Rolf Berndt

Representations
of Linear Groups
Rolf Berndt

Representations
of Linear Groups
An Introduction Based
on Examples from Physics
and Number Theory
Bibliografische information published by Die Deutsche Nationalbibliothek
Die Deutsche Nationalbibliothek lists this publication in the Deutschen Nationalbibliografie;
detailed bibliographic data is available in the Internet at <http://dnb.d-nb.de>.

Prof. Dr. Rolf Berndt

Department of Mathematics
University of Hamburg
Bundesstraße 55
D-20146 Hamburg
Germany

[email protected]

Mathematics Subject Classification

20G05, 22E45, 20C25, 20C35, 11F70

First edition, July 2007

All rights reserved

© Friedr. Vieweg & Sohn Verlag | GWV Fachverlage GmbH, Wiesbaden 2007
Editorial Office: Ulrike Schmickler-Hirzebruch | Susanne Jahnel
Vieweg is a company in the specialist publishing group Springer Science+Business Media.
www.vieweg.de

No part of this publication may be reproduced, stored in a retrieval system

or transmitted, mechanical, photocopying or otherwise without prior
permission of the copyright holder.

Cover design: Ulrike Weigel, www.CorporateDesignGroup.de

Printing and binding: MercedesDruck, Berlin
Printed on acid-free paper
Printed in Germany

ISBN 978-3-8348-0319-1
Preface

There are already many good books on representation theory for all kinds of groups.
Two of the best (in this author’s opinion) are the one by A.W. Knapp: “Representation
Theory for Semisimple Groups. An Overview based on Examples” [Kn1] and by G.W.
Mackey: “Induced Representations in Physics, Probability and Number Theory” [Ma1].
The title of this text is a mixture of both these titles, and our text is meant as a very
elementary introduction to these and, moreover, to the whole topic of group represen-
tations, even infinite-dimensional ones. As is evident from the work of Bargmann [Ba],
Weyl [Wey] and Wigner [Wi], group representations are fundamental for the theory of
atomic spectra and elementary physics. But representation theory has proven to be an
inavoidable ingredient in other fields as well, particularly in number theory, as in the
theory of theta functions, automorphic forms, Galois representations and, finally, the
Langlands program. Hence, we present an approach as elementary as possible, having in
particular these applications in mind.

This book is written as a summary of several courses given in Hamburg for students of
Mathematics and Physics from the fifth semester on. Thus, some knowledge of linear
and multilinear algebra, calculus and analysis in several variables is taken for granted.
Assuming these prerequisites, several groups of particular interest for the applications in
physics and number theory are presented and discussed, including the symmetric group
Sn as the leading example for a finite group, the groups SO(2), SO(3), SU(2), and SU(3)
as examples of compact groups, the Heisenberg groups and SL(2, R), SL(2, C), resp. the
Lorentz group SO(3, 1) as examples for noncompact groups, and the Euclidean groups
E(n) = SO(n)
Rn and the Poincaré group P = SO(3, 1)+
R4 as examples for semidi-
rect products.

This text would not have been possible without the assistance of my students and colleagues; it
is a pleasure for me to thank them all. In particular, D. Bahns, S. Böcherer, O. v. Grudzinski,
M. Hohmann, H. Knorr, J. Michaliček, H. Müller, B. Richter, R. Schmidt, and Chr. Schweigert
helped in many ways, from giving valuable hints to indicating several mistakes. Part of the
material was treated in a joint seminar with Peter Slodowy. I hope that a little bit of his way
of thinking is still felt in this text and that it is apt to participate in keeping alive his memory.
Finally, I am grateful to U. Schmickler-Hirzebruch and S. Jahnel from the Vieweg Verlag for
encouragement and good advice.
Contents

Introduction ix

0 Prologue: Some Groups and their Actions 1

0.1 Several Matrix Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
0.2 Group Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.3 The Symmetric Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1 Basic Algebraic Concepts 7

1.1 Linear Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Equivalent Representations . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 First Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Basic Construction Principles . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.1 Sum of Representations . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.2 Tensor Product of Representations . . . . . . . . . . . . . . . . . . 14
1.4.3 The Contragredient Representation . . . . . . . . . . . . . . . . . . 15
1.4.4 The Factor Representation . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.6 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2 Representations of Finite Groups 23

2.1 Characters as Orthonormal Systems . . . . . . . . . . . . . . . . . . . . . 23
2.2 The Regular Representation . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Characters as Orthonormal Bases . . . . . . . . . . . . . . . . . . . . . . 28

3 Continuous Representations 31
3.1 Topological and Linear Groups . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 The Continuity Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Invariant Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4 Representations of Compact Groups 43

4.1 Basic Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 The Example G = SU(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3 The Example G = SO(3) . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5 Representations of Abelian Groups 59

5.1 Characters and the Pontrjagin Dual . . . . . . . . . . . . . . . . . . . . . 59
5.2 Continuous Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . 60
viii CONTENTS

6 The Infinitesimal Method 63

6.1 Lie Algebras and their Representations . . . . . . . . . . . . . . . . . . . . 63
6.2 The Lie Algebra of a Linear Group . . . . . . . . . . . . . . . . . . . . . . 67
6.3 Derived Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.4 Unitarily Integrable Representations of sl(2, R) . . . . . . . . . . . . . . . 73
6.5 The Examples su(2) and heis(R) . . . . . . . . . . . . . . . . . . . . . . . 82
6.6 Some Structure Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.6.1 Specifications of Groups and Lie Algebras . . . . . . . . . . . . . . 85
6.6.2 Structure Theory for Complex Semisimple Lie Algebras . . . . . . 89
6.6.3 Structure Theory for Compact Real Lie Algebras . . . . . . . . . . 93
6.6.4 Structure Theory for Noncompact Real Lie Algebras . . . . . . . . 95
6.6.5 Representations of Highest Weight . . . . . . . . . . . . . . . . . . 97
6.7 The Example su(3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

7 Induced Representations 117

7.1 The Principle of Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.1.1 Preliminary Approach . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.1.2 Mackey’s Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.1.3 Final Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.1.4 Some Questions and two Easy Examples . . . . . . . . . . . . . . . 126
7.2 Unitary Representations of SL(2, R) . . . . . . . . . . . . . . . . . . . . . 130
7.3 Unitary Representations of SL(2,C) and of the Lorentz Group . . . . . . . 143
7.4 Unitary Representations of Semidirect Products . . . . . . . . . . . . . . . 147
7.5 Unitary Representations of the Poincaré Group . . . . . . . . . . . . . . . 154
7.6 Induced Representations and Vector Bundles . . . . . . . . . . . . . . . . 161

8 Geometric Quantization and the Orbit Method 173

8.1 The Hamiltonian Formalism and its Quantization . . . . . . . . . . . . . . 173
8.2 Coadjoint Orbits and Representations . . . . . . . . . . . . . . . . . . . . 178
8.2.1 Prequantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
8.2.2 Example: Construction of Line Bundles over M = P1 (C) . . . . . 181
8.2.3 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
8.2.4 Coadjoint Orbits and Hamiltonian G-spaces . . . . . . . . . . . . . 186
8.2.5 Construction of an Irreducible Unitary Representation by an Orbit 196
8.3 The Examples SU(2) and SL(2, R) . . . . . . . . . . . . . . . . . . . . . . 197
8.4 The Example Heis(R) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
8.5 Some Hints Concerning the Jacobi Group . . . . . . . . . . . . . . . . . . 209

9 Epilogue: Outlook to Number Theory 215

9.1 Theta Functions and the Heisenberg Group . . . . . . . . . . . . . . . . . 216
9.2 Modular Forms and SL(2,R) . . . . . . . . . . . . . . . . . . . . . . . . . 221
9.3 Theta Functions and the Jacobi Group . . . . . . . . . . . . . . . . . . . . 236
9.4 Hecke’s Theory of L−Functions Associated to Modular Forms . . . . . . . 239
9.5 Elements of Algebraic Number Theory and Hecke L-Functions . . . . . . 246
9.6 Arithmetic L-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
9.7 Summary and Final Reflections . . . . . . . . . . . . . . . . . . . . . . . . 256

Bibliography 261

Index 266
Introduction

In this book, the groups enumerated in the Preface are introduced and treated as matrix
groups to avoid as much as possible the machinery of manifolds, Lie groups and bundles
(though some of it soon creeps in through the backdoor as the theory is further devel-
oped). Parallel to information about the structure of our groups we shall introduce and
develop elements of the representation theory necessary to classify the unitary represen-
tations and to construct concrete models for these representations. As the main tool
for the classification we use the infinitesimal method linearizing the representations of a
group by studying those of the Lie algebra of the group. And as the main tools for the
construction of models for the representations we use
– tensor products of the natural representation,
– representations given by smooth functions (in particular polynomials) living on a
space provided with an action of the group,
and
– the machinery of induced representations.
Moreover, because of the growing importance in physics and the success in deriving
branching relations, the procedure of geometric quantization and the orbit method, de-
veloped and propagated by Kirillov, Kostant, Duflo and many others shall be explained
via its application to some of the examples above.

Besides the sources already mentioned, the author was largely influenced by the now
classical book of Kirillov: “Elements of the Theory of Representations” [Ki] and the
more recent “Introduction to the Orbit Method” [Ki1]. Other sources were the books
by Barut and Raczka: “Theory of Group Representations and Applications” [BR], S.
Lang: “SL(2, R)” [La], and, certainly, Serre: “Linear Representations of Finite Groups”
[Se]. There is also the book by Hein: “Einführung in die Struktur- und Darstellungs-
theorie der klassischen Gruppen” [Hei], which follows the same principle as our text,
namely to do as much as possible for matrix groups, but does not go into the infinite-
dimensional representations necessary for important applications. Whoever is further
interested in the history of the introduction of representation theory into the theory of
automorphic forms and its development is referred to the classical book by Gelfand, Graev
and Pyatetskii-Shapiro: “Representation Theory and Automorphic Forms” [GGP], Gel-
bart’s: “Automorphic Forms on Adèle Groups” [Ge], and Bump’s: “Automorphic Forms
and Representations” [Bu]. More references will be given at the appropriate places in our
text; as already said, we shall start using material only from linear algebra and analysis.
But as we proceed more and more elements from topology, functional analysis, complex
function theory, differential and symplectic geometry will be needed. We will try to in-
troduce these as gently as possible but often will have to be very rudimentary and will
have to cite the hard facts without the proofs, which the reader can find in the more
refined sources.
To sum up, this text is prima facie about real and complex matrices and the nice and
sometimes advanced things one can do with them by elementary means starting from
a certain point of view: Representation theory associates to each matrix from a given
group G another matrix or, in the infinite-dimensional case, an operator acting on a
Hilbert space. One may want to ask, why study these representations by generally more
complicated matrices or operators if the group is already given by possibly rather simple
matrices? An answer to this question is a bit like the one for the pudding: the proof
is in the eating. And we hope our text will give an answer. To the more impatient
reader who wants an answer right away in order to decide whether to read on or not, we
offer the following rough explanation. Certain groups G appear in nature as symmetry
groups leaving invariant a physical or dynamical system. For example, the orthogonal
group O(3) is the symmetry group for the description of the motion of a particle in a
central symmetric force field, and the Poincaré group P is the symmetry group for the
motion of a free particle in Minkowski space. Then the irreducible unitary representa-
tions of G classify indivisible intrinsic descriptions of the system and, boldly spoken, can
be viewed as “elementary particles” for the given situation. Following Wigner and his
contemporaries, the parameters classifying the representations are interpreted as quan-
tum numbers of these elementary particles. . .

The importance of representations for number theory is even more difficult to put into a
nutshell. In the Galois theory of algebraic number fields (of finite degree) Galois groups
appear as symmetry groups G. Important invariants of the fields are introduced via cer-
tain zeta- or L-functions, which are constructed using finite-dimensional representations
of these Galois groups. Another aspect comes from considering smooth (holomorphic
resp. meromorphic) functions in several variables which are periodic or have a more gen-
eral covariant transformation property under the action of a given discrete subgroup of
a continuous group G, like for instance G = SL(2, R) or G a Heisenberg or a symplectic
group. Then these functions with preassigned types, e.g., theta functions or modular
forms, generate representation spaces for (infinite-dimensional) representations of the re-
spective group G.

Finally, we will give an overview over the contents of our text: In a prologue we will fix
some notation concerning the groups and their actions that we later use as our first ex-
amples, namely, the general and special linear groups over the real and complex numbers
and the orthogonal and unitary groups. Moreover, we present the symmetric group Sn
of permutations of n elements and some facts about its structure. We stay on the level of
very elementary algebra and stick to the principle to introduce more general notions and
details from group theory only when needed in our development of the representation
theory. We follow this principle in the first chapter where we introduce the concept of
linear representations using only tools from linear algebra. We define and discuss the
fundamental notions of equivalence, irreducibility, unitarity, direct sums, tensor product,
characters, and give some first examples.

The theory developed thus far is applied in the second chapter to the representations of
finite groups, closely following Serre’s exposition [Se]. We find out that all irreducible
representations may be unitarized and are contained in the regular representation.
In the next step we move on to compact groups. To do this we have to leave the purely
algebraic ground and take in topological considerations. Hence, in the third chapter, we
define the notion of a topological and of a (real or complex) linear group, the central
notion for our text. Following this, we refine the definition of a group representation by
adding the usual continuity condition. Then we adequately modify the general concepts
of the first chapter. We try to take over as much as possible from finite to compact
groups. This requires the introduction of invariant measures on spaces with a (from now
on) continuous group action, and a concept of integration with respect to these measures.
In the forth chapter we concentrate on compact groups and prove that the irreducible
representations are again unitarizable, finite-dimensional, fixed by their characters and
contained in the regular representation. But their number is in general not finite, in
contrast to the situation for finite groups. We state, but do not prove, the Peter-Weyl
Theorem. But to get a (we hope) convincing picture, we illustrate it by reproducing
Wigner’s discussion of the representaions of SU(2) and SO(3). We use and prove that
SU(2) is a double cover of SO(3). Angular momentum, magnetic and spin quantum
numbers make an appearance, but for further application to the theory of atomic spectra
we refer to [Wi] and the physics literature.

In a very short fifth chapter, we assemble some material about the representations of
locally compact abelian groups. We easily get the result that every unitary irreducible
representation is one-dimensional. But as can be seen from the example G = R, their
number need not be denumerable. More functional analysis than we can offer at this
stage is needed to decompose a given reducible representation into a direct integral of
irreducibles, a notion we not consider here.

Before starting the discussion of representations of other noncompact groups, we present

in chapter 6 an important tool for the classification of representations, the infinitesimal
method. Here, at first, we have to explain what a Lie algebra is and how to associate
one to a given linear group. Our main ingredient is the matrix exponential function and
its properties. We also reflect briefly on the notion of representations of Lie algebras.
Here again we are on purely algebraic, at least in examples, easily accessible ground.
We start giving examples by defining the derived representation dπ of a given group
representation π. We do this for the Schrödinger representation of the Heisenberg group
and the standard representation π1 of SU(2). Then we concentrate on the classification
of all unitary irreducible representations of SL(2, R) via a description of all (integrable)
representations of its Lie algebra. Having done this, we consider again the examples su(2)
and heis(R) (relating them to the theory of the harmonic oscillator) and give some hints
concerning the general structure theory of semisimple Lie algebras. The way a general
classification theory works is explained to some extent by considering Lie SU(3); we will
see how quarks show up.

Chapters 7 and 8 are the core of our book. In the seventh chapter we introduce the
concept of induced representations, which allows for the construction of (sometimes
infinite-dimensional) representations of a given group G starting from a (possibly one-
dimensional) representation of a subgroup H of G. To make this work we need again a
bit more Hilbert space theory and have to introduce quasi-invariant measures on spaces
with group action. We illustrate this by considering the examples of the Heisenberg
group and G = SU(2), where we rediscover the representations which we already know.
Then we use the induction process to construct models for the unitary representations of
SL(2, R) and SL(2, C). In particular, we show how holomorphic induction arises in the
discussion of the discrete series of SL(2, R) (here we touch complex function theory). We
insert a brief discussion of the Lorentz group GL = SO(3, 1)0 and prove that SL(2, C)
is a double cover of GL . To get a framework for the discussion of the representations of
the Poincaré group GP , which is a semidirect product of the Lorentz group with R4 , we
define semidirect products and treat Mackey’s theory in a rudimentary form. We outline
a recipe to classify and construct irreducible representations of semidirect products if
one factor is abelian. We do not prove the general validity of this procedure as Mackey’s
Imprimitivity Theorem is beyond the scope of our book, but we apply it to determine
the unitary irreducible representations of the Euclidean and the Poincaré group, which
are fundamental for the classification of elementary particles.

Under the heading of Geometric Quantization, in the eighth chapter we take an alterna-
tive approach to some material from chapter 7 by constructing representations via the
orbit method. Here we have to recall (or introduce) more concepts from higher analysis:
manifolds and bundles, vector fields, differential forms, and in particular the notion of
a symplectic form. We can again use the information and knowledge we already have
of our examples G = SL(2, R), SU(2) and the Heisenberg group to get a feeling what
should be done here. We identify certain spheres and hyperboloids as coadjoint orbits of
the respective groups, and we construct line bundles on these orbits and representation
spaces consisting of polarized sections of the bundles.

Finally, in the nineth and last chapter, we give a brief outlook on some examples where
representations show up in number theory. We present the notion of an automorphic
representation (in a rudimentary form) and explain its relation with theta functions and
automorphic forms. We have a glimpse upon Hecke’s and Artin’s L-functions and men-
tion the Artin conjecture.

We hope that some of the exercises and/or omitted proofs may give a starting point for
a bachelor thesis, and also that this text motivates further studies in a master program
in theoretical physics, algebra or number theory.

¡ Libro, afán
de estar en todas partes,
en soledad!

J. R. Jiménez
Chapter 0

Prologue: Some Groups and

their Actions

This text is mainly on groups which some way or another come from physics and/or
number theory and which can be described in form of a real or complex matrix group.

0.1 Several Matrix Groups

We will use the following notation:
The letter K indicates a field. The reader is invited to think of the field R of real or C
of complex numbers. Most of the things we do at the beginning of our text are valid also
for more general fields at least if they are algebraically closed and of characteristic zero,
but as this is only an introduction for lack of space we will not go into this to a greater
depth.
Mm,n (K) denotes the K-vector space of m × n matrices A = (aij ) with aij ∈ K (i =
1, . . . , m, j = 1, . . . , n) and Mn (K) stands for Mn,n (K).
Our groups will be (for some n) subgroups of of the general linear group of invertible
n × n-matrices
GL(n, K) := {A ∈ Mn (K); det A = 0 }.
As usual, we will denote the special linear group by

SL(n, K) := {A ∈ Mn (K); det A = 1 },

the orthogonal group by

O(n) := {A ∈ Mn (R); t AA = En },

resp. for n = p + q

O(p, q) := {A ∈ Mn (R); t ADp,q A = Dp,q },

where Dp,q is the diagonal matrix having p times 1 and q times −1 in its diagonal, and
the unitary group
U(n) := {A ∈ Mn (C); t AĀ = En },
2 0. Prologue: Some Groups and their Actions

resp. for n = p + q
U(p, q) := {A ∈ Mn (R); t ADp,q Ā = Dp,q }.
Again, addition of the letter S to the symbol of the group indicates that we take only
matrices with determinant 1, e.g.
SO(n) := {A ∈ O(n); det A = 1 }.
These groups together with some other families of groups, in particular the symplectic
groups showing up later, are known as classical groups.

Later on, we will often use subgroups consisting of certain types of block matrices, e.g.
the group of diagonal matrices
An := {D(a1 , . . . , an ); a1 , . . . , an ∈ K∗ },
where D(a1 , . . . , an ) denotes the diagonal matrix with the elements a1 , . . . , an in the di-
agonal, or the group of upper triangular matrices (or standard Borel group) Bn consisting
of matrices with zeros below the diagonal and the standard unipotent group Nn , the sub-
group of Bn where all diagonal elements are 1.
In view of the importance for applications, moreover, we distinguish several types of
Heisenberg groups: Thus the group N3 we just defined, is mostly written as
⎛ ⎞
1 x z
Heis (K) := { g = ⎝ 1 y ⎠ ; x, y, z ∈ K }.
1
In the later application to theta functions it will become clear that, though it may seem
more complicated, we shall better use the following description for the Heisenberg group.
⎛ ⎞
1 0 0 µ
⎜ λ 1 µ κ ⎟
Heis(K) := { g = (λ, µ, κ) := ⎜ ⎟
⎝ 0 0 1 −λ ⎠ ; µ, λ, κ ∈ K },
0 0 0 1
and the “higher dimensional” groups, which for typographical reasons we here do not
write as matrix groups
Heis(Kn ) := { g = (x, y, z); x, y ∈ Kn , z ∈ K }
with the multiplication law given by
gg  = (x + x , y + y  , z + z  + t xy  − t yx ).
We suggest to write elements of Kn as columns.

Exercise 0.1: Write Heis(Kn ) in matrix form. Show that Heis(Kn ) for n = 1 and the
other two Heisenberg groups above are isomorphic.

Exercise 0.2: Verify that the matrices

   
0 1 z 0
, , z ∈ C∗ := C \ {0}
−1 0 0 z̄
generate (by taking all possible finite products) a non-abelian subgroup of GL(2, C).
This is called the Weil group of R and plays an important role in the epilogue at the end
of our text.
0.2 Group Actions 3

0.2 Group Actions

Most groups “appear in nature” as transformation groups acting on some set or space.
This motivates the introduction of the following concepts.

Let G be a group with neutral element e and let X be a set.

Definition 0.1: G acts on X from the left iff a map

G × X −→ X , (g, x) −→ g · x

is given, which satisfies the conditions

g · (g  · x) = (gg  ) · x, e · x = x

for all g, g  ∈ G and x ∈ X .

Remark 0.1: If Aut X denotes the group of all bijections of X onto itself, the definition
says that we have a group homomorphism G −→ Aut X associating to every g ∈ G the
transformation x −→ g · x.

In this case the set X is also called a left G-set.

The group action is called effective iff no element except the neutral element e acts as
the identity, i.e. the homomorphism G −→ Aut X is faithful.

The group action is called transitive iff for every pair x, x ∈ X there is a g ∈ G with
x = g · x.

For x0 ∈ X we call the subset of X

G · x0 := {g · x0 ; g ∈ G }

an orbit of G (through x0 ) and the subgroup

Gx0 := {g ∈ G; g · x0 = x0 }

the isotropy group or the stabilizing group of x0 .

Example 0.1: G = GL(n, K) and its subgroups act on X = Kn from the left by matrix
multiplication
(A, x) −→ Ax

for x ∈ Kn (viewed as a column).

Exercise 0.3: Assure yourself that GL(n, K) acts transitively on X = Kn \ {0}.

Describe the orbits of SO(n).
4 0. Prologue: Some Groups and their Actions

Example 0.2: A group acts on itself from the left in three ways:
a) by left translation (g, g0 ) −→ g · g0 = gg0 =: λg g0 ,
b) by (the inverse of) right translation (g, g0 ) −→ g · g0 = g0 g −1 =: ρg−1 g0 ,
c) by conjugation (g, g0 ) −→ g · g0 = gg0 g −1 =: κg g0 .
Left- and right translations are obviously transitive actions. For a given group, the de-
termination of its conjugacy classes, i.e. the classification of the orbits under conjugation
is a highly interesting question, as we shall see later.

Exercise 0.4: Determine a family of matrices containing exactly one representative for
each conjugacy class of G1 = GL(2, C) and G2 = GL(2, R).

Definition 0.2: A set is called a homogeneous space iff there is a group G acting tran-
sitively on X .

Remark 0.2: In this case one has X = G · x for each x ∈ X and any two stabilizing
groups Gx and Gx
are conjugate. (Prove this as Exercise 0.5.)

Definition 0.3: For any G-set we shall denote by X /G or XG the set of G-orbits in
X and by X G the set of fixed points for G, i.e. those points x ∈ X for which one has
g · x = x for all g ∈ G.

If the set X has some structure, for instance, if X is a vector space, we will (tacidly)
additionally require that a group action preserves this structure, i.e. in this case that
x −→ g · x is a linear map for each g ∈ G (or later on, is continuous if X is a topological
space). It is a fundamental question whether the orbit space X /G inherits the same
properties as the space X may have. For instance, if X is a manifold, is this true also
for X /G? We will come back to this question later several times. Here let us only look
at the following situation: Let X be a homogeneous G-space, x0 ∈ X , and H = Gx0 the
isotropy group. Then we denote by G/H the set of left cosets gH := {gh; h ∈ H}. These
are also to be seen as H-orbits H · g where H acts on G by right translation. G/H is a
G-set, G acting by (g, g0 H) −→ gg0 H. Then we shall identify G/H and X via the map
G/H −→ X , gH −→ g · x0 .
This map is an example for the following general notion.
Definition 0.4: Let X and X  be G-sets and f : X −→ X  be a map. The map f is
called G-equivariant or a G-morphism iff one has for every g ∈ G and x ∈ X
g · f (x) = f (g · x).
Now we come back to the question raised above.
Exercise 0.6: Prove that G/H has the structure of a group iff H is not only a subgroup
but a normal subgroup, i.e. one has ghg −1 ∈ H for all g ∈ G, h ∈ H.

We mention here another useful fact: If H is a subgroup of G, one defines the normalizer
of H in G as
NG (H) := {g ∈ G; gHg −1 = H}.
It is clear that NG (H) is the maximal subgroup in G that has H as a normal sub-
group. Then the group Aut (G/H) of G-equivariant bijections of G/H is isomorphic to
NG (H)/H. (Prove this as Exercise 0.7.)
0.3 The Symmetric Group 5

Parallel to the normalizer, one defines the centralizer

CG (H) := {g ∈ G; ghg −1 = h for all h ∈ H}.

In particular for H = G, the centralizer specializes to the center

CG (G) = {g ∈ G; gh = hg for all h ∈ G} =: C(G).

As well as left actions one defines and studies right actions of a group G on a set or a
space X :

Definition 0.5: G acts from the right iff a map

G × X −→ X , (g, x) −→ x · g,

is given that satisfies the conditions

(x · g) · g  = x · (gg  ), x·e=x

for all g, g  ∈ G and x ∈ X .

Remark 0.3: Obviously it is easy to switch from right action to left action and vice
versa using the antiautomorphism g −→ g −1 of G. Namely, on any right G-space X one
can naturally define a left action by g · x := x · g −1 .

Right actions often show up in number theory and they are written there as xg := x · g
(for instance in the action of the Galois group on a number field).

0.3 The Symmetric Group

Though this text is meant to treat mainly continuous groups like the matrix groups set
up in Section 0.1, some discrete groups as, for instance, Z, SL(n, Z) (where the elements
are matrices with determinant one and integers as entries), and in particular the sym-
metric group Sn , are unavoidable.

Definition 0.6: The symmetric group Sn is the group of permutations, i.e. bijections,
of a set of n elements.

We already introduced Aut X as the group of bijections of a set X , so we can take

Xn := {1, . . . , n} and have Sn = Aut Xn . We emphasize that (with the notions from
Section 0.2) we treat permutations as left actions on the set Xn .
There are several fundamental facts concerning this group, which the reader is supposed
to know from Linear Algebra:

Remark 0.4: The number of elements of the symmetric group Sn is # Sn = n!

Remark 0.5: Each permutation σ ∈ Sn can be written as a product of transpositions,

i.e. permutations, which interchange exactly two elements.
6 0. Prologue: Some Groups and their Actions

For every σ ∈ Sn one defines the signum of σ by

σ(i) − σ(j)
sgn(σ) := (σ) := .
i−j
1≤i<j≤n

Remark 0.6: The map σ −→ (σ) is a group homomorphism of Sn to the subgroup
{1, −1} of the multiplicative group R∗ = R \ {0}. Thus, the number of transpositions in
a representation of σ as a product of transpositions is either even or odd.
The kernel of  is usually written as An . It is a normal subgroup and called the alternating
group.

Permutations are often written as products of cycles, i.e. permutations (i1 , . . . , ir ) with
ij −→ ij+1 for j < r and ir −→ i1 if r > 1 and the identity for r = 1. More precisely,
one has

Remark 0.7: Each permutation is (up to order) a unique product of disjoint cycles.

From here it is not far to the fundamental fact:

Theorem 0.1: The number of conjugacy classes of Sn is equal to the number p(n) of
partitions of n.

A partition of n is a sequence (n1 , . . . , nr ) of natural numbers ni ∈ N with ni ≥ nj for

i < j and Σni = n.

Example 0.3: The partitions of n = 3 are (1,1,1), (2,1), and (3).

The conjugacy classes of S3 are

– {id}

– the three two-element subgroups generated by transpositions, i.e.,

{(1, 2)}, {(2, 3)}, {(3, 1)},

– the two three-element subgroups generated by the 3-cycles

(1, 2, 3) resp. (1, 3, 2).

Exercise 0.8: Determine the conjugacy classes of A4 .

Sn can be realized as a matrix group: we associate to every σ ∈ Sn the matrix

A = A(σ) = (aij ) ∈ GL(n, Z) with aij = δi,σ−1 j .

A matrix like this is called a permutation matrix.

Exercise 0.9: Write down the matrices A(σ) and verify that the map given by associat-
ing σ −→ A(σ) is in fact an isomorphism from S3 to the subgroup of GL(3, R) consisting
of the permutation matrices.
Chapter 1

Basic Algebraic Concepts for

Group Representations

We first collect the basic definitions and concepts concerning representations of groups,
using only algebraic methods. As is common, we introduce the concept of a linear repre-
sentation, which is completely adequate and sufficient for the study of representations of
finite groups. In a later chapter we will enrich the discussion with elements from topology
to refine the notion of a representation to that of a continuous representation.
We start by stating the essential definitions and only afterwards illustrate these by some
examples and constructions.

1.1 Linear Representations

Let G be a group and V a K-vector space. Here, in principle, we mean K = C, but
most things go through as well for an algebraically closed field of characteristic zero.
There are interesting phenomenae in the other cases, which are beyond the scope of this
introduction. For an interested reader wanting some impression of this, we recommend
Chapter 12 ff from Serre’s book [Se], whose first Chapters together with §7-11 from Kir-
illov’s book [Ki] are guidelines for this and our next section.

Definition 1.1: π is a linear representation of G in V iff π is a homomorphism of G

into Aut V i.e. iff we have a map

π : G −→ Aut V, g −→ π(g)

with
π(gg  ) = π(g)π(g  ) for all g, g  ∈ G.

Aut V is equivalently denoted by GL(V ) and stands for the group of all (linear) isomor-
phisms of V .

In the case of a finite-dimensional vector space V , say of dimension n, one says that π is
of degree n or π is a n-dimensional representation.
8 1. Basic Algebraic Concepts

Let B = (v1 , . . . , vn ) be a basis of V . Then every F ∈ Aut V is represented with respect

to B by an invertible n × n-matrix A := MB (F ), and we have an isomorphism of vector
spaces V  Kn and of groups Aut V  GL(n, K). Essentially equivalent to the definition
above, but perhaps more down-to-earth for some readers is the definition:
An n-dimensional linear representation of a group G is a prescription π associating to
each g ∈ G a matrix π(g) = A(g) ∈ GL(n, K) such that

A(gg  ) = A(g) A(g  )

holds for all g, g  ∈ G.

As every group homorphism transforms the neutral elements of the groups into one
another, it is clear that this prescription associates the unit matrix En to the neutral
element e of the group G, and that we have π(e) = idV in the situation of the general
definition above.

If G is a matrix group, G ⊂ GL(n, C), as in Section 0.1, obviously, we have the natural
representation π0 given by
π0 (A) = A
for each A ∈ G. Another even more ubiquous representation is the trivial representation
π = id, associating the identity idV to each g ∈ G.

Using the notion of group action introduced in Section 0.2, we can also say that a lin-
ear representation of G in V is the same thing as a left action of G on the vector space V .

At a first sight, it appears to be the main problem of representation theory to determine

all representations of a given group. To make this a more sensible and accessible prob-
lem, one has to find some building blocks or elementary particles to construct general
representations from, and, moreover, decide when representations are really different.
We begin by searching for the building blocks.

Definition 1.2: Let π be a linear representation of G in V . π is called irreducible, iff

there is no genuine π-invariant subspace V0 in V .
A subspace V0 ⊂ V is π-invariant iff we have π(g)v0 ∈ V0 for all g ∈ G and v0 ∈ V0 .
If this is the case, π0 := π |V0 is a representation of G in V0 and this is called a subrep-
resentation.
So we can also say that π is irreducible, iff π has no genuine subrepresentation.

Next, we restrict our objects still more, in particular in view to the applications in physics.

We assume now that V is a unitary complex vector space, i.e. V is equipped with a scalar
product
< ., . >: V × V −→ C, (v, v  ) −→ < v, v  >
which is
– linear in the second variable and antilinear in the first,
– hermitian: for every v, v  ∈ V one has < v, v  >= < v  , v >, and
– positive definite: for every v ∈ V one has < v, v > ≥ 0 and = 0 exactly for v = 0.
1.2 Equivalent Representations 9

For V = Cn we use the standard scalar product

n
< x, y >:= x̄i yi for all x, y ∈ Cn .
i=1

Definition 1.3: A representation π of G in V is unitary iff each π(g) is unitary, i.e. iff

one has for every v, v  ∈ V and every g ∈ G

< π(g)v, π(g)v  > = < v, v  > .

1.2 Equivalent Representations

If two representations π and π  of G in K-vector spaces V resp. V  are given, it is obvious
to look for G-equivariant maps F : V −→ V  :

Definition 1.4: A K-linear map F : V −→ V  is called an intertwining operator between

π and π  iff one has for every g ∈ G

F π(g) = π  (g)F,

i.e. iff the following diagram commutes

V
F / V

π(g) π
(g)
 
V
F / V .

π and π  are called equivalent iff there is an isomorphism

F : V −→ V 

intertwining π and π  . In this case we write π ∼ π  .

Remark 1.1: The space of intertwining operators between π and π  is again a vec-
tor space. It is denoted by HomG (V, V  ) or C(V, V  ). Moreover, we use the notation
C(V ) := C(V, V ) and c(π, π  ) = c(V, V  ) = dim C(V, V  ). c(π, π  ) is also called the multi-
plicity of π in π  and denoted by mult(π, π  ).

Representations π and π  with c(π, π  ) = c(π  , π) = 0 are called disjoint.

At this point it is possible to state as our principal task:

Determine the unitary (linear) dual Ĝ of a given group G, i.e. the set of equivalence
classes of unitary irreducible representations of G.
10 1. Basic Algebraic Concepts

1.3 First Examples

We hope the reader gets the feeling that the following special cases prepare the ground
for the later more general considerations.

Example 1.1: In Section 1.1 we already mentioned that every matrix group G has its
natural representation π0 , i.e. for each real or complex matrix group G ⊂ GL(n, K) one
has the representation in V = Cn associating to every A ∈ G the matrix A itself. These
natural representations are obviously unitary for G = SO(n) or SU(n) but in general
neither unitary nor irreducible, as the following example shows:

Example 1.2: Let g be an element of the group G = SO(2). Then we usually write

α β
g = r(ϑ) := ( ), α = cos ϑ, β = sin ϑ, ϑ ∈ R.
−β α

For each k ∈ Z we put

χk (r(ϑ)) := exp(ikϑ).
Hence π = χk is a one-dimensional unitary representation in V = C, as, by the addition
theorems for the trigonometric functions, we have

χk (r(ϑ)r(ϑ )) = χk (r(ϑ + ϑ )) = exp(ik(ϑ + ϑ )) = χk (r(ϑ))χk (r(ϑ)).

√ 1 i
For U = (1/ 2)( ) we have the relation
i 1

α − iβ 0 e−iϑ 0
U r(ϑ)U −1 = ( )=( ).
0 α + iβ 0 eiϑ

This can be interpreted as follows. The matrix U is the matrix for an intertwining
operator F intertwining the natural representation π0 with the representation π1 on
V = C2 given by
e−iϑ 0
π1 (r(ϑ)) = ( ).
0 eiϑ
This representation is reducible and we see that the natural representation is equivalent
to a reducible one. Or we realize that the vectors
   

√ 1 
√ i
e1 := U e1 = (1/ 2) , e2 := U e2 = (1/ 2)
i 1

span nontrivial π0 -invariant subspaces of V = C2 . But if we look at π0 as a real repre-

sentation, i.e. in the R-vector space V0 = R2 , π0 is in fact irreducible.

Example 1.3: In Section 0.3 we introduced the symmetric group Sn . G = S3 consists

of the elements

id = (1), (1, 2) =: σ, (1, 3), (2, 3), (1, 2, 3) =: τ, (1, 3, 2).

As there are relations like σ 2 = id, τ 3 = id and

στ σ = τ 2 = (1, 3, 2), στ = (2, 3), τ σ = (1, 3),

1.3 First Examples 11

we see that each element g ∈ S3 can be written as a product of (in general) several
factors σ and τ . We indicate this here (and later on in a similar fashion in other cases)
by the notation
S3 =< σ, τ > .
We easily find one-dimensional representations in V = C, namely the trivial representa-
tion
π1 (g) = 1 for all g ∈ S3
and the signum representation

π2 (g) = sgn g ∈ {±1} for all g ∈ S3 .

One also has a natural three-dimensional representation π0 given in V = C3 by the

permutation matrices from Exercise 0.9. For instance, one has
⎛ ⎞
0 0 1
π0 (τ ) = A(τ ) = ⎝ 1 0 0 ⎠ .
0 1 0

This representation is also called the permutation representation. It can equivalently be

described as follows.
We write

3
V = C3 = ei C
w = e1 z1 + e2 z2 + e3 z3
i=1

with the canonical basis vectors e1 = t (1, 0, 0), . . . , e3 = t (0, 0, 1) and z1 , z2 , z3 ∈ C. Then
π0 is given by

π0 (g)w = eg(i) zi = ei zg−1 (i) .

i

Exercise 1.1: Verify that this is true.

π0 is unitary (verify this also) but not irreducible: V1 := (e1 + e2 + e3 )C is an invariant

subspace, and as we have

π0 (g)(e1 + e2 + e3 ) = eg(1) + eg(2) + eg(3) = e1 + e2 + e3 ,


π0 |V1 is the trivial representation π1 . As one has for w = zi ei

< e1 + e2 + e3 , w >= zi ,
  
V3 := {w, zi = 0} is the orthogonal complement to V1 and as i zi = i zg−1 (i) holds
for all g ∈ S3 , V3 is an invariant subspace.

Exercise 1.2: Show that a := e1 ζ + e2 + e3 ζ 2 and b := e1 + e2 ζ + e3 ζ 2 is a basis for

V3 , if ζ := e2πi/3 . Determine the matrices of π2 := π0 |V3 with respect to this basis and
show that π2 is irreducible.

Remark 1.2: By the general methods to be developed later, it will be clear that all ir-
reducible representations of S3 are equivalent to π1 , π2 , or π3 , i.e. the equivalence classes
of πi (i=1,2,3) constitute the unitary dual of S3 .
12 1. Basic Algebraic Concepts

Exercise 1.3: Prove this in a direct elementary way (see [FH] §1.3).

Example 1.4: A rather general procedure to construct representations is hidden in the

following simple example.

We take G ⊂ GL(2, C) and as vector space V the space Vm = C [u, v]m of homogeneous
polynomials P of degree m in two variables u, v. By counting, we see dim V = m + 1.
There are two essentially equivalent ways to define a representation of G on Vm :
a b a b
For P ∈ Vm and g = ( ) ∈ G resp. g −1 = (  ) we put
c d c d

(λm (g)P )(u, v) := P (a u + b v, c u + d v)

or
(ρm (g)P )(u, v) := P (au + cv, bu + dv).

Obviously λ(g)P and ρ(g)P are both again homogeneous polynomials of degree m. The
functional equations λ(gg  ) = λ(g)λ(g  ) and ρ(gg  ) = ρ(g)ρ(g  ) can be verified directly.
It becomes also clear from the general discussion in the following Example 1.5.
We remark that λm and ρm are irreducible unitary representations of G = SU(2), if Vm
is provided with a suitable scalar product. This will come out later in section 4.2.

Example 1.5: Let X be a G-set with a left G-action x −→ g · x and V = F(X ) a vector
space of complex functions f : X −→ C such that for f ∈ V we also have fg ∈ V where
fg is defined by
fg (x) = f (g −1 x).

Remark 1.3: The prescription

(λ(g)f )(x) := f (g −1 x)

defines a representation λ of G on V .

Proof: One has (gg  ) · x = g · g  · x and hence

−1 −1 −1
λ(gg  )f (x) = f ((gg  )−1 · x) = f (g  g · x) = f (g  · g −1 · x)

and
−1
λ(g)λ(g  )f (x) = λ(g)fg
(x) = fg
(g −1 · x) = f (g  · g −1 · x).

An important special case is given by X = Cn and G ⊂ GL(n, C) acting by matrix mul-

tiplication g · x = gx (as usual x is here a column) on V , a space of continuous functions
in n variables or, say, a subspace of homogeneous polynomials of fixed degree (for n = 2
we then have the last example).

Analogously, one treats a set X with a right G-action x −→ x · g and a vector space
V = F(X ) of complex functions f on X closed under f −→ f g where f g is defined by
f g = f (x · g).
1.3 First Examples 13

Remark 1.4: The prescription

(ρ(g)f )(x) := f (x · g)

defines a representation ρ of G on V .

Proof: One has x · (gg  ) = x · g · g  and hence

ρ(gg  )f (x) = f (x · (gg  )) = f (x · g · g  )

and


ρ(g)ρ(g  )f (x) = ρ(g)f g (x) = f (x · g · g  ).

For X = Cn and V again a space of functions in n variables, this time viewed as a row,
and G ⊂ GL(n, C), we have again the action x −→ x · g = xg by matrix multiplication.
For n = 2 this leads naturally to the second version of Example 1.4.

Another (as, later to be seen, crucial) application of this procedure is given by making G
act on itself, i.e. X = G, by left or right translation and V = L2 (G), the space of square
integrable functions with respect to a left resp. right invariant Haar measure (these no-
tions will be explained later). We then get the left (right) regular representation of G.

Example 1.6: Let G = Heis(R) be the Heisenberg group from Section 0.1, in the form
of the group of triples g = (λ, µ, κ), λ, µ, κ ∈ R with the multiplication law

gg  = (λ + λ , µ + µ , κ + κ + λµ − λ µ),

and V = L2 (R)the space of quadratic Lebesgue integrable functions f on X = R (with

norm  f 2 = ( R f (x)f (x)dx)1/2 ). Then for m ∈ R∗ , π = πm given by

πm (g)f (x) := e2πim(κ+(λ+2x)µ) f (x + λ)

is a unitary irreducible representation of Heis(R). We call it the Schrödinger represen-

tation.

Exercise 1.4: Prove that πm is a unitary representation. (The proof of the irreducibility
will become easier later when we have prepared some more tools.)

Example 1.7: Prove (as Exercise 1.5) that the Weil group of R, WR from Exercise
0.2 in Section 0.1, has the irreducible representations π = π0 and π = πm , m ∈ Z \ {0}
given by
       
0 1 0 1 z 0 z 0
π0 ( ) := det = 1, π0 ( ) := det = z z̄,
−1 0 −1 0 0 z̄ 0 z̄

and for m = 0
   m     
z 0 z 0 0 1 0 1
πm ( ) := , π ( ) := .
0 z̄ 0 z̄ m m
−1 0 (−1)m 0
14 1. Basic Algebraic Concepts

1.4 Basic Construction Principles

Proceeding from a given representation (π, V ) of a group G one can construct several
new representations using standard notions from (multi-)linear algebra. We suppose that
the reader has some knowledge of direct sums, tensor products and dual spaces. If this
is not the case, we hope that one can follow in spite of it as we stay on a very elementary
level and do enough paraphrasing to get some understanding. But sooner or later the
reader should consult any book where these notions appear, for instance take Fischer’s
“Lineare Algebra” [Fi], Lang’s “Algebra” [La], or the appendices to [Ki1] or [Be].

1.4.1 Sum of Representations

Let (π, V ) and (π  , V  ) be (linear) representations of the same group G. Then we define
the direct sum π ⊕ π  of π and π  by

(π ⊕ π  )(g)(v ⊕ v  ) := π(g)v ⊕ π  (g)v  for all v ⊕ v  ∈ V ⊕ V  .

For V = Kn , V  = Km and π(g) = A(g) ∈ GL(n, K), π  (g) = A (g) ∈ GL(m, K) in

matrix form, the sum is given by the block diagonal matrix

A(g) 0
(π ⊕ π  )(g) = ( ) ∈ GL(n + m, K).
0 A (g)

1.4.2 Tensor Product of Representations

Let (π, V ) and (π  , V  ) be as above and V ⊗ V  the tensor product of V and V  . Then
we define the tensor product π ⊗ π  of π and π  by

(π ⊗ π  )(g)(v ⊗ v  ) := π(g)v ⊗ π  (g)v  for all v ⊗ v  ∈ V ⊗ V  .

For V = Kn , V  = Km and π(g) = A(g) ∈ GL(n, K), π  (g) = A (g) ∈ GL(m, K), the
tensor product is given by the Kronecker product of the matrices A(g) and A (g)
⎛ ⎞
a1,1 A (g) . . . a1,n A (g)
(π ⊗ π  )(g) = ⎝ ... ⎠ ∈ GL(nm, K) .
 
an,1 A (g) . . . an,n A (g)

We will not go into the definition of the tensor product (as to be found in the afore
mentioned sources) but simply recall the following. If V has a basis (ei )i∈I and V  a
basis (fj )j∈J , then V ⊗ V  has a basis (ei ⊗ fj )(i,j)∈I×J .
In V ⊗ V  we have the rules

(v1 + v2 ) ⊗ v  = v1 ⊗ v  + v2 ⊗ v  ,
v⊗ (v1
+ v2 )
= v ⊗ v1 + v ⊗ v2 ,
λ(v ⊗ v ) = (λv) ⊗ v  = v ⊗ (λv  ) for all v, v1 , v2 ∈ V, v  , v1 , v2 ∈ V  , λ ∈ K.


It is clear that one can also define tensor products of more than two factors, and in par-
ticular tensor powers like ⊗3 V = V ⊗ V ⊗ V (It is not difficult to see that these products
are associative.)
1.4 Basic Construction Principles 15

Moreover, there are subspaces of tensor powers with a preassigned symmetry property.
In particular we are interested in p-fold symmetric tensors S p V and p-fold antisymmetric
tensors ∧p V .
For instance, if V is three-dimensional with basis (e1 , e2 , e3 ),

• ⊗2 V has dimension 9 and a basis

(e1 ⊗ e1 , e1 ⊗ e2 , e1 ⊗ e3 , e2 ⊗ e1 , . . . , e3 ⊗ e3 ),

• S 2 V has dimension 6 and a basis

e1 ⊗ e1 , e1 ⊗ e2 + e2 ⊗ e1 , e1 ⊗ e3 , e2 ⊗ e2 , e2 ⊗ e3 + e3 ⊗ e2 , e3 ⊗ e3 .

• ∧2 V has dimension 3 and a basis

e1 ∧ e2 := e1 ⊗ e2 − e2 ⊗ e1 , e1 ∧ e3 := e1 ⊗ e3 − e3 ⊗ e1 , e2 ∧ e3 := e2 ⊗ e3 − e3 ⊗ e2 .

More generally, one has symmetric and antisymmetric subspaces in ⊗p V with bases

ei1 · · · · · eip := eig(1) ⊗ · · · ⊗ eig(p) , i1 ≤ · · · ≤ ip ,

g∈Sp

resp.

ei1 ∧ · · · ∧ eip := sgng eig(1) ⊗ · · · ⊗ eig(p) , i1 < · · · < ip .

g∈Sp

For instance, if V is three-dimensional, S p V can be identified with the space K [u, v, w]p
of homogeneous polynomials of degree p in three variables. And if π is a representation
of G on V, the prescription ei −→ π(g)ei defines by linear continuation representations
S p π and ∧p π on S p V resp. ∧p V .
It is one of the most important construction principles for finite-dimensional representa-
tions to take the natural representation π0 and to get (other) irreducible representations
by taking tensor products and reduce these to sums of irreducibles. This will be analysed
later.

1.4.3 The Contragredient Representation

We use the notation V ∗ to denote the dual space of a given K-vector space. This space
consists of K-linear maps ϕ : V −→ K, is again a K-vector space (of the same dimension
as V if the dimension is finite), and is written as

V ∗ = Hom(V, K) = {ϕ : V −→ K, ϕ K − linear }.

We shall also use the notation

ϕ(v) =: < ϕ, v > for all ϕ ∈ V ∗ , v ∈ V

and, if dim V = n with basis (e1 , . . . , en ), we denote (e∗1 , . . . , e∗n ) for the dual basis for
V ∗ , i.e. with < e∗i , ej > = δi,j (= 1 for i = j and = 0 for i = j).
16 1. Basic Algebraic Concepts

If π is a representation of G on V , we define the contragredient representation π ∗ on V ∗

by
(π ∗ (g)ϕ)(v) := ϕ(π(g −1 )v) for all ϕ ∈ V ∗ , v ∈ V,

or equivalently by the condition

< π(g)∗ ϕ, π(g)v >=< ϕ, v > for all ϕ ∈ V ∗ , v ∈ V.

The proof that we really get a representation is already contained in the example 1.5 in
section 1.3.
It is clear from linear algebra that in matrix form we have A∗ (g) = t A(g −1 ).

Exercise 1.6: Verify this.

1.4.4 The Factor Representation

If (π1 , V1 ) is a subrepresentation of the representation (π, V ) of G, one has also an
associated representation on the factor space V /V1 , called the factor representation.

There is still another essential construction principle, namely the concept of induced
representations. It goes back to Frobenius and is also helpful in our present algebraic
context. But we leave it aside here and postpone it for a more general treatment in
Chapter 7.

1.5 Decompositions
We now have at hand procedures to construct representations and we have the notion of
irreducible representations which we want to look at as the building blocks for general
representations. To be able to do this, we still have to clarify some notions.
We defined in Section 1.1 that (π, V ) is irreducible, iff there is no nontrivial subrepre-
sentation. Now we add:

Definition 1.5: (π, V ) is decomposable, iff there exists an invariant subspace V1 ⊂ V

which admits an invariant complementary subspace V2 , i.e. with V = V1 ⊕ V2 . We then
write π = π1 + π2 .
Representations for which every nontrivial subrepresentation admits an invariant com-
plement are called completely reducible.

Complete reducibility is desirable as it leads (at least in the case of finite-dimensional

representations) immediately to the possibility to decompose a given representation into
irreducible representations. Therefore facts like the following are very important.

Theorem 1.1: Let (π, V ) be the representation of a finite group G and (π1 , V1 ) a sub-
representation. Then there exists an invariant complementary subspace V2 .
1.5 Decompositions 17

Proof: Let <, > be any scalar product for V (such a thing exists as is obvious at least
for finite-dimensional V  Kn ) . We can find a G-invariant scalar product <, > , namely
the one defined by

< v, v  >:= < π(g)v, π(g)v  > for all v, v  ∈ V.

g∈G

Then V2 := {v ∈ V, < v, v1 > = 0 for all v1 ∈ V1 } is a complementary subspace to V1

which is π-invariant:
For v ∈ V2 , g ∈ G one has, as <, > is in particular π(g −1 )-invariant,
< π(g)v, v1 > = < π(g −1 )π(g)v, π(g −1 )v1 > = < v, π(g −1 )v1 > = 0
if v1 ∈ V1 and hence also π(g −1 )v ∈ V2 .

Remark 1.5: If G is any group and (π, V ) a unitary representation of G, again every
subrepresentation (π1 , V1 ) has an invariant complement (π2 , V2 ) given as in the proof
above by V2 := V1⊥ = {v ∈ V, < v, v1 > = 0 for all v1 ∈ V1 } because V is already
provided with a π-invariant scalar product.

Remark 1.6: If G is compact, the theorem holds as well, as the sum in the definition
of the invariant scalar product can be replaced by an integral. This will be done later in
Chapter 4.

Remark 1.7: As an immediate consequence we have that finite-dimensional unitary

representations of an arbitrary group are completely reducible.

Remark 1.8: Every finite-dimensional representation of a finite (or compact) group is

unitarizable as the representation space can be given a π-invariant scalar product by the
Theorem 1.1 above.

Example 1.7: We have already seen an example for a decomposition of a representation

in Example 1.3 in Section 1.3, namely the permutation representation of G = S3 . The
standard example that not every representation is completely reducible and not even
decomposable is the following.
Let G = R be the additive group of the real numbers and π a two-dimensional represen-
tation of G in V = C2 given by
1 b
R
b −→ ( ) =: A(b).
0 1
Here V1 := e1 C is an invariant subspace: we have A(b)e1 = e1 for all b ∈ R. If
 
x
V2 := v C = C
y
is an invariant complementary subspace, we have
   
x + by x
A(b)v = =λ
y y
for a certain λ ∈ C, i.e. x + by + λx and y = λy. As y must be nonzero, we have λ = 1.
This leads to by = 0 for all b and hence a contradiction.
18 1. Basic Algebraic Concepts

The following considerations are important in a general representation theory. But as

we will not use them here so much they are only mentioned (for proofs see for instance
[Ki] p.115 f).
Every representation (π, V ) is either indecomposable or the sum of two representations
π = π1 + π2 . The same alternative applies to the representations π1 and π2 and so on. In
the general case this process can continue infinitely. We say that π is finite, if every family
of π-invariant subspaces Vi of V that is strictly monotone with respect to inclusion is
finite. In this case there exists a strictly monotone finite collection of invariant subspaces

V = V0 ⊃ V1 ⊃ · · · ⊃ Vn = {0}

such that that the representations πi appearing on Vi /Vi+1 are irreducible. One has the
Jordan - Hölder Theorem: The length of such a chain of subspaces and the equivalence
classes of the representations πi (up to order) are uniquely defined by the equivalence
class of π.

If we want to discuss the question whether the decomposition of a given representation

(π, V ) is the direct sum of irreducibles, we need criteria for irreducibility. A central cri-
terium is a consequence of the following famous statement.

Theorem 1.2: (Schur’s Lemma): If two linear representations (π, V ) and (π  , V  ) are
irreducible, then every intertwining operator F ∈ C(π, π  ) is either zero or invertible. If
V = V  , π = π  and dim V = n, then F is a homothety, i.e. F = λ id, λ ∈ C.

Proof: i) As defined in Section 1.2 an intertwining operator for π and π  is a linear map
F : V −→ V  with

(1.1) π  (g)F (v) = F (π(g)v) for all g ∈ G, v ∈ V.

Then Ker F = {v ∈ V, F (v) = 0} is a π-invariant subspace of V because for v ∈ Ker F

we have F (v) = 0 and by (1.1) F (π(g)v) = π  (g)F (v) = 0, i.e. π(g)v ∈ Ker F . By a
similar reasoning it is clear that Im F = {F (v), v ∈ V } is an invariant subspace of V  .
The irreducibility of π implies Ker F = {0} or Ker F = V and by the irreducibility of
π  we have Im F = {0} or = V  . Hence F is the zero map or an isomorphism.
ii) If V = V  , the endomorphism F : V −→ V has an eigenvalue λ ∈ C (as the charac-
teristical polynomial PF (t) := det(F − tE) ∈ C[t] has a zero λ ∈ C by the fundamental
theorem of algebra). Hence F  := F − λE is a linear map with Ker F  = {0}. From part
i) we then deduce F  = 0, i.e. F = λE.

Remark 1.9: The first part of the Theorem is also expressed as follows: Two irreducible
representations of the same group are either equivalent or disjoint.
The second part can be written as C(π) = C.

There are several more refined versions of Schur’s Lemma which come up if one sharpens
the notion of a linear representation by adding a continuity requirement as we shall do
later on. For the moment, though it is somewhat redundant, we add another version
which we take over directly from Wigner ([Wi] p. 75) because it contains an irreducibil-
ity criterium used in many physics papers and its proof is a very instructive example for
matrix calculus.
1.5 Decompositions 19

Theorem 1.3 (the finite-dimensional unitary Schur): Let (π, Cn ) be a unitary matrix
representation of a group G, i.e. with π(g) = A(g) ∈ U(n). Let M ∈ GL(n, Cn ) be a
matrix commuting with all A(g), i.e.

(1.2) M A(g) = A(g)M for all g ∈ G.

Then M is a scalar multiple of the unit matrix, M = λEn , λ ∈ C.

Proof: i) First, we show that we may assume M is a Hermitian matrix: We take the
adjoint (with A −→ A∗ := tĀ) of the commutation relation (1.2) and get

A(g)∗ M ∗ = M ∗ A(g)∗ ,

multiply this from both sides by A(g) and, using the unitarity A(g)A(g)∗ = En , get

M ∗ A(g) = A(g)M ∗ .

Hence, if M commutes with all A(g), not only M ∗ does but also the matrices

H1 := M + M ∗ , H2 := i(M − M ∗ ),

which are Hermitian. It is therefore sufficient to show that every Hermitian matrix,
which commutes with all A(g), is a scalar matrix, since if H1 and H2 are multiples of
En , so must be 2M = H1 − iH2 .
ii) If M is Hermitian, one of the fundamental theorems from linear algebra (see for
instance [Ko], p.256 or [Fi] Kapitel 4) tells that M can be diagonalized, i.e. conjugated
into a diagonal matrix D by a matrix U ∈ U(n)

U M U −1 = D.

Then we put Ã(g) := U A(g)U −1 . Ã(g) is unitary as can be easily verified. From the
commutation relation (1.2), we have DÃ(g) = Ã(g)D. With D = (di,j ), di,j = 0 for
i = j, Ã(g) = (ãi,j ), this leads to

di,i ãi,j = ãi,j dj,j .

Hence for di,i = dj,j we have ãi,j = ãj,i = 0, that is, all Ã(g) must have zeros at all
intersections of rows and columns, where the diagonal elements are different. This would
mean that g −→ Ã(g) resp. g −→ A(g) are reducible representations. Hence, as this
is not the case, we have that all diagonal elements are equal and D is a scalar matrix
λEn , λ ∈ C.
The proof shows that we can sharpen the statement from the Theorem to the irreducibiliy
criterium:

Corollary: If there exists a nonscalar matrix which commutes with all matrices of a
finite-dimensional unitary representation π, then the representation is reducible. If there
exists none, it is irreducible.

This Corollary will help to prove the irreducibility in many cases. Here we show the use-
fulness of Schur’s Lemma by proving another fundamental fact (later we shall generalize
this to statements valid also in the infinite-dimensional cases):
20 1. Basic Algebraic Concepts

Theorem 1.4: Any finite-dimensional irreducible representation of an abelian group is

one-dimensional.

Proof: Let (π, V ) be a representation of G. Then

F : V −→ V, v −→ π(g0 )v

is an intertwining operator for each g0 ∈ G, as we have

π(g)F (v) = π(g)π(g0 )v

= π(gg0 )v, as π is a representation,
= π(g0 g)v, as G is abelian,
= π(g0 )π(g)v,
= F (π(g)v).

The second part of Schur’s Lemma says that F is a homothety, i.e. we have a λ ∈ C with

F (v) = π(g0 )v = λv for all v ∈ V.

As g0 ∈ G was chosen arbitrarily, we see that V0 = vC is an invariant subspace of V .

For an irreducible representation π this implies V = V0 , i.e. V is one-dimensional.

To finish our statements on decompositions in this section, we state here without proofs
(for these see [Ki] p.121ff) some standard facts about the kind of uniqueness one can
expect for a decomposition of a completely reducible representation. By the way, this
theorem is a reflex of a central theorem from Algebra about the structure of modules
over a ring (a representation (π, V ) will be seen also as module over the group ring C[G]).

Reminding the notions of disjoint representations from Section 1.2 and of finite repre-
sentations from above, we add the following notion.

Definition 1.6: A representation π is said to be primary, iff it cannot be represented

as a sum of two disjoint representations.

As an example one may think here of a representation of the form π k , k ∈ N, the sum
of k representations (π, V ) given on the k-th sum V k = V ⊕ · · · ⊕ V .
The central fact can be stated like this:

Theorem 1.5: Let π be a completely reducible finite representation in the space V .

Then there exists a unique decomposition of V into a sum of invariant subspaces Wj , for
which the representations πj = π |Wj are primary and pairwise disjoint. Every invariant
subspace V  ⊂ V has the property V  = ⊕j (V ∩ Wj ).

Among other things, the proof uses the notion of projection operators, known from linear
algebra. As we have to use these here later on, we recall the following concept.
1.6 Characters 21

Definition 1.7: Suppose that the space V of a representation π is decomposed into a

direct sum of invariant subspaces Vi , i.e., V = ⊕m
i=1 Vi . Then the projection operator Pi
onto Vi parallel to the remaining Vj is defined by Pi v = vi if v = (v1 , . . . , vi , . . . , vm ),
vi ∈ Vi , i = 1, . . . , m.

Pi has the following properties

- a) Pi2 = Pi ,
- Pi Pj = Pj Pi = 0 for i = j,
b) 
m
- c) i=1 Pi = 1,
- d) Pi ∈ C(π).
One can prove that every collection of operators (i.e. linear maps) in V that enjoys these
properties produces a decomposition of V into a sum of invariant subpaces Vi = Pi V .

Later on, we shall treat explicit examples of decompositions.

1.6 Characters of Finite-dimensional Representations

The following notions need the finite-dimensionality of a representation. For the infinite-
dimensional cases there are generalizations using distributions and/or infinitesimal meth-
ods which we only touch briefly later on because they are beyond the scope of this book.
So in this Section (π, V ) is always a finite-dimensional complex representation of the
group G with dim V = n.

Definition 1.8: We call character of π the complex function χπ (or equivalently χV ) on

G given by
χπ (g) := Tr π(g) for all g ∈ G.

As usual Tr denotes the trace. If π is given by the matrices A(g) = (aij (g)) ∈ GL(n, C),
one has

n
χπ (g) = aii .
i=1

Conjugate matrices A ∼ A , i.e. with A = T AT −1 , T ∈ GL(n, C) have the same trace.

 

Therefore the trace of a representation is well defined independently of the choice of the
matrix representation. And for conjugate g, g  ∈ G we have
χπ (g) = χπ (g  ).
This is also expressed by saying the character is a class function.
Obviously one has for the neutral element e of G
χπ (e) = n = dim V.
We assume moreover that π is unitary (we shall see soon that this is not really a restric-
tion if the group is finite or compact). Then π(g) is in matrix form A(g) conjugate to a
diagonal matrix D = D(λ1 , . . . , λn ), where λ1 , . . . , λn are the eigenvalues of A(g). Hence:
n
Remark 1.10: χπ (g) = i=1 λi .
22 1. Basic Algebraic Concepts

If G is finite, each element g has finite order and so has π(g). Thus the eigenvalues λi
must have | λi |= 1 and we have λ−1
i = λi .

Using this, from Remark 1.10 we deduce an important fact.

Remark 1.11: χπ (g −1 ) = χπ (g) .

Finally we state two very useful formulae.

Theorem 1.6: Let (π, V ) and (π  , V ) be two finite-dimensional representations of a

group G. Then we have

(1.3) χπ⊕π
= χπ + χπ

(1.4) χπ⊗π
= χπ × χπ
.

Proof: The first equation is simply the expression of the fact that the trace of a block
A
matrix ( ) is the sum of the traces of the two matrices A and A . The second
A
relation can be seen by inspection of the trace of the Kronecker product in 1.4.2.

Exercise 1.7: Let (π, V ) be a finite-dimensional representation of G. Show that for

every g ∈ G one has the equations

χ∧2 π (g) = (1/2)(χπ (g)2 − χπ (g 2 )),

χS2 π (g) = (1/2)(χπ (g)2 + χπ (g 2 )).

As we shall find out in the next chapters, the irreducible representations of finite and
compact groups (where in the last case we have to add a continuity condition) are all
finite-dimensional. Then the characters of these representations already contain enough
information to fix these representations in a way which we will discuss more precisely now.
Chapter 2

Representations of Finite
Groups

As we are mainly interested in continuous groups like SO(3) etc., this topic plays no cen-
tral rôle in our text. But here we can learn a lot about how representation theory works.
Thus we present the basic facts. We will not do it by using the abstract notions offered
by modern algebra (as for instance in [FH]), where eventually the whole theory can be
put into the nutshell of one page. Instead we follow very closely the classic presentation
from the first pages of Serre’s book [Se].

In this chapter G is always a finite group of order #G = m.

2.1 Characters as Orthonormal Systems

We introduce the group ring (or group algebra): Let K be a field and G a group with
#G = m. Then we denote the space of maps from G to K by

KG := {u : G −→ K}.
KG is a K-vector space as we have an addition u + u defined by
(u + u )(g) := u(g) + u (g) for all g ∈ G
and scalar multiplication λu by (λu)(g) := λu(g) for u ∈ KG and λ ∈ K. But we also
have a multiplication uu defined by

(uu )(g) := u(a)u (b).

a,b∈G,ab=g

Remark 2.1: One can show that KG is an associative K-algebra (for this notion see
Section 6.1).

There is an alternative way to describe KG, namely as a set of formal sums g∈G u(g)g
built from the values u(g) of u in g ∈ G, i.e.

KG := { u(g)g, u(g) ∈ K}.

g∈G
24 2. Representations of Finite Groups

Here we have addition

u(g)g + u (g)g := (u(g) + u (g))g

and multiplication given by

u(g)g · u (g)g := v(g)g, with v(g) := u(a)u (b).

ab=g

As we restrict our interest in this text to complex representations, in the sequel we write
H = CG and H0 = Ccl G for the algebra of class functions, i.e. with u(g) = u(g  ) for
g ∼ g  . If (π, V ) is a representation of G, by Remark 1.8 in 1.5 we may assume that it is
unitary. Conjugate elements of G have the same character. Hence the character u = χπ
is not only an element in H but also in H0 .
Our first aim is to prove that, for a complete system π1 , . . . , πh of irreducible represen-
tations, the characters χπ1 , . . . , χπh are an orthonormal system (=: ON-system) in H.
Here the scalar product <, > is given by

< u, v >:= (1/m) u(t)v(t) for all u, v ∈ H.

t∈G

(It is clear that this is a scalar product.)

We will also use the following bilinear form

(u, v) := (1/m) u(t−1 )v(t).

t∈G

If u = χπ is a character, by the Remark 1.11 in 1.6 we have u(t) = u(t−1 ) and hence
(χπ , v) = < χπ , v > for v ∈ H.

Theorem 2.1: Let π and π  be irreducible representations of G and χ = χπ as well as

χ = χπ
their characters. Then we have

< χ, χ > = 0, if π ∼ π  ,
< χ, χ > = 1.

This is also written as < χπ , χπ > = δπ,π
.

The proof will be a consequence of two propositions.

Proposition 2.1: Let π, π  be irreducible, F : V −→ V  a linear map, and

F 0 := (1/m) π  (g −1 )F π(g).
g∈G

Then one has

1) F 0 = 0, if π ∼ π  (“case 1”),
2) F 0 = λ id, if V = V  , π = π  , where λ = (1/ dim V )Tr F (“case 2”).
2.1 Characters as Orthonormal Systems 25

Proof: F 0 is an intertwining operator for π and π  : We have

π  (g)−1 F 0 π(g) = (1/m) π  (g)−1 π  (t)−1 F π(t)π(g)

t∈G

= (1/m) π  (tg)−1 F π(tg) = F 0 .

t∈G

By application of Schur’s Lemma, we have F 0 = 0 or F 0 = λ id in the respective cases,

and because of

Tr F 0 = (1/m) Tr(π(t−1 )F π(t)) = Tr F

t∈G

and Tr id = n = dim V, we get λ = (1/n)Tr F .

Proposition 2.2: For π and π  as in Proposition 2.1 given in matrix form by π(g) = A(g)
and π  (g) = B(g), we have

(1/m) Bij (t−1 )Akl (t) = 0 for all i, j, k, l in case 1 ,

t∈G

(1/m) Aij (t−1 )Akl (t) = (1/n)δil δjk for all i, j, k, l in case 2 .
t∈G

Proof: With F = (Fij ) Proposition 2.1 expressed in matrix form

(1/m) Bij (t−1 )Fjk Akl (t) = Fil0

t∈G jk

states

(1/m) Bij (t−1 )Fjk Akl (t) = 0 for all i, l in case 1,

t∈G jk

and this requires that all the coefficients of all Fjk are zero. Hence, we have the first
relation. In the second case we have

(1/m) Aij (t−1 )Fjk Akl (t) = λδil = (1/n) Fjk δjk δil .
t∈G jk jk

Comparing the coefficients of Fjk on both sides leads to the second relation.
 
Proof of the Theorem: We have χ (t) = Bii (t) and χ(t) = Aii (t) and hence

< χ, χ > = (1/m) χ(t−1 )χ(t) = (1/m) Aii (t−1 )Ajj (t)
t∈G t∈G ij

= (1/n) δij δij = 1

ij

by the second relation in Proposition 2.2, and

< χ, χ >= (1/m) χ (t−1 )χ(t) = (1/m) Bii (t−1 )Ajj (t)
t∈G t∈G ij

by the first relation in Proposition 2.2.

26 2. Representations of Finite Groups

Corollary 1: Let (π, V ) be a finite-dimensional representation of G with character

χ := χπ and (πi , Vi ) irreducible representations with characters χi := χπi and

V = V1 ⊕ · · · ⊕ Vk .

Let moreover be (π  , V  ) an irreducible representation with character χ = χπ
. Then we

have

(2.1) # {Vi , Vi  V  } = < χ , χ > .

From 1.2 we know that this number is also recognized as mult(π  , π).

Proof: By Theorem 1.6 in 1.6 we have χ = i χi . Theorem 2.1 in 2.1 says that the
summands in

< χ , χ >= < χ , χi >

i

are either 0 or 1 if χ ∼ χi or χ ∼ χi . Hence < χ , χ > counts the number of irreducible

components πi contained in π which are equivalent to π  .

Corollary 2: mult(π  , π) is independent of the decomposition as in Corollary 1.

Proof: < χ , χ > does not depend on the decomposition.

Corollary 3: If two representations π and π  have the same character χπ = χπ
, they
are equivalent.

Proof: By Corollary 1 the multiplicities are equal for all irreducible components of π
and π  and so both representations are equivalent.

Corollary 4: If π decomposes into the irreducible representations π1 , . . . , πh with re-

spective multiplicities mi , i.e. one has

V = m1 V1 ⊕ · · · ⊕ mh Vh ,

or, as we also write

V = V1m1 ⊕ · · · ⊕ Vhmh ,

one has

h
< χ, χ > = m2i .
i=1

Proof: This follows immediately from Theorem 2.1 in 2.1, as χ = mi χi and the χi
are an ON-system.

This Corollary specializes to the following important irreducibility criterium.

Corollary 5: π is irreducible exactly if < χ, χ > = 1.

2.2 The Regular Representation 27

2.2 Regular Representation and its Decomposition

The regular representation λ of the finite group G with #G = m is defined on the vector
space V spanned by a basis et indexed by the elements t of G, i.e. V =< et >t∈G by the
prescription
λ(g)et := egt ,
 
i.e. we have λ(g)v = t∈G z(g−1 t) et for v = t∈G zt et , zt ∈ C.
V can be identified with H concerning the vector space structure.
Then we have

λ(g)u = u(g −1 t)t for all u = u(t)t ∈ H.

t∈G t∈G

If g = e, we have gt = t for all t ∈ G, which shows that the diagonal elements of λ(g) are
zero. In particular we have Tr λ(g) = 0 and Tr λ(e) = Tr Em = m. So we have proven:
Proposition 2.3: The character χλ of the regular representation λ of G is given by

χλ (e) = #G = m,
χλ (g) = 0 for all g = e.

Proposition 2.4: Every irreducible representation πi is contained in the regular repre-

sentation with multiplicity equal to its dimension ni , i.e. mult(πi , λ) = ni = dim Vi .

Proof: According to Corollary 1 in 2.1 we have

mult(πi , λ) = < χπi , χλ >

and by the definition of the scalar product also

mult(πi , λ) = (1/m) χλ (t−1 )χπi (t).

t∈G

By Proposition 2.3 and the general fact that the character of the neutral element equals
the dimension, we finally get

mult(πi , λ) = (1/m) · m · χπi (e) = ni .

Proposition 2.5: Let π1 , . . . , πh be a complete system representing the equivalence

classes of irreducible representations of G. Then the dimensions ni = dim Vi satisfy the
relations h 2
a) i=1 ni = m,
and, if t ∈ G, t =
e,
h
b) i=1 ni χπi (t) = 0.

Proof: By Proposition 2.4 we have χλ (t) = Σni χi (t) for all t ∈ G. Taking t = e we
obtain a) and for t = e we obtain b).

This result is very useful for the determination of the irreducible representations of a finite
group: suppose one has constructed some mutually nonequivalent irreducible representa-
tions of dimensions n1 , . . . , nh . In order that they be all irreducible representations (up
to equivalence) it is necessary and sufficient that one has n21 + · · · + n2h = m.
28 2. Representations of Finite Groups

In 1.3 we already discussed the example G = S3 . Here we have m = 6. And we

singled out the two one-dimensional representations π1 = 1 and π2 = sgn and the two-
dimensional representation π3 . As we have here 12 + 12 + 22 = 6 = m, we see that we got
all irreducible representations. Moreover, we have a simple way to check the irreducibil-
ity of π3 by using our Corollary 5. in 2.1
As an easy Exercise 2.1, we recommend to do this and to determine the characters of
all the representations π1 , π2 , π3 , and π0 introduced there. One can see that that one has
π0 = π1 + π3 consistent with our Corollary 4 in 2.1.

One can prove more facts in this context, for instance the dimensions ni all divide the
order m of G (see [Se] p.53]).

2.3 Characters as Orthonormal Bases and Number of

Irreducible Representations
We sharpen Theorem 2.1 from 2.1 to the central fact that the characters χ1 , . . . , χh be-
longing to a complete set of equivalence classes of irreducible representations of G form
an orthonormal basis of the space of class functions H0 . We recall that a function f
defined on G is a class function if one has f (g) = f (tgt−1 ) for all g, t ∈ G.

Proposition 2.6: Let f be a class function on G and (π, V ) a representation of G. Let

πf be the endomorphism of V defined by

πf := f (t)π(t).
t∈G

If π is irreducible with dim V = n and character χ, then πf is a scalar multiple of the

identity, i.e. πf = λ idV , with

λ = (1/n) χ(t)f (t) = (m/n) < χ̄, f > .

t∈G

Proof: πf is an intertwining operator for π: We have

π(s)−1 πf π(s) = f (t)π(s)−1 π(t)π(s) = f (t)π(s−1 ts).

t∈G t∈G

As f is a class function, we get by t −→ s−1 ts = t

π(s)−1 πf π(s) = f (t )π(t ) = πf .

t
∈G

By the second part of Schur’s Lemma, πf = λid is a scalar multiple of the identity. We
have Tr λ id = nλ and
Tr πf = Σf (t)Tr π(t) = Σf (t)χ(t).
Hence
λ = (1/n)Σf (t)χ(t) = (m/n) < χ̄, f > .
As the characters are class functions, they are elements of H0 .
2.3 Characters as Orthonormal Bases 29

Theorem 2.2: The characters χ1 , . . . , χh form an ON-basis of H0 .

Proof: Theorem 2.1 from Section 2.1 says that the (χi ) form an ON-system in H0 . We
have to prove that they generate H0 . That is, we have to show that every f ∈ H0 ,
orthogonal to all χ̄i , is zero. Let f be such an element. For each representation π of G,
we put πf = Σf (t)π(t) as above. Proposition 2.6 shows that πf is zero if π is irreducible.
But as each π may be decomposed into irreducible representations, πf is always zero. We
apply this to the regular representation , i.e. to π = λ, and compute for the first basis
vector of the representation space for λ

πf e1 = f (t)λ(t)e1 = f (t)et .
t t

Since πf is zero, we have πf e1 = 0 and this requires f (t) = 0 for all t ∈ G. Hence f is zero.

As each element of H0 is fixed by associating a complex number to each conjugacy class

g∼ of G, the dimension of H0 as a C-vector space equals the number of conjugacy classes.
Hence we have:

Corollary 1: The number of irreducible representations of G (up to equivalence) is

equal to the number of conjugacy classes.

Corollary 2: Let g be an element of G and c(g) := #{g  ; g  ∼ g} the number of elements

in the conjugacy class of g. Then we have
h
a) i=1 χi (g)χi (g  ) = m/c(g) if g  = g,
b) = 0 if g  ∼ g.
Proof: Let fg be the characteristic function of the class of g, i.e. fg (g  ) = 1 if g  ∼ g
and = 0 else. Since we have fg ∈ H0 , by the Theorem, we can write

h
fg = λi χi with λi = < χi , fg > = (c(g)/m)χi (g).
i=1

For each t ∈ G we then have

h
fg (t) = (c(g)/m) χi (g)χi (t).
i=1

And because of fg (t) = 1 for t ∼ g and = 0 for t ∼ g, we get a) and b).

