 0 (i n 1Ii) 0 (Ii X Ii) 0 (Ii 1 U I) 0 (Ii U), (3.3)

Z = (Ii n) 0 (Ii 1 n I) 0 (Ii X Ii) 0 (i u Iii) (u Ii).

0 (3.4)
For a proof, see Figure X.3.10. The following statement gives a presen-
tation by generators and relations of the category of tangle diagrams.
Proposition XII.3.2. The strict tensor category D is generated by the
twelve morphisms U, U, n, n,
X+, X_, Y+, Y_, Z+, Z_, T+ andT_, and the
Relations (2.1), (2.2), (3.1), (3.2), (3.3) and (3.4).
 XII.3 The Category of Tangle Diagrams 303

PROOF. (a) Let F be the set of morphisms listed in Proposition 3.2. We

first have to prove that F generates V. Let II be a generic tangle diagram.
Draw a horizontal line through any vertex of II that is not a local maximum
or minimum. This divides R x [0, 1] into strips such that the restriction of II
to each of them involves only a crossing or a local maximum or minimum,
i.e., it is of the form id 0 f 0 id where f belongs to the set :F. In the
category V the diagram II is the composition of these restrictions. This
presentation of II is unique, i.e., to any generic diagram II we can assign
a unique word an in the alphabet F such that an = II. Since by X.5 any
tangle diagram is isotopic to a generic one, we conclude that F generates
the diagram category.
(b) According to Definition 1.3, we have to check that, given any pair
(a, a') of words in F, we have a ""n a' ? Ii = a' where R is the set of
relations in Proposition 3.2. By the results of Chapter X, we already know
that equivalent words (as defined in Definition 1.1) represent isotopic tangle
diagrams. Similarly, Relations (2.1-2.2) and (3.1-3.4) give rise to isotopic
diagrams (see Figure 2.2 and Figure X.3.1O).
Let now (a, a') be a pair of words in F such that Ii = a'. By Lemma
1.2 (c) we may suppose that a and a' are of the form

Geometrically, this means that Ii = II and a' = II' are generic tangle dia-
grams and that a = an and a' = an' where we use the notation introduced
in Part (a) of this proof. By assumption, II and II' are isotopic diagrams.
Thus, they can be obtained from each other by a finite sequence of oper-
ations taken from the Transformations (A), (B), (C), and (E) of Lemma
X.5.7. In order to show that the words an and an' are congruent modulo
R, it is therefore enough to check that the above-mentioned transforma-
tions do not change the congruence class of words. Let us verify this case
by case.
(A) If II and II' are generically isotopic, then an = all"
(B) If II and II' differ by a Transformation (B), then an "" an' in view
of Relation (1.8).
(C) If II differs from II' by a Reidemeister Transformation (0), then
an ""n an' thanks to (2.1-2.2).
(E) This case is taken care of by Relations (3.1-3.4). 0

Corollary XII.3.3. The strict tensor category V is generated by the six

morphisms U, tI, n, fl, X+' X_, and Relations (2.1), (2.2), (2.3).
PROOF. By Lemma 3.1, V is generated by the previous set of six morphisms.
We now apply Lemma 1.5 to Proposition 3.2: Relations (3.1-3.2) vanish
whereas Relations (3.3-3.4) give rise to Relation (2.3). 0
304 Chapter XII. The Tangle Category

Proof of Theorem 2.2. It will be similar to the proof of Proposition 3.2.

Since any tangle may be represented up to isotopy by a generic tangle
diagram, Corollary 3.3 implies that T is generated by the fiet Fo of the six
morphisms listed in Theorem 2.2.
Let a and a' be words in Fo such that a and a' are isotopic tanglefi. By
Theorem X.5.9 one can pass from a to a' by a finite number of operations
consisting of isotopies of diagrams and Reidemeister Transformations (I),
(II), and (III). Corollary 3.3 implies that isotopies of diagrams do not af-
fect the congruence class modulo (2.1-2.3) of a word. In order to complete
the proof of Theorem 2.2, it is therefore enough to check that Reidemeis-
ter Transformations (I), (II), and (III) also leave the congruence classes
unaltered.
Let us start with Transformation (II): It suffices to check that words of
type L 0 L~ are congruent to II with the right orientation where L is of
the form X, Y, Z, T. When L = X, this follows from Relation (2.4). When
L = Z, it follows from the operations performed in Figure 3.2: the first
and last ones are isotopies of diagrams, the second one is the Reidemeister
Transformation (II) represented by Relation (2.4).

~I

<\
Figure 3.2. Congruence when L = Z

When L = Y or T, it follows from Relations (2.7-2.8) as shown in Fig-

ure 3.3.

Figure 3.3. Congruence when L =Y or T

We now deal with Transformation (III): When all strands are oriented
downwards, it follows from Relation (2.5) and its inverse. In the remaining
cases, one proceeds by reducing to the previous case as for Transformation
(II) above. For details, see [Tur94]' 1.4.5.
 XII.4 Representations of the Category of Tangles 305

Let us take care of Transformation (I). If the diagrams representing this

transformation are oriented downwards, then the desired conclusion fol-
lows from Relation (2.6). We are therefore left with the same diagrams
oriented upwards and we have to check that the corresponding words are
congruent modulo the relations (2.1-2.8). This follows from the operations
performed in Figure 3.4: the first, third, fifth, and seventh ones are iso-
topies of diagrams, the second one is the Reidemeister Transformation (I)
already considered, the fourth one is a Reidemeister Transformation (III),
the sixth one is a Reidemeister Transformation (II) applied twice. D

Figure 3.4. Proof for Transformation (/) "upwards"

XII.4 Representations of the Category of Tangles

c
In XI. 5 we constructed a strict tensor category str from any tensor category
C. Applying this construction to the category Vectf(k) of finite-dimensional
vector spaces over a field k, we get a strict tensor category V.
We define a representation of the tangle category T to be a strict tensor
functor F from the tangle category T into the strict tensor category V. The
main interest of this concept comes from the fact that each representation
F of T produces an isotopy invariant for oriented links with values in
the field k. Indeed, let L be an oriented link in R2 x ]0, 1[ (this space is
diffeomorphic to R 3 ). As we observed in Section 2, we may consider L as
an endomorphism of the unit 0 of the tangle category. Therefore, the image
F (L) of L under the strict tensor functor F is a k-linear endomorphism
of the unit of the category V, which is the ground field k. In other words,
F(L) is the multiplication by a scalar. By definition of the tangle category,
this scalar depends only on the isotopy class of L.
This method of producing isotopy invariants of links is interesting in so
306 Chapter XII. The Tangle Category

far as we can construct representations of the tangle category in a system-

atic way. This will be achieved in this section using the presentation of T
given in Theorem 2.2.

Definition XII.4.1. Let V be a finite-dimensional vector space. An en-

hanced R-matrix on V is a pair (c, IL) where c is an automorphism of V Q9 V
satisfying the Yang-Baxter equation and IL is an automorphism of V such
that
C(IL Q9 IL) = (IL Q9 IL )c, (4.l.a)

tr 2 (c1(id v Q91L)) = id v , (4.l.b)

(Tc=f1)+(id v * Q91L)(clT)+(idv * Q91L- 1) = idv*v (4.l.c)

where T = TV,V'

Here we made use of the partial transpose and of the partial trace defined
in 11.3. We shall also use the evaluation maps ev v , ev v * and the coevalu-
ation maps bv , bv * of 11.3 where we identify any finite-dimensional vector
space with its bidual. We are now ready to state the main theorem of this
section.

Theorem XII.4.2. Given an enhanced R-matrix (c, IL) on a finite-dimen-

sional vector space V, there exists a unique strict tensor functor F from
the tangle category T to V such that F( (+)) = V, F( (-)) = V*, and

(4.2.a)

Then we necessarily have

(4.2.b)

There is a converse statement to Theorem 4.2 so as to have a bijective cor-

respondence between representations of the tangle category and enhanced
R-matrices. We shall not formulate it.
PROOF. Let F be a strict tensor functor from T to V. Set F((+)) = V,
F((-)) = W, and

F(U) = b : k -? V Q9 W, F(U) = b' : k - ? W Q9 V, (4.2.c)

F(n) = d : W Q9 V -? k, F(n) = d' : V Q9 W -? k, (4.2.d)

F(X+) = c = c+, F(X_) = c- : V Q9 V -? V Q9 V. (4.2.e)
By Theorem 2.2, the above six linear maps are related by the relations
obtained by applying F to (2.1-2.8), namely we have

(id v Q9 d)(b Q9 id v ) = id v = (d' Q9 idv)(id v Q9 b'), (4.3.a)

 XII.4 Representations of the Category of Tangles 307

(d Q9 idww)(id w Q9 d Q9 idvww)(idww Q9 c Q9 idww)

(idwwv Q9 b Q9 id w ) (idww Q9 b)
(idww Q9 d')(idwwv Q9 d' Q9 id w ) (idww Q9 c Q9 idww)
(id w Q9b' Q9id v ww)(b' Q9id w w), (4.3.c)

c+ c - = c - c+ = 1dVV' (4.3.d)
(c Q9 idv)(id v Q9 c)(c Q9 id v ) = (id v Q9 c)(c Q9 idv)(id v Q9 c), (4.3.e)
(id v Q9 d')(c Q9 idw)(id v Q9 b) = id v , (4.3.f)
gh = idvw, and hg = idwv (4.3.g)
where the linear maps 9 : W Q9 V ---+ V Q9 Wand h : V Q9 W ---+ W Q9 V are
defined by

(4.3.h)

and
(4.3.i)
The data (V, W, b, b', d, d', c, c-) where V, Ware finite-dimensional vector
spaces and b, b', d, d', c, c- are linear maps satisfying Relations (4.3.a-i) will
be called a representation data for the tangle category T.
Conversely, by Proposition 1.4 and Theorem 2.2, any representation data
(V, W, b, b', d, d', c, c-) for T gives rise to a unique tensor functor F : T ---+ V
such that F((+)) = V, F((-)) = W, and such that Relations (4.2.c-e)
hold.
These considerations imply that Theorem 4.2 is a consequence of the
following proposition. 0

Proposition XII.4.3. Let (c, JL) be an enhanced R-matrix on a finite-

dimensional vector space V. Define b, b', d, d', c, c- by

and c- = c- 1 . Then (V, V*, b, b', d, d', c, c-) is a representation data for T.
There is a converse statement whose formulation and proof are left to
the reader. Before we prove Proposition 4.3, we give a corollary to Theorem
4.2, and we state two lemmas which will be used in the proof of Proposition
4.3 (they may also be used to establish the converse statement).
Let (c, JL) be an enhanced R-matrix on a finite-dimensional vector space
V and F be the unique strict tensor functor from T to V satisfying Relations
(4.2.a-b). Let u be a braid with n strands. Since u is a tangle, we may
evaluate F On u. We get an automorphism F(u) of Vn. Similarly, F can
308 Chapter XII. The Tangle Category

be evaluated On the link a- which is the closure of the braid (J (see X.8).
We get an endomorphism F(a-) of the ground field k, i.e., a scalar. In the
following corollary, we express the automorphism F((J) and the scalar F(a-)
in terms of the representation p~ of the braid group Bn associated to the
R-matrix c in Corollary X.6.9.
Corollary XII.4.4. With the previous notation, we have
F((J) = p~((J) and F(a-) = tr(fLem 0 p~((J)) (4.4)
for any braid (J of Bn'
PROOF. (a) It suffices to prove the first statement for the generators (J l' ... ,
(In-l of Bn" Using the notation of X.6, we have
(Ji = 1i - 1 X+ 1n - i- 1

in the tangle category. Applying the tensor functor F, we get by X.6.2 and
by (4.2.a)
F((Ji) = id V0 (i-l) c id v0 (n-i-l) = p~((J).
(b) We first express the closure a- in the tangle category. We have

a- = n n 0 (rJ r ... noUn

where

and
Un = (l ... 1 U r ... no ... 0 ... (l uno u.
Therefore
F(a-) = F(n n) 0 (F((J) idV*0n) 0 F(UrJ.
Now, it is easy to check that (4.2.a-b) imply that

F(U n ) = 8v0 n and F(n n) = eVV*0n o(l2m idV*0n).

Consequently,

F(a-) = eVV*0n O((fL@n 0 F((J)) idV*0n )8v0 n ,

which is the trace of fL@n 0 F((J) by (11.3.12). o

Let V and W be finite-dimensional vectur spaces equipped with re-
spective bases {VI"'" v m } and {wI"'" w n } and respective dual bases
{vI, ... ,vm} and {wI, ... ,wn}. Suppose also we are given four linear maps
b : k --7 V W, b' : k --7 W V, d : W V --7 k, and d' : V W --7 k.
Define matrices B, B', D, D' by

b(l) = L Bij Vi w j , b'(l) = L B~j wi Vi'

i,j i,j
 XII.4 Representations of the Category of Tangles 309

Lemma XII.4.5. Under the previous hypotheses and with the previous
notation, we have

(id v d)(b id v ) = id v ~ BD = 1, (4.5.a)

(didw)(id w b) = id w ~ DB = 1, (4.5.b)
(d' idv)(id v b') = id v ~ D' B' = 1, (4.5.c)
(id w d')(b' id w ) = id w ~ B'D' =1 (4.5.d)
where 1 represents various matrix units.

PROOF. Simple calculation. D

Now define linear maps a : W* ----+ V and (3 : V* ----+ W by

In the following lemma we assume that a and (3 are isomorphisms, which

is equivalent to assuming Band B' to be invertible matrices. We also take
it that the inverse matrices of Band B' are D and D' respectively. Recall
that T stands for the flip.

Lemma XII.4.6. Let! be an endomorphism o!VV. Under the previous

hypotheses, we have

(d idw@w)(id w d idv@w@w)(idw@w ! idw@w)

(idw@w@v b idw)(idw@w b)
(4.6.a)

(idw@w d')(idw@w@v d' idw)(idw@w ! idw@w)

(id w b' idv@w@w)(b' idw@w)
= T((3 (3)1*((3-1 (3-1)T, (4.6.b)

(id v d')(f idw)(id v b) = tr2 (i (id v f-L)) (4.6.c)

where f-L = a((3-1)*,

(didv@w)(idw!idw)(idw@vb) (ida*) (TV,V 1) x Tv*,v(a- 1id)*,

=
(4.6.d)
(idw@vd')(idw!id w )(b' idv@w) = ((3id v )TV,v* (fTV,v) x (id v (3-1).
(4.6.e)
 310 Chapter XII. The Tangle Category

PROOF. Tedious, but easy computations. D

Proof of Proposition 4.3. Let (c, p,) be an enhanced R-matrix on the finite-
dimensional vector space V. In order to apply Lemmas 4.5-4.6, we set
W = V*, a = id v , and (3 = (p,*)-1. Pick a basis {v 1, ... , v m } of V along
with its dual basis {v 1 , ... ,vm}. Define a matrix B' by

(3(v j ) = L B~j vi.

Since p, and (3 are isomorphisms, the matrix B' is invertible. Let D' be its
inverse. By definition of b, b', d, d' we have

b(l) = LVi C!9 vi, b'(l) = L B~j vi C!9 Vj'

i,j

d(v i C!9 Vj) = {jij' d'(v i C!9 vj) = D~j.

Let us now prove Relations (4.3.a-g).
Relations (4.3.a-b) follow from Lemma 4.5 in view of the fact that the
matrices Band D are identity matrices.
Relation (4.3.c): Relation (4.1.a) implies that f(p, C!9 p,) = (p, C!9 p,)f for
f = c. Taking transposes, we get

which is equivalent to
T j*T = T(p,* C!9 p,*)-1 j* (p,* C!9 p,*)T.

The latter implies (4.3.c) in view of Relations (4.6.a-b), and of a = id v

and (3 = (p,*)-1.
Relation (4.3.d) holds by definition of c- whereas Relation (4.3.e) ex-
presses the fact that c is a solution of the Yang-Baxter equation. Relation
(4.3.f) follows from (4.1. b) in view of (4.6.c).
Relation (4.3.g): In view of (4.6.d-e) and since the expressions in brackets
below are isomorphisms, it is equivalent to show that

1
[((3C!9id v )Tv,v* (CTv,v) x (id v C!9(3-1 ) [(idC!9a*) (Tv,vc'f) x Tv*,v (a -1 C!9id V* ) * 1
is equal to idv*v. Replacing a and (3 by their values, we are reduced to
proving
((p,*)-1 C!9 id v )TV,v * (CTV,v) X (id v C!9 p,*)(TV,vC'f)XTv*,v = idv*v. (4.7)
Relation (4.7) is equivalent to
(id v C!9 (p,*)-1) (CTV,v )X (id v C!9 p,*)(TV,vC'f) X = idvv*. (4.8)
Taking transposes and using Lemma II.3.3, we see that (4.8) is equivalent
to (4.1.c). D
 XII.5 Existence Proof for Jones-Conway Polynomial 311

XII.5 Completion of the Proof of the Existence of

the Jones-Conway Polynomial
The aim of this section is to prove Proposition X.4.7 as a consequence of
Theorem 4.2. Let k be a field and q be an invertible element of k. Fix an
integer m > 1. Let Vm be a vector space over k of dimension m, equipped
with a basis {VI' ... , v m } Define a linear endomorphism cm of Vm (>9 Vm by
Am q Vi (>9 Vi if i = j,
cm(vi (>9 v j ) = { Am Vj (>9 Vi if i < j,
Am Vj (>9 Vi + Am(q - q-1) Vi (>9 Vj if i > j
where Am is a non-zero scalar. The map cm is a special case of the R-
matrices described in Example 3 of VIlLI. Proposition VIIL1.4 implies
that cm is a solution of the Yang-Baxter equation satisfying the additional
quadratic relation
= ( q - q-l)d
\-1
Am Cm -
\ -1
1 V= V= .
Am Cm (5.1)
Define an automorphism fLm of Vm by fLm(V i ) = A;;,lq-2i+l vi. Observe
that
1 qm _ q-m
tr(fLm) = -\--
Amqm q - q
-1 (5.2)

Lemma XII.5.l. If Am = q-m, the pair (c m , fLm) is an enhanced R-

matrix on Vm.
PROOF. We have to check Relations (4.1.a-c). The first one is automatically
verified because of the simple form of fL.
Relation (4.1. b): An immediate computation shows that

tr 2 (Cm(id (>9 fLm)) (vi) (q-2(i-l) + (1 - q-2) L q-2(j-l) )Vi

j<i
(q-2(i-l) + 1 - q-2(i-1) )Vi = vi.
Therefore, tr2(cm(id (>9 fLm)) = id v=. We have to check that the same re-
lation holds when we replace cm by its inverse. Taking advantage of (5.1),
we get

tr2 (C;;,1 (id (>9 fLm))

A;;,2 tr2 (crrJid (>9 fLm)) - A;;,I(q - q-l) tr 2 (id (>9 fLm)

A;;,2 (1 - Am(q - q-l) tr(fLm))id v=

A;;,2 (1 - q-m(qm - q-m) )id Vm

A- 2 -2m id
m q Vrn
= id V=
since Am = q-m and by (5.2).
312 Chapter XII. The Tangle Category

Relation (4.1.c) is proved by a direct computation left to the reader. 0

As a corollary of Theorem 4.2 and of Lemma 5.1, we get

Proposition XII.5.2. Let Am = q-m. There exists a unique strict tensor

functor F m,q from the tangle category T into the strict tensor category V
associated to Vectf(k) such that Fm,q((+)) = V m , Fm,q((-)) = V;" and

L Vi vi, L q2i-l-m vi Vi'

m m
F m ,q(U)(1) = F m ,/U)(l) =
i=l i=l

and the value of F m,q on the trivial knot 0 is given by

(5.4)

PROOF. We apply Theorem 4.2 to the pair (cm,/-lm) of Lemma 5.1. Rela-
tions (4.2.a) imply the desired forms for Fm,q(U) and for Fm,/U). Relation
(5.1) translates immediately to

For the trivial knot, we observe that it is the closure of the trivial braid in
B 1 . We may then appeal to Corollary 4.4, which yields Fm,q(O) = tr(/-lm)'
We conclude with (5.2). 0

We end this section by proving Proposition X.4.7, which completes the

proof of Theorem X.4.2 on the existence of the Jones-Conway polynomial.
Proof of Proposition X.4.7. It is an application of Proposition 5.2 where
k = C is the field of complex numbers and q =f. a is a complex number that
is not a root of unity. We fix an integer m > 1.
Let us denote by F the restriction of the tensor functor F m,q to oriented
links in R2 x ]0,1[. Since oriented links are endomorphisms of 0 in the
tangle category, F takes its values in the endomorphism ring End( C) which
is canonically isomorphic to the field C of complex numbers. Using this
isomorphism, we see that F(L) is a complex number for any oriented link
Lin R2 x ]0,1[. Moreover, by definition of the tangle category, F(L) depends
only on the isotopy class of L. By Proposition 5.2, the value of F on the
trivial knot is
m -m
F(O) = q - q_ =f. O.
q_ q 1
 XII.7 Notes 313

Suppose for a moment that we have proved

qm F(L+) _ q-m F(L_) = (q _ q-l )F(Lo) (5.5)
whenever (L+, L_, La) is a Conway triple. Then the composition cJI m ,q of
F with a diffeomorphism of R3 onto R2 x ]0, 1[ produces a complex-valued
map on the oriented links in R 3 , satisfying the conditions of Proposition
X.4.7. Therefore, the proof will be complete once we have checked Relation
(5.5). Now, by definition of a Conway triple (L+, L_, La), there exist tangles
Li (1 :s; i :s; 4) such that
L+ = Ll 0 (L2 X+ L 3 ) 0 L4 , L_ = Ll 0 (L2 X_ L 3 ) 0 L4 ,
and La = Ll 0 (L 2 11 L3) 0 L 4 . Since Fm,q is a tensor functor, we get
qmF(L+) - q-m F(L_) - (q - q-l)F(Lo)

= Fm ,q(L 1 ) (Fm,q(L 2) S Fm ,q(L3)) Fm,q(L4 )

where
S = qm Fm,q(X+) - q-m Fm,q(X_) - (q - q-l)Fm,q(11).
The latter vanishes by (5.3). This proves Relation (5.5). o

XII.6 Exercises
1. Consider the strict tensor category whose objects are the nonnegative
integers and whose morphisms are the isotopy classes of all braid
diagrams in R x [0, 1]. Show that it is generated by the morphisms
X+, X_ and the relations X+ 0 X_ = X_ 0 X+ = id.
2. Let c E Aut(Vl VI) be an R-matrix as in VIII.2, Example 2. Find
all automorphisms J.L of VI such that (c, J.L) is an enhanced R- matrix.
3. Compute the trace ofthe automorphism (J.Lm J.Lm)cm where (c m , J.L m )
is the enhanced R-matrix of Lemma 5.1. Deduce the value of the
functor F m,q of Proposition 5.2 on the trefoil knot and on the Hopf
link (Hint: use Corollary 4.4 and (5.1)).

XII.7 Notes
The results of this chapter are essentially due to Turaev [Tur89] whose
exposition we followed closely, and to Yetter [Yet88]. Enhanced R-matrices
already appear in [Tur88], though in a slightly different form.
In XIV.5.1 we shall build a strict tensor category Rout offramed tangles
or ribbons (defined in X.8). A presentation ofR by generators and relations
is given in [FY89] [Tur89].
Chapter XIII
Braidings

We define the important concept of a braided tensor category due to Joyal

and Street [JS93]. This concept has been introduced to formalize the char-
acteristic properties of the tensor categories of modules over braided bial-
gebras as well as the idea of crossing in link and tangle diagrams. After
defining braided tensor categories, we show that braids form a braided ten-
sor category that is universal in some precise sense. We also give the "centre
construction" which is the categorical version of Drinfeld's quantum dou-
ble.

XIII. 1 Braided Tensor Categories

XIII. 1. 1 Definitions and main properties
Let C be a category with a tensor product Q9 : C xC -> C and an associativity
constraint a. Denote by T : C x C -> C x C the flip functor defined by
T(V, W) = (W, V) on any pair of objects of the category. A commutativity
constraint c is a natural isomorphism

c: Q9 -> Q9T.

This means that, for any couple (V, W) of objects of the category, we have
an isomorphism
cv,w : V Q9 W -> W Q9 V (1.1)
 XIII.l Braided Tensor Categories 315

such that the square

cV,W
--4

(1.2)

commutes for all morphisms f, g.

The commutativity constraint c satisfies the Hexagon Axiom if the two
hexagonal diagrams
(HI)
CU,vw
U (VW) (VW)U
~

/ au,v,w ~ av,w,u

(UV)W V (WU) (1.3)

~ cu,v0id w / idv0cu,w
av,u,w
(VU)W -----> V(UW)
and (H2)
Cuv,W
(UV)W ~ W(UV)
/ au,'v,w ~ a;:;"~u,v
U(VW) (WU) V (1.4)
~ idu0cv,w / cu,w0 id v
au,'w,v
U (WV) -----> (UW) V

commute for all objects U, V, W of the category.

Observe that the hexagon (H2) can be obtained from (HI) by replacing
the isomorphism c by its inverse c- 1 . The following definition is due to
Joyal and Street. It is central in the theory of quantum groups.

Definition XIII. 1. 1. Let (C, , I, a, l, r) be a tensor category.

(a) A braiding in C is a commutativity constraint satisfying the Hexagon
Axiom, i. e., (1.3-1.4).
(b) A braided tensor category (C, ,I, a, l, r, c) is a tensor category with
a braiding.
Note that if c is a braiding in C, then so is the inverse c- 1 . When the ten-
sor category C is strict, the commutativity of (HI) and (H2) are equivalent
to the relations

(1.5)

and
(1.6)
Let us investigate the relationship of the braiding with the unit con-
straints.
316 Chapter XIII. Braidings

Proposition XIII.1.2. For any object V of a braided tensor category with

unit 1, we have

and

When the category is strict, these relations simply become

(1. 7)

PROOF. Consider the diagram

a c
(V0I)0W -----t V0 (10 W) -----t (10W) 0V
'" Tv0idw lidv0lw llw0 id v "'a
c
V0W -----t W0V 10 (W0 V)
l''d W
/ l v 0 id w 11vw 11wv / id
a id I 0c
(10V) 0W -----t 10 (V0 W) -------> 10 (W0 V)

The outside heptagon commutes by the commutativity of (1.3), the top

square by the naturality of the braiding, the bottom square by the nat-
urality of l, the upper left triangle by the Triangle Axiom (XI.2.9) and
the lower left and the right triangles by Lemma XI.2.2. Consequently, the
middle left triangle commutes, which means that

rv 0 id w = (lv 0 id w ) 0 (CV,I 0 id w ) = (lv 0 CV,I) 0 id w

Set W = 1. Applying the naturality of r, we get rv = lv 0 Cv I' which is

the first equality to be proved. Replacing c by its inverse, we s~e that the
commutativity of (1.4) implies the second relation in a similar way. The
last relation is an immediate consequence of the other two. 0

XIII. 1.2 Relation with the Yang-Baxter equation

One of the main properties of a braided tensor category is stated in the
following theorem which may be considered as the categorical version of
the Yang-Baxter equation.
 XIII.1 Braided Tensor Categories 317

Theorem XIII.1.3. Let U, V, W be objects in a braided tensor category.

Then the dodecagon

(U 129 V) 129 W
,,/ cu,vidw "" au, v, w
(V 129 U) 129 W UQ9 (V 129 W)
1 av,u,w liducv,w

V 129 (U 129 W) U 129 (W 129 V)

lidvcu,w 1 a;:;,'w, v
VQ9 (WQ9U) (U 129 W) 129 V
1 a;:,;,'w,u 1 cu,widv

(VQ9W) Q9U (W 129 U) 129 V

1 cv,widu 1 aw,u,V

(W 129 V) 129 U WQ9 (UQ9V)

~ aw,V,u ,,/ idwcu,v
WQ9 (VQ9U)
commutes.
If the category C is strict, then the commutativity of the dodecagon is
equivalent to the relation

(cv,w 129 id u )(id v 129 Cu,w) (C UY 129 id w )

= (id w 129 Cuy )(cu,w 129 idv)(id u 129 cv,w) (1.8)
This implies that the natural isomorphism Cv v is a solution of the Yang-
Baxter equation for any object V of a braided tensor category.
PROOF, We cut the dodecagon into two hexagons of type (H2) and a square,
According to (1.4), the clockwise composition of the morphisms in the
dodecagon starting from (U 129 V) 129 Wand ending at W 129 (U 129 V) is equal
to cuv,w' Similarly, the counterclockwise composition of the morphisms
from (V 129 U) 129 W to W 129 (V 129 U) is equal to cvu,w' It remains to check
the commutativity of the square
(U 129 V) 129 W CU0V ,w, WQ9 (UQ9 V)
1 cu,vidw 1 idwcu,v
(V 129 U) 129 W CV0U ,w, WQ9 (VQ9U) ,

But this is a special case of the commutative square (1.2) (expressing the
functoriality of the braiding) where f is replaced by Cuy and g by id w
o
We give a few examples of braidings.
318 Chapter XIII. Braidings

XIIL1.3 Braided categories of vector spaces

Example 1. (The flip) The flip T is clearly a braiding in the tensor category
Vect(k). It is also a braiding in the category k[G]-Mod of representations
of a group G and, more generally, in the tensor category A-Mod of modules
over a cocommutative bialgebra A (see Proposition IlL5.1).
The following result relates the notions of a braided tensor category and
of a braided bialgebra as defined in VIIl.2 and justifies the name given to
the latter.

Proposition XIII.1.4. Let (H, IL, "1, Ll, c) be a bialgebra. The tensor cat-
egory H -Mod is braided if and only if the bialgebra H is braided.

PROOF. Let (H, IL, "1, Ll, c, R) be a braided bialgebra with universal R-
matrix R. In VIlL3 we defined isomorphisms c{J w from V W to W V
by ,
c~,w(v w) = TV,W (R(v w))
where v E V and w E W. Proposition VIlL3.1 implies that the family c is
a braiding.
Conversely, let (H, IL, "1, Ll, c) be a bialgebra. Suppose that there exists a
braiding c in the tensor category H-Mod. Define an invertible element R
in HH by
(1.9)
Let us show that R is a universal R-matrix for H.
If v, ware elements of H-modules V, W, the naturality of the braiding
implies the commutativity of the square

HH ~ HH
lV0
w 1 w0v
VW ~ WV
where ii : H ----; V and w : H ----; Ware the H-linear maps defined by
ii(1) = v and w(1) = w. This implies that

cv,w(vw) = (wii) (C H,H(11)) = TV,W ((iiW)(R)) = TV,W (R(vw)).

(1.10)
We express the H-linearity of cH,H: for each a E H we have

By (1.10) we get Ll(a)TH,H(R) = TH,H (RLl(a)). This is equivalent to

LlOP(a)R = RLl(a)

for all a E H.
 XIII.1 Braided Tensor Categories 319

The commutativity of the hexagons (1.3-1.4) implies the relations

(id 6.)(R) = R13R12 and (6. id)(R) = R 13 R 23
respectively. This proves that R satisfies Relations (VIII.2.1) and (VIII.2.3-
2.4) defining a braided bialgebra structure on H. 0
Under the correspondence set up in this proof, the commutativity of the
dodecagon in Theorem 1.3 is equivalent to the equation

R12R13R23 = R23 R 13 R 12
of Theorem VIII. 2.4 (a) .

XIII. 1.4 Crossed G-sets

Given a group G we may form a strict braided tensor category as follows.
Define a (right) crossed G-set as a set X with a right action X x G ---> X
of the group G and a set-theoretic map I I : X ---> G such that
Ixgl = g-llxlg
for all x E X and g E G. A morphism f : X ---> Y of crossed G-sets is a
map f from X to Y such that f(xg) = f(x)g and If(x)1 = Ixl for all x EX
and g E G. Crossed G-sets and their morphisms form a category X (G).
We equip this category with a tensor product as follows. Given crossed
G-sets X and Y, we define X Y as the set-theoretic product X x Y with
G-action given by (x, y)g = (xg, yg) and with map X Y ---> G given by
I(x, y)1 = Ixllyl It is easy to check that X Y belongs to X(G). Similarly,
given morphisms f and g, we define f g = f x g. Then X(G) becomes
a strict tensor category with unit I equal to the crossed G-set {1} with
111 = 1.
For any pair (X, Y) of crossed G-sets, define CX,y : X Y ---> Y X by
CX,y(x, y) = (y, xlyl) (1.11)
where x E X and y E Y. The proof of the following result is left to the
reader.
Proposition XIII.lo5. The maps Cx yare morphisms of crossed G-sets
and form a braiding for the strict tens~r category X ( G).
Let X = G with the group acting on itself by conjugation, so that
(x,g) f--+ g-lxg. Then X = G is a crossed G-set with II = ide. Conse-
quently, Gem = G n is a crossed G-set with I(gl' ... , gn) I = gl ... gn (for
n> 0). The full subcategory of X(G) with objects {1,G,G'81 2 ,G03 ,oo.}
forms a braided subcategory Xe(G) of X(G) with braiding given by
cen,G'n (gl' ... ,gn+m) = (gn+l' ... ,gn+m' g-lglg, ... ,g-lgng) (1.12)

where g = gn+l" gn+m

320 Chapter XIII. Braidings

XIII. 1. 5 Symmetric tensor categories

Braided tensor categories generalize the classical concept of a symmetric
tensor category introduced earlier by category theorists. A tensor category
is symmetric if it is equipped with a braiding c such that

(1.13)

for all objects V, W in the category. If (1.13) holds, we call the braiding c a
symmetry for the category. Notice that the commutativity of the hexagon
(HI) and the commutativity of the hexagon (H2) are equivalent in a sym-
metric tensor category.
We give two examples of symmetric tensor categories.

Proposition XIII.1.6. The strict tensor categories GL and S of XI.3.2

are symmetric.

PROOF. We define automorphisms sn,Tn E GLn+m(k) : n 0 m --* m 0 n as

follows. If { e 1 , ... , e n +m } is the canonical basis of the vector space k n + m , we
set Sn,Tn(e i ) = e Tn +i if 1:::; i:::; n, and sn,m(e i ) = ei - n ifn+ 1:::; i:::; n+m.
The matrix of sn,m in the canonical basis of k n +m is the (n + m) x (n + m)
matrix
( 0
In
1m)0
where In is the unit n x n matrix. This holds when n > 0 and m > O.
Otherwise, we have So n = id kn = sn o
We claim that the f~mily (sn m) i~ a braiding for GL. We have to check
the functoriality and Relations (1.5-1.6). The functoriality is equivalent to
the relation
Sn,m 0 (g 0 h) = (h 0 g) 0 sn,Tn
for all g E G Ln (k) and all h E G Lm (k ). This follows from the matrix
relation

Relation (1.5) is a consequence of the relation

Observe that sm,nsn,m = id n0m . Therefore Relation (1.6) is also verified,

which proves that (sn,m) is a braiding endowing GL with the structure of
a symmetric tensor category.
Since the matrix of sn m is a permutation matrix, the same formulas
define a symmetry on the' category S. 0
 XIII.2 The Braid Category 321

XIII.2 The Braid Category

In X.6 we defined braids as a special case of tangles. When composing
or tensoring two braids as we did for tangles in XII.2, we get another
braid. This proves that braids form a strict tensor category B in which the
objects are finite sequences of + signs. We identify such a finite sequence of
length n with the integer n under the convention that the empty sequence
corresponds to the integer O. Henceforth, we shall consider the set N as the
set of objects of the strict tensor category B. The purpose of this section is
to show that the braid category is a strict braided tensor category.
In order to put a braiding on the braid category, we have to define
isomorphisms cn m : n m -+ m n for any couple (n, m) of non-negative
integers. This is done as follows: Co ,n = idn = cn ,0' and for n, m > 0 we set

where 0'1' ... ,0'm+n-l are the generators of Bm+n defined in X.6. The braid
cn m is represented in Figure 2.1. Observe that the permutation of the braid
cn :m is the permutation sn,m of Proposition 1.6.

n m

Figure 2.1. The braid Cn,m

Theorem XIII.2.1. The family of isomorphisms (c n m)n m>O is a braid-

ing in the braid category B. ' ,-

PROOF. We have to prove that the family (c n m)n m>O is functorial with
respect to all morphisms in B and satisfies Rel~tio~s [1.5) and (1.6).
Let us start with the functoriality. Since any morphism in B is an element
of a braid group, it is enough to check the functoriality with respect to the
generators O'i' More precisely, we must prove that for all i, j such that
1 :::; i :::; n - 1 and 1 :::; j :::; m - 1 we have

Both sides of this relation are represented by the braid diagrams of Figure
2.2. It is clear that one can pass from one braid diagram to the other
by repeated applications of the Reidemeister Transformation (III), which
322 Chapter XIII. Braidings

proves the equality. The reader may replace this topological proof by an
algebraic one using the braid relations of Lemma X.6.4.

Figure 2.2.

As for Relation (1.5) [resp. Relation (1.6)], a graphical proof is given in

Figure 2.3 [resp. in Figure 2.4].

v>
0

,~
n m p n m p

Figure 2.3. Proof of Relation 1.5

n m p n m p

Figure 2.4. Proof of Relation 1.6

Observe that we have the relation

'd 0 (i-l) 'd 0 (n-i-l)
(2.2)
ai =
rO, rO,
1 1 '<Y C1 , 1 '<Y 1 1

in the braid category B.

XIII.3 Universality of the Braid Category

In this section we derive two universality properties for the braid category.
They imply that B is a model for all other braided tensor categories.
 XIII.3 Universality of the Braid Category 323

XIII.3.1 Yang-Baxter operators

We introduce Yang-Baxter operators following Joyal-Street [JS93J.
Definition XIII.3.1. If V is an object of a tensor category (C, 0, I, a, l, r),
an automorphism a of V 0 V is called a Yang-Baxter operator on V if the
dodecagon

(V0 V) 0 V
/0"0 id v "" av,v,v
(V 0 V) 0 V V0 (V0 V)
1 av,v,v l idv0 0"
V0(V0V) V 0 (V0V)
l idv0 0" 1 av,'v,v
V0 (V0 V) (V0V)0V
1 av,'v,v 10"0 id v
(V 0 V) 0 V (V0V) 0V
10"0 idv 1 av,v,v

(V0V) 0 V V0 (V0 V)
~ av,V,V / idv00"

V0(V0V)

commutes.
The commutativity of this dodecagon is equivalent to

(id 0 a)a(a 0 id)a-1(id 0 a)a = a(a 0 id)a-1(id 0 a)a(a 0 id) (3.1)

where a = av v v and id = id v In any braided tensor category, the braiding

cv,v is a Yan'g~Baxter operator. This follows from Theorem 1.3. Here is a
way to generate Yang-Baxter operators.

Lemma XIII.3.2. Let (F, 'Po, 'P2) : C -4 V be a tensor functor between

tensor categories. If a E Aut(V 0 V) is a Yang-Baxter operator on the
object V in C, then

a' = 'P2(V' V)-l 0 F(a) 0 'P2(V, V) (3.2)

is a Yang-Baxter operator on F(V).

PROOF. Clearly a' is an automorphism of F(V) 0 F(V). It is enough to

check Relation (3.1). In other words, if we set

u = (id 0 a')a(a' 0 id)a-1(id 0 a')a

324 Chapter XIII. Braidings

and
v = a(c/ 0 id)a- 1(id 0 (J1)a((J1 0 id),
we have to prove that u = v. Now, by Definition X1.4.1 we have

(3.3)

Relations (3.2-3.3) imply that

u = (id 0 IP21)(id 0 F((J))IP2 1F(a)IP2(F((J) 0 id)

IP21 F(a- 1)IP2(id 0 F((J))IP2 1F(a)IP2(IP2 0 id)

where IP2 = IP2(V, V). Now IP2 is a natural isomorphism. Therefore the
squares
'P2(VV,v)
F(V 0 V) 0 F(V) F((V0V)0V)
1
)

F(u)idp(v) IF(Uid v )
'P2(VV.V)
F(V 0 V) 0 F(V) ) F((V0V)0V)

and
'P2(V,vV)
F(V) 0 F(V 0 V) F(V0(V0V))
1 1
)

idp(v)F(O") F(id v lT)

'P2(V,vV)
F(V) 0 F(V 0 V) ----+ F(V0(V0V))

commute. We rewrite them in the form

Plugging (3.4) into u, we get

U = (id0IP21)IP21F(id0(J)F(a)F((J0id)
F(a- 1)F(id 0 (J)F(a)IP2(IP2 0 id)
(id 0 IP21 )IP2 1F ((id 0 (J )a( (J 0 id)a- 1(id 0 (J)a) IP2( IP2 0 id).

Similarly, we have

The equality u = v results from the fact that (J satisfies Relation (3.1). 0

We now define a new category YB(C) out of the Yang-Baxter operators

on the tensor category (C, 0, I, a, l, r). The objects of YB(C) are pairs (V, (J)
where V is an object of C and (J : V 0 V --7 V 0 V is a Yang-Baxter operator
on V. A morphism f : (V, (J) --7 (V', (JI) in YB(C) is a morphism f : V ---> Vi
in C such that the square
 XIII.3 Universality of the Braid Category 325

VV
1
1@1
V'V'
(3.5)

commutes. The identity of (V, a) in YB(C) is id v .

Let us relate the category of Yang-Baxter operators to the braid category
B. Suppose that (F, 'Po, 'P2) : B ----> C is a tensor functor from B to the tensor
category C. By Theorems 1.3 and 2.1 we know that the automorphism
cLl = a1 of 1 1 = 2 is a Yang-Baxter operator on the object 1 in B. It
follows from Lemma 3.2 that the automorphism

(3.6)

is a Yang-Baxter operator on F(l) in the category C. This defines an object

(F(l), a) in YB(C), which we denote by 8(F).
We claim that 8 extends to a functor Tens(B, C) ----> YB(C). Let us check
that, if ry : (F, 'Po' 'P2) ----> (F', 'P~, 'P;) is a natural tensor transformation,
then ry(l) : F(l) ----> F'(l) is a morphism in the category YB(C). In other
words, we have to show that ry(l) satisfies the following relation

(3.7)

(ry(l) ry(1))'P21F(c1,l)'P2
'P; -lry(2)F(C 1,1)'P2
'P; -1 F'(C 1,1)ry(2)'P2
'P; -IF'(C 1,1)'P;(ry(1) ry(l))
a'(ry(l) ry(l)).

The first and last equalities follow by definition of a and a', the second
and fourth ones by definition of a natural tensor transformation (Definition
XI.4.1), and the third one by Definition XI.1.3.
We can state the first universality property of B.

Theorem XIII.3.3. For any tensor category C the functor, defined above,
8: Tens(B, C) ----> YB(C) is an equivalence of categories.

PROOF. By Proposition XI.1.5 it suffices to check that the functor 8 is

fully faithful and essentially surjective.
In order to establish that 8 is fully faithful, we have to show that 8
induces a bijection on morphisms. We build a map inverse to the map
ry f---+ ry(l) considered above.
Let f : (F(l), 'P21 F(C 1,1)'P2) ----> (F'(l), 'P; -1 F'(C 1,1)'P;) be a morphism
in YB(C) where (F,'PO,'P2) and (F','P~,'P;) are tensor functors from B to
326 Chapter XIII. Braidings

C. We wish to construct a natural tensor transformation 7]j from (F, i.po, i.p2)
to (F', i.p~, i.p~) such that 7]j(1) = f. We proceed as follows. If n = 0,1, we
set 7]j(O) = i.p~i.po1 and 7]j(l) = f. If n > 1 we define 7]j(n) inductively by

(3.8)

Lemma XIII.3.4. The family (7]j(n))n?O is a natural tensor transforma-

tion.

PROOF. We have to check that

(3.9)

for any integer n ~ 0 and any element g of the braid group En' that
7] j (0 )i.po = i.p~ (this holds by definition of 7] j (0)), and that for all n, m ~ 0

7]j(n m)i.p2 = i.p~ (7]j(n) 7]j(m)). (3.10)

It is enough to check Relation (3.9) when g is a generator (Ji of En" A

computation left to the reader shows that Relation (3.9) for g = (Ji is a
consequence of Relations (2.2) and (3.8), of the Pentagon Axiom (XI.2.6),
and of the definition of tensor functors.
As for Relation (3.10), one proceeds by induction on m as in the proof
of Lemma XI.5.2. 0

The full faithfulness of 8 follows from 7]j(1) = f and from 7]T)(1) = 7].
The first relation holds by definition. Let us check the second one. We shall
prove
(3.11)
by induction on n. This is clear for n = 0,1. If n > 1, we use the fact that
7]and 7]T)(1) are natural tensor transformations to write

7]T)(l)((n - 1) 1)

i.p~(7]T)(1)(n-1) 7]1)(1)(1))i.p2
i.p; (7](n - 1) 7](1)) i.p2
7]((n - 1) 1)
7]( n).

This proves (3.11), hence the full faithfulness of 8.

In order to complete the proof of Theorem 3.3, it remains to check that 8
is essentially surjective. By Theorem X1.5.3 we may assume that the tensor
category C is strict. Then the essential surjectivity of 8 is a consequence
of the following lemma. 0
 XIII.3 Universality of the Braid Category 327

Lemma XIII.3.5. Let C be a strict tensor category and (V, a) an object

of YB(C). Then there exists a unique strict tensor functor F : l3 --t C such
that F(1) = V and F(CI,I) = a.

PROOF. If such a functor F exists, then (2.2) implies F(n) = V<8I n and

F( ai ) -- F('d<8l(i-l)
1 I
,0,
'6' CI,1
,0,
'6'
'd<8l(n-i-I)) _ 'd<8l(i-I),o,
1 I-I V '6' a
,0,
'6'
'd<8l(n-i-l)
1 V

for 1 :::; i :::; n - 1. This proves the uniqueness of F in view of the fact that
aI' a z ,"" a n - l generate Bn as a group.
Let us prove the existence of F. Set F(n) = V<8I n . Define automorphisms
cl ,, cn - 1 of F(n) by

ci -_ 1'd<8l(i-l),o,
'6' a
,0,
'6'
'd<8l(n-i-l)
1

when 1 :::; i :::; n-1. The automorphism a being a Yang-Baxter operator, the
automorphisms ci satisfy the braid group relations (X.6.1-6.2). It follows
from Theorem X.6.5 that there exists a unique morphism of groups F from
the braid group Bn to Aut(F(n)) such that F(a i ) = ci for all i. The functor
F is a strict tensor functor from l3 into C and we have F(CI,I) = c i = a.
o

XIII.3.2 Braided tensor functors

In order to state the second universality property of the braid category, we
have to introduce the concept of a braided tensor functor.

Definition XIII.3.6. A tensor functor (F, 'Po, 'P2) from a braided tensor
category C to a braided tensor category V is braided if, for any pair (V, V')
of objects of C, the square

F(V) F(V') ~ F(V V')

1
Q9 Q9
CF(V),F(V') 1 F(cv,v') (3.12)
F(V') Q9 F(V) ~ F(V' Q9 V)
commutes.

We denote by Br(C, V) the category whose objects are the braided ten-
sor functors from C to V and whose morphisms are the natural tensor
transformations.

Theorem XIII.3.7. For any braided tensor category C the functor 8' :
Br(l3, C) --t C defined by 8' (F) = F(l) is an equivalence of categories.

PROOF. By Proposition XI.1.5 again it is enough to prove that 8' is fully

faithful and essentially surjective.
328 Chapter XIII. Braidings

Full faithfulness of 8'. First, we claim that if C is a braided tensor category

and F, F' are braided tensor functors, then

Hom YB(C)(8(F), 8(F')) = Homd8'(F), 8'(F')).

It is clear that the left-hand side sits inside the right-hand side. We have
to prove the opposite inclusion. Let f : F(l) = V --> F'(l) = V' be an
element of Homd8'(F), 8'(F')). We wish to prove that f is a morphism
in the category YB(C), which means that the square (3.5) has to commute
with (f = 'P;-1 F(Cl,l)'P2 and (f' = 'P; -1 F'(C 1,1)'P;. We have

(J 0 J)(f (J 0 J)'P;-1 F(C 1,1)'P2

(J 0 J)cv,v
cv',v,(J 0 J)
'P; -1 F'(C 1,l)'P;(J 0 J)
(f'(J 0 J).

The second and fourth equalities follow from (3.12) whereas the third one
follows from the naturality of the braiding c in C.
Now 8' is fully faithful in view of the isomorphisms

HomBr(B,C)(F, F') HomTcns(B,C)(F, F')

"" Hom YB (c)(8(F),8(F'))
Homc(8'(F), 8' (F'))

where the first one follows by definition, the second one from the full faith-
fulness of 8 (Theorem 3.3), and the last one has just been proved.
Essential surjectivity of 8'. Let V be an object of C. Since C is braided, the
automorphism Cv v is a Yang-Baxter operator by Theorem 1.3. According
to Theorem 3.3, the functor 8 is essentially surjective, which means that
there exists a tensor functor (F, 'Po, 'P2) : B --> C along with an isomorphism
ex: V --> F(l) such that

(3.13)

In order to prove that 8' is essentially surjective, it is enough to prove

that the functor F is braided, i.e., that for any pair (n, m) of nonnegative
integers the square (en m)

F(n) 0F(m) ~ F(n + m)

1
CF(n).F(m) 1
F(c n .m ) (3.14)
F(n) 0 F(m) ~ F(n + m)
commutes. We shall check this by induction on nand m.
 XIII.3 Universality of the Braid Category 329

The commutativity of (Co,o), (C1,0) and (CO,I) is left to the reader. Let
us first check that (C 1,1) commutes. We have

F(C 1,1)'P2 = 'P2(a 12> a)cv,v(a- 1 12> a-I) = 'P2 Cp(I),P(I)

by Relation (3.13) and by naturality of the braiding c.

We now prove that the commutativity of (Cn 1) and of (Cn m) implies
the commutativity of (Cn ,m+l). We h a v e ' ,

F(id 12> cn ,I)F(cn ,m 12> id)'P2

'P2 (id 12> F( Cn,I)) 'P2 1'P2 (F( cn,m) 12> id) 'P2 1'P2
'P2(id 12> 'P2)(id 12> c P(n),P(I)(id 12> 'P2 1)'P2 1
'P2 ('P2 12> id) (cP(n),P(m) 12> id)( 'P2 1 12> id)'P2 1 'P2
'P2(id 12> 'P2)(id 12> c P(n),P(I) a (cP(n),P(m) 12> id)
('P2 1 12> id)'P2 1 'P2
'P2(id 12> 'P2)CP(n),P(m)P(I) a ('P2 1 12> id)'P2 1 'P2
'P2 CP(n),P(m+l)(id 12> 'P2) a ('P2 1 12> id)'P2 1 'P2
'P2 cP(n),P(m+l) .

The first equality is a consequence of the fact that (cn,m)n,m is a braiding

for B, the second one follows by the naturality of 'P2' the third one by the
commutativity of (Cn 1) and (Cn m), the fourth one by Relation (XI.4.1),
the fifth one by Relation (1.3), the sixth one by naturality of the braiding
in C, the seventh one again by Relation (XI.4.1).
A similar computation shows that the commutativity of (C 1 n) and of
(Cn,m) implies the commutativity of (Cn+1,m). This is enough to prove
that all squares (Cn,m) commute. Therefore the functor F is braided and
the proof of Theorem 3.7 is complete. 0

We may interpret Theorem 3.7 as saying that, given any object V in

a braided tensor category C with braiding c, the tensor power vn of V
(whatever parenthesizing is used) is naturally a module over the braid group
Bn. If, moreover, cv,v is an involution, i.e., of square one, then the action
of Bn factorizes through the symmetric group Sn.
The proof of the essential surjectivity of 8' shows that the following more
precise result holds for any strict braided tensor category.

Corollary XIII.3.8. Let V be an object of a strict braided tensor category

C. Then there exists a unique strict braided tensor functor Fv from the braid
category B to C such that Fv(1) = V.
330 Chapter XIII. Braidings

XIII.4 The Centre Construction

We now give a construction which assigns to any strict tensor category
(C, , J) a braided tensor category Z(C), called the centre of C. When C
is the tensor category A-Mod of modules over a finite-dimensional Hopf
algebra A with an invertible antipode, then Z(A-Mod) is tensor equivalent
to the braided tensor category D(A)-Mod of modules over the quantum
double D(A) of A, as described in Chapter IX. In other words, this "centre
construction" is the categorical version of the quantum double construction.

Definition XIII.4.1. An object of Z(C) is a pair (V, c_ v) where V is an

object of C and c_,v is a family of natural isomorphisms'

cx,v : X V ---- V X

defined for all objects X in C such that for all objects X, Y in C we have

(4.1)

A morphism from (V, c_,v) to (W, c_,w) is a morphism f : V ---- W in

C such that for each object X of C we have

(4.2)

The naturality in Definition 4.1 means that the square

XV ~
lfl8i i d v (4.3)
YV ~

commutes for any morphism f : X ---- Y in C.

It is clear that the identity id v is a morphism in Z(C) and that if f, 9
are composable morphisms in Z(C) then the composition go f in C is a
morphism in Z (C). Consequently, Z (C) is a category in which the identity
of (V, c_,v) is id v .
We now state the main theorem of this section.
Theorem XIII.4.2. Let (C, , I) be a strict tensor category. Then Z(C)
is a strict braided tensor category where
(i) the unit is (1, id),
(ii) the tensor product of (V, c_,v) and (W, c_,w) is given by

(V, c_,v) (W, c_,w) = (V W, c_,vl8iw)

where cX ,VI8iW : X V W ---- V W X is the morphism of C defined

for all objects X in C by

(4.4)
 XIII.4 The Centre Construction 331

(iii) and the braiding is given by

Cv,W : (V, C,v) 129 (W, c_,W) ----7 (W, C,W) 129 (V, C_,v).

PROOF. 1. Let (V, c_ v) and (W, C w) be objects in Z(C). We claim that

the pair (V 129 W, c_,v'w) defined iY{ Theorem 4.2 (ii) is an object of Z(C).
Indeed, it follows from the properties of (V, c_,v) and (W, c_,w) that
cx,vw is an isomorphism in C and that cx,vw is natural in X. We have
to check Relation (4.1) for c_,vw. For all objects X, Y of C we have

CXY,VW (id v 129 CXy,w) (CXy,v 129 id w )

(id v 129 cx,w 129 idy)(idvx 129 CY,w)
(cx,v 129 idyw)(id x 129 Cy,v 129 id w )
(id v 129 cx,w 129 idy)(cx,v 129 idwy)
(idxv 129 cy,w)(id x 129 Cy,v 129 id w )
(cx,vw 129 idy)(id x 129 cy,vw).

