SL2(R), serge lang

Some of the main contributors have been: Gelfand-Naimark and Harish-Chandra, who considered the Lorentz group in the late forties; Gelfand-Naimark, who dealt with the classical complex g

Springer Books on Elementary Mathematics by Serge Lang

MATH! Encounters with High School Students

Serge Lang

With 33 Figures

University Ann Arbor, MI 48109

San Francisco, CA 94132 U.S.A

U.S.A

AMS Subject Classification: 22E46

Library of Congress Cataloging in Publication Data

Lang, Serge

SL 2 (R)

(Graduate texts in mathematics; 105)

Originally published: Reading, Mass.:

Addison-Wesley, 1975

Bibliography: p

Includes index

I Lie groups 2 Representations of groups

I Title II Series

QA387.L35 1985 512'.55 85-14802

K.A Ribet Department of Mathematics University of California

at Berkeley Berkeley, CA 94720 U.S.A

This book was originally published in 1975 © Addison-Wesley Publishing Company, Inc., Reading, Massachusetts

© 1985 by Springer-Verlag New York Inc

Softcover reprint of the hardcover 1 st edition 1985

All rights reserved No part of this book may be translated or reproduced in any form without written permission from Springer-Verlag, 175 Fifth Avenue, New York, New York 10010, U.S.A

9 8 7 6 5 4 3 2 (Corrected second printing, 1998)

ISBN-13: 978-1-4612-9581-5

DOl: 10.1007/978-1-4612-5142-2

e-ISBN-13: 978-1-4612-5142-2

Foreword

Starting with Bargmann's paper on the infinite dimensional representations of

SLiR), the theory of representations of semisimple Lie groups has evolved to

a rather extensive production Some of the main contributors have been: Gelfand-Naimark and Harish-Chandra, who considered the Lorentz group in the late forties; Gelfand-Naimark, who dealt with the classical complex groups, while Harish-Chandra worked out the general real case, especially through the derived representation of the Lie algebra, establishing the Plancherel formula (Gelfand-Graev also contributed to the real case); Car-tan, Gelfand-Naimark, Godement, Harish-Chandra, who developed the theory of spherical functions (Godement gave several Bourbaki seminar reports giving proofs for a number of spectral results not accessible other-wise); Selberg, who took the group modulo a discrete subgroup and obtained the trace formula; Gelfand, Fomin, Pjateckii-Shapiro, and Harish-Chandra, who established connections with automorphic forms; lacquet-Langlands, who pushed through the connection with L-series and Hecke theory This history is so involved and so extensive that I am incompetent to give a really good account, and I refer the reader to bibliographies in the books by Warner, Gelfand-Graev-Pjateckii-Shapiro, and Helgason for further infor-mation A few more historical comments will be made in the appropriate places in the book

It is not easy to get into representation theory, especially for someone interested in number theory, for a number of reasons First, the general theorems on higher dimensional groups require massive doses of Lie theory Second, one needs a good background in standard and not so standard analysis on a fairly broad scale Third, the experts have been writing for each other for so long that the literature is somewhat labyrinthine

I got interested because of the obvious connections with number theory, principally through Langlands' conjecture relating representation theory to elliptic curves [La 2] This is a global conjecture, in the adelic theory I

v

Therefore, as I learned the theory myself it seemed a good idea to write

up SL 2 (R) The topics are as follows:

I We first show how a representation decomposes over the maximal compact subgroup K consisting of all matrices

2 We describe the Iwasawa decomposition G = ANK, from which most

of the structure and theorems on G follow In particular, we obtain tations of G induced by characters of A

represen-3 We discuss in detail the case when the trivial representation of K

occurs This is the theory of spherical functions We need only Haar measure for this, thereby making it much more accessible than in other presentations using Lie theory, structure theory, and differential equations

4 We describe a continuous series of representations, the induced ones, some of which are unitary

5 We discuss the derived representation on the Lie algebra, getting into the infinitesimal theory, and proving the uniqueness of any possible unitariza-tion We also characterize the cases when a unitarization is possible, thereby obtaining the classification of Bargmann Although not needed for the Plancherel formula, it is satisfying to know that any unitary irreducible representation is infinitesimally isomorphic to a subrepresentation of an induced one from a quasicharacter of the diagonal group The derived representation of the Lie algebra on the algebraic space of K-finite vectors plays a crucial role, essentially algebraicizing the situation

6 The various representations are related by the Plancherel inversion formula by Harish-Chandra's method of integrating over conjugacy classes

7 We give a method of Harish-Chandra to unitarize the "discrete series," i.e those representations admitting a highest and lowest weight vector

in the space of K-finite vectors

8 We discuss the structure of the algebra of differential operators, with

special cases of Harish-Chandra's results on SL 2 (R) giving the center of the universal enveloping algebra and the commutator of K At this point, we have enough information on differential equations to get the one fact about spherical functions which we could not prove before, namely that there are no other examples besides those exhibited in Chapter IV

FOREWORD va

The above topics in a sense conclude a first part of the book The second part deals with the case when we take the group modulo a discrete subgroup The classical case is SL 2 (Z) This leads to inversion formulas and spectral decomposition theorems on L2(f\ G), which constitute the remaining chap-ters

I had originally intended to include the Selberg trace formula over the reals, but in the case of non-compact quotient this addition would have been sizable, and the book was already getting big I therefore decided to omit it, hoping to return to the matter at a later date

A good portion of the first part of the book depends only on playing with Haar measure and the Iwasawa decomposition, without infinitesimal con-siderations Even when we use these, we are able to carry out the Plancherel formula and the discussion of the various representations without caring whether we have "all" irreducible unitary representations, or "an" spherical functions (although we prove incidentally that we do) A separate chapter deals with those theorems directly involving partial differential equations via the Casimir operator, and analytical considerations using the regularity theorem for elliptic differential equations The organization of the book is therefore designed for maximal flexibility and minimal a priori knowledge The methods used and the notation are carefully chosen to suggest the approach which works in the higher dimensional case