Remark 2.2: We recall the statement from 0.3 that the number of conjugacy classes of
the symmetric group Sn equals the number p(n) of partitions of n.
It is a nice excercise to redo the example of G = S3 in the light of the general facts now
at hand: We have p(3) = 3 and thus know of three irreducible representations π1 , π2 , π3
with respective characters χ1 , χ2 , χ3 . Moreover we know that G = S3 can be generated
by the transposition σ = (1, 2) and the cycle τ = (1, 2, 3). The three conjugacy classes
are represented by e = id, σ, and τ . We have

σ 2 = e, τ 3 = e, στ = τ 2 σ.
30 2. Representations of Finite Groups

Hence, for each representation π the equation π(σ)2 = id holds. So there are two one-
dimensional representations, namely π1 = 1 and π2 = sgn with

χ1 (e) = 1, χ1 (σ) = 1, χ1 (τ ) = 1,
χ2 (e) = 1, χ2 (σ) = −1, χ2 (τ ) = 1.

The third representation π3 must have dim π3 = n with 12 + 12 + n2 = 6, i.e. n = 2, as

we already concluded at the end of 2.2. The value of χ3 can be deduced (Proposition 2.4
in 2.2) from the relation with the character χλ of the regular representation λ

χ1 + χ2 + 2χ3 = χλ .

As we have χλ (e) = 6 and χλ (g) = 0 for g = e, we conclude

χ3 (e) = 2, χ3 (σ) = 0, χ3 (τ ) = −1.

These formulae should also be read off from the results of Exercise 2.1.

There is a lot more to be said about the representations of finite groups in general
and of the symmetric group in particular. Here, we only mention that the irreducible
representations of Sn are classified by the Young tableaus realizing a permutation of n.
And the following theorem about the canonical decomposition of a representation of a
finite group, which is a special case of our Theorem 1.5 in 1.5 and can be proved rather
easily (see [Se] p. 21):
Let π1 , . . . , πh be as before irreducible representations representing each exactly one
equivalence class and let (π, V ) be any representation. Let V = U1 ⊕ · · · ⊕ Uk be a
direct sum decomposition of V into spaces belonging to irreducible representations. For
i = 1, . . . , h denote by Wi the direct sum of those of the U1 , . . . , Uk which are isomorphic
to Vi (the Wi belong to the primary representations introduced at the end of 1.5). Then
we have
V = W 1 ⊕ · · · ⊕ Wh ,
which is called the canonical decomposition. Its properties are as follows:

Theorem 2.3: i) The decomposition V = W1 ⊕· · ·⊕Wh does not depend on the initially
chosen decomposition of V into irreducible subspaces.
ii) The projection Pi of V onto Wi associated to the decomposition is given by

Pi = (ni /m) χi (t)π(t), ni = dim Vi .

t∈G

We stop here our treatment of finite groups and propose to the reader to treat the fol-
lowing examples.

Exercise 2.2: Determine the irreducible representations of the cyclic group Cn and of
the dihedral group Dn (This is the group of rotations and reflections of the plane which
preserve a regular polygon with n vertices, it is generated by a rotation r and a reflection
s fulfilling the relations rn = e, s2 = e, srs = r−1 ).

Exercise 2.3: Determine the characters of S4 .

Chapter 3

Continuous Representations

Among the topological groups, compact groups are the most easy ones to handle. Since
we would like to treat their representations next, it is quite natural that the appearance
of topology leads to an additional requirement for an appropriate definition, namely a
continuity requirement. We will describe this first and indicate the necessary changes
for the general concepts from Sections 1.1 to 1.5 (to be used in the whole text later). In
the next chapter we specialize to compact groups. We will find that their representation
theory has a lot in common with that of the finite groups, in particular the fact that all
irreducible representations are finite-dimensional and contained in the regular represen-
tation (the famous Theorem of Peter and Weyl). But there is an important difference:
the number of equivalence classes of irreducible representations may be infinite.

We do not prove all the modifications in the general theorems when linear representa-
tions are specialized to continuos representations, but concentrate in the next chapters
on an explicit description of the representation theory for SU(2) resp. SO(3) and the
other examples mentioned in the introduction.

3.1 Topological and Linear Groups

We follow [Ki] p.22:
Definition 3.1: A topological group is a set that is simultaneously a group and a topo-
logical space in which the group and topological structures are connected by the following
condition:

The map
G × G −→ G, (a, b) −→ ab−1
is continuous.

Remark 3.1: This requirement is equivalent to the following three conditions which are
more convenient for checking

1.) (a, b) −→ ab is continuous in a and in b,

2.) a −→ a−1 is continuous at the point a = e,
3.) (a, b) − → ab is continuous in both variables together at the point (e, e).
32 3. Continuous Representations

Every abstract group can be regarded as a topological group if one gives it the discrete
topology where every subset is open.

All the matrix groups presented in Section 0.1 and later on are topological groups: As
2 2
they are subsets of Mn (R)  Rn , they inherit the (standard) topology from Rn . Open
balls in Mn (R) (and analogously in Mn (C)) are

Kρ (A) := {B ∈ Mn (R),  B − A < ρ}, ρ ∈ R>0 , A ∈ Mn (R),

where

(3.1)  A  := < A, A >, < A, B >:= Re Tr tĀB.

This topology is consistent with the group structure: multiplication of two matrices gives
a matrix where the coefficients of the product matrix are polynomials in the coefficients of
the factors and thus the two continuity requirements 1.) and 3.) in the remark above are
fulfilled. Similarly with condition 2.), as the inverse of a matrix has coefficients which
are polynomials in the coefficients of the matrix divided by the determinant, another
polynomial here not equal to zero, i.e. we have again a continuous function.

For A ∈ U(n) we have tĀA = E, i.e.  A 2 = n, so we have an example of a compact

group. Heis(R), SL(2, R) and SO(3, 1) are not compact but these groups together with
all others in this text are locally compact, i.e. each element admits a neighbourhood with
compact closure (as Rn does). Moreover, they are closed subgroups of some GL(N, R).
This, in principle, has to be verified in each case. A practical tool is the following stan-
dard criterium.
Remark 3.2: G is closed if for every convergent sequence (Aj ), Aj ∈ G with lim Aj ∈
GL(n, R), we have lim Aj ∈ G.

An example for a different procedure is the following: SL(n, R) is a closed subgroup of

GL(n, R) as det: GL(n, R) −→ R∗ is a continuous map and SL(n, R) is the inverse
image of the closed subset {1} ⊂ R∗ .

Now we have the possibility to define the class of groups we will restrict to in this text:

Definition 3.2: A group is called a linear group if there is an n ∈ N such that G is

isomorphic (as abstract group) to a closed subgroup of GL(n, R) or GL(n, C).

Remark 3.3: More generally, one calls also linear those groups, which are isomorphic
to a closed subgroup of GL(n, H), H the quaternion skewfield, which in this text will
only appear in a later application.

Remark 3.4: In another kind of generality one treats Lie groups, which are topological
groups and at the same time differentiable manifolds such that

G × G −→ G, (a, b) −→ ab and G −→ G, a −→ a−1 ,

are differentiable maps. As we try to work without the (important) notion of a manifold
as long as possible, it is quite adequate to stay with the concept of linear groups as
defined above (which comprises all the groups we are interested in here).
3.2 The Continuity Condition 33

While treating topological notions, let us remind that a group G is called (path-)connected
if for any two a, b ∈ G there is a continuous map of a real interval mapping one endpoint
of the interval to a and the other to b.

Exercise 3.1: Show that SU(2) and SL(2, R) are connected and GL(2, R) is not.

Before we come to the definition of continuous representations, let us adapt the notion
of a G-set from Section 0.2 to the case that G is a topological group.

Definition 3.3: A set X is called a (left-)topological G-space iff it is a left G-set as in

0.2 and X is also provided with a topology such that the map

G × X −→ X , (g, x) −→ gx,

is continuous.

In the sequel, G-space will abbreviate the content of this definition.

If the space X is Hausdorff (as all our examples will be), then the stabilizing subgroup
Gx ⊂ G of each point x ∈ X is a closed subgroup of G. Conversely, if H is a closed
subgroup of a Hausdorff group G, then the space G/H of left cosets gH, g ∈ G, is a
homogeneous Hausdorff G-space when given the usual quotient space topology. If X is
any other topological G-space with the same stabilizing group Gx = H for x ∈ X , then
the natural map
Φ : G/H −→ X , gH −→ gx

is 1-1 and continuos. If G and X are locally compact, then Φ is a homeomorphism and
allows an identification of both spaces.

3.2 The Continuity Condition

From now on a group will mean a locally compact topological group. Then an appropri-
ate representation space V should be a linear topological space. To make things easier
for us, we will suppose that V = H is a separable complex Hilbert space (i.e. having de-
numerable Hilbert space basis (ej )j∈I , I  N), as for instance L2 (S 1 ) with basis given by
the functions ej (t) := exp(2πijt), j ∈ Z. For the finite-dimensional case this is nothing
new: we have as before V = Cn .
If we write again GL(V ), in the finite-dimensional case, we have the same set as be-
fore provided with the topology given by an isomorphism GL(V )  GL(n, C). In the
infinite-dimensional case, V = H, GL(V ) = GL(H) is meant as the group L(H) of linear
bounded operators in H. We recall here the following fundamental notions relevant for
the infinite-dimensional cases (from e.g. [BR] p.641f) :

• An operator T with domain D(T ) ⊂ H is said to be continuous at a point v0 ∈ D(T )

iff for every  > 0 there exists a δ = δ() > 0 such that from  v − v0 < δ with
v ∈ D(T ) we can conclude  T v − T v0  < .
34 3. Continuous Representations

• An operator T is said to be bounded iff there exists a constant C such that one has
 T v ≤ C  v  for all v ∈ D(T ).
• The norm  T  of a bounded operator T is defined as the smallest C such that

 Tv ≤ C  v for all  v  ≤ 1

.
Remark 3.5: A bounded linear operator is uniformly continuous. Conversely, if a linear
operator T is continuous at a point v0 (e.g. v0 = 0), then T is bounded.

Exercise 3.2: Prove this.

• Let A be a linear operator in H with dense domain D(A) ⊂ H. Then one can prove
that there are v, v  ∈ H such that < Au, v > = < u, v  > holds for all u ∈ H. One
sets v  =: A∗ v, verifies that A∗ is a linear operator, and calls A∗ the adjoint of A.
• For operators A and B in H one writes B ⊃ A iff one has D(B) ⊃ D(A) and
Bu = Au for all u ∈ D(A). Then B is called an extension of A.
• A linear operator A with dense domain D(A) is called symmetric iff one has A∗ ⊃ A
and
< Au, v > = < u, Av > for all u, v ∈ D(A).

• A linear operator A with dense domain is called self-adjoint iff one has A∗ = A.

One can prove that a linear symmetric operator A with D(A) = H is bounded and
self-adjoint.

If not mentioned otherwise, L(H) is provided with the strong operator topology ([BR]),
i.e. we have as basis of open neighbourhoods

Bρ,u (T0 ) := {T ∈ L(H);  T u − T0 u  < ρ}, ρ ∈ R>0 , u ∈ H, T0 ∈ L(H).

In the sequel a representation of G shall mean the following.

Definition 3.4: A linear representation (π, H) is said to be continuous iff π is strongly

continuous, i.e. the map
G −→ H, g −→ π(g)v
is continuous for each v ∈ V .

Remark 3.6: It is quite obvious that one could also ask for the following continuity
requirements:
a) The map G × H −→ H, (g, v) −→ π(g)v is continuous.
b) The map G −→ GL(H), g −→ π(g) is continuous.
and
c) For each v ∈ V, the map G −→ H, g −→ π(g)v is continuous in g = e and
there is a uniform bound for  π(g)  in some neighbourhood of e ([Kn] p.10).

Definition 3.5: A representation (π, V ) is bounded iff we have

sup  π(g)  < ∞.

g∈G
3.2 The Continuity Condition 35

In all examples in this section and in most cases of the following sections all these con-
ditions are fulfilled. We will not discuss the hypotheses necessary so that all these
conditions are equivalent and refer to this to [Ki] p.111, [Kn] p.10 and [BR] p.134. But
we will show for some examples how the continuity of a representation comes out. Later
we will not examine continuity and leave this to the reader’s diligence. (To check this
and the question whether a representation from the examples to be treated is bounded
or not may even be (part of) the subject of a bachelor thesis.)

Example 3.1: The linear representation π = χk , k ∈ Z of G = SO(2) in V = C given

by
cos ϑ sin ϑ
χk (r(θ)) := eikϑ for all g = r(ϑ) = ( ), ϑ ∈ R,
− sin ϑ cos ϑ
is obviously continuous: The map
(g, z) −→ eikϑ z
is continuous in g = r(ϑ) for each z ∈ C or even contiuous in both variables g and z,
because the exponential function is (in particular) continuous and an open neighborhood
of e ∈ SO(2) is homeomorphic to an open interval in R containing 0.

Later we will modify the notions of reducibility and irreducibility for continuous rep-
resentations. But here we can already prepare the way how to do this and prove the
following:
Remark 3.7: Each irreducible continuous representation π of G = SO(2) is equivalent
to some χk , k ∈ Z.

Proof: We look at (π, V ) only as a linear representation and apply Schur’s Lemma from
1.5. As G is abelian, we can conclude by Theorem 1.4 following from Schur’s Lemma in
1.5 that V = C. Hence we can assume π : SO(2) −→ C∗ . As SO(2) is homeomorphic to
R/Z, we have a continuous map r : R −→ R/Z. Now we use the assumption that π is
continuous and we have a sequence of continuous maps
R −→ SO(2) −→ C∗ , t −→ r(t) −→ π(r(t)),
and in consequence a continuous map
R
t −→ π(r(t)) =: χ(t) ∈ C∗ .
As π is a linear representation, χ has the property
χ(t1 + t2 ) = χ(t1 )χ(t2 ), χ(t) = 1 for all t ∈ 2πZ.
Here we have to recall a fact from standard calculus (see for instance [Fo] p.75): Every
continuous function χ : R −→ C∗ obeying the functional equation χ(t1 +t2 ) = χ(t1 )χ(t2 )
is an exponential function, i.e. χ(t) = c exp(tζ) for c ∈ C∗ , ζ ∈ C. The condition
χ(2πn) = 1 for all n ∈ Z fixes c = 1 and ζ = ki for k ∈ Z.

By the way, already here, we can can anticipate a generalization of Proposition 2.4 in
2.2 from finite to compact groups: L2 (SO(2)) can be identified with a space of periodic
functions. And from the theory of Fourier series one knows that this space has as a
Hilbert space basis the trigonometric functions ej , j ∈ Z, with
ej (t) := e2πijt , t ∈ R.
36 3. Continuous Representations

This can be translated into the statement that each irreducible continuous representation
of SO(2) is contained (with multiplicity one) in the left- (or right-)regular representation
λ resp. ρ of SO(2) and λ and ρ are direct sums of the χj , j ∈ Z.

Example 3.2: Let G be SO(3) and, for fixed m ∈ N, let πm be the representation of G
given on the space Vm of all homogeneous polynomials of degree m in 3 variables, i.e.

Vm := {P ∈ C[x1 , x2 , x3 ]; P homogeneous, deg P = m},

by
π(g)P (x) := P (g −1 x), x = t (x1 , x2 , x3 ) (a column).

πm is a linear representation as is already clear from the general consideration in 1.3.

The continuity condition is fulfilled since SO(3) inherits the topology from GL(n, R) and
matrix multiplication x −→ g −1 x is a continuous map and polynomials P are continuous
functions in their arguments.

This example is similarly constructed as the example π = πj for G = SU(2) introduced

in 1.3, which is continuous by the same reasoning as are also the other examples in 1.3
and later on. So we will now take the continuity of our representations for granted.

Moreover, let us announce here that we shall have to discuss the relationship of the two
examples G = SO(3) and G = SU(2) very thoroughly later. The representations πj of
SU(2) will come out as irreducible, but here we see already that for instance for m = 2
+ x23 )C since each element g ∈ SO(3)
(πm , Vm ) has an invariant subspace V0 := (x21 + x22
acts on R3 leaving invariant the spheres given by x2i = ρ2 .

The addition of the continuity condition to the definition of the representation has nat-
ural consequences for the notions related to the definition of the linear representation in
Sections 1.1 and 1.2: Here only closed subspaces are relevant. Hence in the future, we
use definitions modified as follows.

• Suppose that there is a closed subspace V1 of the space V of a representation (π, V )

of G, invariant under all π(g), g ∈ G. Then π1 := π|V is called a (topological)
subrepresentation of π.

• The representation π2 on the factor (or quotient) space V /V1 is called a (topological)
factor representation.

The representation (π, V ) is called

• (topologically) irreducible iff it admits no nontrivial subrepresentation,

• (topologically) decomposable iff there exist closed subspaces V1 and V2 in V such

that V = V1 ⊕ V2 . In this case we write π = π1 + π2 , where πi := π |Vi ,
3.2 The Continuity Condition 37

• completely reducible (or discretely decomposable) iff it can be expressed as a direct

sum of irreducible subrepresentations,

• unitary iff V is a Hilbert space H and every π(g) respects the scalar product in H.

• If (π, V ) and (π  , V  ) are continuous representations of G, then a bounded linear

map (or operator) F : V −→ V  is called an intertwining operator for π and π  iff
one has
F π(g) = π  (g)F for all g ∈ G.

Then Schur’s Lemma here appears in the following form.

Theorem 3.1: If π and π  are unitary irreducible representations in the Hilbert spaces
H resp. H and F : H −→ H is an intertwining operator, then F is either an isomor-
phism of Hilbert spaces or F = 0.

This theorem has as a consequence the following irreducibility criterium, which is a

generalization to the infinite-dimensional case of the similar one we proved for the finite-
dimensional linear representations without the continuity condition in 1.5. We will call
it the unitary Schur:

Theorem 3.2: A unitary representation (π, H) is irreducible iff the only operators

commuting with all the π(g) are multiples of the identity.

The proofs of these theorems (see e.g. [BR] p. 143/4 or [Kn] p.12) use some tools from
functional analysis (the spectral theorem) which we do not touch at this stage.

We indicate another very useful notion (in particular concerning unitary representations).

Definition 3.6: A representation π of G in V is said to be cyclic iff there is a v ∈ V

(called a cyclic vector for V ) such that the closure of the linear span of all π(g)v is V
itself.

Theorem 3.3: Every unitary representation (π, H) of G is the direct sum of cyclic rep-
resentations.

As the proof is rather elementary and gives some indications how to work in the infinite-
dimensional case and how the continuity condition works, we give it here (following [BR]
p.146):
Let Hv1 be the closure of the linear span of all π(g)v1 , g ∈ G for any 0 = v1 ∈ H.
Then Hv1 is π-invariant: Indeed, let Hv 1 be the linear span of all π(g)v1 , then for each
u ∈ Hv1 we have a sequence (un ), un ∈ Hv 1 which converges to u. Obviously, one has
π(g)un ∈ Hv 1 . The continuity of each π(g) implies π(g)un −→ π(g)u and hence we have
π(g)u ∈ Hv1 and Hv1 is invariant. Thus the subrepresentation π1 = π |Hv1 is cyclic with
cyclic vector v1 .
If Hv1 = H, choose 0 = v2 ∈ Hv⊥1 = H \ Hv1 , consider the closed linear span Hv2 which
is π-invariant and orthogonal to Hv1 , and continue like this if H = Hv1 ⊕ Hv2 .
Let ξ denote the family of all collections {Hvi }, each composed of a sequence of mutu-
ally orthogonal, invariant and cyclic subspaces. We order the family by means of the
38 3. Continuous Representations

inclusion relation. Then ξ is an ordered set to which Zorn’s Lemma applies: It assures
the existence of a maximal collection {Hvi }max . By the separability of H, there can
be at most a countable number of subspaces in {Hvi }max and their direct sum, by the
maximality of {Hvi }max must coincide with H.

As a consequence of this theorem, we have a very convenient irreducibility criterium for

unitary representations :
Corollary: A unitary representation (π, H) of G is irreducible iff every nonzero u ∈ H
is cyclic for π.

Proof: i) If π is irreducible, every 0 = u ∈ H is cyclic as to be seen from the first part

of the proof above.
ii) Suppose that H1 ⊂ H is a nontrivial invariant subspace of H and 0 = u ∈ H1 .
Due to the invariance of H1 we have π(g)u ∈ H1 . Moreover, the closure of the linear
span of all π(g)u is contained in H1 but by assumption is also equal to H. Hence, we
have a contradiction and π is irreducible.

To close these general considerations let us recall the following. If H and H are Hilbert
spaces with scalar products <, > resp. <, > , then we can define a scalar product in
H ⊗ H by the formula
< v ⊗ v  , w ⊗ w >:=< v, w >< v  , w > .
If either H or H is finite-dimensional, then the space H ⊗ H equipped with this scalar
product is complete. If both H and H are infinite-dimensional, we complete H ⊗ H by
the norm defined by the scalar product above and denote this by H⊗H ˆ .

3.3 Invariant Measures

We want to extend the results from finite  groups to compact groups.
 The tool to do
this is given by replacing the finite sum g∈G by an integration G . To be a bit more
precise, one has to recall some notions from measure and integration theory, which more-
over are inevitable in all the following sections. For the whole topic we recommend the
classical books Measure Theory by Halmos ([Ha]) and Abstract Harmonic Analysis by
Hewitt-Ross ([HR]) or, for summaries, the appropriate chapters in [Ki] p.129ff and [BR]
p.67ff.

If X is a topological space with the family V = {Oi }i∈I of open sets, it is also a Borel
space with a family B of Borel sets B, where these sets are obtained from open and closed
sets by the operation of countable unions, countable intersections and complementation.
A little more generally, a Borel space is a set with a family B of subsets B such that B
is a σ-algebra, i.e. one has

X \ B ∈ B for B ∈ B,
i∈I Bi ∈ B for Bi ∈ B, i ∈ I  N,
i∈I Bi ∈ B for Bi ∈ B, i ∈ I  N.

Let (X , B) and (X  , B ) be two Borel spaces. A map F : X −→ X  is called a Borel map

iff one has F −1 (B  ) ∈ B for all B  ∈ B .
3.3 Invariant Measures 39

Definition 3.7: We call µ a measure on the Borel space (X , B) if µ is a map

µ : B −→ [0, ∞] := R≥0 ∪ {∞},

where µ is σ-additive,
 i.e.
i) µ(∪Bi ) = µ(Bi ) if the Bi are pairwise disjoint (i ∈ I  N), and
ii) there is a covering of X by sets Bi of finite measure, i.e. we have Bi ∈ B, i ∈ I  N
with µ(Bi ) < ∞ and ∪Bi = X .

The most obvious example is X = R provided with the Lebesgue-measure µ =: λ, which

is defined by
λ([a, b)) := b − a for a, b ∈ R, a ≤ b.

As we know from elementary integration theory (or see immediately), λ is translation

invariant. This is a special incidence of the following general machinery.
Let X be a space with a family V of open sets and a family B of Borel sets B and let
µ be a measure on X . We then say (X , B, µ) is a measure space. If we have a right
G-action
(G × X ) −→ X , (g, x) −→ xg,

we can define a measure µg by

µg(B) = µ(Bg) for all B ∈ B.

Then the measure µ is called right-invariant iff one has µg = µ for all g ∈ G. Similarly
µ is called left-invariant iff we have a left action and gµ = µ for all g ∈ G where
gµ(B) = µ(gB) for B ∈ B.

Now we take X = G, a locally compact group. Then a right- or left-invariant measure

on G is called a right- or left-Haar measure. And we have as central result ([Ki] p.130):

Theorem 3.4 (Haar): On every locally compact group with countable basis for the
topology, there exists a nonzero left-invariant (and similarly a right-invariant) measure.
It is defined uniquely up to a numerical factor.

We will not go into the proof of the Theorem (see for instance [HR] Ch.IV, 15) but give
later on several concrete examples for its validity. The most agreable case is the following:

Definition 3.8: G is called unimodular iff it has a measure which is simultaneously left-

and right-invariant. Such a measure is also called biinvariant or simply invariant.

To make these up to now only abstract concepts work in our context, we have to recall
a more practical aspect due to the general integration theory:
Let (X , B, µ) be a measure space and f : X −→ C a measurable function, i.e. f is the
sum f = f1 − f2 + i(f3 − f4 ) of four real nonnegative functions fj (j = 1, .., 4) having the
property
{x ∈ X ; fj (x) > a} ∈ B for all a > 0.
40 3. Continuous Representations

For measurable nonnegative functions f general integration theory provides a linear map

f −→ µ(f ) = f (x)dµ(x) ∈ C ∪ {∞}.
X

We call a measurable function f integrable iff |f (x)| dµ(x) < ∞.
If X = G as above, one usually writes

dµ(g) =: dr g if µ is right − invariant,

=: dl g if µ is left − invariant,
=: dg if µ is bi − invariant.

The following fact is central for applications to representation theory.

Proposition 3.1: Let G be a locally compact group. Then there is a continuous homo-
morphism ∆G , called the modular function of the group G, into the multiplicative group
of positive real numbers for which the following equalities hold

dr (gg  ) = ∆G (g)dr g  , dl (gg  ) = ∆G (g  )−1 dl g,

dr g = const.∆G (g)dl g, dl g = const.∆G (g)−1 dr x,

dr (g −1 ) = ∆G (g)−1 dr g = const.dl g, dl (g −1 ) = ∆G (g)dl g.

As Kirillov does in [Ki] p.130, we propose the proof as an exercise based on the applica-
tion of Haar’s Theorem. There is a proof in [BR] p.68/9, but the modular function there
is the inverse to our function, which here is defined following Kirillov.

Proposition 3.2: If G is compact, then the Haar measure is finite and biinvariant.

Proof: The image of a compact group under the continuous homomorphism ∆G is again
a compact group. The multiplicative group R>0 has only {1} as a compact subgroup.
 So
by Proposition 3.1, we have a biinvariant measure which can be normalized to G dg = 1.

3.4 Examples
Most of the noncompact groups to be treated later on are also unimodular, e.g. the group
SL(2, R) or the Heisenberg group Heis(R). We will show here how this can be proved.

Example 3.3: G = Heis(R) = {g = (λ, µ, κ); λ, µ, κ ∈ R} with

gg  = (λ + λ , µ + µ , κ + κ + λµ − λ µ).

As Heis(R) is isomorphic to R3 as a vector space, we try to use the usual Lebesgue

measure for R3 . So we take
dg := dλdµdκ.
The left translation λg0 : g −→ g0 g =: g  acts as

(λ, µ, κ) −→ (λ := λ0 + λ, µ := µ0 + µ, κ := κ0 + κ + λ0 µ − λµ0 ).

3.4 Examples 41

The Jacobian of this map is

⎛ ⎞
1 −µ0
Jg0 (g) = ⎝ 1 λ0 ⎠
1

with det Jg0 (g) = 1. Hence we see that dg is left-invariant. Similarly we prove the right-
invariance.

Remark 3.8: Without going into the general theory of integration on manifolds, we
indicate that, for calculations like this, it is very convenient to associate the infinitesimal
measure element with an appropriate exterior or alternating differential form and then
check if this form stays invariant. In the example above we take

ω  := dλ ∧ dµ ∧ dκ

and calculate using the the general rules for alternating differential forms (we shall say
a bit more about this later in section 8.1), in particular

dµ ∧ dλ = − dλ ∧ dµ, dµ ∧ dµ = 0,

as follows

ω  ◦ λg0 = dλ ∧ dµ ∧ (dκ + λ0 dµ − µ0 dλ) = dλ ∧ dµ ∧ dκ = ω.

Example 3.4: G = SO(2)

We introduced as standard notation

cos ϑ sin ϑ
SO(2)
r(ϑ) = ( ).
− sin ϑ cos ϑ

Here we have dg = dϑ as biinvariant measure with

  2π
dg = dϑ = 2π.
G 0

Example 3.5: G = SU(2)

This group will be our main example for the representation theory of compact groups.
As standard parametrization we use the same as Wigner does in [Wi] p.158: The general
form of a two–dimensional unitary matrix

a b
g=( ).
c d

of determinant one is

a b
g=( ), a, b ∈ C with | a2 | + | b2 | = 1.
−b̄ ā
42 3. Continuous Representations

With
eiα cos ϑ sin ϑ
s(α) := ( ) and r(ϑ) = ( )
e−iα − sin ϑ cos ϑ
one has a decomposition

a b
SU(2)
g = ( ) = s(−α/2)r(−β/2)s(−γ/2), α, γ ∈ [0, 2π], β ∈ [0, π].
−b̄ ā

In particular, one can see that SU(2) is generated by the matrices s(α), r(β).

Exercise 3.3: Prove this.

Later on we shall see later that the angles α, β, γ correspond to the Euler angles for
the three-dimensional rotation introduced by Wigner in [Wi] p.152. In Hein [He] one
finds all these formulas, but unfortunately in another normalization, namely one has
zHein = a, uHein = −b, αHein = β, βHein = −α, γHein = −γ. Our parametrization
is one-to-one for α, γ ∈ [0, 2π], β ∈ [0, π] up to a set of measure zero. An invariant
normalized measure is given by

dg := (1/16π 2 ) sin β dαdβdγ.

Exercise 3.4: Verify this.

Example 3.6: An example for a group G which is not unimodular is the following

a b
G = {g = ( ) ; a ∈ R∗ , b ∈ R}
1

aa ab + b
We have gg  = ( ) =: g  , hence
1

da ∧ db = a da ∧ (b da + db) = a da ∧ db

a a a b + b
so that dr g = a−1 dadb and, from g  g = ( ) =: g ∗
1

da∗ ∧ db∗ = a da ∧ db,

2

we see dl g = a−2 dadb. Hence, we have here ∆G (g) = a−1 .

Exercise 3.5: Determine ∆G for

1 x y 1/2
G = {g = ( )( ); y > 0, x ∈ R}.
1 y −1/2
Chapter 4

Representations of Compact
Groups

As already mentioned, the representation theory of compact groups generalizes the the-
ory of finite groups. We shall closely follow the presentation in [BR], and cite from [BR]
p.166: “The representation theory of compact groups forms a bridge between the rela-
tively simple representation theory of finite groups and that of noncompact groups. Most
of the theorems for the representations of finite groups have direct analogues for compact
groups and these results in turn serve as the starting point for the representation theory
of noncompact groups.”

4.1 Basic Facts

Let
 G be a compact group provided with an invariant measure µ normalized such that
G
dg = 1 and let π be a representation of G in a Hilbert space H. We recall that all our
representations are meant to be linear und continuous. The following statements are easy
to prove for the finite-dimensional case dim H < ∞. Most times this is quite sufficient
because it will come out that all irreducible representations of compact groups are finite-
dimensional. To prepare a feeling for the noncompact cases where infinite-dimensionality
is essential, we state the results here for general H. But we will report on the proofs only
when not too much general Hilbert space theory is needed.

Proposition 4.1: Let (π, H) be a representation of G in the Hilbert space H with scalar
product <, > . Then there exists a new scalar product <, > defining a norm equivalent to
the initial one, relative to which the map g −→ π(g) defines a unitary representation of G.

Proof: i) As to be expected from the case of finite groups, we define <, > by

< u, v > := < π(g)u, π(g)v > dg.
G

Sesquilinearity and hermiticity of <, > are obvious. Let us check that we can conclude
u = 0 from 
< u, u > = < π(g)u, π(g)u > dg = 0 :
G
44 4. Representations of Compact Groups


From G < π(g)u, π(g)u > dg = 0 one deduces that < π(g)u, π(g)u > = 0 almost every-
where. If g ∈ G is such that π(g)u = 0, then π(g)−1 π(g)u = u = 0.

ii) < , > is π-invariant:


 
< π(g )u, π(g )v > = < π(g)π(g  )u, π(g)π(g  )v > dg

G

= < π(gg  )u, π(gg  )v > dg = < u, v >,

G

because dg is left-invariant.

iii) The norms  .  and  .  induced by <, > resp. <, > are equivalent:
We have

 u 2 = < π(g)u, π(g)u > dg
G

< u, u > dg = N 2  u 
2
≤ sup  π(g)  2
g∈G G

with N := supg∈G  π(g) . And from

 u  2 = < π(g −1 )π(g)u, π(g −1 )π(g)u >

sup  π(g −1 ) 2 < π(g)u, π(g)u > = N 2  π(g)u 
2
≤
g∈G

it follows that

 u  < u, u > dg
2
=
G

≤ N2 < π(g)u, π(g)u > dg = N 2 < u, u >= N 2  u 2 .
G

Hence we have N −1  u  ≤ u ≤ N  u  and  .  and  .  are equivalent.

iv) Equivalent norms define the same families of open balls and thus equivalent topolo-
gies, i.e. if g −→ π(g)v, v ∈ H is continuous for the topology belonging to  .  , it is also
continuous for the topology deduced from  . .

We now turn to the already announced result that every unitary irreducible represen-
tation is finite-dimensional. In its proof we use an interesting tool, namely the Weyl
operator Ku defined for elements u, v from a Hilbert space H with π-invariant scalar
product by

Ku v := < π(g)u, v > π(g)udg.
G

Remark 4.1: The Weyl operator has the following properties

i) Ku is bounded,
ii) Ku commutes with every π(g), i.e. Ku π(g) = π(g)Ku .
4.1 Basic Facts 45

Proof: We perform the standard calculations using Cauchy–Schwarz inequality.


i)  Ku v  ≤ G |< π(g)u, v >| π(g)u  dg
≤ G  π(g)u 2  v  dg = u 2  v ,

ii) π(g  )Ku v = G < π(g)u, v > π(g  g)udg
= G < π(g  g)u, π(g  )v > π(g  g)udg
= G < π(g)u, π(g  )v > π(g)udg = Ku π(g  )v
for all g  ∈ G and v ∈ H.

Theorem 4.1: Every irreducible unitary representation π of G in a Hilbert space H is

finite-dimensional.

Proof: By the last remark, Ku is an intertwining operator for π. Hence by Schur’s

Lemma, Ku is a homothety, i.e. Ku = λ(u)idH and, in consequence,


(4.1) < Ku v, v > = < v, π(g)u >< π(g)u, v > dg
G
= |< π(g)u, v >|2 dg = λ(u)  v 2 .
G

By interchanging the rôles of u and v and using the equality (which is equivalent to the
unimodularity of G)  
f (g −1 )dg = f (g)dg,

we get
 
λ(v)  u 2 = |< π(g)v, u >|2 dg =
|< u, π(g)v >|2 dg
 
−1
= |< π(g )v, u >| dg =
2
|< π(g)u, v >|2 dg

= λ(u)  v 2 .

Hence we have λ(u) = c  u 2 for all u ∈ H with a constant c ∈ C. In (4.1) we put

u = v,  v = 1 and get

|< π(g)u, u >|2 dg = λ(u) = c.

As the nonnegative continuous function g −→ |< π(g)u, u >| assumes the value  u = 1
at g = e, we must have c > 0.
Let {ei }i=1,...,n be a set of ON-vectors in H. In (4.1) we now put u = ek and v = e1 to
obtain 
|< π(g)ek , e1 >|2 dg = λ(ek )  e1 2 = c.

and
n 



n
nc = |< π(g)ek , e1 >|2 dg = |< π(g)ek , e1 >|2 dg
k=1 G G k=1

≤  e1 2 dg = 1.
G
46 4. Representations of Compact Groups

The last inequality is a special case of Parseval’s

 inequality saying that for a unitary ma-
trix representation g −→ A(g) one has k | Ak1 (g) |2 ≤ 1. So finally, one has n ≤ 1/c,
i.e. the dimension of H must stay finite.

We already proved that a finite-dimensional unitary representation is completely re-

ducible. This result can be sharpened for compact groups G:
Theorem 4.2: Every unitary representation π of G is a direct sum of irreducible finite-
dimensional unitary subrepresentations.

The proof uses some more tools from functional analysis and so we skip it here (see [BR]
p.169/170). But we mention the appearance of the important notion of a Hilbert–Schmidt
operator, which is an operator A in H, s.t. for an arbitrary basis {ei }i∈I of H we have

 Aei 2 < ∞.
i∈I

Again, as in the theory of finite groups, we have ON-relations for the matrix elements of
an irreducible representation.

Theorem 4.3: Let π and π  be two irreducible unitary representations of G and A(g)
resp. A (g) their matrices with respect to a basis {ei } of H and {ek } of H . Then one
has relations

Aij (g)Akl (g)dg = 0 if π ∼ π 
= (1/n)δik δjl if π ∼ π  and n := dim H.

Proof: As to be expected, one applies Schur’s Lemma (Theorem 3.1):

We introduce a matrix fij with entries (fij )kl = δik δjl and an operator

Fij := π(g)fij π  (g −1 )dg.
G

This is an operator intertwining the representations π  and π since we have



π(g )Fij = π(g  g)fij π  (g −1 )dg
G

= π(g)fij π  (g −1 g  )dg = Fij π  (g  ).

G

Hence, if π ∼ π  , we have Fij = 0 or in matrix form for all (r, t)


(Fij )rt = Ari (g)Ajt (g −1 )dg

= Ari (g)A tj (g)dg = 0.

For π = π  , Schur’s Lemma requires Fij = λij id. Hence for (r, t) = (i, j), the orthogo-
nality relations just obtained are still satisfied. For (r, t) = (i, j), we have

(Fij )ij =  Aii (g)Ajj (g)dg = λij = 0 if i = j,
= | Aii (g) |2 dg = λii if i = j.
4.1 Basic Facts 47

We have Tr Fii id = nλii and from the definition of Fij ,


Tr Fii = Tr (A(g)fii A(g −1 )dg = Tr fii = 1
G

and hence λii = 1/n.

Similar to the case of finite groups we have a nice fact:

Theorem 4.4: Every irreducible unitary representation π of G is equivalent to a sub-

representation of the right regular representation.

√ of π(g) and let Hπ be the subspace

Proof: Let A(g) = (Ajk (g))j,k=1,...,n be a matrix form
of L2 (G) spanned by the vectors ek with ek (g) := nA1k (g), k = 1, . . . , n (these are an
ON-system by the preceding theorem). We have
√
ρ(g0 )ek (g) = ek (gg0 ) = nA(gg0 )
√ √
= n Σj A1j (g)Ajk (g0 ) = Σj n A1j (g)Ajk (g0 )
= Σj ej (g)Ajk (g0 ).

Hence ρ restricted to Hπ is a subrepresentation equivalent to π.

Exercise 4.1: Show the same fact for the left regular representation.

Remark 4.2: With the same reasoning as in the case of finite groups, the characters χπ
of finite-dimensional  irreducible
 unitary representations π, defined by χπ (g) := Tr π(g)
resp. χπ (g) := i Aii (g) = i < A(g)ei , ei > for a matrix form with respect to a basis
e1 , . . . , en , have the following properties.

1) The characters are class functions, i.e. χ(g0 gg0−1 ) = χ(g) for all g, g0 ∈ G.
2) We have χ(g −1 ) = χ(g).
 
3)  π ∼ π , then χπ = χπ .
If
4) χπ (g)χπ (g)dg = 0 if π ∼ π  and = 1 if π ∼ π  .

If π is any finite-dimensional representation of G, then we can decompose it into irre-

ducible representations πi appearing with multiplicities mi and we have

h
χπ = mi χπi .
i=1

Due to the ON-relation 4) in Remark 4.2 above, we obtain


mi = χπ (g)χπi (g)dg
G

and

h 
m2i = χπ (g)χπ (g)dg.
i=1 G

This formula specializes to the very useful irreducibility criterium:

48 4. Representations of Compact Groups

Corollary: The finite-dimensional representation π is irreducible iff


χπ (g)χπ (g)dg = 1.
G

The big difference to the theory of finite groups is that for compact groups the number of
equivalence classes of irreducible representations no longer necessarily is a finite number.
We already know this from the example G = SO(2) in section 3.2. The central fact is
now the famous Peter–Weyl Theorem:

Theorem 4.5: Let Ĝ = {πi }i∈I be the unitary dual of G, i.e. a complete set of represen-
tatives of the equivalence classes of irreducible unitary representations of G. For every
i ∈ I let Ai (g) = (Aijk (g))j,k=1,...,ni be a matrix form of πi and
i √
Yjk (g) := ni Aijk (g).
Then we have
1) The functions Yjki
, i ∈ I, j, k = 1, . . . , ni form a complete ON-system in L2 (G).
2) Every irreducible unitary representation π of G occurs in the decomposition of
the right regular representation with a multiplicity equal to the dimension of π, i.e.
mult (πi , ρ) = ni .
3) Every C-valued continuous function f on G can be uniformly approximated by a
i
linear combination of the (Yjk ).
4) The characters (χπi )i∈I generate a dense subspace in the space of continuous class
functions on G.

The proof of these facts use standard techniques from higher analysis. We refer to [BR]
p. 173 - 176 or [BtD] p.134.

4.2 The Example G = SU(2)

Representation theory owes a lot to the work of H. Weyl and E. Wigner. As we have
the feeling that we can not do better, we will discuss the example G = SU(2) following
Wigner’s presentation in [Wi] p.163 - 166:

We put
a b
G = SU(2)
g = ( ), | a | 2 + | b | 2 = 1
−b̄ ā
and for j a half integral nonnegative number, we let V (j) be the space C[x, y]2j of
homogeneous polynomials of degree 2j in two variables. Thus we have dim V (j) = 2j + 1.
We choose as a basis of V (j) monomials fp , which are conveniently normalized by
xj+p y j−p
fp (x, y) := , p = −j, −j + 1, . . . , j.
(j + p)! (j − p)!
We define the representation πj by the action
   
x −1 x
(g, ) −→ g ,
y y
4.2 The Example G = SU(2) 49

i.e. we put
πj (g)fp (x, y) := fp (āx − by, b̄x + ay),
and get by a straightforward computation
πj (g)fp (x, y) = Σp
fp
(x, y)Ajp
p (g)
with

j+p
(j + p)!(j − p)!(j + p )!(j − p )!


Ajp
p (g) = (−1) k
aj−p −k āj+p−k bk b̄k+p −p
(j − p − k)!(j + p − k)!k!(k + p − p)!
k=0

in particular

(2j)!
(4.2) Ajjp (g) = .
(j + p)!(j − p)!

Exercise 4.2: Verify these formulae (at least) for j = 1.

Theorem 4.6: These representations πj , j ∈ (1/2) N0 , are unitary, irreducible, and - up

to equivalence - there are no other such representations of SU(2).

Proof: i) Unitarity

Wigner’s proof relies on the fact that the polynomials fp are normalized so that we have

j
1 (| x2 | + | y 2 |)2j
f¯p fp = | x2 |j+p | y 2 |j−p = .
p=−j p=−j
(j + p)!(j − p)! (2j)!

By a similar computation, we conclude from the definition of πj

j
j
| āx − by |2(j+p) | b̄x − ay |2(j−p)
| πj (g)fp (x, y) |2 =
p=−j p=−j
(j + p)!(j − p)!
(| āx − by |2 + | b̄x − ay |2 )2j (| x |2 + | y |2 )2j
= = ,
(2j)! (2j)!
where the last equality follows from the unitarity of g. So we have the invariance
Σp | π(g)fp |2 = Σp | fp |2 .

Substituting here π(g)fp = Σp
fp
Ajp
p (g), we get

Σp (Σp
f¯p
Ājp
p (g)Σp

fp

Ajp

p (g)) = Σp f¯p fp .

If we know that the (2j + 1)2 functions f¯p
fp

are linearly independent, we see that we
have
Σp Ājp
p (g)Ajp

p (g) = Σp Ājp
p (g)tAjpp

(g) = δp
p

,
i.e. the unitarity of Aj (g). Hence we still have to show the linear independence of the
functions f¯p
fp

: We look at a relation








cp
p

x̄j+p ȳ j−p xj+p y j−p = 0, cp
p

∈ C.
p
,p


50 4. Representations of Compact Groups

Then in particular for x ∈ R with q = 2j + p + p , the coefficient of xq has to vanish.

After division by ȳ j y 3j−q this requires for



cp
,q−2j−p
(y/ȳ)p = 0.
p

But this condition implies that cp
,q−2j−p
= 0 (and thus the linear independence of the
f¯p
fp

), because we can write y/ȳ = eit , t ∈ R, and a relation
Σcp
,q−2j−p
eitp = 0 for all t ∈ R
requires that all the coefficients c must vanish.

ii) Irreducibility

Wigner’s proof of the irreducibility of πj is by direct application of the criterium which

here appears as Corollary to the finite-dimensional unitary Schur (Theorem 1.3) in Sec-
tion 1.5, namely by showing that any matrix M , which commutes with all Aj (g), must
necessarily be a constant matrix:
We take
e−iα/2
g = s(−α/2) = ( ).
eiα/2
Then the matrix Aj (g) specializes to the diagonal matrix
(4.3) Ajp
p (s(−α/2)) = δp
p eipα .
It is a standard fact that each matrix M commuting with all such matrices must be
diagonal. Just after the introduction of the matrices Aj (g) we remarked (in (9.9)) that
no element in the last row of Aj (g) vanishes identically. Then by equating the elements
of the j-th row of Aj (g)M and M Aj (g) we conclude that
Ajjk (g)Mkk = Mjj Ajjk (g), i.e. Mjj = Mkk ,
and M is a scalar multiple of the unit matrix.

One can also find a proof by application of the criterium given in the Corollary at the end
of Section 4.1 using the character of πj . We will see this as a byproduct of the following
third part.

iii) Completeness

Wigner’s proof that there are - up to equivalence - no other irreducible representations

of SU(2) than the πj can be seen as a direct proof of (part of) the Peter–Weyl Theorem
from 4.1 in this special case:
Characters are class functions. Since each unitary matrix can be diagonalized by conju-
gation with a unitary matrix, in each conjugacy class of SU(2) we find a diagonal matrix
of the type
eiα
s(α) = ( ), α ∈ [0, 2π).
e−iα
1
Conjugation by w = ( ) transforms
−1
1 ζ −1 −1 ζ
( )( )( )=( ).
−1 ζ 1 ζ −1
4.2 The Example G = SU(2) 51

Hence every conjugacy class is exactly represented by a matrix

eiα
s(α) := ( ), α ∈ [0, π), resp. s(α/2), α ∈ [0, 2π).
e−iα

As we have (see (4.3) in ii))

Ajp
p (s(−α/2)) = δp
p eipα

its trace is

j
χπj (s(−α/2)) = eipα =: ξj (α).
p=−j

So we have

ξ0 (α) = 1,
ξ1/2 (α) = e−iα/2 + eiα/2 = 2 cos(α/2),
ξ1 (α) = e−iα + 1 + eiα = 2 cos α + 1,
ξ3/2 (α) − ξ1/2 (α) = 2 cos(3α/2), etc.

It is now evident that SU(2) can have no other representations than the πj , j ∈ (1/2)N0 :
For the character of such a representation must, after multiplication by a weighting
function, be orthogonal to all ξj , and therefore to ξ0 , ξ1/2 , ξ1 − ξ0 , ξ3/2 − ξ1/2 , . . . . But
a function, which is orthogonal to 1, 2 cos(α/2), 2 cos α, 2 cos(3α/2), . . . in [0, 2π) must
vanish according to Fourier’s Theorem.

Exercise 4.3: Verify the irreducibilty of πj using the character criterium (Corollary to
Remark 4.2 in Section 4.1).

The explicit determination of the characters χπj = ξj of the irreducible representations

πj is very useful if one wants to decompose a given representation π.

Example 4.1: For π = π1 ⊗ π1 , by Theorem 1.6 in Section 1.6, we have

χπ (s(α/2)) = χπ1 (s(α/2))2 = (eiα + 1 + e−iα )2

= ξ2 (α) + ξ1 (α) + ξ0 (α),

i.e.
π1 ⊗ π1 = π0 + π1 + π2 .

Exercise 4.4: Verify the decomposition ([Wi] p.185)

j+j
πj ⊗ π =j
π .
=|j−j
|
52 4. Representations of Compact Groups

4.3 The Example G = SO(3)

One can construct representations of SO(3) by a procedure similar to the one used in
the case of SU(2), namely by using homogeneous polynomials in three variables. But
as we already remarked in the Example 3.2 in Section 3.2, we do not get irreducible
representations without some further considerations. So let us discuss this a bit later
and first follow a method suggested by H. Weyl deducing the representation theory of
SO(3) from the result just obtained in 4.2 and the fact that SU(2) is a double cover of
SO(3):

The elements g of SO(3) are real 3 × 3-matrices consisting of three ON-rows (or columns)
(a1 , a2 , a3 ), i.e. we have six conditions < aj , ak >= δjk for the nine entries in g. Hence
we expect three free parameters. It is a classical fact that the elements of SO(3) are
parametrized by three Euler angles α, β, γ. Again we follow Wigner’s notation ([Wi]
p.152):

Proposition 4.2: For every g ∈ SO(3) there exist angles α, β, γ ∈ R, such that

g = S3 (α)S2 (β)S3 (γ).

Here S3 (α) and S3 (γ) are rotations about the z-axis (or e3 ∈ R3 ) through α resp. γ and
S2 (β) is a rotation about the y-axis (or e2 ∈ R3 ), i.e. we put

⎛ ⎞
cos α − sin α
S3 (α) := ⎝ sin α cos α ⎠,
1
⎛ ⎞
cos α − sin α
S2 (α) := ⎝ 1 ⎠,
sin α cos α
⎛ ⎞
1
S1 (α) := ⎝ cos α − sin α ⎠ .
sin α cos α
Exercise 4.5: Prove this.

Remark 4.3: i) There are several notational conventions for the Euler angles in com-
mon use. In Hein [He] p.56 we find a statement (with proof) analogous to our Proposi-
tion 4.2 with S2 (β) replaced by S1 (β). For a parametrization of SO(3) one needs only
α, γ ∈ [0, 2π) and β ∈ [0, π]. This parametrization is not everywhere injective: for β = 0
α and γ are only fixed up to their sum. Goldstein [Go] p.145-148 gives an overview of
other descriptions.

ii) SO(3) is generated by rotations S3 (t) and S2 (t) or by rotations of type S3 (t) and S1 (t).

The key fact for our discussion here is given by the following theorem (we shall later
see that this is a special case of an analogous statement about SL(2, C) covering the
(restricted) Lorentz group SO(3, 1)+ ):
4.3 The Example G = SO(3) 53

Theorem 4.7: There is a surjective homomorphism

ϕ : SU(2) −→ SO(3)

with ker ϕ = {±E2 }.

This homomorphism can be chosen such that

ϕ(s(α/2)) = S3 (α) and ϕ(r(−β/2)) = S2 (β).

Proof: As in [Wi] p.158 we consider the Pauli matrices

1 i −1
sx := ( ), sy := ( ), sz := ( ).
1 −i 1

The three-dimensional R-vector space V of hermitian 2 × 2-marices H with trace zero is

the linear span of the Pauli matrices

V := {H ∈ M2 (C); t H = H̄, Tr H = 0}
−z x + iy
= {H = ( ) = xsx + ysy + zsz ; x, y, z ∈ R}.
x − iy z

G = SU(2) acts on V by conjugation, i.e. we have a map

G × V −→ V, (g, H) −→ gHg −1 = gH t ḡ =: H  ,

a b
which for H as above and g = ( ) gives by a straightforward computation
−b̄ ā

−z  x + iy 
H = ( )
x − iy  z

with

x = (1/2)(a2 + ā2 − b2 − b̄2 )x + (i/2)(a2 − ā2 + b2 − b̄2 )y + (āb̄ + ab)z

(4.4) y  = (i/2)(ā2 − a2 + b2 − b̄2 )x + (1/2)(a2 + ā2 + b2 + b̄2 )y + i(āb̄ − ab)z
z = −(āb + ab̄)x + i(āb − ab̄)y + (aā − bb̄)z.

E.g. from the fact that the map det is multiplicative, we deduce

det H = x2 + y 2 + z 2 = det H  = x + y  + z 
2 2 2

and if we write (4.4) as a matrix relation with column t (x, y, z) = x

x = A(g)x,

we see that multiplication with A(g) respects the euclidean quadratic form, i.e. we have
A(g) ∈ O(3). Moreover, by a continuity argument, we have A(g) ∈ SO(3): as every
g ∈ SU(2) can be deformed contiuously to g = E2 and det is a continuous function, we
cannot have det A(g) = −1.
54 4. Representations of Compact Groups

From the fact that conjugation is a group action, we see that

ϕ : g −→ A(g) ∈ SO(3)
is a group homomorphism. As ker ϕ = {g ∈ SU(2); gH t ḡ = H for all H ∈ V }, we can
deduce easily that ker ϕ = {±E2 } (by choosing appropriate matrices H).
The fact that ϕ is surjective is more delicate: If g = s(α/2), i.e. a = eiα/2 , b = 0, we see
from (4.4) that we have
A(s(α/2)) = S3 (α).
If g = r(β/2), we see from (4.4) that we have
A(r(β/2)) = S2 (−β).
We already know that SO(3) is generated by matrices of type S2 and S3 and so we are
done.

Now we use this result to describe representations of SO(3) :

If a representation π : SU(2) −→ GL(V ) factorizes through SO(3), i.e. one has a homo-
morphism π  : SO(3) −→ GL(V ) with π = π  · ϕ resp. a commutative diagram
π
SU(2) - GL(V )
@ 
ϕ@ π
@
@
R
SO(3)

then, this way, π induces the representation π  of SO(3). Obviously this is the case for a
representation π with π(−E2 ) = id, where we get a unique prescription by putting
π  (A) := π(g) if A = ϕ(g).
For π(−E2 ) = −id one gets a double valued representation by putting
π  (A) := ±π(g) if A = ϕ(g) = ϕ(−g).
(This is a special case of a projective representation, which will be discussed below.)

In consequence, we know the unitary dual of SO(3): The representations πj of SU(2) with
j ∈ N0 are exactly those with π(−E2 ) = π(E2 ) = id and thus induce (up to equivalence)
all irreducible unitary representations π  of SO(3): If there were any π  not in this family,
it would composed with ϕ produce a representation of SU(2), which by our completeness
result for SU(2) would be equivalent to a πj .
For g ∈ SU(2) and ϕ(g) = g  , Wigner ([Wi] p.167) conjugates the π  (g  ) := πj (g) = Aj (g)
from 4.2 by the diagonal matrix M = ((i)−2k δkl ) and calls the result Dj ({α, β, γ}) with
{α, β, γ} representing the Euler angles of g  ∈ SO(3) in the parametrization chosen above.
Hence we have from the explicit form of the matrices Aj (g) obtained in 4.2 the explicit
form


(j + p)!(j − p)!(j + p )!(j − p )!
j k
D ({α, β, γ})p
p = (−1)
(j + p − k)!(j + p − k)!k!(k + p − p)!
k
ip
α


× e (cos(β/2))2j+p+p −2k (sin(β/2))2k+p −p .
4.3 The Example G = SO(3) 55

The characters of these (projective) representations are

j
χπ
j (S3 (α)) = eipα = 1 + 2 cos α + · · · + 2 cos(jα), j integral
p=−j
= 2 cos(α/2) + 2 cos(3α/2) + · · · + 2 cos(jα), j half − integral.

The double valued representations π j , j half-integral, led to H. Weyl’s explanation of the

spin of an electron. As said above, they are special examples of projective representa-
tions, which are very natural for reasons from physics: In quantum theory the state of a
system is not so much described by an element v of a Hilbert space H but by an element
v∼ of the associated projective space P(H) consisting of the one-dimensional subspaces
of H. In general, one writes for a K-vector space P(V ) := {v∼ ; v ∈ V, v = 0} where v∼
is the class of v for the equivalence v ∼ v  if there is a λ ∈ K∗ such that v  = λv. Having
this in mind, it is appropriate to introduce the following.

Definition 4.1: A projective representation π̃ of G in the K-vector space V is a map

π̃ : G −→ GL(V ),

which enjoys the property

π̃(gg  ) = c(g, g  )π̃(g)π̃(g  ),

where
c : G × G −→ K∗

with
c(g, g  )c(gg  , g  ) = c(g, g  g  )c(g  , g  ) for all g, g  , g  ∈ G.

c is also called a system of multipliers.

Remark 4.4: The functional equation for c is equivalent to the associativity

π̃(g(g  g  )) = π̃((gg  )g  ).

Exercise 4.6: Prove this.

Every linear representation is also a projective representation. And analogously to the

case of linear representations, one also adds a continuity condition and discusses in gen-
eral these representations.

In the second part of his book [Wi], Wigner describes the application of these results
to the theory of spin and atomic spectra. Though we are tempted to do more, from
this we only take over that, roughly, an irreducible unitary (projective) representation πj
of SO(3) corresponds to a particle having three-dimensional rotational symmetry. The
number j specifying the representation then is its angular momentum quantum number
or its spin and the indices numbering a basis of the representation space are interpreted
as magnetic quantum numbers. We strongly recommend to the reader interested in this
topic to look into Wigner’s book or any text from physics.
56 4. Representations of Compact Groups

To finish this chapter, we briefly indicate another approach to the determination of the
representations of G = SO(3), which is a first example for the large theory of eigenspace
representations promoted in particular by Helgason.

From analysis it is well known that the three-dimensional euclidean Laplace operator

∆ = ∂x2 + ∂y2 + ∂z2

is SO(3) invariant, i.e. for f ∈ C ∞ (R3 ) we have

(∆f )(gx) = ∆f (x), for all g ∈ G, x = t (x, y, z) ∈ R3 .

Hence the space V of homogeneous polynomials of degree  in three variables, which

satisfy the Laplace equation
∆f = 0,
is invariant under the action f −→ f g with f g (x) = f (g −1 x) of G = SO(3). We denote
by D (g) the matrix form of the representation π of G on V given by π (g)f = f g .
To solve the Laplace equation, one usually introduces spherical coordinates (r, ϕ, ϑ) with

x = r sin ϑ cos ϕ,
y = r sin ϑ sin ϕ,
z = r cos ϑ.

Homogeneous polynomials f of degree  in x, y, z have the form f = r Y (ϑ, ϕ). If this f

is introduced into the Laplace equation written in the spherical coordinates

(∂r2 + (2/r)∂r + (1/r2 )(∂ϑ2 + cot ϑ ∂ϑ + (1/(sin ϑ)2 )∂ϕ2 ))f = 0

r drops out and the differential equation

(( + 1) + (∂ϑ2 + cot ϑ ∂ϑ + (1/(sin ϑ)2 )∂ϕ2 ))Y = 0

in the variables (ϑ, ϕ) results. The (2 + 1) linearly independent solutions Y = Ym (with
m = −, − + 1, . . . , , ) are known as spherical harmonics of the −th degree. They have
the form
Ym (ϑ, ϕ) = Φm (ϕ)Θm (ϑ)
where
1
Φm (ϕ) := √ eimϕ ,
2π
2 + 1 ( − m)! 1/2 dm
Θm (ϑ) := (−1)m ( ) (sin ϑ)m Pl (cos ϑ),
2 ( + m)! d(cos ϑ)m
2 + 1 ( − m)! 1/2 dm
Θ,−m (ϑ) := ( ) (sin ϑ)m Pl (cos ϑ),
2 ( + m)! d(cos ϑ)m

with m ≥ 0 in the last two equations. The P are the Legendre polynomials defined by

1 d 2
P (x) := (x − 1) .
2 ! dx
4.3 The Example G = SO(3) 57

This leads to the matrix form D for π with


π (g)r Ym (ϑ, ϕ) = 
Ym
(ϑ, ϕ)Dm
m (g).

m
=−

As one easily deduces, we have in particular

 imα
Dm
m (S3 (α)) = e δm
m .

By applying Schur’s commutation criterium, one can prove here, like we did already
above, that D is irreducible for all  = 0, 1, 2, . . . .
The conjugacy classes of SO(3) are parametrized by the angle α, α ∈ [0, π). For the
characters we obtain the formula


χ() (S3 (α)) = eimα = 1 + 2 cos α + · · · + 2 cos(α).
m=−

Again we can see that we got all irreducible representations as {cos nϕ}n∈N0 is a complete
system in L2 ([0, π)) by Fourier’s Theorem.
Chapter 5

Representations of Abelian
Groups

We have seen that

– finite groups have (up to equivalence) finitely many irreducible representations and
these are all finite-dimensional, and
– compact groups have (up to equivalence) denumerably many irreducible representa-
tions and these are also all finite-dimensional.
Abelian topological groups G (with prototype G = R or = C) are something of an ex-
treme in the opposite direction. Here we know that unitary irreducible representations
are all one-dimensional (we proved this as a consequence of the “Unitary Schur” in 1.5).
But in general the unitary dual is not denumerable. Since in this text we are mainly
interested in groups like SL(2, R) and Heis(R), where unitary irreducible representations
will turn out to be infinite-dimensional, we discuss here only briefly some notions of gen-
eral interest, following [BR] p.159 - 165 and [Ki] p.167 - 173.

5.1 Characters and the Pontrjagin Dual

Definition 5.1: A character of an abelian locally compact group G is a continuous
function χ : G −→ C, which satisfies
| χ(g) | = 1 and χ(gg  ) = χ(g)χ(g  ) for all g, g  ∈ G,
i.e. a character is a (one-dimensional) continuous irreducible unitary representation of G.

Remark 5.1: We already defined the unitary dual Ĝ of any group G as the set of
equivalence classes of unitary irreducible representations of G. For G abelian, Ĝ consists
just of all characters of G. It is also called the Pontrjagin dual of G.

Remark 5.2: For G abelian, Ĝ is also a group: We define the composition of two
characters χ and χ by
(χχ )(g) := χ(g)χ (g) for all g ∈ G.
If G is a topological group, Ĝ is also. More precisely, if G is locally compact (resp. com-
ˆ
pact, resp. discrete), Ĝ is locally compact (resp. discrete, resp. compact). One has G  Ĝ.
60 5. Representations of Abelian Groups

Example 5.1: For G = Rn every character χ has the form

χ(x) = exp(i(x̂1 x1 + · · · + x̂n xn )) = exp(i < x̂, x >), with x̂ ∈ Rn , for x ∈ Rn ,

i.e. we have Ĝ = G.
Example 5.2: For G = S 1 = {ζ ∈ C; | ζ | = 1} every character has the form

χ(ζ) = ζ k , k ∈ Z,

as we discussed already in 3.2. I.e., here we have Ĝ = Z.

Example 5.3: For G = Z, every character has the form

χ(m) = eimϕ , ϕ ∈ [0, 2π),

i.e. we have Ĝ = S 1 (and, hence, the self-duality of Z).

5.2 Continuous Decompositions

While studying representation theory for compact groups, we generalized the Fourier
series  2π

f (t) = ikt
c(k)e , c(k) := (1/(2π)) e−ikt f (t)dt
k∈Z 0

for G = SO(2)  S 1 to other compact groups. Now, the analogon of Fourier series for
G = R is the Fourier transform associating to a sufficiently nice function f its Fourier
transform
√  ∞
(5.1) ˆ
Ff (k) = f (k) := (1/ 2π) χk (t)f (t)dt, χk (t) := eikt .
−∞

This may be interpreted in the following way: In the case of compact groups we had a
(denumerable) direct sum decomposition of the regular representation ρ into irreducible
subrepresentations πj

ρ= πj .
j∈J

For G = R we analogously have the regular representation ρ given on H = L2 (R, dx) by

ρ(t)f (x) = f (x + t) for all x, t ∈ R.

Suppose H1 would be the space of a one-dimensional subrepresentation. Then we have

for every f1 ∈ H1

ρ(t)f1 (x) = f1 (x + t) = λ(t)f1 (x) for all x, t ∈ R.

Hence f1 has to be an exponential function and we get a contradiction, because L2 (R)

contains no exponential function = 0. But (5.1) indicates that ρ can be decomposed
into a direct integral of the irreducible one-dimensional representations χk , k ∈ R. For a
proper generalization one here needs more tools from functional analysis, in particular
replacement of the usual measure (as recalled in 3.3) by the notion of a spectral measure
E(.), i.e. a measure which takes operators as values and allows to decompose (linear)
5.2 Continuous Decompositions 61

operators into a direct integral. Following [BR] p.649 resp. [Ki] p.57, we present them
briefly as they are also essential for subsequent chapters.

Spectral Measure

Let [a, b] ⊂ R be a finite or infinite interval and E be a function on [a, b] with values E(λ)
for λ ∈ [a, b], which are operators in a Hilbert space H. E is called a spectral function if
it satisfies the following conditions:
i) E(λ) is self-adjoint for each λ,
ii) E(λ)E(µ) = E(min(λ, µ)),
iii) The operator function E is strongly right continuous, i.e. one has
lim E(λ + )u = E(λ)u for all u ∈ H,

→0+

iv) E(−∞) = 0, E(∞) = Id, i.e. one has

lim E(λ)u = 0, lim E(λ)u = u for all u ∈ H.
λ→−∞ λ→∞

Conditions i) and ii) mean that E(λ), λ ∈ [a, b], are (refining Definition 1.7) orthogonal
projection operators. For an interval ∆ = [λ1 , λ2 ] ⊂ [a, b] one denotes the difference
E(λ2 ) − E(λ1 ) by E(∆).
For ∆1 , ∆2 ⊂ [a, b] one has
E(∆1 )E(∆2 ) = E(∆1 ∩ ∆2 ),
in particular
E(∆1 )E(∆2 ) = 0 if ∆1 ∩ ∆2 = ∅,
i.e. the subspaces H1 = E(∆1 )H and H2 = E(∆2 )H are orthogonal.

Example 5.4: We take G = R, H = L2 (R) and π(a)u(x) := u(x + a) for u ∈ H. Then

 λ  ∞


R
λ −→ E(λ) := (2π)−1 dλ e−iλ a π(a)da
−∞ −∞

is a spectral function.

The properties of the spectral function imply that, for any u ∈ H, the function
σu (λ) := < E(λ)u, u >
is a right continuous, non-decreasing function of bounded variation with
σu (−∞) = 0, σu (∞) =  u 2 .
Moreover it is denumerable additive: For a pairwise disjoint decomposition {∆n } with
∆ = ∪∞
n=1 ∆n , one has
∞

∞

E(∆) = E(∆n ) and σu (∆) = σu (∆n ).

n=1 n=1

The function σu is called the spectral measure. We now can state the fundamental theo-
rem of spectral decomposition theory as follows:
62 5. Representations of Abelian Groups

Theorem 5.1: Every self-adjoint operator A in H has the representation

 ∞
A= λdE(λ).
−∞

Continuous Sums of Hilbert Spaces

The operation of direct sum of Hilbert spaces admits a generalization: Let there be given
a set X with measure µ and a family {Hx }x∈X of Hilbert spaces. It is natural to try to
define a new space 
H= Hx dµ(x),
X
the elements of which are to be functions f on X, assuming values in Hx for all x ∈ X,
and the scalar product being defined by the formula

< f1 , f2 > := < f1 (x), f2 (x) >Hx dµ(x).
X

The difficulty consists in the fact that the integrand in this expression may be non-
measurable. In the case where all of the spaces Hx are separable, one can introduce
an appropriate definition of measurable vector-functions f , which will guarantee this
measurability of the numerical function x −→< f1 (x), f2 (x) >: In the special example
that all Hx coincide with a fixed separable H0 with basis {ej }, the numerical functions
x −→< f (x), ej > are measurable (for more general cases see [Ki] p.57). Once an
appropriate concept of measurable vector-functions
 has been adopted one defines the
continuous sum (or direct integral ) H = X Hx dµ(x) as the set of equivalence classes of
measurable vector-functions f with summable square of norms.

An operator A in H = X Hx dµ(x) is called decomposable iff there exists a family {A(x)}
of operators in the spaces Hx such that

(Af )(x) = A(x)f (x) almost everywhere on X.

By a suitable generalization of the notion of spectral measure to the dual Ĝ (which for
an abelian locally compact group G can be given the structure of a topological space),
one has the general statement:

Theorem 5.2: Let π be a unitary continuous representation of an abelian locally com-

pact group G in a Hilbert space H. Then there exists a spectral measure E(.) on the
character group Ĝ such that 
π(g) = χ(g)dE(χ).
Ĝ
This theorem is called SNAG-theorem, as it goes back to Stone, Naimark, Ambrose, and
Godement. One can find a proof in [BR] p.160ff. We also recommend the Chapter on
locally compact commutative groups in [Ma] p.37ff, which is centered on the notion of
projection valued measure. This is a function E −→ PE associating a projection operator
PE in H to each Borel set E from a Borel space S with the properties
i) P∅ = 0, PS = Id,
ii) PE∩F =PE PF for all E and F,
iii) PE = j∈N PEj for every disjoint decomposition E = j∈N Ej .
Chapter 6

The Infinitesimal Method

Up to now, with exception of the Schrödinger representation of the Heisenberg group,

all irreducible representations treated here were finite-dimensional. Before we go further
into the construction of infinite-dimensional representations, we discuss a method which
allows a linearization of our objects and hence simplifies the task to determine the struc-
ture of the representations and classify them. As we shall see, this is helpful for compact
and noncompact groups as well. The idea is to associate to the linear topological group
G a linear object, its Lie algebra g = Lie G, and study representations π̂ of these alge-
bras, which are open to purely algebraic and combinatorial studies. One can associate
to each representation π of G an infinitesimal representation dπ of g. Conversely, one
can ask which representations π̂ may be integrated to a unitary representation π of G,
i.e., which are of the form π̂ = dπ, π ∈ Ĝ. As we will see in several examples, this
method allows us to classify the (unitary) irreducible representations and, hence, gives a
parametrization of Ĝ. It will further prove to be helpful for the construction of explicit
models for representations (π, H).

6.1 Lie Algebras and their Representations

We introduce some algebraic notions, which will become important in the sequel. The
interested reader can learn more about these in any text book on Algebra (e.g. Lang’s
Algebra [La]).

Definition 6.1: Let K be a field. A K-algebra is a K-vector space A provided with a

K-bilinear map (multiplication)

A × A −→ A, (x, y) −→ xy.

A is called

– associative iff one has

(xy)z = x(yz) for all x, y, z ∈ A,

– commutative iff one has

xy = yx for all x, y ∈ A,
64 6. The Infinitesimal Method

– a Lie algebra iff the multiplication xy, usually written as xy =: [x, y] (the Lie
bracket), is anti-symmetric, i.e.

[x, y] = −[y, x] for all x, y ∈ A,

and fulfills the Jacobi identity, i.e. one has

[x, [y, z]] + [y, [z, x]] + [z, [x, y]] = 0 for all x, y, z ∈ A.

– a Jordan algebra iff one has

xy = yx and x2 (xy) = x(x2 y) for all x, y ∈ A,

– a Poisson algebra iff A is a commutative ring with multiplication

A × A −→ A, (x, y) −→ xy

and one has a bilinear map

A × A −→ A, (x, y) −→ {x, y}

(the Poisson bracket) fulfilling the identities

{x, y} = −{y, x},

{x, yz} = {x, y}z + y{x, z},
{x, {y, z}} = {{x, y}, z} + {y, {x, z}} for all x, y, z ∈ A.

The dimension of A as a K-vector space is called the dimension of the algebra.

A map ϕ : A −→ A is an algebra homomorphism iff it is a K-linear map and respects

the respective composition, i.e. if one has

ϕ(xy) = ϕ(x)ϕ(y) for all x, y ∈ A.

Exercise 6.1: Show that C ∞ (R2n ) is a Poisson algebra with

n
∂f ∂g ∂f ∂g
{f, g} := ( − ) for all f, g ∈ C ∞ (R2n ),
i=1
∂q i ∂p i ∂p i ∂q i

where the coordinates are denoted by (q1 , . . . , qn , p1 , . . . , pn ) ∈ R2n .

Exercise 6.2: Verify that each Poisson algebra is also a Lie algebra.

In the following, we are mainly interested in Lie algebras over K = R or K = C, which

we usually denote by the letter g . We recommend as background any text book on this
topic, for instance Humphreys’ Introduction to Lie Algebras and Representation Theory
[Hu].

Example 6.1: Every Mn (K) is an associative noncommutative algebra with composi-

tion XY of two elements X, Y ∈ Mn (K) given by matrix multiplication.
6.1 Lie Algebras and their Representations 65

Example 6.2: Every associative algebra A is a Lie algebra with

[X, Y ] = XY − Y X for all X, Y ∈ A,

the commutator of X and Y . In particular, one denotes

gl(n, K) := Mn (K).

Example 6.3: If V is any K-vector space, the space End V of linear maps F from V to
V is a Lie algebra with

[F, G] := F G − GF for all F, G ∈ End V.

Example 6.4: If g is a Lie algebra and g0 is a subspace with [X, Y ] ∈ g0 for all X, Y ∈ g0 ,
g0 is again a Lie algebra. One easily verifies this and that

sl(n, K) := {X ∈ gl(n, K); Tr X = 0},

so(n) := {X ∈ gl(n, K); X = −t X},
su(n) := {X ∈ gl(n, K); X = −t X̄, Tr X = 0}

are examples of Lie algebras.

Example 6.5: If A is a K-algebra,

DerA := {D : A −→ A K − linear; D(ab) = a Db + Da b for all a, b ∈ A}

is a Lie algebra, the algebra of derivations.

Exercise 6.3: Verify that the matrices

1 1
F := ( ), G := ( ), H := ( )
1 −1

are a basis of sl2 (K) with the relations

[H, F ] = 2F, [H, G] = −2G, [F, G] = H.

Exercise 6.4: Verify that the matrices

−i −1 i
X1 := (1/2)( ), X2 := (1/2)( ), X3 := (1/2)( )
−i 1 −i

are a basis of su(2) with the relations

[Xi , Xj ] = −Xk , (i, j, k) = (1, 2, 3), (2, 3, 1), (3, 1, 2).

Exercise 6.5: Verify that the matrices

⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0 −1 −1
X1 := ⎝ −1 ⎠ , X2 := ⎝ 0 ⎠ , X3 := ⎝ 1 ⎠
1 1 0
66 6. The Infinitesimal Method

are a basis of so(3) with the relations

[Xi , Xj ] = −Xk , (i, j, k) = (1, 2, 3), (2, 3, 1), (3, 1, 2).

Exercise 6.6: Show that g = R3 provided with the usual vector product as composition
is a Lie algebra isomorphic to so(3).

Exercise 6.7: Show that g = R3 provided with the composition

((p, q, r), (p , q  , r )) −→ (0, 0, 2(pq  − p q))

is a Lie algebra, the Heisenberg algebra heis(R), with basis

P = (1, 0, 0), Q = (0, 1, 0), R = (0, 0, 1)

and relations
[R, P ] = [R, Q] = 0, [P, Q] = 2R.
Remark 6.1: It will soon be clear that gl(n, R), sl(n, R), so(n), su(n), and heis(R) are
associated to the groups GL(n, R), SL(n, R), SO(n), SU(n), resp. Heis(R).

In analogy with the concept of the linear representation of a group G in a vector space
V , one introduces the following notion.

Definition 6.2: A representation of the Lie algebra g in V is a Lie algebra homorphism

π̂ : g −→ End V , i.e. π̂ is K-linear and one has

π̂([X, Y ]) = [π̂(X), π̂(Y )] for all X, Y ∈ g.

Here we have no topology and, hence, no continuity condition.

Each Lie algebra has a trivial representation π̂ = π̂ 0 with π̂ 0 (X) = 0 for all X ∈ g and
an adjoint representation π̂ = ad given by

ad X(Y ) := [X, Y ] for all X, Y ∈ g.

Exercise 6.8: Verify this.

In analogy with the respective notions for group representations, we have the following
concepts.
– (π̂0 , V0 ) is a subrepresentation of (π̂, V ) iff V0 ⊂ V is a π̂-invariant subspace, i.e. π̂(X)v0 ∈
V0 for all v0 ∈ V0 , and π̂0 = π̂ |V .
– π̂ is irreducible iff π̂ has no nontrivial subrepresentation.
If π̂1 and π̂2 are representations of g we have
– the direct sum π̂1 ⊕ π̂2 as the representation on V1 ⊕ V2 given by

(π̂1 ⊕ π̂2 )(X)(v1 , v2 ) := (π̂1 (X)v1 , π̂2 (X)v2 ) for all v1 ∈ V1 , v2 ∈ V2 ,

and
– the tensor product π̂1 ⊗ π̂2 as the representation on V1 ⊗ V2 given by

(π̂1 ⊗ π̂2 )(X)(v1 ⊗ v2 ) := π̂1 (X)v1 ⊗ v2 + v1 ⊗ π̂2 (X)v2 for all v1 ∈ V1 , v2 ∈ V2 .

6.2 The Lie Algebra of a Linear Group 67

Exercise 6.9: Verify that these are in fact representations.

Exercise 6.10: Find the prescription to define a contragredient representation (π̂ ∗ , V ∗ )

to a given representation (π̂, V ) of g.

There is a lot of material about the structure and representations of Lie algebras, much
of it going back to Elie and Henri Cartan. As already indicated, a good reference for this
is the book by Humphreys [Hu]. We will later need some of this and develop parts of it.
Here we only mention that one comes up with facts like this: If g is semisimple (i.e. has
no nontrivial ideals), then each representation π̂ of g is completely reducible.
Before we discuss more examples, we establish the relation between linear groups and
their Lie algebras.