The first and fourth equalities follow from (4.4), the second one from (4.1),
and the third one by (XI.2.3), i.e., by the naturality of the tensor product.
2. Let f: (V,c_,v) ----7 (W,c_,w) and l' : (V',C_,vI) ----7 (W',c,w') be
morphisms in Z(C). We claim that so is f 129 1'. Let us check Relation (4.2)
for f 129 1'. We have

(f 129 idw' 129 id x )(id v 129 l' 129 id x )

(id v 129 CX,v' ) (cx,v 129 id v ')
(f 129 idw' 129 id x )(id v 129 CX,w1)
(id v 129 id x 129 1') (cx,v 129 idvl)
(id w 129 cx,w1)(f 129 id x 129 idwl)
(cx,v 129 idw' )(id x 129 id v 129 1')
(id w 129 cx,w,)(cx,v 129 idwl)
(id x 129 f 129 id w ' )(id x 129 id v 129 1')
cx,ww1(id x 129 f 129 1').

The first and fifth equalities follow from (4.4) and from (XI.2.3), the second
and fourth ones from (4.2), and the third one from (XI.2.1).
Now it is clear that the tensor product is well-defined on the objects and
on the morphisms of Z(C). It is functorial and satisfies all the required
axioms because it already does so in the original category C. Thus, the
category Z(C) is a strict tensor category. We next show that it is braided.
3. Let us start by proving that cv,w is a morphism in Z(C). We have to
check Relation (4.2) for cv,w' namely
332 Chapter XIII. Braidings

for all objects X in C. We have

(cv,w id x )(id v cx,w)(cx,v id w )

cvx,w(cx,v id w )
(id w cx,v )cxv,w
(id w cx,v )(cx,w idv)(id x cvw)
cx,wv(idx cv,w)
The first and last equalities result from (4.4), the second and fourth one
from (4.1), and the third one from the naturality of c_,v.
4. The morphism Cv w is invertible by definition and it is natural with
respect to all morphisr~s of C, hence to those belonging to Z (C). In order
for cv .w to qualify as a braiding, it has to satisfy both Relations (1.5) and
(1.6). Now (1.6) follows from the hypothesis (4.1) and (1.5) from (4.4).
Therefore the tensor category Z (C) is braided with braiding cv. w. D

We give a universal property of the construction Z. For any strict tensor

category C the functor IT : Z (C) --+ C given by
(4.5)
is a strict tensor functor. It is universal in the following sense.
Proposition XIII.4.3. Let F be a strict tensor functor from a strict
braided tensor category C to a strict tensor category C'. Suppose that F
is bijective on objects and surjective on morphisms. Then there exists a
unique strict braided functor Z(F) : C --+ Z(C') such that F = IT 0 Z(F).

PROOF. Let us first prove the existence of Z(F). For any object V of C we
set
Z(F)(V) = (F(V), C-,F(V))
where C,F(V) is defined for all objects X in C' by cX,F(V) = F(C F -l(X),v)'
Here C v is the braiding in C. Relation (4.1) is satisfied because F is a
tensor f~nctor. Therefore Z(F)(V) is an object in Z(C').
If f : V --+ V' is a morphism in C, set Z(F)(f) = F(f). Relation (4.2)
is satisfied because of the naturality of the braiding in C. This proves that
Z(F) is a functor. Clearly, IT 0 Z(F) = F. Let us now check that Z(F) is
a braided tensor functor. It preserves tensor products because of (1.5) and
(4.4). It also respects braidings. Indeed, we have

Z(F)(cv,w) = F(cv,w) = cF(V),F(W)

which is the braiding of Z(C').
The uniqueness of Z(F) is a consequence of the fact that it preserves
braidings. D

Applying Proposition 4.3 to the identity functor of a braided tensor cat-

egory, we get the following result.
 XIII.5 A Categorical Interpretation of the Quantum Double 333

Corollary XIII.4.4. For any strict braided tensor category C there exists
a unique braided tensor functor Z from C to Z (C) such that II 0 Z = ide.

XIII.5 A Categorical Interpretation of the

Quantum Double
We relate the centre construction of Section 4 to the quantum double con-
structed in Chapter IX. This will give us a simple categorical description
of the quantum double.
Let A = (A, /-l, T],~, c, S) be a finite-dimensional Hopf algebra with in-
vertible antipode S. Under this hypothesis, we constructed a braided Hopf
algebra D(A) in IX.4. Let us recall that we have D(A) = A* A as a
vector space, that the Hopf algebras A and (AOP)* are Hopf subalgebras of
D(A), and that the universal R-matrix is given by R = Li ai ai where
{aJi is a basis of A and {aiL is the dual basis. Finally, in Theorem IX.5.2
we proved that a module structure over D(A) is equivalent to a crossed
bimodule structure over A.
We are ready to state the main theorem of this section.

Theorem XIII.5.1. For any finite-dimensional Hopf algebra A with in-

vertible antipode, the braided tensor categories Z(A-Mod) and D(A)-Mod
are equivalent.
We defined the centre construction only for strict tensor categories. But
there is no difficulty in extending it to A-Mod. We start with two prelimi-
nary results before embarking into the proof of Theorem 5.1.

Lemma XIII.5.2. Let (V, c_,v) be an object of Z(A-Mod) and ~v be

the map from V to V A defined for all v E V by ~v(v) = cA,V(1 v).
Then the map ~v endows the left A-module V with a crossed A-bimodule
structure.
PROOF. Let ~v : V ----; V A be defined as above. By convention, we write

~v (v) = L Vv vA E V A (5.1)
(v)

for any v E V. We call ~v the coaction of A on V.

The naturality of c_,v allows us to express cx,v in terms of the coaction
~v for any A-module X. Indeed, given x in X and x : A ----; X the unique
A-linear map sending 1 to x, we have the following commutative square:

AV ~ VA
l~~v l~v~
XV ~ VX.
334 Chapter XIII. Braidings

It implies that for any v E V and x E X we have

cx.v(x Q9 v) = ~v(v)(1 Q9 x) = L Vv Q9 vAx. (5.2)

(v)
Let us show that the coaction ~v is coassociative. By (4.1) we have

L Vv Q9 (vA)'x Q9 (vA)"y
(v)

(c x,v Q9 id y ) ( (id x Q9 Cy,v) (x Q9 y Q9 v) )

L (vv ) v Q9 (vv ) A X Q9 v A y.
(v)

Setting X = Y = A and x = y = 1, we get

L Vv Q9 (VA)' Q9 (VA)" = L (vv)v Q9 (VV)A Q9 VA'

(v) (v)

which proves the coassociativity of ~v.

We also have ck v = id v because k = I is the unit of the tensor category
of k-modules. Thi~ implies c k ,v(1 Q9 v) = L(v) f(vA)vV = v for all v E V.
This means that the coaction ~v is counitary. So far we have proved that
the coaction ~v equips V with a structure of right A-comodule.
Let us now express the fact that Cx v is A-linear. For a E A, v E V and
x E X we have '
acx,v(x Q9 v) = cx,v(a(x Q9 v)).
Replacing cx,v by its expression in ~v, we get

~(a)~v(v)(1 Q9 x) = (L ~v(a"v)(1 Q9 a')) (1 Q9 x).

(a)

Setting X = A and x = 1, we obtain

L a'v v Q9 a"vA = L (a"v)v Q9 (a"v)A a', (5.3)

(a) (v) (a) (v)

which is Relation (5.2) of Chapter IX expressing that V is a crossed A-

bimodule. D

By Theorem IX.5.2, we know that a crosseJ A-bimodule is a left D(A)-

module. Let R = Li a i Q9a i be the universal R-matrix of D(A). We express
the braiding in the braided tensor category Z(A-Mod) in terms of R.
Lemma XIII.5.3. Under the previous hypotheses, if (V, c_ v) is an ob-
ject of Z(A-Mod) and X is an A-module, then the isomorp'hism Cx v is
determined by ,
cx,v(x Q9 v) = Tx ,v(R(xQ9v))
for all x E X and v E V.
 XIII.5 A Categorical Interpretation of the Quantum Double 335

PROOF. By Relations (5.2) and (IX.5.4) we have

Cx,v(X Q9 v) = L Vv Q9 v AX
(v)

L < ai, v A > Vv Q9 aix

(v),i

L ai . v Q9 aix
(v),i

Tx,v(R(XQ9v)).

o
We prove Theorem 5.1 in five steps.
1. We first define a functor F from Z(A-Mod) to D(A)-Mod. Let (V, C_ v)
be an object of Z(A-Mod). By Lemma 5.2 and Theorem IX.5.2, the vector
space F(V, c_ v) = V is a left D(A)-module. Recall from IX.5 that the
action of D(A) on V is determined by

(aa)v = L < a,vA > av v (5.4)

(v)

where a E A, a E A*, and v E V.

If f is a map in Z(A-Mod), then Relation (4.2) implies that f is a map
of A-comodules, hence of A* -modules. Consequently f is D(A)-linear. This
defines F as a faithful functor.
2. Let us show that F is a strict tensor functor. The tensor product of
(V, c_,v) and of (W, c_,w) is (VQ9W, c_,vw) where c_,vw is determined
by cA,vW = (id v Q9 cA,W ) (CA,v Q9id w ) Therefore the coaction on V Q9 W
is given by
~vw(v Q9 w) = L Vv Q9 Ww Q9 wAvA'
(v)(w)

By (5.4) the action of a linear form a on a tensor v Q9 W in V Q9 W can be

expressed as

a'(VQ9w)= L <a,wAvA>vV Q9WW,

(v)(w)

which by definition of the comultiplication ~ of A* (see IX.4) is equal to

L < ~ (a), v A Q9 WA > Vv Q9 Ww = ~(a) . (V Q9 w).

(v)(w)

Therefore the D(A)-action on V Q9 W is given for a E A and a E A* by

(aa)(v Q9 w) = ~(a)(~(a) . (v Q9 w)) = ~(aa)(v Q9 w),

336 Chapter XIII. Braidings

which is exactly the action given by the comultiplication in the quantum

double D(A).
3. By definition of the braiding in Z(A-Mod), Lemma 5.3 can be rein-
terpreted as stating that

F (cv, w )( v c>9 w) = TV, W ( R( v c>9 w) ) ,

which is the braiding in the category of D(A)-modules. Thus, the tensor

functor F is braided.
4. Suppose that V is a left D(A)-module. For any A-module X define
cx,v by
Cx,v(x c>9 v) = TX,v (R(x c>9 v))
where v E V and x E X. Let us show that (V, C,v) is an object of
Z(A-Mod).
The map Cx v is a natural isomorphism because R is invertible. Let us
prove that it is' A-linear. For a E A we have

cx,v(a(x c>9 v)) TX,v (R6.(a)(x v)) c>9

TX,v (6. (a)R(x c>9 v))

0P

6.(a)Tx,v (R(x c>9 v))

acx,v(x c>9 v)
in view of Relation (VIII.2.1).
We also have to check Relation (4.1), namely

CX0)y,v(x c>9 y c>9 v) = (cx,v c>9 id y ) ((id x c>9 cy,v )(x c>9 y c>9 v)).
The left-hand side is equal to

TXY,v ((6. c>9 idA)(R)(x c>9 y c>9 v))

whereas the right-hand side is equal to Txy,v(R 13 R 23 (X c>9 Y c>9 v)). Both
are equal in view of (VIII.2.3). Therefore G(V) = (V, Cv _) is an object in
Z(A-Mod). '
Let f : V ----t W be a map of D(A)-modules. We have to check that
G(j) = f is a morphism in Z(A-Mod). First, it is A-linear since it is
D(A)-linear. Next, we have

TX,W ((id x c>9 j)(R(x c>9 v)))

TX,W (R(x c>9 f(v)))

(cx,v(id x c>9 j))(x c>9 v)
for all x E X and v E V. This proves (4.2).
 XIII.6 Exercises 337

5. Clearly, FG = id. The equality GF = id follows from Lemma 5.3.

This establishes the equivalence of Z(A-Mod) and of D(A)-Mod.
Theorem 5.1 is thus proved. Observe that the same arguments work if
we restrict to finite-dimensional modules.
Remark 5.4. The natural embedding A C D(A) of Hopf algebras induces
a tensor functor D(A)-Mod ---+ A-Mod. It is easy to check that the latter
corresponds to the functor II : Z(A-Mod) ---+ A-Mod of (4.5) under the
equivalence of Theorem 5.1.

XIII. 6 Exercises
1. Let H be a braided bialgebra with universal R-matrix R. Show that
the category H-Mod is symmetric if and only if TH,H(R) = R- 1 .
2. Let C be a strict tensor category. Show that one gets a definition for
a braiding equivalent to Definition 1.1 if one replaces the hexagons
(HI) and H(2) by the square
cU,vid x0y
VUXY
1 idv0UCX, Y

VUYX
for all objects U, V, X, Y.
3. Resume the notation of Exercise XI.S. Define a commutativity con-
straint c by c(vw) = "((n,p)(wv) where v and w are homogeneous
vectors of respective degrees nand p, and where "( is a function with
values in k \ {o}. Show that c is a braiding if and only if the functional
equations
"((m, n + p) = a(n,p, m)-l"((m,p)a(n, m,p),,((m, n)a(m, n,p)-l

and
"((m + n,p) = a(p, m, n)"((m,p)a(m,p, n)-l"((n,p)a(m, n,p)
are satisfied for all integers m, n, p.
c
4. Given a tensor category C, define the reverse category rev as the
category C with tensor product given by V rev W = W V. Prove
that, if C is braided with braiding c, then (id, 'Po = id, 'P2 = c) is a
tensor functor between C and crev .
5. Let C be a braided tensor category. Show that one can equip the
c
strict tensor category str of XI. 5 with a braiding such that the tensor
equivalences constructed in XI.5 between both categories are braided
functors.
338 Chapter XIII. Braidings

6. (Presentation of the braid category) Show that the strict tensor cat-
egory B is generated by the morphisms 0"1' 0"1 1 of B2 and by the
relations 0"10"1 1 = 0"1 1 0"1 = id 2 and

XIII. 7 Notes
Braided tensor categories were introduced by Joyal and Street [JS91a]
[JS93]. They generalize the concept of a symmetric tensor category which
appeared in the 1960's in the work of Benabou [Ben64] and Mac Lane
[Mac63], and was extensively studied in relation to algebraic geometry and
algebraic topology (see, e.g., [DeI90] [DMS2] [KLSO] [Mac63] [SR72]).
The content of Section 3 is taken from [JS93]. Lemma 3.5 is the analogue
of Theorem XII.4.2 for braids. We found the example of crossed G-sets in
[FYS9]. Exercise 4 is from [JS93].
The centre construction of Section 4 is due to Drinfeld (unpublished), to
Joyal and Street [JS91c], and to Majid [Maj91b].
Chapter XIV
Duality in Tensor Categories

In the previous chapter we defined braided tensor categories modelled on

the category of braids. We now introduce a class of tensor categories mod-
elled on framed tangles or ribbons. These are the so-called ribbon cate-
gories. Their definition requires the concept of duality. However, when du-
ality is involved, formulas quickly tend to become obscure and complex. To
overcome this difficulty, we present a graphical calculus in which coloured
tangle diagrams represent morphisms of tensor categories.

XIV.1 Representing Morphisms in a Tensor

Category
We discuss a technique of presenting morphisms of a strict tensor cate-
gory by planar diagrams. Let C be a strict tensor category. We represent
a morphism f : U ----t V in C by a box with two vertical arrows oriented
downwards as in Figure 1.1. Here U and V are treated as the "colours" of
the arrows and f as the "colour" of the box.

Figure 1.1. A morphism f :U ---> V

340 Chapter XIV. Duality in Tensor Categories

The picture for the composition of 1 : U -+ V and of 9 : V -+ W is

obtained by putting the picture of 9 on top of the picture of I, as shown
in Figure 1.2. From now on the symbol ~ displayed in the figures means
that the corresponding morphisms are equal in C. The identity of V will
1
be represented by the vertical arrow v directed downwards.

Figure 1.2. Composition of morphisms

The tensor product of two morphisms 1 and 9 is represented by boxes

placed side by side as in Figure 1.3.

Figure 1.3. Tensor product of morphisms

If we represent a morphism 1 : Ul 0 . 0 Um -+ Vl 0 . 0 Vn as in Figure 1.4,

~
~
Figure 1.4. A morphism f : Ul ... Urn --+ Vl ... Vn

then we have the equality of morphisms of Figure 1.5.

~
' 10gV' ~
U V
cf?'~'
1
U
9
V
Figure 1.5.

The pictorial incarnation of the identity (XI.2.3)

10 9 = (f 0 id) 0 (id 0 g) = (id 0 1) 0 (g 0 id)

is in Figure 1.6.
 XIV.1 Representing Morphisms in a Tensor Category 341

~ 1 tb
CPl
Figure 1.6. The identity (XI.2.3)

This leads to the following "partial isotopy principle": for any figure pre-
senting a morphism of C, the part of the figure lying to the left (or to the
right) of a vertical line may be pushed up or down without changing the
corresponding morphism in C. We shall use this principle frequently and
without any further explanation in the sequel.
Assume now that the tensor category is braided with a braiding c. For
any pair (V, W) of objects in C we represent cv,w and its inverse cv,\.v
respectively by the pictures in Figure 1.7.

V
x x W
cv,W
W
-1
cv,W
V

Figure 1.7. cv, wand its inverse

Figure 1.8 follows from the definitions.

V W V W W V W V
Figure 1.8. Invertibility of cv, w

The naturality of cv,w is expressed in Figure 1.9.

W' V' W' v'

X ~0[]]
1 1
00
1 1 X
V W V W
Figure 1.9. Naturality of cv,W

It implies the naturality of cv\.v

, shown in Figure 1.10.
342 Chapter XIV. Duality in Tensor Categories

V' W' V' w'

X
IT] []J
1 1
[]J0
W
1 1 V
>-<
W V
Figure 1.10. Naturality of CV,lW

Figure 1.11 is a graphical proof of Theorem XIII. 1.3.

~ VU VU
X
>(:~
~~~ ~I
t v
l~
y~
~X
U V W UV W UV W U V W
Figure 1.11. Proof of Theorem XIII. 1.3

XIV.2 Duali ty
We now abstract the notion of duality introduced in I1.3.

Definition XIV.2.1. Let (C, , 1) be a strict tensor category with tensor

product and unit I. It is a tensor category with left duality if for each
object V of C there exist an object V* and morphisms

bv : I ---7 V V* and d v : V* V ---7 I

in the category C such that

(id v d v )(bv id v ) = id v and (d v id v * )(id v * bv ) = id v *. (2.1)

We proceed to give a graphical description of Relations (2.1) using the

conventions of Section 1. Represent the identity of V* by the vertical arrow
rv directed upwards. More generally, we use vertical arrows oriented up-
wards with the convention that the morphism involves not the colour of the
arrow, but rather the dual object. For example, a morphism f : U* ---7 V*
may be represented by the four pictures of Figure 2.1.
 XIV.2 Duality 343

Figure 2.1. A morphism f : U* --t V*

The morphisms bv : I -+ V Q9 V* and d v : V* Q9 V -+ I are respectively

represented by the pictures of Figure 2.2.

f'\v
dv
Figure 2.2. The morphisms bv and d v

Relations (2.1) take the graphical form of Figure 2.3.

Figure 2.3. Relations (2.1)

The above data are enough to extend duality to a functor and to derive
adjunction formulas of the type proved in Chapter II.
Let us first define the transpose 1* : V* -+ U* of a morphism f : U -+ V
in C by
1* = (d v Q9 idu*)(id v * Q9 f Q9 idu*)(id v * Q9 bu ) (2.2)
With our graphical conventions we can represent the transpose 1* of a
morphism f : U -+ V as in Figure 2.4.

Figure 2.4. The transpose 1*

We record a few properties of left duality in the following proposition.

Proposition XIV.2.2. Let C be a strict tensor category with left duality.

(a) If f : V -+ Wand 9 : U -+ V are morphisms of C, then we have
(f 0 g)* = g* 01*, and (id v )* = id v * for any object V.
344 Chapter XIV. Duality in Tensor Categories

(b) For any family U, V, W of objects of C, we have natural bijections

Hom(U Q9 V, W) ~ Hom(U, W Q9 V*)
and
Hom(U* Q9 V, W) ~ Hom(V, U Q9 W).
(c) For any pair V, W of objects of C, (V Q9 W)* and W* Q9 V* are
isomorphic objects.
This proposition implies that the map V 1-+ V* can be extended to a
functor from the category C to the opposite category and that the functor
- Q9 V [resp. the functor V* Q9 - ] is left adjoint to the functor - Q9 V* [resp.
to the functor V Q9 - ].
PROOF. (a) Use the graphical calculus.
(b) Let f E Hom(U Q9 V, W) and g E Hom(U, W Q9 V*). Define elements
fU E Hom(U, W Q9 V*) and l E Hom(U Q9 V, W) by
fU = (J Q9 id v * )(id u Q9 bv ) and l = (id w Q9 d v )(g Q9 id v )

Relations (2.1) imply that (JU)D = f and (l)U = g. A similar proof works
for the other adjunction formula. We invite the reader to give a graphical
proof.
(c) We define a morphism Av,w : W* Q9 V* ---+ (V Q9 W)* by

Av,w = (d w Q9 id(vw)* )(id w * Q9 d v Q9 idw(vw)* )(idw*v* Q9 bvw)

(2.3)
and a morphism Av,~ : (V Q9 W)* ---+ W* Q9 V* by

Av,~ = (dvw Q9 idw*v* )(id(Vw)*v Q9 bw Q9 idv*)(id(vw)* Q9 bv )

(2.4)
The morphisms Av,w and Av,~ are represented by the pictures of Figure
2.5.

VQ9W WV
A
" "

WV VQ9W
Av,w A-I
v,w
Figure 2.5. The morphisms AV,w and AV,\v

Figures 2.6 and 2.7 show that Avw is an isomorphism from W* Q9 V*

onto (V Q9 W)* with inverse given b; Av,~
 XIV.2 Duality 345

wv
" "

wv

wv wv
" " " "

id",-~
lvtZlw
I id v0;l

wv
W v W V

.. ProojojA-V,W
Figure 26 1
0 AV,W = id W*0V*
346 Chapter XIV. Duality in Tensor Categories

VW
"

VW VW
" "

'T~ v v

! idvw!
VW VW

W9W
"

Figure 2.7. Proof of Av,W 0 AV;,\.v = id(vw)*

Note that Figures 1.6 and 2.3 are used in these graphical proofs. 0

There is a similar notion ofright duality: a strict tensor category (C, , 1)

is a tensor category with right duality if for each object V of C there exist
an object *V and morphisms

b~ :I -+ * V V and d~: V *V -+ I
 XIV.2 Duality 347

in the category C such that

Here again V f---t * V may be extended to a functor by defining the mor-

phism * j : * W ----t * V for any j : V ----t W by

Right duality has properties analogous to the ones stated for left duality
in Proposition 2.2. We leave their formulation to the reader. In particular,
right duality implies that the functor V 0 - [resp. the functor - 0 *V] is
left adjoint to the functor *V 0 - [resp. to the functor - 0 V].
In general, right duality is different from left duality unless we add extra
hypotheses on C. Nevertheless, it may happen that C is autonomous, i.e.,
it has left and right duality. In this case, there are isomorphisms

*(V*) ~ V ~ (*V)*

for any object V. We refer to [JS93] for a proof. Hint: the first isomorphism
is a consequence of the following natural isomorphisms

Hom(U, *(V*) 0 W) ~ Hom(V* 0 U, W) ~ Hom(U, V 0 W),

the first one being implied by the right duality and the second one by the
left duality.
Example 1. Let A be a Hopf algebra with antipode S. The category A-
Mod f of left A-modules that are finite-dimensional over the ground field k
is a tensor subcategory of A-Mod. For any left A-module, endow the dual
vector space V* = Hom(V, k) with the A-action given by

< aj,v >=< j,S(a)v > (2.7)

where a E A, v E V and j E V*. For a finite-dimensional A-module V
define maps bv : k ----t V 0 V* and dv : V* 0 V ----t k by

bv (1) = 2:= Vi 0vi and dV(v i 0Vj) =< vi,v j > (2.8)

where {vJ i is any basis of V and {Vi L is the dual basis in V*. The map bv
is the coevaluation map and the map d v is the evaluation map of II.3. It was
proved in Proposition III.5.3 that bv and d v are A-linear. By Proposition
II.3.1 they satisfy Relations (2.1), endowing A-Mod f with the structure of
a tensor category with left duality.
Suppose, furthermore, that the antipode S is invertible. For any left A-
module V, denote by *V the same dual vector space now equipped with
the left A-action given for all a E A, v E V and all linear forms j on V by

< aj,v >=< j,S-l(a)v >. (2.9)

348 Chapter XIV. Duality in Tensor Categories

For a finite-dimensional V define maps b~ : k -+ * V V and d~ : V * V -+

k by

(2.10)

using the same conventions as above. One checks that b~ and d~ are A-
linear and that they satisfy Relations (2.5), endowing A-Mod j also with
the structure of a tensor category with right duality. In other words, the
category A-Mod j is autonomous when A is a Hopf algebra with invertible
antipode.

XIV.3 Ribbon Categories

Let C be a strict tensor category. Suppose that it is braided and has left
duality at the same time. Let V and W be objects of the category. The
following expresses the braiding cv* w for the dual object V* in terms of
the braiding cv, w. '
Proposition XIV.3.1. Under the previous hypothesis, we have

By convention, we represent cV*, wand its inverse by the pictures in

Figure 3.1.

CV*,w

Figure 3.1. CV*, wand its inverse

The pictorial transcription of Proposition 3.1 is then in the next figure.

Figure 3.2. The equality of Proposition 3.1

A graphical proof of it is then:

 X ~ \/ .f'~
r( r }
XIV.3 Ribbon Categories 349

V \ w W. (;'J" w
Figure 3.3. Proof of Proposition 3.1
~
w -
V

The first equality follows from (2.1), the second one from Figure 1.8, and
the last one from the naturality of the braiding c_,w.
We go one step further by introducing the concept of a ribbon category.
Definition XIV.3.2. Let (C, , I) be a strict braided tensor category with
left duality.
(a) A twist is a family ()v : V -t V of natural isomorphisms indexed by
the objects V of C such that
()V0W = (()v ()w)cw,vcv,w (3.1)
and
(3.2)
for all objects V, W in C.
(b) A ribbon category is a strict braided tensor category with left duality
and with a twist.
The naturality of the twist means that for any morphism f : V - t W
we have ()wf = j()v. Using the graphical conventions of Sections 1-2,
Relations (3.1-3.2) may be represented as in Figures 3.4 and 3.5.

VW V W

Figure 3.4. Relation (3.1)

V*
Figure 3.5. Relation (3.2)
350 Chapter XIV. Duality in Tensor Categories

The following gives alternative expressions for Relation (3.1).

Lemma XIV.3.3. (a) Given objects V and W of C we have

BV (8)W = cw,vcv,w(B v Q9 Bw) = cw,v(Bw Q9 Bv)cv,w' (3.3)

(b) We also have BI = id I .

PROOF. (a) See Figure 3.6. All equalities follow by naturality of the braiding
and of the twist.

~~
1

(5
1
IB~
l~ )
x
~~

X '~ X
...:..

V W V W V W

><
~J
~\
...:.. ...:..

~ ...
V W
Figure 3.6. Proof of Lemma 3.3 (a)

(b) Using Relation (3.1) when V = W = I, we get

BI (8)I = (BI Q9 BI )CI,ICI,I'

Since cI I = id I (by Proposition XIII.1.2) and by naturality of the identi-

fication 'of V Q9 I with I, we get BI (8)I = BI Q9 id I = BI Q9 BI , which implies
the statement. D

Example 1. Let Vect f (k) be the category of finite-dimensional vector spaces

over a field k. As we know, it is braided by the flip and it has left duals
given by (2.7-2.8). This category is a ribbon category with trivial twist
Bv = id v
Example 2. Any symmetric tensor category C with left duality is a ribbon
category with twist given by Bv = id v for any object V. In this class falls
the category A-Mod f of finite-dimensional modules over a cocommutative
 XIV.3 Ribbon Categories 351

Hopf algebra or over a braided Hopf algebra A with universal R-matrix R

such that T A,A (R) = R- 1 (see Exercise 1 in Chapter XIII).
Using the braiding and the twist, we define morphisms b;" : 1-+ V* V
and d~ : V V* -+ I for any object V of the ribbon category C by

(3.4)

and
d~ = dvcv,v* (e v id v .). (3.5)
We shall agree to represent b;" and d;" as in Figure 3.7.

V~U
b'v d'v
Figure 3.7. The morphisms b~ and d~

Let us prove that the morphisms b~ and d;" equip C with the structure
of a category with right duality, where *V = V*. Before we give a precise
statement, we shall prove the following technical lemma.
Lemma XIV.3.4. For any object V of a ribbon category, we have

e~? v,\;
(d v idv)(id v * c )(cv,v.b v id v )
(dvcv,v* idv)(id v cv,vb v )
(id v dv cv,v * ) (c v,\; id v * )(id v bv )'

PROOF. The equalities of this lemma are represented in Figure 3.8.

V V V
Figure 3.8. The equalities of Lemma 3.4

It is clear from the pictures that the naturality of the braiding implies the
last two equalities. So it is enough to prove the first one.
By naturality of the twist and by Lemma 3.3 (b), we get

(3.6)

Let us denote by f the second term of the equalities in Lemma 3.4. Figure
3.9 shows that the right-hand side of (3.6) is equal to (e~ f id v * )b v
352 Chapter XIV. Duality in Tensor Categories

...:...
c
X C!!
~
...:...
evv* bv

I~Srl
rn

Figure 3.9. Proof of Bv@v*bv = (B~ f Q9 id v * )bv

Therefore, applying Relations (2.1) twice, we get

v
Therefore, we have f = e 2 , as desired.
As for Figure 3.9, the first equality follows from (3.1), the second one
from Proposition 3.1 and from (3.2), the third one from the naturality of
the tensor product, the fourth one from (2.1), and the last one from the
naturality of the braiding. 0

Proposition XIV.3.5. Under the previous hypotheses, we have

(d~ Q9 idv)(id v Q9 b~) = id v (3.7)

and
(3.8)
 XIV.3 Ribbon Categories 353

PROOF. (a) By (3.4-3.5) and the naturality of the tensor product, we have

(d~ id v )(id v b~)

( dvcv,v* (Bv id v *) id v ) (id v (id v * Bv )cv,v* bv )

BvgBv

where 9 is the third term of the equalities in Lemma 3.4. Consequently, we

have
(d~ id v )(id v b~) = BvB\/Bv = id v .
(b) The proof of (3.8) is in Figure 3.10.

Figure 3.10. Proof of Relation (3.8)

The first equality is by definition, the second one by (2.1), the third one
by naturality of the braiding, the fourth one by Lemma 3.4, the fifth one
follows from Figure 1.8, and the last one from (2.1). 0

It follows from Proposition 3.5 that a ribbon category is autonomous

with *V = V* in the sense of Section 2.

Corollary XIV.3.6. Any object V of a ribbon category is isomorphic to

its bidual V** = (V*)*.

PROOF. We saw in Section 2 that V ~ * (V*) in any autonomous tensor

category. 0
354 Chapter XIV. Duality in Tensor Categories

The pictures in Figure 3.11 represent isomorphisms between V and V**.

V*
L~
Figure 3.11. Isomorphisms between V and V* *

XIV.4 Quantum Trace and Dimension

By analogy with II.3 we define the concept of trace in a ribbon category.

Definition XIV.4.1. Let C be a ribbon category with unit I. For any ob-
ject V of C and any endomorphism f of V, we define the quantum trace
trq(f) of f as the element

trq(f) = d'v(f @ id v * )b v = dvcv,v* (e v f @ id v * )b v

of the monoid End(I), i.e., as the composition of the morphisms

I~V@V* evfid~V@V*~V*@V~I.

This notion coincides with the usual trace when C is the category Vect f (k)
(see Proposition II.3.5). Graphical representations of trq(f) are given in
Figure 4.1.

Figure 4.1. The quantum trace of f

The first equality is by definition, the second one by naturality of the

braiding, the third one by naturality of the twist, and the last one by
definition of b'v.
The quantum trace enjoys the usual properties of the trace in linear
algebra.
 XIV.4 Quantum Trace and Dimension 355

Theorem XIVA.2. Given endomorphisms f and 9 in a ribbon category,

we have
(a) trq(fg) = trq(gf) whenever f and 9 are composable,
(b) trq(f 0 g) = trq(f) trq(g), and
(c) trq(f) = trq(f*) in the monoid End(I).
PROOF. (a) The proof of the first relation is in Figure 4.2.

v
Figure 4.2. The proof oftrq(fg) = trq(gf)

The first equality is by definition, the second one by (2.1), the third
one by naturality of the braiding, the fourth one by (2.1), the fifth one by
naturality of the braiding, the sixth one by naturality of the twist, and the
last one by definition.
(b) We know from Proposition XI.2.4 that the composition in End(I)
coincides with the tensor product. Therefore, it is equivalent to prove that
trq(f 0 g) = trq(f) 0 trq(g). The proof of the latter is in Figure 4.3. The
first and last equalities in that diagram on the next page follow from the
definition, the second one from (3.1), the third one from (2.1), the fourth,
sixth, and seventh ones by naturality of the braiding.
(c) The proof of tr q(f) = tr q(f*) is in Figure 4.4 two pages on.
356 Chapter XIV. Duality in Tensor Categories

~~~~
SVwV
~ c@
V
c@
W
Figure 4.3. The proof of trq(f 0 g) = trq(f) 0 trq(g)
 XIVA Quantum Trace and Dimension 357

I@
V*

Figure 4.4. The proof oftrq(f) = trq(f*)

358 Chapter XIV. Duality in Tensor Categories

The first equality in the diagram on the previous page follows by defini-
tion, the second one by (3.2) and by Proposition 3.1, the third one by (3.8),
the fourth one by naturality of the braiding, the fifth and the sixth ones
by (2.1), the seventh one by definition of d'v, the eighth one by naturality
of the twist, the ninth one by Lemma 3.4. 0

As in the classical case, we can derive a concept of dimension from the

trace.

Definition XIV.4.3. Let C be a ribbon category with unit I. For any ob-
ject V of C we define the quantum dimension dimq(V) as the element

dimq(V) = trq(id v ) = d'vb v

of the monoid End(I).

The quantum dimension of V is represented in Figure 4.5.

Figure 4.5. The quantum dimension of V

As an immediate consequence of Theorem 4.2, we have

Corollary XIV.4.4. Let V, W be objects of a ribbon category. Then

XIV.5 Examples of Ribbon Categories

XIV.5.1 Ribbons
In X.8 we defined the concept of a framed tangle, also called a ribbon,
and we explained how ribbons could be represented by tangle diagrams.
We now wish to show that ribbons allow one to build a ribbon category
R which is universal for ribbon categories, as the category B of braids is
universal for braided tensor categories.
The category R of ribbons is defined in the same way as the category
T of tangles in XII.2: the objects of R are the same as the objects of T;
the morphisms of R are isotopy classes of framed tangles. Composition,
identity, tensor product, and unit are defined as in T. The braiding of the
braid category B (see XIII.2) clearly defines a braiding in R.
Let us endow the strict braided tensor category R with a left duality. Let
e be an object of R, namely a finite sequence (e l' ... ,en) of signs. Define
 XIV.5 Examples of Ribbon Categories 359

a dual object c* by the sequence (-c n , ... , -Cl). The maps be : 0 ~ C 0 c*

and de : c* 0 c ~ 0 are the framed tangles represented in Figure 5.1, the
orientation of the strands being uniquely determined by the sequence of
signs in c.

~W7 A
V /A~
Figure 5.1. The framed tangles be and de

It is easy to check that the maps be and de satisfy Relation (2.1), thus
equipping R with the structure of a strict braided tensor category with left
duality. Observe that the transpose L * of a ribbon L is isotopic to the ribbon
obtained by rotating L through an angle 7r around an axis perpendicular
to the plane of projection.
We define a twist on R as follows: 8(+) is the left ribbon of Figure X.8.1
oriented downwards (also represented by the left tangle diagram of Figure
X.8.2). The right ribbon of Figure X.8.1 oriented downwards defines the
inverse of 8(+) (it is represented by the right tangle diagram of Figure
X.8.2). To define the twist for an arbitrary object, we use Relations (3.1-
3.2). Check that, if c is of length n, then 8e is obtained by twisting by an
angle of 27r the plane containing n vertical fiat ribbons.
Quantum trace and quantum dimension are defined in the ribbon cat-
egory R by the formulas of Section 4. One can check using Reidemeister
Transformations (1') and (II) that if L is a ribbon with s(L) = b(L), then
its quantum trace trq(L) is the closure of L drawn in Figure 5.2. Quantum
dimensions are trivial links with the framing pointing to the reader.

Figure 5.2. The quantum trace trq(L)

The category R has two universal properties similar to the ones given for
the category B in XIII.3. For the first one which corresponds to Theorem
XII.4.2, we refer to PS91c] [Tur89]. We state the second one paralleling
Corollary XIII.3.8.
360 Chapter XIV. Duality in Tensor Categories

Theorem XIV.5.l. Let C be a ribbon category and V be an object of C.

Then there exists a unique strict tensor functor Fv from R to C preserving
the braiding, the left duality and the twist, such that F v (+) = V.

A proof of Theorem 5.1 can be found in several places: [FY89] [JS93]

[RT90] [RT91] [Shu90] [Tur94]. This theorem produces isotopy invariants
for framed links with values in the endomorphism monoid of the unit object
I. Indeed, let L be a framed link. It can be viewed as an endomorphism of
the unit 0 of the category R. Its image Fv(L) is an endomorphism of I. The
isotopy invariant Fv(L) can be computed using the following algorithm:
take a planar diagram representing the framed link L and colour each
connected component with the object V. In the category R, the framed link
r,
L is obtained by composition and tensor product of L X+, u, n, the twist,
etc. Then one gets Fv (L) by replacing the above diagrams respectively by
id v , id v *, cv,v, by, d v , Bv , etc. in the categorical expression for L.
The category T of tangles is also a ribbon category. The only difference
with R lies with the twist: we have B() = id() in T. The twist of a general
object of T can be defined from (3.1-3.2). There is a statement similar to
Theorem 5.1 for the category T. It suffices to replace R by T and to add
the hypothesis Bv = id v to the category C.

XIV.5.2 Crossed G-sets

In XIIL1.4 we considered a category Xc(G) of crossed G-sets where G is a
group. Assume G is finite. We construct a ribbon category Z[Xc( G)] out
of Xc(G). Its objects are the same as for Xc(G) , namely the denumerable
set of crossed sets {I, G, G@2 , ... }. A morphism G@n -+ G@m in Z [Xc (G)]
is an integral matrix Ill{ indexed by all elements (x, y) E G n x G m such
that M xg yg = Mx,y for all g E G and Mx,y = 0 if Ixl i= Iyl. Composition is
defined by multiplying matrices. The category Z[Xc(G)] is a strict tensor
category. Observe that the monoid End(l) is the ring Z. The braiding of
Xc(G) extends linearly to a braiding on Z[Xc(G)].
Define the duality as follows: the dual of G@n has the same underlying
G-set as G@n, but II is replaced by 11- 1 . The maps bcMn and are
defined by

and
du2Jn (gl"'" gn' h 1,, h n ) = I5 g, . h, .. . l5 gn ,h n
Relations (2.1) are satisfied. A twist Bc((n is defined inductively on n by
(3.1) and its initial value Bc = id z1cJ ' One checks that the quantum di-
mension of an object is its cardinality:

dimq(G@rI) = card (G)n.

 XIV.6 Ribbon Algebras 361

Apply Theorem 5.1 to the ribbon category Z[Xa(G)] and to the object
G. We get an endomorphism Fa(L) of the unit object {I}, i.e., an integer
for any (framed) link L. We invite the reader to use the algorithm described
above to compute this isotopy invariant for a few simple links. For instance,
Fa (L) is equal to the number of couples (g1' g2) E G x G with g1g2 = g2g1 if
L is the Hopflink, and with glg2g)"1 = g:;l glg2 if L is the trefoil knot. Freyd
and Yetter [FY89] proved for a general link L that Fa(L) is the number of
group homomorphisms from the fundamental group of L to G.

XIV.6 Ribbon Algebras

We conclude this chapter by giving examples of ribbon categories consisting
of modules over braided Hopf algebras (defined in VlIl.2). Let D be a
braided Hopf algebra with universal R-matrix R = ~i Si 0ti E D0D. Set

(6.1)

We showed in VIll.4 that u was an invertible element of D with inverse

(6.2)

that uS(u) = S(u)u was central in D, and that we had

c(u) = 1 and ~(u) = (R21R)-1(U0U). (6.3)

Moreover, the square of the antipode is given for any x in D by

(6.4)

Definition XIV.6.1. A braided Hopf algebra D is a ribbon algebra if there

exists a central element 0 in D satisfying the relations

~(O) = (R21R)-1(0 0 0), 1::(0) = 1, and S(O) = O. (6.5)

Ribbon algebras produce ribbon categories.

Proposition XIV.6.2. For any ribbon algebra D, the tensor category D-

Mod f is a ribbon category with twist 0v given on any D-module V by the
multiplication by the inverse of the central element O.
Conversely, if D is a finite-dimensional braided Hopf algebra and the
braided category D-Mod f with left duality is a ribbon category, then D is
a ribbon algebra.

PROOF. (a) Let D be a ribbon algebra with the distinguished central ele-
ment O. Braiding and duality in D-Mod f are given as in XIlL1.3 and XlV.2,
362 Chapter XIV. Duality in Tensor Categories

Example 1. Define a twist 0v for any D-module V by 0v(v) = 0- 1 v where

v E V.
The endomorphism 0v is a D-linear automorphism since the element 0
is central and invertible. Let us prove Relations (3.1-3.2) using (6.5). We
have

(0- 1 0-1)(R21R)(vw)
~(O-l)(V w)
0vw(v w).

As for (3.2), we have for any v E V and 00 E V*

< (Ov)* (00), v > < 00, 0v(v) >

< 00, 0-l V >
< oo,5(0-1)v >
< 0- 1 00, v>
< 0v*OO, v > .
(b) We now assume that D is finite-dimensional and that D- Mod f is a
ribbon category. By Proposition XIIL1.4, we know that D is braided. Since
D is assumed to be finite-dimensional, we consider the left D-module D
and the corresponding twist 0D' Define 0 = 0D(l)-l. Then by functoriality
of the twist, we have for any finite-dimensional D-module

The D-linearity of OD implies that 0 is central. From Relation (3.1) we

conclude that
~(0-1) = (0- 1 0-1)(R21R)

whereas Relation (3.2) implies 5(0- 1) = 0- 1. Finally, c(O) = 1 follows from

Lemma 3.3 (b). 0

Corollary XIV.6.3. The central element 02 of a ribbon algebra acts as

u5(u) on any finite-dimensional module.

As a consequence, we see that 0 2 = u5(u) if D is finite-dimensional.

PROOF. By Proposition 6.2 we know that 02 acts as Ov 2 on V. Now by
Lemma 3.4 we have in any ribbon category

It is therefore enough to compute the action on V of the right-hand side of

this equality. Let {vJ i be a basis of V and {Vi L be the dual basis. Then,
 XIV.6 Ribbon Algebras 363

using (VIII.3.1-3.2), (6.1), and (6.4), we have for any v E V

(id v Q9 dvcv,v*)( c v,\; Q9 id v * )(id v Q9 bv )( v)

L < tkv i , sktjV > S(Sj)Vi

i.j,k
L < vi, S(tk)SktjV > S(Sj)Vi
i,j,k

L S(Sj)S(tk)Sktj V
j,k
L S(Sj)utj v
j

L S(Sj)S2(tj)UV
j

J
S(U)UV = uS(U) V.
D

Quantum trace and quantum dimension are given in the category of

modules over a ribbon algebra by

Proposition XIV.6.4. Let V be any finite-dimensional module over a rib-

bon algebra D. We have

trq(f) = tr(v f-+ e-1uf(v))

for any endomorphism f of V. In particular, dimq (V) is equal to the trace

of the multiplication by e-1u on V.
PROOF. Using the definitions of d'v and of U as well as Proposition 6.2, we
immediately get

d'v(vQ9Oo) = L < tiOo, Sie-Iv > = L < a, S(ti)Sie-1v > = < a, ue-1v > .

Therefore,

which is the trace of the linear endomorphism v f-+ e-Iuf(v). D

We end with two examples of ribbon algebras.

Example 1. (5weedler's four-dimensional Hopf algebra) Consider the Hopf
algebra H of Example 2 in VIII.2. It is braided. An immediate computation
364 Chapter XIV. Duality in Tensor Categories

shows that the element u).. corresponding to the universal R-matrix R).. is
independent of the parameter A and is given by u).. = x = S(u)..). Therefore
u)..S(u)..) = x 2 = l. One checks that H is a ribbon algebra with 8 = l.
Example 2. This is due to Reshetikhin and Turaev [RT91]. It deals with the
Hopf algebra U q' i.e., the finite-dimensional quotient of U/5 [( 2)) considered
in VI. 5 for q a root of unity. In IX.6-7 we proved that U q was a braided Hopf
algebra and we computed its universal R-matrix. We resume the notation
and conventions of Chapter IX. In particular, we assume that q is a root
of unity of odd order d > l.

Proposition XIV.6.5. The Hopf algebra U q is a ribbon algebra for which

8 = K-1u = uK-I.

PROOF. The centrality of 8 follows from (6.4) and from the fact that we
also have
S2(X) = KxK- 1
for all x E U q. It is immediate to check that s( 8) = l. As for ~ (8), we have

It remains to check that S(8) = 8. This is equivalent to KS(u) = K-1u.

Now this can be verified directly using the formula given in IX.7 for the
universal R-matrix for U q. Alternatively, there is an argument in [Dri89a],
Section 5 which goes roughly as follows: let V).. be a highest weight module
with highest weight A. Then KS(u) = S(uK- 1) = S(K-1u) acts like K-1u
on the dual module V;. But the latter is isomorphic to V)... Therefore K S( u)
acts like the central element K-1u on V)... A general argument (see [RT91])
extends this to any module. 0

Let Vn = V1,n be the simple U q-module of VI.5. Its quantum dimension

is given by
. qn+l _ q-n-l
dlmq(Vn ) = [n + 1] = -1
q-q
In effect, by Proposition 6.4 it is given as the trace of the multiplication
by 8- 1 u. Here 8- 1 u = K. Now K acts diagonally on Vn with eigenvalues
{ qn ,qn-2 , ... , q-n+2 , q -n} . Therelore,
C

qn+l _ q-n-l
dim (V)=qn+ qn-2+ ... + q-n+2+ q -n= =[n+1].
q n q _ q-l
 XIV.7 Exercises 365

XIV.7 Exercises
l. For any braided Hopf algebra D, define an algebra D(8) as the quo-
tient of the polynomial algebra D[8] by the two-sided ideal generated
by 8 2 - uS(u). Show that D(8) has a unique Hopf algebra structure
such that the natural inclusion of D in D(8) is a Hopf algebra map
and that ~(8) = (R21R)-1(8 8), c(8) = 1, and S(8) = 8. Prove
that D(8) is a ribbon algebra.
2. Under the hypotheses of the previous exercise, show that the category
of left D(8)-modules is equivalent to the category whose objects are
pairs (V, 8v ) where V is a left D-module and 8v is a D-linear auto-
morphism of V such that for all v in V we have 8V2 (v) = uS(u) v,
and whose morphisms (V, 8v ) ---t (W, 8w ) are the D-linear maps f
from V to W such that f8 v = 8w f.
3. Using the definitions and the notation of 5.2, compute 8c @n for n > l.
4. Given a finite abelian group A and a commutative ring K, let K(A) be
the commutative K-algebra of K-valued functions on A. It has a basis
{eo} aEA over K such that the multiplication is given by eoe b = 8a,b
for all a, b E A.
(a) Show that there is a unique Hopf algebra structure on K(A) such
that for all a E A we have

~(eJ = L eb ea-b' c(e a ) = 80 ,' S(e a ) = e_ a

bEA
(b) Let R = La,bEA c( a, b) eo eb where c is a function with values
in the group K X of invertible elements of K. Prove that R equips
K(A) with the structure of a braided Hopf algebra if and only if
c(a + a', b) = c(a, b)c(a', b) and c(a, b + b') = c(a, b)c(a, b')
for all a, a', b, b' E A.
(c) We assume that K (A) with R as defined in (b) is a braided Hopf
algebra. Let X be a group homomorphism from A to K X such that
x(a)2 = 1 for all a E A. Show that 8 = LaEA x(a)c(a, a)e a endows
K(A) with the structure of a ribbon algebra.
(d) Check that the quantum dimension of the K(A)-module eaK(A)
is equal to x(a).
5. (Coribbon algebras) Let H = (H, iL, 7],~, c, S, r) be a cobraided Hopf
algebra with universal R-form r (see VIII.5). It is a coribbon algebra
if there exists an invertible central element ( of the dual algebra H*
such that
366 Chapter XIV. Duality in Tensor Categories

where l' is the inverse of r for the convolution * in (H H)* and

where 1'21 = r 0 TH,H' Prove that the category of finite-dimensional
H-comodules is a ribbon category.

6. Show that the cobraided Hopf algebra SLq(2) is a coribbon algebra

with central linear form ( determined by

((a) = ((d) = q-3/2 and ((b) = ((c) = 0.

XIV.8 Notes
The graphical calculus described in Section 1 was advocated in many pa-
pers, e.g., [FY89] [FY92] [JS91a] [Kau91] [RT90] [RT91].
The concept of duality in a tensor category appeared in the classical
references quoted in Chapter XIII. The examples presented in this book
require distinguishing carefully between left and right duality. In Section 2
we followed Joyal and Street's treatment of duality as proposed in [JS93]
(see also [FY89]). There Joyal and Street also introduced the concept of
a twist in a strict braided tensor category and the concept of a ribbon
category. Actually, they called the latter tortile tensor categories. The name
used here was coined by Turaev [Tur92].
Definition 4.1 is due to Turaev [Tur92] generalizing previous definitions
of [KL80] and [FY89]. We devised a proof of Theorem 4.2 highlighting the
power of the graphical calculus of Section 1 (a different proof can be found
in [Tur94]).
Ribbon algebras were invented by Reshetikhin and Turaev [RT90] who
also showed that the quantum groups of Drinfeld and Jimbo gave birth to
ribbon algebras.
The construction of the ribbon algebra D(B) of Exercise 1 is taken from
[RT90]. Exercise 4 is due to Turaev: this example does not produce any
interesting isotopy invariant. Exercise 5 is from [JS91b]. Exercise 6 is due
to the author.
There exists an elaboration of the centre construction of XIII.4, to be
found in [KT92], which assigns to any strict tensor category C with left
duality a ribbon category D(C). It is related to the quantum double of
a finite-dimensional Hopf algebra A with in-.'ertible antipode and to the
construction of Exercise 1 by the equivalence of ribbon categories

By colouring framed tangle diagrams with objects and morphisms of a

strict tensor category C, we may construct a ribbon category R(C) with
the following property: the construction C f-+ R(C) is functorial and is
left adjoint to the forgetful functor from the category of ribbon categories
to the category of strict tensor categories. In other words, given a ribbon
 XIV.8 Notes 367

category C', there is a natural bijection between the set of strict braided
tensor functors preserving duality and twist from R( C) to C' and the set of
strict tensor functors from C to C'. In particular, if C is a ribbon category,
then the identity functor of C corresponds to a functor Fe : R( C) -> C
preserving tensor products, braidings, duality and twists. For more details
on the category R(C), see [FY89] [JS93] [RT90] [RT91] [Tur92] [Tur94].
The existence of the functor Fe allows one to find isotopy invariants for
framed links with values in the endomorphism monoid of the unit object
of the ribbon category C. Proceed as at the end of 5.1. The main difference
is that we are now permitted to colour the connected components of a link
with different objects of the category rather than with one single object.
Chapter xv
Quasi-Bialgebras

The aim of this chapter is to present Drinfeld's concepts of (braided) quasi-

bialgebra and of gauge transformation. These concepts will be needed to
express the main results of Part IV. The definitions given here are based
on the formalism of tensor categories and tensor functors introduced in
Chapters XI and XIII. In Section 4 we construct braid group representa-
tions for any braided quasi-bialgebra and we show that equivalent braided
quasi-bialgebras give rise to equivalent representations.
We shall make frequent use of the convention of VIII.2 regarding sub-
scripts.

XV.l Quasi-Bialgebras
In XI.3.1 we introduced the notion of an algebra (A, 6., c) with comultipli-
cation and counit: it is an associative unital k-algebra A with a morphism
of algebras 6. : A --> A A (the comultiplication) and a morphism of alge-
bras c : A --> k (the counit). We observed that the classical tensor product
on Vect( k) restricted to a tensor product on the category A-Mod of left
A-modules, for which I = k is a unit.

Definition XV. I. I. Let A = (A, 6., c) be an algebra with comultiplication

and counit. It is a quasi-bialgebra if the category A-Mod equipped with the
tensor product of Vect(k) is a tensor category.
 XV.l Quasi-Bialgebras 369

In other words, (A, 6., 0:) is a quasi-bialgebra if there exists an associa-

tivity constraint a, a left unit constraint l, and a right unit constraint r
satisfying the Pentagon Axiom (XI.2.6) and the Triangle Axiom (XI.2.9).
When these constraints are the usual ones of Vect(k), then A is a bialgebra
by Proposition X1.3.1. We now give a characterization of quasi-bialgebras,
which actually is Drinfeld's original definition in [Dri89b].
Proposition XV.1.2. Let (A, 6., 0:) be an algebra with comultiplication
and c01mit as above. It is a quasi-bialgebra if and only if there exist an
invertible element <I> in A 0 A 0 A and invertible elements l, r in A such
that
(id 0 6.)(6.(a)) = <1>( (6. 0 id)(6.(a)) )<1>-1, (1.1)

(0: 0 id)(6.(a)) = l-l al, (id 0 0:) (6.(a)) = r- 1 ar (1.2)

for all a E A,

(id 0 id 0 6.)(<1 (6. 0 id 0 id)(<I = <1>234 (id 0 6. 0 id)(<I <1>123' (1.3)

and
(1.4)
Here <1>123 = <I> 0 1 and <1>234 = 1 0 <I> according to the conventions of
VIII. 2. When <I> = 1010 1 and I = r = 1, we recover the usual definition of
a bialgebra. From Proposition 1.2 we see that the main difference between
a bialgebra and a quasi-bialgebra lies in the fact that the comultiplication
of a quasi-bialgebra is no longer coassociative. Nevertheless, Relation (1.1)
shows that it is almost. This situation is reminiscent of braided bialgebras
whose non-cocommutativity is also controlled (see VIII.2).
The elements <1>, I and r are part of the definition of a quasi-bialgebra.
Therefore, we shall denote such a quasi-bialgebra by (A,6.,o:,<I>,l,r). The
element <I> is sometimes called the Drinfeld associator of A.
PROOF. (a) Let <1>, land r be elements satisfying Relations (1.1-1.4). For
any triple (U, V, W) of A-modules, define an associativity constraint by

au,v,w((U0V)0W) = <I>(u0(v0 w )) (1.5)

foru E U, v E V and W E W), and unit constraints

Iv(10 v) = Iv and rv(v 01) = rv. (1.6)
The maps a, l, and r are isomorphisms because <1>, I, and r are invertible.
They are A-linear thanks to Relations (1.1-1.2). Relations (1.3) and (1.4)
imply the Pentagon and the Triangle Axioms respectively.
(b) Conversely, suppose that A-Mod is a tensor category with associa-
tivity and unit constraints a, I, and r. From the associativity constraint
define an element <I> in A 0 A 0 A by
(1. 7)
370 Chapter XV. Quasi-Bialgebras

Similarly, from the unit constraints we get elements

(1.8)

of A. Let us check that <1>, I and r satisfy the conditions of the proposi-
tion. First, these elements are invertible because the constraints are iso-
morphisms.
We next prove Relation (1.2). This is done as in the proof of Proposition
XIII.1.4: by functoriality of the associativity constraint, for all u E U, v E V
and w E W, we have the commutative square

(AQ9A)Q9A ~ A Q9 (A Q9 A)
1
(u<Slv)<SI w 1 u<Sl(v<Slw)
(UQ9V)Q9W ~ U Q9 (V Q9 W)

where for any element u of an A-module U we denote by il the unique

A-linear map from A to U sending 1 onto u. Hence

Let us express the A-linearity of au,v,w. On one hand we have

au, v, w ( a ( ( u Q9 v) Q9 W )) = <I> ( (~ Q9 id) (~( a)) ) (u Q9 (v Q9 w) ) .