Since I address this book to those who, like me before I wrote it, don't know anything, I have made considerable efforts to keep it self-contained I reproduce the proofs of a lot of facts from advanced calculus, and also several appendices on various parts of analysis (spectral theorem for bounded and unbounded hermitian operators, elliptic differential equations, etc.) for the convenience of the reader These and my Real Analysis form a sufficient

background

The Faddeev paper on the spectral decomposition of the Laplace tor on the upper half-plane is an exceedingly good introduction to analysis, placing the latter in a nice geometric framework Any good senior under-graduate or first year graduate student should be able to read most of it, and

opera-I have reproduced it (with the addition of many details left out to more expert readers by Faddeev) as Chapter XIV Faddeev's method comes from pertur-bation theory and scattering theory, and as such is interesting for its own sake, as well as to analysts who may know the analytic part and may want to see how it applies in the group theoretic context Kubota's recent book on Eisenstein series (which appeared while the present book was in production) uses a different method (Selberg-Langlands), and assumes most of the details

of functional analysis as known Therefore, neither Kubota's book nor mine makes the other unnecessary

It would have been incoherent to expand the present book to a global context with adeles I hope nevertheless that the reader will be well prepared

Harish-New Haven, Connecticut

September 1974

Serge Lang

Contents

Notation

Chapter I General Results

1 The representation on Cc(G)

2 A criterion for complete reducibility

3 L 2 kernels and operators

4 Plancherel measures

Chapter II Compact Groups

I Decomposition over K for SLiR)

2 Compact groups in general

Chapter HI Induced Representations

Integration on coset spaces

2 Induced representations

3 Associated spherical functions

4 The kernel defining the induced representation

Chapter IV Spherical Functions

Bi-invariance

2 Irreducibility

3 The spherical property

4 Connection with unitary representations

5 Positive definite functions

Chapter V The Spherical Transform

x CONTENTS

2 The Harish transform

3 The Mellin transform

4 The spherical transform

5 Explicit formulas and asymptotic expansions

Chapter VI The Derived Representation on the Lie Algebra

69

74

78

83

2 The derived representation decomposed over K 100

5 Irreducible components of the induced representations 116

6 Classification of all unitary irreducible representations 121

Chapter VII Traces

3 The trace in the induced representation 147

5 Relation between the Harish transforms on A and K 153 Appendix General facts about traces 155 Chapter VIII The Plancherel Formula

Calculus lemma

2 The Harish transforms discontinuities

3 Some lemmas

4 The Plancherel formula

Chapter IX Discrete Series

I Discrete series in L 2( G)

2 Representation in the upper half plane

3 Representation on the disc

4 The lifting of weight m

5 The holomorphic property

Chapter X Partial Differential Operators

The universal enveloping algebra

CONTENTS

Chapter XI The Wei! Representation

1 Some convolutions

2 Generators and relations for SL 2

3 The Wei! representation

Chapter XII Representation on ° L 2(f \ G)

Cusps on the group

2 The Eisenstein series

3 Analytic continuation and functional equation

4 Mellin and zeta transforms

5 Some group theoretic lemmas

6 An expression for TOTrp

7 Analytic continuation of the zeta transform of TOTrp

8 The spectral decomposition

Chapter XIV Spectral Decomposition of the Laplace Operator on r \ ~

XI

.205 209 211

.219 227 232 234

.239 243 245 248 251 253 255 259

I Geometry and differential operators on ~ 266

2 A solution of Irp = s(l - s)cp 272

3 The resolvant of the Laplace operator on ~ for C1 > 1 275

4 Symmetry of the Laplace operator on r\~ 280

5 The Laplace operator on r\~ 284

6 Green's functions and the Whittaker equation 287

7 Decomposition of the resolvant on r\~ for C1 > 3/2 294

s(l - s)

8 The equation -1.);"(y) = 2 1.);(y) on [a, 00) 309

y

9 Eigenfunctions of the Laplacian in L 2(r\~) = H 314

10 The resolvant equations for 0 < C1 < 2 321

11 The kernel of the resolvant for 0 < C1 < 2 328

12 The Eisenstein operator and Eisenstein functions 338

13 The continuous part of the spectrum 346

14 Several cusps 349

xii CONTENTS

Appendix 1 Bounded Hermitian Operators and Schur's Lemma

Continuous functions of operators

2 Projection functions of operators

Appendix 2 Unbounded Operators

I Self-adjoint operators

2 The spectral measure

3 The resolvant formula

Appendix 3 Meromorphic Families of Operators

4 Compactness and regularity on the torus

5 Regularity in Euclidean space

Appendix 5 Weak and Strong Analyticity

.369 377 379

.383 387

.389 395 400 404 407

.411

.415 419 423

.427

Notation

To denote the fact that a function is bounded, we write f = 0(1) If f, g are two functions on a space X and g ;;; 0, we write f = O( g) if there exists a constant C such that If(x)1 ,;;; Cg(x) for all x E X If X = R is the real line,

say, the above relation may hold for x sufficiently large, say x ;;; xo, and then

we express this by writing x ~ 00 Instead of f = O(g), we also use the Vinogradov notation,

f« g

On a topological space X, C(X) is the space of continuous functions If X

is a Coo manifold (nothing worse than open subsets of euclidean space, or something like SL 2 (R), with obvious coordinates, will occur), we let COO(X)

be the space of Coo functions We put a lower index c to indicate compact support Hence Cc(X) and Cc""(X) are the spaces of continuous and Coo

functions with compact support, respectively

By the way, SLz<R) is the group of 2 X 2 real matrices with determinant I

An isomorphism is a morphism (in a category) having an inverse in this category An automorphism is an isomorphism of an object with itself For instance, a continuous linear automorphism of a normed vector space H is a

continuous linear map A: H ~ H for which there exists a continuous linear map B: H ~ H such that AB = BA = I A Coo isomorphism is a Coo

mapping having a CO() inverse

If H is a Banach space, we let En(H) denote the Banach space of continuous linear maps of H into itself If H is a Hilbert space, we let Aut(H)

be the group of unitary automorphisms of H We let GL(H) be the group of continuous linear automorphisms of H with itself