6.2 The Lie Algebra of a Linear Group

In Section 3.1 we introduced the concept of a linear group G as a closed subgroup of a
matrix group GL(n, R) or GL(n, C) for a suitable n ∈ N. To such a group we associate
a Lie algebra g = Lie G using as essential tool the exponential function for matrices. We
rely again on some experience from calculus in several complex variables and we define
the exponential function exp by
∞

exp X := (1/k!)X k for all X ∈ Mn (C).

k=0

Remark 6.2: One has the following facts:

1. exp X is absolutely convergent for each X ∈ Mn (C).

2. exp X : Mn (C) −→ Mn (C) is continuously differentiable.
3. There is an open neighbourhood U of 0 ∈ Mn (R), which is diffeomorphic to an open
neighbourhood V of E ∈ GL(n, R).
4. log X := Σ((−1)k /k)(X − E)k converges for  X − E  < 1 and we have

exp log X = X for  X − E  < 1,

log exp X = X for  exp X − E  < 1.
5. One has

i) exp(X + Y ) = exp X exp Y for commuting X, Y ∈ Mn (C),

ii) (exp X)−1 = exp(−X) for X ∈ Mn (C),
iii) A exp XA−1 = exp(AXA−1 ) for X ∈ Mn (C), A ∈ GL(n, C),
iv) det(exp X) = exp Tr X for X ∈ Mn (C).
The concientious reader should prove all this as Exercise 6.11. She/he should remember
that we introduced in (3.1) in Section 3.1

< X, Y > = Re Tr t X̄Y,  X  = < X, X >1/2 for all X, Y ∈ Mn (C),

and that one has the fundamental inequalities

 X + Y  ≤  X  +  Y , the triangle inequality,

|< X, Y >| ≤  X  Y , the Cauchy Schwarz inequality.
68 6. The Infinitesimal Method

Everyone interested in concrete examples could do the following Exercise 6.12 preparing
the ground for the next definition: Verify
   
1 t 1
exp(tX) = , for X = ,
1
   
cos t sin t 1
exp(tX) = , for X = ,
− sin t cos t −1
 t   
e 1
exp(tX) = , for X = ,
e−t −1
⎛ ⎞ ⎛ ⎞
cos t sin t 1
exp(tX) = ⎝ − sin t cos t ⎠ , for X = ⎝ −1 ⎠.
1 0
Exercise 6.13: Verify that one has the following relations
{exp(tX), X ∈ so(3); t ∈ R} = SO(3),
{exp(tX), X ∈ su(2); t ∈ R} = SU(2),
{exp(tX), X ∈ sl(2, R); t ∈ R} = SL(2, R).
This exercise shows that there are groups, which are examples for groups of exponential
type, i.e. images under the expontial map of their Lie algebras. For G = SL(2, R) one
has
{exp(tX), X ∈ sl(2, R), t ∈ R} = SL(2, R),
but the matrices exp(tX), X ∈ sl(2, R), t ∈ R, generate the identity component of
G = SL(2, R).

We take this as motivation to associate a Lie algebra to each linear group by a kind of
inversion of the exponential map:

Definition 6.3: Let G be a linear group contained in GL(n, K), K = R or K = C.

Then we denote

g = Lie G := {X ∈ Mn (K); exp(tX) ∈ G for all t ∈ R}.

Theorem 6.1: g is a real Lie algebra.

A hurried or less curious reader may simply believe this and waive the proof. (See for
instance [He] p.114: it is, in principle, not too difficult but lengthy. It shows why (in our
approach) we had to restrict ourselves to closed matrix groups.) We emphasize that it is
essentially the heart of the whole theory that not only a linear group but also every Lie
group has an associated Lie algebra. This is based on another important notion, which
we now introduce because it gives us a tool to determine explicitly the Lie algebras for
those groups we are particularly interested in (even if we we have skipped the proof of
the central theorem above).

Definition 6.4: A one-parameter subgroup γ of a topological group G is a continuous

homomorphism
γ : R −→ G, t −→ γ(t).
This notation is a bit misleading: to be precise, G0 = γ(R) is indeed a subgroup of G.
γ is often abbreviated as a 1-PUG (from the German term “Untergruppe”).
6.2 The Lie Algebra of a Linear Group 69

Remark 6.3: Each X ∈ g produces a 1-PUG γ, namely the subgroup given by

γ(t) := exp(tX).
This is obvious since exp(tX) is continuous (and even differentiable) in t and fulfills the
functional equation i) in part 5 of the Remark 6.2 above.

If the 1-PUG γ = γ(t) is differentiable in t, we assign to it an infinitesimal generator

X = Xγ by
d
X := γ(t) |t=0 .
dt
If γ is given as γ(t) = exp(tX), we have X as infinitesimal generator.

This suggests the following procedure to determine the Lie algebra g of a given matrix
group G (which is also the background of our Exercise 6.13 above): Look for “sufficiently
many independent” one-parameter subgroups γ1 , . . . , γd and determine their generators
X1 , . . . , Xd . Then g will come out as the R-vector space generated by these matrices Xi .
Obviously, one needs some explanation:
– Two 1-PUGs γ and γ  are independent iff their infinitesimal generators are linearly
independent over R as matrices in Mn (C).
– The number d of the sufficiently many independent (γi ) is just the dimension of our
group G as a topological manifold: Up to now we could work without this notion, but
later on we will even have to consider differentiable manifolds. Therefore, let us recall (or
introduce) that the notion of a real manifold M of dimension d includes the condition
that each point m ∈ M has a neighbourhood which is homeomorphic to an open set in
Rd . This amounts to the practical recipe that the dimension of a group G is the number
of real parameters needed to describe the elements of G.
For instance for G = SU(2) and SO(3) we have as parameters the three Euler angles
α, β, γ, which we introduced in 4.3 and 4.2. Hence, here we have d = 3, as also in the
next example.

Example 6.6: For G = SL(2, R) we take as a standard parametrization (a special

instance of the important Iwasawa decomposition)
a b
(6.1) SL(2, R)
g = ( ) = n(x)t(y)r(ϑ)
c d
with
1 x y 1/2 cos ϑ sin ϑ
n(x) := ( ), t(y) := ( ), r(ϑ) := ( ),
1 y −1/2 − sin ϑ cos ϑ
where x ∈ R, y > 0, ϑ ∈ [0, 2π). Therefore, we take
1 t et
γ1 (t) := ( ), γ2 (t) := ( ), γ3 (t) := r(t),
1 e−t
and get by differentiating and then putting t = 0
1 1 1
X1 = ( ), X2 = ( ), X3 = ( ).
−1 −1
Obviously, we have with the notation introduced in Exercise 6.3 in 6.1
Lie SL(2, R) = < X1 , X2 , X3 > = < F, G, H > = sl2 (R).
70 6. The Infinitesimal Method

Example 6.7: For

⎛ ⎞
1 x z
G = Heis (R) := {⎝ 1 y ⎠ ; x, y, z ∈ R}
1

we take
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 t 1 1 t
γ1 (t) := ⎝ 1 ⎠ , γ2 (t) := ⎝ 1 t ⎠ , γ3 (t) := ⎝ 1 ⎠,
1 1 1

and get ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1
X1 = ⎝ ⎠ , X2 = ⎝ 1 ⎠ , X3 = ⎝ ⎠.

By an easy computation of the commutators we get the Heisenberg commutation relations

[X1 , X2 ] = X3 , [X1 , X3 ] = [X2 , X3 ] = 0.

Exercise 6.14: Repeat this for G = Heis(R) as a subgroup of GL(4, R).

Exercise 6.15: Determine Lie G for G = {g = n(x)t(y); x ∈ R, y > 0}.

6.3 Derived Representations

Now that we have associated a Lie algebra g = Lie G to a given linear group (using a
differentiation procedure), we want to associate an algebra representation π̂ of g to a
given representation (π, H) of G. Here we will have to use differentiation again. To make
this possible, we have to modify the notion of a group representation:

Definition 6.5: Let (π, H) be a continuous representation of a linear group G. Then

its associated smooth representation (π, H∞ ) is the restriction of π to the space H∞ of
smooth vectors, i.e. those v ∈ H for which the map

G
g −→ π(g)v ∈ H

is differentiable.

This definition makes sense if we accept that a linear group is a differentiable manifold
and extend the notion of differentiability to vector valued functions. In our examples
things are fairly easy as we have functions depending on the parameters of the group
where differentiability is not difficult to check.

Definition 6.6: Let (π, H) be a continuous representation of a linear group G. Then its
associated derived representation is the representation dπ of g given on the space H∞
of smooth vectors by the prescription

d
dπ(X)v := π(exp(tX))v |t=0 for all v ∈ H∞ .
dt
6.3 Derived Representations 71

If it is clear, which representation π is meant, one abbreviates X̂v := dπ(X)v.

It is not difficult to verify that dπ(X) stays in H∞ and rather obvious that dπ(X) is
linear. But there is some more trouble to prove (see [Kn] p.53, [BR] p.320, or [FH] p.16)
that one has
[dπ(X), dπ(Y )] = dπ([X, Y ]) for all X, Y ∈ g.
Example 6.8: We take G = SU(2) and π = π1 given by

π(g)f (x, y) := f (āx − by, b̄x + ay)

for
f ∈ V (1) = < f−1 , f0 , f1 >
with √ √
f−1 (x, y) = (1/ 2)y 2 , f0 (x, y) = xy, f1 (x, y) = (1/ 2)x2 .
As we did in Exercise 6.4 in 6.1, we take as basis of su(2)

−i −1 i
X1 := (1/2)( ), X2 := (1/2)( ), X3 := (1/2)( ).
−i 1 −i

Then we have the one-parameter subgroups

cos (t/2) −i sin (t/2)

exp(tX1 ) = ( ),
−i sin (t/2) cos (t/2)
cos (t/2) − sin (t/2)
exp(tX2 ) = ( ),
sin (t/2) cos (t/2)
eit/2
exp(tX3 ) = ( ).
e−it/2

It is clear that V (1) consists of smooth vectors and hence we can compute

dt π(exp tX1 )f−1 |t=0

d
X̂1 f−1 =
√
= d
dt (1/ 2)(i sin(t/2)x + cos(t/2)y)2 |t=0
√ √
= (i/ 2)xy = (i/ 2)f0 ,

and
dt π(exp tX1 )f0 |t=0
d
X̂1 f0 =
= d
dt (cos(t/2)x + i sin(t/2)y)(i sin(t/2)x + cos(t/2)y) |t=0
√
= (i/2)(x2 + y 2 ) = (i/ 2)(f−1 + f1 ),
dt π(exp tX1 )f1 |t=0
d
X̂1 f1 =
√
= d
+ i sin(t/2)y)2 |t=0
dt (1/ 2)(cos(t/2)x
√ √
= (i/ 2)xy = (i/ 2)f0 .
Similarly we get
√ √ √
X̂2 f−1 = −(1/ 2)f0 , X̂2 f0 = (1/ 2)(f−1 − f1 ), X̂2 f1 = (1/ 2)f0 ,
X̂3 f−1 = if−1 , X̂3 f0 = 0, X̂3 f1 = −if−1 .
72 6. The Infinitesimal Method

Exercise 6.16: i) Verify this and show that for

√
X0 := iX3 , X± := (1/ 2)(−iX1 ∓ X2 )

one has
X̂0 fp = pfp , X̂± fp = fp±1 , p = −1, 0, 1 (f2 = f−2 = 0).
ii) Do the same for π = πj , j > 1.

Example 6.9: We take the Heisenberg group G = Heis(R) ⊂ GL(4, R) and the
Schrödinger representation π = πm , m ∈ R \ {0}, given by

π(g)f (x) = em (κ + (2x + λ)µ)f (x + λ)

with
g = (λ, µ, κ) ∈ G, x ∈ R, f ∈ H = L2 (R), em (u) := e2πimu .
Here H∞ is the (dense) subspace of differentiable functions in L2 (R). We realize the Lie
algebra g = heis(R) by the space spanned by the four-by-four matrices
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1
⎜ 1 ⎟ ⎜ 1 ⎟ ⎜ 1 ⎟
P =⎜⎝
⎟, Q = ⎜
⎠ ⎝
⎟, R = ⎜
⎠ ⎝
⎟.
⎠
−1

Then we have the Heisenberg commutation relations

[P, Q] = 2R, [R, P ] = [R, Q] = 0

and (using again the notation introduced in 0.1)

exp tP = (t, 0, 0), exp tQ = (0, t, 0), exp tR = (0, 0, t).

For differentiable f we compute

dt π(t, 0, 0)f |t=0 dt f (x + t) |t=0 = ∂x f,

d d
P̂ f = =
dt π(0, t, 0)f |t=0 dt e (2xt)f (x) |t=0 = 4πimxf (x),
d d m
Q̂f = =
dt π(0, 0, t)f |t=0 dt e (t)f (x) |t=0 = 2πimf (x),
d d m
R̂f = =

i.e. we have
P̂ = ∂x , Q̂ = 4πimx, R̂ = 2πim.
Exercise 6.17: i) Verify that these operators really fulfill the same commutation rela-
tions as P, Q, R.
2
ii) Take f0 (x) := e−πmx and

Ŷ0 := −iR̂, Ŷ± := (1/2)(P̂ ± iQ̂)

and determine
fp := Ŷ+p f0 , Ŷ− fp and Ŷ0 fp for p = 0, 1, 2, . . . .
These Examples show how to associate a Lie algebra representation to a given group
representation. Now the following question arises naturally: Is there always a nontrivial
subspace H∞ of smooth vectors in a given representation space H? We would even want
6.4 Unitarily Integrable Representations of sl(2, R) 73

that it is a dense subspace. The answer to this question is positive. This is easy in our
examples but rather delicate in general and one has to rely on work by Garding, Nelson
and Harish Chandra. It is not very surprising that equivalent group representations are
infinitesimally equivalent, i.e. their derived representations are equivalent algebra repre-
sentations. The converse is not true in general. But as stated in [Kn] p.209, at least for
semisimple groups one has that infinitesimally equivalent irreducible representations are
unitarily equivalent.
Taking these facts for granted, a discussion of the unitary irreducible representations of
a given linear group G can proceed as follows: We try to classify all irreducible repre-
sentations of the Lie algebra g = Lie G of G and determine those, which can be realized
by derivation from a unitary representation of the group. In general, this demands for
some more information about the structure of the Lie algebra at hand. Before going a
bit into this structure theory, we discuss in extenso three fundamental examples whose
understanding prepares the ground for the more general cases.

6.4 Unitarily Integrable Representations of sl(2, R)

A basic reference for this example is the book “SL(2, R)” [La1] by Serge Lang but in some
way or other our discussion can be found in nearly any text treating the representation
theory of semisimple algebras or groups.
As we already know from Exercise 6.3 in 6.1 and Example 6.5 in 6.2, g = sl(2, R) is the
Lie algebra of G = SL(2, R) and we have

g = sl(2, R) = {X ∈ M2 (R); Tr X = 0} = < F, G, H >

with
1 1
F =( ), G = ( ), H = ( )
1 −1
and the relations
[H, F ] = 2F, [H, G] = −2G, [F, G] = H.
The exercises at the end of the examples in the previous section should give a feeling
that it is convenient to complexify the Lie algebra, i.e. here we go over to

gc := g ⊗R C = < F, G, H >C = < X+ , X− , Z >C

where
1 ±i −i
(6.2) X± := (1/2)(H ± i(F + G)) = (1/2)( ), Z := −i(F − G) = ( )
±i −1 i
with
[Z, X± ] = ±2X± , [X+ , X− ] = Z.
This normalization will soon prove to be adequate and the operators X± will get a
meaning as ladder operators: We consider a representation π̂ of gc = < X± , Z > on a
C-vector space V =< vi >i∈I where we abbreviate

Xv := π̂(X)v for all v ∈ V, X ∈ gc .

We are looking for representations (π̂, V ) of gc , which may be integrated, i.e. which are
complexifications of derived representations dπ for a representation π of SL(2, R).
74 6. The Infinitesimal Method

This has the following consequence: π|K is a representation of K := SO(2). From

Example 3.1 in 3.2 we know that all irreducible unitary representations of SO(2) are
given by
χk (r(ϑ)) = eikϑ , k ∈ Z,
hence one has the derived representation dχk with
1
dχk (Y ) = ik for Y = ( ) = iZ.
−1
This motivates the idea to look for each k ∈ Z at the subspace of V consisting of the
eigenvectors with eigenvalue k with respect to the action of Z, namely
Vk := {v ∈ V ; Zv = kv}.
This space Vk is called a K-isotypic subspace of V . Each v ∈ Vk is said to have weight
k. Now we can explain the meaning of the term ladder operator: The isotypic spaces
Vk may be seen as rungs of a ladder numerated by the weight k and the operators X±
change the weights in a controlled way:

Remark 6.4: We have

X± Vk ⊂ Vk±2 .
Proof: For v ∈ Vk we put v := X± v and check that v  is again a Z-eigenvector of weight


k±2:
Zv  = ZX± v = (X± Z ± 2X± )v = X± (k ± 2)v = (k ± 2)v  .
Here we used the Lie algebra relation [Z, X± ] = ±2X± , which for the operators acting
on V is transformed into the relation
ZX± − X± Z = ±2X± .
This remark has immediate consequences:

i) If V = ΣVk is the space of an irreducible gc -representation, there are only nontrivial

Vk where k is throughout odd or even.
ii) If there is a Vk0 = {0}, then all Vk with k ≥ k0 or all Vk with k ≤ k0 have to be zero.
Hence, there can be only four different configurations for our irreducible gc -representations.
The nontrivial weights k, i.e. those for which Vk = {0}, compose
– case 1: a finite interval
n
Im := {m ≤ k ≤ n; m ≡ k ≡ n mod 2}
of even or odd integers,
– case 2: a half line with an upper bound
I n := {k ≤ n; k ≡ n mod 2}
consisting of even or odd integers,
– case 3: a half line with lower bound
Im := {m ≤ k; m ≡ k mod 2}
consisting of even or odd integers,
– case 4: a full chain of even or odd integers
I+ := {k ∈ 2Z} or I− := {k ∈ 2Z + 1}.
6.4 Unitarily Integrable Representations of sl(2, R) 75

s s
s s
s s
s s
s s
s s s
s s s
s case 3 s s
X+ 6 X−
s ? s
s s s
s s s

case 1 s case 2 s case 4 s

s s
s s
s s

It is quite natural to call an element 0 = v ∈ Vk with k = n in configuration 1 or 2 a

vector of highest weight and similarly for k = m in configuration 1 or 3 a vector of lowest
weight.

The following statement offers the decisive point for the reasoning that we really get
all representations by our procedure. It is a special case of a more general theorem by
Godement on spherical functions. It also is a first example for the famous and important
multiplicity-one statements.

Lemma: If π̂ is irreducible, we have dim Vk ≤ 1.

For a proof we refer to Lang [La1] p.24 or the article by van Dijk [vD].

We look at the configurations 1 or 2: Let 0 = v ∈ Vn be a vector of highest weight,

i.e. with
Zvn = nvn , X+ vn = 0,

and put
k
vn−2k := X− vn .
76 6. The Infinitesimal Method

As we want an irreducible representation and we have accepted to take the muliplicity-

one statement from the Lemma for granted, we can take this vn−2k as a generator for
Vn−2k with vn−2k = 0 for all k ∈ N0 in configuration 1 and all k with n − 2k ≥ m in
configuration 2. We check what X+ does to vn−2k :
Remark 6.5: For k as fixed above, we have
X+ vn−2k = ak vn−2k+2 with ak := kn − k(k − 1).
Proof: Since X+ vn = 0, we know that a0 = 0. As in the proof of Remark 6.4, for the
operators realizing the commutation relation [X+ , X− ] = Z we calculate
X+ vn−2 = X+ X− vn = (X− X+ + Z)vn = nvn .
Hence we get a1 = n. Then we verify by induction
X+ vn−2(k+1) = X+ X− vn−2k = (X− X+ + Z)vn−2k
= X− ((kn − k(k − 1))vn−2k+2 + (n − 2k)vn−2k
= ((k + 1)n − k(k + 1))vn−2k = ak+1 vn−2k .
We see that the irreducibility of π̂ forces ak to be nonzero for all k ∈ N in configuration
2 and for 2k ≤ n − m in configuration 1. Since ak = 0 iff n − (k − 1) = 0, we conclude
that n has to be negative in case 2 and m = −n with n ∈ N in case 1.
Just as well, we can start to treat the configurations 1 and 3 with a vector 0 = vm ∈ Vm
k
of lowest weight. Here we put vm+2k := X+ vm . In parallel to Remark 6.5 we get:

Remark 6.6: For k ∈ N0 in configuration 3 and m + 2k ≤ n in configuration 1, we have

X− vm+2k = bk vm+2k−2 with bk = −(km + k(k − 1)).
Exercise 6.18: Prove this.

As above, we see that we have m = −n with negative n in configuration 1 and that n has
to be positive in configuration 3. For instance, the following are possible configurations.

6 s

4 s

n=2 s m=2 s
s
X+ 6 X−
s ? case 2 case 3
s

m = −2 s n = −2 s

−4 s

case 1 −6 s
s
6.4 Unitarily Integrable Representations of sl(2, R) 77

In configuration 4 we have no distinguished vector of highest or lowest weight. In this

case we choose for each even (or odd) integer n a nonzero vn ∈ Vn , i.e. with Zvn = nvn ,
such that they are related as follows:

Remark 6.7: For all even (resp. odd) n ∈ Z we have

X± vn = a± ±
n vn±2 with an = (1/2)(s + 1 ± n),

where s ∈ C, such that a±n = 0 for all n ∈ 2Z (resp. 2Z + 1), i.e. s ∈ 2Z + 1 for n ∈ 2Z
and s ∈ 2Z for n ∈ 2Z + 1.

Proof: From irreducibility and multiplicity-one we conclude that each X± vn has to be

a nonzero multiple of vn±2 . We verify that the choice of a±
n is consistent with the com-
mutation relations:

[Z, X± ]vn = (ZX± − X± Z)vn = Za± ±

n vn±2 − an nvn±2
±
= ±2an vn±2
= ±2X± vn ,
and
[X+ , X− ]vn = Zvn = nvn .
Verify this as Exercise 6.19.

The conditions for s are obvious since we have a±

n = 0 iff s + 1 = ∓n.

We chose a symmetric procedure for our configuration 4. But as well we could have
j j
chosen any nonzero vn ∈ Vn , and then vn+2j := X+ vn , vn−2j := X− vn for j ∈ N would
constitute an equivalent representation.
Exercise 6.20: Check this.

We denote the representations obtained above as follows

– π̂s,+ or π̂s,− for the even resp. odd case in configuration 4 with s = 2Z + 1 for π̂s,+
and s = 2Z for π̂s,− , s ∈ C,
– π̂n+ for configuration 3 with lowest weight n, n ∈ N0 ,
– π̂n− for configuration 2 with highest weight −n, n ∈ N0 ,
– σ̂n for configuration 1 with highest weight n and lowest weight −n, n ∈ N0 .

Remark 6.8: If in π̂s,+ and π̂s,− we do not exclude the values of s as above, we get all
representations as sub - resp. quotient representations of these π̂s,± .

Proposition 6.1: We have the following equivalences

π̂s,+ ∼ π̂−s,+ and π̂s,− ∼ π̂−s,− .

All other representations listed above are inequivalent.

Proof: i) Let V = < vn > and V  = < vn > be representation spaces for π̂s,+ resp. π̂s
,+
and F : V −→ V  an intertwining operator. Because of Zvn = nvn , we have for each n

ZF vn = F Zvn = nF vn .
78 6. The Infinitesimal Method

Hence one has F vn ∈ Vn , i.e. F vn = bn vn with bn ∈ C. Moreover, from

X± F vn = F X± vn

we deduce
(s + 1 ± n)bn vn±2
 
= bn±2 (s + 1 ± n)vn±2 ,
i.e.
bn /bn−2 = (s + 1 − n)/(s + 1 − n)
and
bn+2 /bn = (s + 1 + n)/(s + 1 + n)
resp.
bn /bn−2 = (s + n − 1)/(s + n − 1).
Now we see that we have

(s + 1 − n)(s + n − 1) = (s + n − 1)(s + 1 − n), i. e. s = ±s.

The same conclusions can be made for π̂s,− .

ii) The inequivalence of the σ̂n ’s among themselves and with respect to all the other
representations is clear because the dimensions are different. The other inequivalences
have to be checked individually. As an example let V =< vn > and V  =< vn > be
spaces of representations π̂k+ and π̂s,+ and F : V −→ V  an intertwining operator. As
above, for k ≥ n we have the relation F vn = bn vn with bn ∈ C. Hence F is not surjective
and both representations are not equivalent. The same reasoning goes through in all the
other cases.
Exercise 6.21: Realize this.

Classification of the Unitarily Integrable Representations of sl(2, R)

At first, we derive a necessary condition for a general representation π̂ of a Lie algebra

g = Lie G to be integrable to a unitary representation π of G.

Proposition 6.2: If (π, H) is a unitary representation of G and X ∈ g = Lie G, the

operator X̂ := dπ(X) on H∞ is skew-Hermitian, i.e. one has X̂ = −X̂ ∗ .

Proof: For v, w ∈ H∞ , unitarity of π implies

< v, π(g)w > = < π(g −1 )v, w > for all g ∈ G

and hence, for t ∈ R, t = 0,

π(exp(tX))w − w π(exp(−tX))v − v
< v, i >=< i ,w > .
t −t
Recalling the definition
d π(exp tX)v − v
dπ(X)v := π(exp(tX))v |t=0 = lim ,
dt t→0 t
we get from the last equation in the limit t −→ 0

< v, i dπ(X)w > = − < i dπ(X)v, w > .

6.4 Unitarily Integrable Representations of sl(2, R) 79

Therefore we have
< v, i dπ(X)w > = − < v, i dπ(X)∗ w >,
i.e. X̂ = dπ(X) is skew-Hermitian.

Now, we apply this to g = sl(2, R). The Proposition leads to the condition:

Remark 6.9: We have dπ(X± )∗ = −dπ(X∓ ).

Proof: Using the notation from the beginning of this section we have
1 ±i
X± = (1/2)(H ± i(F + G)) = (1/2)( ) =: (X1 ± iX2 )
±i −1

with X1 = H, X2 = F + G ∈ sl(2, R). Hence, the skew-Hermiticity of X̂i leads to

(X̂1 ± iX̂2 )∗ = −(X̂1 ∓ iX̂2 ).

Now, if we ask for unitarity, we have to check whether we find a scalar product <, >
defined on the representation space V such that we get

< X̂± v, w > = − < v, X̂∓ w > for all v, w ∈ V.

For V = < vj >j∈J this condition implies

(6.3) < X̂+ vj , vj+2 > = − < vj , X̂− vj+2 > for all j ∈ J.

i) In configuration 1 and 2 we have

X̂− vj+2 = vj

and from Remark 6.5 for j = n − 2k

X̂+ vj = ak vj+2 with ak = k(n − k + 1),

i.e.
ak  vj+2 2 = −  vj 2 .
An equation like this is only possible for ak < 0, which is true for the ak in configuration
2 but not for configuration 1, where we have seen that all ak are positive.

Similarly Remark 6.6 shows that in configuration 3 the condition (6.3) is no obstacle.

Exercise 6.22: Verify this.

ii) Configuration 4 is more delicate: From Remark 6.7 we have

X̂± vj = a± ±
j vj±2 with aj = (1/2)(s + 1 ± j),

i.e. the condition (6.3) has the form

(6.4) (s̄ + 1 + j)  vj+2 2 = −(s − 1 − j)  vj 2 .

At first we treat the even case V =< vj >j∈2Z . Condition (6.4) requires for j = 0

(s̄ + 1)  v2 2 = −(s − 1)  v0 2 .
80 6. The Infinitesimal Method

As the norm has to be positive, a relation like this is only possible if one has

(1 − s)/(s̄ + 1) > 0, i. e. (1 − s)(1 + s) > 0.

We put s = σ + iτ, σ, τ ∈ R, and get the condition

s2 = σ 2 − τ 2 + 2στ i < 1,

i.e. unitarity demands for

σ=0 and τ ∈ R arbitrary

or

τ =0 and σ 2 < 1.
In the odd case V = < vj >j∈2Z+1 we get for j = −1 from (j  )

s̄  v1 2 = −s  v−1 2 .

This is possible only for

−s̄/s > 0, i. e. s̄2 = σ 2 − τ 2 − 2στ i < 0.

Therefore, here we have only one possibility, namely

σ=0 and τ ∈ R.

Putting all this together, we have proved the following statement.

Theorem 6.2: Unitary irreducible representations π of G = SL(2, R) can exist only if

their derived representation π̂ = dπ is among the following list:
– π̂ the trivial representation,
– π̂ = π̂k± , k ∈ N,
– π̂ = π̂is,± , s ∈ R (s = 0 for π̂is,− )
– π̂ = π̂s , s ∈ R, 0 < s2 < 1.

In the next chapter we will prove that, in fact, all these representations π̂ can be inte-
grated to unitary representations π of G. They get the following names
– π = πk∓ discrete series representation of highest (lowest) weight ∓k,
– π = πis,± even (odd ) principal series representation,
– π = πs complementary (or supplementary) series representation.

Because of the equivalences stated in Proposition 6.1, we have for the unitary dual of
G = SL(2, R)
Ĝ = {π0 ; πk± , k ∈ N; πis,± , s ∈ R>0 ; πs , 0 < s < 1}.
We will study this more closely in the next chapter but already indicate a kind of visu-
alization.
6.4 Unitarily Integrable Representations of sl(2, R) 81

iR

πis,+

πk− πk+
πs
s s s s s s s s s s s s s s s

−3 −2 −1 1 2 3

πis,−

We can observe a very nice fact, which will become important later and is the starting
point for a general procedure in the construction of representations:

Exercise 6.23: Show that the operator

Ω := X+ X− + X− X+ + (1/2)Z 2

acts as a scalar for the three types of representations in Theorem 6.2 and determine these
scalars.

Our infinitesimal considerations showed that there is no nontrivial finite-dimensional uni-

tary representation of SL(2, R). There are other proofs for this fact. Because it is so
elegant and instructive, we repeat the one from [KT] p.16.

Theorem 6.3: Every finite-dimensional representation π of G = SL(2, R) is trivial.

Proof: If G has an n-dimensional representation space, we have a homomorphism

π : SL(2, R) −→ U(n).

Since
a 1 b a −1 1 a2 b
( −1 )( )( −1 ) =( )
a 1 a 1
1
b
all n(b) = ( ) are conjugate. So all π(n(b)) with b > 0 are conjugate. It is a not too
1
difficult topological statement that in a compact group all conjugacy classes are closed.
As U(n) is compact, we can deduce

lim π(n(b)) = π(E2 ) = En ,

b→0+
82 6. The Infinitesimal Method

and therefore π(n(b)) = En for all b > 0. But we can repeat this conclusion for b < 0
1
and as well for n̄(b) = ( ) and get
b 1

π(n(b)) = π(n̄(b)) = En for all b ∈ R.

Now the proof is finished if we recall the following

Lemma: SL(2, R) is generated by the elements n(b) and n̄(b), b ∈ R.

Exercise 6.24: Prove this.

6.5 The Examples su(2) and heis(R)

In section 6.3 we computed derived representations for G = SU(2) and Heis(R). We look
at this again on the ground of what we did in the last section.
We refer to Exercise 6.4 in 6.1 and Example 6.8 in 6.3 and (having in mind later appli-
cations) change the basis (Xj ) of su(2) we used there to

−i −i −1
(6.5) H1 = ( ), H2 = ( ), H3 = ( ),
i −i 1

with
[Hj , Hk ] = 2Hl , (j, k, l) = (1, 2, 3), (2, 3, 1), (3, 1, 2).
su(2) and sl(2, R) have the same complexifications. Hence we can express the matrices
Z, X± from (6.2) in the previous section as

Z = iH3 , X± = (1/2)(iH1 ∓ H2 )

and repeat the discussion of the representation space V = < vj >j∈J to get the configu-
rations 1 to 4. Let us look at configuration 1: We take a vector vn ∈ V of highest weight
n ∈ N, put
k
vn−2k = X̂− vn , X̂+ vn−2k =: ak vn−2k+2 ,
get ak = kn − k(k − 1) and m = −n for the lowest weight. But in this case the skew-
hermiticity of the Hj changes things and Remark 6.9 in 6.4 is to be replaced by:

Remark 6.10: For a unitary representation π of SU(2) we must have

dπ(X± )∗ = dπ(X∓ ).

Hence, the discussion in 6.4 shows that for unitarity in this case the ak have to be
positive as, in fact, they are for n ≥ k ≥ 0. We see that our discussion rediscovers the
representation space V = V (j) from section 4.2 with n = 2j. By the way, we get here for
free another proof that the representation πj from 4.2 is irreducible.
Configurations 2 to 4 are not unitarizable.

For G = Heis(R), as in Example 6.8 in 6.3, we have

g = heis(R) = < P, Q, R >

6.5 The Examples su(2) and heis(R) 83

with the Heisenberg commutation relations

[P, Q] = 2R, [R, P ] = [R, Q] = 0.

We complexify g and take as a C-basis

Y0 := iR, Y± := (1/2)(P ± iQ)

to get the relations [Y+ , Y− ] = Y0 , [Y0 , Y± ] = 0.

We want to construct a representation π̂ of g on a space V = < vj >j∈J , (which we

continue linearly to a representation of gc ,) such that π̂ can be unitarily integrated,
i.e. we can find a unitary representation π of Heis(R) with dπ = π̂ : In every group one
has as a distinguished subgroup, the center C(G), defined by

C(G) := {g ∈ G; gg0 = g0 g for all g0 ∈ G}.

Obviously for G = Heis(R) we have

C := C(Heis(R)) = {(0, 0, κ); κ ∈ R}  R.

Hence, a representation π of our G restricted to the center C is built up (see Chapter 5)

from the characters of R. We write here

ψ(κ) := ψm (κ) := exp(2πimκ) = em (κ), m ∈ R \ {0}

with derived representation dψ(κ) = 2πim. Having this in mind (and the discussion of
Example 6.9 in 6.3), we propose the following construction of a representation (π̂, V ),
where we abbreviate again π̂(Y )v =: Y v.

We suppose that we have a vacuum vector, i.e. an element v0 ∈ V with

Y0 v0 = µv0 , Y− v0 = 0, µ = 2πm.

The first relation is inspired by differentiation of the prescription of the Schrödinger

representation
π((0, 0, κ))f = ψ(κ)f
where we have (0, 0, κ) = exp κR and Y0 = iR.

We put vj := Y+j v0 . As Y0 commutes with Y± , we have Y0 vj = µvj . Then we try to

realize a relation
Y− vj = aj vj−1 :
From the commutation relation [Y+ , Y− ] = Y0 we deduce

Y− v1 = Y− Y+ v0 = (Y+ Y− − Y0 )v0 = −µ,

Y− vj = Y− Y+ vj−1 = (Y+ Y− − Y0 )vj−1 = (aj−1 − µ)vj−1 = aj vj−1 ,

and, hence, by induction

aj = −jµ.
Thus we have V spanned by v0 , v1 , v2 , . . . with

Y0 vj = µvj , Y+ vj = vj+1 , Y− vj = −jµvj−1 .

84 6. The Infinitesimal Method

We check whether this representation can be unitarized: We look at the necessary con-
dition in Proposition 6.2 in 6.4. The skew-Hermiticity of P, Q, R leads to

Ŷ±∗ = dπ(Y± )∗ = (1/2)(P ∗ ∓ iQ∗ ) = −(1/2)(P ∓ iQ) = −Ŷ∓ .

Hence, as in the sl-discussion, we get the necessary condition for a scalar product <, >
on V
< Ŷ± v, w > = − < v, Ŷ± w >,
in particular, for v = vj−1 , w = vj , we have

 vj 2 = jµ  vj−1 2 .

We see that unitarity is possible iff µ = 2πm > 0. A model, i.e. a realization of this
representation π̂ by integration, is given by the Schrödinger representation πm , which we
discussed already in several occasions (see Example 6.9 and Exercise 6.17 in 6.3).

Remark 6.11: Our infinitesimal considerations show that πm is irreducible.

Remark 6.12: This discussion above is already a good part of the way to a proof of the
famous Stone-von Neumann Theorem stating that up to equivalence there is no other
irreducible unitary representation π of Heis(R) with π |C = ψm for m = 0 (for a complete
proof see e.g. [LV] p.19ff).

Remark 6.13: What we did here is essentially the same as that appears in the physics
2
literature under the heading of harmonic oscillator. Our v0 = f0 with f0 (x) = e−2πmx
describes the vacuum, and Y+ and Y− are creation resp. annihilation operators producing
higher resp. lower excitations.

Exercise 6.25: Use Exercise 6.17 in 6.3 to verify the equation

(Y+ Y− + Y− Y+ )f = (1/2)(f  − (4πmx)2 f ) = −2πmf

and show that f0 is a solution. Recover the Hermite polynomials.

6.6 Roots, Weights and some Structure Theory

In the examples discussed above in 6.4 and 6.5 we realized the representations of the
complexification gc of the given Lie algebra g = Lie G by ladder operators X± resp. Y±
changing generators vj of our representation space V by going to vj±2 resp. vj±1 in a
controlled way. For G = SL(2, R) and SU(2) the indices of the generators were related to
the weights, i.e. the eigenvalues of the operator Z. For G = Heis(R) we observe a differ-
ent behaviour: We have an operator Y0 producing the same eigenvalue when applied to
all generators of our representation space. This is expression of the fact that sl(2, R) and
su(2) resp. the groups SL(2, R) and SU(2) on the one hand and heis(R) resp. Heis(R) on
the other are examples of two different types for which we have to expect different ways
of generalizations. Let us already guess that we expect to generalize the machinery of the
weights in the first case (the semisimple one) by assuming to have a greater number of
commuting operators whose eigenvalues constitute the weights and to have two different
kinds of operators raising or lowering the weights (ordered, say, lexicographically).
6.6 Some Structure Theory 85

We have to introduce some more notions to get the tools for an appropriate structure
theory. The reader interested mainly in special examples and eager to see the repre-
sentation theory of the other groups mentioned in the introduction may want to skip
this section. Otherwise she or he is strongly recommended to a parallel study of a more
detailed source like [Hu], [HN] or [Ja]. Here we follow [Ki] p.92f, [KT], and [Kn1], which
don’t give proofs neither (for these see [Kn] p.113ff, and/or [Kn2] Ch.I - VI).

6.6.1 Specifications of Groups and Lie Algebras

In this section we mainly treat Lie algebras. But we start with some definitions for
groups, which often correspond to related definitions for algebras.

The reports [Kn1] and [KT] take as their central objects real or complex linear connected
reductive groups. The definition of a reductive group is not the same with all authors
but the following definition ([Kn] p.3, [KT] p.25) is very comprehensive and practical for
our purposes.

Definition 6.7: A linear connected reductive group is a closed connected group of real
or complex matrices that is stable under conjugate transpose, i.e. under the Cartan
involution
Θ : Mn (C) −→ Mn (C), X −→ (t X̄)−1 .
A linear connected semisimple group is a linear connected reductive group with finite
center.
More standard is the following definition: A group is called simple iff it is non-trivial,
and has no normal subgroups other than {e} and G itself.

Example 6.10: The following groups are reductive

a) G = GL(n, C),
b) G = SL(n, C),
c) G = SO(n, C),
En
d) G = Sp(n, C) := {g ∈ SL(2n, C); t gJg = J := ( )}.
−En
The center of GL(n, C) is isomorphic to C∗ , so GL(n, C) is not semisimple. The other
groups are semisimple (with exception n = 2 for c)). More examples come up by the
groups of real matrices in the above complex groups. GL(n, R) is disconnected and there-
fore not reductive in the sense of the above definition, however its identity component is.
We will come back to a classification scheme behind these examples in the discussion of
the corresponding notions for Lie algebras.

For the moment we follow another approach (as in [Ki] p.17). Non-commutative groups
can be classified according to the degree of their non-commutativity: Given two subsets
A and B of the group G, let [A, B] denote the set of all elements of the form aba−1 b−1
as a runs through A and b runs through B. We define two sequences of subgroups:

Let G0 := G and for n ∈ N let Gn be the subgroup generated by the set [Gn−1 , Gn−1 ].
Gn is called the n-th derived group of G.
86 6. The Infinitesimal Method

Let G0 := G and for n ∈ N assume Gn to be the subgroup generated by the set [G, Gn−1 ].
We obviously obtain the following inclusions
G = G0 ⊃ G 1 ⊃ · · · ⊃ G n ⊃ . . . ,
G = G0 ⊃ G 1 ⊃ · · · ⊃ Gn ⊃ . . . .
For a commutative group these sequences are trivial, we have Gn = Gn = {e} for all
n ∈ N.

Definition 6.8: G is called solvable of class k iff Gn = {e} beginning with n = k.

G is called nilpotent of class k iff Gn = {e} beginning with n = k.

Exercise 6.26: Show that Heis(R) is nilpotent and solvable of class 1.

Now we pass to the corresponding concepts for Lie algebras (following [Ki] p.89).

Definition 6.9: A linear subspace g0 in a Lie algebra is called a subalgebra iff

[X, Y ] ⊂ g0 for all X, Y ∈ g0 .
A linear subspace a in a Lie algebra is called an ideal iff
[X, Y ] ⊂ a for all X ∈ g and Y ∈ a.
If a is an ideal in g, then the factor space g/a is provided in a natural way with the
structure of a Lie algebra, which is called the factor algebra of g by a.

In every Lie algebra g we can define two sequences of subspaces, where here the expression
[a, b] denotes the linear hull of all [X, Y ], X ∈ a, Y ∈ b:
g1 := [g, g], g2 := [g, g1 ], . . . , gn+1 := [g, gn ],
g1 := [g, g], g2 := [g1 , g1 ], . . . , gn+1 := [gn , gn ].
It is clear that one has the following inclusions
gn ⊃ gn+1 , gn ⊃ gn+1 , gn ⊃ gn , n ∈ N.
Exercise 6.27: Prove that all gn and gn are ideals in g and that gn /gn+1 and gn /gn+1
are commutative.

For dim g < ∞ the sequences (gn ) and (gn ) must stabilize, beginning with a certain n
we have gn = gn+1 = · · · = g∞ and gn = gn+1 = · · · = g∞ .

Definition 6.10: g is called solvable resp. nilpotent iff g∞ = {0} resp. g∞ = {0}.

Obviously every nilpotent algebra is solvable.

Example 6.11: heis(R) is nilpotent, and hence solvable as we have more generally

bn (K) := {X = (xij ) ∈ Mn (K); xij = 0 for i > j} is solvable,

nn (K) := {X = (xij ) ∈ Mn (K); , xij = 0 for i ≥ j} is nilpotent.
It is clear that every X ∈ nn is nilpotent, i.e. has the property X n = 0. This is generalized
by the following statement justifying the name nilpotent algebra:
6.6 Some Structure Theory 87

Theorem 6.4 (Engel): g is nilpotent iff the operator ad X is nilpotent for all X ∈ g,

i.e. for every X there is an n ∈ N with (ad X)n = 0.
We recall that ad X is defined by ad X(Y ) := [X, Y ] for all Y ∈ g.

Now we pass to the other side of Lie algebra structure theory. As in [Kn1] p.1 ff we
restrict ourselves to the treatment of finite dimensional real or complex algebras g.

Definition 6.11: g is said to be simple iff g is non-abelian and g has no proper non-zero

ideals.

In this case one has [g, g] = g, which shows that we are as far from solvability as possible.

Definition 6.12: g is said to be semisimple iff g has no non-zero abelian ideal.

There are other (equivalent) definitions, for instance g is said to be semisimple iff one
has rad g = 0. In this definition the radical rad g is the sum of all solvable ideals of g.

Semisimple and simple algebras are related a follows.

Theorem 6.5: g is semisimple iff g is the sum of simple ideals. In this case there are
no other simple ideals, the direct sum decomposition is unique up to order of summands
and every ideal is the sum of the simple ideals. Also in this case [g, g] = g.

Finally, from [KT] p.29 we take over

Definition 6.13: A reductive Lie algebra is a Lie algebra that is the direct sum of two
ideals, one equal to a semisimple algebra and the other to an abelian Lie algebra.

We have a practical criterium:

Theorem 6.6: If g is a real Lie algebra of real or complex (even quaternion) matrices
closed under conjugate transposition, then g is reductive. If moreover the center of g is
trivial, i.e. Zg := {X ∈ g; [X, Y ] = 0 for all Y ∈ g} = 0, then g is semisimple.

Reductive Lie algebras have a very convenient property:

Proposition 6.3: A Lie algebra g is reductive iff each ideal a in g has a complementary

ideal, i.e. an ideal b with g = a ⊕ b.

As to be expected, to some extent the notions just defined for groups and algebras fit
together and one has statements like the following:

Proposition 6.4 ([KT] p.29): If G is a linear connected semisimple group, then g = Lie G
is semisimple. More generally, if G is linear connected reductive, then g is reductive with
g = Zg ⊕ [g, g] as a direct sum of ideals. Here Zg denotes the center of g, and the com-
mutator ideal is semisimple.

Example 6.12: gl(n, R) = {scalars} ⊕ sl(n, R).

88 6. The Infinitesimal Method

The main tool for the structure theory of semisimple Lie algebras is the following bilinear
form that was first defined by Killing and then extensively used by E. Cartan.

Definition 6.14: The Killing form B is the symmetric bilinear form on g defined by

B(X, Y ) := Tr (adXadY ) for all X, Y ∈ g.

Remark 6.14: B is invariant in the sense that

B([X, Y ], Z]) = B(X, [Y, Z]) for all X, Y, Z ∈ g.

Exercise 6.28: Determine the matrix of B with respect to the basis

a) F, G, H (from Exercise 6.3 in 6.1) for g = sl(2, R),
b) P, Q, R (from Exercise 6.7 in 6.1) for g = heis(R).

The starting point for structure theory of semisimple Lie algebras is Cartan’s Criterium
for Semisimplicity.

Theorem 6.7: g is semisimple if and only if the Killing form is nondegenerate, that is
B(X, Y ) = 0 for all Y ∈ g implies X = 0.

The proof uses the remarkable fact that ker B is an ideal in g.

There is another important and perhaps a bit more accessible bilinear form on g: The
trace form B0 is given by

B0 (X, Y ) := Tr (XY ) for all X, Y ∈ g.

The trace form is invariant in the same sense as the Killing form. Both forms are related
(see e.g. [Fog] IV.4):

Remark 6.15: We have

B(X, Y ) = 2nTr (XY ) for g = sl(n, K) and n ≥ 2,
= (n − 2)Tr (XY ) for g = so(n, K) and n ≥ 3,
= 2(n + 1)Tr (XY ) for g = sp(2n, K) and n ≥ 1.

A variant of the trace form already appeared in our definition of topology in matrix
spaces. We introduced for X, Y ∈ Mn (C)

< X, Y > := Re Tr t X̄Y, and  X 2 := Re Tr t X̄X.

As infinitesimal object corresponding to the Cartan involution for our matrix groups G

Θ : GL(n, C) −→ GL(n, C), g −→ (t ḡ)−1 ,

we have the map for the Lie algebra g = Lie G

θ : Mn (C) −→ Mn (C), X −→ −t X̄.

Hence we have
< X, Y > := −Re B0 (X, θY )
as a scalar product on g as a real vector space. This will become important later on.
6.6 Some Structure Theory 89

6.6.2 Structure Theory for Complex Semisimple Lie Algebras

A complete classification exists for semisimple Lie algebras, as one knows their building
blocks, the simple algebras: Over C there exist four infinite series of classical simple Lie
algebras and five exceptional simple Lie algebras. Over R we have 12 infinite series and
23 exceptional simple algebras. From all these, here we reproduce only the following
standard list of classical complex simple algebras

1) An := {X ∈ Mn+1 (C); Tr X = 0}, n = 1, 2, 3, . . .

2) Bn := {X ∈ M2n+1 (C); X = −t X}, n = 2, 3, . . .
En
3) Cn := {X ∈ M2n (C); XJ2n + J2n X = 0}, n = 3, 4, . . . , J2n := ( ),
−En
4) Dn := {X ∈ M2n (C); X = − X}, n = 4, 5, . . . .
t

We have the isomorphisms

B1  A1  C1 , C2  B2 , D2  A1 ⊕ A1 , D3  A3
and D1 is commutative. In the sequel we shall in most cases return to our previous
notation and write
An = sl(n + 1, C), Bn = so(2n + 1, C), Cn = sp(n, C), Dn = so(2n, C).
All these algebras are also simple algebras over R. The remaining real classical algebras
remain simple upon complexification. For a complete list see for instance [Ki] p.92/3.
By the way, we introduce here the following also otherwise important concept:

Definition 6.15: Given a complex Lie algebra g, a real algebra g0 is called a real form
of g iff (g0 )c := g0 ⊗ C = g.

Example 6.13: g = sl(2, C) has just two real forms g0 = sl(2, R) and su(2).
More about this is to be found in [Kn1] p.15.

The main ingredients for a classification scheme are the root and Dynkin diagrams. We
give a sketch starting by the case of complex semisimple algebras where the theory is
most accessible, and then go to compact real and finally briefly to noncompact real al-
gebras. The central tool to define roots and their diagrams is a certain distinguished
abelian subalgebra.

Definition 6.16: Let g be a complex semisimple Lie algebra. A Cartan subalgebra h is

a maximal abelian subspace of g in which every ad H, H ∈ h, is diagonizable.

There are other equivalent ways to characterize a Cartan subalgebra h, for instance (see
[Kn1] p.2): h is a nilpotent subalgebra whose normalizer Ng (h) satisfies
Ng (h) := {X ∈ g, [X, H] ∈ h for all H ∈ h} = h.
Each semisimple complex algebra has a Cartan subalgebra. Any two are conjugate via
Int g ([Kn1] p.24), where Int g is a certain analytic subgroup of GL(g) with Lie algebra
ad g. We refer to [Kn2] p.69/70 for the notion of an analytic subgroup, which comes up
quite naturally when one analyzes the relation between Lie algebras and Lie groups.
90 6. The Infinitesimal Method

The theory of Cartan subalgebras for the complex semisimple case extends to a complex
reductive Lie algebra g by just saying that the center of g is to be adjoined to a Cartan
subalgebra of the semisimple part of g.

Example 6.14: For g = sl(n, C) we have as Cartan subalgebra

h := {diagonal matrices in g}.

Using our Cartan subalgebra h, we generalize the decomposition we introduced at the

beginning of Section 6.4 for g = sl(2, C)

g = < Z0 > + < X+ > + < X− >,

[Z0 , X± ] = ±2X± , [X+ , X− ] = Z0 ,

to the root space decomposition 
g=h⊕ gα
α∈∆

as follows: For all H ∈ h the maps ad H are diagonizable, and, as the elements of h
all commute, they are simultaneously diagonizable (by the finite-dimensional spectral
theorem from Linear Algebra). Let V1 , . . . , V be the eigenspaces in g for the different
systems of eigentupels. If h := < H1 , . . . , Hr > and ad Hi acts as λij · id on Vj , define
a linear functional λj on h by λj (Hi ) = λij . If H := Σci Hi , then ad H acts on Vj by
multiplication with

ci λij = ci λj (Hi ) =: λj (H).

i i

In other words, ad h acts in simultaneously diagonal fashion on g and the simultaneous

eigenvalues are members of the dual vector space h∗ := HomC (h, C). There are finitely
many such simultaneous eigenvalues and we write

gλ := {X ∈ g; [H, X] = λ(H)X for all H ∈ h}

for the eigenspace corresponding to λ ∈ h∗ . The nonzero such λ are called roots and the
corresponding gλ a root space. The (finite) set of all roots is denoted by ∆.

The following are some elementary properties of root space decompositions (for proofs
see for instance [Kn2] II.4):

Proposition 6.5: We have

a)[gα , gβ ] ⊂ gα+β .
b) If α, β ∈ ∆ \ {0} and α + β = 0, then B(gα , gβ ) = 0, i.e. root spaces are orthogonal
with respect to the Killing form.
c) B is nonsingular on gα × g−α if α ∈ ∆.
d) −α ∈ ∆ if α ∈ ∆.
e) B |h×h is nondegenerate. We define Hα to be the element of h paired with α.
f) ∆ spans h∗ .

Some deeper properties of root space decompositions are assembled in the following state-
ment (see again [Kn2] II.4).
6.6 Some Structure Theory 91

Theorem 6.8: Root space decompositions have the following properties:

a) dim gα = 1 for α ∈ ∆.
b) nα ∈ ∆ for α ∈ ∆ and any integer n ≥ 2.
c) [gα , gβ ] = gα+β for α + β = 0.
d) The real subspace h0 of h on which all roots are real is a real form of h and B |h0 ×h0
is an inner product.

We transfer B| h0 ×h0 to the real span h∗0 of the roots obtaining a scalar product <, >
and a norm  . . It is not too difficult to see that for α ∈ ∆ there is an orthogonal
transformation sα given on h∗0 by

2 < ϕ, α >
sα (ϕ) := ϕ − α for ϕ ∈ h∗0 .
 α 2

sα is called a root reflection in α and the hyperplane α⊥ a mirror.

The analysis of the root space shows that it has very nice geometrical and combinatorical
properties. The abstraction of these is the following:

Definition 6.17: An abstract root system is a finite set ∆ of nonzero elements in a real
inner product space V such that
a) ∆ spans V ,
b) all reflections sα for α ∈ ∆ carry ∆ to itself,
c) 2 < β, α > /  α 2 ∈ Z for all α, β ∈ ∆.
The abstract root system is called reduced iff α ∈ ∆ implies 2α ∈ ∆.
And it is called reducible iff ∆ = ∆ ∪ ∆ with ∆ ⊥ ∆ , otherwise it is irreducible.

The root system of a complex semisimple Lie algebra g with respect to a Cartan subal-
gebra h forms a reduced abstract root system in h∗0 . And a semisimple Lie algebra g is
simple if the corresponding root system is irreducible.

Definition 6.18: The dimension of the underlying space V of an abstract root system
∆ is called its rank, and if ∆ is the root system of a semisimple Lie algebra g, we also
refer to r = dim h as the rank of g.

To give an illustration, we sketch the reduced root systems of rank 2.

case 1 case 2 case 3 case 4

6
J
]
 I 6
@ Q
kJ
]J

3
-

6
 J
- @ - Q   -

J @  Q

 J^ ?@
R

JJ
+
^Q
s
?
?
92 6. The Infinitesimal Method

Ordering of the Roots, Cartan Matrices, and Dynkin Diagrams

We fix an ordered basis α1 , . . . , αr of h∗0 and define λ = Σci αi to be positive if the first

nonzero ci is positive. The ordering comes from saying λ > µ if λ − µ is positive. Let
∆+ be the set of positive members in ∆.

A root α is called simple if α > 0 and α does not decompose as α = β1 + β2 with β1 and
β2 positive roots. And a root α is called reduced if (1/2)α is not a root.
Relative to a given simple system α1 , . . . , αr , the Cartan matrix C is the r × r-matrix
with entries
cij = 2 < αi , αj > /  αi 2 .

It has the following properties

a) cij ∈ Z for all i, j,
b) cii = 2 for all i,
c) cij ≤ 0 for all i = j,
d) cij = 0 iff cji = 0,
e) there exists a diagonal matrix D with positive diagonal entries such that DCD−1 is
symmetric positive definite.

An abstract Cartan matrix is a square matrix satisfying properties a) through e) as above.

To such a matrix we associate a diagram usually called a Dynkin diagram (historically
more correct is the term CDW-diagram, C indicating Coxeter and W pointing to Witt):
To the elements α1 , . . . , αr of a basis of a root system correspond bijectively r points of
a plane, which are also called α1 , . . . , αr . For i = j one joins the point αi to αj by cij cji
lines, which do not touch any αk , k = i, j. For cij = cji the lines are drawn as arrows
pointing in the direction to αj if cij < cji .

The main facts are that a Cartan matrix and equivalently the CDW-diagram determine
the Lie algebra uniquely (up to isomorphism). In particular this correspondence does
not depend on the choice of a basis of the root system.

Example 6.15: For g = sl(r + 1, C), r ≥ 1 and its standard Cartan subalgebra

h := {D(d1 , . . . , dr ); d1 , . . . , dr ∈ C, Σdi = 0}

we have the Cartan matrix

⎛ ⎞
2 −1
⎜ −1 2 ⎟
⎜ ⎟
⎜ • ⎟
C := ⎜
⎜
⎟
⎟
⎜ • ⎟
⎝ 2 −1 ⎠
−1 2

and the corresponding CDW-diagram

Ar r r r ... r r r r ≥ 1.
α1 α2 αr
6.6 Some Structure Theory 93

The diagrams for the other classical algebras are the following

Br r r r ... r r -
-r r ≥ 2,
α1 α2 αr−1 αr

Cr r r r ... r r
 r r ≥ 3,
α1 α2 αr−1 αr

r αr−1

Dr r r r ... r r r r ≥ 4.
α1 α2 αr−2 αr

The remaining exceptional graphs of indecomposable reduced root systems are the fol-
lowing

E6 r r r r r

E7 r r r r r r

E8 r r r r r r r

r r -
-r r r -r
-
F4 G2 -

6.6.3 Structure Theory for Compact Real Lie Algebras

Up to now, we treated complex Lie algebras. We go over to a real Lie algebra g0 and
(following [Kn1] p.16) we call a subalgebra h0 of g0 a Cartan subalgebra if its complexifi-
cation is a Cartan subalgebra of the complex algebra g = g0 ⊗ C. At first we look at the
compact case: If g0 is the Lie algebra of a compact Lie group G and if t0 is a maximal
abelian subspace of g0 , then t0 is a Cartan subalgebra. We have already discussed the
example g0 = su(2) in 4.2 and we will discuss g0 = su(3) in the next section and see that
we have dim t0 = 2 in this case. To illustrate the general setting, we collect some more
remarks about the background notions (from [Kn1] p.15):
94 6. The Infinitesimal Method

Theorem 6.9: If g0 is semisimple, then the following conditions are equivalent:

a) g0 is the Lie algebra of some compact Lie group.
b) Int g0 is compact.
c) The Killing form of g0 is negative definite.

If g is semisimple complex, a real form g0 of g is said to be compact if the equivalent

conditions of the theorem hold. Our main example su(n) is a compact form of sl(n, C).
The fundamental result is here:

Theorem 6.10: Each complex semisimple Lie algebra has a compact real form.

Another important topic is maximal tori.

Definition 6.19: Let G be a compact connected linear group. A maximal torus in G is

a subgroup T maximal with respect to the property of being compact connected abelian.

From [Kn2] Proposition 4.30, Theorem 4.34 and 4.36 we take over:

Theorem 6.11: If G is a compact connected linear group, then

a) the maximal tori in G are exactly the analytic subgoups corresponding to the maximal
abelian subalgebras of g0 = Lie G;
b) any two maximal abelian subalgebras of g0 are conjugate via Ad G and hence any two
maximal tori are conjugate via G.

Theorem 6.12: If G is compact connected and T a maximal torus, then each element
of G is conjugate to a member of T .

Example 6.16:
1. For G = SU(n) one has as a maximal torus, its Lie algebra and its complexified Lie
algebra
T = {D(eiϑ1 , . . . , eiϑn ); Σϑj = 0},
t0 = {D(iϑ1 , . . . , iϑn ); Σϑj = 0},
t = {D(z1 , . . . , zn ); Σzj = 0}.

2. For G = SO(2n) and SO(2n + 1) one has as a maximal torus T the diagonal block
matrices with 2 × 2-blocks r(ϑj ), j = 1, . . . , n, in the diagonal, resp. additionally 1 in the
second case.

We go back to the general case and use the notation we have introduced. In this setting,
we can form a root-space decomposition

g=t⊕ gα .
α∈∆

Each root is the complexified differential of a multiplicative character χα of the maximal

torus T that corresponds to t0 , with

Ad(t)X = χα (t)X for all X ∈ gα .

6.6 Some Structure Theory 95

Another central concept is the one of the Weyl group. We keep the notation used above:
G is compact connected, g0 = Lie G, T a maximal torus, t0 = Lie T , t = t0 ⊗ C, ∆(g, t) is
the set of roots, and B is the negative of a G invariant scalar product on g0 . We define
tR = it0 . As roots are real on tR , they are in t∗R . The form B, when extended to be
complex bilinear, is positive definite on tR , yielding a scalar product <, > on t∗R . Now, the
Weyl group W = W (∆(g, t)) is in this context defined as the group generated by the root
reflections sα , α ∈ ∆(g, t), given (as already fixed above) on t∗R by sα (λ) = λ − 2<λ,α>
|α|2 α.
This is a finite group, which also can be characterized like this:
One defines W (G, T ) as the quotient of the normalizer by the centralizer

W (G, T ) := NG (T )/ZG (T ).

By Corollary 4.52 in [Kn2] p.260 one has ZG (T ) = T and hence also the formula
W (G, T ) = NG (T )/T.

Theorem 6.13: The group W (G, T ), when considered as acting on t∗R , coincides with
W (∆(g, t)).

6.6.4 Structure Theory for Noncompact Real Lie Algebras

Now, in the final step, we briefly treat the general case of a real Lie algebra g0 of
a noncompact semisimple group G. We already introduced the Killing form B, the
Cartan involution θ with θX = −t X̄ for a matrix X, and the fact that if g0 consists of
real, complex or quaternion matrices and is closed under conjugate transpose, then it is
reductive. More generally we call here a Cartan involution θ any involution of g0 such
that the symmetric bilinear form

Bθ (X, Y ) := −B(X, θY )

is positive definite. Then we have the Cartan decomposition

g0 = k0 ⊕ p0 , k0 := {X ∈ g0 ; θX = X}, p0 := {X ∈ g0 ; θX = −X}

with the bracket relations

[k0 , k0 ] ⊆ k0 , [k0 , p0 ] ⊆ p0 , [p0 , p0 ] ⊆ k0 .

One has the following useful facts ([Kn 2] VI.2):

a) g0 has a Cartan involution.
b) Any two Cartan involutions of g0 are conjugate via Int g0 .
c) If g is a complex semisimple Lie algebra, then any two compact real forms of g are
conjugate via Int g.
d) If g is a complex semisimple Lie algebra and is considered as a real Lie algebra, then
the only Cartan involutions of g are the conjugations with respect to the compact real
forms of g.

The Cartan decomposition of Lie algebras has a global counterpart (see for instance
[Kn1] Theorem 4.3). Its most rudimentary form is the following:
96 6. The Infinitesimal Method

Proposition 6.6: For every A ∈ GL(n, C) resp. GL(n, R), there are S1 , S2 ∈ O(n)
resp. U(n) and a diagonal matrix D with real positive elements in the diagonal such that
A = S1 DS2 . This decomposition is not unique.

Proof as Exercise 6.29 (use the fact that to each positive definite matrix B there is a
unique positive definite matrix P with B = P 2 and/or see [He] p.73).

Restricted Roots

We want to find a way to describe Cartan subalgebras in our general situation: Let
G be semisimple linear group with Lie G = g0 , g = g0 ⊗ C, θ a Cartan involution of
g0 , and g0 = k0 ⊕ p0 the corresponding Cartan decomposition. Let B be the Killing
form (or more generally any nondegenerate symmetric invariant bilinear form on g0 with
B(θX, θY ) = B(X, Y ) such that Bθ (X, Y ) = −B(X, θY ) is positive definite).

Definition 6.20: Let a0 be a maximal abelian subspace of p0 . Restricted roots are the
nonzero λ ∈ a∗0 such that

(g0 )λ := {X ∈ g0 ; (ad H)X = λ(H)X for all H ∈ a0 } = {0}.

Let Σ be the set of restricted roots and m0 := Zk0 . We fix a basis of a0 and an associated
lexicographic ordering in a∗0 and define Σ+ as the set of positive resticted roots. Then

n0 = (g0 )λ
λ∈Σ+

is a nilpotent Lie subalgebra, and we have the following Iwasawa decomposition:

Theorem 6.14: The semisimple Lie algebra g0 is a vector space direct sum

g0 = k0 ⊕ a0 ⊕ n0 .

Here a0 is abelian, n0 is nilpotent, a0 ⊕ n0 is a solvable subalgebra of g0 , and a0 ⊕ n0 has

[a0 ⊕ n0 , a0 ⊕ n0 ] = n0 .

For a proof see for instance [Kn2] Proposition 6.43. There is also a global version ([Kn2]
Theorem 6.46) stating (roughly) that one has a diffeomorphism of the corresponding
groups K × A × N −→ G given by (k, a, n) −→ kan. We already know all this in our
standard example G = SL(2, R): If we take θX = −t X, we have

g0 = sl(2, R) = k0 ⊕ a0 ⊕ n0 ,

with
1 1 1
k0 =< ( ) >, a0 =< ( ) >, n0 =< ( )>
−1 −1
and as in Example 6.6 in 6.2

G = KAN with K = SO(2), A = {t(y); y > 0 }, N = {n(x ); x ∈ R}.

6.6 Some Structure Theory 97

As to be expected, roots and restricted roots are related to each other. If t0 is a maximal
abelian subspace of m0 := Zk0 (a0 ), then h0 := a0 ⊕ t0 is a Cartan subalgebra of g0 in
the sense we defined at the beginning ([Kn2] Proposition 6.47). Roots are real valued on
a0 and imaginary valued on t0 . The nonzero restrictions to a0 of the roots turn out to
be restricted roots. Roots and restricted roots can be ordered compatibly by taking a0
before it0 . Cartan subalgebras in this setting are not always unique up to conjugacy.

1 1
Exercise 6.30: Show that h0 :=< ( ) > and h0 :=< ( ) > are Cartan
1 −1
subagebras of g0 = sl(2, R) and determine the corresponding Iwasawa decomposition.

Every Cartan subalgebra of g0 is conjugate (via Intg0 ) to this h0 or the h0 above.

6.6.5 Representations of Highest Weight

After this excursion into structure theory of complex and real Lie algebras, we finally
come back to show a bit more of the general representation theory, which is behind the
examples discussed in our sections 6.4 and 6.5 and which we shall apply in our next
section to the example g0 = su(3). We start by following the presentation in [Kn1] p.8.
Let at first g be a complex Lie algebra, h a Cartan subalgebra, ∆ = ∆(g, h) the set of
roots, h0 the real form of h where roots are real valued, B the Killing form (or a more
general form as explained above), and Hλ ∈ h0 corresponding to λ ∈ h∗0 .
Let π̂ : g −→ End V be a representation. For λ ∈ h∗ we put

Vλ := {v ∈ V ; (π̂(H) − λ(H)1)n v = 0 for all H ∈ h and some n = n(H, V ) ∈ N}.

If Vλ = {0}, Vλ is called a generalized weight space and λ a weight. If V is finite

dimensional, V is the direct sum of its generalized weight spaces. This is a generalization
of the fact from linear algebra about eigenspace decompositions of a linear transformation
on a finite-dimensional vector space. If λ is a weight, then the subspace

Vλ0 := {v ∈ V ; π̂(H)v = λ(H)v for all H ∈ h}

is nonzero and called the weight space corresponding to λ. One introduces a lexicographic
ordering among the weights and hence has the notion of highest and lowest weights. The
set of weights belonging to a representation π̂ is denoted by Γ(π̂). For dim h = 1 we
have h  h∗  C and so the weights are simply the complex numbers we met in our
examples in 6.4 and 6.5, for instance for g = sl(2, C) we found representations σ̂N with
dim VN = N + 1 and weights −N, −N + 2, . . . , N − 2, N . In continuation of this, we
treat a more general example.

Example 6.17: Let G = SU(n) and, hence, g = su(n)c = sl(n, C) and the Cartan
subalgebra the diagonal subalgebra h :=< H; Tr H = 0 >. We choose as generators of h
the matrices Hj := Ejj − Ej+1,j+1 and (slightly misusing) we write also Hj = ej − ej+1
where ej denote the canonical basis vectors of Cn−1  h. We have three natural types
of representations :
At first, let V be the space of homogeneous polynomials P of degree N in z1 , . . . , zn and
their conjugates and take the action of g ∈ SL(n, C) given by

(π(g)P )(z, z̄) := P (g −1 z, ḡ −1 z̄), z = t (z1 , . . . , zn ).

98 6. The Infinitesimal Method

Hence for H = D(iϑ1 , . . . .iϑn ) with ϑj ∈ R, Σϑj = 0, we come up with

n
(dπ(H)P )(z, z̄) = (−iϑj zj )∂zj P (z, z̄) + (−iϑj z̄j )∂z̄j P (z, z̄).
j=1 j=1

If P is a monomial

n
P (z, z̄) = z1k1 · · · · · znkn · z̄1l1 · · · · · z̄nln with (kj + lj ) = N,
j=1

then we get

n
dπ(H)P = (lj − kj )(iϑj )P.
j=1

Exercise 6.31: Describe the weights for this representation with respect to the lexico-
graphic ordering for the basis elements of h0 =< iH1 , . . . , iHn−1 >R . Do this also for the
two other “natural” representations, namely on the subspaces V1 of holomorphic and V2
of antiholomorphic polynomials in V (i.e. polyomials only in z1 , . . . , zn resp. z̄1 , . . . , z̄n ).
We will come back to this for the case n = 3 in the next section.

In our examples g0 = sl(2, R) and g0 = su(2) in 6.4 resp. 6.5 we saw that the finite-
dimensionality of the representation space V was equivalent to an integrality condition
for the weight. Now we want to look at the general case where the weights are l-tupels
if one fixes a basis of a Cartan subalgebra h resp. its dual h∗ and the real form h∗0 .

Definition 6.21: λ ∈ h∗ is said to be algebraically integral if

2 < λ, α > /|α|2 ∈ Z for all α ∈ ∆.

Then as a generalization of what we saw in 6.4 and 6.5, we have the elementary properties
of the weights for a finite-dimensional representation π̂ on a vector space V :

Proposition 6.7: a) π̂(h) acts diagonally on V , so that every generalized weight vector
is a weight vector and V is the direct sum of all weight spaces.
b) Every weight is real valued on h0 and algebraically integral.
c) Roots and weights are related by π̂(gα )Vλ ⊂ Vλ+α .

We fix a lexicographical ordering and take ∆+ to be the set of positive roots with

Π = {α1 , . . . , αl } as the corresponding simple system of roots. We say that λ ∈ V is
dominant if < λ, αj > ≥ 0 for all αj ∈ Π. The central fact is here the beautiful Theorem
of the Highest Weight ([Kn2] Th. 5.5):

Theorem 6.15: The irreducible finite-dimensional representations π̂ of g stand (up to

equivalence) in one-one correspondence with the algebraically integral dominant linear
functionals λ on h, the correspondence being that λ is the highest weight of π̂λ .
The highest weight λ of π̂λ has these additional properties:
a) λ depends only on the simple system Π and not on the ordering used to define Π.
b) The weight space Vλ for λ is one-dimensional.
c) Each root vector Eα for arbitrary α ∈ ∆+ annihilates the members of Vλ , and the
members of Vλ are the only vectors with this property.
6.6 Some Structure Theory 99

l
d) Every weight of π̂λ is of the form λ − i=1 ni αi with ni ∈ N0 and αi ∈ Π.
e) Each weight space Vµ for π̂λ has dimVwµ = dim Vµ for all w in the Weyl group W (∆),
and each weight µ has | λ |≥| µ | with equality only if µ is in the orbit W (∆)λ.

We already introduced in 3.2 the concept of complete reducibility. Here we can state the
following fact ([Kn2] Th. 5.29).

Theorem 6.16: Let π̂ be a complex linear representation of g on a finite-dimensional

complex vector space V . Then V is completely reducible in the sense that there exist
invariant subspaces U1 , . . . , Ur of V such that V = U1 ⊕ · · · ⊕ Ur and such that the
restriction of the representation to each Ui is irreducible.

The proofs of these two theorems use three tools, which are useful also in other contexts:
– Universal enveloping algebras,
– Casimir elements,
– Verma modules.
Again following [Kn1] p.10f, we briefly present these as they will help us to a better
understanding in our later chapters.

The Universal Enveloping Algebra

This is a general and far reaching concept applicable for any complex Lie algebra g:

We take the tensor algebra

T (g) := C ⊕ g ⊕ (g ⊗ g) ⊕ . . . .

and the two-sided ideal a in T (g) generated by all

X ⊗ Y − Y ⊗ X − [X, Y ], X, Y ∈ T 1 (g).

Here T 1 (g) denotes the space of first order tensors.

Then the universal enveloping algebra is the associative algebra with identity given by

U (g) := T (g)/a.

This formal definition has the practical consequence that U (g) consists of sums of mono-
mials usually written (slightly misusing) as a(j) X1j1 . . . Xnjn , a(j) ∈ C, X1 , . . . , Xn ∈ g
with the usual addition and a multiplication coming up from the Lie algebra relations
by Xi Xj = Xj Xi + [Xi , Xj ]. More carefully and formally correct, one has the following:
Let ι : g −→ U (g) be the composition of natural maps

ι : g  T 1 (g) → T (g) −→ U (g),

so that
ι([X, Y ]) = ι(X)ι(Y ) − ι(Y )ι(X).
100 6. The Infinitesimal Method

ι is in fact injective as a consequence of the fundamental Poincaré-Birkhoff-Witt Theorem:

Theorem 6.17: Let {Xi }i∈I be a basis of g, and suppose that a simple ordering has
been imposed on the index set I. Then the set of all monomials

(ιXi1 )j1 · · · · · (ιXin )jn

with i1 < · · · < in and with all jk ≥ 0, is a basis of U (g). In particular the canonical
map ι : g −→ U (g) is one-to-one.

The name universal enveloping algebra stems from the following universal mapping prop-
erty.

Theorem 6.18: Whenever A is a complex associative algebra with identity and we have
a linear mapping ϕ : g −→ A such that

ϕ(X)ϕ(Y ) − ϕ(Y )ϕ(X) = ϕ([X, Y ]) for all X, Y ∈ g,

then there exists a unique algebra homomorphism ϕ̃ : U (g) −→ A such that ϕ̃(1) = 1
and ϕ = ϕ̃ ◦ ι.

The map ϕ̃ from the theorem may be thought of as an extension of ϕ from g to all of
U (g). This leads to the following useful statement.

Theorem 6.19: Representations of g on complex vector spaces stand in one-one corre-

spondence with left U (g) modules in which 1 acts as the identity.
This fact is essential for the construction of representations and implicit in our examples
in 6.4 and 6.5.

We now come back to the case that g is semisimple and to the notation introduced above.
We enumerate the positive roots as α1 , . . . , αm and we let H1 , . . . , Hl be a basis of h.
Then for the construction of highest weight representations it is appropriate to use the
ordered basis

E−α1 , . . . , E−αm , H1 , . . . , Hl , Eα1 , . . . , Eαm

in the Poincaré-Birkhoff-Witt Theorem. The theorem says that

p1 pm
E−α 1
. . . E−α m
H1k1 . . . Hlkl Eαq11 . . . Eαqm
m

is a basis of U (g). If one applies members of this basis to a nonzero highest weight vector
v0 of V, one gets control of a general member of U (g)v0 : Eαq11 . . . Eαqmm
will act as 0 if
q1 + · · · + qm > 0, and H1 . . . Hl will act as a scalar. Thus one has only to sort out
k1 kl
p1 pm
the effect of E−α 1
. . . E−α m
and most of the conclusions of the Theorem of the Highest
Weight follow readily. If one looks at our examples in 6.4 and 6.5, one can get an idea
how this works even in the case of representations, which are not finite-dimensional, and
how the integrality of the weight leads to finite-dimensionality.
6.6 Some Structure Theory 101

The Casimir Element

For a complex semisimple Lie algebra g with Killing form B, the Casimir element Ω is
the element

Ω0 := B(Xi , Xj )X̃i X̃j

i,j

of U (g), where (Xi ) is a basis of g and (X̃i ) is the dual basis relative to B. One can show
that Ω0 is defined independently of the basis (Xi ) and is an element of the center Z(g)
of U (g).

Exercise 6.32: Check this for the case g = sl(2, C) and determine the Casimir element.

In the general case one has the following statement.

Theorem 6.20: Let Ω0 be the Casimir element, (Hi )i=1,.,l an orthogonal basis of h0
relative to B, and choose root vectors Eα so that B(Eα , E−α ) = 1 for all roots α. Then
l 
a) Ω0 = i=1 Hi2 + α∈∆ Eα E−α .
b) Ω0 operates by the scalar |λ|2 + 2 < λ, δ > = |λ + δ|2 − |δ|2 in an irreducible finite-
dimensional representation of g of highest weight λ, where δ is half the sum of the positive
roots.
c) The scalar by which Ω0 operates in an irreducible finite-dimensional representation of
g is nonzero if the representation is not trivial.

The main point is that ker Ω0 is an invariant subspace of V if V is not irreducible.