On the other hand, we get

a( <1>(uQ9 (v Q9 w)))
( (id Q9 ~) (~( a) ) <I> ) ( U Q9 (v Q9 w) ) .

Setting U = v = W = 1 E A yields Relation (1.1). Similarly, the functoriality

of the unit constraints implies that

Iv(l Q9 v) = Iv and rv(v Q91) = rv. (1.10)

The A-linearity of I A and of r A implies (1.2).

It remains to check Relations (1.3-1.4). By the Pentagon Axiom (XI.2.6)
we have

aA,A,A<SIA 0 aA<SIA,A,A = (idA Q9 aA,A,A) 0 aA,A<SIA,A 0 (aA,A,A Q9 idA)'

Applying both sides of this equation to 1 Q9 1 Q9 1 Q9 1 and using (1.9), we
get Relation (1.3). A similar proof shows that the Triangle Axiom implies
Relation (1.4). 0

We shall see examples of quasi-bialgebras that are not bialgebras later.

All of them will have trivial unit constraints, i.e., I = r = 1.
 XV.2 Braided Quasi-Bialgebras 371

We also need the following concept: a morphism of quasi-bialgebras

a: (A,~,E,<I>,l,r) ---+ (A',~',E',<I>',l',r')

is a morphism of algebras between the underlying algebras such that

(a 0 a)~ = ~Ia, (a 0 a 0 a)(<I = <I>', a(l) = l', a(r) = r'. (1.11)

It is an isomorphism of quasi-bialgebras if, in addition, it is invertible.

XV.2 Braided Quasi-Bialgebras

We now define the counterpart of braided algebras (VIII.2) in the context
of quasi-bialgebras.

Definition XV.2.1. A quasi-bialgebra (A,~,E,<I>,l,r) is braided if the

corresponding tensor category A-Mod is braided.

We characterize braided quasi-algebras (also called quasi-triangular quasi-

bialgebras in the literature) as we did for quasi-bialgebras in Proposition
1.2.

Proposition XV.2.2. (a) A quasi-bialgebra (A,~, E, <I>, l, r) is braided if

and only if there exists an invertible element R in A0A, called the universal
R-matrix, such that for all a E A we have

~OP(a) = R~(a)R-1, (2.1)

(id 0 ~)(R) = (<I>231)-1 R13<I>213R12(<I>123)-1, (2.2)

and
(2.3)
(b) Moreover, the tensor category A-Mod is symmetric if and only if
Relations (2.1-2.3) are satisfied together with the additional relation

(2.4)

As in Section 1, we shall consider R as part of the data of a braided

quasi-bialgebra and write (A,~, E, <I>, l, r, R).
PROOF. We proceed as in the proofs of Proposition 1.2 and Proposition
XIII. 1.4. First, given a braided quasi-bialgebra A with a universal R-matrix
R, we define a braiding on the tensor category A-Mod by

cv,W(v 0 w) = TV,W (R(v 0 w)) (2.5)

372 Chapter XV. Quasi-Bialgebras

where v and w belong to the A-modules V and W respectively. As in the

proof of Proposition VIII.3.1, Relation (2.1) implies that cvw is A-linear
whereas Relations (2.2-2.3) imply the Hexagon Axiom (XIII. 1.3-1.4).
Conversely, if A-Mod is braided with braiding c, set

(2.6)

The naturality of the braiding implies that for any pair V, W of A-modules,
the braiding Cv w is of the form (2.5). As a consequence of the A-linearity
of cA A' we get 6. 0P (a)R = R6.(a) for all a E A, which is equivalent to
Relation (2.1). The commutativity of the hexagons (HI) and (H2) in XIII. 1
implies Relations (2.2) and (2.3), as follows from an easy computation
using (2.5).
By (XIIL1.13) the category A-Mod is symmetric if cw,vcv,w = idvw
for all V and W. Now

Therefore, the category is symmetric if and only if R21 R I, which is

equivalent to R21 being the inverse of R. D

Corollary XV.2.3. In a braided quasi-bialgebra, the universal R-matrix

satisfies the relation

PROOF. This counterpart of Theorem VIIL2.4 (a) follows from (1.9), (2.5),
and from Theorem XIIL1.3. D

Later we shall need the following definition: a morphism of braided quasi-

bialgebras ex : (A,6.,c,c'f>,l,r,R) --+ (A',6.',c',c'f>',l',r',R') is a morphism
of the underlying quasi-bialgebras such that

(ex ex)(R) = R'. (2.7)

XV.3 Gauge Transformations

For simplicity, all quasi- bialgebras (A, 6., c, c'f>, l, r) considered in the sequel
will verify l = r = 1. In other words, the unit constraints of A-Mod will be
the same as the unit constraints of Vect( k). We shall henceforth drop any
reference to land r.
The purpose of this section is to introduce an equivalence relation on
quasi-bialgebras such that the categories of modules of two equivalent
quasi-bialgebras are tensor equivalent.
 XV.3 Gauge Transformations 373

Definition XV.3.1. Let A = (A,,6.,c,<ll) be a quasi-bialgebra. A gauge

transformation on A is an invertible element F of A such that

(c id)(F) = (id c)(F) = 1. (3.1)

Using a gauge transformation F on A, we can build a new quasi-bialgebra

A F as follows. Define an algebra morphism ,6. F
: A -+ A A by

(3.2)

for any a E A, and a new Drinfeld associator <ll F by

Proposition XV.3.2. For any quasi-bialgebra A = (A,,6.,c,<ll) and any

gauge transformation F E A A on A, the algebra A F = (A, ,6. F, c, <ll F) is
a quasi-bialgebra.

Observe that that if A happens to be a bialgebra (i.e., with <ll = 1),

then AF is not in general a bialgebra. This procedure provides non-trivial
examples of quasi-bialgebras.
PROOF. We must check Relations (1.1-1.4) for A F .

Relation (1.1): We have

(id ,6.F) (,6.F (a))<llF

F23 (id ,6.) (F,6.( a )F- 1 )F2-:/ F23 (id ,6.) (F) <ll(,6. id)(F- 1 ) Fi;l
F 2:,(id ,6.)(F)(id ,6.)(,6.(a))<ll(,6. id)(F-l)F23 1
F 23 (id ,6.) (F)<ll(,6. id)(,6.(a))(,6. id)(F-l)Fi;l
F23 (id ,6.) (F) <ll (,6. id) (F- 1 )Fi~/ Fd,6. id) (F,6.( a )F- 1 ) Fi;l
<ll F (,6. F id) (,6.F (a)).

The first and last equalities follow by definition, the third one from Relation
(1.1).
Relation (1.2): For all a E A we have

in view of the counit axiom and of Relation (3.1). We similarly obtain

(id C),6.F = idA-
Relation (1.3): We have to verify the pentagonal relation

(idid,6.F)(<llF) (,6.F idid) (<llF) = (<llF)234 (id,6.Fid) (<llF) (<llF) 123'

(3.4)
374 Chapter XV. Quasi-Bialgebras

Now,

(id Q9 id Q9 ~F)(<I>F) (~F Q9 id Q9 id)(<I>F)

F 34 (id Q9 id Q9~) (F23 (id Q9 ~)(F)<I>(~ Q9 id)(F-l)Fi~nF341
F12(~ Q9 id Q9 id) (F23 (id Q9 ~)(F)<I>(~ Q9 id)(F- 1)Fi:z1 ) Fi;l
F 34 (id Q9 id Q9 ~)(F23)(id Q9 id Q9 ~) ((id Q9 ~)(F)) (id Q9 id Q9 ~)(<I
(~Q9 ~)(F-l)F121 Fi4 1F12F34(~ Q9 ~)(F)
(~ Q9 id Q9 id)(<1(~ Q9 id Q9 id) ((~ Q9 id)(F- 1))
(~Q9 id Q9 id)(F121)FiZ1
F34 (id Q9 id Q9 ~)(F23)(id Q9 id Q9 ~) ((id Q9 ~) (F)) (id Q9 id Q9 ~) (<I

(~Q9 id Q9 id)( <I(~ Q9 id Q9 id) ((~ Q9 id)(F- 1))

(~Q9 id Q9 id)(F121 )F121
F 34 (id Q9 id Q9 ~)(F23)<1>234(id Q9 ~ Q9 id) ((id Q9 ~)(F))
<l>23~(id Q9 id Q9 ~)(<I(~ Q9 id Q9 id)(<I<I>lA
(id Q9 ~ Q9 id) ((~ Q9 id)(F- 1)) <1>123 (~ Q9 id Q9 id) (F121 )F121

F 34 (id Q9 id Q9 ~)(F23)<1>234(id Q9 ~ Q9 id) ((id Q9 ~)(F))

(id Q9 ~ Q9 id) (<I(id Q9 ~ Q9 id) ((~ Q9 id)(F- 1))

<l>123(~ Q9 id Q9 id)(F121)F121
F 34 (id Q9 id Q9 ~)(F23)<1>234(id Q9 ~ Q9 id)(F231)F23 1
F 23 (id Q9 ~ Q9 id) (F23 (id Q9 ~)(F)<I>(~ Q9 id)(F- 1)F121) F23 1
F 23 (id Q9 ~ Q9 id)(Fd<l>123(~ Q9 id Q9 id)(F121 )F121
(<I>Fh34 (id Q9 ~F Q9 id)(<I>F) (<I>F)123'
which proves (3.4). The first and last equalities follow from (3.2-3.3), the
second and sixth ones from the fact that ~ is an algebra morphism, the
third one holds because F12 and F34 commute, the fourth one follows by
applying (1.1) to a = F and F-l, and the fifth one from (1.3).
Relation (1.4): Using the definition of <I> F and Relations (1.4) and (3.1), we
immediately get (id Q9 c Q9 id)(<I>F) = FF- 1 = 1 Q91. 0
When F is an gauge transformation on A, then so is F- 1 and we have

(3.5)
If F' is another gauge transformation, then so is the product F F' and we
 XV.3 Gauge Transformations 375

have
(3.6)
Definition XV.3.3. Two quasi-bialgebras (A,.6, 10, <1 and (A', .6 / , 10 ' , <1>/)
are equivalent if there exist a gauge transformation F on A' and an iso-
morphism a : A ---) A~ of quasi-bialgebras.

Relations (3.5-3.6) imply that this is an equivalence relation. We now

prove that equivalent quasi-bialgebras have equivalent tensor categories of
modules. We start with a preliminary result. Let A = (A,.6, 10, <1 be a
quasi-bialgebra and F E A 129 A be a gauge transformation. Define

iff (V, W)(v 129 w) = F-1(v 129 w) (3.7)

where v and w belong to the A-modules V and W respectively.

Lemma XV.3.4. Under the previous hypothesis, the triple (id, id, iff) is
a tensor functor from the tensor category A-Mod to the tensor category
ArMod.
PROOF. Recall Definition XI.4.1. We have to check Relations (XI.4.1-4.3),
namely if2(k, V) = if2(V, k) = id v and

if2(U, VI29W) (id u l29if2(V, W)) a{;,v,w = au,v,w if2(UI29V, W)(if2(U, V)l29id w )
(3.8)
where a F is the associativity constraint induced by <l>F. The first set of
equalities follows from (3.1) and (3.7). Let us prove (3.8). For all u E U,
v E V and w E W we have

(if2(U, V 129 W)(id u 129 if2(V, W)) a{;,v,w) (u 129 v 129 w)

(id 129 .6) (F- 1)F231<l>F(u 129 v 129 w)
<1>(.6129 id)(F-l)F121(u 129 v 129 w)

(au,v,w if2(U 129 V, W)(if2(U, V) 129 idw))(u 129 v 129 w).

The first and last equalities follow from (1.5) and (3.7), and the second one
from the definition of <I> F. 0

We state the first main result of this section. Let A and A' be equivalent
quasi-bialgebras with a gauge transformation F on A' and an isomorphism
a : A ---) A~ of quasi-bialgebras. The map a induces a strict tensor functor
(a*,id,id) from A~-Mod to A-Mod as explained in Example 2 of XI.4.
Since a is an isomorphism, a* is a tensor equival~nce.

Theorem XV.3.5. The tensor functor (a* , id, iff) is a tensor equivalence
between A'-Mod and A-Mod.
376 Chapter XV. Quasi-Bialgebras

PROOF. Replacing F by F- 1 which is another gauge transformation, we

get a tensor functor (id,id,cpf- 1 ) from A'p.-Mod to A'-Mod which turns
out to be an inverse to (id, id, cpf). The tensor functor (a*, id, cpf) is the
composition of the tensor equivalence (id, id, cpf) : A'-Mod ---t A'p.-Mod
and of the tensor equivalence (a*, id, id). D

We now extend the gauge transformations to braided quasi-bialgebras.

Consider a braided quasi-bialgebra (A,~, c, <P, R) with a universal R-matrix
R. For any gauge transformation F on A, define ~ F and <p F as above. Also
set
(3.9)
Proposition XV.3.6. The algebra AF = (A'~F,C,<PF,RF) is a braided
quasi-bialgebra.

PROOF. One may check Relations (2.1-2.3) directly for R F . Alternatively,

one may also proceed as follows. Let c be the braiding of A-Mod corre-
sponding to the universal R-matrix R. Define c~w : V W ---t W V by
c~w(v w) = Tvw(RF(V w)). An immediat~ computation using (3.9)
sh~ws that '
c~,w = (cpf(V, W))-l 0 cv,w 0 cpf(V, W).

One then checks that cF is a braiding on ArMod as in the proof of Lemma

XIII.3.2. Finally, apply Proposition 2.2. D

Let (A,~,c,<P,R) again be a braided quasi-bialgebra and F be a gauge

transformation on A.

Lemma XV .3.7. Under this hypothesis, the tensor functor (id, id, cpf) is
a braided tensor equivalence from A-Mod to ArMod.

PROOF. In view of Theorem 3.5, it is enough to show that (id, id, cpf) is
braided in the sense of Definition XIII.3.6. We must check that we have
cpf oc{;,v = cu,v ocpf. The latter is equivalent to F-1(RFb = (RF-1b,
which follows from (3.9). D

We adapt Definition 3.3 to braided quasi-bialgebras.

Definition XV.3.8. The two braided quasi-bialgebras (A,~, c, <P, R) and

(A', ~I, c' , <P', R') are equivalent if there exist a gauge transformation F on
A' and an isomorphism a : A ---t A'p. of braided quasi-bialgebras.

Suppose we are in the situation of equivalent braided quasi-bialgebras A

and A' with F and a as in the previous definition.' As a consequence of
Theorem 3.5 and Lemma 3.7, we get the second main result of the section.

Theorem XV.3.9. In the situation just considered, the tensor functor

(a* , id, cpf) is a braided tensor equivalence between the braided tensor cat-
egories A' -Mod and A-Mod.
 XV.4 Braid Group Representations 377

XV.4 Braid Group Representations

Let (A, b., 10, <P, R) be a braided quasi-bialgebra with trivial unit constraints,
V a left A-module, and n an integer> 1. We recall from Chapters X and
XIII how to define a representation of the braid group Bn on V0 n . Since <P
is not necessarily trivial, we have to make precise what we mean by V0 n .
Let us place ourselves in the general situation where we have a braided
tensor category C with associativity constraint a and braiding c. In XI.5 we
c
constructed a strict tensor category str which is tensor equivalent to C. The
c
tensor product of str is denoted *. By definition, we set V0 n = v*n, which
means that V0 n is equipped with the unique system of parentheses opening
only at the extreme left. For instance, we have V0 4 = ((V V) V) V.
Define automorphisms c1 , ... ,cn - 1 of V0 n = v*n in str by c
Ci = idv*(i-l) * cv,v * idv*(n-i-l).
By Theorem XIII.1.3 and Corollary X.6.9 we know that there exists a
unique morphism of groups p~ : Bn ---7 Aut(V0n ) sending the generator (Ji
of Bn to ci for any i = 1, ... , n - 1. The representation p~ will be called
the braid group representation associated to the braided tensor category C.
We now make ci explicit in terms of the original category C.
Lemma XV.4.1. We have
ci = (a~~(i_1),v,v id~(n-i-l) (id V0 (i-l) cv,v id~(n-i-l)
(aV 0 (i-l) ,v,v id~(n-i-l).

PROOF. This follows from Relation (XI.5.4) which expresses the tensor
c
product of morphisms in the strict category str in terms of the tensor
product of morphisms and of the associativity constraint in C. We also use
Relation (XI.5.3) in the following special cases:

'P(S, (V)) = id F (S)0V and 'P(S, (V, V)) = a-;;(S),v,v

D

When C = A-Mod is the braided category of left modules over the braided
quasi-bialgebra A, the braiding C is given by

CVV(vl
,
V2) = (R(VI v2)) 21 .

Consequently, by Lemma 4.1 we have

C1(V1v n )= (R 12 (V 1 V n ))21 (4.1)

and if i > 1
Ci( VI ... Vn ) = <Pi 1((Ri,Hl <Pi)(V 1 ... Vn )) Hl,i (4.2)
378 Chapter XV. Quasi-Bialgebras

where we used the subscript convention of VIII.2 and where

<Pi = ~(Hl)(<p) Q91 0 (n-i-l)

is the invertible element of A0 n expressed in terms of the map ~(Hl) :

A03 - t A0(Hl) defined inductively by ~(3) = id A 3 and by the relations
~ (HI) = (~Q9id~(i-l))~ (i). The corresponding representation P~ = p~-Mod
will be called the braid group representation associated to the braided quasi-
bialgebra A.
Let A' = (A',~', s', <P', R') be another braided bialgebra. We assume that
A and A' are equivalent braided quasi-bialgebras in the sense of Definition
3.8, i.e., there exist a gauge transformation F on A' and an isomorphism
a : A - t (A') F of braided quasi-bialgebras. In particular, we have

(4.3)

Let V be a left A'-module. By a it becomes an A-module. For any integer

n> 1 we have two braid group representations p~,p~' : En - t Aut(V0 n )
associated to A and A' respectively and acting on the same space. The main
result of this section asserts that these representations are equivalent.

Theorem XVA.2. Let A and A' be equivalent braided quasi-bialgebras.

With the previous notation, we have
A' -1 A
Pn (g)(w) = F12 Pn (g)(FI2 w)

for all g E En and wE V0 n .

PROOF. By Theorem 3.9, we know that both braid group representations
are equivalent. It is therefore enough to compute the equivalence on one
special element of the braid group. We choose the generator 0'1' By (4.1)
and (4.3) we have for vl"",vn E V

(R~2(VI Q9 ... Q9 vn )) 21
((F21 1(a Q9 a)(R)Fd(Vl Q9 ... Q9 Vn )) 21

F1-:/((aQ9a)(R)(FI2 (V 1 Q9 00. Q9V n )))21

Fl:/P~(0'1)(FI2(VI Q9 00. Q9V n )).

D

It should be clear that the statement of Theorem 4.2 depends on the way
we put parentheses on V0 n . Other systems of parenthesizing give rise to
different, but equivalent braid group representations.
 XV.5 Quasi-Hopf Algebras 379

XV.5 Quasi-Hopf Algebras

For the sake of completeness we introduce quasi-Hopf algebras as defined
by Drinfeld [Dri89b]. As above, all quasi-bialgebras considered here have
trivial unit constraints, i.e., l = r = 1.
DefinitionXV.5.1. A quasi-bialgebra (A,~,c,<1 is a quasi-Hopfalge-
bra if there exist an invertible anti-automorphism S of the algebra A and
elements a and f3 of A such that for all elements a in A we have

L S(a')aa" = c(a)a, L a'f3S(a") = c(a)f3, (5.1)

(a) (a)

and

where <1> = Li Xi Q9 ~ Q9 Zi and <1>-1 = Li Xi Q9 Q9 Yi Zi' A quasi-Hopf

algebra is braided if the underlying quasi-bialgebra is.

We shall write (A,~, c, <1>, S, a, (3) to express the complete data of a quasi-
Hopf algebra. As in XIV.2 consider the category A-Modf of left A-modules
that are finite-dimensional vector spaces over the ground field k. Equip it
with the tensor category structure induced by ~ and <1>. For any object V of
A-Mod f consider the objects V* and *V as defined in Example 1 ofXIV.2.
We define maps bv : k ----t V Q9 V*, d v : V* Q9 V ----t k, b;,r : k ----t *V Q9 V, and
d;,r : V Q9 *V ----t k by

where {vJ i is a basis of V and {Vi} i the corresponding dual basis.

Proposition XV.5.2. The maps bv , d v , b;,r and d;,r are A-linear and the
composite maps

V ~ k Q9 V bvid)(V Q9 V*) Q9 V~V Q9 (V* Q9 V) iddv )V Q9 k ~ V

V* ~ V* Q9 k idbv )V* (V Q9 V*)~(V* Q9 V) Q9 V* dvid )k Q9 V* ~ V*

'd b' 1 d' 'd
V ~ V Q9 k ! v) V Q9 (*V Q9 V)~(V Q9 *V) Q9 V
V ! )k Q9 V ~ V
b' 'd 'dd'
*V~kQ9*V v! )(*VQ9V)Q9*V~*VQ9(VQ9*V)! v)*Vk~*V
are all identities.
380 Chapter XV. Quasi-Bialgebras

PROOF. The first statement follows from Relations (5.1), and the second
one from Relations (5.2). 0

Consider the braided tensor category A-Mod f associated to the quasi-

Hopf algebra A as well as the strict braided tensor category (A-Mod f ytr.
Proposition 5.2 can be interpreted as follows: the strict braided tensor
category (A-Mod f )str is autonomous, i.e., has left and right dualities given
by the maps bv , d v , b~, and d~.
We end this section with an example of a non-trivial braided quasi-Hopf
algebra close to the quantum double of the algebra k[G] of a finite group
G. Suppose given a normalized 3-cocycle on the group G, i.e., a function
w : G x G x G --+ k \ {O} such that

w(x, y, z)w(tx, y, Z)-lW(t, xy, z)w(t, x, YZ)-lW(t, x, y) =1 (5.3)

for all t, x, y, z E G, and such that w(x, y, z) = 1 whenever x, y, or

z = 1. Consider a finite-dimensional vector space D W (G) with a basis
{egx}(9,X)EGXG indexed by G x G. Define a product on DW(G) by

(5.4)

where 8(g, x, y) = w(g, x, y)w(x, y, (xy)-lgxy)W(x, x-1gx, y)-l. It is easy to

check that this product is associative and has the element 1 = L9EG eg1
as a left and right unit. Observe that when the co cycle w is trivial, i.e.,
w (x, y, z) = 1 for all x, y, z, then D W (G) is isomorphic to the quantum
double D(k[G]) (see IX.4.3). In contrast to the trivial case, the map sending
x to L9EG egx is not a morphism of algebras from k[G] to DW(G) in
general, but the map eg f-+ eg1 is, which will allow us to identify eg1
with ego
We define morphisms of algebras bo : D W(G) --+ D W(G) D W(G) and
c: DW(G) --+ k by

bo(egx) = L 'Y(x, u, v) eux evx and c(egx) = 8g,1 (5.5)

uv=g

where 'Y(x,u,v) = w(u,v,x)w(x,x- 1UX,X- 1 vx)w(u,x,x- 1VX)-1. Set also

X,Y,zEG gEG h

(X = 1, and (J = L9EG w(g,g-l,g)e g. We define an anti-automorphism S

of the algebra D W (G) by

S( egx ) = 8( 9 -1 ,x,x -1)-1 'Y ( x,g,g -1)-1 ex-'gx x -1 . (5.7)

Then (DW( G), bo, c, CP, S, (X, (J) is a braided quasi-Hopf algebra with univer-
sal R-matrix R in the sense of Definitions 1.1, 2.1 and 5.1.
 XV.7 Notes 381

XV.6 Exercises
1. Let (A,~, c, S) be a Hopf algebra and F = 2:i fi gi E A A be a
gauge transformation such that

Consider the element x = 2:i fiS(gi) of A. Show that it is invertible

and that (A'~F,C,SF) is a Hopf algebra where SF(a) = xS(a)x-1
for all a E A.

2. Show that if (A, ~,c, <I is a quasi-bialgebra each of (AOP,~,c,<I>-1),

(A, ~ oP, C, (<I>321)-1) and (AOP, ~ oP, c, <I>321) is a quasi-bialgebra.
3. Let A = (A,~,c,<I>,l,r) and A' = (A',~',c',<I>',l',r') be quasi-bial-
gebras. Let a : A -> A' be a morphism between the underlying al-
gebras. Suppose that the induced functor a* : A'-Mod -> A-Mod
extends to a tensor functor (a*, id, 'P2)' Show that there exists an
invertible element F in A' A' such that 'P2(U v) = F- 1(u v).
Prove that necessarilY'ca = c',

(aa)~(a)F = F~'(a(a)), <I>' F12(~id)(F) = F23(id~)(F)a(<I,

l' = a(l)(c' id)(F), and r' = a(r)(id c')(F).

4. (Gauge transformation of a quasi-Hopf algebra) Let (A,~, c, <I> , S, a, (3)

be a quasi-Hopf algebra and F = 2: i figi be a gauge transformation
on A with inverse F- 1 = 2:i h gi' Set

Prove that (A'~F,c,<I>F,S,aF,{3F) is a quasi-Hopf algebra.

5. Let (A, ~, c, <I>, S, a, (3) be a braided quasi-Hopf algebra with universal
R-matnx R. Suppose that <I> = 2: i XiYiZi and R = 2: j Sjt j .
. -1 - - -

Set
u= L S(i{3S(Zi))S(t j )as j Xi
i,j

Prove that u is an invertible element in A such that c( u) = 1 and

S2(a) = uau- 1 for all a E A.

XV.7 Notes
Quasi-bialgebras, quasi-Hopf algebras, and gauge transformations were in-
vented by Drinfeld [Dri89b][Dri90][Dri89c] in relation with his treatment
382 Chapter XV. Quasi-Bialgebras

of the monodromy of the Knizhnik-Zamolodchikov equations (to be con-

sidered in Chapter XIX). Drinfeld used the term "quasi-triangular quasi-
bialgebra" for a braided quasi-bialgebra. In [DriS9b], Section 1, Drinfeld
showed that one always could reduce a general quasi-Hopf algebra to a
quasi-bialgebra with I = r = l.
Altschuler and Coste [AC92] proved (see Exercise 5) that in any braided
quasi-Hopf algebra the square of the antipode is an inner automorphism
(just as for braided Hopf algebras, c.f. VIlI.4). They also defined ribbon
quasi-Hopf algebras generalizing the ribbon algebras of XIV.6. The braided
quasi-Hopf algebra DW(G) of Section 5 is due to Dijkgraaf, Pasquier, and
Roche [DPR90] and was shown in [AC92] to be a ribbon quasi-Hopf algebra.
Let us remark that when the 3-cocycle w is changed by a coboundary, then
D W (G) is changed by a gauge transformation, so that the tensor category
of modules of DW(G) depends on the cohomology class of w.
Exercise 1 is taken from [Res90] while Exercise 4 is from [DriS9b].
 Part Four

Quantum Groups
and Monodromy
Chapter XVI
Generalities on Quantum
Enveloping Algebras

In order to state the main results of Part IV, we need the concept of a
quantum enveloping algebra. This requires the use of formal series and of
h-adic topology. The chapter is completed by an appendix on inverse limits.

XVI. 1 The Ring of Formal Series and h- Adic

Topology
Consider the complex algebra K = C[[h]] of complex formal series in one
variable h. Any element of K is of the form

(1.1)

where (a o, aI' ... ) is a family of complex numbers indexed by the set N

of non-negative integers. If f' = 2:n>O a~hn is another formal series, then
the sum 1+ f' and the product I f'-of I and f' in K are given by

1+ f' = 2: (an + a~)hn and If' = 2: (2: apa~)hn. (1.2)

n~O n~O p+q=n

Any polynomial in h may be considered as an element of K. In particular,

the constant polynomial 1 is an element of K where it acts as a unit for
the product as can easily be seen from (1.2). The following characterizes
invertible elements in K.
386 Chapter XVI. Generalities on Quantum Enveloping Algebras

Lemma XVI.1.1. A formal series f = Ln>o anh n is invertible in C[[h]]

if and only if ao =I- 0 in C. -

PROOF. The formal series f is invertible if and only if there exists another
series g = Ln>o bnhn such that fg = 1. From (1.1) we see that this
is equivalent to -the existence of an infinite family (b o, bi , ... ) of complex
numbers such that aobo = 1 and

(1.3)

for all n > O. The relation aobo = 1 shows that the invertibility of ao is a
necessary condition for f to be invertible. This condition is also sufficient
since the family (bo,b i , ... ) can be determined inductively from bo = a 1 o
and Relations (1.3). 0

Lemma 1.1 may be interpreted as saying that the ring K is a local ring
whose maximal ideal is the ideal (h) generated by h.
For any integer n > 0 consider the algebra Kn = C[h]/(hn) of truncated
polynomials obtained as the quotient of the algebra of complex polynomi-
als in one variable by the ideal generated by h n. There is a morphism of
algebras 7rn from K to Kn sending a formal series f = Ln>o anh n to the
class of L~:~ akhk modulo (hn). This map is surjective a~d its kernel is
the ideal h n K generated by h n in the ring of formal series. Consequently,
7rn induces an isomorphism of algebras

(1.4)

For n > 0 there is also a surjective morphism of algebras Pn from Kn to

K n- i induced by the inclusion of ideals (hn) c (h n- i ). Consider the inverse
system of algebras (Kn'Pn)n and its inverse limit liI!l Kn as defined in the
n
Appendix. We have Pn 07rn = 7rn - i for all n. It follows from Proposition 9.1
that there exists a unique morphism of algebras 7r from K to l@ Kn whose
n
composition with the projection of the inverse limit onto Kn equals 7r n'
Proposition XVI.1.2. The map 7r : C[[h]] --4 l@ C[h]/(hn) is an iso-
n
morphism of algebras.
PROOF. The map 7r is injective since its kernel, which is the intersection of
all ideals (h n ), is zero in view of (1.2).
In order to prove the surjectivity of 7r we construct a right inverse to it as
follows. Let (fn)n>O be an element of the inverse limit (see the Appendix for
a definition). By definition fn belongs to K n , which allows us to represent
it as
fn =" a(n)hk
n-i

~ k '
k=O
 XVI.1 The Ring of Formal Series and h-Adic Topology 387

and we have Pn(Jn) = fn-l for all n > O. Hence a~n) = a~n-l) for k running
from 0 to n - 2. We can therefore define a formal series f = Ln>o anhn
by an = a~n+2) = a~n+3) = .... We have rr(J) = (In)n' - 0

Proposition 1.2 allows us to equip K with the inverse limit topology

described in the Appendix. This topology is called the h-adic topology.
By definition, the family of subsets rr;;l(Un ), where n > 0 and Un is any
subset of K n , is a basis of open sets of K. Since {O} is a family of open
neighbourhoods in the discrete set K n , the family rr;;l(O) = (hn) is a family
of open neighbourhoods of 0 in K for the h-adic topology. From this it is
easy to see that the h-adic is a metric topology where the metric may be
defined as follows. For any non-zero formal series f = Ln>o anh n , let w(J)
be the unique non-negative integer such that aw(f) -=I- 0 and ak = 0 for all
k < w(J). When f = 0, set w(O) = +00. Extend the natural order of N to
N U {+oo} by requiring that +00 > n for all n E N. We see that

(1.5)

As a consequence, we get
(1.6)

a trivial fact already used in the proof of Proposition 1.2. We also have

w(J + g)::::: min(w(J),w(g)) (1.7)

for all f, 9 E K. Define a map I I from K to the set of non-negative real
numbers by
If I = Tw(f) (1.8)
if f -=I- 0 and 101 = 0 if f = O. The next result is an immediate consequence
of the previous considerations.
Lemma XVI.1.3. For all f and 9 in K we have
Ifl = 0 {==? f = 0, 1- fl = If I, If + gl :::; max(lfl, Igl)
As a corollary we get a distance on K.
Corollary XVI.1.4. Define d(J,g) = If - gl for any f,g E C[[h]]. Then
d is an ultrametric distance on C[[h]J, i.e., we have
(i) d(J,g) = 0 {==? f = g,
(ii) d(J, g) = d(g, j), and
(iii) d(J,h) :::; max(d(J,g),d(g,h)) for any triple (J,g,h) of formal se-
ries.
The distance d puts a metric on C[[h]]. From its definition and from (1.5)
it is clear that the family of ideals (hn) is also a set of open neighbourhoods
of 0 for the metric topology. Therefore the latter is equivalent to the h-adic
topology.
388 Chapter XVI. Generalities on Quantum Enveloping Algebras

XVI. 2 Topologically Free Modules

Let M be a left module over the algebra K = C[[h]]. Consider the family
(hn M)n>O of submodules and the canonical K-linear projections

They form an inverse system of K-modules, and we may consider the in-
verse limit
(2.1)

which has a natural structure as a K-module. The inverse limit 1\/[ has a
natural topology, the inverse limit topology, for which it is easy to see as
in Section 1 that the family of submodules (h n l'vf)n is a family of open
neighbourhoods. The module M is called the h-adic completion of .M.
The projections in : IV! --+ Mn induce a unique K-linear map i : AI --+ /1.1
such that TIn 0 i = in for all n. The kernel of i is given by

Ker (i) = n h M.
n>O
n

Definition XVI.2.1. A K-module M is separated if nn>O hnM = {a}.

It is complete if the map i is surjective.

For any ~dule M the module M/(nn>O hnM) is separated and the
completion M is complete. Indeed, consider the projection TIn : IV[ --+ IV!n'
Its kernel is hnAI, which implies the isomorphism of modules

(2.2)

Taking inverse limits, we get IV! = /1.1, which proves that IV! is complete.
Any separated, complete K-module will be equipped with the topologL
called the h-adic topology,--soming from the inverse limit topology on IV!
via the isomorphism M ~ M.
We now describe an important class of separated, complete K-modules.
It includes K itself, viewed as a K-module by left multiplication. Take any
complex vector space V. Define V[[h]] as the set of all formal series

(2.3)

where (vO,v 1, ... ) is an infinite family of elements of V. Using Formulas

(1.2) we can put a left K-module structure on V[[h]]. Any left K-module
of this form will be called a topologically free module. We recover K by
taking V = C.
 XVI.2 Topologically Free Modules 3S9

Proposition XVI.2.2. Any topologically free module is separated and com-

plete.
PROOF. From the definition we see that the submodule hnV[[h]] is the set
of all elements Ln>o vnhn such that Vo = ... = vn- 1 = 0. It follows that
the intersection of all submodules hnV[[h]] is zero. This implies that V[[hll
is separated.
It is also complete: a proof parallel to the proof of Proposition 1.2 shows
V[[hll isomorphic to the inverse limit of the family (V[[h]lIhnV[[h]]) .
n>O
D

As in the case V = C, the h-adic topology on V[[h]] induced by the

inverse limit topology can be defined by a metric built in the same way as
for C[[h]] (see Section 1).
A topologically free module V[[hll has the following additional properties.
Proposition XVI.2.3. (a) Let {ei}iEI be a basis of the vector space V.
Then the K -submodule generated by the set {eJ iEI is dense in V[[hll for
the h-adic topology.
(b) For any separated, complete K -module N, there is a natural bijection
HomK(V[[h]J, N) ~ Hom(V, N)
where HomK denotes the space of K -linear maps.
Observe that a K-linear map f : M --> N between separated, complete
K-modules is always continuous for the h-adic topology since f(h n M) is
contained in h n N by K-linearity.
PROOF. (a) Let W be the submodule of V[[h]] generated by the set {eJiEI'
Take any element f = Ln>o vnhn of V[[h]]. We have to show that for any
integer n > 0, there exists an element fn E W such that f - fn belongs
to hnV[[h]]. The element fn is constructed as follows: fn = L~:~ vkhk.
Clearly, the difference f - fn lies in hnV[[h]]. It remains to check that fn
belongs to W. Indeed, since {eiLEI is a basis we have
n-l n-l
fn = 2: (2: A~k)ei)hk = 2: (2: A~k)hk)ei E W
k=O iEI iEI k=O
where A~k) is a family of complex numbers, all but finitely many equal to
zero.
(b) Let f be a continuous K-linear map from V[[hll to N. Considering
V as the space of constarit formal series in V[[h]J, we may restrict f to a
C-linear map from V to N. Conversely, let g be a K-linear map from V to
N. Extend it to a Kn-linear map gn from V[[hll/hnV[[hll to N/h n N by
n-l n-l

gn(2: vkhk) = 2: g(vk)h k mod hnN.

k=O k=O
390 Chapter XVI. Generalities on Quantum Enveloping Algebras

Taking inverse limits yields a K-linear map goo between the corresponding
inverse limits. Since V[[h] and N are separated and complete, we get a
map, still denoted goo' from V[[h]] to N. This map restricts to g on V. 0
Topologically free modules can be characterized in a simple way. Recall
that a K-module M is torsion free if hm i- 0 when m is any non-zero
element of M.

Proposition XVI.2.4. A left K -module is topologically free if and only if

it is separated, complete, and torsion-free.

PROOF. By Proposition 2.2 we know that any topologically free module is

separated and complete. It has no torsion in view of (1.2).
Conversely, let M be a separated, complete, and torsion-free module. We
have to show that M is of the form V[[h]]. Choose a vector subspace of M
which is supplementary to hM. Because of the torsion-free assumption, we
get h n M = hnV EEl h n+ 1 A1 for all n 2:: O. Hence,

Taking inverse limits and using the fact that M and V[[h]] are separated
and complete, we get

n n

o
We end with a caveat. We have V[[h]] ~ V C[[h]] only if V is a finite-
dimensional vector space. There is no such isomorphism when V is infinite-
dimensional, in which case V[[h]] is strictly bigger than V C[[h]]. Indeed,
take an infinite family (en)nEN of linearly independent vectors; then the
element Ln::::o enh n of V[[h]] does not belong to V C[[h]].

XVI. 3 Topological Tensor Product

Let M and N be left modules over the algebra K = C[[h]]. Consider the
K - module M K N obtained as the quotient of the vector space M N by
the subspace spanned by all elements of the form fm n - m fn where
f belongs to K, m to M, and n to N.
Definition XVI.3.1. The topological tensor product M0N of M and N
is the h-adic completion of M K N:

M0N = (M K N) = 1,Ll!l (M K N)jhn(M K N).

n>O
 XVI.3 Topological Tensor Product 391

Since it is defined as a completion, the topological tensor product of two

modules is always complete. Given m E M and n E N we denote by mn
the image of m Q9 n under the natural maps M Q9 N ----> M Q9 K N ----> M N.
The subspace of the topological tensor product spanned by all elements
of this form is dense in MN. The usual associativity and commutativity
constraints induce the following K-linear isomorphisms

(MN)P ~ M(NP), (3.1)

MN~NM. (3.2)
We also have
KM~M~MK, (3.3)
which means that K serves as a unit for completions.
The topological tensor product is functorial as can be seen from the
definition: if f : M ----> M' and 9 : N ----> N' are K-linear maps, then there
exists a K-linear map

fg : MN ----> M'N'

enjoying the formal properties of the algebraic tensor product.

Proposition XVI.3.2. If M and N are topologically free modules, then

so is MN. More precisely, we have

V[[h]]W[[h]] = (V Q9 W)[[h]].

PROOF. For any K-module M, the natural maps

are isomorphisms, where the first one is induced by (1.4) and the second
one is given by mQ9f f--* fm (the inverse map being induced by m f--* m(91).
Applying this to M Q9 K N where M = V[[h]] and N = W[[h]] we get

(M Q9K N)/hn(M Q9K N) c,; (M Q9K N) Q9K Kn

(MQ9KKn) Q9Kn (NQ9KKn)
c,; M/hnMQ9KnN/hnN

(VQ9Kn) Q9Kn (WQ9Kn)

c,; (V Q9 W) Q9 Kn
c,; (V Q9 W)[[hll/hn(V Q9 W)[[h]].

Passing to the inverse limit yields the desired result. o

392 Chapter XVI. Generalities on Quantum Enveloping Algebras

XVI. 4 Topological Algebras

We extend the definitions of algebras, quasi-bialgebras, etc. to the setting
of C[[h]]-modules. This is done by replacing the algebraic tensor product
of 11.1 by the topological tensor product of Section 3.
A topological algebra is a triple (A, /-l, 7]) where A is a module over the
ring K = C[[h]L /-l : A~A ----t A and 7] : K ----t A are K-linear maps such
that
(4.1)
and
/-l 0 (7]~idA) = idA = /-l 0 (idA~7]) (4.2)
As in the algebraic case, we use the convention
aa' = /-l(a~a') (4.3)
for the product of two elements a, a' of a topological algebra (A, /-l, 7]). We
also write 1 for the image under 7] of the unit element 1 of K.
Let (A, /-l, 7]) be a topological algebra and f(h) = L:n>O cnh n be a formal
series with complex coefficients. For an element a E A,-the formula

f(ha) = L cnanh n (4.4)

n:2:0
defines a unique element in the inverse limit A= lim A/hn A. Therefore if
n
A is s~parated and complete, it defines an element, still denoted f(ha), in
A ~ A. This procedure can be applied to the classical exponential function
e h = L:n:2:o ~~, yielding elements of the form

(4.5)

in any separated complete topological algebra A. If a' is another element

in A commuting with a, then
(4.6)
As a consequence, we see that e ha is invertible in A with inverse equal
to e- ha .
A morphism f : (A, /-l, 7]) ----t (A', /-l', 7]') of topological algebras is a K-
linear map f : A ----t A' such that
fO/-l=/-l'o(f~f) and f07]=7]'. (4.7)
Using the convention of (4.3), Relations (4.7) can be rewritten as
(4.8)
where a 1 and a 2 are elements of A.
 XVI.4 Topological Algebras 393

Example 1. Let A = K = q[h]]. We identify K'0K with K. Then we see

(K, id K , id K ) is a topological algebra. Moreover, the map 7] : K --; A is a
morphism of topological algebras for any topological algebra (A, p" 7]).
Example 2. Let (A, p" 7]) and (A', p,', 7]') be topological algebras. Then so
is

where TA,A' : A'0A' --; A''0A is the flip. In other words, the product in the
tensor product algebra A'0A' is given by

(4.9)

and the unit is 1'01.

A topological quasi-bialgebra is a sextuple (A, p" 7], A, c, <I where (A, p" 7])
is a topological algebra, A : A --; A'0A and c : A --; K are K-linear maps,
and <I> is an invertible element of the tensor product algebra A'0A'0A such
that
(idA'0A)(A(a)) = <I> ((A'0idA)(A(a))) <I>-1 (4.10)

for all a E A,
(c'0idA)A = idA = (idA'0c)A, (4.11)
(idA'0idA'0A) (<I (A'0idA'0idA)(<I = <I>234 (idA'0A'0id A) (<I <I> 123 ,
(4.12)
and
(4.13)
When <I> = 1'01'01, we call A a topological bialgebra.
A morphism f : (A, p" 7], A, c, <I --; (A', p,', 7]', A', c', <I>') of topological
quasi-bialgebras is a morphism f between the underlying topological alge-
bras such that

(j'0f)A = A'f and (j'0f'0f)(<I = <I>'. (4.14)

A topological quasi-bialgebra (A, p" 7], A, c, <I is a topological braided

quasi-bialgebra if there exists an invertible element R of the tensor product
algebra A'0A such that

AOP(a) = RA(a)R-l, (4.15)

(idA'0A)(R) = (<I>231)-1 R13<I>213R12(<I>123)-1, (4.16)

and
(4.17)
As before, R is called the universal R-matrix of A. It is part of the data of
a topological braided quasi-bialgebra. A morphism of topological braided
394 Chapter XVI. Generalities on Quantum Enveloping Algebras

quasi-bialgebras is a morphism of the underlying topological quasi-bial-

gebras sending the universal R-matrix of the first one to the universal
R-matrix of the second one.
We also need the concept of a gauge transformation on a topological
quasi-bialgebra A: it is an invertible element F of A0A such that
(4.18)
From a topological (braided) quasi-bialgebra A and a gauge transformation
F one can form a new topological (braided) quasi-bialgebra AF by (XV.3.2-
3.3) and (XV.3.9).
Example 3. Let Ao = (Ao, /-Lo, 7)0'6. 0 , co' <Po, Ro) be a braided quasi-bial-
gebra over the field C of complex numbers. Using Proposition 2.3 (b) and
Proposition 3.2, one may define the topological braided quasi-bialgebra

on the space of formal series with coefficients in Ao where /-L, 7), 6. and c
are the unique K-linear maps such that 7)(J) = f7)o(l) = f1 for all f E K,
/-L(a0a') = ILo(a (>9 a'), 6.(a) = 6. o(a), c(a) = co(a)
for all a, a' E Ao. We call Ao[[h]] the trivial topological braided quasi-
bialgebra associated to Ao.
Example 4. Let A = (A, IL, 7),6., c, <P, R) be a topological braided quasi-
bialgebra. Since (A0A)/h(A0A) ~ A/hA (>9 A/hA, the K-linear maps
IL, 7), 6., c induce C-linear maps
fl : A/ hA (>9 A/ hA ---+ A/ hA, fl: C ---+ A/ hA,
LS. : A/hA ---+ A/hA (>9 A/hA, E: A/hA ---+ C.
Define <I? as the class of <P modulo (A/hA)3 and R as the class of R modulo
(A/hA)2. Then A = (A/hA, fl, fl, LS., E, <I?, R) is a braided quasi-bialgebra.
Another concept we need to adapt is the concept of a topological A-
module M over a topological algebra A = (A, /-L, 7)). It is a left K-module
with a K-linear map ILM : A0M ---+ M such that
ILM 0 (IL0id M ) = ILM 0 (id A 0ILM) and ILM 0 (7)0id M ) = id M . (4.19)
We shall write ILM(a0m) = am for a E A and m E M. The definition of a
morphism of topological A-modules is left to the reader.
Let M and N be topological A-modules. Then their topological tensor
product M0N is a topological A0A-module. If A has a comultiplication
6. : A ---+ A0A, we can pull back the A0A-module structure on M0N to
a topological A-module structure given by
a(m0n) = 6.(a)(m0n) (4.20)
for all a E A, m in M, and n in N.
 XVI.5 Quantum Enveloping Algebras 395

If A is a topological braided bialgebra with universal R-matrix R, then

for any topological A-module M the K-linear automorphism eft. M defined
as in VIII.3 by ,
(4.21)

is a solution of the Yang-Baxter equation in M0M0M.

Proceeding as in XV.4, we can show that the universal R-matrix of a
topological braided quasi-bialgebra gives rise ~o a representation of the
braid group Bn on the topological A-module Me)n where n is any integer>
1 and M is any topological A-module. Theorem XV.4.2 can be reformulated
in the present context.
Mimicking IX.5, we say that a topological A-module M over a topological
bialgebra A = (A, p" 'f/,~, c) with left action P,M : A0M -+ M is a topolog-
ical crossed A-bimodule if there exists a K-linear map ~M : M -+ M0A
such that

(4.22)

and

(P, M 0p,) (id A 0TA ,M0idA)(~0~M)

= (idM0P,)(~M0idA)TA,M(idA0P,M )(~0idM) (4.23)

XVI. 5 Quantum Enveloping Algebras

Let g be a complex Lie algebra. In V.2 we defined its enveloping algebra
U(g) and proved that it had a natural bialgebra structure determined by

~(x)=IQ9x+xQ91 and c(x)=O

for all x belonging to g. We equip it with a trivial braided quasi-bialgebra

structure with cI> = 1 Q9 1 Q9 1 and R = 1 Q9 1.

Definition XVI.5.1. A quantum enveloping algebra (QUE) for the Lie al-
gebra g is a topological braided quasi-bialgebra A = (A, p" 'f/,~, c, cI>, R) such
that A is a topologically free module, the induced braided quasi-bialgebra
A = (A/hA, jl, i), L5., if, <1>, R) as in Example 4 of Section 4 coincides with
the trivial braided quasi-bialgebra structure of U(g) and the map 'f/ is triv-
ially extended from i).

Let us be more explicit about this definition. First, a QUE is topologically

free. This means that A = (A/hA)[[h]] as a left K-module. By hypothesis,
we also have A/hA = U(g).
396 Chapter XVI. Generalities on Quantum Enveloping Algebras

Therefore
A = U(g)[[h]] (5.1)
as a K-module. From Proposition 3.2 we derive

A0 n = (U(g)0 n ) [[h]] (5.2)

for all n > O. By Proposition 2.3 (b) we know that the maps p, 'T/, ~ and
C are determined by their restrictions to U(g) U(g), C, U(g) and U(g)
respectively. For elements a, a' E U (g), we have

p(a a') = L Pn(a a')h n (5.3)

n2:0
where (Pn)n>O is a family of linear maps from U(g) U(g) to U(g) such
that Po is the product in the enveloping algebra. Similarly,

~(a) = L ~n(a)hn (5.4)

n2:0
where (~n)n>O is a family of linear maps from U(g) to U(g) U(g) such
that ~o is the comultiplication of the enveloping algebra described above.
We also have
(5.5)
n2:0
where (cn)n>O is a family of linear maps from U(g) to C such that Co is
the counit of the enveloping algebra. The last part of Definition 5.1 means
that the unit 'T/ of A is given by

'T/(f) = 11 (5.6)
for all 1 E C[[h]]. Finally, by Proposition 3.2 again, the elements <P and R
can be written
<P = L<Pn hn (5.7)
n2:0
and
R= L Rn hn (5.8)
n2:0
where (<pn)n>O and (Rn)n>o are families of elements of U(g)03 and U(g)0 2
respectively such that -

<Po = 1 1 1 and Ro = 1 1. (5.9)

It is clear from Lemma 1.1 that (5.9) ensures the invertibility of <P and R.
By definition, a quantum enveloping algebra A is associated to a Lie
algebra g. One recovers 9 from A by

9={XEA/hA I ~o(a)=la+al} (5.10)

 XVI.5 Quantum Enveloping Algebras 397

in view of the fact (stated in V.g) that the subspace of primitive elements
in U (g) is 9 provided that the ground field is of characteristic zero.
We now associate another invariant to a QUE A. If R is its universal
R-matrix and R21 is the image of R under the flip, the formula

(5.11)

defines a unique element t E U(g) 0 U(g). Expressing t using (5.8), we get

t = R1 + (R 1)21' which immediately proves that

t21 = t. (5.12)

Proposition XVI.5.2. The element t is an invariant symmetric element

of 9 0 g, i.e., we have t21 = t and [.6. (x) , tj = 0 for all x E g. It remains
unchanged under any gauge transformation.

The element t E 9 0 9 will be referred to as the canonical 2-tensor of the

quantum enveloping algebra A. Drinfeld calls the pair (g, t) the classical
l'lm'it of A and the quantum enveloping algebra A a quantization of the
pair (g,t).
PROOF. We have already observed that t is symmetric. Let us prove that it
belongs to the subspace 9 0 g. We again use the fact that 9 is the sub-
space of primitive elements in the enveloping algebra. Let us compute
(.6.0id A )(R21R) using (4.16-4.17). We have

( (id A0.6.)(R)) (.6.0idA)(R)

312
<1>-1 R 32 <1>132 R31 (<1>312) -1<1>312R13 (<1>132) -1 R 23 <1>.

Identifying the coefficients of h, we get

(5.13)

Now write t = L:i xi 0 Yi where (Yi)i is a family of linearly independent

elements of U(g). Thus (5.13) becomes

(5.14)

which implies that .6. 0(x i ) = Xi 01 + 1 0 xi for all i. Since the element xi is
primitive for the comultiplication .6. 0 of U(g), it belongs to g. Consequently,
t belongs to 9 0 Ug. Relation (5.12) implies actually that t is in 9 0 g.
Let us check the invariance of t. By (4.15) applied twice, we get

(5.15)

for all a E A. Identifying the coefficients of h, we obtain .6. o(x)t = t.6. o(x)
for all x E g.
398 Chapter XVI. Generalities on Quantum Enveloping Algebras

Finally, let us apply a gauge transformation F to A. Then by (XV.3.9)

we have
(5.16)
Taking the coefficients of h, we get tF = t where tF is the canonical2-tensor
of the QUE AF obtained from A by the gauge-transformation F. D

At this point, the only explicit quantum enveloping algebras for a given
Lie algebra g we know are the trivial QUE U(g)[[h]] constructed from U(g)
as explained in Example 3 of Section 4 and their gauge-transforms. Since
the universal R-matrix of such a QUE is 1 1, the corresponding canonical
2-tensor vanishes: t = O.
We now present an example of a QUE with a non-zero canonical2-tensor.
We shall see more non-trivial examples in Chapters XVII and XIX.
Example 1. (A quantum enveloping algebra associated to the Heisenberg Lie
algebra) We consider the 3-dimensional Lie algebra g with the set {x, y, z}
as a basis and with Lie bracket determined by
[x,y] = z and [x,z] = [y,z] = O.
The symmetric 2-tensor t = z (51 z is invariant because z is central in the Lie
algebra. We claim that there exists a QUE whose classical limit is (g, t).
Indeed, take the trivial bialgebra A = U(g) [[hll as in Example 3 of Section
4, except that we set R = e ht / 2 and <f> = 1~1~1. In order to make sure that
A is a topological braided bialgebra, we have to check Relations (4.15-4.17).
The first one follows from the fact that t is invariant. Relations (4.16-4.17)
with <f> = 1~1~1 are equivalent to
eh (t 13+ t 12)/2 = ehh3/2eht12/2 and e h(h3+t23)/2 = ehh3/2eht23/2. (5.17)
Relations (5.17) hold because the elements t 12 , t 13 and t 23 commute with
one another, due to the centrality of z. Now,
R21 R = e ht == 1 (51 1 + ht mod h 2
shows that t = z z is the canonical 2-tensor of A.