If G' is a subgroup of a group G we let

G'\G

XIII

XIV NOTATION

be the space of right cosets of G' If I' operates on a set ~, we let

be the space of I'-orbits Certain right wingers put their discrete subgroup I'

on the right Gelfand-Graev-Pjateckii-Shapiro and Langlands put it on the left I agree with the latter, and hope to turn the right wingers into left wmgcrs

For the convenience of the reader we also include a summary of objects used frequently throughout the book, with a very brief indication of their respective definitions at the end of the book for quick reference

I General Results

§1 THE REPRESENTATION ON CC<G)

Let G be a locally compact group, always assumed Hausdorff Let H be a

Banach space (which in most of our applications will be a Hilbert space) A representation of G in H is a homomorphism

of open sets for which all functiohals on H are continuous.]

A representation is called bounded if there exists a number C > 0 such that 1'IT(x)1 " C for all x E G If H is a Hilbert space and 'IT(x) is unitary for all x E G, i.e preserves the norm, then the representation 'IT is caned unitary, and is obviously bounded by 1

For a representation, it suffices to verify the continuity condition above

on a dense subset of vectors; in other words:

Let 'IT: G ~ GL(H) be a homomorphism and assume that for a dense set

of v E H the map x ~ 'IT(x)v is continuous Assume that the image of some neighborhood of the unit element e in G under 'IT is bounded in GL(H) Then 'IT is a representation

This is trivially proved by three epsilons Indeed, it suffices to verify the continuity at the unit element Let v E H and select VI close to v such that

2 GENERAL RESULTS [I, § 1]

x H 'IT{x)vJ is continuous We then use the triangle inequality

to prove our assertion

A representation 'IT: G ~ GL(H) is locally bounded, i.e given a compact subset K of G, the set 'IT(K) is bounded in GL(H)

Proof Let K be a compact subset of G For each v E H the set 'IT(K)v is compact, whence bounded By the uniform boundedness theorem (Real Analysis, VIII, §3) it foHows that 'IT(K) is bounded in GL(H)

For the convenience of the reader, we recall briefly the uniform ness theorem

bounded-Let {T;} iEI be a family of bounded operators in a Banach space E, and assume that for each vEE the set {Tjv LEI is bounded Then the family

{ T; LEI is bounded, as a subset of End( E)

Proof Let C n be the set of elements vEE such that

all iE/

Then Cn is closed, and E is the union of the sets Cn' It follows by Baire's theorem that some C n contains an open ball Translating this open baH to the origin yields an open baH B such that the union of the sets T;(B), i E I, is bounded, whence the family {T;};El is bounded, as desired

We let Cc(G) denote the space of continuous functions on G with compact support It is an algebra under convolution, i.e the product is defined by

where dy is a Haar measure on G We shall assume throughout that G is unimodular, meaning that left Haar measure is equal to right Haar measure For any functionf on G we denote by f- the functionj- (x) = j(x-') Then

f j(x) dx = f f(x- I ) dx = J f- (x) dx

Remark When G is not unimodular, then by uniqueness of Haar ure, there is a modular function ~: G ~ R+ which is a continuous

meas-[I, § l] THE REPRESENTATION ON Cc(G) 3 homomorphism into the positive reals, such that

fGf(xa) dx = ~(a) Lf(x) dx

One then has

by an obvious argument It follows that ~(x) dx is right Haar measure The typical non-unimodular group which will concern us, but not until Chapter III, is the group of triangular matrices

or For this chapter, you can forget about the non-unimodular case,

The modular function occurs in a slightly more general context than above Let '1": G ~ G be either an automorphism (group and topological) of

because the expression on the left is a non-trivial invariant positive functional

on C c ( G) We have the obvious composition rule

~('I"(J) = ~('I") ~«J)

In many applications, we have '1"2 = !d, and therefore ~('I") = 1, i.e 'I" is unimodular This occurs in the context of matrices, when for instance 'I" is the transpose

The basic example of a unimodular group is the group of matrices

G = GL,,(R)

The change of variables formula shows that Haar measure on G is equal to

4 GENERAL RESULTS [I, § I J

where d + x is Lebesgue measure on the additive space of n x n matrices The above measure on GLn(R) is therefore both right and left Haar measure Since

where GL 2 +(R) is the group of 2 X 2 matrices with positive determinant, and

R + is the group of positive reals, it follows that left Haar measure on SLiR)

is also right invariant, i.e SLz(R) is unimodular A better proof is to observe that left and right Haar measures differ by a continuous homomorphism of the group into the positive reals, and that SL 2 (R) has no such non-trivial homomorphism (By looking at conjugacy classes of elements and using various decompositions of SL 2 (R) given later in the book, you should be able

to work this out as an exercise.) Later we shan give explicit descriptions of the Haar measure on SL 2 (R) in terms of various choices of coordinates, and hence we do not stop here for a more thorough discussion

We return to an arbitrary locally compact group G Let 17 be a tion of G in H, and let cp E Cc(G) We define what will be an algebra homomorphism

Let aEG and define in this section 'TaCP(x) = cp(a-Ix) Then the left invariance of Haar measure immediately yields

(I)