Remark 6.16: The center of U (g) is important also in the context of the determination
of infinite-dimensional representations. We observed in Exercise 6.23 in 6.4 that Ω, a
multiple of the Casimir element Ω0 , acts as a scalar for the unitary irreducible represen-
tations of sl(2, R). For more general information we recommend [Kn] p.214 where as a
special case of Corollary 8.14 one finds the statement: If π is unitary, then each mem-
ber of the center Z(gc ) of U (gc ) acts as a scalar operator on the K−finite vectors of π.
We will come back to this in 7.2 while explicitly constructing representations of SL(2, R).

The Verma Module

We fix a lexicographic ordering and introduce


b := h ⊕ gα .
α>0

For ν ∈ h∗ , make C into a one-dimensional U (b) module

 Cν by defining the action of
H ∈ h by Hz = ν(H)z for z ∈ C and the action of α>0 gα by zero. For µ ∈ h∗ , we
define the Verma module V (µ) by

V (µ) := U (g) ⊗U (h) Cµ−δ .

where δ is again half the sum of the positive roots and this term is introduced to simplify
calculations with the Weyl group.
102 6. The Infinitesimal Method

Verma modules are essential for the construction of representations. They have the fol-
lowing elementary properties:
a) V (µ) = 0.
b) V (µ) is a universal highest weight module for highest weight modules of U (g) with
highest weight µ − δ.
c) Each weight space of V (µ) is finite-dimensional.
d) V (µ) has a unique irreducible quotient L(µ).

If λ is dominant and algebraically integral, then L(λ + δ) is the irreducible representation

of highest weight λ looked for in the theorem of the Highest Weight (Theorem 6.15).

In our treatment of finite and compact groups we already saw the effectiveness of the
theory of characters of representations. As for the moment we look at finite-dimensional
representations we can use characters here too. To allow for more generalization, we
treat them for now as formal exponential sums (again following [Kn1] p.12/3):
Let again g be a semisimple Lie algebra, h a Cartan subalgebra, ∆ a set of roots provided
with a lexicographic ordering, α1 , . . . , αl the simple roots, and W (∆) the Weyl group.
∗
We regard the set Zh of functions f from h∗ to Z as an abelian group under pointwise
addition. We write

f= f (λ)eλ .
λ∈h∗

∗
The support of f is defined to be the set of λ ∈ h∗ for which f (λ) = 0. Within Zh ,
let Z[h∗ ] be the subgroup of all f of finite support. The subgroup Z[h∗ ] has a natural
commutative ring structure, which is determined by eλ eµ = eλ+µ . Moreover, we introduce
a larger ring Z < h∗ > with
∗
Z[h∗ ] ⊆ Z < h∗ > ⊆ Zh
∗
consisting of all f ∈ Zh whose support is contained in the union of finitely many sets
νi − Q+ , νi ∈ h∗ and

l
Q+ := { ni αi ; ni ∈ N0 }.
i=1

Multiplication in Z < h∗ > is given by

( cλ eλ )( cµ eµ ) := ( cλ cµ )eν .
λ∈h∗ µ∈h∗ λ∈h∗ λ+µ=ν

If V is a representation of g (not necessarily finite-dimensional), one says that V has a

character if V is the direct sum of its weight spaces under h, i.e., V = ⊕µ∈h∗ Vµ , and if
dim Vµ < ∞ for µ ∈ h∗ . In this case the character is

char(V ) := (dim Vµ )eµ

µ∈h∗

∗
as an element of Zh . This definition is meaningful if V is finite-dimensional or if V is a
Verma module.
6.6 Some Structure Theory 103

We have two more important notions:

The Weyl denominator is the element d ∈ Z[h∗ ] given by

d := eδ Πα∈∆+ (1 − e−α ).

Here δ is again half the sum of the positive roots.

The Kostant partition function P is the function from Q+ to N that tells the number of
ways, apart from order, that a member of Q+ can be written as the sum of positve roots.
We put P(0) = 1 and define K := Σγ∈Q+ P(γ)e−γ ∈ Z < h∗ >. Then one can prove that
one has Ke−δ d = 1 in the ring Z < h∗ > , hence d−1 ∈ Z < h∗ >. Then we have as the
last main theorem in this context the famous Weyl Character Formula:

Theorem 6.21: Let (π, V ) be an irreducible finite-dimensional representation of the

complex semisimple Lie algebra g with highest weight λ. Then

char(V ) = d−1 (det w)ew(λ+δ) .

w∈W (∆)

Now we leave the treatment of the complex semisimple case and go over to the follow-
ing situation: G is compact connected, g0 := Lie G, g the complexification of g0 , T a
maximal torus, t0 := Lie T , ∆(g, t) the set of roots, and B the negative of a G invariant
inner product on g0 , and tR := it0 . As we know, roots are real on tR , hence are in t∗R .
The form B, when extended to be complex bilinear, is positive definite on tR , yielding an
inner product <, > on t∗R . W (∆(g, t)) is the Weyl group generated by the root reflections
sα for α ∈ ∆(g, t). Besides the notion of algebraic integrality already exploited in the
complex semisimple case above, we have here still another notion of integrality:
We say that λ ∈ t∗ is analytically integral if the following equivalent conditions hold:
1) Whenever H ∈ t0 satisfies exp H = 1, then λ(H) is in 2πiZ.
2) There is a multiplicative character ψλ of T with ψλ (exp H) = eλ(H) for all H ∈ t0 .
In [Kn1] p.18 one finds a list of properties of these notions. We cite part of it:
a) Weights of finite-dimensional representations of G are analytically integral. In partic-
ular every root is analytically integral.
b) Analytically integral implies algebraically integral.
c) If G is simply connected and semisimple, then algebraically integral implies analyti-
cally integral.

For instance, the half sum δ of positve roots is algebraically integral but not analytically
if G = SO(3).

In our situation the Theorem 6.15 (Theorem of the Highest Weight) comes in the following
form:

Theorem 6.22: Let G be a compact connected Lie group with complexified Lie alge-
bra g, let T be a maximal torus with complexified Lie algebra t, and let ∆+ (g, t) be a
positive system for the roots. Apart from equivalence the irreducible finite-dimensional
representations π of G stand in one-one correspondence with the dominant analytically
integral linear functionals λ on t, the correspondence being that λ is the highest weight
of π.
104 6. The Infinitesimal Method

And we restate Theorem 6.21 (Weyl’s Character Formula) in the form:

Theorem 6.23: The character χλ of the irreducible finite-dimensional representation of

G with highest weight λ is given by

(det w)ψw(λ+δ)−δ (t)
χλ = w∈W
Πα∈∆+ (1 − ψ−α (t))
at every t ∈ T where no ψα takes the value 1 on t.

In the next section we shall illustrate this by treating an example, which was a milestone
in elementary particle physics.

6.7 The Example su(3)

In 1962 Gell-Mann proposed in a seminal paper [Gel] a symmetry scheme for the de-
scription of hadrons, i.e. certain elementary particles (defined by interacting by strong
interaction), which had a great influence in elementary physics and beyond, as it led
to the notion of quarks and the eightfold way. We can not go too much into the phys-
ical content but discuss the mathematical background to give another example of the
general theory. Respecting the historical context, we adopt the notation introduced by
Gell-Mann and used in Cornwell’s presentation in [Co] vol II, p.502ff: As su(3) consists
of tracelass skew hermitian three-by-three matrices, one can use as a basis
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
i 1 i
A1 := ⎝ i ⎠ , A2 := ⎝ −1 ⎠ , A3 := ⎝ −i ⎠ ,
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
i 1
A4 := ⎝ ⎠, A5 := ⎝ ⎠, A6 := ⎝ i ⎠,
i −1 i
⎛ ⎞ ⎛ ⎞
i
A7 := ⎝ 1 ⎠, A8 := √1 ⎝ i ⎠.
3
−1 −2i

This basis is also a basis for the complexification su(3)c = sl(3, C). One may be tempted
to use the elementary matrices Eij with entries (Eij )kl = δik δjl and take as a basis

Eij , i = j, (i, j = 1, 2, 3) H1 := E11 − E22 , H2 := E22 − E33 .

One has the commutation relations [Eij , Ekl ] = δjk Eil − δil Ekj . For a diagonal matrix
H := D(h1 , h2 , h3 ) we get

[H, Eij ] = (ei (H) − ej (H))Eij , ei (H) = hi ,

and hence
[H1 , E12 ] = 2E12 , [H2 , E12 ] = −E12 ,
[H1 , E13 ] = E13 , [H2 , E13 ] = E13 ,
[H1 , E23 ] = −E23 , [H2 , E23 ] = 2E23 ,
and
[E12 , E21 ] = H1 , [E23 , E32 ] = H2 , [E13 , E31 ] = E11 − E33 =: H3 .
6.7 The Example su(3) 105

We see that h :=< H1 , H2 > is a Cartan subalgbra,

gα1 := < E12 >, gα2 := < E13 >, gα3 := < E23 >

are root spaces, i.e. nontrivial eigenspaces for the roots α1 , α2 , α3 ∈ h∗ given by

α̃1 := (α1 (H1 ), α1 (H2 )) = (2, −1),

α̃2 := (α2 (H1 ), α2 (H2 )) = (1, 1),
α̃3 := (α3 (H1 ), α3 (H2 )) = (−1, 2).

(In the physics literature these tuples α̃j themselves sometimes are called roots.)
We choose these roots αj as positive roots and then get in this coordinization a slightly
unsymmetric picture:

α̃3 r2
AK
A
A r α̃2
A 
A 1 2
r r AH r r
-2 -1 HH
HH
r -1 j α̃1
H

Hence we better use Gell-Mann’s matrices and change to Xj := −iAj . Then we get
H1 = X3 , H2 := X8 . Moreover, for X± := (1/2)(X1 ± iX2 ), Y± := (1/2)(X6 ± iX7 ), and
Z± := (1/2)(X4 ±iX5 ), we have X+ = E12 , X− = E21 etc. and the commutation relations

[H1 , X± ] = 2X± , [H2 , X± ] = 0,√

[H1 , Y± ] = ∓Y± , [H2 , Y± ] = ±√3Y± ,
[H1 , Z± ] = ±Z± , [H2 , Z± ] = ± 3Z± .

Now the Cartan subalgebra is h =< H1 , H2 > and the root spaces are

g±α1 =< X± , >, g±α2 =< Y± >, g±α3 =< Z± >

with positive roots given by

ᾱ1 := (α1 (H1 ), α1 (H2 )) = (2, 0),√

ᾱ2 := (α2 (H1 ), α2 (H2 )) = (−1,√ 3),
ᾱ3 := (α3 (H1 ), α3 (H2 )) = (1, 3).

α1 and α2 are simple roots, we have α1 + α2 = α3 and we get a picture with hexagonal
symmetry:
r2
ᾱ2 ᾱ3
AK

A

A r

A

A
ᾱ1
r r A
r -r
-2 -1 1 2

r -1
106 6. The Infinitesimal Method

In the literature (see for instance [Co] p.518) we find still another normalization, which
is motivated like this: The Killing form B for su(3) is given by diagonal matrices

B(Ap , Aq ) = −12δpq , resp. B(Xp , Xq ) = 12δpq .

Exercise 6.33: Verify this.

One uses B to introduce a nondegenerate symmetric bilinear form <, > on h∗ as follows:
We define a map

h∗ ⊃ ∆
α −→ Hα ∈ h by B(Hα , H) = α(H) for all H ∈ h,

and then
< α, β >:= B(Hα , Hβ ) for all α, β ∈ ∆.
This leads to inner products on hR = < Hα1 , . . . , Hαl >R (α1 , . . . , αl simple positive
roots) and its dual space. For su(3) we have l = 2 and with α1 , α2 from above

< α1 , α1 >=< α2 , α2 >= 1/3, < α1 , α2 >=< α2 , α1 >= −1/6.

It is convenient to introduce an orthonormal basis Ȟ1 , . . . , Ȟl of h, i.e. with B(Ȟp , Ȟq ) =
δpq . Then we have the simple rule

< α, β > = α̌j β̌j , α̌j = α(Ȟj ), β̌j = β(Ȟj ).

j

In our case this leads to

√ √
Ȟ1 = (1/(2 3))H1 , Ȟ2 = (1/2)H2 = (1/(2 3))X8

and √ √ √ √
α̌1 = ( 3/6)ᾱ1 = (1/ 3, 0), α̌2 = ( 3/6)ᾱ2 = (−1/(2 3), 1/2).
We see that, for the Cartan matrix C = (Cpq ) = (2 < αp , αq > / < αp , αp >) from 6.6.2,
we have in this case
2 −1
C=( ).
−1 2
The Weyl group W (see the end of 6.6.3) is generated by the reflections

sαp (α̌q ) = α̌q − Cpq α̌p

at the planes in h∗R  Rl orthogonal to the α̌p with l = 2 and p, q = 1, 2 in our case. For
g0 = su(3) we have 6 elements in W .

Exercise 6.34: Determine these explicitely.

Now we proceed to the discussion of the representations (π̂, V ) of su(3). The general
theory tells us that we can assume that π̂(H) is a diagonal matrix for each H ∈ h. The
diagonal elements π̂(H)jj =: λj (H) fix the weights λj , i.e. linear functionals on h. We
write
λ = (λ(Ȟ1 ), . . . , λ(Ȟl ))

for an ON-basis of h. By the Theorem of the Highest Weight (Theorem 6.15 and Theorem
6.22), an irreducible representation is determined by its highest weight Λ and this highest
6.7 The Example su(3) 107

weight is simple, i.e. has multiplicity one. Moreover the general theory says ([Co] p.568)
that these highest weights are linear combinations with non negative integer coefficients
of fundamental weights Λ1 , . . . , Λl defined by

l
Λj (H) = (C −1 )kj αk (H)
k=1

where α1 , . . . , αl are the positive simple roots, fixed at the beginning.

And with δ = (1/2)Σαj the half sum of these positive simple roots Weyl’s dimensionality
formula says ([Co] p.570) that the irreducible finite-dimensional representation (π̂, V )
with highest weight Λ has dimension

< Λ + δ, αj >
d = Πlj=1 .
< δ, αj >

In our case we have

2/3 1/3
C −1 = ( ).
1/3 2/3

and the fundamental weights

√
 1 = (2/3)α̌1 + (1/3)α̌2 = (1/6)( 3, 1), Λ
Λ  2 = (1/3)α̌1 + (2/3)α̌2 = (1/6)(0, 2).

We write π̂(n1 , n2 ) for the representation with highest weight Λ = n1 Λ1 + n2 Λ2 and get
by Weyl’s formula for its dimension

(6.6) d = (n1 + 1)(n2 + 1)((1/2)(n1 + n2 ) + 1).

In the physics literature one often denotes the representation by its dimension. Then one
has

a) π̌(0, 0) = {1}, the trivial representation, which has only one weight, namely the high-
est weight Λ = 0,
b) π̌(1, 0) = {3} with highest weight Λ = Λ1 and dimension 3,
c) π̌(0, 1) = {3∗ } with highest weight Λ = Λ2 and dimension 3,
d) π̌(2, 0) = {6} with highest weight Λ = 2Λ1 and dimension 6,
e) π̌(0, 2) = {6∗ } with highest weight Λ = 2Λ2 and dimension 6,
f) π̌(1, 1) = {8} with highest weight Λ = Λ1 + Λ2 and dimension 8,
g) π̌(3, 0) = {10} with highest weight Λ = 3Λ1 and dimension 10,

and so on. The weights appearing in the representations can be determined by the fact
that they are of the form

λ = Λ − m1 α1 − m2 α2 , m1 , m2 ∈ N0 ,

and that the Weyl group transforms weights into weights (preserving the multiplicity).
We sketch some of the weight diagrams and later discuss an elementary method to get
these diagrams, which generalizes our discussion in 6.4 and 6.5.
108 6. The Infinitesimal Method

π̌(1, 0) = {3} λ(Ȟ2 )

6
√
Λ = (1/6)( 3, 1) = (2/3)α1 + (1/3)α2
t t
√
sα1 Λ = (1/6)(− 3, 1) = −(1/3α1 + (1/3)α2
-
λ(Ȟ1 )

t
sα2 sα1 Λ = (1/6)(0, 2) = −(1/3)α1 − (2/3)α2
∗
π̌(0, 1) = {3 }
λ(Ȟ2 )
6
Λ = (1/6)(0, 2) = (1/3)α1 + (2/3)α2
t

λ(Ȟ1 )
-
√
sα2 Λ = (1/6)( 3, −1) = −(1/3α1 − (1/3)α2
t t
√
sα1 sα2 Λ = (1/6)(− 3, −1) = − (2/3)α1 − (1/3)α2

π̌(2, 0) = {6} λ(Ȟ2 )

6
Λ − α1 = (1/6)(0, 2) = (1/3)α1 + (2/3)α2
t t t
√
sα1 Λ = −(2/3)α1 + (2/3)α2 Λ = (1/6)(2 3, 2) = (4/3)α1 + (2/3)α2

-
sα1 sα2 (Λ − α1 ) = −(2/3)α1 − (1/3)α2 λ(Ȟ1 )
t t
√
sα2 (Λ − α1 ) = (1/6)( 3, −1) = (1/3)α1 − (1/3)α2

sα2 sα1 Λ = (1/6)(0, −4) = −(2/3)α1 − (4/3)α2

t
 6.7 The Example su(3) 109

λ(Ȟ2 )
6
π̌(1, 1) = {8}
u u
√
sα1 Λ = α2 Λ = (1/6)( 3, 3) = α1 + α2

λ(Ȟ1 )
u j
u u -
sα1 sα2 Λ = −α1 0 = (0, 0) sα2 Λ = α1

√
sα1 sα2 sα1 Λ = −α1 − α2 sα2 sα1 Λ = (1/6)( 3, −3) = −α2
u u

π̌(3, 0) = {10} 6
λ(Ȟ2 )

u u u u
√
−α1 + α2 α2 α1 + α2 Λ = (1/6)(3 3, 3) = 2α1 + α2

λ(Ȟ1 )
u u u -
−α1 0 = (0, 0) α1

u u
−α1 − α2 −α2

u −α1 − 2α2

In {8} we find a first example of a weight, which has not multiplicity one, namely the
weight (0, 0).
110 6. The Infinitesimal Method

Similar to our discussion for G = SU(2) and g = su(2) in 4.2, one has here the problem
of the explicit decomposition of representations into irreducible components. This goes
under the heading of Clebsch Gordon series and can be found for instance in [Co] p.611ff.
As examples we cite

{3} ⊗ {3∗ }  {8} ⊕ {1} and {3} ⊗ {3} ⊗ {3}  {10} ⊕ 2{8} ⊕ {1}.

The proofs of these formulae need some skill or at least patience. But it is very tempt-
ing to imagine the bigger weight diagrams in our examples as composition of the tri-
angles {3} and {3∗ }. This leads to an important interpretation in the theory of ele-
mentary particles: We already remarked that the discussion of the irreducible repre-
sentations of SU(2) resp. SO(3) is useful in the description of particles subjected to
SO(3)-symmetry (in particular an electron in an hydrogen atom), namely a representa-
tion πj , j = 0, 1/2, 1, 3/2, . . . describes a multiplet of states of the particle with angular
momentum resp. spin 2j + 1 and the members of the multiplet are distinguished by a
magnetic quantum number m ∈ Z with m = −(2j + 1), −(2j − 1), . . . , (2j + 1). As one
tried to describe and classify the heavier elementary particles, one observed patterns,
which could be related to the representation theory of certain compact groups, in partic-
ular G = SU(3): In Gell-Mann’s eightfold way we take SU(3) as an internal symmetry
group to get the ingredients of an atom, the proton p and the neutron n as members of
a multiplet of eight states, which correspond to our representation {8}. To be little bit
more precise, one looks at a set of hadrons, i.e. particles characterized by an experimen-
tally fixed behaviour (having strong interaction), and describes these by certain quantum
numbers, namely electric charge Q, hypercharge Y, baryon number B, strangeness S,
isospin I with third component I3 . The baryon number is B = 1 for baryons like proton
or the neutron, -1 for antibaryons, and zero for other particles like the π-mesons. In this
context, one has the following relations

Y = B + S, Q = I3 + (1/2)Y

and one relates particle multipets to representations of su(3) by assigning an I3 - axis

to the axis measuring the weight λ(H1 ) and an Y -axis to λ(H2 ). This way, the low-
est dimensional nontrivial (three-dimensional) representations correspond to two triples
¯ s̄), named quarks resp. antiquarks (with u for up, d for down, and s for
(u, d, s) and (ū, d,
nonzero strangeness). We give a list of the quantum numbers characterizing the quarks

quark B I I3 Y S Q
u 1/3 1/2 −1/2 1/3 0 2/3
d 1/3 1/2 1/2 1/3 0 −1/3
s 1/3 0 0 −2/3 −1 −1/3

and for more information refer to p. 50 f of [FS]. In this picture the weight diagrams look
like this:
6.7 The Example su(3) 111

{3} =: quark {3∗ } =: antiquark

Y Y
6 6
2/3 u s̄

u d
u 1/3 r u 1/3 r

r r - r r - I3
−1/2 1/2 I3 −1/2 1/2

−1/3 r u −1/3 r u
ū d¯
−2/3 u s

Up to now the experimental evidence for the existence of the quarks (as far as we know) is
still indirect: they are thought of as building blocks of the more observable particles like
the proton and the neutron. The following diagram shows the quantum numbers of the
baryon octet {8}. By a simple addition process (comparing the (I3 , Y )-coordinates) one
can guess the quark content of the neutron as n = (udd) and of the proton as p = (uud).
One can do as well for the other particles showing up in the diagram.

π̌(1, 1) = {8}

Y
6
u r 1 u
n p

−1/2 ∆0 1/2 I3
u r j
u r u -
Σ− Σ0 Σ+

Ξ− Ξ+
u r −1 u
112 6. The Infinitesimal Method

We shall come back to these interpretations of representations later on when we dis-

cuss the Euclidean groups and the Poincaré group. A broader discussion of the physical
content of these schemes is beyond the scope of this book. But we still reproduce the
baryon decuplet {10}. It became famous in history because it predicted the existence
of an Ω− -particle (the bottom of the diagram) with fixed properties, which after this
prediction really was found experimentally.

π̌(3, 0) = {10} Y
6
u u r 1 u u
∆− ∆0 ∆+ ∆++

−3/2 −1/2 1/2 3/2

r u r u r u r - I3
− 0 +
Σ Σ Σ

u r −1 u
Ξ− Ξ0

Ω− u −2
6.7 The Example su(3) 113

To finish this section, let us look what happens when we apply to our example the
procedure, which was successful in our constructions of the representations of su(2) and
sl(2, R) in 6.4 and 6.5. We stay in the unnormalized picture with

g =< H1 , H2 , X± , Y± , Z± >

with the commutation relations

√ √
[X+ , X− ] = 0, [Y+ , Y− ] = (1/2)( 3H2 − H1 ), [Z+ , Z− ] = (1/2)( 3H2 + H1 ),

[X+ , Y+ ] = Z+ , [X+ , Z− ] = −Y− ,

[X− , Z+ ] = Y+ , [ X− , Y− ] = −Z− ,
[ Y+ , Z+ ] = X− , [ Y− , Z+ ] = −X+ ,
[ H 1 , X± ] = 2X± , [H2 , X± ] = 0,√
[ H1 , Y ± ] = ∓Y± , [ H2 , Y± ] = ±√3Y± ,
[ H 1 , Z± ] = ±Z± , [ H2 , Z± ] = ± 3Z± ,
(the others are zero) and the roots with coordinates
√ √
ᾱ1 = (2, 0), ᾱ2 = (−1, 3), ᾱ2 = (1, 3).
√
(We then get the weights from our diagrams above by multiplication with ( 3/6).) And
as we prefer to think positively, we start our construction of a representation space V by
taking a vector v ∈ V, v = 0, of lowest weight Λ = (k, l), i.e. with

H1 v = kv, H2 = lv, X− v = Y− v = Z− v = 0.

We try to construct V by applying the plus or creation operators X+ , Y+ , Z+ to v, i.e. we

try

r s t
V = vrst C, vrst := X+ Y+ Z+ v.
From the PBW-Theorem we recall that we have to be careful with the order in which we
apply the creation operators. For instance, one has

v110 := Y+ X+ v = (X+ Y+ + [Y+ , X+ ])v = v110 − v001

since [X+ , Y+ ] = Z+ , but Z+ X+ v = X+ Z+ v = v101 as X+ and Z+ commute.

From the commutation relations it is clear that vrst has weight Λ + rᾱ1 + sᾱ2 + tᾱ3 , so,
in particular, the elements v100 , v010 , v001
√ of the first shell
√ have the respective weights
coordinatized by (k + 2, l), (k − 1, l + 3), (k + 1, l + 3). If we apply the minus or
annihilation operators X− , Y− , Z− , using the commutation relations, we get

X− v100 = X− X+ v = ([X− , X+ ] + X+ X− )v = −H1 v + 0 = −kv,

and the same way

Y− v100 = 0,
Z− v100 = 0,
X− v010 = 0, √
Y− v010 = −(1/2)(l 3 − k)v,
Z− v010 = 0,
X− v001 = v010 ,
Y− v001 = −v100 , √
Z− v001 = −(1/2)(l 3 + k)v,
114 6. The Infinitesimal Method

Before we interpret this and get into danger to lose track in the next shells, let us see
how the natural representation of su(3) looks in this picture:
Here we have

3
V = ej C, e1 = t (1, 0, 0), e2 = t (0, 1, 0), e3 = t (0, 0, 1),
j=1

and hence
⎛ ⎞ ⎞⎛
1 1
H1 e1 = ⎝ −1 ⎠ ⎝ 0 ⎠ = e1 ,
0 0
⎛ ⎞⎛ ⎞
√ 1 1 √
H2 e1 = (1/ 3) ⎝ 1 ⎠ ⎝ 0 ⎠ = (1/ 3)e1 .
−2 0

In the same way we get

√
H1 e2 = −e2 , H2 e2 = (1/ √3) e2
H1 e3 = 0, H2 e3 = −(2/ 3) e3 .

Recalling X+ = E12 , X− = E21 and the analogous relations for the other matrices, one
has

X+ e1 = 0, Y+ e1 = 0, Z+ e1 = 0,
X− e1 = e2 , Y− e1 = 0, Z− e1 = e3 ,
X+ e2 = e1 , Y+ e2 = 0, Z+ e2 = 0,
X− e2 = 0, Y− e2 = e3 , Z− e2 = 0,
X+ e3 = 0, Y+ e3 = e2 , Z+ e3 = e1 ,
X− e3 = 0, Y− e3 = 0, Z− e3 = 0.

These relations√ show that, in the coordinates

√ fixed above, e1 is a highest weight vector of
weight (1,
√ (1/ 3)), e2 has weight (−1, 3) and e3 is a lowest weight vector with weight
(0, −(2/ 3). Thus, as to be expected, we find the weight diagram of {3}. This fits into
the construction we started above as follows: we put v1 := v = e3 and get in the first
shell translating the equations just obtained

X+ v = v100 = 0, Y+ v = v010 = e2 := v2 , Z+ v = v001 = e1 := v3 .

Moreover, the equations translate to the fact that the creation operators applied to the
highest weight vector v3 and to v2 give zero with the exception X+ v2 = v3 . The only non-
trivial actions of the annihilation operators are Y− v2 = v1 , X− v3 = v2 and Z− v3 = v1 .
We compare this to the general form for the action obtained at the beginning (now with
v100 = 0, v010 = v2 , v001 = v3 ) and find as neccessary condition √for a three-dimensional
representation in our general scheme the condition k = 0, l = −2/ 3, just as it should be.
6.7 The Example su(3) 115

To show how to come to higher dimensional representations, one has to go to higher shells:

2
We have v200 = X+ v, v110 = X+ Y+ v, v101 = X+ Z+ v and so on. Here we remember
to pay attention to the order if the creation operators do not commute: As already

remarked, we have v110 := Y+ X+ v = v110 − v001 since one has [X+ , Y+ ] = Z+ . Though it
is a bit lengthy we give the list of the action of the annihilation operators (to be obtained
parallel to the case X− v100 above and using the actions of the minus operators on the
first shell)
X− v110 = −(k − 1)v √010 ,
Y− v110 = −(1/2)(l√3 − k)v100 ,
Z− v110 = −(1/2)(l 3 − k)v,
X− v101 = −(k + 1)v001 + v110 ,
Y− v101 = −v200 , √
Z− v101 = −(1/2)(l 3 + k + 2)v100 ,
X− v011 = v200 , √
Y− v011 = −(1/2)(l√3 − k)v001 − v110 ,
Z− v011 = −(1/2)(l 3 + k + 2)v010 ,
X− v200 = −2(k + 1)v100 ,
Y− v200 = 0,
Z− v200 = 0,
X− v020 = 0, √
Y− v020 = −(l 3 − k + 2)v010 ,
Z− v020 = 0,
X− v002 = 2v001 ,
Y− v002 = −2v√ 101 ,
Z− v002 = −(l 3 − k + 2)v001 .
Just as well one can determine the action of the minus operators in the next shell and so
on. As an example, we show how the eight-dimensional representation {8} (with highest
weight Λ = α1 + α2 ) comes up in this picture: As above we start with a lowest weight
vector v1 := v, which has in our coordinates the weight Λ0 = (k, l). In the first shell we
have
v2 := X+ v = v100 , v3 := Y+ v = v010 , v4 = Z+ v = v001
of respective weights Λ0 +α1 , Λ0 +α2 , Λ0 +α1 +α2 . For the second shell we put (secretely
looking at the weight diagram we already constructed using the results from the general
theory, otherwise the reasoning would take a bit more time)
2 
X+ v = v200 = 0, Y+ X+ v = v110 =: v5 , Z+ X+ v =: v6 ,

and using the relation v110 = v110 − v001 of weight Λ0 + α1 + α2 obtained at the beginning

X+ Y+ v = v110 = v5 + v4 , Y+2 v = 0, Z+ Y+ v = v011 ,

and, since X+ and Z+ commute

2
X+ Z+ v = v101 = v6 , Y+ Z+ v =: v7 , Z+ v = v002 =: v8 .

If we look at the relations coming from the application of the minus operators to the
elements of the second shell listed above, we find in particular the equations
√
X− v200 = −2(k + 1)v100 , Y− v020 = (l 3 − k + 2)v010 .
116 6. The Infinitesimal Method

√
Since we have fixed v200 = v020 = 0, these equations lead to k = −1 and l = 2/ 3.
One verifies easily that all the other relations are fulfilled and the weight diagram for the
eightfold way from above in this context leads to the following picture where only part
of the relations between the operators and generators are made explicit:

π̌(1, 1) = {8} Y
6
u r 2/√3 u
v7 = Y+ v4 v8 = Z+ v4

v5 = Y+ v2 1
u r j
u r u - I3
v3 = Y+ v1 −1 v4 = Z+ v1 v6 = Z+ v2

v1 v2 = X+ v1
u r−2/√3 u

Exercise 6.35: Do the same to construct the representations {6} and {10}.

We hope that these examples help to get a feeling how the discreteness of the weights is
related to the finite-dimensionality of the representations. And, as this procedure gets
more and more complicated, one gets perhaps some motivation to look into the proofs
of the general theorems in the previous section about the integrality of the weights, the
existence of fundamental weights, the action of the Weyl group etc.
Chapter 7

Induced Representations

Induction is in this context a method to construct representations of a group starting by

a representation of a subgroup. As we shall see, in its ultimate specialization one redis-
covers Example 1.5 in our section 1.3, where we constructed representations in function
spaces on G-homogeneous spaces. Already in 1898, Frobenius worked with this method
for finite groups. Later on, Wigner, Bargman, and Gelfand-Neumark used it to con-
struct representations of special groups, in particular the Poincaré group. But it was
Mackey, who, from 1950 on, developed a systematic treatment using essentially elements
of functional analysis. Because it is beyond the scope of this book, we do not describe
these here as carefully as we should. In most places we simply give recipes so that we
can construct the representations for the groups we are striving for. We refer to Mackey
[Ma], Kirillov [Ki] p.157ff, Barut-Raczka [BR] p.473ff, or Warner [Wa] p.365ff for the
necessary background.

7.1 The Principle of Induction

Let G be a group, H a subgroup, and π0 a representation of H in a space H0 , all
this with certain properties to be specified soon. From these data one can construct
representations of G in several ways. The guiding principle is the following. We look at a
space of functions H consisting of functions φ : G −→ H0 , which allow for some additional
conditions also to be specified below and for the fundamental functional equation
(7.1) φ(hg) = δ(h)1/2 π0 (h)φ(g), for all h ∈ H, g ∈ G.
Here δ : H −→ R>0 is a normalizing function, which has several different descriptions.
It is identically one in many important cases, so we do not bother with it for the moment
(if δ is left out, one has unnormalized induction, which later on will appear in connection
with an interpretation via line bundles). If H is closed under right translation, this space
is the representation space for the representation
π = indG
H π0

given by right translation π(g0 )φ(g) = φ(gg0 ). Before we make all this more precise, in
the hope to add some motivation, we treat as an example a representation, which we
already treated as one of our first examples in 1.3.
118 7. Induced Representations

Example 7.1: Let G be the Heisenberg group Heis(R) = {g := (λ, µ, κ); λ, µ, κ ∈ R},
H the subgroup consisting of the elements h := (0, µ, κ) and π0 the one-dimensional
representation of H, which for a fixed m ∈ R∗ is given by

π0 (h) := em (κ) = e2πimκ .

As we know from 3.3 that the Heisenberg group and the subgroup H are unimodular,
one has δ ≡ 1 (we shall see this below in 7.1.2) and we have to deal with the functional
equation (7.1)

φ(hg) = π0 (h)φ(g) = em (κ)φ(g), for all h ∈ H, g ∈ G.

By this functional equation the complex function φ(g) = φ(λ, µ, κ) is forced to have the
form

(7.2) φ(λ, µ, κ) = em (κ + λµ)f (λ)

where f is a one-variable function since every g ∈ Heis(R) has a decomposition

g = (λ, µ, κ) = (0, µ, κ + λµ)(λ, 0, 0) =: h0 s(λ)

with h0 ∈ H and s(λ) = (λ, 0, 0) ∈ G. Application of the right translation with g0 ∈ G

(via the muliplication law in the Heisenberg group) leads to

(π(g0 )φ)(g) = φ(gg0 ) = φ(λ + λ0 , µ + µ0 , κ + κ0 + λµ0 − λ0 µ)

= em (κ + κ0 + λµ + λ0 µ0 + 2λµ0 )f (λ + λ0 )
= em (κ + λµ)em (κ0 + (λ0 + 2λ)µ0 )f (λ + λ0 ).

Comparing with (7.2) above, we rediscover the Schrödinger representation from Section
1.3: If we choose H such that in the decomposition of the elements φ above we have
f ∈ L2 (R), then

f −→ π(g0 )f with π(g0 )f (x) = em (κ0 + (λ0 + 2x)µ0 )f (x + λ0 )

is the prescription for the unitary representation from Example 1.6 in 1.3. The repre-
sentation given by right translation on the space of functions φ on G of the form (7.2) is
called Heisenberg representation.

7.1.1 Preliminary Approach

Now we want to see what is behind this example and try a more systematic treatment.
At first we do not consider the most general case but (following [La] p.43f) one sufficient
for several applications. Let
• G be an unimodular connected linear group,
• H be a closed subgroup,
• K be an unimodular closed subgroup such that the map

H × K −→ G, (h, k) −→ hk

is a topological isomorphism,
• δ = ∆H be the modular function of H, and
• π0 a representation of H in H0 .
7.1 The Principle of Induction 119

Then the homogeneous space of right H-cosets X = H\G can be identified with K
(provided with an action of K from the right). Let Hπ be the space of functions
φ : G −→ H0
satisfying the functional condition (7.1)
φ(hg) = δ(h)1/2 π0 (h)φ(g) for all h ∈ H, g ∈ G
and the finiteness condition that the restriction φ |K =: f is in L2 (K). We define a norm
for Hπ by

(7.3)  φ 2Hπ := f 2L2 (K) = |f (k)|2H0 dk,
K

and with φ̃ |K =: f˜ the scalar product by

< φ, φ̃ >:=< f, f˜ > .
We denote π = indG H π0 for π given by right translation (π(g0 )φ)(g) = φ(gg0 ) and call this
the representation of G induced by π0 . We have to show that this really makes sense and
shall see at the same time why the strange normalizing function δ has to be introduced.

Theorem 7.1: If π0 is bounded, π defines a bounded representation of G on Hπ . π is

unitary if π0 is.

Proof: We have to show that φg0 given by φg0 (g) := φ(gg0 ) is in Hπ . One has
φg0 (hg) = φ(hgg0 ) = δ(h)1/2 π0 (h)φ(gg0 ) = δ(h)1/2 π0 (h)φg0 (g),
i.e. φg0 satifies the functional equation. Now to the finiteness condition: We write g = hk
and kg0 =: hk k  with h, hk ∈ H, k, k  ∈ K. Then we have (with δ = ∆H )
φg0 (k) = φ(kg0 ) = φ(hk k  ) = δ(hk )1/2 π0 (hk )φ(k  ),
and, since π0 is bounded,  φg0 2Hπ is bounded by a constant times

∆H (hk ) | φ(k  ) |2 dk,
K
and equality holds if π0 is unitary. The proof is done if we can show that for f ∈ Cc (K)
one has  
f (k)dk = ∆H (hk )f (k  )dk.
K K
We take φ ∈ Cc (G) and use Fubini’s Theorem. Since G is unimodular, we get
   
φ(hk)dkdh = G φ(g)dg = G φ(gg0 )dg
H K  
= H K φ(hkg0 )dkdh
 
= K H φ(hhk k  )dhdk
 
= K H φ(hk  )∆H (hk )dhdk
 
= H K φ(hk  )∆H (hk )dkdh.
Here we take φ such that

φ(hk) = ϕ(h)f (k) with f ∈ C(K), ϕ ∈ Cc (H), ϕ(h)dh = 1,
H
and get the desired result.
120 7. Induced Representations

Exercise 7.1: Verify that the case of the Heisenberg group and its Schrödinger repre-
sentation is covered by this theorem.

Remark 7.1: Using the notions from our section 6.6 where we introduced a bit more
structure theory, one can also state a useful variant of Theorem 7.1 (as in Flicker [Fl]
p.39):
If G has an Iwasawa decomposition G = N AK, K a maximal compact subgroup, A
the maximal torus in H = N A, N the unipotent radical, and χ a character of H, the
representation space consists of smooth functions φ : G −→ C with

φ(nak) = (δ 1/2 χ)(a)φ(k), for all a ∈ A, n ∈ N, k ∈ K,

where the normalizing function can be described by

δ(a) = | det(Ad a |LieN ) | .

The representation is given again by right translation.

7.1.2 Mackey’s Approach

We shall afterwards come back to more easily accessible special cases but now go (fol-
lowing Mackey) to a more general situation (see [Ki] p.187ff, [BR] p.473) where we have
to intensify the measure theoretic material from section 3.3. Let
• G be a connected linear group,
• H be a closed subgroup,
• π0 a unitary representation of H in H0 (later on, in some cases, we will also treat
non-unitary representations),
• δ := ∆H /∆G where ∆H and ∆G are the modular functions of H resp. G introduced
in 3.3,
• X the space of right H-cosets X = H\G,
• p : G −→ X the natural projection g −→ Hg =: x ∈ X .

Theorem 7.2: The normalized induced representation

π := indG
H π0

is given by right translation on the space Hπ , which is defined as the completion of

its dense subspace spanned by the continuous functions φ : G −→ H0 satisfying the
functional equation (7.1)

φ(hg) = (δ(h))1/2 π0 (h)φ(g) for all h ∈ H, g ∈ G

and the finite norm condition


(7.4) φ 2Hπ :=  φ(s(x)) 2H0 dµs (x) < ∞.
X

Before we elaborate on the fact that we really get a unitary representation, this condition
introducing the norm  . Hπ needs some explanation:
7.1 The Principle of Induction 121

Quasi-invariant Measures, Master Equation and the Mackey Decomposition

As above, we denote by X the space of left H-cosets, X = H\G, and p : G −→ X

the natural projection given by g −→ Hg =: x ∈ X . Then it is a general fact that
the homogeneous space X admits quasi-invariant measures µ. These are defined by the
condition that, for every g ∈ G, µ and µg (with µg (B) := µ(Bg) for all Borel sets B)
have the same null sets, i.e. µ(B) = 0 is equivalent to µg (B) = 0. Again we do not prove
this general statement but later on give enough examples to illuminate the situation (we
hope). For f ∈ Cc (X ) we have
 
µg (f ) = f (xg)dµ(x) = f (x)dµ(xg −1 ) with dµ(xg −1 ) =: dµg (x)
X X

and for g ∈ G we have a density function ρg , which may be understood as the quotient
of dµg and dµ (in the general theory this goes under the name of a Radon-Nikodym
derivative). The construction of quasi-invariant measures can be done as follows.
We take over (from [BR] p.70:) [Ma1] part I, Lemma 1.1:

Theorem 7.3 (Mackey’s Decomposition Theorem): Let G be a separable locally compact

group and let H be a closed subgroup of G. Then there exists a Borel set B in G such
that every element g ∈ G can be uniquely represented in the form

g = hs, h ∈ H, s ∈ B.

In our situation where we have

p : G −→ X , g −→ Hg =: x,

this leads to the existence of a Borel section s : X −→ G, i.e. a map preserving Borel
sets with p ◦ s = idX . Hence for g ∈ G, we have a unique Mackey decomposition

(7.5) G
g = h(x)s(x), h(x) ∈ H, x ∈ X .

Then the invariant measures dr h and dr g on H resp. G are related to a quasi-invariant

measure dµs associated to the section ϕ by

∆G (h)
dr g =: dr hdµs (x)
∆H (h)

(by a reasoning, which is a refinement of the last part of the proof of Theorem 7.1, or by
solving the problems in [Ki] p.132) and we have the quasi-measure relation

dµs (xg) ∆H (h(g, x))

(7.6) = ρg (x) = = δ(h(g, x)).
dµs (x) ∆G (h(g, x))

To someone eventually thinking this is complicated, we recommend to look at the case

of the Heisenberg group we treated above: We have the section s : X −→ G given by
X
x −→ s(x) = (x, 0, 0) and the Mackey decomposition

Heis(R)
g = (λ, µ, κ) = h(x)s(x)

with
h(x) = (0, µ, κ + λµ), s(x) = (λ, 0, 0), x = λ.
122 7. Induced Representations

And, since the groups G and H are both unimodular, we simply have

dr g = dλdµdκ, dr h = dµdκ, dµs (x) = dx.

In this special case, one easily verifies directly that dx is not only quasi-invariant but
invariant. And the general finite norm condition above reduces to the condition that f
is an L2 -function on X  R, i.e. the condition we found above while rediscovering the
Schrödinger representation.

The First Realization

In the general case we will need the explicit form of the action of G on X in the framework
of the Mackey decomposition: We apply the Mackey decomposition (7.5) to s(x)g0 and
get a consistent picture by chosing xg0 such that, for h∗ := h(xg0 ) ∈ H (which, for
reasons to be seen later, we also write as h(g0 , x)), we have the Master Equation (the
name is from [Ki1] p.372)

(7.7) s(x)g0 = h∗ s(xg0 ).

Now, let us verify that the space Hπ , as defined above, is in fact the space for our induced
representation with π(g0 )φ = φg0 :
If φ satisfies the functional equation (7.1), by the same reasoning as in the proof of
Theorem 7.1, so does φg0 for every g0 ∈ G.
The main point is the norm condition. The more experienced reader will see that, in
principle, we do the same as in the proof of Theorem 7.1: Using (7.7) above, the functional
equation (7.1) and the unitarity of π0 , one has from (7.4)

 φg0 2Hπ = X  φg0 (s(x)) 2H0 dµs (x)

= X  φ(s(x)g0 ) 2H0 dµs (x)

= X  φ(h∗ s(xg0 )) 2H0 dµs (x)

= X  φ(s(xg0 )) 2H0 δ(h(g0 , x))dµs (x).

Replacing x −→ xg0−1 , we see that translation does not change the norm

 φg0 2Hπ = X  φ(s(x)) 2H0 δ(h(g0 , xg0−1 ))dµs (xg0−1 )

= X  φ(s(x)) 2H0 dµs (x) =  φ 2Hπ ,

since from the Master Equation (7.7) (as expression of the associativity of the group
operation) one deduces the cocycle condition

(7.8) h(g1 , x)h(g2 , xg1 ) = h(g1 g2 , x).

Via the chain rule the quasi-measure relation (7.6) dµs (xg0 )/dµs (x) = ρg0 (x) = δ(h(x, g0 ))
provides for
δ(xg0−1 , g0 )dµs (xg0−1 ) = dµs (x).
It is now evident that π is unitary if we put

< φ, ψ >:= < φ(s(x)), ψ(s(x)) >H0 dµs (x).
X
7.1 The Principle of Induction 123

Thus we have the essential tools to extend the validity of Theorem 7.1 to the more gen-
eral hypothesis stated above (we leave out to show that changing the section ϕ leads to
equivalent representations). This realization of the induced representation by functions
living on the group is usually called the first realization or induced picture.

Before we go over to another realization, let us point out that the representation by
right translation has an equivalent version where G acts by left translation, i.e. we have
π  (g0 )φ(g) = φ(g0−1 g) where here we have the space H consisting of functions φ fulfilling
π

the fundamental functional equation

(7.9) φ(gh) = δ(h−1 )1/2 π0 (h−1 )φ(g) for all h ∈ H, g ∈ G,

and the same finiteness condition.

Exercise 7.2: Fill in missing details.

The Second Realization

We stay with the same hypotheses as above and turn to functions living on the homoge-
neous space X = H\G. We denote by

Hπ := L2 (X , dµs , H0 )

the space of H0 -valued functions f on X , which are quadratic integrable with respect
to the quasi-invariant measure dµs . In this context we have as fundamental fact the
following

Lemma: The map

ψ : Hπ −→ Hπ , φ −→ f
given by f (x) := φ(s(x)) for x ∈ X is bijective and an isometry.

Proof: ψ is linear and has as its inverse the map

ϕ : Hπ −→ Hπ , f −→ φf

for g = h(x)s(x), x ∈ X given by φf (g) := δ(h(x))1/2 π0 (h(x))f (x). One has


 φf 2Hπ = X  φf (s(x))2H0 dµs (x)

= X f (x)2H0 dµs (x) = f 2Hπ .

Hence the maps preserve the norm.

Both spaces have compatible Hilbert space structure since in a complex inner product
space V for u1 , u2 ∈ V one has the relation

4 < u1 , u2 > =  u1 + u2 2 −  u1 − u2 2 −i  u1 + iu2  +i  u1 − iu2 2 .

Exercise 7.3: Prove this.

Using this, we can carry over the induced representation from Hπ to Hπ and from our
Theorem 7.2 come to an equivalent picture as follows.
124 7. Induced Representations

H π0 is realized on Hπ by π̃ with
Theorem 7.4 (Second Realization): π = indG

π̃(g0 )f (x) := A(g0 , x)f (xg0 ) for all f ∈ Hπ

with
A(g0 , x) := δ(h(g0 , x))1/2 π0 (h(g0 , x))
for s(x)g0 = h(g0 , x)s(xg0 ), x ∈ X , g0 ∈ G.

Proof: We evaluate the commutative diagram:

π(g0 )
Hπ - Hπ

6
ϕ ψ

π̃(g0 ) ?
Hπ - Hπ

• f ∈ Hπ is mapped by ϕ onto φ ∈ Hπ with

φ(g) = φ(h(x)s(x)) = δ(h(x))1/2 π0 (h(x))f (x) for g = h(x)s(x) ∈ G, x ∈ X .

• π(g0 ) maps φ to φg0 with φg0 (g) = φ(gg0 ).

• ψ maps φg0 to f˜ with

f˜(x) = φg0 (s(x)) = φ(s(x)g0 ).

By the Master Equation s(x)g0 = h(g0 , x)s(xg0 ) and the definition of φ this can be
written as

f˜(x) = φ(h(x, g0 )s(xg0 )) = δ(h(g0 , x))1/2 π0 (h(g0 , x))f (xg0 ).

Hence, π̃(g0 ) transports f to f˜ = π̃(g0 )f with f˜ as given in the Theorem.

Exercise 7.4: Repeat this discussion to get a representation space consisting of functions
f living on the space of left cosets X  = G/H by starting from the induced representation
π  given by left translation on the space H . Deduce the representation prescription
π

(7.10) π̃(g0 )f (y) := (δ(h(g0 , y))−1/2 π0 ((h(g0 , y)−1 )f (g0−1 y)

based on the Master equation

g0−1 s(y) = s(g0−1 y)h(g0 , y).

Remark 7.2: In the special case that π0 is the trivial one-dimensional representation
with π0 (h) = 1 for all h ∈ H, we fall back to the situation of our Example 1.5 in Section
1.3 where we constructed a representation on a space of functions on a homogeneous
space X = H\G. This may be seen as starting point of the construction of induced
representations.
7.1 The Principle of Induction 125

Remark 7.3: From the cocycle condition (7.8) for h = h(g, x), one can deduce that the
multiplier system A in Theorem 7.4 as well fulfills this relation, i.e. we have
A(g1 g2 , x) = A(g1 , x)A(g2 , xg1 ) for all g1 , g2 ∈ G, x ∈ X .
And, vice versa, one can take such a system as tool to construct a representation on a
homogeneous space X with a right G-action (see for instance [Ki1] p.388).

For instance in [BR] p.478 a proof can be found that essentially all this does not depend
on the choice of the section s. In most practical cases one has a more or less natural
decomposition of the given group into subgroups as at the beginning of this section, so
one can avoid to talk about different sections, since one has a natural decomposition.
We already discussed the Heisenberg group as an example for this. Here the Schrödinger
representation came out as the Second Realization with π̃ = πm acting on Hπ = L2 (R).
The First Realization by functions living on the group is called Heisenberg representation .

7.1.3 Final Approach

Before we give more concrete examples, we reproduce some general considerations (which
may be skipped at first reading) linked to the report on structure theory in section 6.6
and take over the following situation over from [Kn] p.168: Let G be (not uniquely)
decomposed as G = KH and H in Langlands decomposition H = M AN . We do not give
all the details for a proper definition as to be found in [Kn] p.132. But it may suffice
to say that one has a corresponding direct sum decomposition with corresponding Lie
algebras g = k ⊕ m ⊕ a ⊕ n where a is abelian, n nilpotent, and m and a normalize n.
(Our standard example is G = SL(2, R). Here we have as in section 6.6.4 the (unique)
Iwasawa decomposition G = KAN with K = SO(2), A the positive diagonal elements,
N = {n(x); x ∈ R}, and H = M AN with M = {±E2 }.) Let ∆+ denote the roots of
(g, a) positive for N and ρ := (1/2)Σα∈∆+ (dim gα )α. Moreover, let σ be an irreducible
unitary representation of M on V σ , and ν ∈ (a∗ )C , i.e. a complex linear functional on a.
Then in this context, the induced representation is written as

M AN (σ ⊗ exp ν ⊗ 1)
π = indG
and in the induced picture given by left translation π(g0 )φ(g) = φ(g0−1 g) on the represen-
tation space, which is given by the dense subspace of continuous functions φ : G −→ V σ
with functional equation
φ(gman) = e−(ν+ρ) log a σ(m)−1 φ(g)
and the norm 
φ2 := |φ(k)|2 dk.
K
By restriction to the compact group K one gets the compact picture. A dense subspace
is given by continuous functions f : K −→ V σ with
f (km) = σ(m)−1 f (k)
and the same norm as above. Here the representation π̃ is given by
−1
π̃(g0 )f (k) := e−(ν+ρ)H(g0 k)
σ(µ(g0−1 k))−1 f (κ(g0−1 k))
where g decomposes under G = KM AN as g = κ(g)µ(g)eH(g) n.
126 7. Induced Representations

In [Kn] p.169 there is a third picture, the noncompact picture, given by restriction to
N̄ = ΘN . Here the representation π̃ is given by
−1
π̃(g0 )f (n̄) := e−(ν+ρ) log a(g0 n̄)
σ(m(g0−1 n̄))−1 f (n̄(g0−1 n̄))

if one takes account of the fact that N̄ M AN exhausts G except for a lower dimensional
set and that in general one has a decomposition g = n̄(g)m(g)a(g)n. The representation
space is here L2 (N̄ , e2Re νH(n̄) dn̄).

7.1.4 Some Questions and two Easy Examples

Several questions arise immediately:

i) How to construct non-trivial functions in Hπ resp. Hπ ?

ii) When is an induced representation π irreducible?
iii) Which representations can be constructed by induction?
iv) If one restricts an induced representation to the subgroup, does one get back the
representation of the subgroup?
v) What happens by iteration of the induction procedure?
vi) What about the induced representation if the inducing representation is the sum or
tensor product of two representations ?

Here we can not give satisfying answers to these questions (but hope that our examples
later on will be sufficiently illuminating). We only mention some remarks.

Ad i) One can prove ([BR] p.477) the following statement.

Proposition 7.1: Let ϕ be a continuous H0 -valued function on G with compact support.
Define φϕ by 
φϕ (g) := π0 (h−1 )ϕ(hg)dr h.
H
Then φϕ is a continuous function on G whose support goes to a compact set by the
projection p : G −→ X and is an element of Hπ . The set

C π := {φϕ ; ϕ(g) = ξ(g)v, ξ ∈ Cc (G), v ∈ H0 }

is a dense set in Hπ .

Ad ii) We have seen that the Schrödinger representation of the Heisenberg group is an
irreducible induced representation. In general, this is not the case and in our examples
below we will have to discuss this. We shall find conditions to be added to the definition
of our space if one wants to come to an irreducible representation (e.g. holomorphic in-
duction). In [BR] p.499, there are general criteria to get irreducibility (using the concept
of a canonical system of imprimitivity).

Ad iii) This concept of canonical system of imprimitivity is also used to give a criterium
of inducibility stating under which conditions a unitary representation is equivalent to
an induced representation. One finds versions of this in [BR] p.495, [Ki] p.191, and in
[Ki1] p.389 (Mackey Inducibility Criterion). We do not reproduce these because the
statements need more new notions.
7.1 The Principle of Induction 127

Ad iv) Induction and restriction are adjoint functors. We do not try to make this more
precise. At least for finite groups, one finds statements of the Frobenius Reciprocity as in
[Se] p.57 or [Ki] p.185, which, using our notation (see section 1.2), for a representation
π of G can be written as
c(π, indG
H π0 ) = c(π|H , π0 ).

Ad v) One has the concept of induction in stages, which symbolically can be written as

H indK  indK
indG H G

if we have a subgroup chain G ⊃ H ⊃ K ([Ki] p.184, [BR] p.489).

Ad vi) If π0 and π0 are representations of the subgroup H of G the operations of taking
direct sum and induction are interchangeable
 G 
H (π0 ⊕ π0 )  indH π0 ⊕ indH π0
indG G

([BR] p.488). And if π1 and π2 are unitary representations of the closed subgroups H1
and H2 of the separable, locally compact groups G1 resp. G2 , then one has ([BR] p.491)
1 ×G2
H1 ×H2 (π1 ⊗ π2 )  indH1 π1 ⊗ indH2 π2 .
indG G1 G2

Up to now, we treated only the example of the Heisenberg group. Before we do more se-
rious work with G = SL(2, R), we discuss another example we already looked at, namely
G = SU(2), and the two-dimensional form of the Poincaré group, which later on will
be discussed more thoroughly in the four-dimensional form in the context of Mackey’s
method for semidirect products.

Example 7.2: Let

A n a
G = {g = ( ) ∈ SL(3, R); A ∈ SO(1, 1), n = ( ) ∈ R2 }
0 1 t

be the two-dimensional Poincaré group. We abbreviate g =: (A, n), so that matrix

multiplication leads to the composition law

gg  = (AA , An + n).

In analogy to the case of SO(2), we write the elements of SO(1, 1) in the form

cosh ϑ sinh ϑ
A=( ) =: rh(ϑ), ϑ ∈ R.
sinh ϑ cosh ϑ

From the addition theorem we have AA = rh(ϑ)rh(ϑ ) = rh(ϑ + ϑ ), and dϑ is a bi-
invariant measure on SO(1, 1). We take the subgroup

E2 n
H := {h = ( ); n ∈ R2 }
0 1

and the one-dimensional representation π0 defined by

t α
π0 (h) := eπi(aα+tβ) = eπi nn̂
=: χn̂ (n), n̂ := ( ) ∈ R2 .
β
128 7. Induced Representations

Moreover, we denote by K the image of the injection of SO(1, 1) into G

A 0
K := {k = ( ); A ∈ SO(1, 1)}
0 1
Then every g ∈ G has a Mackey decomposition g = (A, n) = (E, n)(A, 0) and we have
G −→ X = H\G  K  R, g = (A, n) −→ A = rh(ϑ) −→ ϑ.
The Master Equation s(x)g0 = h(g0 , x)s(xg0 ) in this case simply is
(A, 0)(A0 , n0 ) = (AA0 , An0 ) = (E, An0 )(AA0 , 0).
We see that dϑ is not only quasi-invariant but invariant and we have a trivial normalizing
function δ(h) = 1 for all h ∈ H. Hence, the induced representation π = indG H χn̂ is given
by right translation on the space Hπ , which is the completion of the space of continuous
functions

φ : G −→ C, φ(hg) = χn̂ (h)φ(g), | φ((rh(ϑ), 0)) |2 dϑ < ∞.
R

The induced representation has its Second Realization on the space Hπ = L2 (X , dµ(x)) =
L2 (R, dϑ). We have the isomorphism given by
Hπ
φ −→ f ∈ Hπ with f (ϑ) := φ((rh(ϑ), 0)),
and the prescription for the representation from Theorem 7.4, namely
π̃(g0 )f (x) = A(g0 , x)f (xg0 ),
in this case leads to
t
π̃((rh(ϑ0 ), n0 ))f (ϑ) = χn̂ (rh(ϑ)n0 )f (ϑ + ϑ0 ) = eπi n0 rh(ϑ)n̂
f (ϑ + ϑ0 ).
The representation is irreducible. This is a byproduct of Mackey’s general theory for
representations of semidirect products, upon which we will report later in section 7.3.
The reader is invited to find a direct proof using the infinitesimal method analogous to
the example in 6.5 (Exercise 7.5). In the next example the induced representation will
come out to be reducible.

Example 7.3: For the group

a b
G = SU(2) = {g = ( ); a, b ∈ C, | a |2 + | b |2 = 1}
−b̄ ā
we obtained in 4.2 the result that every irreducible representation of G is equivalent to
a representation πj , j ∈ (1/2)N0 , given on the space V (j) = C[x, y]2j of homogeneous
polynomials P of degree 2j by the prescription
πj (g0 )P (x, y) := P (ā0 x − by0 , b̄0 x + a0 y).
This formula is based on the action on functions produced by
x x x ā x − b0 y
( ) −→ g0−1 ( ) = t ḡ0 ( )=( 0 ).
y y y b̄0 x + a0 y
We want to see how this fits into our scheme of construction of induced representations:
It is very natural to take as subgroup H the group of diagonal matrices
a
H := {h = ( ); a = eiϕ , ϕ ∈ R}  U(1)  SO(2)
ā
7.1 The Principle of Induction 129

and as its representation π0 given by π0 (h) = eikϕ =: χk (h), k ∈ N0 . Since in πj we have

a left action, we deal here also with an action of G by left translation, i.e.
π(g0 )φ(g) = φ(g0−1 g)
for functions φ from a representation space H , which fulfill the functional equation
π

φ(gh) = π0 (h−1 )φ(g) = λk φ(g), λ = e−iϕ .

(We know that G and H are compact and, hence, unimodular so that the normalizing
factor δ is trivial.) The homogeneous space X  = G/H showing up here is one of the
most fundamental examples of homogeneous spaces: One has
SU(2)/H  P1 (C),
the one-dimensional complex projective space. We already met with projective spaces in
4.3 while introducing the concept of projective representations. Here we use the following
notation: G = SU(2) acts on V = C2 by
x
(g, v) −→ gv, v := ( ).
y
We have P1 (C) = (V \0)∼ where for v, v  ∈ V one has v ∼ v  iff there exists a λ ∈ C∗
with v  = λv. The action of G on V induces an action on P1 (C) given by g(v∼ ) := (gv)∼ .
We write (x : y) = t (x, y)∼ . For v0 = t (0, 1) we have gv0 = t (b, ā), i.e. g(0 : 1) = (b : ā),
and the stabilizing group of (v0 )∼ = (0 : 1) is just H  U(1). Hence we have
a b
SU(2)/H −→ P1 (C), g = ( ) −→ (b : ā).
−b̄ ā
A homogeneous function in two variables F (x, y) induces via f (x : y) := F (x, y) a
function f on P1 (C) , and this in turn induces a function φF living on G given by
φF (g) := F (t (g t (0, 1))) = F ((0, 1)t g). Hence we associate the function φP to a homoge-
neous polynomial P of degree k = 2j from our representation πj , namely
a b
φP (g) := P (b, ā) for g = ( ).
−b̄ ā
The fact that P is homogeneous of degree k translates to the functional equation
φP (gh) = P ((0, 1)t (gh)) = P ((0, e−iϕ )t g) = e−iϕk P ((0, 1)t g) = λk φP (g)
eiϕ
where h = ( ) and λ = e−iϕ as above. One has
e−iϕ
ā0 −b0 a b ā a + b0 b̄ ā0 b − b0 ā
g0−1 g = ( )( )=( 0 )
b̄0 a0 −b̄ ā b̄0 a − a0 b̄ b̄0 b + a0 ā
and hence
φP (g0−1 g) = P (ā0 b − b0 ā, b̄0 b + a0 ā) = φπ(g0 )P (g).
We see that πj is a candidate for a subrepresentation of the induced representation
π = indG H χk . We propose to show as Exercise 7.6 that the finite norm condition is
fulfilled and to search for further conditions to be imposed such that we get a subspace
of Hπ , which is isomorphic to the space V (j) of πj . We shall return to this question
and the example in Section 7.6 in the framework of the more sophisticated and elegant
concept of an interpretation of the induction process via line bundles.
130 7. Induced Representations

7.2 Unitary Representations of SL(2, R)

Serge Lang dedicated a whole book, namely “SL(2, R)” ([La1]), to this group and its
representations. Also, there is a lot of refined material concerning the representations of
SL(2, R) throughout Knapp’s book [Kn]. Summaries presenting models for the different
representations can be found in many places, for instance in [KT] p.1–24 or in [Do]. The
most classical reference is Bargmann’s article [Ba] from 1947. We shall need the repre-
sentations of G = SL(2, R) later in the discussion of the representations of the Poincaré
group and most essentially in the Outlook to Number Theory.

Using the Infinitesimal Method, in Theorem 6.2 in section 6.4, we obtained a list fixing the
possible equivalence classes of irreducible unitary representations of G = SL(2, R). Here
we will show that these representations really exist by presenting concrete realizations
or, as one also says, models. In most occasions it is important to have different models,
depending on the occasion where the representation is showing up. The method we follow
here is just the application of the induction procedure that we developed in the preceding
section. Thus, we have to start by choosing a subgroup H of our G. Already in several
occasions in this text, we specified subgroups of G = SL(2, R). We recall (for instance
from 6.6.4) the notation

cos ϑ sin ϑ
K := SO(2) = {r(ϑ) = ( ); ϑ ∈ R},
− sin ϑ cos ϑ
y 1/2
A := {t(y) = ( ); y ∈ R, y > 0},
y −1/2
1 x
N := {n(x) = ( ); x ∈ R},
1
1
N̄ := {n̄(x) = ( ); x ∈ R},
x 1
M := {±E2 },
a b
B := M AN = {h = ( ); a, b ∈ R, a = 0},
a−1
B̄ := M AN̄ .

The subgroup B of upper triangular matrices is a special example of a parabolic subgroup.

It is also called the standard Borel group. One has the disjoint Iwasawa decomposition
G = KAN = AN K and G = KB = BK with B ∩ K = {±E2 }. It makes sense to start
the induction procedure from all these subgroups but the best thing to do is parabolic
induction:

1. The Principal Series

We take H := B and π0 the one-dimensional representation, which for fixed s ∈ R and

 ∈ {0, 1}, is given by
π0 (h) :=| a |is (sgn a)
.
We put X := H\G and take the Mackey decomposition (7.5)

a b d−1 b 1
g=( ) = ( )( ) = h(x)s(x) for d = 0,
c d d c/d 1
b −a 1
= ( )( ) = h(x)s(x) for d = 0.
b−1 −1
7.2 Unitary Representations of SL(2, R) 131

1 1
Hence X is in bijection to the set S := {( ); x ∈ R} ∪ {( )}. As in our
x 1 −1
1
context a point has measure zero, we can disregard the element ( ) and work
−1
with
G −→ X = H\G ⊃ X   R, g −→ x := c/d, d = 0
1 a b0
with section s : X  −→ G, x −→ s(x) = ( ). For g0 = ( 0 ) our Master
x 1 c0 d0
Equation (7.7)
s(x)g0 = h(g0 , x)s(xg0 )
is realized by

(b0 x + d0 )−1 b0 1 a0 x + c0
h(x, g0 ) = ( ), s(xg0 ) = ( ), xg0 = .
0 b0 x + d 0 xg0 1 b0 x + d0

We see that dµ(x) = dx is a quasi-invariant measure on X   R with the quasi-measure

relation (7.6)
dµ(xg0 ) 1
= =: ρg0 (x) = δ(h(g0 , x)).
dµ(x) (b0 x + d0 )2
Hence, we have δ(h) = a2 and the induced representation π = indG H π0 is given in the
First Realization by right translation on the space Hπ , which is the completion of the
space of smooth functions φ : G −→ C with

(7.11) φ(hg) = (δ(h))1/2 π0 (h)φ(g) for all h ∈ H, g ∈ G

and 
φ :=2
| φ(s(x)) |2 dx < ∞.
R

As we saw in our discussion of the representations of the Heisenberg group, it is more

illuminating to look at the Second Realization. This is given by the representation space
Hπ = L2 (X , dµ(x)) = L2 (R, dx). We see from Theorem 7.4 (or by a direct calculation)
that the representation acts by
a0 x + c0
π̃(g0 )f (x) =| b0 x + d0 |−is−1 (sgn (b0 x + d0 ))
f ( ) for f ∈ L2 (R, dx).
b0 x + d0
Exercise 7.7: Determine the derived representation and verify that this induced rep-
resentation is a model for the up to now only hypothetical representation πis,± from
Theorem 6.2 in 6.4.

Remark 7.4: This exercise provides as byproduct a proof that this induced representa-
tion is irreducible. There are more direct ways to see this (see for instance [Kn] p.36).

Exercise 7.8: We propose to verify in a special situation that the construction es-
sentially does not depend on the choice of the quasi-invariant measure: As in [BR]
p.483, repeat the former discussion for the quasi-invariant measure dµ(x) = ϕ(x) dx
with ϕ(x) = (1 + x2 )−2 to get a representation space H̃π linked to our Hπ by the inter-
twining operator f˜ −→ f, f (x) = f˜(x)/(1 + x2 )1/2 .
132 7. Induced Representations

To get more familiarity with these constructions and because of the general importance of
the result, we also discuss the realization of our induced representation by left translation,
s.t., for functions φ living on G, we have a representation given by π  (g0 )φ(g) = φ(g0−1 g).
a b
We keep H and π0 as above and now put X  = G/H. For g = ( ) with a = 0, we
c d
have a Mackey decomposition
1 a b
g = s(y)h(y), s(y) = ( ), y = c/a, h(y) = ( ),
y 1 a−1

i.e. a map p : G −→ G/H = X  ⊃ Ẋ   R, g −→ y = c/a,

1
with section y −→ s(y) = ( ). The Master Equation is in this case
y 1
g0−1 s(y) = s(g0−1 y)h(y, g0 )

with
1 a0 y − c0
s(g0−1 y) = ( ), g0−1 y = ,
g0−1 y 1 −b0 y + d0
d0 − b 0 y −b0
h(y, g0 ) = ( ) =: h∗ .
(d0 − b0 y)−1

We take dµ(y) = dy as quasi-invariant measure on Ẋ   R and have the quasi-measure

relation
dµ(g0−1 y) 1
= = ρg0 (y).
dµ(y) (−b0 y + d0 )2
To be on the safe side and to provide for some more exercise in the handling of these
notions, we verify that left translation preserves the norm: We take a function φ with

φ(gh) = (δ(h))−1/2 π0 (h−1 )φ(g) for all h ∈ H, g ∈ G,

put φg0 (g) := φ(g0−1 g), and calculate using the Master Equation g0−1 s(y) = s(g0−1 y)h∗
and the unitarity of π0

φg0 2 = R | φg0 (s(y)) |2 dµ(y)

= R | φ(s(g0−1 y)h∗ ) |2 dµ(y)

= R | φ(s(g0−1 y)) |2 δ(h∗ )−1 dµ(y)

= R | φ(s(ỹ)) |2 dµ(ỹ),
since one has by the quasi-measure relation

dµ(ỹ) dµ(g0−1 y) 1
= = = δ(h∗ )−1 .
dµ(y) dµ(y) (−b0 y + d0 )2
So we have as the First Realization the induced representation defined by functions φ on
G with the functional equation

φ(gh) = (δ(h))−1/2 π0 (h−1 )φ(g) for all h ∈ H, g ∈ G,

and the norm condition


φ = 2
| φ(s(y)) |2 dµ(y) < ∞.
R
7.2 Unitary Representations of SL(2, R) 133

1
From here, by the map φ −→ f, f (y) := φ(( )), we come to the Second Realization
y 1
given on the space Hπ = L2 (R, dy) by the prescription

π̃(g0 )f (y) = (δ(h(y, g0 )))−1/2 π(h(y, g0 )−1 )f (g0−1 y)

a0 y − c0
= | d0 − b0 y |−is−1 (sgn(d0 − b0 y))
f ( ).
−b0 y + d0
This is the same formula as in [KT] p.17 and [Kn] p.36 and p.167 where this is an example
for a noncompact picture. We remark that Knapp comes to this formula from his induced
picture, which is prima facie different from our First Realization as he does not use the
definition of the norm we used above but the same definition of the norm as in (7.3)

φ2 = | φ(k) |2 dk.
K

In [Kn] p.168, one finds a proof that this norm and the one we used above are equal. We
propose to search here for a direct proof (Exercise 7.9).

Moreover, in [Kn] p.37, there is a proof of the irreducibility of these induced representa-
tions πis,± except for π0,− , which is based on euclidean Fourier Transform.

2. The Discrete Series

We will discuss several models, which realize the discrete series representations.

The Upper Half Plane Model

We take H = K = SO(2) and π0 (h) = χk (r(ϑ)) = eikϑ , k ∈ Z, for h = r(ϑ) ∈ SO(2).

It is a custom to construct the representations πk± from our list in Theorem 6.2 in 6.4
by functions living on the space of left cosets G/K = SL(2, R)/SO(2). This space is
again an eminent example of a homogeneous space. It is called the Poincaré upper half
plane and we denote it by H := {τ = x + iy ∈ C; y > 0}. Since this is important for
our constructions here but also as a standard example in complex function theory, we
assemble some related standard notions and facts.

a b
For g = ( ) ∈ SL(2, R) and τ ∈ H, we put
c d
aτ + b
g(τ ) :=
cτ + d
and call this a linear fractional transformation.

Exercise 7.10: Prove that this is well defined and provides a transitive left action of
G = SL(2, R) (even of GL(2, R)) on H.
Verify that H = SO(2) is the stabilizing group of τ = i.
Show that the automorphic factor

j(g, τ ) := (cτ + d)−1

fulfills the cocycle relation (7.8) (in the version for the action from the left)

j(gg  , τ ) = j(g, g  (τ ))j(g  , τ ).

134 7. Induced Representations

One has a map

p : G = SL(2, R) −→ SL(2, R)/SO(2)  H, g −→ g(i) =: τ = x + iy
for the Iwasawa decomposition g = n(x)t(y)r(ϑ), which here can be interpreted as a
Mackey decomposition g = s(τ )h, s(τ ) = n(x)t(y), h = r(ϑ). Moreover, one easily verifies
the following formulae
1 ac + bd iϑ d − ic a b
(7.12) y= ,x= 2 , e =√ for g = ( )
c2 +d 2 c +d 2
c2 + d2 c d
and, for g = r(ϑ )t(y  )n(x ),
a −c 1 ab + cd
(7.13) cos ϑ = √ , sin ϑ = √ , y = 2 2
, x = 2 .
a2 +c2 2
a +c2 a +c a + c2
Exercise 7.11: Verify
d(g(τ )) 1
=
dτ (cτ + d)2
and show that dµ(τ ) = y −2 dxdy is a G−invariant measure on H.

This is also an occasion to practise again the calculation of infinitesimal operations

initiated in 6.2 and 6.3: For a smooth complex function φ on G and U ∈ g, we get
a right invariant differential operator RU defined by
d
RU φ(g) := φ((exp tU )−1 g) |t=0
dt
and a left invariant operator LU by
d
LU φ(g) := φ(g (exp tU )) |t=0 .
dt
From 6.4 we recall that we have
g = sl(2, R) = {X ∈ M2 (R); Tr X = 0} = < F, G, H >
with
1 1
F =( ), G = ( ), H = ( )
1 −1
and
gc := g ⊗R C = < F, G, H >C = < X+ , X− , Z >C
where
1 ±i −i
X± = (1/2)(H ± i(F + G)) = (1/2)( ), Z = −i(F − G) = ( )
±i −1 i
with
[Z, X± ] = ±2X± , [X+ , X− ] = Z.
Using the coordinates x, y, ϑ of the Iwasawa decomposition as parameters for g and the
commutation rule t(y)n(x) = n(yx)t(y), one obtains
−1
RH φ(g) = d
dt φ((exp tH) g) |t=0
−t
e
t )n(x)t(y)r(ϑ)) |t=0
d
= dt φ(( e
−2t
= d
dt φ(n(e x)t(e−2t y)r(ϑ)) |t=0
= −(2x∂x + 2y∂y )φ(g).
7.2 Unitary Representations of SL(2, R) 135

The same way one has

−1
RF φ(g) = d
dt φ((exp tF ) g) |t=0
1 −t
= d
dt φ(( )n(x)t(y)r(ϑ)) |t=0
1
dt φ(n(x − t)t(y)r(ϑ)) |t=0
d
=
= −∂x φ(g),
and
−1
RG φ(g) = d
dt φ((exp tG) g) |t=0
1
= φ(( )n(x)t(y)r(ϑ)) |t=0
−t 1
= φ(n(x )t(y )r(ϑ )) |t=0
 

= ((x2 − y 2 )∂x + 2xy∂y + y∂ϑ )φ(g),

where in the last formula we used (7.12) to get

x − t(x2 + y 2 )
x = ,
t2 y 2 + (1 − tx)2
y
y = 2 2 ,
t y + (1 − tx)2

1 − t(x − iy)
eiϑ = eiϑ .
t2 y 2 + (1 − tx)2

Hence we have the differential operators

RF = −∂x ,
(7.14) RG = (x2 − y 2 )∂x + 2xy∂y + y∂ϑ ,
RH = −2x∂x − 2y∂y ,

which combine to
RZ = i((1 + x2 − y 2 )∂x + 2xy∂y + y∂ϑ ),
(7.15)
RX± = ±(i/2)(((x ± i)2 − y 2 )∂x + 2(x ± i)y∂y + y∂ϑ ).

A similar calculation (Exercise 7.12) leads to

LZ = −i∂ϑ ,
(7.16)
LX± = ±(i/2)e±2iϑ (2y(∂x ∓ i∂y ) − ∂ϑ ).

In 6.4 we already met with the (multiple of the Casimir) element

Ω := X+ X− + X− X+ + (1/2)Z 2 ,

which, by the way, is also (Exercise 7.13) Ω = F G + GF + (1/2)H 2 . By the natural

extension of the definition of the differential operators from g to U (gc ) (via RU U
:=
RU RU
) and a small calculation we get the SL(2, R)-Laplacian

(7.17) ∆ := RΩ = LΩ = 2y 2 (∂x2 + ∂y2 ) − 2y∂x ∂ϑ .

136 7. Induced Representations

After these general preparations, we finally come to the discrete series representations.
There are a lot of different models in the literature. We present the most customary one
and then discuss how it fits into the scheme of the induction procedure.
For instance in [La1] p.181, we find the representation πm with the representation space
Hm := L2hol (H, µm ), m ≥ 2. This space consists of holomorphic (i.e. complex differen-
tiable) functions f on the upper half plane H, which have a finite norm f m < ∞,
i.e. with 
dxdy
f 2m := | f (x, y) |2 y m 2 < ∞
H y
and one has an scalar product defined by

dxdy
< f1 , f2 >:= f1 (x, y)f2 (x, y)y m 2 .
H y

As will be apparent soon (even for those readers, which are not so familiar with complex
function theory), the main point is here reduced to the fact that holomorphic functions
f are essentially smooth complex valued functions characterized by the Cauchy Riemann
differential equation
(∂x + i∂y )f = 0.

a b
We write f (τ ) := f (x, y) and put g −1 =: ( ) (this is some trick to get nicer formulae
c d
but, if not remembered carefully, sometimes leads to errors). Then the representation
πm is in the half plane model given by

(7.18) πm (g)f (τ ) := f (g −1 (τ ))jm (g −1 , τ ) = f (g −1 (τ ))(cτ + d)−m .