XVI. 6 Symmetrizing the Universal R-Matrix

The aim of this section is to prove that the universal R-matrix R of a
quantum enveloping algebra will always satisfy R = R21 after a suitable
gauge transformation. We start with the following technical result.
Lemma XVI.6.1. Let A be a topological algebra which is a topologically
free module. Given an element a E A, there exists a unique family (cn)n>O
of complex numbers such that

(1+
n>O
L Cn an h n f= 1 + ah. (6.1)
 XVI.6 Symmetrizing the Universal R-Matrix 399

PROOF. Any formal series 1 + I:n>O cnanh n of the above form defines an
element of the inverse limit A = lim A/hn A, hence of A since A ~ A by
n
hypothesis. Equation (6.1) is equivalent to the system of equations
n-1
2c 1 = 1 and 2c n +L cpcn _ p = 0 (6.2)
p=l
if n > l. This system has a unique solution as can be seen by an easy
induction. 0

The unique element 1+ I:n>O cnanh n satisfying (6.1) is called the square
root of the element 1 + ah and is denoted by (1 + ah)1/2. Its inverse will be
denoted by (1 + ah)-1/2.

Proposition XVI.6.2. Let 9 be a complex Lie algebra and A be a quan-

tum enveloping algebra for g. Then there exists a gauge transformation
F E A0A with F == 1 Q9 1 modulo h such that, if we set R' = F21 RF- 1,
then R;l = R'. Moreover, if A is cocommutative, there exists such an F
satisfying the additional relation F!:::"(a) = !:::"(a)F for all a E A.

PROOF. For any element u E A0A, define u = u 21 . If R' = FRF- 1, then

R' = F RF- 1. We look for an element F such that R' = R'. In other words
we must solve the equation

(6.3)
which can also be written in the form

(6.4)

We claim that
1/2
F = ( R(RR)-1/2 ) (6.5)
is a solution of (6.4) where we use the notation defined after Lemma 6.l.
The element F is invertible and congruent to 1 Q9 1 modulo h since R is.
In order to prove the claim, we observe that ReRR) = (RR)R implies
R(RR)n = (RR)n R for all n 2 1, hence

Rf(RR) = f(RR)R (6.6)

for any complex formal series f in the variable h; in particular, we have

R(RR)-1/2 = (RR)-1/2 R. (6.7)

Let us compute F2F2. By (6.7) we have

F2 F2 = R(RR)-1/2R(RR)-1/2 = (RR)-1/2 RR(RR)-1/2 = 1 Q9 l.

400 Chapter XVI. Generalities on Quantum Enveloping Algebras

Consequently, p2 = F- 2 . By uniqueness of the square root, we obtain

p = F- 1 . Using (6.7) again, we derive
(F- 1FR)2 = (P 2R)2 = (R(RR)-1/2R)2 = (RR(RR)-1/2? = RR,
which proves that (6.4) has a solution. This takes care of the first part of
the proposition.
As for the second one, observe that Relation (XV.2.1) and the cocommu-
tativity of .6. imply that .6.(a) commutes with R and with R for all a E A.
Consequently, .6.(a) commutes with F in view of (6.5). 0

XVI. 7 Exercises
1. Show that
1+ !!. + '"' (_1)n-1 (2n - 3)!! hn
2 ~ 2nn!
n2:2
is a square root of 1 + h in the algebra C[[hll of formal series where
(2n - 3)!! = TI~:i (2k - 1).
2. Let M = V[[hll and N = W[[h]] be topologically free modules.
Show that HomK(M, N) is a topologically free module isomorphic to
Hom(V, W)[[h]]. Deduce that if P is a third topologically free module,
then
Hom K (M0N,P) ~ HomK(M,HomK(N,P)).
3. Let g be a Lie algebra and t E g0g such that [t12' td = [t 13 , t 23 ] = 0
in U(g)3. Consider the gauge transformation F = e ht . Show that
(U g[[hllh is a topological bialgebra.
4. Show that the inverse systems of abelian groups and the maps of
inverse systems form a category Inv such that li-Ill is a functor from
n
Inv to the category Ab of abelian groups. Prove that Ii-Ill is left adjoint
n
to the functor assigning to each abelian group A the constant inverse
system (An,prJ where An = A and Pn = idA for all n.
5. Let (Cn)n>O be a denumerable family of abelian groups. Consider
the inverse system (An,p n ) where An = Co X ... X C n and Pn is
the natural projection. Prove that the inverse limit of this system is
isomorphic to the direct product of all groups Cn.
6. Let (An,p n ) be an inverse system of abelian groups. Use the fact that
its inverse limit can be expressed as the kernel of an endomorphism
of TIn An to prove that for any abelian group C there is a natural
isomorphism
Hom(C,li-Ill An) ~ li-Ill Hom(C, An)
n n
 XVI.9 Appendix. Inverse Limits 401

7. (The ring of p-adic integers) Given a prime p consider the inverse sys-
tem of rings (Zjpnz) equipped with the natural projections induced
by the inclusions of ideals (pn) C (pn-l). Show that the inverse limit
Zp is a ring with a unique maximal ideal. Prove that the inverse limit
topology on Zp can be defined by an ultrametric distance and that
the ring of natural integers Z forms a dense subring of Zp in which
all integers prime to p are invertible.

XVI. 8 Notes
The material of Sections 1-4 is standard. For details on h-adic topology
and completions, read [Bou61], III and [Mat70], Chap. 9. The concept of
a quantum enveloping algebra and the content of Sections 5-6 are due to
Drinfeld (see [Dri87] and [Dri89b], Section 3). Exercise 3 is taken from
[Enr92].

XVI. 9 Appendix. Inverse Limits

An inverse system of abelian groups (An'Pn) is a family (An)nEN of abelian
groups and of morphisms of groups (Pn : An ~ An-1)n>O' Given such a
system we can define its inverse limit 1l!P- An by
n

liI!l An = {(xn)n;:::o E II An I Pn(Xn) = x n- 1 for alln > o}. (9.1)

n n;:::O
The inverse limit has an abelian group structure as a subset of the direct
product ITn>O An whose group structure is defined component-wise. The
natural projection from ITn>O An to Ak restricts to a morphism of groups
7rk : liI!l An ~ A k It is defined by 7rk((xn)n) = Xk' If all maps Pn are
n
surjective, then so are the maps 7rn .
By definition of the inverse limit, we have

(9.2)

for all n > O. The inverse limit has the following universal property.
Proposition XVI.9.1. For any abelian group C and any given family
Un : C ~ An)n>O of morphims of groups such that Pn 0 fn = fn-l for all
n > 0, there exists a unique morphism of groups

such that 7rn 0 f = fn for all n 2: O.

402 Chapter XVI. Generalities on Quantum Enveloping Algebras

PROOF. The family Un)n defines a unique morphism of groups i from C

into the direct product of all groups An' The hypothesis Pn 0 in = in-l
implies that the image of i lies in the subgroup 1,Lrp. An' This proves the
n
existence of f. The condition 7rn 0 i = in implies its uniqueness. 0

The inverse limit is functorial. Define a map from the inverse system
(An'Pn) to the inverse system (A~,p~) as a family Un : An ----t A~)n>O of
morphisms of groups such that p~ 0 in = in-l 0 Pn for all n > O. -
Proposition XVI.9.2. Under the previous hypothesis, there exists a
unique morphism of groups

n n n

such that 7r~ 0 f = fn 07rn for all n ~ O.

PROOF. The family Un 0 7r n : l@ An ----t A~)n satisfies the hypothesis of

n
Proposition 9.1. It follows that there exists a unique morphism f such that
7r~ 0 f = f n 0 7rn for all n. 0

For compos able maps of inverse systems, we have

The inverse limit of any inverse system (An' Pn) possesses a natural topol-
ogy called the inverse limit topology. It is obtained as follows. Put the dis-
crete topology on each An' i.e., the topology for which each subset is an
open set. The inverse limit topology on 1,Lrp. An is the restriction of the di-
n
rect product topology on ITn>O An' In other words, a basis of open sets of
the inverse limit is given by the family of all subsets 7r;;l(Un ) where n runs
over the non-negative integers and Un is any subset of An' By definition of
this topology, the structural maps 7rn from 1,Lrp. An to An are continuous.
n
Moreover, a map i from a topological set to 1,Lrp. An is continuous with
n
respect to the inverse limit topology if and only if the map 7rn 0 i into An
is continuous for all n ~ O.
One may replace the word "abelian group" by "ring", "module" ... in the
above definition. The statements ofthe Appendix remain true in this case, a
fact we have consistently used in this chapter without further explanation.
Chapter XVII
Drinfeld and Jimbo's Quantum
Enveloping Algebras

In Part I we have investigated at length the quantum enveloping alge-

bra of s[(2). In this chapter we give a brief presentation of the algebras
Uh (g) associated by Drinfeld [Dri85][Dri87] and Jimbo [Jim85] to the other
semisimple Lie algebras g. The algebras Uh(g) provide non-trivial examples
of quantum enveloping algebras as defined in XVI.5 as well as examples
of isotopy invariants of links. We shall also need Uh (g) in Chapter XIX
to state the Drinfeld-Kohno theorem on the monodromy of the Knizhnik-
Zamolodchikov systems. Finally, in Section 4 we shall determine an explicit
universal R-matrix for the quantum enveloping algebra of s[(2), using the
crossed bimodules of IX.5.

XVII. 1 Semisimple Lie Algebras

Before we present Drinfeld and Jimbo's quantum enveloping algebras, we
recall a few facts from the theory of complex semis imp Ie Lie algebras.
Let 9 be a finite-dimensional complex Lie algebra. For any finite-dimen-
sional representation p of g, we can define a bilinear form on 9 by

< X,y >p=tr(p(x)p(y)) (1.1 )

where x, y are elements of g. From the properties of the trace, we immedi-

ately see that this bilinear form is symmetric and invariant, i.e., we have

< y,x >p=< X,y >p and < [x,y],z >p=< x, [y,z] >p (1.2)
404 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

for all elements x, y, and z of g. When p = ad is the adjoint representation

of g, the bilinear form <, > ad is called the Killing form of g.
A semisimple Lie algebra is a finite-dimensional complex Lie algebra
whose Killing form is non-degenerate. For any basis {xiL of g, there exists
a unique basis {xiL called the dual basis of {xJi and determined by

for all i, j. Define linear forms aij and (3ij on 9 by

j j

Lemma XVII.I.I. We have (3ij = -aji for all i,j.

PROOF. Applying (1.2) to the Killing form, we get

(1.4)

Expanding the left-hand side of (1.4) gives < [Xi' X], x j >ad = aij(x) whereas
we have < Xi' [X, xj] >ad = -(3ji(X) for the right-hand side. D

We now define the Casimir element C of 9 as the element

C= 2.::: XiX i (1.5)

of the enveloping algebra U(g).

Proposition XVII.I.2. The Casimir element C is independent of the ba-
sis {xi L and belongs to the centre of U (g).
PROOF. The first assertion follows from a well-known fact in linear algebra:
if {yJi is a basis related to the basis {xJi by Yi = I:.i AijXj where the
scalars (Aij)ij form an invertible matrix A, then the dual basis {yiL of the
basis {YiL is related to {xiL by

yi = 2.::: Bjixj
j

where B = (Bij)ij is the inverse of the matrix A. Now,

2.::: (2.::: BkiAij )xjx k

j,k .
2.::: t5 kj Xj Xk
j,k
'L..,
" xxJ
j = C'
j

which proves that C is independent of the choice of bases.

 XVII.1 Semisimple Lie Algebras 405

In order to prove that C is central, it suffices to check that C commutes

with any element x of g. We have

[C,x] L[xixi,X]

L xi[Xi,X] +L [Xi,X]xi

o
by Lemma 1.1. D

Using the comultiplication 6. of the enveloping algebra, we derive the

element

t = 6.( C) - 1 @ C - C @ 1 = """' xi @ xi + xi @ Xi (1.6)

2 ~ 2

of g @ g. This element will playa central r61e in Chapter XIX. It enjoys the
following properties.

Proposition XVII.1.3. The element t is a symmetric g-invariant ele-

ment of g @ g, i. e., we have

t21 =t and [6.(x), t] = 0 (1. 7)

for all x E g, where t21 = Tg,g(t).

PROOF. The symmetry of t is clear from its definition. As for the g-

invariance, it is enough to prove that 6.(x) commutes with 6.( C) and with
l@C+C@1. For the first condition, we have [6.(x), 6.(C)] = 6.([x, CJ) = 0
since 6. is a morphism of algebras and C is central. We also have

[6. (x), 1 @ C + C @ 1] = 1 @ [x, C] + [x, C] @ 1 = 0

again because C is central. D

Example 1. Consider the 3-dimensional simple Lie algebra 5[(2) of Chap-

ter V. It is easy to check that its Killing form is non-degenerate, and that the
dual of the basis {X, Y, H} considered in V.3 is the basis {Y/4, X/4, H/8}.
Consequently, for 5[(2) we get

(1.8)
406 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

Elie Cartan characterized every semisimple Lie algebra by its Cartan

matrix, which is a square matrix A = (aijh::;i,j::;n with the following prop-
erties:
(i) its coefficients aij are non-positive integers when i =I- j, and aii = 2,
(ii) there exists a diagonal matrix D = diag(d l , ... , d n ) with entries,
called the root lengths, in the set {I, 2, 3} such that the matrix DA is
symmetric positive definite.
According to a theorem of Serre's [Ser65], the enveloping algebra U(g) of
g is isomorphic to the algebra generated by 3n generators {Xi' Yi, HJl<i<n
and the relations - -

[Hi' Hj ] = 0, [Xi ,1j] = DijHi' (1.9)

[Hi,Xj ] = aijXj , [Hi ,1j] = -aij 1j, (1.10)
and if i =I- j

(1.11)

and
L
l-aij

k=O
(_l)k C -kaij ) Yik1jYiI-aij-k = O. (1.12)

The Cartan matrix for .5[(2) is the 1 x I-matrix A = (2) with D = (1).
In this case, the presentation above reduces to the formulas (V.3.2).
We end this summary by a few words on the representation theory of a
semisimple Lie algebra g. Any finite-dimensional g-module is semisimple,
i.e., is the direct sum of simple modules. The finite-dimensional simple g-
modules are classified by the set of dominant weights: a dominant weight is a
linear form A on the subspace f) of g spanned by HI"'" Hn such that A(Hi)
is a non-negative integer for all i = 1, ... ,n. For every dominant weight A,
there exists a unique finite-dimensional simple g-module VA generated by
an element VA' called a highest weight vector, such that
(1.13)
for all i = 1, ... ,n. All finite-dimensional simple g-modules are of this form.
The Casimir element C acts by a positive scalar on every simple g-module
VA of dimension> 1, i.e., with A =I- O. We have proved these facts for 5[(2)
in Chapter V. In the case of .5[(2) the set of dominant weights is in bijection
with N, the dominant weight A corresponding to the integer n being defined
by A(H) = n.

XVII. 2 Drinfeld-Jimbo Algebras

Before we describe the quantum enveloping algebra Uh(g) attached by Drin-
feld and Jimbo to any complex semisimple Lie algebra g, we introduce the
 XVII.2 Drinfeld-Jimbo Algebras 407

notion of a topological algebra presented by generators and relations. Recall

the notation K = C[[h]] and Kn = C[h]/(hn) from XV1.l.
Given a set X, we define the topologically free algebra generated by X as
the algebra of formal series over the free complex algebra generated by the
set X:
K(X) = (C(X))[[h]].
We equip K(X) with the h-adic topology. It has the following universal
property which is the topological counterpart of Proposition 1.2.l.

Proposition XVII.2.1. Let f : X ---. A be a map from a set X to a

separated complete K -algebra A. Then there exists a unique continuous K-
linear map 1:K(X) ---. A such that f(x) = f(x) for all x E X.

PROOF. Clearly, f extends to a unique Kn-linear algebra morphism

We then take the inverse limit of the maps f n . The uniqueness of 1 results
from the fact that the K-subalgebra generated by X is dense in K(X). D

Definition XVII.2.2. Let X be a set and R be a subset of the topologically

free algebra K (X) generated by X. A K -algebra A is said to be the K-
algebra topologically generated by the set X of generators and the set R of
relations if A is isomorphic to the quotient of K (X) by the closure (for the
h-adic topology) of the two-sided ideal generated by R.

As an immediate consequence of Proposition 2.1 and of Definition 2.2,

we see that the space of morphisms of K-algebras from A to a separated
complete K-algebra A' is in bijection with the set of maps f : X ---. A' such
that 1 vanishes on R.
We also recall the definition of the following symbols already considered
in V1.l. We added a subscript q in order to stress the dependence on the
parameter q. For any invertible element q and any integer n, define

If r is a non-negative integer, set [O]q! = 1 and if r > 0

[n]q[n - l]q ... [n - r + l]q

[r]q!

We now turn to the definition of the Drinfeld-Jimbo algebra Uh(g). Let

g be a complex semisimple Lie algebra and A = (aij)l::;i,j::;n be its Cartan
matrix, with the diagonal matrix D = diag( d 1 , ... ,dn ) of root lengths.
408 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

Definition XVII.2.3. The algebra Uh(g) is defined to be the K-algebra

topologically generated by the set of generators {Xi'~' HihSciScn and the
relations
(2.1)

(2.2)
and if i -=I j

and

(2.4)

where qi = e hdi / 2 and where sinh is the formal series

eX _ e- x X 2n +1
sinh(x) = 2 = L
n2:0
(2n + I)!'

Note that, although sinh(hd;/2) is not invertible, it is the product of h

with a unique invertible element, so that sinh(hd i H;/2)j sinh(hd;/2) is a
well-defined element of K({Xi'~' Hi}lSciScn)' We have

sinh(hdi H;/2) _
sm 2 = Hi
. h(JLd i j) mod h.

Observe also that Relations (2.2) imply that

eAhHiX. = eAaijX.eAhHi and eAhHiY. = e-AaijY.eAhHi (2.5)
J J J J

for all i, j and any complex number A.

We now state the main result of this section.

Theorem XVII.2.4. The topological algebra Uh(g) is a quantum envelop-

ing algebra
(Uh(g), Mh' 7]h' 6. h , ch' <I>h' R h )
for the Lie algebra g with <I>h = 10101 and comultiplication 6. h and cound
c h determined by
6. h (Hi ) = Hi 1 + 1 Hi' (2.6)
6. h(X)
],
= X ], e hdiH ;/4 + e- hdiH ,j4 X P (2.7)
(2.8)
and
(2.9)
 XVII.2 Drinfeld-Jimbo Algebras 409

We refer to [Dri87] for a proof. Let us make a few remarks. First, if we
set h = in Relations (2.1-2.4) and (2.6-2.9), we recover the enveloping
algebra of 9 in Serre's presentation. In other words, we have an isomorphism
of algebras
(2.lO)
The fact that Uh(g) is a topologically free K-module is not straightforward.
It can be proved by constructing a Poincare-Birkhoff-Witt-type basis. One
has also to check that (2.6-2.9) define morphisms of algebras D..h and ch'
For D..h this follows from (2.5) and the q-binomial formula of Proposition
IV.2.2.
The topological bialgebra Uh (g) has an antipode Sh determined by

Sh(Hi ) = -Hi, Sh(Xi ) = _ ehdi/ 2Xi' Sh(Yj) = _e- hdi/2Yj. (2.11)

Note that the comultiplication of Uh(g) is not cocommutative and that the
antipode is not involutive. Nevertheless, for all a E Uh(g) we have

(2.12)

where p = 2:~=1 J-liHi' the scalars J-li being determined from the inverse
A-I of the Cartan matrix by J-li = 2:7=1 (A-l)jidj.
More importantly, Theorem 2.4 implicitly states that Uh (g) has a uni-
versal R-matrix, which we denote by R h . Drinfeld proved that Rh is of the
form

(2.13)

to = :E (DA)ijl Hi 129 H j (2.14)

lSi,jsn
of 9 129 g, and Pe is a polynomial in the variables Xl 129 1, ... , Xn 129 1 and
in 1129 Yl , ... , 1129 Y n (homogeneous of degree i in Xi 1291 and 1129 Yj). We
have Po = 1 129 1 and
Rh == 1 129 1 mod h. (2.15)
The polynomials Pe can be determined by induction on using Relations
(XVI.4.15-4.17). Explicit expressions for Rh can be found in [KR90] [LS90]
[Ros89] [Ros92].
The representation theory of Uh(g) is parallel to that of the Lie algebra g.
Indeed, for any dominant weight .A of g, there exists a unique topologically
free Uh(g)-module VA satisfying

(2.16)
410 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

and generated by an element VA' called a highest weight vector, such that

(2.17)

for all i = 1, ... ,n, as in the classical case. Rosso [Ros88] proved that any
topologically free Uh(g)-module W with dim(W/hW) < 00 was a direct
sum of modules of the form VA' We shall give an explanation of this fact
in XVIII.4.

XVII.3 Quantum Group Invariants of Links

We now show how to construct an isotopy invariant Qg,v out of any com-
plex semisimple Lie algebra g and of any finite-dimensional simple g-module
V.
Consider the category Uh(g)-Mod fr of finite-rank topologically free Uh(g)-
modules, i.e., of topological modules of the form V[[h]] where V is a finite-
dimensional vector space. This category is a tensor category for the topo-
logical tensor product of XVI.3, the associativity and unit constraints
being the canonical isomorphisms (XVI.3.1) and (XVI.3.3). Actually, the
category Uh (g)-Mod fr is a braided tensor category with left duality: the
braiding is induced by the universal R-matrix Rh while the duality is given
on objects by V[[h]]* = V*[[h]]. The structure maps band d of duality
are C[[h]]-linearly extended from the evaluation and coevaluation maps of
II.2-3.
We claim that Uh (g)-Mod fr is a ribbon category. To sustain the claim,
it suffices to exhibit a twist as defined in XIV.3. We proceed as in XIV.6.
Let u be the invertible element of Uh(g) defined by Formula (VIII.4.1),
which still makes sense in the present context. We have u == 1 modulo h.
By Proposition VIII.4.1 and by (2.12), we get

for all a E Uh (g). This implies that

(3.1)

belongs to the centre of Uh(g).

Proposition XVII.3.1. The central element e satisfies the relations

PROOF. We claim that

e2 = uS(u) = S(u)u. (3.2)

 XVII.3 Quantum Group Invariants of Links 411

Then Proposition 3.1 follows from Proposition VIII.4.5 and from the fact
that uS (u) has a unique square root whose constant term is 1.
Relation (3.2) is reduced in [Dri89a], Proposition 5.1 to showing that
both terms have the same action on all modules of the form VA' It is
enough to evaluate the ~ntral elements S(u)u and 02 = e- 2hP u 2 on a
highest weight vector of VA' Since u can be expressed in such a way that
the generators Xi killing the highest weight vector appear to the right of
Yi, we see that the actions of S(u)u and of 02 are the same as the actions
of the elements obtained from the part of Rh corresponding to I! = 0 in
Formula (2.13). A simple computation shows then that S(u)u and of 0 2 act
by the same scalar on VA' For more details, see [Dri89a], Section 5. 0

Combining Propositions 3.1 and XIV.6.2, we conclude that the action

by 0- 1 induces a twist on the category Uh(g)-Mod jr , thus turning it into
a ribbon category.
By Section 2, any finite-dimensional simple g-module V gives rise to a
unique object V of Uh(g)-Mod jr such that VIhV_= V. Applying Theorem
XIV.5.1 to our ribbon category and to the object V, we get a tensor functor
Fv from the category n of framed tangles to Uh(g)-Mod jr , sending the
object (+) to V. Restricting Fv to framed links yields an isotopy invariant
Qg,v for framed links with values in C[[h]]. It is easy to check that

(3.3)

for any link of order d. Since Uh(g)-Mod jr is a ribbon category, we have

a quantum dimension for any object. Actually, by definition of the functor
Fv above, dimq(V) coincides with the value of Qg,v on the trivial knot. Let
us explain how one may determine dimq (V) when V = VA for a dominant
weight A. By Proposition XIV.6.4, dimq(VA) is the trace of the action of
0-l U = e hp on VA' Since VA has the same weight decomposition as the g-
module VA = VAl h VA and since e hp acts on a vector of weight JL by eh</l,p> ,
we may compute dimq(VA) as follows. Use Weyl's character formula to
determine the character

(3.4)

of the simple module VA (as defined, e.g., in [Bou60], Chap. 8, 9) where JL

runs over the weights of VA' Then

dimq(VA) = L dJ.L eh</l,p>. (3.5)

/l

When h = 0, we recover the dimension of VA' Therefore, the quantum

dimension of VA may be viewed as a q-analogue of the dimension of VA'
412 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

We end this section by stating a special property of the universal R-

matrix Rh of Uh(g). In the next chapter (see Corollary XVIII.4.2) we shall
establish the existence of a unique isomorphism of topological algebras a
from the centre of Uh(g) to the algebra Z(g)[[hll of formal series over the
centre Z(g) of U(g) such that a == idz(g) modulo h. The Casimir element
C in Z(g) C Z(g)[[h]]' as defined by (1.5), can be pulled back to a unique
central element, the quantum Casimir element,

(3.6)

of Uh(g) satisfying
Ch==C modh. (3.7)
Proposition 5.1 of [Dri89a] asserts that

(3.8)

This is proved along the sameJ!nes as Proposition 3.1 above. In order to

determine the action of C h on VA' it is enough to evaluate the action of the
classical Casimir operator C on VA' which is well-known.
Combining (3.8) and Proposition 3.1, we get the following property for
the universal R-matrix of Uh(g).

Proposition XVII.3.2. The universal R-matrix Rh ojUh(g) satisfies the

relation

As an immediate consequence, we see that the canonical 2-tensor of Uh (g)

(as defined in XVI.5) is the symmetric invariant 2-tensor t of (1.6).

XVII.4 The Case of .5[(2)

When the Lie algebra g is the three-dimensional Lie algebra s((2), then
Definition 2.3 implies that Uh = Uh (s((2)) is the K-algebra topologically
generated by the three variables X, Y, H and the relations

[H, X] = 2X, [H, Y] = -2Y (4.1)

and
sinh(hH/2) e hH / 2 _ e- hH / 2
[X, Y] = sinh(h/2) = eh/2 _ e- h/ 2 (4.2)

The following relates the Hopf algebra Uq = Uq (s((2)) of Chapters VI

and VII with Uh . We assume that the ground field k on which Uq is defined
is the field of fractions of the algebra K of complex formal series.
 XVlI.4 The Case of 5[(2) 413

Proposition XVII.4.I. There exists a map of Hopf algebras i : Uq ----+ Uh

such that

i(E) = X e hH / 4 , i(F) = e- hH / 4 y, i(K) = e hH / 2 , i(K- 1 ) = e- hH / 2 ,

(4.3)
and i(q) = eh / 2 .
The proof is left as an exercise (use Relation (2.5) among others). Ac-
tually, the map i is injective, which allows one to identify Uq with the
subalgebra of Uh generated by q = e h / 2 , E = Xe hH / 4 , F = e- hH / 4 y,
K = e hH / 2 , and K- 1 = e- hH / 2 .
We now describe the universal R-matrix Rh of the topological braided
bialgebra Uh .

Theorem XVII.4.2. The element

e h(H;;H) (L (q [ ]q.q~l )e l(e-l)/2 (Ee

e~o
-I! (9 Fe))

of Uh0Uh is a universal R-matrix for Uh(5[(2)).

h(HCZ!H)
Observe that it is because of the factor e -4- that Rh is not well-
defined on the subalgebra Uq , thus preventing Uq from being a braided
Hopf algebra in the purely algebraic sense of VIII.2.
PROOF. The second equality is easy: it follows from the definitions of E
and F. We leave it to the reader and concentrate on the first one. There are
several proofs for it. The first one follows from a direct checking of Relations
(XVI.4.15-4.17). Another method consists in adapting Drinfeld's theory of
the quantum double (as developed in Chapter IX) to the topological setting
and then in proceeding along the lines of the proof of Theorem IX.7.1. This
second method has been used by Rosso in [Ros89]. We shall sketch a third
way using topological crossed bimodules as defined at the end of XVI.4.
This proof goes as follows. As in IX.6 we start by defining a subbialgebra
Eh of Uh . It is the closure of the K-submodule of Uh generated by the
linearly independent set {Hm En}m,n>O where E = X ehH / 4 as above. From
(4.1) and (4.3) it is clear that Eh IS closed under the product and the
coproduct in Uh , and that the multiplication in Eh is determined by the
relation
[H,E] = 2E, (4.4)
and the comultiplication by

6.(H) = 1 (9 H + H (91 and 6.(E) = 1 (9 E + E (9 K (4.5)

414 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

where K = e hH / 2 . Observe also that

KE=lEK (4.6)

for q = e h / 2 . We now characterize topological crossed B h - bimodules.

Proposition XVII.4.3. Let M be a topological crossed Bh -bimodule with

coaction 6. M' Then for any element x E M we have

where 6. 1 and 6. 2 are h-adically locally nilpotent K -linear endomorphisms

of M such that
h
[6. 1 ,6. 2 ] = -26. 2 , (4.7)

[H, 6. 1 ] = 0, [H,6. 2 ] = -26. 2 , (4.8)

h
[E,6. 1 ] = -2E, [E,6. 2 ] = K - e 2 6. 1 (4.9)

A K-linear endomorphism 6. of M is said to be h-adically locally nilpotent

if for all x E M there exists an integer n such that 6. i (x) c hM for all
i ~ n. The h-adic local nilpotence condition on 6. 1 and 6. 2 ensures that
the infinite sum in the statement above converges in the h-adic topology.
We identified E, H, and K with their actions on M in Relations (4.7-4.9).
Proposition 4.3 will be proved later.
Let us denote by Dh the K-algebra topologically generated by E, H, 6. 1 ,
6. 2 and Relations (4.4) and (4.7-4.9). Proposition 4.3 can be interpreted as
saying that a topological crossed Bh-bimodule is the same as a topological
Dh-module with an h-adic local nilpotence condition. The algebra Dh can
be considered as a kind of quantum double for Bh in view of Theorem
IX.5.2.
The next step in the proof of Theorem 4.2 is the following.

Proposition XVII.4.4. There exists a morphism of topological algebras

X: Dh ---> Uh such that

X(E) = E, X(H) = H,

Observe that X(6. 1 ) == X(6. 2 ) == 0 mod h. Consequently, X(6. 1 ) and

X( 6. 2 ) are also h-adically locally nilpotent.
PROOF. It is essentially enough to check that X(E), X(H), X(6. 1 ) and X(6. 2 )
satisfy Relations (4.4) and (4.7-4.9). This is straightforward, except possi-
bly for the second formula in (4.9).
 XVII.4 The Case of ,5[(2) 415

Let us check it: we have

(q - q-l )[E, F]
K-K-l
K _ e- hH/ 2
X(K - e- 2 .6. 1 ).
o
The final step of the proof of Theorem 4.2 goes as follows: By Propo-
sitions 4.3-4.4 we know that any topological Uh-module becomes a topo-
logical crossed Bh-bimodule via x. In view of Relation (IX.5.5) a universal
R-matrix for Uh is given by
qn(n-l)/2 _
Rh = 2:
m,n?O
I[ ]
m. n q.
I m
X(H En)@X(b.r;' b.~). (4.10)

By definition of X we get

"
(q - q -l)n qn(n-l)/2 _hm H m En @ H m F n
~ m![n]! 4m .
m,n2:0 q

(2: 4~:! H m @Hm ) (2: (q -nq~l)n qn(n-l)/2 En @Fn)

m?O n?O []q

eh(HH)/4 (2: (q - q~l )n qn(n-l)/2 En @ Fn).

n?O [n]q.
o
We now prove Proposition 4.3.
Proof of Proposition 4.3. For any x in M, the element b. M (x) is of the form

b.M(x) = 2: b.~,n,p(x) H m En mod h P (4.11)

m,n2:0

for all p > o. In the inverse limit the family (b.~,n,p)p assembles to form a
K-linear endomorphism b.~,n of M. Now the sum in (4.11) is finite, which
implies that b.~n(x) vanishes modulo h for m and n large enough.
The counitarity of b. M yields b.~o = id M whereas the coassociativity
gives

b.i,j b. m,n
M M
= -2nj (j + n)
q n
,,(jh)t
~ 2tt!
(i + t) m -
i
b.i+m-t,j+n
M
(4.12)
q2 t2:0

for all i, j, m, n after using the classical binomial formula as well as the
q-binomial formula of Proposition IV.2.2. Here we agree that b.~n = 0
416 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

when m or n < O. Set .6. 1 = .6.~0 and .6. 2 = .6. f;.:/. By (4.12) and (VI.1.7)
we get
.6. m ,O = ~(.6.1,0)m = ~.6.m, (4.13)
M m! M m! 1
n(n-l) n(n-1)/2
.6.O,n = q (.6. O,l)n = q .6. n (4.14)
M (n)q2! M [n]q! 2'

n(n-l)/2
.6. m,n - .6. m,O .6. O,n - q .6. m .6. n (4.15)
M - M M - m![n]q! 1 2'

From (4.12) and (4.15) we derive

which is equivalent to Relation (4.7).

Let us prove that .6. 1 and .6. 2 are h-adically locally nilpotent. Indeed,
we know that for any x E M we have .6.~o(x) == 0 modulo h for m large
enough. Now, .6.~,o(x) = .6.f'(x). Therefore, .6.f'(x) C hM for m large
enough. The h-adic local nilpotence of .6. 2 is proved similarly.
So far we have expressed the fact that M is a comodule. Now we deal
with Relation (XVI.4.23). Lengthy, but easy, computations using

EH m = (H - 2)mE and EnH = (H - 2n)En

show that (XVI.4.23) is equivalent to the two relations

.6. ~,n H = H.6. ~,n + 2n.6. ~,n (4.16)

and

.6. m,n E
M
+ .6. m,n-l
M
K= "'" ~
~2tt!
E.6. m-t,n + "'" (_2)r (M + r).6. m+r,n-1
M ~ r M
t20 r20
(4.17)
for all m, n ~ O. Specializing the exponents m and n to 0 and 1 in (4.16)
gives Relations (4.8) whereas setting m = 1 and n = 0 in (4.17) gives
[E,.6. 1] = -~E. When we set m = 0 and n = 1 in (4.17), then necessarily
t = 0 and we get

.6. 2E + K = E.6. 2 + L (-2r .6.~.

r20
Therefore,
[E, .6. 2] = K -
r.
L
(-2r .6.: = K - e- 2.6. 1 ,
r20
which is the second formula in (4.9). This completes the proof of Proposi-
tion 4.3. D
 XVIIA The Case of .5[(2) 417

We end this section by an explicit description of the topologically free Uh -

modules extending the simple Uq-modules V(n) ofV.4. For any nonnegative
integer n, consider a (n + I)-dimensional complex vector space V(n) with
a basis {vo,"" v n } and the free K-module Vn = V(n)[[hll = V(n) K.
Consider the three (n + 1) x (n + I)-matrices

0 [n]q 0 0
0 0 [n -1]q 0
Pn(X) =
0 0 1
0 0 0 0

0 0 0 0
1 0 0 0

Pn(Y) = 0 [2]q 0 0

0 0 [n]q 0
and
n 0 0 0
0 n-2 0 0
Pn(H) =
0 0 -n+2 0
0 0 0 -n

where q = eh / 2 and [n]q = S!~:~~!2~)' The matrices Pn(X), Pn(Y), and

Pn (H) satisfy Relations (4.1-4.2) and, hence, define a topological Uh-module
structure on Vn . Observe that

Vn/hVn = V(n) (4.18)

as g-modules and that, when viewed as a Uq-module via the injection i of

Proposition 4.1, Vn is isomorphic to the simple Uq-module V1,n of VI.3.
Let us check by a direct computation that the quantum dimension of Vn
as defined in Section 3 is given by

dimq Vn = [dim(V(n))]q = [n + l]q (4.19)

where q = e h / 2 . Indeed, the element P defined in Section 2 is P = in the If

case of .5[(2). By Section 3, dimq(Vn ) is equal to the trace of the action of
ehp .
418 Chapter XVII. Drinfeld-Jimbo Quantum Enveloping Algebras

Therefore,
+ e(n-2)h/2 + ... + e-(n-2)h/2 + e- nh / 2
e nh / 2

qn + qn-2 + ... + q-(n-2) + q-n

qn+1 _ q-n-1
-=--------=--:-1- = [n + 1] q
q - q-

The elements X and Y, and hence, E and Fact nilpotently on Vn so

that it makes sense to apply the universal R-matrix Rh to it. This allows
us to build a K-linear automorphism cn by

where v 1 , v 2 E Vn as in VIII. 3. The automorphisms cn are solutions of

the Yang-Baxter equation. In the case of V1 , an immediate application of
Theorem 4.2 shows that c 1 is defined in the basis consisting of the vectors
Vo Q9 va, v 1 Q9 v 1 , va Q9 v 1 , v 1 Q9 va by the matrix

q 0 0 0 )
-1/2
( 0 q 0 0
(4.20)
q 0 0 0 1 '
o 0 1 q _ q-1

an R-matrix already encountered in VIII.7 where it allowed us to define

the bialgebra Mq(2) and its quotient SLq(2) using the FRT construction.
This completes our study of the "quantum groups" associated to SL(2)
and of their relations to the R-matrix of (4.20).

XVII. 5 Exercises
1. Compute Pc in Formula (2.13) for Rh when g = (0, ... ,1, .. , ,0) where
1 occurs exactly once.

2. Show that Uh (.5((2)) is topologically free.

3. Let 9 = .5 (( 2) and V = V (1). Relate the isotopy invariant Qg.v to the

invariant <P 2 ,q of Proposition X.4.7.

XVII. 6 Notes
A full account of the theory of semisimple Lie algebras can be found, for
instance, in [Bou60][Dix74][Hum72][Jac79][Ser65][Var74]. See [Bou60] for
the complete list of Cartan matrices.
 XVII.6 Notes 419

The presentation we gave in Section 2 for Uh(g) as an algebra over C[[h]]

is due to Drinfeld [Dri85] [Dri87]. The version considered by Jimbo in
[Jim85] is the algebraic variant Uq(g) already discussed in VI.7. The latter
can be viewed as the Hopf subalgebra of Uh (g) generated by q = e h / 2 , Xi'
Yi "Iie', -- e hd i Hd2 ,an d K-
i lor ~. -- 1, ... , n.
1 l'

In the special case 9 = 13[(2), the algebra U h(13[(2)) had previously been
constructed by Kulish and Reshetikhin [KR81] with the Hopf algebra struc-
ture found by Sklyanin [Sk185].
Drinfeld devised the quantum double construction precisely in order to
find a universal R-matrix for Uh(g). This method was applied by Drinfeld
[Dri87] himself to give an explicit form of Rh in the case 13[(2) and by Rosso
[Ros89] in the case 13[(n). Expressions of the universal R-matrix in the
general case are due to Kirillov-Reshetikhin [KR90] and to Levendorsky-
Soibelman [LS90].
The representation theory of Uhg was elucidated by Lusztig [Lus88] and
Rosso [Ros88].
Chapter XVIII
Cohomology and Rigidity
Theorems

In this chapter we prove two rigidity theorems, both needed in Chapter

XIX. The first one is classical: it asserts that any formal deformation of the
enveloping algebra of a semi simple Lie algebra is trivial. The proof is based
on the vanishing of certain cohomology groups. The second rigidity result is
due to Drinfeld [Dri89b] [Dri90]. It states that if A and AI are quantum en-
veloping algebras with the same underlying cocommutative bialgebras and
the same universal R-matrices, then there exists a gauge transformation
from A to AI. The proof again relies on some co homological considerations,
this time involving the cobar complex of a symmetric coalgebra.
The ground field is assumed to be the field of complex numbers.

XVIII. 1 Cohomology of Lie Algebras

Let g be a Lie algebra and IvI be a left g-module, i.e., a vector space with
a bilinear map g x !vI --+ !'vI such that

[x, y]m = x(ym) ~ y(xm) (1.1 )

for all x, y E g and m E M. It was shown in V.2 that a left g-module is the
same as a left module over the enveloping algebra U(g) of g.
For n > 0, let Cn(g, M) = Hom(A ng, M) be the space of all antisym-
metric n-linear maps from g to M. An n-linear map f is antisymmetric
if f(Xu(l)' ,xcr(n)) = c(O")f(Xl' ... ,x n ) for all Xl' ... 'X n E g and all
permutations 0" of the set {I, ... , n}. If n = 0, we set CO(g, M) = M.
 XVIII.1 Cohomology of Lie Algebras 421

For f E en(g, M) we define a (n + I)-linear map of by

(of)(x l , ,xn+ l )
L (_I)i+j f([xi,xj],X I , ,Xi, ,ij, ... ,x n+ l )
l~i<j~n+l

+ L (_I)i+lxd(XI' ... ,Xi, ... ,xn+l) (1.2)

l~i~n+l

for all Xl" .. 'Xn+l E g. The hat ~ on a letter means that it has been
omitted. If f belongs to eO(g, M) = M, we set (of)(x) = xf. A classical
computation using the Jacobi identity and the definition of a g-module
gives the following.

Lemma XVIII.I.I. If f is in en (g, M), then of is in en+l (g, M). More-

over, 0 0 0 = O.
Let us denote the kernel and the image of 0 in en(g, M) by zn(g, M)
and Bn(g, M) respectively. An element of zn(g, M) is called an n-cocycle
whereas an element of Bn(g, M) is called a n-coboundary. Lemma 1.1 im-
plies that Bn(g, M) is a vector subspace of zn(g, M). This allows us to
consider the quotient space

(1.3)

which is called the n-th cohomology group of the Lie algebra 9 with coeffi-
cients in the g-module M.
Let us describe Hn(g, M) in degree n = 0,1,2. In degree 0 we have

HO(g,M) = ZO(g,M) = {m E M I gm = O}.

A linear map f : 9 --* M is a l-cocycle if and only if

f([x, V]) = xf(y) - yf(x) (1.4)

for all X, y E g. In other words, a l-cocycle is a derivation from 9 to M. It is

a l-coboundary if and only if it is an inner derivation, i.e., there exists an
element m in M such that f(x) = xm for all X in g. Thus, the cohomology
group HI (g, M) classifies all derivations up to inner derivations.
In degree 2 an antisymmetric bilinear map f : 9 x 9 --* M is a co cycle if
and only if

xf(y, z)+yf(z, x)+zf(x, y)- f([x, V], z)- f([y, z], x)- f([z, x], y) =0 (1.5)

for all x, y, z E g. It is a coboundary if and only if there exists a linear map

a : 9 --* M such that for all x, y we have

f(x,y) = xa(y) - ya(x) - a([x,y]). (1.6)

422 Chapter XVIII. Cohomology and Rigidity Theorems

We shall see in the next section that 2-cocycles appear when we "deform"
Lie algebras and their enveloping algebras.
The second cohomology group H2(g, M) has also an interpretation in
terms of extensions of g. These are defined as follows. Let 9 be a Lie algebra
and M be a left g-module. An extension of the Lie algebra 9 with kernel
M is a Lie algebra ~ together with a surjective morphism p : ~ --+ 9 of Lie
algebras such that
(i) the kernel of p (which is a Lie ideal in ~) is M, and
(ii) for any x E ~ and m E M, we have
[x,m] = -[m,x] =p(x)m. (1. 7)

Such an extension is split if there exists a morphism s : 9 --+ ~ of Lie

algebras such that p 0 s = id g . The following relates extensions of 9 to
cohomology.
Proposition XVIII.1.2. If H2(g, M) = 0, then any extension of 9 with
kernel M is split.
PROOF. Let us decompose the vector space ~ as 9 EB M. By definition of
an extension, the Lie bracket on ~ = 9 EB M is necessarily of the form

[(x, m), (y, n)] = ([x, y], xn - ym + f(x, y)) (1.8)

where x, y E g, m, n E M, and f is a bilinear map from 9 x 9 into M. Since

a Lie bracket is antisymmetric, f has to be antisymmetric. The Jacobi
identity for the bracket (1.8) forces another condition on f which is nothing
else than Relation (1.5). In other words, f is a 2-cocycle with values in M.
By hypothesis, f is a 2-coboundary, which means that there exists a linear
map a : 9 --+ M such that (1.6) holds. Define the linear map s = (id, -a)
from 9 to ~ = 9 EB M. We have pos = id g . Let us check that s is a morphism
of Lie algebras, which will show that the extension is split. We have

[s(x), s(y)] [(x, -a(x)), (y, -a(y))]

([x, y], -xa(y) + ya(x) + f(x, y))
([x, y], -a([x, yD)
s([x, YD.
We used (1.6) in the third equality. o
In the next section we shall need the following corollary to Proposition
1.2. Let us consider a Lie algebra 9 and a U(g)-bimodule M, i.e., a vector
space M with a left and a right action of U(g) such that (u 1 m)u 2 = U 1 (mu 2 )
for all U 1 , U 2 E U(g) and m E M. We denote by M the vector space M
equipped with the left g-module structure defined by X m = xm - mx for
all x E 9 and m E M.
 XVIII.1 Cohomology of Lie Algebras 423

Corollary XVIII.lo3. Let f : U(g) x U(g) -+ M be a bilinear map such

that for all x, y, z in U(g), we have f(l, x) = f(x, 1) = 0 and

xf(y, z) - f(xy, z) + f(x, yz) - f(x, y)z = O. (1.9)

Then, if H2(g, M) = 0, there exists a linear map a : U(g) -+ M such that

a(l) = 0 and
f(x, y) = xa(y) - a(xy) + a(x)y (1.10)
for all X,y E U(g).
PROOF. We define a product on U(g) E9 M by

(x, m)(y, n) = (x y, xn + my + f(x, y)) (1.11)

where x, y E U(g) and m, n E M. Relation (1.9) implies that this product

is associative. It has a unit which is (1,0). We get a Lie bracket on the
same space by taking the commutator

[(x, m), (y, n)] (x, m)(y, n) - (y, n)(x, m)

( [x, y], xn - nx + my - ym + f (x, y) - f (y, x) ) .

The subspace 9 = g E9 M is a Lie sub algebra of U(g) E9 M. The first pro-

jection p from 9 onto g is a surjective morphism of Lie algebras. An easy
computation shows that the kernel of this extension is the g-module All.
Since H2(g, M) = 0, we know by Proposition 1.2 that the extension 9 -+ g
is split. Thus, there exists a morphism of Lie algebras 8 : g -+ 9 such
that po 8 = id g . Composing it with the inclusion of 9 into U(g) E9 M,
we obtain a morphism of Lie algebras 8' : g -+ U(g) E9 M which, when
composed with the first projection, is the inclusion of g into its enveloping
algebra. By Theorem V.2.1, 8' extends to a morphism of algebras 0" from
U(g) into U(g) E9 M splitting the firs~ projection. This map is necessar-
ily of the form 0" = (id, -a) where a is a linear map from U(g) to M.
Let us express the fact that 0" is a morphism of algebras. First, we have
(1,0) = 0"(1) = (1, -a(I)), which implies that a(l) = O. Next,

(x, -a(x))(y, -a(y))

(Xy, -xa(y) - a(x)y + f(x, y))
IJ(xy)
(xy, -a(xy)),

from which we derive f(x, y) = xa(y) - a(xy) + a(x)y. o

424 Chapter XVIII. Cohomology and Rigidity Theorems

XVIII. 2 Rigidity for Lie Algebras

We now use the cohomology groups introduced in Section 1 to derive two
classical theorems on topological algebras. The first one is a uniqueness
theorem.

Theorem XVIII.2.1. Let fI and fI' be Lie algebras. Suppose given two
morphisms a and a' of topological algebras from U(fI) [[hll to U(fI')[[h]] such
that a == a' modulo h. If H1(g, U(g')) = 0, there exists an invertible element
F E U(fI')[[hll with F == 1 modulo h such that a'(x) = Fa(x)F-1 for all
x E U(fI)[[hll.
The class modulo h of a (and of a') is an algebra morphism a o from U(g)
to U(g'). We give U(fI') a left I-module structure by setting xu = [ao(x),u]
where x E I and U E U(g'). The cohomological condition in Theorem 2.1
refers precisely to this module structure.
PROOF. Since a is C[[hll-linear, it is determined by its restriction on U(g).
Write the latter in the form

a(x) = L an(x)h n (2.1)

n20

where (an)n is a family of linear maps from U(g) to U(g'). The map a
preserves the unit, which implies that a o(1) = 1 and an (1) = 0 if n > O. It
also preserves the product, which is equivalent to the relations

(2.2)

and
an(xy) = L ap(x)aq(y) (2.3)
p+q=n
if n > O. In particular, we have

(2.4)

Suppose now that x and yare elements of fl. Then Relation (2.4) implies
that
(2.5)
In view of our definition of the fI-action on U (fI') and of (1.4), we see that
a 1 is a 1-cocycle of I with values in U(g'). Since H1(g, U(g')) = 0, the map
a is a 1-coboundary, which means that there exists an element u 1 E U(g')
such that
(2.6)
for all x E g. Set

(2.7)
 XVIII.2 Rigidity for Lie Algebras 425

where x E U(g). This extends C[[h]]-linearly to a new morphism of topo-

logical algebras from U(g)[[h]] to U(g')[[h]]. Modulo h2, we have

a(1)(x) == ao(x) + (ulao(x) - ao(x)u l + al(x))h == ao(x)

in view of (2.6-2.7). This holds for all x E g. Since a o and a(1) are algebra
morphisms defined on U(g), it holds for all elements of the enveloping
algebra. Set a(1) = L:n>O a~l)hn. The previous computation shows that
a6l ) = a o and ail) = o. -
Now apply (2.3) to a(1) and n = 2. We have

a~l)(xy) = ao(x)a~l)(y) + a~l)(x)ao(Y),

which shows that the restriction of a~l) to 9 again is a 1-cocycle with
values in U(g'). For the same reasons as above, there exists an element
u 2 E U(g') such that a~l)(x) = [a O(x),u 2] for all x E g. Set a(2)(x) =
(1 + u 2h2)a(1)(x)(1 + u2h2)-1. A computation as above shows that
(2.8)
Proceeding by induction, we similarly construct elements u 3 , u 4 , ... in U(g')
such that
Una(x)U;:l == ao(x) mod hn+ l (2.9)
for all n > 0 and all x E U(g). Here Un is defined by
Un = (1 + u nhn)(l + un_lh n- l ) ... (1 + u l h).
When we pass to the inverse limit, we see that the family (Un)n defines an
invertible element U E U(g')[[h]] such that U == 1 modulo hand
a(x) = U-1ao(x)U (2.10)
for all x E U(g).
Now let us prove Theorem 2.1. Proceeding as for a, we get an element
U' in U(g')[[h]] such that U' == 1 modulo hand a'(x) = U'-la~(x)U'.
By hypothesis, a~ = ao' From this and from (2.10) we need only to set
F = U,-lU in order to complete the proof. 0

We now consider a topological algebra (A, IL, "I), as defined in XVI. 4 ,

satisfying the following conditions:
(i) as an algebra, A/hA is the enveloping algebra of a complex Lie alge-
bra g:
A/hA = U(g), (2.11)
(ii) as a C[[h]]-module, A is topologically free, i.e.,
A = U(g)[[h]], (2.12)
(iii) the unit "1(1) of A is equal to the constant formal series 1 in U(g) [[h]]
under the identification (2.12).
426 Chapter XVIII. Cohomology and Rigidity Theorems

The second theorem of this section is an existence theorem. Since it states

that any such topological algebra is isomorphic to the trivial topological
algebra associated to U(g), Theorem 2.2 is called a rigidity theorem.

Theorem XVIII.2.2. Under the previous hypotheses, if H2(g, U(g)) is

zero, there exists an isomorphism a : A - t U(g)[[h]] of topological algebras
inducing the identity AI hA - t U(g) modulo h.
Here 9 acts on U(g) by the adjoint representation.
PROOF. We first proceed as in XVI.5 by identifying the C[[h]]-module A
with U(g) [[h]] and by expanding in a formal series the C[[h]]-linear map tL
from A0A = (U(g) U(g))[[h]] to A = U(g)[[h]L i.e.,

tL = L tLn hn (2.13)
n;::O

as in (XVI.5.3), where (tLn)n is a family of bilinear maps from U(g) x U(g)

to U(g) such that tLo is the multiplication of the enveloping algebra of g.
Condition (iii) above can be reformulated as

(2.14)

for all x E U(g) and all n > O. The associativity of the product tL is
expressed by
(2.15)
for all x, y, z E U(g). Expanding tL with (2.13), we obtain the equivalent
system of equations

for all x, y, z E U(g) and all n ~ O. Let N be the smallest integer n > 0
(if it exists) such that tL n i- O. If no such integer exists, we have tL = tLo,
which means that A coincides with U(g)[[h]] as a topological algebra and
the theorem is proved. If N exists, let us rewrite (2.16) for n = N. Using
the customary notation for the product in U(g), we get

tLN(XY, z) + tLN(X, y)z = tLN(X, yz) + XtLN(y, z) (2.17)

for all x, y, z E U(g). In other words, tLN satisfies Condition (1.9) of Corol-
lary 1.3 with M = U(g). Since H2(g, U(g)) vanishes, we may apply Corol-
lary 1.3, which yields a linear endomorphism aN of U(g) with aN(l) = 0
and
(2.18)
for all x, y E U(g). Define a C[[hll-linear automorphism a of U(g)[[h]] by
 XVII1.3 Vanishing Results for Semisimple Lie Algebras 427

its inverse being given by 2:n>O (-l)no:'JvhnN. We have 0:(1) = 1. Define a

new product 11' = 2:n?:O l1~hn on A by

(2.19)

Since 0: == id mod h N , we have 11' == 11 mod hN. Let us compute 11' modulo
h N + 1 . Relation (2.18) implies that

11'(X, y) XY + 11~(x, y)h N

(id + O:NhN)((l1o + I1 N h N )(x - O:N(x)h N , Y - O:N(y)hN))
xy + (O:N(XY) + I1N(x, y) - O:N(X)Y - xO:N(y))h N
xy mod h N + 1 .