Furthermore one also sees that 17 1 is a homomorphism for the convolution product, i.e

(2)

[I, § 1] THE REPRESENTATION ON Cc(G)

In the above proof, for simplicity, we omitted placing a vector v to the right

of '1T1(!p * ~), and to the right of every expression inside the integral signs The integrals are meant in this sense

Since !p has compact support and '1T is locally bounded, it follows that

'1T1(!p) is a bounded operator, i.e '1T1(!p)EEnd(H)

If '1T is a bounded representation, then instead of using functions

!pECc(G), we could have taken functionsfEe'(G) and formulas (1), (2) remain valid In other words, '1T1 extends to el(G), and furthermore we have the inequality

(3)

Thus '1T I is a continuous linear homomorphism (representation) of e I( G) into

End( H), as Banach algebras

If H is a Hilbert space, and'1T is unitary, then we also have the formula

(4)

where !p* is the function such that !p*(x) = !p(x I) This follows at once from the definition of the symbols involved

One can recover the values '1T(a) for aE G by knowing the values '1T1(!p)

for !p E C c ( G), as follows By a Dirac sequence on G we mean a sequence of

functions {!Pn}' real valued, in Cc(G), satisfying the following properties: DIR 1 We have !Pn ; 0 for all n

DIR 2 For all n, we have L !Pn(x) dx = 1

DIR 3 Given a neighborhood V of e in G, the support of!p" is contained in

V for all n sufficiently large

6 GENERAL RESULTS [I, § I]

The third condition shows that for large n, the area under CPn is concentrated near the origin A Dirac sequence looks like Fig 1

Figure 1

I t is obvious that Dirac sequences exist If G has a Coo structure, like SLiR),

one can even take the functions CPn to be Coo It is frequently convenient to use a slightly weaker condition than DIR 3, namely

DIR 3' Given a neighborhcod V of e in G, and its complement Z, and to, we

have

h CPnex) dx < E:

for all n sufficiently large

In other words, instead of assuming that the supports of the functions CPn

shrink to e, we merely assume the corresponding L 1 condition It is slightly more intuitive to work with the stronger condition which suffices for almost all applications When the need arises for the condition DIR 3', we shall assume that the reader can verify for himself the needed convergence state-ments valid with the same proof as for the other case

As will be mentioned later when we discuss analytic vectors, the tion DIR 3' becomes essential if we want the function CPn to be analytic functions (they cannot have compact support)

condi-At the beginning of this book, and for several chapters, we are principally interested in the measure theoretic aspects, or the COO aspects, of representa-tions Consequently we don't need any more about Dirac sequences than their definitions It may nevertheless be helpful to realize explicitly that some

[I, § 1] THE REPRESENTATION ON Cc(G) 7

convolutions arising in the classical literature are taken with Dirac sequences

We have the fan owing examples on R

i) Let cp be a Coo function on R which is positive, has compact support,

and is such that

LOO

oo cp(t) dt = 1

Then the sequence cp,,(t) = ncp(nt) is a Dirac sequence

ii) Let cp(t) = 'IT- 1/ 2 e- I " Let cp" be defined by the same formula as in (i) Then {cp,,} is a Dirac sequence

sequences, even reproducing some basic approximation results from Real

representa-Let a E G If {cp,,} is a Dirac sequence, then {7"aCP,,} is a Dirac sequence at

a (in the obvious sense) It is clear from (1) that

'lT1(7"affJn)v ~ 'IT(a)v

8 GENERAL RESULTS [I, § I]

as n ~ 00 The value 'IT(a)v is therefore obtained as a limit of values 'lT1(qJ)V

for suitable functions qJ E C c ( G)

Let W be a subspace of H (By subspace we shall always mean closed

subspace unless otherwise specified, in which case we sayan algebraic subspace.) By a dense subspace we mean a dense algebraic subspace We say that W is G-invariant if 'IT(x) W c W for aU x E G We make a similar definition for C c (G)-invariant

Quite generally, let S be a family of operators on H We say that W is

S-invariant if AWe W for every A E S Let Wo be a dense algebraic subspace

of W If Wo is S-invariant, then it is clear that W is also S-invariant

From the limiting property obtained above, we conclude:

A subspace W of H is G-invariant if and only if W is Cc(G)-invariant Let (f be a dense subspace of el(G) and assume that 'IT is bounded A subspace W of H is G-invariant if and only if W is also (f -invariant

For the convenience of the reader, we also recall convergence properties

of Dirac convolutions in e'( G)

Let fE el(G) and let Z be a compact set on whichf is continuous Let {qJn}

be a Dirac sequence Then qJn * f converges to f uniformly on Z

[1, §2] A CRITERION FOR COMPLETE REDUCIBILITY 9

If f is continuous with compact support, then {<p" f} converges uniformly

to f on the compact set (supp <p,,)(supp f) and is 0 outside this set Hence {<Pn f} is L I-convergent to f

Since Cc(G) is LI-dense in e\G), we obtain also:

Let f E el ( G) Then {<p" f} is L I-convergent to f

Proof First find <p E Cc( G) such that II <P - fll! < t: Then

1I<p" f - fill';;; II<pH f - <Pn <pIli + II<Pn <P - <pIli + 11<p - fill' Since II g hili';;; II glllllhih for two functions g, hE el(G), and since 11<p,,111

= 1 by DIR 2, our statement is proved by three epsilons

The same argument applies to LP instead of L 1, 1 ;;; p < 00 For our purposes, the most we would want it for is L2

§2 A CRITERION FOR COMPLETE REDUCIBILITY

Let

'TT: G ~ GL(H) and 'TT': G ~ GL(H')

be representations A morphism of 'TT into 'TT' is a continuous linear map

A: H ~ H' such that for every x E G the following diagram is commutative

(In the literature, a morphism is sometimes caned an intertwining operator.)