Here we have used the important notation of the automorphic factor

jm (g, τ ) := j(g, τ )m

extending the one introduced in Exercise 7.10.

Remark 7.5: In [Kn] p.35, one finds a representation given on the same space by the
prescription
a b aτ − c
π̌m ( ))f (τ ) := (−bτ + d)−m f ( ).
c d −bτ + d
And in [GGP] p.53, the authors define a representation Tm given by

a b aτ + c
Tm ( ))f (τ ) := (bτ + d)−m f ( ).
c d bτ + d

The reader is invited to compare this to our definition. We stick to it because we want
to take over directly some proofs from Lang’s presentation in [La1] p.181ff. Namely, one
has to verify several things:
– Hm is complete and non-trivial.
– π fulfills the algebraic relation π(gg  ) = π(g)π(g  ).
– π is continuous and unitary.
7.2 Unitary Representations of SL(2, R) 137

The algebraic relation is an immediate consequence of the facts stated as Exercise 7.10.
We verify the unitarity using Exercise 7.11:
∞∞
π(g)f 22 = 0 −∞ | f (g −1 τ ) |2 | cτ + d |−2m y m−2 dxdy
∞∞
= 0 −∞ | f (τ ) |2 y m−2 dxdy
= f 22 .
For the other statements we refer to Lang. The idea is to transfer the situation from the
upper half plane to the unit disc where the function theory is more lucid. Since this is
an important technique, we report on some facts:

The Unit Disc Model

By the Cayley transform c the Poincaré half plane H is (biholomorphically) equivalent

to the unit disc
D := {ζ ∈ C; | ζ | < 1} :
In all cases, this bijective map sends the point i to 0 but otherwise is not uniquely fixed
in the literature. We take the map
τ −i 1 −i
c : H −→ D, τ −→ ζ := = C(τ ) for C = ( ),
τ +i 1 i
which has as its inverse the map given by
ζ +1 i i
ζ −→ τ = −i = C −1 (ζ) with C −1 = (1/(2i))( ).
ζ −1 −1 1
It is a standard fact that G = SL(2, R) is isomorphic to
α β
G := SU(1, 1) = {g  = ( ); | α |2 − | β |2 = 1}.
β̄ ᾱ
We take
a b α β
c̃ : SL(2, R) −→ SU(1, 1), g = ( ) −→ g  := CgC −1 = ( )
c d β̄ ᾱ
with
(7.19) α = α(g) = (1/2)(a + d + i(b − c)), β = β(g) = (1/2)(a − d − i(b + c)).
Then we have
cos ϑ sin ϑ eiϑ
r(ϑ) := ( ) −→ ( ) =: s(ϑ),
− sin ϑ cos ϑ e−iϑ
i.e. c̃(SO(2)) = {s(ϑ); ϑ ∈ R} = K  , which we identify with U(1). By the linear
fractional transformation G × D −→ D, (g  , ζ) −→ g  (ζ), G acts (transitively) on D.
The stabilizing group of ζ = 0 is U(1) = K  . Hence in analogy to the map
p : G −→ G/SO(2), g −→ g(i) = τ = x + iy
with the Iwasawa decomposition g = n(x)t(y)r(ϑ), we have the map
p : G −→ G /K  = D, g  −→ g  (0) = ζ = u + iv
and the commutative diagram
138 7. Induced Representations

c̃
G = SL(2, R) - G = SU(1, 1)

p p
? c ?
H = SL(2, R)/SO(2) - D = SU(1, 1)/U (1)

with  
1 + iτ̄ −1 − iτ
c̃(n(x)t(y)) = (1/2)y −1/2
−1 + iτ̄ 1 − iτ
τ −i
and c(τ ) = ζ = τ +i .

Now, it is a rather easily acceptable fact that the space of holomorphic functions on D
is spanned by the monomials ζ n , n ∈ N0 . And some calculation shows that

dudv dudv
dν(ζ) := , dνm (ζ) := 41−m (1− | ζ |2 )m ,
1− | ζ |2 1− | ζ |2

is a G −invariant measure resp. a quasi-invariant measure on D and one has an isomor-

phism between the corresponding spaces of holomorphic functions on H and D

Cm : Hm = L2hol (H, dµm ) −→ L2hol (D, dνm ), f −→ Cm f := f˜

given by
ζ + 1 −2i m
f˜(ζ) = f (−i )( ) .
ζ −1 ζ −1
Using a convenient extension of the group action given by (7.18), this can also be written
as
f˜(ζ) = f (C −1 (ζ))jm (C −1 , ζ).

The map Cm intertwines the representation πm from the half plane model given by (7.18)
with the equivalent representation π̃m on the space of functions f˜ ∈ L2hol (D, dνm ) given
by the prescription
π̃m (g)f˜(ζ) := f˜(c̃(g)−1 (ζ))jm (c̃(g)−1 , ζ).

We call this the unit disc model. By the way, if one replaces c̃(g) by g  ∈ G , one also
has a representation of G = SU(1, 1).

Already above, we stated the (plausible) fact that L2hol (D, dνm ) is spanned by functions
f˜n with f˜n (ζ) = ζ n , n ∈ N0 . As to be seen easily, via Cm , these functions correspond to
functions fn living on H with

τ −i n
fn (τ ) := ( ) (τ + i)−m .
τ +i
7.2 Unitary Representations of SL(2, R) 139

The Group Model

Intertwined with these two equivalent models, there is a third one, which we will call the
group model because its representation space consists of functions living on the group
G = SL(2, R). We go back to the map c̃ : G −→ G̃. In the formulae (7.19) one has the
elements of G̃ as functions of g ∈ G. We introduce the parameters x, y, ϑ describing g in
the Iwasawa decomposition as in (7.12) and get by a small calculation

α(g) = (1/2)(a + d + i(b − c)) = (1/2)y −1/2 eiϑ i(τ̄ − i),

β(g) = (1/2)(a − d − i(b + c)) = (1/2)y −1/2 e−iϑ (−i)(τ − i),
and by another one

φn (g) := β(g)n ᾱ(g)−(n+m) (1/(2i))m = y m/2 eimϑ fn (τ ).

Inspired by this formula, we define a lift ϕm , which associates functions φ living on G to

functions f on H via

f (τ ) −→ φf (g) = (ϕm f )(g) := f (g(i))jm (g, i) = f (τ )y m/2 eimϑ .

Here we use the fact that one has jm (g, i) = (cτ + d)−m = y m/2 eimϑ if g is fixed by
our Iwasawa parametrization (7.12). With the help of the cocycle condition for the
automorphic factor (or multiplier) jm , we deduce for these lifted functions

φ(g0−1 g) = f (g0−1 g(i))jm (g0−1 g, i)

= f (g0−1 τ )jm (g0−1 , g(i))jm (g, i)
= (πm (g0 )f )(g(i))jm (g, i).

This shows that the lift ϕm intertwines (algebraically) the representation πm on the space
L2hol (H, dµm ) with the left regular representation λ on the space of functions φ living on
G and coming out by the lift ϕm . In [La1] p.180 one finds a proof that the functions φn
are square integrable with respect to the (biinvariant) measure dg on G given (again in
our Iwasawa parameters) by
dg = y −2 dxdydϑ.
We call this representation, given by left translation on the space spanned by the func-
tions φn living on the group, the group model.

Already earlier, while introducing the other two models, we could have identified the
−
class of our representation with the one of the representation πm of highest weight −m
from the list of (possible) irreducible unitary representations obtained by infinitesimal
considerations in 6.4. We will do this now. The reader is invited to watch very closely
because Lang in [La1] p.184 (erroneously if our reasoning is correct) identifies it as a
lowest weight representation.
There are several ways to proceed: For the isomorphism c̃ : G = SL(2, R) −→ SU(1, 1)
one has c̃(r(ϑ)) = s(ϑ), c̃(r(ϑ)−1 g)) = s(−ϑ)c̃(g). Hence we deduce

α(r(ϑ)−1 g) = e−iϑ α(g), β(r(ϑ)−1 g) = e−iϑ β(g),

and
φn (r(ϑ)−1 g) = e−(2n+m)iϑ φn ,
i.e. our representation has the weights −m, −(m + 2), −(m + 4), . . . . We compare this
with our results from 6.4 and conclude that we have a representation equivalent to the
140 7. Induced Representations

−
highest weight representation πm . But we can do more: In (7.15) we determined the op-
erators RF , RG , RH and RZ , RX± . These operators are just the operators producing the
derived representation of the left regular representation λ. Thus, by a small calculation
(Exercise 7.14), one can verify the formulae

RZ φn = −(2n + m)φn , RX+ φn = −nφn+1 , RX− φn = (n + m)φn+1 .

We can get corresponding formulae by determination of the derived representations of

the other models. We propose the following

Exercise 7.15: Verify the formulae

Rm
Z = π̂m (Z) = i((1 + τ 2 )∂τ + mτ ),
Rm
X± = π̂m (X± ) = ±(i/2)((i ± τ )2 ∂τ + m(τ ± i)),

and
Rm
Z fn = −(m + 2n)fn ,
Rm
X+ fn = −nfn−1 ,
Rm
X− fn = (n + m)fn+1 .
+
Similarly the representations πm of lowest weight can be realized by holomorphic func-
tions on the lower half plane H̄ = {τ = x+iy ∈ C, y < 0} or by antiholomorphic functions
f , i.e. with (∂x − i∂y )f = 0.

Exercise 7.16: Determine the class of the representation π̌m taken over from Knapp in
the Remark 7.5 above.

The Holomorphically Induced Model

Finally, we want to show how these realizations of the discrete series representations fit
into the discussion of the induction procedure. Namely one can obtain them as subrep-
resentations of suitable induced representations:

We start with the subgroup H = K = SO(2) and its representation π0 given by

π0 (r(ϑ)) = χk (r(ϑ)) = eikϑ , k ∈ Z.

The Iwasawa decomposition can be understood as a Mackey decomposition

g = s(τ )r(ϑ), s(τ ) := n(x)t(y),

and we have our standard projection p : G −→ G/SO(2) = H, g −→ g(i) = x + iy = τ

with section s : H −→ G, τ −→ s(τ ). Then the induced representation πkind := indG
SO(2) χk
is given by (inverse) left translation λ on the completion Hχk of the space of smooth
functions φ on G with functional equation

φ(gr(ϑ)) = e−ikϑ φ(g), for all g ∈ G, r(ϑ) ∈ SO(2)

and the norm condition


φ22 = | φ(s(τ )) |2 y −2 dxdy < ∞.
H
7.2 Unitary Representations of SL(2, R) 141

From the functional equation we learn that the functions φ in the representation space
are all of the type
φ(g) = e−ikϑ F (x, y)
with a function F depending only on x, y if g is given in our Iwasawa parametrization.
Comparing this to the functions establishing the group model above, we see that we have
to find a condition, which further restricts these functions F . Here we remember from
the infinitesimal considerations in 6.4 that functions spanning the space of a discrete
series representation πk± have to be eigenfunctions of the Casimir element

Ω = X+ X− + X− X+ + (1/2)Z 2 .

In (7.17) we already determined that Ω acts on smooth functions φ living on G as the

differential operator

∆ = RΩ = LΩ = 4y 2 (∂x2 + ∂y2 ) − 4y∂x ∂ϑ .

The preceding discussion in mind, we try the ansatz

φ(g) = e−ikϑ y −k/2 f (x, y).

We get

∆φ = [(4y 2 (∂x2 + ∂y2 ) + 4yik(∂x + i∂y ) + (k/2)((k/2) + 1))f ]y −k/2 e−ikϑ ,

i.e. we have
∆φ = (k/2)((k/2) + 1)φ
if
∆k f := 4y 2 (∂x2 + ∂y2 ) + 4yik(∂x + i∂y )f = 0,
and this is obviously the case if f fulfills the equation

∂τ̄ f := (1/2)(∂x + i∂y )f = 0,

i.e. if f is holomorphic. Hence for k = −m, we have recovered the lift of the representation
πm as a subrepresentation of our induced representation πkind .
There is a shorter way to single out in Hχk a subspace carrying πm : To the two conditions
characterizing the space of the induced representation, namely the functional equation
and the norm condition, we add simply the condition

LX− φ = 0.

This comes out as follows: From (7.16) we have

LX− = −(i/2)e−2iϑ (2y(∂x + i∂y ) − ∂ϑ ).

For φ(g) = e−ikϑ y −k/2 f (x, y) this again leads to the holomorphicity condition ∂τ̄ f = 0.

Remark 7.6: This is a special example of a general procedure propagated in particular

by W. Schmid and called holomorphic induction.

Remark 7.7: Here we began by stating the prescription for the representation πm in a
space of functions living on the homogeneous space H = G/SO(2). It is a nice exercise to
142 7. Induced Representations

rediscover this formula by our general procedure to change from the First to the Second
Realization of an induced representation:
a b
We start with Mackey’s decomposition as above, g = s(τ )r(ϑ), and for g0−1 = ( )
c d
the Master Equation, using s(τ ) = n(x)t(y), states

g0−1 s(τ ) = s(g0−1 (τ ))r(ϑ∗ )

with 
iϑ∗ cτ + d
e = .
cτ̄ + d
Moreover one has the standard formulae

dµ(g0−1 (τ ) 1 y
= , Im (g0−1 (τ )) = ,
dµ(τ ) (cτ + d)2 | cτ + d |2

such that the left regular representation π(g0 )φ(g) = φ(g0−1 g) in the Second Realization
with F (τ ) = φ(s(τ )) translates to the prescription

dµ(g0−1 (τ )) 1/2 −1
π(g0 )F (τ ) = ( ) π0 (r(ϑ∗ ))F (g0−1 (τ )).
dµ(τ )

With π0 = χk and F (τ ) = y −k/2 f (τ ) for k = −m by a small calculation this leads to the

prescription (7.18)
πm (g0 )f (τ ) = f (g0−1 (τ ))(cτ + d)−m .
Remark 7.8: The functions φ in the representation space of the Ffirst Realization and
the matrix coefficients < π(g)φ, φ > are square-integrable for m ≥ 2. This is a general
criterium to distinguish discrete series representations from the other representations.

The representations π1± are called mock discrete series representations or limits of discrete
±
series. They are realized by the same prescriptions as the πm with m ≥ 0 but here the
−
space of holomorphic functions f on H for π1 gets the new norm

f 2 = sup | f (x + iy) |2 dx,
y>0 R

and analoguously for the other case.

The courageous reader may be tempted to apply this principle of holomorphic induction
to reconstruct as Exercise 7.17 the irreducible unitary representations of G = SU(2).

3. The Complementary Series

For the sake of completeness, we indicate here also a model, but only briefly, as done
in most sources. (The best idea is to go back to Bargmann’s original paper [Ba].) We
take up again the prescriptions as in the construction of the principal series but to get
a model for πs , 0 < s < 1 we change the norm on the space of complex valued functions
on R and introduce  
f (x)f (y)
f s :=
2
dxdy.
R R x−y |
| 1−s
7.3 Unitary Representations of SL(2,C) and of the Lorentz Group 143

a b
For g −1 = ( ) the prescription is here
c d

πs (g)f (x) =| cx + d |−1−s f (g −1 (x)).

Some more models can be found in [vD] p.465 and [GGP] p.36.

A Final General Remark

We close this section by pointing out to the Subrepresentation Theorem, which goes
back to Harish–Chandra and roughly states (for a precise formulation and proof see [Kn]
p.238) that a certain class of representations, the admissible representations, may be
realized as subrepresentations of parabolically induced representations starting eventually
by non-unitary representations of the subgroup. Admissible representations are very
important, in particular in number theory. An admissible representation (π, V ) of G
is (in first approximation) characterized as follows ([Ge] p.55): Let K be a maximal
compact subgroup of G. For an irreducible representation χ of K, denote by V (χ) the
subspace of vectors in V, which transform according some multiple of χ. Then one has a
decomposition
V = ⊕χ V (χ)
where each V (χ) is finite dimensional.

Irreducible unitary representations of reductive groups are admissible. So all represen-

tations in this text have this property (with exception of the representations of the
(non-reductive) Jacobi group appearing later).

The interested reader can exemplify the statement of the subrepresentation theorem in
the case of G = SL(2, R) by constructing as Exercise 7.18 another model of the discrete
series representation by starting the induction procedure from

H = {h = n(x)t(y); x, y ∈ R, y > 0}, π0 (h) = y s , s ∈ C.

7.3 Unitary Representations of SL(2,C) and of the

Lorentz Group
The representations of the group G = SL(2, C) are of particular importance for physi-
cists because this group is the double cover of the identity component of the complete
homogeneous Lorentz group in four dimensions (to which we will restrict ourselves in
this section). At first, we will discuss the representations of SL(2, C) and then the ho-
momorphism into the Lorentz group.

As central fact we have:

Theorem 7.5: For G = SL(2, C) there are (up to equivalence) only two series of irre-
ducible unitary representations:
– the unitary principal series πk,iv depending on two parameters k ∈ Z, v ∈ R,
and
– the complementary series π0,u depending on a real parameter u with 0 < u < 2.
Among these the only equivalences are πk,iv  π−k,−iv .
144 7. Induced Representations

Thus, in contrast to the smaller group SL(2, R), we have no discrete series representa-
tions. We reproduce models for these representations from [Kn] p. 33/4 and discuss their
construction via the induction procedure afterwards.

As representation space we take the space H = L2 (C) of square-integrable complex

functions f on C with measure dzdz̄. For z ∈ C, f ∈ H, the prescription for the unitary
principal series representation is given by

a b −bz + d −k az − c
(7.20) πk,iv (( ))f (z) :=| −bz + d |−2−iv ( ) f( ).
c d | −bz + d | −bz + d

Here, the notion of the fractional linear transformation used in the last section as a map
SL(2, R) × H −→ H, (g, τ ) −→ g(τ ), is naturally extended to a map

SL(2, C) × {C ∪ ∞} −→ {C ∪ ∞}

with
a b
( )(z) := az+b
cz+d for z ∈ C, cz + d = 0,
c d
:= ∞ for z ∈ C, cz + d = 0,

:= a
c for z = ∞.
In a similar way, for k ∈ Z, w = u + iv ∈ C, there is a non–unitary principal series πk,w
given by the prescription

a b −bz + d −k az − c
πk,w (( )f (z) :=| −bz + d |−2−w ( ) f( )
c d | −bz + d | −bz + d

for functions f in the space L2 (C, dµw ) with dµw (z) = (1+ | z |2 )u dzdz̄. However, in the
special case k = 0, w = u ∈ R, 0 < u < 2, the representation πu := π0,u is unitary if the
representation space is provided with the inner product
 
f (z)f˜(ζ)
< f, f˜ >:= dzdζ.
C C | z − ζ |2−u

This is our model for the complementary series.

As to the construction, these representations are induced representations where one takes
the subgroup
α β
H := {h = ( ); α, β ∈ C, α = 0}
α−1
and the representation π0 with π0 (h) :=| α |w ( |α|
α k
) . We have the Mackey decomposition

a b
g=( ) = s(τ )h
c d

where (as in our discussion of the principal series for SL(2, R) in the previous section) it
suffices to treat the case a = 0, such that one has

1 a b
s(τ ) = ( ), z := b/a and h = ( ),
z 1 a−1
7.3 Unitary Representations of SL(2,C) and of the Lorentz Group 145

and the Master Equation g −1 s(z) = s(z ∗ )h∗ with

az − c −bz + d −b
z∗ = , h∗ = ( ).
−bz + d (−bz + d)−1

Moreover, we have
dz ∗ dz̄ ∗ = | −bz + d |−4 dzdz̄.
Hence, since the induced representation in left regular version in its Second Realization
on the space H = L2 (C) has the general prescription

dµ(z ∗ )
π(g)f (z) = π0 (h∗ )f (z ∗ ), for all f ∈ H,
dµ(z)

with w = iv this leads to the formula (7.20).

In [Kn] p.33/4 Knapp proves the irreducibility of πk,iv using Schur’s Lemma in the uni-
tary version cited in 3.2. One could also do this (and verify that really we only have
the above mentioned irreducible unitary representations) by infinitesimal considerations
as in our discussion of SL(2, R). But since sl(2, C) has real dimension 6, this is leads
to considerable non-trivial computations (to be found for instance in Naimark [Na] or
Cornwell [Co] p.668ff in the framework of the discusssion of the Lorentz group).

The interest in the representations of G = SL(2, C) is enlarged by the fact that they
immediately lead to projective representations of the proper orthochronous Lorentz group,
resp. those, which are trivial on −E2 , to genuine representations, the same way as in our
discussion of SU(2) and SO(3) in 4.3. Before we describe the relation between SL(2, C)
and the Lorentz group we have to fix some notation.

We write
⎛ ⎞
1
⎜ 1 ⎟
GL := {A ∈ GL(4, R); t AD3,1 A = D3,1 = ⎜
⎝
⎟}
⎠
1
−1

for the complete homogeneous Lorentz group, i.e. the group of linear transformations,
which leave invariant the Minkowski metric from the theory of special relativity

ds2 = dx21 + dx22 + dx23 − dx24 .

One may be tempted to replace (x1 , x2 , x3 , x4 ) by (x, y, z, ct), (c the velocity of light)
and put
ds2 = c2 dt2 − dx2 − dy 2 − dz 2 ,
as often done in the physics literature, but we will not do this here. In our former notation
from 0.1 we also have GL = O(3, 1) and GL + = SO(3, 1) for the proper homogeneous
Lorentz group, i.e. the invariant subgroup consisting of matrices A with det A = 1. We
write the matrices in block form
B x
A=( t ), B ∈ GL(3, R), x, y ∈ R3 , a ∈ R.
y a
146 7. Induced Representations

GL
+ has two connected components distinguished by a > 0 and a < 0 since one has

t
B y E B x t
BB − y t y t
Bx − ya
( )( 3 )( )=( t ),
t
x a −1 ty a xB − at y t
xx − a2

E3
and the condition that this should be equal to ( ) asks for
−1

(7.21) t
BB − y t y = E3 , t Bx − ya = 0, t xx − a2 = −1,

and, hence, in particular a2 ≥ 1.

+
The subgroup GL + with a > 0 is called the proper orthochronous homogeneous Lorentz
group. It is the connected component of the identity in SO(3, 1). We will abbreviate this
here to Lorentz group and write G0 := SO(3, 1)0 for it. As easily to be seen, one can
imbed SO(3) into G0 by putting

B
SO(3)
B −→ B̃ := ( ).
1

These elements are called Lorentz rotations. Moreover, we introduce the Lorentz boosts,
which are of the type
⎛ ⎞
cosh t sinh t
⎜ 1 ⎟
B(t) = ⎜⎝
⎟ , t ∈ R.
⎠
1
sinh t cosh t

The structure of G0 is clarified by the following remark.

Proposition 7.2: For every A ∈ G0 there are R, S ∈ SO(3) and t ∈ R such that

A = R̃B(t)S̃,

i.e. G0 is generated by Lorentz rotations and Lorentz boosts.

A proof can be found in any source treating the Lorentz group, for instance in [He] p.75.
Our main fact in this context is the next theorem.

Theorem 7.6: There is a surjective homomorphism

ρ : G = SL(2, C) −→ G0 = SO(3, 1)0

with ker ρ = {±E2 }.

Proof (sketch): We put

α z
V := {X ∈ M2 (C); X = t X̄ =: X ∗ } = {( ); α, β ∈ R, z ∈ C}
z̄ β

and look at the map

−a + d b + ic
ϕ : R4 −→ V, v = t (a, b, c, d) −→ ( ).
b − ic a + d
7.4 Unitary Representations of Semidirect Products 147

For X = ϕ(v), Y = ϕ(w) this map transforms the Minkowski (quasi-)inner product

< v, w > = aa + bb + cc − dd

into the corresponding one defined on V by

σ : V × V −→ R, (X, Y ) −→ −(1/2)(det(X + Y ) − det X − det Y )

as easily to be verified by a small calculation. Hence our ϕ induces a bijection between

O(3, 1) and O(V, σ), which can be recognized as an isomorphism and we take this to
identify both sets.
We define a map

ρ : SL(2, C) −→ GL(V ), A −→ ρ(A) with ρ(A)X := AXA∗ for X ∈ V,

which obviously is a homomorphism. We have

ker ρ = {A; AXA∗ = X for all X ∈ V }

1
and deduce ker ρ = {±E2 } by evaluating the defining condition for X = E2 , ( ),
−1
1
and ( ). The main task is to prove the surjectivity of ρ. We go back to the proof
1
of Theorem 4.7 in 4.3 and use the fact that G0 is generated in GL(V ) by elements
S̃2 (α), S̃3 (α) and B(t). By some computation, one verifies that these are images under
ρ of respectively r(α/2), s(α/2) and s(−it/2).

This theorem justifies the statement made above that the representations π of SL(2, C)
with π(−E2 ) = id create the representations of SO(3, 1)0 . E.g., in [Co] p.671ff one finds
a discussion of the finite-dimensional (non-unitary) representations of the Lorentz group.

7.4 Unitary Representations of Semidirect Products

The semidirect products form a class of groups whose importance is also enhanced by its
applications in physics as the Euclidean and Poincaré groups are prominent examples,
which fit into the following general concept:

We take two groups G0 and N and a (left) action of G0 on N given by

ρ : G0 × N −→ N, (a, n) −→ ρ(a)n

with the additional condition that every ρ(a) is an automorphim of N , i.e. ρ(a) ∈ Aut N
for all a ∈ G0 . Then we define the associated semidirect product G := G0
N as the set
of pairs g = (a, n), a ∈ G0 , n ∈ N provided with the composition law

(7.22) gg  := (aa , nρ(a)n ).

G comes out as a group with neutral element e consisting of the pair of neutral elements
in G0 and in N . The inverse to g = (a, n) is g −1 := (a−1 , ρ(a−1 )n−1 ). (Verification of
the associative law as Exercise 7.19.)
148 7. Induced Representations

Examples:

Example 7.4: The Euclidean group GE (n) in n dimensions: We take G0 = SO(n)

and N = Rn . For A ∈ SO(n), x ∈ Rn , the action ρ is given by matrix multiplication
(A, x) −→ Ax. We denote the semidirect product SO(n)
Rn by GE (n) and write this
again as a matrix group

A a
GE (n) = {g = (A, a) = ( ); A ∈ SO(n), a ∈ Rn }.
1

Here the matrix multiplication in Mn+1 (R)

A a A a AA a + Aa
gg  = ( )( )=( )
1 1 1

reflects exactly the composition law (7.22) leading in this case to gg  = (AA , a + Aa ).
This law comes out very naturally if one asks for the composition of the affine transfor-
mations x −→ Ax + a, which preserve the euclidean norm x2 = x21 + · · · + x2n .

Example 7.5: The Poincaré group GP (n) in n dimensions: Paralell to the case above
we take G0 = SO(n − 1, 1) and N = Rn and get

A a
GP (n) = {g = (A, a) = ( ), A ∈ SO(n − 1, 1), a ∈ Rn }
1

with the same composition law. This is the invariance group of the affine transformations
preserving the Minkowski (pseudo-)norm x2 = x21 + · · · + x2n−1 − x2n .

a b
Example 7.6: Interprete the group B := {g = ( ); a, b ∈ R, a > 0} as a semidi-
a−1
rect product (Exercise 7.20).

Example 7.7: Interprete the Heisenberg group

⎛ ⎞
1 x z
Heis (R) := {g = ⎝ 1 y ⎠ ; x, y, z ∈ R}
1

as semidirect product of G0 = R acting on N = R2 . Can you do the same with Heis(R)

as realized in 0.3? (Exercise 7.21)

Example 7.8: The Jacobi group GJ is the semidirect product of G0 = SL(2, R) acting
on the Heisenberg group N = Heis(R) via

a b
(( ), (λ, µ, κ)) −→ (dλ − cµ, −bλ + aµ, κ).
c d

This group can be presented in several ways as a matrix group (see our Section 8.5 or
for instance [EZ] or [BeS] p.1).

With exception of the last one, in all these examples the group N is abelian. In the sequel
we will restrict our treatment to this case as things become easier under this hypothe-
sis and use additive notation in the second item of the pair g = (a, n) (as we already
7.4 Unitary Representations of Semidirect Products 149

did in the examples). Whoever is interested in more general cases is refered to Mackey’s
original papers or the treatment of the representation theory of the Jacobi group in [BeS].

Remark 7.9: We use the following embeddings

G0 −→ G = G0
N, a −→ (a, 0) =: a

and
N −→ G = G0
N, n −→ (e, n) =: n.
Here one has to be a bit careful because we have n · a = (e, n)(a, 0) = (a, n) = g but
a · n = (a, 0)(e, n) = (a, ρ(a)n) = (a, n) = g in general. There is the useful relation

(a, 0)(e, n)(a−1 , 0) = (e, ρ(a)n),

which, using our embeddings, can be stated as ana−1 = ρ(a)n.

Example 7.9: We take

G := Heis (R) = {g = (x, y, z); x, y, z ∈ R},

G0 := {a = (x, 0, 0); x ∈ R},

and
N := {n = (0, y, z); y, z ∈ R}. |=
    
Using the multiplication law gg = (x + x , y + y , z + z + xy ), one calculates

ana−1 = (0, y, z + xy) = ρ(a)n

and this is consistent with the usual semidirect product composition law (7.22) given by
gg  = (a, n)(a , n ) = (aa , nρ(a)n ).

Example 7.10: For a presentation as semidirect product of the group

a b
B := {g = ( ) =: (a, b); a, b ∈ R, a > 0}
a−1

one may be tempted to try G0 := {a = (a, 0); a > 0} and N := {n = (1, b); b ∈ R}.
−1
From matrix multiplication we have gg  = (aa , ab + ba ) and (using the embedding
introduced above) ana−1 = (1, a2 b) = ρ(a)n. But then the composition law would give
gg  = (a, b)(a , b ) = (aa , nρ(a)n ) = (aa , b + a2 b ) which is not the multiplication law in
this group. To make things consistent (and thus solve the problem from Exercise 7.20),
one constructs the semidirect product G = R>0
R = {[a, x], a, x ∈ R, x > 0} with
ρ(a)x = a2 x and defines an isomorphism

ϕ : B −→ G, (a, b) −→ [a, ab].

The agreable thing about these semidirect products G = G0
N, N abelian, is the fact
that Mackey’s Theory provides an explicit way to construct their irreducible unitary
representations as induced representations. We describe this method without going ev-
erywhere into the intricacies of the proofs, which rely on some serious functional analysis:
150 7. Induced Representations

We start by an easy but important fact relating the representations of G = G0
N with

G0 acting on N by ρ to those of its constituents G0 and N .

Proposition 7.3: Let π be a representation of G and π  , π  the restrictions of π to N

resp. G0 , i.e π  (n) = π((e, n)) for n ∈ N and π  (a) = π((a, 0)) for a ∈ G0 . Then π  and
π  completely determine π and are related by the condition

(7.23) π  (ρ(a)n) = π  (a)π  (n)π  (a−1 ) for all a ∈ G0 , n ∈ N.

Proof: π is determined by π  and π  as one has

π((a, n)) = π((e, n)(a, 0)) = π((e, n))π((a, e)) = π  (n)π  (a).

And from the composition law in G

(a1 , n1 )(a2 , n2 ) = (a1 a2 , n1 + ρ(a1 )n2 )

we deduce

π  (n1 )π  (a1 )π  (n2 )π  (a2 ) = π  (n1 )π  (ρ(a1 )n2 )π  (a1 )π  (a2 )

and with a1 = a, n2 = n this leads immediately to (7.23).

Hence to find representations of G, one has to find representations of G0 and N fulfilling

(7.23). This condition is important in the context of Mackey’s system of imprimitivity,
which is the main tool in the proof that the construction we describe in the sequel leads
to a way to construct all irreducible unitary representations of semidirect products. We
essentially follow [BR] p.503ff and [Ki] p.195ff (where the more general case of repre-
sentations of group extensions is treated). A decisive element in the construction comes
from the fact that the action ρ of G0 on N leads also to an action on the unitary dual
N̂ of N , which in our abelian case consists of (classes of) characters n̂ of N. Namely, we
have
G0 × N̂ −→ N̂ , (a, n̂) −→ an̂
given by an̂(n) := n̂(ρ(a−1 )n). Since the (abelian) group N acts (trivially) on N̂ by
conjugation, i.e. via n0 n̂(n) := n̂(n−1
0 nn0 ), one has an action of the semidirect product
G on N̂ . The G−orbit of n̂ ∈ N̂ is written as

Ôn̂ := Gn̂.

Mackey’s construction works under the assumption that the semidirect product is regu-
lar. All the groups we treat as examples are of this type. Hence, since we will not give
complete proofs anyway, a not so ambitious reader at first sight may skip the following
rather technical condition (from [BR] p.506).

Definition 7.1: We say that G is a regular semidirect product of N and G0 if N̂ contains

a countable family Z1 , Z2 , . . . of Borel subsets, each a union of G−orbits, such that every
orbit in N̂ is the intersection of the members of a subfamily Zn1 , Zn2 , . . . containing that
orbit. Without loss of generality, we can suppose that the intersection of a finite number
of Zi is an element of the family (Zi ). This is equivalent to the assumption that any
orbit is the limit of a deceasing subsequence of (Zi ).
7.4 Unitary Representations of Semidirect Products 151

One also has a slightly different looking condition for regularity used by Mackey ([Ma]
p.77):
There is a subset J ⊂ N̂ , which meets each G−orbit Ô exactly once and is an analytic
subset.

Now as our central prescription for the construction of representations in this section, we
take over from [BR] Theorem 4 on p.508:

Theorem 7.7: Let G = G0
N be a regular semidirect product of separable, locally

compact groups G0 and N , N abelian. Then every irreducible unitary representation
of G is unitarily equivalent to an induced representation π = indG H π0 , which can be
constructed as follows.
– Step 1: Determine the set N̂ of characters n̂ of N , i.e. the dual group of N .
– Step 2: Classify and determine the G0 −orbits Ô in N̂ .
– Step 3: Choose an element n̂0 in each orbit Ô and determine the stabilizing group
G0 n̂0 =: H0 . (This group is called the little group in the physics literature.)
– Step 4: Determine the unitary dual Ĥ0 of H0 .
– Step 5: From every element of Ĥ0 take a representation π1 of H0 and extend it to a
representation π0 of H := H0
N by
π0 ((a, n)) := n̂0 (n)π1 (a) for all a ∈ H0 , n ∈ N.
– Step 6: Take H and π0 from the last step and proceed to (normalized) induction
π = indG
H π0 .

It is quite evident that this recipe produces unitary representations. But it is far from
evident that these are irreducible and that we really get the unitary dual of the semidi-
rect product and, hence, have a classification scheme at hand. As already said, we can
not go into the proofs here (which essentially use Mackey’s imprimitivity theorem) and
refer to the Chapters 16 and 17 of [BR] or §13 of [Ki] and Section 10 of [Ma] where
slightly different approaches are adopted. Now, we discuss some examples to show how
the method works. In these examples one can verify directly the validity of the procedure
encapsulated in Theorem 7.7 for instance with the help of the infinitesimal method from
our Chapter 6.

Examples:

Example 7.11: G := Heis (R) = {g = (x, y, z); x, y, z ∈ R} with

gg  = (x + x , y + y  , z + z  + xy  )
is isomorophic to the semidirect product of G0 := {a = (x, 0, 0); x ∈ R} acting on
N := {n = (0, y, z); y, z ∈ R} via ρ(a)n := (0, y, z + xy) as seen in the Example 7.9
above.
– Step 1: The dual N̂ of N  R2 is again to be identified with R2 . With (l, m) ∈ R2
we write
n̂(n) = ei(ly+mz) =: χ(l,m) (y, z).
– Step 2: a ∈ G0  R acts on n̂ ∈ N̂  R2 by
an̂(n) := n̂(ρ(a−1 )n) = ei(ly+m(z−xy)) = χ(l−xm,m) (y, z).
152 7. Induced Representations

Hence N̂ is disjointly decomposed into exactly two types of G0 −orbits:

• Type 1: For m = 0 and n̂0 = (0, m) one has the line Ôn0 = {(−xm, m), x ∈ R},
• Type 2: For m = 0 and n̂0 = (l, 0) there is the point Ôn0 = {(l, 0)}.

We get the following picture

m
6
Ô(0,m)

r r r r r r r r r r-
r l
Ô(l,0)

One can see that G is a regular semidirect product: The set {(l, m) ∈ R2 ; lm = 0} is an
analytic set, which meets each orbit exactly once.
– Step 3: The stabilizing group of n̂0 = (0, m), m = 0 is H0 = G0(0,m) = {0} and that
of n̂0 = (l, 0), l ∈ R is H0 = G0(l,0) = G0  R.
– Step 4: In the first case the unitary dual of H0 simply consists of the trivial identity
representation and in the second case we have Ĥ0  R with representatives given by
π1 (a) = eijx =: χj (x), j ∈ R.
– Step 5 and 6: In the second case, the type 2 orbits, the product of the stabilizing
group H0 = G0 with N already exhausts the whole group G. Hence there is no induction
necessary and one has for (j, l) ∈ R2 a one-dimensional representation π of G = G0
N
given by
π((x, y, z)) = eijx eily .
In the first case, belonging to the type 1 orbits, one has H = H0
N  N  R2 with a
representation π0 given by π0 (h) = eimz for all h = (0, y, z). Here the induction makes
sense and we have a representation π = indG H π0 (for instance) given by right translation
on the completion H of the space of smooth functions φ : G −→ C with functional
equation
φ(hg) = eimz φ(g) for all h ∈ H, g ∈ G,

and φ2 = R | φ((x, 0, 0) |2 dx < ∞. As the composition law says

gg0 = (x + x0 , y + y0 , z + z0 + xy0 ) and (0, y, z)(x, 0, 0) = (x, y, z),

an element φ in H is of the form φ(g) = eimz f (x) with f ∈ L2 (R) and one has

φ(gg0 ) = eimz eim(z0 +xy0 ) f (x + x0 ).

So we have a second realization of this representation given by

π̃(g0 )f (x) = eim(z0 +xy0 ) f (x + x0 ) for f ∈ L2 (R).

This is the Schrödinger representation πm from our Example 1.6 in 1.3 translated to
Heis (R) by the isomorphism

ϕ : Heis (R) −→ Heis(R), g = (x, y, z) −→ (λ, µ, κ) = (x, y, 2z − xy),

from Exercise 0.1. Hence the unitary dual of the Heisenberg group G is

Ĝ = R2  (R \ {0}).
7.4 Unitary Representations of Semidirect Products 153

Example 7.12: G := GE (2) = SO(2)
R2 has G0 = SO(2) and N = R2 . We write

g = (a, n) with a = r(ϑ) and n = x := t (x1 , x2 ). One has ρ(a)n := r(ϑ)x
– Step 1: As in example 7.11 we have N̂ = R2 . This time, for y = (y1 , y2 ) ∈ R2 we
write
n̂(n) = ei(y1 x1 +y2 x2 ) = χ(y1 ,y2 ) (x1 , x2 ).

– Step 2: a = r(ϑ) ∈ G0 = SO(2) acts on n̂ ∈ N̂ by

an̂(n) := n̂(ρ(a−1 )n) = eiyr(−ϑ)x = χyr(−ϑ) (x).

Hence N̂ is again decomposed into two types of orbits:

• Type 1: For n̂0 = (r, 0), r > 0, the circle

Ôn0 = {y ∈ R2 , y2 = r2 }.

• Type 2: For n̂ = (0, 0), the point Ôn0 = {(0, 0)}.

We can verify the regularity of the semidirect product: One has a countable family of
Borel subsets of N̂ consisting of the following sets: Z0,0 := the point Ôn0 = {(0, 0)} and
for any two positive rational numbers r1 < r2 Zr1 ,r2 := the union of all orbits of type 2
with r1 < r < r2 . Then each orbit is the intersection of the members of the subfamily,
which contain the orbit.
– Step 3: For the type 1 orbits the stabilizing group is G0(r,0) = {E2 } and for the type
2 one has G0(0,0) = G0 .
– Step 4: The unitary dual of G0 = SO(2) is Z.
– Step 5 and 6: In the second case, the type 2 orbit, the product of N and the stabilizing
group is G. Hence for every k ∈ Z, we have a one-dimensional representation π of G given
by
π((r(ϑ), x)) = eikϑ .
In the first case, the type 1 orbit, one has for r > 0 the genuine subgroup H  N and
its one-dimensional representation π0 given by

π0 ((E2 , x)) = n̂0 (n) = eirx1 .

Hence, for every r > 0, we have an induced representation π = indG

H π0 .
In consequence, the unitary dual of G = GE (2) is

Ĝ = R>0  Z.

Example 7.13: As Exercise 7.22 apply our construction recipe to construct represen-
tations of
a b
G = B := {g = ( ) =: (a, b); a, b ∈ R, a > 0}
a−1
and
a b
G = B  := {g = ( ) =: (a, b); a, b ∈ R, a = 0}
a−1
and determine the unitary dual.

Exercise 7.23: Do the same for the Euclidean group in three dimensions GE (3).
154 7. Induced Representations

7.5 Unitary Representations of the Poincaré Group

The Poincaré or inhomogeneous Lorentz group is generally known as the symmetry group
of space-time. In continuation to our discussion of the Lorentz group in 7.3, we put

G = GP (4) := G0
N with G0 := SO(3, 1)0 , N := R4 .

Sometimes it is necessary to take more generally G0 = SO(3, 1) or even G0 = O(3, 1),

which also preserve the Minkowski (pseudo)norm | · | given by

E3
| x |2 := x21 + x22 + x23 − x24 = t x( )x, for all x = t (x1 , x2 , x3 , x4 ) ∈ R4 .
−1

We restrict our treatment to a discussion of the simplest case (it should not be too
difficult to extend our results to the more general cases). It should also be not difficult
to translate our procedure to the situation apparently prefered by the physics literature
([Co] p.677ff, [BR] p.513ff]), namely that the coordinates are x0 , x1 , x2 , x3 and the norm
is given by | x |2 = x20 − x21 − x22 − x23 . Moreover, there are reasons from physics (see the
hints in 4.3 while discussing the relation of the representations of SO(3) and SU(2)) and
mathematics to define and treat the semidirect product

G̃ := SL(2, C)
R4 .

For the definition of this product, one goes back to the proof of Theorem 7.6 in 7.3
where we fixed the surjection ρ : SL(2, C) −→ SO(3, 1): There we had an identification
ϕ : R4 −→ V = Herm2 (C) and defined an action of SL(2, C) on V (and hence on R4 )
by
ρ(Ã)X = ÃX Ã∗ for all X ∈ V, Ã ∈ SL(2, C).
We start with the discussion of G = G0
N, G0 = SO(3, 1)0 , N = R4 (say, for aesthetical
reasons) but soon shall be lead to take in also G̃. We follow the construction procedure
from Theorem 7.7 in the last section.

Step 1: The dual N̂ of N = R4 is again to be identified with R4 . We treat N as

consisting of (coordinate) columns x ∈ R4 and N̂ as consisting of (momentum) rows
t
p, p = (p1 , p2 , p3 , p4 ) ∈ R4 . Hence we write
t
n̂(n) = eiΣpj xj = ei px
=: χp (x).

(In the physics literature often minus signs appear in this identfication of N and its dual,
but we want to avoid the discussion of co- and contravariant vectors at this stage.)

Step 2: a ∈ G0 , here given by A ∈ SO(3, 1)0 , acts on n̂ ∈ N̂ , given by p ∈ R4 , by

an̂(n) := n̂(ρ(A−1 )n) = χpA−1 (x),

i.e. A transforms p into pA−1 . It is elementary calculus that the orbits for this action
are subsets in the three types of quadrics

• Type 1: the two-sheeted hyperboloid given by

− | p |2 = p24 − p23 − p22 − p21 = m2 for m ∈ R, m = 0,

7.5 Unitary Representations of the Poincaré Group 155

• Type 2: the one-sheeted hyperboloid given by

− | p |2 = p24 − p23 − p22 − p21 = (im)2 for m ∈ R, m = 0,

• Type 3: the light cone given by

− | p |2 = p24 − p23 − p22 − p21 = 0.

More precisely, one verifies using the explicit description of SO(3, 1) in 7.3 that R4 de-
composes exactly into the following orbits:

• Type 1+ : +
Ôm := Ôn̂0 = G0 n̂0 for n̂0 = (0, 0, 0, m),
− −
• Type 1 : Ôm := Ôn̂0 = G0 n̂0 for n̂0 = (0, 0, 0, −m),
• Type 2 : Ôim := Ôn̂0 = G0 n̂0 for n̂0 = (m, 0, 0, 0),
• Type 3 :
+
Ô0+ = Ôn̂0 = G0 n̂0 for n̂0 = (−1, 0, 0, 1),
−
• Type 3 : Ô0− := Ôn̂0 = G0 n̂0 for n̂0 = (−1, 0, 0, −1),
• Type 3 :
0
Ô00 := Ôn̂0 = G0 n̂0 = {(0, 0, 0, 0)} for n̂0 = (0, 0, 0, 0),
−
The type 1+ orbit Ôm+
is called the positive mass shell, Ôm the negative mass shell and
±
Ô0 the forward resp. backward light cone. The situation is sketched in the picture:

p4
+
Ôm 6
@
@
@
@ Ô0+ (0, m)
@
@ (m, 0)
- p1
@
Ôim @
Ô0− @
@
@
− @
Ôm

The regularity of the semidirect product can be verified analogously as in our example 2
in the last section (see [BR] p.517/8).

Now, for each orbit Ô, we have to determine its little group, i.e. the stabilizing group
H0 := G0 n̂0 .

Step 3: We get the following four cases.

H0 := G0 n̂0  SO(3) for the type 1± , i.e. n̂0 = (0, 0, 0, ±m),

 SO(2, 1) for the type 2, i.e. n̂0 = (m, 0, 0, 0),
±
 G (2)E
for the type 3 , i.e. n̂0 = (−1, 0, 0, ±1),
 SO(3, 1) 0
for the type 3 ,0
i.e. n̂0 = (0, 0, 0, 0).
156 7. Induced Representations

Proof: i) As in 7.3 we write the elements of G0 = SO(3, 1) as

B x
A=( t ), B ∈ SO(3), x, y ∈ R3 , a ∈ R.
y a

For m = 0, the condition

(0, 0, 0, m)A = (0, 0, 0, m)
enforces y = 0 and a = 1. Moreover, from A ∈ SO(3, 1) one deduces (via (7.21)) x = 0
and, hence, the first assertion G0 n̂0  SO(3).

ii) The other assertions can be proved similarly by direct computations. But as an-
nounced above, it is more agreable to use the fact from 7.3 that one has SL(2, C) =: G̃0
via the surjection ρ : G̃0 −→ G0 as a double cover of the Lorentz group. In the proof
of Theorem 7.6 in 7.3 we identified R4 via a map ϕ with the space V of hermitian
2 × 2−matrices
−a + d b + ic
t
(a, b, c, d) −→ ( ),
b − ic a + d
and defined the map

ρ : SL(2, C) −→ GL(V ), Ã −→ ρ(Ã) with ρ(Ã)X = ÃX Ã∗ for X ∈ V.

Hence for m = 0, the condition

(m, 0, 0, 0)A = (m, 0, 0, 0), A ∈ G0

can be translated into the condition

−m
ρ(Ã)X0 = ÃX0 Ã = X0 for X0 := ϕ(m, 0, 0, 0) = ( ),
m

i.e. Ã ∈ SU(1, 1). It is not difficult to see (Exercise 7.24) that the homomorphism
ρ : SL(2, C) −→ SO(3, 1)0 restricts to a surjection SU(1, 1) −→ SO(2, 1)0 .

It will be useful later to take also

mi
X0 := ϕ(0, 0, m, 0) = ( ).
−mi

By a small calculation, one verifies that the stabilizing group {Ã; ÃX0 Ã = X0 } here
i
is SL(2, R): For σ = ( ) one has σ = σ −1 and t A−1 = σAσ −1 . Hence from
−i
AσA∗ = σ we deduce σ −1 Aσ = A∗ −1 = t A−1 , i.e. A = Ā.

2
iii) One has ϕ(−1, 0, 0, 1) = ( ) =: X0 , and ÃX0 Ã = X0 calls for

a b
à = ( ), with a, b ∈ C, | a |= 1.
a−1

We put a =: eiϑ/2 , b =: e−iϑ/2 z, ϑ ∈ [0, 4π], z ∈ C, and get the multiplication law

a b


ÃÃ = (  −1 ), with a = ei(ϑ+ϑ )/2 , b = e−i(ϑ+ϑ )/2 (z + eiϑ z  ).
a
7.5 Unitary Representations of the Poincaré Group 157

This shows that in this case the stabilizing group H̃0 ⊂ G̃ is isomorphic to the semidirect
product of a rotation group {eiϑ } and a translation group R2  C with the composition
law



(eiϑ/2 , z)(eiϑ /2 , z  ) = (ei(ϑ+ϑ )/2 , z + eiϑ z  ).
The homomorphism (eiϑ/2 −→ (r(ϑ)) makes S̃1 := {eiϑ/2 ; 0 < ϑ < 4π} a double cover
of SO(2) and this extends to a surjection of H̃0 to the Euclidean group GE (2).

iv) In the last case there is nothing to prove.

Step 4: We already determined the unitary dual of covering groups H̃0 of all the little
groups coming in here:

i) In the first case (type 1± ), for H̃0 = SU(2), the elements of the unitary dual are rep-
resented by the representations πj , j ∈ (1/2)N0 discussed in 4.2.

ii) In the second case (type 2), one has H̃0 = SU(1, 1) and from 6.2 we know that
SU(1, 1) is isomorphic to SL(2, R) and that the representations appear in three series
πk± , k ∈ N; πis,± , s ∈ R; πs , 0 < s < 1.

iii) For the third case (type 3± ), as Example 7.12 to our construction procedure in 7.4 we
elaborated that for the two-dimensional Euclidean group one has one-dimensional repre-
sentations χk , k ∈ Z (hence χj , j ∈ (1/2)Z for the double cover) and infinite-dimensional
representations πr , r > 0.

iv) For the type 30 , one has the representations of the Lorentz group resp. its covering
SL(2, C) discussed in 7.3, which come in two series πk,iv , k ∈ Z, v ∈ R; π0,u , 0 < u < 2.

Step 5 and 6: Finally, we have to extend the representations of the little group to its
product with N = R4 and, if this product is not the whole group, proceed to an induced
representation. This way we get the following representations (as usual, we list those of
the covering group G̃ of the Poincaré group):

The Unitary Dual of the Poincaré Group

i) In the first case, one has H̃ = SU(2)
R4 . For the type 1+ with m > 0, we have for
this group the representation π0 given by

π0 ((B, x)) = πj (B)eimx4 for B ∈ SU(2), x = t (x1 , x2 , x3 , x4 ) ∈ R4 .

We write for the induced representation

indG̃ π =: πm,+;j .
H̃ 0

Similarly, the type 1− leads to

π0 ((B, x)) = πj (B)e−imx4 for B ∈ SU(2), x = t (x1 , x2 , x3 , x4 ) ∈ R4

and we denote the induced representation by

πm,−;j , m > 0, j ∈ (1/2)N0 .

158 7. Induced Representations

ii) In the second case, one has

H̃ = SL(2, R)
R4
and its representation
π0 ((A, x)) = π(A)eimx3 for A ∈ SL(2, R), x = t (x1 , x2 , x3 , x4 ) ∈ R4
where π here is one of the representations of SL(2, R) listed above in part ii) of Step 4
resp. in Section 7.2. For the corresponding induced representation we write

indG̃ π
H̃ 0
=: πim;k,± for π = πk± , k ∈ N; m > 0,
=: πim;is,± for π = πis,± , s ∈ R; m > 0,
=: πim;s for π = πs , s ∈ R, 0 < s < 1; m > 0.

iii) In the third case, for type 3± , one has

H̃ = (S̃1
R2 )
R4
and its representation
π0 ((Ã, x)) = π(Ã)ei(x1 ±x4 ) for à ∈ (S̃1
R2 ), x = t (x1 , x2 , x3 , x4 ) ∈ R4
where π here is one of the representations of (S̃1
R2 ) indicated in Step 4. For the
corresponding induced representation we write

indG̃ π
H̃ 0
=: π0,±;j for π = χj , j ∈ (1/2)Z,
=: π0,±;r for π = πr , r ∈ R, r > 0.

iv) In the forth case (for type 30 ) the little group is the whole group and we get the
representations of G̃ by the trivial extension of the representations π of SL(2, C), i.e. one
has π((Ã, x)) = π(Ã) for (Ã, x) ∈ G̃0 . For the corresponding representation we write
π =: π0;j,iv,± for π = πj,iv,± , v ∈ R
=: π0;0,u for π = π0,u , u ∈ R, 0 < u < 2.

Among these, the representations πm,+;j , m > 0, j ∈ (1/2)N0 , coming from the type 1+
orbit and π0,±;j , j ∈ (1/2)Z, coming from the type 3± orbit have a meaning for physics
since they are interpreted as describing elementary particles of spin j and mass m resp.
of zero mass.

Example 7.14: As an illustration we discuss an explicit model for the representation

πm,+;j following [BR] p.522:

From the preceding discussion we take over

G̃ = G̃0
N, G̃0 = SL(2, C), N = R4
and write G̃
g = (Ã, x), Ã ∈ SL(2, C), x ∈ R4 (a column). A bit more explicitely than
at the beginning of this section, the action of G̃0 on N is described as follows. We have
the identification
ϕ : R4 −→ V = Herm2 (C),
x4 − x1 x2 + ix3
x = t (x1 , x2 , x3 , x4 ) −→ X = ( ).
x2 − ix3 x4 + x1
7.5 Unitary Representations of the Poincaré Group 159

Via ϕ the action of G̃0 on V given by (Ã, X) −→ ρ(Ã)X = ÃX Ã∗ is translated into the
action of G̃0 on R4 given by

ρ(Ã)x := ϕ−1 (ρ(Ã)ϕ(x)).

By a slight abuse of notation, as we already did above, we also take ρ(Ã) to be a matrix
from G0 = SO(3, 1)0 and understand ρ(Ã)x as multiplication of the column x by the
t
matrix ρ(Ã). We identify the unitary dual N̂ with R4 by writing n̂(n) = ei px , i.e. n̂ is
coordinatized by (the row) t p, p ∈ R4 . Then the action of G̃0 on N̂ is simply given by
the multiplication of the row t p by the matrix ρ(Ã−1 ), i.e.
t
p −→ t pρ(Ã−1 ) =: Ã · p.

The construction of the representation πm,+;j is based on the type 1+ orbit

Ôm = G̃0 · p0 , p0 = (0, 0, 0, m), m > 0.

As we remarked already, p0 corresponds to X0 = mE2 and has the stabilizing group

H̃0 = SU(2). We take the representation πj of SU(2) and extend it to H̃ = H̃0
N to
get π0 given by
t
π0 ((R, x)) = eimx4 π(R) = ei p0 x
πj (R) for all R ∈ SU(2), x ∈ R4 .

To get some more practice and since it is important for application in physics, we work
out the “ Second Realization” of the induced representation indG̃ π , i.e. with a repre-
H̃ 0
sentation space consisting of functions u living on the homogeneous space

G̃/H̃  G̃0 /H̃0  Ôm

+
.

(At this moment, one usually does not bother with any finiteness or integrability condi-
tion and simply takes smooth functions. We shall have occasion to do better in the next
chapter.)
The general theory of the Iwasawa decomposition or a direct calculation say that we have
a decomposition

λ z
G̃0 = SL(2, C)
à = Ap Rp , Rp ∈ SU(2), Ap = ( ), λ ∈ R∗ , z ∈ C.
λ−1

Here it is customary in physics to coordinatize Ap by coordinates t p = (p1 , p2 , p3 , p4 )

such that p = Ap · p0 . Then one has the Mackey decomposition

G̃
g = s(p)h, s(p) = (Ap , 0), h = (Rp , x) ∈ H̃

and the Master Equation

g0−1 s(p) = s(g0−1 (p))h∗ , h∗ = (R∗ , x∗ ) ∈ H̃

with

g0−1 s(p) = (A−1 −1

0 , −ρ(A0 )x0 )(Ap , 0)
= (A0 Ap , −ρ(A−1
−1
0 )x0 )
−1 −1 −1
= (Ag−1 (p) , 0)((Ag−1 (p) ) A−1
0 Ap , −ρ((Ag −1 (p) ) A0 )x0 ).
0 0 0
160 7. Induced Representations

This shows that one has

−1 −1
R∗ = (Ag−1 (p) ) A−1 ∗
0 Ap , x = −ρ((Ag −1 (p) ) A−1
0 )x0 .
0 0

One can verify that dµ(p) = p−1 +

4 dp1 dp2 dp3 is an invariant measure on Ôm and that the
−1
Radon-Nikodym derivative is dµ(g0 (p))/dµ(p) = 1. Hence by (7.10), in this case the
representation is given by the prescription

π̃(g0 )u(p) = π0 (h∗ −1 )u(g0−1 p)

with
t
π0 (h∗ −1 ) = ei px0
πj (R∗ −1 )
t
as one can verify by a small calculation starting from π0 ((R, x)) = ei p0 x πj (R): Namely,
one has
h∗ −1 = (R∗ −1 , x̃) with x̃ = −ρ(R∗ −1 )x∗
and hence
t
p0 x̃ = −t p0 ρ(R∗ −1 )x∗ = t p0 ρ(R∗ −1 (Ag−1 (p) )−1 A0 )x0
0

i.e.
t
p0 x̃ = t p0 ρ(A−1 t
p )x0 = px0
−1
since we have R∗ = (Ag−1 (p) ) A−1
0 Ap .
0

The functions u we are treating here are vector valued functions

+
u : Ôm −→ Vj  C2j+1 ,
j
i.e. can be written as u(p) = (un (p))n=−j,−j+1,...,j . Here we write Dnn
for the matrix

elements of the representation πj from our discussion in 4.2 and then come to

t
∗ −1
(7.24) (π((Ã0 , x0 ))u)n (p) = ei px0 j
Dnn
(R )un
(Ã−1
0 · p).
n

From here one derives the following interpretation that we have a free particle of spin j
and mass m :

In the rest system p = p0 and under rotations (R, 0), R ∈ SU(2) (7.24) restricts to

j
(π̃((R, 0))un )(p0 ) = Dnn
(R)un
(p0 ),
n

i.e. the elementary expression of SU(2)-symmetry belonging to the representation πj .

And for the infinitesimal generators Pµ = ∂µ , µ = 1, .., 4 of the translations in the µ-th
direction x −→ xµ (t) := x + (tδµµ
)µ
one obtains from (7.24)

Pµ un (p) = pµ un (p).

Hence we have M un = mun , i.e. un is an eigenfunction with eigenvalue m for the mass
operator 
M := P42 − P32 − P22 − P12 .
7.6 Induced Representations and Vector Bundles 161

7.6 Induced Representations and Vector Bundles

The classic approach to the inducing mechanism we followed up to now, in the meantime
has found a more geometric version using the notion of bundles. As Warner remarks in
[Wa] p.376, this does not simplify in any essential way the underlying analysis but has
the advantage that important generalizations of the entire process immediately suggest
themselves (e.g. cohomological induction). Bundles play an important rôle in most parts
of contemporary mathematics and we shall have to use them also in our next chapter
concerning the beautiful orbit method. Hence, we present here some rudiments of the
theory, which nowadays appears in most books on differential geometry, analytic and/or
algebraic geometry etc. For instance, for a thourough study we recommend the book
[AM] by Abraham and Marsden or 5.4 in [Ki] or, more briefly, the appendices in [Ki1]
or [Be]. Whoever is in a hurry to get to the applications in Number Theory may skip
this at first lecture. Before we try to shed some light on the notion of line, vector and
Hilbert bundles, we look again, but from a different angle, at the construction of the
representations of SU(2) from 4.2 and Example 7.3 in 7.1.5. We do this in the hope to
create some motivation for the bundle approach.
 
ζ
We take G := SU(2) ⊃ H  U(1) and π0 (h) := χk (ζ) = ζ for h =
k
∈ H, k ∈ N0 .
ζ̄
As we discussed in Example 7.3 in 7.1.5, we have the homogeneous space
 
a b
X = G/H  P1 (C), g = −→ (b : ā)
−b̄ ā

and the unnormalized induction πku = indu G

H χk given by left (inverse) translation on the
space Hπ of smooth functions φ : G −→ C with functional equation

φ(gh) = χk (h−1 )φ(g) for all g ∈ G, h ∈ H.

(In this example everything is unimodular, hence it does not matter whether we treat un-
or normalized induction, but in general several authors stick to unnormalized induction
though one can also treat normalized induction as we shall see below.) For V = C we
introduce as our first example of a line bundle

Bk := G ×H V := {[g, t] := (g, t)∼ ; g ∈ G, t ∈ V }

where the equivalence ∼ of the pairs is defined by

(gh, t) ∼ (g, π0 (h)t) for g ∈ G, h ∈ H, t ∈ V.

G acts on Bk by
g0 [g, t] := [g0 g, t] for all g0 ∈ G, [g, t] ∈ B k .

One has a projection

pr : B k −→ G/H = X , [g, t] −→ x := gH.

Obviously the projection and the action are well defined and one has the fiber over x

(B k )x := pr−1 (x) = {[g, t]; gH = x}  V  C.

162 7. Induced Representations

This leads to the picture that Bk is a bundle standing over the base space X and consisting
of the fibers (B k )x over all base points x ∈ X .
Our main object is the space Γ(B k ) of sections of Bk , i.e. maps

ϕ : X −→ B k with pr ◦ ϕ = idX .

These sections are of the form ϕ(x) = [g, φ(g)] for x = gH with a function φ : G −→ V .
Since the coset gH may be represented also by gh, h ∈ H, the welldefinedness of the
section asks for
[g, φ(g)] = [gh, φ(gh)] = [gh, π0 (h−1 )φ(g)]

where the last equation is expression of the equivalence introduced in the definition of
the bundle Bk . This leads to the fact that the function φ has to fulfill the functional
equation
φ(gh) = π0 (h−1 )φ(g) for all h ∈ H, g ∈ G,

which we know from the construction of the representation space for the induced repre-
sentation. We define an action of G on the space Γ(Bk ) of sections by

(g0 · ϕ)(gH) := g0 ϕ(g0−1 g) = [g, φ(g0−1 g)] for all g0 ∈ G, ϕ ∈ Γ(Bk ),

and this leads to an action of G on the space H of functions φ : G −→ V (satisfying the

functional equation) defined by
g0 · φ(g) := φ(g0−1 g)

Hence at least formally, we recover the prescription for the definition of the induced
representation π. This can be completed to a precise statement if we provide our bundle
with an appropriate topology. And this appears to be largely independent of the special
example G = SU(2), which we proposed at the beginning. In the following, we will try
to show how this works by giving some rudiments of the theory of bundles. We start
by following the treatment of Warner [Wa] p.377, which is adapted to the application in
representation theory, and proceed to another more general approach afterwards.

Bundles, First Approach

Definition 7.2: Let X be a fixed Hausdorff topological space. A bundle B over X is a

pair (B, p) where B is a Hausdorff topological space and p is a continuous open map of
B onto X .

One calls X the base space, B the bundle space, and p the bundle projection. By abuse
of notation the pair (B, p) is often abbreviated by the letter B to indicate the bundle.

For x ∈ X , the set p−1 (x) is called the fiber over x and often denoted by Bx . A cross
section or simply section of the bundle is a map s : X −→ B with p ◦ s = idX .

If b ∈ B and s(p(b)) = b, then we say that s passes through b. If to every b ∈ B there is

a continuous cross section s passing through b, then the bundle is said to have enough
cross sections.
7.6 Induced Representations and Vector Bundles 163

A Hilbert bundle B over X is a bundle (B, p) over X together with operations and a norm
making each fiber Bx , x ∈ X into a complex Hilbert space and satisfying the following
conditions:

1.) The map B −→ R, b −→ b is continuous,

2.) The operation of addition is a continuous map of {(b1 , b2 ) ∈ B × B, p(b1 ) = p(b2 )}
into B,
3.) The operation of scalar multiplication is a continuous map of C × B into B,
4.) The topology of B is such that its restriction to each fiber Bx is the norm topology
of the Hilbert space Bx . (This condition is formulated more precisely in [Wa] p.377.)

Example 7.15: Let E be any Hilbert space. The trivial bundle with constant fiber E
is given by B := X × E equipped with the product topology and the projection p given
by p(x, a) = x for x ∈ X , a ∈ E. The cross sections are given by the functions s : X −→ E.

Let B = (B, p) be a Hilbert bundle over X and Γ(B) the set of continuous cross sections
of B . Then it is clear that Γ(B) is a complex vector space under pointwise addition
and scalar multiplication. In the application to representation theory this space shall
provide the representation space of a representation. So we need a topology on it. And
here things start to be a bit technical. Following [Wa], we assume that X is locally
compact and for Γ(B) define the uniform on compacta topology. For our purposes it is
sufficient that in particular this leads to the consequence that each sequence (sn ) in Γ(B)
converges to a continuous cross section ϕ iff for each compact subset C ⊂ X one has
supx∈C sn (x) − s(x) −→ 0 for n −→ ∞. (In [Wa] the sequence is replaced by the more
general notion of net but we will not go into this here.)

The integration theory in this context is still more delicate. So for details, we also refer
to [Wa]. For our purpose it will suffice to fixa measure µ on X and think of Lq (B, µ)
as the space of sections s of B with sq := ( X s(x)q dµ(x))1/q < ∞. Then the inner
product on the space of sections is given by

< s, t > := < s(x), t(x) > dµ(x) for all s, t ∈ L2 (B, µ).
X

Let G be a linear group, H a closed subgroup, δ the quotient of the modular functions
of G and H treated in 7.1 (normalized by the condition δ(1) = 1) and π0 a unitary
representation of H on the Hilbert space V .

We define a continuous right action of H on the product space G × V as follows:

(g, a)h := (gh, δ(h)1/2 π0 (h−1 )a) for all g ∈ G, a ∈ V.

The orbit under H of a pair (g, a) ∈ G × E will be denoted by [g, a] := (g, a)H. The
space of all such orbits equipped with the quotient topology will be called E(G, H, π0 )
or simply E. This space is Hausdorff and the natural projection map

p : E −→ G/H =: X , [g, a] = (g, a)H −→ gH =: x

is a continuous and open surjection. Therefore E := (E, p) is a bundle over X = G/H,

which becomes a Hilbert bundle by the following natural assignments:
164 7. Induced Representations

We define addition and scalar multiplication in each fiber Ex := {[g, a]; a ∈ V } by

c1 [g, a1 ] + c2 [g, a2 ] := [g, c1 a1 + c2 a2 ] for all c1 , c2 ∈ C

and the scalar product by

< [g, a1 ], [g, a2 ] >Ex := δ(g)−1 < a1 , a2 > .

Obviously, there are some verifications to be done. Whoever does not want to do these
on his own can look into [Wa] p.380/1. The bundle is called the Hilbert bundle induced
by π0 (and δ) and denoted by E(π0 ) or again E. G acts continuously from the left on E
by
g0 [g, a] := [g0 g, a].
Finally, we rediscover the normalized induced representation π = indG H π0 of G as a
representation by unitary operators on the space of square integrable sections of E: To
this end, one first observes that there exists a one-to-one correspondence between cross
sections ϕ for E = E(π0 ) and functions φ : G −→ V fulfilling the functional equation

φ(gh) = δ(h)1/2 π0 (h−1 )φ(g) for all g ∈ G, h ∈ H.

The correspondence is given by

ϕ(gH) = (g, φ(g))H.

If φ and ϕ are connected in this way, one says that φ is the unitary form of ϕ and ϕ is the
bundle form of φ. Evidently a cross section ϕ is continuous on X = G/H iff its unitary
form φ is continuous on G. It is a bit less evident but true (see [Wa] p.381) that E(π0 )
admits enough cross sections. Therefore it makes sense to consider the Hilbert space

Hπ0 := L2 (E, µ)

where µ is the quasi-invariant measure on X associated with δ and the ρ−function on

X as in 7.1. (It can be shown that the construction is independent of the choice of the
function δ.) By a (small) abuse of notation, one denotes the space of bundle forms φ of
elements ϕ in the space of sections as well by Hπ0 . This space inherits the Hilbert space
structure and has as scalar product

< φ, φ̃ > := δ(s(x))−1 < φ(s(x)), φ̃(s(x)) > dµ(x)
X

where s denotes the Mackey section we introduced in 7.1.

The bundle theoretic definition of the unitarily induced representation π is as follows:

For g0 ∈ G and ϕ ∈ Hπ0 , put

π(g0 )ϕ(gH) := g0 ϕ(g0 −1 gH)

and correspondingly for the unitary form φ of ϕ

π(g0 )φ(g) = φ(g0−1 g).

Some routine verifications parallel to those we did in 7.1 (see [Wa] p.382) show that we
have got a unitary representation equivalent to the one constructed in 7.1. And one
7.6 Induced Representations and Vector Bundles 165

should get the feeling that all this is realized by the example of the (complex!) bundle
E = B k over X = P1 (C) from the beginning of this section. We will say more when we
come back to this at the end of the section.

As Warner remarks ([Wa] p.382), the preceding bundle-theoretic construction makes no

use of the local triviality of the constructed bundle. Moreover, “the definition of Hilbert
bundle is, strictly speaking, not in accord with the customaty version - however, it has
the advantage of allowing to place Hilbert space structures on the fibers without having
to worry about transition functions”. As we shall need this later, we present some ele-
ments of this other way to treat bundles.

Bundles, Second Approach

Here we need the notion of a differentiable manifold, which already was alluded to in 6.2
and is to be found in most text books on advanced calculus. We recommend in particular
the Appendix II in [Ki1] (or the Appendix A in [Be]).

Definition 7.3 : A p−manifold M (or a p-dimensional manifold) is a Hausdorff topo-

logical space that admits a covering by open sets Ui , i ∈ I, endowed with one-to-one
continuous maps ϕi : Ui −→ Vi ⊂ Rp .
M is called separable if it can be covered by a countable system of such open sets.
A smooth resp. analytic p-manifold is a separable p-manifold so that all maps

ϕi,j := ϕi ◦ ϕ−1
j |ϕj (Ui ∩Uj )

are infinitely differentiable (resp. analytic).

We use the following terminology:

– the pairs (Ui , ϕi ) (and often also the sets Ui ) are called charts;

– for m ∈ Ui =: U and ϕi =: ϕ the members x1 , . . . , xp of the p−tupel ϕ(m) =: x ∈ Rp

are called local coordinates;

– the functions ϕi,j are called transition functions;

– the collection (Ui , ϕi )i∈I is called an atlas on M .

It has to be clarified which atlases are really different: We say that two atlases (Ui )i∈I
and (U  j )j∈J are called equivalent if the transition functions from any chart of the first
atlas to any chart of the second one are smooth (resp. analytic). The structure of a
smooth (resp. analytic) manifold on M is an equivalence class of atlases.

When dealing with a smooth manifold, we use only one atlas keeping in mind that we
can always replace it by an equivalent one.

In parallel to the just given notions for real manifolds, one defines a smooth (or holo-
morphic) complex p-manifold where Rp is replaced by Cp and the transition functions
are required to be holomorphic, i.e. complex differentiable.
166 7. Induced Representations

Examples:

Example 7.16: M = Rp and open subsets are real p−manifolds.

Example 7.17: M ⊂ Rq is a p-dimensional submanifold if for any m ∈ M there are an

open set V ⊂ Rq and differentiable functions fi : V −→ R, i = 1, . . . , q − p, such that

– a) M ∩ V = {x ∈ V ; f1 (x) = · · · = fq−p (x) = 0},

∂x ) has rank q − p for all x.
– b) the functional matrix ( ∂f

The hyperboloids appearing as orbits in our discussion of the representations of the

Poincaré group in 7.5 are three-dimensional smooth submanifolds of R4 .

Example 7.18: M = P1 (C) (as in Example 7.3 in 7.1.5 and the beginning of this
section) is a smooth one-dimensional complex manifold:
We cover P1 (C) by two open subsets U := {(u : v); u, v ∈ C, v = 0} and U  := {(u :
v); u, v ∈ C, u = 0}. We have the homeomorphisms

ϕ : U −→ C, (u : v) −→ u/v =: w ∈ C,
(7.25) ϕ : U  −→ C, (u : v) −→ v/u =: w ∈ C.

And the map ϕ ◦ ϕ |{w=0} (w) = 1/w = w is complex differentiable. So here one has an
atlas consisting of two charts U and U  .

Certainly, at the same time, P1 (C) is an example for a two-dimensional real manifold.

Example 7.19: Show as Exercise 7.25 that M is a smooth p−dimensional (real)

manifold where

M = Pp (R) := {[u1 : · · · : up+1 ] = (u1 , . . . , up+1 )∼ ; (u1 , . . . , up+1 ) ∈ Rp+1 \{0}

with (u1 , . . . , up+1 ) ∼ (u 1 , . . . , u p+1 ) iff there exists a λ ∈ R∗ with u i = λui for all i.

If M is a smooth p−manifold and M  a smooth p −manifold, there is a natural notion

of a smooth map F : M −→ M  . Namely, for any point m ∈ M with F (m) = m and
any charts U
m and U 
F (m) = m , the local coordinates of m must be smooth
functions of local coordinates of m ∈ U .
A smooth map from one p−manifold to another, which has a smooth inverse map is called
a diffeomorphism. Two manifolds are called diffeomorphic if there is a diffeomorphism
from one to another.

A manifold is called oriented iff it has an oriented atlas, i.e. an atlas where for any two
charts with coordinates x and y the determinant of the functional matrix ( ∂x
∂y ) is positive
whenever it is defined.

As already said, there is much more to be discussed in this context (and we refer again to
e.g. [Ki1]) but, we hope, this rudimentary presentation suffices to grab an understanding
of our second approach to bundles:
7.6 Induced Representations and Vector Bundles 167

Definition 7.4: E is called a (fiber) bundle over the base X with fiber F iff E, F and X
are smooth manifolds and there is given a smooth surjective map p : E −→ X such that
locally E is a direct product of a part of the base and the fiber F .

More precisely, we require that any point x ∈ X has a neighbourhood Ux such that
EUx := p−1 (Ux ) can be identified with Ux ×F via a smooth map hx so that, for m ∈ EUx ,
one has hx (m) = (x , v) if x = p(m), i.e. the following diagram (where p1 denotes the
projection to the first factor) is commutative

h
p−1 (Ux ) −→
x
Ux × F

↓p ↓ p1
∼
Ux −→ Ux .

From the definition it follows that all sets Ex := p−1 (x), x ∈ X , called fibers, are smooth
submanifolds diffeomorphic to F .
F
One often writes E −→ X to denote a bundle with fiber F and base X .
F

Let E −→ X and E  −→ X  be
F
There is the natural notion of a map between bundles:
two bundles. Then a bundle map is given by the pair of smooth maps

Φ : E −→ E  , f : X −→ X 

such that the following diagram is commutative

E
Φ
E −→

p↓ ↓ p

f
X −→ X .

Two bundles over the same base X are called equivalent if there is a bundle map with
f = idX and Φ a diffeomorphism.
F F
A bundle E −→ X is called a trivial bundle if it is equivalent to the bundle X × F −→ X .

We are mainly interested in vector or line bundles where the fiber is F := V a vector
space (with dim V = n = 1 for the case of a line bundle) and the maps hx are linear in
each fiber p−1 (x ), x ∈ Ux .
Primarily, we think of real vector spaces and real manifolds but we also will treat complex
cases.

In any case, the dimension of the fiber as a vector space is called the rank of the vector
bundle.
168 7. Induced Representations

To make this more explicit and show a way to construct bundles, we take an atlas
of the real manifold X given by open sets Ui (i ∈ I), and V = Rn . Then we have
homeomorphisms
h := hi : EUi = p−1 (Ui ) −→ Ui × Rn ,
which a) respect the fibers, i.e. with

h(m) = (x, v) for all m ∈ Ex = p−1 (x), x ∈ Ui ,

and b) restrict to vector space isomorphisms

h |Ex : Ex −→ Rn .

We denote by

hi,j := hi ◦ h−1
j |(Ui ∩Uj )×Rn : (Ui ∩ Uj ) × R −→ (Ui ∩ Uj ) × R
n n

the bundle transition maps and introduce (differentiable) functions, which are called
bundle transition functions

ψi,j : Ui ∩ Uj −→ GL(n, R)  Aut (V )

by
hi,j (x, v) = (x, ψi,j (x)v) for all (x, v) ∈ (Ui ∩ Uj ) × Rn .
It is not difficult to see that these ψi,j are uniquely determined and for i, j, k with

Ui ∩ Uj ∩ Uk = ∅

fulfill the following cocycle condition

(7.26) ψi,j ◦ ψj,k = ψi,k .