Consequently, 11~ = 110 is the multiplication of U(fJ) whereas

11~ = ... = 11~ = O. (2.20)

We use this procedure to construct an isomorphism of algebras between

A and U(fJ)[[h]]. In effect, applying the above considerations to the case
N = 1, we get an isomorphism of the form id + 0: 1 hI from the algebra A to
A equipped with a new product 11(1) such that I1l= O. Applying now to
l)

11(1) and N = 2, we get an isomorphism id+0: 2 h 2 from (A, 11(1)) to (A,I1(2))

where 11(2) is a product with I1l2) = 11~2) = O. Repeating this infinitely many
times and composing all the isomorphisms, we get an isomorphism 0: from
A to A endowed with a product /-L(oo) satisfying /-L~oo) = 0 for all n > O. In
other words, /-L(oo) = /-Lo is the usual product of U(fJ) [[hll D

XVIII.3 Vanishing Results for Semisimple Lie

Algebras
We shall use Theorems 2.1 and 2.2 in Section 4 in case fJ is a finite-
dimensional complex semisimple Lie algebra. To apply them, we have to
prove the vanishing of Hi(fJ, U(fJ)) for i = 1,2. We start with the following
result.

Proposition XVIII.3.1. If fJ is a finite-dimensional complex semisimple

Lie algebra and M is a finite-dimensional non-trivial simple left fJ-module,
then Hn(fJ, M) = 0 for all n ?: O.

Here non-trivial means that M is not the one-dimensional fJ-module on

which the Lie algebra acts by zero.
PROOF. We need the Casimir element C = 2:k Xkxk defined by (XVIL1.5).
We know that C acts on any finite-dimensional non-trivial simple fJ-module
428 Chapter XVIII. Cohomology and Rigidity Theorems

by a non-zero scalar. In order to prove the proposition, we construct for all

n a map h : Cn(g, M) --+ C n - 1(g, M) such that

Cf = bhf + hbf (3.1)

for all f E C"(g, M). By C f we mean the n-linear map defined by

where Y1'" ., Yn belong to g. Let f be an n-cocycle with values in AI, i.e.,

such that bf = O. By (3.1) we get Cf = b(hf), which means that Cf is a
n-coboundary. Since C acts by a non-zero scalar on IVI, we see that f too
is a coboundary. This proves the vanishing of Hn(g, IVI).
We are left with building a map h satisfying (3.1). Given f E C"(g, M)
with n > 0 and the Casimir element, we define an antisymmetric (n - 1)-
linear map hf with values in IVI by

(hf)(Y1""'Yn-1) = L x k f(X k 'Y1,,Yn_1) (3.2)

k

for all Y1"'" Yn -1 E g. If f E CO(g, M), set hf = O. Using (3.2) and (l.2),
we get

(bhf + hbf)(Y1"'" Yn ) = Cf(Y1"'" Yr,) + L (_I)i Zi

l::;,::;n

where

Zi = L ([XklYi]J(X k 'Y1, ... ,Yi, ... ,Yn)+Xkf([X k'Yi],Yl, ... ,Yi, ... ,Yn))'
k

Relation (3.1) will be proved if we show that all Zi vanish. Using the linear
forms ak and (3kf of XVII.l, we get

Zi L(a kP (Yi)xd(X k ,Y1, ... ,Yi, ... ,YrJ

k,

Exchanging k and in the second summand, we obtain

Zi = L (ak(Yi) + (3k(Yi)) xd(xk, Y1,'" , Yi, .. . ,YrJ,

k,

which vanishes in view of Lemma XVII.l.l. D

 XVIII.3 Vanishing Results for Semisimple Lie Algebras 429

As a consequence, we get the so-called "Whitehead lemmas" .

Corollary XVIII.3.2. If g is a semisimple Lie algebra and M is any
finite-dimensional left g-modl1le, then HI(g, M) = H2(g, M) = o.
PROOF. We know that any finite-dimensional module M over a semisimple
Lie algebra is a direct sum M = EEli !vIi of simple modules Mi. Since the
complex C*(g,!vI) is the direct sum of the subcomplexes C*(g, Ali) , we
have Hn(g, !vI) = ECli Hn(g, !vIJ. In view of Proposition 3.1, it is enough
to prove Corollary 3.2 when !vI is the trivial one-dimensional g-module C.
(a) We first prove the vanishing of HI(g, C), which will imply the vanish-
ing of HI (g, M) for all finite-dimensional modules M. Let f be a 1-cocycle
with values in the trivial module C. Relation (1.4) reduces to f([x, y]) = 0
for all x, y E g. Now Serre's relations (XVII. 1.9-1. 10) show that the ele-
ments [x, y] span the vector space g. Therefore f = 0 on the whole space g.
(b) The argument for the vanishing of H2 (g, C) is slightly more involved.
We first claim that if f is a 2-cocycle with values in C, then the linear map
f given by J(x)(y) = f(x,y) for all X,y E g, is a 1-cocycle of g with
values in the dual vector space g*. Such a statement presupposes that we
have defined a left action of g on g*. This is done by taking the coadjoint
representation given by

(xa)(y) = a([y, x]) (3.3)

where :r, y E g and 0 E g*. Indeed, if f is a 2-cocycle with values in C, we

have by (1.5)
f([x, y])(z) = J(y)([z, x]) - J(x)([z, y])
for all x, y, z E g. Reformulating this with (3.3), we get

f([x, y]) = xJ(y) - yJ(x),

which shows that f is a 1-cocycle with values in the fini~e-dimensional g-

module g*. By the first part of Corollary 3.2, the co cycle f is a coboundary,
i.e., there exists a linear form a E g* such that f(x) = xa. We thus get

f(x, y) = J(x)(y) = (xa)(y) = a([y, x]) = -a([x, y]).

In other words, the 2-cocycle f is the coboundary of a. This completes the
proof of the vanishing of H2. Observe that, incidentally, we proved that
H2(g, C) ~ HI(g, g*). 0

Let us equip U(g) with the adjoint representation of g for which the Lie
algebra acts on U(g) on the left by Xl1 = Xl1-l1X = [x,u] where x E g and
11 E U(g). If u = Xl ... xn with xl' . .. ,x n belonging to g, an easy induction
shows that
n
X.U = L Xl .. Xi - l [x, xi]xi+l ... Xn (3.4)
'i=l
430 Chapter XVIII. Cohomology and Rigidity Theorems

We record the following corollary.

Corollary XVIII.3.3. Let 9 be a finite-dimensional complex semisimple
Lie algebra acting on U(g) as above. Then Hl(g, U(g)) = H2(g, U(g)) = O.
PROOF. We use the symmetrization map TJ : S(g) ---+ U(g) defined in V.2
by

where xl' ... 'X n E g. We know that TJ is a linear isomorphism. Moreover,

if we equip S(g) with the left g-module structure given by
n

X (Xl Xn) = LXI ... Xi_I[X, Xi]X Hl Xn (3.5)

i=l

the map TJ becomes an isomorphism of g-modules. Now, as can easily be

seen from (3.5), the action of 9 respects the decomposition of S(g) into its
homogeneous components sn(g). We thus obtain an isomorphism

U(g) ~ EB sn(g)
n::o-O

of g-modules. Consequently, for i = 1,2 we have

Hi(g, U(g)) = EB Hi(g, sn(g)) = 0
n::o-O

by application of Corollary 3.2 to the finite-dimensional modules sn(g).

o

XVIII. 4 Application to Drinfeld-Jimbo Quantum

Enveloping Algebras
Let 9 be a finite-dimensional complex semisimple Lie algebra and let Uh (g)
be the Drinfeld-Jimbo quantum enveloping algebra of XVII.2. The first
three sections of this chapter culminate in the following result.
Theorem XVIII.4.1. There exists an isomorphism a: Uh(g) ---+ U(g)[[h]]
of topological algebras which is congruent to the identity modulo h. If a' is
another such isomorphism, there exists an element Fin U(g) [[h]] such that
F == 1 modulo hand a'(a) = Fa(a)F-I for all elements a of Uh(g).
PROOF. The first statement is a direct consequence of Theorem 2.2 and of
the vanishing of H2(g, U(g)) proved in Corollary 3.3.
As for the second one, observe that a' 0 a-I and the identity are two
automorphisms of the topological algebra U(g) [[h]] inducing the identity on
 XVIII.5 Cohomology of Coalgebras 431

U(g). By Theorem 2.1 and by the vanishing of H1(g, U(g (see Corollary
3.3), there exists an element F == 1 mod h in U(g)[[h]] such that we have
(a' 0 a- 1)(u) = FuF- 1 for all u E U(g) [[h]]. Replacing u by a(a) yields
the conclusion. 0

Since conjugated morphisms restrict to the same map on the centre, we

get the following important result which we have already used in XVII.3
in order to define the quantum Casimir element Gh in the Drinfeld-Jimbo
algebra Uh(g).
Corollary XVIII.4.2. There exists a unique isomorphism a oj topological
algebras Jrom the centre oj Uh (g) to the centre oj U(g)[[h]] such that a == id
modulo h.
We may use the isomorphism a between Uh(g) and U(g) [[h]] to assign a
topologically free Uh (g)-module to any finite-dimensional g-module. Indeed,
let V be such a g-module. We equip V[[h]] with the extended U(g)[[h]]-
module structure. Define 11 as V[[h]] equipped with the Uh(g)-module
structure given by a v = a(a)v where a E Uh(g) and v E V[[h]]. Since
a is congruent to the identity modulo h, we see that 11/ h11 is isomorphic
to V as a g-module. It can be shown that, when VA is the simple g-module
associated to the dominant weight A, then VA is the highest weight Uh(g)-
module alluded to at the end of XVII.2.

XVIII. 5 Cohomology of Coalgebras

In order to prove the second rigidity theorem of the chapter, we set up
a cohomology theory for coalgebras. Let (G,~, E, 1) be a coalgebra over a
field k with an element 1 in G such that ~(1) = 101 (which implies that
E( 1) = 1). Clearly, any bialgebra satisfies these conditions with 1 equal to
the unit of its multiplication.
We set Tn(G) = Gem if n > 0 and TO(G) = k. Define linear maps
8~, ... , 8~+1 from Tn(G) to Tn+1(G) by

8~(X100xn) 10x 1 00x n ,

8~+1(X1 00 x n ) xl 00 xn 0 I,
8~(x1 00 Xn ) xl 00 Xi _ 1 0 ~(Xi) 0 Xi+1 00 xn
if 1 -<; i -<; n. If n = 0, we set 88(1) = 86(1) = 1.
Lemma XVIII.5.1. We have 8~+18~ = 8~+18~-1 Jor all integers i,j such
that 0 -<; i < j -<; n + 2.
PROOF. If j ::::: i + 2, this is straightforward. Let us concentrate on the case
j = i + 1. If 1 -<; i -<; n, we have
8~-t;118~!(X10 0x n ) = x 10 0xi_10(idc0~)(~(xi0xi+10 0x n
432 Chapter XVIII. Cohomology and Rigidity Theorems

On the other hand, we have

6~+16~ (Xl Q9 ... Q9 Xn) = Xl Q9 ... Q9 Xi - l Q9 (~Q9 ide )(~(Xi)) Q9 x'i +1 Q9' .. Q9 x n

Both are equal in view of the coassociativity of ~.

If i = 0, we have

~(l)Q9XIQ9"'@Xn
1 Q9 1 @ Xl Q9 .. Q9 Xn
6~+16~ (Xl Q9 ... @ XrJ

Similar computations prove the desired relations in the remaining cases.

o
As a consequence we can equip T- (C) with a differential of degree +1.
Corollary XVIII.5.2. Define the differential 6 : Tn(c) --+ Tn+1(c) by
6 = L~~Ol (-1)i6~. Then 6 a 6 = O.
The co chain complex (T- (C), 6) is called the cobar complex of the coal-
gebra C.
PROOF. In degree n, we have

606
j=O i=O

i<j j<O,i

,",(-1)i+j(6
L.." n+1 6n _6 n+l 6n-
j i i j l)

i<j
o
by Lemma 5.1. o
The natural isomorphisms Tn(c) Q9 TTn(C) ~ Tn+Tn(C) induce an asso-
ciative graded product on T(C) = EBn>O Tn(c). This product is compati-
ble with the differential 6 in the follOWIng sense.
Lemma XVIII.5.3. If wE Tn(c) and w' E TTn(C), then

6(ww') = 6(w)w' + (-1)"w6(w') (5.1)

for the product ww' in Tn+Tn(C).

We shall use the term differential graded algebra for a graded algebra
with a differential satisfying Relation (5.1). It follows from Lemma 5.3
that the product on T- (C) induces an associative graded product on the
cohomology H- (T- (C), 6) of the co bar complex.
 XVIII.5 Cohomology of Coalgebras 433

PROOF. This results from the relations 6:,+rn (ww' ) = 6~ (w )w' if i ~ n, from
d-n( W ') 1'f Z. >
+ m ( WW ') -- wUrn
u>:in _ n + 1 ,an d f rom

6~+1 (w)w' = w Q9 1 Q9 Wi = w6~ (Wi).

Suppose that C possesses an involution x f-+ x such that 1 = 1 and

,6,(x) = ,6,oP(x), e.g., C is a cocommutative coalgebra with involution equal
to the identity. Then we can put an involution on the complex (Te (C), 6)
as well. Define an automorphism O'n of Tn(c) by 0'0 = id k and by

if n > O. The automorphism O'n is an involution.

Lemma XVIII.5.4. We have 60'n = O'n+16.

PROOF. It is enough to prove that 6~O'n = (_1)n+IO'n+16~+1-i. If i = 0,
we have

On the other hand,

which is the same since (n+I)2(n+2) - (n + 1) = n(n2+1). If 1 ~ i ~ n, we

have

xn)
6~O'n(XI Q9 ... Q9
(_1)n(n+I)/2 xn Q9 ... Q9 ,6,(x n + I -.J Q9 ... Q9 Xl
(_1)(n+l)(n+2)/2-(n+l) Xn Q9 .. Q9 ,6,0P(X n +l_ i ) Q9 ... Q9 Xl
(-1)"+IO'n+16~+I-i(XI Q9 ... Q9 XrJ
The second equality holds by the assumption on ,6,. D

As a consequence of Lemma 5.4, the cochain complex (Te (C), 6) is the

direct sum
T e (C) = T~ (C) ttl T~ (C) (5.3)
of the sub complexes (T~(C), 6) and (Te(CL, 6) defined for all n by

T(C) = {w E Tn(c) I O'n(w) = w}. (5.4)

434 Chapter XVIII. Cohomology and Rigidity Theorems

XVIII. 6 Action of a Semisimple Lie Algebra on

the Cobar Complex
We return to the situation of a cocommutative coalgebra (C,~, E) with an
element 1 such that ~(1) = 1 1. Assume we also have a Lie algebra 9
acting on C such that, if we denote by x . c the action of an element x E 9
on an element c E C, we have x . 1 = 0 and

~(x c) = ~(x) . ~(c) = L (x. c' C" + c' X Cll) (6.1)

(c)

in Sweedler's sigma notation. The examples we have in mind are the coal-
gebras U(g) and 8(g), on which 9 acts by the adjoint representation.
Equipping the tensor powers of C with the induced g-module structures,
we get the following result.

Lemma XVIII.6.1. The cobar complex (Te(C),/5) is a complex made up

of g-modules.

PROOF. It suffices to check that the maps /5~ of Section 5 are maps of
g-modules. Let c l , ... , cn be elements of C and x be in g. For /5~, we get
n

L /5~ (c l ... x . ck ... cn)

k=l
n

L 1 c l ... x . c k ... cn
k=l
X . /5~ (c l ... cn)

since x . 1 = O. There is a similar proof for /5~+1. If 1 ::; i ::; n, we have

n

L /5~ (c ... x . ck ... c

l n )
k=l

L c ... ck ... ~(ci) ... c

l X n
k#i
+ cl ... ~ (x . c i ) ... c n
x . (c l ... ~ (c i ) ... c n )

x . /5~ (c l ... cn )

by (6.1). o
Observe also that the subcomplexes (T(C), /5) are preserved by the g-
action where C is equipped with the identity involution.
We next restrict to the case when 9 is a finite-dimensional semisimple
Lie algebra acting on C such that C is a direct sum of finite-dimensional
 XVIII.7 Computations for Symmetric Coalgebras 435

fl-modules. This is the case for C = 5(fl) and, hence, for the isomorphic
coalgebra U(fl). For any fl-module V, we define a fl-submodule V9 by

V9 = {v E V I x v = 0 \;fx E fl}.

Elements of V9 are called fl-invariant. The linear span flV of the elements
x . v where x runs over fl and v over V is also a fl-submodule of V.

Proposition XVIII.6.2. Under the previous hypotheses, each of the com-

plexes (TJC),6) is the direct sum of the respective subcomplexes (T(C)9)
and flT (C).'
T(C) = T(C)9 EEl gT(C).

PROOF. Since the constructions V f---+ V9 and V f---+ 9 V are functorial, it is

clear that T (C)9 and gT (C) are sub complexes of T (C). Therefore, in
order to prove the proposition, it suffices to check that

V = V9 EElflV (6.2)

for all g-modules V that are direct sums of finite-dimensional fl-modules.

Since 9 is semisimple and Equality (6.2) is preserved by the direct sum of
g-modules, it is enough to check (6.2) when V is finite-dimensional and
simple. If V is the trivial one-dimensional module, then 9 acts by zero,
which implies that V = V9 and gV = O. If V is a non-trivial simple
module, it corresponds to a dominant weight A -I- O. Let v be a highest
weight vector for V. Since A -I- 0, there exists an element Hi of fl such that
Hiv = A(Hi)v -I- O. Consequently, v does not belong to V9 and V9 -I- V.
Since V is simple, the submodule V9 has to be zero. On the other hand,
the same relation shows that fl V -I- O. We again appeal to the simplicity of
V, now obtaining gV = V. In both cases, we get (6.2). D

XVIII. 7 Computations for Symmetric Coalgebras

We assume in this section that k is a field of characteristic zero. We now
compute the cohomology of the complexes (T e (C), 6) and (T (C), 6) in the
special case when C is the symmetric bialgebra C = (5(V),~, E) where V
is a finite-dimensional vector space over k and

~(v)=vl+lv and E(V)=O (7.1)

for any element v of V.

Theorem XVIII.7.1. Under the previous hypotheses,

(a) there exists a unique map f-L : (Te(5(V)),6) -+ (A-(V),O) of differ-
ential graded algebras where the exterior algebra A- (V) is given the zero
436 Chapter XVIII. Cohomology and Rigidity Theorems

differential, such that the restriction of p, to T1 (S (V)) = S (V) is the pro-

jection onto the direct summand Sl (V) = V = A1 (V), and the induced
map p,. : H(r-(S(V)),b) ----+ A(V) on cohomology is an isomorphism.
(b) The antisymmetrization map a : An (V) ----+ Tn (S (V)) given by

a(v 1 1\ ... 1\ vn ) = L c(a)v u (l) ... Vo-(n)

uESn

for all V1, . .. , Vn ' is a map of complexes, i.e., boa = O. Furthermore, we

have p,(a(w)) = n!w fOT all wE An(v).
(c) The map p, induces isomoTphisms

and

in cohomology fOT all n 2: O.

The rest of the section is devoted to the proof of Theorem 7.1. The idea
is to dualize the complex (T(S(V)), b) and to compute the homology of
the dual complex. For a definition of the exterior algebra, see II, Exercise 6.
We first need the concept of a graded dual vector space: if V = ffin>o Vn
is a vector space with a positive grading, we define the graded dual vector
space of V by
~~ = EB V;. (7.2)
n2:0

We can apply this to the vector spaces T(V), S(V), A(V) and T(S(V))
with their natural gradings. If V = ffin>o Vn and W = ffin>o Wn are
vector spaces with gradings, we may consider their tensor prodllct V W
graded by
(VW)n= VpWq' EB (7.3)
p+q=n
Lemma XVIII.7.2. Suppose V = ffin>o v" and W ffin>o Wn have
gmdings fOT which v" and Wn aTe finite-dimensional fOT all n.- Then theTe
is a canonical isomoTph'lsm

PROOF. This is straightforward. It uses the fact proved in II.2 that this
isomorphism holds for finite-dimensional vector spaces. 0

Let (C, ~, c, 1) be a graded coalgebra with unit, meaning that the under-
lying vector space C = ffin>o C n has a grading, that ~ and c are graded
maps (we equip k with the-trivial grading concentrated in degree zero),
 XVIII. 7 Computations for Symmetric Coalgebras 437

and that 1 belongs to Co. If Cn is finite-dimensional for all n - which we

assume henceforth - , then it is clear that A = C;r is an algebra graded by
An = C~, with multiplication given by the transpose .6.* of .6., with unit
given by c*, and with a map of algebras c : A ----* k defined by c(a) = a(l).
The map c will be called the augmentation of A.
From the formula for 8, we see that the complex (Te (C), 8) of Section 5
is a complex with a grading. Applying Lemma 7.2, we get

(7.4)

Lemma XVIII.7.3. The transpose d = 8* of 8 is given by

d(al @ ... @ an) = C(al)a 2 @ ... @ an

n-l
+L (-l)ia l @ ... @ ai- l @ aiai+l @ ai+2 @ ... @ an
i=l

for all elements aI' ... ,an of A.

The chain complex (Te(A), d) is called the bar complex of the augmented
algebra A.
PROOF. Apply both sides to Xl @ ... @ xn where Xl' ... ,xn belong to C.
D

Lemma XVIII. 7.4. Under the previous hypotheses, the cohomology of

(r-(C), 8) is the graded dual of the homology of (Te(A),d).
PROOF. This is a consequence of the fact that the duality functor is exact
and that biduality is a natural isomorphism on finite-dimensional vector
spaces. D

In order to prove Theorem 7.1, it therefore suffices to compute the ho-

mology of the chain complex (Te(S(V);r)' d). We first identity the algebra
S(V);r
Lemma XVIII.7.5. If V is a finite-dimensional vector space over a field
k of characteristic zero, then the graded dual of the graded coalgebra with
unit S(V) is the graded augmented algebra S(W) where W = V* is the
dual vector space of V.

PROOF. Let {VI' ... ,VN} be a basis of V. Then {vr1 ... V~N} <>1 +.+<>N=n
is a basis of sn(V). We define a basis {wr 1 ... w~N}<>1+.+<>N=n of sn(V)*
by

(7.5)
438 Chapter XVIII. Cohomology and Rigidity Theorems

The product map * : sn(V)* Q9 sm(v)* --t sn+m(v)* on S(V);r is by

definition the transpose of the comultiplication ~. We have

< "'N)
( WI"'1 . "WN * (131
WI"
13 N ) 'Y1
'WN ,VI" 'VN
'Y N >
< WI"'1 "'N fQ.. 131
. "WN 'CYWI . "WN
f3N A( 'Y1
,L.l. VI
'YN)
.. 'VN
>
< WI"'1 "'N
... W N fQ..
'CY
f3N , ( VI
WI131 ... W N fQ..
'CY
1 + 1 'CY
fQ..
VI )'Y1 ...

(VN 1291+ 1I29VN)'YN >

131 f3N 'Y1- i 1 'YN-iN

< WI" 'WN ,VI . "VN >
8"'1 +131 m ... 8"'N+f3N,'YN 0: 1 !,Bl! ... 0: N!,BN!

This proves that

(7.6)

which shows that the product on S(V);r is the product of the symmetric
algebra S(V*). The rest of the proof is left to the reader. D

We next compute the homology of the chain complex (T-(S(W)), d).

Define a linear map 0: from A-(W) to T-(S(W)) by

0:( WI 1\ ... 1\ w n ) = L ( a )Wa(l) Q9 ... 129 wa(n) (7.7)

aESn

where wI"'" Wn E Wand (a) is the sign of the permutation a.

Proposition XVIII.7.6. We have do 0: = 0 and also that the induced

map 0:_ : An(w) --t Hn(T-(S(W)),d) is an isomorphism for all n 2: o.
If PI : S(W) --t W is the natural projection onto the direct summand
W = Sl (W) and f-l is defined as the composite map
n
f-l : r(S(W))~w0n --t An(W),

then f-l is a chain map. We have (f-l oo:)(w) = n!w for all elements w
belonging to An(w).

PROOF. We proceed in six steps using the terminology and the results of
the Appendix.
 XVIII. 7 Computations for Symmetric Coalgebras 439

1. We first observe that the bar complex (T(A), d) is obtained from a

complex (T~(A), d') ofleft A-modules by

(7.8)

where T~(A) = A Q9 A@n and the left A-linear differential d' is given by
n-1
d'(a o Q9 ... Q9 an) = L (-l)ia o Q9 .. Q9 a i - 1 Q9 a i a i +1 Q9 ai+2 Q9 .. Q9 an
i=O

(7.9)
for an, ... ,an EA. The reader may easily check that d' 0 d' = O.
2. We claim that the complex (T' (A), d') is a resolution of k by free left
A-modules. It suffices to prove that the complex

... ~T~(A)~T{(A)~TMA)~k -+ 0

is acyclic. Define s : T~ (A) -+ T~+l (A) by

s( ao Q9 .. Q9 an) = 1 Q9 a o Q9 .. Q9 an (7.10)

and s : k -+ TMA) by s(l) = 1. An easy computation shows that

d's + sd' = id (7.11)

on all T;,(A), which proves the acyclicity of (T'(A), d').

3. In case A is the symmetric algebra S(W), there is another resolution
of k by free left A-modules: it is the Koszul resolution (K.(W), 8) defined
by Kn(W) = S(W) Q9 An(w) and
n
8(a Q9 W1 1\ ... 1\ w n ) = L (-1)i+ 1 awi Q9 W 1 1\ ... 1\ Vi, 1\ ... 1\ wn (7.12)
i=l

where a E S(W), w 1 , ... ,wn E Wand where the hat on wi again means
that we omit this element. Check that 808 = o.
We claim that (K.(W), 8) is a resolution of k. This again is due to the
existence of a homotopy: define a map h : Kn (W) -+ Kn+l (W) by
m

h( W 1 ... Wm Q9 w) = L W 1 ... Vi, ... Wm Q9 Wi 1\ W (7.13)

i=O

where W 1 , ... ,wm in Wand w in An (W). Then we have

(8h + h8)(P Q9 w) = (m + n)(P Q9 w) (7.14)

for all P in sm(w) and win An(w). Relation (7.14) shows the acyclicity
of the Koszul resolution in degree> O. As for degree 0, observe that the
440 Chapter XVIII. Cohomology and Rigidity Theorems

cokernel of the map 8 : KI(W) ---- Ko(W) given by 8(a w) = aw is

isomorphic to k.
Since c:(w) = for all W E W, (k S(W) K.(W),id k S(W) 8), the in-
duced complex, is isomorphic to the complex A (W) with zero differential:

(k S(W) K.(W),id k S(W) 8) ~ (A(W),O). (7.15)

4. We compare the resolutions (K.(W), 8) and (T~(S(W)), d).

Lemma XVIII.7.7. The map ids(w) a : K.(W) ---- T~(S(W)) is a

chain map over the identity.

PROOF. We have to prove that

d' 0 (ids(w) a) = (ids(w) a) 08. (7.16)

All maps in (7.16) being S(W)-linear, it is enough to check this relation on

elements of the form 1 WI 1\ ... 1\ wn where wI"'" wn belong to W. By
definition, we have

where
ZI = L C:(O")Wu(l) ... Wu(n) ,
uESn
n-I
Z2 = L( _l)i L C:(o")WU(I) ... wu(i)wu(i+I) ... wu(n)
i=1 uESn
and
Z3 = (_l)n L C:(o")Wu(l) ... Wu(n-I)C:(Wu(n)'
uESn
Let us first deal with ZI' We have
n
ZI =L L C:(O")Wi wu(2) ... wu(n)'
i=l crESn.
u(I)=i

By applying the permutation (12 ... i) of sign (_1)i+1, we get

n
ZI (ids(w) a) (L (-l)i+1wi WI 1\ ... 1\ Wi 1\ ... 1\ Wn)
i=1
 XVIII.7 Computations for Symmetric Coalgebras 441

by (7.12). Relation (7.16) will be proved once we have checked the vanishing
of Z2 and of Z3. Concerning Z2' we have

L C:(O")WO"(I) IZi ... IZi WO"(i)WO"(HI) IZi ... IZi wO"(n)

L c:( 0" )WO"(I) IZi ... IZi wa(i) wO"(Hl) IZi ... IZi wO"(n)
aESn,
O"(i)<O"(Hl)

+
aESn.
O"(Hl)<O"(i)

By exchanging i and i + 1, we see that the second summand is the opposite

of the first one. This proves the vanishing of

L C:(O")WO"(l) 1Zi IZi wO"(i)wO"(i+l) 1Zi IZi WO"(n) ,

O"ESn

hence of Z2. Eventually, Z3 = because c:(w) for all W

= E W = SI(W).
D
5. By Corollary 11.2, there exist S(W)-linear maps (3 : T~(S(W)) ->
Kn(W), hI : T~(S(W)) -> T~+I(S(W)), and h2 : Kn(W) -> Kn+l(W)
such that 8(3 = (3d',

(ids(w) IZi 0:)(3 = id + d'h l + hI d' and (3(id s (w) IZi 0:) = id + 8h 2 + h28.
(7.17)
Tensoring both resolutions on the left with k over S(W), we see that

id k IZis(w) (ids(w) 1Zi0:) = 0:: (A-(W),O) -> (T-(S(W)),d)

induces an isomorphism in homology: indeed, by (7.17) the chain maps

0: 0 (id k IZi SeW) (3) and (id k IZi SeW) (3) 0 0: are homotopic to the identities.
6. It is easy to check that J1 is a chain map, i.e., it annihilates all elements
of the form
n-l
c:(a 1)a 2 1Zi IZi an +L (-1)ia 1 1Zi IZi ai - 1 IZi aiaH1 IZi aH2 1Zi IZi an
i=1

+( -It+ 1a l 1Zi IZi an_1c:(an )

The rest of Proposition 7.6 is obvious. D

We now prove Theorem 7.l.

Proof of Theorem 7.1. (a-b) Dualize Proposition 7.6 using Lemmas 7.3-
7.5. We still have to prove that the map J1 : Tn(s(w)) -> An(w) is the
transpose of the antisymmetrization map 0: : An(v) -> Tn(s(v)). Indeed,
let us check that
< J1(X),w >=< X,o:(w) > (7.18)
442 Chapter XVIII. Cohomology and Rigidity Theorems

for all homogeneous X E Tn(s(w)) and W E An(w). If X belongs to

Sdl (W) 0 ... 0 Sd n (W) where (d l , . .. ,dn ) =f. (1, ... , 1), then both sides of
(7.18) vanish. If X = w il 0 .. 0 win where w il ' ... ,win E W, then

< j1(Wil 00 wiJ, v jl A ... A Vjn >=< w il A ... A win' Vjl A ... A Vjn >
vanishes when (i l , ... , in) is not a permutation of (jl' ... ,jn). If it is, the
right-hand side is equal to the sign of this permutation. On the other hand,

< W il 0 ... 0 Win' a( VJt A ... A vjJ >

L c(a) < W il 0 ... 0 Win' V(J"(jll 0 ... 0 V(J"(jn) >
(J"ES n

yields the same result. This proves (7.18).

(c) Using j1 and a, we determine the action of the involution a on the
cohomology of (Te(s(V)), 8). By (5.2) we have

(j1oanoa)(vIA ... Avn) = (_It(n+l)/2 L c(a)j1(vrr(n)0 ... 0 V(J"(I))

(_1)n(n+l)/2 L c(a)v(J"(n) A AV(J"(I)

(J"ESn

Apply the change of variables effected by a = a'T where T is the permuta-

tion (1,n)(2,n -1) with sign (_1)n(n-I)/2. It follows that

(j1 an a)(v I A A v n ) = (-ltn! vI A A v n . (7.19)

From Parts (a-b) of Theorem 7.1 and from (7.19), we conclude that a 2n
acts as the identity on the cohomology while a 2n +l acts as -id. D

XVIII. 8 Uniqueness Theorem for Quantum

Enveloping Algebras
In this section we deal only with a quantum enveloping algebra whose
underlying quasi-bialgebra is the trivial topological bialgebra U(g)[[h]] of
formal series over the enveloping algebra of a finite-dimensional complex
semisimple Lie algebra g. We state the second rigidity theorem.
Theorem XVIII.8.1. Assume we have A = (U(g) [[h]L ~, c, <1>, R) and
A' = (U(g)[[h]L~, c, <1>', R) which are quantum enveloping algebras for the
same finite-dimensional semisimple Lie algebra g and have the same uni-
versal R-matrix R satisfying the conditions R21 = R, R == 1 0 1 modulo
hand R = ~(u)V, where U E U(g)[[h]] and V is a central element of
(U(g) 0 U(g))[[h]]. Then there exists a gauge transformation F in the al-
gebra (Ug0Ug)[[h]] with F21 = F, F == 101 modulo h, and [F,~(a)] = 0]
for all a E A such that A' = A F .
 XVIII.8 Uniqueness Theorem for Quantum Enveloping Algebras 443

The data ~ and R remain unchanged under a gauge transformation F

verifying the properties of Theorem 8.1. Indeed, by definition of ~F (see
XV.3) and by the last condition on F, we have ~F(a) = F~(a)F-l = ~(a)
for all a E A. As for R, by (XV.3.9) we have RF = F21 RF- 1 = FRF- 1.
Now, in view of the third assumption on R, we have

[R, F] = [~(u)V, F] = [~(u), FlV + ~(u)[V, F] = O.

Consequently, R F = R.
Observe that we may (and shall) apply Theorem 8.1 to the case when
R = eht / 2 where t is the 2-tensor in (XVII.1.6). Indeed, we have R21 = R
and R == 101 modulo h. Moreover, (XVII.1.6) implies that
R = ~(ehC/4) (e- hC / 4 0 e- hC / 4).

Before we prove Theorem 8.1, we establish the following lemma.

Lemma XVIII.8.2. Let (A,~, c, <P, R) be a cocommutative quantum en-
veloping algebra such that R21 = Rand R == 1 0 1 modulo h. Then
<P321 = <P- 1.
PROOF. We first claim that

(8.1)

Indeed, we have

R12<P312R13(<P132)-1 R23<P
<P321 R 23 (<P 231 ) -1 R 13 <P213 R 12
<P321R23(id 0 ~)(R)<p.

The first and last equalities follow from Proposition XV.2.2 while the mid-
dle one follows from Corollary XV.2.3. Next, apply the involution 7 13 to
A 0 A 0 A. Since ~ = ~op and R = R 21 , Relation (8.1) becomes
(8.2)

Combining Relations (8.1-8.2), we get

<p- 1R 23 (id 0 ~)(R)(<P321)-1<p321 R 23 (id 0 ~)(R)<p

<p- 1 (R 23 (id 0 ~)(R)) 2<P.

By uniqueness of the square roots of elements congruent to the unit mod-
ulo h, we get

R12(~ 0 id)(R) = <p- 1R 23 (id 0 ~)(R)<p. (8.3)

Comparing (8.1) and (8.3), we conclude that <P321 = <P- 1. o

444 Chapter XVIII. Cohomology and Rigidity Theorems

Proof of Theorem 8.1. We have to show that we can find a gauge transform-
ation taking <1'> to <1'>'. Suppose <1'> and <1'>' are equal modulo h n for some n 2': 1.
This always holds for n = 1 since <1'> and <1'>' are congruent to 1 1 1 modulo
h. Define 'P E U(I)3 by

Let Ant ('P) be the element

Ant('P) = 'P - 'P213 - 'Pl32 - 'P321 + 'P231 + 'P312 (8.4)

of U (I) 3. The first step in the proof of Theorem 8.1 is the following lemma
with the same hypotheses.

Lemma XVIII.8.3. The element 'P is I-invariant and satisfies the rela-
tions 'P321 = -'P, Ant('P) = 0, and

1 'P - (~idid)('P) + (id~id)('P) - (idid~)('P) + ('P 1) = O.

PROOF. (a) Since ~ is coassociative, Relation (XV.1.1) may be rewritten

in the form
[(~ id)(~(x)), <1'>] =0
for all x E I, which means that <1'> and <1'>' are I-invariant, namely that they
belong to the subspace (U(I)3)g[[hlJ. Consequently, 'P is I-invariant too.
(b) Lemma 8.2 implies that

<1'>~21 == <1'>321 + hn'P321 == (<1'>')-1 == (<1'> + hn'P)-l == <1'> - hn'P

modulo hn+1. It follows that 'P321 = -'P.

(c) We now prove that Ant('P) = O. Consider Relation (XV.2.3) for <1'>
and <1'>':

Reducing the latter modulo h n +1 implies that

'P312 - 'P132 + 'P = O. (8.5)

Since 'P321 = -'P, Relation (8.5) yields

-'P213 + 'P231 - 'P321 = O. (8.6)

Adding Relations (8.5-8.6), we get

Ant( 'P) = 'P - 'P213 - 'P132 - 'P321 + 'P231 + 'P312 = O.

 XVIII.8 Uniqueness Theorem for Quantum Enveloping Algebras 445

(d) Using the pentagonal relations (XV.1.3) for <1> and <1>', we get

(1 <1>')(id ~ id)( <1>')( <1>' 1)(~ id id)( <1>') -1 (id id ~)( <1>')-1

= (l<1>)(id~id)(<1>)(<1>l)(~idid)(<1>)-l(idid~)(<1>)-l = 1.
Reducing these equalities modulo hn+l, we obtain the desired 5-term func-
tional equation for rp. 0

The next step is the following one.

Lemma XVIII.SA. There exists a g-invariant element 1 E U(g) U(g)

such that

rp=lI-(~id)(f)+(id~)(f)-Il and 121=1.

PROOF. Using the co homological language of Section 5, we can paraphrase

Lemma 8.3 by saying that rp is a g-invariant element of U(g)3 satisfying

The first two relations in (8.7) mean that rp is a 3-cocycle in the co-
bar complex (T~(U(g)),b). We claim that rp is a coboundary. Using the
isomorphism TJ : 5(g) ----+ U(g) of co algebras , it suffices to check that
1jJ = (r)-l 7)-1 TJ-1)(rp) is a coboundary in (T~(5(g)),b). We also have
Ant(1jJ) = O. By Theorem 7.1 (c) we have H3(T~(5(g)),b) ~ A3 (g), the
isomorphism being induced by the map JL. It is therefore enough to check
that JL(1jJ) = O. Now Ant(1jJ) = 0 implies Ant((JLJLJL)(1jJ)) = O. An imme-
diate computation shows that a(JL( 1jJ)) = Ant ( (JL JL JL) (1jJ)). Therefore,
by Theorem 7.1 we have

1 1
JL(1jJ) = 6 JL(a(JL(1jJ))) = 6 JL(Ant((JL JL JL)(1jJ))) = 0,

which tells us that the cohomology class of 1jJ is zero.

Since the 3-cocycle rp is a coboundary, there is an element 1 E T~ (U(g)),
i.e., an element 1 E U(g)2 with 121 = 1 such that rp = 15(f), i.e.,

rp = 1 1 - (~ id)(f) + (id ~)(f) - 1 1. (8.8)

We have so far proved Lemma 8.4 up to the fact that we can choose 1 to
be g-invariant.
This last fact is a consequence of Proposition 6.2 applied to the co algebra
U(g) on which 9 acts by the adjoint representation (this is where we use
the assumption that 9 is semisimple). Since (T~(U(g)),I5) splits into the
direct sum of (T~(U(g))g,l5) and of (gT~(U(g)))I5) and since rp belongs to
T~(U(g))g) then 1 necessarily belongs to T~(U(g))g. 0
446 Chapter XVIII. Cohomology and Rigidity Theorems

From the element f whose existence is asserted by Lemma 8.4, we deduce

the gauge transformation

(8.9)

The g-invariance of f implies that [F, ~(x)] = 0 for all x E U(g). We also
have F2I = F. We already know that ~ and R remain unaffected by such
a gauge transformation. Let us compute <l>F modulo hn+l. From (8.8-8.9)
and from (XV.3.3) we get

which can be reexpressed as <1>' == <l>F modulo hn+l. We now define an

element 'PI of U(g)3 by

and start the whole procedure all over again. By composing all the gauge
transformations obtained in this way, we obtain a gauge transformation
between the quasi-bialgebras A and A'. This completes the proof of Theo-
rem 8.1. 0

XVIII. 9 Exercises
1. Compute H2 (g, C) for all complex Lie algebras of dimension:::; 3.

2. Show that the space of primitive elements of a coalgebra C can be

realized as the cohomology group HI (Te (C), 8).

3. Give a direct proof of Theorem 7.1 when V is one-dimensional.

4. Let n 2:: 2 be an integer. Consider the algebra A = k[tl!(t n - 1). Let

N be the element N = 1 + t + ... t n - I of A. Show that

is a resolution of k by free left A-modules.

5. Prove the assertions of the Appendix.

XVIII. 10 Notes
The content of Sections 1-7 is classical. The cohomology of Lie algebras
was introduced by Chevalley and Eilenberg in [CE48] following ideas of E.
Cartan. See [Ger64] for a general deformation theory for algebras.
 XVIII.ll Appendix. Complexes and Resolutions 447

The cobar complex appeared independently in papers by Adams [Ada56]

and Cartier [Car57]. Theorem 7.1 is due to the latter. The reader is advised
to take a look at Cartier's elegant proof using resolutions of comodules in
[Car57]. We gave here a more pedestrian proof based on Eilenberg and Mac
Lane's bar complex [EM53] (see also [SC56]). Drinfeld gives a third one in
[Dri89b], Prop. 2.2 and Prop. 3.11.
The content of Section 8 is entirely due to Drinfeld (see [Dri90], Sec-
tion 3). Observe that Theorems 4.1 and 8.1 are non-constructive. It would
be interesting to find an explicit isomorphism 0: : Uh(g) ---+ U(g)[[h]] and
an explicit gauge transformation F, even in the case 9 = .5[(2).

XVIII.II Appendix. Complexes and Resolutions

We recall some facts from homological algebra. For details and proofs, see,
e.g., Cartan-Eilenberg's book [CE56].
Let A be an algebra. A chain complex of left A-modules (C., d) is a family
(Cn)n>O of left A-modules together with A-linear maps d : C n ---+ C n - 1 ,
defined for all n :::: 1, such that dod = O. The last condition implies that
the image of d sits inside its kernel. We can therefore define the homology
groups H.(C.,d) of the chain complex by
Ker (d : Cn ---+ Cn - 1 )
Hn(C.,d) = I m (d'C C)' (11.1)
. n+l ---+ n
A chain complex is acyclic if all its homology groups vanish.
One similarly defines a cochain complex of left A -modules (C, d) as a
family (C n )n>O of left A-modules with A-linear maps 8 : C n ---+ Cn+l such
that 8 0 8 = 0. The cohomology groups H (C , d) are defined by

Hn(c* d) = Ker (d : C n ---+ Cn+l). (11.2)

, 1m (d : Cn-l ---+ cn)

In both cases, we agree that C_ 1 = C- 1 = O.

Let M be a left A-module. A resolution of M by free left A-modules is
a chain complex (C., d) of free left A-modules together with an A-linear
map c : Co ---+ M such that the chain complex
d
. ----4
d C
C 2 ----4 dC c
1 ----4 0 ----4 M ---+ 0 (11.3)
is acyclic. Any A-module has a resolution by free left A-modules.
A chain map f: (C.,d) ---+ (C~,d') between chain complexes of A-
modules is a family Un : C n ---+ C~)n~O of A-linear maps such that
(11.4)
for all n. A chain map f induces a map f. : H.(C., d) ---+ H.(C~, d) between
the corresponding homology groups.
448 Chapter XVIII. Cohomology and Rigidity Theorems

A homotopy h between two chain maps f,1' : (e.,d) -+ (e~,d') is a

family (h n : en -+ e~+l)n>O of A-linear maps such that

(11.5)

for all n (by convention h_1 = 0). If there exists a homotopy between f
and 1', we say that the two chain maps are homotopic. Homotopy is an
equivalence relation. Homotopic chain maps f, f' induce the same map on
homology: f. = f~
One of the basic results in homological algebra is the following compar-
ison theorem for resolutions.

Theorem XVIII. 11. 1. Let (e., d) [resp. (e~, d')l be a resolution of an

A-module M [resp. of M'l by free left A-modules. Suppose we also have an
A -linear map 1-1 from M to M'. Then there exists a chain map f from
(e., d) to (e~,d') over f-1' i.e., such thatf_1 010 = 10'01-1' If I' is another
chain map over f -1' then f and I' are homotopic.

Corollary XVIII.11.2. Let (e., d) and (e~, d') be resolutions of an A-

module M by free left A-modules. There exist chain maps

f: (e., d) -+ (e~, d') and g: (e~, d') -+ (e., d)

such that g 0 f and fog are homotopic to the identities.

PROOF. Applying the above theorem to f-1 = id, we get chain maps f,g
such that 10 = 10' 0 fo and 10' = 10 0 go' Now go f is a chain map from (e., d)
to itself with 10 0 (go 0 fo) = E. SO is the identity on e .. By the second part
of the theorem, we see that g 0 f is necessarily homotopic to the identity.
A similar argument works for fog. D
Chapter XIX
Monodromy of the
Knizhnik-Zamolodchikov Equations

The purpose of this chapter is twofold:

(i) For any complex Lie algebra 9 and any invariant symmetric 2-tensor
t E g0g, construct a quantum enveloping algebra Ag,t for 9 whose canonical
2-tensor is t.
(ii) Give Drinfeld's reformulation and proof of an important result of
Kohno's which asserts that, if 9 is semisimple, the monodromy of a cer-
tain system of differential equations, called the Knizhnik-Zamolodchikov
system, is equivalent to the braid group representation provided by the
universal R-matrix of the quantum enveloping algebra Uh(g) introduced in
Chapter XVII. In terms of categories, Drinfeld's proof amounts to show-
ing that the braided tensor category Uh(g)-Mod fr of modules over the
Drinfeld-Jimbo algebra (as defined in XVII.3) is equivalent to a braided
category of modules over the trivial deformation U(g)[[h]J, equipped with
a non-trivial associativity constraint.
We shall use some elementary differential geometry in this chapter, but,
as was the case with knot theory earlier in this book, we shall focus on
Drinfeld's ideas and therefore skip the details that are not essential to
their understanding.

XIX.1 Connections
We assume some standard knowledge of differential geometry. Let us never-
theless recall a few facts. For more details, the reader may consult [KN63].
450 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Let X be a complex analytic variety of dimension nand p : E ---t X be

a complex analytic vector bundle of rank d over X. If x is a point of X,
we denote by Fx the fibre at x: Fx = p-1(x). A connection on E is a linear
map V' from the space r(X, E) of sections of the vector bundle E into the
space n1(X, E) of differential I-forms with values in E such that for any
section s and any complex analytic function I on X we have

V'(fs) = (df)s + IV'(s). (1.1)

If V'1 and V' 2 are two connections on E, then the difference V'1 - V' 2 is
O(X)-linear, where O(X) is the ring of complex analytic functions on X.
Locally, we can write a section s in the form

(1.2)
where 11 , ... , Id are complex analytic functions on X and {e 1 , ... ,ed} is a
basis of the fibre. Any connection V' on E can be written locally as

V's = ds - rs (1.3)
where d is the de Rham differential and r is a differential I-form on X with
values in the endomorphism ring of E.
A section s of the bundle is horizontal for the connection V' if V's = 0,
i.e., if locally s is a solution of the system

ds = rs. (1.4)
Let "( : [0, 1]---t X be a smooth path in X from Xo = "((0) to Xl = "((1). We
may pull back the matrix r of differential forms on X along "( to a matrix
A(O)dO = "(*r of differential forms on the interval [0,1]. By the theory of
ordinary differential equations, there exists a unique smooth map A, from
[0, 1] into the group of linear automorphisms of the fibre bundle such that
A,(O) = id and w(O) = A,(O)w(O) is a solution of the differential equation

d~~O) = A( O)w( 0). (1.5)

The automorphism A,(I) defines a linear isomorphism T, : Fxo ---t F X"

called the parallel transport along the path "(. When "(' is a path from
Xl to x 2 we may consider the composed path "("(', as in the Appendix
to Chapter X. The uniqueness theorem on systems of first order linear
differential equations implies that

(1.6)
The holonomy group at Xo is defined as the subgroup of Aut(Fxo) gen-
erated by T, for all loops "( based at Xo at X. In general, the holonomy
depends on the local as well as on the global structure of X. In other words,
 XIX.2 Braid Group Representations from Monodromy 451

T, may change as "( varies (even infinitesimally). Let us give a condition

on the connection \7 under which the parallel transport depends only on
the homotopy class of the path.
We need the notions of covariant derivative and of curvature of a connec-
tion. It is not difficult to show that the connection \7 extends to a unique
endomorphism of degree 1, still denoted \7, and called the covariant deriva-
tive, of 0.. (X, E) such that

\7(ww' ) = (dw)w ' + (-l)Pw\7(w')

for any pair (w, Wi) of differential forms where p is the degree of w.
Lemma XIX. I. I. The curvature K = \7 0 \7 is O(X)-linear.
PROOF. Let w be a differential form and f be a function on X. We have
K(fw) = \7((df)w + f\7(w)) = (d 2 f)w - (df)\7(w) + (df)\7(w) + fK(w)
= fK(w)
since d 2 f = O. D

Locally, the curvature can be expressed in terms of r by

K(s) = d(ds - rs) - r(ds - rs) = (-dr + r A r)s,
which leads to the formula
K = -dr+r Ar. (1.7)
When K = 0 we say that the connection is fiat. In this case, 0.. (X, E)
becomes a cochain complex with differential \7.
Proposition XIX.I.2. Given a connection \7, we have T, = T,I for any
pair b, "(') of homotopic paths in X if and only if the connection is fiat.
This statement implies that, if K = 0, then for any point Xo in X the
parallel transport induces a group morphism T from the fundamental group
1l"l(X,XO) to Aut(Fxo )' It is called the monodromy representation of the
fundamental group acting on the fibre. We shall not prove Proposition 1.2
for which we refer to the classical literature.

XIX.2 Braid Group Representations from

Monodromy
We apply the generalities of Section 1 to the following situation. Suppose
given a finite-dimensional complex vector space W, an integer n > 1 and a
family {Aij h~i<j~n of endomorphisms of W satisfying the conditions
(2.1)
452 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

whenever i, j, k, are distinct integers S n, and

(2.2)

whenever i, j, k are distinct integers. Consider the differential system

dw = '"
L...J A
--'-J-(dzi - dzj)w. (2.3)
l~i<j~n'
z - zJ

According to Section 1, this defines a connection V' = d- r on the trivial

bundle Y n x W where

r= I: A
--'J-(dz. - dz.)
z.-z. ' J
(2.4)
l~i<j~n' J

and Yn is the complex variety

already considered in X.6.

Proposition XIX.2.1. The connection V' =d- r is fiat.

PROOF. By Proposition 1.2 it suffices for us to show that the curvature

K = -dr + r 1\ r vanishes. Since the endomorphisms Aij do not depend
on the variables Zl' ... , zn' we have dr = o. It is therefore enough to check
that r 1\ r = O. Set
dz - dz
u .. =dlog(z-z.)=' J.
'J 'J z., - z.J

We have
r 1\ r = I: AijAk u ij 1\ uk' (2.5)
i<j, k<

The right-hand side of (2.5) is equal to K2 + K3 + K4 where

Kp = I: AijAki' u ij 1\ uk'
i<j, k<

the set of indices {i < j, k < } C {I, ... ,n} running over all such subsets
of cardinality p. We now show that K 2 , K3 and K4 vanish separately due
to Relations (2.1-2.2). For K 2 , this results from Uij 1\ u ij = O. Let us deal
with K 4 : exchanging (i,j) and (k,), we get

K4 = I: AkAij uk 1\ uij'
i<j, k<
 XIX.2 Braid Group Representations from Monodromy 453

The right-hand side is equal to -K4 because uk 1\ u ij = -u ij 1\ Ukf and

AijAk = AkfAij , the latter following from (2.1). Consequently, 2K4 = 0,
which proves the vanishing of K 4 .
Before we prove that K3 = 0, we record the following well-known lemma
(sometimes called Arnold's lemma):
Lemma XIX.2.2. For any triple (i, j, k) of distinct integers, we have

Uij 1\ Ujk + Ujk 1\ Uki + Uki 1\ U ij = O.