We say that A is an embedding if A is a topological linear isomorphism of H

onto a subspace of H' We say that A is an isomorphism if there is a morphism B of 'TT' into 'TT such that AB and BA are the identities of H' and H

respectively An isomorphism is also called an equivalence When H, H' are Hilbert spaces, and 'TT, 'TT' are unitary, then we may deal exclusively with unitary maps, i.e require that A be unitary The context will always make it clear whether this additional restriction is intended We say that 'TT occurs in 'TT'

if there exists an embedding of 'TT in 'TT'

A representation p: G ~ GL(E) is called irreducible if E has no variant subspace other than {OJ and E itself Let S be a set of operators on E

in-10 GENERAL RESULTS [I, §2]

We say that E is S-irreducible if E has no S-invariant subspace other than

{O} and E itself

Let H be a Hilbert space If there exist irreducible subspaces E 1, ••• ,Em

of H which are all G-isomorphic (under 'IT) to (p, E), and such that H can be expressed as a direct sum

and F contains no subspace 'IT ( G)-isomorphic to the Ei' then we say that E

occurs with multiplicity m in H It is easy to see that if this is the case, then in any expression of H as a direct sum,

H = E; E9 E2 E9 E9 E; E9 P,

where the E/ are 'IT ( G)-isomorphic to the E i , and (p, E) does not occur in P,

then r = m For the needed technique to reduce the proof to standard

algebraic arguments of semi-simplicity, see Real Analysis, Chapter VII,

Exer-cise 19 We call m the multiplicity of p in 'IT (or of E in H)

Let H be a Hilbert space and 'IT a representation of G in H We say that

H is completely reducible for 'IT, or that 'IT is completely reducible, if H is the orthogonal direct sum of irreducible subspaces We write such a direct sum as

"

H=EBH,.,

iEI where {i} ranges over a set of indices I, the Hi are subspaces invariant under

G, mutually orthogonal, and H is the closure of the algebraic space generated

by the Hi' This closure is indicated by the roof over the direct sum sign, which signifies algebraic direct sum We also say that the family {Hi} is an orthogonal decomposition of H

Let A; H -? H be an operator (continuous linear map) We recall that A

is called compact if A maps bounded sets into relatively compact sets (sets

whose closure is compact) Alternatively, we could say that if {v n } is a

bounded sequence, then {Avn} has a convergent subsequence A vector v E H

is called an eigenvector for A if Av = AV for some complex number A Given

A EC, the set of elements v E H such that Av = AV, together with 0, is a

subspace H", called the A-eigenspace of A

Spectral theorem for compact operators Let A be a compact hermitian operator on the Hilbert space E Then the family of eigenspaces {E,,}, where

A ranges over all eigenvalues (including 0), is an orthogonal decomposition

of E

Proof Let F be the closure of the subspace generated by all E" Let H

be the orthogonal complement of F Then H is A -invariant, and A induces a

[I, §2] A CRITERION FOR COMPLETE REDUCIBILITY II

compact hermitian operator on H, which has no eigenvalue We must show that H = {O} This will follow from the next lemma

Lemmo Let A be a compact hermitian operator on the Hilbert space

H -=1= {O} Let c = IAI Then cor - c is an eigenvalue for A

Proof There exists a sequence {x n } in H such that Ixnl = 1 and

Selecting a subsequence if necessary, we may assume that

<Axn, xn> ~ a

for some number a, and a = ± IA I Then

o , IAxn - ax,,12 = <Ax" - ax" Ax" - ax,,>

The right-hand side approaches 0 as n tends to infinity Since A is compact, after selecting a subsequence, we may assume that {Ax,,} converges to some

vector y, and then {ax,,} must converge to y also If a = 0, then IA I = 0 and

A = 0, so we are done If a -=1= 0, then {x n } itself must converge to some vector x, and then Ax = ax so that a is the desired eigenvalue for A, thus

proving our lemma, and the theorem

We observe that each EA has a Hilbert basis consisting of eigenvectors, namely any Hilbert basis of EA because all non-zero elements of EA are

eigenvectors Hence E itself has a Hilbert basis consisting of eigenvectors

Thus we recover precisely the analog of the theorem in the finite dimensional case Furthermore, we have some additional information, which follows trivially:

Each EA is finite dimensional if A -=1= 0, otherwise a denumerable subset from a Hilbert basis would provide a sequence contradicting the compactness

of A For a similar reason, given r > 0, there is only a finite number of

eigenvalues A such that IAI ~ r Thus 0 is a limit of the sequence of values if E is infinite dimensional If H is a Hilbert space and A a compact

eigen-operator on H, we may therefore write

00

H = ffi A HA = ffi H A ,

j= I • where the eigenvalues Aj are so ordered that IAi+ Ii ' IAil, and lim A; = o

12 GENERAL RESULTS [I, §3]

A subalgebra <:t of operators on H is said to be * -closed if whenever

A E <:t, then A * E <:t

Theorem 1 Let <:t be a *-closed subalgebra of compact operators on a Hilbert space H Then H is completely reducible for <:t, and each irreducible subspace occurs with finite multiplicity

Proof Let {E;} be a maximal orthogonal family of <:t-irreducible spaces, and let F be the orthogonal complement of the subspace generated by the E j • Since <:t is *-cIosed, it follows that F is <:t-invariant, and therefore we are reduced to proving, under the hypotheses of the theorem, that there exists

sub-an <:t-irreducible subspace We do this as follows

If A E <:t, then

A = 2 + I 2i '

so there exists an element A = A * *- 0 in <:t If M is an invariant subspace

*- to}, then the restriction of A to M satisfies the hypotheses of the theorem

Let A*-O be an eigenvalue for A Among all invariant subspaces M *- {O}, select one such that the eigenspace