Here, our main point is that we can construct bundles with fiber Rn over a manifold
with atlas (Ui )i∈I by pasting together columns Ui × Rn via such a given cocycle system
ψi,j . We have the following construction principle, whose validity is not too difficult to
prove.

Proposition 7.4: Let X be a manifold with atlas (Ui )i∈I and ψi,j : Ui ∩Uj −→ GL(n, R)
be a system of smooth functions with

ψi,i = 1 in Ui ,
ψi,j ◦ ψj,i = 1 in Ui ∩ Uj ,
ψi,j ◦ ψj,k ◦ ψk,i = 1 in Ui ∩ Uj ∩ Uk .

Let Ẽ be the union of the manifolds Ui × Rn . One says that points (xi , vi ) ∈ Ui × Rn
and (xj , vj ) ∈ Uj × Rn are equivalent if xi = xj and vi = ψi,j (xj )vj . Then the factor
space E of Ẽ defined by this equivalence relation is a vector bundle with fiber Rn over X
with projection p induced by the natural projections (xi , vi ) −→ xi . And each such fiber
bundle is equivalent to a bundle of this type.
7.6 Induced Representations and Vector Bundles 169

Proposition 7.5: Let E and E  be vector bundles of the same rank n over X defined

by the systems of bundle transition functions ψi,j and ψi,j (without loss of generality,
by a refinement argument, one can assume that both bundles are defined by the same
atlas (Ui )i∈I ). Then both bundles are equivalent iff there is a family of matrix functions
fi : Ui −→ GL(n, R) such that one has

ψi,j = fi−1 ψi,j fj for all i, j ∈ I.

As in our first approach, for open sets U ⊂ X we introduce (vector) spaces Γ(U, E) of
sections s over U , i.e. smooth maps s : U −→ E with p ◦ s = idU . For U = X we talk of
global sections.
Parallel to our treatment of real bundles, one can look at the complex case by replacing
appropriately R by C.

Examples:

Example 7.20: The most prominent example, which led to the introduction of the
notion of bundles, is the tangent bundle T M , which comes about if for an n-dimensional
manifold M with charts (Ui , ϕi ), i ∈ I, as matrix of bundle transition functions one takes
the Jacobian J of the manifold transition functions ϕi,j

ψi,j (m) := Jϕi,j (ϕj (m)).

Then T M can be thought of as consisting of classes of pairs (m, a)∼ , m ∈ M, a ∈ Rn ,

with (m, a) ∼ (m, b) iff a and b are related by the contravariant transformation property

n
∂yi
bi = (y)aj
j=1
∂x j

when m appears in the chart (U, ϕ) with coordinates x = (x1 , . . . , xn ) and simultane-
ously in the chart (U  , ϕ ) with coordinates y = (y1 , . . . , yn ). This has a geometrical
background: in first approximation, we take the manifold M and to each point m ∈ M
associate as a fiber Em the tangent space Tm M to M at this point m. From higher
calculus one knows that there are different ways to describe the elements Xm ∈ Tm M
i.e. the tangent vectors to a manifold. We follow the one already prepared in 6.2 while
discussing the definition of the Lie algebra associated to a linear group and think of the
tangent space of M at m as the space of all tangent vectors of all smooth curves γ inM
passing through m. More precisely, a tangent vector Xm to M at the point m is defined
as the equivalence class of smooth parametrized curves γ = γ(t) passing through m. Two
curves γ and γ̃ in M with γ(0) = γ̃(0) = m are equivalent iff their tangent vectors in m
coincide in the coordinates of a chart (U, ϕ) with m ∈ U , i.e. one has γ ∼ γ̃ if there is a
chart (U, ϕ) with m ∈ U and coordinates ϕ(m) = x such that

d d
ϕ(γ(t)) |t=0 = ϕ(γ̃(t)) |t=0 .
dt dt
With respect to this chart, a tangent vector can be symbolized by the n-tupel of real
d
numbers ai wherenai is just the i-th component of dt ϕ(γ(t))|t=0 or by the differential
operators Xm = i=1 ai ∂xi acting on smooth functions f in the coordinates x. Addition
and scalar multiplication of these operators resp. tuples define a vector space structure,
which can and will be carried over to the first given description of the space Tm M .
170 7. Induced Representations

If (U  , ϕ ) with coordinates y is another chart for m and (b1 , . . . , bn ) are the components
of the tangent vector to γ expressed in this coordinate system, by the chain rule we have
with ϕ (ϕ−1 (x)) = y(x)

d  d
∂yi
bi = ( ϕ (γ(t)) |t=0 )i = ( (ϕ (ϕ−1 (ϕ(γ(t)))) |t=0 )i = aj ,
dt dt j
∂xj

i.e. the contravariant transformation property from our general construction above.

Definition 7.5: Global sections X : M −→ T M of the tangent bundle T M are called

vector fields on M .

Let (U, ϕ) be a chart with coordinates x = (x1 , . . . , xn ). With a (now usual) slight misuse
of notation we write

n
X |U =: ai (x)∂xi
i=1

where the coefficients ai are smooth functions defined in U . For a chart (U  , ϕ ) with
coordinates y = (y1 , . . . , yn ), one has as well

n
X| = U
bj (y)∂yj
j=1

and on U ∩ U  = ∅ we have the contravariant transformation property

∂yj
bj (y) = (y) ai (x(y)).
i
∂xi

V (M ) denotes the vector space of smooth vector fields X on M .

∗
Example 7.21: The cotangent bundle to M has as its fibers the cotangent spaces Tm M,
i.e. the duals to the tangent spaces Tm M . It turns up if, in the same situation as above
in Example 7.20, as matrix of bundle transition functions we take the transpose of the
Jacobian of the map ϕj ◦ ϕ−1i =: ϕj,i , i.e.

ψi,j (m) = tJϕj,i (ϕi (m)).

The global sections α : M −→ T ∗ M of the cotangent bundle T ∗ M are called differential

forms of first degree or simply one-forms on M . As a parallel to the notation for vector
fields above, with functions a∗i and b∗ j defined in U resp. U  we write

α |U = a∗i (x)dxi , resp. α |U
= b∗j (y)dyj ,

i j

and here have the covariant transformation property

∂xi
(7.27) b∗j (y) = (x(y))a∗i (x(y)).
j
∂yj

Ω1 (M ) denotes the vector space of one-forms α on M .

7.6 Induced Representations and Vector Bundles 171

Example 7.22: The first two examples concerned real bundles but similarly one can
treat the complex case. Here we take the one-dimensional complex manifold M = P1 (C)
covered by the two charts (U, ϕ) and (U  , ϕ ) fixed above in (8.10) while discussing M
as a manifold. For k ∈ Z we get a line bundle Lk by choosing ψ(w) = w−k as bundle
transition function between these two charts. Let us work out the construction of this
bundle along the line of the general construction principle given above: We have two
“columns”, one over U and one over U  , namely

L := {(w, t), w, t ∈ C}, L := {(w , t ), w , t ∈ C}.

Using ψ we paste them together over U ∩ U  = C∗ by identifying (w, t) ∼ (w , t ) iff

w = 1/w, t = w−k t.

Remark 7.9: It can be shown that up to (holomorphic) equivalence there are no other
holomorphic line bundles over P1 (C) than these Lk .

Proposition 7.6: For the space of holomorphic sections one has

dimC Γ(Lk ) = k + 1 for k ≥ 0,

=0 for k < 0.

Proof: Let s be a section of Lk with

s : (u : v) −→  ∈ Lk ,

where h() = (w, t) ∈ L for w = u/v, v = 0, and h () = (w , t ) ∈ L for w = v/u, u = 0.
The section is holomorphic iff t is given by a holomorphic function in w and t by a
holomorphic function in w and we have the compatibility condition t = w−k t. As
examples and prototypes for such a holomorphic function we take the monomials t = wp
p

and t = w with p, p ∈ N0 . The compatibilty condition says for w =  0
p

t (w ) = w = wk t(w) = w−k+p


and, since one has w = 1/w, we get w−p = wp−k and see that everything is consistent
for p = 0, 1, . . . , k if k is not negative. And the holomorphicity of the relation breaks
down for negative k. Hence we have k + 1 essentially different holomorphic sections for
non-negative k and none for negative k. One needs a bit more complex function theory
to be sure that there are no more (see for instance [Ki] p.80).

Bundles and Representations

After this short digression into the world of bundles, which will be useful later, we return
to the problem of the construction of group representations. As we know from our first
approach to bundles, the space of sections is meant to provide a representation space
and, hence, should get the structure of a Hilbert space. In general this needs some
work, which we shall attack in the next chapter. But if the space of sections is a finite-
dimensional vector space, this is no problem. To get an idea we return to our example
G = SU(2) ⊃ H  U(1) and the unitary representation πj = indG H π0 induced from the
representation π0 = χk , k = 2j ≥ 0, of H  U(1). In our first approach we constructed
172 7. Induced Representations

the bundle space Bk = {[g, t]; g ∈ G, t ∈ C} with projection pr onto X = G/H = P1 (C)
given by  
a b
pr([g, t] := gH = (u : v) = (b : ā) for g = .
−b̄ ā
We define a map of Bk to the line bundle Lk constructed by pasting together the sets
L = {(w, t); w, t ∈ C} and L = {(w .t ); w , t ∈ C} via the bundle transition function
ψ(w) = wk as follows: For g with a = 0 we put

B k
[g, t] −→ (w = b/ā; b̄k t) ∈ L

and for g with b = 0

B k
[g, t] −→ (w := ā/b, ak t) ∈ L .
 
ζ
Both maps are well defined since for h = one has
ζ −1
 −1 
aζ bζ
gh−1 =
−b̄ζ −1 āζ

and
π0 (h) = ζ k .
Hence application of the mapping prescription to the element [gh−1 , π0 (h)t], which as
well represents [g, t], leads to the same images. And the images in L and L are consistent
with the pasting condition (w , t ) = (1/w, w−k t) for w = 0, since with w = b/ā one has
w−k (bk t) = āk t. Thus both maps induce a map of Bk to Lk , which obviously is a bijection.

Here we have a complex line bundle over the one-dimensional complex manifold P1 (C)
whose space of holomorphic sections is a complex vector space and the representation
space for the representation πj , 2j = k of G = SU(2). In principle, the base spaces for
the bundles in our first approach are real manifolds and the section spaces real vector
spaces. This example shows that we may get the wanted irreducible representation by
some more restrictions, namely we put more structure on the real construction to get the
more restrictive complex case (as in holomorphic induction). We shall see more of this
under an angle provided from the necessities of physics in the next chapter.
Chapter 8

Geometric Quantization and

the Orbit Method

The orbit method is an object of pure mathematics, but it got a lot of impetus from
physics under the heading of geometric quantization. Historically it was proposed by
Kirillov in [Ki2] for the description of the unitary dual of nilpotent Lie groups. By fur-
ther work of Kirillov and many others, in particular B. Kostant, M. Duflo, M. Vergne and
D. Vogan, it grew into an important tool for explicit constructions of representations of
various types of groups. The construction mechanism uses elements that also appear in
theoretical physics when describing the transition from classical mechanics to quantum
mechanics. We therefore start by recalling some of this background, even though we
cannot give all the necessary definitions.

8.1 The Hamiltonian Formalism and its Quantization

The purpose of theoretical mechanics is to describe the time development of the state
of a physical system. In classical mechanics such a state is identfied with a point in
an n-dimensional manifold Q, the configuration space. The point is usually described by
local coordinates q1 , . . . , qn , called position variables. The time development of the system
then is described by a curve γ given in local coordinates as t −→ (qi (t)), which is fixed
by an ordinary differential equation of second order. In the Hamiltonian formalism one
adds to the position variables q = (q1 , . . . , qn ) the momentum variables p = (p1 , . . . , pn )
(related to the velocity of the particle or system, we do not need to be more exact here)
and the time development t −→ (q(t), p(t)) is described by the hypothesis that the curve
t −→ (q(t), p(t)) =: γ(t) has to fulfill Hamilton’s equations, i.e. the first order system of
partial differential equations

dqi ∂H dpi ∂H
(8.1) q̇i := = , ṗi := =−
dt ∂pi dt ∂qi
where the function H = H(q, p, t) is the Hamiltonian of the system, i.e. encodes all the
relevant information about the system. For instance, if our system is simply a mass one
(point) particle in R3 moving
under the influence of the force with a radial symmetric
potential V (q) = 1/r, r = q12 + q22 + q32 , one has H = p21 + p22 + p23 + V (q).
174 8. Geometric Quantization and the Orbit Method

This Hamilton formalism has several interpretations:

1. The Geometric Picture

A curve γ = γ(t) is called an integral curve for the vector field

X = a(q, p)∂q + b(q, p)∂p := a1 (q, p) ∂q1 + . . . an (q, p) ∂qn + b1 (q, p) ∂p1 + · · · + bn (q, p) ∂pn

iff at each point of the curve its tangent vector is a vector from the vector field. We write
this as
γ̇(t) = X(γ(t)).
The statement that γ is an integral curve for the Hamiltonian vector field
∂H ∂H
X = XH with ai = , bi = − ,
∂pi ∂qi

is an expression of the assertion that γ fulfills Hamiltons equations (8.1).

2. The Symplectic Interpretation

To state this, we have to provide some more general notions about differential forms,
which we will need anyway in the following sections. Hence we enlarge our considera-
tions about the tangent and cotangent bundle and their sections from the last section:
We look again at a (real) smooth n-manifold M .
– A vector field X is defined as a section of the tangent bundle T M and in local coordi-
nates x = (x1 , . . . , xn ) written as

X = a(x)∂x := a1 (x)∂x1 + · · · + an (x)∂xn .

– A differential form of degree one or one-form α is defined as a section of the cotangent

bundle T ∗ M and in local coordinates written as

α = a(x)dx := a1 (x)dx1 + · · · + an (x)dxn .

As one knows from calculus, there also are higher order differential forms of different
types. We shall need the symmetric and the antisymmetric or exterior forms.
– A symmetric r-form β is written in the local coordinates from above as

β := bi1 ...ir (x)dxi1 · · · · · dxir

where the coefficients b are (smooth) functions of the coordinates x and the differentials
dxi are commuting symbols, i.e. with dxi · dxj = dxj · dxi for all i, j, the sum is meant
over all r-tuples i1 , . . . , ir ∈ {1, . . . , n} and the coefficients b are invariant under all
permutations of the indices. For instance, if n = 2, one has the two-form

β = b11 dx21 + b12 dx1 dx2 + b21 dx2 dx1 + b22 dx22 , b12 = b21 .

Certainly, there is the possibility to introduce a reduced form where one restricts the
index-tuples by taking exactly one representative from each orbit of the symmetric group
Sr acting on the index-tuples. Hence, as reduced form one has in the example above

β = b11 dx21 + b̃12 dx1 dx2 + b22 dx22 , b̃12 = 2b12 .

8.1 The Hamiltonian Formalism and its Quantization 175

If one wants to be more sophisticated (and precise), one says that a symmetric r−form
is a (smooth) section of the r−th symmetric power of the cotangent bundle. But for our
treatment here and for the antisymmetric case below the naive description above should
be sufficient.

An antisymmetric or exterior r−form α or form of degree r is given in local coordinates

by

α= ai (x)dxi1 ∧ · · · ∧ dxir
where here the differentials are anticommuting, i.e. with dxi ∧ dxj = −dxj ∧ dxi for all
i, j. The summation is again over all r−tuples i = (i1 , . . . , ir ) and the coefficients have
the special property of antisymmetry

ai = sgn σ aσi for all i = (i1 , . . . , ir ), σ ∈ Sr ,

hence, in particular one has ai = 0 if two of the indices coincide. If we take in our
example n = r = 2, we get

α = a12 dx1 ∧ dx2 + a21 dx2 ∧ dx1 , a12 = −a21 .

It is very convenient to introduce here the reduced form where the sum is taken only over
all r−tuples (i) = (i1 , . . . , ir ) with 1 ≤ i1 < · · · < ir ≤ n. Hence the reduced form of our
example is
α = ã12 dx1 ∧ dx2 , ã12 = 2a12 .
After changing coordinates the differential forms are transformed according to an ap-
propriate generalization of the transformation property (7.27) in Example 7.21 in 7.6
and the complete definition is again given by characterizing exterior r−forms as smooth
sections of the r−th exterior power of the cotangent bundle.

The set of these exterior r−forms on U ⊂ M is denoted by Ωr (U ). By the operations

on the coefficients one has a natural manner to define a vector space structure on Ωr (U )
and one has the possibility to do scalar multiplication with functions. Moreover one has
an associative multiplication if one multiplies a p−form with a q−form distributively re-
garding the anticommutativity of the dxi . It is decisive for the application of differential
forms that we have still two more operations, the differentiation d augmenting the degree
by one and the inner multiplication i(X) with a vector field X lowering the degree by
one (moreover there is the Lie derivative, which we will not consider for the moment).
We will have to be content again with the description in local coordinates.

For α = a(i) dxi1 ∧ · · · ∧ dxir we define the exterior differentiation dα by

dα := da(i) ∧ dxi1 ∧ · · · ∧ dxir

where da denotes the usual total differential of a function a, i.e. is given by

n
∂a
da := dxi .
i=1
∂xi

To handle the exterior differential, one usually reorders it to its reduced form.
176 8. Geometric Quantization and the Orbit Method

A simple but most important example appears if we  look at the coordinates given by
n
x = (q1 , . . . , qn , p1 , . . . , pn ) and take the one-form α = i=1 qi dpi . We get

n
(8.2) ω0 := dα = dqi ∧ dpi
i=1

and, obviously, have dω0 = 0.

We note that the relation ddα = 0 holds in general for all α.

A form α is called closed iff one has dα = 0 and it is called exact iff there is a form β
such that α = dβ.

Moreover, we point to the fact that exterior differentiation commutes with substitution
of the variables and, if Φ : M −→ M  is a smooth map of manifolds and α a form on M  ,
then one can draw back α to a form Φ∗ α on M . Without being more pecise here, for
 ∂yi of the local coordinates x −→ y = y(x)
practical purposes this comes out by substitution
and using the well known formula dyi = j ∂x j
dxj . We do not have enough space to go
into this more deeply but later on shall see in examples how this works.
 
For α = a(i) (x)dxi1 ∧ · · · ∧ dxir and the vector field X = bj (x)∂j , we have an inner
product i(X)α, which comes out (this certainly is not the most elegant definition, but
avoids more generalities) by applying successively the prescriptions

i(X)dxi = bi (x)
and
i(X)(α ∧ β) = (i(X)α) ∧ β + (−1)r α ∧ (i(X)β) for α ∈ Ωr , β ∈ Ωq .
To show how this works, we give an example, which is the most interesting n for us: As
above,
nwe take coordinates x = (q1 , . . . , qn , p 1 , . . . , pn ) and α = ω = i=1 dq i ∧ dpi and
X = i=1 (ai (q, p) ∂qi + bi (q, p) ∂pi ). Then we have

n
i(X)ω = (ai dpi − bi dqi ).
i=1

Finally, we use this for the symplectic formulation of the Hamilton formalism: We go
from the configuration space Q with coordinates q to its cotangent bundle M = T ∗ Q,
calledthe phase space, with coordinates (q, p). We provide this space with the two-form
n
ω = i=1 dqi ∧ dpi (and hence get the protype of a symplectic manifold, which will be
analysed later more thoroughly). Now we can say that a curve γ provides a solution to
Hamilton’s equations (8.1) if it is an integral curve of the vector field X, which fulfills
the equation

(8.3) dH = i(X)ω.

Namely, in this case one has

∂H ∂H
= −bi , = ai ,
∂qi ∂pi
i.e. X is a Hamiltonian vector field XH .
8.1 The Hamiltonian Formalism and its Quantization 177

3. The Poisson Interpretation

In 6.1 we introduced the Poisson bracket for two smooth functions f and g with coordi-
nates (q, p)

n
∂f ∂g ∂f ∂g
{f, g} := − .
i=1
∂q i ∂p i ∂p i ∂qi

Hence Hamilton’s equations (8.1) can also be expressed as

q̇ = {q, H}, ṗ = {p, H}.
And more generally, if one replaces the special observables positions qi and momenta
pi by a more general observable f , i.e. a smooth function of (q, p), as equation of time
development one has
(8.4) f˙ = {f, H}.
Here we find an appropriate starting point to do quantization, i.e. to go over to a quantum
mechanical description of the system. A quantum mechanical system is modeled by a
complex Hilbert space H. The state of the system is represented by a one-dimensional
subspace L = < ψ > of H, ψ ∈ H \ {0}, which can be seen as a point ψ∼ in the projective
space
P(H) := {ψ∼ ; ψ ∈ H \ {0}}, ψ ∼ ψ  iff ψ  = λψ for a λ ∈ C∗ .
Then the time development reflects in a curve γ(t) = ψ(t)∼ in P(H) and the physical
law determining this development is encoded into a one-parameter group G of unitary
operators U (t), t ∈ R, such that γ is a G−orbit in P(H). A physical observable f should
correspond to an operator B = Bf acting on or in H (as there are subtle problems,
we will not be more precise here) and < Bψ, ψ > provides the information about the
probability to measure the observable in the state < ψ >. Similar to our treatment of
infinitesimal generators of one-parameter subgroups in section 6.2, one can look for a
skew-adjoint operator A such that (in some precise sense) one has U (t) = exp tA and,
hence, the counterpart of the classical equation of motion will be
dψ
= Aψ.
dt
In the physics literature, usually one introduces the (self-adjoint) Hamilton operator Ĥ
by another normalization, namely

U (t) =: eitĤ
(we use units such that Planck’s constant is one) and then has the Schrödinger equation
dψ
= iĤψ.
dt
In our standard example of a system consisting of a particle with mass m moving in
R3 in a central force field with potential V, one translates the position variables qj from
the classical world into operators Bqj acting by multiplication with iqj , the momentum
variables pj go to the differential operators Bpj = ∂qj acting on (a subspace of smooth
functions in) the Hilbert space H = L2 (R3 ) and one has the Schrödinger equation

dψ
3
= iĤψ = i((1/2m) ∂q2j − V (q))ψ.
dt j=1
178 8. Geometric Quantization and the Orbit Method

In general, it can be a serious problem to find a Hilbert space H and operators corre-
sponding to the symplectic space and the observables from the classical description. For
some discussion of this we refer to Chapter 5 in [Be]. For our purposes here, we retain
the following observation even though it is somewhat nebulous:

In important cases the behaviour of a physical system is guided by its symmetry group G,
i.e. a group of transformations, which do not change the physical content of the system.
If we determine unitary representations of G we will be provided with a Hilbert space
and a lot of unitary operators.

In the next sections we shall see that important groups, like the Heisenberg group, the
Euclidean and the Poincaré group, SU(2) and SL(2, R), give rise to symplectic manifolds
and line bundles living on these such that the spaces of sections of these bundles are
representation spaces of unitary representation of G.

8.2 Coadjoint Orbits and Representations

We start by analysing the construction of bundles on symplectic manifolds. Beside the
already mentioned sources [Ki] and [Ki1], for further studies we recommend the book by
Woodhouse [Wo]. We return to the realm of mathematics though we still use some terms
from physics, which should help to give some motivation.

8.2.1 Prequantization
In Section 8.1 we already established the cotangent bundle T ∗ Q to a manifold Q and in
particular the space Rn with coordinates (q1 , . . . , qn , p1 , . . . , pn ) and provided with the
standard two-form (8.2)

n
ω0 := dqj ∧ dpj
j=1

as prototypes of a symplectic manifold. In general, one has the fundamental notion:

Definition 8.1: A smooth (real) p-manifold M is called a symplectic manifold if it is

provided with a closed nondegenerate symplectic two-form ω defined on M , that is an
ω ∈ Ω2 (M ), such that

– i) dω = 0,
and
– ii) on each tangent space Tm M, m ∈ M, the condition holds: if for X ∈ Tm M one has

ωm (X, Y ) = 0 for all Y ∈ Tm M,

then X = 0.
8.2 Coadjoint Orbits and Representations 179

Slightly paraphrasing the condition ii), we remark that M neccessarily has to have even
dimension p = 2n and that for ω in m ∈ M with local coordinates (q1 , . . . , qn , p1 , . . . , pn )
given by (8.2)

n
ω0 = dqj ∧ dpj
j=1

and for vector fields X, Y given by

n
X= (aj (q, p) ∂qj + aj (q, p) ∂pj ), Y = (bj (q, p) ∂qj + bj (q, p) ∂pj ),
j=1 j=1

from the general duality formula dxi (∂xj ) = δij and the antisymmetry of the forms one
has the relation

n
ω0 (X, Y ) = (aj (q, p) bj (q, p) − aj (q, p) bj (q, p)).
j=1

At first sight, this form ω0 may look very special. But there is a beautiful theorem stating
that this really a general situation:

Theorem 8.1 (Darboux): For every point m of a symplectic manifold (M, ω) of dimen-
sion 2n, there is an open neighbourhood U of m and a smooth map

ϕ : U −→ R2n with ϕ∗ ω0 = ω|U ,

where ω0 is the standard symplectic form (8.2) on R2n .

The examples of symplectic manifolds, which are of interest for us, are the coadjoint
orbits belonging to the action of a group G on the dual g∗ of its Lie algebra g. For
instance the projective space P1 (C) will come out as such an orbit. Before we go into
this, we elaborate a bit on the way to construct bundles on symplectic manifolds. As we
see it, this construction essentially goes back to Weil’s book Variétés Kählériennes [We].

Definition 8.2: A symplectic manifold (M, ω) is called quantizable if ω is integral.

The integrality of ω can be defined in several equivalent ways. Perhaps the most elemen-
tary one is to require (as in [Wo] p.159]:

(I.) The integral of ω over any closed oriented surface in M is an integral multiple of 2π.

More elegant is the requirement

(II.) ω determines a class [ω] in H 2 (M, Z).

To understand (II.), one has to have at hand the notions of the second cohomology group
H 2 (M, R) and its integral subgroup H 2 (M, Z). Though we are not too far from the
possibility to define these, we renounce to do it here but discuss the construction of the
prequantum line bundle where we can see how this condition works (later we shall be led
to a more practical criterion to decide the quantizability of the manifolds showing up as
coadjoint orbits):
180 8. Geometric Quantization and the Orbit Method

Recalling our definitions from 7.6, this bundle can be constructed as follows. We take a
contractible open cover U = (Uj ) of M . Then Poincaré’s Lemma assures that on each
Uj , there exists a real one–form βj such that

ω = dβj .

The same way, one knows that on each contractible Uj ∩ Uk = ∅ a smooth real function
fjk = −fkj exists such that
dfjk = βj − βk

and, for each contractible Uj ∩ Uk ∩ U = ∅, there is a constant ajk such that

2πajk = fjk + fkl + fj on Uj ∩ Uk ∩ U .

It is the decisive point that the integrality of ω says that the ajk may be assumed to be
integers (this is really the essence of the definition above). If we put

cjk := exp ifjk ,

we have
dcjk = icjk (βj − βk )

and for ajkl ∈ Z on each non–empty Uj ∩ Uk ∩ U

cjk ckl clj = exp(2πiajk ) = 1.

Hence Proposition 7.4 in 7.6 shows that the cjk are transition functions for a line bundle
B over M . This bundle has a compatible Hermitian structure <, > (see [Wo] p.264),
which we define below in 8.2.3. Since in Poincaré’s Lemma a form β with dβ = α is
not unique (it can changed by an exact form), this construction is not unique. The gen-
eral statement is that these constructions of prequantum bundles B are parametrized by
H 1 (M, R)/H 1 (M, Z) ([Wo] p.286).

As it is our goal to construct unitary group representations, we need a Hilbert space.

The Hilbert space H in our prequantization construction (the prequantum Hilbert space)
is the space of all s ∈ Γ(M, B), i.e. smooth global sections of B for which in some sense
an integral of (s, s)εω over M exists and is finite. Here the measure εω is associated to
the Liouville form
1 n
(−1)n(n−1)/2 ω
n!

known from calculus. For s1 , s2 ∈ H the inner product on H is


1
< s1 , s2 > := (s1 , s2 )εω .
(2π)n
M
8.2 Coadjoint Orbits and Representations 181

8.2.2 Example: Construction of Line Bundles over M = P1 (C)

The topic of line bundles on projective space is treated in each text book on algebraic
geometry (for instance, see Griffiths-Harris [GH] p.144) and we already constructed holo-
morphic line bundles over P1 (C) at the end of section 7.6. But to illustrate the procedure
proposed above, for those who – like the author – enjoy explicit descriptions, we add here
the following very elementary constructions (to be skipped if one is in a hurry).

1. Real Analytic Bundles B −→ P1 (C)

We look at P1 (C) as a real two-dimensional manifold and take the covering U = (Uj )j=1,...,6
with
U1 = {(z : 1); z = x + iy ∈ C; |z| < 1},
U2 = {(1 : w); w = u + iv ∈ C; |w| < 1}, w = 1/z = (x − iy)/(x2 + y 2 ),
U3,4 = {(z : 1); z ∈ C; Re z = x ≷ 0},
U5,6 = {(z : 1); z ∈ C; Im z = y ≷ 0}.

This covering is chosen such that all intersections of the covering neighborhoods are
simply connected (this is not the case for the covering by U and U  , which we used in
Example 7.18 in 7.6 as one has U ∩ U   C∗ ). With c = /π ∈ R we take as a symplectic
form on M
dx ∧ dy i dz ∧ dz
ω := c =c for (z : 1) ∈ Uj , j = 2,
(1 + x2 + y 2 )2 2 (1 + |z|2 )2
du ∧ dv i dw ∧ dw
:= c =c for (1 : w) ∈ U2 .
(1 + u2 + v 2 )2 2 (1 + |w|2 )2
Exercise 8.1: Verify that this two–form ω in the appropriate coordinates has potential
forms θz and θw with dθz = dθw = ω given by
 
c xdy − ydx i zdz − zdz
θz = = −c for (z : 1) ∈ Uj , j = 2,
2 1 + x2 + y 2 4 1 + |z|2
 
c udv − vdu i wdw − wdw
θw = = −c for (1 : w) ∈ U2 .
2 1 + u2 + v 2 4 1 + |w|2

On Uj ∩ U2 (j = 2) we have   
i zdz − zdz 1 dz dz
θz − θw = −c − −
4 1 + |z|2 1 + |z|2 z z
 
i zdz − zdz
= −c
4 |z|2
c xdy − ydx
=
2 x2 + y 2
c ˜
= df2j
2
with a potential function f˜2j . If we fix the tangens as a function on (−π/2, π/2) and
f = arctan as its inverse, we can take

−f˜32 (x, y) = f (y/x) = arctan(y/x)

and moreover, for instance,

−f˜52 (x, y) = −f (x/y) + π/2 = − arctan(x/y) + π/2 and − f˜42 (x, y) = π + arctan(y/x).
182 8. Geometric Quantization and the Orbit Method

Then, on U2 ∩ U4 ∩ U5 , we have

f˜24 + f˜45 + f˜52 = −π − arctan(y/x) + π/2 − arctan(x/y)

= −π,
and similarly on the other intersections. Thus, we see that ω is integral in the sense
i) ω = dθj on Uj
ii) θj − θk = dfjk on Uj ∩ Uk
iii) fjk + fk + fj = ajk ∈ 2πZ in Uj ∩ Uk ∩ U exactly for c ∈ 4Z.
This result coincides with the usual volume computation
  ∞ 2π
dx ∧ dy rdrdδ
ω = c = c
(1 + x2 + y 2 )2 (1 + r2 )2
P1 (C) 0 0
∞ ∞
rdr du
= 2πc ”s = πc = πc
(1 + r2 )2 u2
0 1

telling that the volume is an integral multiple of 4π = vol(S 2 ) exactly for c = 4n, n ∈ Z.

Moreoverer, we see that we can construct the C ∞ –prequantum bundle

π M = P1 (C)
B −→
in the standard way as

B = {(m, v, j)∼ ; m ∈ M, v ∈ C, j ∈ I},

where here I = {1, . . . , 6} and

(m , v  , j  ) ∼ (m, v, j)

exactly for
m = m, v  = ψj
j (m)v.
The bundle transition functions ψjk are fixed (in the charts with coordinates x, y) by

ψjk (x, y) = cjk eifjk (x,y) ,

i.e. the different trivializations are glued together by


ei argz .

By the way, the transition function z  comes out if one does not take real potential forms
as above but complex forms like θ z = (1 + |z|2 )−1 z̄dz.

2. Holomorphic Bundles E −→ P1 (C)

Following Weil ([We] p. 90), the same construction can be repeated by looking at P1 (C)
as a one-dimensional holomorphic manifold and searching for holomorphic transition
functions
Fjk : Uj ∩ Uk −→ C∗ .
8.2 Coadjoint Orbits and Representations 183

Here we take the same covering U = (Uj )j=1,...,6 as above and the symplectic form
 dz ∧ dz 
ω := = ddΦj forj = 2,
2πi (1 + |z|2 )2 2πi
with
Φj (z) :=  log(1 + |z|2 )
resp. for j = 2
 dw ∧ dw 
ω := = ddΦ2
2πi (1 + |w|2 )2 2πi
with
Φ2 (w) :=  log(1 + |w|2 ).
On Uj ∩ U2 , j = 2, we have
Φj (z) − Φ2 (1/z) =  log(|z|2 ).
Fixing appropriate branches of the complex log-function, we take for j = 2

f2j := log z.
2πi
Then we have
Φj (z) − Φ2 (1/z) = −2πi(f2j − f 2j )
and
   z dz
ηj = − dΦj = d log(1 + |z|2 ) =
2πi 2πi 2πi 1 + |z|2
with
 dz ∧ dz
ω= .
2πi (1 + |z|2 )2
Using Weil’s notation, for the construction of a holomorphic bundle E = E we have
transition functions
Fjk (z) = exp (2πi fjk (z)) = z 
exactly if  ∈ Z. We see immediately that for  ∈ N0 the holomorphic sections are given
in the z-variable by
s(z) = z λ , λ = 0, 1, . . . , 
(as exactly in these cases we have (1/w)λ w holomorphic in w), i.e. we find
Γ (E ) ∼
= C+1 .
As described in a general procedure below, this space consists of the polarized sections
of the prequantum bundle B = B , where the polarization is given by assigning to the
point m with coordinate z the complex subspace spanned by ∂z in the complexification
Tm P1 (C) ⊗ C  C ∂z ⊕ C ∂z
of the tangent space Tm P1 (C) for each m ∈ P1 (C), i.e. Γ(E ) can be identified with the
space of those sections of B , which are constant “along the direction of the z̄−coordinate”.

We try to make this a bit more comprehensive. But to do it, again we need more general
notions from complex and algebraic geometry. As seen above, these geometric notions
can and shall be made later rather tangible by replacing them by algebraic ones in our
concrete examples (and, hence, may be skipped at first try).
184 8. Geometric Quantization and the Orbit Method

8.2.3 Quantization

The prequantization Hilbert space H built from C ∞ –sections of the prequantum bundle
B over M - as we just have seen - often turns out to be too large to be useful: This is to be
seen again in the case of the coadjoint orbits, as the representations of the group G on H
as a representation space are in general far from being irreducible. From representation
theory procedures are known to construct appropriate representation spaces, for instance
by parabolic or holomorphic induction. But in this approach here, it is “natural” to use
a notion from the Kähler geometry to single out a subspace of H, namely the notion of
the Kähler polarization. We take over from [Wo] p.92 ff:

Definition 8.3: A complex polarization on a 2n–dimensional symplectic manifold (M, ω)

is a complex distribution P with
C
(CP1) For each m ∈ M , Pm is a complex Lagrangian subspace of Tm M , i.e. with
⊥
Pm = Pm ,
(CP2) Dm := Pm ∩ Pm ∩ Tm M has constant dimension for all m ∈ M ,
(CP3) P is integrable.
The polarization P is of type (r, s) if the Hermitian form <, > on Pm defined by

< Z, W > := −4iω(Z, W ) for Z, W ∈ Pm

is of sign (r, s) for all m ∈ M .

P is a Kähler polarization if it is of type (r, s) with r + s = n. Such a polarization induces
a complex structure J on M and (M, ω, J) is a Kähler manifold (in the broad sense that
the real symmetric tensor g determined by

g(X, Y ) = 2ω(X, JY ), X, Y ∈ V (M )

may be indefinite, see the “note 1” to p.93 in [Wo]).

Perhaps the reader wants some more explanations. We follow [Wo] p.269:

– A real distribution on a real manifold M is a subbundle of the tangent bundle T M .

The fiber Pm at m ∈ M is a subspace of Tm M , which varies smoothly with m.
– A vector field X is tangent to P if X(m) ∈ Pm for every m. The space of vector
fields tangent to P is denoted by VP (M ). A smooth function is constant along P if df
vanishes on restriction to P . The space of smooth functions constant along P is denoted
by CP∞ (M ).
– An immersed submanifold N ⊂ M is an integral manifold of P if Pm = Tm N for every
m ∈ N.
– A distribution is integrable if [X, Y ] is tangent to P whenever X and Y are tangent
to P .
– A complex distribution is similarly defined to be a subbundle of the complexified tan-
gent bundle TC M . A complex distribution P is integrable if in some neighbourhood of
each point, one has smooth complex functions fk+1 , . . . , fl with gradients that are inde-
pendent and annihilate all complex vector fields tangent to P , where l − k is the fiber
dimension of P .
8.2 Coadjoint Orbits and Representations 185

– A complex structure on a real vector space V is a linear transformation J : V −→ V

such that J 2 = −1. Complex structures exist only in spaces of even dimension. On a
real manifold a complex structure J is given if one has a complex structure Jm on each
tangent space Tm M and if these Jm vary smoothly with m, i.e. the complex distribution
P spanned by the vector fields X − iJX, X ∈ V (M ), is integrable. A real manifold with
a complex stucture comes out as a complex manifold.
– A connection  on a vector bundle B is an operator that assigns a one-form s with
values in B to each (smooth) section s of B, i.e.,

 : Γ(B) −→ Γ(T ∗ M ⊗ B),

such that for any function f and sections s, s ∈ CB∞ (M ) one has

(s + s ) = s + s and  (f s) = (df )s + f  s.

Vice versa, a given Kähler manifold (M, ω, J), i.e. a manifold which is simultaneously in
a compatible manner a symplectic and a complex manifold, carries two Kähler polariza-
tions: the holomorphic polarization P spanned at each point by the vectors (∂za ) and its
complex conjugate, the anti–holomorphic polarization P spanned by the (∂za )’s. Locally
it is possible to find a real smooth function f such that

ω = i∂∂f

and
Θ = −i∂f resp. Θ = i∂f
is a symplectic potential adapted to P resp. P .
– A Hermitian structure on a complex vector bundle B is a Hermitian inner product
(·, ·) on the fibers, which is smooth in the sense that B −→ C, v −→ (v, v) is a smooth
function. It is compatible with the connection  if for all sections s, s and all real vector
fields X
i(X)d(s, s ) = (X s, s ) + (s, X s ), X s := i(X)  s.

Now, let be given a Kähler polarization P on the quantizable (M, ω) with prequantum
line bundle B over M with connection , Hermitian structure <, > and Hilbert space H
of square integrable sections of B. Then we define a space of polarized sections
∞ ∞
CB (M, P ) := {s ∈ CB (M ); X s = 0 for all X ∈ VC (M, P )} .

Here, for U ⊂ M open,

VC (U, P ) := {X ∈ VC (M ), Xm ∈ Pm for all m ∈ M }

denotes the vector fields tangent to P on U . Then

∞
HP := H ∩ CB (M, P )

comes out as our “new” Hilbert space. As remarked above, it consists of all integrable
holomorphic sections of B.
Alternatively, it can be said that the prequantum bundle B (via P ) is given a structure
of a complex line bundle. Anyway, in important cases, this HP is the representation
space of a discrete series representation.
186 8. Geometric Quantization and the Orbit Method

We leave aside the problem, which of the classical observables have a selfadjoint coun-
terpart acting in HP , discussed to some length in [Wo] and the modification by the
introduction of half–densities and go directly to the application in the construction of
unitary representations via line bundles on the coadjoint orbits.

8.2.4 Coadjoint Orbits and Hamiltonian G-spaces

This is the central topic of this chapter. For background information we recommend the
standard sources by Kirillov [Ki] §15, Kostant [Kos], (at least) the first pages of Kirillov’s
book [Ki1], or, for a comprehensive overview, the article by Vogan [Vo].

Let G be a (real) linear group with Lie algebra g and its dual space g∗ . As usual, we
realize all this by matrices and then have the adjoint representation Ad of G on g given
by
Ad(g)X := gXg −1 .
(This is the derived map of the inner automorphism κg : G −→ G, g0 −→ gg0 g −1 .)
Following our preparation in section 1.4.3, we can construct the contragredient represen-
tation Ad∗ defined on the dual g∗ to g by

Ad∗ (g) := Ad(g −1 )∗ .

The asterisk on the right-hand side indicates a dual operator: If < η, X > denotes the
value η(X) of the linear functional η applied to the vector X and A is an operator on g
the dual A∗ is defined by

< A∗ η, X >:=< η, AX > for all X ∈ g, η ∈ g∗ .

Hence we have as our central prescription

< Ad∗ (g)η, X > = < η, Ad(g −1 )X > for all g ∈ G, X ∈ g, η ∈ g∗ .

This representation Ad∗ is called the coadjoint representation. Often we will abbreviate
g · η := Ad∗ (g)η.

In 6.6.1 we already worked with the trace form for A, B ∈ Mn (C) given by

(A, B) −→ Re Tr (AB) =:< A, B > .

(Here, unfortunately, one has a double meaning of < ., . >.) If g is semisimple or

reductive, one can use this bracket to identify g∗ with g (G-equivariantly) via

g∗
η −→ Xη ∈ g, defined by η(Y ) = < Xη , Y > for all Y ∈ g.

Then the coadjoint representation is fixed by the simple prescription Xη −→ gXη g −1 .

Now we are prepared to define the coadjoint orbit O of η ∈ g∗ by

Oη := G · η.

If Gη denotes the stabilizing group Gη := {g ∈ G; g · η = η}, one has Oη  G/Gη and if

we can identify g∗ with a matrix space as proposed above, we have

Oη = {gXη g −1 , g ∈ G}.
8.2 Coadjoint Orbits and Representations 187

It is an outstanding and most wonderful fact that our coadjoint orbits are symplectic
manifolds. We cite from [Vo] p.187:

Theorem 8.2: Suppose M = Oη is the coadjoint orbit of η ∈ g∗ and gη := Lie Gη . Then

one has
1.) The tangent space to M at η is

Tη M  g/gη .

2.) The skew-symmetric bilinear form

ωη (X, Y ) := η([X, Y ]) for all X, Y ∈ g

on g has radical exactly gη and so defines a symplectic form on Tη M .

3.) The form ωη makes M into a symplectic manifold.

The bilinear form ωη is called Kirillov-Kostant form, sometimes also Kirillov-Kostant-

Souriau form or any permutation of these names. It is also written as ([Ki] p.230 or [Wo]
p.52)

(8.5) ωη (η  )(ξX̂ (η  ), ξŶ (η  )) = < η  , [X̂, Ŷ ] > for all X̂, Ŷ ∈ g, η  ∈ Oη

where ξX̂ denotes the vector field associated to X̂ in the form of a differential operator
acting on smooth functions f on M by the definition
d
(8.6) (ξX̂ f )(η) := (f (exp tX̂ · η))|t=0 .
dt
For a proof we refer to [Ki] 15.2 or [Kos] Theorem 5.4.1. It is obvious that ωη is skew
symmetric. Also it is, in principle, not deep that a form given for the tangent space of
one point can be transported to the whole space if the space is homogeneous. But it may
be an intricate task to give an explicit description of ω in local coordinates. We shall
give an example soon how (8.5) can be used. The fact that ω comes out as a closed form
can be proved by direct calculation or by more sophisticated means as in [Ki1] p.5-10.

The coadjoint orbits still have more power, they are Hamiltonian G-spaces:

Definition 8.4: Suppose (M, ω) is a symplectic manifold and f ∈ C ∞ (M ) is a smooth

function with its Hamiltonian vector field Xf . Moreover suppose that G is a linear group
endowed with a smooth action on M , which respects the form ω. We say that M is a
Hamiltonian G-space if there is a linear map

µ̃ : g −→ C ∞ (M )

with the following properties:

– i) µ̃ intertwines the adjoint action of G on g with its action on C ∞ (M ).
– ii) For each Y ∈ g, Xµ̃(Y ) is the vector field by which Y acts on M .
– iii) µ̃ is a Lie algebra homomorphism.

µ̃ can be reinterpreted as a moment map

µ : G −→ g∗ , µ(m)(Y ) := µ̃(Y )(m) for all m ∈ M, Y ∈ g.

188 8. Geometric Quantization and the Orbit Method

Then the highlight of this theory is the following result to be found also in the above cited
sources. We shall not have occasion to use it here but state it to give a more complete
picture.

Theorem 8.3: The homogeneous Hamiltonian G-spaces are the covering spaces of coad-
joint orbits. More precisely, suppose M is such a space, with map µ̃ : g −→ C ∞ (M ) and
moment map µ. Then µ is a G-equivariant local diffeomorphism onto a coadjoint orbit,
and the Hamiltonian G-space structure on M is pulled back from that on the orbit by
the map µ.

If G is reductive and we have identified g with its dual g∗ as n × n−matrices, one has the
following classification of the coadjoint orbits, which later will be reflected in a mecha-
nism to construct representations of G.

Definition 8.5: An element X ∈ g is called nilpotent if it is nilpotent as a matrix, i.e. if

X N = 0 for N large enough.
X is called semisimple if X regarded as a complex matrix is diagonalizable.
X is called hyperbolic resp. elliptic if it is semisimple and all eigenvalues are real resp. purely
imaginary.
An orbit O = G · η inherits the name from X = Xη .

For instance, the real matrix  

0 1
X=
−1 0
 
i 0
is elliptic, since over C it is conjugate to . Now, this finally is the moment to
0 −i
discuss an example.

Coadjoint Orbits of SL(2, R)

We take
G = SL(2, R) ⊃ K = SO(2)
and
   
∗ x y+z
g  g = Lie G = U (x, y, z) := ; x, y, z ∈ R = < X, Y, Z >R
y−z −x

with      
1 1 1
X= , Y = , Z=
−1 1 −1

[X, Y ] = 2Z, [X, Z] = 2Y, [Y, Z] = −2X

and
X 2 + Y 2 − Z 2 = 3I2 .
1
Obviously, X and Y are hyperbolic, Z is elliptic, and F = ( ) as well as G = ( )
1
are nilpotent.
8.2 Coadjoint Orbits and Representations 189

One can guess the form of the orbits generated by these matrices, but to be on the safe
side we calculate:

a b
gU (x, y, z)g −1 =: U (x̃, ỹ, z̃) for g = ( )
c d

is given by

x̃ = (ad + bc)x + (bd − ac)y − (bd + ac)z,

(8.7) ỹ = (dc − ab)x + (1/2)(a2 + d2 − b2 − c2 )y + (1/2)(a2 + b2 − c2 − d2 )z,
z̃ = −(ab + cd)x + (1/2)(a2 − b2 + c2 − d2 )y + (1/2)(a2 + b2 + c2 + d2 )z.

Hence, for U = αX, α = 0, as a hyperbolic orbit OαX we have the one-sheeted hyper-
boloid in R3 given by
x2 + y 2 − z 2 = α2 .

For U = αZ, α ≷ 0, as elliptic orbits OαZ we have the upper resp. lower half of the
two-sheeted hyperboloid given by

x2 + y 2 − z 2 = −α2 , z ≷ 0.

For U = F and U = G, as nilpotent orbits OF and OG we have the upper resp. lower
half of the cone given by
x2 + y 2 − z 2 = 0, z ≷ 0.

Finally, for U = 0, there is the nilpotent orbit O0 consisting only of the origin in R3 .
Hence we have a disjoint decomposition of R3 into the orbits in their realization as
subsets of R3
  
R3 = OαX OαZ (OF ∪{0} ∪ OG ).
α>0 α≷0

It is an easy exercise to determine the stabilizing groups of the elements we chose to

generate our orbits. We get (again with α = 0)
  
a
GαX = ; a ∈ R∗ ,
1/a

GαZ = SO(2),
  
±1 b
GF = ; b∈R ,
±1
  
±1 0
GG = ; c∈R .
c ±1

From the general theory we have the prediction that the coadjoint orbits are symplectic
manifolds. We can verify this here directly:
190 8. Geometric Quantization and the Orbit Method

Let us take the elliptic orbit

OαZ = G · (αZ)  G/SO(2), α > 0.

We want to give an explict expression for the Kirillov-Kostant form ω in this case. Here
one has to be rather careful. We used the trace form for matrices < A, B >= Tr AB to
identify g∗ with g via η −→ Xη with < Xη , U >= η(U ) for all U ∈ g. Now, we write
g∗ =< X∗ , Y∗ , Z∗ >R with
     
1 1 1
X∗ = , Y∗ = , Z∗ = ,
−1 1 −1

and then have as non-vanishing duality relations

< X∗ , X >= 2, < Y∗ , Y >= 2, < Z∗ , Z >= −2.

We put
g∗
η = x∗ X∗ + y∗ Y∗ + z∗ Z∗
and
X̂ = xX + yY + zZ, Ŷ = x X + y  Y + z  Z ∈ g.
Hence, by the commutation relations for X, Y, Z, we have

[X̂, Ŷ ] = 2((xy  − x y)Z + (xz  − x z)Y − (yz  − y  z)X).

From the Kirillov-Kostant prescription we have

ωη (X̂, Ŷ ) = < η, [X̂, Ŷ ] >

= −2(yz  − yz  )x∗ + 2(xz  − x z)y∗ − 2(xy  − x y)z∗ ,

i.e. in particular for the elliptic orbit through η = kZ∗

ωkZ∗ (X̂, Ŷ ) = −(xy  − x y)k.

There are several ways to go from here to an explicit differential two-form ω on

M : = G · kZ∗
= {(x∗ , y∗ , z∗ ) ∈ R3 ; x2∗ + y∗2 − z∗2 = −k 2 }.

It is perhaps not the shortest one but rather instructive to proceed like this: One knows
that
dẋ ∧ dẏ
ω1 =
ẏ 2
is a G-invariant symplectic form on M  := H  G/SO(2) where

τ̇ = ẋ + iẏ := g(i)

with  
bd + ac 1 a b
ẋ = 2 , ẏ = 2 for g = .
c + d2 c + d2 c d
8.2 Coadjoint Orbits and Representations 191

The coadjoint G-action on M and the usual action of G on M  = H are equivariant (Exer-
cise 8.2: Verify this). From (8.7) we have the description of the elements (x∗ , y∗ , z∗ ) ∈ M
as

x∗ = −(ac + bd)k,
y∗ = (1/2)(a2 + b2 − c2 − d2 )k,
z∗ = (1/2)(a2 + b2 + c2 + d2 )k,

and, hence, a G-equivariant diffeomorphism

ψ : M −→ M  = H, (x∗ , y∗ , z∗ ) −→ (ẋ, ẏ)

given by
x∗ k
ẋ = , ẏ = .
y∗ − z∗ z∗ − y∗
We use ψ
to pull back ω1 to a symplectic form ω0 , which in local coordinates (x∗ , y∗ )
with z∗ = x2∗ + y∗2 + k 2 , i.e. z∗ dz∗ = x∗ dx∗ + y∗ dy∗ comes out as

dx∗ ∧ dy∗
ω0 = ψ ∗ ω1 = .
kz∗

(Verify this as Exercise 8.3.)

Now we have to compare this with the Kirillov-Kostant prescription above and determine
the right scalar factor to get the Kirillov-Kostant symplectic form ω. We shall see that
one has
dx∗ ∧ dy∗
ω=− .
z∗
Perhaps the more experienced reader can directly pick this off the Kirillov-Kostant pre-
scription. As we feel this is a nice occasion to get some more practice in handling the
notions we introduced here, we propose to evaluate the general formula (8.5)

ωη (η  )(ξX̂ (η  ), ξŶ (η  )) := < η  , [X̂, Ŷ ] > for all X̂, Ŷ ∈ g, η  ∈ Oη

in this situation. For the vector fields ξX̂ acting on functions f on M we have the general
formula
d
(ξX̂ f )(η) = f ((exp tX̂) · η)|t=0 .
dt
Putting gX̂ (t) := exp tX̂ one has
 
et
gX (t) = ,
e−t
 
1 t
gF (t) = ,
1
 
1
gG (t) = .
t 1
192 8. Geometric Quantization and the Orbit Method

And, putting A(g) for the matrix of the coadjoint action η −→ g · η, one deduces from
(8.7)
⎛ ⎞
1
A(gX (t)) = ⎝ (1/2)(e2t + e−2t ) (1/2)(e2t − e−2t ) ⎠ ,
(1/2)(e2t − e−2t ) (1/2)(e2t + e−2t )
⎛ ⎞
1 t −t
A(gF (t)) = ⎝−t (1/2)(2 − t2 ) (1/2)t2 ⎠ ,
−t (1/2)t 2
(1/2)(2 + t2 )
⎛ ⎞
1 −t −t
A(gG (t)) = ⎝ t (1/2)(2 − t2 ) −(1/2)t2 ⎠ .
−t (1/2)t 2
(1/2)(2 + t2 )

From (8.6) we get the differential operators (with Y = F + G, Z = F − G)

ξX = 2(z∗ ∂y∗ + y∗ ∂z∗ ),
ξY = −2(z∗ ∂x∗ + x∗ ∂z∗ ),
ξZ = 2(y∗ ∂x∗ − x∗ ∂y∗ ).

When we treat the functions f as functions in the two local coordinates (x∗ , y∗ ) with
z∗2 = x2∗ + y∗2 + k 2 , these operators reduce to
ξX = 2z∗ ∂y∗ ,
ξY = −2z∗ ∂x∗ ,
ξZ = 2(y∗ ∂x∗ − x∗ ∂y∗ ).

And a vector field in these local coordinates comes as

X̂ = α∂x∗ + β∂y∗ = xξX + yξY + zξZ
with
α = 2(zy∗ − yz∗ ), β = 2(xz∗ − zx∗ )
and analogously for
Ŷ = α ∂x∗ + β  ∂y∗ = x ξX + y  ξY + z  ξZ .
One has
αβ  − α β = 4((zx − xz  )y∗ z∗ + (xy  − yx )z∗2 + (yz  − y  z)x∗ z∗ )
and hence in (8.5) for ω = (1/z∗ )dx∗ ∧ dy∗ the left hand side is
(1/z∗ )dx∗ ∧ dy∗ (ξX̂ , ξŶ ) = (1/z∗ )dx∗ ∧ dy∗ (α∂x∗ + β∂y∗ , α ∂x∗ + β  ∂y∗ )
= 4((zx − xz  )y∗ + (xy  − yx )z∗ + (yz  − y  z)x∗ ).
And this is exactly the negative of the outcome of the right hand side evaluated for
η  = x∗ X∗ + y∗ Y∗ + z∗ Z∗
< η  , [X̂, Ŷ ] >= −4(xy  − x y)z∗ + 4(xz  − x z)y∗ − 4(xy  − yx )x∗ .
8.2 Coadjoint Orbits and Representations 193

Coadjoint Orbits for SU(2)

As already done several times, we take

 
a b
G := SU(2) = {g = ; a, b ∈ C, |a|2 + |b|2 = 1}.
−b a

Thus, as a (real) manifold, G is the unit 3–sphere S 3 ⊂ C2 .

One has  
g = su(2) = X ∈ M2 (C); X = −t X
3
= {X = (1/2) aj Hj ; a = (a1 , a2 , a3 ) ∈ R3 }
j=1

with (see (6.5) in Section 6.5)

     
−i −i −1
H1 = , H2 = , H3 = ,
i −i 1

and
[H1 , H2 ] = 2H3 , [H2 , H3 ] = 2H1 , [H3 , H1 ] = 2H2 .
su(2) is isomorphic to so(3) (see the exercises 6.4, 6.5 and 6.6 in Section 6.1) and, as
usual, we identify X ∈ su(2) and a = (a1 , a2 , a3 ) ∈ R3 with
⎛ ⎞
0 −a3 a2
A = ⎝ a3 0 −a1 ⎠ =: a1 A1 + a2 A2 + a3 A3 ∈ so(3)
−a2 a1 0

Using this identification the Lie bracket corresponds to the vector product in R3

[A, B] = a ∧ b

and the trace form <, > in M3 (R) to the scalar product (·, ·) in R3

< A, B > = −2(a, b) = −2 t ab.

Since <, > is proportional to the euclidean scalar product on R3 , we can use the trace
form to identify so(3)∗ with so(3) via the prescription

(8.8) so(3)∗
η −→ Aη such that η(B) =< Aη , B > for all B ∈ so(3) :

We choose a basis (A∗1 , A∗2 , A∗3 ) with < A∗i , Aj >= δij and for an element η ∈ so(3)∗ we
write η ≡ Aη = a∗1 A∗1 + a∗2 A∗2 + a∗3 A∗3 .

As is well known (see the proof of the fact that SU(2) is a twofold covering of SO(3)
in section 4.3), using these identifications the coadjoint operations of SU(2) on su(2)∗
resp. so(3)∗ are simply the rotations of R3 . The orbits are the spheres of radius s,
Ms := G · ηs for ηs := sA∗3 with s > 0, centered at the origin and the origin itself. One
has a symplectic structure on the sphere Ms of radius s given by the volume form, which
is given by restriction to Ms of the 2–form

ω0 := (1/s) a∗i da∗j ∧ da∗k .

{i, j, k}
cyclic
194 8. Geometric Quantization and the Orbit Method

As local coordinates we use (a∗1 , a∗2 ) such that Ms is given by the parametrization

ψ : J := {(a∗1 , a∗2 ); (a∗1 )2 + (a∗2 )2 ≤ s2 } −→ Ms ,

(a∗1 , a∗2 ) −→ (a∗1 , a∗2 , a∗3 ), (a∗3 )2 = s2 − (a∗1 )2 − (a∗2 )2 .

Exercise 8.4: Verify that one has

ω0 = (s/a∗3 )da∗1 ∧ da∗2

and 
vol2 (Ms ) = 2 ψ ∗ ω0 = 4πs2 .
J

We want to compare this with the Kirillov-Kostant recipe

ωη (A, B) = η([A, B]),

resp. (8.5)
ω(ξA , ξB ) =< η  , [A, B] > .
We have
A = a1 A1 + a2 A2 + a3 A3 , B = b1 A1 + b2 A2 + b3 A3 ,
i.e.
[A, B] = (a1 b2 − a2 b1 )A3 + (a2 b3 − a3 b2 )A1 + (a3 b1 − a1 b3 )A2 ,
and with η = a∗1 A∗1 + a∗2 A∗2 + a∗3 A∗3

(8.9) ωη (A, B) = (a1 b2 − a2 b1 )a∗3 + (a2 b3 − a3 b2 )a∗1 + (a3 b1 − a1 b3 )a∗2 .

As in the preceding example we determine the associated vector fields ξAj . Namely using
here gj (t) = exp tAj , j = 1, 2, 3,

d
ξAj f (η) = f (gj (t) · η)|t=0
dt
leads to

ξA1 = −a∗3 ∂a∗2 + a∗2 ∂a∗3 ,

ξA2 = a∗3 ∂a∗1 − a∗1 ∂a∗3 ,
ξA3 = −a∗2 ∂a∗1 + a∗1 ∂a∗2 .

and hence in the local coordinates fixed above

ξA = a1 ξA1 + a2 ξA2 + a3 ξA3

= α∂a∗1 + β∂a∗2 ,

with
α = (a2 a∗3 − a3 a∗2 ), β = (a3 a∗1 − a1 a∗3 )
and the corresponding expressions α , β  for ξB replacing the aj by bj . We get

αβ  − α β = (a2 b3 − a3 b2 )a∗1 a∗3 + (a1 b2 − a2 b1 )a∗3 a∗3 + (a3 b1 − a1 b3 )a∗2 a∗3 .

8.2 Coadjoint Orbits and Representations 195

By evaluation at η = sA∗3 one obtains

ω0 (ξA , ξB ) = (s/a∗3 )(αβ  − α β)
= s2 (a1 b2 − a2 b1 ).
We compare this with (8.9) and see that the Kirillov-Kostant form in this case is
ω = sω0 = (s2 /a∗3 )da∗1 ∧ da∗2 .
The isotropy group of η is
   
eit 0
Gη = h= , t ∈ R  U (1).
0 e−it
G/Gη is identified with the sphere Ms  S 2 via the important Hopf map
ψ : G = SU(2) −→ Ms
given by
(a, b) −→ s(bā + ab̄, ibā − iab̄, aā − bb̄)
By some more calculation using the relation
adā + āda + bdb̄ + b̄db = 0,
which is a consequence of aā + bb̄ = 1, we get for the drawback of the Kirillov-Kostant
form by the Hopf map
ψ ∗ ((s/a∗3 )da∗1 ∧ da∗2 ) = 2is(da ∧ dā + db ∧ db̄).
As we already constructed bundles on P1 (C) (in 7.6 Example 7.19 and in 8.2.2) it is
quite fruitful to compare this description of the coadjoint SU(2)-orbits with the following
alternative: We discussed (in 7.1 Example 7.3) the identification
SU(2)/U(1)  P1 (C) = {(u : v); u, v ∈ C2 \ {(0, 0)}}
essentially given by
g −→ z = u/v = b/ā,
which is a consequence of g acting by multiplication from the left on the column t (u, v).
The same way, one has an identification ψ̃ via
(8.10) g −→ z = v/u = b/a
coming from multiplication from the right to the row (u, v). If we take the symplectic
form ω studied in 8.2.2
dz ∧ dz̄
ω = 2ni , n ∈ Z,
(1 + |z|2 )2
and via ψ̃ draw it back to SU(2), we get again the form obtained above
2ni(da ∧ dā + db ∧ db̄)
and see that we have quantizable orbits for half integral s resp. n.

Now that we have examples for coadjoint orbits and bundles living on them, we continue
in the program to use the sections of these bundles as representation spaces and we have
to discuss how the general polarization business works in practical cases.
196 8. Geometric Quantization and the Orbit Method

8.2.5 Construction of an Irreducible Unitary Representation by

an Orbit
If we want to construct representations from a coadjoint orbit of a group, the orbit has to
be admissible, i.e. “integral” and been provided with a “nice” polarization. Then one can
construct a complex line bundle, whose polarized sections lead to the space for a unitary
(and perhaps even irreducible) representation of the group. Or, in another formulation,
this representation can be got by an induction procedure from a subgroup arising from
the polarized orbit. Up to now, all this is not in a final form if one strives for maximal
generalilty (which we do not do in this text). Though there are now several more refined
procedures using and discussing Duflo’s version of admissibility as for instance in Vogan
[Vo] p.193 ff or in Torasso [To], we will follow the procedure designed by Kirillov [Ki]
p.235 ff and [Ki1] (but in some places adapting our notation to the one used by Vogan).
Similar presentations can be found in [GS1] p.235 and [Wo] p.103.

Step 1. For each orbit Oη ⊂ g∗ we look for real subalgebras n ⊂ g which are subordinate
to η. This means that the condition

< η, [X̌, Y̌ ] > = 0 for all X̌, Y̌ ∈ n

holds or, equivalently (this looks quite different but is not too hard to prove), the map

Y̌ −→ 2πi < η, Y̌ >

defines a one-dimensional representation  = η of n .

Step 2. We say that n is a real algebraic polarization of η if in addition the condition

2 dim(g/n) = dim g + dim gη

is satisfied. The notion of a complex algebraic polarization is defined in the same way:
we extend η to gc = g ⊗ C by complex linearity and consider complex subalgebras n ⊂ gc
that satisfy the corresponding conditions as in the real case.
An algebraic polarization is called admissible if it is invariant under Ad Gη . In the sequel,
we only look for admissible algebraic polarizations.

The relation of these “algebraic” polarizations to the “geometric” ones defined earlier is
explained in [Ki1] p.28/9 and will be made explicit in a simple example later.

Step 3. Moreover, we suppose (as in [Ki] p.238) that in the complex case mc = n + n is
a subalgebra of gc such that we have closed subgroups Q and M of G with Lie algebras
q and m fulfilling the relations

L := Gη ⊂ Q = Gη Q◦ ⊂ M, n ∩ n = qc , n + n = mc .

In the real case, we have q = n.

8.3 The Examples SU(2) and SL(2, R) 197

Step 4. Finally, and this is our practical criterium to guarantee the integrality or
quantizability of the orbit, we suppose that η can be integrated to a unitary character
χQ of Q with dχQ = η . Then we get a unitary representation π of our group G on (the
completion of) the space F+ := C ∞ (G, n, Q, , χQ ) of smooth C–valued functions φ on
G with
φ(gl) = Q (l)−1/2 χQ (l−1 )φ(g) for all l ∈ L, g ∈ G
and
LY̌ φ = 0 for all Y̌ ∈ u, n =: gη + u,
where Q is the usual modular function and LY̌ φ(g) = dt
d
φ(g exptY̌ )|t=0 .
(Here we follow [Vo] p.194 and p.199. For an essentially equivalent approach see [Ki]
p.199ff).

Though this looks (and is) rather intricate, we shall see that the scheme can easily be
realized in our simple examples, in particular for the Heisenberg group in the next but
one section. As already said, there are a lot of subtle questions hidden here. For instance
(see [Ki] p.236), as another condition Pukanszky’s condition, i.e. the additional property
η + n⊥ ⊂ Oη comes in where n⊥ denotes the annihilator of n in g∗ . And, as discussed in
[Ki] p.239, the fundamental group π1 (O) of the orbit plays an important role in the inte-
gration of the representation η of the Lie algebra q to a representation of the group Q.
This leads to the notion of rigged orbits, i.e. orbits in g∗rigg , the set of pairs (η, χ) where
η ∈ g∗ and χ is here a one-dimensional representation of Gη such that dχ = 2πiη|gη (see
[Ki1] p.123 (and Vogan’s criticism of Kirillov’s book on his homepage)). Here we only
want to give an impression, and (from [Ki] p.241) we cite a theorem by Kostant and
Auslander:

Theorem 8.4: Assume G to be a connected and simply connected solvable Lie group.
Then representations that correspond to different orbits or different characters of the
fundamental group of the orbit are necessarily inequivalent.

And (a special version of) the famous Borel-Weil-Bott theorem says:

Theorem 8.5: All irreducible representations of a compact connected, and simply con-
nected Lie group G correspond to integral G-orbits of maximal dimension in g∗ .

In more detail, in our special examples the result of this procedure is as follows.

8.3 The Examples SU(2) and SL(2, R)

1. G = SU(2)

We begin by taking up again the here already often treated example G = SU(2). Using
(6.5) in Section 6.5, we take

g = {X = Σaj Hj ; a1 , a2 , a3 ∈ R}

and, similar to (8.8) in 8.2.4, we choose (Hj∗ ) such that < Hj∗ , Hk > = δjk and

g∗ = {η = Σa∗j Hj∗ ; a∗1 , a∗2 , a∗3 ∈ R}.

198 8. Geometric Quantization and the Orbit Method

From (8.9) for the elliptic orbit Oη  SU(2)/U(1) passing through to η = sH1∗ we have
the Kirillov-Kostant prescription

ωη (X, Y ) = s(a2 b3 − a3 b2 ).

One has gη = Lie U(1)  H1 . As g has no two-dimensional subalgebras n, there is

no real polarization. But one has two complex polarizations given by the subalgebras
n+ = H0 , H+  and n− = H0 , H−  of the complexification gc = H0 , H±  with

H0 = iH1 , H± = H2 ± iH3 .

As one easily verifies, both algebras are subordinate to η in the sense defined above in
Subsection 8.2.5 and, for s = k ∈ N0 , the character of gη

gη
Y −→ 2πi < η, Y >

integrates to a unitary character χk of H = U(1) identified as a subgroup of G = SU(2).

In this case the space F+ is the space belonging to the induced representation indG
H χk
restricted by the polarization condition

LH± φ = 0

since one has n+ = gη + < H+ > or n− = gη + < H− >. And this space can be under-
stood exactly as the subspace of holomorphic resp. antiholomorphic smooth sections of
the real bundle B over P1 (C) from 8.2 and the end of 7.6: To show this, we determine
the appropriate differential operators acting on the smooth sections f of B resp. the func-
tions φ on G with φ(gh) = χ2s (h−1 )φ(g) where we denote φ(g) =: φ(a, b) =: φ(α, α , β, β  )
decomposing our standard coordinates for g into real coordinates

a =: α + iα , b =: β + iβ  .

We put gj (t) := exp tHj and get

 −it     
e cos t −i sin t cos t − sin t
g1 (t) = , g (t) = , g3 (t) = .
eit 2
−i sin t cos t sin t cos t

Application of the prescription LHj φ(g) := dtd

φ(ggj (t))|t=0 to φ as a function of the four
 
real variables α, α , β, β leads to the operators

LH1 = α ∂α − α∂α
− β  ∂β + β∂β
,

LH2 = β  ∂α − β∂α
+ α ∂β − α∂β
,
LH3 = β∂α + β  ∂α
− α∂β − α ∂β
,

and hence, using the Wirtinger relations ∂a = (1/2)(∂α − i∂α
), ∂ā = (1/2)(∂α + i∂α
),
we get the operators

LH2 +iH3 = 2i(b̄∂ā − a∂b ), LH2 −iH3 = −2i(b∂a − ā∂b̄ )

acting on the functions φ here written in the coordinates a, ā, b, b̄. If φ comes from a
section f of the bundle B over P1 (C), for b = 0 it is of the form (see (8.10) in 8.2)

φ(a, ā, b, b̄) = f (z, z̄), z = a/b̄.

8.3 The Examples SU(2) and SL(2, R) 199

Hence we get
LH2 +iH3 φ = (2i/b2 )fz̄ , LH2 −iH3 φ = (2i/b̄2 )fz ,

and we see that the polarization by n± provides that the section f has to be holomorphic
resp. antiholomorphic.

Summary: For G = SU(2) (and SO(3)) all irreducible unitary representations can be
constructed by the orbit method.
Each integral orbit Oη carries a line bundle and has two complex polarizations.
The correponding polarized sections provide equivalent representations.

2. G = SL(2, R)

We continue by completing the results obtained for the orbits from the SL(2)-theory in
the last section. The orbits are all realized as subsets of

Mβ := {(x∗ , y∗ , z∗ ) ∈ R3 ; x2∗ + y∗2 − z∗2 = β}

with convenient β ∈ R.

2.1. Hyperbolic Orbits

For the hyperbolic orbit Oη = G · αX∗ = Mα2  G/L , L = Gη = M A , α = 0 we have

Ad L-invariant real polarizations given by

q1 := X, Y + Z and q2 := X, Y − Z .

The associated characters ρ1 and ρ2 of q1 resp. q2 given by

ρ1 (Y̌ ) = 2πi αX∗ , xX + y̌(Y + Z) = 4πiαx for Y̌ = xX + y̌(Y + Z)

and

ρ2 (Y̌ ) = 2πi αX∗ , xX + ž(Y − Z) = 4πiαx for Y̌ = xX + ž(Y − Z)

integrate uniquely to unitary characters χj,s (j = 1, 2) of the respective groups

Q1 = M N A and Q2 = M N̄ A,

which are trivial on Qj ∩ K. For

b = ε n(x) t(y) resp. = ε n̄ (x) t(y)

these characters are given by

χj,s (b) := (y 1/2 )is with s = 4πiα.

200 8. Geometric Quantization and the Orbit Method

Then by inducing these characters from Qj to G one gets irreducible unitary repre-
sentations. This coincides quantitatively with the fact that one has the principal series
representations P ±,is := πis,± , which we constructed in 7.2 and which have models
consisting of (smooth) functions φ = φ (g) on G with
 
1/2 (is+1) ε ε for −
φ (mn(x)t(y)g) = γ(m) (y ) φ(g) , γ = ,
ε 1 for +

i.e. in a space spanned by functions

φs,2k , k ∈ Z for + .

and
φs,2k+1 , k ∈ Z for −

with

(8.11) φs,j (g) = y (is+1)/2 eijθ , j ∈ Z.

The unitary equivalence P ±,is  P ±,−is can be translated into the fact that both polar-
izations produce equivalent representations (see the discussion in [GS1] p. 299 f). Thus,
by the orbit method, the construction above yields only one half of the representations.
This arises from the fact that the orbit Mα2 is not simply connected but has here a
fundamental group π1 (Mα2 )  Z , so that we have one representation (the even one)
belonging to id ∈ π1 (Mα2 ) and the other one to a generator of π1 (Mα2 ) (see the dis-
cussion in [Ki3] p.463). This may serve as motivation to introduce the notion of rigged
orbits already mentioned at the end of our Subsection 8.2.5.

2.2. Elliptic Orbits

±
For the elliptic orbits M−α 2 := Oη = G · αZ∗  G/SO(2), α = 0 , consisting of the upper
+
half M−α2 (i.e. with z∗ > 0) of the two-sheeted hyperboloid Mα2 for α > 0 and the lower
−
half M−α 2 (with z∗ < 0) for α < 0, we have two purely complex polarizations : We take

   
−i 1 ±i
Z1 = −iZ = , X± = (1/2) (X ± iY ) = (1/2)
i ±i 1

with
[Z1 , X± ] = ±2X± , [ X+ , X− ] = Z1 .

and have
gc = Z1 , X+ , X− .

This establishes the complex bilinear form

Bη ([ X̌, Y̌ ]) = iα(a+ b− − a− b+ )

for
X̌ := a0 Z1 + a+ X+ + a− X− , Y̌ := b0 Z1 + b+ X+ + b− X− .
8.3 The Examples SU(2) and SL(2, R) 201

We have Bη ≡ 0 on the Ad Gη -invariant subalgebras

n := n± = Z1 , X± C

of gc and n + n̄ = gc , n ∩ n̄ = kc = Z1 

n± ∩ g1 = Y − ZR = k .

At first we see that a one-dimensional representation of k is defined by

 
α
ρ (a0 Z1 ) := i  , a0 Z1  = i 2α a0
−α

and this integrates to a representation χk of SO(2) (with χk (r(θ)) = eikθ ) exactly for
2α =: k ∈ Z \ 0 . Thus the elliptic orbits are quantizable for 2α = k ∈ Z \ {0} . Then we
use n+ and n− to define a complex structure on M = Mα±2  G1 /SO(2)  H± . And
by the usual procedure we get the discrete series representations πk± , k ∈ N , realized
±
on the L2 -spaces of holomorphic sections of the holomorphic line bundles L on M−α 2 .

With similar calculations as we did for SU(2), these can also be interpretated as spaces
±
of polarized sections φ of the bundle belonging to χk on M−α 2 , i.e. with

LX+ φ = 0 resp. LX− φ = 0 .

These representations are also representations of K = SO(2) and we have

mult(χk̃ , πk+ ) = 1 for k̃ = k + 2l, l ∈ N0 and = 0 else.

2.3. Nilpotent Orbits

In the nilpotent case Oη = G · (Y∗ + Z∗ ) = M0+ = G/N̄ we have one real polarization.
The form ωY∗ +Z∗ is zero on the Ad N̄ - invariant subalgebra of g given by

n = < X, Y − Z > .

The standard representation ρ of n is trivial

ρ(xX + y(Y − Z)) = < Z∗ , xX + y(Y − Z) > = 0.

a b
The group B = M N A = {±( ); a > 0, b ∈ R} has Lie B = n. We can (at
0 a−1
least formally) proceed as follows. If χ = id denotes the trivial representation of B, one
has here the representation indGB id given by right translation on the space of functions
φ : G −→ C with
φ(bg) = y 1/2 φ(g) for b ∈ B, g ∈ G,
(spanned by φ(g) = y 1/2 eiθ ,  ∈ 2Z), i.e. the representation P +,0 . As in the previous
case, here we have a fundamental group π1  Z, so there is a second representation
P −,0 consisting of odd functions and spanned by φ(g) = y 1/2 eiθ ,  ∈ 1 + 2Z), which
decomposes into irreducible halves

P −,0 = π1+ ⊕ π1−

(see for instance [Kn] p.36). The lower half of the cone M0− = G1 · Y∗ generates an
equivalent situation.
202 8. Geometric Quantization and the Orbit Method

All this is a small item in the big and important theme of how to attach representations
to nilpotent orbits (see in particular the discussion in Example 12.3 of Vogan [Vo1] p.386).

Summary

For the representations of G = SL(2, R), we get a correspondence between

– elliptic orbits and discrete series representations.
– (rigged) hyperbolic orbits and principal series representations,
We do not get the complementary series.
For the nilpotent orbits in the SL(2, R)-theory, it seems still not clear how the correspon-
dence is to be fixed (see the discussion in [Vo1] p.386). To me it seems most probable
(see in particular [Re] p.I.110f) to associate to the nilpotent cone the principal series
P +,0 and the (decomposing) P −,0 = D1 ⊕ D−1 .