_f
PROOF. Let i,j, k be distinct indices. Then
(dzi - dz j ) 1\ (dz j - dz k )
u ij 1\ ujk + ujk 1\ Uki + Uki 1\ Uij - (
Zi - Zj
)(
Zj - zk
)

where the symbol 1 means that we take the sum of the term under the
integral with the other two obtained by circular permutations of the indices.
We have

l.h.s

o
Let us resume the proof of Proposition 2.1. We still have to prove that
K3 = o. We break the sum K3 into three smaller pieces K3 = K5 + K6 + K 7
The first piece is
K5 = L AijAik u ij 1\ uik
i<jopk

Exchanging j and k, we get

K5 = L [A ij , Aikl u ij 1\ uik
i<j<k

Similarly,

K6 = L AikAjk uik 1\ ujk = L [Aik' Ajkl uik 1\ ujk

iopj<k i<j<k

The last piece is

K7 L (AijAjk u ij 1\ ujk + AjkAij ujk 1\ U ij )

i<j<k

L [A ij , Ajkl u ij 1\ ujk
i<j<k
454 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Therefore K3 = ~i<j<k Zijk where

Zijk = [Aij' Aikl uij 1\ uik + [Aik' Ajkl uik 1\ ujk + [Aij, Ajkl uij 1\ ujk
Using Lemma 2.2, we get

Zijk [Aij, Aikl Uij 1\ uik + [Aik , Ajkl uik 1\ ujk

+[A ij , Ajkl (uik 1\ ujk + uij 1\ uik)
[Aij, Aik + Ajkl u ij 1\ uik + [Aij + A ik , Ajkl uik 1\ ujk
o
by (2.2). This implies the vanishing of K3 and completes the proof of Propo-
sition 2.1. 0
Since the connection associated to the differential system (2.3) is flat,
there exists a monodromy representation of the fundamental group of Yn
on the vector space W.
Remark 2.3. It can be proved that the fundamental group of Yn is the
pure braid group Pn defined as the kernel of the natural surjection of the
braid group En onto the symmetric group Sn. Let Pn be the Lie algebra
generated by a set {Xijh::;i<j::;n of generators and Relations (2.1-2.2). A
Pn-module W is nothing but a vector space W with a family (Aijh:S;i<j:S;n
of endomorphisms satisfying (2.1-2.2). For any such module, the connec-
tion corresponding to the differential system (2.3) is flat, therefore inducing
a monodromy representation of the group Pn . It thus makes sense to view
the Lie algebra Pn as the analogue of the Lie algebra of a Lie group for
the pure braid group Pn and monodromy as the analogue of integrating
a representation from the Lie algebra to the Lie group. For these reasons,
Relations (2.1-2.2) are sometimes called the infinitesimal braid group rela-
tions. For more details on the relationship between Pn and Pn , see [Aom78l
[Hai86l [Koh85l
What we actually would like to have is a monodromy representation of
the full braid group En' not only of the subgroup Pn . This can be achieved
as follows. Suppose we have a left action of the symmetric group Sn on the
vector space W. Then there exists a right action of Sn on the trivial vector
bundle Y n x W given by
(2.6)
for a E Sn, (Zl' ... ' zn) E Y n , and w E W. The composition
Y n x W~Yn --> Xn = Yn/Sn
factors through the quotient space E = (YnxW)/Sn of Sn-orbits on Y n xW.
The topological space E thus becomes a non-trivial vector bundle over Xn
with fibre W. In the space E we have
(2.7)
 XIX.3 The Knizhnik-Zamolodchikov Equations 455

Now, if the differential system (2.3) is invariant under the action of Sn'
then the connection 'V = d - r descends to a connection on E. If Rela-
tions (2.1-2.2) are satisfied, then it has a monodromy representation on the
fundamental group of Xn which, by Proposition X.6.14, is the full braid
group Bn'

XIX.3 The Knizhnik-Zamolodchikov Equations

We consider a differential system that is a special case of the systems con-
sidered in Section 2 and that depends on the following data:
(i) a finite-dimensional complex Lie algebra g,
(ii) an invariant symmetric 2-tensor t on g, i.e., an element t = l:r x r0Yr
of 9 0 9 such that
t21 = t and [.6.(X) , tJ = 0 (3.1)
for all x E g,
(iii) a complex parameter h,
(iv) an integer n > 1, and
(v) a finite-dimensional g-module V.
Definition XIX.3.1. The Knizhnik-Zamolodchikov differential system as-
sociated to the above data is the system

dw =
h
L t
_'_J-(dzi - dzj)w (KZn)
2nA l:S;i<j:S;n zi - Zj

where W = w(zl' ... ,zn) is a function on Y n with values in W = vn and

where tij is the element of U(g)n defined for all i #- j by

t
~
= "~ x(1)
r
0 0 x(n)
r
r

where XCi)
r = x r' X(j)
r = Yr and x(k)
r =1 otherwise .
Lemma XIX.3.2. The elements (tijh:S;i<j:S;n induce endomorphisms of
vn satisfying Relations (2.1-2.2).
PROOF. Relations (2.1) hold by definition oft ij . Relations (2.2) follow from
the g-invariance of t. We show this when i = 1, j = 2, and k = 3. We have

[t12' t 13 + t 23 J L [x r 0 Yr 0 1, Xs 010 Y s + 10 Xs 0 YsJ

r,s

s r

by (3.1). D
456 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Lemma 3.2 and Proposition 2.1 imply that the system (KZ r,) defines a
fiat connection on the trivial bundle over Yn with fibre V 18m and, conse-
quently, determines a monodromy representation
p~z : 'if l (Y,-,) --+ Aut(Vn).

Let us prove that this monodromy can be extended to a representation

of the full braid group Bn' As explained at the end of Section 2, we have
to specify a left action of the symmetric group Sn on the space W = vn.
We choose the action given by
(3.2)
for (J E Sn and VI' ... ,Vn E V. The 2-tensor t being symmetric by hypoth-
esis, we have tij = tji for any pair (i, j) of distinct integers. Consequently,
we can rewrite the system (KZn) as

dw = h "6 t
~(dZi - dzj)w. (3.3)
4'ifH ",-z
ls,i#.Js,n r J

It is clear from (3.3) that the system is invariant under the action of the
symmetric group. We thus obtain a monodromy representation
(3.4)
Here we take the point p = (1,2, ... , n) as the base-point of Xn and we
identify the fibre of the bundle E = (Yn x vn) / S n at the point p with
vn.
The main objective of this chapter is to compute the monodromy rep-
resentation p~z as explicitly as possible from the above data. This is a
difficult task. To begin with, let us consider the following special cases.
(a) Case when h = 0: the differential system reduces to dw = 0 which
has constant solutions over Y,-,. The corresponding monodromy is the rep-
resentation of Bn on Vn coming from the action (3.2) of the symmetric
group.
(b) Case when n = 2: the system (KZ 2 ) reduces to
h t
dw =
2'if
H-lzI -Z2 (dz l - dz 2 )w. (3.5)

In order to determine its monodromy, we represent the generator (J I of the

braid group B2 by the loop z( 8) = (Zl (8), Z2 (8)) where 8 varies in [0, 1] and

ZI (8) = ~(3 - e71'v=TS) and z2(8) = ~(3 + e71'v=TS).

2 2
We have z(O) = z(l) = (1,2) = (2,1) in X 2 . Pulling back (3.5) along this
loop leads to the ordinary differential equation

dw = ht W(8) (3.6)
d8 2
 XIX.3 The Knizhnik-Zamolodchikov Equations 457

whose unique solution is given by

w(s) = ehtsj2w(O), (3.7)
where ehtsj2 is the classical entire series ehtsj2 = "" hntns n converging
un>O 2nn!
in Aut(V V) for all values of the complex parameter h. We get the
monodromy action of the generator 0"1 by setting s = 1 in (3.7), namely

(3.8)

for all VI' V2 E V. The flip appears in (3.8) as a consequence of the equality

(2,1; VI V2) = (1,2; v 2 VI)

in the non-trivial bundle E = (Y2 X V0 2 )/S2'
(c) Suppose that n 2: 2 and that the condition

[tij' t k ] = 0 (3.9)
holds for all i, j, k, . We claim that in this case the monodromy action of
En is given for each of the generators 0"1' ... , 0"n-l by

(3.10)

h
were T i ' Hl'd
= 1 V0(i-l) 161 TV V 1'dV0(n-i-l) . 'T'
f(A
.1.0 prove th e cI'
aIm, represent
the generator O"i of the braid group by the loop z( s) = (zl (s), ... , zn (s))
where

(3.11)

and Zj (s) = j for j i= i, i + 1. Then the system obtained by pulling back

(KZn) along this loop can be solved as in Case (b) because the t ij commute
with one another.
We end this section with some observations based on the three special
cases we have just considered. Firstly, in all cases above, the monodromy is
an analytic function in the complex parameter h viewed as a variable. This
holds for any (KZ)-system by the general theory of ordinary differential
equations: the differential system depending linearly on h, its solutions are
analytic in h. We shall henceforth consider the monodromy as an analytic
function in h or rather, as a formal series in the variable h. Denote by
(U (g) [[h]], D., c) the trivial topological bialgebra associated to the bialgebra
(U(g), D., c) as in Example 3 of XVI.4. For any g-module V, put on V[[h]]
the U(g) [[h]]-!?-odule structure induced from the action of U(g) on V. Recall
that (V[[hJl)0 n = V<8In [[h]] by Proposition XVI.3.2. We can then express
the analytical dependence on h of the monodromy of (KZn) by a group
morphism
(3.12)
458 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

From Case (a) we see that p~z is congruent modulo h to the representation
coming from the symmetric group action (3.2).
We next observe that the monodromy is independent of the g-module V.
This again holds in full generality because the system has coefficients in the
tensor powers of U(g). One can prove this using Chen's theory of formal
connections and formal monodromy [Che73] [Che75] [Che77a].
The last remark is the following: in Case (c) above, the monodromy can
be derived from a topological braided bialgebra structure. Indeed, assume
that (U(g)[[h]],~, c) is the trivial topological bialgebra as above. Set R =
e ht / 2 . When t satisfies Relations (3.9), the element R of U(g)[[h]]0U(9)[[h]]
satisfies Relations (XVI. 4. 15-4. 17) with cp = 10101. The proof of this
claim is similar to the one used in XVI.5, Example 1. Therefore,
- - R = eht/2 )
Ag,t = (U(g) [[h]J, ~,c, cp = 1@1@1,
is a topological braided bialgebra whose universal R-matrix is symmetric:
R21 = R. The g-module V extends to the Ag,cmodule V[[h]] defined above.
The universal R-matrix R gives rise to a representation

p~ : Bn -+ AutC[[h]] (V0 n [[h])

of the braid group Bn following the procedure explained in XV.4. A com-

parison with (3.10) gives the following.
Proposition XIX.3.3. When t satisfies (3.9), the monodromy of the sys-
tem (KZn) coincides with the braid group representation induced by the
universal R-matrix R = e ht / 2 of the topological braided bialgebra Ag,t> i.e.,
we have
PKZ
n
= pRn
In the next section, we shall extend this result to the case of an arbitrary
invariant symmetric 2-tensor t in spite of the fact that Ag,t may no longer
be a topological braided bialgebra.

XIX.4 The Drinfeld-Kohno Theorem

In addition to the hypotheses of Section 3, we assume that the Lie algebra 9
is semisimple. In this situation, there exists a topological braided bialgebra

quantizing the enveloping algebra of 9 (see XVII.2). Any finite-dimensional

g-module V can be extended to a canonical Uh(g)-module V such that
V/hV = V. Indeed, if V = VA is the simple g-module associated to a
dominant weight .x, then we define VA as the Uh(g)-module whose existence
is asserted in XVII. 2. If V = EBA VA is a direct sum of simple g-modules,
 XIX.4 The Drinfeld-Kohno Theorem 459

we define V to be V = EIh VA (see also XVIII.4). The universal R-matrix

Rh induces a morphism of groups

p~h : En ----t AutC[[h]](v@n) = AutC[[hll(vn[[h]]) (4.1)

into the automorphism group of Vn with formal series coefficients. On the
other hand, the monodromy of the system (KZn) is a morphism of groups
p~z : En ----t AutC[[h]] (vn[[h]]). (4.2)
We now state the Drinfeld-Kohno theorem.
Theorem XIX.4.1. If 9 is a semisimple Lie algebra and t E 9 9 is
the invariant symmetric 2-tensor given by (XVII.1.6), then the braid group
representations p~z and p~h are equivalent for any n > 1 and any g-module
V. In other words, there exists a C[[h]]-linear automorphism u of vn [[h]]
such that
p~Z(g) = Up~h(g)u-l
for all elements 9 of the braid group En'
The rest of this chapter is devoted to the proof of this important the-
orem which expresses a geometrical problem in terms of quantum groups
and produces an explicit expression for the monodromy of the Knizhnik-
Zamolodchikov equations in the semisimple case.
Drinfeld's proof of Theorem 4.1 relies on two main ideas. The first one
is to consider the trivial topological bialgebra (U(g) [[h]], .6., c) equipped
with the invertible element R KZ = eht / 2 E U(g)2[[h]] as at the end of
Section 3. In general, R KZ does not induce a topological braided bialge-
bra structure on U(g)[[h]]. Nevertheless, it does induce the structure of a
topological braided quasi-bialgebra, making U(g)[[h]] into a quantum en-
veloping algebra. Moreover, this structure contains all the information on
the monodromy of all (KZn)-systems. These assertions are summarized in
the following theorem which generalizes Proposition 3.3.
Theorem XIX.4.2. For any complex Lie algebra and any element t of
9 9 satisfying Conditions (3.1), there exists an element

<I>KZ E U(g)3[[h]] = (U(9)[[h]])@3

with <I>KZ == 1 1 1 mod h such that
(i) the topological (quasi-) bialgebra

Ag,t = (U(g)[[h]],.6.,c,<I>KZ,R KZ = eht / 2)

is a quantum enveloping algebra, and
(ii) for any integer n > 1~ and any g-module V, the monodromy represen-
tation p~z of En on V[[h]]n coincides with the braid group representation
p~KZ induced by the universal R-matrix R KZ = eht / 2 as in XV.4.
460 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Since (RKZhl R KZ = e ht == 1 1 + ht mod h, we see that the canonical

2-tensor of Ag,t coincides with t. Thus, Ag,t provides a solution to the quan-
tization of (I, t). Notice that this solution is obtained without deforming
the multiplication or the comultiplication of the enveloping algebra. This
is the first major property of the QUE Ag,t.
The second one is expressed in Part (ii) of Theorem 4.2: loosely speak-
ing, it means that the algebra Ag,t is universal for the monodromy of all
Knizhnik-Zamolodchikov differential systems. Observe also that the solu-
tion of (KZ 2 ) as given by Formula (3.8) forces the universal R-matrix of
Ag,t to be equal to e ht / 2 .
Drinfeld's second idea for the proof of Theorem 4.1 can be expressed as
follows.

Theorem XIX.4.3. If, furthermore, the Lie algebra I is semisimple and

t is the 2-tensor considered in Theorem 4.1, then there exist a gauge trans-
formation F E Ag,tAg,t and a q[h]]-linear isomorphism

of topological braided quasi-bialgebras.

As a consequence of Theorem 4.3 and of Theorem XV.3.9, we get the

following important categorical interpretation of Drinfeld-Jimbo's algebras.
A more precise statement can be found in Corollary XX.6.2.

Corollary XIX.4.4. The tensor functor (a*, id, rpf) is a braided tensor
equivalence from a braided tensor category of topologically free U(g)[[h]]-
modules equipped with associativity constraint induced by <P KZ and braiding
induced by R KZ , to the category U h (g)-Mod fr of XVII.3.

Theorem 4.3 will be proved in the next section as a consequence of results

of Chapter XVIII. The construction of Ag,t and an indication of the proof
of Theorem 4.2 will be given in Sections 7-8. We now prove the Drinfeld-
Kohno theorem.
Proof of Theorem 4.1. It follows immediately from Theorem 4.2, Theorem
4.3 and Theorem XV.4.2. The latter implies, furthermore, that the auto-
morphism u is given by the action of the element F12 of U(g)n[[h]]. 0
 XIX.5 Equivalence of Uh(g) and Ag t 461

XIX.5 Equivalence of Uh(g) and Ag,t

We start with a semisimple Lie algebra g and the 2-tensor
t = _~--'-(C---,)_-_lQ9_C_-_C_Q9_1 (5.1)
2
associated to the Casimir element C of U(g). The aim of this section is to
prove Theorem 4.3. We shall do this in three steps.
Step 1. Since g is semisimple, we may apply Theorem XVIII.4.1. It gives us
a C[[h]]-linear isomorphism of algebras a : Uh(g) ---> U(g)[[h]] with a == id
modulo h, sending the Drinfeld-Jimbo QUE to the trivial deformation of
U(g). Using a, we may transfer all structure maps of Uh(g) to U(g)[[h]]. In
particular, define
and (5.2)
The map a becomes an isomorphism of topological braided bialgebras from
Uh (g) to
(U(g)[[h]J, ~h' ch' (a 129 a)(Rh))'
The maps ~h and ~ are algebra morphisms, both congruent to ~ modulo
h. Recall from V.2 that U(g) 129 U(g) = U(g x g). Now apply Theorem
XVIII.2.1 with g' = g x g. We get an invertible element F' in (U gQ9 U g) [[h]]
such that F' == 1 129 1 modulo hand
~h(X) = F'-l~(x)F' (5.3)
for all x E U(g)[[h]].
Lemma XIX.5.1. We have ch = C.

PROOF. Since ch is a counit for ~h' it follows that Ch is a counit for ~h'
Therefore
id = (Ch 129 id)~h = (Ch 129 id)(F'-l~F'),
which means that
(Ch 129 id)~(x) = xrl (5.4)
where = (ch 129 id)(F'). Using Sweedler's sigma notation, we get

Ch(x) ch'(2:x'c(x"))
(x)

c(2: Ch(X')X")
(x)

c(xrl)
c()c(x)c()-l
c(x)
by the counit axiom and (5.4). o
462 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

We have proved so far that the QUE (Uh (g),6. h ,E h ,10101,Rh ) is

isomorphic to the topological braided quasi-bialgebra

(U(g) [[h]], F'-l6.F', 10,1010 1, (a 0 O')(Rh))' (5.5)

Using F' as a gauge transform, we get the following, which concludes

Step 1.

Proposition XIX.5.2. The QUE (Uh(g), 6. h , Eh' 10101, R h ) is isomor-

phic to the gauge transform by F' of the topological braided quasi-bialgebra
(U(g) [[h]], 6., 10, <P, R) where

<P = (id 0 6.)(F')(l 0 F')(F,-l 0 1)(6. 0 id)(F,-l)

and

Observe that since 6. is coassociative, <P has to be g-invariant, i.e.,

[(6. 0 id)6.(x), <p] =0

for all x E g. Similarly, R is g-invariant since 6. is cocommutative.
Step 2. We apply the symmetrization procedure of XVI.6. By Proposition
XVI.6.2, there exists a gauge transformation F" on U(g)@2[[hll such that
[6.(x), F"] = 0 for all x E 9 and such that, if we set R' = F~~ RF"-l, then
R;l = R'. We make the following capital claim.

Lemma XIX.5.3. Under the previous hypotheses, we have R' = e ht / 2 .

PROOF. By Proposition 5.2 we have

R,2 R;lR'
F" F'(O' 0 O')((Rhb)F';;} F";;} F"21 F~l (a 0 O')(Rh)F,-l F"-l
F" F'(O' 0 O')((Rhb Rh)F,-l F"-l.

Now by Proposition XVII.3.2 and by Relations (5.2-5.3) and (XVII.3.6)

we have

R,2 F" F' (a 0 a) ( 6.h(ehCh/2)( e- hCh / 2 0 e- hCh / 2 )) F'-l F"-l

F" F' 6. h(e ha (Ch)/2) (e- ha (Ch)/2 0 e- ha (Ch)/2) F,-l F"-l

F" F' F,-16.(e hC / 2)F' (e- hC / 2 0 e- hC / 2)F,-1 F"-l

F"6.(ehC/2)F"-1(e-hC/2 0 e- hC / 2 )
6.(ehC/2)(e-hC/2 0 e- hC / 2)
 XIX.6 Drinfeld's Associator 463

since C is central and F" is g-invariant. Finally, in view of the relationship

between the Casimir element and the 2-tensor t, and of the centrality of C,
we get
R,2 ehA(C)/2(e-hC/2 0 e- hC / 2)
eh (A(C)-10C-C0 1)/2
eht .

Since R' == 1 0 1 mod h, it follows that R' = eht / 2 . D

Remark 5.4. The only property peculiar to Uh(g) used so far in this proof
is the one stated in Proposition XVII.3.2. Drinfeld [Dri90] actually proves
more: if A is any QUE for a complex semisimple Lie algebra g, then A
is necessarily isomorphic to the gauge transform of a topological braided
quasi-bialgebra of the form (U g[[h]],~, c, <I>, R) where R = R21 = ehO / 2
for some invariant symmetric 2-tensor e
on g. The trivial deformation
e
(U g[[h]],~, c, 10101, 101) corresponds to = 0. This concludes Step 2.
Step 3. Summing up Steps 1 and 2, we see that Drinfeld and Jimbo's
QUE Uh(g) is isomorphic as a topological braided bialgebra to the gauge
transform of a topological braided quasi-bialgebra of the form

(U(g)[[h]],~,c,<I>',R' = eht / 2). (5.6)

Now the QUE Ag,t of Theorem 4.2 is of the same form, except that <I>' may
differ from the element <I>KZ of Ag,t. This discrepancy is taken care of by
Theorem XVIII.S.1 which implies the existence of a gauge transformation
F'" on U(g)0 2 [[h]] such that

(U(g)[[h]],~,c,<I>',R' =e ht / 2) = (Ag,t) Fill' (5.7)

Setting F = F'(F,,)-l F"', we obtain an isomorphism between Uh(g) and

(Ag,t)F' which proves Theorem 4.3.

XIX.6 Drinfeld's Associator

In order to construct the element <I>KZ of Theorem 4.2, we investigate the
linear differential equation

G'(z) = h
A (A- + -B)
1 G(z) (6.1)
27r -1 z z-
where G(z) is a formal series in two non-commuting variables A and B
with coefficients which are analytic functions in the complex variable z. As
above, h is a formal parameter.
464 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Equation (6.1) has a singularity at 0 and 1. Changing z into liz shows

that it also has a singularity at 00. These singularities are regular or Fuch-
sian. The theory of such equations is classical (see [Was87]).
Let C' be the simply-connected, connected complement of the union
of the real half-lines] - 00,0] and [1, +oo[ in the complex plane. By the
fundamental theorem of linear differential equations, Equation (6.1) has a
unique analytical solution on C' with specified value at any given point in
C'. Since the equation depends linearly on the parameter h, the solutions
have an analytic dependence on h. They can thus be considered as formal
series in h. Observe that, when h = 0, the equation reduces to G'(z) = 0
whose solutions are constant.
We first examine the asymptotic behaviour of the solutions of Equation
(6.1) at the singularities 0 and 1. Set h = 27r}:-r.

Proposition XIX.6.l. There exist unique solutions Go and G 1 of Equa-

tion (6.1) such that

Go(z) "'z-+O zhA and G1(z) "'z-+l (1- z)hB.

By this we mean that Go(z)z-hA [resp. G 1(z)(1-z)-hB J has an analytic

continuation in a neighbourhood of 0 [resp. of 1J with value at 0 [resp. at 1J
equal to 1. Here zhA and (l_z)hB are well-defined on the simply-connected
subspace C'.
PROOF. We shall give the proof for Go. Let us look for a solution of the
form
(6.2)
with P(z) = Er>o Przr. Suppose we can find such a series, that it is con-
vergent, and thaCP(O) = Po = 1. Then the function Go satisfies the require-
ments of Proposition 6.1. The uniqueness property follows by uniqueness
of the solutions.
Let us now prove that there exists a family (Pr)r>O such that Po = 1
and Go(z) is a formal solution of (6.1). We have -

This can be rewritten under the form

zP'(z) - h[A, P(z)J = _hBzP(z). (6.3)

1-z
Expanding (6.3) in powers of z, we get [A, poJ = 0, and, for r > 0,

(6.4)
 XIX.6 Drinfeld's Associator 465

Equations (6.4) have a solution. Indeed, take Po = 1; then Pr is uniquely

determined by Po' ... 'Pr - 1 due to the fact that the operator rid - Ii ad(A)
is invertible with inverse equal to

The convergence of P(z) results from the general fact that a formal solution
of a regular singular equation is necessarily convergent. We refer to [Was87J,
11.5 for details.
Similarly, one proves that there exists an analytic function Q(z) defined
in a neighbourhood of 0 such that

(6.5)

when z is close to 1. D

Since Go and G 1 are both non-zero solutions of Equation (6.1), they have
to differ by an invertible element.

Definition XIX.6.2. We define <I> (A, B) by Go(z) = G1(z)<I>(A,B).

The element <I> = <I>(A, B) is by definition an element of the algebra S
of formal series in the non-commuting variables A and B with coefficients
in C[[h]] (the variable h commuting with A and B). It is called Drinfeld's
associator, a term which will be justified by the results of Sections 7-8 and
of XX.6.
We end this section by giving an expression for <I> in terms of the iterated
integrals and the multiple zeta values defined in the Appendix to this chap-
ter. For any real number a such that 0 < a < 1, let G a (z) be the unique
solution of Equation (6.1) such that Ga(a) = 1. The element <I> is related
to the solutions G a as follows.

Lemma XIX.6.3. We have

PROOF. Let a be a positive real number sufficiently close to 0 so that P(a)

in (6.2) is defined. Since Go and G a are both solutions of Equation (6.1),
they differ by a constant which one gets by evaluating both solutions at
z = a. We have

(6.6)

for all z. When z is close to 1, we have, by (6.5),

(6.7)
466 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Setting z = 1 - a and using (6.6~6.7) and Definition 6.2, we get

a~hBGo(1- a) a- hA P(a)-la hA
a-hB Q(a) a hB <P a- hA P( a) -la hA .

When a tends to 0, then Q(a) and P(a) tend to l. Consequently, the right-
hand side of the last equation tends to <P. D

Let bE ]0,1[. By Picard's method of approximation (see Appendix), the

value Ga(b) of the solution G a can be computed in terms of iterated inte-
grals. More precisely, we have

(6.8)

where M runs over all monomials in A and Band J:

D(M) is the iterated
integral obtained by replacing each occurrence A in NI by the I-form hDo
and each occurrence B by hDl where

n _ 1 ds and D _ 1 ~
-
21fA s -1'
~~o
21fA s 1 -

By (1l.15) of the Appendix we know that, if the monomial M starts with

A and ends with B, i.e., is of the form API Bq, ... APk Bqk, the limit

lim
a--+O
1a
1 a
- D(M)

exists and is equal to

(6.9)
where the complex numbers T(Pl' ql' ... ,Pk' qk) have been defined and com-

J:-
puted in the Appendix in terms of multiple zeta values. If the monomial M
begins with B or ends with A, then the integral a D(M) diverges as a

tends to O. In order to get rid of such "diverging" monomials, we consider

the C[[hll-submodule S of formal series in S spanned by all monomials be-
ginning with A and ending with B. Let 1f : S --> S be the projection which
is the identity on S and sends the "diverging" monomials to O. Clearly,
1f(G a (l- a)) has a limit r in S when a tends to O. By (6.8-6.9) we get the
following explicit expression for the limit r, namely

(6.10)
 XIX.6 Drinfeld's Associator 467

We now compute <I? in terms of r. Consider the algebra S[a,,8] of polyno-

mials in two commuting variables a and,8 with coefficients in S. Any mono-
mial in S[a,,8] can be written uniquely as ,8P M a q where M is a monomial
in S. Define a C[[hll-linear map f' : S[a,,8] -+ S by f' (,8P M a q) = BP M Aq.
This allows us to build a C[[h]]-linear endomorphism f of S by the formula

f(r(A, B)) = f'(r(A - a, B - ,8)) (6.11)

where r(A, B) is any element of S. Observe that if M is a "diverging"

monomial in S, i.e., starting with B or ending with A, then f(M) = O.
Moreover, if M is any monomial of S, then f (M) = M +N where N is a sum
of "diverging" monomials killed by f. Therefore f2 = f is an idempotent
endomorphism of S. The following result gives an explicit expression for
<I? = <I? (A, B) in terms of the multiple zeta values of the Appendix.

Proposition XIX.6.4. We have

<I?(A, B) = f(r)
= 1+ L L hPl+oOqkT(PI,ql"",Pk,qk)f(APlBql ... APkBqk).
k21 Pl,ql,oO.,Pk,qk21
PROOF. Applying f to both sides of the relation in Lemma 6.3, we get

Since f(BM) = f(M A) = 0, we have

f(<I?) = f(lim Ga (1-

a-->O
a)) = f(r)
In order to complete the proof, it suffices to check that f(<I?) = <I? Let Ho
[resp. HI] be obtained from the solution Go [resp. from G I ] of (6.1) by
replacing A by A - a [resp. by replacing B by B - ,8]. Clearly, Ho and HI
are solutions of the differential equation

G'(z) = h
21fH
(A -z a + B - ,8)G(z).
z-1
(6.12)

Moreover, Ho is asymptotic to zh(A-a) in a neighbourhood of O. Now the

function z-iia(1 - z)-h f3 GO(z) is another solution of Equation (6.12) with
the same asymptotic behaviour as Ho. By uniqueness, we get

(6.13)

Similarly, one has

(6.14)
468 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

It follows from (6.13-6.14) that

<J.>(A - 0:, E - (3) = HI1 Ho = G11Go = <J.>(A, E).

Therefore,

J(<J.>(A, E)) = j'(<J.>(A - 0:, E - (3)) = j'(<J.>(A, E)) = <J.>(A, E). 0

As a consequence of Proposition 6.4 and of Formulas (1l.21-1l.23) of the

Appendix, we get the following expression for the first terms of <J.>(A, E).
Corollary XIX.6.5. Modulo h4 we have

<J.>(A E) == 1- ((2) [A E] h2 + ((3) ([[A E] E]-[A [A E]]) h3

, (27rA)2' (27rA)3" " .
Here ( is Riemann's zeta function. By Euler's formula ((2) = 7r 2 /6, we
see that the coefficient of h2 in the expansion of <J.>(A, E) is 214 [A, E].
Remarks 6.6. (a) If AE = EA, any monomial M -I- 1 in (6.10) can be
rewritten in the form of a "diverging" monomial, hence is killed by J. It
follows that <J.>(A, E) = 1 in this case.
(b) In [Dri90], 2, Drinfeld showed that <J.>(A, E) was the exponential of
a Lie series. He obtained the following formula for the logarithm In <J.> of
<J.>(A, E) modulo L" = [[L, L], [L, L]] where L is the completion of the free
Lie algebra generated by A and E, namely

In <J.> == L ckf! ad(Elad(A)k[A, E] hk+H2 mod L". (6.15)

k,f?O
The complex numbers ckC are given by the generating function

1+ L
k,?O
CkU k+ 1 v H1 = exp (f ((n) n (un
n=2 n(27rA)
+ v n- (u + v)n)).
(6.16)
From (6.16) we get CkC = cCk and ckO = cOk = -((k + 2)/(27rA)k+2 for
all k 2 O.

XIX.7 Construction of the Topological Braided

Quasi-Bialgebra Ag,t
In order to construct A g .t , we have to find an element <J.> = <J.>KZ in U(fl)3[[hll
verifying Relations (XVI.4.10-4.13) and Relations (XVI.4.15-4.17) with
R = R KZ = e htj2 . The element <J.>KZ we are looking for has also to in-
duce the monodromy representations of the (KZ)-systems. We proceed as
in [Dri89b], pp. 1453-1455, [Dri90], Section 2, and [Dri89c].
 XIX.7 Construction of the Topological Braided Quasi-Bialgebra Ag,t 469

We observed in Section 3 that the universal R-matrix R KZ = e ht / 2 of Ag,t

was forced upon us by the monodromy of the system (KZ 2 ). The element
<!>KZ will now come out of the system (KZ3)' In Section 6 we introduced a
formal series <!>(A, B) in two non-commuting variables A and B. We use it
to define <!>KZ'

Definition XIX.7.1. We set <!>KZ = <!>(t12' t 23 )

We claim that <!>KZ E U(g)3[[h]] satisfies all requirements for

(7.1)

to be a topological braided quasi-bialgebra. The proof of this claim will be

sketched in Section 8. We also claim that Ag,t provides the monodromy
of all Knizhnik-Zamolodchikov systems. We have already checked this for
(KZ 2 ) in Section 3.
Consider a solution W(Zl' ... ,zn) of the system (KZn)' By definition, it
satisfies the system of partial differential equations

(i=l, ... ,n) (7.2)

where h = h/(2JrH).
Lemma XIX.7.2. If W(Zl' ... ,zn) is a solution of (7.2)! then it also sat-
isfies the relations

and
l:C;i<j:C;n

PROOF. This follows from (7.2) using tij = t ji . D

As a consequence of this lemma, a solution W of (7.2) depends on n - 2

variables. In particular, a solution w(zl' Z2' z3) of (KZ 3 ) depends on one
variable z. Let us from now on focus on (KZ 3 ) and make the change of
variables
W(Zl' z2' z3) = (z3 - Zl)h(i,2+ t 23+i,3)G(z) (7.3)
where Z = (Z2 - zl)/(z3 - Zl)'

Lemma XIX.7.3. With the above notation, W(Zl,Z2,Z3) is a solution of

(KZ 3 ) if and only if G(z) is a solution of the ordinary differential equation

G'(z) = h(t12 + ~)G(z). (7.4)

z z-l
470 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

PROOF. Relation (7.3) and

~ow = h- (t12 - - W(Zl,z2,z3 )

- - - + - t23)
uZ2 Z2 - Zl Z2 - z3
imply that

Since (Z2 - z3)/(z3 - Zl) = z - 1, we get

(Z3-Z1)h(t12+t2e+it3)(GI(Z)-hC~2 + zt~31)G(z)) =0.

Consequently, G(z) satisfies Equation (7.4). Conversely, one checks easily

that, if G(z) is a solution of (7.4), then W(Zl,Z2,z3) is a solution of (7.2).
o
Equation (7.4) has been studied at length in Section 6. Let Go(z) and
G 1 (z) be the solutions of (7.4) obtained from the solutions Go(z) and G 1 (z)
of (6.1) by replacing A by t12 and B by t 23 . From Proposition 6.1 we get
unique solutions of (7.2)

(i = 0, 1) (7.5)

whose asymptotic behaviours are given by

w:o(z l' z2' z)

3
rv (z 2 - Z1 )hit2 (z 3 - z 1 )h(t 23 +t 13 ) (7.6)

when IZ2 - zll IZ3 - zll, i.e., when IZ2 - zll/lz3 - zll tends to 0, and

W1 (z l' z 2' z)
3
rv (z 3 - Z2 )ht23(Z3 - z 1 )h(t 12 +t 13 ) (7.7)

when IZ2 - z31 IZ1 - z31 In view of Definitions 6.2 and 7.1, Wo and W 1
are related by
W O(zl,z2,z3) = W 1(Zl,Z2,Z3) <I>KZ' (7.8)
Let us determine the monodromy of (KZ3)' The change of variables (7.3)
has the following property: Zl is close to Z2 if and only if z is close to O.
Similarly, z3 is close to zl or to Z2 if and only if z is close to 00 or to 1
respectively. Now consider the generator (J1 of the braid group B3 with the
parametrization given by (3.11). An immediate computation shows that

(7.9)
 XIX.8 Verification of the Axioms 471

In particular, z(O) = 1/2, z(1/2) = (1 + 3H)/5 and z(l) = -1, which

shows that under the change of variables, the generator (J 1 corresponds to
a counterclockwise half-turn around 0 in the complex plane. Similarly, the
generator (J2 of B3 corresponds to a counterclockwise half-turn around 1.
Choose a base-point in the configuration space X3 corresponding under
the change of variables to a point close to 0 in the complex plane. By
definition of the solution Go(z), it is multiplied by ehtr2 when z makes a
complete positive turn around the singularity O. Consequently, the value of
the monodromy of (KZ 3 ) on the generator (J1 is eht12/2. As for (J2' we first
have to move from a neighbourhood of 0 to a neighbourhood of 1 with the
help of <P KZ , then turn around the singularity 1 and come back to a vicinity
of O. This sets the value of the monodromy for (J2 at <PK~ e htr2 / 2 <P KZ . These
are the values of the monodromy exactly predicted by Formula (XV.4.2).
These considerations prove Part (ii) of Theorem 4.2 when n = 3.
We leave the remaining cases n > 3 to the reader. Let us only note that
pulling back the general system (KZn) along the loop (Ji of Bn parametrized
by (3.11) leads to the linear differential equation

dw
ds

This equation can be solved in terms of iterated integrals using Picard's

method of approximation recalled in the Appendix.

XIX.8 Verification of the Axioms

In order to complete the proof of Part (i) of Theorem 4.2, we are left with
showing that Ag,t is a topological braided quasi-bialgebra. Set <P = <P KZ
and R = R KZ . We have to check Relations (XVI.4.10-4.13) and (XVI.4.15-
4.17). Let us write down the as yet unproved relations, namely

(id Q9 6)6(a) = <p(6 Q9 id)6(a)<p-1 and 6 P(a) = R6(a)R-1 (8.1)

for all a E Ag,t = U(g)[[h]J,

(id Q9 E Q9 id)(<p) = 1 Q9 1, (8.2)

(6 Q9 id)(R) = <P312R13(<P132)-1 R 23 <P, (8.3)

(id Q9 6)(R) = (<P 23l )-1 R13<P213R12<P-1, (8.4)
472 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

and

(id0id0~)(<I (~0id0id)(<I = (10<1 (id0~0id)(<I (<1>01). (8.5)

Relations (8.1). Since the comultiplication ~ is coassociative and co com-

mutative and since the Lie algebra g generates Ag,t, Relations (8.1) are
equivalent to the relations

[~(2)(X), <1>] = 0 and [~(x), R] =0 (8.6)

for all x E g, where ~(2) = (~0 id)~ = (id 0 ~)~. By hypothesis, ~(x)
commutes with the 2-tensor t. By the following special case of Leibniz's
formula, we have

which implies [~(x), tn] = 0 by induction on n. Consequently,

This proves the second relation in (8.1).

Let us deal with the first one. We claim that

(8.7)

for all x E g. Indeed, for any element x of g, we have

~(2)(X) = ~(Xh2 +10 10 x = x 0 1 0 1 + ~(X)23'

implying
[~(2)(X), td = [~(x), t]12 + [1 0 1, t] 0 x = O.
One shows that ~(2)(X) commutes with t 23 in a similar way. Now, by re-
peated application of the Leibniz rule to (8.7), we see that ~(2)(X) com-
mutes with all (non-commutative) monomials in the variables t12 and t 23 .
In particular, ~(2)(x) commutes with <I> in view of Proposition 6.4 and
Definition 7.1. This proves Relations (8.1).
Relation (8.2). The element t Egis annihilated by id 0 c: and c: 0 id.
Therefore
(8.8)
Since id0c:0id is a C[[hll-linear morphism of algebras, it kills all non-trivial
monomials in t12 and t 23 . Therefore, again by definition of <1>, we have

(id 0 c: 0 id)(<I = (id 0 c: 0 id)(l 0 1 0 1) = 101.

 XIX.S Verification of the Axioms 473

Relation (8.3). We shall be sketchy. For more details, see [Dri89b], Section
3. Recall the solutions Wo and WI of the system (KZ 3) described with their
asymptotic behaviour in (7.5-7.7). By permuting ZI' Z2, Z3 we get four other
solutions W 2 , W 3 , W 4 , W5 of (KZ3)' uniquely determined by the following
asymptotic behaviour:

W 2 (ZI' z2' z3) rv (z2 - Z3)ht23 (z2 - ZI)h(h2+ t ,3) when IZ3 - z21 IZ2 - zll,

W 3 (Zl' Z2, z3) rv (z3 - Zl) ht 13 (z2 - Zl)h(h2+ t 23) when IZ3 - zll IZ2 - zll,

W 4 (zl' Z2, z3) rv (Zl - Z3) ht 13(Z2 - Z3)h(h2+ t 23) when IZ3 - zll IZ2 - z31,
W5 (zl' Z2, z3) rv (Z2 - ZI)ht12 (Z2 - Z3)h(t 13 +t 23 ) when IZI - z21 IZ2 - z31
We observe that W 2 is obtained from WI by exchanging Z2 and z3' Letting
z3 pass in front of z2 following the loop 0"2 of the braid group B3 yields

W1= W 2 e ht23/2 =
W 2R 23' (8.9)

We next remark that W 2 and W3 are solutions of (KZ 3) where t12 and t 13
have been exchanged. Therefore, by definition of <I>, we have

(8.10)

Similarly, W 4 has been obtained from W3 by having z3 pass in front of ZI'

Consequently,
W3 -- W 4 eht'3/ 2 -- W 4 R 13' (8.11)
An argument analogous to the one applied to W 2 and W3 shows that

(8.12)

Relations (7.8) and (8.9-8.12) imply

Wo = W5 <I>312R13 (<I>132)-1 R 23 <I> (8.13)

Finally, W5 is obtained from Wo by letting z3 pass in front of ZI and of z2'

Hence,
(8.14)
since t 13 + t 23 = (,6. id)(t). By uniqueness of the solutions, Relations
(8.13-8.14) imply Relation (8.3). Figure 8.1 on the next page summarizes
the movements of ZI, Z2 and z3 considered in the previous argument.
474 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Figure 8.1. The movements of Zl, Z2, and Z3

Relation (8.4). One may proceed as for (8.3). An alternate proof consists
in first showing that <1.>321 = <1.>-\ which is done by replacing Z by 1 - Z
in Equation (6.1). Then, as in the proof of Lemma XVIII.8.2, apply the
involution T 13 to Relation (8.3) and use the fact that ~ = ~ op and R = R21
to derive Relation (8.4).
Relation (8.5). In order to prove the "pentagonal" relation we now consider
the system (KZ4)' The following lemma is due to Drinfeld [Dri90], Section
2, to which the reader is referred for a proof.
Lemma XIX.S.l. There exist solutions Xl' X 2, X 3, X 4 and X5 of (KZ 4)
uniquely determined by
X 1 (z l' Z2' Z3' z)
4
rv (z 2 - Z1 )iitl2 (z 3 - z 1 )ii(t l3 +t 23 )(z 4 - z 1 )ii(t,4+ t 24+ t 34) ,

X 2 (z l' z 2' z 3' z)

4
rv (z 3 - Z2 )ii t 23 (z 3 - z 1 )ii(t l2 +t l3 ) (z 4 - z 1 )h,(tl4+t24+t34) ,
X 3 (z l' z 2' z 3' z)
4 rv (z 3 - Z2 )iit23 (z 4 - z 2 )ii(t24+ t 34) (z 4 - z 1 )ii(tl2+tl3+tl4) ,
X 4(zl,z2,z3,z4) rv (Z4 - Z3) iit 34(Z4 - Z2)ii(t 23 +t 24 )(Z4 - Zl)ii(t l 2+t13+tl4),
and
X 5(z l' z 2' z 3' z)
4 rv (z 2 - Z1)iit,2(Z4 - Z3)iit 34(Z4 - z 1)ii(tl3+t14+t23+t24) .

For Xl the sign rv means that there exists an analytic function f (u, v)
such that f(O,O) = 1 and

Xl (Zl , Z2' Z3' Z4) = f (u, v) (Z2 - Zl )iit,2 (Z3 - Zl )ii(t,3+t23) (Z4 - Zl )ii( t'4 +t2d t 34)
where u = (Z2 - Zl)/(z4 - Zl) and v = (z3 - zl)/(z4 - Zl)' The reader will
be able to give a precise meaning to rv in the remaining cases.
The "pentagonal" relation (8.5) is an immediate consequence of the fol-
lowing lemma. This completes the proof of Part (i) of Theorem 4.2.
 XIX.S Verification of the Axioms 475

Lemma XIX.8.2. Under the previous hypotheses, we have

Xl = X 2(<I> (1), X 2 = X 3 (id 0.6. 0 id)(<I, X3 = X 4(10 <1,

X 4 =X5(id0id0.6.)(<I>-1), X5=Xl(.6.0id0id)(<I>-1).

PROOF. (a) We start with the proof of the first relation Xl = X 2 (<I> 0 1).
Set
V1 ( Zl,Z2'Z3,Z4 ) ~
~
X 1 ( Zl,Z2'Z3,Z4 )( Z4 - Zl )-h(h4+ t 24+ t 34)

and

It is enough to prove that VI = V2 . By Lemma 3.2 we have

A similar computation shows that t 23 commutes with t14 + t24 + t 34 . Hence,

<1>01, which is a formal series in t12 and in t 23 , commutes with t14 +t24 +t 34 .
Therefore, V2 can be rewritten as

A simple computation shows that VI and V2 both satisfy the following

system of partial differential equations:

av
-a =
- '"""' t lj
h L.....- ---V(Zl,z2,z3,z4)
-
+ hV(zl,z2,z3,z4) t14 + t24 + t34
,
Zl jil zl - Zj z4 - Zl
(8.15)
for i = 2,3, (8.16)

and
av
-a - '"""' t4j -
h L.....- ---V(zl,z2,z3,z4) - hV(Zl,Z2,z3,Z4)
=
t14 + t24 + t34
.
Z4 ji4 Z4 - Zj z4 - zl
(8.17)
We set z4 = 00 in Equations (8.15-8.16) (this is possible since the equations
are actually defined on the complex projective line). Then V l (Zl,Z2,z3'00)
and V2(Zl,Z2,z3'00) become solutions of the system (KZ3). Moreover, by
Lemma 8.1, V l (Zl,Z2,z3'00) and V2(Zl,Z2,Z3,00)(<I>-1 (1) have the same
asymptotic behaviour as the solutions Wo and WI of (KZ 3 ) respectively.
By uniqueness of these solutions, we get

and
476 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

By definition of <I>, this implies that

(8.18)

for all ZI, z2, z3' Relation (8.18) and Equation (8.17) imply that VI and V2
coincide everywhere.
(b) To prove the second relation of Lemma 8.2, it is enough to check that
the functions U I and U2 coincide when we set

U I (zl,z2,z3,Z4) = X 2 (zl,z2,z3,z4) (z3 - Z2)-ht 23

and

The element (id Q9 6. Q9 id) ( <I is a formal series in the variables

By Lemma 3.2 again, t24 + t34 commutes with t 23 . Therefore U2 can be

rewritten as

Both U I and U 2 are solutions of the system

for i = 1,4, (8.19)

(8.20)

and

(8.21)

When Z2 = Z3' we claim that

for all ZI,z2,z4' Define T i (ZI,z2,z4) = Ui (ZI,z2,z2,z4) for i = 1,2. Equa-

tions (8.19-8.21) imply that TI and T2 are solutions of the system

;:)aT -_ h- (t12 + t I3
+ ~) ( )
T ZI,Z2,Z4 , (8.22)
uZI ZI - Z2 ZI - Z4

(8.23)
 XIX.S Verification of the Axioms 477

and

(8.24)

Now,

and
t24 + t34 = (id 6. id)(t23)'
Therefore Equations (8.22-8.24) imply that Tl and T2 are solutions of the
system (KZ 3 ) in which the coefficients tij have been replaced by new coef-
ficients (id 6. id)(t ij ). By the results of Section 7, there exist solutions
Ho and HI of this modified (KZ 3)-system such that

(8.25)

with the asymptotic behaviour

H
o(z l' z 2' z)
4
rv (z
2
- z )h(t12+h3)(Z
1 4
- z )h(t14+t24+t34)
1

when IZ2 - z41 IZI - z41 It follows from this, from Lemma 8.1, and
from the fact that t 23 commutes with t12 + t 13 , t14 + t24 + t 34 , t24 + t34
and with t12 + t 13 + tw that Tl and T2 (id 6. id)(<I-1 have the same
asymptotic behaviours as Ho and HI respectively. Consequently, Tl = Ho
and T2 (id6.id)(<I-1 = HI' Combining these relations with (8.25), we
conclude that Tl and T2 coincide. Therefore,

(8.26)

for all ZI' Z2' z4' Relation (8.26) and Equation (8.20) imply that the func-
tions U 1 and U2 coincide everywhere.
(c) The remaining relations of Lemma 8.2 are proved in a similar fashion:
in the case of the third relation, we send ZI to 00 whereas for the last two,
we have to identify z3 with z4' and ZI with Z2' respectively. The movements
of ZI' z2' z3 and z4 in this proof can be represented as a system of four
particles moving as in Figure 8.2. 0
478 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Zl Z2 Z3 Z4
Xl

X2

X3

X4

X5

Zl Z2 Z3 Z4 Xl

Figure 8.2. Five configurations of four particles

Remark 8.3. Let us consider variables (t ij h:Si<j:S4 satisfying the infinites-

imal braid group relations (2.1-2.2). Reviewing the proofs of Relations
(8.1-8.5), we see that we have actually established the existence of a for-
mal series <l>(A, B) in two non-commuting variables A and B with constant
term 1, belonging to the algebra 5 of Section 6 and satisfying the three
relations

<l>(tI2' t 23 +t 24 )<l>(t I3 +t 23 , t 34 ) = <l>(t23 , t 34 )<l>(t I2 +t I3 , t24 +t34)<l>(tI2' t 23 ),

(8.27)
(8.28)

(8.29)
Relation (8.27) is the translation of (8.5) while Relations (8.28-8.29) cor-
respond to (8.3-8.4) in view of Rij = e h~ij for 1 ~i <j ~ 3 and of

An element <l>(A, B) of 5, with constant term 1 and satisfying Relations

(8.27-8.29), will be called a Drinjeld series. Drinfeld's associator <l>KZ is the
only explicit Drinfeld series constructed so far. Drinfeld actually established
the existence of a Drinfeld series with rational coefficients (see [Dri90],
Theorems A, A', A"). It would be interesting to have a description of it,
especially in view of the constructions of XX.6-7. In case the elements tij all
come from the invariant symmetric 2-tensor t of a semisimple Lie algebra,
Drinfeld also showed in [Dri89b], Theorem 3.15 that <l> was unique up to
gauge transformation by a symmetric invariant element F.
 XIX.lO Notes 479

XIX.9 Exercises
1. Let 9 be a semisimple Lie algebra and t E 9 @ 9 be the 2-tensor
(XVII.1.6). Show that [tI2' td =f O.
2. (a) Compute Drinfeld's associator <I>(A, B) in the case that A and
B commute with the commutator [A, B].
(b) Show that <I>(A, B) is the exponential of a Lie series (Hint: prove
that Do <I> = <I> @ <I.

3. Prove that the projector f defined by (6.11) coincides with the con-
volution vB *id *v A where v A and vB are the algebra endomorphisms
of S determined by

4. Prove that there exists an analytic function V(z) defined in a neigh-

bourhood of [0,1] such that zhA(l_z) hB V(z) is a solution of Equation
(6.1). Show that <I>(A, B) = V(l)V(O)-I.

5. Let VI' ... ,vn be analytic functions. Consider the differential equation

G'(z) = (t
i=1
Aiui)G(Z) (9.1)

where AI, ... , An are non-commuting variables and ui = vUvi for

i = 1, ... ,n. Set V(z) = v,:;-A n v1A1 G(z) where G(z) is a solution
of Equation (9.1). Establish that V(z) is a solution of the equation
V'(z) = Q(z)V(z) where

XIX.I0 Notes
The material in Sections 1-2 is standard. For more on the configuration
space X n , see [Aom78] [Hai86] [Koh85].
The equations (KZn) were introduced by Knizhnik and Zamolodchikov
[KZ84] in connection with the Wess-Zumino-Witten model in conformal
field theory.
Theorem 4.1, which is the main result of this chapter, first appeared
in [Koh87] [Koh88]. For the proof we followed Drinfeld [Dri89b] [Dri90]
[Dri89c] closely. As a matter of fact, most results of Sections 4-8 are due
to Drinfeld. There is an exception in Section 6 where Proposition 6.4 is
480 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

due to Le and Murakami [LM93b]. We also used [LM93a]. As was shown

in Section 8, Drinfeld's proof that Ag,t is a braided quasi-bialgebra relies
on the asymptotic behaviour of certain solutions of the systems (KZ 3 ) and
(KZ 4 ). Kapranov [Kap93] discussed the asymptotic zones used by Drinfeld
and related them to all possible bracketings of the permutations of a finite
set of letters.
In the Appendix we collected several facts on iterated integrals which
we found in [Aom78] [Che61] [Che73] [Che75] [Che77a] [Che77b] [Gol80]
[Lap53] [Ree58] [Reu93] [Was87] [Zag93].

XIX.II Appendix. Iterated Integrals

Let WI' ... ,wn be complex-valued differential I-forms defined on a real in-
terval [a, b]. We have Wi = ii(s)ds where il"'" in are complex functions.
Define the iterated integral J:WI ... wn inductively by

(1l.1)

and
(1l.2)

if n > l. Iterated integrals enjoy the following formal properties:

(1l.3 )

for a < b < c, and

(1l.5)

where a runs over all (n, m )-shuffies of the symmetric group Sn+m'
Iterated integrals occur in the solution of certain linear differential equa-
tions. Let us consider an equation of the form

dY
- = A(s)Y(s) (11.6)
ds
where Y (s) is a differentiable function defined on the real interval [a, b],
with values in the endomorphism ring of some complex vector space and
 XIX.ll Appendix. Iterated Integrals 481

where A(s) is a linear endomorphism for each s E [a,b]. The differential

equation (11.6) has a unique solution Y(s) such that its initial value Y(a)
is the identity. Picard's method of approximation leads to the following
formal expression for Y (s):

(11. 7)

where the family (Qp)p~o is defined inductively by Qo = id and for p > 0

by
(11.8)

Equivalently, Qp can be defined as an integral over the real p-simplex

by
Qp(s) = r
Jtl.p(a;s)
A(SI)A(S2) ... A(sp)ds I ds 2 ... ds p. (11.9)

We now wish to apply Picard's method to the differential equation

dY = t Aj Y(s)
(11.10)
ds j=1 S - aj

where AI' ... , An are constant linear endomorphisms and aI' ... , an are dis-
tinct complex numbers lying outside the real interval [a, b]. By (11.1-11.2)
and (11.7-11.9) the unique formal solution Y(s) of (11.10) with Y(a) = id
is given by the formal series

Y(s) = id+ L (11.11)

where the complex functions La (all' ... , a jr Is) are defined as the following
iterated integrals

(11.12)

Functions of this kind already appeared in [Poi84], III and were investigated
at length by Lappo-Danilevsky in [Lap53], Memoire II under the name
"hyperlogarithms" .
We now concentrate on the hyperlogarithms built on the particular 1-
forms
no = 27ry11-11dsS- and ~fIIl _- 27ry11-11 ~
s- 1
. (11.13)
482 Chapter XIX. Monodromy of the Knizhnik-Zamolodchikov Equations

Iterated integrals of Do (or of D1 ) are easily computed. For instance, by

induction on k, we get

(11.14a)

I
and
b
Dk _ 1 (10 1- b) k (11.14b)
a 1 - (27rR)kk! g 1- a

when 0 < a < b < 1.

Now, as in (11.11-11.12) we wish to consider "mixed" iterated integrals
fol WI ... wn where each of WI' ... , Wn may be either Do or D1 If WI = Dl
or if wn = Do, the integral fol wI' .. Wn does not converge. However, it does
in the remaining cases. Set

(
T Pl,Ql"",Pk,qk
) = Jrl "p, "q,
HO HI
"Pk "qk
... HO HI (11.15)
o

where PI' Ql' ... , Pk' Qk are integers> O. We shall now compute the iterated
integrals (11.15) in terms of series reminiscent of Riemann's zeta function.
To this end, we introduce the convergent series

(11.16)

where i 1 , ... , i k are positive integers, x is a real number such that 0 <
x < 1, and m 1 , ... , mk run over the set of positive integers. The special
case L( n; x) is the n- th polylogarithm which appears in number theory,
geometry and algebraic K-theory. When n = 1, we have

L(I;x) = L -xm r
= -log(1- x) = -27rvCI In
1
D1 (11.17)
O<mm 0

Taking the derivative of L(i 1 , . . . , i k - 1, ik ; x), we get

dL(i 1 , , i k _ 1 , i k ; x) L(i 1 , ,i k _ 1 ,i k -1;x)

dx x
when i k > 1. Hence,

(
Li 1, ... ,i k- 1,i k ;X)=Jo
r L(i ,,i
1 k _ 1 ,i k
s
-l;s)
ds. (11.18)

If i k = 1 we have
dL(i 1 ,,i k _ 1 ,I;x)
dx
 XIX.ll Appendix. Iterated Integrals 483

Set m = mk - mk-I - 1 ~ O. Then

dL(i j , i k- I, 1; x)
dx
L(il,,ik_I;X)
I-x

It follows that

l
x L(il,,ik_I;S)
L(i l ,,i k_ l ,l;x)= 0 -----'---ds.
l-s
(11.19)

We define the multiple zeta value ((il"'" i k ) by

(11.20)

The special case (( i l ) coincides with the value of Riemann's zeta function
at the positive integer i l . An easy induction using (11.15-11.20) expresses
the mixed iterated integrals T(PI' ql, ... , Pb qk) in terms of multiple zeta
values. To be precise, we get
( _l)Ql+'+Qk
T(PI' ql" .. ,Pk' qk) = (21TR)Pl +ql +"'+Pk+qk

x ((1, ... , 1, Pk + 1,1, ... , 1, Pk-I + 1, ... , 1, ... , 1,PI + 1)

(11.21)
where the first set of l's is of length qk -1, the second one of length qk-I -1,
... and the last one is ql -1 long. In particular, if ql = ... = qk = 1, we get
an expression for the multiple zeta values in terms of the mixed iterated
integrals (11.15), namely

((iI' ... ,i k ) = (_l)k (21TvCI)il ++ik T( i k - 1, 1, i k - I - 1,1, ... , i l - 1, 1).