Ml\ = {vEM,Av = AV}

has minimal dimension Let v E M, v *- O Then <:tv eM, and <:tv IS variant We contend that <:tv is irreducible Suppose that E *- to} is an invariant subspace of <:tv We can write

In-v = v E + v~

where VE is the E-component of v, and v~ E <:tv is perpendicular to E Note that

Av = Av = AVE + Av~ = AVE + AVE'

SO VE and v~ are A-eigenvectors for A If v E or vE = 0, say v~ = 0, then vEE,

whence <:tv = E This must necessarily happen, for otherwise, VE *- 0 and

vE *- 0 imply that E"cM" and E" *- M", so dim E" < dim M",

contradic-tion Hence <:tv is irreducible, and our theorem is proved

Remark To find the irreducible subspace, we needed only one compact hermitian operator in the algebra

§3 L 2 KERNELS AND OPERATORS

A certain type of kernels and operators will recur sufficiently often so that it is worthwhile to mention them independently here, rather than in an

[I, §3] L 2 KERNELS AND OPERATORS 13

appendix, or when we use them for the first time They give examples of compact operators

Theorem 2 Let (X, <:)It, dx) and (Y, '.'It, dy) be measured spaces, and assume that L2(X), L2(y) have countable orthogonal bases Let

q E e2( dx ® dy) Then the operator f ~ Qf such that

jqxE el(Y) We get by Schwarz

IQf(xW " Ilfll~ Ilqxll~,

and integrating,

II Qflli = jIQJ(xW dx " Ilflli jjlq(x,Y)1 2 dy dx

" II qll~ Ilfll~·

This proves that IQI " IIqlb, so that Q is a bounded operator

Let {cp;}, {~} be orthonormal bases for L2(X) and L2(Y) respectively

Let

Then {eli} is an orthonormal basis for L2(X X Y) To see this, it is first clear that the eli are of norm 1, and mutually orthogonal Let g E e 2 (X x Y) be perpendicular to all 0li' Then

Ix cp;(x) dx {I/I;(y)g(x,y) dy = 0

for all i, j Hence

is 0 except for x in a null set S in X If x¢ S, then for almost all y, we have

g(x,y) = O Hence g(x,y) = 0 for almost aU (x,y)EX X Y by Fubini's theorem

Let

14 GENERAL RESULTS [1, §3]

be the expression of q as a series in L 2(X X y), with constants aij' Let

qn = ~ alJij

;,j < n

be a finite truncation of the series It is immediately verified that the

corresponding operator Qn has finite dimensional image In fact, if () is a

function on X X Y such that ()(x,y) = 'P(x)l/I(y), then the image of the corresponding operator has dimension 1

We have already proved the inequality

and the expression on the right tends to 0 as n ? 00 Hence the operators

Qn' which are compact, tend in operator norm to Q, which is therefore also

compact This proves Theorem 2

We now make some comments of a formal nature on the trace of

operators represented by kernels as above Observe that 'Pi®Cjjj is an orthonormal basis for L2(X X X) Take Y = X, and write the Fourier

expansion for q in terms of 'Pi ® Cjjj'

Formany, we then expect the trace of Q to be given by

[I, §4] PLANCHEREL MEASURES 15

What we need to make sense of these computations are sufficient tions to make all the series converge, and the sum

cI>I(y) = IxI(x)cp(x,y) dx,

and a transpose operator,

lcI>g(X) = fy cp{x,y)g(y) dp.(y)

(On occasion, we use the reverse convention, interchanging cI> and 'cI>.) We also write cI>* for '<ii, i.e

cI>*g(x) = fy cp(x,y)g(y) dp.(y)

Then cI>* is the adjoint for the scalar products defined by

in other words, we have

<cI>I, g)1' = <1, cI>*g)

We have

4>* = 4> -I (f and only if 4> is unitary

This follows at once In some applications, we prove that 4;1 is unitary, and then we conclude that 4> -I is given by the starred kernel merely by applying the above formalism to functions g which obviously make the integrals converge, e.g continuous with compact support on locally compact spaces

In the applications of the above formalism, we always specify the function spaces on which the integrals converge In representation theory, we start with X = G (SL 2 (R) for this book), and dx is Haar measure

We shall be given measured spaces (X, dx) and (Y, dy) in a "natural" way We then want to find a positive function P on Y such that

4>* = 4;1 -1

for the measure dp.(y) = P(y) dy In other words, interpret the transpose to

be with respect to dp.(y) by

Icpg(X) = J cp(x, y)g(y)P(y) dy

Then we want CP*<P = Jd, on a suitable space of functions on X If this happens, we call P(y) dy the Plancherel measure for cp, and the formula CP*4;I = Jd is called the Plancherel inversion formula

Actually, in the Plancherel formula on a non-commutative group (as on

SL 2 (R) later) the situation is slightly more complicated, even though formally quite similar, because the map cp is operator valued, and in the inversion, we have to insert a trace Cf the end of the chapter on the Plancherel Formula Let G be the unitary equivalence classes of ir~educible unitary represen-tations of G It is usually possible to parametrize G, or an appropriate subset

of G, by means of an analytic space Y (set of zeros of analytic equations),

[I, §4] PLANCHEREL MEASURES 17

with a "natural" measure dy (for instance if Y is a piece of Euclidean space,

dy is Lebesgue measure) For SLzCR) we shall see that this space Y consist~ of two vertical lines and isolated points, looking like Fig 2 The space G is actually bigger Lebesgue measure is then our dy on the line while the points have discrete measure

Figure 2

Before dealing with the formula in full generality, we shall deal with a somewhat simpler situation of a Plancherel measure only for a special class of functions, bi-invariant under an appropriate compact subgroup of G This leads us to considering the representations of compact groups first, in the next chapter