Let us finally point to a remark by Kirillov in [Ki3] 8.4 and [Ki1] 6.4 that he sees the
possibility to extend the correspondence framework such that even the complementary
series fits in. This should be further elucidated. Up to now one has here an example of an
imperfect matching between the unitary dual and the set of coadjoint orbits. But there
is a conjecture that by the orbit method one gets all representations appearing in the
decomposition of the regular representation of a group ([Ki1] p 204): “Indeed, according
to the ideology of the orbit method, the partition of g∗ into coadjoint orbits corresponds
to the decomposition of the regular decomposition into irreducible components”.

8.4 The Example Heis(R)

The treatment of the Heisenberg group is even easier than that of the two previous
examples. But one has one additional complication as one can not any longer Ad G-
equivariantly identify the Lie algebra with its dual via the trace or Killing form. So
we use the original definition of the coadjoint representation to determine the set O(G)
of coadjoint orbits and compare it with the unitary dual Ĝ of the Heisenberg group
G = Heis(R), which we determined in Example 7.11 in 7.4 as the disjoint union of the
plane R2 with a real line without its origin R \ {0}. We shall get a perfect matching
between Ĝ and O(G).

As the formulae are a bit more smooth, we work with the realization of the Heisenberg
group by three-by-three matrices, i.e. we take
⎛ ⎞
1 a c
G := Heis (R) = {g = ⎝ 1 b ⎠ ; a, b, c ∈ R}
1

and
⎛ ⎞
0 x z
g := {X̌ = ⎝ 0 y ⎠ = xX + yY + zZ; x, y, z ∈ R}.
0
8.4 The Example Heis(R) 203

The Coadjoint Action

As in 8.2, we have the adjoint representation Ad of G given on g by

⎛ ⎞⎛ ⎞⎛ ⎞
1 a c 0 x z 1 −a −c + ab
(Ad g)X̌ = g X̌g −1 =⎝ 1 b⎠ ⎝ 0 y⎠ ⎝ 1 −b ⎠
1 0 1
⎛ ⎞
0 x z + ay − bx
=⎝ 0 y ⎠.
0

The coadjoint representation Ad∗ is the contragredient of Ad (see 1.4.3) and given on g∗
by Ad∗ g = (Ad g −1 )∗ , i.e. if we write

η, X̌ := η(X̌)

for all η ∈ g∗ and X̌ ∈ g, then Ad∗ g is fixed by the relation

Ad∗ g η, X̌ = η, (Ad g −1 )X̌ for all X̌ ∈ g.

We have ⎛ ⎞
0 x z − ay + bx
(Ad g −1 )X̌ = ⎝ 0 y ⎠.
0

And if we take X∗ , Y∗ , Z∗ as a basis for g∗ , dual to X, Y, Z with respect to <, >, and
write
g∗
η = x∗ X∗ + y∗ Y∗ + z∗ Z∗ ,

we have
Ad∗ g η = x̃∗ X∗ + ỹ∗ Y∗ + z̃∗ Z∗

with

(8.12) x̃∗ = x∗ + bz∗ , ỹ∗ = y∗ − az∗ , z̃∗ = z∗ .

There is a more elegant approach to this result: In [Ki1] p.61 Kirillov uses the trace form
on M3 (R) to identify g∗ with the space of lower-triangular matrices of the form
⎛ ⎞
∗ ∗ ∗
η = ⎝x ∗ ∗⎠ .
z y ∗

Here the stars remind us that one actually considers the quotient space of M3 (R) by the
subspace g⊥ of upper-triangular matrices (including the diagonal). This way, one gets
⎛ ⎞ ⎛ ⎞
∗ ∗ ∗ ∗ ∗ ∗
Ad∗ g η = g ⎝x ∗ ∗⎠ g −1 = ⎝x + bz ∗ ∗⎠ .
z y ∗ z y − az ∗
204 8. Geometric Quantization and the Orbit Method

Exercise 8.5: Do the same for the realization of the Heisenberg group and its Lie algebra
by four-by-four matrices as in 0.1 and 6.3. Verify that Ad∗ g is given by
(p∗ , q∗ , r∗ ) −→ (p∗ + 2µr∗ , q∗ − 2λr∗ , r∗ )
for g = (λ, µ, κ) and η = p∗ P∗ + q∗ Q∗ + r∗ R∗ , where P∗ , Q∗ , R∗ is a basis of g∗ dual to
P, Q, R.

The Coadjoint Orbits

From (8.12) it is easy to see that one has just two types of coadjoint orbits Oη = G · η 
G/Gη .
i) For η = mZ∗ , m = 0 we have a two-dimensional orbit
Om := OmZ∗ = {(x∗ , y∗ , m); x∗ , y∗ ∈ R}  R2 .
ii) For η = rX∗ + sY∗ , r, s ∈ R, as orbit we have the point
Ors := OrX∗ +sY∗ = {(r, s, 0)}.
∗
Obviously g  R is the disjoint union of all these orbits and the set of orbits is in
3

one-by-one correspondence with the unitary dual Ĝ described above. Now, we construct
representations following the general procedure outlined in 8.2:

Orbits as Symplectic Manifolds

For M := Om  R2 , m = 0, we have as symplectic forms all multiples of ω0 = dx∗ ∧ dy∗ .

As it is a nice exercise in the handling of our notions, we determine the factor for the
Kirillov-Kostant form ω: Again we use (8.5)
ωη (η  )(ξX̂ (η  ), ξŶ (η  )) = < η  , [X̂, Ŷ ] > for all X̂, Ŷ ∈ g, η  ∈ gη ∗
We put
X̂ := xX + yY + zZ, Ŷ := x X + y  Y + z  Z,
and have
[X̂, Ŷ ] = (xy  − x y) Z
and
< η  , [X̂, Ŷ ] > = (xy  − x y)z∗ for η  = x∗ X∗ + y∗ Y∗ + z∗ Z∗ .
We determine the vector fields ξ using again (8.6)
d
(ξX̂ f )(η) = (f (exp tX̂ · η))|t=0 .
dt
One has
⎛ ⎞
1 t
gX (t) := exp tX = ⎝ 1 ⎠,
1
⎛ ⎞
1
(8.13) gY (t) := exp tY = ⎝ 1 t⎠ ,
1
⎛ ⎞
1 t
gZ (t) := exp tZ = ⎝ 1 ⎠,
1
8.4 The Example Heis(R) 205

and, by (8.12),

gX (t) · η = x∗ X∗ + (y∗ − tz∗ )Y∗ + z∗ Z∗ ,

gY (t) · η = (x∗ + tz∗ )X∗ + y∗ Y∗ + z∗ Z∗ ,
gZ (t) · η = x∗ X∗ + y∗ Y∗ + z∗ Z∗ .

and, hence,
ξX = −z∗ ∂y∗ , ξY = z∗ ∂x∗ , ξZ = 0,

that is

ξX̂ = xξX + yξY + zξZ ,

= α∂x∗ + β∂y∗

with
α = yz∗ , β = −xz∗ .

Thus we have
ω0 (ξX̂ , ξŶ ) = (αβ  − α β) = z∗2 (xy  − x y).

If we compare this with the Kirillov-Kostant prescription above, we see that one has the
symplectic form ω on M = Om given by

ω = (1/m)dx∗ ∧ dy∗ .

Polarizations of the Orbits Om

The algebraic approach to real polarizations asks for subalgebras n of g = X, Y, Z sub-
ordinate to η = mZ∗ . Obviously one has the (abelian) subalgebras n1 = X, Z and
n2 = Y, Z and more generally, for α, β ∈ R with αβ = 0, n = αX + βY, Z. In the
geometric picture these correspond to a polarization given by a splitting of Om  R2
into the union of parallel lines αx∗ + βy∗ = const., which are the orbits of Ad∗ Q where
Q is the subgroup of G with Lie Q = n, i.e. Q = {(αt, βt, c); t, c ∈ R}.

A complex polarization of Om , which is translation invariant, is generated by a constant

vector field ξ = ∂x∗ + τ ∂y∗ , τ ∈ C \ R, in particular τ = i. The functions F satisfying
ξF = 0 are simply holomorphic functions in the variable w := x∗ + τ y∗ . In the algebraic
version the complex polarizations are given by the subalgebra n = (X + τ Y ), Z.

Construction of Representations

We want to follow the scheme outlined at the end of 8.2. As up to now our representations
of the Heisenberg group were realized by right translations and the recipe in 8.2.5 is given
for left translations, we have to make the appropriate changes. For the zero-dimensional
orbits Or,s , r, s ∈ R there is not much to be done: One has η = rX∗ + sY∗ and Gη = G
and hence the one-dimensional representation π given by π(a, b, c) = exp(2πi(ra + sb)).
For the two-dimensional orbits Om , m = 0 we have η = mZ∗ , Gη = {(0, 0, c), c ∈ R} = L
and for each polarization given by n the representation Y̌ −→ 2πi < η, Y̌ >= 2πimz
integrates to a character χm of the group Q belonging to n.
206 8. Geometric Quantization and the Orbit Method

Hence each orbit Om is integral and we get a representation space consisting of sections
of the real line bundle over M = Om (constructed using the character χm of Gη with
χm ((0, 0, c)) = exp (2πimc) = e(mc)) which are constant along the directions fixed by the
polarization n. This leads to consider the following functions: We look at φ : G −→ C
with
φ(lg) = χm (l)φ(g),
i.e. φ is of the form φ(a, b, c) = e(mc)F (a, b) with a smooth function F : R2 −→ C
restricted by the polarization condition, which in our case for n = < U, Z > is given by

RU φ = 0

where RU is the right-invariant differential operator

d
RU φ(g) = φ(gU (t)−1 g)|t=0 , gU (t) := exp tU.
dt
We have
gX (t) = (t, 0, 0), and gX (t)−1 g = (a − t, b, c − tb),
gY (t) = (0, t, 0), gY (t)−1 g = (a, b − t, c),
gZ (t) = (0, 0, t), gZ (t)−1 g = (a, b, c − t).
and hence

(8.14) RX = −∂a − b∂c , RY = −∂b , RZ = −∂c .

– i) The polarization given by n =< Y, Z > leads to

0 = RY φ = −∂b φ = −Fb e(mc),

i.e. φ is of the form φ(a, b, c) = f (a)e(mc). As to be expected, via

π(g0 )φ(g) = φ(gg0 ) = f (a + a0 )e(m(c + c0 + ab0 ))

= e(mc)e(m(c0 + ab0 ))f (a + a0 ),

we recover the Schrödinger representation πm =: πS

f (t) −→ πm (g0 )f (t) = e(m(c0 + b0 t))f (t + a0 ).

– ii) n =< X, Z > leads to

0 = RX φ = −(∂a + b∂c )φ = −(Fa + 2πimbF )e(mc),

i.e. φ is of the form φ(a, b, c) = e(−mab)f (b)e(mc). We get the representation


f (t) −→ e(mc0 − a0 (b0 + t))f (t + b0 ) =: πm (g0 )f (t).

From the Stone-von Neumann Theorem one knows that this representation (and any
other constructed this way via another polarization) is equivalent to the Schrödinger
representation. In our context we can see this directly using the Fourier transform as an
intertwining operator:
8.4 The Example Heis(R) 207

We abbreviate again e(u) := exp(2πiu) and write as Fourier transform


fˆ(b) := f (t)e(mbt)dt.
R

We get 

πm (g0 )fˆ(b) = e(m(c0 − a0 (b + b0 ))) f (t)e(mt(b + b0 ))dt
R

and see that this coincides with the Fourier transform of the application of the Schrödinger
representation πm (g0 )f , namely with a + a0 =: t we get


(πm (g0 )f )(b) = e(m(c0 + ab0 )f (a + a0 )e(mab)da

= e(mc0 )e(−m(a0 (b + b0 ))) e(mtb0 )f (t)e(mbt)dt.

– iii) As another example of this procedure we use a complex polarization to construct

the Fock representation, which (among other things) is fundamental for the construction
of the theta functions. The Fock representation has as its representation space the space
Hm of holomorphic functions f on C subjected to the condition

 f =
2
| f (z)| 2 dµ(z) < ∞, dµ(z) := (1/(2i))dz ∧ dz̄.
C

To adopt our presentation to the more general one given in Igusa [Ig] p.31ff, we introduce
complex coordinates for our Heisenberg group

z := −ia + b, z̄ = ia + b,

i.e.
a = (z̄ − z)/(2i), b = (z + z̄)/2
and
∂z = (1/2)(i∂a + ∂b ), ∂z̄ = (1/2)(−i∂a + ∂b ).
We use the complexification gc =< Y± , Z0 > with Y± := ±iX + Y, Z0 = 2iZ and the
complex polarization given by n =< Y− , Z0 >. Then, in order to construct a representa-
tion space following our general procedure, we come to look at smooth complex functions
φ on G, which are of the form

φ(a, b, c) = F (z, z̄)e(mc)

and subjected to the polarization condition

RY− φ = 0.

From (8.14) we have RX = −(∂a + b∂c ), RY = −∂b . We deduce

RY− = −2∂z̄ + ib∂c

and, hence, that one has 0 = RY− φ = (−2Fz̄ − 2πmbF )e(mc), i.e.

∂z̄ log F = −πmb = −πm(z + z̄)/2.

208 8. Geometric Quantization and the Orbit Method

This shows that F is of the form

2
F (z, z̄) = e−(πm/2)(zz̄+z̄ /2)
f˜(z)

with a holomorphic function f˜. As we soon will see, to get a unitary representation on
the space Hm introduced above, we have to choose here f˜(z) = exp(πmz 2 /4)f (z) with
f ∈ Hm , i.e. our φ is of the form
2
/2−z 2 /2)
φ(a, b, c) = e(mc)e−(πm/2)(zz̄+z̄ f (z).

By a small computation we see that the representation by right translation

π(g0 )φ(g) = φ(gg0 )

leads to the formula of the Fock representation πF

2 2
(8.15) f (z) −→ (πF (g0 )f )(z) = e(mc0 )e−πm((z+z0 /2)z̄0 +(z̄0 −z0 )/4) f (z + z0 ).

with g0 = (a0 , b0 , c0 ), z0 = b0 − ia0 . Now it is easy to verify that πF acts unitarily in

Hm (Exercise 8.6).

Perhaps it is not a complete waste of time and energy to show here how this representation
comes up also (as it must) by the induction procedure we outlined in 7.1: We have the
standard projection

G = Heis (R) −→ H \ G  X , g = (a, b, c) −→ Hg = x = (a, b) ↔ (z, z̄)

with the Mackey section s given by s(x) = (a, b, 0), i.e. g = (0, 0, c)(a, b, 0), and the
master equation (7.7)

s(x)g0 = (a, b, 0)(a0 , b0 , c0 ) = h∗ s(xg0 )

with

s(xg0 ) = (a + a0 , b + b0 , 0), h∗ = (0, 0, c0 + ab0 ) = (0, 0, c0 + (zz0 − z z̄0 + z̄0 z0 − z̄ z̄0 )/(4i)).

dµ(x) = dµ(z) is a quasi-invariant measure on X = C with Radon-Nikodym derivative

dµ(xg0 ) 2
= e−mπ(zz̄0 +z̄z0 +|z0 | ) = δ(h∗ ).
dµ(x)

Hence, injecting these data into the formula from Theorem 7.4 in 7.1.2 for the represen-
tation in the second realization

π̃(g0 )f (x) = δ(h∗ )1/2 π0 (h∗ )f (xg0 )

provides the formula (8.15) for the Fock representation obtained above.

We know that the Fock representation πF is equivalent to the Schrödinger representation

πS but it is perhaps not so evident how the intertwining operator looks like.
8.5 Some Hints Concerning the Jacobi Group 209

Proposition 8.1: The prescription


f (t) −→ F (z) = If (z) := k(t, z)f (t)dt
R

with the kernel

2 2
k(t, z) = 21/4 e−πt e(tz)e(π/2)z

provides an isometry I between the spaces L2 (R) and Hm for m = 1 and intertwines the
corresponding Schrödinger and Fock representations.

The fact that I is an intertwining operator is verified by a straightforward computation

(Exercise 8.7). For the verification of the isometry we refer to [Ig] p.31-35 where one
goes back to explicit Hilbert bases for the spaces.

Summary

For the Heisenberg group (as for each solvable group), one has a perfect matching be-
tween coadjoint orbits and irreducible unitary representations.
Everything goes through as well for the higher-dimensional Heisenberg groups if one
changes the notation appropriately.

8.5 Some Hints Concerning the Jacobi Group

In Kirillov’s book [Ki1] one can find a lot of examples and a thorough discussion of the
merits and demerits of the orbit method. As orbits already turned up naturally in our
presentation of Mackey’s method in 7.1 where we applied it to Euclidean groups and
the Poincaré group, it is a natural topic to inspect the coadjoint orbits for semidirect
products, in particular for the Euclidean groups.

Exercise 8.8: Determine the coadjoint orbits of the Euclidean group

GE (3) = SO(3)
R3 .

The reader will find material for this in [GS] p.124ff. As the author of this text is
particularly fond of the Jacobi group GJ , i.e. – in its simplest version – a semidirect
product of SL(2, R) with Heis(R), as another exercise we propose here to treat some
questions in this direction concerning the Jacobi group. Most anwers to these can be
found in [BeS] and [Be1] where we collected some elements of the representation theory
of GJ and the application of the orbit method to GJ . But at a closer look there are
enough rather easily accessible open problems to do some original new work:

The Jacobi group is in general the semidirect product of the symplectic group with an
appropriate Heisenberg group. Here we look at

GJ (R) := SL(2, R)
Heis(R).

210 8. Geometric Quantization and the Orbit Method

In this case we fix the multiplication law by the embedding into the symplectic group
Sp(2, R) given by
⎛ ⎞
1 µ
⎜ λ 1 µ κ ⎟
Heis(R)
(λ, µ, κ) −→ ⎜
⎝
⎟ ,
1 −λ ⎠
1
⎛ ⎞
  a b
a b ⎜ 1 ⎟
SL(2, R)
M = −→ ⎜
⎝
⎟ .
⎠
c d c d
1

We write
g = (p, q, κ)M or g = M (λ, µ, κ) ∈ GJ (R).
As in [BeS] or [Ya], we can describe the Lie algebra gJ as a subalgebra of g2 = sp(2, R)
by ⎛ ⎞
x 0 y q
⎜p 0 q r ⎟
G(x, y, z, p, q, r) := ⎜
⎝ z 0 −x
⎟
−p⎠
0 0 0 0
and denote
X := G(1, 0, . . . , 0), . . . , R := G(0, . . . , 0, 1).
We get the commutators

[X, Y ] = 2Y, [X, Z] = −2Z, [Y, Z] = X,

[X, P ] = −P, [X, Q] = Q, [P, Q] = 2R,
[ Y, P ] = −Q, [ Z, Q] = −P,

all others are zero. Hence, we have the complexified Lie algebra given by

gJc =< Z1 , X± , Y± , Z0 >

where as in [BeS] p.12

Z1 := −i(Y − Z), Z0 := −iR,

X± := (1/2)(X ± i(Y + Z)), Y± := (1/2)(P ± iQ)

with the commutation relations

[Z1 , X± ] = ±2X± , [Z0 , Y± ] = ±Y± , etc.

Exercise 8.9 : Verify all this and compute left invariant differential operators

LZ0 = i∂κ
LY± = (1/2)y −1/2 e±iθ (∂p − (x ± iy)∂q − (p(x + iy) + q)∂κ )
LX± = ±(i/2)e±2iθ (2y(∂x ∓ i∂y ) − ∂θ )
LZ1 = −i∂θ
8.5 Some Hints Concerning the Jacobi Group 211

acting on differentiable functions φ = φ(g) with the coordinates coming from

g = (p, q, κ)n(x)t(y)r(θ)

where      
1 x y 1/2 α β
n(x) = , t(y) = , r(θ) =
1 y −1/2 −β α
with
α = cos θ, β = sin θ,
and
g(i, 0) = (τ, z) = (x + iy, pτ + q).
As usual, we put
N := {n(x); x ∈ R}, A := {t(a); a ∈ R>0 },
K := SO(2) = {r(θ); θ ∈ R}, M := {±E}.
Elements of the representation theory of GJ (R) can be found in the work of Pyatetskii–
Shapiro, Satake, Kirillov, Howe, Guillemin, Sternberg, Igusa, Mumford and many others.
To a large part, they are summed up in [BeS], the essential outcome being that the
representations π of GJ (R) with nontrivial central character are essentially products
of projective representations of SL(2, R) with a fundamental (projective) representation
m
πSW of GJ called the Schrödinger–Weil representation in [BeS], as it is in turn the
product of the Schrödinger representation πS of the Heisenberg group Heis(R) and the
m
(Segal–Shale–)Weil or oscillator representation πW of SL(2, R). This representation πSW
is a representation of lowest weight k = 1/2 and index m. It is characterized by the fact
that all its K J –types are one-dimensional (K J := SO(2) × C(Heis(R))  SO(2) × R),
m
which in turn is characterized by the fact that the lowest weight vector for πSW satisfies
the heat equation. To be a bit more explicit, the irreducible unitary representations π of
GJ (R) with central character ψ m (with ψ m (x) := em (x)) for m = 0 are infinitesimally
equivalent to

π̂m,s,ν , s ∈ iR, ν = ±1 (π̂m,s,ν  π̂m,−s,ν ),

π̂m,s,ν , s ∈ R, s2 < 1/4,
+
π̂m,k , k ∈ N,
−
π̂m,k , k ∈ N.
Here and in the following, we restrict our treatment to m > 0, but there is a “mir-
ror image” for m < 0 which, to save space, we will not discuss. These infinitesimal
representations are of the form
m
π̂m,s,ν = π̂SW ⊗ π̂s,ν ,
±
π̂m,k m
= π̂SW ⊗ π̂k±0 , k = k0 + 1/2.
m
The Schrödinger–Weil representation π̂SW is given on the space

Vm1/2 := vj j∈N0

by (see [BeS] p.33)

Zvj := (1/2 + j)vj , Z0 vj := µvj , Y+ vj := vj+1 , X+ vj := −1/(2µ)vj+2 , . . . .

212 8. Geometric Quantization and the Orbit Method

+
And, for instance, the discrete series representation π̂k−1/2 of sl(2) is given on

Wk−1/2 := wl l∈2N0

by

Zwl := (k − 1/2 + l)wl , X+ wl := wl+2 , X− wl := (l/2)(k − 3/2 + l)wl−2 .

+
That is, we have as a space for π̂m,k

+
Vm,k := Vm1/2 ⊗ Wk−1/2 = vj ⊗ wl j∈N, l∈2N0

with
Z0 (vj ⊗ wl ) = µ(vj ⊗ wl ),
Z(vj ⊗ wl ) = (k + j + l)(vj ⊗ wl ),
Y+ (vj ⊗ wl ) = vj+1 ⊗ wl ,
X+ (vj ⊗ wl ) = −(1/(2µ))vj+2 ⊗ wl + vj ⊗ wl+2 .
In particular, for the spaces of Z-weight k + λ, λ ≥ 0,

V (λ) := {v ∈ Vm,k
+
; Zv = (k + λ)v}

we have

dim V (0) = 1, dim V (1) = 1, dim V (2) = 2, dim V (3) = 2, dim V (4) = 3, . . .

If we denote by ρm,k̃ the (one-dimensional) representation of K J given by

ρm,k̃ (r(θ), (0, 0, κ)) = e(ik̃θ)+2πimκ , θ ∈ [0, 2π], κ ∈ R,

we have
+
mult (ρm,k̃ , πm,k ) = 1 for k̃ = k,
= 1+l for k̃ = k + 2l or = k + 2l + 1, l ∈ N0
= 0 for k̃ < k,

and
mult (ρm,k̃ , π) = ∞
in the other cases listed above. Now, as to be seen for instance in [BeS] p.28ff, a lowest
+
weight or vacuum vector φ0 for a discrete series representation πm,k of GJ (R) realized by
right translation in a space Hm,k of smooth functions φ = φ(g) on G (R) is characterized
+ J

by the equations

LX− φ0 = LY− φ0 = 0, LZ0 φ0 = 2πmφ0 , LZ φ0 = kφ0 .

The Jacobi group is not semisimple (and indeed a prominent example for a non-reductive
group). We describe its coadjoint orbits using our embedding of GJ into Sp(2, R) :

We identify the dual sp∗ of sp by the Sp(2, R) invariant isomorphism η → Xη given by

η(Y̌ ) = Tr(Xη Y̌ ) =:< Xη , Y̌ > for all Y̌ ∈ sp.

8.5 Some Hints Concerning the Jacobi Group 213

η1 , . . . , η6 denotes a basis of (gJ )∗ dual to

(X1 , . . . , X6 ) := (X, . . . , R).

We realize (gJ )∗ as a subspace of sp by the matrices

⎛ ⎞
x p z 0
⎜ 0 0 0 0 ⎟
M (x, y, z, p, q, r) := ⎜
⎝ y q
⎟ , x, . . . , r ∈ R,
−x 0 ⎠
q r −p 0

and put
X∗ := M (1, 0, . . . , 0), . . . , R∗ := M (0, . . . , 0, 1).

Then
Xη1 = X∗ , . . . , Xη6 = R∗

is a basis of (gJ )∗ with

< X∗ , X >= 2, < Y∗ , Y >= 1, < Z∗ , Z >= 1

< P∗ , P >= 2, < Q∗ , Q >= 2, < R∗ , R >= 1.

By a straightforward computation one obtains the following result:

Lemma: If g −1 is denoted as
⎛ ⎞
a 0 b aµ − bλ
⎜ λ 1 µ κ ⎟
g −1 =⎜
⎝ c
⎟,
0 d cµ − dλ ⎠
0 0 0 1

the coadjoint action of g on M (x, . . . , r) is given by

Ad∗ (g)M (x, . . . , r) = M (x̃, . . . , r̃)

with

x̃ = (ad + bc)x + bdy − acz + (2acµ − (ad + bc)λ)p

+ ((ad + bc)µ − 2bdλ)q + r(aµ − bλ)(cµ − dλ),

ỹ = 2dcx + d2 y − c2 z + 2(cµ − dλ)(cp + dq) + r(cµ − dλ)2 ,

(8.16) z̃ = −2abx − b2 y + a2 z − 2(aµ − bλ)(ap + bq) − r(aµ − bλ)2 ,

p̃ = ap + bq + r(aµ − bλ),

q̃ = cp + dq + r(cµ − dλ),

r̃ = r.
214 8. Geometric Quantization and the Orbit Method

Exercise 8.10: Verify this and determine the coadjoint orbits. For instance, show that
for
U∗ = mR∗ + αX∗ , m = 0, α = 0,
one has the 4-dimensional hyperbolic orbit OU∗ contained in

Mm,−α2 := {(x, y, z.p, q, r) ∈ R6 ; m(x2 + yz − α2 ) = 2pqx − p2 y + q 2 z, r = m}.

Do you get OU∗ = Mm,−α2 ?

It is natural to try to apply the general procedure outlined in this section to realize
representations of GJ by bundles carried by these orbits and to ask, which ones you get
and, if not, why? As already said above, there are some answers in [Be1] and [Ya], but
far from all.
Chapter 9

Epilogue: Outlook to Number

Theory

Representations of groups show up in many places in Number Theory, for instance as

representations of Galois groups. There seems to be no doubt that the relationship of the
two topics culminates in the Langlands Program, which (roughly said) seeks to estab-
lish a correspondence between Galois and automorphic representations. Many eminent
mathematicians have worked and work on this program, and it is now even interesting
for physicists; see Frenkel’s Lectures on the Langlands Program and Conformal Field
Theory [Fr]. We cannot dare to go into this here, but we will try to introduce at least
some initial elements by presenting some representation spaces for the special groups
we treated in our examples, consisting of, or at least containing, theta functions, and
modular and automorphic forms. And we will also introduce the notions of zeta and
L-functions, which ultimately are the foundations of the bridge between the Galois and
automorphic representations. There are many useful books available. We can only cite
some of them, which by now are classic: Representation Theory and Automorphic Func-
tions by Gelfand, Graev and Pyatetskii-Shapiro [GGP], Automorphic Forms on GL(2)
by Jacquet and Langlands [JL], Automorphic Forms on Adele Groups by Gelbart [Ge],
Analytic Properties of Automorphic L-Functions by Gelbart and Shahidi [GS], Automor-
phic Forms and Representations by Bump [Bu], Theta Functions by Igusa [Ig], The Weil
representation, Maslov index and Theta Series by Lion and Vergne [LV], Fourier Analysis
on Number Fields by Ramakrishnan and Valenza [RV], and Mumford’s Tata Lectures on
Theta I, II, and III [Mu].

The goal of these last sections will be to lead a way to automorphic representations:
While we already encountered the decomposition of the regular representation on the
space L2 (G), we shall here take a discrete subgroup Γ of G and ask for the representa-
tions appearing in the decomposition of L2 (Γ \ G), i.e., we consider representation spaces
consisting of functions with a certain periodicity, invariance or covariance property. All
this is most easily understood for the case of the Heisenberg group and leads to the
notion of the theta functions. So we will start by treating this case, and afterwards will
look at the group SL(2, R) and introduce modular forms as examples for more general
automorphic forms. Finally, we shall consider several kinds of L-functions appearing in
connection with representation theory.
216 9. Epilogue: Outlook to Number Theory

Since this is an epilogue and an outlook and the material has such a great extension and
depth, the reader will excuse (we hope) that we shall not be able to give more than some
indications, and even less proofs, than in the preceding chapters.

9.1 Theta Functions and the Heisenberg Group

Whatever we do in this section can easily be extended to higher dimensional cases by
conveniently adopting the notation. As already done before, we denote again z ∈ C and
τ ∈ H and moreover, as usual in this context,

q := e(τ ) = e2πiτ ,  := e(z) = e2πiz .

Then one has the classic Jacobi theta function ϑ given by

2
2
(9.1) ϑ(z, τ ) := eπi(n τ +2nz) = q n /2 n
n∈Z n∈Z

going back to Jacobi in 1828. One has to verify convergence:

Lemma: ϑ converges absolutely and uniformly on compact parts of C × H.

Proof: Exercise 9.1.

Moreover, today, one has variants, which unfortunately are differently normalized by
different authors. For instance in [Ig] p.V, one finds the theta function with characteristics
m = (m , m ) ∈ R2

θm (τ, z) := e((1/2)(n + m )2 τ + (n + m )(z + m )).

n∈Z

In our text, we shall follow the notation of [EZ] p.58 where, for 2m ∈ N, µ = 0, 1, ., 2m−1,
one has

2
(9.2) θm,µ (τ, z) := q r /(4m) r
r∈Z, r≡µ mod 2m

= em ((n + µ/(2m))2 τ + 2(n + µ/(2m))z).

n∈Z

Obviously, we have ϑ(z, τ ) = θ1/2,0 (τ, z). Theta functions have a rather evident quasiperi-
odicity property concerning the complex variable z and a deeper modular property con-
cerning the variable τ ∈ H, which makes them most interesting for applications in number
theory and algebraic geometry, and – concerning both variables – ϑ satisfies the “one-
dimensional” heat equation
4πiϑτ = ϑzz .
At first, we treat the dependance on the complex variable where, as we shall see, the
Heisenberg group comes in. But to give already here at least a tiny hint, for instance,
theta functions are useful if one asks for the number of ways a given natural number can
be written as, say, four squares of integers (see [Ma] p.354 or our Example 9.3 in 9.2).
9.1 Theta Functions and the Heisenberg Group 217

As we keep in mind that the group really behind theta functions is the Jacobi group
introduced in 8.5, we take the version of the Heisenberg group, which comes from its
realization by four-by-four matrices though in other sources different coordinizations are
used (like we did in the last chapter). Thus, we take

G := Heis(R) = {g = (λ, µ, κ) ∈ R3 } with gg  := (λ + λ , µ + µ , κ + κ + λµ − λ µ).

For fixed τ ∈ H, G = Heis(R) acts on C by

C
z −→ g(z) := z + λτ + µ, g = (λ, µ, κ) ∈ G.

Obviously, the stabilizer of z ∈ C is the center C(R) = {(0, 0, κ); κ ∈ R}  R of the

Heisenberg group and one has

X := G(R)/C(R)  C.

This action induces an action on functions F on C in the usual way

F −→ F g with F g (z) := F (g(z)) = F (z + λτ + µ).

Recalling the cocycle condition (7.8) in 7.1 and the first example of an automorphic
factor in 7.2, we refine this action by introducing for m ∈ R as an automorphic factor
for the Heisenberg group

jm (g, z) := em (λ2 τ + 2λz + λµ + κ).

Exercise 9.2: Verify the relation

jm (gg  , z) = jm (g, g  (z))jm (g  , z).

Then we define F |[g] by

F | [g](z) := jm (g, z)F (g(z))
and introduce as our first example of an automorphic form a holomorphic function F on
C with

(9.3) F |[γ] = F for all γ = (r, s, t) ∈ Γ = Heis(Z) = {γ = (r, s, t) ∈ Z3 }.

We denote by Θ(m) the vector space of all these functions F .

Thus, an entire function F (i.e. holomorphic on C) is an element of Θ(m) iff

F (z + s) = F (z) and F (z + rτ ) = e−m (r2 τ + 2rz)F (z) for all r, s ∈ Z.

If we relate this to the theta functions defined above, we are led to the following central
result, which is not too difficult to prove (using the periodicity and the functional equa-
tion (9.3)).

Theorem 9.1: For 2m ∈ N, Θ(m) is spanned by the theta functions

θm,α , α = 0, 1, . . . , 2m − 1

and hence has dimension 2m.

218 9. Epilogue: Outlook to Number Theory

To any function F on C we associate a lifted function φ = φF defined on G = Heis(R)

by
φF (g) := (F |[g])(0) = F (λτ + µ)em (λ2 τ + λµ + κ).
This prescription is to be seen on the background of the relation between the First and
the Second Realization in the induction procedure (see 7.1). Then we come to under-
stand immediately the first two items in the

Proposition 9.1: Under the map F −→ φF , the space Θ(m) is isomorphic to the space
H(m) of functions φ : G −→ C satisfying
i) φ(γg) = φ(g) for all γ ∈ Γ,
ii) φ(gκ) = φ(g)em (κ) for all κ := (0, 0, κ) ∈ C(R),
iii) φ is smooth and satisfies LY− φ = 0.

The third statement is result of a discussion parallel to the one of the Fock representation
in 8.4: here we use Y− := P − τ Q to express the holomorphicity of F by a differential
equation for φ: For

gP (t) = (t, 0, 0),

gQ (t) = (0, t, 0),
gR (t) = (0, 0, t)
one has

ggP (t) = (λ + t, µ, κ − µt),

ggQ (t) = (λ, µ + t, κ + λt),
ggR (t) = (λ, µ, κ + t)

and for the left invariant operators LU acting on functions φ living on G defined by
d
LU φ(g) := φ(ggU (t))|t=0 ,
dt
we get

LP = ∂λ − µ∂κ ,
LQ = ∂µ + λ∂κ ,
LR = ∂κ ,

and
LY− = LP −τ Q = ∂λ − τ ∂µ − (λτ + µ)∂κ .
In the prescription for the lifting from C to G we assume z = λτ + µ, z̄ = λτ̄ + µ and
(via Wirtinger calculus) deduce

∂z̄ = (τ ∂µ − ∂λ )/(τ − τ̄ ).

If we have
φ(λ, µ, κ) = F (z, z̄)em (κ + λ2 τ + λµ),
a tiny computation shows that the condition LY− φ = 0 translates into the holomorphicity
condition ∂z̄ F = 0.
9.1 Theta Functions and the Heisenberg Group 219

This leads us directly to the following representation theoretic interpretation. The con-
ditions i) and ii) of the space H(m) show up if one looks at still another standard
representation of the Heisenberg group, namely the lattice representation:

We take the subgroup

H := {h = (r, s, t); r, s ∈ Z, t ∈ R}
and, for m ∈ Z \ {0}, its character given by

π0 (h) := em (t).

Then we have X = H \ G  Z2 \ R2 . The action of G on C given by

z −→ z + λτ + µ

restricts to the action of H on 0 ∈ C producing the points rτ + s forming a lattice L in

C. By our general procedure from Mackey’s approach in 7.1, we get a representation by
right translation on the space of functions φ on G, which satisfy the functional equation

φ(hg) = em (t)φ(g) for all h = (r, s, t), r, s ∈ Z, t ∈ R, g ∈ G.

(As X is compact, we have no problem with the condition of the finite norm integral.)
This induced representation is not irreducible and we proceed as follows. For Γ = Heis(Z)
we denote by L2 (Γ \ G) the Hilbert space of Γ−invariant measurable functions G −→ C
with the scalar product 
φ, ψ := φ(g)ψ(g)dg.
Γ\G

This is a Hilbert space direct sum

L2 (Γ \ G) = ⊕m∈Z L2 (Γ \ G)m

where L2 (Γ \ G)m denotes the subspace of functions φ that satisfy

φ(gκ) = em (κ)φ(g) for all κ = (0, 0, κ) ∈ Z(R), g ∈ G.

Now for m > 1/2, this induced representation on L2 (Γ \ G)m is not irreducible. We find
here the first example of an important general statement.

Theorem 9.2 (Duality Theorem for the Heisenberg group): For m ∈ N, the multiplicity
of the Schrödinger representation πm in L2 (Γ \ G) is equal to the dimension of the space
Θ(m).

Proof: There are several different approaches. In the our context, we can argue like
this:
The Schrödinger representation πm transforms a function f ∈ L2 (R) into

πm (g0 )f (x) = em (κ0 + (2x + λ0 )µ0 )f (x + λ0 )

and is intertwined with the Heisenberg representation (see Example 7.1 at the beginning
of 7.1) given by right translation on functions φ on G via

f (x) −→ em (κ + λµ)f (λ) = φf (g).

220 9. Epilogue: Outlook to Number Theory

Obviously, one has φ(κg) = em (κ)φ(g) but as the Heisenberg representation is induced
by the two-dimensional subgroup {(0, µ, κ); µ, κ ∈ R} of G, one also has invariance
concerning the µ−variable. Now, to produce a subspace of the lattice representation,
one has to create in- or covariance concerning the λ−variable: We define for n ∈ Z and
α = 0, 1, . . . , 2m − 1
λn,α := (n + α/(2m), 0, 0)
and get
λn,α g = (λ + n + α/(2m), µ, κ + µ(n + α/(2m)).
Hence for every α, we have a map ϑα : f −→ φf,α , called theta transform, with

φf,α (g) := φf (λn,α g)

n∈Z

= em (κ + µ(n + α/(2m)) + (λ + n + α/(2m))µ)f (λ + n + α/(2m)).

n∈Z

It is not difficult to verify that one has for all γ ∈ Γ and κ ∈ R

φf,α (γg) = φf,α (g) and φf,α (gκ) = φf,α (g).
Thus we see that for α = 0, 1, . . . , 2m − 1 every ϑα intertwines the Schrödinger represen-
tation with the lattice representation. We still have to show that these are all different
and that there are no others. To do so, we realize the Schrödinger representation πm by
choosing as vacuum vector f0 not f0 (x) = exp(−2πmx2 ), as in Remark 6.13 in 6.5, but
f0 (x) = em (τ x2 ).
Then this function is annihilated by the differential operator Ŷ− = ∂x − (4πim)τ x from
the derived representation dπm (as in Example 6.9 in 6.3) belonging to Y− = P − τ Q
introduced above. If one applies the theta transform ϑα to f0 one gets
ϑα f0 (g) = em (κ + τ λ2 + λµ)θm,α (τ, λτ + µ),
i.e. exactly the theta basis of the space H(m) appearing in Proposition 9.1 above, which
characterizes the lifting of Θ(m) to functions on G.

The reader could try to find another version of the proof using the bundle approach to
induced representations. One can prove (Theorem of Appell-Humbert) that every line
bundle on the torus X := L \ C, L = Zτ + Z is of the form L(H, β), i.e. the quotient of
the trivial bundle C × C −→ C by the action of L given by
(, (v, z)) −→ (β()eiH(z,)+(1/2)H(,) v, z + ) for all  ∈ L, v, z ∈ C.
Here H : C × C −→ C is a hermitian form such that E = Im (H) is integer valued on L
and β : L −→ S 1 a map with
β(1 + 2 ) = eiπE(1 ,2 ) β(1 )β(2 ).
A computation shows that these hermitian forms are exactly the forms Hm , m ∈ Z, given
by
Hm (z, w) = mz̄w/Im(τ ).
Up to now, we only treated the theta functions as quasiperiodic functions in the complex
variable z. There is also a remarkable modular behaviour concerning the variable τ ∈ H.
Before we discuss this, we return to some more general considerations.
9.2 Modular Forms and SL(2,R) 221

9.2 Modular Forms and SL(2,R)

Besides theta functions and strongly intertwined with it, modular forms is another topic
relating classical function theory to number theory and algebraic and analytic geometry.
There are many competent introductions. A very brief one is given in Serre [Se1]. More-
over we recommend the books by Schöneberg [Sch] and Shimura [Sh] containing more
relevant details.

Classical Theory (Rudiments)

We already met several times with the action of G = SL(2, R) on the upper half plane
X = H := {τ = x + iy ∈ C; y > 0} given by
 
a b aτ + b
(g = , τ ) −→ g(τ ) := .
c d cτ + d
As g and −g produce the same action, it is more convenient to treat the group

Ḡ = PSL(2, R) := SL(2, R)/{±E2 }.

But since we worked so far with G and its representation theory can be rather easily
adopted to Ḡ, we stay with this group.

For k ∈ R (but mainly k ∈ Z) we have (see 7.2.2) the automorphic factor

j(g, τ ) := (cτ + d)−1 resp. jk (g, τ ) := (cτ + d)−k

fulfilling the cocycle relation (7.8) j(gg  , τ ) = j(g, g  (τ ))j(g  , τ ). Let f be a smooth func-
tion on H. Then one writes

f |k [g](τ ) := f (g(τ ))jk (g, τ )

(so that πk (g)f (τ ) := f (g −1 (τ ))jk (g −1 , τ ) at least formally is the prescription of a rep-

resentation).

There are several important types of discrete subgroups Γ of G and the theory becomes
more and more beautiful if one treats more and more of these. For our purpose to give
an introduction, it is sufficient to restrict ourselves to the (full) modular group

Γ = SL(2, Z)

and eventually to the main congruence group of level N

Γ = Γ(N ) := {γ ∈ SL(2, Z); γ ≡ E2 mod N }, N ∈ N,

Hecke’s special congruence subgroup

 
a b
Γ = Γ0 (N ) := {γ = ∈ SL(2, Z); c ≡ 0 mod N }, N ∈ N,
c d

and the theta group Γϑ generated by T 2 and S, i.e.

   
2 0 −1 1 1
Γ = Γϑ := < T , S > with S = and T = .
1 0 0 1
222 9. Epilogue: Outlook to Number Theory

There is a lot to be said about these groups. We mention only two items:

Remark 9.1: Γ = SL(2, Z) is generated by just two elements, namely S and T .

Exercise 9.3: Prove this and realize that T acts on H as a translation τ −→ T (τ ) = τ +1
and S as a reflection at the unit circle τ −→ S(τ ) = −1/τ .
Determine the fixed points for the action of Γ and their stabilizing groups.

Remark 9.2: Γ = SL(2, R) has the following standard fundamental domain

F := {τ ∈ H; |τ | > 1, −1/2 < Re τ < 1/2}.

Here we use the definion from [Sh] p.15: For any discrete subgroup Γ of G = SL(2, R),
we call F a fundamental domain for Γ \ H (or simply for Γ) if
– i) F is a connected open subset of H,
– ii) no two points of F are equivalent under Γ,
– iii) every point of H is equivalent to some point of the closure of F under Γ.
(Other authors sometimes use different definitions.)

Exercise 9.4: Verify this and show that the set of Γ−orbits on H is in bijection to

F  := F ∪ {τ ∈ C; |τ | ≥ 1, Re τ = −1/2} ∪ {τ ∈ C; |τ | = 1, −1/2 ≤ Re τ ≤ 0}.

If you have trouble, search for help in [Sh] p.16 or any other book on modular forms.

We look at functions f : H −→ C with a certain invariance property with respect to the

discrete subgroup. Here we take Γ = SL(2, Z) and k ∈ Z.

Definition 9.1: f is called (elliptic) modular form of weight k iff one has

– i) f |k [γ] = f for all γ ∈ Γ,
– ii) f is holomorphic,
– iii) For Im τ >> 0, f has a Fourier expansion or (q−development)

f (τ ) = c(n)q n , q = e(τ ) = e2πiτ , c(n) ∈ C.

n≥0

f is called cusp form (forme parabolique in French) if one has moreover c(0) = 0.
We denote by Ak (Γ) the space of all elliptic modular forms of weight k and by Sk (Γ)
the space of cusp forms. Both are C vector spaces and their dimensions are important
characteristic numbers attached to Γ. From i) taken for γ = −E2 , one sees immediately
that for odd k there are no non-trivial forms. For even k = 2 one has

dim A2 (Γ) = [/6] if  ≡ 1 mod 6,  ≥ 0

= [/6] + 1 if  ≡
 1 mod 6,  ≥ 0

where [x] denotes the integral part of x, i.e. the largest integer n such that n ≤ x. For a
proof see for instance [Se] p.88 or (with even more general information) [Sh] p.46.

Remark 9.3: One also studies meromorphic modular forms where in ii)  holomorphic is
replaced by meromorphic and in iii) one has a q−development f (τ ) = n>−∞ c(n)q n .
9.2 Modular Forms and SL(2,R) 223

As Γ is generated by S and T , it is sufficient to demand the two conditions

f (τ + 1) = f (τ ), f (−1/τ ) = τ k f (τ ).

Hence a modular form of weight k is given by a series f (τ ) = n≥0 c(n)q n converging
for |q| < 1 and fulfilling the condition

f (−1/τ ) = τ k f (τ ).

Remark 9.4: The condition iii) concerning the Fourier expansion can also be understood
in several different ways, for instance as a certain boundedness condition for y −→ ∞
(in particular for f ∈ Sk (SL(2, Z)), |f (τ )|y k/2 is bounded for all τ ∈ H), or, if k is even,
as holomorphicity of the differential form α = f (τ )dτ k/2 at ∞: one can compactify the
upper half plane H by adjoining as cusps the rational numbers and the “point” ∞, i.e.
one takes H∗ := H ∪ Q ∪ {∞}. Then one extends the action of SL(2, R) on H to an action
on H∗ and identifies Γ \ H∗ with the Riemann sphere or P1 (C). In this sense modular
forms are understood as holomorphic k/2-forms on P1 (C).

All this is made precise in books on the theory of modular forms. Here, for our purpose,
the essential point is that, similar to the theta functions in the previous section, modular
forms are realizations of dominant weight vectors of discrete series representations of
G = SL(2, R). Before we go into this, let us at least give some examples of modular
forms and the way they can be constructed.

Example 9.1: The most well-known elliptic modular form is the ∆−function given by
its q−expansion

(2π)−12 ∆(q) : = q Π∞
n=1 (1 − q )
n 24

= Σ∞
n=1 τ (n)q = q − 24q + 252q − 1472q + . . . .
n 2 3 4

This is a cusp form of weight 12. The function n −→ τ (n) is called the Ramanujan func-
tion. It has many interesting properties, for instance there is the Ramanujan conjecture

τ (n) = O(n(11/2)+ε ), ε > 0,

which has been proved by Deligne in 1974 (as consequence of his proof of the much more
general Weil conjecture concerning Zeta functions of algebraic varieties (see Section 9.6)).
But it is still an open question whether all τ (n) are non-zero.

A modular form in a slightly more refined sense is the η−function, also appearing in the
physics literature in several contexts,

η(τ ) := q 1/24 Π∞
n=1 (1 − q ).
n

Example 9.2: A more comprehensive construction principle leads to the Eisenstein

series Ek resp. Gk . For k = 4, 6, . . . , Gk is defined by

Gk (τ ) := Σ m,n (mτ + n)−k

where the sum is taken over all m, n ∈ Z with exception of (m, n) = (0, 0).
224 9. Epilogue: Outlook to Number Theory

It is not too much trouble to verify that this is really a modular form of weight k and
has a Fourier expansion (see [Sh] p.32)
∞
(2πi)k

Gk (τ ) = 2ζ(k) + 2 σk−1 (n)q n

(k − 1)! n=1

where ζ is Riemann’s Zeta function defined for complex s with Re s > 1 by

∞

ζ(s) := n−s
n=1

and σk (n) denotes the sum of dk for all positive divisors d of n. These functions appear
naturally in the discussion of lattices L = τ Z + Z in C. They can also be introduced by
a far reaching averaging concept similar to the theta transform we used in the previous
section to understand the construction of theta functions:
Every function ϕ on H formally gives rise to an object satisfying the covariance property
i) of a modular form by the averaging

ϕ|k [γ].
γ∈Γ

One has the problem whether this series converges (and has the properties ii) and iii)).
In general this is not to be expected. But one can use the following fact: For g0 = T 
our automorphic factor has the property j(g0 , τ ) = 1 , and, hence, one has

j(g0 g, τ ) = j(g, τ ) = (cτ + d)−1 .

We see that for a constant function ϕ0 , say for ϕ0 (τ ) = 1, we have

ϕ0 |k [T  ] = ϕ0 .

Thus to get a covariant expression, one has only to sum up

Ek (τ ) := (ϕ0 |k [γ])(τ )
γ∈Γ∞ \Γ

with Γ∞ := {T  ;  ∈ Z}. The map

 
a b
Γ −→ Z , g =
2
−→ (c, d)
c d

induces a bijection
Γ∞ \ Γ  {(c, d) ∈ Z2 ; (c, d) = 1}
where here (c, d) denotes the greatest common divisor. Hence, one can write

Ek (τ ) = (cτ + d)−k .
c,d∈Z, (c,d)=1

This Ek has the properties we wanted and is related to Gk by

ζ(k)Ek (τ ) = Gk (τ ).
9.2 Modular Forms and SL(2,R) 225

The same procedure goes through for every Γ∞ -invariant ϕ. For instance, ϕ(τ ) = e(mτ )
leads to the so called Poincaré series, which are used to construct bases for spaces of
cusp forms.

Example 9.3: As already mentioned in 9.1, Jacobi’s theta function ϑ has a modular
behaviour concerning the variable τ . Here we restrict the function to its value at z = 0

2
ϑ(τ ) := ϑ(0, τ ) = eπin τ .
n∈Z

One has the relations

ϑ(τ + 2) = ϑ(τ ), ϑ(−1/τ ) = (τ /i)1/2 ϑ(τ ).

The first one is easy to see and the second one requires some work (see for instance [FB]
p.344/5). These relations entail that this ϑ is a modular form with respect to the theta
group Γϑ if one generalizes the definition of modular forms appropriately. There are
more general theta series associated to quadratic forms resp. lattices leading directly to
modular forms:

Let S ∈ Mn (Z) be a symmetric positive matrix, i.e. with

S[x] := txSx > 0 for all 0 = x ∈ Rn

then one defines an associated theta series by

ϑ(S; τ ) := eπiS[x]τ .
x∈Zn

We get a function with period 2 whose Fourier development

∞

ϑ(S; τ ) = a(S, m)eπimτ

m=0

encodes the numbers

a(S, m) := #{x ∈ Zn ; S[x] = m}
of representations of the natural number m by the quadratic form, which belongs to S.
If we assume that S is moreover unimodular, i.e. S is invertible with S −1 ∈ Mn (Z),
even, i.e. S[x] ∈ 2Z for all x ∈ Zn , and if n is divisible by 8, then ϑ(S; τ ) is an elliptic
modular form of weight n/2. For a proof we refer to [FB] p.352 or [Se1] p.53.

In particular, for

n
an (m) := a(En , m) = #{x ∈ Zn ; x2j = m}
j=1

one has results obtained by Jacobi in 1829 resp. 1828, namely

a8 (m) = 16 (−1)m−d d3 and a4 (m) = 8 d.

d|m d|m,4
d
226 9. Epilogue: Outlook to Number Theory

Proofs can be found for instance in [FB] or [Mu] I. The second relation can be verified
using the nice identity
ϑ4 (τ ) = π −2 (4G2 (2τ ) − G2 (τ /2)).
Example 9.4: As a final example we present the j−invariant or modular invariant,
which is a meromorphic modular form: We put

g2 (τ ) := 60G4 (τ ), g3 (τ ) := 140G6 (τ ).

Then, we have ∆(τ ) = g2 (τ )3 − 27g32 and the j−function is defined by

j(τ ) := 123 g2 (τ )3 /∆(τ ).

j is a modular function with Fourier expansion (see for instance [Sh] p.33)
∞

j(τ ) = q −1 (1 + cn q n ), n ∈ Z.
n=1

This function is of special importance because it induces a bijection Γ\H∗  P1 (C). And
the coefficients cn hide some deep mysteries (essentially assembled under the heading of
monstrous moonshine (see [CN], [FLM])).

Modular Forms as Functions on SL(2,R)

Parallel to the relation between the Second and First Realization in the induction pro-
cedure one defines a lifting of functions f from the homogeneous space X = G/K  Gx0
with K = Gx0 to the group using the automorphic factor j(g, τ ) satisfying the cocycle
condition (7.8) by f −→ φf where we put

φf (g) := f (g(x0 ))j(g, x0 ) = f |[g](x0 ).

Our aim is to translate the conditions i) to iii) from the definition of modular forms f to
appropriate conditions characterizing the lifts of these forms among the functions φ on
G. Classic sources giving all details for this are [GGP], [Ge] and [Bu].

Remark 9.5: If for a subgroup Γ of G and a function f on X one has

(9.4) (f |[γ])(x) = f (γ(x))j(γ, x) = f (x) for all γ ∈ Γ,

the lifted function φ = φf satisfies

ia) φ(γg) = φ(g) for all γ ∈ Γ
and
ib) φ(gκ) = φ(g)j(κ, x0 ) for all κ ∈ K.

Proof: By definition, (9.4), and (7.8), we have

φ(γg) = f (γg(x0 ))j(γg, x0 ),

= f (g(x0 ))j(γ, g(x0 ))−1 j(γg, x0 ),
= f (g(x0 ))j(g, x0 ).
9.2 Modular Forms and SL(2,R) 227

And by definition, (7.8), and K = Gx0 , we have as well

φ(gκ) = f (gκ(x0 ))j(gκ, x0 )

= f (g(x0 ))j(g, x0 )j(κ, x0 )
= φ(g)j(κ, x0 ).

Remark 9.6: In our case G = SL(2, R), x0 = i ∈ X = H = G/K, K = SO(2) and

j(g, τ ) = (cτ + d)−k , for the lifted function we have

φ(g) = f (g(i))(ci + d)−k = f (τ )y k/2 eikϑ

if g is given by the Iwasawa decomposition (7.12)

g = n(x)t(y)r(ϑ).

Proof: In this case one has

ci + d = y −1/2 (−i sin ϑ + cos ϑ) = y −1/2 e−iϑ .

The translation of the holomorphicity condition ii) to a condition restricting a function

φ on G follows the scheme we already used several times: For sl(2, C)c = < Z, X± >
one has the left invariant differential operators (7.16)

LX± = ±(i/2)e±2iϑ (2y(∂x ∓ i∂y ) − ∂ϑ ),

LZ = −i∂ϑ ,

and, hence, for a function φ = φf with φf (g) = ck (g)f (x, y), ck (g) := y k/2 eikϑ

LZ φ = kφ,
LX± φ = ck±2 D̃± f

where
D̃+ = i(∂x − i∂y − ik/y), D̃− = −iy 2 (∂x + i∂y ).
Furthermore, using Wirtinger calculus with

∂τ = (1/2)(∂x − i∂y ), ∂τ̄ = (1/2)(∂x + i∂y )

we get

LZ φ = kφ,
LX± φ = ck±2 D± f

where
D+ = 2i(∂τ + k/(τ − τ̄ )), D− = (i/2)(τ − τ̄ )2 ∂τ̄ .
Hence in particular, we have found:

Remark 9.7: Let φ = φf be the lift of f . The condition ii) that f is holomorphic on H
is equivalent to

ii’) LX− φ = 0.
228 9. Epilogue: Outlook to Number Theory

The last condition iii) is easily seen to be equivalent to the boundedness condition

|f (τ )| < M for Im τ >> 0.

With a bit more work, one translates the cuspidality of a modular form, i.e. f ∈ Sk (Γ),
to the following

Lemma: f ∈ Ak (Γ) is in Sk (Γ) iff for every y0 > 0 there is an M > 0 such that

y k/2 |f (τ )| < M for all y = Im τ > y0 .

Hence one has for the lifted functions φ = φf the following property.

Remark 9.8: The condition iii) translates to

|φ(x, y, ϑ)y −k/2 | < M for y >> 0

and the cusp condition c(0) = 0 to

|φ(g)| < M for all g ∈ G.

The remarks 9.5 to 9.8 together give proof to a fundamental statement.

Theorem 9.3: Via the lifting ϕk , the space Ak (Γ) of modular forms on H is isomorphic
to the space Ak of smooth functions φ on G with
ia) φ(γg) = φ(g) for all γ ∈ Γ = SL(2, Z),
ib) φ(gr(ϑ)) = φ(g)eikϑ for all r(ϑ) ∈ K = SO(2),
ii’) LX− φ = 0,
iii’) |φ(g)y −k/2 | < M for all y >> 0.
Moreover, the space of cusp forms Sk (Γ) is isomorphic to

A0k = {φ ∈ Ak ; |φ| < M }.

Remark 9.9: The cusp condition in Remark 9.8 and in the Theorem above is equivalent
to

(cusp) Γ∩N \N
φ(n(x)g)dx = 0 for all g ∈ G.

This follows by identifying N = {n(x); x ∈ R} with R and computing

  1
φ(n(x)g)dx = φ(n(x)g)dx
Γ∩N \N 0
 1
= f (τ + x) eikϑ y k/2 dx
0
= eikϑ y k/2 c(0).

Now we can refine Theorem 9.3 above:

9.2 Modular Forms and SL(2,R) 229

Theorem 9.4: The space of cusp forms A0k is characterized by the following conditions
ia) φ(γg) = φ(g) for all γ ∈ Γ = SL(2, Z),
ib) φ(gr(ϑ)) = φ(g)eikϑ for all r(ϑ) ∈ K = SO(2),
ii) ∆φ = LΩ φ = k((k/2) − 1)φ,
iii) φ is bounded,

iv) Γ∩N \N
φ(n(x)g)dx = 0 for all g ∈ G.

Here
∆ = LΩ = 2y 2 (∂x2 + ∂y2 ) − 2y∂x ∂ϑ
is the Laplacian (7.17) from 7.2 and, by a small calculation, it is easy to see that the lift
φf of a cusp form f ∈ Sk (Γ) satisfies the condition ii) and hence all the conditions from
the theorem. The proof of the fact that the lifting map ϕk is a surjection onto the space
given by the conditions ia) to iv) reduces to the verification that for φ ∈ A0k the function
f with

(9.5) f (τ ) := φ(g)y −k/2 e−ikϑ

is well defined for each g with g(i) = τ (which is quite obvious) and that f is a cusp form.
This assertion is based on a considerably deeper theorem, the Theorem of the discrete
spectrum, which we shall state below.

Decomposition of L2 (Γ \ G)

For G = SL(2, R) ⊃ Γ = SL(2, Z), we put H = L2 (Γ \ G) with respect to the biinvariant

measure dġ on X = Γ \ G fixed by
1 dxdy
dg = dϑ.
2π y 2
We want to discuss the decomposition of the representation ρ of G given by right trans-
lation on H and show how the cusp forms come in here. A first indication is given by the

Remark 9.10: For f ∈ Sk (Γ), the lift φf = ϕk (f ) is in H.

This follows from  

|φf (ġ)|2 dġ = |f (τ )|2 y k−2 dxdy < ∞
X F
by the boundedness condition iii’) for the cusp form f .

We need another tool from functional analysis (for background see for instance [La1]
p.389ff):

Remark 9.11: The minimal closed extension of the Laplacian ∆ is a selfadjoint operator
ˆ As ∆ does by constuction, the extension ∆
∆. ˆ also commutes with the right translation
and hence, by Schur’s Lemma (“the unitary Schur” mentioned in 3.2) ∆|H0 is scalar for
each G-invariant subspace H0 of H. From functional analysis we take over that ∆ˆ and ρ
have a discrete spectrum only for the subspace of cuspidal functions

H0 := {φ ∈ L2 (Γ \ G) ; φ(n(x)g)dx = 0 for almost all g ∈ G}.
Γ∩N \N
230 9. Epilogue: Outlook to Number Theory

The central fact is as follows.

Theorem 9.5 (Theorem of the discrete spectrum): The right regular representation ρ of
SL(2, R) on H0 is completely reducible

H0 = ⊕j∈Z Hj

and each irreducible component has finite multiplicity.

As this leads too much into (most fascinating) functional analysis, we can not go into the
proof of this theorem and refer to [La1] and [GGP] p.94ff. A main point is a multiplicity-
one statement going back to Godement, which we already mentioned (without proof)
in 6.4. But we try to give an indication of the power of this theorem: For the proof of
characterization of cusp forms by the Laplacian in the theorem above, we have to show
that the function f from (9.5)

f (τ ) = φ(g)y −k/2 e−ikϑ

is holomorphic. If one decomposes φ as in Theorem 9.5

φ= φj ,
j

the components φj have to fulfill the same relation ib) as φ does. And one has

∆φj = λk φj , λk = k((k/2) − 1),

because by the last remark above, ∆ is scalar on Hj , i.e. one has a λj with

∆φj = λj φj .

And then using the inner product in H one has

< φ, ∆φ > = λj < φ, φj >

ˆ is self adjoint,
and, as ∆

< φ, ∆φj > = < ∆φ, φj > = λk < φ, φj >

and therefore λj = λk . Hence, for the restriction of ρ to its irreducible subrepresentations

the components φj are vectors of lowest weight k for the discrete series representation
πk+ . Such vectors are annihilated by LX− and, hence, f is holomorphic.

As another consequence we come to the statement parallel to the one we proved in 9.1
for theta functions.

Theorem 9.6 (Duality Theorem for the discrete series of SL(2, R)): The multiplicity
of πk− in the right regular representation ρ on the space H0 of cuspidal functions is equal
to the number of linear independant cusp forms of weight k

mult(πk− , ρ) = dim Sk (Γ).

9.2 Modular Forms and SL(2,R) 231

Proof: There is a nice and comprehensive classic proof in [GGP] p.53-57. As we al-
ready have at hand the necessary tools, we reproduce its highlights changing only the
parametrization of G = SL(2, R) used by these authors to the one used here all the time.

a)At first, we show that mk := mult(πk− , ρ) is equal to the number m̃k of linear indepen-
dent functions Ψ on X = Γ \ G in H0 with

(9.6) ∆Ψ = λk Ψ, λk := k((k/2) − 1), ρ(r(ϕ))Ψ = e−ikϕ Ψ for all ϕ ∈ R.

ai) For each irreducible subspace Hj of H0 , which is a model for πk− , there is - up to a
nonzero factor - exactly one highest weight vector where ρ restricted to Hj has highest
weight −k and, hence ∆ has eigenvalue λk .
aii) Every Ψ ∈ H0 satisfying (9.6) is a linear combination of vectors of highest weight
from irreducible subspaces of H0 equivalent to a representation space of πk− .

b) In the second step, we show that m̃k is equal to the number of cusp forms:
The space of functions Ψ as above is in bijection to the space of functions Φ on G with

(9.7) ∆Φ = λk Φ, Φ(gr(ϕ)) = e−ikϕ Φ(g), Φ(γ −1 g) = Φ(g) for all ϕ ∈ R, γ ∈ Γ.

We describe G by our parameters τ, ϑ with τ = g(i), e−iϑ = (ci + d)/ | ci + d | .

Hence, for g0−1 with components a0 , b0 , c0 , d0 , the matrix g0−1 g has parameters τ1 , ϑ1 with
τ1 = g0−1 (τ ), ϑ1 = ϑ − arg(c0 τ + d0 ). And gr(ϕ) has the parameters τ, ϑ + ϕ. Thus the
second equation in (9.7) shows that Φ(g) is a function of type f1 (τ )e−ikϑ . Guided by the
proof of our Remark 9.7, in order to also fulfill the other conditions in (9.7) we try the
ansatz
Φ(g) = e−ikϑ y k/2 f (τ ).
−k
Using our previous notation, this can also be written as Φ(g) = j(g, i) f (τ ). We see
that using this ansatz, the third condition in (9.7) is equivalent to the functional equation
for modular forms

f (γ −1 (τ ))j(γ −1 , τ )−k = f (τ ) for all γ ∈ Γ.

For the real coordinates x, y of τ the first equation in (9.7) comes down to

y 2+(k/2) e−ikϑ (f¯xx + f¯yy ) + iky 1+(k/2) e−ikϑ (f¯x − if¯y ) = 0

resp.
y(fxx + fyy ) − ik(fx + ify ) = 0
If f is a modular form, ergo holomorphic, one has fτ̄ = 0 and this equation is fulfilled.
On the other hand, if Φ fulfills (9.7), by the Decomposition Theorem, it is a linear
combination of highest weight vectors with higest weight −k, which are annihilated by
LX+ = e2iϑ (2y(∂x − i∂y ) − ∂ϑ ). And this again leads just to the condition fτ̄ = 0 showing
that Φ corresponds to a linear combination of cusp forms.

The reader again is invited to watch very closely whether all signs are correct. It has to
be emphasized that the notions highest and lowest weight depend on our parametrization
of K = SO(2) and interchange if one uses r(−ϑ) in place of our r(ϑ) (as Gelbart does in
[Ge]). [GGP] wisely use the term vector of dominant weight.
232 9. Epilogue: Outlook to Number Theory

Remark 9.12: It is a natural question to ask under which conditions the lift φ = φf
of a cusp form f is a cyclic vector, i.e. generates an irreducible representation. There
is a nice answer to this: if f is an eigenfunction to the Hecke operators T (p). We will
introduce these operators later in the context of associating L−functions to modular
forms and here only mention that the space of cusp forms has a basis consisting of such
eigenfunctions.

The big difference between this duality statement and the one for the Heisenberg group
in 9.1 is that in this case we have no way to an explicit construction of the modular forms
starting from their property of being a dominant weight vector. Apparently, this reflects
the fact that the structure of the group at hand is considerably more complicated than
that of the Heisenberg group. Some progress here should give a deep insight into the
modular group.

The characterization theorem is the starting point for several generalizations leading to
the notion of automorphic forms. It can be extended to other groups, even adelic ones
as to be found for instance in [Ge]. There it is also pointed out that the right habitat
for the modern theory of modular forms is not SL(2) but GL(2). Here we can not be too
ambitious and we stay with G = SL(2, R), K = SO(2) and Γ = SL(2, Z) and only treat
the following generalization.

Definition 9.2: The smooth function φ on G is called an automorphic form for Γ iff φ

satisfies the following conditions
ia) φ(γg) = φ(g) for all γ ∈ Γ,
ib) φ is right K−finite, i.e. dim φ(gr(ϑ)); r(ϑ) ∈ K < ∞,
ii) φ is an eigenfunction of ∆,
iii) φ fulfills the growth condition that there are C, M with |φ(g)| ≤ Cy M for y −→ ∞.

φ is called cusp form if moreover one has


(cusp) Γ∩N \N
φ(n(x)g)dx = 0.

Besides the cusp forms realizing highest weight vectors of πk− we just treated, as their
counterparts there come in forms realizing the lowest weight vectors of πk+ , corresponding
to antiholomorphic modular forms on H and the famous Maass wave forms: these are
bounded right K-invariant eigenfunctions of the Laplacian ∆φ = λs φ with
λs = −(1 − s2 )/2, s ∈ iR.
In analogy to the duality theorem stated above, one has here the fact that the dimension
of the space Ws (Γ) of these wave forms is equal to the multiplicity of the principal series
representation πis in ρ (for this and in general more precise information see [GGP] in
particular p.50). In [Ge] p.29 we find the statement: one knows that Ws (Γ) is non-trivial
for infinitely many values of s, but almost nothing is known concerning the specific val-
ues of s for which Ws (Γ) is non-trivial. Still now, Maass wave forms are an interesting
subject. For information on more recent research we recommend the paper by Lewis and
Zagier [LZ].

We finish this section by some remarks concerning the continuous spectrum of the Lapla-
cian, resp. the orthocomplement Hc of the subspace Hd of H = L2 (Γ \ G) consisting of
9.2 Modular Forms and SL(2,R) 233

the cuspidal and the constant functions. One has rather good control about this space
as it can be spanned by the so called non-holomorphic Eisenstein series. As sources for
this we go back to [Ku] and [La1] p.239ff.

The classical object is fixed as follows. One takes

fµ (τ ) := y µ , µ ∈ C

and for k ∈ N0

Ek (τ, µ) : = (f(µ−k)/2 |k [γ])(τ ),

γ∈Γ∞ \Γ

y (µ−k)/2 1
=
|cτ + d|µ−k (cτ + d)k
c,d∈Z, (c,d)=1

 
1 b
with Γ∞ = {γ = ; b ∈ Z}.
1
This prescription formally has the transformation property of a modular form of weight
k and it can be verified that for µ with Re µ > 2 it defines a (real analytic) function
on H, which is non-cuspidal. Using the standard lift ϕk , this function can be lifted to a
function on G

Ek (g, µ) : = ϕk Ek (τ, µ),

cτ̄ + d k/2 1
= y µ/2 eikϑ ( ) ,
cτ + d |cτ + d|µ
c,d∈Z, (c,d)=1

= φs,k (γg),
γ∈Γ∞ \Γ

where for µ = is + 1 the functions (8.11) from 8.3 φs,k , k ∈ Z, with

φs,k (g) = y (is+1)/2 eikϑ

come in, which span the space of the principal series representations. We have

∆Ek (., µ) = −((s2 + 1)/2) Ek (., µ)

and these functions are further examples for the definition of automorphic forms above.

Already in our (brief) chapter on abelian groups, we met with a continuous decomposi-
tion of the regular representation for G = R by throwing in elements of Fourier analysis
without really explaining the background from functional analysis. Now, this gets worse
here: There is the central statement ([Ge] p.33):

Theorem 9.7: The restriction ρc of ρ to Hc is the continuous sum of the principal series
representations πis,+ , i.e.
 ∞
c
ρ (g) = πis,+ (g)ds.
0
234 9. Epilogue: Outlook to Number Theory

Using the inner product for functions f, f˜ on H given by


(f, f˜) := f (τ )f˜(τ )y −2 dxdy
F

one can say that any slowly increasing function f on H, which is orthogonal to all cusp
forms, may be expressed by a generalized Fourier integral
 ∞
y (is)/2
f (τ ) = fˆ(s)E(τ, s)ds, fˆ(s) := (f, E(., s)), E(τ, s) := .
0 |cτ + d|2s
c,d∈Z;(c,d)=1

Though we will not try to get to the bottom of the proofs, we indicate the following also
otherwise useful tools:

The Theta Transform θ

We already used the averaging procedure while introducing Eisenstein series: If ϕ is a

function on N \ G, one gets, at least formally, a function θϕ on Γ \ G by the prescription

θϕ(g) := ϕ(γg).
γ∈Γ∞ \Γ

It is part of nice analysis to discuss under which conditions this leads to convergent ob-
jects and it is no wonder that it works at least if ϕ is a Schwartz function.

The Constant Term θ∗

Let φ be an element of H := L2 (Γ \ G). To this φ we associate θ∗ φ (again at least

formally) by

θ∗ φ(g) := φ(ng)dn.
Γ∞ \N

As φ can be assigned a periodic function, which one denotes again by φ, θ∗ φ may be

understood as the constant term in a Fourier expansion

φ(n(x)g) = φj (g)e(jx).
j

θ and θ∗ are adjoint operators if we introduce an inner product


< ϕ, ϕ̃ >N \G := ϕ(ġ)ϕ̃(ġ)dġ,
N \G

i.e. we have for functions assuring the convergence of our prescriptions the relation

< θϕ, φ >Γ\G = < ϕ, θ∗ φ >N \G .

Exercise 9.5: Verify this.

9.2 Modular Forms and SL(2,R) 235

The Zeta Transform Z

This is a prescription to produce elements in the representation space His,± of a principal

series representation πis,± : For µ ∈ C and ϕ a function on N \ G, we put (again at least
formally)
 ∞
dy
Z(ϕ, g, µ) := ϕ(t(y)g)y −µ .
0 y
For µ = (is + 1)/2, we get
 ∞
dy
ϕ(t(y0 y)g)y −µ
(is+1)/2
Z(ϕ, n(x0 )t(y0 )g, µ) = = y0 Z(ϕ, g, µ),
0 y

i.e. an element fulfilling the equation (7.11) in 7.2 characterizing the elements of His,+ .

The following statements are easy to prove.

Remark 9.13: For ϕ in the Schwartz space S(N \G) and Re µ > 0, the integral defining
Z(ϕ, g, µ) converges absolutely and Z(ϕ, g, µ) is an entire function in µ if moreover ϕ has
finite support.

With some more work composing the Zeta and the Theta transform, one obtains a
Γ−invariant function:

Remark 9.14: For ϕ ∈ S(N \ G) and Re µ > 1, the Eisenstein series

E(ϕ, g, µ) := θZ(ϕ, g, µ) = Z(ϕ, γg, µ)

γ∈Γ∞ \Γ

is absolutely convergent.

By a small computation, we get the relation between this Eisenstein series and the one
we defined above.

Remark 9.15: For

ϕ(g) := ϕ,k (g) = y − e−π/y eikϑ ,  ∈ N0 , k ∈ Z,

one has
E(ϕ, g, µ) = π −(+µ) Γ( + µ)Ek (g, µ).

The functions ϕ,k ,  ∈ N0 , k ∈ Z are a Hilbert basis of the space L2 (N \ G) with respect
to the measure dν = y −2 dydϑ. For every fixed  and µ = is + 1, the images of these ϕ,k
by the Zeta transform are a basis for the representation spaces His,± if one takes all even
resp. odd k ∈ Z.

There is still much to be said here. We shall come back to the Zeta transform in the
context of a discussion of ζ− and L−functions. But before doing this, we shall finish our
outlook to automorphic forms by returning to theta functions.
236 9. Epilogue: Outlook to Number Theory

9.3 Theta Functions and the Jacobi Group

In 9.1 we introduced theta functions and studied their behaviour as functions of the
complex variable z. But as we already indicated and partially explored, there also is
an important dependance on the modular variable τ ∈ H. This leads to the ultimate
interpretation of (lifts of) theta functions as automorphic forms for the Jacobi group GJ ,
which we introduced in 8.5. We defined

q := e(τ ) = e2πiτ ,  := e(z) = e2πiz .

and the classic Jacobi theta function ϑ (9.1) given by

2
2
ϑ(z, τ ) := eπi(n τ +2nz) = q n /2 n ,
n∈Z n∈Z

the theta function with characteristics m = (m , m ) ∈ R2

θm (τ, z) := e((1/2)(n + m )2 τ + (n + m )(z + m )),

n∈Z

and its variant (9.2) (following the notation of [EZ] p.58) with 2m ∈ N, µ = 0, 1, . . . , 2m − 1

2
θm,µ (τ, z) := q r /(4m) r
r∈Z, r≡µ mod 2m

= em ((n + µ/(2m))2 τ + 2(n + µ/(2m))z).

n∈Z

One has ϑ(z, τ ) = θ1/2,0 (τ, z). We already explored the quasiperiodicity property con-
cerning the complex variable z
2
ϑ(z + rτ + s, τ ) = e−πi(r τ +2rz)
ϑ(z, τ )

and (in Example 9.3 in 9.2) the modular property of the value ϑ(τ ) := θ(0, τ ) for z = 0
(also called by the German word Thetanullwert)

ϑ(τ + 2) = ϑ(τ ), ϑ(−1/τ ) = (τ /i)1/2 ϑ(τ ).

To be consistent with [EZ] and the usual Jacobi Theory, we have to change (from Jacobi’s
and Mumford’s) notation and put

θ(τ, z) := ϑ(z, τ ).

Then one has the fundamental modular transformation relation.

Theorem 9.8: For a, b, c, d ∈ Z with ad − cd = 1 and ab, cd ∈ 2Z, there is an eighth

root of unity ζ such that one has

aτ + b z
(9.8) θ( , ) = (cτ + d)1/2 ζ e((1/2)cz 2 /(cτ + d))θ(τ, z).
cτ + d cτ + d
The unit root ζ can be determined as follows:
9.3 Theta Functions and the Jacobi Group 237

We
 may  assume c > 0 or c = 0 and d > 0 (otherwise one would multiply the matrix
a b
by −1) and hence have (cτ + d)1/2 in the first quadrant, i.e. with real- and
c d
imaginary part > 0.
We take the Jacobi symbol ( ab ), i.e. the multiplicative extension of the Legendre symbol,
which for a, p ∈ Z, p prime is given by

( ap ) = 0 if p | a,
= 1 p a and there is an x ∈ Z with x2 ≡ a mod p,
= −1 p a and there is no x ∈ Z with x2 ≡ a mod p.

Then one has

c
ζ = i(d−1)/2 ( |d| ) for c even and d odd,
= e−πic/4 ( dc ) for c odd and d even.