(11.22)
As a consequence of (11.3), we get the inversion formula

(11.23)

Relations (11.21-11.23) imply ((1,2) = ((3). We used the last equality to

derive Corollary 6.5.
Chapter xx
Postlude. A Universal Knot
Invariant

In Section 1 we present the concept of a knot invariant of finite type and

prove that all quantum group invariants are of finite type. Then we con-
struct a universal knot invariant Z(K) of finite type, with values in a com-
mutative algebra built on pairs of points on a circle. We also show that the
quantum group invariants of XVII. 3 can be recovered from Z(K) in a sim-
ple combinatorial way. The proof of this fact, as well as the construction of
Z(K), use the formalism of the KZ-equations and Drinfeld's results stated
in XIX.4.
These new, fascinating developments have now reached a state of clarity
and simplicity which allows us to conclude this book with a brief account.

XX.1 Knot Invariants of Finite Type

We start with singular knots. Consider an immersion f of the circle 51 into
the 3-dimensional oriented Euclidean space R 3 . Assume that for any m in
the image of f, the cardinality of r1(m) is 1 or 2. If it is 1, the point m
will be called an ordinary point; if it is 2, the point m will be called a double
point. We restrict to immersions with finitely many double points, and such
that locally at any double point both branches meet transversally. We also
assume that the image of f comes with an orientation. These conditions
define a singular knot. If the singular knot is equipped with a framing (as
defined in X.8), we say that the singular knot is framed. Singular links and
framed singular links are defined as immersions of a finite number of circles
 XX.1 Knot Invariants of Finite Type 485

with similar restrictions on the singularities. Singular knots and links are
represented by planar singular knot and singular link diagrams defined in
the same way as ordinary knot and link diagrams are (see X.3). One also
has an obvious notion of isotopy generalizing the one introduced in X.1.
Now any double point in a singular link diagram can be "desingularized"
by locally replacing the pattern X formed by the double point and the two
downwards oriented branches, by the patterns X+ and X_ described in
X.4. This observation allows us to extend any isotopy invariant of links to
any singular link. Indeed, let P be such an invariant with values in some
complex vector space V. Then the rule

(1.1)

defines the invariant P on the set of isotopy classes of all singular links
with one double point. Here L is a link diagram with one double point and
the ordinary link diagrams L+ and L_ are obtained from L by replacing
a neighbourhood X of the double point by X+ and X_, respectively. By
induction on the number of double points we may extend P to an isotopy
invariant for all singular links (with a finite number of double points).
Definition XX. I. I. Let m be a non-negative integer. An isotopy invari-
ant of oriented links is an invariant of degree ::; m if it vanishes on all
singular links with more than m double points.
There is a similar definition for framed links. Observe that an invari-
ant P is of degree 0 if and only if we have P(L+) = P(L_) on all link
diagrams, which means that the invariant P does not distinguish between
under crossings and overcrossings. Therefore it depends only on the number
of connected components of the link. In padicular, P is constant on the
space of all ordinary knots.
The first question we wish to address is the following: Are there any
non-trivial examples of finite-degree invariants of higher degree? Before
we answer this question, let us say that a (framed) link isotopy invariant
P(L) = L:m>O Pm(L) hm with values in V[[h]J, where V is a complex
vector space, IS of finite type if, for all m 2: 0, the V -valued invariant Pm is
of degree::; m. We now state the main source of invariants of finite type.
Proposition XX.I.2. Let P be a link isotopy invariant with values in
V[[h]] where V is a complex vector space. If, for any link L, we have
P(L+) == P(L_) modulo h at any crossing point of a link diagram of L,
then P is of finite type.

PROOF. Define V-valued invariants Pm by P(L) = L:m>O Pm(L)h m. We

have to show that each Pm is of degree ::; m. Let L be a singular link
diagram with one double point. By Relation (1.1) and by the assumption
on P, the series P( L) is divisible by h. An easy induction on m shows that
P(L) is divisible by hm+l whenever L is a singular link with m + 1 (or
486 Chapter XX. Postlude. A Universal Knot Invariant

more) double points. Consequently, the coefficient Pm (L) is zero on such a

singular link. 0

As a consequence of Proposition 1.2, all quantum group invariants are

of finite type. Recall from XVII.3 that a quantum group invariant is asso-
ciated to any semisimple Lie algebra 9 and any finite-dimensional simple
g-module V.

Corollary XX.1.3. The isotopy framed link invariant Q g,v of XVII. 3 is

of finite type.

PROOF. By construction of Qg,v, the congruence Rh == 1 1 implies that

Qg,v(L+) == Qg,v(L_) modulo h at any crossing point (one may also use
(XVII.3.3)). Then use Proposition 1.2. 0

Proposition 1.2 may also be applied to the Jones-Conway polynomial and

hence to the Alexander and the Jones polynomials which are specializations
thereof. Recall from X.4 that the Jones-Conway polynomial satisfies the
skein relation
(1.2)
for any Conway triple (L+,L_,L o ). Replace x and y by formal series x(h)
and y(h) in h such that

x(h) - 1 == y(h) == 0 modh.

Then Relation (1.2) simplifies to P(L+) == P(L_) modulo h, which allows

us to apply Proposition 1.2. In particular, using Taylor expansions, we
see that the m-th derivative of the Jones polynomial [resp. of the Conway
polynomial] at the point 1 [resp. at the point 0] is an invariant of degree
:::; m. The invariants <I> m,q of Proposition X.4.7 give also rise to invariants
of finite degree.
For some open questions on finite-degree invariants, see [BN92], Section 7
and [Bir93], Section 8.

XX.2 Chord Diagrams and Kontsevich's Theorem

Let us restrict to knots in this section. Given a complex vector space V
and a non-negative integer m, we denote by v(m) (V) the vector space of
all knot invariants of degree:::; m with values in V. Since

v(m) (V) = v(m) V (2.1)

where v(m) = v(m) (C), it is enough to consider complex-valued invariants.

We have the following inclusion of vector spaces

yeO) c v(1) c ... c V(m-I) c v(m) c ....

 XX.2 Chord Diagrams and Kontsevich's Theorem 487

One similarly defines the vector space vt) of complex-valued framed knot
invariants of degree :S m. In this section we shall show that the spaces v(m)
and vjr;:) are all finite-dimensional and give a combinatorial description of
the quotients v(m)/v(m-l) and V(m)/V(m-1)
fT fT
To this end, we need the notion of a chord diagram on a circle: it is a
finite set of unordered pairs of distinct points on the circle considered up to
homeomorphisms preserving the orientation. To specify a pair one draws
a dashed line, called a chord, between the two points. Given 2m distinct
points there are
(2m - I)!! = 1 35 .. (2m - 1)
different ways to pair them. Indeed, given one point among 2m, we may
pair it with (2m - 1) points. Take another point among the remaining
(2m - 2); it may be paired to (2m - 3) other points, etc.
There is a relationship between invariants of finite degree and chord
diagrams on a circle which we explain now. Let D be a chord diagram
on the circle with m chords (Le., with 2m points paired two by two). By
an embedding of D into R3 we mean any singular knot f : Sl ....... R3 with
exactly m double points such that f(s) = f(s') if and only if s = s' or sand
s' are the two endpoints of a chord in D. There always exists an embedding
K D of D. If K'v is another embedding of D, then it can be obtained from
K D by a series of operations consisting in replacing an undercrossing by an
overcrossing and vice-versa. Suppose we are given a complex-valued knot
invariant P of degree :S m. Since P vanishes on singular knots with at least
m+ 1 double points, P remains constant by Rule (1.1) under the operations
transforming K D into K'v, which means that P( K D) is independent of the
embedding of D chosen to compute it.
Define Em as the complex vector space with a basis given by all chord
diagrams on the circle with m chords. The dimension of Em is finite and
:S (2m - I)!!. Then the evaluation of an invariant of degree :S m on an
embedding of a chord diagram with m chords gives rise to a pairing
< , > : v(m) Em ....... C. (2.2)

Suppose that < P, D > = 0 for all chord diagram with m chords. Since any
singular knot with m double points can be represented as an embedding
of a chord diagram, we see that P vanishes on all singular knots with m
double points, which means that P is an invariant of degree :S m - 1.
Consequently, the map P 1-+ < P, - > induces an injection
Y m : v(m) /v(m-l) ....... Hom(Em, C) (2.3)
of the quotient v(m) /v(m-1) into a finite-dimensional space. A similar
argument works for vjr;:) /Vjr;:-l). We get the following result.

Proposition XX.2.1. The spaces v(m) and vjrr:) are finite-dimensional.

488 Chapter XX. Postlude. A Universal Knot Invariant

PROOF. We have already noted that knot invariants of degree 0 are con-
stant. Therefore, V(O) = vj~ = C, which proves the assertion for m = O.
The rest follows by an easy induction on m using the injection (2.3). 0

Actually, the proof of Proposition 2.1 shows that the dimensions of v(m)
and vjr;) are bounded by 1 + 2::Z"=1 (2m - 1)!!.
What we next aim to, is to restrict the size of the image of the map
Ym . More precisely, we shall show that any linear form in the image of Ym
satisfies an important four-term relation. Let D be a chord diagram with
m - 2 chords. Consider the four pictures in Figure 2.1 involving each 2
chords.

v v v v v v v v v v v v

Figure 2.1. The chord diagrams defining the four-term relation

Denote by D 1 , D 2, D3 and D4 the chord diagrams obtained by adding to

D successively the pictures of Figure 2.1 at the same place (the vertical lines
in the pictures represent portions of the circle carrying the chord diagram).
We claim the following.

Proposition XX.2.2. (a) If P is an element ofv(m) or ofVjr;), we have

< P, Dl > - < P, D2 > + < P, D3 > - < P, D4 > = 0 (2.4)

for any chord diagram D with m - 2 chords.
(b) Any element P of v(m) vanishes on any chord diagram with an
isolated chord, i.e., a chord that does not intersect any other one in the
diagram.
Relation (2.4) is called the four-term relation for invariants of finite de-
gree.
PROOF. (a) Let K 1 , K 2 , K3 and K4 be singular knots differing locally by
the pictures in Figure 2.2. They are embeddings of chord diagrams D 1 , D 2 ,
D3 and D 4 as described above.

Figure 2.2. The local differences of K 1 , K 2 , K3, and K4

 XX.2 Chord Diagrams and Kontsevich's Theorem 489

Then the left-hand side of (2.4) equals

(2.5)

To compute (2.5) we apply Relation (1.1) to all double points in Figure

2.2. Then (2.5) becomes a sum of 16 terms where all 23 = 8 possible
configurations of three crossing points occur twice with opposite signs.
(b) Let K be an embedding of the chord diagram D. Let us focus on
the double point of K corresponding to the isolated chord. It separates the
knot into two distinct unrelated parts. From this we see that K+ and K_
are isotopic. It follows from Relation (1.1) that

The argument in Part (b) of the proof above does not work for framed
knot invariants since K+ and K_ are not necessarily isotopic as framed
knots.
The relations in Proposition 2.2 appear as universal relations satisfied by
all invariants of finite degree. We may wonder whether there are more such
relations. The answer is negative. In order to make this more precise, we
define a vector space Am as the quotient of Em by the subspace generated
by all elements of the form

(2.6)

where D is any chord diagram on the circle with m - 2 chords and D 1 , D 2 ,

D 3 , D4 have been defined above. If m = 0,1 we set Am = Em' Moreover,
we define Am as the quotient of Am by all chord diagrams with isolated
chords.
As a consequence of this definition and of Proposition 2.2, we see that
Y m embeds V(m)/V(m-l)
]1']1'
in Hom(A m' C) and also embeds v(m)/v(m-l)
in Hom(Am, C). The following deep theorem due to Kontsevich [Kon93]
signifies that Am and Am capture all universal relations for knot invariants
of finite degree.

Theorem XX.2.3. The maps Y m

v(m)/v(m-l) ---+
]1']1'
Hom (A m' C) and v(m) /v(m-l) ---+ Hom(Am, C)

are isomorphisms.

Let us explain Kontsevich's proof of Theorem 2.3. Define

(2.7)
m;:,D m;:,D
490 Chapter XX. Postlude. A Universal Knot Invariant

We also need the direct products

(2.8)

We shall treat an element of A or of A as a formal series Lm::::o Dm h m

where Dm belongs to Am or to Am' With this notation fixed, Kontsevich
[Kon93] assigns to each framed knot K an element

Z(K) = L Zm(K) h m E A (2.9)

m::::O
with the following properties:
(i) the map K 1-+ Z(K) is a framed knot invariant of finite type, i.e., Zm
is of degree:::; m for all m ~ O.
(ii) for each chord diagram D with m chords and each embedding K D
of D, we have
(2.10)

The invariant Z(K) and its image Z(K) in A are called the Kontse-
vich universal (fmmed) knot invariants. We now use Z(K) to prove Theo-
rem 2.3.
Proof of Theorem 2.3. We assume the existence of such an invariant Z. Let
w be a linear form on Am' By Property (i) above, Pw(K) = w(Zm(K)) is
an element of vjr;l.Define a map Xm from Hom(Am, C) to vjr;l/V)';-ll
by composing w 1-+ Pw with the projection onto vjr;l /vjr;-ll. By (2.10),
we have

< Pw,D >

Pw(KD )
w(Zm(Kd) = w(D)
for any linear form w on Am and any chord diagram D with m chords.
This shows that Ym 0 Xm = id. Consequently, Ym is surjective. We already
know it is injective. Therefore, both Ym and Xm are isomorphisms. There
is a similar proof for (unframed) knot invariants. 0

As a consequence of the proof of Theorem 2.3, we see that Xm 0 Y m = id,

which means that, if P is any framed knot invariant of degree:::; m, then
there exists a unique linear form w = Ym(P) on Am such that P - w 0 Zm
is an invariant of degree :::; m - 1. It follows by induction that

for a unique family of linear maps (wi: Ai ----+ C)O<i<m' Consequently, for
any framed knot invariant P = Lm::::o Pmh m of finite type, with values in
 XX.3 Algebra Structures on Chord Diagrams 491

the formal series algebra C[[h]], there exists a unique linear map w from A
to C[[h]] such that
P(K) = w(Z(K))
for all framed knots K. The bijection set up by Z(K) between framed
knot isotopy invariants of finite type and linear maps defined on A justifies
the qualifier "universal" for Z(K). We have a similar formulation for knot
invariants after replacing A by A: and Z(K) by Z(K). In particular, any
quantum group invariant can be obtained in this way (we shall give details
in Section 8).
Kontsevich's original definition ofthe universal knot invariant Z(K) used
complicated multiple integrals depending on the realization of the knot as
a smooth curve in the three-dimensional space. In Section 7 we shall give a
combinatorial construction of Z(K) using a planar diagram of the knot and
category theory in the spirit of what we did in Chapters X, XII, XIV.5.1
and XVII.3. The combinatorial construction is due to Cartier [Car93], Le-
Murakami [LM93c] and Piunikhin [Piu93].

XX.3 Algebra Structures on Chord Diagrams

We extend the notion of a chord diagram. Let T be an oriented (framed)
tangle as defined in X.5. A chord diagram on T is a finite set of unordered
pairs of distinct points of T \ aT (where aT is the boundary of the tangle).
Again, as in the previous section, the pairs, called chords and represented
by dashed lines, are considered up to homeomorphisms preserving each
connected component and the orientation of the tangle.
Let E(T) be the complex vector space with a basis given by all chord
diagrams on T. The vector space E(T) has a grading

E(T) = EB Em(T) (3.1)

m~O

where Em(T) is spanned by all chord diagrams with m chords. The sub-
space Eo(T) is the one-dimensional subspace spanned by the unique chord
diagram without chords. If f : T ---+ T' is a homeomorphism of tangles,
then f sends any chord diagram on T to a chord diagram on T', thus in-
ducing an isomorphism E(T) ~ E(T'). In particular, since any tangle is
homeomorphic to an "unknotted" tangle, the isomorphism class of E(T)
depends only on the number of circles and segments composing T.
Let T and T' be tangles such that s(T) = b(T') in the notation of X.5
and XII.2. Under this condition, the composition ToT' is defined. Placing
a chord diagram of T on top of a chord diagram of T', we get a chord
492 Chapter XX. Postlude. A Universal Knot Invariant

diagram of ToT'. This construction extends to a linear map

E(T) Q9 E(T') ----7 E(T 0 T') (3.2)

sending Em(T) Q9Em,(T') into Em+m,(ToT').

We now define the vector space A(T) as the quotient of E(T) by the four-
term relation (2.6) that already served us to define Am in Section 2. Here
again the pictures in Figure 2.1 have to be understood as local modifications
of a chord diagram with the vertical full lines representing portions of the
tangle. The graded structure of E(T) passes to A(T) and we have

A(T) = E9 Am (T) (3.3)

m;o.()

where m counts the number of chords.

Similarly, one defines A(T) = EBm>O Am (T) as the quotient of A(T)
by all chord diagrams with isolated dlords. When T is a circle, we have
isomorphisms
(3.4)
Since the four-term relation is local, the composition (3.2) induces linear
maps

A(T) Q9 A(T') ----7 A(T 0 T') and A(T) Q9 A(T') ----7 A(T 0 T'), (3.5)

defined when s(T) = b(T'). The maps (3.5) preserve the gradings.
Next put a graded algebra structure on the vector space A = EBm>O Am'
Consider the braid In with n > 0 vertical segments oriented downwards
(defined in X.6). In the tangle category, In is the identity of the sequence
consisting of n +-signs. Since In 0 In = In the maps (3.2) and (3.5) yield
algebra structures on E(1n), A(I.,J and A(1rJ whose units are the chord
diagrams without chords.
We use these algebra structures to produce a family of elements of A(lrJ
satisfying the infinitesimal braid group relations (XIX.2.1-2.2). For integers
1 <:::: i =f j <:::: n, let tij be the unique chord diagram on In with a single
chord between the i-th and the j-th strands. We have t ji = tiJ by definition.
Using the algebra structure of E(lrJ we also have

(3.6)

whenever i, j, k, f. are distinct. In view of the definition of the product on

E(ln) and on A(1n), and of the four-term relation (2.6), we have
[tij, e k + tjk] = tije k _ tikt ij + tijt,ik - tikt ij =0 (3.7)

in the quotient algebras A(1n) and A(ln) when i,j, k are distinct integers.
Consequently, the classes of the elements (tijh<;i<j<;n satisfy the infinites-
imal braid group relations in A(1n) and in A(1n).
 XX.3 Algebra Structures on Chord Diagrams 493

The algebras A(1n) and A(ln) also have bialgebra structures. The co-
multiplication ~ is given by the formula

~(D) = L D' D" (3.8)

0C;D'C;D

where D' runs over all sub diagrams of D including the chordless diagram
and D" is the subdiagram complementary to D' in D. The co unit is zero
on all chord diagrams with at least one chord and is 1 on the chordless
diagram 0. The reader may check that the comultiplication and the counit
are well-defined on A(IT/) and A(1n) and satisfy all the required axioms.
Observe that ~ is cocommutative. Actually, these bialgebras are Hopf al-
gebras as are all graded bialgebras whose zero-th part is equal to C. The
antipode S is defined inductively on the number of chords by S(0) = 0 and

S(D) = -D - L S(D')D". (3.9)

0#D'#D

We now consider the special case n = 1 and denote A(11) and A(1 1) by
AU) and AU) respectively. The following lemma holds in A(l).

Lemma XX.3.l. Let D be a chord diagram on 1= 11 with at least two

chords. Let p be the highest point of D and {p, q} be the corresponding
chard. Let p' be a point of 1 not in D and lower than all points of D.
Define a new chord diagram D' by D' = (D \ {p, q}) U{p', q}. Then D and
D' define the same element in A(l).

PROOF. We first reformulate the four-term relation (2.6). Let D be a chord

diagram on 1 with at least one chord a = {y, z} and let q be a point of 1
not in D. Define four points Xl' X 2 , X 3 ' x 4 by their heights ht(x i ) as follows:

ht(x 1 ) = ht(y) + E, ht(x 2 ) = ht(y) - E,

ht(x 3 ) = ht(z) + E, ht(x 4 ) = ht(z) - E

where E is a positive, small enough, real number. Consider the diagrams

D~q = DU{ q, xJ for i = 1, ... ,4. With this notation the four-term relation
(2.6) translates into the relation

(3.10)

The proof of the lemma now follows from (3.10) and from the equality

where the sum is taken over all chords a of 15 = D \ {p, q}. D

494 Chapter XX. Postlude. A Universal Knot Invariant

Corollary XX.3.2. (a) The algebra A(l) is commutative.

(b) Closing the braid 1 induces an isomorphism

PROOF. (a) This results from a repeated application of Lemma 3.l.

(b) Let D be a chord diagram on the circle. By slitting the latter at some
point distinct of the endpoints of the chords of D we get a chord diagram
on 11' Lemma 3.1 shows that this is independent of the place where the
circle is slit. D

We already know that A(l) has a co commutative Hopf algebra structure.

It is also commutative by the previous result. We use the isomorphism of
Corollary 3.2 to transport this structure on A. Now, by a well-known re-
sult of Milnor and Moore [MM65], any commutative co commutative Hopf
algebra A over a field of characteristic zero is a symmetric algebra over the
subspace Prim(A) = EBm>O Prim(A)m of primitive elements. This applies
to the isomorphic Hopf algebras A and A(l). Despite the fact that the al-
gebras A and A(l) are polynomial algebras, not much is known about their
generators, not even dm = dim(Prim(A)rr,}, which is the (finite) number
of generators of A in degree m. The dimension dm has been computed in
degrees up to m = 8. According to [BN92]' Section 6 we have the following
table for dm .

1 2 345 6 7
1 1 123 5 8
A final observation is in order: denote by C the image in A of the unique
chord diagram with one single chord and by (C) the two-sided ideal it
generates. We have A = Aj(C) and A ~ A[C].

XX.4 Infinitesimal Symmetric Categories

Let S = (S, @,I) be a strict tensor category whose sets of morphisms
Horns (V, W) are all complex vector spaces and where the composition and
the tensor product of morphisms are C-bilinear maps. We assume that S
is symmetric with an involutive braiding (CTv,w)v,w'
Definition XX.4.1. Under the previous hypotheses, define an infinitesi-
mal braiding on S as a family of functorial endomorphisms in S

tv,w : V @ W ---+ V @ W, (4.1)

defined for all pairs (V, W) of objects of S, such that

CTv,W 0 tv,w = tw,v 0 CTv,w, (4.2)

 XX.4 Infinitesimal Symmetric Categories 495

and
tu,vw = tu,v 0 id w + (au,v 0id w )-1 0 (id v 0tu,w) 0 (au,v 0id w ) (4.3)
for all objects U, V, W in S.
A symmetric category as above equipped with an infinitesimal braiding is
called an infinitesimal symmetric category.
Observe that in view of (4.2), Relation (4.3) is equivalent to

tuv,w = idu0tv,w+(idu0av,w)-1o(tu,w0idv)o(idu0av,w). (4.4)

Relations (4.3-4.4) are infinitesimal versions of the relations (XIII.1.5-1.6)
defining a braiding in a braided tensor category. Indeed, suppose we have
a braided tensor category in which the morphisms depend on a formal
parameter h and, in particular, the braiding cv,w is of the form

cv,w = av,w (idvw + htv,w + terms of higher degree in h)

for some symmetry a V W An immediate computation shows that if Cv w

satisfies Relations (XIII. 1.5-1.6) , then the endomorphisms tvw satisfy Re-
lations (4.3-4.4). '
If, in addition, the infinitesimal symmetric category S has a left duality
V f---> V* with structure maps b~ : I --+ V 0 V* and d~ : V* 0 V --+ I (as
defined in XIV.2), then the infinitesimal braiding is of the form

(4.5)

where (Cv : V --+ V)v is a natural family of endomorphisms of S defined

by
(4.6)
Let us give an example of an infinitesimal braiding. We know that if H =
(H,~, c, S) is a cocommutative Hopf algebra, then the category H-Mod of
H-modules is a symmetric tensor category, with the flip as symmetry. Let
Prim(H) be the vector space of primitive elements in H. We have the
following characterization of infinitesimal braidings on H-Mod.
Proposition XX.4.2. (a) Let t be an element of Prim(H) 0 Prim(H)
satisfying the conditions t21 = t and [~(a), tJ = 0 for all a E H. For any
pair (V, W) of H -modules define the endomorphism tv,w of V 0 W by
tv,w(V 0 w) = t(v 0w) (4.7)
where v E V and w E W. Then (tv w ) v w is an infinitesimal braiding on
the category H-Mod. "
(b) Conversely, any infinitesimal braiding (tvw)vw on H-Mod is of
the form (4.7) with t = tH H(1 0 1) E H 0 H. The 'element t belongs to
Prim(H) 0 Prim(H) and s~tisfies the two conditions of Part (a).
496 Chapter XX. Postlude. A Universal Knot Invariant

PROOF. Part (a) follows by direct checking. To prove Part (b) we proceed as
in the proof of Proposition XIII. 1.4. The functoriality of the infinitesimal
braiding forces it to be of the form (4.7) with t = t H H (1 @ 1). The H-
linearity of the infinitesimal braiding implies that [,6. (~), t] = 0 for all a
in H. Conditions (4.2) and (4.3) yield t21 = t and (id@ ,6.)(t) = t12 + t 13
respectively. The fact that t belongs to the subspace generated by primitive
elements follows from an argument already used in the proof of Proposition
XVI.5.2. 0

Let us restrict to the subcategory H -Mod f of finite-dimensional H-

modules. It has left duality. If H-Mod has an infinitesimal braiding induced
by the element t = Li Xi @Yi where xi' Yi are primitive, then the endomor-
phisms Cv of (4.6) are induced by the action of a single element, namely
C = Li xiYi E H. This follows from (4.6) and the fact that the antipode
of a primitive element is equal to its opposite.
We may apply Proposition 4.2 to the enveloping algebra H = U(g) of
a semisimple Lie algebra 9 with t E 9 @ 9 being equal to the symmet-
ric invariant 2-tensor (XVII. 1.6). In this case, C is the Casimir element
(XVII. 1.5).
We shall need the following result in Section 5.

Lemma XX.4.3. If U, V, Ware objects of an infinitesimal symmetric cat-

egory with symmetry (o-v,w)v,w and infinitesimal braiding (tv.w )v.w! then
we have
[t UY @ id w , cr- 1 (t u ,w @ idv)cr + id u @t v .w ] = 0 (4.8)
where cr = id u @ crv,w'

PROOF. The square

U@V@W t u 09 V W U@V@W
1 1
)

tu.vidw tu,vidw
U @ V @ W tu 09 V.w) u@v@w

commutes by functoriality of the infinitesimal braiding. In other words, we

have
[t UY @id w , tuv,w] = O.
Replacing tuv,w by its expression in RelatiGIl (4.4) yields (4.8). 0

XX.5 A Universal Category for Infinitesimal

Braidings
We now construct an infinitesimal symmetric category AB of special inter-
est. The objects of the category AB are the objects of the braid category
B of XIII.2, namely nonnegative integers. A morphism in AB is an element
 XX.5 A Universal Category for Infinitesimal Braidings 497

of the complex vector space AhoT(T) for some braid T where AhoT(T) has
the same definition as A(T) (see Section 3), except that we allow only hor-
izontal chords. The source [resp. the target] of such a chord diagram is the
sequence s(T) [resp. the sequence b(T)] defined in X.5. The composition of
morphisms is given by the map (3.5). The identity of an integer n is the
chord less diagram on the braid In (defined in X.6).
We put the same tensor product on AB as the one we put on the braid
category, namely, we have n @ m = n + m on objects while the tensor
product of morphisms is defined by placing chord diagrams side by side.
The tensor product is well-defined and strictly associative with unit I = O.
The braiding (XIII.2.1) of the braid category induces a braiding on the
category AB: it suffices to take the chord less diagrams on the correspond-
ing braids. Since we are considering braid chord diagrams up to homeo-
morphisms, we see that this braiding is symmetric in AB although it is not
in the braid category.
Given objects n, m of AB, define an endomorphism tn,m of n@m = n+m
as follows. If n or m = 0, set tn,m = O. Otherwise, set
n m
t n, Tn = ""'
L..-t ""'
L-t ti,n+j (5.1)
i=l j=l

where tij is the chord diagram (already defined in Section 3) with a unique
chord between the i-th and the j-th strands.

Proposition XX.5.l. The family (tn.m)n.m>O is an infinitesimal braiding

on the category AB. -
PROOF. Relations (4.2-4.3) are easy to check. It remains to prove that the
family (tn.m)n.m is functorial with respect to all morphisms of AB. Since
the category is symmetric, it is enough to show that the square
tn ~
n@m -~

If0id~ (5.2)
tn =
n@m ~

commutes for all morphisms f. Now the endomorphisms of n in AB are

clearly generated by the generators 0'1"'" O'n-1 of the braid group Bn and
by the chord diagrams (tij)l~i<j~n of A h or(1rJ. Therefore, it suffices to
check the commutativity of (5.2) when f is of type O'i and when it is of
type tij. This is easy in the first case. In the second case, using Relations
(4.3-4.4), we see it is enough to consider the case n = 2, m = 1 and f = t 12 .
We have

and
498 Chapter XX. Postlude. A Universal Knot Invariant

Consequently, the commutativity of (5.2) in this special case is equivalent

to the relation
[t 12 , t 13 + t 23 ] = O. (5.3)
The latter follows from (3.7). o
This proof shows that the four-term relation is imposed by the naturality
of the family (tn.m)n m>O in the category AB. We now state a universality
property for AB whi~h IS the infinitesimal analogue of Corollary XIII.3.8.

Proposition XX.5.2. Let S be an infinitesimal symmetric category with

symmetry ((J v w) v wand infinitesimal braiding (tv w ) v w' For any object
V of S there 'exist~ a unique braided strict tensor functor Fv : AB ---7 S
such that

Fv(l) = V and FV(t 12 : 1 1 ---7 1 1) = tv,v' (5.4)

PROOF. One proceeds as for Lemma XIII.3.5. Define Fv on the generators

(tijh:;i<j:;n by

F v (t ij) -- ((J i j )-l('d

1 V0(i-l) t 'd
v,v 1 V0(n-i-l) ) (J ij (5.5)

where (Jij = id v0i (JV 0 (J-i-l) v id v0 (n-j). We have to check the relations
defining the morphisms of AB, including (3.6-3.7). Relation (3.6) is clear
while Relation (3.7) follows from (4.8). 0

Using Proposition 5.2, one may derive an equivalence between the cat-
egory S and a category of braided tensor functors preserving infinitesimal
braidings from AB to S.

XX.6 Formal Integration of Infinitesimal

Symmetric Categories
We review a categorical construction due to Cartier [Car93]. Let <I> be a
Drinfeld series as defined in Remark XIX.8.3, for instance Drinfeld's asso-
ciator <I>KZ'
Given an infinitesimal symmetric category S with symmetry ((Jvw)vw
and infinitesimal braiding (tvw)vw, we con-;truct a braided tens~r cate-
gory S[[h]] as follows. The objects' of S[[h]] are the same as the objects of
S. A morphism from V to Win S[[h]] is a formal series 2:71>0 fnhn where
fo,f1,f2"" are morphisms from V to W in S. The composition in S[[h]]
extends the composition in S and the multiplication of formal series. The
identity of V in S[[h]] is the constant formal series id v .
Theorem XX.6.1. Under the previous hypotheses, there exists a unique
structure of braided tensor category on S[[h]] such that the tensor product on
objects and the unit are the same as in S, the tensor product on morphisms
 XX.7 Construction of Kontsevich's Universal Invariant 499

extends C[[h]]-linear/y the tensor product in S, the associativity constraint

a is given by
au.vw, = <P(tu v' t vw ), ,~
(6.1)
and the braiding c is given by
cV,W =0"
V,w oe htv ,w/ 2 . (6.2)
PROOF. We have to check the Pentagon Axiom (XI.2.6) and the Hexagon
Axioms (XIII. 1.3-1.4). Now this follows from Relations (XIX.S.27-S.29)
:,mtisfied by <P. 0

Applying this construction to the category S = AE of Section 5, we

get a braided tensor category AE[[h]]. Choose the object 1 in it. Then by
Corollary XIII.3.S there exists a unique strict braided tensor functor Z
from the braid category E to the category AE[[h]] which sends 1 to 1 and,
therefore, is the identity on objects. Restricting to the endomorphisms of
n in E, namely to the braid group B n , we get a group morphism

Z : Bn --+ Bn x (1 + L A~r(1r,)hm)
7n:;,1

which by Lemma XV.4.1 is defined on the generators of the braid group by

(6.3)
and
Z() -1 'dQ9C n - i - 1)) ht"i+l /2 ( 'dQ9cn-i-1))
O"i = ( a V3 (i-l),V,V 01 v O"i e aV3(,-I),V,V 01 v

(6.4)
when 2 ::; i ::; n - 1. The associativity isomorphisms aV3(i-l),v,v have
to be computed from the Drinfeld series <P using (6.1) and (4.3-4.4). The
composition of Z with the projection onto Bn is the surjection sending each
braid to its permutation. In the next section, we shall extend the map Z
to all tangles.
Let 9 be a semisimple Lie algebra and t E g0g be the invariant symmetric
2-tensor given by (XVII. 1.6). Consider the infinitesimal symmetric category
U(g)-Mod f . We can reformulate precisely Drinfeld's Theorem XIX.4.3 and
Corollary XIX.4.4 as follows.
Corollary XX.6.2. In case <P = <P KZ , there is a braided tensor equiva-
lence between the braided tensor category U h (g)-Mod fT of XVII.3 and the
braided tensor category (U(g)-Modf)[[hl].

XX.7 Construction of Kontsevich's Universal

Invariant
We first state a complement to Theorem 6.1 in the case when the infinites-
imal symmetric category S has a left duality V f-+ V* with structure maps
500 Chapter XX. Postlude. A Universal Knot Invariant

b~ : I -+ V c>9 V* and d~ : V* c>9 V -+ I. In Section 6 we constructed a

non-strict braided tensor category S[[h]]. Let S[[hW tr be the strict braided
tensor category associated to S[[h]] by the procedure of XI.5. We keep the
notations of Section 6.

Theorem XX.7.1. Under these hypotheses, the strict braided tensor cat-
egory S[[hW tr is a ribbon category with twist Bv given by

(7.1)

and with left duality defined as follows: for any object V the dual object V*
is the same as in the category S; the structure maps bv and d v are defined
by
bv = b~ and d v = d~ 0 (AV: c>9 id v ) (7.2)
where AV* is the automorphism of V* defined by

Ribbon categories were defined in XIV.3 and the endomorphisms C v

by (4.6).
PROOF. The axioms (XIV.2.1) for the duality follow from a computation
and the axioms (XIV.3.1-3.2) for the twist essentially from Relation (4.5).
o
The importance of this theorem lies in the fact explained in XIV.5.1
that, by colouring links with any object of S, the ribbon category S[[hW tr
provides us with a framed link invariant with values in the endomorphism
ring Ends (I) [[h]] of the unit object I in S[[h]]"tr. This fact will now be used
to construct Kontsevich's universal invariant. From now on, we assume that
the Drinfeld series we want Theorems 6.1 and 7.1 to work with is Drinfeld's
associator <P KZ '
We first define an infinitesimal symmetric category A with left duality. It
is built in the same way as the category AB of Section 5, except that braids
are now replaced by framed tangles and chords are no longer assumed to
be horizontal. More precisely, the objects of the category A are the objects
of the tangle category I, namely finite sequences of + and -, including the
empty sequence 0. A morphism in A is an element of the complex vector
space A(T) for some framed tangle T (as defined in Section 3). Its source
[resp. its target] is the sequence s(T) [resp. the sequence b(T)] defined in
X.5. The composition of morphisms is given by the map (3.5). The identity
of a sequence S is the chordless diagram on the tangle ids.
We define a strictly associative tensor product on A as on AB. Its unit is
the empty sequence: I = 0. Remember that the monoid of endomorphisms
of 0 in the framed tangle category is the set of all isotopy classes of framed
links in R2 x ]0,1[. Here, the monoid of endomorphisms of 0 in the category
A is a complex associative algebra since the sets of morphisms are complex
 XX.7 Construction of Kontsevich's Universal Invariant 501

vector spaces. This algebra is bigraded by the number of chords and the
number of connected components of the link. We have

EndA(I) 2;! EB Am(00 n ) (7.4)

m,n;:::O
where 00 n denotes the disjoint union of n circles.
The symmetry and the infinitesimal braiding of the category AB de-
fine a symmetry and an infinitesimal braiding on A. The latter also has a
left duality induced by the left duality of the category R of ribbons (see
XIV.5.1).
The category A satisfying the hypotheses of Theorem 7.1, we get a ribbon
category A[[h]]'tr. Take the object (+) in it. By Theorem XIV.5.1 there
exists a unique strict braided tensor functor
Z :R -t A[[h]]str

preserving the duality and the twist, sending the object (+) of the category
R of ribbons to the object (+) of A[[h]]str. Consequently, the functor Z is
the identity on objects. The restriction of Z to braids is the morphism
defined by (6.3-6.4).
Let K be a framed link. It can be viewed as an endomorphism of the unit
object in the category R. Its image Z(K) is an isotopy invariant living in

End A[[hll",,(0) = EB A(on).

n>O
When K is a framed knot, the invariant Z(K) lies in A = I1m>O Am'
This is the universal invariant we are after. Indeed, by the definition of the
braiding (6.2) in A[[h]], the invariant Z(K) = L:m>O Zm(K)h m satisfies
the hypotheses of Proposition 1.2. It results that Z is an invariant of finite
type. In order to check Relation (2.10), we have to extend the invariant
Z to singular knots. This is done using (1.1). At each double point of a
singular knot, we have for Z a local contribution of the form cs,s' - <5,~,.
By (6.2) this looks like
ehts ,s,/2 - e- hts ,s,/2 = hts,s' + terms of degree > 1. (7.5)

Relation (7.5) and an induction on the number of double points imply

Relation (2.10).
Remarks 7.2. (a) Le and Murakami [LM93c] showed that Z(K) coincided
with the invariant originally constructed by Kontsevich with multiple in-
tegrals. The reader is advised to read [LM93c] where Z(K) is defined in a
slightly different way using the concept of quasi-tangles.
(b) The appearance of <I>KZ in the definition of Z (K) makes it difficult
to compute for any framed knot. Nevertheless, the first terms of the formal
series Z(K) may be determined using Corollary XIX.6.5.
502 Chapter XX. Postlude. A Universal Knot Invariant

XX.8 Recovering Quantum Group Invariants

The aim of this section is to show how one recovers the quantum group
invariants of XVII.3 from Kontsevich's universal invariant Z. We first state
a universal property for the category A of Section 7, parallel to Proposi-
tion 5.2.
Proposition XX.B.l. Let S be an infinitesimal symmetric category with
left duality, with symmetry (0' v. W ) v, wand infinitesimal braiding (tv. w ) v, w
For any object of S there exists a unique functor Fv from the category A
to the category S such that

Fv(8 8') = Fv(8) 0 Fv(8'), Fv(0) = I, Fv( +) = V, Fv( -) = V*,

(8.1)
Fv(O's,s') = O'FV(S),Fv(S') , Fv(b s ) = bFv(S)' Fv(d s ) = dFv(s) (8.2)
and
Fv(ts,s') = tFv(S),Fv(S') (8.3)
for all objects 8 and 8' of S.

PROOF. One proceeds as for Proposition 5.2. The main difference lies in the
existence of general chord diagrams in A. In order to show that Relation
(8.3) determines Fv on any chord diagram, we observe that any chord may
be arranged so as to be horizontal after possibly adding some maxima and
minima to the diagram. D

We wish to illustrate Proposition 8.1 in the case when S = H-Mod f

where H = (H, 6., c) is a complex Hopf algebra along with an element
t = Lj Xi 0 Yi in Prim(H)0Prim(H) such that t21 = t and [t,6.(a)] =
o for all a E H. By Proposition 4.2 we know that S is an infinitesimal
symmetric category with left duality, the symmetry being the flip and the
infinitesimal braiding given by (4.7). Fix a finite-dimensional left H-module
V. By Proposition 8.1, there exists a well-defined functor Fv : A --> H-
Mod f such that F v (+) = V. Consequently, if D is a chord diagram on
1= id+, then Fv(D) is an H-linear endomorphism of V. We now determine
this endomorphism.
Let D be a chord diagram on 1 with m > 0 chords. Define an element CD
of H by the following combinatorial rule. Running down along the strand
L write x jk whenever you come across the k-th upper endpoint of a chord
and write Yjk when you meet its lower endpoint. In this way one gets a
word W D' Suppose the word is

WD = Xj,XhxhYhXj4Yj,Yj4Xj5YhYj5
(here m = 5). Then the element CD is by definition

CD = (_l)m "'"
~
XJ, x J2 XJ3 y.J2 x J4 y.J' y.J4 X.15 y.J3 y..15' (8.4)
]1,,]5
 XX.8 Recovering Quantum Group Invariants 503

Proposition XX.8.2. For any chord diagram D on L the element CD is

central in H and depends only on the class of D in A(l). Moreover, the
endomorphism Fv(D) is the action of the central element CD of H on V.
PROOF. We first deform D into a chord diagram whose chords are hori-
zontal. We claim that (8.1-8.3) imply that Fv(D) is equal to the action
of CD on V. Now F v is defined on the equivalence classes of chord dia-
grams. Therefore CD depends only on the equivalence class of D. Finally,
the endomorphism Fv(D) being H-linear, CD is central.
Let us prove the claim in the special case when D is the unique chord
diagram on 1 with two intersecting chords. This diagram can be expressed
as
D = (id+ Q9 d+)(t+,_ Q9 id+)(id+ Q9 C,+)(b+ Q9 id+) (8.5)
in the category A. Its image under the functor F v is the endomorphism

Let v be an element of V and {vJi be a basis of V. We denote the dual

basis by {viL. We have

2: XjVi < YjXk Vi , Yk v >

i,j,k

2: XjVi < vi, S(YjXk)YkV >

i,j,k

i,j,k

(2: XjXkYjYk)V
j,k
CDv.

The third equality follows from S(YjXk) = S(Xk)S(Yj) = (-1)2 xkyj , which
holds because x k and Yj are primitive elements of H. 0

Proposition 8.2 provides an interesting way of constructing central el-

ements of H. For instance, if H is the enveloping algebra of a complex
semisimple Lie algebra fJ with its canonical 2-tensor t, then we recover the
Casimir operator (XVIL1.5) C = CD from the chord diagram D with one
chord. It would be interesting to characterize the subspace of the centre of
U(fJ) spanned by all elements CD.
We are now ready to indicate how one recovers the quantum group in-
variant Qg,v from Kontsevich's universal invariant Z. Recall that Qg,v is
defined for a semisimple Lie algebra fJ and a finite-dimensional simple fJ-
module V. To the data (fJ, V), we associate a linear map Wg,V on the space
of all chord diagrams as follows. Let D be a chord diagram on the circle. To
it corresponds a uniquely defined chord diagram, still denoted D, on the
504 Chapter XX. Postlude. A Universal Knot Invariant

line 1. By Proposition 8.1 we know how to build a central element CD of

U(g). Since V is simple, CD acts as a scalar J1D on V. We define wg,v(D)
as
(8.7)
The relationship between the quantum group invariant Q g,v and Kont-
sevich's universal knot invariant Z is given by the following statement.

Theorem XX.8.3. Under the previous hypotheses, for all framed knots K
we have
Qg,v(K) = dimq(V) L
Wg,v(Zm(K))hm (8.8)
m:;,D

where dimq(V) is the quantum dimension defined in XVII.3.

PROOF. Applying Proposition 8.1 to S = U(g)-Mod l and to the given

simple module V, we get a functor Fv from A to U(g)-Mod I with Fv (+) =
V. By Theorem 7.1, Fv extends to a ribbon functor Fv from A[[h]]'tT to
(U(g)-Mod l ) [[hW tT such that Fv (+) = V. By a ribbon functor, we mean a
sJ;rict braided tensor functor preserving left duality and twist. Com2osing
Fv with the functor Z of Section 7, we get the ribbon functor Fv 0 Z
from the category R of framed tangles to (U(g)-Mod I) [[hWtT such that
(Fv 0 Z)(+) = V. Now by Corollary 6.2, the latter category is equivalent
to the category Uh (g)- Mod Ir of XVII.3 via a strict braided tensor functor
E sending the simple g-module V to the topologically free Uh(g)-module
V. Actually, the equivalence E preserves also the duality and the twist (see
[Dri89b]). Therefore, EoFvoZ is a ribbon functor from R to Uh(g)-Mod lr ,
sending (+) to V. By the uniqueness statement in Theorem XIV. 5.1, we
have
EoFv oZ = Fv (8.9)
where Fv is the ribbon functor introduced in XVII.3. Let K be a framed
knot. By construction of the invariant Q g,v, we get

(8.10)

Let us evaluate Eo Fv on a chord diagram D on the circle. By Proposition

8.2, by (8.2-8.3), (8.7) and by XIV.4, we have

(E 0 Fv)(D) E(trq( C DIV)) = J1D E( dimq(V))

J1D dimq(E(V)) = dimq(V) wg,v(D) (8.11)

where the quantum trace and dimension are taken first in the ribbon cat-
egory (U(g)-Modj)[[hW tr , then in the equivalent category Uh(g)-Mod lr .
Combining the last set of equalities with (8.10) yields Theorem 8.3. D
 XX.lO Notes 505

XX.9 Exercises
1. Find all primitive elements of degree :'S 4 in the Hopf algebra A of
Section 3.

2. Let O(N) be the framed trivial knot whose framing twists the knot by
27rN. Compute its Kontsevich invariant Z(O(N)) modulo h4. (Hint:
use Corollary XIX.6.5.)

3. Compute the Kontsevich invariant of the closure of the braid ai N + 1

of B2 modulo h4.

XX.I0 Notes
The concept of a knot invariant of finite degree (also called "Vassiliev in-
variant" in the literature) was introduced by Gusarov [Gus91] and Vas-
siliev [Vas90] [Vas92] around 1989-90. Vassiliev's approch was based on
the theory of singularities. Soon after, a number of mathematicians made
substantial contributions to this new theory such as D. Bar-Natan, J. Bir-
man, P. Cartier, M. Kontsevich, Le T.Q.T., X.S. Lin, J. Murakami, S. Pi-
unikhin, T. Stanford (see [BN92] [Bir93] [BL93] [Car93] [Kon93] [LM93b]
[LM93a] [LM93c] [Lin91] [Piu92] [Piu93] [Sta92] [Sta93]). One will find a
review of their results in [Vog93]. A major step forward was undertaken by
Kontsevich who constructed the universal knot invariant Z(K) and proved
Theorem 2.3. Kontsevich's definition of Z(K) used complicated multiple
integrals. It was proved by Cartier [Car93], Le-Murakami [LM93c] and Pi-
unikhin [Piu93] that it could be defined in a simpler way using tangle
diagrams. Theorems 6.1 and 7.1 are due to Cartier [Car93].
The contents of Sections 5 and 8 seem to be new. For a generalization,
see [KT94].
References

[Abe80] E. Abe. Hopf Algebras. Cambridge University Press, Cam-

bridge, 1980.

[AC92] D. Altschuler and A. Coste. Quasi-quantum groups, knots,

three-manifolds, and topological field theory. Comm. Math.
Phys. 150 (1992), 83-107.

[Ada29] C.R. Adams. On the linear ordinary q-difference equation.

Ann. of Math. (2) 30 (1929), 195-205.

[Ada56] J.F. Adams. On the cobar construction. Pmc. Nat. Acad. Sci.
USA 42 (1956), 409-412.

[Ale23] J. W. Alexander. A lemma on systems of knotted curves. Pmc.

Nat. Acad. Sci. U.S.A. 9 (1923), 93-95.

[Ale28] J.W. Alexander. Topological invariants of knots and links.

Trans. Amer. Math. Soc. 30 (1928), 275-306.

[And76] G.E. Andrews. The Theory of Partitions, volume 2 of Ency-

clopedia of Mathematics and its Applications. Addison Wesley
Publ. Co, Reading, MA, 1976.

[Aom78] K. Aomoto. Fonctions hyperlogarithmes et groupes de mono-

dromie unipotents. J. Fac. Sci. Univ. Tokyo Sect. fA Math.
25 (1978), 149-156.
 References 507

[Art25] E. Artin. Theorie der Zopfe. Abh. Math. Sem. Univ. Hamburg
4 (1925), 47-72.

[Art47] E. Artin. Theory of braids. Ann. of Math. (2) 48 (1947),

101-126.

[AST91] M. Artin, W. Schelters, and J. Tate. Quantum deformations

of GL n . Comm. Pure Appl. Math. 44 (1991), 879-895.

[Bax72] R.J. Baxter. Partition function for the eight-vertex lattice

model. Ann. Physics 70 (1972), 193-228.

[Bax82] R.J. Baxter. Exactly Solved Models in Statistical Mechanics.

Academic Press, London, 1982.

[Ben63] J. Benabou. Categories avec multiplication. C. R. Acad. Sci.

Paris Ser. I Math. 256 (1963), 1887-1890.

[Ben64] J. Benabou. Algebre elementaire dans les categories avec mul-

tiplication. C. R. Acad. Sci. Paris Ser. I Math. 258 (1964),
771-774.

[Bir74] J. Birman. Braids, Links and Mapping Class Groups, vol-

ume 82 of Ann. of Math. Stud. Princeton University Press,
Princeton, NJ, 1974.

[Bir93] J. Birman. New points of view in knot theory. Bull. Amer.

Math. Soc. 28 (1993), 253-287.

[BL93] J. Birman and X.S. Lin. Knot polynomials and Vassiliev's

invariants. Invent. Math. 111 (1993), 225-270.

[BN92] D. Bar-Natan. On the Vassiliev knot invariants. Harvard Uni-

versity, 1992. Topology, to appear.

[Bou60] N. Bourbaki. Groupes et Algebres de Lie. Hermann, Paris,

1960.

[Bou61] N. Bourbaki. Algebre Commutative. Hermann, Paris, 1961.

[Bou70] N. Bourbaki. Algebre, Chap. 1-3. Hermann, Paris, 1970.

[Bur36] W. Burau. Uber Zopfgruppen und gleichsinnig verdrillte Ver-

kettungen. Abh. Math. Sem. Univ. Hamburg 11 (1936), 179-
186.

[BZ85] G. Burde and H. Zieschang. Knots. W. de Gruyter, Berlin,

1985.
508 References

[Car57] P. Cartier. Cohomologie des coalgebres (Expose 5). In

Seminaire Sophus Lie, J 955~56. Faculte des Sciences de Paris,
Paris, 1957.
[Car93] P. Cartier. Construction combinatoire des invariants de
Vassiliev-Kontsevich des nCBuds. C. R. Acad. Sci. Paris Ser.
I Math. 316 (1993), 1205~121O.

[CE48] C. Chevalley and S. Eilenberg. Cohomology theory of Lie

groups and Lie algebras. Trans. Amer. Math. Soc. 63 (1948),
85-124.

[CE56] H. Cartan and S. Eilenberg. Homological Algebra. Princeton

University Press, Princeton, NJ, 1956.

[Che61] K.T. Chen. Formal differential equations. Ann. of Math. (2)

73 (1961), 110-l33.

[Che73] K.T. Chen. Iterated integrals of differential forms and loop

space homology. Ann. of Math. (2) 97 (1973), 217-246.

[Che75] K.T. Chen. Connections, holonomy and path space homology.

In Differential Geometry, Part I, volume 27 of Proc. Sympos.
Pure Math., pages 39-52. Amer. Math. Soc., Providence, RI,
1975.

[Che77a] K.T. Chen. Extensions of Coo function algebra by integrals

and Malcev completion of ]fl' Adv. Math. 23 (1977), 181-210.

[Che77b] K.T. Chen. Iterated path integrals. Bull. Amer. Math. Soc.
83 (1977),831-879.

[Cig79] J. Cigler. Operatormethoden fur q-Identitaten. Monatsh.

Math. 88 (1979), 87-105. MR 81h:05009.

[Coh7l] P.M. Cohn. Free Rings and their Relations, volume 2 of Lon-
don Math. Soc. Monographs. Academic Press, London, New
York, 1971.

[CK90] C. de Concini and V.G. Kac. Representations of quantum

groups at roots of l. In Operator Algebras, Unitary Repre-
sentations, Enveloping Algebras, and Invariant Theory (Paris,
1989), volume 92 of Progr. Math., pages 471-506. Birkhiiuser,
Boston, 1990. MR 92g: 17012.

[CKP92] C. de Concini, V.G. Kac, and C. Procesi. Quantum coadjoint

orbit. J. Amer. Math. Soc. 5 (1992), 151-189. MR 93f:17020.
 References 509

[Con70] J.H. Conway. An enumeration of knots and links, and some

of their algebraic properties. In Computational Fmblems in
Abstract Algebra, pages 329-358. Pergamon Press, New York,
1970.

[Cur52] C.W. Curtis. A note on noncommutative polynomials. Fmc.

Amer. Math. Soc. 3 (1952), 965-969.

[DehI4] M. Dehn. Die beiden Kleeblattschlingen. Math. Ann. 75

(1914),402-413.

[De190] P. Deligne. Categories tannakiennes. In The Gmthend'ieck

Festschrift, vol. II, volume 87 of Fmgr. Math., pages 111-195.
Birkhiiuser, Boston, 1990. MR 92d:14002.

[DG70] M. Demazure and P. GabrieL G-roupes Algebriques. North-

Holland PubL Co, Amsterdam, 1970.

[Dix74] J. Dixmier. Algebres Enveloppantes. Bordas (Gauthier-

Villars), Paris, 1974. English transL: Enveloping Algebras,
North-Holland PubL Co, Amsterdam, 1977.

[DJMM91] E. Date, M. Jimbo, K. Miki, and T. Miwa. Cyclic represen-

tations of Uls((n + 1, C)) at qN = 1. Fubl. Res. Inst. Math.
Sci. 27 (1991), 347-366. MR 92i:17018.

[DM82] P. Deligne and J.S. Milne. Tannakian categories. In Hodge

cycles, Motives and Shimura Varieties, volume 900 of Lecture
Notes in Math., pages 101-228. Springer-Verlag, Berlin, 1982.
MR 84m:14046.