II Compact Groups

§ 1 DECOMPOSITION OVER K FOR SL 2 (R)

In this section we essentially work out a special case of representation theory over compact groups, but in the context of SL 2 (R), providing a good introduction for what follows We bring out immediately the important role

of a maximal compact subgroup, the circle group K, i.e the group of matrices

f(r(O)yr(O'» = e- i "'j(y)e- im8'

for aU y E G and all real 0, Of

Lemma 1 The algebraic sum ~ S" m is LI-dense in CC<G) In fact, given f and fECc(G), there exi~l; a'function gE~SII.m such that the support of g is contained in K(supp f)K, and such that Ilf - glloo < f

19

20 COMPACT GROUPS [II, § I]

Proof Let

be the (n, m) Fourier coefficient of p Then fn m has support contained in

K(supp f)K The Cesaro Fejer kernels in one variable

! ~ ~ e,n8

M N-O Inl<N

form a Dirac sequence The product of the kernel in 0 and the kernel in 0'

form a Dirac sequence in two variables, say {CM(O, O')}, M = 1, 2, From the definitions, we see that (p CM)(O, 0) (convolution taken on the

product of the circle with itself) is the sum of terms cn, mfn, m(Y) with appropriate constant coefficients Cn, m arising from the sum in the Cesaro kernel We check that the argument giving the convergence of the convolu-tion toward

P(O,O) = f(y)

is uniform in y We have to estimate the difference

fflP(O, O')CM( - 0, - 0') - P(O, O)CM( - 0, - 0' )] dO dO'

Given E:, there exists a neighborhood U of (0, 0) such that

[II, § 1] DECOMPOSITION OVER K FOR SL 2 (R.)

Since G is unimodular, an integral with respect to y over G is invariant under

the transformationy ~ y-' Now lety ~ r(O)y From this invariance and the invariance under right and left translations, it follows that the above value

f * g(x) remains the same when multiplied by factors e im8 and e- ile This is

possible only when it is equal to 0, so (i) is proved The other two assertions are proved in an analogous way, left to the reader

The above lemma shows that Sn, n is an algebra under convolution The arguments are quite formal We now come to a more specialized property

Lemma 3 The algebra Sn, n is commutative

As we are concerned here with the arbitrary Sn, n' and not just So,o, we give the proof in a general context The reader will find it profitable to look at the simpler case of bi-invariant functions given at the beginning of Chapter IV,

due to Gelfand The generalization we give here is due to Silberger, Proc

AMS 1969, p 437 (The result was designed to workp-adically.)

Let 0 be an automorphism of a unimodular group G, or an automorphism By the uniqueness of Haar measure, there exists a positive

anti-number Ll(o) such that for allfE Cc(G) we have

f/(x") dx = Ll(o) f/(x) dx

We must have Ll(02) = Ll(o) Ll(o), and therefore if 0 2 = 1, it follows that

Ll(o) = 1 Thus the Haar integral is invariant under the transformation

x ~ x" We also write Ox instead of x"

Theorem 1 Let G be a a unimodular locally compact group Let K be a

compact subgroup Assume:

i) That there exists an anti-automorphism 'T of G, of order 2, such that

k T = k -I for all k E K

ii) If S is the set of elements s E G such that s., = s, then G = SK

iii) There exists an automorphism 0 of order 2 such that k" = k -I for all

k E K, and if s E S, then

for some kl E K

22 COMPACT GROUPS [II, § 1]

Let p: K ~ Cl be a character of K, and let Sp, p be the set of functions

fE Cc(G) such that

for all x E G and k1, k2 E K Then Sp, p is commutative

Proof Define j*(x) = f(x T) Then

(j*g)* = g* *j*

On the other hand, define f'(x) = f(x") Then

We prove the former (the latter is easier) We have

The conditions of Theorem 1 are verified, in view of the standard polar decomposition of a matrix, which we recall If x E GLn(R), we let y = x'x, so

[II, §l] DECOMPOSITION OVER K FOR SL 2 (R) 23

y is symmetric positive definite There is a basis for R" consisting of vectors such that

eigen-Let S2 = y, so that s has eigenvalues ±~i on Vi' and choose s so that

sign det s = sign det x

Let k = s -IX Then x = sk, and det k = 1 Also

Hence k is real unitary and we are done

Let us return to G = SL 2 (R) or GLi(R) Let

'IT: G ~ GL{H)

A; > O

be a representation of G into a Banach space H For each integer n let H" be the set of elements V E H such that

Then H" is a subspace (obviously closed)

LeIlUlUl 4 Assume that H is a Hilbert space and 'IT is unitary on K If

m *" n, then H" is perpendicular to Hm'

Proof For vEH" and wEHm we have 'IT(r(O»* = 'IT(r(-O», so

('IT(r(O»v, w> = eillil(v, w>

= (v, 'IT(r( - O»w> = ei""'(v, w>

The assertion follows

24 COMPACT GROUPS [II, § 1]

for k = r(fJ),

7T(k)7T 1 (f)V = 7T(k) f/(Y)7T(Y)V dy

= f/(Y)7T(ky)V dy

= f/(k-Y)7T(Y)V dy

This proves our lemma

If w lies in a finite direct sum of spaces H q , we let Wq denote its component in H q • Lemma 5 shows that 7T 1(f) for f E ~ Sn, m maps H into such

a direct sum

Lemmtl 6 Assume that 7T is irreducible Then the space Hq is irreducible for

Sq, q' and if Hq ¥= {O}, then 7T 1 (Sq, q)Hq ¥= {O}

Proof Let W be a proper subspace of Hq, invariant for 7T 1 (Sq, q)' If wE W

and f is a finite sum of functions fn, m E Sn, m' then by Lemma 5,

The algebra (£ = ~Sn,m is LI-dense in Cc(G) by Lemma I, and the algebraic space of elements 7T 1(f)W withfE (£ has its q-component contained

in W This is impossible because of the possibility of Dirac sequence

approximations (cf I, § I)