The proof of this theorem is a very nice piece of analysis but too lengthy for this text.
So we refer to the books of Eichler [Ei], in particular p.59-62, Lion-Vergne [LV] p.145ff,
or Mumford [Mu] I.

Another most remarkable fact is the following characterization of θ, which comes about
as an exercise in complex function theory.

Theorem 9.9: Let

f : H × C −→ C
be a holomorphic function satisfying
a) f (τ, z + 1) = f (τ, z),
b) f (τ, z + τ ) = f (τ, z)e(−(1/2)τ − z),
c) f (τ + 1, z + 1/2) = f (τ, z),
d) f (−1/τ, z/τ ) = f (τ, z)(−iτ )1/2 e((1/2)z 2 /τ ) for all (z, τ ) ∈ C × H,
e) limImτ −→∞ f (τ, z) = 1.
Then one has f = θ.

By a closer look one sees that the quasi-periodicity and the modular transformation
property of θ can be combined to a transformation property under an action of the
Jacobi group GJ , which is a semi-direct product of SL(2, R) and Heis(R) and in 8.5 was
realized as a subgroup of Sp(2, R) given by four-by-four matrices. As in 8.5, we use two
notations    
a b a b
GJ
g = (λ, µ, κ) = (p, q, r) .
c d c d
By a short calculation one can verify that for (τ, z) ∈ H × C

aτ + b z + λτ + µ
(τ, z) −→ g(τ, z) := ( , )
cτ + d cτ + d

defines an action of GJ on H × C. The stabilizing group of (i, 0) is K J := SO(2) × C(R),

C(R) the center of the Heisenberg group. One has

GJ /K J  H × C, g(i, 0) = (τ = x + iy, z = pτ + q) where g = (p, q, r)n(x)t(y)r(ϑ).

238 9. Epilogue: Outlook to Number Theory

And for k, m ∈ Q one has (see the first pages of [EZ]) an automorphic factor

c(z + λτ + µ)2
(9.9) jk,m (g; (τ, z)) := (cτ + d)−k em (− + λ2 τ + 2λz + λµ + κ).
cτ + d

This leads to an action of GJ on functions f living on H × C via

f −→ f |k,m [g] with f |k,m [g](τ, z) := f (g(τ, z))jk,m (g; (τ, z)).

Similar to modular forms and theta functions as automorphic forms for SL(2, R) resp.
Heis(R), for the Jacobi group GJ , its discrete subgroup ΓJ := SL(2, Z)
Z3 , and
k, m ∈ N0 one defines (as in [EZ]):

Definition 9.3: A function f : H × C −→ C is called a Jacobi form of weight k and

index m if one has
– i)f |k,m [γ](τ, z) = f (τ, z) for all γ ∈ ΓJ ,
– ii) f is holomorphic,
– iii) for Im τ >> 0, f has a Fourier development

f (τ, z) = c(n, r)q n εr , q = e(τ ), ε = e(z), c(n, r) ∈ C.

n,r∈Z, 4mn−r 2 ≥0

f is called a cusp form if it satisfies moreover

– iii’) c(n, r) = 0 unless 4mn > r2 .
cusp
The vector spaces of all such functions f are denoted by Jk,m resp. Jk,m . They are
finite dimensional by Theorem 1.1 of [EZ]. As usual, one lifts a function f on H × C to
a function φ = φf on GJ by

φf (g) := f (g(i, 0))jk,m (g; (i, 0)) = f (x, y, p, q)em (κ+pz)eikϑ y k/2 , g = (p, q, κ)n(x)t(y)r(ϑ).

Similarly to the analysis in the former cases, one can characterize the images of Jacobi
forms.

Proposition 9.2: Jk,m is isomorphic to the space Ak,m of complex functions C ∞ (GJ )
with
– i) φ(γg) = φ(g) for all γ ∈ ΓJ ,
– ii) φ(gr(ϑ)(0, 0, κ)) = φ(g)em (κ)eikϑ ,
– iii) LX− φ = LY− φ = 0,
– iv) φ(g)y −k/2 is bounded in domains of type y > y0 .
cusp
Jk,m is isomorphic to the subspace A0k,m of Ak,m with
– iv’) φ(g) is bounded.

For a proof of this and the following statement see [BeB] or [BeS] p. 76ff.
9.4 Hecke’s Theory of L−Functions Associated to Modular Forms 239

cusp
Theorem 9.10 (Duality Theorem for the Jacobi Group): For m, k ∈ N, dim Jk,m is
J
equal to the multiplicity of the appropriate discrete series representation of G in the
right regular representation 0 on the space Hm0
of cuspidal functions of type m, which
consists of the elements φ ∈ L (Γ \ G ) satisfying the functional equation
2 J J

φ(g(0, 0, κ)) = e(mκ)φ(g), for all κ ∈ R

and fulfilling the cusp condition


φ(gλ/(2m) g)dn = 0 for almost all g ∈ GJ and λ = 0, . . . , 2m − 1
N J ∩ΓJ \N J

where N J := {(0, q, 0)n(x); q, x ∈ R} and gλ := (λ, 0, 0).

Further information concerning the continuous part of the decomposition of L2 (ΓJ \ GJ )

is collected in [BeS].

At a closer look, one can see that the classic ϑ has the transformation property of a Jacobi
form of weight and index 1/2 for a certain subgroup of ΓJ . As already the infinitesi-
mal considerations indicated, one has to multiply with modular forms of half-integral
weight to get Jacobi forms in the sense of the definition above. This is very nicely ex-
ploited in [EZ]. The appearance of the half-integral weights can be explained by the fact
that, somewhat similar to the relation of the representation theory of SO(3) and SU(2)
resp. SO(3, 1) and SL(2, C), here one has coming in a double cover of SL(2, R) called
the metaplectic group Mp(2, R). But this group is no more a linear group and hence,
unfortunately, outside the scope of this book.

9.4 Hecke’s Theory of L−Functions of Modular Forms

∞
It is a classical topic of analysis to study the relation of a power series f := n=1 c(n)q n
to a Dirichlet series

∞
c(n)
Lf (s) := .
n=1
ns

In our context, we take a cusp form f ∈ Sk (Γ) and get an associated Dirichlet series
convergent for s with Re s > k as the coefficients c(n) of the Fourier expansion of f fulfill
an appropriate growth condition. The most important Dirichlet series have an Euler
product similar to the prototype of a Dirichlet series, namely the Riemann Zeta series
∞

ζ(s) := n−s ,
n=1

which is convergent for s with Re s > 1 and has the Euler product (taken over all prime
numbers p)
1
ζ(s) = Π .
1 − p−s
Moreover, it is known that ζ fulfills a functional equation under s −→ 1 − s and can be
continued to a meromorphic function on the whole plane C with a simple pole at s = 1.
240 9. Epilogue: Outlook to Number Theory

Roughly said, Hecke’s theory analyses how these and more general properties are related
to properties of modular forms. The main tools are the following operators.

Hecke Operators

Definition 9.4: For a natural number n one introduces the Hecke operator T (n) as an
operator acting on functions f on the upper half plane H by

aτ + b
(9.10) T (n)f (τ ) := nk−1 d−k f ( ).
d
a≥1,ad=n,0≤b<d

The origin of this prescription can be understood in several ways. There is the more
general background asking for the decomposition of double cosets into cosets for Γ: For
instance, for a prime p one has
    
1 1 b
Γ Γ= Γ
p p
0≤b<p

(as exploited in [Sh] p.51ff). But we follow the presentation in [Se] p.98 based on the
notion of correspondences: Let E be a set and let GE be the free abelian group generated
by E. A correspondence on E with integer coefficients is a homomorphism T of GE to
itself. We can describe T by its values on the elements x of E:

T (x) = ny (x)y, ny (x) ∈ Z, ny (x) = 0 for almost all y.

y∈E

A C-valued function F on E can be extended linearly to a function on GE , which is

again denoted by F . The transform of F by T , denoted T F , is the restriction to E of
the function F ◦ T . With the notation introduced above, it is given by

T F (x) = F (T (x)) = ny (x)F (y).

y∈E

Let R be the set of lattices L in C and n ∈ N. We denote by T (n) the correspondence

on R which transforms a lattice to the sum (in GR ) of its sublattices Ln of index n.
Thus we have for L ∈ R

T (n)L = Ln .
[L:Ln ]

The sum on the right side is finite. If n is prime, one sees that one has n + 1 such
sublattices. For the general case one has

Lemma: There is a bijection between the set of matrices

 
a b
M= ; ad = n, a ≥ 1, 0 ≤ b < d,
0 d

and the set of lattices Ln of index n in L. For L with basis (τ1 , τ2 ) this bijection is given
by associating to the matrix M the lattice with basis τ  1 = aτ1 + bτ2 , τ  2 = d τ2 .

Exercise 9.6: Prove this.

9.4 Hecke’s Theory of L−Functions Associated to Modular Forms 241

One also uses homothety operators Rλ , λ ∈ C∗ defined by

Rλ L = λ L.

It makes sense to compose the correspondences T (n) and Rλ , since they are endomor-
phisms of the abelian group GR .

Proposition 9.3: The correspondences T (n) and Rλ verify the identities

Rλ Rµ = Rλµ , λ, µ ∈ C∗ ,
Rλ T (n) = T (n)Rλ , n ∈ N, λ ∈ C∗ ,
T (m)T (n) = T (mn), if (m, n) = 1,
T (pn )T (p) = T (pn+1 ) + pT (pn−1 )Rp , p prime, n ∈ N.

Proof: We refer to [Se] p.98.

Corollary 1: The T (pn ) are polynomials in T (p) and Rp .

Corollary 2: The algebra generated by the Rλ and the T (p), p prime, is commutative;
it contains all the T (n).

For the description of the action of these operators on modular forms of weight k, we
associate to f a corresponding function F defined on the set R of lattices by

F (L) := (τ1 /τ2 )−k f (τ1 /τ2 )

if (τ1 , τ2 ) is a basis of L. We define T (n)f as the function on H associated to the function

nk−1 T (n)F on R, i.e.
T (n)f (τ ) := nk−1 T (n)F (L(τ, 1))
where L(τ, 1) denotes the lattice with basis (τ := τ1 /τ2 , 1) (it can be arranged and is
silently understood that the description of the lattices is chosen such that this τ is in H).
Using the Lemma above, we get the formula (9.10) in the definition at the beginning.
Certainly, one has to and can verify that this T (n) transforms the spaces of modular
forms of weight k into itself (and similarly for cusp forms). We leave this to the reader
and only note the relations, which are immediate consequences of Proposition 9.3

(9.11) T (m)T (n)f = T (mn)f, if (m, n) = 1,

(9.12) T (pn )T (p)f = T (pn+1 )f + pk−1 T (pn−1 )f, p prime n ∈ N.

The central point of all this is the action

 of the Hecke operators expressed by the coeffi-
cients of the q−development f (τ ) = c(n)q n :

Proposition 9.4 We have

T (n)f (τ ) = γ(m)q m
m∈Z

with

(9.13) γ(m) = ak−1 c(nm/a2 ).

a∈N,a|(n,m)
242 9. Epilogue: Outlook to Number Theory

Using the relations obtained earlier, the proof is straightforward. Again we refer to [Se]
p.101. In particular, we come to

γ(0) = σk−1 (n)c(0), σk (n) := dk ,

d∈N,d|n
γ(1) = c(n),
γ(m) = c(pm) if n = p prime and m ≡ 0 mod p,
= c(pm) + p k−1
c(m/p) m≡0 mod p.

Since the Hecke operators commute, one can hope for simultaneous diagonalizability.

Hecke Eigenforms

Let f be a Hecke eigenform of weight k, i.e. a modular form of weight k, which is not
identically zero and an eigenfunction for all T (n), n ∈ N,

T (n)f (τ ) = λ(n)f (τ ).

Examples are the Eisenstein series Gk and the discriminant ∆. One has

T (n)Gk = σk−1 (n)Gk , T (n)∆ = (2π)12 τ (n)∆, for all n ∈ N.

The verification is non-trivial but standard ([Se] p.104).

Our central statement is as follows:

Theorem 9.11: The coefficient c(1) of a Hecke eigenform f is non-zero. If f is normalized

by c(1) = 1, one has
λ(n) = c(n) for all n ∈ N.
Proof: The formula (9.13) in Proposition 9.4 above shows that the coefficient of q in
T (n)f is c(n). On the other hand, as f is a T (n)−eigenfunction, it is also λ(n)c(1).
Thus we have c(n) = λ(n)c(1). If c(1) were zero, all the c(n), n > 0 would be zero, and
f would be a constant, which is absurd.

Applying the first statement of the Theorem to the difference of two normalized Hecke
eigenforms, we get

Corollary 1: Two modular forms of weight k, k > 0, which are both normalized Hecke
eigenforms with the same eigenvalues, coincide.

By the second statement of the Theorem the relations (9.11) for the Hecke operators
translate immediately to relations for the Fourier coefficients. We get

Corollary 2: For the Fourier coefficients of a normalized Hecke eigenform f of weight

k one has the relations

(9.14) c(m)c(n) = c(mn), if (m, n) = 1,

(9.15) c(pn )c(p) = c(pn+1 ) + pk−1 c(pn−1 ), if p is prime and n ∈ N.
9.4 Hecke’s Theory of L−Functions Associated to Modular Forms 243

 n
Now, these relations for the Fourier coefficients of a Hecke eigenform f = −sc(n)q trans-
late to the statement that the associated Dirichlet series Lf (s) = c(n)n (convergent,
as already mentioned, for Re s > k) has an Euler product.

Corollary 3: The Dirichlet series Lf of a Hecke eigenform f of weight k has the Euler
product

1
(9.16) Lf (s) = Πp∈P .
1− c(p)p−s + pk−1−2s

Proof: From Corollary 2 we know that n −→ c(n) is a multiplicative function, so with
t := p−s we can write
∞

Lf (s) = Πp∈P ( c(pn )tn ).

n=0

Hence it is sufficient to show that we have

∞

c(pn )tn = Lf,p (t) for Lf,p (t) := (1 − c(p)t + pk−1 t2 )−1 .
n=0

We compute
∞

ψ(t) := ( c(pn )tn )(1 − c(p)t + pk−1 t2 ) =: d(n)tn

n=0

and as coefficient of t get d(1) = c(p) − c(p) = 0 and, for n ≥ 1, as coefficient of tn+1

d(n + 1) = c(pn+1 ) − c(p)c(pn ) + pk−1 c(pn−1 ) = 0

in consequence of the second relation in Corollary 2. I.e., we have ψ(t) = c(1) = 1.

Remark 9.16: The statement that Lf has an Euler product as in Corollary 3 is equiv-
alent to the fact that the Fourier coefficients of f fulfill the relations in Corollary 2.

Remark 9.17: Hecke proved that Lf can be analytically continued to a meromorphic

function on the whole complex plane, resp. to a holomorphic function if f is a cusp form,
and that the completed L-function

L∗f (s) := (2π)−s Γ(s)Lf (s)

satisfies the functional equation

(9.17) L∗f (s) = (−1)k/2 L∗f (k − s).

The proof is based on the classical Mellin transform: Let f be a normalized Hecke cusp
form of weight k, i.e. we have
∞

f (τ ) = c(n)q n with f (−1/τ ) = τ k f (τ ).

n=1
244 9. Epilogue: Outlook to Number Theory

Using the Γ−function

 ∞
Γ(s) := e−t ts−1 dt
0

we determine the Mellin transform of f , namely

 ∞  ∞
dy
f (iy)y s−1 dy = ( c(n)e(niy)) y s
y
0

0
 ∞
= c(n) e−2πny y s−1 dy
0

= c(n)(2πn)−s Γ(s)

= (2π)−s Γ(s)Lf (s) = L∗f (s).

In the integral on the left hand side, we put u = 1/y and use the transformation formula
for f to get
 ∞ 0 
du
f (iy)y s−1
dy = − f (i/u)u−s
u
0
 ∞∞
du
= f (iu)ik uk−s
0 u
= ik L∗f (k − s).

Remark 9.18: The whole theory has been extended to other groups than Γ = SL(2, Z),
in particular to Γ0 (N ) (see for instance [Ge]).
There is also a converse to the statement in Remark 9.17: Every Dirichlet series L,
which fulfills a functional equation of the type just treated and has certain regularity
and growth properties, comes from a modular form of f of weight k. This was first re-
marked by Hecke and then refined considerably by Weil in [We1] and is now cited under
the heading of the famous Converse Theorem.

Remark 9.19: The Hecke operators are hermitian with respect to the Petersson scalar
product for forms f1 , f2 of weight k given by

< f1 , f2 > := f1 (τ )f2 (τ )y k−2 dxdy, F as in Remark 9.2,
F

i.e. one has

< T (n)f1 , f2 > = < f1 , T (n)f2 > .

Arithmetic Modular Forms

Though it does not directly belong to our topic, we can not resist the temptation to
complete this section by indicating a concept of arithmetically distinguished modular
forms, which has been exploited by Shimura and many others (see [Sh]): We look at
modular forms (for the modular group Γ, but this is open to wide generalizations) with
integral Fourier coefficients

Ak (Γ, Z) := {f = c(n)q n ∈ Ak (Γ); c(n) ∈ Z}.

9.4 Hecke’s Theory of L−Functions Associated to Modular Forms 245

One can show that one can find a Z-basis of Ak (Γ, Z), which is also a C-basis of Ak (Γ).
Prototypes of these forms are, up to factors, forms we already met with, namely
∞

F (τ ) := q (1 − q n )24 ,
n=1
∞

E4 (τ ) := 1 + 240 σ3 (n)q n ,
n=1
∞

E6 (τ ) := 1 − 504 σ5 (n)q n .
n=1

For k ∈ 4N Ak (Γ, Z) has as Z-basis (see [Se1] p.105)

E4α F β , α, β ∈ N0 with α + 3β = k/4,

and for k ∈ 2(2N + 1)

E6 E4α F β , α, β ∈ N0 with α + 3β = ((k/2) − 3)/2.

Our analysis of the action of the Hecke operators above shows that Ak (Γ, Z) is stable
under all T (n). Hence the coefficients of the characteristic polynomial of T (n) acting
on Ak (Γ, Z) are integers and one can deduce that the eigenvalues are algebraic integers.
There are explicit formulas for the traces of the T (n), the Eichler-Selberg trace formulas,
starting by [Ei1] and [Sel].

Summary

In this section we looked at L-functions L(s) = Lf (s) coming from modular cusp forms f .
Via a functional equation these functions have analytic continuation to the whole com-
plex plane. This functional equation is a consequence of the covariant transformation
property of the modular form. The modular form is related to an (infinite-dimensional)
automorphic representation of SL(2, R) (or GL(2, R)). We call such L-functions auto-
morphic L-functions. If f is a Hecke eigenform, Lf has an Euler product with factors of
degree 2, i.e. with

(1 − c(p)p−s + pk−1−2s )−1 = (1 − c(p)t + pk−1 t2 )−1 , t := p−s .

There is a general procedure to associate such functions to automorphic representations

of other groups leading to Euler products with other degrees. In particular, one has
a functional equation and factors of degree 1 in the theory of Hecke’s L-series with
grössencharacters (from the German word “Grössencharakter”). This is fairly easy to be
understood ([Ge] p.99f) as a theory of automorphic L-functions for the group GL(1) if
one uses the more functional analytic idelic approach initiated by Tate’s thesis (repro-
duced in [CF] or described in [La2] and, provided with the necessary analytic background,
in [RV]). Hecke’s classic approach is carefully displayed in [Ne] p.515ff. Here we shall
indicate the definition of this function later when we have more number theory at hand.
For an introduction to more general automorphic L-functions we refer to [Ge] and [GeS].
246 9. Epilogue: Outlook to Number Theory

9.5 Elements of Algebraic Number Theory and Hecke

L-Functions
Any book on number theory is good as a background to the following survey, in particu-
lar we recommend the books by Lang [La2] and by Neukirch [Ne] (now also in English).

Number Fields and their Galois Groups

Definition 9.5: An algebraic number field is a finite extension K of the field Q of ra-
tional numbers, i.e. a field containing Q, which is a finite-dimensional vector space over Q.

Such a field is obtained by adjoining to Q roots of polynomials with coefficients in Q.

For example, the field
Q(i) := {a + bi; a, b ∈ Q}
is obtained by adjoining the roots of the polynomial x2 + 1, denoted as usual by i and
−i. We obtain a field, which has dimension 2 as a vector space over Q, the field of Gauss
numbers. This is the most elementary
√ example of an imaginary quadratic field, which
has the general form K = Q( −d), d ∈ N, d squarefree.

Similarly, for N ∈ N, adjoining to Q a primitive N −th root of unity ζN we obtain the

N -th cyclotomic field QN := Q(ζN ). Its dimension as a vector space over Q is ϕ(N ),
the Euler function of N , i.e.

ϕ(N ) := # {m ∈ N; 1 ≤ m < N, (m, N ) = 1}.

We can embed Q(ζN ) into C in such a way that ζN −→ e(1/N ). But this is not the
only possible embedding of QN into C, since we could also send ζN −→ e(m/N ) where
(m, N ) = 1.

Suppose now that K is an algebraic number field and F a finite extension of K, i.e. another
field containing K, which has a finite dimension as a vector space over K. This dimension
is called the degree of F over K and denoted by degK F, in particular deg F for K = Q.

Definition 9.6: The group of all field automorphisms σ of F, preserving the field struc-
ture and such that σ(x) = x for all x ∈ K, is called the Galois group of F over K and
denoted by Gal(F/K), in particular Gal(F) for K = Q.

Example 9.5: The Galois group Gal(QN ) is naturally identified with the group

(Z/N Z)× := {n̄ := n + N Z ∈ Z/N Z; (n, N ) = 1}

with respect to multiplication. The element n̄ ∈ (Z/N Z)× gives rise to the automor-
n m mn
phism of QN sending ζN to ζN , and hence ζN to ζN for all m. If M divides N , then
QM is contained in QN , and the corresponding homomorphism of the Galois groups
Gal(QN ) −→ Gal(QM ) coincides under the above identification with the natural surjec-
tive homomorphism

(Z/N Z)× −→ (Z/M Z)× ; n̄ −→ n mod M.

A central notion from number theory is the notion of a Galois extension:

9.5 Elements of Algebraic Number Theory and Hecke L-Functions 247

Definition 9.7: An algebraic field extension F/K is called a Galois extension iff the
order of the Galois group G = Gal(F/K) is equal to the relative degree degK F.

It is a main part of the theory to give different versions of this definition. Then, Galois
theory consists in establishing for a Galois extension a (contravariant) bijection between
subgroups of the Galois group and subfields of F containing K.

Next, one has class field theory, which tries to collect information about a Galois field
extension F/K, in particular the structure of the Galois group, only from the structure
of the basefield K. We shall need some of this and have to recall some more notions and
facts from algebra:

The Ideal Class Group

Let R be an integral domain, i.e. a commutative ring with unit e and without zero-
divisors. A subring a in R is an ideal in R iff one has ra ∈ a for all r ∈ R and a ∈ a.
An ideal a is a principal ideal iff it can be generated by one element, i.e. is of the form
a = aR for an element a ∈ R. An ideal p in R is a prime ideal iff from ab ∈ p one can
conclude a or b ∈ p, or, equivalently, iff the factor ring R/p has no zero-divisors.

For instance, every normalized irreducible polynomial f of degree n in the polynomial

ring R = Q[x] generates a prime ideal p = f R. This ideal is maximal and the factor ring
R/p is an algebraic number field of degree n. Every algebraic number field of degree n
can be described like this.

If for two ideals we have a ⊃ b, we say a divides b. And (a, b) denotes the greatest
common divisor of a and b (corresponding to the usual notions for integers if R = Z and
a = aZ, b = bZ).

From two ideals a and b one can build new ideals, the sum a + b and the product a · b
(the last one consisting of finite sums of products ab, a ∈ a, b ∈ b).

In number theory, the main occasion to use these notions is the maximal order of an
algebraic number field K, i.e. the ring o of integral elements of K: an element a ∈ K
is called integral iff it is the root of a normalized polynomial f ∈ Z[x] (one has to show
that the set of these elements is a ring). The determination of the maximal orders is
a fundamental task of number theory. For instance, for the field K = Q(i) of Gauss
numbers one has the Gauss integers o = Z[i] = {a + bi; a, b ∈ Z} (this is not difficult, try
to prove it as Exercise 9.7). For the cyclotomic field K = Q(ζN ) one has also the nice
result o = Z[ζN ].

Main tools in this context are the maps norm and trace: For x ∈ F one has
λx : F −→ F, a −→ xa for all a ∈ F,
the (additive) trace map
T rF/K : F −→ K, x −→ Tr λx
and the (multiplicative) norm map
NF/K : F −→ K, x −→ det λx .
248 9. Epilogue: Outlook to Number Theory

An ideal in o is also called an ideal of K and this notion is extended to the notion of a
fractional ideal of K: This is an o-submodule a = 0 of K, which has a common denom-
inator, i.e. there is an element 0 = d ∈ o with da ⊆ o.

All this is done with the intention to associate another finite group to an algebraic num-
ber field: It is not difficult to see that the set JK of fractional ideals provided with
multiplication as composition has the structure of an abelian group. It is called the ideal
group of K. The (almost unique) decomposition of an integer in Z into a product of
powers of primes is extended to the statement (see for instance [Ne] p.23):

Every fractional ideal a ∈ JK has a unique representation as a product of powers of prime

ideals
a = Πp pνp , νp ∈ Z, almost all νp = 0.
The principal fractional ideals PK := {(a) = ao; a ∈ K∗ } form a subgroup of JK and,
finally, one introduces the factor group

ClK := JK /PK .

This comes out as a finite abelian group and is called the ideal class group or simply
class group of K.

Class field theory tries to relate the Galois group of a field extension to the ideal class
group of the ground field. An extension F/K with abelian Galois group G is called a
Hilbert class field iff one has
G(F/K) = ClK .
Example 9.6: The cyclotomic field QN is a Hilbert class field over Q. We have
ClQN  (Z/N Z)× .

Dirichlet Characters and Hecke’s L-Functions

For 0 = m ∈ Z, the classical Dirichlet character mod m is a character of the group

(Z/mZ)×
χ : (Z/mZ)× −→ S 1 .
The character χ is called primitive if there is no genuine divisor m of m such that
χ factorizes via a character of (Z/m Z)× . Given such a χ, we get a (multiplicative)
function for all natural numbers n, which we denote again by χ, by mapping n to χ(n̄)
if (n, m) = 1 and n̄ is the class of n mod m, and to zero if (n, m) = 1. Hence, to such a
primitive character χ, one can associate an L-series

∞
χ(n)
L(χ, s) := .
n=1
ns

Obviously, for the trivial character χ0 mod 1 with χ0 (n) = 1 for all n, we obtain the
Riemann Zeta series. Hecke proved that these series have an Euler product (with degree
one factors)
1
L(χ, s) = Πp∈P ,
1 − χ(p)p−s
analytic continuation to C (resp. C\{1} for χ0 ), and a functional equation for s −→ 1−s.
(The proof relies again on the transformation formula of a (generalized) theta series.)
9.5 Elements of Algebraic Number Theory and Hecke L-Functions 249

These are examples of Hecke’s L-functions, which can be defined more generally: Let K
be an algebraic number field with maximal order o, m ⊂ o an ideal, and J m the group of
fractional ideals a in K, which are prime to m

J m := {a ⊂ K; (a, m) = 1}.

Let χ : J m −→ S 1 be a character. Then we define an L-series

χ(a)
L(χ, s) := .
a⊂o
N (a)s

Here the sum is taken over all ideals a in o, N (a) denotes the number of elements of the
residue class ring o/a, and again we put χ(a) = 0 for (a, m) = 1. The assumption that χ
is a character translates easily to the existence of an Euler product taken over all prime
ideals p in o
1
L(χ, s) := Πp .
1 − χ(p)N (p)−s

If we take χ = χ0 with χ0 (a) = 1 for all a, we get the usual Dedekind Zeta function of
the number field K

L(χ0 , s) = N (a)−s =: ζK (s).

a⊂o

The question asking for the biggest class of characters, for which one can prove a func-
tional equation (and hence has analytic continuation) leads to Hecke’s definition of the
notion “Grössencharakter”:

Definition 9.8: A character χ : J m −→ S 1 is called a grössencharacter mod m iff there

is a pair of characters

χf : (o/m)∗ −→ S 1 , χ∞ : R∗ −→ S 1

such that
χ((a)) = χf (a)χ∞ (a) for all a ∈ o with ((a), m) = 1.

We have remarked (in 5.1) that the set of characters of the additive group R is simply
in bijection to R. A sign that multiplicative structure is more intricate than additive
structure is again given by the fact that the set of these characters χ∞ of R∗ is more
complicated (for a complete statement see [Ne] p.497).

As already said, a more elegant way to understand these constructions is based on the
theory of adèles and idèles introduced into number theory by Tate and then pursued by
many others. Here one has the central notion of a Hecke character as a character of the
idel class group. The relation to the notions presented here is also to be found in [Ne]
p.501. There are further refinements using ray class groups and their characters but we
will stop here and change the topic to get again more concrete examples.
250 9. Epilogue: Outlook to Number Theory

9.6 Arithmetic L-Functions

In classical number theory and in arithmetic geometry there are many occasions to in-
troduce Zeta- and L-functions as Dirichlet series encoding diophantine or arithmetic in-
formation. By their nature as Dirichlet series, these series converge in right half planes.
Then, as the main task, one has to find a way to extend the function given by such a
series to a function on the whole plane, eventually with some poles, and study the values
of this function at special, in particular, integral points. This may turn out to be the
expression of a mysterious regularity property inherent in the arithmetic or diophantine
problem we started with. A way to prove this comes up if this arithmetic L-function can
be seen as an automorphic L-function. For instance the proof of the Fermat conjecture
by Wiles (and Taylor) incorporates a proof that the L-function of an elliptic curve can
be interpreted as an L-function of a modular form. Here we are far from making this
sufficiently precise but to give at least some flavour and define the L-functions for number
fields Emil Artin discussed around 1923, we borrow heavily from an expository article
on the role of representation theory in number theory by Langlands [L].

Two Examples:

We look at diophantine equations, i.e. polynomial equations with integral coefficients to

which integral solutions are sought. A famous example is the Fermat equation

xm + y m = z m .

For m = 2 there are infinitely many solutions, the Pythagorian triples, e.g. (3, 4, 5), and
for m > 2, as Wiles (and Taylor) proved, there are no non-trivial integral solutions. But
if there are no integral solutions to a diophantine problem, one can look for solutions
mod p, p prime, try to count their number and encode these numbers into a function.

Example 9.7: Probably the simplest example is the equation

(9.18) x2 + 1 = 0.

The primes p for which the congruence x2 + 1 ≡ 0 mod p can be solved are 2, 5, 13, . . .
all of which leave the remainder 1 upon division by 4, whereas primes like 7, 19, 23, . . .
that leave the remainder 3 upon division by 4 never do so. In 9.3 we already presented
the Legendre symbol
a
( ) = ±1 for all a ∈ Z, with (p, a) = 1,
p

which indicates whether the congruence x2 ≡ a mod p is solvable or not. Having at

hand the notion of a Dirichlet character, we see that here we have an example with a
multiplicative function χ on Z given by

χ(p) = ( −1
p ) =1 for p ≡ 1 mod 4,
= −1 for p ≡ 3 mod 4,
= 0  1,
for (p, 4) =

i.e. with χ(1) = 1, χ(2) = 0, χ(3) = −1, χ(4) = 0 . . . .

9.6 Arithmetic L-Functions 251

We have
∞
χ(p) −1
χ(n) 1 1 1 1
L(χ, s) = (1 − s
) = s
= 1 − s + s − s + s + ... .
p
p n=1
n 3 5 7 9

Like every Dirichlet series, this series defines a function in a right half plane. It is an
expression of the mysterious symmetry of our problem that one can bring into play a
theta function, which, by its transformation property, produces a function valid for all
s ∈ C.
Now, having prepared some material from algebraic number theory, we can look at the
function L(χ, s) from another point of view: The polynomial f (x) = x2 + 1 defines the
Gauss field
K = Q(i) = {a + bi, a, b ∈ Q}
and this field has the Galois group G = G(K) of automorphisms elementwise fixing Q
consisting
√ just of the identity map and the map ϕ, uniquely fixed by changing the root
θ1 := −1 = i into θ2 := −i. Hence, one has G  {±1} and this group has exactly two
one-dimensional representations, the trivial one and a representation π with π(ϕ) = −1.
One can use this to define another type of L-function, the Artin L-function. To explain
this (following Langlands’ article), we go back some steps:
The solvability of the congruence f (x) = x2 + 1 ≡ 0 mod p is equivalent to the possi-
bility to factorize f mod p into a product of two linear factors, e.g.

x2 + 1 ≡ x2 + 2x + 1 = (x + 1)2 mod 2
x + 1 = x − 4 + 5 ≡ (x − 2)(x + 2)
2 2
mod 5.

For the primes p = 7 and 11 and, in general, for all primes, for which the congruence has
no solutions, the polynomial x2 + 1 stays irreducible mod p.

Example 9.8: Another famous example is

(9.19) f (x) = x5 + 10x3 − 10x2 + 35x − 18.

The polynomial is irreducible mod p for p = 7, 13, 19, 29, 43, 47, 59, . . . and factorizes into
linear factors mod p for p = 2063, 2213, 2953, 3631, . . . .

In general, we can look at a polynomial with integral coefficients

f (x) = xn + an−1 xn−1 + · · · + a1 x + a0

with roots α1 , . . . , αn , which we assume to be pairwise different. There will be various

relations between these roots with coefficients that are rational,

F (α1 , . . . , αn ) = 0.
√ √
For example, the roots of x3 − 1 = 0 are α1 = 1, α2 = (−1 + −3)/2, α3 = (−1 − −3)/2
and two of the valid relations for these roots are

α1 = 1, α2 α3 = α1 .

To the equation f (x) = 0 we can associate the group of all permutations of its roots that
preserve all valid relations. In the last example the sole possibility in addition to the
252 9. Epilogue: Outlook to Number Theory

trivial permutation is the permutation that fixes α1 and interchanges α2 and α3 . The
group G that we get is called the Galois group of the equation f (x) = 0. It is identical
with the Galois group of the algebraic number field K defined by the polynomial f .
We introduce the discriminant ∆ of the equation (resp. the number field) as defined by
the equation

∆ := (αi − αj ).
i=j

It is an integer and one of the most important characteristic numbers to be associated to

a number field. Now, one can attach to any prime p that does not divide ∆ an element
Fp ∈ G, called the Frobenius automorphism, that determines among other things how
the equation factors mod p. More precisely, it is the conjugacy class of Fp within G
that is determined (we will indicate below how this is done). To construct an L-series
we need a prescription associating numerical information to the primes p. We can come
to this by choosing a (finite-dimensional) representation ρ of G and taking the trace or
the determinant of the matrix ρ(Fp ). Things are particularly easy if we have a one-
dimensional representation ρ as for instance for the group G in our first example (9.18)
above: We can define an Euler product

L(ρ, s) := (1 − ρ(Fp )p−s )−1
p

where the primes dividing ∆ are omitted. By closer analysis, one can realize that in the
example we get just the Hecke L-function L(χ, s) if we take for ρ the non-trivial character
χ of G  {±1}. This is another expression of (a special case of) the main theorem of
abelian class field theory.

In Chapter 2 we studied very briefly representations ρ of finite groups. Now, here, we

find an occasion to expand the subject slightly. Beyond one-dimensional representations,
the next possibility is that ρ is a two-dimensional representation, which we may suppose
to be unitary. We know that SU(2) is a double cover of SO(3). It is customary to classify
finite subgroups of the unitary group by their image in the group of rotations. Taking
for an irreducible ρ the finite subgroup to be ρ(G), we obtain dihedral, tetrahedral, octa-
hedral, and icosahedral representations if G has as its image in SU(2) the corresponding
group. Dihedral representations can be treated by the classical theory, and also the
other representations were treated at the time of Langlands’ article with the exception
of a complete treatment of the icosahedral case. An example of an equation with an
icosahedral representation ρ is given by our equation (9.19). For all primes, which do not
divide 800, the conductor (related to the discriminant in a way we do not explain here),
one can form the matrix ρ(Fp ), which has two eigenvalues, λp and µp . Artin’s L-function
is then
1 1
L(ρ, s) := −s 1 − µ p−s
p
1 − λ p p p

the primes dividing the conductor being omitted from the product. This product con-
verges for Re s > 1 and it is a special case of fhe famous Artin Conjecture from 1923 to
show that it can be continued to an analytic function for all s ∈ C. For the equation
(9.19) this was done in 1970 by J. Buhler. The general case is still open.
9.6 Arithmetic L-Functions 253

Decomposition Theory and Artin’s L-Function

To start with, we look again at the example K = QN = Q(ζN ): Here we have the abelian
Galois group G = Gal(QN )  (Z/N Z)× . This isomorphism relates every irreducible
representation ρ of G uniquely to a character χ of (Z/N Z)× . To each prime number p,
which does not divide N , we associate the class p̄ := p mod N in (Z/N Z)× and then
p
the Frobenius automorphism Fp ∈ G given by ζN −→ ζN . Now we can define the Artin
L-function
1
L(ρ, s) :=
1 − ρ(Fp )p−s
p
N

and it is clear that it coincides with the Hecke L-function

1
L(χ, s) := .
1 − χ(p)p−s
p
N

For the general case, we look at an algebraic field extension F/K with Galois group
G = Gal(F/K) and rings of integral elements O ⊃ o. For a prime ideal P in O and a
prime ideal p in o we say that P lies over p if P contains p and hence P ∩ o = p. We
say that p decomposes into the primes Pi if p generates in O an ideal (p), which has the
decomposition
h
(p) = Pei i
i=1

into the primes Pi lying over p with the ramification indices ei ∈ N. We denote by k(P)
the residue field O/P and similarly k(p) := o/p. Both fields are finite fields. If Pi lies
over p, k(Pi ) is a finite extension of k(p) and we denote fi := degk(p) k(Pi ). These degrees
and the ramification indices are related by the central formula of Hilbert’s ramification
theory

h
ei fi = n.
i=1

From now on, we assume that the extension F/K is a Galois extension. Then the
decomposition simplifies to

h
(p) = ( Pi )e
i=1

and all degrees fi coincide (=: f ). The Galois group G acts on O and moreover even
transitively on the primes Pi lying over a given p ⊂ o. We introduce the decomposition
group
GP := {σ ∈ G; σP = P}.

The index of the decomposition group in G is equal to the ramification index, which is
defined as e = [G : GP ]. One has the two extreme cases:

i) If GP contains only the identity, we say that p is totally decomposed, and

ii) GP = G where p stays indecomposed.
254 9. Epilogue: Outlook to Number Theory

Every σ ∈ GP induces an automorphism of the residue class field k(P) := O/P by

a mod P −→ σa mod P

Since this automorphism preserves the elements of the subfield k(p), we get a homomor-
phism
ϕP : GP −→ Gal(k(P)/k(p)),
which is surjective. We define the inertia group as its kernel, IP := ker ϕP . If p is
unramified, IP consists only of the identity, one has GP  Gal(k(P)/k(p)) and, hence, this
group may be seen as a subgroup of G. In this case we can find exacly one automorphism
FP ∈ G such that
FP a ≡ aq mod P for all a ∈ O
with q := N (p). This is the Frobenius automorphism, it generates the cyclic group GP .
If IP is not trivial, we take as Frobenius FP an element of G such that its image in
Gal(k(P)/k(p)) has again the action a mod P −→ aq mod P.

Now we are ready to define Artin’s L-function for this general situation. We take a
representation ρ of G in the finite-dimensional vector space V and denote by V P the
subspace of elements fixed by ρ(IP ). The characteristic polynomial

det(E − ρ(FP )t; V P )

for the action of ρ(FP ) on V P depends only on the prime p and not on the prime P lying
over p.

Definition 9.9: Let F/K be a Galois extension of algebraic number fields and (ρ, V ) a
representation of its Galois group G. Then Artin’s L-series for ρ is defined by
1
L(ρ, s) := L(F/K, ρ, s) :=
p
det(1 − ρ(FP )N (p)−s ; V IP )

where the product is taken over all prime ideals p in K (i.e. in o).

This series converges in the right half plane Re s > 1 to an analytic function. For K = Q
and the trivial one-dimensional representation ρ0 with ρ0 (σ) = 1 for all σ ∈ G, Artin’s
L-function is just the Dedekind Zeta-function, one has

L(F/Q, ρ0 , s) = ζF (s).

In [Ne] Kapitel VII, we find a lot of results concerning the Artin L-function. To finish
with this topic, we only mention Theorem (10.6) in [Ne], clarifying the relation between
Artin’s and Hecke’s L-function in the abelian case.

Theorem 9.12: Let F/K be an abelian extension, f the conductor of F/K, ρ a nontrivial
character of Gal(F/K) and χ the associated grössencharacter mod f. Then we have the
relation
1
L(F/K, ρ, s) = L(χ, s), S := {p | f, χ(IP ) = 1}.
1 − ρ(FP )N (p)−s
p∈S

Here we should still explain the notion of a conductor, but again we have to refer to [Ne].
9.6 Arithmetic L-Functions 255

L-Functions of Elliptic Curves

Again we follow Langlands’ article [L] and look at another diophantine problem: We ask
for integral or at least rational solutions of the equation

(9.20) y 2 = x3 + Dx,

with D ∈ Z, D = 0, say D = −1. This is a special case of an affine equation defining an

elliptic curve E = ED . More generally, one looks at a smooth projective curve defined
by the homogeneous equation

y 2 z = x3 + axz 2 + bz 3 , a, b ∈ Q, 4a3 + 27b2 = 0.

A curve like this is distinguished among all smooth curves because it can be given the
structure of an abelian group. Elliptic curves now are very popular even outside of
number theory because of their importance for cryptography. On the way to define an
L-function via Euler factors for the primes p, it seems a good idea to count again the
number Np of solutions of (9.20) mod p (or of the associated homogeneous equation, but
we stay here with Langlands in his article). For instance, for D = −1 and p = 5, we get
N5 = 7. We define αp and βp by the conditions

Np = p + αp + βp ; βp = ᾱp ; αp βp = p.
√
(These conditions will define αp and βp only if |Np − p| ≤ 2 p, but it can be proven that
this is true.) Then one should be tempted to define as L-function for the elliptic curve
1 1
L(ED , s) =
p
1 − αp p 1 − βp p−s
−s

omitting perhaps some primes p related to D.

For D = (6577)2 , one omits p = 2 and 3 and gets a function for any s and not only for
Re s > 3/2 where the Euler product converges.

At this point it is to be emphasized that the value of an elliptic L-function at s = 1 has a

particular significance: It is a (special case of a) conjecture of Birch and Swinnerton-Dyer
that the equation (9.20) must have a solution if L(ED , 1) = 0. This conjecture from 1965
is based on numerical calculations and in general is still open.

The L-functions of elliptic curves are special cases of the Hasse-Weil Zeta functions asso-
ciated to projective varieties over number fields. Starting with work by Hasse, Weil, and
Dwork, and up to work by Deligne, Langlands and many others, there are deep theorems
and conjectures for these relating them again to automorphic L-functions when the vari-
eties are Shimura varieties, i.e. come as homogeneous spaces from a linear group, as for
instance (roughly said) the completion of the modular curve Γ \ H = Γ \ SL(2, R)/SO(2).
There was a famous conjecture, connected mainly with the names of Taniyama, Shimura,
and A. Weil, indicating (again roughly) that each elliptic L-function is an automorphic
L-function belonging to some modular form, which has been proved by Wiles and Tay-
lor in the framework of their proof of the Fermat conjecture already mentioned at the
beginning of this section.
256 9. Epilogue: Outlook to Number Theory

9.7 Summary and Final Reflections

There are still much more L- and Zeta functions than we mentioned up to now (Dwork
once remarked to the author that he had the impression that every mathematician feels
obliged to define his own one). As in this text we restricted our treatment to the dis-
cussion of real and complex groups, we have to leave aside the true story, which happens
using the groups G(Qp ) defined over Qp , the non-archimedian completions of Q, and the
adelic groups G(A), which come up as the restricted direct product of the groups over all
archimedian and non-archimedian completions of Q and/or as groups of matrices with
adelic entries.

Hence we finish our story by giving a rudimentary survey of what Langlands says in [L]
concerning the archimedian case, in particular as several notions, which we treated in our
chapters 6 and 7, reappear in a language using intuition from elementary particle physics.

On the Way to (Archimedian) Langlands Parameters

We look at representations π of a subgroup G of GL(n, R), in particular G = GL(n, R).

There are essentially two ways to analyse or construct representations:
i) the infinitesimal method using the Lie algebra of G, as exploited in Chapter 6,
and
ii) the method, which came up several times in Chapter 4, 7, and 8 and which starts
with the action of G on a manifold M , passes to an associated action on functions living
on the manifold, and finally decomposes this action into irreducibles.

If G is compact, as for instance G = SO(n), one takes M = G and decomposes the

right-regular representation on H = L2 (G). This way, one comes to representations π
with square integrable matrix coefficients

< π(g)v, w >, v, w ∈ H,

and discrete parameters characterizing π.

If G is not compact, the situation changes considerably as we saw in our discussion of the
example SL(2, R) in 7.2. Langlands proposes that the system being treated is best com-
pared to a Schrödinger equation for which both asymptotically independent and bound
states appear (for instance, in the case of the hydrogen atom, for the electron one has
discrete orbits and a continuous spectrum):

Since every element g ∈ GL(n, R) can be written as a product k1 ak2 where k1 and k2 are
orthogonal matrices and a := D(α1 , . . . , αn ) a diagonal matrix with positive eigenvalues
αj , the simplest representations should have n freely assignable parameters, and they are
analogous to a system of n interacting but asymptotically independent particles to which
arbitrary momenta can be assigned. In addition, the presence of the factors k1 and k2
may entail the presence of discrete quantum numbers. We exemplified this in 7.2.1 while
constructing the principal series representations of SL(2, R) where we had the purely
imaginary “momentum” is and the discrete parameter  ∈ {0, 1}. In the GL(n, R) case,
the general induction procedure from 7.1 proposes to look at first at the group B of
9.7 Summary and Final Reflections 257

superdiagonal matrices
⎛ ⎞
α1 ∗ ∗ ... ∗
⎜ α2 ∗ ... ∗ ⎟
⎜ ⎟
⎜ α3 ... ∗ ⎟
(9.21) b = ⎜
⎜
⎟.
⎟
⎜ ... ⎟
⎝ ⎠
αn

Given n real parameters s1 , . . . , sn and n numbers k = ±1 (the supplementary discrete

quantum numbers), one introduces the characters

χk : α −→ sgn(α)
k |α|isk

of R∗ , as well as the character χ of B that sends the matrix b to

n
χk (αk )|αk |δ−k .
k=1

Here we have δ = (n + 1)/2 and the supplementary exponent comes from the modular
function and makes things unitary: Similar to our treatment in the special case in 7.2 we
associate to χ the induced representation indGB χ given by right translation on the space
of functions φ on G that satisfy the functional equation

φ(bg) = χ(b)φ(g) for all b ∈ B, g ∈ G.

The representations associated to two sequences of parameters χk are equivalent iff one

sequence is a permutation of the other.

If one replaces the exponents isk by arbitrary complex numbers, one gets representations,
which in general will be neither irreducible nor unitary. But there is a well-determined
process for choosing a specific irreducible factor of the representation that is then taken
as the representation associated to the characters χ1 , . . . , χn . In the intuitive language,
this means that one allows complex momenta.

In our study of SL(2, R) in 7.2 we met with the discrete series representations. Keeping
this in mind, one can deduce that for GL(2, R) there are representations associated to
one continuous parameter (or, if the analogy is pursued, one momentum) and one dis-
crete parameter (or quantum number).

In the treatment of GL(n, R) the case n = 2 plays a special role because there is just
one algebraic extension of the field R, namely C, and this is of degree 2 over R.

For the group GL(n, C) of complex matrices, the representation theory is simpler. One
has no discrete series and therefore no bound states. In Langlands’ language, for K = R
or C the general irreducible representation of GL(n, K) is analogous to r interacting but
asymptotically bound states with n1 , . . . , nr particles, Σnk = n and the nk being subject
to constraints appropriate to the field. To construct the representation, one introduces
the group P of matrices of the form (9.21) where n is replaced by r, each αk is a square
nk –matrix, and the asterisks represent block matrices of the appropriate size. From
here, we can again apply the induction process but we have to replace the character χ by
258 9. Epilogue: Outlook to Number Theory

the tensor product of (in general infinite-dimensional) representations χk of GL(n, K),

so that the functions φ with their functional equation as above take their values in an
infinite-dimensional space. If the momenta are all real, then the representation is irre-
ducible. Otherwise it is again necessary to pass to a specific factor. As before, the order
of the χk is irrelevant.
After some more consideration, finally one can conclude that the classification of irre-
ducible representations of GL(n, K) has an extreme formal simplicity. For K = C, to
specify representations we specify the n-dimensional representations of GL(1, C) = C× .
As Langlands emphasizes, here we compare two very different objects: on one hand, irre-
ducible and in general infinite-dimensional representations of the group GL(n, C), and on
the other, finite-dimensional but in general reducible representations of GL(1, C) = C× .

A similar statement for K = R requires a group that is not commutative but whose
irreducible representations are of degree at most two, in order to accomodate the existence
of two particle bound states. The appropriate group is the Weil group of R, WR . It is
obtained by adjoining to the group C× an element w such that

w2 = −1, and wz = z̄w for all z ∈ C× .

This is a kind of dihedral group and, in our text, it appeared as an example already in
the Chapters 0 and 1. It has as non-trivial irreducible representations the character π0
given by
z −→ z z̄, w −→ −1
and the two-dimensional representations πm , for 0 = m ∈ Z given by
 m   
z 0 0 1
z −→ , w −→ .
0 z̄ m (−1)m 0

The Weil group WC of the complex number field C is just C× . One has the nice result
that for K = R and C the irreducible representations of GL(n, K) are parametrized by
n-dimensional representations of the Weil group WK . The Weil group can be attached
to many of the fields appearing in number theory. It can be rather complicated because,
as Langlands assures, it incorporates many of the deepest facts about the theory of equa-
tions known at the present.

To get some more flavour, we add a remark concerning orthogonal subgroups G of

GL(2n + 1, R) defined as stability groups of quadratic forms. As we know, there are
several types of groups depending on the index of the quadratic form and the represen-
tation theory of these will be quite different. But one can get a unified point of view
by a parametrization of the representations via homomorphims of the Weil group WR
into the L-group L G of G. The general definition of the L-group is given in terms of the
theory of equations and of root systems. For G = GL(n, R) it is simply L G = GL(n, C)
and for G = SO(2n + 1) it is the symplectic group of 2n × 2n-matrices, L G = Sp(n, R).

The principle that the irreducible representations of the real group G are classified by
the (continuous) homomorphisms of WR into L G is valid, but this has consequences. A
reducible n-dimensional representation of WR is a homomorphism of WR into GL(n, C)
that is isomorphic to C× and not in the center of GL(n, C). Hence the notion of irre-
ducibility has an obvious analogue for the homomorphisms of WR into L G, and it can
be verified that the homomorphisms which are irreducible in this sense correspond to
9.7 Summary and Final Reflections 259

representations whose matrix coefficients are square-integrable over the group. If the
quadratic form chosen is Euclidean, then G is compact and all representations have this
property, so that any homomorphism of WR into L G that is not irreducible in this sense
will have to be excluded from the list of parameters. A similar but less restrictive con-
dition must also be imposed for the groups with other indices.
It also turns out that the classification provided by WR and L G is coarse. Some homo-
morphisms correspond to several irreducible representations of G. This finally leads to
the notion of L-packets which indicates representations with the same L-functions.

L-Groups and Functoriality

For Langlands the introduction of the L-group is an essential step towards his famous
principle of functoriality. A homomorphism from a group H to a group G does not pro-
vide any way to transfer irreducible representations from one of these groups to another,
unless G is abelian, nor are homorphisms between H and G usually related to homomor-
phisms between L H and L G. On the other hand, a homomorphism from L H to L G, by
composition,
WR −→ L H −→ L G
yields a map from the set of parameters for the irreducible representations of H to the
irreducible representations of G, and thus implicitly a way to pass from irreducible rep-
resentations of H to irreducible representations of G. Langlands calls this passage the
principle of functoriality in the L-group. A possibility to realize this principle in general
would have deepest consequences.

The Standard L-Function

In 9.6 we already insisted on the possibility to associate to a given problem or object

an L-function via an Euler product by encoding information about the object in such
a way that one gets a reasonable “Euler polynomial” in the variable t = p−s , which
can be put in the denominator for an Euler factor. While constructing representations,
say of GL(n, K), via induction in Chapter 7, the main idea was that at least some
representations are fixed by choosing characters for the group H of diagonal matrices
D(α1 , . . . , αn ), which can be understood as distinguishing matrices

D(αis1 , . . . , αisn ), s1 , . . . , sn ∈ R.

Even if we still avoid to go into the representation theory of the non-archimedian field
Qp , it should not be too hard to accept that the analogon of the matrix above in the
p-adic theory is the matrix

A(πp ) := D(pis1 , . . . , pisn ), s1 , . . . , sn ∈ R.

To this matrix one can attach the function

1
L(πp , s) :=
det(En − A(πp )p−s )

and then form the Euler product

L(π, s) := L(πp , s).
260 9. Epilogue: Outlook to Number Theory

In this product perhaps some “special” primes are left out or are represented by special
factors and one has to find an appropriate “factor at ∞” associated to the archimedian
representation. Such a product has the chance to converge for, say, Re s ≥ a + 1, pro-
vided that the eigenvalues of each A(πp ) are less than pa in absolute value. The outcome
is called a standard L-function. Whoever wants to see that this really happens (and
even more...) should be motivated to take a look into the non-archimedian theory, as for
instance in [Ge] or [GeS].

L-Groups and Automorphic L-Functions

To finish our fantastic and magical story (the German word Märchen appears well in the
title of one of Langlands’ articles ([L1])), we look at an (automorphic) representation π
of a group G (in particular G = GL(n)), take a finite-dimensional representation ρ of
the L-group L G belonging to G. Then we have again a finite Euler polynomial given by
det(E − ρ(A(πp )t) and one can introduce as automorphic L-function
1
L(π, ρ, s) :=
det(E − ρ(A(πp ))p−s )

where again for some primes p we have to put in special factors, which we do not discuss
here. In particular, one has to find a completing factor belonging to the archimedian
theory. The final goal is to generalize what we sketched in the last section by showing
that to given π and ρ there is an automorphic representation Π such that

A(Πp ) = ρ(πp )

for almost all p. Then one could have

L(π, ρ, s) = L(Π, s)

and use properties of one type of functions to be transfered to the other.

There are many other important topics coming up in this context. We only mention the
topic of the trace formula but can not touch this here. The only theme whose formulation
is within our reach is the following: Given a representation π of a group G = GL(n) and
a representation π  of G = GL(n ), by the methods from of our text one can construct
representations π + π  of GL(n + n ) and π ⊗ π  of GL(nn ). It is natural to ask for a
similar construction for automorphic representations. This is related to functoriality and
up to now there are very few results.

We repeat our hope that the vagueness of the presentation of some of the material in
the last sections would make the reader curious to know more about all this and to be
stimulated to more intense studies of this most fascinating part of mathematics. All the
time, we were driven by J. R. Jiménez’ words:

¡Voz mı́a, canta, canta;

que mientras haya algo
que no hayas dicho tú,
tú nada has dicho!
Bibliography

[AM] Abraham, R., Marsden, J.E.: Foundations of Mechanics. Benjamin/Cummings,

Reading 1978.

[BR] Barut, A.O., Raczka, R.: Theory of Group Representations and Applications.
PWN Polish Scientific Publishers Warszawa 1980.

[Ba] Bargmann, V.: Irreducible Unitary Representations of the Lorentz Group. An-
nals of Math. 48 (1947) 568-640.

[Be] Berndt, R.: Einführung in die Symplektische Geometrie. Vieweg, Braun-

schweig/Wiesbaden 1998.
Now also translated: An Introduction to Symplectic Geometry. GSM 26, AMS
2001.

[Be1] Berndt, R.: The Heat Equation and Representations of the Jacobi Group. Con-
temporary Mathematics 389 (2006) 47 – 68.

[BeS] Berndt, R. Schmidt, R.: Elements of the Representation Theory of the Jacobi
Group. PM 163, Birkhäuser, Basel 1998.

[BR] Brateli, O., Robinson, D.W.: Operator Algebra and Quantum Statistical Me-
chanics. Springer, New York 1979.

[BtD] Bröcker, T., tom Dieck, T.: Representations of Compact Lie Groups. Springer,
New York 1985.

[Bu] Bump, D.: Automorphic Forms and Representations. Cambridge University

Press, Cambridge 1997.

[CF] Cassels, J.W.S., Fröhlich, A.: Algebraic Number Theory. Thompson, Washing-
ton, D.C. 1967.

[CN] Conway, J.H., Norton, S.P.: Monstrous Moonshine. Bull. London Math. Soc.
11 (1979) 308 - 339.

[Co] Cornwell, J. F.: Group Theory in Physics. Academic Press, London 1984.

[Do] Donley, R. W.: Irreducible Representations of SL(2, R). p.51 - 59 in: Repre-
sentation Theory and Automorphic Forms (Bailey, T.N., Knapp, A.W., eds.),
PSPM Vol. 61 AMS 1997.

[Du] Duflo, M.: Théorie de Mackey pour les groupes de Lie algébriques. Acta Math.
149 (1982) 153 - 213.
262 BIBLIOGRAPHY

[Ei] Eichler, M.: Einführung in die Theorie der algebraischen Zahlen und Funktio-
nen. Birkhäuser, Basel 1963.
[Ei1] Eichler, M.: Einige Anwendungen der Spurformel im Bereich der Modularkor-
respondenzen. Math. Ann. 168 (1967) 128 - 137.
[EZ] Eichler, M., Zagier, D.: The Theory of Jacobi Forms. Birkhäuser, Boston 1985.
[Fi] Fischer, G.: Lineare Algebra. Vieweg, Wiesbaden 2005.
[Fl] Flicker. Y. Z.: Automorphic Forms and Shimura Varieties of PGSp(2). World
Scientific, New Jersey 2005.
[Fog] Fogarty,J.: Invariant Theory. Benjamin,New York 1969.
[Fo] Forster, O.: Analysis 1. Vieweg, Wiesbaden 2006.
[FB] Freitag, E., Busam, R.: Funktionentheorie. Springer, Berlin 1993.
[Fr] Frenkel, E.: Lectures on the Langlands Program and Conformal Field Theory.
arXiv:hep-th/0512172v1 15 Dec 2005.
[FLM] Frenkel, E., Lepowski, J., Meurman, A.: Vertex Operator Algebras and the
Monster. Academic Press, Boston 1988.
[FS] Fuchs, J., Schweigert, Chr.: Symmetries, Lie Algebras and Representations.
Cambridge University Press 1997.
[FH] Fulton, W., Harris, J.: Representation Theory. GTM 129, Springer, New York
1991.
[Ge] Gelbart, S.: Automorphic Forms on Adele Groups. Annals of Math. Studies 83,
Princeton University Press 1975.
[GeS] Gelbart, S., Shahidi, F.: Analytic Properties of Automorphic L-Functions. Aca-
demic Press, Boston 1988.
[Gel] Gell-Mann, M.: Phys. Rev. 125 1067-1084, 1962.
[GH] Griffiths, P., Harris, H.: Principles of Algebraic Geometry. Wiley, New York
1978.
[GGP] Gelfand, I., Graev, M., Pyatetskii-Shapiro, I.: Representation Theory and Au-
tomorphic Functions. W.B. Saunders, Philadelphia 1963.
[Go] Goldstein, H.: Classical Mechanics. Addison-Wesley, Reading 1980.
[GS] Guillemin, V., Sternberg, S.: Symplectic Techniques in Physics. Cambridge
University Press 1984.
[GS1] Guillemin, V., Sternberg, S.: Geometric Asymptotics. Math. Surv. and Mono-
graphs 14, AMS 1990.
[Ha] Halmos, P.: Measure Theory. Van Nostrand, New York 1950.
[He] Hein, W.: Struktur- und Darstellungstheorie der klassischen Gruppen. Springer
HT Berlin 1990.
BIBLIOGRAPHY 263

[HN] Hilgert, J., Neeb, K.-H.: Lie-Gruppen und Lie-Algebren. Vieweg, Braun-
schweig/Wiesbaden 1991.
[HR] Hewitt, E., Ross, K.A.: Abstract Harmonic Analysis I. Springer, Berlin 1963.
[Hu] Humphreys, J.E.: Introduction to Lie Algebras and Representation Theory.
GTM 9, Springer New York 1972.
[Ig] Igusa, J.: Theta Functions. Springer, Berlin 1972.
[JL] Jacquet, H., Langlands, R.P.: Automorphic Forms on GL(2). LNM 114,
Springer, New York 1970.
[Ja] Jacobson, N.: Lie Algebras. Interscience, New York 1962.
[Ki] Kirillov, A.A.: Elements of the Theory of Representations. Springer, Berlin
1976.
[Ki1] Kirillov, A.A.: Lectures on the Orbit Method. GSM 64, AMS 2004.
[Ki2] Kirillov, A.A.: Unitary Representations of Nilpotent Lie Groups. Uspekhi Mat.
Nauk 17 (1962), 57 - 110; English transl. in Russian Math. Surveys 17 (1962).
[Ki3] Kirillov, A.A.: Merits and Demerits of the Orbit Method. Bulletin of the AMS
36 (1999) 433 - 488.
[Ko] Koecher, M.: Lineare Algebra. Springer, Berlin 1983.
[Kos] Kostant, B.: Quantization and Unitary Representations. In Lectures in Modern
Analysis III. (ed. Taam, C.T.) LNM 170, Springer, Berlin 1970.
[Kn] Knapp. A.W.: Representation Theory of Semisimple Groups. An Overview
Based on Examples. Princeton University Press 1986.
[Kn1] Knapp. A.W.: Structure Theory of Semisimple Lie Groups. p.1 - 27 in: Repre-
sentation Theory and Automorphic Forms (Bailey, T.N., Knapp, A.W., eds.),
PSPM Vol. 61, AMS 1997.
[Kn2] Knapp. A.W.: Lie Groups Beyond an Introduction. PM 140, Birkhäuser, Boston
1996.
[KT] Knapp, A.W., Trapa, P.E.: Representations of Semisimple Lie Groups. p. 7
- 87 in: Representation Theory of Lie Groups (Adams, J., Vogan, D., eds.),
IAS/Park City Math. Series 8, AMS 2000.
[Ku] Kubota, T.: Elementary Theory of Eisenstein Series. Halsted Press, New York
1973.
[La] Lang, S.: Algebra. Addison-Wesley, Reading, Mass. 1965.
[La1] Lang, S.: SL(2,R). Springer, New York 1985.
[La2] Lang, S.: Algebraic Number Theory. Addison - Wesley, Reading, Mass. 1970.
[L] Langlands, R.P.: Representation Theory: Its Rise and Its Role in Number The-
ory. p. 181-210 in Proceedings of the Gibbs Symposium, Yale University 1989,
AMS 1990.
264 BIBLIOGRAPHY

[L1] Langlands, R.P.: Automorphic Representations, Shimura Varieties, and Mo-

tives. Ein Märchen. p. 205-246 in PSPM Vol 33, Part 2, AMS 1979.

[LV] Lion, G., Vergne, M.: The Weil representation, Maslov index and Theta series.
Birkhäuser, Boston 1980.

[LZ] Lewis, J., Zagier, D.: Period functions for Maass wave forms I. Ann. Math.
153 (2001) 191-253.

[Ma] Mackey, G.W.: Unitary Group Representations in Physics, Probability, and

Number Theory. Benjamin/Cummings Publishing Co., Reading, Mass. 1978.

[Ma1] Mackey, G.W.: Induced Representations of Locally Compact Groups I. Ann. of

Math. 55 (1952) 101-139.

[Mu] Mumford, D.: Tata Lectures on Theta I,II,III. PM 28, 43, 97, Birkhäuser, Boston
1983, 1984, 1991.

[Na] Naimark, M.A.: Linear Representations of the Lorentz Group. Pergamon Press,
London 1964.

[Ne] Neukirch, J.: Algebraische Zahlentheorie. Springer, Berlin 2002.

[RV] Ramakrishnan, D., Valenza, R.J.: Fourier Analysis on Number Fields. GTM
186, Springer, New York 1999.

[Re] Renouard, P.: Variétés symplectiques et quantification. Thèse. Orsay 1969.

[Sch] Schöneberg, B.: Elliptic Modular Functions. Springer, Berlin 1974.

[Se] Serre, J.P.: Linear Representations of Finite Groups. Springer, New York 1977.

[Se1] Serre, J.P.: A Course in Arithmetics. GTM 7, Springer, New York 1973.

[Sel] Selberg, A.: Harmonic analysis and discontinuous groups in weakly symmetric
riemannian spaces with application to Dirichlet series. J. Indian Math. Soc. 20
(1956) 47 - 87.

[Sh] Shimura, G.: Introduction to the Arithmetic Theory of Automorphic Functions.

Iwanami Shoten and Princeton University Press 1971.

[To] Torasso, P.: Méthode des orbites de Kirillov–Duflo et représentations minimales

des groupes simples sur un corps local de caractéristique nulle. Duke Math. J.
90 (1997) 261 – 377.

[vD] van Dijk, G.: The irreducible Unitary Representations of SL(2, R). In: Rep-
resentations of Locally Compact Groups with Applications. ed. Koornwinder,
T.H. Mathematisch Centrum Amsterdam 1979.

[Ve] Vergne, M.: Geometric Quantization and Equivariant Cohomology. First Eu-
ropean Congress of Mathematics, Vol I, p.249–298. PM 119, Birkhäuser, Basel
1994.

[Ve1] Vergne, M.: Quantification géométrique et réduction symplectique. Séminaire

Bourbaki 888 (2001).
BIBLIOGRAPHY 265

[Vo] Vogan, D.A.: The Method of Coadjoint Orbits for Real Reductive Groups. p.179
– 238 in: Representation Theory of Lie Groups (Adams, J., Vogan, D., eds.),
IAS/Park City Math. Series 8 AMS 2000.
[Vo1] Vogan, D.A.: Associated Varieties and Unipotent Representations. p.315 – 388
in: Harmonic Analysis on Reductive Lie Groups (Barker, W., Sally, P., eds.),
Birkhäuser, Boston 1991.
[Vo2] Vogan, D.A.: Cohomology and group representations. PSPM Vol 61, 219–243
(1997).
[Wa] Warner, G.: Harmonic Analysis on Semi-Simple Lie Groups. Springer, Berlin
1972.
[We] Weil, A.: Variétés Kählériennes. Hermann, Paris 1957.
[We1] Weil, A.: Über die Bestimmung Dirichletscher Reihen durch ihre Funktionalglei-
chung. Math, Ann. 168 149-156 (1967).
[Wo] Woodhouse, N.: Geometric Quantization. (Second Edition) Oxford University
Press 1991.
[Ya] Yang, Y.-H.: The Method of Orbits for Real Lie Groups. Kyungpook Math. J.
42 (2002) 199–272.
Index

algebraic number field, 246 of Heis(R), 83

analytic subgroup, 89 of a group, 5, 83
arithmetic modular form, 244 of a Lie algebra, 87
Artin L-function, 252, 254 centralizer, 5
Artin conjecture, 252 character
automorphic L-function, 245 of a representation, 21
automorphic factor, 136 of an abelian group, 59
for SL(2, R), 221 coadjoint orbit, 186
for the Heisenberg group, 217 cocycle condition, 122
for the Jacobi group, 238 complementary series
automorphic form, 217, 232 of SL(2, C), 143
of SL(2, R), 142
Borel section, 121 complex structure, 185
Borel space, 38 configuration space, 173
Borel subgroup, 130 conjugacy class, 4
bundle, 162 conjugation, 4
cotangent bundle, 170 connection on a vector bundle, 185
equivalent, 167 continuous decomposition, 60
fiber, 167 continuous sum, 62
Hilbert bundle, 163 correspondence, 240
holomorphic line bundle over P1 (C), cusp form, 222
171, 183 cyclotomic field, 246
line bundle, 161
prequantum line bundle, 179 Darboux theorem, 179
real analytic line bundles over P1 (C), decomposition group, 253
182 Dedekind Zeta function, 249
section of a bundle, 162 degree of a field extension, 246
tangent bundle, 169 delta function, 223
transition function, 168 derivation, 65
vector bundle, 167 derived representation, 70
bundle map, 167 differential form, 174
closed, 176
Cartan decomposition of g, 95 exact, 176
Cartan involution, 85 exterior differentiation, 175
Cartan matrix, 92 inner product with a vector field, 176
Cartan subalgebra, 89, 93 of first degree, 170
Casimir element, 101, 135 direct integral, 62
Cauchy Riemann differential equation, 136 Dirichlet character, 248
Cayley transform, 137 Dirichlet series, 239
CDW diagram, 92 discriminant, 252
center disrete series of SL(2, R), 133
INDEX 267

distribution on a manifold, 184 Poincaré group GP (n), 148

duality theorem reductive, 85
for the discrete series representation of simple, 85
GJ , 239 solvable, 86
for the discrete Series of SL(2, R), 230 standard Borel, 2
for the Schrödinger representation of standard unipotent, 2
Heis(R), 219 symmetric group, 5
Dynkin diagram, 92 symmetry group, 178
symplectic, 85
Eisenstein series, 223 theta group, 221
non-holomorphic, 233 topological, 31
elliptic curve, 255 unimodular, 39
eta function, 223 unitary group, 1
Euler angles, 42, 52 Weil group of R, 2, 258
Euler function, 246 Weyl group, 95
Euler product, 239, 245, 249 group action
exponential function for matrices, 67 effective, 3
from the left, 3
factor representation, 36 from the right, 5
fractional ideal, 248 transitive, 3
Frobenius automorphism, 254 group ring, 23
Frobenuis reciprocity, 127
fundamental domain, 222 Haar measure, 39
fundamental functional equation, 117 hadron, 110
half plane model, 136
G-orbit, 3 Hamilton operator, 177
G-space, 33 Hamilton’s equations, 173
Galois extension, 247 Hamiltonian G-space, 187
Galois group, 246 Hamiltonian formalism, 173
Gamma function, 244 Hecke L-function, 249
Gauss integer, 247 Hecke eigenform, 242
Gell-Mann’s eightfold way, 110 Hecke operator, 232, 240
general linear group, 1 Heisenberg algebra, 66
generalized weight space, 97 Heisenberg commutation relations, 70, 72
Grössencharakter, 249 Heisenberg representation, 118
group Highest Weight Theorem, 98, 103
alternating, 6 Hilbert class field, 248
classical, 2 homogeneous space, 4
congruence group, 221 homothety, 18
derived, 85 Hopf map, 195
euclidean group GE (n), 148
Heisenberg group, 2 ideal
Jacobi group, 209 fractional, 248
linear group, 32 of a commutative ring, 247
Lorentz group GL , 145 of a Lie algebra, 86
modular group, 221 ideal class group, 248
nilpotent, 86 induced representation, 119
of exponential type, 68 compact picture, 125
orthogonal group, 1 first realization, 123
Poincaré group GP , 154 induced picture, 125
268 INDEX

noncompact picture, 126 Lorentz

normalized, 120 boost, 146
second realization, 124 rotation, 146
induction Lorentz group GL , 145
holomorphic, 141
parabolic, 130 Maass wave forms, 232
induction in stages, 127 Mackey decomposition, 121
infinitesimal generator, 69 Mackey Theory for semidirect products, 150
Int g, 89 manifold, 165
integral curve, 174 analytic, 165
integral element, 247 diffeomorphic, 166
internal symmetry group, 110 Kähler, 185
intertwining operator, 9 oriented, 166
isospin, 110 quantizable symplectic, 179
isotypic subspace, 74 smooth, 165
Iwasawa decomposition, 69, 96, 130 symplectic, 176, 178
master equation, 122
Jacobi form, 238 maximal order, 247
Jacobi group, 148, 209, 236, 237 maximal torus, 94
Jacobi identity, 64 measure
Jacobi symbol, 237 invariant, 39
Jacobi theta function, 216, 236 projection valued, 62
Jordan-Hölder theorem, 18 quasi-invariant, 121
spectral, 60
Killing form, 88 Mellin transform, 244
Kirillov-Kostant form, 187 Minkowski (pseudo)norm, 154
Minkowski metric, 145
ladder operator, 73 mock discrete series of SL(2, R), 142
Langlands decomposition, 125 modular form, 222
Laplace operator arithmetic, 244
euclidean, 56 modular function ∆, 40
Laplacian on SL(2, R), 135, 229 modular group, 221
left translation, 4 modular invariant, 226
Legendre symbol, 237 moment map, 187
Lie algebra, 64 multiplicity, 9
classical, 89 multiplier system, 55, 125
exceptional, 89
nilpotent, 86 neutron, 110
of SL(2, R), 65, 69 normal subgroup, 4
of SO(3), 66 normalizer
of SU(2), 65 of h in g, 89
of a linear group, 68 of H in G, 4
of the Heisenberg group, 66, 70
radical of a Lie algebra, 87 one-parameter subgroup, 68
semisimple, 87 operator
simple, 87 adjoint, 34
solvable, 86 bounded, 33
linear fractional transformation, 133 continuous, 33
linear group, 32 norm, 33
little group, 151 self-adjoint, 34
INDEX 269

symmetric, 34 continuous, 34
orbit contragredient, 16
admissible, 196 cyclic, 37
coadjoint, 186 decomposable, 16, 36
elliptic, 188 derived, 70
hyperbolic, 188 discrete series, 142
nilpotent, 188 discrete series of SL(2, R), 133
rigged, 197 disjoint, 9
eigenspace, 56
parabolic induction, 130 equivalent, 9
parabolic subgroup, 130 factor representation, 16
partition, 6 Fock representation, 207
Pauli matrices, 53 induced, 119, 164
permutation matrix, 6 irreducible, 8, 36
Peter–Weyl Theorem, 48 lattice, 219
Petersson scalar product, 244 Lie algebra, 66
phase space, 176 linear, 7
Poincaré group, 154 permutation representation , 11
two-dimensional, 127 primary, 20
Poincaré half plane, 133
principal series, 130
Poincaré series, 225
projective representation, 55
Poincaré-Birkhoff-Witt Theorem, 100
regular representation of a compact group,
Poisson algebra, 64
47
Poisson bracket, 64, 177
regular representation of a finite group,
polarization
27
admissible, 196
Schrödinger representation, 13, 118, 152,
algebraic approach, 196
206
geometric approach, 184
Schrödinger-Weil representation, 211
Pontrjagin dual, 59
smooth, 70
prequantum Hilbert space, 180
unitary, 9
prime ideal, 247
representations
principal series of SL(2, C), 143
of SL(2, C), 143
principal series of SL(2, R), 130, 200, 235
principle of functoriality, 259 of SL(2, R), 130
projection operator, 20 of SO(3), 54
projective representation, 55, 145 of SU(2), 48
projective space, 55, 129, 166 of the Poincaré group GP , 157
proton, 110 of the two-dimensional Poincaré group,
127
q-development, 222 unitarily integrable representations of
quantization, 177 sl(2, R), 78
quasi-measure relation, 121 unitarily integrable representations of
su(2), 82
Radon-Nikodym derivative, 121 Riemann Zeta function, 224, 239
Ramanujan function, 223 right translation, 4
representation root, 90
adjoint, 186 positive, 92
admissible, 143 reduced, 92
coadjoint, 186 restricted, 96
completely reducible, 16, 36 simple, 92
270 INDEX

root reflection, 91 Weyl character formula, 103, 104

root space, 90 Weyl denominator, 103
root system, 91 Weyl group, 95
reduced, 91 Weyl operator, 44
Weyl’s dimensionality formula, 107
Schrödinger equation, 177
Schrödinger representation, 118 zeta transform, 235
Schur’s Lemma, 18
finite-dimensional unitary Schur, 19
the unitary Schur, 37
semidirect product, 147
regular, 150
separable Hilbert space, 33
special linear group, 1
spectral measure, 61
spherical harmonics, 56
spin, 55
standard L-function, 260
Stone-von Neumann Theorem, 84
subrepresentation, 8, 36
subrepresentation theorem, 143
symplectic form ω, 178
integral, 179

tangent space, 169

tensor algebra, 99
tensor product of representations, 14
theta function with characteristics, 216
theta group, 221
theta transform, 234
total differential, 175
trace form, 88, 186

unit disc model, 138

unitary (linear) dual, 9
unitary dual
of SL(2, R), 80, 130
of the Heisenberg group, 152
of the Poincaré group GP , 157
universal enveloping algebra, 99

vacuum vector
of Heis(R), 83
vector field, 170, 174
Hamiltonian, 174
vector of lowest weight, 75
Verma module, 101

weight, 74, 97
algebraically integral, 98
Weil group of R, 2, 13, 258