[DPR90] R. Dijkgraaf, V. Pasquier, and P. Roche. QuasiHopf algebras,

group cohomology and orbifold models. Nuclear Fhys. B Fmc.
Suppl. 18B (1990), 60-72. MR 92m:81238.

[Dri85] V.G. Drinfeld. Hopf algebras and the quantum Yang-Baxter

equation. Dokl. Akad. Nauk SSSR 283:5 (1985), 1060-1064.
English transL: Soviet Math. Dokl. 32 (1985), 254-258. MR
87h:58080.

[Dri87] V.G. Drinfeld. Quantum groups. In Fmc. Int. Congo Math.

(Berkeley, 1986), pages 798-820. Amer. Math. Soc., Provi-
dence, RI, 1987. MR 89f:17017.

[Dri89a] V.G. Drinfeld. On almost cocommutative Hopf algebras. Al-

gebra i Analiz 1:2 (1989), 30-46. English transL: Leningrad
Math J. 1 (1990),321-342. MR 91b:16046.
510 References

[Dri89b] V.G. Drinfeld. Quasi-Hopf algebras. Algebra i Analiz 1:6

(1989),114-148. English transl.: Leningrad Math. J. 1 (1990),
1419-1457. MR 91b:17016.

[Dri89c] V.G. Drinfeld. Quasi-Hopf algebras and Knizhnik-

Zamolodchikov equations. In Problems of modern quantum
field theory (Alushta, 1989), Res. Rep. Phys., pages 1-l3.
Springer-Verlag, Berlin, 1989. MR 92j:170l1.

[Dri90] V.G. Drinfeld. On quasitriangular quasi-Hopf algebras and a

group closely connected with Gal(Q/Q). Algebra i Analiz 2:4
(1990),149-181. English transl.: Leningrad Math. J. 2 (1991),
829-860. MR 92f:16047.

[EM 53] S. Eilenberg and S. Mac Lane. On HJTr, n), 1. Ann. of Math.
(2) 58 (1953), 55-106.
[Enr92] B. Enriquez. Rational forms for twistings of enveloping al-
gebras of simple Lie algebras. Lett. Math. Phys. 25 (1992),
111-120.

[Fad84] L.D. Faddeev. Integrable models in (1 + I)-dimensional quan-

tum field theory. Lectures at Les Houches 1982, Elsevier Sci-
ence Publishers B.V., Amsterdam, New York, 1984.

[FY89] P.J. Freyd and D.N. Yetter. Braided compact closed categories
with applications to low-dimensional topology. Adv. Math. 77
(1989), 156-182. MR 91c:57019.

[FY92] P. Freyd and D. Yetter. Coherence theorems via knot theory.

1. Pure Appl. Algebra 78 (1992), 49-76. MR 93d:180l3.
[FYH+85] P. Freyd, D. Yetter, J. Hoste, W.B.R. Lickorish, K.C. Mil-
lett, and A. Ocneanu. A new polynomial invariant of knots
and links. Bull. Amer. Math. Soc. 12 (1985), 239-246. MR
86e:57007.

[Ger64] M. Gerstenhaber. On the deformation of rings and algebras.

Ann. of Math. (2) 79 (1964), 59-103.

[Go180] V.A. Golubeva. On the recovery of a Pfaffian system of

Fuchsian type from the generators of the monodromy group.
Izv. Akad. Nauk SSSR 44:5 (1980), 979-998. English transl.:
Math. USSR-Izv. 17 (1981), 227-241.

[GR90] G. Gasper and M. Rahman. Basic Hypergeometric Series, vol-

ume 35 of Encyclopedia of Mathematics and its Applications.
Cambridge University Press, Cambridge, 1990. MR 91d:33034.
 References 511

[Gus91] M.N. Gusarov. A new form of the Conway-Jones polynomial

of oriented links. Zap. Nauchn. Sem. Leningrad. Otdel. Mat.
Inst. Steklov. (LOMI) 193 (1991). Geom. i TopoL, 1,4-9,161
(in Russian). MR 93b:57007.

[Hai86] R.M. Hain. On a generalization of Hilbert's 21st problem.

Ann. Sci. Ecole Norm. Sup. (4) 19 (1986), 609-627.

[HKW86] P. de la Harpe, M. Kervaire, and C. Weber. On the Jones

polynomial. Enseign. Math. (2) 32 (1986), 271-335. MR
88f:57004.

[Hay92] T. Hayashi. Quantum groups and quantum determinants. J.

Algebra 152 (1992), 146-165.

[Hei46] E. Heine. Uber die Reihe

(q"-l)(qil_l) (q"_1)(q<>+1_1)(qil_l)(qil+1_l) 2
1 + (q-l)(q"l-l) X + (q-l)(q2-1)(q"l-1)(q"l+'-1) X +.... J.
Reine Angew. Math. 32 (1846), 210-212.

[Hoc81] G.P. Hochschild. Basic Theory of Algebraic Groups and Lie

Algebras, volume 75 of Graduate Texts in Math. Springer-
Verlag, New York, 1981. MR 82i:20002.

[Hop41] H. Hopf. Uber die Topologie der Gruppen-Mannigfaltigkeiten

und ihren Verallgemeinerungen. Ann. of Math. (2) 42 (1941),
22-52.

[Hos86] J. Hoste. A polynomial invariant of knots and links. Pacific

J. Math. 124 (1986), 295-320. MR 88d:57004.

[Hug51] V. Hugo. Pierres (Vers et Prose). Editions du Milieu du

Monde, Geneve, 1951.

[Hum72] J.E. Humphreys. Introduction to Lie Algebras and their Rep-

resentation Theory, volume 9 of Graduate Texts in Math.
Springer Verlag, New York, 1972.

[Jac10] F.H. Jackson. q-difference equations. Amer. J. Math. 32

(1910), 305-314.

[Jac79] N. Jacobson. Lie Algebras. Dover Publ., New York, 1979.

[Jim85] M. Jimbo. A q-difference analogue of U(g) and the Yang-

Baxter equation. Lett. Math. Phys. 10 (1985), 63-69. MR
86k:17008.

[Jim86a] M. Jimbo. A q-analogue of U(gl(N + 1)), Hecke algebra and

the Yang-Baxter equation. Lett. Math. Phys. 11 (1986), 247-
252. MR 87k:17011.
512 References

[Jim86b] M. Jimbo. Quantum R-matrix for the generalised Toda

system. Comm. Math. Phys. 102 (1986), 537-547. MR
87h:58086.

[JL92] A. Joseph and G. Letzter. Local finiteness of the adjoint action

for quantized enveloping algebras. J. Algebra 153 (1992), 289-
318.
[Jon85] V.F.R. Jones. A polynomial invariant for links via von Neu-
mann algebras. Bull. Amer. Math. Soc. 12 (1985), 103-111.
MR 86e:57006.

[Jon87] V.F.R. Jones. Recke algebra representations of braid groups

and link polynomials. Ann. of Math. (2) 126 (1987),335-388.
MR 89c:46092.

[JS91a] A. Joyal and R. Street. The geometry of tensor calculus, I.

Adv. Math. 88 (1991), 55-112. MR 92d:18011.

[JS91b] A. Joyal and R. Street. An introduction to Tannakian dual-

ity. In Category Theory (Como, 1990), volume 1488 of Lecture
Notes in Math., pages 413-492. Springer-Verlag, Berlin, 1991.
MR 93f:18015.

[JS91c] A. Joyal and R. Street. Tortile Yang-Baxter operators in ten-

sor categories. J. Pure Appl. Algebra 71 (1991), 43-51. MR
92e:18006.

[JS93] A. Joyal and R. Street. Braided tensor categories. Adv. Math.

102 (1993), 20-78. Formerly Macquarie Math. Reports No.
850067 (Dec. 1985) and No. 860081 (Nov. 86).

[Kap93] M.M. Kapranov. The permutoassociahedron, Mac Lane's co-

herence theorem and asymptotic zones for the KZ equation.
J. Pure Appl. Algebra 85 (1993), 119-142.
[Kas92] C. Kassel. Cours sur les groupes quantiques, Ie partie. Publ.
I.R.M.A., Strasbourg, 1992. MR 93k:17030.

[Kau87a] L.R. Kauffman. On Knots, volume 115 of Ann. of Math.

Stud. Princeton University Press, Princeton, NJ, 1987. MR
89c:57005.

[Kau87b] L.R. Kauffman. State models and the Jones polynomial.

Topology 26 (1987), 395-407. MR 88f:57006.

[Kau91] L.R. Kauffman. Knots and Physics, volume 1 of Series on

Knots and Everything. World Scientific Publ. Co, River Edge,
NJ, 1991. MR 93b:57010.
 References 513

[KeI64] G.M. Kelly. On Mac Lane's conditions for coherence of natural

associativities, commutativities, etc. J. Algebra 1 (1964),397-
402.

[KK89] H.T. Koelink and T.H. Koornwinder. The Clebsch-Gordan co-

efficients for the quantum group S"U(2) and q-Hahn polyno-
mials. Proc. Kon. Nederl. Akad. Wetensch. A92, Indag. Math.
(N.S.) 51 (1989), 443-456. MR 91c:17009.
[KL80] G.M. Kelly and M.L. Laplaza. Coherence for compact closed
categories. J. Pure Appl. Algebra 19 (1980), 193-213. MR
81m:18008.

[KM91] R. Kirby and P. Melvin. The 3-manifold invariants of Wit-

ten and Reshetikhin-Turaev for 81(2, C). Invent. Math. 105
(1991),473-545. MR 92e:57011.

[KN63] S. Kobayashi and K. Nomizu. Foundations of Differential Ge-

ometry I,ll. Wiley (Interscience), New York, 1963.

[Koh85] T. Kohno. Serie de Poincare-Koszul associee aux groupes de

tresses pures. Invent. Math. 82 (1985), 57-75. MR 87c:32015a.

[Koh87] T. Kohno. Monodromy representations of braid groups and

Yang-Baxter equations. Ann. Inst. Fourier (Grenoble) 37
(1987),139-160. MR 89h:17030.

[Koh88] T. Kohno. Hecke algebra representations of braid groups and

classical Yang-Baxter equations. In Conformal Field Theory
and Solvable Lattice Models (Kyoto, 1986), volume 16 of Adv.
Stud. Pure Math., pages 255-269. Academic Press, Boston,
MA, 1988. MR 90a:17016.

[Kon93] M. Kontsevich. Vassiliev's knot invariants. In Gelfand Semi-

nar, Part 2, volume 16 of Adv. Soviet Math., pages 137-150.
Amer. Math. Soc., Providence, RI, 1993.

[Koo90] T.H. Koornwinder. Orthogonal polynomials in connection

with quantum groups. In Orthogonal Polynomials: Theory and
Practice (Columbus, Ohio, 1989), volume 294 of NATO Adv.
Sci. Inst. Ser. C: Math. Phys. Sci., pages 257-292. Kluwer
Acad. Publ., Dordrecht, 1990.

[KR81] P.P. Kulish and N. Yu. Reshetikhin. Quantum linear problem

for the sine-Gordon equation and higher representations. Zap.
Nauchn. Sem. Leningrad. Otdel. Mat. Inst. Steklov. (LOMI)
101 (1981), 101-110. English trans!.: J. Soviet Math. 23
(1983), 2435-2441.
514 References

[KR89] A.N. Kirillov and N. Yu. Reshetikhin. Representations of

the algebra Uq (sl(2)), q-orthogonal polynomials and invari-
ant of links. In Infinite-dimensional Lie algebras and groups
(Luminy-Marseille, 1988), volume 7 of Adv. Ser. Math. Phys.,
pages 285~339. World Scientific, Teaneck, NJ, 1989. MR
90m:17022.

[KR90] A.N. Kirillov and N. Yu. Reshetikhin. q-Weyl group and a

multiplicative formula for universal R-matrices. Comm. Math.
Phys. 134 (1990), 421~431. MR 92c:17023.

[KS80] P.P. Kulish and E.K. Sklyanin. Solutions of the Yang-Baxter

equation. Zap. Nauchn. Sem. Leningrad. Otdel. Mat. Inst.
Steklov. (LOMI) 95 (1980), 129~160, 161. English transl.: J.
Soviet Math. 19 (1982), 1596~1620. MR 82i:82018.

[KT92] C. Kassel and V. Turaev. Double construction for monoidal

categories. Preprint, I.R.M.A., Strasbourg, 1992, Acta Math.

[KT94] C. Kassel and V. Turaev. Chord diagram invariants for ribbon

graphs. Preprint, I.R.M.A., Strasbourg, 1994.

[KZ84] V.G. Knizhnik and A.B. Zamolodchikov. Current algebra and

Wess-Zumino models in two dimensions. Nuclear Phys. B 247
(1984), 83~103. MR 87h:81129.

[Lan65] S. Lang. Algebra. Addison Wesley, Reading, Mass., 1965.

[Lap53] J.A. Lappo-Danilevsky. Memoires sur la TMorie des

Systemes Differentiels Lineaires. Chelsea Publ. Co, New York,
NY, 1953.

[Lin91] x.S. Lin. Vertex models, quantum groups and Vassiliev's knot
invariants. Columbia University, Preprint, 1991.

[LM87] W.B.R. Lickorish and K.C. Millett. A polynomial invariant of

oriented links. Topology 26 (1987), 107~141. MR 88b:57012.

[LM93a] T.Q.T. Le and J. Murakami. Kontsevich's integral for Kauff-

man polynomial. Preprint, M.P.I. Bonn, 1993.

[LM93b] T.Q.T. Le and J. Murakami. Kontsevich's integral for the

Homfiy polynomial and relations between values of multiple
zeta functions. Preprint, M.P.I. Bonn, 1993.

[LM93c] T.Q.T. Le and J. Murakami. Representations of the category

of tangles by Kontsevich's iterated integral. Preprint, M.P.I.
Bonn, 1993.
 References 515

[LS90] S.Z. Levendorsky and Ya. S. Soibelman. Some applications of

the quantum Weyl groups. J. Geom. Phys. 7 (1990), 241-254.
MR 92g:17016.

[LT91] R.G. Larson and J. Towber. Two dual classes of bialgebras

related to the concepts of "quantum group" and "quantum
Lie algebra". Comm. Algebra 19 (1991), 3295-3345. MR
93b:16070.

[Lus88] G. Lusztig. Quantum deformations of certain simple modules

over enveloping algebras. Adv. Math. 70 (1988), 237-249. MR
89k:17029.

[Lus89] G. Lusztig. Modular representations and quantum groups.

Contemp. Math. 82 (1989), 59-77. MR 90a:16008.

[Lus90a] G. Lusztig. Finite-dimensional Hopf algebras arising from

quantized enveloping algebras. J. Amer. Math. Soc. 3 (1990),
257-296. MR 91e:17009.

[Lus90b] G. Lusztig. Quantum groups at roots of 1. Geom. Dedicata

35 (1990), 89-113. MR 91j:17018.

[Lus93] G. Lusztig. Introduction to Quantum Groups, volume 110 of

Progr. Math. Birkhiiuser, Boston, 1993.
[Mac63] S. Mac Lane. Natural associativity and commutativity. Rice
Univ. Studies 49 (1963), 28-46.
[Mac71] S. Mac Lane. Categories for the Working Mathematician, vol-
ume 5 of Graduate Texts in Math. Springer-Verlag, New York,
1971.
[Maj90a] S. Majid. Physics for algebraists: non-commutative and non-
cocommutative Hopf algebras by a bicrossproduct construc-
tion. J. Algebra 130 (1990), 17-64. MR 91j:16050.

[Maj90b] S. Majid. Quasi-triangular Hopf algebras and Yang-Baxter

equations. Intern. J. Modern Phys. A 5 (1990), 1-91. MR
90k:16008.
[Maj91a] S. Majid. Doubles of quasitriangular Hopf algebras. Comm.
Algebra 19 (1991), 3061-3073. MR 92k:16052.

[Maj91b] S. Majid. Representations, duals and quantum doubles of

monoidal categories. Rend. Circ. Math. Palermo (2) Suppl.
26 (1991), 197-206. MR 93c:18008.
516 References

[Ma190] G. Maltsiniotis. Groupes quantiques et structures

differentielles. C. R. Acad. Sci. Paris Ser. I Math. 311 (1990),
831-834. MR 91m:58010.

[Mal93] G. Maltsiniotis. Le lang age des espaces et des groupes quan-

tiques. Comm. Math. Phys. 151 (1993), 275-302.
[Man87] Yu.1. Manin. Some remarks on Koszul algebras and quantum
groups. Ann. lnst. Fourier (Grenoble) 37 (1987), 191-205.
MR 8ge:16022.

[Man88] Yu.1. Manin. Quantum Groups and Noncommutative Geo-

metry. Universite de Montreal, Centre de Recherches
Mathematiques, Montreal, 1988. MR 91e:17001.

[Man89] Yu.1. Manin. Multiparametric quantum definition of the gen-

erallinear supergroup. Comm. Math. Phys. 123 (1989), 163-
175. MR 90h: 17018.

[Mar36] A. Markoff. -ober die freie Aquivalenz der geschlossenen Zopfe.

Recueil Mathematique, Mat. Sb. 1 (43) (1936), 73-78.
[Mat70] H. Matsumura. Commutative Algebra. W.A. Benjamin, New
York, 1970.

[MM65] J. Milnor and J. Moore. On the structure of Hopf algebras.

Ann. of Math. (2) 81 (1965),211-264.
[MMN+90] T. Masuda, K. Mimachi, Y. Nagakami, M. Noumi, Y. Saburi,
and K. Ueno. Unitary representations of the quantum group
SU q (l, I): Structure of the dual space of Uq (sl(2)). Lett. Math.
Phys. 19 (1990), 187-194. MR 91h:33028.
[Moo91] J.A. Moody. The Burau representation of the braid group Bn
is unfaithful for large n. Bull. Amer. Math. Soc. 25 (1991),
379-384. MR 92b:20041.

[MR87] J.C. McConnell and J.C. Robson. Noncommutative Noethe-

rian Rings. John Wiley and Sons, Ltd, Chichester, 1987. MR
89j:16023.

[Ore33] O. Ore. Theory of non-commutative polynomials. Ann. Math.

(2) 34 (1933), 480-508.
[Par81] B. Pareigis. A non-commutative non-cocommutative Hopf
algebra in "nature". J. Algebra 70 (1981), 356-374. MR
83g:16021.

[Piu92] S. Piunikhin. Weights of Feynman diagrams, link polynomials

and Vassiliev knot invariants. Preprint, 1992.
 References 517

[Piu93] S. Piunikhin. Combinatorial expression for universal Vassiliev

link invariant. Harvard University, Preprint, 1993.

[Poi84] H. Poincare. Sur les groupes d'equations lineaires. Acta Math.

4 (1884), 201-311. (also in CEuvres, tome II, Gauthier-Villars,
Paris 1916).
[PT87] J.H. Przytycki and P. Traczyk. Invariants of links of Conway
type. Kobe J. Math. 4 (1987), 115-139.

[PW90] P. Po dIes and S. Woronowicz. Quantum deformation of

Lorentz group. Comm. Math. Phys 130 (1990), 383-431. MR
91:46100.
[PW91] B. Parshall and J.P. Wang. Quantum Linear Groups, volume
439 of Mem. Amer. Math. Soc. Amer. Math. Soc., Providence,
RI, 1991. MR 91g:16028.

[RA89] P. Roche and D. Arnaudon. Irreducible representations of the

quantum analogue of SU(2). Lett. Math. Phys. 17 (1989),
295-300. MR 90f:17013.

[Rad76] D.E. Radford. The order of the antipode of a finite-

dimensional Hopf algebra is finite. Amer. J. Math. 98 (1976),
333-355.

[Rad93a] D.E. Radford. Minimal quasitriangular Hopf algebras. J. Al-

gebra 157 (1993), 285-315.
[Rad93b] D.E. Radford. Solutions to the quantum Yang-Baxter equa-
tion and the Drinfeld double. J. Algebra 161 (1993), 20-32.

[Ree58] R. Ree. Lie elements and an algebra associated with shuffles.

Ann. of Math (2) 68 (1958), 210-220.

[Rei32] K. Reidemeister. Knotentheorie, volume 1 of Ergebnisse

der Mathematik und ihrer Grenzgebiete (Alte Folge). Julius
Springer, Berlin, 1932. English transl.: Knot Theory, BSC As-
sociates, Moscow, Idaho, 1983.

[Res90] N.Yu. Reshetikhin. Multiparametric quantum groups and

twisted quasitriangular Hopf algebras. Lett. Math. Phys. 20
(1990), 331-335. MR 91k:17012.

[Reu93] C. Reutenauer. Free Lie Algebras. Clarendon Press, Oxford,

1993.
[RoI76] D. Rolfsen. Knots and Links, volume 7 of Mathematics Lecture
Series. Publish or Perish, Berkeley, 1976.
518 References

[Ros88] M. Rosso. Finite-dimensional representations of the quantum

analog of the enveloping algebra of a complex simple Lie alge-
bra. Comm. Math. Phys. 117 (1988), 581-593. MR 90c:17019.
[Ros89] M. Rosso. An analogue of P.B.W. theorem and the universal
R-matrix for Uhs[(N + 1). Comm. Math. Phys. 124 (1989),
307-318. MR 90h:17019.
[Ros90] M. Rosso. Analogues de la formule de Killing et du theoreme
d'Harish-Chandra pour les groupes quantiques. Ann. Sci.
Ecole. Norm. Sup. {4} 23 (1990), 445-467. MR 93e:17026.
[Ros92] M. Rosso. Representations des groupes quantiques. In
Seminaire Bourbaki, volume 201-203 of Asterisque, pages 443-
483. S.M.F., Paris, 1992. MR 93b:17050.
[RSTS88] N.Yu. Reshetikhin and M.A. Semenov-Tian-Shansky. Quan-
tum R-matrices and factorization problems. J. Geom. Phys.
5 (1988), 533-550. MR 92g:17019.
[RT90] N.Yu. Reshetikhin and V.G. Turaev. Ribbon graphs and their
invariants derived from quantum groups. Comm. Math. Phys.
127 (1990), 1-26. MR 91c:57016.

[RT91] N.Yu. Reshetikhin and V.G. Turaev. Invariants of 3-manifolds

via link polynomials and quantum groups. Invent. Math. 103
(1991), 547-597. MR 92b:57024.

[RTF89] N.Yu. Reshetikhin, L.A. Takhtadjian, and L.D. Faddeev.

Quantization of Lie groups and Lie algebras. Algebra i Analiz 1
(1989),178-206. English transl.: Leningrad Math. J. 1 (1990),
193-225. MR 90j:17039.

[Sa190] H. Saleur. Representations of Uq (s[(2)) for q a root of unity.

In Number theory and physics {Les Houches, 1989}, volume 47
of Springer Proc. Phys., pages 68-76. Springer, Berlin, 1990.
MR 91i:17023.

[Sch92] P. Schauenburg. On Coquasitriangular Hop! Algebras and

the Quantum Yang-Baxter Equation, volume 67 of Algebra
Berichte. Verlag Reinhard Fischer, Miinchen, 1992.

[SC56] Seminaire Cartan. Algebres d'Eilenberg-Mac Lane et homo-

topie {1954-55}. Ecole Normale Superieure, Paris, 1956. (also
published by W.A. Benjamin, Inc., New York, Amsterdam,
1967).

[Ser65] J.-P. Serre. Lie Algebras and Lie Groups. W.A. Benjamin,
Inc., New York, Amsterdam, 1965.
 References 519

[Ser93] J.-P. Serre. Gebres. Enseign. Math. (2) 39 (1993), 33-85.

[Shu90] M.C. Shum. Tortile tensor categories. Ph.D. Thesis, Mac-
quarie University, Nov. 1989, Macquarie Math. Report No.
900047, 1990. J. Pure Appl. Algebra, to appear in 1994.

[Sk185] E.K. Sklyanin. ~n an algebra generated by quadratic rela-

tions. Uspekhi Mat. Nauk 40:2 (242) (1985), 214. (in Rus-
sian).

[Sla66] L.J. Slater. Generalized Hypergeometric Functions. Cam-

bridge University Press, Cambridge, 1966.

[SR72] N. Saavedra Rivano. Categories Tannakiennes, volume 265

of Lecture Notes in Math. Springer-Verlag, Berlin, New York,
1972.

[Sta92] T. Stanford. Finite-type invariants of knots, links, and graphs.

Columbia University, Preprint, 1992.

[Sta93] T. Stanford. Braid commutators and Vassiliev invariants.

Preprint, 1993.

[Sud90] A. Sudbery. Consistent multiparametric quantisation of

GL(n). J. Phys. A 23 (1990), L697-L704. MR 91m:17022.

[Swe69] M. Sweedler. Hopf Algebras. W.A. Benjamin, Inc., New York,

1969.

[Taf71] E.J. Taft. The order of the antipode of finite-dimensional Hopf

algebras. Proc. Nat. Acad. Sci. U.S.A. 68 (1971), 2631-2633.

[Tak81] M. Takeuchi. Matched pairs of groups and bismash products

of Hopf algebras. Comm. Algebra 9 (1981), 841-882. MR
83f:16013.
[Tak85] M. Takeuchi. Topological coalgebras. J. Algebra 97 (1985),
505-539.
[Tak89] M. Takeuchi. Quantum orthogonal and symplectic groups and
their embedding into quantum GL. Proc. Japan Acad. Ser. A
Math. Sci. 65 (1989), 55-58. MR 90j:16024.
[Tak92a] M. Takeuchi. Finite-dimensional representations of the quan-
tum Lorentz group. Comm. Math. Phys. 144 (1992), 557-580.
MR 93c:17030.

[Tak92b] M. Takeuchi. Hopf algebra techniques applied to the quantum

group U/5[(2. In Deformation Theory and Quantum Groups
with Applications to Mathematical Physics (Amherst, 1990),
520 References

volume 134 of Contemp. Math., pages 309-323. Amer. Math.

Soc., Providence, RI, 1992.
[Tak92c] M. Takeuchi. Some topics on GLq(n). J. Algebra 147 (1992),
379-410. MR 93b:17055.
[Tan90] T. Tanisaki. Harish-Chandra isomorphisms for quantum alge-
bras. Comm. Math. Phys. 127 (1990), 555-571.
[Trj33] W.J. Trjitzinsky. Analytic theory of linear q-difference equa-
tions. Acta Math. 61 (1933), 1-38.
[Tur88] V.G. Turaev. The Yang-Baxter equation and invariants of
links. Invent. Math. 92 (1988), 527-553. MR 8ge:57003.
[Tur89] V.G. Turaev. Operator invariants of tangles and R-matrices.
Izv. Akad. Nauk SSSR Ser. Math. 53:5 (1989), 1073-1107.
English transl.: Math. USSR-Izv. 35 (1990), 411-444. MR
91e:17011.
[Tur92] V.G. Turaev. Modular categories and 3-manifold invariants.
Intern. J. Modern Phys. B 6 (1992), 1807-1824.
[Tur94] V.G. Turaev. Quantum invariants of knots and 3-manifolds.
W. de Gruyter, Berlin, 1994.
[TW80] E.J. Taft and R.L. Wilson. There exist finite-dimensional Hopf
algebras with antipodes of arbitrary even order. J. Algebra 62
(1980), 283-291. MR 81d:16008.
[Vak89] L.L. Vaksman. q-analogues of Clebsch-Gordan coefficients,
and the algebra of functions on the quantum SU(2) group.
Dokl. Akad. Nauk SSSR 306:2 (1989), 269-271. English
transl.: Soviet Math. Dokl. 39 (1989), 467-470. MR 90k:33018.
[Var74] V.S. Varadarajan. Lie groups, Lie algebras, and their repre-
sentations. Prentice Hall, Inc., Englewood Cliffs, NJ, 1974.
[Vas90] V.A. Vassiliev. Cohomology of knot spaces. In V.I. Arnold,
editor, Theory of singularities and applications, volume 1 of
Adv. Soviet Math., pages 23-69. Amer. Math. Soc., Providence,
RI, 1990. MR 92a:57016.
[Vas92] V.A. Vassiliev. Complements of discriminants of smooth
maps, volume 98 of Transl. Math. Monographs. Amer. Math.
Soc., Providence, RI, 1992.
[Vog93] P. Vogel. Invariants de Vassiliev des namds (d'apres D. Bar-
Natan, M. Kontsevich et V.A. Vassiliev). In Seminaire Bour-
baki, volume 216 of Asterisque, pages 213-232. S.M.F., Paris,
1993.
 References 521

[Was87] W. Wasow. Asymptotic Expansions for Ordinary Differential

Equations. Dover Pub!., New York, 1987.
[Wor87a] S.L. Woronowicz. Compact matrix pseudo-groups. Comm.
Math. Phys. 111 (1987), 613-665. MR 88m:46079.
[Wor87b] S.L. Woronowicz. Twisted SU(2) group. An example of non-
commutative differential calculus. Publ. Res. Inst. Math. Sci.
23 (1987), 117-181. MR 88h:46130.

[Wor88] S.L. Woronowicz. Tannaka-Krein duality for compact ma-

trix pseudo-groups. twisted SU(N) groups. Invent. Math. 93
(1988), 35-76. MR 90e:22033.

[Yam89] H. Yamane. A Poincare-Birkhoff-Witt theorem for quantized

enveloping algebras of type AN' Publ. Res. Inst. Math. Sci.
25 (1989), 503-520. MR 91a:17016.

[Yan67] C.N. Yang. Some exact results for the many-body problem in
one dimension with repulsive delta-function interaction. Phys.
Rev. Lett. 19 (1967), 1312-1315.
[Yet88] D.N. Yetter. Markov algebras. In Braids (Santa Cruz, 1986),
volume 78 of Contemp. Math., pages 705-730. Amer. Math.
Soc., Providence, RI, 1988.

[Yet90] D.N. Yetter. Quantum groups and representations of monoidal

categories. Math. Proc. Cambridge Philos. Soc. 108 (1990),
261-290. MR 91k:16028.

[Zag93] D. Zagier. Values of zeta functions and their applications.

M.P.I., Bonn, Preprint, 1993. Proc. E.C.M., to appear.
Index

0,390,410 C, 100, 404, 412, 461

C(F), 297
~-operation, 243, 259 Ch , 412
p~, 378 q[h]J, 385
p~, 377 C q , 131-133
p~, 307 Cn(g, M), 420
p~z, 456, 457, 459 crev , 337
p~h, 459 cstr , 288, 337, 377
T(Pl,ql, ... ,Pk,qk)' 466, 482 D(()), 365, 366
<P, 369, 443, 465 D(A),333
<p(A,B), 465, 469, 479 det q , 79
<P F , 373 dim q , 358, 363, 364, 411
<P KZ ' 459, 463, 469, 471, 478, 498 g[(2), 99
<Pm,q' 255, 313, 418, 486 GL(2), 11,33, 57
g[(n), 94
A,500 g[(V),94
AB,496 GLq(2), 83, 87, 196
A(c), 188, 194, 197 GL q(n),91
A X ,9 gr(A), 14
Ag,t, 449, 458, 459, 468, 471 Hn(g, M), 421
B,321 L(A),94
Bn , 263, 377, 454, 459 M, 388
Bn(g, M), 421
M(2), 10,33,47,57,65, 112
B q , 224, 227
M(n),47
Br(C, V), 327
M p ,q(2), 90, 198
524 Index

Mq(2), 78, 83, 89, 150, 194,418 X n , 267, 269, 454, 479
Mq(n), 91, 198 Y", 267, 270, 452, 454, 456
[n], 121 YB(C), 324
(n)q,74 Z(K), 490, 505
P n , 454 zn(fl, M), 421
Pn , 454 Zq' 133
Prim(C),48 Z(C),330
Prim(H), 495 Z[Xc(G)], 360
CJg,v, 411, 418, 486, 503
R, 313, 358, 411, 501
abelian Lie algebra, 94, 95, 117
Rh , 409, 412, 458, 459
acyclic chain complex, 447
R KZ , 471
adjoint corepresentation, 2:36
Sn' 454, 456
adjoint functors, 279, 291, 400
sl(2), 99, 405, 412
adjoint representation, 99, 107,208,
SL(2), 11,33,57,65, 114
404, 429
s[(n),94
of a Lie algebra, 208
SL p,q(2),91
affine
SLq(2), 83, 87, 90, 153, 196, 366,
line, 8
418
plane, 8, 65, 109
SLq(n), 91, 163
Alexander polynomial, 246, 486
T, 299, 305, 312, 360
T'(V),67' . algebra, 3, 39
of differential operators, 21
Tens(C, D), 288
of pseudo-differential opera-
Tensstr(C, D), 288
tors, 21
tr q' 354, 363
topologically generated by gen-
U~, 125, 139, 143
erators and relations, 407
U(s[(2)), 99, 109, 114, 118
antilinear, 86
Uh , 412
antipode, 50, 52, 55, 58, 84, 197
Uh(g), 407, 419, 430, 458
antisymmetric tensors, 37
Uh (s[(2)), 412, 419
antisymmetrization, 436
Uq , 122, 140, 412
associativity constraint, 281, 369,
Uq , 136, 224, 230, 238, 364 377, 460, 499
Uq (g), 138, 419 autonomous tensor category, 347,
Uq (s[(2)), 121, 122, 138, 140, 153, 353, 380
162, 168, 223, 412
Uq(s[(f + 1)), 139
Uq(s[(n)), 163 bar complex, 437, 447
V(A), 129, 138 basic hypergeometric series, 90
V(n), 102, 118, 417 basis of a free module, 26
Vn ,417 bialgebr~ 46, 58, 83, 285, 493

VA' 406, 409, 417 of a monoid, 46, 51

VA ,409 bicrossed product
X(G), 319 of bialgebras, 204, 213
Xc(G), 319, 360
of groups, 201
bimodule, 71
 Index 525

binomial coefficients, 74, 122 centre (cont.)

braid, 262, 307 of a tensor category, 330, 333,
category, 321, 325 338, 366
diagram, 263, 313 of the braid group, 269
group, 264, 272, 307,377,395, chain
454, 499 complex, 447
relations, 265, 327 map, 447
representation, 395, 459 chord diagram, 487, 491
associated to a braided classical limit of a quantum
quasi-bialgebra, 378 enveloping algebra, 397
associated to a braided Clebsch-Gordan, 157
tensor category, 377 formula, 105
braided closed polygonal arc, 242
bialgebra, 173, 175, 177, 178, closure
198, 318 of a braid, 272, 307
Hopf algebra, 173, 175, 183, of a ribbon, 359
216, 229, 333, 361, 365 coaction, 62
quasi-bialgebra, 371, 376, 377, coadjoint
394 corepresentation, 236
quasi-Hopf algebra, 379, 380, representation, 210, 429
382 coalgebra, 40, 431
tensor of a set, 41, 42, 46, 292
category, 315,327,329,330, coassociative, 40, 43
338, 341, 371, 376, 377, cobar complex, 432, 434, 445, 447
410, 460, 495, 498 coboundary, 421
equivalence, 376, 460 cobraided
functor, 327, 329 bialgebra, 184, 191, 195, 197,
braiding, 315, 341, 377, 410, 460, 198, 237, 238
495, 499 Hopf algebra, 185, 196, 197,
Burau representation, 269, 272 365
co chain complex, 447
canonical 2-tensor, 412, 449, 460 co commutative
of a quantum enveloping bialgebra, 58, 174, 318
algebra, 397 coalgebra, 40, 44
Cartan matrix, 139, 405, 407 Hopf algebra, 350, 495
Casimir element, 100, 103, 404, cocycle, 421
412, 427, 461, 496 coefficient space of a representa-
category, 275 tion, 71
associated to a family of groups, co evaluation map, 29, 60, 306,347
276 cohomology
of ribbons, 358, 501 group, 447
central element, 503 of a coalgebra, 431
centre, 4, 90, 100, 117, 130, 133, of a Lie algebra, 421, 446
134, 139, 181, 404, 410, coideal, 43
412, 431, 503
526 Index

combinatorially equivalent differential

links, 243 form, 450, 480
tangles, 259 graded algebra, 432, 435
commutative direct
algebra, 4, 40 product
bialgebra, 63 of groups, 401
commutativity constraint, 314 of vector spaces, 25, 292
comodule, 62, 366 sum of vector spaces, 25, 292
comodule-algebra, 64, 65, 85, 89, dominant weight, 406, 409, 431
91 Drinfeld
comodule-coalgebra, 236 associator, 369, 373, 465, 478,
comparison theorem, 448 479, 498
complete module, 388, 390 quantum double, 333, 380
completely reducible series, 478, 498
representation, 5 Drinfeld-Jimbo algebra, 139, 163,
composition of tangles, 261, 299 366, 407, 430, 449
comultiplication, 40, 285, 368 Drinfeld-Kohno theorem, 459, 460
concatenation, 7, 299 dual vector space, 27, 41, 46, 59,
configuration space, 267, 269, 479 62, 66, 71, 99, 110, 115,
congruent words, 298 118, 154, 177, 182,347
connected component of a link, duality, 120
242 between bialgebras, 110, 112,
connection, 450 114
convolution, 49, 69, 237, 479 between Hopf algebras, 150,
algebra, 68 153, 163
Conway
polynomial, 246, 253, 486 edge, 242
triple, 252, 313, 486 element
coproduct, 40 B,41O
coribbon algebra, 365 t, 405, 455, 459, 461
counit, 40, 285, 368 u, 180, 183, 410
counital, 40, 44 embedding of a chord diagram,
crossed 487
bimodule, 220, 237, 238, 333, endomorphism, 276
414 endpoint of a simple polygonal arc,
G-set, 319, 338, 360 242
product ofbialgebras, 207, 215, enhanced R-matrix, 306, 307, 311,
238 313
crossing point, 246, 260 enveloping algebra, 94, 97, 292,
curvature of a connection, 451 395, 406
equivalence of categories, 278
equivalent
Dehn's theorem, 245 braided quasi-bialgebras, 376,
derivation, 14, 21, 108, 117, 118, 378
146, 421 links, 244
 Index 527

equivalent (cont.) generic (cont.)

quasi-bialgebras, 375 link diagram, 249
tangles, 260 tangle diagram, 261
words, 295 graded
essentially surjective functor, 278 algebra, 12
evaluation map, 28, 60, 182, 306, coalgebra, 68, 436
347 dual vector space, 67, 436
exponential function, 392 tensor product, 292, 436
extension of Lie algebras, 422 vector space, 292
exterior algebra, 37, 435 grouplike element, 56, 292
groupoid, 276, 279
of isomorphisms, 277
faithful functor, 278
filtered algebra, 13 h-adic
flat connection, 451, 452, 456 completion, 388, 390, 401
flip, 4, 24, 40, 58, 63, 168, 314, topology, 387, 388, 401, 407
318, 393, 457, 495 Harish-Chandra homomorphism,
formal series, 385 131, 139
four-term relation, 488, 492, 498 Heisenberg Lie algebra, 398
framed hexagon axiom, 315, 372
singular knot, 484 highest weight
tangle, 270, 313, 358 module, 127, 130, 131, 138
free vector, 101, 127, 154,406,410
algebra, 7, 13, 119, 280 Hilbert's basis theorem, 19, 22
comodule, 63, 70 holonomy group, 450
Lie algebra, 119 Homfly polynomial, 253
module, 26 homology group, 447
tensor category on a class of homotopy, 273, 448, 451
morphisms, 297 Hopf
FRT construction, 188, 194, 198 *-algebra, 87, 90, 92, 118, 119,
fully faithful functor, 278 162, 163
functor, 277 algebra, 51, 58, 84, 97, 347,
fundamental group 381, 493
of a link, 245, 361 of a group, 51, 56, 219
of a topological space, 269, link, 245, 256, 313, 361
270, 274, 451, 454 hyperlogarithm, 481

g-invariant, 435 identity morphism, 276

gauge transformation, 373, 381, 394, indecomposable module, 5
399, 442, 460 infinitesimal
Gauss polynomials, 74 braid group relations, 454, 478,
generators of a strict tensor cate- 492
gory, 297 braiding, 494, 497
generic symmetric category, 495, 498,
isotopy, 250 500, 502
528 Index

inner derivation, 421 linking number, 245

invariant loop, 267, 273, 456
of finite degree, 485, 505
symmetric 2-tensor, 397, 499 Mac Lane's coherence theorem, 291,
inverse 293
limit, 386, 401 Markov's theorem, 273
topology, 402 matched pair
system, 400, 401 of bialgebras, 203, 211
irreducible representation, 5 of groups, 200, 207
isomorphism, 276 Milnor and Moore, theorem of, 494
isotopic minimal braided Hopf algebra, 238
link diagrams, 249 mirror image, 253, 257
links, 243 module, 4, 61
tangles, 260 over a Lie algebra, 98
isotopy, 243, 260, 268, 299 module-algebra, 107-109, 117, 118,
invariant, 244, 273, 305, 360, 148
367, 410, 418, 485 module-coalgebra, 203, 236
of finite type, 485 monodromy,451,454-456,458,459,
iterated integral, 465, 480 469
monoidal category, 293
Jones polynomial, 253, 486 morphism
two-variable, 253 of a category, 275
Jones-Conway polynomial, 253, 312, of algebras, 3, 40
486 of bialgebras, 46, 69
of braided quasi-bialgebras, 372
Kauffman's bracket, 270, 272 of coalgebras, 41, 45
Killing form, 404, 405 of comodules, 62
Knizhnik-Zamolodchikovequations, of Hopf algebras, 51, 69
382, 449, 455, 459, 469, of modules, 5, 61
479 of quasi-bialgebras, 371
knot, 242 of topological algebras, 392
Kontsevich universal knot invari- of topological braided quasi-
ant, 490, 500, 504, 505 bialgebras, 394
Koszul complex, 439 of topological quasi-bialgebras,
393
Laurent polynomials, 9 multiple zeta value, 465, 466, 483
left
duality, 342, 366, 380, 410, natural
502 isomorphism, 277
unit constraint, 282, 315, 369 tensor
Leibniz formula, 21 isomorphism, 288
Lie algebra, 93, 292, 395, 420, 455 transformation, 287
link, 242, 258, 300, 305 transformation, 277
diagram, 246 Noetherian, 18, 21, 73, 81, 118,
projection, 246 123
 Index 529

non-commutative residue, 22 q-analogue, 411

q- binomial formula, 74
object of a category, 275 q-Chu-Vandermonde formula, 75
opposite q-difference equation, 90
algebra, 4 q-differentiation, 89, 90
coalgebra, 41 q-exponential, 76, 90
Lie algebra, 94 q- factorial, 74
order of a link, 242, 244 q-hypergeometric series, 90
Ore extension, 15, 19, 22, 73, 81, q- Pascal identity, 74
91, 123 quadratic algebra, 91
oriented link, 242 quantization, 397, 460
origin of a simple polygonal arc, quantum
242 3j-symbols, 158
overcrossing, 246, 260 Casimir element, 130, 132, 139,
144, 412
parallel transport, 450 Clebsch-Gordan coefficients,
partial 158, 161, 163
trace, 31, 306 determinant, 79, 84, 91, 152,
transpose, 306 195
transposition, 30 dimension, 358-360, 363-365,
pentagon axiom, 282, 293, 369 411, 417, 504
perfect duality, 110 double, 213, 215, 216, 219,
permutation of a braid, 267, 321, 227, 238, 366, 414, 419
499 enveloping algebra, 395, 399,
Picard's method of approximation, 401, 408, 430, 442, 449,
466, 481 459
piecewise-linear, 242 group invariant, 486, 503
Poincare series, 20 partial derivative, 147
Poincare-Birkhoff-Witt Theorem, plane, 72, 85, 118, 146, 148
98 trace, 354, 359, 363, 504
polygonal arc, 242 Yang-Baxter module, 238
poly logarithm, 482 quasi-bialgebra, 368, 369, 373, 381
polynomial algebra, 8, 13 quasi-cocommutative
presentation of a strict tensor cat- bialgebra, 173
egory by generators and Hopf algebra, 173, 174, 180,
relations, 298, 300, 302, 182
303, 313 quasi-commutative bialgebra, 185
primitive element, 48, 119, 397, quasi-Hopf algebra, 379, 381
446, 494, 495, 505 quasi-triangular bialgebra, 174, 198
product quasi-triangular quasi-bialgebra, 371,
algebra, 4 382
of categories, 276 QUE,395
of groups, 202 quotient-coalgebra, 43
of loops, 268, 273
pure braid group, 270, 454
530 Index

R-matrix, 167, 169, 171, 179, 194, sign of a permutation, 438

198, 234, 311 simple
real form of a complex Lie alge- module, 5, 102, 103, 109, 115,
bra, 119 118, 128, 134, 136, 138,
Rees algebra, 20 139, 154, 155, 168, 234,
regular 364, 406, 410, 417, 427,
link projection, 246 431,503
tangle projection, 260 polygonal arc, 242
Reidemeister moves, 248, 261, 271, singular knot, 484
273 skein module, 254
representation, 5 relation, 253, 486
of the tangle category, 305 skew
resolution, 447 antipode, 55
restricted dual, 71, 120, 163 polynomial ring, 22
of a Hopf algebra, 71 source of a morphism, 276
reverse category, 337 split extension of Lie algebras, 422
ribbon, 270, 313, 358 square
algebra, 361, 365, 366, 382 of the antipode, 70, 85, 142,
category, 349, 360, 361, 366, 179, 382
410, 500 root, 399, 400, 443
functor, 504 strict tensor
Riemann's zeta function, 468, 483 category, 282, 286, 288, 295,
right 299, 315, 319, 329, 332,
comodule, 62 337, 339, 377, 380, 500
duality, 346, 351, 366, 380 functor, 287, 306, 312, 360
unit constraint, 282, 315, 369 structure theorem for bimodules
rigidity theorem, 426, 442 over a Hopf algebra, 71
root subalgebra, 3
length, 406, 407 subcategory, 276
of unity, 85, 90, 122, 134, 139 subcomodule, 62
submodule, 5
Sweedler's
scalar product, 155 Hopf algebra, 68, 174, 196,
Schur's lemma, 20 198, 363
semi direct product of groups, 202 sigma notation, 43, 173, 434,
semisimplicity, 143, 163 461
semisimple symmetric
Lie algebra, 138,403,407,410, algebra, 35, 56, 95, 435, 494
418, 427, 430, 434, 442, group, 267, 454, 456, 480
449, 459, 460, 479, 496, invariant 2-tensor, 405, 412,
499, 503 449, 455, 459, 463
module, 5, 104, 144, 406 tensor category, 320, 337,338,
separated module, 388, 390 350, 371, 494
shuffie, 47, 68, 97, 480 tensors, 37
bialgebra, 68
 Index 531

symmetrization trace, 21
map, 98 of an endomorphism, 31
of a universal R-matrix, 462 transpose, 343
symmetry, 320 transposition, 30
trefoil knot, 244, 256, 313, 361
tangle, 257, 299 triangle axiom, 282, 293, 369
category, 300, 306, 312, 360 trivial
diagram, 260, 302 comodule, 63
projection, 260 knot, 244, 312
target of a morphism, 276 link, 244
tensor module, 58, 59, 98, 102
algebra, 34, 47, 56 topological
category, 282, 293, 368 bialgebra, 457
coalgebra, 67 braided quasi-bialgebra, 394
equivalence, 288, 375 twist, 349, 359, 361, 366, 411, 500
equivalent categories, 288 two-variable Jones polynomial, 253
functor, 287, 375
product under crossing , 246, 260
algebra, 393 unit of a tensor category, 282, 284,
of algebras, 32 305, 316
of bialgebras, 207 universal
of coalgebras, 42, 45, 66 R-form, 185, 191, 195-197,365
of comodules, 63 R-matrix, 173, 175, 177, 179,
of linear maps, 26 198, 216, 220, 230, 371,
of modules, 57, 98 376, 393, 398, 409, 412,
of vector spaces, 23, 280 413, 419, 458, 459
theorem of Milnor and Moore, 494
topological Vassiliev invariant, 505
algebra, 392, 406, 424 Verma module, 130, 138
bialgebra, 393, 458 vertex, 242
braided quasi-bialgebra, 393,
395, 460, 469, 471 weight, 101, 127
crossed bimodule, 395, 414 Weyl algebra, 18
module, 394 Whitehead lemmas, 117, 429
quasi-bialgebra, 393 word,295
tensor product, 390, 410
Yang-Baxter
topologically free
equation, 167, 171, 178, 179,
algebra, 407
185, 191, 196, 198, 234,
module, 388, 390, 400, 409,
266, 306, 311, 317, 395,
418
418
torsion free module, 390
operator, 323
tortile tensor category, 366
Graduate Texts in Mathematics

continued from page ii

65 WELLS. Differential Analysis on Complex 94 WARNER. Foundations of Differentiable

Manifolds. 2nd ed. Manifolds and Lie Groups.
66 WATERHOUSE. Introduction to Affine Group 95 SHIRY AYEV. Probability.
Schemes. 96 CONWAY. A Course in Functional Analysis.
67 SERRE. Local Fields. 2nd ed.
68 WEIDMANN. Linear Operators in Hilbert 97 KOBLITZ. Introduction to Elliptic Curves and
Spaces. Modular Forms. 2nd ed.
69 LANG. Cyclotomic Fields II. 98 BROCKERiTOM DIECK. Representations of
70 MASSEY. Singular Homology Theory. Compact Lie Groups.
71 FARKAs/KRA. Riemann Surfaces. 2nd ed. 99 GRovE/BENSON. Finite Reflection Groups.
72 STILLWELL. Classical Topology and 2nd ed.
Combinatorial Group Theory. 2nd ed. 100 BERG/CHRISTENSEN/RESSEL. Harmonic
73 HUNGERFORD. Algebra. Analysis on Semigroups: Theory of Positive
74 DAVENPORT. Multiplicative Number Theory. Definite and Related Functions.
2nd ed. 101 EDWARDS. Galois Theory.
75 HOCHSCHILD. Basic Theory of Algebraic 102 V ARADARAJAN. Lie Groups, Lie Algebras and
Groups and Lie Algebras. Their Representations.
76 IITAKA. Algebraic Geometry. 103 LANG. Complex Analysis. 3rd ed.
77 HECKE. Lectures on the Theory of Algebraic 104 DUBROVIN/FoMENKO/NoVIKOV. Modern
Numbers. Geometry-Methods and Applications.
78 BURRIS/SANKAPPANAVAR. A Course in Part II.
Universal Algebra. 105 LANG. SL,(R).
79 WALTERS. An Introduction to Ergodic 106 SILVERMAN. The Arithmetic of Elliptic
Theory. Curves.
80 ROBINSON. A Course in the Theory of 107 OLVER. Applications of Lie Groups to
Groups. Differential Equations. 2nd ed.
81 FORSTER. Lectures on Riemann Surfaces. 108 RANGE. Holomorphic Functions and Integral
82 BOTT/Tu. Differential Forms in Algebraic Representations in Several Complex
Topology. Variables.
83 WASHINGTON. Introduction to Cyclotomic 109 LEHTO. Univalent Functions and Teichmiiller
Fields. Spaces.
84 IRELAND/RoSEN. A Classical Introduction to 110 LANG. Algebraic Number Theory.
Modern Number Theory. 2nd ed. 111 HUSEMOLLER. Elliptic Curves.
85 EDWARDS. Fourier Series. Vol. II. 2nd ed. 112 LANG. Elliptic Functions.
86 VAN LINT. Introduction to Coding Theory. 113 KARATZAS/SHREVE. Brownian Motion and
2nd ed. Stochastic Calculus. 2nd ed.
87 BROWN. Cohomology of Groups. 114 KOBLITZ. A Course in Number Theory and
88 PIERCE. Associative Algebras. Cryptography. 2nd ed.
89 LANG. Introduction to Algebraic and Abelian 115 BERGERlGOSTlAUX. Differential Geometry:
Functions. 2nd ed. Manifolds, Curves, and Surfaces.
90 BR0NDSTED. An Introduction to Convex 116 KELLEy/SRINIVASAN. Mea,ure and Integral.
Polytopes. Vol. I.
91 BEARDON. On the Geometry of Discrete 117 SERRE. Algebraic Groups and Class Fields.
Groups. 118 PEDERSEN. Analysis Now.
92 DIESTEL. Sequences and Series in Banach 119 ROTMAN. An Introduction to Algebraic
Spaces. Topology.
93 DUBROVIN/FoMENKO/NoVIKOV. Modern 120 ZIEMER. Weakly Differentiable Functions:
Geometry-Methods and Applications. Part Sobolev Spaces and Functions of Bounded
I. 2nd ed. Variation.
121 LANG. Cyclotomic Fields I and II. Combined 140 AUBIN. Optima and Equilibria. An
2nd ed. Introduction to Nonlinear Analysis.
122 REMMERT. Theory of Complex Functions. 141 BECKERiWEISPFENNING/KREDEL. Griibner
Readings in Mathemntics Bases. A Computational Approach to
123 EBBINGHAUslHERMES et al. Numbers. Commutative Algebra.
Readings in Mathemntics 142 LANG. Real and Functional Analysis. 3rd ed.
124 DUBROVINlFoMENKO!NOVIKOV. Modern 143 DOOB. Measure Theory.
Geometry-Methods and Applications. Part 144 DENNIS/FARB. Noncommutative
III. Algebra.
125 BERENSTEIN/GAY. Complex Variables: An 145 VICK. Homology Theory. An
Introduction. Introduction to Algebraic Topology.
126 BOREL. Linear Algebraic Groups. 2nd ed.
127 MASSEY. A Basic Course in Algebraic 146 BRIDGES. Computability: A
Topology. Mathematical Sketchbook.
128 RAUCH. Partial Differential Equations. 147 ROSENBERG. Algebraic K- Theory and
129 fuLTON/HARRIS. Representation Theory: A Its Applications.
First Course. 148 ROTMAN. An Introduction to the
Readings in Mathemntics Theory of Groups. 4th ed.
130 DODSON/POSTON. Tensor Geometry. 149 RATCLIFFE. Foundations of Hyperbolic
131 LAM. A First Course in Noncommutative Manifolds.
Rings. 150 EISENBUD. Commutative Algebra with
132 BEARDON. Iteration of Rational Functions. a View Toward Algebraic Geometry.
133 HARRIS. Algebraic Geometry: A First Course. 151 SILVERMAN. Advanced Topics in the
134 ROMAN. Coding and Information Theory. Arithmetic of Elliptic Curves.
135 ROMAN. Advanced Linear Algebra. 152 ZIEGLER. Lectures on Polytopes.
136 ADKINS/WEINTRAUB. Algebra: An Approach 153 FULTON. Algebraic Topology: A First
via Module Theory. Course.
137 AxLERIBoURDON/RAMEY. Harmonic Function 154 BROWNIPEARCY. An Introduction to
Theory. Analysis.
138 COHEN. A Course in Computational 155 KASSEL. Quantum Groups.
Algebraic Number Theory. 156 KECHRIS. Classical Descriptive Set
139 BREDON. Topology and Geometry. Theory.