Theorem 2 Let 7T be an irreducible representation of G on a Banach space

H Let Hn be the subspace of vectors v such that

1*(x) = f(x- I ) It is immediately verified that 1* E Sn n (cf Lemma 2, ii) Hence, 7T 1(Sn n) is * -closed, and Schur's lemma implies that dim Hn = 0 or 1,

cf Appendix 1

Theorem 3 Let 7T be an irreducible representation of G on a Banach space

[II, § 1] DECOMPOSITION OVER K FOR SL 2 (R) 25

H Then the sum "i.H" is dense in H If H is a Hilbert space and 'IT is unitary on K, this sum is an orthogonal decomposition of H

Proof Let E be the (closed) subspace generated by the H" By Lemma 5

and the fact that the sum "i Sm n is dense in Cc( G), we conclude that E is

C c ( G)-invariant, whence is G-invariant Since 'IT is irreducible, it follows that

E = H If 'IT is unitary on K, we know from Lemma 4 that the H" are

mutually orthogonal This proves our theorem

Theorems 2 and 3 give us an indication of what will happen to the representations of SL 2 (R) Up to a point, they will be classified by the

presence or absence of appropriate H" In the theory of spherical functions,

we study the case when Ho occurs This is equivalent to the existence of a fixed vector under K, i.e a vector v E H, v =1= 0 such that 'IT(K)v = v In the alternative case, we are led to the discrete series

In this section we dealt with the K-decomposition of the representation

by means of the abstract nonsense of Haar measure and convolution In Chapter VI we return to this decomposition from the point of view of the derived representation on the Lie algebra, and get much more precise infor-mation on the way the group operates, via the exponential map This later chapter is mostly logically independent of the material on spherical functions, and the reader can easily read most of it immediately following the present discussion, to see how differentiability can be used

Let 'IT be a representation of G in a Banach space H, and suppose that H

is a direct sum

where H" is the n-th eigenspace of K as defined above Then the algebraic sum

is an algebraic subspace of H, dense in H It has an algebraic

characteriza-tion Let us say that an element v E H is X-finite if 'IT(K)v generates a finite

dimensional vector space

The algebraic space "i.H" is the space of K-finite vectors

Proof It is clear that every element of "i.H" is K-finite Conversely, suppose that an element v E H is K-finite A finite dimensional representation

of K in a space W decomposes into a direct sum of spaces W", and W" c H,

It is therefore clear that v is contained in "i.H"

The algebraic sum "i.H" will be denoted by H(K) Theorem 2 shows the importance of knowing that the dimensions of the components H" are finite

26 COMPACT GROUPS [II, §2]

Because of this, we define a representation 'IT to be admissible if dim Hn is

finite for all n Theorem 2 with this terminology then implies that every irreducible unitary representation is admissible We say that the representa-

tion is strictly admissible if the dimensions dim Hn are bounded

§2 COMPACT GROUPS IN GENERAL

In the case of SLiR), the circle group discussed in § 1 is commutative, and consequently one does not need the general theory of compact groups (which, however, follows closely the pattern given in the commutative case) However, the non-commutative aspects illustrate other principles which will

arise in a much more complicated fashion for the non-compact SL 2 (R), e.g the formalism of the trace Hence it is worthwhile to go through the theory of compact groups as an introduction to the other

Let K be a compact group with Haar measure equal to 1, and let

'IT: K ~ GL(H)

be a representation in a Hilbert space H By a remark at the beginning of Chapter I, § 1, we know that 'IT is bounded

We shall now see that we can find an equivalent norm on H such that 'IT

is unitary with respect to this norm For v E H define

Ivl; = fKI'IT (k)vI 2 dk

Then Ivl; ~ C21vl2 if C is a bound for 'IT Hence Ivl" ~ Clvl On the other hand, for k E K,

-I

Ivl = 1'IT(k) 'IT(k)vl ~ CI'IT(k)vl,

whence

and This proves that I I." is equivalent to I I, and it is clear that 'IT is unitary with respect to the norm II" This proves what we wanted

On L 2(K) (with respect to Haar measure), we have an operation of right translation T, defined by

T(y)f(x) = f(xy)

Then T is unitary because

f)f(x Y )1 2 dx = f)f(xW dx,

[II, §2] COMPACT GROUPS IN GENERAL 27

since a compact group is unimodular (a homomorphism of a compact group

into the positive reals must be trivial) We also call T the regular

representa-tion (on the right)

Let <p E C c ( G) Then

Tl(<p)f(x) = JKf(xy)<p(y) dy

= Lf(Y)<p(x-y) dy

=f*<p-(x)

where <p-(x) = <p(x- 1) We see that Tl(<p) arises from a kernel

which is continuous on K X K By the Weierstrass-Stone theorem, any continuous function on K X K can be uniformly approximated by finite sums

and the operator arising from the kernel <p/i9t/J;, i.e the function

for each t, has a one-dimensional image Consequently, T'(<p) can be

approximated in norm by operators with finite dimensional image, whence

Tl(<p) is compact By I, §2, Th 1 we get:

Theorem 1 Under the regular representation, L 2(K) is the orthogonal direct

sum of irreducible subs paces , i.e the regular representation is completely

reducible

Theorem 2 Let w: K ~ Aut(H) be a unitary irreducible representation of

a compact group K Then H is finite dimensional

Proof Let u be a unit vector in H and let P be the orthogonal projection

on the one-dimensional space (u) Let Q: H -4 H be the continuous linear map defined by

Qv = L w(x) - I Pw(x)v dx

Then Q commutes with all operators w(y), y E K (immediate by the right and